dm thin: performance improvement to discard processing

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@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c
index 0686078..8c3d048 100644 (file)
--- 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);