[PATCH v2 25/31] arm64: Performance counters support

[Catalin Marinas <catalin.marinas@arm.com>]

From: Will Deacon <will.deacon@arm.com>

This patch adds support for the AArch64 performance counters.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
---
 arch/arm64/include/asm/perf_event.h |   22 +
 arch/arm64/include/asm/pmu.h        |   82 +++
 arch/arm64/kernel/perf_event.c      | 1368 +++++++++++++++++++++++++++++++++++
 tools/perf/perf.h                   |    6 +
 4 files changed, 1478 insertions(+), 0 deletions(-)
 create mode 100644 arch/arm64/include/asm/perf_event.h
 create mode 100644 arch/arm64/include/asm/pmu.h
 create mode 100644 arch/arm64/kernel/perf_event.c

diff --git a/arch/arm64/include/asm/perf_event.h b/arch/arm64/include/asm/perf_event.h
new file mode 100644
index 0000000..a6fffd5
--- /dev/null
+++ b/arch/arm64/include/asm/perf_event.h
@@ -0,0 +1,22 @@
+/*
+ * Copyright (C) 2012 ARM Ltd.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef __ASM_PERF_EVENT_H
+#define __ASM_PERF_EVENT_H
+
+/* It's quiet around here... */
+
+#endif
diff --git a/arch/arm64/include/asm/pmu.h b/arch/arm64/include/asm/pmu.h
new file mode 100644
index 0000000..e6f0878
--- /dev/null
+++ b/arch/arm64/include/asm/pmu.h
@@ -0,0 +1,82 @@
+/*
+ * Based on arch/arm/include/asm/pmu.h
+ *
+ * Copyright (C) 2009 picoChip Designs Ltd, Jamie Iles
+ * Copyright (C) 2012 ARM Ltd.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+#ifndef __ASM_PMU_H
+#define __ASM_PMU_H
+
+#ifdef CONFIG_HW_PERF_EVENTS
+
+/* The events for a given PMU register set. */
+struct pmu_hw_events {
+	/*
+	 * The events that are active on the PMU for the given index.
+	 */
+	struct perf_event	**events;
+
+	/*
+	 * 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;
+
+	/*
+	 * Hardware lock to serialize accesses to PMU registers. Needed for the
+	 * read/modify/write sequences.
+	 */
+	raw_spinlock_t		pmu_lock;
+};
+
+struct arm_pmu {
+	struct pmu		pmu;
+	cpumask_t		active_irqs;
+	const char		*name;
+	irqreturn_t		(*handle_irq)(int irq_num, void *dev);
+	void			(*enable)(struct hw_perf_event *evt, int idx);
+	void			(*disable)(struct hw_perf_event *evt, int idx);
+	int			(*get_event_idx)(struct pmu_hw_events *hw_events,
+						 struct hw_perf_event *hwc);
+	int			(*set_event_filter)(struct hw_perf_event *evt,
+						    struct perf_event_attr *attr);
+	u32			(*read_counter)(int idx);
+	void			(*write_counter)(int idx, u32 val);
+	void			(*start)(void);
+	void			(*stop)(void);
+	void			(*reset)(void *);
+	int			(*map_event)(struct perf_event *event);
+	int			num_events;
+	atomic_t		active_events;
+	struct mutex		reserve_mutex;
+	u64			max_period;
+	struct platform_device	*plat_device;
+	struct pmu_hw_events	*(*get_hw_events)(void);
+};
+
+#define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu))
+
+int __init armpmu_register(struct arm_pmu *armpmu, char *name, int type);
+
+u64 armpmu_event_update(struct perf_event *event,
+			struct hw_perf_event *hwc,
+			int idx);
+
+int armpmu_event_set_period(struct perf_event *event,
+			    struct hw_perf_event *hwc,
+			    int idx);
+
+#endif /* CONFIG_HW_PERF_EVENTS */
+#endif /* __ASM_PMU_H */
diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c
new file mode 100644
index 0000000..ecbf2d8
--- /dev/null
+++ b/arch/arm64/kernel/perf_event.c
@@ -0,0 +1,1368 @@
+/*
+ * PMU support
+ *
+ * Copyright (C) 2012 ARM Limited
+ * Author: Will Deacon <will.deacon@arm.com>
+ *
+ * This code is based heavily on the ARMv7 perf event code.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+#define pr_fmt(fmt) "hw perfevents: " fmt
+
+#include <linux/bitmap.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/export.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/uaccess.h>
+
+#include <asm/cputype.h>
+#include <asm/irq.h>
+#include <asm/irq_regs.h>
+#include <asm/pmu.h>
+#include <asm/stacktrace.h>
+
+/*
+ * ARMv8 supports a maximum of 32 events.
+ * The cycle counter is included in this total.
