OHCI: add PowerPC 440EP/440EPx support

PATCH In review Valentine Barshak 4 participants 11 messages Oct 8, 2007

Subsystems (2)

USB USB ohci driver

lore

Share & tools

lore

Valentine Barshak

author

2007-10-08 14:27 UTC #0 lore raw

PATCH 1/1 OHCI: add PowerPC 440EP/440EPx support

PowerPC 440EP 440EPx OHCI controller needs frame number value shift.

Signed-off-by Valentine Barshak <[email protected]>

---
 drivers/usb/host/ohci.h |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

drivers/usb/host/ohci.h +1 -1

diff -pruN linux-2.6.orig/drivers/usb/host/ohci.h linux-2.6/drivers/usb/host/ohci.h
@@ -612,7 +612,7 @@ static inline u32 hc32_to_cpup (const st
  * to arch/powerpc
  */
 
-#ifdef CONFIG_STB03xxx
+#if defined(CONFIG_STB03xxx) || defined(CONFIG_440EP) || defined(CONFIG_440EPX)
 #define OHCI_BE_FRAME_NO_SHIFT	16
 #else
 #define OHCI_BE_FRAME_NO_SHIFT	0

David Brownell

2007-10-08 15:01 UTC #1 lore raw

drivers/usb/host/ohci.h static inline u32 hc32_to_cpup (const st · line 618

Quoted 9 lines

drivers/usb/host/ohci.h +1 -1

@@ -612,7 +612,7 @@ static inline u32 hc32_to_cpup (const st
  * to arch/powerpc
  */
 
-#ifdef CONFIG_STB03xxx
+#if defined(CONFIG_STB03xxx) || defined(CONFIG_440EP) || defined(CONFIG_440EPX)
 #define OHCI_BE_FRAME_NO_SHIFT	16
 #else
 #define OHCI_BE_FRAME_NO_SHIFT	0

Near as I can tell, the original code is wrong ... the hcca->frame_no byte offset is fully specified, so that shift should always be 16.

- Dave

Valentine Barshak

author

2007-10-08 16:08 UTC #2 lore raw

Quoted 13 lines · David Brownell

drivers/usb/host/ohci.h static inline u32 hc32_to_cpup (const st · line 618
Quoted 9 lines
drivers/usb/host/ohci.h +1 -1
@@ -612,7 +612,7 @@ static inline u32 hc32_to_cpup (const st
  * to arch/powerpc
  */
 
-#ifdef CONFIG_STB03xxx
+#if defined(CONFIG_STB03xxx) || defined(CONFIG_440EP) || defined(CONFIG_440EPX)
 #define OHCI_BE_FRAME_NO_SHIFT	16
 #else
 #define OHCI_BE_FRAME_NO_SHIFT	0
Near as I can tell, the original code is wrong ... the hcca->frame_no byte offset is fully specified, so that shift should always be 16.

Quoted 2 lines

- Dave

Are you saying that it should always be #define OHCI_BE_FRAME_NO_SHIFT 16 for big endian platforms?

David Brownell

2007-10-08 16:19 UTC #3 lore raw

Quoted 5 lines

Quoted 2 lines
Near as I can tell, the original code is wrong ... the hcca->frame_no byte offset is fully specified, so that shift should always be 16.
Are you saying that it should always be #define OHCI_BE_FRAME_NO_SHIFT 16 for big endian platforms?

More than that, I'm saying that shouldn't even be a #define! The default should be that drivers expect chips to follow the interface specs. Any value other than 16 violates the OHCI spec.

However, based on one other post, I suspect at least one Freescale part will need to declare a chip quirk for this case.

- Dave

Valentine Barshak

author

2007-10-08 16:55 UTC #4 lore raw

PATCH 1/1 USB: Add frame_no big endian OHCI quirk.

Add OHCI big endian frame_no quirk.
The frame_no value stored in the HCCA is a 16 bit field at a specific offset, but since not all CPUs can do 16-bit memory accesses it's used as a 32 bit field.
And that's why big-endian OHCI will shift 16 bits, unless the spec is not followed. Currently there's one MPC52xx platform that doesn't need the shift. This case is hanled using big endian frame_no quirk.

Signed-off-by Valentine Barshak <[email protected]>

---
 drivers/usb/host/ohci-ppc-of.c |    5 ++++-
 drivers/usb/host/ohci.h        |   13 ++++++-------
 2 files changed, 10 insertions(+), 8 deletions(-)

drivers/usb/host/ohci.h +6 -7

diff -pruN linux-2.6.orig/drivers/usb/host/ohci.h linux-2.6/drivers/usb/host/ohci.h
@@ -398,6 +398,7 @@ struct ohci_hcd {
 #define	OHCI_QUIRK_BE_MMIO	0x10			/* BE registers */
 #define	OHCI_QUIRK_ZFMICRO	0x20			/* Compaq ZFMicro chipset*/
 #define	OHCI_QUIRK_NEC		0x40			/* lost interrupts */
+#define	OHCI_QUIRK_FRAME_NO	0x80			/* no big endian frame_no shift */
 	// there are also chip quirks/bugs in init logic
 
 	struct work_struct	nec_work;	/* Worker for NEC quirk */

@@ -607,15 +608,12 @@ static inline u32 hc32_to_cpup (const st
 /* HCCA frame number is 16 bits, but is accessed as 32 bits since not all
  * hardware handles 16 bit reads.  That creates a different confusion on
  * some big-endian SOC implementations.  Same thing happens with PSW access.
- *
- * FIXME: Deal with that as a runtime quirk when STB03xxx is ported over
- * to arch/powerpc
  */
 
