diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index a3c922288cc063..8e91cf34a9c886 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -1608,6 +1608,11 @@ static const struct stacktrace_ops backtrace_ops = {
 void
 perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
 {
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* TODO: We don't support guest os callchain now */
+		return NULL;
+	}
+
 	perf_callchain_store(entry, regs->ip);
 
 	dump_trace(NULL, regs, NULL, regs->bp, &backtrace_ops, entry);
@@ -1656,6 +1661,10 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
 	struct stack_frame frame;
 	const void __user *fp;
 
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* TODO: We don't support guest os callchain now */
+		return NULL;
+	}
 
 	fp = (void __user *)regs->bp;
 
@@ -1681,19 +1690,6 @@ perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
 	}
 }
 
-struct perf_callchain_entry *perf_callchain_buffer(void)
-{
-	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
-		/* TODO: We don't support guest os callchain now */
-		return NULL;
-	}
-
-	if (in_nmi())
-		return &__get_cpu_var(perf_callchain_entry_nmi);
-
-	return &__get_cpu_var(perf_callchain_entry);
-}
-
 unsigned long perf_instruction_pointer(struct pt_regs *regs)
 {
 	unsigned long ip;
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 4db61dded388d3..d7e8ea6908642f 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -983,7 +983,6 @@ extern void perf_callchain_user(struct perf_callchain_entry *entry,
 				struct pt_regs *regs);
 extern void perf_callchain_kernel(struct perf_callchain_entry *entry,
 				  struct pt_regs *regs);
-extern struct perf_callchain_entry *perf_callchain_buffer(void);
 
 
 static inline void
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 615d024894cf6c..75ab8a2df6b2dc 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -1763,6 +1763,216 @@ static u64 perf_event_read(struct perf_event *event)
 	return perf_event_count(event);
 }
 
+/*
+ * Callchain support
+ */
+
+struct callchain_cpus_entries {
+	struct rcu_head			rcu_head;
+	struct perf_callchain_entry	*cpu_entries[0];
+};
+
+static DEFINE_PER_CPU(int, callchain_recursion[4]);
+static atomic_t nr_callchain_events;
+static DEFINE_MUTEX(callchain_mutex);
+struct callchain_cpus_entries *callchain_cpus_entries;
+
+
+__weak void perf_callchain_kernel(struct perf_callchain_entry *entry,
+				  struct pt_regs *regs)
+{
+}
+
+__weak void perf_callchain_user(struct perf_callchain_entry *entry,
+				struct pt_regs *regs)
+{
+}
+
+static void release_callchain_buffers_rcu(struct rcu_head *head)
+{
+	struct callchain_cpus_entries *entries;
+	int cpu;
+
+	entries = container_of(head, struct callchain_cpus_entries, rcu_head);
+
+	for_each_possible_cpu(cpu)
+		kfree(entries->cpu_entries[cpu]);
+
+	kfree(entries);
+}
+
+static void release_callchain_buffers(void)
+{
+	struct callchain_cpus_entries *entries;
+
+	entries = callchain_cpus_entries;
+	rcu_assign_pointer(callchain_cpus_entries, NULL);
+	call_rcu(&entries->rcu_head, release_callchain_buffers_rcu);
+}
+
+static int alloc_callchain_buffers(void)
+{
+	int cpu;
+	int size;
+	struct callchain_cpus_entries *entries;
+
+	/*
+	 * We can't use the percpu allocation API for data that can be
+	 * accessed from NMI. Use a temporary manual per cpu allocation
+	 * until that gets sorted out.
+	 */
+	size = sizeof(*entries) + sizeof(struct perf_callchain_entry *) *
+		num_possible_cpus();
+
+	entries = kzalloc(size, GFP_KERNEL);
+	if (!entries)
+		return -ENOMEM;
+
+	size = sizeof(struct perf_callchain_entry) * 4;
+
+	for_each_possible_cpu(cpu) {
+		entries->cpu_entries[cpu] = kmalloc_node(size, GFP_KERNEL,
+							 cpu_to_node(cpu));
+		if (!entries->cpu_entries[cpu])
+			goto fail;
+	}
+
+	rcu_assign_pointer(callchain_cpus_entries, entries);
+
+	return 0;
+
+fail:
+	for_each_possible_cpu(cpu)
+		kfree(entries->cpu_entries[cpu]);
+	kfree(entries);
+
+	return -ENOMEM;
+}
+
+static int get_callchain_buffers(void)
+{
+	int err = 0;
+	int count;
+
+	mutex_lock(&callchain_mutex);
+
+	count = atomic_inc_return(&nr_callchain_events);
+	if (WARN_ON_ONCE(count < 1)) {
+		err = -EINVAL;
+		goto exit;
+	}
+
+	if (count > 1) {
+		/* If the allocation failed, give up */
+		if (!callchain_cpus_entries)
+			err = -ENOMEM;
+		goto exit;
+	}
+
+	err = alloc_callchain_buffers();
+	if (err)
+		release_callchain_buffers();
+exit:
+	mutex_unlock(&callchain_mutex);
+
+	return err;
+}
+
+static void put_callchain_buffers(void)
+{
+	if (atomic_dec_and_mutex_lock(&nr_callchain_events, &callchain_mutex)) {
+		release_callchain_buffers();
+		mutex_unlock(&callchain_mutex);
+	}
+}
+
+static int get_recursion_context(int *recursion)
+{
+	int rctx;
+
+	if (in_nmi())
+		rctx = 3;
+	else if (in_irq())
+		rctx = 2;
+	else if (in_softirq())
+		rctx = 1;
+	else
+		rctx = 0;
+
+	if (recursion[rctx])
+		return -1;
+
+	recursion[rctx]++;
+	barrier();
+
+	return rctx;
+}
+
+static inline void put_recursion_context(int *recursion, int rctx)
+{
+	barrier();
+	recursion[rctx]--;
+}
+
+static struct perf_callchain_entry *get_callchain_entry(int *rctx)
+{
+	int cpu;
+	struct callchain_cpus_entries *entries;
+
+	*rctx = get_recursion_context(__get_cpu_var(callchain_recursion));
+	if (*rctx == -1)
+		return NULL;
+
+	entries = rcu_dereference(callchain_cpus_entries);
+	if (!entries)
+		return NULL;
+
+	cpu = smp_processor_id();
+
+	return &entries->cpu_entries[cpu][*rctx];
+}
+
+static void
+put_callchain_entry(int rctx)
+{
+	put_recursion_context(__get_cpu_var(callchain_recursion), rctx);
+}
+
+static struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
+{
+	int rctx;
+	struct perf_callchain_entry *entry;
+
+
+	entry = get_callchain_entry(&rctx);
+	if (rctx == -1)
+		return NULL;
+
+	if (!entry)
+		goto exit_put;
+
+	entry->nr = 0;
+
+	if (!user_mode(regs)) {
+		perf_callchain_store(entry, PERF_CONTEXT_KERNEL);
+		perf_callchain_kernel(entry, regs);
+		if (current->mm)
+			regs = task_pt_regs(current);
+		else
+			regs = NULL;
+	}
+
+	if (regs) {
+		perf_callchain_store(entry, PERF_CONTEXT_USER);
+		perf_callchain_user(entry, regs);
+	}
+
+exit_put:
+	put_callchain_entry(rctx);
+
+	return entry;
+}
+
 /*
  * Initialize the perf_event context in a task_struct:
  */
@@ -1895,6 +2105,8 @@ static void free_event(struct perf_event *event)
 			atomic_dec(&nr_comm_events);
 		if (event->attr.task)
 			atomic_dec(&nr_task_events);
+		if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)
+			put_callchain_buffers();
 	}
 
