diff --git a/arch/s390/include/asm/processor.h b/arch/s390/include/asm/processor.h
index 073e18b..d6fd22e 100644
--- a/arch/s390/include/asm/processor.h
+++ b/arch/s390/include/asm/processor.h
@@ -184,6 +184,10 @@
 struct mm_struct;
 struct seq_file;
 
+typedef int (*dump_trace_func_t)(void *data, unsigned long address);
+void dump_trace(dump_trace_func_t func, void *data,
+		struct task_struct *task, unsigned long sp);
+
 void show_cacheinfo(struct seq_file *m);
 
 /* Free all resources held by a thread. */
diff --git a/arch/s390/kernel/dumpstack.c b/arch/s390/kernel/dumpstack.c
index e1c786d..2150b01 100644
--- a/arch/s390/kernel/dumpstack.c
+++ b/arch/s390/kernel/dumpstack.c
@@ -19,28 +19,28 @@
 #include <asm/ipl.h>
 
 /*
- * For show_trace we have tree different stack to consider:
+ * For dump_trace we have tree different stack to consider:
  *   - the panic stack which is used if the kernel stack has overflown
  *   - the asynchronous interrupt stack (cpu related)
  *   - the synchronous kernel stack (process related)
- * The stack trace can start at any of the three stack and can potentially
+ * The stack trace can start at any of the three stacks and can potentially
  * touch all of them. The order is: panic stack, async stack, sync stack.
  */
 static unsigned long
-__show_trace(unsigned long sp, unsigned long low, unsigned long high)
+__dump_trace(dump_trace_func_t func, void *data, unsigned long sp,
+	     unsigned long low, unsigned long high)
 {
 	struct stack_frame *sf;
 	struct pt_regs *regs;
-	unsigned long addr;
 
 	while (1) {
 		if (sp < low || sp > high - sizeof(*sf))
 			return sp;
 		sf = (struct stack_frame *) sp;
-		addr = sf->gprs[8];
-		printk("([<%016lx>] %pSR)\n", addr, (void *)addr);
 		/* Follow the backchain. */
 		while (1) {
+			if (func(data, sf->gprs[8]))
+				return sp;
 			low = sp;
 			sp = sf->back_chain;
 			if (!sp)
@@ -48,45 +48,58 @@
 			if (sp <= low || sp > high - sizeof(*sf))
 				return sp;
 			sf = (struct stack_frame *) sp;
-			addr = sf->gprs[8];
-			printk(" [<%016lx>] %pSR\n", addr, (void *)addr);
 		}
 		/* Zero backchain detected, check for interrupt frame. */
 		sp = (unsigned long) (sf + 1);
 		if (sp <= low || sp > high - sizeof(*regs))
 			return sp;
 		regs = (struct pt_regs *) sp;
-		addr = regs->psw.addr;
-		printk(" [<%016lx>] %pSR\n", addr, (void *)addr);
+		if (!user_mode(regs)) {
+			if (func(data, regs->psw.addr))
+				return sp;
+		}
 		low = sp;
 		sp = regs->gprs[15];
 	}
 }
 
-static void show_trace(struct task_struct *task, unsigned long *stack)
+void dump_trace(dump_trace_func_t func, void *data, struct task_struct *task,
+		unsigned long sp)
 {
-	const unsigned long frame_size =
-		STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
-	unsigned long sp;
+	unsigned long frame_size;
 
-	sp = (unsigned long) stack;
-	if (!sp)
-		sp = task ? task->thread.ksp : current_stack_pointer();
-	printk("Call Trace:\n");
+	frame_size = STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
 #ifdef CONFIG_CHECK_STACK
-	sp = __show_trace(sp,
+	sp = __dump_trace(func, data, sp,
 			  S390_lowcore.panic_stack + frame_size - 4096,
 			  S390_lowcore.panic_stack + frame_size);
 #endif
-	sp = __show_trace(sp,
+	sp = __dump_trace(func, data, sp,
 			  S390_lowcore.async_stack + frame_size - ASYNC_SIZE,
 			  S390_lowcore.async_stack + frame_size);
 	if (task)
-		__show_trace(sp, (unsigned long) task_stack_page(task),
-			     (unsigned long) task_stack_page(task) + THREAD_SIZE);
+		__dump_trace(func, data, sp,
+			     (unsigned long)task_stack_page(task),
+			     (unsigned long)task_stack_page(task) + THREAD_SIZE);
 	else
-		__show_trace(sp, S390_lowcore.thread_info,
+		__dump_trace(func, data, sp,
+			     S390_lowcore.thread_info,
 			     S390_lowcore.thread_info + THREAD_SIZE);
+}
+EXPORT_SYMBOL_GPL(dump_trace);
+
+static int show_address(void *data, unsigned long address)
+{
+	printk("([<%016lx>] %pSR)\n", address, (void *)address);
+	return 0;
+}
+
+static void show_trace(struct task_struct *task, unsigned long sp)
+{
+	if (!sp)
+		sp = task ? task->thread.ksp : current_stack_pointer();
+	printk("Call Trace:\n");
+	dump_trace(show_address, NULL, task, sp);
 	if (!task)
 		task = current;
 	debug_show_held_locks(task);
@@ -112,7 +125,7 @@
 		printk("%016lx ", *stack++);
 	}
 	printk("\n");
-	show_trace(task, sp);
+	show_trace(task, (unsigned long)sp);
 }
 
