Re: [PATCH 3/4] io_uring: tweak iopoll return for REQ_F_CQE_SKIP

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

On 12/4/21 23:20, Jens Axboe wrote:
> On 12/4/21 3:48 PM, Pavel Begunkov wrote:
>> On 12/4/21 22:21, Jens Axboe wrote:
>>> On 12/4/21 1:49 PM, Pavel Begunkov wrote:
>>>> Currently, IOPOLL returns the number of completed requests, but with
>>>> REQ_F_CQE_SKIP there are not the same thing anymore. That may be
>>>> confusing as non-iopoll wait cares only about CQEs, so make io_do_iopoll
>>>> return the number of posted CQEs.
>>>>
>>>> Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
>>>> ---
>>>>    fs/io_uring.c | 6 +++---
>>>>    1 file changed, 3 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>>> index 64add8260abb..ea7a0daa0b3b 100644
>>>> --- a/fs/io_uring.c
>>>> +++ b/fs/io_uring.c
>>>> @@ -2538,10 +2538,10 @@ static int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
>>>>    		/* order with io_complete_rw_iopoll(), e.g. ->result updates */
>>>>    		if (!smp_load_acquire(&req->iopoll_completed))
>>>>    			break;
>>>> +		if (unlikely(req->flags & REQ_F_CQE_SKIP))
>>>> +			continue;
>>>>    
>>>> -		if (!(req->flags & REQ_F_CQE_SKIP))
>>>> -			__io_fill_cqe(ctx, req->user_data, req->result,
>>>> -				      io_put_kbuf(req));
>>>> +		__io_fill_cqe(ctx, req->user_data, req->result, io_put_kbuf(req));
>>>>    		nr_events++;
>>>>    	}
>>>>    
>>>
>>> Not sure I follow the logic behind this change. Places like
>>> io_iopoll_try_reap_events() just need a "did we find anything" return,
>>> which is independent on whether or not we actually posted CQEs or not.
>>> Other callers either don't care what the return value is or if it's < 0
>>> or not (which this change won't affect).
>>>
>>> I feel like I'm missing something here, or that the commit message
>>> better needs to explain why this change is done.
>>
>> I was wrong on how I described it, but it means that the problem is in
>> a different place.
>>
>> int io_do_iopoll() {
>> 	return nr_events;
>> }
>>
>> int io_iopoll_check() {
>> 	do {
>> 		nr_events += io_do_iopoll();
>> 	while (nr_events < min && ...);
>> }
>>
>> And "events" there better to be CQEs, otherwise the semantics
>> of @min + CQE_SKIP is not very clear and mismatches non-IOPOLL.
> 
> Can you do a v2 of this patch? Rest of them look good to me.

I'll send it tomorrow with clarifications added

-- 
Pavel Begunkov