forked from llvm/llvm-project
-
Notifications
You must be signed in to change notification settings - Fork 1
[LLDB][GPU] Connect synchronously to the gpu process #10
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Merged
walter-erquinigo
merged 1 commit into
llvm-server-plugins
from
clayborg/walter/sync-connection
Jul 1, 2025
Merged
[LLDB][GPU] Connect synchronously to the gpu process #10
walter-erquinigo
merged 1 commit into
llvm-server-plugins
from
clayborg/walter/sync-connection
Jul 1, 2025
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
As preliminary context, everything belows applies to the client. The server side is all fine. For NVIDIA, after connecting to the GPU process we issue a resume_gpu GPUAction. We were doing that after connecting to the process asynchronously, which resulted in a race condition in which the following actions were happening at the same time: - the gpu process was handling its stop state, fetching the state of its threads - the GPUAction was issuing a Resume request this caused that very often, after resuming, the gpu process was left in a Running state but with the thread metadata not cleared. Then, in the next stop event, it would reuse the previous thread metadata instead of using the new one. Not only that, this sometimes caused many event listeners that were processing the original stop event to not complete their job after the GPUAction issued the Resume request. A simple way to fix this is to connect to the gpu process synchronously. This ensures that all the stop event handlers have been processed and that we can safely issue the Resume without race conditions. A better fix would be to add a WaitForProcessToStop right before processing GPUActions, but I wasn't able to make it work that way. I'm too ignorant.
clayborg
approved these changes
Jul 1, 2025
pveras
pushed a commit
to pveras/llvm-project
that referenced
this pull request
Oct 10, 2025
A recent change adding a new sanitizer kind (via Sanitizers.def) was reverted in c74fa20 ("Revert "[Clang][CodeGen] Introduce the AllocToken SanitizerKind" (llvm#162413)"). The reason was this ASan report, when running the test cases in clang/test/Preprocessor/print-header-json.c: ``` ==clang==483265==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x7d82b97e8b58 at pc 0x562cd432231f bp 0x7fff3fad0850 sp 0x7fff3fad0848 READ of size 16 at 0x7d82b97e8b58 thread T0 #0 0x562cd432231e in __copy_non_overlapping_range<const unsigned long *, const unsigned long *> zorg-test/libcxx_install_asan_ubsan/include/c++/v1/string:2144:38 clayborg#1 0x562cd432231e in void std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>::__init_with_size[abi:nn220000]<unsigned long const*, unsigned long const*>(unsigned long const*, unsigned long const*, unsigned long) zorg-test/libcxx_install_asan_ubsan/include/c++/v1/string:2685:18 clayborg#2 0x562cd41e2797 in __init<const unsigned long *, 0> zorg-test/libcxx_install_asan_ubsan/include/c++/v1/string:2673:3 clayborg#3 0x562cd41e2797 in basic_string<const unsigned long *, 0> zorg-test/libcxx_install_asan_ubsan/include/c++/v1/string:1174:5 clayborg#4 0x562cd41e2797 in clang::ASTReader::ReadString(llvm::SmallVectorImpl<unsigned long> const&, unsigned int&) clang/lib/Serialization/ASTReader.cpp:10171:15 clayborg#5 0x562cd41fd89a in clang::ASTReader::ParseLanguageOptions(llvm::SmallVector<unsigned long, 64u> const&, llvm::StringRef, bool, clang::ASTReaderListener&, bool) clang/lib/Serialization/ASTReader.cpp:6475:28 clayborg#6 0x562cd41eea53 in clang::ASTReader::ReadOptionsBlock(llvm::BitstreamCursor&, llvm::StringRef, unsigned