20 files changed, 376 insertions, 125 deletions

diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index 3345c6d0fd1e..92873cdee31f 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -47,7 +47,8 @@ comma = ,
 # testing for a specific architecture or later rather impossible.
 arch-$(CONFIG_CPU_32v6)         :=-D__LINUX_ARM_ARCH__=6 $(call cc-option,-march=armv6,-march=armv5t -Wa$(comma)-march=armv6)
 arch-$(CONFIG_CPU_32v6K)        :=-D__LINUX_ARM_ARCH__=6 $(call cc-option,-march=armv6k,-march=armv5t -Wa$(comma)-march=armv6k)
-arch-$(CONFIG_CPU_32v5)         :=-D__LINUX_ARM_ARCH__=5 $(call cc-option,-march=armv5te,-march=armv4)
+arch-$(CONFIG_CPU_32v5)         :=-D__LINUX_ARM_ARCH__=5 $(call cc-option,-march=armv5te,-march=armv4t)
+arch-$(CONFIG_CPU_32v4T)        :=-D__LINUX_ARM_ARCH__=4 -march=armv4t
 arch-$(CONFIG_CPU_32v4)         :=-D__LINUX_ARM_ARCH__=4 -march=armv4
 arch-$(CONFIG_CPU_32v3)         :=-D__LINUX_ARM_ARCH__=3 -march=armv3
 
diff --git a/arch/arm/common/dmabounce.c b/arch/arm/common/dmabounce.c
index 5b7c26395b44..028bdc9228fb 100644
--- a/arch/arm/common/dmabounce.c
+++ b/arch/arm/common/dmabounce.c
@@ -179,17 +179,19 @@ alloc_safe_buffer(struct dmabounce_device_info *device_info, void *ptr,
 static inline struct safe_buffer *
 find_safe_buffer(struct dmabounce_device_info *device_info, dma_addr_t safe_dma_addr)
 {
-        struct safe_buffer *b = NULL;
+        struct safe_buffer *b, *rb = NULL;
         unsigned long flags;
 
         read_lock_irqsave(&device_info->lock, flags);
 
         list_for_each_entry(b, &device_info->safe_buffers, node)
-                if (b->safe_dma_addr == safe_dma_addr)
+                if (b->safe_dma_addr == safe_dma_addr) {
+                        rb = b;
                         break;
+                }
 
         read_unlock_irqrestore(&device_info->lock, flags);
-        return b;
+        return rb;
 }
 
 static inline void
diff --git a/arch/arm/common/sa1111.c b/arch/arm/common/sa1111.c
index a331c12cead9..29818bd3248f 100644
--- a/arch/arm/common/sa1111.c
+++ b/arch/arm/common/sa1111.c
@@ -618,7 +618,7 @@ __sa1111_probe(struct device *me, struct resource *mem, int irq)
 {
         struct sa1111 *sachip;
         unsigned long id;
-        unsigned int has_devs, val;
+        unsigned int has_devs;
         int i, ret = -ENODEV;
 
         sachip = kzalloc(sizeof(struct sa1111), GFP_KERNEL);
@@ -669,6 +669,9 @@ __sa1111_probe(struct device *me, struct resource *mem, int irq)
         sa1111_wake(sachip);
 
 #ifdef CONFIG_ARCH_SA1100
+        {
+        unsigned int val;
+
         /*
          * The SDRAM configuration of the SA1110 and the SA1111 must
          * match.  This is very important to ensure that SA1111 accesses
@@ -692,6 +695,7 @@ __sa1111_probe(struct device *me, struct resource *mem, int irq)
          * Enable the SA1110 memory bus request and grant signals.
          */
         sa1110_mb_enable();
+        }
 #endif
 
         /*
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index f0c0cdb1c183..1320a0efca73 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -13,12 +13,11 @@ obj-y		:= compat.o entry-armv.o entry-common.o irq.o \
 obj-$(CONFIG_APM)               += apm.o
 obj-$(CONFIG_ISA_DMA_API)       += dma.o
 obj-$(CONFIG_ARCH_ACORN)        += ecard.o 
-obj-$(CONFIG_FOOTBRIDGE)        += isa.o
 obj-$(CONFIG_FIQ)               += fiq.o
 obj-$(CONFIG_MODULES)           += armksyms.o module.o
 obj-$(CONFIG_ARTHUR)            += arthur.o
 obj-$(CONFIG_ISA_DMA)           += dma-isa.o
-obj-$(CONFIG_PCI)               += bios32.o
+obj-$(CONFIG_PCI)               += bios32.o isa.o
 obj-$(CONFIG_SMP)               += smp.o
 obj-$(CONFIG_OABI_COMPAT)       += sys_oabi-compat.o
 
diff --git a/arch/arm/kernel/entry-armv.S b/arch/arm/kernel/entry-armv.S
index 7ea5f01dfc7b..de4e33137901 100644
--- a/arch/arm/kernel/entry-armv.S
+++ b/arch/arm/kernel/entry-armv.S
@@ -634,6 +634,14 @@ ENTRY(__switch_to)
  * purpose.
  */
 
+        .macro  usr_ret, reg
+#ifdef CONFIG_ARM_THUMB
+        bx      \reg
+#else
+        mov     pc, \reg
+#endif
+        .endm
+
         .align  5
         .globl  __kuser_helper_start
 __kuser_helper_start:
@@ -675,7 +683,7 @@ __kuser_memory_barrier:				@ 0xffff0fa0
 #if __LINUX_ARM_ARCH__ >= 6 && defined(CONFIG_SMP)
         mcr     p15, 0, r0, c7, c10, 5  @ dmb
 #endif
-        mov     pc, lr
+        usr_ret lr
 
         .align  5
 
@@ -778,7 +786,7 @@ __kuser_cmpxchg:				@ 0xffff0fc0
         mov     r0, #-1
         adds    r0, r0, #0
 #endif
-        mov     pc, lr
+        usr_ret lr
 
 #else
 
@@ -792,7 +800,7 @@ __kuser_cmpxchg:				@ 0xffff0fc0
 #ifdef CONFIG_SMP
         mcr     p15, 0, r0, c7, c10, 5  @ dmb
 #endif
-        mov     pc, lr
+        usr_ret lr
 
 #endif
 
@@ -834,16 +842,11 @@ __kuser_cmpxchg:				@ 0xffff0fc0
 __kuser_get_tls:                                @ 0xffff0fe0
 
