On 05/29/2014 01:00 PM, Peter Zijlstra wrote: > On Thu, May 29, 2014 at 06:50:57PM +0200, Peter Zijlstra wrote: >> > On Thu, May 29, 2014 at 12:44:23PM -0400, Sasha Levin wrote: >>> > > On 05/29/2014 11:07 AM, Peter Zijlstra wrote: >>>> > > > On Thu, May 29, 2014 at 10:47:09AM -0400, Sasha Levin wrote: >>>>> > > >> It doesn't work out well because we later lock a mutex in >>>>> > > >> sync_child_event(). >>>>> > > >> >>>> > > > >>>> > > > Urgh, right you are. I'll go stare at it more. It shouldn't have >>>> > > > mattered, because the mutex we take just before should ensure >>>> > > > existence, >>>> > > > but.. you know.. :-) >>>> > > > >>> > > >>> > > So the only caller to sync_child_event() is that loop. According to >>> > > what you said >>> > > it should be safe to remove that mutex lock, but doing that triggers a >>> > > list >>> > > corruption: >>> > > >>> > > [ 1204.341887] WARNING: CPU: 20 PID: 12839 at lib/list_debug.c:62 >>> > > __list_del_entry+0xa1/0xe0() >>> > > [ 1204.347597] list_del corruption. next->prev should be >>> > > ffff8806ca68b108, but was ffff88051a67c398 >>> > > [...] >>> > > >>> > > I don't see how that would happen :/ >> > >> > No, what I said is that the mutex in perf_event_exit_task() should be >> > sufficient to guard the list iteration calling __perf_event_exit_task(). >> > >> > Ading the RCU was a bit of paranoia.. > Hmm, so can you try this.. > > While that mutex should guard the elements, it doesn't guard against the > use-after-free that's from list_for_each_entry_rcu(). > __perf_event_exit_task() can actually free the event. > > And because list addition/deletion is guarded by both ctx->mutex and > ctx->lock, holding ctx->mutex is sufficient for reading the list, so we > don't actually need the rcu list iteration.
Works for me, thanks! Thanks, Sasha -- 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/
