Re: perf: fuzzer KASAN perf_callchain_store on amd

[Vince Weaver]

On Wed, 16 Nov 2016, Dmitry Vyukov wrote:

> On Wed, Nov 16, 2016 at 5:33 PM, Vince Weaver <vincent.weaver@xxxxxxxxx> wrote:
> >
> > [118725.787218]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
> > [118725.793967]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
> > [118725.800690]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
> > [118725.807393]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
> > ...
> 
> 
> This is heap OOB rather than stack OOB.
> Is there an allocation stack/object size/shadow in the report? It
> would greatly help debugging.

No, the report just trailed off and then launched into the next KASAN 
report without anything in between (see below).

I am reporting what was in dmesg, I lost the serial console output.  
I'm currently re-running on the same machine, hopefully I can reproduce it.

[118726.110077]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
[118726.115849]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
[118726.121615]  [<ffffffff811d1512>] ? save_stack+0x33/0xa6
[118726.127385]  [<ffffffff811d

[118726.139200] BAD LUCK: lost 11949 message(s) from NMI context!
[118726.139201] 
==================================================================
[118726.153327] BUG: KASAN: slab-out-of-bounds in perf_callchain_store+0x69/0x84 at addr ffff8801d4fbe800