Interceptors using ScopedInteceptor should never call into user's code before the ScopedInterceptor is out of scope (and its destructor is called). Let's add a DCHECK to enforce that.
Differential Revision: http://reviews.llvm.org/D15381
llvm-svn: 255996
Some interceptors in tsan_libdispatch_mac.cc currently wrongly use TSAN_SCOPED_INTERCEPTOR/ScopedInterceptor. Its constructor can start ignoring memory accesses, and the destructor the stops this -- however, e.g. dispatch_sync can call user's code, so the ignoring will extend to user's code as well. This is not expected and we should only limit the scope of ScopedInterceptor to TSan code. This patch introduces annotations that mark the beginning and ending of a callback into user's code.
Differential Revision: http://reviews.llvm.org/D15419
llvm-svn: 255995
check_memcpy test added in r254959 fails on some configurations due to
memcpy() calls inserted by Clang. Try harder to avoid them by using
internal_memcpy() where applicable.
llvm-svn: 255287
Summary:
Android doesn't have __libc_malloc and related allocation
functions. As its dynamic linker doesn't use malloc, so
we can use REAL(malloc) to replace __libc_malloc safely.
Reviewers: kcc, eugenis, dvyukov
Subscribers: llvm-commits, tberghammer, danalbert, srhines
Differential Revision: http://reviews.llvm.org/D15297
llvm-svn: 255167
This patch provides the assembly support for setjmp/longjmp for use
with the thread sanitizer. This is a big more complicated than for
aarch64, because sibcalls are only legal under our ABIs if the TOC
pointer is unchanged. Since the true setjmp function trashes the TOC
pointer, and we have to leave the stack in a correct state, we emulate
the setjmp function rather than branching to it.
We also need to materialize the TOC for cases where the _setjmp code
is called from libc. This is done differently under the ELFv1 and
ELFv2 ABIs.
llvm-svn: 255059
This patch is by Simone Atzeni with portions by Adhemerval Zanella.
This contains the LLVM patches to enable the thread sanitizer for
PPC64, both big- and little-endian. Two different virtual memory
sizes are supported: Old kernels use a 44-bit address space, while
newer kernels require a 46-bit address space.
There are two companion patches that will be added shortly. There is
a Clang patch to actually turn on the use of the thread sanitizer for
PPC64. There is also a patch that I wrote to provide interceptor
support for setjmp/longjmp on PPC64.
Patch discussion at reviews.llvm.org/D12841.
llvm-svn: 255057
Another attempt at fixing tsan_invisible_barrier.
Current implementation causes:
https://llvm.org/bugs/show_bug.cgi?id=25643
There were several unsuccessful iterations for this functionality:
Initially it was implemented in user code using REAL(pthread_barrier_wait). But pthread_barrier_wait is not supported on MacOS. Futexes are linux-specific for this matter.
Then we switched to atomics+usleep(10). But usleep produced parasitic "as-if synchronized via sleep" messages in reports which failed some output tests.
Then we switched to atomics+sched_yield. But this produced tons of tsan- visible events, which lead to "failed to restore stack trace" failures.
Move implementation into runtime and use internal_sched_yield in the wait loop.
This way tsan should see no events from the barrier, so not trace overflows and
no "as-if synchronized via sleep" messages.
llvm-svn: 255030
1) There's a few wrongly defined things in tsan_interceptors.cc,
2) a typo in tsan_rtl_amd64.S which calls setjmp instead of sigsetjmp in the interceptor, and
3) on OS X, accessing an mprotected page results in a SIGBUS (and not SIGSEGV).
Differential Revision: http://reviews.llvm.org/D15052
llvm-svn: 254299
This patch ports the assembly file tsan_rtl_amd64.S to OS X, where we need several changes:
* Some assembler directives are not available on OS X (.hidden, .type, .size)
* Symbol names need to start with an underscore (added a ASM_TSAN_SYMBOL macro for that).
* To make the interceptors work, we ween to name the function "_wrap_setjmp" (added ASM_TSAN_SYMBOL_INTERCEPTOR for that).
* Calling the original setjmp is done with a simple "jmp _setjmp".
* __sigsetjmp doesn't exist on OS X.
Differential Revision: http://reviews.llvm.org/D14947
llvm-svn: 254228
This patch fixes the __cxa_guard_acquire, __cxa_guard_release and __cxa_guard_abort interceptors on OS X. They apparently work on Linux just by having the same name, but on OS X, we actually need to use TSAN_INTERCEPTOR.
Differential Revision: http://reviews.llvm.org/D14868
llvm-svn: 253776
On OS X, the thread finalization is fragile due to thread-local variables destruction order. I've seen cases where the we destroy the ThreadState too early and subsequent thread-local values' destructors call interceptors again. Let's replace the TLV-based thread finalization method with libpthread hooks. The notification PTHREAD_INTROSPECTION_THREAD_TERMINATE is called *after* all TLVs have been destroyed.
