[PATCH] re-fix sata_sil quirk (w/ description)

[PATCH] re-fix sata_sil quirk (w/ description)

[Jeff Garzik]

[-- Attachment #1: Type: text/plain, Size: 452 bytes --]


The patch itself is unchanged, but I have checked it in with a detailed 
technical description, including my rationale.

Bart, I'm not going to bother with a 'better fix'.  Since this only 
affects 3112 controllers, I will be adding code to make sure that the 
mod15write quirk does not activate for the other controllers (3512, 3114).

See the changeset description accompanying the patch for more info.

Pushing to Andrew and Linus right now...





[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 2908 bytes --]

#   2004/06/22 20:42:16-04:00 jgarzik@pobox.com 
#   [libata sata_sil] Re-fix mod15write bug
#   
#   Certain early SATA drives have problems with write requests whose
#   length satisfy the equation "sectors % 15 == 1", on the SiI 3112.
#   Other drives, and other SiI controllers, are not affected.
#   
#   The fix for this problem is to avoid such requests, in one of three
#   ways, for the affect drive+controller combos:
#   1) Limit all writes to 15 sectors
#   2) Use block layer features to avoid creating requests whose
#      length satisfies the above equation.
#   3) When a request satisfies the above equation, split the request
#      into two writes, neither of which satisfies the equation.
#   
#   I chose fix #1, the most simple to implement.  After discussion with
#   Silicon Image and others regarding the impact of this fix, I have
#   decided to remain with fix #1, and will not be implementing a
#   "better fix".  This means that the affected SATA drives will see
#   decreased performance, but set of affected drives is small and will
#   never grow larger.
#   
#   Further, the complexity of implementing solution #2 or
#   solution #3 is rather large.
#   
#   When implementing lba48 'large request' support, I unintentionally
#   broke the fix for these affected drives.  Kudos to Ricky Beam for
#   noticing this.
#   
#   This change restores the fix, by adding a flag ATA_DFLAG_LOCK_SECTORS
#   to indicate that the max_sectors value set by the low-level driver
#   should never be changed.
# 
diff -Nru a/drivers/scsi/libata-scsi.c b/drivers/scsi/libata-scsi.c
--- a/drivers/scsi/libata-scsi.c	2004-06-22 20:44:51 -04:00
+++ b/drivers/scsi/libata-scsi.c	2004-06-22 20:44:51 -04:00
@@ -182,7 +182,8 @@
 		 * 65534 when Jens Axboe's patch for dynamically
 		 * determining max_sectors is merged.
 		 */
-		if (dev->flags & ATA_DFLAG_LBA48) {
+		if ((dev->flags & ATA_DFLAG_LBA48) &&
+		    ((dev->flags & ATA_DFLAG_LOCK_SECTORS) == 0)) {
 			sdev->host->max_sectors = 2048;
 			blk_queue_max_sectors(sdev->request_queue, 2048);
 		}
diff -Nru a/drivers/scsi/sata_sil.c b/drivers/scsi/sata_sil.c
--- a/drivers/scsi/sata_sil.c	2004-06-22 20:44:51 -04:00
+++ b/drivers/scsi/sata_sil.c	2004-06-22 20:44:51 -04:00
@@ -302,6 +302,7 @@
 		       ap->id, dev->devno);
 		ap->host->max_sectors = 15;
 		ap->host->hostt->max_sectors = 15;
+		dev->flags |= ATA_DFLAG_LOCK_SECTORS;
 		return;
 	}
 
diff -Nru a/include/linux/libata.h b/include/linux/libata.h
--- a/include/linux/libata.h	2004-06-22 20:44:51 -04:00
+++ b/include/linux/libata.h	2004-06-22 20:44:51 -04:00
@@ -91,6 +91,7 @@
 	ATA_DFLAG_MASTER	= (1 << 2), /* is device 0? */
 	ATA_DFLAG_WCACHE	= (1 << 3), /* has write cache we can
 					     * (hopefully) flush? */
+	ATA_DFLAG_LOCK_SECTORS	= (1 << 4), /* don't adjust max_sectors */
 
 	ATA_DEV_UNKNOWN		= 0,	/* unknown device */
 	ATA_DEV_ATA		= 1,	/* ATA device */

Re: [PATCH] re-fix sata_sil quirk (w/ description)

[Ricky Beam]

On Tue, 22 Jun 2004, Jeff Garzik wrote:
>Bart, I'm not going to bother with a 'better fix'.  Since this only
>affects 3112 controllers, I will be adding code to make sure that the
>mod15write quirk does not activate for the other controllers (3512, 3114).

Actually, it does effect other controllers ... I'm using a 3114.  And it's
currently only known to effect Seagate drives.  It may still happen to
other drives and my not effect Seagate drives with some future firmware
revision.

Current activity using buffered writes:
Device:            tps    kB_read/s    kB_wrtn/s    kB_read    kB_wrtn
sda            1694.95         0.00     10438.68          0     626112
sdb            1694.33         0.00     10438.68          0     626112
sdc            1643.23         0.00     10438.68          0     626112
sdd            1631.01         0.00     10438.68          0     626112
md_d0         26096.70         0.00     46974.06          0    2817504
md_d0p2       20877.36         0.00     41754.72          0    2504448

That's certainly acceptable -- and in fact right at the edge of what the
3114 can do with all four port lit up.

