[PATCH 03/20] Revert "bcache: Kill btree_io_wq"

[PATCH 03/20] Revert "bcache: Kill btree_io_wq"

[Coly Li]

From: Kai Krakow <kai@kaishome.de>

This reverts commit 56b30770b27d54d68ad51eccc6d888282b568cee.

With the btree using the `system_wq`, I seem to see a lot more desktop
latency than I should.

After some more investigation, it looks like the original assumption
of 56b3077 no longer is true, and bcache has a very high potential of
congesting the `system_wq`. In turn, this introduces laggy desktop
performance, IO stalls (at least with btrfs), and input events may be
delayed.

So let's revert this. It's important to note that the semantics of
using `system_wq` previously mean that `btree_io_wq` should be created
before and destroyed after other bcache wqs to keep the same
assumptions.

Cc: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org # 5.4+
Signed-off-by: Kai Krakow <kai@kaishome.de>
Signed-off-by: Coly Li <colyli@suse.de>
---
 drivers/md/bcache/bcache.h |  2 ++
 drivers/md/bcache/btree.c  | 21 +++++++++++++++++++--
 drivers/md/bcache/super.c  |  4 ++++
 3 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index d7a84327b7f1..2b8c7dd2cfae 100644
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -1046,5 +1046,7 @@ void bch_debug_exit(void);
 void bch_debug_init(void);
 void bch_request_exit(void);
 int bch_request_init(void);
+void bch_btree_exit(void);
+int bch_btree_init(void);
 
 #endif /* _BCACHE_H */
diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index 910df242c83d..952f022db5a5 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -99,6 +99,8 @@
 #define PTR_HASH(c, k)							\
 	(((k)->ptr[0] >> c->bucket_bits) | PTR_GEN(k, 0))
 
+static struct workqueue_struct *btree_io_wq;
+
 #define insert_lock(s, b)	((b)->level <= (s)->lock)
 
 
@@ -308,7 +310,7 @@ static void __btree_node_write_done(struct closure *cl)
 	btree_complete_write(b, w);
 
 	if (btree_node_dirty(b))
-		schedule_delayed_work(&b->work, 30 * HZ);
+		queue_delayed_work(btree_io_wq, &b->work, 30 * HZ);
 
 	closure_return_with_destructor(cl, btree_node_write_unlock);
 }
@@ -481,7 +483,7 @@ static void bch_btree_leaf_dirty(struct btree *b, atomic_t *journal_ref)
 	BUG_ON(!i->keys);
 
 	if (!btree_node_dirty(b))
-		schedule_delayed_work(&b->work, 30 * HZ);
+		queue_delayed_work(btree_io_wq, &b->work, 30 * HZ);
 
 	set_btree_node_dirty(b);
 
@@ -2764,3 +2766,18 @@ void bch_keybuf_init(struct keybuf *buf)
 	spin_lock_init(&buf->lock);
 	array_allocator_init(&buf->freelist);
 }
+
+void bch_btree_exit(void)
+{
+	if (btree_io_wq)
+		destroy_workqueue(btree_io_wq);
+}
+
+int __init bch_btree_init(void)
+{
+	btree_io_wq = create_singlethread_workqueue("bch_btree_io");
+	if (!btree_io_wq)
+		return -ENOMEM;
+
+	return 0;
+}
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index e7d1b52c5cc8..85a44a0cffe0 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c
@@ -2821,6 +2821,7 @@ static void bcache_exit(void)
 		destroy_workqueue(bcache_wq);
 	if (bch_journal_wq)
 		destroy_workqueue(bch_journal_wq);
+	bch_btree_exit();
 
 	if (bcache_major)
 		unregister_blkdev(bcache_major, "bcache");
@@ -2876,6 +2877,9 @@ static int __init bcache_init(void)
 		return bcache_major;
 	}
 
+	if (bch_btree_init())
+		goto err;
+
 	bcache_wq = alloc_workqueue("bcache", WQ_MEM_RECLAIM, 0);
 	if (!bcache_wq)
 		goto err;
-- 
2.26.2

[PATCH 04/20] bcache: Give btree_io_wq correct semantics again

[Coly Li]

From: Kai Krakow <kai@kaishome.de>

Before killing `btree_io_wq`, the queue was allocated using
`create_singlethread_workqueue()` which has `WQ_MEM_RECLAIM`. After
killing it, it no longer had this property but `system_wq` is not
single threaded.

Let's combine both worlds and make it multi threaded but able to
reclaim memory.

