Adding upstream version 5.10.209.upstream/5.10.209upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

23 files changed, 17400 insertions, 0 deletions

diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
new file mode 100644
index 000000000..130327ff0
--- /dev/null
+++ b/drivers/perf/Kconfig
@@ -0,0 +1,135 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Performance Monitor Drivers
+#
+
+menu "Performance monitor support"
+	depends on PERF_EVENTS
+
+config ARM_CCI_PMU
+	tristate "ARM CCI PMU driver"
+	depends on (ARM && CPU_V7) || ARM64
+	select ARM_CCI
+	help
+	  Support for PMU events monitoring on the ARM CCI (Cache Coherent
+	  Interconnect) family of products.
+
+	  If compiled as a module, it will be called arm-cci.
+
+config ARM_CCI400_PMU
+	bool "support CCI-400"
+	default y
+	depends on ARM_CCI_PMU
+	select ARM_CCI400_COMMON
+	help
+	  CCI-400 provides 4 independent event counters counting events related
+	  to the connected slave/master interfaces, plus a cycle counter.
+
+config ARM_CCI5xx_PMU
+	bool "support CCI-500/CCI-550"
+	default y
+	depends on ARM_CCI_PMU
+	help
+	  CCI-500/CCI-550 both provide 8 independent event counters, which can
+	  count events pertaining to the slave/master interfaces as well as the
+	  internal events to the CCI.
+
+config ARM_CCN
+	tristate "ARM CCN driver support"
+	depends on ARM || ARM64
+	help
+	  PMU (perf) driver supporting the ARM CCN (Cache Coherent Network)
+	  interconnect.
+
+config ARM_CMN
+	tristate "Arm CMN-600 PMU support"
+	depends on ARM64 || (COMPILE_TEST && 64BIT)
+	help
+	  Support for PMU events monitoring on the Arm CMN-600 Coherent Mesh
+	  Network interconnect.
+
+config ARM_PMU
+	depends on ARM || ARM64
+	bool "ARM PMU framework"
+	default y
+	help
+	  Say y if you want to use CPU performance monitors on ARM-based
+	  systems.
+
+config ARM_PMU_ACPI
+	depends on ARM_PMU && ACPI
+	def_bool y
+
+config ARM_SMMU_V3_PMU
+	 tristate "ARM SMMUv3 Performance Monitors Extension"
+	 depends on ARM64 && ACPI && ARM_SMMU_V3
+	   help
+	   Provides support for the ARM SMMUv3 Performance Monitor Counter
+	   Groups (PMCG), which provide monitoring of transactions passing
+	   through the SMMU and allow the resulting information to be filtered
+	   based on the Stream ID of the corresponding master.
+
+config ARM_DSU_PMU
+	tristate "ARM DynamIQ Shared Unit (DSU) PMU"
+	depends on ARM64
+	  help
+	  Provides support for performance monitor unit in ARM DynamIQ Shared
+	  Unit (DSU). The DSU integrates one or more cores with an L3 memory
+	  system, control logic. The PMU allows counting various events related
+	  to DSU.
+
+config FSL_IMX8_DDR_PMU
+	tristate "Freescale i.MX8 DDR perf monitor"
+	depends on ARCH_MXC
+	  help
+	  Provides support for the DDR performance monitor in i.MX8, which
+	  can give information about memory throughput and other related
+	  events.
+
+config QCOM_L2_PMU
+	bool "Qualcomm Technologies L2-cache PMU"
+	depends on ARCH_QCOM && ARM64 && ACPI
+	select QCOM_KRYO_L2_ACCESSORS
+	  help
+	  Provides support for the L2 cache performance monitor unit (PMU)
+	  in Qualcomm Technologies processors.
+	  Adds the L2 cache PMU into the perf events subsystem for
+	  monitoring L2 cache events.
+
+config QCOM_L3_PMU
+	bool "Qualcomm Technologies L3-cache PMU"
+	depends on ARCH_QCOM && ARM64 && ACPI
+	select QCOM_IRQ_COMBINER
+	help
+	   Provides support for the L3 cache performance monitor unit (PMU)
+	   in Qualcomm Technologies processors.
+	   Adds the L3 cache PMU into the perf events subsystem for
+	   monitoring L3 cache events.
