Merge branch 'for-linus' of git://git.monstr.eu/linux-2.6-microblaze

* 'for-linus' of git://git.monstr.eu/linux-2.6-microblaze: (49 commits)
  microblaze: Add KGDB support
  microblaze: Support brki rX, 0x18 for user application debugging
  microblaze: Remove nop after MSRCLR/SET, MTS, MFS instructions
  microblaze: Simplify syscall rutine
  microblaze: Move PT_MODE saving to delay slot
  microblaze: Fix _interrupt function
  microblaze: Fix _user_exception function
  microblaze: Put together addik instructions
  microblaze: Use delay slot in syscall macros
  microblaze: Save kernel mode in delay slot
  microblaze: Do not mix register saving and mode setting
  microblaze: Move SAVE_STATE upward
  microblaze: entry.S: Macro optimization
  microblaze: Optimize hw exception rutine
  microblaze: Implement clear_ums macro and fix SAVE_STATE macro
  microblaze: Remove additional setup for kernel_mode
  microblaze: Optimize SAVE_STATE macro
  microblaze: Remove additional loading
  microblaze: Completely remove working with R11 register
  microblaze: Do not setup BIP in _debug_exception
  ...

33 files changed, 974 insertions, 541 deletions

diff --git a/arch/microblaze/Kconfig b/arch/microblaze/Kconfig
index 505a0859242..be3855250db 100644
--- a/arch/microblaze/Kconfig
+++ b/arch/microblaze/Kconfig
@@ -14,6 +14,7 @@ config MICROBLAZE
 	select USB_ARCH_HAS_EHCI
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select HAVE_OPROFILE
+	select HAVE_ARCH_KGDB
 	select HAVE_DMA_ATTRS
 	select HAVE_DMA_API_DEBUG
 	select TRACING_SUPPORT
@@ -223,6 +224,36 @@ config TASK_SIZE
 	hex "Size of user task space" if TASK_SIZE_BOOL
 	default "0x80000000"
 
+choice
+	prompt "Page size"
+	default MICROBLAZE_4K_PAGES
+	depends on ADVANCED_OPTIONS && !MMU
+	help
+	  Select the kernel logical page size. Increasing the page size
+	  will reduce software overhead at each page boundary, allow
+	  hardware prefetch mechanisms to be more effective, and allow
+	  larger dma transfers increasing IO efficiency and reducing
+	  overhead. However the utilization of memory will increase.
+	  For example, each cached file will using a multiple of the
+	  page size to hold its contents and the difference between the
+	  end of file and the end of page is wasted.
+
+	  If unsure, choose 4K_PAGES.
+
+config MICROBLAZE_4K_PAGES
+	bool "4k page size"
+
+config MICROBLAZE_8K_PAGES
+	bool "8k page size"
+
+config MICROBLAZE_16K_PAGES
+	bool "16k page size"
+
+config MICROBLAZE_32K_PAGES
+	bool "32k page size"
+
+endchoice
+
 endmenu
 
 source "mm/Kconfig"
diff --git a/arch/microblaze/Kconfig.debug b/arch/microblaze/Kconfig.debug
index 9dc708a7f70..e6e5e0da28c 100644
--- a/arch/microblaze/Kconfig.debug
+++ b/arch/microblaze/Kconfig.debug
@@ -10,6 +10,7 @@ source "lib/Kconfig.debug"
 
 config EARLY_PRINTK
 	bool "Early printk function for kernel"
+	depends on SERIAL_UARTLITE_CONSOLE
 	default n
 	help
 	  This option turns on/off early printk messages to console.
diff --git a/arch/microblaze/boot/Makefile b/arch/microblaze/boot/Makefile
index 57f50c2371c..be01d78750d 100644
--- a/arch/microblaze/boot/Makefile
+++ b/arch/microblaze/boot/Makefile
@@ -35,13 +35,14 @@ quiet_cmd_cp = CP      $< $@$2
 	cmd_cp = cat $< >$@$2 || (rm -f $@ && echo false)
 
 quiet_cmd_strip = STRIP   $@
-      cmd_strip = $(STRIP) -K _start -K _end -K __log_buf -K _fdt_start vmlinux -o $@
+	cmd_strip = $(STRIP) -K microblaze_start -K _end -K __log_buf \
+				-K _fdt_start vmlinux -o $@
 
 quiet_cmd_uimage = UIMAGE  $@.ub
-      cmd_uimage = $(CONFIG_SHELL) $(MKIMAGE) -A microblaze -O linux -T kernel \
-                   -C none -n 'Linux-$(KERNELRELEASE)' \
-                   -a $(CONFIG_KERNEL_BASE_ADDR) -e $(CONFIG_KERNEL_BASE_ADDR) \
-                   -d $@ $@.ub
+	cmd_uimage = $(CONFIG_SHELL) $(MKIMAGE) -A microblaze -O linux -T kernel \
+		-C none -n 'Linux-$(KERNELRELEASE)' \
+		-a $(CONFIG_KERNEL_BASE_ADDR) -e $(CONFIG_KERNEL_BASE_ADDR) \
+		-d $@ $@.ub
 
 $(obj)/simpleImage.%: vmlinux FORCE
 	$(call if_changed,cp,.unstrip)
diff --git a/arch/microblaze/include/asm/cacheflush.h b/arch/microblaze/include/asm/cacheflush.h
index a6edd356cd0..7ebd955460d 100644
--- a/arch/microblaze/include/asm/cacheflush.h
+++ b/arch/microblaze/include/asm/cacheflush.h
@@ -17,6 +17,7 @@
 
 /* Somebody depends on this; sigh... */
 #include <linux/mm.h>
+#include <linux/io.h>
 
 /* Look at Documentation/cachetlb.txt */
 
@@ -60,7 +61,6 @@ void microblaze_cache_init(void);
 #define invalidate_icache()				mbc->iin();
 #define invalidate_icache_range(start, end)		mbc->iinr(start, end);
 
-
 #define flush_icache_user_range(vma, pg, adr, len)	flush_icache();
 #define flush_icache_page(vma, pg)			do { } while (0)
 
@@ -72,9 +72,15 @@ void microblaze_cache_init(void);
 #define flush_dcache()					mbc->dfl();
 #define flush_dcache_range(start, end)			mbc->dflr(start, end);
 
-#define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 0
-/* D-cache aliasing problem can't happen - cache is between MMU and ram */
-#define flush_dcache_page(page)			do { } while (0)
+#define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1
+/* MS: We have to implement it because of rootfs-jffs2 issue on WB */
+#define flush_dcache_page(page) \
+do { \
+	unsigned long addr = (unsigned long) page_address(page); /* virtual */ \
+	addr = (u32)virt_to_phys((void *)addr); \
+	flush_dcache_range((unsigned) (addr), (unsigned) (addr) + PAGE_SIZE); \
+} while (0);
+
 #define flush_dcache_mmap_lock(mapping)		do { } while (0)
 #define flush_dcache_mmap_unlock(mapping)	do { } while (0)
 
@@ -97,8 +103,10 @@ void microblaze_cache_init(void);
 
 #define copy_to_user_page(vma, page, vaddr, dst, src, len)		\
 do {									\
+	u32 addr = virt_to_phys(dst);					\
+	invalidate_icache_range((unsigned) (addr), (unsigned) (addr) + (len));\
 	memcpy((dst), (src), (len));					\
-	flush_icache_range((unsigned) (dst), (unsigned) (dst) + (len));	\
+	flush_dcache_range((unsigned) (addr), (unsigned) (addr) + (len));\
 } while (0)
 
 #define copy_from_user_page(vma, page, vaddr, dst, src, len)		\
diff --git a/arch/microblaze/include/asm/dma-mapping.h b/arch/microblaze/include/asm/dma-mapping.h
index 18b3731c850..50738958070 100644
--- a/arch/microblaze/include/asm/dma-mapping.h
+++ b/arch/microblaze/include/asm/dma-mapping.h
@@ -79,12 +79,6 @@ static inline int dma_supported(struct device *dev, u64 mask)
 	return ops->dma_supported(dev, mask);
 }
 
-#ifdef CONFIG_PCI
-/* We have our own implementation of pci_set_dma_mask() */
-#define HAVE_ARCH_PCI_SET_DMA_MASK
-
-#endif
-
 static inline int dma_set_mask(struct device *dev, u64 dma_mask)
 {
 	struct dma_map_ops *ops = get_dma_ops(dev);
diff --git a/arch/microblaze/include/asm/elf.h b/arch/microblaze/include/asm/elf.h
index 7d4acf2b278..732caf1be74 100644
--- a/arch/microblaze/include/asm/elf.h
+++ b/arch/microblaze/include/asm/elf.h
@@ -77,7 +77,7 @@ typedef elf_fpreg_t elf_fpregset_t[ELF_NFPREG];
 #define ELF_DATA	ELFDATA2MSB
 #endif
 
-#define ELF_EXEC_PAGESIZE	4096
+#define ELF_EXEC_PAGESIZE	PAGE_SIZE
 
 
 #define ELF_CORE_COPY_REGS(_dest, _regs)			\
diff --git a/arch/microblaze/include/asm/exceptions.h b/arch/microblaze/include/asm/exceptions.h
index 4c7b5d037c8..6479097b802 100644
--- a/arch/microblaze/include/asm/exceptions.h
+++ b/arch/microblaze/include/asm/exceptions.h
@@ -14,6 +14,11 @@
 #define _ASM_MICROBLAZE_EXCEPTIONS_H
 
 #ifdef __KERNEL__
+
+#ifndef CONFIG_MMU
+#define EX_HANDLER_STACK_SIZ	(4*19)
+#endif
+
 #ifndef __ASSEMBLY__
 
 /* Macros to enable and disable HW exceptions in the MSR */
@@ -64,22 +69,6 @@ asmlinkage void full_exception(struct pt_regs *regs, unsigned int type,
 void die(const char *str, struct pt_regs *fp, long err);
 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr);
 
-#if defined(CONFIG_KGDB)
-void (*debugger)(struct pt_regs *regs);
-int (*debugger_bpt)(struct pt_regs *regs);
-int (*debugger_sstep)(struct pt_regs *regs);
-int (*debugger_iabr_match)(struct pt_regs *regs);
-int (*debugger_dabr_match)(struct pt_regs *regs);
-void (*debugger_fault_handler)(struct pt_regs *regs);
-#else
-#define debugger(regs)			do { } while (0)
-#define debugger_bpt(regs)		0
-#define debugger_sstep(regs)		0
-#define debugger_iabr_match(regs)	0
-#define debugger_dabr_match(regs)	0
-#define debugger_fault_handler		((void (*)(struct pt_regs *))0)
-#endif
-
 #endif /*__ASSEMBLY__ */
 #endif /* __KERNEL__ */
 #endif /* _ASM_MICROBLAZE_EXCEPTIONS_H */
diff --git a/arch/microblaze/include/asm/kgdb.h b/arch/microblaze/include/asm/kgdb.h
new file mode 100644
index 00000000000..78b17d40b23
--- /dev/null
+++ b/arch/microblaze/include/asm/kgdb.h
@@ -0,0 +1,28 @@
+#ifdef __KERNEL__
+#ifndef __MICROBLAZE_KGDB_H__
+#define __MICROBLAZE_KGDB_H__
+
+#ifndef __ASSEMBLY__
+
+#define CACHE_FLUSH_IS_SAFE	1
+#define BUFMAX			2048
+
+/*
+ * 32 32-bit general purpose registers (r0-r31)
+ *  6 32-bit special registers (pc, msr, ear, esr, fsr, btr)
+ * 12 32-bit PVR
+ *   7 32-bit MMU Regs (redr, rpid, rzpr, rtlbx, rtlbsx, rtlblo, rtlbhi)
+ * ------
+ *  57 registers
+ */
+#define NUMREGBYTES	(57 * 4)
+
+#define BREAK_INSTR_SIZE	4
+static inline void arch_kgdb_breakpoint(void)
+{
+	__asm__ __volatile__("brki r16, 0x18;");
+}
+
+#endif /* __ASSEMBLY__ */
+#endif /* __MICROBLAZE_KGDB_H__ */
+#endif /* __KERNEL__ */
diff --git a/arch/microblaze/include/asm/page.h b/arch/microblaze/include/asm/page.h
index 464ff32bee3..c12c6dfafd9 100644
--- a/arch/microblaze/include/asm/page.h
+++ b/arch/microblaze/include/asm/page.h
@@ -23,8 +23,16 @@
 #ifdef __KERNEL__
 
