Add Intel DRM backed GpuMemoryBuffer implementation.

Add Intel DRM backed GpuMemoryBuffer implementation.

NOT FOR REVIEW

BUG=356871

[reveman, 2014-04-04 17:00:37]

This is based on https://codereview.chromium.org/186123006/ and   
https://codereview.chromium.org/211133005/

It's really just that some code refactored into something that we can
potentially land.

It supports both browser compositor and renderer compositor but in the second
case would only work without a sandbox.

[fjhenigman, 2014-04-09 16:33:51]

https://codereview.chromium.org/225023009/diff/1/content/common/gpu/client/gp...
File content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc (right):

https://codereview.chromium.org/225023009/diff/1/content/common/gpu/client/gp...
content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc:189: if
(buffer_size > std::numeric_limits<int32>::max())
should be uint32

[reveman, 2014-04-09 18:22:22]

https://codereview.chromium.org/225023009/diff/1/content/common/gpu/client/gp...
File content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc (right):

https://codereview.chromium.org/225023009/diff/1/content/common/gpu/client/gp...
content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc:189: if
(buffer_size > std::numeric_limits<int32>::max())
On 2014/04/09 16:33:52, fjhenigman wrote:
> should be uint32

Fixed by using base::IsValueInRangeForNumericType instead.

[fjhenigman, 2014-04-10 16:26:00]

With the two tests corrected (see inline comments) zero copy seems to 
work in the browser.  However when the sense of the tests was reversed (or
now if one of the corrected conditions fails) it should quietly fall back
to shm.  Instead it was using a texture target of GL_TEXTURE_EXTERNAL_OES
with a non-EGLImage texture, and mesa objected to that.

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
File content/browser/gpu/browser_gpu_channel_host_factory.cc (right):

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
content/browser/gpu/browser_gpu_channel_host_factory.cc:390: if
(!GpuMemoryBufferImplIntelDRM::IsFormatSupported(internalformat)) {
remove the !

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
content/browser/gpu/browser_gpu_channel_host_factory.cc:398: if
(!buffer->InitializeFromBufferObject(buffer_object))
remove this ! too

[reveman, 2014-04-10 17:48:28]

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
File content/browser/gpu/browser_gpu_channel_host_factory.cc (right):

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
content/browser/gpu/browser_gpu_channel_host_factory.cc:390: if
(!GpuMemoryBufferImplIntelDRM::IsFormatSupported(internalformat)) {
On 2014/04/10 16:26:01, fjhenigman wrote:
> remove the !

Done.

https://codereview.chromium.org/225023009/diff/60001/content/browser/gpu/brow...
content/browser/gpu/browser_gpu_channel_host_factory.cc:398: if
(!buffer->InitializeFromBufferObject(buffer_object))
On 2014/04/10 16:26:01, fjhenigman wrote:
> remove this ! too

Done.

[fjhenigman, 2014-04-11 02:45:28]

https://codereview.chromium.org/225023009/diff/80001/content/common/gpu/clien...
File content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc (right):

https://codereview.chromium.org/225023009/diff/80001/content/common/gpu/clien...
content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc:293:
ExportBufferObject(buffer_object_, &handle.handle);
GetHandle() should cache the result of drm_intel_bo_gem_export_to_prime()
instead of calling it every time,
like in https://codereview.chromium.org/186123006/patch/40001/50004

I tried the change and it conserved filed descriptors and cured rendering
glitches.

[reveman, 2014-04-11 05:54:26]

I think this latest patch might fix the file descriptor leakage you were seeing.
Please give it a try when you have a chance.

https://codereview.chromium.org/225023009/diff/80001/content/common/gpu/clien...
File content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc (right):

https://codereview.chromium.org/225023009/diff/80001/content/common/gpu/clien...
content/common/gpu/client/gpu_memory_buffer_impl_intel_drm.cc:293:
ExportBufferObject(buffer_object_, &handle.handle);
On 2014/04/11 02:45:28, fjhenigman wrote:
> GetHandle() should cache the result of drm_intel_bo_gem_export_to_prime()
> instead of calling it every time,
> like in https://codereview.chromium.org/186123006/patch/40001/50004
> 
> I tried the change and it conserved filed descriptors and cured rendering
> glitches.

Done.

[reveman, 2014-04-11 07:15:51]

Note: you'll need to need to update to latest version of
https://codereview.chromium.org/215143002/ for this patch to work.

[fjhenigman, 2014-04-11 16:45:00]

On 2014/04/11 07:15:51, reveman wrote:
> Note: you'll need to need to update to latest version of
> https://codereview.chromium.org/215143002/ for this patch to work.

To test on chrome os edit the mesa ebuild:
-               $(use gbm && echo "--with-egl-platforms=drm")
+               $(use gbm && echo "--with-egl-platforms=drm,x11")
and rebuild mesa with USE="gbm egl gles1 gles2 shared-glapi" (maybe don't need
gles1).

And run chrome with:
  export EGL_PLATFORM=x11
  --use-gl=egl --no-sandbox

To test the browser compositor add --ui-enable-map-image
--ui-enable-impl-side-painting

To test the renderer compositor add --enable-map-image   
--enable-impl-side-painting

For more debug output add
  export EGL_LOG_LEVEL=debug

To change file descriptor limit: ulimit -n 8192