Skip to contentSkip to conversation

Git tree updated to 2.6.19-rc2

PATCH In review Tony Lindgren 6 participants 14 messages Oct 23, 2006 – Oct 29, 2006

Subsystems (7)
lore
Share & tools
lore

Tony Lindgren

author reviewer
#0 lore raw

Hi all,

I've updated our git tree to 2.6.19-rc2. At least following have problems on omap1 (OSK):

- CONFIG_DEBUG_LL hangs system
- BUG in omap1_mbox_enable_irq
- Aic23 NULL pointer exception

So YMMV :) Anybody got patches for these issues already?

Tony

#1 lore raw
Quoted 5 lines · --- Tony Lindgren

Hi all,

I've updated our git tree to 2.6.19-rc2. At least following have problems on omap1 (OSK):

..

Quoted 2 lines

- Aic23 NULL pointer exception

Anything related to as discussed on davinci mailing list for I2C?

http://linux.omap.com/pipermail/davinci-linux-open-source/2006-October/001312.html

If that is the case, then please try adding zero-length transaction and SMBUS_QUICK code as temporary hack.

---Komal Shah
http://komalshah.blogspot.com/

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

#2 lore raw

Aic23 NULL pointer exception, was: Git tree updated to 2.6.19-rc2

Quoted 3 lines · Komal Shah
Quoted 1 line

- Aic23 NULL pointer exception

Anything related to as discussed on davinci mailing list for I2C?

Oops is

Unable to handle kernel NULL pointer dereference at virtual address 00000004
PC is at i2c_smbus_write_byte_data+0x10/0x50
LR is at aic23_write_value+0x40/0x50
Quoted 1 line

http://linux.omap.com/pipermail/davinci-linux-open-source/2006-October/001312.html

I think this is the same?

Quoted 2 lines

If that is the case, then please try adding zero-length transaction and SMBUS_QUICK code as temporary hack.

Anywhere a (hackish) patch for this available? Would be nice to have sound working again, even if a hack, until clean solution is available.

Best regards

Dirk

#3 lore raw

PATCH 1/1 ARM: OMAP: Fixup omap-rtc (was: Git tree updated to 2.6.19-rc2)

In reply to Tony Lindgren
Quoted 8 lines · Tony Lindgren

I've updated our git tree to 2.6.19-rc2. At least following have problems on omap1 (OSK):

- CONFIG_DEBUG_LL hangs system
- BUG in omap1_mbox_enable_irq
- Aic23 NULL pointer exception

So YMMV :) Anybody got patches for these issues already?

Haven't hit any of these with the Amstrad Delta defconfig so far (just done a compile and boot so far), but did need the patch below to get omap-rtc to compile.

-----

drivers/char/omap-rtc.c +1 -1
index f48563f..d57402f 100644
@@ -120,7 +120,7 @@ static const unsigned char days_in_mo[] 
  *	A very tiny interrupt handler. It runs with SA_INTERRUPT set.
  */
 
-irqreturn_t rtc_interrupt(int irq, void *dev_id, struct pt_regs *regs)
+irqreturn_t rtc_interrupt(int irq, void *dev_id)
 {
 	/*
 	 *	Either an alarm interrupt or update complete interrupt.

-----

J.

Signature · 4 lines
] http://www.earth.li/~noodles/ []   How does it work?  I don't know   [
]  PGP/GPG Key @ the.earth.li   []            but it does!             [
] via keyserver, web or email.  []                                     [
] RSA: 4DC4E7FD / DSA: 5B430367 []                                     [
#4 lore raw

Re: [PATCH] ARM: OMAP: Fixup omap-rtc (was: Git tree updated to 2.6.19-rc2)

Quoted 5 lines · Jonathan McDowell

Haven't hit any of these with the Amstrad Delta defconfig so far (just done a compile and boot so far), but did need the patch below to get omap-rtc to compile.

Looks right. In fact, that RTC should IMO be added to defconfig for all the OMAP platforms ...

author reviewer
#5 lore raw

Re: [PATCH] ARM: OMAP: Fixup omap-rtc (was: Git tree updated to 2.6.19-rc2)

Quoted 10 lines · David Brownell
Quoted 5 lines · Jonathan McDowell

Haven't hit any of these with the Amstrad Delta defconfig so far (just done a compile and boot so far), but did need the patch below to get omap-rtc to compile.

Looks right. In fact, that RTC should IMO be added to defconfig for all the OMAP platforms ...

OK, I've pushed Jonathan's patch and also removed other unnecessary pt_regs.

