Summary:
Follow the clang-tidy suggestion to replace reset-release with move assignment.
Move assignment's effect for unique_ptr:
> Effects: Transfers ownership from `u` to `*this` as if by calling `reset(u.release())`
followed by an assignment from `std::forward<D>(u.get_deleter())`.
Summary:
Remove X86MCPlusBuilder code that duplicates checks in X86BaseInfo.
Remove isINC and isDEC as redundant.
The new code of `X86MCPlusBuilder::isMacroOpFusionPair` is functionally
equivalent to `X86AsmBackend::isMacroFused`. However, as the method is
declared/defined in X86AsmBackend.cpp and not exported in a header file,
there's no way to use it in BOLT without changes in LLVM code.
(cherry picked from FBD33440373)
Summary:
Remove patterns ineligible for macro-fusion:
- First instruction has a memory destination
This is a temporary commit to align BOLT with LLVM MC interfaces.
(cherry picked from FBD33479340)
Summary:
When building with GCC and linking against jemalloc there is an
assertion when Debug Patches unique pointers are being freed from an std vector.
The base destructor gets invoked which results in type mismatch.
Added virtual destructor to the base class.
(cherry picked from FBD33489259)
Summary:
Reformat code and put options in lexicographical order.
Comparing to clang-format output, manual formatting looks cleaner to me.
(cherry picked from FBD33481692)
Summary:
Adding support for DW_FORM_data_2, DW_FORM_data_1, DW_FORM_udata.
With new .debug_info code only need to modify the check.
(cherry picked from FBD33302731)
Summary:
Now that we are re-writing .debug_info we are not longer restricted to have same size patches.
Simplifying logic to use direct forms.
(cherry picked from FBD32971159)
Summary:
If `addUnknownControlFlow` in `BinaryFunction::postProcessIndirectBranches`
is invoked with a basic block that has multiple edges to the same successor,
it leads to an assertion in `BinaryBasicBlock::removePredecessor`.
For basic blocks with multiple edges to the same successor, the default
behavior of removePredecessor is to remove all occurrences of the
predecessor block in its predecessor list (Multiple=true).
Example:
```A -> B (two edges)
A->removeAllSuccessors()
for each successor of block A: // B twice
// this removes both occurrences of A in B's predecessors list
B->removePredecessor(A);
// this invocation triggers an assert as A is no longer in B's
// predecessor list
B->removePredecessor(A);
```
This issue is not fixed by NormalizeCFG as `removeAllSuccessor` is called
earlier (from `buildCFG` -> `postProcessIndirectBranches`).
Solve this issue by collecting the successors into a set (`SmallPtrSet`) first,
before invoking `SuccessorBB->removePredecessor(this)`.
GitHub issue: https://github.com/facebookincubator/BOLT/issues/187
(cherry picked from FBD30796979)
Summary:
Changed the behavior of how we handle .debug_info section.
Instead of patching it will now rewrite it.
With this approach we are no longer constrained to having new values
of the same size.
It handles re-writing by treating .debug_info as raw data.
It copies chunks of data between patches, with new data written in
between.
(cherry picked from FBD32519952)
Summary:
Refactor remaining bolt sources to follow the braces rule for if/else/loop from
[LLVM Coding Standards](https://llvm.org/docs/CodingStandards.html).
(cherry picked from FBD33345885)
Summary:
Refactor bolt/*/Profile to follow the braces rule for if/else/loop from
[LLVM Coding Standards](https://llvm.org/docs/CodingStandards.html).
(cherry picked from FBD33345741)
Summary:
Refactor bolt/lib/Target to follow the braces rule for if/else/loop from
[LLVM Coding Standards](https://llvm.org/docs/CodingStandards.html).
(cherry picked from FBD33345353)
Summary:
Refactor bolt/*/Passes to follow the braces rule for if/else/loop from
[LLVM Coding Standards](https://llvm.org/docs/CodingStandards.html).
(cherry picked from FBD33344642)
Summary:
The lower_bound might return the end iterator, the ignoring of which will
cause memory corruption.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD33307803)
Summary:
Fix missing string header file inclusion and link_fdata find
problem in lit tests. Change root-level tests to require
linux. Re-enable Windows in our root CMakeLists.txt.
(cherry picked from FBD33296290)
Summary:
Fix according to Coding Standards doc, section Don't Use
Braces on Simple Single-Statement Bodies of if/else/loop Statements.
This set of changes applies to lib Core only.
(cherry picked from FBD33240028)
Summary:
Since nops are now removed in a separate pass, the profile is consumed
on a CFG with nops. If previously a profile was generated without nops,
the offsets in the profile could be different if branches included nops
either as a source or a destination.
This diff adjust offsets to make the profile reading backwards
compatible.
(cherry picked from FBD33231254)
Summary:
The patch moves the shortenInstructions and nop remove to separate binary
passes. As a result when llvm-bolt optimizations stage will begin the
instructions of the binary functions will be absolutely the same as it
was in the binary. This is needed for the golang support by llvm-bolt.
Some of the tests must be changed, since bb alignment nops might create
unreachable BBs in original functions.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD32896517)
Summary:
This patch adds AArch64 relocations handling in case updating of
debug sections is enabled
Elvina Yakubova,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD33077609)
Summary:
Create a new high-level target named bolt that builds all
BOLT artifacts, as well as a install-bolt target that installs them.
(cherry picked from FBD33133002)
Summary:
Gracefully handle binaries with split functions where two fragments are folded
into one, resulting in a fragment with two parent functions.
