[PATCH] libata: get rid of ATA_MAX_QUEUE loop in ata_qc_complete_multiple()

[PATCH] libata: get rid of ATA_MAX_QUEUE loop in ata_qc_complete_multiple()

[Jens Axboe]

Hi,

We very rarely (if ever) complete more than one command in the
sactive mask at the time, even for extremely high IO rates. So
looping over the entire range of possible tags is pointless,
instead use __ffs() to just find the completed tags directly.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
---
 drivers/ata/libata-core.c |   15 +++++++++------
 1 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
index 17c5d48..8de0081 100644
--- a/drivers/ata/libata-core.c
+++ b/drivers/ata/libata-core.c
@@ -5027,9 +5027,9 @@ void ata_qc_complete(struct ata_queued_cmd *qc)
  */
 int ata_qc_complete_multiple(struct ata_port *ap, u32 qc_active)
 {
+	unsigned int i = 0;
 	int nr_done = 0;
 	u32 done_mask;
-	int i;
 
 	done_mask = ap->qc_active ^ qc_active;
 
@@ -5039,16 +5039,19 @@ int ata_qc_complete_multiple(struct ata_port *ap, u32 qc_active)
 		return -EINVAL;
 	}
 
-	for (i = 0; i < ATA_MAX_QUEUE; i++) {
+	while (done_mask) {
 		struct ata_queued_cmd *qc;
+		unsigned int next = __ffs(done_mask);
 
-		if (!(done_mask & (1 << i)))
-			continue;
-
-		if ((qc = ata_qc_from_tag(ap, i))) {
+		qc = ata_qc_from_tag(ap, i + next);
+		if (qc) {
 			ata_qc_complete(qc);
 			nr_done++;
 		}
+		if (++next >= ATA_MAX_QUEUE)
+			break;
+		i += next;
+		done_mask >>= next;
 	}
 
 	return nr_done;
-- 
1.6.3.rc0.1.gf800


-- 
Jens Axboe

Re: [PATCH] libata: get rid of ATA_MAX_QUEUE loop in ata_qc_complete_multiple()

[Tejun Heo]

Hello, Jens.

Jens Axboe wrote:
> @@ -5039,16 +5039,19 @@ int ata_qc_complete_multiple(struct ata_port *ap, u32 qc_active)
>  		return -EINVAL;
>  	}
>  
> -	for (i = 0; i < ATA_MAX_QUEUE; i++) {
> +	while (done_mask) {
>  		struct ata_queued_cmd *qc;
> +		unsigned int next = __ffs(done_mask);
>  
> -		if (!(done_mask & (1 << i)))
> -			continue;
> -
> -		if ((qc = ata_qc_from_tag(ap, i))) {
> +		qc = ata_qc_from_tag(ap, i + next);
> +		if (qc) {
>  			ata_qc_complete(qc);
>  			nr_done++;
>  		}
> +		if (++next >= ATA_MAX_QUEUE)
> +			break;

Given that __ffs doesn't care how far the first bit is on most archs
(even the generic one), I think it would be simpler and more efficient
to simply do done_mask &= ~(1 << i) and loop without adding to index
and shifting.

Thanks.

-- 
tejun

Re: [PATCH] libata: get rid of ATA_MAX_QUEUE loop in ata_qc_complete_multiple()

[Jens Axboe]

On Wed, May 20 2009, Tejun Heo wrote:
> Hello, Jens.
> 
> Jens Axboe wrote:
> > @@ -5039,16 +5039,19 @@ int ata_qc_complete_multiple(struct ata_port *ap, u32 qc_active)
> >  		return -EINVAL;
> >  	}
> >  
> > -	for (i = 0; i < ATA_MAX_QUEUE; i++) {
> > +	while (done_mask) {
> >  		struct ata_queued_cmd *qc;
> > +		unsigned int next = __ffs(done_mask);
> >  
> > -		if (!(done_mask & (1 << i)))
> > -			continue;
> > -
> > -		if ((qc = ata_qc_from_tag(ap, i))) {
> > +		qc = ata_qc_from_tag(ap, i + next);
> > +		if (qc) {
> >  			ata_qc_complete(qc);
> >  			nr_done++;
> >  		}
> > +		if (++next >= ATA_MAX_QUEUE)
> > +			break;
> 
> Given that __ffs doesn't care how far the first bit is on most archs
> (even the generic one), I think it would be simpler and more efficient
> to simply do done_mask &= ~(1 << i) and loop without adding to index
> and shifting.

It would be easier to read as well, agree. I'll repost it in 2 secs.

-- 
Jens Axboe