Merge branch 'master' into sh/hwblk

16 files changed, 1387 insertions, 60 deletions

diff --git a/arch/sh/kernel/Makefile_32 b/arch/sh/kernel/Makefile_32
index 94ed99b6800..f2245ebf0b3 100644
--- a/arch/sh/kernel/Makefile_32
+++ b/arch/sh/kernel/Makefile_32
@@ -9,10 +9,10 @@ ifdef CONFIG_FUNCTION_TRACER
 CFLAGS_REMOVE_ftrace.o = -pg
 endif
 
-obj-y	:= debugtraps.o idle.o io.o io_generic.o irq.o			\
+obj-y	:= debugtraps.o dumpstack.o idle.o io.o io_generic.o irq.o	\
 	   machvec.o process_32.o ptrace_32.o setup.o signal_32.o	\
-	   sys_sh.o sys_sh32.o syscalls_32.o time.o topology.o	\
-	   traps.o traps_32.o
+	   sys_sh.o sys_sh32.o syscalls_32.o time.o topology.o		\
+	   traps.o traps_32.o unwinder.o
 
 obj-y				+= cpu/
 obj-$(CONFIG_VSYSCALL)		+= vsyscall/
@@ -33,6 +33,7 @@ obj-$(CONFIG_FTRACE_SYSCALLS)	+= ftrace.o
 obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o
 obj-$(CONFIG_DUMP_CODE)		+= disassemble.o
 obj-$(CONFIG_HIBERNATION)	+= swsusp.o
+obj-$(CONFIG_DWARF_UNWINDER)	+= dwarf.o
 
 obj-$(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST)	+= localtimer.o
 
diff --git a/arch/sh/kernel/Makefile_64 b/arch/sh/kernel/Makefile_64
index 67b9f6c6326..639ee514266 100644
--- a/arch/sh/kernel/Makefile_64
+++ b/arch/sh/kernel/Makefile_64
@@ -2,7 +2,7 @@ extra-y	:= head_64.o init_task.o vmlinux.lds
 
 obj-y	:= debugtraps.o idle.o io.o io_generic.o irq.o machvec.o process_64.o \
 	   ptrace_64.o setup.o signal_64.o sys_sh.o sys_sh64.o \
-	   syscalls_64.o time.o topology.o traps.o traps_64.o
+	   syscalls_64.o time.o topology.o traps.o traps_64.o unwinder.o
 
 obj-y				+= cpu/
 obj-$(CONFIG_SMP)		+= smp.o
@@ -13,6 +13,7 @@ obj-$(CONFIG_CRASH_DUMP)	+= crash_dump.o
 obj-$(CONFIG_STACKTRACE)	+= stacktrace.o
 obj-$(CONFIG_IO_TRAPPED)	+= io_trapped.o
 obj-$(CONFIG_GENERIC_GPIO)	+= gpio.o
+obj-$(CONFIG_DWARF_UNWINDER)	+= dwarf.o
 
 obj-$(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST)	+= localtimer.o
 
diff --git a/arch/sh/kernel/cpu/init.c b/arch/sh/kernel/cpu/init.c
index ad85421099c..d40b9db5be0 100644
--- a/arch/sh/kernel/cpu/init.c
+++ b/arch/sh/kernel/cpu/init.c
@@ -3,7 +3,7 @@
  *
  * CPU init code
  *
- * Copyright (C) 2002 - 2007  Paul Mundt
+ * Copyright (C) 2002 - 2009  Paul Mundt
  * Copyright (C) 2003  Richard Curnow
  *
  * This file is subject to the terms and conditions of the GNU General Public
@@ -62,6 +62,37 @@ static void __init speculative_execution_init(void)
 #define speculative_execution_init()	do { } while (0)
 #endif
 
+#ifdef CONFIG_CPU_SH4A
+#define EXPMASK			0xff2f0004
+#define EXPMASK_RTEDS		(1 << 0)
+#define EXPMASK_BRDSSLP		(1 << 1)
+#define EXPMASK_MMCAW		(1 << 4)
+
+static void __init expmask_init(void)
+{
+	unsigned long expmask = __raw_readl(EXPMASK);
+
+	/*
+	 * Future proofing.
+	 *
+	 * Disable support for slottable sleep instruction
+	 * and non-nop instructions in the rte delay slot.
+	 */
+	expmask &= ~(EXPMASK_RTEDS | EXPMASK_BRDSSLP);
+
+	/*
+	 * Enable associative writes to the memory-mapped cache array
+	 * until the cache flush ops have been rewritten.
+	 */
+	expmask |= EXPMASK_MMCAW;
+
+	__raw_writel(expmask, EXPMASK);
+	ctrl_barrier();
+}
+#else
+#define expmask_init()	do { } while (0)
+#endif
+
 /* 2nd-level cache init */
 void __uses_jump_to_uncached __attribute__ ((weak)) l2_cache_init(void)
 {
@@ -321,4 +352,5 @@ asmlinkage void __init sh_cpu_init(void)
 #endif
 
 	speculative_execution_init();
+	expmask_init();
 }
diff --git a/arch/sh/kernel/cpu/sh2/entry.S b/arch/sh/kernel/cpu/sh2/entry.S
index becc54c4569..c8a4331d9b8 100644
--- a/arch/sh/kernel/cpu/sh2/entry.S
+++ b/arch/sh/kernel/cpu/sh2/entry.S
@@ -227,8 +227,9 @@ ENTRY(sh_bios_handler)
 	mov.l	@r15+, r14
 	add	#8,r15
 	lds.l	@r15+, pr
+	mov.l	@r15+,r15
 	rte
-	 mov.l	@r15+,r15
+	 nop
 	.align	2
 1:	.long	gdb_vbr_vector
 #endif /* CONFIG_SH_STANDARD_BIOS */
diff --git a/arch/sh/kernel/cpu/sh2a/entry.S b/arch/sh/kernel/cpu/sh2a/entry.S
index ab3903eeda5..222742ddc0d 100644
--- a/arch/sh/kernel/cpu/sh2a/entry.S
+++ b/arch/sh/kernel/cpu/sh2a/entry.S
@@ -176,8 +176,9 @@ ENTRY(sh_bios_handler)
 	movml.l	@r15+,r14
 	add	#8,r15
 	lds.l	@r15+, pr
+	mov.l	@r15+,r15
 	rte
-	 mov.l	@r15+,r15
+	 nop
 	.align	2
 1:	.long	gdb_vbr_vector
 #endif /* CONFIG_SH_STANDARD_BIOS */
diff --git a/arch/sh/kernel/cpu/sh3/entry.S b/arch/sh/kernel/cpu/sh3/entry.S
index 3cb531f233f..67ad6467c69 100644
--- a/arch/sh/kernel/cpu/sh3/entry.S
+++ b/arch/sh/kernel/cpu/sh3/entry.S
@@ -137,6 +137,7 @@ ENTRY(tlb_protection_violation_store)
 	 mov	#1, r5
 
 call_dpf:
+	setup_frame_reg
 	mov.l	1f, r0
 	mov	r5, r8
 	mov.l	@r0, r6
diff --git a/arch/sh/kernel/cpu/shmobile/sleep.S b/arch/sh/kernel/cpu/shmobile/sleep.S
index 5d888ef53d8..baf2d7d46b0 100644
--- a/arch/sh/kernel/cpu/shmobile/sleep.S
+++ b/arch/sh/kernel/cpu/shmobile/sleep.S
@@ -26,8 +26,30 @@ ENTRY(sh_mobile_standby)
 
 	tst	#SUSP_SH_SF, r0
 	bt	skip_set_sf
+#ifdef CONFIG_CPU_SUBTYPE_SH7724
+	/* DBSC: put memory in self-refresh mode */
 
