On 8/20/24 6:08 PM, Pavel Begunkov wrote: > On 8/20/24 23:58, Jens Axboe wrote: >> On 8/20/24 4:47 PM, Pavel Begunkov wrote: >>> On 8/20/24 23:46, Pavel Begunkov wrote: >>>> On 8/20/24 00:28, Jens Axboe wrote: >>>>> Waiting for events with io_uring has two knobs that can be set: >>>>> >>>>> 1) The number of events to wake for >>>>> 2) The timeout associated with the event >>>>> >>>>> Waiting will abort when either of those conditions are met, as expected. >>>>> >>>>> This adds support for a third event, which is associated with the number >>>>> of events to wait for. Applications generally like to handle batches of >>>>> completions, and right now they'd set a number of events to wait for and >>>>> the timeout for that. If no events have been received but the timeout >>>>> triggers, control is returned to the application and it can wait again. >>>>> However, if the application doesn't have anything to do until events are >>>>> reaped, then it's possible to make this waiting more efficient. >>>>> >>>>> For example, the application may have a latency time of 50 usecs and >>>>> wanting to handle a batch of 8 requests at the time. If it uses 50 usecs >>>>> as the timeout, then it'll be doing 20K context switches per second even >>>>> if nothing is happening. >>>>> >>>>> This introduces the notion of min batch wait time. If the min batch wait >>>>> time expires, then we'll return to userspace if we have any events at all. >>>>> If none are available, the general wait time is applied. Any request >>>>> arriving after the min batch wait time will cause waiting to stop and >>>>> return control to the application. >>>>> >>>>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx> >>>>> --- >>>>> io_uring/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++++------ >>>>> io_uring/io_uring.h | 2 ++ >>>>> 2 files changed, 67 insertions(+), 10 deletions(-) >>>>> >>>>> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c >>>>> index ddfbe04c61ed..d09a7c2e1096 100644 >>>>> --- a/io_uring/io_uring.c >>>>> +++ b/io_uring/io_uring.c >>>>> @@ -2363,13 +2363,62 @@ static enum hrtimer_restart io_cqring_timer_wakeup(struct hrtimer *timer) >>>>> return HRTIMER_NORESTART; >>>>> } >>>>> +/* >>>>> + * Doing min_timeout portion. If we saw any timeouts, events, or have work, >>>>> + * wake up. If not, and we have a normal timeout, switch to that and keep >>>>> + * sleeping. >>>>> + */ >>>>> +static enum hrtimer_restart io_cqring_min_timer_wakeup(struct hrtimer *timer) >>>>> +{ >>>>> + struct io_wait_queue *iowq = container_of(timer, struct io_wait_queue, t); >>>>> + struct io_ring_ctx *ctx = iowq->ctx; >>>>> + >>>>> + /* no general timeout, or shorter, we are done */ >>>>> + if (iowq->timeout == KTIME_MAX || >>>>> + ktime_after(iowq->min_timeout, iowq->timeout)) >>>>> + goto out_wake; >>>>> + /* work we may need to run, wake function will see if we need to wake */ >>>>> + if (io_has_work(ctx)) >>>>> + goto out_wake; >>>>> + /* got events since we started waiting, min timeout is done */ >>>>> + if (iowq->cq_min_tail != READ_ONCE(ctx->rings->cq.tail)) >>>>> + goto out_wake; >>>>> + /* if we have any events and min timeout expired, we're done */ >>>>> + if (io_cqring_events(ctx)) >>>>> + goto out_wake; >>>>> + >>>>> + /* >>>>> + * If using deferred task_work running and application is waiting on >>>>> + * more than one request, ensure we reset it now where we are switching >>>>> + * to normal sleeps. Any request completion post min_wait should wake >>>>> + * the task and return. >>>>> + */ >>>>> + if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) >>>>> + atomic_set(&ctx->cq_wait_nr, 1); >>>> >>>> racy >>>> >>>> atomic_set(&ctx->cq_wait_nr, 1); >>>> smp_mb(); >>>> if (llist_empty(&ctx->work_llist)) >>>> // wake; >>> >>> rather if _not_ empty >> >> Yep that one was a given :-) >> >> Updated it, we'll punt to out_wake at that point. > > Another concern is racing with the task [re]setting ->cq_wait_nr > in io_cqring_wait(), e.g. because of a spurious wake up. I tried to close that up somewhat in the next iteration. -- Jens Axboe