diff options
Diffstat (limited to 'mozglue/build/TsanOptions.cpp')
| -rw-r--r-- | mozglue/build/TsanOptions.cpp | 302 |
1 files changed, 302 insertions, 0 deletions
diff --git a/mozglue/build/TsanOptions.cpp b/mozglue/build/TsanOptions.cpp new file mode 100644 index 0000000000..e94f5a04d0 --- /dev/null +++ b/mozglue/build/TsanOptions.cpp @@ -0,0 +1,302 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this file, + * You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "mozilla/Attributes.h" +#include "mozilla/TsanOptions.h" + +#ifndef _MSC_VER // Not supported by clang-cl yet + +// +// When running with ThreadSanitizer, we sometimes need to suppress existing +// races. However, in any case, it should be either because +// +// 1) a bug is on file. In this case, the bug number should always be +// included with the suppression. +// +// or 2) this is an intentional race. Please be very careful with judging +// races as intentional and benign. Races in C++ are undefined behavior +// and compilers increasingly rely on exploiting this for optimizations. +// Hence, many seemingly benign races cause harmful or unexpected +// side-effects. +// +// See also: +// https://software.intel.com/en-us/blogs/2013/01/06/benign-data-races-what-could-possibly-go-wrong +// +// +// Also, when adding any race suppressions here, make sure to always add +// a signature for each of the two race stacks. Sometimes, TSan fails to +// symbolize one of the two traces and this can cause suppressed races to +// show up intermittently. +// +// clang-format off +extern "C" const char* __tsan_default_suppressions() { + return "# Add your suppressions below\n" + + // External uninstrumented libraries + MOZ_TSAN_DEFAULT_EXTLIB_SUPPRESSIONS + + // TSan internals + "race:__tsan::ProcessPendingSignals\n" + "race:__tsan::CallUserSignalHandler\n" + + + + + + // Uninstrumented code causing false positives + + // These libraries are uninstrumented and cause mutex false positives. + // However, they can be unloaded by GTK early which we cannot avoid. + "mutex:libGL.so\n" + "mutex:libGLdispatch\n" + "mutex:libGLX\n" + // Bug 1637707 - permanent + "mutex:libEGL_mesa.so\n" + // ~GLContextGLX unlocks a libGL mutex. + "mutex:GLContextGLX::~GLContextGLX\n" + // Bug 1651446 - permanent (ffmpeg) + "race:libavcodec.so*\n" + "race:libavutil.so*\n" + // For some reason, the suppressions on libpulse.so + // through `called_from_lib` only work partially. + "race:libpulse.so\n" + "race:pa_context_suspend_source_by_index\n" + "race:pa_context_unref\n" + "race:pa_format_info_set_prop_string_array\n" + "race:pa_stream_get_index\n" + "race:pa_stream_update_timing_info\n" + // This is a callback from libglib-2 that is apparently + // not fully suppressed through `called_from_lib`. + "race:g_main_context_dispatch\n" + // This is likely a false positive involving a mutex from GTK. + // See also bug 1642653 - permanent. + "mutex:GetMaiAtkType\n" + // Bug 1688716 - Failure due to fire_glxtest_process + // calling into uninstrumented external graphics driver code. + // For example: iris_dri.so and swrast_dri.so. + "race:fire_glxtest_process\n" + // Bug 1722721 - WebRender using uninstrumented Mesa drivers + "race:swrast_dri.so\n" + + + + + // Deadlock reports on single-threaded runtime. + // + // This is a known false positive from TSan where it reports + // a potential deadlock even though all mutexes are only + // taken by a single thread. For applications/tasks where we + // are absolutely sure that no second thread will be involved + // we should suppress these issues. + // + // See also https://github.com/google/sanitizers/issues/488 + + // Bug 1614605 - permanent + "deadlock:SanctionsTestServer\n" + "deadlock:OCSPStaplingServer\n" + // Bug 1643087 - permanent + "deadlock:BadCertAndPinningServer\n" + // Bug 1606804 - permanent + "deadlock:cert_storage::SecurityState::open_db\n" + "deadlock:cert_storage::SecurityState::add_certs\n" + // Bug 1651770 - permanent + "deadlock:mozilla::camera::LockAndDispatch\n" + // Bug 1606804 - permanent + "deadlock:third_party/rust/rkv/src/env.rs\n" + // Bug 1680655 - permanent + "deadlock:EncryptedClientHelloServer\n" + // Bug 1682861 - permanent + "deadlock:nsDOMWindowUtils::CompareCanvases\n" + + + + + + // Benign races in third-party code (should be fixed upstream) + + // No Bug - permanent + // No Upstream Bug Filed! + // + // SIMD Initialization in libjpeg, potentially runs + // initialization twice, but otherwise benign. Init + // routine itself is in native assembler. + "race:init_simd\n" + "race:simd_support\n" + "race:jsimd_can_ycc_rgb\n" + // Bug 1615228 - permanent + // No Upstream Bug Filed! + // + // Likely benign race in ipc/chromium/ where we set + // `message_loop_` to `NULL` on two threads when stopping + // a thread at the same time it is already finishing. + "race:base::Thread::Stop\n" + // Bug 1615569 - permanent + // No Upstream Bug Filed! + // + // NSS is using freebl from two different threads but freebl isn't + // that threadsafe. + "race:mp_exptmod.max_window_bits\n" + // Bug 1652499 - permanent + // No Upstream Bug Filed! + // + // Likely benign race in webrtc.org code - race while updating the + // minimum log severity. + "race:Loggable\n" + "race:UpdateMinLogSeverity\n" + // Bug 1652174 - permanent + // Upstream Bug: https://github.com/libevent/libevent/issues/777 + // + // Likely benign write-write race in libevent to set a sticky boolean + // flag to true. + "race:event_debug_mode_too_late\n" + + // Bug 1653618 - permanent + // Upstream Bug: https://github.com/sctplab/usrsctp/issues/507 + // + // Might lead to scheduled timers in libusrsctp getting dropped? + "race:sctp_handle_tick\n" + "race:sctp_handle_sack\n" + // Bug 1648604 - permanent + // Upstream Bug: https://github.com/sctplab/usrsctp/issues/482 + // + // Likely benign race in libusrsctp allocator during a free. + "race:system_base_info\n" + // Benign lock-order-inversion in libusrsctp + // No upstream bug filed! + "deadlock:sctp_add_to_readq\n" + + // Bug 1153409 - permanent + // No Upstream Bug Filed! + // + // Probably benign - sqlite has a few optimizations where it does + // racy reads and then does properly synchronized integrity checks + // afterwards. Some concern of compiler optimizations messing this + // up due to "volatile" being too weak for this. + "race:third_party/sqlite3/*\n" + "deadlock:third_party/sqlite3/*\n" + // Bug 1674770 - permanent + // Upstream Bug: https://github.com/Amanieu/parking_lot/issues/257 + // + // parking_lot using incorrect atomic orderings in RwLock, upstream + // fix already up for review. + "race:StrongRuleNode::ensure_child\n" + // No Bug - permanent + // Upstream Bugs: + // + // * https://github.com/rayon-rs/rayon/issues/812 + // * https://github.com/crossbeam-rs/crossbeam/issues/589 + // + // Probably a false-positive from crossbeam's deque not being + // understood by tsan. + "race:crossbeam_deque*::resize\n" + "race:crossbeam_deque*::push\n" + "race:crossbeam_deque*::write\n" + "race:crossbeam_deque*::read\n" + "race:crossbeam_deque*::steal\n" + // Bug 1805819 - permanent + // No Upstream Bug Filed! + // + // False positive in libc's tzset_internal + // See https://crbug.com/379738 also + "race:tzset_internal\n" + + + + + + // The rest of these suppressions are miscellaneous issues in gecko + // that should be investigated and ideally fixed. + + // Bug 1671574 - Permanent + // The StartupCache thread intentionally races with the main thread to + // trigger OS-level paging. It is never joined with the main thread. + "thread:StartupCache\n" + + // Bug 1734262 - Permanent + // When spawning async processes, we create a helper thread to wait for + // the process to terminate in order to asynchronously report the exit + // code to Gecko. This thread waits on a syscall for the process to end, + // which means there's no easy way to cancel and join it during Gecko + // shutdown. Suppress thread leak reports for this thread. + "thread:CreateMonitorThread\n" + + // Bug 1601600 + "race:SkARGB32_Blitter\n" + "race:SkARGB32_Shader_Blitter\n" + "race:SkARGB32_Opaque_Blitter\n" + "race:SkRasterPipelineBlitter\n" + "race:Clamp_S32_D32_nofilter_trans_shaderproc\n" + "race:SkSpriteBlitter_Memcpy\n" + + // Bug 1606800 + "race:CallInitFunc\n" + + // Bug 1606803 + "race:ipv6_is_present\n" + + // Bug 1615123 + "race:_dl_deallocate_tls\n" + "race:__libc_memalign\n" + + // Bug 1664803 + "race:Sampler::sSigHandlerCoordinator\n" + + // Bug 1656068 + "race:WebRtcAec_Create\n" + + // No Bug - Logging bug in Mochitests + "race:mochitest/ssltunnel/ssltunnel.cpp\n" + + // This thread does not seem to be stopped/joined. + // ImageBridgeChild should be turned back into a background + // task queue in bug 1647628, in which case these suppressions + // can be removed. + "race:mozilla::layers::ImageBridgeChild::ShutDown\n" + + // Bug 1652530 + "mutex:XErrorTrap\n" + + // Bug 1671601 + "race:CamerasParent::ActorDestroy\n" + "race:CamerasParent::DispatchToVideoCaptureThread\n" + + // Bug 1623541 + "race:VRShMem::PullSystemState\n" + "race:VRShMem::PushSystemState\n" + "race:VRShMem::PullBrowserState\n" + "race:VRShMem::PushBrowserState\n" + + // Bug 1682951 + "race:storage::Connection::Release\n" + + // Bug 1683357 + "race:image::ImageSurfaceCache::SuggestedSizeInternal\n" + "race:image::RasterImage::SetMetadata\n" + "race:image::RasterImage::GetWidth\n" + + // Bug 1722721 - This is a benign race creating worker/SW compositor threads. + "race:webrender::profiler::register_thread\n" + + // Bug 1722721 - This is a false positive during SW-WR rendering. + "race:scale_blit\n" + + "race:mozilla::gl::MesaMemoryLeakWorkaround\n" + + // Bug 1733908 + "race:js::wasm::Code::bestTier\n" + "race:js::wasm::Code::commitTier2\n" + "race:js::wasm::Code::setTier2\n" + "race:js::wasm::Code::setAndBorrowTier2\n" + + // Bug 1755449 + // The Glean init thread is used to perform I/O and other blocking operations. + // It is never joined with the main thread, but this is being re-evaluated. + "thread:glean::initialize\n" + + // End of suppressions. + ; // Please keep this semicolon. +} +// clang-format on +#endif // _MSC_VER |