This behavior is expected in GCC8+ with -O2 optimization level, where both
function splitting and ICF are enabled by default.
On the BOLT side, the changes are:
- BinaryFunction: allow multiple parent fragments:
- `ParentFragment` --> `ParentFragments`,
- `setParentFragment` --> `addParentFragment`.
- BinaryContext:
- `populateJumpTables`: mark fragments to be skipped later,
- `registerFragment`: add a name heuristic check, return false if it failed,
- `processInterproceduralReferences`: check if `registerFragment`
succeeded, otherwise issue a warning,
- `skipMarkedFragments`: move out fragment traversal and skipping from
`populateJumpTables` into a separate function.
This change fixes an issue where unrelated functions might be registered
as fragments:
```
BOLT-WARNING: interprocedural reference between unrelated fragments:
bad_gs/1(*2) and amd_decode_mce.cold.27/1(*2)
```
(Linux kernel binary)
(cherry picked from FBD32786688)
Summary:
Refactor members of BinaryBasicBlock. Replace some std containers with
ADT equivalents. The size of BinaryBasicBlock on x86-64 Linux is reduced
from 232 bytes to 192 bytes.
(cherry picked from FBD33081850)
Summary:
Build is already broken because VS fails to locate
llvm-boltdiff when running tests, and VS also complains that
include/bolt/Passes/InstrumentationSummary.h is lacking an include
string header. Disable this until we have a Windows buildbot to make
sure this build is sane.
(cherry picked from FBD33039972)
Summary:
Switched members of BinaryFunction to ADT where it was possible and
made sense. As a result, the size of BinaryFunction on x86-64 Linux
reduced from 1624 bytes to 1448.
(cherry picked from FBD32981555)
Summary:
For DWP case the AbbreviationsOffset is the offset of the abbrev
contribution in the DWP file, so can be none zero.
(cherry picked from FBD32961240)
Summary:
Currently, RuntimeDyld will not allocate a section without relocations
even if such a section is marked allocatable and defines symbols.
When we emit .debug_line for compile units with unchanged code, we
output original (input) data, without relocations. If all units are
emitted in this way, we will have no relocations in the emitted
.debug_line. RuntimeDyld will not allocate the section and as a result
we will write an empty .debug_line section.
To workaround the issue, always emit a relocation of RELOC_NONE type
when emitting raw contents to debug_line.
(cherry picked from FBD32909869)
Summary:
Some optimizations may remove all instructions in a basic block.
The pass will cleanup the CFG afterwards by removing empty basic
blocks and merging duplicate CFG edges.
The normalized CFG is printed under '-print-normalized' option.
(cherry picked from FBD32774360)
Summary:
The debug message for the last fall-through block was printed under the
reverse condition, i.e. when the block was not a fall-through. Remove
the debug message. If we'll need such information, we can add a pass
with more analysis, i.e. checking the last instruction, if the block is
reachable, etc.
(cherry picked from FBD32670816)
Summary:
TailDuplication::isInCacheLine makes the assumption that the block
has a valid layout index, which is not the case for unreachable blocks.
Add a check for a valid layout index.
(cherry picked from FBD32659755)
Summary:
In some of the system stack protection is enabled by default, which will
lead in extra symbols dependencies, which we want to avoid.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD32478426)
Summary:
As pointed out by Vladislav in issue 217, if our RTDyld-based
linker fails to locate a symbol, it will crash with segfault. Fix that.
(cherry picked from FBD32481543)
Summary:
Currently there are two issues rendering the use of bughunter/BOLT on a binary
with a large number of functions (100k) impossible:
1) `selectFunctionsToProcess` has O(binary_fn * force_fn) run-time, which is up
to quadratic with the number of functions in the binary.
2) It unnecessarily treats supplied function names as regexes.
This diff proposes the following changes to address the issue:
1. Add two options that treat function names as is, not as regexes, matching
bughunter usage model: `-funcs-no-regex`/`-funcs-file-no-regex`.
These options are complementary to `-funcs`/`-funcs-file` and `-skip-funcs`/
`-skip-funcs-file`. `funcs` takes precedence over `funcs-no-regex`.
2. Use string set to speed up function eligibility checking with
`-funcs-file-no-regex` to O(binary_fn * log force_fn).
(cherry picked from FBD28917225)
Summary:
The push and pop instructions might have wrong reorder due to this
error. Thanks rafaelauler for the provided test case.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD32478348)
Summary:
We were not tracking -DBUILD_SHARED_LIBS=ON and we introduced
some commits that break that by not specifying the correct dependencies.
Fix that.
(cherry picked from FBD32453377)
Summary: In some cases __open() is returning 0 for me. The open
syscall will return a negative number (-1) on error so 0 should be
valid.
This happens when one compiles a BOLT instrumented executable of
pyston (python implementation) and afterwards pip installs a package
which needs to be compiled and sets CFLAGS=-pipe. I could not reduce
it down to a small testcase but I guess it one can trigger when
manually closing std{in, out, err}. Everything seems to work
normally when disabling the assert for 0 in getBinaryPath() - I
decided to also modify the second case in readDescriptions() even
though I did not run into that one yet.
(cherry picked from FBD32409548)
Summary:
Make BOLT build in VisualStudio compiler and run without
crashing on a simple test. Other tests are not running.
(cherry picked from FBD32378736)
Summary:
Moved the FDATA printing under the condition of non-empty profile data.
The change reduces the assembly dumps.
(cherry picked from FBD32262675)
Summary:
Import the test. The assembly input has three functions with associated fdata.
