ARC

Fixes https://github.com/rui314/mold/issues/1442

.github/FUNDING.yml

@@ -0,0 +1,13 @@
+# These are supported funding model platforms
+
+github: [rui314] # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+patreon: # Replace with a single Patreon username
+open_collective: # Replace with a single Open Collective username
+ko_fi: # Replace with a single Ko-fi username
+tidelift: # Replace with a single Tidelift platform-name/package-name e.g., npm/babel
+community_bridge: # Replace with a single Community Bridge project-name e.g., cloud-foundry
+liberapay: # Replace with a single Liberapay username
+issuehunt: # Replace with a single IssueHunt username
+otechie: # Replace with a single Otechie username
+lfx_crowdfunding: # Replace with a single LFX Crowdfunding project-name e.g., cloud-foundry
+custom: # Replace with up to 4 custom sponsorship URLs e.g., ['link1', 'link2']

.github/workflows/build-all.yml

@@ -0,0 +1,51 @@
+name: Build all tarballs
+
+on:
+  schedule:
+    - cron: '0 0 * * *'
+  release:
+    types: [ published ]
+  workflow_dispatch:
+
+jobs:
+  build-tarballs:
+    strategy:
+      matrix:
+        include:
+          - { target: x86_64,      os: ubuntu-24.04 }
+          - { target: aarch64,     os: ubuntu-24.04-arm }
+          - { target: arm,         os: ubuntu-24.04-arm }
+          - { target: riscv64,     os: ubuntu-24.04 }
+          - { target: ppc64le,     os: ubuntu-24.04 }
+          - { target: s390x,       os: ubuntu-24.04 }
+          - { target: loongarch64, os: ubuntu-24.04 }
+
+    runs-on: ${{ matrix.os }}
+
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+    - name: Checkout Repository
+      uses: actions/checkout@v4
+
+    - name: Install Podman
+      run: sudo apt-get update && sudo apt-get install -y podman qemu-user-static
+
+    - name: Login to GitHub Container Registry
+      uses: redhat-actions/podman-login@v1
+      with:
+        registry: ghcr.io
+        username: ${{ github.repository_owner }}
+        password: ${{ secrets.GITHUB_TOKEN }}
+
+    - name: Build a tarball
+      run: ./dist.sh ${{ matrix.target }}
+
+    - name: Upload artifact
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ matrix.target }}
+        path: dist/mold-*.tar.gz
+        compression-level: 0

.github/workflows/build-native.yml

@@ -0,0 +1,44 @@
+name: Build native tarballs
+
+on:
+  push:
+  workflow_dispatch:
+
+jobs:
+  build-tarballs:
+    strategy:
+      matrix:
+        include:
+          - { target: x86_64,  os: ubuntu-24.04 }
+          - { target: aarch64, os: ubuntu-24.04-arm }
+          - { target: arm,     os: ubuntu-24.04-arm }
+
+    runs-on: ${{ matrix.os }}
+
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+    - name: Checkout Repository
+      uses: actions/checkout@v4
+
+    - name: Install Podman
+      run: sudo apt-get update && sudo apt-get install -y podman qemu-user-static
+
+    - name: Login to GitHub Container Registry
+      uses: redhat-actions/podman-login@v1
+      with:
+        registry: ghcr.io
+        username: ${{ github.repository_owner }}
+        password: ${{ secrets.GITHUB_TOKEN }}
+
+    - name: Build a tarball
+      run: ./dist.sh ${{ matrix.target }}
+
+    - name: Upload artifact
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ matrix.target }}
+        path: dist/mold-*.tar.gz
+        compression-level: 0

.github/workflows/ci.yml

@@ -1,56 +1,149 @@
 name: CI
 on:
   push:
-    branches: [ main ]
   pull_request:
-    branches: [ main ]
+env:
+  UBSAN_OPTIONS: print_stacktrace=1:halt_on_error=1
 jobs:
-  build-clang:
-    runs-on: ubuntu-20.04
+  build-sanitizers:
+    strategy:
+      matrix:
+        target:
+        - ''
+        - '-DMOLD_USE_ASAN=On'
+        - '-DMOLD_USE_TSAN=On'
+    runs-on: ubuntu-24.04
     steps:
-    - uses: actions/checkout@v2
-    - name: apt-get
-      run: sudo dpkg --add-architecture i386 && sudo apt-get update && sudo apt-get install build-essential libstdc++-10-dev cmake clang libssl-dev zlib1g-dev libxxhash-dev libtbb-dev git bsdmainutils dwarfdump libc6-dev:i386 lib32gcc-10-dev libstdc++-10-dev-arm64-cross gcc-aarch64-linux-gnu g++-aarch64-linux-gnu qemu-user
-    - name: ccache
-      uses: hendrikmuhs/ccache-action@v1
-    - name: make
+    - uses: actions/checkout@v4
+    - uses: rui314/setup-mold@staging
+    - run: sudo ./install-build-deps.sh
+    - name: build
       run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make CC=clang CXX=clang++ DEBUG=1 -j$(nproc)
-    - name: make test
-      run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make test -j$(nproc)
-  build-gcc11:
-    runs-on: ubuntu-20.04
-    container: gcc:11.1.0
+        sudo apt-get install -y clang-18 clang gcc-multilib gdb dwarfdump zstd
+        mkdir build
+        cd build
+        cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_C_COMPILER=clang-18 -DCMAKE_CXX_COMPILER=clang++-18 ${{ matrix.target }} ..
+        cmake --build . -j$(nproc)
+    - run: cd build; ctest --output-on-failure -j$(nproc)
+    - name: archive test results
+      uses: actions/upload-artifact@v4
+      if: failure()
+      with:
+        name: test-results-clang
+        path: |
+          build
+          !build/CMakeFiles
+
+  build-multi-archs:
+    runs-on: ubuntu-24.04
     steps:
-    - uses: actions/checkout@v2
-    - name: apt-get
-      run: dpkg --add-architecture i386 && apt-get update && apt-get install -y sudo libssl-dev zlib1g-dev libxxhash-dev cmake clang dwarfdump bsdextrautils libgcc-9-dev gcc-aarch64-linux-gnu g++-aarch64-linux-gnu libc6-dev:i386 lib32gcc-10-dev qemu-user
-    - name: ccache
-      uses: hendrikmuhs/ccache-action@v1
-    - name: make
+    - uses: actions/checkout@v4
+    - name: install-build-deps
       run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make CC=gcc CXX=g++ DEBUG=1 -j$(nproc)
-    - name: make test
+        sudo ./install-build-deps.sh
+        sudo ./install-cross-tools.sh
+        sudo .github/workflows/install-extras.sh
+    - name: build
       run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make test -j$(nproc)
+        mkdir build
+        cd build
+        cmake ..
+        cmake --build . -j$(nproc)
+    - run: cd build; ctest --output-on-failure -j$(nproc)
+    - name: archive test results
+      uses: actions/upload-artifact@v4
+      if: failure()
+      with:
+        name: test-results-multi-archs
+        path: |
+          build
+          !build/CMakeFiles
+
+  build-distros:
+    strategy:
+      matrix:
+        distro:
+        - alpine
+        - archlinux
+        - debian:11 # GCC 10 and CMake 3.18 - the minimum supported by mold
+        - fedora
+        - gentoo/stage3
+        - opensuse/tumbleweed
+        - ubuntu:22.04
+        - ubuntu:25.04
+    runs-on: ubuntu-latest
+    container: ${{ matrix.distro }}
+    steps:
+    - uses: actions/checkout@v4
+    - run: ./install-build-deps.sh
+    - name: build
+      run: |
+        mkdir build
+        cd build
+        cmake ..
+        cmake --build . -j$(nproc)
+    - run: cd build; ctest --output-on-failure -j$(nproc)
+    - name: archive test results
+      uses: actions/upload-artifact@v4
+      if: failure()
+      with:
+        name: test-results-${{ matrix.distro }}
+        path: |
+          build
+          !build/CMakeFiles
+
   build-macos:
-    runs-on: macos-11
+    runs-on: macos-latest
     steps:
-    - uses: actions/checkout@v2
-    - name: brew
-      run: brew install xxhash
-    - name: ccache
-      uses: hendrikmuhs/ccache-action@v1
-    - name: make
+    - uses: actions/checkout@v4
+    - name: build
       run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make DEBUG=1 -j$(sysctl -n hw.logicalcpu)
-    - name: make test
+        mkdir build
+        cd build
+        cmake ..
+        cmake --build . -j$(sysctl -n hw.physicalcpu)
+
+  build-windows:
+    runs-on: windows-latest
+    steps:
+    - uses: actions/checkout@v4
+    - name: build
       run: |
-        export PATH="/usr/lib/ccache:/usr/local/opt/ccache/libexec:$PATH"
-        make test -j$(sysctl -n hw.logicalcpu)
+        mkdir build
+        cd build
+        cmake -T clangcl ..
+        cmake --build . -j $Env:NUMBER_OF_PROCESSORS
+
+  build-msys:
+    runs-on: windows-latest
+    steps:
+    - uses: actions/checkout@v4
+    - name: Setup MSYS2
+      uses: msys2/setup-msys2@v2
+      with:
+        msystem: UCRT64
+        update: true
+        pacboy: gcc-libs:p libwinpthread-git:p tbb:p zlib:p zstd:p dlfcn:p cc:p cmake:p ninja:p
+    - name: build
+      shell: msys2 {0}
+      run: |
+        mkdir build
+        cd build
+        cmake -GNinja -DMOLD_USE_MIMALLOC=OFF -DMOLD_USE_SYSTEM_TBB=ON ..
+        cmake --build . -j $(nproc)
+
+  build-freebsd:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - name: Build and test
+      uses: vmactions/freebsd-vm@v1
+      with:
+        usesh: true
+        run: |
+          ./install-build-deps.sh
+          mkdir build
+          cd build
+          cmake ..
+          cmake --build . -j$(nproc)
+          ctest --output-on-failure -j$(nproc)

.github/workflows/install-extras.sh

@@ -0,0 +1,61 @@
+#!/bin/bash -x
+
+apt-get update
+apt-get install -y wget xz-utils
+
+# Install a 32-bit RISC-V toolchain
+mkdir /rv32
+wget -O- -q https://github.com/riscv-collab/riscv-gnu-toolchain/releases/download/2025.01.20/riscv32-glibc-ubuntu-22.04-gcc-nightly-2025.01.20-nightly.tar.xz | tar -C /rv32 --strip-components=1 --xz -xf -
+
+ln -sf /rv32/sysroot /usr/riscv32-linux-gnu
+echo '/rv32/bin/riscv32-unknown-linux-gnu-gcc -L/usr/riscv32-linux-gnu "$@"' > /usr/bin/riscv32-linux-gnu-gcc
+echo '/rv32/bin/riscv32-unknown-linux-gnu-g++ -L/usr/riscv32-linux-gnu "$@"' > /usr/bin/riscv32-linux-gnu-g++
+chmod 755 /usr/bin/riscv32-linux-gnu-{gcc,g++}
+
+for i in objdump objcopy strip; do
+  ln -sf /rv32/bin/riscv32-unknown-linux-gnu-$i /usr/bin/riscv32-linux-gnu-$i
+done
+
+# Install a LoongArch toolchain
+mkdir /larch
+wget -O- -q https://github.com/loongson/build-tools/releases/download/2025.02.21/x86_64-cross-tools-loongarch64-binutils_2.44-gcc_14.2.0-glibc_2.41.tar.xz | tar -C /larch --strip-components=1 --xz -xf -
+
+cp -r /larch/loongarch64-unknown-linux-gnu/lib/* /larch/target/lib64
+ln -sf /larch/target /usr/loongarch64-linux-gnu
+
+for i in gcc g++ objdump objcopy strip; do
+  ln -sf /larch/bin/loongarch64-unknown-linux-gnu-$i /usr/bin/loongarch64-linux-gnu-$i
+done
+
+wget -O /usr/local/bin/qemu-loongarch64 -q https://github.com/loongson/build-tools/releases/download/2024.11.01/qemu-loongarch64
+chmod 755 /usr/local/bin/qemu-loongarch64
+
+# Install ARM64 big-endian toolchain
+mkdir /aarch64be
+wget -O- -q https://sources.buildroot.net/toolchain-external-arm-aarch64-be/arm-gnu-toolchain-13.3.rel1-x86_64-aarch64_be-none-linux-gnu.tar.xz | tar -C /aarch64be --strip-components=1 --xz -xf -
+
+ln -sf /aarch64be/aarch64_be-none-linux-gnu/libc /usr/aarch64_be-linux-gnu
+echo '/aarch64be/bin/aarch64_be-none-linux-gnu-gcc -L/usr/aarch64_be-linux-gnu "$@"' > /usr/bin/aarch64_be-linux-gnu-gcc
+echo '/aarch64be/bin/aarch64_be-none-linux-gnu-g++ -L/usr/aarch64_be-linux-gnu "$@"' > /usr/bin/aarch64_be-linux-gnu-g++
+chmod 755 /usr/bin/aarch64_be-linux-gnu-{gcc,g++}
+
+for i in objdump objcopy strip; do
+  ln -sf /aarch64be/bin/aarch64_be-none-linux-gnu-$i /usr/bin/aarch64_be-linux-gnu-$i
+done
+
+# Install SH4 big-endian toolchain
+mkdir /sh4aeb
+wget -O- -q https://toolchains.bootlin.com/downloads/releases/toolchains/sh-sh4aeb/tarballs/sh-sh4aeb--glibc--stable-2024.05-1.tar.xz | tar -C /sh4aeb --strip-components=1 --xz -xf -
+ln -sf /sh4aeb/sh4aeb-buildroot-linux-gnu/sysroot /usr/sh4aeb-linux-gnu
+echo '/sh4aeb/bin/sh4aeb-linux-gcc -L/usr/sh4aeb-linux-gnu "$@"' > /usr/bin/sh4aeb-linux-gnu-gcc
+echo '/sh4aeb/bin/sh4aeb-linux-g++ -L/usr/sh4aeb-linux-gnu "$@"' > /usr/bin/sh4aeb-linux-gnu-g++
+chmod 755 /usr/bin/sh4aeb-linux-gnu-{gcc,g++}
+
+for i in objdump objcopy strip; do
+  ln -sf /sh4aeb/bin/sh4aeb-linux-$i /usr/bin/sh4aeb-linux-gnu-$i
+done
+
+# Install Intel SDE CPU emulator for CET-related tests
+mkdir /sde
+wget -O- -q https://downloadmirror.intel.com/843185/sde-external-9.48.0-2024-11-25-lin.tar.xz | tar -C /sde --strip-components=1 --xz -xf -
+ln -s /sde/sde64 /usr/bin

.github/workflows/update-manpage.yml

@@ -0,0 +1,33 @@
+name: Update manpage
+
+on:
+  push:
+    paths:
+      - 'docs/mold.md'
+    branches:
+      - main
+  workflow_dispatch:
+
+jobs:
+  update-manpage:
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Check out repository
+      uses: actions/checkout@v4
+
+    - name: Install ronn
+      run: sudo apt-get update && sudo apt-get install -y ronn
+
+    - name: Generate mold.1 from mold.md
+      run: ronn --roff docs/mold.md
+
+    - name: Configure Git
+      run: |
+        git config --global user.name 'Rui Ueyama'
+        git config --global user.email '[email protected]'
+
+    - name: Commit and push if mold.1 is updated
+      run: |
+        git add docs/mold.1
+        git diff --staged --quiet || (git commit -m "Update mold.1 (automated commit)" && git push)

.gitignore

@@ -1,13 +1,11 @@
 *.o
 *.so
 *~
+**/.DS_Store
 .cache
-compile_commands.json
-/mold
-/test/tmp
-options.inc
 core
-ld
-ld64.mold
 gentoo
-out
+/mold-*.tar.gz
+/build*
+/mold
+/dist

CMakeLists.txt

@@ -0,0 +1,458 @@
+# You can customize a build by specifying CMake options. An option may be
+# given in the -Dvariable=value form. For a boolean variable, `ON` or `1`
+# means true, while `OFF` or `0` means false.
+#
+# Here are a couple of common cmake options:
+#
+#  -DCMAKE_C_COMPILER=<command-name>
+#
+#   Specifies the C compiler name to use. The default value is `cc`.
+#
+#  -DCMAKE_CXX_COMPILER=<command-name>
+#
+#   Specifies the C++ compiler name to use. The default value is `c++`.
+#
+#  -DCMAKE_INSTALL_PREFIX=<directory>
+#
+#   Specifies the install target directory. The default value is `/usr/local`.
+#
+#  -DCMAKE_BUILD_TYPE=[Debug | Release | RelWithDebInfo | MinSizeRel]
+#
+#   Specifies the build type. The default is `Release`, which is the right
+#   option unless you are debugging mold.
+#
+# An example of a cmake command line is shown below:
+#
+#   $ cmake -DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_CXX_COMPILER=g++-12 ..
+#
+# where `..` refers to this directory.
+#
+# With cmake, you may run `cmake --install .` instead of `make install` to
+# install build artifacts to system directories. If you want to install
+# artifacts to a temporary target directory, run `cmake --install . --prefix
+# <dir>`.
+#
+# You can see the current cmake variables and their values by running
+# `cmake -N -L .` in a build directory.
+#
+# Note that in this file, we provide various dials and knobs to configure
+# how to build mold. However, as a policy, we do not provide a way to
+# enable/disable any individual mold's feature. In other words, we do not
+# provide options like `--enable-foo` or `--disable-foo`. The motivation
+# behind this is build reproducibility. We want to guarantee that all builds
+# of the mold linker of the same version will have the exact same set of
+# features and behave exactly the same.
+
+cmake_minimum_required(VERSION 3.14)
+project(mold VERSION 2.37.1)
+
+include(CMakeDependentOption)
+include(CheckSymbolExists)
+include(GNUInstallDirs)
+
+add_executable(mold)
+target_compile_features(mold PRIVATE cxx_std_20)
+
+if(MINGW)
+  target_link_libraries(mold PRIVATE dl)
+else()
+  target_link_libraries(mold PRIVATE ${CMAKE_DL_LIBS})
+endif()
+
+# Build mold itself using mold if -DMOLD_USE_MOLD=ON
+option(MOLD_USE_MOLD "Use mold to build mold" OFF)
+if(MOLD_USE_MOLD)
+  target_link_options(mold PRIVATE -fuse-ld=mold -Wl,--gdb-index)
+
+  if(CMAKE_BUILD_TYPE MATCHES "^Rel")
+    target_link_options(mold PRIVATE -Wl,--gc-sections,--icf=safe)
+  endif()
+endif()
+
+if(NOT "${CMAKE_CXX_COMPILER_FRONTEND_VARIANT}" STREQUAL "MSVC")
+  target_compile_options(mold PRIVATE
+    -fno-exceptions
+    -fno-unwind-tables
+    -fno-asynchronous-unwind-tables
+    -ffunction-sections
+    -fdata-sections
+    -Wall
+    -Wextra
+    -Wno-sign-compare
+    -Wno-unused-function
+    -Wno-unused-parameter
+    -Wno-missing-field-initializers
+    -ggnu-pubnames)
+endif()
+
+if(CMAKE_BUILD_TYPE STREQUAL "Debug")
+  target_compile_options(mold PRIVATE -D_GLIBCXX_ASSERTIONS)
+endif()
+
+if(CMAKE_SYSTEM_NAME STREQUAL "OpenBSD")
+  set(OPENBSD ON)
+endif()
+
+# Build mold with -flto if -DMOLD_LTO=ON
+option(MOLD_LTO "Build mold with link-time optimization enabled")
+if(MOLD_LTO)
+  set_property(TARGET mold PROPERTY INTERPROCEDURAL_OPTIMIZATION ON)
+endif()
+
+# Enable AddressSanitizer if -DMOLD_USE_ASAN=ON
+option(MOLD_USE_ASAN "Build mold with AddressSanitizer" OFF)
+if(MOLD_USE_ASAN)
+  target_compile_options(mold PRIVATE -fsanitize=address -fsanitize=undefined)
+  target_link_options(mold PRIVATE -fsanitize=address -fsanitize=undefined)
+endif()
+
+# Enabled ThreadSanitizer if -DMOLD_USE_TSAN=ON
+option(MOLD_USE_TSAN "Build mold with ThreadSanitizer" OFF)
+if(MOLD_USE_TSAN)
+  target_compile_options(mold PRIVATE -fsanitize=thread)
+  target_link_options(mold PRIVATE -fsanitize=thread)
+endif()
+
+# Statically-link libstdc++ if -DMOLD_MOSTLY_STATIC=ON.
+#
+# This option is intended to be used by `./dist.sh` script to create a
+# mold binary that works on various Linux distros. You probably don't
+# need nor want to set this to ON.
+option(MOLD_MOSTLY_STATIC "Statically link libstdc++ and some other libraries" OFF)
+if(MOLD_MOSTLY_STATIC)
+  target_link_options(mold PRIVATE -static-libstdc++)
+endif()
+
+# Find zlib. If libz.so is not found, we compile a bundled one and
+# statically-link it to mold.
+find_package(ZLIB QUIET)
+if(ZLIB_FOUND AND NOT MOLD_MOSTLY_STATIC)
+  target_link_libraries(mold PRIVATE ZLIB::ZLIB)
+else()
+  set(ZLIB_BUILD_EXAMPLES OFF CACHE INTERNAL "")
+  add_subdirectory(third-party/zlib EXCLUDE_FROM_ALL)
+  target_include_directories(zlibstatic INTERFACE third-party/zlib
+    $<TARGET_PROPERTY:zlibstatic,BINARY_DIR>)
+  target_link_libraries(mold PRIVATE zlibstatic)
+endif()
+
+# Find BLAKE3 cryptographic hash library. Just like zlib, if libblkae3.so
+# is not found, we compile a bundled one and statically-link it to mold.
+find_package(BLAKE3 QUIET)
+if(BLAKE3_FOUND AND NOT MOLD_MOSTLY_STATIC)
+  target_link_libraries(mold PRIVATE BLAKE3::blake3)
+else()
+  function(mold_add_blake3)
+    set(BUILD_SHARED_LIBS OFF)
+    add_subdirectory(third-party/blake3/c EXCLUDE_FROM_ALL)
+    target_link_libraries(mold PRIVATE blake3)
+    target_include_directories(mold PUBLIC third-party/blake3/c)
+  endfunction()
+
+  mold_add_blake3()
+endif()
+
+# Find zstd compression library. If zstd.h is not found, we compile a
+# bundled one and statically-link it to mold.
+include(CheckIncludeFile)
+check_include_file(zstd.h HAVE_ZSTD_H)
+
+if(HAVE_ZSTD_H AND NOT MOLD_MOSTLY_STATIC)
+  target_link_libraries(mold PRIVATE zstd)
+else()
+  add_subdirectory(third-party/zstd/build/cmake EXCLUDE_FROM_ALL)
+  target_compile_definitions(libzstd_static PRIVATE
+    ZSTD_BUILD_STATIC=1
+    ZSTD_BUILD_SHARED=0
+    ZSTD_BUILD_PROGRAMS=0
+    ZSTD_MULTITHREAD_SUPPORT=0
+    ZSTD_BUILD_TESTS=0)
+  target_include_directories(mold PUBLIC third-party/zstd/lib)
+  target_link_libraries(mold PRIVATE libzstd_static)
+endif()
+
+# Find mimalloc. mimalloc is an alternative malloc implementation
+# optimized for multi-threaded applications.
+#
+# If you want to use the usual libc's malloc, pass -DMOLD_USE_MIMALLOC=OFF.
+#
+# We enable mimalloc by default for 64-bit targets. It doesn't seem to
+# be stable on 32-bit targets.
+cmake_dependent_option(
+  MOLD_USE_MIMALLOC "Use mimalloc" ON
+  "CMAKE_SIZEOF_VOID_P EQUAL 8; NOT APPLE; NOT ANDROID; NOT OPENBSD; NOT MOLD_USE_ASAN; NOT MOLD_USE_TSAN" OFF)
+
+cmake_dependent_option(
+  MOLD_USE_SYSTEM_MIMALLOC "Use system or vendored mimalloc" OFF
+  MOLD_USE_MIMALLOC OFF)
+
+# By default, we build a bundled mimalloc and statically-link it to
+# mold. If you want to dynamically link to the system's
+# libmimalloc.so, pass -DMOLD_USE_SYSTEM_MIMALLOC=ON.
+if(MOLD_USE_MIMALLOC)
+  if(MOLD_USE_SYSTEM_MIMALLOC)
+    find_package(mimalloc REQUIRED)
+    target_link_libraries(mold PRIVATE mimalloc)
+  else()
+    function(mold_add_mimalloc)
+      set(MI_BUILD_STATIC ON CACHE INTERNAL "")
+      set(MI_BUILD_TESTS OFF CACHE INTERNAL "")
+      add_subdirectory(third-party/mimalloc EXCLUDE_FROM_ALL)
+      target_compile_definitions(mimalloc-static PRIVATE MI_USE_ENVIRON=0)
+      target_link_libraries(mold PRIVATE mimalloc-static)
+    endfunction()
+
+    mold_add_mimalloc()
+  endif()
+endif()
+
+# Find TBB. TBB (OneTBB or Intel TBB) is a high-level threading library.
+# Use of this library is mandatory.
+#
+# By default, we build a bundled one and statically-link the library
+# to mold. If you want to link to the system's libtbb2.so, pass
+# -DMOLD_USE_SYSTEM_TBB=ON.
+option(MOLD_USE_SYSTEM_TBB "Use system or vendored TBB" OFF)
+if(MOLD_USE_SYSTEM_TBB)
+  find_package(TBB REQUIRED)
+  target_link_libraries(mold PRIVATE TBB::tbb)
+else()
+  function(mold_add_tbb)
+    set(BUILD_SHARED_LIBS OFF)
+    set(TBB_TEST OFF CACHE INTERNAL "")
+    set(TBB_STRICT OFF CACHE INTERNAL "")
+    add_subdirectory(third-party/tbb EXCLUDE_FROM_ALL)
+    target_compile_definitions(tbb PRIVATE __TBB_DYNAMIC_LOAD_ENABLED=0)
+    target_link_libraries(mold PRIVATE TBB::tbb)
+  endfunction()
+
+  mold_add_tbb()
+endif()
+
+# We always use Clang to build mold on Windows. MSVC can't compile mold.
+if(WIN32)
+  if(MSVC AND NOT CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+    message(FATAL_ERROR
+      "Your compiler is not supported; install Clang from Visual Studio Installer and re-run cmake with '-T clangcl'")
+  endif()
+
+  target_compile_definitions(mold PRIVATE NOGDI NOMINMAX)
+  if(MINGW)
+    target_compile_definitions(mold PRIVATE _WIN32_WINNT=0xA00)
+    target_link_libraries(mold PRIVATE bcrypt)
+  endif()
+else()
+  include(CheckLibraryExists)
+  check_library_exists(m pow "" LIBM_FOUND)
+  if(LIBM_FOUND)
+    target_link_libraries(mold PRIVATE m)
+  endif()
+endif()
+
+# Build mold-wrapper.so
+if(NOT APPLE AND NOT WIN32)
+  add_library(mold-wrapper SHARED)
+  install(TARGETS mold-wrapper DESTINATION ${CMAKE_INSTALL_LIBDIR}/mold)
+
+  # Remove the default `lib` prefix
+  set_target_properties(mold-wrapper PROPERTIES PREFIX "")
+  target_link_libraries(mold-wrapper PRIVATE ${CMAKE_DL_LIBS})
+  target_sources(mold-wrapper PRIVATE src/mold-wrapper.c)
+endif()
+
+# If atomics doesn't work by default, add -latomic.
+# We need the flag on riscv, armv6 and m68k.
+include(CheckCXXSourceCompiles)
+check_cxx_source_compiles("#include <atomic>
+int main() {
+  std::atomic_uint8_t  a;
+  std::atomic_uint16_t b;
+  std::atomic_uint32_t c;
+  std::atomic_uint64_t d;
+  return ++a + ++b + ++c + ++d;
+}" HAVE_FULL_ATOMIC_SUPPORT)
+
+if(NOT HAVE_FULL_ATOMIC_SUPPORT)
+  target_link_libraries(mold PRIVATE atomic)
+endif()
+
+# Add -pthread
+if(NOT APPLE AND NOT MSVC)
+  target_compile_options(mold PRIVATE -pthread)
+  target_link_options(mold PRIVATE -pthread)
+endif()
+
+check_symbol_exists(madvise sys/mman.h HAVE_MADVISE)
+check_symbol_exists(uname sys/utsname.h HAVE_UNAME)
+
+# Create a .cc file containing the current git hash for `mold --version`.
+add_custom_target(git_hash
+  COMMAND ${CMAKE_COMMAND}
+    -DSOURCE_DIR=${CMAKE_SOURCE_DIR}
+    -DOUTPUT_FILE=${CMAKE_BINARY_DIR}/git-hash.cc
+    -P ${CMAKE_SOURCE_DIR}/lib/update-git-hash.cmake
+  DEPENDS lib/update-git-hash.cmake
+  BYPRODUCTS git-hash.cc
+  VERBATIM)
+
+add_dependencies(mold git_hash)
+
+# Create config.h file
+configure_file(lib/config.h.in config.h)
+include_directories(${CMAKE_CURRENT_BINARY_DIR})
+
+# Almost all functions are template in mold which take a target type
+# (e.g. X86_64) as its type parameter. Since we suport more than 10
+# targets, compiling a single source file for all the targets is very
+# slow.
+#
+# As a workaround, we create a .cc file for each target and spawn many
+# compiler instances. This is hacky but greatly reduces compile time
+# on a multicore machine.
+list(APPEND MOLD_ELF_TARGETS
+  X86_64 I386 ARM64LE ARM64BE ARM32 RV32LE RV32BE RV64LE RV64BE PPC32
+  PPC64V1 PPC64V2 S390X SPARC64 M68K SH4LE SH4BE LOONGARCH32 LOONGARCH64 ARC)
+
+list(APPEND MOLD_ELF_TEMPLATE_FILES
+  src/arch-arm64.cc
+  src/arch-loongarch.cc
+  src/arch-riscv.cc
+  src/arch-sh4.cc
+  src/cmdline.cc
+  src/filetype.cc
+  src/gc-sections.cc
+  src/gdb-index.cc
+  src/icf.cc
+  src/input-files.cc
+  src/input-sections.cc
+  src/linker-script.cc
+  src/main.cc
+  src/mapfile.cc
+  src/output-chunks.cc
+  src/passes.cc
+  src/relocatable.cc
+  src/shrink-sections.cc
+  src/thunks.cc
+  src/tls.cc
+  )
+
+if(WIN32 AND NOT MINGW)
+  list(APPEND MOLD_ELF_TEMPLATE_FILES src/lto-win32.cc)
+else()
+  list(APPEND MOLD_ELF_TEMPLATE_FILES src/lto-unix.cc)
+endif()
+
+if(WIN32)
+  list(APPEND MOLD_ELF_TEMPLATE_FILES
+    src/output-file-win32.cc
+    src/subprocess-win32.cc
+    )
+else()
+  list(APPEND MOLD_ELF_TEMPLATE_FILES
+    src/output-file-unix.cc
+    src/subprocess-unix.cc
+  )
+endif()
+
+function(mold_instantiate_templates SOURCE TARGET)
+  set(PATH ${CMAKE_BINARY_DIR}/${SOURCE}.${TARGET}.cc)
+  if(NOT EXISTS ${PATH})
+    file(WRITE ${PATH} "#define MOLD_${TARGET} 1
+#define MOLD_TARGET ${TARGET}
+#include \"${CMAKE_SOURCE_DIR}/${SOURCE}\"
+")
+  endif()
+  target_sources(mold PRIVATE ${PATH})
+endfunction()
+
+foreach (SOURCE IN LISTS MOLD_ELF_TEMPLATE_FILES)
+  foreach(TARGET IN LISTS MOLD_ELF_TARGETS)
+    mold_instantiate_templates(${SOURCE} ${TARGET})
+  endforeach()
+endforeach()
+
+# Add other non-template source files.
+target_sources(mold PRIVATE
+  git-hash.cc
+  lib/compress.cc
+  lib/crc32.cc
+  lib/demangle.cc
+  lib/filepath.cc
+  lib/glob.cc
+  lib/hyperloglog.cc
+  lib/mimalloc.cc
+  lib/multi-glob.cc
+  lib/perf.cc
+  lib/random.cc
+  lib/tar.cc
+  src/arch-arc.cc
+  src/arch-arm32.cc
+  src/arch-i386.cc
+  src/arch-m68k.cc
+  src/arch-ppc32.cc
+  src/arch-ppc64v1.cc
+  src/arch-ppc64v2.cc
+  src/arch-s390x.cc
+  src/arch-sparc64.cc
+  src/arch-x86-64.cc
+  src/config.cc
+  src/elf.cc
+  third-party/rust-demangle/rust-demangle.c
+  )
+
+if(WIN32)
+  target_sources(mold PRIVATE
+    lib/jobs-win32.cc
+    lib/mapped-file-win32.cc
+    lib/signal-win32.cc
+    )
+else()
+  target_sources(mold PRIVATE
+    lib/jobs-unix.cc
+    lib/mapped-file-unix.cc
+    lib/signal-unix.cc
+    )
+endif()
+
+include(CTest)
+
+if(BUILD_TESTING)
+  # Create the ld symlinks required for testing
+  if(NOT WIN32)
+    add_custom_command(
+      TARGET mold POST_BUILD
+      COMMAND ${CMAKE_COMMAND} -E create_symlink mold ld
+      BYPRODUCTS ld
+      WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
+      VERBATIM)
+  endif()
+
+  if(${UNIX})
+    add_subdirectory(test)
+  endif()
+endif()
+
+if(NOT CMAKE_SKIP_INSTALL_RULES)
+  install(TARGETS mold RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
+  install(FILES docs/mold.1 DESTINATION ${CMAKE_INSTALL_MANDIR}/man1/)
+  install(FILES LICENSE DESTINATION ${CMAKE_INSTALL_DOCDIR})
+
+  function(mold_install_relative_symlink OLD NEW)
+    install(CODE "
+      get_filename_component(PREFIX_ABS \${CMAKE_INSTALL_PREFIX}/ ABSOLUTE)
+      get_filename_component(OLD_ABS ${OLD} ABSOLUTE BASE_DIR \${PREFIX_ABS})
+      get_filename_component(NEW_ABS ${NEW} ABSOLUTE BASE_DIR \${PREFIX_ABS})
+      get_filename_component(NEW_DIR \${NEW_ABS} DIRECTORY)
+      file(RELATIVE_PATH OLD_REL \${NEW_DIR} \${OLD_ABS})
+      message(STATUS \"Installing symlink: \$ENV{DESTDIR}\${NEW_ABS} -> \${OLD_REL}\")
+      file(MAKE_DIRECTORY \$ENV{DESTDIR}\${NEW_DIR})
+      file(CREATE_LINK \${OLD_REL} \$ENV{DESTDIR}\${NEW_ABS} SYMBOLIC)")
+  endfunction()
+
+  mold_install_relative_symlink(${CMAKE_INSTALL_BINDIR}/mold${CMAKE_EXECUTABLE_SUFFIX}
+    ${CMAKE_INSTALL_LIBEXECDIR}/mold/ld${CMAKE_EXECUTABLE_SUFFIX})
+  mold_install_relative_symlink(${CMAKE_INSTALL_BINDIR}/mold${CMAKE_EXECUTABLE_SUFFIX}
+    ${CMAKE_INSTALL_BINDIR}/ld.mold${CMAKE_EXECUTABLE_SUFFIX})
+  mold_install_relative_symlink(${CMAKE_INSTALL_MANDIR}/man1/mold.1
+    ${CMAKE_INSTALL_MANDIR}/man1/ld.mold.1)
+endif()

CONTRIBUTING.md

@@ -1,54 +0,0 @@
-# Contributing to mold
-
-Summary: When you send a patch, you must agree that the patch is
-released under the dual license of the GNU AGPLv3 and the MIT license.
-As an indication of agreement, please add a `Signed-off-by` line to
-your commit message using `git commit -s`.
-
-Here is the long official version:
-
-mold is currently licensed under the GNU AGPLv3, but I want to relicense
-it under a more liberal license in the future. To make this possible,
-I want all patches to be submitted under the dual license of the GNU
-AGPLv3 and the MIT license (the "Dual License") from the beginning.
-
-To make it clear that contributors have agreed to release their
-patches under the Dual License, I introduced a sign-off procedure just
-like the Linux kernel project did. The sign-off is a simple line at
-the end of the git commit message, which certifies that you wrote it
-or otherwise have the right to pass it on as an open-source patch. The
-rules are pretty simple: if you can certify the below:
-
-> ## Developer's Certificate of Origin
->
-> By making a contribution to this project, I certify that:
->
->  1. The contribution was created in whole or in part by me and I
->     have the right to submit it under the Dual License; or
->
->  2. The contribution is based upon previous work that, to the best
->     of my knowledge, is covered under the the Dual License and I
->     have the right under the Dual License to submit that work with
->     modifications, whether created in whole or in part by me, under
->     the Dual License (unless I am permitted to submit under
->     different licenses), as indicated in the file; or
->
->  3. The contribution was provided directly to me by some other
->     person who certified (1), (2) or (3) and I have not modified it.
->
->  4. I understand and agree that this project and the contribution
->     are public and that a record of the contribution (including all
->     personal information I submit with it, including my sign-off) is
->     maintained indefinitely and may be redistributed consistent with
->     this project or the open source license(s) involved.
-
-then you just add a line saying:
-
-```
-Signed-off-by: Random J Developer <[email protected]>
-```
-
-using your real name (sorry, no pseudonyms or anonymous contributions.)
-This will be done for you automatically if you use ``git commit -s``.
-Reverts should also include "Signed-off-by". ``git revert -s`` does that
-for you.

LICENSE

@@ -1,671 +1,21 @@
-Note: I'm looking for a sponsor who wants to purchase the copyright
-of this work and relicense it under a more liberal license such as
-the MIT license. For now, mold is released under the GNU AGPL v3.
-
-If you just want to use mold but do not want to use AGPL software
-in your organization, a commercial license is available. Please
-contact [email protected].
-
---
-
-                    GNU AFFERO GENERAL PUBLIC LICENSE
-                       Version 3, 19 November 2007
-
- Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-                            Preamble
-
-  The GNU Affero General Public License is a free, copyleft license for
-software and other kinds of works, specifically designed to ensure
-cooperation with the community in the case of network server software.
-
-  The licenses for most software and other practical works are designed
-to take away your freedom to share and change the works.  By contrast,
-our General Public Licenses are intended to guarantee your freedom to
-share and change all versions of a program--to make sure it remains free
-software for all its users.
-
-  When we speak of free software, we are referring to freedom, not
-price.  Our General Public Licenses are designed to make sure that you
-have the freedom to distribute copies of free software (and charge for
-them if you wish), that you receive source code or can get it if you
-want it, that you can change the software or use pieces of it in new
-free programs, and that you know you can do these things.
-
-  Developers that use our General Public Licenses protect your rights
-with two steps: (1) assert copyright on the software, and (2) offer
-you this License which gives you legal permission to copy, distribute
-and/or modify the software.
-
-  A secondary benefit of defending all users' freedom is that
-improvements made in alternate versions of the program, if they
-receive widespread use, become available for other developers to
-incorporate.  Many developers of free software are heartened and
-encouraged by the resulting cooperation.  However, in the case of
-software used on network servers, this result may fail to come about.
-The GNU General Public License permits making a modified version and
-letting the public access it on a server without ever releasing its
-source code to the public.
-
-  The GNU Affero General Public License is designed specifically to
-ensure that, in such cases, the modified source code becomes available
-to the community.  It requires the operator of a network server to
-provide the source code of the modified version running there to the
-users of that server.  Therefore, public use of a modified version, on
-a publicly accessible server, gives the public access to the source
-code of the modified version.
-
-  An older license, called the Affero General Public License and
-published by Affero, was designed to accomplish similar goals.  This is
-a different license, not a version of the Affero GPL, but Affero has
-released a new version of the Affero GPL which permits relicensing under
-this license.
-
-  The precise terms and conditions for copying, distribution and
-modification follow.
-
-                       TERMS AND CONDITIONS
-
-  0. Definitions.
-
-  "This License" refers to version 3 of the GNU Affero General Public License.
-
-  "Copyright" also means copyright-like laws that apply to other kinds of
-works, such as semiconductor masks.
-
-  "The Program" refers to any copyrightable work licensed under this
-License.  Each licensee is addressed as "you".  "Licensees" and
-"recipients" may be individuals or organizations.
-
-  To "modify" a work means to copy from or adapt all or part of the work
-in a fashion requiring copyright permission, other than the making of an
-exact copy.  The resulting work is called a "modified version" of the
-earlier work or a work "based on" the earlier work.
-
-  A "covered work" means either the unmodified Program or a work based
-on the Program.
-
-  To "propagate" a work means to do anything with it that, without
-permission, would make you directly or secondarily liable for
-infringement under applicable copyright law, except executing it on a
-computer or modifying a private copy.  Propagation includes copying,
-distribution (with or without modification), making available to the
-public, and in some countries other activities as well.
-
-  To "convey" a work means any kind of propagation that enables other
-parties to make or receive copies.  Mere interaction with a user through
-a computer network, with no transfer of a copy, is not conveying.
-
-  An interactive user interface displays "Appropriate Legal Notices"
-to the extent that it includes a convenient and prominently visible
-feature that (1) displays an appropriate copyright notice, and (2)
-tells the user that there is no warranty for the work (except to the
-extent that warranties are provided), that licensees may convey the
-work under this License, and how to view a copy of this License.  If
-the interface presents a list of user commands or options, such as a
-menu, a prominent item in the list meets this criterion.
-
-  1. Source Code.
-
-  The "source code" for a work means the preferred form of the work
-for making modifications to it.  "Object code" means any non-source
-form of a work.
-
-  A "Standard Interface" means an interface that either is an official
-standard defined by a recognized standards body, or, in the case of
-interfaces specified for a particular programming language, one that
-is widely used among developers working in that language.
-
-  The "System Libraries" of an executable work include anything, other
-than the work as a whole, that (a) is included in the normal form of
-packaging a Major Component, but which is not part of that Major
-Component, and (b) serves only to enable use of the work with that
-Major Component, or to implement a Standard Interface for which an
-implementation is available to the public in source code form.  A
-"Major Component", in this context, means a major essential component
-(kernel, window system, and so on) of the specific operating system
-(if any) on which the executable work runs, or a compiler used to
-produce the work, or an object code interpreter used to run it.
-
-  The "Corresponding Source" for a work in object code form means all
-the source code needed to generate, install, and (for an executable
-work) run the object code and to modify the work, including scripts to
-control those activities.  However, it does not include the work's
-System Libraries, or general-purpose tools or generally available free
-programs which are used unmodified in performing those activities but
-which are not part of the work.  For example, Corresponding Source
-includes interface definition files associated with source files for
-the work, and the source code for shared libraries and dynamically
-linked subprograms that the work is specifically designed to require,
-such as by intimate data communication or control flow between those
-subprograms and other parts of the work.
-
-  The Corresponding Source need not include anything that users
-can regenerate automatically from other parts of the Corresponding
-Source.
-
-  The Corresponding Source for a work in source code form is that
-same work.
-
-  2. Basic Permissions.
-
-  All rights granted under this License are granted for the term of
-copyright on the Program, and are irrevocable provided the stated
-conditions are met.  This License explicitly affirms your unlimited
-permission to run the unmodified Program.  The output from running a
-covered work is covered by this License only if the output, given its
-content, constitutes a covered work.  This License acknowledges your
-rights of fair use or other equivalent, as provided by copyright law.
-
-  You may make, run and propagate covered works that you do not
-convey, without conditions so long as your license otherwise remains
-in force.  You may convey covered works to others for the sole purpose
-of having them make modifications exclusively for you, or provide you
-with facilities for running those works, provided that you comply with
-the terms of this License in conveying all material for which you do
-not control copyright.  Those thus making or running the covered works
-for you must do so exclusively on your behalf, under your direction
-and control, on terms that prohibit them from making any copies of
-your copyrighted material outside their relationship with you.
-
-  Conveying under any other circumstances is permitted solely under
-the conditions stated below.  Sublicensing is not allowed; section 10
-makes it unnecessary.
-
-  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
-
-  No covered work shall be deemed part of an effective technological
-measure under any applicable law fulfilling obligations under article
-11 of the WIPO copyright treaty adopted on 20 December 1996, or
-similar laws prohibiting or restricting circumvention of such
-measures.
-
-  When you convey a covered work, you waive any legal power to forbid
-circumvention of technological measures to the extent such circumvention
-is effected by exercising rights under this License with respect to
-the covered work, and you disclaim any intention to limit operation or
-modification of the work as a means of enforcing, against the work's
-users, your or third parties' legal rights to forbid circumvention of
-technological measures.
-
-  4. Conveying Verbatim Copies.
-
-  You may convey verbatim copies of the Program's source code as you
-receive it, in any medium, provided that you conspicuously and
-appropriately publish on each copy an appropriate copyright notice;
-keep intact all notices stating that this License and any
-non-permissive terms added in accord with section 7 apply to the code;
-keep intact all notices of the absence of any warranty; and give all
-recipients a copy of this License along with the Program.
-
-  You may charge any price or no price for each copy that you convey,
-and you may offer support or warranty protection for a fee.
-
-  5. Conveying Modified Source Versions.
-
-  You may convey a work based on the Program, or the modifications to
-produce it from the Program, in the form of source code under the
-terms of section 4, provided that you also meet all of these conditions:
-
-    a) The work must carry prominent notices stating that you modified
-    it, and giving a relevant date.
-
-    b) The work must carry prominent notices stating that it is
-    released under this License and any conditions added under section
-    7.  This requirement modifies the requirement in section 4 to
-    "keep intact all notices".
-
-    c) You must license the entire work, as a whole, under this
-    License to anyone who comes into possession of a copy.  This
-    License will therefore apply, along with any applicable section 7
-    additional terms, to the whole of the work, and all its parts,
-    regardless of how they are packaged.  This License gives no
-    permission to license the work in any other way, but it does not
-    invalidate such permission if you have separately received it.
-
-    d) If the work has interactive user interfaces, each must display
-    Appropriate Legal Notices; however, if the Program has interactive
-    interfaces that do not display Appropriate Legal Notices, your
-    work need not make them do so.
-
-  A compilation of a covered work with other separate and independent
-works, which are not by their nature extensions of the covered work,
-and which are not combined with it such as to form a larger program,
-in or on a volume of a storage or distribution medium, is called an
-"aggregate" if the compilation and its resulting copyright are not
-used to limit the access or legal rights of the compilation's users
-beyond what the individual works permit.  Inclusion of a covered work
-in an aggregate does not cause this License to apply to the other
-parts of the aggregate.
-
-  6. Conveying Non-Source Forms.
-
-  You may convey a covered work in object code form under the terms
-of sections 4 and 5, provided that you also convey the
-machine-readable Corresponding Source under the terms of this License,
-in one of these ways:
-
-    a) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by the
-    Corresponding Source fixed on a durable physical medium
-    customarily used for software interchange.
-
-    b) Convey the object code in, or embodied in, a physical product
-    (including a physical distribution medium), accompanied by a
-    written offer, valid for at least three years and valid for as
-    long as you offer spare parts or customer support for that product
-    model, to give anyone who possesses the object code either (1) a
-    copy of the Corresponding Source for all the software in the
-    product that is covered by this License, on a durable physical
-    medium customarily used for software interchange, for a price no
-    more than your reasonable cost of physically performing this
-    conveying of source, or (2) access to copy the
-    Corresponding Source from a network server at no charge.
-
-    c) Convey individual copies of the object code with a copy of the
-    written offer to provide the Corresponding Source.  This
-    alternative is allowed only occasionally and noncommercially, and
-    only if you received the object code with such an offer, in accord
-    with subsection 6b.
-
-    d) Convey the object code by offering access from a designated
-    place (gratis or for a charge), and offer equivalent access to the
-    Corresponding Source in the same way through the same place at no
-    further charge.  You need not require recipients to copy the
-    Corresponding Source along with the object code.  If the place to
-    copy the object code is a network server, the Corresponding Source
-    may be on a different server (operated by you or a third party)
-    that supports equivalent copying facilities, provided you maintain
-    clear directions next to the object code saying where to find the
-    Corresponding Source.  Regardless of what server hosts the
-    Corresponding Source, you remain obligated to ensure that it is
-    available for as long as needed to satisfy these requirements.
-
-    e) Convey the object code using peer-to-peer transmission, provided
-    you inform other peers where the object code and Corresponding
-    Source of the work are being offered to the general public at no
-    charge under subsection 6d.
-
-  A separable portion of the object code, whose source code is excluded
-from the Corresponding Source as a System Library, need not be
-included in conveying the object code work.
-
-  A "User Product" is either (1) a "consumer product", which means any
-tangible personal property which is normally used for personal, family,
-or household purposes, or (2) anything designed or sold for incorporation
-into a dwelling.  In determining whether a product is a consumer product,
-doubtful cases shall be resolved in favor of coverage.  For a particular
-product received by a particular user, "normally used" refers to a
-typical or common use of that class of product, regardless of the status
-of the particular user or of the way in which the particular user
-actually uses, or expects or is expected to use, the product.  A product
-is a consumer product regardless of whether the product has substantial
-commercial, industrial or non-consumer uses, unless such uses represent
-the only significant mode of use of the product.
-
-  "Installation Information" for a User Product means any methods,
-procedures, authorization keys, or other information required to install
-and execute modified versions of a covered work in that User Product from
-a modified version of its Corresponding Source.  The information must
-suffice to ensure that the continued functioning of the modified object
-code is in no case prevented or interfered with solely because
-modification has been made.
-
-  If you convey an object code work under this section in, or with, or
-specifically for use in, a User Product, and the conveying occurs as
-part of a transaction in which the right of possession and use of the
-User Product is transferred to the recipient in perpetuity or for a
-fixed term (regardless of how the transaction is characterized), the
-Corresponding Source conveyed under this section must be accompanied
-by the Installation Information.  But this requirement does not apply
-if neither you nor any third party retains the ability to install
-modified object code on the User Product (for example, the work has
-been installed in ROM).
-
-  The requirement to provide Installation Information does not include a
-requirement to continue to provide support service, warranty, or updates
-for a work that has been modified or installed by the recipient, or for
-the User Product in which it has been modified or installed.  Access to a
-network may be denied when the modification itself materially and
-adversely affects the operation of the network or violates the rules and
-protocols for communication across the network.
-
-  Corresponding Source conveyed, and Installation Information provided,
-in accord with this section must be in a format that is publicly
-documented (and with an implementation available to the public in
-source code form), and must require no special password or key for
-unpacking, reading or copying.
-
-  7. Additional Terms.
-
-  "Additional permissions" are terms that supplement the terms of this
-License by making exceptions from one or more of its conditions.
-Additional permissions that are applicable to the entire Program shall
-be treated as though they were included in this License, to the extent
-that they are valid under applicable law.  If additional permissions
-apply only to part of the Program, that part may be used separately
-under those permissions, but the entire Program remains governed by
-this License without regard to the additional permissions.
-
-  When you convey a copy of a covered work, you may at your option
-remove any additional permissions from that copy, or from any part of
-it.  (Additional permissions may be written to require their own
-removal in certain cases when you modify the work.)  You may place
-additional permissions on material, added by you to a covered work,
-for which you have or can give appropriate copyright permission.
-
-  Notwithstanding any other provision of this License, for material you
-add to a covered work, you may (if authorized by the copyright holders of
-that material) supplement the terms of this License with terms:
-
-    a) Disclaiming warranty or limiting liability differently from the
-    terms of sections 15 and 16 of this License; or
-
-    b) Requiring preservation of specified reasonable legal notices or
-    author attributions in that material or in the Appropriate Legal
-    Notices displayed by works containing it; or
-
-    c) Prohibiting misrepresentation of the origin of that material, or
-    requiring that modified versions of such material be marked in
-    reasonable ways as different from the original version; or
-
-    d) Limiting the use for publicity purposes of names of licensors or
-    authors of the material; or
-
-    e) Declining to grant rights under trademark law for use of some
-    trade names, trademarks, or service marks; or
-
-    f) Requiring indemnification of licensors and authors of that
-    material by anyone who conveys the material (or modified versions of
-    it) with contractual assumptions of liability to the recipient, for
-    any liability that these contractual assumptions directly impose on
-    those licensors and authors.
-
-  All other non-permissive additional terms are considered "further
-restrictions" within the meaning of section 10.  If the Program as you
-received it, or any part of it, contains a notice stating that it is
-governed by this License along with a term that is a further
-restriction, you may remove that term.  If a license document contains
-a further restriction but permits relicensing or conveying under this
-License, you may add to a covered work material governed by the terms
-of that license document, provided that the further restriction does
-not survive such relicensing or conveying.
-
-  If you add terms to a covered work in accord with this section, you
-must place, in the relevant source files, a statement of the
-additional terms that apply to those files, or a notice indicating
-where to find the applicable terms.
-
-  Additional terms, permissive or non-permissive, may be stated in the
-form of a separately written license, or stated as exceptions;
-the above requirements apply either way.
-
-  8. Termination.
-
-  You may not propagate or modify a covered work except as expressly
-provided under this License.  Any attempt otherwise to propagate or
-modify it is void, and will automatically terminate your rights under
-this License (including any patent licenses granted under the third
-paragraph of section 11).
-
-  However, if you cease all violation of this License, then your
-license from a particular copyright holder is reinstated (a)
-provisionally, unless and until the copyright holder explicitly and
-finally terminates your license, and (b) permanently, if the copyright
-holder fails to notify you of the violation by some reasonable means
-prior to 60 days after the cessation.
-
-  Moreover, your license from a particular copyright holder is
-reinstated permanently if the copyright holder notifies you of the
-violation by some reasonable means, this is the first time you have
-received notice of violation of this License (for any work) from that
-copyright holder, and you cure the violation prior to 30 days after
-your receipt of the notice.
-
-  Termination of your rights under this section does not terminate the
-licenses of parties who have received copies or rights from you under
-this License.  If your rights have been terminated and not permanently
-reinstated, you do not qualify to receive new licenses for the same
-material under section 10.
-
-  9. Acceptance Not Required for Having Copies.
-
-  You are not required to accept this License in order to receive or
-run a copy of the Program.  Ancillary propagation of a covered work
-occurring solely as a consequence of using peer-to-peer transmission
-to receive a copy likewise does not require acceptance.  However,
-nothing other than this License grants you permission to propagate or
-modify any covered work.  These actions infringe copyright if you do
-not accept this License.  Therefore, by modifying or propagating a
-covered work, you indicate your acceptance of this License to do so.
-
-  10. Automatic Licensing of Downstream Recipients.
-
-  Each time you convey a covered work, the recipient automatically
-receives a license from the original licensors, to run, modify and
-propagate that work, subject to this License.  You are not responsible
-for enforcing compliance by third parties with this License.
-
-  An "entity transaction" is a transaction transferring control of an
-organization, or substantially all assets of one, or subdividing an
-organization, or merging organizations.  If propagation of a covered
-work results from an entity transaction, each party to that
-transaction who receives a copy of the work also receives whatever
-licenses to the work the party's predecessor in interest had or could
-give under the previous paragraph, plus a right to possession of the
-Corresponding Source of the work from the predecessor in interest, if
-the predecessor has it or can get it with reasonable efforts.
-
-  You may not impose any further restrictions on the exercise of the
-rights granted or affirmed under this License.  For example, you may
-not impose a license fee, royalty, or other charge for exercise of
-rights granted under this License, and you may not initiate litigation
-(including a cross-claim or counterclaim in a lawsuit) alleging that
-any patent claim is infringed by making, using, selling, offering for
-sale, or importing the Program or any portion of it.
-
-  11. Patents.
-
-  A "contributor" is a copyright holder who authorizes use under this
-License of the Program or a work on which the Program is based.  The
-work thus licensed is called the contributor's "contributor version".
-
-  A contributor's "essential patent claims" are all patent claims
-owned or controlled by the contributor, whether already acquired or
-hereafter acquired, that would be infringed by some manner, permitted
-by this License, of making, using, or selling its contributor version,
-but do not include claims that would be infringed only as a
-consequence of further modification of the contributor version.  For
-purposes of this definition, "control" includes the right to grant
-patent sublicenses in a manner consistent with the requirements of
-this License.
-
-  Each contributor grants you a non-exclusive, worldwide, royalty-free
-patent license under the contributor's essential patent claims, to
-make, use, sell, offer for sale, import and otherwise run, modify and
-propagate the contents of its contributor version.
-
-  In the following three paragraphs, a "patent license" is any express
-agreement or commitment, however denominated, not to enforce a patent
-(such as an express permission to practice a patent or covenant not to
-sue for patent infringement).  To "grant" such a patent license to a
-party means to make such an agreement or commitment not to enforce a
-patent against the party.
-
-  If you convey a covered work, knowingly relying on a patent license,
-and the Corresponding Source of the work is not available for anyone
-to copy, free of charge and under the terms of this License, through a
-publicly available network server or other readily accessible means,
-then you must either (1) cause the Corresponding Source to be so
-available, or (2) arrange to deprive yourself of the benefit of the
-patent license for this particular work, or (3) arrange, in a manner
-consistent with the requirements of this License, to extend the patent
-license to downstream recipients.  "Knowingly relying" means you have
-actual knowledge that, but for the patent license, your conveying the
-covered work in a country, or your recipient's use of the covered work
-in a country, would infringe one or more identifiable patents in that
-country that you have reason to believe are valid.
-
-  If, pursuant to or in connection with a single transaction or
-arrangement, you convey, or propagate by procuring conveyance of, a
-covered work, and grant a patent license to some of the parties
-receiving the covered work authorizing them to use, propagate, modify
-or convey a specific copy of the covered work, then the patent license
-you grant is automatically extended to all recipients of the covered
-work and works based on it.
-
-  A patent license is "discriminatory" if it does not include within
-the scope of its coverage, prohibits the exercise of, or is
-conditioned on the non-exercise of one or more of the rights that are
-specifically granted under this License.  You may not convey a covered
-work if you are a party to an arrangement with a third party that is
-in the business of distributing software, under which you make payment
-to the third party based on the extent of your activity of conveying
-the work, and under which the third party grants, to any of the
-parties who would receive the covered work from you, a discriminatory
-patent license (a) in connection with copies of the covered work
-conveyed by you (or copies made from those copies), or (b) primarily
-for and in connection with specific products or compilations that
-contain the covered work, unless you entered into that arrangement,
-or that patent license was granted, prior to 28 March 2007.
-
-  Nothing in this License shall be construed as excluding or limiting
-any implied license or other defenses to infringement that may
-otherwise be available to you under applicable patent law.
-
-  12. No Surrender of Others' Freedom.
-
-  If conditions are imposed on you (whether by court order, agreement or
-otherwise) that contradict the conditions of this License, they do not
-excuse you from the conditions of this License.  If you cannot convey a
-covered work so as to satisfy simultaneously your obligations under this
-License and any other pertinent obligations, then as a consequence you may
-not convey it at all.  For example, if you agree to terms that obligate you
-to collect a royalty for further conveying from those to whom you convey
-the Program, the only way you could satisfy both those terms and this
-License would be to refrain entirely from conveying the Program.
-
-  13. Remote Network Interaction; Use with the GNU General Public License.
-
-  Notwithstanding any other provision of this License, if you modify the
-Program, your modified version must prominently offer all users
-interacting with it remotely through a computer network (if your version
-supports such interaction) an opportunity to receive the Corresponding
-Source of your version by providing access to the Corresponding Source
-from a network server at no charge, through some standard or customary
-means of facilitating copying of software.  This Corresponding Source
-shall include the Corresponding Source for any work covered by version 3
-of the GNU General Public License that is incorporated pursuant to the
-following paragraph.
-
-  Notwithstanding any other provision of this License, you have
-permission to link or combine any covered work with a work licensed
-under version 3 of the GNU General Public License into a single
-combined work, and to convey the resulting work.  The terms of this
-License will continue to apply to the part which is the covered work,
-but the work with which it is combined will remain governed by version
-3 of the GNU General Public License.
-
-  14. Revised Versions of this License.
-
-  The Free Software Foundation may publish revised and/or new versions of
-the GNU Affero General Public License from time to time.  Such new versions
-will be similar in spirit to the present version, but may differ in detail to
-address new problems or concerns.
-
-  Each version is given a distinguishing version number.  If the
-Program specifies that a certain numbered version of the GNU Affero General
-Public License "or any later version" applies to it, you have the
-option of following the terms and conditions either of that numbered
-version or of any later version published by the Free Software
-Foundation.  If the Program does not specify a version number of the
-GNU Affero General Public License, you may choose any version ever published
-by the Free Software Foundation.
-
-  If the Program specifies that a proxy can decide which future
-versions of the GNU Affero General Public License can be used, that proxy's
-public statement of acceptance of a version permanently authorizes you
-to choose that version for the Program.
-
-  Later license versions may give you additional or different
-permissions.  However, no additional obligations are imposed on any
-author or copyright holder as a result of your choosing to follow a
-later version.
-
-  15. Disclaimer of Warranty.
-
-  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
-APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
-HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
-OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
-THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
-PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
-IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
-ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
-
-  16. Limitation of Liability.
-
-  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
-WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
-THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
-GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
-USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
-DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
-PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
-EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
-SUCH DAMAGES.
-
-  17. Interpretation of Sections 15 and 16.
-
-  If the disclaimer of warranty and limitation of liability provided
-above cannot be given local legal effect according to their terms,
-reviewing courts shall apply local law that most closely approximates
-an absolute waiver of all civil liability in connection with the
-Program, unless a warranty or assumption of liability accompanies a
-copy of the Program in return for a fee.
-
-                     END OF TERMS AND CONDITIONS
-
-            How to Apply These Terms to Your New Programs
-
-  If you develop a new program, and you want it to be of the greatest
-possible use to the public, the best way to achieve this is to make it
-free software which everyone can redistribute and change under these terms.
-
-  To do so, attach the following notices to the program.  It is safest
-to attach them to the start of each source file to most effectively
-state the exclusion of warranty; and each file should have at least
-the "copyright" line and a pointer to where the full notice is found.
-
-    <one line to give the program's name and a brief idea of what it does.>
-    Copyright (C) <year>  <name of author>
-
-    This program is free software: you can redistribute it and/or modify
-    it under the terms of the GNU Affero General Public License as published
-    by the Free Software Foundation, either version 3 of the License, or
-    (at your option) any later version.
-
-    This program is distributed in the hope that it will be useful,
-    but WITHOUT ANY WARRANTY; without even the implied warranty of
-    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-    GNU Affero General Public License for more details.
-
-    You should have received a copy of the GNU Affero General Public License
-    along with this program.  If not, see <https://www.gnu.org/licenses/>.
-
-Also add information on how to contact you by electronic and paper mail.
-
-  If your software can interact with users remotely through a computer
-network, you should also make sure that it provides a way for users to
-get its source.  For example, if your program is a web application, its
-interface could display a "Source" link that leads users to an archive
-of the code.  There are many ways you could offer source, and different
-solutions will be better for different programs; see section 13 for the
-specific requirements.
-
-  You should also get your employer (if you work as a programmer) or school,
-if any, to sign a "copyright disclaimer" for the program, if necessary.
-For more information on this, and how to apply and follow the GNU AGPL, see
-<https://www.gnu.org/licenses/>.
+MIT License
+
+Copyright (c) 2023 Rui Ueyama
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

Makefile

@@ -1,197 +0,0 @@
-PREFIX = /usr/local
-BINDIR = $(PREFIX)/bin
-LIBDIR = $(PREFIX)/lib
-LIBEXECDIR = $(PREFIX)/libexec
-MANDIR = $(PREFIX)/share/man
-
-INSTALL = install
-INSTALL_PROGRAM = $(INSTALL)
-INSTALL_DATA = $(INSTALL) -m 644
-
-D = $(DESTDIR)
-
-# CXX defaults to `g++`. Rewrite it with a vendor-neutral compiler
-# name `c++`.
-ifeq ($(origin CXX), default)
-  CXX = c++
-endif
-
-# If you want to keep symbols in the installed binary, run make with
-# `STRIP=true` to run /bin/true instead of the strip command.
-STRIP = strip
-
-SRCS=$(wildcard *.cc elf/*.cc macho/*.cc)
-HEADERS=$(wildcard *.h elf/*.h macho/*.h)
-OBJS=$(SRCS:%.cc=out/%.o)
-
-OS = $(shell uname -s)
-ARCH = $(shell uname -m)
-
-IS_ANDROID = 0
-ifneq ($(findstring -android,$(shell $(CC) -dumpmachine)),)
-  IS_ANDROID = 1
-endif
-
-# If you want to compile mold for debugging, invoke make as
-# `make CXXFLAGS=-g`.
-CFLAGS = -O2
-CXXFLAGS = -O2
-
-MOLD_CXXFLAGS = -std=c++20 -fno-exceptions -fno-unwind-tables \
-                -fno-asynchronous-unwind-tables -DMOLD_VERSION=\"1.0.3\" \
-                -DLIBDIR="\"$(LIBDIR)\""
-
-MOLD_LDFLAGS = -pthread -lz -lm
-
-GIT_HASH = $(shell [ -d .git ] && git rev-parse HEAD)
-ifneq ($(GIT_HASH),)
-  MOLD_CXXFLAGS += -DGIT_HASH=\"$(GIT_HASH)\"
-endif
-
-LTO = 0
-
-ifeq ($(LTO), 1)
-  CXXFLAGS += -flto -O3
-  LDFLAGS  += -flto
-endif
-
-# By default, we want to use mimalloc as a memory allocator. mimalloc
-# is disabled on macOS and Android because it didn't work on those hosts.
-USE_MIMALLOC = 1
-ifeq ($(OS), Darwin)
-  USE_MIMALLOC = 0
-else ifeq ($(IS_ANDROID), 1)
-  USE_MIMALLOC = 0
-endif
-
-ifeq ($(USE_MIMALLOC), 1)
-  ifdef SYSTEM_MIMALLOC
-    MOLD_LDFLAGS += -lmimalloc
-  else
-    MIMALLOC_LIB = out/mimalloc/libmimalloc.a
-    MOLD_CXXFLAGS += -Ithird-party/mimalloc/include
-    MOLD_LDFLAGS += -Wl,-whole-archive $(MIMALLOC_LIB) -Wl,-no-whole-archive
-  endif
-endif
-
-ifdef SYSTEM_TBB
-  MOLD_LDFLAGS += -ltbb
-else
-  TBB_LIB = out/tbb/libs/libtbb.a
-  MOLD_LDFLAGS += $(TBB_LIB)
-  MOLD_CXXFLAGS += -Ithird-party/tbb/include
-endif
-
-ifdef SYSTEM_XXHASH
-  MOLD_LDFLAGS += -lxxhash
-else
-  XXHASH_LIB = third-party/xxhash/libxxhash.a
-  MOLD_LDFLAGS += $(XXHASH_LIB)
-  MOLD_CXXFLAGS += -Ithird-party/xxhash
-endif
-
-ifeq ($(OS), Linux)
-  ifeq ($(IS_ANDROID), 0)
-    # glibc before 2.17 need librt for clock_gettime
-    MOLD_LDFLAGS += -lrt
-  endif
-endif
-
-# Use pkg-config to know where libcrypto resides.
-ifneq ($(OS), Darwin)
-  MOLD_CXXFLAGS += $(shell pkg-config --cflags-only-I openssl)
-  MOLD_LDFLAGS += $(shell pkg-config --libs-only-L openssl) -lcrypto
-endif
-
-# '-latomic' flag is needed building on riscv64 system
-# RV32 system doesn't tested yet
-# seems like '-atomic' would be better but not working.
-ifeq ($(ARCH), riscv64)
-  MOLD_LDFLAGS += -latomic
-endif
-
-# -Wc++11-narrowing is a fatal error on Android, so disable it.
-ifeq ($(IS_ANDROID), 1)
-  MOLD_CXXFLAGS += -Wno-c++11-narrowing
-endif
-
-ifeq ($(OS), Linux)
-  MOLD_WRAPPER_LDFLAGS = -Wl,-push-state -Wl,-no-as-needed -ldl -Wl,-pop-state
-endif
-
-all: mold mold-wrapper.so
-
-mold: $(OBJS) $(MIMALLOC_LIB) $(TBB_LIB) $(XXHASH_LIB)
-	$(CXX) $(OBJS) -o $@ $(MOLD_LDFLAGS) $(LDFLAGS)
-	ln -sf mold ld
-	ln -sf mold ld64.mold
-
-mold-wrapper.so: elf/mold-wrapper.c Makefile
-	$(CC) $(CFLAGS) -fPIC -shared -o $@ $< $(MOLD_WRAPPER_LDFLAGS) $(LDFLAGS)
-
-out/%.o: %.cc $(HEADERS) Makefile out/elf/.keep out/macho/.keep
-	$(CXX) $(MOLD_CXXFLAGS) $(CXXFLAGS) -c -o $@ $<
-
-out/elf/.keep:
-	mkdir -p out/elf
-	touch $@
-
-out/macho/.keep:
-	mkdir -p out/macho
-	touch $@
-
-$(MIMALLOC_LIB):
-	mkdir -p out/mimalloc
-	(cd out/mimalloc; CFLAGS=-DMI_USE_ENVIRON=0 cmake -G'Unix Makefiles' ../../third-party/mimalloc)
-	$(MAKE) -C out/mimalloc mimalloc-static
-
-$(TBB_LIB):
-	mkdir -p out/tbb
-	(cd out/tbb; cmake -G'Unix Makefiles' -DBUILD_SHARED_LIBS=OFF -DTBB_TEST=OFF -DCMAKE_CXX_FLAGS=-D__TBB_DYNAMIC_LOAD_ENABLED=0 -DTBB_STRICT=OFF ../../third-party/tbb)
-	$(MAKE) -C out/tbb tbb
-	(cd out/tbb; ln -sf *_relwithdebinfo libs)
-
-$(XXHASH_LIB):
-	$(MAKE) -C third-party/xxhash libxxhash.a
-
-test tests check: all
-ifeq ($(OS), Darwin)
-	$(MAKE) -C test -f Makefile.darwin --no-print-directory
-else
-	$(MAKE) -C test -f Makefile.linux --no-print-directory --output-sync
-endif
-
-	@if test -t 1; then \
-	  printf '\e[32mPassed all tests\e[0m\n'; \
-	else \
-	  echo 'Passed all tests'; \
-	fi
-
-install: all
-	$(INSTALL) -d $D$(BINDIR)
-	$(INSTALL_PROGRAM) mold $D$(BINDIR)
-	$(STRIP) $D$(BINDIR)/mold
-
-	$(INSTALL) -d $D$(LIBDIR)/mold
-	$(INSTALL_DATA) mold-wrapper.so $D$(LIBDIR)/mold
-	$(STRIP) $D$(LIBDIR)/mold/mold-wrapper.so
-
-	$(INSTALL) -d $D$(LIBEXECDIR)/mold
-	ln -sf $(BINDIR)/mold $D$(LIBEXECDIR)/mold/ld
-
-	$(INSTALL) -d $D$(MANDIR)/man1
-	$(INSTALL_DATA) docs/mold.1 $D$(MANDIR)/man1
-
-	ln -sf mold $D$(BINDIR)/ld.mold
-	ln -sf mold $D$(BINDIR)/ld64.mold
-
-uninstall:
-	rm -f $D$(BINDIR)/mold $D$(BINDIR)/ld.mold $D$(BINDIR)/ld64.mold
-	rm -f $D$(MANDIR)/man1/mold.1
-	rm -rf $D$(LIBDIR)/mold
-
-clean:
-	rm -rf *~ mold mold-wrapper.so out ld ld64.mold
-	$(MAKE) -C third-party/xxhash clean
-
-.PHONY: all test tests check clean

README.md

@@ -1,133 +1,117 @@
 # mold: A Modern Linker
 
-mold is a faster drop-in replacement for existing Unix linkers.
-It is several times faster than LLVM lld linker, the second-fastest
-open-source linker which I originally created a few years ago.
-mold is created for increasing developer productivity by reducing
-build time especially in rapid debug-edit-rebuild cycles.
+mold is a faster drop-in replacement for existing Unix linkers. It is several
+times quicker than the LLVM lld linker, the second-fastest open-source linker,
+which I initially developed a few years ago. mold aims to enhance developer
+productivity by minimizing build time, particularly in rapid
+debug-edit-rebuild cycles.
 
-Here is a performance comparison of GNU gold, LLVM lld, and mold for
-linking final debuginfo-enabled executables of major large programs
-on a simulated 8-core 16-threads machine.
+Here is a performance comparison of GNU ld, GNU gold, LLVM lld, and
+mold when linking final debuginfo-enabled executables for major large
+programs on a simulated 16-core, 32-thread machine.
 
-![Link speed comparison](docs/comparison.png)
+![Link speed comparison](docs/chart.svg)
 
-| Program (linker output size)  | GNU gold | LLVM lld | mold
-|-------------------------------|----------|----------|--------
-| Chrome 96 (1.89 GiB)          | 53.86s   | 11.74s   | 2.21s
-| Clang 13 (3.18 GiB)           | 64.12s   | 5.82s    | 2.90s
-| Firefox 89 libxul (1.64 GiB)  | 32.95s   | 6.80s    | 1.42s
+| Program (linker output size)  | GNU ld | GNU gold | LLVM lld | mold
+|-------------------------------|--------|----------|----------|------
+| MySQL 8.3 (0.47 GiB)          | 10.84s | 7.47s    | 1.64s    | 0.46s
+| Clang 19 (1.56 GiB)           | 42.07s | 33.13s   | 5.20s    | 1.35s
+| Chromium 124 (1.35 GiB)       | N/A    | 27.40s   | 6.10s    | 1.52s
 
-mold is so fast that it is only 2x _slower_ than `cp` on the same
-machine. Feel free to [file a bug](https://github.com/rui314/mold/issues)
-if you find mold is not faster than other linkers.
+mold is so fast that it is only 2x _slower_ than the `cp` command on the same
+machine. If you find that mold is not faster than other linkers, feel
+free to [file a bug report](https://github.com/rui314/mold/issues).
 
-mold currently supports x86-64, i386 and ARM64.
+mold supports x86-64, i386, ARM64, ARM32, 64-bit/32-bit little/big-endian
+RISC-V, 32-bit PowerPC, 64-bit big-endian PowerPC ELFv1, 64-bit little-endian
+PowerPC ELFv2, s390x, 64-bit/32-bit LoongArch, SPARC64, m68k, and SH-4.
 
-## Why does the speed of linking matter?
+## Why does linking speed matter?
 
-If you are using a compiled language such as C, C++ or Rust, a build
-consists of two phases. In the first phase, a compiler compiles a
-source file into an object file (`.o` files). In the second phase,
-a linker takes all object files to combine them into a single executable
-or a shared library file.
+If you are using a compiled language such as C, C++, or Rust, a build consists
+of two phases. In the first phase, a compiler compiles source files into
+object files (`.o` files). In the second phase, a linker takes all object
+files and combines them into a single executable or shared library file.
 
-The second phase takes a long time if your build output is large.
-mold can make it faster, saving your time and keeping you from being
-distracted while waiting for a long build to finish. The difference is
-most noticeable when you are in rapid debug-edit-rebuild cycles.
+The second phase can be time-consuming if your build output is large. mold can
+speed up this process, saving you time and preventing distractions while
+waiting for a lengthy build to finish. The difference is most noticeable
+during rapid debug-edit-rebuild cycles.
 
-## Install
+## Installation
 
-Binary packages for the following distros are currently available.
+Binary packages for the following systems are currently available:
 
 [![Packaging status](https://repology.org/badge/vertical-allrepos/mold.svg)](https://repology.org/project/mold/versions)
 
-## How to build
+## How to Build
 
-mold is written in C++20, so if you build mold yourself, you need a
-recent version of a C++ compiler and a C++ standard library. GCC 10.2
-or Clang 12.0.0 as well as libstdc++ 10 or libc++ 7 are recommended.
+mold is written in C++20, so if you build mold yourself, you will need a
+recent version of a C++ compiler and a C++ standard library. We recommend GCC
+10.2 or Clang 12.0.0 (or later) and libstdc++ 10 or libc++ 7 (or later).
 
-I'm using Ubuntu 20.04 as a development platform. In that environment,
-you can build mold by the following commands.
+### Install Dependencies
 
-### Install dependencies
-
-#### Ubuntu 20.04 and later / Debian 11 and later
-
-```shell
-sudo apt-get update
-sudo apt-get install -y build-essential git clang cmake libstdc++-10-dev libssl-dev libxxhash-dev zlib1g-dev pkg-config
-```
-
-#### Fedora 34 and later
-
-```shell
-sudo dnf install -y git clang cmake openssl-devel xxhash-devel zlib-devel libstdc++-devel
-```
+To install build dependencies, run `./install-build-deps.sh` in this
+directory. It will detect your Linux distribution and attempt to install the
+necessary packages. You may need to run it as root.
 
 ### Compile mold
 
 ```shell
-git clone https://github.com/rui314/mold.git
+git clone --branch stable https://github.com/rui314/mold.git
 cd mold
-git checkout v1.0.3
-make -j$(nproc) CXX=clang++
-sudo make install
+./install-build-deps.sh
+cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=c++ -B build
+cmake --build build -j$(nproc)
+sudo cmake --build build --target install
 ```
 
-You may need to pass a C++20 compiler command name to `make`.
-In the above case, `clang++` is passed. If it doesn't work for you,
-try a specific version of a compiler such as `g++-10` or `clang++-12`.
+You might need to pass a C++20 compiler command name to `cmake`. In the
+example above, `c++` is passed. If that doesn't work for you, try a specific
+version of a compiler, such as `g++-10` or `clang++-12`.
 
-By default, `mold` is installed to `/usr/local/bin`.
+By default, `mold` is installed to `/usr/local/bin`. You can change the
+installation location by passing `-DCMAKE_INSTALL_PREFIX=<directory>`.
+For other cmake options, see the comments in `CMakeLists.txt`.
 
-If you don't use a recent enough Linux distribution, or if for any reason `make`
-in the above commands doesn't work for you, you can use Docker to build it in
-a Docker environment. To do so, just run `./build-static.sh` in this
-directory instead of running `make -j$(nproc)`. The shell script creates a
-Ubuntu 20.04 Docker image, installs necessary tools and libraries to it,
-and builds mold as a statically-linked executable.
-
-`make test` depends on a few more packages. To install, run the following commands:
-
-```shell
-sudo dpkg --add-architecture i386
-sudo apt update
-sudo apt-get install bsdmainutils dwarfdump libc6-dev:i386 lib32gcc-10-dev libstdc++-10-dev-arm64-cross gcc-10-aarch64-linux-gnu g++-10-aarch64-linux-gnu
-```
+If you are not using a recent enough Linux distribution, or if `cmake` does
+not work for you for any reason, you can use Podman to build mold in a
+container. To do so, run `./dist.sh` in this directory instead of using
+`cmake`. The shell script will pull a container image, build mold and auxiliary
+files inside it, and package them into a single tar file named
+`dist/mold-$version-$arch-linux.tar.gz`. You can extract the tar file anywhere
+and use the mold executable in it.
 
 ## How to use
 
 <details><summary>A classic way to use mold</summary>
 
-On Unix, the linker command (which is usually `/usr/bin/ld`) is
-invoked indirectly by the compiler driver (which is usually `cc`,
-`gcc` or `clang`), which is typically in turn indirectly invoked by
-`make` or some other build system command.
+On Unix, the linker command (usually `/usr/bin/ld`) is indirectly invoked by
+the compiler driver (typically `cc`, `gcc`, or `clang`), which is in turn
+indirectly invoked by `make` or other build system commands.
 
-If you can specify an additional command line option to your compiler
-driver by modifying build system's config files, add one of the
-following flags to use `mold` instead of `/usr/bin/ld`:
+If you can specify an additional command line option for your compiler driver
+by modifying the build system's config files, add one of the following flags
+to use mold instead of `/usr/bin/ld`:
 
-- Clang: pass `-fuse-ld=mold`
+- For Clang: pass `-fuse-ld=mold`
 
-- GCC 12.1.0 (upcoming version) or later: pass `-fuse-ld=mold`
+- For GCC 12.1.0 or later: pass `-fuse-ld=mold`
 
-- GCC before 12.1.0: `-fuse-ld` does not accept `mold` as a valid
-  argument, so you need to use `-B` option instead. `-B` is an option
-  to tell GCC where to look for external commands such as `ld`.
+- For GCC before 12.1.0: the `-fuse-ld` option does not accept `mold` as a
+  valid argument, so you need to use the `-B` option instead. The `-B` option
+  tells GCC where to look for external commands like `ld`.
 
-  If you have installed mold with `make install`, there should be a
-  directory named `/usr/libexec/mold` (or `/usr/local/libexec/mold`,
-  depending on your $PREFIX), and `ld` command should be there. The
-  `ld` is actually a symlink to `mold`. So, all you need is to pass
-  `-B/usr/libexec/mold` (or `-B/usr/local/libexec/mold`) to GCC.
+  If you have installed mold with `make install`, there should be a directory
+  named `/usr/libexec/mold` (or `/usr/local/libexec/mold`, depending on your
+  `$PREFIX`), and the `ld` command should be there. The `ld` is actually a
+  symlink to `mold`. So, all you need is to pass `-B/usr/libexec/mold` (or
+  `-B/usr/local/libexec/mold`) to GCC.
 
-If you haven't installed `mold` to any `$PATH`, you can still pass
-`-fuse-ld=/absolute/path/to/mold` to clang to use mold. GCC does not
-take an absolute path as an argument for `-fuse-ld` though.
+If you haven't installed `ld.mold` to any `$PATH`, you can still pass
+`-fuse-ld=/absolute/path/to/mold` to clang to use mold. However, GCC does not
+accept an absolute path as an argument for `-fuse-ld`.
 
 </details>
 
@@ -135,77 +119,157 @@ take an absolute path as an argument for `-fuse-ld` though.
 
 Create `.cargo/config.toml` in your project directory with the following:
 
-```
+```toml
 [target.x86_64-unknown-linux-gnu]
 linker = "clang"
 rustflags = ["-C", "link-arg=-fuse-ld=/path/to/mold"]
 ```
 
-where `/path/to/mold` is an absolute path to `mold` exectuable.
+where `/path/to/mold` is an absolute path to the mold executable. In the
+example above, we use `clang` as a linker driver since it always accepts the
+`-fuse-ld` option. If your GCC is recent enough to recognize the option, you
+may be able to remove the `linker = "clang"` line.
 
-If you want to use mold for all projects, put the above snippet to
+```toml
+[target.x86_64-unknown-linux-gnu]
+rustflags = ["-C", "link-arg=-fuse-ld=mold"]
+```
+
+If you want to use mold for all projects, add the above snippet to
 `~/.cargo/config.toml`.
 
 </details>
 
+<details><summary>If you are using Nim</summary>
+
+Create `config.nims` in your project directory with the following:
+
+```nim
+when findExe("mold").len > 0 and defined(linux):
+  switch("passL", "-fuse-ld=mold")
+```
+
+where `mold` must be included in the `PATH` environment variable. In this
+example, `gcc` is used as the linker driver. Use the `-fuse-ld` option if your
+GCC is recent enough to recognize this option.
+
+If you want to use mold for all projects, add the above snippet to
+`~/.config/config.nims`.
+
+</details>
+
+<details><summary>If you are using Conan package manager</summary>
+
+You can configure [Conan](https://github.com/conan-io) to download the latest
+version of `mold` and use it as the linker when building your dependencies and
+projects from source. Please see the instructions [here](https://conan.io/center/recipes/mold).
+
+</details>
+
 <details><summary>mold -run</summary>
 
-It is sometimes very hard to pass an appropriate command line option
-to `cc` to specify an alternative linker.  To deal with the situation,
-mold has a feature to intercept all invocations of `ld`, `ld.lld` or
-`ld.gold` and redirect it to itself. To use the feature, run `make`
-(or another build command) as a subcommand of mold as follows:
+It is sometimes very hard to pass an appropriate command line option to `cc`
+to specify an alternative linker. To address this situation, mold has a
+feature to intercept all invocations of `ld`, `ld.bfd`, `ld.lld`, or `ld.gold`
+and redirect them to itself. To use this feature, run `make` (or another build
+command) as a subcommand of mold as follows:
 
 ```shell
 mold -run make <make-options-if-any>
 ```
 
-Here's an example showing how to link Rust code when using the
-cargo package manager:
-
-```shell
-mold -run cargo build
-```
-
-Internally, mold invokes a given command with `LD_PRELOAD` environment
-variable set to its companion shared object file. The shared object
-file intercepts all function calls to `exec(3)`-family functions to
-replace `argv[0]` with `mold` if it is `ld`, `ld.gold` or `ld.lld`.
+Internally, mold invokes a given command with the `LD_PRELOAD` environment
+variable set to its companion shared object file. The shared object file
+intercepts all function calls to `exec(3)`-family functions to replace
+`argv[0]` with `mold` if it is `ld`, `ld.bf`, `ld.gold`, or `ld.lld`.
 
 </details>
 
-mold leaves its identification string in `.comment` section in an output
+<details><summary>GitHub Actions</summary>
+
+You can use our [setup-mold](https://github.com/rui314/setup-mold) GitHub
+Action to speed up GitHub-hosted continuous builds. Although GitHub Actions
+run on a 4 core machine, mold is still significantly faster than the default
+GNU linker, especially when linking large programs.
+
+</details>
+
+<details><summary>Verify that you are using mold</summary>
+
+mold leaves its identification string in the `.comment` section of an output
 file. You can print it out to verify that you are actually using mold.
 
 ```shell
-readelf -p .comment <executable-file>
+$ readelf -p .comment <executable-file>
 
