Re: [RFC] Deadlock via recursive wakeup via RCU with threadirqs

From: Peter Zijlstra
Date: Mon Jul 01 2019 - 05:42:53 EST

On Fri, Jun 28, 2019 at 10:20:56AM -0700, Paul E. McKenney wrote:
> On Fri, Jun 28, 2019 at 06:04:08PM +0200, Peter Zijlstra wrote:
> > On Fri, Jun 28, 2019 at 08:54:04AM -0700, Paul E. McKenney wrote:
> > > Thank you! Plus it looks like scheduler_ipi() takes an early exit if
> > > ->wake_list is empty, regardless of TIF_NEED_RESCHED, right?
> >
> > Yes, TIF_NEED_RESCHED is checked in the interrupt return path.
> OK, got it. So the following sequence would be a valid way to get the
> scheduler's attention on the current CPU shortly after interrupts
> are re-enabled, even if the current CPU is already holding some
> rq or pi locks, correct?
> set_tsk_need_resched(current);
> set_preempt_need_resched();
> smp_send_reschedule(smp_processor_id());

I'm not sure if smp_send_reschedule() can be used as self-IPI, some
hardware doesn't particularly like that IIRC. That is, hardware might
only have interfaces to IPI _other_ CPUs, but not self.

The normal scheduler code takes care to not call smp_send_reschedule()
to self.