Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6

Conflicts:

	include/asm-x86/statfs.h

31 files changed, 820 insertions, 274 deletions

diff --git a/arch/avr32/boards/atngw100/setup.c b/arch/avr32/boards/atngw100/setup.c
index b8286f1ce854..6c54580a66df 100644
--- a/arch/avr32/boards/atngw100/setup.c
+++ b/arch/avr32/boards/atngw100/setup.c
@@ -9,6 +9,7 @@
  */
 #include <linux/clk.h>
 #include <linux/etherdevice.h>
+#include <linux/gpio.h>
 #include <linux/irq.h>
 #include <linux/i2c.h>
 #include <linux/i2c-gpio.h>
@@ -53,8 +54,11 @@ static struct spi_board_info spi0_board_info[] __initdata = {
 };
 
 static struct mci_platform_data __initdata mci0_data = {
-	.detect_pin	= GPIO_PIN_PC(25),
-	.wp_pin		= GPIO_PIN_PE(0),
+	.slot[0] = {
+		.bus_width	= 4,
+		.detect_pin	= GPIO_PIN_PC(25),
+		.wp_pin		= GPIO_PIN_PE(0),
+	},
 };
 
 /*
@@ -190,7 +194,7 @@ static int __init atngw100_init(void)
 	 * PB28/EXTINT3 doesn't; it should be SMBALERT# (for PMBus),
 	 * but it's not available off-board.
 	 */
-	at32_select_periph(GPIO_PIN_PB(28), 0, AT32_GPIOF_PULLUP);
+	at32_select_periph(GPIO_PIOB_BASE, 1 << 28, 0, AT32_GPIOF_PULLUP);
 	at32_select_gpio(i2c_gpio_data.sda_pin,
 		AT32_GPIOF_MULTIDRV | AT32_GPIOF_OUTPUT | AT32_GPIOF_HIGH);
 	at32_select_gpio(i2c_gpio_data.scl_pin,
@@ -204,6 +208,15 @@ postcore_initcall(atngw100_init);
 
 static int __init atngw100_arch_init(void)
 {
+	/* PB30 is the otherwise unused jumper on the mainboard, with an
+	 * external pullup; the jumper grounds it.  Use it however you
+	 * like, including letting U-Boot or Linux tweak boot sequences.
+	 */
+	at32_select_gpio(GPIO_PIN_PB(30), 0);
+	gpio_request(GPIO_PIN_PB(30), "j15");
+	gpio_direction_input(GPIO_PIN_PB(30));
+	gpio_export(GPIO_PIN_PB(30), false);
+
 	/* set_irq_type() after the arch_initcall for EIC has run, and
 	 * before the I2C subsystem could try using this IRQ.
 	 */
diff --git a/arch/avr32/boards/atstk1000/atstk1002.c b/arch/avr32/boards/atstk1000/atstk1002.c
index ee4c292683e1..29e5b51a7fd2 100644
--- a/arch/avr32/boards/atstk1000/atstk1002.c
+++ b/arch/avr32/boards/atstk1000/atstk1002.c
@@ -232,7 +232,7 @@ static void __init atstk1002_setup_extdac(void)
 		goto err_set_clk;
 	}
 
-	at32_select_periph(GPIO_PIN_PA(30), GPIO_PERIPH_A, 0);
+	at32_select_periph(GPIO_PIOA_BASE, (1 << 30), GPIO_PERIPH_A, 0);
 	at73c213_data.dac_clk = gclk;
 
 err_set_clk:
@@ -264,16 +264,20 @@ void __init setup_board(void)
 
 #ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
 
+static struct mci_platform_data __initdata mci0_data = {
+	.slot[0] = {
+		.bus_width	= 4,
+
 /* MMC card detect requires MACB0 *NOT* be used */
 #ifdef CONFIG_BOARD_ATSTK1002_SW6_CUSTOM
-static struct mci_platform_data __initdata mci0_data = {
-	.detect_pin	= GPIO_PIN_PC(14),	/* gpio30/sdcd */
-	.wp_pin		= GPIO_PIN_PC(15),	/* gpio31/sdwp */
-};
-#define MCI_PDATA	&mci0_data
+		.detect_pin	= GPIO_PIN_PC(14), /* gpio30/sdcd */
+		.wp_pin		= GPIO_PIN_PC(15), /* gpio31/sdwp */
 #else
-#define MCI_PDATA	NULL
+		.detect_pin	= -ENODEV,
+		.wp_pin		= -ENODEV,
 #endif	/* SW6 for sd{cd,wp} routing */
+	},
+};
 
