On Thu, May 21, 2015 at 11:49 PM, Ingo Molnar <[email protected]> wrote: > > > * Stephane Eranian <[email protected]> wrote: > > > On Thu, May 21, 2015 at 7:03 AM, Peter Zijlstra <[email protected]> > > wrote: > > > On Thu, 2015-05-21 at 06:36 -0700, Stephane Eranian wrote: > > >> On Thu, May 21, 2015 at 6:29 AM, Peter Zijlstra <[email protected]> > > >> wrote: > > >> > On Thu, 2015-05-21 at 06:27 -0700, Stephane Eranian wrote: > > >> >> Or are you talking about a preemption while executing > > >> >> x86_schedule_events()? > > >> > > > >> > That. > > >> > > > >> > And we can of course cure that by an earlier patch I send; but I find > > >> > it > > >> > a much simpler rule to just never allow modifying global state for > > >> > validation. > > >> > > >> I can see validation being preempted, but not the context switch code > > >> path. > > >> Is that what you are talking about? > > >> > > >> You are saying validate_group() is in the middle of x86_schedule_events() > > >> using fake_cpuc, when it gets preempted. The context switch code when it > > >> loads > > >> the new thread's PMU state calls x86_schedule_events() which modifies the > > >> cpuc->event_list[]->hwc. But this is cpuc vs. fake_cpuc again. So yes, > > >> the calls > > >> nest but they do not touch the same state. > > > > > > They both touch event->hw->constraint. > > > > > >> And when you eventually come back > > >> to validate_group() you are back to using the fake_cpuc. So I am still > > >> not clear > > >> on how the corruption can happen. > > > > > > validate_group() > > > x86_schedule_events() > > > event->hw.constraint = c; # store > > > > > > <context switch> > > > perf_task_event_sched_in() > > > ... > > > x86_schedule_events(); > > > event->hw.constraint = c2; # store > > > > > > ... > > > > > > put_event_constraints(event); # assume failure to schedule > > > intel_put_event_constraints() > > > event->hw.constraint = NULL; > > > > > > <context switch end> > > > > > > c = event->hw.constraint; # read -> NULL > > > > > > if (!test_bit(hwc->idx, c->idxmsk)) # <- *BOOM* NULL deref > > > > > > > > > This in particular is possible when the event in question is a cpu-wide > > > event and group-leader, where the validate_group() tries to add an event > > > to the group. > > > > Ok, I think I get it now. It is not related to fake_cpuc vs. cpuc, > > it is related to the fact that the constraint is cached in the event > > struct itself and that one is shared between validate_group() and > > x86_schedule_events() because cpu_hw_event->event_list[] is an array > > of pointers to events and not an array of events. > > Btw., comments and the code structure should be greatly enhanced to > make all that very clear and hard to mess up. > Peter and I will clean this up.
> > A month ago perf became fuzzing-proof, and now that's down the drain > again... > It will be fixed this week. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