 /* PAGE_SHIFT determines the page size */
-#define PAGE_SHIFT	(12)
-#define PAGE_SIZE	(_AC(1, UL) << PAGE_SHIFT)
+#if defined(CONFIG_MICROBLAZE_32K_PAGES)
+#define PAGE_SHIFT		15
+#elif defined(CONFIG_MICROBLAZE_16K_PAGES)
+#define PAGE_SHIFT		14
+#elif defined(CONFIG_MICROBLAZE_8K_PAGES)
+#define PAGE_SHIFT		13
+#else
+#define PAGE_SHIFT		12
+#endif
+#define PAGE_SIZE	(ASM_CONST(1) << PAGE_SHIFT)
 #define PAGE_MASK	(~(PAGE_SIZE-1))
 
 #define LOAD_OFFSET	ASM_CONST((CONFIG_KERNEL_START-CONFIG_KERNEL_BASE_ADDR))
diff --git a/arch/microblaze/include/asm/pvr.h b/arch/microblaze/include/asm/pvr.h
index e38abc7714b..9578666e98b 100644
--- a/arch/microblaze/include/asm/pvr.h
+++ b/arch/microblaze/include/asm/pvr.h
@@ -16,7 +16,7 @@
 #define PVR_MSR_BIT 0x400
 
 struct pvr_s {
-	unsigned pvr[16];
+	unsigned pvr[12];
 };
 
 /* The following taken from Xilinx's standalone BSP pvr.h */
diff --git a/arch/microblaze/include/asm/setup.h b/arch/microblaze/include/asm/setup.h
index 7f31394985e..782b5c89248 100644
--- a/arch/microblaze/include/asm/setup.h
+++ b/arch/microblaze/include/asm/setup.h
@@ -28,8 +28,6 @@ void disable_early_printk(void);
 void heartbeat(void);
 void setup_heartbeat(void);
 
-unsigned long long sched_clock(void);
-
 #   ifdef CONFIG_MMU
 extern void mmu_reset(void);
 extern void early_console_reg_tlb_alloc(unsigned int addr);
diff --git a/arch/microblaze/include/asm/system.h b/arch/microblaze/include/asm/system.h
index 81e1f7d5b4c..e6a2284571d 100644
--- a/arch/microblaze/include/asm/system.h
+++ b/arch/microblaze/include/asm/system.h
@@ -45,7 +45,6 @@ extern struct task_struct *_switch_to(struct thread_info *prev,
 #define smp_rmb()		rmb()
 #define smp_wmb()		wmb()
 
-void show_trace(struct task_struct *task, unsigned long *stack);
 void __bad_xchg(volatile void *ptr, int size);
 
 static inline unsigned long __xchg(unsigned long x, volatile void *ptr,
diff --git a/arch/microblaze/include/asm/uaccess.h b/arch/microblaze/include/asm/uaccess.h
index 26460d15b33..d840f4a2d3c 100644
--- a/arch/microblaze/include/asm/uaccess.h
+++ b/arch/microblaze/include/asm/uaccess.h
@@ -359,7 +359,7 @@ extern long __user_bad(void);
 	__copy_tofrom_user((__force void __user *)(to), \
 				(void __user *)(from), (n))
 #define __copy_from_user_inatomic(to, from, n) \
-		copy_from_user((to), (from), (n))
+		__copy_from_user((to), (from), (n))
 
 static inline long copy_from_user(void *to,
 		const void __user *from, unsigned long n)
@@ -373,7 +373,7 @@ static inline long copy_from_user(void *to,
 #define __copy_to_user(to, from, n)	\
 		__copy_tofrom_user((void __user *)(to), \
 			(__force const void __user *)(from), (n))
-#define __copy_to_user_inatomic(to, from, n)	copy_to_user((to), (from), (n))
+#define __copy_to_user_inatomic(to, from, n) __copy_to_user((to), (from), (n))
 
 static inline long copy_to_user(void __user *to,
 		const void *from, unsigned long n)
diff --git a/arch/microblaze/include/asm/unwind.h b/arch/microblaze/include/asm/unwind.h
new file mode 100644
index 00000000000..d248b7de4b1
--- /dev/null
+++ b/arch/microblaze/include/asm/unwind.h
@@ -0,0 +1,29 @@
+/*
+ * Backtrace support for Microblaze
+ *
+ * Copyright (C) 2010  Digital Design Corporation
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+
+#ifndef __MICROBLAZE_UNWIND_H
+#define __MICROBLAZE_UNWIND_H
+
+struct stack_trace;
+
+struct trap_handler_info {
+	unsigned long start_addr;
+	unsigned long end_addr;
+	const char    *trap_name;
+};
+extern struct trap_handler_info microblaze_trap_handlers;
+
+extern const char _hw_exception_handler;
+extern const char ex_handler_unhandled;
+
+void microblaze_unwind(struct task_struct *task, struct stack_trace *trace);
+
+#endif	/* __MICROBLAZE_UNWIND_H */
+
diff --git a/arch/microblaze/kernel/Makefile b/arch/microblaze/kernel/Makefile
index e51bc152082..5eecc9f1fbd 100644
--- a/arch/microblaze/kernel/Makefile
+++ b/arch/microblaze/kernel/Makefile
@@ -17,7 +17,7 @@ extra-y := head.o vmlinux.lds
 obj-y += dma.o exceptions.o \
 	hw_exception_handler.o init_task.o intc.o irq.o of_device.o \
 	of_platform.o process.o prom.o prom_parse.o ptrace.o \
-	setup.o signal.o sys_microblaze.o timer.o traps.o reset.o
+	reset.o setup.o signal.o sys_microblaze.o timer.o traps.o unwind.o
 
 obj-y += cpu/
 
@@ -28,5 +28,6 @@ obj-$(CONFIG_MODULES)		+= microblaze_ksyms.o module.o
 obj-$(CONFIG_MMU)		+= misc.o
 obj-$(CONFIG_STACKTRACE)	+= stacktrace.o
 obj-$(CONFIG_FUNCTION_TRACER)	+= ftrace.o mcount.o
+obj-$(CONFIG_KGDB)		+= kgdb.o
 
 obj-y	+= entry$(MMU).o
diff --git a/arch/microblaze/kernel/cpu/mb.c b/arch/microblaze/kernel/cpu/mb.c
index 4216eb1eaa3..7086e356428 100644
--- a/arch/microblaze/kernel/cpu/mb.c
+++ b/arch/microblaze/kernel/cpu/mb.c
@@ -126,6 +126,7 @@ static int show_cpuinfo(struct seq_file *m, void *v)
 			cpuinfo.pvr_user1,
 			cpuinfo.pvr_user2);
 
+	count += seq_printf(m, "Page size:\t%lu\n", PAGE_SIZE);
 	return 0;
 }
 
diff --git a/arch/microblaze/kernel/entry-nommu.S b/arch/microblaze/kernel/entry-nommu.S
index 8cc18cd2cce..ca84368570b 100644
--- a/arch/microblaze/kernel/entry-nommu.S
+++ b/arch/microblaze/kernel/entry-nommu.S
@@ -588,3 +588,31 @@ sys_rt_sigsuspend_wrapper:
 #include "syscall_table.S"
 
 syscall_table_size=(.-sys_call_table)
+
+type_SYSCALL:
+	.ascii "SYSCALL\0"
+type_IRQ:
+	.ascii "IRQ\0"
+type_IRQ_PREEMPT:
+	.ascii "IRQ (PREEMPTED)\0"
+type_SYSCALL_PREEMPT:
+	.ascii " SYSCALL (PREEMPTED)\0"
+
+	/*
+	 * Trap decoding for stack unwinder
+	 * Tuples are (start addr, end addr, string)
+	 * If return address lies on [start addr, end addr],
+	 * unwinder displays 'string'
+	 */
+
+	.align 4
+.global microblaze_trap_handlers
+microblaze_trap_handlers:
+	/* Exact matches come first */
+	.word ret_to_user  ; .word ret_to_user    ; .word type_SYSCALL
+	.word ret_from_intr; .word ret_from_intr  ; .word type_IRQ
+	/* Fuzzy matches go here */
+	.word ret_from_intr; .word no_intr_resched; .word type_IRQ_PREEMPT
+	.word work_pending ; .word no_work_pending; .word type_SYSCALL_PREEMPT
+	/* End of table */
+	.word 0             ; .word 0               ; .word 0
diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
index c0ede25c5b9..304882e5645 100644
--- a/arch/microblaze/kernel/entry.S
+++ b/arch/microblaze/kernel/entry.S
@@ -48,128 +48,107 @@
  */
 #if CONFIG_XILINX_MICROBLAZE0_USE_MSR_INSTR
 	.macro	clear_bip
-	msrclr	r11, MSR_BIP
-	nop
+	msrclr	r0, MSR_BIP
 	.endm
 
 	.macro	set_bip
-	msrset	r11, MSR_BIP
-	nop
+	msrset	r0, MSR_BIP
 	.endm
 
 	.macro	clear_eip
-	msrclr	r11, MSR_EIP
-	nop
+	msrclr	r0, MSR_EIP
 	.endm
 
 	.macro	set_ee
-	msrset	r11, MSR_EE
-	nop
+	msrset	r0, MSR_EE
 	.endm
 
 	.macro	disable_irq
-	msrclr	r11, MSR_IE
-	nop
+	msrclr	r0, MSR_IE
 	.endm
 
 	.macro	enable_irq
-	msrset	r11, MSR_IE
-	nop
+	msrset	r0, MSR_IE
 	.endm
 
 	.macro	set_ums
-	msrset	r11, MSR_UMS
-	nop
-	msrclr	r11, MSR_VMS
-	nop
+	msrset	r0, MSR_UMS
+	msrclr	r0, MSR_VMS
 	.endm
 
 	.macro	set_vms
-	msrclr	r11, MSR_UMS
-	nop
-	msrset	r11, MSR_VMS
-	nop
+	msrclr	r0, MSR_UMS
+	msrset	r0, MSR_VMS
+	.endm
+
+	.macro	clear_ums
+	msrclr	r0, MSR_UMS
 	.endm
 
 	.macro	clear_vms_ums
-	msrclr	r11, MSR_VMS
-	nop
-	msrclr	r11, MSR_UMS
-	nop
+	msrclr	r0, MSR_VMS | MSR_UMS
 	.endm
 #else
 	.macro	clear_bip
 	mfs	r11, rmsr
-	nop
 	andi	r11, r11, ~MSR_BIP
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	set_bip
 	mfs	r11, rmsr
-	nop
 	ori	r11, r11, MSR_BIP
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	clear_eip
 	mfs	r11, rmsr
-	nop
 	andi	r11, r11, ~MSR_EIP
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	set_ee
 	mfs	r11, rmsr
-	nop
 	ori	r11, r11, MSR_EE
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	disable_irq
 	mfs	r11, rmsr
-	nop
 	andi	r11, r11, ~MSR_IE
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	enable_irq
 	mfs	r11, rmsr
-	nop
 	ori	r11, r11, MSR_IE
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro set_ums
 	mfs	r11, rmsr
-	nop
 	ori	r11, r11, MSR_VMS
 	andni	r11, r11, MSR_UMS
 	mts	rmsr, r11
-	nop
 	.endm
 
 	.macro	set_vms
 	mfs	r11, rmsr
-	nop
 	ori	r11, r11, MSR_VMS
 	andni	r11, r11, MSR_UMS
 	mts	rmsr, r11
-	nop
+	.endm
+
+	.macro	clear_ums
+	mfs	r11, rmsr
+	andni	r11, r11, MSR_UMS
+	mts	rmsr,r11
 	.endm
 
 	.macro	clear_vms_ums
 	mfs	r11, rmsr
-	nop
 	andni	r11, r11, (MSR_VMS|MSR_UMS)
 	mts	rmsr,r11
-	nop
 	.endm
 #endif
 
@@ -180,18 +159,22 @@
 
 /* turn on virtual protected mode save */
 #define VM_ON		\
-	set_ums;		\
+	set_ums;	\
 	rted	r0, 2f;	\
-2: nop;
+	nop; \
+2:
 
 /* turn off virtual protected mode save and user mode save*/
 #define VM_OFF			\
-	clear_vms_ums;			\
+	clear_vms_ums;		\
 	rted	r0, TOPHYS(1f);	\
-1: nop;
+	nop; \
+1:
 
 #define SAVE_REGS \
 	swi	r2, r1, PTO+PT_R2;	/* Save SDA */			\
+	swi	r3, r1, PTO+PT_R3;					\
+	swi	r4, r1, PTO+PT_R4;					\
 	swi	r5, r1, PTO+PT_R5;					\
 	swi	r6, r1, PTO+PT_R6;					\
 	swi	r7, r1, PTO+PT_R7;					\
@@ -218,14 +201,14 @@
 	swi	r30, r1, PTO+PT_R30;					\
 	swi	r31, r1, PTO+PT_R31;	/* Save current task reg */	\
 	mfs	r11, rmsr;		/* save MSR */			\
-	nop;								\
 	swi	r11, r1, PTO+PT_MSR;
 
