bcache: Kill btree write work item, flush via journal reclaim

10 files changed, 34 insertions, 52 deletions

diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index 3f604fcd8560..ae2efdeedd24 100644
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -758,7 +758,6 @@ struct cache_set {
 	mempool_t		btree_sort_pool;
 
 	struct journal		journal;
-	unsigned		btree_flush_delay;
 
 	/* CACHING OTHER BLOCK DEVICES */
 	mempool_t		search;
diff --git a/drivers/md/bcache/btree_cache.c b/drivers/md/bcache/btree_cache.c
index 52cd2105add4..3ae92ef40039 100644
--- a/drivers/md/bcache/btree_cache.c
+++ b/drivers/md/bcache/btree_cache.c
@@ -87,7 +87,6 @@ static struct btree *mca_bucket_alloc(struct cache_set *c, gfp_t gfp)
 
 	six_lock_init(&b->lock);
 	INIT_LIST_HEAD(&b->list);
-	INIT_DELAYED_WORK(&b->work, btree_node_write_work);
 	b->c = c;
 	b->writes[1].index = 1;
 	INIT_LIST_HEAD(&b->write_blocked);
@@ -351,7 +350,6 @@ void bch_btree_cache_free(struct cache_set *c)
 		b = list_first_entry(&c->btree_cache_freed,
 				     struct btree, list);
 		list_del(&b->list);
-		cancel_delayed_work_sync(&b->work);
 		kfree(b);
 	}
 
diff --git a/drivers/md/bcache/btree_io.c b/drivers/md/bcache/btree_io.c
index 59ba43e604e1..4b03e6b32243 100644
--- a/drivers/md/bcache/btree_io.c
+++ b/drivers/md/bcache/btree_io.c
@@ -587,11 +587,6 @@ static void btree_node_write_done(struct closure *cl)
 		bch_journal_halt(&c->journal);
 
 	bch_btree_complete_write(c, b, w);
-
-	if (btree_node_dirty(b) && c->btree_flush_delay)
-		queue_delayed_work(system_freezable_wq, &b->work,
-				   c->btree_flush_delay * HZ);
-
 	closure_return_with_destructor(cl, btree_node_write_unlock);
 }
 
@@ -634,8 +629,6 @@ static void do_btree_node_write(struct closure *cl)
 	BUG_ON(btree_bset_first(b)->seq != i->seq);
 	BUG_ON(BSET_BIG_ENDIAN(i) != CPU_BIG_ENDIAN);
 
-	cancel_delayed_work(&b->work);
-
 	change_bit(BTREE_NODE_write_idx, &b->flags);
 
 	i->version	= cpu_to_le16(BCACHE_BSET_VERSION);
@@ -812,13 +805,6 @@ void bch_btree_node_write_lazy(struct btree *b, struct btree_iter *iter)
 		bch_btree_node_write(b, NULL, iter);
 }
 
-void btree_node_write_work(struct work_struct *w)
-{
-	struct btree *b = container_of(to_delayed_work(w), struct btree, work);
-
-	bch_btree_node_write_dirty(b, NULL);
-}
-
 /*
  * Write all dirty btree nodes to disk, including roots
  */
diff --git a/drivers/md/bcache/btree_io.h b/drivers/md/bcache/btree_io.h
index d061ac36f88e..9fe2ec488537 100644
--- a/drivers/md/bcache/btree_io.h
+++ b/drivers/md/bcache/btree_io.h
@@ -36,7 +36,6 @@ void __bch_btree_node_write(struct btree *, struct closure *, int);
 void bch_btree_node_write(struct btree *, struct closure *,
 			  struct btree_iter *);
 void bch_btree_node_write_lazy(struct btree *, struct btree_iter *);
-void btree_node_write_work(struct work_struct *);
 