+
+config THUNDERX2_PMU
+	tristate "Cavium ThunderX2 SoC PMU UNCORE"
+	depends on ARCH_THUNDER2 && ARM64 && ACPI && NUMA
+	default m
+	help
+	   Provides support for ThunderX2 UNCORE events.
+	   The SoC has PMU support in its L3 cache controller (L3C) and
+	   in the DDR4 Memory Controller (DMC).
+
+config XGENE_PMU
+        depends on ARCH_XGENE
+        bool "APM X-Gene SoC PMU"
+        default n
+        help
+          Say y if you want to use APM X-Gene SoC performance monitors.
+
+config ARM_SPE_PMU
+	tristate "Enable support for the ARMv8.2 Statistical Profiling Extension"
+	depends on ARM64
+	help
+	  Enable perf support for the ARMv8.2 Statistical Profiling
+	  Extension, which provides periodic sampling of operations in
+	  the CPU pipeline and reports this via the perf AUX interface.
+
+source "drivers/perf/hisilicon/Kconfig"
+
+endmenu
diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
new file mode 100644
index 000000000..5365fd56f
--- /dev/null
+++ b/drivers/perf/Makefile
@@ -0,0 +1,15 @@
+# SPDX-License-Identifier: GPL-2.0
+obj-$(CONFIG_ARM_CCI_PMU) += arm-cci.o
+obj-$(CONFIG_ARM_CCN) += arm-ccn.o
+obj-$(CONFIG_ARM_CMN) += arm-cmn.o
+obj-$(CONFIG_ARM_DSU_PMU) += arm_dsu_pmu.o
+obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o
+obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o
+obj-$(CONFIG_ARM_SMMU_V3_PMU) += arm_smmuv3_pmu.o
+obj-$(CONFIG_FSL_IMX8_DDR_PMU) += fsl_imx8_ddr_perf.o
+obj-$(CONFIG_HISI_PMU) += hisilicon/
+obj-$(CONFIG_QCOM_L2_PMU)	+= qcom_l2_pmu.o
+obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o
+obj-$(CONFIG_THUNDERX2_PMU) += thunderx2_pmu.o
+obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o
+obj-$(CONFIG_ARM_SPE_PMU) += arm_spe_pmu.o
diff --git a/drivers/perf/arm-cci.c b/drivers/perf/arm-cci.c
new file mode 100644
index 000000000..87c4be9dd
--- /dev/null
+++ b/drivers/perf/arm-cci.c
@@ -0,0 +1,1729 @@
+// SPDX-License-Identifier: GPL-2.0
+// CCI Cache Coherent Interconnect PMU driver
+// Copyright (C) 2013-2018 Arm Ltd.
+// Author: Punit Agrawal <punit.agrawal@arm.com>, Suzuki Poulose <suzuki.poulose@arm.com>
+
+#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_device.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>
+
+#define DRIVER_NAME		"ARM-CCI PMU"
+
+#define CCI_PMCR		0x0100
+#define CCI_PID2		0x0fe8
+
+#define CCI_PMCR_CEN		0x00000001
+#define CCI_PMCR_NCNT_MASK	0x0000f800
+#define CCI_PMCR_NCNT_SHIFT	11
+
+#define CCI_PID2_REV_MASK	0xf0
+#define CCI_PID2_REV_SHIFT	4
+
+#define CCI_PMU_EVT_SEL		0x000
+#define CCI_PMU_CNTR		0x004
+#define CCI_PMU_CNTR_CTRL	0x008
+#define CCI_PMU_OVRFLW		0x00c
+
+#define CCI_PMU_OVRFLW_FLAG	1
+
+#define CCI_PMU_CNTR_SIZE(model)	((model)->cntr_size)
+#define CCI_PMU_CNTR_BASE(model, idx)	((idx) * CCI_PMU_CNTR_SIZE(model))
+#define CCI_PMU_CNTR_MASK		((1ULL << 32) -1)
+#define CCI_PMU_CNTR_LAST(cci_pmu)	(cci_pmu->num_cntrs - 1)
+
+#define CCI_PMU_MAX_HW_CNTRS(model) \
+	((model)->num_hw_cntrs + (model)->fixed_hw_cntrs)
+
+/* Types of interfaces that can generate events */
+enum {
+	CCI_IF_SLAVE,