The old link_fdata.sh script only replaces the symbol names with symbol values,
whereas fdata format expects to have symbol offsets against the anchor symbol.
Introduce the link_fdata.py script which is able to parse the input and produce
either an offset or an absolute symbol value.
(cherry picked from FBD32256351)
Summary:
The majority of tests in LLVM projects are using - instead of _ in the name,
i.e. `check-something.test` is preferred over `check_something.test`.
It makes sense for us to adopt the same naming scheme for our future tests and
to rename existing ones.
(cherry picked from FBD32185879)
Summary:
BOLT meta test using merge-fdata tool.
This tests BOLT instrumentation for a non-trivial binary, running instrumented
binary, and using the instrumentation profile for BOLT optimizations.
The results are verified between original, instrumented, and optimized binaries.
Additional tested features: boltdiff mode and merge-fdata for two profiles.
merge-fdata tool is linked with relocs on Linux to support this test.
(cherry picked from FBD32141812)
Summary:
Replace erroneous check for function eligibility from `Function.isIgnored()`
to `shouldOptimize(Function)`. This prevents non-simple functions from being
processed.
(cherry picked from FBD32301958)
Summary:
With LTO, it's possible for multiple DWARF compile units to share the
same abbreviation section set, i.e. to have the same abbrev_offset.
When units sharing the same abbrev set are located next to each other
and neither of them is being processed (i.e. contain processed
functions), it can trigger a bug in BOLT. When this happened,
the abbrev set is considered empty. Additionally, different units
may patch abbrev section differently.
The fix is to not rely on the next unit offset when detecting
abbreviation set boundaries and to delay writing abbrev section
until all units are processed.
(cherry picked from FBD31985046)
Summary:
This commit uses reviews.llvm.org/D6629 as a reference to optimize
X86::EFLAGS load/store in the instrumentation snippet by using lahf/sahf
instructions instead of pushf/popf.
(cherry picked from FBD31662303)
Summary:
Added new functionality of dumping simple functions into assembly.
This includes:
- function control flow (basic blocks, instructions),
- profile information as `FDATA` directives, to be consumed by link_fdata,
- data labels,
- CFI directives,
- symbols for callee functions,
- jump table symbols.
Envisioned usage:
1. Find a function that triggers BOLT crash (e.g. with `bughunter.sh`).
2. Generate reproducer asm source for that function (using `-funcs`).
3. Attach it to an issue.
4. Reduce and include as a test case.
Current limitations:
1. Emitted assembly won't match input file relocations.
2. No DWARF support.
3. Data is not emitted.
(cherry picked from FBD32746857)
Summary:
BinaryContext is available via BinaryFunction::getBinaryContext(),
hence there's no reason to pass both as arguments to a function.
In a similar fashion, BinaryBasicBlock has an access to BinaryFunction
via getFunction(). Eliminate unneeded arguments.
(cherry picked from FBD31921680)
Summary:
Change cmake config in BOLT to only support Linux. In other
platforms, we print a warning that we won't build BOLT. Change
configs to determine whether we will build BOLT runtime libs. This
only happens in x86 hosts. If true, we will build the runtime and
enable bolt-runtime tests. New tests that depend on the bolt_rt lib
needs to be marked REQUIRES:bolt-runtime. I updated the relevant
tests. Fix cmake to do not crash when building llvm with a target
that BOLT does not support.
(cherry picked from FBD31935760)
Summary:
Moves source files into separate components, and make explicit
component dependency on each other, so LLVM build system knows how to
build BOLT in BUILD_SHARED_LIBS=ON.
Please use the -c merge.renamelimit=230 git option when rebasing your
work on top of this change.
To achieve this, we create a new library to hold core IR files (most
classes beginning with Binary in their names), a new library to hold
Utils, some command line options shared across both RewriteInstance
and core IR files, a new library called Rewrite to hold most classes
concerned with running top-level functions coordinating the binary
rewriting process, and a new library called Profile to hold classes
dealing with profile reading and writing.
To remove the dependency from BinaryContext into X86-specific classes,
we do some refactoring on the BinaryContext constructor to receive a
reference to the specific backend directly from RewriteInstance. Then,
the dependency on X86 or AArch64-specific classes is transfered to the
Rewrite library. We can't have the Core library depend on targets
because targets depend on Core (which would create a cycle).
Files implementing the entry point of a tool are transferred to the
tools/ folder. All header files are transferred to the include/
folder. The src/ folder was renamed to lib/.
(cherry picked from FBD32746834)
Summary:
Increase the hard limit from 256 to 4096.
This fixes the 'Assertion failed: failed to open binary path' error I'm seeing.
(cherry picked from FBD31911946)
Summary:
This patch introduces remove-symtab option to be able to skip emitting
symtab section in the final binary.
Also this patch adds ".zdebug_*" (compressed debug section) in the list
of debug section names.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31738239)
Summary:
Some compilation units will contain only code that is left unmodified by
BOLT, e.g. there is no profile data available for any function from such
units as they are rarely or never executed.
To save processing time and memory, we disable building line info tables
for such units and write unmodified tables to the output file.
(cherry picked from FBD31599759)
Summary:
This patch temporarily disables instrumentation and higufy build not for
x86 platforms to be able to build llvm-bolt tool on aarch64.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31738306)
Summary:
Sync the file with storage device on data dump to stabilize
instrumentation testing
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31738021)
Summary:
After the "Allocate memory for constant islands on-demand" patch there
are couple of problems found in constant islands handling:
1. When creating constant island dependency we need to check that we
already allocated IslandInfo for BF.
