Skip to contentSkip to conversation

task work optimization

PATCH Superseded Hao Xu 2 participants 16 messages Oct 27, 2021 – Oct 29, 2021

Subsystems (2)

v3 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Hao Xu

author
#0 lore raw

v3 cover task work optimization

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:

   depth     no 8/8   include 8/8      before this patchset
    1        7.05         7.82              7.10
    2        8.47         8.48              8.60
    4        10.42        9.99              10.42
    8        13.78        13.13             13.22
    16       27.41        27.92             24.33
    32       49.40        46.16             53.08
    64       102.53       105.68            103.36
    128      196.98       202.76            205.61
    256      372.99       375.61            414.88
    512      747.23       763.95            791.30
    1024     1472.59      1527.46           1538.72
    2048     3153.49      3129.22           3329.01
    4096     6387.86      5899.74           6682.54
    8192     12150.25     12433.59          12774.14
    16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
io-wq: add helper to merge two wq_lists
io_uring: add a priority tw list for irq completion work
io_uring: add helper for task work execution code
io_uring: split io_req_complete_post() and add a helper
io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
io_uring: add nr_ctx to record the number of ctx in a task
io_uring: batch completion in prior_task_list
io_uring: add limited number of TWs to priority task list

 fs/io-wq.h    |  21 +++++++
 fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------
 2 files changed, 138 insertions(+), 51 deletions(-)
Signature · 1 line
2.24.4
author
#1 lore raw

PATCH v3 1/8 io-wq: add helper to merge two wq_lists

add a helper to merge two wq_lists, it will be useful in the next patches.

---
 fs/io-wq.h | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
fs/io-wq.h +21 -0
index 41bf37674a49..a7b0b505db9d 100644
@@ -52,6 +52,27 @@ static inline void wq_list_add_after(struct io_wq_work_node *node,
 		list->last = node;
 }
 
+/**
+ * wq_list_merge - merge the second list to the first one.
+ * @list0: the first list
+ * @list1: the second list
+ * Return list0 if list1 is NULL, and vice versa.
+ * Otherwise after merge, list0 contains the merged list.
+ */
+static inline struct io_wq_work_list *wq_list_merge(struct io_wq_work_list *list0,
+						    struct io_wq_work_list *list1)
+{
+	if (!list1 || !list1->first)
+		return list0;
+
+	if (!list0 || !list0->first)
+		return list1;
+
+	list0->last->next = list1->first;
+	list0->last = list1->last;
+	return list0;
+}
+
 static inline void wq_list_add_tail(struct io_wq_work_node *node,
 				    struct io_wq_work_list *list)
 {
Signature · 1 line
2.24.4
author
#2 lore raw

PATCH v3 2/8 io_uring: add a priority tw list for irq completion work

In reply to Hao Xu

Now we have a lot of task_work users, some are just to complete a req and generate a cqe. Let's put the work to a new tw list which has a higher priority, so that it can be handled quickly and thus to reduce avg req latency. an explanatory case:

origin timeline:

    submit_sqe-->irq-->add completion task_work
    -->run heavy work0~n-->run completion task_work
now timeline:
    submit_sqe-->irq-->add completion task_work
    -->run completion task_work-->run heavy work0~n

Tested this patch(and the following ones) by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

Tried various iodepth.
The peak IOPS for this patch is 710K, while the old one is 665K. For avg latency, difference shows when iodepth grow:
depth and avg latency(usec):

	depth      new          old
	 1        7.05         7.10
	 2        8.47         8.60
	 4        10.42        10.42
	 8        13.78        13.22
	 16       27.41        24.33
	 32       49.40        53.08
	 64       102.53       103.36
	 128      196.98       205.61
	 256      372.99       414.88
         512      747.23       791.30
         1024     1472.59      1538.72
         2048     3153.49      3329.01
         4096     6387.86      6682.54
         8192     12150.25     12774.14
         16384    23085.58     26044.71
---
 fs/io_uring.c | 38 +++++++++++++++++++++++++-------------
 1 file changed, 25 insertions(+), 13 deletions(-)
fs/io_uring.c +25 -13
index 88c5ee4dc242..9fdac3b9a560 100644
@@ -467,6 +467,7 @@ struct io_uring_task {
 
 	spinlock_t		task_lock;
 	struct io_wq_work_list	task_list;
+	struct io_wq_work_list	prior_task_list;
 	struct callback_head	task_work;
 	bool			task_running;
 };
@@ -2147,13 +2148,17 @@ static void tctx_task_work(struct callback_head *cb)
 
 	while (1) {
 		struct io_wq_work_node *node;
+		struct io_wq_work_list *merged_list;
 
