bcache: allow quick writeback when backing idle

If the control system would wait for at least half a second, and there's
been no reqs hitting the backing disk for awhile: use an alternate mode
where we have at most one contiguous set of writebacks in flight at a
time. (But don't otherwise delay).  If front-end IO appears, it will
still be quick, as it will only have to contend with one real operation
in flight.  But otherwise, we'll be sending data to the backing disk as
quickly as it can accept it (with one op at a time).

Signed-off-by: Michael Lyle <mlyle@lyle.org>
Reviewed-by: Tang Junhui <tang.junhui@zte.com.cn>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index 3be0fcc..5f7b0b2 100644 (file)
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -320,6 +320,13 @@ struct cached_dev {
         */
        atomic_t                has_dirty;
 
+       /*
+        * Set to zero by things that touch the backing volume-- except
+        * writeback.  Incremented by writeback.  Used to determine when to
+        * accelerate idle writeback.
+        */
+       atomic_t                backing_idle;
+
        struct bch_ratelimit    writeback_rate;
        struct delayed_work     writeback_rate_update;
 

diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
index 52b4ce2..ddd9410 100644 (file)
--- a/drivers/md/bcache/request.c
+++ b/drivers/md/bcache/request.c
@@ -996,6 +996,7 @@ static blk_qc_t cached_dev_make_request(struct request_queue *q,
        struct cached_dev *dc = container_of(d, struct cached_dev, disk);
        int rw = bio_data_dir(bio);
 
+       atomic_set(&dc->backing_idle, 0);
        generic_start_io_acct(q, rw, bio_sectors(bio), &d->disk->part0);
 
        bio_set_dev(bio, dc->bdev);

diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
index 6e1d2fd..f82ffb2 100644 (file)
--- a/drivers/md/bcache/writeback.c
+++ b/drivers/md/bcache/writeback.c
@@ -356,6 +356,27 @@ static void read_dirty(struct cached_dev *dc)
 
                delay = writeback_delay(dc, size);
 
+               /* If the control system would wait for at least half a
+                * second, and there's been no reqs hitting the backing disk
+                * for awhile: use an alternate mode where we have at most
+                * one contiguous set of writebacks in flight at a time.  If
+                * someone wants to do IO it will be quick, as it will only
+                * have to contend with one operation in flight, and we'll
+                * be round-tripping data to the backing disk as quickly as
+                * it can accept it.
+                */
+               if (delay >= HZ / 2) {
+                       /* 3 means at least 1.5 seconds, up to 7.5 if we
+                        * have slowed way down.
+                        */
+                       if (atomic_inc_return(&dc->backing_idle) >= 3) {
+                               /* Wait for current I/Os to finish */
+                               closure_sync(&cl);
+                               /* And immediately launch a new set. */
+                               delay = 0;
+                       }
+               }
+
                while (!kthread_should_stop() && delay) {
                        schedule_timeout_interruptible(delay);
                        delay = writeback_delay(dc, 0);