 #endif	/* SW2 for MMC signal routing */
 
@@ -325,14 +329,15 @@ static int __init atstk1002_init(void)
 #ifdef CONFIG_BOARD_ATSTK100X_SPI1
 	at32_add_device_spi(1, spi1_board_info, ARRAY_SIZE(spi1_board_info));
 #endif
-#ifndef CONFIG_BOARD_ATSTK1002_SW2_CUSTOM
-	at32_add_device_mci(0, MCI_PDATA);
+#ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
+	at32_add_device_mci(0, &mci0_data);
 #endif
 #ifdef CONFIG_BOARD_ATSTK1002_SW5_CUSTOM
 	set_hw_addr(at32_add_device_eth(1, &eth_data[1]));
 #else
 	at32_add_device_lcdc(0, &atstk1000_lcdc_data,
-			     fbmem_start, fbmem_size, 0);
+			     fbmem_start, fbmem_size,
+			     ATMEL_LCDC_PRI_24BIT | ATMEL_LCDC_PRI_CONTROL);
 #endif
 	at32_add_device_usba(0, NULL);
 #ifndef CONFIG_BOARD_ATSTK100X_SW3_CUSTOM
diff --git a/arch/avr32/boards/atstk1000/atstk1003.c b/arch/avr32/boards/atstk1000/atstk1003.c
index 0cf664174c17..be089d7f37eb 100644
--- a/arch/avr32/boards/atstk1000/atstk1003.c
+++ b/arch/avr32/boards/atstk1000/atstk1003.c
@@ -19,6 +19,7 @@
 #include <linux/spi/spi.h>
 
 #include <asm/setup.h>
+#include <asm/atmel-mci.h>
 
 #include <mach/at32ap700x.h>
 #include <mach/board.h>
@@ -66,6 +67,16 @@ static struct spi_board_info spi1_board_info[] __initdata = { {
 } };
 #endif
 
+#ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
+static struct mci_platform_data __initdata mci0_data = {
+	.slot[0] = {
+		.bus_width	= 4,
+		.detect_pin	= -ENODEV,
+		.wp_pin		= -ENODEV,
+	},
+};
+#endif
+
 #ifdef CONFIG_BOARD_ATSTK1000_EXTDAC
 static void __init atstk1003_setup_extdac(void)
 {
@@ -84,7 +95,7 @@ static void __init atstk1003_setup_extdac(void)
 		goto err_set_clk;
 	}
 
-	at32_select_periph(GPIO_PIN_PA(30), GPIO_PERIPH_A, 0);
+	at32_select_periph(GPIO_PIOA_BASE, (1 << 30), GPIO_PERIPH_A, 0);
 	at73c213_data.dac_clk = gclk;
 
 err_set_clk:
@@ -154,7 +165,7 @@ static int __init atstk1003_init(void)
 	at32_add_device_spi(1, spi1_board_info, ARRAY_SIZE(spi1_board_info));
 #endif
 #ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
-	at32_add_device_mci(0, NULL);
+	at32_add_device_mci(0, &mci0_data);
 #endif
 	at32_add_device_usba(0, NULL);
 #ifndef CONFIG_BOARD_ATSTK100X_SW3_CUSTOM
diff --git a/arch/avr32/boards/atstk1000/atstk1004.c b/arch/avr32/boards/atstk1000/atstk1004.c
index 50a5273e5916..248ef237c167 100644
--- a/arch/avr32/boards/atstk1000/atstk1004.c
+++ b/arch/avr32/boards/atstk1000/atstk1004.c
@@ -21,6 +21,7 @@
 #include <video/atmel_lcdc.h>
 
 #include <asm/setup.h>
+#include <asm/atmel-mci.h>
 
 #include <mach/at32ap700x.h>
 #include <mach/board.h>
@@ -71,6 +72,16 @@ static struct spi_board_info spi1_board_info[] __initdata = { {
 } };
 #endif
 
+#ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
+static struct mci_platform_data __initdata mci0_data = {
+	.slot[0] = {
+		.bus_width	= 4,
+		.detect_pin	= -ENODEV,
+		.wp_pin		= -ENODEV,
+	},
+};
+#endif
+
 #ifdef CONFIG_BOARD_ATSTK1000_EXTDAC
 static void __init atstk1004_setup_extdac(void)
 {
@@ -89,7 +100,7 @@ static void __init atstk1004_setup_extdac(void)
 		goto err_set_clk;
 	}
 
