Re: perf: fuzzer triggered trouble on AMD, maybe ibs related

[Stephane Eranian]

Hi Vince,


Sorry for late reply. I have reactivated my AMD Buldozer machine and
updated it to latest kernel.
I will run the fuzzer to see if I can reproduce the problem.

On Sat, Oct 24, 2015 at 12:12 AM, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Thu, Oct 22, 2015 at 12:46:53PM -0400, Vince Weaver wrote:
> > Hello
> >
> > I've been busy but finally had a chance to run perf_fuzzer on current git.
> > I am running on an AMD A10 system (my traditional Haswell system is
> > otherwise occupied).
> >
> > I got the following WARNING which was followed by an NMI storm which
> > eventually managed to confuse ext4 enough that my / partition was
> > remounted read-only? Very alarming.
> >
> > This is in static void perf_ibs_start(struct perf_event *event, int flags)
> >
> >       if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
> >                 return;
>
> Hurm, its been a long while since I looked at the AMD IBS stuff. But
> looking at it I would not be surprised if there's weird interaction
> between the HES flags and IBS_flags.
>
> But I'll have to look more careful..
>
> > [  468.763231] Dazed and confused, but trying to continue
> > [  468.794184] Uhhuh. NMI received for unknown reason 2d on CPU 2.
> > [  468.794184] Do you have a strange power saving mode enabled?
> > [  468.794184] Dazed and confused, but trying to continue
> > [  473.190535] sd 0:0:0:0: [sda] tag#2 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_TIMEOUT
> > [  473.199631] sd 0:0:0:0: [sda] tag#2 CDB: Write(10) 2a 00 39 93 49 d0 00 00 18 00
>
> This might just be the NMI storm starving a device interrupt.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/