Latest git not seeing HD with sata_nv

It seems that commit 305d2a1a ("libata: unify mechanism to request follow-up SRST") causes sata_nv not to see the HD on a system I have here; with current git (80750147), I get this in my bootlog:

[    2.425108] Driver 'sd' needs updating - please use bus_type methods
[    2.435751] ACPI: PCI Interrupt Link [LSA0] enabled at IRQ 23
[    2.441508] ACPI: PCI Interrupt 0000:00:05.0[A] -> Link [LSA0] -> GSI 23 (level, low) -> IRQ 23
[    2.452929] sata_nv 0000:00:05.0: Using SWNCQ mode
[    2.457892] scsi0 : sata_nv
[    2.461813] scsi1 : sata_nv
[    2.465891] ata1: SATA max UDMA/133 cmd 0xd480 ctl 0xd400 bmdma 0xcc00 irq 23
[    2.473034] ata2: SATA max UDMA/133 cmd 0xd080 ctl 0xd000 bmdma 0xcc08 irq 23
[    2.565962] ata1: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[    3.003517] ata2: SATA link down (SStatus 0 SControl 300)
[    3.137171] ACPI: PCI Interrupt Link [LSA1] enabled at IRQ 22
[    3.142926] ACPI: PCI Interrupt 0000:00:05.1[B] -> Link [LSA1] -> GSI 22 (level, low) -> IRQ 22
[    3.151675] sata_nv 0000:00:05.1: Using SWNCQ mode
[    3.156582] scsi2 : sata_nv
[    3.160553] scsi3 : sata_nv

ie no HD is found, even thought he ata1 link comes up, which of course leads to:

[    4.763274] VFS: Cannot open root device "sda1" or unknown-block(0,0)
[    4.769724] Please append a correct "root=" boot option; here are the available partitions:

and the system is dead.

I did a bisection and came up with 305d2a1a as the culprit. Then I applied the revert patch below by hand (since 305d2a1a doesn't revert entirely cleanly), and things worked again:

index b0b00af..e0321ce 100644
@@ -3683,6 +3683,7 @@ int sata_std_hardreset(struct ata_link *link, unsigned int *class,
 
 	/* do hardreset */
 	rc = sata_link_hardreset(link, timing, deadline, &online, NULL);
+	*class = ATA_DEV_NONE;
 	return online ? -EAGAIN : rc;
 }
 

index d94359a..667d0e6 100644
@@ -2053,10 +2053,20 @@ static int ata_do_reset(struct ata_link *link, ata_reset_fn_t reset,
 
 	rc = reset(link, classes, deadline);
 
-	/* convert all ATA_DEV_UNKNOWN to ATA_DEV_NONE */
+	/* If any class isn't ATA_DEV_UNKNOWN, consider classification
+	 * is complete and convert all ATA_DEV_UNKNOWN to
+	 * ATA_DEV_NONE.
+	 */
 	ata_link_for_each_dev(dev, link)
-		if (classes[dev->devno] == ATA_DEV_UNKNOWN)
-			classes[dev->devno] = ATA_DEV_NONE;
+		if (classes[dev->devno] != ATA_DEV_UNKNOWN)
+			break;
+
+	if (dev) {
+		ata_link_for_each_dev(dev, link) {
+			if (classes[dev->devno] == ATA_DEV_UNKNOWN)
+				classes[dev->devno] = ATA_DEV_NONE;
+		}
+	}
 
 	return rc;
 }

@@ -2067,15 +2077,16 @@ static int ata_eh_followup_srst_needed(struct ata_link *link,
 {
 	if ((link->flags & ATA_LFLAG_NO_SRST) || ata_link_offline(link))
 		return 0;
-	if (rc == -EAGAIN) {
-		if (classify)
-			return 1;
-		rc = 0;
-	}
+	if (rc == -EAGAIN)
+		return 1;
 	if (rc != 0)
 		return 0;
 	if (sata_pmp_supported(link->ap) && ata_is_host_link(link))
 		return 1;
+	if (classify && !(link->flags & ATA_LFLAG_ASSUME_CLASS) &&
+	    classes[0] == ATA_DEV_UNKNOWN)
+		return 1;
+
 	return 0;
 }
 

index 1549952..11b2adc 100644
@@ -1943,6 +1943,7 @@ int sata_sff_hardreset(struct ata_link *link, unsigned int *class,
 
 	rc = sata_link_hardreset(link, timing, deadline, &online,
 				 ata_sff_check_ready);
+	*class = ATA_DEV_NONE;
 	if (online)
 		*class = ata_sff_dev_classify(link->device, 1, NULL);
 

And here's the bootlog with current git with my revert patch applied:

