Merge pull request #9441 from skylersaleh/master

Apple M1 Support for MacOS

BuildMacOSUniversalBinary.py

@@ -0,0 +1,330 @@
+#!/usr/bin/env python3
+"""
+The current tooling supported in CMake, Homebrew, and Qt5 are insufficient for
+creating macOS universal binaries automatically for applications like Dolphin
+which have more complicated build requirements (like different libraries, build
+flags and source files for each target architecture).
+
+So instead, this script manages the configuration and compilation of distinct
+builds and project files for each target architecture and then merges the two
+binaries into a single universal binary.
+
+Running this script will:
+1) Generate Xcode project files for the ARM build (if project files don't
+   already exist)
+2) Generate Xcode project files for the x64 build (if project files don't
+   already exist)
+3) Build the ARM project for the selected build_target
+4) Build the x64 project for the selected build_target
+5) Generate universal .app packages combining the ARM and x64 packages
+6) Use the lipo tool to combine the binary objects inside each of the
+   packages into universal binaries
+7) Code sign the final universal binaries using the specified
+   codesign_identity
+"""
+
+import argparse
+import filecmp
+import glob
+import json
+import multiprocessing
+import os
+import shutil
+import subprocess
+
+# The config variables listed below are the defaults, but they can be
+# overridden by command line arguments see parse_args(), or run:
+# BuildMacOSUniversalBinary.py --help
+DEFAULT_CONFIG = {
+
+    # Location of destination universal binary
+    "dst_app": "universal/",
+    # Build Target (dolphin-emu to just build the emulator and skip the tests)
+    "build_target": "ALL_BUILD",
+
+    # Location for CMake to search for files (default is for homebrew)
+    "arm64_cmake_prefix":  "/opt/homebrew",
+    "x86_64_cmake_prefix": "/usr/local",
+
+    # Locations to qt5 directories for arm and x64 libraries
+    # The default values of these paths are taken from the default
+    # paths used for homebrew
+    "arm64_qt5_path":  "/opt/homebrew/opt/qt5",
+    "x86_64_qt5_path": "/usr/local/opt/qt5",
+
+    # Identity to use for code signing. "-" indicates that the app will not
+    # be cryptographically signed/notarized but will instead just use a
+    # SHA checksum to verify the integrity of the app. This doesn't
+    # protect against malicious actors, but it does protect against
+    # running corrupted binaries and allows for access to the extended
+    # permisions needed for ARM builds
+    "codesign_identity":  "-",
+    # Entitlements file to use for code signing
+    "entitlements": "../Source/Core/DolphinQt/DolphinEmu.entitlements",
+
+    # Minimum macOS version for each architecture slice
+    "arm64_mac_os_deployment_target":  "11.0.0",
+    "x86_64_mac_os_deployment_target": "10.12.0",
+
+    # CMake Generator to use for building
+    "generator": "Unix Makefiles",
+    "build_type": "Release",
+
+}
+
+# Architectures to build for. This is explicity left out of the command line
+# config options for several reasons:
+# 1) Adding new architectures will generally require more code changes
+# 2) Single architecture builds should utilize the normal generated cmake
+#    project files rather than this wrapper script
+
+ARCHITECTURES = ["x86_64", "arm64"]
+
+
+def parse_args(conf=DEFAULT_CONFIG):
+    """
+    Parses the command line arguments into a config dictionary.
+    """
+
+    parser = argparse.ArgumentParser(
+        formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+
+    parser.add_argument(
+        "--target",
+        help="Build target in generated project files",
+        default=conf["build_target"],
+        dest="build_target")
+    parser.add_argument(
+        "-G",
+        help="CMake Generator to use for creating project files",
+        default=conf["generator"],
+        dest="generator")
+    parser.add_argument(
+        "--build_type",
+        help="CMake build type [Debug, Release, RelWithDebInfo, MinSizeRel]",
+        default=conf["build_type"],
+        dest="build_type")
+    parser.add_argument(
+        "--dst_app",
+        help="Directory where universal binary will be stored",
+        default=conf["dst_app"])
+
+    parser.add_argument(
+        "--entitlements",
+        help="Path to .entitlements file for code signing",
+        default=conf["entitlements"])
+
+    parser.add_argument(
+        "--codesign",
+        help="Code signing identity to use to sign the applications",
+        default=conf["codesign_identity"],
+        dest="codesign_identity")
+
+    for arch in ARCHITECTURES:
+        parser.add_argument(
+             f"--{arch}_cmake_prefix",
+             help="Folder for cmake to search for packages",
+             default=conf[arch+"_cmake_prefix"],
+             dest=arch+"_cmake_prefix")
+
+        parser.add_argument(
+             f"--{arch}_qt5_path",
+             help=f"Install path for {arch} qt5 libraries",
+             default=conf[arch+"_qt5_path"])
+
+        parser.add_argument(
+             f"--{arch}_mac_os_deployment_target",
+             help=f"Deployment architecture for {arch} slice",
+             default=conf[arch+"_mac_os_deployment_target"])
+
+    return vars(parser.parse_args())
+
+
+def lipo(path0, path1, dst):
+    if subprocess.call(["lipo", "-create", "-output", dst, path0, path1]) != 0:
+        print(f"WARNING: {path0} and {path1} cannot be lipo'd")
+
+        shutil.copy(path0, dst)
+
+
+def recursive_merge_binaries(src0, src1, dst):
+    """
+    Merges two build trees together for different architectures into a single
+    universal binary.
+
+    The rules for merging are:
+
+    1) Files that exist in either src tree are copied into the dst tree
+    2) Files that exist in both trees and are identical are copied over
+       unmodified
+    3) Files that exist in both trees and are non-identical are lipo'd
+    4) Symlinks are created in the destination tree to mirror the hierarchy in
+       the source trees
+    """
+
+    # Check that all files present in the folder are of the same type and that
+    # links link to the same relative location
+    for newpath0 in glob.glob(src0+"/*"):
+        filename = os.path.basename(newpath0)
