dm optimize: make map_lock percpu

This patch changes map_lock to make it a percpu lock similar to io_lock.
Majority of code for managing io_lock from the previous patch is
reused.

Note: this patch triggers a false lockdep warning, the reason is that
there is no "write_lock_nested" function to anotate nested rw spinlocks.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

---
 drivers/md/dm.c |   26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

Index: linux-3.3-devel/drivers/md/dm.c
===================================================================
--- linux-3.3-devel.orig/drivers/md/dm.c	2012-04-18 05:06:54.000000000 +0200
+++ linux-3.3-devel/drivers/md/dm.c	2012-04-18 05:07:46.000000000 +0200
@@ -148,6 +148,9 @@ static void up_write_percpu_##name(struc
 
 declare_percpu_rw_lock(rw_semaphore, struct rw_semaphore, init_rwsem, down_read, up_read, down_write_nested, up_write)
 
+#define write_lock_nested(l, i)	write_lock(l)
+declare_percpu_rw_lock(rw_spinlock, rwlock_t, rwlock_init, read_lock, read_unlock, write_lock_nested, write_unlock)
+
 /*
  * Cookies are numeric values sent with CHANGE and REMOVE
  * uevents while resuming, removing or renaming the device.
@@ -242,7 +245,7 @@ EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
 struct mapped_device {
 	struct percpu_rw_semaphore io_lock;
 	struct mutex suspend_lock;
-	rwlock_t map_lock;
+	struct percpu_rw_spinlock map_lock;
 	atomic_t holders;
 	atomic_t open_count;
 
@@ -660,13 +663,14 @@ static void queue_io(struct mapped_devic
 struct dm_table *dm_get_live_table(struct mapped_device *md)
 {
 	struct dm_table *t;
+	int lock_cpu;
 
 	BUG_ON(in_interrupt());
-	read_lock(&md->map_lock);
+	lock_cpu = down_read_percpu_rw_spinlock(&md->map_lock);
 	t = md->map;
 	if (t)
 		dm_table_get(t);
-	read_unlock(&md->map_lock);
+	up_read_percpu_rw_spinlock(&md->map_lock, lock_cpu);
 
 	return t;
 }
@@ -1957,11 +1961,14 @@ static struct mapped_device *alloc_dev(i
 	if (r < 0)
 		goto bad_semaphore;
 
+	r = init_percpu_rw_spinlock(&md->map_lock);
+	if (r < 0)
+		goto bad_spinlock;
+
 	md->type = DM_TYPE_NONE;
 	mutex_init(&md->suspend_lock);
 	mutex_init(&md->type_lock);
 	spin_lock_init(&md->deferred_lock);
-	rwlock_init(&md->map_lock);
 	atomic_set(&md->holders, 1);
 	atomic_set(&md->open_count, 0);
 	atomic_set(&md->event_nr, 0);
@@ -2024,6 +2031,8 @@ bad_thread:
 bad_disk:
 	blk_cleanup_queue(md->queue);
 bad_queue:
+	free_percpu_rw_spinlock(&md->map_lock);
+bad_spinlock:
 	free_percpu_rw_semaphore(&md->io_lock);
 bad_semaphore:
 	free_minor(minor);
@@ -2051,6 +2060,7 @@ static void free_dev(struct mapped_devic
 		bioset_free(md->bs);
 	blk_integrity_unregister(md->disk);
 	del_gendisk(md->disk);
+	free_percpu_rw_spinlock(&md->map_lock);
 	free_percpu_rw_semaphore(&md->io_lock);
 	free_minor(minor);
 
@@ -2206,7 +2216,7 @@ static struct dm_table *__bind(struct ma
 
 	merge_is_optional = dm_table_merge_is_optional(t);
 
-	write_lock(&md->map_lock);
+	down_write_percpu_rw_spinlock(&md->map_lock);
 	old_map = md->map;
 	md->map = t;
 	md->immutable_target_type = dm_table_get_immutable_target_type(t);
@@ -2216,7 +2226,7 @@ static struct dm_table *__bind(struct ma
 		set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
 	else
 		clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
-	write_unlock(&md->map_lock);
+	up_write_percpu_rw_spinlock(&md->map_lock);
 
 	return old_map;
 }
@@ -2232,9 +2242,9 @@ static struct dm_table *__unbind(struct
 		return NULL;
 
 	dm_table_event_callback(map, NULL, NULL);
-	write_lock(&md->map_lock);
+	down_write_percpu_rw_spinlock(&md->map_lock);
 	md->map = NULL;
-	write_unlock(&md->map_lock);
+	up_write_percpu_rw_spinlock(&md->map_lock);
 
 	return map;
 }