[PATCH 0/6] Various perf fixes

[PATCH 0/6] Various perf fixes

[Peter Zijlstra]

My current perf queue..

[PATCH 1/6] percpu: {get,put}_cpu_ptr

[Peter Zijlstra]

[-- Attachment #1: percpu-get-cpu-ptr.patch --]
[-- Type: text/plain, Size: 769 bytes --]

Similar to {get,put}_cpu_var() except for dynamically allocated per-cpu memory.

Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 include/linux/percpu.h |    9 +++++++++
 1 file changed, 9 insertions(+)

Index: linux-2.6/include/linux/percpu.h
===================================================================
--- linux-2.6.orig/include/linux/percpu.h
+++ linux-2.6/include/linux/percpu.h
@@ -39,6 +39,15 @@
 	preempt_enable();				\
 } while (0)
 
+#define get_cpu_ptr(var) ({				\
+	preempt_disable();				\
+	this_cpu_ptr(var); })
+
+#define put_cpu_ptr(var) do {				\
+	(void)(var);					\
+	preempt_enable();				\
+} while (0)
+
 #ifdef CONFIG_SMP
 
 /* minimum unit size, also is the maximum supported allocation size */

Re: [PATCH 1/6] percpu: {get,put}_cpu_ptr

[Tejun Heo]

On 09/17/2010 11:28 AM, Peter Zijlstra wrote:
> Similar to {get,put}_cpu_var() except for dynamically allocated per-cpu memory.
> 
> Cc: Tejun Heo <tj@kernel.org>
> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>

Acked-by: Tejun Heo <tj@kernel.org>

How do you want to route this?  Given that perf is the only user at
this point, feel free to push this with other perf changes.

Thanks.

-- 
tejun

[tip:perf/core] percpu: Add {get,put}_cpu_ptr

[tip-bot for Peter Zijlstra]

Commit-ID:  8b8e2ec1eeca7f6941bc81cefc9663018d6ceb57
Gitweb:     http://git.kernel.org/tip/8b8e2ec1eeca7f6941bc81cefc9663018d6ceb57
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Thu, 16 Sep 2010 19:21:28 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 21 Sep 2010 13:55:43 +0200

percpu: Add {get,put}_cpu_ptr

These are similar to {get,put}_cpu_var() except for dynamically
allocated per-cpu memory.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Tejun Heo <tj@kernel.org>
LKML-Reference: <20100917093009.252867712@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 include/linux/percpu.h |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/include/linux/percpu.h b/include/linux/percpu.h
index 49466b1..0eb5083 100644
--- a/include/linux/percpu.h
+++ b/include/linux/percpu.h
@@ -39,6 +39,15 @@
 	preempt_enable();				\
 } while (0)
 
+#define get_cpu_ptr(var) ({				\
+	preempt_disable();				\
+	this_cpu_ptr(var); })
+
+#define put_cpu_ptr(var) do {				\
+	(void)(var);					\
+	preempt_enable();				\
+} while (0)
+
 #ifdef CONFIG_SMP
 
 /* minimum unit size, also is the maximum supported allocation size */

[PATCH 2/6] perf: Avoid RCU vs preemption assumptions

[Peter Zijlstra]

[-- Attachment #1: perf-fix-preempt-iteration.patch --]
[-- Type: text/plain, Size: 2377 bytes --]

The per-pmu per-cpu context patch converted things from get_cpu_var() to
this_cpu_ptr(), but that only works if rcu_read_lock() actually disables
preemption, and since there is no such guarantee, we need to fix that.

Use the newly introduced {get,put}_cpu_ptr().

Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/perf_event.c |   18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -3814,18 +3814,20 @@ static void perf_event_task_event(struct
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_task_ctx(&cpuctx->ctx, task_event);
 
 		ctx = task_event->task_ctx;
 		if (!ctx) {
 			ctxn = pmu->task_ctx_nr;
 			if (ctxn < 0)
-				continue;
+				goto next;
 			ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		}
 		if (ctx)
 			perf_event_task_ctx(ctx, task_event);
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 }
@@ -3947,16 +3949,18 @@ static void perf_event_comm_event(struct
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_comm_ctx(&cpuctx->ctx, comm_event);
 
 		ctxn = pmu->task_ctx_nr;
 		if (ctxn < 0)
-			continue;
+			goto next;
 
 		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		if (ctx)
 			perf_event_comm_ctx(ctx, comm_event);
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 }
@@ -4130,19 +4134,21 @@ static void perf_event_mmap_event(struct
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_mmap_ctx(&cpuctx->ctx, mmap_event,
 					vma->vm_flags & VM_EXEC);
 
 		ctxn = pmu->task_ctx_nr;
 		if (ctxn < 0)
-			continue;
+			goto next;
 
 		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		if (ctx) {
 			perf_event_mmap_ctx(ctx, mmap_event,
 					vma->vm_flags & VM_EXEC);
 		}
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 

[tip:perf/core] perf: Avoid RCU vs preemption assumptions

[tip-bot for Peter Zijlstra]

Commit-ID:  41945f6ccf1e86f87fddf6b32db9cf431c05fb54
Gitweb:     http://git.kernel.org/tip/41945f6ccf1e86f87fddf6b32db9cf431c05fb54
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Thu, 16 Sep 2010 19:17:24 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 21 Sep 2010 13:55:44 +0200

perf: Avoid RCU vs preemption assumptions

The per-pmu per-cpu context patch converted things from
get_cpu_var() to this_cpu_ptr(), but that only works if
rcu_read_lock() actually disables preemption, and since
there is no such guarantee, we need to fix that.

