Re: Q: perf_install_in_context/perf_event_enable are racy?

[Oleg Nesterov <oleg@redhat.com>]

On 01/27, Peter Zijlstra wrote:
>
> +void task_function_trampoline(void *data)
> +{
> +	struct task_function_call *tfc = data;
> +	struct task_struct *p = tfc->p;
> +	struct rq *rq = this_rq();
> +
> +#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
> +	if (rq->in_ctxsw)
> +		return;
> +#endif
> +
> +	if (rq->curr != p)
> +		return;

Yes, I think this should solve the problem.

>  prepare_task_switch(struct rq *rq, struct task_struct *prev,
>  		    struct task_struct *next)
>  {
> +#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
> +	rq->in_ctxsw = 1;
> +#endif
> +	sched_info_switch(prev, next);
> +	perf_event_task_sched_out(prev, next);
>  	fire_sched_out_preempt_notifiers(prev, next);
>  	prepare_lock_switch(rq, next);
>  	prepare_arch_switch(next);
> +	trace_sched_switch(prev, next);
>  }

Yes, I was wondering why schedule() calls perf_event_task_sched_out().
This way the code looks more symmetrical/understandable.

>  /**
> @@ -2823,6 +2860,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
>  	perf_event_task_sched_in(current);
>  #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
>  	local_irq_enable();
> +	rq->in_ctxsw = 0;

If we think that context_switch finishes here, probably it would be
more clean to clear ->in_ctxsw before local_irq_enable().

>  #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
>  	finish_lock_switch(rq, prev);

But, otoh, maybe finish_lock_switch() can clear in_ctxsw, it already
checks __ARCH_WANT_INTERRUPTS_ON_CTXSW. Likewise, perhaps it can be
set in prepare_lock_switch() which enables irqs.

But this is cosmetic and up to you.

Oleg.