dm cache policy mq: implement writeback_work() and mq_{set,clear}_dirty()

There are now two multiqueues for in cache blocks.  A clean one and a
dirty one.

writeback_work comes from the dirty one.  Demotions come from the clean
one.

There are two benefits:
- Performance improvement, since demoting a clean block is a noop.
- The cache cleans itself when io load is light.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

Documentation/device-mapper/cache-policies.txt

@@ -30,8 +30,10 @@ multiqueue
 
 This policy is the default.
 
-The multiqueue policy has two sets of 16 queues: one set for entries
-waiting for the cache and another one for those in the cache.
+The multiqueue policy has three sets of 16 queues: one set for entries
+waiting for the cache and another two for those in the cache (a set for
+clean entries and a set for dirty entries).
+
 Cache entries in the queues are aged based on logical time. Entry into
 the cache is based on variable thresholds and queue selection is based
 on hit count on entry. The policy aims to take different cache miss

drivers/md/dm-cache-policy-mq.c

@@ -224,6 +224,7 @@ struct entry {
 	 * FIXME: pack these better
 	 */
 	bool in_cache:1;
+	bool dirty:1;
 	unsigned hit_count;
 	unsigned generation;
 	unsigned tick;
@@ -238,13 +239,15 @@ struct mq_policy {
 	struct io_tracker tracker;
 
 	/*
-	 * We maintain two queues of entries.  The cache proper contains
-	 * the currently active mappings.  Whereas the pre_cache tracks
-	 * blocks that are being hit frequently and potential candidates
-	 * for promotion to the cache.
+	 * We maintain three queues of entries.  The cache proper,
+	 * consisting of a clean and dirty queue, contains the currently
+	 * active mappings.  Whereas the pre_cache tracks blocks that
+	 * are being hit frequently and potential candidates for promotion
+	 * to the cache.
 	 */
 	struct queue pre_cache;
-	struct queue cache;
+	struct queue cache_clean;
+	struct queue cache_dirty;
 
 	/*
 	 * Keeps track of time, incremented by the core.  We use this to
@@ -324,7 +327,8 @@ static void free_entries(struct mq_policy *mq)
 	struct entry *e, *tmp;
 
 	concat_queue(&mq->free, &mq->pre_cache);
-	concat_queue(&mq->free, &mq->cache);
+	concat_queue(&mq->free, &mq->cache_clean);
+	concat_queue(&mq->free, &mq->cache_dirty);
 
 	list_for_each_entry_safe(e, tmp, &mq->free, list)
 		kmem_cache_free(mq_entry_cache, e);
@@ -508,7 +512,8 @@ static void push(struct mq_policy *mq, struct entry *e)
 
 	if (e->in_cache) {
 		alloc_cblock(mq, e->cblock);
-		queue_push(&mq->cache, queue_level(e), &e->list);
+		queue_push(e->dirty ? &mq->cache_dirty : &mq->cache_clean,
+			   queue_level(e), &e->list);
 	} else
 		queue_push(&mq->pre_cache, queue_level(e), &e->list);
 }
@@ -558,7 +563,8 @@ static bool updated_this_tick(struct mq_policy *mq, struct entry *e)
  * of the entries.
  *
  * At the moment the threshold is taken by averaging the hit counts of some
- * of the entries in the cache (the first 20 entries of the first level).
+ * of the entries in the cache (the first 20 entries across all levels in
+ * ascending order, giving preference to the clean entries at each level).
  *
  * We can be much cleverer than this though.  For example, each promotion
  * could bump up the threshold helping to prevent churn.  Much more to do
@@ -580,7 +586,16 @@ static void check_generation(struct mq_policy *mq)
 		mq->generation++;
 
