Add GL strings to CastSysInfo interface and use to configure GPU

Add GL strings to CastSysInfo interface and use to configure GPU

Linux GPU info collector uses PCI scanning, which doesn't work on Chromecast.  So initial browser-process decisions on GPU feature blacklisting are incorrect.  This provides the info instead through CastSysInfo and will allow us for example to add exceptions to the accelerated canvas blacklist for our GPU.

BUG=

Committed: https://crrev.com/a69ea541d69b4e1d2ed77a5a546b5a2fc7ce7f0f
Cr-Commit-Position: refs/heads/master@{#328186}

[halliwell, 2015-04-30 05:58:52]

halliwell@chromium.org changed reviewers:
+ zmo@chromium.org

[halliwell, 2015-04-30 06:01:37]

On 2015/04/30 05:58:52, halliwell wrote:
> mailto:halliwell@chromium.org changed reviewers:
> + mailto:zmo@chromium.org

Hey Mo - this is something we discussed a while back in March, but I got
distracted with other things.

The CL here is definitely not finished - no tests and I haven't really tested
with it much.  Think of it more as a concrete proposal for how to solve this
problem of gpu config on Chromecast :)

Please let me know what you think of the general direction.  Then I can work on
the details :)

[Zhenyao Mo, 2015-04-30 21:42:24]

https://codereview.chromium.org/1117813002/diff/1/chromecast/browser/cast_bro...
File chromecast/browser/cast_browser_main_parts.cc (right):

https://codereview.chromium.org/1117813002/diff/1/chromecast/browser/cast_bro...
chromecast/browser/cast_browser_main_parts.cc:233:
content::GpuDataManager::GetInstance()->SetGpuDeviceName(
Please make sure this is called later than the GpuDataManager::Initialize().

https://codereview.chromium.org/1117813002/diff/1/content/browser/gpu/gpu_dat...
File content/browser/gpu/gpu_data_manager_impl.h (right):

https://codereview.chromium.org/1117813002/diff/1/content/browser/gpu/gpu_dat...
content/browser/gpu/gpu_data_manager_impl.h:90: void SetGpuDeviceName(const
std::string& vendor,
Per our discussion, you can just use existing SetGLStrings().
your "vendor" can be gl_vendor, and your "device" can be gl_renderer.  This will
make this CL much simpler. Just make sure the vendor supplies these two values
the same as they supply GL_VENDOR and GL_RENDERER in the OpenGL driver.
(Otherwise when you launch GPU process and create a context, these two fields
may be overwritten with a different value, and cause the blacklisting to change.

[halliwell, 2015-05-01 17:35:36]

halliwell@chromium.org changed reviewers:
+ gunsch@chromium.org, lcwu@chromium.org

[halliwell, 2015-05-01 17:37:09]

On 2015/04/30 21:42:24, Zhenyao Mo wrote:
>
https://codereview.chromium.org/1117813002/diff/1/chromecast/browser/cast_bro...
> File chromecast/browser/cast_browser_main_parts.cc (right):
> 
>
https://codereview.chromium.org/1117813002/diff/1/chromecast/browser/cast_bro...
> chromecast/browser/cast_browser_main_parts.cc:233:
> content::GpuDataManager::GetInstance()->SetGpuDeviceName(
> Please make sure this is called later than the GpuDataManager::Initialize().
> 

I put the SetGLStrings in PreMainMessageLoopRun now, it looks like that's after
Initialize (and verified it works).

>
https://codereview.chromium.org/1117813002/diff/1/content/browser/gpu/gpu_dat...
> File content/browser/gpu/gpu_data_manager_impl.h (right):
> 
>
https://codereview.chromium.org/1117813002/diff/1/content/browser/gpu/gpu_dat...
> content/browser/gpu/gpu_data_manager_impl.h:90: void SetGpuDeviceName(const
> std::string& vendor,
> Per our discussion, you can just use existing SetGLStrings().
> your "vendor" can be gl_vendor, and your "device" can be gl_renderer.  This
will
> make this CL much simpler. Just make sure the vendor supplies these two values
> the same as they supply GL_VENDOR and GL_RENDERER in the OpenGL driver.
> (Otherwise when you launch GPU process and create a context, these two fields
> may be overwritten with a different value, and cause the blacklisting to
change.