+        newpath1 = os.path.join(src1, filename)
+        if not os.path.exists(newpath1):
+            continue
+
+        if os.path.islink(newpath0) and os.path.islink(newpath1):
+            if os.path.relpath(newpath0, src0) == os.path.relpath(newpath1, src1):
+                continue
+
+        if os.path.isdir(newpath0) and os.path.isdir(newpath1):
+            continue
+
+        # isfile() can be true for links so check that both are not links
+        # before checking if they are both files
+        if (not os.path.islink(newpath0)) and (not os.path.islink(newpath1)):
+            if os.path.isfile(newpath0) and os.path.isfile(newpath1):
+                continue
+
+        raise Exception(f"{newpath0} and {newpath1} cannot be " +
+                        "merged into a universal binary because they are of " +
+                        "incompatible types. Perhaps the installed libraries" +
+                        " are from different versions for each architecture")
+
+    for newpath0 in glob.glob(src0+"/*"):
+        filename = os.path.basename(newpath0)
+        newpath1 = os.path.join(src1, filename)
+        new_dst_path = os.path.join(dst, filename)
+        if os.path.islink(newpath0):
+            # Symlinks will be fixed after files are resolved
+            continue
+
+        if not os.path.exists(newpath1):
+            if os.path.isdir(newpath0):
+                shutil.copytree(newpath0, new_dst_path)
+            else:
+                shutil.copy(newpath0, new_dst_path)
+
+            continue
+
+        if os.path.isdir(newpath1):
+            os.mkdir(new_dst_path)
+            recursive_merge_binaries(newpath0, newpath1, new_dst_path)
+            continue
+
+        if filecmp.cmp(newpath0, newpath1):
+            shutil.copy(newpath0, new_dst_path)
+        else:
+            lipo(newpath0, newpath1, new_dst_path)
+
+    # Loop over files in src1 and copy missing things over to dst
+    for newpath1 in glob.glob(src1+"/*"):
+        filename = os.path.basename(newpath1)
+        newpath0 = os.path.join(src0, filename)
+        new_dst_path = os.path.join(dst, filename)
+        if (not os.path.exists(newpath0)) and (not os.path.islink(newpath1)):
+            if os.path.isdir(newpath1):
+                shutil.copytree(newpath1, new_dst_path)
+            else:
+                shutil.copy(newpath1, new_dst_path)
+
+    # Fix up symlinks for path0
+    for newpath0 in glob.glob(src0+"/*"):
+        filename = os.path.basename(newpath0)
+        new_dst_path = os.path.join(dst, filename)
+        if os.path.islink(newpath0):
+            relative_path = os.path.relpath(os.path.realpath(newpath0), src0)
+            os.symlink(relative_path, new_dst_path)
+    # Fix up symlinks for path1
+    for newpath1 in glob.glob(src1+"/*"):
+        filename = os.path.basename(newpath1)
+        new_dst_path = os.path.join(dst, filename)
+        newpath0 = os.path.join(src0, filename)
+        if os.path.islink(newpath1) and not os.path.exists(newpath0):
+            relative_path = os.path.relpath(os.path.realpath(newpath1), src1)
+            os.symlink(relative_path, new_dst_path)
+
+
+def build(config):
+    """
+    Builds the project with the parameters specified in config.
+    """
+
+    print("Building config:")
+    print(json.dumps(config, indent=4))
+
+    # Configure and build single architecture builds for each architecture
+    for arch in ARCHITECTURES:
+        if not os.path.exists(arch):
+            os.mkdir(arch)
+
+        env = os.environ.copy()
+        env["Qt5_DIR"] = config[arch+"_qt5_path"]
+        env["CMAKE_OSX_ARCHITECTURES"] = arch
+        env["CMAKE_PREFIX_PATH"] = config[arch+"_cmake_prefix"]
+
+        # Add the other architecture's prefix path to the ignore path so that
+        # CMake doesn't try to pick up the wrong architecture's libraries when
+        # cross compiling.
+        ignore_path = ""
+        for a in ARCHITECTURES:
+            if a != arch:
+                ignore_path = config[a+"_cmake_prefix"]
+
+        subprocess.check_call([
+                "cmake", "../../", "-G", config["generator"],
+                "-DCMAKE_BUILD_TYPE=" + config["build_type"],
+                '-DCMAKE_CXX_FLAGS="-DMACOS_UNIVERSAL_BUILD=1"',
+                '-DCMAKE_C_FLAGS="-DMACOS_UNIVERSAL_BUILD=1"',
+                # System name needs to be specified for CMake to use
+                # the specified CMAKE_SYSTEM_PROCESSOR
+                "-DCMAKE_SYSTEM_NAME=Darwin",
+                "-DCMAKE_PREFIX_PATH="+config[arch+"_cmake_prefix"],
+                "-DCMAKE_SYSTEM_PROCESSOR="+arch,
+                "-DCMAKE_IGNORE_PATH="+ignore_path,
+                "-DCMAKE_OSX_DEPLOYMENT_TARGET="
+                + config[arch+"_mac_os_deployment_target"],
+                "-DMACOS_CODE_SIGNING_IDENTITY="
+                + config["codesign_identity"],
+                "-DMACOS_CODE_SIGNING_IDENTITY_UPDATER="
+                + config["codesign_identity"],
+                '-DMACOS_CODE_SIGNING="ON"'
+            ],
+            env=env, cwd=arch)
+
+        threads = multiprocessing.cpu_count()
+        subprocess.check_call(["cmake", "--build", ".",
+                               "--config", config["build_type"],
+                               "--parallel", f"{threads}"], cwd=arch)
+
+    dst_app = config["dst_app"]
+
+    if os.path.exists(dst_app):
+        shutil.rmtree(dst_app)
+
+    # Create and codesign the universal binary/
+    os.mkdir(dst_app)
+
+    # Source binary trees to merge together
+    src_app0 = ARCHITECTURES[0]+"/Binaries/"
+    src_app1 = ARCHITECTURES[1]+"/Binaries/"
+
+    recursive_merge_binaries(src_app0, src_app1, dst_app)
+    for path in glob.glob(dst_app+"/*"):
+        if os.path.isdir(path) and os.path.splitext(path)[1] != ".app":
+            continue
+
+        subprocess.check_call([
+            "codesign",
+            "-d",
+            "--force",
+            "-s",
+            config["codesign_identity"],
+            "--options=runtime",
+            "--entitlements", config["entitlements"],
+            "--deep",
+            "--verbose=2",
+            path])
+
+
+if __name__ == "__main__":
+    conf = parse_args()
+    build(conf)
+    print("Built Universal Binary successfully!")