 String dump of section '.comment':
   [     0]  GCC: (Ubuntu 10.2.0-5ubuntu1~20.04) 10.2.0
   [    2b]  mold 9a1679b47d9b22012ec7dfbda97c8983956716f7
 ```
 
-If `mold` is in `.comment`, the file is created by mold.
+If `mold` is present in the `.comment` section, the file was created by mold.
+
+</details>
+
+<details><summary>Online manual</summary>
+
+Since mold is a drop-in replacement, you should be able to use it without
+reading its manual. However, if you need it, [mold's man page](docs/mold.md)
+is available online. You can read the same manual by running `man mold`.
+
+</details>
 
 ## Why is mold so fast?
 
-One reason is because it simply uses faster algorithms and efficient
-data structures than other linkers do. The other reason is that the
-new linker is highly parallelized.
+One reason is that it utilizes faster algorithms and more efficient data
+structures compared to other linkers. Another reason is that mold is highly
+parallelized.
 
-Here is a side-by-side comparison of per-core CPU usage of lld (left)
-and mold (right). They are linking the same program, Chromium
-executable.
+Here is a side-by-side comparison of per-core CPU usage for lld (left) and
+mold (right), linking the same program, a Chromium executable.
 
 ![CPU usage comparison in htop animation](docs/htop.gif)
 
-As you can see, mold uses all available cores throughout its execution
-and finishes quickly. On the other hand, lld failed to use available
-cores most of the time. In this demo, the maximum parallelism is
-artificially capped to 16 so that the bars fit in the GIF.
+As you can see, mold uses all available cores throughout its execution and
+finishes quickly. In contrast, lld fails to utilize available cores most of
+the time. In this demo, the maximum parallelism is artificially capped at 16,
+so that the bars fit in the GIF.
 
-For details, please read [design notes](docs/design.md).
+For details, please see the [design notes](docs/design.md).
 
-# Logo
+## Sponsors
 
--![mold image](docs/mold.jpg)
+It is taken for granted nowadays that compiler toolchains can be easily
+installed and used for free, and people may not think too much about the
+individuals behind these "free tools". mold supports many projects, but it
+is essentially a one-person project. This situation is similar to the one
+depicted in the following xkcd illustration.
+
+[![xkcd 2347](https://imgs.xkcd.com/comics/dependency.png)](https://xkcd.com/2347)
+
+If you think that the "Nebraska guy" should be rewarded, please consider
+becoming our [GitHub sponsor](https://github.com/sponsors/rui314)!
+
+We thank everyone who sponsors our project. In particular, we'd like to acknowledge
+the following people and organizations who have sponsored $128/month or more:
+
+### Corporate sponsors
+
+<a href="https://mercury.com"><img src="docs/mercury-logo.png" align=center height=120 width=400 alt=Mercury></a>
+
+<a href="https://cybozu-global.com"><img src="docs/cyboze-logo.png" align=center height=120 width=133 alt=Cybozu></a>
+
+<a href="https://www.emergetools.com"><img src="docs/emerge-tools-logo.png" align=center height=120 width=240 alt="Emerge Tools"></a><br>
+
+- [G-Research](https://www.gresearch.co.uk)
+- [Signal Slot Inc.](https://github.com/signal-slot)
+- [GlareDB](https://github.com/GlareDB)
+
+### Individual sponsors
+
+- [Wei Wu](https://github.com/lazyparser)
+- [kyle-elliott](https://github.com/kyle-elliott)
+- [Bryant Biggs](https://github.com/bryantbiggs)
+- [kraptor23](https://github.com/kraptor23)
+- [Jinkyu Yi](https://github.com/jincreator)
+- [Pedro Navarro](https://github.com/pedronavf)

arc.sh

@@ -0,0 +1,29 @@
+#!/bin/bash -x
+set -e
+apt-get install -y git bzip2 ninja-build pkg-config libglib2.0-dev python3-distutils-extra libpixman-1-dev
+
+cd /
+git clone --depth=1 https://github.com/foss-for-synopsys-dwc-arc-processors/qemu.git
+mkdir /qemu/build
+cd /qemu/build
+../configure --static --extra-cflags=-Wno-error
+ninja
+ln -s /qemu/build/qemu-arc /qemu/build/qemu-arc64 /usr/local/bin
+
+
+apt-get install -y libgmp-dev libmpfr-dev texinfo bison yacc flex
+cd /
+git clone --depth=1 https://github.com/foss-for-synopsys-dwc-arc-processors/binutils-gdb.git
+cd /binutils-gdb
+mkdir build
+cd build
+../configure --target=arc-linux
+make -j$(nproc)
+
+
+apt-get install -y mold
+mkdir /build
+cd /build
+cmake -GNinja -DCMAKE_BUILD_TYPE=Debug -DMOLD_USE_MOLD=1 /mold
+ninja
+TRIPLE=arc-linux-gnu bash -x /mold/test/hello-dynamic.sh

build-static.sh

@@ -1,44 +0,0 @@
-#!/bin/bash -x
-# This is a shell script to build a statically-linked mold
-# executable using Docker, so that it is easy to build mold
-# on non-Ubuntu 20.04 machines.
-#
-# A docker image used for building mold is persistent.
-# Run `docker image rm mold-build-ubuntu20` to remove the image
-# from disk.
-
-set -e
-
-# If the existing file is not statically-linked, remove it.
-[ -f mold ] && ! ldd mold 2>&1 | grep -q 'not a dynamic executable' && \
-  rm mold
-
-cat <<EOF | docker build -t mold-build-ubuntu20 -
-FROM ubuntu:20.04
-RUN apt-get update && \
-  TZ=Europe/London apt-get install -y tzdata && \
-  apt-get install -y --no-install-recommends build-essential clang lld \
-    bsdmainutils file gcc-multilib git pkg-config \
-    cmake libstdc++-10-dev zlib1g-dev libssl-dev && \
-  apt clean && \
-  rm -rf /var/lib/apt/lists/*
-EOF
-
-LDFLAGS='-fuse-ld=lld -static'
-
-# libstdc++'s `std::__glibcxx_rwlock_rdlock` refers these symbols
-# as weak symbols, although they need to be defined. Otherwise,
-# the program crashes after juping to address 0.
-# So, we force loading symbols as a workaround.
-LDFLAGS="$LDFLAGS -Wl,-u,pthread_rwlock_rdlock"
-LDFLAGS="$LDFLAGS -Wl,-u,pthread_rwlock_unlock"
-LDFLAGS="$LDFLAGS -Wl,-u,pthread_rwlock_wrlock"
-
-docker_args=(-v "`pwd`:/mold:Z" -u "$(id -u)":"$(id -g)")
-if docker --version | grep -q podman; then
-  docker_args+=(--userns=keep-id)
-fi
-
-docker run -it --rm "${docker_args[@]}" \
-  mold-build-ubuntu20 \
-  make -C /mold -j"$(nproc)" CC=clang CXX=clang++ LDFLAGS="$LDFLAGS" "$@"

byteorder.h

@@ -1,75 +0,0 @@
-#pragma once
-
-#include <cstdint>
-
-namespace mold {
-
-template <typename T>
-class BigEndian {
-public:
-  BigEndian() : BigEndian(0) {}
-
-  BigEndian(T x) {
-    *this = x;
-  }
-
-  operator T() const {
-    // We don't need to optimize this code because compilers are
-    // usually smart enough to compile this loop into a single
-    // byte-swap instruction such as x86's bswap.
-    T ret = 0;
-    for (int i = 0; i < sizeof(T); i++)
-      ret = (ret << 8) | val[i];
-    return ret;
-  }
-
-  BigEndian &operator=(T x) {
-    for (int i = 0; i < sizeof(T); i++)
-      val[sizeof(T) - 1 - i] = x >> (i * 8);
-    return *this;
-  }
-
-  BigEndian &operator++() {
-    return *this = *this + 1;
-  }
-
-  BigEndian operator++(int) {
-    T ret = *this;
-    *this = *this + 1;
-    return ret;
-  }
-
-  BigEndian &operator--() {
-    return *this = *this - 1;
-  }
-
-  BigEndian operator--(int) {
-    T ret = *this;
-    *this = *this - 1;
-    return ret;
-  }
-
-  BigEndian &operator+=(T x) {
-    return *this = *this + x;
-  }
-
-  BigEndian &operator&=(T x) {
-    return *this = *this & x;
-  }
-
-  BigEndian &operator|=(T x) {
-    return *this = *this | x;
-  }
-
-private:
-  uint8_t val[sizeof(T)];
-};
-
-using ibig16 = BigEndian<int16_t>;
-using ibig32 = BigEndian<int32_t>;
-using ibig64 = BigEndian<int64_t>;
-using ubig16 = BigEndian<uint16_t>;
-using ubig32 = BigEndian<uint32_t>;
-using ubig64 = BigEndian<uint64_t>;
-
-} // namespace mold

cmdline.h

@@ -1,63 +0,0 @@
-#pragma once
-
-#include "mold.h"
-
-namespace mold {
-
-template <typename C>
-std::vector<std::string_view>
-read_response_file(C &ctx, std::string_view path) {
-  std::vector<std::string_view> vec;
-  MappedFile<C> *mf = MappedFile<C>::must_open(ctx, std::string(path));
-  u8 *data = mf->data;
-
-  auto read_quoted = [&](i64 i, char quote) {
-    std::string buf;
-    while (i < mf->size && data[i] != quote) {
-      if (data[i] == '\\') {
-        buf.append(1, data[i + 1]);
-        i += 2;
-      } else {
-        buf.append(1, data[i++]);
-      }
-    }
-    if (i >= mf->size)
-      Fatal(ctx) << path << ": premature end of input";
-    vec.push_back(save_string(ctx, buf));
-    return i + 1;
-  };
-
-  auto read_unquoted = [&](i64 i) {
-    std::string buf;
-    while (i < mf->size && !isspace(data[i]))
-      buf.append(1, data[i++]);
-    vec.push_back(save_string(ctx, buf));
-    return i;
-  };
-
-  for (i64 i = 0; i < mf->size;) {
-    if (isspace(data[i]))
-      i++;
-    else if (data[i] == '\'')
-      i = read_quoted(i + 1, '\'');
-    else if (data[i] == '\"')
-      i = read_quoted(i + 1, '\"');
-    else
-      i = read_unquoted(i);
-  }
-  return vec;
-}
-
-template <typename C>
-std::vector<std::string_view> expand_response_files(C &ctx, char **argv) {
-  std::vector<std::string_view> vec;
-  for (i64 i = 0; argv[i]; i++) {
-    if (argv[i][0] == '@')
-      append(vec, read_response_file(ctx, argv[i] + 1));
-    else
-      vec.push_back(argv[i]);
-  }
-  return vec;
-}
-
-} // namespace mold

compress.cc

@@ -1,166 +0,0 @@
-// This file implements a multi-threaded zlib compression routine.
-//
-// Multiple pieces of raw compressed data in zlib-format can be merged
-// just by concatenation as long as each zlib stream is flushed with
-// Z_SYNC_FLUSH. In this file, we split input data into multiple
-// shards, compress them individually and concatenate them. We then
-// append a header, a trailer and a checksum so that the concatenated
-// data is valid zlib-format data.
-//
-// Using threads to compress data has a downside. Since the dictionary
-// is reset on boundaries of shards, compression ratio is sacrificed
-// a little bit. However, if a shard size is large enough, that loss
-// is negligible in practice.
-
-#include "mold.h"
-
-#include <tbb/parallel_for_each.h>
-#include <zlib.h>
-
-namespace mold {
-
-static constexpr i64 SHARD_SIZE = 1024 * 1024;
-
-static std::vector<std::string_view> split(std::string_view input) {
-  std::vector<std::string_view> shards;
-
-  while (input.size() >= SHARD_SIZE) {
-    shards.push_back(input.substr(0, SHARD_SIZE));
-    input = input.substr(SHARD_SIZE);
-  }
-  if (!input.empty())
-    shards.push_back(input);
-  return shards;
-}
-
-static std::vector<u8> do_compress(std::string_view input) {
-  // Initialize zlib stream. Since debug info is generally compressed
-  // pretty well, we chose compression level 3.
-  z_stream strm;
-  strm.zalloc = Z_NULL;
-  strm.zfree = Z_NULL;
-  strm.opaque = Z_NULL;
-  int r = deflateInit2(&strm, 3, Z_DEFLATED, -15, 8, Z_DEFAULT_STRATEGY);
-  assert(r == Z_OK);
-
-  // Set an input buffer
-  strm.avail_in = input.size();
-  strm.next_in = (u8 *)input.data();
-
-  // Set an output buffer. deflateBound() returns an upper bound
-  // on the compression size. +16 for Z_SYNC_FLUSH.
-  std::vector<u8> buf(deflateBound(&strm, strm.avail_in) + 16);
-
-  strm.avail_out = buf.size();
-  strm.next_out = buf.data();
-
-  r = deflate(&strm, Z_SYNC_FLUSH);
-  assert(r == Z_OK);
-  assert(strm.avail_out > 0);
-
-  buf.resize(buf.size() - strm.avail_out);
-  deflateEnd(&strm);
-  return buf;
-}
-
-ZlibCompressor::ZlibCompressor(std::string_view input) {
-  std::vector<std::string_view> inputs = split(input);
-  std::vector<u64> adlers(inputs.size());
-  shards.resize(inputs.size());
-
-  // Compress each shard
-  tbb::parallel_for((i64)0, (i64)inputs.size(), [&](i64 i) {
-    adlers[i] = adler32(1, (u8 *)inputs[i].data(), inputs[i].size());
-    shards[i] = do_compress(inputs[i]);
-  });
-
-  // Combine checksums
-  checksum = adlers[0];
-  for (i64 i = 1; i < inputs.size(); i++)
-    checksum = adler32_combine(checksum, adlers[i], inputs[i].size());
-}
-
-i64 ZlibCompressor::size() const {
-  i64 size = 2;    // +2 for header
-  for (const std::vector<u8> &shard : shards)
-    size += shard.size();
-  return size + 6; // +6 for trailer and checksum
-}
-
-void ZlibCompressor::write_to(u8 *buf) {
-  // Write a zlib-format header
-  buf[0] = 0x78;
-  buf[1] = 0x9c;
-
-  // Copy compressed data
-  std::vector<i64> offsets(shards.size());
-  offsets[0] = 2; // +2 for header
-  for (i64 i = 1; i < shards.size(); i++)
-    offsets[i] = offsets[i - 1] + shards[i - 1].size();
-
-  tbb::parallel_for((i64)0, (i64)shards.size(), [&](i64 i) {
-    memcpy(&buf[offsets[i]], shards[i].data(), shards[i].size());
-  });
-
-  // Write a trailer
-  u8 *end = buf + size();
-  end[-6] = 3;
-  end[-5] = 0;
-
-  // Write a checksum
-  *(ubig32 *)(end - 4) = checksum;
-}
-
-GzipCompressor::GzipCompressor(std::string_view input) {
-  std::vector<std::string_view> inputs = split(input);
-  std::vector<u32> crc(inputs.size());
-  shards.resize(inputs.size());
-
-  // Compress each shard
-  tbb::parallel_for((i64)0, (i64)inputs.size(), [&](i64 i) {
-    crc[i] = crc32(0, (u8 *)inputs[i].data(), inputs[i].size());
-    shards[i] = do_compress(inputs[i]);
-  });
-
-  // Combine checksums
-  checksum = crc[0];
-  for (i64 i = 1; i < inputs.size(); i++)
-    checksum = crc32_combine(checksum, crc[i], inputs[i].size());
-
-  uncompressed_size = input.size();
-}
-
-i64 GzipCompressor::size() const {
-  i64 size = 10;    // +10 for header
-  for (const std::vector<u8> &shard : shards)
-    size += shard.size();
-  return size + 10; // +10 for trailer and checksum
-}
-
-void GzipCompressor::write_to(u8 *buf) {
-  // Write a zlib-format header
-  memset(buf, 0, 10);
-  buf[0] = 0x1f; // magic
-  buf[1] = 0x8b; // magic
-  buf[2] = 0x08; // compression method is zlib
-  buf[9] = 0xff; // made on unknown OS
-
-  // Copy compressed data
-  std::vector<i64> offsets(shards.size());
-  offsets[0] = 10; // +10 for header
-  for (i64 i = 1; i < shards.size(); i++)
-    offsets[i] = offsets[i - 1] + shards[i - 1].size();
-
-  tbb::parallel_for((i64)0, (i64)shards.size(), [&](i64 i) {
-    memcpy(&buf[offsets[i]], shards[i].data(), shards[i].size());
-  });
-
-  // Write a trailer
-  u8 *end = buf + size();
-  end[-10] = 0x3; // two-byte zlib stream terminator
-  end[-9] = 0;
-  *(u32 *)(end - 8) = checksum;
-  *(u32 *)(end - 4) = uncompressed_size;
-}
-
-} // namespace mold

demangle.cc

@@ -1,24 +0,0 @@
-#include "mold.h"
-
-#include <cstdlib>
-#include <cxxabi.h>
-
-namespace mold {
-
-std::string_view demangle(std::string_view name) {
-  if (name.starts_with("_Z")) {
-    static thread_local char *buf;
-    static thread_local size_t buflen;
-
-    int status;
-    char *p = abi::__cxa_demangle(std::string(name).c_str(), buf, &buflen, &status);
-    if (status == 0) {
-      buf = p;
-      return p;
-    }
-  }
-
-  return name;
-}
-
-} // namespace mold

dist.sh

@@ -0,0 +1,219 @@
+#!/bin/bash
+#
+# This script creates a mold binary distribution. The output is written in
+# `dist` directory as `mold-$version-$arch-linux.tar.gz` (e.g.,
+# `mold-1.0.3-x86_64-linux.tar.gz`).
+#
+# The mold executable created by this script is statically linked to
+# libstdc++ but dynamically linked to libc, libm, libz, and librt, as
+# these libraries almost always exist on any Linux system. We can't
+# statically link libc because doing so would disable dlopen(), which is
+# necessary to open the LTO linker plugin.
+#
+# This script aims to produce reproducible outputs. That means if you run
+# the script twice on the same git commit, it should produce bit-by-bit
+# identical binary files. This property is crucial as a countermeasure
+# against supply chain attacks. With this, you can verify that the binary
+# files distributed on the GitHub release pages are created from the
+# commit with release tags by rebuilding the binaries yourself.
+#
+# Debian provides snapshot.debian.org to host all historical binary
+# packages. We use it to construct Podman images pinned to a
+# particular timestamp.
+#
+# We aim to use a reasonably old Debian version because we'll dynamically
+# link glibc to mold, and a binary linked against a newer version of glibc
+# won't work on a system with an older version of glibc.
+#
+# We want to build mold with Clang rather than GCC because mold's
+# Identical Code Folding works best with the LLVM address significance
+# table (.llvm_addrsig). Building a release binary with GCC yields a
+# slightly larger binary than Clang's.
+#
+# We need a recent version of Clang to build mold. If it's not available
+# via apt-get, we'll build it ourselves.
+#
+# You may need to install qemu-user-static package to build non-native
+# binaries.
+
+set -e -x
+cd "$(dirname $0)"
+
+usage() {
+  echo "Usage: $0 [ x86_64 | aarch64 | arm | riscv64 | ppc64le | s390x | loongarch64 ]"
+  exit 1
+}
+
+case $# in
+0)
+  arch=$(uname -m)
+  if [ $arch = arm64 ]; then
+    arch=aarch64
+  elif [[ $arch = arm* ]]; then
+    arch=arm
+  fi
+  ;;
+1)
+  arch="$1"
+  ;;
+*)
+  usage
+esac
+
+# Create a Podman image.
+if [ "$GITHUB_REPOSITORY" = '' ]; then
+  image=mold-builder-$arch
+  image_build="podman build --arch $arch -t $image -"
+else
+  # If this script is running on GitHub Actions, we want to cache
+  # the created container image in GitHub's container repostiory.
+  image=ghcr.io/$GITHUB_REPOSITORY/mold-builder-$arch
+  image_build="podman build --arch $arch -t $image --output=type=registry --layers --cache-to $image --cache-from $image -"
+fi
+
+case $arch in
+x86_64)
+  # Debian 9 (Stretch) released in June 2017.
+  # We use a Google-provided mirror (gcr.io) of the official Docker hub
+  # (docker.io) because docker.io has a strict rate limit policy.
+  cat <<EOF | $image_build
+FROM mirror.gcr.io/library/debian:stretch@sha256:c5c5200ff1e9c73ffbf188b4a67eb1c91531b644856b4aefe86a58d2f0cb05be
+ENV DEBIAN_FRONTEND=noninteractive TZ=UTC
+RUN sed -i -e '/^deb/d' -e 's/^# deb /deb [trusted=yes] /g' /etc/apt/sources.list && \
+  echo 'Acquire::Retries "10"; Acquire::http::timeout "10"; Acquire::Check-Valid-Until "false";' > /etc/apt/apt.conf.d/80-retries && \
+  apt-get update && \
+  apt-get install -y --no-install-recommends wget file make gcc g++ zlib1g-dev libssl-dev && \
+  rm -rf /var/lib/apt/lists
+
+# Build CMake 3.27
+RUN mkdir /build && \
+  cd /build && \
+  wget -O- --no-check-certificate --progress=dot:mega https://cmake.org/files/v3.27/cmake-3.27.7.tar.gz | tar xzf - --strip-components=1 && \
+  ./bootstrap --parallel=\$(nproc) && \
+  make -j\$(nproc) && \
+  make install && \
+  rm -rf /build
+
+# Build GCC 14
+RUN mkdir /build && \
+  cd /build && \
+  wget -O- --no-check-certificate --progress=dot:mega https://ftpmirror.gnu.org/gcc/gcc-14.2.0/gcc-14.2.0.tar.gz | tar xzf - --strip-components=1 && \
+  mkdir gmp mpc mpfr && \
+  wget -O- --no-check-certificate --progress=dot:mega https://ftpmirror.gnu.org/gmp/gmp-6.3.0.tar.gz | tar xzf - --strip-components=1 -C gmp && \
+  wget -O- --no-check-certificate --progress=dot:mega https://ftpmirror.gnu.org/mpc/mpc-1.3.1.tar.gz | tar xzf - --strip-components=1 -C mpc && \
+  wget -O- --no-check-certificate --progress=dot:mega https://ftpmirror.gnu.org/mpfr/mpfr-4.2.1.tar.gz | tar xzf - --strip-components=1 -C mpfr && \
+  ./configure --prefix=/usr --enable-languages=c,c++ --disable-bootstrap --disable-multilib && \
+  make -j\$(nproc) && \
+  make install && \
+  ln -sf /usr/lib64/libstdc++.so.6 /usr/lib/x86_64-linux-gnu/libstdc++.so.6 && \
+  rm -rf /build
+
+# Build GNU binutils 2.43
+RUN mkdir /build && \
+  cd /build && \
+  wget -O- --no-check-certificate --progress=dot:mega https://ftpmirror.gnu.org/binutils/binutils-2.43.tar.gz | tar xzf - --strip-components=1 && \
+  ./configure --prefix=/usr && \
+  make -j\$(nproc) && \
+  make install && \
+  rm -fr /build
+
+# Build Python 3.12.7
+RUN mkdir /build && \
+  cd /build && \
+  wget -O- --no-check-certificate --progress=dot:mega https://www.python.org/ftp/python/3.12.7/Python-3.12.7.tgz | tar xzf - --strip-components=1 && \
+  ./configure && \
+  make -j\$(nproc) && \
+  make install && \
+  rm -rf /build
+
+# Build LLVM 19
+RUN mkdir /build && \
+  cd /build && \
+  wget -O- --no-check-certificate --progress=dot:mega https://github.com/llvm/llvm-project/archive/refs/tags/llvmorg-19.1.7.tar.gz | tar xzf - --strip-components=1 && \
+  mkdir b && \
+  cd b && \
+  cmake -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_PROJECTS=clang ../llvm && \
+  cmake --build . -j\$(nproc) && \
+  cmake --install . --strip && \
+  rm -rf /build
+EOF
+  ;;
+aarch64 | arm | ppc64le | s390x)
+  # Debian 11 (Bullseye) released in August 2021
+  #
+  # We don't want to build Clang for these targets with Qemu becuase
+  # that'd take extremely long time. Also I believe old build machines
+  # are usually x86-64.
+  cat <<EOF | $image_build
+FROM mirror.gcr.io/library/debian:bullseye-20240904@sha256:8ccc486c29a3ad02ad5af7f1156e2152dff3ba5634eec9be375269ef123457d8
+ENV DEBIAN_FRONTEND=noninteractive TZ=UTC
+RUN sed -i -e '/^deb/d' -e 's/^# deb /deb [trusted=yes] /g' /etc/apt/sources.list && \
+  echo 'Acquire::Retries "10"; Acquire::http::timeout "10"; Acquire::Check-Valid-Until "false";' > /etc/apt/apt.conf.d/80-retries && \
+  apt-get update && \
+  apt-get install -y --no-install-recommends build-essential gcc-10 g++-10 clang-16 cmake && \
+  ln -sf /usr/bin/clang-16 /usr/bin/clang && \
+  ln -sf /usr/bin/clang++-16 /usr/bin/clang++ && \
+  rm -rf /var/lib/apt/lists
+EOF
+  ;;
+riscv64)
+  cat <<EOF | $image_build
+FROM mirror.gcr.io/riscv64/debian:unstable-20240926@sha256:25654919c2926f38952cdd14b3300d83d13f2d820715f78c9f4b7a1d9399bf48
+ENV DEBIAN_FRONTEND=noninteractive TZ=UTC
+RUN sed -i -e '/^URIs/d' -e 's/^# http/URIs: http/' /etc/apt/sources.list.d/debian.sources && \
+  echo 'Acquire::Retries "10"; Acquire::http::timeout "10"; Acquire::Check-Valid-Until "false";' > /etc/apt/apt.conf.d/80-retries && \
+  apt-get update && \
+  apt-get install -y --no-install-recommends build-essential gcc-14 g++-14 clang-18 cmake && \
+  ln -sf /usr/bin/clang-18 /usr/bin/clang && \
+  ln -sf /usr/bin/clang++-18 /usr/bin/clang++ && \
+  rm -rf /var/lib/apt/lists
+EOF
+  ;;
+loongarch64)
+  # LoongArch build is not reproducible yet
+  cat <<EOF | $image_build
+FROM mirror.gcr.io/loongarch64/debian:sid
+ENV DEBIAN_FRONTEND=noninteractive TZ=UTC
+RUN apt-get update && \
+  apt-get install -y --no-install-recommends build-essential gcc-14 g++-14 clang-19 cmake && \
+  ln -sf /usr/bin/clang-19 /usr/bin/clang && \
+  ln -sf /usr/bin/clang++-19 /usr/bin/clang++ && \
+  rm -rf /var/lib/apt/lists
+EOF
+  ;;
+*)
+  usage
+  ;;
+esac
+
+version=$(sed -n 's/^project(mold VERSION \(.*\))/\1/p' CMakeLists.txt)
+dest=mold-$version-$arch-linux
+
+# Source tarballs available on GitHub don't contain .git history.
+# Clone the repo if missing.
+[ -d .git ] || git clone --branch v$version --depth 1 --bare https://github.com/rui314/mold .git
+
+# We use the timestamp of the last Git commit as the file timestamp
+# for build artifacts.
+timestamp="$(git log -1 --format=%ci)"
+
+# Build mold in a container.
+mkdir -p dist
+
+podman run --arch $arch -it --rm --userns=host --pids-limit=-1 --network=none \
+  -v "$(pwd):/mold:ro" -v "$(pwd)/dist:/dist" $image bash -c "
+set -e
+mkdir /build
+cd /build
+cmake -DCMAKE_BUILD_TYPE=Release -DMOLD_MOSTLY_STATIC=1 -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ /mold
+cmake --build . -j\$(nproc)
+cmake --install .
+cmake -DMOLD_USE_MOLD=1 .
+cmake --build . -j\$(nproc)
+ctest --output-on-failure -j\$(nproc)
+cmake --install . --prefix $dest --strip
+find $dest -print | xargs touch --no-dereference --date='$timestamp'
+find $dest -print | sort | tar -cf - --no-recursion --files-from=- | gzip -9nc > /dist/$dest.tar.gz
+cp mold /dist
+sha256sum /dist/$dest.tar.gz
+"

docs/chart.svg

@@ -0,0 +1 @@
+<svg version="1.1" viewBox="0.0 0.0 600.0 371.0" fill="none" stroke="none" stroke-linecap="square" stroke-miterlimit="10" width="600" height="371" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns="http://www.w3.org/2000/svg"><path fill="#ffffff" d="M0 0L600.0 0L600.0 371.0L0 371.0L0 0Z" fill-rule="nonzero"/><path stroke="#333333" stroke-width="1.0" stroke-linecap="butt" d="M82.5 275.5L581.5 275.5" fill-rule="nonzero"/><path stroke="#cccccc" stroke-width="1.0" stroke-linecap="butt" d="M82.5 231.5L581.5 231.5" fill-rule="nonzero"/><path stroke="#cccccc" stroke-width="1.0" stroke-linecap="butt" d="M82.5 186.5L581.5 186.5" fill-rule="nonzero"/><path stroke="#cccccc" stroke-width="1.0" stroke-linecap="butt" d="M82.5 142.5L581.5 142.5" fill-rule="nonzero"/><path stroke="#cccccc" stroke-width="1.0" stroke-linecap="butt" d="M82.5 98.5L581.5 98.5" fill-rule="nonzero"/><path stroke="#cccccc" stroke-width="1.0" stroke-linecap="butt" d="M82.5 53.5L581.5 53.5" fill-rule="nonzero"/><clipPath id="id_0"><path d="M82.55 53.916668L581.45 53.916668L581.45 275.45L82.55 275.45L82.55 53.916668Z" clip-rule="nonzero"/></clipPath><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M158.0 275.0L136.0 275.0L136.0 229.0C136.0 227.89543 136.89543 227.0 138.0 227.0L156.0 227.0C157.10457 227.0 158.0 227.89543 158.0 229.0Z" fill-rule="nonzero"/><path fill="#4285f4" clip-path="url(#id_0)" d="M158.0 275.0L136.0 275.0L136.0 229.0C136.0 227.89543 136.89543 227.0 138.0 227.0L156.0 227.0C157.10457 227.0 158.0 227.89543 158.0 229.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M307.0 275.0L285.0 275.0L285.0 91.0C285.0 89.89543 285.89542 89.0 287.0 89.0L305.0 89.0C306.10458 89.0 307.0 89.89543 307.0 91.0Z" fill-rule="nonzero"/><path fill="#4285f4" clip-path="url(#id_0)" d="M307.0 275.0L285.0 275.0L285.0 91.0C285.0 89.89543 285.89542 89.0 287.0 89.0L305.0 89.0C306.10458 89.0 307.0 89.89543 307.0 91.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M182.0 275.0L160.0 275.0L160.0 244.0C160.0 242.89543 160.89543 242.0 162.0 242.0L180.0 242.0C181.10457 242.0 182.0 242.89543 182.0 244.0Z" fill-rule="nonzero"/><path fill="#ea4335" clip-path="url(#id_0)" d="M182.0 275.0L160.0 275.0L160.0 244.0C160.0 242.89543 160.89543 242.0 162.0 242.0L180.0 242.0C181.10457 242.0 182.0 242.89543 182.0 244.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M331.0 275.0L309.0 275.0L309.0 131.0C309.0 129.89543 309.89542 129.0 311.0 129.0L329.0 129.0C330.10458 129.0 331.0 129.89543 331.0 131.0Z" fill-rule="nonzero"/><path fill="#ea4335" clip-path="url(#id_0)" d="M331.0 275.0L309.0 275.0L309.0 131.0C309.0 129.89543 309.89542 129.0 311.0 129.0L329.0 129.0C330.10458 129.0 331.0 129.89543 331.0 131.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M480.0 275.0L458.0 275.0L458.0 156.0C458.0 154.89543 458.89542 154.0 460.0 154.0L478.0 154.0C479.10458 154.0 480.0 154.89543 480.0 156.0Z" fill-rule="nonzero"/><path fill="#ea4335" clip-path="url(#id_0)" d="M480.0 275.0L458.0 275.0L458.0 156.0C458.0 154.89543 458.89542 154.0 460.0 154.0L478.0 154.0C479.10458 154.0 480.0 154.89543 480.0 156.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M206.0 275.0L184.0 275.0L184.0 270.0C184.0 268.89542 184.89543 268.0 186.0 268.0L204.0 268.0C205.10457 268.0 206.0 268.89542 206.0 270.0Z" fill-rule="nonzero"/><path fill="#fbbc04" clip-path="url(#id_0)" d="M206.0 275.0L184.0 275.0L184.0 270.0C184.0 268.89542 184.89543 268.0 186.0 268.0L204.0 268.0C205.10457 268.0 206.0 268.89542 206.0 270.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M355.0 275.0L333.0 275.0L333.0 254.0C333.0 252.89543 333.89542 252.0 335.0 252.0L353.0 252.0C354.10458 252.0 355.0 252.89543 355.0 254.0Z" fill-rule="nonzero"/><path fill="#fbbc04" clip-path="url(#id_0)" d="M355.0 275.0L333.0 275.0L333.0 254.0C333.0 252.89543 333.89542 252.0 335.0 252.0L353.0 252.0C354.10458 252.0 355.0 252.89543 355.0 254.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroke-opacity="0.0" clip-path="url(#id_0)" d="M504.0 275.0L482.0 275.0L482.0 250.0C482.0 248.89543 482.89542 248.0 484.0 248.0L502.0 248.0C503.10458 248.0 504.0 248.89543 504.0 250.0Z" fill-rule="nonzero"/><path fill="#fbbc04" clip-path="url(#id_0)" d="M504.0 275.0L482.0 275.0L482.0 250.0C482.0 248.89543 482.89542 248.0 484.0 248.0L502.0 248.0C503.10458 248.0 504.0 248.89543 504.0 250.0Z" fill-rule="nonzero"/><path stroke="#000000" stroke-width="2.0" stroke-linecap="butt" stroLine truncated

docs/coding-guidelines.md

@@ -0,0 +1,46 @@
+# Coding Guidelines
+
+mold is written in C++20, but as is the case with every C++ project,
+it has local coding rules. In this document, I'll explain some of them
+and try to give justifications for why I chose such rules.
+
+## DOs
+
+- Always use `i64` (which is a type alias for `int64_t` in mold) for
+  integers unless you have a reason not to. For example, even if you know
+  that a loop counter won't exceed 100, you should stop thinking about it
+  and just use `i64`.
+
+  Justification: Local variables are usually on CPU registers, so on
+  64-bit CPUs, there's no performance peanlty on choosing `i64` over
+  `i32`. Even if a compiler has to spill register values to the stack,
+  I don't think there's an observable difference between `i32` and
+  `i64`. Therefore, extra 32 bits are essentially free. On 32-bit CPUs,
+  they are not free, but that's OK because we are writing mold for modern
+  computers. mold will still run on 32-bit computers but a bit slowly.
+
+  By always using `i64`, we can eliminate the need to think about the
+  "right" size for each variable. It also reduces the risk of integer
+  overflow.
+
+  Exceptions: If you have to allocate a very large number (e.g. millions)
+  of the same object, its size matters. In that case, use a smaller type.
+
+## DON'Ts
+
+- Don't use `auto` unless its actual type is obvious in the very narrow
+  context. Currently, we use `auto` only for lambdas.
+
+  Justification: I think `auto` makes code writing easier but code reading
+  harder, because readers have to make a guess as to what is the actual
+  type of `auto`. If you are already familiar with the existing codebase,
+  you may be able to guess it easily, but that's not always the case.
+  I want to keep the mold codebase friendly to first-time readers.
+
+- Don't over-use inheritance. In mold, most classes don't have parents,
+  and even if they do, their class hierarchy is very shallow. Currently
+  its height is just two (i.e. abstract classes and their implementations).
+
+  Justification: Designing class hierarchies is fun as it feels like
+  taxonomy, but I don't think that always help writing code. It looks like
+  simpler class hierarchy makes its code simpler.

docs/design.md

@@ -1,3 +1,8 @@
+[This document was written in 2020, and the contents are outdated.
+Specifically, we no longer believe that object preloading is a good
+idea. That being said, most of the points in this document still hold
+even today. Therefore, I'll keep this document as-is.]
+
 ## Design and implementation of mold
 
 For the rest of this documentation, I'll explain the design and the

docs/execstack.md

@@ -0,0 +1,55 @@
+This page explains the following warning message and how to fix it.
+mold emits this message when it sees an object file that may not be
+compatible with mold.
+
+
+```
+mold: warning: foo.o: this file may cause a segmentation fault because it requires an executable stack. See https://github.com/rui314/mold/tree/main/docs/execstack.md for more info.
+```
+
+# Background
+
+On modern computers, the stack area (to which local variables are
+stored) cannot contain executable code. If the control reaches the
+stack area, the CPU refuses to execute any code there and the program
+is usually terminated due to segmentation fault.
+
+This is a security measure. The stack area used to be executable (old
+CPUs generally execute any code as long as it is in a readable memory
+region), but that provided an easy attack vector to a malicious user.
+They wrote executable code to the stack area using some buffer
+overflow bug and jumped there to run arbitrary code in a remote server
+process.
+
+To prevent this type of attack, the stack area is no longer executable
+since the early 2000s. On Linux, the stack's executable-ness is
+controlled by a bit in an executable, and the loader respects that
+bit. The bit is set by the linker.
+
+GCC had (and still has) a feature that depends on the executable
+stack, so they invented a way to tell the linker to mark the stack
+executable. Specifically, if an object file contains a
+`.note.GNU-stack` section with the `SHF_EXECSTACK` bit, GNU linker
+silently makes the stack of an output file executable.
+
+But the GNU linker's behavior is dangerous. If you accidentally link
+an object file that has that marker section, the entire stack area
+silently becomes executable, disabling the security mechanism.
+
+Therefore, mold simply ignores that marker section. If you are using
+mold, you need to explicitly pass `-z execstack` to the linker to make
+the stack executable.
+
+# What caused this issue?
+
+You are likely to use GCC's [Nested
+Functions](https://gcc.gnu.org/onlinedocs/gcc/Nested-Functions.html)
+feature which still depends on the executable stack.
+
+# How to fix it?
+
+If you know what you are doing, pass `-z execstack` to mold. Beware
+that this will significantly weaken your program's security.
+
+If you don't want to pass `-z execstack`, rewrite your code so that
+your code does not depend on the executable stack.

docs/glossary.md

@@ -0,0 +1,158 @@
+The very concept of linking is simple: a compiler compiles a piece of
+source code into an object file (a file containing machine code), and
+a linker combines object files into a single executable or a shared
+library file. However, the actual implementation of the linker for
+modern systems is much more complicated because hardware, operating
+system, compiler and linker all have many more features.
+
+In this file, I'll explain random topics in the glossary format that
+you need to understand to read mold code.
+
+## DSO
+
+A .so file. Short for Dynamic Shared Object. Often called as a
+shared library, a dynamic libray or a shared object as well.
+
+An DSO contains common functions and data that are used by multiple
+executables and/or other DSOs. At runtime, a DSO is loaded to a
+contiguous region in the virtual address.
+
+## Object file
+
+A .o file. An object file contains machine code and data, but it
+cannot be executed because it's not self-contained. For example,
+if you compile a C source file containing a call of `printf`,
+the actual function code of `printf` is not included in the resulting
+object file. You include `stdio.h`, but that teaches the compiler
+only about `printf`'s type, and the compiler still don't know what
+`printf` actually does. Therefore, it cannot emit code for `printf`.
+
+You need to link an object file with other object file or a shared
+library to make it exectuable.
+
+## Virtual address space
+
+A pointer has a value like 0x803020 which is an address of the
+pointee. But it doesn't mean that the pointee resides at the
+physical memory address 0x803020 on the computer. Modern CPUs
+contains so-called Mmeory Management Unit (MMU), and all access to
+the memory are first translated by MMU to the physical address.
+The address before translation is called the "virtual address".
+Unless you are doing the kernel programming, all addresses you
+handle are virtual addresses.
+
+The OS kernel controls the MMU so that each process owns the entire
+virtual address space. So, even if two process uses the same virtual
+address, they don't conflict. They are mapped to different physical
+addresses.
+
+The existence of MMU has several implications to the linker. First,
+we can link the main executable to a specific address. On process
+startup, there's no code or data in the virtual address space, so
+the mapping of the main executable always succeed. However, it's not
+true to DSOs because they are loaded after the main executable and
+possibly other DSOs. Therefore, shared libraries must be linked in a
+way that they can be loaded to any address in the virtual address
+space.
+
+## Relocation
+
+A piece of information for the linker as to how to link object files
+or a dynamic objects.
+
+Object files can refer functions or data in other object files. For
+example, if you compile a function which calls a non-local function
+`foo`, the resulting code contains something like this:
+
+```
+  26:   e8 00 00 00 00          callq  2b <bar+0xb>
+                        27: R_X86_64_PLT32      foo-0x4
+```
+
+The above `callq` is the instruction to call a function at the
+machine code level. It's opcode is `0xe8` in x86-64, so the
+instruction begins with `0xe8`. The following four bytes are
+displacement; that is, the address of the branch target relative to
+the end of this `callq` instruction. Notice that the displacement is
+0. The compiler couldn't fill the displacement because it has no
+idea as to where `foo` will be at runtime. So, the compiler write 0
+as a placeholder and instead write a relocation `R_X86_64_PLT32`
+with `foo` as its associated symbol. The linker reads this
+relocation, computes the offsets between this call instruction and
+function `foo` and overwrite the placeholder value 0 with an actual
+displacement.
+
+There are many different types of relocations. For example, if you
+want to fix up not with a displacement but with an absolute address
+of a symbol, you need to use `R_X86_64_ABS64` instead.
+
+## Static library
+
+A .a file. Often called as an archive file or just archive as well.
+
+A static library is a container just like tar or zip. Actually,
+there's no technical reason to not use tar or (uncompressed) zip,
+but traditionally the .a file format is used by the linker.
+
+A static library contains object files and can be passed to the
+linker along with other object files and/or archives.
+
+A linker pulls out object files from an archive only if it is needed
+to resolve undefined symbols. In other words, object files in an
+archive are not linked by default and used as a complement to supply
+missing definitions. This is ideal for a library because you don't
+want to link library functions unless you are actually using them.
+
+Contrary to archive files, object files directly given to a linker
+are always linked to the output.
+
+To maximize the benefit of archive files, a library often used as a
+static library is broken down to small files to separate each
+function individually (for example, look at
+https://git.musl-libc.org/cgit/musl/tree/src/stdio). By doing this,
+you import only used functions.
+
+A static file is created by `ar`, whose command line arguments are
+similar to `tar`. A static library contains the symbol table which
+offers a quick way to look up an object file for a defined symbol,
+but mold does not use the static library's symbol table. mold
+doesdn't need a symbol table to exist in an archive, and if exists,
+mold just ignores it.
+
+See also: DSO (dynamic library)
+
+## Symbol
+
+A symbol is a label assigned to a specific location in an input file
+or an output file. For example, if you define function `foo` and
+compile it, the resulting object file contains a symbol `foo`
+pointing to the beginning of the machine code for `foo`.
+
+Usually, a symbol name is a function or a variable name. If an
+object is anonymous (such the one for a string literal), a compiler
+generated a unique symbol, which often starts with `.` to avoid
+conflict with user-defined symbols.
+
+For C++, symbol name is a complex "mangled" name. We need to mangle
+identifiers because a simple name such as `foo` cannot be uniquely
+identify a function or a data in C++, because for example `foo` may
+be in a namespace or defined as a static member in some class. If
+`foo` is an overloaded function, we need to distinguish different
+`foo`s by its type. Therefore, C++ compiler mangles an identifier by
+appending nmaepsace names, type information and such so that
+different things get different names.
+
+For example, a function `int foo(int)` in a namespace `bar` is
+mangled as `_ZN3bar3fooEi`.
+
+A symbol can be either defined or undefined. A defined symbol points
+to some location in a file which may contain the function's machine
+code or the variable's initial value. An undefined symbol does not
+point to anywhere. It needs to be merged with a defined symbol with
+the same name at link-time. This merging process is called "name
+resolution".
+
+For example, if your program is using `printf`, it usually contains
+`printf` as an undefined symbol. You need to link it with `libc.a`
+or `libc.so`, which contain a defined symbol of `printf`, to make a
+complete program.

docs/mold.md

@@ -0,0 +1,935 @@
+mold(1) -- a modern linker
+==========================
+
+## SYNOPSIS
+
+`mold` [_option_...] _file_...
+
+## DESCRIPTION
+`mold` is a faster drop-in replacement for the default GNU ld(1).
+
+### How to use
+See <https://github.com/rui314/mold#how-to-use>.
+
+### Compatibility
+
+`mold` is designed to be a drop-in replacement for the GNU linkers for
+linking user-land programs. If your user-land program cannot be built due to
+missing command-line options, please file a bug at
+<https://github.com/rui314/mold/issues>.
+
+`mold` supports a very limited set of linker script features, which is just
+sufficient to read `/usr/lib/x86_64-linux-gnu/libc.so` on Linux systems (on
+Linux, that file is contrary to its name not a shared library but an ASCII
+linker script that loads a real `libc.so` file.)
+
+Beyond that, we have no plan to support any additional linker script features.
+The linker script is an ad-hoc, over-designed, complex language which we
+believe needs to be replaced by a simpler mechanism. We have a plan to add a
+replacement for the linker script to `mold` instead.
+
+### Archive symbol resolution
+
+Traditionally, Unix linkers are sensitive to the order in which input files
+appear on the command line. They process input files from the first (leftmost)
+file to the last (rightmost) file one-by-one. While reading input files, they
+maintain sets of defined and undefined symbols. When visiting an archive file
+(`.a` files), they pull out object files to resolve as many undefined symbols
+as possible and move on to the next input file. Object files that weren't
+pulled out will never have a chance for a second look.
+
+Due to this behavior, you usually have to add archive files at the end of a
+command line, so that when a linker reaches archive files, it knows what
+symbols remain as undefined.
+
+If you put archive files at the beginning of a command line, a linker doesn't
+have any undefined symbols, and thus no object files will be pulled out from
+archives. You can change the processing order by using the `--start-group` and
+`--end-group` options, though they make a linker slower.
+
+`mold`, as well as the LLVM lld(1) linker, takes a different approach. They
+remember which symbols can be resolved from archive files instead of
+forgetting them after processing each archive. Therefore, `mold` and lld(1)
+can "go back" in a command line to pull out object files from archives if they
+are needed to resolve remaining undefined symbols. They are not sensitive to
+the input file order.
+
+`--start-group` and `--end-group` are still accepted by `mold` and lld(1)
+for compatibility with traditional linkers, but they are silently ignored.
+
+### Dynamic symbol resolution
+
+Some Unix linker features are difficult to understand without comprehending
+the semantics of dynamic symbol resolution. Therefore, even though it's not
+specific to `mold`, we'll explain it here.
+
+We use "ELF module" or just "module" as a collective term to refer to an
+executable or a shared library file in the ELF format.
+
+An ELF module may have lists of imported symbols and exported symbols, as well
+as a list of shared library names from which imported symbols should be
+imported. The point is that imported symbols are not bound to any specific
+shared library until runtime.
+
+Here is how the Unix dynamic linker resolves dynamic symbols. Upon the start
+of an ELF program, the dynamic linker constructs a list of ELF modules which,
+as a whole, consist of a complete program. The executable file is always at
+the beginning of the list followed by its dependent shared libraries. An
+imported symbol is searched from the beginning of the list to the end. If two
+or more modules define the same symbol, the one that appears first in the list
+takes precedence over the others.
+
+This Unix semantics are contrary to systems such as Windows that have a
+two-level namespace for dynamic symbols. On Windows, for example, dynamic
+symbols are represented as a tuple of (`symbol-name`, `shared-library-name`),
+so that each dynamic symbol is guaranteed to be resolved from some specific
+library.
+
+Typically, an ELF module that exports a symbol also imports the same symbol.
+Such a symbol is usually resolved to itself, but that's not the case if a
+module that appears before it in the symbol search list provides another
+definition of the same symbol.
+
+Let's take `malloc` as an example. Assume that you define your version of
+`malloc` in your main executable file. Then, all `malloc` calls from any
+module are resolved to your function instead of the one in libc, because the
+executable is always at the beginning of the dynamic symbol search list. Note
+that even `malloc` calls within libc are resolved to your definition since
+libc exports and imports `malloc`. Therefore, by defining `malloc` yourself,
+you can overwrite a library function, and the `malloc` in libc becomes dead
+code.
+
+These Unix semantics are tricky and sometimes considered harmful. For example,
+assume that you accidentally define `atoi` as a global function in your
+executable that behaves completely differently from the one in the C standard.
+Then, all `atoi` function calls from any modules (even function calls within
+libc) are redirected to your function instead of the one in libc, which will
+very likely cause a problem. That is a somewhat surprising consequence for an
+accidental name conflict. On the other hand, this semantic is sometimes useful
+because it allows users to override library functions without rebuilding
+modules containing them.
+
+Whether good or bad, you should keep these semantics in mind to understand
+Unix linkers' behaviors.
+
+### Build reproducibility
+
+`mold`'s output is deterministic. That is, if you pass the same object files
+and the same command-line options to the same version of `mold`, it is
+guaranteed that `mold` produces the bit-for-bit identical output. The linker's
+internal randomness, such as the timing of thread scheduling or iteration
+orders of hash tables, doesn't affect the output.
+
+`mold` does not have any host-specific default settings. This is contrary to the
+GNU linkers, for which some configurable values, such as system-dependent
+library search paths, are hard-coded. `mold` depends only on its command-line
+arguments.
+
+## OPTION NOTATIONS
+
+Multi-letter long options may precede either a single dash or double dashes,
+except for those starting with the letter "o". For historical reasons, long
+options beginning with "o" must precede double dashes.
+
+For example, you can spell `--as-needed` as `-as-needed`, but `--omagic` must
+not be spelled as `-omagic`. `-omagic` will be interpreted not as `--omagic`
+but as `-o magic`.
+
+## MOLD-SPECIFIC OPTIONS
+
+* `--chroot`=_dir_:
+  Set _dir_ as the root directory.
+
+* `--color-diagnostics`=[ _auto_ | _always_ | _never_ ]:
+  Show diagnostic messages in color using ANSI escape sequences. `auto` means
+  that `mold` prints out messages in color only if the standard output is
+  connected to a TTY. Default is `auto`.
+
+* `--color-diagnostics`:
+  Synonym for `--color-diagnostics=auto`.
+
+* `--no-color-diagnostics`:
+  Synonym for `--color-diagnostics=never`.
+
+* `--detach`, `--no-detach`:
+  Permit or do not permit mold to create a debug info file in the background.
+
+* `--fork`, `--no-fork`:
+  Spawn a child process and let it do the actual linking. When linking a large
+  program, the OS kernel can take a few hundred milliseconds to terminate a
+  `mold` process. `--fork` hides that latency. By default, it does fork.
+
+* `--perf`:
+  Print performance statistics.
+
+* `--print-dependencies`:
+  Print out dependency information for input files.
+
+  Each line of the output for this option shows which file depends on which
+  file to use a specific symbol. This option is useful for debugging why some
+  object file in a static archive got linked or why some shared library is
+  kept in an output file's dependency list even with `--as-needed`.
+
+* `--relocatable-merge-sections`:
+  By default, `mold` doesn't merge input sections by name when merging input
+  object files into a single output object file for `-r`. For example,
+  `.text.foo` and `.text.bar` aren't merged for `-r` even though they are
+  merged into `.text` based on the default section merging rules.
+
+  This option changes the behavior so that `mold` merges input sections by
+  name by the default section merging rules.
+
+* `--repro`:
+  Archive input files, as well as a text file containing command line options,
+  in a tar file so that you can run `mold` with the exact same inputs again.
+  This is useful for reporting a bug with a reproducer. The output filename is
+  `path/to/output.tar`, where `path/to/output` is an output filename specified
+  by `-o`.
+
+* `--reverse-sections`:
+  Reverse the order of input sections before assigning them the offsets in the
+  output file.
+
+  This option is useful for finding bugs that depend on the initialization
+  order of global objects. In C++, constructors of global objects in a single
+  source file are guaranteed to be executed in the source order, but there's
+  no such guarantee across compilation units. Usually, constructors are
+  executed in the order given to the linker, but depending on it is a mistake.
+
+  By reversing the order of input sections using `--reverse-sections`, you can
+  easily test that your program works in the reversed initialization order.
+
+* `--run` _command_ _arg_...:
+  Run _command_ with `mold` as `/usr/bin/ld`. Specifically, `mold` runs a
+  given command with the `LD_PRELOAD` environment set to intercept exec(3)
+  family functions and replaces `argv[0]` with itself if it is `ld`,
+  `ld.gold`, or `ld.lld`.
+
+* `--separate-debug-file`, `--separate-debug-file`=_file_:
+  Bundle debug info sections into a separate file instead of embedding them in
+  an output executable or a shared library. mold creates a debug info file in
+  the background by default, so that you can start running your executable as
+  soon as possible.
+
+  By default, the debug info file is created in the same directory as is the
+  output file, with the `.dbg` file extension. That filename is embedded into
+  the output file so that `gdb` can automatically find the debug info file for
+  the output file. For more info about gdb features related to separate debug
+  files, see
+  <https://sourceware.org/gdb/current/onlinedocs/gdb.html/Separate-Debug-Files.html>.
+
+  mold holds a file lock with flock(2) while creating a debug info file in the
+  background.
+
+  If you don't want to create a debug info file in the background, pass the
+  `--no-detach` option.
+
+* `--shuffle-sections`, `--shuffle-sections`=_number_:
+  Randomize the output by shuffling the order of input sections before
+  assigning them the offsets in the output file. If a _number_ is given, it's
+  used as a seed for the random number generator, so that the linker produces
+  the same output for the same seed. If no seed is given, a random number is
+  used as a seed.
+
+  This option is useful for benchmarking. Modern CPUs are sensitive to a
+  program's memory layout. A seemingly benign change in program layout, such
+  as a small size increase of a function in the middle of a program, can
+  affect the program's performance. Therefore, even if you write new code and
+  get a good benchmark result, it is hard to say whether the new code improves
+  the program's performance; it is possible that the new memory layout happens
+  to perform better.
+
+  By running a benchmark multiple times with randomized memory layouts using
+  `--shuffle-sections`, you can isolate your program's real performance number
+  from the randomness caused by memory layout changes.
+
+* `--spare-program-headers`=_number_:
+  Append the given number of `PT_NULL` entries to the end of the program
+  header, so that post-link processing tools can easily add new segments by
+  overwriting the null entries.
+
+  Note that ELF requires all `PT_LOAD` segments to be sorted by `p_vaddr`.
+  Therefore, if you add a new LOAD segment, you may need to sort the entire
+  program header.
+
+* `--stats`:
+  Print input statistics.
+
+* `--thread-count`=_count_:
+  Use _count_ number of threads.
+
+* `--threads`, `--no-threads`:
+  Use multiple threads. By default, `mold` uses as many threads as the number of
+  cores or 32, whichever is smaller. The reason it is capped at 32 is because
+  `mold` doesn't scale well beyond that point. To use only one thread, pass
+  `--no-threads` or `--thread-count=1`.
+
+* `--quick-exit`, `--no-quick-exit`:
+  Use or do not use `quick_exit` to exit.
+
+* `-z rewrite-endbr`, `-z norewrite-endbr`:
+  As a security measure, some CPU instruction sets have recently gained a
+  feature to protect control flow integrity by disallowing indirect branches
+  by default. If the feature is enabled, the instruction that is executed
+  immediately after an indirect branch must be an branch target marker
+  instruction, or a CPU-level fault will raise. The marker instruction is also
+  known as "landing pad" instruction, to which indirect branches can land.
+  This feature makes ROP attacks harder to conduct.
+
+  To use the feature, a function whose pointer is taken needs to begin with a
+  landing pad because a function call via a function pointer is compiled to an
+  indirect branch. On the other hand, if a function is called only directly
+  (i.e. referred to only by _direct_ branch instructions), it doesn't have to
+  begin with it.
+
+  By default, the compiler always emits a landing pad at the beginning of each
+  global function because it doesn't know whether or not the function's
+  pointer is taken in another translation unit. As a result, the resulting
+  binary has more attack surface than necessary.
+
+  If `--rewrite-endbr` is given, mold conducts a whole program analysis
+  to identify functions whose addresses are actually taken and rewrites
+  landing pads with no-ops for non-address-taken functions, reducing the
+  attack surface.
+
+  This feature is currently available only on x86-64.
+
+## GNU-COMPATIBLE OPTIONS
+
+* `--help`:
+  Report usage information to stdout and exit.
+
+* `-v`, `--version`:
+  Report version information to stdout.
+
+* `-V`:
+  Report version and target information to stdout.
+
+* `-E`, `--export-dynamic`, `--no-export-dynamic`:
+  When creating an executable, using the `-E` option causes all global symbols
+  to be put into the dynamic symbol table, so that the symbols are visible
+  from other ELF modules at runtime.
+
+  By default, or if `--no-export-dynamic` is given, only symbols that are
+  referenced by DSOs at link-time are exported from an executable.
+
+* `-F` _libname_, `--filter`=_libname_:
+  Set the `DT_FILTER` dynamic section field to _libname_.
+
+* `-I` _file_, `--dynamic-linker`=_file_, `--no-dynamic-linker`:
+  Set the dynamic linker path to _file_. If no `-I` option is given, or if
+  `--no-dynamic-linker` is given, no dynamic linker path is set to an output
+  file. This is contrary to the GNU linkers which set a default dynamic linker
+  path in that case. This difference doesn't usually make any difference
+  because the compiler driver always passes `-I` to the linker.
+
+* `-L` _dir_, `--library-path`=_dir_:
+  Add _dir_ to the list of library search paths from which `mold` searches
+  libraries for the `-l` option.
+
+  Unlike the GNU linkers, `mold` does not have default search paths. This
+  difference doesn't usually make any difference because the compiler driver
+  always passes all necessary search paths to the linker.
+
+* `-M`, `--print-map`:
+  Write a map file to stdout.
+
+* `-N`, `--omagic`, `--no-omagic`:
+  Force `mold` to emit an output file with an old-fashioned memory layout.
+  First, it makes the first data segment not aligned to a page boundary.
+  Second, text segments are marked as writable if the option is given.
+
+* `-S`, `--strip-debug`:
+  Omit `.debug_*` sections from the output file.
+
+* `-T` _file_, `--script`=_file_:
+  Read linker script from _file_.
+
+* `-X`, `--discard-locals`:
+  Discard temporary local symbols to reduce the sizes of the symbol table and
+  the string table. Temporary local symbols are local symbols starting with
+  `.L`. Compilers usually generate such symbols for unnamed program elements
+  such as string literals or floating-point literals.
+
+* `-e` _symbol_, `--entry`=_symbol_:
+
+  Use _symbol_ as the entry point symbol instead of the default entry
+  point symbol _start.
+
+* `-f` _shlib_, `--auxiliary`=_shlib_:
+  Set the `DT_AUXILIARY` dynamic section field to _shlib_.
+
+* `-h` _libname_, `--soname`=_libname_:
+  Set the `DT_SONAME` dynamic section field to _libname_. This option is used
+  when creating a shared object file. Typically, when you create `libfoo.so`,
+  you want to pass `--soname=foo` to a linker.
+
+* `-l` _libname_:
+  Search for `lib`_libname_`.so` or `lib`_libname_`.a` from library search
+  paths.
+
+* `-m` _target_:
+  Choose a _target_.
+
+* `-o` _file_, `--output`=_file_:
+  Use _file_ as the output file name instead of the default name `a.out`.
+
+* `-r`, `--relocatable`:
+  Instead of generating an executable or a shared object file, combine input
+  object files to generate another object file that can be used as an input to
+  a linker.
+
+* `-s`, `--strip-all`:
+  Omit `.symtab` section from the output file.
+
+* `-u` _symbol_, `--undefined`=_symbol_:
+  If _symbol_ remains as an undefined symbol after reading all object files,
+  and if there is a static archive that contains an object file defining
+  _symbol_, pull out the object file and link it so that the output file
+  contains a definition of _symbol_.
+
+* `-y` _symbol_, `--trace-symbol`=_symbol_:
+  Trace references to _symbol_.
+
+* `--Bdynamic`:
+  Link against shared libraries.
+
+* `--Bstatic`:
+  Do not link against shared libraries.
+
+* `--Bsymbolic`:
+  When creating a shared library, make global symbols export-only (i.e. do not
+  import the same symbol). As a result, references within a shared library are
+  always resolved locally, negating symbol override at runtime. See "Dynamic
+  symbol resolution" for more information about symbol imports and exports.
+
+* `--Bsymbolic-functions`:
+  This option has the same effect as `--Bsymbolic` but works only for function
+  symbols. Data symbols remain being both imported and exported.
+
+* `--Bsymbolic-non-weak`:
+  This option has the same effect as `--Bsymbolic` but works only for non-weak
+  symbols. Weak symbols remain being both imported and exported.
+
+* `--Bsymbolic-non-weak-functions`:
+  This option has the same effect as `--Bsymbolic` but works only for non-weak
+  function symbols. Data symbols and weak function symbols remain being both
+  imported and exported.
+
+* `--Bno-symbolic`:
+  Cancel `--Bsymbolic`, `--Bsymbolic-functions`, `--Bsymbolic-non-weak` and
+  `--Bsymbolic-non-weak-functions`.
+
+* `--Map`=_file_:
+  Write map file to _file_.
+
+* `--Tbss`=_address_:
+  Alias for `--section-start=.bss=`_address_.
+
+* `--Tdata`=_address_:
+  Alias for `--section-start=.data=`_address_.
+
+* `--Ttext`=_address_:
+  Alias for `--section-start=.text=`_address_.
+
+* `--allow-multiple-definition`:
+  Normally, the linker reports an error if there are more than one definition
+  of a symbol. This option changes the default behavior so that it doesn't
+  report an error for duplicate definitions and instead use the first
+  definition.
+
+* `--allow-shlib-undefined`, `--no-allow-shlib-undefined`:
+  Even if mold succeeds in linking a main executable without undefined symbol
+  errors, you may still encounter symbol lookup errors at runtime because the
+  dynamic linker cannot find some symbols in shared libraries in any ELF
+  module. This occurs because mold ignores undefined symbols in shared
+  libraries by default.
+
+  If you pass `--no-allow-shlib-undefined`, mold verifies that undefined
+  symbols in shared libraries given to the linker can be resolved at
+  link-time. In other words, this converts the runtime error to a link-time
+  error.
+
+  Note that you need to pass all shared libraries, including indirectly
+  dependent ones, to the linker as arguments for `-l`. If a shared library
+  depends on a library that's not passed to the linker, the verification will
+  be skipped for that file.
+
+* `--as-needed`, `--no-as-needed`:
+  By default, shared libraries given to the linker are unconditionally added
+  to the list of required libraries in an output file. However, shared
+  libraries after `--as-needed` are added to the list only when at least one
+  symbol is actually used by the output file. In other words, shared libraries
+  after `--as-needed` are not added to the list of needed libraries if they
+  are not needed by a program.
+
+  The `--no-as-needed` option restores the default behavior for subsequent
+  files.
+
+* `--build-id`=[ `md5` | `sha1` | `sha256` | `fast` | `uuid` | `0x`_hexstring_ | `none` ]:
+  Create a `.note.gnu.build-id` section containing a byte string to uniquely
+  identify an output file. `sha256` compute a 256-bit cryptographic hash of an
+  output file and set it to build-id. `md5` and `sha1` compute the same hash
+  but truncate it to 128 and 160 bits, respectively, before setting it to
+  build-id. `uuid` sets a random 128-bit UUID. `0x`_hexstring_ sets
+  _hexstring_. `fast` is a synonym for `sha256`.
+
+* `--build-id`:
+  Synonym for `--build-id=sha256`.
+
+* `--no-build-id`:
+  Synonym for `--build-id=none`.
+
+* `--compress-debug-sections`=[ `zlib` | `zlib-gabi` | `zstd` | `none` ]:
+  Compress DWARF debug info (`.debug_*` sections) using the zlib or zstd
+  compression algorithm. `zlib-gabi` is an alias for `zlib`.
+
+* `--defsym`=_symbol_=_value_:
+  Define _symbol_ as an alias for _value_.
+
+  _value_ is either an integer (in decimal or hexadecimal with `0x` prefix) or
+  a symbol name. If an integer is given as a value, _symbol_ is defined as an
+  absolute symbol with the given value.
+
+* `--default-symver`:
+  Use soname as a symbol version and append that version to all symbols.
+
+* `--demangle`, `--no-demangle`:
+  Demangle C++ and Rust symbols in log messages.
+
+* `--dependency-file`=_file_:
+  Write a dependency file to _file_. The contents of the written file is
+  readable by make(1), which defines only one rule with the linker's output
+  file as a target and all input files as its prerequisites. Users are
+  expected to include the generated dependency file into a Makefile to
+  automate the dependency management. This option is analogous to the
+  compiler's `-MM -MF` options.
+
+* `--dynamic-list`=_file_:
+  Read a list of dynamic symbols from _file_. Same as
+  `--export-dynamic-symbol-list`, except that it implies `--Bsymbolic`. If
+  _file_ does not exist in the current directory, it is searched from library
+  search paths for the sake of compatibility with GNU ld.
+
+* `--eh-frame-hdr`, `--no-eh-frame-hdr`:
+  Create `.eh_frame_hdr` section.
+
+* `--emit-relocs`:
+  The linker usually "consumes" relocation sections. That is, the linker
+  applies relocations to other sections, and relocation sections themselves
+  are discarded.
+
+  The `--emit-relocs` instructs the linker to leave relocation sections in the
+  output file. Some post-link binary analysis or optimization tools such as
+  LLVM Bolt need them.
+
+* `--enable-new-dtags`, `--disable-new-dtags`:
+  By default, `mold` emits `DT_RUNPATH` for `--rpath`. If you pass
+  `--disable-new-dtags`, `mold` emits `DT_RPATH` for `--rpath` instead.
+
+* `--execute-only`:
+
+  Traditionally, setting the executable bit to 1 for a memory page implies
+  that the page also become readable, which allows machine code to be read
+  as data at runtime. That is actually what an attacker often does after
+  gaining a limited control of a process to find pieces of machine code
+  they can use to gain the full control of the process. As a mitigation,
+  recent processors including some ARM64 ones allows "execute-only" pages.
+  If a page is execute-only, you can call a function there as long as you
+  know its address but can't read it as data.
+
+  This option marks text segments as execute-only by setting just the "X"
+  bit instead of "RX". Note that on most systems, the absence of the "R"
+  bit in the text segment serves just as a hint. If you run a program
+  linked with `--execute-only` on a processor that doesn't support
+  execute-only pages, your executable will likely still function normally,
+  but the text segment will remain readable.
+
+* `--exclude-libs`=_libraries_ ...:
+  Mark all symbols in the given _libraries_ hidden.
+
+* `--export-dynamic-symbol`=_symbol_:
+  Put symbols matching _symbol_ in the dynamic symbol table. _symbol_ may be a
+  glob pattern in the same syntax as for the `--export-dynamic-symbol-list` or
+  `--version-script` options.
+
+* `--export-dynamic-symbol-list`=_file_:
+  Read a list of dynamic symbols from _file_.
+
+* `--fatal-warnings`, `--no-fatal-warnings`:
+  Treat warnings as errors.
+
+* `--fini`=_symbol_:
+  Call _symbol_ at unload-time.
+
+* `--gc-sections`, `--no-gc-sections`:
+  Remove unreferenced sections.
+
+* `--gdb-index`:
+  Create a `.gdb_index` section to speed up GNU debugger. To use this, you
+  need to compile source files with the `-ggnu-pubnames` compiler flag.
+
+* `--hash-style`=[ `sysv` | `gnu` | `both` | `none` ]:
+  Set hash style.
+
+* `--icf`=[ `safe` | `all` | `none` ], `--no-icf`:
+  It is not uncommon for a program to contain many identical functions that
+  differ only in name. For example, a C++ template `std::vector` is very
+  likely to be instantiated to the identical code for `std::vector<int>` and
+  `std::vector<unsigned>` because the container cares only about the size of
+  the parameter type. Identical Code Folding (ICF) is a size optimization to
+  identify and merge such identical functions.
+
+  If `--icf=all` is given, `mold` tries to merge all identical functions. This
+  reduces the size of the output most, but it is not a "safe" optimization. It
+  is guaranteed in C and C++ that two pointers pointing two different
+  functions will never be equal, but `--icf=all` breaks that assumption as two
+  identical functions have the same address after merging. So a care must be
+  taken when you use this flag that your program does not depend on the
+  function pointer uniqueness.
+
+  `--icf=safe` is a flag to merge functions only when it is safe to do so.
+  That is, if a program does not take an address of a function, it is safe to
+  merge that function with other function, as you cannot compare a function
+  pointer with something else without taking an address of a function.
+
+  `--icf=safe` needs to be used with a compiler that supports `.llvm_addrsig`
+  section which contains the information as to what symbols are address-taken.
+  LLVM/Clang supports that section by default. Since GCC does not support it
+  yet, you cannot use `--icf=safe` with GCC (it doesn't do any harm but can't
+  optimize at all.)
+
+  `--icf=none` and `--no-icf` disables ICF.
+
+* `--ignore-data-address-equality`:
+  Make ICF to merge not only functions but also data. This option should be
+  used in combination with `--icf=all`.
+
+* `--image-base`=_addr_:
+  Set the base address to _addr_.
+
+* `--init`=_symbol_:
+  Call _symbol_ at load-time.
+
+* `--no-undefined`:
+  Report undefined symbols (even with `--shared`).
+
+* `--noinhibit-exec`:
+  Create an output file even if errors occur.
+
+* `--package-metadata`=_percent-encoded-string_:
+  Embed a specified string into the `.note.package` section. This option
+  is designed for build scripts that generate binary packages, such as
+  `.rpm` or `.deb`, to include package metadata in each executable. It
+  simplifies the process of identifying the corresponding package for a
+  given executable or core file.
+
+  An argument to this option is treated as percent-encoded and decoded
+  before being inserted into the section, allowing you to avoid the use of
+  the comma (`,`) character in the argument. This is useful because the
+  compiler replaces all occurrences of commas in `-Wl,` with spaces before
+  forwarding them to the linker. Note that `mold` always interprets the
+  argument as percent-encoded, so you also need to escape all occurrences
+  of `%` as `%25`.
+
+* `--pack-dyn-relocs`=[ `relr` | `none` ]:
+  If `relr` is specified, all `R_*_RELATIVE` relocations are put into
+  `.relr.dyn` section instead of `.rel.dyn` or `.rela.dyn` section. Since
+  `.relr.dyn` section uses a space-efficient encoding scheme, specifying this
+  flag can reduce the size of the output. This is typically most effective for
+  position-independent executable.
+
+  Note that a runtime loader has to support `.relr.dyn` to run executables or
+  shared libraries linked with `--pack-dyn-relocs=relr`. As of 2022, only
+  ChromeOS, Android and Fuchsia support it.
+
+* `--pie`, `--pic-executable`, `--no-pie`, `--no-pic-executable`:
+  Create a position-independent executable.
+
+* `--print-gc-sections`, `--no-print-gc-sections`:
+  Print removed unreferenced sections.
+
+* `--print-icf-sections`, `--no-print-icf-sections`:
+  Print folded identical sections.
+
+* `--push-state`, `--pop-state`:
+  `--push-state` saves the current values of `--as-needed`, `--whole-archive`,
+  `--static`, and `--start-lib`. The saved values can be restored by
+  pop-state.
+
+  `--push-state` and `--pop-state` pairs can nest.
+
+  These options are useful when you want to construct linker command line
+  options programmatically. For example, if you want to link `libfoo.so` by
+  as-needed basis but don't want to change the global state of `--as-needed`,
+  you can append `--push-state --as-needed -lfoo --pop-state` to the linker
+  command line options.
+
+* `--relax, --no-relax`:
+  Rewrite machine instructions with more efficient ones for some relocations.
+  The feature is enabled by default.
+
+* `--require-defined`=_symbol_:
+  Like `--undefined`, except the new symbol must be defined by the end of the
+  link.
+
+* `--retain-symbols-file`=_file_:
+  Keep only symbols listed in _file_. _file_ is a text file containing a
+  symbol name on each line. `mold` discards all local symbols as well as
+  global symbol that are not in _file_. Note that this option removes symbols
+  only from `.symtab` section and does not affect `.dynsym` section, which is
+  used for dynamic linking.
+
+* `--rpath`=_dir_:
+  Add _dir_ to runtime search paths.
+
+* `--section-start`=_section_=_address_:
+  Set _address_ to section. _address_ is a hexadecimal number that may start
+  with an optional `0x`.
+
+* `--shared`, `--Bshareable`:
+  Create a share library.
+
+* `--spare-dynamic-tags`=_number_:
+  Append the given number of `DT_NULL` entries to the end of the `.dynamic`
+  section, so that post-link processing tools can easily add new dynamic tags
+  by overwriting the null entries.
+
+* `--start-lib`, `--end-lib`:
+  Handle object files between `--start-lib` and `--end-lib` as if they were in
+  an archive file. That means object files between them are linked only when
+  they are needed to resolve undefined symbols. The options are useful if you
+  want to link object files only when they are needed but want to avoid the
+  overhead of running ar(3).
+
+* `--static`:
+  Do not link against shared libraries.
+
+* `--sysroot`=_dir_:
+  Set target system root directory to _dir_.
+
+* `--trace`:
+  Print name of each input file.
+
+* `--undefined-glob`=_pattern_:
+  Synonym for `--undefined`, except that `--undefined-glob` takes a glob
+  pattern instead of just a single symbol name.
+
+* `--undefined-version`, `--no-undefined-version`:
+  By default, `mold` warns on a symbol specified by a version script or by
+  `--export-dynamic-symbol` if it is not defined. You can silence the warning
+  by `--undefined-version`.
+
+* `--unique`=_pattern_:
+  Don't merge input sections that match the given glob pattern _pattern_.
+
+* `--unresolved-symbols`=[ `report-all` | `ignore-all` | `ignore-in-object-files` | `ignore-in-shared-libs` ]:
+  How to handle undefined symbols.
+
+* `--version-script`=_file_:
+  Read version script from _file_. If _file_ does not exist in the current
+  directory, it is searched from library search paths for the sake of
+  compatibility with GNU ld.
+
+* `--warn-common`, `--no-warn-common`:
+  Warn about common symbols.
+
+* `--warn-once`:
+  Only warn once for each undefined symbol instead of warn for each relocation
+  referring an undefined symbol.
+
+* `--warn-unresolved-symbols`, `--error-unresolved-symbols`:
+  Normally, the linker reports an error for unresolved symbols.
+  `--warn-unresolved-symbols` option turns it into a warning.
+  `--error-unresolved-symbols` option restores the default behavior.
+
+* `--whole-archive`, `--no-whole-archive`:
+  When archive files (`.a` files) are given to the linker, only object files
+  that are needed to resolve undefined symbols are extracted from them and
+  linked to an output file. `--whole-archive` changes that behavior for
+  subsequent archives so that the linker extracts all object files and links
+  them to an output. For example, if you are creating a shared object file and
+  you want to include all archive members to the output, you should pass
+  `--whole-archive`. `--no-whole-archive` restores the default behavior for
+  subsequent archives.
+
+* `--wrap`=_symbol_:
+  Make _symbol_ be resolved to `__wrap_`_symbol_. The original symbol can be
+  resolved as `__real_`_symbol_. This option is typically used for wrapping an
+  existing function.
+
+* `-z cet-report`=[ `warning` | `error` | `none` ]:
+  Intel Control-flow Enforcement Technology (CET) is a new x86 feature
+  available since Tiger Lake which is released in 2020. It defines new
+  instructions to harden security to protect programs from control hijacking
+  attacks. You can tell the compiler to use the feature by specifying the
+  `-fcf-protection` flag.
+
+  `-z cet-report` flag is used to make sure that all object files were
+  compiled with a correct `-fcf-protection` flag. If `warning` or `error` are
+  given, `mold` prints out a warning or an error message if an object file was
+  not compiled with the compiler flag.
+
+  `mold` looks for `GNU_PROPERTY_X86_FEATURE_1_IBT` bit and
+  `GNU_PROPERTY_X86_FEATURE_1_SHSTK` bit in `.note.gnu.property` section to
+  determine whether or not an object file was compiled with `-fcf-protection`.
+
+* `-z now`, `-z lazy`:
+  By default, functions referring to other ELF modules are resolved by the
+  dynamic linker when they are called for the first time. `-z now` marks an
+  executable or a shared library file so that all dynamic symbols are resolved
+  when a file is loaded to memory. `-z lazy` restores the default behavior.
+
+* `-z origin`:
+  Mark object requiring immediate `$ORIGIN` processing at runtime.
+
+* `-z ibt`:
+  Turn on `GNU_PROPERTY_X86_FEATURE_1_IBT` bit in `.note.gnu.property` section
+  to indicate that the output uses IBT-enabled PLT. This option implies `-z
+  ibtplt`.
+
+* `-z ibtplt`:
+  Generate Intel Branch Tracking (IBT)-enabled PLT which is the default on
+  x86-64. This is the default.
+
+* `-z execstack`, `-z noexecstack`:
+  By default, the pages for the stack area (i.e. the pages where local
+  variables reside) are not executable for security reasons. `-z execstack`
+  makes it executable. `-z noexecstack` restores the default behavior.
+
+* `-z keep-text-section-prefix`, `-z nokeep-text-section-prefix`:
+  Keep `.text.hot`, `.text.unknown`, `.text.unlikely`, `.text.startup`, and
+  `.text.exit` as separate sections in the final binary instead of merging
+  them as `.text`.
+
+* `-z rodynamic`:
+  Make the `.dynamic` section read-only.
+
+* `-z relro`, `-z norelro`:
+  Some sections such as `.dynamic` have to be writable only during a module is
+  being loaded to memory. Once the dynamic linker finishes its job, such
+  sections won't be mutated by anyone. As a security mitigation, it is
+  preferred to make such segments read-only during program execution.
+
+  `-z relro` puts such sections into a special segment called `relro`. The
+  dynamic linker makes a relro segment read-only after it finishes its job.
+
+  By default, `mold` generates a relro segment. `-z norelro` disables the
+  feature.
+
+* `-z sectionheader`, `-z nosectionheader`:
+  `-z nosectionheader` tell the linker to omit the section header.
+  By default, the linker does not omit the section header.
+
+* `-z separate-loadable-segments`, `-z separate-code`, `-z noseparate-code`:
+  If one memory page contains multiple segments, the page protection bits are
+  set in such a way that the needed attributes (writable or executable) are
+  satisfied for all segments. This usually happens at a boundary of two
+  segments with two different attributes.
+
+  `separate-loadable-segments` adds paddings between segments with different
+  attributes so that they do not share the same page. This is the default.
+
+  `separate-code` adds paddings only between executable and non-executable
+  segments.
+
+  `noseparate-code` does not add any paddings between segments.
+
+* `-z defs`, `-z nodefs`:
+  Report undefined symbols (even with `--shared`).
+
+* `-z shstk`:
+  Enforce shadow stack by turning `GNU_PROPERTY_X86_FEATURE_1_SHSTK` bit in
+  `.note.gnu.property` output section. Shadow stack is part of Intel
+  Control-flow Enforcement Technology (CET), which is available since Tiger
+  Lake (2020).
+
+* `-z start_stop_visibility`=[ `hidden` | `protected` ]:
+  If a section name is valid as a C identifier (i.e., it matches
+  `/^[_a-zA-Z][_a-zA-Z0-9]*$/`), mold creates `__start_SECNAME` and
+  `__stop_SECNAME` symbols to mark the beginning and end of the section,
+  where `SECNAME` is the section name.
+
+  You can make these marker symbols visible from other ELF modules by passing
+  `-z start_stop_visibility=protected`. Default is `hidden`.
+
+* `-z text`, `-z notext`, `-z textoff`:
+  `mold` by default reports an error if dynamic relocations are created in
+  read-only sections. If `-z notext` or `-z textoff` are given, `mold` creates
+  such dynamic relocations without reporting an error. `-z text` restores the
+  default behavior.
+
+* `-z max-page-size`=_number_:
+  Some CPU ISAs support multiple memory page sizes. This option specifies the
+  maximum page size that an output binary can run on. In general, binaries
+  built for a larger page size can run on a system with a smaller page size,
+  but not vice versa. The default value is 4 KiB for i386, x86-64, and RISC-V,
+  and 64 KiB for ARM64.
+
+* `-z nodefaultlib`:
+  Make the dynamic loader ignore default search paths.
+
+* `-z nodelete`:
+  Mark DSO non-deletable at runtime.
+
+* `-z nodlopen`:
+  Mark DSO not available to dlopen(3). This option makes it possible for the
+  linker to optimize thread-local variable accesses by rewriting instructions
+  for some targets.
+
+* `-z nodump`:
+  Mark DSO not available to dldump(3).
+
+* `-z nocopyreloc`:
+  Do not create copy relocations.
+
+* `-z initfirst`:
+  Mark DSO to be initialized first at runtime.
+
+* `-z interpose`:
+  Mark object to interpose all DSOs but executable.
+
+* `-(`, `-)`, `-EL`, `-O`_number_, `--dc`, `--dp`, `--end-group`, `--no-add-needed`, `--no-copy-dt-needed-entries`, `--nostdlib`, `--rpath-link=Ar dir`, `--sort-common`, `--sort-section`, `--start-group`, `--warn-constructors`, `--warn-once`, `--fix-cortex-a53-835769`, `--fix-cortex-a53-843419`, `-z combreloc`, `-z common-page-size`, `-z nocombreloc`:
+  Ignored
+
+## ENVIRONMENT VARIABLES
+
+* `MOLD_JOBS`:
+  If this variable is set to `1`, only one `mold` process will run at a time.
+  If a new mold process is initiated while another is already active, the new
+  process will wait until the active one completes before starting.
+
+  The primary reason for this environment variable is to minimize peak memory
+  usage. Since mold is designed to operate with high parallelism, running
+  multiple mold instances simultaneously may not be beneficial. If you execute
+  N instances of mold concurrently, it could require N times the time and N
+  times the memory. On the other hand, running them one after the other might
+  still take N times longer, but the peak memory usage would be the same as
+  running just a single instance.
+
+  If your build system invokes multiple linker processes simultaneously and
+  some of them often get killed due to out-of-memory errors, you might
+  consider setting this environment variable to `1` to see if it addresses the
+  OOM issue.
+
+  Currently, any value other than `1` is silently ignored.
+
+* `MOLD_DEBUG`:
+  If this variable is set to a non-empty string, `mold` embeds its
+  command-line options in the output file's `.comment` section.
+
+* `MOLD_REPRO`:
+  Setting this variable to a non-empty string has the same effect as passing
+  the `--repro` option.
+
+## SEE ALSO
+
+gold(1), ld(1), elf(5), ld.so(8)
+
+## AUTHOR
+
+Rui Ueyama <ruiu@cs.stanford.edu>
+
+## BUGS
+
+Report bugs to <https://github.com/rui314/mold/issues>.

elf/arch-arm64.cc

@@ -1,794 +0,0 @@
-#include "mold.h"
-
-#include <tbb/parallel_for.h>
-#include <tbb/parallel_for_each.h>
-
-namespace mold::elf {
-
-using E = ARM64;
-
-static void write_adr(u8 *buf, u64 val) {
-  u32 hi = (val & 0x1ffffc) << 3;
-  u32 lo = (val & 3) << 29;
-  *(u32 *)buf = (*(u32 *)buf & 0x9f00001f) | hi | lo;
-}
-
-// Returns [hi:lo] bits of val.
-static u64 bits(u64 val, u64 hi, u64 lo) {
-  return (val >> lo) & (((u64)1 << (hi - lo + 1)) - 1);
-}
-
-static u64 page(u64 val) {
-  return val & ~(u64)0xfff;
-}
-
-template <>
-void GotPltSection<E>::copy_buf(Context<E> &ctx) {
-  u64 *buf = (u64 *)(ctx.buf + this->shdr.sh_offset);
-
-  // The first slot of .got.plt points to _DYNAMIC.
-  buf[0] = ctx.dynamic ? ctx.dynamic->shdr.sh_addr : 0;
-  buf[1] = 0;
-  buf[2] = 0;
-
-  for (Symbol<E> *sym : ctx.plt->symbols)
-    buf[sym->get_gotplt_idx(ctx)] = ctx.plt->shdr.sh_addr;
-}
-
-static void write_plt_header(Context<E> &ctx, u8 *buf) {
-  // Write PLT header
-  static const u8 plt0[] = {
-    0xf0, 0x7b, 0xbf, 0xa9, // stp    x16, x30, [sp,#-16]!
-    0x10, 0x00, 0x00, 0x90, // adrp   x16, .got.plt[2]
-    0x11, 0x02, 0x40, 0xf9, // ldr    x17, [x16, .got.plt[2]]
-    0x10, 0x02, 0x00, 0x91, // add    x16, x16, .got.plt[2]
-    0x20, 0x02, 0x1f, 0xd6, // br     x17
-    0x1f, 0x20, 0x03, 0xd5, // nop
-    0x1f, 0x20, 0x03, 0xd5, // nop
-    0x1f, 0x20, 0x03, 0xd5, // nop
-  };
-
-  u64 gotplt = ctx.gotplt->shdr.sh_addr + 16;
-  u64 plt = ctx.plt->shdr.sh_addr;
-
-  memcpy(buf, plt0, sizeof(plt0));
-  write_adr(buf + 4, bits(page(gotplt) - page(plt + 4), 32, 12));
-  *(u32 *)(buf + 8) |= bits(gotplt, 11, 3) << 10;
-  *(u32 *)(buf + 12) |= ((gotplt) & 0xfff) << 10;
-}
-
-static void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
-  u8 *ent = buf + ctx.plt_hdr_size + sym.get_plt_idx(ctx) * ctx.plt_size;
-
-  static const u8 data[] = {
-    0x10, 0x00, 0x00, 0x90, // adrp x16, .got.plt[n]
-    0x11, 0x02, 0x40, 0xf9, // ldr  x17, [x16, .got.plt[n]]
-    0x10, 0x02, 0x00, 0x91, // add  x16, x16, .got.plt[n]
-    0x20, 0x02, 0x1f, 0xd6, // br   x17
-  };
-
-  u64 gotplt = sym.get_gotplt_addr(ctx);
-  u64 plt = sym.get_plt_addr(ctx);
-
-  memcpy(ent, data, sizeof(data));
-  write_adr(ent, bits(page(gotplt) - page(plt), 32, 12));
-  *(u32 *)(ent + 4) |= bits(gotplt, 11, 3) << 10;
-  *(u32 *)(ent + 8) |= (gotplt & 0xfff) << 10;
-}
-
-template <>
-void PltSection<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + this->shdr.sh_offset;
-  write_plt_header(ctx, buf);
-  for (Symbol<E> *sym : symbols)
-    write_plt_entry(ctx, buf, *sym);
-}
-
-template <>
-void PltGotSection<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + this->shdr.sh_offset;
-
-  for (Symbol<E> *sym : symbols) {
-    u8 *ent = buf + sym->get_pltgot_idx(ctx) * ARM64::pltgot_size;
-
-    static const u8 data[] = {
-      0x10, 0x00, 0x00, 0x90, // adrp x16, GOT[n]
-      0x11, 0x02, 0x40, 0xf9, // ldr  x17, [x16, GOT[n]]
-      0x20, 0x02, 0x1f, 0xd6, // br   x17
-      0x1f, 0x20, 0x03, 0xd5, // nop
-    };
-
-    u64 got = sym->get_got_addr(ctx);
-    u64 plt = sym->get_plt_addr(ctx);
-
-    memcpy(ent, data, sizeof(data));
-    write_adr(ent, bits(page(got) - page(plt), 32, 12));
-    *(u32 *)(ent + 4) |= bits(got, 11, 3) << 10;
-  }
-}
-
-template <>
-void EhFrameSection<E>::apply_reloc(Context<E> &ctx, ElfRel<E> &rel,
-                                    u64 loc, u64 val) {
-  u8 *base = ctx.buf + this->shdr.sh_offset;
-
-  switch (rel.r_type) {
-  case R_AARCH64_ABS64:
-    *(u64 *)(base + loc) = val;
-    return;
-  case R_AARCH64_PREL32:
-    *(u32 *)(base + loc) = val - this->shdr.sh_addr - loc;
-    return;
-  case R_AARCH64_PREL64:
-    *(u64 *)(base + loc) = val - this->shdr.sh_addr - loc;
-    return;
-  }
-  Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
-}
-
-template <>
-void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
-  ElfRel<E> *dynrel = nullptr;
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-
-  i64 frag_idx = 0;
-
-  if (ctx.reldyn)
-    dynrel = (ElfRel<E> *)(ctx.buf + ctx.reldyn->shdr.sh_offset +
-                           file.reldyn_offset + this->reldyn_offset);
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_AARCH64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    const SectionFragmentRef<E> *frag_ref = nullptr;
-    if (rel_fragments && rel_fragments[frag_idx].idx == i)
-      frag_ref = &rel_fragments[frag_idx++];
-
-    auto overflow_check = [&](i64 val, i64 lo, i64 hi) {
-      if (val < lo || hi <= val)
-        Error(ctx) << *this << ": relocation " << rel << " against "
-                   << sym << " out of range: " << val << " is not in ["
-                   << lo << ", " << hi << ")";
-    };
-
-#define S   (frag_ref ? frag_ref->frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A   (frag_ref ? frag_ref->addend : rel.r_addend)
-#define P   (output_section->shdr.sh_addr + offset + rel.r_offset)
-#define G   (sym.get_got_addr(ctx) - ctx.got->shdr.sh_addr)
-#define GOT ctx.got->shdr.sh_addr
-
-    if (needs_dynrel[i]) {
-      *dynrel++ = {P, R_AARCH64_ABS64, (u32)sym.get_dynsym_idx(ctx), A};
-      *(u64 *)loc = A;
-      continue;
-    }
-
-    if (needs_baserel[i]) {
-      if (!is_relr_reloc(ctx, rel))
-        *dynrel++ = {P, R_AARCH64_RELATIVE, 0, (i64)(S + A)};
-      *(u64 *)loc = S + A;
-      continue;
-    }
-
-    switch (rel.r_type) {
-    case R_AARCH64_ABS64:
-      *(u64 *)loc = S + A;
-      continue;
-    case R_AARCH64_LDST8_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 0) << 10;
-      continue;
-    case R_AARCH64_LDST16_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 1) << 10;
-      continue;
-    case R_AARCH64_LDST32_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 2) << 10;
-      continue;
-    case R_AARCH64_LDST64_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 3) << 10;
-      continue;
-    case R_AARCH64_LDST128_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 4) << 10;
-      continue;
-    case R_AARCH64_ADD_ABS_LO12_NC:
-      *(u32 *)loc |= bits(S + A, 11, 0) << 10;
-      continue;
-    case R_AARCH64_MOVW_UABS_G0_NC:
-      *(u32 *)loc |= bits(S + A, 15, 0) << 5;
-      continue;
-    case R_AARCH64_MOVW_UABS_G1_NC:
-      *(u32 *)loc |= bits(S + A, 31, 16) << 5;
-      continue;
-    case R_AARCH64_MOVW_UABS_G2_NC:
-      *(u32 *)loc |= bits(S + A, 47, 32) << 5;
-      continue;
-    case R_AARCH64_MOVW_UABS_G3:
-      *(u32 *)loc |= bits(S + A, 63, 48) << 5;
-      continue;
-    case R_AARCH64_ADR_GOT_PAGE: {
-      i64 val = page(G + GOT + A) - page(P);
-      overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-      write_adr(loc, bits(val, 32, 12));
-      continue;
-    }
-    case R_AARCH64_ADR_PREL_PG_HI21: {
-      i64 val = page(S + A) - page(P);
-      overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-      write_adr(loc, bits(val, 32, 12));
-      continue;
-    }
-    case R_AARCH64_ADR_PREL_LO21: {
-      i64 val = S + A - P;
-      overflow_check(val, -((i64)1 << 20), (i64)1 << 20);
-      write_adr(loc, val);
-      continue;
-    }
-    case R_AARCH64_CALL26:
-    case R_AARCH64_JUMP26: {
-      if (sym.esym().is_undef_weak()) {
-        // On ARM, calling an weak undefined symbol jumps to the
-        // next instruction.
-        *(u32 *)loc |= 1;
-        continue;
-      }
-
-      i64 lo = -(1 << 27);
-      i64 hi = 1 << 27;
-      i64 val = S + A - P;
-
-      if (val < lo || hi <= val) {
-        RangeExtensionRef ref = range_extn[i];
-        val = output_section->thunks[ref.thunk_idx]->get_addr(ref.sym_idx) + A - P;
-        assert(lo <= val && val < hi);
-      }
-
-      *(u32 *)loc |= (val >> 2) & 0x3ffffff;
-      continue;
-    }
-    case R_AARCH64_CONDBR19: {
-      i64 val = S + A - P;
-      overflow_check(val, -((i64)1 << 20), (i64)1 << 20);
-      *(u32 *)loc |= bits(val, 20, 2) << 5;
-      continue;
-    }
-    case R_AARCH64_PREL16: {
-      i64 val = S + A - P;
-      overflow_check(val, -((i64)1 << 15), (i64)1 << 15);
-      *(u16 *)loc = val;
-      continue;
-    }
-    case R_AARCH64_PREL32: {
-      i64 val = S + A - P;
-      overflow_check(val, -((i64)1 << 31), (i64)1 << 32);
-      *(u32 *)loc = val;
-      continue;
-    }
-    case R_AARCH64_PREL64:
-      *(u64 *)loc = S + A - P;
-      continue;
-    case R_AARCH64_LD64_GOT_LO12_NC:
-      *(u32 *)loc |= bits(G + GOT + A, 11, 3) << 10;
-      continue;
-    case R_AARCH64_LD64_GOTPAGE_LO15: {
-      i64 val = G + GOT + A - page(GOT);
-      overflow_check(val, 0, 1 << 15);
-      *(u32 *)loc |= bits(val, 14, 3) << 10;
-      continue;
-    }
-    case R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21: {
-      i64 val = page(sym.get_gottp_addr(ctx) + A) - page(P);
-      overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-      write_adr(loc, bits(val, 32, 12));
-      continue;
-    }
-    case R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC:
-      *(u32 *)loc |= bits(sym.get_gottp_addr(ctx) + A, 11, 3) << 10;
-      continue;
-    case R_AARCH64_TLSLE_ADD_TPREL_HI12: {
-      i64 val = S + A - ctx.tls_begin + 16;
-      overflow_check(val, 0, (i64)1 << 24);
-      *(u32 *)loc |= bits(val, 23, 12) << 10;
-      continue;
-    }
