Merge with Linux 2.4.2.

54 files changed, 1044 insertions, 517 deletions

diff --git a/arch/arm/Makefile b/arch/arm/Makefile
index c19ccddec..0320dba31 100644
--- a/arch/arm/Makefile
+++ b/arch/arm/Makefile
@@ -12,25 +12,18 @@
 #
 # Copyright (C) 1995-2000 by Russell King
 
-OBJCOPY		:= $(CROSS_COMPILE)objcopy -O binary -R .note -R .comment -S
-CPP		:= $(CC) -E
-LINKFLAGS	:= -p -X -T arch/arm/vmlinux.lds
-ARCHCC		:= $(word 1,$(CC))
-
-AFLAGS		+= -mno-fpu
-CFLAGS_PIPE	:= -pipe
-CFLAGS		:= $(CFLAGS) $(CFLAGS_PIPE) -msoft-float
+LINKFLAGS	:=-p -X -T arch/arm/vmlinux.lds
+GZFLAGS		:=-9
+CFLAGS		+=-fno-common -pipe
 
 ifdef CONFIG_FRAME_POINTER
-CFLAGS		:= $(CFLAGS:-fomit-frame-pointer=)
+CFLAGS		:=$(CFLAGS:-fomit-frame-pointer=)
 endif
 
 ifdef CONFIG_DEBUG_INFO
-CFLAGS		+= -g
+CFLAGS		+=-g
 endif
 
-GZFLAGS		= -9
-
 # Ensure this is ld "2.9.4" or later
 NEW_LINKER	:= $(shell $(LD) --gc-sections --version >/dev/null 2>&1; echo $$?)
 
@@ -40,61 +33,42 @@ dummy:;		@echo '*** ${VERSION}.${PATCHLEVEL} kernels no longer build correctly w
 		@false
 endif
 
-# GCC 2.7 uses different options to later compilers; sort out which we have
-NEW_GCC		:= $(shell $(CC) --version 2>&1 | grep '^2\.7' > /dev/null; echo $$?)
-
-#
-# select flags depending on the compiler
+# Select CPU dependent flags.  Note that order of declaration is important;
+# the options further down the list override previous items.
 #
-ifneq ($(NEW_GCC),0)
-CFLAGS			+= -mshort-load-bytes
-CFLAGS_PROC_CPU_26	:= -mcpu=arm3 -mapcs-26 -Os
-CFLAGS_PROC_CPU_32v3	:= -march=armv3
-CFLAGS_PROC_CPU_32v4	:= -march=armv4
-else
-CFLAGS			+= -DNO_TEXT_SECTIONS
-CFLAGS_PROC_CPU_26	:= -m3
-CFLAGS_PROC_CPU_32v3	:= -m6
-CFLAGS_PROC_CPU_32v4	:= -m6
-endif
+apcs-$(CONFIG_CPU_26)		:=-mapcs-26 -mcpu=arm3 -Os
+apcs-$(CONFIG_CPU_32)		:=-mapcs-32
+
+arch-$(CONFIG_CPU_32v3)		:=-march=armv3
+arch-$(CONFIG_CPU_32v4)		:=-march=armv4
+arch-$(CONFIG_CPU_32v5)		:=-march=armv5
+
+proc-$(CONFIG_CPU_32v3)		:=-marmv3m
+proc-$(CONFIG_CPU_32v4)		:=-marmv4
+proc-$(CONFIG_CPU_32v5)		:=-marmv5
+
+tune-$(CONFIG_CPU_ARM610)	:=-mtune=arm610
+tune-$(CONFIG_CPU_ARM710)	:=-mtune=arm710
+tune-$(CONFIG_CPU_ARM720T)	:=-mtune=arm7tdmi
+tune-$(CONFIG_CPU_ARM920T)	:=-mtune=arm9tdmi
+tune-$(CONFIG_CPU_SA110)	:=-mtune=strongarm110
+tune-$(CONFIG_CPU_SA1100)	:=-mtune=strongarm1100
+
+CFLAGS		+= $(apcs-y) $(arch-y) $(tune-y) -mshort-load-bytes -msoft-float
+AFLAGS		+= $(apcs-y) $(proc-y) -mno-fpu
+
+LIBGCC		:= $(shell $(CC) $(CFLAGS) --print-libgcc-file-name)
 
-#
-# Select CPU dependent flags
-#
 ifeq ($(CONFIG_CPU_26),y)
- PROCESSOR	 = armo
- TEXTADDR	 = 0x02080000
- CFLAGS		+= $(CFLAGS_PROC_CPU_26)
- AFLAGS		+= -mapcs-26
+PROCESSOR	 = armo
+TEXTADDR	 = 0x02080000
 endif
 
 ifeq ($(CONFIG_CPU_32),y)
- PROCESSOR	 = armv
- TEXTADDR	 = 0xC0008000
- ifeq ($(CONFIG_CPU_32v4),y)
-  CFLAGS	+= $(CFLAGS_PROC_CPU_32v4)
-  AFLAGS	+= -mapcs-32 -marmv4
- else
-  CFLAGS	+= $(CFLAGS_PROC_CPU_32v3)
-  AFLAGS	+= -mapcs-32 -marmv3m
- endif
-
- opt-$(CONFIG_CPU_ARM6)		:= -mtune=arm6
- opt-$(CONFIG_CPU_ARM7)		:= -mtune=arm7
- opt-$(CONFIG_CPU_ARM720)	:= -mtune=arm7tdmi
- opt-$(CONFIG_CPU_ARM920)	:= -mtune=arm9tdmi
- opt-$(CONFIG_CPU_SA110)	:= -mtune=strongarm110
- opt-$(CONFIG_CPU_SA1100)	:= -mtune=strongarm110
-
- ifneq ($(NEW_GCC),0)
- CFLAGS	+= $(opt-y)
- endif
+PROCESSOR	 = armv
+TEXTADDR	 = 0xC0008000
 endif
 
-LIBGCC		:= $(shell $(CC) $(CFLAGS) --print-libgcc-file-name)
-
-export	LIBGCC MACHINE PROCESSOR TEXTADDR GZFLAGS
-
 ifeq ($(CONFIG_ARCH_ARCA5K),y)
 MACHINE		 = arc
 endif
@@ -132,6 +106,10 @@ MACHINE		 = shark
 endif
 
 ifeq ($(CONFIG_ARCH_SA1100),y)
+ifeq ($(CONFIG_SA1111),y)
+# SA1111 DMA bug: we don't want the kernel to live in precious DMA-able memory
+TEXTADDR	 = 0xc0208000
+endif
 MACHINE		 = sa1100
 endif
 
@@ -143,6 +121,13 @@ ifeq ($(CONFIG_ARCH_INTEGRATOR),y)
 MACHINE		 = integrator
 endif
 
+ifeq ($(CONFIG_ARCH_CLPS711X),y)
+TEXTADDR	 = 0xc0018000
+MACHINE		 = clps711x
+endif
+
+export	LIBGCC MACHINE PROCESSOR TEXTADDR GZFLAGS
+
 # Only set INCDIR if its not already defined above
 # Grr, ?= doesn't work as all the other assignment operators do.  Make bug?
 ifeq ($(origin INCDIR), undefined)
@@ -158,8 +143,7 @@ endif
 
 HEAD		:= arch/arm/kernel/head-$(PROCESSOR).o \
 		   arch/arm/kernel/init_task.o
-SUBDIRS		+= arch/arm/kernel arch/arm/mm arch/arm/lib \
-		   arch/arm/nwfpe
+SUBDIRS		+= arch/arm/kernel arch/arm/mm arch/arm/lib arch/arm/nwfpe
 CORE_FILES	:= arch/arm/kernel/kernel.o arch/arm/mm/mm.o $(CORE_FILES)
 LIBS		:= arch/arm/lib/lib.o arch/arm/lib/lib.a $(LIBS) $(LIBGCC)
 
@@ -167,14 +151,6 @@ ifeq ($(CONFIG_NWFPE),y)
 LIBS		:= arch/arm/nwfpe/math-emu.o $(LIBS)
 endif
 
-ifeq ($(CONFIG_ARCH_ACORN),y)
-SUBDIRS		+= drivers/acorn
-DRIVERS		+= drivers/acorn/block/acorn-block.a
-DRIVERS		+= drivers/acorn/char/acorn-char.o
-DRIVERS		+= drivers/acorn/net/acorn-net.o
-DRIVERS		+= drivers/acorn/scsi/acorn-scsi.a
-endif
-
 ifeq ($(CONFIG_ARCH_CLPS7500),y)
 SUBDIRS		+= drivers/acorn/char
 DRIVERS		+= drivers/acorn/char/acorn-char.o
@@ -230,6 +206,7 @@ zImg:;	@$(MAKEBOOT) zImage
 Img:;	@$(MAKEBOOT) Image
 i:;	@$(MAKEBOOT) install
 zi:;	@$(MAKEBOOT) zinstall
+bp:;	@$(MAKEBOOT) bootpImage
 
 #
 # Configuration targets.  Use these to select a
@@ -237,9 +214,10 @@ zi:;	@$(MAKEBOOT) zinstall
 CFGS=	a5k_config		ebsa110_config		\
 	footbridge_config	rpc_config		\
 	brutus_config		victor_config		\
-	empeg_config		thinclient_config	\
+	empeg_config		graphicsclient_config	\
 	assabet_config		lart_config		\
-	cerf_config		lusl7200_config 
+	cerf_config		lusl7200_config 	\
+	sherman_config		pangolin_config
 
 $(CFGS):
 	@( \
diff --git a/arch/arm/boot/Makefile b/arch/arm/boot/Makefile
index f74033b36..7b13b1bcb 100644
--- a/arch/arm/boot/Makefile
+++ b/arch/arm/boot/Makefile
@@ -83,15 +83,18 @@ ZTEXTADDR	 = 0xc0008000
 ZRELADDR	 = 0xc0008000
 ifeq ($(CONFIG_SA1100_VICTOR),y)
   ZTEXTADDR	 = 0x00002000
-  ZBSSADDR	 = 0xc0100000
+  ZBSSADDR	 = 0xc0200000
 endif
 ifeq ($(CONFIG_SA1100_SHERMAN),y)
   ZTEXTADDR      = 0x00050000   
-  ZBSSADDR       = 0xc0100000
+  ZBSSADDR       = 0xc0200000
 endif
 ifeq ($(CONFIG_SA1100_GRAPHICSCLIENT),y)
   ZTEXTADDR	 = 0xC0200000
 endif
+ifeq ($(CONFIG_SA1111),y)
+  ZRELADDR	 = 0xc0208000
+endif
 endif
 
 #
@@ -105,15 +108,15 @@ endif
 export	SYSTEM ZTEXTADDR ZBSSADDR ZRELADDR INITRD_PHYS INITRD_VIRT PARAMS_PHYS
 
 Image:	$(CONFIGURE) $(SYSTEM)
-	$(OBJCOPY) $(SYSTEM) $@
+	$(OBJCOPY) -O binary -R .note -R .comment -S $(SYSTEM) $@
 
 bzImage: zImage
 
 zImage:	$(CONFIGURE) compressed/vmlinux
-	$(OBJCOPY) compressed/vmlinux $@
+	$(OBJCOPY) -O binary -R .note -R .comment -S compressed/vmlinux $@
 
 bootpImage: bootp/bootp
-	$(OBJCOPY) bootp/bootp $@
+	$(OBJCOPY) -O binary -R .note -R .comment -S bootp/bootp $@
 
 compressed/vmlinux: $(TOPDIR)/vmlinux dep
 	@$(MAKE) -C compressed vmlinux
diff --git a/arch/arm/boot/bootp/Makefile b/arch/arm/boot/bootp/Makefile
index 918e3eb9e..adeeeddc5 100644
--- a/arch/arm/boot/bootp/Makefile
+++ b/arch/arm/boot/bootp/Makefile
@@ -7,7 +7,8 @@ INITRD		=$(ZSYSTEM)
 ZLDFLAGS	=-p -X -T bootp.lds \
 		 --defsym initrd_addr=$(INITRD_PHYS) \
 		 --defsym initrd_virt=$(INITRD_VIRT) \
-		 --defsym params=$(PARAMS_PHYS)
+		 --defsym params=$(PARAMS_PHYS) \
+		 --defsym kernel_addr=$(ZTEXTADDR)
 
 all:		bootp
 
diff --git a/arch/arm/boot/bootp/init.S b/arch/arm/boot/bootp/init.S
index 2b498b7bb..9963fbc2b 100644
--- a/arch/arm/boot/bootp/init.S
+++ b/arch/arm/boot/bootp/init.S
@@ -12,33 +12,59 @@
  */
 		.section .start,#alloc,#execinstr
 		.type	_entry, #function
-_entry:
-kernel_addr:	adr	r10, initdata
-		ldmia	r10, {r11, r12}
+_entry:		adr	r10, initdata
+		ldr	r11, initdata
 		sub	r11, r10, r11		@ work out exec offset
-		add	r12, r12, r11		@ correct "splitify"
-		mov	pc, r12			@ jump to splitify
+		b	splitify
 		.size	_entry,. - _entry
 
 		.type	initdata, #object
 initdata:	.word	initdata		@ compiled address of this
-		.word	splitify
 		.size	initdata,. - initdata
 
 		.text
 splitify:	adr	r13, data
-		ldmia	r13!, {r4-r6}		@ move the kernel
+		ldmia	r13!, {r4-r6}		@ move the initrd
 		add	r4, r4, r11		@ correction
-		mov	r12, r5
 		bl	move
 
-		ldmia	r13!, {r4-r6}		@ then the initrd
+		ldmia	r13!, {r4-r6}		@ then the kernel
+		mov	r12, r5
 		add	r4, r4, r11		@ correction
 		bl	move
 
-		ldmib	r13, {r5,r6,r7}		@ get size and addr of initrd
-		add	r7, r7, #16*4		@ offset of initrd_start in param_struct
-		stmia	r7, {r5,r6}		@ save in param_struct
+/*
+ * Setup the initrd parameters to pass to the kernel.  This can either be
+ * passed in via a param_struct or a tag list.  We spot the param_struct
+ * method by looking at the first word; this should either indicate a page
+ * size of 4K, 16K or 32K.
+ */
+		ldmia	r13, {r5-r8}		@ get size and addr of initrd
+						@ r5 = ATAG_INITRD
+						@ r6 = initrd start
+						@ r7 = initrd end
+						@ r8 = param_struct address
+		ldr	r9, [r8, #0]		@ no param struct?
+		teq	r9, #0x1000		@ 4K?
+		teqne	r9, #0x4000		@ 16K?
+		teqne	r9, #0x8000		@ 32K?
+		beq	no_taglist
+
+/*
+ * find the end of the tag list, and then add an INITRD tag on the end.
+ */
+taglist:	ldr	r9, [r8, #0]		@ tag length
+		teq	r9, #0			@ last tag?
+		addne	r8, r8, r9
+		bne	taglist
+
+		mov	r4, #16			@ length of initrd tag
+		mov	r9, #0			@ end of tag list terminator
+		stmia	r8, {r4, r5, r6, r7, r9}
+		mov	pc, r12			@ call kernel
+
+no_taglist:	add	r8, r8, #16*4
+		stmia	r8, {r6,r7}		@ save in param_struct
 		mov	pc, r12			@ call kernel
 
 move:		ldmia	r4!, {r7 - r10}		@ move 32-bytes at a time
@@ -49,17 +75,18 @@ move:		ldmia	r4!, {r7 - r10}		@ move 32-bytes at a time
 		bcs	move
 		mov	pc, lr
 
