852251 - Multiple crashes seen in "layers" while on call and attempting FTP over DUN

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Greg Grisco]

Attachment: decoded minidump of crash

Breaking up bug 848003 into individual reports


1. Make MO call using QXDM
2. Make Dialup Networking (DUN) call and start FTP download test case

After the weekend run, device collected many minidumps.  Minidump attached.

Comment 1

[Greg Grisco]

Attachment: EXTRA file attachment

Comment 2

[Michael Vines [:m1] [:evilmachines]]

(tef+ due to regression in stability)

Comment 3

[Jason Duell]

What's the "FTP download case"?  Is that downloading a file in the browser using FTP, or an system-level app update?   I'm afraid I also don't know what a "MO call using QXDM" is.  (I'm just trying to get a sense for 1) how common the use case is here and 2) how the parts interact and 3) a clear set of STR so we can debug).

From the stack trace, I'm not clear on how the FTP download could be triggering this.  The crash stack is all Layers code (crashing at PCompositorChild::SendPLayersConstructor).  And on the child, FTP only operates on the main thread, and none of the FTP code is on the main thread stack in this crash.  But of course earlier FTP events on the main thread might be changing state somehow to cause this.  But I suspect the brokenness is all up above the FTP code itself.

Assigning to Jeff just because he seems to be the (latest) person to take the other Layers IPDL bug (bug 851664).

Comment 4

[Jeff Muizelaar [:jrmuizel]]

I'm not sure how actionable this is without being able to reproduce this locally.

Comment 5

[Michael Vines [:m1] [:evilmachines]]

Please try to look at the backtrace and attempt to determine how this could happen.  Reproducing these types of issues locally can be difficult and an time-consuming due to test setup.   The "FTP download case" stuff is the name of the stability test that cause this crash to occur /somehow/, but obviously the crash has nothing to do with FTP.  We just got (un)lucky and hit this layers bug as a part of that test.

Comment 6

[Greg Grisco]

Clearing my ni since I can't provide more clear STR.  This has been reproduced in last two AUs that we ran stabiilty tests on.

Comment 7

[Andrew Overholt [:overholt]]

Anthony or Nick, can one of you take a look?

Comment 8

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

It looks like CompositorChild::Get() may be returning a null. That would happen if it is called before CompositorChild::Create() or after CompositorChild::Destroy(). There is no null check in TabChild::InitRenderingState(). Something is happening out of order but without a way to reproduce the problem all I can do is put in a null check in InitRederingState().

Comment 9

[Nick Cameron [:nrc]]

I got nothing, sorry

Comment 10

[Justin Lebar (not reading bugmail)]

Anthony asked me to see if I've got something here, but I need to run now.

Comment 11

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Attachment: Check for null on CompositorChild::Get()

Comment 12

[Justin Lebar (not reading bugmail)]

I think this patch is probably right, based on the comment in TabChild::RecvShow.

Anyway it seems difficult to imagine how this patch would make things worse than they already are.

Comment 13

[Justin Lebar (not reading bugmail)]

Comment on attachment 729829 [details] [diff] [review]
Check for null on CompositorChild::Get()

r=me

Comment 14

[Joe Cheng [:jcheng] (please needinfo)]

is this ready to land? thanks

Comment 15

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Public holidays have delayed the landing.

https://tbpl.mozilla.org/?tree=Try&rev=d3a5fe89fb9b

Comment 16

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/9ecc1799f6f4

Comment 17

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/9ecc1799f6f4

Comment 18

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-b2g18/rev/1e5002f7a205
https://hg.mozilla.org/releases/mozilla-b2g18_v1_0_1/rev/f35275c90b13