-    case R_AARCH64_TLSLE_ADD_TPREL_LO12_NC:
-      *(u32 *)loc |= bits(S + A - ctx.tls_begin + 16, 11, 0) << 10;
-      continue;
-    case R_AARCH64_TLSGD_ADR_PAGE21: {
-      i64 val = page(sym.get_tlsgd_addr(ctx) + A) - page(P);
-      overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-      write_adr(loc, bits(val, 32, 12));
-      continue;
-    }
-    case R_AARCH64_TLSGD_ADD_LO12_NC:
-      *(u32 *)loc |= bits(sym.get_tlsgd_addr(ctx) + A, 11, 0) << 10;
-      continue;
-    case R_AARCH64_TLSDESC_ADR_PAGE21: {
-      if (ctx.relax_tlsdesc && !sym.is_imported) {
-        // adrp x0, 0 -> movz x0, #tls_ofset_hi, lsl #16
-        i64 val = (S + A - ctx.tls_begin + 16);
-        overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-        *(u32 *)loc = 0xd2a00000 | (bits(val, 32, 16) << 5);
-      } else {
-        i64 val = page(sym.get_tlsdesc_addr(ctx) + A) - page(P);
-        overflow_check(val, -((i64)1 << 32), (i64)1 << 32);
-        write_adr(loc, bits(val, 32, 12));
-      }
-      continue;
-    }
-    case R_AARCH64_TLSDESC_LD64_LO12:
-      if (ctx.relax_tlsdesc && !sym.is_imported) {
-        // ldr x2, [x0] -> movk x0, #tls_ofset_lo
-        u32 offset_lo = (S + A - ctx.tls_begin + 16) & 0xffff;
-        *(u32 *)loc = 0xf2800000 | (offset_lo << 5);
-      } else {
-        *(u32 *)loc |= bits(sym.get_tlsdesc_addr(ctx) + A, 11, 3) << 10;
-      }
-      continue;
-    case R_AARCH64_TLSDESC_ADD_LO12:
-      if (ctx.relax_tlsdesc && !sym.is_imported) {
-        // add x0, x0, #0 -> nop
-        *(u32 *)loc = 0xd503201f;
-      } else {
-        *(u32 *)loc |= bits(sym.get_tlsdesc_addr(ctx) + A, 11, 0) << 10;
-      }
-      continue;
-    case R_AARCH64_TLSDESC_CALL:
-      if (ctx.relax_tlsdesc && !sym.is_imported) {
-        // blr x2 -> nop
-        *(u32 *)loc = 0xd503201f;
-      }
-      continue;
-    default:
-      unreachable();
-    }
-
-#undef S
-#undef A
-#undef P
-#undef G
-#undef GOT
-  }
-}
-
-template <>
-void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-  i64 frag_idx = 0;
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_AARCH64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    SectionFragment<E> *frag;
-    i64 addend;
-    std::tie(frag, addend) = get_fragment(ctx, rel);
-
-#define S (frag ? frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A (frag ? addend : rel.r_addend)
-
-    switch (rel.r_type) {
-    case R_AARCH64_ABS64:
-      *(u64 *)loc = S + A;
-      continue;
-    case R_AARCH64_ABS32:
-      *(u32 *)loc = S + A;
-      continue;
-    default:
-      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
-                 << rel;
-      break;
-    }
-
-#undef S
-#undef A
-  }
-}
-
-template <>
-void InputSection<E>::scan_relocations(Context<E> &ctx) {
-  assert(shdr.sh_flags & SHF_ALLOC);
-
-  this->reldyn_offset = file.num_dynrel * sizeof(ElfRel<E>);
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-
-  // Scan relocations
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_AARCH64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    if (sym.get_type() == STT_GNU_IFUNC) {
-      sym.flags |= NEEDS_GOT;
-      sym.flags |= NEEDS_PLT;
-    }
-
-    switch (rel.r_type) {
-    case R_AARCH64_ABS64: {
-      Action table[][4] = {
-        // Absolute  Local    Imported data  Imported code
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // DSO
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // PIE
-        {  NONE,     NONE,    COPYREL,       PLT    },     // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_AARCH64_ADR_GOT_PAGE:
-    case R_AARCH64_LD64_GOT_LO12_NC:
-    case R_AARCH64_LD64_GOTPAGE_LO15:
-      sym.flags |= NEEDS_GOT;
-      break;
-    case R_AARCH64_CALL26:
-    case R_AARCH64_JUMP26:
-      if (sym.is_imported)
-        sym.flags |= NEEDS_PLT;
-      break;
-    case R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21:
-    case R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC:
-      sym.flags |= NEEDS_GOTTP;
-      break;
-    case R_AARCH64_ADR_PREL_PG_HI21: {
-      Action table[][4] = {
-        // Absolute  Local    Imported data  Imported code
-        {  NONE,     NONE,    ERROR,         ERROR },      // DSO
-        {  NONE,     NONE,    COPYREL,       PLT   },      // PIE
-        {  NONE,     NONE,    COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_AARCH64_TLSGD_ADR_PAGE21:
-      sym.flags |= NEEDS_TLSGD;
-      break;
-    case R_AARCH64_TLSDESC_ADR_PAGE21:
-    case R_AARCH64_TLSDESC_LD64_LO12:
-    case R_AARCH64_TLSDESC_ADD_LO12:
-      if (!ctx.relax_tlsdesc || sym.is_imported)
-        sym.flags |= NEEDS_TLSDESC;
-      break;
-    case R_AARCH64_ADD_ABS_LO12_NC:
-    case R_AARCH64_ADR_PREL_LO21:
-    case R_AARCH64_CONDBR19:
-    case R_AARCH64_LDST16_ABS_LO12_NC:
-    case R_AARCH64_LDST32_ABS_LO12_NC:
-    case R_AARCH64_LDST64_ABS_LO12_NC:
-    case R_AARCH64_LDST128_ABS_LO12_NC:
-    case R_AARCH64_LDST8_ABS_LO12_NC:
-    case R_AARCH64_MOVW_UABS_G0_NC:
-    case R_AARCH64_MOVW_UABS_G1_NC:
-    case R_AARCH64_MOVW_UABS_G2_NC:
-    case R_AARCH64_MOVW_UABS_G3:
-    case R_AARCH64_PREL16:
-    case R_AARCH64_PREL32:
-    case R_AARCH64_PREL64:
-    case R_AARCH64_TLSLE_ADD_TPREL_HI12:
-    case R_AARCH64_TLSLE_ADD_TPREL_LO12_NC:
-    case R_AARCH64_TLSGD_ADD_LO12_NC:
-    case R_AARCH64_TLSDESC_CALL:
-      break;
-    default:
-      Error(ctx) << *this << ": unknown relocation: " << rel;
-    }
-  }
-}
-
-static void reset_thunk(RangeExtensionThunk<E> &thunk) {
-  for (Symbol<E> *sym : thunk.symbols) {
-    sym->thunk_idx = -1;
-    sym->thunk_sym_idx = -1;
-    sym->flags &= (u8)~NEEDS_THUNK;
-  }
-}
-
-static bool is_reachable(Context<E> &ctx, Symbol<E> &sym,
-                         InputSection<E> &isec, const ElfRel<E> &rel) {
-  // We always create a thunk for an absolute symbol conservatively
-  // because `shrink_sections` may increase a distance between a
-  // branch instruction and an absolute symbol. Branching to an
-  // absolute location is extremely rare in real code, though.
-  if (sym.is_absolute())
-    return false;
-
-  // Compute a distance between the relocated place and the symbol
-  // and check if they are within reach.
-  i64 S = sym.get_addr(ctx);
-  i64 A = rel.r_addend;
-  i64 P = isec.get_addr() + rel.r_offset;
-  i64 val = S + A - P;
-  return -(1 << 27) <= val && val < (1 << 27);
-}
-
-// We create a thunk no further than 100 MiB from any section.
-constexpr i64 MAX_DISTANCE = 100 * 1024 * 1024;
-
-// We create a thunk for each 10 MiB input sections.
-static constexpr i64 GROUP_SIZE = 10 * 1024 * 1024;
-
-static void create_thunks(Context<E> &ctx, OutputSection<E> &osec) {
-  std::span<InputSection<E> *> members = osec.members;
-  members[0]->offset = 0;
-
-  // Initialize input sections with very large dummy offsets so that
-  // sections that have got real offsets are separated from the ones
-  // without in the virtual address space.
-  tbb::parallel_for((i64)1, (i64)members.size(), [&](i64 i) {
-    members[i]->offset = 1 << 31;
-  });
-
-  // We create thunks from the beginning of the section to the end.
-  // We manage progress using four offsets which increase monotonically.
-  // The locations they point to are always A <= B <= C <= D.
-  i64 a = 0;
-  i64 b = 0;
-  i64 c = 0;
-  i64 d = 0;
-  i64 offset = 0;
-
-  while (b < members.size()) {
-    // Move D foward as far as we can jump from B to D.
-    while (d < members.size() && offset - members[b]->offset < MAX_DISTANCE) {
-      offset = align_to(offset, members[d]->shdr.sh_addralign);
-      members[d]->offset = offset;
-      offset += members[d]->shdr.sh_size;
-      d++;
-    }
-
-    // Move C forward so that C is apart from B by GROUP_SIZE.
-    while (c < members.size() &&
-           members[c]->offset - members[b]->offset < GROUP_SIZE)
-      c++;
-
-    // Move A forward so that A is reachable from C.
-    if (c > 0) {
-      i64 c_end = members[c - 1]->offset + members[c - 1]->shdr.sh_size;
-      while (a < osec.thunks.size() &&
-             osec.thunks[a]->offset < c_end - MAX_DISTANCE)
-        reset_thunk(*osec.thunks[a++]);
-    }
-
-    // Create a thunk for input sections between B and C and place it at D.
-    osec.thunks.emplace_back(new RangeExtensionThunk<E>{osec});
-
-    RangeExtensionThunk<E> &thunk = *osec.thunks.back();
-    thunk.thunk_idx = osec.thunks.size() - 1;
-    thunk.offset = offset;
-
-    // Scan relocations between B and C to collect symbols that need thunks.
-    tbb::parallel_for_each(&members[b], &members[c], [&](InputSection<E> *isec) {
-      std::span<ElfRel<E>> rels = isec->get_rels(ctx);
-      isec->range_extn.resize(rels.size());
-
-      for (i64 i = 0; i < rels.size(); i++) {
-        const ElfRel<E> &rel = rels[i];
-        if (rel.r_type != R_AARCH64_CALL26 && rel.r_type != R_AARCH64_JUMP26)
-          continue;
-
-        Symbol<E> &sym = *isec->file.symbols[rel.r_sym];
-
-        // Skip if the destination is within reach.
-        if (is_reachable(ctx, sym, *isec, rel))
-          continue;
-
-        // If the symbol is already in another thunk, reuse it.
-        if (sym.thunk_idx != -1) {
-          isec->range_extn[i] = {sym.thunk_idx, sym.thunk_sym_idx};
-          continue;
-        }
-
-        // Otherwise, add the symbol to this thunk if it's not added already.
-        isec->range_extn[i] = {thunk.thunk_idx, -1};
-
-        if (!(sym.flags.fetch_or(NEEDS_THUNK) & NEEDS_THUNK)) {
-          std::scoped_lock lock(thunk.mu);
-          thunk.symbols.push_back(&sym);
-        }
-      }
-    });
-
-    // Now that we know the number of symbols in the thunk, we can compute
-    // its size.
-    offset += thunk.size();
-
-    // Sort symbols added to the thunk to make the output deterministic.
-    sort(thunk.symbols, [](Symbol<E> *a, Symbol<E> *b) { return *a < *b; });
-
-    // Assign offsets within the thunk to the symbols.
-    for (i64 i = 0; Symbol<E> *sym : thunk.symbols) {
-      sym->thunk_idx = thunk.thunk_idx;
-      sym->thunk_sym_idx = i++;
-    }
-
-    // Scan relocations again to fix symbol offsets in the last thunk.
-    tbb::parallel_for_each(&members[b], &members[c], [&](InputSection<E> *isec) {
-      std::span<ElfRel<E>> rels = isec->get_rels(ctx);
-
-      for (i64 i = 0; i < rels.size(); i++) {
-        RangeExtensionRef &ref = isec->range_extn[i];
-
-        if (ref.thunk_idx == thunk.thunk_idx) {
-          Symbol<E> &sym = *isec->file.symbols[rels[i].r_sym];
-          ref.sym_idx = sym.thunk_sym_idx;
-        }
-      }
-    });
-
-    // Move B forward to point to the begining of the next group.
-    b = c;
-  }
-
-  while (a < osec.thunks.size())
-    reset_thunk(*osec.thunks[a++]);
-
-  osec.shdr.sh_size = offset;
-}
-
-static void mark_thunk_symbols(Context<E> &ctx, OutputSection<E> &osec) {
-  for (std::unique_ptr<RangeExtensionThunk<E>> &thunk : osec.thunks) {
-    i64 sz = thunk->symbols.size();
-    thunk->symbol_map.resize(sz);
-    thunk->used.reset(new std::atomic_bool[sz]{});
-  }
-
-  // Mark referenced thunk symbols
-  tbb::parallel_for_each(osec.members, [&](InputSection<E> *isec) {
-    std::span<ElfRel<E>> rels = isec->get_rels(ctx);
-
-    for (i64 i = 0; i < rels.size(); i++) {
-      RangeExtensionRef &ref = isec->range_extn[i];
-      if (ref.thunk_idx == -1)
-        continue;
-
-      Symbol<E> &sym = *isec->file.symbols[rels[i].r_sym];
-      if (!is_reachable(ctx, sym, *isec, rels[i]))
-        osec.thunks[ref.thunk_idx]->used[ref.sym_idx] = true;
-    }
-  });
-
-  // Remove unreferenced thunk symbols
-  tbb::parallel_for_each(osec.thunks,
-                         [&](std::unique_ptr<RangeExtensionThunk<E>> &thunk) {
-    i64 i = 0;
-    for (i64 j = 0; j < thunk->symbols.size(); j++) {
-      if (thunk->used[j]) {
-        thunk->symbol_map[j] = i;
-        thunk->symbols[i] = thunk->symbols[j];
-        i++;
-      }
-    }
-    thunk->symbols.resize(i);
-  });
-}
-
-static void shrink_section(Context<E> &ctx, OutputSection<E> &osec) {
-  std::span<std::unique_ptr<RangeExtensionThunk<E>>> thunks = osec.thunks;
-  std::span<InputSection<E> *> members = osec.members;
-
-  i64 offset = 0;
-
-  auto add_thunk = [&]() {
-    thunks[0]->offset = offset;
-    offset += thunks[0]->size();
-    thunks = thunks.subspan(1);
-  };
-
-  auto add_isec = [&]() {
-    offset = align_to(offset, members[0]->shdr.sh_addralign);
-    members[0]->offset = offset;
-    offset += members[0]->shdr.sh_size;
-    members = members.subspan(1);
-  };
-
-  while (!thunks.empty() && !members.empty()) {
-    if (thunks[0]->offset < members[0]->offset)
-      add_thunk();
-    else
-      add_isec();
-  }
-
-  while (!thunks.empty())
-    add_thunk();
-  while (!members.empty())
-    add_isec();
-
-  assert(offset <= osec.shdr.sh_size);
-  osec.shdr.sh_size = offset;
-}
-
-// ARM64's call/jump instructions take 27 bits displacement, so they
-// can refer only up to ±128 MiB. If a branch target is further than
-// that, we need to let it branch to a linker-synthesized code
-// sequence that construct a full 32 bit address in a register and
-// jump there. That linker-synthesized code is called "thunk".
-i64 create_range_extension_thunks(Context<E> &ctx) {
-  Timer t(ctx, "create_range_extension_thunks");
-
-  // First, we create thunks with a pessimistic assumption that all
-  // out-of-section relocations would need thunks. To do so, we start
-  // with an initial layout in which output sections are separated far
-  // apart.
-  for (i64 i = 0; Chunk<E> *chunk : ctx.chunks)
-    chunk->shdr.sh_addr = i++ << 31;
-
-  std::vector<OutputSection<E> *> sections;
-  for (std::unique_ptr<OutputSection<E>> &osec : ctx.output_sections)
-    if (!osec->members.empty() && (osec->shdr.sh_flags & SHF_EXECINSTR))
-      sections.push_back(osec.get());
-
-  for (OutputSection<E> *osec : sections)
-    create_thunks(ctx, *osec);
-
-  // Recompute file layout.
-  set_osec_offsets(ctx);
-
-  // Based on the current file layout, remove thunk symbols that turned
-  // to be unnecessary.
-  tbb::parallel_for_each(sections, [&](OutputSection<E> *osec) {
-    mark_thunk_symbols(ctx, *osec);
-  });
-
-  // Recompute output section sizes that contain thunks. New section
-  // sizes must be equal to or smaller than previous values, so all
-  // relocations that were previously reachable will still be reachable
-  // after this step.
-  for (OutputSection<E> *osec : sections)
-    shrink_section(ctx, *osec);
-
-  // Compute the final layout.
-  return set_osec_offsets(ctx);
-}
-
-void RangeExtensionThunk<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + output_section.shdr.sh_offset + offset;
-
-  static const u8 insn[] = {
-    0x10, 0x00, 0x00, 0x90, // adrp x16, 0   # R_AARCH64_ADR_PREL_PG_HI21
-    0x10, 0x02, 0x00, 0x91, // add  x16, x16 # R_AARCH64_ADD_ABS_LO12_NC
-    0x00, 0x02, 0x1f, 0xd6, // br   x16
-  };
-
-  static_assert(ENTRY_SIZE == sizeof(insn));
-
-  for (i64 i = 0; i < symbols.size(); i++) {
-    u64 S = symbols[i]->get_addr(ctx);
-    u64 P = output_section.shdr.sh_addr + offset + i * ENTRY_SIZE;
-
-    u8 *loc = buf + i * ENTRY_SIZE;
-    memcpy(loc , insn, sizeof(insn));
-    write_adr(loc, bits(page(S) - page(P), 32, 12));
-    *(u32 *)(loc + 4) |= bits(S, 11, 0) << 10;
-  }
-}
-
-void write_thunks(Context<E> &ctx) {
-  Timer t(ctx, "write_thunks");
-
-  tbb::parallel_for_each(ctx.output_sections,
-                         [&](std::unique_ptr<OutputSection<E>> &osec) {
-    tbb::parallel_for_each(osec->thunks,
-                           [&](std::unique_ptr<RangeExtensionThunk<E>> &thunk) {
-      thunk->copy_buf(ctx);
-    });
-  });
-}
-
-} // namespace mold::elf

elf/arch-i386.cc

@@ -1,473 +0,0 @@
-#include "mold.h"
-
-namespace mold::elf {
-
-using E = I386;
-
-template <>
-void GotPltSection<E>::copy_buf(Context<E> &ctx) {
-  u32 *buf = (u32 *)(ctx.buf + this->shdr.sh_offset);
-
-  // The first slot of .got.plt points to _DYNAMIC.
-  buf[0] = ctx.dynamic ? ctx.dynamic->shdr.sh_addr : 0;
-  buf[1] = 0;
-  buf[2] = 0;
-
-  for (Symbol<E> *sym : ctx.plt->symbols)
-    buf[sym->get_gotplt_idx(ctx)] = sym->get_plt_addr(ctx) + 6;
-}
-
-static void write_plt_header(Context<E> &ctx, u8 *buf) {
-  if (ctx.arg.pic) {
-    static const u8 plt0[] = {
-      0xff, 0xb3, 0x04, 0, 0, 0, // pushl 4(%ebx)
-      0xff, 0xa3, 0x08, 0, 0, 0, // jmp *8(%ebx)
-      0x90, 0x90, 0x90, 0x90,    // nop
-    };
-    memcpy(buf, plt0, sizeof(plt0));
-  } else {
-    static const u8 plt0[] = {
-      0xff, 0x35, 0, 0, 0, 0, // pushl (GOTPLT+4)
-      0xff, 0x25, 0, 0, 0, 0, // jmp *(GOTPLT+8)
-      0x90, 0x90, 0x90, 0x90, // nop
-    };
-    memcpy(buf, plt0, sizeof(plt0));
-    *(u32 *)(buf + 2) = ctx.gotplt->shdr.sh_addr + 4;
-    *(u32 *)(buf + 8) = ctx.gotplt->shdr.sh_addr + 8;
-  }
-}
-
-static void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym,
-                            i64 idx) {
-  u8 *ent = buf + ctx.plt_hdr_size + sym.get_plt_idx(ctx) * ctx.plt_size;
-
-  if (ctx.arg.pic) {
-    static const u8 data[] = {
-      0xff, 0xa3, 0, 0, 0, 0, // jmp *foo@GOT(%ebx)
-      0x68, 0,    0, 0, 0,    // pushl $reloc_offset
-      0xe9, 0,    0, 0, 0,    // jmp .PLT0@PC
-    };
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 2) = sym.get_gotplt_addr(ctx) - ctx.gotplt->shdr.sh_addr;
-  } else {
-    static const u8 data[] = {
-      0xff, 0x25, 0, 0, 0, 0, // jmp *foo@GOT
-      0x68, 0,    0, 0, 0,    // pushl $reloc_offset
-      0xe9, 0,    0, 0, 0,    // jmp .PLT0@PC
-    };
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 2) = sym.get_gotplt_addr(ctx);
-  }
-
-  *(u32 *)(ent + 7) = idx * sizeof(ElfRel<E>);
-  *(u32 *)(ent + 12) = ctx.plt->shdr.sh_addr - sym.get_plt_addr(ctx) - 16;
-}
-
-template <>
-void PltSection<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + this->shdr.sh_offset;
-  write_plt_header(ctx, buf);
-
-  for (i64 i = 0; i < symbols.size(); i++)
-    write_plt_entry(ctx, buf, *symbols[i], i);
-}
-
-template <>
-void PltGotSection<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + this->shdr.sh_offset;
-
-  if (ctx.arg.pic) {
-    static const u8 data[] = {
-      0xff, 0xa3, 0, 0, 0, 0, // jmp   *foo@GOT(%ebx)
-      0x66, 0x90,             // nop
-    };
-
-    for (i64 i = 0; i < symbols.size(); i++) {
-      u8 *ent = buf + i * sizeof(data);
-      memcpy(ent, data, sizeof(data));
-      *(u32 *)(ent + 2) = symbols[i]->get_got_addr(ctx) - ctx.gotplt->shdr.sh_addr;
-    }
-  } else {
-    static const u8 data[] = {
-      0xff, 0x25, 0, 0, 0, 0, // jmp   *foo@GOT
-      0x66, 0x90,             // nop
-    };
-
-    for (i64 i = 0; i < symbols.size(); i++) {
-      u8 *ent = buf + i * sizeof(data);
-      memcpy(ent, data, sizeof(data));
-      *(u32 *)(ent + 2) = symbols[i]->get_got_addr(ctx);
-    }
-  }
-}
-
-template <>
-void EhFrameSection<E>::apply_reloc(Context<E> &ctx, ElfRel<E> &rel,
-                                    u64 loc, u64 val) {
-  u8 *base = ctx.buf + this->shdr.sh_offset;
-
-  switch (rel.r_type) {
-  case R_386_NONE:
-    return;
-  case R_386_32:
-    *(u32 *)(base + loc) = val;
-    return;
-  case R_386_PC32:
-    *(u32 *)(base + loc) = val - this->shdr.sh_addr - loc;
-    return;
-  }
-  unreachable();
-}
-
-template <>
-void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
-  ElfRel<E> *dynrel = nullptr;
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-  i64 frag_idx = 0;
-
-  if (ctx.reldyn)
-    dynrel = (ElfRel<E> *)(ctx.buf + ctx.reldyn->shdr.sh_offset +
-                              file.reldyn_offset + this->reldyn_offset);
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_386_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    const SectionFragmentRef<E> *frag_ref = nullptr;
-    if (rel_fragments && rel_fragments[frag_idx].idx == i)
-      frag_ref = &rel_fragments[frag_idx++];
-
-    auto overflow_check = [&](i64 val, i64 lo, i64 hi) {
-      if (val < lo || hi <= val)
-        Error(ctx) << *this << ": relocation " << rel << " against "
-                   << sym << " out of range: " << val << " is not in ["
-                   << lo << ", " << hi << ")";
-    };
-
-    auto write8 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 8);
-      *loc = val;
-    };
-
-    auto write8s = [&](u64 val) {
-      overflow_check(val, -(1 << 7), 1 << 7);
-      *loc = val;
-    };
-
-    auto write16 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 16);
-      *(u16 *)loc = val;
-    };
-
-    auto write16s = [&](u64 val) {
-      overflow_check(val, -(1 << 15), 1 << 15);
-      *(u16 *)loc = val;
-    };
-
-    auto write32 = [&](u64 val) {
-      *(u32 *)loc = val;
-    };
-
-#define S      (frag_ref ? frag_ref->frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A      (frag_ref ? frag_ref->addend : this->get_addend(rel))
-#define P      (output_section->shdr.sh_addr + offset + rel.r_offset)
-#define G      (sym.get_got_addr(ctx) - ctx.got->shdr.sh_addr)
-#define GOTPLT ctx.gotplt->shdr.sh_addr
-
-    if (needs_dynrel[i]) {
-      *dynrel++ = {P, R_386_32, (u32)sym.get_dynsym_idx(ctx)};
-      write32(A);
-      continue;
-    }
-
-    if (needs_baserel[i]) {
-      if (!is_relr_reloc(ctx, rel))
-        *dynrel++ = {P, R_386_RELATIVE, 0};
-      write32(S + A);
-      continue;
-    }
-
-    switch (rel.r_type) {
-    case R_386_8:
-      write8(S + A);
-      continue;
-    case R_386_16:
-      write16(S + A);
-      continue;
-    case R_386_32:
-      write32(S + A);
-      continue;
-    case R_386_PC8:
-      write8s(S + A);
-      continue;
-    case R_386_PC16:
-      write16s(S + A);
-      continue;
-    case R_386_PC32:
-    case R_386_PLT32:
-      write32(S + A - P);
-      continue;
-    case R_386_GOT32:
-    case R_386_GOT32X:
-      write32(sym.get_got_addr(ctx) + A - GOTPLT);
-      continue;
-    case R_386_GOTOFF:
-      write32(S + A - GOTPLT);
-      continue;
-    case R_386_GOTPC:
-      write32(GOTPLT + A - P);
-      continue;
-    case R_386_TLS_GOTIE:
-      write32(sym.get_gottp_addr(ctx) + A - GOTPLT);
-      continue;
-    case R_386_TLS_LE:
-      write32(S + A - ctx.tls_end);
-      continue;
-    case R_386_TLS_IE:
-      write32(sym.get_gottp_addr(ctx) + A);
-      continue;
-    case R_386_TLS_GD:
-      write32(sym.get_tlsgd_addr(ctx) + A - GOTPLT);
-      continue;
-    case R_386_TLS_LDM:
-      write32(ctx.got->get_tlsld_addr(ctx) + A - GOTPLT);
-      continue;
-    case R_386_TLS_LDO_32:
-      write32(S + A - ctx.tls_begin);
-      continue;
-    case R_386_SIZE32:
-      write32(sym.esym().st_size + A);
-      continue;
-    case R_386_TLS_GOTDESC:
-      if (sym.get_tlsdesc_idx(ctx) == -1) {
-        static const u8 insn[] = {
-          0x8d, 0x05, 0, 0, 0, 0, // lea 0, %eax
-        };
-        memcpy(loc - 2, insn, sizeof(insn));
-        write32(S + A - ctx.tls_end);
-      } else {
-        write32(sym.get_tlsdesc_addr(ctx) + A - GOTPLT);
-      }
-      continue;
-    case R_386_TLS_DESC_CALL:
-      if (ctx.arg.relax && !ctx.arg.shared) {
-        // call *(%rax) -> nop
-        loc[0] = 0x66;
-        loc[1] = 0x90;
-      }
-      continue;
-    default:
-      unreachable();
-    }
-
-#undef S
-#undef A
-#undef P
-#undef G
-#undef GOTPLT
-  }
-}
-
-template <>
-void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-  i64 frag_idx = 0;
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_386_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    SectionFragment<E> *frag;
-    i64 addend;
-    std::tie(frag, addend) = get_fragment(ctx, rel);
-
-    auto overflow_check = [&](i64 val, i64 lo, i64 hi) {
-      if (val < lo || hi <= val)
-        Error(ctx) << *this << ": relocation " << rel << " against "
-                   << sym << " out of range: " << val << " is not in ["
-                   << lo << ", " << hi << ")";
-    };
-
-    auto write8 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 8);
-      *loc = val;
-    };
-
-    auto write8s = [&](u64 val) {
-      overflow_check(val, -(1 << 7), 1 << 7);
-      *loc = val;
-    };
-
-    auto write16 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 16);
-      *(u16 *)loc = val;
-    };
-
-    auto write16s = [&](u64 val) {
-      overflow_check(val, -(1 << 15), 1 << 15);
-      *(u16 *)loc = val;
-    };
-
-#define S      (frag ? frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A      (frag ? addend : this->get_addend(rel))
-#define G      (sym.get_got_addr(ctx) - ctx.got->shdr.sh_addr)
-#define GOTPLT ctx.gotplt->shdr.sh_addr
-
-    switch (rel.r_type) {
-    case R_386_8:
-      write8(S + A);
-      continue;
-    case R_386_16:
-      write16(S + A);
-      continue;
-    case R_386_32:
-      *(u32 *)loc = S + A;
-      continue;
-    case R_386_PC8:
-      write8s(S + A);
-      continue;
-    case R_386_PC16:
-      write16s(S + A);
-      continue;
-    case R_386_PC32:
-      *(u32 *)loc = S + A;
-      continue;
-    case R_386_GOTPC:
-      *(u32 *)loc = GOTPLT + A;
-      continue;
-    case R_386_GOTOFF:
-      *(u32 *)loc = S + A - GOTPLT;
-      continue;
-    case R_386_TLS_LDO_32:
-      *(u32 *)loc = S + A - ctx.tls_begin;
-      continue;
-    case R_386_SIZE32:
-      *(u32 *)loc = sym.esym().st_size + A;
-      continue;
-    default:
-      unreachable();
-    }
-
-#undef S
-#undef A
-#undef GOTPLT
-  }
-}
-
-template <>
-void InputSection<E>::scan_relocations(Context<E> &ctx) {
-  assert(shdr.sh_flags & SHF_ALLOC);
-
-  this->reldyn_offset = file.num_dynrel * sizeof(ElfRel<E>);
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-
-  // Scan relocations
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_386_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    if (sym.get_type() == STT_GNU_IFUNC) {
-      sym.flags |= NEEDS_GOT;
-      sym.flags |= NEEDS_PLT;
-    }
-
-    switch (rel.r_type) {
-    case R_386_8:
-    case R_386_16: {
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  NONE,     ERROR, ERROR,         ERROR },      // DSO
-        {  NONE,     ERROR, ERROR,         ERROR },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_386_32: {
-      Action table[][4] = {
-        // Absolute  Local    Imported data  Imported code
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // DSO
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // PIE
-        {  NONE,     NONE,    COPYREL,       PLT },        // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_386_PC8:
-    case R_386_PC16: {
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  ERROR,    NONE,  ERROR,         ERROR },      // DSO
-        {  ERROR,    NONE,  COPYREL,       PLT   },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_386_PC32: {
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  BASEREL,  NONE,  DYNREL,        DYNREL },     // DSO
-        {  BASEREL,  NONE,  COPYREL,       PLT   },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_386_GOT32:
-    case R_386_GOT32X:
-    case R_386_GOTPC:
-      sym.flags |= NEEDS_GOT;
-      break;
-    case R_386_PLT32:
-      if (sym.is_imported)
-        sym.flags |= NEEDS_PLT;
-      break;
-    case R_386_TLS_GOTIE:
-    case R_386_TLS_LE:
-    case R_386_TLS_IE:
-      sym.flags |= NEEDS_GOTTP;
-      break;
-    case R_386_TLS_GD:
-      sym.flags |= NEEDS_TLSGD;
-      break;
-    case R_386_TLS_LDM:
-      sym.flags |= NEEDS_TLSLD;
-      break;
-    case R_386_TLS_GOTDESC:
-      if (!ctx.arg.relax || ctx.arg.shared)
-        sym.flags |= NEEDS_TLSDESC;
-      break;
-    case R_386_GOTOFF:
-    case R_386_TLS_LDO_32:
-    case R_386_SIZE32:
-    case R_386_TLS_DESC_CALL:
-      break;
-    default:
-      Error(ctx) << *this << ": unknown relocation: " << rel;
-    }
-  }
-}
-
-} // namespace mold::elf

elf/arch-x86-64.cc

@@ -1,749 +0,0 @@
-#include "mold.h"
-
-namespace mold::elf {
-
-using E = X86_64;
-
-template <>
-void GotPltSection<E>::copy_buf(Context<E> &ctx) {
-  u64 *buf = (u64 *)(ctx.buf + this->shdr.sh_offset);
-
-  // The first slot of .got.plt points to _DYNAMIC, as requested by
-  // the x86-64 psABI. The second and the third slots are reserved by
-  // the psABI.
-  buf[0] = ctx.dynamic ? ctx.dynamic->shdr.sh_addr : 0;
-  buf[1] = 0;
-  buf[2] = 0;
-
-  for (Symbol<E> *sym : ctx.plt->symbols) {
-    if (ctx.arg.z_ibtplt)
-      buf[sym->get_gotplt_idx(ctx)] = sym->get_plt_addr(ctx) + 10;
-    else
-      buf[sym->get_gotplt_idx(ctx)] = sym->get_plt_addr(ctx) + 6;
-  }
-}
-
-// The compact PLT format is used when `-z now` is given. If the flag
-// is given, all PLT symbols are resolved eagerly on startup, so we
-// can omit code for lazy symbol resolution from PLT in that case.
-static void write_compact_plt(Context<E> &ctx) {
-  u8 *buf = ctx.buf + ctx.plt->shdr.sh_offset;
-
-  static const u8 data[] = {
-    0xff, 0x25, 0, 0, 0, 0, // jmp *foo@GOT
-    0x66, 0x90,             // nop
-  };
-
-  for (Symbol<E> *sym : ctx.plt->symbols) {
-    u8 *ent = buf + sym->get_plt_idx(ctx) * ctx.plt_size;
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 2) = sym->get_gotplt_addr(ctx) - sym->get_plt_addr(ctx) - 6;
-  }
-}
-
-// The IBTPLT is a security-enhanced version of the regular PLT.
-// It uses Indirect Branch Tracking (IBT) feature which is part of
-// Intel Control-Flow Enforcement (CET). IBTPLT is slightly larger
-// than the regular PLT (24 bytes vs 16 bytes for each entry).
-//
-// Note that our IBTPLT instruction sequence is different from the one
-// used in GNU ld. GNU's IBTPLT implementation uses two separate
-// sections (.plt and .plt.sec) in which one PLT entry takes 32 bytes
-// in total.
-static void write_ibtplt(Context<E> &ctx) {
-  u8 *buf = ctx.buf + ctx.plt->shdr.sh_offset;
-
-  // Write PLT header
-  static const u8 plt0[] = {
-    0xff, 0x35, 0, 0, 0, 0, // push GOTPLT+8(%rip)
-    0xff, 0x25, 0, 0, 0, 0, // jmp *GOTPLT+16(%rip)
-    0x0f, 0x1f, 0x40, 0x00, // nop
-  };
-
-  memcpy(buf, plt0, sizeof(plt0));
-  *(u32 *)(buf + 2) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 2;
-  *(u32 *)(buf + 8) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 4;
-
-  // Write PLT entries
-  i64 relplt_idx = 0;
-
-  static const u8 data[] = {
-    0xf3, 0x0f, 0x1e, 0xfa, // endbr64
-    0xff, 0x25, 0, 0, 0, 0, // jmp *foo@GOTPLT
-    0x68, 0, 0, 0, 0,       // push $index_in_relplt
-    0xf2, 0xe9, 0, 0, 0, 0, // jmp PLT[0]
-    0x0f, 0x1f, 0x00,       // nop
-  };
-
-  for (Symbol<E> *sym : ctx.plt->symbols) {
-    u8 *ent = buf + ctx.plt_hdr_size + sym->get_plt_idx(ctx) * ctx.plt_size;
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 6) = sym->get_gotplt_addr(ctx) - sym->get_plt_addr(ctx) - 10;
-    *(u32 *)(ent + 11) = relplt_idx++;
-    *(u32 *)(ent + 17) = ctx.plt->shdr.sh_addr - sym->get_plt_addr(ctx) - 21;
-  }
-}
-
-// The regular PLT.
-static void write_plt(Context<E> &ctx) {
-  u8 *buf = ctx.buf + ctx.plt->shdr.sh_offset;
-
-  // Write PLT header
-  static const u8 plt0[] = {
-    0xff, 0x35, 0, 0, 0, 0, // pushq GOTPLT+8(%rip)
-    0xff, 0x25, 0, 0, 0, 0, // jmp *GOTPLT+16(%rip)
-    0x0f, 0x1f, 0x40, 0x00, // nop
-  };
-
-  memcpy(buf, plt0, sizeof(plt0));
-  *(u32 *)(buf + 2) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 2;
-  *(u32 *)(buf + 8) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 4;
-
-  // Write PLT entries
-  i64 relplt_idx = 0;
-
-  static const u8 data[] = {
-    0xff, 0x25, 0, 0, 0, 0, // jmp   *foo@GOTPLT
-    0x68, 0,    0, 0, 0,    // push  $index_in_relplt
-    0xe9, 0,    0, 0, 0,    // jmp   PLT[0]
-  };
-
-  for (Symbol<E> *sym : ctx.plt->symbols) {
-    u8 *ent = buf + ctx.plt_hdr_size + sym->get_plt_idx(ctx) * ctx.plt_size;
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 2) = sym->get_gotplt_addr(ctx) - sym->get_plt_addr(ctx) - 6;
-    *(u32 *)(ent + 7) = relplt_idx++;
-    *(u32 *)(ent + 12) = ctx.plt->shdr.sh_addr - sym->get_plt_addr(ctx) - 16;
-  }
-}
-
-template <>
-void PltSection<E>::copy_buf(Context<E> &ctx) {
-  if (ctx.arg.z_now)
-    write_compact_plt(ctx);
-  else if (ctx.arg.z_ibtplt)
-    write_ibtplt(ctx);
-  else
-    write_plt(ctx);
-}
-
-template <>
-void PltGotSection<E>::copy_buf(Context<E> &ctx) {
-  u8 *buf = ctx.buf + this->shdr.sh_offset;
-
-  static const u8 data[] = {
-    0xff, 0x25, 0, 0, 0, 0, // jmp *foo@GOT
-    0x66, 0x90,             // nop
-  };
-
-  for (Symbol<E> *sym : symbols) {
-    u8 *ent = buf + sym->get_pltgot_idx(ctx) * X86_64::pltgot_size;
-    memcpy(ent, data, sizeof(data));
-    *(u32 *)(ent + 2) = sym->get_got_addr(ctx) - sym->get_plt_addr(ctx) - 6;
-  }
-}
-
-template <>
-void EhFrameSection<E>::apply_reloc(Context<E> &ctx, ElfRel<E> &rel,
-                                    u64 loc, u64 val) {
-  u8 *base = ctx.buf + this->shdr.sh_offset;
-
-  switch (rel.r_type) {
-  case R_X86_64_NONE:
-    return;
-  case R_X86_64_32:
-    *(u32 *)(base + loc) = val;
-    return;
-  case R_X86_64_64:
-    *(u64 *)(base + loc) = val;
-    return;
-  case R_X86_64_PC32:
-    *(u32 *)(base + loc) = val - this->shdr.sh_addr - loc;
-    return;
-  case R_X86_64_PC64:
-    *(u64 *)(base + loc) = val - this->shdr.sh_addr - loc;
-    return;
-  }
-  unreachable();
-}
-
-static u32 relax_gotpcrelx(u8 *loc) {
-  switch ((loc[0] << 8) | loc[1]) {
-  case 0xff15: return 0x90e8; // call *0(%rip) -> call 0
-  case 0xff25: return 0x90e9; // jmp  *0(%rip) -> jmp  0
-  }
-  return 0;
-}
-
-static u32 relax_rex_gotpcrelx(u8 *loc) {
-  switch ((loc[0] << 16) | (loc[1] << 8) | loc[2]) {
-  case 0x488b05: return 0x488d05; // mov 0(%rip), %rax -> lea 0(%rip), %rax
-  case 0x488b0d: return 0x488d0d; // mov 0(%rip), %rcx -> lea 0(%rip), %rcx
-  case 0x488b15: return 0x488d15; // mov 0(%rip), %rdx -> lea 0(%rip), %rdx
-  case 0x488b1d: return 0x488d1d; // mov 0(%rip), %rbx -> lea 0(%rip), %rbx
-  case 0x488b25: return 0x488d25; // mov 0(%rip), %rsp -> lea 0(%rip), %rsp
-  case 0x488b2d: return 0x488d2d; // mov 0(%rip), %rbp -> lea 0(%rip), %rbp
-  case 0x488b35: return 0x488d35; // mov 0(%rip), %rsi -> lea 0(%rip), %rsi
-  case 0x488b3d: return 0x488d3d; // mov 0(%rip), %rdi -> lea 0(%rip), %rdi
-  case 0x4c8b05: return 0x4c8d05; // mov 0(%rip), %r8  -> lea 0(%rip), %r8
-  case 0x4c8b0d: return 0x4c8d0d; // mov 0(%rip), %r9  -> lea 0(%rip), %r9
-  case 0x4c8b15: return 0x4c8d15; // mov 0(%rip), %r10 -> lea 0(%rip), %r10
-  case 0x4c8b1d: return 0x4c8d1d; // mov 0(%rip), %r11 -> lea 0(%rip), %r11
-  case 0x4c8b25: return 0x4c8d25; // mov 0(%rip), %r12 -> lea 0(%rip), %r12
-  case 0x4c8b2d: return 0x4c8d2d; // mov 0(%rip), %r13 -> lea 0(%rip), %r13
-  case 0x4c8b35: return 0x4c8d35; // mov 0(%rip), %r14 -> lea 0(%rip), %r14
-  case 0x4c8b3d: return 0x4c8d3d; // mov 0(%rip), %r15 -> lea 0(%rip), %r15
-  }
-  return 0;
-}
-
-static u32 relax_gottpoff(u8 *loc) {
-  switch ((loc[0] << 16) | (loc[1] << 8) | loc[2]) {
-  case 0x488b05: return 0x48c7c0; // mov 0(%rip), %rax -> mov $0, %rax
-  case 0x488b0d: return 0x48c7c1; // mov 0(%rip), %rcx -> mov $0, %rcx
-  case 0x488b15: return 0x48c7c2; // mov 0(%rip), %rdx -> mov $0, %rdx
-  case 0x488b1d: return 0x48c7c3; // mov 0(%rip), %rbx -> mov $0, %rbx
-  case 0x488b25: return 0x48c7c4; // mov 0(%rip), %rsp -> mov $0, %rsp
-  case 0x488b2d: return 0x48c7c5; // mov 0(%rip), %rbp -> mov $0, %rbp
-  case 0x488b35: return 0x48c7c6; // mov 0(%rip), %rsi -> mov $0, %rsi
-  case 0x488b3d: return 0x48c7c7; // mov 0(%rip), %rdi -> mov $0, %rdi
-  case 0x4c8b05: return 0x49c7c0; // mov 0(%rip), %r8  -> mov $0, %r8
-  case 0x4c8b0d: return 0x49c7c1; // mov 0(%rip), %r9  -> mov $0, %r9
-  case 0x4c8b15: return 0x49c7c2; // mov 0(%rip), %r10 -> mov $0, %r10
-  case 0x4c8b1d: return 0x49c7c3; // mov 0(%rip), %r11 -> mov $0, %r11
-  case 0x4c8b25: return 0x49c7c4; // mov 0(%rip), %r12 -> mov $0, %r12
-  case 0x4c8b2d: return 0x49c7c5; // mov 0(%rip), %r13 -> mov $0, %r13
-  case 0x4c8b35: return 0x49c7c6; // mov 0(%rip), %r14 -> mov $0, %r14
-  case 0x4c8b3d: return 0x49c7c7; // mov 0(%rip), %r15 -> mov $0, %r15
-  }
-  return 0;
-}
-
-// Apply relocations to SHF_ALLOC sections (i.e. sections that are
-// mapped to memory at runtime) based on the result of
-// scan_relocations().
-template <>
-void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
-  ElfRel<E> *dynrel = nullptr;
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-  i64 frag_idx = 0;
-
-  if (ctx.reldyn)
-    dynrel = (ElfRel<E> *)(ctx.buf + ctx.reldyn->shdr.sh_offset +
-                                file.reldyn_offset + this->reldyn_offset);
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_X86_64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    const SectionFragmentRef<E> *frag_ref = nullptr;
-    if (rel_fragments && rel_fragments[frag_idx].idx == i)
-      frag_ref = &rel_fragments[frag_idx++];
-
-    auto overflow_check = [&](i64 val, i64 lo, i64 hi) {
-      if (val < lo || hi <= val)
-        Error(ctx) << *this << ": relocation " << rel << " against "
-                   << sym << " out of range: " << val << " is not in ["
-                   << lo << ", " << hi << ")";
-    };
-
-    auto write8 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 8);
-      *loc = val;
-    };
-
-    auto write8s = [&](u64 val) {
-      overflow_check(val, -(1 << 7), 1 << 7);
-      *loc = val;
-    };
-
-    auto write16 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 16);
-      *(u16 *)loc = val;
-    };
-
-    auto write16s = [&](u64 val) {
-      overflow_check(val, -(1 << 15), 1 << 15);
-      *(u16 *)loc = val;
-    };
-
-    auto write32 = [&](u64 val) {
-      overflow_check(val, 0, (i64)1 << 32);
-      *(u32 *)loc = val;
-    };
-
-    auto write32s = [&](u64 val) {
-      overflow_check(val, -((i64)1 << 31), (i64)1 << 31);
-      *(u32 *)loc = val;
-    };
-
-    auto write64 = [&](u64 val) {
-      *(u64 *)loc = val;
-    };
-
-#define S   (frag_ref ? frag_ref->frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A   (frag_ref ? frag_ref->addend : rel.r_addend)
-#define P   (output_section->shdr.sh_addr + offset + rel.r_offset)
-#define G   (sym.get_got_addr(ctx) - ctx.gotplt->shdr.sh_addr)
-#define GOT ctx.gotplt->shdr.sh_addr
-
-    if (needs_dynrel[i]) {
-      *dynrel++ = {P, R_X86_64_64, (u32)sym.get_dynsym_idx(ctx), A};
-      write64(A);
-      continue;
-    }
-
-    if (needs_baserel[i]) {
-      if (!is_relr_reloc(ctx, rel))
-        *dynrel++ = {P, R_X86_64_RELATIVE, 0, (i64)(S + A)};
-      write64(S + A);
-      continue;
-    }
-
-    switch (rel.r_type) {
-    case R_X86_64_8:
-      write8(S + A);
-      continue;
-    case R_X86_64_16:
-      write16(S + A);
-      continue;
-    case R_X86_64_32:
-      write32(S + A);
-      continue;
-    case R_X86_64_32S:
-      write32s(S + A);
-      continue;
-    case R_X86_64_64:
-      write64(S + A);
-      continue;
-    case R_X86_64_PC8:
-      write8s(S + A - P);
-      continue;
-    case R_X86_64_PC16:
-      write16s(S + A - P);
-      continue;
-    case R_X86_64_PC32:
-      write32s(S + A - P);
-      continue;
-    case R_X86_64_PC64:
-      write64(S + A - P);
-      continue;
-    case R_X86_64_PLT32:
-      write32s(S + A - P);
-      continue;
-    case R_X86_64_PLTOFF64:
-      write64(S + A - GOT);
-      break;
-    case R_X86_64_GOT32:
-      write32s(G + A);
-      continue;
-    case R_X86_64_GOT64:
-      write64(G + A);
-      continue;
-    case R_X86_64_GOTOFF64:
-      write64(S + A - GOT);
-      continue;
-    case R_X86_64_GOTPC32:
-      write32s(GOT + A - P);
-      continue;
-    case R_X86_64_GOTPC64:
-      write64(GOT + A - P);
-      continue;
-    case R_X86_64_GOTPCREL:
-      write32s(G + GOT + A - P);
-      continue;
-    case R_X86_64_GOTPCREL64:
-      write64(G + GOT + A - P);
-      continue;
-    case R_X86_64_GOTPCRELX:
-      if (sym.get_got_idx(ctx) == -1) {
-        u32 insn = relax_gotpcrelx(loc - 2);
-        loc[-2] = insn >> 8;
-        loc[-1] = insn;
-        write32s(S + A - P);
-      } else {
-        write32s(G + GOT + A - P);
-      }
-      continue;
-    case R_X86_64_REX_GOTPCRELX:
-      if (sym.get_got_idx(ctx) == -1) {
-        u32 insn = relax_rex_gotpcrelx(loc - 3);
-        loc[-3] = insn >> 16;
-        loc[-2] = insn >> 8;
-        loc[-1] = insn;
-        write32s(S + A - P);
-      } else {
-        write32s(G + GOT + A - P);
-      }
-      continue;
-    case R_X86_64_TLSGD:
-      if (sym.get_tlsgd_idx(ctx) == -1) {
-        // Relax GD to LE
-        static const u8 insn[] = {
-          0x64, 0x48, 0x8b, 0x04, 0x25, 0, 0, 0, 0, // mov %fs:0, %rax
-          0x48, 0x8d, 0x80, 0,    0,    0, 0,       // lea 0(%rax), %rax
-        };
-        memcpy(loc - 4, insn, sizeof(insn));
-
-        i64 val = S - ctx.tls_end + A + 4;
-        overflow_check(val, -((i64)1 << 31), (i64)1 << 31);
-        *(u32 *)(loc + 8) = val;
-        i++;
-      } else {
-        write32s(sym.get_tlsgd_addr(ctx) + A - P);
-      }
-      continue;
-    case R_X86_64_TLSLD:
-      if (ctx.got->tlsld_idx == -1) {
-        // Relax LD to LE
-        static const u8 insn[] = {
-          0x66, 0x66, 0x66,                         // (padding)
-          0x64, 0x48, 0x8b, 0x04, 0x25, 0, 0, 0, 0, // mov %fs:0, %rax
-        };
-        memcpy(loc - 3, insn, sizeof(insn));
-        i++;
-      } else {
-        write32s(ctx.got->get_tlsld_addr(ctx) + A - P);
-      }
-      continue;
-    case R_X86_64_DTPOFF32:
-      if (ctx.arg.relax && !ctx.arg.shared)
-        write32s(S + A - ctx.tls_end);
-      else
-        write32s(S + A - ctx.tls_begin);
-      continue;
-    case R_X86_64_DTPOFF64:
-      if (ctx.arg.relax && !ctx.arg.shared)
-        write64(S + A - ctx.tls_end);
-      else
-        write64(S + A - ctx.tls_begin);
-      continue;
-    case R_X86_64_TPOFF32:
-      write32s(S + A - ctx.tls_end);
-      continue;
-    case R_X86_64_TPOFF64:
-      write64(S + A - ctx.tls_end);
-      continue;
-    case R_X86_64_GOTTPOFF:
-      if (sym.get_gottp_idx(ctx) == -1) {
-        u32 insn = relax_gottpoff(loc - 3);
-        loc[-3] = insn >> 16;
-        loc[-2] = insn >> 8;
-        loc[-1] = insn;
-        write32s(S + A - ctx.tls_end + 4);
-      } else {
-        write32s(sym.get_gottp_addr(ctx) + A - P);
-      }
-      continue;
-    case R_X86_64_GOTPC32_TLSDESC:
-      if (sym.get_tlsdesc_idx(ctx) == -1) {
-        static const u8 insn[] = {
-          0x48, 0xc7, 0xc0, 0, 0, 0, 0, // mov $0, %rax
-        };
-        memcpy(loc - 3, insn, sizeof(insn));
-        write32s(S + A - ctx.tls_end + 4);
-      } else {
-        write32s(sym.get_tlsdesc_addr(ctx) + A - P);
-      }
-      continue;
-    case R_X86_64_SIZE32:
-      write32(sym.esym().st_size + A);
-      continue;
-    case R_X86_64_SIZE64:
-      write64(sym.esym().st_size + A);
-      continue;
-    case R_X86_64_TLSDESC_CALL:
-      if (sym.get_tlsdesc_idx(ctx) == -1) {
-        // call *(%rax) -> nop
-        loc[0] = 0x66;
-        loc[1] = 0x90;
-      }
-      continue;
-    default:
-      unreachable();
-    }
-
-#undef S
-#undef A
-#undef P
-#undef G
-#undef GOT
-  }
-}
-
-// This function is responsible for applying relocations against
-// non-SHF_ALLOC sections (i.e. sections that are not mapped to memory
-// at runtime).
-//
-// Relocations against non-SHF_ALLOC sections are much easier to
-// handle than that against SHF_ALLOC sections. It is because, since
-// they are not mapped to memory, they don't contain any variable or
-// function and never need PLT or GOT. Non-SHF_ALLOC sections are
-// mostly debug info sections.
-//
-// Relocations against non-SHF_ALLOC sections are not scanned by
-// scan_relocations.
-template <>
-void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_X86_64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = base + rel.r_offset;
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    SectionFragment<E> *frag;
-    i64 addend;
-    std::tie(frag, addend) = get_fragment(ctx, rel);
-
-    auto overflow_check = [&](i64 val, i64 lo, i64 hi) {
-      if (val < lo || hi <= val)
-        Error(ctx) << *this << ": relocation " << rel << " against "
-                   << sym << " out of range: " << val << " is not in ["
-                   << lo << ", " << hi << ")";
-    };
-
-    auto write8 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 8);
-      *loc = val;
-    };
-
-    auto write16 = [&](u64 val) {
-      overflow_check(val, 0, 1 << 16);
-      *(u16 *)loc = val;
-    };
-
-    auto write32 = [&](u64 val) {
-      overflow_check(val, 0, (i64)1 << 32);
-      *(u32 *)loc = val;
-    };
-
-    auto write32s = [&](u64 val) {
-      overflow_check(val, -((i64)1 << 31), (i64)1 << 31);
-      *(u32 *)loc = val;
-    };
-
-#define S (frag ? frag->get_addr(ctx) : sym.get_addr(ctx))
-#define A (frag ? addend : rel.r_addend)
-
-    switch (rel.r_type) {
-    case R_X86_64_8:
-      write8(S + A);
-      break;
-    case R_X86_64_16:
-      write16(S + A);
-      break;
-    case R_X86_64_32:
-      write32(S + A);
-      break;
-    case R_X86_64_32S:
-      write32s(S + A);
-      break;
-    case R_X86_64_64:
-      *(u64 *)loc = S + A;
-      break;
-    case R_X86_64_DTPOFF32:
-      write32s(S + A - ctx.tls_begin);
-      break;
-    case R_X86_64_DTPOFF64:
-      *(u64 *)loc = S + A - ctx.tls_begin;
-      break;
-    case R_X86_64_SIZE32:
-      write32(sym.esym().st_size + A);
-      break;
-    case R_X86_64_SIZE64:
-      *(u64 *)loc = sym.esym().st_size + A;
-      break;
-    default:
-      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
-                 << rel;
-      break;
-    }
-
-#undef S
-#undef A
-  }
-}
-
-// Linker has to create data structures in an output file to apply
-// some type of relocations. For example, if a relocation refers a GOT
-// or a PLT entry of a symbol, linker has to create an entry in .got
-// or in .plt for that symbol. In order to fix the file layout, we
-// need to scan relocations.
-template <>
-void InputSection<E>::scan_relocations(Context<E> &ctx) {
-  assert(shdr.sh_flags & SHF_ALLOC);
-
-  this->reldyn_offset = file.num_dynrel * sizeof(ElfRel<E>);
-  std::span<ElfRel<E>> rels = get_rels(ctx);
-
-  // Scan relocations
-  for (i64 i = 0; i < rels.size(); i++) {
-    const ElfRel<E> &rel = rels[i];
-    if (rel.r_type == R_X86_64_NONE)
-      continue;
-
-    Symbol<E> &sym = *file.symbols[rel.r_sym];
-    u8 *loc = (u8 *)(contents.data() + rel.r_offset);
-
-    if (!sym.file) {
-      report_undef(ctx, sym);
-      continue;
-    }
-
-    if (sym.get_type() == STT_GNU_IFUNC) {
-      sym.flags |= NEEDS_GOT;
-      sym.flags |= NEEDS_PLT;
-    }
-
-    switch (rel.r_type) {
-    case R_X86_64_8:
-    case R_X86_64_16:
-    case R_X86_64_32:
-    case R_X86_64_32S: {
-      // Dynamic linker does not support 8, 16 or 32-bit dynamic
-      // relocations for these types of relocations. We report an
-      // error if we cannot relocate them even at load-time.
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  NONE,     ERROR, ERROR,         ERROR },      // DSO
-        {  NONE,     ERROR, ERROR,         ERROR },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_X86_64_64: {
-      // Unlike the above, we can use R_X86_64_RELATIVE and R_86_64_64
-      // relocations.
-      Action table[][4] = {
-        // Absolute  Local    Imported data  Imported code
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // DSO
-        {  NONE,     BASEREL, DYNREL,        DYNREL },     // PIE
-        {  NONE,     NONE,    COPYREL,       PLT    },     // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_X86_64_PC8:
-    case R_X86_64_PC16:
-    case R_X86_64_PC32: {
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  ERROR,    NONE,  ERROR,         ERROR },      // DSO
-        {  ERROR,    NONE,  COPYREL,       PLT   },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_X86_64_PC64: {
-      Action table[][4] = {
-        // Absolute  Local  Imported data  Imported code
-        {  BASEREL,  NONE,  DYNREL,        DYNREL },     // DSO
-        {  BASEREL,  NONE,  COPYREL,       PLT   },      // PIE
-        {  NONE,     NONE,  COPYREL,       PLT   },      // PDE
-      };
-      dispatch(ctx, table, i, rel, sym);
-      break;
-    }
-    case R_X86_64_GOT32:
-    case R_X86_64_GOT64:
-    case R_X86_64_GOTPC32:
-    case R_X86_64_GOTPC64:
-    case R_X86_64_GOTPCREL:
-    case R_X86_64_GOTPCREL64:
-      sym.flags |= NEEDS_GOT;
-      break;
-    case R_X86_64_GOTPCRELX: {
-      if (rel.r_addend != -4)
-        Fatal(ctx) << *this << ": bad r_addend for R_X86_64_GOTPCRELX";
-
-      bool do_relax = ctx.arg.relax && !sym.is_imported &&
-                      sym.is_relative() && relax_gotpcrelx(loc - 2);
-      if (!do_relax)
-        sym.flags |= NEEDS_GOT;
-      break;
-    }
-    case R_X86_64_REX_GOTPCRELX: {
-      if (rel.r_addend != -4)
-        Fatal(ctx) << *this << ": bad r_addend for R_X86_64_REX_GOTPCRELX";
-
-      bool do_relax = ctx.arg.relax && !sym.is_imported &&
-                      sym.is_relative() && relax_rex_gotpcrelx(loc - 3);
-      if (!do_relax)
-        sym.flags |= NEEDS_GOT;
-      break;
-    }
-    case R_X86_64_PLT32:
-    case R_X86_64_PLTOFF64:
-      if (sym.is_imported)
-        sym.flags |= NEEDS_PLT;
-      break;
-    case R_X86_64_TLSGD:
-      if (i + 1 == rels.size())
-        Fatal(ctx) << *this
-                   << ": TLSGD reloc must be followed by PLT32 or GOTPCREL";
-
-      if (ctx.arg.relax && !ctx.arg.shared && !sym.is_imported)
-        i++;
-      else
-        sym.flags |= NEEDS_TLSGD;
-      break;
-    case R_X86_64_TLSLD:
-      if (i + 1 == rels.size())
-        Fatal(ctx) << *this
-                   << ": TLSLD reloc must be followed by PLT32 or GOTPCREL";
-
-      if (ctx.arg.relax && !ctx.arg.shared)
-        i++;
-      else
-        sym.flags |= NEEDS_TLSLD;
-      break;
-    case R_X86_64_GOTTPOFF: {
-      ctx.has_gottp_rel = true;
-
-      bool do_relax = ctx.arg.relax && !ctx.arg.shared &&
-                      !sym.is_imported && relax_gottpoff(loc - 3);
-      if (!do_relax)
-        sym.flags |= NEEDS_GOTTP;
-      break;
-    }
-    case R_X86_64_GOTPC32_TLSDESC: {
-      if (memcmp(loc - 3, "\x48\x8d\x05", 3))
-        Fatal(ctx) << *this << ": GOTPC32_TLSDESC relocation is used"
-                   << " against an invalid code sequence";
-
-      bool do_relax = ctx.relax_tlsdesc && !sym.is_imported;
-      if (!do_relax)
-        sym.flags |= NEEDS_TLSDESC;
-      break;
-    }
-    case R_X86_64_GOTOFF64:
-    case R_X86_64_DTPOFF32:
-    case R_X86_64_DTPOFF64:
-    case R_X86_64_TPOFF32:
-    case R_X86_64_TPOFF64:
-    case R_X86_64_SIZE32:
-    case R_X86_64_SIZE64:
-    case R_X86_64_TLSDESC_CALL:
-      break;
-    default:
-      Error(ctx) << *this << ": unknown relocation: " << rel;
-    }
-  }
-}
-
-} // namespace mold::elf

elf/cmdline.cc

@@ -1,973 +0,0 @@
-#include "mold.h"
-
-#include <sstream>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <unistd.h>
-#include <unordered_set>
-
-namespace mold::elf {
-
-static const char helpmsg[] = R"(
-Options:
-  --help                      Report usage information
-  -v, --version               Report version information
-  -V                          Report version and target information
-  -(, --start-group           Ignored
-  -), --end-group             Ignored
-  -C DIR, --directory DIR     Change to DIR before doing anything
-  -E, --export-dynamic        Put symbols in the dynamic symbol table
-    --no-export-dynamic
-  -F LIBNAME, --filter LIBNAME
-                              Set DT_FILTER to the specified value
-  -I FILE, --dynamic-linker FILE
-                              Set dynamic linker path
-    --no-dynamic-linker
-  -L DIR, --library-path DIR  Add DIR to library search path
-  -M, --print-map             Write map file to stdout
-  -N, --omagic                Do not page align data, do not make text readonly
-    --no-omagic
-  -O NUMBER                   Ignored
-  -S, --strip-debug           Strip .debug_* sections
-  -T FILE, --script FILE      Read linker script
-  -X, --discard-locals        Discard temporary local symbols
-  -e SYMBOL, --entry SYMBOL   Set program entry point
-  -f SHLIB, --auxiliary SHLIB Set DT_AUXILIARY to the specified value
-  -h LIBNAME, --soname LIBNAME
-                              Set shared library name
-  -l LIBNAME                  Search for a given library
-  -m TARGET                   Set target
-  -o FILE, --output FILE      Set output filename
-  -r, --relocatable           Generate relocatable output
-  -s, --strip-all             Strip .symtab section
-  -u SYMBOL, --undefined SYMBOL
-                              Force to resolve SYMBOL
-  --Bdynamic                  Link against shared libraries (default)
-  --Bstatic                   Do not link against shared libraries
-  --Bsymbolic                 Bind global symbols locally
-  --Bsymbolic-functions       Bind global functions locally
-  --Bno-symbolic              Cancel --Bsymbolic and --Bsymbolic-functions
-  --Map FILE                  Write map file to a given file
-  --allow-multiple-definition Allow multiple definitions
-  --as-needed                 Only set DT_NEEDED if used
-    --no-as-needed
-  --build-id [none,md5,sha1,sha256,uuid,HEXSTRING]
-                              Generate build ID
-    --no-build-id
-  --chroot DIR                Set a given path to root directory
-  --color-diagnostics=[auto,always,never]
-                              Use colors in diagnostics
-  --color-diagnostics         Alias for --color-diagnostics=always
-  --compress-debug-sections [none,zlib,zlib-gabi,zlib-gnu]
-                              Compress .debug_* sections
-  --dc                        Ignored
-  --defsym=SYMBOL=VALUE       Define a symbol alias
-  --demangle                  Demangle C++ symbols in log messages (default)
-    --no-demangle
-  --disable-new-dtags         Ignored
-  --dp                        Ignored
-  --dynamic-list              Read a list of dynamic symbols
-  --eh-frame-hdr              Create .eh_frame_hdr section
-    --no-eh-frame-hdr
-  --enable-new-dtags          Ignored
-  --exclude-libs LIB,LIB,..   Mark all symbols in given libraries hidden
-  --fatal-warnings            Ignored
-    --no-fatal-warnings       Ignored
-  --fini SYMBOL               Call SYMBOL at unload-time
-  --fork                      Spawn a child process (default)
-    --no-fork
-  --gc-sections               Remove unreferenced sections
-    --no-gc-sections
-  --gdb-index                 Ignored
-  --hash-style [sysv,gnu,both]
-                              Set hash style
-  --icf                       Fold identical code
-    --no-icf
-  --image-base ADDR           Set the base address to a given value
-  --init SYMBOL               Call SYMBOl at load-time
-  --no-undefined              Report undefined symbols (even with --shared)
-  --pack-dyn-relocs=[relr,none]
-                              Pack dynamic relocations
-  --perf                      Print performance statistics
-  --pie, --pic-executable     Create a position independent executable
-    --no-pie, --no-pic-executable
-  --plugin                    Ignored
-  --plugin-opt                Ignored
-  --pop-state                 Pop state of flags governing input file handling
-  --preload
-    --no-preload
-  --print-gc-sections         Print removed unreferenced sections
-    --no-print-gc-sections
-  --print-icf-sections        Print folded identical sections
-    --no-print-icf-sections
-  --push-state                Pop state of flags governing input file handling
-  --quick-exit                Use quick_exit to exit (default)
-    --no-quick-exit
-  --relax                     Optimize instructions (default)
-    --no-relax
-  --repro                     Embed input files to .repro section
-  --require-defined SYMBOL    Require SYMBOL be defined in the final output
-  --retain-symbols-file FILE  Keep only symbols listed in FILE
-  --rpath DIR                 Add DIR to runtime search path
-  --rpath-link DIR            Ignored
-  --run COMMAND ARG...        Run COMMAND with mold as /usr/bin/ld
-  --shared, --Bshareable      Create a share library
-  --sort-common               Ignored
-  --sort-section              Ignored
-  --spare-dynamic-tags NUMBER Reserve give number of tags in .dynamic section
-  --start-lib                 Give following object files in-archive-file semantics
-    --end-lib                 End the effect of --start-lib
-  --static                    Do not link against shared libraries
-  --stats                     Print input statistics
-  --sysroot DIR               Set target system root directory
-  --thread-count COUNT, --threads=COUNT
-                              Use COUNT number of threads
-  --threads                   Use multiple threads (default)
-    --no-threads
-  --trace                     Print name of each input file
-  --unique PATTERN            Don't merge input sections that match a given pattern
-  --unresolved-symbols [report-all,ignore-all,ignore-in-object-files,ignore-in-shared-libs]
-                              How to handle unresolved symbols
-  --version-script FILE       Read version script
-  --warn-common               Warn about common symbols
-    --no-warn-common
-  --warn-unresolved-symbols   Report unresolved symbols as warnings
-    --error-unresolved-symbols
-                              Report unresolved symbols as errors (default)
-  --whole-archive             Include all objects from static archives
-    --no-whole-archive
-  --wrap SYMBOL               Use wrapper function for a given symbol
-  -z defs                     Report undefined symbols (even with --shared)
-    -z nodefs
-  -z common-page-size=VALUE   Ignored
-  -z execstack                Require executable stack
-    -z noexecstack
-  -z initfirst                Mark DSO to be initialized first at runtime
-  -z interpose                Mark object to interpose all DSOs but executable
-  -z keep-text-section-prefix Keep .text.{hot,unknown,unlikely,startup,exit} as separate sections in the final binary
-    -z nokeep-text-section-prefix
-  -z lazy                     Enable lazy function resolution (default)
-  -z max-page-size=VALUE      Use VALUE as the memory page size
-  -z nocopyreloc              Do not create copy relocations
-  -z nodefaultlib             Make the dynamic loader to ignore default search paths
-  -z nodelete                 Mark DSO non-deletable at runtime
-  -z nodlopen                 Mark DSO not available to dlopen
-  -z nodump                   Mark DSO not available to dldump
-  -z now                      Disable lazy function resolution
-  -z origin                   Mark object requiring immediate $ORIGIN processing at runtime
-  -z separate-loadable-segments
-                              Separate all loadable segments to different pages
-    -z separate-code          Separate code and data into different pages
-    -z noseparate-code        Allow overlap in pages
-  -z relro                    Make some sections read-only after relocation (default)
-    -z norelro
-  -z text                     Report error if DT_TEXTREL is set
-    -z notext
-    -z textoff
-
-mold: supported targets: elf32-i386 elf64-x86-64 elf64-littleaarch64
-mold: supported emulations: elf_i386 elf_x86_64 aarch64linux aarch64elf)";
-
-static std::vector<std::string> add_dashes(std::string name) {
-  // Multi-letter linker options can be preceded by either a single
-  // dash or double dashes except ones starting with "o", which must
-  // be preceded by double dashes. For example, "-omagic" is
-  // interpreted as "-o magic". If you really want to specify the
-  // "omagic" option, you have to pass "--omagic".
-  if (name[0] == 'o')
-    return {"--" + name};
-  return {"-" + name, "--" + name};
-}
-
-template <typename E>
-bool read_arg(Context<E> &ctx, std::span<std::string_view> &args,
-              std::string_view &arg, std::string name) {
-  if (name.size() == 1) {
-    if (args[0] == "-" + name) {
-      if (args.size() == 1)
-        Fatal(ctx) << "option -" << name << ": argument missing";
-      arg = args[1];
-      args = args.subspan(2);
-      return true;
-    }
-
-    if (args[0].starts_with("-" + name)) {
-      arg = args[0].substr(name.size() + 1);
-      args = args.subspan(1);
-      return true;
-    }
-    return false;
-  }
-
-  for (std::string opt : add_dashes(name)) {
-    if (args[0] == opt) {
-      if (args.size() == 1)
-        Fatal(ctx) << "option -" << name << ": argument missing";
-      arg = args[1];
-      args = args.subspan(2);
-      return true;
-    }
-
-    if (args[0].starts_with(opt + "=")) {
-      arg = args[0].substr(opt.size() + 1);
-      args = args.subspan(1);
-      return true;
-    }
-  }
-  return false;
-}
-
-bool read_flag(std::span<std::string_view> &args, std::string name) {
-  for (std::string opt : add_dashes(name)) {
-    if (args[0] == opt) {
-      args = args.subspan(1);
-      return true;
-    }
-  }
-  return false;
-}
-
-static bool read_z_flag(std::span<std::string_view> &args, std::string name) {
-  if (args.size() >= 2 && args[0] == "-z" && args[1] == name) {
-    args = args.subspan(2);
-    return true;
-  }
-
-  if (!args.empty() && args[0] == "-z" + name) {
-    args = args.subspan(1);
-    return true;
-  }
-
-  return false;
-}
-
-template <typename E>
-bool read_z_arg(Context<E> &ctx, std::span<std::string_view> &args,
-                std::string_view &arg, std::string name) {
-  if (args.size() >= 2 && args[0] == "-z" && args[1].starts_with(name + "=")) {
-    arg = args[1].substr(name.size() + 1);
-    args = args.subspan(2);
-    return true;
-  }
-
-  if (!args.empty() && args[0].starts_with("-z" + name + "=")) {
-    arg = args[0].substr(name.size() + 3);
-    args = args.subspan(1);
-    return true;
-  }
-
-  return false;
-}
-
-template <typename E>
-std::string create_response_file(Context<E> &ctx) {
-  std::string buf;
-  std::stringstream out;
-
-  std::string cwd = std::filesystem::current_path();
-  out << "-C " << cwd.substr(1) << "\n";
-
-  if (cwd != "/") {
-    out << "--chroot ..";
-    i64 depth = std::count(cwd.begin(), cwd.end(), '/');
-    for (i64 i = 1; i < depth; i++)
-      out << "/..";
-    out << "\n";
-  }
-
-  for (i64 i = 1; i < ctx.cmdline_args.size(); i++) {
-    std::string_view arg = ctx.cmdline_args[i];
-
-    if (arg == "-repro" || arg == "--repro") {
-      i++;
-      continue;
-    }
-
-    out << arg << "\n";
-  }
-
-  return out.str();
-}
-
-template <typename E>
-static i64 parse_hex(Context<E> &ctx, std::string opt, std::string_view value) {
-  if (!value.starts_with("0x") && !value.starts_with("0X"))
-    Fatal(ctx) << "option -" << opt << ": not a hexadecimal number";
-  value = value.substr(2);
-  if (value.find_first_not_of("0123456789abcdefABCDEF") != value.npos)
-    Fatal(ctx) << "option -" << opt << ": not a hexadecimal number";
-  return std::stol(std::string(value), nullptr, 16);
-}
-
-template <typename E>
-static i64 parse_number(Context<E> &ctx, std::string opt,
-                        std::string_view value) {
-  size_t nread;
-  i64 ret = std::stol(std::string(value), &nread, 0);
-  if (value.size() != nread)
-    Fatal(ctx) << "option -" << opt << ": not a number: " << value;
-  return ret;
-}
-
-template <typename E>
-static std::vector<u8> parse_hex_build_id(Context<E> &ctx,
-                                          std::string_view arg) {
-  assert(arg.starts_with("0x") || arg.starts_with("0X"));
-
-  if (arg.size() % 2)
-    Fatal(ctx) << "invalid build-id: " << arg;
-  if (arg.substr(2).find_first_not_of("0123456789abcdefABCDEF") != arg.npos)
-    Fatal(ctx) << "invalid build-id: " << arg;
-
-  arg = arg.substr(2);
-
-  auto fn = [](char c) {
-    if ('0' <= c && c <= '9')
-      return c - '0';
-    if ('a' <= c && c <= 'f')
-      return c - 'a' + 10;
-    assert('A' <= c && c <= 'F');
-    return c - 'A' + 10;
-  };
-
-  std::vector<u8> vec(arg.size() / 2);
-  for (i64 i = 0; i < vec.size(); i++)
-    vec[i] = (fn(arg[i * 2]) << 4) | fn(arg[i * 2 + 1]);
-  return vec;
-}
-
-static std::vector<std::string_view>
-split_by_comma_or_colon(std::string_view str) {
-  std::vector<std::string_view> vec;
-
-  for (;;) {
-    i64 pos = str.find_first_of(",:");
-    if (pos == str.npos) {
-      vec.push_back(str);
-      break;
-    }
-    vec.push_back(str.substr(0, pos));
-    str = str.substr(pos);
-  }
-  return vec;
-}
-
-static std::string_view trim(std::string_view str) {
-  size_t pos = str.find_first_not_of(" \t");
-  if (pos == str.npos)
-    return "";
-  str = str.substr(pos);
-
-  pos = str.find_last_not_of(" \t");
-  if (pos == str.npos)
-    return str;
-  return str.substr(0, pos + 1);
-}
-
-template <typename E>
-static void read_retain_symbols_file(Context<E> &ctx, std::string_view path) {
-  MappedFile<Context<E>> *mf =
-    MappedFile<Context<E>>::must_open(ctx, std::string(path));
-  std::string_view data((char *)mf->data, mf->size);
-
-  ctx.arg.retain_symbols_file.reset(new std::unordered_set<std::string_view>);
-
-  while (!data.empty()) {
-    size_t pos = data.find('\n');
-    std::string_view name;
-
-    if (pos == data.npos) {
-      name = data;
-      data = "";
-    } else {
-      name = data.substr(0, pos);
-      data = data.substr(pos + 1);
-    }
-
-    name = trim(name);
-    if (!name.empty())
-      ctx.arg.retain_symbols_file->insert(name);
-  }
-}
-
-static bool is_file(std::string_view path) {
-  struct stat st;
-  return stat(std::string(path).c_str(), &st) == 0 &&
-         (st.st_mode & S_IFMT) != S_IFDIR;
-}
-
-// Returns a PLT header size and a PLT entry size.
-template <typename E>
-static std::pair<i64, i64> get_plt_size(Context<E> &ctx) {
-  switch (E::e_machine) {
-  case EM_X86_64:
-    if (ctx.arg.z_now)
-      return {0, 8};
-    if (ctx.arg.z_ibtplt)
-      return {16, 24};
-    return {16, 16};
-  case EM_386:
-    return {16, 16};
-  case EM_AARCH64:
-    return {32, 16};
-  }
-  unreachable();
-}
-
-template <typename E>
-void parse_nonpositional_args(Context<E> &ctx,
-                              std::vector<std::string_view> &remaining) {
-  std::span<std::string_view> args = ctx.cmdline_args;
-  args = args.subspan(1);
-
-  ctx.arg.color_diagnostics = isatty(STDERR_FILENO);
-  ctx.page_size = E::page_size;
-
-  bool version_shown = false;
-
-  while (!args.empty()) {
-    std::string_view arg;
-
-    if (read_flag(args, "help")) {
-      SyncOut(ctx) << "Usage: " << ctx.cmdline_args[0]
-                   << " [options] file...\n" << helpmsg;
-      exit(0);
-    }
-
-    if (read_arg(ctx, args, arg, "o") || read_arg(ctx, args, arg, "output")) {
-      ctx.arg.output = arg;
-    } else if (read_arg(ctx, args, arg, "dynamic-linker") ||
-               read_arg(ctx, args, arg, "I")) {
-      ctx.arg.dynamic_linker = arg;
-    } else if (read_flag(args, "no-dynamic-linker")) {
-      ctx.arg.dynamic_linker = "";
-    } else if (read_flag(args, "v")) {
-      SyncOut(ctx) << mold_version;
-      version_shown = true;
-    } else if (read_flag(args, "version")) {
-      SyncOut(ctx) << mold_version;
-      exit(0);
-    } else if (read_flag(args, "V")) {
-      SyncOut(ctx) << mold_version
-                   << "\n  Supported emulations:\n   elf_x86_64\n   elf_i386";
-      version_shown = true;
-    } else if (read_arg(ctx, args, arg, "m")) {
-      if (arg == "elf_x86_64") {
-        ctx.arg.emulation = EM_X86_64;
-      } else if (arg == "elf_i386") {
-        ctx.arg.emulation = EM_386;
-      } else if (arg == "aarch64linux") {
-        ctx.arg.emulation = EM_AARCH64;
-      } else {
-        Fatal(ctx) << "unknown -m argument: " << arg;
-      }
-    } else if (read_flag(args, "end-lib")) {
-      remaining.push_back("-end-lib");
-    } else if (read_flag(args, "export-dynamic") || read_flag(args, "E")) {
-      ctx.arg.export_dynamic = true;
-    } else if (read_flag(args, "no-export-dynamic")) {
-      ctx.arg.export_dynamic = false;
-    } else if (read_flag(args, "Bsymbolic")) {
-      ctx.arg.Bsymbolic = true;
-    } else if (read_flag(args, "Bsymbolic-functions")) {
-      ctx.arg.Bsymbolic_functions = true;
-    } else if (read_flag(args, "Bno-symbolic")) {
-      ctx.arg.Bsymbolic = false;
-      ctx.arg.Bsymbolic_functions = false;
-    } else if (read_arg(ctx, args, arg, "exclude-libs")) {
-      append(ctx.arg.exclude_libs, split_by_comma_or_colon(arg));
-    } else if (read_arg(ctx, args, arg, "e") ||
-               read_arg(ctx, args, arg, "entry")) {
-      ctx.arg.entry = arg;
-    } else if (read_arg(ctx, args, arg, "Map")) {
-      ctx.arg.Map = arg;
-      ctx.arg.print_map = true;
-    } else if (read_flag(args, "print-map") || read_flag(args, "M")) {
-      ctx.arg.print_map = true;
-    } else if (read_flag(args, "static") || read_flag(args, "Bstatic")) {
-      ctx.arg.is_static = true;
-      remaining.push_back("-Bstatic");
-    } else if (read_flag(args, "Bdynamic")) {
-      ctx.arg.is_static = false;
-      remaining.push_back("-Bdynamic");
-    } else if (read_flag(args, "shared") || read_flag(args, "Bshareable")) {
-      ctx.arg.shared = true;
-    } else if (read_arg(ctx, args, arg, "spare-dynamic-tags")) {
-      ctx.arg.spare_dynamic_tags = parse_number(ctx, "spare-dynamic-tags", arg);
-    } else if (read_flag(args, "start-lib")) {
-      remaining.push_back("-start-lib");
-    } else if (read_arg(ctx, args, arg, "defsym")) {
-      size_t pos = arg.find('=');
-      if (pos == arg.npos || pos == arg.size() - 1)
-        Fatal(ctx) << "-defsym: syntax error: " << arg;
-      ctx.arg.defsyms.push_back({arg.substr(0, pos), arg.substr(pos + 1)});
-    } else if (read_flag(args, "demangle")) {
-      ctx.arg.demangle = true;
-    } else if (read_flag(args, "no-demangle")) {
-      ctx.arg.demangle = false;
-    } else if (read_flag(args, "default-symver")) {
-      ctx.arg.default_symver = true;
-    } else if (read_arg(ctx, args, arg, "y") ||
-               read_arg(ctx, args, arg, "trace-symbol")) {
-      ctx.arg.trace_symbol.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "filler")) {
-      ctx.arg.filler = parse_hex(ctx, "filler", arg);
-    } else if (read_arg(ctx, args, arg, "L") ||
-               read_arg(ctx, args, arg, "library-path")) {
-      ctx.arg.library_paths.push_back(std::string(arg));
-    } else if (read_arg(ctx, args, arg, "sysroot")) {
-      ctx.arg.sysroot = arg;
-    } else if (read_arg(ctx, args, arg, "unique")) {
-      std::optional<GlobPattern> pat = GlobPattern::compile(arg);
-      if (!pat)
-        Fatal(ctx) << "-unique: invalid glob pattern: " << arg;
-      ctx.arg.unique = std::move(*pat);
-    } else if (read_arg(ctx, args, arg, "unresolved-symbols")) {
-      if (arg == "report-all" || arg == "ignore-in-shared-libs")
-        ctx.arg.unresolved_symbols = UNRESOLVED_ERROR;
-      else if (arg == "ignore-all" || arg == "ignore-in-object-files")
-        ctx.arg.unresolved_symbols = UNRESOLVED_IGNORE;
-      else
-        Fatal(ctx) << "unknown --unresolved-symbols argument: " << arg;
-    } else if (read_arg(ctx, args, arg, "u") ||
-               read_arg(ctx, args, arg, "undefined")) {
-      ctx.arg.undefined.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "require-defined")) {
-      ctx.arg.require_defined.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "init")) {
-      ctx.arg.init = arg;
-    } else if (read_arg(ctx, args, arg, "fini")) {
-      ctx.arg.fini = arg;
-    } else if (read_arg(ctx, args, arg, "hash-style")) {
-      if (arg == "sysv") {
-        ctx.arg.hash_style_sysv = true;
-        ctx.arg.hash_style_gnu = false;
-      } else if (arg == "gnu") {
-        ctx.arg.hash_style_sysv = false;
-        ctx.arg.hash_style_gnu = true;
-      } else if (arg == "both") {
-        ctx.arg.hash_style_sysv = true;
-        ctx.arg.hash_style_gnu = true;
-      } else {
-        Fatal(ctx) << "invalid --hash-style argument: " << arg;
-      }
-    } else if (read_arg(ctx, args, arg, "soname") ||
-               read_arg(ctx, args, arg, "h")) {
-      ctx.arg.soname = arg;
-    } else if (read_flag(args, "allow-multiple-definition")) {
-      ctx.arg.allow_multiple_definition = true;
-    } else if (read_flag(args, "trace")) {
-      ctx.arg.trace = true;
-    } else if (read_flag(args, "eh-frame-hdr")) {
-      ctx.arg.eh_frame_hdr = true;
-    } else if (read_flag(args, "no-eh-frame-hdr")) {
-      ctx.arg.eh_frame_hdr = false;
-    } else if (read_flag(args, "pie") || read_flag(args, "pic-executable")) {
-      ctx.arg.pic = true;
-      ctx.arg.pie = true;
-    } else if (read_flag(args, "no-pie") ||
-               read_flag(args, "no-pic-executable")) {
-      ctx.arg.pic = false;
-      ctx.arg.pie = false;
-    } else if (read_flag(args, "relax")) {
-      ctx.arg.relax = true;
-    } else if (read_flag(args, "no-relax")) {
-      ctx.arg.relax = false;
-    } else if (read_flag(args, "r") || read_flag(args, "relocatable")) {
-      ctx.arg.relocatable = true;
-    } else if (read_flag(args, "perf")) {
-      ctx.arg.perf = true;
-    } else if (read_flag(args, "pack-dyn-relocs=relr")) {
-      ctx.arg.pack_dyn_relocs_relr = true;
-    } else if (read_flag(args, "pack-dyn-relocs=none")) {
-      ctx.arg.pack_dyn_relocs_relr = false;
-    } else if (read_flag(args, "stats")) {
-      ctx.arg.stats = true;
-      Counter::enabled = true;
-    } else if (read_arg(ctx, args, arg, "C") ||
-               read_arg(ctx, args, arg, "directory")) {
-      ctx.arg.directory = arg;
-    } else if (read_arg(ctx, args, arg, "chroot")) {
-      ctx.arg.chroot = arg;
-    } else if (args[0] == "-color-diagnostics=auto" ||
-               args[0] == "--color-diagnostics=auto") {
-      ctx.arg.color_diagnostics = isatty(STDERR_FILENO);
-      args = args.subspan(1);
-    } else if (args[0] == "-color-diagnostics=always" ||
-               args[0] == "--color-diagnostics=always") {
-      ctx.arg.color_diagnostics = true;
-      args = args.subspan(1);
-    } else if (args[0] == "-color-diagnostics=never" ||
-               args[0] == "--color-diagnostics=never") {
-      ctx.arg.color_diagnostics = false;
-      args = args.subspan(1);
-    } else if (read_flag(args, "color-diagnostics")) {
-      ctx.arg.color_diagnostics = true;
-    } else if (read_flag(args, "warn-common")) {
-      ctx.arg.warn_common = true;
-    } else if (read_flag(args, "no-warn-common")) {
-      ctx.arg.warn_common = false;
-    } else if (read_arg(ctx, args, arg, "compress-debug-sections")) {
-      if (arg == "zlib" || arg == "zlib-gabi")
-        ctx.arg.compress_debug_sections = COMPRESS_GABI;
-      else if (arg == "zlib-gnu")
-        ctx.arg.compress_debug_sections = COMPRESS_GNU;
-      else if (arg == "none")
-        ctx.arg.compress_debug_sections = COMPRESS_NONE;
-      else
-        Fatal(ctx) << "invalid --compress-debug-sections argument: " << arg;
-    } else if (read_arg(ctx, args, arg, "wrap")) {
-      ctx.arg.wrap.insert(arg);
-    } else if (read_flag(args, "omagic") || read_flag(args, "N")) {
-      ctx.arg.omagic = true;
-      ctx.arg.is_static = true;
-    } else if (read_flag(args, "no-omagic")) {
-      ctx.arg.omagic = false;
-    } else if (read_arg(ctx, args, arg, "retain-symbols-file")) {
-      read_retain_symbols_file(ctx, arg);
-    } else if (read_flag(args, "repro")) {
-      ctx.arg.repro = true;
-    } else if (read_z_flag(args, "now")) {
-      ctx.arg.z_now = true;
-    } else if (read_z_flag(args, "lazy")) {
-      ctx.arg.z_now = false;
-    } else if (read_z_flag(args, "cet-report=none")) {
-      ctx.arg.z_cet_report = CET_REPORT_NONE;
-    } else if (read_z_flag(args, "cet-report=warning")) {
-      ctx.arg.z_cet_report = CET_REPORT_WARNING;
-    } else if (read_z_flag(args, "cet-report=error")) {
-      ctx.arg.z_cet_report = CET_REPORT_ERROR;
-    } else if (read_z_flag(args, "execstack")) {
-      ctx.arg.z_execstack = true;
-    } else if (read_z_arg(ctx, args, arg, "max-page-size")) {
-      ctx.page_size = parse_number(ctx, "-z max-page-size", arg);
-      if (__builtin_popcountll(ctx.page_size) != 1)
-        Fatal(ctx) << "-z max-page-size " << arg << ": value must be a power of 2";
-    } else if (read_z_flag(args, "noexecstack")) {
-      ctx.arg.z_execstack = false;
-    } else if (read_z_flag(args, "relro")) {
-      ctx.arg.z_relro = true;
-    } else if (read_z_flag(args, "norelro")) {
-      ctx.arg.z_relro = false;
-    } else if (read_z_flag(args, "defs")) {
-      ctx.arg.z_defs = true;
-    } else if (read_z_flag(args, "nodefs")) {
-      ctx.arg.z_defs = false;
-    } else if (read_z_flag(args, "nodlopen")) {
-      ctx.arg.z_dlopen = false;
-    } else if (read_z_flag(args, "nodelete")) {
-      ctx.arg.z_delete = false;
-    } else if (read_z_flag(args, "nocopyreloc")) {
-      ctx.arg.z_copyreloc = false;
-    } else if (read_z_flag(args, "nodump")) {
-      ctx.arg.z_dump = false;
-    } else if (read_z_flag(args, "initfirst")) {
-      ctx.arg.z_initfirst = true;
-    } else if (read_z_flag(args, "interpose")) {
-      ctx.arg.z_interpose = true;
-    } else if (read_z_flag(args, "ibt")) {
-      ctx.arg.z_ibt = true;
-      ctx.arg.z_ibtplt = true;
-    } else if (read_z_flag(args, "ibtplt")) {
-      ctx.arg.z_ibtplt = true;
-    } else if (read_z_flag(args, "muldefs")) {
-      ctx.arg.allow_multiple_definition = true;
-    } else if (read_z_flag(args, "keep-text-section-prefix")) {
-      ctx.arg.z_keep_text_section_prefix = true;
-    } else if (read_z_flag(args, "nokeep-text-section-prefix")) {
-      ctx.arg.z_keep_text_section_prefix = false;
-    } else if (read_z_flag(args, "shstk")) {
-      ctx.arg.z_shstk = true;
-    } else if (read_z_flag(args, "text")) {
-      ctx.arg.z_text = true;
-    } else if (read_z_flag(args, "notext") || read_z_flag(args, "textoff")) {
-      ctx.arg.z_text = false;
-    } else if (read_z_flag(args, "origin")) {
-      ctx.arg.z_origin = true;
-    } else if (read_z_flag(args, "nodefaultlib")) {
-      ctx.arg.z_nodefaultlib = true;
-    } else if (read_z_flag(args, "separate-loadable-segments")) {
-      ctx.arg.z_separate_code = SEPARATE_LOADABLE_SEGMENTS;
-    } else if (read_z_flag(args, "separate-code")) {
-      ctx.arg.z_separate_code = SEPARATE_CODE;
-    } else if (read_z_flag(args, "noseparate-code")) {
-      ctx.arg.z_separate_code = NOSEPARATE_CODE;
-    } else if (read_flag(args, "no-undefined")) {
-      ctx.arg.z_defs = true;
-    } else if (read_flag(args, "fatal-warnings")) {
-      ctx.arg.fatal_warnings = true;
-    } else if (read_flag(args, "no-fatal-warnings")) {
-      ctx.arg.fatal_warnings = false;
-    } else if (read_flag(args, "fork")) {
-      ctx.arg.fork = true;
-    } else if (read_flag(args, "no-fork")) {
-      ctx.arg.fork = false;
-    } else if (read_flag(args, "gc-sections")) {
-      ctx.arg.gc_sections = true;
-    } else if (read_flag(args, "no-gc-sections")) {
-      ctx.arg.gc_sections = false;
-    } else if (read_flag(args, "print-gc-sections")) {
-      ctx.arg.print_gc_sections = true;
-    } else if (read_flag(args, "no-print-gc-sections")) {
-      ctx.arg.print_gc_sections = false;
-    } else if (read_arg(ctx, args, arg, "icf")) {
-      if (arg == "all")
-        ctx.arg.icf = true;
-      else if (arg == "none")
-        ctx.arg.icf = false;
-      else
-        Fatal(ctx) << "unknown --icf argument: " << arg;
-    } else if (read_flag(args, "no-icf")) {
-      ctx.arg.icf = false;
-    } else if (read_arg(ctx, args, arg, "image-base")) {
-      ctx.arg.image_base = parse_number(ctx, "image-base", arg);
-    } else if (read_flag(args, "print-icf-sections")) {
-      ctx.arg.print_icf_sections = true;
-    } else if (read_flag(args, "no-print-icf-sections")) {
-      ctx.arg.print_icf_sections = false;
-    } else if (read_flag(args, "quick-exit")) {
-      ctx.arg.quick_exit = true;
-    } else if (read_flag(args, "no-quick-exit")) {
-      ctx.arg.quick_exit = false;
-    } else if (read_arg(ctx, args, arg, "thread-count")) {
-      ctx.arg.thread_count = parse_number(ctx, "thread-count", arg);
-    } else if (read_flag(args, "threads")) {
-      ctx.arg.thread_count = 0;
-    } else if (read_flag(args, "no-threads")) {
-      ctx.arg.thread_count = 1;
-    } else if (args[0].starts_with("-threads=")) {
-      ctx.arg.thread_count = parse_number(ctx, "threads=", args[0].substr(9));
-      args = args.subspan(1);
-    } else if (args[0].starts_with("--threads=")) {
-      ctx.arg.thread_count = parse_number(ctx, "threads=", args[0].substr(10));
-      args = args.subspan(1);
-    } else if (read_flag(args, "discard-all") || read_flag(args, "x")) {
-      ctx.arg.discard_all = true;
-    } else if (read_flag(args, "discard-locals") || read_flag(args, "X")) {
-      ctx.arg.discard_locals = true;
-    } else if (read_flag(args, "strip-all") || read_flag(args, "s")) {
-      ctx.arg.strip_all = true;
-    } else if (read_flag(args, "strip-debug") || read_flag(args, "S")) {
-      ctx.arg.strip_all = true;
-    } else if (read_flag(args, "warn-unresolved-symbols")) {
-      ctx.arg.unresolved_symbols = UNRESOLVED_WARN;
-    } else if (read_flag(args, "error-unresolved-symbols")) {
-      ctx.arg.unresolved_symbols = UNRESOLVED_ERROR;
-    } else if (read_arg(ctx, args, arg, "rpath")) {
-      if (!ctx.arg.rpaths.empty())
-        ctx.arg.rpaths += ":";
-      ctx.arg.rpaths += arg;
-    } else if (read_arg(ctx, args, arg, "R")) {
-      if (is_file(arg))
-        Fatal(ctx) << "-R" << arg
-                   << ": -R as an alias for --just-symbols is not supported";
-
-      if (!ctx.arg.rpaths.empty())
-        ctx.arg.rpaths += ":";
-      ctx.arg.rpaths += arg;
-    } else if (read_flag(args, "build-id")) {
-      ctx.arg.build_id.kind = BuildId::HASH;
-      ctx.arg.build_id.hash_size = 20;
-    } else if (read_arg(ctx, args, arg, "build-id")) {
-      if (arg == "none") {
-        ctx.arg.build_id.kind = BuildId::NONE;
-      } else if (arg == "uuid") {
-        ctx.arg.build_id.kind = BuildId::UUID;
-      } else if (arg == "md5") {
-        ctx.arg.build_id.kind = BuildId::HASH;
-        ctx.arg.build_id.hash_size = 16;
-      } else if (arg == "sha1") {
-        ctx.arg.build_id.kind = BuildId::HASH;
-        ctx.arg.build_id.hash_size = 20;
-      } else if (arg == "sha256") {
-        ctx.arg.build_id.kind = BuildId::HASH;
-        ctx.arg.build_id.hash_size = 32;
-      } else if (arg.starts_with("0x") || arg.starts_with("0X")) {
-        ctx.arg.build_id.kind = BuildId::HEX;
-        ctx.arg.build_id.value = parse_hex_build_id(ctx, arg);
-      } else {
-        Fatal(ctx) << "invalid --build-id argument: " << arg;
-      }
-    } else if (read_flag(args, "no-build-id")) {
-      ctx.arg.build_id.kind = BuildId::NONE;
-    } else if (read_arg(ctx, args, arg, "format") ||
-               read_arg(ctx, args, arg, "b")) {
-      if (arg == "binary")
-        Fatal(ctx)
-          << "mold does not suppor `-b binary`. If you want to convert a binary"
-          " file into an object file, use `objcopy -I binary -O elf64-x86-64"
-          << " <input-file> <output-file.o>` instead.";
-      Fatal(ctx) << "unknown command line option: -b " << arg;
-    } else if (read_arg(ctx, args, arg, "auxiliary") ||
-               read_arg(ctx, args, arg, "f")) {
-      ctx.arg.auxiliary.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "filter") ||
-               read_arg(ctx, args, arg, "F")) {
-      ctx.arg.filter.push_back(arg);
-    } else if (read_flag(args, "preload")) {
-      ctx.arg.preload = true;
-    } else if (read_flag(args, "no-preload")) {
-      ctx.arg.preload = false;
-    } else if (read_flag(args, "apply-dynamic-relocs")) {
-    } else if (read_arg(ctx, args, arg, "O")) {
-    } else if (read_flag(args, "O0")) {
-    } else if (read_flag(args, "O1")) {
-    } else if (read_flag(args, "O2")) {
-    } else if (read_flag(args, "verbose")) {
-    } else if (read_arg(ctx, args, arg, "plugin")) {
-    } else if (read_arg(ctx, args, arg, "plugin-opt")) {
-    } else if (read_flag(args, "color-diagnostics")) {
-    } else if (read_flag(args, "gdb-index")) {
-    } else if (read_flag(args, "eh-frame-hdr")) {
-    } else if (read_flag(args, "start-group")) {
-    } else if (read_flag(args, "end-group")) {
-    } else if (read_flag(args, "(")) {
-    } else if (read_flag(args, ")")) {
-    } else if (read_flag(args, "fatal-warnings")) {
-    } else if (read_flag(args, "enable-new-dtags")) {
-    } else if (read_flag(args, "disable-new-dtags")) {
-    } else if (read_flag(args, "nostdlib")) {
-    } else if (read_flag(args, "allow-shlib-undefined")) {
-    } else if (read_flag(args, "no-allow-shlib-undefined")) {
-    } else if (read_flag(args, "no-add-needed")) {
-    } else if (read_flag(args, "no-call-graph-profile-sort")) {
-    } else if (read_flag(args, "no-copy-dt-needed-entries")) {
-    } else if (read_flag(args, "no-undefined-version")) {
-    } else if (read_arg(ctx, args, arg, "sort-section")) {
-    } else if (read_flag(args, "sort-common")) {
-    } else if (read_flag(args, "dc")) {
-    } else if (read_flag(args, "dp")) {
-    } else if (read_flag(args, "fix-cortex-a53-835769")) {
-    } else if (read_flag(args, "fix-cortex-a53-843419")) {
-    } else if (read_flag(args, "EL")) {
-    } else if (read_flag(args, "warn-once")) {
-    } else if (read_flag(args, "nodefaultlibs")) {
-    } else if (read_flag(args, "warn-constructors")) {
-    } else if (read_flag(args, "warn-execstack")) {
-    } else if (read_flag(args, "no-warn-execstack")) {
-    } else if (read_arg(ctx, args, arg, "rpath-link")) {
-    } else if (read_z_flag(args, "combreloc")) {
-    } else if (read_z_flag(args, "nocombreloc")) {
-    } else if (read_z_arg(ctx, args, arg, "common-page-size")) {
-    } else if (read_arg(ctx, args, arg, "version-script")) {
-      remaining.push_back("--version-script");
-      remaining.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "dynamic-list")) {
-      remaining.push_back("--dynamic-list");
-      remaining.push_back(arg);
-    } else if (read_flag(args, "as-needed")) {
-      remaining.push_back("-as-needed");
-    } else if (read_flag(args, "no-as-needed")) {
-      remaining.push_back("-no-as-needed");
-    } else if (read_flag(args, "whole-archive")) {
-      remaining.push_back("-whole-archive");
-    } else if (read_flag(args, "no-whole-archive")) {
-      remaining.push_back("-no-whole-archive");
-    } else if (read_arg(ctx, args, arg, "l")) {
-      remaining.push_back("-l");
-      remaining.push_back(arg);
-    } else if (read_arg(ctx, args, arg, "script") ||
-               read_arg(ctx, args, arg, "T")) {
-      remaining.push_back(arg);
-    } else if (read_flag(args, "push-state")) {
-      remaining.push_back("-push-state");
-    } else if (read_flag(args, "pop-state")) {
-      remaining.push_back("-pop-state");
-    } else if (args[0].starts_with("-z") && args[0].size() > 2) {
-      Warn(ctx) << "unknown command line option: " << args[0];
-      args = args.subspan(1);
-    } else if (args[0] == "-z" && args.size() >= 2) {
-      Warn(ctx) << "unknown command line option: -z " << args[1];
-      args = args.subspan(2);
-    } else {
-      if (args[0][0] == '-')
-        Fatal(ctx) << "unknown command line option: " << args[0];
-      remaining.push_back(args[0]);
-      args = args.subspan(1);
-    }
-  }
-
-  if (!ctx.arg.sysroot.empty()) {
-    for (std::string &path : ctx.arg.library_paths) {
-      if (std::string_view(path).starts_with('='))
-        path = ctx.arg.sysroot + path.substr(1);
-      else if (std::string_view(path).starts_with("$SYSROOT"))
-        path = ctx.arg.sysroot + path.substr(8);
-    }
-  }
-
-  if (ctx.arg.shared) {
-    ctx.arg.pic = true;
-    ctx.arg.dynamic_linker = "";
-  }
-
-  if (ctx.arg.pic)
-    ctx.arg.image_base = 0;
-
-  if (ctx.arg.retain_symbols_file) {
-    ctx.arg.strip_all = false;
-    ctx.arg.discard_all = false;
-  }
-
-  if (ctx.arg.relocatable)
-    ctx.arg.is_static = true;
-
-  if (!ctx.arg.shared) {
-    if (!ctx.arg.filter.empty())
-      Fatal(ctx) << "-filter may not be used without -shared";
-    if (!ctx.arg.auxiliary.empty())
-      Fatal(ctx) << "-auxiliary may not be used without -shared";
-  }
-
-  if (ctx.arg.image_base % ctx.page_size)
-    Fatal(ctx) << "-image-base msut be a multiple of -max-page-size";
-
-  if (char *env = getenv("MOLD_REPRO"); env && env[0])
-    ctx.arg.repro = true;
-
-  if (ctx.arg.output.empty())
-    ctx.arg.output = "a.out";
-
-  if (ctx.arg.shared || ctx.arg.export_dynamic)
-    ctx.default_version = VER_NDX_GLOBAL;
-  else
-    ctx.default_version = VER_NDX_LOCAL;
-
-  if (ctx.arg.default_symver) {
-    std::string ver = ctx.arg.soname.empty() ?
-      filepath(ctx.arg.output).filename().string() : std::string(ctx.arg.soname);
-    ctx.arg.version_definitions.push_back(ver);
-    ctx.default_version = VER_NDX_LAST_RESERVED + 1;
-  }
-
-  std::tie(ctx.plt_hdr_size, ctx.plt_size) = get_plt_size(ctx);
-
-  ctx.arg.undefined.push_back(ctx.arg.entry);
-
-  // TLSDESC relocs must be always relaxed for statically-linked
-  // executables even if -no-relax is given. It is because a
-  // statically-linked executable doesn't contain a tranpoline
-  // function needed for TLSDESC.
-  ctx.relax_tlsdesc = ctx.arg.is_static || (ctx.arg.relax && !ctx.arg.shared);
-
-  if (version_shown && remaining.empty())
-    exit(0);
-}
-
-#define INSTANTIATE(E)                                                  \
-  template                                                              \
-  bool read_arg(Context<E> &ctx, std::span<std::string_view> &args,     \
-                std::string_view &arg,                                  \
-                std::string name);                                      \
-                                                                        \
-  template std::string create_response_file(Context<E> &ctx);           \
-                                                                        \
-  template                                                              \
-  void parse_nonpositional_args(Context<E> &ctx,                        \
-                                std::vector<std::string_view> &remaining)
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/elf.h

@@ -1,953 +0,0 @@
-#pragma once
-
-#include <cstdint>
-#include <ostream>
-#include <string>
-
-#ifdef __BIG_ENDIAN__
-#error "mold does not support big-endian hosts"
-#endif
-
-namespace mold::elf {
-
-typedef uint8_t u8;
-typedef uint16_t u16;
-typedef uint32_t u32;
-typedef uint64_t u64;
-
-typedef int8_t i8;
-typedef int16_t i16;
-typedef int32_t i32;
-typedef int64_t i64;
-
-struct X86_64;
-struct I386;
-struct ARM64;
-
-template <typename E> struct ElfSym;
-template <typename E> struct ElfShdr;
-template <typename E> struct ElfEhdr;
-template <typename E> struct ElfPhdr;
-template <typename E> struct ElfRel;
-template <typename E> struct ElfDyn;
-template <typename E> struct ElfChdr;
-
-template <typename E>
-std::string rel_to_string(u32 r_type);
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const ElfRel<E> &rel) {
