1685046 - macOS Crash in [@ mozilla::wr::RenderCompositorNativeOGL::Bind]

Status: RESOLVED WORKSFORME

(Core :: Graphics: WebRender, defect)

Description

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Crash report: https://crash-stats.mozilla.org/report/index/8de1b37d-8823-4dcf-bbda-f5cef0210105

MOZ_CRASH Reason: MOZ_RELEASE_ASSERT(fbo)

Top 10 frames of crashing thread:

0 XUL mozilla::wr::RenderCompositorNativeOGL::Bind gfx/webrender_bindings/RenderCompositorNative.cpp:559
1 XUL <webrender_bindings::bindings::WrCompositor as webrender::composite::Compositor>::bind gfx/webrender_bindings/src/bindings.rs:1305
2 XUL webrender::renderer::Renderer::draw_frame gfx/wr/webrender/src/renderer.rs:6026
3 XUL webrender::renderer::Renderer::render_impl gfx/wr/webrender/src/renderer.rs:3488
4 XUL webrender::renderer::Renderer::render gfx/wr/webrender/src/renderer.rs:3244
5 XUL wr_renderer_render gfx/webrender_bindings/src/bindings.rs:614
6 XUL mozilla::wr::RenderThread::UpdateAndRender gfx/webrender_bindings/RenderThread.cpp:493
7 XUL mozilla::wr::RenderThread::HandleFrameOneDoc gfx/webrender_bindings/RenderThread.cpp:327
8 XUL mozilla::detail::RunnableMethodImpl<mozilla::wr::RenderThread*, void  xpcom/threads/nsThreadUtils.h:1201
9 XUL MessageLoop::DoWork ipc/chromium/src/base/message_loop.cc:548

Comment 1

[Brad Werth [:bradwerth]]

Failure is at https://searchfox.org/mozilla-central/rev/519f913527b0d9d5097d290d5731cff6b2991fe0/gfx/webrender_bindings/RenderCompositorNative.cpp#530. If we're going to treat the return value of NextSurfaceAsFramebuffer as a release assert, we should hoist that assert into the implementation of the function itself, so we can understand the failure better. I'll prepare a patch that does that, and we can keep this bug open.

Comment 2

[Brad Werth [:bradwerth]]

Attachment: Bug 1685046: Hoist release asserts in RenderCompositorNativeOGL::Bind to crash earlier.

This turns a browser crash into rendering blank spots. The sources of failure
have multiple callers, so hoisting release asserts to those functions could
expand the crashes to more cases.

Comment 3

[Brad Werth [:bradwerth]]

(In reply to Brad Werth [:bradwerth] from comment #2)

This turns a browser crash into rendering blank spots. The sources of failure
have multiple callers, so hoisting release asserts to those functions could
expand the crashes to more cases.

I couldn't find a way to hoist the assert earlier without expanding the cases where the crash might occur, so I instead opted to make it not crash. It would be very helpful to have Steps to Reproduce so we can check the debug logs that indicate why this is happening (empty texture size? ObtainSurfaceFromPool returning null?)

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b9665e89a3b4
Hoist release asserts in RenderCompositorNativeOGL::Bind to crash earlier. r=jrmuizel

Comment 5

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/b9665e89a3b4

Comment 6

[Daniel Bodea [:danibodea]]

Are there some steps to reproduce in order to reproduce, check tracking flags and verify the possible fix?

Comment 7

[Brad Werth [:bradwerth]]

(In reply to Bodea Daniel [:danibodea] from comment #6)

Are there some steps to reproduce in order to reproduce, check tracking flags and verify the possible fix?

I don't have steps to reproduce. The patch that's been landed is largely an effort to give us a better signature to help create a testcase.

Comment 8

[Brad Werth [:bradwerth]]

We've changed the signature enough (for example, in Bug 1707832, also patched) that this bug doesn't need to remain open.