-	at32_select_periph(GPIO_PIN_PA(30), GPIO_PERIPH_A, 0);
+	at32_select_periph(GPIO_PIOA_BASE, (1 << 30), GPIO_PERIPH_A, 0);
 	at73c213_data.dac_clk = gclk;
 
 err_set_clk:
@@ -137,10 +148,11 @@ static int __init atstk1004_init(void)
 	at32_add_device_spi(1, spi1_board_info, ARRAY_SIZE(spi1_board_info));
 #endif
 #ifndef CONFIG_BOARD_ATSTK100X_SW2_CUSTOM
-	at32_add_device_mci(0, NULL);
+	at32_add_device_mci(0, &mci0_data);
 #endif
 	at32_add_device_lcdc(0, &atstk1000_lcdc_data,
-			     fbmem_start, fbmem_size, 0);
+			     fbmem_start, fbmem_size,
+			     ATMEL_LCDC_PRI_24BIT | ATMEL_LCDC_PRI_CONTROL);
 	at32_add_device_usba(0, NULL);
 #ifndef CONFIG_BOARD_ATSTK100X_SW3_CUSTOM
 	at32_add_device_ssc(0, ATMEL_SSC_TX);
diff --git a/arch/avr32/boot/images/.gitignore b/arch/avr32/boot/images/.gitignore
new file mode 100644
index 000000000000..64ea9d0141d2
--- /dev/null
+++ b/arch/avr32/boot/images/.gitignore
@@ -0,0 +1,4 @@
+uImage
+uImage.srec
+vmlinux.cso
+sfdwarf.log
diff --git a/arch/avr32/include/asm/atmel-mci.h b/arch/avr32/include/asm/atmel-mci.h
index c2ea6e1c9aa1..59f3fadd0b68 100644
--- a/arch/avr32/include/asm/atmel-mci.h
+++ b/arch/avr32/include/asm/atmel-mci.h
@@ -1,9 +1,39 @@
 #ifndef __ASM_AVR32_ATMEL_MCI_H
 #define __ASM_AVR32_ATMEL_MCI_H
 
-struct mci_platform_data {
+#define ATMEL_MCI_MAX_NR_SLOTS	2
+
+struct dma_slave;
+
+/**
+ * struct mci_slot_pdata - board-specific per-slot configuration
+ * @bus_width: Number of data lines wired up the slot
+ * @detect_pin: GPIO pin wired to the card detect switch
+ * @wp_pin: GPIO pin wired to the write protect sensor
+ *
+ * If a given slot is not present on the board, @bus_width should be
+ * set to 0. The other fields are ignored in this case.
+ *
+ * Any pins that aren't available should be set to a negative value.
+ *
+ * Note that support for multiple slots is experimental -- some cards
+ * might get upset if we don't get the clock management exactly right.
+ * But in most cases, it should work just fine.
+ */
+struct mci_slot_pdata {
+	unsigned int		bus_width;
 	int			detect_pin;
 	int			wp_pin;
 };
 
+/**
+ * struct mci_platform_data - board-specific MMC/SDcard configuration
+ * @dma_slave: DMA slave interface to use in data transfers, or NULL.
+ * @slot: Per-slot configuration data.
+ */
+struct mci_platform_data {
+	struct dma_slave	*dma_slave;
+	struct mci_slot_pdata	slot[ATMEL_MCI_MAX_NR_SLOTS];
+};
+
 #endif /* __ASM_AVR32_ATMEL_MCI_H */
diff --git a/arch/avr32/include/asm/byteorder.h b/arch/avr32/include/asm/byteorder.h
index d77b48ba7338..8e3af02076dd 100644
--- a/arch/avr32/include/asm/byteorder.h
+++ b/arch/avr32/include/asm/byteorder.h
@@ -7,6 +7,9 @@
 #include <asm/types.h>
 #include <linux/compiler.h>
 
+#define __BIG_ENDIAN
+#define __SWAB_64_THRU_32__
+
 #ifdef __CHECKER__
 extern unsigned long __builtin_bswap_32(unsigned long x);
 extern unsigned short __builtin_bswap_16(unsigned short x);
@@ -17,15 +20,18 @@ extern unsigned short __builtin_bswap_16(unsigned short x);
  * the result.
  */
 #if !(__GNUC__ == 4 && __GNUC_MINOR__ < 2)
-#define __arch__swab32(x) __builtin_bswap_32(x)
-#define __arch__swab16(x) __builtin_bswap_16(x)
-#endif
+static inline __attribute_const__ __u16 __arch_swab16(__u16 val)
+{
+	return __builtin_bswap_16(val);
+}
+#define __arch_swab16 __arch_swab16
 