 #define RESTORE_REGS \
 	lwi	r11, r1, PTO+PT_MSR;					\
 	mts	rmsr , r11;						\
-	nop;								\
 	lwi	r2, r1, PTO+PT_R2;	/* restore SDA */		\
+	lwi	r3, r1, PTO+PT_R3;					\
+	lwi	r4, r1, PTO+PT_R4;					\
 	lwi	r5, r1, PTO+PT_R5;					\
 	lwi	r6, r1, PTO+PT_R6;					\
 	lwi	r7, r1, PTO+PT_R7;					\
@@ -252,6 +235,39 @@
 	lwi	r30, r1, PTO+PT_R30;					\
 	lwi	r31, r1, PTO+PT_R31;	/* Restore cur task reg */
 
+#define SAVE_STATE	\
+	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)); /* save stack */	\
+	/* See if already in kernel mode.*/				\
+	mfs	r1, rmsr;						\
+	andi	r1, r1, MSR_UMS;					\
+	bnei	r1, 1f;						\
+	/* Kernel-mode state save.  */					\
+	/* Reload kernel stack-ptr. */					\
+	lwi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP));			\
+	/* FIXME: I can add these two lines to one */			\
+	/* tophys(r1,r1); */						\
+	/* addik	r1, r1, -STATE_SAVE_SIZE; */			\
+	addik	r1, r1, CONFIG_KERNEL_BASE_ADDR - CONFIG_KERNEL_START - STATE_SAVE_SIZE; \
+	SAVE_REGS							\
+	brid	2f;							\
+	swi	r1, r1, PTO+PT_MODE; 	 				\
+1:	/* User-mode state save.  */					\
+	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE)); /* get saved current */\
+	tophys(r1,r1);							\
+	lwi	r1, r1, TS_THREAD_INFO;	/* get the thread info */	\
+	/* MS these three instructions can be added to one */		\
+	/* addik	r1, r1, THREAD_SIZE; */				\
+	/* tophys(r1,r1); */						\
+	/* addik	r1, r1, -STATE_SAVE_SIZE; */			\
+	addik r1, r1, THREAD_SIZE + CONFIG_KERNEL_BASE_ADDR - CONFIG_KERNEL_START - STATE_SAVE_SIZE; \
+	SAVE_REGS							\
+	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));			\
+	swi	r11, r1, PTO+PT_R1; /* Store user SP.  */		\
+	swi	r0, r1, PTO + PT_MODE; /* Was in user-mode.  */		\
+	/* MS: I am clearing UMS even in case when I come from kernel space */ \
+	clear_ums; 							\
+2:	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
+
 .text
 
 /*
@@ -267,45 +283,23 @@
  * are masked. This is nice, means we don't have to CLI before state save
  */
 C_ENTRY(_user_exception):
-	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 	addi	r14, r14, 4	/* return address is 4 byte after call */
-	swi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11);	/* Save r11 */
-
-	lwi	r11, r0, TOPHYS(PER_CPU(KM));/* See if already in kernel mode.*/
-	beqi	r11, 1f;		/* Jump ahead if coming from user */
-/* Kernel-mode state save. */
-	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP)); /* Reload kernel stack-ptr*/
-	tophys(r1,r11);
-	swi	r11, r1, (PT_R1-PT_SIZE); /* Save original SP. */
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* restore r11 */
-
-	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack. */
-	SAVE_REGS
-
-	addi	r11, r0, 1; 		/* Was in kernel-mode. */
-	swi	r11, r1, PTO+PT_MODE; /* pt_regs -> kernel mode */
-	brid	2f;
-	nop;				/* Fill delay slot */
+	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 
-/* User-mode state save.  */
-1:
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11);	/* restore r11 */
 	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE)); /* get saved current */
 	tophys(r1,r1);
 	lwi	r1, r1, TS_THREAD_INFO;	/* get stack from task_struct */
-/* calculate kernel stack pointer from task struct 8k */
-	addik	r1, r1, THREAD_SIZE;
-	tophys(r1,r1);
-
-	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack.  */
+	/* MS these three instructions can be added to one */
+	/* addik	r1, r1, THREAD_SIZE; */
+	/* tophys(r1,r1); */
+	/* addik	r1, r1, -STATE_SAVE_SIZE; */
+	addik r1, r1, THREAD_SIZE + CONFIG_KERNEL_BASE_ADDR - CONFIG_KERNEL_START - STATE_SAVE_SIZE;
 	SAVE_REGS
 
-	swi	r0, r1, PTO+PT_MODE;			/* Was in user-mode. */
 	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));
 	swi	r11, r1, PTO+PT_R1;		/* Store user SP.  */
-	addi	r11, r0, 1;
-	swi	r11, r0, TOPHYS(PER_CPU(KM));	/* Now we're in kernel-mode.  */
-2:	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
+	clear_ums;
+	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
 	/* Save away the syscall number.  */
 	swi	r12, r1, PTO+PT_R0;
 	tovirt(r1,r1)
@@ -316,10 +310,8 @@ C_ENTRY(_user_exception):
  * register should point to the location where
  * the called function should return.  [note that MAKE_SYS_CALL uses label 1] */
 
-	# Step into virtual mode.
-	set_vms;
-	addik	r11, r0, 3f
-	rtid	r11, 0
+	/* Step into virtual mode */
+	rtbd	r0, 3f
 	nop
 3:
 	lwi	r11, CURRENT_TASK, TS_THREAD_INFO /* get thread info */
@@ -363,24 +355,17 @@ C_ENTRY(_user_exception):
 	# Find and jump into the syscall handler.
 	lwi	r12, r12, sys_call_table
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
-	la	r15, r0, ret_from_trap-8
+	addi	r15, r0, ret_from_trap-8
 	bra	r12
 
 	/* The syscall number is invalid, return an error.  */
 5:
+	rtsd	r15, 8;		/* looks like a normal subroutine return */
 	addi	r3, r0, -ENOSYS;
-	rtsd	r15,8;		/* looks like a normal subroutine return */
-	or 	r0, r0, r0
-
 
 /* Entry point used to return from a syscall/trap */
 /* We re-enable BIP bit before state restore */
 C_ENTRY(ret_from_trap):
-	set_bip;			/*  Ints masked for state restore*/
-	lwi	r11, r1, PTO+PT_MODE;
-/* See if returning to kernel mode, if so, skip resched &c.  */
-	bnei	r11, 2f;
-
 	swi	r3, r1, PTO + PT_R3
 	swi	r4, r1, PTO + PT_R4
 
@@ -413,32 +398,19 @@ C_ENTRY(ret_from_trap):
 	andi	r11, r11, _TIF_SIGPENDING;
 	beqi	r11, 1f;		/* Signals to handle, handle them */
 
-	la	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
+	addik	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
 	addi	r7, r0, 1;		/* Arg 3: int in_syscall */
 	bralid	r15, do_signal;	/* Handle any signals */
 	add	r6, r0, r0;		/* Arg 2: sigset_t *oldset */
 
 /* Finally, return to user state.  */
-1:
-	lwi	r3, r1, PTO + PT_R3; /* restore syscall result */
-	lwi	r4, r1, PTO + PT_R4;
-
-	swi	r0, r0, PER_CPU(KM);	/* Now officially in user state. */
+1:	set_bip;			/*  Ints masked for state restore */
 	swi	CURRENT_TASK, r0, PER_CPU(CURRENT_SAVE); /* save current */
 	VM_OFF;
 	tophys(r1,r1);
 	RESTORE_REGS;
 	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
 	lwi	r1, r1, PT_R1 - PT_SIZE;/* Restore user stack pointer. */
-	bri	6f;
-
-/* Return to kernel state.  */
-2:	VM_OFF;
-	tophys(r1,r1);
-	RESTORE_REGS;
-	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
-	tovirt(r1,r1);
-6:
 TRAP_return:		/* Make global symbol for debugging */
 	rtbd	r14, 0;	/* Instructions to return from an IRQ */
 	nop;
@@ -450,12 +422,11 @@ TRAP_return:		/* Make global symbol for debugging */
 C_ENTRY(sys_fork_wrapper):
 	addi	r5, r0, SIGCHLD			/* Arg 0: flags */
 	lwi	r6, r1, PTO+PT_R1	/* Arg 1: child SP (use parent's) */
-	la	r7, r1, PTO			/* Arg 2: parent context */
+	addik	r7, r1, PTO			/* Arg 2: parent context */
 	add	r8. r0, r0			/* Arg 3: (unused) */
 	add	r9, r0, r0;			/* Arg 4: (unused) */
-	add	r10, r0, r0;			/* Arg 5: (unused) */
 	brid	do_fork		/* Do real work (tail-call) */
-	nop;
+	add	r10, r0, r0;			/* Arg 5: (unused) */
 
 /* This the initial entry point for a new child thread, with an appropriate
    stack in place that makes it look the the child is in the middle of an
@@ -466,35 +437,31 @@ C_ENTRY(ret_from_fork):
 	bralid	r15, schedule_tail; /* ...which is schedule_tail's arg */
 	add	r3, r5, r0;	/* switch_thread returns the prev task */
 				/* ( in the delay slot ) */
-	add	r3, r0, r0;	/* Child's fork call should return 0. */
 	brid	ret_from_trap;	/* Do normal trap return */
-	nop;
+	add	r3, r0, r0;	/* Child's fork call should return 0. */
 
 C_ENTRY(sys_vfork):
 	brid	microblaze_vfork	/* Do real work (tail-call) */
-	la	r5, r1, PTO
+	addik	r5, r1, PTO
 
 C_ENTRY(sys_clone):
 	bnei	r6, 1f;			/* See if child SP arg (arg 1) is 0. */
-	lwi	r6, r1, PTO+PT_R1;	/* If so, use paret's stack ptr */
-1:	la	r7, r1, PTO;			/* Arg 2: parent context */
+	lwi	r6, r1, PTO + PT_R1;	/* If so, use paret's stack ptr */
+1:	addik	r7, r1, PTO;			/* Arg 2: parent context */
 	add	r8, r0, r0;			/* Arg 3: (unused) */
 	add	r9, r0, r0;			/* Arg 4: (unused) */
-	add	r10, r0, r0;			/* Arg 5: (unused) */
 	brid	do_fork		/* Do real work (tail-call) */
-	nop;
+	add	r10, r0, r0;			/* Arg 5: (unused) */
 
 C_ENTRY(sys_execve):
-	la	r8, r1, PTO;		/* add user context as 4th arg */
 	brid	microblaze_execve;	/* Do real work (tail-call).*/
-	nop;
+	addik	r8, r1, PTO;		/* add user context as 4th arg */
 
 C_ENTRY(sys_rt_sigreturn_wrapper):
 	swi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
 	swi	r4, r1, PTO+PT_R4;
-	la	r5, r1, PTO;		/* add user context as 1st arg */
 	brlid	r15, sys_rt_sigreturn	/* Do real work */
-	nop;
+	addik	r5, r1, PTO;		/* add user context as 1st arg */
 	lwi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
 	lwi	r4, r1, PTO+PT_R4;
 	bri ret_from_trap /* fall through will not work here due to align */
@@ -503,83 +470,23 @@ C_ENTRY(sys_rt_sigreturn_wrapper):
 /*
  * HW EXCEPTION rutine start
  */
-
-#define SAVE_STATE	\
-	swi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* Save r11 */	\
-	set_bip;	/*equalize initial state for all possible entries*/\
-	clear_eip;							\
-	enable_irq;							\
-	set_ee;								\
-	/* See if already in kernel mode.*/				\
-	lwi	r11, r0, TOPHYS(PER_CPU(KM));				\
-	beqi	r11, 1f;		/* Jump ahead if coming from user */\
-	/* Kernel-mode state save.  */					\
-	/* Reload kernel stack-ptr. */					\
-	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));			\
-	tophys(r1,r11);							\
-	swi	r11, r1, (PT_R1-PT_SIZE); /* Save original SP. */	\
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* restore r11 */\
-	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack.  */\
-	/* store return registers separately because			\
-	 * this macros is use for others exceptions */			\
-	swi	r3, r1, PTO + PT_R3;					\
-	swi	r4, r1, PTO + PT_R4;					\
-	SAVE_REGS							\
-	/* PC, before IRQ/trap - this is one instruction above */	\
-	swi	r17, r1, PTO+PT_PC;					\
-									\
-	addi	r11, r0, 1; 		/* Was in kernel-mode.  */	\
-	swi	r11, r1, PTO+PT_MODE; 	 				\
-	brid	2f;							\
-	nop;				/* Fill delay slot */		\
-1:	/* User-mode state save.  */					\
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* restore r11 */\
-	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE)); /* get saved current */\
-	tophys(r1,r1);							\
-	lwi	r1, r1, TS_THREAD_INFO;	/* get the thread info */	\
-	addik	r1, r1, THREAD_SIZE;	/* calculate kernel stack pointer */\
-	tophys(r1,r1);							\
-									\
-	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack.  */\
-	/* store return registers separately because this macros	\
-	 * is use for others exceptions */				\
-	swi	r3, r1, PTO + PT_R3; 					\
-	swi	r4, r1, PTO + PT_R4;					\
-	SAVE_REGS							\
-	/* PC, before IRQ/trap - this is one instruction above FIXME*/	\
-	swi	r17, r1, PTO+PT_PC;					\
-									\
-	swi	r0, r1, PTO+PT_MODE; /* Was in user-mode.  */		\
-	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));			\
-	swi	r11, r1, PTO+PT_R1; /* Store user SP.  */		\
-	addi	r11, r0, 1;						\
-	swi	r11, r0, TOPHYS(PER_CPU(KM)); /* Now we're in kernel-mode.*/\
-2:	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));	\
-	/* Save away the syscall number.  */				\
-	swi	r0, r1, PTO+PT_R0;					\
-	tovirt(r1,r1)
-
 C_ENTRY(full_exception_trap):
-	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 	/* adjust exception address for privileged instruction
 	 * for finding where is it */
 	addik	r17, r17, -4
 	SAVE_STATE /* Save registers */
+	/* PC, before IRQ/trap - this is one instruction above */
+	swi	r17, r1, PTO+PT_PC;
+	tovirt(r1,r1)
 	/* FIXME this can be store directly in PT_ESR reg.
 	 * I tested it but there is a fault */
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
-	la	r15, r0, ret_from_exc - 8
-	la	r5, r1, PTO		 /* parameter struct pt_regs * regs */
+	addik	r15, r0, ret_from_exc - 8
 	mfs	r6, resr
-	nop
 	mfs	r7, rfsr;		/* save FSR */
-	nop
 	mts	rfsr, r0;	/* Clear sticky fsr */
-	nop
-	la	r12, r0, full_exception
-	set_vms;
-	rtbd	r12, 0;
-	nop;
+	rted	r0, full_exception
+	addik	r5, r1, PTO		 /* parameter struct pt_regs * regs */
 