CMakeLists.txt

@@ -2,13 +2,17 @@
 # General setup
 #
 cmake_minimum_required(VERSION 3.10)
-set(CMAKE_OSX_ARCHITECTURES "x86_64")
+
 # Minimum OS X version.
 # This is inserted into the Info.plist as well.
 
-# MacOS prior to 10.12 did not fully support C++17, which is used to
-# handle configuration options
-set(CMAKE_OSX_DEPLOYMENT_TARGET "10.12.0" CACHE STRING "")
+# MacOS prior to 10.14 did not support aligned alloc which is used to implement
+# std::unique_ptr in the arm64 C++ standard library. x86_64 builds can override
+# this to 10.12.0 using -DCMAKE_OSX_DEPLOYMENT_TARGET="10.12.0" without issue.
+# This is done in the universal binary building script to build a binary that 
+# runs on 10.12 on x86_64 computers, while still containing an arm64 slice.
+
+set(CMAKE_OSX_DEPLOYMENT_TARGET "10.14.0" CACHE STRING "")
 
 set(CMAKE_USER_MAKE_RULES_OVERRIDE "CMake/FlagsOverride.cmake")
 
@@ -68,8 +72,12 @@ else()
 endif()
 
 if(APPLE)
-  option(OSX_USE_DEFAULT_SEARCH_PATH "Don't prioritize system library paths" OFF)
-  option(SKIP_POSTPROCESS_BUNDLE "Skip postprocessing bundle for redistributability" OFF)
+  option(MACOS_USE_DEFAULT_SEARCH_PATH "Don't prioritize system library paths" OFF)
+  option(SKIP_POSTPROCESS_BUNDLE "Skip postprocessing bundle for redistributability" OFF) 
+  # Enable adhoc code signing by default (otherwise makefile builds on ARM will not work)
+  option(MACOS_CODE_SIGNING "Enable codesigning" ON)
+  set(MACOS_CODE_SIGNING_IDENTITY "-" CACHE STRING "The identity used for codesigning.")
+  set(MACOS_CODE_SIGNING_IDENTITY_UPDATER "-" CACHE STRING "The identity used for codesigning, for the updater.")
 endif()
 
 if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
@@ -290,7 +298,7 @@ else()
 endif()
 
 if(CMAKE_SYSTEM_NAME MATCHES "Darwin")
-  if(NOT OSX_USE_DEFAULT_SEARCH_PATH)
+  if(NOT MACOS_USE_DEFAULT_SEARCH_PATH)
     # Hack up the path to prioritize the path to built-in OS libraries to
     # increase the chance of not depending on a bunch of copies of them
     # installed by MacPorts, Fink, Homebrew, etc, and ending up copying
@@ -306,9 +314,10 @@ if(CMAKE_SYSTEM_NAME MATCHES "Darwin")
   set(CMAKE_XCODE_ATTRIBUTE_GCC_STRICT_ALIASING NO)
 
   # Specify target CPUs.
-  check_and_add_flag(HAVE_MSSSE3 -mssse3)
-  check_and_add_flag(HAVE_ARCH_CORE2 -march=core2)
-
+  if(_ARCH_64 AND CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64|amd64|AMD64")
+    check_and_add_flag(HAVE_MSSSE3 -mssse3)
+    check_and_add_flag(HAVE_ARCH_CORE2 -march=core2)
+  endif()
   # Linker flags.
   # Drop unreachable code and data.
   set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -Wl,-dead_strip,-dead_strip_dylibs")

Externals/MoltenVK/version.txt

@@ -1 +1 @@
-MoltenVK from https://github.com/KhronosGroup/MoltenVK, commit b9b78def172074872bfbb1015ccf75eeec554ae2
+MoltenVK from https://vulkan.lunarg.com/sdk/home#mac, version 1.2.170.0

Readme.md

@@ -65,6 +65,8 @@ missing packages yourself.
 