-		if (!tctx->task_list.first && locked)
+		if (!tctx->prior_task_list.first &&
+		    !tctx->task_list.first && locked)
 			io_submit_flush_completions(ctx);
 
 		spin_lock_irq(&tctx->task_lock);
-		node = tctx->task_list.first;
+		merged_list = wq_list_merge(&tctx->prior_task_list, &tctx->task_list);
+		node = merged_list->first;
 		INIT_WQ_LIST(&tctx->task_list);
+		INIT_WQ_LIST(&tctx->prior_task_list);
 		if (!node)
 			tctx->task_running = false;
 		spin_unlock_irq(&tctx->task_lock);
@@ -2182,19 +2187,23 @@ static void tctx_task_work(struct callback_head *cb)
 	ctx_flush_and_put(ctx, &locked);
 }
 
-static void io_req_task_work_add(struct io_kiocb *req)
+static void io_req_task_work_add(struct io_kiocb *req, bool priority)
 {
 	struct task_struct *tsk = req->task;
 	struct io_uring_task *tctx = tsk->io_uring;
 	enum task_work_notify_mode notify;
 	struct io_wq_work_node *node;
+	struct io_wq_work_list *merged_list;
 	unsigned long flags;
 	bool running;
 
 	WARN_ON_ONCE(!tctx);
 
 	spin_lock_irqsave(&tctx->task_lock, flags);
-	wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
+	if (priority)
+		wq_list_add_tail(&req->io_task_work.node, &tctx->prior_task_list);
+	else
+		wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
 	running = tctx->task_running;
 	if (!running)
 		tctx->task_running = true;
@@ -2219,8 +2228,10 @@ static void io_req_task_work_add(struct io_kiocb *req)
 
 	spin_lock_irqsave(&tctx->task_lock, flags);
 	tctx->task_running = false;
-	node = tctx->task_list.first;
+	merged_list = wq_list_merge(&tctx->prior_task_list, &tctx->task_list);
+	node = merged_list->first;
 	INIT_WQ_LIST(&tctx->task_list);
+	INIT_WQ_LIST(&tctx->prior_task_list);
 	spin_unlock_irqrestore(&tctx->task_lock, flags);
 
 	while (node) {
@@ -2257,19 +2268,19 @@ static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
 {
 	req->result = ret;
 	req->io_task_work.func = io_req_task_cancel;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, true);
 }
 
 static void io_req_task_queue(struct io_kiocb *req)
 {
 	req->io_task_work.func = io_req_task_submit;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, false);
 }
 
 static void io_req_task_queue_reissue(struct io_kiocb *req)
 {
 	req->io_task_work.func = io_queue_async_work;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, false);
 }
 
 static inline void io_queue_next(struct io_kiocb *req)
@@ -2374,7 +2385,7 @@ static inline void io_put_req_deferred(struct io_kiocb *req)
 {
 	if (req_ref_put_and_test(req)) {
 		req->io_task_work.func = io_free_req_work;
-		io_req_task_work_add(req);
+		io_req_task_work_add(req, false);
 	}
 }
 
@@ -2677,7 +2688,7 @@ static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
 		return;
 	req->result = res;
 	req->io_task_work.func = io_req_task_complete;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, true);
 }
 
 static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
@@ -5248,7 +5259,7 @@ static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
 	 * of executing it. We can't safely execute it anyway, as we may not
 	 * have the needed state needed for it anyway.
 	 */
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, false);
 	return 1;
 }
 
@@ -5916,7 +5927,7 @@ static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
 	spin_unlock_irqrestore(&ctx->timeout_lock, flags);
 
 	req->io_task_work.func = io_req_task_timeout;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, false);
 	return HRTIMER_NORESTART;
 }
 
@@ -6859,7 +6870,7 @@ static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
 	spin_unlock_irqrestore(&ctx->timeout_lock, flags);
 
 	req->io_task_work.func = io_req_task_link_timeout;
-	io_req_task_work_add(req);
+	io_req_task_work_add(req, false);
 	return HRTIMER_NORESTART;
 }
 
@@ -8563,6 +8574,7 @@ static __cold int io_uring_alloc_task_context(struct task_struct *task,
 	task->io_uring = tctx;
 	spin_lock_init(&tctx->task_lock);
 	INIT_WQ_LIST(&tctx->task_list);
+	INIT_WQ_LIST(&tctx->prior_task_list);
 	init_task_work(&tctx->task_work, tctx_task_work);
 	return 0;
 }
Signature · 1 line
2.24.4
author
#3 lore raw

PATCH v3 3/8 io_uring: add helper for task work execution code

In reply to Hao Xu

Add a helper for task work execution code. We will use it later.