-  out << rel_to_string<E>(rel.r_type);
-  return out;
-}
-
-static constexpr u32 SHN_UNDEF = 0;
-static constexpr u32 SHN_ABS = 0xfff1;
-static constexpr u32 SHN_COMMON = 0xfff2;
-static constexpr u32 SHN_XINDEX = 0xffff;
-
-static constexpr u32 SHT_NULL = 0;
-static constexpr u32 SHT_PROGBITS = 1;
-static constexpr u32 SHT_SYMTAB = 2;
-static constexpr u32 SHT_STRTAB = 3;
-static constexpr u32 SHT_RELA = 4;
-static constexpr u32 SHT_HASH = 5;
-static constexpr u32 SHT_DYNAMIC = 6;
-static constexpr u32 SHT_NOTE = 7;
-static constexpr u32 SHT_NOBITS = 8;
-static constexpr u32 SHT_REL = 9;
-static constexpr u32 SHT_SHLIB = 10;
-static constexpr u32 SHT_DYNSYM = 11;
-static constexpr u32 SHT_INIT_ARRAY = 14;
-static constexpr u32 SHT_FINI_ARRAY = 15;
-static constexpr u32 SHT_PREINIT_ARRAY = 16;
-static constexpr u32 SHT_GROUP = 17;
-static constexpr u32 SHT_SYMTAB_SHNDX = 18;
-static constexpr u32 SHT_RELR = 19;
-static constexpr u32 SHT_GNU_HASH = 0x6ffffff6;
-static constexpr u32 SHT_GNU_VERDEF = 0x6ffffffd;
-static constexpr u32 SHT_GNU_VERNEED = 0x6ffffffe;
-static constexpr u32 SHT_GNU_VERSYM = 0x6fffffff;
-static constexpr u32 SHT_X86_64_UNWIND = 0x70000001;
-
-static constexpr u32 SHF_WRITE = 0x1;
-static constexpr u32 SHF_ALLOC = 0x2;
-static constexpr u32 SHF_EXECINSTR = 0x4;
-static constexpr u32 SHF_MERGE = 0x10;
-static constexpr u32 SHF_STRINGS = 0x20;
-static constexpr u32 SHF_INFO_LINK = 0x40;
-static constexpr u32 SHF_LINK_ORDER = 0x80;
-static constexpr u32 SHF_GROUP = 0x200;
-static constexpr u32 SHF_TLS = 0x400;
-static constexpr u32 SHF_COMPRESSED = 0x800;
-static constexpr u32 SHF_EXCLUDE = 0x80000000;
-
-static constexpr u32 GRP_COMDAT = 1;
-
-static constexpr u32 STT_NOTYPE = 0;
-static constexpr u32 STT_OBJECT = 1;
-static constexpr u32 STT_FUNC = 2;
-static constexpr u32 STT_SECTION = 3;
-static constexpr u32 STT_FILE = 4;
-static constexpr u32 STT_COMMON = 5;
-static constexpr u32 STT_TLS = 6;
-static constexpr u32 STT_GNU_IFUNC = 10;
-
-static constexpr u32 STB_LOCAL = 0;
-static constexpr u32 STB_GLOBAL = 1;
-static constexpr u32 STB_WEAK = 2;
-static constexpr u32 STB_GNU_UNIQUE = 10;
-
-static constexpr u32 STV_DEFAULT = 0;
-static constexpr u32 STV_INTERNAL = 1;
-static constexpr u32 STV_HIDDEN = 2;
-static constexpr u32 STV_PROTECTED = 3;
-
-static constexpr u32 VER_NDX_LOCAL = 0;
-static constexpr u32 VER_NDX_GLOBAL = 1;
-static constexpr u32 VER_NDX_LAST_RESERVED = 1;
-
-static constexpr u32 VER_FLG_BASE = 1;
-static constexpr u32 VER_FLG_WEAK = 2;
-static constexpr u32 VER_FLG_INFO = 4;
-
-static constexpr u32 VERSYM_HIDDEN = 0x8000;
-
-static constexpr u32 PT_NULL = 0;
-static constexpr u32 PT_LOAD = 1;
-static constexpr u32 PT_DYNAMIC = 2;
-static constexpr u32 PT_INTERP = 3;
-static constexpr u32 PT_NOTE = 4;
-static constexpr u32 PT_SHLIB = 5;
-static constexpr u32 PT_PHDR = 6;
-static constexpr u32 PT_TLS = 7;
-static constexpr u32 PT_GNU_EH_FRAME = 0x6474e550;
-static constexpr u32 PT_GNU_STACK = 0x6474e551;
-static constexpr u32 PT_GNU_RELRO = 0x6474e552;
-
-static constexpr u32 PF_X = 1;
-static constexpr u32 PF_W = 2;
-static constexpr u32 PF_R = 4;
-
-static constexpr u32 ET_NONE = 0;
-static constexpr u32 ET_REL = 1;
-static constexpr u32 ET_EXEC = 2;
-static constexpr u32 ET_DYN = 3;
-
-static constexpr u32 ELFDATA2LSB = 1;
-static constexpr u32 ELFDATA2MSB = 2;
-
-static constexpr u32 ELFCLASS32 = 1;
-static constexpr u32 ELFCLASS64 = 2;
-
-static constexpr u32 EV_CURRENT = 1;
-
-static constexpr u32 EM_386 = 3;
-static constexpr u32 EM_X86_64 = 62;
-static constexpr u32 EM_AARCH64 = 183;
-
-static constexpr u32 EI_CLASS = 4;
-static constexpr u32 EI_DATA = 5;
-static constexpr u32 EI_VERSION = 6;
-static constexpr u32 EI_OSABI = 7;
-static constexpr u32 EI_ABIVERSION = 8;
-
-static constexpr u32 ELFOSABI_NONE = 0;
-static constexpr u32 ELFOSABI_GNU = 3;
-
-static constexpr u32 DT_NULL = 0;
-static constexpr u32 DT_NEEDED = 1;
-static constexpr u32 DT_PLTRELSZ = 2;
-static constexpr u32 DT_PLTGOT = 3;
-static constexpr u32 DT_HASH = 4;
-static constexpr u32 DT_STRTAB = 5;
-static constexpr u32 DT_SYMTAB = 6;
-static constexpr u32 DT_RELA = 7;
-static constexpr u32 DT_RELASZ = 8;
-static constexpr u32 DT_RELAENT = 9;
-static constexpr u32 DT_STRSZ = 10;
-static constexpr u32 DT_SYMENT = 11;
-static constexpr u32 DT_INIT = 12;
-static constexpr u32 DT_FINI = 13;
-static constexpr u32 DT_SONAME = 14;
-static constexpr u32 DT_RPATH = 15;
-static constexpr u32 DT_SYMBOLIC = 16;
-static constexpr u32 DT_REL = 17;
-static constexpr u32 DT_RELSZ = 18;
-static constexpr u32 DT_RELENT = 19;
-static constexpr u32 DT_PLTREL = 20;
-static constexpr u32 DT_DEBUG = 21;
-static constexpr u32 DT_TEXTREL = 22;
-static constexpr u32 DT_JMPREL = 23;
-static constexpr u32 DT_BIND_NOW = 24;
-static constexpr u32 DT_INIT_ARRAY = 25;
-static constexpr u32 DT_FINI_ARRAY = 26;
-static constexpr u32 DT_INIT_ARRAYSZ = 27;
-static constexpr u32 DT_FINI_ARRAYSZ = 28;
-static constexpr u32 DT_RUNPATH = 29;
-static constexpr u32 DT_FLAGS = 30;
-static constexpr u32 DT_RELRSZ = 35;
-static constexpr u32 DT_RELR = 36;
-static constexpr u32 DT_RELRENT = 37;
-static constexpr u32 DT_GNU_HASH = 0x6ffffef5;
-static constexpr u32 DT_VERSYM = 0x6ffffff0;
-static constexpr u32 DT_RELACOUNT = 0x6ffffff9;
-static constexpr u32 DT_RELCOUNT = 0x6ffffffa;
-static constexpr u32 DT_FLAGS_1 = 0x6ffffffb;
-static constexpr u32 DT_VERDEF = 0x6ffffffc;
-static constexpr u32 DT_VERDEFNUM = 0x6ffffffd;
-static constexpr u32 DT_VERNEED = 0x6ffffffe;
-static constexpr u32 DT_VERNEEDNUM = 0x6fffffff;
-static constexpr u32 DT_AUXILIARY = 0x7ffffffd;
-static constexpr u32 DT_FILTER = 0x7fffffff;
-
-static constexpr u32 DF_ORIGIN = 0x01;
-static constexpr u32 DF_SYMBOLIC = 0x02;
-static constexpr u32 DF_TEXTREL = 0x04;
-static constexpr u32 DF_BIND_NOW = 0x08;
-static constexpr u32 DF_STATIC_TLS = 0x10;
-
-static constexpr u32 DF_1_NOW = 0x00000001;
-static constexpr u32 DF_1_NODELETE = 0x00000008;
-static constexpr u32 DF_1_INITFIRST = 0x00000020;
-static constexpr u32 DF_1_NOOPEN = 0x00000040;
-static constexpr u32 DF_1_ORIGIN = 0x00000080;
-static constexpr u32 DF_1_INTERPOSE = 0x00000400;
-static constexpr u32 DF_1_NODEFLIB = 0x00000800;
-static constexpr u32 DF_1_NODUMP = 0x00001000;
-static constexpr u32 DF_1_PIE = 0x08000000;
-
-static constexpr u32 NT_GNU_ABI_TAG = 1;
-static constexpr u32 NT_GNU_HWCAP = 2;
-static constexpr u32 NT_GNU_BUILD_ID = 3;
-static constexpr u32 NT_GNU_GOLD_VERSION = 4;
-static constexpr u32 NT_GNU_PROPERTY_TYPE_0 = 5;
-
-static constexpr u32 GNU_PROPERTY_AARCH64_FEATURE_1_AND = 0xc0000000;
-static constexpr u32 GNU_PROPERTY_X86_FEATURE_1_AND = 0xc0000002;
-
-static constexpr u32 GNU_PROPERTY_X86_FEATURE_1_IBT = 1;
-static constexpr u32 GNU_PROPERTY_X86_FEATURE_1_SHSTK = 2;
-
-static constexpr u32 ELFCOMPRESS_ZLIB = 1;
-
-static constexpr u32 R_X86_64_NONE = 0;
-static constexpr u32 R_X86_64_64 = 1;
-static constexpr u32 R_X86_64_PC32 = 2;
-static constexpr u32 R_X86_64_GOT32 = 3;
-static constexpr u32 R_X86_64_PLT32 = 4;
-static constexpr u32 R_X86_64_COPY = 5;
-static constexpr u32 R_X86_64_GLOB_DAT = 6;
-static constexpr u32 R_X86_64_JUMP_SLOT = 7;
-static constexpr u32 R_X86_64_RELATIVE = 8;
-static constexpr u32 R_X86_64_GOTPCREL = 9;
-static constexpr u32 R_X86_64_32 = 10;
-static constexpr u32 R_X86_64_32S = 11;
-static constexpr u32 R_X86_64_16 = 12;
-static constexpr u32 R_X86_64_PC16 = 13;
-static constexpr u32 R_X86_64_8 = 14;
-static constexpr u32 R_X86_64_PC8 = 15;
-static constexpr u32 R_X86_64_DTPMOD64 = 16;
-static constexpr u32 R_X86_64_DTPOFF64 = 17;
-static constexpr u32 R_X86_64_TPOFF64 = 18;
-static constexpr u32 R_X86_64_TLSGD = 19;
-static constexpr u32 R_X86_64_TLSLD = 20;
-static constexpr u32 R_X86_64_DTPOFF32 = 21;
-static constexpr u32 R_X86_64_GOTTPOFF = 22;
-static constexpr u32 R_X86_64_TPOFF32 = 23;
-static constexpr u32 R_X86_64_PC64 = 24;
-static constexpr u32 R_X86_64_GOTOFF64 = 25;
-static constexpr u32 R_X86_64_GOTPC32 = 26;
-static constexpr u32 R_X86_64_GOT64 = 27;
-static constexpr u32 R_X86_64_GOTPCREL64 = 28;
-static constexpr u32 R_X86_64_GOTPC64 = 29;
-static constexpr u32 R_X86_64_GOTPLT64 = 30;
-static constexpr u32 R_X86_64_PLTOFF64 = 31;
-static constexpr u32 R_X86_64_SIZE32 = 32;
-static constexpr u32 R_X86_64_SIZE64 = 33;
-static constexpr u32 R_X86_64_GOTPC32_TLSDESC = 34;
-static constexpr u32 R_X86_64_TLSDESC_CALL = 35;
-static constexpr u32 R_X86_64_TLSDESC = 36;
-static constexpr u32 R_X86_64_IRELATIVE = 37;
-static constexpr u32 R_X86_64_GOTPCRELX = 41;
-static constexpr u32 R_X86_64_REX_GOTPCRELX = 42;
-
-template <>
-inline std::string rel_to_string<X86_64>(u32 r_type) {
-  switch (r_type) {
-  case R_X86_64_NONE: return "R_X86_64_NONE";
-  case R_X86_64_64: return "R_X86_64_64";
-  case R_X86_64_PC32: return "R_X86_64_PC32";
-  case R_X86_64_GOT32: return "R_X86_64_GOT32";
-  case R_X86_64_PLT32: return "R_X86_64_PLT32";
-  case R_X86_64_COPY: return "R_X86_64_COPY";
-  case R_X86_64_GLOB_DAT: return "R_X86_64_GLOB_DAT";
-  case R_X86_64_JUMP_SLOT: return "R_X86_64_JUMP_SLOT";
-  case R_X86_64_RELATIVE: return "R_X86_64_RELATIVE";
-  case R_X86_64_GOTPCREL: return "R_X86_64_GOTPCREL";
-  case R_X86_64_32: return "R_X86_64_32";
-  case R_X86_64_32S: return "R_X86_64_32S";
-  case R_X86_64_16: return "R_X86_64_16";
-  case R_X86_64_PC16: return "R_X86_64_PC16";
-  case R_X86_64_8: return "R_X86_64_8";
-  case R_X86_64_PC8: return "R_X86_64_PC8";
-  case R_X86_64_DTPMOD64: return "R_X86_64_DTPMOD64";
-  case R_X86_64_DTPOFF64: return "R_X86_64_DTPOFF64";
-  case R_X86_64_TPOFF64: return "R_X86_64_TPOFF64";
-  case R_X86_64_TLSGD: return "R_X86_64_TLSGD";
-  case R_X86_64_TLSLD: return "R_X86_64_TLSLD";
-  case R_X86_64_DTPOFF32: return "R_X86_64_DTPOFF32";
-  case R_X86_64_GOTTPOFF: return "R_X86_64_GOTTPOFF";
-  case R_X86_64_TPOFF32: return "R_X86_64_TPOFF32";
-  case R_X86_64_PC64: return "R_X86_64_PC64";
-  case R_X86_64_GOTOFF64: return "R_X86_64_GOTOFF64";
-  case R_X86_64_GOTPC32: return "R_X86_64_GOTPC32";
-  case R_X86_64_GOT64: return "R_X86_64_GOT64";
-  case R_X86_64_GOTPCREL64: return "R_X86_64_GOTPCREL64";
-  case R_X86_64_GOTPC64: return "R_X86_64_GOTPC64";
-  case R_X86_64_GOTPLT64: return "R_X86_64_GOTPLT64";
-  case R_X86_64_PLTOFF64: return "R_X86_64_PLTOFF64";
-  case R_X86_64_SIZE32: return "R_X86_64_SIZE32";
-  case R_X86_64_SIZE64: return "R_X86_64_SIZE64";
-  case R_X86_64_GOTPC32_TLSDESC: return "R_X86_64_GOTPC32_TLSDESC";
-  case R_X86_64_TLSDESC_CALL: return "R_X86_64_TLSDESC_CALL";
-  case R_X86_64_TLSDESC: return "R_X86_64_TLSDESC";
-  case R_X86_64_IRELATIVE: return "R_X86_64_IRELATIVE";
-  case R_X86_64_GOTPCRELX: return "R_X86_64_GOTPCRELX";
-  case R_X86_64_REX_GOTPCRELX: return "R_X86_64_REX_GOTPCRELX";
-  }
-  return "unknown (" + std::to_string(r_type) + ")";
-}
-
-static constexpr u32 R_386_NONE = 0;
-static constexpr u32 R_386_32 = 1;
-static constexpr u32 R_386_PC32 = 2;
-static constexpr u32 R_386_GOT32 = 3;
-static constexpr u32 R_386_PLT32 = 4;
-static constexpr u32 R_386_COPY = 5;
-static constexpr u32 R_386_GLOB_DAT = 6;
-static constexpr u32 R_386_JUMP_SLOT = 7;
-static constexpr u32 R_386_RELATIVE = 8;
-static constexpr u32 R_386_GOTOFF = 9;
-static constexpr u32 R_386_GOTPC = 10;
-static constexpr u32 R_386_32PLT = 11;
-static constexpr u32 R_386_TLS_TPOFF = 14;
-static constexpr u32 R_386_TLS_IE = 15;
-static constexpr u32 R_386_TLS_GOTIE = 16;
-static constexpr u32 R_386_TLS_LE = 17;
-static constexpr u32 R_386_TLS_GD = 18;
-static constexpr u32 R_386_TLS_LDM = 19;
-static constexpr u32 R_386_16 = 20;
-static constexpr u32 R_386_PC16 = 21;
-static constexpr u32 R_386_8 = 22;
-static constexpr u32 R_386_PC8 = 23;
-static constexpr u32 R_386_TLS_GD_32 = 24;
-static constexpr u32 R_386_TLS_GD_PUSH = 25;
-static constexpr u32 R_386_TLS_GD_CALL = 26;
-static constexpr u32 R_386_TLS_GD_POP = 27;
-static constexpr u32 R_386_TLS_LDM_32 = 28;
-static constexpr u32 R_386_TLS_LDM_PUSH = 29;
-static constexpr u32 R_386_TLS_LDM_CALL = 30;
-static constexpr u32 R_386_TLS_LDM_POP = 31;
-static constexpr u32 R_386_TLS_LDO_32 = 32;
-static constexpr u32 R_386_TLS_IE_32 = 33;
-static constexpr u32 R_386_TLS_LE_32 = 34;
-static constexpr u32 R_386_TLS_DTPMOD32 = 35;
-static constexpr u32 R_386_TLS_DTPOFF32 = 36;
-static constexpr u32 R_386_TLS_TPOFF32 = 37;
-static constexpr u32 R_386_SIZE32 = 38;
-static constexpr u32 R_386_TLS_GOTDESC = 39;
-static constexpr u32 R_386_TLS_DESC_CALL = 40;
-static constexpr u32 R_386_TLS_DESC = 41;
-static constexpr u32 R_386_IRELATIVE = 42;
-static constexpr u32 R_386_GOT32X = 43;
-
-template <>
-inline std::string rel_to_string<I386>(u32 r_type) {
-  switch (r_type) {
-  case R_386_NONE: return "R_386_NONE";
-  case R_386_32: return "R_386_32";
-  case R_386_PC32: return "R_386_PC32";
-  case R_386_GOT32: return "R_386_GOT32";
-  case R_386_PLT32: return "R_386_PLT32";
-  case R_386_COPY: return "R_386_COPY";
-  case R_386_GLOB_DAT: return "R_386_GLOB_DAT";
-  case R_386_JUMP_SLOT: return "R_386_JUMP_SLOT";
-  case R_386_RELATIVE: return "R_386_RELATIVE";
-  case R_386_GOTOFF: return "R_386_GOTOFF";
-  case R_386_GOTPC: return "R_386_GOTPC";
-  case R_386_32PLT: return "R_386_32PLT";
-  case R_386_TLS_TPOFF: return "R_386_TLS_TPOFF";
-  case R_386_TLS_IE: return "R_386_TLS_IE";
-  case R_386_TLS_GOTIE: return "R_386_TLS_GOTIE";
-  case R_386_TLS_LE: return "R_386_TLS_LE";
-  case R_386_TLS_GD: return "R_386_TLS_GD";
-  case R_386_TLS_LDM: return "R_386_TLS_LDM";
-  case R_386_16: return "R_386_16";
-  case R_386_PC16: return "R_386_PC16";
-  case R_386_8: return "R_386_8";
-  case R_386_PC8: return "R_386_PC8";
-  case R_386_TLS_GD_32: return "R_386_TLS_GD_32";
-  case R_386_TLS_GD_PUSH: return "R_386_TLS_GD_PUSH";
-  case R_386_TLS_GD_CALL: return "R_386_TLS_GD_CALL";
-  case R_386_TLS_GD_POP: return "R_386_TLS_GD_POP";
-  case R_386_TLS_LDM_32: return "R_386_TLS_LDM_32";
-  case R_386_TLS_LDM_PUSH: return "R_386_TLS_LDM_PUSH";
-  case R_386_TLS_LDM_CALL: return "R_386_TLS_LDM_CALL";
-  case R_386_TLS_LDM_POP: return "R_386_TLS_LDM_POP";
-  case R_386_TLS_LDO_32: return "R_386_TLS_LDO_32";
-  case R_386_TLS_IE_32: return "R_386_TLS_IE_32";
-  case R_386_TLS_LE_32: return "R_386_TLS_LE_32";
-  case R_386_TLS_DTPMOD32: return "R_386_TLS_DTPMOD32";
-  case R_386_TLS_DTPOFF32: return "R_386_TLS_DTPOFF32";
-  case R_386_TLS_TPOFF32: return "R_386_TLS_TPOFF32";
-  case R_386_SIZE32: return "R_386_SIZE32";
-  case R_386_TLS_GOTDESC: return "R_386_TLS_GOTDESC";
-  case R_386_TLS_DESC_CALL: return "R_386_TLS_DESC_CALL";
-  case R_386_TLS_DESC: return "R_386_TLS_DESC";
-  case R_386_IRELATIVE: return "R_386_IRELATIVE";
-  case R_386_GOT32X: return "R_386_GOT32X";
-  }
-  return "unknown (" + std::to_string(r_type) + ")";
-}
-
-static constexpr u32 R_AARCH64_NONE = 0;
-static constexpr u32 R_AARCH64_ABS64 = 0x101;
-static constexpr u32 R_AARCH64_ABS32 = 0x102;
-static constexpr u32 R_AARCH64_ABS16 = 0x103;
-static constexpr u32 R_AARCH64_PREL64 = 0x104;
-static constexpr u32 R_AARCH64_PREL32 = 0x105;
-static constexpr u32 R_AARCH64_PREL16 = 0x106;
-static constexpr u32 R_AARCH64_MOVW_UABS_G0 = 0x107;
-static constexpr u32 R_AARCH64_MOVW_UABS_G0_NC = 0x108;
-static constexpr u32 R_AARCH64_MOVW_UABS_G1 = 0x109;
-static constexpr u32 R_AARCH64_MOVW_UABS_G1_NC = 0x10a;
-static constexpr u32 R_AARCH64_MOVW_UABS_G2 = 0x10b;
-static constexpr u32 R_AARCH64_MOVW_UABS_G2_NC = 0x10c;
-static constexpr u32 R_AARCH64_MOVW_UABS_G3 = 0x10d;
-static constexpr u32 R_AARCH64_MOVW_SABS_G0 = 0x10e;
-static constexpr u32 R_AARCH64_MOVW_SABS_G1 = 0x10f;
-static constexpr u32 R_AARCH64_MOVW_SABS_G2 = 0x110;
-static constexpr u32 R_AARCH64_LD_PREL_LO19 = 0x111;
-static constexpr u32 R_AARCH64_ADR_PREL_LO21 = 0x112;
-static constexpr u32 R_AARCH64_ADR_PREL_PG_HI21 = 0x113;
-static constexpr u32 R_AARCH64_ADR_PREL_PG_HI21_NC = 0x114;
-static constexpr u32 R_AARCH64_ADD_ABS_LO12_NC = 0x115;
-static constexpr u32 R_AARCH64_LDST8_ABS_LO12_NC = 0x116;
-static constexpr u32 R_AARCH64_TSTBR14 = 0x117;
-static constexpr u32 R_AARCH64_CONDBR19 = 0x118;
-static constexpr u32 R_AARCH64_JUMP26 = 0x11a;
-static constexpr u32 R_AARCH64_CALL26 = 0x11b;
-static constexpr u32 R_AARCH64_LDST16_ABS_LO12_NC = 0x11c;
-static constexpr u32 R_AARCH64_LDST32_ABS_LO12_NC = 0x11d;
-static constexpr u32 R_AARCH64_LDST64_ABS_LO12_NC = 0x11e;
-static constexpr u32 R_AARCH64_MOVW_PREL_G0 = 0x11f;
-static constexpr u32 R_AARCH64_MOVW_PREL_G0_NC = 0x120;
-static constexpr u32 R_AARCH64_MOVW_PREL_G1 = 0x121;
-static constexpr u32 R_AARCH64_MOVW_PREL_G1_NC = 0x122;
-static constexpr u32 R_AARCH64_MOVW_PREL_G2 = 0x123;
-static constexpr u32 R_AARCH64_MOVW_PREL_G2_NC = 0x124;
-static constexpr u32 R_AARCH64_MOVW_PREL_G3 = 0x125;
-static constexpr u32 R_AARCH64_LDST128_ABS_LO12_NC = 0x12b;
-static constexpr u32 R_AARCH64_ADR_GOT_PAGE = 0x137;
-static constexpr u32 R_AARCH64_LD64_GOT_LO12_NC = 0x138;
-static constexpr u32 R_AARCH64_LD64_GOTPAGE_LO15 = 0x139;
-static constexpr u32 R_AARCH64_PLT32 = 0x13a;
-static constexpr u32 R_AARCH64_TLSGD_ADR_PREL21 = 0x200;
-static constexpr u32 R_AARCH64_TLSGD_ADR_PAGE21 = 0x201;
-static constexpr u32 R_AARCH64_TLSGD_ADD_LO12_NC = 0x202;
-static constexpr u32 R_AARCH64_TLSGD_MOVW_G1 = 0x203;
-static constexpr u32 R_AARCH64_TLSGD_MOVW_G0_NC = 0x204;
-static constexpr u32 R_AARCH64_TLSLD_ADR_PREL21 = 0x205;
-static constexpr u32 R_AARCH64_TLSLD_ADR_PAGE21 = 0x206;
-static constexpr u32 R_AARCH64_TLSLD_ADD_LO12_NC = 0x207;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_G1 = 0x208;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_G0_NC = 0x209;
-static constexpr u32 R_AARCH64_TLSLD_LD_PREL19 = 0x20a;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_DTPREL_G2 = 0x20b;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_DTPREL_G1 = 0x20c;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_DTPREL_G1_NC = 0x20d;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_DTPREL_G0 = 0x20e;
-static constexpr u32 R_AARCH64_TLSLD_MOVW_DTPREL_G0_NC = 0x20f;
-static constexpr u32 R_AARCH64_TLSLD_ADD_DTPREL_HI12 = 0x210;
-static constexpr u32 R_AARCH64_TLSLD_ADD_DTPREL_LO12 = 0x211;
-static constexpr u32 R_AARCH64_TLSLD_ADD_DTPREL_LO12_NC = 0x212;
-static constexpr u32 R_AARCH64_TLSLD_LDST8_DTPREL_LO12 = 0x213;
-static constexpr u32 R_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC = 0x214;
-static constexpr u32 R_AARCH64_TLSLD_LDST16_DTPREL_LO12 = 0x215;
-static constexpr u32 R_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC = 0x216;
-static constexpr u32 R_AARCH64_TLSLD_LDST32_DTPREL_LO12 = 0x217;
-static constexpr u32 R_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC = 0x218;
-static constexpr u32 R_AARCH64_TLSLD_LDST64_DTPREL_LO12 = 0x219;
-static constexpr u32 R_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC = 0x21a;
-static constexpr u32 R_AARCH64_TLSIE_MOVW_GOTTPREL_G1 = 0x21b;
-static constexpr u32 R_AARCH64_TLSIE_MOVW_GOTTPREL_G0_NC = 0x21c;
-static constexpr u32 R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21 = 0x21d;
-static constexpr u32 R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC = 0x21e;
-static constexpr u32 R_AARCH64_TLSIE_LD_GOTTPREL_PREL19 = 0x21f;
-static constexpr u32 R_AARCH64_TLSLE_MOVW_TPREL_G2 = 0x220;
-static constexpr u32 R_AARCH64_TLSLE_MOVW_TPREL_G1 = 0x221;
-static constexpr u32 R_AARCH64_TLSLE_MOVW_TPREL_G1_NC = 0x222;
-static constexpr u32 R_AARCH64_TLSLE_MOVW_TPREL_G0 = 0x223;
-static constexpr u32 R_AARCH64_TLSLE_MOVW_TPREL_G0_NC = 0x224;
-static constexpr u32 R_AARCH64_TLSLE_ADD_TPREL_HI12 = 0x225;
-static constexpr u32 R_AARCH64_TLSLE_ADD_TPREL_LO12 = 0x226;
-static constexpr u32 R_AARCH64_TLSLE_ADD_TPREL_LO12_NC = 0x227;
-static constexpr u32 R_AARCH64_TLSLE_LDST8_TPREL_LO12 = 0x228;
-static constexpr u32 R_AARCH64_TLSLE_LDST8_TPREL_LO12_NC = 0x229;
-static constexpr u32 R_AARCH64_TLSLE_LDST16_TPREL_LO12 = 0x22a;
-static constexpr u32 R_AARCH64_TLSLE_LDST16_TPREL_LO12_NC = 0x22b;
-static constexpr u32 R_AARCH64_TLSLE_LDST32_TPREL_LO12 = 0x22c;
-static constexpr u32 R_AARCH64_TLSLE_LDST32_TPREL_LO12_NC = 0x22d;
-static constexpr u32 R_AARCH64_TLSLE_LDST64_TPREL_LO12 = 0x22e;
-static constexpr u32 R_AARCH64_TLSLE_LDST64_TPREL_LO12_NC = 0x22f;
-static constexpr u32 R_AARCH64_TLSDESC_ADR_PAGE21 = 0x232;
-static constexpr u32 R_AARCH64_TLSDESC_LD64_LO12 = 0x233;
-static constexpr u32 R_AARCH64_TLSDESC_ADD_LO12 = 0x234;
-static constexpr u32 R_AARCH64_TLSDESC_CALL = 0x239;
-static constexpr u32 R_AARCH64_TLSLE_LDST128_TPREL_LO12_NC = 0x23b;
-static constexpr u32 R_AARCH64_COPY = 0x400;
-static constexpr u32 R_AARCH64_GLOB_DAT = 0x401;
-static constexpr u32 R_AARCH64_JUMP_SLOT = 0x402;
-static constexpr u32 R_AARCH64_RELATIVE = 0x403;
-static constexpr u32 R_AARCH64_TLS_DTPMOD64 = 0x404;
-static constexpr u32 R_AARCH64_TLS_DTPREL64 = 0x405;
-static constexpr u32 R_AARCH64_TLS_TPREL64 = 0x406;
-static constexpr u32 R_AARCH64_TLSDESC = 0x407;
-static constexpr u32 R_AARCH64_IRELATIVE = 0x408;
-
-template <>
-inline std::string rel_to_string<ARM64>(u32 r_type) {
-  switch (r_type) {
-  case R_AARCH64_NONE: return "R_AARCH64_NONE";
-  case R_AARCH64_ABS64: return "R_AARCH64_ABS64";
-  case R_AARCH64_ABS32: return "R_AARCH64_ABS32";
-  case R_AARCH64_ABS16: return "R_AARCH64_ABS16";
-  case R_AARCH64_PREL64: return "R_AARCH64_PREL64";
-  case R_AARCH64_PREL32: return "R_AARCH64_PREL32";
-  case R_AARCH64_PREL16: return "R_AARCH64_PREL16";
-  case R_AARCH64_MOVW_UABS_G0: return "R_AARCH64_MOVW_UABS_G0";
-  case R_AARCH64_MOVW_UABS_G0_NC: return "R_AARCH64_MOVW_UABS_G0_NC";
-  case R_AARCH64_MOVW_UABS_G1: return "R_AARCH64_MOVW_UABS_G1";
-  case R_AARCH64_MOVW_UABS_G1_NC: return "R_AARCH64_MOVW_UABS_G1_NC";
-  case R_AARCH64_MOVW_UABS_G2: return "R_AARCH64_MOVW_UABS_G2";
-  case R_AARCH64_MOVW_UABS_G2_NC: return "R_AARCH64_MOVW_UABS_G2_NC";
-  case R_AARCH64_MOVW_UABS_G3: return "R_AARCH64_MOVW_UABS_G3";
-  case R_AARCH64_MOVW_SABS_G0: return "R_AARCH64_MOVW_SABS_G0";
-  case R_AARCH64_MOVW_SABS_G1: return "R_AARCH64_MOVW_SABS_G1";
-  case R_AARCH64_MOVW_SABS_G2: return "R_AARCH64_MOVW_SABS_G2";
-  case R_AARCH64_LD_PREL_LO19: return "R_AARCH64_LD_PREL_LO19";
-  case R_AARCH64_ADR_PREL_LO21: return "R_AARCH64_ADR_PREL_LO21";
-  case R_AARCH64_ADR_PREL_PG_HI21: return "R_AARCH64_ADR_PREL_PG_HI21";
-  case R_AARCH64_ADR_PREL_PG_HI21_NC: return "R_AARCH64_ADR_PREL_PG_HI21_NC";
-  case R_AARCH64_ADD_ABS_LO12_NC: return "R_AARCH64_ADD_ABS_LO12_NC";
-  case R_AARCH64_LDST8_ABS_LO12_NC: return "R_AARCH64_LDST8_ABS_LO12_NC";
-  case R_AARCH64_TSTBR14: return "R_AARCH64_TSTBR14";
-  case R_AARCH64_CONDBR19: return "R_AARCH64_CONDBR19";
-  case R_AARCH64_JUMP26: return "R_AARCH64_JUMP26";
-  case R_AARCH64_CALL26: return "R_AARCH64_CALL26";
-  case R_AARCH64_LDST16_ABS_LO12_NC: return "R_AARCH64_LDST16_ABS_LO12_NC";
-  case R_AARCH64_LDST32_ABS_LO12_NC: return "R_AARCH64_LDST32_ABS_LO12_NC";
-  case R_AARCH64_LDST64_ABS_LO12_NC: return "R_AARCH64_LDST64_ABS_LO12_NC";
-  case R_AARCH64_MOVW_PREL_G0: return "R_AARCH64_MOVW_PREL_G0";
-  case R_AARCH64_MOVW_PREL_G0_NC: return "R_AARCH64_MOVW_PREL_G0_NC";
-  case R_AARCH64_MOVW_PREL_G1: return "R_AARCH64_MOVW_PREL_G1";
-  case R_AARCH64_MOVW_PREL_G1_NC: return "R_AARCH64_MOVW_PREL_G1_NC";
-  case R_AARCH64_MOVW_PREL_G2: return "R_AARCH64_MOVW_PREL_G2";
-  case R_AARCH64_MOVW_PREL_G2_NC: return "R_AARCH64_MOVW_PREL_G2_NC";
-  case R_AARCH64_MOVW_PREL_G3: return "R_AARCH64_MOVW_PREL_G3";
-  case R_AARCH64_LDST128_ABS_LO12_NC: return "R_AARCH64_LDST128_ABS_LO12_NC";
-  case R_AARCH64_ADR_GOT_PAGE: return "R_AARCH64_ADR_GOT_PAGE";
-  case R_AARCH64_LD64_GOT_LO12_NC: return "R_AARCH64_LD64_GOT_LO12_NC";
-  case R_AARCH64_LD64_GOTPAGE_LO15: return "R_AARCH64_LD64_GOTPAGE_LO15";
-  case R_AARCH64_PLT32: return "R_AARCH64_PLT32";
-  case R_AARCH64_TLSGD_ADR_PREL21: return "R_AARCH64_TLSGD_ADR_PREL21";
-  case R_AARCH64_TLSGD_ADR_PAGE21: return "R_AARCH64_TLSGD_ADR_PAGE21";
-  case R_AARCH64_TLSGD_ADD_LO12_NC: return "R_AARCH64_TLSGD_ADD_LO12_NC";
-  case R_AARCH64_TLSGD_MOVW_G1: return "R_AARCH64_TLSGD_MOVW_G1";
-  case R_AARCH64_TLSGD_MOVW_G0_NC: return "R_AARCH64_TLSGD_MOVW_G0_NC";
-  case R_AARCH64_TLSLD_ADR_PREL21: return "R_AARCH64_TLSLD_ADR_PREL21";
-  case R_AARCH64_TLSLD_ADR_PAGE21: return "R_AARCH64_TLSLD_ADR_PAGE21";
-  case R_AARCH64_TLSLD_ADD_LO12_NC: return "R_AARCH64_TLSLD_ADD_LO12_NC";
-  case R_AARCH64_TLSLD_MOVW_G1: return "R_AARCH64_TLSLD_MOVW_G1";
-  case R_AARCH64_TLSLD_MOVW_G0_NC: return "R_AARCH64_TLSLD_MOVW_G0_NC";
-  case R_AARCH64_TLSLD_LD_PREL19: return "R_AARCH64_TLSLD_LD_PREL19";
-  case R_AARCH64_TLSLD_MOVW_DTPREL_G2: return "R_AARCH64_TLSLD_MOVW_DTPREL_G2";
-  case R_AARCH64_TLSLD_MOVW_DTPREL_G1: return "R_AARCH64_TLSLD_MOVW_DTPREL_G1";
-  case R_AARCH64_TLSLD_MOVW_DTPREL_G1_NC: return "R_AARCH64_TLSLD_MOVW_DTPREL_G1_NC";
-  case R_AARCH64_TLSLD_MOVW_DTPREL_G0: return "R_AARCH64_TLSLD_MOVW_DTPREL_G0";
-  case R_AARCH64_TLSLD_MOVW_DTPREL_G0_NC: return "R_AARCH64_TLSLD_MOVW_DTPREL_G0_NC";
-  case R_AARCH64_TLSLD_ADD_DTPREL_HI12: return "R_AARCH64_TLSLD_ADD_DTPREL_HI12";
-  case R_AARCH64_TLSLD_ADD_DTPREL_LO12: return "R_AARCH64_TLSLD_ADD_DTPREL_LO12";
-  case R_AARCH64_TLSLD_ADD_DTPREL_LO12_NC: return "R_AARCH64_TLSLD_ADD_DTPREL_LO12_NC";
-  case R_AARCH64_TLSLD_LDST8_DTPREL_LO12: return "R_AARCH64_TLSLD_LDST8_DTPREL_LO12";
-  case R_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC: return "R_AARCH64_TLSLD_LDST8_DTPREL_LO12_NC";
-  case R_AARCH64_TLSLD_LDST16_DTPREL_LO12: return "R_AARCH64_TLSLD_LDST16_DTPREL_LO12";
-  case R_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC: return "R_AARCH64_TLSLD_LDST16_DTPREL_LO12_NC";
-  case R_AARCH64_TLSLD_LDST32_DTPREL_LO12: return "R_AARCH64_TLSLD_LDST32_DTPREL_LO12";
-  case R_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC: return "R_AARCH64_TLSLD_LDST32_DTPREL_LO12_NC";
-  case R_AARCH64_TLSLD_LDST64_DTPREL_LO12: return "R_AARCH64_TLSLD_LDST64_DTPREL_LO12";
-  case R_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC: return "R_AARCH64_TLSLD_LDST64_DTPREL_LO12_NC";
-  case R_AARCH64_TLSIE_MOVW_GOTTPREL_G1: return "R_AARCH64_TLSIE_MOVW_GOTTPREL_G1";
-  case R_AARCH64_TLSIE_MOVW_GOTTPREL_G0_NC: return "R_AARCH64_TLSIE_MOVW_GOTTPREL_G0_NC";
-  case R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21: return "R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21";
-  case R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC: return "R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC";
-  case R_AARCH64_TLSIE_LD_GOTTPREL_PREL19: return "R_AARCH64_TLSIE_LD_GOTTPREL_PREL19";
-  case R_AARCH64_TLSLE_MOVW_TPREL_G2: return "R_AARCH64_TLSLE_MOVW_TPREL_G2";
-  case R_AARCH64_TLSLE_MOVW_TPREL_G1: return "R_AARCH64_TLSLE_MOVW_TPREL_G1";
-  case R_AARCH64_TLSLE_MOVW_TPREL_G1_NC: return "R_AARCH64_TLSLE_MOVW_TPREL_G1_NC";
-  case R_AARCH64_TLSLE_MOVW_TPREL_G0: return "R_AARCH64_TLSLE_MOVW_TPREL_G0";
-  case R_AARCH64_TLSLE_MOVW_TPREL_G0_NC: return "R_AARCH64_TLSLE_MOVW_TPREL_G0_NC";
-  case R_AARCH64_TLSLE_ADD_TPREL_HI12: return "R_AARCH64_TLSLE_ADD_TPREL_HI12";
-  case R_AARCH64_TLSLE_ADD_TPREL_LO12: return "R_AARCH64_TLSLE_ADD_TPREL_LO12";
-  case R_AARCH64_TLSLE_ADD_TPREL_LO12_NC: return "R_AARCH64_TLSLE_ADD_TPREL_LO12_NC";
-  case R_AARCH64_TLSLE_LDST8_TPREL_LO12: return "R_AARCH64_TLSLE_LDST8_TPREL_LO12";
-  case R_AARCH64_TLSLE_LDST8_TPREL_LO12_NC: return "R_AARCH64_TLSLE_LDST8_TPREL_LO12_NC";
-  case R_AARCH64_TLSLE_LDST16_TPREL_LO12: return "R_AARCH64_TLSLE_LDST16_TPREL_LO12";
-  case R_AARCH64_TLSLE_LDST16_TPREL_LO12_NC: return "R_AARCH64_TLSLE_LDST16_TPREL_LO12_NC";
-  case R_AARCH64_TLSLE_LDST32_TPREL_LO12: return "R_AARCH64_TLSLE_LDST32_TPREL_LO12";
-  case R_AARCH64_TLSLE_LDST32_TPREL_LO12_NC: return "R_AARCH64_TLSLE_LDST32_TPREL_LO12_NC";
-  case R_AARCH64_TLSLE_LDST64_TPREL_LO12: return "R_AARCH64_TLSLE_LDST64_TPREL_LO12";
-  case R_AARCH64_TLSLE_LDST64_TPREL_LO12_NC: return "R_AARCH64_TLSLE_LDST64_TPREL_LO12_NC";
-  case R_AARCH64_TLSDESC_ADR_PAGE21: return "R_AARCH64_TLSDESC_ADR_PAGE21";
-  case R_AARCH64_TLSDESC_LD64_LO12: return "R_AARCH64_TLSDESC_LD64_LO12";
-  case R_AARCH64_TLSDESC_ADD_LO12: return "R_AARCH64_TLSDESC_ADD_LO12";
-  case R_AARCH64_TLSDESC_CALL: return "R_AARCH64_TLSDESC_CALL";
-  case R_AARCH64_TLSLE_LDST128_TPREL_LO12_NC: return "R_AARCH64_TLSLE_LDST128_TPREL_LO12_NC";
-  case R_AARCH64_COPY: return "R_AARCH64_COPY";
-  case R_AARCH64_GLOB_DAT: return "R_AARCH64_GLOB_DAT";
-  case R_AARCH64_JUMP_SLOT: return "R_AARCH64_JUMP_SLOT";
-  case R_AARCH64_RELATIVE: return "R_AARCH64_RELATIVE";
-  case R_AARCH64_TLS_DTPMOD64: return "R_AARCH64_TLS_DTPMOD64";
-  case R_AARCH64_TLS_DTPREL64: return "R_AARCH64_TLS_DTPREL64";
-  case R_AARCH64_TLS_TPREL64: return "R_AARCH64_TLS_TPREL64";
-  case R_AARCH64_TLSDESC: return "R_AARCH64_TLSDESC";
-  case R_AARCH64_IRELATIVE: return "R_AARCH64_IRELATIVE";
-  }
-  return "unknown (" + std::to_string(r_type) + ")";
-}
-
-static constexpr u32 DW_EH_PE_absptr = 0;
-static constexpr u32 DW_EH_PE_omit = 0xff;
-static constexpr u32 DW_EH_PE_uleb128 = 0x01;
-static constexpr u32 DW_EH_PE_udata2 = 0x02;
-static constexpr u32 DW_EH_PE_udata4 = 0x03;
-static constexpr u32 DW_EH_PE_udata8 = 0x04;
-static constexpr u32 DW_EH_PE_signed = 0x08;
-static constexpr u32 DW_EH_PE_sleb128 = 0x09;
-static constexpr u32 DW_EH_PE_sdata2 = 0x0a;
-static constexpr u32 DW_EH_PE_sdata4 = 0x0b;
-static constexpr u32 DW_EH_PE_sdata8 = 0x0c;
-static constexpr u32 DW_EH_PE_pcrel = 0x10;
-static constexpr u32 DW_EH_PE_textrel = 0x20;
-static constexpr u32 DW_EH_PE_datarel = 0x30;
-static constexpr u32 DW_EH_PE_funcrel = 0x40;
-static constexpr u32 DW_EH_PE_aligned = 0x50;
-
-struct Elf64Sym {
-  bool is_defined() const { return !is_undef(); }
-  bool is_undef() const { return st_shndx == SHN_UNDEF; }
-  bool is_abs() const { return st_shndx == SHN_ABS; }
-  bool is_common() const { return st_shndx == SHN_COMMON; }
-  bool is_weak() const { return st_bind == STB_WEAK; }
-
-  bool is_undef_strong() const {
-    return st_shndx == SHN_UNDEF && st_bind != STB_WEAK;
-  }
-
-  bool is_undef_weak() const {
-    return st_shndx == SHN_UNDEF && st_bind == STB_WEAK;
-  }
-
-  u32 st_name;
-  u8 st_type : 4;
-  u8 st_bind : 4;
-  u8 st_visibility : 2;
-  u16 st_shndx;
-  u64 st_value;
-  u64 st_size;
-};
-
-struct Elf32Sym {
-  bool is_defined() const { return !is_undef(); }
-  bool is_undef() const { return st_shndx == SHN_UNDEF; }
-  bool is_abs() const { return st_shndx == SHN_ABS; }
-  bool is_common() const { return st_shndx == SHN_COMMON; }
-  bool is_weak() const { return st_bind == STB_WEAK; }
-
-  bool is_undef_strong() const {
-    return st_shndx == SHN_UNDEF && st_bind != STB_WEAK;
-  }
-
-  bool is_undef_weak() const {
-    return st_shndx == SHN_UNDEF && st_bind == STB_WEAK;
-  }
-
-  u32 st_name;
-  u32 st_value;
-  u32 st_size;
-  u8 st_type : 4;
-  u8 st_bind : 4;
-  u8 st_visibility : 2;
-  u16 st_shndx;
-};
-
-struct Elf64Shdr {
-  u32 sh_name;
-  u32 sh_type;
-  u64 sh_flags;
-  u64 sh_addr;
-  u64 sh_offset;
-  u64 sh_size;
-  u32 sh_link;
-  u32 sh_info;
-  u64 sh_addralign;
-  u64 sh_entsize;
-};
-
-struct Elf32Shdr {
-  u32 sh_name;
-  u32 sh_type;
-  u32 sh_flags;
-  u32 sh_addr;
-  u32 sh_offset;
-  u32 sh_size;
-  u32 sh_link;
-  u32 sh_info;
-  u32 sh_addralign;
-  u32 sh_entsize;
-};
-
-struct Elf64Ehdr {
-  u8 e_ident[16];
-  u16 e_type;
-  u16 e_machine;
-  u32 e_version;
-  u64 e_entry;
-  u64 e_phoff;
-  u64 e_shoff;
-  u32 e_flags;
-  u16 e_ehsize;
-  u16 e_phentsize;
-  u16 e_phnum;
-  u16 e_shentsize;
-  u16 e_shnum;
-  u16 e_shstrndx;
-};
-
-struct Elf32Ehdr {
-  u8 e_ident[16];
-  u16 e_type;
-  u16 e_machine;
-  u32 e_version;
-  u32 e_entry;
-  u32 e_phoff;
-  u32 e_shoff;
-  u32 e_flags;
-  u16 e_ehsize;
-  u16 e_phentsize;
-  u16 e_phnum;
-  u16 e_shentsize;
-  u16 e_shnum;
-  u16 e_shstrndx;
-};
-
-struct Elf64Phdr {
-  u32 p_type;
-  u32 p_flags;
-  u64 p_offset;
-  u64 p_vaddr;
-  u64 p_paddr;
-  u64 p_filesz;
-  u64 p_memsz;
-  u64 p_align;
-};
-
-struct Elf32Phdr {
-  u32 p_type;
-  u32 p_offset;
-  u32 p_vaddr;
-  u32 p_paddr;
-  u32 p_filesz;
-  u32 p_memsz;
-  u32 p_flags;
-  u32 p_align;
-};
-
-struct Elf64Rel {
-  u64 r_offset;
-  u32 r_type;
-  u32 r_sym;
-};
-
-struct Elf32Rel {
-  u32 r_offset;
-  u32 r_type : 8;
-  u32 r_sym : 24;
-};
-
-struct Elf64Rela {
-  u64 r_offset;
-  u32 r_type;
-  u32 r_sym;
-  i64 r_addend;
-};
-
-struct Elf32Rela {
-  u32 r_offset;
-  u32 r_type : 8;
-  u32 r_sym : 24;
-  i32 r_addend;
-};
-
-struct Elf64Dyn {
-  u64 d_tag;
-  u64 d_val;
-};
-
-struct Elf32Dyn {
-  u32 d_tag;
-  u32 d_val;
-};
-
-struct ElfVerneed {
-  u16 vn_version;
-  u16 vn_cnt;
-  u32 vn_file;
-  u32 vn_aux;
-  u32 vn_next;
-};
-
-struct ElfVernaux {
-  u32 vna_hash;
-  u16 vna_flags;
-  u16 vna_other;
-  u32 vna_name;
-  u32 vna_next;
-};
-
-struct ElfVerdef {
-  u16 vd_version;
-  u16 vd_flags;
-  u16 vd_ndx;
-  u16 vd_cnt;
-  u32 vd_hash;
-  u32 vd_aux;
-  u32 vd_next;
-};
-
-struct ElfVerdaux {
-  u32 vda_name;
-  u32 vda_next;
-};
-
-struct Elf64Chdr {
-  u32 ch_type;
-  u32 ch_reserved;
-  u64 ch_size;
-  u64 ch_addralign;
-};
-
-struct Elf32Chdr {
-  u32 ch_type;
-  u32 ch_size;
-  u32 ch_addralign;
-};
-
-struct ElfNhdr {
-  u32 n_namesz;
-  u32 n_descsz;
-  u32 n_type;
-};
-
-struct X86_64 {
-  using WordTy = u64;
-
-  static constexpr u32 R_NONE = R_X86_64_NONE;
-  static constexpr u32 R_COPY = R_X86_64_COPY;
-  static constexpr u32 R_GLOB_DAT = R_X86_64_GLOB_DAT;
-  static constexpr u32 R_JUMP_SLOT = R_X86_64_JUMP_SLOT;
-  static constexpr u32 R_RELATIVE = R_X86_64_RELATIVE;
-  static constexpr u32 R_IRELATIVE = R_X86_64_IRELATIVE;
-  static constexpr u32 R_DTPOFF = R_X86_64_DTPOFF64;
-  static constexpr u32 R_TPOFF = R_X86_64_TPOFF64;
-  static constexpr u32 R_DTPMOD = R_X86_64_DTPMOD64;
-  static constexpr u32 R_TLSDESC = R_X86_64_TLSDESC;
-
-  static constexpr u32 word_size = 8;
-  static constexpr u32 page_size = 4096;
-  static constexpr u32 e_machine = EM_X86_64;
-  static constexpr u32 pltgot_size = 8;
-  static constexpr bool is_rel = false;
-  static constexpr bool is_le = true;
-};
-
-template <> struct ElfSym<X86_64> : public Elf64Sym {};
-template <> struct ElfShdr<X86_64> : public Elf64Shdr {};
-template <> struct ElfEhdr<X86_64> : public Elf64Ehdr {};
-template <> struct ElfPhdr<X86_64> : public Elf64Phdr {};
-template <> struct ElfRel<X86_64> : public Elf64Rela {};
-template <> struct ElfDyn<X86_64> : public Elf64Dyn {};
-template <> struct ElfChdr<X86_64> : public Elf64Chdr {};
-
-struct I386 {
-  using WordTy = u32;
-
-  static constexpr u32 R_NONE = R_386_NONE;
-  static constexpr u32 R_COPY = R_386_COPY;
-  static constexpr u32 R_GLOB_DAT = R_386_GLOB_DAT;
-  static constexpr u32 R_JUMP_SLOT = R_386_JUMP_SLOT;
-  static constexpr u32 R_RELATIVE = R_386_RELATIVE;
-  static constexpr u32 R_IRELATIVE = R_386_IRELATIVE;
-  static constexpr u32 R_DTPOFF = R_386_TLS_DTPOFF32;
-  static constexpr u32 R_TPOFF = R_386_TLS_TPOFF;
-  static constexpr u32 R_DTPMOD = R_386_TLS_DTPMOD32;
-  static constexpr u32 R_TLSDESC = R_386_TLS_DESC;
-
-  static constexpr u32 word_size = 4;
-  static constexpr u32 page_size = 4096;
-  static constexpr u32 e_machine = EM_386;
-  static constexpr u32 pltgot_size = 8;
-  static constexpr bool is_rel = true;
-  static constexpr bool is_le = true;
-};
-
-template <> struct ElfSym<I386> : public Elf32Sym {};
-template <> struct ElfShdr<I386> : public Elf32Shdr {};
-template <> struct ElfEhdr<I386> : public Elf32Ehdr {};
-template <> struct ElfPhdr<I386> : public Elf32Phdr {};
-template <> struct ElfRel<I386> : public Elf32Rel {};
-template <> struct ElfDyn<I386> : public Elf32Dyn {};
-template <> struct ElfChdr<I386> : public Elf32Chdr {};
-
-struct ARM64 {
-  using WordTy = u64;
-
-  static constexpr u32 R_NONE = R_AARCH64_NONE;
-  static constexpr u32 R_COPY = R_AARCH64_COPY;
-  static constexpr u32 R_GLOB_DAT = R_AARCH64_GLOB_DAT;
-  static constexpr u32 R_JUMP_SLOT = R_AARCH64_JUMP_SLOT;
-  static constexpr u32 R_RELATIVE = R_AARCH64_RELATIVE;
-  static constexpr u32 R_IRELATIVE = R_AARCH64_IRELATIVE;
-  static constexpr u32 R_DTPOFF = R_AARCH64_TLS_DTPREL64;
-  static constexpr u32 R_TPOFF = R_AARCH64_TLS_TPREL64;
-  static constexpr u32 R_DTPMOD = R_AARCH64_TLS_DTPMOD64;
-  static constexpr u32 R_TLSDESC = R_AARCH64_TLSDESC;
-
-  static constexpr u32 word_size = 8;
-  static constexpr u32 page_size = 65536;
-  static constexpr u32 e_machine = EM_AARCH64;
-  static constexpr u32 pltgot_size = 16;
-  static constexpr bool is_rel = false;
-  static constexpr bool is_le = true;
-};
-
-template <> struct ElfSym<ARM64> : public Elf64Sym {};
-template <> struct ElfShdr<ARM64> : public Elf64Shdr {};
-template <> struct ElfEhdr<ARM64> : public Elf64Ehdr {};
-template <> struct ElfPhdr<ARM64> : public Elf64Phdr {};
-template <> struct ElfRel<ARM64> : public Elf64Rela {};
-template <> struct ElfDyn<ARM64> : public Elf64Dyn {};
-template <> struct ElfChdr<ARM64> : public Elf64Chdr {};
-
-} // namespace mold::elf

elf/input-sections.cc

@@ -1,162 +0,0 @@
-#include "mold.h"
-
-#include <limits>
-
-namespace mold::elf {
-
-template <typename E>
-bool CieRecord<E>::equals(const CieRecord<E> &other) const {
-  if (get_contents() != other.get_contents())
-    return false;
-
-  std::span<ElfRel<E>> x = get_rels();
-  std::span<ElfRel<E>> y = other.get_rels();
-  if (x.size() != y.size())
-    return false;
-
-  for (i64 i = 0; i < x.size(); i++) {
-    if (x[i].r_offset - input_offset != y[i].r_offset - other.input_offset ||
-        x[i].r_type != y[i].r_type ||
-        file.symbols[x[i].r_sym] != other.file.symbols[y[i].r_sym] ||
-        input_section.get_addend(x[i]) != other.input_section.get_addend(y[i]))
-      return false;
-  }
-  return true;
-}
-
-template <typename E>
-InputSection<E>::InputSection(Context<E> &ctx, ObjectFile<E> &file,
-                              const ElfShdr<E> &shdr, std::string_view name,
-                              std::string_view contents, i64 section_idx)
-  : file(file), shdr(shdr), contents(contents), nameptr(name.data()),
-    namelen(name.size()), section_idx(section_idx) {
-  output_section =
-    OutputSection<E>::get_instance(ctx, name, shdr.sh_type, shdr.sh_flags);
-}
-
-template <typename E>
-void InputSection<E>::write_to(Context<E> &ctx, u8 *buf) {
-  if (shdr.sh_type == SHT_NOBITS || shdr.sh_size == 0)
-    return;
-
-  // Copy data
-  memcpy(buf, contents.data(), contents.size());
-
-  // Apply relocations
-  if (shdr.sh_flags & SHF_ALLOC)
-    apply_reloc_alloc(ctx, buf);
-  else
-    apply_reloc_nonalloc(ctx, buf);
-}
-
-template <typename E>
-static i64 get_output_type(Context<E> &ctx) {
-  if (ctx.arg.shared)
-    return 0;
-  if (ctx.arg.pie)
-    return 1;
-  return 2;
-}
-
-template <typename E>
-static i64 get_sym_type(Symbol<E> &sym) {
-  if (sym.is_absolute())
-    return 0;
-  if (!sym.is_imported)
-    return 1;
-  if (sym.get_type() != STT_FUNC)
-    return 2;
-  return 3;
-}
-
-template <typename E>
-void InputSection<E>::dispatch(Context<E> &ctx, Action table[3][4], i64 i,
-                               const ElfRel<E> &rel, Symbol<E> &sym) {
-  Action action = table[get_output_type(ctx)][get_sym_type(sym)];
-  bool is_code = (shdr.sh_flags & SHF_EXECINSTR);
-  bool is_writable = (shdr.sh_flags & SHF_WRITE);
-
-  auto error = [&]() {
-    Error(ctx) << *this << ": " << rel << " relocation against symbol `"
-               << sym << "' can not be used; recompile with -fPIC";
-  };
-
-  switch (action) {
-  case NONE:
-    return;
-  case ERROR:
-    error();
-    return;
-  case COPYREL:
-    if (!ctx.arg.z_copyreloc) {
-      error();
-      return;
-    }
-
-    if (sym.esym().st_visibility == STV_PROTECTED) {
-      Error(ctx) << *this
-                 << ": cannot make copy relocation for protected symbol '"
-                 << sym << "', defined in " << *sym.file
-                 << "; recompile with -fPIC";
-      return;
-    }
-
-    sym.flags |= NEEDS_COPYREL;
-    return;
-  case PLT:
-    sym.flags |= NEEDS_PLT;
-    return;
-  case DYNREL:
-    if (!is_writable) {
-      if (!is_code || ctx.arg.z_text) {
-        error();
-        return;
-      }
-      ctx.has_textrel = true;
-    }
-
-    assert(sym.is_imported);
-    needs_dynrel[i] = true;
-    file.num_dynrel++;
-    return;
-  case BASEREL:
-    if (!is_writable) {
-      if (!is_code || ctx.arg.z_text) {
-        error();
-        return;
-      }
-      ctx.has_textrel = true;
-    }
-
-    needs_baserel[i] = true;
-    if (!is_relr_reloc(ctx, rel))
-      file.num_dynrel++;
-    return;
-  default:
-    unreachable();
-  }
-}
-
-template <typename E>
-void InputSection<E>::report_undef(Context<E> &ctx, Symbol<E> &sym) {
-  switch (ctx.arg.unresolved_symbols) {
-  case UNRESOLVED_ERROR:
-    Error(ctx) << "undefined symbol: " << file << ": " << sym;
-    break;
-  case UNRESOLVED_WARN:
-    Warn(ctx) << "undefined symbol: " << file << ": " << sym;
-    break;
-  case UNRESOLVED_IGNORE:
-    break;
-  }
-}
-
-#define INSTANTIATE(E)                          \
-  template struct CieRecord<E>;                 \
-  template class InputSection<E>;
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/linker-script.cc

@@ -1,412 +0,0 @@
-// On Linux, /usr/lib/x86_64-linux-gnu/libc.so is not actually
-// a shared object file but an ASCII text file containing a linker
-// script to include a "real" libc.so file. Therefore, we need to
-// support a (very limited) subset of the linker script language.
-
-#include "mold.h"
-
-#include <cctype>
-#include <iomanip>
-
-namespace mold::elf {
-
-template <typename E>
-static thread_local MappedFile<Context<E>> *current_file;
-
-template <typename E>
-void read_version_script(Context<E> &ctx, std::span<std::string_view> &tok);
-
-static std::string_view get_line(std::string_view input, const char *pos) {
-  assert(input.data() <= pos);
-  assert(pos < input.data() + input.size());
-
-  i64 start = input.rfind('\n', pos - input.data());
-  if (start == input.npos)
-    start = 0;
-  else
-    start++;
-
-  i64 end = input.find('\n', pos - input.data());
-  if (end == input.npos)
-    end = input.size();
-
-  return input.substr(start, end - start);
-}
-
-template <typename E>
-class SyntaxError {
-public:
-  SyntaxError(Context<E> &ctx, std::string_view errpos) : out(ctx) {
-    std::string_view contents = current_file<E>->get_contents();
-    std::string_view line = get_line(contents, errpos.data());
-
-    i64 lineno = 1;
-    for (i64 i = 0; contents.data() + i < line.data(); i++)
-      if (contents[i] == '\n')
-        lineno++;
-
-    i64 column = errpos.data() - line.data();
-
-    std::stringstream ss;
-    ss << current_file<E>->name << ":" << lineno << ": ";
-    i64 indent = (i64)ss.tellp() + strlen("mold: ");
-    ss << line << "\n" << std::setw(indent + column) << " " << "^ ";
-    out << ss.str();
-  }
-
-  template <class T> SyntaxError &operator<<(T &&val) {
-    out << std::forward<T>(val);
-    return *this;
-  }
-
-  [[noreturn]] ~SyntaxError() = default;
-
-  Fatal<Context<E>> out;
-};
-
-template <typename E>
-static std::vector<std::string_view>
-tokenize(Context<E> &ctx, std::string_view input) {
-  std::vector<std::string_view> vec;
-  while (!input.empty()) {
-    if (isspace(input[0])) {
-      input = input.substr(1);
-      continue;
-    }
-
-    if (input.starts_with("/*")) {
-      i64 pos = input.find("*/", 2);
-      if (pos == std::string_view::npos)
-        SyntaxError(ctx, input) << "unclosed comment";
-      input = input.substr(pos + 2);
-      continue;
-    }
-
-    if (input[0] == '#') {
-      i64 pos = input.find("\n", 1);
-      if (pos == std::string_view::npos)
-        break;
-      input = input.substr(pos + 1);
-      continue;
-    }
-
-    if (input[0] == '"') {
-      i64 pos = input.find('"', 1);
-      if (pos == std::string_view::npos)
-        SyntaxError(ctx, input) << "unclosed string literal";
-      vec.push_back(input.substr(0, pos + 1));
-      input = input.substr(pos + 1);
-      continue;
-    }
-
-    i64 pos = input.find_first_not_of(
-      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
-      "0123456789_.$/\\~=+[]*?-!^:");
-
-    if (pos == 0)
-      pos = 1;
-    else if (pos == input.npos)
-      pos = input.size();
-
-    vec.push_back(input.substr(0, pos));
-    input = input.substr(pos);
-  }
-  return vec;
-}
-
-template <typename E>
-static std::span<std::string_view>
-skip(Context<E> &ctx, std::span<std::string_view> tok, std::string_view str) {
-  if (tok.empty())
-    Fatal(ctx) << current_file<E>->name << ": expected '" << str
-               << "', but got EOF";
-  if (tok[0] != str)
-    SyntaxError(ctx, tok[0]) << "expected '" << str << "'";
-  return tok.subspan(1);
-}
-
-static std::string_view unquote(std::string_view s) {
-  if (s.size() > 0 && s[0] == '"') {
-    assert(s[s.size() - 1] == '"');
-    return s.substr(1, s.size() - 2);
-  }
-  return s;
-}
-
-template <typename E>
-static std::span<std::string_view>
-read_output_format(Context<E> &ctx, std::span<std::string_view> tok) {
-  tok = skip(ctx, tok, "(");
-  while (!tok.empty() && tok[0] != ")")
-    tok = tok.subspan(1);
-  if (tok.empty())
-    Fatal(ctx) << current_file<E>->name << ": expected ')', but got EOF";
-  return tok.subspan(1);
-}
-
-template <typename E>
-static bool is_in_sysroot(Context<E> &ctx, std::string path) {
-  std::string rel =
-    to_abs_path(path).lexically_relative(to_abs_path(ctx.arg.sysroot));
-  return rel != "." && !rel.starts_with("../");
-}
-
-template <typename E>
-static MappedFile<Context<E>> *resolve_path(Context<E> &ctx, std::string_view tok) {
-  std::string str(unquote(tok));
-
-  // GNU ld prepends the sysroot if a pathname starts with '/' and the
-  // script being processed is in the sysroot. We do the same.
-  if (str.starts_with('/') && is_in_sysroot(ctx, current_file<E>->name))
-    return MappedFile<Context<E>>::must_open(ctx, ctx.arg.sysroot + str);
-
-  if (str.starts_with('=')) {
-    std::string path;
-    if (ctx.arg.sysroot.empty())
-      path = str.substr(1);
-    else
-      path = ctx.arg.sysroot + str.substr(1);
-    return MappedFile<Context<E>>::must_open(ctx, path);
-  }
-
-  if (str.starts_with("-l"))
-    return find_library(ctx, str.substr(2));
-
-  if (MappedFile<Context<E>> *mf = open_library(ctx, str))
-    return mf;
-
-  for (std::string_view dir : ctx.arg.library_paths) {
-    std::string path = std::string(dir) + "/" + str;
-    if (MappedFile<Context<E>> *mf = open_library(ctx, path))
-      return mf;
-  }
-
-  SyntaxError(ctx, tok) << "library not found: " << str;
-}
-
-template <typename E>
-static std::span<std::string_view>
-read_group(Context<E> &ctx, std::span<std::string_view> tok) {
-  tok = skip(ctx, tok, "(");
-
-  while (!tok.empty() && tok[0] != ")") {
-    if (tok[0] == "AS_NEEDED") {
-      bool orig = ctx.as_needed;
-      ctx.as_needed = true;
-      tok = read_group(ctx, tok.subspan(1));
-      ctx.as_needed = orig;
-      continue;
-    }
-
-    MappedFile<Context<E>> *mf = resolve_path(ctx, tok[0]);
-    read_file(ctx, mf);
-    tok = tok.subspan(1);
-  }
-
-  if (tok.empty())
-    Fatal(ctx) << current_file<E>->name << ": expected ')', but got EOF";
-  return tok.subspan(1);
-}
-
-template <typename E>
-void parse_linker_script(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  current_file<E> = mf;
-
-  std::vector<std::string_view> vec = tokenize(ctx, mf->get_contents());
-  std::span<std::string_view> tok = vec;
-
-  while (!tok.empty()) {
-    if (tok[0] == "OUTPUT_FORMAT") {
-      tok = read_output_format(ctx, tok.subspan(1));
-    } else if (tok[0] == "INPUT" || tok[0] == "GROUP") {
-      tok = read_group(ctx, tok.subspan(1));
-    } else if (tok[0] == "VERSION") {
-      tok = tok.subspan(1);
-      tok = skip(ctx, tok, "{");
-      read_version_script(ctx, tok);
-      tok = skip(ctx, tok, "}");
-    } else if (tok[0] == ";") {
-      tok = tok.subspan(1);
-    } else {
-      SyntaxError(ctx, tok[0]) << "unknown linker script token";
-    }
-  }
-}
-
-template <typename E>
-i64 get_script_output_type(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  current_file<E> = mf;
-
-  std::vector<std::string_view> vec = tokenize(ctx, mf->get_contents());
-  std::span<std::string_view> tok = vec;
-
-  if (tok.size() >= 3 && tok[0] == "OUTPUT_FORMAT" && tok[1] == "(") {
-    if (tok[2] == "elf64-x86-64")
-      return EM_X86_64;
-    if (tok[2] == "elf32-i386")
-      return EM_386;
-  }
-  return -1;
-}
-
-static bool read_label(std::span<std::string_view> &tok,
-                       std::string label) {
-  if (tok.size() >= 1 && tok[0] == label + ":") {
-    tok = tok.subspan(1);
-    return true;
-  }
-
-  if (tok.size() >= 2 && tok[0] == label && tok[1] == ":") {
-    tok = tok.subspan(2);
-    return true;
-  }
-  return false;
-}
-
-template <typename E>
-static void
-read_version_script_commands(Context<E> &ctx, std::span<std::string_view> &tok,
-                             u16 ver_idx, bool is_cpp) {
-  bool is_global = true;
-
-  while (!tok.empty() && tok[0] != "}") {
-    if (read_label(tok, "global")) {
-      is_global = true;
-      continue;
-    }
-
-    if (read_label(tok, "local")) {
-      is_global = false;
-      continue;
-    }
-
-    if (tok[0] == "extern") {
-      tok = tok.subspan(1);
-
-      if (!tok.empty() && tok[0] == "\"C\"") {
-        tok = tok.subspan(1);
-        tok = skip(ctx, tok, "{");
-        read_version_script_commands( ctx, tok, ver_idx, false);
-      } else {
-        tok = skip(ctx, tok, "\"C++\"");
-        tok = skip(ctx, tok, "{");
-        read_version_script_commands(ctx, tok, ver_idx, true);
-      }
-
-      tok = skip(ctx, tok, "}");
-      tok = skip(ctx, tok, ";");
-      continue;
-    }
-
-    if (tok[0] == "*") {
-      ctx.default_version = (is_global ? ver_idx : VER_NDX_LOCAL);
-    } else if (is_global) {
-      ctx.version_patterns.push_back({unquote(tok[0]), ver_idx, is_cpp});
-    } else {
-      ctx.version_patterns.push_back({unquote(tok[0]), VER_NDX_LOCAL, is_cpp});
-    }
-
-    tok = tok.subspan(1);
-
-    if (!tok.empty() && tok[0] == "}")
-      return;
-    tok = skip(ctx, tok, ";");
-  }
-}
-
-template <typename E>
-void read_version_script(Context<E> &ctx, std::span<std::string_view> &tok) {
-  u16 next_ver = VER_NDX_LAST_RESERVED + ctx.arg.version_definitions.size() + 1;
-
-  while (!tok.empty() && tok[0] != "}") {
-    u16 ver_idx;
-    if (tok[0] == "{") {
-      ver_idx = VER_NDX_GLOBAL;
-    } else {
-      ver_idx = next_ver++;
-      ctx.arg.version_definitions.push_back(std::string(tok[0]));
-      tok = tok.subspan(1);
-    }
-
-    tok = skip(ctx, tok, "{");
-    read_version_script_commands(ctx, tok, ver_idx, false);
-    tok = skip(ctx, tok, "}");
-    if (!tok.empty() && tok[0] != ";")
-      tok = tok.subspan(1);
-    tok = skip(ctx, tok, ";");
-  }
-}
-
-template <typename E>
-void parse_version_script(Context<E> &ctx, std::string path) {
-  current_file<E> = MappedFile<Context<E>>::must_open(ctx, path);
-  std::vector<std::string_view> vec =
-    tokenize(ctx, current_file<E>->get_contents());
-  std::span<std::string_view> tok = vec;
-  read_version_script(ctx, tok);
-  if (!tok.empty())
-    SyntaxError(ctx, tok[0]) << "trailing garbage token";
-}
-
-template <typename E>
-void read_dynamic_list_commands(Context<E> &ctx, std::span<std::string_view> &tok,
-                                bool is_cpp) {
-  while (!tok.empty() && tok[0] != "}") {
-    if (tok[0] == "extern") {
-      tok = tok.subspan(1);
-
-      if (!tok.empty() && tok[0] == "\"C\"") {
-        tok = tok.subspan(1);
-        tok = skip(ctx, tok, "{");
-        read_dynamic_list_commands(ctx, tok, false);
-      } else {
-        tok = skip(ctx, tok, "\"C++\"");
-        tok = skip(ctx, tok, "{");
-        read_dynamic_list_commands(ctx, tok, true);
-      }
-
-      tok = skip(ctx, tok, "}");
-      tok = skip(ctx, tok, ";");
-      continue;
-    }
-
-    if (tok[0] == "*")
-      ctx.default_version = VER_NDX_GLOBAL;
-    else
-      ctx.version_patterns.push_back({unquote(tok[0]), VER_NDX_GLOBAL, is_cpp});
-
-    tok = skip(ctx, tok.subspan(1), ";");
-  }
-}
-
-template <typename E>
-void parse_dynamic_list(Context<E> &ctx, std::string path) {
-  current_file<E> = MappedFile<Context<E>>::must_open(ctx, path);
-  std::vector<std::string_view> vec =
-    tokenize(ctx, current_file<E>->get_contents());
-
-  std::span<std::string_view> tok = vec;
-  tok = skip(ctx, tok, "{");
-  read_dynamic_list_commands(ctx, tok, false);
-  tok = skip(ctx, tok, "}");
-  tok = skip(ctx, tok, ";");
-
-  if (!tok.empty())
-    SyntaxError(ctx, tok[0]) << "trailing garbage token";
-}
-
-#define INSTANTIATE(E)                                                  \
-  template                                                              \
-  void parse_linker_script(Context<E> &ctx, MappedFile<Context<E>> *mf); \
-  template                                                              \
-  i64 get_script_output_type(Context<E> &ctx, MappedFile<Context<E>> *mf); \
-  template                                                              \
-  void parse_version_script(Context<E> &ctx, std::string path);         \
-  template                                                              \
-  void parse_dynamic_list(Context<E> &ctx, std::string path)
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/main.cc

@@ -1,731 +0,0 @@
-#include "mold.h"
-#include "../archive-file.h"
-#include "../cmdline.h"
-
-#include <cstring>
-#include <functional>
-#include <iomanip>
-#include <map>
-#include <signal.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <tbb/global_control.h>
-#include <tbb/parallel_for_each.h>
-#include <unistd.h>
-#include <unordered_set>
-
-namespace mold::elf {
-
-template <typename E>
-static ObjectFile<E> *new_object_file(Context<E> &ctx, MappedFile<Context<E>> *mf,
-                                      std::string archive_name) {
-  if (i64 type = ((ElfEhdr<E> *)mf->data)->e_machine; type != E::e_machine)
-    Fatal(ctx) << mf->name << ": incompatible file type: " << type;
-
-  static Counter count("parsed_objs");
-  count++;
-
-  bool in_lib = ctx.in_lib || (!archive_name.empty() && !ctx.whole_archive);
-  ObjectFile<E> *file = ObjectFile<E>::create(ctx, mf, archive_name, in_lib);
-  file->priority = ctx.file_priority++;
-  ctx.tg.run([file, &ctx]() { file->parse(ctx); });
-  if (ctx.arg.trace)
-    SyncOut(ctx) << "trace: " << *file;
-  return file;
-}
-
-template <typename E>
-static SharedFile<E> *
-new_shared_file(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  if (i64 type = ((ElfEhdr<E> *)mf->data)->e_machine; type != E::e_machine)
-    Fatal(ctx) << mf->name << ": incompatible file type: " << type;
-
-  SharedFile<E> *file = SharedFile<E>::create(ctx, mf);
-  file->priority = ctx.file_priority++;
-  ctx.tg.run([file, &ctx]() { file->parse(ctx); });
-  if (ctx.arg.trace)
-    SyncOut(ctx) << "trace: " << *file;
-  return file;
-}
-
-template <typename E>
-void read_file(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  if (ctx.visited.contains(mf->name))
-    return;
-
-  FileType type = get_file_type(mf);
-  switch (type) {
-  case FileType::ELF_OBJ:
-    ctx.objs.push_back(new_object_file(ctx, mf, ""));
-    return;
-  case FileType::ELF_DSO:
-    ctx.dsos.push_back(new_shared_file(ctx, mf));
-    ctx.visited.insert(mf->name);
-    return;
-  case FileType::AR:
-  case FileType::THIN_AR:
-    for (MappedFile<Context<E>> *child : read_archive_members(ctx, mf))
-      if (get_file_type(child) == FileType::ELF_OBJ)
-        ctx.objs.push_back(new_object_file(ctx, child, mf->name));
-    ctx.visited.insert(mf->name);
-    return;
-  case FileType::TEXT:
-    parse_linker_script(ctx, mf);
-    return;
-  case FileType::LLVM_BITCODE:
-    Warn(ctx) << mf->name << ": looks like this is an LLVM bitcode, "
-              << "but mold does not support LTO";
-    ctx.llvm_lto = true;
-    return;
-  default:
-    Fatal(ctx) << mf->name << ": unknown file type: " << type;
-  }
-}
-
-// Read the beginning of a given file and returns its machine type
-// (e.g. EM_X86_64 or EM_386). Return -1 if unknown.
-template <typename E>
-static i64 get_machine_type(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  switch (get_file_type(mf)) {
-  case FileType::ELF_OBJ:
-  case FileType::ELF_DSO:
-    return ((ElfEhdr<E> *)mf->data)->e_machine;
-  case FileType::AR:
-    for (MappedFile<Context<E>> *child : read_fat_archive_members(ctx, mf))
-      if (get_file_type(child) == FileType::ELF_OBJ)
-        return ((ElfEhdr<E> *)child->data)->e_machine;
-    return -1;
-  case FileType::THIN_AR:
-    for (MappedFile<Context<E>> *child : read_thin_archive_members(ctx, mf))
-      if (get_file_type(child) == FileType::ELF_OBJ)
-        return ((ElfEhdr<E> *)child->data)->e_machine;
-    return -1;
-  case FileType::TEXT:
-    return get_script_output_type(ctx, mf);
-  default:
-    return -1;
-  }
-}
-
-template <typename E>
-static i64
-deduce_machine_type(Context<E> &ctx, std::span<std::string_view> args) {
-  for (std::string_view arg : args)
-    if (!arg.starts_with('-'))
-      if (auto *mf = MappedFile<Context<E>>::open(ctx, std::string(arg)))
-        if (i64 type = get_machine_type(ctx, mf); type != -1)
-          return type;
-  Fatal(ctx) << "-m option is missing";
-}
-
-template <typename E>
-MappedFile<Context<E>> *open_library(Context<E> &ctx, std::string path) {
-  MappedFile<Context<E>> *mf = MappedFile<Context<E>>::open(ctx, path);
-  if (!mf)
-    return nullptr;
-
-  i64 type = get_machine_type(ctx, mf);
-  if (type == -1 || type == E::e_machine)
-    return mf;
-  Warn(ctx) << path << ": skipping incompatible file " << (int)type
-            << " " << (int)E::e_machine;
-  return nullptr;
-}
-
-template <typename E>
-MappedFile<Context<E>> *find_library(Context<E> &ctx, std::string name) {
-  if (name.starts_with(':')) {
-    for (std::string_view dir : ctx.arg.library_paths) {
-      std::string path = std::string(dir) + "/" + name.substr(1);
-      if (MappedFile<Context<E>> *mf = open_library(ctx, path))
-        return mf;
-    }
-    Fatal(ctx) << "library not found: " << name;
-  }
-
-  for (std::string_view dir : ctx.arg.library_paths) {
-    std::string stem = std::string(dir) + "/lib" + name;
-    if (!ctx.is_static)
-      if (MappedFile<Context<E>> *mf = open_library(ctx, stem + ".so"))
-        return mf;
-    if (MappedFile<Context<E>> *mf = open_library(ctx, stem + ".a"))
-      return mf;
-  }
-  Fatal(ctx) << "library not found: " << name;
-}
-
-template <typename E>
-static void read_input_files(Context<E> &ctx, std::span<std::string_view> args) {
-  Timer t(ctx, "read_input_files");
-
-  std::vector<std::tuple<bool, bool, bool, bool>> state;
-  ctx.is_static = ctx.arg.is_static;
-
-  while (!args.empty()) {
-    std::string_view arg;
-
-    if (read_flag(args, "as-needed")) {
-      ctx.as_needed = true;
-    } else if (read_flag(args, "no-as-needed")) {
-      ctx.as_needed = false;
-    } else if (read_flag(args, "whole-archive")) {
-      ctx.whole_archive = true;
-    } else if (read_flag(args, "no-whole-archive")) {
-      ctx.whole_archive = false;
-    } else if (read_flag(args, "Bstatic")) {
-      ctx.is_static = true;
-    } else if (read_flag(args, "Bdynamic")) {
-      ctx.is_static = false;
-    } else if (read_flag(args, "start-lib")) {
-      ctx.in_lib = true;
-    } else if (read_flag(args, "end-lib")) {
-      ctx.in_lib = false;
-    } else if (read_arg(ctx, args, arg, "version-script")) {
-      parse_version_script(ctx, std::string(arg));
-    } else if (read_arg(ctx, args, arg, "dynamic-list")) {
-      parse_dynamic_list(ctx, std::string(arg));
-    } else if (read_flag(args, "push-state")) {
-      state.push_back({ctx.as_needed, ctx.whole_archive, ctx.is_static,
-                       ctx.in_lib});
-    } else if (read_flag(args, "pop-state")) {
-      if (state.empty())
-        Fatal(ctx) << "no state pushed before popping";
-      std::tie(ctx.as_needed, ctx.whole_archive, ctx.is_static, ctx.in_lib) =
-        state.back();
-      state.pop_back();
-    } else if (read_arg(ctx, args, arg, "l")) {
-      MappedFile<Context<E>> *mf = find_library(ctx, std::string(arg));
-      mf->given_fullpath = false;
-      read_file(ctx, mf);
-    } else {
-      read_file(ctx, MappedFile<Context<E>>::must_open(ctx, std::string(args[0])));
-      args = args.subspan(1);
-    }
-  }
-
-  if (ctx.objs.empty())
-    Fatal(ctx) << "no input files";
-
-  ctx.tg.wait();
-}
-
-template <typename E>
-static i64 get_mtime(Context<E> &ctx, std::string path) {
-  struct stat st;
-  if (stat(path.c_str(), &st) < 0)
-    Fatal(ctx) << path << ": stat failed: " << errno_string();
-  return st.st_mtime;
-}
-
-template <typename E>
-static bool reload_input_files(Context<E> &ctx) {
-  Timer t(ctx, "reload_input_files");
-
-  std::vector<ObjectFile<E> *> objs;
-  std::vector<SharedFile<E> *> dsos;
-
-  // Reload updated .o files
-  for (ObjectFile<E> *file : ctx.objs) {
-    if (file->mf->parent) {
-      if (get_mtime(ctx, file->mf->parent->name) != file->mf->parent->mtime)
-        return false;
-      objs.push_back(file);
-      continue;
-    }
-
-    if (get_mtime(ctx, file->mf->name) == file->mf->mtime) {
-      objs.push_back(file);
-      continue;
-    }
-
-    MappedFile<Context<E>> *mf =
-      MappedFile<Context<E>>::must_open(ctx, file->mf->name);
-    objs.push_back(new_object_file(ctx, mf, file->mf->name));
-  }
-
-  // Reload updated .so files
-  for (SharedFile<E> *file : ctx.dsos) {
-    if (get_mtime(ctx, file->mf->name) == file->mf->mtime) {
-      dsos.push_back(file);
-    } else {
-      MappedFile<Context<E>> *mf =
-        MappedFile<Context<E>>::must_open(ctx, file->mf->name);
-      dsos.push_back(new_shared_file(ctx, mf));
-    }
-  }
-
-  ctx.objs = objs;
-  ctx.dsos = dsos;
-  return true;
-}
-
-template <typename E>
-static void show_stats(Context<E> &ctx) {
-  for (ObjectFile<E> *obj : ctx.objs) {
-    static Counter defined("defined_syms");
-    defined += obj->first_global - 1;
-
-    static Counter undefined("undefined_syms");
-    undefined += obj->symbols.size() - obj->first_global;
-
-    for (std::unique_ptr<InputSection<E>> &sec : obj->sections) {
-      if (!sec || !sec->is_alive)
-        continue;
-
-      static Counter alloc("reloc_alloc");
-      static Counter nonalloc("reloc_nonalloc");
-
-      if (sec->shdr.sh_flags & SHF_ALLOC)
-        alloc += sec->get_rels(ctx).size();
-      else
-        nonalloc += sec->get_rels(ctx).size();
-    }
-
-    static Counter comdats("comdats");
-    comdats += obj->comdat_groups.size();
-
-    static Counter removed_comdats("removed_comdat_mem");
-    for (auto &pair : obj->comdat_groups)
-      if (ComdatGroup *group = pair.first; group->owner != obj->priority)
-        removed_comdats += pair.second.size();
-
-    static Counter num_cies("num_cies");
-    num_cies += obj->cies.size();
-
-    static Counter num_unique_cies("num_unique_cies");
-    for (CieRecord<E> &cie : obj->cies)
-      if (cie.is_leader)
-        num_unique_cies++;
-
-    static Counter num_fdes("num_fdes");
-    num_fdes +=  obj->fdes.size();
-  }
-
-  static Counter num_bytes("total_input_bytes");
-  for (std::unique_ptr<MappedFile<Context<E>>> &mf : ctx.mf_pool)
-    num_bytes += mf->size;
-
-  static Counter num_input_sections("input_sections");
-  for (ObjectFile<E> *file : ctx.objs)
-    num_input_sections += file->sections.size();
-
-  static Counter num_output_chunks("output_chunks", ctx.chunks.size());
-  static Counter num_objs("num_objs", ctx.objs.size());
-  static Counter num_dsos("num_dsos", ctx.dsos.size());
-
-  if constexpr (E::e_machine == EM_AARCH64) {
-    static Counter num_thunks("num_thunks");
-    for (std::unique_ptr<OutputSection<E>> &osec : ctx.output_sections)
-      for (std::unique_ptr<RangeExtensionThunk<E>> &thunk : osec->thunks)
-        num_thunks += thunk->symbols.size();
-  }
-
-  Counter::print();
-}
-
-static i64 get_default_thread_count() {
-  // mold doesn't scale above 32 threads.
-  int n = tbb::global_control::active_value(
-    tbb::global_control::max_allowed_parallelism);
-  return std::min(n, 32);
-}
-
-template <typename E>
-static int elf_main(int argc, char **argv) {
-  Context<E> ctx;
-
-  // Process -run option first. process_run_subcommand() does not return.
-  if (argc >= 2)
-    if (argv[1] == "-run"sv || argv[1] == "--run"sv)
-      process_run_subcommand(ctx, argc, argv);
-
-  // Parse non-positional command line options
-  ctx.cmdline_args = expand_response_files(ctx, argv);
-  std::vector<std::string_view> file_args;
-  parse_nonpositional_args(ctx, file_args);
-
-  // If no -m option is given, deduce it from input files.
-  if (ctx.arg.emulation == -1)
-    ctx.arg.emulation = deduce_machine_type(ctx, file_args);
-
-  // Redo if -m is not x86-64.
-  if (ctx.arg.emulation != E::e_machine) {
-    switch (ctx.arg.emulation) {
-    case EM_386:
-      return elf_main<I386>(argc, argv);
-    case EM_AARCH64:
-      return elf_main<ARM64>(argc, argv);
-    }
-    unreachable();
-  }
-
-  Timer t_all(ctx, "all");
-
-  if (ctx.arg.relocatable) {
-    combine_objects(ctx, file_args);
-    return 0;
-  }
-
-  if (!ctx.arg.preload)
-    try_resume_daemon(ctx);
-
-  i64 thread_count = ctx.arg.thread_count;
-  if (thread_count == 0)
-    thread_count = get_default_thread_count();
-  tbb::global_control tbb_cont(tbb::global_control::max_allowed_parallelism,
-                               thread_count);
-
-  install_signal_handler();
-
-  if (!ctx.arg.directory.empty())
-    if (chdir(ctx.arg.directory.c_str()) == -1)
-      Fatal(ctx) << "chdir failed: " << ctx.arg.directory
-                 << ": " << errno_string();
-
-  // Handle --wrap options if any.
-  for (std::string_view name : ctx.arg.wrap)
-    get_symbol(ctx, name)->wrap = true;
-
-  // Handle --retain-symbols-file options if any.
-  if (ctx.arg.retain_symbols_file)
-    for (std::string_view name : *ctx.arg.retain_symbols_file)
-      get_symbol(ctx, name)->write_to_symtab = true;
-
-  // Preload input files
-  std::function<void()> on_complete;
-  std::function<void()> wait_for_client;
-
-  if (ctx.arg.preload)
-    daemonize(ctx, &wait_for_client, &on_complete);
-  else if (ctx.arg.fork)
-    on_complete = fork_child();
-
-  for (std::string_view arg : ctx.arg.trace_symbol)
-    get_symbol(ctx, arg)->traced = true;
-
-  // Parse input files
-  read_input_files(ctx, file_args);
-
-  if (ctx.arg.preload) {
-    wait_for_client();
-    if (!reload_input_files(ctx)) {
-      std::vector<char *> args(argv, argv + argc);
-      args.push_back((char *)"--no-preload");
-      args.push_back(nullptr);
-      return elf_main<E>(argc + 1, args.data());
-    }
-  }
-
-  {
-    Timer t(ctx, "register_section_pieces");
-    tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
-      file->register_section_pieces(ctx);
-    });
-  }
-
-  // Uniquify shared object files by soname
-  {
-    std::unordered_set<std::string_view> seen;
-    std::erase_if(ctx.dsos, [&](SharedFile<E> *file) {
-      return !seen.insert(file->soname).second;
-    });
-  }
-
-  Timer t_total(ctx, "total");
-  Timer t_before_copy(ctx, "before_copy");
-
-  // Apply -exclude-libs
-  apply_exclude_libs(ctx);
-
-  // Create instances of linker-synthesized sections such as
-  // .got or .plt.
-  create_synthetic_sections(ctx);
-
-  // Resolve symbols and fix the set of object files that are
-  // included to the final output.
-  resolve_symbols(ctx);
-
-  // We currently do not natively support LTO.
-  // If LLVM LTO is in use, fallback to the lld linker by invoking
-  // it with the exact same command line arguments as we got.
-  if (ctx.llvm_lto) {
-    Warn(ctx) << "LLVM LTO is detected, so falling back to ld.lld";
-    argv[0] = (char *)"ld.lld";
-    execvp("ld.lld", argv);
-    Fatal(ctx) << "execvp failed: ld.lld: " << errno_string();
-  }
-
-  // Do the same for GCC LTO.
-  if (Symbol<E> *sym = get_symbol(ctx, "__gnu_lto_slim"); sym->file) {
-    Warn(ctx) << *sym->file
-              << "GCC LTO is detected, so falling back to ld.bfd";
-    execvp("ld.bfd", argv);
-    Fatal(ctx) << "execvp failed: ld.bfd: " << errno_string();
-  }
-
-  // Remove redundant comdat sections (e.g. duplicate inline functions).
-  eliminate_comdats(ctx);
-
-  // Create .bss sections for common symbols.
-  convert_common_symbols(ctx);
-
-  // Apply version scripts.
-  apply_version_script(ctx);
-
-  // Parse symbol version suffixes (e.g. "foo@ver1").
-  parse_symbol_version(ctx);
-
-  // Set is_import and is_export bits for each symbol.
-  compute_import_export(ctx);
-
-  // Garbage-collect unreachable sections.
-  if (ctx.arg.gc_sections)
-    gc_sections(ctx);
-
-  // Merge identical read-only sections.
-  if (ctx.arg.icf)
-    icf_sections(ctx);
-
-  // Compute sizes of sections containing mergeable strings.
-  compute_merged_section_sizes(ctx);
-
-  // Bin input sections into output sections.
-  bin_sections(ctx);
-
-  // Get a list of output sections.
-  append(ctx.chunks, collect_output_sections(ctx));
-
-  // Create a dummy file containing linker-synthesized symbols
-  // (e.g. `__bss_start`).
-  ctx.internal_obj = create_internal_file(ctx);
-  ctx.internal_obj->resolve_symbols(ctx);
-  ctx.objs.push_back(ctx.internal_obj);
-
-  // Beyond this point, no new files will be added to ctx.objs
-  // or ctx.dsos.
-
-  // Handle `-z cet-report`.
-  if (ctx.arg.z_cet_report != CET_REPORT_NONE)
-    check_cet_errors(ctx);
-
-  // If we are linking a .so file, remaining undefined symbols does
-  // not cause a linker error. Instead, they are treated as if they
-  // were imported symbols.
-  //
-  // If we are linking an executable, weak undefs are converted to
-  // weakly imported symbol so that they'll have another chance to be
-  // resolved.
-  claim_unresolved_symbols(ctx);
-
-  // Beyond this point, no new symbols will be added to the result.
-
-  // Make sure that all symbols have been resolved.
-  if (!ctx.arg.allow_multiple_definition)
-    check_duplicate_symbols(ctx);
-
-  // Handle --require-defined
-  for (std::string_view name : ctx.arg.require_defined)
-    if (!get_symbol(ctx, name)->file)
-      Error(ctx) << "--require-defined: undefined symbol: " << name;
-
-  // .init_array and .fini_array contents have to be sorted by
-  // a special rule. Sort them.
-  sort_init_fini(ctx);
-
-  // Compute sizes of output sections while assigning offsets
-  // within an output section to input sections.
-  compute_section_sizes(ctx);
-
-  // Sort sections by section attributes so that we'll have to
-  // create as few segments as possible.
-  sort(ctx.chunks, [&](Chunk<E> *a, Chunk<E> *b) {
-    return get_section_rank(ctx, a) < get_section_rank(ctx, b);
-  });
-
-  // Copy string referred by .dynamic to .dynstr.
-  for (SharedFile<E> *file : ctx.dsos)
-    ctx.dynstr->add_string(file->soname);
-  for (std::string_view str : ctx.arg.auxiliary)
-    ctx.dynstr->add_string(str);
-  for (std::string_view str : ctx.arg.filter)
-    ctx.dynstr->add_string(str);
-  if (!ctx.arg.rpaths.empty())
-    ctx.dynstr->add_string(ctx.arg.rpaths);
-  if (!ctx.arg.soname.empty())
-    ctx.dynstr->add_string(ctx.arg.soname);
-
-  // Scan relocations to find symbols that need entries in .got, .plt,
-  // .got.plt, .dynsym, .dynstr, etc.
-  scan_rels(ctx);
-
-  // If --packed_dyn_relocs=relr was given, base relocations are stored
-  // to a .relr.dyn section in a compressed form. Construct a compressed
-  // relocations now so that we can fix section sizes and file layout.
-  if (ctx.arg.pack_dyn_relocs_relr)
-    construct_relr(ctx);
-
-  // Reserve a space for dynamic symbol strings in .dynstr and sort
-  // .dynsym contents if necessary. Beyond this point, no symbol will
-  // be added to .dynsym.
-  ctx.dynsym->finalize(ctx);
-
-  // Fill .gnu.version_d section contents.
-  ctx.verdef->construct(ctx);
-
-  // Fill .gnu.version_r section contents.
-  ctx.verneed->construct(ctx);
-
-  // Compute .symtab and .strtab sizes for each file.
-  {
-    Timer t(ctx, "compute_symtab");
-    tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
-      file->compute_symtab(ctx);
-    });
-  }
-
-  // .eh_frame is a special section from the linker's point of view,
-  // as its contents are parsed and reconstructed by the linker,
-  // unlike other sections that are regarded as opaque bytes.
-  // Here, we transplant .eh_frame sections from a regular output
-  // section to the special EHFrameSection.
-  {
-    Timer t(ctx, "eh_frame");
-    std::erase_if(ctx.chunks, [](Chunk<E> *chunk) {
-      return chunk->kind == Chunk<E>::REGULAR && chunk->name == ".eh_frame";
-    });
-    ctx.eh_frame->construct(ctx);
-  }
-
-  // Update shdr.sh_size for each chunk and remove empty ones.
-  for (Chunk<E> *chunk : ctx.chunks)
-    chunk->update_shdr(ctx);
-
-  std::erase_if(ctx.chunks, [](Chunk<E> *chunk) {
-    return chunk->kind == Chunk<E>::SYNTHETIC && chunk->shdr.sh_size == 0;
-  });
-
-  // Set section indices.
-  for (i64 i = 0, shndx = 1; i < ctx.chunks.size(); i++)
-    if (ctx.chunks[i]->kind != Chunk<E>::HEADER)
-      ctx.chunks[i]->shndx = shndx++;
-
-  // Some types of section header refer other section by index.
-  // Recompute the section header to fill such fields with correct values.
-  for (Chunk<E> *chunk : ctx.chunks)
-    chunk->update_shdr(ctx);
-
-  // Assign offsets to output sections
-  i64 filesize = set_osec_offsets(ctx);
-
-  // On ARM64, we may need to create so-called "range extension thunks"
-  // to extend branch instructions reach, as they can jump only to
-  // ±128 MiB.
-  if constexpr (E::e_machine == EM_AARCH64)
-    filesize = create_range_extension_thunks(ctx);
-
-  // Fix linker-synthesized symbol addresses.
-  fix_synthetic_symbols(ctx);
-
-  // If --compress-debug-sections is given, compress .debug_* sections
-  // using zlib.
-  if (ctx.arg.compress_debug_sections != COMPRESS_NONE) {
-    compress_debug_sections(ctx);
-    filesize = set_osec_offsets(ctx);
-  }
-
-  // At this point, file layout is fixed.
-  // Beyond this, you can assume that symbol addresses including their
-  // GOT or PLT addresses have a correct final value.
-
-  // Some types of TLS relocations are defined relative to the beginning
-  // or the end of the TLS segment address. Find these addresses now.
-  for (ElfPhdr<E> phdr : create_phdr(ctx)) {
-    if (phdr.p_type == PT_TLS) {
-      ctx.tls_begin = phdr.p_vaddr;
-      ctx.tls_end = align_to(phdr.p_vaddr + phdr.p_memsz, phdr.p_align);
-      break;
-    }
-  }
-
-  t_before_copy.stop();
-
-  // Create an output file
-  ctx.output_file = OutputFile<E>::open(ctx, ctx.arg.output, filesize, 0777);
-  ctx.buf = ctx.output_file->buf;
-
-  Timer t_copy(ctx, "copy");
-
-  // Copy input sections to the output file
-  {
-    Timer t(ctx, "copy_buf");
-
-    tbb::parallel_for_each(ctx.chunks, [&](Chunk<E> *chunk) {
-      std::string name(chunk->name);
-      if (name.empty())
-        name = "(header)";
-      Timer t2(ctx, name, &t);
-
-      chunk->copy_buf(ctx);
-    });
-
-    ctx.checkpoint();
-  }
-
-  if constexpr (E::e_machine == EM_AARCH64)
-    write_thunks(ctx);
-
-  // Dynamic linker works better with sorted .rela.dyn section,
-  // so we sort them.
-  ctx.reldyn->sort(ctx);
-
-  // Zero-clear paddings between sections
-  clear_padding(ctx);
-
-  if (ctx.buildid) {
-    Timer t(ctx, "build_id");
-    ctx.buildid->write_buildid(ctx);
-  }
-
-  t_copy.stop();
-
-  // Close the output file. This is the end of the linker's main job.
-  ctx.output_file->close(ctx);
-
-  t_total.stop();
-  t_all.stop();
-
-  if (ctx.arg.print_map)
-    print_map(ctx);
-
-  // Show stats numbers
-  if (ctx.arg.stats)
-    show_stats(ctx);
-
-  if (ctx.arg.perf)
-    print_timer_records(ctx.timer_records);
-
-  std::cout << std::flush;
-  std::cerr << std::flush;
-  if (on_complete)
-    on_complete();
-
-  if (ctx.arg.quick_exit)
-    _exit(0);
-
-  for (std::function<void()> &fn : ctx.on_exit)
-    fn();
-  return 0;
-}
-
-int main(int argc, char **argv) {
-  return elf_main<X86_64>(argc, argv);
-}
-
-#define INSTANTIATE(E)                                                  \
-  template void read_file(Context<E> &, MappedFile<Context<E>> *);
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/mapfile.cc

@@ -1,111 +0,0 @@
-#include "mold.h"
-
-#include <fstream>
-#include <iomanip>
-#include <ios>
-#include <sstream>
-#include <tbb/parallel_for_each.h>
-#include <unordered_map>
-
-namespace mold::elf {
-
-template <typename E>
-using Map =
-  tbb::concurrent_hash_map<InputSection<E> *, std::vector<Symbol<E> *>>;
-
-template <typename E>
-static std::unique_ptr<std::ofstream> open_output_file(Context<E> &ctx) {
-  std::unique_ptr<std::ofstream> file(new std::ofstream);
-  file->open(ctx.arg.Map.c_str());
-  if (!file->is_open())
-    Fatal(ctx) << "cannot open " << ctx.arg.Map << ": " << errno_string();
-  return file;
-}
-
-template <typename E>
-static Map<E> get_map(Context<E> &ctx) {
-  Map<E> map;
-
-  tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
-    for (Symbol<E> *sym : file->symbols) {
-      if (sym->file == file && sym->input_section &&
-          sym->get_type() != STT_SECTION) {
-        assert(file == &sym->input_section->file);
-
-        typename Map<E>::accessor acc;
-        map.insert(acc, {sym->input_section, {}});
-        acc->second.push_back(sym);
-      }
-    }
-  });
-
-  tbb::parallel_for(map.range(), [](const typename Map<E>::range_type &range) {
-    for (auto it = range.begin(); it != range.end(); it++) {
-      std::vector<Symbol<E> *> &vec = it->second;
-      sort(vec, [](Symbol<E> *a, Symbol<E> *b) { return a->value < b->value; });
-    }
-  });
-  return map;
-}
-
-template <typename E>
-void print_map(Context<E> &ctx) {
-  std::ostream *out = &std::cout;
-  std::unique_ptr<std::ofstream> file;
-
-  if (!ctx.arg.Map.empty()) {
-    file = open_output_file(ctx);
-    out = file.get();
-  }
-
-  // Construct a section-to-symbol map.
-  Map<E> map = get_map(ctx);
-
-  // Print a mapfile.
-  *out << "             VMA       Size Align Out     In      Symbol\n";
-
-  for (Chunk<E> *osec : ctx.chunks) {
-    *out << std::setw(16) << (u64)osec->shdr.sh_addr
-         << std::setw(11) << (u64)osec->shdr.sh_size
-         << std::setw(6) << (u64)osec->shdr.sh_addralign
-         << " " << osec->name << "\n";
-
-    if (osec->kind != Chunk<E>::REGULAR)
-      continue;
-
-    std::span<InputSection<E> *> members = ((OutputSection<E> *)osec)->members;
-    std::vector<std::string> bufs(members.size());
-
-    tbb::parallel_for((i64)0, (i64)members.size(), [&](i64 i) {
-      InputSection<E> *mem = members[i];
-      std::ostringstream ss;
-      opt_demangle = ctx.arg.demangle;
-
-      ss << std::setw(16) << (osec->shdr.sh_addr + mem->offset)
-         << std::setw(11) << (u64)mem->shdr.sh_size
-         << std::setw(6) << (u64)mem->shdr.sh_addralign
-         << "         " << *mem << "\n";
-
-      typename Map<E>::const_accessor acc;
-      if (map.find(acc, mem))
-        for (Symbol<E> *sym : acc->second)
-          ss << std::setw(16) << sym->get_addr(ctx)
-             << "          0     0                 "
-             << *sym << "\n";
-
-      bufs[i] = ss.str();
-    });
-
-    for (std::string &str : bufs)
-      *out << str;
-  }
-}
-
-#define INSTANTIATE(E)                          \
-  template void print_map(Context<E> &ctx);
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/output-file.cc

@@ -1,142 +0,0 @@
-#include "mold.h"
-
-#include <fcntl.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-namespace mold::elf {
-
-inline u32 get_umask() {
-  u32 orig_umask = umask(0);
-  umask(orig_umask);
-  return orig_umask;
-}
-
-template <typename C>
-static std::pair<i64, char *>
-open_or_create_file(C &ctx, std::string path, i64 filesize, i64 perm) {
-  std::string tmpl = filepath(path).parent_path() / ".mold-XXXXXX";
-  char *path2 = (char *)save_string(ctx, tmpl).data();
-
-  i64 fd = mkstemp(path2);
-  if (fd == -1)
-    Fatal(ctx) << "cannot open " << path2 <<  ": " << errno_string();
-
-  // Reuse an existing file if exists and writable.
-  if (rename(path.c_str(), path2) == 0) {
-    ::close(fd);
-    fd = ::open(path2, O_RDWR | O_CREAT, perm);
-    if (fd != -1 && !ftruncate(fd, filesize) && !fchmod(fd, perm & ~get_umask()))
-      return {fd, path2};
-
-    unlink(path2);
-    fd = ::open(path2, O_RDWR | O_CREAT, perm);
-    if (fd == -1)
-      Fatal(ctx) << "cannot open " << path2 << ": " << errno_string();
-  }
-
-  if (ftruncate(fd, filesize))
-    Fatal(ctx) << "ftruncate failed: " << errno_string();
-
-  if (fchmod(fd, (perm & ~get_umask())) == -1)
-    Fatal(ctx) << "fchmod failed: " << errno_string();
-  return {fd, path2};
-}
-
-template <typename E>
-class MemoryMappedOutputFile : public OutputFile<E> {
-public:
-  MemoryMappedOutputFile(Context<E> &ctx, std::string path, i64 filesize, i64 perm)
-    : OutputFile<E>(path, filesize, true) {
-    i64 fd;
-    std::tie(fd, output_tmpfile) = open_or_create_file(ctx, path, filesize, perm);
-
-    this->buf = (u8 *)mmap(nullptr, filesize, PROT_READ | PROT_WRITE,
-                           MAP_SHARED, fd, 0);
-    if (this->buf == MAP_FAILED)
-      Fatal(ctx) << path << ": mmap failed: " << errno_string();
-    ::close(fd);
-  }
-
-  void close(Context<E> &ctx) override {
-    Timer t(ctx, "close_file");
-
-    if (!this->is_unmapped)
-      munmap(this->buf, this->filesize);
-
-    if (rename(output_tmpfile, this->path.c_str()) == -1)
-      Fatal(ctx) << this->path << ": rename failed: " << errno_string();
-    output_tmpfile = nullptr;
-  }
-};
-
-template <typename E>
-class MallocOutputFile : public OutputFile<E> {
-public:
-  MallocOutputFile(Context<E> &ctx, std::string path, i64 filesize, i64 perm)
-    : OutputFile<E>(path, filesize, false), perm(perm) {
-    this->buf = (u8 *)mmap(NULL, filesize, PROT_READ | PROT_WRITE,
-                           MAP_SHARED | MAP_ANONYMOUS, -1, 0);
-    if (this->buf == MAP_FAILED)
-      Fatal(ctx) << "mmap failed: " << errno_string();
-  }
-
-  void close(Context<E> &ctx) override {
-    Timer t(ctx, "close_file");
-
-    if (this->path == "-") {
-      fwrite(this->buf, this->filesize, 1, stdout);
-      fclose(stdout);
-      return;
-    }
-
-    i64 fd = ::open(this->path.c_str(), O_RDWR | O_CREAT, perm);
-    if (fd == -1)
-      Fatal(ctx) << "cannot open " << this->path << ": " << errno_string();
-
-    FILE *fp = fdopen(fd, "w");
-    fwrite(this->buf, this->filesize, 1, fp);
-    fclose(fp);
-  }
-
-private:
-  i64 perm;
-};
-
-template <typename E>
-std::unique_ptr<OutputFile<E>>
-OutputFile<E>::open(Context<E> &ctx, std::string path, i64 filesize, i64 perm) {
-  Timer t(ctx, "open_file");
-
-  if (path.starts_with('/') && !ctx.arg.chroot.empty())
-    path = ctx.arg.chroot + "/" + path_clean(path);
-
-  bool is_special = false;
-  if (path == "-") {
-    is_special = true;
-  } else {
-    struct stat st;
-    if (stat(path.c_str(), &st) == 0 && (st.st_mode & S_IFMT) != S_IFREG)
-      is_special = true;
-  }
-
-  std::unique_ptr<OutputFile<E>> file;
-  if (is_special)
-    file = std::make_unique<MallocOutputFile<E>>(ctx, path, filesize, perm);
-  else
-    file = std::make_unique<MemoryMappedOutputFile<E>>(ctx, path, filesize, perm);
-
-  if (ctx.arg.filler != -1)
-    memset(file->buf, ctx.arg.filler, filesize);
-  return file;
-}
-
-#define INSTANTIATE(E)                          \
-  template class OutputFile<E>;
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/relocatable.cc

@@ -1,595 +0,0 @@
-// This file implements the -r or ---relocatable option. If the option
-// is given to a linker, the linker creates not an executable nor a
-// shared library file but instead create another object file by
-// combining input object files.
-//
-// The behavior of a linker with -r is pretty different from the one
-// without -r, and adding code for -r to the regular code path could
-// severely complicate it. Therefore, we implement -r as an
-// independent feature from others. That's why this file share only a
-// small amount of code with other files.
-//
-// Since the -r option is a minor feature, we don't aim for speed in
-// this file. This is a "better than nothing" implementation of the -r
-// option.
-//
-// Here is the strategy as to how to combine multiple object files
-// into one:
-//
-//  - Regular sections containing opaque data (e.g. ".text" or
-//    ".data") are just copied as-is without merging.
-//
-//  - .symtab, .strtab and .shstrtab are merged.
-//
-//  - COMDAT groups are uniquified.
-//
-//  - Relocation sections are copied one by one, but we need to fix
-//    symbol indices.
-
-#include "mold.h"
-#include "../archive-file.h"
-
-#include <unordered_map>
-#include <unordered_set>
-
-namespace mold::elf {
-
-template <typename E> class RObjectFile;
-
-template <typename E>
-class RChunk {
-public:
-  RChunk() {
-    out_shdr.sh_addralign = 1;
-  }
-
-  virtual ~RChunk() = default;
-  virtual void update_shdr(Context<E> &ctx) {}
-  virtual void write_to(Context<E> &ctx) = 0;
-
-  std::string_view name;
-  i64 shndx = 0;
-  ElfShdr<E> in_shdr = {};
-  ElfShdr<E> out_shdr = {};
-};
-
-template <typename E>
-class RInputSection : public RChunk<E> {
-public:
-  RInputSection(Context<E> &ctx, RObjectFile<E> &file, const ElfShdr<E> &shdr);
-  void update_shdr(Context<E> &ctx) override;
-  void write_to(Context<E> &ctx) override;
-
-  RObjectFile<E> &file;
-};
-
-template <typename E>
-class RSymtabSection : public RChunk<E> {
-public:
-  RSymtabSection() {
-    this->name = ".symtab";
-    this->out_shdr.sh_type = SHT_SYMTAB;
-    this->out_shdr.sh_entsize = sizeof(ElfSym<E>);
-    this->out_shdr.sh_addralign = E::word_size;
-  }
-
-  void add_local_symbol(Context<E> &ctx, RObjectFile<E> &file, i64 idx);
-  void add_global_symbol(Context<E> &ctx, RObjectFile<E> &file, i64 idx);
-  void update_shdr(Context<E> &ctx) override;
-  void write_to(Context<E> &ctx) override;
-
-  std::unordered_map<std::string_view, i64> sym_map;
-  std::vector<ElfSym<E>> syms{1};
-};
-
-template <typename E>
-class RStrtabSection : public RChunk<E> {
-public:
-  RStrtabSection(std::string_view name) {
-    this->name = name;
-    this->out_shdr.sh_type = SHT_STRTAB;
-    this->out_shdr.sh_size = 1;
-  }
-
-  i64 add_string(std::string_view str);
-  void write_to(Context<E> &ctx) override;
-
-  std::unordered_map<std::string_view, i64> strings;
-};
-
-template <typename E>
-class ROutputEhdr : public RChunk<E> {
-public:
-  ROutputEhdr() {
-    this->out_shdr.sh_size = sizeof(ElfEhdr<E>);
-  }
-
-  void write_to(Context<E> &ctx) override;
-};
-
-template <typename E>
-class ROutputShdr : public RChunk<E> {
-public:
-  ROutputShdr() {
-    this->out_shdr.sh_size = sizeof(ElfShdr<E>);
-  }
-
-  void update_shdr(Context<E> &ctx) override;
-  void write_to(Context<E> &ctx) override;
-};
-
-template <typename E>
-class RObjectFile {
-public:
-  RObjectFile(Context<E> &ctx, MappedFile<Context<E>> &mf, bool is_alive);
-
-  void remove_comdats(Context<E> &ctx,
-                      std::unordered_set<std::string_view> &groups);
-
-  template <typename T>
-  std::span<T> get_data(Context<E> &ctx, const ElfShdr<E> &shdr);
-
-  MappedFile<Context<E>> &mf;
-  std::span<ElfShdr<E>> elf_sections;
-  std::vector<std::unique_ptr<RInputSection<E>>> sections;
-  std::span<const ElfSym<E>> syms;
-  std::vector<i64> symidx;
-  std::unordered_set<std::string_view> defined_syms;
-  std::unordered_set<std::string_view> undef_syms;
-  i64 symtab_shndx = 0;
-  i64 first_global = 0;
-  bool is_alive;
-  const char *strtab = nullptr;
-  const char *shstrtab = nullptr;
-};
-
-template <typename E>
-void RSymtabSection<E>::add_local_symbol(Context<E> &ctx, RObjectFile<E> &file,
-                                         i64 idx) {
-  ElfSym<E> sym = file.syms[idx];
-  assert(sym.st_bind == STB_LOCAL);
-
-  if (!sym.is_undef() && !sym.is_abs() && !sym.is_common()) {
-    if (!file.sections[sym.st_shndx])
-      return;
-    sym.st_shndx = file.sections[sym.st_shndx]->shndx;
-  }
-
-  std::string_view name = file.strtab + sym.st_name;
-  sym.st_name = ctx.r_strtab->add_string(name);
-
-  file.symidx[idx] = syms.size();
-  syms.push_back(sym);
-}
-
-template <typename E>
-void RSymtabSection<E>::add_global_symbol(Context<E> &ctx, RObjectFile<E> &file,
-                                          i64 idx) {
-  ElfSym<E> sym = file.syms[idx];
-  assert(sym.st_bind != STB_LOCAL);
-
-  std::string_view name = file.strtab + sym.st_name;
-  auto [it, inserted] = sym_map.insert({name, syms.size()});
-
-  if (inserted) {
-    if (!sym.is_undef() && !sym.is_abs() && !sym.is_common())
-      sym.st_shndx = file.sections[sym.st_shndx]->shndx;
-    sym.st_name = ctx.r_strtab->add_string(name);
-    file.symidx[idx] = syms.size();
-    syms.push_back(sym);
-    return;
-  }
-
-  file.symidx[idx] = it->second;
-
-  ElfSym<E> &existing = syms[it->second];
-  if (existing.is_undef() && !sym.is_undef()) {
-    if (!sym.is_abs() && !sym.is_common())
-      sym.st_shndx = file.sections[sym.st_shndx]->shndx;
-    sym.st_name = existing.st_name;
-    existing = sym;
-  }
-}
-
-template <typename E>
-void RSymtabSection<E>::update_shdr(Context<E> &ctx) {
-  this->out_shdr.sh_size = syms.size() * sizeof(ElfSym<E>);
-  this->out_shdr.sh_link = ctx.r_strtab->shndx;
-}
-
-template <typename E>
-void RSymtabSection<E>::write_to(Context<E> &ctx) {
-  ElfSym<E> *buf = (ElfSym<E> *)(ctx.buf + this->out_shdr.sh_offset);
-  for (i64 i = 1; i < syms.size(); i++)
-    buf[i] = syms[i];
-}
-
-template <typename E>
-RInputSection<E>::RInputSection(Context<E> &ctx, RObjectFile<E> &file,
-                                const ElfShdr<E> &shdr)
-  : file(file) {
-  this->name = file.shstrtab + shdr.sh_name;
-  this->in_shdr = shdr;
-  this->out_shdr = shdr;
-}
-
-template <typename E>
-void RInputSection<E>::update_shdr(Context<E> &ctx) {
-  switch (this->in_shdr.sh_type) {
-  case SHT_GROUP:
-    this->out_shdr.sh_link = ctx.r_symtab->shndx;
-    this->out_shdr.sh_info = file.symidx[this->in_shdr.sh_info];
-    break;
-  case SHT_REL:
-  case SHT_RELA:
-    this->out_shdr.sh_link = ctx.r_symtab->shndx;
-    this->out_shdr.sh_info = file.sections[this->in_shdr.sh_info]->shndx;
-    break;
-  default:
-    if (this->in_shdr.sh_link) {
-      std::unique_ptr<RInputSection<E>> &sec =
-        file.sections[this->in_shdr.sh_info];
-
-      if (sec)
-        this->out_shdr.sh_link = sec->shndx;
-      else if (this->in_shdr.sh_link == file.symtab_shndx)
-        this->out_shdr.sh_link = ctx.r_symtab->shndx;
-    }
-  }
-}
-
-template <typename E>
-void RInputSection<E>::write_to(Context<E> &ctx) {
-  if (this->in_shdr.sh_type == SHT_NOBITS)
-    return;
-
-  std::span<u8> contents = file.template get_data<u8>(ctx, this->in_shdr);
-  memcpy(ctx.buf + this->out_shdr.sh_offset, contents.data(), contents.size());
