759653 - BasicLayers.cpp ABORTs in DEBUG mode on B2G

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Dave Hylands [:dhylands]]

All DEBUG builds after the following commit from our mozilla-central git repository:

7fe94a9 Sat May 26 12:38:17 2012 +0800 Kan-Ru Chen Bug 743182 - Use gfxPlatform::OptimalFormatForContent everywhere. r=joedrew

cause the following ABORT to occur in B2G:

###!!! ABORT: Bad! Did we create a buffer twice without painting?: '!mIsNewBuffer', file /home/work/B2G/gecko/gfx/layers/basic/BasicLayers.cpp, line 2531

Comment 1

[Dave Hylands [:dhylands]]

Attachment: gdb backtrace from the ABORT

Comment 2

[Dave Hylands [:dhylands]]

And oh yeah - the ABORT is only seen when OMTC is enabled.

Comment 3

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

This abort is a very serious logic error: it means our double-buffering code became discombobulated and probably would have not have synchronized front->back buffers correctly.  If that happens we'll get horrendous rendering artifacts.

Very interested to know how we're getting into this state.

Comment 4

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

It would be instructive to try re-applying the patch from bug 743182, but with the gfx/layers hunks removed.

Comment 5

[Dave Hylands [:dhylands]]

I also confirmed that a non-debug build of the same commit fixes bug 744238 (so no artifacts seen when sliding up the lock screen when OMTC is enabled).

Comment 6

[Dave Hylands [:dhylands]]

Removing gfx/layers/basic/BasicLayers.cpp and gfx/layers/ipc/ShadowLayers.cpp from the patchset, and adding back in the gfxASurface::FormatFromContent function cause the ABORT to not occur (and the lock screen artifact still shows)

Comment 7

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

That is very very odd.

Comment 8

[Dave Hylands [:dhylands]]

In the working code case, the image format that ShadowLayerForwarder::AllocBuffer creates is ImageFormatRGB16_565.

In the case that aborts, the image format that ShadowLayerForwarder::AllocBuffer creates is ImageFormatARGB32.

And if I go back to the version of the code that doesn't ABORT and modify OptimalFormatFor to return ImageFormatARGB32 then I also get the ABORT.

Comment 9

[Dave Hylands [:dhylands]]

And all of this testing was being done on a Samsung Galaxy S II running the ICS variant of gonk.

Comment 10

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

dhylands, do you have a patch for the ImageFormat problem?

Comment 11

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Actually, I think this bug only affects sgs2, so it's not a top priority.  But if you already have a patch we should land it.

Comment 12

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Attachment: Use RGB24 for offscreen surfaces on android-esque platforms

Comment 13

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a9479d148ce5

Comment 14

[Graeme McCutcheon [:graememcc]]

https://hg.mozilla.org/mozilla-central/rev/a9479d148ce5