int, bool, clang::ASTReaderListener&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>&) clang/lib/Serialization/ASTReader.cpp:3069:11 clayborg#7 0x562cd4204ab8 in clang::ASTReader::ReadControlBlock(clang::serialization::ModuleFile&, llvm::SmallVectorImpl<clang::ASTReader::ImportedModule>&, clang::serialization::ModuleFile const*, unsigned int) clang/lib/Serialization/ASTReader.cpp:3249:15 clayborg#8 0x562cd42097d2 in clang::ASTReader::ReadASTCore(llvm::StringRef, clang::serialization::ModuleKind, clang::SourceLocation, clang::serialization::ModuleFile*, llvm::SmallVectorImpl<clang::ASTReader::ImportedModule>&, long, long, clang::ASTFileSignature, unsigned int) clang/lib/Serialization/ASTReader.cpp:5182:15 clayborg#9 0x562cd421ec77 in clang::ASTReader::ReadAST(llvm::StringRef, clang::serialization::ModuleKind, clang::SourceLocation, unsigned int, clang::serialization::ModuleFile**) clang/lib/Serialization/ASTReader.cpp:4828:11 clayborg#10 0x562cd3d07b74 in clang::CompilerInstance::findOrCompileModuleAndReadAST(llvm::StringRef, clang::SourceLocation, clang::SourceLocation, bool) clang/lib/Frontend/CompilerInstance.cpp:1805:27 clayborg#11 0x562cd3d0b2ef in clang::CompilerInstance::loadModule(clang::SourceLocation, llvm::ArrayRef<clang::IdentifierLoc>, clang::Module::NameVisibilityKind, bool) clang/lib/Frontend/CompilerInstance.cpp:1956:31 clayborg#12 0x562cdb04eb1c in clang::Preprocessor::HandleHeaderIncludeOrImport(clang::SourceLocation, clang::Token&, clang::Token&, clang::SourceLocation, clang::detail::SearchDirIteratorImpl<true>, clang::FileEntry const*) clang/lib/Lex/PPDirectives.cpp:2423:49 clayborg#13 0x562cdb042222 in clang::Preprocessor::HandleIncludeDirective(clang::SourceLocation, clang::Token&, clang::detail::SearchDirIteratorImpl<true>, clang::FileEntry const*) clang/lib/Lex/PPDirectives.cpp:2101:17 clayborg#14 0x562cdb043366 in clang::Preprocessor::HandleDirective(clang::Token&) clang/lib/Lex/PPDirectives.cpp:1338:14 clayborg#15 0x562cdafa84bc in clang::Lexer::LexTokenInternal(clang::Token&, bool) clang/lib/Lex/Lexer.cpp:4512:7 clayborg#16 0x562cdaf9f20b in clang::Lexer::Lex(clang::Token&) clang/lib/Lex/Lexer.cpp:3729:24 clayborg#17 0x562cdb0d4ffa in clang::Preprocessor::Lex(clang::Token&) clang/lib/Lex/Preprocessor.cpp:896:11 clayborg#18 0x562cd77da950 in clang::ParseAST(clang::Sema&, bool, bool) clang/lib/Parse/ParseAST.cpp:163:7 [...] 0x7d82b97e8b58 is located 0 bytes after 3288-byte region [0x7d82b97e7e80,0x7d82b97e8b58) allocated by thread T0 here: #0 0x562cca76f604 in malloc zorg-test/llvm-project/compiler-rt/lib/asan/asan_malloc_linux.cpp:67:3 clayborg#1 0x562cd1cce452 in safe_malloc llvm/include/llvm/Support/MemAlloc.h:26:18 clayborg#2 0x562cd1cce452 in llvm::SmallVectorBase<unsigned int>::grow_pod(void*, unsigned long, unsigned long) llvm/lib/Support/SmallVector.cpp:151:15 clayborg#3 0x562cdbe1768b in grow_pod llvm/include/llvm/ADT/SmallVector.h:139:11 clayborg#4 0x562cdbe1768b in grow llvm/include/llvm/ADT/SmallVector.h:525:41 clayborg#5 0x562cdbe1768b in reserve llvm/include/llvm/ADT/SmallVector.h:665:13 clayborg#6 0x562cdbe1768b in llvm::BitstreamCursor::readRecord(unsigned int, llvm::SmallVectorImpl<unsigned long>&, llvm::StringRef*) llvm/lib/Bitstream/Reader/BitstreamReader.cpp:230:10 clayborg#7 0x562cd41ee8ab in clang::ASTReader::ReadOptionsBlock(llvm::BitstreamCursor&, llvm::StringRef, unsigned int, bool, clang::ASTReaderListener&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>&) clang/lib/Serialization/ASTReader.cpp:3060:49 clayborg#8 0x562cd4204ab8 in