-	/* SDRAM: disable power down and put in self-refresh mode */
+	mov.l	dben_reg, r4
+	mov.l	dben_data0, r1
+	mov.l	r1, @r4
+
+	mov.l	dbrfpdn0_reg, r4
+	mov.l	dbrfpdn0_data0, r1
+	mov.l	r1, @r4
+
+	mov.l	dbcmdcnt_reg, r4
+	mov.l	dbcmdcnt_data0, r1
+	mov.l	r1, @r4
+
+	mov.l	dbcmdcnt_reg, r4
+	mov.l	dbcmdcnt_data1, r1
+	mov.l	r1, @r4
+
+	mov.l	dbrfpdn0_reg, r4
+	mov.l	dbrfpdn0_data1, r1
+	mov.l	r1, @r4
+#else
+	/* SBSC: disable power down and put in self-refresh mode */
 	mov.l	1f, r4
 	mov.l	2f, r1
 	mov.l	@r4, r2
@@ -35,6 +57,7 @@ ENTRY(sh_mobile_standby)
 	mov.l   3f, r3
 	and	r3, r2
 	mov.l	r2, @r4
+#endif
 
 skip_set_sf:
 	tst	#SUSP_SH_SLEEP, r0
@@ -84,7 +107,36 @@ done_sleep:
 	tst	#SUSP_SH_SF, r0
 	bt	skip_restore_sf
 
-	/* SDRAM: set auto-refresh mode */
+#ifdef CONFIG_CPU_SUBTYPE_SH7724
+	/* DBSC: put memory in auto-refresh mode */
+
+	mov.l	dbrfpdn0_reg, r4
+	mov.l	dbrfpdn0_data0, r1
+	mov.l	r1, @r4
+
+	/* sleep 140 ns */
+	nop
+	nop
+	nop
+	nop
+
+	mov.l	dbcmdcnt_reg, r4
+	mov.l	dbcmdcnt_data0, r1
+	mov.l	r1, @r4
+
+	mov.l	dbcmdcnt_reg, r4
+	mov.l	dbcmdcnt_data1, r1
+	mov.l	r1, @r4
+
+	mov.l	dben_reg, r4
+	mov.l	dben_data1, r1
+	mov.l	r1, @r4
+
+	mov.l	dbrfpdn0_reg, r4
+	mov.l	dbrfpdn0_data2, r1
+	mov.l	r1, @r4
+#else
+	/* SBSC: set auto-refresh mode */
 	mov.l	1f, r4
 	mov.l	@r4, r2
 	mov.l   4f, r3
@@ -98,15 +150,29 @@ done_sleep:
 	add	r4, r3
 	or	r2, r3
 	mov.l	r3, @r1
+#endif
 skip_restore_sf:
 	rts
 	 nop
 
