Remove the blacklist on newer Nvidia Linux drivers

Remove the blacklist on newer Nvidia Linux drivers

The initial reason for the blacklist was a bug that is now fixed.
A workaround was first developed in Chromium and then later
Nvidia actually fixed the root of the problem in their driver.
This patch removes the blacklist for 331.38 and newer drivers.

BUG=469979
Committed: https://crrev.com/5486eb809770fa68ac25c9e4e30a1cdf76225877
Cr-Commit-Position: refs/heads/master@{#322527}

[dbdaniel42, 2015-03-23 14:27:45]

dbdaniel42@gmail.com changed reviewers:
+ kbr@chromium.org

[Ken Russell (switch to Gerrit), 2015-03-23 18:11:25]

kbr@chromium.org changed reviewers:
+ ccameron@chromium.org

[Ken Russell (switch to Gerrit), 2015-03-23 18:11:26]

Can you point to a specific NVIDIA driver release that improved its behavior
under low VRAM conditions? If there is one, then the blacklist entry should be
updated to apply only to earlier driver releases, rather than removing the entry
entirely.

[dbdaniel42, 2015-03-24 01:19:24]

On 2015/03/23 18:11:26, Ken Russell wrote:
> Can you point to a specific NVIDIA driver release that improved its behavior
> under low VRAM conditions? If there is one, then the blacklist entry should be
> updated to apply only to earlier driver releases, rather than removing the
entry
> entirely.

If I'm understanding the comments on the bug report correctly, now that we're
using Aura there's a workaround in place so that we won't be affected by
Nvidia's bug even in the bad drivers.

"We'll be able to enable FCM when Aura ships, because (IIUC) that will only have
1 GLX window, which will sidestep NVIDIA's bug."

https://code.google.com/p/chromium/issues/detail?id=145600#c110

As far as the drivers go, I'm seeing two possible fixes on Nvidia's end.  The
first was in 319.17

"Fixed a memory leak that occurred when destroying a GLX window but not its
associated X window."

http://www.nvidia.com/object/linux-display-amd64-319.17-driver

And the second in 331.38

"Fixed a bug that caused the X server to crash if video memory is exhausted and
the GPU does not support rendering to system memory."

http://www.nvidia.com/Download/driverResults.aspx/72249/en-us

[Ken Russell (switch to Gerrit), 2015-03-24 02:21:03]

On 2015/03/24 01:19:24, dbdaniel42 wrote:
> On 2015/03/23 18:11:26, Ken Russell wrote:
> > Can you point to a specific NVIDIA driver release that improved its behavior
> > under low VRAM conditions? If there is one, then the blacklist entry should
be
> > updated to apply only to earlier driver releases, rather than removing the
> entry
> > entirely.
> 
> If I'm understanding the comments on the bug report correctly, now that we're
> using Aura there's a workaround in place so that we won't be affected by
> Nvidia's bug even in the bad drivers.
> 
> "We'll be able to enable FCM when Aura ships, because (IIUC) that will only
have
> 1 GLX window, which will sidestep NVIDIA's bug."
> 
> https://code.google.com/p/chromium/issues/detail?id=145600#c110
> 
> As far as the drivers go, I'm seeing two possible fixes on Nvidia's end.  The
> first was in 319.17
> 
> "Fixed a memory leak that occurred when destroying a GLX window but not its
> associated X window."
> 
> http://www.nvidia.com/object/linux-display-amd64-319.17-driver

This fix was motivated by ccameron's discovery of the leak, as you mentioned
above. It isn't directly related to this blacklist entry. You're correct that
the switch to Aura has improved the situation even if this bug were still
present, but force-compositing-mode has been enabled for a long time in
Chromium. This blacklist entry only disables hardware accelerated video
decoding, Flash's use of the 3D APIs, and Flash's Stage 3D.

> And the second in 331.38
> 
> "Fixed a bug that caused the X server to crash if video memory is exhausted
and
> the GPU does not support rendering to system memory."
> 
> http://www.nvidia.com/Download/driverResults.aspx/72249/en-us

This behavior also isn't what we were seeing -- we observed the GPU slowing to a
crawl when lots of video memory was allocated. However, it's possible this fix
may have improved things.

Rather than delete this blacklist entry wholesale, since it was found to be
necessary with older drivers, let's do the following:

 - File a bug about limiting this blacklist entry to older drivers and block it
on Issue 145600 (email me the bug ID once you've filed it; I can help update the
fields)
 - Change this CL to apply the blacklist entry only to drivers older than 331.38
using the driver_version directive
 - Update the comment and bug ID of the blacklist entry to point to your new bug

Also please make sure that if you run Chrome with --ignore-gpu-blacklist on your
hardware that it behaves as expected. That will have much of the effect of
removing this blacklist entry.

[dbdaniel42, 2015-03-24 04:10:18]

Done.  Yes, it works correctly with --ignore-gpu-blacklist but I'm on a GTX 770
with 2GB so if it's an issue with running out of VRAM I probably won't be
affected.  Btw I'm also sending you a CL dealing with Nouveau.  I've been doing
some testing with both drivers.

[Ken Russell (switch to Gerrit), 2015-03-24 22:20:20]

This change looks better. Could you please update the issue description?

@ccameron, any comments?

[dbdaniel42, 2015-03-25 02:09:34]

On 2015/03/24 22:20:20, Ken Russell wrote:
> This change looks better. Could you please update the issue description?
> 
> @ccameron, any comments?

updated

[Ken Russell (switch to Gerrit), 2015-03-26 01:47:33]

https://codereview.chromium.org/1032483002/diff/20001/gpu/config/software_ren...
File gpu/config/software_rendering_list_json.cc (right):

https://codereview.chromium.org/1032483002/diff/20001/gpu/config/software_ren...
gpu/config/software_rendering_list_json.cc:21: "version": "10.0",
Sorry, I forgot to mention: please update this number. You'll need to sync as
it's already been changed relative to your CL.

[dbdaniel42, 2015-03-26 12:20:44]

On 2015/03/26 01:47:33, Ken Russell wrote:
>
https://codereview.chromium.org/1032483002/diff/20001/gpu/config/software_ren...
> File gpu/config/software_rendering_list_json.cc (right):
> 
>
https://codereview.chromium.org/1032483002/diff/20001/gpu/config/software_ren...
> gpu/config/software_rendering_list_json.cc:21: "version": "10.0",
> Sorry, I forgot to mention: please update this number. You'll need to sync as
> it's already been changed relative to your CL.

version number updated

[Ken Russell (switch to Gerrit), 2015-03-27 00:47:37]

lgtm

I'll send this to the commit queue for you.

[Ken Russell (switch to Gerrit), 2015-03-27 00:47:42]

The CQ bit was checked by kbr@chromium.org

[commit-bot: I haz the power, 2015-03-27 00:47:56]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1032483002/40001

[commit-bot: I haz the power, 2015-03-27 04:21:04]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2015-03-27 04:22:01]

Patchset 3 (id:??) landed as
https://crrev.com/5486eb809770fa68ac25c9e4e30a1cdf76225877
Cr-Commit-Position: refs/heads/master@{#322527}