clang::ASTReader::ReadControlBlock(clang::serialization::ModuleFile&, llvm::SmallVectorImpl<clang::ASTReader::ImportedModule>&, clang::serialization::ModuleFile const*, unsigned int) clang/lib/Serialization/ASTReader.cpp:3249:15 clayborg#9 0x562cd42097d2 in clang::ASTReader::ReadASTCore(llvm::StringRef, clang::serialization::ModuleKind, clang::SourceLocation, clang::serialization::ModuleFile*, llvm::SmallVectorImpl<clang::ASTReader::ImportedModule>&, long, long, clang::ASTFileSignature, unsigned int) clang/lib/Serialization/ASTReader.cpp:5182:15 clayborg#10 0x562cd421ec77 in clang::ASTReader::ReadAST(llvm::StringRef, clang::serialization::ModuleKind, clang::SourceLocation, unsigned int, clang::serialization::ModuleFile**) clang/lib/Serialization/ASTReader.cpp:4828:11 clayborg#11 0x562cd3d07b74 in clang::CompilerInstance::findOrCompileModuleAndReadAST(llvm::StringRef, clang::SourceLocation, clang::SourceLocation, bool) clang/lib/Frontend/CompilerInstance.cpp:1805:27 clayborg#12 0x562cd3d0b2ef in clang::CompilerInstance::loadModule(clang::SourceLocation, llvm::ArrayRef<clang::IdentifierLoc>, clang::Module::NameVisibilityKind, bool) clang/lib/Frontend/CompilerInstance.cpp:1956:31 clayborg#13 0x562cdb04eb1c in clang::Preprocessor::HandleHeaderIncludeOrImport(clang::SourceLocation, clang::Token&, clang::Token&, clang::SourceLocation, clang::detail::SearchDirIteratorImpl<true>, clang::FileEntry const*) clang/lib/Lex/PPDirectives.cpp:2423:49 clayborg#14 0x562cdb042222 in clang::Preprocessor::HandleIncludeDirective(clang::SourceLocation, clang::Token&, clang::detail::SearchDirIteratorImpl<true>, clang::FileEntry const*) clang/lib/Lex/PPDirectives.cpp:2101:17 clayborg#15 0x562cdb043366 in clang::Preprocessor::HandleDirective(clang::Token&) clang/lib/Lex/PPDirectives.cpp:1338:14 clayborg#16 0x562cdafa84bc in clang::Lexer::LexTokenInternal(clang::Token&, bool) clang/lib/Lex/Lexer.cpp:4512:7 clayborg#17 0x562cdaf9f20b in clang::Lexer::Lex(clang::Token&) clang/lib/Lex/Lexer.cpp:3729:24 clayborg#18 0x562cdb0d4ffa in clang::Preprocessor::Lex(clang::Token&) clang/lib/Lex/Preprocessor.cpp:896:11 clayborg#19 0x562cd77da950 in clang::ParseAST(clang::Sema&, bool, bool) clang/lib/Parse/ParseAST.cpp:163:7 [...] SUMMARY: AddressSanitizer: heap-buffer-overflow clang/lib/Serialization/ASTReader.cpp:10171:15 in clang::ASTReader::ReadString(llvm::SmallVectorImpl<unsigned long> const&, unsigned int&) ``` The reason is this particular RUN line: ``` // RUN: env CC_PRINT_HEADERS_FORMAT=json CC_PRINT_HEADERS_FILTERING=direct-per-file CC_PRINT_HEADERS_FILE=%t.txt %clang -fsyntax-only -I %S/Inputs/print-header-json -isystem %S/Inputs/print-header-json/system -fmodules -fimplicit-module-maps -fmodules-cache-path=%t %s -o /dev/null ``` which was added in 8df194f ("[Clang] Support includes translated to module imports in -header-include-filtering=direct-per-file (llvm#156756)"). The problem is caused by an incremental build reusing stale cached module files (.pcm) that are no longer binary-compatible with the updated compiler. Adding a new sanitizer option altered the implicit binary layout of the serialized LangOptions data structure. The build + test system is oblivious to such changes. When the new compiler attempted to read the old module file (from the previous test invocation), it misinterpreted the data due to the layout mismatch, resulting in a heap-buffer-overflow. Unfortunately Clang's PCM format does not encode nor detect version mismatches here; a more graceful failure mode would be preferable. For now, fix the test to be more robust with incremental build + test.