+	CCI_IF_MASTER,
+#ifdef CONFIG_ARM_CCI5xx_PMU
+	CCI_IF_GLOBAL,
+#endif
+	CCI_IF_MAX,
+};
+
+#define NUM_HW_CNTRS_CII_4XX	4
+#define NUM_HW_CNTRS_CII_5XX	8
+#define NUM_HW_CNTRS_MAX	NUM_HW_CNTRS_CII_5XX
+
+#define FIXED_HW_CNTRS_CII_4XX	1
+#define FIXED_HW_CNTRS_CII_5XX	0
+#define FIXED_HW_CNTRS_MAX	FIXED_HW_CNTRS_CII_4XX
+
+#define HW_CNTRS_MAX		(NUM_HW_CNTRS_MAX + FIXED_HW_CNTRS_MAX)
+
+struct event_range {
+	u32 min;
+	u32 max;
+};
+
+struct cci_pmu_hw_events {
+	struct perf_event **events;
+	unsigned long *used_mask;
+	raw_spinlock_t pmu_lock;
+};
+
+struct cci_pmu;
+/*
+ * struct cci_pmu_model:
+ * @fixed_hw_cntrs - Number of fixed event counters
+ * @num_hw_cntrs - Maximum number of programmable event counters
+ * @cntr_size - Size of an event counter mapping
+ */
+struct cci_pmu_model {
+	char *name;
+	u32 fixed_hw_cntrs;
+	u32 num_hw_cntrs;
+	u32 cntr_size;
+	struct attribute **format_attrs;
+	struct attribute **event_attrs;
+	struct event_range event_ranges[CCI_IF_MAX];
+	int (*validate_hw_event)(struct cci_pmu *, unsigned long);
+	int (*get_event_idx)(struct cci_pmu *, struct cci_pmu_hw_events *, unsigned long);
+	void (*write_counters)(struct cci_pmu *, unsigned long *);
+};
+
+static struct cci_pmu_model cci_pmu_models[];
+
+struct cci_pmu {
+	void __iomem *base;
+	void __iomem *ctrl_base;
+	struct pmu pmu;
+	int cpu;
+	int nr_irqs;
+	int *irqs;
+	unsigned long active_irqs;
+	const struct cci_pmu_model *model;
+	struct cci_pmu_hw_events hw_events;
+	struct platform_device *plat_device;
+	int num_cntrs;
+	atomic_t active_events;
+	struct mutex reserve_mutex;
+};
+
+#define to_cci_pmu(c)	(container_of(c, struct cci_pmu, pmu))
+
+static struct cci_pmu *g_cci_pmu;
+
+enum cci_models {
+#ifdef CONFIG_ARM_CCI400_PMU
+	CCI400_R0,
+	CCI400_R1,
+#endif
+#ifdef CONFIG_ARM_CCI5xx_PMU
+	CCI500_R0,
+	CCI550_R0,
+#endif
+	CCI_MODEL_MAX
+};
+
+static void pmu_write_counters(struct cci_pmu *cci_pmu,
+				 unsigned long *mask);
+static ssize_t __maybe_unused cci_pmu_format_show(struct device *dev,
+			struct device_attribute *attr, char *buf);
+static ssize_t __maybe_unused cci_pmu_event_show(struct device *dev,
+			struct device_attribute *attr, char *buf);
+
+#define CCI_EXT_ATTR_ENTRY(_name, _func, _config) 				\
+	&((struct dev_ext_attribute[]) {					\
+		{ __ATTR(_name, S_IRUGO, _func, NULL), (void *)_config }	\
+	})[0].attr.attr
+
+#define CCI_FORMAT_EXT_ATTR_ENTRY(_name, _config) \
+	CCI_EXT_ATTR_ENTRY(_name, cci_pmu_format_show, (char *)_config)
+#define CCI_EVENT_EXT_ATTR_ENTRY(_name, _config) \
+	CCI_EXT_ATTR_ENTRY(_name, cci_pmu_event_show, (unsigned long)_config)
+
+/* CCI400 PMU Specific definitions */
+
+#ifdef CONFIG_ARM_CCI400_PMU
+
+/* Port ids */
+#define CCI400_PORT_S0		0
+#define CCI400_PORT_S1		1
+#define CCI400_PORT_S2		2
+#define CCI400_PORT_S3		3
+#define CCI400_PORT_S4		4
+#define CCI400_PORT_M0		5
+#define CCI400_PORT_M1		6
+#define CCI400_PORT_M2		7
+
+#define CCI400_R1_PX		5
+
+/*
+ * 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
+ * make use of this event in hardware.