Differential Revision: http://reviews.llvm.org/D14777
llvm-svn: 253560
Reimplement dispatch_once in an interceptor to solve these issues that may produce false positives with TSan on OS X:
1) there is a racy load inside an inlined part of dispatch_once,
2) the fast path in dispatch_once doesn't perform an acquire load, so we don't properly synchronize the initialization and subsequent uses of whatever is initialized,
3) dispatch_once is already used in a lot of already-compiled code, so TSan doesn't see the inlined fast-path.
This patch uses a trick to avoid ever taking the fast path (by never storing ~0 into the predicate), which means the interceptor will always be called even from already-compiled code. Within the interceptor, our own atomic reads and writes are not written into shadow cells, so the race in the inlined part is not reported (because the accesses are only loads).
Differential Revision: http://reviews.llvm.org/D14811
llvm-svn: 253552
This patch adds assembly routines to enable setjmp/longjmp for aarch64
on linux. It fixes:
* test/tsan/longjmp2.cc
* test/tsan/longjmp3.cc
* test/tsan/longjmp4.cc
* test/tsan/signal_longjmp.cc
I also checked with perlbench from specpu2006 (it fails to run
with missing setjmp/longjmp intrumentation).
llvm-svn: 253205
Fixing `tsan_interceptors.cc`, which on OS X produces a bunch of warnings about unused constants and functions.
Differential Revision: http://reviews.llvm.org/D14381
llvm-svn: 252165
On OS X, memcpy and memmove are actually aliases of the same implementation, which means the interceptor of memcpy is also invoked when memmove is called. The current implementation of the interceptor uses `internal_memcpy` to perform the actual memory operation, which can produce an incorrect result when memmove semantics are expected. Let's call `internal_memmove` instead.
Differential Revision: http://reviews.llvm.org/D14336
llvm-svn: 252162
A call to memmove is used early during new thread initialization on OS X. This patch uses the `COMMON_INTERCEPTOR_NOTHING_IS_INITIALIZED` check, similarly to how we deal with other early-used interceptors.
Differential Revision: http://reviews.llvm.org/D14377
llvm-svn: 252161
TSan has a re-implementation of `pthread_once` in its interceptor, which assumes that the `pthread_once_t *once_control` pointer is actually pointing to a "storage" which is zero-initialized and used for the atomic operations. However, that's not true on OS X, where pthread_once_t is a structure, that contains a header (with a magic value) and the actual storage follows after that. This patch skips the header to make the interceptor work on OS X.
Differential Revision: http://reviews.llvm.org/D14379
llvm-svn: 252160
This implements a "poor man's TLV" to be used for TSan's ThreadState on OS X. Based on the fact that `pthread_self()` is always available and reliable and returns a valid pointer to memory, we'll use the shadow memory of this pointer as a thread-local storage. No user code should ever read/write to this internal libpthread structure, so it's safe to use it for this purpose. We lazily allocate the ThreadState object and store the pointer here.
Differential Revision: http://reviews.llvm.org/D14288
llvm-svn: 252159
TSan needs to use a custom malloc zone on OS X, which is already implemented in ASan. This patch uses the sanitizer_common implementation in `sanitizer_malloc_mac.inc` for TSan as well.
Reviewed at http://reviews.llvm.org/D14330
llvm-svn: 252155
On OS X, GCD worker threads are created without a call to pthread_create. We need to properly register these threads with ThreadCreate and ThreadStart. This patch uses a libpthread API (`pthread_introspection_hook_install`) to get notifications about new threads and about threads that are about to be destroyed.
Differential Revision: http://reviews.llvm.org/D14328
llvm-svn: 252049
This patch modifies `tsan_interceptors.cc` to be buildable on OS X. Several of the intercepted methods are not available on OS X, so we need to `#if !SANITIZER_MAC` them. Plus a few other fixes, e.g. `pthread_yield` doesn't exist, let's use `internal_sched_yield` instead.
This is part of an effort to port TSan to OS X, and it's one the very first steps. Don't expect TSan on OS X to actually work or pass tests at this point.
Differential Revision: http://reviews.llvm.org/D14237
llvm-svn: 251916
Race deduplication code proved to be a performance bottleneck in the past if suppressions/annotations are used, or just some races left unaddressed. And we still get user complaints about this:
https://groups.google.com/forum/#!topic/thread-sanitizer/hB0WyiTI4e4
ReportRace already has several layers of caching for racy pcs/addresses to make deduplication faster. However, ReportRace still takes a global mutex (ThreadRegistry and ReportMutex) during deduplication and also calls mmap/munmap (which take process-wide semaphore in kernel), this makes deduplication non-scalable.