 #if !defined(CONFIG_HAS_TLS_REG) && !defined(CONFIG_TLS_REG_EMUL)
-
         ldr     r0, [pc, #(16 - 8)]             @ TLS stored at 0xffff0ff0
-        mov     pc, lr
-
 #else
-
         mrc     p15, 0, r0, c13, c0, 3          @ read TLS register
-        mov     pc, lr
-
 #endif
+        usr_ret lr
 
         .rep    5
         .word   0                       @ pad up to __kuser_helper_version
diff --git a/arch/arm/kernel/head.S b/arch/arm/kernel/head.S
index 4fe386eea4b4..5365d4e5949e 100644
--- a/arch/arm/kernel/head.S
+++ b/arch/arm/kernel/head.S
@@ -118,7 +118,7 @@ ENTRY(secondary_startup)
         sub     r4, r4, r5                      @ mmu has been enabled
         ldr     r4, [r7, r4]                    @ get secondary_data.pgdir
         adr     lr, __enable_mmu                @ return address
-        add     pc, r10, #12                    @ initialise processor
+        add     pc, r10, #PROCINFO_INITFUNC     @ initialise processor
                                                 @ (return control reg)
 
         /*
diff --git a/arch/arm/kernel/isa.c b/arch/arm/kernel/isa.c
index 685c3e591a7e..54bbd9fe255c 100644
--- a/arch/arm/kernel/isa.c
+++ b/arch/arm/kernel/isa.c
@@ -3,21 +3,14 @@
  *
  *  Copyright (C) 1999 Phil Blundell
  *
- *  ISA shared memory and I/O port support
- */
-
-/*
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version
  * 2 of the License, or (at your option) any later version.
+ *
+ *  ISA shared memory and I/O port support, and is required to support
+ *  iopl, inb, outb and friends in userspace via glibc emulation.
  */
-
-/* 
- * Nothing about this is actually ARM specific.  One day we could move
- * it into kernel/resource.c or some place like that.  
- */
-
 #include <linux/stddef.h>
 #include <linux/types.h>
 #include <linux/fs.h>
@@ -27,21 +20,49 @@
 static unsigned int isa_membase, isa_portbase, isa_portshift;
 
 static ctl_table ctl_isa_vars[4] = {
-        {BUS_ISA_MEM_BASE, "membase", &isa_membase, 
-         sizeof(isa_membase), 0444, NULL, &proc_dointvec},
-        {BUS_ISA_PORT_BASE, "portbase", &isa_portbase, 
-         sizeof(isa_portbase), 0444, NULL, &proc_dointvec},
-        {BUS_ISA_PORT_SHIFT, "portshift", &isa_portshift, 
-         sizeof(isa_portshift), 0444, NULL, &proc_dointvec},
-        {0}
+        {
+                .ctl_name       = BUS_ISA_MEM_BASE,
+                .procname       = "membase",
+                .data           = &isa_membase, 
+                .maxlen         = sizeof(isa_membase),
+                .mode           = 0444,
+                .proc_handler   = &proc_dointvec,
+        }, {
+                .ctl_name       = BUS_ISA_PORT_BASE,
+                .procname       = "portbase",
+                .data           = &isa_portbase, 
+                .maxlen         = sizeof(isa_portbase),
+                .mode           = 0444,
+                .proc_handler   = &proc_dointvec,
+        }, {
+                .ctl_name       = BUS_ISA_PORT_SHIFT,
+                .procname       = "portshift",
+                .data           = &isa_portshift, 
+                .maxlen         = sizeof(isa_portshift),
+                .mode           = 0444,
+                .proc_handler   = &proc_dointvec,
+        }, {0}
 };
 
 static struct ctl_table_header *isa_sysctl_header;
 
-static ctl_table ctl_isa[2] = {{CTL_BUS_ISA, "isa", NULL, 0, 0555, ctl_isa_vars},
-                               {0}};
-static ctl_table ctl_bus[2] = {{CTL_BUS, "bus", NULL, 0, 0555, ctl_isa},
-                               {0}};
+static ctl_table ctl_isa[2] = {
+        {
+                .ctl_name       = CTL_BUS_ISA,
+                .procname       = "isa",
+                .mode           = 0555,
+                .child          = ctl_isa_vars,
+        }, {0}
+};
+
+static ctl_table ctl_bus[2] = {
+        {
+                .ctl_name       = CTL_BUS,
+                .procname       = "bus",
+                .mode           = 0555,
+                .child          = ctl_isa,
+        }, {0}
+};
 
 void __init
 register_isa_ports(unsigned int membase, unsigned int portbase, unsigned int portshift)
diff --git a/arch/arm/mach-footbridge/dc21285.c b/arch/arm/mach-footbridge/dc21285.c
index 607ed1f5b3f8..823e25d4547e 100644
--- a/arch/arm/mach-footbridge/dc21285.c
+++ b/arch/arm/mach-footbridge/dc21285.c
@@ -35,7 +35,6 @@
 
 extern int setup_arm_irq(int, struct irqaction *);
 extern void pcibios_report_status(u_int status_mask, int warn);
-extern void register_isa_ports(unsigned int, unsigned int, unsigned int);
 
 static unsigned long
 dc21285_base_address(struct pci_bus *bus, unsigned int devfn)
diff --git a/arch/arm/mach-integrator/pci_v3.c b/arch/arm/mach-integrator/pci_v3.c
index f9043592e299..4418f6d7572d 100644
--- a/arch/arm/mach-integrator/pci_v3.c
+++ b/arch/arm/mach-integrator/pci_v3.c
@@ -600,4 +600,6 @@ void __init pci_v3_postinit(void)
                 printk(KERN_ERR "PCI: unable to grab local bus timeout "
                        "interrupt: %d\n", ret);
 #endif
+
+        register_isa_ports(PHYS_PCI_MEM_BASE, PHYS_PCI_IO_BASE, 0);
 }
diff --git a/arch/arm/mach-pxa/corgi_ssp.c b/arch/arm/mach-pxa/corgi_ssp.c
index f9421318cb7a..ff6b4ee037f5 100644
--- a/arch/arm/mach-pxa/corgi_ssp.c
+++ b/arch/arm/mach-pxa/corgi_ssp.c
@@ -47,14 +47,15 @@ static struct corgissp_machinfo *ssp_machinfo;
  */
 unsigned long corgi_ssp_ads7846_putget(ulong data)
 {
-        unsigned long ret,flag;
+        unsigned long flag;
+        u32 ret = 0;
 
