search

llvm / llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies.
http://llvm.org
Other
29.41k stars 12.15k forks source link

LLVM occassionally crashes when compiling project #100084

Closed JakeChampion closed 1 month ago

JakeChampion commented 4 months ago

The project sometimes builds and sometimes fails to build with the below backtrace

I'm not sure what other information to give here, other than the backtrace - please let me know what (if any) other information would be useful to include and I will try my best to return with that information 🙇 

0:12.22 PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
 0:12.32  #0 0x000000010aba7df0 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb83df0)
 0:12.32  #1 0x000000010aba835c SignalHandler(int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb8435c)
 0:12.32  #2 0x000000019436b584 (/usr/lib/system/libsystem_platform.dylib+0x180477584)
 0:12.32  #3 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784)
 0:12.32  #4 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784)
 0:12.32  #5 0x0000000104f97e18 lld::wasm::CustomSection::writeTo(unsigned char*) (/Users/jakechampion/.mozbuild/clang/bin/lld+0x10033be18)
 0:12.32  #6 0x000000010ab41c34 std::__1::__function::__func<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0, std::__1::allocator<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1dc34)
 0:12.32  #7 0x000000010ab41b60 std::__1::__function::__func<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0, std::__1::allocator<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1db60)
 0:12.32  #8 0x000000010ab411c4 llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::work(llvm::ThreadPoolStrategy, unsigned int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d1c4)
 0:12.33  #9 0x000000010ab4136c void* std::__1::__thread_proxy[abi:un170006]<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::ThreadPoolExecutor(llvm::ThreadPoolStrategy)::'lambda'()::operator()() const::'lambda'()>>(void*) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d36c)
 0:12.33 #10 0x000000019433af94 (/usr/lib/system/libsystem_pthread.dylib+0x180446f94)
 0:12.33 #11 0x0000000194335d34 (/usr/lib/system/libsystem_pthread.dylib+0x180441d34)
llvmbot commented 4 months ago

@llvm/issue-subscribers-backend-webassembly

Author: Jake Champion (JakeChampion)

The project sometimes builds and sometimes fails to build with the below backtrace I'm not sure what other information to give here, other than the backtrace - please let me know what (if any) other information would be useful to include and I will try my best to return with that information 🙇 ``` 0:12.22 PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace. 0:12.32 #0 0x000000010aba7df0 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb83df0) 0:12.32 #1 0x000000010aba835c SignalHandler(int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb8435c) 0:12.32 #2 0x000000019436b584 (/usr/lib/system/libsystem_platform.dylib+0x180477584) 0:12.32 #3 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784) 0:12.32 #4 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784) 0:12.32 #5 0x0000000104f97e18 lld::wasm::CustomSection::writeTo(unsigned char*) (/Users/jakechampion/.mozbuild/clang/bin/lld+0x10033be18) 0:12.32 #6 0x000000010ab41c34 std::__1::__function::__func<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0, std::__1::allocator<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1dc34) 0:12.32 #7 0x000000010ab41b60 std::__1::__function::__func<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0, std::__1::allocator<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1db60) 0:12.32 #8 0x000000010ab411c4 llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::work(llvm::ThreadPoolStrategy, unsigned int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d1c4) 0:12.33 #9 0x000000010ab4136c void* std::__1::__thread_proxy[abi:un170006]<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::ThreadPoolExecutor(llvm::ThreadPoolStrategy)::'lambda'()::operator()() const::'lambda'()>>(void*) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d36c) 0:12.33 #10 0x000000019433af94 (/usr/lib/system/libsystem_pthread.dylib+0x180446f94) 0:12.33 #11 0x0000000194335d34 (/usr/lib/system/libsystem_pthread.dylib+0x180441d34) ```
aheejin commented 4 months ago

Even if it does not always fail, we'd appreciate a reproducer. Can you provide one?

JakeChampion commented 4 months ago

I've now landed enough of the patches to make the reproduction hopefully simpler to do.

This happens when compiling Mozilla's SpiderMonkey to WASI in debug mode and with Intl included

A full log output can be found at https://treeherder.mozilla.org/logviewer?job_id=468809046&repo=try&lineNumber=5901

Gecko-Dev, which contains SpiderMonkey is a rather large codebase though, the steps to get it building are quite a bit -- https://firefox-source-docs.mozilla.org/setup/macos_build.html

Once that is done you can create a mozconfig file named 'mozconfig-intl-debug' containing:

ac_add_options --enable-project=js
ac_add_options --enable-application=js
ac_add_options --target=wasm32-unknown-wasi
ac_add_options --without-system-zlib
ac_add_options --disable-jit
ac_add_options --disable-shared-js
ac_add_options --disable-shared-memory
ac_add_options --disable-tests
ac_add_options --disable-clang-plugin
ac_add_options --enable-jitspew
ac_add_options --enable-optimize=-O3
ac_add_options --enable-portable-baseline-interp
mk_add_options AUTOCLOBBER=1
ac_add_options --host=aarch64-apple-darwin
ac_add_options --enable-debug

And building the project with MOZCONFIG=./mozconfig-intl-debug ./mach build

aheejin commented 4 months ago

Can you possibly generate a reproducer using wasm-ld --reproduce?

JakeChampion commented 4 months ago

Thanks, I've generated that file, it's 574M in size, not too sure where I could upload that

aheejin commented 4 months ago

You can upload it to a Google drive or Dropbox or something and paste the link here.