 	.balign 4
+#ifdef CONFIG_CPU_SUBTYPE_SH7724
+dben_reg:	.long	0xfd000010 /* DBEN */
+dben_data0:	.long	0
+dben_data1:	.long	1
+dbrfpdn0_reg:	.long	0xfd000040 /* DBRFPDN0 */
+dbrfpdn0_data0:	.long	0
+dbrfpdn0_data1:	.long	1
+dbrfpdn0_data2:	.long	0x00010000
+dbcmdcnt_reg:	.long	0xfd000014 /* DBCMDCNT */
+dbcmdcnt_data0:	.long	2
+dbcmdcnt_data1:	.long	4
+#else
 1:	.long	0xfe400008 /* SDCR0 */
 2:	.long	0x00000400
 3:	.long	0xffff7fff
 4:	.long	0xfffffbff
+#endif
 5:	.long	0xa4150020 /* STBCR */
 6:	.long   0xfe40001c /* RTCOR */
 7:	.long   0xfe400018 /* RTCNT */
diff --git a/arch/sh/kernel/dumpstack.c b/arch/sh/kernel/dumpstack.c
new file mode 100644
index 00000000000..6f5ad151340
--- /dev/null
+++ b/arch/sh/kernel/dumpstack.c
@@ -0,0 +1,123 @@
+/*
+ *  Copyright (C) 1991, 1992  Linus Torvalds
+ *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
+ *  Copyright (C) 2009  Matt Fleming
+ */
+#include <linux/kallsyms.h>
+#include <linux/ftrace.h>
+#include <linux/debug_locks.h>
+#include <asm/unwinder.h>
+#include <asm/stacktrace.h>
+
+void printk_address(unsigned long address, int reliable)
+{
+	printk(" [<%p>] %s%pS\n", (void *) address,
+			reliable ? "" : "? ", (void *) address);
+}
+
+#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+static void
+print_ftrace_graph_addr(unsigned long addr, void *data,
+			const struct stacktrace_ops *ops,
+			struct thread_info *tinfo, int *graph)
+{
+	struct task_struct *task = tinfo->task;
+	unsigned long ret_addr;
+	int index = task->curr_ret_stack;
+
+	if (addr != (unsigned long)return_to_handler)
+		return;
+
+	if (!task->ret_stack || index < *graph)
+		return;
+
+	index -= *graph;
+	ret_addr = task->ret_stack[index].ret;
+
+	ops->address(data, ret_addr, 1);
+
+	(*graph)++;
+}
+#else
+static inline void
+print_ftrace_graph_addr(unsigned long addr, void *data,
+			const struct stacktrace_ops *ops,
+			struct thread_info *tinfo, int *graph)
+{ }
+#endif
+
+void
+stack_reader_dump(struct task_struct *task, struct pt_regs *regs,
+		  unsigned long *sp, const struct stacktrace_ops *ops,
+		  void *data)
+{
+	struct thread_info *context;
+	int graph = 0;
+
+	context = (struct thread_info *)
+		((unsigned long)sp & (~(THREAD_SIZE - 1)));
+
+	while (!kstack_end(sp)) {
+		unsigned long addr = *sp++;
+
+		if (__kernel_text_address(addr)) {
+			ops->address(data, addr, 1);
+
+			print_ftrace_graph_addr(addr, data, ops,
+						context, &graph);
+		}
+	}
+}
+
+static void
+print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
+{
+	printk(data);
+	print_symbol(msg, symbol);
+	printk("\n");
+}
+
+static void print_trace_warning(void *data, char *msg)
+{
+	printk("%s%s\n", (char *)data, msg);
+}
+
+static int print_trace_stack(void *data, char *name)
+{
+	printk("%s <%s> ", (char *)data, name);
+	return 0;
+}
+
+/*
+ * Print one address/symbol entries per line.
+ */
+static void print_trace_address(void *data, unsigned long addr, int reliable)
+{
+	printk(data);
+	printk_address(addr, reliable);
+}
+
+static const struct stacktrace_ops print_trace_ops = {
+	.warning = print_trace_warning,
+	.warning_symbol = print_trace_warning_symbol,
+	.stack = print_trace_stack,
+	.address = print_trace_address,
+};
+
+void show_trace(struct task_struct *tsk, unsigned long *sp,
+		struct pt_regs *regs)
+{
+	if (regs && user_mode(regs))
+		return;
+
+	printk("\nCall trace:\n");
+
+	unwind_stack(tsk, regs, sp, &print_trace_ops, "");
+
+	printk("\n");
+
+	if (!tsk)
+		tsk = current;
+
+	debug_show_held_locks(tsk);
+}
diff --git a/arch/sh/kernel/dwarf.c b/arch/sh/kernel/dwarf.c
new file mode 100644
index 00000000000..c6c5764a8ab
--- /dev/null
+++ b/arch/sh/kernel/dwarf.c
@@ -0,0 +1,902 @@
+/*
+ * Copyright (C) 2009 Matt Fleming <matt@console-pimps.org>
+ *
+ * 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.
+ *
+ * This is an implementation of a DWARF unwinder. Its main purpose is
+ * for generating stacktrace information. Based on the DWARF 3
+ * specification from http://www.dwarfstd.org.
+ *
+ * TODO:
+ *	- DWARF64 doesn't work.
+ */
+
+/* #define DEBUG */
+#include <linux/kernel.h>
+#include <linux/io.h>
+#include <linux/list.h>
+#include <linux/mm.h>
+#include <asm/dwarf.h>
+#include <asm/unwinder.h>
+#include <asm/sections.h>
+#include <asm/unaligned.h>
+#include <asm/dwarf.h>
+#include <asm/stacktrace.h>
+
+static LIST_HEAD(dwarf_cie_list);
+DEFINE_SPINLOCK(dwarf_cie_lock);
+
+static LIST_HEAD(dwarf_fde_list);
+DEFINE_SPINLOCK(dwarf_fde_lock);
+
+static struct dwarf_cie *cached_cie;
+
+/*
+ * Figure out whether we need to allocate some dwarf registers. If dwarf
+ * registers have already been allocated then we may need to realloc
+ * them. "reg" is a register number that we need to be able to access
+ * after this call.
+ *
+ * Register numbers start at zero, therefore we need to allocate space
+ * for "reg" + 1 registers.
+ */
+static void dwarf_frame_alloc_regs(struct dwarf_frame *frame,
+				   unsigned int reg)
+{
+	struct dwarf_reg *regs;
+	unsigned int num_regs = reg + 1;
+	size_t new_size;
+	size_t old_size;
+
+	new_size = num_regs * sizeof(*regs);
+	old_size = frame->num_regs * sizeof(*regs);
+
+	/* Fast path: don't allocate any regs if we've already got enough. */
+	if (frame->num_regs >= num_regs)
+		return;
+
+	regs = kzalloc(new_size, GFP_ATOMIC);
+	if (!regs) {
+		printk(KERN_WARNING "Unable to allocate DWARF registers\n");
+		/*
+		 * Let's just bomb hard here, we have no way to
+		 * gracefully recover.
+		 */
+		BUG();
+	}
+
+	if (frame->regs) {
+		memcpy(regs, frame->regs, old_size);
+		kfree(frame->regs);
+	}
+
+	frame->regs = regs;
+	frame->num_regs = num_regs;
+}
+
+/**
+ *	dwarf_read_addr - read dwarf data
+ *	@src: source address of data
+ *	@dst: destination address to store the data to
+ *
+ *	Read 'n' bytes from @src, where 'n' is the size of an address on
+ *	the native machine. We return the number of bytes read, which
+ *	should always be 'n'. We also have to be careful when reading
+ *	from @src and writing to @dst, because they can be arbitrarily
+ *	aligned. Return 'n' - the number of bytes read.
+ */
+static inline int dwarf_read_addr(unsigned long *src, unsigned long *dst)
+{
+	u32 val = get_unaligned(src);
+	put_unaligned(val, dst);
+	return sizeof(unsigned long *);
+}
+
+/**
+ *	dwarf_read_uleb128 - read unsigned LEB128 data
+ *	@addr: the address where the ULEB128 data is stored
+ *	@ret: address to store the result
+ *
+ *	Decode an unsigned LEB128 encoded datum. The algorithm is taken
+ *	from Appendix C of the DWARF 3 spec. For information on the
+ *	encodings refer to section "7.6 - Variable Length Data". Return
+ *	the number of bytes read.
+ */
+static inline unsigned long dwarf_read_uleb128(char *addr, unsigned int *ret)
+{
+	unsigned int result;
+	unsigned char byte;
+	int shift, count;
+
+	result = 0;
+	shift = 0;