2. In ADRRelaxationPass we need to set constant island check under new
hasIslandsInfo condition.
3. In binaryemitter we need to replace hasConstantIsland with
hasIslandsInfo check since originally the BF might not have constant
island, but might have access to other's BF CI.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31737935)
Summary:
Refactored Loc and LocList writers to write out entries during finalization phase,
and hid some of the details in a class.
This simplifies things from impelementation details, and also will be needed for
DWARF5 where we need to know how many locLists entries there are there.
(cherry picked from FBD31563795)
Summary:
Add lit.local.cfg to X86 and AArch64 folders.
Fix host_arch in lit config for AArch64.
Fix AArch64 and X86 tests.
Elvina Yakubova,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31702068)
Summary:
This patch handles 2 problems with LongJmp pass:
1. The pass should be executed before FinalizeFunctions, since the pass
may add new entry points for the function, and the
BinaryFunction::addEntryPoint has an assert "CurrentState == State::CFG"
2. Replaced shortJmp implementation with position-independent code.
Currently we could handle PIC binaries with max +-4Gb offsets, the
longJmp uses absolute addreses and could could be used only in non-PIE
binaries.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31416925)
Summary:
In non-PIC binaries compiler could save absolute addresses in constant
isalnd which we should handle properly. This patch adds relocations
handling in constant islands.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31416848)
Summary:
Change sed separator to allow replacing symbols with slash in the name.
This is required for symbol names produced by BOLT which include
"/1" suffix.
(cherry picked from FBD31324540)
Summary:
This patch adds R_AARCH64_TSTBR14 and R_AARCH64_CONDBR19 relocations
support in order to handle condition branches, cbz/cnbz and tbz/tbnz
instructions correctly
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31416734)
Summary:
The ELF::R_AARCH64_TLSDESC_LD64_LO12 and
ELF::R_AARCH64_TLSDESC_ADR_PAGE21 relocations might also be relaxed to
mov instructions, handle these cases
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31353063)
Summary:
Use the new API introduced in https://reviews.llvm.org/D106624
to request LLVM do not process relocations for debug sections, since
BOLT processes final binaries that are already relocated.
(cherry picked from FBD31449206)
Summary:
When the compiler emits line table program, it emits EOS using the label
at the end of the containing code section. Since each compilation unit
has its own set of code sections it works as expected (* see the excerpt
from the standard below). However, in BOLT the code from many CUs is
combined into a common section, such as hot text or cold text.
As a result, the symbol at the end of the section may point way past the
code sequence for a given unit.
Since we can emit functions in any order, we conservatively emit
end-of-sequence at the end of every emitted function.
Fixes a problem while intermixing source code with disassembly in
binutils' objdump.
(*) DWARF v4 6.2.5.3:
"Every line number program sequence must end with a DW_LNE_end_sequence
instruction which creates a row whose address is that of the byte after
the last target machine instruction of the sequence."
(cherry picked from FBD31347870)
Summary:
Generate line tables for original/unmodified functions directly from
input line tables, bypassing conversion into intermediate structures,
such as BinaryLineDivisions.
Emit end-of-sequence markers only when necessary, i.e. when the line
sequence is not adjacent to the next one, or at the end of the line
sequence for the compilation unit.
If the sequence starts with ambiguous line info (multiple lines per
address), make sure we emit all such lines.
Reduce memory consumption when updating debug info by eliminating
intermediate data structures allocation.
(cherry picked from FBD30829448)
Summary:
BOLT needs to generate line info tables using absolute addresses as well
as using the standard MC way of labels attached to instructions. Move
line table generation code under BOLT.
Ideally, we should be able to extend existing interfaces in LLVM, but
without other users of the interface it will be hard to justify the
change.
(cherry picked from FBD30723466)
Summary:
For historical reasons, we are populating FailedAddresses twice in
RewriteInstance. Remove the second (happening later) call to avoid the
confusion.
(cherry picked from FBD31278956)
Summary:
When rewriting .debug_abbrev section, update abbrev offsets for type
units in addition to compile units.
Reuse abbreviation entries if they were shared by multiple compile/type
units.
(cherry picked from FBD31262326)
Summary:
Create bolt/test/runtime folder and move tests that execute the binary.
Move lit.local.cfg with host_arch check to the corresponding folder.
Addresses issue facebookincubator/BOLT#132.
AArch64/tls.c shows a different behavior with clang hence marked as XFAIL
TODO: add a check for non-exec tests for a corresponding LLVM_TARGETS_TO_BUILD.
(cherry picked from FBD31132234)
Summary:
Previously, we were registering all CUs with aranges writer. Since DWO
CUs have offsets set to 0, and we were registering them after the
skeleton unit at offset 0 was already registered, it was mostly
harmless as DWO CUs were effectively ignored.
(cherry picked from FBD31162621)
Summary:
Instead of patching the original .debug_abbrev section contents,
generate new section data based on parsed compilation unit
abbreviations.
This eliminates the dependency on the LLVM extension that records
abbreviation attribute offsets while parsing .debug_abbrev contents.
The output with this patch should stay the same (NFC).
(cherry picked from FBD31133611)
Summary:
There are some cases, when relocations must not be processed by bolt.
This patch handles three of such cases:
* The linker might eliminate the instruction and replace it with NOP
* The linker might perform TLS relocations relaxations, replacing the
got to direct TP + offset access.
