Re: [PATCH v2] io_uring: reduce latency by reissueing the operation

[Pavel Begunkov]

On 6/20/21 8:05 PM, Olivier Langlois wrote:
> It is quite frequent that when an operation fails and returns EAGAIN,
> the data becomes available between that failure and the call to
> vfs_poll() done by io_arm_poll_handler().
> 
> Detecting the situation and reissuing the operation is much faster
> than going ahead and push the operation to the io-wq.
> 
> Signed-off-by: Olivier Langlois <olivier@xxxxxxxxxxxxxx>
> ---
>  fs/io_uring.c | 26 +++++++++++++++++---------
>  1 file changed, 17 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index fa8794c61af7..6e037304429a 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -5143,7 +5143,10 @@ static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
>  	return mask;
>  }
>  
> -static bool io_arm_poll_handler(struct io_kiocb *req)
> +#define IO_ARM_POLL_OK    0
> +#define IO_ARM_POLL_ERR   1
> +#define IO_ARM_POLL_READY 2

Please add a new line here. Can even be moved somewhere
to the top, but it's a matter of taste.

Also, how about to rename it to apoll? io_uring internal
rw/send/recv polling is often abbreviated as such around
io_uring.c
IO_APOLL_OK and so on.

> +static int io_arm_poll_handler(struct io_kiocb *req)
>  {
>  	const struct io_op_def *def = &io_op_defs[req->opcode];
>  	struct io_ring_ctx *ctx = req->ctx;
> @@ -5153,22 +5156,22 @@ static bool io_arm_poll_handler(struct io_kiocb *req)
>  	int rw;
>  
>  	if (!req->file || !file_can_poll(req->file))
> -		return false;
> +		return IO_ARM_POLL_ERR;

It's not really an error. Maybe IO_APOLL_ABORTED or so?

>  	if (req->flags & REQ_F_POLLED)
> -		return false;
> +		return IO_ARM_POLL_ERR;
>  	if (def->pollin)
>  		rw = READ;
>  	else if (def->pollout)
>  		rw = WRITE;
>  	else
> -		return false;
> +		return IO_ARM_POLL_ERR;
>  	/* if we can't nonblock try, then no point in arming a poll handler */
>  	if (!io_file_supports_async(req, rw))
> -		return false;
> +		return IO_ARM_POLL_ERR;
>  
>  	apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
>  	if (unlikely(!apoll))
> -		return false;
> +		return IO_ARM_POLL_ERR;
>  	apoll->double_poll = NULL;
>  
>  	req->flags |= REQ_F_POLLED;
> @@ -5194,12 +5197,12 @@ static bool io_arm_poll_handler(struct io_kiocb *req)
>  	if (ret || ipt.error) {
>  		io_poll_remove_double(req);
>  		spin_unlock_irq(&ctx->completion_lock);
> -		return false;
> +		return ret?IO_ARM_POLL_READY:IO_ARM_POLL_ERR;

spaces would be great.

>  	}
>  	spin_unlock_irq(&ctx->completion_lock);
>  	trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
>  					apoll->poll.events);
> -	return true;
> +	return IO_ARM_POLL_OK;
>  }
>  
>  static bool __io_poll_remove_one(struct io_kiocb *req,
> @@ -6416,6 +6419,7 @@ static void __io_queue_sqe(struct io_kiocb *req)
>  	struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
>  	int ret;
>  
> +issue_sqe:
>  	ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
>  
>  	/*
> @@ -6435,12 +6439,16 @@ static void __io_queue_sqe(struct io_kiocb *req)
>  			io_put_req(req);
>  		}
>  	} else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
> -		if (!io_arm_poll_handler(req)) {
> +		switch (io_arm_poll_handler(req)) {
> +		case IO_ARM_POLL_READY:
> +			goto issue_sqe;

Checked assembly, the fast path looks ok (i.e. not affected).
Also, a note, linked_timeout is handled correctly.

> +		case IO_ARM_POLL_ERR:
>  			/*
>  			 * Queued up for async execution, worker will release
>  			 * submit reference when the iocb is actually submitted.
>  			 */
>  			io_queue_async_work(req);
> +			break;
>  		}
>  	} else {
>  		io_req_complete_failed(req, ret);
> 

-- 
Pavel Begunkov