Skip to contentSkip to conversation

pata_hpt366: fix clock detection

PATCH In review Tejun Heo 4 participants 9 messages Dec 8, 2008 – Dec 9, 2008 libata

lore
Share & tools
lore

Tejun Heo

author
#0 lore raw

PATCH 1/3 pata_hpt366: fix clock detection

pata_hpt366 had its clock detection wrong and detected 25Mhz as 40Mhz and vice-versa. Fix it.

---
 drivers/ata/pata_hpt366.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/ata/pata_hpt366.c +2 -2
Index: work/drivers/ata/pata_hpt366.c
===================================================================
@@ -382,10 +382,10 @@ static int hpt36x_init_one(struct pci_de
 	/* PCI clocking determines the ATA timing values to use */
 	/* info_hpt366 is safe against re-entry so we can scribble on it */
 	switch((reg1 & 0x700) >> 8) {
-		case 5:
+		case 9:
 			hpriv = &hpt366_40;
 			break;
-		case 9:
+		case 5:
 			hpriv = &hpt366_25;
 			break;
 		default:
author
#1 lore raw

PATCH 2/3 pata_hpt366: fix cable detection

hpt366 is strange in that its cable detection register uses the higher bit for master, so the testing should be 2 >> port_no instead of 1 << port_no. Fix it.

Info provided by Alan Cox.

---
 drivers/ata/pata_hpt366.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/ata/pata_hpt366.c +1 -1
Index: work/drivers/ata/pata_hpt366.c
===================================================================
@@ -215,7 +215,7 @@ static int hpt36x_cable_detect(struct at
 	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 
 	pci_read_config_byte(pdev, 0x5A, &ata66);
-	if (ata66 & (1 << ap->port_no))
+	if (ata66 & (2 >> ap->port_no))
 		return ATA_CBL_PATA40;
 	return ATA_CBL_PATA80;
 }
author
#2 lore raw

PATCH 3/3 pata_hpt366: update mode programming

Update mode programming logic of pata_hpt366 such that it's identical to that of IDE hpt366 driver. The differences were...

* pata_hpt366 used 0xCFFF8FFFF to mask pio modes and 0x3FFFFFFF dma modes. IDE hpt366 uses 0xC1F8FFFF for PIO, 0x303800FF for MWDMA and 0x30070000 for UDMA.

* pata_hpt366 doesn't set 0x08000000 for PIO unless it's already set and always turns it on for MWDMA/UDMA. IDE hpt366 doesn't bother with the bit. It always uses what was there.

* IDE hpt366 always clears 0xC0000000. pata_hpt366 doesn't.

---

Alan, can you please ack this if you think it's okay. Jeff, please don't commit till Alan acks. Thanks.

 drivers/ata/pata_hpt366.c |   50 ++++++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 21 deletions(-)
drivers/ata/pata_hpt366.c +29 -21
Index: work/drivers/ata/pata_hpt366.c
===================================================================
@@ -188,25 +188,41 @@ static unsigned long hpt366_filter(struc
 }
 
 /**
- *	hpt36x_find_mode	-	reset the hpt36x bus
+ *	hpt36x_calc_timing	-	calculate timing value for transfer mode
  *	@ap: ATA port
- *	@speed: transfer mode
+ *	@cur_timing: current timing value
+ *	@speed: target transfer mode
  *
  *	Return the 32bit register programming information for this channel
  *	that matches the speed provided.
  */
 
-static u32 hpt36x_find_mode(struct ata_port *ap, int speed)
+static u32 hpt36x_calc_timing(struct ata_port *ap, u32 cur_timing, int speed)
 {
 	struct hpt_clock *clocks = ap->host->private_data;
+	u32 mask;
 
-	while(clocks->xfer_speed) {
+	if (speed < XFER_MW_DMA_0)
+		mask = 0xc1f8ffff;
+	else if (speed < XFER_UDMA_0)
+		mask = 0x303800ff;
+	else
+		mask = 0x30070000;
+
+	while (clocks->xfer_speed) {
 		if (clocks->xfer_speed == speed)
-			return clocks->timing;
+			break;
 		clocks++;
 	}
-	BUG();
-	return 0xffffffffU;	/* silence compiler warning */
+	if (!clocks->xfer_speed)
+		BUG();
+
+	/*
+	 * Combine new mode bits with old config bits and disable
+	 * on-chip PIO FIFO/buffer (and PIO MST mode as well) to avoid
+	 * problems handling I/O errors later.
+	 */
+	return ((cur_timing & ~mask) | (clocks->timing & mask)) & ~0xc0000000;
 }
 