Use the newly introduced {get,put}_cpu_ptr().

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Tejun Heo <tj@kernel.org>
LKML-Reference: <20100917093009.308453028@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/perf_event.c |   18 ++++++++++++------
 1 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index baae136..c16158c 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -3836,18 +3836,20 @@ static void perf_event_task_event(struct perf_task_event *task_event)
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_task_ctx(&cpuctx->ctx, task_event);
 
 		ctx = task_event->task_ctx;
 		if (!ctx) {
 			ctxn = pmu->task_ctx_nr;
 			if (ctxn < 0)
-				continue;
+				goto next;
 			ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		}
 		if (ctx)
 			perf_event_task_ctx(ctx, task_event);
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 }
@@ -3969,16 +3971,18 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_comm_ctx(&cpuctx->ctx, comm_event);
 
 		ctxn = pmu->task_ctx_nr;
 		if (ctxn < 0)
-			continue;
+			goto next;
 
 		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		if (ctx)
 			perf_event_comm_ctx(ctx, comm_event);
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 }
@@ -4152,19 +4156,21 @@ got_name:
 
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		perf_event_mmap_ctx(&cpuctx->ctx, mmap_event,
 					vma->vm_flags & VM_EXEC);
 
 		ctxn = pmu->task_ctx_nr;
 		if (ctxn < 0)
-			continue;
+			goto next;
 
 		ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
 		if (ctx) {
 			perf_event_mmap_ctx(ctx, mmap_event,
 					vma->vm_flags & VM_EXEC);
 		}
+next:
+		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
 	rcu_read_unlock();
 

[PATCH 3/6] perf_events: Fix broken event grouping

[Peter Zijlstra]

[-- Attachment #1: stephane-perf_events-fix_broken_event_grouping.patch --]
[-- Type: text/plain, Size: 1746 bytes --]

Author: Stephane Eranian <eranian@google.com>

Events were not grouped anymore. The reason was that in
perf_event_open(), the field event->group_leader was
initialized before the function looked up the group_fd
to find the event leader. This patch fixes this by
reordering the code correctly.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <4c90ea16.21edd80a.4b94.2495@mx.google.com>
---
 kernel/perf_event.c |   15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -5543,17 +5543,11 @@ SYSCALL_DEFINE5(perf_event_open,
 	if (event_fd < 0)
 		return event_fd;
 
-	event = perf_event_alloc(&attr, cpu, group_leader, NULL, NULL);
-	if (IS_ERR(event)) {
-		err = PTR_ERR(event);
-		goto err_fd;
-	}
-
 	if (group_fd != -1) {
 		group_leader = perf_fget_light(group_fd, &fput_needed);
 		if (IS_ERR(group_leader)) {
 			err = PTR_ERR(group_leader);
-			goto err_alloc;
+			goto err_fd;
 		}
 		group_file = group_leader->filp;
 		if (flags & PERF_FLAG_FD_OUTPUT)
@@ -5562,6 +5556,12 @@ SYSCALL_DEFINE5(perf_event_open,
 			group_leader = NULL;
 	}
 
+	event = perf_event_alloc(&attr, cpu, group_leader, NULL, NULL);
+	if (IS_ERR(event)) {
+		err = PTR_ERR(event);
+		goto err_fd;
+	}
+
 	/*
 	 * Special case software events and allow them to be part of
 	 * any hardware group.
@@ -5643,7 +5643,6 @@ SYSCALL_DEFINE5(perf_event_open,
 	put_ctx(ctx);
 err_group_fd:
 	fput_light(group_file, fput_needed);
-err_alloc:
 	free_event(event);
 err_fd:
 	put_unused_fd(event_fd);

[tip:perf/core] perf_events: Fix broken event grouping

[tip-bot for Stephane Eranian]

Commit-ID:  d14b12d7adbf214f33eb59f800b5c3d5ed9268e8
Gitweb:     http://git.kernel.org/tip/d14b12d7adbf214f33eb59f800b5c3d5ed9268e8
Author:     Stephane Eranian <eranian@google.com>
AuthorDate: Fri, 17 Sep 2010 11:28:47 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Fri, 17 Sep 2010 12:48:47 +0200

perf_events: Fix broken event grouping

Events were not grouped anymore. The reason was that in
perf_event_open(), the field event->group_leader was
initialized before the function looked up the group_fd
to find the event leader. This patch fixes this by
reordering the code correctly.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <20100917093009.360420946@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/perf_event.c |   15 +++++++--------
 1 files changed, 7 insertions(+), 8 deletions(-)