---
 fs/io_uring.c | 36 ++++++++++++++++++++----------------
 1 file changed, 20 insertions(+), 16 deletions(-)
fs/io_uring.c +20 -16
index 9fdac3b9a560..98abf94b2015 100644
@@ -2139,6 +2139,25 @@ static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked)
 	percpu_ref_put(&ctx->refs);
 }
 
+static void handle_tw_list(struct io_wq_work_node *node, struct io_ring_ctx **ctx, bool *locked)
+{
+	do {
+		struct io_wq_work_node *next = node->next;
+		struct io_kiocb *req = container_of(node, struct io_kiocb,
+						    io_task_work.node);
+
+		if (req->ctx != *ctx) {
+			ctx_flush_and_put(*ctx, locked);
+			*ctx = req->ctx;
+			/* if not contended, grab and improve batching */
+			*locked = mutex_trylock(&(*ctx)->uring_lock);
+			percpu_ref_get(&(*ctx)->refs);
+		}
+		req->io_task_work.func(req, locked);
+		node = next;
+	} while (node);
+}
+
 static void tctx_task_work(struct callback_head *cb)
 {
 	bool locked = false;
@@ -2165,22 +2184,7 @@ static void tctx_task_work(struct callback_head *cb)
 		if (!node)
 			break;
 
-		do {
-			struct io_wq_work_node *next = node->next;
-			struct io_kiocb *req = container_of(node, struct io_kiocb,
-							    io_task_work.node);
-
-			if (req->ctx != ctx) {
-				ctx_flush_and_put(ctx, &locked);
-				ctx = req->ctx;
-				/* if not contended, grab and improve batching */
-				locked = mutex_trylock(&ctx->uring_lock);
-				percpu_ref_get(&ctx->refs);
-			}
-			req->io_task_work.func(req, &locked);
-			node = next;
-		} while (node);
-
+		handle_tw_list(node, &ctx, &locked);
 		cond_resched();
 	}
 
Signature · 1 line
2.24.4
author
#4 lore raw

PATCH v3 4/8 io_uring: split io_req_complete_post() and add a helper

In reply to Hao Xu

Split io_req_complete_post(), this is a prep for the next patch.

---
 fs/io_uring.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
fs/io_uring.c +10 -2
index 98abf94b2015..3c621798dd2f 100644
@@ -1818,12 +1818,11 @@ static noinline bool io_cqring_fill_event(struct io_ring_ctx *ctx, u64 user_data
 	return __io_cqring_fill_event(ctx, user_data, res, cflags);
 }
 
-static void io_req_complete_post(struct io_kiocb *req, s32 res,
+static void __io_req_complete_post(struct io_kiocb *req, s32 res,
 				 u32 cflags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 
-	spin_lock(&ctx->completion_lock);
 	__io_cqring_fill_event(ctx, req->user_data, res, cflags);
 	/*
 	 * If we're the last reference to this request, add to our locked
@@ -1844,6 +1843,15 @@ static void io_req_complete_post(struct io_kiocb *req, s32 res,
 		wq_list_add_head(&req->comp_list, &ctx->locked_free_list);
 		ctx->locked_free_nr++;
 	}
+}
+
+static void io_req_complete_post(struct io_kiocb *req, long res,
+				 unsigned int cflags)
+{
+	struct io_ring_ctx *ctx = req->ctx;
+
+	spin_lock(&ctx->completion_lock);
+	__io_req_complete_post(req, res, cflags);
 	io_commit_cqring(ctx);
 	spin_unlock(&ctx->completion_lock);
 	io_cqring_ev_posted(ctx);
Signature · 1 line
2.24.4
author
#5 lore raw

PATCH v3 5/8 io_uring: move up io_put_kbuf() and io_put_rw_kbuf()

In reply to Hao Xu

Move them up to avoid explicit declaration. We will use them in later patches.

---
 fs/io_uring.c | 36 ++++++++++++++++++------------------
 1 file changed, 18 insertions(+), 18 deletions(-)
fs/io_uring.c +18 -18
index 3c621798dd2f..db5d9189df3a 100644
@@ -2147,6 +2147,24 @@ static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked)
 	percpu_ref_put(&ctx->refs);
 }
 
+static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
+{
+	unsigned int cflags;
+
+	cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
+	cflags |= IORING_CQE_F_BUFFER;
+	req->flags &= ~REQ_F_BUFFER_SELECTED;
+	kfree(kbuf);
+	return cflags;
+}
+
+static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
+{
+	if (likely(!(req->flags & REQ_F_BUFFER_SELECTED)))
+		return 0;
+	return io_put_kbuf(req, req->kbuf);
+}
+
 static void handle_tw_list(struct io_wq_work_node *node, struct io_ring_ctx **ctx, bool *locked)
 {
 	do {
@@ -2416,24 +2434,6 @@ static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
 	return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
 }
 
-static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
-{
-	unsigned int cflags;
-
-	cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
-	cflags |= IORING_CQE_F_BUFFER;
-	req->flags &= ~REQ_F_BUFFER_SELECTED;
-	kfree(kbuf);
-	return cflags;
-}
-
-static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
-{
-	if (likely(!(req->flags & REQ_F_BUFFER_SELECTED)))
-		return 0;
-	return io_put_kbuf(req, req->kbuf);
-}
-
 static inline bool io_run_task_work(void)
 {
 	if (test_thread_flag(TIF_NOTIFY_SIGNAL) || current->task_works) {
Signature · 1 line
2.24.4
author
#6 lore raw

PATCH v3 6/8 io_uring: add nr_ctx to record the number of ctx in a task

In reply to Hao Xu

This is used in the next patch for task_work batch optimization.

