Re: IORING_OP_POLL_ADD slower than linux-aio IOCB_CMD_POLL

[Pavel Begunkov <asml.silence@gmail.com>]

On 6/15/22 12:07, Avi Kivity wrote:
> Sorry, hit send too quiclky.
>>>> If this is correct (it's not unreasonable, but should be documented), then there should also be a simple way to force a kernel entry. But how to do this using liburing? IIUC if I the following apply:
>>>>
>>>>
>>>>   1. I have no pending sqes
>>>>
>>>>   2. There are pending completions
>>>>
>>>>   3. There is a completed event for which a completion has not been appended to the completion queue ring
>>>>
>>>>
>>>> Then io_uring_wait_cqe() will elide io_uring_enter() and the completed-but-not-reported event will be delayed.
>>>
>>> One way is to process all CQEs and then it'll try to enter the
>>> kernel and do the job.
> 
> 
> But I don't want it to wait. I want it to generate pending completions, and return immediately even if no completions were generated. I have some background computations I'm happy to perform if no events are pending, but I would like those events to be generated promptly.

Ok, then you may want IORING_SETUP_TASKRUN_FLAG described below.
btw IORING_SETUP_TASKRUN_FLAG has some additional overhead.


>>> Another way is to also set IORING_SETUP_TASKRUN_FLAG, then when
>>> there is work that requires to enter the kernel io_uring will
>>> set IORING_SQ_TASKRUN in sq_flags.
>>> Actually, I'm not mistaken io_uring has some automagic handling
>>> of it internally
>>>
>>> https://github.com/axboe/liburing/blob/master/src/queue.c#L36
>>>
> 
> Is there documentation about this flag?

Unfortunately, I don't see any.
Here is a link to the kernel commit if it helps:

https://git.kernel.dk/cgit/linux-block/commit/?h=ef060ea9e4fd3b763e7060a3af0a258d2d5d7c0d

I think it's more interesting what support liburing has,
but would need to look up in the code. Maybe Jens
remembers and can tell.

-- 
Pavel Begunkov