md: incorporate new plugging into raid5.
authorNeilBrown <neilb@suse.de>
Mon, 18 Apr 2011 08:25:43 +0000 (18:25 +1000)
committerNeilBrown <neilb@suse.de>
Mon, 18 Apr 2011 08:25:43 +0000 (18:25 +1000)
In raid5 plugging is used for 2 things:
 1/ collecting writes that require a bitmap update
 2/ collecting writes in the hope that we can create full
    stripes - or at least more-full.

We now release these different sets of stripes when plug_cnt
is zero.

Also in make_request, we call mddev_check_plug to hopefully increase
plug_cnt, and wake up the thread at the end if plugging wasn't
achieved for some reason.

Signed-off-by: NeilBrown <neilb@suse.de>
drivers/md/raid5.c

index a1755a6..f301e6a 100644 (file)
  *
  * We group bitmap updates into batches.  Each batch has a number.
  * We may write out several batches at once, but that isn't very important.
- * conf->bm_write is the number of the last batch successfully written.
- * conf->bm_flush is the number of the last batch that was closed to
+ * conf->seq_write is the number of the last batch successfully written.
+ * conf->seq_flush is the number of the last batch that was closed to
  *    new additions.
  * When we discover that we will need to write to any block in a stripe
  * (in add_stripe_bio) we update the in-memory bitmap and record in sh->bm_seq
- * the number of the batch it will be in. This is bm_flush+1.
+ * the number of the batch it will be in. This is seq_flush+1.
  * When we are ready to do a write, if that batch hasn't been written yet,
  *   we plug the array and queue the stripe for later.
  * When an unplug happens, we increment bm_flush, thus closing the current
@@ -459,7 +459,7 @@ get_active_stripe(raid5_conf_t *conf, sector_t sector,
                                                     < (conf->max_nr_stripes *3/4)
                                                     || !conf->inactive_blocked),
                                                    conf->device_lock,
-                                                   md_wakeup_thread(conf->mddev->thread));
+                                                   );
                                conf->inactive_blocked = 0;
                        } else
                                init_stripe(sh, sector, previous);
@@ -3927,6 +3927,7 @@ static int make_request(mddev_t *mddev, struct bio * bi)
        struct stripe_head *sh;
        const int rw = bio_data_dir(bi);
        int remaining;
+       int plugged;
 
        if (unlikely(bi->bi_rw & REQ_FLUSH)) {
                md_flush_request(mddev, bi);
@@ -3945,6 +3946,7 @@ static int make_request(mddev_t *mddev, struct bio * bi)
        bi->bi_next = NULL;
        bi->bi_phys_segments = 1;       /* over-loaded to count active stripes */
 
+       plugged = mddev_check_plugged(mddev);
        for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) {
                DEFINE_WAIT(w);
                int disks, data_disks;
@@ -4059,6 +4061,9 @@ static int make_request(mddev_t *mddev, struct bio * bi)
                }
                        
        }
+       if (!plugged)
+               md_wakeup_thread(mddev->thread);
+
        spin_lock_irq(&conf->device_lock);
        remaining = raid5_dec_bi_phys_segments(bi);
        spin_unlock_irq(&conf->device_lock);
@@ -4472,14 +4477,18 @@ static void raid5d(mddev_t *mddev)
        while (1) {
                struct bio *bio;
 
-               if (conf->seq_flush != conf->seq_write) {
-                       int seq = conf->seq_flush;
+               if (atomic_read(&mddev->plug_cnt) == 0 &&
+                   !list_empty(&conf->bitmap_list)) {
+                       /* Now is a good time to flush some bitmap updates */
+                       conf->seq_flush++;
                        spin_unlock_irq(&conf->device_lock);
                        bitmap_unplug(mddev->bitmap);
                        spin_lock_irq(&conf->device_lock);
-                       conf->seq_write = seq;
+                       conf->seq_write = conf->seq_flush;
                        activate_bit_delay(conf);
                }
+               if (atomic_read(&mddev->plug_cnt) == 0)
+                       raid5_activate_delayed(conf);
 
                while ((bio = remove_bio_from_retry(conf))) {
                        int ok;