---
 fs/io_uring.c | 7 +++++++
 1 file changed, 7 insertions(+)
fs/io_uring.c +7 -0
index db5d9189df3a..7c6d90d693b8 100644
@@ -470,6 +470,7 @@ struct io_uring_task {
 	struct io_wq_work_list	prior_task_list;
 	struct callback_head	task_work;
 	bool			task_running;
+	unsigned int		nr_ctx;
 };
 
 /*
@@ -9655,6 +9656,9 @@ static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx)
 		mutex_lock(&ctx->uring_lock);
 		list_add(&node->ctx_node, &ctx->tctx_list);
 		mutex_unlock(&ctx->uring_lock);
+		spin_lock_irq(&tctx->task_lock);
+		tctx->nr_ctx++;
+		spin_unlock_irq(&tctx->task_lock);
 	}
 	tctx->last = ctx;
 	return 0;
@@ -9692,6 +9696,9 @@ static __cold void io_uring_del_tctx_node(unsigned long index)
 	mutex_lock(&node->ctx->uring_lock);
 	list_del(&node->ctx_node);
 	mutex_unlock(&node->ctx->uring_lock);
+	spin_lock_irq(&tctx->task_lock);
+	tctx->nr_ctx--;
+	spin_unlock_irq(&tctx->task_lock);
 
 	if (tctx->last == node->ctx)
 		tctx->last = NULL;
Signature · 1 line
2.24.4
author
#7 lore raw

Re: [PATCH 6/8] io_uring: add nr_ctx to record the number of ctx in a task

Quoted 3 lines · Hao Xu

This is used in the next patch for task_work batch optimization.

haven't done the sqpoll part. nr_ctx works for
non-sqpoll mode for now.

Quoted 37 lines
---
  fs/io_uring.c | 7 +++++++
  1 file changed, 7 insertions(+)
fs/io_uring.c +7 -0
index db5d9189df3a..7c6d90d693b8 100644
@@ -470,6 +470,7 @@ struct io_uring_task {
  	struct io_wq_work_list	prior_task_list;
  	struct callback_head	task_work;
  	bool			task_running;
+	unsigned int		nr_ctx;
  };
 
  /*
@@ -9655,6 +9656,9 @@ static int __io_uring_add_tctx_node(struct io_ring_ctx *ctx)
  		mutex_lock(&ctx->uring_lock);
  		list_add(&node->ctx_node, &ctx->tctx_list);
  		mutex_unlock(&ctx->uring_lock);
+		spin_lock_irq(&tctx->task_lock);
+		tctx->nr_ctx++;
+		spin_unlock_irq(&tctx->task_lock);
  	}
  	tctx->last = ctx;
  	return 0;
@@ -9692,6 +9696,9 @@ static __cold void io_uring_del_tctx_node(unsigned long index)
  	mutex_lock(&node->ctx->uring_lock);
  	list_del(&node->ctx_node);
  	mutex_unlock(&node->ctx->uring_lock);
+	spin_lock_irq(&tctx->task_lock);
+	tctx->nr_ctx--;
+	spin_unlock_irq(&tctx->task_lock);
 
  	if (tctx->last == node->ctx)
  		tctx->last = NULL;
author
#8 lore raw

PATCH v3 7/8 io_uring: batch completion in prior_task_list

In reply to Hao Xu

In previous patches, we have already gathered some tw with io_req_task_complete() as callback in prior_task_list, let's complete them in batch. This is better than before in cases where !locked.

---
 fs/io_uring.c | 43 ++++++++++++++++++++++++++++++++++++-------
 1 file changed, 36 insertions(+), 7 deletions(-)
fs/io_uring.c +36 -7
index 7c6d90d693b8..bf1b730df158 100644
@@ -2166,6 +2166,26 @@ static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
 	return io_put_kbuf(req, req->kbuf);
 }
 
+static void handle_prior_tw_list(struct io_wq_work_node *node)
+{
+	struct io_kiocb *req = container_of(node, struct io_kiocb, io_task_work.node);
+	struct io_ring_ctx *ctx = req->ctx;
+
+	spin_lock(&ctx->completion_lock);
+	do {
+		struct io_wq_work_node *next = node->next;
+		struct io_kiocb *req = container_of(node, struct io_kiocb,
+						    io_task_work.node);
+
+		__io_req_complete_post(req, req->result, io_put_rw_kbuf(req));
+		node = next;
+	} while (node);
+
+	io_commit_cqring(ctx);
+	spin_unlock(&ctx->completion_lock);
+	io_cqring_ev_posted(ctx);
+}
+
 static void handle_tw_list(struct io_wq_work_node *node, struct io_ring_ctx **ctx, bool *locked)
 {
 	do {
@@ -2193,25 +2213,34 @@ static void tctx_task_work(struct callback_head *cb)
 						  task_work);
 