-data:		.word	kernel_start
-		.word	kernel_addr
-		.word	kernel_len
-
-		.word	initrd_start
+data:		.word	initrd_start
 		.word	initrd_addr
 		.word	initrd_len
 
-		.word	initrd_virt
-		.word	initrd_len
-		.word	params
+		.word	kernel_start
+		.word	kernel_addr
+		.word	kernel_len
+
+		.word	0x54410005		@ r5 = ATAG_INITRD
+		.word	initrd_virt		@ r6
+		.word	initrd_len		@ r7
+		.word	params			@ r8
 
 		.type	kernel_start,#object
 		.type	initrd_start,#object
diff --git a/arch/arm/boot/compressed/Makefile b/arch/arm/boot/compressed/Makefile
index bf8f1c947..7f077cd21 100644
--- a/arch/arm/boot/compressed/Makefile
+++ b/arch/arm/boot/compressed/Makefile
@@ -66,7 +66,7 @@ $(HEAD): 	$(HEAD:.o=.S)
 		$(CC) $(AFLAGS) -traditional -c $(HEAD:.o=.S)
 
 piggy.o:	$(SYSTEM)
-		$(OBJCOPY) $(SYSTEM) piggy
+		$(OBJCOPY) -O binary -R .note -R .comment -S $(SYSTEM) piggy
 		gzip $(GZFLAGS) < piggy > piggy.gz
 		$(LD) -r -o $@ -b binary piggy.gz
 		rm -f piggy piggy.gz
diff --git a/arch/arm/boot/compressed/head-ftvpci.S b/arch/arm/boot/compressed/head-ftvpci.S
index a8c806ef3..8e10667ca 100644
--- a/arch/arm/boot/compressed/head-ftvpci.S
+++ b/arch/arm/boot/compressed/head-ftvpci.S
@@ -6,6 +6,13 @@
  * Special startup code for FTV PCI board.
  */
 
+/*
+ * 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.
+ */
+
 	.section        ".start", #alloc, #execinstr
 ftv_start:
 	mcr	p15, 0, r0, c7, c5, 0		@ flush I cache
diff --git a/arch/arm/boot/compressed/head-l7200.S b/arch/arm/boot/compressed/head-l7200.S
index ecf4d6881..8ab9eafdb 100644
--- a/arch/arm/boot/compressed/head-l7200.S
+++ b/arch/arm/boot/compressed/head-l7200.S
@@ -3,7 +3,7 @@
  * 
  * Copyright (C) 2000 Steve Hill <sjhill@cotw.com>
  * 
- * Some code borrowed from Nicola Pitre's 'head-sa1100.S' file. This
+ * Some code borrowed from Nicolas Pitre's 'head-sa1100.S' file. This
  * is merged with head.S by the linker.
  */
 
@@ -16,15 +16,14 @@
 		.section        ".start", #alloc, #execinstr
 
 __L7200_start:
-
 		mov	r0, #0x00100000		@ FLASH address of initrd
 		mov	r2, #0xf1000000		@ RAM address of initrd
-		add	r1, r2, #0x00700000	@ Size of initrd 
+		add	r3, r2, #0x00700000	@ Size of initrd 
 1:
-		ldmia   r0!, {r3, r4, r5, r6}
-		stmia   r2!, {r3, r4, r5, r6}
-		cmp	r2, r1
+		ldmia   r0!, {r4, r5, r6, r7}
+		stmia   r2!, {r4, r5, r6, r7}
+		cmp	r2, r3
 		ble	1b
-		
+
 		mov	r8, #0			@ Zero it out
 		mov	r7, #19			@ Set architecture ID
diff --git a/arch/arm/boot/compressed/head-sa1100.S b/arch/arm/boot/compressed/head-sa1100.S
index 73cef4bd8..f49ceb5e6 100644
--- a/arch/arm/boot/compressed/head-sa1100.S
+++ b/arch/arm/boot/compressed/head-sa1100.S
@@ -11,10 +11,6 @@
 #include <linux/linkage.h>
 #include <asm/mach-types.h>
 
-#ifndef CONFIG_ARCH_SA1100
-#error What am I doing here...
-#endif
-
 		.section        ".start", #alloc, #execinstr
 
 __SA1100_start:
@@ -54,6 +50,8 @@ __SA1100_start:
 		bic	r0, r0, #0x0d		@ clear WB, DC, MMU
 		bic	r0, r0, #0x1000		@ clear Icache
 		mcr	p15, 0, r0, c1, c0, 0
+
+#ifdef CONFIG_ANGELBOOT
 /*
  * Pause for a short time so that we give enough time
  * for the host to start a terminal up.
@@ -61,3 +59,5 @@ __SA1100_start:
 		mov	r0, #0x00200000
 1:		subs	r0, r0, #1
 		bne	1b
+#endif
+
diff --git a/arch/arm/boot/compressed/head-shark.S b/arch/arm/boot/compressed/head-shark.S
new file mode 100644
index 000000000..817b75366
--- /dev/null
+++ b/arch/arm/boot/compressed/head-shark.S
@@ -0,0 +1,121 @@
+/* The head-file for the Shark
+ * by Alexander.Schulz@stud.uni-karlsruhe.de
+ *
+ * Does the following:
+ * - get the memory layout from firmware. This can only be done as long as the mmu
+ *   is still on.
+ * - switch the mmu off, so we have physical addresses
+ * - copy the kernel to 0x08508000. This is done to have a fixed address where the
+ *   C-parts (misc.c) are executed. This address must be known at compile-time,
+ *   but the load-address of the kernel depends on how much memory is installed.
+ * - Jump to this location.
+ * - Set r8 with 0, r7 with the architecture ID for head.S
+ */
+
+#include <linux/linkage.h>
+
+#include <asm/assembler.h>
+	
+		.section	".start", #alloc, #execinstr
+
+		b	__beginning
+	
+__serial_addr:	.long	0xf7eff3f8
+		.long	0				@ space
+__ofw_data:	.long	0				@ the number of memory blocks
+		.space  128				@ (startaddr,size) ...
+		.space  128				@ bootargs
+		.align
+
+__beginning:	mov	r4, r0				@ save the entry to the firmware
+
+		mov	r0, #0xC0			@ disable irq and fiq
+		mov	r1, r0
+		mrs	r3, cpsr_all
+		bic	r2, r3, r0
+		eor	r2, r2, r1
+		msr	cpsr_all, r2
+
+		ldr	r0, __serial_addr		@ disable serial interrupt
+		mov	r1, #0				@ hangs the machine, I don t know why.
+		strb	r1, [r0, #0x01]
+
+		mov	r0, r4				@ get the Memory layout from firmware
+		adr	r1, __ofw_data
+		add	r2, r1, #4
+		mov	lr, pc
+		b	SYMBOL_NAME(ofw_init)
+		mov	r1, #0
+
+		adr	r2, __mmu_off			@ calculate physical address
+		sub	r2, r2, #0xf0000000		@ openprom maps us at f000 virt, 0e50 phys
+		adr	r0, __ofw_data
+		ldr	r0, [r0, #4]
+		add	r2, r2, r0
+		add	r2, r2, #0x00500000
+
+		mrc	p15, 0, r3, c1, c0
+		bic	r3, r3, #0xC			@ Write Buffer and DCache
+		bic	r3, r3, #0x1000			@ ICache
+		mcr	p15, 0, r3, c1, c0		@ disabled
+
+		mov	r0, #0
+		mcr	p15, 0, r0, c7, c7		@ flush I,D caches on v4
+		mcr	p15, 0, r0, c7, c10, 4		@ drain write buffer on v4
+		mcr	p15, 0, r0, c8, c7		@ flush I,D TLBs on v4
+
+		bic	r3, r3, #0x1			@ MMU
+		mcr	p15, 0, r3, c1, c0		@ disabled
+
+		mov	pc, r2
+
+__copy_target:	.long	0x08508000
+__copy_end:	.long	0x08608000
+		
+		.word	_start
+		.word	__bss_start
+
+		.align
+__temp_stack:	.space 128
+
+__mmu_off:
+		adr	r0, __ofw_data
+		ldr	r0, [r0, #4]
+		orr	r0, r0, #0x00600000
+	
+		ldr	r1, __copy_end
+		ldr	r3, __copy_target
+
+/* r0 = 0x0e600000 (current end of kernelcode)
+ * r3 = 0x08508000 (where it should begin)
+ * r1 = 0x08608000 (end of copying area, 1MB)
+ * The kernel is compressed, so 1 MB should be enough.
+ * copy the kernel to the beginning of physical memory
+ * We start from the highest address, so we can copy
+ * from 0x08500000 to 0x08508000 if we have only 8MB
+ */
+
+	
+__Copy:		ldr	r2, [r0], #-4
+		str	r2, [r1], #-4
+		teq	r1, r3
+		bne	__Copy
+		/* and jump to it */
+		adr	r2, __go_on
+		adr	r0, __ofw_data
+		ldr	r0, [r0, #4]
+		sub	r2, r2, r0
+		sub	r2, r2, #0x00500000
+		ldr	r0, __copy_target
+		add	r2, r2, r0
+		mov	pc, r2
+
+__go_on:
+		adr	sp, __temp_stack
+		add	sp, sp, #128
+		adr	r0, __ofw_data
+		mov	lr, pc
+		b	SYMBOL_NAME(create_params)
+	
+		mov	r8, #0
+		mov	r7, #15
diff --git a/arch/arm/boot/compressed/head.S b/arch/arm/boot/compressed/head.S
index 21bdb43c0..0e384ec72 100644
--- a/arch/arm/boot/compressed/head.S
+++ b/arch/arm/boot/compressed/head.S
@@ -112,8 +112,7 @@ start:
 		 */
 
 		.text
-1:		mrc	p15, 0, r6, c0, c0	@ get processor ID
-		adr	r2, LC0
+1:		adr	r2, LC0
 		ldmia	r2, {r2, r3, r4, r5, sp}
 
 		mov	r0, #0
@@ -124,14 +123,15 @@ start:
 		cmp	r2, r3
 		blt	1b
 
+		mrc	p15, 0, r6, c0, c0	@ get processor ID
 		bl	cache_on
 
 		mov	r1, sp			@ malloc space above stack
 		add	r2, sp, #0x10000	@ 64k max
 
 		teq	r4, r5			@ will we overwrite ourselves?
-		moveq	r5, r2
-		movne	r5, r4
+		moveq	r5, r2			@ decompress after image
+		movne	r5, r4			@ decompress to final location
 
 		mov	r0, r5
 		mov	r3, r7
diff --git a/arch/arm/boot/compressed/ofw-shark.c b/arch/arm/boot/compressed/ofw-shark.c
new file mode 100644
index 000000000..0b4474c59
--- /dev/null
+++ b/arch/arm/boot/compressed/ofw-shark.c
@@ -0,0 +1,216 @@
+/*
+ * linux/arch/arm/boot/compressed/ofw-shark.c
+ *
+ * by Alexander.Schulz@stud.uni-karlsruhe.de
+ *
+ * This file is used to get some basic information
+ * about the memory layout of the shark we are running
+ * on. Memory is usually divided in blocks a 8 MB.
+ * And bootargs are copied from OpenFirmware.
+ */
+
+
+#include <linux/kernel.h>
+#include <asm/setup.h>
+#include <asm/page.h>
+
+
+asmlinkage void
+create_params (unsigned long *buffer)
+{
+	/* Is there a better address? Also change in kernel/arch.c */
+	struct param_struct *params = (struct param_struct *) 0x08003000;
+	int j,i,m,k,nr_banks,size;
+
+	for (j=0;j<256;j++) params->u1.unused[j]=0;
+
+	size=0;
+	nr_banks=(unsigned int) buffer[0];
+	if (nr_banks > NR_BANKS) nr_banks = NR_BANKS;
+	for (j=0;j<nr_banks;j++){
+		/* search the lowest address and put it into the next entry   */
+		/* not a fast sort algorithm, but there are at most 8 entries */
+		/* and this is used only once anyway                          */
+		m=0xffffffff;
+		for (i=0;i<(unsigned int) buffer[0];i++){
+			if (buffer[2*i+1]<m) {
+				m=buffer[2*i+1];
+				k=i;
+			}
+		}
+	  
+		/* using pages_in_bank for startaddress and size               */
+		/* start is in bytes, size in pages and not bigger than 0x2000 */
+		/* This is decoded in fixup_shark in arch/arm/kernel/arch.c    */
+		params->u1.s.pages_in_bank[j]=buffer[2*k+1]|(buffer[2*k+2]/PAGE_SIZE);
+		size += buffer[2*k+2];
+
+		buffer[2*k+1]=0xffffffff;                    /* mark as copied             */
+	}
+	
+	params->u1.s.page_size = PAGE_SIZE;
+	params->u1.s.nr_pages  = size/PAGE_SIZE;
+	params->u1.s.flags     = FLAG_READONLY;
+	
+	/* Copy over the bootargs */
+	for (j=0;j<128/4;j++) {
+		((unsigned long *) params->commandline)[j]=buffer[33+j];
+	}
+}
+
+
+typedef int (*ofw_handle_t)(void *);
+
+/* Everything below is called with a wrong MMU setting.
+ * This means: no string constants, no initialization of
+ * arrays, no global variables! This is ugly but I didn't
+ * want to write this in assembler :-)
+ */
+
+int
+of_decode_int(const unsigned char *p)
+{
+	unsigned int i = *p++ << 8;
+	i = (i + *p++) << 8;
+	i = (i + *p++) << 8;
+	return (i + *p);
+}
+  
+int
+OF_finddevice(ofw_handle_t openfirmware, char *name)
+{
+	unsigned int args[8];
+	char service[12];
+
+	service[0]='f';
+	service[1]='i';
+	service[2]='n';
+	service[3]='d';
+	service[4]='d';
+	service[5]='e';
+	service[6]='v';
+	service[7]='i';
+	service[8]='c';
+	service[9]='e';
+	service[10]='\0';
+
+	args[0]=(unsigned int)service;
+	args[1]=1;
+	args[2]=1;
+	args[3]=(unsigned int)name;
+
+	if (openfirmware(args) == -1)
+		return -1;
+	return args[4];
+}
+
+int
+OF_getproplen(ofw_handle_t openfirmware, int handle, char *prop)
+{
+	unsigned int args[8];
+	char service[12];
+
+	service[0]='g';
+	service[1]='e';
+	service[2]='t';
+	service[3]='p';
+	service[4]='r';
+	service[5]='o';
+	service[6]='p';
+	service[7]='l';
+	service[8]='e';
+	service[9]='n';
+	service[10]='\0';
+
+	args[0] = (unsigned int)service;
+	args[1] = 2;
+	args[2] = 1;
+	args[3] = (unsigned int)handle;
+	args[4] = (unsigned int)prop;
+
+	if (openfirmware(args) == -1)
+		return -1;
+	return args[5];
+}
+  
+int
+OF_getprop(ofw_handle_t openfirmware, int handle, char *prop, void *buf, unsigned int buflen)
+{
+	unsigned int args[8];
+	char service[8];
+
+	service[0]='g';
+	service[1]='e';
+	service[2]='t';
+	service[3]='p';
+	service[4]='r';
+	service[5]='o';
+	service[6]='p';
+	service[7]='\0';
+
+	args[0] = (unsigned int)service;
+	args[1] = 4;
+	args[2] = 1;
+	args[3] = (unsigned int)handle;
+	args[4] = (unsigned int)prop;
+	args[5] = (unsigned int)buf;
+	args[6] = buflen;
+
+	if (openfirmware(args) == -1)
+		return -1;
+	return args[7];
+}
+  
+asmlinkage void ofw_init(ofw_handle_t o, int *nomr, int *pointer)
+{
+	int phandle,i,mem_len,buffer[32];
+	char temp[12];
+  
+	temp[0]='/';
+	temp[1]='m';
+	temp[2]='e';
+	temp[3]='m';
+	temp[4]='o';
+	temp[5]='r';
+	temp[6]='y';
+	temp[7]='\0';
+
+	phandle=OF_finddevice(o,temp);
+
+	temp[0]='r';
+	temp[1]='e';
+	temp[2]='g';
+	temp[3]='\0';
+
+	mem_len = OF_getproplen(o,phandle, temp);
+	OF_getprop(o,phandle, temp, buffer, mem_len);
+	*nomr=mem_len >> 3;
+
+	for (i=0; i<=mem_len/4; i++) pointer[i]=of_decode_int((const unsigned char *)&buffer[i]);
+
+	temp[0]='/';
+	temp[1]='c';
+	temp[2]='h';
+	temp[3]='o';
+	temp[4]='s';
+	temp[5]='e';
+	temp[6]='n';
+	temp[7]='\0';
+
+	phandle=OF_finddevice(o,temp);
+
+	temp[0]='b';
+	temp[1]='o';
+	temp[2]='o';
+	temp[3]='t';
+	temp[4]='a';
+	temp[5]='r';
+	temp[6]='g';
+	temp[7]='s';
+	temp[8]='\0';
+
+	mem_len = OF_getproplen(o,phandle, temp);
+	OF_getprop(o,phandle, temp, buffer, mem_len);
+	for (i=0; i<=mem_len/4; i++) pointer[i+32]=buffer[i];
+  
+}
diff --git a/arch/arm/boot/compressed/setup-sa1100.S b/arch/arm/boot/compressed/setup-sa1100.S
index 3babde180..32e77c285 100644
--- a/arch/arm/boot/compressed/setup-sa1100.S
+++ b/arch/arm/boot/compressed/setup-sa1100.S
@@ -33,13 +33,11 @@ UART3_BASE:	.long	0x80050000
 #define UTSR0           0x1c
 #define UTSR1           0x20
 