Some boards use external RTC, but I agree we should add it in general.

Tony

#6 lore raw

Re: omap-keypad issues with 2.6.19-rc?

In reply to Jonathan McDowell
Quoted 4 lines · Jonathan McDowell
Quoted 3 lines · Tony Lindgren

I've updated our git tree to 2.6.19-rc2. At least following have problems on omap1 (OSK):

Quoted 8 lines
Quoted 5 lines

- CONFIG_DEBUG_LL hangs system
- BUG in omap1_mbox_enable_irq
- Aic23 NULL pointer exception

So YMMV :) Anybody got patches for these issues already?

Haven't hit any of these with the Amstrad Delta defconfig so far (just done a compile and boot so far),

However having prodded further I'm seeing incorrect key parsing with the omap-keypad driver under both -rc2 and -rc3 (works fine in 2.6.18-omap1). Numerous keys now returns strings of characters rather than the single expected character. A few keys still work as expected however. Before I dig in further is anyone else seeing anything similar?

J.

Signature · 3 lines
 After a number of decimal places,  |       Black Cat Networks Ltd
       nobody gives a damn.         | http://www.blackcatnetworks.co.uk/
                                    |  UK Web, domain and email hosting
#7 lore raw

PATCH 1/1 ARM: OMAP: Fix Amstrad Delta omap-keypad usage.

Quoted 5 lines · Jonathan McDowell

However having prodded further I'm seeing incorrect key parsing with the omap-keypad driver under both -rc2 and -rc3 (works fine in 2.6.18-omap1). Numerous keys now returns strings of characters rather than the single expected character. A few keys still work as expected however. Before I dig in further is anyone else seeing anything similar?

This appears to be due to the addition of keymapsize to the omap_kp_platform_data structure. The patch below fixes things up for the Delta; other boards appear to have already had this done.

-----

arch/arm/mach-omap1/board-ams-delta.c +5 -4
index 0a0c06f..c73ca61 100644
@@ -187,10 +187,11 @@ static struct resource ams_delta_kp_reso
 };
 
 static struct omap_kp_platform_data ams_delta_kp_data = {
-	.rows	= 8,
-	.cols	= 8,
-	.keymap = ams_delta_keymap,
-	.rep	= 1,
+	.rows		= 8,
+	.cols		= 8,
+	.keymap 	= ams_delta_keymap,
+	.keymapsize	= ARRAY_SIZE(ams_delta_keymap),
+	.delay		= 9,
 };
 
 static struct platform_device ams_delta_kp_device = {

-----

J.

Signature · 4 lines
   101 things you can't have too   |  .''`.  Debian GNU/Linux Developer
  much of : 9 - Processing power.  | : :' :  Happy to accept PGP signed
                                   | `. `'   or encrypted mail - RSA +
                                   |   `-    DSA keys on the keyservers.
#8 lore raw
In reply to Tony Lindgren
Quoted 1 line · Tony Lindgren

- CONFIG_DEBUG_LL hangs system

For reference see

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008025.html

and

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008054.html

for an ugly workaround if you really need it. I really have no clue what is broken here.

Quoted 1 line

- BUG in omap1_mbox_enable_irq

I think you mean

Unbalanced enable for IRQ 10

BUG: warning at kernel/irq/manage.c:118/enable_irq()

printed in bootmessages.

I spent some minutes for this as well. It is MPU "dsp" interrupt which is enabled in omap1_mbox_enable_irq() by

enable_irq(mbox->irq);

In kernel/irq/manage.c function enable_irq() checks for desc->depth and if it is 0 it prints above message. Seems that desc->depth has to be >= 1.

The comment tells us

* Undoes the effect of one call to disable_irq(). If this
* matches the last disable, processing of interrupts on this
* IRQ line is re-enabled.

so seems that before calling enable_irq() at least one disable_irq() is needed.

I don't know if patch in attachment is a valid patch or only a workaround. What do you think?

Best regards

Dirk

arm/plat-omap/mailbox.c renamed from arm/plat-omap/mailbox.c_orig +1 -0
@@ -237,6 +237,7 @@ static int omap_mbox_init(struct omap_mb
 		       "failed to register mailbox interrupt:%d\n", ret);
 		goto fail1;
 	}
+	disable_mbox_irq(mbox, IRQ_RX);
 	enable_mbox_irq(mbox, IRQ_RX);
 
 	spin_lock_init(&mbox->lock);
author reviewer
#9 lore raw
Quoted 5 lines · Dirk Behme
Quoted 1 line · Tony Lindgren

- CONFIG_DEBUG_LL hangs system

For reference see

Quoted 8 lines

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008025.html

and

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008054.html

for an ugly workaround if you really need it. I really have no clue what is broken here.