-
-  switch (this->in_shdr.sh_type) {
-  case SHT_GROUP: {
-    u32 *mem = (u32 *)(ctx.buf + this->out_shdr.sh_offset);
-    for (i64 i = 1; i < this->out_shdr.sh_size / sizeof(u32); i++)
-      mem[i] = file.sections[mem[i]]->shndx;
-    break;
-  }
-  case SHT_REL:
-  case SHT_RELA: {
-    ElfRel<E> *rel = (ElfRel<E> *)(ctx.buf + this->out_shdr.sh_offset);
-    i64 size = this->out_shdr.sh_size / sizeof(ElfRel<E>);
-
-    for (i64 i = 0; i < size; i++) {
-      const ElfSym<E> &sym = file.syms[rel[i].r_sym];
-      if (sym.is_undef() || sym.is_abs() || sym.is_common() ||
-          file.sections[sym.st_shndx])
-        rel[i].r_sym = file.symidx[rel[i].r_sym];
-      else
-        memset(rel + i, 0, sizeof(ElfRel<E>));
-    }
-
-    i64 i = 0;
-    i64 j = 0;
-    for (; j < size; j++)
-      if (rel[j].r_type)
-        rel[i++] = rel[j];
-    for (; i < size; i++)
-      memset(rel + i, 0, sizeof(ElfRel<E>));
-    break;
-  }
-  }
-}
-
-template <typename E>
-i64 RStrtabSection<E>::add_string(std::string_view str) {
-  auto [it, inserted] = strings.insert({str, this->out_shdr.sh_size});
-  if (inserted)
-    this->out_shdr.sh_size += str.size() + 1;
-  return it->second;
-}
-
-template <typename E>
-void RStrtabSection<E>::write_to(Context<E> &ctx) {
-  for (auto [str, offset] : strings)
-    memcpy(ctx.buf + this->out_shdr.sh_offset + offset, str.data(), str.size());
-}
-
-template <typename E>
-void ROutputEhdr<E>::write_to(Context<E> &ctx) {
-  ElfEhdr<E> &hdr = *(ElfEhdr<E> *)(ctx.buf + this->out_shdr.sh_offset);
-  memcpy(&hdr.e_ident, "\177ELF", 4);
-  hdr.e_ident[EI_CLASS] = (E::word_size == 8) ? ELFCLASS64 : ELFCLASS32;
-  hdr.e_ident[EI_DATA] = E::is_le ? ELFDATA2LSB : ELFDATA2MSB;
-  hdr.e_ident[EI_VERSION] = EV_CURRENT;
-  hdr.e_type = ET_REL;
-  hdr.e_machine = E::e_machine;
-  hdr.e_version = EV_CURRENT;
-  hdr.e_shoff = ctx.r_shdr->out_shdr.sh_offset;
-  hdr.e_ehsize = sizeof(ElfEhdr<E>);
-  hdr.e_shentsize = sizeof(ElfShdr<E>);
-  hdr.e_shstrndx = ctx.r_shstrtab->shndx;
-  hdr.e_shnum = ctx.r_shdr->out_shdr.sh_size / sizeof(ElfShdr<E>);
-  hdr.e_shstrndx = ctx.r_shstrtab->shndx;
-}
-
-template <typename E>
-void ROutputShdr<E>::update_shdr(Context<E> &ctx) {
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    if (chunk->shndx)
-      this->out_shdr.sh_size += sizeof(ElfShdr<E>);
-}
-
-template <typename E>
-void ROutputShdr<E>::write_to(Context<E> &ctx) {
-  ElfShdr<E> *hdr = (ElfShdr<E> *)(ctx.buf + this->out_shdr.sh_offset);
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    if (chunk->shndx)
-      hdr[chunk->shndx] = chunk->out_shdr;
-}
-
-template <typename E>
-RObjectFile<E>::RObjectFile(Context<E> &ctx, MappedFile<Context<E>> &mf,
-                            bool is_alive)
-  : mf(mf), is_alive(is_alive) {
-  // Read ELF header and section header
-  ElfEhdr<E> &ehdr = *(ElfEhdr<E> *)mf.data;
-  ElfShdr<E> *sh_begin = (ElfShdr<E> *)(mf.data + ehdr.e_shoff);
-  i64 num_sections = (ehdr.e_shnum == 0) ? sh_begin->sh_size : ehdr.e_shnum;
-  elf_sections = {sh_begin, sh_begin + num_sections};
-  sections.resize(num_sections);
-
-  // Read .shstrtab
-  i64 shstrtab_idx = (ehdr.e_shstrndx == SHN_XINDEX)
-    ? sh_begin->sh_link : ehdr.e_shstrndx;
-  shstrtab = (char *)(mf.data + elf_sections[shstrtab_idx].sh_offset);
-
-  // Read .symtab
-  for (i64 i = 1; i < elf_sections.size(); i++) {
-    ElfShdr<E> &shdr = elf_sections[i];
-    if (shdr.sh_type == SHT_SYMTAB) {
-      syms = get_data<const ElfSym<E>>(ctx, shdr);
-      strtab = (char *)(mf.data + elf_sections[shdr.sh_link].sh_offset);
-      symtab_shndx = i;
-      first_global = shdr.sh_info;
-      break;
-    }
-  }
-
-  symidx.resize(syms.size());
-
-  // Read sections
-  for (i64 i = 1; i < elf_sections.size(); i++) {
-    ElfShdr<E> &shdr = elf_sections[i];
-    switch (shdr.sh_type) {
-    case SHT_NULL:
-    case SHT_SYMTAB:
-    case SHT_STRTAB:
-      break;
-    default:
-      sections[i].reset(new RInputSection(ctx, *this, shdr));
-    }
-  }
-
-  // Read global symbols
-  for (i64 i = first_global; i < syms.size(); i++) {
-    std::string_view name = strtab + syms[i].st_name;
-    if (syms[i].is_defined())
-      defined_syms.insert(name);
-    else
-      undef_syms.insert(name);
-  }
-}
-
-// Remove duplicate comdat groups
-template <typename E>
-void RObjectFile<E>::remove_comdats(Context<E> &ctx,
-                                    std::unordered_set<std::string_view> &groups) {
-  for (i64 i = 1; i < sections.size(); i++) {
-    if (!sections[i])
-      continue;
-    ElfShdr<E> &shdr = sections[i]->in_shdr;
-    if (shdr.sh_type != SHT_GROUP)
-      continue;
-
-    // Get a comdat group signature and insert it into a set.
-    const ElfSym<E> &sym = syms[shdr.sh_info];
-    std::string_view signature = strtab + sym.st_name;
-    if (groups.insert(signature).second)
-      continue;
-
-    // If it is a duplicate, remove it and its members.
-    for (i64 j : this->template get_data<u32>(ctx, shdr).subspan(1))
-      sections[j] = nullptr;
-    sections[i] = nullptr;
-  }
-}
-
-template <typename E>
-template <typename T>
-std::span<T> RObjectFile<E>::get_data(Context<E> &ctx, const ElfShdr<E> &shdr) {
-  T *begin = (T *)(mf.data + shdr.sh_offset);
-  T *end   = (T *)(mf.data + shdr.sh_offset + shdr.sh_size);
-  return {begin, end};
-}
-
-template <typename E>
-static std::vector<std::unique_ptr<RObjectFile<E>>>
-open_files(Context<E> &ctx, std::span<std::string_view> args) {
-  std::vector<std::unique_ptr<RObjectFile<E>>> files;
-  bool whole_archive = false;
-
-  while (!args.empty()) {
-    if (read_flag(args, "whole-archive")) {
-      whole_archive = true;
-      continue;
-    }
-
-    if (read_flag(args, "no-whole-archive")) {
-      whole_archive = false;
-      continue;
-    }
-
-    std::string_view arg;
-    if (read_arg(ctx, args, arg, "version-script") ||
-        read_arg(ctx, args, arg, "dynamic-list"))
-      continue;
-
-    MappedFile<Context<E>> *mf = nullptr;
-
-    if (read_arg(ctx, args, arg, "l")) {
-      mf = find_library(ctx, std::string(arg));
-    } else {
-      if (arg.starts_with('-'))
-        continue;
-      arg = args[0];
-      args = args.subspan(1);
-      mf = MappedFile<Context<E>>::must_open(ctx, std::string(arg));
-    }
-
-    switch (get_file_type(mf)) {
-    case FileType::ELF_OBJ:
-      files.emplace_back(new RObjectFile<E>(ctx, *mf, true));
-      break;
-    case FileType::AR:
-    case FileType::THIN_AR:
-      for (MappedFile<Context<E>> *child : read_archive_members(ctx, mf))
-        if (get_file_type(child) == FileType::ELF_OBJ)
-          files.emplace_back(new RObjectFile<E>(ctx, *child, whole_archive));
-      break;
-    default:
-      break;
-    }
-  }
-  return files;
-}
-
-template <typename E>
-static i64 assign_offsets(Context<E> &ctx) {
-  i64 offset = 0;
-  for (RChunk<E> *chunk : ctx.r_chunks) {
-    offset = align_to(offset, chunk->out_shdr.sh_addralign);
-    chunk->out_shdr.sh_offset = offset;
-    offset += chunk->out_shdr.sh_size;
-  }
-  return offset;
-}
-
-static bool contains(std::unordered_set<std::string_view> &a,
-                     std::unordered_set<std::string_view> &b) {
-  for (std::string_view x : b)
-    if (a.contains(x))
-      return true;
-  return false;
-}
-
-template <typename E>
-void combine_objects(Context<E> &ctx, std::span<std::string_view> file_args) {
-  // Read object files
-  std::vector<std::unique_ptr<RObjectFile<E>>> files = open_files(ctx, file_args);
-
-  // Identify needed objects
-  std::unordered_set<std::string_view> undef_syms;
-  auto add_syms = [&](RObjectFile<E> &file) {
-    undef_syms.insert(file.undef_syms.begin(), file.undef_syms.end());
-    for (std::string_view name : file.defined_syms)
-      undef_syms.erase(name);
-    file.is_alive = true;
-  };
-
-  for (std::unique_ptr<RObjectFile<E>> &file : files)
-    if (file->is_alive)
-      add_syms(*file);
-
-  for (;;) {
-    bool added = false;
-    for (std::unique_ptr<RObjectFile<E>> &file : files) {
-      if (!file->is_alive && contains(undef_syms, file->defined_syms)) {
-        add_syms(*file);
-        added = true;
-      }
-    }
-    if (!added)
-      break;
-  }
-
-  std::erase_if(files, [](std::unique_ptr<RObjectFile<E>> &file) {
-    return !file->is_alive;
-  });
-
-  // Remove duplicate comdat groups
-  std::unordered_set<std::string_view> comdat_groups;
-  for (std::unique_ptr<RObjectFile<E>> &file : files)
-    file->remove_comdats(ctx, comdat_groups);
-
-  // Create headers and linker-synthesized sections
-  ROutputEhdr<E> ehdr;
-  ROutputShdr<E> shdr;
-  RSymtabSection<E> symtab;
-  RStrtabSection<E> shstrtab(".shstrtab");
-  RStrtabSection<E> strtab(".strtab");
-
-  ctx.r_chunks.push_back(&ehdr);
-  ctx.r_chunks.push_back(&shstrtab);
-  ctx.r_chunks.push_back(&strtab);
-
-  ctx.r_ehdr = &ehdr;
-  ctx.r_shdr = &shdr;
-  ctx.r_shstrtab = &shstrtab;
-  ctx.r_strtab = &strtab;
-  ctx.r_symtab = &symtab;
-
-  // Add input sections to output sections
-  for (std::unique_ptr<RObjectFile<E>> &file : files)
-    for (std::unique_ptr<RInputSection<E>> &sec : file->sections)
-      if (sec)
-        ctx.r_chunks.push_back(sec.get());
-
-  ctx.r_chunks.push_back(&symtab);
-  ctx.r_chunks.push_back(&shdr);
-
-  // Assign output section indices
-  i64 shndx = 1;
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    if (chunk != &ehdr && chunk != &shdr)
-      chunk->shndx = shndx++;
-
-  // Add section names to .shstrtab
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    if (chunk->shndx)
-      chunk->out_shdr.sh_name = shstrtab.add_string(chunk->name);
-
-  // Copy symbols from input objects to an output object
-  for (std::unique_ptr<RObjectFile<E>> &file : files)
-    for (i64 i = 1; i < file->first_global; i++)
-      symtab.add_local_symbol(ctx, *file, i);
-
-  symtab.out_shdr.sh_info = symtab.syms.size();
-
-  for (std::unique_ptr<RObjectFile<E>> &file : files)
-    for (i64 i = file->first_global; i < file->syms.size(); i++)
-      symtab.add_global_symbol(ctx, *file, i);
-
-  // Finalize section header
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    chunk->update_shdr(ctx);
-
-  // Open an output file
-  i64 filesize = assign_offsets(ctx);
-  std::unique_ptr<OutputFile<E>> out =
-    OutputFile<E>::open(ctx, ctx.arg.output, filesize, 0666);
-  memset(out->buf, 0, filesize);
-  ctx.buf = out->buf;
-
-  // Write to the output file
-  for (RChunk<E> *chunk : ctx.r_chunks)
-    chunk->write_to(ctx);
-  out->close(ctx);
-}
-
-#define INSTANTIATE(E)                                                  \
-  template void combine_objects(Context<E> &, std::span<std::string_view>);
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

elf/subprocess.cc

@@ -1,325 +0,0 @@
-#include "mold.h"
-
-#include <filesystem>
-#include <signal.h>
-#include <sys/select.h>
-#include <sys/socket.h>
-#include <sys/stat.h>
-#include <sys/time.h>
-#include <sys/types.h>
-#include <sys/un.h>
-#include <sys/wait.h>
-#include <unistd.h>
-
-#ifdef __APPLE__
-#  define COMMON_DIGEST_FOR_OPENSSL
-#  include <CommonCrypto/CommonDigest.h>
-#else
-#  include <openssl/sha.h>
-#endif
-
-#define DAEMON_TIMEOUT 30
-
-namespace mold::elf {
-
-// Exiting from a program with large memory usage is slow --
-// it may take a few hundred milliseconds. To hide the latency,
-// we fork a child and let it do the actual linking work.
-std::function<void()> fork_child() {
-  int pipefd[2];
-  if (pipe(pipefd) == -1) {
-    perror("pipe");
-    exit(1);
-  }
-
-  pid_t pid = fork();
-  if (pid == -1) {
-    perror("fork");
-    exit(1);
-  }
-
-  if (pid > 0) {
-    // Parent
-    close(pipefd[1]);
-
-    char buf[1];
-    if (read(pipefd[0], buf, 1) == 1)
-      _exit(0);
-
-    int status;
-    waitpid(pid, &status, 0);
-
-    if (WIFEXITED(status))
-      _exit(WEXITSTATUS(status));
-    if (WIFSIGNALED(status))
-      raise(WTERMSIG(status));
-    _exit(1);
-  }
-
-  // Child
-  close(pipefd[0]);
-
-  return [=]() {
-    char buf[] = {1};
-    int n = write(pipefd[1], buf, 1);
-    assert(n == 1);
-  };
-}
-
-static std::string base64(u8 *data, u64 size) {
-  static const char chars[] =
-    "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+_";
-
-  std::ostringstream out;
-
-  auto encode = [&](u32 x) {
-    out << chars[x & 0b111111]
-        << chars[(x >> 6) & 0b111111]
-        << chars[(x >> 12) & 0b111111]
-        << chars[(x >> 18) & 0b111111];
-  };
-
-  i64 i = 0;
-  for (; i < size - 3; i += 3)
-    encode((data[i + 2] << 16) | (data[i + 1] << 8) | data[i]);
-
-  if (i == size - 1)
-    encode(data[i]);
-  else if (i == size - 2)
-    encode((data[i + 1] << 8) | data[i]);
-  return out.str();
-}
-
-static std::string compute_sha256(std::span<std::string_view> argv) {
-  SHA256_CTX sha;
-  SHA256_Init(&sha);
-
-  for (std::string_view arg : argv) {
-    if (arg != "-preload" && arg != "--preload") {
-      SHA256_Update(&sha, arg.data(), arg.size());
-      char buf[] = {0};
-      SHA256_Update(&sha, buf, 1);
-    }
-  }
-
-  u8 digest[SHA256_SIZE];
-  SHA256_Final(digest, &sha);
-  return base64(digest, SHA256_SIZE);
-}
-
-template <typename E>
-static void send_fd(Context<E> &ctx, i64 conn, i64 fd) {
-  struct iovec iov;
-  char dummy = '1';
-  iov.iov_base = &dummy;
-  iov.iov_len = 1;
-
-  struct msghdr msg = {};
-  msg.msg_iov = &iov;
-  msg.msg_iovlen = 1;
-
-  char buf[CMSG_SPACE(sizeof(int))];
-  msg.msg_control = buf;
-  msg.msg_controllen = CMSG_LEN(sizeof(int));
-
-  struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
-  cmsg->cmsg_level = SOL_SOCKET;
-  cmsg->cmsg_type = SCM_RIGHTS;
-  cmsg->cmsg_len = CMSG_LEN(sizeof(int));
-  *(int *)CMSG_DATA(cmsg) = fd;
-
-  if (sendmsg(conn, &msg, 0) == -1)
-    Fatal(ctx) << "sendmsg failed: " << errno_string();
-}
-
-template <typename E>
-static i64 recv_fd(Context<E> &ctx, i64 conn) {
-  struct iovec iov;
-  char buf[1];
-  iov.iov_base = buf;
-  iov.iov_len = sizeof(buf);
-
-  struct msghdr msg = {};
-  msg.msg_iov = &iov;
-  msg.msg_iovlen = 1;
-
-  char cmsgbuf[CMSG_SPACE(sizeof(int))];
-  msg.msg_control = (caddr_t)cmsgbuf;
-  msg.msg_controllen = sizeof(cmsgbuf);
-
-  i64 len = recvmsg(conn, &msg, 0);
-  if (len <= 0)
-    Fatal(ctx) << "recvmsg failed: " << errno_string();
-
-  struct cmsghdr *cmsg;
-  cmsg = CMSG_FIRSTHDR(&msg);
-  return *(int *)CMSG_DATA(cmsg);
-}
-
-template <typename E>
-void try_resume_daemon(Context<E> &ctx) {
-  i64 conn = socket(AF_UNIX, SOCK_STREAM, 0);
-  if (conn == -1)
-    Fatal(ctx) << "socket failed: " << errno_string();
-
-  std::string path = "/tmp/mold-" + compute_sha256(ctx.cmdline_args);
-
-  struct sockaddr_un name = {};
-  name.sun_family = AF_UNIX;
-  memcpy(name.sun_path, path.data(), path.size());
-
-  if (connect(conn, (struct sockaddr *)&name, sizeof(name)) != 0) {
-    close(conn);
-    return;
-  }
-
-  send_fd(ctx, conn, STDOUT_FILENO);
-  send_fd(ctx, conn, STDERR_FILENO);
-
-  char buf[1];
-  i64 r = read(conn, buf, 1);
-  close(conn);
-  if (r == 1)
-    exit(0);
-}
-
-template <typename E>
-void daemonize(Context<E> &ctx, std::function<void()> *wait_for_client,
-               std::function<void()> *on_complete) {
-  if (daemon(1, 0) == -1)
-    Fatal(ctx) << "daemon failed: " << errno_string();
-
-  i64 sock = socket(AF_UNIX, SOCK_STREAM, 0);
-  if (sock == -1)
-    Fatal(ctx) << "socket failed: " << errno_string();
-
-  socket_tmpfile =
-    strdup(("/tmp/mold-" + compute_sha256(ctx.cmdline_args)).c_str());
-
-  struct sockaddr_un name = {};
-  name.sun_family = AF_UNIX;
-  strcpy(name.sun_path, socket_tmpfile);
-
-  u32 orig_mask = umask(0177);
-
-  if (bind(sock, (struct sockaddr *)&name, sizeof(name)) == -1) {
-    if (errno != EADDRINUSE)
-      Fatal(ctx) << "bind failed: " << errno_string();
-
-    unlink(socket_tmpfile);
-    if (bind(sock, (struct sockaddr *)&name, sizeof(name)) == -1)
-      Fatal(ctx) << "bind failed: " << errno_string();
-  }
-
-  umask(orig_mask);
-
-  if (listen(sock, 0) == -1)
-    Fatal(ctx) << "listen failed: " << errno_string();
-
-  static i64 conn = -1;
-
-  *wait_for_client = [=, &ctx]() {
-    fd_set rfds;
-    FD_ZERO(&rfds);
-    FD_SET(sock, &rfds);
-
-    struct timeval tv;
-    tv.tv_sec = DAEMON_TIMEOUT;
-    tv.tv_usec = 0;
-
-    i64 res = select(sock + 1, &rfds, NULL, NULL, &tv);
-    if (res == -1)
-      Fatal(ctx) << "select failed: " << errno_string();
-
-    if (res == 0) {
-      std::cout << "timeout\n";
-      exit(0);
-    }
-
-    conn = accept(sock, NULL, NULL);
-    if (conn == -1)
-      Fatal(ctx) << "accept failed: " << errno_string();
-    unlink(socket_tmpfile);
-
-    dup2(recv_fd(ctx, conn), STDOUT_FILENO);
-    dup2(recv_fd(ctx, conn), STDERR_FILENO);
-  };
-
-  *on_complete = [=]() {
-    char buf[] = {1};
-    int n = write(conn, buf, 1);
-    assert(n == 1);
-  };
-}
-
-static std::string get_self_path() {
-  return std::filesystem::read_symlink("/proc/self/exe");
-}
-
-static bool is_regular_file(const std::string &path) {
-  struct stat st;
-  return !stat(path.c_str(), &st) && (st.st_mode & S_IFMT) == S_IFREG;
-}
-
-template <typename E>
-std::string find_dso(Context<E> &ctx, std::filesystem::path self) {
-  // Look for mold-wrapper.so from the same directory as the executable is.
-  std::filesystem::path path = self.parent_path() / "mold-wrapper.so";
-  std::error_code ec;
-  if (std::filesystem::is_regular_file(path, ec) && !ec)
-    return path;
-
-#ifdef LIBDIR
-  // If not found, search $(LIBDIR)/mold, which is /usr/local/lib/mold
-  // by default.
-  path = LIBDIR "/mold/mold-wrapper.so";
-  if (std::filesystem::is_regular_file(path, ec) && !ec)
-    return path;
-#endif
-
-  Fatal(ctx) << "mold-wrapper.so is missing";
-}
-
-template <typename E>
-[[noreturn]]
-void process_run_subcommand(Context<E> &ctx, int argc, char **argv) {
-  std::string_view arg1 = argv[1];
-  assert(arg1 == "-run" || arg1 == "--run");
-  if (!argv[2])
-    Fatal(ctx) << "-run: argument missing";
-
-  // Get the mold-wrapper.so path
-  std::string self = get_self_path();
-  std::string dso_path = find_dso(ctx, self);
-
-  // Set environment variables
-  putenv(strdup(("LD_PRELOAD=" + dso_path).c_str()));
-  putenv(strdup(("MOLD_PATH=" + self).c_str()));
-
-  // If ld, ld.lld or ld.gold is specified, run mold itself
-  if (std::string cmd = filepath(argv[2]).filename();
-      cmd == "ld" || cmd == "ld.lld" || cmd == "ld.gold") {
-    std::vector<char *> args;
-    args.push_back(argv[0]);
-    args.insert(args.end(), argv + 3, argv + argc);
-    args.push_back(nullptr);
-    execv(self.c_str(), args.data());
-    Fatal(ctx) << "mold -run failed: " << self << ": " << errno_string();
-  }
-
-  // Execute a given command
-  execvp(argv[2], argv + 2);
-  Fatal(ctx) << "mold -run failed: " << argv[2] << ": " << errno_string();
-}
-
-#define INSTANTIATE(E)                                                  \
-  template void try_resume_daemon(Context<E> &);                        \
-  template void daemonize(Context<E> &, std::function<void()> *,        \
-                          std::function<void()> *);                     \
-  template void process_run_subcommand(Context<E> &, int, char **)
-
-INSTANTIATE(X86_64);
-INSTANTIATE(I386);
-INSTANTIATE(ARM64);
-
-} // namespace mold::elf

filepath.cc

@@ -1,27 +0,0 @@
-#include "mold.h"
-
-#include <filesystem>
-#include <sys/stat.h>
-
-namespace mold {
-
-std::string get_realpath(std::string_view path) {
-  std::error_code ec;
-  std::string ret = std::filesystem::canonical(path, ec);
-  return ec ? std::string(path) : ret;
-}
-
-// Removes redundant '/..' or '/.' from a given path.
-// The transformation is done purely by lexical processing.
-// This function does not access file system.
-std::string path_clean(std::string_view path) {
-  return filepath(path).lexically_normal();
-}
-
-std::filesystem::path to_abs_path(std::filesystem::path path) {
-  if (path.is_absolute())
-    return path;
-  return (std::filesystem::current_path() / path).lexically_normal();
-}
-
-} // namespace mold

filetype.h

@@ -1,96 +0,0 @@
-#pragma once
-
-#include "mold.h"
-
-namespace mold {
-
-enum class FileType {
-  UNKNOWN,
-  EMPTY,
-  ELF_OBJ,
-  ELF_DSO,
-  MACH_OBJ,
-  MACH_DYLIB,
-  MACH_UNIVERSAL,
-  AR,
-  THIN_AR,
-  TAPI,
-  TEXT,
-  LLVM_BITCODE,
-};
-
-template <typename C>
-bool is_text_file(MappedFile<C> *mf) {
-  u8 *data = mf->data;
-  return mf->size >= 4 && isprint(data[0]) && isprint(data[1]) &&
-         isprint(data[2]) && isprint(data[3]);
-}
-
-template <typename C>
-FileType get_file_type(MappedFile<C> *mf) {
-  std::string_view data = mf->get_contents();
-
-  if (data.empty())
-    return FileType::EMPTY;
-
-  if (data.starts_with("\177ELF")) {
-    switch (*(u16 *)(data.data() + 16)) {
-    case 1: // ET_REL
-      return FileType::ELF_OBJ;
-    case 3: // ET_DYN
-      return FileType::ELF_DSO;
-    }
-    return FileType::UNKNOWN;
-  }
-
-  if (data.starts_with("\xcf\xfa\xed\xfe")) {
-    switch (*(u32 *)(data.data() + 12)) {
-    case 1: // MH_OBJECT
-      return FileType::MACH_OBJ;
-    case 6: // MH_DYLIB
-      return FileType::MACH_DYLIB;
-    }
-    return FileType::UNKNOWN;
-  }
-
-  if (data.starts_with("!<arch>\n"))
-    return FileType::AR;
-  if (data.starts_with("!<thin>\n"))
-    return FileType::THIN_AR;
-  if (data.starts_with("--- !tapi-tbd"))
-    return FileType::TAPI;
-  if (data.starts_with("\xca\xfe\xba\xbe"))
-    return FileType::MACH_UNIVERSAL;
-  if (is_text_file(mf))
-    return FileType::TEXT;
-  if (data.starts_with("\xde\xc0\x17\x0b"))
-    return FileType::LLVM_BITCODE;
-  if (data.starts_with("BC\xc0\xde"))
-    return FileType::LLVM_BITCODE;
-  return FileType::UNKNOWN;
-}
-
-inline std::string filetype_to_string(FileType type) {
-  switch (type) {
-  case FileType::UNKNOWN: return "UNKNOWN";
-  case FileType::EMPTY: return "EMPTY";
-  case FileType::ELF_OBJ: return "ELF_OBJ";
-  case FileType::ELF_DSO: return "ELF_DSO";
-  case FileType::MACH_OBJ: return "MACH_OBJ";
-  case FileType::MACH_DYLIB: return "MACH_DYLIB";
-  case FileType::MACH_UNIVERSAL: return "MACH_UNIVERSAL";
-  case FileType::AR: return "AR";
-  case FileType::THIN_AR: return "THIN_AR";
-  case FileType::TAPI: return "TAPI";
-  case FileType::TEXT: return "TEXT";
-  case FileType::LLVM_BITCODE: return "LLVM_BITCODE";
-  }
-  return "UNKNOWN";
-}
-
-inline std::ostream &operator<<(std::ostream &out, FileType type) {
-  out << filetype_to_string(type);
-  return out;
-}
-
-} // namespace mold

install-build-deps.sh

@@ -0,0 +1,52 @@
+#!/bin/sh
+# This script installs binary packages needed to build mold.
+# Feel free to send me a PR if your OS is not on this list.
+
+set -e
+. /etc/os-release
+
+set -x
+
+case "$ID" in
+ubuntu | pop | linuxmint | debian | raspbian | neon)
+  apt-get update
+  apt-get install -y cmake gcc g++ clang gdb
+  ;;
+fedora | amzn | rhel | centos)
+  dnf install -y gcc-g++ cmake glibc-static libstdc++-static diffutils util-linux tar
+  ;;
+rocky)
+  dnf install -y gcc-g++ cmake diffutils util-linux
+  ;;
+opensuse-*)
+  zypper install -y make cmake gcc-c++ glibc-devel-static tar diffutils util-linux
+  ;;
+gentoo)
+  emerge-webrsync
+  FEATURES='getbinpkg binpkg-request-signature' emerge dev-build/cmake
+  ;;
+arch | archarm | artix | endeavouros)
+  pacman -Sy --needed --noconfirm base-devel cmake util-linux
+  ;;
+void)
+  xbps-install -Sy xbps bash make cmake gcc tar diffutils util-linux
+  ;;
+alpine)
+  apk update
+  apk add bash make linux-headers cmake gcc g++
+  ;;
+clear-linux-os)
+  swupd update
+  swupd bundle-add c-basic diffutils
+  ;;
+almalinux)
+  dnf install -y gcc-toolset-13-gcc-c++ gcc-toolset-13-libstdc++-devel cmake diffutils
+  ;;
+freebsd)
+  pkg update
+  pkg install -y cmake bash binutils gcc
+  ;;
+*)
+  echo "Error: don't know anything about build dependencies on $ID-$VERSION_ID"
+  exit 1
+esac

install-cross-tools.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+set -e
+. /etc/os-release
+
+set -x
+
+# This script install packages for -DMOLD_ENABLE_QEMU_TESTS=1
+# to enable cross-target tests.
+#
+# Feel free to send me a PR if your OS is not on this list.
+
+case "$ID" in
+ubuntu | pop | linuxmint | debian | raspbian)
+  apt-get install -y qemu-user {gcc,g++}-{i686,aarch64,riscv64,powerpc,powerpc64,powerpc64le,s390x,sparc64,m68k,sh4,arc}-linux-gnu {gcc,g++}-arm-linux-gnueabihf
+  ;;
+*)
+  echo "Error: don't know anything about build dependencies on $ID-$VERSION_ID"
+  exit 1
+esac

lib/archive-file.h

@@ -1,7 +1,31 @@
+// This file contains functions to read an archive file (.a file).
+// An archive file is just a bundle of object files. It's similar to
+// tar or zip, but the contents are not compressed.
+//
+// An archive file is either "regular" or "thin". A regular archive
+// contains object files directly, while a thin archive contains only
+// pathnames. In the latter case, actual file contents have to be read
+// from given pathnames. A regular archive is sometimes called "fat"
+// archive as opposed to "thin".
+//
+// If an archive file is given to the linker, the linker pulls out
+// object files that are needed to resolve undefined symbols. So,
+// bunding object files as an archive and giving that archive to the
+// linker has a different meaning than directly giving the same set of
+// object files to the linker. The former links only needed object
+// files, while the latter links all the given object files.
+//
+// Therefore, if you link libc.a for example, not all the libc
+// functions are linked to your binary. Instead, only object files
+// that provides functions and variables used in your program get
+// linked. To make this efficient, static library functions are
+// usually separated to each object file in an archive file. You can
+// see the contents of libc.a by running `ar t
+// /usr/lib/x86_64-linux-gnu/libc.a`.
+
 #pragma once
 
-#include "mold.h"
-#include "filetype.h"
+#include "common.h"
 
 namespace mold {
 
@@ -51,12 +75,12 @@ struct ArHdr {
   }
 };
 
-template <typename C>
-std::vector<MappedFile<C> *>
-read_thin_archive_members(C &ctx, MappedFile<C> *mf) {
+template <typename Context>
+std::vector<MappedFile *>
+read_thin_archive_members(Context &ctx, MappedFile *mf) {
   u8 *begin = mf->data;
   u8 *data = begin + 8;
-  std::vector<MappedFile<C> *> vec;
+  std::vector<MappedFile *> vec;
   std::string_view strtab;
 
   while (data < begin + mf->size) {
@@ -91,19 +115,19 @@ read_thin_archive_members(C &ctx, MappedFile<C> *mf) {
       continue;
 
     std::string path = name.starts_with('/') ?
-      name : (filepath(mf->name).parent_path() / name).string();
-    vec.push_back(MappedFile<C>::must_open(ctx, path));
+      name : (path_dirname(mf->name) / name).string();
+    vec.push_back(must_open_file(ctx, path));
+    vec.back()->thin_parent = mf;
     data = body;
   }
   return vec;
 }
 
-template <typename C>
-std::vector<MappedFile<C> *>
-read_fat_archive_members(C &ctx, MappedFile<C> *mf) {
+template <typename Context>
+std::vector<MappedFile *> read_fat_archive_members(Context &ctx, MappedFile *mf) {
   u8 *begin = mf->data;
   u8 *data = begin + 8;
-  std::vector<MappedFile<C> *> vec;
+  std::vector<MappedFile *> vec;
   std::string_view strtab;
 
   while (begin + mf->size - data >= 2) {
@@ -137,17 +161,13 @@ read_fat_archive_members(C &ctx, MappedFile<C> *mf) {
   return vec;
 }
 
-template <typename C>
-std::vector<MappedFile<C> *>
-read_archive_members(C &ctx, MappedFile<C> *mf) {
-  switch (get_file_type(mf)) {
-  case FileType::AR:
+template <typename Context>
+std::vector<MappedFile *> read_archive_members(Context &ctx, MappedFile *mf) {
+  std::string_view str = mf->get_contents();
+  if (str.starts_with("!<arch>\n"))
     return read_fat_archive_members(ctx, mf);
-  case FileType::THIN_AR:
-    return read_thin_archive_members(ctx, mf);
-  default:
-    unreachable();
-  }
+  assert(str.starts_with("!<thin>\n"));
+  return read_thin_archive_members(ctx, mf);
 }
 
 } // namespace mold

lib/common.h

@@ -0,0 +1,979 @@
+#pragma once
+
+#include "config.h"
+#include "integers.h"
+
+#include <array>
+#include <atomic>
+#include <bit>
+#include <bitset>
+#include <cassert>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <fcntl.h>
+#include <filesystem>
+#include <iostream>
+#include <mutex>
+#include <optional>
+#include <span>
+#include <sstream>
+#include <string>
+#include <string_view>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <tbb/concurrent_vector.h>
+#include <tbb/enumerable_thread_specific.h>
+#include <tbb/parallel_for.h>
+#include <vector>
+
+#ifdef _WIN32
+# include <io.h>
+#else
+# include <sys/mman.h>
+# include <unistd.h>
+#endif
+
+#define XXH_INLINE_ALL 1
+#include "../third-party/xxhash/xxhash.h"
+
+#ifdef NDEBUG
+# define unreachable() __builtin_unreachable()
+#else
+# define unreachable() assert(0 && "unreachable")
+#endif
+
+inline uint64_t hash_string(std::string_view str) {
+  return XXH3_64bits(str.data(), str.size());
+}
+
+class HashCmp {
+public:
+  static size_t hash(const std::string_view &k) {
+    return hash_string(k);
+  }
+
+  static bool equal(const std::string_view &k1, const std::string_view &k2) {
+    return k1 == k2;
+  }
+};
+
+namespace mold {
+
+namespace ranges = std::ranges;
+using namespace std::literals::string_literals;
+using namespace std::literals::string_view_literals;
+
+inline char *output_tmpfile;
+
+inline u8 *output_buffer_start = nullptr;
+inline u8 *output_buffer_end = nullptr;
+
+extern std::string mold_git_hash;
+
+std::string errno_string();
+void cleanup();
+void install_signal_handler();
+
+inline u64 combine_hash(u64 a, u64 b) {
+  return a ^ (b + 0x9e3779b9 + (a << 6) + (a >> 2));
+}
+
+//
+// Error output
+//
+
+// Some C++ stdlibs don't support std::osyncstream even though
+// it's is in the C++20 standard. So we implement it ourselves.
+class SyncStream {
+public:
+  SyncStream(std::ostream &out) : out(out) {}
+
+  ~SyncStream() {
+    emit();
+  }
+
+  template <typename T> SyncStream &operator<<(T &&val) {
+    ss << std::forward<T>(val);
+    return *this;
+  }
+
+  void emit() {
+    if (emitted)
+      return;
+
+    std::scoped_lock lock(mu);
+    out << ss.str() << '\n';
+    emitted = true;
+  }
+
+private:
+  std::ostream &out;
+  std::stringstream ss;
+  bool emitted = false;
+  static inline std::mutex mu;
+};
+
+template <typename Context>
+class Out {
+public:
+  Out(Context &ctx) {}
+
+  template <typename T> Out &operator<<(T &&val) {
+    out << std::forward<T>(val);
+    return *this;
+  }
+
+private:
+  SyncStream out{std::cout};
+};
+
+static std::string_view fatal_mono = "mold: fatal: ";
+static std::string_view fatal_color = "mold: \033[0;1;31mfatal:\033[0m ";
+static std::string_view error_mono = "mold: error: ";
+static std::string_view error_color = "mold: \033[0;1;31merror:\033[0m ";
+static std::string_view warning_mono = "mold: warning: ";
+static std::string_view warning_color = "mold: \033[0;1;35mwarning:\033[0m ";
+
+template <typename Context>
+class Fatal {
+public:
+  Fatal(Context &ctx) {
+    out << (ctx.arg.color_diagnostics ? fatal_color : fatal_mono);
+  }
+
+  [[noreturn]] ~Fatal() {
+    out.emit();
+    cleanup();
+    _exit(1);
+  }
+
+  template <typename T> Fatal &operator<<(T &&val) {
+    out << std::forward<T>(val);
+    return *this;
+  }
+
+private:
+  SyncStream out{std::cerr};
+};
+
+template <typename Context>
+class Error {
+public:
+  Error(Context &ctx) {
+    if (ctx.arg.noinhibit_exec) {
+      out << (ctx.arg.color_diagnostics ? warning_color : warning_mono);
+    } else {
+      out << (ctx.arg.color_diagnostics ? error_color : error_mono);
+      ctx.has_error = true;
+    }
+  }
+
+  template <typename T> Error &operator<<(T &&val) {
+    out << std::forward<T>(val);
+    return *this;
+  }
+
+private:
+  SyncStream out{std::cerr};
+};
+
+template <typename Context>
+class Warn {
+public:
+  Warn(Context &ctx) {
+    if (ctx.arg.suppress_warnings)
+      return;
+
+    out.emplace(std::cerr);
+
+    if (ctx.arg.fatal_warnings) {
+      *out << (ctx.arg.color_diagnostics ? error_color : error_mono);
+      ctx.has_error = true;
+    } else {
+      *out << (ctx.arg.color_diagnostics ? warning_color : warning_mono);
+    }
+  }
+
+  template <typename T> Warn &operator<<(T &&val) {
+    if (out)
+      *out << std::forward<T>(val);
+    return *this;
+  }
+
+private:
+  std::optional<SyncStream> out;
+};
+
+//
+// Atomics
+//
+
+// This is the same as std::atomic except that the default memory
+// order is relaxed instead of sequential consistency.
+template <typename T>
+struct Atomic : std::atomic<T> {
+  static constexpr std::memory_order relaxed = std::memory_order_relaxed;
+
+  using std::atomic<T>::atomic;
+
+  Atomic(const Atomic<T> &other) : std::atomic<T>(other.load()) {}
+
+  Atomic<T> &operator=(const Atomic<T> &other) {
+    store(other.load());
+    return *this;
+  }
+
+  void operator=(T val) { store(val); }
+  operator T() const { return load(); }
+
+  void store(T val, std::memory_order order = relaxed) {
+    std::atomic<T>::store(val, order);
+  }
+
+  T load(std::memory_order order = relaxed) const {
+    return std::atomic<T>::load(order);
+  }
+
+  T exchange(T val) { return std::atomic<T>::exchange(val, relaxed); }
+  T operator|=(T val) { return std::atomic<T>::fetch_or(val, relaxed); }
+  T operator++() { return std::atomic<T>::fetch_add(1, relaxed) + 1; }
+  T operator--() { return std::atomic<T>::fetch_sub(1, relaxed) - 1; }
+  T operator++(int) { return std::atomic<T>::fetch_add(1, relaxed); }
+  T operator--(int) { return std::atomic<T>::fetch_sub(1, relaxed); }
+
+  bool test_and_set() {
+    // A relaxed load + branch (assuming miss) takes only around 20 cycles,
+    // while an atomic RMW can easily take hundreds on x86. We note that it's
+    // common that another thread beat us in marking, so doing an optimistic
+    // early test tends to improve performance in the ~20% ballpark.
+    return load() || exchange(true);
+  }
+};
+
+//
+// mimalloc.cc
+//
+
+void set_mimalloc_options();
+
+//
+// perf.cc
+//
+
+// Counter is used to collect statistics numbers.
+class Counter {
+public:
+  Counter(std::string_view name, i64 value = 0) : name(name), values(value) {
+    static std::mutex mu;
+    std::scoped_lock lock(mu);
+    instances.push_back(this);
+  }
+
+  Counter &operator++(int) {
+    if (enabled) [[unlikely]]
+      values.local()++;
+    return *this;
+  }
+
+  Counter &operator+=(int delta) {
+    if (enabled) [[unlikely]]
+      values.local() += delta;
+    return *this;
+  }
+
+  static void print();
+
+  static inline bool enabled = false;
+
+private:
+  i64 get_value();
+
+  std::string_view name;
+  tbb::enumerable_thread_specific<i64> values;
+
+  static inline std::vector<Counter *> instances;
+};
+
+// Timer and TimeRecord records elapsed time (wall clock time)
+// used by each pass of the linker.
+struct TimerRecord {
+  TimerRecord(std::string name, TimerRecord *parent = nullptr);
+  void stop();
+
+  std::string name;
+  TimerRecord *parent;
+  tbb::concurrent_vector<TimerRecord *> children;
+  i64 start;
+  i64 end;
+  i64 user;
+  i64 sys;
+  bool stopped = false;
+};
+
+void
+print_timer_records(tbb::concurrent_vector<std::unique_ptr<TimerRecord>> &);
+
+template <typename Context>
+class Timer {
+public:
+  Timer(Context &ctx, std::string name, Timer *parent = nullptr) {
+    record = new TimerRecord(name, parent ? parent->record : nullptr);
+    ctx.timer_records.emplace_back(record);
+  }
+
+  Timer(const Timer &) = delete;
+
+  ~Timer() {
+    record->stop();
+  }
+
+  void stop() {
+    record->stop();
+  }
+
+private:
+  TimerRecord *record;
+};
+
+//
+// Utility functions
+//
+
+// Some C++ libraries haven't implemented std::has_single_bit yet.
+inline bool has_single_bit(u64 val) {
+  return std::popcount(val) == 1;
+}
+
+// Some C++ libraries haven't implemented std::bit_ceil yet.
+inline u64 bit_ceil(u64 val) {
+  if (has_single_bit(val))
+    return val;
+  return 1LL << (64 - std::countl_zero(val));
+}
+
+inline u64 align_to(u64 val, u64 align) {
+  if (align == 0)
+    return val;
+  assert(has_single_bit(align));
+  return (val + align - 1) & ~(align - 1);
+}
+
+inline u64 align_down(u64 val, u64 align) {
+  assert(has_single_bit(align));
+  return val & ~(align - 1);
+}
+
+inline u64 bit(u64 val, i64 pos) {
+  return (val >> pos) & 1;
+};
+
+// Returns [hi:lo] bits of val.
+inline u64 bits(u64 val, u64 hi, u64 lo) {
+  return (val >> lo) & ((1LL << (hi - lo + 1)) - 1);
+}
+
+// Cast val to a signed N bit integer.
+// For example, sign_extend(x, 32) == (i32)x for any integer x.
+inline i64 sign_extend(u64 val, i64 n) {
+  return (i64)(val << (64 - n)) >> (64 - n);
+}
+
+inline bool is_int(u64 val, i64 n) {
+  return sign_extend(val, n) == val;
+}
+
+template <typename T, typename Compare = std::less<T>>
+void update_minimum(std::atomic<T> &atomic, u64 new_val, Compare cmp = {}) {
+  T old_val = atomic.load(std::memory_order_relaxed);
+  while (cmp(new_val, old_val) &&
+         !atomic.compare_exchange_weak(old_val, new_val,
+                                       std::memory_order_relaxed));
+}
+
+template <typename T, typename Compare = std::less<T>>
+void update_maximum(std::atomic<T> &atomic, u64 new_val, Compare cmp = {}) {
+  T old_val = atomic.load(std::memory_order_relaxed);
+  while (cmp(old_val, new_val) &&
+         !atomic.compare_exchange_weak(old_val, new_val,
+                                       std::memory_order_relaxed));
+}
+
+template <typename T>
+inline void append(std::vector<T> &x, const auto &y) {
+  x.insert(x.end(), y.begin(), y.end());
+}
+
+template <typename T>
+inline std::vector<T> flatten(std::vector<std::vector<T>> &vec) {
+  i64 size = 0;
+  for (std::vector<T> &v : vec)
+    size += v.size();
+
+  std::vector<T> ret;
+  ret.reserve(size);
+  for (std::vector<T> &v : vec)
+    append(ret, v);
+  return ret;
+}
+
+template <typename T>
+inline void remove_duplicates(std::vector<T> &vec) {
+  vec.erase(std::unique(vec.begin(), vec.end()), vec.end());
+}
+
+inline i64 write_string(void *buf, std::string_view str) {
+  memcpy(buf, str.data(), str.size());
+  *((u8 *)buf + str.size()) = '\0';
+  return str.size() + 1;
+}
+
+template <typename T>
+inline void write_vector(void *buf, const std::vector<T> &vec) {
+  if (!vec.empty())
+    memcpy(buf, vec.data(), vec.size() * sizeof(T));
+}
+
+inline void encode_uleb(std::vector<u8> &vec, u64 val) {
+  do {
+    u8 byte = val & 0x7f;
+    val >>= 7;
+    vec.push_back(val ? (byte | 0x80) : byte);
+  } while (val);
+}
+
+inline void encode_sleb(std::vector<u8> &vec, i64 val) {
+  for (;;) {
+    u8 byte = val & 0x7f;
+    val >>= 7;
+
+    bool neg = (byte & 0x40);
+    if ((val == 0 && !neg) || (val == -1 && neg)) {
+      vec.push_back(byte);
+      break;
+    }
+    vec.push_back(byte | 0x80);
+  }
+}
+
+inline i64 write_uleb(u8 *buf, u64 val) {
+  i64 i = 0;
+  do {
+    u8 byte = val & 0x7f;
+    val >>= 7;
+    buf[i++] = val ? (byte | 0x80) : byte;
+  } while (val);
+  return i;
+}
+
+inline u64 read_uleb(u8 **buf) {
+  u64 val = 0;
+  u8 shift = 0;
+  u8 byte;
+  do {
+    byte = *(*buf)++;
+    val |= (byte & 0x7f) << shift;
+    shift += 7;
+  } while (byte & 0x80);
+  return val;
+}
+
+inline u64 read_uleb(u8 *buf) {
+  u8 *tmp = buf;
+  return read_uleb(&tmp);
+}
+
+inline i64 read_sleb(u8 **buf) {
+  u64 val = 0;
+  u8 shift = 0;
+  u8 byte;
+  do {
+    byte = *(*buf)++;
+    val |= (byte & 0x7f) << shift;
+    shift += 7;
+  } while (byte & 0x80);
+  return sign_extend(val, shift);
+}
+
+inline i64 read_sleb(u8 *buf) {
+  u8 *tmp = buf;
+  return read_sleb(&tmp);
+}
+
+inline u64 read_uleb(std::string_view *str) {
+  u8 *start = (u8 *)str->data();
+  u8 *ptr = start;
+  u64 val = read_uleb(&ptr);
+  *str = str->substr(ptr - start);
+  return val;
+}
+
+inline u64 read_uleb(std::string_view str) {
+  std::string_view tmp = str;
+  return read_uleb(&tmp);
+}
+
+inline i64 uleb_size(u64 val) {
+  for (int i = 1; i < 9; i++)
+    if (val < (1LL << (7 * i)))
+      return i;
+  return 9;
+}
+
+inline void overwrite_uleb(u8 *loc, u64 val) {
+  while (*loc & 0b1000'0000) {
+    *loc++ = 0b1000'0000 | (val & 0b0111'1111);
+    val >>= 7;
+  }
+  *loc = val & 0b0111'1111;
+}
+
+static inline void pause() {
+#if defined(__x86_64__)
+  asm volatile("pause");
+#elif defined(__aarch64__)
+  asm volatile("yield");
+#elif defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_8A__)
+  asm volatile("yield");
+#endif
+}
+
+//
+// Concurrent Map
+//
+
+// This is an implementation of a fast concurrent hash map. Unlike
+// ordinary hash tables, this impl just aborts if it becomes full.
+// So you need to give a correct estimation of the final size before
+// using it. We use this hash map to uniquify pieces of data in
+// mergeable sections.
+//
+// We've implemented this ourselves because the performance of
+// conrurent hash map is critical for our linker.
+template <typename T>
+class ConcurrentMap {
+public:
+  ConcurrentMap() = default;
+
+  ConcurrentMap(i64 nbuckets) {
+    resize(nbuckets);
+  }
+
+  ~ConcurrentMap() {
+    if (entries) {
+#ifdef _WIN32
+      _aligned_free(entries);
+#else
+      munmap(entries, sizeof(Entry) * nbuckets);
+#endif
+    }
+  }
+
+  // In order to avoid unnecessary cache-line false sharing, we want
+  // to make this object to be aligned to a reasonably large
+  // power-of-two address.
+  struct alignas(32) Entry {
+    Atomic<const char *> key;
+    u32 keylen;
+    T value;
+  };
+
+  void resize(i64 nbuckets) {
+    assert(!entries);
+    this->nbuckets = std::max<i64>(MIN_NBUCKETS, bit_ceil(nbuckets));
+    i64 bufsize = sizeof(Entry) * this->nbuckets;
+
+    // Allocate a zero-initialized buffer. We use mmap() if available
+    // because it's faster than malloc() and memset().
+#ifdef _WIN32
+    entries = (Entry *)_aligned_malloc(bufsize, alignof(Entry));
+    memset((void *)entries, 0, bufsize);
+#else
+    entries = (Entry *)mmap(nullptr, bufsize, PROT_READ | PROT_WRITE,
+                            MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
+#endif
+  }
+
+  std::pair<T *, bool> insert(std::string_view key, u64 hash, const T &val) {
+    assert(has_single_bit(nbuckets));
+
+    u64 begin = hash & (nbuckets - 1);
+    u64 mask = nbuckets / NUM_SHARDS - 1;
+
+    for (i64 i = 0; i < MAX_RETRY; i++) {
+      u64 idx = (begin & ~mask) | ((begin + i) & mask);
+      Entry &ent = entries[idx];
+
+      // It seems avoiding compare-and-swap is faster overall at least
+      // on my Zen4 machine, so do it.
+      if (const char *ptr = ent.key.load(std::memory_order_acquire);
+          ptr != nullptr && ptr != (char *)-1) {
+        if (key == std::string_view(ptr, ent.keylen))
+          return {&ent.value, false};
+        continue;
+      }
+
+      // Otherwise, use CAS to atomically claim the ownership of the slot.
+      const char *ptr = nullptr;
+      bool claimed = ent.key.compare_exchange_strong(ptr, (char *)-1,
+                                                     std::memory_order_acquire);
+
+      // If we successfully claimed the ownership of the slot,
+      // copy values to it.
+      if (claimed) {
+        new (&ent.value) T(val);
+        ent.keylen = key.size();
+        ent.key.store(key.data(), std::memory_order_release);
+        return {&ent.value, true};
+      }
+
+      // If someone is copying values to the slot, do busy wait.
+      while (ptr == (char *)-1) {
+        pause();
+        ptr = ent.key.load(std::memory_order_acquire);
+      }
+
+      // If the same key is already present, this is the slot we are
+      // looking for.
+      if (key == std::string_view(ptr, ent.keylen))
+        return {&ent.value, false};
+    }
+
+    std::cerr << "ConcurrentMap is full\n";
+    abort();
+  }
+
+  i64 get_idx(T *value) const {
+    uintptr_t addr = (uintptr_t)value - (uintptr_t)value % sizeof(Entry);
+    return (Entry *)addr - entries;
+  }
+
+  // Return a list of map entries sorted in a deterministic order.
+  std::vector<Entry *> get_sorted_entries(i64 shard_idx) {
+    if (nbuckets == 0)
+      return {};
+
+    i64 shard_size = nbuckets / NUM_SHARDS;
+    i64 begin = shard_idx * shard_size;
+    i64 end = begin + shard_size;
+
+    i64 sz = 0;
+    for (i64 i = begin; i < end; i++)
+      if (entries[i].key)
+        sz++;
+
+    std::vector<Entry *> vec;
+    vec.reserve(sz);
+
+    // Since the shard is circular, we need to handle the last entries
+    // as if they were next to the first entries.
+    while (entries[end - 1].key)
+      vec.push_back(entries + --end);
+
+    // Find entries contiguous in the buckets and sort them.
+    i64 last = 0;
+    for (i64 i = begin; i < end;) {
+      while (i < end && entries[i].key)
+        vec.push_back(entries + i++);
+
+      std::sort(vec.begin() + last, vec.end(), [](Entry *a, Entry *b) {
+        if (a->keylen != b->keylen)
+          return a->keylen < b->keylen;
+        return memcmp(a->key, b->key, a->keylen) < 0;
+      });
+
+      last = vec.size();
+
+      while (i < end && !entries[i].key)
+        i++;
+    }
+    return vec;
+  }
+
+  std::vector<Entry *> get_sorted_entries_all() {
+    std::vector<std::vector<Entry *>> vec(NUM_SHARDS);
+    tbb::parallel_for((i64)0, NUM_SHARDS, [&](i64 i) {
+      vec[i] = get_sorted_entries(i);
+    });
+    return flatten(vec);
+  }
+
+  static constexpr i64 MIN_NBUCKETS = 4096;
+  static constexpr i64 NUM_SHARDS = 16;
+  static constexpr i64 MAX_RETRY = 256;
+
+  Entry *entries = nullptr;
+  u64 nbuckets = 0;
+};
+
+//
+// random.cc
+//
+
+void get_random_bytes(u8 *buf, i64 size);
+
+//
+// hyperloglog.cc
+//
+
+class HyperLogLog {
+public:
+  void insert(u64 hash) {
+    update_maximum(buckets[hash & (NBUCKETS - 1)], std::countl_zero(hash) + 1);
+  }
+
+  i64 get_cardinality() const;
+
+  void merge(const HyperLogLog &other) {
+    for (i64 i = 0; i < NBUCKETS; i++)
+      update_maximum(buckets[i], other.buckets[i]);
+  }
+
+private:
+  static constexpr i64 NBUCKETS = 2048;
+  static constexpr double ALPHA = 0.79402;
+
+  Atomic<u8> buckets[NBUCKETS];
+};
+
+//
+// glob.cc
+//
+
+class Glob {
+  typedef enum { STRING, STAR, QUESTION, BRACKET } Kind;
+
+  struct Element {
+    Element(Kind k) : kind(k) {}
+    Kind kind;
+    std::string str;
+    std::bitset<256> bitset;
+  };
+
+public:
+  static std::optional<Glob> compile(std::string_view pat);
+  bool match(std::string_view str);
+
+private:
+  Glob(std::vector<Element> &&vec) : elements(vec) {}
+  static bool do_match(std::string_view str, std::span<Element> elements);
+
+  std::vector<Element> elements;
+};
+
+//
+// multi-glob.cc
+//
+
+class MultiGlob {
+public:
+  bool add(std::string_view pat, i64 val);
+  bool empty() const { return strings.empty(); }
+  std::optional<i64> find(std::string_view str);
+
+private:
+  struct TrieNode {
+    i64 value = -1;
+    TrieNode *suffix_link = nullptr;
+    std::unique_ptr<TrieNode> children[256];
+  };
+
+  void compile();
+  void fix_suffix_links(TrieNode &node);
+  void fix_values();
+  i64 find_aho_corasick(std::string_view str);
+
+  std::vector<std::string> strings;
+  std::unique_ptr<TrieNode> root;
+  std::vector<std::pair<Glob, i64>> globs;
+  std::once_flag once;
+  bool is_compiled = false;
+  bool prefix_match = false;
+};
+
+//
+// filepath.cc
+//
+
+inline std::filesystem::path path_dirname(std::string_view path) {
+  return std::filesystem::path(path).parent_path();
+}
+
+inline std::string path_filename(std::string_view path) {
+  return std::filesystem::path(path).filename().string();
+}
+
+inline std::string path_clean(std::string_view path) {
+  return std::filesystem::path(path).lexically_normal().string();
+}
+
+std::string get_self_path();
+
+//
+// demangle.cc
+//
+
+std::optional<std::string_view> demangle_cpp(std::string_view name);
+std::optional<std::string_view> demangle_rust(std::string_view name);
+
+//
+// jbos.cc
+//
+
+void acquire_global_lock();
+void release_global_lock();
+
+//
+// crc32.cc
+//
+
+u32 compute_crc32(u32 crc, u8 *buf, i64 len);
+std::vector<u8> crc32_solve(u32 current, u32 desired);
+
+//
+// compress.cc
+//
+
+class Compressor {
+public:
+  virtual void write_to(u8 *buf) = 0;
+  virtual ~Compressor() {}
+  i64 compressed_size = 0;
+};
+
+class ZlibCompressor : public Compressor {
+public:
+  ZlibCompressor(u8 *buf, i64 size);
+  void write_to(u8 *buf) override;
+
+private:
+  std::vector<std::vector<u8>> shards;
+  u64 checksum = 0;
+};
+
+class ZstdCompressor : public Compressor {
+public:
+  ZstdCompressor(u8 *buf, i64 size);
+  void write_to(u8 *buf) override;
+
+private:
+  std::vector<std::vector<u8>> shards;
+};
+
+//
+// tar.cc
+//
+
+// TarFile is a class to create a tar file.
+//
+// If you pass `--repro` to mold, mold collects all input files and
+// put them into `<output-file-path>.repro.tar`, so that it is easy to
+// run the same command with the same command line arguments.
+class TarWriter {
+public:
+  static std::unique_ptr<TarWriter>
+  open(std::string output_path, std::string basedir);
+
+  ~TarWriter();
+  void append(std::string path, std::string_view data);
+
+private:
+  TarWriter(FILE *out, std::string basedir) : out(out), basedir(basedir) {}
+
+  FILE *out = nullptr;
+  std::string basedir;
+};
+
+//
+// Memory-mapped file
+//
+
+// MappedFile represents an mmap'ed input file.
+// mold uses mmap-IO only.
+class MappedFile {
+public:
+  ~MappedFile() { unmap(); }
+  void unmap();
+  void close_fd();
+  void reopen_fd(const std::string &path);
+
+  template <typename Context>
+  MappedFile *slice(Context &ctx, std::string name, u64 start, u64 size) {
+    MappedFile *mf = new MappedFile;
+    mf->name = name;
+    mf->data = data + start;
+    mf->size = size;
+    mf->parent = this;
+
+    ctx.mf_pool.emplace_back(mf);
+    return mf;
+  }
+
+  std::string_view get_contents() {
+    return std::string_view((char *)data, size);
+  }
+
+  i64 get_offset() const {
+    return parent ? (data - parent->data + parent->get_offset()) : 0;
+  }
+
+  // Returns a string that uniquely identify a file that is possibly
+  // in an archive.
+  std::string get_identifier() const {
+    if (parent) {
+      // We use the file offset within an archive as an identifier
+      // because archive members may have the same name.
+      return parent->name + ":" + std::to_string(get_offset());
+    }
+
+    if (thin_parent) {
+      // If this is a thin archive member, the filename part is
+      // guaranteed to be unique.
+      return thin_parent->name + ":" + name;
+    }
+    return name;
+  }
+
+  std::string name;
+  u8 *data = nullptr;
+  i64 size = 0;
+  bool given_fullpath = true;
+  MappedFile *parent = nullptr;
+  MappedFile *thin_parent = nullptr;
+
+  // For --dependency-file
+  bool is_dependency = true;
+
+#ifdef _WIN32
+  HANDLE fd = INVALID_HANDLE_VALUE;
+#else
+  int fd = -1;
+#endif
+};
+
+MappedFile *open_file_impl(const std::string &path, std::string &error);
+
+template <typename Context>
+MappedFile *open_file(Context &ctx, std::string path) {
+  if (path.starts_with('/') && !ctx.arg.chroot.empty())
+    path = ctx.arg.chroot + "/" + path_clean(path);
+
+  std::string error;
+  MappedFile *mf = open_file_impl(path, error);
+  if (!error.empty())
+    Fatal(ctx) << error;
+
+  if (mf)
+    ctx.mf_pool.emplace_back(mf);
+  return mf;
+}
+
+template <typename Context>
+MappedFile *must_open_file(Context &ctx, std::string path) {
+  MappedFile *mf = open_file(ctx, path);
+  if (!mf)
+    Fatal(ctx) << "cannot open " << path << ": " << errno_string();
+  return mf;
+}
+
+} // namespace mold

lib/compress.cc

@@ -0,0 +1,185 @@
+// This file implements a multi-threaded zlib and zstd compression
+// routine.
+//
+// zlib-compressed data can be merged just by concatenation as long as
+// each piece of data is flushed with Z_SYNC_FLUSH. In this file, we
+// split input data into multiple shards, compress them individually
+// and concatenate them. We then append a header, a trailer and a
+// checksum so that the concatenated data is valid zlib-format data.
+//
+// zstd-compressed data can be merged in the same way.
+//
+// Using threads to compress data has a downside. Since the dictionary
+// is reset on boundaries of shards, compression ratio is sacrificed
+// a little bit. However, if a shard size is large enough, that loss
+// is negligible in practice.
+
+#include "common.h"
+
+#include <tbb/parallel_for_each.h>
+#include <zlib.h>
+#include <zstd.h>
+
+#define CHECK(fn)                               \
+  do {                                          \
+    [[maybe_unused]] int r = (fn);              \
+    assert(r == Z_OK);                          \
+  } while (0)
+
+namespace mold {
+
+static constexpr i64 SHARD_SIZE = 1024 * 1024;
+
+static std::vector<std::string_view> split(std::string_view input) {
+  std::vector<std::string_view> shards;
+
+  while (input.size() >= SHARD_SIZE) {
+    shards.push_back(input.substr(0, SHARD_SIZE));
+    input = input.substr(SHARD_SIZE);
+  }
+  if (!input.empty())
+    shards.push_back(input);
+  return shards;
+}
+
+static std::vector<u8> zlib_compress(std::string_view input) {
+  // Initialize zlib stream. Since debug info is generally compressed
+  // pretty well with lower compression levels, we chose compression
+  // level 1.
+  z_stream strm;
+  strm.zalloc = Z_NULL;
+  strm.zfree = Z_NULL;
+  strm.opaque = Z_NULL;
+
+  CHECK(deflateInit2(&strm, 1, Z_DEFLATED, -15, 8, Z_DEFAULT_STRATEGY));
+
+  // Set an input buffer
+  strm.avail_in = input.size();
+  strm.next_in = (u8 *)input.data();
+
+  // Set an output buffer. deflateBound() returns an upper bound
+  // on the compression size. +16 for Z_SYNC_FLUSH.
+  std::vector<u8> buf(deflateBound(&strm, strm.avail_in) + 16);
+
+  // Compress data. It writes all compressed bytes except the last
+  // partial byte, so up to 7 bits can be held to be written to the
+  // buffer.
+  strm.avail_out = buf.size();
+  strm.next_out = buf.data();
+  CHECK(deflate(&strm, Z_BLOCK));
+
+  // This is a workaround for libbacktrace before 2022-04-06.
+  //
+  // Zlib is a bit stream, and what Z_SYNC_FLUSH does is to write a
+  // three bit value indicating the start of an uncompressed data
+  // block followed by four byte data 00 00 ff ff which indicates that
+  // the length of the block is zero. libbacktrace uses its own zlib
+  // inflate routine, and it had a bug that if that particular three
+  // bit value happens to end at a byte boundary, it accidentally
+  // skipped the next byte.
+  //
+  // In order to avoid triggering that bug, we should avoid calling
+  // deflate() with Z_SYNC_FLUSH if the current bit position is 5.
+  // If it's 5, we insert an empty block consisting of 10 bits so
+  // that the bit position is 7 in the next byte.
+  //
+  // https://github.com/ianlancetaylor/libbacktrace/pull/87
+  int nbits;
+  deflatePending(&strm, Z_NULL, &nbits);
+  if (nbits == 5)
+    CHECK(deflatePrime(&strm, 10, 2));
+  CHECK(deflate(&strm, Z_SYNC_FLUSH));
+
+  assert(strm.avail_out > 0);
+  buf.resize(buf.size() - strm.avail_out);
+  buf.shrink_to_fit();
+  deflateEnd(&strm);
+  return buf;
+}
+
+ZlibCompressor::ZlibCompressor(u8 *buf, i64 size) {
+  std::string_view input{(char *)buf, (size_t)size};
+  std::vector<std::string_view> inputs = split(input);
+  std::vector<u64> adlers(inputs.size());
+  shards.resize(inputs.size());
+
+  // Compress each shard
+  tbb::parallel_for((i64)0, (i64)inputs.size(), [&](i64 i) {
+    adlers[i] = adler32(1, (u8 *)inputs[i].data(), inputs[i].size());
+    shards[i] = zlib_compress(inputs[i]);
+  });
+
+  // Combine checksums
+  checksum = adlers[0];
+  for (i64 i = 1; i < inputs.size(); i++)
+    checksum = adler32_combine(checksum, adlers[i], inputs[i].size());
+
+  // Comput the total size
+  compressed_size = 8; // the header and the trailer
+  for (std::vector<u8> &shard : shards)
+    compressed_size += shard.size();
+}
+
+void ZlibCompressor::write_to(u8 *buf) {
+  // Write a zlib-format header
+  buf[0] = 0x78;
+  buf[1] = 0x9c;
+
+  // Copy compressed data
+  std::vector<i64> offsets(shards.size());
+  offsets[0] = 2; // +2 for header
+  for (i64 i = 1; i < shards.size(); i++)
+    offsets[i] = offsets[i - 1] + shards[i - 1].size();
+
+  tbb::parallel_for((i64)0, (i64)shards.size(), [&](i64 i) {
+    memcpy(buf + offsets[i], shards[i].data(), shards[i].size());
+  });
+
+  // Write a trailer
+  u8 *end = buf + compressed_size;
+  end[-6] = 3;
+  end[-5] = 0;
+
+  // Write a checksum
+  *(ub32 *)(end - 4) = checksum;
+}
+
+static std::vector<u8> zstd_compress(std::string_view input) {
+  std::vector<u8> buf(ZSTD_COMPRESSBOUND(input.size()));
+  constexpr int LEVEL = 3; // compression level; must be between 1 to 22
+
+  size_t sz = ZSTD_compress(buf.data(), buf.size(), input.data(), input.size(),
+                            LEVEL);
+  assert(!ZSTD_isError(sz));
+  buf.resize(sz);
+  buf.shrink_to_fit();
+  return buf;
+}
+
+ZstdCompressor::ZstdCompressor(u8 *buf, i64 size) {
+  std::string_view input{(char *)buf, (size_t)size};
+  std::vector<std::string_view> inputs = split(input);
+  shards.resize(inputs.size());
+
+  // Compress each shard
+  tbb::parallel_for((i64)0, (i64)inputs.size(), [&](i64 i) {
+    shards[i] = zstd_compress(inputs[i]);
+  });
+
+  compressed_size = 0;
+  for (std::vector<u8> &shard : shards)
+    compressed_size += shard.size();
+}
+
+void ZstdCompressor::write_to(u8 *buf) {
+  // Copy compressed data
+  std::vector<i64> offsets(shards.size());
+  for (i64 i = 1; i < shards.size(); i++)
+    offsets[i] = offsets[i - 1] + shards[i - 1].size();
+
+  tbb::parallel_for((i64)0, (i64)shards.size(), [&](i64 i) {
+    memcpy(buf + offsets[i], shards[i].data(), shards[i].size());
+  });
+}
+
+} // namespace mold

lib/config.h.in

@@ -0,0 +1,7 @@
+#define MOLD_VERSION "@mold_VERSION@"
+#define MOLD_LIBDIR "@CMAKE_INSTALL_FULL_LIBDIR@"
+
+#cmakedefine HAVE_MADVISE 1
+#cmakedefine HAVE_UNAME 1
+#cmakedefine MOLD_USE_MIMALLOC 1
+#cmakedefine MOLD_USE_SYSTEM_MIMALLOC 1

lib/crc32.cc

@@ -0,0 +1,60 @@
+#include "common.h"
+
+#include <tbb/parallel_for_each.h>
+#include <zlib.h>
+
+namespace mold {
+
+// This function "forges" a CRC. That is, given the current and a desired
+// CRC32 value, crc32_solve() returns a binary blob to add to the end of
+// the original data to yield the desired CRC. Trailing garbage is ignored
+// by many bianry file formats, so you can create a file with a desired
+// CRC using crc32_solve(). We need it for --separate-debug-file.
+std::vector<u8> crc32_solve(u32 current, u32 desired) {
+  constexpr u32 poly = 0xedb88320;
+  u32 x = ~desired;
+
+  // Each iteration computes x = (x * x^-1) mod poly.
+  for (i64 i = 0; i < 32; i++) {
+    x = std::rotl(x, 1);
+    x ^= (x & 1) * (poly << 1);
+  }
+
+  x ^= ~current;
+
+  std::vector<u8> out(4);
+  out[0] = x;
+  out[1] = x >> 8;
+  out[2] = x >> 16;
+  out[3] = x >> 24;
+  return out;
+}
+
+// Compute a CRC for given data in parallel
+u32 compute_crc32(u32 crc, u8 *buf, i64 len) {
+  struct Shard {
+    u8 *buf;
+    i64 len;
+    u32 crc;
+  };
+
+  constexpr i64 shard_size = 1024 * 1024; // 1 MiB
+  std::vector<Shard> shards;
+
+  while (len > 0) {
+    i64 sz = std::min(len, shard_size);
+    shards.push_back({buf, sz, 0});
+    buf += sz;
+    len -= sz;
+  }
+
+  tbb::parallel_for_each(shards, [](Shard &shard) {
+    shard.crc = crc32(0, shard.buf, shard.len);
+  });
+
+  for (Shard &shard : shards)
+    crc = crc32_combine(crc, shard.crc, shard.len);
+  return crc;
+}
+
+} // namespace mold

lib/demangle.cc

@@ -0,0 +1,42 @@
+#include "common.h"
+
+#include <cstdlib>
+
+#ifndef _WIN32
+#include <cxxabi.h>
+#endif
+
+#include "../third-party/rust-demangle/rust-demangle.h"
+
+namespace mold {
+
+std::optional<std::string_view> demangle_cpp(std::string_view name) {
+  static thread_local char *buf;
+  static thread_local size_t buflen;
+
+  // TODO(cwasser): Actually demangle Symbols on Windows using e.g.
+  // `UnDecorateSymbolName` from Dbghelp, maybe even Itanium symbols?
+#ifndef _WIN32
+  if (name.starts_with("_Z")) {
+    int status;
+    char *p = abi::__cxa_demangle(std::string(name).c_str(), buf, &buflen, &status);
+    if (status == 0) {
+      buf = p;
+      return p;
+    }
+  }
+#endif
+
+  return {};
+}
+
+std::optional<std::string_view> demangle_rust(std::string_view name) {
+  static thread_local char *buf;
+  free(buf);
+  buf = rust_demangle(std::string(name).c_str(), 0);
+  if (buf)
+    return buf;
+  return {};
+}
+
+} // namespace mold

lib/filepath.cc

@@ -0,0 +1,42 @@
+#include "common.h"
+
+#include <filesystem>
+#include <sys/stat.h>
+
+#ifdef __APPLE__
+# include <mach-o/dyld.h>
+#endif
+
+#ifdef __FreeBSD__
+# include <sys/sysctl.h>
+#endif
+
+namespace mold {
+
+// Returns the path of the mold executable itself
+std::string get_self_path() {
+#if __APPLE__ || _WIN32
+  fprintf(stderr, "mold: get_self_path is not supported");
+  exit(1);
+#elif __FreeBSD__
+  // /proc may not be mounted on FreeBSD. The proper way to get the
+  // current executable's path is to use sysctl(2).
+  int mib[4];
+  mib[0] = CTL_KERN;
+  mib[1] = KERN_PROC;
+  mib[2] = KERN_PROC_PATHNAME;
+  mib[3] = -1;
+
+  size_t size;
+  sysctl(mib, 4, NULL, &size, NULL, 0);
+
+  std::string path;
+  path.resize(size);
+  sysctl(mib, 4, path.data(), &size, NULL, 0);
+  return path;
+#else
+  return std::filesystem::read_symlink("/proc/self/exe").string();
+#endif
+}
+
+} // namespace mold

lib/gentoo-test.sh

@@ -0,0 +1,69 @@
+#!/bin/bash
+#
+# This test script takes a Gentoo package name and tries to build it
+# using mold in a Podman environment. We chose Gentoo Linux as a test
+# target, because its source-based package allows us to build programs
+# locally and run their test suites without any hassle.
+#
+# You can get a complete list of Gentoo packages availalbe for testing
+# with the following command:
+#
+# podman run --rm mold-gentoo emerge --color n -s '' | \
+#   perl -ne 'next unless m!^\*\s+(\S+/\S+)!; print "$1\n"'
+
+package="$1"
+
+if [ "$package" = "" ]; then
+  echo "Usage: $0 gentoo-package-name"
+  exit 1
+fi
+
+set -x
+
+# Create a Podman image
+if ! podman image ls mold-gentoo | grep -q mold-gentoo; then
+  set -e
+  cat <<EOF | podman build -t mold-gentoo -
+FROM docker.io/gentoo/stage3
+RUN emerge-webrsync
+RUN echo 'USE="X ssl elogind -systemd corefonts truetype jpeg jpeg2k tiff zstd static-libs binary -perl"' >> /etc/portage/make.conf && \
+    echo 'ACCEPT_KEYWORDS="~amd64"' >> /etc/portage/make.conf && \
+    echo 'ACCEPT_LICENSE="* -@EULA"' >> /etc/portage/make.conf && \
+    echo 'FEATURES="\${FEATURE} noclean nostrip ccache -ipc-sandbox -network-sandbox -pid-sandbox -sandbox"' >> /etc/portage/make.conf && \
+    echo 'CCACHE_DIR="/ccache"' >> /etc/portage/make.conf && \
+    FEATURES='getbinpkg binpkg-request-signature' emerge gdb lld llvm-core/clang vim emacs strace ccache xeyes dev-build/cmake dev-vcs/git && \
+    rm -rf /var/tmp/portage
+EOF
+  set +e
+fi
+
+git_hash=$(./dist/mold --version | perl -ne '/\((\w+)/; print $1;')
+
+if [ "$package" = dev-libs/concurrencykit ]; then
+  echo "Skipping known broken package: $package"
+  exit 0
+fi
+
+# Build a given package in Podman
+cmd1='(cd /usr/bin; ln -sf /mold/dist/mold $(realpath ld))'
+cmd2="MAKEOPTS=-'j$(nproc) --load-average=100' emerge --onlydeps $package"
+cmd3="MAKEOPTS='-j$(nproc) --load-average=100' FEATURES=test emerge $package"
+filename=`echo "$package" | sed 's!/!_!g'`
+podman="podman run --rm --pids-limit=-1 --cap-add=SYS_PTRACE -v `pwd`:/mold:ro -v /var/cache/ccache-gentoo:/ccache mold-gentoo timeout -v -k 15s 3h"
+dir=gentoo/$git_hash
+
+mkdir -p "$dir"/success "$dir"/failure
+
+$podman nice -n 19 bash -c "$cmd1 && $cmd2 && $cmd3" >& "$dir"/"$filename".mold
+if [ $? = 0 ]; then
+  mv "$dir"/"$filename".mold "$dir"/success
+else
+  mv "$dir"/"$filename".mold "$dir"/failure
+fi
+
+$podman nice -n 19 bash -c "$cmd2 && $cmd3" >& "$dir"/"$filename".ld
+if [ $? = 0 ]; then
+  mv "$dir"/"$filename".ld "$dir"/success
+else
+  mv "$dir"/"$filename".ld "$dir"/failure
+fi

lib/glob.cc

@@ -1,10 +1,10 @@
-#include "mold.h"
+#include "common.h"
 
 #include <cstring>
 
-namespace mold::elf {
+namespace mold {
 
-std::optional<GlobPattern> GlobPattern::compile(std::string_view pat) {
+std::optional<Glob> Glob::compile(std::string_view pat) {
   std::vector<Element> vec;
 
   while (!pat.empty()) {
@@ -19,9 +19,8 @@ std::optional<GlobPattern> GlobPattern::compile(std::string_view pat) {
       // [$\]!]: $, ] or !
       // [a-czg-i]: a, b, c, z, g, h, or i
       // [^a-z]: Any character except lowercase letters
-      vec.push_back({BRACKET});
-      std::vector<bool> &bitmap = vec.back().bitmap;
-      bitmap.resize(256);
+      vec.emplace_back(BRACKET);
+      std::bitset<256> &bitset = vec.back().bitset;
 
       bool negate = false;
       if (!pat.empty() && pat[0] == '^') {
@@ -60,9 +59,9 @@ std::optional<GlobPattern> GlobPattern::compile(std::string_view pat) {
             return {};
 
           for (i64 i = start; i <= end; i++)
-            bitmap[i] = true;
+            bitset[i] = true;
         } else {
-          bitmap[(u8)pat[0]] = true;
+          bitset[(u8)pat[0]] = true;
           pat = pat.substr(1);
         }
       }
@@ -71,39 +70,46 @@ std::optional<GlobPattern> GlobPattern::compile(std::string_view pat) {
         return {};
 
       if (negate)
-        for (i64 i = 0; i < 256; i++)
-          bitmap[i] = !bitmap[i];
+        bitset.flip();
       break;
     }
     case '?':
-      vec.push_back({QUESTION});
+      vec.emplace_back(QUESTION);
       break;
     case '*':
-      vec.push_back({STAR});
+      vec.emplace_back(STAR);
+      break;
+    case '\\':
+      if (pat.empty())
+        return {};
+      if (vec.empty() || vec.back().kind != STRING)
+        vec.emplace_back(STRING);
+      vec.back().str += pat[0];
+      pat = pat.substr(1);
       break;
     default:
       if (vec.empty() || vec.back().kind != STRING)
-        vec.push_back({STRING});
+        vec.emplace_back(STRING);
       vec.back().str += c;
       break;
     }
   }
 
-  return {GlobPattern{std::move(vec)}};
+  return {Glob{std::move(vec)}};
 }
 
-bool GlobPattern::match(std::string_view str) {
+bool Glob::match(std::string_view str) {
   return do_match(str, elements);
 }
 
-bool GlobPattern::do_match(std::string_view str, std::span<Element> elements) {
+bool Glob::do_match(std::string_view str, std::span<Element> elements) {
   while (!elements.empty()) {
     Element &e = elements[0];
     elements = elements.subspan(1);
 
     switch (e.kind) {
     case STRING:
-      if (str.empty() || !str.starts_with(e.str))
+      if (!str.starts_with(e.str))
         return false;
       str = str.substr(e.str.size());
       break;
@@ -118,13 +124,12 @@ bool GlobPattern::do_match(std::string_view str, std::span<Element> elements) {
         for (;;) {
           size_t pos = str.find(elements[0].str);
           if (pos == str.npos)
-            break;
+            return false;
           if (do_match(str.substr(pos + elements[0].str.size()),
                        elements.subspan(1)))
             return true;
           str = str.substr(pos + 1);
         }
-        return false;
       }
 
       // Other cases are handled here.
@@ -138,7 +143,7 @@ bool GlobPattern::do_match(std::string_view str, std::span<Element> elements) {
       str = str.substr(1);
       break;
     case BRACKET:
-      if (str.empty() || !e.bitmap[str[0]])
+      if (str.empty() || !e.bitset[str[0]])
         return false;
       str = str.substr(1);
       break;
@@ -148,4 +153,4 @@ bool GlobPattern::do_match(std::string_view str, std::span<Element> elements) {
   return str.empty();
 }
 
-} // namespace mold::elf
+} // namespace mold

lib/hyperloglog.cc

@@ -4,7 +4,7 @@
 // For more info, read
 // https://engineering.fb.com/2018/12/13/data-infrastructure/hyperloglog
 
-#include "mold.h"
+#include "common.h"
 
 #include <cmath>
 
@@ -13,7 +13,7 @@ namespace mold {
 i64 HyperLogLog::get_cardinality() const {
   double z = 0;
   for (i64 val : buckets)
-    z += pow(2, -val);
+    z += std::ldexp(1.0, -val);
   return ALPHA * NBUCKETS * NBUCKETS / z;
 }
 

lib/integers.h

@@ -0,0 +1,144 @@
+// This file defines integral types for file input/output. We need to use
+// these types instead of the plain integers (such as uint32_t or int32_t)
+// when reading from/writing to an mmap'ed file area for the following
+// reasons:
+//
+// 1. mold is always a cross linker and should not depend on what host it
+//    is running on. For example, users should be able to run mold on a
+//    big-endian SPARC machine to create a little-endian RV64 binary.
+//
+// 2. Even though data members in all ELF data strucutres are naturally
+//    aligned, they are not guaranteed to be aligned on memory because of
+//    archive files. Archive files (.a files) align each member only to a
+//    2 byte boundary, so anything larger than 2 bytes may be misaligned
+//    in an mmap'ed memory. Misaligned access is an undefined behavior in
+//    C/C++, so we shouldn't cast an arbitrary pointer to a uint32_t, for
+//    example, to read a 32 bit value.
+//
+// The data types defined in this file don't depend on host byte order and
+// don't do unaligned access. Note that modern compilers are smart enough
+// to recognize shift and bitwise OR patterns and compile them into a
+// single load instruction.
+
+#pragma once
+
+#include <cstdint>
+
+#if !defined(__LITTLE_ENDIAN__) && !defined(__BIG_ENDIAN__)
+# if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#  define __LITTLE_ENDIAN__ 1
+# elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#  define __BIG_ENDIAN__ 1
+# else
+#  error "unknown host byte order"
+# endif
+#endif
+
+namespace mold {
+
+typedef uint8_t u8;
+typedef uint16_t u16;
+typedef uint32_t u32;
+typedef uint64_t u64;
+
+typedef int8_t i8;
+typedef int16_t i16;
+typedef int32_t i32;
+typedef int64_t i64;
+
+template <typename T, bool is_le, int size = sizeof(T)>
+class Integer {
+public:
+  constexpr Integer() = default;
+
+  constexpr Integer(T x) requires (is_le && size == 2)
+    : buf{(u8)x, (u8)(x >> 8)} {}
+
+  constexpr Integer(T x) requires (is_le && size == 3)
+    : buf{(u8)x, (u8)(x >> 8), (u8)(x >> 16)} {}
+
+  constexpr Integer(T x) requires (is_le && size == 4)
+    : buf{(u8)x, (u8)(x >> 8), (u8)(x >> 16), (u8)(x >> 24)} {}
+
+  constexpr Integer(T x) requires (is_le && size == 8)
+    : buf{(u8)x,         (u8)(x >> 8),  (u8)(x >> 16), (u8)(x >> 24),
+          (u8)(x >> 32), (u8)(x >> 40), (u8)(x >> 48), (u8)(x >> 56)} {}
+
+  constexpr Integer(T x) requires (!is_le && size == 2)
+    : buf{(u8)(x >> 8), (u8)x} {}
+
+  constexpr Integer(T x) requires (!is_le && size == 3)
+    : buf{(u8)(x >> 16), (u8)(x >> 8), (u8)x} {}
+
+  constexpr Integer(T x) requires (!is_le && size == 4)
+    : buf{(u8)(x >> 24), (u8)(x >> 16), (u8)(x >> 8), (u8)x} {}
+
+  constexpr Integer(T x) requires (!is_le && size == 8)
+    : buf{(u8)(x >> 56), (u8)(x >> 48), (u8)(x >> 40), (u8)(x >> 32),
+          (u8)(x >> 24), (u8)(x >> 16), (u8)(x >> 8),  (u8)x} {}
+
+  operator T() const {
+    if constexpr (is_le) {
+      if constexpr (size == 2)
+        return buf[1] << 8 | buf[0];
+      else if constexpr (size == 3)
+        return buf[2] << 16 | buf[1] << 8 | buf[0];
+      else if constexpr (size == 4)
+        return buf[3] << 24 | buf[2] << 16 | buf[1] << 8 | buf[0];
+      else
+        return (u64)buf[7] << 56 | (u64)buf[6] << 48 |
+               (u64)buf[5] << 40 | (u64)buf[4] << 32 |
+               (u64)buf[3] << 24 | (u64)buf[2] << 16 |
+               (u64)buf[1] << 8  | (u64)buf[0];
+    } else {
+      if constexpr (size == 2)
+        return buf[0] << 8 | buf[1];
+      else if constexpr (size == 3)
+        return buf[0] << 16 | buf[1] << 8 | buf[2];
+      else if constexpr (size == 4)
+        return buf[0] << 24 | buf[1] << 16 | buf[2] << 8 | buf[3];
+      else
+        return (u64)buf[0] << 56 | (u64)buf[1] << 48 |
+               (u64)buf[2] << 40 | (u64)buf[3] << 32 |
+               (u64)buf[4] << 24 | (u64)buf[5] << 16 |
+               (u64)buf[6] << 8  | (u64)buf[7];
+    }
+  }
+
+  Integer &operator=(T x) {
+    new (this) Integer(x);
+    return *this;
+  }
+
+  Integer &operator++()    { return *this = *this + 1; }
+  Integer operator++(int)  { return ++*this - 1; }
+  Integer &operator--()    { return *this = *this - 1; }
+  Integer operator--(int)  { return --*this + 1; }
+  Integer &operator+=(T x) { return *this = *this + x; }
+  Integer &operator-=(T x) { return *this = *this - x; }
+  Integer &operator&=(T x) { return *this = *this & x; }
+  Integer &operator|=(T x) { return *this = *this | x; }
+
+private:
+  u8 buf[size];
+};
+
+using il16 = Integer<i16, true>;
+using il32 = Integer<i32, true>;
+using il64 = Integer<i64, true>;
+
+using ul16 = Integer<u16, true>;
+using ul24 = Integer<u32, true, 3>;
+using ul32 = Integer<u32, true>;
+using ul64 = Integer<u64, true>;
+
+using ib16 = Integer<i16, false>;
+using ib32 = Integer<i32, false>;
+using ib64 = Integer<i64, false>;
+
+using ub16 = Integer<u16, false>;
+using ub24 = Integer<u32, false, 3>;
+using ub32 = Integer<u32, false>;
+using ub64 = Integer<u64, false>;
+
+} // namespace mold

lib/jobs-unix.cc

@@ -0,0 +1,50 @@
+// Many build systems attempt to invoke as many linker processes as there
+// are cores, based on the assumption that the linker is single-threaded.
+// However, since mold is multi-threaded, such build systems' behavior is
+// not beneficial and just increases the overall peak memory usage.
+// On machines with limited memory, this could lead to an out-of-memory
+// error.
+//
+// This file implements a feature that limits the number of concurrent
+// mold processes to just 1 for each user. It is intended to be used as
+// `MOLD_JOBS=1 ninja` or `MOLD_JOBS=1 make -j$(nproc)`.
+
+#include "common.h"
+
+#include <cstdlib>
+#include <fcntl.h>
+#include <pwd.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+namespace mold {
+
+static int lock_fd = -1;
+
+void acquire_global_lock() {
+  char *jobs = getenv("MOLD_JOBS");
+  if (!jobs || jobs != "1"s)
+    return;
+
+  std::string path;
+  if (char *dir = getenv("XDG_RUNTIME_DIR"))
+    path = dir + "/mold-lock"s;
+  else
+    path = "/tmp/mold-lock-"s + getpwuid(getuid())->pw_name;
+
+  int fd = open(path.c_str(), O_WRONLY | O_CREAT | O_CLOEXEC, 0600);
+  if (fd == -1)
+    return;
+
+  if (lockf(fd, F_LOCK, 0) == -1)
+    return;
+  lock_fd = fd;
+}
+
+void release_global_lock() {
+  if (lock_fd != -1)
+    close(lock_fd);
+}
+
+} // namespace mold

lib/jobs-win32.cc

@@ -0,0 +1,6 @@
+namespace mold {
+
+void acquire_global_lock() {}
+void release_global_lock() {}
+
+} // namespace mold

lib/mapped-file-unix.cc

@@ -0,0 +1,51 @@
+#include "common.h"
+
+namespace mold {
+
+MappedFile *open_file_impl(const std::string &path, std::string &error) {
+  i64 fd = ::open(path.c_str(), O_RDONLY);
+  if (fd == -1) {
+    if (errno != ENOENT)
+      error = "opening " + path + " failed: " + errno_string();
+    return nullptr;
+  }
+
+  struct stat st;
+  if (fstat(fd, &st) == -1)
+    error = path + ": fstat failed: " + errno_string();
+
+  MappedFile *mf = new MappedFile;
+  mf->name = path;
+  mf->size = st.st_size;
+
+  if (st.st_size > 0) {
+    mf->data = (u8 *)mmap(nullptr, st.st_size, PROT_READ | PROT_WRITE,
+                          MAP_PRIVATE, fd, 0);
+    if (mf->data == MAP_FAILED)
+      error = path + ": mmap failed: " + errno_string();
+  }
+
+  close(fd);
+  return mf;
+}
+
+void MappedFile::unmap() {
+  if (size == 0 || parent || !data)
+    return;
+  munmap(data, size);
+  data = nullptr;
+}
+
+void MappedFile::close_fd() {
+  if (fd == -1)
+    return;
+  close(fd);
+  fd = -1;
+}
+
+void MappedFile::reopen_fd(const std::string &path) {
+  if (fd == -1)
+    fd = open(path.c_str(), O_RDONLY);
+}
+
+} // namespace mold

lib/mapped-file-win32.cc

@@ -0,0 +1,81 @@
+#include "common.h"
+
+namespace mold {
+
+MappedFile *open_file_impl(const std::string &path, std::string &error) {
+  HANDLE fd = CreateFileA(path.c_str(), GENERIC_READ,
+                          FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
+                          nullptr, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, nullptr);
+
+  if (fd == INVALID_HANDLE_VALUE) {
+    auto err = GetLastError();
+    if (err != ERROR_FILE_NOT_FOUND)
+      error = "opening " + path + " failed: " + errno_string();
+    return nullptr;
+  }
+
+  if (GetFileType(fd) != FILE_TYPE_DISK) {
+    CloseHandle(fd);
+    return nullptr;
+  }
+
+  DWORD size_hi;
+  DWORD size_lo = GetFileSize(fd, &size_hi);
+
+  if (size_lo == INVALID_FILE_SIZE) {
+    error = path + ": GetFileSize failed: " + errno_string();
+    return nullptr;
+  }
+
+  u64 size = ((u64)size_hi << 32) + size_lo;
+
+  MappedFile *mf = new MappedFile;
+  mf->name = path;
+  mf->size = size;
+  mf->fd = fd;
+
+  if (size > 0) {
+    HANDLE h = CreateFileMapping(fd, nullptr, PAGE_READONLY, 0, size, nullptr);
+    if (!h) {
+      error = path + ": CreateFileMapping failed: " + errno_string();
+      return nullptr;
+    }
+
+    mf->data = (u8 *)MapViewOfFile(h, FILE_MAP_COPY, 0, 0, size);
+    CloseHandle(h);
+
+    if (!mf->data) {
+      error = path + ": MapViewOfFile failed: " + errno_string();
+      return nullptr;
+    }
+  }
+
+  return mf;
+}
+
+void MappedFile::unmap() {
+  if (size == 0 || parent || !data)
+    return;
+
+  UnmapViewOfFile(data);
+  if (fd != INVALID_HANDLE_VALUE)
+    CloseHandle(fd);
+
+  data = nullptr;
+}
+
+void MappedFile::close_fd() {
+  if (fd == INVALID_HANDLE_VALUE)
+    return;
+  CloseHandle(fd);
+  fd = INVALID_HANDLE_VALUE;
+}
+
+void MappedFile::reopen_fd(const std::string &path) {
+  if (fd == INVALID_HANDLE_VALUE)
+    fd = CreateFileA(path.c_str(), GENERIC_READ,
+                     FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
+                     nullptr, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, nullptr);
+}
+
+} // namespace mold

lib/mimalloc.cc

@@ -0,0 +1,24 @@
+#include "common.h"
+
+// Including mimalloc-new-delete.h overrides new/delete operators.
+// We need it only when we are using mimalloc as a dynamic library.
+#if MOLD_USE_SYSTEM_MIMALLOC
+# include <mimalloc-new-delete.h>
+#endif
+
+// Silence mimalloc warning messages that users can just ignore.
+#if MOLD_USE_MIMALLOC
+# include <mimalloc.h>
+
+namespace mold {
+void set_mimalloc_options() {
+  mi_option_disable(mi_option_verbose);
+  mi_option_disable(mi_option_show_errors);
+}
+}
+
+#else
+namespace mold {
+void set_mimalloc_options() {}
+}
+#endif

lib/multi-glob.cc

@@ -1,65 +1,75 @@
-// This file implements the Aho-Corasick algorithm to match version
-// script patterns to symbol strings as quickly as possible.
+// This file implements the Aho-Corasick algorithm to match multiple
+// glob patterns to symbol strings as quickly as possible.
 //
-// Here are some examples of version script patterns:
+// Here are some examples of glob patterns:
 //
 //    qt_private_api_tag*
 //    *16QAccessibleCache*
 //    *32QAbstractFileIconProviderPrivate*
 //    *17QPixmapIconEngine*
 //
-// The pattern is a glob pattern, so `*` is a wildcard that matches
-// any substring. We sometimes have hundreds of version script
-// patterns and have to match them against millions of symbol strings.
+// `*` is a wildcard that matches any substring. We sometimes have
+// hundreds of glob patterns and have to match them against millions
+// of symbol strings.
 //
 // Aho-Corasick cannot handle complex patterns such as `*foo*bar*`.
-// We handle such patterns with GlobPattern. GlobPattern is relatively
-// slow, but complex patterns are rare in practice, so it should be OK.
+// We handle such patterns with the Glob class. Glob is relatively
+// slow, but complex patterns are rare in practice, so it should be
+// OK.
 
-#include "mold.h"
+#include "common.h"
 
 #include <queue>
 #include <regex>
 
-namespace mold::elf {
+namespace mold {
 
-std::optional<u16> VersionMatcher::find(std::string_view str) {
-  u32 idx = UINT32_MAX;
+std::optional<i64> MultiGlob::find(std::string_view str) {
+  std::call_once(once, [&] { compile(); });
+  i64 val = -1;
 
-  if (root) {
-    std::call_once(once_flag, [&]() { compile(); });
-
-    // Match against simple glob patterns
-    TrieNode *node = root.get();
-
-    auto walk = [&](u8 c) {
-      for (;;) {
-        if (node->children[c]) {
-          node = node->children[c].get();
-          idx = std::min(idx, node->value);
-          return;
-        }
-
-        if (!node->suffix_link)
-          return;
-        node = node->suffix_link;
-      }
-    };
-
-    walk('\0');
-    for (u8 c : str)
-      walk(c);
-    walk('\0');
-  }
+  // Match against simple glob patterns
+  if (root)
+    val = find_aho_corasick(str);
 
   // Match against complex glob patterns
-  for (std::pair<GlobPattern, u32> &glob : globs)
+  for (std::pair<Glob, i64> &glob : globs)
     if (glob.first.match(str))
-      idx = std::min(idx, glob.second);
+      val = std::max(val, glob.second);
 
-  if (idx == UINT32_MAX)
+  if (val == -1)
     return {};
-  return versions[idx];
+  return val;
+}
+
+i64 MultiGlob::find_aho_corasick(std::string_view str) {
+  TrieNode *node = root.get();
+  i64 val = -1;
+
+  auto walk = [&](u8 c) {
+    for (;;) {
+      if (node->children[c]) {
+        node = node->children[c].get();
+        val = std::max(val, node->value);
+        return;
+      }
+
+      if (!node->suffix_link)
+        return;
+      node = node->suffix_link;
+    }
+  };
+
+  walk('\0');
+
+  for (u8 c : str) {
+    if (prefix_match && node == root.get())
+      return val;
+    walk(c);
+  }
+
+  walk('\0');
+  return val;
 }
 
 static bool is_simple_pattern(std::string_view pat) {
@@ -82,18 +92,16 @@ static std::string handle_stars(std::string_view pat) {
   return "\0"s + str + "\0"s;
 }
 
-bool VersionMatcher::add(std::string_view pat, u16 ver) {
-  assert(!compiled);
+bool MultiGlob::add(std::string_view pat, i64 val) {
+  assert(!is_compiled);
   assert(!pat.empty());
 
-  u32 idx = strings.size();
-  strings.push_back(std::string(pat));
-  versions.push_back(ver);
+  strings.emplace_back(pat);
 
   // Complex glob pattern
   if (!is_simple_pattern(pat)) {
-    if (std::optional<GlobPattern> glob = GlobPattern::compile(pat)) {
-      globs.push_back({std::move(*glob), idx});
+    if (std::optional<Glob> glob = Glob::compile(pat)) {
+      globs.emplace_back(std::move(*glob), val);
       return true;
     }
     return false;
@@ -110,17 +118,29 @@ bool VersionMatcher::add(std::string_view pat, u16 ver) {
     node = node->children[c].get();
   }
 
-  node->value = std::min(node->value, idx);
+  node->value = std::max(node->value, val);
   return true;
 }
 
-void VersionMatcher::compile() {
-  fix_suffix_links(*root);
-  fix_values();
-  compiled = true;
+void MultiGlob::compile() {
+  is_compiled = true;
+  if (root) {
+    fix_suffix_links(*root);
+    fix_values();
+
+    // If no pattern starts with '*', set prefix_match to true.
+    // We'll use this flag for optimization.
+    prefix_match = true;
+    for (i64 i = 1; i < 256; i++) {
+      if (root->children[i]) {
+        prefix_match = false;
+        break;
+      }
+    }
+  }
 }
 
-void VersionMatcher::fix_suffix_links(TrieNode &node) {
+void MultiGlob::fix_suffix_links(TrieNode &node) {
   for (i64 i = 0; i < 256; i++) {
     if (!node.children[i])
       continue;
@@ -146,7 +166,7 @@ void VersionMatcher::fix_suffix_links(TrieNode &node) {
   }
 }
 
-void VersionMatcher::fix_values() {
+void MultiGlob::fix_values() {
   std::queue<TrieNode *> queue;
   queue.push(root.get());
 
@@ -157,10 +177,10 @@ void VersionMatcher::fix_values() {
     for (std::unique_ptr<TrieNode> &child : node->children) {
       if (!child)
         continue;
-      child->value = std::min(child->value, child->suffix_link->value);
+      child->value = std::max(child->value, child->suffix_link->value);
       queue.push(child.get());
     }
   } while (!queue.empty());
 }
 
-} // namespace mold::elf
+} // namespace mold

lib/perf.cc

@@ -1,10 +1,14 @@
-#include "mold.h"
+#include "common.h"
 
 #include <functional>
 #include <iomanip>
 #include <ios>
+#include <tbb/concurrent_vector.h>
+
+#ifndef _WIN32
 #include <sys/resource.h>
 #include <sys/time.h>
+#endif
 
 namespace mold {
 
@@ -13,9 +17,7 @@ i64 Counter::get_value() {
 }
 
 void Counter::print() {
-  sort(instances, [](Counter *a, Counter *b) {
-    return a->get_value() > b->get_value();
-  });
+  ranges::stable_sort(instances, {}, [](Counter *x) { return x->get_value(); });
 
   for (Counter *c : instances)
     std::cout << std::setw(20) << std::right << c->name
@@ -23,24 +25,33 @@ void Counter::print() {
 }
 
 static i64 now_nsec() {
-  struct timespec t;
-  clock_gettime(CLOCK_MONOTONIC, &t);
-  return (i64)t.tv_sec * 1000000000 + t.tv_nsec;
+  return (i64)std::chrono::steady_clock::now().time_since_epoch().count();
 }
 
-static i64 to_nsec(struct timeval t) {
-  return (i64)t.tv_sec * 1000000000 + t.tv_usec * 1000;
+static std::pair<i64, i64> get_usage() {
+#ifdef _WIN32
+  auto to_nsec = [](FILETIME t) -> i64 {
+    return (((u64)t.dwHighDateTime << 32) + (u64)t.dwLowDateTime) * 100;
+  };
+
+  FILETIME creation, exit, kernel, user;
+  GetProcessTimes(GetCurrentProcess(), &creation, &exit, &kernel, &user);
+  return {to_nsec(user), to_nsec(kernel)};
+#else
+  auto to_nsec = [](struct timeval t) -> i64 {
+    return (i64)t.tv_sec * 1'000'000'000 + t.tv_usec * 1'000;
+  };
+
+  struct rusage ru;
+  getrusage(RUSAGE_SELF, &ru);
+  return {to_nsec(ru.ru_utime), to_nsec(ru.ru_stime)};
+#endif
 }
 
 TimerRecord::TimerRecord(std::string name, TimerRecord *parent)
   : name(name), parent(parent) {
-  struct rusage usage;
-  getrusage(RUSAGE_SELF, &usage);
-
   start = now_nsec();
-  user = to_nsec(usage.ru_utime);
-  sys = to_nsec(usage.ru_stime);
-
+  std::tie(user, sys) = get_usage();
   if (parent)
     parent->children.push_back(this);
 }
@@ -50,25 +61,24 @@ void TimerRecord::stop() {
     return;
   stopped = true;
 
-  struct rusage usage;
-  getrusage(RUSAGE_SELF, &usage);
+  i64 user2;
+  i64 sys2;
+  std::tie(user2, sys2) = get_usage();
 
   end = now_nsec();
-  user = to_nsec(usage.ru_utime) - user;
-  sys = to_nsec(usage.ru_stime) - sys;
+  user = user2 - user;
+  sys = sys2 - sys;
 }
 
 static void print_rec(TimerRecord &rec, i64 indent) {
   printf(" % 8.3f % 8.3f % 8.3f  %s%s\n",
-         ((double)rec.user / 1000000000),
-         ((double)rec.sys / 1000000000),
-         (((double)rec.end - rec.start) / 1000000000),
+         ((double)rec.user / 1'000'000'000),
+         ((double)rec.sys / 1'000'000'000),
+         (((double)rec.end - rec.start) / 1'000'000'000),
          std::string(indent * 2, ' ').c_str(),
          rec.name.c_str());
 
-  sort(rec.children, [](TimerRecord *a, TimerRecord *b) {
-    return a->start < b->start;
-  });
+  ranges::stable_sort(rec.children, {}, &TimerRecord::start);
 
   for (TimerRecord *child : rec.children)
     print_rec(*child, indent + 1);

lib/random.cc

@@ -0,0 +1,20 @@
+#include "common.h"
+
+#include <random>
+
+namespace mold {
+
+void get_random_bytes(u8 *buf, i64 size) {
+  std::random_device rand;
+  i64 i = 0;
+
+  for (; i < size - 4; i += 4) {
+    u32 val = rand();
+    memcpy(buf + i, &val, 4);
+  }
+
+  u32 val = rand();
+  memcpy(buf + i, &val, size - i);
+}
+
+} // namespace mold

lib/signal-unix.cc

@@ -0,0 +1,88 @@
+#include "common.h"
+
+#include <csignal>
+#include <tbb/version.h>
+
+#ifdef __FreeBSD__
+# include <sys/sysctl.h>
+# include <unistd.h>
+#endif
+
+namespace mold {
+
+std::string errno_string() {
+  // strerror is not thread-safe, so guard it with a lock.
+  static std::mutex mu;
+  std::scoped_lock lock(mu);
+  return strerror(errno);
+}
+
+void cleanup() {
+  if (output_tmpfile)
+    unlink(output_tmpfile);
+}
+
+// mold mmap's an output file, and the mmap succeeds even if there's
+// no enough space left on the filesystem. The actual disk blocks are
+// not allocated on the mmap call but when the program writes to it
+// for the first time.
+//
+// If a disk becomes full as a result of a write to an mmap'ed memory
+// region, the failure of the write is reported as a SIGBUS or structured
+// exeption with code EXCEPTION_IN_PAGE_ERROR on Windows. This
+// signal handler catches that signal and prints out a user-friendly
+// error message. Without this, it is very hard to realize that the
+// disk might be full.
+static std::string sigabrt_msg;
+
+static void sighandler(int signo, siginfo_t *info, void *ucontext) {
+  static std::mutex mu;
+  std::scoped_lock lock{mu};
+
+  // Handle disk full error
+  switch (signo) {
+  case SIGSEGV:
+  case SIGBUS:
+    if (output_buffer_start <= info->si_addr &&
+        info->si_addr < output_buffer_end) {
+      const char msg[] = "mold: failed to write to an output file. Disk full?\n";
+      (void)!write(STDERR_FILENO, msg, sizeof(msg) - 1);
+    }
+    break;
+  case SIGABRT: {
+    (void)!write(STDERR_FILENO, &sigabrt_msg[0], sigabrt_msg.size());
+    break;
+  }
+  }
+
+  // Re-throw the signal
+  signal(SIGSEGV, SIG_DFL);
+  signal(SIGBUS, SIG_DFL);
+  signal(SIGABRT, SIG_DFL);
+
+  cleanup();
+  raise(signo);
+}
+
+void install_signal_handler() {
+  struct sigaction action;
+  action.sa_sigaction = sighandler;
+  sigemptyset(&action.sa_mask);
+  action.sa_flags = SA_SIGINFO;
+
+  sigaction(SIGSEGV, &action, NULL);
+  sigaction(SIGBUS, &action, NULL);
+
+  // OneTBB 2021.9.0 has the interface version 12090.
+  if (TBB_runtime_interface_version() < 12090) {
+    sigabrt_msg = "mold: aborted\n"
+      "mold: mold with libtbb version 2021.9.0 or older is known to be unstable "
+      "under heavy load. Your libtbb version is " +
+      std::string(TBB_runtime_version()) +
+      ". Please upgrade your libtbb library and try again.\n";
+
+    sigaction(SIGABRT, &action, NULL);
+  }
+}
+
+} // namespace mold

lib/signal-win32.cc

@@ -0,0 +1,55 @@
+#include "common.h"
+
+#include <windows.h>
+
+namespace mold {
+
+void cleanup() {
+  if (output_tmpfile)
+    _unlink(output_tmpfile);
+}
+
+std::string errno_string() {
+  LPVOID buf;
+  DWORD dw = GetLastError();
+
+  FormatMessage(FORMAT_MESSAGE_ALLOCATE_BUFFER |
+                FORMAT_MESSAGE_FROM_SYSTEM |
+                FORMAT_MESSAGE_IGNORE_INSERTS,
+                nullptr, dw,
+                MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
+                (LPTSTR)&buf, 0, nullptr);
+
+  std::string ret = (char *)buf;
+  LocalFree(buf);
+  return ret;
+}
+
+static LONG WINAPI vectored_handler(_EXCEPTION_POINTERS *exception_info) {
+  static std::mutex mu;
+  std::scoped_lock lock{mu};
+
+  PEXCEPTION_RECORD rec = exception_info->ExceptionRecord;
+  ULONG_PTR *p = rec->ExceptionInformation;
+
+  if (rec->ExceptionCode == EXCEPTION_IN_PAGE_ERROR &&
