Commit Graph

8 Commits

Author SHA1 Message Date
Jonas Hahnfeld 0a9cb239a6 [compiler-rt] Fix broken sanitizer bots (hopefully)
According to the logs and local debugging there were two issues:
1) tsan tests listed libc++.a before the source file. That's usually
   ok for shared libraries, but the linker will not add symbols from
   a static library unless needed at that time. As a result the tests
   that rely upon symbols from the library (and not only include the
   headers) had undefined references.
   To solve this I'm adding a new substitution %link_libcxx_tsan which
   expands to libc++.a if available.
2) The target Fuzzer-x86_64-Test linked in SANITIZER_TEST_CXX_LIBRARIES
   which defaults to -lstdc++. This resulted in error messages like
     hidden symbol '_ZdlPv' is not defined locally
     hidden symbol '_Znwm' is not defined locally
   when using GNU gold (ld.bfd and lld are fine). Removing the linkage
   is fine because we build a custom libc++ for that purpose.

llvm-svn: 354231
2019-02-17 18:47:33 +00:00
Dmitry Vyukov eee690b29a tsan: fix test invisible barrier
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
2015-12-08 17:54:47 +00:00
Dmitry Vyukov 3ab6b2347e tsan: remove sleeps from tests
Even sleep(1) lead to episodical flakes on some machines.
Use an invisible by tsan barrier to enforce required execution order instead.
This makes the tests deterministic and faster.

llvm-svn: 226659
2015-01-21 13:50:02 +00:00
Alexey Samsonov 2c8dda4e53 [TSan] Make lit-tests more self-contained
llvm-svn: 210370
2014-06-06 21:04:55 +00:00
Dmitry Vyukov 233f401c2b tsan: make positive tests more robust
Add a script that is used to deflake inherently flaky tsan tests.
It is invoked from lit tests as:
%deflake %run %t
The script runs the target program up to 10 times,
until it produces a tsan warning.

llvm-svn: 209898
2014-05-30 14:08:51 +00:00
Greg Fitzgerald b8aae5405b Add %run to all lit tests
llvm-svn: 207709
2014-04-30 21:34:17 +00:00
Dmitry Vyukov eac8cc7a42 tsan: fix flaky test
llvm-svn: 204119
2014-03-18 06:37:31 +00:00
Alexey Samsonov e6a6183e9b Move TSan lit-tests under test/tsan
llvm-svn: 201414
2014-02-14 14:35:48 +00:00