Sounds like some clock is not enabled somewhere and serial write hangs.

Quoted 39 lines
Quoted 1 line

- BUG in omap1_mbox_enable_irq

I think you mean

Unbalanced enable for IRQ 10

BUG: warning at kernel/irq/manage.c:118/enable_irq()

printed in bootmessages.

I spent some minutes for this as well. It is MPU "dsp" interrupt which is enabled in omap1_mbox_enable_irq() by

enable_irq(mbox->irq);

In kernel/irq/manage.c function enable_irq() checks for desc->depth and if it is 0 it prints above message. Seems that desc->depth has to be >= 1.

The comment tells us

* Undoes the effect of one call to disable_irq(). If this
* matches the last disable, processing of interrupts on this
* IRQ line is re-enabled.

so seems that before calling enable_irq() at least one disable_irq() is needed.

I don't know if patch in attachment is a valid patch or only a workaround. What do you think?

Best regards

Dirk

arm/plat-omap/mailbox.c static int omap_mbox_init(struct omap_mb · line 243

Quoted 7 lines
arm/plat-omap/mailbox.c renamed from arm/plat-omap/mailbox.c_orig +1 -0
@@ -237,6 +237,7 @@ static int omap_mbox_init(struct omap_mb
 		       "failed to register mailbox interrupt:%d\n", ret);
 		goto fail1;
 	}
+	disable_mbox_irq(mbox, IRQ_RX);
 	enable_mbox_irq(mbox, IRQ_RX);
 
 	spin_lock_init(&mbox->lock);

Let's see what Toshihiro and Hiroshi think.

Tony

#10 lore raw
Quoted 12 lines · [email protected] wrote:
Quoted 10 lines · Dirk Behme
Quoted 1 line · Tony Lindgren

- CONFIG_DEBUG_LL hangs system

For reference see

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008025.html

and

Quoted 7 lines
Quoted 4 lines

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008054.html

for an ugly workaround if you really need it. I really have no clue what is broken here.

Sounds like some clock is not enabled somewhere and serial write hangs.

Good idea! I think we had some clock rework meanwhile. Maybe we have to move a clock enable to some earlier point if using CONFIG_DEBUG_LL. Any idea which clock is worth looking at?

Best regards

Dirk

#11 lore raw
In reply to [email protected]

From: "ext [email protected]" <[email protected]>
Subject: Re: Git tree updated to 2.6.19-rc2
Date: Fri, 27 Oct 2006 15:50:17 +0300

Quoted 61 lines · Dirk Behme
Quoted 5 lines
Quoted 1 line · Tony Lindgren

- CONFIG_DEBUG_LL hangs system

For reference see

Quoted 8 lines

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008025.html

and

http://linux.omap.com/pipermail/linux-omap-open-source/2006-October/008054.html

for an ugly workaround if you really need it. I really have no clue what is broken here.

Sounds like some clock is not enabled somewhere and serial write hangs.

Quoted 44 lines
Quoted 1 line

- BUG in omap1_mbox_enable_irq

I think you mean

Unbalanced enable for IRQ 10

BUG: warning at kernel/irq/manage.c:118/enable_irq()

printed in bootmessages.

I spent some minutes for this as well. It is MPU "dsp" interrupt which is enabled in omap1_mbox_enable_irq() by

enable_irq(mbox->irq);

In kernel/irq/manage.c function enable_irq() checks for desc->depth and if it is 0 it prints above message. Seems that desc->depth has to be >= 1.

The comment tells us

* Undoes the effect of one call to disable_irq(). If this
* matches the last disable, processing of interrupts on this
* IRQ line is re-enabled.

so seems that before calling enable_irq() at least one disable_irq() is needed.

I don't know if patch in attachment is a valid patch or only a workaround. What do you think?

Best regards

Dirk

arm/plat-omap/mailbox.c renamed from arm/plat-omap/mailbox.c_orig +1 -0
@@ -237,6 +237,7 @@ static int omap_mbox_init(struct omap_mb
 		       "failed to register mailbox interrupt:%d\n", ret);
 		goto fail1;
 	}
+	disable_mbox_irq(mbox, IRQ_RX);
 	enable_mbox_irq(mbox, IRQ_RX);
 
Quoted 4 lines

arm/plat-omap/mailbox.c static int omap_mbox_init(struct omap_mb · line 243

Quoted 2 lines
arm/plat-omap/mailbox.c renamed from arm/plat-omap/mailbox.c_orig +0 -0
 	spin_lock_init(&mbox->lock);

Let's see what Toshihiro and Hiroshi think.

