1529995 - Crash with MOZ_CRASH(All runnables destined for a worker thread must be cancelable!), in [@ mozilla::dom::WorkerPrivate::MaybeWrapAsWorkerRunnable ] with offscreenCanvas

Status: RESOLVED DUPLICATE of bug 1736177

(Core :: Graphics: CanvasWebGL, defect, P4)

Description

[Daniel Holbert [:dholbert]]

This bug is for crash report bp-bb291515-9497-416b-a28c-d22aa0190219.

Top 10 frames of crashing thread:

0 libxul.so libxul.so@0x2845535 
1 libxul.so libxul.so@0x2854e2a 
2 libxul.so libxul.so@0x8f4197 
3 libxul.so libxul.so@0x205b5eb 
4 libxul.so libxul.so@0x205b6a2 
5 libxul.so libxul.so@0x5890b9f 
6 libxul.so libxul.so@0x1d54bda 
7 libxul.so libxul.so@0x6badf4f 
8 libxul.so libxul.so@0x1fe03e7 
9 libxul.so libxul.so@0x1fe01ef 

Not sure why we're missing symbols; that's unfortunate. :( But the crash reason is:

MOZ_CRASH(All runnables destined for a worker thread must be cancelable!)

...which seems to be this line of code:
https://searchfox.org/mozilla-central/rev/dc0adc07db3df9431a0876156f50c65d580010cb/dom/workers/WorkerPrivate.cpp#1469

This crash report came from someone named "Wes" in IRC #developers today. (no longer on, unfortunately)

Comment 1

[Wesley W. Garland]

dholbert - I'm watching this bug now.

I'm investigating two crashers right now; I believe this one is the root cause I'm chasing (the other one is that tabs crash when you use the debugger to inspect a typed array; I discussed that on slack bhackett this morning).

I think this was triggered by offscreenCanvas(...).getContext('webgl2').fenceSync() from a WebWorker.

Comment 2

[Wesley W. Garland]

Here is a fairly minimal example which reproduces the bug. Note that you must have gfx.offscreenCanvas enabled in your prefs: http://people.kingsds.network/wesgarland/crashers/ian-chew-webgl2-rendering-context-fenceSync/

Comment 3

[Wesley W. Garland]

Attachment: Web content which reproduces bug

Untar the file into a directory and point your browser at index.html. You need gfx.offscreenCanvas enabled. When you click "run", your tab will crash.

I have reproduced on the latest release as well as 826b59e57fe4 pulled from m-c a couple of days ago.

Comment 4

[Daniel Holbert [:dholbert]]

That's great -- I was able to reproduce the crash immediately with the URL (and the pref flip) in comment 2.

Plus, my crash report has symbols, so might be easier to analyze: bp-9d0f6208-f4c0-40ee-8cf1-ceafd0190226

My crash report has signature [@ mozilla::dom::WorkerPrivate::MaybeWrapAsWorkerRunnable ]

Comment 5

[Daniel Holbert [:dholbert]]

jgilbert, maybe you know what's going on here? The JS in question seems to be doing webgl stuff.

Comment 6

[Kelsey Gilbert [:jgilbert]]

OffscreenCanvas is unsupported.

Comment 7

[Daniel Holbert [:dholbert]]

Wes, do you know if any of your folks has hit this with gfx.offscreenCanvas at its default (false) value?

(with another testcase that doesn't depend on the offscreenCanvas API, I mean)

Comment 8

[Wesley W. Garland]

dholbert - We did some digging this morning. It does not appear that it is possible to access this functionality without using offscreenCanvas, because the WebGL2RenderingContext object is not added to the top of the prototype chain (global object) in a worker until you |new OffscreenCanvas().getContext()|.

Comment 9

[Daniel Holbert [:dholbert]]

Cool, thanks for that info, and for bringing this to our attention in the first place!

I've chained this to bug 1390089 (the offscreenCanvas metabug).

Comment 10

[Andrew Osmond [:aosmond] (he/him)]

I went through the runnables as part of bug 1736177 and made them cancelable.