- User Since
- Aug 30 2015, 11:51 AM (119 w, 1 d)
Sat, Dec 9
******************** Testing Time: 94.77s ******************** Failing Tests (7): MemorySanitizer-x86_64 :: chained_origin_with_signals.cc MemorySanitizer-x86_64 :: dtls_test.c MemorySanitizer-x86_64 :: fork.cc MemorySanitizer-x86_64 :: sem_getvalue.cc MemorySanitizer-x86_64 :: signal_stress_test.cc MemorySanitizer-x86_64 :: textdomain.cc MemorySanitizer-x86_64 :: tzset.cc
Fri, Dec 8
Reverted, it breaks Linux.
As discussed with Joerg, this TLS handling is questionable by design, but the proposed NetBSD code is not worse than any other OS.
I'm not reproducing the issues on NetBSD, so for now I will move this to background and keep patches locally.
Reverted, as it broke Linux.
Thu, Dec 7
Wed, Dec 6
I've addressed the comments in review.
I think it's done now.
Apply changes from review.
Tue, Dec 5
Just reminding about my side question:
- rename sigaction_symname to SIGACTION_SYMNAME
- put SIGACTION_SYMNAME in sanitizer_common/sanitizer_platform_limits_posix.h and ..netbsd.h
Remove comments stating obvious.
@joerg done. I will fix improve handling of dynamically allocated vectors in another revision.
- Fix bug with static TLS vector index.
- Reuse dl_iterate_phdr().
- Simplify code.
Mon, Dec 4
Back to it as required for MSan.
Rebase to trunk.
Sat, Dec 2
While there.. is lib/sanitizer_common/sanitizer_common_syscalls.inc required for MSan?
Fri, Dec 1
Remove unrelated local changes from this patch.
Thu, Nov 30
OK, so we can reschedule it for later. Excuse me for noise. It looks like compiler-rt wasn't built (in a while) on NetBSD/evbmips (release engineering machines test x86_64, i386, ppc, sparc, sparc64, arm). Once it will be done, we will address it.
It looks like current NetBSD has it as <mips/cachectl.h>, my mistake.
Additionally NetBSD requires symbol mangling: clock_gettime -> __clock_gettime50.
NetBSD does not ship with vDSO and the code should/could be disabled for it.
Please correct this commit. This header is also present on NetBSD/MIPS.
Wed, Nov 29
Thank you again for the great support!
- rename TSanMaybeSpawnBackgroundThread() to MaybeSpawnBackgroundThread ()
- simplify atomic check in MaybeSpawnBackgroundThread()
Tue, Nov 28
Try to make the code prettier.
Try to share more code between NetBSD and Linux.
I will give it a try.
Thanks! I've uploaded a slightly older diff, that has another renaming in sanitizer_signal_interceptors.inc. I will commit the newer version and check the remaining portability issues on buildbots.
Mon, Nov 27
I think Memory should be removed as it's impossible to make it portable as is to every combination of OS and hardware. It's too low-level interface.
Sun, Nov 26