 static void show_last_breaking_event(struct pt_regs *regs)
@@ -152,7 +165,7 @@
 	show_registers(regs);
 	/* Show stack backtrace if pt_regs is from kernel mode */
 	if (!user_mode(regs))
-		show_trace(NULL, (unsigned long *) regs->gprs[15]);
+		show_trace(NULL, regs->gprs[15]);
 	show_last_breaking_event(regs);
 }
 
diff --git a/arch/s390/kernel/perf_event.c b/arch/s390/kernel/perf_event.c
index 0943b11..a1b7086 100644
--- a/arch/s390/kernel/perf_event.c
+++ b/arch/s390/kernel/perf_event.c
@@ -222,64 +222,20 @@
 }
 arch_initcall(service_level_perf_register);
 
-/* See also arch/s390/kernel/traps.c */
-static unsigned long __store_trace(struct perf_callchain_entry *entry,
-				   unsigned long sp,
-				   unsigned long low, unsigned long high)
+static int __perf_callchain_kernel(void *data, unsigned long address)
 {
-	struct stack_frame *sf;
-	struct pt_regs *regs;
+	struct perf_callchain_entry *entry = data;
 
-	while (1) {
-		if (sp < low || sp > high - sizeof(*sf))
-			return sp;
-		sf = (struct stack_frame *) sp;
-		perf_callchain_store(entry, sf->gprs[8]);
-		/* Follow the backchain. */
-		while (1) {
-			low = sp;
-			sp = sf->back_chain;
-			if (!sp)
-				break;
-			if (sp <= low || sp > high - sizeof(*sf))
-				return sp;
-			sf = (struct stack_frame *) sp;
-			perf_callchain_store(entry, sf->gprs[8]);
-		}
-		/* Zero backchain detected, check for interrupt frame. */
-		sp = (unsigned long) (sf + 1);
-		if (sp <= low || sp > high - sizeof(*regs))
-			return sp;
-		regs = (struct pt_regs *) sp;
-		perf_callchain_store(entry, sf->gprs[8]);
-		low = sp;
-		sp = regs->gprs[15];
-	}
+	perf_callchain_store(entry, address);
+	return 0;
 }
 
 void perf_callchain_kernel(struct perf_callchain_entry *entry,
 			   struct pt_regs *regs)
 {
-	unsigned long head, frame_size;
-	struct stack_frame *head_sf;
-
 	if (user_mode(regs))
 		return;
-
-	frame_size = STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
-	head = regs->gprs[15];
-	head_sf = (struct stack_frame *) head;
-
-	if (!head_sf || !head_sf->back_chain)
-		return;
-
-	head = head_sf->back_chain;
-	head = __store_trace(entry, head,
-			     S390_lowcore.async_stack + frame_size - ASYNC_SIZE,
-			     S390_lowcore.async_stack + frame_size);
-
-	__store_trace(entry, head, S390_lowcore.thread_info,
-		      S390_lowcore.thread_info + THREAD_SIZE);
+	dump_trace(__perf_callchain_kernel, entry, NULL, regs->gprs[15]);
 }
 
 /* Perf defintions for PMU event attributes in sysfs */
diff --git a/arch/s390/kernel/stacktrace.c b/arch/s390/kernel/stacktrace.c
index 7dd8360..44f84b2 100644
--- a/arch/s390/kernel/stacktrace.c
+++ b/arch/s390/kernel/stacktrace.c
@@ -10,69 +10,31 @@
 #include <linux/kallsyms.h>
 #include <linux/module.h>
 
