diff --git a/arch/avr32/boards/atngw100/flash.c b/arch/avr32/boards/atngw100/flash.c
index f9b32a8eab9..b07ae63aa54 100644
--- a/arch/avr32/boards/atngw100/flash.c
+++ b/arch/avr32/boards/atngw100/flash.c
@@ -15,7 +15,7 @@
 
 #include <asm/arch/smc.h>
 
-static struct smc_config flash_config __initdata = {
+static struct smc_timing flash_timing __initdata = {
 	.ncs_read_setup		= 0,
 	.nrd_setup		= 40,
 	.ncs_write_setup	= 0,
@@ -28,7 +28,9 @@ static struct smc_config flash_config __initdata = {
 
 	.read_cycle		= 120,
 	.write_cycle		= 120,
+};
 
+static struct smc_config flash_config __initdata = {
 	.bus_width		= 2,
 	.nrd_controlled		= 1,
 	.nwe_controlled		= 1,
@@ -82,6 +84,7 @@ static int __init atngw100_flash_init(void)
 {
 	int ret;
 
+	smc_set_timing(&flash_config, &flash_timing);
 	ret = smc_set_configuration(0, &flash_config);
 	if (ret < 0) {
 		printk(KERN_ERR "atngw100: failed to set NOR flash timing\n");
diff --git a/arch/avr32/boards/atngw100/setup.c b/arch/avr32/boards/atngw100/setup.c
index ef801563bbf..52987c81d66 100644
--- a/arch/avr32/boards/atngw100/setup.c
+++ b/arch/avr32/boards/atngw100/setup.c
@@ -125,8 +125,11 @@ static struct platform_device ngw_gpio_leds = {
 };
 
 static struct i2c_gpio_platform_data i2c_gpio_data = {
-	.sda_pin	= GPIO_PIN_PA(6),
-	.scl_pin	= GPIO_PIN_PA(7),
+	.sda_pin		= GPIO_PIN_PA(6),
+	.scl_pin		= GPIO_PIN_PA(7),
+	.sda_is_open_drain	= 1,
+	.scl_is_open_drain	= 1,
+	.udelay			= 2,	/* close to 100 kHz */
 };
 
 static struct platform_device i2c_gpio_device = {
@@ -154,6 +157,7 @@ static int __init atngw100_init(void)
 	set_hw_addr(at32_add_device_eth(1, &eth_data[1]));
 
 	at32_add_device_spi(0, spi0_board_info, ARRAY_SIZE(spi0_board_info));
+	at32_add_device_usba(0, NULL);
 
 	for (i = 0; i < ARRAY_SIZE(ngw_leds); i++) {
 		at32_select_gpio(ngw_leds[i].gpio,
@@ -161,8 +165,10 @@ static int __init atngw100_init(void)
 	}
 	platform_device_register(&ngw_gpio_leds);
 
-	at32_select_gpio(i2c_gpio_data.sda_pin, 0);
-	at32_select_gpio(i2c_gpio_data.scl_pin, 0);
+	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,
+		AT32_GPIOF_MULTIDRV | AT32_GPIOF_OUTPUT | AT32_GPIOF_HIGH);
 	platform_device_register(&i2c_gpio_device);
 
 	return 0;
diff --git a/arch/avr32/boards/atstk1000/atstk1002.c b/arch/avr32/boards/atstk1000/atstk1002.c
index c9981b731ef..6b9e466104a 100644
--- a/arch/avr32/boards/atstk1000/atstk1002.c
+++ b/arch/avr32/boards/atstk1000/atstk1002.c
@@ -241,6 +241,7 @@ static int __init atstk1002_init(void)
 	at32_add_device_lcdc(0, &atstk1000_lcdc_data,
 			     fbmem_start, fbmem_size);
 #endif
+	at32_add_device_usba(0, NULL);
 #ifndef CONFIG_BOARD_ATSTK1002_SW3_CUSTOM
 	at32_add_device_ssc(0, ATMEL_SSC_TX);
 #endif
diff --git a/arch/avr32/boards/atstk1000/flash.c b/arch/avr32/boards/atstk1000/flash.c
index aac4300cca1..3d0a102ad45 100644
--- a/arch/avr32/boards/atstk1000/flash.c
+++ b/arch/avr32/boards/atstk1000/flash.c
@@ -15,7 +15,7 @@
 
 #include <asm/arch/smc.h>
 
-static struct smc_config flash_config __initdata = {
+static struct smc_timing flash_timing __initdata = {
 	.ncs_read_setup		= 0,
 	.nrd_setup		= 40,
 	.ncs_write_setup	= 0,
@@ -28,7 +28,9 @@ static struct smc_config flash_config __initdata = {
 