[    2.417741] Driver 'sd' needs updating - please use bus_type methods
[    2.424142] sata_nv 0000:00:05.0: version 3.5
[    2.428387] ACPI: PCI Interrupt Link [LSA0] enabled at IRQ 23
[    2.434134] ACPI: PCI Interrupt 0000:00:05.0[A] -> Link [LSA0] -> GSI 23 (level, low) -> IRQ 23
[    2.445558] sata_nv 0000:00:05.0: Using SWNCQ mode
[    2.450390] PCI: Setting latency timer of device 0000:00:05.0 to 64
[    2.450526] scsi0 : sata_nv
[    2.454445] scsi1 : sata_nv
[    2.458523] ata1: SATA max UDMA/133 cmd 0xd480 ctl 0xd400 bmdma 0xcc00 irq 23
[    2.465664] ata2: SATA max UDMA/133 cmd 0xd080 ctl 0xd000 bmdma 0xcc08 irq 23
[    2.923281] ata1: SATA link up 3.0 Gbps (SStatus 123 SControl 300)
[    2.966595] ata1.00: ATA-7: WDC WD3200YS-01PGB0, 21.00M21, max UDMA/133
[    2.977218] ata1.00: 625142448 sectors, multi 16: LBA48 NCQ (depth 1)
[    3.008489] ata1.00: configured for UDMA/133
[    3.289737] ata2: SATA link down (SStatus 0 SControl 300)
[    3.341584] isa bounce pool size: 16 pages
[    3.349571] scsi 0:0:0:0: Direct-Access     ATA      WDC WD3200YS-01P 21.0 PQ: 0 ANSI: 5
[    3.357780] sd 0:0:0:0: [sda] 625142448 512-byte hardware sectors (320073 MB)
[    3.364921] sd 0:0:0:0: [sda] Write Protect is off
[    3.369717] sd 0:0:0:0: [sda] Mode Sense: 00 3a 00 00
[    3.369732] sd 0:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
[    3.378822] sd 0:0:0:0: [sda] 625142448 512-byte hardware sectors (320073 MB)
[    3.385964] sd 0:0:0:0: [sda] Write Protect is off
[    3.390761] sd 0:0:0:0: [sda] Mode Sense: 00 3a 00 00
[    3.390776] sd 0:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
[    3.399816]  sda: sda1 sda2 < sda5 sda6 >
[    3.445543] sd 0:0:0:0: [sda] Attached SCSI disk
[    3.450580] ACPI: PCI Interrupt Link [LSA1] enabled at IRQ 22
[    3.457951] ACPI: PCI Interrupt 0000:00:05.1[B] -> Link [LSA1] -> GSI 22 (level, low) -> IRQ 22
[    3.466708] sata_nv 0000:00:05.1: Using SWNCQ mode
[    3.471529] PCI: Setting latency timer of device 0000:00:05.1 to 64

good. my test system had the same problem. just want to start the bisect

other things in the bisect list
1. clocktest is broken.
2. kexec RHEL 5.1 the mpt sas only can support 2 cards...

YH

I wonder if this is related to the code

static int nv_hardreset(struct ata_link *link, unsigned int *class,
unsigned long deadline)

{
	unsigned int dummy;

	/* SATA hardreset fails to retrieve proper device signature on
	 * some controllers.  Don't classify on hardreset.  For more
	 * info, see http://bugzilla.kernel.org/show_bug.cgi?id=3352
	 */
	return sata_sff_hardreset(link, &dummy, deadline);
}

It seems especially suspicious given that 305d2a1a touches sata_sff_hardreset...

- R.

FWIW, the patch below also makes my box start working again... I have no idea what the right fix is, given that bug 3352 mostly seems to be about nforce2 and nforce3, and my box has a more modern nvidia chipset:

    00:04.0 IDE interface [0101]: nVidia Corporation MCP55 IDE [10de:036e] (rev a1)

Anyway, it seems 305d2a1a broke the logic that nv_hardreset() relies on when it doesn't let sata_sff_hardreset() set class.

index 109b074..d0d85a3 100644
@@ -1597,7 +1597,7 @@ static int nv_hardreset(struct ata_link *link, unsigned int *class,
 	 * some controllers.  Don't classify on hardreset.  For more
 	 * info, see http://bugzilla.kernel.org/show_bug.cgi?id=3352
 	 */
-	return sata_sff_hardreset(link, &dummy, deadline);
+	return sata_sff_hardreset(link, class, deadline);
 }
 
 static void nv_adma_error_handler(struct ata_port *ap)

Ah... okay now I see it. I neglected whole class of hardresets which was using &dummy as class. I'll verify and prep a patch soon. Thanks.

tejun

Thanks for the bisection. Hmmm... it's not supposed to change probe behavior (well, of course). I'll try to reproduce the problem here and report back.

tejun