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 86f394e..ce95617 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -5550,17 +5550,11 @@ SYSCALL_DEFINE5(perf_event_open,
 	if (event_fd < 0)
 		return event_fd;
 
-	event = perf_event_alloc(&attr, cpu, group_leader, NULL, NULL);
-	if (IS_ERR(event)) {
-		err = PTR_ERR(event);
-		goto err_fd;
-	}
-
 	if (group_fd != -1) {
 		group_leader = perf_fget_light(group_fd, &fput_needed);
 		if (IS_ERR(group_leader)) {
 			err = PTR_ERR(group_leader);
-			goto err_alloc;
+			goto err_fd;
 		}
 		group_file = group_leader->filp;
 		if (flags & PERF_FLAG_FD_OUTPUT)
@@ -5569,6 +5563,12 @@ SYSCALL_DEFINE5(perf_event_open,
 			group_leader = NULL;
 	}
 
+	event = perf_event_alloc(&attr, cpu, group_leader, NULL, NULL);
+	if (IS_ERR(event)) {
+		err = PTR_ERR(event);
+		goto err_fd;
+	}
+
 	/*
 	 * Special case software events and allow them to be part of
 	 * any hardware group.
@@ -5653,7 +5653,6 @@ err_context:
 	put_ctx(ctx);
 err_group_fd:
 	fput_light(group_file, fput_needed);
-err_alloc:
 	free_event(event);
 err_fd:
 	put_unused_fd(event_fd);

[PATCH 4/6] perf: Complete software pmu grouping

[Peter Zijlstra]

[-- Attachment #1: perf-cross-pmu-group.patch --]
[-- Type: text/plain, Size: 4722 bytes --]

Aside from allowing software events into a !software group, allow
adding !software events to pure software groups.

Once we've moved the software group and attached the first !software
event, the group will no longer be a pure software group and hence no
longer be eligible for movement, at which point the straight ctx
comparison is correct again.

Cc: Paul Mackerras <paulus@samba.org>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 include/linux/perf_event.h |    6 ++++
 kernel/perf_event.c        |   65 +++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 66 insertions(+), 5 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -5190,6 +5190,7 @@ int perf_pmu_register(struct pmu *pmu)
 
 		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
 		__perf_event_init_context(&cpuctx->ctx);
+		cpuctx->ctx.type = cpu_context;
 		cpuctx->ctx.pmu = pmu;
 		cpuctx->timer_interval = TICK_NSEC;
 		hrtimer_init(&cpuctx->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
@@ -5523,7 +5524,8 @@ SYSCALL_DEFINE5(perf_event_open,
 		struct perf_event_attr __user *, attr_uptr,
 		pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
 {
-	struct perf_event *event, *group_leader = NULL, *output_event = NULL;
+	struct perf_event *group_leader = NULL, *output_event = NULL;
+	struct perf_event *event, *sibling;
 	struct perf_event_attr attr;
 	struct perf_event_context *ctx;
 	struct file *event_file = NULL;
@@ -5531,6 +5533,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	struct task_struct *task = NULL;
 	struct pmu *pmu;
 	int event_fd;
+	int move_group = 0;
 	int fput_needed = 0;
 	int err;
 
@@ -5580,8 +5583,29 @@ SYSCALL_DEFINE5(perf_event_open,
 	 * any hardware group.
 	 */
 	pmu = event->pmu;
-	if ((pmu->task_ctx_nr == perf_sw_context) && group_leader)
-		pmu = group_leader->pmu;
+
+	if (group_leader &&
+	    (is_software_event(event) != is_software_event(group_leader))) {
+		if (is_software_event(event)) {
+			/*
+			 * If event and group_leader are not both a software
+			 * event, and event is, then group leader is not.
+			 *
+			 * Allow the addition of software events to !software
+			 * groups, this is safe because software events never
+			 * fail to schedule.
+			 */
+			pmu = group_leader->pmu;
+		} else if (is_software_event(group_leader) &&
+			   (group_leader->group_flags & PERF_GROUP_SOFTWARE)) {
+			/*
+			 * In case the group is a pure software group, and we
+			 * try to add a hardware event, move the whole group to
+			 * the hardware context.
+			 */
+			move_group = 1;
+		}
+	}
 
 	if (pid != -1)
 		task = find_lively_task_by_vpid(pid);
@@ -5611,8 +5635,14 @@ SYSCALL_DEFINE5(perf_event_open,
 		 * Do not allow to attach to a group in a different
 		 * task or CPU context:
 		 */
-		if (group_leader->ctx != ctx)
-			goto err_context;
+		if (move_group) {
+			if (group_leader->ctx->type != ctx->type)
+				goto err_context;
+		} else {
+			if (group_leader->ctx != ctx)
+				goto err_context;
+		}
+
 		/*
 		 * Only a group leader can be exclusive or pinned
 		 */
@@ -5632,9 +5662,34 @@ SYSCALL_DEFINE5(perf_event_open,
 		goto err_context;
 	}
 