Cc: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org # 5.4+
Signed-off-by: Kai Krakow <kai@kaishome.de>
Signed-off-by: Coly Li <colyli@suse.de>
---
 drivers/md/bcache/btree.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index 952f022db5a5..fe6dce125aba 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -2775,7 +2775,7 @@ void bch_btree_exit(void)
 
 int __init bch_btree_init(void)
 {
-	btree_io_wq = create_singlethread_workqueue("bch_btree_io");
+	btree_io_wq = alloc_workqueue("bch_btree_io", WQ_MEM_RECLAIM, 0);
 	if (!btree_io_wq)
 		return -ENOMEM;
 
-- 
2.26.2

[PATCH 05/20] bcache: Move journal work to new flush wq

[Coly Li]

From: Kai Krakow <kai@kaishome.de>

This is potentially long running and not latency sensitive, let's get
it out of the way of other latency sensitive events.

As observed in the previous commit, the `system_wq` comes easily
congested by bcache, and this fixes a few more stalls I was observing
every once in a while.

Let's not make this `WQ_MEM_RECLAIM` as it showed to reduce performance
of boot and file system operations in my tests. Also, without
`WQ_MEM_RECLAIM`, I no longer see desktop stalls. This matches the
previous behavior as `system_wq` also does no memory reclaim:

> // workqueue.c:
> system_wq = alloc_workqueue("events", 0, 0);

Cc: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org # 5.4+
Signed-off-by: Kai Krakow <kai@kaishome.de>
Signed-off-by: Coly Li <colyli@suse.de>
---
 drivers/md/bcache/bcache.h  |  1 +
 drivers/md/bcache/journal.c |  4 ++--
 drivers/md/bcache/super.c   | 16 ++++++++++++++++
 3 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index 2b8c7dd2cfae..848dd4db1659 100644
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -1005,6 +1005,7 @@ void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent);
 
 extern struct workqueue_struct *bcache_wq;
 extern struct workqueue_struct *bch_journal_wq;
+extern struct workqueue_struct *bch_flush_wq;
 extern struct mutex bch_register_lock;
 extern struct list_head bch_cache_sets;
 
diff --git a/drivers/md/bcache/journal.c b/drivers/md/bcache/journal.c
index aefbdb7e003b..c6613e817333 100644
--- a/drivers/md/bcache/journal.c
+++ b/drivers/md/bcache/journal.c
@@ -932,8 +932,8 @@ atomic_t *bch_journal(struct cache_set *c,
 		journal_try_write(c);
 	} else if (!w->dirty) {
 		w->dirty = true;
-		schedule_delayed_work(&c->journal.work,
-				      msecs_to_jiffies(c->journal_delay_ms));
+		queue_delayed_work(bch_flush_wq, &c->journal.work,
+				   msecs_to_jiffies(c->journal_delay_ms));
 		spin_unlock(&c->journal.lock);
 	} else {
 		spin_unlock(&c->journal.lock);
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index 85a44a0cffe0..0228ccb293fc 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c
@@ -49,6 +49,7 @@ static int bcache_major;
 static DEFINE_IDA(bcache_device_idx);
 static wait_queue_head_t unregister_wait;
 struct workqueue_struct *bcache_wq;
+struct workqueue_struct *bch_flush_wq;
 struct workqueue_struct *bch_journal_wq;
 
 
@@ -2821,6 +2822,8 @@ static void bcache_exit(void)
 		destroy_workqueue(bcache_wq);
 	if (bch_journal_wq)
 		destroy_workqueue(bch_journal_wq);
+	if (bch_flush_wq)
+		destroy_workqueue(bch_flush_wq);
 	bch_btree_exit();
 
 	if (bcache_major)
@@ -2884,6 +2887,19 @@ static int __init bcache_init(void)
 	if (!bcache_wq)
 		goto err;
 
+	/*
+	 * Let's not make this `WQ_MEM_RECLAIM` for the following reasons:
+	 *
+	 * 1. It used `system_wq` before which also does no memory reclaim.
+	 * 2. With `WQ_MEM_RECLAIM` desktop stalls, increased boot times, and
+	 *    reduced throughput can be observed.
+	 *
+	 * We still want to user our own queue to not congest the `system_wq`.
+	 */
+	bch_flush_wq = alloc_workqueue("bch_flush", 0, 0);
+	if (!bch_flush_wq)
+		goto err;
+
 	bch_journal_wq = alloc_workqueue("bch_journal", WQ_MEM_RECLAIM, 0);
 	if (!bch_journal_wq)
 		goto err;
-- 
2.26.2