Ozone: Keep the native pixmap alive till gl_image is destroyed

Ozone: Keep the native pixmap alive till gl_image is destroyed

While testing with support for HW Overlays enabled and playing video,
I observed that browser crashes when you refresh the tab or try to close
the window (Debug builds). This was the sequence:
1)GLImage is scheduled to be displayed with GLSurfaceOzone.(HandleScheduleOverlayPlaneCHROMIUM)
2)If the page is refreshed, we start releasing the resources of
  Video and VaapiDrmPicture calls destroy on the GLImage.
3)In the swapbuffer request, surfaceOzone tries to use the GLImage to
  display the buffer on screen but the pixmap_is not valid anymore as
  it has been destroyed in step 2 above. (We hit the DCHECK in
  GLImageOzoneNativePixmap::ScheduleOverlayPlane)

Now, we release only EGL resources during Destroy call while
keeping around NativePixmap till ImageOzoneNativePixmap is
alive.

BUG=

Committed: https://crrev.com/862a9d9c5b475b05b6bb1741f782b8459967591e
Cr-Commit-Position: refs/heads/master@{#342280}

[kalyank, 2015-08-06 08:00:24]

kalyan.kondapally@intel.com changed reviewers:
+ achaulk@chromium.org, alexst@chromium.org, piman@chromium.org

[kalyank, 2015-08-06 08:00:24]

@piman I was thinking if it would be better to guarantee that the native pixmap
is not destroyed till SurfaceOzone is done with it. I am not sure who should be
able to guarantee this and destroy the resources at the right
time(texture_manager ?)

[alexst (slow to review), 2015-08-06 11:50:24]

On 2015/08/06 08:00:24, kalyank wrote:
> @piman I was thinking if it would be better to guarantee that the native
pixmap
> is not destroyed till SurfaceOzone is done with it. I am not sure who should
be
> able to guarantee this and destroy the resources at the right
> time(texture_manager ?)

Why not just release the EGL binding with GLImageEGL::Destroy(have_context); but
hold onto the pixmap_ in gl_image_ozone_native_pixmap until the destructor is
called?

[kalyank, 2015-08-06 15:19:13]

On 2015/08/06 11:50:24, alexst wrote:
> On 2015/08/06 08:00:24, kalyank wrote:
> > @piman I was thinking if it would be better to guarantee that the native
> pixmap
> > is not destroyed till SurfaceOzone is done with it. I am not sure who should
> be
> > able to guarantee this and destroy the resources at the right
> > time(texture_manager ?)
> 
> Why not just release the EGL binding with GLImageEGL::Destroy(have_context);
but
> hold onto the pixmap_ in gl_image_ozone_native_pixmap until the destructor is
> called?

Done, thanks.

[piman, 2015-08-06 20:20:21]

lgtm

[kalyank, 2015-08-06 22:08:54]

On 2015/08/06 20:20:21, piman (Very slow to review) wrote:
> lgtm

@alexst, Is this what you had in  mind?

[alexst (slow to review), 2015-08-06 22:10:01]

yep, lgtm

[kalyank, 2015-08-07 01:37:58]

The CQ bit was checked by kalyan.kondapally@intel.com

[commit-bot: I haz the power, 2015-08-07 01:39:09]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1278523005/60001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1278523005/60001

[commit-bot: I haz the power, 2015-08-07 03:05:41]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2015-08-07 03:06:32]

Patchset 4 (id:??) landed as
https://crrev.com/862a9d9c5b475b05b6bb1741f782b8459967591e
Cr-Commit-Position: refs/heads/master@{#342280}