https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket.appspot.com/8934337943248186144/+/steps/browser_tests/0/logs/CopyBetweenWindows__x2f_FilesAppBrowserTest.Test__x2f_copyBetweenWindowsDriveToLocal_DriveFs/0

This appears to be saying that a member of GpuMemoryBufferHandle is not initialized, and it's being serialized to mojo. I think this is where the GpuMemoryBufferHandle is created: https://cs.chromium.org/chromium/src/components/viz/host/gpu_client.cc?type=cs&q=components/viz/host/gpu_client.cc:182&g=0&l=186

So this might be fixed by providing default values for members of GpuMemoryBufferHandle.

I'll add an initialization for GpuMemoryBufferHandle members that aren't default-initialized. It doesn't explain though why we started seeing this issue only now. 

There was one recent change in GpuMemoryBufferHandle serialization: https://crrev.com/c/1251322

thakis@ could you take a look? It seems that the code path you've added is executed in the failing test.

https://chromium-review.googlesource.com/c/chromium/src/+/1251322 didn't add any executable code, so I don't think it could cause any behavior change.

Ok, I have a patch in the CQ that may fix the issue, let's wait.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/6ae13a865a0a4fc22113708f19004f1194ade19c

commit 6ae13a865a0a4fc22113708f19004f1194ade19c
Author: Alexandr Ilin <alexilin@chromium.org>
Date: Tue Oct 02 15:35:11 2018

Initialize all members of GpuMemoryBufferHandle in the default ctor

GpuMemoryBufferHandle::offset and GpuMemoryBufferHandle::stride are not
initialized in the default constructor. These uninitialized members may be
serialized later by mojo that makes MSAN bot unhappy.

Bug:  890930 
Change-Id: I0663de62997d673b97907a8bd470e9834bdaf638
Reviewed-on: https://chromium-review.googlesource.com/1256786
Reviewed-by: Daniele Castagna <dcastagna@chromium.org>
Commit-Queue: Alexandr Ilin <alexilin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#595850}
[modify] https://crrev.com/6ae13a865a0a4fc22113708f19004f1194ade19c/ui/gfx/gpu_memory_buffer.cc

The test is no longer flaky on the "Linux ChromiumOS MSan Tests" bot. 

I've checked that on the flakiness dashboard: https://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=CopyBetweenWindows%2FFilesAppBrowserTest.Test%2FcopyBetweenWindowsDriveToLocal_DriveFs