Re: [ISSUE] The time cost of IOSQE_IO_LINK

[Jens Axboe <axboe@kernel.dk>]

On 2/14/20 8:47 AM, Jens Axboe wrote:
>> I suspect you meant to put that in finish_task_switch() which is the
>> tail end of every schedule(), schedule_tail() is the tail end of
>> clone().
>>
>> Or maybe you meant to put it in (and rename) sched_update_worker() which
>> is after every schedule() but in a preemptible context -- much saner
>> since you don't want to go add an unbounded amount of work in a
>> non-preemptible context.
>>
>> At which point you already have your callback: io_wq_worker_running(),
>> or is this for any random task?
> 
> Let me try and clarify - this isn't for the worker tasks, this is for
> any task that is using io_uring. In fact, it's particularly not for the
> worker threads, just the task itself.
> 
> I basically want the handler to be called when:
> 
> 1) The task is scheduled in. The poll will complete and stuff some items
>    on that task list, and I want to task to process them as it wakes up.
> 
> 2) The task is going to sleep, don't want to leave entries around while
>    the task is sleeping.
> 
> 3) I need it to be called from "normal" context, with ints enabled,
>    preempt enabled, etc.
> 
> sched_update_worker() (with a rename) looks ideal for #1, and the
> context is sane for me. Just need a good spot to put the hook call for
> schedule out. I think this:
> 
> 	if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) {
> 		preempt_disable();
> 		if (tsk->flags & PF_WQ_WORKER)
> 			wq_worker_sleeping(tsk);
> 		else
> 			io_wq_worker_sleeping(tsk);
> 		preempt_enable_no_resched();
> 	}
> 
> just needs to go into another helper, and then I can call it there
> outside of the preempt.
> 
> I'm sure there are daemons lurking here, but I'll test and see how it
> goes...

Here's a stab at cleaning it up:

https://git.kernel.dk/cgit/linux-block/log/?h=io_uring-task-poll

top two patches. First one simply cleans up the sched_update_worker(),
so we now have sched_in_update() and sched_out_update(). No changes in
this patch, just moves the worker sched-out handling into a helper.

2nd patch then utilizes this to flush the per-task requests that may
have been queued up.

-- 
Jens Axboe