scsi: target: Save a few cycles in transport_lookup_[cmd|tmr]_lun()

[PATCH v2] scsi: target: Save a few cycles in transport_lookup_[cmd|tmr]_lun()

Posted by Christophe JAILLET 3 years, 7 months ago

Use percpu_ref_tryget_live_rcu() instead of percpu_ref_tryget_live() to
save a few cycles when it is known that the rcu lock is already
taken/released.

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
---
Change in v2:
  * sync with -next

v1:
  * https://lore.kernel.org/all/e4a21bc607c39935cb98d4825cd63ba349820550.1635974637.git.christophe.jaillet@wanadoo.fr/
---
 drivers/target/target_core_device.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/target/target_core_device.c b/drivers/target/target_core_device.c
index b7f16ee8aa0e..76c6613a1da7 100644
--- a/drivers/target/target_core_device.c
+++ b/drivers/target/target_core_device.c
@@ -77,7 +77,7 @@ transport_lookup_cmd_lun(struct se_cmd *se_cmd)
 
 		se_lun = deve->se_lun;
 
-		if (!percpu_ref_tryget_live(&se_lun->lun_ref)) {
+		if (!percpu_ref_tryget_live_rcu(&se_lun->lun_ref)) {
 			se_lun = NULL;
 			goto out_unlock;
 		}
@@ -154,7 +154,7 @@ int transport_lookup_tmr_lun(struct se_cmd *se_cmd)
 	if (deve) {
 		se_lun = deve->se_lun;
 
-		if (!percpu_ref_tryget_live(&se_lun->lun_ref)) {
+		if (!percpu_ref_tryget_live_rcu(&se_lun->lun_ref)) {
 			se_lun = NULL;
 			goto out_unlock;
 		}
-- 
2.34.1

Re: [PATCH v2] scsi: target: Save a few cycles in transport_lookup_[cmd|tmr]_lun()

Posted by Chaitanya Kulkarni 3 years, 7 months ago

On 8/15/22 13:52, Christophe JAILLET wrote:
> Use percpu_ref_tryget_live_rcu() instead of percpu_ref_tryget_live() to
> save a few cycles when it is known that the rcu lock is already
> taken/released.
> 
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> ---

do you have a quantitative data that shows actual savings of cycles?

-ck

Re: [PATCH v2] scsi: target: Save a few cycles in transport_lookup_[cmd|tmr]_lun()

Posted by Christophe JAILLET 3 years, 7 months ago

Le 16/08/2022 à 08:06, Chaitanya Kulkarni a écrit :
> On 8/15/22 13:52, Christophe JAILLET wrote:
>> Use percpu_ref_tryget_live_rcu() instead of percpu_ref_tryget_live() to
>> save a few cycles when it is known that the rcu lock is already
>> taken/released.
>>
>> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
>> ---
> do you have a quantitative data that shows actual savings of cycles?
>
> -ck
>

Some numbers were given for io_uring by the one who introduced 
percpu_ref_tryget_live_rcu().

See [1].

I don't have specific numbers for the patch against scsi.

CJ


[1]: 
https://lore.kernel.org/linux-kernel/cover.1634822969.git.asml.silence@gmail.com/