+ */
+#define ARMPMU_MAX_HWEVENTS		32
+
+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);
+
+#define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu))
+
+/* Set at runtime when we know what CPU type we are. */
+static struct arm_pmu *cpu_pmu;
+
+int
+armpmu_get_max_events(void)
+{
+	int max_events = 0;
+
+	if (cpu_pmu != NULL)
+		max_events = cpu_pmu->num_events;
+
+	return max_events;
+}
+EXPORT_SYMBOL_GPL(armpmu_get_max_events);
+
+int perf_num_counters(void)
+{
+	return armpmu_get_max_events();
+}
+EXPORT_SYMBOL_GPL(perf_num_counters);
+
+#define HW_OP_UNSUPPORTED		0xFFFF
+
+#define C(_x) \
+	PERF_COUNT_HW_CACHE_##_x
+
+#define CACHE_OP_UNSUPPORTED		0xFFFF
+
+static int
+armpmu_map_cache_event(const unsigned (*cache_map)
+				      [PERF_COUNT_HW_CACHE_MAX]
+				      [PERF_COUNT_HW_CACHE_OP_MAX]
+				      [PERF_COUNT_HW_CACHE_RESULT_MAX],
+		       u64 config)
+{
+	unsigned int cache_type, cache_op, cache_result, ret;
+
+	cache_type = (config >>  0) & 0xff;
+	if (cache_type >= PERF_COUNT_HW_CACHE_MAX)
+		return -EINVAL;
+
+	cache_op = (config >>  8) & 0xff;
+	if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX)
+		return -EINVAL;
+
+	cache_result = (config >> 16) & 0xff;
+	if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
+		return -EINVAL;
+
+	ret = (int)(*cache_map)[cache_type][cache_op][cache_result];
+
+	if (ret == CACHE_OP_UNSUPPORTED)
+		return -ENOENT;
+
+	return ret;
+}
+
+static int
+armpmu_map_event(const unsigned (*event_map)[PERF_COUNT_HW_MAX], u64 config)
+{
+	int mapping = (*event_map)[config];
+	return mapping == HW_OP_UNSUPPORTED ? -ENOENT : mapping;
+}
+
+static int
+armpmu_map_raw_event(u32 raw_event_mask, u64 config)
+{
+	return (int)(config & raw_event_mask);
+}
+
+static int map_cpu_event(struct perf_event *event,
+			 const unsigned (*event_map)[PERF_COUNT_HW_MAX],
+			 const unsigned (*cache_map)
+					[PERF_COUNT_HW_CACHE_MAX]
+					[PERF_COUNT_HW_CACHE_OP_MAX]
+					[PERF_COUNT_HW_CACHE_RESULT_MAX],
+			 u32 raw_event_mask)
+{
+	u64 config = event->attr.config;
+
+	switch (event->attr.type) {
+	case PERF_TYPE_HARDWARE:
+		return armpmu_map_event(event_map, config);
+	case PERF_TYPE_HW_CACHE:
+		return armpmu_map_cache_event(cache_map, config);
+	case PERF_TYPE_RAW:
+		return armpmu_map_raw_event(raw_event_mask, config);
+	}
+
+	return -ENOENT;
+}
+
+int
+armpmu_event_set_period(struct perf_event *event,
+			struct hw_perf_event *hwc,
+			int idx)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	s64 left = local64_read(&hwc->period_left);
+	s64 period = hwc->sample_period;
+	int ret = 0;
+
+	if (unlikely(left <= -period)) {
+		left = period;
+		local64_set(&hwc->period_left, left);
+		hwc->last_period = period;
+		ret = 1;
+	}
+
+	if (unlikely(left <= 0)) {
+		left += period;
+		local64_set(&hwc->period_left, left);
+		hwc->last_period = period;
+		ret = 1;
+	}
+
+	if (left > (s64)armpmu->max_period)
+		left = armpmu->max_period;
+
+	local64_set(&hwc->prev_count, (u64)-left);
+
+	armpmu->write_counter(idx, (u64)(-left) & 0xffffffff);
+
+	perf_event_update_userpage(event);
+
+	return ret;
+}
+
+u64
+armpmu_event_update(struct perf_event *event,
+		    struct hw_perf_event *hwc,
+		    int idx)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	u64 delta, prev_raw_count, new_raw_count;
+
+again:
+	prev_raw_count = local64_read(&hwc->prev_count);
+	new_raw_count = armpmu->read_counter(idx);
+
+	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+			     new_raw_count) != prev_raw_count)
+		goto again;
+
+	delta = (new_raw_count - prev_raw_count) & armpmu->max_period;
+
+	local64_add(delta, &event->count);
+	local64_sub(delta, &hwc->period_left);
+
+	return new_raw_count;
+}
+
+static void
+armpmu_read(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	/* Don't read disabled counters! */
+	if (hwc->idx < 0)
+		return;
+
+	armpmu_event_update(event, hwc, hwc->idx);
+}
+
+static void
+armpmu_stop(struct perf_event *event, int flags)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	/*
+	 * ARM pmu always has to update the counter, so ignore
+	 * PERF_EF_UPDATE, see comments in armpmu_start().
+	 */
+	if (!(hwc->state & PERF_HES_STOPPED)) {
+		armpmu->disable(hwc, hwc->idx);
+		barrier(); /* why? */
+		armpmu_event_update(event, hwc, hwc->idx);
+		hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	}
+}
+
+static void
+armpmu_start(struct perf_event *event, int flags)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	/*
+	 * ARM pmu always has to reprogram the period, so ignore
+	 * PERF_EF_RELOAD, see the comment below.
+	 */
+	if (flags & PERF_EF_RELOAD)
+		WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+
+	hwc->state = 0;
+	/*
+	 * Set the period again. Some counters can't be stopped, so when we
+	 * were stopped we simply disabled the IRQ source and the counter
+	 * may have been left counting. If we don't do this step then we may
+	 * get an interrupt too soon or *way* too late if the overflow has
+	 * happened since disabling.