 static int hpt36x_cable_detect(struct ata_port *ap)
@@ -232,8 +248,7 @@ static void hpt366_set_piomode(struct at
 {
 	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 	u32 addr1, addr2;
-	u32 reg;
-	u32 mode;
+	u32 reg, timing;
 	u8 fast;
 
 	addr1 = 0x40 + 4 * (adev->devno + 2 * ap->port_no);
@@ -247,11 +262,8 @@ static void hpt366_set_piomode(struct at
 	}
 
 	pci_read_config_dword(pdev, addr1, &reg);
-	mode = hpt36x_find_mode(ap, adev->pio_mode);
-	mode &= ~0x8000000;	/* No FIFO in PIO */
-	mode &= ~0x30070000;	/* Leave config bits alone */
-	reg &= 0x30070000;	/* Strip timing bits */
-	pci_write_config_dword(pdev, addr1, reg | mode);
+	timing = hpt36x_calc_timing(ap, reg, adev->pio_mode);
+	pci_write_config_dword(pdev, addr1, timing);
 }
 
 /**
@@ -267,8 +279,7 @@ static void hpt366_set_dmamode(struct at
 {
 	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 	u32 addr1, addr2;
-	u32 reg;
-	u32 mode;
+	u32 reg, timing;
 	u8 fast;
 
 	addr1 = 0x40 + 4 * (adev->devno + 2 * ap->port_no);
@@ -282,11 +293,8 @@ static void hpt366_set_dmamode(struct at
 	}
 
 	pci_read_config_dword(pdev, addr1, &reg);
-	mode = hpt36x_find_mode(ap, adev->dma_mode);
-	mode |= 0x8000000;	/* FIFO in MWDMA or UDMA */
-	mode &= ~0xC0000000;	/* Leave config bits alone */
-	reg &= 0xC0000000;	/* Strip timing bits */
-	pci_write_config_dword(pdev, addr1, reg | mode);
+	timing = hpt36x_calc_timing(ap, reg, adev->dma_mode);
+	pci_write_config_dword(pdev, addr1, timing);
 }
 
 static struct scsi_host_template hpt36x_sht = {
Acked-by on v1 2/3
#3 lore raw

Re: [PATCH #upstream-fixes 2/3] pata_hpt366: fix cable detection

In reply to Tejun Heo

Hello.

Quoted 3 lines · Tejun Heo

hpt366 is strange in that its cable detection register uses the higher bit for master, so the testing should be 2 >> port_no instead of 1 <<

For primary -- the cable bits are per-channel, not per-device.

Quoted 6 lines

port_no. Fix it.

Info provided by Alan Cox.

drivers/ata/pata_hpt366.c static int hpt36x_cable_detect(struct at · line 219

Quoted 11 lines

Index: work/drivers/ata/pata_hpt366.c
===================================================================

drivers/ata/pata_hpt366.c +1 -1
@@ -215,7 +215,7 @@ static int hpt36x_cable_detect(struct at
 	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 
 	pci_read_config_byte(pdev, 0x5A, &ata66);
-	if (ata66 & (1 << ap->port_no))
+	if (ata66 & (2 >> ap->port_no))
 

HPT36x are single channel per function, so the shift can be removed.

MBR, Sergei

#4 lore raw

Re: [PATCH #upstream-fixes 2/3] pata_hpt366: fix cable detection

Quoted 6 lines

drivers/ata/pata_hpt366.c static int hpt36x_cable_detect(struct at · line 218

Quoted 4 lines
drivers/ata/pata_hpt366.c +1 -1
 	pci_read_config_byte(pdev, 0x5A, &ata66);
-	if (ata66 & (1 << ap->port_no))
+	if (ata66 & (2 >> ap->port_no))

HPT36x are single channel per function, so the shift can be removed.

In the case of two channels are you sure the bits appear as bit 2 in both cases ?

Alan

#5 lore raw

Re: [PATCH #upstream-fixes 2/3] pata_hpt366: fix cable detection

Hello.

Quoted 8 lines · Alan Cox
Quoted 5 lines

drivers/ata/pata_hpt366.c static int hpt36x_cable_detect(struct at · line 218

Quoted 3 lines
	pci_read_config_byte(pdev, 0x5A, &ata66);
drivers/ata/pata_hpt366.c +1 -1
-	if (ata66 & (1 << ap->port_no))
+	if (ata66 & (2 >> ap->port_no))

HPT36x are single channel per function, so the shift can be removed.

In the case of two channels are you sure the bits appear as bit 2 in both cases ?

