[PATCH v7] io_uring: releasing CPU resources when polling

[hexue]

This patch add a new hybrid poll at io_uring level, it also set a signal
"IORING_SETUP_HY_POLL" to application, aim to provide a interface for users
to enable use new hybrid polling flexibly.

io_uring use polling mode could improve the IO performence, but it will
spend 100% of CPU resources to do polling.

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.

Test Result
fio-3.35, 16 poll queues, 1 thread
-------------------------------------------------------------------------
Performance
-------------------------------------------------------------------------
                write         read        randwrite  randread
regular poll BW=3939MiB/s  BW=6613MiB/s  IOPS=190K  IOPS=470K
IRQ          BW=3927MiB/s  BW=6612MiB/s  IOPS=181K  IOPS=203K
hybrid poll  BW=3937MiB/s  BW=6623MiB/s  IOPS=190K  IOPS=358K(suboptimal)
-------------------------------------------------------------------------
CPU Utilization
------------------------------------------------------
                write   read    randwrite   randread
regular poll    100%    100%    100%        100%
IRQ             50%     53%     100%        100%
hybrid poll     70%     37%     70%         85%
------------------------------------------------------

--
changes since v6:
- Modified IO path, distinct iopoll and uring_cmd_iopoll
- update test results

changes since v5:
- Remove cstime recorder
- Use minimize sleep time in different drivers
- Use the half of whole runtime to do schedule
- Consider as a suboptimal solution between
  regular poll and IRQ

changes since v4:
- Rewrote the commit
- Update the test results
- Reorganized the code basd on 6.11

changes since v3:
- Simplified the commit
- Add some comments on code

changes since v2:
- Modified some formatting errors
- Move judgement to poll path

changes since v1:
- Extend hybrid poll to async polled io

Signed-off-by: hexue <xue01.he@xxxxxxxxxxx>
---
 include/linux/io_uring_types.h |   6 ++
 include/uapi/linux/io_uring.h  |   1 +
 io_uring/io_uring.c            |   3 +-
 io_uring/rw.c                  | 100 +++++++++++++++++++++++++++++----
 4 files changed, 99 insertions(+), 11 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 91224bbcfa73..0897126fb2d7 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -428,6 +428,8 @@ struct io_ring_ctx {
 	unsigned short			n_sqe_pages;
 	struct page			**ring_pages;
 	struct page			**sqe_pages;
+	/* for hybrid poll*/
+	u64			available_time;
 };
 
 struct io_tw_state {
@@ -665,6 +667,10 @@ struct io_kiocb {
 		u64			extra1;
 		u64			extra2;
 	} big_cqe;
+    /* for hybrid iopoll */
+	bool		poll_state;
+	u64			iopoll_start;
+	u64			iopoll_end;
 };
 
 struct io_overflow_cqe {
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 994bf7af0efe..ef32ec319d1f 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -199,6 +199,7 @@ enum io_uring_sqe_flags_bit {
  * Removes indirection through the SQ index array.
  */
 #define IORING_SETUP_NO_SQARRAY		(1U << 16)
+#define IORING_SETUP_HY_POLL	(1U << 17)
 
 enum io_uring_op {
 	IORING_OP_NOP,
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 816e93e7f949..b38f8af118c5 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -299,6 +299,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 		goto err;
 
 	ctx->flags = p->flags;
+	ctx->available_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);
@@ -3637,7 +3638,7 @@ 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_HY_POLL))
 		return -EINVAL;
 
 	return io_uring_create(entries, &p, params);
diff --git a/io_uring/rw.c b/io_uring/rw.c
index 1a2128459cb4..eff040f5e5f6 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -772,6 +772,13 @@ static bool need_complete_io(struct io_kiocb *req)
 		S_ISBLK(file_inode(req->file)->i_mode);
 }
 
+static void init_hybrid_poll(struct io_ring_ctx *ctx, struct io_kiocb *req)
+{
+	/* make sure every req only block once*/
+	req->poll_state = false;
+	req->iopoll_start = ktime_get_ns();
+}
+
 static int io_rw_init_file(struct io_kiocb *req, fmode_t mode)
 {
 	struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw);
@@ -809,6 +816,8 @@ static int io_rw_init_file(struct io_kiocb *req, fmode_t mode)
 		kiocb->ki_flags |= IOCB_HIPRI;
 		kiocb->ki_complete = io_complete_rw_iopoll;
 		req->iopoll_completed = 0;
+		if (ctx->flags & IORING_SETUP_HY_POLL)
+			init_hybrid_poll(ctx, req);
 	} else {
 		if (kiocb->ki_flags & IOCB_HIPRI)
 			return -EINVAL;
@@ -1106,6 +1115,83 @@ 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)
+{
+	int ret;
+	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);
+		ret = file->f_op->uring_cmd_iopoll(ioucmd, iob,
+										poll_flags);
+	} else {
+		struct io_rw *rw = io_kiocb_to_cmd(req, struct io_rw);
+
+		ret = file->f_op->iopoll(&rw->kiocb, iob, poll_flags);
+	}
+	return ret;
+}
+
+static u64 io_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->poll_state)
+		return 0;
+
+	if (ctx->available_time == LLONG_MAX)
+		return 0;
+
+	/* Using half running time to do schedul */
+	sleep_time = ctx->available_time / 2;
+
+	kt = ktime_set(0, sleep_time);
+	req->poll_state = true;
+
+	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;
+	int ret;
+	u64 runtime, sleep_time;
+
+	sleep_time = io_delay(ctx, req);
+	ret = io_uring_classic_poll(req, iob, poll_flags);
+	req->iopoll_end = ktime_get_ns();
+	runtime = req->iopoll_end - req->iopoll_start - sleep_time;
+	if (runtime < 0)
+		return 0;
+
+	/* use minimize sleep time if there are different speed
+	 * drivers, it could get more completions from fast one
+	 */
+	if (ctx->available_time > runtime)
+		ctx->available_time = runtime;
+	return ret;
+}
+
 int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
 {
 	struct io_wq_work_node *pos, *start, *prev;
@@ -1133,17 +1219,11 @@ 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_HY_POLL)
+			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