 void bch_btree_flush(struct cache_set *);
 void bch_btree_node_flush_journal_entries(struct cache_set *, struct btree *,
diff --git a/drivers/md/bcache/btree_types.h b/drivers/md/bcache/btree_types.h
index 1274b3b8c7de..fc6bfe0dad46 100644
--- a/drivers/md/bcache/btree_types.h
+++ b/drivers/md/bcache/btree_types.h
@@ -76,8 +76,6 @@ struct btree {
 	struct list_head	list;
 
 	struct closure		io;
-	struct delayed_work	work;
-
 	struct btree_write	writes[2];
 };
 
diff --git a/drivers/md/bcache/btree_update.c b/drivers/md/bcache/btree_update.c
index b845d82bb9e5..75d5ed086546 100644
--- a/drivers/md/bcache/btree_update.c
+++ b/drivers/md/bcache/btree_update.c
@@ -168,7 +168,6 @@ static void __btree_node_free(struct cache_set *c, struct btree *b,
 	if (btree_node_dirty(b))
 		bch_btree_complete_write(c, b, btree_current_write(b));
 	clear_btree_node_dirty(b);
-	cancel_delayed_work(&b->work);
 
 	mca_hash_remove(c, b);
 
@@ -775,13 +774,8 @@ void bch_btree_journal_key(struct btree_iter *iter,
 			cpu_to_le64(bch_journal_res_seq(j, res));
 	}
 
-	if (!btree_node_dirty(b)) {
+	if (!btree_node_dirty(b))
 		set_btree_node_dirty(b);
-
-		if (c->btree_flush_delay)
-			queue_delayed_work(system_freezable_wq, &b->work,
-					   c->btree_flush_delay * HZ);
-	}
 }
 
 static void verify_keys_sorted(struct keylist *l)
diff --git a/drivers/md/bcache/journal.c b/drivers/md/bcache/journal.c
index d37130c70396..07a51df9ed58 100644
--- a/drivers/md/bcache/journal.c
+++ b/drivers/md/bcache/journal.c
@@ -1216,7 +1216,7 @@ static int journal_entry_open(struct journal *j)
 		if (!old.prev_buf_unwritten)
 			mod_delayed_work(system_freezable_wq,
 					 &j->write_work,
-					 msecs_to_jiffies(j->delay_ms));
+					 msecs_to_jiffies(j->write_delay_ms));
 	}
 
 	return ret;
@@ -1271,7 +1271,7 @@ void bch_journal_start(struct cache_set *c)
 
 	spin_unlock(&j->lock);
 
-	queue_work(system_freezable_wq, &j->reclaim_work);
+	queue_delayed_work(system_freezable_wq, &j->reclaim_work, 0);
 }
 
 int bch_journal_replay(struct cache_set *c, struct list_head *list)
@@ -1597,14 +1597,15 @@ static bool should_discard_bucket(struct journal *j, struct journal_device *ja)
  */
 static void journal_reclaim_work(struct work_struct *work)
 {
-	struct cache_set *c = container_of(work, struct cache_set,
-					   journal.reclaim_work);
+	struct cache_set *c = container_of(to_delayed_work(work),
+				struct cache_set, journal.reclaim_work);
 	struct journal *j = &c->journal;
 	struct cache *ca;
 	struct journal_entry_pin *pin;
 	u64 seq_to_flush = 0;
 	unsigned iter, nr, bucket_to_flush;
 	bool reclaim_lock_held = false;
+	bool flushed = false;
 
 	/*
 	 * Advance last_idx to point to the oldest journal entry containing
@@ -1665,10 +1666,17 @@ static void journal_reclaim_work(struct work_struct *work)
 	seq_to_flush = max_t(s64, seq_to_flush,
 			     (s64) j->seq - (j->pin.size >> 1));
 
-	while ((pin = journal_get_next_pin(j, seq_to_flush))) {
+	/* Ensure we always flush at least one journal pin: */
+
+	while ((pin = journal_get_next_pin(j, flushed ? seq_to_flush : U64_MAX))) {
 		__set_current_state(TASK_RUNNING);
 		pin->flush(j, pin);
+		flushed = true;
 	}
+
+	if (flushed && !test_bit(CACHE_SET_RO, &c->flags))
+		queue_delayed_work(system_freezable_wq, &j->reclaim_work,
+				   msecs_to_jiffies(j->reclaim_delay_ms));
 }
 