Oops, they don't. The cable select register (0x5a) seems to be shared by
both functions (with function 0 representing "primary" and the function 1 "secondary" channel). But in this case, the logic remains broken -- I don't think that function #1 has its ap->port_no set to 1. The IDE driver appears broken as well then. Damn Highpoint for their brain damaged design...

Quoted 1 line

Alan

MBR, Sergei

#6 lore raw

Re: [PATCH #upstream-fixes 2/3] pata_hpt366: fix cable detection

In reply to Sergei Shtylyov
Quoted 5 lines · Hello, I wrote:
Quoted 2 lines

hpt366 is strange in that its cable detection register uses the higher bit for master, so the testing should be 2 >> port_no instead of 1 <<

For primary -- the cable bits are per-channel, not per-device.

Quoted 7 lines
Quoted 5 lines

port_no. Fix it.

Info provided by Alan Cox.

I have to NAK this. HPT366 design is just too brain damaged for both IDE
and libata drivers to get it right so far.

Quoted 7 lines
Quoted 7 lines

Index: work/drivers/ata/pata_hpt366.c
===================================================================

drivers/ata/pata_hpt366.c +0 -0
@@ -215,7 +215,7 @@ static int hpt36x_cable_detect(struct at
     struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 
Quoted 6 lines

drivers/ata/pata_hpt366.c static int hpt36x_cable_detect(struct at

Quoted 4 lines
drivers/ata/pata_hpt366.c +1 -1
     pci_read_config_byte(pdev, 0x5A, &ata66);
-    if (ata66 & (1 << ap->port_no))
+    if (ata66 & (2 >> ap->port_no))
 

HPT36x are single channel per function, so the shift can be removed.

It should be replaced by more sophisticated logic for function 1 in fact...

MBR, Sergei

author
#7 lore raw

Re: [PATCH #upstream-fixes 2/3] pata_hpt366: fix cable detection

Quoted 32 lines · Sergei Shtylyov
Quoted 5 lines · Hello, I wrote:
Quoted 2 lines

hpt366 is strange in that its cable detection register uses the higher bit for master, so the testing should be 2 >> port_no instead of 1 <<

For primary -- the cable bits are per-channel, not per-device.

Quoted 7 lines
Quoted 5 lines

port_no. Fix it.

Info provided by Alan Cox.

I have to NAK this. HPT366 design is just too brain damaged for both
IDE and libata drivers to get it right so far.

Quoted 7 lines
Quoted 7 lines

Index: work/drivers/ata/pata_hpt366.c
===================================================================

drivers/ata/pata_hpt366.c +0 -0
@@ -215,7 +215,7 @@ static int hpt36x_cable_detect(struct at
     struct pci_dev *pdev = to_pci_dev(ap->host->dev);
 
Quoted 6 lines

drivers/ata/pata_hpt366.c static int hpt36x_cable_detect(struct at

Quoted 4 lines
drivers/ata/pata_hpt366.c +1 -1
     pci_read_config_byte(pdev, 0x5A, &ata66);
-    if (ata66 & (1 << ap->port_no))
+    if (ata66 & (2 >> ap->port_no))
 

HPT36x are single channel per function, so the shift can be removed.

Quoted 2 lines

It should be replaced by more sophisticated logic for function 1 in
fact...

Ah... crap. I don't know anything about this controller. Care to post the correct patch? I'll be happy to test.

Thanks.

Signature · 1 line
tejun
#8 lore raw
In reply to Tejun Heo
Quoted 13 lines · Tejun Heo

pata_hpt366 had its clock detection wrong and detected 25Mhz as 40Mhz and vice-versa. Fix it.

---
 drivers/ata/pata_hpt366.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: work/drivers/ata/pata_hpt366.c
===================================================================

drivers/ata/pata_hpt366.c static int hpt36x_init_one(struct pci_de · line 389

Quoted 10 lines
drivers/ata/pata_hpt366.c +2 -2
@@ -382,10 +382,10 @@ static int hpt36x_init_one(struct pci_de
 	/* PCI clocking determines the ATA timing values to use */
 	/* info_hpt366 is safe against re-entry so we can scribble on it */
 	switch((reg1 & 0x700) >> 8) {
-		case 5:
+		case 9:
 			hpriv = &hpt366_40;
 			break;
-		case 9:
+		case 5:
 			hpriv = &hpt366_25;

applied #upstream-fixes

Avoiding patch #2 in this series b/c the thread seemed to indicate it needed more work

Avoiding patch #3 in this series awaiting Alan's ack (did I miss it?)