-static unsigned long save_context_stack(struct stack_trace *trace,
-					unsigned long sp,
-					unsigned long low,
-					unsigned long high,
-					int nosched)
+static int __save_address(void *data, unsigned long address, int nosched)
 {
-	struct stack_frame *sf;
-	struct pt_regs *regs;
-	unsigned long addr;
+	struct stack_trace *trace = data;
 
-	while(1) {
-		if (sp < low || sp > high)
-			return sp;
-		sf = (struct stack_frame *)sp;
-		while(1) {
-			addr = sf->gprs[8];
-			if (!trace->skip)
-				trace->entries[trace->nr_entries++] = addr;
-			else
-				trace->skip--;
-			if (trace->nr_entries >= trace->max_entries)
-				return sp;
-			low = sp;
-			sp = sf->back_chain;
-			if (!sp)
-				break;
-			if (sp <= low || sp > high - sizeof(*sf))
-				return sp;
-			sf = (struct stack_frame *)sp;
-		}
-		/* Zero backchain detected, check for interrupt frame. */
-		sp = (unsigned long)(sf + 1);
-		if (sp <= low || sp > high - sizeof(*regs))
-			return sp;
-		regs = (struct pt_regs *)sp;
-		addr = regs->psw.addr;
-		if (!nosched || !in_sched_functions(addr)) {
-			if (!trace->skip)
-				trace->entries[trace->nr_entries++] = addr;
-			else
-				trace->skip--;
-		}
-		if (trace->nr_entries >= trace->max_entries)
-			return sp;
-		low = sp;
-		sp = regs->gprs[15];
+	if (nosched && in_sched_functions(address))
+		return 0;
+	if (trace->skip > 0) {
+		trace->skip--;
+		return 0;
 	}
+	if (trace->nr_entries < trace->max_entries) {
+		trace->entries[trace->nr_entries++] = address;
+		return 0;
+	}
+	return 1;
 }
 
-static void __save_stack_trace(struct stack_trace *trace, unsigned long sp)
+static int save_address(void *data, unsigned long address)
 {
-	unsigned long new_sp, frame_size;
+	return __save_address(data, address, 0);
+}
 
-	frame_size = STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
-	new_sp = save_context_stack(trace, sp,
-			S390_lowcore.panic_stack + frame_size - PAGE_SIZE,
-			S390_lowcore.panic_stack + frame_size, 0);
-	new_sp = save_context_stack(trace, new_sp,
-			S390_lowcore.async_stack + frame_size - ASYNC_SIZE,
-			S390_lowcore.async_stack + frame_size, 0);
-	save_context_stack(trace, new_sp,
-			   S390_lowcore.thread_info,
-			   S390_lowcore.thread_info + THREAD_SIZE, 0);
+static int save_address_nosched(void *data, unsigned long address)
+{
+	return __save_address(data, address, 1);
 }
 
 void save_stack_trace(struct stack_trace *trace)
@@ -80,7 +42,7 @@
 	unsigned long sp;
 
 	sp = current_stack_pointer();
-	__save_stack_trace(trace, sp);
+	dump_trace(save_address, trace, NULL, sp);
 	if (trace->nr_entries < trace->max_entries)
 		trace->entries[trace->nr_entries++] = ULONG_MAX;
 }
@@ -88,14 +50,12 @@
 
 void save_stack_trace_tsk(struct task_struct *tsk, struct stack_trace *trace)
 {
-	unsigned long sp, low, high;
+	unsigned long sp;
 
 	sp = tsk->thread.ksp;
 	if (tsk == current)
 		sp = current_stack_pointer();
-	low = (unsigned long) task_stack_page(tsk);
-	high = (unsigned long) task_pt_regs(tsk);
-	save_context_stack(trace, sp, low, high, 1);
+	dump_trace(save_address_nosched, trace, tsk, sp);
 	if (trace->nr_entries < trace->max_entries)
 		trace->entries[trace->nr_entries++] = ULONG_MAX;
 }
@@ -106,7 +66,7 @@
 	unsigned long sp;
 
 	sp = kernel_stack_pointer(regs);
-	__save_stack_trace(trace, sp);
+	dump_trace(save_address, trace, NULL, sp);
 	if (trace->nr_entries < trace->max_entries)
 		trace->entries[trace->nr_entries++] = ULONG_MAX;
 }
diff --git a/arch/s390/oprofile/Makefile b/arch/s390/oprofile/Makefile
index 1bd2301..496e4a7 100644
--- a/arch/s390/oprofile/Makefile
+++ b/arch/s390/oprofile/Makefile
@@ -6,5 +6,5 @@
 		oprofilefs.o oprofile_stats.o  \
 		timer_int.o )
 