* Due to errata 843419 the linker might create a veneer, replacing the
load/store instruction with branching.
In both cases linker leaves old relocations, that are no longer matches
the instruction emmited to binary, so we must avoid processing of these
relocations.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31002384)
Summary:
There are few problems found when dealing with TLS relocations for
aarch64.
* RewriteInstance.cpp
** While analyzing TLS relocation we don't have to modify
SymbolAddress (which is the offset from the TLS section), so we need to
just skip verifiction
** The non-got related TLS relocations on aarch64 might be skipped too
** The forse relocation must be applied for GOT relocations on
Aarch64. The symbol adress for GOT relocation might no be pointing
on GOT section (for example ADRP GOT may point to the wrong section,
since GOT table is not page-aligned), so we won't try to get section by
the symbol address.
* Relocation.cpp - Remove R_AARCH64_TLSLE_ADD_TPREL_HI12 and
R_AARCH64_TLSLE_ADD_TPREL_LO12_NC from isGOT check, since they are not
got-related relocations
* BinaryFunction.h
** Remove R_AARCH64_TLSLE_ADD_TPREL_HI12 and
R_AARCH64_TLSLE_ADD_TPREL_LO12_NC from adding to relocation list, since
this is actually an offset in TLS section and BOLT does not change it we
don't need to do something with this relocations, the value won't change
in new binary files
** Refactor the code, separating aarch64 and x86 relocations
* AArch64MCPlusBuilder.cpp
** Add forgotten LO12 relocations to switch case to getTargetExprFor
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD31003349)
Summary:
LLVM started printing warnings when DWARFDebugInfoEntry::extractFast()
is invoked trying to read a DIE past the current unit limits. This
results in verbose warnings from BOLT which are harmless but confusing
to the user. Check the boundaries before calling the API above.
(cherry picked from FBD31097271)
Summary:
In "Add initial function injection support", Laith added this
code because injected functions would use the original text section as
the section to emit their code to. Now, what happens is that functions
are mapped to either their own section in non-reloc mode, or mapped to
a particular section in the pass reassign sections. So this section does
not need to have an output address anymore and this code is obsolete.
(cherry picked from FBD30980450)
Summary:
We have a problem where we will emit sections that we are not supposed
to emit (with no output offset assigned). This will make us write at
file offset 0 and corrupt the first sections in the binary (usually
.interp section will be corrupted and bash will refuse to run the
binary).
This only happens in non-reloc mode when using JTS_BASIC and when we
do not emit a function that has a jump table (if it gets too large).
Using -update-debug-sections will trigger the pass
check-large-functions, which will mark large funcs as non-simple
and will hide this bug.
(cherry picked from FBD30882012)
Summary:
This commit introduces TryLock usage for SimpleHashTable getter to
avoid deadlock and relax syscalls usage which causes significant
overhead in runtime.
The old behavior left under -conservative-instrumentation option passed
to instrumentation library.
Also, this commit includes a corresponding test case: instrumentation of
executable which performs indirect calls from common code and signal
handler.
Note: in case if TryLock was failed to acquire the lock - this indirect
call will not be accounted in the resulting profile.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30821949)
Summary:
This commit adds checking if maxIndividualTestTime is availabe on
the platform. If available - it sets per test timeout to 60sec and
declares lit-max-individual-test-time feature for further checking
by particular test cases.
Based on https://reviews.llvm.org/D64251 implementation.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30821986)
Summary:
The clang 12 doesn't want to build this place due to unrelated
types of iterator element and std vector.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30821177)
Summary: Changing to use the new APIs for getting offset of attribute from .debug_info. They were split in to multiple ones so that Offset can be gotten seperatly.
(cherry picked from FBD30616705)
Summary:
Three way branches commonly appear
in HHVM. They have one test and then two jumps. The
jump's destinations are not currently optimized.
This pass attempts to optimize which is the first branch.
(cherry picked from FBD30460441)
Summary:
There are 2 problems found when handling ADR instruction:
1. When extracting value from the ADR instruction we need to do
it another way, then we do it for ADRP instruction.
2. When creating target expression the VariantKind should be other for
ADR instruction.
And we introduces R_AARCH64_ADR_PREL_LO21,
R_AARCH64_TLSDESC_ADR_PREL21 and R_AARCH64_ADR_PREL_PG_HI21_NC
relocations support.
Also this patch introduces AdrPass, which will replace non-local
pointing ADR instructions with ADRP + ADD instructions sequence due to
small offset range of ADR instruction, so after BOLT magic there are no
guarantees that ADR instruction will still be in the range of
just +- 1MB from its target. The instruction replacement needs
relocations to be avalailable, so we won't remove "IsFromCode"
relocations after disassembly from BF anymore. Also we need original
offset of ADR instruction to be available so we add offset annotation
for these instructions.
The last thing this patch adds is ARM testing directory, which will be
used only on ARM testing servers. The common tests (non-assembler tests
which are platform-independent) might be moved from the X86 directory to
the parent one in the future, so such tests could be tested on both X86
and ARM machines.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30497379)
Summary:
Currently most of the warnings are printed only in debug mode. Since
relocations are very important for binary correct work I suggest to
print number of failed to process relocations to pay extra attention in
case some problems with them were met
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30500629)
Summary:
Added a function in TailDuplication
that will do Constant and Copy Propagation for blocks that
we duplicated as a part of tail duplication. Added supporting
functions to MCPlusBuilder to find src registers and replace
registers
(cherry picked from FBD30231907)
Summary:
This patch is part of preparation for golang support. The golang symbols
might have spaces in the name (for example "type..eq.[10]interface {}").