clayborg
pushed a commit
that referenced
this pull request
Nov 14, 2025
…am (llvm#167724) This got exposed by `09262656f32ab3f2e1d82e5342ba37eecac52522`. The underlying stream of `m_os` is referenced by the `TextDiagnostic` member of `TextDiagnosticPrinter`. It got turned into a `llvm::formatted_raw_ostream` in the commit above. When `~TextDiagnosticPrinter` (and thus `~TextDiagnostic`) is invoked, we now call `~formatted_raw_ostream`, which tries to access the underlying stream. But `m_os` was already deleted because it is earlier in the order of destruction in `TextDiagnosticPrinter`. Move the `m_os` member before the `TextDiagnosticPrinter` to avoid a use-after-free. Drive-by: * Also move the `m_output` member which the `m_os` holds a reference to. The fact it's a reference indicates the expectation is most likely that the string outlives the stream. The ASAN macOS bot is currently failing with this: ``` 08:15:39 ================================================================= 08:15:39 ==61103==ERROR: AddressSanitizer: heap-use-after-free on address 0x60600012cf40 at pc 0x00012140d304 bp 0x00016eecc850 sp 0x00016eecc848 08:15:39 READ of size 8 at 0x60600012cf40 thread T0 08:15:39 #0 0x00012140d300 in llvm::formatted_raw_ostream::releaseStream() FormattedStream.h:205 08:15:39 #1 0x00012140d3a4 in llvm::formatted_raw_ostream::~formatted_raw_ostream() FormattedStream.h:145 08:15:39 #2 0x00012604abf8 in clang::TextDiagnostic::~TextDiagnostic() TextDiagnostic.cpp:721 08:15:39 #3 0x00012605dc80 in clang::TextDiagnosticPrinter::~TextDiagnosticPrinter() TextDiagnosticPrinter.cpp:30 08:15:39 #4 0x00012605dd5c in clang::TextDiagnosticPrinter::~TextDiagnosticPrinter() TextDiagnosticPrinter.cpp:27 08:15:39 #5 0x0001231fb210 in (anonymous namespace)::StoringDiagnosticConsumer::~StoringDiagnosticConsumer() ClangModulesDeclVendor.cpp:47 08:15:39 #6 0x0001231fb3bc in (anonymous namespace)::StoringDiagnosticConsumer::~StoringDiagnosticConsumer() ClangModulesDeclVendor.cpp:47 08:15:39 #7 0x000129aa9d70 in clang::DiagnosticsEngine::~DiagnosticsEngine() Diagnostic.cpp:91 08:15:39 #8 0x0001230436b8 in llvm::RefCountedBase<clang::DiagnosticsEngine>::Release() const IntrusiveRefCntPtr.h:103 08:15:39 #9 0x0001231fe6c8 in (anonymous namespace)::ClangModulesDeclVendorImpl::~ClangModulesDeclVendorImpl() ClangModulesDeclVendor.cpp:93 08:15:39 #10 0x0001231fe858 in (anonymous namespace)::ClangModulesDeclVendorImpl::~ClangModulesDeclVendorImpl() ClangModulesDeclVendor.cpp:93 ... 08:15:39 08:15:39 0x60600012cf40 is located 32 bytes inside of 56-byte region [0x60600012cf20,0x60600012cf58) 08:15:39 freed by thread T0 here: 08:15:39 #0 0x0001018abb88 in _ZdlPv+0x74 (libclang_rt.asan_osx_dynamic.dylib:arm64e+0x4bb88) 08:15:39 #1 0x0001231fb1c0 in (anonymous namespace)::StoringDiagnosticConsumer::~StoringDiagnosticConsumer() ClangModulesDeclVendor.cpp:47 08:15:39 #2 0x0001231fb3bc in (anonymous namespace)::StoringDiagnosticConsumer::~StoringDiagnosticConsumer() ClangModulesDeclVendor.cpp:47 08:15:39 #3 0x000129aa9d70 in clang::DiagnosticsEngine::~DiagnosticsEngine() Diagnostic.cpp:91 08:15:39 #4 0x0001230436b8 in llvm::RefCountedBase<clang::DiagnosticsEngine>::Release() const IntrusiveRefCntPtr.h:103 08:15:39 #5 0x0001231fe6c8 in (anonymous namespace)::ClangModulesDeclVendorImpl::~ClangModulesDeclVendorImpl() ClangModulesDeclVendor.cpp:93 08:15:39 #6 0x0001231fe858 in (anonymous namespace)::ClangModulesDeclVendorImpl::~ClangModulesDeclVendorImpl() ClangModulesDeclVendor.cpp:93 ... 