+	count = 0;
+
+	while (1) {
+		byte = __raw_readb(addr);
+		addr++;
+		count++;
+
+		result |= (byte & 0x7f) << shift;
+		shift += 7;
+
+		if (!(byte & 0x80))
+			break;
+	}
+
+	*ret = result;
+
+	return count;
+}
+
+/**
+ *	dwarf_read_leb128 - read signed LEB128 data
+ *	@addr: the address of the LEB128 encoded data
+ *	@ret: address to store the result
+ *
+ *	Decode signed LEB128 data. The algorithm is taken from Appendix
+ *	C of the DWARF 3 spec. Return the number of bytes read.
+ */
+static inline unsigned long dwarf_read_leb128(char *addr, int *ret)
+{
+	unsigned char byte;
+	int result, shift;
+	int num_bits;
+	int count;
+
+	result = 0;
+	shift = 0;
+	count = 0;
+
+	while (1) {
+		byte = __raw_readb(addr);
+		addr++;
+		result |= (byte & 0x7f) << shift;
+		shift += 7;
+		count++;
+
+		if (!(byte & 0x80))
+			break;
+	}
+
+	/* The number of bits in a signed integer. */
+	num_bits = 8 * sizeof(result);
+
+	if ((shift < num_bits) && (byte & 0x40))
+		result |= (-1 << shift);
+
+	*ret = result;
+
+	return count;
+}
+
+/**
+ *	dwarf_read_encoded_value - return the decoded value at @addr
+ *	@addr: the address of the encoded value
+ *	@val: where to write the decoded value
+ *	@encoding: the encoding with which we can decode @addr
+ *
+ *	GCC emits encoded address in the .eh_frame FDE entries. Decode
+ *	the value at @addr using @encoding. The decoded value is written
+ *	to @val and the number of bytes read is returned.
+ */
+static int dwarf_read_encoded_value(char *addr, unsigned long *val,
+				    char encoding)
+{
+	unsigned long decoded_addr = 0;
+	int count = 0;
+
+	switch (encoding & 0x70) {
+	case DW_EH_PE_absptr:
+		break;
+	case DW_EH_PE_pcrel:
+		decoded_addr = (unsigned long)addr;
+		break;
+	default:
+		pr_debug("encoding=0x%x\n", (encoding & 0x70));
+		BUG();
+	}
+
+	if ((encoding & 0x07) == 0x00)
+		encoding |= DW_EH_PE_udata4;
+
+	switch (encoding & 0x0f) {
+	case DW_EH_PE_sdata4:
+	case DW_EH_PE_udata4:
+		count += 4;
+		decoded_addr += get_unaligned((u32 *)addr);
+		__raw_writel(decoded_addr, val);
+		break;
+	default:
+		pr_debug("encoding=0x%x\n", encoding);
+		BUG();
+	}
+
+	return count;
+}
+
+/**
+ *	dwarf_entry_len - return the length of an FDE or CIE
+ *	@addr: the address of the entry
+ *	@len: the length of the entry
+ *
+ *	Read the initial_length field of the entry and store the size of
+ *	the entry in @len. We return the number of bytes read. Return a
+ *	count of 0 on error.
+ */
+static inline int dwarf_entry_len(char *addr, unsigned long *len)
+{
+	u32 initial_len;
+	int count;
+
+	initial_len = get_unaligned((u32 *)addr);
+	count = 4;
+
+	/*
+	 * An initial length field value in the range DW_LEN_EXT_LO -
+	 * DW_LEN_EXT_HI indicates an extension, and should not be
+	 * interpreted as a length. The only extension that we currently
+	 * understand is the use of DWARF64 addresses.
+	 */
+	if (initial_len >= DW_EXT_LO && initial_len <= DW_EXT_HI) {
+		/*
+		 * The 64-bit length field immediately follows the
+		 * compulsory 32-bit length field.
+		 */
+		if (initial_len == DW_EXT_DWARF64) {
+			*len = get_unaligned((u64 *)addr + 4);
+			count = 12;
+		} else {
+			printk(KERN_WARNING "Unknown DWARF extension\n");
+			count = 0;
+		}
+	} else
+		*len = initial_len;
+
+	return count;
+}
+
+/**
+ *	dwarf_lookup_cie - locate the cie
+ *	@cie_ptr: pointer to help with lookup
+ */
+static struct dwarf_cie *dwarf_lookup_cie(unsigned long cie_ptr)
+{
+	struct dwarf_cie *cie, *n;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dwarf_cie_lock, flags);
+
+	/*
+	 * We've cached the last CIE we looked up because chances are
+	 * that the FDE wants this CIE.
+	 */
+	if (cached_cie && cached_cie->cie_pointer == cie_ptr) {
+		cie = cached_cie;
+		goto out;
+	}
+
+	list_for_each_entry_safe(cie, n, &dwarf_cie_list, link) {
+		if (cie->cie_pointer == cie_ptr) {
+			cached_cie = cie;
+			break;
+		}
+	}
+
+	/* Couldn't find the entry in the list. */
+	if (&cie->link == &dwarf_cie_list)
+		cie = NULL;
+out:
+	spin_unlock_irqrestore(&dwarf_cie_lock, flags);
+	return cie;
+}
+
+/**
+ *	dwarf_lookup_fde - locate the FDE that covers pc
+ *	@pc: the program counter
+ */
+struct dwarf_fde *dwarf_lookup_fde(unsigned long pc)
+{
+	unsigned long flags;
+	struct dwarf_fde *fde, *n;
+
+	spin_lock_irqsave(&dwarf_fde_lock, flags);
+	list_for_each_entry_safe(fde, n, &dwarf_fde_list, link) {
+		unsigned long start, end;
+
+		start = fde->initial_location;
+		end = fde->initial_location + fde->address_range;
+
+		if (pc >= start && pc < end)
+			break;
+	}
+
+	/* Couldn't find the entry in the list. */
+	if (&fde->link == &dwarf_fde_list)
+		fde = NULL;
+
+	spin_unlock_irqrestore(&dwarf_fde_lock, flags);
+
+	return fde;
+}
+
+/**
+ *	dwarf_cfa_execute_insns - execute instructions to calculate a CFA
+ *	@insn_start: address of the first instruction
+ *	@insn_end: address of the last instruction
+ *	@cie: the CIE for this function
+ *	@fde: the FDE for this function
+ *	@frame: the instructions calculate the CFA for this frame
+ *	@pc: the program counter of the address we're interested in
+ *	@define_ra: keep executing insns until the return addr reg is defined?
+ *
+ *	Execute the Call Frame instruction sequence starting at
+ *	@insn_start and ending at @insn_end. The instructions describe
+ *	how to calculate the Canonical Frame Address of a stackframe.
+ *	Store the results in @frame.
+ */
+static int dwarf_cfa_execute_insns(unsigned char *insn_start,
+				   unsigned char *insn_end,
+				   struct dwarf_cie *cie,
+				   struct dwarf_fde *fde,
+				   struct dwarf_frame *frame,
+				   unsigned long pc,
+				   bool define_ra)
+{
+	unsigned char insn;
+	unsigned char *current_insn;
+	unsigned int count, delta, reg, expr_len, offset;
+	bool seen_ra_reg;
+
+	current_insn = insn_start;
+
+	/*
+	 * If we're executing instructions for the dwarf_unwind_stack()
+	 * FDE we need to keep executing instructions until the value of
+	 * DWARF_ARCH_RA_REG is defined. See the comment in
+	 * dwarf_unwind_stack() for more details.
+	 */
+	if (define_ra)
+		seen_ra_reg = false;
+	else
+		seen_ra_reg = true;
+
+	while (current_insn < insn_end && (frame->pc <= pc || !seen_ra_reg) ) {
+		insn = __raw_readb(current_insn++);
+
+		if (!seen_ra_reg) {
+			if (frame->num_regs >= DWARF_ARCH_RA_REG &&
+			    frame->regs[DWARF_ARCH_RA_REG].flags)
+				seen_ra_reg = true;
+		}
+
+		/*
+		 * Firstly, handle the opcodes that embed their operands
+		 * in the instructions.
+		 */
+		switch (DW_CFA_opcode(insn)) {
+		case DW_CFA_advance_loc:
+			delta = DW_CFA_operand(insn);
+			delta *= cie->code_alignment_factor;
+			frame->pc += delta;
+			continue;
+			/* NOTREACHED */
+		case DW_CFA_offset:
+			reg = DW_CFA_operand(insn);
+			count = dwarf_read_uleb128(current_insn, &offset);
+			current_insn += count;