+      (ULONG_PTR)output_buffer_start <= p[1] &&
+      p[1] < (ULONG_PTR)output_buffer_end) {
+    static const char msg[] =
+      "mold: failed to write to an output file. Disk full?\n";
+    (void)!_write(_fileno(stderr), msg, sizeof(msg) - 1);
+  } else if (rec->ExceptionCode == EXCEPTION_STACK_OVERFLOW) {
+    static const char msg[] =
+      "mold: stack overflow\n";
+    (void)!_write(_fileno(stderr), msg, sizeof(msg) - 1);
+  }
+
+  cleanup();
+  _exit(1);
+}
+
+void install_signal_handler() {
+  AddVectoredExceptionHandler(0, vectored_handler);
+}
+
+} // namespace mold

lib/siphash.h

@@ -0,0 +1,144 @@
+// This is a header-only C++20 implementation of SipHash based on the
+// reference implementation. To use, just copy this header file into
+// your project and #include it.
+//
+// https://github.com/rui314/siphash/blob/main/siphash.h
+
+#include <bit>
+#include <cstdint>
+#include <cstring>
+
+template <int C_ROUNDS, int D_ROUNDS, int OUTLEN>
+class SipHashTmpl {
+public:
+  static_assert(OUTLEN == 64 || OUTLEN == 128);
+
+  SipHashTmpl(void *key) {
+    uint64_t k0 = read64(key);
+    uint64_t k1 = read64((char *)key + 8);
+
+    v0 = 0x736f6d6570736575 ^ k0;
+    v1 = 0x646f72616e646f6d ^ k1;
+    v2 = 0x6c7967656e657261 ^ k0;
+    v3 = 0x7465646279746573 ^ k1;
+
+    if (OUTLEN == 128)
+      v1 ^= 0xee;
+  }
+
+  void update(void *msgp, int64_t msglen) {
+    char *msg = (char *)msgp;
+    sum += msglen;
+
+    if (buflen) {
+      if (buflen + msglen < 8) {
+        memcpy(buf + buflen, msg, msglen);
+        buflen += msglen;
+        return;
+      }
+
+      int j = 8 - buflen;
+      memcpy(buf + buflen, msg, j);
+      compress(read64(buf));
+
+      msg += j;
+      msglen -= j;
+      buflen = 0;
+    }
+
+    while (msglen >= 8) {
+      compress(read64(msg));
+      msg += 8;
+      msglen -= 8;
+    }
+
+    memcpy(buf, msg, msglen);
+    buflen = msglen;
+  }
+
+  void finish(void *out) {
+    memset(buf + buflen, 0, 8 - buflen);
+    compress(((uint64_t)sum << 56) | read64(buf));
+
+    v2 ^= (OUTLEN == 128) ? 0xee : 0xff;
+    finalize();
+    write64(out, v0 ^ v1 ^ v2 ^ v3);
+
+    if (OUTLEN == 128) {
+      v1 ^= 0xdd;
+      finalize();
+      write64((char *)out + 8, v0 ^ v1 ^ v2 ^ v3);
+    }
+  }
+
+  static void hash(void *out, void *key, void *in, int inlen) {
+    SipHashTmpl<C_ROUNDS, D_ROUNDS, OUTLEN> h(key);
+    h.update(in, inlen);
+    h.finish(out);
+  }
+
+private:
+  uint64_t v0, v1, v2, v3;
+  uint8_t buf[8];
+  uint8_t buflen = 0;
+  uint8_t sum = 0;
+
+  uint64_t read64(void *loc) {
+    uint64_t val;
+    memcpy(&val, loc, 8);
+    if (std::endian::native == std::endian::big)
+      val = bswap(val);
+    return val;
+  }
+
+  void write64(void *loc, uint64_t val) {
+    if (std::endian::native == std::endian::big)
+      val = bswap(val);
+    memcpy(loc, &val, 8);
+  }
+
+  uint64_t bswap(uint64_t val) {
+    return ((val << 56) & 0xff00000000000000) |
+           ((val << 40) & 0x00ff000000000000) |
+           ((val << 24) & 0x0000ff0000000000) |
+           ((val << 8)  & 0x000000ff00000000) |
+           ((val >> 8)  & 0x00000000ff000000) |
+           ((val >> 24) & 0x0000000000ff0000) |
+           ((val >> 40) & 0x000000000000ff00) |
+           ((val >> 56) & 0x00000000000000ff);
+  }
+
+  void round() {
+    v0 += v1;
+    v1 = std::rotl(v1, 13);
+    v1 ^= v0;
+    v0 = std::rotl(v0, 32);
+    v2 += v3;
+    v3 = std::rotl(v3, 16);
+    v3 ^= v2;
+    v0 += v3;
+    v3 = std::rotl(v3, 21);
+    v3 ^= v0;
+    v2 += v1;
+    v1 = std::rotl(v1, 17);
+    v1 ^= v2;
+    v2 = std::rotl(v2, 32);
+  }
+
+  void compress(uint64_t m) {
+    v3 ^= m;
+    for (int i = 0; i < C_ROUNDS; i++)
+      round();
+    v0 ^= m;
+  }
+
+  void finalize() {
+    for (int i = 0; i < D_ROUNDS; i++)
+      round();
+  }
+};
+
+using SipHash = SipHashTmpl<2, 4, 64>;
+using SipHash128 = SipHashTmpl<2, 4, 128>;
+using SipHash13 = SipHashTmpl<1, 3, 64>;
+using SipHash13_128 = SipHashTmpl<1, 3, 128>;

lib/tar.cc

@@ -0,0 +1,114 @@
+// This file contains functions to create a tar file.
+
+#include "common.h"
+
+#ifdef _WIN32
+# define ftruncate _chsize_s
+#endif
+
+namespace mold {
+
+static constexpr i64 BLOCK_SIZE = 512;
+
+// A tar file consists of one or more Ustar header followed by data.
+// Each Ustar header represents a single file in an archive.
+//
+// tar is an old file format, and its `name` field is only 100 bytes long.
+// If `name` is longer than 100 bytes, we can emit a PAX header before a
+// Ustar header to store a long filename.
+//
+// For simplicity, we always emit a PAX header even for a short filename.
+struct UstarHeader {
+  char name[100];
+  char mode[8];
+  char uid[8];
+  char gid[8];
+  char size[12];
+  char mtime[12];
+  char checksum[8];
+  char typeflag[1];
+  char linkname[100];
+  char magic[6];
+  char version[2];
+  char uname[32];
+  char gname[32];
+  char devmajor[8];
+  char devminor[8];
+  char prefix[155];
+  char pad[12];
+};
+
+static_assert(sizeof(UstarHeader) == BLOCK_SIZE);
+
+static void finalize(UstarHeader &hdr) {
+  memset(hdr.checksum, ' ', sizeof(hdr.checksum));
+  memcpy(hdr.magic, "ustar", 5);
+  memcpy(hdr.version, "00", 2);
+
+  // Compute checksum
+  int sum = 0;
+  for (i64 i = 0; i < sizeof(hdr); i++)
+    sum += ((u8 *)&hdr)[i];
+
+  // We need to convince the compiler that sum isn't too big to silence
+  // -Werror=format-truncation.
+  if (sum >= 01'000'000)
+    unreachable();
+  snprintf(hdr.checksum, sizeof(hdr.checksum), "%06o", sum);
+}
+
+static std::string encode_path(std::string basedir, std::string path) {
+  path = path_clean(basedir + "/" + path);
+
+  // Construct a string which contains something like
+  // "16 path=foo/bar\n" where 16 is the size of the string
+  // including the size string itself.
+  i64 len = std::string(" path=\n").size() + path.size();
+  i64 total = std::to_string(len).size() + len;
+  total = std::to_string(total).size() + len;
+  return std::to_string(total) + " path=" + path + "\n";
+}
+
+std::unique_ptr<TarWriter>
+TarWriter::open(std::string output_path, std::string basedir) {
+  FILE *out = fopen(output_path.c_str(), "w");
+  if (!out)
+    return nullptr;
+  return std::unique_ptr<TarWriter>(new TarWriter(out, basedir));
+}
+
+TarWriter::~TarWriter() {
+  fclose(out);
+}
+
+void TarWriter::append(std::string path, std::string_view data) {
+  // Write PAX header
+  UstarHeader pax = {};
+
+  std::string attr = encode_path(basedir, path);
+  snprintf(pax.size, sizeof(pax.size), "%011zo", attr.size());
+  pax.name[0] = '/';
+  pax.typeflag[0] = 'x';
+  finalize(pax);
+  fwrite(&pax, sizeof(pax), 1, out);
+
+  // Write pathname
+  fwrite(attr.data(), attr.size(), 1, out);
+  fseek(out, align_to(ftell(out), BLOCK_SIZE), SEEK_SET);
+
+  // Write Ustar header
+  UstarHeader ustar = {};
+  memcpy(ustar.mode, "0000664", 8);
+  snprintf(ustar.size, sizeof(ustar.size), "%011zo", data.size());
+  finalize(ustar);
+  fwrite(&ustar, sizeof(ustar), 1, out);
+
+  // Write file contents
+  fwrite(data.data(), data.size(), 1, out);
+  fseek(out, align_to(ftell(out), BLOCK_SIZE), SEEK_SET);
+
+  // A tar file must ends with two empty blocks
+  (void)!ftruncate(fileno(out), ftell(out) + BLOCK_SIZE * 2);
+}
+
+} // namespace mold

lib/update-git-hash.cmake

@@ -0,0 +1,33 @@
+# Get a git hash value. We do not want to use git command here
+# because we don't want to make git a build-time dependency.
+if(EXISTS "${SOURCE_DIR}/.git/HEAD")
+  file(READ "${SOURCE_DIR}/.git/HEAD" HASH)
+  string(STRIP "${HASH}" HASH)
+
+  if(HASH MATCHES "^ref: (.*)")
+    set(HEAD "${CMAKE_MATCH_1}")
+    if(EXISTS "${SOURCE_DIR}/.git/${HEAD}")
+      file(READ "${SOURCE_DIR}/.git/${HEAD}" HASH)
+      string(STRIP "${HASH}" HASH)
+    else()
+      file(READ "${SOURCE_DIR}/.git/packed-refs" PACKED_REFS)
+      string(REGEX REPLACE ".*\n([0-9a-f]+) ${HEAD}\n.*" "\\1" HASH "\n${PACKED_REFS}")
+    endif()
+  endif()
+endif()
+
+# Create new file contents and update a given file if necessary.
+set(NEW_CONTENTS "#include <string>
+namespace mold {
+std::string mold_git_hash = \"${HASH}\";
+}
+")
+
+if(EXISTS "${OUTPUT_FILE}")
+  file(READ "${OUTPUT_FILE}" OLD_CONTENTS)
+  if(NOT "${NEW_CONTENTS}" STREQUAL "${OLD_CONTENTS}")
+    file(WRITE "${OUTPUT_FILE}" "${NEW_CONTENTS}")
+  endif()
+else()
+  file(WRITE "${OUTPUT_FILE}" "${NEW_CONTENTS}")
+endif()

macho/arch-arm64.cc

@@ -1,258 +0,0 @@
-#include "mold.h"
-
-namespace mold::macho {
-
-// Returns [hi:lo] bits of val.
-static u64 bits(u64 val, u64 hi, u64 lo) {
-  return (val >> lo) & (((u64)1 << (hi - lo + 1)) - 1);
-}
-
-static u64 page(u64 val) {
-  return val & 0xffff'ffff'ffff'f000;
-}
-
-static u64 encode_page(u64 val) {
-  return (bits(val, 13, 12) << 29) | (bits(val, 32, 14) << 5);
-}
-
-template <>
-void StubsSection<ARM64>::copy_buf(Context<ARM64> &ctx) {
-  u32 *buf = (u32 *)(ctx.buf + this->hdr.offset);
-
-  for (i64 i = 0; i < syms.size(); i++) {
-    static const u32 insn[] = {
-      0x90000010, // adrp x16, $ptr@PAGE
-      0xf9400210, // ldr  x16, [x16, $ptr@PAGEOFF]
-      0xd61f0200, // br   x16
-    };
-
-    static_assert(sizeof(insn) == ARM64::stub_size);
-
-    u64 la_addr = ctx.lazy_symbol_ptr.hdr.addr + ARM64::word_size * i;
-    u64 this_addr = this->hdr.addr + ARM64::stub_size * i;
-
-    memcpy(buf, insn, sizeof(insn));
-    buf[0] |= encode_page(page(la_addr) - page(this_addr));
-    buf[1] |= bits(la_addr, 11, 3) << 10;
-    buf += 3;
-  }
-}
-
-template <>
-void StubHelperSection<ARM64>::copy_buf(Context<ARM64> &ctx) {
-  u32 *start = (u32 *)(ctx.buf + this->hdr.offset);
-  u32 *buf = start;
-
-  static const u32 insn0[] = {
-    0x90000011, // adrp x17, $__dyld_private@PAGE
-    0x91000231, // add  x17, x17, $__dyld_private@PAGEOFF
-    0xa9bf47f0, // stp	x16, x17, [sp, #-16]!
-    0x90000010, // adrp x16, $dyld_stub_binder@PAGE
-    0xf9400210, // ldr  x16, [x16, $dyld_stub_binder@PAGEOFF]
-    0xd61f0200, // br   x16
-  };
-
-  static_assert(sizeof(insn0) == ARM64::stub_helper_hdr_size);
-  memcpy(buf, insn0, sizeof(insn0));
-
-  u64 dyld_private = get_symbol(ctx, "__dyld_private")->get_addr(ctx);
-  buf[0] |= encode_page(page(dyld_private) - page(this->hdr.addr));
-  buf[1] |= bits(dyld_private, 11, 0) << 10;
-
-  u64 stub_binder = get_symbol(ctx, "dyld_stub_binder")->get_addr(ctx);
-  buf[3] |= encode_page(page(stub_binder) - page(this->hdr.addr - 12));
-  buf[4] |= bits(stub_binder, 11, 0) << 10;
-
-  buf += 6;
-
-  for (i64 i = 0; i < ctx.stubs.syms.size(); i++) {
-    static const u32 insn[] = {
-      0x18000050, // ldr  w16, addr
-      0x14000000, // b    stubHelperHeader
-      0x00000000, // addr: .long <idx>
-    };
-
-    static_assert(sizeof(insn) == ARM64::stub_helper_size);
-
-    memcpy(buf, insn, sizeof(insn));
-    buf[1] |= bits((start - buf - 1) * 4, 27, 2);
-    buf[2] = ctx.stubs.bind_offsets[i];
-    buf += 3;
-  }
-}
-
-static i64 read_addend(u8 *buf, const MachRel &r) {
-  if (r.p2size == 2)
-    return *(i32 *)(buf + r.offset);
-  if (r.p2size == 3)
-    return *(i64 *)(buf + r.offset);
-  unreachable();
-}
-
-static Relocation<ARM64>
-read_reloc(Context<ARM64> &ctx, ObjectFile<ARM64> &file,
-           const MachSection &hdr, MachRel *rels, i64 &idx) {
-  i64 addend = 0;
-
-  switch (rels[idx].type) {
-  case ARM64_RELOC_UNSIGNED:
-  case ARM64_RELOC_SUBTRACTOR:
-    addend = read_addend((u8 *)file.mf->data + hdr.offset, rels[idx]);
-    break;
-  case ARM64_RELOC_ADDEND:
-    addend = rels[idx++].offset;
-    break;
-  }
-
-  MachRel &r = rels[idx];
-  Relocation<ARM64> rel{r.offset, (u8)r.type, (u8)r.p2size, (bool)r.is_pcrel};
-
-  if (r.is_extern) {
-    rel.sym = file.syms[r.idx];
-    rel.addend = addend;
-    return rel;
-  }
-
-  u32 addr;
-  if (r.is_pcrel)
-    addr = hdr.addr + r.offset + addend;
-  else
-    addr = addend;
-
-  Subsection<ARM64> *target = file.find_subsection(ctx, addr);
-  if (!target)
-    Fatal(ctx) << file << ": bad relocation: " << r.offset;
-
-  rel.subsec = target;
-  rel.addend = addr - target->input_addr;
-  return rel;
-}
-
-template <>
-std::vector<Relocation<ARM64>>
-read_relocations(Context<ARM64> &ctx, ObjectFile<ARM64> &file,
-                 const MachSection &hdr) {
-  std::vector<Relocation<ARM64>> vec;
-
-  MachRel *rels = (MachRel *)(file.mf->data + hdr.reloff);
-  for (i64 i = 0; i < hdr.nreloc; i++)
-    vec.push_back(read_reloc(ctx, file, hdr, rels, i));
-  return vec;
-}
-
-template <>
-void Subsection<ARM64>::scan_relocations(Context<ARM64> &ctx) {
-  for (Relocation<ARM64> &r : get_rels()) {
-    Symbol<ARM64> *sym = r.sym;
-    if (!sym)
-      continue;
-
-    switch (r.type) {
-    case ARM64_RELOC_GOT_LOAD_PAGE21:
-    case ARM64_RELOC_GOT_LOAD_PAGEOFF12:
-    case ARM64_RELOC_POINTER_TO_GOT:
-      sym->flags |= NEEDS_GOT;
-      break;
-    case ARM64_RELOC_TLVP_LOAD_PAGE21:
-    case ARM64_RELOC_TLVP_LOAD_PAGEOFF12:
-      sym->flags |= NEEDS_THREAD_PTR;
-      break;
-    }
-
-    if (sym->file && sym->file->is_dylib) {
-      sym->flags |= NEEDS_STUB;
-      ((DylibFile<ARM64> *)sym->file)->is_needed = true;
-    }
-  }
-}
-
-template <>
-void Subsection<ARM64>::apply_reloc(Context<ARM64> &ctx, u8 *buf) {
-  std::span<Relocation<ARM64>> rels = get_rels();
-
-  for (i64 i = 0; i < rels.size(); i++) {
-    Relocation<ARM64> &r = rels[i];
-
-    if (r.sym && !r.sym->file) {
-      Error(ctx) << "undefined symbol: " << isec.file << ": " << *r.sym;
-      continue;
-    }
-
-    u64 val = 0;
-
-    switch (r.type) {
-    case ARM64_RELOC_UNSIGNED:
-    case ARM64_RELOC_BRANCH26:
-    case ARM64_RELOC_PAGE21:
-    case ARM64_RELOC_PAGEOFF12:
-      val = r.sym ? r.sym->get_addr(ctx) : r.subsec->get_addr(ctx);
-      break;
-    case ARM64_RELOC_SUBTRACTOR: {
-      Relocation<ARM64> s = rels[++i];
-      assert(s.type == ARM64_RELOC_UNSIGNED);
-      u64 val1 = r.sym ? r.sym->get_addr(ctx) : r.subsec->get_addr(ctx);
-      u64 val2 = s.sym ? s.sym->get_addr(ctx) : s.subsec->get_addr(ctx);
-      val = val2 - val1;
-      break;
-    }
-    case ARM64_RELOC_GOT_LOAD_PAGE21:
-    case ARM64_RELOC_GOT_LOAD_PAGEOFF12:
-    case ARM64_RELOC_POINTER_TO_GOT:
-      val = r.sym->get_got_addr(ctx);
-      break;
-    case ARM64_RELOC_TLVP_LOAD_PAGE21:
-    case ARM64_RELOC_TLVP_LOAD_PAGEOFF12:
-      val = r.sym->get_tlv_addr(ctx);
-      break;
-    default:
-      Fatal(ctx) << isec << ": unknown reloc: " << (int)r.type;
-    }
-
-    val += r.addend;
-
-    switch (r.type) {
-    case ARM64_RELOC_UNSIGNED:
-    case ARM64_RELOC_SUBTRACTOR:
-    case ARM64_RELOC_POINTER_TO_GOT:
-      if (r.is_pcrel)
-        val -= get_addr(ctx) + r.offset;
-
-      if (r.p2size == 2)
-        *(i32 *)(buf + r.offset) = val;
-      else if (r.p2size == 3)
-        *(i64 *)(buf + r.offset) = val;
-      else
-        unreachable();
-      break;
-    case ARM64_RELOC_BRANCH26:
-      if (r.is_pcrel)
-        val -= get_addr(ctx) + r.offset;
-      *(u32 *)(buf + r.offset) |= bits(val, 27, 2);
-      break;
-    case ARM64_RELOC_PAGE21:
-    case ARM64_RELOC_GOT_LOAD_PAGE21:
-    case ARM64_RELOC_TLVP_LOAD_PAGE21:
-      assert(r.is_pcrel);
-      *(u32 *)(buf + r.offset) |=
-        encode_page(page(val) - page(get_addr(ctx) + r.offset));
-      break;
-    case ARM64_RELOC_PAGEOFF12:
-    case ARM64_RELOC_GOT_LOAD_PAGEOFF12:
-    case ARM64_RELOC_TLVP_LOAD_PAGEOFF12: {
-      if (r.is_pcrel)
-        val -= get_addr(ctx) + r.offset;
-
-      u32 insn = *(u32 *)(buf + r.offset);
-      i64 scale = 0;
-      if ((insn & 0x3b000000) == 0x39000000)
-        scale = insn >> 30;
-      *(u32 *)(buf + r.offset) |= bits(val, 11, scale) << 10;
-      break;
-    }
-    default:
-      Fatal(ctx) << isec << ": unknown reloc: " << (int)r.type;
-    }
-  }
-}
-
-} // namespace mold::macho

macho/arch-x86-64.cc

@@ -1,222 +0,0 @@
-#include "mold.h"
-
-namespace mold::macho {
-
-template <>
-void StubsSection<X86_64>::copy_buf(Context<X86_64> &ctx) {
-  u8 *buf = ctx.buf + this->hdr.offset;
-
-  for (i64 i = 0; i < syms.size(); i++) {
-    // `ff 25 xx xx xx xx` is a RIP-relative indirect jump instruction,
-    // i.e., `jmp *IMM(%rip)`. It loads an address from la_symbol_ptr
-    // and jump there.
-    static_assert(X86_64::stub_size == 6);
-    buf[i * 6] = 0xff;
-    buf[i * 6 + 1] = 0x25;
-    *(u32 *)(buf + i * 6 + 2) =
-      ctx.lazy_symbol_ptr.hdr.addr + i * X86_64::word_size -
-      (this->hdr.addr + i * 6 + 6);
-  }
-}
-
-template <>
-void StubHelperSection<X86_64>::copy_buf(Context<X86_64> &ctx) {
-  u8 *start = ctx.buf + this->hdr.offset;
-  u8 *buf = start;
-
-  static const u8 insn0[] = {
-    0x4c, 0x8d, 0x1d, 0, 0, 0, 0, // lea $__dyld_private(%rip), %r11
-    0x41, 0x53,                   // push %r11
-    0xff, 0x25, 0, 0, 0, 0,       // jmp *$dyld_stub_binder@GOT(%rip)
-    0x90,                         // nop
-  };
-
-  static_assert(sizeof(insn0) == X86_64::stub_helper_hdr_size);
-
-  memcpy(buf, insn0, sizeof(insn0));
-  *(u32 *)(buf + 3) =
-    get_symbol(ctx, "__dyld_private")->get_addr(ctx) - this->hdr.addr - 7;
-  *(u32 *)(buf + 11) =
-    get_symbol(ctx, "dyld_stub_binder")->get_got_addr(ctx) - this->hdr.addr - 15;
-
-  buf += 16;
-
-  for (i64 i = 0; i < ctx.stubs.syms.size(); i++) {
-    u8 insn[] = {
-      0x68, 0, 0, 0, 0, // push $bind_offset
-      0xe9, 0, 0, 0, 0, // jmp $__stub_helper
-    };
-
-    static_assert(sizeof(insn) == X86_64::stub_helper_size);
-
-    memcpy(buf, insn, sizeof(insn));
-    *(u32 *)(buf + 1) = ctx.stubs.bind_offsets[i];
-    *(u32 *)(buf + 6) = start - buf - 10;
-    buf += 10;
-  }
-}
-
-static i64 get_reloc_addend(u32 type) {
-  switch (type) {
-  case X86_64_RELOC_SIGNED_1:
-    return 1;
-  case X86_64_RELOC_SIGNED_2:
-    return 2;
-  case X86_64_RELOC_SIGNED_4:
-    return 4;
-  default:
-    return 0;
-  }
-}
-
-static i64 read_addend(u8 *buf, const MachRel &r) {
-  if (r.p2size == 2)
-    return *(i32 *)(buf + r.offset) + get_reloc_addend(r.type);
-  if (r.p2size == 3)
-    return *(i64 *)(buf + r.offset) + get_reloc_addend(r.type);
-  unreachable();
-}
-
-static Relocation<X86_64>
-read_reloc(Context<X86_64> &ctx, ObjectFile<X86_64> &file,
-           const MachSection &hdr, MachRel &r) {
-  if (r.p2size != 2 && r.p2size != 3)
-    Fatal(ctx) << file << ": invalid r.p2size: " << (u32)r.p2size;
-
-  if (r.is_pcrel) {
-    if (r.p2size != 2)
-      Fatal(ctx) << file << ": invalid PC-relative reloc: " << r.offset;
-  } else {
-    if (r.p2size != 3)
-      Fatal(ctx) << file << ": invalid non-PC-relative reloc: " << r.offset;
-  }
-
-  u8 *buf = (u8 *)file.mf->data + hdr.offset;
-  Relocation<X86_64> rel{r.offset, (u8)r.type, (u8)r.p2size, (bool)r.is_pcrel};
-  i64 addend = read_addend(buf, r);
-
-  if (r.is_extern) {
-    rel.sym = file.syms[r.idx];
-    rel.addend = addend;
-    return rel;
-  }
-
-  u32 addr;
-  if (r.is_pcrel)
-    addr = hdr.addr + r.offset + 4 + addend;
-  else
-    addr = addend;
-
-  Subsection<X86_64> *target = file.find_subsection(ctx, addr);
-  if (!target)
-    Fatal(ctx) << file << ": bad relocation: " << r.offset;
-
-  rel.subsec = target;
-  rel.addend = addr - target->input_addr;
-  return rel;
-}
-
-template <>
-std::vector<Relocation<X86_64>>
-read_relocations(Context<X86_64> &ctx, ObjectFile<X86_64> &file,
-                 const MachSection &hdr) {
-  std::vector<Relocation<X86_64>> vec;
-  vec.reserve(hdr.nreloc);
-
-  MachRel *rels = (MachRel *)(file.mf->data + hdr.reloff);
-  for (i64 i = 0; i < hdr.nreloc; i++)
-    vec.push_back(read_reloc(ctx, file, hdr, rels[i]));
-  return vec;
-}
-
-template <>
-void Subsection<X86_64>::scan_relocations(Context<X86_64> &ctx) {
-  for (Relocation<X86_64> &r : get_rels()) {
-    Symbol<X86_64> *sym = r.sym;
-    if (!sym)
-      continue;
-
-    switch (r.type) {
-    case X86_64_RELOC_GOT_LOAD:
-      if (sym->file && sym->file->is_dylib)
-        sym->flags |= NEEDS_GOT;
-      break;
-    case X86_64_RELOC_GOT:
-      sym->flags |= NEEDS_GOT;
-      break;
-    case X86_64_RELOC_TLV:
-      if (sym->file && sym->file->is_dylib)
-        sym->flags |= NEEDS_THREAD_PTR;
-      break;
-    }
-
-    if (sym->file && sym->file->is_dylib) {
-      sym->flags |= NEEDS_STUB;
-      ((DylibFile<X86_64> *)sym->file)->is_needed = true;
-    }
-  }
-}
-
-template <>
-void Subsection<X86_64>::apply_reloc(Context<X86_64> &ctx, u8 *buf) {
-  for (const Relocation<X86_64> &r : get_rels()) {
-    if (r.sym && !r.sym->file) {
-      Error(ctx) << "undefined symbol: " << isec.file << ": " << *r.sym;
-      continue;
-    }
-
-    u64 val = 0;
-
-    switch (r.type) {
-    case X86_64_RELOC_UNSIGNED:
-    case X86_64_RELOC_SIGNED:
-    case X86_64_RELOC_BRANCH:
-    case X86_64_RELOC_SIGNED_1:
-    case X86_64_RELOC_SIGNED_2:
-    case X86_64_RELOC_SIGNED_4:
-      val = r.sym ? r.sym->get_addr(ctx) : r.subsec->get_addr(ctx);
-      break;
-    case X86_64_RELOC_GOT_LOAD:
-      if (r.sym->got_idx != -1) {
-        val = r.sym->get_got_addr(ctx);
-      } else {
-        // Relax MOVQ into LEAQ
-        if (buf[r.offset - 2] != 0x8b)
-          Error(ctx) << isec << ": invalid GOT_LOAD relocation";
-        buf[r.offset - 2] = 0x8d;
-        val = r.sym->get_addr(ctx);
-      }
-      break;
-    case X86_64_RELOC_GOT:
-      val = r.sym->get_got_addr(ctx);
-      break;
-    case X86_64_RELOC_TLV:
-      if (r.sym->tlv_idx != -1) {
-        val = r.sym->get_tlv_addr(ctx);
-      } else {
-        // Relax MOVQ into LEAQ
-        if (buf[r.offset - 2] != 0x8b)
-          Error(ctx) << isec << ": invalid TLV relocation";
-        buf[r.offset - 2] = 0x8d;
-        val = r.sym->get_addr(ctx);
-      }
-      break;
-    default:
-      Fatal(ctx) << isec << ": unknown reloc: " << (int)r.type;
-    }
-
-    val += r.addend;
-
-    if (r.is_pcrel)
-      val -= get_addr(ctx) + r.offset + 4 + get_reloc_addend(r.type);
-
-    if (r.p2size == 2)
-      *(u32 *)(buf + r.offset) = val;
-    else if (r.p2size == 3)
-      *(u64 *)(buf + r.offset) = val;
-    else
-      unreachable();
-  }
-}
-
-} // namespace mold::macho

macho/cmdline.cc

@@ -1,322 +0,0 @@
-#include "mold.h"
-#include "../cmdline.h"
-
-#include <optional>
-#include <sstream>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <regex>
-#include <unistd.h>
-#include <unordered_set>
-
-namespace mold::macho {
-
-static const char helpmsg[] = R"(
-Options:
-  -F<PATH>                    Add DIR to framework search path
-  -L<PATH>                    Add DIR to library search path
-  -Z                          Do not search the standard directories when
-                              searching for libraries and frameworks
-  -ObjC                       Load all static archive members that implement
-                              an Objective-C class or category
-  -adhoc_codesign             Add ad-hoc code signature to the output file
-    -no_adhoc_codesign
-  -arch <ARCH_NAME>           Specify target architecture
-  -bundle                     Produce a mach-o bundle
-  -dead_strip                 Remove unreachable functions and data
-  -dead_strip_dylibs          Remove unreachable dylibs from dependencies
-  -demangle                   Demangle C++ symbols in log messages (default)
-  -dylib                      Produce a dynamic library
-  -dynamic                    Link against dylibs (default)
-  -e <SYMBOL>                 Specify the entry point of a main executable
-  -execute                    Produce an executable (default)
-  -filelist <FILE>[,<DIR>]    Specify the list of input file names
-  -framework <NAME>,[,<SUFFIX>]
-                              Search for a given framework
-  -headerpad <SIZE>           Allocate the size of padding after load commands
-  -headerpad_max_install_names
-                              Allocate MAXPATHLEN byte padding after load commands
-  -help                       Report usage information
-  -l<LIB>                     Search for a given library
-  -lto_library <FILE>         Ignored
-  -map <FILE>                 Write map file to a given file
-  -needed-l<LIB>              Search for a given library
-  -needed-framework <NAME>[,<SUFFIX>]
-                              Search for a given framework
-  -no_deduplicate             Ignored
-  -o <FILE>                   Set output filename
-  -pagezero_size <SIZE>       Specify the size of the __PAGEZERO segment
-  -platform_version <PLATFORM> <MIN_VERSION> <SDK_VERSION>
-                              Set platform, platform version and SDK version
-  -rpath <PATH>               Add PATH to the runpath search path list
-  -syslibroot <DIR>           Prepend DIR to library search paths
-  -t                          Print out each file the linker loads
-  -v                          Report version information)";
-
-template <typename E>
-static i64 parse_platform(Context<E> &ctx, std::string_view arg) {
-  static std::regex re(R"(\d+)", std::regex_constants::ECMAScript);
-  if (std::regex_match(arg.begin(), arg.end(), re))
-    return stoi(std::string(arg));
-
-  if (arg == "macos")
-    return PLATFORM_MACOS;
-  if (arg == "ios")
-    return PLATFORM_IOS;
-  if (arg == "tvos")
-    return PLATFORM_TVOS;
-  if (arg == "watchos")
-    return PLATFORM_WATCHOS;
-  if (arg == "bridgeos")
-    return PLATFORM_BRIDGEOS;
-  if (arg == "mac-catalyst")
-    return PLATFORM_MACCATALYST;
-  if (arg == "ios-simulator")
-    return PLATFORM_IOSSIMULATOR;
-  if (arg == "tvos-simulator")
-    return PLATFORM_TVOSSIMULATOR;
-  if (arg == "watchos-simulator")
-    return PLATFORM_WATCHOSSIMULATOR;
-  if (arg == "driverkit")
-    return PLATFORM_DRIVERKIT;
-  Fatal(ctx) << "unknown -platform_version name: " << arg;
-}
-
-template <typename E>
-static i64 parse_version(Context<E> &ctx, std::string_view arg) {
-  static std::regex re(R"((\d+)(?:\.(\d+))?(?:\.(\d+))?)",
-                       std::regex_constants::ECMAScript);
-  std::cmatch m;
-  if (!std::regex_match(arg.begin(), arg.end(), m, re))
-    Fatal(ctx) << "malformed version number: " << arg;
-
-  i64 major = (m[1].length() == 0) ? 0 : stoi(m[1]);
-  i64 minor = (m[2].length() == 0) ? 0 : stoi(m[2]);
-  i64 patch = (m[3].length() == 0) ? 0 : stoi(m[3]);
-  return (major << 16) | (minor << 8) | patch;
-}
-
-static bool is_directory(std::filesystem::path path) {
-  std::error_code ec;
-  return std::filesystem::is_directory(path, ec) && !ec;
-}
-
-template <typename E>
-void parse_nonpositional_args(Context<E> &ctx,
-                              std::vector<std::string> &remaining) {
-  std::vector<std::string_view> &args = ctx.cmdline_args;
-  i64 i = 1;
-
-  std::vector<std::string> framework_paths;
-  std::vector<std::string> library_paths;
-  bool nostdlib = false;
-  std::optional<i64> pagezero_size;
-
-  while (i < args.size()) {
-    std::string_view arg;
-    std::string_view arg2;
-    std::string_view arg3;
-
-    auto read_arg = [&](std::string name) {
-      if (args[i] == name) {
-        if (args.size() <= i + 1)
-          Fatal(ctx) << "option -" << name << ": argument missing";
-        arg = args[i + 1];
-        i += 2;
-        return true;
-      }
-      return false;
-    };
-
-    auto read_arg3 = [&](std::string name) {
-      if (args[i] == name) {
-        if (args.size() <= i + 3)
-          Fatal(ctx) << "option -" << name << ": argument missing";
-        arg = args[i + 1];
-        arg2 = args[i + 2];
-        arg3 = args[i + 3];
-        i += 4;
-        return true;
-      }
-      return false;
-    };
-
-    auto read_joined = [&](std::string name) {
-      if (read_arg(name))
-        return true;
-      if (args[i].starts_with(name)) {
-        arg = args[i].substr(name.size());
-        i++;
-        return true;
-      }
-      return false;
-    };
-
-    auto read_flag = [&](std::string name) {
-      if (args[i] == name) {
-        i++;
-        return true;
-      }
-      return false;
-    };
-
-    if (args[i].starts_with('@')) {
-      std::vector<std::string_view> vec =
-        read_response_file(ctx, args[i].substr(1));
-      args.erase(args.begin() + i);
-      args.insert(args.begin() + i, vec.begin(), vec.end());
-      continue;
-    }
-
-    if (read_flag("-help") || read_flag("--help")) {
-      SyncOut(ctx) << "Usage: " << ctx.cmdline_args[i]
-                   << " [options] file...\n" << helpmsg;
-      exit(0);
-    }
-
-    if (read_joined("-F")) {
-      framework_paths.push_back(std::string(arg));
-    } else if (read_joined("-L")) {
-      library_paths.push_back(std::string(arg));
-    } else if (read_flag("-Z")) {
-      nostdlib = true;
-    } else if (read_flag("-ObjC")) {
-      ctx.arg.ObjC = true;
-    } else if (read_flag("-adhoc_codesign")) {
-      ctx.arg.adhoc_codesign = true;
-    } else if (read_flag("-no_adhoc_codesign")) {
-      ctx.arg.adhoc_codesign = false;
-    } else if (read_arg("-arch")) {
-      if (arg == "x86_64")
-        ctx.arg.arch = CPU_TYPE_X86_64;
-      else if (arg == "arm64")
-        ctx.arg.arch = CPU_TYPE_ARM64;
-      else
-        Fatal(ctx) << "unknown -arch: " << arg;
-    } else if (read_flag("-bundle")) {
-      ctx.output_type = MH_BUNDLE;
-    } else if (read_flag("-color-diagnostics") ||
-               read_flag("--color-diagnostics")) {
-    } else if (read_flag("-dead_strip")) {
-      ctx.arg.dead_strip = true;
-    } else if (read_flag("-dead_strip_dylibs")) {
-      ctx.arg.dead_strip_dylibs = true;
-    } else if (read_flag("-demangle")) {
-      ctx.arg.demangle = true;
-    } else if (read_flag("-dylib")) {
-      ctx.output_type = MH_DYLIB;
-    } else if (read_arg("-headerpad")) {
-      size_t pos;
-      ctx.arg.headerpad = std::stol(std::string(arg), &pos, 16);
-      if (pos != arg.size())
-        Fatal(ctx) << "malformed -headerpad: " << arg;
-    } else if (read_flag("-headerpad_max_install_names")) {
-      ctx.arg.headerpad = 1024;
-    } else if (read_flag("-dynamic")) {
-      ctx.arg.dynamic = true;
-    } else if (read_arg("-e")) {
-      ctx.arg.entry = arg;
-    } else if (read_flag("-execute")) {
-      ctx.output_type = MH_EXECUTE;
-    } else if (read_arg("-fatal_warnings")) {
-    } else if (read_arg("-filelist")) {
-      remaining.push_back("-filelist");
-      remaining.push_back(std::string(arg));
-    } else if (read_arg("-framework")) {
-      remaining.push_back("-framework");
-      remaining.push_back(std::string(arg));
-    } else if (read_arg("-lto_library")) {
-    } else if (read_joined("-l")) {
-      remaining.push_back("-l");
-      remaining.push_back(std::string(arg));
-    } else if (read_arg("-map")) {
-      ctx.arg.map = arg;
-    } else if (read_joined("-needed-l")) {
-      remaining.push_back("-needed-l");
-      remaining.push_back(std::string(arg));
-    } else if (read_arg("-needed_framework")) {
-      remaining.push_back("-needed_framework");
-      remaining.push_back(std::string(arg));
-    } else if (read_flag("-no_deduplicate")) {
-    } else if (read_arg("-o")) {
-      ctx.arg.output = arg;
-    } else if (read_arg("-pagezero_size")) {
-      size_t pos;
-      pagezero_size = std::stol(std::string(arg), &pos, 16);
-      if (pos != arg.size())
-        Fatal(ctx) << "malformed -pagezero_size: " << arg;
-    } else if (read_arg3("-platform_version")) {
-      ctx.arg.platform = parse_platform(ctx, arg);
-      ctx.arg.platform_min_version = parse_version(ctx, arg2);
-      ctx.arg.platform_sdk_version = parse_version(ctx, arg3);
-    } else if (read_arg("-rpath")) {
-      ctx.arg.rpath.push_back(std::string(arg));
-    } else if (read_flag("-search_dylibs_first")) {
-      Fatal(ctx) << "-search_dylibs_first is not supported";
-    } else if (read_flag("-search_paths_first")) {
-    } else if (read_arg("-syslibroot")) {
-      ctx.arg.syslibroot.push_back(std::string(arg));
-    } else if (read_flag("-t")) {
-      ctx.arg.trace = true;
-    } else if (read_flag("-v")) {
-      SyncOut(ctx) << mold_version;
-    } else {
-      if (args[i][0] == '-')
-        Fatal(ctx) << "unknown command line option: " << args[i];
-      remaining.push_back(std::string(args[i]));
-      i++;
-    }
-  }
-
-  auto add_search_path = [&](std::vector<std::string> &vec, std::string path) {
-    if (!path.starts_with('/') || ctx.arg.syslibroot.empty()) {
-      if (is_directory(path))
-        vec.push_back(path);
-      return;
-    }
-
-    bool found = false;
-    for (std::string &dir : ctx.arg.syslibroot) {
-      std::string str = path_clean(dir + "/" + path);
-      if (is_directory(str)) {
-        vec.push_back(str);
-        found = true;
-      }
-    }
-    if (!found && is_directory(path))
-      vec.push_back(path);
-  };
-
-  for (std::string &path : library_paths)
-    add_search_path(ctx.arg.library_paths, path);
-
-  if (!nostdlib) {
-    add_search_path(ctx.arg.library_paths, "/usr/lib");
-    add_search_path(ctx.arg.library_paths, "/usr/local/lib");
-  }
-
-  for (std::string &path : framework_paths)
-    add_search_path(ctx.arg.framework_paths, path);
-
-  if (!nostdlib) {
-    add_search_path(ctx.arg.framework_paths, "/Library/Frameworks");
-    add_search_path(ctx.arg.framework_paths, "/System/Library/Frameworks");
-  }
-
-  if (pagezero_size.has_value()) {
-    if (ctx.output_type != MH_EXECUTE)
-      Error(ctx) << "-pagezero_size option can only be used when"
-                 << " linking a main executable";
-    ctx.arg.pagezero_size = *pagezero_size;
-  } else {
-    ctx.arg.pagezero_size = (ctx.output_type == MH_EXECUTE) ? 0x100000000 : 0;
-  }
-}
-
-#define INSTANTIATE(E) \
-  template void parse_nonpositional_args(Context<E> &, std::vector<std::string> &)
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/dead-strip.cc

@@ -1,112 +0,0 @@
-#include "mold.h"
-
-namespace mold::macho {
-
-template <typename E>
-static std::vector<Subsection<E> *> collect_root_set(Context<E> &ctx) {
-  std::vector<Subsection<E> *> rootset;
-
-  auto mark = [&](Symbol<E> *sym) {
-    if (sym && sym->subsec)
-      rootset.push_back(sym->subsec);
-  };
-
-  mark(get_symbol(ctx, ctx.arg.entry));
-
-  if (ctx.output_type == MH_DYLIB || ctx.output_type == MH_BUNDLE)
-    for (ObjectFile<E> *file : ctx.objs)
-      for (Symbol<E> *sym : file->syms)
-        if (sym->file == file && sym->is_extern)
-          mark(sym);
-
-  return rootset;
-}
-
-template <typename E>
-static void visit(Context<E> &ctx, Subsection<E> &subsec) {
-  if (subsec.is_alive.exchange(true))
-    return;
-
-  for (Relocation<E> &rel : subsec.get_rels()) {
-    if (rel.sym) {
-      if (rel.sym->subsec)
-        visit(ctx, *rel.sym->subsec);
-    } else {
-      visit(ctx, *rel.subsec);
-    }
-  }
-
-  for (UnwindRecord<E> &rec : subsec.get_unwind_records()) {
-    rec.is_alive = true;
-    visit(ctx, *rec.subsec);
-    if (rec.lsda)
-      visit(ctx, *rec.lsda);
-    if (Symbol<E> *sym = rec.personality; sym && sym->subsec)
-      visit(ctx, *sym->subsec);
-  }
-}
-
-template <typename E>
-static bool refers_live_subsection(Subsection<E> &subsec) {
-  for (Relocation<E> &rel : subsec.get_rels()) {
-    if (rel.sym) {
-      if (!rel.sym->subsec || rel.sym->subsec->is_alive)
-        return true;
-    } else {
-      if (rel.subsec->is_alive)
-        return true;
-    }
-  }
-  return false;
-}
-
-template <typename E>
-static void mark(Context<E> &ctx, const std::vector<Subsection<E> *> &rootset) {
-  for (Subsection<E> *subsec : rootset)
-    visit(ctx, *subsec);
-
-  bool repeat;
-  do {
-    repeat = false;
-    for (ObjectFile<E> *file : ctx.objs) {
-      for (std::unique_ptr<Subsection<E>> &subsec : file->subsections) {
-        if ((subsec->isec.hdr.attr & S_ATTR_LIVE_SUPPORT) &&
-            !subsec->is_alive &&
-            refers_live_subsection(*subsec)) {
-          visit(ctx, *subsec);
-          repeat = true;
-        }
-      }
-    }
-  } while (repeat);
-}
-
-template <typename E>
-static void sweep(Context<E> &ctx) {
-  for (ObjectFile<E> *file : ctx.objs) {
-    std::erase_if(file->subsections,
-                  [](const std::unique_ptr<Subsection<E>> &subsec) {
-      return !subsec->is_alive;
-    });
-  }
-
-  for (ObjectFile<E> *file : ctx.objs)
-    for (Symbol<E> *&sym : file->syms)
-      if (sym->file == file && sym->subsec && !sym->subsec->is_alive)
-        sym = nullptr;
-}
-
-template <typename E>
-void dead_strip(Context<E> &ctx) {
-  std::vector<Subsection<E> *> rootset = collect_root_set(ctx);
-  mark(ctx, rootset);
-  sweep(ctx);
-}
-
-#define INSTANTIATE(E)                          \
-  template void dead_strip(Context<E> &)
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/dumper.cc

@@ -1,511 +0,0 @@
-#include "mold.h"
-
-#include <cstdlib>
-#include <fcntl.h>
-#include <iomanip>
-#include <iostream>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-namespace mold::macho {
-
-static u8 *open_file(std::string path) {
-  i64 fd = ::open(path.c_str(), O_RDONLY);
-  if (fd == -1)
-    return nullptr;
-
-  struct stat st;
-  if (fstat(fd, &st) == -1) {
-    std::cerr << path << ": fstat failed: " << errno_string();
-    exit(1);
-  }
-
-  void *ptr = mmap(nullptr, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
-  if (ptr == MAP_FAILED) {
-    std::cerr << path << ": mmap failed: " << errno_string();
-    exit(1);
-  }
-
-  close(fd);
-  return (u8 *)ptr;
-}
-
-static void print_bytes(u8 *buf, i64 size) {
-  if (size == 0) {
-    std::cout << "[]\n";
-    return;
-  }
-
-  std::cout << "[" << std::setw(2) << std::setfill('0') << (u32)buf[0];
-  for (i64 i = 1; i < size; i++)
-    std::cout << " " << std::setw(2) << std::setfill('0') << (u32)buf[i];
-  std::cout << "]\n";
-}
-
-struct ExportEntry {
-  std::string name;
-  u32 flags;
-  u64 addr;
-};
-
-static void read_trie(std::vector<ExportEntry> &vec, u8 *start, i64 offset = 0,
-                      const std::string &prefix = "") {
-  u8 *buf = start + offset;
-
-  if (read_uleb(buf)) {
-    ExportEntry ent;
-    ent.name = prefix;
-    ent.flags = read_uleb(buf);
-    ent.addr = read_uleb(buf);
-    vec.push_back(std::move(ent));
-  }
-
-  for (i64 i = 0, n = read_uleb(buf); i < n; i++) {
-    std::string suffix((char *)buf);
-    buf += suffix.size() + 1;
-    i64 off = read_uleb(buf);
-    read_trie(vec, start, off, prefix + suffix);
-  }
-}
-
-void dump_unwind_info(u8 *buf, MachSection &sec) {
-  UnwindSectionHeader &hdr = *(UnwindSectionHeader *)(buf + sec.offset);
-
-  std::cout << std::hex << "  Unwind info:"
-            << "\n   version: 0x"   << hdr.version
-            << "\n   encoding_offset: 0x"   << hdr.encoding_offset
-            << "\n   encoding_count: 0x"   << hdr.encoding_count
-            << "\n   personality_offset: 0x"   << hdr.personality_offset
-            << "\n   personality_count: 0x"   << hdr.personality_count
-            << "\n   page_offset: 0x"   << hdr.page_offset
-            << "\n   page_count: 0x"   << hdr.page_count;
-
-  std::cout << "\n   encoding:";
-  u32 *enc = (u32 *)(buf + sec.offset + hdr.encoding_offset);
-  for (i64 i = 0; i < hdr.encoding_count; i++)
-    std::cout << std::hex << "\n    0x" << enc[i];
-
-  std::cout << "\n   personality:";
-  u32 *pers = (u32 *)(buf + sec.offset + hdr.personality_offset);
-  for (i64 i = 0; i < hdr.personality_count; i++)
-    std::cout << std::hex << "\n    0x" << pers[i];
-
-  UnwindFirstLevelPage *ent =
-    (UnwindFirstLevelPage *)(buf + sec.offset + hdr.page_offset);
-
-  for (i64 i = 0; i < hdr.page_count; i++) {
-    std::cout << std::hex << "\n   function:"  
-              << "\n    func_addr: 0x"   << ent[i].func_addr
-              << "\n    page_offset: 0x"   << ent[i].page_offset
-              << "\n    lsda_offset: 0x"   << ent[i].lsda_offset;
-
-    if (i != hdr.page_count - 1) {
-      UnwindLsdaEntry *lsda =
-        (UnwindLsdaEntry *)(buf + sec.offset + ent[i].lsda_offset);
-      i64 lsda_size = ent[i + 1].lsda_offset - ent[i].lsda_offset;
-      for (i64 j = 0; j < lsda_size / sizeof(UnwindLsdaEntry); j++)
-        std::cout << std::hex
-                  << "\n    lsda:"
-                  << "\n     func_addr: 0x" << lsda[j].func_addr
-                  << "\n     lsda_addr: 0x" << lsda[j].lsda_addr;
-    }
-
-    if (ent[i].page_offset == 0)
-      break;
-
-    u8 *addr = buf + sec.offset + ent[i].page_offset;
-
-    switch (u32 kind = *addr; kind) {
-    case UNWIND_SECOND_LEVEL_REGULAR: {
-      std::cout << "\n    UNWIND_SECOND_LEVEL_REGULAR:"  ;
-      break;
-    }
-    case UNWIND_SECOND_LEVEL_COMPRESSED: {
-      std::cout << "\n    UNWIND_SECOND_LEVEL_COMPRESSED"  ;
-      UnwindSecondLevelPage &hdr2 = *(UnwindSecondLevelPage *)addr;
-      std::cout << std::hex
-                << "\n     page_offset: 0x"   << hdr2.page_offset
-                << "\n     page_count: 0x"   << hdr2.page_count
-                << "\n     encoding_offset: 0x"   << hdr2.encoding_offset
-                << "\n     encoding_count: 0x"   << hdr2.encoding_count;
-
-      UnwindPageEntry *ent2 = (UnwindPageEntry *)(addr + hdr2.page_offset);
-      for (i64 j = 0; j < hdr2.page_count; j++)
-        std::cout << std::hex << "\n      ent 0x"  
-                  << (ent[i].func_addr + ent2[j].func_addr)
-                  << " 0x" << ent2[j].encoding;
-
-      u32 *enc = (u32 *)(addr + hdr2.encoding_offset);
-      for (i64 j = 0; j < hdr2.encoding_count; j++)
-        std::cout << std::hex << "\n      0x"   << enc[j];
-      break;
-    }
-    default:
-      std::cout << "\n    bad 2nd-level unwind info header: " << kind;
-    }
-  }
-
-  std::cout << "\n";
-}
-
-void dump_compact_unwind(u8 *buf, MachSection &sec) {
-  CompactUnwindEntry *ent = (CompactUnwindEntry *)(buf + sec.offset);
-  i64 nentry = sec.size / sizeof(CompactUnwindEntry);
-
-  std::cout << "  Compact unwind:"
-            << "\n   num_entry: " << nentry;
-
-  for (i64 i = 0; i < nentry; i++) {
-    std::cout << std::hex
-              << "\n   entry: 0x" << (i * sizeof(CompactUnwindEntry))
-              << "\n    code_start: 0x" << ent[i].code_start
-              << "\n    code_len: 0x" << ent[i].code_len
-              << "\n    encoding: 0x" << ent[i].encoding
-              << "\n    personality: 0x" << ent[i].personality
-              << "\n    lsda: 0x" << ent[i].lsda;
-  }
-
-  std::cout << "\n";
-}
-
-void dump_file(std::string path) {
-  u8 *buf = open_file(path);
-  if (!buf) {
-    std::cerr << "cannot open " << path << "\n";
-    exit(1);
-  }
-
-  MachHeader &hdr = *(MachHeader *)buf;
-  std::cout << "magic: 0x" << std::hex << hdr.magic
-            << "\ncputype: 0x" << hdr.cputype
-            << "\ncpusubtype: 0x" << hdr.cpusubtype
-            << "\nfiletype: 0x" << hdr.filetype
-            << "\nncmds: 0x" << hdr.ncmds
-            << "\nsizeofcmds: 0x" << hdr.sizeofcmds
-            << "\nflags: 0x" << hdr.flags
-            << "\n\n";
-
-  u8 *p = buf + sizeof(MachHeader);
-
-  for (i64 i = 0; i < hdr.ncmds; i++) {
-    std::cout << "fileoff: 0x" << std::hex << (p - buf) << "\n";
-
-    LoadCommand &lc = *(LoadCommand *)p;
-    p += lc.cmdsize;
-
-    auto print_dylib = [&]() {
-      DylibCommand &cmd = *(DylibCommand *)&lc;
-      std::cout << " cmdsize: 0x" << cmd.cmdsize
-                << "\n nameoff: 0x" << cmd.nameoff
-                << "\n timestamp: 0x" << cmd.timestamp
-                << "\n current_version: 0x" << cmd.current_version
-                << "\n compatibility_version: 0x" << cmd.compatibility_version
-                << "\n data: " << (char *)((char *)&cmd + cmd.nameoff)
-                << "\n";
-    };
-
-    switch (lc.cmd) {
-    case LC_SYMTAB: {
-      std::cout << "LC_SYMTAB\n";
-      SymtabCommand &cmd = *(SymtabCommand *)&lc;
-      std::cout << " cmdsize: " << cmd.cmdsize
-                << "\n symoff: 0x" << std::hex << cmd.symoff
-                << "\n nsyms: " << std::dec << cmd.nsyms
-                << "\n stroff: 0x" << std::hex << cmd.stroff
-                << "\n strsize: 0x" << cmd.strsize
-                << "\n symdata: ";
-      print_bytes(buf + cmd.symoff, cmd.nsyms * sizeof(MachSym));
-
-      std::cout << " strdata: ";
-      print_bytes(buf + cmd.stroff, cmd.strsize);
-
-      MachSym *syms = (MachSym *)(buf + cmd.symoff);
-      for (i64 j = 0; j < cmd.nsyms; j++) {
-        std::cout << " symbol:"
-                  << "\n  name: " << (char *)(buf + cmd.stroff + syms[j].stroff)
-                  << "\n  stub: " << (u32)syms[j].stub
-                  << "\n  pext: " << (u32)syms[j].pext
-                  << "\n  type: " << (u32)syms[j].type
-                  << "\n  ext: " << (u32)syms[j].ext
-                  << "\n  sect: 0x" << (u32)syms[j].sect
-                  << "\n  desc: 0x" << (u32)syms[j].desc
-                  << "\n  value: 0x" << syms[j].value
-                  << "\n";
-      }
-      break;
-    }
-    case LC_DYSYMTAB: {
-      std::cout << "LC_DYSYMTAB\n";
-      DysymtabCommand &cmd = *(DysymtabCommand *)&lc;
-      std::cout << " cmdsize: 0x" << cmd.cmdsize
-                << "\n ilocalsym: 0x" << cmd.ilocalsym
-                << "\n nlocalsym: 0x" << cmd.nlocalsym
-                << "\n iextdefsym: 0x" << cmd.iextdefsym
-                << "\n nextdefsym: 0x" << cmd.nextdefsym
-                << "\n iundefsym: 0x" << cmd.iundefsym
-                << "\n nundefsym: 0x" << cmd.nundefsym
-                << "\n tocoff: 0x" << cmd.tocoff
-                << "\n ntoc: 0x" << cmd.ntoc
-                << "\n modtaboff: 0x" << cmd.modtaboff
-                << "\n nmodtab: 0x" << cmd.nmodtab
-                << "\n extrefsymoff: 0x" << cmd.extrefsymoff
-                << "\n nextrefsyms: 0x" << cmd.nextrefsyms
-                << "\n indirectsymoff: 0x" << cmd.indirectsymoff
-                << "\n nindirectsyms: 0x" << cmd.nindirectsyms
-                << "\n extreloff: 0x" << cmd.extreloff
-                << "\n nextrel: 0x" << cmd.nextrel
-                << "\n locreloff: 0x" << cmd.locreloff
-                << "\n nlocrel: 0x" << cmd.nlocrel
-                << "\n";
-
-      if (cmd.indirectsymoff) {
-        std::cout << " indirectsymdata: ";
-        print_bytes(buf + cmd.indirectsymoff, 4 * cmd.nindirectsyms);
-      }
-      break;
-    }
-    case LC_LOAD_DYLIB:
-      std::cout << "LC_LOAD_DYLIB\n";
-      print_dylib();
-      break;
-    case LC_LOAD_WEAK_DYLIB:
-      std::cout << "LC_LOAD_WEAK_DYLIB\n";
-      print_dylib();
-      break;
-    case LC_ID_DYLIB:
-      std::cout << "LC_ID_DYLIB\n";
-      print_dylib();
-      break;
-    case LC_LOAD_DYLINKER: {
-      std::cout << "LC_LOAD_DYLINKER\n";
-      DylinkerCommand &cmd = *(DylinkerCommand *)&lc;
-      std::cout << " cmdsize: 0x" << cmd.cmdsize
-                << "\n nameoff: 0x" << cmd.nameoff
-                << "\n data: " << (char *)((char *)&cmd + cmd.nameoff)
-                << "\n";
-      break;
-    }
-    case LC_SEGMENT_64: {
-      std::cout << "LC_SEGMENT_64\n";
-      SegmentCommand &cmd = *(SegmentCommand *)&lc;
-      std::cout << " cmdsize: " << cmd.cmdsize
-                << "\n segname: " << cmd.get_segname()
-                << "\n vmaddr: 0x" << std::hex << cmd.vmaddr
-                << "\n vmsize: 0x" << cmd.vmsize
-                << "\n fileoff: 0x" << cmd.fileoff
-                << "\n filesize: 0x" << cmd.filesize
-                << "\n maxprot: " << std::dec << cmd.maxprot
-                << "\n initprot: " << cmd.initprot
-                << "\n nsects: " << cmd.nsects
-                << "\n flags: 0x" << std::hex << cmd.flags
-                << "\n";
-
-      MachSection *sec = (MachSection *)((u8 *)&lc + sizeof(cmd));
-      for (i64 j = 0; j < cmd.nsects; j++) {
-        std::cout << " section:\n  sectname: " << sec[j].get_sectname()
-                  << "\n  segname: " << sec[j].get_segname()
-                  << "\n  addr: 0x" << std::hex << sec[j].addr
-                  << "\n  size: 0x" << sec[j].size
-                  << "\n  offset: 0x" << sec[j].offset
-                  << "\n  p2align: " << std::dec << sec[j].p2align
-                  << "\n  reloff: " << std::hex << sec[j].reloff
-                  << "\n  nreloc: " << std::dec << sec[j].nreloc
-                  << "\n  type: 0x" << std::hex << sec[j].type
-                  << "\n  attr: 0x" << std::hex << sec[j].attr
-                  << "\n";
-
-        if (sec[j].type != S_ZEROFILL) {
-          std::cout << "  contents: ";
-          print_bytes(buf + sec[j].offset, sec[j].size);
-        }
-
-        if (sec[j].reloff) {
-          MachRel *rel = (MachRel *)(buf + sec[j].reloff);
-          for (i64 k = 0; k < sec[j].nreloc; k++) {
-            std::cout << "  reloc: "
-                      << "\n   offset: 0x" << rel[k].offset
-                      << "\n   idx: 0x" << rel[k].idx
-                      << "\n   is_pcrel: " << rel[k].is_pcrel
-                      << "\n   p2size: 0x" << rel[k].p2size
-                      << "\n   is_extern: " << rel[k].is_extern
-                      << "\n   type: " << rel[k].type
-                      << "\n";
-          }
-        }
-
-        if (sec[j].match("__TEXT", "__unwind_info"))
-          dump_unwind_info(buf, sec[j]);
-
-        if (sec[j].match("__LD", "__compact_unwind"))
-          dump_compact_unwind(buf, sec[j]);
-      }
-      break;
-    }
-    case LC_UUID: {
-      std::cout << "LC_UUID\n";
-      UUIDCommand &cmd = *(UUIDCommand *)&lc;
-      std::cout << " data: ";
-      print_bytes(cmd.uuid, sizeof(cmd.uuid));
-      break;
-    }
-    case LC_DYLD_INFO_ONLY: {
-      std::cout << "LC_DYLD_INFO_ONLY\n";
-      DyldInfoCommand &cmd = *(DyldInfoCommand *)&lc;
-
-      if (cmd.rebase_off) {
-        std::cout << "  rebase: ";
-        print_bytes(buf + cmd.rebase_off, cmd.rebase_size);
-        std::cout << "  rebase_off: 0x" << std::hex << cmd.rebase_off
-                  << "\n  rebase_size: 0x" << cmd.rebase_size
-                  << "\n";
-      }
-
-      if (cmd.bind_off) {
-        std::cout << "  bind: ";
-        print_bytes(buf + cmd.bind_off, cmd.bind_size);
-        std::cout << "  bind_off: 0x" << std::hex << cmd.bind_off
-                  << "\n  bind_size: 0x" << cmd.bind_size
-                  << "\n";
-      }
-
-      if (cmd.weak_bind_off) {
-        std::cout << "  weak_bind: ";
-        print_bytes(buf + cmd.weak_bind_off, cmd.weak_bind_size);
-        std::cout << "  weak_bind_off: 0x" << std::hex << cmd.weak_bind_off
-                  << "\n  weak_bind_size: 0x" << cmd.weak_bind_size
-                  << "\n";
-      }
-
-      if (cmd.lazy_bind_off) {
-        std::cout << "  lazy_bind: ";
-        print_bytes(buf + cmd.lazy_bind_off, cmd.lazy_bind_size);
-        std::cout << "  lazy_bind_off: 0x" << std::hex << cmd.lazy_bind_off
-                  << "\n  lazy_bind_size: 0x" << cmd.lazy_bind_size
-                  << "\n";
-      }
-
-      if (cmd.export_off) {
-        std::cout << "  export: ";
-        print_bytes(buf + cmd.export_off, cmd.export_size);
-        std::cout << "  export_off: 0x" << std::hex << cmd.export_off
-                  << "\n  export_size: 0x" << cmd.export_size
-                  << "\n";
-
-        std::vector<ExportEntry> vec;
-        read_trie(vec, buf + cmd.export_off);
-        for (ExportEntry &ent : vec)
-          std::cout << "  export_sym: " << ent.name << " 0x" << ent.addr << "\n";
-      }
-      break;
-    }
-    case LC_FUNCTION_STARTS: {
-      std::cout << "LC_FUNCTION_STARTS\n";
-      LinkEditDataCommand &cmd = *(LinkEditDataCommand *)&lc;
-      std::cout << " dataoff: 0x" << cmd.dataoff
-                << "\n datasize: 0x" << cmd.datasize
-                << "\n data:";
-
-      u8 *p = buf + cmd.dataoff;
-      u64 addr = 0;
-      for (;;) {
-        u64 delta = read_uleb(p);
-        if (!delta)
-          break;
-        addr += delta;
-        std::cout << std::hex << " 0x" << addr;
-      }
-      std::cout << "\n";
-      break;
-    }
-    case LC_MAIN: {
-      std::cout << "LC_MAIN\n";
-      EntryPointCommand &cmd = *(EntryPointCommand *)&lc;
-      std::cout << " cmdsize: 0x" << cmd.cmdsize
-                << "\n entryoff: 0x" << cmd.entryoff
-                << "\n stacksize: 0x" << cmd.stacksize
-                << "\n";
-      break;
-    }
-    case LC_DATA_IN_CODE: {
-      std::cout << "LC_DATA_IN_CODE\n";
-      LinkEditDataCommand &cmd = *(LinkEditDataCommand *)&lc;
-      std::cout << " dataoff: 0x" << cmd.dataoff
-                << "\n datasize: 0x" << cmd.datasize
-                << "\n";
-      break;
-    }
-    case LC_SOURCE_VERSION: {
-      std::cout << "LC_SOURCE_VERSION\n";
-      SourceVersionCommand &cmd = *(SourceVersionCommand *)&lc;
-      std::cout << " version: 0x" << cmd.version
-                << "\n";
-      break;
-    }
-      break;
-    case LC_BUILD_VERSION: {
-      std::cout << "LC_BUILD_VERSION\n";
-      BuildVersionCommand &cmd = *(BuildVersionCommand *)&lc;
-      std::cout << " cmdsize: 0x" << cmd.cmdsize
-                << "\n platform: 0x" << cmd.platform
-                << "\n minos: 0x" << cmd.minos
-                << "\n sdk: 0x" << cmd.sdk
-                << "\n ntools: 0x" << cmd.ntools
-                << "\n";
-
-      BuildToolVersion *tools =
-        (BuildToolVersion *)((u8 *)&lc + sizeof(BuildVersionCommand));
-      for (i64 i = 0; i < cmd.ntools; i++)
-        std::cout << "  tool: 0x" << tools[i].tool
-                  << "\n  version: 0x" << tools[i].version
-                  << "\n";
-      break;
-    }
-    case LC_VERSION_MIN_MACOSX: {
-      std::cout << "LC_VERSION_MIN_MACOSX\n";
-      VersionMinCommand &cmd = *(VersionMinCommand *)&lc;
-      std::cout << " version: " << (int)cmd.version
-                << "\n sdk: " << (int)cmd.sdk
-                << "\n";
-      break;
-    }
-    case LC_CODE_SIGNATURE: {
-      std::cout << "LC_CODE_SIGNATURE\n";
-      LinkEditDataCommand &cmd = *(LinkEditDataCommand *)&lc;
-
-      CodeSignatureHeader &sig = *(CodeSignatureHeader *)(buf + cmd.dataoff);
-      std::cout << " magic: " << sig.magic
-                << "\n length: " << sig.length
-                << "\n count: " << sig.count
-                << "\n";
-
-      for (i64 i = 0; i < sig.count; i++) {
-        CodeSignatureBlobIndex &idx = ((CodeSignatureBlobIndex *)(&sig + 1))[i];
-        std::cout << " idx type: " << idx.type
-                  << "\n idx offset: " << idx.offset
-                  << "\n";
-
-        CodeSignatureDirectory &dir =
-          *(CodeSignatureDirectory *)((u8 *)&sig + idx.offset);
-        std::cout << std::hex
-                  << " magic: 0x" << dir.magic
-                  << "\n version: 0x" << dir.version
-                  << "\n flags: 0x" << dir.flags
-                  << "\n hash_offset: 0x" << dir.hash_offset
-                  << "\n n_code_slots: 0x" << dir.n_code_slots
-                  << "\n hash_size: 0x" << (u32)dir.hash_size
-                  << "\n hash_type: 0x" << (u32)dir.hash_type
-                  << "\n page_size: 0x" << (1 << dir.page_size)
-                  << "\n";
-      }
-
-      break;
-    }
-    default:
-      std::cout << "UNKNOWN (0x" << std::hex << lc.cmd << ")\n";
-      break;
-    }
-  }
-}
-
-} // namespace mold::macho

macho/input-sections.cc

@@ -1,60 +0,0 @@
-#include "mold.h"
-
-namespace mold::macho {
-
-template <typename E>
-InputSection<E>::InputSection(Context<E> &ctx, ObjectFile<E> &file,
-                              const MachSection &hdr)
-  : file(file), hdr(hdr),
-    osec(*OutputSection<E>::get_instance(ctx, hdr.get_segname(),
-                                         hdr.get_sectname())) {
-  if (hdr.type != S_ZEROFILL)
-    contents = file.mf->get_contents().substr(hdr.offset, hdr.size);
-}
-
-template <typename E>
-void InputSection<E>::parse_relocations(Context<E> &ctx) {
-  rels.reserve(hdr.nreloc);
-
-  // Parse mach-o relocations to fill `rels` vector
-  rels = read_relocations(ctx, file, hdr);
-
-  // Sort `rels` vector
-  sort(rels, [](const Relocation<E> &a, const Relocation<E> &b) {
-    return a.offset < b.offset;
-  });
-
-  // Find subsections for this section
-  auto begin = std::lower_bound(
-      file.subsections.begin(), file.subsections.end(), hdr.addr,
-      [](std::unique_ptr<Subsection<E>> &subsec, u32 addr) {
-    return subsec->input_addr < addr;
-  });
-
-  auto end = std::lower_bound(
-      begin, file.subsections.end(), hdr.addr + hdr.size,
-      [](std::unique_ptr<Subsection<E>> &subsec, u32 addr) {
-    return subsec->input_addr < addr;
-  });
-
-  // Assign each subsection a group of relocations
-  i64 i = 0;
-  for (auto it = begin; it < end; it++) {
-    Subsection<E> &subsec = **it;
-    subsec.rel_offset = i;
-    while (i < rels.size() &&
-           rels[i].offset < subsec.input_offset + subsec.input_size) {
-      rels[i].offset -= subsec.input_offset;
-      i++;
-    }
-    subsec.nrels = i - subsec.rel_offset;
-  }
-}
-
-#define INSTANTIATE(E)                          \
-  template class InputSection<E>
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/macho.h

@@ -1,693 +0,0 @@
-#pragma once
-
-#include "../byteorder.h"
-
-#include <cassert>
-#include <cstdint>
-#include <cstring>
-#include <string_view>
-
-namespace mold::macho {
-
-typedef uint8_t u8;
-typedef uint16_t u16;
-typedef uint32_t u32;
-typedef uint64_t u64;
-
-typedef int8_t i8;
-typedef int16_t i16;
-typedef int32_t i32;
-typedef int64_t i64;
-
-static constexpr u32 FAT_MAGIC = 0xcafebabe;
-
-static constexpr u32 MH_OBJECT = 0x1;
-static constexpr u32 MH_EXECUTE = 0x2;
-static constexpr u32 MH_FVMLIB = 0x3;
-static constexpr u32 MH_CORE = 0x4;
-static constexpr u32 MH_PRELOAD = 0x5;
-static constexpr u32 MH_DYLIB = 0x6;
-static constexpr u32 MH_DYLINKER = 0x7;
-static constexpr u32 MH_BUNDLE = 0x8;
-static constexpr u32 MH_DYLIB_STUB = 0x9;
-static constexpr u32 MH_DSYM = 0xa;
-static constexpr u32 MH_KEXT_BUNDLE = 0xb;
-
-static constexpr u32 MH_NOUNDEFS = 0x1;
-static constexpr u32 MH_INCRLINK = 0x2;
-static constexpr u32 MH_DYLDLINK = 0x4;
-static constexpr u32 MH_BINDATLOAD = 0x8;
-static constexpr u32 MH_PREBOUND = 0x10;
-static constexpr u32 MH_SPLIT_SEGS = 0x20;
-static constexpr u32 MH_LAZY_INIT = 0x40;
-static constexpr u32 MH_TWOLEVEL = 0x80;
-static constexpr u32 MH_FORCE_FLAT = 0x100;
-static constexpr u32 MH_NOMULTIDEFS = 0x200;
-static constexpr u32 MH_NOFIXPREBINDING = 0x400;
-static constexpr u32 MH_PREBINDABLE = 0x800;
-static constexpr u32 MH_ALLMODSBOUND = 0x1000;
-static constexpr u32 MH_SUBSECTIONS_VIA_SYMBOLS = 0x2000;
-static constexpr u32 MH_CANONICAL = 0x4000;
-static constexpr u32 MH_WEAK_DEFINES = 0x8000;
-static constexpr u32 MH_BINDS_TO_WEAK = 0x10000;
-static constexpr u32 MH_ALLOW_STACK_EXECUTION = 0x20000;
-static constexpr u32 MH_ROOT_SAFE = 0x40000;
-static constexpr u32 MH_SETUID_SAFE = 0x80000;
-static constexpr u32 MH_NO_REEXPORTED_DYLIBS = 0x100000;
-static constexpr u32 MH_PIE = 0x200000;
-static constexpr u32 MH_DEAD_STRIPPABLE_DYLIB = 0x400000;
-static constexpr u32 MH_HAS_TLV_DESCRIPTORS = 0x800000;
-static constexpr u32 MH_NO_HEAP_EXECUTION = 0x1000000;
-static constexpr u32 MH_APP_EXTENSION_SAFE = 0x02000000;
-static constexpr u32 MH_NLIST_OUTOFSYNC_WITH_DYLDINFO = 0x04000000;
-static constexpr u32 MH_SIM_SUPPORT = 0x08000000;
-
-static constexpr u32 VM_PROT_READ = 0x1;
-static constexpr u32 VM_PROT_WRITE = 0x2;
-static constexpr u32 VM_PROT_EXECUTE = 0x4;
-static constexpr u32 VM_PROT_NO_CHANGE = 0x8;
-static constexpr u32 VM_PROT_COPY = 0x10;
-static constexpr u32 VM_PROT_WANTS_COPY = 0x10;
-
-static constexpr u32 LC_REQ_DYLD = 0x80000000;
-
-static constexpr u32 LC_SEGMENT = 0x1;
-static constexpr u32 LC_SYMTAB = 0x2;
-static constexpr u32 LC_SYMSEG = 0x3;
-static constexpr u32 LC_THREAD = 0x4;
-static constexpr u32 LC_UNIXTHREAD = 0x5;
-static constexpr u32 LC_LOADFVMLIB = 0x6;
-static constexpr u32 LC_IDFVMLIB = 0x7;
-static constexpr u32 LC_IDENT = 0x8;
-static constexpr u32 LC_FVMFILE = 0x9;
-static constexpr u32 LC_PREPAGE = 0xa;
-static constexpr u32 LC_DYSYMTAB = 0xb;
-static constexpr u32 LC_LOAD_DYLIB = 0xc;
-static constexpr u32 LC_ID_DYLIB = 0xd;
-static constexpr u32 LC_LOAD_DYLINKER = 0xe;
-static constexpr u32 LC_ID_DYLINKER = 0xf;
-static constexpr u32 LC_PREBOUND_DYLIB = 0x10;
-static constexpr u32 LC_ROUTINES = 0x11;
-static constexpr u32 LC_SUB_FRAMEWORK = 0x12;
-static constexpr u32 LC_SUB_UMBRELLA = 0x13;
-static constexpr u32 LC_SUB_CLIENT = 0x14;
-static constexpr u32 LC_SUB_LIBRARY = 0x15;
-static constexpr u32 LC_TWOLEVEL_HINTS = 0x16;
-static constexpr u32 LC_PREBIND_CKSUM = 0x17;
-static constexpr u32 LC_LOAD_WEAK_DYLIB = (0x18 | LC_REQ_DYLD);
-static constexpr u32 LC_SEGMENT_64 = 0x19;
-static constexpr u32 LC_ROUTINES_64 = 0x1a;
-static constexpr u32 LC_UUID = 0x1b;
-static constexpr u32 LC_RPATH = (0x1c | LC_REQ_DYLD);
-static constexpr u32 LC_CODE_SIGNATURE = 0x1d;
-static constexpr u32 LC_SEGMENT_SPLIT_INFO = 0x1e;
-static constexpr u32 LC_REEXPORT_DYLIB = (0x1f | LC_REQ_DYLD);
-static constexpr u32 LC_LAZY_LOAD_DYLIB = 0x20;
-static constexpr u32 LC_ENCRYPTION_INFO = 0x21;
-static constexpr u32 LC_DYLD_INFO = 0x22;
-static constexpr u32 LC_DYLD_INFO_ONLY = (0x22 | LC_REQ_DYLD);
-static constexpr u32 LC_LOAD_UPWARD_DYLIB = (0x23 | LC_REQ_DYLD);
-static constexpr u32 LC_VERSION_MIN_MACOSX = 0x24;
-static constexpr u32 LC_VERSION_MIN_IPHONEOS = 0x25;
-static constexpr u32 LC_FUNCTION_STARTS = 0x26;
-static constexpr u32 LC_DYLD_ENVIRONMENT = 0x27;
-static constexpr u32 LC_MAIN = (0x28 | LC_REQ_DYLD);
-static constexpr u32 LC_DATA_IN_CODE = 0x29;
-static constexpr u32 LC_SOURCE_VERSION = 0x2A;
-static constexpr u32 LC_DYLIB_CODE_SIGN_DRS = 0x2B;
-static constexpr u32 LC_ENCRYPTION_INFO_64 = 0x2C;
-static constexpr u32 LC_LINKER_OPTION = 0x2D;
-static constexpr u32 LC_LINKER_OPTIMIZATION_HINT = 0x2E;
-static constexpr u32 LC_VERSION_MIN_TVOS = 0x2F;
-static constexpr u32 LC_VERSION_MIN_WATCHOS = 0x30;
-static constexpr u32 LC_NOTE = 0x31;
-static constexpr u32 LC_BUILD_VERSION = 0x32;
-static constexpr u32 LC_DYLD_EXPORTS_TRIE = (0x33 | LC_REQ_DYLD);
-static constexpr u32 LC_DYLD_CHAINED_FIXUPS = (0x34 | LC_REQ_DYLD);
-
-static constexpr u32 SG_HIGHVM = 0x1;
-static constexpr u32 SG_FVMLIB = 0x2;
-static constexpr u32 SG_NORELOC = 0x4;
-static constexpr u32 SG_PROTECTED_VERSION_1 = 0x8;
-static constexpr u32 SG_READ_ONLY = 0x10;
-
-static constexpr u32 S_REGULAR = 0x0;
-static constexpr u32 S_ZEROFILL = 0x1;
-static constexpr u32 S_CSTRING_LITERALS = 0x2;
-static constexpr u32 S_4BYTE_LITERALS = 0x3;
-static constexpr u32 S_8BYTE_LITERALS = 0x4;
-static constexpr u32 S_LITERAL_POINTERS = 0x5;
-static constexpr u32 S_NON_LAZY_SYMBOL_POINTERS = 0x6;
-static constexpr u32 S_LAZY_SYMBOL_POINTERS = 0x7;
-static constexpr u32 S_SYMBOL_STUBS = 0x8;
-static constexpr u32 S_MOD_INIT_FUNC_POINTERS = 0x9;
-static constexpr u32 S_MOD_TERM_FUNC_POINTERS = 0xa;
-static constexpr u32 S_COALESCED = 0xb;
-static constexpr u32 S_GB_ZEROFILL = 0xc;
-static constexpr u32 S_INTERPOSING = 0xd;
-static constexpr u32 S_16BYTE_LITERALS = 0xe;
-static constexpr u32 S_DTRACE_DOF = 0xf;
-static constexpr u32 S_LAZY_DYLIB_SYMBOL_POINTERS = 0x10;
-static constexpr u32 S_THREAD_LOCAL_REGULAR = 0x11;
-static constexpr u32 S_THREAD_LOCAL_ZEROFILL = 0x12;
-static constexpr u32 S_THREAD_LOCAL_VARIABLES = 0x13;
-static constexpr u32 S_THREAD_LOCAL_VARIABLE_POINTERS = 0x14;
-static constexpr u32 S_THREAD_LOCAL_INIT_FUNCTION_POINTERS = 0x15;
-static constexpr u32 S_INIT_FUNC_OFFSETS = 0x16;
-
-static constexpr u32 S_ATTR_LOC_RELOC = 0x000001;
-static constexpr u32 S_ATTR_EXT_RELOC = 0x000002;
-static constexpr u32 S_ATTR_SOME_INSTRUCTIONS = 0x000004;
-
-static constexpr u32 S_ATTR_DEBUG = 0x020000;
-static constexpr u32 S_ATTR_SELF_MODIFYING_CODE = 0x040000;
-static constexpr u32 S_ATTR_LIVE_SUPPORT = 0x080000;
-static constexpr u32 S_ATTR_NO_DEAD_STRIP = 0x100000;
-static constexpr u32 S_ATTR_STRIP_STATIC_SYMS = 0x200000;
-static constexpr u32 S_ATTR_NO_TOC = 0x400000;
-static constexpr u32 S_ATTR_PURE_INSTRUCTIONS = 0x800000;
-
-static constexpr u32 CPU_TYPE_X86_64 = 0x1000007;
-static constexpr u32 CPU_TYPE_ARM64 = 0x100000c;
-
-static constexpr u32 CPU_SUBTYPE_X86_64_ALL = 3;
-static constexpr u32 CPU_SUBTYPE_ARM64_ALL = 0;
-
-static constexpr u32 REBASE_TYPE_POINTER = 1;
-static constexpr u32 REBASE_TYPE_TEXT_ABSOLUTE32 = 2;
-static constexpr u32 REBASE_TYPE_TEXT_PCREL32 = 3;
-
-static constexpr u32 REBASE_OPCODE_MASK = 0xf0;
-static constexpr u32 REBASE_IMMEDIATE_MASK = 0x0f;
-static constexpr u32 REBASE_OPCODE_DONE = 0x00;
-static constexpr u32 REBASE_OPCODE_SET_TYPE_IMM = 0x10;
-static constexpr u32 REBASE_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB = 0x20;
-static constexpr u32 REBASE_OPCODE_ADD_ADDR_ULEB = 0x30;
-static constexpr u32 REBASE_OPCODE_ADD_ADDR_IMM_SCALED = 0x40;
-static constexpr u32 REBASE_OPCODE_DO_REBASE_IMM_TIMES = 0x50;
-static constexpr u32 REBASE_OPCODE_DO_REBASE_ULEB_TIMES = 0x60;
-static constexpr u32 REBASE_OPCODE_DO_REBASE_ADD_ADDR_ULEB = 0x70;
-static constexpr u32 REBASE_OPCODE_DO_REBASE_ULEB_TIMES_SKIPPING_ULEB = 0x80;
-
-static constexpr u32 BIND_TYPE_POINTER = 1;
-static constexpr u32 BIND_TYPE_TEXT_ABSOLUTE32 = 2;
-static constexpr u32 BIND_TYPE_TEXT_PCREL32 = 3;
-
-static constexpr u32 BIND_SPECIAL_DYLIB_SELF = 0;
-static constexpr u32 BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE = -1;
-static constexpr u32 BIND_SPECIAL_DYLIB_FLAT_LOOKUP = -2;
-static constexpr u32 BIND_SPECIAL_DYLIB_WEAK_LOOKUP = -3;
-
-static constexpr u32 BIND_SYMBOL_FLAGS_WEAK_IMPORT = 0x1;
-static constexpr u32 BIND_SYMBOL_FLAGS_NON_WEAK_DEFINITION = 0x8;
-
-static constexpr u32 BIND_OPCODE_MASK = 0xF0;
-static constexpr u32 BIND_IMMEDIATE_MASK = 0x0F;
-static constexpr u32 BIND_OPCODE_DONE = 0x00;
-static constexpr u32 BIND_OPCODE_SET_DYLIB_ORDINAL_IMM = 0x10;
-static constexpr u32 BIND_OPCODE_SET_DYLIB_ORDINAL_ULEB = 0x20;
-static constexpr u32 BIND_OPCODE_SET_DYLIB_SPECIAL_IMM = 0x30;
-static constexpr u32 BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM = 0x40;
-static constexpr u32 BIND_OPCODE_SET_TYPE_IMM = 0x50;
-static constexpr u32 BIND_OPCODE_SET_ADDEND_SLEB = 0x60;
-static constexpr u32 BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB = 0x70;
-static constexpr u32 BIND_OPCODE_ADD_ADDR_ULEB = 0x80;
-static constexpr u32 BIND_OPCODE_DO_BIND = 0x90;
-static constexpr u32 BIND_OPCODE_DO_BIND_ADD_ADDR_ULEB = 0xA0;
-static constexpr u32 BIND_OPCODE_DO_BIND_ADD_ADDR_IMM_SCALED = 0xB0;
-static constexpr u32 BIND_OPCODE_DO_BIND_ULEB_TIMES_SKIPPING_ULEB = 0xC0;
-static constexpr u32 BIND_OPCODE_THREADED = 0xD0;
-static constexpr u32 BIND_SUBOPCODE_THREADED_SET_BIND_ORDINAL_TABLE_SIZE_ULEB = 0x00;
-static constexpr u32 BIND_SUBOPCODE_THREADED_APPLY = 0x01;
-
-static constexpr u32 EXPORT_SYMBOL_FLAGS_KIND_MASK = 0x03;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_KIND_REGULAR = 0x00;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_KIND_THREAD_LOCAL = 0x01;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_KIND_ABSOLUTE = 0x02;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_WEAK_DEFINITION = 0x04;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_REEXPORT = 0x08;
-static constexpr u32 EXPORT_SYMBOL_FLAGS_STUB_AND_RESOLVER = 0x10;
-
-static constexpr u32 DICE_KIND_DATA = 1;
-static constexpr u32 DICE_KIND_JUMP_TABLE8 = 2;
-static constexpr u32 DICE_KIND_JUMP_TABLE16 = 3;
-static constexpr u32 DICE_KIND_JUMP_TABLE32 = 4;
-static constexpr u32 DICE_KIND_ABS_JUMP_TABLE32 = 5;
-
-static constexpr u32 N_UNDF = 0;
-static constexpr u32 N_ABS = 1;
-static constexpr u32 N_INDR = 5;
-static constexpr u32 N_PBUD = 6;
-static constexpr u32 N_SECT = 7;
-
-static constexpr u32 N_GSYM = 0x20;
-static constexpr u32 N_FNAME = 0x22;
-static constexpr u32 N_FUN = 0x24;
-static constexpr u32 N_STSYM = 0x26;
-static constexpr u32 N_LCSYM = 0x28;
-static constexpr u32 N_BNSYM = 0x2e;
-static constexpr u32 N_AST = 0x32;
-static constexpr u32 N_OPT = 0x3c;
-static constexpr u32 N_RSYM = 0x40;
-static constexpr u32 N_SLINE = 0x44;
-static constexpr u32 N_ENSYM = 0x4e;
-static constexpr u32 N_SSYM = 0x60;
-static constexpr u32 N_SO = 0x64;
-static constexpr u32 N_OSO = 0x66;
-static constexpr u32 N_LSYM = 0x80;
-static constexpr u32 N_BINCL = 0x82;
-static constexpr u32 N_SOL = 0x84;
-static constexpr u32 N_PARAMS = 0x86;
-static constexpr u32 N_VERSION = 0x88;
-static constexpr u32 N_OLEVEL = 0x8A;
-static constexpr u32 N_PSYM = 0xa0;
-static constexpr u32 N_EINCL = 0xa2;
-static constexpr u32 N_ENTRY = 0xa4;
-static constexpr u32 N_LBRAC = 0xc0;
-static constexpr u32 N_EXCL = 0xc2;
-static constexpr u32 N_RBRAC = 0xe0;
-static constexpr u32 N_BCOMM = 0xe2;
-static constexpr u32 N_ECOMM = 0xe4;
-static constexpr u32 N_ECOML = 0xe8;
-static constexpr u32 N_LENG = 0xfe;
-static constexpr u32 N_PC = 0x30;
-
-static constexpr u32 REFERENCE_TYPE = 0xf;
-static constexpr u32 REFERENCE_FLAG_UNDEFINED_NON_LAZY = 0;
-static constexpr u32 REFERENCE_FLAG_UNDEFINED_LAZY = 1;
-static constexpr u32 REFERENCE_FLAG_DEFINED = 2;
-static constexpr u32 REFERENCE_FLAG_PRIVATE_DEFINED = 3;
-static constexpr u32 REFERENCE_FLAG_PRIVATE_UNDEFINED_NON_LAZY = 4;
-static constexpr u32 REFERENCE_FLAG_PRIVATE_UNDEFINED_LAZY = 5;
-
-static constexpr u32 REFERENCED_DYNAMICALLY = 0x0010;
-
-static constexpr u32 N_NO_DEAD_STRIP = 0x0020;
-static constexpr u32 N_DESC_DISCARDED = 0x0020;
-static constexpr u32 N_WEAK_REF = 0x0040;
-static constexpr u32 N_WEAK_DEF = 0x0080;
-static constexpr u32 N_REF_TO_WEAK = 0x0080;
-static constexpr u32 N_ARM_THUMB_DEF = 0x0008;
-static constexpr u32 N_SYMBOL_RESOLVER = 0x0100;
-static constexpr u32 N_ALT_ENTRY = 0x0200;
-
-static constexpr u32 PLATFORM_MACOS = 1;
-static constexpr u32 PLATFORM_IOS = 2;
-static constexpr u32 PLATFORM_TVOS = 3;
-static constexpr u32 PLATFORM_WATCHOS = 4;
-static constexpr u32 PLATFORM_BRIDGEOS = 5;
-static constexpr u32 PLATFORM_MACCATALYST = 6;
-static constexpr u32 PLATFORM_IOSSIMULATOR = 7;
-static constexpr u32 PLATFORM_TVOSSIMULATOR = 8;
-static constexpr u32 PLATFORM_WATCHOSSIMULATOR = 9;
-static constexpr u32 PLATFORM_DRIVERKIT = 10;
-
-static constexpr u32 TOOL_CLANG = 1;
-static constexpr u32 TOOL_SWIFT = 2;
-static constexpr u32 TOOL_LD = 3;
-
-static constexpr u32 ARM64_RELOC_UNSIGNED = 0;
-static constexpr u32 ARM64_RELOC_SUBTRACTOR = 1;
-static constexpr u32 ARM64_RELOC_BRANCH26 = 2;
-static constexpr u32 ARM64_RELOC_PAGE21 = 3;
-static constexpr u32 ARM64_RELOC_PAGEOFF12 = 4;
-static constexpr u32 ARM64_RELOC_GOT_LOAD_PAGE21 = 5;
-static constexpr u32 ARM64_RELOC_GOT_LOAD_PAGEOFF12 = 6;
-static constexpr u32 ARM64_RELOC_POINTER_TO_GOT = 7;
-static constexpr u32 ARM64_RELOC_TLVP_LOAD_PAGE21 = 8;
-static constexpr u32 ARM64_RELOC_TLVP_LOAD_PAGEOFF12 = 9;
-static constexpr u32 ARM64_RELOC_ADDEND = 10;
-
-static constexpr u32 X86_64_RELOC_UNSIGNED = 0;
-static constexpr u32 X86_64_RELOC_SIGNED = 1;
-static constexpr u32 X86_64_RELOC_BRANCH = 2;
-static constexpr u32 X86_64_RELOC_GOT_LOAD = 3;
-static constexpr u32 X86_64_RELOC_GOT = 4;
-static constexpr u32 X86_64_RELOC_SUBTRACTOR = 5;
-static constexpr u32 X86_64_RELOC_SIGNED_1 = 6;
-static constexpr u32 X86_64_RELOC_SIGNED_2 = 7;
-static constexpr u32 X86_64_RELOC_SIGNED_4 = 8;
-static constexpr u32 X86_64_RELOC_TLV = 9;
-
-struct FatHeader {
-  ubig32 magic;
-  ubig32 nfat_arch;
-};
-
-struct FatArch {
-  ubig32 cputype;
-  ubig32 cpusubtype;
-  ubig32 offset;
-  ubig32 size;
-  ubig32 align;
-};
-
-struct MachHeader {
-  u32 magic;
-  u32 cputype;
-  u32 cpusubtype;
-  u32 filetype;
-  u32 ncmds;
-  u32 sizeofcmds;
-  u32 flags;
-  u32 reserved;
-};
-
-struct LoadCommand {
-  u32 cmd;
-  u32 cmdsize;
-};
-
-struct SegmentCommand {
-  std::string_view get_segname() const {
-    return {segname, strnlen(segname, sizeof(segname))};
-  }
-
-  u32 cmd;
-  u32 cmdsize;
-  char segname[16];
-  u64 vmaddr;
-  u64 vmsize;
-  u64 fileoff;
-  u64 filesize;
-  u32 maxprot;
-  u32 initprot;
-  u32 nsects;
-  u32 flags;
-};
-
-struct MachSection {
-  void set_segname(std::string_view name) {
-    assert(name.size() <= sizeof(segname));
-    memcpy(segname, name.data(), name.size());
-  }
-
-  std::string_view get_segname() const {
-    return {segname, strnlen(segname, sizeof(segname))};
-  }
-
-  void set_sectname(std::string_view name) {
-    assert(name.size() <= sizeof(sectname));
-    memcpy(sectname, name.data(), name.size());
-  }
-
-  std::string_view get_sectname() const {
-    return {sectname, strnlen(sectname, sizeof(sectname))};
-  }
-
-  bool match(std::string_view segname, std::string_view sectname) const {
-    return get_segname() == segname && get_sectname() == sectname;
-  }
-
-  char sectname[16];
-  char segname[16];
-  u64 addr;
-  u64 size;
-  u32 offset;
-  u32 p2align;
-  u32 reloff;
-  u32 nreloc;
-  u32 type : 8;
-  u32 attr : 24;
-  u32 reserved1;
-  u32 reserved2;
-  u32 reserved3;
-};
-
-struct DylibCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 nameoff;
-  u32 timestamp;
-  u32 current_version;
-  u32 compatibility_version;
-};
-
-struct DylinkerCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 nameoff;
-};
-
-struct SymtabCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 symoff;
-  u32 nsyms;
-  u32 stroff;
-  u32 strsize;
-};
-
-struct DysymtabCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 ilocalsym;
-  u32 nlocalsym;
-  u32 iextdefsym;
-  u32 nextdefsym;
-  u32 iundefsym;
-  u32 nundefsym;
-  u32 tocoff;
-  u32 ntoc;
-  u32 modtaboff;
-  u32 nmodtab;
-  u32 extrefsymoff;
-  u32 nextrefsyms;
-  u32 indirectsymoff;
-  u32 nindirectsyms;
-  u32 extreloff;
-  u32 nextrel;
-  u32 locreloff;
-  u32 nlocrel;
-};
-
-struct VersionMinCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 version;
-  u32 sdk;
-};
-
-struct DyldInfoCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 rebase_off;
-  u32 rebase_size;
-  u32 bind_off;
-  u32 bind_size;
-  u32 weak_bind_off;
-  u32 weak_bind_size;
-  u32 lazy_bind_off;
-  u32 lazy_bind_size;
-  u32 export_off;
-  u32 export_size;
-};
-
-struct UUIDCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u8 uuid[16];
-};
-
-struct RpathCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 path_off;
-};
-
-struct LinkEditDataCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 dataoff;
-  u32 datasize;
-};
-
-struct BuildToolVersion {
-  u32 tool;
-  u32 version;
-};
-
-struct BuildVersionCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u32 platform;
-  u32 minos;
-  u32 sdk;
-  u32 ntools;
-};
-
-struct EntryPointCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u64 entryoff;
-  u64 stacksize;
-};
-
-struct SourceVersionCommand {
-  u32 cmd;
-  u32 cmdsize;
-  u64 version;
-};
-
-struct DataInCodeEntry {
-  u32 offset;
-  u16 length;
-  u16 kind;
-};
-
-// This struct is named `n_list` on BSD and macOS.
-struct MachSym {
-  bool is_undef() const {
-    return type == N_UNDF && !is_common();
-  }
-
-  bool is_common() const {
-    return type == N_UNDF && ext && value;
-  }
-
-  u32 stroff;
-  u8 ext : 1;
-  u8 type : 3;
-  u8 pext : 1;
-  u8 stub : 3;
-  u8 sect;
-  union {
-    u16 desc;
-    struct {
-      u16 : 8;
-      u16 p2align : 4;
-      u16 : 4;
-    };
-  };
-  u64 value;
-};
-
-// This struct is named `relocation_info` on BSD and macOS.
-struct MachRel {
-  u32 offset;
-  u32 idx : 24;
-  u32 is_pcrel : 1;
-  u32 p2size : 2;
-  u32 is_extern : 1;
-  u32 type : 4;
-};
-
-// __TEXT,__unwind_info section contents
-
-static constexpr u32 UNWIND_SECTION_VERSION = 1;
-static constexpr u32 UNWIND_SECOND_LEVEL_REGULAR = 2;
-static constexpr u32 UNWIND_SECOND_LEVEL_COMPRESSED = 3;
-static constexpr u32 UNWIND_PERSONALITY_MASK = 0x30000000;
-
-struct UnwindSectionHeader {
-  u32 version;
-  u32 encoding_offset;
-  u32 encoding_count;
-  u32 personality_offset;
-  u32 personality_count;
-  u32 page_offset;
-  u32 page_count;
-};
-
-struct UnwindFirstLevelPage {
-  u32 func_addr;
-  u32 page_offset;
-  u32 lsda_offset;
-};
-
-struct UnwindSecondLevelPage {
-  u32 kind;
-  u16 page_offset;
-  u16 page_count;
-  u16 encoding_offset;
-  u16 encoding_count;
-};
-
-struct UnwindLsdaEntry {
-  u32 func_addr;
-  u32 lsda_addr;
-};
-
-struct UnwindPageEntry {
-  u32 func_addr : 24;
-  u32 encoding : 8;
-};
-
-// __LD,__compact_unwind section contents
-
-struct CompactUnwindEntry {
-  u64 code_start;
-  u32 code_len;
-  u32 encoding;
-  u64 personality;
-  u64 lsda;
-};
-
-// __LINKEDIT,__code_signature
-
-static constexpr u32 CSMAGIC_EMBEDDED_SIGNATURE = 0xfade0cc0;
-static constexpr u32 CS_SUPPORTSEXECSEG = 0x20400;
-static constexpr u32 CSMAGIC_CODEDIRECTORY = 0xfade0c02;
-static constexpr u32 CSSLOT_CODEDIRECTORY = 0;
-static constexpr u32 CS_ADHOC = 0x00000002;
-static constexpr u32 CS_LINKER_SIGNED = 0x00020000;
-static constexpr u32 CS_EXECSEG_MAIN_BINARY = 1;
-static constexpr u32 CS_HASHTYPE_SHA256 = 2;
-
-struct CodeSignatureHeader {
-  ubig32 magic;
-  ubig32 length;
-  ubig32 count;
-};
-
-struct CodeSignatureBlobIndex {
-  ubig32 type;
-  ubig32 offset;
-  u32 padding;
-};
-
-struct CodeSignatureDirectory {
-  ubig32 magic;
-  ubig32 length;
-  ubig32 version;
-  ubig32 flags;
-  ubig32 hash_offset;
-  ubig32 ident_offset;
-  ubig32 n_special_slots;
-  ubig32 n_code_slots;
-  ubig32 code_limit;
-  u8 hash_size;
-  u8 hash_type;
-  u8 platform;
-  u8 page_size;
-  ubig32 spare2;
-  ubig32 scatter_offset;
-  ubig32 team_offset;
-  ubig32 spare3;
-  ubig64 code_limit64;
-  ubig64 exec_seg_base;
-  ubig64 exec_seg_limit;
-  ubig64 exec_seg_flags;
-};
-
-struct ARM64 {
-  static constexpr u32 cputype = CPU_TYPE_ARM64;
-  static constexpr u32 cpusubtype = CPU_SUBTYPE_ARM64_ALL;
-  static constexpr u32 abs_rel = ARM64_RELOC_UNSIGNED;
-  static constexpr u32 word_size = 8;
-  static constexpr u32 stub_size = 12;
-  static constexpr u32 stub_helper_hdr_size = 24;
-  static constexpr u32 stub_helper_size = 12;
-};
-
-struct X86_64 {
-  static constexpr u32 cputype = CPU_TYPE_X86_64;
-  static constexpr u32 cpusubtype = CPU_SUBTYPE_X86_64_ALL;
-  static constexpr u32 abs_rel = X86_64_RELOC_UNSIGNED;
-  static constexpr u32 word_size = 8;
-  static constexpr u32 stub_size = 6;
-  static constexpr u32 stub_helper_hdr_size = 16;
-  static constexpr u32 stub_helper_size = 10;
-};
-
-} // namespace mold::macho

macho/main.cc

@@ -1,480 +0,0 @@
-#include "mold.h"
-#include "../archive-file.h"
-#include "../cmdline.h"
-
-#include <cstdlib>
-#include <fcntl.h>
-#include <iomanip>
-#include <iostream>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-namespace mold::macho {
-
-static std::pair<std::string_view, std::string_view>
-split_string(std::string_view str, char sep) {
-  size_t pos = str.find(sep);
-  if (pos == str.npos)
-    return {str, ""};
-  return {str.substr(0, pos), str.substr(pos + 1)};
-}
-
-template <typename E>
-static void create_internal_file(Context<E> &ctx) {
-  ObjectFile<E> *obj = new ObjectFile<E>;
-  ctx.obj_pool.push_back(std::unique_ptr<ObjectFile<E>>(obj));
-  ctx.objs.push_back(obj);
-
-  auto add = [&](std::string_view name) {
-    Symbol<E> *sym = get_symbol(ctx, name);
-    sym->file = obj;
-    obj->syms.push_back(sym);
-    return sym;
-  };
-
-  add("__dyld_private");
-
-  switch (ctx.output_type) {
-  case MH_EXECUTE: {
-    Symbol<E> *sym = add("__mh_execute_header");
-    sym->is_extern = true;
-    sym->referenced_dynamically = true;
-    sym->value = ctx.arg.pagezero_size;
-    break;
-  }
-  case MH_DYLIB:
-    add("__mh_dylib_header");
-    break;
-  case MH_BUNDLE:
-    add("__mh_bundle_header");
-    break;
-  default:
-    unreachable();
-  }
-}
-
-template <typename E>
-static bool compare_segments(const std::unique_ptr<OutputSegment<E>> &a,
-                             const std::unique_ptr<OutputSegment<E>> &b) {
-  // We want to sort output segments in the following order:
-  // __TEXT, __DATA_CONST, __DATA, <other segments>, __LINKEDIT
-  auto get_rank = [](std::string_view name) {
-    if (name == "__TEXT")
-      return 0;
-    if (name == "__DATA_CONST")
-      return 1;
-    if (name == "__DATA")
-      return 2;
-    return INT_MAX;
-  };
-
-  std::string_view na = a->cmd.get_segname();
-  std::string_view nb = b->cmd.get_segname();
-  i64 ra = get_rank(na);
-  i64 rb = get_rank(nb);
-  if (ra != INT_MAX || rb != INT_MAX)
-    return ra < rb;
-
-  if (na == "__LINKEDIT" || nb == "__LINKEDIT")
-    return na != "__LINKEDIT";
-  return na < nb;
-}
-
-template <typename E>
-static bool compare_chunks(const Chunk<E> *a, const Chunk<E> *b) {
-  assert(a->hdr.get_segname() == b->hdr.get_segname());
-
-  if ((a->hdr.type == S_ZEROFILL) != (b->hdr.type == S_ZEROFILL))
-    return a->hdr.type != S_ZEROFILL;
-
-  static const std::string_view rank[] = {
-    // __TEXT
-    "__mach_header",
-    "__text",
-    "__stubs",
-    "__stub_helper",
-    "__gcc_except_tab",
-    "__cstring",
-    "__unwind_info",
-    // __DATA_CONST
-    "__got",
-    // __DATA
-    "__la_symbol_ptr",
-    "__data",
-    "__thread_ptr",
-    "__thread_data",
-    "__thread_vars",
-    "__thread_bss",
-    "__common",
-    "__bss",
-    // __LINKEDIT
-    "__rebase",
-    "__binding",
-    "__weak_binding",
-    "__lazy_binding",
-    "__export",
-    "__func_starts",
-    "__data_in_code",
-    "__symbol_table",
-    "__ind_sym_tab",
-    "__string_table",
-    "__code_signature",
-  };
-
-  auto get_rank = [](const Chunk<E> *chunk) -> i64 {
-    std::string_view name = chunk->hdr.get_sectname();
-    i64 i = 0;
-    for (; i < sizeof(rank) / sizeof(rank[0]); i++)
-      if (name == rank[i])
-        return i;
-    return INT_MAX;
-  };
-
-  i64 ra = get_rank(a);
-  i64 rb = get_rank(b);
-  if (ra == INT_MAX && rb == INT_MAX)
-    return a->hdr.get_sectname() < b->hdr.get_sectname();
-  return ra < rb;
-}
-
-template <typename E>
-static void create_synthetic_chunks(Context<E> &ctx) {
-  for (ObjectFile<E> *file : ctx.objs)
-    for (std::unique_ptr<Subsection<E>> &subsec : file->subsections)
-      subsec->isec.osec.add_subsec(subsec.get());
-
-  for (Chunk<E> *chunk : ctx.chunks) {
-    if (chunk != ctx.data && chunk->is_regular &&
-        ((OutputSection<E> *)chunk)->members.empty())
-      continue;
-
-    OutputSegment<E> *seg =
-      OutputSegment<E>::get_instance(ctx, chunk->hdr.get_segname());
-    seg->chunks.push_back(chunk);
-  }
-
-  sort(ctx.segments, compare_segments<E>);
-
-  for (std::unique_ptr<OutputSegment<E>> &seg : ctx.segments)
-    sort(seg->chunks, compare_chunks<E>);
-}
-
-template <typename E>
-static void scan_unwind_info(Context<E> &ctx) {
-  for (ObjectFile<E> *file : ctx.objs)
-    for (UnwindRecord<E> &rec : file->unwind_records)
-      if (!ctx.arg.dead_strip || rec.is_alive)
-        if (rec.personality)
-          rec.personality->flags |= NEEDS_GOT;
-}
-
-template <typename E>
-static void export_symbols(Context<E> &ctx) {
-  ctx.got.add(ctx, get_symbol(ctx, "dyld_stub_binder"));
-
-  for (ObjectFile<E> *file : ctx.objs) {
-    for (Symbol<E> *sym : file->syms) {
-      if (sym && sym->file == file) {
-        if (sym->flags & NEEDS_GOT)
-          ctx.got.add(ctx, sym);
-        if (sym->flags & NEEDS_THREAD_PTR)
-          ctx.thread_ptrs.add(ctx, sym);
-      }
-    }
-  }
-
-  for (DylibFile<E> *file : ctx.dylibs) {
-    for (Symbol<E> *sym : file->syms) {
-      if (sym && sym->file == file) {
-        if (sym->flags & NEEDS_STUB)
-          ctx.stubs.add(ctx, sym);
-        if (sym->flags & NEEDS_GOT)
-          ctx.got.add(ctx, sym);
-        if (sym->flags & NEEDS_THREAD_PTR)
-          ctx.thread_ptrs.add(ctx, sym);
-      }
-    }
-  }
-}
-
-template <typename E>
-static i64 assign_offsets(Context<E> &ctx) {
-  i64 sect_idx = 1;
-  for (std::unique_ptr<OutputSegment<E>> &seg : ctx.segments)
-    for (Chunk<E> *chunk : seg->chunks)
-      if (!chunk->is_hidden)
-        chunk->sect_idx = sect_idx++;
-
-  i64 fileoff = 0;
-  i64 vmaddr = ctx.arg.pagezero_size;
-
-  for (std::unique_ptr<OutputSegment<E>> &seg : ctx.segments) {
-    seg->set_offset(ctx, fileoff, vmaddr);
-    fileoff += seg->cmd.filesize;
-    vmaddr += seg->cmd.vmsize;
-  }
-  return fileoff;
-}
-
-template <typename E>
-static void fix_synthetic_symbol_values(Context<E> &ctx) {
-  get_symbol(ctx, "__dyld_private")->value = ctx.data->hdr.addr;
-  get_symbol(ctx, "__mh_dylib_header")->value = ctx.data->hdr.addr;
-  get_symbol(ctx, "__mh_bundle_header")->value = ctx.data->hdr.addr;
-}
-
-template <typename E>
-MappedFile<Context<E>> *find_framework(Context<E> &ctx, std::string name) {
-  std::string suffix;
-  std::tie(name, suffix) = split_string(name, ',');
-
-  for (std::string path : ctx.arg.framework_paths) {
-    path = get_realpath(path + "/" + name + ".framework/" + name);
-
-    if (!suffix.empty())
-      if (auto *mf = MappedFile<Context<E>>::open(ctx, path + suffix))
-        return mf;
-
-    if (auto *mf = MappedFile<Context<E>>::open(ctx, path + ".tbd"))
-      return mf;
-
-    if (auto *mf = MappedFile<Context<E>>::open(ctx, path))
-      return mf;
-  }
-  Fatal(ctx) << "-framework not found: " << name;
-}
-
-template <typename E>
-MappedFile<Context<E>> *find_library(Context<E> &ctx, std::string name) {
-  for (std::string dir : ctx.arg.library_paths) {
-    for (std::string ext : {".tbd", ".dylib", ".a"}) {
-      std::string path = dir + "/lib" + name + ext;
-      if (MappedFile<Context<E>> *mf = MappedFile<Context<E>>::open(ctx, path))
-        return mf;
-    }
-  }
-  return nullptr;
-}
-
-template <typename E>
-static MappedFile<Context<E>> *
-strip_universal_header(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  FatHeader &hdr = *(FatHeader *)mf->data;
-  assert(hdr.magic == FAT_MAGIC);
-
-  FatArch *arch = (FatArch *)(mf->data + sizeof(hdr));
-  for (i64 i = 0; i < hdr.nfat_arch; i++)
-    if (arch[i].cputype == E::cputype)
-      return mf->slice(ctx, mf->name, arch[i].offset, arch[i].size);
-  Fatal(ctx) << mf->name << ": fat file contains no matching file";
-}
-
-template <typename E>
-static void read_file(Context<E> &ctx, MappedFile<Context<E>> *mf,
-                      bool is_needed = false) {
-  if (get_file_type(mf) == FileType::MACH_UNIVERSAL)
-    mf = strip_universal_header(ctx, mf);
-
-  switch (get_file_type(mf)) {
-  case FileType::TAPI:
-  case FileType::MACH_DYLIB:
-    ctx.dylibs.push_back(DylibFile<E>::create(ctx, mf));
-    if (is_needed || !ctx.arg.dead_strip_dylibs)
-      ctx.dylibs.back()->is_needed = true;
-    break;
-  case FileType::MACH_OBJ:
-    ctx.objs.push_back(ObjectFile<E>::create(ctx, mf, ""));
-    break;
-  case FileType::AR:
-    for (MappedFile<Context<E>> *child : read_archive_members(ctx, mf))
-      if (get_file_type(child) == FileType::MACH_OBJ)
-        ctx.objs.push_back(ObjectFile<E>::create(ctx, child, mf->name));
-    break;
-  default:
-    break;
-    // Fatal(ctx) << mf->name << ": unknown file type";
-  }
-}
-
-template <typename E>
-static std::vector<std::string>
-read_filelist(Context<E> &ctx, std::string arg) {
-  std::string path;
-  std::string dir;
-
-  if (size_t pos = arg.find(','); pos != arg.npos) {
-    path = arg.substr(0, pos);
-    dir = arg.substr(pos + 1) + "/";
-  } else {
-    path = arg;
-  }
-
-  MappedFile<Context<E>> *mf = MappedFile<Context<E>>::open(ctx, path);
-  if (!mf)
-    Fatal(ctx) << "-filepath: cannot open " << path;
-
-  std::vector<std::string> vec;
-  for (std::string_view str = mf->get_contents(); !str.empty();) {
-    std::string_view path;
-    std::tie(path, str) = split_string(str, '\n');
-    vec.push_back(path_clean(dir + std::string(path)));
-  }
-  return vec;
-}
-
-template <typename E>
-static void read_input_files(Context<E> &ctx, std::span<std::string> args) {
-  while (!args.empty()) {
-    if (args[0].starts_with("-filelist")) {
-      for (std::string &path : read_filelist(ctx, args[1])) {
-        MappedFile<Context<E>> *mf = MappedFile<Context<E>>::open(ctx, path);
-        if (!mf)
-          Fatal(ctx) << "-filepath " << args[1] << ": cannot open file: " << path;
-        read_file(ctx, mf);
-      }
-      args = args.subspan(2);
-    } else if (args[0] == "-framework" || args[0] == "-needed_framework") {
-      bool needed = (args[0] == "-needed_framework");
-      read_file(ctx, find_framework(ctx, args[1]), needed);
-      args = args.subspan(2);
-    } else if (args[0] == "-l" || args[0] == "-needed-l") {
-      MappedFile<Context<E>> *mf = find_library(ctx, args[1]);
-      if (!mf)
-        Fatal(ctx) << "library not found: -l" << args[1];
-      bool needed = (args[0] == "-needed-l");
-      read_file(ctx, mf, needed);
-      args = args.subspan(2);
-    } else {
-      read_file(ctx, MappedFile<Context<E>>::must_open(ctx, args[0]));
-      args = args.subspan(1);
-    }
-  }
-
-  if (ctx.objs.empty())
-    Fatal(ctx) << "no input files";
-}
-
-template <typename E>
-static int do_main(int argc, char **argv) {
-  Context<E> ctx;
-
-  if (argc > 1 && std::string_view(argv[1]) == "-dump") {
-    if (argc != 3)
-      Fatal(ctx) << "usage: ld64.mold -dump <executable-name>\n";
-    dump_file(argv[2]);
-    exit(0);
-  }
-
-  for (i64 i = 0; i < argc; i++)
-    ctx.cmdline_args.push_back(argv[i]);
-
-  std::vector<std::string> file_args;
-  parse_nonpositional_args(ctx, file_args);
-
-  if (ctx.arg.arch != E::cputype) {
-    if (ctx.arg.arch == CPU_TYPE_X86_64)
-      return do_main<X86_64>(argc, argv);
-  }
-
-  read_input_files(ctx, file_args);
-
-  i64 priority = 1;
-  for (ObjectFile<E> *file : ctx.objs)
-    file->priority = priority++;
-  for (DylibFile<E> *dylib : ctx.dylibs)
-    dylib->priority = priority++;
-
-  for (i64 i = 0; i < ctx.dylibs.size(); i++)
-    ctx.dylibs[i]->dylib_idx = i + 1;
-
-  for (ObjectFile<E> *file : ctx.objs)
-    file->parse(ctx);
-  for (DylibFile<E> *dylib : ctx.dylibs)
-    dylib->parse(ctx);
-
-  if (ctx.arg.ObjC)
-    for (ObjectFile<E> *file : ctx.objs)
-      if (!file->archive_name.empty() && file->is_objc_object(ctx))
-        file->is_alive = true;
-
-  for (ObjectFile<E> *file : ctx.objs) {
-    if (file->is_alive)
-      file->resolve_regular_symbols(ctx);
-    else
-      file->resolve_lazy_symbols(ctx);
-  }
-
-  std::vector<ObjectFile<E> *> live_objs;
-  for (ObjectFile<E> *file : ctx.objs)
-    if (file->is_alive)
-      live_objs.push_back(file);
-
-  for (i64 i = 0; i < live_objs.size(); i++)
-    append(live_objs, live_objs[i]->mark_live_objects(ctx));
-
-  for (DylibFile<E> *dylib : ctx.dylibs)
-    dylib->resolve_symbols(ctx);
-
-  if (ctx.output_type == MH_EXECUTE && !get_symbol(ctx, ctx.arg.entry)->file)
-    Error(ctx) << "undefined entry point symbol: " << ctx.arg.entry;
-
-  create_internal_file(ctx);
-
-  std::erase_if(ctx.objs, [](ObjectFile<E> *file) { return !file->is_alive; });
-  std::erase_if(ctx.dylibs, [](DylibFile<E> *file) { return !file->is_alive; });
-
-  if (ctx.arg.trace) {
-    for (ObjectFile<E> *file : ctx.objs)
-      SyncOut(ctx) << *file;
-    for (DylibFile<E> *file : ctx.dylibs)
-      SyncOut(ctx) << *file;
-  }
-
-  for (ObjectFile<E> *file : ctx.objs)
-    file->convert_common_symbols(ctx);
-
-  if (ctx.arg.dead_strip)
-    dead_strip(ctx);
-
-  create_synthetic_chunks(ctx);
-
