linux-raid.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Shaohua Li <shli@kernel.org>
To: linux-raid@vger.kernel.org
Cc: neilb@suse.de, axboe@kernel.dk, dan.j.williams@intel.com,
	shli@fusionio.com
Subject: [patch 08/10 v3] raid5: make_request use batch stripe release
Date: Mon, 25 Jun 2012 15:24:55 +0800	[thread overview]
Message-ID: <20120625072702.921605418@kernel.org> (raw)
In-Reply-To: 20120625072447.268095276@kernel.org

[-- Attachment #1: raid5-make_request-relase_stripe-batch.patch --]
[-- Type: text/plain, Size: 4473 bytes --]

make_request() does stripe release for every stripe and the stripe usually has
count 1, which makes previous release_stripe() optimization not work. In my
test, this release_stripe() becomes the heaviest pleace to take
conf->device_lock after previous patches applied.

Below patch makes stripe release batch. All the stripes will be released in
unplug. The STRIPE_ON_UNPLUG_LIST bit is to protect concurrent access stripe
lru.

Signed-off-by: Shaohua Li <shli@fusionio.com>
---
 drivers/md/raid5.c |   64 ++++++++++++++++++++++++++++++++++++++++++++++++-----
 drivers/md/raid5.h |    1 
 2 files changed, 60 insertions(+), 5 deletions(-)

Index: linux/drivers/md/raid5.c
===================================================================
--- linux.orig/drivers/md/raid5.c	2012-06-25 14:38:33.110889008 +0800
+++ linux/drivers/md/raid5.c	2012-06-25 14:38:37.378835415 +0800
@@ -484,7 +484,8 @@ get_active_stripe(struct r5conf *conf, s
 		} else {
 			if (atomic_read(&sh->count)) {
 				BUG_ON(!list_empty(&sh->lru)
-				    && !test_bit(STRIPE_EXPANDING, &sh->state));
+				    && !test_bit(STRIPE_EXPANDING, &sh->state)
+				    && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state));
 			} else {
 				if (!test_bit(STRIPE_HANDLE, &sh->state))
 					atomic_inc(&conf->active_stripes);
@@ -3984,6 +3985,51 @@ static struct stripe_head *__get_priorit
 	return sh;
 }
 
+#define raid5_unplug_list(mdcb) (struct list_head *)(mdcb + 1)
+static void raid5_unplug(struct md_plug_cb *mdcb)
+{
+	struct list_head *list = raid5_unplug_list(mdcb);
+	struct stripe_head *sh;
+	struct r5conf *conf = mdcb->mddev->private;
+
+	if (list->next == NULL || list_empty(list))
+		return;
+	spin_lock_irq(&conf->device_lock);
+	while (!list_empty(list)) {
+		sh = list_entry(list->next, struct stripe_head, lru);
+		list_del_init(&sh->lru);
+		/*
+		 * avoid race release_stripe_plug() sees STRIPE_ON_UNPLUG_LIST
+		 * clear but the stripe is still in our list
+		 */
+		smp_mb__before_clear_bit();
+		clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state);
+		__release_stripe(conf, sh);
+	}
+	spin_unlock_irq(&conf->device_lock);
+}
+
+static void release_stripe_plug(struct md_plug_cb *mdcb,
+				struct stripe_head *sh)
+{
+	struct list_head *list = raid5_unplug_list(mdcb);
+
+	if (!mdcb) {
+		release_stripe(sh);
+		return;
+	}
+
+	if (list->next == NULL) {
+		INIT_LIST_HEAD(list);
+		mdcb->unplug = raid5_unplug;
+	}
+
+	if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
+		list_add_tail(&sh->lru, list);
+	else
+		release_stripe(sh);
+}
+
 static void make_request(struct mddev *mddev, struct bio * bi)
 {
 	struct r5conf *conf = mddev->private;
@@ -3993,7 +4039,7 @@ static void make_request(struct mddev *m
 	struct stripe_head *sh;
 	const int rw = bio_data_dir(bi);
 	int remaining;
-	int plugged;
+	struct md_plug_cb *mdcb;
 
 	if (unlikely(bi->bi_rw & REQ_FLUSH)) {
 		md_flush_request(mddev, bi);
@@ -4012,7 +4058,8 @@ static void make_request(struct mddev *m
 	bi->bi_next = NULL;
 	bi->bi_phys_segments = 1;	/* over-loaded to count active stripes */
 