 /*
  * Unaligned data trap.
@@ -592,19 +499,27 @@ C_ENTRY(full_exception_trap):
  * The assembler routine is in "arch/microblaze/kernel/hw_exception_handler.S"
  */
 C_ENTRY(unaligned_data_trap):
-	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
+	/* MS: I have to save r11 value and then restore it because
+	 * set_bit, clear_eip, set_ee use r11 as temp register if MSR
+	 * instructions are not used. We don't need to do if MSR instructions
+	 * are used and they use r0 instead of r11.
+	 * I am using ENTRY_SP which should be primary used only for stack
+	 * pointer saving. */
+	swi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));
+	set_bip;        /* equalize initial state for all possible entries */
+	clear_eip;
+	set_ee;
+	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));
 	SAVE_STATE		/* Save registers.*/
+	/* PC, before IRQ/trap - this is one instruction above */
+	swi	r17, r1, PTO+PT_PC;
+	tovirt(r1,r1)
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
-	la	r15, r0, ret_from_exc-8
+	addik	r15, r0, ret_from_exc-8
 	mfs	r3, resr		/* ESR */
-	nop
 	mfs	r4, rear		/* EAR */
-	nop
-	la	r7, r1, PTO		/* parameter struct pt_regs * regs */
-	la	r12, r0, _unaligned_data_exception
-	set_vms;
-	rtbd	r12, 0;	/* interrupts enabled */
-	nop;
+	rtbd	r0, _unaligned_data_exception
+	addik	r7, r1, PTO		/* parameter struct pt_regs * regs */
 
 /*
  * Page fault traps.
@@ -625,38 +540,32 @@ C_ENTRY(unaligned_data_trap):
  */
 /* data and intruction trap - which is choose is resolved int fault.c */
 C_ENTRY(page_fault_data_trap):
-	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 	SAVE_STATE		/* Save registers.*/
+	/* PC, before IRQ/trap - this is one instruction above */
+	swi	r17, r1, PTO+PT_PC;
+	tovirt(r1,r1)
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
-	la	r15, r0, ret_from_exc-8
-	la	r5, r1, PTO		/* parameter struct pt_regs * regs */
+	addik	r15, r0, ret_from_exc-8
 	mfs	r6, rear		/* parameter unsigned long address */
-	nop
 	mfs	r7, resr		/* parameter unsigned long error_code */
-	nop
-	la	r12, r0, do_page_fault
-	set_vms;
-	rtbd	r12, 0;	/* interrupts enabled */
-	nop;
+	rted	r0, do_page_fault
+	addik	r5, r1, PTO		/* parameter struct pt_regs * regs */
 
 C_ENTRY(page_fault_instr_trap):
-	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 	SAVE_STATE		/* Save registers.*/
+	/* PC, before IRQ/trap - this is one instruction above */
+	swi	r17, r1, PTO+PT_PC;
+	tovirt(r1,r1)
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
-	la	r15, r0, ret_from_exc-8
-	la	r5, r1, PTO		/* parameter struct pt_regs * regs */
+	addik	r15, r0, ret_from_exc-8
 	mfs	r6, rear		/* parameter unsigned long address */
-	nop
 	ori	r7, r0, 0		/* parameter unsigned long error_code */
-	la	r12, r0, do_page_fault
-	set_vms;
-	rtbd	r12, 0;	/* interrupts enabled */
-	nop;
+	rted	r0, do_page_fault
+	addik	r5, r1, PTO		/* parameter struct pt_regs * regs */
 
 /* Entry point used to return from an exception.  */
 C_ENTRY(ret_from_exc):
-	set_bip;			/*  Ints masked for state restore*/
-	lwi	r11, r1, PTO+PT_MODE;
+	lwi	r11, r1, PTO + PT_MODE;
 	bnei	r11, 2f;		/* See if returning to kernel mode, */
 					/* ... if so, skip resched &c.  */
 
@@ -687,32 +596,27 @@ C_ENTRY(ret_from_exc):
 	 * traps), but signal handlers may want to examine or change the
 	 * complete register state.  Here we save anything not saved by
 	 * the normal entry sequence, so that it may be safely restored
-	 * (in a possibly modified form) after do_signal returns.
-	 * store return registers separately because this macros is use
-	 * for others exceptions */
-	la	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
+	 * (in a possibly modified form) after do_signal returns. */
+	addik	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
 	addi	r7, r0, 0;		/* Arg 3: int in_syscall */
 	bralid	r15, do_signal;	/* Handle any signals */
 	add	r6, r0, r0;		/* Arg 2: sigset_t *oldset */
 
 /* Finally, return to user state.  */
-1:	swi	r0, r0, PER_CPU(KM);	/* Now officially in user state. */
+1:	set_bip;			/* Ints masked for state restore */
 	swi	CURRENT_TASK, r0, PER_CPU(CURRENT_SAVE); /* save current */
 	VM_OFF;
 	tophys(r1,r1);
 
-	lwi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
-	lwi	r4, r1, PTO+PT_R4;
 	RESTORE_REGS;
 	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
 
 	lwi	r1, r1, PT_R1 - PT_SIZE; /* Restore user stack pointer. */
 	bri	6f;
 /* Return to kernel state.  */
-2:	VM_OFF;
+2:	set_bip;			/* Ints masked for state restore */
+	VM_OFF;
 	tophys(r1,r1);
-	lwi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
-	lwi	r4, r1, PTO+PT_R4;
 	RESTORE_REGS;
 	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
 
@@ -736,36 +640,23 @@ C_ENTRY(_interrupt):
 /* MS: we are in physical address */
 /* Save registers, switch to proper stack, convert SP to virtual.*/
 	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP))
-	swi	r11, r0, TOPHYS(PER_CPU(R11_SAVE));
 	/* MS: See if already in kernel mode. */
-	lwi	r11, r0, TOPHYS(PER_CPU(KM));
-	beqi	r11, 1f; /* MS: Jump ahead if coming from user */
+	mfs	r1, rmsr
+	nop
+	andi	r1, r1, MSR_UMS
+	bnei	r1, 1f
 
 /* Kernel-mode state save. */
-	or	r11, r1, r0
-	tophys(r1,r11); /* MS: I have in r1 physical address where stack is */
-/* MS: Save original SP - position PT_R1 to next stack frame 4 *1 - 152*/
-	swi	r11, r1, (PT_R1 - PT_SIZE);
-/* MS: restore r11 because of saving in SAVE_REGS */
-	lwi	r11, r0, TOPHYS(PER_CPU(R11_SAVE));
+	lwi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP))
+	tophys(r1,r1); /* MS: I have in r1 physical address where stack is */
 	/* save registers */
 /* MS: Make room on the stack -> activation record */
 	addik	r1, r1, -STATE_SAVE_SIZE;
-/* MS: store return registers separately because
- * this macros is use for others exceptions */
-	swi	r3, r1, PTO + PT_R3;
-	swi	r4, r1, PTO + PT_R4;
 	SAVE_REGS
-	/* MS: store mode */
-	addi	r11, r0, 1; /* MS: Was in kernel-mode. */
-	swi	r11, r1, PTO + PT_MODE; /* MS: and save it */
 	brid	2f;
-	nop; /* MS: Fill delay slot */
-
+	swi	r1, r1, PTO + PT_MODE; /* 0 - user mode, 1 - kernel mode */
 1:
 /* User-mode state save. */
-/* MS: restore r11 -> FIXME move before SAVE_REG */
-	lwi	r11, r0, TOPHYS(PER_CPU(R11_SAVE));
  /* MS: get the saved current */
 	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
 	tophys(r1,r1);
@@ -774,27 +665,18 @@ C_ENTRY(_interrupt):
 	tophys(r1,r1);
 	/* save registers */
 	addik	r1, r1, -STATE_SAVE_SIZE;
-	swi	r3, r1, PTO+PT_R3;
-	swi	r4, r1, PTO+PT_R4;
 	SAVE_REGS
 	/* calculate mode */
 	swi	r0, r1, PTO + PT_MODE;
 	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));
 	swi	r11, r1, PTO+PT_R1;
-	/* setup kernel mode to KM */
-	addi	r11, r0, 1;
-	swi	r11, r0, TOPHYS(PER_CPU(KM));
-
+	clear_ums;
 2:
 	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
-	swi	r0, r1, PTO + PT_R0;
 	tovirt(r1,r1)
-	la	r5, r1, PTO;
-	set_vms;
-	la	r11, r0, do_IRQ;
-	la	r15, r0, irq_call;
-irq_call:rtbd	r11, 0;
-	nop;
+	addik	r15, r0, irq_call;
+irq_call:rtbd	r0, do_IRQ;
+	addik	r5, r1, PTO;
 
 /* MS: we are in virtual mode */
 ret_from_irq:
@@ -815,7 +697,7 @@ ret_from_irq:
 	beqid	r11, no_intr_resched
 /* Handle a signal return; Pending signals should be in r18. */
 	addi	r7, r0, 0; /* Arg 3: int in_syscall */
-	la	r5, r1, PTO; /* Arg 1: struct pt_regs *regs */
+	addik	r5, r1, PTO; /* Arg 1: struct pt_regs *regs */
 	bralid	r15, do_signal;	/* Handle any signals */
 	add	r6, r0, r0; /* Arg 2: sigset_t *oldset */
 
@@ -823,12 +705,9 @@ ret_from_irq:
 no_intr_resched:
     /* Disable interrupts, we are now committed to the state restore */
 	disable_irq
-	swi	r0, r0, PER_CPU(KM); /* MS: Now officially in user state. */
 	swi	CURRENT_TASK, r0, PER_CPU(CURRENT_SAVE);
 	VM_OFF;
 	tophys(r1,r1);
-	lwi	r3, r1, PTO + PT_R3; /* MS: restore saved r3, r4 registers */
-	lwi	r4, r1, PTO + PT_R4;
 	RESTORE_REGS
 	addik	r1, r1, STATE_SAVE_SIZE /* MS: Clean up stack space. */
 	lwi	r1, r1, PT_R1 - PT_SIZE;
@@ -857,8 +736,6 @@ restore:
 #endif
 	VM_OFF /* MS: turn off MMU */
 	tophys(r1,r1)
-	lwi	r3, r1, PTO + PT_R3; /* MS: restore saved r3, r4 registers */
-	lwi	r4, r1, PTO + PT_R4;
 	RESTORE_REGS
 	addik	r1, r1, STATE_SAVE_SIZE	/* MS: Clean up stack space. */
 	tovirt(r1,r1);
@@ -868,86 +745,91 @@ IRQ_return: /* MS: Make global symbol for debugging */
 	nop
 
 /*
- * `Debug' trap
- *  We enter dbtrap in "BIP" (breakpoint) mode.
- *  So we exit the breakpoint mode with an 'rtbd' and proceed with the
- *  original dbtrap.
- *  however, wait to save state first
+ * Debug trap for KGDB. Enter to _debug_exception by brki r16, 0x18
+ * and call handling function with saved pt_regs
  */
 C_ENTRY(_debug_exception):
 	/* BIP bit is set on entry, no interrupts can occur */
 	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP))
 
