If the thread receives a signal concurrently with PTRACE_ATTACH,
we can get notification about the signal before notification about stop.
In such case we need to forward the signal to the thread, otherwise
the signal will be missed (as we do PTRACE_DETACH with arg=0) and
any logic relying on signals will break. After forwarding we need to
continue to wait for stopping, because the thread is not stopped yet.
We do ignore delivery of SIGSTOP, because we want to make stop-the-world
as invisible as possible.
http://reviews.llvm.org/D7723
--This line, and those below, will be ignored--
M lib/sanitizer_common/sanitizer_stoptheworld_linux_libcdep.cc
M test/tsan/signal_segv_handler.cc
llvm-svn: 229832
Long story short: stop-the-world briefly resets SIGSEGV handler to SIG_DFL.
This breaks programs that handle and continue after SIGSEGV (namely JVM).
See the test and comments for details.
http://reviews.llvm.org/D7722
llvm-svn: 229678
Enabling internal ptrace for mips, which fixes some
ptrace related tests. Along with this fixing some
other failures.
Reviewers: Reviewers: eugenis, kcc, samsonov
Subscribers: dsanders, sagar, lldb-commits
Differential Revision: http://reviews.llvm.org/D7332
llvm-svn: 229656
Provide defaults for TSAN_COLLECT_STATS and TSAN_NO_HISTORY.
Replace #ifdef directives with #if. This fixes a bug introduced
in r229112, where building TSan runtime with -DTSAN_COLLECT_STATS=0
would still enable stats collection and reporting.
llvm-svn: 229581
In general, this is a reasonable warning, except real_pthread_create is
weak and can be null. The existing usage is correct as it the function
is declared with SANITIZER_WEAK, but MSVC can't know that because it is
defined to nothing on Windows.
llvm-svn: 229562
They autotools build has a number of missing features, supports less
OS, architectures, build configurations, doesn't have any tests and
is hard to support in sync with CMake build.
llvm-svn: 229556
Summary:
LSan can be combined with a parent tool (for now it's only ASan).
Also, we allow LSAN_OPTIONS to override certain common flags. It means
we have to parse LSAN_OPTIONS early enough, before the rest of the
parent tool (including chunks of sanitizer_common) is initialized.
In future, we can use the same approach for UBSan, after we embed it
into ASan runtime in a similar way.
Test Plan: regression test suite
Reviewers: earthdok, eugenis
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D7577
llvm-svn: 229519
In Go mode the background thread is not started (internal_thread_start is empty).
There is no sense in having this code compiled in.
Also removes dependency on sanitizer_linux_libcdep.cc which is good,
ideally Go runtime does not depend on libc at all.
llvm-svn: 229396
mingw gcc complains:
warning: 'err' may be used uninitialized in this function [-Wmaybe-uninitialized]
Printf("Failed to read options from '%s': error %d\n", value, err);
llvm-svn: 229392
Android devices may not support 'adb root', but be rooted with 'su'
binary. This patch makes it possible to install ASAN to such
devices. When --use-su flag is specified, most 'adb ...' commangs are
changed to 'adb su -c "..."'.
Some other notes:
* 'readlink' changed to 'ls -l', since not all devices have readlink
in their firmware.
* removing ASan library step moved to very end, because 'su' may not
run properly without this library until shell will be restarted.
Patch by Dmitry <ripp at yandex-team dot ru>.
llvm-svn: 229368
Revision 229127 introduced a bug:
zero value is not OK for trace headers,
because stack0 needs constructor call.
Instead unmap the unused part of trace after
all ctors have been executed.
llvm-svn: 229263
We are going to use only a small part of the trace with the default
value of history_size. However, the constructor writes to the whole trace.
It writes mostly zeros, so freshly mmaped memory will do.
The only non-zero field if mutex type used for debugging.
Reduces per-goroutine overhead by 8K.
https://code.google.com/p/thread-sanitizer/issues/detail?id=89
llvm-svn: 229127
The ContainsSameAccess optimization substantially reduces pressure
on trace by eliminating duplicate accesses. So now we can reduce
default trace size to reduce per-goroutine memory consumption.
Current default size is 64K events, new -- 32K events.
In either case user can change it with GORACE env var.
Reduces per-goroutine memory consumption from 356K to 226K.
llvm-svn: 229117
and even before that, it was never implemented. Just define it to zero
instead, so compiler-rt can compile on FreeBSD 11 and later.
Differential Revision: http://reviews.llvm.org/D7485
llvm-svn: 228871
Summary:
__ubsan::getFunctionLocation() used to issue a call to symbolizer, and
convert the result (SymbolizedStack) to one of UBSan structures:
SourceLocation, ModuleLocation or MemoryLocation. This:
(1) is inefficient: we do an extra allocation/deallocation to copy data,
while we can instead can just pass SymbolizedStack around (which
contains all the necessary data).
(2) leaks memory: strings stored in SourceLocation/MemoryLocation are
never deallocated, and Filipe Cabecinhas suggests this causes crashes
of UBSan-ified programs in the wild.
Instead, let Location store a pointer to SymbolizedStack object, and
make sure it's properly deallocated when UBSan handler exits.
ModuleLocation is made obsolete by this change, and is deleted.
Test Plan: check-ubsan test suite
Reviewers: rsmith, filcab
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D7548
llvm-svn: 228869
We're still using global constructors when not on linux, so this was
causing warnings on Darwin. Reverting for now.
This reverts r228384.
llvm-svn: 228795
Summary:
Make sure we don't print the error report from -fsanitize=function
twice for the same source location, as we do in another UBSan handlers.
Test Plan: check-ubsan test suite
Reviewers: rsmith, pcc
Reviewed By: pcc
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D7524
llvm-svn: 228772
Make sure clear_cache() builtin has an appropriate definition for Linux.
Call syscall(NR_cacheflush, ...).
Differential Revision: http://reviews.llvm.org/D7205
llvm-svn: 228767
We currently skip all "Noinst" unit tests on OS X, which was probably caused when we removed the "allow_reexec" flag. The MaybeReexec function fails to re-execute when the runtime is linked statically, because there is no dylib to use. This patch adds an explicit DisableReexec function that is used from asan_noinst_test.cc and the runtime then doesn't try to re-execute.
Reviewed at http://reviews.llvm.org/D7493
llvm-svn: 228740
Alexey Samsonov