-#define BAUD_DIV_230400	0x000
-#define BAUD_DIV_115200	0x001
-#define BAUD_DIV_57600	0x003
-#define BAUD_DIV_38400	0x005
-#define BAUD_DIV_19200	0x00b
-#define BAUD_DIV_9600	0x017
-#define BAUD_DIV	BAUD_DIV_9600
+#ifndef CONFIG_SA1100_DEFAULT_BAUDRATE
+#define CONFIG_SA1100_DEFAULT_BAUDRATE	9600
+#endif
+
+#define BAUD_DIV	((230400/CONFIG_SA1100_DEFAULT_BAUDRATE)-1)
 
 SCR_loc:	.long	SYMBOL_NAME(SCR_value)
 #define GPIO_2_9	0x3fc
@@ -92,7 +90,7 @@ skip_SCR:
 		bne	skip_uart
 
 		@ UART3 if Assabet is used with Neponset
-		teq	r3, #25			@ if Assabet 
+		teq	r3, #MACH_TYPE_ASSABET	@ if Assabet 
 		tsteq	r2, #(1 << 9)		@ ... and Neponset present
 		ldreq	r0, UART3_BASE
 		beq	uart_init
diff --git a/arch/arm/boot/compressed/vmlinux.lds.in b/arch/arm/boot/compressed/vmlinux.lds.in
index 86ed5e6e3..8608996e9 100644
--- a/arch/arm/boot/compressed/vmlinux.lds.in
+++ b/arch/arm/boot/compressed/vmlinux.lds.in
@@ -19,12 +19,13 @@ SECTIONS
 
   .text : {
     _start = .;
-    head.o(.start)
     *(.start)
-    head.o(.text)
     *(.text)
     *(.fixup)
     *(.gnu.warning)
+    *(.rodata)
+    *(.glue_7)
+    *(.glue_7t)
     input_data = .;
     piggy.o
     input_data_end = .;
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index ed5115c53..45956ea07 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -10,16 +10,14 @@ USE_STANDARD_AS_RULE := true
 HEAD_OBJ  = head-$(PROCESSOR).o
 ENTRY_OBJ = entry-$(PROCESSOR).o
 
-AFLAGS_head-armv.o := -DTEXTADDR=$(TEXTADDR) -traditional
-AFLAGS_head-armo.o := -DTEXTADDR=$(TEXTADDR) -traditional
+AFLAGS_head-armv.o := -DTEXTADDR=$(TEXTADDR)
+AFLAGS_head-armo.o := -DTEXTADDR=$(TEXTADDR)
 
 O_OBJS_arc        = dma-arc.o oldlatches.o
 O_OBJS_rpc        = dma-rpc.o
 O_OBJS_footbridge = dma-footbridge.o isa.o
 O_OBJS_l7200      = fiq.o
 
-leds-ebsa110	  = leds-ebsa110.o
-
 pci-nexuspci	  = plx90x0.o
 pci-footbridge	  = dec21285.o
 pci-shark	  = via82c505.o
@@ -31,37 +29,29 @@ O_TARGET := kernel.o
 
 # Object file lists.
 
-obj-y			:= arch.o dma.o $(ENTRY_OBJ) irq.o process.o ptrace.o \
-			   semaphore.o setup.o signal.o sys_arm.o time.o \
-			   traps.o $(O_OBJS_$(MACHINE))
-obj-m			:=
-obj-n			:=
-obj-			:=
+obj-y		:= arch.o dma.o $(ENTRY_OBJ) irq.o process.o ptrace.o    \
+		   semaphore.o setup.o signal.o sys_arm.o time.o traps.o \
+		   $(O_OBJS_$(MACHINE))
+obj-m		:=
+obj-n		:=
+obj-		:=
+
+export-objs	:= armksyms.o dma.o ecard.o fiq.o oldlatches.o time.o
 
-export-objs		:= armksyms.o dma.o ecard.o \
-			   $(leds-$(MACHINE)) oldlatches.o \
-			   time.o
+no-irq-arch	:= $(CONFIG_ARCH_INTEGRATOR) $(CONFIG_ARCH_CLPS711X) \
+		   $(CONFIG_ARCH_FOOTBRIDGE) $(CONFIG_ARCH_EBSA110)
+
+ifneq ($(findstring y,$(no-irq-arch)),y)
+  obj-y		+= irq-arch.o
+endif
 
 obj-$(CONFIG_ARCH_ACORN) += ecard.o fiq.o time-acorn.o
 obj-$(CONFIG_DEBUG_LL)	+= debug-$(PROCESSOR).o
 obj-$(CONFIG_MODULES)	+= armksyms.o
-obj-$(CONFIG_LEDS)	+= $(leds-$(MACHINE))
 obj-$(CONFIG_ARTHUR)	+= arthur.o
 obj-$(CONFIG_ISA_DMA)	+= dma-isa.o
-
 obj-$(CONFIG_PCI)	+= bios32.o $(pci-$(MACHINE)) $(pci-y)
 
-# Files that are both resident and modular; remove from modular.
-
-obj-m		:= $(filter-out $(obj-y), $(obj-m))
-
-# Translate to Rules.make lists.
-
-O_OBJS		:= $(filter-out $(export-objs), $(obj-y))
-OX_OBJS 	:= $(filter	$(export-objs), $(obj-y))
-M_OBJS		:= $(sort $(filter-out $(export-objs), $(obj-m)))
-MX_OBJS 	:= $(sort $(filter     $(export-objs), $(obj-m)))
-
 all: kernel.o $(HEAD_OBJ) init_task.o
 
 include $(TOPDIR)/Rules.make
diff --git a/arch/arm/kernel/bios32.c b/arch/arm/kernel/bios32.c
index 868fbd4a7..bbc43c509 100644
--- a/arch/arm/kernel/bios32.c
+++ b/arch/arm/kernel/bios32.c
@@ -257,11 +257,30 @@ void __init pcibios_fixup_bus(struct pci_bus *bus)
 			(struct arm_pci_sysdata *)bus->sysdata;
 	struct arm_bus_sysdata *busdata;
 
-	if (bus->number < MAX_NR_BUS)
-		busdata = sysdata->bus + bus->number;
-	else
+	if (bus->number >= MAX_NR_BUS)
 		BUG();
 
+	if (bus->self) {
+		struct pci_dev *dev = bus->self;
+		int i;
+
+		for (i = 0; i < 3; i++) {
+			bus->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES + i];
+			bus->resource[i]->name = bus->name;
+		}
+		bus->resource[0]->start = ioport_resource.start;
+		bus->resource[0]->end   = ioport_resource.end;
+		bus->resource[0]->flags |= pci_bridge_check_io(dev);
+		bus->resource[1]->start = iomem_resource.start;
+		bus->resource[1]->end   = iomem_resource.end;
+		bus->resource[1]->flags |= IORESOURCE_MEM;
+
+		/* Turn off downsteam prefetchable memory address range */
+		bus->resource[2]->start = 1024*1024;
+		bus->resource[2]->end   = bus->resource[2]->start - 1;
+	}
+
+	busdata = sysdata->bus + bus->number;
 	busdata->max_lat = 255;
 
 	/*
@@ -363,10 +382,6 @@ void __init pcibios_fixup_bus(struct pci_bus *bus)
 void __init
 pcibios_fixup_pbus_ranges(struct pci_bus *bus, struct pbus_set_ranges_data *ranges)
 {
-	ranges->io_start -= bus->resource[0]->start;
-	ranges->io_end -= bus->resource[0]->start;
-	ranges->mem_start -= bus->resource[1]->start;
-	ranges->mem_end -= bus->resource[1]->start;
 }
 
 u8 __init no_swizzle(struct pci_dev *dev, u8 *pin)
@@ -442,7 +457,8 @@ void __init pcibios_init(void)
 	hw_pci->init(&sysdata);
 
 	/*
-	 * Other architectures don't seem to do this... should we?
+	 * Claim the currently allocated resources.  This ensures
+	 * that we will not allocate an already inuse region.
 	 */
 	pcibios_claim_resources();
 
diff --git a/arch/arm/kernel/dma-arc.c b/arch/arm/kernel/dma-arc.c
index 121a2792b..ad9ca3de5 100644
--- a/arch/arm/kernel/dma-arc.c
+++ b/arch/arm/kernel/dma-arc.c
@@ -40,7 +40,7 @@ static void arc_floppy_data_enable_dma(dmach_t channel, dma_t *dma)
 		memcpy ((void *)0x1c, (void *)&fdc1772_dma_read,
 			&fdc1772_dma_read_end - &fdc1772_dma_read);
 		fdc1772_setupdma(dma->buf.length, dma->buf.address); /* Sets data pointer up */
-		enable_irq (64);
+		enable_fiq(FIQ_FLOPPYDATA);
 		restore_flags(flags);
 	   }
 	   break;
@@ -55,7 +55,7 @@ static void arc_floppy_data_enable_dma(dmach_t channel, dma_t *dma)
 		memcpy ((void *)0x1c, (void *)&fdc1772_dma_write,
 			&fdc1772_dma_write_end - &fdc1772_dma_write);
 		fdc1772_setupdma(dma->buf.length, dma->buf.address); /* Sets data pointer up */
-		enable_irq (64);
+		enable_fiq(FIQ_FLOPPYDATA;
 
 		restore_flags(flags);
 	    }
@@ -102,7 +102,7 @@ static int arc_floppy_cmdend_get_dma_residue(dmach_t channel, dma_t *dma)
 
 static void arc_disable_dma(dmach_t channel, dma_t *dma)
 {
-	disable_irq(dma->dma_irq);
+	disable_fiq(dma->dma_irq);
 }
 
 static struct dma_ops arc_floppy_data_dma_ops = {
@@ -158,12 +158,12 @@ static void a5k_floppy_enable_dma(dmach_t channel, dma_t *dma)
 	regs.ARM_r10 = dma->buf.address;
 	regs.ARM_fp = FLOPPYDMA_BASE;
 	set_fiq_regs(&regs);
-	enable_irq(dma->dma_irq);
+	enable_fiq(dma->dma_irq);
 }
 
 static void a5k_floppy_disable_dma(dmach_t channel, dma_t *dma)
 {
-	disable_irq(dma->dma_irq);
+	disable_fiq(dma->dma_irq);
 	release_fiq(&fh);
 }
 
