diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c
index 2a946a160cac7d7050fc1e6644aa00a590c1ebce..c74e20d593a77ede26b3c47de5d90b2811ee818b 100644
--- a/arch/x86/kernel/cpu/perf_counter.c
+++ b/arch/x86/kernel/cpu/perf_counter.c
@@ -1088,6 +1088,7 @@ perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry)
 {
 	unsigned long bp;
 	char *stack;
+	int nr = entry->nr;
 
 	callchain_store(entry, instruction_pointer(regs));
 
@@ -1099,6 +1100,8 @@ perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry)
 #endif
 
 	dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, entry);
+
+	entry->kernel = entry->nr - nr;
 }
 
 
@@ -1128,6 +1131,7 @@ perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
 {
 	struct stack_frame frame;
 	const void __user *fp;
+	int nr = entry->nr;
 
 	regs = (struct pt_regs *)current->thread.sp0 - 1;
 	fp   = (void __user *)regs->bp;
@@ -1147,6 +1151,8 @@ perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
 		callchain_store(entry, frame.return_address);
 		fp = frame.next_fp;
 	}
+
+	entry->user = entry->nr - nr;
 }
 
 static void
@@ -1182,6 +1188,9 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 		entry = &__get_cpu_var(irq_entry);
 
 	entry->nr = 0;
+	entry->hv = 0;
+	entry->kernel = 0;
+	entry->user = 0;
 
 	perf_do_callchain(regs, entry);
 
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 5428ba120d78cfd917a9be7ff1f90bc48d9b4307..90cce0c74a034664092cf510ec7a4cc57a6e7443 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -513,10 +513,10 @@ extern void perf_counter_mmap(unsigned long addr, unsigned long len,
 extern void perf_counter_munmap(unsigned long addr, unsigned long len,
 				unsigned long pgoff, struct file *file);
 
-#define MAX_STACK_DEPTH		255
+#define MAX_STACK_DEPTH		254
 
 struct perf_callchain_entry {
-	u64	nr;
+	u32	nr, hv, kernel, user;
 	u64	ip[MAX_STACK_DEPTH];
 };
 
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 9bcab10e735dfa476b124e64bad9835a923d7587..f105a6e696c20c178de06fa332c9407e6bb4ce0f 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -1819,7 +1819,7 @@ void perf_counter_output(struct perf_counter *counter,
 		callchain = perf_callchain(regs);
 
 		if (callchain) {
-			callchain_size = (1 + callchain->nr) * sizeof(u64);
+			callchain_size = (2 + callchain->nr) * sizeof(u64);
 
 			header.type |= __PERF_EVENT_CALLCHAIN;
 			header.size += callchain_size;