 	.read_cycle		= 120,
 	.write_cycle		= 120,
+};
 
+static struct smc_config flash_config __initdata = {
 	.bus_width		= 2,
 	.nrd_controlled		= 1,
 	.nwe_controlled		= 1,
@@ -82,6 +84,7 @@ static int __init atstk1000_flash_init(void)
 {
 	int ret;
 
+	smc_set_timing(&flash_config, &flash_timing);
 	ret = smc_set_configuration(0, &flash_config);
 	if (ret < 0) {
 		printk(KERN_ERR "atstk1000: failed to set NOR flash timing\n");
diff --git a/arch/avr32/kernel/Makefile b/arch/avr32/kernel/Makefile
index 90e5afff54a..989fcd1fef7 100644
--- a/arch/avr32/kernel/Makefile
+++ b/arch/avr32/kernel/Makefile
@@ -11,8 +11,3 @@ obj-y				+= signal.o sys_avr32.o process.o time.o
 obj-y				+= init_task.o switch_to.o cpu.o
 obj-$(CONFIG_MODULES)		+= module.o avr32_ksyms.o
 obj-$(CONFIG_KPROBES)		+= kprobes.o
-
-USE_STANDARD_AS_RULE		:= true
-
-%.lds: %.lds.c FORCE
-	$(call if_changed_dep,cpp_lds_S)
diff --git a/arch/avr32/kernel/entry-avr32b.S b/arch/avr32/kernel/entry-avr32b.S
index 42657f1703b..ccadfd9b438 100644
--- a/arch/avr32/kernel/entry-avr32b.S
+++ b/arch/avr32/kernel/entry-avr32b.S
@@ -159,11 +159,18 @@ handle_vmalloc_miss:
 
 	.section .scall.text,"ax",@progbits
 system_call:
+#ifdef CONFIG_PREEMPT
+	mask_interrupts
+#endif
 	pushm	r12		/* r12_orig */
 	stmts	--sp, r0-lr
-	zero_fp
+
 	mfsr	r0, SYSREG_RAR_SUP
 	mfsr	r1, SYSREG_RSR_SUP
+#ifdef CONFIG_PREEMPT
+	unmask_interrupts
+#endif
+	zero_fp
 	stm	--sp, r0-r1
 
 	/* check for syscall tracing */
@@ -638,6 +645,13 @@ irq_level\level:
 	stmts	--sp,r0-lr
 	mfsr	r8, rar_int\level
 	mfsr	r9, rsr_int\level
+
+#ifdef CONFIG_PREEMPT
+	sub	r11, pc, (. - system_call)
+	cp.w	r11, r8
+	breq	4f
+#endif
+
 	pushm	r8-r9
 
 	mov	r11, sp
@@ -668,6 +682,16 @@ irq_level\level:
 	sub	sp, -4		/* ignore r12_orig */
 	rete
 
+#ifdef CONFIG_PREEMPT
+4:	mask_interrupts
+	mfsr	r8, rsr_int\level
+	sbr	r8, 16
+	mtsr	rsr_int\level, r8
+	ldmts	sp++, r0-lr
+	sub	sp, -4		/* ignore r12_orig */
+	rete
+#endif
+
 2:	get_thread_info	r0
 	ld.w	r1, r0[TI_flags]
 	bld	r1, TIF_CPU_GOING_TO_SLEEP
diff --git a/arch/avr32/kernel/setup.c b/arch/avr32/kernel/setup.c
index d08b0bc6b2b..4b4c1884e1c 100644
--- a/arch/avr32/kernel/setup.c
+++ b/arch/avr32/kernel/setup.c
@@ -248,7 +248,7 @@ static int __init early_parse_fbmem(char *p)
 
 	fbmem_size = memparse(p, &p);
 	if (*p == '@') {
-		fbmem_start = memparse(p, &p);
+		fbmem_start = memparse(p + 1, &p);
 		ret = add_reserved_region(fbmem_start,
 					  fbmem_start + fbmem_size - 1,
 					  "Framebuffer");
diff --git a/arch/avr32/kernel/vmlinux.lds.c b/arch/avr32/kernel/vmlinux.lds.S
similarity index 95%
rename from arch/avr32/kernel/vmlinux.lds.c
rename to arch/avr32/kernel/vmlinux.lds.S
index db0438f35c0..ce9ac965988 100644
--- a/arch/avr32/kernel/vmlinux.lds.c
+++ b/arch/avr32/kernel/vmlinux.lds.S
@@ -9,6 +9,8 @@
  */
 #define LOAD_OFFSET 0x00000000
 #include <asm-generic/vmlinux.lds.h>
+#include <asm/cache.h>
+#include <asm/thread_info.h>
 
