Skip to contentSkip to conversation

libata-dev: make the returned err_mask more accurate

PATCH/RFC Albert Lee 2 participants 4 messages Apr 6, 2006

Subsystems (2)
lore
Share & tools
lore

Albert Lee

author
#0 lore raw

make the returned err_mask more accurate
Changes:
- add ac_err_drq() and ac_err_idle() to map err_mask from drv_status for data xfer and idle states.
- fix ata_hsm_move() to use the functions for err_mask mapping.

---

The current err_mask returned by HSM doesn't look right.
e.g. When we found DRQ=0 when data transfer is expected, AC_ERR_HSM is returned. However, DRQ=0 maybe caused by device error. So, during data transfer, if we see DRQ=0 and ERR=1, returning AC_ERR_DEV should be more accurate.

e.g. In HSM_ST_LAST state, AC_ERR_OTHER is returned for BSY=0 DRQ=1 currently. AC_ERR_HSM shoule be more accurate.

Patch against irq-pio (79fa1b677be3a985cc66b9218a4dd09818f1051b). For your review, thanks.

drivers/scsi/libata-core.c +11 -8
@@ -3906,7 +3906,7 @@ fsm_start:
 		/* check device status */
 		if (unlikely((status & (ATA_BUSY | ATA_DRQ)) != ATA_DRQ)) {
 			/* Wrong status. Let EH handle this */
-			qc->err_mask |= AC_ERR_HSM;
+			qc->err_mask |= ac_err_drq(status);
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}
@@ -3920,7 +3920,7 @@ fsm_start:
 		if (unlikely(status & (ATA_ERR | ATA_DF))) {
 			printk(KERN_WARNING "ata%d: DRQ=1 with device error, dev_stat 0x%X\n",
 			       ap->id, status);
-			qc->err_mask |= AC_ERR_DEV;
+			qc->err_mask |= AC_ERR_HSM;
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}
@@ -3976,7 +3976,7 @@ fsm_start:
 			if (unlikely(status & (ATA_ERR | ATA_DF))) {
 				printk(KERN_WARNING "ata%d: DRQ=1 with device error, dev_stat 0x%X\n",
 				       ap->id, status);
-				qc->err_mask |= AC_ERR_DEV;
+				qc->err_mask |= AC_ERR_HSM;
 				ap->hsm_task_state = HSM_ST_ERR;
 				goto fsm_start;
 			}
@@ -3991,7 +3991,7 @@ fsm_start:
 			/* ATA PIO protocol */
 			if (unlikely((status & ATA_DRQ) == 0)) {
 				/* handle BSY=0, DRQ=0 as error */
-				qc->err_mask |= AC_ERR_HSM;
+				qc->err_mask |= ac_err_drq(status);
 				ap->hsm_task_state = HSM_ST_ERR;
 				goto fsm_start;
 			}
@@ -4007,13 +4007,16 @@ fsm_start:
 			 * transferred to the device.
 			 */
 			if (unlikely(status & (ATA_ERR | ATA_DF))) {
-				/* data might be corrputed */
-				qc->err_mask |= AC_ERR_DEV;
 
-				if (!(qc->tf.flags & ATA_TFLAG_WRITE)) {
+				if (qc->tf.flags & ATA_TFLAG_WRITE)
+					qc->err_mask |= AC_ERR_HSM;
+				else {
+					/* data might be corrputed */
+					qc->err_mask |= AC_ERR_DEV;
 					ata_pio_sectors(qc);
 					ata_altstatus(ap);
 					status = ata_wait_idle(ap);
+					qc->err_mask |= ac_err_idle(status);
 				}
 
 				/* ata_pio_sectors() might change the
@@ -4041,7 +4044,7 @@ fsm_start:
 
 	case HSM_ST_LAST:
 		if (unlikely(!ata_ok(status))) {
-			qc->err_mask |= __ac_err_mask(status);
+			qc->err_mask |= ac_err_idle(status);
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}
include/linux/libata.h +40 -0
@@ -942,6 +942,46 @@ static inline int ata_try_flush_cache(co
 	       ata_id_has_flush_ext(dev->id);
 }
 
