New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 671728 link

Starred by 1 user

Issue metadata

Status: Duplicate
Owner:
Closed: Dec 2016
Cc:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 3
Type: Bug



Sign in to add a comment

Multiple flakes on gpu::gles2::GLES2Implementation::BufferDataHelper on Mac GPU ASAN FYI bot

Project Member Reported by ccameron@chromium.org, Dec 6 2016

Issue description

This function has the ominous text from crrev.com/314098 and  issue 445745 

#if defined(MEMORY_SANITIZER) && !defined(OS_NACL)
  // Do not upload uninitialized data. Even if it's not a bug, it can cause a
  // bogus MSan report during a readback later. This is because MSan doesn't
  // understand shared memory and would assume we were reading back the same
  // unintialized data.
  if (data) __msan_check_mem_is_initialized(data, size);
#endif

Currently we're seeing at least one failure every run on
https://build.chromium.org/p/chromium.gpu.fyi/builders/Mac%20GPU%20ASAN%20Release

An example failure:
==18553==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7fff5aecf370 at pc 0x0001070b6438 bp 0x7fff5aecef30 sp 0x7fff5aece6e0
READ of size 128 at 0x7fff5aecf370 thread T0
    #0 0x1070b6437 in __asan_memcpy (in libclang_rt.asan_osx_dynamic.dylib) + 855
    #1 0x1051c65f9 in gpu::gles2::GLES2Implementation::BufferDataHelper(unsigned int, long, void const*, unsigned int) (in gles2_conform_test_windowless) + 1833
    #2 0x104e0a750 in GTFExtensionTestVertexArrayObjectApply (in gles2_conform_test_windowless) + 13920
    #3 0x104f656c0 in GTFTestExtensionApply (in gles2_conform_test_windowless) + 1248
    #4 0x104f5eb8a in GTFRunTest (in gles2_conform_test_windowless) + 4922
    #5 0x104f59caa in GTFRunTestDriver (in gles2_conform_test_windowless) + 5722
    #6 0x104f2c933 in GTFRun (in gles2_conform_test_windowless) + 803
    #7 0x104f21593 in GTFInitEGL (in gles2_conform_test_windowless) + 1747
    #8 0x104f2b4c4 in GTFMain (in gles2_conform_test_windowless) + 5732
    #9 0x104f99474 in main (in gles2_conform_test_windowless) + 772
    #10 0x7fff8f75b5ac in start (in libdyld.dylib) + 0
 
Cc: kbr@chromium.org
Is this a real bug, or should we be adding a suppression?

Comment 2 by zmo@chromium.org, Dec 6 2016

Owner: zmo@chromium.org
Status: Assigned (was: Untriaged)
Let me take a look

Comment 3 by kbr@chromium.org, Dec 8 2016

Cc: ynovikov@chromium.org
Owner: ynovikov@chromium.org
Yuly's already working on these. There have been a couple of bugs in the gles2_conform tests which he's fixed already.

Yuly, if you were tracking these under another bug already, please feel free to dup this into the other bug.

Mergedinto: 646538
Status: Duplicate (was: Assigned)
I've fixed 2 buffer overruns in the test code.
Bot went green for a first time after that.
https://build.chromium.org/p/chromium.gpu.fyi/builders/Mac%20GPU%20ASAN%20Release/builds/877

Comment 5 by kbr@chromium.org, Dec 9 2016

Awesome job Yuly!!!

Sign in to add a comment