@@ -192,15 +192,15 @@ void __init arch_dma_init(dma_t *dma)
 {
 #if defined(CONFIG_BLK_DEV_FD1772) || defined(CONFIG_BLK_DEV_FD1772_MODULE)
 	if (machine_is_archimedes()) {
-		dma[DMA_VIRTUAL_FLOPPY0].dma_irq = 64;
+		dma[DMA_VIRTUAL_FLOPPY0].dma_irq = FIQ_FLOPPYDATA;
 		dma[DMA_VIRTUAL_FLOPPY0].d_ops   = &arc_floppy_data_dma_ops;
-		dma[DMA_VIRTUAL_FLOPPY1].dma_irq = 65;
+		dma[DMA_VIRTUAL_FLOPPY1].dma_irq = 1;
 		dma[DMA_VIRTUAL_FLOPPY1].d_ops   = &arc_floppy_cmdend_dma_ops;
 	}
 #endif
 #ifdef CONFIG_ARCH_A5K
 	if (machine_is_a5k()) {
-		dma[DMA_VIRTUAL_FLOPPY0].dma_irq = 64;
+		dma[DMA_VIRTUAL_FLOPPY0].dma_irq = FIQ_FLOPPYDATA;
 		dma[DMA_VIRTUAL_FLOPPY0].d_ops   = &a5k_floppy_dma_ops;
 	}
 #endif
diff --git a/arch/arm/kernel/dma-rpc.c b/arch/arm/kernel/dma-rpc.c
index b46ca7753..402b71cdb 100644
--- a/arch/arm/kernel/dma-rpc.c
+++ b/arch/arm/kernel/dma-rpc.c
@@ -10,7 +10,7 @@
  *  DMA functions specific to RiscPC architecture
  */
 #include <linux/sched.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/mman.h>
 #include <linux/init.h>
 #include <linux/pci.h>
@@ -42,11 +42,11 @@ typedef struct {
 #define TRANSFER_SIZE	2
 
 #define CURA	(0)
-#define ENDA	((IOMD_IO0ENDA - IOMD_IO0CURA) << 2)
-#define CURB	((IOMD_IO0CURB - IOMD_IO0CURA) << 2)
-#define ENDB	((IOMD_IO0ENDB - IOMD_IO0CURA) << 2)
-#define CR	((IOMD_IO0CR - IOMD_IO0CURA) << 2)
-#define ST	((IOMD_IO0ST - IOMD_IO0CURA) << 2)
+#define ENDA	(IOMD_IO0ENDA - IOMD_IO0CURA)
+#define CURB	(IOMD_IO0CURB - IOMD_IO0CURA)
+#define ENDB	(IOMD_IO0ENDB - IOMD_IO0CURA)
+#define CR	(IOMD_IO0CR - IOMD_IO0CURA)
+#define ST	(IOMD_IO0ST - IOMD_IO0CURA)
 
 #define state_prog_a	0
 #define state_wait_a	1
@@ -93,14 +93,14 @@ static void iomd_get_next_sg(struct scatterlist *sg, dma_t *dma)
 
 static inline void iomd_setup_dma_a(struct scatterlist *sg, dma_t *dma)
 {
-	outl_t(sg->dma_address, dma->dma_base + CURA);
-	outl_t(sg->length, dma->dma_base + ENDA);
+	iomd_writel(sg->dma_address, dma->dma_base + CURA);
+	iomd_writel(sg->length, dma->dma_base + ENDA);
 }
 
 static inline void iomd_setup_dma_b(struct scatterlist *sg, dma_t *dma)
 {
-	outl_t(sg->dma_address, dma->dma_base + CURB);
-	outl_t(sg->length, dma->dma_base + ENDB);
+	iomd_writel(sg->dma_address, dma->dma_base + CURB);
+	iomd_writel(sg->length, dma->dma_base + ENDB);
 }
 
 static void iomd_dma_handle(int irq, void *dev_id, struct pt_regs *regs)
@@ -116,7 +116,7 @@ static void iomd_dma_handle(int irq, void *dev_id, struct pt_regs *regs)
 			dma->state = state_wait_a;
 
 		case state_wait_a:
-			status = inb_t(dma->dma_base + ST);
+			status = iomd_readb(dma->dma_base + ST);
 			switch (status & (DMA_ST_OFL|DMA_ST_INT|DMA_ST_AB)) {
 			case DMA_ST_OFL|DMA_ST_INT:
 				iomd_get_next_sg(&dma->cur_sg, dma);
@@ -137,7 +137,7 @@ static void iomd_dma_handle(int irq, void *dev_id, struct pt_regs *regs)
 			break;
 
 		case state_wait_b:
-			status = inb_t(dma->dma_base + ST);
+			status = iomd_readb(dma->dma_base + ST);
 			switch (status & (DMA_ST_OFL|DMA_ST_INT|DMA_ST_AB)) {
 			case DMA_ST_OFL|DMA_ST_INT|DMA_ST_AB:
 				iomd_get_next_sg(&dma->cur_sg, dma);
@@ -193,14 +193,14 @@ static void iomd_enable_dma(dmach_t channel, dma_t *dma)
 				PCI_DMA_FROMDEVICE : PCI_DMA_TODEVICE);
 		}
 
-		outb_t(DMA_CR_C, dma_base + CR);
+		iomd_writeb(DMA_CR_C, dma_base + CR);
 		dma->state = state_prog_a;
 	}
 		
 	if (dma->dma_mode == DMA_MODE_READ)
 		ctrl |= DMA_CR_D;
 
-	outb_t(ctrl, dma_base + CR);
+	iomd_writeb(ctrl, dma_base + CR);
 	enable_irq(dma->dma_irq);
 }
 
@@ -210,8 +210,8 @@ static void iomd_disable_dma(dmach_t channel, dma_t *dma)
 	unsigned int ctrl;
 
 	disable_irq(dma->dma_irq);
-	ctrl = inb_t(dma_base + CR);
-	outb_t(ctrl & ~DMA_CR_E, dma_base + CR);
+	ctrl = iomd_readb(dma_base + CR);
+	iomd_writeb(ctrl & ~DMA_CR_E, dma_base + CR);
 }
 
 static int iomd_set_dma_speed(dmach_t channel, dma_t *dma, int cycle)
@@ -227,7 +227,7 @@ static int iomd_set_dma_speed(dmach_t channel, dma_t *dma, int cycle)
 	else
 		speed = 0;
 
-	tcr = inb(IOMD_DMATCR);
+	tcr = iomd_readb(IOMD_DMATCR);
 	speed &= 3;
 
 	switch (channel) {
@@ -251,7 +251,7 @@ static int iomd_set_dma_speed(dmach_t channel, dma_t *dma, int cycle)
 		break;
 	}
 
-	outb(tcr, IOMD_DMATCR);
+	iomd_writeb(tcr, IOMD_DMATCR);
 
 	return speed;
 }
@@ -287,7 +287,7 @@ static void floppy_enable_dma(dmach_t channel, dma_t *dma)
 
 	regs.ARM_r9  = dma->buf.length;
 	regs.ARM_r10 = (unsigned long)dma->buf.address;
-	regs.ARM_fp  = (unsigned long)PCIO_FLOPPYDMABASE;
+	regs.ARM_fp  = FLOPPYDMA_BASE;
 
 	if (claim_fiq(&fh)) {
 		printk("floppydma: couldn't claim FIQ.\n");
@@ -296,12 +296,12 @@ static void floppy_enable_dma(dmach_t channel, dma_t *dma)
 
 	set_fiq_handler(fiqhandler_start, fiqhandler_length);
 	set_fiq_regs(&regs);
-	enable_irq(dma->dma_irq);
+	enable_fiq(dma->dma_irq);
 }
 
 static void floppy_disable_dma(dmach_t channel, dma_t *dma)
 {
-	disable_irq(dma->dma_irq);
+	disable_fiq(dma->dma_irq);
 	release_fiq(&fh);
 }
 
@@ -334,32 +334,32 @@ static struct dma_ops sound_dma_ops = {
 
 void __init arch_dma_init(dma_t *dma)
 {
-	outb(0, IOMD_IO0CR);
-	outb(0, IOMD_IO1CR);
-	outb(0, IOMD_IO2CR);
-	outb(0, IOMD_IO3CR);
+	iomd_writeb(0, IOMD_IO0CR);
+	iomd_writeb(0, IOMD_IO1CR);
+	iomd_writeb(0, IOMD_IO2CR);
+	iomd_writeb(0, IOMD_IO3CR);
 
-	outb(0xa0, IOMD_DMATCR);
+	iomd_writeb(0xa0, IOMD_DMATCR);
 
-	dma[DMA_0].dma_base		= ioaddr(IOMD_IO0CURA);
+	dma[DMA_0].dma_base		= IOMD_IO0CURA;
 	dma[DMA_0].dma_irq		= IRQ_DMA0;
 	dma[DMA_0].d_ops		= &iomd_dma_ops;
-	dma[DMA_1].dma_base		= ioaddr(IOMD_IO1CURA);
+	dma[DMA_1].dma_base		= IOMD_IO1CURA;
 	dma[DMA_1].dma_irq		= IRQ_DMA1;
 	dma[DMA_1].d_ops		= &iomd_dma_ops;
-	dma[DMA_2].dma_base		= ioaddr(IOMD_IO2CURA);
+	dma[DMA_2].dma_base		= IOMD_IO2CURA;
 	dma[DMA_2].dma_irq		= IRQ_DMA2;
 	dma[DMA_2].d_ops		= &iomd_dma_ops;
-	dma[DMA_3].dma_base		= ioaddr(IOMD_IO3CURA);
+	dma[DMA_3].dma_base		= IOMD_IO3CURA;
 	dma[DMA_3].dma_irq		= IRQ_DMA3;
 	dma[DMA_3].d_ops		= &iomd_dma_ops;
-	dma[DMA_S0].dma_base		= ioaddr(IOMD_SD0CURA);
+	dma[DMA_S0].dma_base		= IOMD_SD0CURA;
 	dma[DMA_S0].dma_irq		= IRQ_DMAS0;
 	dma[DMA_S0].d_ops		= &iomd_dma_ops;
-	dma[DMA_S1].dma_base		= ioaddr(IOMD_SD1CURA);
+	dma[DMA_S1].dma_base		= IOMD_SD1CURA;
 	dma[DMA_S1].dma_irq		= IRQ_DMAS1;
 	dma[DMA_S1].d_ops		= &iomd_dma_ops;
-	dma[DMA_VIRTUAL_FLOPPY].dma_irq	= 64;
+	dma[DMA_VIRTUAL_FLOPPY].dma_irq	= FIQ_FLOPPYDATA;
 	dma[DMA_VIRTUAL_FLOPPY].d_ops	= &floppy_dma_ops;
 	dma[DMA_VIRTUAL_SOUND].d_ops	= &sound_dma_ops;
 
@@ -367,5 +367,5 @@ void __init arch_dma_init(dma_t *dma)
 	 * Setup DMA channels 2,3 to be for podules
 	 * and channels 0,1 for internal devices
 	 */
-	outb(DMA_EXT_IO3|DMA_EXT_IO2, IOMD_DMAEXT);
+	iomd_writeb(DMA_EXT_IO3|DMA_EXT_IO2, IOMD_DMAEXT);
 }
diff --git a/arch/arm/kernel/dma.c b/arch/arm/kernel/dma.c
index ad311020f..7f0106953 100644
--- a/arch/arm/kernel/dma.c
+++ b/arch/arm/kernel/dma.c
@@ -12,7 +12,7 @@
  *  DMA facilities.
  */
 #include <linux/module.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/sched.h>
 #include <linux/mman.h>
 #include <linux/init.h>
diff --git a/arch/arm/kernel/ecard.c b/arch/arm/kernel/ecard.c
index 3584c3227..07de0170c 100644
--- a/arch/arm/kernel/ecard.c
+++ b/arch/arm/kernel/ecard.c
@@ -35,7 +35,7 @@
 #include <linux/sched.h>
 #include <linux/interrupt.h>
 #include <linux/mm.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/proc_fs.h>
 #include <linux/init.h>
 
diff --git a/arch/arm/kernel/entry-armo.S b/arch/arm/kernel/entry-armo.S
index e4df9fe1a..65f70c727 100644
--- a/arch/arm/kernel/entry-armo.S
+++ b/arch/arm/kernel/entry-armo.S
@@ -197,8 +197,7 @@ irq_prio_h:	.byte	 0, 8, 9, 8,10,10,10,10,11,11,11,11,10,10,10,10
 #endif
 #define FAULT_CODE_PREFETCH	0x04
 #define FAULT_CODE_WRITE	0x02
-#define FAULT_CODE_USER		0x01
-
+#define FAULT_CODE_FORCECOW	0x01
 
 #define SVC_SAVE_ALL				\
 		str	sp, [sp, #-16]!		;\
@@ -487,7 +486,6 @@ vector_data:	sub	lr, lr, #8		@ Correct lr
 		save_user_regs
 		teqp	pc, #0x00000003		@ NOT a problem - doesnt change mode
 		mask_pc	r0, lr
-		mov	r2, #FAULT_CODE_USER
 		bl	Ldata_do
 		b	ret_from_exception
 
@@ -499,7 +497,6 @@ Ldata_not_user:
 		tst	lr, #0x08000000
 		teqeqp	pc, #0x00000003		@ NOT a problem - doesnt change mode
 		mask_pc	r0, lr
-		mov	r2, #0
 		bl	Ldata_do
 		SVC_RESTORE_ALL
 
@@ -510,6 +507,7 @@ Ldata_illegal_mode:
 
 Ldata_do:	mov	r3, sp
 		ldr	r4, [r0]		@ Get instruction
+		mov	r2, #0
 		tst	r4, #1 << 20		@ Check to see if it is a write instruction
 		orreq	r2, r2, #FAULT_CODE_WRITE @ Indicate write instruction
 		mov	r1, r4, lsr #22		@ Now branch to the relevent processing routine
diff --git a/arch/arm/kernel/fiq.c b/arch/arm/kernel/fiq.c
index 1be23e8fb..da32d69f9 100644
--- a/arch/arm/kernel/fiq.c
+++ b/arch/arm/kernel/fiq.c
@@ -36,12 +36,14 @@
  *  6. Goto 3
  */
 #include <linux/config.h>
+#include <linux/module.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/init.h>
 
 #include <asm/fiq.h>
 #include <asm/io.h>
+#include <asm/irq.h>
 #include <asm/pgalloc.h>
 #include <asm/system.h>
 #include <asm/uaccess.h>
@@ -71,7 +73,7 @@ static inline void protect_page_0(void)
  * - we always relinquish FIQ control
  * - we always reacquire FIQ control
  */
-int fiq_def_op(void *ref, int relinquish)
+static int fiq_def_op(void *ref, int relinquish)
 {
 	if (!relinquish) {
 		unprotect_page_0();
@@ -214,6 +216,24 @@ void release_fiq(struct fiq_handler *f)
 	while (current_fiq->fiq_op(current_fiq->dev_id, 0));
 }
 
