Bug 1896513 Comment 36 Edit History

Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.

Original comment by
on 
Using mozregression with `--find-fix` narrowed me to [this range](https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=9142f728c689228ac4b5a2f55b2d50a3c5440d1b&tochange=0e8cd4eeebfd40e0572904479820de539ac43065). I couldn't get a tighter range because builds kept crashing on launch. But that contains bug 1894929!

And if I build and run current mozilla central I cannot reproduce, but if I revert bug 1894929 then I can again. So that's definitely the culprit.

This makes sense - that bug is an issue with our recycling of SurfaceTextures for webgl/canvas. (PDF.js uses canvas). It could lead to us attempting to render in to a SurfaceTexture in the canvas thread before it has been released by the renderer thread. On this GPU it appears that causes a deadlock. Unfortunately a debugger did not give me any useful callstacks to identify where exactly the hang occurs, but I verified this is in fact what is happening with some printfs.

So this will be fixed in the next nightly, and I'll make sure that the fix gets uplifted to beta
Revision 1 by
on 
Using mozregression with `--find-fix` narrowed me to [this range](https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=9142f728c689228ac4b5a2f55b2d50a3c5440d1b&tochange=0e8cd4eeebfd40e0572904479820de539ac43065). I couldn't get a tighter range because builds kept crashing on launch. But that contains bug 1894929!

And if I build and run current mozilla central I cannot reproduce, but if I revert bug 1894929 then I can again. So that's definitely the culprit.

This makes sense - that bug is an issue with our recycling of SurfaceTextures for webgl/canvas. (PDF.js uses canvas). It could lead to us attempting to render in to a SurfaceTexture in the canvas thread before it has been released by the renderer thread. On this GPU it appears that causes a deadlock. Or perhaps on all phones, but it’s more likely to reproduce on some due to timings. Unfortunately a debugger did not give me any useful callstacks to identify where exactly the hang occurs, but I verified this is in fact what is happening with some printfs.

So this will be fixed in the next nightly, and I'll make sure that the fix gets uplifted to beta

Back to Bug 1896513 Comment 36