+static inline unsigned int ac_err_drq(u8 status)
+{
+	/* BSY = 1*/
+	if (status & ATA_BUSY)
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 1 */
+	if (status & ATA_DRQ)
+		return 0;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+
+	/* BSY = 0, DRQ = 0, no device err
+	 * when we are expecting data transfer.
+	 */
+	return AC_ERR_HSM;
+}
+
+static inline unsigned int ac_err_idle(u8 status)
+{
+	/* BSY = 1 or DRQ = 1 */
+	if (status & (ATA_BUSY | ATA_DRQ))
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+
+	/* BSY = 0, DRQ = 0, no device err. DRDY = 1 */
+	if (status & ATA_DRDY)
+		return 0;
+
+	/* thing looks good.
+	 * don't know why DRDY not set.
+	 */
+	return AC_ERR_OTHER
+}
+
 static inline unsigned int ac_err_mask(u8 status)
 {
 	if (status & ATA_BUSY)
#1 lore raw

Hello, Albert.

On Thu, Apr 06, 2006 at 03:02:36PM +0800, Albert Lee wrote:
[-- snip --]

Quoted 36 lines
include/linux/libata.h +30 -0
@@ -942,6 +942,46 @@ static inline int ata_try_flush_cache(co
 	       ata_id_has_flush_ext(dev->id);
 }
 
+static inline unsigned int ac_err_drq(u8 status)
+{
+	/* BSY = 1*/
+	if (status & ATA_BUSY)
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 1 */
+	if (status & ATA_DRQ)
+		return 0;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+
+	/* BSY = 0, DRQ = 0, no device err
+	 * when we are expecting data transfer.
+	 */
+	return AC_ERR_HSM;
+}
+
+static inline unsigned int ac_err_idle(u8 status)
+{
+	/* BSY = 1 or DRQ = 1 */
+	if (status & (ATA_BUSY | ATA_DRQ))
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+

include/linux/libata.h static inline int ata_try_flush_cache(co · line 984

Quoted 10 lines
include/linux/libata.h +10 -0
+	/* BSY = 0, DRQ = 0, no device err. DRDY = 1 */
+	if (status & ATA_DRDY)
+		return 0;
+
+	/* thing looks good.
+	 * don't know why DRDY not set.
+	 */
+	return AC_ERR_OTHER
+}
+

I don't think putting these two functions into libata.h as inlines is a good idea. As currently there are no users, it would be better to put them where they are used as static functions and export it later as necessary.

Signature · 1 line
tejun
author
#2 lore raw
Quoted 6 lines · Tejun Heo

I don't think putting these two functions into libata.h as inlines is a good idea. As currently there are no users, it would be better to put them where they are used as static functions and export it later as necessary.

You are right. Will revise it.

author
#3 lore raw

[PATCH/RFC] libata-dev: make the returned err_mask more accurate (revised)

In reply to Tejun Heo

make the returned err_mask more accurate
Changes:
- add ac_err_drq() and ac_err_idle() to map err_mask from drv_status for data xfer and idle states.
- fix ata_hsm_move() to use the functions for err_mask mapping.

---
(revised to not inline per Tejun's comments.)

The current err_mask returned by HSM doesn't look right.
e.g. When we found DRQ=0 when data transfer is expected, AC_ERR_HSM is returned. However, DRQ=0 maybe caused by device error. So, during data transfer, if we see DRQ=0 and ERR=1, returning AC_ERR_DEV should be more accurate.

e.g. In HSM_ST_LAST state, AC_ERR_OTHER is returned for BSY=0 DRQ=1 currently. AC_ERR_HSM shoule be more accurate.