+	if (move_group) {
+		struct perf_event_context *gctx = group_leader->ctx;
+
+		mutex_lock(&gctx->mutex);
+		perf_event_remove_from_context(group_leader);
+		list_for_each_entry(sibling, &group_leader->sibling_list,
+				    group_entry) {
+			perf_event_remove_from_context(sibling);
+			put_ctx(gctx);
+		}
+		mutex_unlock(&gctx->mutex);
+		put_ctx(gctx);
+	}
+
 	event->filp = event_file;
 	WARN_ON_ONCE(ctx->parent_ctx);
 	mutex_lock(&ctx->mutex);
+
+	if (move_group) {
+		perf_install_in_context(ctx, group_leader, cpu);
+		get_ctx(ctx);
+		list_for_each_entry(sibling, &group_leader->sibling_list,
+				    group_entry) {
+			perf_install_in_context(ctx, sibling, cpu);
+			get_ctx(ctx);
+		}
+	}
+
 	perf_install_in_context(ctx, event, cpu);
 	++ctx->generation;
 	mutex_unlock(&ctx->mutex);
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -804,12 +804,18 @@ struct perf_event {
 #endif /* CONFIG_PERF_EVENTS */
 };
 
+enum perf_event_context_type {
+	task_context,
+	cpu_context,
+};
+
 /**
  * struct perf_event_context - event context structure
  *
  * Used as a container for task events and CPU events as well:
  */
 struct perf_event_context {
+	enum perf_event_context_type	type;
 	struct pmu			*pmu;
 	/*
 	 * Protect the states of the events in the list,

[tip:perf/core] perf: Complete software pmu grouping

[tip-bot for Peter Zijlstra]

Commit-ID:  b04243ef7006cda301819f54ee7ce0a3632489e3
Gitweb:     http://git.kernel.org/tip/b04243ef7006cda301819f54ee7ce0a3632489e3
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Fri, 17 Sep 2010 11:28:48 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Fri, 17 Sep 2010 12:48:48 +0200

perf: Complete software pmu grouping

Aside from allowing software events into a !software group,
allow adding !software events to pure software groups.

Once we've moved the software group and attached the first
!software event, the group will no longer be a pure software
group and hence no longer be eligible for movement, at which
point the straight ctx comparison is correct again.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <20100917093009.410784731@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 include/linux/perf_event.h |    6 ++++
 kernel/perf_event.c        |   65 ++++++++++++++++++++++++++++++++++++++++---
 2 files changed, 66 insertions(+), 5 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 39d8860..165287f 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -804,12 +804,18 @@ struct perf_event {
 #endif /* CONFIG_PERF_EVENTS */
 };
 
+enum perf_event_context_type {
+	task_context,
+	cpu_context,
+};
+
 /**
  * struct perf_event_context - event context structure
  *
  * Used as a container for task events and CPU events as well:
  */
 struct perf_event_context {
+	enum perf_event_context_type	type;
 	struct pmu			*pmu;
 	/*
 	 * Protect the states of the events in the list,
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index ce95617..6d7eef5 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -5184,6 +5184,7 @@ int perf_pmu_register(struct pmu *pmu)
 
 		cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
 		__perf_event_init_context(&cpuctx->ctx);
+		cpuctx->ctx.type = cpu_context;
 		cpuctx->ctx.pmu = pmu;
 		cpuctx->timer_interval = TICK_NSEC;
 		hrtimer_init(&cpuctx->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
@@ -5517,7 +5518,8 @@ SYSCALL_DEFINE5(perf_event_open,
 		struct perf_event_attr __user *, attr_uptr,
 		pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
 {
-	struct perf_event *event, *group_leader = NULL, *output_event = NULL;
+	struct perf_event *group_leader = NULL, *output_event = NULL;
+	struct perf_event *event, *sibling;
 	struct perf_event_attr attr;
 	struct perf_event_context *ctx;
 	struct file *event_file = NULL;
@@ -5525,6 +5527,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	struct task_struct *task = NULL;
 	struct pmu *pmu;
 	int event_fd;
+	int move_group = 0;
 	int fput_needed = 0;
 	int err;
 
@@ -5574,8 +5577,29 @@ SYSCALL_DEFINE5(perf_event_open,
 	 * any hardware group.
 	 */
 	pmu = event->pmu;
-	if ((pmu->task_ctx_nr == perf_sw_context) && group_leader)
-		pmu = group_leader->pmu;
+
+	if (group_leader &&
+	    (is_software_event(event) != is_software_event(group_leader))) {
+		if (is_software_event(event)) {
+			/*
+			 * If event and group_leader are not both a software
+			 * event, and event is, then group leader is not.
+			 *
+			 * Allow the addition of software events to !software
+			 * groups, this is safe because software events never
+			 * fail to schedule.
+			 */
+			pmu = group_leader->pmu;
+		} else if (is_software_event(group_leader) &&
+			   (group_leader->group_flags & PERF_GROUP_SOFTWARE)) {
+			/*
+			 * In case the group is a pure software group, and we
+			 * try to add a hardware event, move the whole group to
+			 * the hardware context.
+			 */
+			move_group = 1;
+		}
+	}
 
 	if (pid != -1)
 		task = find_lively_task_by_vpid(pid);
@@ -5605,8 +5629,14 @@ SYSCALL_DEFINE5(perf_event_open,
 		 * Do not allow to attach to a group in a different
 		 * task or CPU context:
 		 */
-		if (group_leader->ctx != ctx)
-			goto err_context;
+		if (move_group) {
+			if (group_leader->ctx->type != ctx->type)
+				goto err_context;
+		} else {
+			if (group_leader->ctx != ctx)
+				goto err_context;
+		}
+
 		/*
 		 * Only a group leader can be exclusive or pinned
 		 */
@@ -5626,9 +5656,34 @@ SYSCALL_DEFINE5(perf_event_open,
 		goto err_context;
 	}
 