 	while (1) {
-		struct io_wq_work_node *node;
-		struct io_wq_work_list *merged_list;
+		unsigned int nr_ctx;
+		struct io_wq_work_node *node1, *node2;
 
 		if (!tctx->prior_task_list.first &&
 		    !tctx->task_list.first && locked)
 			io_submit_flush_completions(ctx);
 
 		spin_lock_irq(&tctx->task_lock);
-		merged_list = wq_list_merge(&tctx->prior_task_list, &tctx->task_list);
-		node = merged_list->first;
+		node1 = tctx->prior_task_list.first;
+		node2 = tctx->task_list.first;
 		INIT_WQ_LIST(&tctx->task_list);
 		INIT_WQ_LIST(&tctx->prior_task_list);
-		if (!node)
+		nr_ctx = tctx->nr_ctx;
+		if (!node1 && !node2)
 			tctx->task_running = false;
 		spin_unlock_irq(&tctx->task_lock);
-		if (!node)
+		if (!node1 && !node2)
 			break;
 
-		handle_tw_list(node, &ctx, &locked);
+		if (node1) {
+			if (nr_ctx == 1)
+				handle_prior_tw_list(node1);
+			else
+				handle_tw_list(node1, &ctx, &locked);
+		}
+
+		if (node2)
+			handle_tw_list(node2, &ctx, &locked);
 		cond_resched();
 	}
 
Signature · 1 line
2.24.4
author
#9 lore raw

PATCH v3 8/8 io_uring: add limited number of TWs to priority task list

In reply to Hao Xu

One thing to watch out is sometimes irq completion TWs comes overwhelmingly, which makes the new tw list grows fast, and TWs in the old list are starved. So we have to limit the length of the new tw list. A practical value is 1/3:
len of new tw list < 1/3 * (len of new + old tw list)

In this way, the new tw list has a limited length and normal task get there chance to run.
Say MAX_PRIORITY_TW_RATIO is k, the number of TWs in priority list is x, in non-priority list in is y. Then a TW can be inserted to the priority list in the condition:

            x <= 1/k * (x + y)
          =>k * x <= x + y
          =>(1 - k) * x + y >= 0

So we just need a variable z = (1 - k) * x + y. Everytime a new TW comes,
if z >= 0, we add it to prio list, and z += (1 - k) if z < 0, we add it to non-prio list, and z++

So we just one extra operation, and we can simplify the check to:

       if (priority && k >= 0) add to prio list;
---
 fs/io_uring.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
fs/io_uring.c +8 -2
index bf1b730df158..0099decac71d 100644
@@ -471,6 +471,7 @@ struct io_uring_task {
 	struct callback_head	task_work;
 	bool			task_running;
 	unsigned int		nr_ctx;
+	int			factor;
 };
 
 /*
@@ -2225,6 +2226,7 @@ static void tctx_task_work(struct callback_head *cb)
 		node2 = tctx->task_list.first;
 		INIT_WQ_LIST(&tctx->task_list);
 		INIT_WQ_LIST(&tctx->prior_task_list);
+		tctx->factor = 0;
 		nr_ctx = tctx->nr_ctx;
 		if (!node1 && !node2)
 			tctx->task_running = false;
@@ -2247,6 +2249,7 @@ static void tctx_task_work(struct callback_head *cb)
 	ctx_flush_and_put(ctx, &locked);
 }
 
+#define MAX_PRIORITY_TW_RATIO 3
 static void io_req_task_work_add(struct io_kiocb *req, bool priority)
 {
 	struct task_struct *tsk = req->task;
@@ -2260,10 +2263,13 @@ static void io_req_task_work_add(struct io_kiocb *req, bool priority)
 	WARN_ON_ONCE(!tctx);
 
 	spin_lock_irqsave(&tctx->task_lock, flags);
-	if (priority)
+	if (priority && tctx->factor >= 0) {
 		wq_list_add_tail(&req->io_task_work.node, &tctx->prior_task_list);
-	else
+		tctx->factor += (1 - MAX_PRIORITY_TW_RATIO);
+	} else {
 		wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
+		tctx->factor++;
+	}
 	running = tctx->task_running;
 	if (!running)
 		tctx->task_running = true;
Signature · 1 line
2.24.4
author
#10 lore raw
In reply to Hao Xu
Quoted 18 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:

Quoted 17 lines
    depth     no 8/8   include 8/8      before this patchset
     1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42
     8        13.78        13.13             13.22
     16       27.41        27.92             24.33
     32       49.40        46.16             53.08
     64       102.53       105.68            103.36
     128      196.98       202.76            205.61
     256      372.99       375.61            414.88
     512      747.23       763.95            791.30
     1024     1472.59      1527.46           1538.72
     2048     3153.49      3129.22           3329.01
     4096     6387.86      5899.74           6682.54
     8192     12150.25     12433.59          12774.14
     16384    23085.58     24342.84          26044.71

The data in the previous v2 version seems not right, I may miss something at that time.