 	if (event->buffer) {
@@ -2937,55 +3149,6 @@ void perf_event_do_pending(void)
 	__perf_pending_run();
 }
 
-DEFINE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);
-
-/*
- * Callchain support -- arch specific
- */
-
-__weak struct perf_callchain_entry *perf_callchain_buffer(void)
-{
-	return &__get_cpu_var(perf_callchain_entry);
-}
-
-__weak void perf_callchain_kernel(struct perf_callchain_entry *entry,
-				  struct pt_regs *regs)
-{
-}
-
-__weak void perf_callchain_user(struct perf_callchain_entry *entry,
-				struct pt_regs *regs)
-{
-}
-
-static struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
-{
-	struct perf_callchain_entry *entry;
-
-	entry = perf_callchain_buffer();
-	if (!entry)
-		return NULL;
-
-	entry->nr = 0;
-
-	if (!user_mode(regs)) {
-		perf_callchain_store(entry, PERF_CONTEXT_KERNEL);
-		perf_callchain_kernel(entry, regs);
-		if (current->mm)
-			regs = task_pt_regs(current);
-		else
-			regs = NULL;
-	}
-
-	if (regs) {
-		perf_callchain_store(entry, PERF_CONTEXT_USER);
-		perf_callchain_user(entry, regs);
-	}
-
-	return entry;
-}
-
-
 /*
  * We assume there is only KVM supporting the callbacks.
  * Later on, we might change it to a list if there is
@@ -3480,14 +3643,20 @@ static void perf_event_output(struct perf_event *event, int nmi,
 	struct perf_output_handle handle;
 	struct perf_event_header header;
 
+	/* protect the callchain buffers */
+	rcu_read_lock();
+
 	perf_prepare_sample(&header, data, event, regs);
 
 	if (perf_output_begin(&handle, event, header.size, nmi, 1))
-		return;
+		goto exit;
 
 	perf_output_sample(&handle, &header, data, event);
 
 	perf_output_end(&handle);
+
+exit:
+	rcu_read_unlock();
 }
 
 /*
@@ -4243,32 +4412,16 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id,
 int perf_swevent_get_recursion_context(void)
 {
 	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
-	int rctx;
 
-	if (in_nmi())
-		rctx = 3;
-	else if (in_irq())
-		rctx = 2;
-	else if (in_softirq())
-		rctx = 1;
-	else
-		rctx = 0;
-
-	if (cpuctx->recursion[rctx])
-		return -1;
-
-	cpuctx->recursion[rctx]++;
-	barrier();
-
-	return rctx;
+	return get_recursion_context(cpuctx->recursion);
 }
 EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context);
 
 void inline perf_swevent_put_recursion_context(int rctx)
 {
 	struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
-	barrier();
-	cpuctx->recursion[rctx]--;
+
+	put_recursion_context(cpuctx->recursion, rctx);
 }
 
 void __perf_sw_event(u32 event_id, u64 nr, int nmi,
@@ -4968,6 +5121,13 @@ perf_event_alloc(struct perf_event_attr *attr,
 			atomic_inc(&nr_comm_events);
 		if (event->attr.task)
 			atomic_inc(&nr_task_events);
+		if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) {
+			err = get_callchain_buffers();
+			if (err) {
+				free_event(event);
+				return ERR_PTR(err);
+			}
+		}
 	}
 
 	return event;