-#if !defined(__STRICT_ANSI__) || defined(__KERNEL__)
-# define __BYTEORDER_HAS_U64__
-# define __SWAB_64_THRU_32__
+static inline __attribute_const__ __u32 __arch_swab32(__u32 val)
+{
+	return __builtin_bswap_32(val);
+}
+#define __arch_swab32 __arch_swab32
 #endif
 
-#include <linux/byteorder/big_endian.h>
-
+#include <linux/byteorder.h>
 #endif /* __ASM_AVR32_BYTEORDER_H */
diff --git a/arch/avr32/include/asm/io.h b/arch/avr32/include/asm/io.h
index a520f77ead96..22c97ef92201 100644
--- a/arch/avr32/include/asm/io.h
+++ b/arch/avr32/include/asm/io.h
@@ -160,6 +160,14 @@ BUILDIO_IOPORT(l, u32)
 #define readw_relaxed			readw
 #define readl_relaxed			readl
 
+#define readb_be			__raw_readb
+#define readw_be			__raw_readw
+#define readl_be			__raw_readl
+
+#define writeb_be			__raw_writeb
+#define writew_be			__raw_writew
+#define writel_be			__raw_writel
+
 #define __BUILD_MEMORY_STRING(bwl, type)				\
 static inline void writes##bwl(volatile void __iomem *addr,		\
 			       const void *data, unsigned int count)	\
diff --git a/arch/avr32/kernel/.gitignore b/arch/avr32/kernel/.gitignore
new file mode 100644
index 000000000000..c5f676c3c224
--- /dev/null
+++ b/arch/avr32/kernel/.gitignore
@@ -0,0 +1 @@
+vmlinux.lds
diff --git a/arch/avr32/kernel/asm-offsets.c b/arch/avr32/kernel/asm-offsets.c
index e4796c67a831..d6a8193a1d2f 100644
--- a/arch/avr32/kernel/asm-offsets.c
+++ b/arch/avr32/kernel/asm-offsets.c
@@ -4,6 +4,8 @@
  * to extract and format the required data.
  */
 
+#include <linux/mm.h>
+#include <linux/sched.h>
 #include <linux/thread_info.h>
 #include <linux/kbuild.h>
 
@@ -17,4 +19,8 @@ void foo(void)
 	OFFSET(TI_rar_saved, thread_info, rar_saved);
 	OFFSET(TI_rsr_saved, thread_info, rsr_saved);
 	OFFSET(TI_restart_block, thread_info, restart_block);
+	BLANK();
+	OFFSET(TSK_active_mm, task_struct, active_mm);
+	BLANK();
+	OFFSET(MM_pgd, mm_struct, pgd);
 }
diff --git a/arch/avr32/kernel/avr32_ksyms.c b/arch/avr32/kernel/avr32_ksyms.c
index 84a7d44edc67..11e310c567a9 100644
--- a/arch/avr32/kernel/avr32_ksyms.c
+++ b/arch/avr32/kernel/avr32_ksyms.c
@@ -58,6 +58,7 @@ EXPORT_SYMBOL(find_first_zero_bit);
 EXPORT_SYMBOL(find_next_zero_bit);
 EXPORT_SYMBOL(find_first_bit);
 EXPORT_SYMBOL(find_next_bit);
+EXPORT_SYMBOL(generic_find_next_le_bit);
 EXPORT_SYMBOL(generic_find_next_zero_le_bit);
 
 /* I/O primitives (lib/io-*.S) */
diff --git a/arch/avr32/kernel/entry-avr32b.S b/arch/avr32/kernel/entry-avr32b.S
index 2b398cae110c..33d49377b8be 100644
--- a/arch/avr32/kernel/entry-avr32b.S
+++ b/arch/avr32/kernel/entry-avr32b.S
@@ -334,9 +334,64 @@ save_full_context_ex:
 