         spin_lock_irqsave(&corgi_ssp_lock, flag);
         if (ssp_machinfo->cs_ads7846 >= 0)
                 GPCR(ssp_machinfo->cs_ads7846) = GPIO_bit(ssp_machinfo->cs_ads7846);
 
         ssp_write_word(&corgi_ssp_dev,data);
-        ret = ssp_read_word(&corgi_ssp_dev);
+        ssp_read_word(&corgi_ssp_dev, &ret);
 
         if (ssp_machinfo->cs_ads7846 >= 0)
                 GPSR(ssp_machinfo->cs_ads7846) = GPIO_bit(ssp_machinfo->cs_ads7846);
@@ -88,7 +89,9 @@ void corgi_ssp_ads7846_put(ulong data)
 
 unsigned long corgi_ssp_ads7846_get(void)
 {
-        return ssp_read_word(&corgi_ssp_dev);
+        u32 ret = 0;
+        ssp_read_word(&corgi_ssp_dev, &ret);
+        return ret;
 }
 
 EXPORT_SYMBOL(corgi_ssp_ads7846_putget);
@@ -104,6 +107,7 @@ EXPORT_SYMBOL(corgi_ssp_ads7846_get);
 unsigned long corgi_ssp_dac_put(ulong data)
 {
         unsigned long flag, sscr1 = SSCR1_SPH;
+        u32 tmp;
 
         spin_lock_irqsave(&corgi_ssp_lock, flag);
 
@@ -118,7 +122,7 @@ unsigned long corgi_ssp_dac_put(ulong data)
                 GPCR(ssp_machinfo->cs_lcdcon) = GPIO_bit(ssp_machinfo->cs_lcdcon);
         ssp_write_word(&corgi_ssp_dev,data);
         /* Read null data back from device to prevent SSP overflow */
-        ssp_read_word(&corgi_ssp_dev);
+        ssp_read_word(&corgi_ssp_dev, &tmp);
         if (ssp_machinfo->cs_lcdcon >= 0)
                 GPSR(ssp_machinfo->cs_lcdcon) = GPIO_bit(ssp_machinfo->cs_lcdcon);
 
@@ -150,7 +154,7 @@ EXPORT_SYMBOL(corgi_ssp_blduty_set);
 int corgi_ssp_max1111_get(ulong data)
 {
         unsigned long flag;
-        int voltage,voltage1,voltage2;
+        long voltage = 0, voltage1 = 0, voltage2 = 0;
 
         spin_lock_irqsave(&corgi_ssp_lock, flag);
         if (ssp_machinfo->cs_max1111 >= 0)
@@ -163,15 +167,15 @@ int corgi_ssp_max1111_get(ulong data)
 
         /* TB1/RB1 */
         ssp_write_word(&corgi_ssp_dev,data);
-        ssp_read_word(&corgi_ssp_dev); /* null read */
+        ssp_read_word(&corgi_ssp_dev, (u32*)&voltage1); /* null read */
 
         /* TB12/RB2 */
         ssp_write_word(&corgi_ssp_dev,0);
-        voltage1=ssp_read_word(&corgi_ssp_dev);
+        ssp_read_word(&corgi_ssp_dev, (u32*)&voltage1);
 
         /* TB13/RB3*/
         ssp_write_word(&corgi_ssp_dev,0);
-        voltage2=ssp_read_word(&corgi_ssp_dev);
+        ssp_read_word(&corgi_ssp_dev, (u32*)&voltage2);
 
         ssp_disable(&corgi_ssp_dev);
         ssp_config(&corgi_ssp_dev, (SSCR0_National | (SSCR0_DSS & 0x0b )), 0, 0, SSCR0_SerClkDiv(ssp_machinfo->clk_ads7846));
diff --git a/arch/arm/mach-pxa/ssp.c b/arch/arm/mach-pxa/ssp.c
index 93096befd017..1fddfeaa630d 100644
--- a/arch/arm/mach-pxa/ssp.c
+++ b/arch/arm/mach-pxa/ssp.c
@@ -40,6 +40,8 @@
 
 #define PXA_SSP_PORTS   3
 
+#define TIMEOUT 100000
+
 struct ssp_info_ {
         int irq;
         u32 clock;
@@ -92,13 +94,18 @@ static irqreturn_t ssp_interrupt(int irq, void *dev_id, struct pt_regs *regs)
  * The caller is expected to perform the necessary locking.
  *
  * Returns:
- *   %-ETIMEDOUT        timeout occurred (for future)
+ *   %-ETIMEDOUT        timeout occurred
  *   0                  success
  */
 int ssp_write_word(struct ssp_dev *dev, u32 data)
 {
-        while (!(SSSR_P(dev->port) & SSSR_TNF))
+        int timeout = TIMEOUT;
+
+        while (!(SSSR_P(dev->port) & SSSR_TNF)) {
+                if (!--timeout)
+                        return -ETIMEDOUT;
                 cpu_relax();
+        }
 
         SSDR_P(dev->port) = data;
 
@@ -117,15 +124,21 @@ int ssp_write_word(struct ssp_dev *dev, u32 data)
  * The caller is expected to perform the necessary locking.
  *
  * Returns:
- *   %-ETIMEDOUT        timeout occurred (for future)
+ *   %-ETIMEDOUT        timeout occurred
  *   32-bit data        success
  */
-int ssp_read_word(struct ssp_dev *dev)
+int ssp_read_word(struct ssp_dev *dev, u32 *data)
 {
-        while (!(SSSR_P(dev->port) & SSSR_RNE))
+        int timeout = TIMEOUT;
+
+        while (!(SSSR_P(dev->port) & SSSR_RNE)) {
+                if (!--timeout)
+                        return -ETIMEDOUT;
                 cpu_relax();
+        }
 
-        return SSDR_P(dev->port);
+        *data = SSDR_P(dev->port);
+        return 0;
 }
 
 /**
@@ -136,13 +149,21 @@ int ssp_read_word(struct ssp_dev *dev)
  *
  * The caller is expected to perform the necessary locking.
  */
-void ssp_flush(struct ssp_dev *dev)
+int ssp_flush(struct ssp_dev *dev)
 {
+        int timeout = TIMEOUT * 2;
+
         do {
                 while (SSSR_P(dev->port) & SSSR_RNE) {
+                        if (!--timeout)
+                                return -ETIMEDOUT;
                         (void) SSDR_P(dev->port);
                 }
+                if (!--timeout)
+                        return -ETIMEDOUT;
         } while (SSSR_P(dev->port) & SSSR_BSY);
+
+        return 0;
 }
 
