Merge tag 'arm-perf-3.19' of git://git.kernel.org/pub/scm/linux/kernel/git/will/linux into next/drivers

Pull "ARM: perf: updates for 3.19" from Will Deacon:

This patch series takes us slightly further on the road to big.LITTLE
support in perf. The main change enabling this is moving the CCI PMU
driver away from the arm-pmu abstraction, allowing the arch code to
focus specifically on support for CPU PMUs.

* tag 'arm-perf-3.19' of git://git.kernel.org/pub/scm/linux/kernel/git/will/linux:
  arm: perf: fold hotplug notifier into arm_pmu
  arm: perf: dynamically allocate cpu hardware data
  arm: perf: fold percpu_pmu into pmu_hw_events
  arm: perf: kill get_hw_events()
  arm: perf: limit size of accounting data
  arm: perf: use IDR types for CPU PMUs
  arm: perf: make PMU probing data-driven
  arm: perf: add missing pr_info newlines
  arm: perf: factor out callchain code
  ARM: perf: use pr_* instead of printk
  ARM: perf: remove useless return and check of idx in counter handling
  bus: cci: move away from arm_pmu framework

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

10 files changed, 772 insertions, 403 deletions

diff --git a/arch/arm/include/asm/perf_event.h b/arch/arm/include/asm/perf_event.h
index c3a83691af8e..d9cf138fd7d4 100644
--- a/arch/arm/include/asm/perf_event.h
+++ b/arch/arm/include/asm/perf_event.h
@@ -12,7 +12,7 @@
 #ifndef __ARM_PERF_EVENT_H__
 #define __ARM_PERF_EVENT_H__
 
-#ifdef CONFIG_HW_PERF_EVENTS
+#ifdef CONFIG_PERF_EVENTS
 struct pt_regs;
 extern unsigned long perf_instruction_pointer(struct pt_regs *regs);
 extern unsigned long perf_misc_flags(struct pt_regs *regs);
diff --git a/arch/arm/include/asm/pmu.h b/arch/arm/include/asm/pmu.h
index 0b648c541293..b1596bd59129 100644
--- a/arch/arm/include/asm/pmu.h
+++ b/arch/arm/include/asm/pmu.h
@@ -15,6 +15,8 @@
 #include <linux/interrupt.h>
 #include <linux/perf_event.h>
 
+#include <asm/cputype.h>
+
 /*
  * struct arm_pmu_platdata - ARM PMU platform data
  *
@@ -66,19 +68,25 @@ struct pmu_hw_events {
 	/*
 	 * The events that are active on the PMU for the given index.
 	 */
-	struct perf_event	**events;
+	struct perf_event	*events[ARMPMU_MAX_HWEVENTS];
 
 	/*
 	 * A 1 bit for an index indicates that the counter is being used for
 	 * an event. A 0 means that the counter can be used.
 	 */
-	unsigned long           *used_mask;
+	DECLARE_BITMAP(used_mask, ARMPMU_MAX_HWEVENTS);
 
 	/*
 	 * Hardware lock to serialize accesses to PMU registers. Needed for the
 	 * read/modify/write sequences.
 	 */
 	raw_spinlock_t		pmu_lock;
+
+	/*
+	 * When using percpu IRQs, we need a percpu dev_id. Place it here as we
+	 * already have to allocate this struct per cpu.
+	 */
+	struct arm_pmu		*percpu_pmu;
 };
 
 struct arm_pmu {
@@ -107,7 +115,8 @@ struct arm_pmu {
 	struct mutex	reserve_mutex;
 	u64		max_period;
 	struct platform_device	*plat_device;
-	struct pmu_hw_events	*(*get_hw_events)(void);
+	struct pmu_hw_events	__percpu *hw_events;
+	struct notifier_block	hotplug_nb;
 };
 
 #define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu))
@@ -127,6 +136,27 @@ int armpmu_map_event(struct perf_event *event,
 						[PERF_COUNT_HW_CACHE_RESULT_MAX],
 		     u32 raw_event_mask);
 
+struct pmu_probe_info {
+	unsigned int cpuid;
+	unsigned int mask;
+	int (*init)(struct arm_pmu *);
+};
+
+#define PMU_PROBE(_cpuid, _mask, _fn)	\
+{					\
+	.cpuid = (_cpuid),		\
+	.mask = (_mask),		\
+	.init = (_fn),			\
+}
+
+#define ARM_PMU_PROBE(_cpuid, _fn) \
+	PMU_PROBE(_cpuid, ARM_CPU_PART_MASK, _fn)
+
+#define ARM_PMU_XSCALE_MASK	((0xff << 24) | ARM_CPU_XSCALE_ARCH_MASK)
+
+#define XSCALE_PMU_PROBE(_version, _fn) \
+	PMU_PROBE(ARM_CPU_IMP_INTEL << 24 | _version, ARM_PMU_XSCALE_MASK, _fn)
+
 #endif /* CONFIG_HW_PERF_EVENTS */
 
 #endif /* __ARM_PMU_H__ */
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 38ddd9f83d0e..8dcbed5016ac 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -82,7 +82,7 @@ obj-$(CONFIG_CPU_MOHAWK)	+= xscale-cp0.o
 obj-$(CONFIG_CPU_PJ4)		+= pj4-cp0.o
 obj-$(CONFIG_CPU_PJ4B)		+= pj4-cp0.o
 obj-$(CONFIG_IWMMXT)		+= iwmmxt.o
-obj-$(CONFIG_PERF_EVENTS)	+= perf_regs.o
+obj-$(CONFIG_PERF_EVENTS)	+= perf_regs.o perf_callchain.o
 obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event.o perf_event_cpu.o
 AFLAGS_iwmmxt.o			:= -Wa,-mcpu=iwmmxt
 obj-$(CONFIG_ARM_CPU_TOPOLOGY)  += topology.o
diff --git a/arch/arm/kernel/perf_callchain.c b/arch/arm/kernel/perf_callchain.c
new file mode 100644
index 000000000000..4e02ae5950ff
--- /dev/null
+++ b/arch/arm/kernel/perf_callchain.c
@@ -0,0 +1,136 @@
+/*
+ * ARM callchain support
+ *
+ * Copyright (C) 2009 picoChip Designs, Ltd., Jamie Iles
+ * Copyright (C) 2010 ARM Ltd., Will Deacon <will.deacon@arm.com>
+ *
+ * This code is based on the ARM OProfile backtrace code.
+ */
+#include <linux/perf_event.h>
+#include <linux/uaccess.h>
+
+#include <asm/stacktrace.h>
+
+/*
+ * The registers we're interested in are at the end of the variable
+ * length saved register structure. The fp points at the end of this
+ * structure so the address of this struct is:
+ * (struct frame_tail *)(xxx->fp)-1
+ *
+ * This code has been adapted from the ARM OProfile support.
+ */
+struct frame_tail {
+	struct frame_tail __user *fp;
+	unsigned long sp;
+	unsigned long lr;
+} __attribute__((packed));
+
+/*
+ * Get the return address for a single stackframe and return a pointer to the
+ * next frame tail.
+ */
+static struct frame_tail __user *
+user_backtrace(struct frame_tail __user *tail,
+	       struct perf_callchain_entry *entry)
+{
+	struct frame_tail buftail;
+	unsigned long err;
+
+	if (!access_ok(VERIFY_READ, tail, sizeof(buftail)))
+		return NULL;
+
+	pagefault_disable();
+	err = __copy_from_user_inatomic(&buftail, tail, sizeof(buftail));
+	pagefault_enable();
+
+	if (err)
+		return NULL;
+
+	perf_callchain_store(entry, buftail.lr);
+
+	/*
+	 * Frame pointers should strictly progress back up the stack
+	 * (towards higher addresses).
+	 */
+	if (tail + 1 >= buftail.fp)
+		return NULL;
+
+	return buftail.fp - 1;
+}
+
+void
+perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
+{
+	struct frame_tail __user *tail;
+
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* We don't support guest os callchain now */
+		return;
+	}
+
+	perf_callchain_store(entry, regs->ARM_pc);
+
+	if (!current->mm)
+		return;
+
+	tail = (struct frame_tail __user *)regs->ARM_fp - 1;
+
+	while ((entry->nr < PERF_MAX_STACK_DEPTH) &&
+	       tail && !((unsigned long)tail & 0x3))
+		tail = user_backtrace(tail, entry);
+}
+
+/*
+ * Gets called by walk_stackframe() for every stackframe. This will be called
+ * whist unwinding the stackframe and is like a subroutine return so we use
+ * the PC.
+ */
+static int
+callchain_trace(struct stackframe *fr,
+		void *data)
+{
+	struct perf_callchain_entry *entry = data;
+	perf_callchain_store(entry, fr->pc);
+	return 0;
+}
+
+void
+perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
+{
+	struct stackframe fr;
+
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		/* We don't support guest os callchain now */
+		return;
+	}
+
+	arm_get_current_stackframe(regs, &fr);
+	walk_stackframe(&fr, callchain_trace, entry);
+}
+
+unsigned long perf_instruction_pointer(struct pt_regs *regs)
+{
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest())
+		return perf_guest_cbs->get_guest_ip();
+
+	return instruction_pointer(regs);
+}
+
+unsigned long perf_misc_flags(struct pt_regs *regs)
+{
+	int misc = 0;
+
+	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
+		if (perf_guest_cbs->is_user_mode())
+			misc |= PERF_RECORD_MISC_GUEST_USER;
+		else
+			misc |= PERF_RECORD_MISC_GUEST_KERNEL;
+	} else {
+		if (user_mode(regs))
+			misc |= PERF_RECORD_MISC_USER;
+		else
+			misc |= PERF_RECORD_MISC_KERNEL;
+	}
+
+	return misc;
+}
diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index 266cba46db3e..e34934f63a49 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -7,21 +7,18 @@
  * Copyright (C) 2010 ARM Ltd., Will Deacon <will.deacon@arm.com>
  *
  * This code is based on the sparc64 perf event code, which is in turn based
- * on the x86 code. Callchain code is based on the ARM OProfile backtrace
- * code.
+ * on the x86 code.
  */
 #define pr_fmt(fmt) "hw perfevents: " fmt
 