Since fdata uses spaces as a field separator such names brakes the fdata
format, so we need to escape whitespaces and backslashes in symbol names
using the backslash character.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD29999491)
Summary:
Remove unused code introduced a while ago (2016), with its use removed
since then.
PR facebookincubator/BOLT#198
Author: Amir Aupov <aaupov@fb.com>
(cherry picked from FBD30376537)
Summary:
The ADRP instructions has 21 bits to store page offsets + 12 lowest bits
are zero, that give us a total of 33 bits (32 bits for address + 1 sign
bit, to address +- 4GB).
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30283044)
Summary:
This commit adds dummy tests for checking instrumentation
support for PIE executables and shared libraries.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092729)
Summary:
To avoid RELATIVE relocations avoid using of GOT table
by using hidden visibility for all symbols in library.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092712)
Summary:
The trampolines are no loger pointers to the functions. For
propper name resolving by bolt use extern "C" for all external symbols
in instr.cpp
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092698)
Summary:
This commit introduces -instrumentation-binpath argument used
to point instuqmented binary in runtime in case if /proc/self/map_files
path is not accessible due to access restriction issues.
Vasily Leonenko
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092681)
Summary:
This commit introduces static binaries instrumentation
support. Note that current implementation does not support profile
output on the instrumented binary finalization. So it requires to use
-instrumentation-sleep-time=N (N>0) option usage. Note: There is
unhandled case with static PIE executable which might have dynamic
header.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092471)
Summary:
This commit adds support for opening libs based on links
/proc/self/map_files. For this we're getting current virtual address
and searching the lib in the directory with such address range. After
that, we're getting full path to the binary by using readlink
function. Direct read from link in /proc/self/map_files entries is not
possible because of lack of permissions.
Elvina Yakubova,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092422)
Summary:
This commit adds support for getting directory entries and
reading value of a symbolic link in instrumentation runtime library
Elvina Yakubova,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092362)
Summary:
This commit implements new method for _start & _fini functions hooking
which allows to use relative jumps for future PIE & .so library support.
Instead of using absolute address of _start & _fini functions known on
linking stage - we'll use dynamically created trampoline functions and
use corresponding symbols in instrumentation runtime library.
As we would like to use instrumentation for dynamically loaded binaries
(with PIE & .so), thus we need to compile instrumentation library with
"-fPIC" flag to support relative address resolution for functions and
data.
For shared libraries we need to handle initialization of instrumentation
library case by using DT_INIT section entry point.
Also this commit adds detection if the binary is executable or shared
library based on existence of PT_INTERP header. In case of shared
library we save information about real library init function address
for further usage for instrumentation library init trampoline function
creation and also update DT_INIT to point instrumentation library init
function.
Functions called from init/fini functions should be called with forced
stack alignment to avoid issues with instructions which relies on it.
E.g. optimized string operations.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD30092316)
Summary:
Move the common code into MCPlusBuilder.h.
Use group 1 `kTailCall` MCAnnotation instead of dynamically allocated
annotation.
This diff reduces the processing time overhead to 1.5% vs using
TAILJMP opcode.
(cherry picked from FBD30055585)
Summary:
The linker can generate 8- or 16-byte entries in .plt.got and .plt.sec
sections. On X86, the main differentiator is the presence of endbr64
instruction at the beginning of the entry. Detect the instruction and
adjust the size accordingly.
(cherry picked from FBD29847639)
Summary:
.stab and .stabstr are special sections containing debugging
information and strings associated with the debugging information.
This commit adds them to the list of debugging sections, so
these sections can be removed for output binary.
Vasily Leonenko,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD29746153)
Summary:
Match new direct call generated during ICP to correct pseudo probe
New call is matched to the probes of original call instruction.
(cherry picked from FBD29591662)
Summary:
Created a binary pass that records how many
times tail duplication would be used and how many cache
misses it would theoretically stop
(cherry picked from FBD29619858)
Summary:
We extended DynoStats to dump the histogram per instruction opcode. By
default the dump is turned off. Use '-print-dyno-opcode-stats' to enable
the dump.
BOLT also dumps for each instruction opcode the maximum execution count and
corresponding function name and basic block offsets where the instruction
occurs. Below is a sample of the dump:
Opcode, Execution Count, Max Exec Count, Function Name:Offset
SHR8rCL, 232, 232, _ZNK5folly14AsyncSSLSocket4goodEv:53
VPADDDYrr, 13956, 388, chacha20_encrypt_bytes.part.0/3:736
PMOVSXBWrr, 4, 2, ares_expand_name/1:264
VMOVAPSmr, 1082, 43, chacha20_encrypt_bytes.part.0/3:2864
VPSHUFBrr, 9540, 1667, chacha20_encrypt_bytes.part.0/3:4416
VPUNPCKLDQYrr, 1102, 188, jsimd_ycc_rgb_convert_avx2/1:125
VPBROADCASTQYrm, 39, 39, chacha20_encrypt_bytes.part.0/3:400
PMOVSXWDrr, 8, 2, ares_expand_name/1:264
VPORrr, 817, 129, jsimd_idct_islow_avx2/1:41
PSLLDri, 8690752, 65644, blockmix_salsa8_xor/1:1424
(cherry picked from FBD28859624)
Summary:
A binary can contain multiple PLT sections with different name and
attributes (such as an entry size). Extend the support to .plt.sec and
refactor the code to make future extensions simpler.