+void enable_fiq(int fiq)
+{
+	enable_irq(fiq + FIQ_START);
+}
+
+void disable_fiq(int fiq)
+{
+	disable_irq(fiq + FIQ_START);
+}
+
+EXPORT_SYMBOL(set_fiq_handler);
+EXPORT_SYMBOL(set_fiq_regs);
+EXPORT_SYMBOL(get_fiq_regs);
+EXPORT_SYMBOL(claim_fiq);
+EXPORT_SYMBOL(release_fiq);
+EXPORT_SYMBOL(enable_fiq);
+EXPORT_SYMBOL(disable_fiq);
+
 void __init init_FIQ(void)
 {
 	no_fiq_insn = *(unsigned long *)FIQ_VECTOR;
diff --git a/arch/arm/kernel/irq.c b/arch/arm/kernel/irq.c
index 3c6899264..ae599d255 100644
--- a/arch/arm/kernel/irq.c
+++ b/arch/arm/kernel/irq.c
@@ -24,7 +24,7 @@
 #include <linux/sched.h>
 #include <linux/ioport.h>
 #include <linux/interrupt.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/random.h>
 #include <linux/smp.h>
 #include <linux/init.h>
diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
index 9f50ab6fd..6b045845f 100644
--- a/arch/arm/kernel/process.c
+++ b/arch/arm/kernel/process.c
@@ -17,7 +17,7 @@
 #include <linux/stddef.h>
 #include <linux/unistd.h>
 #include <linux/ptrace.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/user.h>
 #include <linux/delay.h>
 #include <linux/reboot.h>
@@ -92,8 +92,10 @@ void cpu_idle(void)
 		void (*idle)(void) = pm_idle;
 		if (!idle)
 			idle = arch_idle;
+		leds_event(led_idle_start);
 		while (!current->need_resched)
 			idle();
+		leds_event(led_idle_end);
 		schedule();
 #ifndef CONFIG_NO_PGT_CACHE
 		check_pgt_cache();
@@ -364,20 +366,23 @@ void dump_thread(struct pt_regs * regs, struct user * dump)
  */
 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 {
-	extern long sys_exit(int) __attribute__((noreturn));
 	pid_t __ret;
 
 	__asm__ __volatile__(
-	"mov	r0, %1		@ kernel_thread sys_clone
+	"orr	r0, %1, %2	@ kernel_thread sys_clone
 	mov	r1, #0
 	"__syscall(clone)"
-	teq	r0, #0		@ if we are the child
-	moveq	fp, #0		@ ensure that fp is zero
-	mov	%0, r0"
+	movs	%0, r0		@ if we are the child
+	bne	1f
+	mov	fp, #0		@ ensure that fp is zero
+	mov	r0, %4
+	mov	lr, pc
+	mov	pc, %3
+	b	sys_exit
+1:	"
         : "=r" (__ret)
-        : "Ir" (flags | CLONE_VM) : "r0", "r1");
-	if (__ret == 0)
-		sys_exit((fn)(arg));
+        : "Ir" (flags), "I" (CLONE_VM), "r" (fn), "r" (arg)
+	: "r0", "r1", "lr");
 	return __ret;
 }
 
diff --git a/arch/arm/kernel/sys_arm.c b/arch/arm/kernel/sys_arm.c
index 36bbf3ac8..16b8e59ea 100644
--- a/arch/arm/kernel/sys_arm.c
+++ b/arch/arm/kernel/sys_arm.c
@@ -14,7 +14,7 @@
  */
 #include <linux/errno.h>
 #include <linux/sched.h>
-#include <linux/malloc.h>
+#include <linux/slab.h>
 #include <linux/mm.h>
 #include <linux/sem.h>
 #include <linux/msg.h>
diff --git a/arch/arm/kernel/time.c b/arch/arm/kernel/time.c
index 926cdff2f..be24bdcfc 100644
--- a/arch/arm/kernel/time.c
+++ b/arch/arm/kernel/time.c
@@ -134,32 +134,17 @@ void (*leds_event)(led_event_t) = dummy_leds_event;
 #ifdef CONFIG_MODULES
 EXPORT_SYMBOL(leds_event);
 #endif
+#endif
 
+#ifdef CONFIG_LEDS_TIMER
 static void do_leds(void)
 {
-#ifdef CONFIG_LEDS_CPU
-	{
-		static int last_pid;
-
-		if (current->pid != last_pid) {
-			last_pid = current->pid;
-			if (last_pid)
-				leds_event(led_idle_end);
-			else
-				leds_event(led_idle_start);
-		}
-	}
-#endif
-#ifdef CONFIG_LEDS_TIMER
-	{
-		static unsigned int count = 50;
+	static unsigned int count = 50;
 
-		if (--count == 0) {
-			count = 50;
-			leds_event(led_timer);
-		}
+	if (--count == 0) {
+		count = 50;
+		leds_event(led_timer);
 	}
-#endif
 }
 #else
 #define do_leds()
diff --git a/arch/arm/kernel/traps.c b/arch/arm/kernel/traps.c
index 9d0cc5c69..8455caf78 100644
--- a/arch/arm/kernel/traps.c
+++ b/arch/arm/kernel/traps.c
@@ -107,7 +107,7 @@ static void dump_instr(struct pt_regs *regs)
 			bad = __get_user(val, &((u32 *)addr)[i]);
 
 		if (!bad)
-			printk(i == 0 ? "(%0*x) " : "%0*x", width, val);
+			printk(i == 0 ? "(%0*x) " : "%0*x ", width, val);
 		else {
 			printk("bad PC value.");
 			break;
@@ -133,7 +133,7 @@ static void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk)
 		printk("no frame pointer");
 		ok = 0;
 	} else if (verify_stack(fp)) {
-		printk("invalid frame pointer %08lx", fp);
+		printk("invalid frame pointer 0x%08x", fp);
 		ok = 0;
 	} else if (fp < 4096+(unsigned long)tsk)
 		printk("frame pointer underflow");
diff --git a/arch/arm/lib/Makefile b/arch/arm/lib/Makefile
index fa7f10117..ee111801f 100644
--- a/arch/arm/lib/Makefile
+++ b/arch/arm/lib/Makefile
@@ -1,58 +1,58 @@
 #
 # linux/arch/arm/lib/Makefile
 #
-# Copyright (C) 1995-1999 Russell King
+# Copyright (C) 1995-2000 Russell King
 #
 
 USE_STANDARD_AS_RULE := true
 
-L_TARGET := lib.a
-L_OBJS   := changebit.o csumipv6.o csumpartial.o csumpartialcopy.o \
-	    csumpartialcopyuser.o clearbit.o copy_page.o findbit.o \
-	    memchr.o memcpy.o memset.o memzero.o setbit.o          \
-	    strncpy_from_user.o strnlen_user.o strchr.o strrchr.o  \
-	    testchangebit.o testclearbit.o testsetbit.o uaccess.o
-
-l-obj-y	:=
-l-obj-n	:=
-
-O_TARGET := lib.o
-O_OBJS   := backtrace.o delay.o
-
-ifeq ($(CONFIG_ARCH_ACORN),y)
-  half := n
-  full := y
+L_TARGET	:= lib.a
+
+obj-y		:= backtrace.o changebit.o csumipv6.o csumpartial.o   \
+		   csumpartialcopy.o csumpartialcopyuser.o clearbit.o \
+		   copy_page.o delay.o findbit.o memchr.o memcpy.o    \
+		   memset.o memzero.o setbit.o strncpy_from_user.o    \
+		   strnlen_user.o strchr.o strrchr.o testchangebit.o  \
+		   testclearbit.o testsetbit.o uaccess.o
+obj-m		:=
+obj-n		:=
+
+export-objs	:= io.o
+
+obj-arc		:= ecard.o io-acorn.o floppydma.o
+obj-rpc		:= ecard.o io-acorn.o floppydma.o
+obj-clps7500	:= io-acorn.o
+obj-footbridge	:= io-pcio.o
+obj-l7200     	:= io-acorn.o
+obj-nexuspci	:= io-pcio.o
+obj-sa1100	:= io-pcio.o
+obj-shark	:= io-shark.o
+obj-integrator	:= io-pcio.o
+obj-clps711x	:= io-shark.o
+
+obj-y		+= $(obj-$(MACHINE))
+
+ifeq ($(CONFIG_CPU_32v4),y)
+  v3		:= n
+  v4		:= y
 else
-  half := y
-  full := n
+  v3		:= y
+  v4		:= n
 endif
 
-L_OBJS_arc		:= ecard.o io-acorn.o floppydma.o
-L_OBJS_rpc		:= ecard.o io-acorn.o floppydma.o
-L_OBJS_clps7500		:= io-acorn.o
-L_OBJS_footbridge	:= io-pcio.o
-L_OBJS_l7200     	:= io-acorn.o
-L_OBJS_nexuspci		:= io-pcio.o
-L_OBJS_sa1100		:= io-pcio.o
-L_OBJS_shark		:= io-shark.o
-L_OBJS_integrator	:= io-pcio.o
-L_OBJS_clps711x		:= io-shark.o
-
-l-obj-y			+= io-readsb.o io-writesb.o
-l-obj-$(full)		+= io-readsw-armv3.o io-writesw-armv3.o
-l-obj-$(half)		+= io-readsw-armv4.o io-writesw-armv4.o
-l-obj-y			+= io-readsl.o io-writesl.o
+obj-y		+= io-readsb.o io-writesb.o
+obj-$(v3)	+= io-readsw-armv3.o io-writesw-armv3.o io-readsl-armv3.o
+obj-$(v4)	+= io-readsw-armv4.o io-writesw-armv4.o io-readsl-armv4.o
+obj-y		+= io-writesl.o
 
 ifeq ($(PROCESSOR),armo)
-  L_OBJS += uaccess-armo.o
+  obj-y		+= uaccess-armo.o
 endif
 
 ifneq ($(MACHINE),ebsa110)
-  OX_OBJS += io.o
+  obj-y		+= io.o
 endif
 
-L_OBJS	+= $(L_OBJS_$(MACHINE)) $(l-obj-y)
-
 include $(TOPDIR)/Rules.make
 
 constants.h: getconsdata.o extractconstants.pl
diff --git a/arch/arm/lib/io-acorn.S b/arch/arm/lib/io-acorn.S
index a40272fdb..db19d932e 100644
--- a/arch/arm/lib/io-acorn.S
+++ b/arch/arm/lib/io-acorn.S
@@ -60,32 +60,32 @@ ENTRY(outsl)
  */
 
 ENTRY(insw)
-		teq	r2, #0
-		RETINSTR(moveq, pc, lr)
 		addr	r0
-		b	__arch_readsw
+		teq	r2, #0
+		bne	__arch_readsw
+		RETINSTR(mov, pc, lr)
 
 ENTRY(insb)
-		teq	r2, #0
-		RETINSTR(moveq, pc, lr)
 		addr	r0
-		b	__arch_readsb
+		teq	r2, #0
+		bne	__arch_readsb
+		RETINSTR(mov, pc, lr)
 
 @ Purpose: write a block of data from memory to a hardware register.
 @ Proto  : outsw(int to_reg, void *from, int len_in_words);
 @ Notes  : increments from
 
 ENTRY(outsw)
-		teq	r2, #0
-		RETINSTR(moveq, pc, lr)
 		addr	r0
-		b	__arch_writesw
+		teq	r2, #0
+		bne	__arch_writesw
+		RETINSTR(mov, pc, lr)
 
 ENTRY(outsb)
-		teq	r2, #0
-		RETINSTR(moveq, pc, lr)
 		addr	r0
-		b	__arch_writesb
+		teq	r2, #0
+		bne	__arch_writesb
+		RETINSTR(mov, pc, lr)
 
 @ Purpose: write a memc register
 @ Proto  : void memc_write(int register, int value);
diff --git a/arch/arm/lib/io-pcio.S b/arch/arm/lib/io-pcio.S
index f57abfc10..243366b86 100644
--- a/arch/arm/lib/io-pcio.S
+++ b/arch/arm/lib/io-pcio.S
@@ -1,4 +1,5 @@
 #include <linux/linkage.h>
+#include <asm/assembler.h>
 #include <asm/hardware.h>
 
 		.equ	pcio_high, PCIO_BASE & 0xff000000
@@ -13,26 +14,37 @@
 
 ENTRY(insl)
 		ioaddr	r0, r0
-		b	__arch_readsl
+		teq	r2, #0
+		bne	__arch_readsl
+		RETINSTR(mov, pc, lr)
 
 ENTRY(outsl)
 		ioaddr	r0, r0
-		b	__arch_writesl
-
-		/* Nobody could say these are optimal, but not to worry. */
+		teq	r2, #0
+		bne	__arch_writesl
+		RETINSTR(mov, pc, lr)
 
 ENTRY(outsw)
 		ioaddr	r0, r0
-		b	__arch_writesw
+		teq	r2, #0
+		bne	__arch_writesw
+		RETINSTR(mov, pc, lr)
 
 ENTRY(insw)
 		ioaddr	r0, r0
-		b	__arch_readsw
+		teq	r2, #0
+		bne	__arch_readsw
+		RETINSTR(mov, pc, lr)
 
 ENTRY(insb)
 		ioaddr	r0, r0
-		b	__arch_readsb
+		teq	r2, #0
+		bne	__arch_readsb
+		RETINSTR(mov, pc, lr)
 
 ENTRY(outsb)
 		ioaddr	r0, r0
-		b	__arch_writesb
+		teq	r2, #0
+		bne	__arch_writesb
+		RETINSTR(mov, pc, lr)
+
diff --git a/arch/arm/lib/io-readsb.S b/arch/arm/lib/io-readsb.S
index 23cca8929..a31381fd5 100644
--- a/arch/arm/lib/io-readsb.S
+++ b/arch/arm/lib/io-readsb.S
@@ -62,6 +62,7 @@ ENTRY(__arch_readsb)
 		ldrb	ip, [r0]
 		orr	r6, r6, ip, lsl #24
 		stmia	r1!, {r3 - r6}
+
 		subs	r2, r2, #16
 		bpl	.insb_16_lp
 
@@ -88,7 +89,7 @@ ENTRY(__arch_readsb)
 		stmia	r1!, {r3, r4}
 
 .insb_no_8:	tst	r2, #4
-		bne	.insb_no_4
+		beq	.insb_no_4
 
 		ldrb	r3, [r0]
 		ldrb	r4, [r0]
@@ -101,6 +102,7 @@ ENTRY(__arch_readsb)
 