 #include <linux/kernel.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
-#include <linux/uaccess.h>
 #include <linux/irq.h>
 #include <linux/irqdesc.h>
 
 #include <asm/irq_regs.h>
 #include <asm/pmu.h>
-#include <asm/stacktrace.h>
 
 static int
 armpmu_map_cache_event(const unsigned (*cache_map)
@@ -80,8 +77,12 @@ armpmu_map_event(struct perf_event *event,
 		 u32 raw_event_mask)
 {
 	u64 config = event->attr.config;
+	int type = event->attr.type;
 
-	switch (event->attr.type) {
+	if (type == event->pmu->type)
+		return armpmu_map_raw_event(raw_event_mask, config);
+
+	switch (type) {
 	case PERF_TYPE_HARDWARE:
 		return armpmu_map_hw_event(event_map, config);
 	case PERF_TYPE_HW_CACHE:
@@ -200,7 +201,7 @@ static void
 armpmu_del(struct perf_event *event, int flags)
 {
 	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *hw_events = armpmu->get_hw_events();
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 
@@ -217,7 +218,7 @@ static int
 armpmu_add(struct perf_event *event, int flags)
 {
 	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *hw_events = armpmu->get_hw_events();
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
 	struct hw_perf_event *hwc = &event->hw;
 	int idx;
 	int err = 0;
@@ -274,14 +275,12 @@ validate_group(struct perf_event *event)
 {
 	struct perf_event *sibling, *leader = event->group_leader;
 	struct pmu_hw_events fake_pmu;
-	DECLARE_BITMAP(fake_used_mask, ARMPMU_MAX_HWEVENTS);
 
 	/*
 	 * Initialise the fake PMU. We only need to populate the
 	 * used_mask for the purposes of validation.
 	 */
-	memset(fake_used_mask, 0, sizeof(fake_used_mask));
-	fake_pmu.used_mask = fake_used_mask;
+	memset(&fake_pmu.used_mask, 0, sizeof(fake_pmu.used_mask));
 
 	if (!validate_event(&fake_pmu, leader))
 		return -EINVAL;
@@ -305,17 +304,21 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
 	int ret;
 	u64 start_clock, finish_clock;
 
-	if (irq_is_percpu(irq))
-		dev = *(void **)dev;
-	armpmu = dev;
+	/*
+	 * we request the IRQ with a (possibly percpu) struct arm_pmu**, but
+	 * the handlers expect a struct arm_pmu*. The percpu_irq framework will
+	 * do any necessary shifting, we just need to perform the first
+	 * dereference.
+	 */
+	armpmu = *(void **)dev;
 	plat_device = armpmu->plat_device;
 	plat = dev_get_platdata(&plat_device->dev);
 
 	start_clock = sched_clock();
 	if (plat && plat->handle_irq)
-		ret = plat->handle_irq(irq, dev, armpmu->handle_irq);
+		ret = plat->handle_irq(irq, armpmu, armpmu->handle_irq);
 	else
-		ret = armpmu->handle_irq(irq, dev);
+		ret = armpmu->handle_irq(irq, armpmu);
 	finish_clock = sched_clock();
 
 	perf_sample_event_took(finish_clock - start_clock);
@@ -468,7 +471,7 @@ static int armpmu_event_init(struct perf_event *event)
 static void armpmu_enable(struct pmu *pmu)
 {
 	struct arm_pmu *armpmu = to_arm_pmu(pmu);
-	struct pmu_hw_events *hw_events = armpmu->get_hw_events();
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
 	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);
 
 	if (enabled)
@@ -533,130 +536,3 @@ int armpmu_register(struct arm_pmu *armpmu, int type)
 	return perf_pmu_register(&armpmu->pmu, armpmu->name, type);
 }
 
-/*
- * Callchain handling code.
- */
-
-/*
- * The registers we're interested in are at the end of the variable
- * length saved register structure. The fp points at the end of this
- * structure so the address of this struct is:
- * (struct frame_tail *)(xxx->fp)-1
- *
- * This code has been adapted from the ARM OProfile support.
- */
-struct frame_tail {
-	struct frame_tail __user *fp;
-	unsigned long sp;
-	unsigned long lr;
-} __attribute__((packed));
-
-/*
- * Get the return address for a single stackframe and return a pointer to the
- * next frame tail.
- */
-static struct frame_tail __user *
-user_backtrace(struct frame_tail __user *tail,
-	       struct perf_callchain_entry *entry)
-{
-	struct frame_tail buftail;
-	unsigned long err;
-
-	if (!access_ok(VERIFY_READ, tail, sizeof(buftail)))
-		return NULL;
-
-	pagefault_disable();
-	err = __copy_from_user_inatomic(&buftail, tail, sizeof(buftail));
-	pagefault_enable();
-
-	if (err)
-		return NULL;
-
-	perf_callchain_store(entry, buftail.lr);
-
-	/*
-	 * Frame pointers should strictly progress back up the stack
-	 * (towards higher addresses).
-	 */
-	if (tail + 1 >= buftail.fp)
-		return NULL;
-
-	return buftail.fp - 1;
-}
-
-void
-perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
-{
-	struct frame_tail __user *tail;
-
-	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
-		/* We don't support guest os callchain now */
-		return;
-	}
-
-	perf_callchain_store(entry, regs->ARM_pc);
-
-	if (!current->mm)
-		return;
-
-	tail = (struct frame_tail __user *)regs->ARM_fp - 1;
-
-	while ((entry->nr < PERF_MAX_STACK_DEPTH) &&
-	       tail && !((unsigned long)tail & 0x3))
-		tail = user_backtrace(tail, entry);
-}
-
-/*
- * Gets called by walk_stackframe() for every stackframe. This will be called
- * whist unwinding the stackframe and is like a subroutine return so we use
- * the PC.
- */
-static int
-callchain_trace(struct stackframe *fr,
-		void *data)
-{
-	struct perf_callchain_entry *entry = data;
-	perf_callchain_store(entry, fr->pc);
-	return 0;
-}
-
-void
-perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
-{
-	struct stackframe fr;
-
-	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
-		/* We don't support guest os callchain now */
-		return;
-	}
-
-	arm_get_current_stackframe(regs, &fr);
-	walk_stackframe(&fr, callchain_trace, entry);
-}
-
-unsigned long perf_instruction_pointer(struct pt_regs *regs)
-{
-	if (perf_guest_cbs && perf_guest_cbs->is_in_guest())
-		return perf_guest_cbs->get_guest_ip();
-
-	return instruction_pointer(regs);
-}
-
-unsigned long perf_misc_flags(struct pt_regs *regs)
-{
-	int misc = 0;
-
-	if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
-		if (perf_guest_cbs->is_user_mode())
-			misc |= PERF_RECORD_MISC_GUEST_USER;
-		else
-			misc |= PERF_RECORD_MISC_GUEST_KERNEL;
-	} else {
-		if (user_mode(regs))
-			misc |= PERF_RECORD_MISC_USER;
-		else
-			misc |= PERF_RECORD_MISC_KERNEL;
-	}
-
-	return misc;
-}
diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c
index eb2c4d55666b..dd9acc95ebc0 100644
--- a/arch/arm/kernel/perf_event_cpu.c
+++ b/arch/arm/kernel/perf_event_cpu.c
@@ -35,11 +35,6 @@
 /* Set at runtime when we know what CPU type we are. */
 static struct arm_pmu *cpu_pmu;
 