 /**
@@ -1844,13 +1852,13 @@ static void journal_write_done(struct closure *cl)
 	if (journal_entry_is_open(j))
 		mod_delayed_work(system_freezable_wq, &j->write_work,
 				 test_bit(JOURNAL_NEED_WRITE, &j->flags)
-				 ? 0 : msecs_to_jiffies(j->delay_ms));
+				 ? 0 : msecs_to_jiffies(j->write_delay_ms));
 
 	/*
 	 * Updating last_seq_ondisk may let journal_reclaim_work() discard more
 	 * buckets:
 	 */
-	queue_work(system_freezable_wq, &j->reclaim_work);
+	mod_delayed_work(system_freezable_wq, &j->reclaim_work, 0);
 }
 
 static void journal_write(struct closure *cl)
@@ -2046,7 +2054,7 @@ retry:
 	 * Direct reclaim - can't rely on reclaim from work item
 	 * due to freezing..
 	 */
-	journal_reclaim_work(&j->reclaim_work);
+	journal_reclaim_work(&j->reclaim_work.work);
 
 	trace_bcache_journal_full(c);
 blocked:
@@ -2229,14 +2237,15 @@ int bch_journal_alloc(struct journal *j)
 	spin_lock_init(&j->pin_lock);
 	init_waitqueue_head(&j->wait);
 	INIT_DELAYED_WORK(&j->write_work, journal_write_work);
-	INIT_WORK(&j->reclaim_work, journal_reclaim_work);
+	INIT_DELAYED_WORK(&j->reclaim_work, journal_reclaim_work);
 	mutex_init(&j->blacklist_lock);
 	INIT_LIST_HEAD(&j->seq_blacklist);
 	mutex_init(&j->reclaim_lock);
 
 	lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
 
-	j->delay_ms = 10;
+	j->write_delay_ms	= 100;
+	j->reclaim_delay_ms	= 100;
 
 	bkey_extent_init(&j->key);
 
diff --git a/drivers/md/bcache/journal_types.h b/drivers/md/bcache/journal_types.h
index b40a2dfae25a..032b02e78625 100644
--- a/drivers/md/bcache/journal_types.h
+++ b/drivers/md/bcache/journal_types.h
@@ -128,8 +128,6 @@ struct journal {
 	struct closure		io;
 	struct delayed_work	write_work;
 
-	unsigned		delay_ms;
-
 	/* Sequence number of most recent journal entry (last entry in @pin) */
 	u64			seq;
 
@@ -166,13 +164,15 @@ struct journal {
 
 	BKEY_PADDED(key);
 
-	struct work_struct	reclaim_work;
+	struct delayed_work	reclaim_work;
 	/* protects advancing ja->last_idx: */
 	struct mutex		reclaim_lock;
 
 	__le64			prio_buckets[MAX_CACHES_PER_SET];
 	unsigned		nr_prio_buckets;
 
+	unsigned		write_delay_ms;
+	unsigned		reclaim_delay_ms;
 
 	u64			res_get_blocked_start;
 	u64			need_write_time;
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index c834c57b0274..e1080b9e6e66 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c
@@ -709,6 +709,7 @@ static void __bch_cache_set_read_only(struct cache_set *c)
 		bch_journal_meta(&c->journal);
 
 	cancel_delayed_work_sync(&c->journal.write_work);
+	cancel_delayed_work_sync(&c->journal.reclaim_work);
 }
 
 static void bch_writes_disabled(struct percpu_ref *writes)
@@ -1111,8 +1112,6 @@ static struct cache_set *bch_cache_set_alloc(struct cache_sb *sb,
 
 	c->writeback_pages_max = (256 << 10) / PAGE_SIZE;
 