 .insb_no_4:	ands	r2, r2, #3
 		LOADREGS(eqfd, sp!, {r4 - r6, pc})
+
 		cmp	r2, #2
 		ldrb	r3, [r0]
 		strb	r3, [r1], #1
@@ -108,4 +110,5 @@ ENTRY(__arch_readsb)
 		strgeb	r3, [r1], #1
 		ldrgtb	r3, [r0]
 		strgtb	r3, [r1]
+
 		LOADREGS(fd, sp!, {r4 - r6, pc})
diff --git a/arch/arm/lib/io-readsl-armv3.S b/arch/arm/lib/io-readsl-armv3.S
new file mode 100644
index 000000000..0751c3157
--- /dev/null
+++ b/arch/arm/lib/io-readsl-armv3.S
@@ -0,0 +1,76 @@
+/*
+ *  linux/arch/arm/lib/io-readsl-armv3.S
+ *
+ *  Copyright (C) 1995-2000 Russell King
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+#include <asm/hardware.h>
+
+/*
+ * Note that some reads can be aligned on half-word boundaries.
+ */
+ENTRY(__arch_readsl)
+		ands	ip, r1, #3
+		bne	2f
+
+1:		ldr	r3, [r0]
+		str	r3, [r1], #4
+		subs	r2, r2, #1
+		bne	1b
+		mov	pc, lr
+
+2:		cmp	ip, #2
+		ldr	ip, [r0]
+		blt	4f
+		bgt	6f
+
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+3:		subs	r2, r2, #1
+		ldrne	r3, [r0]
+		orrne	ip, ip, r3, lsl #16
+		strne	ip, [r1], #4
+		movne	ip, r3, lsr #16
+		bne	3b
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	pc, lr
+
+4:		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+5:		subs	r2, r2, #1
+		ldrne	r3, [r0]
+		orrne	ip, ip, r3, lsl #8
+		strne	ip, [r1], #4
+		movne	ip, r3, lsr #24
+		bne	5b
+		strb	ip, [r1], #1
+		mov	pc, lr
+
+6:		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+7:		subs	r2, r2, #1
+		ldrne	r3, [r0]
+		orrne	ip, ip, r3, lsl #24
+		strne	ip, [r1], #4
+		movne	ip, r3, lsr #8
+		bne	7b
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	ip, ip, lsr #8
+		strb	ip, [r1], #1
+		mov	pc, lr
+
diff --git a/arch/arm/lib/io-readsl.S b/arch/arm/lib/io-readsl-armv4.S
index 530175485..08b3d350f 100644
--- a/arch/arm/lib/io-readsl.S
+++ b/arch/arm/lib/io-readsl-armv4.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-readsl-armv4.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -11,6 +11,9 @@
 #include <asm/assembler.h>
 #include <asm/hardware.h>
 
+/*
+ * Note that some reads can be aligned on half-word boundaries.
+ */
 ENTRY(__arch_readsl)
 		ands	ip, r1, #3
 		bne	2f
diff --git a/arch/arm/lib/io-readsw-armv3.S b/arch/arm/lib/io-readsw-armv3.S
index 65fb94e2f..cb2d5cee9 100644
--- a/arch/arm/lib/io-readsw-armv3.S
+++ b/arch/arm/lib/io-readsw-armv3.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-readsw-armv3.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -62,8 +62,10 @@ ENTRY(__arch_readsw)
 		orr	r6, r6, lr, lsl #16
 
 		stmia	r1!, {r3 - r6}
+
 		subs	r2, r2, #8
 		bpl	.insw_8_lp
+
 		tst	r2, #7
 		LOADREGS(eqfd, sp!, {r4, r5, r6, pc})
 
@@ -97,6 +99,7 @@ ENTRY(__arch_readsw)
 		strneb	r3, [r1], #1
 		movne	r3, r3, lsr #8
 		strneb	r3, [r1]
+
 		LOADREGS(fd, sp!, {r4, r5, r6, pc})
 
 
diff --git a/arch/arm/lib/io-readsw-armv4.S b/arch/arm/lib/io-readsw-armv4.S
index 6f1750a49..10eeee99f 100644
--- a/arch/arm/lib/io-readsw-armv4.S
+++ b/arch/arm/lib/io-readsw-armv4.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-readsw-armv4.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -46,16 +46,18 @@ ENTRY(__arch_readsw)
 		orr	r4, r4, r5, lsl #16
 
 		ldrh	r5, [r0]
-		ldrh	r6, [r0]
-		orr	r5, r5, r6, lsl #16
+		ldrh	ip, [r0]
+		orr	r5, r5, ip, lsl #16
 
 		ldrh	ip, [r0]
 		ldrh	lr, [r0]
 		orr	ip, ip, lr, lsl #16
 
 		stmia	r1!, {r3 - r5, ip}
+
 		subs	r2, r2, #8
 		bpl	.insw_8_lp
+
 		tst	r2, #7
 		LOADREGS(eqfd, sp!, {r4, r5, pc})
 
@@ -84,4 +86,5 @@ ENTRY(__arch_readsw)
 .no_insw_2:	tst	r2, #1
 		ldrneh	r3, [r0]
 		strneh	r3, [r1]
+
 		LOADREGS(fd, sp!, {r4, r5, pc})
diff --git a/arch/arm/lib/io-writesb.S b/arch/arm/lib/io-writesb.S
index b919fdaad..a9a8c4af9 100644
--- a/arch/arm/lib/io-writesb.S
+++ b/arch/arm/lib/io-writesb.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-writesb.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -13,7 +13,7 @@
 
 .outsb_align:	rsb	ip, ip, #4
 		cmp	ip, r2
-		mov	ip, r2
+		movgt	ip, r2
 		cmp	ip, #2
 		ldrb	r3, [r1], #1
 		strb	r3, [r0]
@@ -34,6 +34,7 @@ ENTRY(__arch_writesb)
 		bmi	.outsb_no_16
 
 .outsb_16_lp:	ldmia	r1!, {r3 - r6}
+
 		strb	r3, [r0]
 		mov	r3, r3, lsr #8
 		strb	r3, [r0]
@@ -65,6 +66,7 @@ ENTRY(__arch_writesb)
 		strb	r6, [r0]
 		mov	r6, r6, lsr #8
 		strb	r6, [r0]
+
 		subs	r2, r2, #16
 		bpl	.outsb_16_lp
 
@@ -74,7 +76,8 @@ ENTRY(__arch_writesb)
 .outsb_no_16:	tst	r2, #8
 		beq	.outsb_no_8
 
-		ldmia	r1, {r3, r4}
+		ldmia	r1!, {r3, r4}
+
 		strb	r3, [r0]
 		mov	r3, r3, lsr #8
 		strb	r3, [r0]
@@ -92,7 +95,7 @@ ENTRY(__arch_writesb)
 		strb	r4, [r0]
 
 .outsb_no_8:	tst	r2, #4
-		bne	.outsb_no_4
+		beq	.outsb_no_4
 
 		ldr	r3, [r1], #4
 		strb	r3, [r0]
@@ -105,6 +108,7 @@ ENTRY(__arch_writesb)
 
 .outsb_no_4:	ands	r2, r2, #3
 		LOADREGS(eqfd, sp!, {r4 - r6, pc})
+
 		cmp	r2, #2
 		ldrb	r3, [r1], #1
 		strb	r3, [r0]
@@ -112,4 +116,5 @@ ENTRY(__arch_writesb)
 		strgeb	r3, [r0]
 		ldrgtb	r3, [r1]
 		strgtb	r3, [r0]
+
 		LOADREGS(fd, sp!, {r4 - r6, pc})
diff --git a/arch/arm/lib/io-writesl.S b/arch/arm/lib/io-writesl.S
index ccda08c39..aa3340bdd 100644
--- a/arch/arm/lib/io-writesl.S
+++ b/arch/arm/lib/io-writesl.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-writesl.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
diff --git a/arch/arm/lib/io-writesw-armv3.S b/arch/arm/lib/io-writesw-armv3.S
index 29c94cc3d..fce7d5ef5 100644
--- a/arch/arm/lib/io-writesw-armv3.S
+++ b/arch/arm/lib/io-writesw-armv3.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-writesw-armv3.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -39,6 +39,7 @@ ENTRY(__arch_writesw)
 
 		subs	r2, r2, #8
 		bmi	.no_outsw_8
+
 .outsw_8_lp:	ldmia	r1!, {r3, r4, r5, r6}
 
 		mov	ip, r3, lsl #16
@@ -75,6 +76,7 @@ ENTRY(__arch_writesw)
 
 		subs	r2, r2, #8
 		bpl	.outsw_8_lp
+
 		tst	r2, #7
 		LOADREGS(eqfd, sp!, {r4, r5, r6, pc})
 
@@ -121,5 +123,3 @@ ENTRY(__arch_writesw)
 		strne	ip, [r0]
 
 		LOADREGS(fd, sp!, {r4, r5, r6, pc})
-
-
diff --git a/arch/arm/lib/io-writesw-armv4.S b/arch/arm/lib/io-writesw-armv4.S
index 0f96d1325..bac67a42e 100644
--- a/arch/arm/lib/io-writesw-armv4.S
+++ b/arch/arm/lib/io-writesw-armv4.S
@@ -1,5 +1,5 @@
 /*
- *  linux/arch/arm/lib/io-readsb.S
+ *  linux/arch/arm/lib/io-writesw-armv4.S
  *
  *  Copyright (C) 1995-2000 Russell King
  *
@@ -32,11 +32,11 @@ ENTRY(__arch_writesw)
 		tst	r1, #3
 		bne	.outsw_align
 
+		stmfd	sp!, {r4, r5, lr}
+
 		subs	r2, r2, #8
 		bmi	.no_outsw_8
 
-		stmfd	sp!, {r4, r5, lr}
-
 .outsw_8_lp:	ldmia	r1!, {r3, r4, r5, ip}
 
 		strh	r3, [r0]
@@ -54,26 +54,31 @@ ENTRY(__arch_writesw)
 
 		subs	r2, r2, #8
 		bpl	.outsw_8_lp
-		ldmfd	sp!, {r4, r5, lr}
+
 		tst	r2, #7
-		RETINSTR(moveq, pc, lr)
+		LOADREGS(eqfd, sp!, {r4, r5, pc})
 
 .no_outsw_8:	tst	r2, #4
-		ldmneia	r1!, {r3, ip}
-		strneh	r3, [r0]
-		movne	r3, r3, lsr #16
-		strneh	r3, [r0]
-		strneh	ip, [r0]
-		movne	ip, ip, lsr #16
-		strneh	ip, [r0]
-		tst	r2, #2
-		ldrne	r3, [r1], #4
-		strneh	r3, [r0]
-		movne	r3, r3, lsr #16
-		strneh	r3, [r0]
-		tst	r2, #1
+		beq	.no_outsw_4
+
+		ldmia	r1!, {r3, ip}
+		strh	r3, [r0]
+		mov	r3, r3, lsr #16
+		strh	r3, [r0]
+		strh	ip, [r0]
+		mov	ip, ip, lsr #16
+		strh	ip, [r0]
+
+.no_outsw_4:	tst	r2, #2
+		beq	.no_outsw_2
+
+		ldr	r3, [r1], #4
+		strh	r3, [r0]
+		mov	r3, r3, lsr #16
+		strh	r3, [r0]
+
+.no_outsw_2:	tst	r2, #1
 		ldrneh	r3, [r1]
 		strneh	r3, [r0]
-		RETINSTR(mov, pc, lr)
-
 
+		LOADREGS(fd, sp!, {r4, r5, pc})
diff --git a/arch/arm/mach-sa1100/Makefile b/arch/arm/mach-sa1100/Makefile
index 441608295..3a61edf39 100644
--- a/arch/arm/mach-sa1100/Makefile
+++ b/arch/arm/mach-sa1100/Makefile
@@ -20,15 +20,4 @@ export-objs		:= hw.o leds.o
 
 obj-$(CONFIG_LEDS)	+= leds.o
 
-# Files that are both resident and modular; remove from modular.
-
-obj-m		:= $(filter-out $(obj-y), $(obj-m))
-
-# Translate to Rules.make lists.
-
-O_OBJS		:= $(filter-out $(export-objs), $(obj-y))
-OX_OBJS 	:= $(filter	$(export-objs), $(obj-y))
-M_OBJS		:= $(sort $(filter-out $(export-objs), $(obj-m)))
-MX_OBJS 	:= $(sort $(filter     $(export-objs), $(obj-m)))
-
 include $(TOPDIR)/Rules.make
diff --git a/arch/arm/mach-sa1100/hw.c b/arch/arm/mach-sa1100/hw.c
index 1ccc61fe3..a28b28177 100644
--- a/arch/arm/mach-sa1100/hw.c
+++ b/arch/arm/mach-sa1100/hw.c
@@ -90,12 +90,12 @@ static int bitsy_egpio = EGPIO_BITSY_RS232_ON;
 void clr_bitsy_egpio(unsigned long x) 
 {
   bitsy_egpio &= ~x;
-  *(volatile int *)0xdc000000 = bitsy_egpio;
+  BITSY_EGPIO = bitsy_egpio;
 }
 void set_bitsy_egpio(unsigned long x) 
 {
   bitsy_egpio |= x;
-  *(volatile int *)0xdc000000 = bitsy_egpio;
+  BITSY_EGPIO = bitsy_egpio;
 }
 EXPORT_SYMBOL(clr_bitsy_egpio);
 EXPORT_SYMBOL(set_bitsy_egpio);
@@ -119,8 +119,8 @@ static void __init sa1111_init(void){
   /* First, set up the 3.6864MHz clock on GPIO 27 for the SA-1111:
    * (SA-1110 Developer's Manual, section 9.1.2.1)
    */
-  GAFR |= GPIO_GPIO27;
-  GPDR |= GPIO_GPIO27;
+  GAFR |= GPIO_32_768kHz;
+  GPDR |= GPIO_32_768kHz;
   TUCR = TUCR_3_6864MHz;
 
   /* Now, set up the PLL and RCLK in the SA-1111: */
@@ -133,32 +133,32 @@ static void __init sa1111_init(void){
    * using the SKPCR.
    */
 
-  {
-  /*
-   * SA1111 DMA bus master setup 
+  /* If the system is going to use the SA-1111 DMA engines, set up
+   * the memory bus request/grant pins. Also configure the shared
+   * memory controller on the SA-1111 (SA-1111 Developer's Manual,
+   * section 3.2.3) and power up the DMA bus clock:
    */
-	int cas;
-
-	/* SA1111 side */
-	switch ( (MDCNFG>>12) & 0x03 ) {
-	case 0x02:
-		cas = 0; break;
-	case 0x03:
-		cas = 1; break;
-	default:
-		cas = 1; break;
-	}
-	SMCR = 1		/* 1: memory is SDRAM */
-		| ( 1 << 1 )	/* 1:MBGNT is enable */
-		| ( ((MDCNFG >> 4) & 0x07) << 2 )	/* row address lines */
-		| ( cas << 5 );	/* CAS latency */
-
-	/* SA1110 side */
-	GPDR |= 1<<21;
-	GPDR &= ~(1<<22);
-	GAFR |= ( (1<<21) | (1<<22) );
-
-	TUCR |= (1<<10);
+  if(machine_is_assabet()){
+
+    GAFR |= (GPIO_MBGNT | GPIO_MBREQ);
+    GPDR |= GPIO_MBGNT;
+    GPDR &= ~GPIO_MBREQ;
+    TUCR |= TUCR_MR;
+
+    /* Assabet is populated by default with two Samsung KM416S8030T-G8
+     * 128Mb SDRAMs, which are organized as 12-bit (row addr) x 9-bit
+     * (column addr), according to the data sheet. Apparently, the 
+     * bank selects factor into the row address, as Angel sets up the
+     * SA-1110 to use 14x9 addresses. The SDRAM datasheet specifies
+     * that when running at 100-125MHz, the CAS latency for -8 parts
+     * is 3 cycles, which is consistent with Angel.
+     */
+
+    SMCR = (SMCR_DTIM | SMCR_MBGE | 
+	    FInsrt(FExtr(MDCNFG, MDCNFG_SA1110_DRAC0), SMCR_DRAC) |
+	    ((FExtr(MDCNFG, MDCNFG_SA1110_TDL0)==3) ? SMCR_CLAT : 0));
+
+    SKPCR |= SKPCR_DCLKEN;
   }
 }
 
diff --git a/arch/arm/mach-shark/Makefile b/arch/arm/mach-shark/Makefile
index 414d0c2e1..c6a4d4023 100644
--- a/arch/arm/mach-shark/Makefile
+++ b/arch/arm/mach-shark/Makefile
@@ -5,6 +5,8 @@
 # removes any old dependencies. DON'T put your own dependencies here
 # unless it's something special (ie not a .c file).
 
