I confirmed VDA unittest passed on Chrome ToT by merely reverting this CL.

What bots are these test run on?

It would be preferable to fix the crash. What files do I need to copy to the device to reproduce? I tried copying video_decode_accelerator_unittest + test-25fps.vp9 and I'm getting a different error:

[9439:9445:0409/110120.301299:1321557822:FATAL:video_decode_accelerator_unittest.cc(639)] Assert failed: decoder_. Failed creating a VDA
#0 0x5c66bacf284c base::debug::StackTrace::StackTrace()
#1 0x5c66bacb1180 logging::LogMessage::~LogMessage()
#2 0x5c66baa30fe5 media::(anonymous namespace)::GLRenderingVDAClient::CreateAndStartDecoder()
#3 0x5c66bacf30fc base::debug::TaskAnnotator::RunTask()
#4 0x5c66bacb3527 base::MessageLoop::RunTask()
#5 0x5c66bacb3a17 base::MessageLoop::DoWork()
#6 0x5c66bacb4ca9 base::MessagePumpLibevent::Run()
#7 0x5c66bacc2675 base::RunLoop::Run()
#8 0x5c66bacd95aa base::Thread::ThreadMain()
#9 0x5c66bacd78bd base::(anonymous namespace)::ThreadFunc()
#10 0x7b24fb22f2b8 <unknown>
#11 0x7b24fa4f0fad clone

#3: make sure you have both the executable and the test file on the same
folder (e.g. /tmp) and that you're on that very folder as well (otherwise
arcane errors might occur).  I usually copy all needed files, e.g. from a
chromium checkout:

scp media/test/data/test-25fps.* root@192.168.1.10:/tmp

Also JIC when runnning `stop ui`, let it sink for a few seconds.

sadrul@, these are autotests running on the Chromebooks fleet:
https://stainless.corp.google.com/search?view=matrix&row=board&col=build&first_date=2018-03-28&last_date=2018-04-10&test=video_VideoDecodeAccelerator&exclude_cts=false&exclude_not_run=false&exclude_non_release=false&exclude_au=true&exclude_acts=true&exclude_retried=true&exclude_non_production=true

I got the test to run (and fail). I think there was some version mismatch between Chrome OS and simple Chrome binary.

Anyways, I see the problem. The task that hands over the DRM display controller handle to the GPU runs on the UI thread instead of IO thread now. video_decode_accelerator_unittest starts running tests before that task has run, so the GPU thread isn't ready yet and things fail.

It should be safe enough to add a RunLoop to OzoneGpuTestHelper::Initialize() so the UI thread has a chance to finish initialization.

I uploaded https://crrev.com/c/1003076 which fixes video_decode_accelerator_unittest on eve.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/2121a5d2b55687562961efed87d35b99340ab257

commit 2121a5d2b55687562961efed87d35b99340ab257
Author: kylechar <kylechar@chromium.org>
Date: Tue Apr 10 14:32:07 2018

Ensure Ozone DRM initialization has finished.

Make OzoneGpuTestHelper::Initialize() run UI thread tasks after
OnGpuProcessLaunched(). In https://crrev.com/c/980574 Ozone DRM was
changed so the DRM display controller handle was only accessed on the UI
thread to prevent races. This means that when GpuPlatformSupportHost
OnGpuProcessLaunched() is called from the IO thread there is a UI thread
task that needs to run to finish initialization.

This CL adds a RunLoop that runs on the UI thread and immediately posts
a task to quit that run loop. That will run any tasks on the UI thread
that are pending and then continue, ensuring the Ozone DRM
initialization is finished. This should be safe for other Ozone
platforms as well.

Bug:  830233 
Change-Id: I5b2e45c36f7042d59800c0822ca97b13a60a7aab
Reviewed-on: https://chromium-review.googlesource.com/1003076
Reviewed-by: Robert Kroeger <rjkroege@chromium.org>
Commit-Queue: kylechar <kylechar@chromium.org>
Cr-Commit-Position: refs/heads/master@{#549523}
[modify] https://crrev.com/2121a5d2b55687562961efed87d35b99340ab257/ui/ozone/public/ozone_gpu_test_helper.cc

ToT works at least locally now on my eve, thanks kylechar@!

Great, thanks.

The NextAction date has arrived: 2018-04-20

Not crashing anymore, so: Verified.