1291163 - Intermittent dom/canvas/test/webgl-mochitest/test_capture.html | application crashed [@ mozilla::layers::PersistentBufferProviderShared::BorrowDrawTarget(mozilla::gfx::IntRectTyped<mozilla::gfx::UnknownUnits> const &)]

Status: RESOLVED FIXED

(Core :: Graphics: CanvasWebGL, defect)

Description

[Treeherder Bug Filer]

Filed by: philringnalda [at] gmail.com

https://treeherder.mozilla.org/logviewer.html#?job_id=1272753&repo=autoland

http://archive.mozilla.org/pub/firefox/tinderbox-builds/autoland-win64-debug/1470095809/autoland_win8_64-debug_test-mochitest-gl-bm112-tests1-windows-build58.txt.gz

Comment 1

[Jerry Shih[:jerry] (UTC+8) (inactive)]

Nical,

Here is an assert at:
https://hg.mozilla.org/mozilla-central/annotate/ffac2798999c5b84f1b4605a1280994bb665a406/gfx/layers/PersistentBufferProvider.cpp#l196

Comment 2

[Nicolas Silva [:nical]]

orangefactor shows 3 instances of this intermittent failure, all of which happening just after a device reset.
This is when trying to get a texture for the canvas to draw into, and we already have buffered 4 textures which are marked as busy (the compositor hasn't released them). In theory we throttle the main thread so that this never happens, but I can imagine that the device reset puts us in buggy situation where we didn't unlock the textures on the compositor side.

Comment 3

[Nicolas Silva [:nical]]

Attachment: ReadUnlock if we are not going to call EndFrame

This should do. If after this the intermittent persists, we can turn the MOZ_ASSERT(false) into a non-fatal gfxCriticalNote, but I'd rather keep crashing debug builds when something like that happens if possible.

Comment 4

[Intermittent Failures Robot]

7 automation job failures were associated with this bug in the last 7 days.

Repository breakdown:
* mozilla-inbound: 3
* autoland: 2
* try: 1
* fx-team: 1

Platform breakdown:
* windows8-64: 5
* android-4-3-armv7-api15: 2

For more details, see:
https://brasstacks.mozilla.com/orangefactor/?display=Bug&bugid=1291163&startday=2016-08-01&endday=2016-08-07&tree=all

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/9beaa7589405
Make sure TextureHosts are read-unlock'ed if Compositor::EndFrame is not called. r=sotaro

Comment 6

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/9beaa7589405

Comment 7

[Ryan VanderMeulen [:RyanVM]]

Should we consider landing this on Aurora as well?

Comment 8

[Nicolas Silva [:nical]]

Comment on attachment 8777767 [details] [diff] [review]
ReadUnlock if we are not going to call EndFrame

Approval Request Comment
[Feature/regressing bug #]:
[User impact if declined]: Some crashes related to device resets.
[Describe test coverage new/current, TreeHerder]: None.
[Risks and why]: low risk, this patch is simple and assumes the worst cases so it tends to be more robust than before patch.
[String/UUID change made/needed]: None.

Comment 9

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

Comment on attachment 8777767 [details] [diff] [review]
ReadUnlock if we are not going to call EndFrame

Fixes an intermittent failure, Aurora50+

Comment 10

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/0ed2f1ac5617