On 9/12/21 7:24 PM, Hao Xu wrote:
> 在 2021/9/11 下午9:52, Pavel Begunkov 写道:
>> Split io_cqring_fill_event() into a couple of more targeted functions.
>> The first on is io_fill_cqe_aux() for completions that are not
>> associated with request completions and doing the ->cq_extra accounting.
>> Examples are additional CQEs from multishot poll and rsrc notifications.
>>
>> The second is io_fill_cqe_req(), should be called when it's a normal
>> request completion. Nothing more to it at the moment, will be used in
>> later patches.
>>
>> The last one is inlined __io_fill_cqe() for a finer grained control,
>> should be used with caution and in hottest places.
>>
>> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
>> ---
[...]
>> @@ -5293,13 +5297,12 @@ static bool __io_poll_complete(struct io_kiocb *req, __poll_t mask)
>> }
>> if (req->poll.events & EPOLLONESHOT)
>> flags = 0;
>> - if (!io_cqring_fill_event(ctx, req->user_data, error, flags)) {
>> + if (!(flags & IORING_CQE_F_MORE)) {
>> + io_fill_cqe_req(req, error, flags);
> We should check the return value of io_fill_cqe_req() and do
> req->poll.done = true if the return value is false, which means ocqe
> allocation failed. Though I think the current poll.done logic itself
> is not right.(I've changed it in another patch)
.done was serving a purpose for cancellations and some other places of
convenience. There is not much difference for the machinery whether we
set it here or not, because the success case doesn't do it, and in both
cases requests will be put to end. If there is a bug it will be just
triggerable with successfully emitting a CQE.
I saw the poll fixes, will be reading through later unless Jens beats
me on that and will keep them in mind when rebasing the series.
>> + } else if (!io_fill_cqe_aux(ctx, req->user_data, error, flags)) {
>> req->poll.done = true;
>> flags = 0;
>> }
>> - if (flags & IORING_CQE_F_MORE)
>> - ctx->cq_extra++;
>> -
>> return !(flags & IORING_CQE_F_MORE);
>> }
--
Pavel Begunkov
