1087270 - Not all D3D11 result codes are checked.

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Nicolas Silva [:nical]]

Also, it's not clear what we are supposed to do when a command fails. Some cases should be recoverable like device resets (although we don't want to get into a loop of loosing and recreating devices so we should detect). In some cases we should crash debug builds, or maybe try to fallback to another backend (we don't have the infrastructure to switch backends at runtime, though).
We need a way to separate failures causing occasional glitches from the ones making the browser unusable.

Comment 1

[Nicolas Silva [:nical]]

Attachment: Check more d3d11 calls and add a hook to react in case of failure.

By default a failing operation will crash debug builds. We can use this as a base to add more checks and logging (for instance we could crash release builds after N failures with N a pref or just a hard coded threshold).

Comment 2

[Milan Sreckovic [:milan] (needinfo for best results)]

Comment on attachment 8509552 [details] [diff] [review]
Check more d3d11 calls and add a hook to react in case of failure.

Review of attachment 8509552 [details] [diff] [review]:
-----------------------------------------------------------------

You will want to use gfxCriticalError() and gfxWarning() to message what's going on, even before doing a MOZ_CRASH.  gfxCriticalError() will show up in the crash reports as well.  See also bug 1074952.

Comment 3

[Nicolas Silva [:nical]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/8a39622aa24a

Comment 4

[Nicolas Silva [:nical]]

(In reply to Milan Sreckovic [:milan] from comment #2)
> You will want to use gfxCriticalError() and gfxWarning() to message what's
> going on, even before doing a MOZ_CRASH.  gfxCriticalError() will show up in
> the crash reports as well.  See also bug 1074952.

This would be awesome, I'll happily write the patch as soon as we can use gfxCriticalError off the main thread (or is it already the case?)

Comment 5

[Carsten Book [:Tomcat]]

Hi Nical, sorry had to back this out for bustage on windows like https://treeherder.mozilla.org/ui/logviewer.html#?job_id=3240120&repo=mozilla-inbound

Comment 6

[Nicolas Silva [:nical]]

Ah, -WError did catch an actual bug this time, my bad.

Comment 7

[Nicolas Silva [:nical]]

green try push: https://tbpl.mozilla.org/?tree=Try&rev=49b03746dc55 (with the fix)

Comment 8

[Nicolas Silva [:nical]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/934e7c6d8811

Comment 9

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/934e7c6d8811