+	if (move_group) {
+		struct perf_event_context *gctx = group_leader->ctx;
+
+		mutex_lock(&gctx->mutex);
+		perf_event_remove_from_context(group_leader);
+		list_for_each_entry(sibling, &group_leader->sibling_list,
+				    group_entry) {
+			perf_event_remove_from_context(sibling);
+			put_ctx(gctx);
+		}
+		mutex_unlock(&gctx->mutex);
+		put_ctx(gctx);
+	}
+
 	event->filp = event_file;
 	WARN_ON_ONCE(ctx->parent_ctx);
 	mutex_lock(&ctx->mutex);
+
+	if (move_group) {
+		perf_install_in_context(ctx, group_leader, cpu);
+		get_ctx(ctx);
+		list_for_each_entry(sibling, &group_leader->sibling_list,
+				    group_entry) {
+			perf_install_in_context(ctx, sibling, cpu);
+			get_ctx(ctx);
+		}
+	}
+
 	perf_install_in_context(ctx, event, cpu);
 	++ctx->generation;
 	mutex_unlock(&ctx->mutex);

[PATCH 5/6] perf: Fix perf_event_exit_cpu_context()

[Peter Zijlstra]

[-- Attachment #1: perf-fix-unplug.patch --]
[-- Type: text/plain, Size: 865 bytes --]

Use the right cpu-context.. spotted by preempt warning on hot-unplug

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/perf_event.c |    3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -6259,14 +6259,13 @@ static void perf_event_exit_cpu_context(
 
 	idx = srcu_read_lock(&pmus_srcu);
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		ctx = &this_cpu_ptr(pmu->pmu_cpu_context)->ctx;
+		ctx = &per_cpu_ptr(pmu->pmu_cpu_context, cpu)->ctx;
 
 		mutex_lock(&ctx->mutex);
 		smp_call_function_single(cpu, __perf_event_exit_context, ctx, 1);
 		mutex_unlock(&ctx->mutex);
 	}
 	srcu_read_unlock(&pmus_srcu, idx);
-
 }
 
 static void perf_event_exit_cpu(int cpu)

[tip:perf/core] perf: Fix perf_event_exit_cpu_context()

[tip-bot for Peter Zijlstra]

Commit-ID:  917bdd1c9b7b0f4c22f2504c2f0c1074c8ab9df7
Gitweb:     http://git.kernel.org/tip/917bdd1c9b7b0f4c22f2504c2f0c1074c8ab9df7
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Fri, 17 Sep 2010 11:28:49 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Fri, 17 Sep 2010 12:48:48 +0200

perf: Fix perf_event_exit_cpu_context()

Use the right cpu-context.. spotted by preempt warning on
hot-unplug

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <20100917093009.461794357@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/perf_event.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 6d7eef5..27332e5 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -6269,14 +6269,13 @@ static void perf_event_exit_cpu_context(int cpu)
 
 	idx = srcu_read_lock(&pmus_srcu);
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		ctx = &this_cpu_ptr(pmu->pmu_cpu_context)->ctx;
+		ctx = &per_cpu_ptr(pmu->pmu_cpu_context, cpu)->ctx;
 
 		mutex_lock(&ctx->mutex);
 		smp_call_function_single(cpu, __perf_event_exit_context, ctx, 1);
 		mutex_unlock(&ctx->mutex);
 	}
 	srcu_read_unlock(&pmus_srcu, idx);
-
 }
 
 static void perf_event_exit_cpu(int cpu)

[PATCH 6/6] perf: Undo the per cpu-context timer stuff

[Peter Zijlstra]

[-- Attachment #1: perf-fix-rotation-timers.patch --]
[-- Type: text/plain, Size: 6736 bytes --]

Revert the timer per cpu-context timers because of unfortunate nohz
interaction. Fixing that would have been somewhat ugly, so go back to
driving things from the regular tick. Provide a jiffies interval
feature for people who want slower rotations.

Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 include/linux/perf_event.h |    6 ++-
 kernel/perf_event.c        |   79 +++++++++++++++++++++++++++------------------
 kernel/sched.c             |    2 +
 3 files changed, 55 insertions(+), 32 deletions(-)

Index: linux-2.6/kernel/perf_event.c
===================================================================
--- linux-2.6.orig/kernel/perf_event.c
+++ linux-2.6/kernel/perf_event.c
@@ -77,23 +77,22 @@ void perf_pmu_enable(struct pmu *pmu)
 		pmu->pmu_enable(pmu);
 }
 
+static DEFINE_PER_CPU(struct list_head, rotation_list);
+
+/*
+ * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
+ * because they're strictly cpu affine and rotate_start is called with IRQs
+ * disabled, while rotate_context is called from IRQ context.
+ */
 static void perf_pmu_rotate_start(struct pmu *pmu)
 {
 	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+	struct list_head *head = &__get_cpu_var(rotation_list);
 
-	if (hrtimer_active(&cpuctx->timer))
-		return;
+	WARN_ON(!irqs_disabled());
 
-	__hrtimer_start_range_ns(&cpuctx->timer,
-			ns_to_ktime(cpuctx->timer_interval), 0,
-			HRTIMER_MODE_REL_PINNED, 0);
-}
-
-static void perf_pmu_rotate_stop(struct pmu *pmu)
-{
-	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-
-	hrtimer_cancel(&cpuctx->timer);
+	if (list_empty(&cpuctx->rotation_list))
+		list_add(&cpuctx->rotation_list, head);
 }
 
 static void get_ctx(struct perf_event_context *ctx)
@@ -1607,36 +1606,33 @@ static void rotate_ctx(struct perf_event
 }
 
 /*
- * Cannot race with ->pmu_rotate_start() because this is ran from hardirq
- * context, and ->pmu_rotate_start() is called with irqs disabled (both are
- * cpu affine, so there are no SMP races).
+ * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
+ * because they're strictly cpu affine and rotate_start is called with IRQs
+ * disabled, while rotate_context is called from IRQ context.
  */
-static enum hrtimer_restart perf_event_context_tick(struct hrtimer *timer)
+static void perf_rotate_context(struct perf_cpu_context *cpuctx)
 {
-	enum hrtimer_restart restart = HRTIMER_NORESTART;
-	struct perf_cpu_context *cpuctx;
+	u64 interval = (u64)cpuctx->jiffies_interval * TICK_NSEC;
 	struct perf_event_context *ctx = NULL;
-	int rotate = 0;
-
-	cpuctx = container_of(timer, struct perf_cpu_context, timer);
+	int rotate = 0, remove = 1;
 
 	if (cpuctx->ctx.nr_events) {
-		restart = HRTIMER_RESTART;
+		remove = 0;
 		if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
 			rotate = 1;
 	}
 
 	ctx = cpuctx->task_ctx;
 	if (ctx && ctx->nr_events) {
-		restart = HRTIMER_RESTART;
+		remove = 0;
 		if (ctx->nr_events != ctx->nr_active)
 			rotate = 1;
 	}
 
 	perf_pmu_disable(cpuctx->ctx.pmu);
-	perf_ctx_adjust_freq(&cpuctx->ctx, cpuctx->timer_interval);
+	perf_ctx_adjust_freq(&cpuctx->ctx, interval);
 	if (ctx)
-		perf_ctx_adjust_freq(ctx, cpuctx->timer_interval);
+		perf_ctx_adjust_freq(ctx, interval);
 
 	if (!rotate)
 		goto done;
@@ -1654,10 +1650,24 @@ static enum hrtimer_restart perf_event_c
 		task_ctx_sched_in(ctx, EVENT_FLEXIBLE);
 
 done:
+	if (remove)
+		list_del_init(&cpuctx->rotation_list);
+
 	perf_pmu_enable(cpuctx->ctx.pmu);
-	hrtimer_forward_now(timer, ns_to_ktime(cpuctx->timer_interval));
+}
+
+void perf_event_task_tick(void)
+{
+	struct list_head *head = &__get_cpu_var(rotation_list);
+	struct perf_cpu_context *cpuctx, *tmp;
 
-	return restart;
+	WARN_ON(!irqs_disabled());
+
+	list_for_each_entry_safe(cpuctx, tmp, head, rotation_list) {
+		if (cpuctx->jiffies_interval == 1 ||
+				!(jiffies % cpuctx->jiffies_interval))
+			perf_rotate_context(cpuctx);
+	}
 }
 
 static int event_enable_on_exec(struct perf_event *event,
@@ -5180,9 +5190,8 @@ int perf_pmu_register(struct pmu *pmu)
 		__perf_event_init_context(&cpuctx->ctx);
 		cpuctx->ctx.type = cpu_context;
 		cpuctx->ctx.pmu = pmu;
-		cpuctx->timer_interval = TICK_NSEC;
-		hrtimer_init(&cpuctx->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
-		cpuctx->timer.function = perf_event_context_tick;
+		cpuctx->jiffies_interval = 1;
+		INIT_LIST_HEAD(&cpuctx->rotation_list);
 	}
 
 got_cpu_context:
@@ -6219,6 +6228,7 @@ static void __init perf_event_init_all_c
 	for_each_possible_cpu(cpu) {
 		swhash = &per_cpu(swevent_htable, cpu);
 		mutex_init(&swhash->hlist_mutex);
+		INIT_LIST_HEAD(&per_cpu(rotation_list, cpu));
 	}
 }
 