Quoted 17 lines

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
io-wq: add helper to merge two wq_lists
io_uring: add a priority tw list for irq completion work
io_uring: add helper for task work execution code
io_uring: split io_req_complete_post() and add a helper
io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
io_uring: add nr_ctx to record the number of ctx in a task
io_uring: batch completion in prior_task_list
io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------
  2 files changed, 138 insertions(+), 51 deletions(-)
#11 lore raw
In reply to Hao Xu
Quoted 3 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

If submissions and completions are done by the same task it doesn't really matter in which order they're executed because the task won't get back to userspace execution to see CQEs until tw returns. Furthermore, it even might be worse because the earlier you submit the better with everything else equal.

IIRC, that's how it's with fio, right? If so, you may get better numbers with a test that does submissions and completions in different threads.

Also interesting to find an explanation for you numbers assuming they're stable. 7/8 batching? How often it does it go this path? If only one task submits requests it should already be covered with existing batching.

Quoted 48 lines

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:

    depth     no 8/8   include 8/8      before this patchset
     1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42
     8        13.78        13.13             13.22
     16       27.41        27.92             24.33
     32       49.40        46.16             53.08
     64       102.53       105.68            103.36
     128      196.98       202.76            205.61
     256      372.99       375.61            414.88
     512      747.23       763.95            791.30
     1024     1472.59      1527.46           1538.72
     2048     3153.49      3129.22           3329.01
     4096     6387.86      5899.74           6682.54
     8192     12150.25     12433.59          12774.14
     16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
io-wq: add helper to merge two wq_lists
io_uring: add a priority tw list for irq completion work
io_uring: add helper for task work execution code
io_uring: split io_req_complete_post() and add a helper
io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
io_uring: add nr_ctx to record the number of ctx in a task
io_uring: batch completion in prior_task_list
io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------
  2 files changed, 138 insertions(+), 51 deletions(-)
Signature · 1 line
Pavel Begunkov
author
#12 lore raw
Quoted 8 lines · Pavel Begunkov
Quoted 3 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

If submissions and completions are done by the same task it doesn't really matter in which order they're executed because the task won't get back to userspace execution to see CQEs until tw returns.

It may matter, it depends on the time cost of submittion and the DMA IO time. Pick up sqpoll mode as example, we submit 10 reqs:
t1 io_submit_sqes

            -->io_req_task_queue

t2 io_task_work_run
we actually do the submittion in t2, but if the workload is big engough, the 'irq completion TW' will be inserted to the TW list after t2 is fully done, then those 'irq completion TW' will be delayed to the next round. With this patchset, we can handle them first.

Quoted 6 lines

Furthermore, it even might be worse because the earlier you submit the better with everything else equal.

IIRC, that's how it's with fio, right? If so, you may get better numbers with a test that does submissions and completions in different threads.

Because of the completion cache, I doubt if it works. For single ctx, it seems we always update the cqring pointer after all the TWs in the list are done.

Quoted 2 lines

Also interesting to find an explanation for you numbers assuming

The reason may be what I said above, but I don't have a strict proof now.

Quoted 3 lines

they're stable. 7/8 batching? How often it does it go this path? If only one task submits requests it should already be covered with existing batching.

the problem of the existing batch is(given there is only one ctx):
1. we flush it after all the TWs done
2. we batch them if we have uring lock.
the new batch is:
1. don't care about uring lock
2. we can flush the completions in the priority list in advance.(which means userland can see it earlier.)

Quoted 51 lines
Quoted 50 lines

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:
    depth     no 8/8   include 8/8      before this patchset 1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42 8        13.78        13.13             13.22 16       27.41        27.92             24.33 32       49.40        46.16             53.08 64       102.53       105.68            103.36 128      196.98       202.76            205.61 256      372.99       375.61            414.88 512      747.23       763.95            791.30 1024     1472.59      1527.46           1538.72 2048     3153.49      3129.22           3329.01 4096     6387.86      5899.74           6682.54 8192     12150.25     12433.59          12774.14 16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
   io-wq: add helper to merge two wq_lists
   io_uring: add a priority tw list for irq completion work
   io_uring: add helper for task work execution code
   io_uring: split io_req_complete_post() and add a helper
   io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
   io_uring: add nr_ctx to record the number of ctx in a task
   io_uring: batch completion in prior_task_list
   io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------

  2 files changed, 138 insertions(+), 51 deletions(-)
author
#13 lore raw
Quoted 23 lines · Hao Xu
Quoted 8 lines · Pavel Begunkov
Quoted 3 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

If submissions and completions are done by the same task it doesn't really matter in which order they're executed because the task won't get back to userspace execution to see CQEs until tw returns.

