All of lore.kernel.org
 help / color / mirror / Atom feed
From: Jens Axboe <axboe@kernel.dk>
To: linux-block@vger.kernel.org, osandov@osandov.com
Cc: Jens Axboe <axboe@kernel.dk>
Subject: [PATCH 1/2] sbitmap: ammortize cost of clearing bits
Date: Thu, 29 Nov 2018 19:09:07 -0700	[thread overview]
Message-ID: <20181130020908.7325-2-axboe@kernel.dk> (raw)
In-Reply-To: <20181130020908.7325-1-axboe@kernel.dk>

sbitmap maintains a set of words that we use to set and clear bits, with
each bit representing a tag for blk-mq. Even though we spread the bits
out and maintain a hint cache, one particular bit allocated will end up
being cleared in the exact same spot.

This introduces batched clearing of bits. Instead of clearing a given
bit, the same bit is set in a cleared/free mask instead. If we fail
allocating a bit from a given word, then we check the free mask, and
batch move those cleared bits at that time. This trades 64 atomic bitops
for 2 cmpxchg().

In a threaded poll test case, half the overhead of getting and clearing
tags is removed with this change. On another poll test case with a
single thread, performance is unchanged.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/sbitmap.h | 26 +++++++++++++++---
 lib/sbitmap.c           | 60 ++++++++++++++++++++++++++++++++++++++---
 2 files changed, 78 insertions(+), 8 deletions(-)

diff --git a/include/linux/sbitmap.h b/include/linux/sbitmap.h
index 804a50983ec5..cec685b89998 100644
--- a/include/linux/sbitmap.h
+++ b/include/linux/sbitmap.h
@@ -30,14 +30,19 @@ struct seq_file;
  */
 struct sbitmap_word {
 	/**
-	 * @word: The bitmap word itself.
+	 * @depth: Number of bits being used in @word/@cleared
 	 */
-	unsigned long word;
+	unsigned long depth;
 
 	/**
-	 * @depth: Number of bits being used in @word.
+	 * @word: word holding free bits
 	 */
-	unsigned long depth;
+	unsigned long word ____cacheline_aligned_in_smp;
+
+	/**
+	 * @cleared: word holding cleared bits
+	 */
+	unsigned long cleared ____cacheline_aligned_in_smp;
 } ____cacheline_aligned_in_smp;
 
 /**
@@ -310,6 +315,19 @@ static inline void sbitmap_clear_bit(struct sbitmap *sb, unsigned int bitnr)
 	clear_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr));
 }
 
+/*
+ * This one is special, since it doesn't actually clear the bit, rather it
+ * sets the corresponding bit in the ->cleared mask instead. Paired with
+ * the caller doing sbitmap_batch_clear() if a given index is full, which
+ * will clear the previously freed entries in the corresponding ->word.
+ */
+static inline void sbitmap_deferred_clear_bit(struct sbitmap *sb, unsigned int bitnr)
+{
+	unsigned long *addr = &sb->map[SB_NR_TO_INDEX(sb, bitnr)].cleared;
+
+	set_bit(SB_NR_TO_BIT(sb, bitnr), addr);
+}
+
 static inline void sbitmap_clear_bit_unlock(struct sbitmap *sb,
 					    unsigned int bitnr)
 {
diff --git a/lib/sbitmap.c b/lib/sbitmap.c
index 45cab6bbc1c7..2316f53f3e1d 100644
--- a/lib/sbitmap.c
+++ b/lib/sbitmap.c
@@ -111,6 +111,58 @@ static int __sbitmap_get_word(unsigned long *word, unsigned long depth,
 	return nr;
 }
 
+/*
+ * See if we have deferred clears that we can batch move
+ */
+static inline bool sbitmap_deferred_clear(struct sbitmap *sb, int index)
+{
+	unsigned long mask, val;
+
+	if (!sb->map[index].cleared)
+		return false;
+
+	/*
+	 * First get a stable cleared mask, setting the old mask to 0.
+	 */
+	do {
+		mask = sb->map[index].cleared;
+	} while (cmpxchg(&sb->map[index].cleared, mask, 0) != mask);
+
+	/*
+	 * Now clear the masked bits in our free word
+	 */
+	do {
+		val = sb->map[index].word;
+	} while (cmpxchg(&sb->map[index].word, val, val & ~mask) != val);
+
+	/*
+	 * If someone found ->cleared == 0 before we wrote ->word, then
+	 * they could have failed when they should not have. Check for
+	 * waiters.
+	 */
+	smp_mb__after_atomic();
+	sbitmap_queue_wake_up(container_of(sb, struct sbitmap_queue, sb));
+	return true;
+}
+
+static int sbitmap_find_bit_in_index(struct sbitmap *sb, int index,
+				     unsigned int alloc_hint, bool round_robin)
+{
+	int nr;
+
+	do {
+		nr = __sbitmap_get_word(&sb->map[index].word,
+					sb->map[index].depth, alloc_hint,
+					!round_robin);
+		if (nr != -1)
+			break;
+		if (!sbitmap_deferred_clear(sb, index))
+			break;
+	} while (1);
+
+	return nr;
+}
+
 int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin)
 {
 	unsigned int i, index;
@@ -129,9 +181,8 @@ int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin)
 		alloc_hint = 0;
 
 	for (i = 0; i < sb->map_nr; i++) {
-		nr = __sbitmap_get_word(&sb->map[index].word,
-					sb->map[index].depth, alloc_hint,
-					!round_robin);
+		nr = sbitmap_find_bit_in_index(sb, index, alloc_hint,
+						round_robin);
 		if (nr != -1) {
 			nr += index << sb->shift;
 			break;
@@ -514,7 +565,8 @@ EXPORT_SYMBOL_GPL(sbitmap_queue_wake_up);
 void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr,
 			 unsigned int cpu)
 {
-	sbitmap_clear_bit_unlock(&sbq->sb, nr);
+	sbitmap_deferred_clear_bit(&sbq->sb, nr);
+
 	/*
 	 * Pairs with the memory barrier in set_current_state() to ensure the
 	 * proper ordering of clear_bit_unlock()/waitqueue_active() in the waker
-- 
2.17.1


  reply	other threads:[~2018-11-30  2:09 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-11-30  2:09 [PATCHSET v3] sbitmap optimizations Jens Axboe
2018-11-30  2:09 ` Jens Axboe [this message]
2018-11-30  2:09 ` [PATCH 2/2] sbitmap: optimize wakeup check Jens Axboe
2018-11-30 16:01 [PATCHSET v4] sbitmap optimizations Jens Axboe
2018-11-30 16:01 ` [PATCH 1/2] sbitmap: ammortize cost of clearing bits Jens Axboe
2018-11-30 20:03   ` Omar Sandoval
2018-11-30 20:10     ` Jens Axboe
2018-11-30 21:41       ` Omar Sandoval

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=20181130020908.7325-2-axboe@kernel.dk \
    --to=axboe@kernel.dk \
    --cc=linux-block@vger.kernel.org \
    --cc=osandov@osandov.com \
    /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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.