+			offset *= cie->data_alignment_factor;
+			dwarf_frame_alloc_regs(frame, reg);
+			frame->regs[reg].addr = offset;
+			frame->regs[reg].flags |= DWARF_REG_OFFSET;
+			continue;
+			/* NOTREACHED */
+		case DW_CFA_restore:
+			reg = DW_CFA_operand(insn);
+			continue;
+			/* NOTREACHED */
+		}
+
+		/*
+		 * Secondly, handle the opcodes that don't embed their
+		 * operands in the instruction.
+		 */
+		switch (insn) {
+		case DW_CFA_nop:
+			continue;
+		case DW_CFA_advance_loc1:
+			delta = *current_insn++;
+			frame->pc += delta * cie->code_alignment_factor;
+			break;
+		case DW_CFA_advance_loc2:
+			delta = get_unaligned((u16 *)current_insn);
+			current_insn += 2;
+			frame->pc += delta * cie->code_alignment_factor;
+			break;
+		case DW_CFA_advance_loc4:
+			delta = get_unaligned((u32 *)current_insn);
+			current_insn += 4;
+			frame->pc += delta * cie->code_alignment_factor;
+			break;
+		case DW_CFA_offset_extended:
+			count = dwarf_read_uleb128(current_insn, &reg);
+			current_insn += count;
+			count = dwarf_read_uleb128(current_insn, &offset);
+			current_insn += count;
+			offset *= cie->data_alignment_factor;
+			break;
+		case DW_CFA_restore_extended:
+			count = dwarf_read_uleb128(current_insn, &reg);
+			current_insn += count;
+			break;
+		case DW_CFA_undefined:
+			count = dwarf_read_uleb128(current_insn, &reg);
+			current_insn += count;
+			break;
+		case DW_CFA_def_cfa:
+			count = dwarf_read_uleb128(current_insn,
+						   &frame->cfa_register);
+			current_insn += count;
+			count = dwarf_read_uleb128(current_insn,
+						   &frame->cfa_offset);
+			current_insn += count;
+
+			frame->flags |= DWARF_FRAME_CFA_REG_OFFSET;
+			break;
+		case DW_CFA_def_cfa_register:
+			count = dwarf_read_uleb128(current_insn,
+						   &frame->cfa_register);
+			current_insn += count;
+			frame->cfa_offset = 0;
+			frame->flags |= DWARF_FRAME_CFA_REG_OFFSET;
+			break;
+		case DW_CFA_def_cfa_offset:
+			count = dwarf_read_uleb128(current_insn, &offset);
+			current_insn += count;
+			frame->cfa_offset = offset;
+			break;
+		case DW_CFA_def_cfa_expression:
+			count = dwarf_read_uleb128(current_insn, &expr_len);
+			current_insn += count;
+
+			frame->cfa_expr = current_insn;
+			frame->cfa_expr_len = expr_len;
+			current_insn += expr_len;
+
+			frame->flags |= DWARF_FRAME_CFA_REG_EXP;
+			break;
+		case DW_CFA_offset_extended_sf:
+			count = dwarf_read_uleb128(current_insn, &reg);
+			current_insn += count;
+			count = dwarf_read_leb128(current_insn, &offset);
+			current_insn += count;
+			offset *= cie->data_alignment_factor;
+			dwarf_frame_alloc_regs(frame, reg);
+			frame->regs[reg].flags |= DWARF_REG_OFFSET;
+			frame->regs[reg].addr = offset;
+			break;
+		case DW_CFA_val_offset:
+			count = dwarf_read_uleb128(current_insn, &reg);
+			current_insn += count;
+			count = dwarf_read_leb128(current_insn, &offset);
+			offset *= cie->data_alignment_factor;
+			frame->regs[reg].flags |= DWARF_REG_OFFSET;
+			frame->regs[reg].addr = offset;
+			break;
+		default:
+			pr_debug("unhandled DWARF instruction 0x%x\n", insn);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+/**
+ *	dwarf_unwind_stack - recursively unwind the stack
+ *	@pc: address of the function to unwind
+ *	@prev: struct dwarf_frame of the previous stackframe on the callstack
+ *
+ *	Return a struct dwarf_frame representing the most recent frame
+ *	on the callstack. Each of the lower (older) stack frames are
+ *	linked via the "prev" member.
+ */
+struct dwarf_frame *dwarf_unwind_stack(unsigned long pc,
+				       struct dwarf_frame *prev)
+{
+	struct dwarf_frame *frame;
+	struct dwarf_cie *cie;
+	struct dwarf_fde *fde;
+	unsigned long addr;
+	int i, offset;
+	bool define_ra = false;
+
+	/*
+	 * If this is the first invocation of this recursive function we
+	 * need get the contents of a physical register to get the CFA
+	 * in order to begin the virtual unwinding of the stack.
+	 *
+	 * Setting "define_ra" to true indictates that we want
+	 * dwarf_cfa_execute_insns() to continue executing instructions
+	 * until we know how to calculate the value of DWARF_ARCH_RA_REG
+	 * (which we need in order to kick off the whole unwinding
+	 * process).
+	 *
+	 * NOTE: the return address is guaranteed to be setup by the
+	 * time this function makes its first function call.
+	 */
+	if (!pc && !prev) {
+		pc = (unsigned long)&dwarf_unwind_stack;
+		define_ra = true;
+	}
+
+	frame = kzalloc(sizeof(*frame), GFP_ATOMIC);
+	if (!frame)
+		return NULL;
+
+	frame->prev = prev;
+
+	fde = dwarf_lookup_fde(pc);
+	if (!fde) {
+		/*
+		 * This is our normal exit path - the one that stops the
+		 * recursion. There's two reasons why we might exit
+		 * here,
+		 *
+		 *	a) pc has no asscociated DWARF frame info and so
+		 *	we don't know how to unwind this frame. This is
+		 *	usually the case when we're trying to unwind a
+		 *	frame that was called from some assembly code
+		 *	that has no DWARF info, e.g. syscalls.
+		 *
+		 *	b) the DEBUG info for pc is bogus. There's
+		 *	really no way to distinguish this case from the
+		 *	case above, which sucks because we could print a
+		 *	warning here.
+		 */
+		return NULL;
+	}
+
+	cie = dwarf_lookup_cie(fde->cie_pointer);
+
+	frame->pc = fde->initial_location;
+
+	/* CIE initial instructions */
+	dwarf_cfa_execute_insns(cie->initial_instructions,
+				cie->instructions_end, cie, fde,
+				frame, pc, false);
+
+	/* FDE instructions */
+	dwarf_cfa_execute_insns(fde->instructions, fde->end, cie,
+				fde, frame, pc, define_ra);
+
+	/* Calculate the CFA */
+	switch (frame->flags) {
+	case DWARF_FRAME_CFA_REG_OFFSET:
+		if (prev) {
+			BUG_ON(!prev->regs[frame->cfa_register].flags);
+
+			addr = prev->cfa;
+			addr += prev->regs[frame->cfa_register].addr;
+			frame->cfa = __raw_readl(addr);
+
+		} else {
+			/*
+			 * Again, this is the first invocation of this
+			 * recurisve function. We need to physically
+			 * read the contents of a register in order to
+			 * get the Canonical Frame Address for this
+			 * function.
+			 */
+			frame->cfa = dwarf_read_arch_reg(frame->cfa_register);
+		}
+
+		frame->cfa += frame->cfa_offset;
+		break;
+	default:
+		BUG();
+	}
+
+	/* If we haven't seen the return address reg, we're screwed. */
+	BUG_ON(!frame->regs[DWARF_ARCH_RA_REG].flags);
+
+	for (i = 0; i <= frame->num_regs; i++) {
+		struct dwarf_reg *reg = &frame->regs[i];
+
+		if (!reg->flags)
+			continue;
+
+		offset = reg->addr;
+		offset += frame->cfa;
+	}
+
+	addr = frame->cfa + frame->regs[DWARF_ARCH_RA_REG].addr;
+	frame->return_addr = __raw_readl(addr);
+
+	frame->next = dwarf_unwind_stack(frame->return_addr, frame);
+	return frame;
+}
+
+static int dwarf_parse_cie(void *entry, void *p, unsigned long len,
+			   unsigned char *end)
+{
+	struct dwarf_cie *cie;
+	unsigned long flags;
+	int count;
+