-oprofile-y :=	$(DRIVER_OBJS) init.o backtrace.o
+oprofile-y :=	$(DRIVER_OBJS) init.o
 oprofile-y +=	hwsampler.o
diff --git a/arch/s390/oprofile/backtrace.c b/arch/s390/oprofile/backtrace.c
deleted file mode 100644
index 1884e17..0000000
--- a/arch/s390/oprofile/backtrace.c
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
- * S390 Version
- *   Copyright IBM Corp. 2005
- *   Author(s): Andreas Krebbel <Andreas.Krebbel@de.ibm.com>
- */
-
-#include <linux/oprofile.h>
-
-#include <asm/processor.h> /* for struct stack_frame */
-
-static unsigned long
-__show_trace(unsigned int *depth, unsigned long sp,
-	     unsigned long low, unsigned long high)
-{
-	struct stack_frame *sf;
-	struct pt_regs *regs;
-
-	while (*depth) {
-		if (sp < low || sp > high - sizeof(*sf))
-			return sp;
-		sf = (struct stack_frame *) sp;
-		(*depth)--;
-		oprofile_add_trace(sf->gprs[8]);
-
-		/* Follow the backchain.  */
-		while (*depth) {
-			low = sp;
-			sp = sf->back_chain;
-			if (!sp)
-				break;
-			if (sp <= low || sp > high - sizeof(*sf))
-				return sp;
-			sf = (struct stack_frame *) sp;
-			(*depth)--;
-			oprofile_add_trace(sf->gprs[8]);
-
-		}
-
-		if (*depth == 0)
-			break;
-
-		/* Zero backchain detected, check for interrupt frame.  */
-		sp = (unsigned long) (sf + 1);
-		if (sp <= low || sp > high - sizeof(*regs))
-			return sp;
-		regs = (struct pt_regs *) sp;
-		(*depth)--;
-		oprofile_add_trace(sf->gprs[8]);
-		low = sp;
-		sp = regs->gprs[15];
-	}
-	return sp;
-}
-
-void s390_backtrace(struct pt_regs * const regs, unsigned int depth)
-{
-	unsigned long head, frame_size;
-	struct stack_frame* head_sf;
-
-	if (user_mode(regs))
-		return;
-
-	frame_size = STACK_FRAME_OVERHEAD + sizeof(struct pt_regs);
-	head = regs->gprs[15];
-	head_sf = (struct stack_frame*)head;
-
-	if (!head_sf->back_chain)
-		return;
-
-	head = head_sf->back_chain;
-
-	head = __show_trace(&depth, head,
-			    S390_lowcore.async_stack + frame_size - ASYNC_SIZE,
-			    S390_lowcore.async_stack + frame_size);
-
-	__show_trace(&depth, head, S390_lowcore.thread_info,
-		     S390_lowcore.thread_info + THREAD_SIZE);
-}
diff --git a/arch/s390/oprofile/init.c b/arch/s390/oprofile/init.c
index 9cfa2ff..d85d1e6 100644
--- a/arch/s390/oprofile/init.c
+++ b/arch/s390/oprofile/init.c
@@ -20,8 +20,6 @@
 
 #include "../../../drivers/oprofile/oprof.h"
 
-extern void s390_backtrace(struct pt_regs * const regs, unsigned int depth);
-
 #include "hwsampler.h"
 #include "op_counter.h"
 
@@ -494,6 +492,24 @@
 	hwsampler_shutdown();
 }
 
+static int __s390_backtrace(void *data, unsigned long address)
+{
+	unsigned int *depth = data;
+
+	if (*depth == 0)
+		return 1;
+	(*depth)--;
+	oprofile_add_trace(address);
+	return 0;
+}
+
+static void s390_backtrace(struct pt_regs *regs, unsigned int depth)
+{
+	if (user_mode(regs))
+		return;
+	dump_trace(__s390_backtrace, &depth, NULL, regs->gprs[15]);
+}
+
 int __init oprofile_arch_init(struct oprofile_operations *ops)
 {
 	ops->backtrace = s390_backtrace;