Re: [ISSUE] The time cost of IOSQE_IO_LINK

[Oleg Nesterov <oleg@redhat.com>]

On 02/18, Oleg Nesterov wrote:
>
> otherwise I think this is correct, but how about the patch below?
> Then this code can be changed to use try_cmpxchg().

You have already sent the patch which adds the generic try_cmpxchg,
so the patch below can be trivially adapted.

But I'd prefer another change, I think both task_work_add() and
task_work_cancel() can use try_cmpxchg() too.

> 
> Oleg.
> 
> --- a/kernel/task_work.c
> +++ b/kernel/task_work.c
> @@ -97,17 +97,24 @@ void task_work_run(void)
>  		 * work->func() can do task_work_add(), do not set
>  		 * work_exited unless the list is empty.
>  		 */
> -		raw_spin_lock_irq(&task->pi_lock);
>  		do {
> +			head = NULL;
>  			work = READ_ONCE(task->task_works);
> -			head = !work && (task->flags & PF_EXITING) ?
> -				&work_exited : NULL;
> +			if (!work) {
> +				if (task->flags & PF_EXITING)
> +					head = &work_exited;
> +				else
> +					break;
> +			}
>  		} while (cmpxchg(&task->task_works, work, head) != work);
> -		raw_spin_unlock_irq(&task->pi_lock);
>  
>  		if (!work)
>  			break;
>  
> +		// Synchronize with task_work_cancel()
> +		raw_spin_lock_irq(&task->pi_lock);
> +		raw_spin_unlock_irq(&task->pi_lock);
> +
>  		do {
>  			next = work->next;
>  			work->func(work);