-  for (ObjectFile<E> *file : ctx.objs)
-    file->check_duplicate_symbols(ctx);
-
-  for (i64 i = 0; i < ctx.segments.size(); i++)
-    ctx.segments[i]->seg_idx = i + 1;
-
-  for (ObjectFile<E> *file : ctx.objs)
-    for (std::unique_ptr<Subsection<E> > &subsec : file->subsections)
-      subsec->scan_relocations(ctx);
-
-  scan_unwind_info(ctx);
-
-  if (ctx.arg.dead_strip_dylibs)
-    std::erase_if(ctx.dylibs, [](DylibFile<E> *file) { return !file->is_needed; });
-
-  export_symbols(ctx);
-  i64 output_size = assign_offsets(ctx);
-  fix_synthetic_symbol_values(ctx);
-
-  ctx.output_file = OutputFile<E>::open(ctx, ctx.arg.output, output_size, 0777);
-  ctx.buf = ctx.output_file->buf;
-
-  for (std::unique_ptr<OutputSegment<E>> &seg : ctx.segments)
-    seg->copy_buf(ctx);
-  ctx.code_sig.write_signature(ctx);
-
-  ctx.output_file->close(ctx);
-  ctx.checkpoint();
-
-  if (!ctx.arg.map.empty())
-    print_map(ctx);
-  return 0;
-}
-
-int main(int argc, char **argv) {
-  return do_main<ARM64>(argc, argv);
-}
-
-}

macho/mapfile.cc

@@ -1,80 +0,0 @@
-#include "mold.h"
-
-#include <iomanip>
-#include <fstream>
-
-namespace mold::macho {
-
-struct Sym {
-  u64 addr;
-  u32 size;
-  u32 fileidx;
-  std::string_view name;
-};
-
-template <typename E>
-void print_map(Context<E> &ctx) {
-  std::ofstream out(ctx.arg.map.c_str());
-  if (!out.is_open())
-    Fatal(ctx) << "cannot open " << ctx.arg.map << ": " << errno_string();
-
-  out << "# Path: " << ctx.arg.output << "\n"
-      << "# Arch: x86-64\n"
-      << "# Object files:\n";
-
-  std::vector<Sym> syms;
-
-  for (i64 i = 0; i < ctx.objs.size(); i++){
-    ObjectFile<E> &file = *ctx.objs[i];
-
-    if (file.is_alive) {
-      out << "[" << std::setw(3) << i << "] " << file << "\n";
-      for (Symbol<E> *sym : file.syms)
-        if (sym->file == &file)
-          syms.push_back({sym->get_addr(ctx), 0, (u32)i, sym->name});
-    }
-  }
-
-  sort(syms, [](const Sym &a, const Sym &b) {
-    return a.addr < b.addr;
-  });
-
-  out << "# Sections:\n"
-      << "# Address       Size            Segment Section\n";
-
-  for (std::unique_ptr<OutputSegment<E>> &seg : ctx.segments) {
-    for (Chunk<E> *chunk : seg->chunks) {
-      if (!chunk->is_hidden) {
-        out << "0x" << std::hex
-            << std::setw(8) << std::setfill('0') << chunk->hdr.addr
-            << "     0x"
-            << std::setw(8) << std::setfill('0') << chunk->hdr.size
-            << "      " << std::setw(7) << std::left << std::setfill(' ')
-            << chunk->hdr.get_segname()
-            << " " << chunk->hdr.get_sectname() << "\n";
-      }
-    }
-  }
-
-  out << "# Symbols:\n"
-      << "# Address       Size            File  Name\n";
-
-  for (Sym &sym : syms) {
-    out << "0x" << std::hex
-        << std::setw(8) << std::setfill('0') << sym.addr
-        << "     0x"
-        << std::setw(8) << std::setfill('0') << 0
-        << "      ["
-        << std::setw(3) << std::right << std::setfill(' ') << std::dec
-        << sym.fileidx
-        << "] " << sym.name << "\n";
-  }
-}
-
-#define INSTANTIATE(E)                          \
-  template void print_map(Context<E> &)
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/mold.h

@@ -1,914 +0,0 @@
-#pragma once
-
-#include "macho.h"
-#include "../mold.h"
-
-#include <map>
-#include <memory>
-#include <span>
-#include <tbb/concurrent_hash_map.h>
-#include <tbb/spin_mutex.h>
-#include <unordered_map>
-#include <variant>
-
-namespace mold::macho {
-
-static constexpr i64 COMMON_PAGE_SIZE = 0x4000;
-static constexpr i64 SHA256_SIZE = 32;
-
-template <typename E> class Chunk;
-template <typename E> class InputSection;
-template <typename E> class OutputSection;
-template <typename E> class Subsection;
-template <typename E> struct Context;
-template <typename E> struct Symbol;
-
-//
-// object-file.cc
-//
-
-template <typename E>
-struct Relocation {
-  u32 offset = 0;
-  u8 type = -1;
-  u8 p2size = 0;
-  bool is_pcrel = false;
-  i64 addend = 0;
-  Symbol<E> *sym = nullptr;
-  Subsection<E> *subsec = nullptr;
-};
-
-template <typename E>
-struct UnwindRecord {
-  UnwindRecord(u32 len, u32 enc) : code_len(len), encoding(enc) {}
-
-  inline u64 get_func_raddr(Context<E> &ctx) const;
-
-  Subsection<E> *subsec = nullptr;
-  u32 offset = 0;
-  u32 code_len;
-  u32 encoding;
-  Symbol<E> *personality = nullptr;
-  Subsection<E> *lsda = nullptr;
-  u32 lsda_offset = 0;
-  bool is_alive = false;
-};
-
-template <typename E>
-class InputFile {
-public:
-  MappedFile<Context<E>> *mf = nullptr;
-  std::vector<Symbol<E> *> syms;
-  i64 priority = 0;
-  bool is_dylib = false;
-  std::atomic_bool is_alive = false;
-  std::string archive_name;
-
-protected:
-  InputFile() = default;
-};
-
-template <typename E>
-class ObjectFile : public InputFile<E> {
-public:
-  ObjectFile() = default;
-
-  static ObjectFile *create(Context<E> &ctx, MappedFile<Context<E>> *mf,
-                            std::string archive_name);
-  void parse(Context<E> &ctx);
-  Subsection<E> *find_subsection(Context<E> &ctx, u32 addr);
-  void parse_compact_unwind(Context<E> &ctx, MachSection &hdr);
-  void resolve_regular_symbols(Context<E> &ctx);
-  void resolve_lazy_symbols(Context<E> &ctx);
-  bool is_objc_object(Context<E> &ctx);
-  std::vector<ObjectFile *> mark_live_objects(Context<E> &ctx);
-  void convert_common_symbols(Context<E> &ctx);
-  void check_duplicate_symbols(Context<E> &ctx);
-
-  Relocation<E> read_reloc(Context<E> &ctx, const MachSection &hdr, MachRel r);
-
-  std::vector<std::unique_ptr<InputSection<E>>> sections;
-  std::vector<std::unique_ptr<Subsection<E>>> subsections;
-  std::vector<u32> sym_to_subsec;
-  std::span<MachSym> mach_syms;
-  std::vector<Symbol<E>> local_syms;
-  std::vector<UnwindRecord<E>> unwind_records;
-  std::span<DataInCodeEntry> data_in_code_entries;
-
-private:
-  void parse_sections(Context<E> &ctx);
-  void parse_symtab(Context<E> &ctx);
-  void split_subsections(Context<E> &ctx);
-  void parse_data_in_code(Context<E> &ctx);
-  LoadCommand *find_load_command(Context<E> &ctx, u32 type);
-  i64 find_subsection_idx(Context<E> &ctx, u32 addr);
-  void override_symbol(Context<E> &ctx, i64 symidx);
-  InputSection<E> *get_common_sec(Context<E> &ctx);
-
-  MachSection *unwind_sec = nullptr;
-  std::unique_ptr<MachSection> common_hdr;
-  InputSection<E> *common_sec = nullptr;
-};
-
-template <typename E>
-class DylibFile : public InputFile<E> {
-public:
-  static DylibFile *create(Context<E> &ctx, MappedFile<Context<E>> *mf);
-  void parse(Context<E> &ctx);
-  void resolve_symbols(Context<E> &ctx);
-
-  std::string_view install_name;
-  i64 dylib_idx = 0;
-  std::atomic_bool is_needed = false;
-
-private:
-  void parse_dylib(Context<E> &ctx);
-  void read_trie(Context<E> &ctx, u8 *start, i64 offset = 0,
-                 const std::string &prefix = "");
-
-  DylibFile() {
-    this->is_dylib = true;
-    this->is_alive = true;
-  }
-};
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const InputFile<E> &file);
-
-//
-// input-sections.cc
-//
-
-template <typename E>
-class InputSection {
-public:
-  InputSection(Context<E> &ctx, ObjectFile<E> &file, const MachSection &hdr);
-  void parse_relocations(Context<E> &ctx);
-
-  ObjectFile<E> &file;
-  const MachSection &hdr;
-  OutputSection<E> &osec;
-  std::string_view contents;
-  std::vector<Symbol<E> *> syms;
-  std::vector<Relocation<E>> rels;
-};
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const InputSection<E> &sec);
-
-template <typename E>
-class Subsection {
-public:
-  inline u64 get_addr(Context<E> &ctx) const;
-
-  std::string_view get_contents() {
-    assert(isec.hdr.type != S_ZEROFILL);
-    return isec.contents.substr(input_offset, input_size);
-  }
-
-  std::span<UnwindRecord<E>> get_unwind_records() {
-    return std::span<UnwindRecord<E>>(isec.file.unwind_records)
-      .subspan(unwind_offset, nunwind);
-  }
-
-  std::span<Relocation<E>> get_rels() const {
-    return std::span<Relocation<E>>(isec.rels).subspan(rel_offset, nrels);
-  }
-
-  void scan_relocations(Context<E> &ctx);
-  void apply_reloc(Context<E> &ctx, u8 *buf);
-
-  InputSection<E> &isec;
-  u32 input_offset = 0;
-  u32 input_size = 0;
-  u32 input_addr = 0;
-  u32 rel_offset = 0;
-  u32 nrels = 0;
-  u32 unwind_offset = 0;
-  u32 nunwind = 0;
-  u32 raddr = -1;
-  u16 p2align = 0;
-  std::atomic_bool is_alive = false;
-};
-
-template <typename E>
-std::vector<Relocation<E>>
-read_relocations(Context<E> &ctx, ObjectFile<E> &file, const MachSection &hdr);
-
-//
-// Symbol
-//
-
-enum {
-  NEEDS_GOT        = 1 << 0,
-  NEEDS_STUB       = 1 << 1,
-  NEEDS_THREAD_PTR = 1 << 2,
-};
-
-template <typename E>
-struct Symbol {
-  Symbol() = default;
-  Symbol(std::string_view name) : name(name) {}
-  Symbol(const Symbol<E> &other) : name(other.name) {}
-
-  std::string_view name;
-  InputFile<E> *file = nullptr;
-  Subsection<E> *subsec = nullptr;
-  u64 value = 0;
-
-  i32 stub_idx = -1;
-  i32 got_idx = -1;
-  i32 tlv_idx = -1;
-
-  tbb::spin_mutex mu;
-
-  std::atomic_uint8_t flags = 0;
-
-  u8 is_extern : 1 = false;
-  u8 is_lazy : 1 = false;
-  u8 is_common : 1 = false;
-  u8 referenced_dynamically : 1 = false;
-
-  inline u64 get_addr(Context<E> &ctx) const;
-  inline u64 get_got_addr(Context<E> &ctx) const;
-  inline u64 get_tlv_addr(Context<E> &ctx) const;
-};
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const Symbol<E> &sym);
-
-//
-// output-chunks.cc
-//
-
-template <typename E>
-class OutputSegment {
-public:
-  static OutputSegment<E> *
-  get_instance(Context<E> &ctx, std::string_view name);
-
-  void set_offset(Context<E> &ctx, i64 fileoff, u64 vmaddr);
-  void copy_buf(Context<E> &ctx);
-
-  std::string_view name;
-  SegmentCommand cmd = {};
-  i32 seg_idx = -1;
-  std::vector<Chunk<E> *> chunks;
-
-private:
-  OutputSegment(std::string_view name);
-};
-
-template <typename E>
-class Chunk {
-public:
-  inline Chunk(Context<E> &ctx, std::string_view segname,
-               std::string_view sectname);
-
-  virtual ~Chunk() = default;
-  virtual void compute_size(Context<E> &ctx) {};
-  virtual void copy_buf(Context<E> &ctx) {}
-
-  MachSection hdr = {};
-  u32 sect_idx = 0;
-  bool is_hidden = false;
-  bool is_regular = false;
-};
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const Chunk<E> &chunk);
-
-template <typename E>
-class OutputMachHeader : public Chunk<E> {
-public:
-  OutputMachHeader(Context<E> &ctx)
-    : Chunk<E>(ctx, "__TEXT", "__mach_header") {
-    this->is_hidden = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-};
-
-template <typename E>
-class OutputSection : public Chunk<E> {
-public:
-  static OutputSection<E> *
-  get_instance(Context<E> &ctx, std::string_view segname,
-               std::string_view sectname);
-
-  OutputSection(Context<E> &ctx, std::string_view segname,
-                std::string_view sectname)
-    : Chunk<E>(ctx, segname, sectname) {
-    this->is_regular = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  void add_subsec(Subsection<E> *subsec) {
-    members.push_back(subsec);
-    this->hdr.p2align = std::max<u32>(this->hdr.p2align, subsec->p2align);
-    this->hdr.attr |= subsec->isec.hdr.attr;
-    this->hdr.type = subsec->isec.hdr.type;
-  }
-
-  std::vector<Subsection<E> *> members;
-};
-
-class RebaseEncoder {
-public:
-  RebaseEncoder();
-  void add(i64 seg_idx, i64 offset);
-  void flush();
-  void finish();
-
-  std::vector<u8> buf;
-
-private:
-  i64 cur_seg = -1;
-  i64 cur_off = 0;
-  i64 times = 0;
-};
-
-template <typename E>
-class OutputRebaseSection : public Chunk<E> {
-public:
-  OutputRebaseSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__rebase") {
-    this->is_hidden = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<u8> contents;
-};
-
-class BindEncoder {
-public:
-  BindEncoder();
-  void add(i64 dylib_idx, std::string_view sym, i64 flags, i64 seg_idx,
-           i64 offset);
-  void finish();
-
-  std::vector<u8> buf;
-
-private:
-  std::string_view last_sym;
-  i64 last_flags = -1;
-  i64 last_dylib = -1;
-  i64 last_seg = -1;
-  i64 last_off = -1;
-};
-
-template <typename E>
-class OutputBindSection : public Chunk<E> {
-public:
-  OutputBindSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__binding") {
-    this->is_hidden = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<u8> contents;
-};
-
-template <typename E>
-class OutputLazyBindSection : public Chunk<E> {
-public:
-  OutputLazyBindSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__lazy_binding") {
-    this->is_hidden = true;
-    this->hdr.p2align = __builtin_ctz(8);
-  }
-
-  void add(Context<E> &ctx, Symbol<E> &sym, i64 flags);
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<u8> contents;
-};
-
-class ExportEncoder {
-public:
-  void add(std::string_view name, u32 flags, u64 addr);
-  i64 finish();
-
-  void write_trie(u8 *buf) {
-    write_trie(buf, root);
-  }
-
-private:
-  struct Entry {
-    std::string_view name;
-    u32 flags;
-    u64 addr;
-  };
-
-  struct TrieNode {
-    std::string_view prefix;
-    bool is_leaf = false;
-    u32 flags = 0;
-    u64 addr = 0;
-    u32 offset = -1;
-    std::vector<TrieNode> children;
-  };
-
-  static i64 common_prefix_len(std::span<Entry> entries, i64 len);
-  static TrieNode construct_trie(std::span<Entry> entries, i64 len);
-  static i64 set_offset(TrieNode &node, i64 offset);
-  void write_trie(u8 *buf, TrieNode &node);
-
-  TrieNode root;
-  std::vector<Entry> entries;
-};
-
-template <typename E>
-class OutputExportSection : public Chunk<E> {
-public:
-  OutputExportSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__export") {
-    this->is_hidden = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-private:
-  ExportEncoder enc;
-};
-
-template <typename E>
-class OutputFunctionStartsSection : public Chunk<E> {
-public:
-  OutputFunctionStartsSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__func_starts") {
-    this->is_hidden = true;
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<u8> contents;
-};
-
-template <typename E>
-class OutputSymtabSection : public Chunk<E> {
-public:
-  OutputSymtabSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__symbol_table") {
-    this->is_hidden = true;
-    this->hdr.p2align = __builtin_ctz(8);
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  struct Entry {
-    Symbol<E> *sym;
-    i64 stroff;
-  };
-
-  std::vector<Entry> locals;
-  std::vector<Entry> globals;
-  std::vector<Entry> undefs;
-};
-
-template <typename E>
-class OutputStrtabSection : public Chunk<E> {
-public:
-  OutputStrtabSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__string_table") {
-    this->is_hidden = true;
-    this->hdr.p2align = __builtin_ctz(8);
-  }
-
-  i64 add_string(std::string_view str);
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::string contents{1, '\0'};
-};
-
-template <typename E>
-class OutputIndirectSymtabSection : public Chunk<E> {
-public:
-  OutputIndirectSymtabSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__ind_sym_tab") {
-    this->is_hidden = true;
-  }
-
-  static constexpr i64 ENTRY_SIZE = 4;
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  struct Entry {
-    Symbol<E> *sym;
-    i64 symtab_idx;
-  };
-
-  std::vector<Entry> stubs;
-  std::vector<Entry> gots;
-};
-
-template <typename E>
-class CodeSignatureSection : public Chunk<E> {
-public:
-  CodeSignatureSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__code_signature") {
-    this->is_hidden = true;
-    this->hdr.p2align = __builtin_ctz(16);
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void write_signature(Context<E> &ctx);
-
-  static constexpr i64 BLOCK_SIZE = 4096;
-};
-
-template <typename E>
-class DataInCodeSection : public Chunk<E> {
-public:
-  DataInCodeSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__LINKEDIT", "__data_in_code") {
-    this->is_hidden = true;
-    this->hdr.p2align = __builtin_ctz(alignof(DataInCodeEntry));
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<DataInCodeEntry> contents;
-};
-
-template <typename E>
-class StubsSection : public Chunk<E> {
-public:
-  StubsSection(Context<E> &ctx) : Chunk<E>(ctx, "__TEXT", "__stubs") {
-    this->hdr.p2align = __builtin_ctz(2);
-    this->hdr.type = S_SYMBOL_STUBS;
-    this->hdr.attr = S_ATTR_SOME_INSTRUCTIONS | S_ATTR_PURE_INSTRUCTIONS;
-    this->hdr.reserved2 = E::stub_size;
-  }
-
-  void add(Context<E> &ctx, Symbol<E> *sym);
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<Symbol<E> *> syms;
-  std::vector<u32> bind_offsets;
-};
-
-template <typename E>
-class StubHelperSection : public Chunk<E> {
-public:
-  StubHelperSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__TEXT", "__stub_helper") {
-    this->hdr.p2align = __builtin_ctz(4);
-    this->hdr.attr = S_ATTR_SOME_INSTRUCTIONS | S_ATTR_PURE_INSTRUCTIONS;
-  }
-
-  void copy_buf(Context<E> &ctx) override;
-};
-
-template <typename E>
-class UnwindEncoder {
-public:
-  std::vector<u8> encode(Context<E> &ctx, std::span<UnwindRecord<E>> records);
-
-private:
-  u32 encode_personality(Context<E> &ctx, Symbol<E> *sym);
-
-  std::vector<std::span<UnwindRecord<E>>>
-  split_records(Context<E> &ctx, std::span<UnwindRecord<E>>);
-
-  std::vector<Symbol<E> *> personalities;
-};
-
-template <typename E>
-class UnwindInfoSection : public Chunk<E> {
-public:
-  UnwindInfoSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__TEXT", "__unwind_info") {
-    this->hdr.p2align = __builtin_ctz(4);
-  }
-
-  void compute_size(Context<E> &ctx) override;
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<u8> contents;
-};
-
-template <typename E>
-class GotSection : public Chunk<E> {
-public:
-  GotSection(Context<E> &ctx) : Chunk<E>(ctx, "__DATA_CONST", "__got") {
-    this->hdr.p2align = __builtin_ctz(8);
-    this->hdr.type = S_NON_LAZY_SYMBOL_POINTERS;
-  }
-
-  void add(Context<E> &ctx, Symbol<E> *sym);
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<Symbol<E> *> syms;
-};
-
-template <typename E>
-class LazySymbolPtrSection : public Chunk<E> {
-public:
-  LazySymbolPtrSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__DATA", "__la_symbol_ptr") {
-    this->hdr.p2align = __builtin_ctz(8);
-    this->hdr.type = S_LAZY_SYMBOL_POINTERS;
-  }
-
-  void copy_buf(Context<E> &ctx) override;
-};
-
-template <typename E>
-class ThreadPtrsSection : public Chunk<E> {
-public:
-  ThreadPtrsSection(Context<E> &ctx)
-    : Chunk<E>(ctx, "__DATA", "__thread_ptrs") {
-    this->hdr.p2align = __builtin_ctz(8);
-    this->hdr.type = S_THREAD_LOCAL_VARIABLE_POINTERS;
-  }
-
-  void add(Context<E> &ctx, Symbol<E> *sym);
-  void copy_buf(Context<E> &ctx) override;
-
-  std::vector<Symbol<E> *> syms;
-};
-
-//
-// mapfile.cc
-//
-
-template <typename E>
-void print_map(Context<E> &ctx);
-
-//
-// dumper.cc
-//
-
-void dump_file(std::string path);
-
-//
-// output-file.cc
-//
-
-template <typename E>
-class OutputFile {
-public:
-  static std::unique_ptr<OutputFile>
-  open(Context<E> &ctx, std::string path, i64 filesize, i64 perm);
-
-  virtual void close(Context<E> &ctx) = 0;
-  virtual ~OutputFile() {}
-
-  u8 *buf = nullptr;
-  std::string path;
-  i64 filesize;
-  bool is_mmapped;
-  bool is_unmapped = false;
-
-protected:
-  OutputFile(std::string path, i64 filesize, bool is_mmapped)
-    : path(path), filesize(filesize), is_mmapped(is_mmapped) {}
-};
-
-//
-// yaml.cc
-//
-
-struct YamlNode {
-  std::variant<std::string_view,
-               std::vector<YamlNode>,
-               std::map<std::string_view, YamlNode>> data;
-};
-
-struct YamlError {
-  std::string msg;
-  i64 pos;
-};
-
-std::variant<std::vector<YamlNode>, YamlError>
-parse_yaml(std::string_view str);
-
-//
-// tapi.cc
-//
-
-struct TextDylib {
-  std::string_view uuid;
-  std::string_view install_name;
-  std::string_view current_version = "1.0";
-  std::string_view parent_umbrella;
-  std::vector<std::string_view> reexported_libs;
-  std::vector<std::string_view> exports;
-};
-
-template <typename E>
-TextDylib parse_tbd(Context<E> &ctx, MappedFile<Context<E>> *mf);
-
-//
-// cmdline.cc
-//
-
-template <typename E>
-void parse_nonpositional_args(Context<E> &ctx,
-                              std::vector<std::string> &remaining);
-
-//
-// dead-strip.cc
-//
-
-template <typename E>
-void dead_strip(Context<E> &ctx);
-
-//
-// main.cc
-//
-
-template <typename E>
-struct Context {
-  Context() {
-    text_seg = OutputSegment<E>::get_instance(*this, "__TEXT");
-    data_const_seg = OutputSegment<E>::get_instance(*this, "__DATA_CONST");
-    data_seg = OutputSegment<E>::get_instance(*this, "__DATA");
-    linkedit_seg = OutputSegment<E>::get_instance(*this, "__LINKEDIT");
-
-    text = OutputSection<E>::get_instance(*this, "__TEXT", "__text");
-    data = OutputSection<E>::get_instance(*this, "__DATA", "__data");
-    bss = OutputSection<E>::get_instance(*this, "__DATA", "__bss");
-    cstring = OutputSection<E>::get_instance(*this, "__TEXT", "__cstring");
-    common = OutputSection<E>::get_instance(*this, "__DATA", "__common");
-
-    bss->hdr.type = S_ZEROFILL;
-    cstring->hdr.type = S_CSTRING_LITERALS;
-    common->hdr.type = S_ZEROFILL;
-  }
-
-  Context(const Context<E> &) = delete;
-
-  void checkpoint() {
-    if (has_error) {
-      cleanup();
-      _exit(1);
-    }
-  }
-
-  // Command-line arguments
-  struct {
-    bool ObjC = false;
-    bool adhoc_codesign = true;
-    bool color_diagnostics = false;
-    bool dead_strip = true;
-    bool dead_strip_dylibs = false;
-    bool deduplicate = true;
-    bool demangle = false;
-    bool dylib = false;
-    bool dynamic = true;
-    bool fatal_warnings = false;
-    bool trace = false;
-    i64 arch = CPU_TYPE_ARM64;
-    i64 headerpad = 256;
-    i64 pagezero_size = 0;
-    i64 platform = PLATFORM_MACOS;
-    i64 platform_min_version = 0;
-    i64 platform_sdk_version = 0;
-    std::string chroot;
-    std::string entry = "_main";
-    std::string map;
-    std::string output = "a.out";
-    std::vector<std::string> framework_paths;
-    std::vector<std::string> library_paths;
-    std::vector<std::string> rpath;
-    std::vector<std::string> syslibroot;
-  } arg;
-
-  std::vector<std::string_view> cmdline_args;
-  u32 output_type = MH_EXECUTE;
-
-  bool has_error = false;
-
-  tbb::concurrent_hash_map<std::string_view, Symbol<E>> symbol_map;
-
-  std::unique_ptr<OutputFile<E>> output_file;
-  u8 *buf;
-
-  tbb::concurrent_vector<std::unique_ptr<ObjectFile<E>>> obj_pool;
-  tbb::concurrent_vector<std::unique_ptr<DylibFile<E>>> dylib_pool;
-  tbb::concurrent_vector<std::unique_ptr<u8[]>> string_pool;
-  tbb::concurrent_vector<std::unique_ptr<MappedFile<Context<E>>>> mf_pool;
-  std::vector<std::unique_ptr<OutputSection<E>>> osec_pool;
-
-  tbb::concurrent_vector<std::unique_ptr<TimerRecord>> timer_records;
-
-  std::vector<ObjectFile<E> *> objs;
-  std::vector<DylibFile<E> *> dylibs;
-
-  OutputSegment<E> *text_seg = nullptr;
-  OutputSegment<E> *data_const_seg = nullptr;
-  OutputSegment<E> *data_seg = nullptr;
-  OutputSegment<E> *linkedit_seg = nullptr;
-
-  std::vector<std::unique_ptr<OutputSegment<E>>> segments;
-  std::vector<Chunk<E> *> chunks;
-
-  OutputMachHeader<E> mach_hdr{*this};
-  StubsSection<E> stubs{*this};
-  StubHelperSection<E> stub_helper{*this};
-  UnwindInfoSection<E> unwind_info{*this};
-  GotSection<E> got{*this};
-  LazySymbolPtrSection<E> lazy_symbol_ptr{*this};
-  CodeSignatureSection<E> code_sig{*this};
-  DataInCodeSection<E> data_in_code{*this};
-  ThreadPtrsSection<E> thread_ptrs{*this};
-
-  OutputRebaseSection<E> rebase{*this};
-  OutputBindSection<E> bind{*this};
-  OutputLazyBindSection<E> lazy_bind{*this};
-  OutputExportSection<E> export_{*this};
-  OutputFunctionStartsSection<E> function_starts{*this};
-  OutputSymtabSection<E> symtab{*this};
-  OutputIndirectSymtabSection<E> indir_symtab{*this};
-  OutputStrtabSection<E> strtab{*this};
-
-  OutputSection<E> *text = nullptr;
-  OutputSection<E> *data = nullptr;
-  OutputSection<E> *bss = nullptr;
-  OutputSection<E> *cstring = nullptr;
-  OutputSection<E> *common = nullptr;
-};
-
-int main(int argc, char **argv);
-
-//
-// Inline functions
-//
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const InputSection<E> &sec) {
-  out << sec.file << "(" << sec.hdr.get_segname() << ","
-      << sec.hdr.get_sectname() << ")";
-  return out;
-}
-
-template <typename E>
-u64 Subsection<E>::get_addr(Context<E> &ctx) const {
-  return ctx.arg.pagezero_size + raddr;
-}
-
-template <typename E>
-u64 Symbol<E>::get_addr(Context<E> &ctx) const {
-  if (subsec)
-    return subsec->get_addr(ctx) + value;
-  if (stub_idx != -1)
-    return ctx.stubs.hdr.addr + stub_idx * E::stub_size;
-  return value;
-}
-
-template <typename E>
-u64 Symbol<E>::get_got_addr(Context<E> &ctx) const {
-  assert(got_idx != -1);
-  return ctx.got.hdr.addr + got_idx * E::word_size;
-}
-
-template <typename E>
-u64 Symbol<E>::get_tlv_addr(Context<E> &ctx) const {
-  assert(tlv_idx != -1);
-  return ctx.thread_ptrs.hdr.addr + tlv_idx * E::word_size;
-}
-
-template <typename E>
-inline Symbol<E> *get_symbol(Context<E> &ctx, std::string_view name) {
-  typename decltype(ctx.symbol_map)::const_accessor acc;
-  ctx.symbol_map.insert(acc, {name, Symbol<E>(name)});
-  return (Symbol<E> *)(&acc->second);
-}
-
-template <typename E>
-inline std::ostream &operator<<(std::ostream &out, const Symbol<E> &sym) {
-  out << sym.name;
-  return out;
-}
-
-template <typename E>
-Chunk<E>::Chunk(Context<E> &ctx, std::string_view segname,
-                std::string_view sectname) {
-  ctx.chunks.push_back(this);
-  hdr.set_segname(segname);
-  hdr.set_sectname(sectname);
-}
-
-template <typename E>
-u64 UnwindRecord<E>::get_func_raddr(Context<E> &ctx) const {
-  return subsec->raddr + offset;
-}
-
-} // namespace mold::macho

macho/object-file.cc

@@ -1,621 +0,0 @@
-#include "mold.h"
-
-#include "../archive-file.h"
-
-namespace mold::macho {
-
-template <typename E>
-std::ostream &operator<<(std::ostream &out, const InputFile<E> &file) {
-  if (file.archive_name.empty())
-    out << path_clean(file.mf->name);
-  else
-    out << path_clean(file.archive_name) << "(" << path_clean(file.mf->name) + ")";
-  return out;
-}
-
-template <typename E>
-ObjectFile<E> *
-ObjectFile<E>::create(Context<E> &ctx, MappedFile<Context<E>> *mf,
-                      std::string archive_name) {
-  ObjectFile<E> *obj = new ObjectFile<E>;
-  obj->mf = mf;
-  obj->archive_name = archive_name;
-  obj->is_alive = archive_name.empty();
-  ctx.obj_pool.push_back(std::unique_ptr<ObjectFile<E>>(obj));
-  return obj;
-};
-
-template <typename E>
-void ObjectFile<E>::parse(Context<E> &ctx) {
-  parse_sections(ctx);
-  parse_symtab(ctx);
-  split_subsections(ctx);
-  parse_data_in_code(ctx);
-
-  for (std::unique_ptr<InputSection<E>> &isec : sections)
-    if (isec)
-      isec->parse_relocations(ctx);
-
-  if (unwind_sec)
-    parse_compact_unwind(ctx, *unwind_sec);
-}
-
-template <typename E>
-void ObjectFile<E>::parse_sections(Context<E> &ctx) {
-  MachHeader &hdr = *(MachHeader *)this->mf->data;
-  u8 *p = this->mf->data + sizeof(hdr);
-
-  // Read all but a symtab section
-  for (i64 i = 0; i < hdr.ncmds; i++) {
-    LoadCommand &lc = *(LoadCommand *)p;
-    p += lc.cmdsize;
-    if (lc.cmd != LC_SEGMENT_64)
-      continue;
-
-    SegmentCommand &cmd = *(SegmentCommand *)&lc;
-    MachSection *mach_sec = (MachSection *)((u8 *)&lc + sizeof(cmd));
-
-    for (MachSection &msec : std::span(mach_sec, mach_sec + cmd.nsects)) {
-      sections.push_back(nullptr);
-
-      if (msec.match("__LD", "__compact_unwind")) {
-        unwind_sec = &msec;
-        continue;
-      }
-
-      if (msec.attr & S_ATTR_DEBUG)
-        continue;
-
-      sections.back().reset(new InputSection<E>(ctx, *this, msec));
-    }
-  }
-}
-
-template <typename E>
-void ObjectFile<E>::parse_symtab(Context<E> &ctx) {
-  SymtabCommand *cmd = (SymtabCommand *)find_load_command(ctx, LC_SYMTAB);
-  if (!cmd)
-    return;
-
-  mach_syms = {(MachSym *)(this->mf->data + cmd->symoff), cmd->nsyms};
-  this->syms.reserve(mach_syms.size());
-
-  i64 nlocal = 0;
-  for (MachSym &msym : mach_syms)
-    if (!msym.ext)
-      nlocal++;
-  local_syms.reserve(nlocal);
-
-  for (MachSym &msym : mach_syms) {
-    std::string_view name = (char *)(this->mf->data + cmd->stroff + msym.stroff);
-
-    if (msym.ext) {
-      this->syms.push_back(get_symbol(ctx, name));
-    } else {
-      local_syms.emplace_back(name);
-      this->syms.push_back(&local_syms.back());
-    }
-  }
-}
-
-struct SplitRegion {
-  u32 offset;
-  u32 size;
-  u32 symidx;
-  bool is_alt_entry;
-};
-
-template <typename E>
-struct SplitInfo {
-  InputSection<E> *isec = nullptr;
-  std::vector<SplitRegion> regions;
-};
-
-template <typename E>
-static std::vector<SplitInfo<E>> split(Context<E> &ctx, ObjectFile<E> &file) {
-  std::vector<SplitInfo<E>> vec;
-
-  for (std::unique_ptr<InputSection<E>> &isec : file.sections)
-    vec.push_back({isec.get()});
-
-  for (i64 i = 0; i < file.mach_syms.size(); i++) {
-    MachSym &msym = file.mach_syms[i];
-    if (msym.type == N_SECT && file.sections[msym.sect - 1]) {
-      SplitRegion r;
-      r.offset = msym.value - file.sections[msym.sect - 1]->hdr.addr;
-      r.symidx = i;
-      r.is_alt_entry = (msym.desc & N_ALT_ENTRY);
-      vec[msym.sect - 1].regions.push_back(r);
-    }
-  }
-
-  std::erase_if(vec, [](const SplitInfo<E> &info) { return !info.isec; });
-
-  sort(vec, [](const SplitInfo<E> &a, const SplitInfo<E> &b) {
-    return a.isec->hdr.addr < b.isec->hdr.addr;
-  });
-
-  for (SplitInfo<E> &info : vec) {
-    std::vector<SplitRegion> &r = info.regions;
-
-    if (r.empty()) {
-      r.push_back({0, (u32)info.isec->hdr.size, (u32)-1, false});
-      continue;
-    }
-
-    sort(r, [](const SplitRegion &a, const SplitRegion &b) {
-      return a.offset < b.offset;
-    });
-
-    if (r[0].offset > 0)
-      r.insert(r.begin(), {0, r[0].offset, (u32)-1, false});
-
-    for (i64 i = 1; i < r.size(); i++)
-      if (r[i - 1].offset == r[i].offset)
-        r[i++].is_alt_entry = true;
-
-    i64 last = -1;
-    for (i64 i = 0; i < r.size(); i++) {
-      if (!r[i].is_alt_entry) {
-        if (last != -1)
-          r[last].size = r[i].offset - r[last].offset;
-        last = i;
-      }
-    }
-
-    if (last != -1)
-      r[last].size = info.isec->hdr.size - r[last].offset;
-  }
-  return vec;
-}
-
-template <typename E>
-void ObjectFile<E>::split_subsections(Context<E> &ctx) {
-  sym_to_subsec.resize(mach_syms.size());
-
-  for (SplitInfo<E> &info : split(ctx, *this)) {
-    InputSection<E> &isec = *info.isec;
-
-    for (SplitRegion &r : info.regions) {
-      if (!r.is_alt_entry) {
-        Subsection<E> *subsec = new Subsection<E>{
-          .isec = isec,
-          .input_offset = r.offset,
-          .input_size = r.size,
-          .input_addr = (u32)(isec.hdr.addr + r.offset),
-          .p2align = (u8)isec.hdr.p2align,
-        };
-        subsections.push_back(std::unique_ptr<Subsection<E>>(subsec));
-      }
-
-      if (r.symidx != -1)
-        sym_to_subsec[r.symidx] = subsections.size() - 1;
-    }
-  }
-
-  for (i64 i = 0; i < mach_syms.size(); i++)
-    if (!mach_syms[i].ext)
-      override_symbol(ctx, i);
-}
-
-template <typename E>
-void ObjectFile<E>::parse_data_in_code(Context<E> &ctx) {
-  if (auto *cmd = (LinkEditDataCommand *)find_load_command(ctx, LC_DATA_IN_CODE))
-    data_in_code_entries = {
-      (DataInCodeEntry *)(this->mf->data + cmd->dataoff),
-      cmd->datasize / sizeof(DataInCodeEntry)
-    };
-}
-
-template <typename E>
-LoadCommand *ObjectFile<E>::find_load_command(Context<E> &ctx, u32 type) {
-  MachHeader &hdr = *(MachHeader *)this->mf->data;
-  u8 *p = this->mf->data + sizeof(hdr);
-
-  for (i64 i = 0; i < hdr.ncmds; i++) {
-    LoadCommand &lc = *(LoadCommand *)p;
-    if (lc.cmd == type)
-      return &lc;
-    p += lc.cmdsize;
-  }
-  return nullptr;
-}
-
-template <typename E>
-i64 ObjectFile<E>::find_subsection_idx(Context<E> &ctx, u32 addr) {
-  auto it = std::upper_bound(
-      subsections.begin(), subsections.end(), addr,
-      [&](u32 addr, const std::unique_ptr<Subsection<E>> &subsec) {
-    return addr < subsec->input_addr;
-  });
-
-  if (it == subsections.begin())
-    return -1;
-  return it - subsections.begin() - 1;
-}
-
-template <typename E>
-Subsection<E> *ObjectFile<E>::find_subsection(Context<E> &ctx, u32 addr) {
-  i64 i = find_subsection_idx(ctx, addr);
-  return (i == -1) ? nullptr : subsections[i].get();
-}
-
-template <typename E>
-void ObjectFile<E>::parse_compact_unwind(Context<E> &ctx, MachSection &hdr) {
-  if (hdr.size % sizeof(CompactUnwindEntry))
-    Fatal(ctx) << *this << ": invalid __compact_unwind section size";
-
-  i64 num_entries = hdr.size / sizeof(CompactUnwindEntry);
-  unwind_records.reserve(num_entries);
-
-  CompactUnwindEntry *src = (CompactUnwindEntry *)(this->mf->data + hdr.offset);
-
-  // Read compact unwind entries
-  for (i64 i = 0; i < num_entries; i++)
-    unwind_records.emplace_back(src[i].code_len, src[i].encoding);
-
-  // Read relocations
-  MachRel *mach_rels = (MachRel *)(this->mf->data + hdr.reloff);
-  for (i64 i = 0; i < hdr.nreloc; i++) {
-    MachRel &r = mach_rels[i];
-    if (r.offset >= hdr.size)
-      Fatal(ctx) << *this << ": relocation offset too large: " << i;
-
-    i64 idx = r.offset / sizeof(CompactUnwindEntry);
-    UnwindRecord<E> &dst = unwind_records[idx];
-
-    auto error = [&]() {
-      Fatal(ctx) << *this << ": __compact_unwind: unsupported relocation: " << i;
-    };
-
-    switch (r.offset % sizeof(CompactUnwindEntry)) {
-    case offsetof(CompactUnwindEntry, code_start): {
-      if (r.is_pcrel || r.p2size != 3 || r.is_extern || r.type)
-        error();
-
-      Subsection<E> *target = find_subsection(ctx, src[idx].code_start);
-      if (!target)
-        error();
-      dst.subsec = target;
-      dst.offset = src[idx].code_start - target->input_addr;
-      break;
-    }
-    case offsetof(CompactUnwindEntry, personality):
-      if (r.is_pcrel || r.p2size != 3 || !r.is_extern || r.type)
-        error();
-      dst.personality = this->syms[r.idx];
-      break;
-    case offsetof(CompactUnwindEntry, lsda): {
-      if (r.is_pcrel || r.p2size != 3 || r.is_extern || r.type)
-        error();
-
-      i32 addr = *(i32 *)((u8 *)this->mf->data + hdr.offset + r.offset);
-      Subsection<E> *target = find_subsection(ctx, addr);
-      if (!target)
-        error();
-      dst.lsda = target;
-      dst.lsda_offset = addr - target->input_addr;
-      break;
-    }
-    default:
-      Fatal(ctx) << *this << ": __compact_unwind: unsupported relocation: " << i;
-    }
-  }
-
-  for (i64 i = 0; i < num_entries; i++)
-    if (!unwind_records[i].subsec)
-      Fatal(ctx) << ": __compact_unwind: missing relocation at " << i;
-
-  // Sort unwind entries by offset
-  sort(unwind_records, [](const UnwindRecord<E> &a, const UnwindRecord<E> &b) {
-    return std::tuple(a.subsec->input_addr, a.offset) <
-           std::tuple(b.subsec->input_addr, b.offset);
-  });
-
-  // Associate unwind entries to subsections
-  for (i64 i = 0; i < num_entries;) {
-    Subsection<E> &subsec = *unwind_records[i].subsec;
-    subsec.unwind_offset = i;
-
-    i64 j = i + 1;
-    while (j < num_entries && unwind_records[j].subsec == &subsec)
-      j++;
-    subsec.nunwind = j - i;
-    i = j;
-  }
-}
-
-// Symbols with higher priorities overwrites symbols with lower priorities.
-// Here is the list of priorities, from the highest to the lowest.
-//
-//  1. Strong defined symbol
-//  2. Weak defined symbol
-//  3. Strong defined symbol in a DSO
-//  4. Weak defined symbol in a DSO
-//  5. Strong or weak defined symbol in an archive
-//  6. Common symbol
-//  7. Unclaimed (nonexistent) symbol
-//
-// Ties are broken by file priority.
-template <typename E>
-static u64 get_rank(InputFile<E> *file, MachSym &msym, bool is_lazy) {
-  if (msym.is_common())
-    return (6 << 24) + file->priority;
-  if (is_lazy)
-    return (5 << 24) + file->priority;
-  if (file->is_dylib)
-    return (3 << 24) + file->priority;
-  return (1 << 24) + file->priority;
-}
-
-template <typename E>
-static u64 get_rank(Symbol<E> &sym) {
-  InputFile<E> *file = sym.file;
-  if (!file)
-    return 7 << 24;
-  if (sym.is_common)
-    return (6 << 24) + file->priority;
-  if (!file->archive_name.empty())
-    return (5 << 24) + file->priority;
-  if (file->is_dylib)
-    return (3 << 24) + file->priority;
-  return (1 << 24) + file->priority;
-}
-
-template <typename E>
-void ObjectFile<E>::override_symbol(Context<E> &ctx, i64 idx) {
-  Symbol<E> &sym = *this->syms[idx];
-  MachSym &msym = mach_syms[idx];
-
-  sym.file = this;
-  sym.is_extern = msym.ext;
-  sym.is_lazy = false;
-
-  switch (msym.type) {
-  case N_UNDF:
-    assert(msym.is_common());
-    sym.subsec = nullptr;
-    sym.value = msym.value;
-    sym.is_common = true;
-    break;
-  case N_ABS:
-    sym.subsec = nullptr;
-    sym.value = msym.value;
-    sym.is_common = false;
-    break;
-  case N_SECT:
-    sym.subsec = subsections[sym_to_subsec[idx]].get();
-    sym.value = msym.value - sym.subsec->input_addr;
-    sym.is_common = false;
-    break;
-  default:
-    Fatal(ctx) << sym << ": unknown symbol type: " << (u64)msym.type;
-  }
-}
-
-template <typename E>
-void ObjectFile<E>::resolve_regular_symbols(Context<E> &ctx) {
-  for (i64 i = 0; i < this->syms.size(); i++) {
-    MachSym &msym = mach_syms[i];
-    if (!msym.ext || msym.is_undef())
-      continue;
-
-    Symbol<E> &sym = *this->syms[i];
-    std::scoped_lock lock(sym.mu);
-    if (get_rank(this, msym, false) < get_rank(sym))
-      override_symbol(ctx, i);
-  }
-}
-
-template <typename E>
-void ObjectFile<E>::resolve_lazy_symbols(Context<E> &ctx) {
-  for (i64 i = 0; i < this->syms.size(); i++) {
-    MachSym &msym = mach_syms[i];
-    if (!msym.ext || msym.is_undef() || msym.is_common())
-      continue;
-
-    Symbol<E> &sym = *this->syms[i];
-    std::scoped_lock lock(sym.mu);
-
-    if (get_rank(this, msym, true) < get_rank(sym)) {
-      sym.file = this;
-      sym.subsec = nullptr;
-      sym.value = 0;
-      sym.is_extern = false;
-      sym.is_lazy = true;
-      sym.is_common = false;
-    }
-  }
-}
-
-template <typename E>
-bool ObjectFile<E>::is_objc_object(Context<E> &ctx) {
-  for (std::unique_ptr<InputSection<E>> &isec : sections)
-    if (isec->hdr.match("__DATA", "__objc_catlist") ||
-        isec->hdr.match("__TEXT", "__swift"))
-      return true;
-
-  for (i64 i = 0; i < this->syms.size(); i++)
-    if (!mach_syms[i].is_undef() && mach_syms[i].ext &&
-        this->syms[i]->name.starts_with("_OBJC_CLASS_$_"))
-      return true;
-  return false;
-}
-
-template <typename E>
-std::vector<ObjectFile<E> *> ObjectFile<E>::mark_live_objects(Context<E> &ctx) {
-  std::vector<ObjectFile<E> *> vec;
-  assert(this->is_alive);
-
-  for (i64 i = 0; i < this->syms.size(); i++) {
-    MachSym &msym = mach_syms[i];
-    if (!msym.ext)
-      continue;
-
-    Symbol<E> &sym = *this->syms[i];
-    std::scoped_lock lock(sym.mu);
-
-    if (msym.is_undef()) {
-      if (sym.file && !sym.file->is_alive.exchange(true))
-        vec.push_back((ObjectFile<E> *)sym.file);
-      continue;
-    }
-
-    if (get_rank(this, msym, false) < get_rank(sym))
-      override_symbol(ctx, i);
-  }
-  return vec;
-}
-
-template <typename E>
-void ObjectFile<E>::convert_common_symbols(Context<E> &ctx) {
-  for (i64 i = 0; i < this->syms.size(); i++) {
-    Symbol<E> &sym = *this->syms[i];
-    MachSym &msym = mach_syms[i];
-
-    if (sym.file == this && sym.is_common) {
-      InputSection<E> *isec = get_common_sec(ctx);
-      Subsection<E> *subsec = new Subsection<E>{
-        .isec = *isec,
-        .input_size = (u32)msym.value,
-        .p2align = (u8)msym.p2align,
-      };
-
-      subsections.push_back(std::unique_ptr<Subsection<E>>(subsec));
-
-      sym.subsec = subsec;
-      sym.value = 0;
-      sym.is_common = false;
-    }
-  }
-}
-
-template <typename E>
-void ObjectFile<E>::check_duplicate_symbols(Context<E> &ctx) {
-  for (i64 i = 0; i < this->syms.size(); i++) {
-    Symbol<E> *sym = this->syms[i];
-    MachSym &msym = mach_syms[i];
-    if (sym && !msym.is_undef() && !msym.is_common() && sym->file != this)
-      Error(ctx) << "duplicate symbol: " << *this << ": " << *sym->file
-                 << ": " << *sym;
-  }
-}
-
-template <typename E>
-InputSection<E> *ObjectFile<E>::get_common_sec(Context<E> &ctx) {
-  if (!common_sec) {
-    MachSection *hdr = new MachSection;
-    common_hdr.reset(hdr);
-
-    memset(hdr, 0, sizeof(*hdr));
-    hdr->set_segname("__DATA");
-    hdr->set_sectname("__common");
-    hdr->type = S_ZEROFILL;
-
-    common_sec = new InputSection<E>(ctx, *this, *hdr);
-    sections.push_back(std::unique_ptr<InputSection<E>>(common_sec));
-  }
-  return common_sec;
-}
-
-template <typename E>
-DylibFile<E> *DylibFile<E>::create(Context<E> &ctx, MappedFile<Context<E>> *mf) {
-  DylibFile<E> *dylib = new DylibFile<E>;
-  dylib->mf = mf;
-  ctx.dylib_pool.push_back(std::unique_ptr<DylibFile<E>>(dylib));
-  return dylib;
-};
-
-template <typename E>
-void DylibFile<E>::read_trie(Context<E> &ctx, u8 *start, i64 offset,
-                             const std::string &prefix) {
-  u8 *buf = start + offset;
-
-  if (*buf) {
-    read_uleb(buf); // size
-    read_uleb(buf); // flags
-    read_uleb(buf); // addr
-    this->syms.push_back(get_symbol(ctx, prefix));
-  } else {
-    buf++;
-  }
-
-  i64 nchild = *buf++;
-
-  for (i64 i = 0; i < nchild; i++) {
-    std::string suffix((char *)buf);
-    buf += suffix.size() + 1;
-    i64 off = read_uleb(buf);
-    read_trie(ctx, start, off, prefix + suffix);
-  }
-}
-
-template <typename E>
-void DylibFile<E>::parse_dylib(Context<E> &ctx) {
-  MachHeader &hdr = *(MachHeader *)this->mf->data;
-  u8 *p = this->mf->data + sizeof(hdr);
-
-  for (i64 i = 0; i < hdr.ncmds; i++) {
-    LoadCommand &lc = *(LoadCommand *)p;
-
-    switch (lc.cmd) {
-    case LC_ID_DYLIB: {
-      DylibCommand &cmd = *(DylibCommand *)p;
-      install_name = (char *)p + cmd.nameoff;
-      break;
-    }
-    case LC_DYLD_INFO_ONLY: {
-      DyldInfoCommand &cmd = *(DyldInfoCommand *)p;
-      if (cmd.export_off)
-        read_trie(ctx, this->mf->data + cmd.export_off);
-      break;
-    }
-    case LC_DYLD_EXPORTS_TRIE: {
-      LinkEditDataCommand &cmd = *(LinkEditDataCommand *)p;
-      read_trie(ctx, this->mf->data + cmd.dataoff);
-      break;
-    }
-    }
-
-    p += lc.cmdsize;
-  }
-}
-
-template <typename E>
-void DylibFile<E>::parse(Context<E> &ctx) {
-  switch (get_file_type(this->mf)) {
-  case FileType::TAPI: {
-    TextDylib tbd = parse_tbd(ctx, this->mf);
-    for (std::string_view sym : tbd.exports)
-      this->syms.push_back(get_symbol(ctx, sym));
-    install_name = tbd.install_name;
-    break;
-  }
-  case FileType::MACH_DYLIB:
-    parse_dylib(ctx);
-    break;
-  default:
-    Fatal(ctx) << this->mf->name << ": is not a dylib";
-  }
-}
-
-template <typename E>
-void DylibFile<E>::resolve_symbols(Context<E> &ctx) {
-  for (Symbol<E> *sym : this->syms) {
-    std::scoped_lock lock(sym->mu);
-    if (sym->file && sym->file->priority < this->priority)
-      continue;
-    sym->file = this;
-    sym->is_extern = true;
-  }
-}
-
-#define INSTANTIATE(E)                                                  \
-  template class ObjectFile<E>;                                         \
-  template class DylibFile<E>;                                          \
-  template std::ostream &operator<<(std::ostream &, const InputFile<E> &)
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/output-file.cc

@@ -1,120 +0,0 @@
-// This file implements the same functionality as elf/output-file.cc.
-// We have two different implemetntations for ELF and Mach-O because
-// we implemented a special optimization for ELF or Linux.
-//
-// On Linux, it is faster to write to an existing file than creating a
-// new file and writing to it. So, we overwrite an existing executable
-// if exists.
-//
-// On macOS, it looks like the system does not assume that an executable
-// is mutated once it is created. Due to some code signing mechanism or
-// something, we always have to create a fresh file.
-
-#include "mold.h"
-
-#include <fcntl.h>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-namespace mold::macho {
-
-template <typename E>
-class MemoryMappedOutputFile : public OutputFile<E> {
-public:
-  MemoryMappedOutputFile(Context<E> &ctx, std::string path, i64 filesize, i64 perm)
-    : OutputFile<E>(path, filesize, true) {
-    std::string tmp_path = filepath(path).parent_path() / ".mold-XXXXXX";
-    output_tmpfile = (char *)save_string(ctx, tmp_path).data();
-
-    i64 fd = mkstemp(output_tmpfile);
-    if (fd == -1)
-      Fatal(ctx) << "cannot open " << output_tmpfile <<  ": " << errno_string();
-
-    if (ftruncate(fd, filesize))
-      Fatal(ctx) << "ftruncate failed: " << errno_string();
-    if (fchmod(fd, perm) == -1)
-      Fatal(ctx) << "fchmod failed: " << errno_string();
-
-    this->buf = (u8 *)mmap(nullptr, filesize, PROT_READ | PROT_WRITE,
-                           MAP_SHARED, fd, 0);
-    if (this->buf == MAP_FAILED)
-      Fatal(ctx) << path << ": mmap failed: " << errno_string();
-    ::close(fd);
-  }
-
-  void close(Context<E> &ctx) override {
-    Timer t(ctx, "close_file");
-
-    munmap(this->buf, this->filesize);
-    if (rename(output_tmpfile, this->path.c_str()) == -1)
-      Fatal(ctx) << this->path << ": rename failed: " << errno_string();
-    output_tmpfile = nullptr;
-  }
-};
-
-template <typename E>
-class MallocOutputFile : public OutputFile<E> {
-public:
-  MallocOutputFile(Context<E> &ctx, std::string path, i64 filesize, i64 perm)
-    : OutputFile<E>(path, filesize, false), perm(perm) {
-    this->buf = (u8 *)mmap(NULL, filesize, PROT_READ | PROT_WRITE,
-                           MAP_SHARED | MAP_ANONYMOUS, -1, 0);
-    if (this->buf == MAP_FAILED)
-      Fatal(ctx) << "mmap failed: " << errno_string();
-  }
-
-  void close(Context<E> &ctx) override {
-    Timer t(ctx, "close_file");
-
-    if (this->path == "-") {
-      fwrite(this->buf, this->filesize, 1, stdout);
-      fclose(stdout);
-      return;
-    }
-
-    i64 fd = ::open(this->path.c_str(), O_RDWR | O_CREAT, perm);
-    if (fd == -1)
-      Fatal(ctx) << "cannot open " << this->path << ": " << errno_string();
-
-    FILE *fp = fdopen(fd, "w");
-    fwrite(this->buf, this->filesize, 1, fp);
-    fclose(fp);
-  }
-
-private:
-  i64 perm;
-};
-
-template <typename E>
-std::unique_ptr<OutputFile<E>>
-OutputFile<E>::open(Context<E> &ctx, std::string path, i64 filesize, i64 perm) {
-  Timer t(ctx, "open_file");
-
-  if (path.starts_with('/') && !ctx.arg.chroot.empty())
-    path = ctx.arg.chroot + "/" + path_clean(path);
-
-  bool is_special = false;
-  if (path == "-") {
-    is_special = true;
-  } else {
-    struct stat st;
-    if (stat(path.c_str(), &st) == 0 && (st.st_mode & S_IFMT) != S_IFREG)
-      is_special = true;
-  }
-
-  std::unique_ptr<OutputFile> file;
-  if (is_special)
-    file = std::make_unique<MallocOutputFile<E>>(ctx, path, filesize, perm);
-  else
-    file = std::make_unique<MemoryMappedOutputFile<E>>(ctx, path, filesize, perm);
-  return file;
-}
-
-#define INSTANTIATE(E)                          \
-  template class OutputFile<E>
-
-INSTANTIATE(ARM64);
-INSTANTIATE(X86_64);
-
-} // namespace mold::macho

macho/tapi.cc

@@ -1,127 +0,0 @@
-#include "mold.h"
-
-#include <optional>
-
-namespace mold::macho {
-
-static std::vector<YamlNode>
-get_vector(YamlNode &node, std::string_view key) {
-  if (auto *map = std::get_if<std::map<std::string_view, YamlNode>>(&node.data))
-    if (auto it = map->find(key); it != map->end())
-      if (auto *vec = std::get_if<std::vector<YamlNode>>(&it->second.data))
-        return *vec;
-  return {};
-}
-
-static std::optional<std::string_view>
-get_string(YamlNode &node, std::string_view key) {
-  if (auto *map = std::get_if<std::map<std::string_view, YamlNode>>(&node.data))
-    if (auto it = map->find(key); it != map->end())
-      if (auto *str = std::get_if<std::string_view>(&it->second.data))
-        return *str;
-  return {};
-}
-
-static bool contains(const std::vector<YamlNode> &vec, std::string_view key) {
-  for (const YamlNode &mem : vec)
-    if (const std::string_view *s = std::get_if<std::string_view>(&mem.data))
-      if (*s == key)
-        return true;
-  return false;
-}
-
-static std::optional<TextDylib> to_tbd(YamlNode &node, std::string_view arch) {
-  if (!contains(get_vector(node, "targets"), arch))
-    return {};
-
-  TextDylib tbd;
-
-  for (YamlNode &mem : get_vector(node, "uuids"))
-    if (auto target = get_string(mem, "target"))
-      if (*target == arch)
-        if (auto value = get_string(mem, "value"))
-          tbd.uuid = *value;
-
-  if (auto val = get_string(node, "install-name"))
-    tbd.install_name = *val;
-
-  if (auto val = get_string(node, "current-version"))
-    tbd.current_version = *val;
-
-  for (YamlNode &mem : get_vector(node, "parent-umbrella"))
-    if (contains(get_vector(mem, "targets"), arch))
-      if (auto val = get_string(mem, "umbrella"))
-        tbd.parent_umbrella = *val;
-
-  for (YamlNode &mem : get_vector(node, "reexported-libraries"))
-    if (contains(get_vector(mem, "targets"), arch))
-      for (YamlNode &mem : get_vector(mem, "libraries"))
-        if (auto *lib = std::get_if<std::string_view>(&mem.data))
-          tbd.reexported_libs.push_back(*lib);
-
-  for (std::string_view key : {"exports", "reexports"})
-    for (YamlNode &mem : get_vector(node, key))
-      if (contains(get_vector(mem, "targets"), arch))
-        for (YamlNode &mem : get_vector(mem, "symbols"))
-          if (auto *sym = std::get_if<std::string_view>(&mem.data))
-            tbd.exports.push_back(*sym);
-
-  return tbd;
-}
-
-template <typename E>
-static TextDylib squash(Context<E> &ctx, std::span<TextDylib> tbds) {
-  std::unordered_map<std::string_view, TextDylib> map;
-
-  TextDylib main = std::move(tbds[0]);
-  for (TextDylib &tbd : tbds.subspan(1))
-    map[tbd.install_name] = std::move(tbd);
-
-  std::vector<std::string_view> libs;
-
-  for (std::string_view lib : main.reexported_libs) {
-    auto it = map.find(lib);
-    if (it != map.end())
-      append(main.exports, it->second.exports);
-    else
-      libs.push_back(lib);
-  }
-
-  main.reexported_libs = std::move(libs);
-  return main;
-}
-
-template <typename E>
-static TextDylib parse(Context<E> &ctx, MappedFile<Context<E>> *mf,
-                       std::string_view arch) {
-  std::string_view contents = mf->get_contents();
-  std::variant<std::vector<YamlNode>, YamlError> res = parse_yaml(contents);
-
-  if (YamlError *err = std::get_if<YamlError>(&res)) {
-    i64 lineno = std::count(contents.begin(), contents.begin() + err->pos, '\n');
-    Fatal(ctx) << mf->name << ":" << (lineno + 1)
-               << ": YAML parse error: " << err->msg;
-  }
-
-  std::vector<YamlNode> &nodes = std::get<std::vector<YamlNode>>(res);
-  if (nodes.empty())
-    Fatal(ctx) << mf->name << ": malformed TBD file";
-
-  std::vector<TextDylib> vec;
-  for (YamlNode &node : nodes)
-    if (std::optional<TextDylib> dylib = to_tbd(node, arch))
-      vec.push_back(*dylib);
-  return squash(ctx, vec);
-}
-
-template <>
-TextDylib parse_tbd(Context<ARM64> &ctx, MappedFile<Context<ARM64>> *mf) {
-  return parse(ctx, mf, "arm64-macos");
-}
-
-template <>
-TextDylib parse_tbd(Context<X86_64> &ctx, MappedFile<Context<X86_64>> *mf) {
-  return parse(ctx, mf, "x86_64-macos");
-}
-
-} // namespace mold::macho

macho/yaml.cc

@@ -1,360 +0,0 @@
-#include "mold.h"
-
-#include <optional>
-
-namespace mold::macho {
-
-struct Token {
-  enum { STRING = 1, INDENT, DEDENT, END };
-
-  u8 kind = 0;
-  std::string_view str;
-};
-
-class YamlParser {
-public:
-  YamlParser(std::string_view input) : input(input) {}
-
-  std::variant<std::vector<YamlNode>, YamlError> parse();
-
-private:
-  std::optional<YamlError> tokenize();
-
-  void tokenize_bare_string(std::string_view &str);
-
-  std::optional<YamlError> tokenize_list(std::string_view &str);
-  std::optional<YamlError> tokenize_string(std::string_view &str, char end);
-
-  std::variant<YamlNode, YamlError> parse_element(std::span<Token> &tok);
-  std::variant<YamlNode, YamlError> parse_list(std::span<Token> &tok);
-  std::variant<YamlNode, YamlError> parse_map(std::span<Token> &tok);
-  std::variant<YamlNode, YamlError> parse_flow_element(std::span<Token> &tok);
-  std::variant<YamlNode, YamlError> parse_flow_list(std::span<Token> &tok);
-
-  std::string_view input;
-  std::vector<Token> tokens;
-};
-
-std::optional<YamlError> YamlParser::tokenize() {
-  std::vector<i64> indents = {0};
-
-  auto indent = [&](std::string_view str, i64 depth) {
-    tokens.push_back({Token::INDENT, str});
-    indents.push_back(depth);
-  };
-
-  auto dedent = [&](std::string_view str) {
-    assert(indents.size() > 1);
-    tokens.push_back({Token::DEDENT, str});
-    indents.pop_back();
-  };
-
-  auto skip_line = [](std::string_view &str) {
-    size_t pos = str.find('\n');
-    if (pos == str.npos)
-      str = str.substr(str.size());
-    else
-      str = str.substr(pos + 1);
-  };
-
-  auto tokenize_line = [&](std::string_view &str) -> std::optional<YamlError> {
-    const char *start = str.data();
-
-    if (str.starts_with("---")) {
-      while (indents.size() > 1)
-        dedent(str);
-      tokens.push_back({Token::END, str.substr(0, 3)});
-      skip_line(str);
-      return {};
-    }
-
-    if (str.starts_with("...")) {
-      while (indents.size() > 1)
-        dedent(str);
-      tokens.push_back({Token::END, str.substr(0, 3)});
-      str = str.substr(str.size());
-      return {};
-    }
-
-    size_t pos = str.find_first_not_of(" \t");
-    if (pos == str.npos || str[pos] == '#' || str[pos] == '\n') {
-      skip_line(str);
-      return {};
-    }
-
-    if (indents.back() != pos) {
-      if (indents.back() < pos) {
-        indent(str, pos);
-      } else {
-        while (indents.back() != pos) {
-          if (pos < indents.back())
-            dedent(str);
-          else
-            return YamlError{"bad indentation", start - input.data()};
-        }
-      }
-    }
-
-    str = str.substr(pos);
-
-    while (!str.empty()) {
-      if (str[0] == '\n') {
-        str = str.substr(1);
-        return {};
-      }
-
-      if (str.starts_with("- ")) {
-        tokens.push_back({'-', str.substr(0, 1)});
-
-        size_t pos = str.find_first_not_of(" \t", 1);
-        if (pos == str.npos || str[pos] == '\n') {
-          skip_line(str);
-          return {};
-        }
-
-        str = str.substr(pos);
-        indent(str, str.data() - start);
-        continue;
-      }
-
-      if (str.starts_with('['))
-        return tokenize_list(str);
-
-      if (str.starts_with('\'')) {
-        if (std::optional<YamlError> err = tokenize_string(str, '\''))
-          return err;
-        continue;
-      }
-
-      if (str.starts_with('"')) {
-        if (std::optional<YamlError> err = tokenize_string(str, '"'))
-          return err;
-        continue;
-      }
-
-      if (str.starts_with('#')) {
-        skip_line(str);
-        return {};
-      }
-
-      if (str.starts_with(':')) {
-        tokens.push_back({':', str.substr(0, 1)});
-
-        size_t pos = str.find_first_not_of(" \t", 1);
-        if (pos == str.npos || str[pos] == '\n') {
-          skip_line(str);
-          return {};
-        }
-
-        str = str.substr(pos);
-        continue;
-      }
-
-      tokenize_bare_string(str);
-    }
-    return {};
-  };
-
-  std::string_view str = input;
-  while (!str.empty())
-    if (std::optional<YamlError> err = tokenize_line(str))
-      return err;
-  return {};
-}
-
-std::optional<YamlError> YamlParser::tokenize_list(std::string_view &str) {
-  const char *start = str.data();
-
-  tokens.push_back({'[', str.substr(0, 1)});
-  str = str.substr(1);
-
-  while (!str.empty() && str[0] != ']') {
-    if (size_t pos = str.find_first_not_of(" \t\n"); pos) {
-      str = str.substr(pos);
-      continue;
-    }
-
-    if (str.starts_with('\'')) {
-      if (std::optional<YamlError> err = tokenize_string(str, '\''))
-        return err;
-      continue;
-    }
-
-    if (str.starts_with('"')) {
-      if (std::optional<YamlError> err = tokenize_string(str, '"'))
-        return err;
-      continue;
-    }
-
-    if (str.starts_with(',')) {
-      tokens.push_back({',', str.substr(0, 1)});
-      str = str.substr(1);
-      continue;
-    }
-
-    tokenize_bare_string(str);
-  }
-
-  if (str.empty())
-    return YamlError{"unclosed list", start - input.data()};
-
-  const char *bracket = str.data();
-  tokens.push_back({']', str.substr(0, 1)});
-  str = str.substr(1);
-
-  while (!str.empty() && (str[0] == ' ' || str[0] == '\t'))
-    str = str.substr(1);
-  if (str.empty() || str[0] != '\n')
-    return YamlError{"no newline after ']'", bracket - input.data()};
-  str = str.substr(1);
-  return {};
-}
-
-std::optional<YamlError>
-YamlParser::tokenize_string(std::string_view &str, char end) {
-  const char *start = str.data();
-  size_t pos = str.find(end, 1);
-  if (pos == str.npos)
-    return YamlError{"unterminated string literal", start - input.data()};
-
-  tokens.push_back({Token::STRING, str.substr(1, pos - 1)});
-  str = str.substr(pos + 1);
-  return {};
-}
-
-void
-YamlParser::tokenize_bare_string(std::string_view &str) {
-  size_t pos = str.find_first_not_of(
-    "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_-/.");
-  if (pos == str.npos)
-    pos = str.size();
-  tokens.push_back({Token::STRING, str.substr(0, pos)});
-  str = str.substr(pos);
-}
-
-std::variant<std::vector<YamlNode>, YamlError> YamlParser::parse() {
-  if (std::optional<YamlError> err = tokenize())
-    return *err;
-
-  std::span<Token> tok(tokens);
-  std::vector<YamlNode> vec;
-
-  while (!tok.empty()) {
-    if (tok[0].kind == Token::END) {
-      tok = tok.subspan(1);
-      continue;
-    }
-
-    std::variant<YamlNode, YamlError> elem = parse_element(tok);
-    if (YamlError *err = std::get_if<YamlError>(&elem))
-      return *err;
-    vec.push_back(std::get<YamlNode>(elem));
-
-    if (tok[0].kind != Token::END)
-      return YamlError{"stray token", tok[0].str.data() - input.data()};
-  }
-  return vec;
-}
-
-std::variant<YamlNode, YamlError>
-YamlParser::parse_element(std::span<Token> &tok) {
-  if (tok[0].kind == Token::INDENT) {
-    tok = tok.subspan(1);
-
-    std::variant<YamlNode, YamlError> elem = parse_element(tok);
-    assert(tok[0].kind == Token::DEDENT);
-    tok = tok.subspan(1);
-    return elem;
-  }
-
-  if (tok[0].kind == '-')
-    return parse_list(tok);
-
-  if (tok.size() > 2 && tok[0].kind == Token::STRING && tok[1].kind == ':')
-    return parse_map(tok);
-
-  return parse_flow_element(tok);
-}
-
-std::variant<YamlNode, YamlError>
-YamlParser::parse_list(std::span<Token> &tok) {
-  std::vector<YamlNode> vec;
-
-  while (tok[0].kind != Token::END && tok[0].kind != Token::DEDENT) {
-    if (tok[0].kind != '-')
-      return YamlError{"list element expected", tok[0].str.data() - input.data()};
-    tok = tok.subspan(1);
-
-    std::variant<YamlNode, YamlError> elem = parse_element(tok);
-    if (YamlError *err = std::get_if<YamlError>(&elem))
-      return *err;
-    vec.push_back(std::get<YamlNode>(elem));
-  }
-  return YamlNode{vec};
-}
-
-std::variant<YamlNode, YamlError>
-YamlParser::parse_map(std::span<Token> &tok) {
-  std::map<std::string_view, YamlNode> map;
-
-  while (tok[0].kind != Token::END && tok[0].kind != Token::DEDENT) {
-    if (tok.size() < 2 || tok[0].kind != Token::STRING || tok[1].kind != ':')
-      return YamlError{"map key expected", tok[0].str.data() - input.data()};
-
-    std::string_view key = tok[0].str;
-    tok = tok.subspan(2);
-
-    std::variant<YamlNode, YamlError> elem = parse_element(tok);
-    if (YamlError *err = std::get_if<YamlError>(&elem))
-      return *err;
-    map[key] = std::get<YamlNode>(elem);
-  }
-  return YamlNode{map};
-}
-
-std::variant<YamlNode, YamlError>
-YamlParser::parse_flow_element(std::span<Token> &tok) {
-  if (tok[0].kind == '[') {
-    tok = tok.subspan(1);
-    return parse_flow_list(tok);
-  }
-
-  if (tok[0].kind != Token::STRING)
-    return YamlError{"scalar expected", tok[0].str.data() - input.data()};
-
-  std::string_view val = tok[0].str;
-  tok = tok.subspan(1);
-  return YamlNode{val};
-}
-
-std::variant<YamlNode, YamlError>
-YamlParser::parse_flow_list(std::span<Token> &tok) {
-  std::vector<YamlNode> vec;
-  const char *start = tok[0].str.data();
-
-  while (tok[0].kind != ']' && tok[0].kind != Token::END) {
-    std::variant<YamlNode, YamlError> elem = parse_flow_element(tok);
-    if (YamlError *err = std::get_if<YamlError>(&elem))
-      return *err;
-    vec.push_back(std::get<YamlNode>(elem));
-
-    if (tok[0].kind == ']')
-      break;
-    if (tok[0].kind != ',')
-      return YamlError{"comma expected", tok[0].str.data() - input.data()};
-    tok = tok.subspan(1);
-  }
-
-  if (tok[0].kind == Token::END)
-    return YamlError{"unterminated flow list", start - input.data()};
-
-  tok = tok.subspan(1);
-  return YamlNode{vec};
-}
-
-std::variant<std::vector<YamlNode>, YamlError>
-parse_yaml(std::string_view str) {
-  return YamlParser(str).parse();
-}
-
-} // namespace mold::macho

main.cc

@@ -1,43 +0,0 @@
-#include "elf/mold.h"
-#include "macho/mold.h"
-
-#include <cstring>
-#include <signal.h>
-
-namespace mold {
-
-#ifdef GIT_HASH
-const std::string mold_version =
-  "mold " MOLD_VERSION " (" GIT_HASH "; compatible with GNU ld)";
-#else
-const std::string mold_version =
-  "mold " MOLD_VERSION " (compatible with GNU ld)";
-#endif
-
-void cleanup() {
-  if (output_tmpfile)
-    unlink(output_tmpfile);
-  if (socket_tmpfile)
-    unlink(socket_tmpfile);
-}
-
-static void signal_handler(int) {
-  cleanup();
-  _exit(1);
-}
-
-void install_signal_handler() {
-  signal(SIGINT, signal_handler);
-  signal(SIGTERM, signal_handler);
-}
-
-} // namespace mold
-
-int main(int argc, char **argv) {
-  std::string cmd = mold::filepath(argv[0]).filename();
-
-  if (cmd == "ld64" || cmd == "ld64.mold")
-    return mold::macho::main(argc, argv);
-
-  return mold::elf::main(argc, argv);
-}

mold.h

@@ -1,684 +0,0 @@
-#pragma once
-
-#include "byteorder.h"
-
-#include <atomic>
-#include <cassert>
-#include <cstdio>
-#include <cstring>
-#include <fcntl.h>
-#include <filesystem>
-#include <iostream>
-#include <mutex>
-#include <span>
-#include <sstream>
-#include <string>
-#include <string_view>
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <tbb/concurrent_vector.h>
-#include <tbb/enumerable_thread_specific.h>
-#include <unistd.h>
-#include <vector>
-
-#ifdef NDEBUG
-#  define unreachable() __builtin_unreachable()
-#else
-#  define unreachable() assert(0 && "unreachable")
-#endif
-
-namespace mold {
-
-using namespace std::literals::string_literals;
-using namespace std::literals::string_view_literals;
-
-typedef uint8_t u8;
-typedef uint16_t u16;
-typedef uint32_t u32;
-typedef uint64_t u64;
-
-typedef int8_t i8;
-typedef int16_t i16;
-typedef int32_t i32;
-typedef int64_t i64;
-
-template <typename C> class OutputFile;
-
-inline char *output_tmpfile;
-inline char *socket_tmpfile;
-inline thread_local bool opt_demangle;
-
-extern const std::string mold_version;
-
-std::string_view errno_string();
-void cleanup();
-void install_signal_handler();
-
-//
-// Error output
-//
-
-template <typename C>
-class SyncOut {
-public:
-  SyncOut(C &ctx, std::ostream &out = std::cout) : out(out) {
-    opt_demangle = ctx.arg.demangle;
-  }
-
-  ~SyncOut() {
-    std::scoped_lock lock(mu);
-    out << ss.str() << "\n";
-  }
-
-  template <class T> SyncOut &operator<<(T &&val) {
-    ss << std::forward<T>(val);
-    return *this;
-  }
-
-  static inline std::mutex mu;
-
-private:
-  std::ostream &out;
-  std::stringstream ss;
-};
-
-template <typename C>
-class Fatal {
-public:
-  Fatal(C &ctx) : out(ctx, std::cerr) {
-    out << "mold: ";
-  }
-
-  [[noreturn]] ~Fatal() {
-    out.~SyncOut();
-    cleanup();
-    _exit(1);
-  }
-
-  template <class T> Fatal &operator<<(T &&val) {
-    out << std::forward<T>(val);
-    return *this;
-  }
-
-private:
-  SyncOut<C> out;
-};
-
-template <typename C>
-class Error {
-public:
-  Error(C &ctx) : out(ctx, std::cerr) {
-    if (ctx.arg.color_diagnostics)
-      out << "mold: \033[0;1;31merror:\033[0m ";
-    else
-      out << "mold: error: ";
-
-    ctx.has_error = true;
-  }
-
-  template <class T> Error &operator<<(T &&val) {
-    out << std::forward<T>(val);
-    return *this;
-  }
-
-private:
-  SyncOut<C> out;
-};
-
-template <typename C>
-class Warn {
-public:
-  Warn(C &ctx) : out(ctx, std::cerr) {
-    if (ctx.arg.fatal_warnings) {
-      if (ctx.arg.color_diagnostics)
-        out << "mold: \033[0;1;31merror:\033[0m ";
-      else
-        out << "mold: error: ";
-      ctx.has_error = true;
-    } else {
-      if (ctx.arg.color_diagnostics)
-        out << "mold: \033[0;1;35mwarning:\033[0m ";
-      else
-        out << "mold: warning: ";
-    }
-  }
-
-  template <class T> Warn &operator<<(T &&val) {
-    out << std::forward<T>(val);
-    return *this;
-  }
-
-private:
-  SyncOut<C> out;
-};
-
-//
-// Utility functions
-//
-
-inline u64 align_to(u64 val, u64 align) {
-  if (align == 0)
-    return val;
-  assert(__builtin_popcount(align) == 1);
-  return (val + align - 1) & ~(align - 1);
-}
-
-inline u64 align_down(u64 val, u64 align) {
-  assert(__builtin_popcount(align) == 1);
-  return val & ~(align - 1);
-}
-
-inline u64 next_power_of_two(u64 val) {
-  assert(val >> 63 == 0);
-  if (val == 0 || val == 1)
-    return 1;
-  return (u64)1 << (64 - __builtin_clzl(val - 1));
-}
-
-template <typename T, typename Compare = std::less<T>>
-void update_minimum(std::atomic<T> &atomic, u64 new_val,
-                    Compare cmp = {}) {
-  T old_val = atomic;
-  while (cmp(new_val, old_val) &&
-         !atomic.compare_exchange_weak(old_val, new_val));
-}
-
-template <typename T, typename Compare = std::less<T>>
-void update_maximum(std::atomic<T> &atomic, u64 new_val,
-                    Compare cmp = {}) {
-  T old_val = atomic;
-  while (cmp(old_val, new_val) &&
-         !atomic.compare_exchange_weak(old_val, new_val));
-}
-
-template <typename T, typename U>
-inline void append(std::vector<T> &vec1, std::vector<U> vec2) {
-  vec1.insert(vec1.end(), vec2.begin(), vec2.end());
-}
-
-template <typename T>
-inline std::vector<T> flatten(std::vector<std::vector<T>> &vec) {
-  std::vector<T> ret;
-  for (std::vector<T> &v : vec)
-    append(ret, v);
-  return ret;
-}
-
-template <typename T>
-inline void sort(T &vec) {
-  std::stable_sort(vec.begin(), vec.end());
-}
-
-template <typename T, typename U>
-inline void sort(T &vec, U less) {
-  std::stable_sort(vec.begin(), vec.end(), less);
-}
-
-inline i64 write_string(u8 *buf, std::string_view str) {
-  memcpy(buf, str.data(), str.size());
-  buf[str.size()] = '\0';
-  return str.size() + 1;
-}
-
-template <typename T>
-inline i64 write_vector(u8 *buf, const std::vector<T> &vec) {
-  i64 sz = vec.size() * sizeof(T);
-  memcpy(buf, vec.data(), sz);
-  return sz;
-}
-
-inline void encode_uleb(std::vector<u8> &vec, u64 val) {
-  do {
-    u8 byte = val & 0x7f;
-    val >>= 7;
-    vec.push_back(val ? (byte | 0x80) : byte);
-  } while (val);
-}
-
-inline i64 write_uleb(u8 *buf, u64 val) {
-  i64 i = 0;
-  do {
-    u8 byte = val & 0x7f;
-    val >>= 7;
-    buf[i++] = val ? (byte | 0x80) : byte;
-  } while (val);
-  return i;
-}
-
-inline u64 read_uleb(u8 *&buf) {
-  u64 val = 0;
-  u8 shift = 0;
-  u8 byte;
-  do {
-    byte = *buf++;
-    val |= (byte & 0x7f) << shift;
-    shift += 7;
-  } while (byte & 0x80);
-  return val;
-}
-
-inline i64 uleb_size(u64 val) {
-  i64 i = 0;
-  do {
-    i++;
-    val >>= 7;
-  } while (val);
-  return i;
-}
-
-template <typename C>
-std::string_view save_string(C &ctx, const std::string &str) {
-  u8 *buf = new u8[str.size() + 1];
-  memcpy(buf, str.data(), str.size());
-  buf[str.size()] = '\0';
-  ctx.string_pool.push_back(std::unique_ptr<u8[]>(buf));
-  return {(char *)buf, str.size()};
-}
-
-//
-// Concurrent Map
-//
-
-template <typename T>
-class ConcurrentMap {
-public:
-  ConcurrentMap() {}
-
-  ConcurrentMap(i64 nbuckets) {
-    resize(nbuckets);
-  }
-
-  ~ConcurrentMap() {
-    if (keys) {
-      free((void *)keys);
-      free((void *)sizes);
-      free((void *)values);
-    }
-  }
-
-  void resize(i64 nbuckets) {
-    this->~ConcurrentMap();
-
-    nbuckets = std::max<i64>(MIN_NBUCKETS, next_power_of_two(nbuckets));
-
-    this->nbuckets = nbuckets;
-    keys = (std::atomic<const char *> *)calloc(nbuckets, sizeof(keys[0]));
-    sizes = (u32 *)calloc(nbuckets, sizeof(sizes[0]));
-    values = (T *)calloc(nbuckets, sizeof(values[0]));
-  }
-
-  std::pair<T *, bool> insert(std::string_view key, u64 hash, const T &val) {
-    if (!keys)
-      return {nullptr, false};
-
-    assert(__builtin_popcount(nbuckets) == 1);
-    i64 idx = hash & (nbuckets - 1);
-    i64 retry = 0;
-
-    while (retry < MAX_RETRY) {
-      const char *ptr = keys[idx];
-      if (ptr == locked) {
-#ifdef __x86_64__
-        asm volatile("pause" ::: "memory");
-#endif
-        continue;
-      }
-
-      if (ptr == nullptr) {
-        if (!keys[idx].compare_exchange_weak(ptr, locked))
-          continue;
-        new (values + idx) T(val);
-        sizes[idx] = key.size();
-        keys[idx] = key.data();
-        return {values + idx, true};
-      }
-
-      if (key.size() == sizes[idx] && memcmp(ptr, key.data(), sizes[idx]) == 0)
-        return {values + idx, false};
-
-      u64 mask = nbuckets / NUM_SHARDS - 1;
-      idx = (idx & ~mask) | ((idx + 1) & mask);
-      retry++;
-    }
-
-    assert(false && "ConcurrentMap is full");
-    return {nullptr, false};
-  }
-
-  bool has_key(i64 idx) {
-    return keys[idx];
-  }
-
-  static constexpr i64 MIN_NBUCKETS = 2048;
-  static constexpr i64 NUM_SHARDS = 16;
-  static constexpr i64 MAX_RETRY = 128;
-
-  i64 nbuckets = 0;
-  std::atomic<const char *> *keys = nullptr;
-  u32 *sizes = nullptr;
-  T *values = nullptr;
-
-private:
-  static constexpr const char *locked = "marker";
-};
-
-//
-// Bit vector
-//
-
-class BitVector {
-  class BitRef {
-  public:
-    BitRef(u8 &byte, u8 bitpos) : byte(byte), bitpos(bitpos) {}
-
-    BitRef &operator=(bool val) {
-      if (val)
-        byte |= (1 << bitpos);
-      else
-        byte &= ~(1 << bitpos);
-      return *this;
-    }
-
-    BitRef &operator=(const BitRef &other) {
-      *this = (bool)other;
-      return *this;
-    }
-
-    operator bool() const {
-      return byte & (1 << bitpos);
-    }
-
-  private:
-    u8 &byte;
-    u8 bitpos;
-  };
-
-public:
-  void resize(i64 size) {
-    vec.reset(new u8[(size + 7) / 8]);
-    memset(vec.get(), 0, (size + 7) / 8);
-  }
-
-  BitRef operator[](i64 i) {
-    return BitRef(vec[i / 8], i % 8);
-  }
-
-private:
-  std::unique_ptr<u8[]> vec = nullptr;
-};
-
-//
-// threads.cc
-//
-
-void set_thread_count(i64 n);
-
-//
-// hyperloglog.cc
-//
-
-class HyperLogLog {
-public:
-  HyperLogLog() : buckets(NBUCKETS) {}
-
-  void insert(u32 hash) {
-    update_maximum(buckets[hash & (NBUCKETS - 1)], __builtin_clz(hash) + 1);
-  }
-
-  i64 get_cardinality() const;
-
-  void merge(const HyperLogLog &other) {
-    for (i64 i = 0; i < NBUCKETS; i++)
-      update_maximum(buckets[i], other.buckets[i]);
-  }
-
-private:
-  static constexpr i64 NBUCKETS = 2048;
-  static constexpr double ALPHA = 0.79402;
-
-  std::vector<std::atomic_uint8_t> buckets;
-};
-
-//
-// filepath.cc
-//
-
-template <typename T>
-std::filesystem::path filepath(const T &path) {
-  return {path, std::filesystem::path::format::generic_format};
-}
-
-std::string get_realpath(std::string_view path);
-std::string path_clean(std::string_view path);
-std::filesystem::path to_abs_path(std::filesystem::path path);
-
-//
-// demangle.cc
-//
-
-std::string_view demangle(std::string_view name);
-
-//
-// compress.cc
-//
-
-class ZlibCompressor {
-public:
-  ZlibCompressor(std::string_view input);
-  void write_to(u8 *buf);
-  i64 size() const;
-
-private:
-  std::vector<std::vector<u8>> shards;
-  u64 checksum = 0;
-};
-
-class GzipCompressor {
-public:
-  GzipCompressor(std::string_view input);
-  void write_to(u8 *buf);
-  i64 size() const;
-
-private:
-  std::vector<std::vector<u8>> shards;
-  u32 checksum = 0;
-  u32 uncompressed_size = 0;
-};
-
-//
-// perf.cc
-//
-
-// Counter is used to collect statistics numbers.
-class Counter {
-public:
-  Counter(std::string_view name, i64 value = 0) : name(name), values(value) {
-    static std::mutex mu;
-    std::scoped_lock lock(mu);
-    instances.push_back(this);
-  }
-
-  Counter &operator++(int) {
-    if (enabled)
-      values.local()++;
-    return *this;
-  }
-
-  Counter &operator+=(int delta) {
-    if (enabled)
-      values.local() += delta;
-    return *this;
-  }
-
-  static void print();
-
-  static inline bool enabled = false;
-
-private:
-  i64 get_value();
-
-  std::string_view name;
-  tbb::enumerable_thread_specific<i64> values;
-
-  static inline std::vector<Counter *> instances;
-};
-
-// Timer and TimeRecord records elapsed time (wall clock time)
-// used by each pass of the linker.
-struct TimerRecord {
-  TimerRecord(std::string name, TimerRecord *parent = nullptr);
-  void stop();
-
-  std::string name;
-  TimerRecord *parent;
-  tbb::concurrent_vector<TimerRecord *> children;
-  i64 start;
-  i64 end;
-  i64 user;
-  i64 sys;
-  bool stopped = false;
-};
-
-void
-print_timer_records(tbb::concurrent_vector<std::unique_ptr<TimerRecord>> &);
-
-template <typename C>
-class Timer {
-public:
-  Timer(C &ctx, std::string name, Timer *parent = nullptr) {
-    record = new TimerRecord(name, parent ? parent->record : nullptr);
-    ctx.timer_records.push_back(std::unique_ptr<TimerRecord>(record));
-  }
-
-  Timer(const Timer &) = delete;
-
-  ~Timer() {
-    record->stop();
-  }
-
-  void stop() {
-    record->stop();
-  }
-
-private:
-  TimerRecord *record;
-};
-
-//
-// tar.cc
-//
-
-// TarFile is a class to create a tar file.
-//
-// If you pass `--reproduce=repro.tar` to mold, mold collects all
-// input files and put them into `repro.tar`, so that it is easy to
-// run the same command with the same command line arguments.
-class TarFile {
-public:
-  TarFile(std::string basedir) : basedir(basedir) {}
-  void append(std::string path, std::string_view data);
-  void write_to(u8 *buf);
-  i64 size() const { return size_; }
-
-private:
-  static constexpr i64 BLOCK_SIZE = 512;
-
-  std::string encode_path(std::string path);
-
-  std::string basedir;
-  std::vector<std::pair<std::string, std::string_view>> contents;
-  i64 size_ = BLOCK_SIZE * 2;
-};
-
-//
-// Memory-mapped file
-//
-
-// MappedFile represents an mmap'ed input file.
-// mold uses mmap-IO only.
-template <typename C>
-class MappedFile {
-public:
-  static MappedFile *open(C &ctx, std::string path);
-  static MappedFile *must_open(C &ctx, std::string path);
-
-  ~MappedFile();
-
-  MappedFile *slice(C &ctx, std::string name, u64 start, u64 size);
-
-  std::string_view get_contents() {
-    return std::string_view((char *)data, size);
-  }
-
-  std::string name;
-  u8 *data = nullptr;
-  i64 size = 0;
-  i64 mtime = 0;
-  bool given_fullpath = true;
-  MappedFile *parent = nullptr;
-};
-
-template <typename C>
-MappedFile<C> *MappedFile<C>::open(C &ctx, std::string path) {
-  MappedFile *mf = new MappedFile;
-  mf->name = path;
-
-  ctx.mf_pool.push_back(std::unique_ptr<MappedFile>(mf));
-
-  if (path.starts_with('/') && !ctx.arg.chroot.empty())
-    path = ctx.arg.chroot + "/" + path_clean(path);
-
-  i64 fd = ::open(path.c_str(), O_RDONLY);
-  if (fd == -1)
-    return nullptr;
-
-  struct stat st;
-  if (fstat(fd, &st) == -1)
-    Fatal(ctx) << path << ": fstat failed: " << errno_string();
-
-  mf->size = st.st_size;
-
-#ifdef __APPLE__
-  mf->mtime = (u64)st.st_mtimespec.tv_sec * 1000000000 + st.st_mtimespec.tv_nsec;
-#else
-  mf->mtime = (u64)st.st_mtim.tv_sec * 1000000000 + st.st_mtim.tv_nsec;
-#endif
-
-  if (st.st_size > 0) {
-    mf->data = (u8 *)mmap(nullptr, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
-    if (mf->data == MAP_FAILED)
-      Fatal(ctx) << path << ": mmap failed: " << errno_string();
-  }
-
-  close(fd);
-  return mf;
-}
-
-template <typename C>
-MappedFile<C> *MappedFile<C>::must_open(C &ctx, std::string path) {
-  if (MappedFile *mf = MappedFile::open(ctx, path))
-    return mf;
-  Fatal(ctx) << "cannot open " << path;
-}
-
-template <typename C>
-MappedFile<C> *
-MappedFile<C>::slice(C &ctx, std::string name, u64 start, u64 size) {
-  MappedFile *mf = new MappedFile<C>;
-  mf->name = name;
-  mf->data = data + start;
-  mf->size = size;
-  mf->parent = this;
-
-  ctx.mf_pool.push_back(std::unique_ptr<MappedFile>(mf));
-  return mf;
-}
-
-template <typename C>
-MappedFile<C>::~MappedFile() {
-  if (size && !parent)
-    munmap(data, size);
-}
-
-} // namespace mold

src/arch-arc.cc

@@ -0,0 +1,285 @@
+// r31 is blink. r30 is I believe a scratch register.
+//
+// Programs might use one of several branch, jump, and link instructions
+// to control execution flow through direct and indirect function calls
+// and branching. For function calling, the conditional branch-and-link
+// instruction has a maximum branch range of ±1 MiB, and the target
+// address is 32-bit-aligned. The unconditional branch-and-link format has
+// a maximum branch range of ±16 MiB.
+//
+// https://github.com/foss-for-synopsys-dwc-arc-processors/arc-ABI-manual/blob/master/ARCv2_ABI.pdf
+
+#include "mold.h"
+
+namespace mold {
+
+using E = ARC;
+
+// An integer-like class for the mixed-endian encoding
+class M32 {
+public:
+  operator u32() const {
+    return (buf[1] << 24) | (buf[0] << 16) | (buf[3] << 8) | buf[2];
+  }
+
+  M32 &operator=(u32 x) {
+    buf[0] = x >> 16;
+    buf[1] = x >> 24;
+    buf[2] = x;
+    buf[3] = x >> 8;
+    return *this;
+  }
+
+  M32 &operator&=(u32 x) { return *this = *this & x; }
+  M32 &operator|=(u32 x) { return *this = *this | x; }
+  u8 buf[4];
+};
+
+static void write_disp7u(u8 *loc, u32 val) {
+  *(ul16 *)loc |= (bits(val, 6, 3) << 4) | bits(val, 2, 0);
+}
+
+static void write_disp9(u8 *loc, u32 val) {
+  *(M32 *)loc |= bits(val, 8, 0);
+}
+
+static void write_disp9ls(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 7, 0) << 16) | (bit(val, 8) << 15);
+}
+
+static void write_disp9s(u8 *loc, u32 val) {
+  *(ul16 *)loc |= bits(val, 8, 0);
+}
+
+static void write_disp10u(u8 *loc, u32 val) {
+  *(ul16 *)loc |= bits(val, 9, 0);
+}
+
+static void write_disp13s(u8 *loc, u32 val) {
+  *(ul16 *)loc |= bits(val, 12, 0);
+}
+
+static void write_disp21h(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 10, 1) << 17) | (bits(val, 20, 11) << 6);
+}
+
+static void write_disp21w(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 10, 2) << 18) | (bits(val, 20, 11) << 6);
+}
+
+static void write_disp25h(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 10, 1) << 17) | (bits(val, 20, 11) << 6) |
+                  bits(val, 24, 21);
+}
+
+static void write_disp25w(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 10, 2) << 18) | (bits(val, 20, 11) << 6) |
+                  bits(val, 24, 21);
+}
+
+static void write_disps9(u8 *loc, u32 val) {
+  *(ul16 *)loc |= bits(val, 10, 2);
+}
+
+static void write_disps12(u8 *loc, u32 val) {
+  *(M32 *)loc |= (bits(val, 5, 0) << 6) | bits(val, 11, 6);
+}
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  static const ul16 insn[] = {
+    0x2730, 0x7f8b, 0, 0, // ld r11, [pcl,0]
+    0x2730, 0x7f8a, 0, 0, // ld r10, [pcl,0]
+    0x2020, 0x0280,       // j       [r10]
+    0, 0,                 // (address of GOTPLT)
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(M32 *)(buf + 4) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 4;
+  *(M32 *)(buf + 12) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr;
+  *(M32 *)(buf + 20) = ctx.gotplt->shdr.sh_addr;
+}
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static const ul16 insn[] = {
+    0x2730, 0x7f8c, 0, 0, // ld  r12, [pcl,0]
+    0x2021, 0x0300,       // j.d [r12]
+    0x240a, 0x1fc0,       // mov r12,pcl
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(M32 *)(buf + 4) = sym.get_gotplt_addr(ctx) - sym.get_plt_addr(ctx);
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static const ul16 insn[] = {
+    0x2730, 0x7f8c, 0, 0, // ld  r12, [pcl,0]
+    0x2021, 0x0300,       // j.d [r12]
+    0x240a, 0x1fc0,       // mov r12,pcl
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(M32 *)(buf + 4) = sym.get_got_pltgot_addr(ctx) - sym.get_plt_addr(ctx);
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_ARC_32:
+    *(ul32 *)loc = val;
+    break;
+  case R_ARC_32_PCREL:
+    *(ul32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = rel.r_addend;
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    switch (rel.r_type) {
+    case R_ARC_32:
+      *(ul32 *)loc = S + A;
+      break;
+    case R_ARC_32_ME:
+      *(M32 *)loc = S + A;
+      break;
+    case R_ARC_S25H_PCREL:
+    case R_ARC_S25H_PCREL_PLT:
+      write_disp25h(loc, align_to(S + A - P, 2));
+      break;
+    case R_ARC_S25W_PCREL:
+    case R_ARC_S25W_PCREL_PLT:
+      write_disp25w(loc, align_to(S + A - P, 4));
+      break;
+    case R_ARC_PC32:
+      *(M32 *)loc = S + A - P;
+      break;
+    case R_ARC_32_PCREL:
+      *(ul32 *)loc = S + A - P;
+      break;
+    case R_ARC_GOTPC32:
+      *(M32 *)loc = GOT + G + A - P;
+      break;
+    case R_ARC_TLS_LE_32:
+      *(M32 *)loc = S + A - ctx.tp_addr;
+      break;
+    case R_ARC_TLS_IE_GOT:
+      *(M32 *)loc = sym.get_gottp_addr(ctx) + A - P;
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : (i64)rel.r_addend;
+
+    switch (rel.r_type) {
+    case R_ARC_32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ul32 *)loc = *val;
+      else
+        *(ul32 *)loc = S + A;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+
+    if (sym.is_ifunc())
+      Error(ctx) << sym << ": GNU ifunc symbol is not supported on m68k";
+
+    switch (rel.r_type) {
+    case R_ARC_32:
+      break;
+    case R_ARC_32_ME:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_ARC_S25H_PCREL:
+    case R_ARC_S25W_PCREL:
+    case R_ARC_32_PCREL:
+    case R_ARC_PC32:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_ARC_S25H_PCREL_PLT:
+    case R_ARC_S25W_PCREL_PLT:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_ARC_GOTPC32:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_ARC_TLS_LE_32:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_ARC_TLS_IE_GOT:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+u64 get_eflags(Context<E> &ctx) {
+  return 0x406;
+}
+
+} // namespace mold

src/arch-arm32.cc

@@ -0,0 +1,849 @@
+// ARM32 is a bit special from the linker's viewpoint because ARM
+// processors support two different instruction encodings: Thumb and
+// ARM (in a narrower sense). Thumb instructions are either 16 bits or
+// 32 bits, while ARM instructions are all 32 bits. Feature-wise,
+// Thumb is a subset of ARM, so not all ARM instructions are
+// representable in Thumb.
+//
+// ARM processors originally supported only ARM instructions. Thumb
+// instructions were later added to increase code density.
+//
+// ARM processors runs in either ARM mode or Thumb mode. The mode can
+// be switched using BX (branch and mode exchange)-family instructions.
+// We need to use that instructions to, for example, call a function
+// encoded in Thumb from a function encoded in ARM. Sometimes, the
+// linker even has to emit interworking thunk code to switch mode.
+//
+// ARM instructions are aligned to 4 byte boundaries. Thumb are to 2
+// byte boundaries. So the least significant bit of a function address
+// is always 0.
+//
+// To distinguish Thumb functions from ARM fucntions, the LSB of a
+// function address is repurposed as a boolean flag. If the LSB is 0,
+// the function referred to by the address is encoded in ARM;
+// otherwise, Thumb.
+//
+// For example, if a symbol `foo` is of type STT_FUNC and has value
+// 0x2001, `foo` is a function using Thumb instructions whose address
+// is 0x2000 (not 0x2001, as Thumb instructions are always 2-byte
+// aligned). Likewise, if a function pointer has value 0x2001, it
+// refers a Thumb function at 0x2000.
+//
+// https://github.com/ARM-software/abi-aa/blob/main/aaelf32/aaelf32.rst
+
+#include "mold.h"
+
+#include <tbb/parallel_for.h>
+
+namespace mold {
+
+using E = ARM32;
+
+template <>
+i64 get_addend(u8 *loc, const ElfRel<E> &rel) {
+  ul32 *arm = (ul32 *)loc;
+  ul16 *thm = (ul16 *)loc;
+
+  switch (rel.r_type) {
+  case R_ARM_ABS32:
+  case R_ARM_REL32:
+  case R_ARM_BASE_PREL:
+  case R_ARM_GOTOFF32:
+  case R_ARM_GOT_PREL:
+  case R_ARM_GOT_BREL:
+  case R_ARM_TLS_GD32:
+  case R_ARM_TLS_LDM32:
+  case R_ARM_TLS_LDO32:
+  case R_ARM_TLS_IE32:
+  case R_ARM_TLS_LE32:
+  case R_ARM_TLS_GOTDESC:
+  case R_ARM_TARGET1:
+  case R_ARM_TARGET2:
+    return (il32)*arm;
+  case R_ARM_THM_JUMP8:
+    return sign_extend(thm[0], 8) << 1;
+  case R_ARM_THM_JUMP11:
+    return sign_extend(thm[0], 11) << 1;
+  case R_ARM_THM_JUMP19: {
+    // https://developer.arm.com/documentation/ddi0597/2024-12/Base-Instructions/B--Branch-
+    u32 S = bit(thm[0], 10);
+    u32 J2 = bit(thm[1], 11);
+    u32 J1 = bit(thm[1], 13);
+    u32 imm6 = bits(thm[0], 5, 0);
+    u32 imm11 = bits(thm[1], 10, 0);
+    u32 val = (S << 20) | (J2 << 19) | (J1 << 18) | (imm6 << 12) | (imm11 << 1);
+    return sign_extend(val, 21);
+  }
+  case R_ARM_THM_CALL:
+  case R_ARM_THM_JUMP24:
+  case R_ARM_THM_TLS_CALL: {
+    // https://developer.arm.com/documentation/ddi0597/2024-12/Base-Instructions/BL--BLX--immediate---Branch-with-Link-and-optional-Exchange--immediate--
+    u32 S = bit(thm[0], 10);
+    u32 J1 = bit(thm[1], 13);
+    u32 J2 = bit(thm[1], 11);
+    u32 I1 = !(J1 ^ S);
+    u32 I2 = !(J2 ^ S);
+    u32 imm10 = bits(thm[0], 9, 0);
+    u32 imm11 = bits(thm[1], 10, 0);
+    u32 val = (S << 24) | (I1 << 23) | (I2 << 22) | (imm10 << 12) | (imm11 << 1);
+    return sign_extend(val, 25);
+  }
+  case R_ARM_CALL:
+  case R_ARM_JUMP24:
+  case R_ARM_PLT32:
+  case R_ARM_TLS_CALL:
+    return sign_extend(*arm, 24) << 2;
+  case R_ARM_MOVW_PREL_NC:
+  case R_ARM_MOVW_ABS_NC:
+  case R_ARM_MOVT_PREL:
+  case R_ARM_MOVT_ABS: {
+    // https://developer.arm.com/documentation/ddi0597/2024-12/Base-Instructions/MOV--MOVS--immediate---Move--immediate--
+    u32 imm4 = bits(*arm, 19, 16);
+    u32 imm12 = bits(*arm, 11, 0);
+    u32 val = (imm4 << 12) | imm12;
+    return sign_extend(val, 16);
+  }
+  case R_ARM_PREL31:
+    return sign_extend(*arm, 31);
+  case R_ARM_THM_MOVW_PREL_NC:
+  case R_ARM_THM_MOVW_ABS_NC:
+  case R_ARM_THM_MOVT_PREL:
+  case R_ARM_THM_MOVT_ABS: {
+    // https://developer.arm.com/documentation/ddi0597/2024-12/Base-Instructions/MOVT--Move-Top-
+    u32 imm4 = bits(thm[0], 3, 0);
+    u32 i = bit(thm[0], 10);
+    u32 imm3 = bits(thm[1], 14, 12);
+    u32 imm8 = bits(thm[1], 7, 0);
+    u32 val = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
+    return sign_extend(val, 16);
+  }
+  default:
+    return 0;
+  }
+}
+
+static void write_arm_mov(u8 *loc, u32 val) {
+  u32 imm12 = bits(val, 11, 0);
+  u32 imm4 = bits(val, 15, 12);
+  *(ul32 *)loc = (*(ul32 *)loc & 0xfff0'f000) | (imm4 << 16) | imm12;
+}
+
+static void write_thm_b21(u8 *loc, u32 val) {
+  u32 S = bit(val, 20);
+  u32 J2 = bit(val, 19);
+  u32 J1 = bit(val, 18);
+  u32 imm6 = bits(val, 17, 12);
+  u32 imm11 = bits(val, 11, 1);
+
+  ul16 *buf = (ul16 *)loc;
+  buf[0] = (buf[0] & 0b1111'1011'1100'0000) | (S << 10) | imm6;
+  buf[1] = (buf[1] & 0b1101'0000'0000'0000) | (J1 << 13) | (J2 << 11) | imm11;
+}
+
+static void write_thm_b25(u8 *loc, u32 val) {
+  u32 S = bit(val, 24);
+  u32 I1 = bit(val, 23);
+  u32 I2 = bit(val, 22);
+  u32 J1 = !I1 ^ S;
+  u32 J2 = !I2 ^ S;
+  u32 imm10 = bits(val, 21, 12);
+  u32 imm11 = bits(val, 11, 1);
+
+  ul16 *buf = (ul16 *)loc;
+  buf[0] = (buf[0] & 0b1111'1000'0000'0000) | (S << 10) | imm10;
+  buf[1] = (buf[1] & 0b1101'0000'0000'0000) | (J1 << 13) | (J2 << 11) | imm11;
+}
+
+static void write_thm_mov(u8 *loc, u32 val) {
+  u32 imm4 = bits(val, 15, 12);
+  u32 i = bit(val, 11);
+  u32 imm3 = bits(val, 10, 8);
+  u32 imm8 = bits(val, 7, 0);
+
+  ul16 *buf = (ul16 *)loc;
+  buf[0] = (buf[0] & 0b1111'1011'1111'0000) | (i << 10) | imm4;
+  buf[1] = (buf[1] & 0b1000'1111'0000'0000) | (imm3 << 12) | imm8;
+}
+
+template <>
+void write_addend(u8 *loc, i64 val, const ElfRel<E> &rel) {
+  switch (rel.r_type) {
+  case R_ARM_NONE:
+    break;
+  case R_ARM_ABS32:
+  case R_ARM_REL32:
+  case R_ARM_BASE_PREL:
+  case R_ARM_GOTOFF32:
+  case R_ARM_GOT_PREL:
+  case R_ARM_GOT_BREL:
+  case R_ARM_TLS_GD32:
+  case R_ARM_TLS_LDM32:
+  case R_ARM_TLS_LDO32:
+  case R_ARM_TLS_IE32:
+  case R_ARM_TLS_LE32:
+  case R_ARM_TLS_GOTDESC:
+  case R_ARM_TARGET1:
+  case R_ARM_TARGET2:
+    *(ul32 *)loc = val;
+    break;
+  case R_ARM_THM_JUMP8:
+    *(ul16 *)loc = (*(ul16 *)loc & 0xff00) | bits(val, 8, 1);
+    break;
+  case R_ARM_THM_JUMP11:
+    *(ul16 *)loc = (*(ul16 *)loc & 0xf800) | bits(val, 11, 1);
+    break;
+  case R_ARM_THM_CALL:
+  case R_ARM_THM_JUMP24:
+  case R_ARM_THM_TLS_CALL:
+    write_thm_b25(loc, val);
+    break;
+  case R_ARM_CALL:
+  case R_ARM_JUMP24:
+  case R_ARM_PLT32:
+    *(ul32 *)loc = (*(ul32 *)loc & 0xff00'0000) | bits(val, 25, 2);
+    break;
+  case R_ARM_MOVW_PREL_NC:
+  case R_ARM_MOVW_ABS_NC:
+  case R_ARM_MOVT_PREL:
+  case R_ARM_MOVT_ABS:
+    write_arm_mov(loc, val);
+    break;
+  case R_ARM_PREL31:
+    *(ul32 *)loc = (*(ul32 *)loc & 0x8000'0000) | (val & 0x7fff'ffff);
+    break;
+  case R_ARM_THM_MOVW_PREL_NC:
+  case R_ARM_THM_MOVW_ABS_NC:
+  case R_ARM_THM_MOVT_PREL:
+  case R_ARM_THM_MOVT_ABS:
+    write_thm_mov(loc, val);
+    break;
+  default:
+    unreachable();
+  }
+}
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  constexpr ul32 insn[] = {
+    0xe52d'e004, //    push {lr}
+    0xe59f'e004, //    ldr lr, 2f
+    0xe08f'e00e, // 1: add lr, pc, lr
+    0xe5be'f008, //    ldr pc, [lr, #8]!
+    0x0000'0000, // 2: .word .got.plt - 1b - 8
+    0x0000'0000, //    (padding)
+    0x0000'0000, //    (padding)
+    0x0000'0000, //    (padding)
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(ul32 *)(buf + 16) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr - 16;
+}
+
+static constexpr ul32 plt_entry[] = {
+  0xe59f'c004, // 1: ldr ip, 2f
+  0xe08c'c00f, //    add ip, ip, pc
+  0xe59c'f000, //    ldr pc, [ip]
+  0x0000'0000, // 2: .word sym@GOT - 1b
+};
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  memcpy(buf, plt_entry, sizeof(plt_entry));
+  *(ul32 *)(buf + 12) = sym.get_gotplt_addr(ctx) - sym.get_plt_addr(ctx) - 12;
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  memcpy(buf, plt_entry, sizeof(plt_entry));