I guess that the above should work for now but I think that there's another way, for example, I can have initialization flag in enable_mbox_irq() in omap1, though I am sure that there's better way.

Hiroshi DOYU

#12 lore raw
In reply to Tony Lindgren

I measured it at "rc3", myself ... :)

Here's an update to the musb/tusb code fixing, among other things, compile glitches that have recently joined us. Like removing pt_regs in all the IRQ paths, and a rename to <linux/usb/otg.h> ...

It's possible that (finally) removing the MIN_DMA_REQUEST thing would make trouble; best way to test that is probably with g_file_storage. If that fails, the obvious (temporary) workaround is disabling DMA.

- Dave

Various bugfixes:
- gadget side does dma_sync_single_for_*() if the buffer
was mapped by the gadget driver (none do that just yet)
- host side rx, always force dma request off "just in case", instead of just for Mentor's DMA (i.e. not just OMAP2430)
- complain louder about the board needing different driver config
- iounmap() registers on rmmod
- fault handling fixes in tusb_start()
- (from TI) save dma mask so modprobe after use_dma=n can use dma
- (from TI) gadget side zero-length IN likes TXPKTRDY before status stage

Other updates:
- catch up to pt_regs finally leaving the irq path, other build fixes
- remove pointless headers like <linux/pci.h>
- RemoveMoreCamelCaseNastiness
- resolve some config-specific compiler warning/error messages
- dump tusb irq status register too
- remove that MIN_DMA thing; dma logic must now handle zlps in all cases, not just to terminate a transfer
- comment fixes

drivers/usb/musb/musbdefs.h +2 -6
Index: h4/drivers/usb/musb/musbdefs.h
===================================================================
@@ -43,7 +43,7 @@
 #include <linux/usb_ch9.h>
 #include <linux/usb_gadget.h>
 #include <linux/usb.h>
-#include <linux/usb_otg.h>
+#include <linux/usb/otg.h>
 #include <linux/usb/musb.h>
 
 struct musb;
@@ -75,9 +75,6 @@ struct musb_ep;
 #include "otg.h"
 
 
-/* REVISIT tune this */
-#define	MIN_DMA_REQUEST		1	/* use PIO below this xfer size */
-
 
 #ifdef CONFIG_USB_MUSB_OTG
 
@@ -374,7 +371,7 @@ static inline struct usb_request *next_o
 struct musb {
 	spinlock_t		Lock;
 	struct clk		*clock;
-	irqreturn_t		(*isr)(int, void *, struct pt_regs *);
+	irqreturn_t		(*isr)(int, void *);
 	struct work_struct	irq_work;
 
 #ifdef CONFIG_USB_MUSB_HDRC_HCD
@@ -418,7 +415,6 @@ struct musb {
 	u8			int_usb;
 	u16			int_rx;
 	u16			int_tx;
-	struct pt_regs		*int_regs;
 
 	struct otg_transceiver	xceiv;
 
drivers/usb/musb/g_ep0.c +4 -2
Index: h4/drivers/usb/musb/g_ep0.c
===================================================================
@@ -556,9 +556,11 @@ musb_read_setup(struct musb *pThis, stru
 	 */
 	pThis->bSetAddress = FALSE;
 	pThis->ackpend = MGC_M_CSR0_P_SVDRXPKTRDY;
-	if (req->wLength == 0)
+	if (req->wLength == 0) {
+		if (req->bRequestType & USB_DIR_IN)
+			pThis->ackpend |= MGC_M_CSR0_TXPKTRDY;
 		pThis->ep0_state = MGC_END0_STAGE_ACKWAIT;
-	else if (req->bRequestType & USB_DIR_IN) {
+	} else if (req->bRequestType & USB_DIR_IN) {
 		pThis->ep0_state = MGC_END0_STAGE_TX;
 		musb_writew(regs, MGC_O_HDRC_CSR0, MGC_M_CSR0_P_SVDRXPKTRDY);
 		while ((musb_readw(regs, MGC_O_HDRC_CSR0)
drivers/usb/musb/musb_gadget.h +1 -0
Index: h4/drivers/usb/musb/musb_gadget.h
===================================================================
@@ -74,6 +74,7 @@ struct musb_ep {
 	/* later things are modified based on usage */
 	struct list_head		req_list;
 
