1286840 - Tabs go black after repeatedly toggle full screen

Status: RESOLVED WORKSFORME

(Core :: Graphics: Layers, defect, P3)

Description

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

[Affected versions]:
- latest Nightly 50.0a1 (2016-07-14)
- latest Aurora 49.0a2 (2016-07-14)
- 48.0b7 build1 (20160711002726)
- 47.0.1 build1 (20160623154057)
 
[Affected platforms]:
- Windows 10 x64

[Steps to reproduce]:
1. Launch Firefox with a clean profile
2. Open several random pages in different tabs
3. Go to http://mystaticself.com/ and repeatedly toggle full screen (by pressing F11 key)
4. Switch to the other open tabs and toggle full screen (by pressing F11 key)

[Expected result]:
- The browser successfully toggles full screen option, no matter how many times it is required; there are no visual issues

[Actual result]:
- (step 3) After several attempts, the active tab turns black
- (step 4) Switching to the other open tabs, their content also turns black 
- See the screencast https://drive.google.com/file/d/0B0nYKG6PRiCcanUzVEtPR1NPb1k/view?usp=sharing

[Regression range]:
- I will investigate this as soon as possible, but as far as I managed to search into, this is not a recent regression

[Additional notes]:
- The issue is reproducible with e10s on/off for 50, 49 and 48 builds - the difference is that with e10s off the browser hangs and gets unresponsive, but only for a few moments (the graphics driver is affected)
- For 47 build, the issue is reproducible only with e10s on 
- Under certain conditions, Firefox crashes (https://crash-stats.mozilla.com/report/index/1f27df09-3307-4809-aaa5-3b9df2160713), but I reproduced it only once
- The issue is reproduced under Intel(R) HD Graphics 4600 - it seems that the issue depends on the graphics card model; I will further investigate this

Comment 1

[David Anderson [:dvander] - inactive, e-mail if emergency]

Looks WebGL related.

Comment 2

[u279076]

Too late for Fx47 and I suspect too late for Fx48 at this point.

Iulia please find a regression window ASAP.

Comment 3

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

Hello! Sorry for the delay! 
Regression range (I used the mozilla-aurora project branch because of some errors thrown by mozregression on old mozilla-central builds)
======================
Last good revision: 6c52c5091ef26e2327e5c7172a3f12809f224517
First bad revision: 8fffb73940213a6b5b14d2cb9e7f49857eea7d6c
Pushlog:
https://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?fromchange=6c52c5091ef26e2327e5c7172a3f12809f224517&tochange=8fffb73940213a6b5b14d2cb9e7f49857eea7d6c 

Potential regressor: 
Bug 1233784 - Disallow illegal characters in cookie names. r=jduell a=sylvestre (I have to mention that I am not authorized to access this bug).

Comment 4

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

The issue is also reproducible on latest Nightly 51.0a1 (2016-08-30).

Comment 5

[Jeff Muizelaar [:jrmuizel]]

Mason Chang — Bug 1233576. Ensure vsync timestamps are monotonic on Windows 10. r=kats a=sylvestre 
Jeff Gilbert — Bug 1233557 - Allow RGB8 to be renderable again for web-compat. - r=jrmuizel a=ritu 

These are the more likely candidates.

Comment 6

[u279076]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #5)
> Mason Chang — Bug 1233576. Ensure vsync timestamps are monotonic on Windows
> 10. r=kats a=sylvestre 
> Jeff Gilbert — Bug 1233557 - Allow RGB8 to be renderable again for
> web-compat. - r=jrmuizel a=ritu 
> 
> These are the more likely candidates.

Mason and/or Jeff, can one of you take a look at this?

Comment 7

[Mason Chang [Inactive] [:mchang]]

I am unable to reproduce this locally. Can you try doing a local build and backing out either bug? I'm doubtful it is bug 1233576 as we've rewritten that code a bit already and the patch properly fixes just starting vsync. Since this site is continually rendering, we shouldn't be hitting that code.

Comment 8

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Mason Chang [:mchang] from comment #7)
> I am unable to reproduce this locally. Can you try doing a local build and
> backing out either bug? I'm doubtful it is bug 1233576 as we've rewritten
> that code a bit already and the patch properly fixes just starting vsync.
> Since this site is continually rendering, we shouldn't be hitting that code.

Hello! No, I don't think I can do this. But I can try, if you please give some instructions. Thank you!