(cherry picked from FBD29502107)
Summary:
clang-12 now compiles bolt without warnings.
Some warnings were fixed if possible while others were suppressed by
doing (void)variable for unused variable warnings or moving code inside
assert statements of LLVM_DEBUG blocks.
(cherry picked from FBD29469054)
Summary:
Add code to read more dynamic relocations (DT_JMPREL) and enforce strict
checks that corresponding sections sizes match .dynamic entry
description.
(cherry picked from FBD29502109)
Summary:
The code for writing out dwo files wasn't handling case where DWP is an input.
Because all the sections are part of the same binary.
One note with current implementation. .debug-str.dwo will have strings for all the dwo objects.
This is because llvm-dwp de-duplicates strings and combines them in to one section. It then re-writes .debug-str-offsets.dwo to point to new .debug-str.dwo section.
(cherry picked from FBD29244835)
Summary:
Our YAML objects contain references to dynamic relocations via .dynamic,
but there are no corresponding relocation sections. Change .dynamic
contents to specify no dynamic relocations.
(cherry picked from FBD29502108)
Summary:
Move the code that handles true external references (non-unreachable)
out of a for-loop in `BinaryFunction::disassemble`.
(cherry picked from FBD29411345)
Summary:
Handle R_X86_64_64 the same way as R_X86_64_32;
`getSizeForType` takes care of the size:
```x86_64 ABI relocation types
Name Value Field Calculation
R_X86_64_64 1 word64 S + A
R_X86_64_32 10 word32 S + A
```
(cherry picked from FBD29370417)
Summary:
When we fold a function in relocation mode, make sure to clear its state
to avoid emitting relocations against undefined symbols.
(cherry picked from FBD29245320)
Summary:
Dived more in to DWARF APIs and llvm-symbolizer this is a more streamline way of doing it, and address base gets set properly.
Writing out dwo files with dwp input will be separate patch.
(cherry picked from FBD31361529)
Summary:
When indirect call is instrmented it locks SimpleHashTable's mutex on get() call.
If while locked we we receive a signal and signal handler also will call
indirect function we will end up with deadlock.
PR facebookincubator/BOLT#167
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD28909921)
Summary:
Suppresses the warning
```
src/DebugData.h:338:20: warning: 'addList' overrides a member function but is not marked 'override' [-Wsuggest-override]
```
(cherry picked from FBD28858201)
Summary:
Make bolt decode pseudo probe section in binary
For more detail of pseudo probe, check https://reviews.llvm.org/D86490.
(cherry picked from FBD28856316)
Summary:
While printing debug info for instructions, we should use line tables
from the corresponding DWARF CU which could be different from the
containing function CU in case of inlined instructions.
(cherry picked from FBD28908324)
Summary:
FBD55943 changed the way ProcessAllSections works in RuntimeDyld. After
the change, all sections, including symbol table, section table, etc.
are loaded into memory whenever ProcessAllSections is enabled.
In BOLT we rely on RuntimeDyld for processing sections with relocations.
These include most allocatable sections and additionally .debug_line.
The latter is skipped by RuntimeDyld without ProcessAllSections flag.
If we enable ProcessAllSections, we will have to deal with allocating
memory for more sections than we need (see above) and later to filter
them out.
The alternative is to mark all sections that we actually plan to use as
"required for execution" (using RuntimeDyld terminology). For
.debug_line section on ELF it means adding SHF_ALLOC flag. On MachO,
RuntimeDyld currently treats all sections as required.
(cherry picked from FBD28729398)
Summary:
This patch introduces LoopInversionPass. Its main purpose is to ensure
that the loop layout is optimal depending on the profile information. So
if profile information shows that the loop is used, the unconditional
jump instruction must be executed only once and vice-versa. Please take
a look to the pass header file and test for more details.
Also change link_fdata script a bit, to be able to change FDATA prefix,
like FileCheck does.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
PR facebookincubator/BOLT#153
(cherry picked from FBD28391811)
Summary:
Implemented support for Debug Fission.
For the most part it doesn't impact Monolithic execution path.
One area that was changed is the DW_AT_low_pc/DW_AT_high_pc conversion. Before it was to DW_AT_ranges/DW_AT_low_pc, now DW_AT_low_pc is kept in same place.
Another more visible impact is in Skeleton CU the DW_AT_low_pc is replaced with DW_AT_ranges_base if it's not originally present and bolt converted ranges conversion inside the dwo units.
Output of this are multiple .dwo files with updated debug information.
(cherry picked from FBD29569788)
Summary:
Remove relocations against internal function labels, e.g. jump table
relocations, only when overwriting them.
While reading an input file with relocations, we create internal
relocations against code references (we skip PIC relocations).
Later, when we discover jump tables, we remove corresponding relocations
with the assumption that original relocations will either be ignored or
replaced by new relocations. However, it is possible to miss some
references to the jump table, in which case the original entries will
not be ignored. While such situation is abnormal, it is still a
better/safer approach to preserve relocations if we are not replacing
them with new ones.
(cherry picked from FBD28406628)
Summary:
Explicit assignment operator can be replaced with an implicit one.