 ### macOS Build Steps:
 
+A binary supporting a single architecture can be built using the following steps: 
+
 1. `mkdir build`
 2. `cd build`
 3. `cmake ..`
@@ -72,6 +74,18 @@ missing packages yourself.
 
 An application bundle will be created in `./Binaries`.
 
+A script is also provided to build universal binaries supporting both x64 and ARM in the same
+application bundle using the following steps:
+
+1. `mkdir build`
+2. `cd build`
+3. `python ../BuildMacOSUniversalBinary.py`
+4. Universal binaries will be available in the `universal` folder
+
+Doing this is more complex as it requires installation of library dependencies for both x64 and ARM (or universal library
+equivalents) and may require specifying additional arguments to point to relevant library locations. 
+Execute BuildMacOSUniversalBinary.py --help for more details.  
+
 ### Linux Global Build Steps:
 
 To install to your system.

Source/Core/Common/Arm64Emitter.cpp

@@ -21,6 +21,9 @@
 #ifdef _WIN32
 #include <Windows.h>
 #endif
+#ifdef __APPLE__
+#include <libkern/OSCacheControl.h>
+#endif
 
 namespace Arm64Gen
 {
@@ -342,7 +345,7 @@ void ARM64XEmitter::FlushIcacheSection(u8* start, u8* end)
   if (start == end)
     return;
 
-#if defined(IOS)
+#if defined(IOS) || defined(__APPLE__)
   // Header file says this is equivalent to: sys_icache_invalidate(start, end - start);
   sys_cache_control(kCacheFunctionPrepareForExecution, start, end - start);
 #elif defined(WIN32)

Source/Core/Common/ArmCPUDetect.cpp

@@ -8,7 +8,7 @@
 #include <string>
 #include <thread>
 
-#ifndef _WIN32
+#if !defined(_WIN32) && !defined(__APPLE__)
 #ifndef __FreeBSD__
 #include <asm/hwcap.h>
 #endif
@@ -71,7 +71,17 @@ void CPUInfo::Detect()
   vendor = CPUVendor::ARM;
   bFlushToZero = true;
 
-#ifdef _WIN32
+#ifdef __APPLE__
+  num_cores = std::thread::hardware_concurrency();
+
+  // M-series CPUs have all of these
+  bFP = true;
+  bASIMD = true;
+  bAES = true;
+  bSHA1 = true;
+  bSHA2 = true;
+  bCRC32 = true;
+#elif defined(_WIN32)
   num_cores = std::thread::hardware_concurrency();
 
   // Windows does not provide any mechanism for querying the system registers on ARMv8, unlike Linux

Source/Core/Common/MemoryUtil.cpp

@@ -16,6 +16,7 @@
 #include <windows.h>
 #include "Common/StringUtil.h"
 #else
+#include <pthread.h>
 #include <stdio.h>
 #include <sys/mman.h>
 #include <sys/types.h>
@@ -38,9 +39,15 @@ void* AllocateExecutableMemory(size_t size)
 #if defined(_WIN32)
   void* ptr = VirtualAlloc(nullptr, size, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
 #else
-  void* ptr =
-      mmap(nullptr, size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANON | MAP_PRIVATE, -1, 0);
-
+  int map_flags = MAP_ANON | MAP_PRIVATE;
+#if defined(__APPLE__)
+  // This check is in place to prepare for x86_64 MAP_JIT support. While MAP_JIT did exist
+  // prior to 10.14, it had restrictions on the number of JIT allocations that were removed
+  // in 10.14.
+  if (__builtin_available(macOS 10.14, *))
+    map_flags |= MAP_JIT;
+#endif
+  void* ptr = mmap(nullptr, size, PROT_READ | PROT_WRITE | PROT_EXEC, map_flags, -1, 0);
   if (ptr == MAP_FAILED)
     ptr = nullptr;
 #endif
@@ -50,6 +57,79 @@ void* AllocateExecutableMemory(size_t size)
 