+	/* true if lock must be dropped but req_list may not be advanced */
 	u8				busy;
 };
 
drivers/usb/musb/musb_gadget.c +42 -26

107 lines not on this page · raw message

drivers/usb/musb/musb_host.c +11 -15
Index: h4/drivers/usb/musb/musb_host.c
===================================================================
@@ -33,7 +33,6 @@
  ******************************************************************/
 
 #include <linux/module.h>
-#include <linux/pci.h>
 #include <linux/kernel.h>
 #include <linux/delay.h>
 #include <linux/sched.h>
@@ -284,7 +283,7 @@ __acquires(musb->Lock)
 			);
 
 	spin_unlock(&musb->Lock);
-	usb_hcd_giveback_urb(musb_to_hcd(musb), urb, musb->int_regs);
+	usb_hcd_giveback_urb(musb_to_hcd(musb), urb);
 	spin_lock(&musb->Lock);
 }
 
@@ -1485,8 +1484,8 @@ void musb_host_rx(struct musb *pThis, u8
 	}
 
 	if (unlikely(dma_channel_status(dma) == MGC_DMA_STATUS_BUSY)) {
-		/* SHOULD NEVER HAPPEN */
-		ERR("RX%d dma busy\n", bEnd);
+		/* SHOULD NEVER HAPPEN ... but at least DaVinci has done it */
+		ERR("RX%d dma busy, csr %04x\n", bEnd, wRxCsrVal);
 		goto finish;
 	}
 
@@ -1521,9 +1520,15 @@ void musb_host_rx(struct musb *pThis, u8
 	}
 #endif
 	if (dma && (wRxCsrVal & MGC_M_RXCSR_DMAENAB)) {
+		xfer_len = dma->dwActualLength;
+
+		wVal &= ~(MGC_M_RXCSR_DMAENAB
+			| MGC_M_RXCSR_H_AUTOREQ
+			| MGC_M_RXCSR_AUTOCLEAR
+			| MGC_M_RXCSR_RXPKTRDY);
+		musb_writew(pEnd->regs, MGC_O_HDRC_RXCSR, wVal);
 
 #ifdef CONFIG_USB_INVENTRA_DMA
-		xfer_len = dma->dwActualLength;
 		pUrb->actual_length += xfer_len;
 		qh->offset += xfer_len;
 
@@ -1531,14 +1536,6 @@ void musb_host_rx(struct musb *pThis, u8
 		bDone = (pUrb->actual_length >= pUrb->transfer_buffer_length)
 			|| (dma->dwActualLength & (qh->maxpacket - 1));
 
-		wVal &= ~(MGC_M_RXCSR_DMAENAB
-			| MGC_M_RXCSR_H_AUTOREQ
-			| MGC_M_RXCSR_AUTOCLEAR
-			| MGC_M_RXCSR_RXPKTRDY);
-
-		MGC_WriteCsr16(pBase, MGC_O_HDRC_RXCSR, bEnd, wVal);
-		MGC_WriteCsr16(pBase, MGC_O_HDRC_RXCSR, bEnd, wVal);
-
 		/* send IN token for next packet, without AUTOREQ */
 		if (!bDone) {
 			wVal |= MGC_M_RXCSR_H_REQPKT;
@@ -1552,7 +1549,6 @@ void musb_host_rx(struct musb *pThis, u8
 			MGC_ReadCsr16(pBase, MGC_O_HDRC_RXCOUNT, bEnd));
 #else
 		bDone = TRUE;
-		xfer_len = dma->dwActualLength;
 #endif
 	} else if (pUrb->status == -EINPROGRESS) {
 		/* if no errors, be sure a packet is ready for unloading */
@@ -1682,7 +1678,7 @@ static int musb_schedule(
 	int			idle;
 	int			wBestDiff;
 	int			nBestEnd, nEnd;
-	struct musb_hw_ep	*hw_ep;
+	struct musb_hw_ep	*hw_ep = NULL;
 	struct list_head	*head = NULL;
 
 	/* use fixed hardware for control and bulk */
drivers/usb/musb/plat_uds.c +21 -9
Index: h4/drivers/usb/musb/plat_uds.c
===================================================================
@@ -94,7 +94,6 @@
  */
 
 #include <linux/module.h>
-#include <linux/pci.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
@@ -665,6 +664,7 @@ void musb_start(struct musb *musb)
 	musb_writeb(regs, MGC_O_HDRC_POWER, MGC_M_POWER_ISOUPDATE
 						| MGC_M_POWER_SOFTCONN
 						| MGC_M_POWER_HSENAB
+						/* ENSUSPEND wedges tusb */
 						// | MGC_M_POWER_ENSUSPEND
 						);
 
