}
static int __io_queue_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
- struct sqe_submit *s)
+ struct sqe_submit *s, bool force_nonblock)
{
int ret;
- ret = __io_submit_sqe(ctx, req, s, true);
+ ret = __io_submit_sqe(ctx, req, s, force_nonblock);
if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
struct io_uring_sqe *sqe_copy;
}
static int io_queue_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
- struct sqe_submit *s)
+ struct sqe_submit *s, bool force_nonblock)
{
int ret;
return 0;
}
- return __io_queue_sqe(ctx, req, s);
+ return __io_queue_sqe(ctx, req, s, force_nonblock);
}
static int io_queue_link_head(struct io_ring_ctx *ctx, struct io_kiocb *req,
- struct sqe_submit *s, struct io_kiocb *shadow)
+ struct sqe_submit *s, struct io_kiocb *shadow,
+ bool force_nonblock)
{
int ret;
int need_submit = false;
if (!shadow)
- return io_queue_sqe(ctx, req, s);
+ return io_queue_sqe(ctx, req, s, force_nonblock);
/*
* Mark the first IO in link list as DRAIN, let all the following
spin_unlock_irq(&ctx->completion_lock);
if (need_submit)
- return __io_queue_sqe(ctx, req, s);
+ return __io_queue_sqe(ctx, req, s, force_nonblock);
return 0;
}
#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK)
static void io_submit_sqe(struct io_ring_ctx *ctx, struct sqe_submit *s,
- struct io_submit_state *state, struct io_kiocb **link)
+ struct io_submit_state *state, struct io_kiocb **link,
+ bool force_nonblock)
{
struct io_uring_sqe *sqe_copy;
struct io_kiocb *req;
INIT_LIST_HEAD(&req->link_list);
*link = req;
} else {
- io_queue_sqe(ctx, req, s);
+ io_queue_sqe(ctx, req, s, force_nonblock);
}
}
* that's the end of the chain. Submit the previous link.
*/
if (!prev_was_link && link) {
- io_queue_link_head(ctx, link, &link->submit, shadow_req);
+ io_queue_link_head(ctx, link, &link->submit, shadow_req,
+ true);
link = NULL;
}
prev_was_link = (sqes[i].sqe->flags & IOSQE_IO_LINK) != 0;
sqes[i].has_user = has_user;
sqes[i].needs_lock = true;
sqes[i].needs_fixed_file = true;
- io_submit_sqe(ctx, &sqes[i], statep, &link);
+ io_submit_sqe(ctx, &sqes[i], statep, &link, true);
submitted++;
}
}
if (link)
- io_queue_link_head(ctx, link, &link->submit, shadow_req);
+ io_queue_link_head(ctx, link, &link->submit, shadow_req, true);
if (statep)
io_submit_state_end(&state);
return 0;
}
-static int io_ring_submit(struct io_ring_ctx *ctx, unsigned int to_submit)
+static int io_ring_submit(struct io_ring_ctx *ctx, unsigned int to_submit,
+ bool block_for_last)
{
struct io_submit_state state, *statep = NULL;
struct io_kiocb *link = NULL;
}
for (i = 0; i < to_submit; i++) {
+ bool force_nonblock = true;
struct sqe_submit s;
if (!io_get_sqring(ctx, &s))
* that's the end of the chain. Submit the previous link.
*/
if (!prev_was_link && link) {
- io_queue_link_head(ctx, link, &link->submit, shadow_req);
+ io_queue_link_head(ctx, link, &link->submit, shadow_req,
+ force_nonblock);
link = NULL;
}
prev_was_link = (s.sqe->flags & IOSQE_IO_LINK) != 0;
s.needs_lock = false;
s.needs_fixed_file = false;
submit++;
- io_submit_sqe(ctx, &s, statep, &link);
+
+ /*
+ * The caller will block for events after submit, submit the
+ * last IO non-blocking. This is either the only IO it's
+ * submitting, or it already submitted the previous ones. This
+ * improves performance by avoiding an async punt that we don't
+ * need to do.
+ */
+ if (block_for_last && submit == to_submit)
+ force_nonblock = false;
+
+ io_submit_sqe(ctx, &s, statep, &link, force_nonblock);
}
io_commit_sqring(ctx);
if (link)
- io_queue_link_head(ctx, link, &link->submit, shadow_req);
+ io_queue_link_head(ctx, link, &link->submit, shadow_req,
+ block_for_last);
if (statep)
io_submit_state_end(statep);
ret = 0;
if (to_submit) {
+ bool block_for_last = false;
+
to_submit = min(to_submit, ctx->sq_entries);
+ /*
+ * Allow last submission to block in a series, IFF the caller
+ * asked to wait for events and we don't currently have
+ * enough. This potentially avoids an async punt.
+ */
+ if (to_submit == min_complete &&
+ io_cqring_events(ctx->rings) < min_complete)
+ block_for_last = true;
+
mutex_lock(&ctx->uring_lock);
- submitted = io_ring_submit(ctx, to_submit);
+ submitted = io_ring_submit(ctx, to_submit, block_for_last);
mutex_unlock(&ctx->uring_lock);
}
if (flags & IORING_ENTER_GETEVENTS) {