-	swi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* Save r11 */
-	set_bip;	/*equalize initial state for all possible entries*/
-	clear_eip;
-	enable_irq;
-	lwi	r11, r0, TOPHYS(PER_CPU(KM));/* See if already in kernel mode.*/
-	beqi	r11, 1f;		/* Jump ahead if coming from user */
-	/* Kernel-mode state save.  */
-	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP)); /* Reload kernel stack-ptr*/
-	tophys(r1,r11);
-	swi	r11, r1, (PT_R1-PT_SIZE); /* Save original SP. */
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* restore r11 */
+	mfs	r1, rmsr
+	nop
+	andi	r1, r1, MSR_UMS
+	bnei	r1, 1f
+/* MS: Kernel-mode state save - kgdb */
+	lwi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)); /* Reload kernel stack-ptr*/
 
-	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack.  */
-	swi	r3, r1, PTO + PT_R3;
-	swi	r4, r1, PTO + PT_R4;
+	/* BIP bit is set on entry, no interrupts can occur */
+	addik   r1, r1, CONFIG_KERNEL_BASE_ADDR - CONFIG_KERNEL_START - STATE_SAVE_SIZE;
 	SAVE_REGS;
+	/* save all regs to pt_reg structure */
+	swi	r0, r1, PTO+PT_R0;	/* R0 must be saved too */
+	swi	r14, r1, PTO+PT_R14	/* rewrite saved R14 value */
+	swi	r16, r1, PTO+PT_R16
+	swi	r16, r1, PTO+PT_PC; /* PC and r16 are the same */
+	swi	r17, r1, PTO+PT_R17
+	/* save special purpose registers to pt_regs */
+	mfs	r11, rear;
+	swi	r11, r1, PTO+PT_EAR;
+	mfs	r11, resr;
+	swi	r11, r1, PTO+PT_ESR;
+	mfs	r11, rfsr;
+	swi	r11, r1, PTO+PT_FSR;
+
+	/* stack pointer is in physical address at it is decrease
+	 * by STATE_SAVE_SIZE but we need to get correct R1 value */
+	addik   r11, r1, CONFIG_KERNEL_START - CONFIG_KERNEL_BASE_ADDR + STATE_SAVE_SIZE;
+	swi	r11, r1, PTO+PT_R1
+	/* MS: r31 - current pointer isn't changed */
+	tovirt(r1,r1)
+#ifdef CONFIG_KGDB
+	addi	r5, r1, PTO /* pass pt_reg address as the first arg */
+	la	r15, r0, dbtrap_call; /* return address */
+	rtbd	r0, microblaze_kgdb_break
+	nop;
+#endif
+	/* MS: Place handler for brki from kernel space if KGDB is OFF.
+	 * It is very unlikely that another brki instruction is called. */
+	bri 0
 
-	addi	r11, r0, 1; 		/* Was in kernel-mode.  */
-	swi	r11, r1, PTO + PT_MODE;
-	brid	2f;
-	nop;				/* Fill delay slot */
-1:      /* User-mode state save.  */
-	lwi	r11, r0, TOPHYS(r0_ram + PTO + PT_R11); /* restore r11 */
-	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE)); /* get saved current */
+/* MS: User-mode state save - gdb */
+1:	lwi	r1, r0, TOPHYS(PER_CPU(CURRENT_SAVE)); /* get saved current */
 	tophys(r1,r1);
 	lwi	r1, r1, TS_THREAD_INFO;	/* get the thread info */
 	addik	r1, r1, THREAD_SIZE;	/* calculate kernel stack pointer */
 	tophys(r1,r1);
 
 	addik	r1, r1, -STATE_SAVE_SIZE; /* Make room on the stack.  */
-	swi	r3, r1, PTO + PT_R3;
-	swi	r4, r1, PTO + PT_R4;
 	SAVE_REGS;
-
-	swi	r0, r1, PTO+PT_MODE; /* Was in user-mode.  */
+	swi	r17, r1, PTO+PT_R17;
+	swi	r16, r1, PTO+PT_R16;
+	swi	r16, r1, PTO+PT_PC;	/* Save LP */
+	swi	r0, r1, PTO + PT_MODE; /* Was in user-mode.  */
 	lwi	r11, r0, TOPHYS(PER_CPU(ENTRY_SP));
 	swi	r11, r1, PTO+PT_R1; /* Store user SP.  */
-	addi	r11, r0, 1;
-	swi	r11, r0, TOPHYS(PER_CPU(KM));	/* Now we're in kernel-mode.  */
-2:	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
-	/* Save away the syscall number.  */
-	swi	r0, r1, PTO+PT_R0;
+	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
 	tovirt(r1,r1)
-
-	addi	r5, r0, SIGTRAP		     /* send the trap signal */
-	add	r6, r0, CURRENT_TASK; /* Get current task ptr into r11 */
-	addk	r7, r0, r0		     /* 3rd param zero */
-
 	set_vms;
-	la	r11, r0, send_sig;
-	la	r15, r0, dbtrap_call;
-dbtrap_call:	rtbd	r11, 0;
-	nop;
+	addik	r5, r1, PTO;
+	addik	r15, r0, dbtrap_call;
+dbtrap_call: /* Return point for kernel/user entry + 8 because of rtsd r15, 8 */
+	rtbd	r0, sw_exception
+	nop
 
-	set_bip;			/*  Ints masked for state restore*/
-	lwi	r11, r1, PTO+PT_MODE;
+	/* MS: The first instruction for the second part of the gdb/kgdb */
+	set_bip; /* Ints masked for state restore */
+	lwi	r11, r1, PTO + PT_MODE;
 	bnei	r11, 2f;
-
+/* MS: Return to user space - gdb */
 	/* Get current task ptr into r11 */
 	lwi	r11, CURRENT_TASK, TS_THREAD_INFO;	/* get thread info */
 	lwi	r11, r11, TI_FLAGS;	/* get flags in thread info */
 	andi	r11, r11, _TIF_NEED_RESCHED;
 	beqi	r11, 5f;
 
-/* Call the scheduler before returning from a syscall/trap. */
-
+	/* Call the scheduler before returning from a syscall/trap. */
 	bralid	r15, schedule;	/* Call scheduler */
 	nop;				/* delay slot */
-	/* XXX Is PT_DTRACE handling needed here? */
-	/* XXX m68knommu also checks TASK_STATE & TASK_COUNTER here.  */
 
 	/* Maybe handle a signal */
 5:	lwi	r11, CURRENT_TASK, TS_THREAD_INFO;	/* get thread info */
@@ -955,54 +837,40 @@ dbtrap_call:	rtbd	r11, 0;
 	andi	r11, r11, _TIF_SIGPENDING;
 	beqi	r11, 1f;		/* Signals to handle, handle them */
 
-/* Handle a signal return; Pending signals should be in r18.  */
-	/* Not all registers are saved by the normal trap/interrupt entry
-	   points (for instance, call-saved registers (because the normal
-	   C-compiler calling sequence in the kernel makes sure they're
-	   preserved), and call-clobbered registers in the case of
-	   traps), but signal handlers may want to examine or change the
-	   complete register state.  Here we save anything not saved by
-	   the normal entry sequence, so that it may be safely restored
-	   (in a possibly modified form) after do_signal returns.  */
-
-	la	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
+	addik	r5, r1, PTO;		/* Arg 1: struct pt_regs *regs */
 	addi  r7, r0, 0;	/* Arg 3: int in_syscall */
 	bralid	r15, do_signal;	/* Handle any signals */
 	add	r6, r0, r0;		/* Arg 2: sigset_t *oldset */
 
-
 /* Finally, return to user state.  */
-1:	swi	r0, r0, PER_CPU(KM);	/* Now officially in user state. */
-	swi	CURRENT_TASK, r0, PER_CPU(CURRENT_SAVE); /* save current */
+1:	swi	CURRENT_TASK, r0, PER_CPU(CURRENT_SAVE); /* save current */
 	VM_OFF;
 	tophys(r1,r1);
-
-	lwi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
-	lwi	r4, r1, PTO+PT_R4;
+	/* MS: Restore all regs */
 	RESTORE_REGS
-	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
-
-
-	lwi	r1, r1, PT_R1 - PT_SIZE;
-					/* Restore user stack pointer. */
-	bri	6f;
+	lwi	r17, r1, PTO+PT_R17;
+	lwi	r16, r1, PTO+PT_R16;
+	addik	r1, r1, STATE_SAVE_SIZE	 /* Clean up stack space */
+	lwi	r1, r1, PT_R1 - PT_SIZE; /* Restore user stack pointer */
+DBTRAP_return_user: /* MS: Make global symbol for debugging */
+	rtbd	r16, 0; /* MS: Instructions to return from a debug trap */
+	nop;
 
-/* Return to kernel state.  */
+/* MS: Return to kernel state - kgdb */
 2:	VM_OFF;
 	tophys(r1,r1);
-	lwi	r3, r1, PTO+PT_R3; /* restore saved r3, r4 registers */
-	lwi	r4, r1, PTO+PT_R4;
+	/* MS: Restore all regs */
 	RESTORE_REGS
-	addik	r1, r1, STATE_SAVE_SIZE		/* Clean up stack space.  */
-
+	lwi	r14, r1, PTO+PT_R14;
+	lwi	r16, r1, PTO+PT_PC;
+	lwi	r17, r1, PTO+PT_R17;
+	addik	r1, r1, STATE_SAVE_SIZE; /* MS: Clean up stack space */
 	tovirt(r1,r1);
-6:
-DBTRAP_return:		/* Make global symbol for debugging */
-	rtbd	r14, 0;	/* Instructions to return from an IRQ */
+DBTRAP_return_kernel: /* MS: Make global symbol for debugging */
+	rtbd	r16, 0; /* MS: Instructions to return from a debug trap */
 	nop;
 
 
-
 ENTRY(_switch_to)
 	/* prepare return value */
 	addk	r3, r0, CURRENT_TASK
@@ -1037,16 +905,12 @@ ENTRY(_switch_to)
 	swi	r30, r11, CC_R30
 	/* special purpose registers */
 	mfs	r12, rmsr
-	nop
 	swi	r12, r11, CC_MSR
 	mfs	r12, rear
-	nop
 	swi	r12, r11, CC_EAR
 	mfs	r12, resr
-	nop
 	swi	r12, r11, CC_ESR
 	mfs	r12, rfsr
-	nop
 	swi	r12, r11, CC_FSR
 
 	/* update r31, the current-give me pointer to task which will be next */
@@ -1085,10 +949,8 @@ ENTRY(_switch_to)
 	/* special purpose registers */
 	lwi	r12, r11, CC_FSR
 	mts	rfsr, r12
-	nop
 	lwi	r12, r11, CC_MSR
 	mts	rmsr, r12
-	nop
 
 	rtsd	r15, 8
 	nop
@@ -1096,15 +958,6 @@ ENTRY(_switch_to)
 ENTRY(_reset)
 	brai	0x70; /* Jump back to FS-boot */
 
-ENTRY(_break)
-	mfs	r5, rmsr
-	nop
-	swi	r5, r0, 0x250 + TOPHYS(r0_ram)
-	mfs	r5, resr
-	nop
-	swi	r5, r0, 0x254 + TOPHYS(r0_ram)
-	bri	0
-
 	/* These are compiled and loaded into high memory, then
 	 * copied into place in mach_early_setup */
 	.section	.init.ivt, "ax"
@@ -1116,14 +969,38 @@ ENTRY(_break)
 	nop
 	brai	TOPHYS(_user_exception); /* syscall handler */
 	brai	TOPHYS(_interrupt);	/* Interrupt handler */
-	brai	TOPHYS(_break);		/* nmi trap handler */
+	brai	TOPHYS(_debug_exception);	/* debug trap handler */
 	brai	TOPHYS(_hw_exception_handler);	/* HW exception handler */
 
-	.org	0x60
-	brai	TOPHYS(_debug_exception);	/* debug trap handler*/
-
 .section .rodata,"a"
 #include "syscall_table.S"
 
 syscall_table_size=(.-sys_call_table)
 
