@@ -1273,7 +1273,12 @@ static void io_req_local_work_add(struct io_kiocb *req)
if (ctx->has_evfd)
io_eventfd_signal(ctx);
- __io_cqring_wake(ctx);
+
+ if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
+ wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE);
+ } else {
+ __io_cqring_wake(ctx);
+ }
}
void __io_req_task_work_add(struct io_kiocb *req, bool allow_local)
@@ -2558,12 +2563,17 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
do {
unsigned long check_cq;
- prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq,
- TASK_INTERRUPTIBLE);
+ if (ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
+ set_current_state(TASK_INTERRUPTIBLE);
+ } else {
+ prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq,
+ TASK_INTERRUPTIBLE);
+ }
+
ret = io_cqring_wait_schedule(ctx, &iowq, timeout);
+ __set_current_state(TASK_RUNNING);
if (ret < 0)
break;
- __set_current_state(TASK_RUNNING);
/*
* Run task_work after scheduling and before io_should_wake().
* If we got woken because of task_work being processed, run it
@@ -2591,7 +2601,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
cond_resched();
} while (1);
- finish_wait(&ctx->cq_wait, &iowq.wq);
+ if (!(ctx->flags & IORING_SETUP_DEFER_TASKRUN))
+ finish_wait(&ctx->cq_wait, &iowq.wq);
restore_saved_sigmask_unless(ret == -EINTR);
return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
With DEFER_TASKRUN only ctx->submitter_task might be waiting for CQEs, we can use this to optimise io_cqring_wait(). Replace ->cq_wait waitqueue with waking the task directly. It works but misses an important optimisation covered by the following patch, so this patch without follow ups might hurt performance. Signed-off-by: Pavel Begunkov <asml.silence@gmail.com> --- io_uring/io_uring.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-)