We started using MediaCodec.setSurface() in M56 for M+ devices, so I'd guess this is some device(s) handling this incorrectly.

blacklist-able?

Yeah, we'll have to plumb some blacklist bits, but is otherwise fairly easy since it's M+ only:

https://cs.chromium.org/chromium/src/media/filters/gpu_video_decoder.cc?l=229

You filtered the crashes to Mali-450MP, any reason why? If we blacklist all M,N Mali-450MP that will affect ~0.7% of M,N watch time. 

> You filtered the crashes to Mali-450MP, any reason why?

Err... pasted wrong link? Removing ARM and Mali, 344 out of 383 reports contain libstagefright_foundation.so

https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20AND%20custom_data.ChromeCrashProto.ptype%3D%27gpu-process%27%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%3D%27GPU%20hang%27%20AND%20product.Version%3D%2756.0.2924.59%27&ignore_case=false&enable_rewrite=true&omit_field_name=&omit_field_value=&omit_field_opt=%3D

vs that + libstagefright_foundation.so:

https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20AND%20custom_data.ChromeCrashProto.ptype%3D%27gpu-process%27%20AND%20custom_data.ChromeCrashProto.magic_signature_1.name%3D%27GPU%20hang%27%20AND%20product.Version%3D%2756.0.2924.59%27%20OMIT%20RECORD%20IF%20SUM(thread.StackTrace.StackFrame.Module.CodeFile%3D%27%2Fsystem%2Flib%2Flibstagefright_foundation.so%27)%20%3D%200&ignore_case=false&enable_rewrite=true&omit_field_name=thread.StackTrace.StackFrame.Module.CodeFile&omit_field_value=%2Fsystem%2Flib%2Flibstagefright_foundation.so&omit_field_opt=%3D#samplereports

So should be ok to block m57 then?

Sure, we can figure out some blacklisting here. In case it's a video specific issue, here's the one SFW video I could find in crash URLs:

https://endlessvideo.com/watch?v=kiYVifQ90kE

Looks like the #1 failing device is the ALE-L21 (actually the whole series of L21 devices account for 61% of crashes). We've had issues with this device in the past. I think liberato@ was going to order one per issue 650863.

liberato@, did you ever order this? I'll grab it from you tomorrow if so.

I didn't.  I think that i fixed 650863 then promptly forgot to order the phone.  Sorry about that.

liberato@ hadn't ordered one, I have one on the way and will be here Friday. Will see if I can repro and if this is an issue we can workaround or if we need to blacklist it. Given the long tail (~40%) of crashes, I'm hoping there is something we could do to prevent this hang.

Just got the device updated to M, ALE-L21 + setSurface == immediate failure. MediaCodec gives and takes a few buffers after the switch but then hangs shortly after and never returns any more input or output buffers. Seems like it's straight up broken. CTS test helpfully says "TODO" :|

https://android.googlesource.com/platform/cts/+/master/tests/tests/media/src/android/media/cts/AdaptivePlaybackTest.java#918

Will look into blacklisting options.

Looks like we'll have to go with blacklisting by Mali-450MP, just using *-L21 will only capture ~45% of the crashes versus ~72% with the GL renderer filter. Unfortunately to do that I need to plumb a gpu preferences bit since we can't easily access the GL renderer value from media/filters. Will start on that tomorrow.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/8a5efb12a76b24de92a20a23f56d20aa9e22338b

commit 8a5efb12a76b24de92a20a23f56d20aa9e22338b
Author: dalecurtis <dalecurtis@chromium.org>
Date: Tue Feb 07 02:17:21 2017

Make setOutputSurface, encrypted capabilities. Blacklist hi6210sft.

No need for platform specific define for encrypted hardware decode
support when we have a capability system; thus make support for
encrypted streams a capability.

We now have a device with broken MediaCodec.setOutputSurface(), so
make it a capability and blacklist the hardware platform common to
99% of the crashes (hi6210sft).

BUG= 683401 
TEST=p8lite no longer hangs on fullscreen transition.
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2677983004
Cr-Commit-Position: refs/heads/master@{#448514}

[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/base/android/java/src/org/chromium/media/MediaCodecUtil.java
[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/base/android/media_codec_util.cc
[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/base/android/media_codec_util.h
[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/filters/gpu_video_decoder.cc
[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/gpu/android_video_decode_accelerator.cc
[modify] https://crrev.com/8a5efb12a76b24de92a20a23f56d20aa9e22338b/media/video/video_decode_accelerator.h

Looks good on dev, marking as merge request.

Your change meets the bar and is auto-approved for M57. Please go ahead and merge the CL to branch 2987 manually. Please contact milestone owner if you have questions.
Owners: amineer@(clank), cmasso@(bling), ketakid@(cros), govind@(desktop)

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/e3f399dec4ba3384929172a2ca92ab92a2a5132e

commit e3f399dec4ba3384929172a2ca92ab92a2a5132e
Author: Dale Curtis <dalecurtis@chromium.org>
Date: Sat Feb 11 01:01:24 2017

Merge M57: "Make setOutputSurface, encrypted capabilities. Blacklist hi6210sft."

No need for platform specific define for encrypted hardware decode
support when we have a capability system; thus make support for
encrypted streams a capability.

We now have a device with broken MediaCodec.setOutputSurface(), so
make it a capability and blacklist the hardware platform common to
99% of the crashes (hi6210sft).

BUG= 683401 
TEST=p8lite no longer hangs on fullscreen transition.
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2677983004
Cr-Commit-Position: refs/heads/master@{#448514}
(cherry picked from commit 8a5efb12a76b24de92a20a23f56d20aa9e22338b)

Review-Url: https://codereview.chromium.org/2693653003 .
Cr-Commit-Position: refs/branch-heads/2987@{#459}
Cr-Branched-From: ad51088c0e8776e8dcd963dbe752c4035ba6dab6-refs/heads/master@{#444943}

[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/base/android/java/src/org/chromium/media/MediaCodecUtil.java
[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/base/android/media_codec_util.cc
[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/base/android/media_codec_util.h
[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/filters/gpu_video_decoder.cc
[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/gpu/android_video_decode_accelerator.cc
[modify] https://crrev.com/e3f399dec4ba3384929172a2ca92ab92a2a5132e/media/video/video_decode_accelerator.h

We can merge back to M56 if there's going to be another update, it's currently ~12% of GPU crashes on stable. +amineer to make the call.

I'll apply a Merge-Review-56 tag to this to keep it on my radar for respins, but as of now, nothing planned.

No more 56.