+	 */
+	armpmu_event_set_period(event, hwc, hwc->idx);
+	armpmu->enable(hwc, hwc->idx);
+}
+
+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 hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	WARN_ON(idx < 0);
+
+	armpmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[idx] = NULL;
+	clear_bit(idx, hw_events->used_mask);
+
+	perf_event_update_userpage(event);
+}
+
+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 hw_perf_event *hwc = &event->hw;
+	int idx;
+	int err = 0;
+
+	perf_pmu_disable(event->pmu);
+
+	/* If we don't have a space for the counter then finish early. */
+	idx = armpmu->get_event_idx(hw_events, hwc);
+	if (idx < 0) {
+		err = idx;
+		goto out;
+	}
+
+	/*
+	 * If there is an event in the counter we are going to use then make
+	 * sure it is disabled.
+	 */
+	event->hw.idx = idx;
+	armpmu->disable(hwc, idx);
+	hw_events->events[idx] = event;
+
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	if (flags & PERF_EF_START)
+		armpmu_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 int
+validate_event(struct pmu_hw_events *hw_events,
+	       struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event fake_event = event->hw;
+	struct pmu *leader_pmu = event->group_leader->pmu;
+
+	if (event->pmu != leader_pmu || event->state <= PERF_EVENT_STATE_OFF)
+		return 1;
+
+	return armpmu->get_event_idx(hw_events, &fake_event) >= 0;
+}
+
+static int
+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;
+
+	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;
+	}
+
+	if (!validate_event(&fake_pmu, event))
+		return -EINVAL;
+
+	return 0;
+}
+
+static void
+armpmu_release_hardware(struct arm_pmu *armpmu)
+{
+	int i, irq, irqs;
+	struct platform_device *pmu_device = armpmu->plat_device;
+
+	irqs = min(pmu_device->num_resources, num_possible_cpus());
+
+	for (i = 0; i < irqs; ++i) {
+		if (!cpumask_test_and_clear_cpu(i, &armpmu->active_irqs))
+			continue;
+		irq = platform_get_irq(pmu_device, i);
+		if (irq >= 0)
+			free_irq(irq, armpmu);
+	}
+}
+
+static int
+armpmu_reserve_hardware(struct arm_pmu *armpmu)
+{
+	int i, err, irq, irqs;
+	struct platform_device *pmu_device = armpmu->plat_device;
+
+	if (!pmu_device) {
+		pr_err("no PMU device registered\n");
+		return -ENODEV;
+	}
+
+	irqs = min(pmu_device->num_resources, num_possible_cpus());
+	if (irqs < 1) {
+		pr_err("no irqs for PMUs defined\n");
+		return -ENODEV;
+	}
+
+	for (i = 0; i < irqs; ++i) {
+		err = 0;
+		irq = platform_get_irq(pmu_device, i);
+		if (irq < 0)
+			continue;
+
+		/*
+		 * If we have a single PMU interrupt that we can't shift,
+		 * assume that we're running on a uniprocessor machine and
+		 * continue. Otherwise, continue without this interrupt.
+		 */
+		if (irq_set_affinity(irq, cpumask_of(i)) && irqs > 1) {
+			pr_warning("unable to set irq affinity (irq=%d, cpu=%u)\n",
+				    irq, i);
+			continue;
+		}
+
+		err = request_irq(irq, armpmu->handle_irq,
+				  IRQF_NOBALANCING,
+				  "arm-pmu", armpmu);
+		if (err) {
+			pr_err("unable to request IRQ%d for ARM PMU counters\n",
+				irq);
+			armpmu_release_hardware(armpmu);
+			return err;
+		}
+
+		cpumask_set_cpu(i, &armpmu->active_irqs);
+	}
+
+	return 0;
+}
+
+static void
+hw_perf_event_destroy(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	atomic_t *active_events	 = &armpmu->active_events;
+	struct mutex *pmu_reserve_mutex = &armpmu->reserve_mutex;
+
+	if (atomic_dec_and_mutex_lock(active_events, pmu_reserve_mutex)) {
+		armpmu_release_hardware(armpmu);
+		mutex_unlock(pmu_reserve_mutex);
+	}
+}
+
+static int
+event_requires_mode_exclusion(struct perf_event_attr *attr)
+{
+	return attr->exclude_idle || attr->exclude_user ||
+	       attr->exclude_kernel || attr->exclude_hv;
+}
+
+static int
+__hw_perf_event_init(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int mapping, err;
+
+	mapping = armpmu->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. For SMP systems, each core has it's own PMU so we can't do any
+	 * clever allocation or constraints checking at this point.
+	 */
+	hwc->idx		= -1;
+	hwc->config_base	= 0;
+	hwc->config		= 0;
+	hwc->event_base		= 0;
+
+	/*
+	 * Check whether we need to exclude the counter from certain modes.
+	 */
+	if ((!armpmu->set_event_filter ||
+	     armpmu->set_event_filter(hwc, &event->attr)) &&
+	     event_requires_mode_exclusion(&event->attr)) {
+		pr_debug("ARM performance counters do not support mode exclusion\n");
+		return -EPERM;
+	}
+
+	/*
+	 * Store the event encoding into the config_base field.