 	/* Low-level exception handlers */
 handle_critical:
+	/*
+	 * AT32AP700x errata:
+	 *
+	 * After a Java stack overflow or underflow trap, any CPU
+	 * memory access may cause erratic behavior. This will happen
+	 * when the four least significant bits of the JOSP system
+	 * register contains any value between 9 and 15 (inclusive).
+	 *
+	 * Possible workarounds:
+	 *   - Don't use the Java Extension Module
+	 *   - Ensure that the stack overflow and underflow trap
+	 *     handlers do not do any memory access or trigger any
+	 *     exceptions before the overflow/underflow condition is
+	 *     cleared (by incrementing or decrementing the JOSP)
+	 *   - Make sure that JOSP does not contain any problematic
+	 *     value before doing any exception or interrupt
+	 *     processing.
+	 *   - Set up a critical exception handler which writes a
+	 *     known-to-be-safe value, e.g. 4, to JOSP before doing
+	 *     any further processing.
+	 *
+	 * We'll use the last workaround for now since we cannot
+	 * guarantee that user space processes don't use Java mode.
+	 * Non-well-behaving userland will be terminated with extreme
+	 * prejudice.
+	 */
+#ifdef CONFIG_CPU_AT32AP700X
+	/*
+	 * There's a chance we can't touch memory, so temporarily
+	 * borrow PTBR to save the stack pointer while we fix things
+	 * up...
+	 */
+	mtsr	SYSREG_PTBR, sp
+	mov	sp, 4
+	mtsr	SYSREG_JOSP, sp
+	mfsr	sp, SYSREG_PTBR
+	sub	pc, -2
+
+	/* Push most of pt_regs on stack. We'll do the rest later */
 	sub	sp, 4
-	stmts	--sp, r0-lr
-	rcall	save_full_context_ex
+	pushm	r0-r12
+
+	/* PTBR mirrors current_thread_info()->task->active_mm->pgd */
+	get_thread_info r0
+	ld.w	r1, r0[TI_task]
+	ld.w	r2, r1[TSK_active_mm]
+	ld.w	r3, r2[MM_pgd]
+	mtsr	SYSREG_PTBR, r3
+#else
+	sub	sp, 4
+	pushm	r0-r12
+#endif
+	sub	r0, sp, -(14 * 4)
+	mov	r1, lr
+	mfsr	r2, SYSREG_RAR_EX
+	mfsr	r3, SYSREG_RSR_EX
+	pushm	r0-r3
+
 	mfsr	r12, SYSREG_ECR
 	mov	r11, sp
 	rcall	do_critical_exception
diff --git a/arch/avr32/kernel/process.c b/arch/avr32/kernel/process.c
index 2c08ac992ac3..134d5302b6dd 100644
--- a/arch/avr32/kernel/process.c
+++ b/arch/avr32/kernel/process.c
@@ -9,6 +9,7 @@
 #include <linux/module.h>
 #include <linux/kallsyms.h>
 #include <linux/fs.h>
+#include <linux/pm.h>
 #include <linux/ptrace.h>
 #include <linux/reboot.h>
 #include <linux/tick.h>
@@ -20,7 +21,7 @@
 
 #include <mach/pm.h>
 
-void (*pm_power_off)(void) = NULL;
+void (*pm_power_off)(void);
 EXPORT_SYMBOL(pm_power_off);
 
 /*
diff --git a/arch/avr32/kernel/setup.c b/arch/avr32/kernel/setup.c
index d8e623c426c1..5c7083916c33 100644
--- a/arch/avr32/kernel/setup.c
+++ b/arch/avr32/kernel/setup.c
@@ -283,6 +283,25 @@ static int __init early_parse_fbmem(char *p)
 }
 early_param("fbmem", early_parse_fbmem);
 
