On Fri, May 15, 2020 at 01:17:32PM +0200, Peter Zijlstra wrote:
> On Fri, May 15, 2020 at 09:47:40AM +0100, Mel Gorman wrote:
>
> > However, the wakeups are so rapid that the wakeup
> > happens while the server is descheduling. That forces the waker to spin
> > on smp_cond_load_acquire for longer. In this case, it can be cheaper to
> > add the task to the rq->wake_list even if that potentially requires an IPI.
>
> Right, I think Rik ran into that as well at some point. He wanted to
> make ->on_cpu do a hand-off, but simply queueing the wakeup on the prev
> cpu (which is currently in the middle of schedule()) should be an easier
> proposition.
>
> Maybe something like this untested thing... could explode most mighty,
> didn't thing too hard.
>
> ---
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index fa6c19d38e82..c07b92a0ee5d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -2312,7 +2312,7 @@ static void wake_csd_func(void *info)
> sched_ttwu_pending();
> }
>
> -static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
> +static void __ttwu_queue_remote(struct task_struct *p, int cpu, int
> wake_flags)
> {
> struct rq *rq = cpu_rq(cpu);
>
> @@ -2354,6 +2354,17 @@ bool cpus_share_cache(int this_cpu, int that_cpu)
> {
> return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
> }
> +
> +static bool ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
> +{
> + if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(),
> cpu)) {
> + sched_clock_cpu(cpu); /* Sync clocks across CPUs */
> + __ttwu_queue_remote(p, cpu, wake_flags);
> + return true;
> + }
> +
> + return false;
> +}
> #endif /* CONFIG_SMP */
>
> static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
> @@ -2362,11 +2373,8 @@ static void ttwu_queue(struct task_struct *p, int cpu,
> int wake_flags)
> struct rq_flags rf;
>
> #if defined(CONFIG_SMP)
> - if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(),
> cpu)) {
> - sched_clock_cpu(cpu); /* Sync clocks across CPUs */
> - ttwu_queue_remote(p, cpu, wake_flags);
> + if (ttwu_queue_remote(p, cpu, wake_flags))
> return;
> - }
> #endif
>
> rq_lock(rq, &rf);
> @@ -2550,7 +2558,15 @@ try_to_wake_up(struct task_struct *p, unsigned int
> state, int wake_flags)
> if (p->on_rq && ttwu_remote(p, wake_flags))
> goto unlock;
>
> + if (p->in_iowait) {
> + delayacct_blkio_end(p);
> + atomic_dec(&task_rq(p)->nr_iowait);
> + }
> +
> #ifdef CONFIG_SMP
> + p->sched_contributes_to_load = !!task_contributes_to_load(p);
> + p->state = TASK_WAKING;
> +
> /*
> * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
> * possible to, falsely, observe p->on_cpu == 0.
> @@ -2581,15 +2597,10 @@ try_to_wake_up(struct task_struct *p, unsigned int
> state, int wake_flags)
> * This ensures that tasks getting woken will be fully ordered against
> * their previous state and preserve Program Order.
> */
> - smp_cond_load_acquire(&p->on_cpu, !VAL);
> -
> - p->sched_contributes_to_load = !!task_contributes_to_load(p);
> - p->state = TASK_WAKING;
> + if (READ_ONCE(p->on_cpu) && __ttwu_queue_remote(p, cpu, wake_flags))
> + goto unlock;
>
> - if (p->in_iowait) {
> - delayacct_blkio_end(p);
> - atomic_dec(&task_rq(p)->nr_iowait);
> - }
> + smp_cond_load_acquire(&p->on_cpu, !VAL);
>
> cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
> if (task_cpu(p) != cpu) {
I don't see a problem with moving the updating of p->state to the other
side of the barrier but I'm relying on the comment that the barrier is
only related to on_rq and on_cpu.
However, I'm less sure about what exactly you intended to do.
__ttwu_queue_remote is void so maybe you meant to use ttwu_queue_remote.
In that case, we potentially avoid spinning on on_rq for wakeups between
tasks that do not share CPU but it's not clear why it would be specific to
remote tasks. If you meant to call __ttwu_queue_remote unconditionally,
it's not clear why that's now safe when smp_cond_load_acquire()
cared about on_rq being 0 before queueing a task for wakup or directly
waking it up.
Also because __ttwu_queue_remote() now happens before select_task_rq(), is
there not a risk that in some cases we end up stacking tasks unnecessarily?
> @@ -2597,14 +2608,6 @@ try_to_wake_up(struct task_struct *p, unsigned int
> state, int wake_flags)
> psi_ttwu_dequeue(p);
> set_task_cpu(p, cpu);
> }
> -
> -#else /* CONFIG_SMP */
> -
> - if (p->in_iowait) {
> - delayacct_blkio_end(p);
> - atomic_dec(&task_rq(p)->nr_iowait);
> - }
> -
> #endif /* CONFIG_SMP */
>
> ttwu_queue(p, cpu, wake_flags);
--
Mel Gorman
SUSE Labs