+	 */
+	hwc->config_base	    |= (unsigned long)mapping;
+
+	if (!hwc->sample_period) {
+		/*
+		 * For non-sampling runs, 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  = armpmu->max_period >> 1;
+		hwc->last_period    = hwc->sample_period;
+		local64_set(&hwc->period_left, hwc->sample_period);
+	}
+
+	err = 0;
+	if (event->group_leader != event) {
+		err = validate_group(event);
+		if (err)
+			return -EINVAL;
+	}
+
+	return err;
+}
+
+static int armpmu_event_init(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	int err = 0;
+	atomic_t *active_events = &armpmu->active_events;
+
+	if (armpmu->map_event(event) == -ENOENT)
+		return -ENOENT;
+
+	event->destroy = hw_perf_event_destroy;
+
+	if (!atomic_inc_not_zero(active_events)) {
+		mutex_lock(&armpmu->reserve_mutex);
+		if (atomic_read(active_events) == 0)
+			err = armpmu_reserve_hardware(armpmu);
+
+		if (!err)
+			atomic_inc(active_events);
+		mutex_unlock(&armpmu->reserve_mutex);
+	}
+
+	if (err)
+		return err;
+
+	err = __hw_perf_event_init(event);
+	if (err)
+		hw_perf_event_destroy(event);
+
+	return err;
+}
+
+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();
+	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);
+
+	if (enabled)
+		armpmu->start();
+}
+
+static void armpmu_disable(struct pmu *pmu)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(pmu);
+	armpmu->stop();
+}
+
+static void __init armpmu_init(struct arm_pmu *armpmu)
+{
+	atomic_set(&armpmu->active_events, 0);
+	mutex_init(&armpmu->reserve_mutex);
+
+	armpmu->pmu = (struct pmu) {
+		.pmu_enable	= armpmu_enable,
+		.pmu_disable	= armpmu_disable,
+		.event_init	= armpmu_event_init,
+		.add		= armpmu_add,
+		.del		= armpmu_del,
+		.start		= armpmu_start,
+		.stop		= armpmu_stop,
+		.read		= armpmu_read,
+	};
+}
+
+int __init armpmu_register(struct arm_pmu *armpmu, char *name, int type)
+{
+	armpmu_init(armpmu);
+	return perf_pmu_register(&armpmu->pmu, name, type);
+}
+
+/*
+ * ARMv8 PMUv3 Performance Events handling code.
+ * Common event types.
+ */
+enum armv8_pmuv3_perf_types {
+	/* Required events. */
+	ARMV8_PMUV3_PERFCTR_PMNC_SW_INCR			= 0x00,
+	ARMV8_PMUV3_PERFCTR_L1_DCACHE_REFILL			= 0x03,
+	ARMV8_PMUV3_PERFCTR_L1_DCACHE_ACCESS			= 0x04,
+	ARMV8_PMUV3_PERFCTR_PC_BRANCH_MIS_PRED			= 0x10,
+	ARMV8_PMUV3_PERFCTR_CLOCK_CYCLES			= 0x11,
+	ARMV8_PMUV3_PERFCTR_PC_BRANCH_PRED			= 0x12,
+
+	/* At least one of the following is required. */
+	ARMV8_PMUV3_PERFCTR_INSTR_EXECUTED			= 0x08,
+	ARMV8_PMUV3_PERFCTR_OP_SPEC				= 0x1B,
+
+	/* Common architectural events. */
+	ARMV8_PMUV3_PERFCTR_MEM_READ				= 0x06,
+	ARMV8_PMUV3_PERFCTR_MEM_WRITE				= 0x07,
+	ARMV8_PMUV3_PERFCTR_EXC_TAKEN				= 0x09,
+	ARMV8_PMUV3_PERFCTR_EXC_EXECUTED			= 0x0A,
+	ARMV8_PMUV3_PERFCTR_CID_WRITE				= 0x0B,
+	ARMV8_PMUV3_PERFCTR_PC_WRITE				= 0x0C,
+	ARMV8_PMUV3_PERFCTR_PC_IMM_BRANCH			= 0x0D,
+	ARMV8_PMUV3_PERFCTR_PC_PROC_RETURN			= 0x0E,
+	ARMV8_PMUV3_PERFCTR_MEM_UNALIGNED_ACCESS		= 0x0F,
+	ARMV8_PMUV3_PERFCTR_TTBR_WRITE				= 0x1C,
+
+	/* Common microarchitectural events. */
+	ARMV8_PMUV3_PERFCTR_L1_ICACHE_REFILL			= 0x01,
+	ARMV8_PMUV3_PERFCTR_ITLB_REFILL				= 0x02,
+	ARMV8_PMUV3_PERFCTR_DTLB_REFILL				= 0x05,
+	ARMV8_PMUV3_PERFCTR_MEM_ACCESS				= 0x13,
+	ARMV8_PMUV3_PERFCTR_L1_ICACHE_ACCESS			= 0x14,
+	ARMV8_PMUV3_PERFCTR_L1_DCACHE_WB			= 0x15,
+	ARMV8_PMUV3_PERFCTR_L2_CACHE_ACCESS			= 0x16,
+	ARMV8_PMUV3_PERFCTR_L2_CACHE_REFILL			= 0x17,
+	ARMV8_PMUV3_PERFCTR_L2_CACHE_WB				= 0x18,
+	ARMV8_PMUV3_PERFCTR_BUS_ACCESS				= 0x19,
+	ARMV8_PMUV3_PERFCTR_MEM_ERROR				= 0x1A,
+	ARMV8_PMUV3_PERFCTR_BUS_CYCLES				= 0x1D,
+
+	/*
+	 * This isn't an architected event.
+	 * We detect this event number and use the cycle counter instead.