+ */
+enum cci400_perf_events {
+	CCI400_PMU_CYCLES = 0xff
+};
+
+#define CCI400_PMU_CYCLE_CNTR_IDX	0
+#define CCI400_PMU_CNTR0_IDX		1
+
+/*
+ * CCI PMU event id is an 8-bit value made of two parts - bits 7:5 for one of 8
+ * ports and bits 4:0 are event codes. There are different event codes
+ * associated with each port type.
+ *
+ * Additionally, the range of events associated with the port types changed
+ * between Rev0 and Rev1.
+ *
+ * The constants below define the range of valid codes for each port type for
+ * the different revisions and are used to validate the event to be monitored.
+ */
+
+#define CCI400_PMU_EVENT_MASK		0xffUL
+#define CCI400_PMU_EVENT_SOURCE_SHIFT	5
+#define CCI400_PMU_EVENT_SOURCE_MASK	0x7
+#define CCI400_PMU_EVENT_CODE_SHIFT	0
+#define CCI400_PMU_EVENT_CODE_MASK	0x1f
+#define CCI400_PMU_EVENT_SOURCE(event) \
+	((event >> CCI400_PMU_EVENT_SOURCE_SHIFT) & \
+			CCI400_PMU_EVENT_SOURCE_MASK)
+#define CCI400_PMU_EVENT_CODE(event) \
+	((event >> CCI400_PMU_EVENT_CODE_SHIFT) & CCI400_PMU_EVENT_CODE_MASK)
+
+#define CCI400_R0_SLAVE_PORT_MIN_EV	0x00
+#define CCI400_R0_SLAVE_PORT_MAX_EV	0x13
+#define CCI400_R0_MASTER_PORT_MIN_EV	0x14
+#define CCI400_R0_MASTER_PORT_MAX_EV	0x1a
+
+#define CCI400_R1_SLAVE_PORT_MIN_EV	0x00
+#define CCI400_R1_SLAVE_PORT_MAX_EV	0x14
+#define CCI400_R1_MASTER_PORT_MIN_EV	0x00
+#define CCI400_R1_MASTER_PORT_MAX_EV	0x11
+
+#define CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(_name, _config) \
+	CCI_EXT_ATTR_ENTRY(_name, cci400_pmu_cycle_event_show, \
+					(unsigned long)_config)
+
+static ssize_t cci400_pmu_cycle_event_show(struct device *dev,
+			struct device_attribute *attr, char *buf);
+
+static struct attribute *cci400_pmu_format_attrs[] = {
+	CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"),
+	CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-7"),
+	NULL
+};
+
+static struct attribute *cci400_r0_pmu_event_attrs[] = {
+	/* Slave events */
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_any, 0x0),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_device, 0x01),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_normal_or_nonshareable, 0x2),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_inner_or_outershareable, 0x3),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maintenance, 0x4),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_mem_barrier, 0x5),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_sync_barrier, 0x6),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg_sync, 0x8),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_tt_full, 0x9),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_last_hs_snoop, 0xA),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall_rvalids_h_rready_l, 0xB),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_any, 0xC),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_device, 0xD),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_normal_or_nonshareable, 0xE),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_inner_or_outershare_wback_wclean, 0xF),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_unique, 0x10),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_line_unique, 0x11),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_evict, 0x12),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall_tt_full, 0x13),
+	/* Master events */
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_retry_speculative_fetch, 0x14),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_addr_hazard, 0x15),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_id_hazard, 0x16),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_tt_full, 0x17),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_barrier_hazard, 0x18),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_barrier_hazard, 0x19),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_tt_full, 0x1A),
+	/* Special event for cycles counter */
+	CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(cycles, 0xff),
+	NULL
+};
+
+static struct attribute *cci400_r1_pmu_event_attrs[] = {
+	/* Slave events */