+/*
+ * Pick out the memory size.  We look for mem=size@start,
+ * where start and size are "size[KkMmGg]"
+ */
+static int __init early_mem(char *p)
+{
+	resource_size_t size, start;
+
+	start = system_ram->start;
+	size  = memparse(p, &p);
+	if (*p == '@')
+		start = memparse(p + 1, &p);
+
+	system_ram->start = start;
+	system_ram->end = system_ram->start + size - 1;
+	return 0;
+}
+early_param("mem", early_mem);
+
 static int __init parse_tag_core(struct tag *tag)
 {
 	if (tag->hdr.size > 2) {
diff --git a/arch/avr32/kernel/syscall-stubs.S b/arch/avr32/kernel/syscall-stubs.S
index 890286a1e62b..673178e235f3 100644
--- a/arch/avr32/kernel/syscall-stubs.S
+++ b/arch/avr32/kernel/syscall-stubs.S
@@ -109,3 +109,12 @@ __sys_epoll_pwait:
 	rcall	sys_epoll_pwait
 	sub	sp, -4
 	popm	pc
+
+	.global __sys_sync_file_range
+	.type	__sys_sync_file_range,@function
+__sys_sync_file_range:
+	pushm	lr
+	st.w	--sp, ARG6
+	rcall	sys_sync_file_range
+	sub	sp, -4
+	popm	pc
diff --git a/arch/avr32/kernel/syscall_table.S b/arch/avr32/kernel/syscall_table.S
index 478bda4c4a09..7ee0057613b3 100644
--- a/arch/avr32/kernel/syscall_table.S
+++ b/arch/avr32/kernel/syscall_table.S
@@ -275,7 +275,7 @@ sys_call_table:
 	.long	sys_set_robust_list
 	.long	sys_get_robust_list	/* 260 */
 	.long	__sys_splice
-	.long	sys_sync_file_range
+	.long	__sys_sync_file_range
 	.long	sys_tee
 	.long	sys_vmsplice
 	.long	__sys_epoll_pwait	/* 265 */
diff --git a/arch/avr32/kernel/traps.c b/arch/avr32/kernel/traps.c
index b835c4c01368..0d987373bc01 100644
--- a/arch/avr32/kernel/traps.c
+++ b/arch/avr32/kernel/traps.c
@@ -116,15 +116,15 @@ asmlinkage void do_nmi(unsigned long ecr, struct pt_regs *regs)
 	switch (ret) {
 	case NOTIFY_OK:
 	case NOTIFY_STOP:
-		return;
+		break;
 	case NOTIFY_BAD:
 		die("Fatal Non-Maskable Interrupt", regs, SIGINT);
 	default:
+		printk(KERN_ALERT "Got NMI, but nobody cared. Disabling...\n");
+		nmi_disable();
 		break;
 	}
-
-	printk(KERN_ALERT "Got NMI, but nobody cared. Disabling...\n");
-	nmi_disable();
+	nmi_exit();
 }
 
 asmlinkage void do_critical_exception(unsigned long ecr, struct pt_regs *regs)
diff --git a/arch/avr32/lib/findbit.S b/arch/avr32/lib/findbit.S
index c6b91dee857c..997b33b2288a 100644
--- a/arch/avr32/lib/findbit.S
+++ b/arch/avr32/lib/findbit.S
@@ -123,6 +123,36 @@ ENTRY(find_next_bit)
 	brgt	1b
 	retal	r11
 
+ENTRY(generic_find_next_le_bit)
+	lsr	r8, r10, 5
+	sub	r9, r11, r10
+	retle	r11
+
+	lsl	r8, 2
+	add	r12, r8
+	andl	r10, 31, COH
+	breq	1f
+
+	/* offset is not word-aligned. Handle the first (32 - r10) bits */
+	ldswp.w	r8, r12[0]
+	sub	r12, -4
+	lsr	r8, r8, r10
+	brne	.L_found
+
+	/* r9 = r9 - (32 - r10) = r9 + r10 - 32 */
+	add	r9, r10
+	sub	r9, 32
+	retle	r11
+
+	/* Main loop. offset must be word-aligned */
+1:	ldswp.w	r8, r12[0]
+	cp.w	r8, 0
+	brne	.L_found
+	sub	r12, -4
+	sub	r9, 32
+	brgt	1b
+	retal	r11
+
 ENTRY(generic_find_next_zero_le_bit)
 	lsr	r8, r10, 5
 	sub	r9, r11, r10
diff --git a/arch/avr32/mach-at32ap/at32ap700x.c b/arch/avr32/mach-at32ap/at32ap700x.c
index e01dbe4ebb40..813b6844cdf6 100644
--- a/arch/avr32/mach-at32ap/at32ap700x.c
+++ b/arch/avr32/mach-at32ap/at32ap700x.c
@@ -82,8 +82,9 @@ static struct platform_device _name##_id##_device = {		\
 	.num_resources	= ARRAY_SIZE(_name##_id##_resource),	\
 }
 
-#define select_peripheral(pin, periph, flags)			\
-	at32_select_periph(GPIO_PIN_##pin, GPIO_##periph, flags)
+#define select_peripheral(port, pin_mask, periph, flags)	\
+	at32_select_periph(GPIO_##port##_BASE, pin_mask,	\
+			   GPIO_##periph, flags)
 
