Skip to contentSkip to conversation

sata_sil: fix spurious IRQ handling

PATCH In review Tejun Heo 2 participants 6 messages Dec 7, 2007 – Dec 18, 2007 libata

lore
Share & tools
lore

Tejun Heo

author
#0 lore raw

PATCH 1/1 sata_sil: fix spurious IRQ handling

Interestingly, sata_sil raises spurious interrupts if it's coupled with Sil SATA_PATA bridge. Currently, sata_sil interrupt handler is strict about spurious interrupts and freezes the port when it occurs. This patch makes it more forgiving.

* On SATA PHY event interrupt, serror value is checked to see whether it really is PHYRDY CHG event. If not, SATA PHY event interrupt is ignored.

* If ATA interrupt occurs while no command is in progress, it's cleared and ignored.

This fixes bugzilla bug 9505.

http://bugzilla.kernel.org/show_bug.cgi?id=9505

Signed-off-by Tejun Heo <[email protected]>
---
 drivers/ata/sata_sil.c |   29 +++++++++++++++++------------
 1 file changed, 17 insertions(+), 12 deletions(-)
drivers/ata/sata_sil.c +17 -12
Index: work/drivers/ata/sata_sil.c
===================================================================
@@ -390,23 +390,28 @@ static void sil_host_intr(struct ata_por
 		sil_scr_read(ap, SCR_ERROR, &serror);
 		sil_scr_write(ap, SCR_ERROR, serror);
 
-		/* Trigger hotplug and accumulate SError only if the
-		 * port isn't already frozen.  Otherwise, PHY events
-		 * during hardreset makes controllers with broken SIEN
-		 * repeat probing needlessly.
+		/* Sometimes spurious interrupts occur, double check
+		 * it's PHYRDY CHG.
 		 */
-		if (!(ap->pflags & ATA_PFLAG_FROZEN)) {
-			ata_ehi_hotplugged(&ap->link.eh_info);
-			ap->link.eh_info.serror |= serror;
+		if (serror & SERR_PHYRDY_CHG) {
+			/* Trigger hotplug and accumulate SError only
+			 * if the port isn't already frozen.
+			 * Otherwise, PHY events during hardreset
+			 * makes controllers with broken SIEN repeat
+			 * probing needlessly.
+			 */
+			if (!(ap->pflags & ATA_PFLAG_FROZEN)) {
+				ata_ehi_hotplugged(&ap->link.eh_info);
+				ap->link.eh_info.serror |= serror;
+			}
+			goto freeze;
 		}
 
-		goto freeze;
+		if (!(bmdma2 & SIL_DMA_COMPLETE))
+			return;
 	}
 
-	if (unlikely(!qc))
-		goto freeze;
-
-	if (unlikely(qc->tf.flags & ATA_TFLAG_POLLING)) {
+	if (unlikely(!qc || (qc->tf.flags & ATA_TFLAG_POLLING))) {
 		/* this sometimes happens, just clear IRQ */
 		ata_chk_status(ap);
 		return;
author
#1 lore raw

PATCH 1/1 libata: clear link->eh_info.serror from ata_std_postreset()

link->eh_info.serror is used to cache SError for controllers which need it cleared from interrupt handler to clear IRQ. It also should be cleared after reset just like SError itself.

Make ata_std_postreset() clear link->eh_info.serror too and update sata_sil such that it doesn't care about bookkeeping the value.

Signed-off-by Tejun Heo <[email protected]>
---
 drivers/ata/libata-core.c |    1 +
 drivers/ata/sata_sil.c    |   11 +----------
 2 files changed, 2 insertions(+), 10 deletions(-)
drivers/ata/libata-core.c +1 -0
Index: work/drivers/ata/libata-core.c
===================================================================
@@ -3923,6 +3923,7 @@ void ata_std_postreset(struct ata_link *
 	/* clear SError */
 	if (sata_scr_read(link, SCR_ERROR, &serror) == 0)
 		sata_scr_write(link, SCR_ERROR, serror);
+	link->eh_info.serror = 0;
 
 	/* is double-select really necessary? */
 	if (classes[0] != ATA_DEV_NONE)
drivers/ata/sata_sil.c +1 -10
Index: work/drivers/ata/sata_sil.c
===================================================================
@@ -394,16 +394,7 @@ static void sil_host_intr(struct ata_por
 		 * it's PHYRDY CHG.
 		 */
 		if (serror & SERR_PHYRDY_CHG) {
-			/* Trigger hotplug and accumulate SError only
-			 * if the port isn't already frozen.
-			 * Otherwise, PHY events during hardreset
-			 * makes controllers with broken SIEN repeat
-			 * probing needlessly.
-			 */
-			if (!(ap->pflags & ATA_PFLAG_FROZEN)) {
-				ata_ehi_hotplugged(&ap->link.eh_info);
-				ap->link.eh_info.serror |= serror;
-			}
+			ap->link.eh_info.serror |= serror;
 			goto freeze;
 		}
#2 lore raw

Re: [PATCH #upstream-fixes] libata: clear link->eh_info.serror from ata_std_postreset()

Quoted 18 lines · Tejun Heo

link->eh_info.serror is used to cache SError for controllers which need it cleared from interrupt handler to clear IRQ. It also should be cleared after reset just like SError itself.