08:15:39 08:15:39 previously allocated by thread T0 here: 08:15:39 #0 0x0001018ab760 in _Znwm+0x74 (libclang_rt.asan_osx_dynamic.dylib:arm64e+0x4b760) 08:15:39 #1 0x0001231f8dec in lldb_private::ClangModulesDeclVendor::Create(lldb_private::Target&) ClangModulesDeclVendor.cpp:732 08:15:39 #2 0x00012320af58 in lldb_private::ClangPersistentVariables::GetClangModulesDeclVendor() ClangPersistentVariables.cpp:124 08:15:39 #3 0x0001232111f0 in lldb_private::ClangUserExpression::PrepareForParsing(lldb_private::DiagnosticManager&, lldb_private::ExecutionContext&, bool) ClangUserExpression.cpp:536 08:15:39 #4 0x000123213790 in lldb_private::ClangUserExpression::Parse(lldb_private::DiagnosticManager&, lldb_private::ExecutionContext&, lldb_private::ExecutionPolicy, bool, bool) ClangUserExpression.cpp:647 08:15:39 #5 0x00012032b258 in lldb_private::UserExpression::Evaluate(lldb_private::ExecutionContext&, lldb_private::EvaluateExpressionOptions const&, llvm::StringRef, llvm::StringRef, std::__1::shared_ptr<lldb_private::ValueObject>&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>*, lldb_private::ValueObject*) UserExpression.cpp:280 08:15:39 #6 0x000120724010 in lldb_private::Target::EvaluateExpression(llvm::StringRef, lldb_private::ExecutionContextScope*, std::__1::shared_ptr<lldb_private::ValueObject>&, lldb_private::EvaluateExpressionOptions const&, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>>*, lldb_private::ValueObject*) Target.cpp:2905 08:15:39 #7 0x00011fc7bde0 in lldb::SBTarget::EvaluateExpression(char const*, lldb::SBExpressionOptions const&) SBTarget.cpp:2305 08:15:39 ==61103==ABORTING ... ```
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
As preliminary context, everything belows applies to the client. The server side is all fine.
For NVIDIA, after connecting to the GPU process we issue a resume_gpu GPUAction. We were doing that after connecting to the process asynchronously, which resulted in a race condition in which the following actions were happening at the same time:
this caused that very often, after resuming, the gpu process was left in a Running state but with the thread metadata not cleared. Then, in the next stop event, it would reuse the previous thread metadata instead of using the new one.
Not only that, this sometimes caused many event listeners that were processing the original stop event to not complete their job after the GPUAction issued the Resume request.
A simple way to fix this is to connect to the gpu process synchronously. This ensures that all the stop event handlers have been processed and that we can safely issue the Resume without race conditions.
A better fix would be to add a WaitForProcessToStop right before processing GPUActions, but I wasn't able to make it work that way. I'm too ignorant.