 OUTPUT_FORMAT("elf32-avr32", "elf32-avr32", "elf32-avr32")
 OUTPUT_ARCH(avr32)
@@ -58,11 +60,10 @@ SECTIONS
 			*(.init.ramfs)
 		__initramfs_end = .;
 #endif
-		. = ALIGN(4096);
+		. = ALIGN(PAGE_SIZE);
 		__init_end = .;
 	}
 
-	. = ALIGN(8192);
 	.text		: AT(ADDR(.text) - LOAD_OFFSET) {
 		_evba = .;
 		_text = .;
@@ -96,7 +97,7 @@ SECTIONS
 
 	RODATA
 
-	. = ALIGN(8192);
+	. = ALIGN(THREAD_SIZE);
 
 	.data		: AT(ADDR(.data) - LOAD_OFFSET) {
 		_data = .;
@@ -107,7 +108,7 @@ SECTIONS
 		*(.data.init_task)
 
 		/* Then, the cacheline aligned data */
-		. = ALIGN(32);
+		. = ALIGN(L1_CACHE_BYTES);
 		*(.data.cacheline_aligned)
 
 		/* And the rest... */
diff --git a/arch/avr32/mach-at32ap/at32ap7000.c b/arch/avr32/mach-at32ap/at32ap7000.c
index 64cc5583ddf..f6d154ca4d2 100644
--- a/arch/avr32/mach-at32ap/at32ap7000.c
+++ b/arch/avr32/mach-at32ap/at32ap7000.c
@@ -25,12 +25,6 @@
 #include "pio.h"
 #include "pm.h"
 
-/*
- * We can reduce the code size a bit by using a constant here. Since
- * this file is completely chip-specific, it's safe to not use
- * ioremap. Generic drivers should of course never do this.
- */
-#define AT32_PM_BASE	0xfff00000
 
 #define PBMEM(base)					\
 	{						\
@@ -1167,6 +1161,72 @@ at32_add_device_ssc(unsigned int id, unsigned int flags)
 	return pdev;
 }
 