+USE_STANDARD_AS_RULE := true
+
 O_TARGET		:= shark.o
 
 # Object file lists.
@@ -18,15 +20,4 @@ export-objs		:=
 
 #obj-$(CONFIG_LEDS)	+= leds.o
 
-# Files that are both resident and modular; remove from modular.
-
-obj-m		:= $(filter-out $(obj-y), $(obj-m))
-
-# Translate to Rules.make lists.
-
-O_OBJS		:= $(filter-out $(export-objs), $(obj-y))
-OX_OBJS 	:= $(filter	$(export-objs), $(obj-y))
-M_OBJS		:= $(sort $(filter-out $(export-objs), $(obj-m)))
-MX_OBJS 	:= $(sort $(filter     $(export-objs), $(obj-m)))
-
 include $(TOPDIR)/Rules.make
diff --git a/arch/arm/mm/Makefile b/arch/arm/mm/Makefile
index fe3cf5eac..5cc3ff520 100644
--- a/arch/arm/mm/Makefile
+++ b/arch/arm/mm/Makefile
@@ -9,12 +9,11 @@
 
 USE_STANDARD_AS_RULE := true
 
-EXTRA_AFLAGS	:= -traditional
 O_TARGET	:= mm.o
 
 # Object file lists.
 
-obj-y		:= extable.o fault-$(PROCESSOR).o init.o \
+obj-y		:= extable.o fault-common.o fault-$(PROCESSOR).o init.o \
 		   mm-$(PROCESSOR).o small_page.o
 obj-m		:=
 obj-n		:=
@@ -22,10 +21,10 @@ obj-		:=
 export-objs	:= proc-syms.o
 
 p-$(CONFIG_CPU_26)	+= proc-arm2,3.o
-p-$(CONFIG_CPU_ARM6)	+= proc-arm6,7.o
-p-$(CONFIG_CPU_ARM7)	+= proc-arm6,7.o
-p-$(CONFIG_CPU_ARM720)	+= proc-arm720.o
-p-$(CONFIG_CPU_ARM920)	+= proc-arm920.o
+p-$(CONFIG_CPU_ARM610)	+= proc-arm6,7.o
+p-$(CONFIG_CPU_ARM710)	+= proc-arm6,7.o
+p-$(CONFIG_CPU_ARM720T)	+= proc-arm720.o
+p-$(CONFIG_CPU_ARM920T)	+= proc-arm920.o
 p-$(CONFIG_CPU_ARM10)	+= proc-arm10.o
 p-$(CONFIG_CPU_SA110)	+= proc-sa110.o
 p-$(CONFIG_CPU_SA1100)	+= proc-sa110.o
@@ -37,29 +36,16 @@ endif
 
 # Integrator follows "new style"
 # Soon, others will do too, and we can get rid of this
-MMMACH		:= mm-$(MACHINE).o
+MMMACH		:= mm-$(MACHINE).c
 ifeq ($(MMMACH),$(wildcard $(MMMACH)))
-obj-$(CONFIG_CPU_32)	+= $(MMMACH)
+obj-$(CONFIG_CPU_32)	+= $(MMMACH:.c=.o)
 endif
 
 obj-y		+= $(sort $(p-y))
 
-# Files that are both resident and modular; remove from modular.
-
-obj-m		:= $(filter-out $(obj-y), $(obj-m))
-
-# Translate to Rules.make lists.
-
-O_OBJS		:= $(filter-out $(export-objs), $(obj-y))
-OX_OBJS 	:= $(filter	$(export-objs), $(obj-y))
-M_OBJS		:= $(sort $(filter-out $(export-objs), $(obj-m)))
-MX_OBJS 	:= $(sort $(filter     $(export-objs), $(obj-m)))
-
 include $(TOPDIR)/Rules.make
 
 # Special dependencies
-fault-armv.o:   fault-common.c
-fault-armo.o:   fault-common.c
 proc-arm2,3.o:	../lib/constants.h
 proc-arm6,7.o:	../lib/constants.h
 proc-arm720.o:	../lib/constants.h
diff --git a/arch/arm/mm/fault-armo.c b/arch/arm/mm/fault-armo.c
index 85b2dc2f4..a37c625f4 100644
--- a/arch/arm/mm/fault-armo.c
+++ b/arch/arm/mm/fault-armo.c
@@ -23,14 +23,20 @@
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
 
-#define FAULT_CODE_FORCECOW	0x80
+#define FAULT_CODE_LDRSTRPOST	0x80
+#define FAULT_CODE_LDRSTRPRE	0x40
+#define FAULT_CODE_LDRSTRREG	0x20
+#define FAULT_CODE_LDMSTM	0x10
+#define FAULT_CODE_LDCSTC	0x08
 #define FAULT_CODE_PREFETCH	0x04
 #define FAULT_CODE_WRITE	0x02
+#define FAULT_CODE_FORCECOW	0x01
 
 #define DO_COW(m)		((m) & (FAULT_CODE_WRITE|FAULT_CODE_FORCECOW))
 #define READ_FAULT(m)		(!((m) & FAULT_CODE_WRITE))
 
-#include "fault-common.c"
+extern int do_page_fault(unsigned long addr, int mode, struct pt_regs *regs);
+extern void show_pte(struct mm_struct *mm, unsigned long addr);
 
 /*
  * Handle a data abort.  Note that we have to handle a range of addresses
diff --git a/arch/arm/mm/fault-armv.c b/arch/arm/mm/fault-armv.c
index f184b4e68..a6a45f091 100644
--- a/arch/arm/mm/fault-armv.c
+++ b/arch/arm/mm/fault-armv.c
@@ -27,27 +27,9 @@
 #include <asm/pgtable.h>
 #include <asm/unaligned.h>
 
-#define FAULT_CODE_READ		0x02
-
-#define DO_COW(m)		(!((m) & FAULT_CODE_READ))
-#define READ_FAULT(m)		((m) & FAULT_CODE_READ)
-
 extern void die_if_kernel(const char *str, struct pt_regs *regs, int err);
-
-#include "fault-common.c"
-
-#ifdef DEBUG
-static int sp_valid(unsigned long *sp)
-{
-	unsigned long addr = (unsigned long) sp;
-
-	if (addr >= 0xb0000000 && addr < 0xd0000000)
-		return 1;
-	if (addr >= 0x03ff0000 && addr < 0x04000000)
-		return 1;
-	return 0;
-}
-#endif
+extern void show_pte(struct mm_struct *mm, unsigned long addr);
+extern int do_page_fault(unsigned long addr, int mode, struct pt_regs *regs);
 
 #ifdef CONFIG_ALIGNMENT_TRAP
 /*
@@ -178,6 +160,15 @@ do_alignment(unsigned long addr, int error_code, struct pt_regs *regs)
 					eaddr -= offset.un;
 			}
 
+			/*
+			 * This is a "hint" - we already have eaddr worked out by the
+			 * processor for us.
+			 */
+			if (addr != eaddr)
+				printk(KERN_ERR "LDRHSTRH: PC = %08lx, instr = %08x, "
+					"addr = %08lx, eaddr = %08lx\n",
+					 instruction_pointer(regs), instr, addr, eaddr);
+
 			if (LDST_L_BIT(instr))
 				regs->uregs[rd] = get_unaligned((unsigned short *)eaddr);
 			else
@@ -253,8 +244,15 @@ do_alignment(unsigned long addr, int error_code, struct pt_regs *regs)
 			}
 		}
 
-if (addr != eaddr)
-printk("PC = %08lx, instr = %08x, addr = %08lx, eaddr = %08lx\n", instruction_pointer(regs), instr, addr, eaddr);
+		/*
+		 * This is a "hint" - we already have eaddr worked out by the
+		 * processor for us.
+		 */
+		if (addr != eaddr)
+			printk(KERN_ERR "LDRSTR: PC = %08lx, instr = %08x, "
+				"addr = %08lx, eaddr = %08lx\n",
+				 instruction_pointer(regs), instr, addr, eaddr);
+
 		if (LDST_L_BIT(instr)) {
 			regs->uregs[rd] = get_unaligned((unsigned long *)eaddr);
 			if (rd == 15)
@@ -284,6 +282,15 @@ printk("PC = %08lx, instr = %08x, addr = %08lx, eaddr = %08lx\n", instruction_po
 		if  (!LDST_U_BIT(instr))
 			eaddr -= nr_regs;
 
+		/*
+		 * This is a "hint" - we already have eaddr worked out by the
+		 * processor for us.
+		 */
+		if (addr != eaddr)
+			printk(KERN_ERR "LDMSTM: PC = %08lx, instr = %08x, "
+				"addr = %08lx, eaddr = %08lx\n",
+				 instruction_pointer(regs), instr, addr, eaddr);
+
 		if ((LDST_U_BIT(instr) == 0 && LDST_P_BIT(instr) == 0) ||
 		    (LDST_U_BIT(instr)      && LDST_P_BIT(instr)))
 			eaddr += 4;
@@ -322,39 +329,39 @@ printk("PC = %08lx, instr = %08x, addr = %08lx, eaddr = %08lx\n", instruction_po
 
 #endif
 
-#ifdef CONFIG_DEBUG_USER
-
+/*
+ * Some section permission faults need to be handled gracefully, for
+ * instance, when they happen due to a __{get,put}_user during an oops).
+ * In this case, we should return an error to the __{get,put}_user caller
+ * instead of causing another oops.  We should also fixup this fault as
+ * the user could pass a pointer to a section to the kernel.
+ */
 static int
 do_sect_fault(unsigned long addr, int error_code, struct pt_regs *regs)
 {
+	unsigned long fixup;
+
 	if (user_mode(regs)) {
+#ifdef CONFIG_DEBUG_USER
 		printk("%s: permission fault on section, "
 		       "address=0x%08lx, code %d\n",
 		       current->comm, addr, error_code);
+#endif
+		goto fail;
+	}
+
+	fixup = search_exception_table(instruction_pointer(regs));
+	if (fixup != 0) {
 #ifdef DEBUG
-		{
-			unsigned int i, j;
-			unsigned long *sp;
-
-			sp = (unsigned long *) (regs->ARM_sp - 128);
-			for (j = 0; j < 20 && sp_valid(sp); j++) {
-				printk("%p: ", sp);
-				for (i = 0; i < 8 && sp_valid(sp); i += 1, sp++)
-					printk("%08lx ", *sp);
-				printk("\n");
-			}
-			show_regs(regs);
-			c_backtrace(regs->ARM_fp, regs->ARM_cpsr);
-		}
+		printk(KERN_DEBUG "%s: Exception at [<%lx>] addr=%lx (fixup: %lx)\n",
+			tsk->comm, regs->ARM_pc, addr, fixup);
 #endif
+		regs->ARM_pc = fixup;
+		return 0;
 	}
+fail:
 	return 1;	/* not fixed up */
 }
-#else
-
-#define do_sect_fault NULL
-
-#endif
 
 static const struct fsr_info {
 	int	(*fn)(unsigned long addr, int error_code, struct pt_regs *regs);
@@ -382,18 +389,19 @@ static const struct fsr_info {
 /*
  * Currently dropped down to debug level
  */
-#define BUG_PROC_MSG \
-  KERN_DEBUG "Weird data abort (%08X).\n" \
-  KERN_DEBUG "Please see http://www.arm.linux.org.uk/state.html for " \
-	"more information\n"
-
 asmlinkage void
 do_DataAbort(unsigned long addr, int error_code, struct pt_regs *regs, int fsr)
 {
 	const struct fsr_info *inf = fsr_info + (fsr & 15);
 
+#if defined(CONFIG_CPU_SA110) || defined(CONFIG_CPU_SA1100)
 	if (addr == regs->ARM_pc)
-		goto weirdness;
+		goto sa1_weirdness;
+#endif
+#if defined(CONFIG_CPU_ARM720T) && defined(CONFIG_ALIGNMENT_TRAP)
+	if (addr & 3 && (fsr & 13) != 1)
+		goto arm720_weirdness;
+#endif
 
 	if (!inf->fn)
 		goto bad;
@@ -409,26 +417,51 @@ bad:
 	die_if_kernel("Oops", regs, 0);
 	return;
 
-weirdness:
+#if defined(CONFIG_CPU_SA110) || defined(CONFIG_CPU_SA1100)
+sa1_weirdness:
 	if (user_mode(regs)) {
 		static int first = 1;
 		if (first)
-			/*
-			 * I want statistical information on this problem,
-			 * but we don't want to hastle the users too much.
-			 */
-			printk(BUG_PROC_MSG, fsr);
+			printk(KERN_DEBUG "Weird data abort detected\n");
 		first = 0;
 		return;
 	}
 
 	if (!inf->fn || inf->fn(addr, error_code, regs))
 		goto bad;
+	return;
+#endif
+#if defined(CONFIG_CPU_ARM720T) && defined(CONFIG_ALIGNMENT_TRAP)
+arm720_weirdness:
+	if (!user_mode(regs)) {
+		unsigned long instr;
+
+		instr = *(unsigned long *)instruction_pointer(regs);
+
+		if ((instr & 0x04400000) != 0x04400000) {
+			static int first = 1;
+			if (first)
+				printk("Mis-reported alignment fault at "
+					"0x%08lx, fsr 0x%02x, code 0x%02x, "
+					"PC = 0x%08lx, instr = 0x%08lx\n",
+					addr, fsr, error_code, regs->ARM_pc,
+					instr);
+			first = 0;
+			cpu_tlb_invalidate_all();
+			cpu_cache_clean_invalidate_all();
+			return;
+		}
+	}
+
+	if (!inf->fn || inf->fn(addr, error_code, regs))
+		goto bad;
+	return;
+#endif
 }
 
 asmlinkage int
 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
 {
-	do_page_fault(addr, FAULT_CODE_READ, regs);
+	do_page_fault(addr, 0, regs);
 	return 1;
 }
diff --git a/arch/arm/mm/fault-common.c b/arch/arm/mm/fault-common.c
index 1cbdd7790..f79b618a2 100644
--- a/arch/arm/mm/fault-common.c
+++ b/arch/arm/mm/fault-common.c
@@ -9,8 +9,41 @@
  * published by the Free Software Foundation.
  */
 #include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/interrupt.h>
+#include <linux/proc_fs.h>
+#include <linux/init.h>
+
+#include <asm/system.h>
+#include <asm/uaccess.h>
+#include <asm/pgtable.h>
+#include <asm/unaligned.h>
+
+#ifdef CONFIG_CPU_26
+#define FAULT_CODE_WRITE	0x02
+#define FAULT_CODE_FORCECOW	0x01
+#define DO_COW(m)		((m) & (FAULT_CODE_WRITE|FAULT_CODE_FORCECOW))
+#define READ_FAULT(m)		(!((m) & FAULT_CODE_WRITE))
+#else
+/*
+ * On 32-bit processors, we define "mode" to be zero when reading,
+ * non-zero when writing.  This now ties up nicely with the polarity
+ * of the 26-bit machines, and also means that we avoid the horrible
+ * gcc code for "int val = !other_val;".
+ */
+#define DO_COW(m)		(m)
+#define READ_FAULT(m)		(!(m))
+#endif
 
