[PATCH v9 1/1] io_uring: releasing CPU resources when polling

[hexue <xue01.he@samsung.com>]

A new hybrid poll is implemented on the io_uring layer. Once IO issued,
it will not polling immediately, but block first and re-run before IO
complete, then poll to reap IO. This poll function could be a suboptimal
solution when running on a single thread, it offers the performance lower
than regular polling but higher than IRQ, and CPU utilization is also lower
than polling.

Signed-off-by: hexue <xue01.he@samsung.com>
---
 include/linux/io_uring_types.h | 19 ++++++-
 include/uapi/linux/io_uring.h  |  3 ++
 io_uring/io_uring.c            |  8 ++-
 io_uring/rw.c                  | 92 ++++++++++++++++++++++++++++++----
 4 files changed, 108 insertions(+), 14 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 4b9ba523978d..4a85a823b888 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -302,6 +302,11 @@ struct io_ring_ctx {
 		 * ->uring_cmd() by io_uring_cmd_insert_cancelable()
 		 */
 		struct hlist_head	cancelable_uring_cmd;
+		/*
+		 * For Hybrid IOPOLL, runtime in hybrid polling, without
+		 * scheduling time
+		 */
+		u64					hybrid_poll_time;
 	} ____cacheline_aligned_in_smp;
 
 	struct {
@@ -447,6 +452,7 @@ enum {
 	REQ_F_LINK_TIMEOUT_BIT,
 	REQ_F_NEED_CLEANUP_BIT,
 	REQ_F_POLLED_BIT,
+	REQ_F_HYBRID_IOPOLL_STATE_BIT,
 	REQ_F_BUFFER_SELECTED_BIT,
 	REQ_F_BUFFER_RING_BIT,
 	REQ_F_REISSUE_BIT,
@@ -506,6 +512,8 @@ enum {
 	REQ_F_NEED_CLEANUP	= IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
 	/* already went through poll handler */
 	REQ_F_POLLED		= IO_REQ_FLAG(REQ_F_POLLED_BIT),
+	/* every req only blocks once in hybrid poll */
+	REQ_F_IOPOLL_STATE        = IO_REQ_FLAG(REQ_F_HYBRID_IOPOLL_STATE_BIT),
 	/* buffer already selected */
 	REQ_F_BUFFER_SELECTED	= IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
 	/* buffer selected from ring, needs commit */
@@ -643,8 +651,15 @@ struct io_kiocb {
 	atomic_t			refs;
 	bool				cancel_seq_set;
 	struct io_task_work		io_task_work;
-	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
-	struct hlist_node		hash_node;
+	union {
+		/*
+		 * for polled requests, i.e. IORING_OP_POLL_ADD and async armed
+		 * poll
+		 */
+		struct hlist_node	hash_node;
+		/* For IOPOLL setup queues, with hybrid polling */
+		u64                     iopoll_start;
+	};
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 1fe79e750470..ddd6e42b134d 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -200,6 +200,9 @@ enum io_uring_sqe_flags_bit {
  */
 #define IORING_SETUP_NO_SQARRAY		(1U << 16)
 
+/* Use hybrid poll in iopoll process */
+#define IORING_SETUP_HYBRID_IOPOLL	(1U << 17)
+
 enum io_uring_op {
 	IORING_OP_NOP,
 	IORING_OP_READV,
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 4199fbe6ce13..ed131fc824a0 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -301,6 +301,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 		goto err;
 
 	ctx->flags = p->flags;
+	ctx->hybrid_poll_time = LLONG_MAX;
 	atomic_set(&ctx->cq_wait_nr, IO_CQ_WAKE_INIT);
 	init_waitqueue_head(&ctx->sqo_sq_wait);
 	INIT_LIST_HEAD(&ctx->sqd_list);
@@ -3545,6 +3546,11 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 	ctx->clockid = CLOCK_MONOTONIC;
 	ctx->clock_offset = 0;
 
+	/* HYBRID_IOPOLL only valid with IOPOLL */
+	if ((ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_HYBRID_IOPOLL)) ==
+			IORING_SETUP_HYBRID_IOPOLL)
+		return -EINVAL;
+
 	if ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) &&
 	    !(ctx->flags & IORING_SETUP_IOPOLL) &&
 	    !(ctx->flags & IORING_SETUP_SQPOLL))
@@ -3724,7 +3730,7 @@ static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
 			IORING_SETUP_SQE128 | IORING_SETUP_CQE32 |
 			IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN |
 			IORING_SETUP_NO_MMAP | IORING_SETUP_REGISTERED_FD_ONLY |
-			IORING_SETUP_NO_SQARRAY))
+			IORING_SETUP_NO_SQARRAY | IORING_SETUP_HYBRID_IOPOLL))
 		return -EINVAL;
 