+/* --------------------------------------------------------------------
+ *  USB Device Controller
+ * -------------------------------------------------------------------- */
+static struct resource usba0_resource[] __initdata = {
+	{
+		.start		= 0xff300000,
+		.end		= 0xff3fffff,
+		.flags		= IORESOURCE_MEM,
+	}, {
+		.start		= 0xfff03000,
+		.end		= 0xfff033ff,
+		.flags		= IORESOURCE_MEM,
+	},
+	IRQ(31),
+};
+static struct clk usba0_pclk = {
+	.name		= "pclk",
+	.parent		= &pbb_clk,
+	.mode		= pbb_clk_mode,
+	.get_rate	= pbb_clk_get_rate,
+	.index		= 12,
+};
+static struct clk usba0_hclk = {
+	.name		= "hclk",
+	.parent		= &hsb_clk,
+	.mode		= hsb_clk_mode,
+	.get_rate	= hsb_clk_get_rate,
+	.index		= 6,
+};
+
+struct platform_device *__init
+at32_add_device_usba(unsigned int id, struct usba_platform_data *data)
+{
+	struct platform_device *pdev;
+
+	if (id != 0)
+		return NULL;
+
+	pdev = platform_device_alloc("atmel_usba_udc", 0);
+	if (!pdev)
+		return NULL;
+
+	if (platform_device_add_resources(pdev, usba0_resource,
+					  ARRAY_SIZE(usba0_resource)))
+		goto out_free_pdev;
+
+	if (data) {
+		if (platform_device_add_data(pdev, data, sizeof(*data)))
+			goto out_free_pdev;
+
+		if (data->vbus_pin != GPIO_PIN_NONE)
+			at32_select_gpio(data->vbus_pin, 0);
+	}
+
+	usba0_pclk.dev = &pdev->dev;
+	usba0_hclk.dev = &pdev->dev;
+
+	platform_device_add(pdev);
+
+	return pdev;
+
+out_free_pdev:
+	platform_device_put(pdev);
+	return NULL;
+}
+
 /* --------------------------------------------------------------------
  *  GCLK
  * -------------------------------------------------------------------- */
@@ -1252,6 +1312,8 @@ struct clk *at32_clock_list[] = {
 	&ssc0_pclk,
 	&ssc1_pclk,
 	&ssc2_pclk,
+	&usba0_hclk,
+	&usba0_pclk,
 	&gclk0,
 	&gclk1,
 	&gclk2,
diff --git a/arch/avr32/mach-at32ap/clock.c b/arch/avr32/mach-at32ap/clock.c
index 0f8c89c9f83..4642117cc9a 100644
--- a/arch/avr32/mach-at32ap/clock.c
+++ b/arch/avr32/mach-at32ap/clock.c
@@ -150,3 +150,119 @@ struct clk *clk_get_parent(struct clk *clk)
 	return clk->parent;
 }
 EXPORT_SYMBOL(clk_get_parent);
+
+
+
+#ifdef CONFIG_DEBUG_FS
+
+/* /sys/kernel/debug/at32ap_clk */
+
+#include <linux/io.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+#include "pm.h"
+
+
+#define	NEST_DELTA	2
+#define	NEST_MAX	6
+
+struct clkinf {
+	struct seq_file	*s;
+	unsigned	nest;
+};
+
+static void
+dump_clock(struct clk *parent, struct clkinf *r)
+{
+	unsigned	nest = r->nest;
+	char		buf[16 + NEST_MAX];
+	struct clk	*clk;
+	unsigned	i;
+
+	/* skip clocks coupled to devices that aren't registered */
+	if (parent->dev && !parent->dev->bus_id[0] && !parent->users)
+		return;
+
+	/* <nest spaces> name <pad to end> */
+	memset(buf, ' ', sizeof(buf) - 1);
+	buf[sizeof(buf) - 1] = 0;
+	i = strlen(parent->name);
+	memcpy(buf + nest, parent->name,
+			min(i, (unsigned)(sizeof(buf) - 1 - nest)));
+
+	seq_printf(r->s, "%s%c users=%2d %-3s %9ld Hz",
+		buf, parent->set_parent ? '*' : ' ',
+		parent->users,
+		parent->users ? "on" : "off",	/* NOTE: not-paranoid!! */
+		clk_get_rate(parent));
+	if (parent->dev)
+		seq_printf(r->s, ", for %s", parent->dev->bus_id);
+	seq_printf(r->s, "\n");
+
+	/* cost of this scan is small, but not linear... */
+	r->nest = nest + NEST_DELTA;
+	for (i = 3; i < at32_nr_clocks; i++) {
+		clk = at32_clock_list[i];
+		if (clk->parent == parent)
+			dump_clock(clk, r);
+	}
+	r->nest = nest;
+}
+
+static int clk_show(struct seq_file *s, void *unused)
+{
+	struct clkinf	r;
+	int		i;
+
+	/* show all the power manager registers */
+	seq_printf(s, "MCCTRL  = %8x\n", pm_readl(MCCTRL));
+	seq_printf(s, "CKSEL   = %8x\n", pm_readl(CKSEL));
+	seq_printf(s, "CPUMASK = %8x\n", pm_readl(CPU_MASK));
+	seq_printf(s, "HSBMASK = %8x\n", pm_readl(HSB_MASK));
+	seq_printf(s, "PBAMASK = %8x\n", pm_readl(PBA_MASK));
+	seq_printf(s, "PBBMASK = %8x\n", pm_readl(PBB_MASK));
+	seq_printf(s, "PLL0    = %8x\n", pm_readl(PLL0));
+	seq_printf(s, "PLL1    = %8x\n", pm_readl(PLL1));
+	seq_printf(s, "IMR     = %8x\n", pm_readl(IMR));
+	for (i = 0; i < 8; i++) {
+		if (i == 5)
+			continue;
+		seq_printf(s, "GCCTRL%d = %8x\n", i, pm_readl(GCCTRL(i)));
+	}
+
+	seq_printf(s, "\n");
+
+	/* show clock tree as derived from the three oscillators
+	 * we "know" are at the head of the list
+	 */
+	r.s = s;
+	r.nest = 0;
+	dump_clock(at32_clock_list[0], &r);
+	dump_clock(at32_clock_list[1], &r);
+	dump_clock(at32_clock_list[2], &r);
+
+	return 0;
+}
+
+static int clk_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, clk_show, NULL);
+}
+
+static const struct file_operations clk_operations = {
+	.open		= clk_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+
+static int __init clk_debugfs_init(void)
+{
+	(void) debugfs_create_file("at32ap_clk", S_IFREG | S_IRUGO,
+			NULL, NULL, &clk_operations);
+
+	return 0;
+}
+postcore_initcall(clk_debugfs_init);
+
+#endif
diff --git a/arch/avr32/mach-at32ap/hsmc.c b/arch/avr32/mach-at32ap/hsmc.c
index 5e22a750632..704607fbcc6 100644
--- a/arch/avr32/mach-at32ap/hsmc.c
+++ b/arch/avr32/mach-at32ap/hsmc.c
@@ -29,16 +29,25 @@ struct hsmc {
 
 static struct hsmc *hsmc;
 