-static DEFINE_PER_CPU(struct arm_pmu *, percpu_pmu);
-static DEFINE_PER_CPU(struct perf_event * [ARMPMU_MAX_HWEVENTS], hw_events);
-static DEFINE_PER_CPU(unsigned long [BITS_TO_LONGS(ARMPMU_MAX_HWEVENTS)], used_mask);
-static DEFINE_PER_CPU(struct pmu_hw_events, cpu_hw_events);
-
 /*
  * Despite the names, these two functions are CPU-specific and are used
  * by the OProfile/perf code.
@@ -69,11 +64,6 @@ EXPORT_SYMBOL_GPL(perf_num_counters);
 #include "perf_event_v6.c"
 #include "perf_event_v7.c"
 
-static struct pmu_hw_events *cpu_pmu_get_cpu_events(void)
-{
-	return this_cpu_ptr(&cpu_hw_events);
-}
-
 static void cpu_pmu_enable_percpu_irq(void *data)
 {
 	int irq = *(int *)data;
@@ -92,20 +82,21 @@ static void cpu_pmu_free_irq(struct arm_pmu *cpu_pmu)
 {
 	int i, irq, irqs;
 	struct platform_device *pmu_device = cpu_pmu->plat_device;
+	struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
 
 	irqs = min(pmu_device->num_resources, num_possible_cpus());
 
 	irq = platform_get_irq(pmu_device, 0);
 	if (irq >= 0 && irq_is_percpu(irq)) {
 		on_each_cpu(cpu_pmu_disable_percpu_irq, &irq, 1);
-		free_percpu_irq(irq, &percpu_pmu);
+		free_percpu_irq(irq, &hw_events->percpu_pmu);
 	} else {
 		for (i = 0; i < irqs; ++i) {
 			if (!cpumask_test_and_clear_cpu(i, &cpu_pmu->active_irqs))
 				continue;
 			irq = platform_get_irq(pmu_device, i);
 			if (irq >= 0)
-				free_irq(irq, cpu_pmu);
+				free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, i));
 		}
 	}
 }
@@ -114,19 +105,21 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
 {
 	int i, err, irq, irqs;
 	struct platform_device *pmu_device = cpu_pmu->plat_device;
+	struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
 
 	if (!pmu_device)
 		return -ENODEV;
 
 	irqs = min(pmu_device->num_resources, num_possible_cpus());
 	if (irqs < 1) {
-		printk_once("perf/ARM: No irqs for PMU defined, sampling events not supported\n");
+		pr_warn_once("perf/ARM: No irqs for PMU defined, sampling events not supported\n");
 		return 0;
 	}
 
 	irq = platform_get_irq(pmu_device, 0);
 	if (irq >= 0 && irq_is_percpu(irq)) {
-		err = request_percpu_irq(irq, handler, "arm-pmu", &percpu_pmu);
+		err = request_percpu_irq(irq, handler, "arm-pmu",
+					 &hw_events->percpu_pmu);
 		if (err) {
 			pr_err("unable to request IRQ%d for ARM PMU counters\n",
 				irq);
@@ -153,7 +146,7 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
 
 			err = request_irq(irq, handler,
 					  IRQF_NOBALANCING | IRQF_NO_THREAD, "arm-pmu",
-					  cpu_pmu);
+					  per_cpu_ptr(&hw_events->percpu_pmu, i));
 			if (err) {
 				pr_err("unable to request IRQ%d for ARM PMU counters\n",
 					irq);
@@ -167,18 +160,50 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
 	return 0;
 }
 
-static void cpu_pmu_init(struct arm_pmu *cpu_pmu)
+/*
+ * PMU hardware loses all context when a CPU goes offline.
+ * When a CPU is hotplugged back in, since some hardware registers are
+ * UNKNOWN at reset, the PMU must be explicitly reset to avoid reading
+ * junk values out of them.
+ */
+static int cpu_pmu_notify(struct notifier_block *b, unsigned long action,
+			  void *hcpu)
+{
+	struct arm_pmu *pmu = container_of(b, struct arm_pmu, hotplug_nb);
+
+	if ((action & ~CPU_TASKS_FROZEN) != CPU_STARTING)
+		return NOTIFY_DONE;
+
+	if (pmu->reset)
+		pmu->reset(pmu);
+	else
+		return NOTIFY_DONE;
+
+	return NOTIFY_OK;
+}
+
+static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
 {
+	int err;
 	int cpu;
+	struct pmu_hw_events __percpu *cpu_hw_events;
+
+	cpu_hw_events = alloc_percpu(struct pmu_hw_events);
+	if (!cpu_hw_events)
+		return -ENOMEM;
+
+	cpu_pmu->hotplug_nb.notifier_call = cpu_pmu_notify;
+	err = register_cpu_notifier(&cpu_pmu->hotplug_nb);
+	if (err)
+		goto out_hw_events;
+
 	for_each_possible_cpu(cpu) {
-		struct pmu_hw_events *events = &per_cpu(cpu_hw_events, cpu);
-		events->events = per_cpu(hw_events, cpu);
-		events->used_mask = per_cpu(used_mask, cpu);
+		struct pmu_hw_events *events = per_cpu_ptr(cpu_hw_events, cpu);
 		raw_spin_lock_init(&events->pmu_lock);
-		per_cpu(percpu_pmu, cpu) = cpu_pmu;
+		events->percpu_pmu = cpu_pmu;
 	}
 
-	cpu_pmu->get_hw_events	= cpu_pmu_get_cpu_events;
+	cpu_pmu->hw_events	= cpu_hw_events;
 	cpu_pmu->request_irq	= cpu_pmu_request_irq;
 	cpu_pmu->free_irq	= cpu_pmu_free_irq;
 
@@ -189,31 +214,19 @@ static void cpu_pmu_init(struct arm_pmu *cpu_pmu)
 	/* If no interrupts available, set the corresponding capability flag */
 	if (!platform_get_irq(cpu_pmu->plat_device, 0))
 		cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
-}
-
-/*
- * PMU hardware loses all context when a CPU goes offline.
- * When a CPU is hotplugged back in, since some hardware registers are
- * UNKNOWN at reset, the PMU must be explicitly reset to avoid reading
- * junk values out of them.
- */
-static int cpu_pmu_notify(struct notifier_block *b, unsigned long action,
-			  void *hcpu)
-{
-	if ((action & ~CPU_TASKS_FROZEN) != CPU_STARTING)
-		return NOTIFY_DONE;
 
-	if (cpu_pmu && cpu_pmu->reset)
-		cpu_pmu->reset(cpu_pmu);
-	else
-		return NOTIFY_DONE;
+	return 0;
 
-	return NOTIFY_OK;
+out_hw_events:
+	free_percpu(cpu_hw_events);
+	return err;
 }
 
-static struct notifier_block cpu_pmu_hotplug_notifier = {
-	.notifier_call = cpu_pmu_notify,
-};
+static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu)
+{
+	unregister_cpu_notifier(&cpu_pmu->hotplug_nb);
+	free_percpu(cpu_pmu->hw_events);
+}
 