@@ -6238,6 +6248,15 @@ static void __cpuinit perf_event_init_cp
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
+static void perf_pmu_rotate_stop(struct pmu *pmu)
+{
+	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+
+	WARN_ON(!irqs_disabled());
+
+	list_del_init(&cpuctx->rotation_list);
+}
+
 static void __perf_event_exit_context(void *__info)
 {
 	struct perf_event_context *ctx = __info;
Index: linux-2.6/include/linux/perf_event.h
===================================================================
--- linux-2.6.orig/include/linux/perf_event.h
+++ linux-2.6/include/linux/perf_event.h
@@ -870,8 +870,8 @@ struct perf_cpu_context {
 	struct perf_event_context	*task_ctx;
 	int				active_oncpu;
 	int				exclusive;
-	u64				timer_interval;
-	struct hrtimer			timer;
+	struct list_head		rotation_list;
+	int				jiffies_interval;
 };
 
 struct perf_output_handle {
@@ -1065,6 +1065,7 @@ extern int perf_swevent_get_recursion_co
 extern void perf_swevent_put_recursion_context(int rctx);
 extern void perf_event_enable(struct perf_event *event);
 extern void perf_event_disable(struct perf_event *event);
+extern void perf_event_task_tick(void);
 #else
 static inline void
 perf_event_task_sched_in(struct task_struct *task)			{ }
@@ -1099,6 +1100,7 @@ static inline int  perf_swevent_get_recu
 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
 static inline void perf_event_enable(struct perf_event *event)		{ }
 static inline void perf_event_disable(struct perf_event *event)		{ }
+static inline void perf_event_task_tick(void)				{ }
 #endif
 
 #define perf_output_put(handle, x) \
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -3581,6 +3581,8 @@ void scheduler_tick(void)
 	curr->sched_class->task_tick(rq, curr, 0);
 	raw_spin_unlock(&rq->lock);
 
+	perf_event_task_tick();
+
 #ifdef CONFIG_SMP
 	rq->idle_at_tick = idle_cpu(cpu);
 	trigger_load_balance(rq, cpu);

[tip:perf/core] perf: Undo the per cpu-context timer stuff

[tip-bot for Peter Zijlstra]

Commit-ID:  e9d2b064149ff7ef4acbc65a1b9374ac8b218d3e
Gitweb:     http://git.kernel.org/tip/e9d2b064149ff7ef4acbc65a1b9374ac8b218d3e
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Fri, 17 Sep 2010 11:28:50 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Fri, 17 Sep 2010 12:48:48 +0200

perf: Undo the per cpu-context timer stuff

Revert the timer per cpu-context timers because of unfortunate
nohz interaction. Fixing that would have been somewhat ugly, so
go back to driving things from the regular tick. Provide a
jiffies interval feature for people who want slower rotations.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephane Eranian <eranian@google.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Yinghai Lu <yinghai@kernel.org>
LKML-Reference: <20100917093009.519845633@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 include/linux/perf_event.h |    6 ++-
 kernel/perf_event.c        |   79 +++++++++++++++++++++++++++----------------
 kernel/sched.c             |    2 +
 3 files changed, 55 insertions(+), 32 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 165287f..61b1e2d 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -870,8 +870,8 @@ struct perf_cpu_context {
 	struct perf_event_context	*task_ctx;
 	int				active_oncpu;
 	int				exclusive;
-	u64				timer_interval;
-	struct hrtimer			timer;
+	struct list_head		rotation_list;
+	int				jiffies_interval;
 };
 
 struct perf_output_handle {
@@ -1065,6 +1065,7 @@ extern int perf_swevent_get_recursion_context(void);
 extern void perf_swevent_put_recursion_context(int rctx);
 extern void perf_event_enable(struct perf_event *event);
 extern void perf_event_disable(struct perf_event *event);
+extern void perf_event_task_tick(void);
 #else
 static inline void
 perf_event_task_sched_in(struct task_struct *task)			{ }
@@ -1099,6 +1100,7 @@ static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
 static inline void perf_event_enable(struct perf_event *event)		{ }
 static inline void perf_event_disable(struct perf_event *event)		{ }
+static inline void perf_event_task_tick(void)				{ }
 #endif
 
 #define perf_output_put(handle, x) \
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 27332e5..baae136 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -77,23 +77,22 @@ void perf_pmu_enable(struct pmu *pmu)
 		pmu->pmu_enable(pmu);
 }
 
+static DEFINE_PER_CPU(struct list_head, rotation_list);
+
+/*
+ * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
+ * because they're strictly cpu affine and rotate_start is called with IRQs
+ * disabled, while rotate_context is called from IRQ context.
+ */
 static void perf_pmu_rotate_start(struct pmu *pmu)
 {
 	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+	struct list_head *head = &__get_cpu_var(rotation_list);
 
-	if (hrtimer_active(&cpuctx->timer))
-		return;
+	WARN_ON(!irqs_disabled());
 