-int smc_set_configuration(int cs, const struct smc_config *config)
+void smc_set_timing(struct smc_config *config,
+		    const struct smc_timing *timing)
 {
-	unsigned long mul;
-	unsigned long offset;
-	u32 setup, pulse, cycle, mode;
+	int recover;
+	int cycle;
 
-	if (!hsmc)
-		return -ENODEV;
-	if (cs >= NR_CHIP_SELECTS)
-		return -EINVAL;
+	unsigned long mul;
+
+	/* Reset all SMC timings */
+	config->ncs_read_setup	= 0;
+	config->nrd_setup	= 0;
+	config->ncs_write_setup	= 0;
+	config->nwe_setup	= 0;
+	config->ncs_read_pulse	= 0;
+	config->nrd_pulse	= 0;
+	config->ncs_write_pulse	= 0;
+	config->nwe_pulse	= 0;
+	config->read_cycle	= 0;
+	config->write_cycle	= 0;
 
 	/*
 	 * cycles = x / T = x * f
@@ -50,16 +59,102 @@ int smc_set_configuration(int cs, const struct smc_config *config)
 
 #define ns2cyc(x) ((((x) * mul) + 65535) >> 16)
 
-	setup = (HSMC_BF(NWE_SETUP, ns2cyc(config->nwe_setup))
-		 | HSMC_BF(NCS_WR_SETUP, ns2cyc(config->ncs_write_setup))
-		 | HSMC_BF(NRD_SETUP, ns2cyc(config->nrd_setup))
-		 | HSMC_BF(NCS_RD_SETUP, ns2cyc(config->ncs_read_setup)));
-	pulse = (HSMC_BF(NWE_PULSE, ns2cyc(config->nwe_pulse))
-		 | HSMC_BF(NCS_WR_PULSE, ns2cyc(config->ncs_write_pulse))
-		 | HSMC_BF(NRD_PULSE, ns2cyc(config->nrd_pulse))
-		 | HSMC_BF(NCS_RD_PULSE, ns2cyc(config->ncs_read_pulse)));
-	cycle = (HSMC_BF(NWE_CYCLE, ns2cyc(config->write_cycle))
-		 | HSMC_BF(NRD_CYCLE, ns2cyc(config->read_cycle)));
+	if (timing->ncs_read_setup > 0)
+		config->ncs_read_setup = ns2cyc(timing->ncs_read_setup);
+
+	if (timing->nrd_setup > 0)
+		config->nrd_setup = ns2cyc(timing->nrd_setup);
+
+	if (timing->ncs_write_setup > 0)
+		config->ncs_write_setup = ns2cyc(timing->ncs_write_setup);
+
+	if (timing->nwe_setup > 0)
+		config->nwe_setup = ns2cyc(timing->nwe_setup);
+
+	if (timing->ncs_read_pulse > 0)
+		config->ncs_read_pulse = ns2cyc(timing->ncs_read_pulse);
+
+	if (timing->nrd_pulse > 0)
+		config->nrd_pulse = ns2cyc(timing->nrd_pulse);
+
+	if (timing->ncs_write_pulse > 0)
+		config->ncs_write_pulse = ns2cyc(timing->ncs_write_pulse);
+
+	if (timing->nwe_pulse > 0)
+		config->nwe_pulse = ns2cyc(timing->nwe_pulse);
+
+	if (timing->read_cycle > 0)
+		config->read_cycle = ns2cyc(timing->read_cycle);
+
+	if (timing->write_cycle > 0)
+		config->write_cycle = ns2cyc(timing->write_cycle);
+
+	/* Extend read cycle in needed */
+	if (timing->ncs_read_recover > 0)
+		recover = ns2cyc(timing->ncs_read_recover);
+	else
+		recover = 1;
+
+	cycle = config->ncs_read_setup + config->ncs_read_pulse + recover;
+
+	if (config->read_cycle < cycle)
+		config->read_cycle = cycle;
+
+	/* Extend read cycle in needed */
+	if (timing->nrd_recover > 0)
+		recover = ns2cyc(timing->nrd_recover);
+	else
+		recover = 1;
+
+	cycle = config->nrd_setup + config->nrd_pulse + recover;
+
+	if (config->read_cycle < cycle)
+		config->read_cycle = cycle;
+
+	/* Extend write cycle in needed */
+	if (timing->ncs_write_recover > 0)
+		recover = ns2cyc(timing->ncs_write_recover);
+	else
+		recover = 1;
+
+	cycle = config->ncs_write_setup + config->ncs_write_pulse + recover;
+
+	if (config->write_cycle < cycle)
+		config->write_cycle = cycle;
+
+	/* Extend write cycle in needed */
+	if (timing->nwe_recover > 0)
+		recover = ns2cyc(timing->nwe_recover);
+	else
+		recover = 1;
+
+	cycle = config->nwe_setup + config->nwe_pulse + recover;
+
+	if (config->write_cycle < cycle)
+		config->write_cycle = cycle;
+}
+EXPORT_SYMBOL(smc_set_timing);
+
+int smc_set_configuration(int cs, const struct smc_config *config)
+{
+	unsigned long offset;
+	u32 setup, pulse, cycle, mode;
+
+	if (!hsmc)
+		return -ENODEV;
+	if (cs >= NR_CHIP_SELECTS)
+		return -EINVAL;
+
+	setup = (HSMC_BF(NWE_SETUP, config->nwe_setup)
+		 | HSMC_BF(NCS_WR_SETUP, config->ncs_write_setup)
+		 | HSMC_BF(NRD_SETUP, config->nrd_setup)
+		 | HSMC_BF(NCS_RD_SETUP, config->ncs_read_setup));
+	pulse = (HSMC_BF(NWE_PULSE, config->nwe_pulse)
+		 | HSMC_BF(NCS_WR_PULSE, config->ncs_write_pulse)
+		 | HSMC_BF(NRD_PULSE, config->nrd_pulse)
+		 | HSMC_BF(NCS_RD_PULSE, config->ncs_read_pulse));
+	cycle = (HSMC_BF(NWE_CYCLE, config->write_cycle)
+		 | HSMC_BF(NRD_CYCLE, config->read_cycle));
 
