Is it that useful to abstract? That code that creates GpuChannelHost and GpuMemoryBufferManager is already in code that isn't used by mus - e.g. in the renderer that is done inside the RenderWidgetCompositorDelegate implementation, which differs between the mus version and the content version (AIUI). Abstracting it at a lower level is only useful if we can share code, but can we?

Assigning to Sadrul. I feel like this may be out of date at this point. Feel free to mark as WontFix.

I have updated chrome-renderer to use mus-gpu[*] on all platform, whether running mus or not. (crrev.com/439282)

I am currently updating chrome-gpu process to do the same (crrev.com/2612623002), where chrome-gpu will provide the impl. to chrome-browser using the mus implementation of the gpu API.

chrome-browser too will be updated to use the gpu-host implementation from mus.

*: i.e. mus implementation of the gpu API. When running in mus, the mus server provides the implementation of the API. Otherwise, chrome-browser provides the impl.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a4447c621719211e9d8624de2fe62c59f1e1384e

commit a4447c621719211e9d8624de2fe62c59f1e1384e
Author: sadrul <sadrul@chromium.org>
Date: Wed Jul 05 18:17:54 2017

vr/gpu: BrowserGpuChannelHostFactory should be called on the same thread.

MailboxToSurfaceBridge in vr code lives in the vr-gl thread. It requests
a cc::ContextProvider from content::Compositor, which in turn gets a gpu
channel from BrowserGpuChannelHostFactory. However, the factory lives in
the UI thread, and is not thread-safe. So, impose a restriction on the
factory and content::Compositor that the relevant functions can only be
called from the UI thread. Update the vr code to do the necessary thread
hopping to make sure it calls the functions on the right threads.

BUG=609316

Review-Url: https://codereview.chromium.org/2969153002
Cr-Commit-Position: refs/heads/master@{#484311}

[modify] https://crrev.com/a4447c621719211e9d8624de2fe62c59f1e1384e/chrome/browser/android/vr_shell/mailbox_to_surface_bridge.cc
[modify] https://crrev.com/a4447c621719211e9d8624de2fe62c59f1e1384e/chrome/browser/android/vr_shell/mailbox_to_surface_bridge.h
[modify] https://crrev.com/a4447c621719211e9d8624de2fe62c59f1e1384e/content/browser/gpu/browser_gpu_channel_host_factory.cc
[modify] https://crrev.com/a4447c621719211e9d8624de2fe62c59f1e1384e/content/browser/renderer_host/compositor_impl_android.cc
[modify] https://crrev.com/a4447c621719211e9d8624de2fe62c59f1e1384e/content/public/browser/android/compositor.h

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca

commit 059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca
Author: sadrul <sadrul@chromium.org>
Date: Wed Jul 05 21:22:16 2017

gpu: Reduce of BrowserGpuChannelHostFactory.

BrowserGpuChannelHostFactory will be replaced by ui::Gpu. So reduce
its usage. Most of the call-sites need just the base interface of
gpu::GpuChannelHostFactory. So provide a way to get to that from the
BrowserMainLoop singleton instance.

BUG=609316

Review-Url: https://codereview.chromium.org/2967963002
Cr-Commit-Position: refs/heads/master@{#484354}

[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/browser_main_loop.cc
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/browser_main_loop.h
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/gpu/browser_gpu_channel_host_factory.cc
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/gpu/browser_gpu_channel_host_factory.h
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/gpu/gpu_ipc_browsertests.cc
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/renderer_host/compositor_impl_android.cc
[modify] https://crrev.com/059bc5352b1cd9e29e8fc89ceffdb767a2bd2cca/content/browser/renderer_host/media/video_capture_gpu_jpeg_decoder.cc

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/519814bd8b4235d2f1702ebdc306a16c5b03707f

commit 519814bd8b4235d2f1702ebdc306a16c5b03707f
Author: sadrul <sadrul@chromium.org>
Date: Thu Jul 06 17:58:30 2017

mus-gpu: Fix handling channel establish requests.

A couple of changes:
. If a request for establishing a gpu channel is made from the callback
  of a failed previous request, then make sure the request is processed
  correctly. Without this fix, |ui::Gpu::establish_callbacks_| is
  changed while iterating over it, causing errors.
. If a request for a gpu channel was successful, then the subsequent
  requests should be responded synchronously. This is the current
  behaviour of BrowserGpuChannelHostFactory.

BUG=609316

Review-Url: https://codereview.chromium.org/2968053002
Cr-Commit-Position: refs/heads/master@{#484664}

[modify] https://crrev.com/519814bd8b4235d2f1702ebdc306a16c5b03707f/services/ui/public/cpp/gpu/gpu.cc
[modify] https://crrev.com/519814bd8b4235d2f1702ebdc306a16c5b03707f/services/ui/public/cpp/tests/gpu_unittest.cc

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/49c9257113c2de95b7a16506c297de1767b36931

commit 49c9257113c2de95b7a16506c297de1767b36931
Author: Sadrul Habib Chowdhury <sadrul@chromium.org>
Date: Fri Jul 14 19:17:41 2017

gpu: Retry gpu channel connection in the host.

If establishing a gpu channel fails, then it could be because access to
gpu is not allowed, in which case the client should use software fallback.
If, on the other hand, the failure happens because the gpu process was
killed, or the gpu process crashed, then the attempt to establish the
channel should be retried. But instead of retrying in the clients, retry
in the host, which has better information to know whether a retry attempt
may be successful or not.

Remove the retry logic in the renderer (in RenderWidgetCompositor), since
the host (GpuClient) will be taking care of that.

BUG=609316

Change-Id: Ibcc1d5e79f354ccb282bad2321fca26d1c06402c
Reviewed-on: https://chromium-review.googlesource.com/571490
Reviewed-by: Antoine Labour <piman@chromium.org>
Commit-Queue: Sadrul Chowdhury <sadrul@chromium.org>
Cr-Commit-Position: refs/heads/master@{#486835}
[modify] https://crrev.com/49c9257113c2de95b7a16506c297de1767b36931/content/browser/gpu/gpu_client.cc
[modify] https://crrev.com/49c9257113c2de95b7a16506c297de1767b36931/content/renderer/gpu/render_widget_compositor.cc
[modify] https://crrev.com/49c9257113c2de95b7a16506c297de1767b36931/content/renderer/gpu/render_widget_compositor.h
[modify] https://crrev.com/49c9257113c2de95b7a16506c297de1767b36931/content/renderer/gpu/render_widget_compositor_unittest.cc

Cleaning up old Proj-Mustash labels.