null_blk: unalign call_single_data

Commit Message

Jens Axboe Dec. 20, 2017, 7:34 p.m. UTC

Commit 966a967116e6 randomly added alignment to this structure, but
it's actually detrimental to performance of null_blk. Test case:

# modprobe null_blk queue_mode=1 irqmode=1 home_node={0,1}
# echo noop > /sys/block/nullb0/queue/scheduler
# fio --name=csd --filename=/dev/nullb0 --numjobs=8 --direct=1 --rw=randread --norandommap --cpus_allowed=$(cat /sys/devices/system/node/node0/cpulist) --group_reporting=1

Running on both the home and remote node shows a ~5% degradation
in performance.

Fixes: 966a967116e69 ("smp: Avoid using two cache lines for struct call_single_data")
Signed-off-by: Jens Axboe <[e-mail]>

---

Patch

diff --git a/drivers/block/null_blk.c b/drivers/block/null_blk.c
index ccb9975a97fa..8c6912dd5142 100644
--- a/drivers/block/null_blk.c
+++ b/drivers/block/null_blk.c
@@ -35,7 +35,7 @@  static inline u64 mb_per_tick(int mbps)
 struct nullb_cmd {
 	struct list_head list;
 	struct llist_node ll_list;
-	call_single_data_t csd;
+	struct __call_single_data csd;
 	struct request *rq;
 	struct bio *bio;
 	unsigned int tag;