+	cie = kzalloc(sizeof(*cie), GFP_KERNEL);
+	if (!cie)
+		return -ENOMEM;
+
+	cie->length = len;
+
+	/*
+	 * Record the offset into the .eh_frame section
+	 * for this CIE. It allows this CIE to be
+	 * quickly and easily looked up from the
+	 * corresponding FDE.
+	 */
+	cie->cie_pointer = (unsigned long)entry;
+
+	cie->version = *(char *)p++;
+	BUG_ON(cie->version != 1);
+
+	cie->augmentation = p;
+	p += strlen(cie->augmentation) + 1;
+
+	count = dwarf_read_uleb128(p, &cie->code_alignment_factor);
+	p += count;
+
+	count = dwarf_read_leb128(p, &cie->data_alignment_factor);
+	p += count;
+
+	/*
+	 * Which column in the rule table contains the
+	 * return address?
+	 */
+	if (cie->version == 1) {
+		cie->return_address_reg = __raw_readb(p);
+		p++;
+	} else {
+		count = dwarf_read_uleb128(p, &cie->return_address_reg);
+		p += count;
+	}
+
+	if (cie->augmentation[0] == 'z') {
+		unsigned int length, count;
+		cie->flags |= DWARF_CIE_Z_AUGMENTATION;
+
+		count = dwarf_read_uleb128(p, &length);
+		p += count;
+
+		BUG_ON((unsigned char *)p > end);
+
+		cie->initial_instructions = p + length;
+		cie->augmentation++;
+	}
+
+	while (*cie->augmentation) {
+		/*
+		 * "L" indicates a byte showing how the
+		 * LSDA pointer is encoded. Skip it.
+		 */
+		if (*cie->augmentation == 'L') {
+			p++;
+			cie->augmentation++;
+		} else if (*cie->augmentation == 'R') {
+			/*
+			 * "R" indicates a byte showing
+			 * how FDE addresses are
+			 * encoded.
+			 */
+			cie->encoding = *(char *)p++;
+			cie->augmentation++;
+		} else if (*cie->augmentation == 'P') {
+			/*
+			 * "R" indicates a personality
+			 * routine in the CIE
+			 * augmentation.
+			 */
+			BUG();
+		} else if (*cie->augmentation == 'S') {
+			BUG();
+		} else {
+			/*
+			 * Unknown augmentation. Assume
+			 * 'z' augmentation.
+			 */
+			p = cie->initial_instructions;
+			BUG_ON(!p);
+			break;
+		}
+	}
+
+	cie->initial_instructions = p;
+	cie->instructions_end = end;
+
+	/* Add to list */
+	spin_lock_irqsave(&dwarf_cie_lock, flags);
+	list_add_tail(&cie->link, &dwarf_cie_list);
+	spin_unlock_irqrestore(&dwarf_cie_lock, flags);
+
+	return 0;
+}
+
+static int dwarf_parse_fde(void *entry, u32 entry_type,
+			   void *start, unsigned long len)
+{
+	struct dwarf_fde *fde;
+	struct dwarf_cie *cie;
+	unsigned long flags;
+	int count;
+	void *p = start;
+
+	fde = kzalloc(sizeof(*fde), GFP_KERNEL);
+	if (!fde)
+		return -ENOMEM;
+
+	fde->length = len;
+
+	/*
+	 * In a .eh_frame section the CIE pointer is the
+	 * delta between the address within the FDE
+	 */
+	fde->cie_pointer = (unsigned long)(p - entry_type - 4);
+
+	cie = dwarf_lookup_cie(fde->cie_pointer);
+	fde->cie = cie;
+
+	if (cie->encoding)
+		count = dwarf_read_encoded_value(p, &fde->initial_location,
+						 cie->encoding);
+	else
+		count = dwarf_read_addr(p, &fde->initial_location);
+
+	p += count;
+
+	if (cie->encoding)
+		count = dwarf_read_encoded_value(p, &fde->address_range,
+						 cie->encoding & 0x0f);
+	else
+		count = dwarf_read_addr(p, &fde->address_range);
+
+	p += count;
+
+	if (fde->cie->flags & DWARF_CIE_Z_AUGMENTATION) {
+		unsigned int length;
+		count = dwarf_read_uleb128(p, &length);
+		p += count + length;
+	}
+
+	/* Call frame instructions. */
+	fde->instructions = p;
+	fde->end = start + len;
+
+	/* Add to list. */
+	spin_lock_irqsave(&dwarf_fde_lock, flags);
+	list_add_tail(&fde->link, &dwarf_fde_list);
+	spin_unlock_irqrestore(&dwarf_fde_lock, flags);
+
+	return 0;
+}
+
+static void dwarf_unwinder_dump(struct task_struct *task, struct pt_regs *regs,
+				unsigned long *sp,
+				const struct stacktrace_ops *ops, void *data)
+{
+	struct dwarf_frame *frame;
+
+	frame = dwarf_unwind_stack(0, NULL);
+
+	while (frame && frame->return_addr) {
+		ops->address(data, frame->return_addr, 1);
+		frame = frame->next;
+	}
+}
+
+static struct unwinder dwarf_unwinder = {
+	.name = "dwarf-unwinder",
+	.dump = dwarf_unwinder_dump,
+	.rating = 150,
+};
+
+static void dwarf_unwinder_cleanup(void)
+{
+	struct dwarf_cie *cie, *m;
+	struct dwarf_fde *fde, *n;
+	unsigned long flags;
+
+	/*
+	 * Deallocate all the memory allocated for the DWARF unwinder.
+	 * Traverse all the FDE/CIE lists and remove and free all the
+	 * memory associated with those data structures.
+	 */
+	spin_lock_irqsave(&dwarf_cie_lock, flags);
+	list_for_each_entry_safe(cie, m, &dwarf_cie_list, link)
+		kfree(cie);
+	spin_unlock_irqrestore(&dwarf_cie_lock, flags);
+
+	spin_lock_irqsave(&dwarf_fde_lock, flags);
+	list_for_each_entry_safe(fde, n, &dwarf_fde_list, link)
+		kfree(fde);
+	spin_unlock_irqrestore(&dwarf_fde_lock, flags);
+}
+
+/**
+ *	dwarf_unwinder_init - initialise the dwarf unwinder
+ *
+ *	Build the data structures describing the .dwarf_frame section to
+ *	make it easier to lookup CIE and FDE entries. Because the
+ *	.eh_frame section is packed as tightly as possible it is not
+ *	easy to lookup the FDE for a given PC, so we build a list of FDE
+ *	and CIE entries that make it easier.
+ */
+void dwarf_unwinder_init(void)
+{
+	u32 entry_type;
+	void *p, *entry;
+	int count, err;
+	unsigned long len;
+	unsigned int c_entries, f_entries;
+	unsigned char *end;
+	INIT_LIST_HEAD(&dwarf_cie_list);
+	INIT_LIST_HEAD(&dwarf_fde_list);
+
+	c_entries = 0;
+	f_entries = 0;
+	entry = &__start_eh_frame;
+
+	while ((char *)entry < __stop_eh_frame) {
+		p = entry;
+
+		count = dwarf_entry_len(p, &len);
+		if (count == 0) {
+			/*
+			 * We read a bogus length field value. There is
+			 * nothing we can do here apart from disabling
+			 * the DWARF unwinder. We can't even skip this
+			 * entry and move to the next one because 'len'
+			 * tells us where our next entry is.
+			 */
+			goto out;
+		} else
+			p += count;
+
+		/* initial length does not include itself */
+		end = p + len;
+
+		entry_type = get_unaligned((u32 *)p);
+		p += 4;
+
+		if (entry_type == DW_EH_FRAME_CIE) {
+			err = dwarf_parse_cie(entry, p, len, end);
+			if (err < 0)
+				goto out;
+			else
+				c_entries++;
+		} else {
+			err = dwarf_parse_fde(entry, entry_type, p, len);
+			if (err < 0)
+				goto out;
+			else
+				f_entries++;
+		}
+
+		entry = (char *)entry + len + 4;
+	}
+
+	printk(KERN_INFO "DWARF unwinder initialised: read %u CIEs, %u FDEs\n",
+	       c_entries, f_entries);
+
+	err = unwinder_register(&dwarf_unwinder);
+	if (err)
+		goto out;
+
+	return;
+
+out:
+	printk(KERN_ERR "Failed to initialise DWARF unwinder: %d\n", err);
+	dwarf_unwinder_cleanup();
+}
diff --git a/arch/sh/kernel/early_printk.c b/arch/sh/kernel/early_printk.c
index 64f2746baf9..81a46145ffa 100644
--- a/arch/sh/kernel/early_printk.c
+++ b/arch/sh/kernel/early_printk.c
@@ -223,6 +223,7 @@ static int __init setup_early_printk(char *buf)
 #if defined(CONFIG_CPU_SH4) || defined(CONFIG_CPU_SH3)
 		scif_sercon_init(buf + 6);
 #endif
+#endif
 	}
 #endif
 