 /*
  * PMU platform driver and devicetree bindings.
@@ -241,48 +254,34 @@ static struct platform_device_id cpu_pmu_plat_device_ids[] = {
 	{},
 };
 
+static const struct pmu_probe_info pmu_probe_table[] = {
+	ARM_PMU_PROBE(ARM_CPU_PART_ARM1136, armv6_1136_pmu_init),
+	ARM_PMU_PROBE(ARM_CPU_PART_ARM1156, armv6_1156_pmu_init),
+	ARM_PMU_PROBE(ARM_CPU_PART_ARM1176, armv6_1176_pmu_init),
+	ARM_PMU_PROBE(ARM_CPU_PART_ARM11MPCORE, armv6mpcore_pmu_init),
+	ARM_PMU_PROBE(ARM_CPU_PART_CORTEX_A8, armv7_a8_pmu_init),
+	ARM_PMU_PROBE(ARM_CPU_PART_CORTEX_A9, armv7_a9_pmu_init),
+	XSCALE_PMU_PROBE(ARM_CPU_XSCALE_ARCH_V1, xscale1pmu_init),
+	XSCALE_PMU_PROBE(ARM_CPU_XSCALE_ARCH_V2, xscale2pmu_init),
+	{ /* sentinel value */ }
+};
+
 /*
  * CPU PMU identification and probing.
  */
 static int probe_current_pmu(struct arm_pmu *pmu)
 {
 	int cpu = get_cpu();
+	unsigned int cpuid = read_cpuid_id();
 	int ret = -ENODEV;
+	const struct pmu_probe_info *info;
 
 	pr_info("probing PMU on CPU %d\n", cpu);
 
-	switch (read_cpuid_part()) {
-	/* ARM Ltd CPUs. */
-	case ARM_CPU_PART_ARM1136:
-		ret = armv6_1136_pmu_init(pmu);
-		break;
-	case ARM_CPU_PART_ARM1156:
-		ret = armv6_1156_pmu_init(pmu);
-		break;
-	case ARM_CPU_PART_ARM1176:
-		ret = armv6_1176_pmu_init(pmu);
-		break;
-	case ARM_CPU_PART_ARM11MPCORE:
-		ret = armv6mpcore_pmu_init(pmu);
-		break;
-	case ARM_CPU_PART_CORTEX_A8:
-		ret = armv7_a8_pmu_init(pmu);
-		break;
-	case ARM_CPU_PART_CORTEX_A9:
-		ret = armv7_a9_pmu_init(pmu);
-		break;
-
-	default:
-		if (read_cpuid_implementor() == ARM_CPU_IMP_INTEL) {
-			switch (xscale_cpu_arch_version()) {
-			case ARM_CPU_XSCALE_ARCH_V1:
-				ret = xscale1pmu_init(pmu);
-				break;
-			case ARM_CPU_XSCALE_ARCH_V2:
-				ret = xscale2pmu_init(pmu);
-				break;
-			}
-		}
+	for (info = pmu_probe_table; info->init != NULL; info++) {
+		if ((cpuid & info->mask) != info->cpuid)
+			continue;
+		ret = info->init(pmu);
 		break;
 	}
 
@@ -299,13 +298,13 @@ static int cpu_pmu_device_probe(struct platform_device *pdev)
 	int ret = -ENODEV;
 
 	if (cpu_pmu) {
-		pr_info("attempt to register multiple PMU devices!");
+		pr_info("attempt to register multiple PMU devices!\n");
 		return -ENOSPC;
 	}
 
 	pmu = kzalloc(sizeof(struct arm_pmu), GFP_KERNEL);
 	if (!pmu) {
-		pr_info("failed to allocate PMU device!");
+		pr_info("failed to allocate PMU device!\n");
 		return -ENOMEM;
 	}
 
@@ -320,18 +319,24 @@ static int cpu_pmu_device_probe(struct platform_device *pdev)
 	}
 
 	if (ret) {
-		pr_info("failed to probe PMU!");
+		pr_info("failed to probe PMU!\n");
 		goto out_free;
 	}
 
-	cpu_pmu_init(cpu_pmu);
-	ret = armpmu_register(cpu_pmu, PERF_TYPE_RAW);
+	ret = cpu_pmu_init(cpu_pmu);
+	if (ret)
+		goto out_free;
 
-	if (!ret)
-		return 0;
+	ret = armpmu_register(cpu_pmu, -1);
+	if (ret)
+		goto out_destroy;
 
+	return 0;
+
+out_destroy:
+	cpu_pmu_destroy(cpu_pmu);
 out_free:
-	pr_info("failed to register PMU devices!");
+	pr_info("failed to register PMU devices!\n");
 	kfree(pmu);
 	return ret;
 }
@@ -348,16 +353,6 @@ static struct platform_driver cpu_pmu_driver = {
 
 static int __init register_pmu_driver(void)
 {
-	int err;
-
-	err = register_cpu_notifier(&cpu_pmu_hotplug_notifier);
-	if (err)
-		return err;
-
-	err = platform_driver_register(&cpu_pmu_driver);
-	if (err)
-		unregister_cpu_notifier(&cpu_pmu_hotplug_notifier);
-
-	return err;
+	return platform_driver_register(&cpu_pmu_driver);
 }
 device_initcall(register_pmu_driver);
diff --git a/arch/arm/kernel/perf_event_v6.c b/arch/arm/kernel/perf_event_v6.c
index abfeb04f3213..f2ffd5c542ed 100644
--- a/arch/arm/kernel/perf_event_v6.c
+++ b/arch/arm/kernel/perf_event_v6.c
@@ -262,7 +262,7 @@ static void armv6pmu_enable_event(struct perf_event *event)
 	unsigned long val, mask, evt, flags;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	if (ARMV6_CYCLE_COUNTER == idx) {
@@ -300,7 +300,7 @@ armv6pmu_handle_irq(int irq_num,
 	unsigned long pmcr = armv6_pmcr_read();
 	struct perf_sample_data data;
 	struct arm_pmu *cpu_pmu = (struct arm_pmu *)dev;
-	struct pmu_hw_events *cpuc = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
 	struct pt_regs *regs;
 	int idx;
 
@@ -356,7 +356,7 @@ armv6pmu_handle_irq(int irq_num,
 static void armv6pmu_start(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = armv6_pmcr_read();
@@ -368,7 +368,7 @@ static void armv6pmu_start(struct arm_pmu *cpu_pmu)
 static void armv6pmu_stop(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = armv6_pmcr_read();
@@ -409,7 +409,7 @@ static void armv6pmu_disable_event(struct perf_event *event)
 	unsigned long val, mask, evt, flags;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	if (ARMV6_CYCLE_COUNTER == idx) {
@@ -444,7 +444,7 @@ static void armv6mpcore_pmu_disable_event(struct perf_event *event)
 	unsigned long val, mask, flags, evt = 0;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	if (ARMV6_CYCLE_COUNTER == idx) {
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
index 116758b77f93..8993770c47de 100644
--- a/arch/arm/kernel/perf_event_v7.c
+++ b/arch/arm/kernel/perf_event_v7.c
@@ -564,13 +564,11 @@ static inline int armv7_pmnc_counter_has_overflowed(u32 pmnc, int idx)
 	return pmnc & BIT(ARMV7_IDX_TO_COUNTER(idx));
 }
 
-static inline int armv7_pmnc_select_counter(int idx)
+static inline void armv7_pmnc_select_counter(int idx)
 {
 	u32 counter = ARMV7_IDX_TO_COUNTER(idx);
 	asm volatile("mcr p15, 0, %0, c9, c12, 5" : : "r" (counter));
 	isb();
-
-	return idx;
 }
 
 static inline u32 armv7pmu_read_counter(struct perf_event *event)
@@ -580,13 +578,15 @@ static inline u32 armv7pmu_read_counter(struct perf_event *event)
 	int idx = hwc->idx;
 	u32 value = 0;
 
