public inbox for [email protected]
 help / color / mirror / Atom feed
From: Pavel Begunkov <[email protected]>
To: Jens Axboe <[email protected]>,
	[email protected], [email protected]
Subject: [PATCH v2] io-wq: handle hashed writes in chains
Date: Mon, 23 Mar 2020 22:57:22 +0300	[thread overview]
Message-ID: <c0e464556675bd40ea47d61ce12e2393603ce43c.1584993396.git.asml.silence@gmail.com> (raw)

We always punt async buffered writes to an io-wq helper, as the core
kernel does not have IOCB_NOWAIT support for that. Most buffered async
writes complete very quickly, as it's just a copy operation. This means
that doing multiple locking roundtrips on the shared wqe lock for each
buffered write is wasteful. Additionally, buffered writes are hashed
work items, which means that any buffered write to a given file is
serialized.

Keep identicaly hashed work items contiguously in @wqe->work_list, and
track a tail for each hash bucket. On dequeue of a hashed item, splice
all of the same hash in one go using the tracked tail. Until the batch
is done, the caller doesn't have to synchronize with the wqe or worker
locks again.

Signed-off-by: Pavel Begunkov <[email protected]>
---

v2 (since diff):
- add wq_next_work() helper
- rename list helpers to match list.h naming
- reshuffle io_worker_handle_work() bits, but keeping
  functionally identical

 fs/io-wq.c | 68 ++++++++++++++++++++++++++++++++++++++----------------
 fs/io-wq.h | 45 +++++++++++++++++++++++++++++-------
 2 files changed, 85 insertions(+), 28 deletions(-)

diff --git a/fs/io-wq.c b/fs/io-wq.c
index b3fb61ec0870..cc5cf2209fb0 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -69,6 +69,8 @@ struct io_worker {
 #define IO_WQ_HASH_ORDER	5
 #endif
 
+#define IO_WQ_NR_HASH_BUCKETS	(1u << IO_WQ_HASH_ORDER)
+
 struct io_wqe_acct {
 	unsigned nr_workers;
 	unsigned max_workers;
@@ -98,6 +100,7 @@ struct io_wqe {
 	struct list_head all_list;
 
 	struct io_wq *wq;
+	struct io_wq_work *hash_tail[IO_WQ_NR_HASH_BUCKETS];
 };
 
 /*
@@ -384,7 +387,7 @@ static struct io_wq_work *io_get_next_work(struct io_wqe *wqe)
 	__must_hold(wqe->lock)
 {
 	struct io_wq_work_node *node, *prev;
-	struct io_wq_work *work;
+	struct io_wq_work *work, *tail;
 	unsigned int hash;
 
 	wq_list_for_each(node, prev, &wqe->work_list) {
@@ -392,7 +395,7 @@ static struct io_wq_work *io_get_next_work(struct io_wqe *wqe)
 
 		/* not hashed, can run anytime */
 		if (!io_wq_is_hashed(work)) {
-			wq_node_del(&wqe->work_list, node, prev);
+			wq_list_del(&wqe->work_list, node, prev);
 			return work;
 		}
 
@@ -400,7 +403,10 @@ static struct io_wq_work *io_get_next_work(struct io_wqe *wqe)
 		hash = io_get_work_hash(work);
 		if (!(wqe->hash_map & BIT(hash))) {
 			wqe->hash_map |= BIT(hash);
-			wq_node_del(&wqe->work_list, node, prev);
+			/* all items with this hash lie in [work, tail] */
+			tail = wqe->hash_tail[hash];
+			wqe->hash_tail[hash] = NULL;
+			wq_list_cut(&wqe->work_list, &tail->list, prev);
 			return work;
 		}
 	}
@@ -485,7 +491,7 @@ static void io_worker_handle_work(struct io_worker *worker)
 	struct io_wq *wq = wqe->wq;
 
 	do {
-		struct io_wq_work *work, *assign_work;
+		struct io_wq_work *work;
 		unsigned int hash;
 get_next:
 		/*
@@ -508,8 +514,9 @@ static void io_worker_handle_work(struct io_worker *worker)
 
 		/* handle a whole dependent link */
 		do {
-			struct io_wq_work *old_work;
+			struct io_wq_work *old_work, *next_hashed, *linked;
 
+			next_hashed = wq_next_work(work);
 			io_impersonate_work(worker, work);
 			/*
 			 * OK to set IO_WQ_WORK_CANCEL even for uncancellable
@@ -518,22 +525,23 @@ static void io_worker_handle_work(struct io_worker *worker)
 			if (test_bit(IO_WQ_BIT_CANCEL, &wq->state))
 				work->flags |= IO_WQ_WORK_CANCEL;
 
-			old_work = work;
 			hash = io_get_work_hash(work);
-			work->func(&work);
-			work = (old_work == work) ? NULL : work;
-
-			assign_work = work;
-			if (work && io_wq_is_hashed(work))
-				assign_work = NULL;
-			io_assign_current_work(worker, assign_work);
+			linked = old_work = work;
+			linked->func(&linked);
+			linked = (old_work == linked) ? NULL : linked;
+
+			work = next_hashed;
+			if (!work && linked && !io_wq_is_hashed(linked)) {
+				work = linked;
+				linked = NULL;
+			}
+			io_assign_current_work(worker, work);
 			wq->free_work(old_work);
 
