1227548 - [XP] Firefox crashes or doesn't play WebGL content after unlocking the computer

Status: VERIFIED FIXED

(Core :: Graphics: CanvasWebGL, defect)

Description

[Petruta Horea [:phorea], Desktop QA]

Reproduced with Firefox 43 beta 6, latest DevEdition 44.0a2 and Nightly 45.0a1 2015-11-24 under Win XP 32-bit.
The issue doesn't reproduce under Win 7 64-bit, Mac OS X 10.9.5 and Ubuntu 14.04 32-bit.

Steps to reproduce:
1. Have tabs containing WebGL samples open (eg http://madebyevan.com/webgl-water/ , http://webglsamples.org/aquarium/aquarium.html , http://www.wordsaretoys.com/apps/fissure/)
2. While the game/animation is playing, lock the computer (Win Logo + L)
3. Login and check the WebGL sites

Expected results: WebGL continues to play

Actual results:
Firefox crashes (with no reliable crash signature -/throttling) or the WebGL content is not rendered (context lost log.js:55:5)

In Firefox 40.0.3 WebGL content is not loaded, the following message appears in the browser console:
Error: WebGL: Refused to create native OpenGL context because of blacklisting. webgl.js:156:16
Error: WebGL: WebGL creation failed.

In Firefox 41.0.2 I encountered the crash only once (https://crash-stats.mozilla.com/about/throttling) so it seems that this is not a regression although it's easier to reproduce on Firefox 43 beta.

Graphics
Adapter Description	Intel(R) HD Graphics
Adapter Drivers	igxprd32
Adapter RAM	Unknown
Asynchronous Pan/Zoom	none
Device ID	0x0152
DirectWrite Enabled	false (0.0.0.0)
Driver Date	1-16-2012
Driver Version	6.14.10.5398

Comment 1

[Petruta Horea [:phorea], Desktop QA]

On another Win XP x86 machine, the crashes were correctly registered:

https://crash-stats.mozilla.com/report/index/8cc0b935-db56-4f3d-b677-799272151124
https://crash-stats.mozilla.com/report/index/bp-ef9eb173-d009-465f-811b-33c5b2151124

Graphics
Adapter Description	ATI Radeon 3000 Graphics
Adapter Drivers	ati2dvag
Adapter RAM	Unknown
Asynchronous Pan/Zoom	none
Device ID	0x9616
DirectWrite Enabled	false (0.0.0.0)
Driver Date	7-3-2012
Driver Version	8.970.100.3000

Comment 2

[Robert Kaiser]

Interestingly, the crash signature of the crashes mentioned in comment #1 also exists on Win7 and higher and on Firefox 42, even if the specific case to reproduce might not work there.

Comment 3

[Liz Henry (:lizzard) (relman/hg->git project)]

Tracking this as it's a crash that may have become more common recently. It doesn't look very high volume on beta, a bit more so on 42 release.

Comment 4

[Arthur K. (he/him)]

(In reply to Petruta Rasa [QA] [:petruta] from comment #0)
> Graphics
> Adapter Description	Intel(R) HD Graphics
> Adapter Drivers	igxprd32
> Adapter RAM	Unknown
> Asynchronous Pan/Zoom	none
> Device ID	0x0152
> DirectWrite Enabled	false (0.0.0.0)
> Driver Date	1-16-2012
> Driver Version	6.14.10.5398

It might be fixed by installing driver 14.51.11.5437 (6.14.10.5437) from 2/13/2013.

Comment 5

[Arthur K. (he/him)]

(In reply to Petruta Rasa [QA] [:petruta] from comment #1)
> On another Win XP x86 machine, the crashes were correctly registered:
> 
> https://crash-stats.mozilla.com/report/index/8cc0b935-db56-4f3d-b677-
> 799272151124
> https://crash-stats.mozilla.com/report/index/bp-ef9eb173-d009-465f-811b-
> 33c5b2151124
> 
> Graphics
> Adapter Description	ATI Radeon 3000 Graphics
> Adapter Drivers	ati2dvag
> Adapter RAM	Unknown
> Asynchronous Pan/Zoom	none
> Device ID	0x9616
> DirectWrite Enabled	false (0.0.0.0)
> Driver Date	7-3-2012
> Driver Version	8.970.100.3000

This could likely be fix by updating to 8.97.100.7 from Catalyst 14.4 from 04/25/2014.