 /**
diff --git a/arch/arm/mach-s3c2410/Makefile b/arch/arm/mach-s3c2410/Makefile
index 0c7938645df6..273e05f2b8de 100644
--- a/arch/arm/mach-s3c2410/Makefile
+++ b/arch/arm/mach-s3c2410/Makefile
@@ -10,45 +10,47 @@ obj-m			:=
 obj-n                   :=
 obj-                    :=
 
+# DMA
+obj-$(CONFIG_S3C2410_DMA)       += dma.o
+
 # S3C2400 support files
-obj-$(CONFIG_CPU_S3C2400)  += s3c2400-gpio.o
+obj-$(CONFIG_CPU_S3C2400)       += s3c2400-gpio.o
 
 # S3C2410 support files
 
-obj-$(CONFIG_CPU_S3C2410)  += s3c2410.o
-obj-$(CONFIG_CPU_S3C2410)  += s3c2410-gpio.o
-obj-$(CONFIG_S3C2410_DMA)  += dma.o
+obj-$(CONFIG_CPU_S3C2410)       += s3c2410.o
+obj-$(CONFIG_CPU_S3C2410)       += s3c2410-gpio.o
 
 # Power Management support
 
-obj-$(CONFIG_PM)           += pm.o sleep.o
-obj-$(CONFIG_PM_SIMTEC)    += pm-simtec.o
+obj-$(CONFIG_PM)                += pm.o sleep.o
+obj-$(CONFIG_PM_SIMTEC)         += pm-simtec.o
 
 # S3C2412 support
-obj-$(CONFIG_CPU_S3C2412)  += s3c2412.o
-obj-$(CONFIG_CPU_S3C2412)  += s3c2412-clock.o
+obj-$(CONFIG_CPU_S3C2412)       += s3c2412.o
+obj-$(CONFIG_CPU_S3C2412)       += s3c2412-clock.o
 
 #
 # S3C244X support
 
-obj-$(CONFIG_CPU_S3C244X)  += s3c244x.o
-obj-$(CONFIG_CPU_S3C244X)  += s3c244x-irq.o
+obj-$(CONFIG_CPU_S3C244X)       += s3c244x.o
+obj-$(CONFIG_CPU_S3C244X)       += s3c244x-irq.o
 
 # Clock control
 
-obj-$(CONFIG_S3C2410_CLOCK) += s3c2410-clock.o
+obj-$(CONFIG_S3C2410_CLOCK)     += s3c2410-clock.o
 
 # S3C2440 support
 
-obj-$(CONFIG_CPU_S3C2440)  += s3c2440.o s3c2440-dsc.o
-obj-$(CONFIG_CPU_S3C2440)  += s3c2440-irq.o
-obj-$(CONFIG_CPU_S3C2440)  += s3c2440-clock.o
-obj-$(CONFIG_CPU_S3C2440)  += s3c2410-gpio.o
+obj-$(CONFIG_CPU_S3C2440)       += s3c2440.o s3c2440-dsc.o
+obj-$(CONFIG_CPU_S3C2440)       += s3c2440-irq.o
+obj-$(CONFIG_CPU_S3C2440)       += s3c2440-clock.o
+obj-$(CONFIG_CPU_S3C2440)       += s3c2410-gpio.o
 
 # S3C2442 support
 
-obj-$(CONFIG_CPU_S3C2442)  += s3c2442.o
-obj-$(CONFIG_CPU_S3C2442)  += s3c2442-clock.o
+obj-$(CONFIG_CPU_S3C2442)       += s3c2442.o
+obj-$(CONFIG_CPU_S3C2442)       += s3c2442-clock.o
 
 # bast extras
 
diff --git a/arch/arm/mach-s3c2410/dma.c b/arch/arm/mach-s3c2410/dma.c
index 094cc52745c5..25855452fe8c 100644
--- a/arch/arm/mach-s3c2410/dma.c
+++ b/arch/arm/mach-s3c2410/dma.c
@@ -112,7 +112,7 @@ dmadbg_capture(s3c2410_dma_chan_t *chan, struct s3c2410_dma_regstate *regs)
 }
 
 static void
-dmadbg_showregs(const char *fname, int line, s3c2410_dma_chan_t *chan,
+dmadbg_dumpregs(const char *fname, int line, s3c2410_dma_chan_t *chan,
                  struct s3c2410_dma_regstate *regs)
 {
         printk(KERN_DEBUG "dma%d: %s:%d: DCSRC=%08lx, DISRC=%08lx, DSTAT=%08lx DMT=%02lx, DCON=%08lx\n",
@@ -132,7 +132,16 @@ dmadbg_showchan(const char *fname, int line, s3c2410_dma_chan_t *chan)
                chan->number, fname, line, chan->load_state,
                chan->curr, chan->next, chan->end);
 
-        dmadbg_showregs(fname, line, chan, &state);
+        dmadbg_dumpregs(fname, line, chan, &state);
+}
+
+static void
+dmadbg_showregs(const char *fname, int line, s3c2410_dma_chan_t *chan)
+{
+        struct s3c2410_dma_regstate state;
+
+        dmadbg_capture(chan, &state);
+        dmadbg_dumpregs(fname, line, chan, &state);
 }
 
 #define dbg_showregs(chan) dmadbg_showregs(__FUNCTION__, __LINE__, (chan))
@@ -253,10 +262,14 @@ s3c2410_dma_loadbuffer(s3c2410_dma_chan_t *chan,
                          buf->next);
                 reload = (buf->next == NULL) ? S3C2410_DCON_NORELOAD : 0;
         } else {
-                pr_debug("load_state is %d => autoreload\n", chan->load_state);
+                //pr_debug("load_state is %d => autoreload\n", chan->load_state);
                 reload = S3C2410_DCON_AUTORELOAD;
         }
 
+        if ((buf->data & 0xf0000000) != 0x30000000) {
+                dmawarn("dmaload: buffer is %p\n", (void *)buf->data);
+        }
+
         writel(buf->data, chan->addr_reg);
 
         dma_wrreg(chan, S3C2410_DMA_DCON,
@@ -370,7 +383,7 @@ static int s3c2410_dma_start(s3c2410_dma_chan_t *chan)
         tmp |= S3C2410_DMASKTRIG_ON;
         dma_wrreg(chan, S3C2410_DMA_DMASKTRIG, tmp);
 