Patch against irq-pio (79fa1b677be3a985cc66b9218a4dd09818f1051b). For your review, thanks.

drivers/scsi/libata-core.c +67 -8
@@ -3838,6 +3838,62 @@ err_out:
 }
 
 /**
+ *	ac_err_drq - check status when PIO data transfer expected.
+ *	@status: device status
+ *
+ *	RETURNS:
+ *	0 if status looks good, err_mask otherwise.
+ */
+
+static unsigned int ac_err_drq(u8 status)
+{
+	/* BSY = 1*/
+	if (status & ATA_BUSY)
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 1 */
+	if (status & ATA_DRQ)
+		return 0;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+
+	/* BSY = 0, DRQ = 0, no device err.
+	 * but we are expecting data transfer...
+	 */
+	return AC_ERR_HSM;
+}
+
+/**
+ *	ac_err_idle - check status when the transaction is finished.
+ *	@status: device status
+ *
+ *	RETURNS:
+ *	0 if status looks good, err_mask otherwise.
+ */
+
+static unsigned int ac_err_idle(u8 status)
+{
+	/* BSY = 1 or DRQ = 1 */
+	if (status & (ATA_BUSY | ATA_DRQ))
+		return AC_ERR_HSM;
+
+	/* BSY = 0, DRQ = 0, device err */
+	if (status & (ATA_ERR | ATA_DF))
+		return AC_ERR_DEV;
+
+	/* BSY = 0, DRQ = 0, no device err. DRDY = 1 */
+	if (status & ATA_DRDY)
+		return 0;
+
+	/* thing looks good.
+	 * don't know why DRDY not set.
+	 */
+	return AC_ERR_OTHER;
+}
+
+/**
  *	ata_hsm_ok_in_wq - Check if the qc can be handled in the workqueue.
  *	@ap: the target ata_port
  *	@qc: qc on going
@@ -3906,7 +3962,7 @@ fsm_start:
 		/* check device status */
 		if (unlikely((status & (ATA_BUSY | ATA_DRQ)) != ATA_DRQ)) {
 			/* Wrong status. Let EH handle this */
-			qc->err_mask |= AC_ERR_HSM;
+			qc->err_mask |= ac_err_drq(status);
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}
@@ -3920,7 +3976,7 @@ fsm_start:
 		if (unlikely(status & (ATA_ERR | ATA_DF))) {
 			printk(KERN_WARNING "ata%d: DRQ=1 with device error, dev_stat 0x%X\n",
 			       ap->id, status);
-			qc->err_mask |= AC_ERR_DEV;
+			qc->err_mask |= AC_ERR_HSM;
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}
@@ -3976,7 +4032,7 @@ fsm_start:
 			if (unlikely(status & (ATA_ERR | ATA_DF))) {
 				printk(KERN_WARNING "ata%d: DRQ=1 with device error, dev_stat 0x%X\n",
 				       ap->id, status);
-				qc->err_mask |= AC_ERR_DEV;
+				qc->err_mask |= AC_ERR_HSM;
 				ap->hsm_task_state = HSM_ST_ERR;
 				goto fsm_start;
 			}
@@ -3991,7 +4047,7 @@ fsm_start:
 			/* ATA PIO protocol */
 			if (unlikely((status & ATA_DRQ) == 0)) {
 				/* handle BSY=0, DRQ=0 as error */
-				qc->err_mask |= AC_ERR_HSM;
+				qc->err_mask |= ac_err_drq(status);
 				ap->hsm_task_state = HSM_ST_ERR;
 				goto fsm_start;
 			}
@@ -4007,13 +4063,16 @@ fsm_start:
 			 * transferred to the device.
 			 */
 			if (unlikely(status & (ATA_ERR | ATA_DF))) {
-				/* data might be corrputed */
-				qc->err_mask |= AC_ERR_DEV;
 
-				if (!(qc->tf.flags & ATA_TFLAG_WRITE)) {
+				if (qc->tf.flags & ATA_TFLAG_WRITE)
+					qc->err_mask |= AC_ERR_HSM;
+				else {
+					/* data might be corrputed */
+					qc->err_mask |= AC_ERR_DEV;
 					ata_pio_sectors(qc);
 					ata_altstatus(ap);
 					status = ata_wait_idle(ap);
+					qc->err_mask |= ac_err_idle(status);
 				}
 
 				/* ata_pio_sectors() might change the
@@ -4041,7 +4100,7 @@ fsm_start:
 
 	case HSM_ST_LAST:
 		if (unlikely(!ata_ok(status))) {
-			qc->err_mask |= __ac_err_mask(status);
+			qc->err_mask |= ac_err_idle(status);
 			ap->hsm_task_state = HSM_ST_ERR;
 			goto fsm_start;
 		}