Comment 9

[Ryan VanderMeulen [:RyanVM]]

bug 1233576 (rev 8e41d75f63fa): https://treeherder.mozilla.org/#/jobs?repo=try&revision=9155e836338b
bug 1233557 (rev 001ddf39af08): https://treeherder.mozilla.org/#/jobs?repo=try&revision=ce808a585105

When these builds finish, you should be able to run them via |mozregression --launch <revision>|

Comment 10

[Ryan VanderMeulen [:RyanVM]]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #9)

These builds hit Windows infra bustage. New pushes:

Bug 1233576 (rev 8e41d75f63fa): https://treeherder.mozilla.org/#/jobs?repo=try&revision=834514c70e0e
Bug 1233557 (rev 001ddf39af08): https://treeherder.mozilla.org/#/jobs?repo=try&revision=2948c5e591dd

Comment 11

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #10)
> (In reply to Ryan VanderMeulen [:RyanVM] from comment #9)
> 
> These builds hit Windows infra bustage. New pushes:
> 
> Bug 1233576 (rev 8e41d75f63fa):
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=834514c70e0e
> Bug 1233557 (rev 001ddf39af08):
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=2948c5e591dd

The issue is reproducible with both of above builds.

Comment 12

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

So, it isn't bug 1233576, but could still be bug 1233557 - Ryan, can you do a try build for rev 56de1b2cdd2b, just before bug 1233557 got pushed?  Not sure I know how to do it :)

Comment 13

[Ryan VanderMeulen [:RyanVM]]

rev 56de1b2cdd2b: https://treeherder.mozilla.org/#/jobs?repo=try&revision=e4abd5eb909f

Comment 14

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #13)
> rev 56de1b2cdd2b:
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=e4abd5eb909f

Thanks, Ryan! :) 
This build doesn't load the tested page (http://mystaticself.com/) completely and so, the problem doesn't reproduce.

Comment 15

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

Hi Ryan, Milan, please see comment 14. Do we need to give SV folks another try build?

Comment 16

[Ryan VanderMeulen [:RyanVM]]

Let's see if bisecting on trunk works any better.

rev 82cd6a486698 (push prior):  https://treeherder.mozilla.org/#/jobs?repo=try&revision=8023eda23670
rev 40bdbd4eff2a (bug 1233557): https://treeherder.mozilla.org/#/jobs?repo=try&revision=5b6ab0809a1f

Comment 17

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #16)
> Let's see if bisecting on trunk works any better.
> 
> rev 82cd6a486698 (push prior): 
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=8023eda23670
> rev 40bdbd4eff2a (bug 1233557):
> https://treeherder.mozilla.org/#/jobs?repo=try&revision=5b6ab0809a1f

Tried again to reproduce: 
- with rev 82cd6a486698 - the issue isn't reproducible (the site is not completely loaded) 
- with rev 40bdbd4eff2a - the issue is reproducible

Comment 18

[Ryan VanderMeulen [:RyanVM]]

Calling this a regression from bug 1233557 in that case, thanks for testing Iulia!

Comment 19

[Gerry Chang [:gchang]]

Hi Peter,
Per comment #18, can you help take a look at this one?

Comment 20

[Peter Chang[:pchang]]

Michael, please help to check this.

Comment 21

[Michael Leu[:Lenzak](UTC+8)]

Hi Iulia,

I cannot reproduce it on Windows 10 with HD4000 and i7-3770.
Can you post your about:support here?
Maybe it's a driver-related issue.

Comment 22

[Michael Leu[:Lenzak](UTC+8)]

I prepared a platform with HD4600 but still cannot reproduce it.
Here is my about:support here, the gfx driver is 20.19.15.4531.

BTW, I also tried NVidia graphics card (750Ti) and no luck, either.

Graphics
Features
Compositing	Direct3D 11
Asynchronous Pan/Zoom	wheel input enabled; touch input enabled
WebGL Renderer	Google Inc. -- ANGLE (Intel(R) HD Graphics 4600 Direct3D11 vs_5_0 ps_5_0)
WebGL2 Renderer	Google Inc. -- ANGLE (Intel(R) HD Graphics 4600 Direct3D11 vs_5_0 ps_5_0)
Hardware H264 Decoding	Yes; Failed to create D3D11 device for decoder; Using D3D9 API
Audio Backend	wasapi
Direct2D	true
DirectWrite	true (10.0.10240.17146)
GPU #1
Active	Yes
Description	Intel(R) HD Graphics 4600
Vendor ID	0x8086
Device ID	0x0412
Driver Version	20.19.15.4531
Driver Date	9-29-2016
Drivers	igdumdim64 igd10iumd64 igd10iumd64 igd12umd64 igdumdim32 igd10iumd32 igd10iumd32 igd12umd32
Subsys ID	00000000
RAM	Unknown
Diagnostics
AzureCanvasAccelerated	0
AzureCanvasBackend	skia
AzureContentBackend	skia
AzureFallbackCanvasBackend	cairo
GPUProcessPid	8144
GPUProcess	
Decision Log
D3D9_COMPOSITING	
disabled by default: Disabled by default