+  *(ul32 *)(buf + 12) = sym.get_got_pltgot_addr(ctx) - sym.get_plt_addr(ctx) - 12;
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_ARM_ABS32:
+    *(ul32 *)loc = val;
+    break;
+  case R_ARM_REL32:
+    *(ul32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+static Thunk<E> &get_reachable_thunk(OutputSection<E> &osec, u64 addr) {
+  auto it = std::upper_bound(osec.thunks.begin(), osec.thunks.end(), addr,
+                             [](u64 addr, std::unique_ptr<Thunk<E>> &thunk) {
+    return addr < thunk->get_addr();
+  });
+  return **it;
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || rel.r_type == R_ARM_V4BX)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = get_addend(*this, rel);
+    u64 P = get_addr() + rel.r_offset;
+    u64 T = S & 1;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, i, val, lo, hi);
+    };
+
+    auto get_thumb_thunk_addr = [&] { return sym.get_thunk_addr(ctx, P); };
+    auto get_arm_thunk_addr   = [&] { return sym.get_thunk_addr(ctx, P) + 4; };
+
+    auto get_tlsdesc_trampoline_addr = [&] {
+      return get_reachable_thunk(*output_section, P).get_addr();
+    };
+
+    switch (rel.r_type) {
+    case R_ARM_ABS32:
+    case R_ARM_TARGET1:
+      break;
+    case R_ARM_REL32:
+      *(ul32 *)loc = S + A - P;
+      break;
+    case R_ARM_THM_CALL: {
+      if (sym.is_remaining_undef_weak()) {
+        // On ARM, calling an weak undefined symbol jumps to the
+        // next instruction.
+        *(ul32 *)loc = 0x8000'f3af; // NOP.W
+        break;
+      }
+
+      // THM_CALL relocation refers to either BL or BLX instruction.
+      // They are different in only one bit. We need to use BL if
+      // the jump target is Thumb. Otherwise, use BLX.
+      i64 val1 = S + A - P;
+      i64 val2 = align_to(S + A - P, 4);
+
+      if (T && is_int(val1, 25)) {
+        *(ul16 *)(loc + 2) |= 0x1000;  // BL
+        write_thm_b25(loc, val1);
+      } else if (!T && is_int(val2, 25)) {
+        *(ul16 *)(loc + 2) &= ~0x1000; // BLX
+        write_thm_b25(loc, val2);
+      } else {
+        *(ul16 *)(loc + 2) |= 0x1000;  // BL
+        write_thm_b25(loc, get_thumb_thunk_addr() + A - P);
+      }
+      break;
+    }
+    case R_ARM_BASE_PREL:
+      *(ul32 *)loc = GOT + A - P;
+      break;
+    case R_ARM_GOTOFF32:
+      *(ul32 *)loc = ((S + A) | T) - GOT;
+      break;
+    case R_ARM_GOT_PREL:
+    case R_ARM_TARGET2:
+      *(ul32 *)loc = GOT + G + A - P;
+      break;
+    case R_ARM_GOT_BREL:
+      *(ul32 *)loc = G + A;
+      break;
+    case R_ARM_CALL: {
+      if (sym.is_remaining_undef_weak()) {
+        *(ul32 *)loc = 0xe320'f000; // NOP
+        break;
+      }
+
+      // Just like THM_CALL, ARM_CALL relocation refers to either BL or
+      // BLX instruction. We may need to rewrite BL → BLX or BLX → BL.
+      bool is_bl = ((*(ul32 *)loc & 0xff00'0000) == 0xeb00'0000);
+      bool is_blx = ((*(ul32 *)loc & 0xfe00'0000) == 0xfa00'0000);
+      if (!is_bl && !is_blx)
+        Fatal(ctx) << *this << ": R_ARM_CALL refers to neither BL nor BLX";
+
+      i64 val = S + A - P;
+      if (is_int(val, 26)) {
+        if (T) {
+          *(ul32 *)loc = 0xfa00'0000; // BLX
+          *(ul32 *)loc |= (bit(val, 1) << 24) | bits(val, 25, 2);
+        } else {
+          *(ul32 *)loc = 0xeb00'0000; // BL
+          *(ul32 *)loc |= bits(val, 25, 2);
+        }
+      } else {
+        *(ul32 *)loc = 0xeb00'0000; // BL
+        *(ul32 *)loc |= bits(get_arm_thunk_addr() + A - P, 25, 2);
+      }
+      break;
+    }
+    case R_ARM_JUMP24: {
+      if (sym.is_remaining_undef_weak()) {
+        *(ul32 *)loc = 0xe320'f000; // NOP
+        break;
+      }
+
+      // These relocs refers to a B (unconditional branch) instruction.
+      // Unlike BL or BLX, we can't rewrite B to BX in place when the
+      // processor mode switch is required because BX doesn't takes an
+      // immediate; it takes only a register. So if mode switch is
+      // required, we jump to a linker-synthesized thunk which does the
+      // job with a longer code sequence.
+      i64 val = S + A - P;
+      if (T || !is_int(val, 26))
+        val = get_arm_thunk_addr() + A - P;
+      *(ul32 *)loc = (*(ul32 *)loc & 0xff00'0000) | bits(val, 25, 2);
+      break;
+    }
+    case R_ARM_PLT32:
+      if (sym.is_remaining_undef_weak()) {
+        *(ul32 *)loc = 0xe320'f000; // NOP
+      } else {
+        u64 val = (T ? get_arm_thunk_addr() : S) + A - P;
+        *(ul32 *)loc = (*(ul32 *)loc & 0xff00'0000) | bits(val, 25, 2);
+      }
+      break;
+    case R_ARM_THM_JUMP8:
+      check(S + A - P, -(1 << 8), 1 << 8);
+      *(ul16 *)loc &= 0xff00;
+      *(ul16 *)loc |= bits(S + A - P, 8, 1);
+      break;
+    case R_ARM_THM_JUMP11:
+      check(S + A - P, -(1 << 11), 1 << 11);
+      *(ul16 *)loc &= 0xf800;
+      *(ul16 *)loc |= bits(S + A - P, 11, 1);
+      break;
+    case R_ARM_THM_JUMP19:
+      check(S + A - P, -(1 << 20), 1 << 20);
+      write_thm_b21(loc, S + A - P);
+      break;
+    case R_ARM_THM_JUMP24: {
+      if (sym.is_remaining_undef_weak()) {
+        *(ul32 *)loc = 0x8000'f3af; // NOP
+        break;
+      }
+
+      // Just like R_ARM_JUMP24, we need to jump to a thunk if we need to
+      // switch processor mode.
+      i64 val = S + A - P;
+      if (!T || !is_int(val, 25))
+        val = get_thumb_thunk_addr() + A - P;
+      write_thm_b25(loc, val);
+      break;
+    }
+    case R_ARM_MOVW_PREL_NC:
+      write_arm_mov(loc, ((S + A) | T) - P);
+      break;
+    case R_ARM_MOVW_ABS_NC:
+      write_arm_mov(loc, (S + A) | T);
+      break;
+    case R_ARM_THM_MOVW_PREL_NC:
+      write_thm_mov(loc, ((S + A) | T) - P);
+      break;
+    case R_ARM_PREL31:
+      check(S + A - P, -(1LL << 30), 1LL << 30);
+      *(ul32 *)loc &= 0x8000'0000;
+      *(ul32 *)loc |= (S + A - P) & 0x7fff'ffff;
+      break;
+    case R_ARM_THM_MOVW_ABS_NC:
+      write_thm_mov(loc, (S + A) | T);
+      break;
+    case R_ARM_MOVT_PREL:
+      write_arm_mov(loc, (S + A - P) >> 16);
+      break;
+    case R_ARM_THM_MOVT_PREL:
+      write_thm_mov(loc, (S + A - P) >> 16);
+      break;
+    case R_ARM_MOVT_ABS:
+      write_arm_mov(loc, (S + A) >> 16);
+      break;
+    case R_ARM_THM_MOVT_ABS:
+      write_thm_mov(loc, (S + A) >> 16);
+      break;
+    case R_ARM_TLS_GD32:
+      *(ul32 *)loc = sym.get_tlsgd_addr(ctx) + A - P;
+      break;
+    case R_ARM_TLS_LDM32:
+      *(ul32 *)loc = ctx.got->get_tlsld_addr(ctx) + A - P;
+      break;
+    case R_ARM_TLS_LDO32:
+      *(ul32 *)loc = S + A - ctx.dtp_addr;
+      break;
+    case R_ARM_TLS_IE32:
+      *(ul32 *)loc = sym.get_gottp_addr(ctx) + A - P;
+      break;
+    case R_ARM_TLS_LE32:
+      *(ul32 *)loc = S + A - ctx.tp_addr;
+      break;
+    case R_ARM_TLS_GOTDESC:
+      // ARM32 TLSDESC uses the following code sequence to materialize
+      // a TP-relative address in r0.
+      //
+      //       ldr     r0, .L2
+      //  .L1: bl      foo
+      //           R_ARM_TLS_CALL
+      //  .L2: .word   foo + . - .L1
+      //           R_ARM_TLS_GOTDESC
+      //
+      // We may relax the instructions to the following if its TP-relative
+      // address is known at link-time
+      //
+      //       ldr     r0, .L2
+      //  .L1: nop
+      //       ...
+      //  .L2: .word   foo(tpoff)
+      //
+      // or to the following if the TP-relative address is known at
+      // process startup time.
+      //
+      //       ldr     r0, .L2
+      //  .L1: ldr r0, [pc, r0]
+      //       ...
+      //  .L2: .word   foo(gottpoff) + . - .L1
+      if (sym.has_tlsdesc(ctx)) {
+        // A is odd if the corresponding TLS_CALL is Thumb.
+        *(ul32 *)loc = sym.get_tlsdesc_addr(ctx) - P + A - ((A & 1) ? 6 : 4);
+      } else if (sym.has_gottp(ctx)) {
+        *(ul32 *)loc = sym.get_gottp_addr(ctx) - P + A - ((A & 1) ? 5 : 8);
+      } else {
+        *(ul32 *)loc = S - ctx.tp_addr;
+      }
+      break;
+    case R_ARM_TLS_CALL:
+      if (sym.has_tlsdesc(ctx)) {
+        *(ul32 *)loc = 0xeb00'0000; // bl 0
+        *(ul32 *)loc |= bits(get_tlsdesc_trampoline_addr() - P - 8, 25, 2);
+      } else if (sym.has_gottp(ctx)) {
+        *(ul32 *)loc = 0xe79f'0000; // ldr r0, [pc, r0]
+      } else {
+        *(ul32 *)loc = 0xe320'f000; // nop
+      }
+      break;
+    case R_ARM_THM_TLS_CALL:
+      if (sym.has_tlsdesc(ctx)) {
+        u64 val = align_to(get_tlsdesc_trampoline_addr() - P - 4, 4);
+        write_thm_b25(loc, val);
+        *(ul16 *)(loc + 2) &= ~0x1000; // rewrite BL with BLX
+      } else if (sym.has_gottp(ctx)) {
+        // Since `ldr r0, [pc, r0]` is not representable in Thumb,
+        // we use two instructions instead.
+        *(ul16 *)loc = 0x4478;         // add r0, pc
+        *(ul16 *)(loc + 2) = 0x6800;   // ldr r0, [r0]
+      } else {
+        *(ul32 *)loc = 0x8000'f3af;    // nop.w
+      }
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : get_addend(*this, rel);
+
+    switch (rel.r_type) {
+    case R_ARM_ABS32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ul32 *)loc = *val;
+      else
+        *(ul32 *)loc = S + A;
+      break;
+    case R_ARM_TLS_LDO32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ul32 *)loc = *val;
+      else
+        *(ul32 *)loc = S + A - ctx.dtp_addr;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+      break;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  // Scan relocations
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+
+    if (sym.is_ifunc())
+      sym.flags |= NEEDS_GOT | NEEDS_PLT;
+
+    switch (rel.r_type) {
+    case R_ARM_MOVW_ABS_NC:
+    case R_ARM_THM_MOVW_ABS_NC:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_ARM_THM_CALL:
+    case R_ARM_CALL:
+    case R_ARM_JUMP24:
+    case R_ARM_PLT32:
+    case R_ARM_THM_JUMP24:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_ARM_GOT_PREL:
+    case R_ARM_GOT_BREL:
+    case R_ARM_TARGET2:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_ARM_MOVT_PREL:
+    case R_ARM_THM_MOVT_PREL:
+    case R_ARM_PREL31:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_ARM_TLS_GD32:
+      sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_ARM_TLS_LDM32:
+      ctx.needs_tlsld = true;
+      break;
+    case R_ARM_TLS_IE32:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_ARM_TLS_CALL:
+    case R_ARM_THM_TLS_CALL:
+      scan_tlsdesc(ctx, sym);
+      break;
+    case R_ARM_TLS_LE32:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_ARM_ABS32:
+    case R_ARM_TARGET1:
+    case R_ARM_MOVT_ABS:
+    case R_ARM_THM_MOVT_ABS:
+    case R_ARM_REL32:
+    case R_ARM_BASE_PREL:
+    case R_ARM_GOTOFF32:
+    case R_ARM_THM_JUMP8:
+    case R_ARM_THM_JUMP11:
+    case R_ARM_THM_JUMP19:
+    case R_ARM_MOVW_PREL_NC:
+    case R_ARM_THM_MOVW_PREL_NC:
+    case R_ARM_TLS_LDO32:
+    case R_ARM_V4BX:
+    case R_ARM_TLS_GOTDESC:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+void Thunk<E>::copy_buf(Context<E> &ctx) {
+  // TLS trampoline code. ARM32's TLSDESC is designed so that this
+  // common piece of code is factored out from object files to reduce
+  // output size. Since no one provide, the linker has to synthesize it.
+  constexpr ul32 hdr[] = {
+    0xe08e'0000, // add r0, lr, r0
+    0xe590'1004, // ldr r1, [r0, #4]
+    0xe12f'ff11, // bx  r1
+    0xe320'f000, // nop
+  };
+
+  // This is a range extension and mode switch thunk.
+  // It has two entry points: +0 for Thumb and +4 for ARM.
+  constexpr u8 entry[] = {
+    // .thumb
+    0x78, 0x47,             //    bx   pc  # jumps to 1f
+    0xc0, 0x46,             //    nop
+    // .arm
+    0x00, 0xc0, 0x9f, 0xe5, // 1: ldr  ip, 3f
+    0x0f, 0xf0, 0x8c, 0xe0, // 2: add  pc, ip, pc
+    0x00, 0x00, 0x00, 0x00, // 3: .word sym - 2b
+  };
+
+  static_assert(E::thunk_hdr_size == sizeof(hdr));
+  static_assert(E::thunk_size == sizeof(entry));
+
+  u8 *buf = ctx.buf + output_section.shdr.sh_offset + offset;
+  memcpy(buf, hdr, sizeof(hdr));
+  buf += sizeof(hdr);
+
+  u64 P = output_section.shdr.sh_addr + offset + sizeof(hdr);
+
+  for (Symbol<E> *sym : symbols) {
+    memcpy(buf, entry, sizeof(entry));
+    *(ul32 *)(buf + 12) = sym->get_addr(ctx) - P - 16;
+
+    buf += sizeof(entry);
+    P += sizeof(entry);
+  }
+}
+
+template <>
+u64 get_eflags(Context<E> &ctx) {
+  return EF_ARM_EABI_VER5;
+}
+
+void create_arm_exidx_section(Context<E> &ctx) {
+  for (i64 i = 0; i < ctx.chunks.size(); i++) {
+    OutputSection<E> *osec = ctx.chunks[i]->to_osec();
+
+    if (osec && osec->shdr.sh_type == SHT_ARM_EXIDX) {
+      auto *sec = new Arm32ExidxSection(*osec);
+      ctx.extra.exidx = sec;
+      ctx.chunks[i] = sec;
+      ctx.chunk_pool.emplace_back(sec);
+
+      for (InputSection<E> *isec : osec->members)
+        isec->is_alive = false;
+      break;
+    }
+  }
+}
+
+void Arm32ExidxSection::compute_section_size(Context<E> &ctx) {
+  output_section.compute_section_size(ctx);
+  this->shdr.sh_size = output_section.shdr.sh_size + 8; // +8 for sentinel
+}
+
+void Arm32ExidxSection::update_shdr(Context<E> &ctx) {
+  // .ARM.exidx's sh_link should be set to the .text section index.
+  // Runtime doesn't care about it, but the binutils's strip command does.
+  if (Chunk<E> *chunk = find_chunk(ctx, ".text"))
+    this->shdr.sh_link = chunk->shndx;
+}
+
+void Arm32ExidxSection::remove_duplicate_entries(Context<E> &ctx) {
+  this->shdr.sh_size = get_contents(ctx).size();
+}
+
+void Arm32ExidxSection::copy_buf(Context<E> &ctx) {
+  std::vector<u8> contents = get_contents(ctx);
+  assert(this->shdr.sh_size == contents.size());
+  write_vector(ctx.buf + this->shdr.sh_offset, contents);
+}
+
+// Returns the end of the text segment
+static u64 get_text_end(Context<E> &ctx) {
+  u64 ret = 0;
+  for (Chunk<E> *chunk : ctx.chunks)
+    if (chunk->shdr.sh_flags & SHF_EXECINSTR)
+      ret = std::max<u64>(ret, chunk->shdr.sh_addr + chunk->shdr.sh_size);
+  return ret;
+}
+
+// ARM executables use an .ARM.exidx section to look up an exception
+// handling record for the current instruction pointer. The table needs
+// to be sorted by their addresses.
+//
+// Other target uses .eh_frame_hdr instead for the same purpose.
+// I don't know why only ARM uses the different mechanism, but it's
+// likely that it's due to some historical reason.
+//
+// This function returns contents of .ARM.exidx.
+std::vector<u8> Arm32ExidxSection::get_contents(Context<E> &ctx) {
+  // .ARM.exidx records consists of a signed 31-bit relative address
+  // and a 32-bit value. The relative address indicates the start
+  // address of a function that the record covers. The value is one of
+  // the followings:
+  //
+  // 1. CANTUNWIND indicating that there's no unwinding info for the function,
+  // 2. a compact unwinding record encoded into a 32-bit value, or
+  // 3. a 31-bit relative address which points to a larger record in
+  //    the .ARM.extab section.
+  //
+  // CANTUNWIND is value 1. The most significant bit is set in (2) but
+  // not in (3). So we can distinguished them just by looking at a value.
+  const u32 CANTUNWIND = 1;
+
+  struct Entry {
+    ul32 addr;
+    ul32 val;
+  };
+
+  // We reserve one extra slot for the sentinel
+  i64 num_entries = output_section.shdr.sh_size / sizeof(Entry) + 1;
+  std::vector<u8> buf(num_entries * sizeof(Entry));
+  Entry *ent = (Entry *)buf.data();
+
+  // Write section contents to the buffer
+  output_section.shdr.sh_addr = this->shdr.sh_addr;
+  output_section.write_to(ctx, buf.data());
+
+  // Fill in sentinel fields
+  u64 sentinel_addr = this->shdr.sh_addr + sizeof(Entry) * (num_entries - 1);
+  ent[num_entries - 1].addr = get_text_end(ctx) - sentinel_addr;
+  ent[num_entries - 1].val = CANTUNWIND;
+
+  // Entry's addresses are relative to themselves. In order to sort
+  // records by address, we first translate them so that the addresses
+  // are relative to the beginning of the section.
+  auto is_relative = [](u32 val) {
+    return val != CANTUNWIND && !(val & 0x8000'0000);
+  };
+
+  tbb::parallel_for((i64)0, num_entries, [&](i64 i) {
+    i64 offset = sizeof(Entry) * i;
+    ent[i].addr = sign_extend(ent[i].addr, 31) + offset;
+    if (is_relative(ent[i].val))
+      ent[i].val = 0x7fff'ffff & (ent[i].val + offset);
+  });
+
+  std::sort(ent, ent + num_entries, [](const Entry &a, const Entry &b) {
+    return a.addr < b.addr;
+  });
+
+  // Remove duplicate adjacent entries. That is, if two adjacent functions
+  // have the same compact unwind info or are both CANTUNWIND, we can
+  // merge them into a single address range.
+  auto it = std::unique(ent, ent + num_entries,
+                        [](const Entry &a, const Entry &b) {
+    return a.val == b.val;
+  });
+
+  num_entries = it - ent;
+  buf.resize(num_entries * sizeof(Entry));
+
+  // Make addresses relative to themselves.
+  tbb::parallel_for((i64)0, num_entries, [&](i64 i) {
+    i64 offset = sizeof(Entry) * i;
+    ent[i].addr = 0x7fff'ffff & (ent[i].addr - offset);
+    if (is_relative(ent[i].val))
+      ent[i].val = 0x7fff'ffff & (ent[i].val - offset);
+  });
+
+  return buf;
+}
+
+} // namespace mold

src/arch-arm64.cc

@@ -0,0 +1,617 @@
+// This file contains ARM64-specific code. Being new, the ARM64's ELF
+// psABI doesn't have anything peculiar. ARM64 is a clean RISC
+// instruction set that supports PC-relative load/store instructions.
+//
+// Unlike ARM32, instructions length doesn't vary. All ARM64
+// instructions are 4 bytes long.
+//
+// Branch instructions used for function call can jump within ±128 MiB.
+// We need to create range extension thunks to support binaries whose
+// .text is larger than that.
+//
+// Unlike most other targets, the TLSDESC access model is used by default
+// for -fPIC to access thread-local variables instead of the less
+// efficient GD model. You can still enable GD but it needs the
+// -mtls-dialect=trad flag. Since GD is used rarely, we don't need to
+// implement GD → LE relaxation.
+//
+// https://github.com/ARM-software/abi-aa/blob/main/aaelf64/aaelf64.rst
+
+#if MOLD_ARM64LE || MOLD_ARM64BE
+
+#include "mold.h"
+
+namespace mold {
+
+using E = MOLD_TARGET;
+
+static void write_adrp(u8 *buf, u64 val) {
+  *(ul32 *)buf |= (bits(val, 13, 12) << 29) | (bits(val, 32, 14) << 5);
+}
+
+static void write_adr(u8 *buf, u64 val) {
+  *(ul32 *)buf |= (bits(val, 1, 0) << 29) | (bits(val, 20, 2) << 5);
+}
+
+static void write_movn_movz(u8 *buf, i64 val) {
+  *(ul32 *)buf &= 0b0000'0000'0110'0000'0000'0000'0001'1111;
+
+  if (val >= 0)
+    *(ul32 *)buf |= 0xd280'0000 | (bits(val, 15, 0) << 5);  // rewrite to movz
+  else
+    *(ul32 *)buf |= 0x9280'0000 | (bits(~val, 15, 0) << 5); // rewrite to movn
+}
+
+static u64 page(u64 val) {
+  return val & 0xffff'ffff'ffff'f000;
+}
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  constexpr ul32 insn[] = {
+    0xa9bf'7bf0, // stp  x16, x30, [sp,#-16]!
+    0x9000'0010, // adrp x16, .got.plt[2]
+    0xf940'0211, // ldr  x17, [x16, .got.plt[2]]
+    0x9100'0210, // add  x16, x16, .got.plt[2]
+    0xd61f'0220, // br   x17
+    0xd420'7d00, // brk
+    0xd420'7d00, // brk
+    0xd420'7d00, // brk
+ };
+
+  u64 gotplt = ctx.gotplt->shdr.sh_addr + 16;
+  u64 plt = ctx.plt->shdr.sh_addr;
+
+  memcpy(buf, insn, sizeof(insn));
+  write_adrp(buf + 4, page(gotplt) - page(plt + 4));
+  *(ul32 *)(buf + 8) |= bits(gotplt, 11, 3) << 10;
+  *(ul32 *)(buf + 12) |= (gotplt & 0xfff) << 10;
+}
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  constexpr ul32 insn[] = {
+    0x9000'0010, // adrp x16, .got.plt[n]
+    0xf940'0211, // ldr  x17, [x16, .got.plt[n]]
+    0x9100'0210, // add  x16, x16, .got.plt[n]
+    0xd61f'0220, // br   x17
+  };
+
+  u64 gotplt = sym.get_gotplt_addr(ctx);
+  u64 plt = sym.get_plt_addr(ctx);
+
+  memcpy(buf, insn, sizeof(insn));
+  write_adrp(buf, page(gotplt) - page(plt));
+  *(ul32 *)(buf + 4) |= bits(gotplt, 11, 3) << 10;
+  *(ul32 *)(buf + 8) |= (gotplt & 0xfff) << 10;
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  constexpr ul32 insn[] = {
+    0x9000'0010, // adrp x16, GOT[n]
+    0xf940'0211, // ldr  x17, [x16, GOT[n]]
+    0xd61f'0220, // br   x17
+    0xd420'7d00, // brk
+  };
+
+  u64 got = sym.get_got_pltgot_addr(ctx);
+  u64 plt = sym.get_plt_addr(ctx);
+
+  memcpy(buf, insn, sizeof(insn));
+  write_adrp(buf, page(got) - page(plt));
+  *(ul32 *)(buf + 4) |= bits(got, 11, 3) << 10;
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_AARCH64_ABS64:
+    *(U64<E> *)loc = val;
+    break;
+  case R_AARCH64_PREL32:
+    *(U32<E> *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  case R_AARCH64_PREL64:
+    *(U64<E> *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+static bool is_adrp(u8 *loc) {
+  // https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/ADRP--Form-PC-relative-address-to-4KB-page-
+  u32 insn = *(ul32 *)loc;
+  return (bits(insn, 31, 24) & 0b1001'1111) == 0b1001'0000;
+}
+
+static bool is_ldr(u8 *loc) {
+  // https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/LDR--immediate---Load-Register--immediate--
+  u32 insn = *(ul32 *)loc;
+  return (bits(insn, 31, 20) & 0b1111'1111'1100) == 0b1111'1001'0100;
+}
+
+static bool is_add(u8 *loc) {
+  // https://developer.arm.com/documentation/ddi0596/2021-12/Base-Instructions/ADD--immediate---Add--immediate--
+  u32 insn = *(ul32 *)loc;
+  return (bits(insn, 31, 20) & 0b1111'1111'1100) == 0b1001'0001'0000;
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = rel.r_addend;
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, i, val, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_AARCH64_ABS64:
+      break;
+    case R_AARCH64_LDST8_ABS_LO12_NC:
+    case R_AARCH64_ADD_ABS_LO12_NC:
+      *(ul32 *)loc |= bits(S + A, 11, 0) << 10;
+      break;
+    case R_AARCH64_LDST16_ABS_LO12_NC:
+      *(ul32 *)loc |= bits(S + A, 11, 1) << 10;
+      break;
+    case R_AARCH64_LDST32_ABS_LO12_NC:
+      *(ul32 *)loc |= bits(S + A, 11, 2) << 10;
+      break;
+    case R_AARCH64_LDST64_ABS_LO12_NC:
+      *(ul32 *)loc |= bits(S + A, 11, 3) << 10;
+      break;
+    case R_AARCH64_LDST128_ABS_LO12_NC:
+      *(ul32 *)loc |= bits(S + A, 11, 4) << 10;
+      break;
+    case R_AARCH64_MOVW_UABS_G0:
+      check(S + A, 0, 1 << 16);
+      *(ul32 *)loc |= bits(S + A, 15, 0) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G0_NC:
+      *(ul32 *)loc |= bits(S + A, 15, 0) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G1:
+      check(S + A, 0, 1LL << 32);
+      *(ul32 *)loc |= bits(S + A, 31, 16) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G1_NC:
+      *(ul32 *)loc |= bits(S + A, 31, 16) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G2:
+      check(S + A, 0, 1LL << 48);
+      *(ul32 *)loc |= bits(S + A, 47, 32) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G2_NC:
+      *(ul32 *)loc |= bits(S + A, 47, 32) << 5;
+      break;
+    case R_AARCH64_MOVW_UABS_G3:
+      *(ul32 *)loc |= bits(S + A, 63, 48) << 5;
+      break;
+    case R_AARCH64_ADR_GOT_PAGE:
+      if (sym.has_got(ctx)) {
+        i64 val = page(G + GOT + A) - page(P);
+        check(val, -(1LL << 32), 1LL << 32);
+        write_adrp(loc, val);
+      } else {
+        // Relax GOT-loading ADRP+LDR to an immediate ADRP+ADD
+        i64 val = page(S + A) - page(P);
+        check(val, -(1LL << 32), 1LL << 32);
+        write_adrp(loc, val);
+
+        u32 reg = bits(*(ul32 *)loc, 4, 0);
+        *(ul32 *)(loc + 4) = 0x9100'0000 | (reg << 5) | reg; // ADD
+        *(ul32 *)(loc + 4) |= bits(S + A, 11, 0) << 10;
+        i++;
+      }
+      break;
+    case R_AARCH64_ADR_PREL_PG_HI21:
+    case R_AARCH64_ADR_PREL_PG_HI21_NC: {
+      // The ARM64 psABI defines that an `ADRP x0, foo` and `ADD x0, x0,
+      // :lo12: foo` instruction pair to materialize a PC-relative address
+      // in a register can be relaxed to `NOP` followed by `ADR x0, foo`
+      // if foo is in PC ± 1 MiB.
+      if (ctx.arg.relax && sym.is_pcrel_linktime_const(ctx) &&
+          i + 1 < rels.size()) {
+        i64 val = S + A - P - 4;
+        const ElfRel<E> &rel2 = rels[i + 1];
+        if (is_int(val, 21) &&
+            rel2.r_type == R_AARCH64_ADD_ABS_LO12_NC &&
+            rel2.r_sym == rel.r_sym &&
+            rel2.r_offset == rel.r_offset + 4 &&
+            rel2.r_addend == rel.r_addend &&
+            is_adrp(loc) &&
+            is_add(loc + 4)) {
+          u32 reg1 = bits(*(ul32 *)loc, 4, 0);
+          u32 reg2 = bits(*(ul32 *)(loc + 4), 4, 0);
+          if (reg1 == reg2) {
+            *(ul32 *)loc = 0xd503'201f;              // nop
+            *(ul32 *)(loc + 4) = 0x1000'0000 | reg1; // adr
+            write_adr(loc + 4, val);
+            i++;
+            break;
+          }
+        }
+      }
+
+      i64 val = page(S + A) - page(P);
+      if (rel.r_type == R_AARCH64_ADR_PREL_PG_HI21)
+        check(val, -(1LL << 32), 1LL << 32);
+      write_adrp(loc, val);
+      break;
+    }
+    case R_AARCH64_ADR_PREL_LO21:
+      check(S + A - P, -(1LL << 20), 1LL << 20);
+      write_adr(loc, S + A - P);
+      break;
+    case R_AARCH64_CALL26:
+    case R_AARCH64_JUMP26: {
+      if (sym.is_remaining_undef_weak()) {
+        // On ARM, calling an weak undefined symbol jumps to the
+        // next instruction.
+        *(ul32 *)loc = 0xd503'201f; // nop
+        break;
+      }
+
+      i64 val = S + A - P;
+      if (!is_int(val, 28))
+        val = sym.get_thunk_addr(ctx, P) + A - P;
+      *(ul32 *)loc |= bits(val, 27, 2);
+      break;
+    }
+    case R_AARCH64_PLT32:
+      check(S + A - P, -(1LL << 31), 1LL << 31);
+      *(U32<E> *)loc = S + A - P;
+      break;
+    case R_AARCH64_CONDBR19:
+    case R_AARCH64_LD_PREL_LO19:
+      check(S + A - P, -(1LL << 20), 1LL << 20);
+      *(ul32 *)loc |= bits(S + A - P, 20, 2) << 5;
+      break;
+    case R_AARCH64_PREL16:
+      check(S + A - P, -(1LL << 15), 1LL << 16);
+      *(U16<E> *)loc = S + A - P;
+      break;
+    case R_AARCH64_PREL32:
+      check(S + A - P, -(1LL << 31), 1LL << 32);
+      *(U32<E> *)loc = S + A - P;
+      break;
+    case R_AARCH64_PREL64:
+      *(U64<E> *)loc = S + A - P;
+      break;
+    case R_AARCH64_LD64_GOT_LO12_NC:
+      *(ul32 *)loc |= bits(G + GOT + A, 11, 3) << 10;
+      break;
+    case R_AARCH64_LD64_GOTPAGE_LO15: {
+      i64 val = G + GOT + A - page(GOT);
+      check(val, 0, 1 << 15);
+      *(ul32 *)loc |= bits(val, 14, 3) << 10;
+      break;
+    }
+    case R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21: {
+      i64 val = page(sym.get_gottp_addr(ctx) + A) - page(P);
+      check(val, -(1LL << 32), 1LL << 32);
+      write_adrp(loc, val);
+      break;
+    }
+    case R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC:
+      *(ul32 *)loc |= bits(sym.get_gottp_addr(ctx) + A, 11, 3) << 10;
+      break;
+    case R_AARCH64_TLSLE_MOVW_TPREL_G0: {
+      i64 val = S + A - ctx.tp_addr;
+      check(val, -(1 << 15), 1 << 15);
+      write_movn_movz(loc, val);
+      break;
+    }
+    case R_AARCH64_TLSLE_MOVW_TPREL_G0_NC:
+      *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 15, 0) << 5;
+      break;
+    case R_AARCH64_TLSLE_MOVW_TPREL_G1: {
+      i64 val = S + A - ctx.tp_addr;
+      check(val, -(1LL << 31), 1LL << 31);
+      write_movn_movz(loc, val >> 16);
+      break;
+    }
+    case R_AARCH64_TLSLE_MOVW_TPREL_G1_NC:
+      *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 31, 16) << 5;
+      break;
+    case R_AARCH64_TLSLE_MOVW_TPREL_G2: {
+      i64 val = S + A - ctx.tp_addr;
+      check(val, -(1LL << 47), 1LL << 47);
+      write_movn_movz(loc, val >> 32);
+      break;
+    }
+    case R_AARCH64_TLSLE_ADD_TPREL_HI12: {
+      i64 val = S + A - ctx.tp_addr;
+      check(val, 0, 1LL << 24);
+      *(ul32 *)loc |= bits(val, 23, 12) << 10;
+      break;
+    }
+    case R_AARCH64_TLSLE_ADD_TPREL_LO12:
+      check(S + A - ctx.tp_addr, 0, 1 << 12);
+      *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 11, 0) << 10;
+      break;
+    case R_AARCH64_TLSLE_ADD_TPREL_LO12_NC:
+      *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 11, 0) << 10;
+      break;
+    case R_AARCH64_TLSGD_ADR_PAGE21: {
+      i64 val = page(sym.get_tlsgd_addr(ctx) + A) - page(P);
+      check(val, -(1LL << 32), 1LL << 32);
+      write_adrp(loc, val);
+      break;
+    }
+    case R_AARCH64_TLSGD_ADD_LO12_NC:
+      *(ul32 *)loc |= bits(sym.get_tlsgd_addr(ctx) + A, 11, 0) << 10;
+      break;
+    case R_AARCH64_TLSDESC_ADR_PAGE21:
+      // ARM64 TLSDESC uses the following code sequence to materialize
+      // a TP-relative address in x0.
+      //
+      //   adrp    x0, 0
+      //       R_AARCH64_TLSDESC_ADR_PAGE21 foo
+      //   ldr     x1, [x0]
+      //       R_AARCH64_TLSDESC_LD64_LO12  foo
+      //   add     x0, x0, #0
+      //       R_AARCH64_TLSDESC_ADD_LO12   foo
+      //   blr     x1
+      //       R_AARCH64_TLSDESC_CALL       foo
+      //
+      // We may relax the instructions to the following if its TP-relative
+      // address is known at link-time
+      //
+      //   nop
+      //   nop
+      //   movz    x0, :tls_offset_hi:foo, lsl #16
+      //   movk    x0, :tls_offset_lo:foo
+      //
+      // or to the following if the TP-relative address is known at
+      // process startup time.
+      //
+      //   nop
+      //   nop
+      //   adrp    x0, :gottprel:foo
+      //   ldr     x0, [x0, :gottprel_lo12:foo]
+      if (sym.has_tlsdesc(ctx)) {
+        i64 val = page(sym.get_tlsdesc_addr(ctx) + A) - page(P);
+        check(val, -(1LL << 32), 1LL << 32);
+        write_adrp(loc, val);
+      } else {
+        *(ul32 *)loc = 0xd503'201f; // nop
+      }
+      break;
+    case R_AARCH64_TLSDESC_LD64_LO12:
+      if (sym.has_tlsdesc(ctx))
+        *(ul32 *)loc |= bits(sym.get_tlsdesc_addr(ctx) + A, 11, 3) << 10;
+      else
+        *(ul32 *)loc = 0xd503'201f; // nop
+      break;
+    case R_AARCH64_TLSDESC_ADD_LO12:
+      if (sym.has_tlsdesc(ctx)) {
+        *(ul32 *)loc |= bits(sym.get_tlsdesc_addr(ctx) + A, 11, 0) << 10;
+      } else if (sym.has_gottp(ctx)) {
+        *(ul32 *)loc = 0x9000'0000; // adrp x0, 0
+        write_adrp(loc, page(sym.get_gottp_addr(ctx) + A) - page(P));
+      } else {
+        *(ul32 *)loc = 0xd2a0'0000; // movz x0, 0, lsl #16
+        *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 32, 16) << 5;
+      }
+      break;
+    case R_AARCH64_TLSDESC_CALL:
+      if (sym.has_tlsdesc(ctx)) {
+        // Do nothing
+      } else if (sym.has_gottp(ctx)) {
+        *(ul32 *)loc = 0xf940'0000; // ldr x0, [x0, 0]
+        *(ul32 *)loc |= bits(sym.get_gottp_addr(ctx) + A, 11, 3) << 10;
+      } else {
+        *(ul32 *)loc = 0xf280'0000; // movk x0, 0
+        *(ul32 *)loc |= bits(S + A - ctx.tp_addr, 15, 0) << 5;
+      }
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : (i64)rel.r_addend;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, val, i, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_AARCH64_ABS64:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(U64<E> *)loc = *val;
+      else
+        *(U64<E> *)loc = S + A;
+      break;
+    case R_AARCH64_ABS32:
+      check(S + A, 0, 1LL << 32);
+      *(U32<E> *)loc = S + A;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+      break;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  // Scan relocations
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = (u8 *)(contents.data() + rel.r_offset);
+
+    if (sym.is_ifunc())
+      sym.flags |= NEEDS_GOT | NEEDS_PLT;
+
+    switch (rel.r_type) {
+    case R_AARCH64_MOVW_UABS_G3:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_AARCH64_ADR_GOT_PAGE:
+      // An ADR_GOT_PAGE and GOT_LO12_NC relocation pair is used to load a
+      // symbol's address from GOT. If the GOT value is a link-time
+      // constant, we may be able to rewrite the ADRP+LDR instruction pair
+      // with an ADRP+ADD, eliminating a GOT memory load.
+      if (ctx.arg.relax && sym.is_pcrel_linktime_const(ctx) &&
+          i + 1 < rels.size()) {
+        // ADRP+LDR must be consecutive and use the same register to relax.
+        const ElfRel<E> &rel2 = rels[i + 1];
+        if (rel2.r_type == R_AARCH64_LD64_GOT_LO12_NC &&
+            rel2.r_offset == rel.r_offset + 4 &&
+            rel2.r_sym == rel.r_sym &&
+            rel.r_addend == 0 &&
+            rel2.r_addend == 0 &&
+            is_adrp(loc) &&
+            is_ldr(loc + 4)) {
+          u32 rd = bits(*(ul32 *)loc, 4, 0);
+          u32 rn = bits(*(ul32 *)(loc + 4), 9, 5);
+          u32 rt = bits(*(ul32 *)(loc + 4), 4, 0);
+          if (rd == rn && rn == rt) {
+            i++;
+            break;
+          }
+        }
+      }
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_AARCH64_LD64_GOT_LO12_NC:
+    case R_AARCH64_LD64_GOTPAGE_LO15:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_AARCH64_CALL26:
+    case R_AARCH64_JUMP26:
+    case R_AARCH64_PLT32:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_AARCH64_TLSIE_ADR_GOTTPREL_PAGE21:
+    case R_AARCH64_TLSIE_LD64_GOTTPREL_LO12_NC:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_AARCH64_ADR_PREL_PG_HI21:
+    case R_AARCH64_ADR_PREL_PG_HI21_NC:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_AARCH64_TLSGD_ADR_PAGE21:
+      sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_AARCH64_TLSDESC_CALL:
+      scan_tlsdesc(ctx, sym);
+      break;
+    case R_AARCH64_TLSLE_MOVW_TPREL_G2:
+    case R_AARCH64_TLSLE_ADD_TPREL_LO12:
+    case R_AARCH64_TLSLE_ADD_TPREL_LO12_NC:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_AARCH64_ABS64:
+    case R_AARCH64_ADD_ABS_LO12_NC:
+    case R_AARCH64_ADR_PREL_LO21:
+    case R_AARCH64_CONDBR19:
+    case R_AARCH64_LD_PREL_LO19:
+    case R_AARCH64_LDST16_ABS_LO12_NC:
+    case R_AARCH64_LDST32_ABS_LO12_NC:
+    case R_AARCH64_LDST64_ABS_LO12_NC:
+    case R_AARCH64_LDST128_ABS_LO12_NC:
+    case R_AARCH64_LDST8_ABS_LO12_NC:
+    case R_AARCH64_MOVW_UABS_G0:
+    case R_AARCH64_MOVW_UABS_G0_NC:
+    case R_AARCH64_MOVW_UABS_G1:
+    case R_AARCH64_MOVW_UABS_G1_NC:
+    case R_AARCH64_MOVW_UABS_G2:
+    case R_AARCH64_MOVW_UABS_G2_NC:
+    case R_AARCH64_PREL16:
+    case R_AARCH64_PREL32:
+    case R_AARCH64_PREL64:
+    case R_AARCH64_TLSGD_ADD_LO12_NC:
+    case R_AARCH64_TLSLE_MOVW_TPREL_G0:
+    case R_AARCH64_TLSLE_MOVW_TPREL_G0_NC:
+    case R_AARCH64_TLSLE_MOVW_TPREL_G1:
+    case R_AARCH64_TLSLE_MOVW_TPREL_G1_NC:
+    case R_AARCH64_TLSLE_ADD_TPREL_HI12:
+    case R_AARCH64_TLSDESC_ADR_PAGE21:
+    case R_AARCH64_TLSDESC_LD64_LO12:
+    case R_AARCH64_TLSDESC_ADD_LO12:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+void Thunk<E>::copy_buf(Context<E> &ctx) {
+  constexpr ul32 insn[] = {
+    0x9000'0010, // adrp x16, 0   # R_AARCH64_ADR_PREL_PG_HI21
+    0x9100'0210, // add  x16, x16 # R_AARCH64_ADD_ABS_LO12_NC
+    0xd61f'0200, // br   x16
+    0xd420'7d00, // brk
+  };
+
+  static_assert(E::thunk_size == sizeof(insn));
+
+  u8 *buf = ctx.buf + output_section.shdr.sh_offset + offset;
+  u64 P = output_section.shdr.sh_addr + offset;
+
+  for (Symbol<E> *sym : symbols) {
+    u64 S = sym->get_addr(ctx);
+    memcpy(buf, insn, E::thunk_size);
+    write_adrp(buf, page(S) - page(P));
+    *(ul32 *)(buf + 4) |= bits(S, 11, 0) << 10;
+
+    buf += E::thunk_size;
+    P += E::thunk_size;
+  }
+}
+
+} // namespace mold
+
+#endif

src/arch-i386.cc

@@ -0,0 +1,598 @@
+// i386 is similar to x86-64 but lacks PC-relative memory access
+// instructions. So it's not straightforward to support position-
+// independent code (PIC) on that target.
+//
+// If an object file is compiled with -fPIC, a function that needs to load
+// a value from memory first obtains its own address with the following
+// code
+//
+//   call __x86.get_pc_thunk.bx
+//
+// where __x86.get_pc_thunk.bx is defined as
+//
+//   __x86.get_pc_thunk.bx:
+//     mov (%esp), %ebx  # move the return address to %ebx
+//     ret
+//
+// . With the function's own address (or, more precisely, the address
+// immediately after the call instruction), the function can compute an
+// absolute address of a variable with its address + link-time constant.
+//
+// Executing call-mov-ret isn't very cheap, and allocating one register to
+// store PC isn't cheap too, especially given that i386 has only 8
+// general-purpose registers. But that's the cost of PIC on i386. You need
+// to pay it when creating a .so and a position-independent executable.
+//
+// When a position-independent function calls another function, it sets
+// %ebx to the address of .got. Position-independent PLT entries use that
+// register to load values from .got.plt/.got.
+//
+// If we are creating a position-dependent executable (PDE), we can't
+// assume that %ebx is set to .got. For PDE, we need to create position-
+// dependent PLT entries which don't use %ebx.
+//
+// https://github.com/rui314/psabi/blob/main/i386.pdf
+
+#include "mold.h"
+
+namespace mold {
+
+using E = I386;
+
+template <>
+i64 get_addend(u8 *loc, const ElfRel<E> &rel) {
+  switch (rel.r_type) {
+  case R_386_8:
+  case R_386_PC8:
+    return *loc;
+  case R_386_16:
+  case R_386_PC16:
+    return *(ul16 *)loc;
+  case R_386_32:
+  case R_386_PC32:
+  case R_386_GOT32:
+  case R_386_GOT32X:
+  case R_386_PLT32:
+  case R_386_GOTOFF:
+  case R_386_GOTPC:
+  case R_386_TLS_LDM:
+  case R_386_TLS_GOTIE:
+  case R_386_TLS_LE:
+  case R_386_TLS_IE:
+  case R_386_TLS_GD:
+  case R_386_TLS_LDO_32:
+  case R_386_SIZE32:
+  case R_386_TLS_GOTDESC:
+    return *(ul32 *)loc;
+  default:
+    return 0;
+  }
+}
+
+template <>
+void write_addend(u8 *loc, i64 val, const ElfRel<E> &rel) {
+  switch (rel.r_type) {
+  case R_386_NONE:
+    break;
+  case R_386_8:
+  case R_386_PC8:
+    *loc = val;
+    break;
+  case R_386_16:
+  case R_386_PC16:
+    *(ul16 *)loc = val;
+    break;
+  case R_386_32:
+  case R_386_PC32:
+  case R_386_GOT32:
+  case R_386_GOT32X:
+  case R_386_PLT32:
+  case R_386_GOTOFF:
+  case R_386_GOTPC:
+  case R_386_TLS_LDM:
+  case R_386_TLS_GOTIE:
+  case R_386_TLS_LE:
+  case R_386_TLS_IE:
+  case R_386_TLS_GD:
+  case R_386_TLS_LDO_32:
+  case R_386_SIZE32:
+  case R_386_TLS_GOTDESC:
+    *(ul32 *)loc = val;
+    break;
+  default:
+    unreachable();
+  }
+}
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  if (ctx.arg.pic) {
+    static const u8 insn[] = {
+      0x51,                   // push   %ecx
+      0x8d, 0x8b, 0, 0, 0, 0, // lea    GOTPLT+4(%ebx), %ecx
+      0xff, 0x31,             // push   (%ecx)
+      0xff, 0x61, 0x04,       // jmp    *0x4(%ecx)
+      0xcc, 0xcc, 0xcc, 0xcc, // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 3) = ctx.gotplt->shdr.sh_addr - ctx.got->shdr.sh_addr + 4;
+  } else {
+    static const u8 insn[] = {
+      0x51,                         // push   %ecx
+      0xb9, 0, 0, 0, 0,             // mov    GOTPLT+4, %ecx
+      0xff, 0x31,                   // push   (%ecx)
+      0xff, 0x61, 0x04,             // jmp    *0x4(%ecx)
+      0xcc, 0xcc, 0xcc, 0xcc, 0xcc, // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 2) = ctx.gotplt->shdr.sh_addr + 4;
+  }
+}
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  if (ctx.arg.pic) {
+    static const u8 insn[] = {
+      0xb9, 0, 0, 0, 0,             // mov $reloc_offset, %ecx
+      0xff, 0xa3, 0, 0, 0, 0,       // jmp *foo@GOT(%ebx)
+      0xcc, 0xcc, 0xcc, 0xcc, 0xcc, // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 1) = sym.get_plt_idx(ctx) * sizeof(ElfRel<E>);
+    *(ul32 *)(buf + 7) = sym.get_gotplt_addr(ctx) - ctx.got->shdr.sh_addr;
+  } else {
+    static const u8 insn[] = {
+      0xb9, 0, 0, 0, 0,             // mov $reloc_offset, %ecx
+      0xff, 0x25, 0, 0, 0, 0,       // jmp *foo@GOT
+      0xcc, 0xcc, 0xcc, 0xcc, 0xcc, // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 1) = sym.get_plt_idx(ctx) * sizeof(ElfRel<E>);
+    *(ul32 *)(buf + 7) = sym.get_gotplt_addr(ctx);
+  }
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  if (ctx.arg.pic) {
+    static const u8 insn[] = {
+      0xff, 0xa3, 0, 0, 0, 0, // jmp *foo@GOT(%ebx)
+      0xcc, 0xcc,             // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 2) = sym.get_got_pltgot_addr(ctx) - ctx.got->shdr.sh_addr;
+  } else {
+    static const u8 insn[] = {
+      0xff, 0x25, 0, 0, 0, 0, // jmp *foo@GOT
+      0xcc, 0xcc,             // (padding)
+    };
+    memcpy(buf, insn, sizeof(insn));
+    *(ul32 *)(buf + 2) = sym.get_got_pltgot_addr(ctx);
+  }
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_386_32:
+    *(ul32 *)loc = val;
+    break;
+  case R_386_PC32:
+    *(ul32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+static u32 relax_got32x(u8 *loc) {
+  // mov imm(%reg1), %reg2 -> lea imm(%reg1), %reg2
+  if (loc[0] == 0x8b)
+    return 0x8d00 | loc[1];
+  return 0;
+}
+
+// Relax GD to LE
+static void relax_gd_to_le(u8 *loc, ElfRel<E> rel, u64 val) {
+  static const u8 insn[] = {
+    0x65, 0xa1, 0, 0, 0, 0, // mov %gs:0, %eax
+    0x81, 0xc0, 0, 0, 0, 0, // add $tp_offset, %eax
+  };
+
+  switch (rel.r_type) {
+  case R_386_PLT32:
+  case R_386_PC32:
+    memcpy(loc - 3, insn, sizeof(insn));
+    *(ul32 *)(loc + 5) = val;
+    break;
+  case R_386_GOT32:
+  case R_386_GOT32X:
+    memcpy(loc - 2, insn, sizeof(insn));
+    *(ul32 *)(loc + 6) = val;
+    break;
+  default:
+    unreachable();
+  }
+}
+
+// Relax LD to LE
+static void relax_ld_to_le(u8 *loc, ElfRel<E> rel, u64 tls_size) {
+  switch (rel.r_type) {
+  case R_386_PLT32:
+  case R_386_PC32: {
+    static const u8 insn[] = {
+      0x65, 0xa1, 0, 0, 0, 0, // mov %gs:0, %eax
+      0x2d, 0, 0, 0, 0,       // sub $tls_size, %eax
+    };
+    memcpy(loc - 2, insn, sizeof(insn));
+    *(ul32 *)(loc + 5) = tls_size;
+    break;
+  }
+  case R_386_GOT32:
+  case R_386_GOT32X: {
+    static const u8 insn[] = {
+      0x65, 0xa1, 0, 0, 0, 0, // mov %gs:0, %eax
+      0x81, 0xe8, 0, 0, 0, 0, // sub $tls_size, %eax
+    };
+    memcpy(loc - 2, insn, sizeof(insn));
+    *(ul32 *)(loc + 6) = tls_size;
+    break;
+  }
+  default:
+    unreachable();
+  }
+}
+
+static u32 relax_tlsdesc_to_ie(u8 *loc) {
+  switch ((loc[0] << 8) | loc[1]) {
+  case 0x8d83: return 0x8b83; // lea 0(%ebx), %eax -> mov 0(%ebx), %eax
+  case 0x8d9b: return 0x8b9b; // lea 0(%ebx), %ebx -> mov 0(%ebx), %ebx
+  case 0x8d8b: return 0x8b8b; // lea 0(%ebx), %ecx -> mov 0(%ebx), %ecx
+  case 0x8d93: return 0x8b93; // lea 0(%ebx), %edx -> mov 0(%ebx), %edx
+  case 0x8db3: return 0x8bb3; // lea 0(%ebx), %esi -> mov 0(%ebx), %esi
+  case 0x8dbb: return 0x8bbb; // lea 0(%ebx), %edi -> mov 0(%ebx), %edi
+  case 0x8da3: return 0x8ba3; // lea 0(%ebx), %esp -> mov 0(%ebx), %esp
+  case 0x8dab: return 0x8bab; // lea 0(%ebx), %ebp -> mov 0(%ebx), %ebp
+  }
+  return 0;
+}
+
+static u32 relax_tlsdesc_to_le(u8 *loc) {
+  switch ((loc[0] << 8) | loc[1]) {
+  case 0x8d83: return 0x90b8; // lea 0(%ebx), %eax -> mov $0, %eax
+  case 0x8d9b: return 0x90bb; // lea 0(%ebx), %ebx -> mov $0, %ebx
+  case 0x8d8b: return 0x90b9; // lea 0(%ebx), %ecx -> mov $0, %ecx
+  case 0x8d93: return 0x90ba; // lea 0(%ebx), %edx -> mov $0, %edx
+  case 0x8db3: return 0x90be; // lea 0(%ebx), %esi -> mov $0, %esi
+  case 0x8dbb: return 0x90bf; // lea 0(%ebx), %edi -> mov $0, %edi
+  case 0x8da3: return 0x90bc; // lea 0(%ebx), %esp -> mov $0, %esp
+  case 0x8dab: return 0x90bd; // lea 0(%ebx), %ebp -> mov $0, %ebp
+  }
+  return 0;
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = get_addend(*this, rel);
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, i, val, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_386_8:
+      check(S + A, 0, 1 << 8);
+      *loc = S + A;
+      break;
+    case R_386_16:
+      check(S + A, 0, 1 << 16);
+      *(ul16 *)loc = S + A;
+      break;
+    case R_386_32:
+      break;
+    case R_386_PC8:
+      check(S + A - P, -(1 << 7), 1 << 7);
+      *loc = S + A - P;
+      break;
+    case R_386_PC16:
+      check(S + A - P, -(1 << 15), 1 << 15);
+      *(ul16 *)loc = S + A - P;
+      break;
+    case R_386_PC32:
+    case R_386_PLT32:
+      *(ul32 *)loc = S + A - P;
+      break;
+    case R_386_GOT32:
+      *(ul32 *)loc = G + A;
+      break;
+    case R_386_GOT32X:
+      if (sym.has_got(ctx)) {
+        *(ul32 *)loc = G + A;
+      } else {
+        u32 insn = relax_got32x(loc - 2);
+        assert(insn);
+        loc[-2] = insn >> 8;
+        loc[-1] = insn;
+        *(ul32 *)loc = S + A - GOT;
+      }
+      break;
+    case R_386_GOTOFF:
+      *(ul32 *)loc = S + A - GOT;
+      break;
+    case R_386_GOTPC:
+      *(ul32 *)loc = GOT + A - P;
+      break;
+    case R_386_TLS_GOTIE:
+      *(ul32 *)loc = sym.get_gottp_addr(ctx) + A - GOT;
+      break;
+    case R_386_TLS_LE:
+      *(ul32 *)loc = S + A - ctx.tp_addr;
+      break;
+    case R_386_TLS_IE:
+      *(ul32 *)loc = sym.get_gottp_addr(ctx) + A;
+      break;
+    case R_386_TLS_GD:
+      if (sym.has_tlsgd(ctx))
+        *(ul32 *)loc = sym.get_tlsgd_addr(ctx) + A - GOT;
+      else
+        relax_gd_to_le(loc, rels[++i], S - ctx.tp_addr);
+      break;
+    case R_386_TLS_LDM:
+      if (ctx.got->has_tlsld(ctx))
+        *(ul32 *)loc = ctx.got->get_tlsld_addr(ctx) + A - GOT;
+      else
+        relax_ld_to_le(loc, rels[++i], ctx.tp_addr - ctx.tls_begin);
+      break;
+    case R_386_TLS_LDO_32:
+      *(ul32 *)loc = S + A - ctx.dtp_addr;
+      break;
+    case R_386_SIZE32:
+      *(ul32 *)loc = sym.esym().st_size + A;
+      break;
+    case R_386_TLS_GOTDESC:
+      // i386 TLSDESC uses the following code sequence to materialize
+      // a TP-relative address in %eax.
+      //
+      //   lea    0(%ebx), %eax
+      //       R_386_TLS_GOTDESC   foo
+      //   call   *(%eax)
+      //       R_386_TLS_DESC_CALL foo
+      //
+      // We may relax the instructions to the following if its TP-relative
+      // address is known at link-time
+      //
+      //   mov     $foo@TPOFF, %eax
+      //   nop
+      //
+      // or to the following if the TP-relative address is known at
+      // process startup time.
+      //
+      //   mov     foo@GOTTPOFF(%ebx), %eax
+      //   nop
+      //
+      // We allow the following alternative code sequence too because
+      // LLVM emits such code.
+      //
+      //   lea    0(%ebx), %reg
+      //       R_386_TLS_GOTDESC   foo
+      //   mov    %reg, %eax
+      //   call   *(%eax)
+      //       R_386_TLS_DESC_CALL foo
+      //
+      // Note that the compiler always uses the local-exec TLS model
+      // for -fno-pic, so TLSDESC code is always PIC (i.e. uses %ebx to
+      // store the address of GOT.)
+      if (sym.has_tlsdesc(ctx)) {
+        *(ul32 *)loc = sym.get_tlsdesc_addr(ctx) + A - GOT;
+      } else if (sym.has_gottp(ctx)) {
+        u32 insn = relax_tlsdesc_to_ie(loc - 2);
+        if (!insn)
+          Fatal(ctx) << *this << ": illegal instruction sequence for TLSDESC";
+        loc[-2] = insn >> 8;
+        loc[-1] = insn;
+        *(ul32 *)loc = sym.get_gottp_addr(ctx) + A - GOT;
+      } else {
+        u32 insn = relax_tlsdesc_to_le(loc - 2);
+        if (!insn)
+          Fatal(ctx) << *this << ": illegal instruction sequence for TLSDESC";
+        loc[-2] = insn >> 8;
+        loc[-1] = insn;
+        *(ul32 *)loc = S + A - ctx.tp_addr;
+      }
+      break;
+    case R_386_TLS_DESC_CALL:
+      if (!sym.has_tlsdesc(ctx)) {
+        // call *(%eax) -> nop
+        loc[0] = 0x66;
+        loc[1] = 0x90;
+      }
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : get_addend(*this, rel);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, val, i, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_386_8:
+      check(S + A, 0, 1 << 8);
+      *loc = S + A;
+      break;
+    case R_386_16:
+      check(S + A, 0, 1 << 16);
+      *(ul16 *)loc = S + A;
+      break;
+    case R_386_32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ul32 *)loc = *val;
+      else
+        *(ul32 *)loc = S + A;
+      break;
+    case R_386_PC8:
+      check(S + A, -(1 << 7), 1 << 7);
+      *loc = S + A;
+      break;
+    case R_386_PC16:
+      check(S + A, -(1 << 15), 1 << 15);
+      *(ul16 *)loc = S + A;
+      break;
+    case R_386_PC32:
+      *(ul32 *)loc = S + A;
+      break;
+    case R_386_GOTPC:
+      *(ul32 *)loc = GOT + A;
+      break;
+    case R_386_GOTOFF:
+      *(ul32 *)loc = S + A - GOT;
+      break;
+    case R_386_TLS_LDO_32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ul32 *)loc = *val;
+      else
+        *(ul32 *)loc = S + A - ctx.dtp_addr;
+      break;
+    case R_386_SIZE32:
+      *(ul32 *)loc = sym.esym().st_size + A;
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  // Scan relocations
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = (u8 *)(contents.data() + rel.r_offset);
+
+    if (sym.is_ifunc())
+      sym.flags |= NEEDS_GOT | NEEDS_PLT;
+
+    if (rel.r_type == R_386_TLS_GD || rel.r_type == R_386_TLS_LDM) {
+      if (i + 1 == rels.size())
+        Fatal(ctx) << *this << ": " << rel << " must be followed by PLT or GOT32";
+
+      if (u32 ty = rels[i + 1].r_type;
+          ty != R_386_PLT32 && ty != R_386_PC32 &&
+          ty != R_386_GOT32 && ty != R_386_GOT32X)
+        Fatal(ctx) << *this << ": " << rel << " must be followed by PLT or GOT32";
+    }
+
+    switch (rel.r_type) {
+    case R_386_8:
+    case R_386_16:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_386_PC8:
+    case R_386_PC16:
+    case R_386_PC32:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_386_GOT32:
+    case R_386_GOTPC:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_386_GOT32X:
+      // We always want to relax GOT32X even if --no-relax is given
+      // because static PIE doesn't work without it.
+      if (sym.is_pcrel_linktime_const(ctx) && relax_got32x(loc - 2)) {
+        // Do nothing
+      } else {
+        sym.flags |= NEEDS_GOT;
+      }
+      break;
+    case R_386_PLT32:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_386_TLS_GOTIE:
+    case R_386_TLS_IE:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_386_TLS_GD:
+      // We always relax if -static because libc.a doesn't contain
+      // __tls_get_addr().
+      if (ctx.arg.static_ || (ctx.arg.relax && sym.is_tprel_linktime_const(ctx)))
+        i++;
+      else
+        sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_386_TLS_LDM:
+      // We always relax if -static because libc.a doesn't contain
+      // __tls_get_addr().
+      if (ctx.arg.static_ || (ctx.arg.relax && !ctx.arg.shared))
+        i++;
+      else
+        ctx.needs_tlsld = true;
+      break;
+    case R_386_TLS_GOTDESC:
+      scan_tlsdesc(ctx, sym);
+      break;
+    case R_386_TLS_LE:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_386_32:
+    case R_386_GOTOFF:
+    case R_386_TLS_LDO_32:
+    case R_386_SIZE32:
+    case R_386_TLS_DESC_CALL:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+} // namespace mold

src/arch-m68k.cc

@@ -0,0 +1,312 @@
+// This file contains code for the Motorola 68000 series microprocessors,
+// which is often abbreviated as m68k. Running a Unix-like system on a
+// m68k-based machine today is a retro-computing hobby activity, but the
+// processor was a popular choice to build Unix computers during '80s.
+// Early Sun workstations for example used m68k. Macintosh until 1994 were
+// based on m68k as well until they switched to PowerPC (and then to x86
+// and to ARM.)
+//
+// From the linker's point of view, it is not hard to support m68k. It's
+// just a 32-bit big-endian CISC ISA. Compared to comtemporary i386,
+// m68k's psABI is actually simpler because m68k has PC-relative memory
+// access instructions and therefore can support position-independent
+// code without too much hassle.
+//
+// https://github.com/rui314/psabi/blob/main/m68k.pdf
+
+#include "mold.h"
+
+namespace mold {
+
+using E = M68K;
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  static const u8 insn[] = {
+    0x2f, 0x00,                         // move.l %d0, -(%sp)
+    0x2f, 0x3b, 0x01, 0x70, 0, 0, 0, 0, // move.l (GOTPLT+4, %pc), -(%sp)
+    0x4e, 0xfb, 0x01, 0x71, 0, 0, 0, 0, // jmp    ([GOTPLT+8, %pc])
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(ub32 *)(buf + 6) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr;
+  *(ub32 *)(buf + 14) = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr - 4;
+}
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static const u8 insn[] = {
+    0x20, 0x3c, 0, 0, 0, 0,             // move.l PLT_OFFSET, %d0
+    0x4e, 0xfb, 0x01, 0x71, 0, 0, 0, 0, // jmp    ([GOTPLT_ENTRY, %pc])
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(ub32 *)(buf + 2) = sym.get_plt_idx(ctx) * sizeof(ElfRel<E>);
+  *(ub32 *)(buf + 10) = sym.get_gotplt_addr(ctx) - sym.get_plt_addr(ctx) - 8;
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static const u8 insn[] = {
+    0x4e, 0xfb, 0x01, 0x71, 0, 0, 0, 0, // jmp ([GOT_ENTRY, %pc])
+  };
+
+  memcpy(buf, insn, sizeof(insn));
+  *(ub32 *)(buf + 4) = sym.get_got_pltgot_addr(ctx) - sym.get_plt_addr(ctx) - 2;
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_68K_32:
+    *(ub32 *)loc = val;
+    break;
+  case R_68K_PC32:
+    *(ub32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = rel.r_addend;
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, i, val, lo, hi);
+    };
+
+    auto write16 = [&](u64 val) {
+      check(val, 0, 1 << 16);
+      *(ub16 *)loc = val;
+    };
+
+    auto write16s = [&](u64 val) {
+      check(val, -(1 << 15), 1 << 15);
+      *(ub16 *)loc = val;
+    };
+
+    auto write8 = [&](u64 val) {
+      check(val, 0, 1 << 8);
+      *loc = val;
+    };
+
+    auto write8s = [&](u64 val) {
+      check(val, -(1 << 7), 1 << 7);
+      *loc = val;
+    };
+
+    switch (rel.r_type) {
+    case R_68K_32:
+      break;
+    case R_68K_16:
+      write16(S + A);
+      break;
+    case R_68K_8:
+      write8(S + A);
+      break;
+    case R_68K_PC32:
+    case R_68K_PLT32:
+      *(ub32 *)loc = S + A - P;
+      break;
+    case R_68K_PC16:
+    case R_68K_PLT16:
+      write16s(S + A - P);
+      break;
+    case R_68K_PC8:
+    case R_68K_PLT8:
+      write8s(S + A - P);
+      break;
+    case R_68K_GOTPCREL32:
+      *(ub32 *)loc = GOT + A - P;
+      break;
+    case R_68K_GOTPCREL16:
+      write16s(GOT + A - P);
+      break;
+    case R_68K_GOTPCREL8:
+      write8s(GOT + A - P);
+      break;
+    case R_68K_GOTOFF32:
+      *(ub32 *)loc = G + A;
+      break;
+    case R_68K_GOTOFF16:
+      write16(G + A);
+      break;
+    case R_68K_GOTOFF8:
+      write8(G + A);
+      break;
+    case R_68K_TLS_GD32:
+      *(ub32 *)loc = sym.get_tlsgd_addr(ctx) + A - GOT;
+      break;
+    case R_68K_TLS_GD16:
+      write16(sym.get_tlsgd_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_GD8:
+      write8(sym.get_tlsgd_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_LDM32:
+      *(ub32 *)loc = ctx.got->get_tlsld_addr(ctx) + A - GOT;
+      break;
+    case R_68K_TLS_LDM16:
+      write16(ctx.got->get_tlsld_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_LDM8:
+      write8(ctx.got->get_tlsld_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_LDO32:
+      *(ub32 *)loc = S + A - ctx.dtp_addr;
+      break;
+    case R_68K_TLS_LDO16:
+      write16s(S + A - ctx.dtp_addr);
+      break;
+    case R_68K_TLS_LDO8:
+      write8s(S + A - ctx.dtp_addr);
+      break;
+    case R_68K_TLS_IE32:
+      *(ub32 *)loc = sym.get_gottp_addr(ctx) + A - GOT;
+      break;
+    case R_68K_TLS_IE16:
+      write16(sym.get_gottp_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_IE8:
+      write8(sym.get_gottp_addr(ctx) + A - GOT);
+      break;
+    case R_68K_TLS_LE32:
+      *(ub32 *)loc = S + A - ctx.tp_addr;
+      break;
+    case R_68K_TLS_LE16:
+      write16(S + A - ctx.tp_addr);
+      break;
+    case R_68K_TLS_LE8:
+      write8(S + A - ctx.tp_addr);
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : (i64)rel.r_addend;
+
+    switch (rel.r_type) {
+    case R_68K_32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ub32 *)loc = *val;
+      else
+        *(ub32 *)loc = S + A;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+
+    if (sym.is_ifunc())
+      Error(ctx) << sym << ": GNU ifunc symbol is not supported on m68k";
+
+    switch (rel.r_type) {
+    case R_68K_16:
+    case R_68K_8:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_68K_PC32:
+    case R_68K_PC16:
+    case R_68K_PC8:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_68K_GOTPCREL32:
+    case R_68K_GOTPCREL16:
+    case R_68K_GOTPCREL8:
+    case R_68K_GOTOFF32:
+    case R_68K_GOTOFF16:
+    case R_68K_GOTOFF8:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_68K_PLT32:
+    case R_68K_PLT16:
+    case R_68K_PLT8:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_68K_TLS_GD32:
+    case R_68K_TLS_GD16:
+    case R_68K_TLS_GD8:
+      sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_68K_TLS_LDM32:
+    case R_68K_TLS_LDM16:
+    case R_68K_TLS_LDM8:
+      ctx.needs_tlsld = true;
+      break;
+    case R_68K_TLS_IE32:
+    case R_68K_TLS_IE16:
+    case R_68K_TLS_IE8:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_68K_TLS_LE32:
+    case R_68K_TLS_LE16:
+    case R_68K_TLS_LE8:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_68K_32:
+    case R_68K_TLS_LDO32:
+    case R_68K_TLS_LDO16:
+    case R_68K_TLS_LDO8:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+} // namespace mold

src/arch-ppc32.cc

@@ -0,0 +1,442 @@
+// This file implements the PowerPC 32-bit ISA. For 64-bit PowerPC, see
+// arch-ppc64v1.cpp and arch-ppc64v2.cpp.
+//
+// PPC32 is a RISC ISA. It has 32 general-purpose registers (GPRs).
+// r0, r11 and r12 are reserved for static linkers, so we can use these
+// registers in PLTs and range extension thunks. In addition to that, it
+// has a few special registers. Notable ones are LR which holds a return
+// address and CTR which we can use to store a branch target address.
+//
+// It feels that the PPC32 psABI is unnecessarily complicated at first
+// glance, but that is mainly stemmed from the fact that the ISA lacks
+// PC-relative load/store instructions. Since machine instructions cannot
+// load data relative to its own address, it is not straightforward to
+// support position-independent code (PIC) on PPC32.
+//
+// A position-independent function typically contains the following code
+// in the prologue to obtain its own address:
+//
+//    mflr  r0        // save the current return address to %r0
+//    bcl   20, 31, 4 // call the next instruction as if it were a function
+//    mtlr  r12       // save the return address to %r12
+//    mtlr  r0        // restore the original return address
+//
+// An object file compiled with -fPIC contains a data section named
+// `.got2` to store addresses of locally-defined global variables and
+// constants. A PIC function usually computes its .got2+0x8000 and set it
+// to %r30. This scheme allows the function to access global objects
+// defined in the same input file with a single %r30-relative load/store
+// instruction with a 16-bit offset, given that .got2 is smaller than
+// 0x10000 (or 65536) bytes.
+//
+// Since each object file has its own .got2, %r30 refers to different
+// places in a merged .got2 for two functions that came from different
+// input files. Therefore, %r30 makes sense only within a single function.
+//
+// Technically, we can reuse a %r30 value in our PLT if we create a PLT
+// _for each input file_ (that's what GNU ld seems to be doing), but that
+// doesn't seems to be worth its complexity. Our PLT simply doesn't rely
+// on a %r30 value.
+//
+// https://github.com/rui314/psabi/blob/main/ppc32.pdf
+
+#include "mold.h"
+
+namespace mold {
+
+using E = PPC32;
+
+static u64 lo(u64 x)    { return x & 0xffff; }
+static u64 hi(u64 x)    { return x >> 16; }
+static u64 ha(u64 x)    { return (x + 0x8000) >> 16; }
+static u64 high(u64 x)  { return (x >> 16) & 0xffff; }
+static u64 higha(u64 x) { return ((x + 0x8000) >> 16) & 0xffff; }
+
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  constexpr ub32 insn[] = {
+    // Get the address of this PLT section
+    0x7c08'02a6, //    mflr    r0
+    0x429f'0005, //    bcl     20, 31, 4
+    0x7d88'02a6, // 1: mflr    r12
+    0x7c08'03a6, //    mtlr    r0
+
+    // Compute the runtime address of GOTPLT+12
+    0x3d8c'0000, //    addis   r12, r12, (GOTPLT - 1b)@higha
+    0x398c'0000, //    addi    r12, r12, (GOTPLT - 1b)@lo
+
+    // Compute the PLT entry offset
+    0x7d6c'5850, //    sub     r11, r11, r12
+    0x1d6b'0003, //    mulli   r11, r11, 3
+
+    // Load GOTPLT[2] and branch to GOTPLT[1]
+    0x800c'fff8, //    lwz     r0,  -8(r12)
+    0x7c09'03a6, //    mtctr   r0
+    0x818c'fffc, //    lwz     r12, -4(r12)
+    0x4e80'0420, //    bctr
+    0x6000'0000, //    nop
+    0x6000'0000, //    nop
+    0x6000'0000, //    nop
+    0x6000'0000, //    nop
+  };
+
+  static_assert(sizeof(insn) == E::plt_hdr_size);
+  memcpy(buf, insn, sizeof(insn));
+
+  ub32 *loc = (ub32 *)buf;
+  loc[4] |= higha(ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 4);
+  loc[5] |= lo(ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr + 4);
+}
+
+static constexpr ub32 plt_entry[] = {
+  // Get the address of this PLT entry
+  0x7c08'02a6, // mflr    r0
+  0x429f'0005, // bcl     20, 31, 4
+  0x7d88'02a6, // mflr    r12
+  0x7c08'03a6, // mtlr    r0
+
+  // Load an address from the GOT/GOTPLT entry and jump to that address
+  0x3d6c'0000, // addis   r11, r12, OFFSET@higha
+  0x396b'0000, // addi    r11, r11, OFFSET@lo
+  0x818b'0000, // lwz     r12, 0(r11)
+  0x7d89'03a6, // mtctr   r12
+  0x4e80'0420, // bctr
+};
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static_assert(E::plt_size == sizeof(plt_entry));
+  memcpy(buf, plt_entry, sizeof(plt_entry));
+
+  ub32 *loc = (ub32 *)buf;
+  i64 offset = sym.get_gotplt_addr(ctx) - sym.get_plt_addr(ctx) - 8;
+  loc[4] |= higha(offset);
+  loc[5] |= lo(offset);
+}
+
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  static_assert(E::pltgot_size == sizeof(plt_entry));
+  memcpy(buf, plt_entry, sizeof(plt_entry));
+
+  ub32 *loc = (ub32 *)buf;
+  i64 offset = sym.get_got_pltgot_addr(ctx) - sym.get_plt_addr(ctx) - 8;
+  loc[4] |= higha(offset);
+  loc[5] |= lo(offset);
+}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_PPC_ADDR32:
+    *(ub32 *)loc = val;
+    break;
+  case R_PPC_REL32:
+    *(ub32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  u64 GOT2 = file.extra.got2 ? file.extra.got2->get_addr() : 0;
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = rel.r_addend;
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+
+    switch (rel.r_type) {
+    case R_PPC_ADDR14:
+      *(ub32 *)loc |= bits(S + A, 15, 2) << 2;
+      break;
+    case R_PPC_ADDR16:
+    case R_PPC_UADDR16:
+    case R_PPC_ADDR16_LO:
+      *(ub16 *)loc = lo(S + A);
+      break;
+    case R_PPC_ADDR16_HI:
+      *(ub16 *)loc = hi(S + A);
+      break;
+    case R_PPC_ADDR16_HA:
+      *(ub16 *)loc = ha(S + A);
+      break;
+    case R_PPC_ADDR24:
+      *(ub32 *)loc |= bits(S + A, 25, 2) << 2;
+      break;
+    case R_PPC_ADDR30:
+      *(ub32 *)loc |= bits(S + A, 31, 2) << 2;
+      break;
+    case R_PPC_PLT16_LO:
+      *(ub16 *)loc = lo(G + GOT - A - GOT2);
+      break;
+    case R_PPC_PLT16_HI:
+      *(ub16 *)loc = hi(G + GOT - A - GOT2);
+      break;
+    case R_PPC_PLT16_HA:
+      *(ub16 *)loc = ha(G + GOT - A - GOT2);
+      break;
+    case R_PPC_PLT32:
+      *(ub32 *)loc = G + GOT - A - GOT2;
+      break;
+    case R_PPC_REL14:
+      *(ub32 *)loc |= bits(S + A - P, 15, 2) << 2;
+      break;
+    case R_PPC_REL16:
+    case R_PPC_REL16_LO:
+      *(ub16 *)loc = lo(S + A - P);
+      break;
+    case R_PPC_REL16_HI:
+      *(ub16 *)loc = hi(S + A - P);
+      break;
+    case R_PPC_REL16_HA:
+      *(ub16 *)loc = ha(S + A - P);
+      break;
+    case R_PPC_REL24:
+    case R_PPC_LOCAL24PC: {
+      i64 val = S + A - P;
+      if (!is_int(val, 26))
+        val = sym.get_thunk_addr(ctx, P) - P;
+      *(ub32 *)loc |= bits(val, 25, 2) << 2;
+      break;
+    }
+    case R_PPC_PLTREL24: {
+      i64 val = S - P;
+      if (sym.has_plt(ctx) || !is_int(val, 26))
+        val = sym.get_thunk_addr(ctx, P) - P;
+      *(ub32 *)loc |= bits(val, 25, 2) << 2;
+      break;
+    }
+    case R_PPC_REL32:
+    case R_PPC_PLTREL32:
+      *(ub32 *)loc = S + A - P;
+      break;
+    case R_PPC_GOT16:
+    case R_PPC_GOT16_LO:
+      *(ub16 *)loc = lo(G + A);
+      break;
+    case R_PPC_GOT16_HI:
+      *(ub16 *)loc = hi(G + A);
+      break;
+    case R_PPC_GOT16_HA:
+      *(ub16 *)loc = ha(G + A);
+      break;
+    case R_PPC_TPREL16_LO:
+      *(ub16 *)loc = lo(S + A - ctx.tp_addr);
+      break;
+    case R_PPC_TPREL16_HI:
+      *(ub16 *)loc = hi(S + A - ctx.tp_addr);
+      break;
+    case R_PPC_TPREL16_HA:
+      *(ub16 *)loc = ha(S + A - ctx.tp_addr);
+      break;
+    case R_PPC_DTPREL16_LO:
+      *(ub16 *)loc = lo(S + A - ctx.dtp_addr);
+      break;
+    case R_PPC_DTPREL16_HI:
+      *(ub16 *)loc = hi(S + A - ctx.dtp_addr);
+      break;
+    case R_PPC_DTPREL16_HA:
+      *(ub16 *)loc = ha(S + A - ctx.dtp_addr);
+      break;
+    case R_PPC_GOT_TLSGD16:
+      *(ub16 *)loc = sym.get_tlsgd_addr(ctx) - GOT;
+      break;
+    case R_PPC_GOT_TLSLD16:
+      *(ub16 *)loc = ctx.got->get_tlsld_addr(ctx) - GOT;
+      break;
+    case R_PPC_GOT_TPREL16:
+      *(ub16 *)loc = sym.get_gottp_addr(ctx) - GOT;
+      break;
+    case R_PPC_ADDR32:
+    case R_PPC_UADDR32:
+    case R_PPC_TLS:
+    case R_PPC_TLSGD:
+    case R_PPC_TLSLD:
+    case R_PPC_PLTSEQ:
+    case R_PPC_PLTCALL:
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : (i64)rel.r_addend;
+
+    switch (rel.r_type) {
+    case R_PPC_ADDR32:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ub32 *)loc = *val;
+      else
+        *(ub32 *)loc = S + A;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  // Scan relocations
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+
+    if (sym.is_ifunc())
+      sym.flags |= NEEDS_GOT | NEEDS_PLT;
+
+    switch (rel.r_type) {
+    case R_PPC_ADDR14:
+    case R_PPC_ADDR16:
+    case R_PPC_UADDR16:
+    case R_PPC_ADDR16_LO:
+    case R_PPC_ADDR16_HI:
+    case R_PPC_ADDR16_HA:
+    case R_PPC_ADDR24:
+    case R_PPC_ADDR30:
+      scan_absrel(ctx, sym, rel);
+      break;
+    case R_PPC_REL14:
+    case R_PPC_REL16:
+    case R_PPC_REL16_LO:
+    case R_PPC_REL16_HI:
+    case R_PPC_REL16_HA:
+    case R_PPC_REL32:
+      scan_pcrel(ctx, sym, rel);
+      break;
+    case R_PPC_GOT16:
+    case R_PPC_GOT16_LO:
+    case R_PPC_GOT16_HI:
+    case R_PPC_GOT16_HA:
+    case R_PPC_PLT16_LO:
+    case R_PPC_PLT16_HI:
+    case R_PPC_PLT16_HA:
+    case R_PPC_PLT32:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_PPC_REL24:
+    case R_PPC_PLTREL24:
+    case R_PPC_PLTREL32:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_PPC_GOT_TLSGD16:
+      sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_PPC_GOT_TLSLD16:
+      ctx.needs_tlsld = true;
+      break;
+    case R_PPC_GOT_TPREL16:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_PPC_TPREL16_LO:
+    case R_PPC_TPREL16_HI:
+    case R_PPC_TPREL16_HA:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_PPC_ADDR32:
+    case R_PPC_UADDR32:
+    case R_PPC_LOCAL24PC:
+    case R_PPC_TLS:
+    case R_PPC_TLSGD:
+    case R_PPC_TLSLD:
+    case R_PPC_DTPREL16_LO:
+    case R_PPC_DTPREL16_HI:
+    case R_PPC_DTPREL16_HA:
+    case R_PPC_PLTSEQ:
+    case R_PPC_PLTCALL:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+void Thunk<E>::copy_buf(Context<E> &ctx) {
+  constexpr ub32 local_thunk[] = {
+    // Get this thunk's address
+    0x7c08'02a6, // mflr    r0
+    0x429f'0005, // bcl     20, 31, 4
+    0x7d88'02a6, // mflr    r12
+    0x7c08'03a6, // mtlr    r0
+
+    // Materialize the destination's address in %r11 and jump to that address
+    0x3d6c'0000, // addis   r11, r12, OFFSET@higha
+    0x396b'0000, // addi    r11, r11, OFFSET@lo
+    0x7d69'03a6, // mtctr   r11
+    0x4e80'0420, // bctr
+    0x6000'0000, // nop
+  };
+
+  static_assert(E::thunk_size == sizeof(plt_entry));
+  static_assert(E::thunk_size == sizeof(local_thunk));
+
+  u8 *buf = ctx.buf + output_section.shdr.sh_offset + offset;
+  u64 P = output_section.shdr.sh_addr + offset;
+
+  for (Symbol<E> *sym : symbols) {
+    if (sym->has_plt(ctx)) {
+      u64 got =
+        sym->has_got(ctx) ? sym->get_got_addr(ctx) : sym->get_gotplt_addr(ctx);
+      i64 val = got - P - 8;
+      memcpy(buf, plt_entry, sizeof(plt_entry));
+      *(ub32 *)(buf + 16) |= higha(val);
+      *(ub32 *)(buf + 20) |= lo(val);
+    } else {
+      i64 val = sym->get_addr(ctx) - P - 8;
+      memcpy(buf, local_thunk, sizeof(local_thunk));
+      *(ub32 *)(buf + 16) |= higha(val);
+      *(ub32 *)(buf + 20) |= lo(val);
+    }
+
+    buf += E::thunk_size;
+    P += E::thunk_size;
+  }
+}
+
+} // namespace mold

src/arch-ppc64v1.cc

@@ -0,0 +1,672 @@
+// This file contains code for the 64-bit PowerPC ELFv1 ABI that is
+// commonly used for big-endian PPC systems. Modern PPC systems that use
+// the processor in the little-endian mode use the ELFv2 ABI instead. For
+// ELFv2, see arch-ppc64v2.cc.
+//
+// Even though they are similiar, ELFv1 isn't only different from ELFv2 in
+// endianness. The most notable difference is, in ELFv1, a function
+// pointer doesn't directly refer to the entry point of a function but
+// instead refers to a data structure so-called "function descriptor".
+//
+// The function descriptor is essentially a pair of a function entry point
+// address and a value that should be set to %r2 before calling that
+// function. There is also a third member for "the environment pointer for
+// languages such as Pascal and PL/1" according to the psABI, but it looks
+// like no one acutally uses it. In total, the function descriptor is 24
+// bytes long. Here is why we need it.
+//
+// PPC generally lacks PC-relative data access instructions. Position-
+// independent code sets GOT + 0x8000 to %r2 and access global variables
+// relative to %r2.
+//
+// Each ELF file has its own GOT. If a function calls another function in
+// the same ELF file, it doesn't have to reset %r2. However, if it is in
+// other file (e.g. other .so), it has to set a new value to %r2 so that
+// the register contains the callee's GOT + 0x8000.
+//
+// In this way, you can't call a function just by knowing the function's
+// entry point address. You also need to know a proper %r2 value for the
+// function. This is why a function pointer refers to a tuple of an
+// address and a %r2 value.
+//
+// If a function call is made through PLT, PLT takes care of restoring %r2.
+// Therefore, the caller has to restore %r2 only for function calls
+// through function pointers.
+//
+// .opd (short for "official procedure descriptors") contains function
+// descriptors.
+//
+// You can think OPD as this: even in other targets, a function can have a
+// few different addresses for different purposes. It may not only have an
+// entry point address but may also have PLT and/or GOT addresses.
+// In PPCV1, it may have an OPD address in addition to these. OPD address
+// is used for relocations that refers to the address of a function as a
+// function pointer.
+//
+// https://github.com/rui314/psabi/blob/main/ppc64v1.pdf
+
+#include "mold.h"
+
+#include <algorithm>
+#include <tbb/parallel_for_each.h>
+
+namespace mold {
+
+using E = PPC64V1;
+
+static u64 lo(u64 x)    { return x & 0xffff; }
+static u64 hi(u64 x)    { return x >> 16; }
+static u64 ha(u64 x)    { return (x + 0x8000) >> 16; }
+static u64 high(u64 x)  { return (x >> 16) & 0xffff; }
+static u64 higha(u64 x) { return ((x + 0x8000) >> 16) & 0xffff; }
+
+// .plt is used only for lazy symbol resolution on PPC64. All PLT
+// calls are made via range extension thunks even if they are within
+// reach. Thunks read addresses from .got.plt and jump there.
+// Therefore, once PLT symbols are resolved and final addresses are
+// written to .got.plt, thunks just skip .plt and directly jump to the
+// resolved addresses.
+template <>
+void write_plt_header(Context<E> &ctx, u8 *buf) {
+  constexpr ub32 insn[] = {
+    0x7d88'02a6, // mflr    r12
+    0x429f'0005, // bcl     20, 31, 4 // obtain PC
+    0x7d68'02a6, // mflr    r11
+    0x7d88'03a6, // mtlr    r12
+    0x3d6b'0000, // addis   r11, r11, GOTPLT_OFFSET@ha
+    0x396b'0000, // addi    r11, r11, GOTPLT_OFFSET@lo
+    0xe98b'0000, // ld      r12,0(r11)
+    0xe84b'0008, // ld      r2,8(r11)
+    0x7d89'03a6, // mtctr   r12
+    0xe96b'0010, // ld      r11,16(r11)
+    0x4e80'0420, // bctr
+  };
+
+  static_assert(sizeof(insn) == E::plt_hdr_size);
+  memcpy(buf, insn, sizeof(insn));
+
+  i64 val = ctx.gotplt->shdr.sh_addr - ctx.plt->shdr.sh_addr - 8;
+  *(ub32 *)(buf + 16) |= higha(val);
+  *(ub32 *)(buf + 20) |= lo(val);
+}
+
+template <>
+void write_plt_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {
+  ub32 *loc = (ub32 *)buf;
+  i64 idx = sym.get_plt_idx(ctx);
+
+  // The PPC64 ELFv1 ABI requires PLT entries to be vary in size depending
+  // on their indices. Unlike other targets, .got.plt is filled not by us
+  // but by the loader, so we don't have a control over where the initial
+  // call to the PLT entry jumps to. So we need to strictly follow the PLT
+  // section layout as the loader expect it to be.
+  if (idx < 0x8000) {
+    constexpr ub32 insn[] = {
+      0x3800'0000, // li      r0, PLT_INDEX
+      0x4b00'0000, // b       plt0
+    };
+
+    memcpy(loc, insn, sizeof(insn));
+    loc[0] |= idx;
+    loc[1] |= (ctx.plt->shdr.sh_addr - sym.get_plt_addr(ctx) - 4) & 0x00ff'ffff;
+  } else {
+    constexpr ub32 insn[] = {
+      0x3c00'0000, // lis     r0, PLT_INDEX@high
+      0x6000'0000, // ori     r0, r0, PLT_INDEX@lo
+      0x4b00'0000, // b       plt0
+    };
+
+    memcpy(loc, insn, sizeof(insn));
+    loc[0] |= high(idx);
+    loc[1] |= lo(idx);
+    loc[2] |= (ctx.plt->shdr.sh_addr - sym.get_plt_addr(ctx) - 8) & 0x00ff'ffff;
+  }
+}
+
+// .plt.got is not necessary on PPC64 because range extension thunks
+// directly read GOT entries and jump there.
+template <>
+void write_pltgot_entry(Context<E> &ctx, u8 *buf, Symbol<E> &sym) {}
+
+template <>
+void EhFrameSection<E>::apply_eh_reloc(Context<E> &ctx, const ElfRel<E> &rel,
+                                       u64 offset, u64 val) {
+  u8 *loc = ctx.buf + this->shdr.sh_offset + offset;
+
+  switch (rel.r_type) {
+  case R_NONE:
+    break;
+  case R_PPC64_ADDR64:
+    *(ub64 *)loc = val;
+    break;
+  case R_PPC64_REL32:
+    *(ub32 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  case R_PPC64_REL64:
+    *(ub64 *)loc = val - this->shdr.sh_addr - offset;
+    break;
+  default:
+    Fatal(ctx) << "unsupported relocation in .eh_frame: " << rel;
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_alloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE)
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    u64 S = sym.get_addr(ctx);
+    u64 A = rel.r_addend;
+    u64 P = get_addr() + rel.r_offset;
+    u64 G = sym.get_got_idx(ctx) * sizeof(Word<E>);
+    u64 GOT = ctx.got->shdr.sh_addr;
+    u64 TOC = ctx.extra.TOC->value;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, i, val, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_PPC64_TOC:
+      break;
+    case R_PPC64_TOC16_HA:
+      *(ub16 *)loc = ha(S + A - TOC);
+      break;
+    case R_PPC64_TOC16_LO:
+      *(ub16 *)loc = lo(S + A - TOC);
+      break;
+    case R_PPC64_TOC16_DS:
+      check(S + A - TOC, -(1 << 15), 1 << 15);
+      *(ub16 *)loc |= (S + A - TOC) & 0xfffc;
+      break;
+    case R_PPC64_TOC16_LO_DS:
+      *(ub16 *)loc |= (S + A - TOC) & 0xfffc;
+      break;
+    case R_PPC64_REL24: {
+      i64 val = sym.get_addr(ctx, NO_OPD) + A - P;
+      if (sym.has_plt(ctx) || !is_int(val, 26))
+        val = sym.get_thunk_addr(ctx, P) + A - P;
+
+      check(val, -(1 << 25), 1 << 25);
+      *(ub32 *)loc |= bits(val, 25, 2) << 2;
+
+      // If a callee is an external function, PLT saves %r2 to the
+      // caller's r2 save slot. We need to restore it after function
+      // return. To do so, there's usually a NOP as a placeholder
+      // after a BL. 0x6000'0000 is a NOP.
+      if (sym.has_plt(ctx) && *(ub32 *)(loc + 4) == 0x6000'0000)
+        *(ub32 *)(loc + 4) = 0xe841'0028; // ld r2, 40(r1)
+      break;
+    }
+    case R_PPC64_REL32:
+      *(ub32 *)loc = S + A - P;
+      break;
+    case R_PPC64_REL64:
+      *(ub64 *)loc = S + A - P;
+      break;
+    case R_PPC64_REL16_HA:
+      *(ub16 *)loc = ha(S + A - P);
+      break;
+    case R_PPC64_REL16_LO:
+      *(ub16 *)loc = lo(S + A - P);
+      break;
+    case R_PPC64_PLT16_HA:
+      *(ub16 *)loc = ha(G + GOT - TOC);
+      break;
+    case R_PPC64_PLT16_HI:
+      *(ub16 *)loc = hi(G + GOT - TOC);
+      break;
+    case R_PPC64_PLT16_LO:
+      *(ub16 *)loc = lo(G + GOT - TOC);
+      break;
+    case R_PPC64_PLT16_LO_DS:
+      *(ub16 *)loc |= (G + GOT - TOC) & 0xfffc;
+      break;
+    case R_PPC64_GOT_TPREL16_HA:
+      *(ub16 *)loc = ha(sym.get_gottp_addr(ctx) - TOC);
+      break;
+    case R_PPC64_GOT_TLSGD16_HA:
+      *(ub16 *)loc = ha(sym.get_tlsgd_addr(ctx) - TOC);
+      break;
+    case R_PPC64_GOT_TLSGD16_LO:
+      *(ub16 *)loc = lo(sym.get_tlsgd_addr(ctx) - TOC);
+      break;
+    case R_PPC64_GOT_TLSLD16_HA:
+      *(ub16 *)loc = ha(ctx.got->get_tlsld_addr(ctx) - TOC);
+      break;
+    case R_PPC64_GOT_TLSLD16_LO:
+      *(ub16 *)loc = lo(ctx.got->get_tlsld_addr(ctx) - TOC);
+      break;
+    case R_PPC64_DTPREL16_HA:
+      *(ub16 *)loc = ha(S + A - ctx.dtp_addr);
+      break;
+    case R_PPC64_DTPREL16_LO:
+      *(ub16 *)loc = lo(S + A - ctx.dtp_addr);
+      break;
+    case R_PPC64_DTPREL16_LO_DS:
+      *(ub16 *)loc |= (S + A - ctx.dtp_addr) & 0xfffc;
+      break;
+    case R_PPC64_TPREL16_HA:
+      *(ub16 *)loc = ha(S + A - ctx.tp_addr);
+      break;
+    case R_PPC64_TPREL16_LO:
+      *(ub16 *)loc = lo(S + A - ctx.tp_addr);
+      break;
+    case R_PPC64_TPREL16_LO_DS:
+      *(ub16 *)loc |= (S + A - ctx.tp_addr) & 0xfffc;
+      break;
+    case R_PPC64_GOT_TPREL16_LO_DS:
+      *(ub16 *)loc |= (sym.get_gottp_addr(ctx) - TOC) & 0xfffc;
+      break;
+    case R_PPC64_ADDR64:
+    case R_PPC64_PLTSEQ:
+    case R_PPC64_PLTCALL:
+    case R_PPC64_TLS:
+    case R_PPC64_TLSGD:
+    case R_PPC64_TLSLD:
+      break;
+    default:
+      unreachable();
+    }
+  }
+}
+
+template <>
+void InputSection<E>::apply_reloc_nonalloc(Context<E> &ctx, u8 *base) {
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+    u8 *loc = base + rel.r_offset;
+
+    SectionFragment<E> *frag;
+    i64 frag_addend;
+    std::tie(frag, frag_addend) = get_fragment(ctx, rel);
+
+    u64 S = frag ? frag->get_addr(ctx) : sym.get_addr(ctx);
+    u64 A = frag ? frag_addend : (i64)rel.r_addend;
+
+    auto check = [&](i64 val, i64 lo, i64 hi) {
+      check_range(ctx, val, i, lo, hi);
+    };
+
+    switch (rel.r_type) {
+    case R_PPC64_ADDR64:
+      if (std::optional<u64> val = get_tombstone(sym, frag))
+        *(ub64 *)loc = *val;
+      else
+        *(ub64 *)loc = S + A;
+      break;
+    case R_PPC64_ADDR32:
+      check(S + A, 0, 1LL << 32);
+      *(ub32 *)loc = S + A;
+      break;
+    case R_PPC64_DTPREL64:
+      *(ub64 *)loc = S + A - ctx.dtp_addr;
+      break;
+    default:
+      Fatal(ctx) << *this << ": invalid relocation for non-allocated sections: "
+                 << rel;
+    }
+  }
+}
+
+template <>
+void InputSection<E>::scan_relocations(Context<E> &ctx) {
+  assert(shdr().sh_flags & SHF_ALLOC);
+  std::span<const ElfRel<E>> rels = get_rels(ctx);
+
+  // Scan relocations
+  for (i64 i = 0; i < rels.size(); i++) {
+    const ElfRel<E> &rel = rels[i];
+    if (rel.r_type == R_NONE || record_undef_error(ctx, rel))
+      continue;
+
+    Symbol<E> &sym = *file.symbols[rel.r_sym];
+
+    if (sym.is_ifunc())
+      sym.flags |= NEEDS_GOT | NEEDS_PLT | NEEDS_PPC_OPD;
+
+    // Any relocation except R_PPC64_REL24 is considered as an
+    // address-taking relocation.
+    if (rel.r_type != R_PPC64_REL24 && sym.get_type() == STT_FUNC)
+      sym.flags |= NEEDS_PPC_OPD;
+
+    switch (rel.r_type) {
+    case R_PPC64_GOT_TPREL16_HA:
+      sym.flags |= NEEDS_GOTTP;
+      break;
+    case R_PPC64_REL24:
+      if (sym.is_imported)
+        sym.flags |= NEEDS_PLT;
+      break;
+    case R_PPC64_PLT16_HA:
+      sym.flags |= NEEDS_GOT;
+      break;
+    case R_PPC64_GOT_TLSGD16_HA:
+      sym.flags |= NEEDS_TLSGD;
+      break;
+    case R_PPC64_GOT_TLSLD16_HA:
+      ctx.needs_tlsld = true;
+      break;
+    case R_PPC64_TPREL16_HA:
+    case R_PPC64_TPREL16_LO:
+    case R_PPC64_TPREL16_LO_DS:
+      check_tlsle(ctx, sym, rel);
+      break;
+    case R_PPC64_ADDR64:
+    case R_PPC64_TOC:
+    case R_PPC64_REL32:
+    case R_PPC64_REL64:
+    case R_PPC64_TOC16_HA:
+    case R_PPC64_TOC16_LO:
+    case R_PPC64_TOC16_LO_DS:
+    case R_PPC64_TOC16_DS:
+    case R_PPC64_REL16_HA:
+    case R_PPC64_REL16_LO:
+    case R_PPC64_PLT16_HI:
+    case R_PPC64_PLT16_LO:
+    case R_PPC64_PLT16_LO_DS:
+    case R_PPC64_PLTSEQ:
+    case R_PPC64_PLTCALL:
+    case R_PPC64_GOT_TPREL16_LO_DS:
+    case R_PPC64_GOT_TLSGD16_LO:
+    case R_PPC64_GOT_TLSLD16_LO:
+    case R_PPC64_TLS:
+    case R_PPC64_TLSGD:
+    case R_PPC64_TLSLD:
+    case R_PPC64_DTPREL16_HA:
+    case R_PPC64_DTPREL16_LO:
+    case R_PPC64_DTPREL16_LO_DS:
+      break;
+    default:
+      Error(ctx) << *this << ": unknown relocation: " << rel;
+    }
+  }
+}
+
+template <>
+void Thunk<E>::copy_buf(Context<E> &ctx) {
+  // If the destination is .plt.got, we save the current r2, read an
+  // address of a function descriptor from .got, restore %r2 and jump
+  // to the function.
+  constexpr ub32 pltgot_thunk[] = {
+    // Store the caller's %r2
+    0xf841'0028, // std   %r2, 40(%r1)
+
+    // Load an address of a function descriptor
+    0x3d82'0000, // addis %r12, %r2,  foo@got@toc@ha
+    0xe98c'0000, // ld    %r12, foo@got@toc@lo(%r12)
+
+    // Restore the callee's %r2
+    0xe84c'0008, // ld    %r2,  8(%r12)
+
+    // Jump to the function
+    0xe98c'0000, // ld    %r12, 0(%r12)
+    0x7d89'03a6, // mtctr %r12
+    0x4e80'0420, // bctr
+  };
+
+  // If the destination is .plt, read a function descriptor from .got.plt.
+  constexpr ub32 plt_thunk[] = {
+    // Store the caller's %r2
+    0xf841'0028, // std   %r2, 40(%r1)
+
+    // Materialize an address of a function descriptor
+    0x3d82'0000, // addis %r12, %r2,  foo@gotplt@toc@ha
+    0x398c'0000, // addi  %r12, %r12, foo@gotplt@toc@lo
+
+    // Restore the callee's %r2
+    0xe84c'0008, // ld    %r2,  8(%r12)
+
+    // Jump to the function
+    0xe98c'0000, // ld    %r12, 0(%r12)
+    0x7d89'03a6, // mtctr %r12
+    0x4e80'0420, // bctr
+  };
+
+  // If the destination is a non-imported function, we directly jump
+  // to the function entry address.
+  constexpr ub32 local_thunk[] = {
+    0x3d82'0000, // addis r12, r2,  foo@toc@ha
+    0x398c'0000, // addi  r12, r12, foo@toc@lo
+    0x7d89'03a6, // mtctr r12
+    0x4e80'0420, // bctr
+    0x6000'0000, // nop
+    0x6000'0000, // nop
+    0x6000'0000, // nop
+  };
+
+  static_assert(E::thunk_size == sizeof(pltgot_thunk));
+  static_assert(E::thunk_size == sizeof(plt_thunk));
+  static_assert(E::thunk_size == sizeof(local_thunk));
+
+  u8 *buf = ctx.buf + output_section.shdr.sh_offset + offset;
+
+  for (Symbol<E> *sym : symbols) {
+    if (sym->has_got(ctx)) {
+      i64 val = sym->get_got_addr(ctx) - ctx.extra.TOC->value;
+      memcpy(buf, pltgot_thunk, sizeof(pltgot_thunk));
+      *(ub32 *)(buf + 4) |= higha(val);
+      *(ub32 *)(buf + 8) |= lo(val);
+    } else if(sym->has_plt(ctx)) {
+      i64 val = sym->get_gotplt_addr(ctx) - ctx.extra.TOC->value;
+      memcpy(buf, plt_thunk, sizeof(plt_thunk));
+      *(ub32 *)(buf + 4) |= higha(val);
+      *(ub32 *)(buf + 8) |= lo(val);
+    } else {
+      i64 val = sym->get_addr(ctx, NO_OPD) - ctx.extra.TOC->value;
+      memcpy(buf, local_thunk, sizeof(local_thunk));
+      *(ub32 *)buf |= higha(val);
+      *(ub32 *)(buf + 4) |= lo(val);
+    }
+
+    buf += E::thunk_size;
+  }
+}
+
+static InputSection<E> *get_opd_section(ObjectFile<E> &file) {
+  for (std::unique_ptr<InputSection<E>> &isec : file.sections)
+    if (isec && isec->name() == ".opd")
+      return isec.get();
+  return nullptr;
+}
+
+static ElfRel<E> *
+get_relocation_at(Context<E> &ctx, InputSection<E> &isec, i64 offset) {
+  std::span<ElfRel<E>> rels = isec.get_rels(ctx);
+
+  auto it = std::lower_bound(rels.begin(), rels.end(), offset,
+                             [](const ElfRel<E> &r, i64 offset) {
+    return r.r_offset < offset;
+  });
+
+  if (it == rels.end())
+    return nullptr;
+  if (it->r_offset != offset)
+    return nullptr;
+  return &*it;
+}
+
+namespace {
+struct OpdSymbol {
+  u64 r_offset = 0;
+  Symbol<E> *sym = nullptr;
+};
+}
+
+static Symbol<E> *
+get_opd_sym_at(std::span<OpdSymbol> syms, u64 offset) {
+  auto it = ranges::lower_bound(syms, offset, {}, &OpdSymbol::r_offset);
+  if (it == syms.end())
+    return nullptr;
+  if (it->r_offset != offset)
+    return nullptr;
+  return it->sym;
+}
+
+// Compiler creates an .opd entry for each function symbol. The intention
+// is to make it possible to create an output .opd section just by linking
+// input .opd sections in the same manner as we do to other normal input
+// sections.
+//
+// However, in reality, .opd isn't a normal input section. It needs many
+// special treatments as follows:
+//
+// 1. A function symbol refers to not a .text but an .opd. Its address
+//    works fine for address-taking relocations such as R_PPC64_ADDR64.
+//    However, R_PPC64_REL24 (which is used for branch instruction) needs
+//    a function's real address instead of the function's .opd address.
+//    We need to read .opd contents to find out a function entry point
+//    address to apply R_PPC64_REL24.
+//
+// 2. Output .opd entries are needed only for functions whose addresses
+//    are taken. Just copying input .opd sections to an output would
+//    produces lots of dead .opd entries.
+//
+// 3. In this design, all function symbols refer to an .opd section, and
+//    that doesn't work well with graph traversal optimizations such as
+//    garbage collection or identical comdat folding. For example, garbage
+//    collector would mark an .opd alive which in turn mark all functions
+//    thatare referenced by .opd as alive, effectively keeping all
+//    functions as alive.
+//
+// The problem is that the compiler creates a half-baked .opd section, and
+// the linker has to figure out what all these .opd entries and
+// relocations are trying to achieve. It's like the compiler would emit a
+// half-baked .plt section in an object file and the linker has to deal
+// with that. That's not a good design.
+//
+// So, in this function, we undo what the compiler did to .opd. We remove
+// function symbols from .opd and reattach them to their function entry
+// points. We also rewrite relocations that directly refer to an input
+// .opd  section so that they refer to function symbols instead. We then
+// mark input .opd sections as dead.
+//
+// After this function, we mark symbols with the NEEDS_PPC_OPD flag if the
+// symbol needs an .opd entry. We then create an output .opd just like we
+// do for .plt or .got.
+void ppc64v1_rewrite_opd(Context<E> &ctx) {
+  tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
+    InputSection<E> *opd = get_opd_section(*file);
+    if (!opd)
+      return;
+    opd->is_alive = false;
+
+    // Move symbols from .opd to .text.
+    std::vector<OpdSymbol> opd_syms;
+
+    for (Symbol<E> *sym : file->symbols) {
+      if (sym->file != file || sym->get_input_section() != opd)
+        continue;
+
+      if (u32 ty = sym->get_type(); ty != STT_FUNC && ty != STT_GNU_IFUNC)
+        continue;
+
+      ElfRel<E> *rel = get_relocation_at(ctx, *opd, sym->value);
+      if (!rel)
+        Fatal(ctx) << *file << ": cannot find a relocation in .opd for "
+                   << *sym << " at offset 0x" << std::hex << (u64)sym->value;
+
+      Symbol<E> *sym2 = file->symbols[rel->r_sym];
+      if (sym2->get_type() != STT_SECTION)
+        Fatal(ctx) << *file << ": bad relocation in .opd referring to " << *sym2;
+
+      opd_syms.push_back({sym->value, sym});
+
+      sym->set_input_section(sym2->get_input_section());
+      sym->value = rel->r_addend;
+    }
+
+    // Sort symbols so that get_opd_sym_at() can do binary search.
+    ranges::stable_sort(opd_syms, {}, &OpdSymbol::r_offset);
+
+    // Rewrite relocations so that they directly refer to .opd.
+    for (std::unique_ptr<InputSection<E>> &isec : file->sections) {
+      if (!isec || !isec->is_alive || isec.get() == opd)
+        continue;
+
+      for (ElfRel<E> &r : isec->get_rels(ctx)) {
+        Symbol<E> &sym = *file->symbols[r.r_sym];
+        if (sym.get_input_section() != opd)
+          continue;
+
+        Symbol<E> *real_sym = get_opd_sym_at(opd_syms, r.r_addend);
+        if (!real_sym)
+          Fatal(ctx) << *isec << ": cannot find a symbol in .opd for " << r
+                     << " at offset 0x" << std::hex << (u64)r.r_addend;
+
+        r.r_sym = real_sym->sym_idx;
+        r.r_addend = 0;
+      }
+    }
+  });
+}
+
+// When a function is exported, the dynamic symbol for the function should
+// refers to the function's .opd entry. This function marks such symbols
+// with NEEDS_PPC_OPD.
+void ppc64v1_scan_symbols(Context<E> &ctx) {
+  tbb::parallel_for_each(ctx.objs, [&](ObjectFile<E> *file) {
+    for (Symbol<E> *sym : file->symbols)
+      if (sym->file == file && sym->is_exported)
+        if (u32 ty = sym->get_type(); ty == STT_FUNC || ty == STT_GNU_IFUNC)
+          sym->flags |= NEEDS_PPC_OPD;
+  });
+
+  // Functions referenced by the ELF header also have to have .opd entries.
+  if (!ctx.arg.entry->is_imported)
+    ctx.arg.entry->flags |= NEEDS_PPC_OPD;
+
+  if (!ctx.arg.init->is_imported)
+    ctx.arg.init->flags |= NEEDS_PPC_OPD;
+
+  if (!ctx.arg.fini->is_imported)
+    ctx.arg.fini->flags |= NEEDS_PPC_OPD;
+}
+
+void PPC64OpdSection::add_symbol(Context<E> &ctx, Symbol<E> *sym) {
+  sym->set_opd_idx(ctx, symbols.size());
+  symbols.push_back(sym);
+  this->shdr.sh_size += ENTRY_SIZE;
+}
+
+i64 PPC64OpdSection::get_reldyn_size(Context<E> &ctx) const {
+  if (ctx.arg.pic)
+    return symbols.size() * 2;
+  return 0;
+}
+
+void PPC64OpdSection::copy_buf(Context<E> &ctx) {
+  ub64 *buf = (ub64 *)(ctx.buf + this->shdr.sh_offset);
+
+  ElfRel<E> *rel = nullptr;
+  if (ctx.arg.pic)
+    rel = (ElfRel<E> *)(ctx.buf + ctx.reldyn->shdr.sh_offset + reldyn_offset);
+
+  for (Symbol<E> *sym : symbols) {
+    u64 addr = sym->get_addr(ctx, NO_PLT | NO_OPD);
+    *buf++ = addr;
+    *buf++ = ctx.extra.TOC->value;
+    *buf++ = 0;
+
+    if (ctx.arg.pic) {
+      u64 loc = sym->get_opd_addr(ctx);
+      *rel++ = ElfRel<E>(loc, E::R_RELATIVE, 0, addr);
+      *rel++ = ElfRel<E>(loc + 8, E::R_RELATIVE, 0, ctx.extra.TOC->value);
+    }
+  }
+}
+
+} // namespace mold