 		for (level = 0; level < NR_QUEUE_LEVELS && count < MAX_TO_AVERAGE; level++) {
-			head = mq->cache.qs + level;
+			head = mq->cache_clean.qs + level;
+			list_for_each_entry(e, head, list) {
+				nr++;
+				total += e->hit_count;
+
+				if (++count >= MAX_TO_AVERAGE)
+					break;
+			}
+
+			head = mq->cache_dirty.qs + level;
 			list_for_each_entry(e, head, list) {
 				nr++;
 				total += e->hit_count;
@@ -633,19 +648,28 @@ static void requeue_and_update_tick(struct mq_policy *mq, struct entry *e)
  * - set the hit count to a hard coded value other than 1, eg, is it better
  *   if it goes in at level 2?
  */
-static dm_cblock_t demote_cblock(struct mq_policy *mq, dm_oblock_t *oblock)
+static int demote_cblock(struct mq_policy *mq, dm_oblock_t *oblock, dm_cblock_t *cblock)
 {
-	dm_cblock_t result;
-	struct entry *demoted = pop(mq, &mq->cache);
+	struct entry *demoted = pop(mq, &mq->cache_clean);
 
-	BUG_ON(!demoted);
-	result = demoted->cblock;
+	if (!demoted)
+		/*
+		 * We could get a block from mq->cache_dirty, but that
+		 * would add extra latency to the triggering bio as it
+		 * waits for the writeback.  Better to not promote this
+		 * time and hope there's a clean block next time this block
+		 * is hit.
+		 */
+		return -ENOSPC;
+
+	*cblock = demoted->cblock;
 	*oblock = demoted->oblock;
 	demoted->in_cache = false;
+	demoted->dirty = false;
 	demoted->hit_count = 1;
 	push(mq, demoted);
 
-	return result;
+	return 0;
 }
 
 /*
@@ -705,11 +729,16 @@ static int cache_entry_found(struct mq_policy *mq,
 static int pre_cache_to_cache(struct mq_policy *mq, struct entry *e,
 			      struct policy_result *result)
 {
+	int r;
 	dm_cblock_t cblock;
 
 	if (find_free_cblock(mq, &cblock) == -ENOSPC) {
 		result->op = POLICY_REPLACE;
-		cblock = demote_cblock(mq, &result->old_oblock);
+		r = demote_cblock(mq, &result->old_oblock, &cblock);
+		if (r) {
+			result->op = POLICY_MISS;
+			return 0;
+		}
 	} else
 		result->op = POLICY_NEW;
 
@@ -717,6 +746,7 @@ static int pre_cache_to_cache(struct mq_policy *mq, struct entry *e,
 
 	del(mq, e);
 	e->in_cache = true;
+	e->dirty = false;
 	push(mq, e);
 
 	return 0;
@@ -760,6 +790,7 @@ static void insert_in_pre_cache(struct mq_policy *mq,
 	}
 
 	e->in_cache = false;
+	e->dirty = false;
 	e->oblock = oblock;
 	e->hit_count = 1;
 	e->generation = mq->generation;
@@ -787,6 +818,7 @@ static void insert_in_cache(struct mq_policy *mq, dm_oblock_t oblock,
 	e->oblock = oblock;
 	e->cblock = cblock;
 	e->in_cache = true;
+	e->dirty = false;
 	e->hit_count = 1;
 	e->generation = mq->generation;
 	push(mq, e);
@@ -917,6 +949,40 @@ static int mq_lookup(struct dm_cache_policy *p, dm_oblock_t oblock, dm_cblock_t
 	return r;
 }
 
+/*
+ * FIXME: __mq_set_clear_dirty can block due to mutex.
+ * Ideally a policy should not block in functions called
+ * from the map() function.  Explore using RCU.
+ */
+static void __mq_set_clear_dirty(struct dm_cache_policy *p, dm_oblock_t oblock, bool set)
+{
+	struct mq_policy *mq = to_mq_policy(p);
+	struct entry *e;
+
+	mutex_lock(&mq->lock);
+	e = hash_lookup(mq, oblock);
+	if (!e)
+		DMWARN("__mq_set_clear_dirty called for a block that isn't in the cache");
+	else {
+		BUG_ON(!e->in_cache);
+
+		del(mq, e);
+		e->dirty = set;
+		push(mq, e);
+	}
+	mutex_unlock(&mq->lock);
+}
+
+static void mq_set_dirty(struct dm_cache_policy *p, dm_oblock_t oblock)
+{
+	__mq_set_clear_dirty(p, oblock, true);
+}
+
+static void mq_clear_dirty(struct dm_cache_policy *p, dm_oblock_t oblock)
+{
+	__mq_set_clear_dirty(p, oblock, false);
+}
+
 static int mq_load_mapping(struct dm_cache_policy *p,
 			   dm_oblock_t oblock, dm_cblock_t cblock,
 			   uint32_t hint, bool hint_valid)
@@ -931,6 +997,7 @@ static int mq_load_mapping(struct dm_cache_policy *p,
 	e->cblock = cblock;
 	e->oblock = oblock;
 	e->in_cache = true;
+	e->dirty = false;	/* this gets corrected in a minute */
 	e->hit_count = hint_valid ? hint : 1;
 	e->generation = mq->generation;
 	push(mq, e);
@@ -949,7 +1016,14 @@ static int mq_walk_mappings(struct dm_cache_policy *p, policy_walk_fn fn,
 	mutex_lock(&mq->lock);
 
