Fix crashes when loading gles2.cc on browser startup, and during playback.

Fix crashes when loading gles2.cc on browser startup, and during playback.

Startup crash was being caused by deleting the Context3D object out from under a
running OmxVideoDecodeAccelerator, which would trigger a SEGV in the GPU
process, which would cause a Graphics3DContextLost callback to be fired on the
plugin, which is implemented as an assert(false).  The delete that kicked all
this off was not actually necessary, so removed it.

During-playback crash was being caused by a lack of synchronization between the
GPU command buffer mechanism and the PPAPI impl IPC mechanism (triggered much
more commonly in DEBUG mode, which explains why these weren't seen so much
before).

While debugging this cleaned up and documented some of the bogusity I found that
I didn't want to clean all in one CL:
- Emit type of unhandled message in GpuChannel, and document TODO to fix poor
  code structure that confused me.
- Emit EGL error code on CHECK-failures in EGL<->GLES translator.
- Simplify GpuVideoService a bit and remove redundant map lookups.
- Fix ppapi_tests.gypi: my r89636 was too ambitious and ran into the limits of
  my understanding of .gyp.  This version is less factored but has the benefit
  of producing actually-working example plugins.

BUG=none
TEST=manually running gles2.{html,cc} works every time.  No crashes.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=89775

[Ami GONE FROM CHROMIUM, 2011-06-20 18:39:19]

apatrick: please review gpu_channel.cc
brettw: FYI see ppapi_tests (FYI since you reviewed the CL that this fixes)
scherkus: please review everything.

[apatrick_chromium, 2011-06-20 18:47:54]

LGTM for gpu_channel.cc.

[brettw, 2011-06-20 19:21:55]

gypi LGTM

[Ami GONE FROM CHROMIUM, 2011-06-20 20:22:27]

+vjain,cfreeman,vhiremath as FYIs.

[vrk (LEFT CHROMIUM), 2011-06-20 22:22:17]

LGTM w/ nits

Coool I'm super excited for the noncrashiness!

http://codereview.chromium.org/7200033/diff/2001/content/common/gpu/gpu_chann...
File content/common/gpu/gpu_channel.cc (right):

http://codereview.chromium.org/7200033/diff/2001/content/common/gpu/gpu_chann...
content/common/gpu/gpu_channel.cc:333: // the GpuCommandBufferStub that owns the
commandbuffer GVDA is using.
Yes, this makes a lot more sense to me!

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.cc
File ppapi/examples/gles2/gles2.cc (right):

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:310: if (surface_)
delete these lines too

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:337: FinishGL();
is this really necessary here?

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:358: FinishGL();
Why do we serialize after PaintFinished instead of before? (i.e. before the call
to SwapBuffers)

[Ami GONE FROM CHROMIUM, 2011-06-20 22:27:22]

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.cc
File ppapi/examples/gles2/gles2.cc (right):

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:310: if (surface_)
On 2011/06/20 22:22:17, Victoria Kirst wrote:
> delete these lines too

Done.

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:337: FinishGL();
On 2011/06/20 22:22:17, Victoria Kirst wrote:
> is this really necessary here?

Yes, by observation (otherwise we get crash).

http://codereview.chromium.org/7200033/diff/2001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:358: FinishGL();
On 2011/06/20 22:22:17, Victoria Kirst wrote:
> Why do we serialize after PaintFinished instead of before? (i.e. before the
call
> to SwapBuffers)

The call to Render() means the decoder will not touch the texture until the next
call to ReusePictureBuffer, so the SwapBuffers is safe.
OTOH, we must guarantee the SwapBuffer is *actually* done before we return the
texture to the decoder, or we get crashes.

[scherkus (not reviewing), 2011-06-21 01:17:18]

LGTM

http://codereview.chromium.org/7200033/diff/9001/ppapi/examples/gles2/gles2.cc
File ppapi/examples/gles2/gles2.cc (right):

http://codereview.chromium.org/7200033/diff/9001/ppapi/examples/gles2/gles2.c...
ppapi/examples/gles2/gles2.cc:74: gles2_if_->Finish(context_->pp_resource());
wow yeah that's a lot of manual FinishGL-ing :\