-extern void die(const char *msg, struct pt_regs *regs, int err);
+NORET_TYPE void die(const char *msg, struct pt_regs *regs, int err) ATTRIB_NORET;
 
 /*
  * This is useful to dump out the page tables associated with
@@ -60,7 +93,9 @@ void show_pte(struct mm_struct *mm, unsigned long addr)
 	printk("\n");
 }
 
-static int __do_page_fault(struct mm_struct *mm, unsigned long addr, int mode, struct task_struct *tsk)
+static int
+__do_page_fault(struct mm_struct *mm, unsigned long addr, int mode,
+		struct task_struct *tsk)
 {
 	struct vm_area_struct *vma;
 	int fault, mask;
@@ -159,7 +194,7 @@ bad_area:
 	return -2;
 }
 
-static int do_page_fault(unsigned long addr, int mode, struct pt_regs *regs)
+int do_page_fault(unsigned long addr, int mode, struct pt_regs *regs)
 {
 	struct task_struct *tsk;
 	struct mm_struct *mm;
@@ -278,8 +313,10 @@ no_context:
 	 * Oops. The kernel tried to access some bad page. We'll have to
 	 * terminate things with extreme prejudice.
 	 */
-	printk(KERN_ALERT "Unable to handle kernel %s at virtual address %08lx\n",
-		(addr < PAGE_SIZE) ? "NULL pointer dereference" : "paging request", addr);
+	printk(KERN_ALERT
+		"Unable to handle kernel %s at virtual address %08lx\n",
+		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
+		"paging request", addr);
 
 	show_pte(mm, addr);
 	die("Oops", regs, mode);
diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c
index bfa085a52..6ceec5a22 100644
--- a/arch/arm/mm/init.c
+++ b/arch/arm/mm/init.c
@@ -49,7 +49,7 @@
 #define TABLE_SIZE	((TABLE_OFFSET + PTRS_PER_PTE) * sizeof(void *))
 
 static unsigned long totalram_pages;
-pgd_t swapper_pg_dir[PTRS_PER_PGD];
+extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
 extern char _stext, _text, _etext, _end, __init_begin, __init_end;
 
 /*
@@ -418,7 +418,7 @@ static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int boot
 	if (machine_is_archimedes() || machine_is_a5k())
 		reserve_bootmem_node(pgdat, 0x02000000, 0x00080000);
 	if (machine_is_p720t())
-		reserve_bootmem_node(pgdat, 0xc0000000, 0x00014000);
+		reserve_bootmem_node(pgdat, PAGE_OFFSET, 0x00014000);
 }
 
 /*
@@ -450,8 +450,28 @@ void __init bootmem_init(struct meminfo *mi)
 	initrd_node   = check_initrd(mi);
 
 	map_pg = bootmap_pfn;
-	
-	for (node = 0; node < numnodes; node++, np++) {
+
+	/*
+	 * Initialise the bootmem nodes.
+	 *
+	 * What we really want to do is:
+	 *
+	 *   unmap_all_regions_except_kernel();
+	 *   for_each_node_in_reverse_order(node) {
+	 *     map_node(node);
+	 *     allocate_bootmem_map(node);
+	 *     init_bootmem_node(node);
+	 *     free_bootmem_node(node);
+	 *   }
+	 *
+	 * but this is a 2.5-type change.  For now, we just set
+	 * the nodes up in reverse order.
+	 *
+	 * (we could also do with rolling bootmem_init and paging_init
+	 * into one generic "memory_init" type function).
+	 */
+	np += numnodes - 1;
+	for (node = numnodes - 1; node >= 0; node--, np--) {
 		/*
 		 * If there are no pages in this node, ignore it.
 		 * Note that node 0 must always have some pages.
diff --git a/arch/arm/mm/mm-armv.c b/arch/arm/mm/mm-armv.c
index d96fcbccb..a2fc69ea4 100644
--- a/arch/arm/mm/mm-armv.c
+++ b/arch/arm/mm/mm-armv.c
@@ -275,7 +275,7 @@ static void __init create_mapping(struct map_desc *md)
 	off    = md->physical - virt;
 	length = md->length;
 
-	while ((virt & 1048575 || (virt + off) & 1048575) && length >= PAGE_SIZE) {
+	while ((virt & 0xfffff || (virt + off) & 0xfffff) && length >= PAGE_SIZE) {
 		alloc_init_page(virt, virt + off, md->domain, prot_pte);
 
 		virt   += PAGE_SIZE;
diff --git a/arch/arm/mm/proc-arm6,7.S b/arch/arm/mm/proc-arm6,7.S
index 61e135e19..d26a98f75 100644
--- a/arch/arm/mm/proc-arm6,7.S
+++ b/arch/arm/mm/proc-arm6,7.S
@@ -7,10 +7,8 @@
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
  *
- *  MMU functions for ARM6
- *
  *  These are the low level assembler for performing cache and TLB
- *  functions on the ARM6 & ARM7.
+ *  functions on the ARM610 & ARM710.
  */
 #include <linux/linkage.h>
 #include <asm/assembler.h>
@@ -105,9 +103,9 @@ ENTRY(cpu_arm7_tlb_invalidate_page)
 
 ENTRY(cpu_arm6_data_abort)
 		ldr	r4, [r0]			@ read instruction causing problem
-		mov	r1, r4, lsr #19			@ r1 b1 = L
+		tst	r4, r4, lsr #21			@ C = bit 20
+		sbc	r1, r1, r1			@ r1 = C - 1
 		and	r2, r4, #14 << 24
-		and	r1, r1, #2			@ check read/write bit
 		teq	r2, #8 << 24			@ was it ldm/stm
 		bne	Ldata_simple
 
@@ -138,9 +136,9 @@ Ldata_simple:	mrc	p15, 0, r0, c6, c0, 0		@ get FAR
 
 ENTRY(cpu_arm7_data_abort)
 		ldr	r4, [r0]			@ read instruction causing problem
-		mov	r1, r4, lsr #19			@ r1 b1 = L
+		tst	r4, r4, lsr #21			@ C = bit 20
+		sbc	r1, r1, r1			@ r1 = C - 1
 		and	r2, r4, #15 << 24
-		and	r1, r1, #2			@ check read/write bit
 		add	pc, pc, r2, lsr #22		@ Now branch to the relevent processing routine
 		movs	pc, lr
 
diff --git a/arch/arm/mm/proc-arm720.S b/arch/arm/mm/proc-arm720.S
index e6fc86bd1..77b689bdf 100644
--- a/arch/arm/mm/proc-arm720.S
+++ b/arch/arm/mm/proc-arm720.S
@@ -144,9 +144,9 @@ Ldata_simple:	mrc	p15, 0, r0, c6, c0, 0		@ get FAR
 
 ENTRY(cpu_arm720_data_abort)
 		ldr	r4, [r0]			@ read instruction causing problem
-		mov	r1, r4, lsr #19			@ r1 b1 = L
+		tst	r4, r4, lsr #21			@ C = bit 20
+		sbc	r1, r1, r1			@ r1 = C - 1
 		and	r2, r4, #15 << 24
-		and	r1, r1, #2			@ check read/write bit
 		add	pc, pc, r2, lsr #22		@ Now branch to the relevent processing routine
 		movs	pc, lr
 
@@ -283,23 +283,11 @@ ENTRY(cpu_arm720_proc_fin)
 		ldmfd	sp!, {pc}
 
 /*
- * Function: arm720_proc_do_idle (void)
- *
- * Params  : r0 = call type:
- *           0 = slow idle
- *           1 = fast idle
- *           2 = switch to slow processor clock
- *           3 = switch to fast processor clock
- *
+ * Function: arm720_proc_do_idle(void)
+ * Params  : r0 = unused
  * Purpose : put the processer in proper idle mode
  */
 ENTRY(cpu_arm720_do_idle)
-#if 0 /* FIXME: is this part of the processor? */
-		ldr	r2, =IO_BASE			@ Virt addr of IO
-		add	r2, r2, #0x00050000		@ Start of PMU regs
-		mov	r1, #0x01			@ Idle mode
-		str	r1, [r2, #4]
-#endif
 		mov	pc, lr
 
 /*
diff --git a/arch/arm/mm/proc-arm920.S b/arch/arm/mm/proc-arm920.S
index a4306f08a..6f1f11601 100644
--- a/arch/arm/mm/proc-arm920.S
+++ b/arch/arm/mm/proc-arm920.S
@@ -68,9 +68,9 @@
 ENTRY(cpu_arm920_data_abort)
 	ldr	r1, [r0]			@ read aborted instruction
 	mrc	p15, 0, r0, c6, c0, 0		@ get FAR
-	mov	r1, r1, lsr #19			@ b1 = L
+	tst	r1, r1, lsr #21			@ C = bit 20
 	mrc	p15, 0, r3, c5, c0, 0		@ get FSR
-	and	r1, r1, #2
+	sbc	r1, r1, r1			@ r1 = C - 1
 	and	r3, r3, #255
 	mov	pc, lr
 
@@ -330,7 +330,7 @@ ENTRY(cpu_arm920_icache_invalidate_range)
 	mov	r0, #0
 	mcr	p15, 0, r0, c7, c10, 4		@ drain WB
 ENTRY(cpu_arm920_icache_invalidate_page)
-	/* why no invalidate I cache --rmk */
+	mcr	p15, 0, r0, c7, c5, 0		@ invalidate I cache
 	mov	pc, lr
 
 
diff --git a/arch/arm/mm/proc-sa110.S b/arch/arm/mm/proc-sa110.S
index 031ba648a..45b0a31fe 100644
--- a/arch/arm/mm/proc-sa110.S
+++ b/arch/arm/mm/proc-sa110.S
@@ -83,9 +83,9 @@ ENTRY(cpu_sa110_data_abort)
 ENTRY(cpu_sa1100_data_abort)
 	ldr	r1, [r0]			@ read aborted instruction
 	mrc	p15, 0, r0, c6, c0, 0		@ get FAR
-	mov	r1, r1, lsr #19			@ b1 = L
+	tst	r1, r1, lsr #21			@ C = bit 20
 	mrc	p15, 0, r3, c5, c0, 0		@ get FSR
-	and	r1, r1, #2
+	sbc	r1, r1, r1			@ r1 = C - 1
 	and	r3, r3, #255
 	mov	pc, lr
 
diff --git a/arch/arm/nwfpe/Makefile b/arch/arm/nwfpe/Makefile
index a7f69d405..16c7c0aed 100644
--- a/arch/arm/nwfpe/Makefile
+++ b/arch/arm/nwfpe/Makefile
@@ -6,27 +6,27 @@
 
 USE_STANDARD_AS_RULE := true
 
-NWFPE_OBJS := fpa11.o fpa11_cpdo.o fpa11_cpdt.o fpa11_cprt.o \
-	      fpmodule.o fpopcode.o softfloat.o \
-	      single_cpdo.o double_cpdo.o extended_cpdo.o
+O_TARGET		:= math-emu.o
 
-ifeq ($(CONFIG_CPU_26),y)
-NWFPE_OBJS += entry26.o
-else
-NWFPE_OBJS += entry.o
-endif
+obj-y			:=
+obj-m			:=
+obj-n			:=
+
+list-multi		:= nwfpe.o
 
-ifeq ($(CONFIG_NWFPE),y)
-O_TARGET := math-emu.o
-O_OBJS = $(NWFPE_OBJS)
+obj-$(CONFIG_NWFPE)	+= nwfpe.o
+
+nwfpe-objs		:= fpa11.o fpa11_cpdo.o fpa11_cpdt.o fpa11_cprt.o \
+			   fpmodule.o fpopcode.o softfloat.o \
+			   single_cpdo.o double_cpdo.o extended_cpdo.o
+
+ifeq ($(CONFIG_CPU_26),y)
+nwfpe-objs		+= entry26.o
 else
-  ifeq ($(CONFIG_NWFPE),m)
-    M_OBJS = nwfpe.o
-    MI_OBJS = $(NWFPE_OBJS)
-  endif
+nwfpe-objs		+= entry.o
 endif
 
 include $(TOPDIR)/Rules.make
 
-nwfpe.o: $(MI_OBJS) $(MIX_OBJS)
-	 $(LD) $(LD_RFLAG) -r -o $@ $(MI_OBJS) $(MIX_OBJS)
+nwfpe.o: $(nwfpe-objs)
+	$(LD) -r -o $@ $(nwfpe-objs)
diff --git a/arch/arm/vmlinux-armo.lds.in b/arch/arm/vmlinux-armo.lds.in
index 0c0e720aa..f42deb72c 100644
--- a/arch/arm/vmlinux-armo.lds.in
+++ b/arch/arm/vmlinux-armo.lds.in
@@ -47,6 +47,8 @@ SECTIONS
 			*(.gnu.warning)
 			*(.text.lock)	/* out-of-line lock text */
 			*(.rodata)
+			*(.glue_7)
+			*(.glue_7t)
 			*(.kstrtab)
 		. = ALIGN(16);		/* Exception table		*/
 		__start___ex_table = .;
diff --git a/arch/arm/vmlinux-armv.lds.in b/arch/arm/vmlinux-armv.lds.in
index b0398852d..10661f414 100644
--- a/arch/arm/vmlinux-armv.lds.in
+++ b/arch/arm/vmlinux-armv.lds.in
@@ -42,6 +42,8 @@ SECTIONS
 			*(.gnu.warning)
 			*(.text.lock)	/* out-of-line lock text */
 			*(.rodata)
+			*(.glue_7)
+			*(.glue_7t)
 			*(.kstrtab)
 		. = ALIGN(16);
 		__start___ex_table = .;	/* Exception table		*/