721321 - Minidump writer process ends with SIGILL after writing the minidump

Status: RESOLVED FIXED

(Toolkit :: Crash Reporting, defect)

Description

[Mike Hommey [:glandium]]

I ended up reliably triggering this when applying part 2 from bug 686805, but I suspect this may happen under some circumstances, which effectively makes us miss crash reports.

It appears this was fixed upstream: http://codereview.chromium.org/8165010

Comment 1

[Mike Hommey [:glandium]]

Attachment: Fix sys_clone() on ARM to work reliably in Thumb-2

Comment 2

[Mike Hommey [:glandium]]

In fact, it happens after the minidump is written, not before, but it'd still be better to do the right thing here.

Comment 3

[(not currently active) Ted Mielczarek]

I really don't feel confident reviewing this. Can you get someone who actually understands ARM assembly to review?

Comment 4

[(not currently active) Ted Mielczarek]

Also, upstream for this file is: http://code.google.com/p/linux-syscall-support/

Comment 5

[Jacob Bramley [:jbramley]]

Comment on attachment 591726 [details] [diff] [review]
Fix sys_clone() on ARM to work reliably in Thumb-2

Review of attachment 591726 [details] [diff] [review]:
-----------------------------------------------------------------

::: toolkit/crashreporter/google-breakpad/src/common/linux/linux_syscall_support.h
@@ +2196,5 @@
> +                              */
> +                           #ifdef __thumb2__
> +                             "ldr   r7,[sp]\n"
> +                             "blx   r7\n"
> +                           #else

Yep, this is fine.

I'd normally argue for using BLX in the ARM case too (except on ARMv4T, where it doesn't exist), but this doesn't exactly look like performance-critical code. I'd probably make it "__thumb2__ || !__armv4t__" (or whatever the macro is called), but it's up to you.

Comment 6

[Mike Hommey [:glandium]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/04a179077054

Comment 7

[(not currently active) Ted Mielczarek]

Can you upload a patch against the linux-syscall-support repo to codereview.chromium.org and ask for review from [e-mail] ?

Comment 8

[Mike Hommey [:glandium]]

(In reply to Ted Mielczarek [:ted, :luser] from comment #7)
> Can you upload a patch against the linux-syscall-support repo to
> codereview.chromium.org and ask for review from [e-mail] ?

The patch comes from there.

Comment 9

[Marco Bonardo [:mak]]

https://hg.mozilla.org/mozilla-central/rev/04a179077054