Re: [nohz] 2a16fc93d2c: kernel lockup on idle injection

[Peter Zijlstra]

On Tue, Dec 16, 2014 at 10:21:27PM +0100, Thomas Gleixner wrote:
> DEFINE_PER_CPU(nohz_full_must_tick, unsigned long);
> 
> enum {
>      NOHZ_SCHED_NEEDS_TICK,
>      NOHZ_POSIXTIMER_NEEDS_TICK,
>      NOHZ_PERF_NEEEDS_TICK,
> };
> 
> /* rq->lock is held for evaluating rq->nr_running */
> static void sched_ttwu_nohz(struct rq *rq)
> {
> 	if (nohz_full_disabled())
> 		return;
> 
> 	if (rq->nr_running != 2)
> 		return;
> 	set_bit(NOHZ_SCHED_NEEDS_TICK, this_cpu_ptr(nohz_full_must_tick);
> }
> 
> /* rq->lock is held for evaluating rq->nr_running */
> static void sched_ttwu_remote_nohz(struct rq *rq)
> {
> 	if (nohz_full_disabled())
> 		return;
> 
> 	if (rq->nr_running != 2)
> 	   	return;
> 	/*
> 	 * Force smp_send_reschedule(). irq_exit() on the
> 	 * remote cpu will handle the rest.
> 	 */

smp_send_reschedule() is magic and does not guarantee irq_{enter,exit}()
being called, although we could audit and fix that.

> 	set_bit(NOHZ_SCHED_NEEDS_TICK, per_cpu_ptr(nohz_full_must_tick, rq->cpu);
> 	rq->force_send_resched = true;
> }

I'd make that force_send_resched the return value or so..

> /* rq->lock is held for evaluating rq->nr_running */
> static void sched_out_nohz(struct rq *rq)
> {
> 	if (nohz_full_disabled())
> 		return;
> 
> 	if (rq->nr_running >= 2)
> 	   	return;
> 	clear_bit(NOHZ_SCHED_NEEDS_TICK, this_cpu_ptr(nohz_full_must_tick);
> }
> 
> /* preeemption is disabled */
> static void sched_in_nohz(struct task_struct *task)
> {
> 	if (nohz_full_disabled())
> 		return;
> 
> 	if (!task_uses_posix_timers(task))
> 		clear_bit(NOHZ_POSIXTIMER_NEEDS_TICK,
> 			  this_cpu_ptr(nohz_full_must_tick));
> 	else
> 		set_bit(NOHZ_POSIXTIMER_NEEDS_TICK,
> 			this_cpu_ptr(nohz_full_must_tick));
> 

/me hands you a few spare {} :-)

Arguably test state before doing a possibly pointless update?

> 	local_irq_disable();
> 	tick_full_nohz_update_state();
> 	local_irq_enable();
> }

But yes, that should work just fine..
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/