1306426 - On content process crash, restart unaccelerated if we crashed in the driver

Status: NEW

(Core :: Graphics, enhancement, P3)

Description

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

If we're running accelerated graphics in the content process and hit (a single, or a sequence of) crashes, which point to the driver, disable acceleration, either just for the session, or until the driver/Firefox update.

Comment 1

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

We may just need to report the top of the stack, or unwind and report the whole thing - I'm not sure.

There is a dependency on "send the top of the stack to the parent process on crash" - Ted, would you know if there is already a bug for something like that, and if not, what the best location for that bug would be?  I know if bug 1225851, and understand there is a bug for doing some kind of unwinding on the client, but I can't quite find it.

Comment 2

[(not currently active) Ted Mielczarek]

bug 1280469 covers client-side stack unwinding. I don't know of anything like what you're describing, though. The work in bug 1280469 will be able to read minidumps on the client, but it doesn't support content processes currently.

Comment 3

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

Thanks Ted!

Gabriel, is bug 1293656 something that would help in this case, or perhaps exactly what we need?

Comment 4

[Gabriele Svelto [:gsvelto]]

Yes! Once I'll have implemented client-stack unwinding of content process crashes we should be able to easily identify which module caused the crash (and possibly even get the crash signature). With that information I suppose you should be able to trace it to GPU driver in case the crash happened there.