+type_SYSCALL:
+	.ascii "SYSCALL\0"
+type_IRQ:
+	.ascii "IRQ\0"
+type_IRQ_PREEMPT:
+	.ascii "IRQ (PREEMPTED)\0"
+type_SYSCALL_PREEMPT:
+	.ascii " SYSCALL (PREEMPTED)\0"
+
+	/*
+	 * Trap decoding for stack unwinder
+	 * Tuples are (start addr, end addr, string)
+	 * If return address lies on [start addr, end addr],
+	 * unwinder displays 'string'
+	 */
+
+	.align 4
+.global microblaze_trap_handlers
+microblaze_trap_handlers:
+	/* Exact matches come first */
+	.word ret_from_trap; .word ret_from_trap   ; .word type_SYSCALL
+	.word ret_from_irq ; .word ret_from_irq    ; .word type_IRQ
+	/* Fuzzy matches go here */
+	.word ret_from_irq ; .word no_intr_resched ; .word type_IRQ_PREEMPT
+	.word ret_from_trap; .word TRAP_return     ; .word type_SYSCALL_PREEMPT
+	/* End of table */
+	.word 0               ; .word 0               ; .word 0
diff --git a/arch/microblaze/kernel/exceptions.c b/arch/microblaze/kernel/exceptions.c
index 02cbdfe5aa8..b98ee8d0c1c 100644
--- a/arch/microblaze/kernel/exceptions.c
+++ b/arch/microblaze/kernel/exceptions.c
@@ -48,12 +48,17 @@ void die(const char *str, struct pt_regs *fp, long err)
 	do_exit(err);
 }
 
+/* for user application debugging */
+void sw_exception(struct pt_regs *regs)
+{
+	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->r16);
+}
+
 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
 {
 	siginfo_t info;
 
 	if (kernel_mode(regs)) {
-		debugger(regs);
 		die("Exception in kernel mode", regs, signr);
 	}
 	info.si_signo = signr;
@@ -143,7 +148,7 @@ asmlinkage void full_exception(struct pt_regs *regs, unsigned int type,
 #ifdef CONFIG_MMU
 	case MICROBLAZE_PRIVILEGED_EXCEPTION:
 		pr_debug(KERN_WARNING "Privileged exception\n");
-		/* "brk r0,r0" - used as debug breakpoint */
+		/* "brk r0,r0" - used as debug breakpoint - old toolchain */
 		if (get_user(code, (unsigned long *)regs->pc) == 0
 			&& code == 0x980c0000) {
 			_exception(SIGTRAP, regs, TRAP_BRKPT, addr);
diff --git a/arch/microblaze/kernel/head.S b/arch/microblaze/kernel/head.S
index 1bf73988826..42434008209 100644
--- a/arch/microblaze/kernel/head.S
+++ b/arch/microblaze/kernel/head.S
@@ -43,10 +43,10 @@
 .global empty_zero_page
 .align 12
 empty_zero_page:
-	.space	4096
+	.space	PAGE_SIZE
 .global swapper_pg_dir
 swapper_pg_dir:
-	.space	4096
+	.space	PAGE_SIZE
 
 #endif /* CONFIG_MMU */
 
diff --git a/arch/microblaze/kernel/hw_exception_handler.S b/arch/microblaze/kernel/hw_exception_handler.S
index 995a2123635..781195438ee 100644
--- a/arch/microblaze/kernel/hw_exception_handler.S
+++ b/arch/microblaze/kernel/hw_exception_handler.S
@@ -78,9 +78,6 @@
 #include <asm/asm-offsets.h>
 
 /* Helpful Macros */
-#ifndef CONFIG_MMU
-#define EX_HANDLER_STACK_SIZ	(4*19)
-#endif
 #define NUM_TO_REG(num)		r ## num
 
 #ifdef CONFIG_MMU
@@ -988,6 +985,7 @@ ex_unaligned_fixup:
 .end _unaligned_data_exception
 #endif /* CONFIG_MMU */
 
+.global ex_handler_unhandled
 ex_handler_unhandled:
 /* FIXME add handle function for unhandled exception - dump register */
 	bri 0
diff --git a/arch/microblaze/kernel/irq.c b/arch/microblaze/kernel/irq.c
index 8f120aca123..598f1fd61c8 100644
--- a/arch/microblaze/kernel/irq.c
+++ b/arch/microblaze/kernel/irq.c
@@ -37,6 +37,7 @@ void __irq_entry do_IRQ(struct pt_regs *regs)
 {
 	unsigned int irq;
 	struct pt_regs *old_regs = set_irq_regs(regs);
+	trace_hardirqs_off();
 
 	irq_enter();
 	irq = get_irq(regs);
@@ -53,6 +54,7 @@ next_irq:
 
 	irq_exit();
 	set_irq_regs(old_regs);
+	trace_hardirqs_on();
 }
 
 int show_interrupts(struct seq_file *p, void *v)
diff --git a/arch/microblaze/kernel/kgdb.c b/arch/microblaze/kernel/kgdb.c
new file mode 100644
index 00000000000..bfc006b7f2d
--- /dev/null
+++ b/arch/microblaze/kernel/kgdb.c
@@ -0,0 +1,147 @@
+/*
+ * Microblaze KGDB support
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/kgdb.h>
+#include <linux/kdebug.h>
+#include <linux/irq.h>
+#include <linux/io.h>
+#include <asm/cacheflush.h>
+#include <asm/asm-offsets.h>
+#include <asm/pvr.h>
+
+#define GDB_REG		0
+#define GDB_PC		32
+#define GDB_MSR		33
+#define GDB_EAR		34
+#define GDB_ESR		35
+#define GDB_FSR		36
+#define GDB_BTR		37
+#define GDB_PVR		38
+#define GDB_REDR	50
+#define GDB_RPID	51
+#define GDB_RZPR	52
+#define GDB_RTLBX	53
+#define GDB_RTLBSX	54 /* mfs can't read it */
+#define GDB_RTLBLO	55
+#define GDB_RTLBHI	56
+
+/* keep pvr separately because it is unchangeble */
+struct pvr_s pvr;
+
+void pt_regs_to_gdb_regs(unsigned long *gdb_regs, struct pt_regs *regs)
+{
+	int i;
+	unsigned long *pt_regb = (unsigned long *)regs;
+	int temp;
+	/* registers r0 - r31, pc, msr, ear, esr, fsr + do not save pt_mode */
+	for (i = 0; i < (sizeof(struct pt_regs) / 4) - 1; i++)
+		gdb_regs[i] = pt_regb[i];
+
+	/* Branch target register can't be changed */
+	__asm__ __volatile__ ("mfs %0, rbtr;" : "=r"(temp) : );
+	gdb_regs[GDB_BTR] = temp;
+
+	/* pvr part  - we have 11 pvr regs */
+	for (i = 0; i < sizeof(struct pvr_s)/4; i++)
+		gdb_regs[GDB_PVR + i] = pvr.pvr[i];
+
+	/* read special registers - can't be changed */
+	__asm__ __volatile__ ("mfs %0, redr;" : "=r"(temp) : );
+	gdb_regs[GDB_REDR] = temp;
+	__asm__ __volatile__ ("mfs %0, rpid;" : "=r"(temp) : );
+	gdb_regs[GDB_RPID] = temp;
+	__asm__ __volatile__ ("mfs %0, rzpr;" : "=r"(temp) : );
+	gdb_regs[GDB_RZPR] = temp;
+	__asm__ __volatile__ ("mfs %0, rtlbx;" : "=r"(temp) : );
+	gdb_regs[GDB_RTLBX] = temp;
+	__asm__ __volatile__ ("mfs %0, rtlblo;" : "=r"(temp) : );
+	gdb_regs[GDB_RTLBLO] = temp;
+	__asm__ __volatile__ ("mfs %0, rtlbhi;" : "=r"(temp) : );
+	gdb_regs[GDB_RTLBHI] = temp;
+}
+
+void gdb_regs_to_pt_regs(unsigned long *gdb_regs, struct pt_regs *regs)
+{
+	int i;
+	unsigned long *pt_regb = (unsigned long *)regs;
+
+	/* pt_regs and gdb_regs have the same 37 values.
+	 * The rest of gdb_regs are unused and can't be changed.
+	 * r0 register value can't be changed too. */
+	for (i = 1; i < (sizeof(struct pt_regs) / 4) - 1; i++)
+		pt_regb[i] = gdb_regs[i];
+}
+
+void microblaze_kgdb_break(struct pt_regs *regs)
+{
+	if (kgdb_handle_exception(1, SIGTRAP, 0, regs) != 0)
+		return 0;
+
+	/* Jump over the first arch_kgdb_breakpoint which is barrier to
+	 * get kgdb work. The same solution is used for powerpc */
+	if (*(u32 *) (regs->pc) == *(u32 *) (&arch_kgdb_ops.gdb_bpt_instr))
+		regs->pc += BREAK_INSTR_SIZE;
+}
+
+/* untested */
+void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *p)
+{
+	int i;
+	unsigned long *pt_regb = (unsigned long *)(p->thread.regs);
+
+	/* registers r0 - r31, pc, msr, ear, esr, fsr + do not save pt_mode */
+	for (i = 0; i < (sizeof(struct pt_regs) / 4) - 1; i++)
+		gdb_regs[i] = pt_regb[i];
+
+	/* pvr part  - we have 11 pvr regs */
+	for (i = 0; i < sizeof(struct pvr_s)/4; i++)
+		gdb_regs[GDB_PVR + i] = pvr.pvr[i];
+}
+
+void kgdb_arch_set_pc(struct pt_regs *regs, unsigned long ip)
+{
+	regs->pc = ip;
+}
+
+int kgdb_arch_handle_exception(int vector, int signo, int err_code,
+			       char *remcom_in_buffer, char *remcom_out_buffer,
+			       struct pt_regs *regs)
+{
+	char *ptr;
+	unsigned long address;
+	int cpu = smp_processor_id();
+
+	switch (remcom_in_buffer[0]) {
+	case 'c':
+		/* handle the optional parameter */
+		ptr = &remcom_in_buffer[1];
+		if (kgdb_hex2long(&ptr, &address))
+			regs->pc = address;
+
+		return 0;
+	}
+	return -1; /* this means that we do not want to exit from the handler */
+}
+
+int kgdb_arch_init(void)
+{
+	get_pvr(&pvr); /* Fill PVR structure */
+	return 0;
+}
+
+void kgdb_arch_exit(void)
+{
+	/* Nothing to do */
+}
+
+/*
+ * Global data
+ */
+struct kgdb_arch arch_kgdb_ops = {
+	.gdb_bpt_instr = {0xba, 0x0c, 0x00, 0x18}, /* brki r16, 0x18 */
+};
diff --git a/arch/microblaze/kernel/misc.S b/arch/microblaze/kernel/misc.S
index 0fb5fc6c1fc..206da3da361 100644
--- a/arch/microblaze/kernel/misc.S
+++ b/arch/microblaze/kernel/misc.S
@@ -76,7 +76,7 @@ early_console_reg_tlb_alloc:
 	 * the UARTs nice and early.  We use a 4k real==virtual mapping.
 	 */
 	ori	r4, r0, MICROBLAZE_TLB_SIZE - 1
-	mts	rtlbx, r4 /* TLB slot 2 */
+	mts	rtlbx, r4 /* TLB slot 63 */
 
 	or	r4,r5,r0
 	andi	r4,r4,0xfffff000
diff --git a/arch/microblaze/kernel/process.c b/arch/microblaze/kernel/process.c
index 09bed44dfcd..ba7c4b16ed3 100644
--- a/arch/microblaze/kernel/process.c
+++ b/arch/microblaze/kernel/process.c
@@ -76,8 +76,11 @@ __setup("hlt", hlt_setup);
 void default_idle(void)
 {
 	if (likely(hlt_counter)) {
-		while (!need_resched())
-			cpu_relax();
+		local_irq_disable();
+		stop_critical_timings();
+		cpu_relax();
+		start_critical_timings();
+		local_irq_enable();
 	} else {
 		clear_thread_flag(TIF_POLLING_NRFLAG);
 		smp_mb__after_clear_bit();
diff --git a/arch/microblaze/kernel/ptrace.c b/arch/microblaze/kernel/ptrace.c
index a4a7770c614..dc03ffc8174 100644
--- a/arch/microblaze/kernel/ptrace.c
+++ b/arch/microblaze/kernel/ptrace.c
@@ -38,6 +38,8 @@
 #include <asm/processor.h>
 #include <linux/uaccess.h>
 #include <asm/asm-offsets.h>
+#include <asm/cacheflush.h>
+#include <asm/io.h>
 
 /* Returns the address where the register at REG_OFFS in P is stashed away. */
 static microblaze_reg_t *reg_save_addr(unsigned reg_offs,
@@ -101,8 +103,21 @@ long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 			microblaze_reg_t *reg_addr = reg_save_addr(addr, child);
 			if (request == PTRACE_PEEKUSR)
 				val = *reg_addr;
-			else
+			else {
+#if 1
 				*reg_addr = data;
+#else
+				/* MS potential problem on WB system
+				 * Be aware that reg_addr is virtual address
+				 * virt_to_phys conversion is necessary.
+				 * This could be sensible solution.
+				 */
+				u32 paddr = virt_to_phys((u32)reg_addr);
+				invalidate_icache_range(paddr, paddr + 4);
+				*reg_addr = data;
+				flush_dcache_range(paddr, paddr + 4);
+#endif
+			}
 		} else
 			rval = -EIO;
 
diff --git a/arch/microblaze/kernel/stacktrace.c b/arch/microblaze/kernel/stacktrace.c
index 123692f2264..84bc6686102 100644
--- a/arch/microblaze/kernel/stacktrace.c
+++ b/arch/microblaze/kernel/stacktrace.c
@@ -14,52 +14,18 @@
 #include <linux/thread_info.h>
 #include <linux/ptrace.h>
 #include <linux/module.h>
+#include <asm/unwind.h>
 
