On Fri, Mar 16, 2018 at 11:50:17AM +0100, Peter Zijlstra wrote: > On Fri, Mar 16, 2018 at 11:39:46AM +0100, Jiri Olsa wrote: > > On Fri, Mar 16, 2018 at 11:31:29AM +0100, Peter Zijlstra wrote:
> > > There is at least one more known issue with that patch, but neither Mark > > > nor me could reproduce so far, so we don't know if we're right about the > > > cause. > > > > is there more info about that issue? I could try it > > Find below, 0day report didn't go out to lkml. We think moving the > list_del_init() out from the !RB_NODE_EMPTY() test might fix, but since > we can't repro so far, its all guesses. I've managed to reproduce this using the 0day scripts. From the 0day logs, it looks like it's possible to hit it ~6% of the time. I added a WARN_ON(RB_NODE_EMPTY(...)), and I see that fire in the exit path: [ 76.287197] perf_remove_from_context+0x9a/0xc0 [ 76.287552] perf_event_release_kernel+0x214/0x3e0 [ 76.287928] ? _raw_spin_unlock+0x8/0x10 [ 76.288237] ? locks_remove_file+0x219/0x230 [ 76.288572] perf_release+0xe/0x20 [ 76.288842] __fput+0x1c9/0x340 [ 76.289089] ____fput+0x8/0x10 [ 76.289332] task_work_run+0x9a/0xd0 [ 76.289613] do_exit+0x6cc/0x1220 [ 76.289877] ? __might_sleep+0xcb/0x150 [ 76.290183] do_group_exit+0x96/0x110 [ 76.290473] get_signal+0x8c3/0xb60 [ 76.290750] ? __perf_event_task_sched_in+0x20d/0x250 [ 76.291143] do_signal+0x19/0x950 [ 76.291408] ? finish_task_switch+0x212/0x480 [ 76.291750] ? __switch_to+0x414/0x730 [ 76.292051] ? _raw_spin_unlock_irqrestore+0x45/0x60 [ 76.292439] ? trace_hardirqs_on+0x63/0x100 [ 76.292769] ? prepare_to_wait_event+0x23a/0x250 [ 76.293132] ? do_int80_syscall_32+0x271/0x290 [ 76.293478] exit_to_usermode_loop+0xb9/0x190 [ 76.293819] do_int80_syscall_32+0x271/0x290 [ 76.294161] entry_INT80_32+0x36/0x36 ... then we subsequently hit the initial splat, which looks promisingly like our initial theory. However, I don't currently understand how we can see a group leader with an empty RB node in this path. I can only see that being the case for siblings that we promoted to being leaders, and those should have an empty list at the point we promote them... Thanks, Mark.