 #define DEV_CLK(_name, devname, bus, _index)			\
 static struct clk devname##_##_name = {				\
@@ -871,6 +872,7 @@ static struct clk atmel_psif1_pclk = {
 struct platform_device *__init at32_add_device_psif(unsigned int id)
 {
 	struct platform_device *pdev;
+	u32 pin_mask;
 
 	if (!(id == 0 || id == 1))
 		return NULL;
@@ -881,20 +883,22 @@ struct platform_device *__init at32_add_device_psif(unsigned int id)
 
 	switch (id) {
 	case 0:
+		pin_mask  = (1 << 8) | (1 << 9); /* CLOCK & DATA */
+
 		if (platform_device_add_resources(pdev, atmel_psif0_resource,
 					ARRAY_SIZE(atmel_psif0_resource)))
 			goto err_add_resources;
 		atmel_psif0_pclk.dev = &pdev->dev;
-		select_peripheral(PA(8), PERIPH_A, 0); /* CLOCK */
-		select_peripheral(PA(9), PERIPH_A, 0); /* DATA  */
+		select_peripheral(PIOA, pin_mask, PERIPH_A, 0);
 		break;
 	case 1:
+		pin_mask  = (1 << 11) | (1 << 12); /* CLOCK & DATA */
+
 		if (platform_device_add_resources(pdev, atmel_psif1_resource,
 					ARRAY_SIZE(atmel_psif1_resource)))
 			goto err_add_resources;
 		atmel_psif1_pclk.dev = &pdev->dev;
-		select_peripheral(PB(11), PERIPH_A, 0); /* CLOCK */
-		select_peripheral(PB(12), PERIPH_A, 0); /* DATA  */
+		select_peripheral(PIOB, pin_mask, PERIPH_A, 0);
 		break;
 	default:
 		return NULL;
@@ -958,26 +962,30 @@ DEV_CLK(usart, atmel_usart3, pba, 6);
 
 static inline void configure_usart0_pins(void)
 {
-	select_peripheral(PA(8),  PERIPH_B, 0);	/* RXD	*/
-	select_peripheral(PA(9),  PERIPH_B, 0);	/* TXD	*/
+	u32 pin_mask = (1 << 8) | (1 << 9); /* RXD & TXD */
+
+	select_peripheral(PIOA, pin_mask, PERIPH_B, 0);
 }
 
 static inline void configure_usart1_pins(void)
 {
-	select_peripheral(PA(17), PERIPH_A, 0);	/* RXD	*/
-	select_peripheral(PA(18), PERIPH_A, 0);	/* TXD	*/
+	u32 pin_mask = (1 << 17) | (1 << 18); /* RXD & TXD */
+
+	select_peripheral(PIOA, pin_mask, PERIPH_A, 0);
 }
 
 static inline void configure_usart2_pins(void)
 {
-	select_peripheral(PB(26), PERIPH_B, 0);	/* RXD	*/
-	select_peripheral(PB(27), PERIPH_B, 0);	/* TXD	*/
+	u32 pin_mask = (1 << 26) | (1 << 27); /* RXD & TXD */
+
+	select_peripheral(PIOB, pin_mask, PERIPH_B, 0);
 }
 
 static inline void configure_usart3_pins(void)
 {
-	select_peripheral(PB(18), PERIPH_B, 0);	/* RXD	*/
-	select_peripheral(PB(17), PERIPH_B, 0);	/* TXD	*/
+	u32 pin_mask = (1 << 18) | (1 << 17); /* RXD & TXD */
+
+	select_peripheral(PIOB, pin_mask, PERIPH_B, 0);
 }
 
 static struct platform_device *__initdata at32_usarts[4];
@@ -1057,59 +1065,69 @@ struct platform_device *__init
 at32_add_device_eth(unsigned int id, struct eth_platform_data *data)
 {
 	struct platform_device *pdev;
+	u32 pin_mask;
 
 	switch (id) {
 	case 0:
 		pdev = &macb0_device;
 
-		select_peripheral(PC(3),  PERIPH_A, 0);	/* TXD0	*/
-		select_peripheral(PC(4),  PERIPH_A, 0);	/* TXD1	*/
-		select_peripheral(PC(7),  PERIPH_A, 0);	/* TXEN	*/
-		select_peripheral(PC(8),  PERIPH_A, 0);	/* TXCK */
-		select_peripheral(PC(9),  PERIPH_A, 0);	/* RXD0	*/
-		select_peripheral(PC(10), PERIPH_A, 0);	/* RXD1	*/
-		select_peripheral(PC(13), PERIPH_A, 0);	/* RXER	*/
-		select_peripheral(PC(15), PERIPH_A, 0);	/* RXDV	*/
-		select_peripheral(PC(16), PERIPH_A, 0);	/* MDC	*/
-		select_peripheral(PC(17), PERIPH_A, 0);	/* MDIO	*/
+		pin_mask  = (1 << 3);	/* TXD0 */
+		pin_mask |= (1 << 4);	/* TXD1 */
+		pin_mask |= (1 << 7);	/* TXEN */
+		pin_mask |= (1 << 8);	/* TXCK */
+		pin_mask |= (1 << 9);	/* RXD0 */
+		pin_mask |= (1 << 10);	/* RXD1 */
+		pin_mask |= (1 << 13);	/* RXER */
+		pin_mask |= (1 << 15);	/* RXDV */
+		pin_mask |= (1 << 16);	/* MDC  */
+		pin_mask |= (1 << 17);	/* MDIO */
 
