986836 - Parent process crash when shutting down compositor @ 0xa5a5a5a4

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Gregor Wagner [:gwagner]]

Happens every time during shutdown after mochitests on emulators with debug gecko.

Main thread sends libxul.so!nsBaseWidget::DestroyCompositor()

Compositor thread crashes via libxul.so!mozilla::gl::GLLibraryEGL::fDestroySurface(void*, void*) [GLLibraryEGL.h:9fbc75ee6cfa : 187 + 0xd] at 0xa5a5a5a4

We finally have better stack coverage during mochitests so its hard to tell if this is a regression or not.


For example:
https://tbpl.mozilla.org/php/getParsedLog.php?id=36542254&full=1&branch=pine

There is also another assertion present in the _child_ process: MOZ_Assert: Assertion failure: work_queue_.empty(), at ../../../gecko/ipc/chromium/src/base/message_loop.cc:393
Maybe they are related?

Comment 1

[(no longer active)]

(In reply to Gregor Wagner [:gwagner] from comment #0)
> Happens every time during shutdown after mochitests on emulators with debug
> gecko.
> 
> Main thread sends libxul.so!nsBaseWidget::DestroyCompositor()
> 
> Compositor thread crashes via
> libxul.so!mozilla::gl::GLLibraryEGL::fDestroySurface(void*, void*)
> [GLLibraryEGL.h:9fbc75ee6cfa : 187 + 0xd] at 0xa5a5a5a4
> 
> We finally have better stack coverage during mochitests so its hard to tell
> if this is a regression or not.
> 
> 
> For example:
> https://tbpl.mozilla.org/php/getParsedLog.php?id=36542254&full=1&branch=pine
> 
> There is also another assertion present in the _child_ process: MOZ_Assert:
> Assertion failure: work_queue_.empty(), at
> ../../../gecko/ipc/chromium/src/base/message_loop.cc:393
> Maybe they are related?

It's plausible, as the latter shows that some work posted to the MessageLoop did not run.  It could be something that signals the parent process to do proper cleanup etc.

Comment 2

[Gregor Wagner [:gwagner]]

Blocking a blocker.

Comment 3

[(no longer active)]

Milan, can you please help us find an owner here?  Thanks!

Comment 4

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

New feature for 1.4, better support for debug emulators.  I need to get this approved before messing with 2.0 work.

Comment 5

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

Got it approved.  Benoit, all yours.

Comment 6

[Al Billings [:abillings - ex-MoCo]]

How far back does this issue go?

Comment 7

[Benoit Jacob [:bjacob] (mostly away)]

Got it, looking.

Comment 8

[Benoit Jacob [:bjacob] (mostly away)]

Attachment: Don't destroy the native framebuffer's eglSurface

Comment 9

[Jeff Muizelaar [:jrmuizel]]

Comment on attachment 8405006 [details] [diff] [review]
Don't destroy the native framebuffer's eglSurface

Review of attachment 8405006 [details] [diff] [review]:
-----------------------------------------------------------------

Make a more thorough comment

Comment 10

[Benoit Jacob [:bjacob] (mostly away)]

Attachment: Don't destroy the native framebuffer's eglSurface

Comment 11

[Gregor Wagner [:gwagner]]

I see more crashes with this patch applied: https://tbpl.mozilla.org/?showall=1&tree=Pine&rev=fed9d5539ef0
Are these followups?

Comment 12

[(no longer active)]

(In reply to Gregor Wagner [:gwagner] from comment #11)
> I see more crashes with this patch applied:
> https://tbpl.mozilla.org/?showall=1&tree=Pine&rev=fed9d5539ef0
> Are these followups?

Aren't those crashes bug 983489?

Comment 13

[Benoit Jacob [:bjacob] (mostly away)]

https://hg.mozilla.org/integration/b2g-inbound/rev/20e45d4b88f4

Comment 14

[Benoit Jacob [:bjacob] (mostly away)]

(In reply to Gregor Wagner [:gwagner] from comment #11)
> I see more crashes with this patch applied:
> https://tbpl.mozilla.org/?showall=1&tree=Pine&rev=fed9d5539ef0
> Are these followups?

These crashes seem different. Clearly, trying to run DEBUG builds when we didn't until now, is bound to run into many crashes. The patch just landed fixes one specific crash, the one originally described in comment 0.

(In reply to :Ehsan Akhgari (lagging on bugmail, needinfo? me!) from comment #12)
> (In reply to Gregor Wagner [:gwagner] from comment #11)
> > I see more crashes with this patch applied:
> > https://tbpl.mozilla.org/?showall=1&tree=Pine&rev=fed9d5539ef0
> > Are these followups?
> 
> Aren't those crashes bug 983489?

Just clicking on a few orange numbers on the TBPL link in comment 11, suggests that we are looking at more than one different bug here.

Comment 15

[Benoit Jacob [:bjacob] (mostly away)]

mwu points out bug 930884 which has a patch fixing the ICS emulator bug that we were working around here.

Comment 16

[Michael Wu [:mwu]]

Bug 930884 landed on March 15th, but this was filed a week after. So, it sounds like either bug 930884 didn't work or this patch won't make any difference and there's a different cause for this crash.

Comment 17

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/20e45d4b88f4

Comment 18

[Al Billings [:abillings - ex-MoCo]]

(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #17)
> https://hg.mozilla.org/mozilla-central/rev/20e45d4b88f4

Did this only affect trunk and therefore not need sec-approval? The b2g flags kind of indicate otherwise.

Comment 19

[Benoit Jacob [:bjacob] (mostly away)]

This bug only affects ICS emulator, not real devices/users, and unhiding it will benefit discussion on another bug.

Comment 20

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/7a869c7d9b19

Comment 21

[Ryan VanderMeulen [:RyanVM]]

Backed out for causing perma-hangs on B2G debug mochitest-3 and mochitest-4.
https://hg.mozilla.org/releases/mozilla-aurora/rev/1c796697dca3

https://tbpl.mozilla.org/php/getParsedLog.php?id=37963987&tree=Mozilla-Aurora
https://tbpl.mozilla.org/php/getParsedLog.php?id=37964461&tree=Mozilla-Aurora

Comment 22

[Benoit Jacob [:bjacob] (mostly away)]

Ah. So, what are next steps here?

Do we need to re-land this? If not, do we not need this at all anymore?

We probably need to sync on this with people working on bug 930884 ... Michael, do you know what we need to do here?

Comment 23

[Michael Wu [:mwu]]

Bug 930884 should've fixed it, and there haven't been any oranges reported on that bug after landing, so I think we're done. Relanding shouldn't be necessary.

Comment 24

[Ryan VanderMeulen [:RyanVM]]

Calling this wontfix for v1.4 based on comment 22 and comment 23.

Comment 25

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/b2g-inbound/rev/fdd739afef78

Comment 26

[Ryan VanderMeulen [:RyanVM]]

ignore that ^