diff mbox

[v5,6/7] scsi_io_completion hints on fastpatch

Message ID 20180329054702.28644-7-dgilbert@interlog.com (mailing list archive)
State Changes Requested
Headers show

Commit Message

Douglas Gilbert March 29, 2018, 5:47 a.m. UTC 
Add likely() and unlikely() hints to conditionals on or near the fastpath.

Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
---
A reviewer wanted any performance improvement (or otherwise) quantified.
The improvement was so small, that ftrace ignored it. Inline timing code
suggests the improvement from this whole patchset is around 7 nanoseconds
per invocation (tested on a Lenovo X270 (i5-7200U)). Not exactly huge.
Another win might be the smaller size of scsi_io_completion() after the
refactoring; this should allow more other code to fit in the instruction
cache.

 drivers/scsi/scsi_lib.c | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

Comments

Bart Van Assche March 29, 2018, 11:49 p.m. UTC | #1 
On Thu, 2018-03-29 at 01:47 -0400, Douglas Gilbert wrote:
> -	/* no bidi support for !blk_rq_is_passthrough yet */

> +	/* no bidi support yet, other than in pass-through */

> 

> [ ... ]

>

> -	 /* Kill remainder if no retries. */

> +	/* Kill remainder if no retries. */


If you would repost this patch series, please fold these two changes into a
previous patch since these changes are not relately to branch optimizing.

Anyway:

Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
diff mbox

Patch

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index d24bd29cd94a..60b57324a244 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1028,17 +1028,17 @@  void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 	struct request *req = cmd->request;
 	blk_status_t blk_stat = BLK_STS_OK;
 
-	if (result)	/* does not necessarily mean there is an error */
+	if (unlikely(result))	/* a nz result may or may not be an error */
 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
 
-	if (blk_rq_is_passthrough(req)) {
+	if (unlikely(blk_rq_is_passthrough(req))) {
 		/*
 		 * __scsi_error_from_host_byte may have reset the host_byte
 		 */
 		scsi_req(req)->result = cmd->result;
 		scsi_req(req)->resid_len = scsi_get_resid(cmd);
 
-		if (scsi_bidi_cmnd(cmd)) {
+		if (unlikely(scsi_bidi_cmnd(cmd))) {
 			/*
 			 * Bidi commands Must be complete as a whole,
 			 * both sides at once.
@@ -1051,7 +1051,7 @@  void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 		}
 	}
 
-	/* no bidi support for !blk_rq_is_passthrough yet */
+	/* no bidi support yet, other than in pass-through */
 	BUG_ON(blk_bidi_rq(req));
 
 	/*
@@ -1067,13 +1067,13 @@  void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 	 * handle. Failed, zero length commands always need to drop down
 	 * to retry code. Fast path should return in this block.
 	 */
-	if (blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK) {
-		if (!scsi_end_request(req, blk_stat, good_bytes, 0))
+	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
+		if (likely(!scsi_end_request(req, blk_stat, good_bytes, 0)))
 			return; /* no bytes remaining */
 	}
 
-	 /* Kill remainder if no retries. */
-	if (blk_stat && scsi_noretry_cmd(cmd)) {
+	/* Kill remainder if no retries. */
+	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req), 0))
 			BUG();
 		return;
@@ -1083,7 +1083,7 @@  void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 	 * If there had been no error, but we have leftover bytes in the
 	 * requeues just queue the command up again.
 	 */
-	if (result == 0)
+	if (likely(result == 0))
 		scsi_io_completion_reprep(cmd, q);
 	else
 		scsi_io_completion_action(cmd, result);