This patch moves race deduplication outside of global mutexes and also removes all mmap/munmap calls.
As the result, race_stress.cc with 100 threads and 10000 iterations become 30x faster:
before:
real 0m21.673s
user 0m5.932s
sys 0m34.885s
after:
real 0m0.720s
user 0m23.646s
sys 0m1.254s
http://reviews.llvm.org/D12554
llvm-svn: 246758
This patch enabled TSAN for aarch64 with 39-bit VMA layout. As defined by
tsan_platform.h the layout used is:
0000 4000 00 - 0200 0000 00: main binary
2000 0000 00 - 4000 0000 00: shadow memory
4000 0000 00 - 5000 0000 00: metainfo
5000 0000 00 - 6000 0000 00: -
6000 0000 00 - 6200 0000 00: traces
6200 0000 00 - 7d00 0000 00: -
7d00 0000 00 - 7e00 0000 00: heap
7e00 0000 00 - 7fff ffff ff: modules and main thread stack
Which gives it about 8GB for main binary, 4GB for heap and 8GB for
modules and main thread stack.
Most of tests are passing, with the exception of:
* ignore_lib0, ignore_lib1, ignore_lib3 due a kernel limitation for
no support to make mmap page non-executable.
* longjmp tests due missing specialized assembly routines.
These tests are xfail for now.
The only tsan issue still showing is:
rtl/TsanRtlTest/Posix.ThreadLocalAccesses
Which still required further investigation. The test is disable for
aarch64 for now.
llvm-svn: 244055
Summary:
PTHREAD_CREATE_DETACHED has a different value on Mac OS X. Since both
PTHREAD_CREATE_JOINABLE and PTHREAD_CREATE_DETACHED are non-zero,
`__tsan::ThreadCreate` always creates detached threads.
Reviewers: kcc, samsonov, glider
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D10606
llvm-svn: 243151
POSIX states that "It shall be safe to destroy an initialized condition
variable upon which no threads are currently blocked", and later clarifies
"A condition variable can be destroyed immediately after all the threads
that are blocked on it are awakened) (in examples section). Tsan reported
such destruction as a data race.
Fixes https://llvm.org/bugs/show_bug.cgi?id=23616
Reviewed in http://reviews.llvm.org/D10693
llvm-svn: 241082
Previously tsan modelled dup2(oldfd, newfd) as write on newfd.
We hit several cases where the write lead to false positives:
1. Some software dups a closed pipe in place of a socket before closing
the socket (to prevent races actually).
2. Some daemons dup /dev/null in place of stdin/stdout.
On the other hand we have not seen cases when write here catches real bugs.
So model dup2 as read on newfd instead.
llvm-svn: 240687
We see false reports between dlopen and dl_iterate_phdr.
This happens because tsan does not see dynamic linker
internal synchronization. Unpoison module names
in dl_iterate_phdr callback.
llvm-svn: 240576
Current code tries to find the dynamic TLS header to the left of the
TLS block without checking that it's not a static TLS allocation.
llvm-svn: 237495
Fixes https://llvm.org/bugs/show_bug.cgi?id=23235
If pthread_create is followed by pthread_detach,
the new thread may not acquire synchronize with
the parent thread.
llvm-svn: 235293
This patch is related to Issue 346: moar string interceptors: strstr, strcasestr, strcspn, strpbrk
As was suggested in original review http://reviews.llvm.org/D6056 a new "strict_string_checks" run-time flag introduced.
The flag support applied for existing common, asan, msan and tsan interceptors. New asan tests added.
Change by Maria Guseva reviewed in http://reviews.llvm.org/D7123
llvm-svn: 234187
MetaMap::ResetRange/FreeRange rounds the range up to at least kMetaShadowSize.
This is requried for e.g. free(malloc(0)). However, munmap returns EINVAL
and do not unmap any memory when length arguments is equal to 0.
So don't free meta shadow in this case as well.
llvm-svn: 234145
Munmap interceptor did not reset meta shadow for the range,
and __tsan_java_move crashed because it encountered
non-zero meta shadow for the destination.
llvm-svn: 232029
The problem is that without SA_RESTORER flag, kernel ignores the handler. So tracer actually did not setup any handler.
Add SA_RESTORER flag when setting up handlers.
Add a test that causes SIGSEGV in stoptheworld callback.
Move SignalContext from asan to sanitizer_common to print better diagnostics about signal in the tracer thread.
http://reviews.llvm.org/D8005
llvm-svn: 230978
SuppressionContext is no longer a singleton, shared by all sanitizers,
but a regular class. Each of ASan, LSan, UBSan and TSan now have their
own SuppressionContext, which only parses suppressions specific to
that sanitizer.
