I think the issue is that the no sync tokens are being verified in the proxy. If I do "UpdateVerifiedReleases(channel_->GetHighestValidatedFlushID(stream_id_));" in OrderingBarrier once the list of flushes goes over 100, then that seems to solve the problem.

+piman@

I can't see any use of flushed_release_flush_id_ other than updating it. Maybe it isn't really required?

It is used in UpdateVerifiedReleases - the map is used to keep track of what we have flushed but not yet verified, so that we can know if we need to do the round trip in IsFenceSyncFlushReceived

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/c5faeccf26cc5e05407c6375231aef4e736138f9

commit c5faeccf26cc5e05407c6375231aef4e736138f9
Author: jbauman <jbauman@chromium.org>
Date: Tue Aug 09 21:50:31 2016

Clear old verified flush IDs when doing OrderingBarrier.

If a command buffer generates sync tokens but doesn't validate them
(letting a different command buffer in the same stream do that), then
its queue of unverified flush ids will grow without bound. It can clear
them out with little overhead by checking which ones are verified after
every GpuChannelHost::OrderingBarrier call.

BUG= 635051 
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/2222663002
Cr-Commit-Position: refs/heads/master@{#410840}

[modify] https://crrev.com/c5faeccf26cc5e05407c6375231aef4e736138f9/gpu/ipc/client/command_buffer_proxy_impl.cc
[modify] https://crrev.com/c5faeccf26cc5e05407c6375231aef4e736138f9/gpu/ipc/client/gpu_channel_host.cc
[modify] https://crrev.com/c5faeccf26cc5e05407c6375231aef4e736138f9/gpu/ipc/client/gpu_channel_host.h

Your change meets the bar and is auto-approved for M53 (branch: 2785)

Please merge your change by today 5:00 PM PT if possible so we can take it in for next week this release. Thank you.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9987bc7fb26e0807237f31cf33d6b7729714542b

commit 9987bc7fb26e0807237f31cf33d6b7729714542b
Author: John Bauman <jbauman@chromium.org>
Date: Mon Aug 15 21:33:00 2016

Clear old verified flush IDs when doing OrderingBarrier.

If a command buffer generates sync tokens but doesn't validate them
(letting a different command buffer in the same stream do that), then
its queue of unverified flush ids will grow without bound. It can clear
them out with little overhead by checking which ones are verified after
every GpuChannelHost::OrderingBarrier call.

BUG= 635051 
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/2222663002
Cr-Commit-Position: refs/heads/master@{#410840}
(cherry picked from commit c5faeccf26cc5e05407c6375231aef4e736138f9)

Review URL: https://codereview.chromium.org/2241253003 .

Cr-Commit-Position: refs/branch-heads/2785@{#609}
Cr-Branched-From: 68623971be0cfc492a2cb0427d7f478e7b214c24-refs/heads/master@{#403382}

[modify] https://crrev.com/9987bc7fb26e0807237f31cf33d6b7729714542b/gpu/ipc/client/command_buffer_proxy_impl.cc
[modify] https://crrev.com/9987bc7fb26e0807237f31cf33d6b7729714542b/gpu/ipc/client/gpu_channel_host.cc
[modify] https://crrev.com/9987bc7fb26e0807237f31cf33d6b7729714542b/gpu/ipc/client/gpu_channel_host.h

Still may be broken in some circumstances.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0f79f41c428a89593c81deaf5bbbd952b80195d6

commit 0f79f41c428a89593c81deaf5bbbd952b80195d6
Author: jbauman <jbauman@chromium.org>
Date: Thu Sep 08 02:20:33 2016

Validate all flush IDs if the unvalidated list gets too long.

The browser process normally never needs to validate sync tokens unless
it's returning resources to a renderer, so it's possible the highest
verified flush ID will almost never be updated and the list of
unvalidated flush IDs will grow indefinitely. Force a validation if that
happens to avoid wasting memory.

BUG= 635051 
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/2318933005
Cr-Commit-Position: refs/heads/master@{#417168}

[modify] https://crrev.com/0f79f41c428a89593c81deaf5bbbd952b80195d6/gpu/ipc/client/command_buffer_proxy_impl.cc
[modify] https://crrev.com/0f79f41c428a89593c81deaf5bbbd952b80195d6/gpu/ipc/client/command_buffer_proxy_impl.h

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0f79f41c428a89593c81deaf5bbbd952b80195d6

commit 0f79f41c428a89593c81deaf5bbbd952b80195d6
Author: jbauman <jbauman@chromium.org>
Date: Thu Sep 08 02:20:33 2016

Validate all flush IDs if the unvalidated list gets too long.

The browser process normally never needs to validate sync tokens unless
it's returning resources to a renderer, so it's possible the highest
verified flush ID will almost never be updated and the list of
unvalidated flush IDs will grow indefinitely. Force a validation if that
happens to avoid wasting memory.

BUG= 635051 
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/2318933005
Cr-Commit-Position: refs/heads/master@{#417168}

[modify] https://crrev.com/0f79f41c428a89593c81deaf5bbbd952b80195d6/gpu/ipc/client/command_buffer_proxy_impl.cc
[modify] https://crrev.com/0f79f41c428a89593c81deaf5bbbd952b80195d6/gpu/ipc/client/command_buffer_proxy_impl.h