1212011 - Use AutoEnterOOMUnsafeRegion in a couple more places

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jon Coppeard (:jonco)]

Attachment: auto-enter-oom-unsafe-region-2

This converts the new CrashAtUnhandlableOOM()s.

Comment 1

[Terrence Cole [:terrence]]

Comment on attachment 8670398 [details] [diff] [review]
auto-enter-oom-unsafe-region-2

Review of attachment 8670398 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/gc/Zone.h
@@ +369,2 @@
>          if (!runtimeFromAnyThread()->gc.nursery.addedUniqueIdToCell(cell))
> +            oomUnsafe.crash("failed to allocate tracking data for a nursery uid");

Gah! Sorry! I think I even wrote this part of the patch after oomUnsafe.crash was a thing. I think I may have even used it elsewhere in the patch.

Comment 2

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f2eb600ca47a

Comment 3

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/f2eb600ca47a

Comment 4

[Robert Kaiser]

Just out of interest, what info does oomUnsafe.crash leave in annotations? Anything that makes it clear to us that it's an OOM?

Comment 5

[Jon Coppeard (:jonco)]

(In reply to Robert Kaiser (:[e-mail]) from comment #4)
> Just out of interest, what info does oomUnsafe.crash leave in annotations?
> Anything that makes it clear to us that it's an OOM?

Yes, it prepends the "[unhandlable oom]" tag to the message.

Comment 6

[Robert Kaiser]

(In reply to Jon Coppeard (:jonco) from comment #5)
> Yes, it prepends the "[unhandlable oom]" tag to the message.

Where can we see that message in the crash report? Do we get any info of the size of the allocation that is failing? We'll need to hook our OOM signature generation to detecting that and adopting signatures accordingly.

Comment 7

[Jon Coppeard (:jonco)]

(In reply to Robert Kaiser (:[e-mail]) from comment #6)
This bug doesn't change the previous behaviour at all, just the code path via which the crash is generated.  The purpose is so that we can avoid triggering this in our OOM tests, since it is unhandlable.

> Do we get any info of the size of the allocation that is failing?

No there is currently no indication of this.

This is generally used for small allocations that are not expected to fail however.