-	__hrtimer_start_range_ns(&cpuctx->timer,
-			ns_to_ktime(cpuctx->timer_interval), 0,
-			HRTIMER_MODE_REL_PINNED, 0);
-}
-
-static void perf_pmu_rotate_stop(struct pmu *pmu)
-{
-	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-
-	hrtimer_cancel(&cpuctx->timer);
+	if (list_empty(&cpuctx->rotation_list))
+		list_add(&cpuctx->rotation_list, head);
 }
 
 static void get_ctx(struct perf_event_context *ctx)
@@ -1607,36 +1606,33 @@ static void rotate_ctx(struct perf_event_context *ctx)
 }
 
 /*
- * Cannot race with ->pmu_rotate_start() because this is ran from hardirq
- * context, and ->pmu_rotate_start() is called with irqs disabled (both are
- * cpu affine, so there are no SMP races).
+ * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
+ * because they're strictly cpu affine and rotate_start is called with IRQs
+ * disabled, while rotate_context is called from IRQ context.
  */
-static enum hrtimer_restart perf_event_context_tick(struct hrtimer *timer)
+static void perf_rotate_context(struct perf_cpu_context *cpuctx)
 {
-	enum hrtimer_restart restart = HRTIMER_NORESTART;
-	struct perf_cpu_context *cpuctx;
+	u64 interval = (u64)cpuctx->jiffies_interval * TICK_NSEC;
 	struct perf_event_context *ctx = NULL;
-	int rotate = 0;
-
-	cpuctx = container_of(timer, struct perf_cpu_context, timer);
+	int rotate = 0, remove = 1;
 
 	if (cpuctx->ctx.nr_events) {
-		restart = HRTIMER_RESTART;
+		remove = 0;
 		if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
 			rotate = 1;
 	}
 
 	ctx = cpuctx->task_ctx;
 	if (ctx && ctx->nr_events) {
-		restart = HRTIMER_RESTART;
+		remove = 0;
 		if (ctx->nr_events != ctx->nr_active)
 			rotate = 1;
 	}
 
 	perf_pmu_disable(cpuctx->ctx.pmu);
-	perf_ctx_adjust_freq(&cpuctx->ctx, cpuctx->timer_interval);
+	perf_ctx_adjust_freq(&cpuctx->ctx, interval);
 	if (ctx)
-		perf_ctx_adjust_freq(ctx, cpuctx->timer_interval);
+		perf_ctx_adjust_freq(ctx, interval);
 
 	if (!rotate)
 		goto done;
@@ -1654,10 +1650,24 @@ static enum hrtimer_restart perf_event_context_tick(struct hrtimer *timer)
 		task_ctx_sched_in(ctx, EVENT_FLEXIBLE);
 
 done:
+	if (remove)
+		list_del_init(&cpuctx->rotation_list);
+
 	perf_pmu_enable(cpuctx->ctx.pmu);
-	hrtimer_forward_now(timer, ns_to_ktime(cpuctx->timer_interval));
+}
+
+void perf_event_task_tick(void)
+{
+	struct list_head *head = &__get_cpu_var(rotation_list);
+	struct perf_cpu_context *cpuctx, *tmp;
 
-	return restart;
+	WARN_ON(!irqs_disabled());
+
+	list_for_each_entry_safe(cpuctx, tmp, head, rotation_list) {
+		if (cpuctx->jiffies_interval == 1 ||
+				!(jiffies % cpuctx->jiffies_interval))
+			perf_rotate_context(cpuctx);
+	}
 }
 
 static int event_enable_on_exec(struct perf_event *event,
@@ -5186,9 +5196,8 @@ int perf_pmu_register(struct pmu *pmu)
 		__perf_event_init_context(&cpuctx->ctx);
 		cpuctx->ctx.type = cpu_context;
 		cpuctx->ctx.pmu = pmu;
-		cpuctx->timer_interval = TICK_NSEC;
-		hrtimer_init(&cpuctx->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
-		cpuctx->timer.function = perf_event_context_tick;
+		cpuctx->jiffies_interval = 1;
+		INIT_LIST_HEAD(&cpuctx->rotation_list);
 	}
 
 got_cpu_context:
@@ -6229,6 +6238,7 @@ static void __init perf_event_init_all_cpus(void)
 	for_each_possible_cpu(cpu) {
 		swhash = &per_cpu(swevent_htable, cpu);
 		mutex_init(&swhash->hlist_mutex);
+		INIT_LIST_HEAD(&per_cpu(rotation_list, cpu));
 	}
 }
 
@@ -6248,6 +6258,15 @@ static void __cpuinit perf_event_init_cpu(int cpu)
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
+static void perf_pmu_rotate_stop(struct pmu *pmu)
+{
+	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
+
+	WARN_ON(!irqs_disabled());
+
+	list_del_init(&cpuctx->rotation_list);
+}
+
 static void __perf_event_exit_context(void *__info)
 {
 	struct perf_event_context *ctx = __info;
diff --git a/kernel/sched.c b/kernel/sched.c
index 1c3ea7a..794819e 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -3584,6 +3584,8 @@ void scheduler_tick(void)
 	curr->sched_class->task_tick(rq, curr, 0);
 	raw_spin_unlock(&rq->lock);
 
+	perf_event_task_tick();
+
 #ifdef CONFIG_SMP
 	rq->idle_at_tick = idle_cpu(cpu);
 	trigger_load_balance(rq, cpu);