+	 */
+	ARMV8_PMUV3_PERFCTR_CPU_CYCLES				= 0xFF,
+};
+
+/* PMUv3 HW events mapping. */
+static const unsigned armv8_pmuv3_perf_map[PERF_COUNT_HW_MAX] = {
+	[PERF_COUNT_HW_CPU_CYCLES]		= ARMV8_PMUV3_PERFCTR_CPU_CYCLES,
+	[PERF_COUNT_HW_INSTRUCTIONS]		= ARMV8_PMUV3_PERFCTR_INSTR_EXECUTED,
+	[PERF_COUNT_HW_CACHE_REFERENCES]	= ARMV8_PMUV3_PERFCTR_L1_DCACHE_ACCESS,
+	[PERF_COUNT_HW_CACHE_MISSES]		= ARMV8_PMUV3_PERFCTR_L1_DCACHE_REFILL,
+	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS]	= HW_OP_UNSUPPORTED,
+	[PERF_COUNT_HW_BRANCH_MISSES]		= ARMV8_PMUV3_PERFCTR_PC_BRANCH_MIS_PRED,
+	[PERF_COUNT_HW_BUS_CYCLES]		= HW_OP_UNSUPPORTED,
+	[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND]	= HW_OP_UNSUPPORTED,
+	[PERF_COUNT_HW_STALLED_CYCLES_BACKEND]	= HW_OP_UNSUPPORTED,
+};
+
+static const unsigned armv8_pmuv3_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
+						[PERF_COUNT_HW_CACHE_OP_MAX]
+						[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
+	[C(L1D)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= ARMV8_PMUV3_PERFCTR_L1_DCACHE_ACCESS,
+			[C(RESULT_MISS)]	= ARMV8_PMUV3_PERFCTR_L1_DCACHE_REFILL,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= ARMV8_PMUV3_PERFCTR_L1_DCACHE_ACCESS,
+			[C(RESULT_MISS)]	= ARMV8_PMUV3_PERFCTR_L1_DCACHE_REFILL,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(L1I)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(LL)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(DTLB)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(ITLB)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(BPU)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= ARMV8_PMUV3_PERFCTR_PC_BRANCH_PRED,
+			[C(RESULT_MISS)]	= ARMV8_PMUV3_PERFCTR_PC_BRANCH_MIS_PRED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= ARMV8_PMUV3_PERFCTR_PC_BRANCH_PRED,
+			[C(RESULT_MISS)]	= ARMV8_PMUV3_PERFCTR_PC_BRANCH_MIS_PRED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+	[C(NODE)] = {
+		[C(OP_READ)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_WRITE)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+		[C(OP_PREFETCH)] = {
+			[C(RESULT_ACCESS)]	= CACHE_OP_UNSUPPORTED,
+			[C(RESULT_MISS)]	= CACHE_OP_UNSUPPORTED,
+		},
+	},
+};
+
+/*
+ * Perf Events' indices
+ */
+#define	ARMV8_IDX_CYCLE_COUNTER	0
+#define	ARMV8_IDX_COUNTER0	1
+#define	ARMV8_IDX_COUNTER_LAST	(ARMV8_IDX_CYCLE_COUNTER + cpu_pmu->num_events - 1)
+
+#define	ARMV8_MAX_COUNTERS	32
+#define	ARMV8_COUNTER_MASK	(ARMV8_MAX_COUNTERS - 1)
+
+/*
+ * ARMv8 low level PMU access
+ */
+
+/*
+ * Perf Event to low level counters mapping
+ */
+#define	ARMV8_IDX_TO_COUNTER(x)	\
+	(((x) - ARMV8_IDX_COUNTER0) & ARMV8_COUNTER_MASK)
+
+/*
+ * Per-CPU PMCR: config reg
+ */
+#define ARMV8_PMCR_E		(1 << 0) /* Enable all counters */
+#define ARMV8_PMCR_P		(1 << 1) /* Reset all counters */
+#define ARMV8_PMCR_C		(1 << 2) /* Cycle counter reset */
+#define ARMV8_PMCR_D		(1 << 3) /* CCNT counts every 64th cpu cycle */
+#define ARMV8_PMCR_X		(1 << 4) /* Export to ETM */
+#define ARMV8_PMCR_DP		(1 << 5) /* Disable CCNT if non-invasive debug*/
+#define	ARMV8_PMCR_N_SHIFT	11	 /* Number of counters supported */
+#define	ARMV8_PMCR_N_MASK	0x1f
+#define	ARMV8_PMCR_MASK		0x3f	 /* Mask for writable bits */
+
+/*
+ * PMOVSR: counters overflow flag status reg
+ */
+#define	ARMV8_OVSR_MASK		0xffffffff	/* Mask for writable bits */
+#define	ARMV8_OVERFLOWED_MASK	ARMV8_OVSR_MASK
+
+/*
+ * PMXEVTYPER: Event selection reg
+ */
+#define	ARMV8_EVTYPE_MASK	0xc00000ff	/* Mask for writable bits */
+#define	ARMV8_EVTYPE_EVENT	0xff		/* Mask for EVENT bits */
+
+/*
+ * Event filters for PMUv3
+ */
+#define	ARMV8_EXCLUDE_EL1	(1 << 31)
+#define	ARMV8_EXCLUDE_EL0	(1 << 30)
+#define	ARMV8_INCLUDE_EL2	(1 << 27)
+
+static inline u32 armv8pmu_pmcr_read(void)
+{
+	u32 val;
+	asm volatile("mrs %0, pmcr_el0" : "=r" (val));
+	return val;
+}
+
+static inline void armv8pmu_pmcr_write(u32 val)
+{
+	val &= ARMV8_PMCR_MASK;
+	isb();
+	asm volatile("msr pmcr_el0, %0" :: "r" (val));
+}
+
+static inline int armv8pmu_has_overflowed(u32 pmovsr)
+{
+	return pmovsr & ARMV8_OVERFLOWED_MASK;
+}
+
+static inline int armv8pmu_counter_valid(int idx)
+{
+	return idx >= ARMV8_IDX_CYCLE_COUNTER && idx <= ARMV8_IDX_COUNTER_LAST;
+}
+
+static inline int armv8pmu_counter_has_overflowed(u32 pmnc, int idx)
+{
+	int ret = 0;
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u checking wrong counter %d overflow status\n",
+			smp_processor_id(), idx);
+	} else {
+		counter = ARMV8_IDX_TO_COUNTER(idx);
+		ret = pmnc & BIT(counter);
+	}
+
+	return ret;
+}
+
+static inline int armv8pmu_select_counter(int idx)
+{
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u selecting wrong PMNC counter %d\n",
+			smp_processor_id(), idx);
+		return -EINVAL;
+	}
+
+	counter = ARMV8_IDX_TO_COUNTER(idx);
+	asm volatile("msr pmselr_el0, %0" :: "r" (counter));
+	isb();
+
+	return idx;
+}
+
+static inline u32 armv8pmu_read_counter(int idx)
+{
+	u32 value = 0;
+
+	if (!armv8pmu_counter_valid(idx))
+		pr_err("CPU%u reading wrong counter %d\n",
+			smp_processor_id(), idx);
+	else if (idx == ARMV8_IDX_CYCLE_COUNTER)
+		asm volatile("mrs %0, pmccntr_el0" : "=r" (value));
+	else if (armv8pmu_select_counter(idx) == idx)
+		asm volatile("mrs %0, pmxevcntr_el0" : "=r" (value));
+
+	return value;
+}
+
+static inline void armv8pmu_write_counter(int idx, u32 value)
+{
+	if (!armv8pmu_counter_valid(idx))
+		pr_err("CPU%u writing wrong counter %d\n",
+			smp_processor_id(), idx);
+	else if (idx == ARMV8_IDX_CYCLE_COUNTER)
+		asm volatile("msr pmccntr_el0, %0" :: "r" (value));
+	else if (armv8pmu_select_counter(idx) == idx)
+		asm volatile("msr pmxevcntr_el0, %0" :: "r" (value));
+}
+
+static inline void armv8pmu_write_evtype(int idx, u32 val)
+{
+	if (armv8pmu_select_counter(idx) == idx) {
+		val &= ARMV8_EVTYPE_MASK;
+		asm volatile("msr pmxevtyper_el0, %0" :: "r" (val));
+	}
+}
+
+static inline int armv8pmu_enable_counter(int idx)
+{
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u enabling wrong PMNC counter %d\n",
+			smp_processor_id(), idx);
+		return -EINVAL;
+	}
+
+	counter = ARMV8_IDX_TO_COUNTER(idx);
+	asm volatile("msr pmcntenset_el0, %0" :: "r" (BIT(counter)));
+	return idx;
+}
+
+static inline int armv8pmu_disable_counter(int idx)
+{
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u disabling wrong PMNC counter %d\n",
+			smp_processor_id(), idx);
+		return -EINVAL;
+	}
+
+	counter = ARMV8_IDX_TO_COUNTER(idx);
+	asm volatile("msr pmcntenclr_el0, %0" :: "r" (BIT(counter)));
+	return idx;
+}
+
+static inline int armv8pmu_enable_intens(int idx)
+{
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u enabling wrong PMNC counter IRQ enable %d\n",
+			smp_processor_id(), idx);
+		return -EINVAL;
+	}
+
+	counter = ARMV8_IDX_TO_COUNTER(idx);
+	asm volatile("msr pmintenset_el1, %0" :: "r" (BIT(counter)));
+	return idx;
+}
+
+static inline int armv8pmu_disable_intens(int idx)
+{
+	u32 counter;
+
+	if (!armv8pmu_counter_valid(idx)) {
+		pr_err("CPU%u disabling wrong PMNC counter IRQ enable %d\n",
+			smp_processor_id(), idx);
+		return -EINVAL;
+	}
+
+	counter = ARMV8_IDX_TO_COUNTER(idx);
+	asm volatile("msr pmintenclr_el1, %0" :: "r" (BIT(counter)));
+	isb();
+	/* Clear the overflow flag in case an interrupt is pending. */
+	asm volatile("msr pmovsclr_el0, %0" :: "r" (BIT(counter)));
+	isb();
+	return idx;
+}
+
+static inline u32 armv8pmu_getreset_flags(void)
+{
+	u32 value;
+
+	/* Read */
+	asm volatile("mrs %0, pmovsclr_el0" : "=r" (value));
+
+	/* Write to clear flags */
+	value &= ARMV8_OVSR_MASK;
+	asm volatile("msr pmovsclr_el0, %0" :: "r" (value));
+
+	return value;
+}
+
+static void armv8pmu_enable_event(struct hw_perf_event *hwc, int idx)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+
+	/*
+	 * Enable counter and interrupt, and set the counter to count
+	 * the event that we're interested in.
+	 */
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+
+	/*
+	 * Disable counter
+	 */
+	armv8pmu_disable_counter(idx);
+
+	/*
+	 * Set event (if destined for PMNx counters).
+	 */
+	armv8pmu_write_evtype(idx, hwc->config_base);
+
+	/*
+	 * Enable interrupt for this counter
+	 */
+	armv8pmu_enable_intens(idx);
+
+	/*
+	 * Enable counter
+	 */
+	armv8pmu_enable_counter(idx);
+
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static void armv8pmu_disable_event(struct hw_perf_event *hwc, int idx)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+
+	/*
+	 * Disable counter and interrupt
+	 */
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+
+	/*
+	 * Disable counter
+	 */
+	armv8pmu_disable_counter(idx);
+
+	/*
+	 * Disable interrupt for this counter
+	 */
+	armv8pmu_disable_intens(idx);
+
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static irqreturn_t armv8pmu_handle_irq(int irq_num, void *dev)
+{
+	u32 pmovsr;
+	struct perf_sample_data data;
+	struct pmu_hw_events *cpuc;
+	struct pt_regs *regs;
+	int idx;
+
+	/*
+	 * Get and reset the IRQ flags
+	 */
+	pmovsr = armv8pmu_getreset_flags();
+
+	/*
+	 * Did an overflow occur?
+	 */
+	if (!armv8pmu_has_overflowed(pmovsr))
+		return IRQ_NONE;
+
+	/*
+	 * Handle the counter(s) overflow(s)
+	 */
+	regs = get_irq_regs();
+
+	cpuc = &__get_cpu_var(cpu_hw_events);
+	for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
+		struct perf_event *event = cpuc->events[idx];
+		struct hw_perf_event *hwc;
+
+		/* Ignore if we don't have an event. */
+		if (!event)
+			continue;
+
+		/*
+		 * We have a single interrupt for all counters. Check that
+		 * each counter has overflowed before we process it.
+		 */
+		if (!armv8pmu_counter_has_overflowed(pmovsr, idx))
+			continue;
+
+		hwc = &event->hw;
+		armpmu_event_update(event, hwc, idx);
+		perf_sample_data_init(&data, 0, hwc->last_period);
+		if (!armpmu_event_set_period(event, hwc, idx))
+			continue;
+
+		if (perf_event_overflow(event, &data, regs))
+			cpu_pmu->disable(hwc, idx);
+	}
+
+	/*
+	 * Handle the pending perf events.
+	 *
+	 * Note: this call *must* be run with interrupts disabled. For
+	 * platforms that can have the PMU interrupts raised as an NMI, this
+	 * will not work.
+	 */
+	irq_work_run();
+
+	return IRQ_HANDLED;
+}
+
+static void armv8pmu_start(void)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	/* Enable all counters */
+	armv8pmu_pmcr_write(armv8pmu_pmcr_read() | ARMV8_PMCR_E);
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static void armv8pmu_stop(void)
+{
+	unsigned long flags;
+	struct pmu_hw_events *events = cpu_pmu->get_hw_events();
+
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+	/* Disable all counters */
+	armv8pmu_pmcr_write(armv8pmu_pmcr_read() & ~ARMV8_PMCR_E);
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static int armv8pmu_get_event_idx(struct pmu_hw_events *cpuc,
+				  struct hw_perf_event *event)
+{
+	int idx;
+	unsigned long evtype = event->config_base & ARMV8_EVTYPE_EVENT;
+
+	/* Always place a cycle counter into the cycle counter. */
+	if (evtype == ARMV8_PMUV3_PERFCTR_CPU_CYCLES) {
+		if (test_and_set_bit(ARMV8_IDX_CYCLE_COUNTER, cpuc->used_mask))
+			return -EAGAIN;
+
+		return ARMV8_IDX_CYCLE_COUNTER;
+	}
+
+	/*
+	 * For anything other than a cycle counter, try and use
+	 * the events counters
+	 */
+	for (idx = ARMV8_IDX_COUNTER0; idx < cpu_pmu->num_events; ++idx) {
+		if (!test_and_set_bit(idx, cpuc->used_mask))
+			return idx;
+	}
+
+	/* The counters are all in use. */
+	return -EAGAIN;
+}
+
+/*
+ * Add an event filter to a given event. This will only work for PMUv2 PMUs.
+ */
+static int armv8pmu_set_event_filter(struct hw_perf_event *event,
+				     struct perf_event_attr *attr)
+{
+	unsigned long config_base = 0;
+
+	if (attr->exclude_idle)
+		return -EPERM;
+	if (attr->exclude_user)
+		config_base |= ARMV8_EXCLUDE_EL0;
+	if (attr->exclude_kernel)
+		config_base |= ARMV8_EXCLUDE_EL1;
+	if (!attr->exclude_hv)
+		config_base |= ARMV8_INCLUDE_EL2;
+
+	/*
+	 * Install the filter into config_base as this is used to
+	 * construct the event type.
+	 */
+	event->config_base = config_base;
+
+	return 0;
+}
+
+static void armv8pmu_reset(void *info)
+{
+	u32 idx, nb_cnt = cpu_pmu->num_events;
+
+	/* The counter and interrupt enable registers are unknown at reset. */
+	for (idx = ARMV8_IDX_CYCLE_COUNTER; idx < nb_cnt; ++idx)
+		armv8pmu_disable_event(NULL, idx);
+
+	/* Initialize & Reset PMNC: C and P bits. */
+	armv8pmu_pmcr_write(ARMV8_PMCR_P | ARMV8_PMCR_C);
+
+	/* Disable access from userspace. */
+	asm volatile("msr pmuserenr_el0, %0" :: "r" (0));
+}
+
+static int armv8_pmuv3_map_event(struct perf_event *event)
+{
+	return map_cpu_event(event, &armv8_pmuv3_perf_map,
+				&armv8_pmuv3_perf_cache_map, 0xFF);
+}
+
+static struct arm_pmu armv8pmu = {
+	.handle_irq		= armv8pmu_handle_irq,
+	.enable			= armv8pmu_enable_event,
+	.disable		= armv8pmu_disable_event,
+	.read_counter		= armv8pmu_read_counter,
+	.write_counter		= armv8pmu_write_counter,
+	.get_event_idx		= armv8pmu_get_event_idx,
+	.start			= armv8pmu_start,
+	.stop			= armv8pmu_stop,
+	.reset			= armv8pmu_reset,
+	.max_period		= (1LLU << 32) - 1,
+};
+
+static u32 __init armv8pmu_read_num_pmnc_events(void)
+{
+	u32 nb_cnt;
+
+	/* Read the nb of CNTx counters supported from PMNC */
+	nb_cnt = (armv8pmu_pmcr_read() >> ARMV8_PMCR_N_SHIFT) & ARMV8_PMCR_N_MASK;
+
+	/* Add the CPU cycles counter and return */
+	return nb_cnt + 1;
+}
+
+static struct arm_pmu *__init armv8_pmuv3_pmu_init(void)
+{
+	armv8pmu.name			= "arm/armv8-pmuv3";
+	armv8pmu.map_event		= armv8_pmuv3_map_event;
+	armv8pmu.num_events		= armv8pmu_read_num_pmnc_events();
+	armv8pmu.set_event_filter	= armv8pmu_set_event_filter;
+	return &armv8pmu;
+}
+
+/*
+ * Ensure the PMU has sane values out of reset.
+ * This requires SMP to be available, so exists as a separate initcall.
+ */
+static int __init
+cpu_pmu_reset(void)
+{
+	if (cpu_pmu && cpu_pmu->reset)
+		return on_each_cpu(cpu_pmu->reset, NULL, 1);
+	return 0;
+}
+arch_initcall(cpu_pmu_reset);
+
+/*
+ * PMU platform driver and devicetree bindings.
+ */
+static struct of_device_id armpmu_of_device_ids[] = {
+	{.compatible = "arm,armv8-pmuv3"},
+	{},
+};
+
+static int __devinit armpmu_device_probe(struct platform_device *pdev)
+{
+	if (!cpu_pmu)
+		return -ENODEV;
+
+	cpu_pmu->plat_device = pdev;
+	return 0;
+}
+
+static struct platform_driver armpmu_driver = {
+	.driver		= {
+		.name	= "arm-pmu",
+		.of_match_table = armpmu_of_device_ids,
+	},
+	.probe		= armpmu_device_probe,
+};
+
+static int __init register_pmu_driver(void)
+{
+	return platform_driver_register(&armpmu_driver);
+}
+device_initcall(register_pmu_driver);
+
+static struct pmu_hw_events *armpmu_get_cpu_events(void)
+{
+	return &__get_cpu_var(cpu_hw_events);
+}
+
+static void __init cpu_pmu_init(struct arm_pmu *armpmu)
+{
+	int cpu;
+	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);
+		raw_spin_lock_init(&events->pmu_lock);
+	}
+	armpmu->get_hw_events = armpmu_get_cpu_events;
+}
+
+static int __init init_hw_perf_events(void)
+{
+	u64 dfr = read_cpuid(ID_AA64DFR0_EL1);
+
+	switch ((dfr >> 8) & 0xf) {
+	case 0x1:	/* PMUv3 */
+		cpu_pmu = armv8_pmuv3_pmu_init();
+		break;
+	}
+
+	if (cpu_pmu) {
+		pr_info("enabled with %s PMU driver, %d counters available\n",
+			cpu_pmu->name, cpu_pmu->num_events);
+		cpu_pmu_init(cpu_pmu);
+		armpmu_register(cpu_pmu, "cpu", PERF_TYPE_RAW);
+	} else {
+		pr_info("no hardware support available\n");
+	}
+
+	return 0;
+}
+early_initcall(init_hw_perf_events);
+
+/*
+ * Callchain handling code.
+ */
+struct frame_tail {
+	struct frame_tail   __user *fp;
+	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;
+
+	/* Also check accessibility of one struct frame_tail beyond */
+	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 >= buftail.fp)
+		return NULL;
+
+	return buftail.fp;
+}
+
+void perf_callchain_user(struct perf_callchain_entry *entry,
+			 struct pt_regs *regs)
+{
+	struct frame_tail __user *tail;
+
+	tail = (struct frame_tail __user *)regs->regs[29];
+
+	while (entry->nr < PERF_MAX_STACK_DEPTH &&
+	       tail && !((unsigned long)tail & 0xf))
+		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 *frame, void *data)
+{
+	struct perf_callchain_entry *entry = data;
+	perf_callchain_store(entry, frame->pc);
+	return 0;
+}
+
+void perf_callchain_kernel(struct perf_callchain_entry *entry,
+			   struct pt_regs *regs)
+{
+	struct stackframe frame;
+
+	frame.fp = regs->regs[29];
+	frame.sp = regs->sp;
+	frame.pc = regs->pc;
+	walk_stackframe(&frame, callchain_trace, entry);
+}
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index f960ccb..8c36763 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -88,6 +88,12 @@ void get_term_dimensions(struct winsize *ws);
 #define CPUINFO_PROC	"Processor"
 #endif
 
+#ifdef __aarch64__
+#include "../../arch/arm64/include/asm/unistd.h"
+#define rmb()		asm volatile("dmb ld" ::: "memory")
+#define cpu_relax()	asm volatile("yield" ::: "memory")
+#endif
+
 #ifdef __mips__
 #include "../../arch/mips/include/asm/unistd.h"
 #define rmb()		asm volatile(					\