It may matter, it depends on the time cost of submittion and the DMA IO time. Pick up sqpoll mode as example, we submit 10 reqs:
t1          io_submit_sqes
            -->io_req_task_queue
t2          io_task_work_run
we actually do the submittion in t2,  but if the workload is big engough, the 'irq completion TW' will be inserted to the TW list after t2 is fully done, then those 'irq completion TW' will be delayed to the next round. With this patchset, we can handle them first.

Quoted 3 lines

Furthermore, it even might be worse because the earlier you submit the better with everything else equal.

Quoted 6 lines
Quoted 3 lines

IIRC, that's how it's with fio, right? If so, you may get better numbers with a test that does submissions and completions in different threads.

Because of the completion cache, I doubt if it works. For single ctx, it seems we always update the cqring pointer after all the TWs in the list are done.

I suddenly realized sqpoll mode does submissions and completions in different threads, and in this situation this patchset always first commit_cqring() after handling TWs in priority list. So this is the right test, do I miss something?

Quoted 66 lines
Quoted 2 lines

Also interesting to find an explanation for you numbers assuming

The reason may be what I said above, but I don't have a strict proof now.

Quoted 3 lines

they're stable. 7/8 batching? How often it does it go this path? If only one task submits requests it should already be covered with existing batching.

the problem of the existing batch is(given there is only one ctx):
1. we flush it after all the TWs done
2. we batch them if we have uring lock.
the new batch is:
1. don't care about uring lock
2. we can flush the completions in the priority list in advance.(which means userland can see it earlier.)

Quoted 51 lines
Quoted 50 lines

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:
    depth     no 8/8   include 8/8      before this patchset 1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42 8        13.78        13.13             13.22 16       27.41        27.92             24.33 32       49.40        46.16             53.08 64       102.53       105.68            103.36 128      196.98       202.76            205.61 256      372.99       375.61            414.88 512      747.23       763.95            791.30 1024     1472.59      1527.46           1538.72 2048     3153.49      3129.22           3329.01 4096     6387.86      5899.74           6682.54 8192     12150.25     12433.59          12774.14 16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
   io-wq: add helper to merge two wq_lists
   io_uring: add a priority tw list for irq completion work
   io_uring: add helper for task work execution code
   io_uring: split io_req_complete_post() and add a helper
   io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
   io_uring: add nr_ctx to record the number of ctx in a task
   io_uring: batch completion in prior_task_list
   io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------

  2 files changed, 138 insertions(+), 51 deletions(-)
#14 lore raw
Quoted 27 lines · Hao Xu
Quoted 26 lines · Hao Xu
Quoted 8 lines · Pavel Begunkov
Quoted 3 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

If submissions and completions are done by the same task it doesn't really matter in which order they're executed because the task won't get back to userspace execution to see CQEs until tw returns.

It may matter, it depends on the time cost of submittion and the DMA IO time. Pick up sqpoll mode as example, we submit 10 reqs:
t1          io_submit_sqes
             -->io_req_task_queue
t2          io_task_work_run
we actually do the submittion in t2,  but if the workload is big engough, the 'irq completion TW' will be inserted to the TW list after t2 is fully done, then those 'irq completion TW' will be delayed to the next round. With this patchset, we can handle them first.

Quoted 6 lines

Furthermore, it even might be worse because the earlier you submit the better with everything else equal.

IIRC, that's how it's with fio, right? If so, you may get better numbers with a test that does submissions and completions in different threads.

Quoted 7 lines
Quoted 3 lines

Because of the completion cache, I doubt if it works. For single ctx, it seems we always update the cqring pointer after all the TWs in the list are done.

I suddenly realized sqpoll mode does submissions and completions in different threads, and in this situation this patchset always first commit_cqring() after handling TWs in priority list. So this is the right test, do I miss something?

Yep, should be it. So the scope of the feature is SQPOLL or completion/submission with different tasks.

Quoted 67 lines
Quoted 66 lines
Quoted 2 lines

Also interesting to find an explanation for you numbers assuming

The reason may be what I said above, but I don't have a strict proof now.

Quoted 3 lines

they're stable. 7/8 batching? How often it does it go this path? If only one task submits requests it should already be covered with existing batching.

the problem of the existing batch is(given there is only one ctx):
1. we flush it after all the TWs done
2. we batch them if we have uring lock.
the new batch is:
1. don't care about uring lock
2. we can flush the completions in the priority list in advance.(which means userland can see it earlier.)