-        pr_debug("wrote %08lx to DMASKTRIG\n", tmp);
+        pr_debug("dma%d: %08lx to DMASKTRIG\n", chan->number, tmp);
 
 #if 0
         /* the dma buffer loads should take care of clearing the AUTO
@@ -384,7 +397,30 @@ static int s3c2410_dma_start(s3c2410_dma_chan_t *chan)
 
         dbg_showchan(chan);
 
+        /* if we've only loaded one buffer onto the channel, then chec
+         * to see if we have another, and if so, try and load it so when
+         * the first buffer is finished, the new one will be loaded onto
+         * the channel */
+
+        if (chan->next != NULL) {
+                if (chan->load_state == S3C2410_DMALOAD_1LOADED) {
+
+                        if (s3c2410_dma_waitforload(chan, __LINE__) == 0) {
+                                pr_debug("%s: buff not yet loaded, no more todo\n",
+                                         __FUNCTION__);
+                        } else {
+                                chan->load_state = S3C2410_DMALOAD_1RUNNING;
+                                s3c2410_dma_loadbuffer(chan, chan->next);
+                        }
+
+                } else if (chan->load_state == S3C2410_DMALOAD_1RUNNING) {
+                        s3c2410_dma_loadbuffer(chan, chan->next);
+                }
+        }
+
+
         local_irq_restore(flags);
+
         return 0;
 }
 
@@ -436,12 +472,11 @@ int s3c2410_dma_enqueue(unsigned int channel, void *id,
         buf = kmem_cache_alloc(dma_kmem, GFP_ATOMIC);
         if (buf == NULL) {
                 pr_debug("%s: out of memory (%ld alloc)\n",
-                         __FUNCTION__, sizeof(*buf));
+                         __FUNCTION__, (long)sizeof(*buf));
                 return -ENOMEM;
         }
 
-        pr_debug("%s: new buffer %p\n", __FUNCTION__, buf);
-
+        //pr_debug("%s: new buffer %p\n", __FUNCTION__, buf);
         //dbg_showchan(chan);
 
         buf->next  = NULL;
@@ -537,14 +572,20 @@ s3c2410_dma_lastxfer(s3c2410_dma_chan_t *chan)
         case S3C2410_DMALOAD_1LOADED:
                 if (s3c2410_dma_waitforload(chan, __LINE__) == 0) {
                                 /* flag error? */
-                        printk(KERN_ERR "dma%d: timeout waiting for load\n",
-                               chan->number);
+                        printk(KERN_ERR "dma%d: timeout waiting for load (%s)\n",
+                               chan->number, __FUNCTION__);
                         return;
                 }
                 break;
 
+        case S3C2410_DMALOAD_1LOADED_1RUNNING:
+                /* I belive in this case we do not have anything to do
+                 * until the next buffer comes along, and we turn off the
+                 * reload */
+                return;
+
         default:
-                pr_debug("dma%d: lastxfer: unhandled load_state %d with no next",
+                pr_debug("dma%d: lastxfer: unhandled load_state %d with no next\n",
                          chan->number, chan->load_state);
                 return;
 
@@ -629,7 +670,14 @@ s3c2410_dma_irq(int irq, void *devpw, struct pt_regs *regs)
         } else {
         }
 
-        if (chan->next != NULL) {
+        /* only reload if the channel is still running... our buffer done
+         * routine may have altered the state by requesting the dma channel
+         * to stop or shutdown... */
+
+        /* todo: check that when the channel is shut-down from inside this
+         * function, we cope with unsetting reload, etc */
+
+        if (chan->next != NULL && chan->state != S3C2410_DMA_IDLE) {
                 unsigned long flags;
 
                 switch (chan->load_state) {
@@ -644,8 +692,8 @@ s3c2410_dma_irq(int irq, void *devpw, struct pt_regs *regs)
                 case S3C2410_DMALOAD_1LOADED:
                         if (s3c2410_dma_waitforload(chan, __LINE__) == 0) {
                                 /* flag error? */
-                                printk(KERN_ERR "dma%d: timeout waiting for load\n",
-                                       chan->number);
+                                printk(KERN_ERR "dma%d: timeout waiting for load (%s)\n",
+                                       chan->number, __FUNCTION__);
                                 return IRQ_HANDLED;
                         }
 
@@ -678,8 +726,6 @@ s3c2410_dma_irq(int irq, void *devpw, struct pt_regs *regs)
         return IRQ_HANDLED;
 }
 
-
-
 /* s3c2410_request_dma
  *
  * get control of an dma channel
@@ -718,11 +764,17 @@ int s3c2410_dma_request(unsigned int channel, s3c2410_dma_client_t *client,
                 pr_debug("dma%d: %s : requesting irq %d\n",
                          channel, __FUNCTION__, chan->irq);
 
+                chan->irq_claimed = 1;
+                local_irq_restore(flags);
+
                 err = request_irq(chan->irq, s3c2410_dma_irq, IRQF_DISABLED,
                                   client->name, (void *)chan);
 
+                local_irq_save(flags);
+
                 if (err) {
                         chan->in_use = 0;
+                        chan->irq_claimed = 0;
                         local_irq_restore(flags);
 
                         printk(KERN_ERR "%s: cannot get IRQ %d for DMA %d\n",
@@ -730,7 +782,6 @@ int s3c2410_dma_request(unsigned int channel, s3c2410_dma_client_t *client,
                         return err;
                 }
 
-                chan->irq_claimed = 1;
                 chan->irq_enabled = 1;
         }
 
@@ -810,6 +861,7 @@ static int s3c2410_dma_dostop(s3c2410_dma_chan_t *chan)
 
         tmp = dma_rdreg(chan, S3C2410_DMA_DMASKTRIG);
         tmp |= S3C2410_DMASKTRIG_STOP;
+        //tmp &= ~S3C2410_DMASKTRIG_ON;
         dma_wrreg(chan, S3C2410_DMA_DMASKTRIG, tmp);
 
 #if 0
@@ -819,6 +871,7 @@ static int s3c2410_dma_dostop(s3c2410_dma_chan_t *chan)
         dma_wrreg(chan, S3C2410_DMA_DCON, tmp);
 #endif
 
+        /* should stop do this, or should we wait for flush? */
         chan->state      = S3C2410_DMA_IDLE;
         chan->load_state = S3C2410_DMALOAD_NONE;
 