Remove it to allow an implicit copy constructor:
```
bolt/src/Passes/DataflowAnalysis.h:74:8: warning: definition of
implicit copy constructor for 'ProgramPoint' is deprecated because it
has a user-declared copy assignment operator [-Wdeprecated-copy]
void operator=(const ProgramPoint &PP) {
^
bolt/src/Passes/DataflowAnalysis.h:62:14: note: in implicit copy
constructor for 'llvm::bolt::ProgramPoint' first required here
return ProgramPoint(&*Last);
```
(cherry picked from FBD28335138)
Summary:
Since gcc/ld could produce and expect PIE files we need to pass -no-pie option to avoid linking errors for tests.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
(cherry picked from FBD28360045)
Summary:
Reorder-blocks optimization pass doesn't take into account that
available offset for legacy Jcc instructions (for example,
JRCXZ - operand 8 bits) has to be less than 255 bytes.
It's rare case and to exclude such functions with unsupported
instructions from optimization passes added extra checking
Alexey Moksyakov
Advanced Software Technology Lab, Huawei
(cherry picked from FBD28264117)
Summary:
Ran iwyu multiple times, manually picked header remove lines.
Reached fixed point wrt removal: iwyu doesn't automatically remove
any more headers or forward declarations.
(cherry picked from FBD29569221)
Summary:
Previously, we used p_align value of the code segment to predict the
mapping of the segment at runtime. However, at times the reported
value is not aligned and at other times the actual aligned value will
be different because of the different page size used.
All we know is that the page size used at runtime should not exceed
p_align value. Adjust our segment address matching accordingly.
(cherry picked from FBD28133066)
Summary:
Addressing comments from the review for "Expand auto types".
Use const reference in MCPlusBuilder for MCInstrDesc where the copy
is not necessary.
(cherry picked from FBD27844344)
Summary:
We may have a CU with empty ranges, so accept errors coming
from DWARFDie::getAddressRanges(). This happens when using tools that
selectively strip debuginfo from the binary.
(cherry picked from FBD27602731)
Summary:
Refactor SectionPatches to avoid the use of extra map and a cast
from StringRef to std::string.
cherry-picked from FBD26756560
(cherry picked from FBD27490641)
Summary:
The user may wish to run BOLT for printing statistics only
(i.e. to check that the profile is valid). Add an option to run BOLT
without writing any output file, similar to a dry run. This option
is triggered by supplying -o with "/dev/null".
(cherry picked from FBD29568632)
Summary:
During the initial indirect jump analysis, we used to assert that the
discovered jump table type matched the pattern of the corresponding
instruction sequence. E.g., for PIC jump table memory we expected the
PIC jump table instruction sequence. The assertions were too
conservative, as in the case of a mismatch we can mark the indirect jump
as having an unknown control flow. That should be sufficient to either
skip the function processing or rely on relocation information for
possible recovery of the control flow.
(cherry picked from FBD27255816)
Summary:
Fix a bug with instrumentation when trying to instrument
functions that share a jump table with multiple indirect
jumps. Usually, each indirect jump that uses a JT will have its own
copy of it. When this does not happen, we need to duplicate the jump
table safely, so we can split the edges correctly (each copy of the
jump table may have different split edges). For this to happen, we
need to correctly match the sequence of instructions that perform the
indirect jump to identify the base address of the jump table and patch
it to point to the new cloned JT. It was reported to us a case in
which the compiler generated suboptimal code to do an indirect jump
which our matcher failed to identify.
Fixesfacebookincubator/BOLT#126
(cherry picked from FBD27065579)
Summary:
Whenever BOLT encounters a data reference in code, it tries to convert
it into <Object+Offset> form. The primary reason behind this approach is
to support read-only data-reordering optimization. However, with the
current level of the linker and compiler support we don't have enough
information to always correctly restore the original <Object+Offset>.
E.g. with zero-sized symbols we have to speculate that the actual size
of the underlying object extends to the next symbol. Most of the time,
there will be an object pointed by a zero-sized symbol and even
if we are guessing incorrectly, there will be no harm in creating
references of such form.
The problem happens when there's no object corresponding to the original
symbol and the next object is an (unmarked) jump table:
A: # <- zero-sized object
.LJUMP_TABLE:
.long <entry1>
.long <entry2>
....
.LB:
.long 21
.LC:
.long 42
The jump table will be moved and all references past it (up to the next
named object) will be incorrectly updated.
We should not speculate about the size of A in a case like that and
treat all discovered data objects (and thus references) independently.
(cherry picked from FBD27005660)
Summary:
This PR introduces 2 new instrumentation options:
1. instrumentation-no-counters-clear: Discussed at https://github.com/facebookincubator/BOLT/issues/121
2. instrumentation-wait-forks: Since the instrumentation counters are mapped as MAP_SHARED it will be nice to add ability to wait until all forks of the parent process will die using tracking of process group.
The last patch is just emitBinary code refactor.
Vladislav Khmelevsky,
Advanced Software Technology Lab, Huawei
Pull Request resolved: https://github.com/facebookincubator/BOLT/pull/125
GitHub Author: Vladislav Khmelevskyi <Vladislav.Khmelevskyi@huawei.com>
(cherry picked from FBD26919011)
Summary:
TBSS section is a "virtual" section that does not take memory or file
space. Ignore it completely while adjusting section sizes.
(cherry picked from FBD26824484)
Summary:
There is no real link between CU and TU, so relying on fact
that address are the same, and we are updating all of them.
(cherry picked from FBD28112114)
Summary:
This commit is the first step in rebasing all of BOLT
history in the LLVM monorepo. It also solves trivial build issues
by updating BOLT codebase to use current LLVM. There is still work
left in rebasing some BOLT features and in making sure everything
is working as intended.
History has been rewritten to put BOLT in the /bolt folder, as
opposed to /tools/llvm-bolt.
(cherry picked from FBD33289252)