-#ifdef CONFIG_STB03xxx
-#define OHCI_BE_FRAME_NO_SHIFT	16
+#ifdef CONFIG_PPC_MPC52xx
+#define big_endian_frame_no_quirk(ohci)	(ohci->flags & OHCI_QUIRK_FRAME_NO)
 #else
-#define OHCI_BE_FRAME_NO_SHIFT	0
+#define big_endian_frame_no_quirk(ohci)	0
 #endif
 
 static inline u16 ohci_frame_no(const struct ohci_hcd *ohci)

@@ -623,7 +621,8 @@ static inline u16 ohci_frame_no(const st
 	u32 tmp;
 	if (big_endian_desc(ohci)) {
 		tmp = be32_to_cpup((__force __be32 *)&ohci->hcca->frame_no);
-		tmp >>= OHCI_BE_FRAME_NO_SHIFT;
+		if (!big_endian_frame_no_quirk(ohci))
+			tmp >>= 16;
 	} else
 		tmp = le32_to_cpup((__force __le32 *)&ohci->hcca->frame_no);

drivers/usb/host/ohci-ppc-of.c +4 -1

diff -pruN linux-2.6.orig/drivers/usb/host/ohci-ppc-of.c linux-2.6/drivers/usb/host/ohci-ppc-of.c
@@ -134,8 +134,11 @@ ohci_hcd_ppc_of_probe(struct of_device *
 	}
 
 	ohci = hcd_to_ohci(hcd);
-	if (is_bigendian)
+	if (is_bigendian) {
 		ohci->flags |= OHCI_QUIRK_BE_MMIO | OHCI_QUIRK_BE_DESC;
+		if (of_device_is_compatible(dn, "mpc5200-ohci"))
+			ohci->flags |= OHCI_QUIRK_FRAME_NO;
+	}
 
 	ohci_hcd_init(ohci);

Valentine Barshak

author

2007-10-08 16:59 UTC #5 lore raw

Re: [linux-usb-devel] [PATCH] USB: Add frame_no big endian OHCI quirk.

I've tested this patch only on PowerPC 440EPx Sequoia. So, this needs to be tested at least on mpc52xx. Thanks,
Valentine.

Valentine Barshak

author

2007-10-08 18:43 UTC #6 lore raw

PATCH 1/1 USB: Add frame_no big endian OHCI quirk

In reply to Valentine Barshak

Signed-off-by Valentine Barshak <[email protected]>

---
 drivers/usb/host/ohci-ppc-of.c  |    5 ++++-
 drivers/usb/host/ohci-ppc-soc.c |    5 +++++
 drivers/usb/host/ohci.h         |   13 ++++++-------
 3 files changed, 15 insertions(+), 8 deletions(-)

drivers/usb/host/ohci.h +6 -7

diff -pruN linux-2.6.orig/drivers/usb/host/ohci.h linux-2.6/drivers/usb/host/ohci.h
@@ -398,6 +398,7 @@ struct ohci_hcd {
 #define	OHCI_QUIRK_BE_MMIO	0x10			/* BE registers */
 #define	OHCI_QUIRK_ZFMICRO	0x20			/* Compaq ZFMicro chipset*/
 #define	OHCI_QUIRK_NEC		0x40			/* lost interrupts */
+#define	OHCI_QUIRK_FRAME_NO	0x80			/* no big endian frame_no shift */
 	// there are also chip quirks/bugs in init logic
 
 	struct work_struct	nec_work;	/* Worker for NEC quirk */

@@ -607,15 +608,12 @@ static inline u32 hc32_to_cpup (const st
 /* HCCA frame number is 16 bits, but is accessed as 32 bits since not all
  * hardware handles 16 bit reads.  That creates a different confusion on
  * some big-endian SOC implementations.  Same thing happens with PSW access.
- *
- * FIXME: Deal with that as a runtime quirk when STB03xxx is ported over
- * to arch/powerpc
  */
 
-#ifdef CONFIG_STB03xxx
-#define OHCI_BE_FRAME_NO_SHIFT	16
+#ifdef CONFIG_PPC_MPC52xx
+#define big_endian_frame_no_quirk(ohci)	(ohci->flags & OHCI_QUIRK_FRAME_NO)
 #else
-#define OHCI_BE_FRAME_NO_SHIFT	0
+#define big_endian_frame_no_quirk(ohci)	0
 #endif
 
 static inline u16 ohci_frame_no(const struct ohci_hcd *ohci)