@@ -827,6 +880,22 @@ static int s3c2410_dma_dostop(s3c2410_dma_chan_t *chan)
         return 0;
 }
 
+void s3c2410_dma_waitforstop(s3c2410_dma_chan_t *chan)
+{
+        unsigned long tmp;
+        unsigned int timeout = 0x10000;
+
+        while (timeout-- > 0) {
+                tmp = dma_rdreg(chan, S3C2410_DMA_DMASKTRIG);
+
+                if (!(tmp & S3C2410_DMASKTRIG_ON))
+                        return;
+        }
+
+        pr_debug("dma%d: failed to stop?\n", chan->number);
+}
+
+
 /* s3c2410_dma_flush
  *
  * stop the channel, and remove all current and pending transfers
@@ -837,7 +906,9 @@ static int s3c2410_dma_flush(s3c2410_dma_chan_t *chan)
         s3c2410_dma_buf_t *buf, *next;
         unsigned long flags;
 
-        pr_debug("%s:\n", __FUNCTION__);
+        pr_debug("%s: chan %p (%d)\n", __FUNCTION__, chan, chan->number);
+
+        dbg_showchan(chan);
 
         local_irq_save(flags);
 
@@ -864,11 +935,64 @@ static int s3c2410_dma_flush(s3c2410_dma_chan_t *chan)
                 }
         }
 
+        dbg_showregs(chan);
+
+        s3c2410_dma_waitforstop(chan);
+
+#if 0
+        /* should also clear interrupts, according to WinCE BSP */
+        {
+                unsigned long tmp;
+
+                tmp = dma_rdreg(chan, S3C2410_DMA_DCON);
+                tmp |= S3C2410_DCON_NORELOAD;
+                dma_wrreg(chan, S3C2410_DMA_DCON, tmp);
+        }
+#endif
+
+        dbg_showregs(chan);
+
         local_irq_restore(flags);
 
         return 0;
 }
 
+int
+s3c2410_dma_started(s3c2410_dma_chan_t *chan)
+{
+        unsigned long flags;
+
+        local_irq_save(flags);
+
+        dbg_showchan(chan);
+
+        /* if we've only loaded one buffer onto the channel, then chec
+         * to see if we have another, and if so, try and load it so when
+         * the first buffer is finished, the new one will be loaded onto
+         * the channel */
+
+        if (chan->next != NULL) {
+                if (chan->load_state == S3C2410_DMALOAD_1LOADED) {
+
+                        if (s3c2410_dma_waitforload(chan, __LINE__) == 0) {
+                                pr_debug("%s: buff not yet loaded, no more todo\n",
+                                         __FUNCTION__);
+                        } else {
+                                chan->load_state = S3C2410_DMALOAD_1RUNNING;
+                                s3c2410_dma_loadbuffer(chan, chan->next);
+                        }
+
+                } else if (chan->load_state == S3C2410_DMALOAD_1RUNNING) {
+                        s3c2410_dma_loadbuffer(chan, chan->next);
+                }
+        }
+
+
+        local_irq_restore(flags);
+
+        return 0;
+
+}
 
 int
 s3c2410_dma_ctrl(dmach_t channel, s3c2410_chan_op_t op)
@@ -885,14 +1009,15 @@ s3c2410_dma_ctrl(dmach_t channel, s3c2410_chan_op_t op)
                 return s3c2410_dma_dostop(chan);
 
         case S3C2410_DMAOP_PAUSE:
-                return -ENOENT;
-
         case S3C2410_DMAOP_RESUME:
                 return -ENOENT;
 
         case S3C2410_DMAOP_FLUSH:
                 return s3c2410_dma_flush(chan);
 
+        case S3C2410_DMAOP_STARTED:
+                return s3c2410_dma_started(chan);
+
         case S3C2410_DMAOP_TIMEOUT:
                 return 0;
 
diff --git a/arch/arm/mach-sa1100/ssp.c b/arch/arm/mach-sa1100/ssp.c
index 1604dadf27fc..5eba5fbbb561 100644
--- a/arch/arm/mach-sa1100/ssp.c
+++ b/arch/arm/mach-sa1100/ssp.c
@@ -23,6 +23,8 @@
 #include <asm/hardware.h>
 #include <asm/hardware/ssp.h>
 
+#define TIMEOUT 100000
+
 static irqreturn_t ssp_interrupt(int irq, void *dev_id, struct pt_regs *regs)
 {
         unsigned int status = Ser4SSSR;
@@ -47,18 +49,27 @@ static irqreturn_t ssp_interrupt(int irq, void *dev_id, struct pt_regs *regs)
  * The caller is expected to perform the necessary locking.
  *
  * Returns:
- *   %-ETIMEDOUT        timeout occurred (for future)
+ *   %-ETIMEDOUT        timeout occurred
  *   0                  success
  */
 int ssp_write_word(u16 data)
 {
-        while (!(Ser4SSSR & SSSR_TNF))
+        int timeout = TIMEOUT;
+
+        while (!(Ser4SSSR & SSSR_TNF)) {
+                if (!--timeout)
+                        return -ETIMEDOUT;
                 cpu_relax();
+        }
 
         Ser4SSDR = data;
 
-        while (!(Ser4SSSR & SSSR_BSY))
+        timeout = TIMEOUT;
+        while (!(Ser4SSSR & SSSR_BSY)) {
+                if (!--timeout)
+                        return -ETIMEDOUT;
                 cpu_relax();
+        }
 
         return 0;
 }
@@ -75,15 +86,22 @@ int ssp_write_word(u16 data)
  * The caller is expected to perform the necessary locking.
  *
  * Returns:
- *   %-ETIMEDOUT        timeout occurred (for future)
+ *   %-ETIMEDOUT        timeout occurred
  *   16-bit data        success
  */
-int ssp_read_word(void)
+int ssp_read_word(u16 *data)
 {
-        while (!(Ser4SSSR & SSSR_RNE))
+        int timeout = TIMEOUT;
+
+        while (!(Ser4SSSR & SSSR_RNE)) {
+                if (!--timeout)
+                        return -ETIMEDOUT;
                 cpu_relax();
+        }
+
+        *data = (u16)Ser4SSDR;
 