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_any, 0x0),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_device, 0x01),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_normal_or_nonshareable, 0x2),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_inner_or_outershareable, 0x3),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maintenance, 0x4),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_mem_barrier, 0x5),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_sync_barrier, 0x6),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg_sync, 0x8),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_tt_full, 0x9),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_last_hs_snoop, 0xA),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall_rvalids_h_rready_l, 0xB),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_any, 0xC),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_device, 0xD),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_normal_or_nonshareable, 0xE),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_inner_or_outershare_wback_wclean, 0xF),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_unique, 0x10),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_line_unique, 0x11),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_evict, 0x12),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall_tt_full, 0x13),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_slave_id_hazard, 0x14),
+	/* Master events */
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_retry_speculative_fetch, 0x0),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_stall_cycle_addr_hazard, 0x1),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_master_id_hazard, 0x2),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_hi_prio_rtq_full, 0x3),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_barrier_hazard, 0x4),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_barrier_hazard, 0x5),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_wtq_full, 0x6),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_low_prio_rtq_full, 0x7),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_mid_prio_rtq_full, 0x8),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn0, 0x9),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn1, 0xA),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn2, 0xB),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn3, 0xC),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn0, 0xD),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn1, 0xE),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn2, 0xF),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn3, 0x10),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_unique_or_line_unique_addr_hazard, 0x11),
+	/* Special event for cycles counter */
+	CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(cycles, 0xff),
+	NULL
+};
+
+static ssize_t cci400_pmu_cycle_event_show(struct device *dev,
+			struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+				struct dev_ext_attribute, attr);
+	return snprintf(buf, PAGE_SIZE, "config=0x%lx\n", (unsigned long)eattr->var);
+}
+
+static int cci400_get_event_idx(struct cci_pmu *cci_pmu,
+				struct cci_pmu_hw_events *hw,
+				unsigned long cci_event)
+{
+	int idx;
+
+	/* cycles event idx is fixed */
+	if (cci_event == CCI400_PMU_CYCLES) {
+		if (test_and_set_bit(CCI400_PMU_CYCLE_CNTR_IDX, hw->used_mask))
+			return -EAGAIN;
+
+		return CCI400_PMU_CYCLE_CNTR_IDX;
+	}
+
+	for (idx = CCI400_PMU_CNTR0_IDX; idx <= CCI_PMU_CNTR_LAST(cci_pmu); ++idx)
+		if (!test_and_set_bit(idx, hw->used_mask))
+			return idx;
+
+	/* No counters available */
+	return -EAGAIN;
+}
+
+static int cci400_validate_hw_event(struct cci_pmu *cci_pmu, unsigned long hw_event)
+{
+	u8 ev_source = CCI400_PMU_EVENT_SOURCE(hw_event);
+	u8 ev_code = CCI400_PMU_EVENT_CODE(hw_event);
+	int if_type;
+
+	if (hw_event & ~CCI400_PMU_EVENT_MASK)
+		return -ENOENT;
+
+	if (hw_event == CCI400_PMU_CYCLES)
+		return hw_event;
+
+	switch (ev_source) {
+	case CCI400_PORT_S0:
+	case CCI400_PORT_S1:
+	case CCI400_PORT_S2:
+	case CCI400_PORT_S3:
+	case CCI400_PORT_S4:
+		/* Slave Interface */
+		if_type = CCI_IF_SLAVE;
+		break;
+	case CCI400_PORT_M0:
+	case CCI400_PORT_M1:
+	case CCI400_PORT_M2:
+		/* Master Interface */
+		if_type = CCI_IF_MASTER;
+		break;
+	default:
+		return -ENOENT;
+	}
+
+	if (ev_code >= cci_pmu->model->event_ranges[if_type].min &&
+		ev_code <= cci_pmu->model->event_ranges[if_type].max)
+		return hw_event;
+
+	return -ENOENT;
+}
+
+static int probe_cci400_revision(struct cci_pmu *cci_pmu)
+{
+	int rev;
+	rev = readl_relaxed(cci_pmu->ctrl_base + CCI_PID2) & CCI_PID2_REV_MASK;
+	rev >>= CCI_PID2_REV_SHIFT;
+
+	if (rev < CCI400_R1_PX)
+		return CCI400_R0;
+	else
+		return CCI400_R1;
+}
+
+static const struct cci_pmu_model *probe_cci_model(struct cci_pmu *cci_pmu)
+{
+	if (platform_has_secure_cci_access())
+		return &cci_pmu_models[probe_cci400_revision(cci_pmu)];
+	return NULL;
+}
+#else	/* !CONFIG_ARM_CCI400_PMU */
+static inline struct cci_pmu_model *probe_cci_model(struct cci_pmu *cci_pmu)
+{
+	return NULL;
+}
+#endif	/* CONFIG_ARM_CCI400_PMU */
+
+#ifdef CONFIG_ARM_CCI5xx_PMU
+
+/*
+ * CCI5xx PMU event id is an 9-bit value made of two parts.