@@ -623,7 +621,8 @@ static inline u16 ohci_frame_no(const st
 	u32 tmp;
 	if (big_endian_desc(ohci)) {
 		tmp = be32_to_cpup((__force __be32 *)&ohci->hcca->frame_no);
-		tmp >>= OHCI_BE_FRAME_NO_SHIFT;
+		if (!big_endian_frame_no_quirk(ohci))
+			tmp >>= 16;
 	} else
 		tmp = le32_to_cpup((__force __le32 *)&ohci->hcca->frame_no);

drivers/usb/host/ohci-ppc-of.c +4 -1

diff -pruN linux-2.6.orig/drivers/usb/host/ohci-ppc-of.c linux-2.6/drivers/usb/host/ohci-ppc-of.c
@@ -134,8 +134,11 @@ ohci_hcd_ppc_of_probe(struct of_device *
 	}
 
 	ohci = hcd_to_ohci(hcd);
-	if (is_bigendian)
+	if (is_bigendian) {
 		ohci->flags |= OHCI_QUIRK_BE_MMIO | OHCI_QUIRK_BE_DESC;
+		if (of_device_is_compatible(dn, "mpc5200-ohci"))
+			ohci->flags |= OHCI_QUIRK_FRAME_NO;
+	}
 
 	ohci_hcd_init(ohci);

drivers/usb/host/ohci-ppc-soc.c +5 -0

diff -pruN linux-2.6.orig/drivers/usb/host/ohci-ppc-soc.c linux-2.6/drivers/usb/host/ohci-ppc-soc.c
@@ -73,6 +73,11 @@ static int usb_hcd_ppc_soc_probe(const s
 
 	ohci = hcd_to_ohci(hcd);
 	ohci->flags |= OHCI_QUIRK_BE_MMIO | OHCI_QUIRK_BE_DESC;
+
+#ifdef CONFIG_PPC_MPC52xx
+	/* MPC52xx doesn't need frame_no shift */
+	ohci->flags |= OHCI_QUIRK_FRAME_NO;
+#endif
 	ohci_hcd_init(ohci);
 
 	retval = usb_add_hcd(hcd, irq, IRQF_DISABLED);

Dale Farnsworth

Acked-by on v1 1/1

2007-10-08 18:52 UTC #7 lore raw

Re: [linux-usb-devel] [PATCH] USB: Add frame_no big endian OHCI quirk

Quoted 10 lines · Valentine Barshak

Add OHCI big endian frame_no quirk.
The frame_no value stored in the HCCA is a 16 bit field at a specific offset, but since not all CPUs can do 16-bit memory accesses it's used as a 32 bit field.
And that's why big-endian OHCI will shift 16 bits, unless the spec is not followed. Currently there's one MPC52xx platform that doesn't need the shift. This case is hanled using big endian frame_no quirk.
Signed-off-by Valentine Barshak <[email protected]>

Acked-by Dale Farnsworth <[email protected]>

Kumar Gala

2007-10-08 17:39 UTC #8 lore raw

In reply to David Brownell

Quoted 7 lines · David Brownell

Quoted 6 lines
Quoted 3 lines
Near as I can tell, the original code is wrong ... the hcca-
Quoted 1 line
frame_no
byte offset is fully specified, so that shift should always be 16.
Are you saying that it should always be #define OHCI_BE_FRAME_NO_SHIFT 16 for big endian platforms?

Quoted 6 lines

More than that, I'm saying that shouldn't even be a #define! The default should be that drivers expect chips to follow the interface specs. Any value other than 16 violates the OHCI spec.
However, based on one other post, I suspect at least one Freescale part will need to declare a chip quirk for this case.

Which Freescale part do you think needs this?

- k

Valentine Barshak

author

2007-10-08 17:43 UTC #9 lore raw

Quoted 13 lines · Kumar Gala

Quoted 10 lines · David Brownell
Quoted 5 lines
Quoted 2 lines
Near as I can tell, the original code is wrong ... the hcca->frame_no byte offset is fully specified, so that shift should always be 16.
Are you saying that it should always be #define OHCI_BE_FRAME_NO_SHIFT 16 for big endian platforms?
More than that, I'm saying that shouldn't even be a #define! The default should be that drivers expect chips to follow the interface specs. Any value other than 16 violates the OHCI spec.

Quoted 6 lines

Quoted 2 lines
However, based on one other post, I suspect at least one Freescale part will need to declare a chip quirk for this case.
Which Freescale part do you think needs this?
- k

I've looked at drivers/usb/host/Kconfig
It seems to be MPC52xx.
David, is that the one you mentioned?
Thanks,
Valentine.

David Brownell

2007-10-08 19:45 UTC #10 lore raw

Quoted 8 lines

Quoted 4 lines
Quoted 2 lines
However, based on one other post, I suspect at least one Freescale part will need to declare a chip quirk for this case.
Which Freescale part do you think needs this?
I've looked at drivers/usb/host/Kconfig
It seems to be MPC52xx.
David, is that the one you mentioned?

I wasn't the one to mention it. It was someone else in this thread.