 	switch (config->bus_width) {
 	case 1:
diff --git a/arch/avr32/mach-at32ap/pio.c b/arch/avr32/mach-at32ap/pio.c
index 1eb99b814f5..d61a02da898 100644
--- a/arch/avr32/mach-at32ap/pio.c
+++ b/arch/avr32/mach-at32ap/pio.c
@@ -110,6 +110,10 @@ void __init at32_select_gpio(unsigned int pin, unsigned long flags)
 			pio_writel(pio, SODR, mask);
 		else
 			pio_writel(pio, CODR, mask);
+		if (flags & AT32_GPIOF_MULTIDRV)
+			pio_writel(pio, MDER, mask);
+		else
+			pio_writel(pio, MDDR, mask);
 		pio_writel(pio, PUDR, mask);
 		pio_writel(pio, OER, mask);
 	} else {
diff --git a/arch/avr32/mach-at32ap/pm.h b/arch/avr32/mach-at32ap/pm.h
index a1f8aced0a8..47efd0d1951 100644
--- a/arch/avr32/mach-at32ap/pm.h
+++ b/arch/avr32/mach-at32ap/pm.h
@@ -4,6 +4,14 @@
 #ifndef __ARCH_AVR32_MACH_AT32AP_PM_H__
 #define __ARCH_AVR32_MACH_AT32AP_PM_H__
 
+/*
+ * We can reduce the code size a bit by using a constant here. Since
+ * this file is only used on AVR32 AP CPUs with segmentation enabled,
+ * it's safe to not use ioremap. Generic drivers should of course
+ * never do this.
+ */
+#define AT32_PM_BASE	0xfff00000
+
 /* PM register offsets */
 #define PM_MCCTRL				0x0000
 #define PM_CKSEL				0x0004
diff --git a/arch/avr32/mm/init.c b/arch/avr32/mm/init.c
index 82cf70854b9..480760bde63 100644
--- a/arch/avr32/mm/init.c
+++ b/arch/avr32/mm/init.c
@@ -224,19 +224,9 @@ void free_initmem(void)
 
 #ifdef CONFIG_BLK_DEV_INITRD
 
-static int keep_initrd;
-
 void free_initrd_mem(unsigned long start, unsigned long end)
 {
-	if (!keep_initrd)
-		free_area(start, end, "initrd");
+	free_area(start, end, "initrd");
 }
 