 	for (level = 0; level < NR_QUEUE_LEVELS; level++)
-		list_for_each_entry(e, &mq->cache.qs[level], list) {
+		list_for_each_entry(e, &mq->cache_clean.qs[level], list) {
+			r = fn(context, e->cblock, e->oblock, e->hit_count);
+			if (r)
+				goto out;
+		}
+
+	for (level = 0; level < NR_QUEUE_LEVELS; level++)
+		list_for_each_entry(e, &mq->cache_dirty.qs[level], list) {
 			r = fn(context, e->cblock, e->oblock, e->hit_count);
 			if (r)
 				goto out;
@@ -974,9 +1048,39 @@ static void mq_remove_mapping(struct dm_cache_policy *p, dm_oblock_t oblock)
 
 	del(mq, e);
 	e->in_cache = false;
+	e->dirty = false;
+	push(mq, e);
+
+	mutex_unlock(&mq->lock);
+}
+
+static int __mq_writeback_work(struct mq_policy *mq, dm_oblock_t *oblock,
+			      dm_cblock_t *cblock)
+{
+	struct entry *e = pop(mq, &mq->cache_dirty);
+
+	if (!e)
+		return -ENODATA;
+
+	*oblock = e->oblock;
+	*cblock = e->cblock;
+	e->dirty = false;
 	push(mq, e);
 
+	return 0;
+}
+
+static int mq_writeback_work(struct dm_cache_policy *p, dm_oblock_t *oblock,
+			     dm_cblock_t *cblock)
+{
+	int r;
+	struct mq_policy *mq = to_mq_policy(p);
+
+	mutex_lock(&mq->lock);
+	r = __mq_writeback_work(mq, oblock, cblock);
 	mutex_unlock(&mq->lock);
+
+	return r;
 }
 
 static void force_mapping(struct mq_policy *mq,
@@ -988,6 +1092,7 @@ static void force_mapping(struct mq_policy *mq,
 
 	del(mq, e);
 	e->oblock = new_oblock;
+	e->dirty = true;
 	push(mq, e);
 }
 
@@ -1063,10 +1168,12 @@ static void init_policy_functions(struct mq_policy *mq)
 	mq->policy.destroy = mq_destroy;
 	mq->policy.map = mq_map;
 	mq->policy.lookup = mq_lookup;
+	mq->policy.set_dirty = mq_set_dirty;
+	mq->policy.clear_dirty = mq_clear_dirty;
 	mq->policy.load_mapping = mq_load_mapping;
 	mq->policy.walk_mappings = mq_walk_mappings;
 	mq->policy.remove_mapping = mq_remove_mapping;
-	mq->policy.writeback_work = NULL;
+	mq->policy.writeback_work = mq_writeback_work;
 	mq->policy.force_mapping = mq_force_mapping;
 	mq->policy.residency = mq_residency;
 	mq->policy.tick = mq_tick;
@@ -1099,7 +1206,9 @@ static struct dm_cache_policy *mq_create(dm_cblock_t cache_size,
 	mq->find_free_last_word = 0;
 
 	queue_init(&mq->pre_cache);
-	queue_init(&mq->cache);
+	queue_init(&mq->cache_clean);
+	queue_init(&mq->cache_dirty);
+
 	mq->generation_period = max((unsigned) from_cblock(cache_size), 1024U);
 
 	mq->nr_entries = 2 * from_cblock(cache_size);