[dm-devel] [PATCH 3/6] sched, perf, kvm: Fix preemption condition

[Mathieu Desnoyers]

----- On Jun 2, 2021, at 9:12 AM, Peter Zijlstra peterz at infradead.org wrote:

> When ran from the sched-out path (preempt_notifier or perf_event),
> p->state is irrelevant to determine preemption. You can get preempted
> with !task_is_running() just fine.
> 
> The right indicator for preemption is if the task is still on the
> runqueue in the sched-out path.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz at infradead.org>
> ---
> kernel/events/core.c |    7 +++----
> virt/kvm/kvm_main.c  |    2 +-
> 2 files changed, 4 insertions(+), 5 deletions(-)
> 
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -8568,13 +8568,12 @@ static void perf_event_switch(struct tas
> 		},
> 	};
> 
> -	if (!sched_in && task->state == TASK_RUNNING)
> +	if (!sched_in && current->on_rq) {

This changes from checking task->state to current->on_rq, but this change
from "task" to "current" is not described in the commit message, which is odd.

Are we really sure that task == current here ?

Thanks,

Mathieu

> 		switch_event.event_id.header.misc |=
> 				PERF_RECORD_MISC_SWITCH_OUT_PREEMPT;
> +	}
> 
> -	perf_iterate_sb(perf_event_switch_output,
> -		       &switch_event,
> -		       NULL);
> +	perf_iterate_sb(perf_event_switch_output, &switch_event, NULL);
> }
> 
> /*
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -4869,7 +4869,7 @@ static void kvm_sched_out(struct preempt
> {
> 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
> 
> -	if (current->state == TASK_RUNNING) {
> +	if (current->on_rq) {
> 		WRITE_ONCE(vcpu->preempted, true);
> 		WRITE_ONCE(vcpu->ready, true);
>  	}

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com