-/* FIXME initial support */
 void save_stack_trace(struct stack_trace *trace)
 {
-	unsigned long *sp;
-	unsigned long addr;
-	asm("addik %0, r1, 0" : "=r" (sp));
-
-	while (!kstack_end(sp)) {
-		addr = *sp++;
-		if (__kernel_text_address(addr)) {
-			if (trace->skip > 0)
-				trace->skip--;
-			else
-				trace->entries[trace->nr_entries++] = addr;
-
-			if (trace->nr_entries >= trace->max_entries)
-				break;
-		}
-	}
+	/* Exclude our helper functions from the trace*/
+	trace->skip += 2;
+	microblaze_unwind(NULL, trace);
 }
 EXPORT_SYMBOL_GPL(save_stack_trace);
 
 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
 {
-	unsigned int *sp;
-	unsigned long addr;
-
-	struct thread_info *ti = task_thread_info(tsk);
-
-	if (tsk == current)
-		asm("addik %0, r1, 0" : "=r" (sp));
-	else
-		sp = (unsigned int *)ti->cpu_context.r1;
-
-	while (!kstack_end(sp)) {
-		addr = *sp++;
-		if (__kernel_text_address(addr)) {
-			if (trace->skip > 0)
-				trace->skip--;
-			else
-				trace->entries[trace->nr_entries++] = addr;
-
-			if (trace->nr_entries >= trace->max_entries)
-				break;
-		}
-	}
+	microblaze_unwind(tsk, trace);
 }
 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
diff --git a/arch/microblaze/kernel/timer.c b/arch/microblaze/kernel/timer.c
index ed61b2f1771..b1380ae93ae 100644
--- a/arch/microblaze/kernel/timer.c
+++ b/arch/microblaze/kernel/timer.c
@@ -28,6 +28,7 @@
 #include <asm/prom.h>
 #include <asm/irq.h>
 #include <asm/system.h>
+#include <linux/cnt32_to_63.h>
 
 #ifdef CONFIG_SELFMOD_TIMER
 #include <asm/selfmod.h>
@@ -135,7 +136,7 @@ static void microblaze_timer_set_mode(enum clock_event_mode mode,
 static struct clock_event_device clockevent_microblaze_timer = {
 	.name		= "microblaze_clockevent",
 	.features       = CLOCK_EVT_FEAT_ONESHOT | CLOCK_EVT_FEAT_PERIODIC,
-	.shift		= 24,
+	.shift		= 8,
 	.rating		= 300,
 	.set_next_event	= microblaze_timer_set_next_event,
 	.set_mode	= microblaze_timer_set_mode,
@@ -195,7 +196,7 @@ static cycle_t microblaze_cc_read(const struct cyclecounter *cc)
 static struct cyclecounter microblaze_cc = {
 	.read = microblaze_cc_read,
 	.mask = CLOCKSOURCE_MASK(32),
-	.shift = 24,
+	.shift = 8,
 };
 
 int __init init_microblaze_timecounter(void)
@@ -213,7 +214,7 @@ static struct clocksource clocksource_microblaze = {
 	.rating		= 300,
 	.read		= microblaze_read,
 	.mask		= CLOCKSOURCE_MASK(32),
-	.shift		= 24, /* I can shift it */
+	.shift		= 8, /* I can shift it */
 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
 };
 
@@ -235,6 +236,12 @@ static int __init microblaze_clocksource_init(void)
 	return 0;
 }
 
+/*
+ * We have to protect accesses before timer initialization
+ * and return 0 for sched_clock function below.
+ */
+static int timer_initialized;
+
 void __init time_init(void)
 {
 	u32 irq, i = 0;
@@ -289,4 +296,15 @@ void __init time_init(void)
 #endif
 	microblaze_clocksource_init();
 	microblaze_clockevent_init();
+	timer_initialized = 1;
+}
+
+unsigned long long notrace sched_clock(void)
+{
+	if (timer_initialized) {
+		struct clocksource *cs = &clocksource_microblaze;
+		cycle_t cyc = cnt32_to_63(cs->read(NULL));
+		return clocksource_cyc2ns(cyc, cs->mult, cs->shift);
+	}
+	return 0;
 }
diff --git a/arch/microblaze/kernel/traps.c b/arch/microblaze/kernel/traps.c
index 75e49202a5e..ba034d421ec 100644
--- a/arch/microblaze/kernel/traps.c
+++ b/arch/microblaze/kernel/traps.c
@@ -16,13 +16,14 @@
 
 #include <asm/exceptions.h>
 #include <asm/system.h>
+#include <asm/unwind.h>
 
 void trap_init(void)
 {
 	__enable_hw_exceptions();
 }
 
-static unsigned long kstack_depth_to_print = 24;
+static unsigned long kstack_depth_to_print;	/* 0 == entire stack */
 
 static int __init kstack_setup(char *s)
 {
@@ -30,31 +31,47 @@ static int __init kstack_setup(char *s)
 }
 __setup("kstack=", kstack_setup);
 
-void show_trace(struct task_struct *task, unsigned long *stack)
+void show_stack(struct task_struct *task, unsigned long *sp)
 {
-	unsigned long addr;
-
-	if (!stack)
-		stack = (unsigned long *)&stack;
+	unsigned long words_to_show;
+	u32 fp = (u32) sp;
+
+	if (fp == 0) {
+		if (task) {
+			fp = ((struct thread_info *)
+				(task->stack))->cpu_context.r1;
+		} else {
+			/* Pick up caller of dump_stack() */
+			fp = (u32)&sp - 8;
+		}
+	}
 
-	printk(KERN_NOTICE "Call Trace: ");
-#ifdef CONFIG_KALLSYMS
-	printk(KERN_NOTICE "\n");
-#endif
-	while (!kstack_end(stack)) {
-		addr = *stack++;
-		/*
-		 * If the address is either in the text segment of the
-		 * kernel, or in the region which contains vmalloc'ed
-		 * memory, it *may* be the address of a calling
-		 * routine; if so, print it so that someone tracing
-		 * down the cause of the crash will be able to figure
-		 * out the call path that was taken.
-		 */
-		if (kernel_text_address(addr))
-			print_ip_sym(addr);
+	words_to_show = (THREAD_SIZE - (fp & (THREAD_SIZE - 1))) >> 2;
+	if (kstack_depth_to_print && (words_to_show > kstack_depth_to_print))
+		words_to_show = kstack_depth_to_print;
+
+	pr_info("Kernel Stack:\n");
+
+	/*
+	 * Make the first line an 'odd' size if necessary to get
+	 * remaining lines to start at an address multiple of 0x10
+	 */
+	if (fp & 0xF) {
+		unsigned long line1_words = (0x10 - (fp & 0xF)) >> 2;
+		if (line1_words < words_to_show) {
+			print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 32,
+				       4, (void *)fp, line1_words << 2, 0);
+			fp += line1_words << 2;
+			words_to_show -= line1_words;
+		}
 	}
-	printk(KERN_NOTICE "\n");
+	print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 32, 4, (void *)fp,
+		       words_to_show << 2, 0);
+	printk(KERN_INFO "\n\n");
+
+	pr_info("Call Trace:\n");
+	microblaze_unwind(task, NULL);
+	pr_info("\n");
 
 	if (!task)
 		task = current;
@@ -62,34 +79,6 @@ void show_trace(struct task_struct *task, unsigned long *stack)
 	debug_show_held_locks(task);
 }
 