 		if (!data->is_rmii) {
-			select_peripheral(PC(0),  PERIPH_A, 0);	/* COL	*/
-			select_peripheral(PC(1),  PERIPH_A, 0);	/* CRS	*/
-			select_peripheral(PC(2),  PERIPH_A, 0);	/* TXER	*/
-			select_peripheral(PC(5),  PERIPH_A, 0);	/* TXD2	*/
-			select_peripheral(PC(6),  PERIPH_A, 0);	/* TXD3 */
-			select_peripheral(PC(11), PERIPH_A, 0);	/* RXD2	*/
-			select_peripheral(PC(12), PERIPH_A, 0);	/* RXD3	*/
-			select_peripheral(PC(14), PERIPH_A, 0);	/* RXCK	*/
-			select_peripheral(PC(18), PERIPH_A, 0);	/* SPD	*/
+			pin_mask |= (1 << 0);	/* COL  */
+			pin_mask |= (1 << 1);	/* CRS  */
+			pin_mask |= (1 << 2);	/* TXER */
+			pin_mask |= (1 << 5);	/* TXD2 */
+			pin_mask |= (1 << 6);	/* TXD3 */
+			pin_mask |= (1 << 11);	/* RXD2 */
+			pin_mask |= (1 << 12);	/* RXD3 */
+			pin_mask |= (1 << 14);	/* RXCK */
+			pin_mask |= (1 << 18);	/* SPD  */
 		}
+
+		select_peripheral(PIOC, pin_mask, PERIPH_A, 0);
+
 		break;
 
 	case 1:
 		pdev = &macb1_device;
 
-		select_peripheral(PD(13), PERIPH_B, 0);		/* TXD0	*/
-		select_peripheral(PD(14), PERIPH_B, 0);		/* TXD1	*/
-		select_peripheral(PD(11), PERIPH_B, 0);		/* TXEN	*/
-		select_peripheral(PD(12), PERIPH_B, 0);		/* TXCK */
-		select_peripheral(PD(10), PERIPH_B, 0);		/* RXD0	*/
-		select_peripheral(PD(6),  PERIPH_B, 0);		/* RXD1	*/
-		select_peripheral(PD(5),  PERIPH_B, 0);		/* RXER	*/
-		select_peripheral(PD(4),  PERIPH_B, 0);		/* RXDV	*/
-		select_peripheral(PD(3),  PERIPH_B, 0);		/* MDC	*/
-		select_peripheral(PD(2),  PERIPH_B, 0);		/* MDIO	*/
+		pin_mask  = (1 << 13);	/* TXD0 */
+		pin_mask |= (1 << 14);	/* TXD1 */
+		pin_mask |= (1 << 11);	/* TXEN */
+		pin_mask |= (1 << 12);	/* TXCK */
+		pin_mask |= (1 << 10);	/* RXD0 */
+		pin_mask |= (1 << 6);	/* RXD1 */
+		pin_mask |= (1 << 5);	/* RXER */
+		pin_mask |= (1 << 4);	/* RXDV */
+		pin_mask |= (1 << 3);	/* MDC  */
+		pin_mask |= (1 << 2);	/* MDIO */
+
+		if (!data->is_rmii)
+			pin_mask |= (1 << 15);	/* SPD  */
+
+		select_peripheral(PIOD, pin_mask, PERIPH_B, 0);
 
 		if (!data->is_rmii) {
-			select_peripheral(PC(19), PERIPH_B, 0);	/* COL	*/
-			select_peripheral(PC(23), PERIPH_B, 0);	/* CRS	*/
-			select_peripheral(PC(26), PERIPH_B, 0);	/* TXER	*/
-			select_peripheral(PC(27), PERIPH_B, 0);	/* TXD2	*/
-			select_peripheral(PC(28), PERIPH_B, 0);	/* TXD3 */