Loosen the GLFenceARB assertions when glClientWaitSync fails.

Loosen the GLFenceARB assertions when glClientWaitSync fails.

If the driver supports one of the {ARB,EXT,KHR}_robustness extensions,
then the context will have been allocated with it. In this case, if
glClientWaitSync fails, don't crash, but assert that the context's
been lost. This will be discovered by the command buffer later,
leading to a lost context.

BUG=650138
Committed: https://crrev.com/dec2cf265c13989c3e0f112af844c0259daa173d
Cr-Commit-Position: refs/heads/master@{#426409}

[Ken Russell (switch to Gerrit), 2016-10-08 01:31:55]

kbr@chromium.org changed reviewers:
+ zmo@chromium.org

[Ken Russell (switch to Gerrit), 2016-10-08 01:31:56]

Mo: Please review.

Others: FYI. Any comments are welcome.

This implements piman's suggestion on the bug and goes in conjunction with
https://chromium-review.googlesource.com/394233 . I tested it by manually
injecting a failure in ANGLE's FenceSync11::clientWait and verifying that Chrome
detected and recovered from it.

[Ken Russell (switch to Gerrit), 2016-10-08 02:04:09]

CC'ing a few other potentially interested people.

This was tested in a Release build with dcheck_always_on=true and the ANGLE CL
https://chromium-review.googlesource.com/394233 with a manual failure injected
into FenceSync11::clientWait. The lost context was detected cooperatively by the
command buffer, and the GPU process exited and restarted cooperatively.

Since the failure was deterministic the GPU process exited repeatedly, and at
least one time, it crashed in a different way (I didn't get a stack trace from
it). Therefore I don't think this will address all the crashes, but should
reduce the crash rate.

[piman, 2016-10-11 00:29:02]

piman@chromium.org changed reviewers:
+ piman@chromium.org

[piman, 2016-10-11 00:29:02]

lgtm

[Zhenyao Mo, 2016-10-11 01:01:25]

On 2016/10/11 00:29:02, piman extended OOO until Oct17 wrote:
> lgtm

lgtm

[Ken Russell (switch to Gerrit), 2016-10-20 03:14:19]

The CQ bit was checked by kbr@chromium.org

[Ken Russell (switch to Gerrit), 2016-10-20 03:14:20]

The patchset sent to the CQ was uploaded after l-g-t-m from zmo@chromium.org,
piman@chromium.org
Link to the patchset: https://codereview.chromium.org/2408543003/#ps20001
(title: "Preemptive rebase.")

[commit-bot: I haz the power, 2016-10-20 03:14:56]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-10-20 05:05:36]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-10-21 13:15:50]

Description was changed from

==========
Loosen the GLFenceARB assertions when glClientWaitSync fails.

If the driver supports one of the {ARB,EXT,KHR}_robustness extensions,
then the context will have been allocated with it. In this case, if
glClientWaitSync fails, don't crash, but assert that the context's
been lost. This will be discovered by the command buffer later,
leading to a lost context.

BUG=650138
==========

to

==========
Loosen the GLFenceARB assertions when glClientWaitSync fails.

If the driver supports one of the {ARB,EXT,KHR}_robustness extensions,
then the context will have been allocated with it. In this case, if
glClientWaitSync fails, don't crash, but assert that the context's
been lost. This will be discovered by the command buffer later,
leading to a lost context.

BUG=650138

Committed: https://crrev.com/dec2cf265c13989c3e0f112af844c0259daa173d
Cr-Commit-Position: refs/heads/master@{#426409}
==========

[commit-bot: I haz the power, 2016-10-21 13:15:51]

Patchset 2 (id:??) landed as
https://crrev.com/dec2cf265c13989c3e0f112af844c0259daa173d
Cr-Commit-Position: refs/heads/master@{#426409}