Quoted 51 lines
Quoted 50 lines

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:
    depth     no 8/8   include 8/8      before this patchset 1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42 8        13.78        13.13             13.22 16       27.41        27.92             24.33 32       49.40        46.16             53.08 64       102.53       105.68            103.36 128      196.98       202.76            205.61 256      372.99       375.61            414.88 512      747.23       763.95            791.30 1024     1472.59      1527.46           1538.72 2048     3153.49      3129.22           3329.01 4096     6387.86      5899.74           6682.54 8192     12150.25     12433.59          12774.14 16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
   io-wq: add helper to merge two wq_lists
   io_uring: add a priority tw list for irq completion work
   io_uring: add helper for task work execution code
   io_uring: split io_req_complete_post() and add a helper
   io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
   io_uring: add nr_ctx to record the number of ctx in a task
   io_uring: batch completion in prior_task_list
   io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168 +++++++++++++++++++++++++++++++++++---------------

  2 files changed, 138 insertions(+), 51 deletions(-)
Signature · 1 line
Pavel Begunkov
author
#15 lore raw
Quoted 31 lines · Pavel Begunkov
Quoted 30 lines · Hao Xu
Quoted 29 lines · Hao Xu
Quoted 8 lines · Pavel Begunkov
Quoted 3 lines · Hao Xu

Tested this patchset by manually replace __io_queue_sqe() in io_queue_sqe() by io_req_task_queue() to construct 'heavy' task works. Then test with fio:

If submissions and completions are done by the same task it doesn't really matter in which order they're executed because the task won't get back to userspace execution to see CQEs until tw returns.

It may matter, it depends on the time cost of submittion and the DMA IO time. Pick up sqpoll mode as example, we submit 10 reqs:
t1          io_submit_sqes
             -->io_req_task_queue
t2          io_task_work_run
we actually do the submittion in t2,  but if the workload is big engough, the 'irq completion TW' will be inserted to the TW list after t2 is fully done, then those 'irq completion TW' will be delayed to the next round. With this patchset, we can handle them first.

Quoted 6 lines

Furthermore, it even might be worse because the earlier you submit the better with everything else equal.

IIRC, that's how it's with fio, right? If so, you may get better numbers with a test that does submissions and completions in different threads.

Because of the completion cache, I doubt if it works. For single ctx, it seems we always update the cqring pointer after all the TWs in the list are done.

Quoted 7 lines
Quoted 4 lines

I suddenly realized sqpoll mode does submissions and completions in different threads, and in this situation this patchset always first commit_cqring() after handling TWs in priority list. So this is the right test, do I miss something?

Yep, should be it. So the scope of the feature is SQPOLL or completion/submission with different tasks.

From the test results, it's a bit risk to apply this feature to normal mode(no good, but have to ensure no bad), so I'd like to apply it to sqpoll mode for now. For completion/submission decoupled situation, maybe we can include it later.

Quoted 70 lines
Quoted 69 lines
Quoted 68 lines
Quoted 3 lines

Also interesting to find an explanation for you numbers assuming

The reason may be what I said above, but I don't have a strict proof now.

Quoted 3 lines

they're stable. 7/8 batching? How often it does it go this path? If only one task submits requests it should already be covered with existing batching.

the problem of the existing batch is(given there is only one ctx):
1. we flush it after all the TWs done
2. we batch them if we have uring lock.
the new batch is:
1. don't care about uring lock
2. we can flush the completions in the priority list in advance.(which means userland can see it earlier.)

Quoted 52 lines
Quoted 51 lines

ioengine=io_uring
sqpoll=1
thread=1
bs=4k
direct=1
rw=randread
time_based=1
runtime=600
randrepeat=0
group_reporting=1
filename=/dev/nvme0n1

2/8 set unlimited priority_task_list, 8/8 set a limitation of 1/3 * (len_prority_list + len_normal_list), data below:
    depth     no 8/8   include 8/8      before this patchset 1        7.05         7.82              7.10
     2        8.47         8.48              8.60
     4        10.42        9.99              10.42 8        13.78        13.13             13.22 16       27.41        27.92             24.33 32       49.40        46.16             53.08 64       102.53       105.68            103.36 128      196.98       202.76            205.61 256      372.99       375.61            414.88 512      747.23       763.95            791.30 1024     1472.59      1527.46           1538.72 2048     3153.49      3129.22           3329.01 4096     6387.86      5899.74           6682.54 8192     12150.25     12433.59          12774.14 16384    23085.58     24342.84          26044.71

It seems 2/8 is better, haven't tried other choices other than 1/3, still put 8/8 here for people's further thoughts.

Hao Xu (8):
   io-wq: add helper to merge two wq_lists
   io_uring: add a priority tw list for irq completion work
   io_uring: add helper for task work execution code
   io_uring: split io_req_complete_post() and add a helper
   io_uring: move up io_put_kbuf() and io_put_rw_kbuf()
   io_uring: add nr_ctx to record the number of ctx in a task
   io_uring: batch completion in prior_task_list
   io_uring: add limited number of TWs to priority task list

  fs/io-wq.h    |  21 +++++++
  fs/io_uring.c | 168
+++++++++++++++++++++++++++++++++++---------------

  2 files changed, 138 insertions(+), 51 deletions(-)