diff --git a/arch/sh/kernel/entry-common.S b/arch/sh/kernel/entry-common.S
index fc26ccd8278..e63178fefb9 100644
--- a/arch/sh/kernel/entry-common.S
+++ b/arch/sh/kernel/entry-common.S
@@ -43,6 +43,7 @@
  *	syscall #
  *
  */
+#include <asm/dwarf.h>
 
 #if defined(CONFIG_PREEMPT)
 #  define preempt_stop()	cli ; TRACE_IRQS_OFF
@@ -66,6 +67,11 @@ ENTRY(exception_error)
 
 	.align	2
 ret_from_exception:
+	CFI_STARTPROC simple
+	CFI_DEF_CFA r14, 0
+	CFI_REL_OFFSET 17, 64
+	CFI_REL_OFFSET 15, 0
+	CFI_REL_OFFSET 14, 56
 	preempt_stop()
 ENTRY(ret_from_irq)
 	!
@@ -240,6 +246,7 @@ debug_trap:
 	 nop
 	bra	__restore_all
 	 nop
+	CFI_ENDPROC
 
 	.align	2
 1:	.long	debug_trap_table
@@ -285,6 +292,7 @@ ret_from_fork:
  * system calls and debug traps through their respective jump tables.
  */
 ENTRY(system_call)
+	setup_frame_reg
 #if !defined(CONFIG_CPU_SH2)
 	mov.l	1f, r9
 	mov.l	@r9, r8		! Read from TRA (Trap Address) Register
diff --git a/arch/sh/kernel/irq.c b/arch/sh/kernel/irq.c
index 278c68c6048..2bb43dc74f2 100644
--- a/arch/sh/kernel/irq.c
+++ b/arch/sh/kernel/irq.c
@@ -14,6 +14,7 @@
 #include <asm/processor.h>
 #include <asm/machvec.h>
 #include <asm/uaccess.h>
+#include <asm/dwarf.h>
 #include <asm/thread_info.h>
 #include <cpu/mmu_context.h>
 
@@ -261,6 +262,9 @@ void __init init_IRQ(void)
 		sh_mv.mv_init_irq();
 
 	irq_ctx_init(smp_processor_id());
+
+	/* This needs to be early, but not too early.. */
+	dwarf_unwinder_init();
 }
 
 #ifdef CONFIG_SPARSE_IRQ
diff --git a/arch/sh/kernel/stacktrace.c b/arch/sh/kernel/stacktrace.c
index 1a2a5eb76e4..c2e45c48409 100644
--- a/arch/sh/kernel/stacktrace.c
+++ b/arch/sh/kernel/stacktrace.c
@@ -13,47 +13,93 @@
 #include <linux/stacktrace.h>
 #include <linux/thread_info.h>
 #include <linux/module.h>
+#include <asm/unwinder.h>
 #include <asm/ptrace.h>
+#include <asm/stacktrace.h>
+
+static void save_stack_warning(void *data, char *msg)
+{
+}
+
+static void
+save_stack_warning_symbol(void *data, char *msg, unsigned long symbol)
+{
+}
+
+static int save_stack_stack(void *data, char *name)
+{
+	return 0;
+}
 
 /*
  * Save stack-backtrace addresses into a stack_trace buffer.
  */
+static void save_stack_address(void *data, unsigned long addr, int reliable)
+{
+	struct stack_trace *trace = data;
+
+	if (!reliable)
+		return;
+
+	if (trace->skip > 0) {
+		trace->skip--;
+		return;
+	}
+
+	if (trace->nr_entries < trace->max_entries)
+		trace->entries[trace->nr_entries++] = addr;
+}
+
+static const struct stacktrace_ops save_stack_ops = {
+	.warning = save_stack_warning,
+	.warning_symbol = save_stack_warning_symbol,
+	.stack = save_stack_stack,
+	.address = save_stack_address,
+};
+
 void save_stack_trace(struct stack_trace *trace)
 {
 	unsigned long *sp = (unsigned long *)current_stack_pointer;
 
-	while (!kstack_end(sp)) {
-		unsigned long 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;
-		}
-	}
+	unwind_stack(current, NULL, sp,  &save_stack_ops, trace);
+	if (trace->nr_entries < trace->max_entries)
+		trace->entries[trace->nr_entries++] = ULONG_MAX;
 }
 EXPORT_SYMBOL_GPL(save_stack_trace);
 
+static void
+save_stack_address_nosched(void *data, unsigned long addr, int reliable)
+{
+	struct stack_trace *trace = (struct stack_trace *)data;
+
+	if (!reliable)
+		return;
+
+	if (in_sched_functions(addr))
+		return;
+
+	if (trace->skip > 0) {
+		trace->skip--;
+		return;
+	}
+
+	if (trace->nr_entries < trace->max_entries)
+		trace->entries[trace->nr_entries++] = addr;
+}
+
+static const struct stacktrace_ops save_stack_ops_nosched = {
+	.warning = save_stack_warning,
+	.warning_symbol = save_stack_warning_symbol,
+	.stack = save_stack_stack,
+	.address = save_stack_address_nosched,
+};
+
 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
 {
 	unsigned long *sp = (unsigned long *)tsk->thread.sp;
 
-	while (!kstack_end(sp)) {
-		unsigned long addr = *sp++;
-
-		if (__kernel_text_address(addr)) {
-			if (in_sched_functions(addr))
-				break;
-			if (trace->skip > 0)
-				trace->skip--;
-			else
-				trace->entries[trace->nr_entries++] = addr;
-			if (trace->nr_entries >= trace->max_entries)
-				break;
-		}
-	}
+	unwind_stack(current, NULL, sp,  &save_stack_ops_nosched, trace);
+	if (trace->nr_entries < trace->max_entries)
+		trace->entries[trace->nr_entries++] = ULONG_MAX;
 }
 EXPORT_SYMBOL_GPL(save_stack_trace_tsk);
diff --git a/arch/sh/kernel/traps_32.c b/arch/sh/kernel/traps_32.c
index 2b772776fcd..563426487c6 100644
--- a/arch/sh/kernel/traps_32.c
+++ b/arch/sh/kernel/traps_32.c
@@ -858,30 +858,6 @@ void __init trap_init(void)
 	per_cpu_trap_init();
 }
 