-	if (!armv7_pmnc_counter_valid(cpu_pmu, idx))
+	if (!armv7_pmnc_counter_valid(cpu_pmu, idx)) {
 		pr_err("CPU%u reading wrong counter %d\n",
 			smp_processor_id(), idx);
-	else if (idx == ARMV7_IDX_CYCLE_COUNTER)
+	} else if (idx == ARMV7_IDX_CYCLE_COUNTER) {
 		asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r" (value));
-	else if (armv7_pmnc_select_counter(idx) == idx)
+	} else {
+		armv7_pmnc_select_counter(idx);
 		asm volatile("mrc p15, 0, %0, c9, c13, 2" : "=r" (value));
+	}
 
 	return value;
 }
@@ -597,45 +597,43 @@ static inline void armv7pmu_write_counter(struct perf_event *event, u32 value)
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 
-	if (!armv7_pmnc_counter_valid(cpu_pmu, idx))
+	if (!armv7_pmnc_counter_valid(cpu_pmu, idx)) {
 		pr_err("CPU%u writing wrong counter %d\n",
 			smp_processor_id(), idx);
-	else if (idx == ARMV7_IDX_CYCLE_COUNTER)
+	} else if (idx == ARMV7_IDX_CYCLE_COUNTER) {
 		asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" (value));
-	else if (armv7_pmnc_select_counter(idx) == idx)
+	} else {
+		armv7_pmnc_select_counter(idx);
 		asm volatile("mcr p15, 0, %0, c9, c13, 2" : : "r" (value));
+	}
 }
 
 static inline void armv7_pmnc_write_evtsel(int idx, u32 val)
 {
-	if (armv7_pmnc_select_counter(idx) == idx) {
-		val &= ARMV7_EVTYPE_MASK;
-		asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (val));
-	}
+	armv7_pmnc_select_counter(idx);
+	val &= ARMV7_EVTYPE_MASK;
+	asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (val));
 }
 
-static inline int armv7_pmnc_enable_counter(int idx)
+static inline void armv7_pmnc_enable_counter(int idx)
 {
 	u32 counter = ARMV7_IDX_TO_COUNTER(idx);
 	asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (BIT(counter)));
-	return idx;
 }
 
-static inline int armv7_pmnc_disable_counter(int idx)
+static inline void armv7_pmnc_disable_counter(int idx)
 {
 	u32 counter = ARMV7_IDX_TO_COUNTER(idx);
 	asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (BIT(counter)));
-	return idx;
 }
 
-static inline int armv7_pmnc_enable_intens(int idx)
+static inline void armv7_pmnc_enable_intens(int idx)
 {
 	u32 counter = ARMV7_IDX_TO_COUNTER(idx);
 	asm volatile("mcr p15, 0, %0, c9, c14, 1" : : "r" (BIT(counter)));
-	return idx;
 }
 
-static inline int armv7_pmnc_disable_intens(int idx)
+static inline void armv7_pmnc_disable_intens(int idx)
 {
 	u32 counter = ARMV7_IDX_TO_COUNTER(idx);
 	asm volatile("mcr p15, 0, %0, c9, c14, 2" : : "r" (BIT(counter)));
@@ -643,8 +641,6 @@ static inline int armv7_pmnc_disable_intens(int idx)
 	/* Clear the overflow flag in case an interrupt is pending. */
 	asm volatile("mcr p15, 0, %0, c9, c12, 3" : : "r" (BIT(counter)));
 	isb();
-
-	return idx;
 }
 
 static inline u32 armv7_pmnc_getreset_flags(void)
@@ -667,34 +663,34 @@ static void armv7_pmnc_dump_regs(struct arm_pmu *cpu_pmu)
 	u32 val;
 	unsigned int cnt;
 
-	printk(KERN_INFO "PMNC registers dump:\n");
+	pr_info("PMNC registers dump:\n");
 
 	asm volatile("mrc p15, 0, %0, c9, c12, 0" : "=r" (val));
-	printk(KERN_INFO "PMNC  =0x%08x\n", val);
+	pr_info("PMNC  =0x%08x\n", val);
 
 	asm volatile("mrc p15, 0, %0, c9, c12, 1" : "=r" (val));
-	printk(KERN_INFO "CNTENS=0x%08x\n", val);
+	pr_info("CNTENS=0x%08x\n", val);
 
 	asm volatile("mrc p15, 0, %0, c9, c14, 1" : "=r" (val));
-	printk(KERN_INFO "INTENS=0x%08x\n", val);
+	pr_info("INTENS=0x%08x\n", val);
 
 	asm volatile("mrc p15, 0, %0, c9, c12, 3" : "=r" (val));
-	printk(KERN_INFO "FLAGS =0x%08x\n", val);
+	pr_info("FLAGS =0x%08x\n", val);
 
 	asm volatile("mrc p15, 0, %0, c9, c12, 5" : "=r" (val));
-	printk(KERN_INFO "SELECT=0x%08x\n", val);
+	pr_info("SELECT=0x%08x\n", val);
 
 	asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r" (val));
-	printk(KERN_INFO "CCNT  =0x%08x\n", val);
+	pr_info("CCNT  =0x%08x\n", val);
 
 	for (cnt = ARMV7_IDX_COUNTER0;
 			cnt <= ARMV7_IDX_COUNTER_LAST(cpu_pmu); cnt++) {
 		armv7_pmnc_select_counter(cnt);
 		asm volatile("mrc p15, 0, %0, c9, c13, 2" : "=r" (val));
-		printk(KERN_INFO "CNT[%d] count =0x%08x\n",
+		pr_info("CNT[%d] count =0x%08x\n",
 			ARMV7_IDX_TO_COUNTER(cnt), val);
 		asm volatile("mrc p15, 0, %0, c9, c13, 1" : "=r" (val));
-		printk(KERN_INFO "CNT[%d] evtsel=0x%08x\n",
+		pr_info("CNT[%d] evtsel=0x%08x\n",
 			ARMV7_IDX_TO_COUNTER(cnt), val);
 	}
 }
@@ -705,7 +701,7 @@ static void armv7pmu_enable_event(struct perf_event *event)
 	unsigned long flags;
 	struct hw_perf_event *hwc = &event->hw;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	if (!armv7_pmnc_counter_valid(cpu_pmu, idx)) {
@@ -751,7 +747,7 @@ static void armv7pmu_disable_event(struct perf_event *event)
 	unsigned long flags;
 	struct hw_perf_event *hwc = &event->hw;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	if (!armv7_pmnc_counter_valid(cpu_pmu, idx)) {
@@ -783,7 +779,7 @@ static irqreturn_t armv7pmu_handle_irq(int irq_num, void *dev)
 	u32 pmnc;
 	struct perf_sample_data data;
 	struct arm_pmu *cpu_pmu = (struct arm_pmu *)dev;
-	struct pmu_hw_events *cpuc = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
 	struct pt_regs *regs;
 	int idx;
 
@@ -843,7 +839,7 @@ static irqreturn_t armv7pmu_handle_irq(int irq_num, void *dev)
 static void armv7pmu_start(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	/* Enable all counters */
@@ -854,7 +850,7 @@ static void armv7pmu_start(struct arm_pmu *cpu_pmu)
 static void armv7pmu_stop(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	/* Disable all counters */
@@ -1287,7 +1283,7 @@ static void krait_pmu_disable_event(struct perf_event *event)
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	/* Disable counter and interrupt */
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
@@ -1313,7 +1309,7 @@ static void krait_pmu_enable_event(struct perf_event *event)
 	struct hw_perf_event *hwc = &event->hw;
 	int idx = hwc->idx;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	/*
 	 * Enable counter and interrupt, and set the counter to count
diff --git a/arch/arm/kernel/perf_event_xscale.c b/arch/arm/kernel/perf_event_xscale.c
index 08da0af550b7..8af9f1f82c68 100644
--- a/arch/arm/kernel/perf_event_xscale.c
+++ b/arch/arm/kernel/perf_event_xscale.c
@@ -138,7 +138,7 @@ xscale1pmu_handle_irq(int irq_num, void *dev)
 	unsigned long pmnc;
 	struct perf_sample_data data;
 	struct arm_pmu *cpu_pmu = (struct arm_pmu *)dev;
-	struct pmu_hw_events *cpuc = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
 	struct pt_regs *regs;
 	int idx;
 
@@ -198,7 +198,7 @@ static void xscale1pmu_enable_event(struct perf_event *event)
 	unsigned long val, mask, evt, flags;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	switch (idx) {
@@ -234,7 +234,7 @@ static void xscale1pmu_disable_event(struct perf_event *event)
 	unsigned long val, mask, evt, flags;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	switch (idx) {
@@ -287,7 +287,7 @@ xscale1pmu_get_event_idx(struct pmu_hw_events *cpuc,
 static void xscale1pmu_start(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = xscale1pmu_read_pmnc();
@@ -299,7 +299,7 @@ static void xscale1pmu_start(struct arm_pmu *cpu_pmu)
 static void xscale1pmu_stop(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = xscale1pmu_read_pmnc();
@@ -485,7 +485,7 @@ xscale2pmu_handle_irq(int irq_num, void *dev)
 	unsigned long pmnc, of_flags;
 	struct perf_sample_data data;
 	struct arm_pmu *cpu_pmu = (struct arm_pmu *)dev;
-	struct pmu_hw_events *cpuc = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
 	struct pt_regs *regs;
 	int idx;
 
@@ -539,7 +539,7 @@ static void xscale2pmu_enable_event(struct perf_event *event)
 	unsigned long flags, ien, evtsel;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	ien = xscale2pmu_read_int_enable();
@@ -585,7 +585,7 @@ static void xscale2pmu_disable_event(struct perf_event *event)
 	unsigned long flags, ien, evtsel, of_flags;
 	struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
 	struct hw_perf_event *hwc = &event->hw;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 	int idx = hwc->idx;
 
 	ien = xscale2pmu_read_int_enable();
@@ -651,7 +651,7 @@ out:
 static void xscale2pmu_start(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = xscale2pmu_read_pmnc() & ~XSCALE_PMU_CNT64;
@@ -663,7 +663,7 @@ static void xscale2pmu_start(struct arm_pmu *cpu_pmu)
 static void xscale2pmu_stop(struct arm_pmu *cpu_pmu)
 {
 	unsigned long flags, val;
-	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+	struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
 	val = xscale2pmu_read_pmnc();
diff --git a/drivers/bus/arm-cci.c b/drivers/bus/arm-cci.c
index 7af78df241f2..860da40b78ef 100644
--- a/drivers/bus/arm-cci.c
+++ b/drivers/bus/arm-cci.c
@@ -16,17 +16,17 @@
 
 #include <linux/arm-cci.h>
 #include <linux/io.h>
+#include <linux/interrupt.h>
 #include <linux/module.h>
 #include <linux/of_address.h>
 #include <linux/of_irq.h>
 #include <linux/of_platform.h>
+#include <linux/perf_event.h>
 #include <linux/platform_device.h>
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 
 #include <asm/cacheflush.h>
-#include <asm/irq_regs.h>
-#include <asm/pmu.h>
 #include <asm/smp_plat.h>
 
 #define DRIVER_NAME		"CCI-400"
@@ -98,6 +98,8 @@ static unsigned long cci_ctrl_phys;
 
 #define CCI_PMU_CNTR_BASE(idx)	((idx) * SZ_4K)
 
+#define CCI_PMU_CNTR_MASK	((1ULL << 32) -1)
+
 /*
  * Instead of an event id to monitor CCI cycles, a dedicated counter is
  * provided. Use 0xff to represent CCI cycles and hope that no future revisions
@@ -170,18 +172,29 @@ static char *const pmu_names[] = {
 	[CCI_REV_R1] = "CCI_400_r1",
 };
 
-struct cci_pmu_drv_data {
+struct cci_pmu_hw_events {
+	struct perf_event *events[CCI_PMU_MAX_HW_EVENTS];
+	unsigned long used_mask[BITS_TO_LONGS(CCI_PMU_MAX_HW_EVENTS)];
+	raw_spinlock_t pmu_lock;
+};
+
+struct cci_pmu {
 	void __iomem *base;
-	struct arm_pmu *cci_pmu;
+	struct pmu pmu;
 	int nr_irqs;
 	int irqs[CCI_PMU_MAX_HW_EVENTS];
 	unsigned long active_irqs;
-	struct perf_event *events[CCI_PMU_MAX_HW_EVENTS];
-	unsigned long used_mask[BITS_TO_LONGS(CCI_PMU_MAX_HW_EVENTS)];
 	struct pmu_port_event_ranges *port_ranges;
-	struct pmu_hw_events hw_events;
+	struct cci_pmu_hw_events hw_events;
+	struct platform_device *plat_device;
+	int num_events;
+	atomic_t active_events;
+	struct mutex reserve_mutex;
+	cpumask_t cpus;
 };
-static struct cci_pmu_drv_data *pmu;
+static struct cci_pmu *pmu;
+
+#define to_cci_pmu(c)	(container_of(c, struct cci_pmu, pmu))
 
 static bool is_duplicate_irq(int irq, int *irqs, int nr_irqs)
 {
@@ -252,7 +265,7 @@ static int pmu_validate_hw_event(u8 hw_event)
 	return -ENOENT;
 }
 
-static int pmu_is_valid_counter(struct arm_pmu *cci_pmu, int idx)
+static int pmu_is_valid_counter(struct cci_pmu *cci_pmu, int idx)
 {
 	return CCI_PMU_CYCLE_CNTR_IDX <= idx &&
 		idx <= CCI_PMU_CNTR_LAST(cci_pmu);
@@ -293,14 +306,9 @@ static u32 pmu_get_max_counters(void)
 	return n_cnts + 1;
 }
 
-static struct pmu_hw_events *pmu_get_hw_events(void)
-{
-	return &pmu->hw_events;
-}
-
-static int pmu_get_event_idx(struct pmu_hw_events *hw, struct perf_event *event)
+static int pmu_get_event_idx(struct cci_pmu_hw_events *hw, struct perf_event *event)
 {
-	struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
 	struct hw_perf_event *hw_event = &event->hw;
 	unsigned long cci_event = hw_event->config_base & CCI_PMU_EVENT_MASK;
 	int idx;
@@ -336,7 +344,7 @@ static int pmu_map_event(struct perf_event *event)
 	return mapping;
 }
 
-static int pmu_request_irq(struct arm_pmu *cci_pmu, irq_handler_t handler)
+static int pmu_request_irq(struct cci_pmu *cci_pmu, irq_handler_t handler)
 {
 	int i;
 	struct platform_device *pmu_device = cci_pmu->plat_device;
@@ -371,17 +379,91 @@ static int pmu_request_irq(struct arm_pmu *cci_pmu, irq_handler_t handler)
 	return 0;
 }
 
+static void pmu_free_irq(struct cci_pmu *cci_pmu)
+{
+	int i;
+
+	for (i = 0; i < pmu->nr_irqs; i++) {
+		if (!test_and_clear_bit(i, &pmu->active_irqs))
+			continue;
+
+		free_irq(pmu->irqs[i], cci_pmu);
+	}
+}
+
+static u32 pmu_read_counter(struct perf_event *event)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct hw_perf_event *hw_counter = &event->hw;
+	int idx = hw_counter->idx;
+	u32 value;
+
+	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) {
+		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
+		return 0;
+	}
+	value = pmu_read_register(idx, CCI_PMU_CNTR);
+
+	return value;
+}
+
+static void pmu_write_counter(struct perf_event *event, u32 value)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct hw_perf_event *hw_counter = &event->hw;
+	int idx = hw_counter->idx;
+
+	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx)))
+		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
+	else
+		pmu_write_register(value, idx, CCI_PMU_CNTR);
+}
+
+static u64 pmu_event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev_raw_count, new_raw_count;
+
+	do {
+		prev_raw_count = local64_read(&hwc->prev_count);
+		new_raw_count = pmu_read_counter(event);
+	} while (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+		 new_raw_count) != prev_raw_count);
+
+	delta = (new_raw_count - prev_raw_count) & CCI_PMU_CNTR_MASK;
+
+	local64_add(delta, &event->count);
+
+	return new_raw_count;
+}
+
+static void pmu_read(struct perf_event *event)
+{
+	pmu_event_update(event);
+}
+
+void pmu_event_set_period(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	/*
+	 * The CCI PMU counters have a period of 2^32. To account for the
+	 * possiblity of extreme interrupt latency we program for a period of
+	 * half that. Hopefully we can handle the interrupt before another 2^31
+	 * events occur and the counter overtakes its previous value.
+	 */
+	u64 val = 1ULL << 31;
+	local64_set(&hwc->prev_count, val);
+	pmu_write_counter(event, val);
+}
+
 static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
 {
 	unsigned long flags;
-	struct arm_pmu *cci_pmu = (struct arm_pmu *)dev;
-	struct pmu_hw_events *events = cci_pmu->get_hw_events();
-	struct perf_sample_data data;
-	struct pt_regs *regs;
+	struct cci_pmu *cci_pmu = dev;
+	struct cci_pmu_hw_events *events = &pmu->hw_events;
 	int idx, handled = IRQ_NONE;
 
 	raw_spin_lock_irqsave(&events->pmu_lock, flags);
-	regs = get_irq_regs();
 	/*
 	 * Iterate over counters and update the corresponding perf events.
 	 * This should work regardless of whether we have per-counter overflow
@@ -403,154 +485,407 @@ static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
 
 		pmu_write_register(CCI_PMU_OVRFLW_FLAG, idx, CCI_PMU_OVRFLW);
 
+		pmu_event_update(event);
+		pmu_event_set_period(event);
 		handled = IRQ_HANDLED;
-
-		armpmu_event_update(event);
-		perf_sample_data_init(&data, 0, hw_counter->last_period);
-		if (!armpmu_event_set_period(event))
-			continue;
-
-		if (perf_event_overflow(event, &data, regs))
-			cci_pmu->disable(event);
 	}
 	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
 
 	return IRQ_RETVAL(handled);
 }
 
-static void pmu_free_irq(struct arm_pmu *cci_pmu)
+static int cci_pmu_get_hw(struct cci_pmu *cci_pmu)
 {
-	int i;
+	int ret = pmu_request_irq(cci_pmu, pmu_handle_irq);
+	if (ret) {
+		pmu_free_irq(cci_pmu);
+		return ret;
+	}
+	return 0;
+}
 
-	for (i = 0; i < pmu->nr_irqs; i++) {
-		if (!test_and_clear_bit(i, &pmu->active_irqs))
-			continue;
+static void cci_pmu_put_hw(struct cci_pmu *cci_pmu)
+{
+	pmu_free_irq(cci_pmu);
+}
 
-		free_irq(pmu->irqs[i], cci_pmu);
+static void hw_perf_event_destroy(struct perf_event *event)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	atomic_t *active_events = &cci_pmu->active_events;
+	struct mutex *reserve_mutex = &cci_pmu->reserve_mutex;
+
+	if (atomic_dec_and_mutex_lock(active_events, reserve_mutex)) {
+		cci_pmu_put_hw(cci_pmu);
+		mutex_unlock(reserve_mutex);
 	}
 }
 
-static void pmu_enable_event(struct perf_event *event)
+static void cci_pmu_enable(struct pmu *pmu)
 {
+	struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
+	struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+	int enabled = bitmap_weight(hw_events->used_mask, cci_pmu->num_events);
 	unsigned long flags;
-	struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
-	struct pmu_hw_events *events = cci_pmu->get_hw_events();
-	struct hw_perf_event *hw_counter = &event->hw;
-	int idx = hw_counter->idx;
+	u32 val;
+
+	if (!enabled)
+		return;
+
+	raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
+
+	/* Enable all the PMU counters. */
+	val = readl_relaxed(cci_ctrl_base + CCI_PMCR) | CCI_PMCR_CEN;
+	writel(val, cci_ctrl_base + CCI_PMCR);
+	raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
+
+}
+
+static void cci_pmu_disable(struct pmu *pmu)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
+	struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+	unsigned long flags;
+	u32 val;
+
+	raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
+
+	/* Disable all the PMU counters. */
+	val = readl_relaxed(cci_ctrl_base + CCI_PMCR) & ~CCI_PMCR_CEN;
+	writel(val, cci_ctrl_base + CCI_PMCR);
+	raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
+}
+
+static void cci_pmu_start(struct perf_event *event, int pmu_flags)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+	unsigned long flags;
+
+	/*
+	 * To handle interrupt latency, we always reprogram the period
+	 * regardlesss of PERF_EF_RELOAD.
+	 */
+	if (pmu_flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+
+	hwc->state = 0;
 
 	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) {
 		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
 		return;
 	}
 
-	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
 
 	/* Configure the event to count, unless you are counting cycles */
 	if (idx != CCI_PMU_CYCLE_CNTR_IDX)
-		pmu_set_event(idx, hw_counter->config_base);
+		pmu_set_event(idx, hwc->config_base);
 
+	pmu_event_set_period(event);
 	pmu_enable_counter(idx);
 
-	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+	raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
 }
 
-static void pmu_disable_event(struct perf_event *event)
+static void cci_pmu_stop(struct perf_event *event, int pmu_flags)
 {
-	struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
-	struct hw_perf_event *hw_counter = &event->hw;
-	int idx = hw_counter->idx;
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	if (hwc->state & PERF_HES_STOPPED)
+		return;
 
 	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) {
 		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
 		return;
 	}
 
+	/*
+	 * We always reprogram the counter, so ignore PERF_EF_UPDATE. See
+	 * cci_pmu_start()
+	 */
 	pmu_disable_counter(idx);
+	pmu_event_update(event);
+	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
 }
 
-static void pmu_start(struct arm_pmu *cci_pmu)
+static int cci_pmu_add(struct perf_event *event, int flags)
 {
-	u32 val;
-	unsigned long flags;
-	struct pmu_hw_events *events = cci_pmu->get_hw_events();
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+	int err = 0;
 
-	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	perf_pmu_disable(event->pmu);
 
-	/* Enable all the PMU counters. */
-	val = readl_relaxed(cci_ctrl_base + CCI_PMCR) | CCI_PMCR_CEN;
-	writel(val, cci_ctrl_base + CCI_PMCR);
+	/* If we don't have a space for the counter then finish early. */
+	idx = pmu_get_event_idx(hw_events, event);
+	if (idx < 0) {
+		err = idx;
+		goto out;
+	}
 
-	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+	event->hw.idx = idx;
+	hw_events->events[idx] = event;
+
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	if (flags & PERF_EF_START)
+		cci_pmu_start(event, PERF_EF_RELOAD);
+
+	/* Propagate our changes to the userspace mapping. */
+	perf_event_update_userpage(event);
+
+out:
+	perf_pmu_enable(event->pmu);
+	return err;
 }
 
-static void pmu_stop(struct arm_pmu *cci_pmu)
+static void cci_pmu_del(struct perf_event *event, int flags)
 {
-	u32 val;
-	unsigned long flags;
-	struct pmu_hw_events *events = cci_pmu->get_hw_events();
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
 
-	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	cci_pmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[idx] = NULL;
+	clear_bit(idx, hw_events->used_mask);
 
-	/* Disable all the PMU counters. */
-	val = readl_relaxed(cci_ctrl_base + CCI_PMCR) & ~CCI_PMCR_CEN;
-	writel(val, cci_ctrl_base + CCI_PMCR);
+	perf_event_update_userpage(event);
+}
 
-	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+static int
+validate_event(struct cci_pmu_hw_events *hw_events,
+	       struct perf_event *event)
+{
+	if (is_software_event(event))
+		return 1;
+
+	if (event->state < PERF_EVENT_STATE_OFF)
+		return 1;
+
+	if (event->state == PERF_EVENT_STATE_OFF && !event->attr.enable_on_exec)
+		return 1;
+
+	return pmu_get_event_idx(hw_events, event) >= 0;
 }
 
-static u32 pmu_read_counter(struct perf_event *event)
+static int
+validate_group(struct perf_event *event)
 {
-	struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
-	struct hw_perf_event *hw_counter = &event->hw;
-	int idx = hw_counter->idx;
-	u32 value;
+	struct perf_event *sibling, *leader = event->group_leader;
+	struct cci_pmu_hw_events fake_pmu = {
+		/*
+		 * Initialise the fake PMU. We only need to populate the
+		 * used_mask for the purposes of validation.
+		 */
+		.used_mask = CPU_BITS_NONE,
+	};
 
-	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) {
-		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
-		return 0;
+	if (!validate_event(&fake_pmu, leader))
+		return -EINVAL;
+
+	list_for_each_entry(sibling, &leader->sibling_list, group_entry) {
+		if (!validate_event(&fake_pmu, sibling))
+			return -EINVAL;
 	}
-	value = pmu_read_register(idx, CCI_PMU_CNTR);
 
-	return value;
+	if (!validate_event(&fake_pmu, event))
+		return -EINVAL;
+
+	return 0;
 }
 
-static void pmu_write_counter(struct perf_event *event, u32 value)
+static int
+__hw_perf_event_init(struct perf_event *event)
 {
-	struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
-	struct hw_perf_event *hw_counter = &event->hw;
-	int idx = hw_counter->idx;
+	struct hw_perf_event *hwc = &event->hw;
+	int mapping;
 
-	if (unlikely(!pmu_is_valid_counter(cci_pmu, idx)))
-		dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
-	else
-		pmu_write_register(value, idx, CCI_PMU_CNTR);
+	mapping = pmu_map_event(event);
+
+	if (mapping < 0) {
+		pr_debug("event %x:%llx not supported\n", event->attr.type,
+			 event->attr.config);
+		return mapping;
+	}
+
+	/*
+	 * We don't assign an index until we actually place the event onto
+	 * hardware. Use -1 to signify that we haven't decided where to put it
+	 * yet.
+	 */
+	hwc->idx		= -1;
+	hwc->config_base	= 0;
+	hwc->config		= 0;
+	hwc->event_base		= 0;
+
+	/*
+	 * Store the event encoding into the config_base field.
+	 */
+	hwc->config_base	    |= (unsigned long)mapping;
+
+	/*
+	 * Limit the sample_period to half of the counter width. That way, the
+	 * new counter value is far less likely to overtake the previous one
+	 * unless you have some serious IRQ latency issues.
+	 */
+	hwc->sample_period  = CCI_PMU_CNTR_MASK >> 1;
+	hwc->last_period    = hwc->sample_period;
+	local64_set(&hwc->period_left, hwc->sample_period);
+
+	if (event->group_leader != event) {
+		if (validate_group(event) != 0)
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int cci_pmu_event_init(struct perf_event *event)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	atomic_t *active_events = &cci_pmu->active_events;
+	int err = 0;
+	int cpu;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/* Shared by all CPUs, no meaningful state to sample */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EOPNOTSUPP;
+
+	/* We have no filtering of any kind */
+	if (event->attr.exclude_user	||
+	    event->attr.exclude_kernel	||
+	    event->attr.exclude_hv	||
+	    event->attr.exclude_idle	||
+	    event->attr.exclude_host	||
+	    event->attr.exclude_guest)
+		return -EINVAL;
+
+	/*
+	 * Following the example set by other "uncore" PMUs, we accept any CPU
+	 * and rewrite its affinity dynamically rather than having perf core
+	 * handle cpu == -1 and pid == -1 for this case.
+	 *
+	 * The perf core will pin online CPUs for the duration of this call and
+	 * the event being installed into its context, so the PMU's CPU can't
+	 * change under our feet.
+	 */
+	cpu = cpumask_first(&cci_pmu->cpus);
+	if (event->cpu < 0 || cpu < 0)
+		return -EINVAL;
+	event->cpu = cpu;
+
+	event->destroy = hw_perf_event_destroy;
+	if (!atomic_inc_not_zero(active_events)) {
+		mutex_lock(&cci_pmu->reserve_mutex);
+		if (atomic_read(active_events) == 0)
+			err = cci_pmu_get_hw(cci_pmu);
+		if (!err)
+			atomic_inc(active_events);
+		mutex_unlock(&cci_pmu->reserve_mutex);
+	}
+	if (err)
+		return err;
+
+	err = __hw_perf_event_init(event);
+	if (err)
+		hw_perf_event_destroy(event);
+
+	return err;
 }
 
-static int cci_pmu_init(struct arm_pmu *cci_pmu, struct platform_device *pdev)
+static ssize_t pmu_attr_cpumask_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
 {
-	*cci_pmu = (struct arm_pmu){
-		.name		  = pmu_names[probe_cci_revision()],
-		.max_period       = (1LLU << 32) - 1,
-		.get_hw_events    = pmu_get_hw_events,
-		.get_event_idx    = pmu_get_event_idx,
-		.map_event        = pmu_map_event,
-		.request_irq      = pmu_request_irq,
-		.handle_irq       = pmu_handle_irq,
-		.free_irq         = pmu_free_irq,
-		.enable           = pmu_enable_event,
-		.disable          = pmu_disable_event,
-		.start            = pmu_start,
-		.stop             = pmu_stop,
-		.read_counter     = pmu_read_counter,
-		.write_counter    = pmu_write_counter,
+	int n = cpulist_scnprintf(buf, PAGE_SIZE - 2, &pmu->cpus);
+
+	buf[n++] = '\n';
+	buf[n] = '\0';
+	return n;
+}
+
+static DEVICE_ATTR(cpumask, S_IRUGO, pmu_attr_cpumask_show, NULL);
+
+static struct attribute *pmu_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group pmu_attr_group = {
+	.attrs = pmu_attrs,
+};
+
+static const struct attribute_group *pmu_attr_groups[] = {
+	&pmu_attr_group,
+	NULL
+};
+
+static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
+{
+	char *name = pmu_names[probe_cci_revision()];
+	cci_pmu->pmu = (struct pmu) {
+		.name		= pmu_names[probe_cci_revision()],
+		.task_ctx_nr	= perf_invalid_context,
+		.pmu_enable	= cci_pmu_enable,
+		.pmu_disable	= cci_pmu_disable,
+		.event_init	= cci_pmu_event_init,
+		.add		= cci_pmu_add,
+		.del		= cci_pmu_del,
+		.start		= cci_pmu_start,
+		.stop		= cci_pmu_stop,
+		.read		= pmu_read,
+		.attr_groups	= pmu_attr_groups,
 	};
 
 	cci_pmu->plat_device = pdev;
 	cci_pmu->num_events = pmu_get_max_counters();
 
-	return armpmu_register(cci_pmu, -1);
+	return perf_pmu_register(&cci_pmu->pmu, name, -1);
 }
 
+static int cci_pmu_cpu_notifier(struct notifier_block *self,
+				unsigned long action, void *hcpu)
+{
+	unsigned int cpu = (long)hcpu;
+	unsigned int target;
+
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_DOWN_PREPARE:
+		if (!cpumask_test_and_clear_cpu(cpu, &pmu->cpus))
+			break;
+		target = cpumask_any_but(cpu_online_mask, cpu);
+		if (target < 0) // UP, last CPU
+			break;
+		/*
+		 * TODO: migrate context once core races on event->ctx have
+		 * been fixed.
+		 */
+		cpumask_set_cpu(target, &pmu->cpus);
+	default:
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block cci_pmu_cpu_nb = {
+	.notifier_call	= cci_pmu_cpu_notifier,
+	/*
+	 * to migrate uncore events, our notifier should be executed
+	 * before perf core's notifier.
+	 */
+	.priority	= CPU_PRI_PERF + 1,
+};
+
 static const struct of_device_id arm_cci_pmu_matches[] = {
 	{
 		.compatible = "arm,cci-400-pmu",
@@ -604,15 +939,16 @@ static int cci_pmu_probe(struct platform_device *pdev)
 		return -EINVAL;
 	}
 
-	pmu->cci_pmu = devm_kzalloc(&pdev->dev, sizeof(*(pmu->cci_pmu)), GFP_KERNEL);
-	if (!pmu->cci_pmu)
-		return -ENOMEM;
-
-	pmu->hw_events.events = pmu->events;
-	pmu->hw_events.used_mask = pmu->used_mask;
 	raw_spin_lock_init(&pmu->hw_events.pmu_lock);
+	mutex_init(&pmu->reserve_mutex);
+	atomic_set(&pmu->active_events, 0);
+	cpumask_set_cpu(smp_processor_id(), &pmu->cpus);
+
+	ret = register_cpu_notifier(&cci_pmu_cpu_nb);
+	if (ret)
+		return ret;
 
-	ret = cci_pmu_init(pmu->cci_pmu, pdev);
+	ret = cci_pmu_init(pmu, pdev);
 	if (ret)
 		return ret;