Thanks Mo - much simpler with SetGLStrings :)

[Zhenyao Mo, 2015-05-01 18:12:03]

LGTM

Please also verify with about:gpu page that the strings show up correctly.

https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
File chromecast/browser/cast_browser_main_parts.cc (right):

https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
chromecast/browser/cast_browser_main_parts.cc:263:
content::GpuDataManager::GetInstance()->SetGLStrings(
Please add a comment here that this has to be called after
content::GpuDataManager::Initialize(). In case someone in the future moves this
around to break the contract.

[gunsch, 2015-05-01 18:23:45]

lgtm

https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
File chromecast/browser/cast_browser_main_parts.cc (right):

https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
chromecast/browser/cast_browser_main_parts.cc:263:
content::GpuDataManager::GetInstance()->SetGLStrings(
On 2015/05/01 18:12:03, Zhenyao Mo wrote:
> Please add a comment here that this has to be called after
> content::GpuDataManager::Initialize(). In case someone in the future moves
this
> around to break the contract.

I don't suppose we have a public API via which we can assert that GpuDataManager
is initialized by this point?

[Zhenyao Mo, 2015-05-01 18:29:14]

On 2015/05/01 18:23:45, gunsch wrote:
> lgtm
> 
>
https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
> File chromecast/browser/cast_browser_main_parts.cc (right):
> 
>
https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
> chromecast/browser/cast_browser_main_parts.cc:263:
> content::GpuDataManager::GetInstance()->SetGLStrings(
> On 2015/05/01 18:12:03, Zhenyao Mo wrote:
> > Please add a comment here that this has to be called after
> > content::GpuDataManager::Initialize(). In case someone in the future moves
> this
> > around to break the contract.
> 
> I don't suppose we have a public API via which we can assert that
GpuDataManager
> is initialized by this point?

Not really. Let me see if I can fix that, but that should not affect this CL.

[halliwell, 2015-05-01 18:33:17]

On 2015/05/01 18:29:14, Zhenyao Mo wrote:
> On 2015/05/01 18:23:45, gunsch wrote:
> > lgtm
> > 
> >
>
https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
> > File chromecast/browser/cast_browser_main_parts.cc (right):
> > 
> >
>
https://codereview.chromium.org/1117813002/diff/20001/chromecast/browser/cast...
> > chromecast/browser/cast_browser_main_parts.cc:263:
> > content::GpuDataManager::GetInstance()->SetGLStrings(
> > On 2015/05/01 18:12:03, Zhenyao Mo wrote:
> > > Please add a comment here that this has to be called after
> > > content::GpuDataManager::Initialize(). In case someone in the future moves
> > this
> > > around to break the contract.
> > 
> > I don't suppose we have a public API via which we can assert that
> GpuDataManager
> > is initialized by this point?
> 
> Not really. Let me see if I can fix that, but that should not affect this CL.

I will add a comment for now.  Another option would be for GpuDataManager itself
to assert if SetGLStrings is called before Initialize?

[halliwell, 2015-05-04 21:06:06]

The CQ bit was checked by halliwell@chromium.org

[halliwell, 2015-05-04 21:06:06]

The patchset sent to the CQ was uploaded after l-g-t-m from zmo@chromium.org,
gunsch@chromium.org
Link to the patchset: https://codereview.chromium.org/1117813002/#ps40001
(title: "Add comment on ordering of GpuDataManager calls")

[commit-bot: I haz the power, 2015-05-04 21:07:06]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1117813002/40001

[commit-bot: I haz the power, 2015-05-04 21:20:19]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-05-04 21:22:13]

Patchset 3 (id:??) landed as
https://crrev.com/a69ea541d69b4e1d2ed77a5a546b5a2fc7ce7f0f
Cr-Commit-Position: refs/heads/master@{#328186}