[PATCH libata-dev-2.6] libata: flush COMRESET write

[PATCH libata-dev-2.6] libata: flush COMRESET write

[Brett Russ]

Jeff,

I'm seeing single drives getting stuck in ata_busy_sleep() with these
log messages on boot:

<4>ata2 is slow to respond, please be patient
<3>ata2 failed to respond (30 secs)

One thing I found was the bug fixed below, where the COMRESET wasn't
getting flushed properly.  This may or may not be the reason, but it
can't hurt.  I have to keep looking and testing.

I wanted to also send a separate patch to ensure the port was stopped
before issuing COMRESET, since this is required by at least the ICH6R,
but the port_stop() routines seemed too heavy to call from
__sata_phy_reset() so I'm not sending anything like that yet.

Signed-off-by: Brett Russ <russb@emc.com>

Index: libata-dev-2.6/drivers/scsi/libata-core.c
===================================================================
--- libata-dev-2.6.orig/drivers/scsi/libata-core.c	2005-03-23 16:44:48.000000000 -0500
+++ libata-dev-2.6/drivers/scsi/libata-core.c	2005-03-25 16:58:22.000000000 -0500
@@ -1262,7 +1262,7 @@
 
 	if (ap->flags & ATA_FLAG_SATA_RESET) {
 		scr_write(ap, SCR_CONTROL, 0x301); /* issue phy wake/reset */
-		scr_read(ap, SCR_STATUS);	/* dummy read; flush */
+		scr_read(ap, SCR_CONTROL);	/* dummy read; flush */
 		udelay(400);			/* FIXME: a guess */
 	}
 	scr_write(ap, SCR_CONTROL, 0x300);	/* issue phy wake/clear reset */

[PATCH libata-2.6] libata: flush COMRESET set and clear

[Brett Russ]

Updated patch to fix erroneous flush of COMRESET set and missing flush
of COMRESET clear.  Created a new routine scr_write_flush() to try to
prevent this in the future.  Also, this patch is based on libata-2.6
instead of the previous libata-dev-2.6 based patch.

Signed-off-by: Brett Russ <russb@emc.com>

Index: libata-2.6/drivers/scsi/libata-core.c
===================================================================
--- libata-2.6.orig/drivers/scsi/libata-core.c	2005-03-28 15:06:41.000000000 -0500
+++ libata-2.6/drivers/scsi/libata-core.c	2005-03-28 15:06:41.000000000 -0500
@@ -1253,11 +1253,11 @@
 	unsigned long timeout = jiffies + (HZ * 5);
 
 	if (ap->flags & ATA_FLAG_SATA_RESET) {
-		scr_write(ap, SCR_CONTROL, 0x301); /* issue phy wake/reset */
-		scr_read(ap, SCR_STATUS);	/* dummy read; flush */
+		/* issue phy wake/reset */
+		scr_write_flush(ap, SCR_CONTROL, 0x301);
 		udelay(400);			/* FIXME: a guess */
 	}
-	scr_write(ap, SCR_CONTROL, 0x300);	/* issue phy wake/clear reset */
+	scr_write_flush(ap, SCR_CONTROL, 0x300); /* phy wake/clear reset */
 
 	/* wait for phy to become ready, if necessary */
 	do {
Index: libata-2.6/include/linux/libata.h
===================================================================
--- libata-2.6.orig/include/linux/libata.h	2005-03-28 15:06:42.000000000 -0500
+++ libata-2.6/include/linux/libata.h	2005-03-28 15:06:42.000000000 -0500
@@ -584,6 +584,13 @@
 	ap->ops->scr_write(ap, reg, val);
 }
 
+static inline void scr_write_flush(struct ata_port *ap, unsigned int reg, 
+				   u32 val)
+{
+	ap->ops->scr_write(ap, reg, val);
+	(void) ap->ops->scr_read(ap, reg);
+}
+
 static inline unsigned int sata_dev_present(struct ata_port *ap)
 {
 	return ((scr_read(ap, SCR_STATUS) & 0xf) == 0x3) ? 1 : 0;

Re: [PATCH libata-2.6] libata: flush COMRESET set and clear

[Jeff Garzik]

Brett Russ wrote:
> Updated patch to fix erroneous flush of COMRESET set and missing flush
> of COMRESET clear.  Created a new routine scr_write_flush() to try to
> prevent this in the future.  Also, this patch is based on libata-2.6
> instead of the previous libata-dev-2.6 based patch.
> 
> Signed-off-by: Brett Russ <russb@emc.com>

applied