JakeChampion commented 3 months ago

@aheejin here it is https://mega.nz/file/VlNnRTSK#2kj3qBJFjlgQjl_vOT2MRfZQAG849r1sC9iXvF4qkRs

guybedford commented 3 months ago

I'll be picking up from @JakeChampion on this Fastly / BytecodeAlliance error report. If there is any more info that can be provided please just let me know.

aheejin commented 1 month ago

Sorry for the long delay. I tried to build the reproducer but I don't seem to have the right WASI environment set..?

$ wasm-ld @response.txt 
wasm-ld: warning: unknown -z value: noexecstack
wasm-ld: warning: unknown -z value: text
wasm-ld: warning: unknown -z value: relro
wasm-ld: warning: unknown -z value: now
wasm-ld: warning: unknown -z value: nocopyreloc
wasm-ld: error: unable to find library -lm
wasm-ld: error: unable to find library -lwasi-emulated-process-clocks
wasm-ld: error: unable to find library -lwasi-emulated-getpid
wasm-ld: error: unable to find library -lc++
wasm-ld: error: unable to find library -lc++abi
wasm-ld: error: unable to find library -lc

How am I supposed to set up the libraries?

JakeChampion commented 1 month ago

Do you have WASI SDK installed? https://github.com/WebAssembly/wasi-sdk

aheejin commented 1 month ago

Thanks I installed WASI SDK using the prebuilt binarires.

By the way, what's your LLVM version (release version or git commit)?

sbc100 commented 1 month ago

Sorry for the long delay. I tried to build the reproducer but I don't seem to have the right WASI environment set..?

$ wasm-ld @response.txt 
wasm-ld: warning: unknown -z value: noexecstack
wasm-ld: warning: unknown -z value: text
wasm-ld: warning: unknown -z value: relro
wasm-ld: warning: unknown -z value: now
wasm-ld: warning: unknown -z value: nocopyreloc
wasm-ld: error: unable to find library -lm
wasm-ld: error: unable to find library -lwasi-emulated-process-clocks
wasm-ld: error: unable to find library -lwasi-emulated-getpid
wasm-ld: error: unable to find library -lc++
wasm-ld: error: unable to find library -lc++abi
wasm-ld: error: unable to find library -lc

How am I supposed to set up the libraries?

Looks like there could be an absolute path in the response file that you would need to update?

aheejin commented 1 month ago

@sbc100 Yeah I installed WASI SDK and updated the -L path in response.txt.

aheejin commented 1 month ago

@sbc100 While you're here, do you have any idea what the problem could be?

llvmbot commented 1 month ago

@llvm/issue-subscribers-lld-wasm

Author: Jake Champion (JakeChampion)

The project sometimes builds and sometimes fails to build with the below backtrace I'm not sure what other information to give here, other than the backtrace - please let me know what (if any) other information would be useful to include and I will try my best to return with that information 🙇 ``` 0:12.22 PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace. 0:12.32 #0 0x000000010aba7df0 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb83df0) 0:12.32 #1 0x000000010aba835c SignalHandler(int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb8435c) 0:12.32 #2 0x000000019436b584 (/usr/lib/system/libsystem_platform.dylib+0x180477584) 0:12.32 #3 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784) 0:12.32 #4 0x0000000104f7e784 lld::wasm::InputChunk::relocate(unsigned char*) const (/Users/jakechampion/.mozbuild/clang/bin/lld+0x100322784) 0:12.32 #5 0x0000000104f97e18 lld::wasm::CustomSection::writeTo(unsigned char*) (/Users/jakechampion/.mozbuild/clang/bin/lld+0x10033be18) 0:12.32 #6 0x000000010ab41c34 std::__1::__function::__func<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0, std::__1::allocator<llvm::parallelFor(unsigned long, unsigned long, llvm::function_ref<void (unsigned long)>)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1dc34) 0:12.32 #7 0x000000010ab41b60 std::__1::__function::__func<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0, std::__1::allocator<llvm::parallel::TaskGroup::spawn(std::__1::function<void ()>, bool)::$_0>, void ()>::operator()() (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1db60) 0:12.32 #8 0x000000010ab411c4 llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::work(llvm::ThreadPoolStrategy, unsigned int) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d1c4) 0:12.33 #9 0x000000010ab4136c void* std::__1::__thread_proxy[abi:un170006]<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, llvm::parallel::detail::(anonymous namespace)::ThreadPoolExecutor::ThreadPoolExecutor(llvm::ThreadPoolStrategy)::'lambda'()::operator()() const::'lambda'()>>(void*) (/Users/jakechampion/.mozbuild/clang/lib/libLLVM.dylib+0xb1d36c) 0:12.33 #10 0x000000019433af94 (/usr/lib/system/libsystem_pthread.dylib+0x180446f94) 0:12.33 #11 0x0000000194335d34 (/usr/lib/system/libsystem_pthread.dylib+0x180441d34) ```
sbc100 commented 1 month ago

Looks like a crash in wasm-ld but I can't see much more from the backtrace. Do you have a backtrace from a LLVM build that includes asserts?

Feel free to assign to me, although I'm not sure when I'll be able to get to it.

aheejin commented 1 month ago

What's your LLVM version? Actually I think this bug might have been already fixed in #96134.

When I tried to reproduce the bug, the tip-of-tree LLVM didn't error out but the last release (LLVM 18) errored out for me. So I bisected this on the LLVM repo and #96134 was the first ~bad~good commit.

If you use LLVM from the source repo, I guess you can simply update the LLVM to the tip of the tree (or any revision equal to or later than #96134). If you use an official LLVM release, LLVM 19 should contain this.

aheejin commented 1 month ago

Will close this issue as completed. Feel free to reopen the issue if #96134 does not fix your problem.