-	c->btree_flush_delay = 30;
-
 	c->copy_gc_enabled = 1;
 	c->tiering_enabled = 1;
 	c->tiering_percent = 10;
diff --git a/drivers/md/bcache/sysfs.c b/drivers/md/bcache/sysfs.c
index 1c5f9f6f9689..985be520df52 100644
--- a/drivers/md/bcache/sysfs.c
+++ b/drivers/md/bcache/sysfs.c
@@ -128,7 +128,9 @@ sysfs_pd_controller_attribute(writeback);
 read_attribute(stripe_size);
 read_attribute(partial_stripes_expensive);
 
-rw_attribute(journal_delay_ms);
+rw_attribute(journal_write_delay_ms);
+rw_attribute(journal_reclaim_delay_ms);
+
 rw_attribute(discard);
 rw_attribute(running);
 rw_attribute(label);
@@ -146,7 +148,6 @@ sysfs_pd_controller_attribute(tiering);
 
 sysfs_pd_controller_attribute(foreground_write);
 
-rw_attribute(btree_flush_delay);
 rw_attribute(pd_controllers_update_seconds);
 
 rw_attribute(foreground_target_percent);
@@ -647,7 +648,8 @@ SHOW(bch_cache_set)
 
 	sysfs_print(minor,			c->minor);
 
-	sysfs_print(journal_delay_ms,		c->journal.delay_ms);
+	sysfs_print(journal_write_delay_ms,	c->journal.write_delay_ms);
+	sysfs_print(journal_reclaim_delay_ms,	c->journal.reclaim_delay_ms);
 
 	sysfs_hprint(block_size,		block_bytes(c));
 	sysfs_print(block_size_bytes,		block_bytes(c));
@@ -698,8 +700,6 @@ SHOW(bch_cache_set)
 	sysfs_print(tiering_percent,		c->tiering_percent);
 	sysfs_pd_controller_show(tiering,	&c->tiering_pd);
 
-	sysfs_print(btree_flush_delay,		c->btree_flush_delay);
-
 	sysfs_printf(meta_replicas_have, "%llu",
 		     CACHE_SET_META_REPLICAS_HAVE(&c->disk_sb));
 	sysfs_printf(data_replicas_have, "%llu",
@@ -771,7 +771,9 @@ STORE(__bch_cache_set)
 		return size;
 	}
 
-	sysfs_strtoul(journal_delay_ms, c->journal.delay_ms);
+	sysfs_strtoul(journal_write_delay_ms, c->journal.write_delay_ms);
+	sysfs_strtoul(journal_reclaim_delay_ms, c->journal.reclaim_delay_ms);
+
 	sysfs_strtoul(foreground_write_ratelimit_enabled,
 		      c->foreground_write_ratelimit_enabled);
 
@@ -798,8 +800,6 @@ STORE(__bch_cache_set)
 
 	sysfs_pd_controller_store(foreground_write, &c->foreground_write_pd);
 
-	sysfs_strtoul(btree_flush_delay, c->btree_flush_delay);
-
 	if (attr == &sysfs_journal_flush) {
 		bch_journal_meta_async(&c->journal, NULL);
 
@@ -875,7 +875,8 @@ STORE(bch_cache_set)
 static struct attribute *bch_cache_set_files[] = {
 	&sysfs_unregister,
 	&sysfs_stop,
-	&sysfs_journal_delay_ms,
+	&sysfs_journal_write_delay_ms,
+	&sysfs_journal_reclaim_delay_ms,
 	&sysfs_blockdev_volume_create,
 	&sysfs_add_device,
 
@@ -901,7 +902,6 @@ static struct attribute *bch_cache_set_files[] = {
 	&sysfs_meta_replicas_have,
 	&sysfs_data_replicas_have,
 
-	&sysfs_btree_flush_delay,
 	&sysfs_foreground_target_percent,
 	&sysfs_tiering_percent,