Make ata_std_postreset() clear link->eh_info.serror too and update sata_sil such that it doesn't care about bookkeeping the value.

Signed-off-by Tejun Heo <[email protected]>
---
 drivers/ata/libata-core.c |    1 +
 drivers/ata/sata_sil.c    |   11 +----------
 2 files changed, 2 insertions(+), 10 deletions(-)

Index: work/drivers/ata/libata-core.c
===================================================================

drivers/ata/libata-core.c void ata_std_postreset(struct ata_link * · line 3926

Quoted 4 lines
drivers/ata/libata-core.c +1 -0
@@ -3923,6 +3923,7 @@ void ata_std_postreset(struct ata_link *
 	/* clear SError */
 	if (sata_scr_read(link, SCR_ERROR, &serror) == 0)
 		sata_scr_write(link, SCR_ERROR, serror);
+	link->eh_info.serror = 0;

IMO it would make more sense to record the state of the hardware following sata_scr_write() than simply zeroing the cache value.

Just a gut feeling... it seems like having a manufactured value (zero) rather than the last known from-the-hardware value could lead to inconsistencies.

Comments?

Jeff

author
#3 lore raw

Re: [PATCH #upstream-fixes] libata: clear link->eh_info.serror from ata_std_postreset()

Quoted 27 lines · Jeff Garzik
Quoted 18 lines · Tejun Heo

link->eh_info.serror is used to cache SError for controllers which need it cleared from interrupt handler to clear IRQ. It also should be cleared after reset just like SError itself.

Make ata_std_postreset() clear link->eh_info.serror too and update sata_sil such that it doesn't care about bookkeeping the value.

Signed-off-by Tejun Heo <[email protected]>
---
 drivers/ata/libata-core.c |    1 +
 drivers/ata/sata_sil.c    |   11 +----------
 2 files changed, 2 insertions(+), 10 deletions(-)

Index: work/drivers/ata/libata-core.c
===================================================================

drivers/ata/libata-core.c void ata_std_postreset(struct ata_link * · line 3926

Quoted 4 lines
drivers/ata/libata-core.c +1 -0
@@ -3923,6 +3923,7 @@ void ata_std_postreset(struct ata_link *
     /* clear SError */
     if (sata_scr_read(link, SCR_ERROR, &serror) == 0)
         sata_scr_write(link, SCR_ERROR, serror);
+    link->eh_info.serror = 0;

IMO it would make more sense to record the state of the hardware following sata_scr_write() than simply zeroing the cache value.

Quoted 5 lines

Just a gut feeling... it seems like having a manufactured value (zero) rather than the last known from-the-hardware value could lead to inconsistencies.

Comments?

link->eh_info.serror is always used as complement to the hardware SError value. It's a temporary storage to dump/accumulate SError value when for whatever reason the hardware can't hold the value. link->eh_info.serror and the hardware SError value are always OR'd before being used, so no reason to dup the value.

Thanks.

Signature · 1 line
tejun
#4 lore raw

Re: [PATCH #upstream-fixes] libata: clear link->eh_info.serror from ata_std_postreset()

In reply to Tejun Heo
Quoted 4 lines · Tejun Heo

link->eh_info.serror is used to cache SError for controllers which need it cleared from interrupt handler to clear IRQ. It also should be cleared after reset just like SError itself.

Quoted 8 lines

Make ata_std_postreset() clear link->eh_info.serror too and update sata_sil such that it doesn't care about bookkeeping the value.

Signed-off-by Tejun Heo <[email protected]>
---
 drivers/ata/libata-core.c |    1 +
 drivers/ata/sata_sil.c    |   11 +----------
 2 files changed, 2 insertions(+), 10 deletions(-)

applied #upstream-fixes

#5 lore raw
In reply to Tejun Heo
Quoted 9 lines · Tejun Heo

Interestingly, sata_sil raises spurious interrupts if it's coupled with Sil SATA_PATA bridge. Currently, sata_sil interrupt handler is strict about spurious interrupts and freezes the port when it occurs. This patch makes it more forgiving.

* On SATA PHY event interrupt, serror value is checked to see whether it really is PHYRDY CHG event. If not, SATA PHY event interrupt is ignored.

Quoted 8 lines

* If ATA interrupt occurs while no command is in progress, it's cleared and ignored.

This fixes bugzilla bug 9505.

http://bugzilla.kernel.org/show_bug.cgi?id=9505

Signed-off-by Tejun Heo <[email protected]>

applied #upstream-fixes