-void show_trace(struct task_struct *tsk, unsigned long *sp,
-		struct pt_regs *regs)
-{
-	unsigned long addr;
-
-	if (regs && user_mode(regs))
-		return;
-
-	printk("\nCall trace:\n");
-
-	while (!kstack_end(sp)) {
-		addr = *sp++;
-		if (kernel_text_address(addr))
-			print_ip_sym(addr);
-	}
-
-	printk("\n");
-
-	if (!tsk)
-		tsk = current;
-
-	debug_show_held_locks(tsk);
-}
-
 void show_stack(struct task_struct *tsk, unsigned long *sp)
 {
 	unsigned long stack;
diff --git a/arch/sh/kernel/unwinder.c b/arch/sh/kernel/unwinder.c
new file mode 100644
index 00000000000..2b30fa28b44
--- /dev/null
+++ b/arch/sh/kernel/unwinder.c
@@ -0,0 +1,162 @@
+/*
+ * Copyright (C) 2009  Matt Fleming
+ *
+ * Based, in part, on kernel/time/clocksource.c.
+ *
+ * This file provides arbitration code for stack unwinders.
+ *
+ * Multiple stack unwinders can be available on a system, usually with
+ * the most accurate unwinder being the currently active one.
+ */
+#include <linux/errno.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <asm/unwinder.h>
+#include <asm/atomic.h>
+
+/*
+ * This is the most basic stack unwinder an architecture can
+ * provide. For architectures without reliable frame pointers, e.g.
+ * RISC CPUs, it can be implemented by looking through the stack for
+ * addresses that lie within the kernel text section.
+ *
+ * Other CPUs, e.g. x86, can use their frame pointer register to
+ * construct more accurate stack traces.
+ */
+static struct list_head unwinder_list;
+static struct unwinder stack_reader = {
+	.name = "stack-reader",
+	.dump = stack_reader_dump,
+	.rating = 50,
+	.list = {
+		.next = &unwinder_list,
+		.prev = &unwinder_list,
+	},
+};
+
+/*
+ * "curr_unwinder" points to the stack unwinder currently in use. This
+ * is the unwinder with the highest rating.
+ *
+ * "unwinder_list" is a linked-list of all available unwinders, sorted
+ * by rating.
+ *
+ * All modifications of "curr_unwinder" and "unwinder_list" must be
+ * performed whilst holding "unwinder_lock".
+ */
+static struct unwinder *curr_unwinder = &stack_reader;
+
+static struct list_head unwinder_list = {
+	.next = &stack_reader.list,
+	.prev = &stack_reader.list,
+};
+
+static DEFINE_SPINLOCK(unwinder_lock);
+
+static atomic_t unwinder_running = ATOMIC_INIT(0);
+
+/**
+ * select_unwinder - Select the best registered stack unwinder.
+ *
+ * Private function. Must hold unwinder_lock when called.
+ *
+ * Select the stack unwinder with the best rating. This is useful for
+ * setting up curr_unwinder.
+ */
+static struct unwinder *select_unwinder(void)
+{
+	struct unwinder *best;
+
+	if (list_empty(&unwinder_list))
+		return NULL;
+
+	best = list_entry(unwinder_list.next, struct unwinder, list);
+	if (best == curr_unwinder)
+		return NULL;
+
+	return best;
+}
+
+/*
+ * Enqueue the stack unwinder sorted by rating.
+ */
+static int unwinder_enqueue(struct unwinder *ops)
+{
+	struct list_head *tmp, *entry = &unwinder_list;
+
+	list_for_each(tmp, &unwinder_list) {
+		struct unwinder *o;
+
+		o = list_entry(tmp, struct unwinder, list);
+		if (o == ops)
+			return -EBUSY;
+		/* Keep track of the place, where to insert */
+		if (o->rating >= ops->rating)
+			entry = tmp;
+	}
+	list_add(&ops->list, entry);
+
+	return 0;
+}
+
+/**
+ * unwinder_register - Used to install new stack unwinder
+ * @u: unwinder to be registered
+ *
+ * Install the new stack unwinder on the unwinder list, which is sorted
+ * by rating.
+ *
+ * Returns -EBUSY if registration fails, zero otherwise.
+ */
+int unwinder_register(struct unwinder *u)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&unwinder_lock, flags);
+	ret = unwinder_enqueue(u);
+	if (!ret)
+		curr_unwinder = select_unwinder();
+	spin_unlock_irqrestore(&unwinder_lock, flags);
+
+	return ret;
+}
+
+/*
+ * Unwind the call stack and pass information to the stacktrace_ops
+ * functions. Also handle the case where we need to switch to a new
+ * stack dumper because the current one faulted unexpectedly.
+ */
+void unwind_stack(struct task_struct *task, struct pt_regs *regs,
+		  unsigned long *sp, const struct stacktrace_ops *ops,
+		  void *data)
+{
+	unsigned long flags;
+
+	/*
+	 * The problem with unwinders with high ratings is that they are
+	 * inherently more complicated than the simple ones with lower
+	 * ratings. We are therefore more likely to fault in the
+	 * complicated ones, e.g. hitting BUG()s. If we fault in the
+	 * code for the current stack unwinder we try to downgrade to
+	 * one with a lower rating.
+	 *
+	 * Hopefully this will give us a semi-reliable stacktrace so we
+	 * can diagnose why curr_unwinder->dump() faulted.
+	 */
+	if (atomic_inc_return(&unwinder_running) != 1) {
+		spin_lock_irqsave(&unwinder_lock, flags);
+
+		if (!list_is_singular(&unwinder_list)) {
+			list_del(&curr_unwinder->list);
+			curr_unwinder = select_unwinder();
+		}
+
+		spin_unlock_irqrestore(&unwinder_lock, flags);
+		atomic_dec(&unwinder_running);
+	}
+
+	curr_unwinder->dump(task, regs, sp, ops, data);
+
+	atomic_dec(&unwinder_running);
+}
diff --git a/arch/sh/kernel/vmlinux.lds.S b/arch/sh/kernel/vmlinux.lds.S
index 80dc9f8d941..1b7d9d541e0 100644
--- a/arch/sh/kernel/vmlinux.lds.S
+++ b/arch/sh/kernel/vmlinux.lds.S
@@ -12,7 +12,7 @@ OUTPUT_ARCH(sh)
 
 #include <asm/thread_info.h>
 #include <asm/cache.h>
-#include <asm-generic/vmlinux.lds.h>
+#include <asm/vmlinux.lds.h>
 
 ENTRY(_start)
 SECTIONS
@@ -70,6 +70,8 @@ SECTIONS
 
 	_edata = .;			/* End of data section */
 
+	DWARF_EH_FRAME
+
 	. = ALIGN(PAGE_SIZE);		/* Init code and data */
 	__init_begin = .;
 	INIT_TEXT_SECTION(PAGE_SIZE)