@@ -672,7 +672,7 @@ void musb_start(struct musb *musb)
 	devctl = musb_readb(regs, MGC_O_HDRC_DEVCTL);
 	devctl &= ~MGC_M_DEVCTL_SESSION;
 
-	if (is_otg_enabled(pThis)) {
+	if (is_otg_enabled(musb)) {
 		/* session started after:
 		 * (a) ID-grounded irq, host mode;
 		 * (b) vbus present/connect IRQ, peripheral mode;
@@ -683,7 +683,7 @@ void musb_start(struct musb *musb)
 		else
 			devctl |= MGC_M_DEVCTL_SESSION;
 
-	} else if (is_host_enabled(pThis)) {
+	} else if (is_host_enabled(musb)) {
 		/* assume ID pin is hard-wired to ground */
 		devctl |= MGC_M_DEVCTL_SESSION;
 
@@ -1276,7 +1276,7 @@ static int __devinit musb_core_init(u16 
 
 #ifdef CONFIG_ARCH_OMAP243X
 
-static irqreturn_t generic_interrupt(int irq, void *__hci, struct pt_regs *r)
+static irqreturn_t generic_interrupt(int irq, void *__hci)
 {
 	unsigned long	flags;
 	irqreturn_t	retval = IRQ_NONE;
@@ -1287,7 +1287,6 @@ static irqreturn_t generic_interrupt(int
 	musb->int_usb = musb_readb(musb->pRegs, MGC_O_HDRC_INTRUSB);
 	musb->int_tx = musb_readw(musb->pRegs, MGC_O_HDRC_INTRTX);
 	musb->int_rx = musb_readw(musb->pRegs, MGC_O_HDRC_INTRRX);
-	musb->int_regs = r;
 
 	if (musb->int_usb || musb->int_tx || musb->int_rx)
 		retval = musb_interrupt(musb);
@@ -1604,7 +1603,6 @@ static void musb_free(struct musb *musb)
 	if (is_dma_capable() && musb->pDmaController) {
 		struct dma_controller	*c = musb->pDmaController;
 
-//
 		(void) c->stop(c->pPrivateData);
 		dma_controller_factory.destroy(c);
 	}
@@ -1667,7 +1665,7 @@ musb_init_controller(struct device *dev,
 	bad_config:
 #endif
 	default:
-		dev_dbg(dev, "incompatible Kconfig role setting\n");
+		dev_err(dev, "incompatible Kconfig role setting\n");
 		return -EINVAL;
 	}
 
@@ -1811,8 +1809,9 @@ fail:
 	INIT_WORK(&pThis->irq_work, musb_irq_work, pThis);
 
 #ifdef CONFIG_SYSFS
-	device_create_file(dev, &dev_attr_mode);
-	device_create_file(dev, &dev_attr_cable);
+	status = device_create_file(dev, &dev_attr_mode);
+	status = device_create_file(dev, &dev_attr_cable);
+	status = 0;
 #endif
 
 	return status;
@@ -1828,6 +1827,10 @@ fail2:
  * bridge to a platform device; this driver then suffices.
  */
 
+#ifndef CONFIG_USB_INVENTRA_FIFO
+static u64	*orig_dma_mask;
+#endif
+
 static int __devinit musb_probe(struct platform_device *pdev)
 {
 	struct device	*dev = &pdev->dev;
@@ -1845,12 +1848,17 @@ static int __devinit musb_probe(struct p
 		return -ENOMEM;
 	}
 
+#ifndef CONFIG_USB_INVENTRA_FIFO
+	/* clobbered by use_dma=n */
+	orig_dma_mask = dev->dma_mask;
+#endif
 	return musb_init_controller(dev, irq, base);
 }
 
 static int __devexit musb_remove(struct platform_device *pdev)
 {
 	struct musb	*musb = dev_to_musb(&pdev->dev);
+	void __iomem	*ctrl_base = musb->ctrl_base;
 
 	/* this gets called on rmmod.
 	 *  - Host mode: host may still be active
@@ -1864,7 +1872,11 @@ static int __devexit musb_remove(struct 
 		usb_remove_hcd(musb_to_hcd(musb));
 #endif
 	musb_free(musb);
+	iounmap(ctrl_base);
 	device_init_wakeup(&pdev->dev, 0);
+#ifndef CONFIG_USB_INVENTRA_FIFO
+	dev->dma_mask = orig_dma_mask;
+#endif
 	return 0;
 }
 
drivers/usb/musb/musb_procfs.c +2 -1
Index: h4/drivers/usb/musb/musb_procfs.c
===================================================================
@@ -558,7 +558,7 @@ static int dump_header_stats(struct musb
 	code = sprintf(buffer,
 			"TUSB6010: devconf %08x, phy enable %08x drive %08x"
 			"\n\totg %03x timer %08x"
-			"\n\tprcm conf %08x mgmt %08x; intmask %08x"
+			"\n\tprcm conf %08x mgmt %08x; int src %08x mask %08x"
 			"\n",
 			musb_readl(pThis->ctrl_base, TUSB_DEV_CONF),
 			musb_readl(pThis->ctrl_base, TUSB_PHY_OTG_CTRL_ENABLE),
@@ -567,6 +567,7 @@ static int dump_header_stats(struct musb
 			musb_readl(pThis->ctrl_base, TUSB_DEV_OTG_TIMER),
 			musb_readl(pThis->ctrl_base, TUSB_PRCM_CONF),
 			musb_readl(pThis->ctrl_base, TUSB_PRCM_MNGMT),
+			musb_readl(pThis->ctrl_base, TUSB_INT_SRC),
 			musb_readl(pThis->ctrl_base, TUSB_INT_MASK));
 	if (code <= 0)
 		goto done;
drivers/usb/musb/tusb6010.c +15 -7
Index: h4/drivers/usb/musb/tusb6010.c
===================================================================
@@ -146,6 +146,7 @@ void musb_read_fifo(struct musb_hw_ep *h
  * at most mA current to be drawn from VBUS during a Default-B session
  * (that is, while VBUS exceeds 4.4V).  In Default-A (including pure host
  * mode), or low power Default-B sessions, something else supplies power.
+ * Caller must take care of locking.
  */
 static int tusb_set_power(struct otg_transceiver *x, unsigned mA)
 {
@@ -178,7 +179,7 @@ static int tusb_set_power(struct otg_tra
  * (to be fixed in rev3 silicon) ... symptoms include disconnect
  * or looping suspend/resume cycles
  */
-static void tusb_set_clock_source(struct musb *musb, int mode)
+static void tusb_set_clock_source(struct musb *musb, unsigned mode)
 {
 	void __iomem	*base = musb->ctrl_base;
 	u32		reg;
@@ -186,10 +187,17 @@ static void tusb_set_clock_source(struct
 	reg = musb_readl(base, TUSB_PRCM_CONF);
 	reg &= ~TUSB_PRCM_CONF_SYS_CLKSEL(0x3);
 
+	/* 0 = refclk (clkin, XI)
+	 * 1 = PHY 60 MHz (internal PLL)
+	 * 2 = not supported
+	 * 3 = NOR clock (huh?)
+	 */
 	if (mode > 0)
 		reg |= TUSB_PRCM_CONF_SYS_CLKSEL(mode & 0x3);
 
 	musb_writel(base, TUSB_PRCM_CONF, reg);
+
+	// FIXME tusb6010_platform_retime(mode == 0);
 }
 
 /*
@@ -477,7 +485,7 @@ tusb_otg_ints(struct musb *musb, u32 int
 	}
 }
 
-static irqreturn_t tusb_interrupt(int irq, void *__hci, struct pt_regs *r)
+static irqreturn_t tusb_interrupt(int irq, void *__hci)
 {
 	struct musb	*musb = __hci;
 	void __iomem	*base = musb->ctrl_base;
@@ -489,7 +497,6 @@ static irqreturn_t tusb_interrupt(int ir
 	int_src = musb_readl(base, TUSB_INT_SRC) & ~TUSB_INT_SRC_RESERVED_BITS;
 	DBG(3, "TUSB IRQ %08x\n", int_src);
 
-	musb->int_regs = r;
 	musb->int_usb = (u8) int_src;
 
 	/* Acknowledge wake-up source interrupts */
@@ -570,7 +577,6 @@ static irqreturn_t tusb_interrupt(int ir
 	musb_writel(base, TUSB_INT_SRC_CLEAR,
 		int_src & ~TUSB_INT_MASK_RESERVED_BITS);
 
-	musb->int_regs = NULL;
 	musb_platform_try_idle(musb);
 	spin_unlock_irqrestore(&musb->Lock, flags);
 
@@ -690,10 +696,10 @@ static int tusb_print_revision(struct mu
 	return TUSB_REV_MAJOR(musb_readl(base, TUSB_INT_CTRL_REV));
 }
 
-static int tusb_start(struct musb *musb)
+static int __devinit tusb_start(struct musb *musb)
 {
 	void __iomem	*base = musb->ctrl_base;
-	int		ret = -1;
+	int		ret = 0;
 	unsigned long	flags;
 	u32		reg;
 
@@ -701,7 +707,7 @@ static int tusb_start(struct musb *musb)
 		ret = musb->board_set_power(1);
 	if (ret != 0) {
 		printk(KERN_ERR "tusb: Cannot enable TUSB6010\n");
-		goto err;
+		return ret;
 	}
 
 	spin_lock_irqsave(&musb->Lock, flags);
@@ -752,6 +758,8 @@ static int tusb_start(struct musb *musb)
 	return 0;
 
 err:
+	spin_unlock_irqrestore(&musb->Lock, flags);
+
 	if (musb->board_set_power)
 		musb->board_set_power(0);
 
drivers/usb/musb/omap2430.c +0 -4
Index: h4/drivers/usb/musb/omap2430.c
===================================================================
@@ -21,7 +21,6 @@
  *
  */
 #include <linux/module.h>
-#include <linux/pci.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
@@ -29,9 +28,6 @@
 #include <linux/list.h>
 #include <linux/clk.h>
 
-#include <linux/module.h>
-#include <linux/kernel.h>
-#include <linux/init.h>
 #include <asm/io.h>
 #include <asm/mach-types.h>
 #include <asm/arch/hardware.h>
drivers/usb/musb/davinci.c +1 -4
Index: h4/drivers/usb/musb/davinci.c
===================================================================
@@ -22,7 +22,6 @@
  */
 
 #include <linux/module.h>
-#include <linux/pci.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
@@ -246,7 +245,7 @@ static void davinci_set_vbus(struct musb
 	return davinci_vbus_power(musb, is_on, 0);
 }
 
-static irqreturn_t davinci_interrupt(int irq, void *__hci, struct pt_regs *r)
+static irqreturn_t davinci_interrupt(int irq, void *__hci)
 {
 	unsigned long	flags;
 	irqreturn_t	retval = IRQ_NONE;
@@ -289,7 +288,6 @@ static irqreturn_t davinci_interrupt(int
 			>> DAVINCI_USB_TXINT_SHIFT;
 	musb->int_usb = (tmp & DAVINCI_USB_USBINT_MASK)
 			>> DAVINCI_USB_USBINT_SHIFT;
-	musb->int_regs = r;
 
 	/* treat DRVVBUS irq like an ID change IRQ (for now) */
 	if (tmp & (1 << (8 + DAVINCI_USB_USBINT_SHIFT))) {
@@ -317,7 +315,6 @@ static irqreturn_t davinci_interrupt(int
 	/* irq stays asserted until EOI is written */
 	musb_writel(tibase, DAVINCI_USB_EOI_REG, 0);
 
-	musb->int_regs = NULL;
 	spin_unlock_irqrestore(&musb->Lock, flags);
 
 	/* REVISIT we sometimes get unhandled IRQs
author reviewer
#13 lore raw
Quoted 24 lines · David Brownell

I measured it at "rc3", myself ... :)

Here's an update to the musb/tusb code fixing, among other things, compile glitches that have recently joined us. Like removing pt_regs in all the IRQ paths, and a rename to <linux/usb/otg.h> ...

It's possible that (finally) removing the MIN_DMA_REQUEST thing would make trouble; best way to test that is probably with g_file_storage. If that fails, the obvious (temporary) workaround is disabling DMA.

- Dave

Various bugfixes:
- gadget side does dma_sync_single_for_*() if the buffer
was mapped by the gadget driver (none do that just yet)
- host side rx, always force dma request off "just in case", instead of just for Mentor's DMA (i.e. not just OMAP2430)
- complain louder about the board needing different driver config
- iounmap() registers on rmmod
- fault handling fixes in tusb_start()
- (from TI) save dma mask so modprobe after use_dma=n can use dma
- (from TI) gadget side zero-length IN likes TXPKTRDY before status stage

Quoted 9 lines

Other updates:
- catch up to pt_regs finally leaving the irq path, other build fixes
- remove pointless headers like <linux/pci.h>
- RemoveMoreCamelCaseNastiness
- resolve some config-specific compiler warning/error messages
- dump tusb irq status register too
- remove that MIN_DMA thing; dma logic must now handle zlps in all cases, not just to terminate a transfer
- comment fixes

Pushing today.

Tony