1202748 - [e10s] 42 crashed on startup on Mac OSX 10.10.5

Status: RESOLVED DUPLICATE of bug 1218070

(Core :: Graphics, defect)

Description

[Nils Ohlmeier [:drno]]

DevEdition with e10s on crashed during startup on my MacBook Pro with 10.10.5:

https://crash-stats.mozilla.com/report/index/2b9f92a9-251c-47ab-b4e1-c56cc2150906

Comment 1

[Tracy Walker [:tracy]]

Nils, is your DevEdition up to date?  I believe this should have been fixed, a week ago, in bug 1198594.

Comment 2

[Nils Ohlmeier [:drno]]

From the build ID 20150904004016 in the crash report it looks like I was running the DevEdition build from Friday. Certainly well after Ryan uplifted bug 1198594 to DevEdition https://bugzilla.mozilla.org/show_bug.cgi?id=1198594#c7

Comment 3

[u279076]

Based on stats this has not moved significantly since bug 1198594 was fixed. Looking at that bug the stack here is much different. Also looking at other stacks reported there's a lot of variety. I'm guessing this signature is conflating a bunch of different issues. I'm not sure of a good way to separate these out though.

For what it's worth here are the stats for the conflated signature:
* it is the #2 Mac OS X crash on Dev Edition, accounting for 6.58% of the crashes
* it is 100% isolated to Mac OS 10.10 (no other Mac OS versions are affected)
* it is roughly 50-50 distributed between Firefox and Thunderbird
* top URLs are facebook, yahoo, youtube, and gmail

Comment 4

[Jim Mathies [:jimm]]

Top crash in compositor code on dev edition. Milan, is this something you are tracking?

Comment 5

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

I can look at it and scratch my head.

As Anthony points out, there are a number of crashes showing up under here, that kernel address may very well be a generic "kernel panic" type of thing?  For example, https://crash-stats.mozilla.com/report/index/ece6a4f9-aa32-4350-bd21-68d9e2150910 is in GC, https://crash-stats.mozilla.com/report/index/4caeaf8c-2492-4fce-96f5-a31292150910 is in some special place, https://crash-stats.mozilla.com/report/index/52090eb8-46c4-43bd-b0e8-d49112150910 in print.

Now, we have seen problems in the driver when the system switches from integrated to discrete graphics; perhaps that is somehow common to all bugs, but otherwise not sure how make this actionable.

Nils, if you force your graphics to discrete all the time, does this problem go away?

Comment 6

[Nils Ohlmeier [:drno]]

How do I force my card? By something like https://gfx.io/ ?

FWIW I first encountered this crash bug here. After restarting Fx it crashed right away again in bug 1202750, which is using the discrete Nvidia card (I filled two separate bugs because of the different crash signatures). So I guess the short answer is no, as it crashed also with the discrete card (even though I don't know why the graphic card switched in between the two starts).

But these two crashes were also not repeatable for me. I have since turned off e10s in my DevEdition profile, because that seems to cause a lot of graphics related crashes for me.

Comment 7

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

Thanks.  But, yes, something like gfx.io, and the crash in integrated or discrete could happen when a switch is made from one to the other.  Although, the start up one seems like there wouldn't have been a time to switch.  Either way, this is good info, thanks.

Comment 8

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

Perhaps it is GL context lost, for at least some of these (e.g., the one from comment 0 and the one from bug 1202750.)  I'm not sure we track lost GL contexts on OS X.

Comment 9

[Jeff Muizelaar [:jrmuizel]]

OS X doesn't really have concept of a lost context afaik. (It's possible it's been added in more recent OS releases)

Comment 10

[Ryan VanderMeulen [:RyanVM]]

Duping over to bug 1218070 as it appears to be the same signature with more activity.