   return ptr;
 }
+// This function is used to provide a counter for the JITPageWrite*Execute*
+// functions to enable nesting. The static variable is wrapped in a a function
+// to allow those functions to be called inside of the constructor of a static
+// variable portably.
+//
+// The variable is thread_local as the W^X mode is specific to each running thread.
+static int& JITPageWriteNestCounter()
+{
+  static thread_local int nest_counter = 0;
+  return nest_counter;
+}
+
+// Certain platforms (Mac OS on ARM) enforce that a single thread can only have write or
+// execute permissions to pages at any given point of time. The two below functions
+// are used to toggle between having write permissions or execute permissions.
+//
+// The default state of these allocations in Dolphin is for them to be executable,
+// but not writeable. So, functions that are updating these pages should wrap their
+// writes like below:
+
+// JITPageWriteEnableExecuteDisable();
+// PrepareInstructionStreamForJIT();
+// JITPageWriteDisableExecuteEnable();
+
+// These functions can be nested, in which case execution will only be enabled
+// after the call to the JITPageWriteDisableExecuteEnable from the top most
+// nesting level. Example:
+
+// [JIT page is in execute mode for the thread]
+// JITPageWriteEnableExecuteDisable();
+//   [JIT page is in write mode for the thread]
+//   JITPageWriteEnableExecuteDisable();
+//     [JIT page is in write mode for the thread]
+//   JITPageWriteDisableExecuteEnable();
+//   [JIT page is in write mode for the thread]
+// JITPageWriteDisableExecuteEnable();
+// [JIT page is in execute mode for the thread]
+
+// Allows a thread to write to executable memory, but not execute the data.
+void JITPageWriteEnableExecuteDisable()
+{
+#if defined(_M_ARM_64) && defined(__APPLE__)
+  if (JITPageWriteNestCounter() == 0)
+  {
+    if (__builtin_available(macOS 11.0, *))
+    {
+      pthread_jit_write_protect_np(0);
+    }
+  }
+#endif
+  JITPageWriteNestCounter()++;
+}
+// Allows a thread to execute memory allocated for execution, but not write to it.
+void JITPageWriteDisableExecuteEnable()
+{
+  JITPageWriteNestCounter()--;
+
+  // Sanity check the NestCounter to identify underflow
+  // This can indicate the calls to JITPageWriteDisableExecuteEnable()
+  // are not matched with previous calls to JITPageWriteEnableExecuteDisable()
+  if (JITPageWriteNestCounter() < 0)
+    PanicAlertFmt("JITPageWriteNestCounter() underflowed");
+
+#if defined(_M_ARM_64) && defined(__APPLE__)
+  if (JITPageWriteNestCounter() == 0)
+  {
+    if (__builtin_available(macOS 11.0, *))
+    {
+      pthread_jit_write_protect_np(1);
+    }
+  }
+#endif
+}
 
 void* AllocateMemoryPages(size_t size)
 {
@@ -128,7 +208,10 @@ void WriteProtectMemory(void* ptr, size_t size, bool allowExecute)
   DWORD oldValue;
   if (!VirtualProtect(ptr, size, allowExecute ? PAGE_EXECUTE_READ : PAGE_READONLY, &oldValue))
     PanicAlertFmt("WriteProtectMemory failed!\nVirtualProtect: {}", GetLastErrorString());
-#else
+#elif !(defined(_M_ARM_64) && defined(__APPLE__))
+  // MacOS 11.2 on ARM does not allow for changing the access permissions of pages
+  // that were marked executable, instead it uses the protections offered by MAP_JIT
+  // for write protection.
   if (mprotect(ptr, size, allowExecute ? (PROT_READ | PROT_EXEC) : PROT_READ) != 0)
     PanicAlertFmt("WriteProtectMemory failed!\nmprotect: {}", LastStrerrorString());
 #endif
@@ -140,7 +223,10 @@ void UnWriteProtectMemory(void* ptr, size_t size, bool allowExecute)
   DWORD oldValue;
   if (!VirtualProtect(ptr, size, allowExecute ? PAGE_EXECUTE_READWRITE : PAGE_READWRITE, &oldValue))
     PanicAlertFmt("UnWriteProtectMemory failed!\nVirtualProtect: {}", GetLastErrorString());
-#else
+#elif !(defined(_M_ARM_64) && defined(__APPLE__))
+  // MacOS 11.2 on ARM does not allow for changing the access permissions of pages
+  // that were marked executable, instead it uses the protections offered by MAP_JIT
+  // for write protection.
   if (mprotect(ptr, size,
                allowExecute ? (PROT_READ | PROT_WRITE | PROT_EXEC) : PROT_WRITE | PROT_READ) != 0)
   {

Source/Core/Common/MemoryUtil.h

@@ -10,6 +10,23 @@
 namespace Common
 {
 void* AllocateExecutableMemory(size_t size);
+
+// These two functions control the executable/writable state of the W^X memory
+// allocations. More detailed documentation about them is in the .cpp file.
+// In general where applicable the ScopedJITPageWriteAndNoExecute wrapper
+// should be used to prevent bugs from not pairing up the calls properly.
+
+// Allows a thread to write to executable memory, but not execute the data.
+void JITPageWriteEnableExecuteDisable();
+// Allows a thread to execute memory allocated for execution, but not write to it.
+void JITPageWriteDisableExecuteEnable();
+// RAII Wrapper around JITPageWrite*Execute*(). When this is in scope the thread can
+// write to executable memory but not execute it.
+struct ScopedJITPageWriteAndNoExecute
+{
+  ScopedJITPageWriteAndNoExecute() { JITPageWriteEnableExecuteDisable(); }
+  ~ScopedJITPageWriteAndNoExecute() { JITPageWriteDisableExecuteEnable(); }
+};
 void* AllocateMemoryPages(size_t size);
 void FreeMemoryPages(void* ptr, size_t size);
 void* AllocateAlignedMemory(size_t size, size_t alignment);

