Re: perf/tracepoint: another fuzzer generated lockup

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2013/11/15 0:33), Peter Zijlstra wrote:
> On Thu, Nov 14, 2013 at 04:23:04PM +0100, Peter Zijlstra wrote:
>>  /*
>> + * We must dis-allow sampling irq_work_exit() because perf event sampling
>> + * itself can cause irq_work, which would lead to an infinite loop;
>> + *
>> + *  1) irq_work_exit happens
>> + *  2) generates perf sample
>> + *  3) generates irq_work
>> + *  4) goto 1
>> + */
>> +TRACE_EVENT_PERF_PERM(irq_work_exit, is_sampling_event(p_event) ? -EPERM : 0);
> 
> And the only reason this doesn't feed fwd itself into oblivion for
> irq_work_enter() is because the irq_work_list must not be empty when the
> interrupt is raised, and queueing further work does not re-raise the
> IPI.
> 
> 
> Also, we should probably do something 'smart' for kprobes, as all of
> irq_work.c and plenty of perf itself is not __kprobe marked so you're
> all free to insert kprobes in the middle of perf and then attach perf to
> such a thing.

Kprobes itself can detect nested call by using per-cpu current-running
kprobe pointer. And if it is nested, it just skips calling handlers.
Anyway, I don't recommend to probe inside the handlers, but yes,
you can trace perf-handler by ftrace B). I actually traced a kprobe-bug
by kprobe-tracer last night, that was amazing :)

Thank you,

-- 
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com