Comment 6

[Petruta Horea [:phorea], Desktop QA]

(In reply to Arthur K. from comment #4)
> It might be fixed by installing driver 14.51.11.5437 (6.14.10.5437) from
> 2/13/2013.

After updating, the browser only crashed once. 
Driver Date	2-4-2013
Driver Version	6.14.10.5437

(In reply to Arthur K. from comment #5)
> (In reply to Petruta Rasa [QA] [:petruta] from comment #1)
> > On another Win XP x86 machine, the crashes were correctly registered:
> This could likely be fix by updating to 8.97.100.7 from Catalyst 14.4 from
> 04/25/2014.

I updated to: 
Driver Date	11-16-2012
Driver Version	8.970.100.7000
and still had crashes on almost every try.

Comment 7

[Liz Henry (:lizzard) (relman/hg->git project)]

Too late for a fix for 43. This doesn't look very high volume but the signature is still showing up on release as well as on 44 and 45.

Comment 8

[Ritu Kothari (:ritu) (Inactive, please n-i to RyanVM, jcristau, or pascal)]

Milan, this is a pretty low volume crash, so for 44 it seems like a wontfix. I am just pinging you on whether we need to re-prioritize this one as it hasn't gotten much attention since mid December. Thanks!

Comment 9

[Milan Sreckovic [:milan] (needinfo for best results)]

We'll get to it; not the highest urgency based on some other bugs, but we want to get to it.

Comment 10

[Edwin Flores [inactive from 2016-12-01] [:eflores] [:edwin]]

Attachment: 1227548.patch

When our GLContext dies, the subclass destructor calls GLContext::MarkDestroyed. This zeroes out |mSymbols|, so no GL calls can be made.

Sometimes the GLContext can still have a GLScreenBuffer, which still has a DrawBuffer or ReadBuffer. When these latter two are destroyed they try to call glDelete{Frame,Render}buffers, which leads to a crash because |mSymbols| is zeroed.

I'm not sure why the GLContext is dying here -- probably driver weirdness -- but the least we can do here is not crash. The attached patch just changes ~{Read,Draw}Buffer() to check the return value of GLContext::MakeCurrent().

Comment 11

[Kelsey Gilbert [:jgilbert]]

Comment on attachment 8714902 [details] [diff] [review]
1227548.patch

Review of attachment 8714902 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/gl/GLScreenBuffer.cpp
@@ +840,5 @@
>  
>  DrawBuffer::~DrawBuffer()
>  {
> +    if (!mGL->MakeCurrent()) {
> +      return;

4-space indents.
Also you don't need the brackets for returns in this area of the code.

Comment 12

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/8e6090909a8a

Comment 13

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/8e6090909a8a

Comment 14

[Edwin Flores [inactive from 2016-12-01] [:eflores] [:edwin]]

Comment on attachment 8714902 [details] [diff] [review]
1227548.patch

Approval Request Comment
[Feature/regressing bug #]: WebGL.
[User impact if declined]: Crashes on some WebGL content.
[Describe test coverage new/current, TreeHerder]: On m-c.
[Risks and why]: None. Just adds a sanity check.
[String/UUID change made/needed]: None.

Comment 15

[Sylvestre Ledru [:Sylvestre]]

Comment on attachment 8714902 [details] [diff] [review]
1227548.patch

Fix a crash, taking it.
Should be in 45 beta 3.

Comment 16

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/e5f8331bcd2c

Comment 17

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/releases/mozilla-beta/rev/e0e447f81f40

Comment 18

[Petruta Horea [:phorea], Desktop QA]

I could no longer reproduce the crash with Firefox 45 beta 6, latest Aurora 46.a02 and latest Nightly 47.0a1 2016-02-18.
Also there are no new crashes registered in Socorro after the fix landed, so I'm marking this as verified.