-void show_stack(struct task_struct *task, unsigned long *sp)
-{
-	unsigned long *stack;
-	int i;
-
-	if (sp == NULL) {
-		if (task)
-			sp = (unsigned long *) ((struct thread_info *)
-						(task->stack))->cpu_context.r1;
-		else
-			sp = (unsigned long *)&sp;
-	}
-
-	stack = sp;
-
-	printk(KERN_INFO "\nStack:\n  ");
-
-	for (i = 0; i < kstack_depth_to_print; i++) {
-		if (kstack_end(sp))
-			break;
-		if (i && ((i % 8) == 0))
-			printk("\n  ");
-		printk("%08lx ", *sp++);
-	}
-	printk("\n");
-	show_trace(task, stack);
-}
-
 void dump_stack(void)
 {
 	show_stack(NULL, NULL);
diff --git a/arch/microblaze/kernel/unwind.c b/arch/microblaze/kernel/unwind.c
new file mode 100644
index 00000000000..fefac5c3358
--- /dev/null
+++ b/arch/microblaze/kernel/unwind.c
@@ -0,0 +1,318 @@
+/*
+ * Backtrace support for Microblaze
+ *
+ * Copyright (C) 2010  Digital Design Corporation
+ *
+ * Based on arch/sh/kernel/cpu/sh5/unwind.c code which is:
+ * Copyright (C) 2004  Paul Mundt
+ * Copyright (C) 2004  Richard Curnow
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+
+/* #define DEBUG 1 */
+#include <linux/kallsyms.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/stacktrace.h>
+#include <linux/types.h>
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/io.h>
+#include <asm/sections.h>
+#include <asm/exceptions.h>
+#include <asm/unwind.h>
+
+struct stack_trace;
+
+/*
+ * On Microblaze, finding the previous stack frame is a little tricky.
+ * At this writing (3/2010), Microblaze does not support CONFIG_FRAME_POINTERS,
+ * and even if it did, gcc (4.1.2) does not store the frame pointer at
+ * a consistent offset within each frame. To determine frame size, it is
+ * necessary to search for the assembly instruction that creates or reclaims
+ * the frame and extract the size from it.
+ *
+ * Microblaze stores the stack pointer in r1, and creates a frame via
+ *
+ *     addik r1, r1, -FRAME_SIZE
+ *
+ * The frame is reclaimed via
+ *
+ *     addik r1, r1, FRAME_SIZE
+ *
+ * Frame creation occurs at or near the top of a function.
+ * Depending on the compiler, reclaim may occur at the end, or before
+ * a mid-function return.
+ *
+ * A stack frame is usually not created in a leaf function.
+ *
+ */
+
+/**
+ * get_frame_size - Extract the stack adjustment from an
+ *                  "addik r1, r1, adjust" instruction
+ * @instr : Microblaze instruction
+ *
+ * Return - Number of stack bytes the instruction reserves or reclaims
+ */
+inline long get_frame_size(unsigned long instr)
+{
+	return abs((s16)(instr & 0xFFFF));
+}
+
+/**
+ * find_frame_creation - Search backward to find the instruction that creates
+ *                       the stack frame (hopefully, for the same function the
+ *                       initial PC is in).
+ * @pc : Program counter at which to begin the search
+ *
+ * Return - PC at which stack frame creation occurs
+ *          NULL if this cannot be found, i.e. a leaf function
+ */
+static unsigned long *find_frame_creation(unsigned long *pc)
+{
+	int i;
+
+	/* NOTE: Distance to search is arbitrary
+	 *	 250 works well for most things,
+	 *	 750 picks up things like tcp_recvmsg(),
+	 *	1000 needed for fat_fill_super()
+	 */
+	for (i = 0; i < 1000; i++, pc--) {
+		unsigned long instr;
+		s16 frame_size;
+
+		if (!kernel_text_address((unsigned long) pc))
+			return NULL;
+
+		instr = *pc;
+
+		/* addik r1, r1, foo ? */
+		if ((instr & 0xFFFF0000) != 0x30210000)
+			continue;	/* No */
+
+		frame_size = get_frame_size(instr);
+		if ((frame_size < 8) || (frame_size & 3)) {
+			pr_debug("    Invalid frame size %d at 0x%p\n",
+				 frame_size, pc);
+			return NULL;
+		}
+
+		pr_debug("    Found frame creation at 0x%p, size %d\n", pc,
+			 frame_size);
+		return pc;
+	}
+
+	return NULL;
+}
+
+/**
+ * lookup_prev_stack_frame - Find the stack frame of the previous function.
+ * @fp          : Frame (stack) pointer for current function
+ * @pc          : Program counter within current function
+ * @leaf_return : r15 value within current function. If the current function
+ *		  is a leaf, this is the caller's return address.
+ * @pprev_fp    : On exit, set to frame (stack) pointer for previous function
+ * @pprev_pc    : On exit, set to current function caller's return address
+ *
+ * Return - 0 on success, -EINVAL if the previous frame cannot be found
+ */
+static int lookup_prev_stack_frame(unsigned long fp, unsigned long pc,
+				   unsigned long leaf_return,
+				   unsigned long *pprev_fp,
+				   unsigned long *pprev_pc)
+{
+	unsigned long *prologue = NULL;
+
+	/* _switch_to is a special leaf function */
+	if (pc != (unsigned long) &_switch_to)
+		prologue = find_frame_creation((unsigned long *)pc);
+
+	if (prologue) {
+		long frame_size = get_frame_size(*prologue);
+
+		*pprev_fp = fp + frame_size;
+		*pprev_pc = *(unsigned long *)fp;
+	} else {
+		if (!leaf_return)
+			return -EINVAL;
+		*pprev_pc = leaf_return;
+		*pprev_fp = fp;
+	}
+
+	/* NOTE: don't check kernel_text_address here, to allow display
+	 *	 of userland return address
+	 */
+	return (!*pprev_pc || (*pprev_pc & 3)) ? -EINVAL : 0;
+}
+
+static void microblaze_unwind_inner(struct task_struct *task,
+				    unsigned long pc, unsigned long fp,
+				    unsigned long leaf_return,
+				    struct stack_trace *trace);
+
+/**
+ * unwind_trap - Unwind through a system trap, that stored previous state
+ *		 on the stack.
+ */
+#ifdef CONFIG_MMU
+static inline void unwind_trap(struct task_struct *task, unsigned long pc,
+				unsigned long fp, struct stack_trace *trace)
+{
+	/* To be implemented */
+}
+#else
+static inline void unwind_trap(struct task_struct *task, unsigned long pc,
+				unsigned long fp, struct stack_trace *trace)
+{
+	const struct pt_regs *regs = (const struct pt_regs *) fp;
+	microblaze_unwind_inner(task, regs->pc, regs->r1, regs->r15, trace);
+}
+#endif
+
+/**
+ * microblaze_unwind_inner - Unwind the stack from the specified point
+ * @task  : Task whose stack we are to unwind (may be NULL)
+ * @pc    : Program counter from which we start unwinding
+ * @fp    : Frame (stack) pointer from which we start unwinding
+ * @leaf_return : Value of r15 at pc. If the function is a leaf, this is
+ *				  the caller's return address.
+ * @trace : Where to store stack backtrace (PC values).
+ *	    NULL == print backtrace to kernel log
+ */
+void microblaze_unwind_inner(struct task_struct *task,
+			     unsigned long pc, unsigned long fp,
+			     unsigned long leaf_return,
+			     struct stack_trace *trace)
+{
+	int ofs = 0;
+
+	pr_debug("    Unwinding with PC=%p, FP=%p\n", (void *)pc, (void *)fp);
+	if (!pc || !fp || (pc & 3) || (fp & 3)) {
+		pr_debug("    Invalid state for unwind, aborting\n");
+		return;
+	}
+	for (; pc != 0;) {
+		unsigned long next_fp, next_pc = 0;
+		unsigned long return_to = pc +  2 * sizeof(unsigned long);
+		const struct trap_handler_info *handler =
+			&microblaze_trap_handlers;
+
+		/* Is previous function the HW exception handler? */
+		if ((return_to >= (unsigned long)&_hw_exception_handler)
+		    &&(return_to < (unsigned long)&ex_handler_unhandled)) {
+			/*
+			 * HW exception handler doesn't save all registers,
+			 * so we open-code a special case of unwind_trap()
+			 */
+#ifndef CONFIG_MMU
+			const struct pt_regs *regs =
+				(const struct pt_regs *) fp;
+#endif
+			pr_info("HW EXCEPTION\n");
+#ifndef CONFIG_MMU
+			microblaze_unwind_inner(task, regs->r17 - 4,
+						fp + EX_HANDLER_STACK_SIZ,
+						regs->r15, trace);
+#endif
+			return;
+		}
+
+		/* Is previous function a trap handler? */
+		for (; handler->start_addr; ++handler) {
+			if ((return_to >= handler->start_addr)
+			    && (return_to <= handler->end_addr)) {
+				if (!trace)
+					pr_info("%s\n", handler->trap_name);
+				unwind_trap(task, pc, fp, trace);
+				return;
+			}
+		}
+		pc -= ofs;
+
+		if (trace) {
+#ifdef CONFIG_STACKTRACE
+			if (trace->skip > 0)
+				trace->skip--;
+			else
+				trace->entries[trace->nr_entries++] = pc;
+
+			if (trace->nr_entries >= trace->max_entries)
+				break;
+#endif
+		} else {
+			/* Have we reached userland? */
+			if (unlikely(pc == task_pt_regs(task)->pc)) {
+				pr_info("[<%p>] PID %lu [%s]\n",
+					(void *) pc,
+					(unsigned long) task->pid,
+					task->comm);
+				break;
+			} else
+				print_ip_sym(pc);
+		}
+
+		/* Stop when we reach anything not part of the kernel */
+		if (!kernel_text_address(pc))
+			break;
+
+		if (lookup_prev_stack_frame(fp, pc, leaf_return, &next_fp,
+					    &next_pc) == 0) {
+			ofs = sizeof(unsigned long);
+			pc = next_pc & ~3;
+			fp = next_fp;
+			leaf_return = 0;
+		} else {
+			pr_debug("    Failed to find previous stack frame\n");
+			break;
+		}
+
+		pr_debug("    Next PC=%p, next FP=%p\n",
+			 (void *)next_pc, (void *)next_fp);
+	}
+}
+
+/**
+ * microblaze_unwind - Stack unwinder for Microblaze (external entry point)
+ * @task  : Task whose stack we are to unwind (NULL == current)
+ * @trace : Where to store stack backtrace (PC values).
+ *	    NULL == print backtrace to kernel log
+ */
+void microblaze_unwind(struct task_struct *task, struct stack_trace *trace)
+{
+	if (task) {
+		if (task == current) {
+			const struct pt_regs *regs = task_pt_regs(task);
+			microblaze_unwind_inner(task, regs->pc, regs->r1,
+						regs->r15, trace);
+		} else {
+			struct thread_info *thread_info =
+				(struct thread_info *)(task->stack);
+			const struct cpu_context *cpu_context =
+				&thread_info->cpu_context;
+
+			microblaze_unwind_inner(task,
+						(unsigned long) &_switch_to,
+						cpu_context->r1,
+						cpu_context->r15, trace);
+		}
+	} else {
+		unsigned long pc, fp;
+
+		__asm__ __volatile__ ("or %0, r1, r0" : "=r" (fp));
+
+		__asm__ __volatile__ (
+			"brlid %0, 0f;"
+			"nop;"
+			"0:"
+			: "=r" (pc)
+		);
+
+		/* Since we are not a leaf function, use leaf_return = 0 */
+		microblaze_unwind_inner(current, pc, fp, 0, trace);
+	}
+}
+
diff --git a/arch/microblaze/kernel/vmlinux.lds.S b/arch/microblaze/kernel/vmlinux.lds.S
index db72d712460..a09f2962fbe 100644
--- a/arch/microblaze/kernel/vmlinux.lds.S
+++ b/arch/microblaze/kernel/vmlinux.lds.S
@@ -10,7 +10,7 @@
 
 OUTPUT_FORMAT("elf32-microblaze", "elf32-microblaze", "elf32-microblaze")
 OUTPUT_ARCH(microblaze)
-ENTRY(_start)
+ENTRY(microblaze_start)
 
 #include <asm/page.h>
 #include <asm-generic/vmlinux.lds.h>
@@ -20,7 +20,7 @@ jiffies = jiffies_64 + 4;
 
 SECTIONS {
 	. = CONFIG_KERNEL_START;
-	_start = CONFIG_KERNEL_BASE_ADDR;
+	microblaze_start = CONFIG_KERNEL_BASE_ADDR;
 	.text : AT(ADDR(.text) - LOAD_OFFSET) {
 		_text = . ;
 		_stext = . ;
@@ -55,7 +55,7 @@ SECTIONS {
 	 */
 	.sdata2 : AT(ADDR(.sdata2) - LOAD_OFFSET) {
 		_ssrw = .;
-		. = ALIGN(4096); /* page aligned when MMU used - origin 0x8 */
+		. = ALIGN(PAGE_SIZE); /* page aligned when MMU used */
 		*(.sdata2)
 	. = ALIGN(8);
 	_essrw = .;
@@ -70,7 +70,7 @@ SECTIONS {
 	/* Reserve some low RAM for r0 based memory references */
 	. = ALIGN(0x4) ;
 	r0_ram = . ;
-	. = . +  4096;	/* a page should be enough */
+	. = . +  PAGE_SIZE;	/* a page should be enough */
 
 	/* Under the microblaze ABI, .sdata and .sbss must be contiguous */
 	. = ALIGN(8);
@@ -120,7 +120,7 @@ SECTIONS {
 
 	__init_end_before_initramfs = .;
 
-	.init.ramfs ALIGN(4096) : AT(ADDR(.init.ramfs) - LOAD_OFFSET) {
+	.init.ramfs ALIGN(PAGE_SIZE) : AT(ADDR(.init.ramfs) - LOAD_OFFSET) {
 		__initramfs_start = .;
 		*(.init.ramfs)
 		__initramfs_end = .;
@@ -132,11 +132,11 @@ SECTIONS {
  * so that __init_end == __bss_start. This will make image.elf
  * consistent with the image.bin
  */
-		/* . = ALIGN(4096); */
+		/* . = ALIGN(PAGE_SIZE); */
 	}
 	__init_end = .;
 
-	.bss ALIGN (4096) : AT(ADDR(.bss) - LOAD_OFFSET) {
+	.bss ALIGN (PAGE_SIZE) : AT(ADDR(.bss) - LOAD_OFFSET) {
 		/* page aligned when MMU used */
 		__bss_start = . ;
 			*(.bss*)
@@ -145,7 +145,7 @@ SECTIONS {
 		__bss_stop = . ;
 		_ebss = . ;
 	}
-	. = ALIGN(4096);
+	. = ALIGN(PAGE_SIZE);
 	_end = .;
 
 	DISCARDS
diff --git a/arch/microblaze/mm/fault.c b/arch/microblaze/mm/fault.c
index bab92299318..57bd2a09610 100644
--- a/arch/microblaze/mm/fault.c
+++ b/arch/microblaze/mm/fault.c
@@ -37,10 +37,6 @@
 #include <linux/uaccess.h>
 #include <asm/exceptions.h>
 
-#if defined(CONFIG_KGDB)
-int debugger_kernel_faults = 1;
-#endif
-
 static unsigned long pte_misses;	/* updated by do_page_fault() */
 static unsigned long pte_errors;	/* updated by do_page_fault() */
 
@@ -81,10 +77,6 @@ void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 	}
 
 	/* kernel has accessed a bad area */
-#if defined(CONFIG_KGDB)
-	if (debugger_kernel_faults)
-		debugger(regs);
-#endif
 	die("kernel access of bad area", regs, sig);
 }
 
@@ -115,13 +107,6 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 	if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
 		is_write = 0;
 
-#if defined(CONFIG_KGDB)
-	if (debugger_fault_handler && regs->trap == 0x300) {
-		debugger_fault_handler(regs);
-		return;
-	}
-#endif /* CONFIG_KGDB */
-
 	if (unlikely(in_atomic() || !mm)) {
 		if (kernel_mode(regs))
 			goto bad_area_nosemaphore;
@@ -226,7 +211,6 @@ good_area:
 	 * make sure we exit gracefully rather than endlessly redo
 	 * the fault.
 	 */
-survive:
 	fault = handle_mm_fault(mm, vma, address, is_write ? FAULT_FLAG_WRITE : 0);
 	if (unlikely(fault & VM_FAULT_ERROR)) {
 		if (fault & VM_FAULT_OOM)
diff --git a/arch/microblaze/mm/init.c b/arch/microblaze/mm/init.c
index db593498992..65eb00419d1 100644
--- a/arch/microblaze/mm/init.c
+++ b/arch/microblaze/mm/init.c
@@ -134,13 +134,8 @@ void __init setup_memory(void)
 	 * for 4GB of memory, using 4kB pages), plus 1 page
 	 * (in case the address isn't page-aligned).
 	 */
-#ifndef CONFIG_MMU
-	map_size = init_bootmem_node(NODE_DATA(0), PFN_UP(TOPHYS((u32)klimit)),
-					min_low_pfn, max_low_pfn);
-#else
-	map_size = init_bootmem_node(&contig_page_data,
+	map_size = init_bootmem_node(NODE_DATA(0),
 		PFN_UP(TOPHYS((u32)klimit)), min_low_pfn, max_low_pfn);
-#endif
 	memblock_reserve(PFN_UP(TOPHYS((u32)klimit)) << PAGE_SHIFT, map_size);
 
 	/* free bootmem is whole main memory */