 	return io_uring_create(entries, &p, params);
diff --git a/io_uring/rw.c b/io_uring/rw.c
index f023ff49c688..340dc4b7b84f 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -808,6 +808,11 @@ static int io_rw_init_file(struct io_kiocb *req, fmode_t mode, int rw_type)
 		kiocb->ki_flags |= IOCB_HIPRI;
 		kiocb->ki_complete = io_complete_rw_iopoll;
 		req->iopoll_completed = 0;
+		if (ctx->flags & IORING_SETUP_HYBRID_IOPOLL) {
+			/* make sure every req only blocks once*/
+			req->flags &= ~REQ_F_IOPOLL_STATE;
+			req->iopoll_start = ktime_get_ns();
+		}
 	} else {
 		if (kiocb->ki_flags & IOCB_HIPRI)
 			return -EINVAL;
@@ -1112,6 +1117,78 @@ void io_rw_fail(struct io_kiocb *req)
 	io_req_set_res(req, res, req->cqe.flags);
 }
 
+static int io_uring_classic_poll(struct io_kiocb *req, struct io_comp_batch *iob,
+				unsigned int poll_flags)
+{
+	struct file *file = req->file;
+
+	if (req->opcode == IORING_OP_URING_CMD) {
+		struct io_uring_cmd *ioucmd;
+
+		ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
+		return file->f_op->uring_cmd_iopoll(ioucmd, iob, poll_flags);
+	} else {
+		struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw);
+
+		return file->f_op->iopoll(&rw->kiocb, iob, poll_flags);
+	}
+}
+
+static u64 io_hybrid_iopoll_delay(struct io_ring_ctx *ctx, struct io_kiocb *req)
+{
+	struct hrtimer_sleeper timer;
+	enum hrtimer_mode mode;
+	ktime_t kt;
+	u64 sleep_time;
+
+	if (req->flags & REQ_F_IOPOLL_STATE)
+		return 0;
+
+	if (ctx->hybrid_poll_time == LLONG_MAX)
+		return 0;
+
+	/* Using half the running time to do schedule */
+	sleep_time = ctx->hybrid_poll_time / 2;
+
+	kt = ktime_set(0, sleep_time);
+	req->flags |= REQ_F_IOPOLL_STATE;
+
+	mode = HRTIMER_MODE_REL;
+	hrtimer_init_sleeper_on_stack(&timer, CLOCK_MONOTONIC, mode);
+	hrtimer_set_expires(&timer.timer, kt);
+	set_current_state(TASK_INTERRUPTIBLE);
+	hrtimer_sleeper_start_expires(&timer, mode);
+
+	if (timer.task)
+		io_schedule();
+
+	hrtimer_cancel(&timer.timer);
+	__set_current_state(TASK_RUNNING);
+	destroy_hrtimer_on_stack(&timer.timer);
+	return sleep_time;
+}
+
+static int io_uring_hybrid_poll(struct io_kiocb *req,
+				struct io_comp_batch *iob, unsigned int poll_flags)
+{
+	struct io_ring_ctx *ctx = req->ctx;
+	u64 runtime, sleep_time;
+	int ret;
+
+	sleep_time = io_hybrid_iopoll_delay(ctx, req);
+	ret = io_uring_classic_poll(req, iob, poll_flags);
+	runtime = ktime_get_ns() - req->iopoll_start - sleep_time;
+
+	/*
+	 * Use minimum sleep time if we're polling devices with different
+	 * latencies. We could get more completions from the faster ones.
+	 */
+	if (ctx->hybrid_poll_time > runtime)
+		ctx->hybrid_poll_time = runtime;
+
+	return ret;
+}
+
 int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
 {
 	struct io_wq_work_node *pos, *start, *prev;
@@ -1128,7 +1205,6 @@ int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
 
 	wq_list_for_each(pos, start, &ctx->iopoll_list) {
 		struct io_kiocb *req = container_of(pos, struct io_kiocb, comp_list);
-		struct file *file = req->file;
 		int ret;
 
 		/*
@@ -1139,17 +1215,11 @@ int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
 		if (READ_ONCE(req->iopoll_completed))
 			break;
 
-		if (req->opcode == IORING_OP_URING_CMD) {
-			struct io_uring_cmd *ioucmd;
-
-			ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
-			ret = file->f_op->uring_cmd_iopoll(ioucmd, &iob,
-								poll_flags);
-		} else {
-			struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw);
+		if (ctx->flags & IORING_SETUP_HYBRID_IOPOLL)
+			ret = io_uring_hybrid_poll(req, &iob, poll_flags);
+		else
+			ret = io_uring_classic_poll(req, &iob, poll_flags);
 
-			ret = file->f_op->iopoll(&rw->kiocb, &iob, poll_flags);
-		}
 		if (unlikely(ret < 0))
 			return ret;
 		else if (ret)
-- 
2.40.1