"suppressions" flag is moved away from common flags into tool-specific
flags, so the user now may pass
ASAN_OPTIONS=suppressions=asan_supp.txt LSAN_OPIONS=suppressions=lsan_supp.txt
in a single invocation.
llvm-svn: 230026
Let each LibIgnore user (for now it's only TSan) manually go
through SuppressionContext and pass ignored library templates to
LibIgnore.
llvm-svn: 229924
signal handler reads sa_sigaction when a concurrent sigaction call can modify it
as the result in could try to call SIG_DFL or a partially overwritten function pointer
llvm-svn: 224530
Summary:
Turn "allocator_may_return_null" common flag into an
Allocator::may_return_null bool flag. We want to make sure
that common flags are immutable after initialization. There
are cases when we want to change this flag in the allocator
at runtime: e.g. in unit tests and during ASan activation
on Android.
Test Plan: regression test suite, real-life applications
Reviewers: kcc, eugenis
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6623
llvm-svn: 224148
Summary:
This change removes `__tsan::StackTrace` class. There are
now three alternatives:
# Lightweight `__sanitizer::StackTrace`, which doesn't own a buffer
of PCs. It is used in functions that need stack traces in read-only
mode, and helps to prevent unnecessary allocations/copies (e.g.
for StackTraces fetched from StackDepot).
# `__sanitizer::BufferedStackTrace`, which stores buffer of PCs in
a constant array. It is used in TraceHeader (non-Go version)
# `__tsan::VarSizeStackTrace`, which owns buffer of PCs, dynamically
allocated via TSan internal allocator.
Test Plan: compiler-rt test suite
Reviewers: dvyukov, kcc
Reviewed By: kcc
Subscribers: llvm-commits, kcc
Differential Revision: http://reviews.llvm.org/D6004
llvm-svn: 221194
The current handling (manual execution of atexit callbacks)
is overly complex and leads to constant problems due to mutual ordering of callbacks.
Instead simply wrap callbacks into our wrapper to establish
the necessary synchronization.
Fixes issue https://code.google.com/p/thread-sanitizer/issues/detail?id=80
llvm-svn: 219675
On some tests we see that signals are not delivered
when a thread is blocked in epoll_wait. The hypothesis
is that the signal is delivered right before epoll_wait
call. The signal is queued as in_blocking_func is not set
yet, and then the thread just blocks in epoll_wait forever.
So double check pending signals *after* setting
in_blocking_func. This way we either queue a signal
and handle it in the beginning of a blocking func,
or process the signal synchronously if it's delivered
when in_blocking_func is set.
llvm-svn: 218070
There interceptors do not seem to be strictly necessary for tsan.
But we see cases where the interceptors consume 70% of execution time.
Memory blocks passed to fgetgrent_r are "written to" by tsan several times.
First, there is some recursion (getgrnam_r calls fgetgrent_r), and each
function "writes to" the buffer. Then, the same memory is "written to"
twice, first as buf and then as pwbufp (both of them refer to the same addresses).
llvm-svn: 216904
Convert TSan and LSan to the new interface. More changes will follow:
1) "suppressions" should become a common runtime flag.
2) Code for parsing suppressions file should be moved to SuppressionContext::Init().
llvm-svn: 214334
The former used to crash with a null deref if it was given a not owned pointer,
while the latter returned 0. Now they both return 0. This is still not the best possible
behavior: it is better to print an error report with a stack trace, pointing
to the error in user code, as we do in ASan.
llvm-svn: 212112
The new storage (MetaMap) is based on direct shadow (instead of a hashmap + per-block lists).
This solves a number of problems:
- eliminates quadratic behaviour in SyncTab::GetAndLock (https://code.google.com/p/thread-sanitizer/issues/detail?id=26)
- eliminates contention in SyncTab
- eliminates contention in internal allocator during allocation of sync objects
- removes a bunch of ad-hoc code in java interface
- reduces java shadow from 2x to 1/2x
- allows to memorize heap block meta info for Java and Go
- allows to cleanup sync object meta info for Go
- which in turn enabled deadlock detector for Go
llvm-svn: 209810
The refactoring makes suppressions more flexible
and allow to suppress based on arbitrary number of stacks.
In particular it fixes:
https://code.google.com/p/thread-sanitizer/issues/detail?id=64
"Make it possible to suppress deadlock reports by any stack (not just first)"
llvm-svn: 209757
This way does not require a __sanitizer_cov_dump() call. That's
important on Android, where apps can be killed at arbitrary time.
We write raw PCs to disk instead of module offsets; we also write
memory layout to a separate file. This increases dump size by the
factor of 2 on 64-bit systems.
llvm-svn: 209653
Move fflush and fclose interceptors to sanitizer_common.
Use a metadata map to keep information about the external locations
that must be updated when the file is written to.
llvm-svn: 208676
Kostya Serebryany