[dm-devel] [for-3.19 PATCH v2 12/17] dm thin: performance improvement to discard processing

[Mike Snitzer]

From: Joe Thornber <ejt at redhat.com>

When processing a discard bio, if the block is already quiesced do the
discard immediately rather than adding the mapping to a list for the
next iteration of the worker thread.

Discarding a fully provisioned 100G thin volume with 64k block size goes
from 860s to 95s with this change.

Clearly there's something wrong with the worker architecture, more
investigation needed.

Signed-off-by: Joe Thornber <ejt at redhat.com>
Signed-off-by: Mike Snitzer <snitzer at redhat.com>
---
 drivers/md/dm-thin.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index 73a1c96..fffdebc 100644
--- a/drivers/md/dm-thin.c
+++ b/drivers/md/dm-thin.c
@@ -1194,7 +1194,6 @@ static void retry_bios_on_resume(struct pool *pool, struct dm_bio_prison_cell *c
 static void process_discard(struct thin_c *tc, struct bio *bio)
 {
 	int r;
-	unsigned long flags;
 	struct pool *pool = tc->pool;
 	struct dm_bio_prison_cell *cell, *cell2;
 	struct dm_cell_key key, key2;
@@ -1235,12 +1234,9 @@ static void process_discard(struct thin_c *tc, struct bio *bio)
 			m->cell2 = cell2;
 			m->bio = bio;
 
-			if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) {
-				spin_lock_irqsave(&pool->lock, flags);
-				list_add_tail(&m->list, &pool->prepared_discards);
-				spin_unlock_irqrestore(&pool->lock, flags);
-				wake_worker(pool);
-			}
+			if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list))
+				pool->process_prepared_discard(m);
+
 		} else {
 			inc_all_io_entry(pool, bio);
 			cell_defer_no_holder(tc, cell);
-- 
1.8.3.1