--Ricky

Re: [PATCH] re-fix sata_sil quirk (w/ description)

[Ricky Beam]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 706 bytes --]

On Tue, 22 Jun 2004, Jeff Garzik wrote:
>The patch itself is unchanged, but I have checked it in with a detailed
>technical description, including my rationale.
...

#   ... This means that the affected SATA drives will see
#   decreased performance, but set of affected drives is small and will
#   never grow larger.

Actually, the set of affected drives is about 1/3rd of the SATA drive
market: all current model Seagate drives. (Maxtor and WD being the other
thirds.)  My money says their new 400G monster will have the exact same
problems.

I'm using ST3160023AS drives (3 FW 3.18, 1 FW 3.05.) I'd include their
manufacture date, but they're rather busy right now so I cannot pull them
out.

--Ricky


[-- Attachment #2: Type: TEXT/PLAIN, Size: 2908 bytes --]

#   2004/06/22 20:42:16-04:00 jgarzik@pobox.com 
#   [libata sata_sil] Re-fix mod15write bug
#   
#   Certain early SATA drives have problems with write requests whose
#   length satisfy the equation "sectors % 15 == 1", on the SiI 3112.
#   Other drives, and other SiI controllers, are not affected.
#   
#   The fix for this problem is to avoid such requests, in one of three
#   ways, for the affect drive+controller combos:
#   1) Limit all writes to 15 sectors
#   2) Use block layer features to avoid creating requests whose
#      length satisfies the above equation.
#   3) When a request satisfies the above equation, split the request
#      into two writes, neither of which satisfies the equation.
#   
#   I chose fix #1, the most simple to implement.  After discussion with
#   Silicon Image and others regarding the impact of this fix, I have
#   decided to remain with fix #1, and will not be implementing a
#   "better fix".  This means that the affected SATA drives will see
#   decreased performance, but set of affected drives is small and will
#   never grow larger.
#   
#   Further, the complexity of implementing solution #2 or
#   solution #3 is rather large.
#   
#   When implementing lba48 'large request' support, I unintentionally
#   broke the fix for these affected drives.  Kudos to Ricky Beam for
#   noticing this.
#   
#   This change restores the fix, by adding a flag ATA_DFLAG_LOCK_SECTORS
#   to indicate that the max_sectors value set by the low-level driver
#   should never be changed.
# 
diff -Nru a/drivers/scsi/libata-scsi.c b/drivers/scsi/libata-scsi.c
--- a/drivers/scsi/libata-scsi.c	2004-06-22 20:44:51 -04:00
+++ b/drivers/scsi/libata-scsi.c	2004-06-22 20:44:51 -04:00
@@ -182,7 +182,8 @@
 		 * 65534 when Jens Axboe's patch for dynamically
 		 * determining max_sectors is merged.
 		 */
-		if (dev->flags & ATA_DFLAG_LBA48) {
+		if ((dev->flags & ATA_DFLAG_LBA48) &&
+		    ((dev->flags & ATA_DFLAG_LOCK_SECTORS) == 0)) {
 			sdev->host->max_sectors = 2048;
 			blk_queue_max_sectors(sdev->request_queue, 2048);
 		}
diff -Nru a/drivers/scsi/sata_sil.c b/drivers/scsi/sata_sil.c
--- a/drivers/scsi/sata_sil.c	2004-06-22 20:44:51 -04:00
+++ b/drivers/scsi/sata_sil.c	2004-06-22 20:44:51 -04:00
@@ -302,6 +302,7 @@
 		       ap->id, dev->devno);
 		ap->host->max_sectors = 15;
 		ap->host->hostt->max_sectors = 15;
+		dev->flags |= ATA_DFLAG_LOCK_SECTORS;
 		return;
 	}
 
diff -Nru a/include/linux/libata.h b/include/linux/libata.h
--- a/include/linux/libata.h	2004-06-22 20:44:51 -04:00
+++ b/include/linux/libata.h	2004-06-22 20:44:51 -04:00
@@ -91,6 +91,7 @@
 	ATA_DFLAG_MASTER	= (1 << 2), /* is device 0? */
 	ATA_DFLAG_WCACHE	= (1 << 3), /* has write cache we can
 					     * (hopefully) flush? */
+	ATA_DFLAG_LOCK_SECTORS	= (1 << 4), /* don't adjust max_sectors */
 
 	ATA_DEV_UNKNOWN		= 0,	/* unknown device */
 	ATA_DEV_ATA		= 1,	/* ATA device */

Re: [PATCH] re-fix sata_sil quirk (w/ description)

[Jeff Garzik]

Ricky Beam wrote:
> On Tue, 22 Jun 2004, Jeff Garzik wrote:
> 
>> The patch itself is unchanged, but I have checked it in with a detailed
>> technical description, including my rationale.
> 
> ...
> 
> #   ... This means that the affected SATA drives will see
> #   decreased performance, but set of affected drives is small and will
> #   never grow larger.
> 
> Actually, the set of affected drives is about 1/3rd of the SATA drive
> market: all current model Seagate drives. (Maxtor and WD being the other
> thirds.)  My money says their new 400G monster will have the exact same
> problems.
> 
> I'm using ST3160023AS drives (3 FW 3.18, 1 FW 3.05.) I'd include their
> manufacture date, but they're rather busy right now so I cannot pull them
> out.

Seagate claims they fixed the problem, so I doubt it...

	Jeff