-			if (work && !assign_work) {
-				io_wqe_enqueue(wqe, work);
-				work = NULL;
-			}
-			if (hash != -1U) {
+			if (linked)
+				io_wqe_enqueue(wqe, linked);
+
+			if (hash != -1U && !next_hashed) {
 				spin_lock_irq(&wqe->lock);
 				wqe->hash_map &= ~BIT_ULL(hash);
 				wqe->flags &= ~IO_WQE_FLAG_STALLED;
@@ -776,6 +784,26 @@ static void io_run_cancel(struct io_wq_work *work, struct io_wqe *wqe)
 	} while (work);
 }
 
+static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work)
+{
+	unsigned int hash;
+	struct io_wq_work *tail;
+
+	if (!io_wq_is_hashed(work)) {
+append:
+		wq_list_add_tail(&work->list, &wqe->work_list);
+		return;
+	}
+
+	hash = io_get_work_hash(work);
+	tail = wqe->hash_tail[hash];
+	wqe->hash_tail[hash] = work;
+	if (!tail)
+		goto append;
+
+	wq_list_add_after(&work->list, &tail->list, &wqe->work_list);
+}
+
 static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
 {
 	struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
@@ -795,7 +823,7 @@ static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
 
 	work_flags = work->flags;
 	spin_lock_irqsave(&wqe->lock, flags);
-	wq_list_add_tail(&work->list, &wqe->work_list);
+	io_wqe_insert_work(wqe, work);
 	wqe->flags &= ~IO_WQE_FLAG_STALLED;
 	spin_unlock_irqrestore(&wqe->lock, flags);
 
@@ -914,7 +942,7 @@ static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe,
 		work = container_of(node, struct io_wq_work, list);
 
 		if (match->fn(work, match->data)) {
-			wq_node_del(&wqe->work_list, node, prev);
+			wq_list_del(&wqe->work_list, node, prev);
 			found = true;
 			break;
 		}
diff --git a/fs/io-wq.h b/fs/io-wq.h
index d2a5684bf673..3ee7356d6be5 100644
--- a/fs/io-wq.h
+++ b/fs/io-wq.h
@@ -28,6 +28,18 @@ struct io_wq_work_list {
 	struct io_wq_work_node *last;
 };
 
+static inline void wq_list_add_after(struct io_wq_work_node *node,
+				     struct io_wq_work_node *pos,
+				     struct io_wq_work_list *list)
+{
+	struct io_wq_work_node *next = pos->next;
+
+	pos->next = node;
+	node->next = next;
+	if (!next)
+		list->last = node;
+}
+
 static inline void wq_list_add_tail(struct io_wq_work_node *node,
 				    struct io_wq_work_list *list)
 {
@@ -40,17 +52,26 @@ static inline void wq_list_add_tail(struct io_wq_work_node *node,
 	}
 }
 
-static inline void wq_node_del(struct io_wq_work_list *list,
-			       struct io_wq_work_node *node,
+static inline void wq_list_cut(struct io_wq_work_list *list,
+			       struct io_wq_work_node *last,
 			       struct io_wq_work_node *prev)
 {
-	if (node == list->first)
-		WRITE_ONCE(list->first, node->next);
-	if (node == list->last)
+	/* first in the list, if prev==NULL */
+	if (!prev)
+		WRITE_ONCE(list->first, last->next);
+	else
+		prev->next = last->next;
+
+	if (last == list->last)
 		list->last = prev;
-	if (prev)
-		prev->next = node->next;
-	node->next = NULL;
+	last->next = NULL;
+}
+
+static inline void wq_list_del(struct io_wq_work_list *list,
+			       struct io_wq_work_node *node,
+			       struct io_wq_work_node *prev)
+{
+	wq_list_cut(list, node, prev);
 }
 
 #define wq_list_for_each(pos, prv, head)			\
@@ -78,6 +99,14 @@ struct io_wq_work {
 		*(work) = (struct io_wq_work){ .func = _func };	\
 	} while (0)						\
 
+static inline struct io_wq_work *wq_next_work(struct io_wq_work *work)
+{
+	if (!work->list.next)
+		return NULL;
+
+	return container_of(work->list.next, struct io_wq_work, list);
+}
+
 typedef void (free_work_fn)(struct io_wq_work *);
 
 struct io_wq_data {
-- 
2.24.0


             reply	other threads:[~2020-03-23 19:58 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-23 19:57 Pavel Begunkov [this message]
2020-03-24  2:31 ` [PATCH v2] io-wq: handle hashed writes in chains Jens Axboe
  -- strict thread matches above, loose matches on Subject: below --
2020-03-19 18:56 Jens Axboe
2020-03-22 16:09 ` Pavel Begunkov
2020-03-22 16:24   ` Pavel Begunkov
2020-03-22 17:08     ` Jens Axboe
2020-03-22 18:54     ` Pavel Begunkov
2020-03-22 19:51       ` Jens Axboe
2020-03-22 20:05         ` Jens Axboe
2020-03-22 20:15           ` Jens Axboe
2020-03-22 20:20             ` Pavel Begunkov
2020-03-22 21:16               ` Pavel Begunkov
2020-03-22 21:31                 ` Pavel Begunkov
2020-03-22 20:25         ` Pavel Begunkov
2020-03-23  1:37           ` Jens Axboe
2020-03-23  8:38             ` Pavel Begunkov
2020-03-23 14:26               ` Jens Axboe
2020-03-22 17:08   ` Jens Axboe
2020-03-22 17:37     ` Pavel Begunkov
2020-03-22 20:56 ` Pavel Begunkov

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=c0e464556675bd40ea47d61ce12e2393603ce43c.1584993396.git.asml.silence@gmail.com \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox