usb: gadget: pxa25x: prevent ixp4xx + pxa25x combination

From: Arnd Bergmann <[email protected]>

The USB gadget driver is one of the holdouts of Arm single-platform support that conflicts with a multiplatform kernel. Trying to build this for both Intel PXA and IXP fails with:

drivers/usb/gadget/udc/pxa25x_udc.c:144:9: error: 'UDCCS_IO_ROF' redefined [-Werror]
144 | #define UDCCS_IO_ROF (1 << 2) /* Receive overflow */

      |         ^~~~~~~~~~~~
drivers/usb/gadget/udc/pxa25x_udc.c:141:9: note: this is the location of the previous definition
  141 | #define UDCCS_IO_ROF    (1 << 3)        /* Receive overflow */
      |         ^~~~~~~~~~~~

drivers/usb/gadget/udc/pxa25x_udc.c:232:2: error: #error "Can't configure both IXP and PXA"
232 | #error "Can't configure both IXP and PXA"

Until recently this was not a problem, as ixp4xx was only big-endian and pxa is only little-endian, so these could never be in the same kernel image.

As this has changed recently, an allmodconfig kernel for ARMv5 runs into the above failure. I don't think this is an important scenario in practice, especially as pxa25x support is already rather broken, so prevent this configuration using a Kconfig dependency, and rework the driver to allow a combined ixp4xx/pxa3xx kernel that only uses this driver on ixp.

---
 drivers/usb/gadget/udc/Kconfig      | 1 +
 drivers/usb/gadget/udc/pxa25x_udc.c | 8 ++++----
 2 files changed, 5 insertions(+), 4 deletions(-)

index 7ddbc2606686..5142f70b6f3a 100644
@@ -126,6 +126,7 @@ config USB_OMAP
 config USB_PXA25X
 	tristate "PXA 25x or IXP 4xx"
 	depends on (ARCH_PXA && PXA25x) || ARCH_IXP4XX
+	depends on !(PXA25x && ARCH_IXP4XX)
 	depends on HAS_IOMEM
 	help
 	   Intel's PXA 25x series XScale ARM-5TE processors include

index 594d67193763..7e158ca73fa1 100644
@@ -140,7 +140,7 @@
 #ifdef CONFIG_ARCH_IXP4XX /* FIXME: is this right?, datasheed says '2' */
 #define UDCCS_IO_ROF	(1 << 3)	/* Receive overflow */
 #endif
-#ifdef CONFIG_ARCH_PXA
+#ifdef CONFIG_PXA25x
 #define UDCCS_IO_ROF	(1 << 2)	/* Receive overflow */
 #endif
 #define UDCCS_IO_DME	(1 << 3)	/* DMA enable */

@@ -228,7 +228,7 @@ static const char ep0name [] = "ep0";
 #ifdef CONFIG_ARCH_IXP4XX
 
 /* cpu-specific register addresses are compiled in to this code */
-#ifdef CONFIG_ARCH_PXA
+#ifdef CONFIG_PXA25x
 #error "Can't configure both IXP and PXA"
 #endif
 

@@ -2230,7 +2230,7 @@ static struct pxa25x_udc memory = {
 
 #define CP15R0_VENDOR_MASK	0xffffe000
 
-#if	defined(CONFIG_ARCH_PXA)
+#if	defined(CONFIG_PXA25x)
 #define CP15R0_XSCALE_VALUE	0x69052000	/* intel/arm/xscale */
 
 #elif	defined(CONFIG_ARCH_IXP4XX)

@@ -2282,7 +2282,7 @@ static int pxa25x_udc_probe(struct platform_device *pdev)
 
 	/* trigger chiprev-specific logic */
 	switch (chiprev & CP15R0_PRODREV_MASK) {
-#if	defined(CONFIG_ARCH_PXA)
+#if	defined(CONFIG_PXA25x)
 	case PXA255_A0:
 		dev->has_cfr = 1;
 		break;

2.53.0

That's a reasonable fix. It's not like we would want an actual multi_* kernel on these devices anyway.

Yours,
Linus Walleij