-static int __init keepinitrd_setup(char *__unused)
-{
-	keep_initrd = 1;
-	return 1;
-}
-
-__setup("keepinitrd", keepinitrd_setup);
 #endif
diff --git a/include/asm-avr32/arch-at32ap/board.h b/include/asm-avr32/arch-at32ap/board.h
index 0215965dc58..7dbd603c38c 100644
--- a/include/asm-avr32/arch-at32ap/board.h
+++ b/include/asm-avr32/arch-at32ap/board.h
@@ -6,6 +6,8 @@
 
 #include <linux/types.h>
 
+#define GPIO_PIN_NONE	(-1)
+
 /* Add basic devices: system manager, interrupt controller, portmuxes, etc. */
 void at32_add_system_devices(void);
 
@@ -36,6 +38,12 @@ struct platform_device *
 at32_add_device_lcdc(unsigned int id, struct atmel_lcdfb_info *data,
 		     unsigned long fbmem_start, unsigned long fbmem_len);
 
+struct usba_platform_data {
+	int vbus_pin;
+};
+struct platform_device *
+at32_add_device_usba(unsigned int id, struct usba_platform_data *data);
+
 /* depending on what's hooked up, not all SSC pins will be used */
 #define	ATMEL_SSC_TK		0x01
 #define	ATMEL_SSC_TF		0x02
diff --git a/include/asm-avr32/arch-at32ap/portmux.h b/include/asm-avr32/arch-at32ap/portmux.h
index 9930871decd..b1abe6b4e4e 100644
--- a/include/asm-avr32/arch-at32ap/portmux.h
+++ b/include/asm-avr32/arch-at32ap/portmux.h
@@ -19,6 +19,7 @@
 #define AT32_GPIOF_OUTPUT	0x00000002	/* (OUT) Enable output driver */
 #define AT32_GPIOF_HIGH		0x00000004	/* (OUT) Set output high */
 #define AT32_GPIOF_DEGLITCH	0x00000008	/* (IN) Filter glitches */
+#define AT32_GPIOF_MULTIDRV	0x00000010	/* Enable multidriver option */
 
 void at32_select_periph(unsigned int pin, unsigned int periph,
 			unsigned long flags);
diff --git a/include/asm-avr32/arch-at32ap/smc.h b/include/asm-avr32/arch-at32ap/smc.h
index 07152b7fd9c..c98eea44a70 100644
--- a/include/asm-avr32/arch-at32ap/smc.h
+++ b/include/asm-avr32/arch-at32ap/smc.h
@@ -15,22 +15,50 @@
 /*
  * All timing parameters are in nanoseconds.
  */
-struct smc_config {
+struct smc_timing {
 	/* Delay from address valid to assertion of given strobe */
-	u16		ncs_read_setup;
-	u16		nrd_setup;
-	u16		ncs_write_setup;
-	u16		nwe_setup;
+	int ncs_read_setup;
+	int nrd_setup;
+	int ncs_write_setup;
+	int nwe_setup;
 
 	/* Pulse length of given strobe */
-	u16		ncs_read_pulse;
-	u16		nrd_pulse;
-	u16		ncs_write_pulse;
-	u16		nwe_pulse;
+	int ncs_read_pulse;
+	int nrd_pulse;
+	int ncs_write_pulse;
+	int nwe_pulse;
 
 	/* Total cycle length of given operation */
-	u16		read_cycle;
-	u16		write_cycle;
+	int read_cycle;
+	int write_cycle;
+
+	/* Minimal recovery times, will extend cycle if needed */
+	int ncs_read_recover;
+	int nrd_recover;
+	int ncs_write_recover;
+	int nwe_recover;
+};
+
+/*
+ * All timing parameters are in clock cycles.
+ */
+struct smc_config {
+
+	/* Delay from address valid to assertion of given strobe */
+	u8		ncs_read_setup;
+	u8		nrd_setup;
+	u8		ncs_write_setup;
+	u8		nwe_setup;
+
+	/* Pulse length of given strobe */
+	u8		ncs_read_pulse;
+	u8		nrd_pulse;
+	u8		ncs_write_pulse;
+	u8		nwe_pulse;
+
+	/* Total cycle length of given operation */
+	u8		read_cycle;
+	u8		write_cycle;
 
 	/* Bus width in bytes */
 	u8		bus_width;
@@ -76,6 +104,9 @@ struct smc_config {
 	unsigned int	tdf_mode:1;
 };
 