-	plugged = !!mddev_check_plugged(mddev, NULL, 0);
+	mdcb = mddev_check_plugged(mddev, raid5_unplug,
+				   sizeof(struct list_head));
 	for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
 		DEFINE_WAIT(w);
 		int previous;
@@ -4114,7 +4161,14 @@ static void make_request(struct mddev *m
 			if ((bi->bi_rw & REQ_SYNC) &&
 			    !test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state))
 				atomic_inc(&conf->preread_active_stripes);
-			release_stripe(sh);
+			/*
+			 * We must recheck here. schedule() might be called
+			 * above which makes unplug invoked already, so the old
+			 * mdcb is invalid
+			 */
+			mdcb = mddev_check_plugged(mddev, raid5_unplug,
+						   sizeof(struct list_head));
+			release_stripe_plug(mdcb, sh);
 		} else {
 			/* cannot get stripe for read-ahead, just give-up */
 			clear_bit(BIO_UPTODATE, &bi->bi_flags);
@@ -4123,7 +4177,7 @@ static void make_request(struct mddev *m
 		}
 			
 	}
-	if (!plugged)
+	if (!mdcb)
 		md_wakeup_thread(mddev->thread);
 
 	remaining = raid5_dec_bi_active_stripes(bi);
Index: linux/drivers/md/raid5.h
===================================================================
--- linux.orig/drivers/md/raid5.h	2012-06-25 14:37:13.651888057 +0800
+++ linux/drivers/md/raid5.h	2012-06-25 14:38:37.382835318 +0800
@@ -320,6 +320,7 @@ enum {
 	STRIPE_BIOFILL_RUN,
 	STRIPE_COMPUTE_RUN,
 	STRIPE_OPS_REQ_PENDING,
+	STRIPE_ON_UNPLUG_LIST,
 };
 
 /*


  parent reply	other threads:[~2012-06-25  7:24 UTC|newest]

Thread overview: 32+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-06-25  7:24 [patch 00/10 v3] raid5: improve write performance for fast storage Shaohua Li
2012-06-25  7:24 ` [patch 01/10 v3] raid5: use wake_up_all for overlap waking Shaohua Li
2012-06-28  7:26   ` NeilBrown
2012-06-28  8:53     ` Shaohua Li
2012-06-25  7:24 ` [patch 02/10 v3] raid5: delayed stripe fix Shaohua Li
2012-07-02  0:46   ` NeilBrown
2012-07-02  0:49     ` Shaohua Li
2012-07-02  0:55       ` NeilBrown
2012-06-25  7:24 ` [patch 03/10 v3] raid5: add a per-stripe lock Shaohua Li
2012-07-02  0:50   ` NeilBrown
2012-07-02  3:16     ` Shaohua Li
2012-07-02  7:39       ` NeilBrown
2012-07-03  1:27         ` Shaohua Li
2012-07-03 12:16         ` majianpeng
2012-07-03 23:56           ` NeilBrown
2012-07-04  1:09             ` majianpeng
2012-06-25  7:24 ` [patch 04/10 v3] raid5: lockless access raid5 overrided bi_phys_segments Shaohua Li
2012-06-25  7:24 ` [patch 05/10 v3] raid5: remove some device_lock locking places Shaohua Li
2012-06-25  7:24 ` [patch 06/10 v3] raid5: reduce chance release_stripe() taking device_lock Shaohua Li
2012-07-02  0:57   ` NeilBrown
2012-06-25  7:24 ` [patch 07/10 v3] md: personality can provide unplug private data Shaohua Li
2012-07-02  1:06   ` NeilBrown
2012-06-25  7:24 ` Shaohua Li [this message]
2012-07-02  2:31   ` [patch 08/10 v3] raid5: make_request use batch stripe release NeilBrown
2012-07-02  2:59     ` Shaohua Li
2012-07-02  5:07       ` NeilBrown
2012-06-25  7:24 ` [patch 09/10 v3] raid5: raid5d handle stripe in batch way Shaohua Li
2012-07-02  2:32   ` NeilBrown
2012-06-25  7:24 ` [patch 10/10 v3] raid5: create multiple threads to handle stripes Shaohua Li
2012-07-02  2:39   ` NeilBrown
2012-07-02 20:03   ` Dan Williams
2012-07-03  8:04     ` Shaohua Li

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=20120625072702.921605418@kernel.org \
    --to=shli@kernel.org \
    --cc=axboe@kernel.dk \
    --cc=dan.j.williams@intel.com \
    --cc=linux-raid@vger.kernel.org \
    --cc=neilb@suse.de \
    --cc=shli@fusionio.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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).