Source/Core/Core/DolphinAnalytics.cpp

@@ -289,11 +289,17 @@ void DolphinAnalytics::MakeBaseBuilder()
       s64 minor_version;  // NSInteger minorVersion
       s64 patch_version;  // NSInteger patchVersion
     };
-
+    // Under arm64, we need to call objc_msgSend to recieve a struct.
+    // On x86_64, we need to explicitly call objc_msgSend_stret for a struct.
+#if _M_ARM_64
+#define msgSend objc_msgSend
+#else
+#define msgSend objc_msgSend_stret
+#endif
     // NSOperatingSystemVersion version = [processInfo operatingSystemVersion]
-    OSVersion version = reinterpret_cast<OSVersion (*)(id, SEL)>(objc_msgSend_stret)(
+    OSVersion version = reinterpret_cast<OSVersion (*)(id, SEL)>(msgSend)(
         processInfo, sel_getUid("operatingSystemVersion"));
-
+#undef msgSend
     builder.AddData("osx-ver-major", version.major_version);
     builder.AddData("osx-ver-minor", version.minor_version);
     builder.AddData("osx-ver-bugfix", version.patch_version);

Source/Core/Core/MachineContext.h

@@ -67,6 +67,12 @@ typedef x86_thread_state64_t SContext;
 #define CTX_R14 __r14
 #define CTX_R15 __r15
 #define CTX_RIP __rip
+#elif _M_ARM_64
+typedef arm_thread_state64_t SContext;
+#define CTX_REG(x) __x[x]
+#define CTX_LR __x[30]
+#define CTX_SP __sp
+#define CTX_PC __pc
 #else
 #error No context definition for architecture
 #endif

Source/Core/Core/MemTools.cpp

@@ -25,6 +25,20 @@
 #include <unistd.h>  // Needed for _POSIX_VERSION
 #endif
 
+#if defined(__APPLE__)
+#ifdef _M_X86_64
+#define THREAD_STATE64_COUNT x86_THREAD_STATE64_COUNT
+#define THREAD_STATE64 x86_THREAD_STATE64
+#define thread_state64_t x86_thread_state64_t
+#elif defined(_M_ARM_64)
+#define THREAD_STATE64_COUNT ARM_THREAD_STATE64_COUNT
+#define THREAD_STATE64 ARM_THREAD_STATE64
+#define thread_state64_t arm_thread_state64_t
+#else
+#error Unsupported architecture
+#endif
+#endif
+
 namespace EMM
 {
 #ifdef _WIN32
@@ -123,7 +137,7 @@ static void ExceptionThread(mach_port_t port)
     int64_t code[2];
     int flavor;
     mach_msg_type_number_t old_stateCnt;
-    natural_t old_state[x86_THREAD_STATE64_COUNT];
+    natural_t old_state[THREAD_STATE64_COUNT];
     mach_msg_trailer_t trailer;
   } msg_in;
 
@@ -134,7 +148,7 @@ static void ExceptionThread(mach_port_t port)
     kern_return_t RetCode;
     int flavor;
     mach_msg_type_number_t new_stateCnt;
-    natural_t new_state[x86_THREAD_STATE64_COUNT];
+    natural_t new_state[THREAD_STATE64_COUNT];
   } msg_out;
 #pragma pack()
   memset(&msg_in, 0xee, sizeof(msg_in));
@@ -165,13 +179,13 @@ static void ExceptionThread(mach_port_t port)
       return;
     }
 
-    if (msg_in.flavor != x86_THREAD_STATE64)
+    if (msg_in.flavor != THREAD_STATE64)
     {
-      PanicAlertFmt("unknown flavor {} (expected {})", msg_in.flavor, x86_THREAD_STATE64);
+      PanicAlertFmt("unknown flavor {} (expected {})", msg_in.flavor, THREAD_STATE64);
       return;
     }
 
-    x86_thread_state64_t* state = (x86_thread_state64_t*)msg_in.old_state;
+    thread_state64_t* state = (thread_state64_t*)msg_in.old_state;
 
     bool ok = JitInterface::HandleFault((uintptr_t)msg_in.code[1], state);
 
@@ -184,9 +198,9 @@ static void ExceptionThread(mach_port_t port)
     if (ok)
     {
       msg_out.RetCode = KERN_SUCCESS;
-      msg_out.flavor = x86_THREAD_STATE64;
-      msg_out.new_stateCnt = x86_THREAD_STATE64_COUNT;
-      memcpy(msg_out.new_state, msg_in.old_state, x86_THREAD_STATE64_COUNT * sizeof(natural_t));
+      msg_out.flavor = THREAD_STATE64;
+      msg_out.new_stateCnt = THREAD_STATE64_COUNT;
+      memcpy(msg_out.new_state, msg_in.old_state, THREAD_STATE64_COUNT * sizeof(natural_t));
     }
     else
     {
@@ -218,7 +232,7 @@ void InstallExceptionHandler()
   // Debuggers set the task port, so we grab the thread port.
   CheckKR("thread_set_exception_ports",
           thread_set_exception_ports(mach_thread_self(), EXC_MASK_BAD_ACCESS, port,
-                                     EXCEPTION_STATE | MACH_EXCEPTION_CODES, x86_THREAD_STATE64));
+                                     EXCEPTION_STATE | MACH_EXCEPTION_CODES, THREAD_STATE64));
   // ...and get rid of our copy so that MACH_NOTIFY_NO_SENDERS works.
   CheckKR("mach_port_mod_refs",
           mach_port_mod_refs(mach_task_self(), port, MACH_PORT_RIGHT_SEND, -1));

Source/Core/Core/PowerPC/JitArm64/Jit.cpp

@@ -73,6 +73,8 @@ void JitArm64::Init()
 
 bool JitArm64::HandleFault(uintptr_t access_address, SContext* ctx)
 {
+  // Ifdef this since the exception handler runs on a separate thread on macOS (ARM)
+#if !(defined(__APPLE__) && defined(_M_ARM_64))
   // We can't handle any fault from other threads.
   if (!Core::IsCPUThread())
   {
@@ -80,6 +82,7 @@ bool JitArm64::HandleFault(uintptr_t access_address, SContext* ctx)
     DoBacktrace(access_address, ctx);
     return false;
   }
+#endif
 
   bool success = false;
 
@@ -124,6 +127,7 @@ void JitArm64::ClearCache()
   m_handler_to_loc.clear();
 
   blocks.Clear();
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
   ClearCodeSpace();
   farcode.ClearCodeSpace();
   UpdateMemoryOptions();
@@ -596,6 +600,7 @@ void JitArm64::Jit(u32)
   {
     ClearCache();
   }
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
 
   std::size_t block_size = m_code_buffer.size();
   const u32 em_address = PowerPC::ppcState.pc;

Source/Core/Core/PowerPC/JitArm64/JitArm64Cache.cpp

@@ -59,11 +59,11 @@ void JitArm64BlockCache::WriteLinkBlock(Arm64Gen::ARM64XEmitter& emit,
 
 void JitArm64BlockCache::WriteLinkBlock(const JitBlock::LinkData& source, const JitBlock* dest)
 {
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
   u8* location = source.exitPtrs;
   ARM64XEmitter emit(location);
 
   WriteLinkBlock(emit, source, dest);
-
   emit.FlushIcache();
 }
 
@@ -71,9 +71,8 @@ void JitArm64BlockCache::WriteDestroyBlock(const JitBlock& block)
 {
   // Only clear the entry points as we might still be within this block.
   ARM64XEmitter emit(block.checkedEntry);
-
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
   while (emit.GetWritableCodePtr() <= block.normalEntry)
     emit.BRK(0x123);
-
   emit.FlushIcache();
 }

Source/Core/Core/PowerPC/JitArm64/JitArm64_BackPatch.cpp

@@ -289,6 +289,7 @@ bool JitArm64::HandleFastmemFault(uintptr_t access_address, SContext* ctx)
   if ((const u8*)ctx->CTX_PC - fault_location > fastmem_area_length)
     return false;
 
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
   ARM64XEmitter emitter((u8*)fault_location);
 
   emitter.BL(slow_handler_iter->second.slowmem_code);
@@ -300,6 +301,7 @@ bool JitArm64::HandleFastmemFault(uintptr_t access_address, SContext* ctx)
   m_fault_to_handler.erase(slow_handler_iter);
 
   emitter.FlushIcache();
+
   ctx->CTX_PC = reinterpret_cast<std::uintptr_t>(fault_location);
   return true;
 }

Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp

@@ -25,6 +25,8 @@ using namespace Arm64Gen;
 
 void JitArm64::GenerateAsm()
 {
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
+
   // This value is all of the callee saved registers that we are required to save.
   // According to the AACPS64 we need to save R19 ~ R30 and Q8 ~ Q15.
   const u32 ALL_CALLEE_SAVED = 0x7FF80000;

Source/Core/Core/PowerPC/JitInterface.cpp

@@ -71,9 +71,8 @@ CPUCoreBase* InitJitCore(PowerPC::CPUCore core)
     break;
 
   default:
-    PanicAlertFmtT("The selected CPU emulation core ({0}) is not available. "
-                   "Please select a different CPU emulation core in the settings.",
-                   core);
+    // Under this case the caller overrides the CPU core to the default and logs that
+    // it performed the override.
     g_jit = nullptr;
     return nullptr;
   }

Source/Core/DolphinQt/CMakeLists.txt

@@ -473,6 +473,8 @@ if(APPLE)
   set_target_properties(dolphin-emu PROPERTIES
     MACOSX_BUNDLE true
     MACOSX_BUNDLE_INFO_PLIST ${CMAKE_CURRENT_SOURCE_DIR}/Info.plist.in
+    XCODE_ATTRIBUTE_CODE_SIGN_ENTITLEMENTS "${CMAKE_CURRENT_SOURCE_DIR}/DolphinEmu.entitlements"
+    XCODE_ATTRIBUTE_OTHER_CODE_SIGN_FLAGS "--deep --options=runtime"
     OUTPUT_NAME Dolphin
     )
 