+extern void smc_set_timing(struct smc_config *config,
+			   const struct smc_timing *timing);
+
 extern int smc_set_configuration(int cs, const struct smc_config *config);
 extern struct smc_config *smc_get_configuration(int cs);
 
diff --git a/include/asm-avr32/dma-mapping.h b/include/asm-avr32/dma-mapping.h
index 21bb60bbb9a..81e342636ac 100644
--- a/include/asm-avr32/dma-mapping.h
+++ b/include/asm-avr32/dma-mapping.h
@@ -264,7 +264,11 @@ static inline void
 dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle,
 			size_t size, enum dma_data_direction direction)
 {
-	dma_cache_sync(dev, bus_to_virt(dma_handle), size, direction);
+	/*
+	 * No need to do anything since the CPU isn't supposed to
+	 * touch this memory after we flushed it at mapping- or
+	 * sync-for-device time.
+	 */
 }
 
 static inline void
@@ -309,12 +313,11 @@ static inline void
 dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 		    int nents, enum dma_data_direction direction)
 {
-	int i;
-
-	for (i = 0; i < nents; i++) {
-		dma_cache_sync(dev, page_address(sg[i].page) + sg[i].offset,
-			       sg[i].length, direction);
-	}
+	/*
+	 * No need to do anything since the CPU isn't supposed to
+	 * touch this memory after we flushed it at mapping- or
+	 * sync-for-device time.
+	 */
 }
 
 static inline void
diff --git a/include/asm-avr32/system.h b/include/asm-avr32/system.h
index a8236bacc87..dc2d527cef4 100644
--- a/include/asm-avr32/system.h
+++ b/include/asm-avr32/system.h
@@ -73,11 +73,16 @@ extern struct task_struct *__switch_to(struct task_struct *,
 
 extern void __xchg_called_with_bad_pointer(void);
 
-#ifdef __CHECKER__
-extern unsigned long __builtin_xchg(void *ptr, unsigned long x);
-#endif
+static inline unsigned long xchg_u32(u32 val, volatile u32 *m)
+{
+	u32 ret;
 
-#define xchg_u32(val, m) __builtin_xchg((void *)m, val)
+	asm volatile("xchg %[ret], %[m], %[val]"
+			: [ret] "=&r"(ret), "=m"(*m)
+			: "m"(*m), [m] "r"(m), [val] "r"(val)
+			: "memory");
+	return ret;
+}
 
 static inline unsigned long __xchg(unsigned long x,
 				       volatile void *ptr,
diff --git a/include/asm-avr32/unistd.h b/include/asm-avr32/unistd.h
index 3b4e35b55c8..de09009593f 100644
--- a/include/asm-avr32/unistd.h
+++ b/include/asm-avr32/unistd.h
@@ -303,6 +303,19 @@
 #ifdef __KERNEL__
 #define NR_syscalls		282
 
+/* Old stuff */
+#define __IGNORE_uselib
+#define __IGNORE_mmap
+
+/* NUMA stuff */
+#define __IGNORE_mbind
+#define __IGNORE_get_mempolicy
+#define __IGNORE_set_mempolicy
+#define __IGNORE_migrate_pages
+#define __IGNORE_move_pages
+
+/* SMP stuff */
+#define __IGNORE_getcpu
 
 #define __ARCH_WANT_IPC_PARSE_VERSION
 #define __ARCH_WANT_STAT64
diff --git a/scripts/checkstack.pl b/scripts/checkstack.pl
index f7844f6aa48..66315862715 100755
--- a/scripts/checkstack.pl
+++ b/scripts/checkstack.pl
@@ -12,6 +12,7 @@
 #	sh64 port by Paul Mundt
 #	Random bits by Matt Mackall <mpm@selenic.com>
 #	M68k port by Geert Uytterhoeven and Andreas Schwab
+#	AVR32 port by Haavard Skinnemoen <hskinnemoen@atmel.com>
 #
 #	Usage:
 #	objdump -d vmlinux | stackcheck.pl [arch]
@@ -37,6 +38,10 @@ my (@stack, $re, $x, $xs);
 	if ($arch eq 'arm') {
 		#c0008ffc:	e24dd064	sub	sp, sp, #100	; 0x64
 		$re = qr/.*sub.*sp, sp, #(([0-9]{2}|[3-9])[0-9]{2})/o;
+	} elsif ($arch eq 'avr32') {
+		#8000008a:       20 1d           sub sp,4
+		#80000ca8:       fa cd 05 b0     sub sp,sp,1456
+		$re = qr/^.*sub.*sp.*,([0-9]{1,8})/o;
 	} elsif ($arch =~ /^i[3456]86$/) {
 		#c0105234:       81 ec ac 05 00 00       sub    $0x5ac,%esp
 		$re = qr/^.*[as][du][db]    \$(0x$x{1,8}),\%esp$/o;