[PATCH] md: Improve comments about locking situation in raid5 make_request
authorNeilBrown <neilb@suse.de>
Mon, 27 Mar 2006 09:18:15 +0000 (01:18 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Mon, 27 Mar 2006 16:45:02 +0000 (08:45 -0800)
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/md/raid5.c

index bb16ac2..8eded08 100644 (file)
@@ -1768,6 +1768,14 @@ static int make_request(request_queue_t *q, struct bio * bi)
                if (likely(conf->expand_progress == MaxSector))
                        disks = conf->raid_disks;
                else {
+                       /* spinlock is needed as expand_progress may be
+                        * 64bit on a 32bit platform, and so it might be
+                        * possible to see a half-updated value
+                        * Ofcourse expand_progress could change after
+                        * the lock is dropped, so once we get a reference
+                        * to the stripe that we think it is, we will have
+                        * to check again.
+                        */
                        spin_lock_irq(&conf->device_lock);
                        disks = conf->raid_disks;
                        if (logical_sector >= conf->expand_progress)
@@ -1791,7 +1799,12 @@ static int make_request(request_queue_t *q, struct bio * bi)
                if (sh) {
                        if (unlikely(conf->expand_progress != MaxSector)) {
                                /* expansion might have moved on while waiting for a
-                                * stripe, so we much do the range check again.
+                                * stripe, so we must do the range check again.
+                                * Expansion could still move past after this
+                                * test, but as we are holding a reference to
+                                * 'sh', we know that if that happens,
+                                *  STRIPE_EXPANDING will get set and the expansion
+                                * won't proceed until we finish with the stripe.
                                 */
                                int must_retry = 0;
                                spin_lock_irq(&conf->device_lock);