-        return Ser4SSDR;
+        return 0;
 }
 
 /**
@@ -93,14 +111,26 @@ int ssp_read_word(void)
  * is empty.
  *
  * The caller is expected to perform the necessary locking.
+ *
+ * Returns:
+ *   %-ETIMEDOUT        timeout occurred
+ *   0                  success
  */
-void ssp_flush(void)
+int ssp_flush(void)
 {
+        int timeout = TIMEOUT * 2;
+
         do {
                 while (Ser4SSSR & SSSR_RNE) {
+                        if (!--timeout)
+                                return -ETIMEDOUT;
                         (void) Ser4SSDR;
                 }
+                if (!--timeout)
+                        return -ETIMEDOUT;
         } while (Ser4SSSR & SSSR_BSY);
+
+        return 0;
 }
 
 /**
diff --git a/arch/arm/mach-versatile/core.c b/arch/arm/mach-versatile/core.c
index c4e3f8c68479..f2bbef07b1e4 100644
--- a/arch/arm/mach-versatile/core.c
+++ b/arch/arm/mach-versatile/core.c
@@ -285,7 +285,7 @@ static struct flash_platform_data versatile_flash_data = {
 
 static struct resource versatile_flash_resource = {
         .start                  = VERSATILE_FLASH_BASE,
-        .end                    = VERSATILE_FLASH_BASE + VERSATILE_FLASH_SIZE,
+        .end                    = VERSATILE_FLASH_BASE + VERSATILE_FLASH_SIZE - 1,
         .flags                  = IORESOURCE_MEM,
 };
 
diff --git a/arch/arm/mm/Kconfig b/arch/arm/mm/Kconfig
index 5f80f184cd32..b4f220dd5eb8 100644
--- a/arch/arm/mm/Kconfig
+++ b/arch/arm/mm/Kconfig
@@ -46,7 +46,7 @@ config CPU_ARM710
 config CPU_ARM720T
         bool "Support ARM720T processor" if !ARCH_CLPS711X && !ARCH_L7200 && !ARCH_CDB89712 && ARCH_INTEGRATOR
         default y if ARCH_CLPS711X || ARCH_L7200 || ARCH_CDB89712 || ARCH_H720X
-        select CPU_32v4
+        select CPU_32v4T
         select CPU_ABRT_LV4T
         select CPU_CACHE_V4
         select CPU_CACHE_VIVT
@@ -64,7 +64,7 @@ config CPU_ARM920T
         bool "Support ARM920T processor"
         depends on ARCH_EP93XX || ARCH_INTEGRATOR || CPU_S3C2410 || CPU_S3C2440 || CPU_S3C2442 || ARCH_IMX || ARCH_AAEC2000 || ARCH_AT91RM9200
         default y if CPU_S3C2410 || CPU_S3C2440 || CPU_S3C2442 || ARCH_AT91RM9200
-        select CPU_32v4
+        select CPU_32v4T
         select CPU_ABRT_EV4T
         select CPU_CACHE_V4WT
         select CPU_CACHE_VIVT
@@ -85,7 +85,7 @@ config CPU_ARM922T
         bool "Support ARM922T processor" if ARCH_INTEGRATOR
         depends on ARCH_LH7A40X || ARCH_INTEGRATOR
         default y if ARCH_LH7A40X
-        select CPU_32v4
+        select CPU_32v4T
         select CPU_ABRT_EV4T
         select CPU_CACHE_V4WT
         select CPU_CACHE_VIVT
@@ -104,7 +104,7 @@ config CPU_ARM925T
         bool "Support ARM925T processor" if ARCH_OMAP1
         depends on ARCH_OMAP15XX
         default y if ARCH_OMAP15XX
-        select CPU_32v4
+        select CPU_32v4T
         select CPU_ABRT_EV4T
         select CPU_CACHE_V4WT
         select CPU_CACHE_VIVT
@@ -285,6 +285,11 @@ config CPU_32v4
         select TLS_REG_EMUL if SMP || !MMU
         select NEEDS_SYSCALL_FOR_CMPXCHG if SMP
 
+config CPU_32v4T
+        bool
+        select TLS_REG_EMUL if SMP || !MMU
+        select NEEDS_SYSCALL_FOR_CMPXCHG if SMP
+
 config CPU_32v5
         bool
         select TLS_REG_EMUL if SMP || !MMU
diff --git a/arch/arm/vfp/vfp.h b/arch/arm/vfp/vfp.h
index 4b97950984e9..5fbdf81a8aaf 100644
--- a/arch/arm/vfp/vfp.h
+++ b/arch/arm/vfp/vfp.h
@@ -353,3 +353,11 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand);
  * A special flag to tell the normalisation code not to normalise.
  */
 #define VFP_NAN_FLAG    0x100
+
+/*
+ * A bit pattern used to indicate the initial (unset) value of the
+ * exception mask, in case nothing handles an instruction.  This
+ * doesn't include the NAN flag, which get masked out before
+ * we check for an error.
+ */
+#define VFP_EXCEPTION_ERROR     ((u32)-1 & ~VFP_NAN_FLAG)
diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c
index 009038c8113e..04bd3425b29b 100644
--- a/arch/arm/vfp/vfpdouble.c
+++ b/arch/arm/vfp/vfpdouble.c
@@ -465,7 +465,7 @@ static u32 vfp_double_fcvts(int sd, int unused, int dm, u32 fpscr)
          */
         if (tm & (VFP_INFINITY|VFP_NAN)) {
                 vsd.exponent = 255;
-                if (tm & VFP_NAN)
+                if (tm == VFP_QNAN)
                         vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN;
                 goto pack_nan;
         } else if (tm & VFP_ZERO)
@@ -1127,7 +1127,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
 {
         u32 op = inst & FOP_MASK;
         u32 exceptions = 0;
-        unsigned int dd = vfp_get_dd(inst);
+        unsigned int dest;
         unsigned int dn = vfp_get_dn(inst);
         unsigned int dm = vfp_get_dm(inst);
         unsigned int vecitr, veclen, vecstride;
@@ -1137,10 +1137,20 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
         vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2;
 
         /*
+         * fcvtds takes an sN register number as destination, not dN.
+         * It also always operates on scalars.
+         */
+        if ((inst & FEXT_MASK) == FEXT_FCVT) {
+                veclen = 0;
+                dest = vfp_get_sd(inst);
+        } else
+                dest = vfp_get_dd(inst);
+
+        /*
          * If destination bank is zero, vector length is always '1'.
          * ARM DDI0100F C5.1.3, C5.3.2.
          */
-        if (FREG_BANK(dd) == 0)
+        if (FREG_BANK(dest) == 0)
                 veclen = 0;
 
         pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
@@ -1153,16 +1163,20 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
         for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
                 u32 except;
 