Comment 23

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

I reproduced the issue on Windows 10 x64, with Intel Core i5-4570 CPU and using latest Nightly 53.0a1 (2016-11-28). 
My about:support Graphics section (after the issue was reproduced):

Graphics
Features
Compositing	Direct3D 11
Asynchronous Pan/Zoom	wheel input enabled; touch input enabled
WebGL Renderer	Google Inc. -- ANGLE (Intel(R) HD Graphics 4600 Direct3D11 vs_5_0 ps_5_0)
WebGL2 Renderer	Google Inc. -- ANGLE (Intel(R) HD Graphics 4600 Direct3D11 vs_5_0 ps_5_0)
Hardware H264 Decoding	Yes; Failed to create D3D11 device for decoder; Using D3D9 API
Audio Backend	wasapi
Direct2D	true
DirectWrite	true (10.0.10586.633)
GPU #1
Active	Yes
Description	Intel(R) HD Graphics 4600
Vendor ID	0x8086
Device ID	0x0412
Driver Version	20.19.15.4285
Driver Date	9-11-2015
Drivers	igdumdim64 igd10iumd64 igd10iumd64 igd12umd64 igdumdim32 igd10iumd32 igd10iumd32 igd12umd32
Subsys ID	18e7103c
RAM	Unknown
Diagnostics
AzureCanvasAccelerated	0
AzureCanvasBackend	skia
AzureContentBackend	skia
AzureFallbackCanvasBackend	cairo
GPUProcessPid	7304
GPUProcess	
failures	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
Decision Log
D3D9_COMPOSITING	
disabled by default: Disabled by default
Failure Log
(#0) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3207) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3208) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3209) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3210) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3211) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3212) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3213) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3214) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3215) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3216) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3217) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3218) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3219) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.
(#3220) 	CP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6
(#3221) 	CP+[GFX1-]: (gfxWindowsPlatform) Finished device reset.

Comment 24

[Michael Leu[:Lenzak](UTC+8)]

"failuresCP+[GFX1-]: (gfxWindowsPlatform) Detected device reset: 6"

It seems to be the reason why we got black screen.

Since the PC is a desktop, a gfx device reset usually means that something goes wrong.

And the DeviceResetReason is 6 which seems to be Out Of Memory error.

However, I have checked my about:memory, the webgl site does not consume much memory (about 30~40MB).

I think maybe it is a driver bug.

Iulia, could you update your Intel GFX driver to the latest and test this bug again?

Thank you :)

Comment 25

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Michael Leu[:lenzak800](UTC+8) from comment #24)
> Iulia, could you update your Intel GFX driver to the latest and test this
> bug again?
> 
> Thank you :)

I tried to update the gpu driver before I posted comment 23. I tried again in several ways, but Windows refuses to update it (it states that the driver software for this product is up to date). A possible cause is this issue https://answers.microsoft.com/en-us/windows/forum/windows_10-hardware/intel-graphics-4600-driver-2019154531/57aed147-16a1-48ca-be94-fe80e15aa3bd. Maybe Microsoft stopped pushing the 20.19.15.4531 driver update.

Comment 26

[Michael Leu[:Lenzak](UTC+8)]

Sorry for late reply,

I think you should download the latest Intel GFX driver from their official website.
https://downloadcenter.intel.com/download/26404/Intel-Graphics-Driver-15-45-

Sometimes the driver database from Microsoft doesn't update as fast as hardware vendors.

Comment 27

[Ryan VanderMeulen [:RyanVM]]

Also note that this isn't an action we can realistically expect our users to take.

Comment 28

[Michael Leu[:Lenzak](UTC+8)]

I know, I just want to know whether this bug is caused by driver or not.

Comment 29

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

Too late to fix in 50.1.0 release

Comment 30

[Iulia Cristescu, QA [:JuliaC] (away, please needinfo? [e-mail])]

(In reply to Michael Leu[:lenzak800](UTC+8) from comment #26)
> Sorry for late reply,
> 
> I think you should download the latest Intel GFX driver from their official
> website.
> https://downloadcenter.intel.com/download/26404/Intel-Graphics-Driver-15-45-
> 
> Sometimes the driver database from Microsoft doesn't update as fast as
> hardware vendors.

Tried this, too, but I encountered this error "The driver being installed is not validated for this computer. Obtain the appropriate driver from the computer's manufacturer. Setup will exit.". This means I should uninstall the manufacturer's driver, then install the Intel Graphics Driver. However, I'll risk losing functionality or causing other technical issues.

Comment 31

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

Too late for a fix for 53, as we are in the last week of the 53 beta cycle.

Comment 32

[Gerry Chang [:gchang]]

Mark 54 won't fix as 54 was released.

Comment 33

[Gregory Pappas [:gregp]]

Unable to reproduce