diff options
Diffstat (limited to 'mozglue/build/AsanOptions.cpp')
| -rw-r--r-- | mozglue/build/AsanOptions.cpp | 194 |
1 files changed, 194 insertions, 0 deletions
diff --git a/mozglue/build/AsanOptions.cpp b/mozglue/build/AsanOptions.cpp new file mode 100644 index 0000000000..a2539e450d --- /dev/null +++ b/mozglue/build/AsanOptions.cpp @@ -0,0 +1,194 @@ +/* -*- 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" + +#ifndef _MSC_VER // Not supported by clang-cl yet + +// When running with AddressSanitizer, we need to explicitly set some +// options specific to our codebase to prevent errors during runtime. +// To override these, set the ASAN_OPTIONS environment variable. +// +// Currently, these are: +// +// allow_user_segv_handler=1 - Tell ASan to allow our code to use its +// own SIGSEGV handlers. This is required by ASM.js internally. +// +// alloc_dealloc_mismatch=0 - Disable alloc-dealloc mismatch checking +// in ASan. This is required because we define our own new/delete +// operators that are backed by malloc/free. If one of them gets inlined +// while the other doesn't, ASan will report false positives. +// +// detect_leaks=0 - Disable LeakSanitizer. This is required because +// otherwise leak checking will be enabled for various building and +// testing executables where we don't care much about leaks. +// +// allocator_may_return_null=1 - Tell ASan to return NULL when an allocation +// fails instead of aborting the program. This allows us to handle failing +// allocations the same way we would handle them with a regular allocator and +// also uncovers potential bugs that might occur in these situations. +// +// max_malloc_fill_size - Tell ASan to initialize memory to a certain value +// when it is allocated. This option specifies the maximum allocation size +// for which ASan should still initialize the memory. The value we specify +// here is exactly 256MiB. +// +// max_free_fill_size - Similar to max_malloc_fill_size, tell ASan to +// overwrite memory with a certain value when it is freed. Again, the value +// here specifies the maximum allocation size, larger allocations will +// skipped. +// +// malloc_fill_byte / free_fill_byte - These values specify the byte values +// used to initialize/overwrite memory in conjunction with the previous +// options max_malloc_fill_size and max_free_fill_size. The values used here +// are 0xe4 and 0xe5 to match the kAllocPoison and kAllocJunk constants used +// by mozjemalloc. +// +// malloc_context_size - This value specifies how many stack frames are +// stored for each malloc and free call. Since Firefox can have lots of deep +// stacks with allocations, we limit the default size here further to save +// some memory. +// +// fast_unwind_on_check - Use the fast (frame-pointer-based) stack unwinder +// for internal CHECK failures. The slow unwinder doesn't work on Android. +// +// fast_unwind_on_fatal - Use the fast (frame-pointer-based) stack unwinder +// to print fatal error reports. The slow unwinder doesn't work on Android. +// +// detect_stack_use_after_return=0 - Work around bug 1768099. +// +// intercept_tls_get_addr=0 - Work around +// https://github.com/google/sanitizers/issues/1322 (bug 1635327). +// +// !! Note: __asan_default_options is not used on Android! (bug 1576213) +// These should be updated in: +// mobile/android/geckoview/src/asan/resources/lib/*/wrap.sh +// +extern "C" MOZ_ASAN_BLACKLIST const char* __asan_default_options() { + return "allow_user_segv_handler=1:alloc_dealloc_mismatch=0:detect_leaks=0" +# ifdef MOZ_ASAN_REPORTER + ":malloc_context_size=20" +# endif +# ifdef __ANDROID__ + ":fast_unwind_on_check=1:fast_unwind_on_fatal=1" +# endif + ":max_free_fill_size=268435456:max_malloc_fill_size=268435456" + ":malloc_fill_byte=228:free_fill_byte=229" + ":handle_sigill=1" + ":allocator_may_return_null=1" + ":detect_stack_use_after_return=0" + ":intercept_tls_get_addr=0"; +} + +// !!! Please do not add suppressions for new leaks in Gecko code, unless they +// are intentional !!! +extern "C" const char* __lsan_default_suppressions() { + return "# Add your suppressions below\n" + + // LSan runs with a shallow stack depth and no debug symbols, so some + // small intentional leaks in system libraries show up with this. You + // do not want this enabled when running locally with a deep stack, as + // it can catch too much. + "leak:libc.so\n" + + // nsComponentManagerImpl intentionally leaks factory entries, and + // probably some other stuff. + "leak:nsComponentManagerImpl\n" + + // Bug 981220 - Pixman fails to free TLS memory. + "leak:pixman_implementation_lookup_composite\n" + + // Bug 987918 - Font shutdown leaks when CLEANUP_MEMORY is not enabled. + "leak:libfontconfig.so\n" + "leak:libfreetype.so\n" + "leak:GI___strdup\n" + // The symbol is really __GI___strdup, but if you have the leading _, + // it doesn't suppress it. + + // Bug 1078015 - If the process terminates during a PR_Sleep, LSAN + // detects a leak + "leak:PR_Sleep\n" + + // Bug 1363976 - Stylo holds some global data alive forever. + "leak:style::global_style_data\n" + "leak:style::sharing::SHARING_CACHE_KEY\n" + "leak:style::bloom::BLOOM_KEY\n" + + // + // Many leaks only affect some test suites. The suite annotations are + // not checked. + // + + // Bug 979928 - WebRTC leaks in different mochitest suites. + "leak:NR_reg_init\n" + // nr_reg_local_init should be redundant with NR_reg_init, but on + // Aurora we get fewer stack frames for some reason. + "leak:nr_reg_local_init\n" + "leak:r_log_register\n" + "leak:nr_reg_set\n" + + // This is a one-time leak in mochitest-bc, so it is probably okay to + // ignore. + "leak:GlobalPrinters::InitializeGlobalPrinters\n" + "leak:nsPSPrinterList::GetPrinterList\n" + + // Bug 1028456 - Various NSPR fd-related leaks in different mochitest + // suites. + "leak:_PR_Getfd\n" + + // Bug 1028483 - The XML parser sometimes leaks an object. Mostly + // happens in toolkit/components/thumbnails. + "leak:processInternalEntity\n" + + // Bug 1187421 - NSS does not always free the error stack in different + // mochitest suites. + "leak:nss_ClearErrorStack\n" + + // Bug 1602689 - leak at mozilla::NotNull, RacyRegisteredThread, + // RegisteredThread::RegisteredThread, mozilla::detail::UniqueSelector + "leak:RegisteredThread::RegisteredThread\n" + + // + // Leaks with system libraries in their stacks. These show up across a + // number of tests. Better symbols and disabling fast stackwalking may + // help diagnose these. + // + "leak:libcairo.so\n" + // https://github.com/OpenPrinting/cups/pull/317 + "leak:libcups.so\n" + "leak:libdl.so\n" + "leak:libdricore.so\n" + "leak:libdricore9.2.1.so\n" + "leak:libGL.so\n" + "leak:libEGL_mesa.so\n" + "leak:libglib-2.0.so\n" + "leak:libglsl.so\n" + "leak:libp11-kit.so\n" + "leak:libpixman-1.so\n" + "leak:libpulse.so\n" + // lubpulsecommon 1.1 is Ubuntu 12.04 + "leak:libpulsecommon-1.1.so\n" + // lubpulsecommon 1.1 is Ubuntu 16.04 + "leak:libpulsecommon-8.0.so\n" + "leak:libresolv.so\n" + "leak:libstdc++.so\n" + "leak:libXrandr.so\n" + "leak:libX11.so\n" + "leak:pthread_setspecific_internal\n" + "leak:swrast_dri.so\n" + + "leak:js::frontend::BytecodeEmitter:\n" + "leak:js::frontend::GeneralParser\n" + "leak:js::frontend::Parse\n" + "leak:xpc::CIGSHelper\n" + "leak:mozJSModuleLoader\n" + "leak:mozilla::xpcom::ConstructJSMComponent\n" + "leak:XPCWrappedNativeJSOps\n" + + // End of suppressions. + ; // Please keep this semicolon. +} + +#endif // _MSC_VER |