-                if (op == FOP_EXT)
+                if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT)
+                        pr_debug("VFP: itr%d (s%u) = op[%u] (d%u)\n",
+                                 vecitr >> FPSCR_LENGTH_BIT,
+                                 dest, dn, dm);
+                else if (op == FOP_EXT)
                         pr_debug("VFP: itr%d (d%u) = op[%u] (d%u)\n",
                                  vecitr >> FPSCR_LENGTH_BIT,
-                                 dd, dn, dm);
+                                 dest, dn, dm);
                 else
                         pr_debug("VFP: itr%d (d%u) = (d%u) op[%u] (d%u)\n",
                                  vecitr >> FPSCR_LENGTH_BIT,
-                                 dd, dn, FOP_TO_IDX(op), dm);
+                                 dest, dn, FOP_TO_IDX(op), dm);
 
-                except = fop(dd, dn, dm, fpscr);
+                except = fop(dest, dn, dm, fpscr);
                 pr_debug("VFP: itr%d: exceptions=%08x\n",
                          vecitr >> FPSCR_LENGTH_BIT, except);
 
@@ -1180,7 +1194,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
                  * we encounter an exception.  We continue.
                  */
 
-                dd = FREG_BANK(dd) + ((FREG_IDX(dd) + vecstride) & 6);
+                dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6);
                 dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6);
                 if (FREG_BANK(dm) != 0)
                         dm = FREG_BANK(dm) + ((FREG_IDX(dm) + vecstride) & 6);
diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
index 9d265d5e748c..4178f6cc3d37 100644
--- a/arch/arm/vfp/vfpmodule.c
+++ b/arch/arm/vfp/vfpmodule.c
@@ -131,7 +131,7 @@ static void vfp_raise_exceptions(u32 exceptions, u32 inst, u32 fpscr, struct pt_
 
         pr_debug("VFP: raising exceptions %08x\n", exceptions);
 
-        if (exceptions == (u32)-1) {
+        if (exceptions == VFP_EXCEPTION_ERROR) {
                 vfp_panic("unhandled bounce");
                 vfp_raise_sigfpe(0, regs);
                 return;
@@ -170,7 +170,7 @@ static void vfp_raise_exceptions(u32 exceptions, u32 inst, u32 fpscr, struct pt_
  */
 static u32 vfp_emulate_instruction(u32 inst, u32 fpscr, struct pt_regs *regs)
 {
-        u32 exceptions = (u32)-1;
+        u32 exceptions = VFP_EXCEPTION_ERROR;
 
         pr_debug("VFP: emulate: INST=0x%08x SCR=0x%08x\n", inst, fpscr);
 
diff --git a/arch/arm/vfp/vfpsingle.c b/arch/arm/vfp/vfpsingle.c
index dae2c2f46052..78d7cac5f36b 100644
--- a/arch/arm/vfp/vfpsingle.c
+++ b/arch/arm/vfp/vfpsingle.c
@@ -506,7 +506,7 @@ static u32 vfp_single_fcvtd(int dd, int unused, s32 m, u32 fpscr)
          */
         if (tm & (VFP_INFINITY|VFP_NAN)) {
                 vdd.exponent = 2047;
-                if (tm & VFP_NAN)
+                if (tm == VFP_QNAN)
                         vdd.significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
                 goto pack_nan;
         } else if (tm & VFP_ZERO)
@@ -514,10 +514,6 @@ static u32 vfp_single_fcvtd(int dd, int unused, s32 m, u32 fpscr)
         else
                 vdd.exponent = vsm.exponent + (1023 - 127);
 
-        /*
-         * Technically, if bit 0 of dd is set, this is an invalid
-         * instruction.  However, we ignore this for efficiency.
-         */
         return vfp_double_normaliseround(dd, &vdd, fpscr, exceptions, "fcvtd");
 
  pack_nan:
@@ -1174,7 +1170,7 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
 {
         u32 op = inst & FOP_MASK;
         u32 exceptions = 0;
-        unsigned int sd = vfp_get_sd(inst);
+        unsigned int dest;
         unsigned int sn = vfp_get_sn(inst);
         unsigned int sm = vfp_get_sm(inst);
         unsigned int vecitr, veclen, vecstride;
@@ -1184,10 +1180,22 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
         vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
 
         /*
+         * fcvtsd takes a dN register number as destination, not sN.
+         * Technically, if bit 0 of dd is set, this is an invalid
+         * instruction.  However, we ignore this for efficiency.
+         * It also only operates on scalars.
+         */
+        if ((inst & FEXT_MASK) == FEXT_FCVT) {
+                veclen = 0;
+                dest = vfp_get_dd(inst);
+        } else
+                dest = vfp_get_sd(inst);
+
+        /*
          * If destination bank is zero, vector length is always '1'.
          * ARM DDI0100F C5.1.3, C5.3.2.
          */
-        if (FREG_BANK(sd) == 0)
+        if (FREG_BANK(dest) == 0)
                 veclen = 0;
 
         pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
@@ -1201,15 +1209,18 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
                 s32 m = vfp_get_float(sm);
                 u32 except;
 
-                if (op == FOP_EXT)
+                if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT)
+                        pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n",
+                                 vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
+                else if (op == FOP_EXT)
                         pr_debug("VFP: itr%d (s%u) = op[%u] (s%u=%08x)\n",
-                                 vecitr >> FPSCR_LENGTH_BIT, sd, sn, sm, m);
+                                 vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
                 else
                         pr_debug("VFP: itr%d (s%u) = (s%u) op[%u] (s%u=%08x)\n",
-                                 vecitr >> FPSCR_LENGTH_BIT, sd, sn,
+                                 vecitr >> FPSCR_LENGTH_BIT, dest, sn,
                                  FOP_TO_IDX(op), sm, m);
 
-                except = fop(sd, sn, m, fpscr);
+                except = fop(dest, sn, m, fpscr);
                 pr_debug("VFP: itr%d: exceptions=%08x\n",
                          vecitr >> FPSCR_LENGTH_BIT, except);
 
@@ -1227,7 +1238,7 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
                  * we encounter an exception.  We continue.
                  */
 
-                sd = FREG_BANK(sd) + ((FREG_IDX(sd) + vecstride) & 7);
+                dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7);
                 sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
                 if (FREG_BANK(sm) != 0)
                         sm = FREG_BANK(sm) + ((FREG_IDX(sm) + vecstride) & 7);