@@ -516,6 +518,22 @@ if(APPLE)
     POST_BUILD COMMAND
     ${CMAKE_INSTALL_NAME_TOOL} -add_rpath "@executable_path/../Frameworks/"
     $<TARGET_FILE:dolphin-emu>)
+
+  if(MACOS_CODE_SIGNING)
+    # Code sign make file builds
+    add_custom_command(TARGET dolphin-emu
+      POST_BUILD COMMAND
+      /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY}" --deep --options=runtime --entitlements "${CMAKE_SOURCE_DIR}/Source/Core/DolphinQt/DolphinEmu.entitlements" "${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/Dolphin.app" || true)
+    
+    # Code sign builds for build systems that do have release/debug variants (Xcode)
+    add_custom_command(TARGET dolphin-emu
+      POST_BUILD COMMAND
+      /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY}" --deep --options=runtime --entitlements "${CMAKE_SOURCE_DIR}/Source/Core/DolphinQt/DolphinEmu.entitlements" "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_DEBUG}/Dolphin.app" || true)
+      
+    add_custom_command(TARGET dolphin-emu
+      POST_BUILD COMMAND
+      /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY}" --deep --options=runtime --entitlements "${CMAKE_SOURCE_DIR}/Source/Core/DolphinQt/DolphinEmu.entitlements" "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_RELEASE}/Dolphin.app" || true)
+  endif()
 else()
   install(TARGETS dolphin-emu RUNTIME DESTINATION ${bindir})
 endif()

Source/Core/DolphinQt/DolphinEmu.entitlements

@@ -0,0 +1,17 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>com.apple.security.cs.allow-jit</key>
+	<true/>
+	<!-- Needed for GameCube microphone emulation -->
+	<key>com.apple.security.device.audio-input</key>
+	<true/>
+	<!-- TODO: It is likely this requirement is coming from Qt, but should confirm -->
+  	<key>com.apple.security.automation.apple-events</key>
+  	<true/>
+	<!-- This is needed to use adhoc signed linked libraries -->
+ 	<key>com.apple.security.cs.disable-library-validation</key> 
+  	<true/>
+</dict>
+</plist>

Source/Core/MacUpdater/CMakeLists.txt

@@ -13,6 +13,8 @@ set(SOURCES
 add_executable(MacUpdater ${SOURCES})
 
 set(MacUpdater_NAME "Dolphin Updater")
+set(MacUpdater_BIN_DIR ${CMAKE_BINARY_DIR}/Binaries)
+set(MacUpdater_BUNDLE_PATH ${MacUpdater_BIN_DIR}/${MacUpdater_NAME}.app)
 
 set_target_properties(MacUpdater PROPERTIES
   MACOSX_BUNDLE true
@@ -53,8 +55,24 @@ foreach(sb ${STORYBOARDS})
 
   add_custom_command(TARGET MacUpdater POST_BUILD
                      COMMAND ${IBTOOL} --errors --warnings --notices --output-format human-readable-text
-                                       --compile ${MacUpdater_BIN_DIR}/${MacUpdater_NAME}.app/Contents/Resources/${sb}c
+                                       --compile ${MacUpdater_BUNDLE_PATH}/Contents/Resources/${sb}c
                      ${CMAKE_CURRENT_SOURCE_DIR}/${sb}
                      COMMENT "Compiling Storyboard ${sb}...")
 endforeach()
 
+if(MACOS_CODE_SIGNING)
+  if (MACOS_CODE_SIGNING_IDENTITY_UPDATER STREQUAL "")
+    set(MACOS_CODE_SIGNING_IDENTITY_UPDATER "${MACOS_CODE_SIGNING_IDENTITY}")
+  endif()
+
+  # Make file build code sign
+  add_custom_command(TARGET MacUpdater POST_BUILD
+    COMMAND test ${MacUpdater_BUNDLE_PATH} || /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY_UPDATER}" --deep --options runtime ${MacUpdater_BUNDLE_PATH})
+  
+  # Xcode build code sign
+  add_custom_command(TARGET MacUpdater POST_BUILD
+    COMMAND test "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_DEBUG}/${MacUpdater_NAME}.app" || /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY_UPDATER}" --deep --options runtime "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_DEBUG}/${MacUpdater_NAME}.app")
+    
+  add_custom_command(TARGET MacUpdater POST_BUILD
+    COMMAND test "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_RELEASE}/${MacUpdater_NAME}.app" || /usr/bin/codesign -f -s "${MACOS_CODE_SIGNING_IDENTITY_UPDATER}" --deep --options runtime "${CMAKE_RUNTIME_OUTPUT_DIRECTORY_RELEASE}/${MacUpdater_NAME}.app")
+endif()

Source/Core/UICommon/AutoUpdate.cpp

@@ -140,7 +140,11 @@ static std::string GetPlatformID()
 #if defined _WIN32
   return "win";
 #elif defined __APPLE__
+#if defined(MACOS_UNIVERSAL_BUILD)
+  return "macos-universal";
+#else
   return "macos";
+#endif
 #else
   return "unknown";
 #endif

Source/Core/VideoCommon/VertexLoaderARM64.cpp

@@ -54,6 +54,7 @@ VertexLoaderARM64::VertexLoaderARM64(const TVtxDesc& vtx_desc, const VAT& vtx_at
     : VertexLoaderBase(vtx_desc, vtx_att), m_float_emit(this)
 {
   AllocCodeSpace(4096);
+  const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
   ClearCodeSpace();
   GenerateVertexLoader();
   WriteProtect();