+ *	 bits [8:5] - Source for the event
+ *	 bits [4:0] - Event code (specific to type of interface)
+ *
+ *
+ */
+
+/* Port ids */
+#define CCI5xx_PORT_S0			0x0
+#define CCI5xx_PORT_S1			0x1
+#define CCI5xx_PORT_S2			0x2
+#define CCI5xx_PORT_S3			0x3
+#define CCI5xx_PORT_S4			0x4
+#define CCI5xx_PORT_S5			0x5
+#define CCI5xx_PORT_S6			0x6
+
+#define CCI5xx_PORT_M0			0x8
+#define CCI5xx_PORT_M1			0x9
+#define CCI5xx_PORT_M2			0xa
+#define CCI5xx_PORT_M3			0xb
+#define CCI5xx_PORT_M4			0xc
+#define CCI5xx_PORT_M5			0xd
+#define CCI5xx_PORT_M6			0xe
+
+#define CCI5xx_PORT_GLOBAL		0xf
+
+#define CCI5xx_PMU_EVENT_MASK		0x1ffUL
+#define CCI5xx_PMU_EVENT_SOURCE_SHIFT	0x5
+#define CCI5xx_PMU_EVENT_SOURCE_MASK	0xf
+#define CCI5xx_PMU_EVENT_CODE_SHIFT	0x0
+#define CCI5xx_PMU_EVENT_CODE_MASK	0x1f
+
+#define CCI5xx_PMU_EVENT_SOURCE(event)	\
+	((event >> CCI5xx_PMU_EVENT_SOURCE_SHIFT) & CCI5xx_PMU_EVENT_SOURCE_MASK)
+#define CCI5xx_PMU_EVENT_CODE(event)	\
+	((event >> CCI5xx_PMU_EVENT_CODE_SHIFT) & CCI5xx_PMU_EVENT_CODE_MASK)
+
+#define CCI5xx_SLAVE_PORT_MIN_EV	0x00
+#define CCI5xx_SLAVE_PORT_MAX_EV	0x1f
+#define CCI5xx_MASTER_PORT_MIN_EV	0x00
+#define CCI5xx_MASTER_PORT_MAX_EV	0x06
+#define CCI5xx_GLOBAL_PORT_MIN_EV	0x00
+#define CCI5xx_GLOBAL_PORT_MAX_EV	0x0f
+
+
+#define CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(_name, _config) \
+	CCI_EXT_ATTR_ENTRY(_name, cci5xx_pmu_global_event_show, \
+					(unsigned long) _config)
+
+static ssize_t cci5xx_pmu_global_event_show(struct device *dev,
+				struct device_attribute *attr, char *buf);
+
+static struct attribute *cci5xx_pmu_format_attrs[] = {
+	CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"),
+	CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-8"),
+	NULL,
+};
+
+static struct attribute *cci5xx_pmu_event_attrs[] = {
+	/* Slave events */
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_arvalid, 0x0),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_dev, 0x1),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_nonshareable, 0x2),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_shareable_non_alloc, 0x3),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_shareable_alloc, 0x4),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_invalidate, 0x5),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maint, 0x6),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_rval, 0x8),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_rlast_snoop, 0x9),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_awalid, 0xA),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_dev, 0xB),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_non_shareable, 0xC),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wb, 0xD),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wlu, 0xE),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wunique, 0xF),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_evict, 0x10),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_wrevict, 0x11),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_w_data_beat, 0x12),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_srq_acvalid, 0x13),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_srq_read, 0x14),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_srq_clean, 0x15),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_srq_data_transfer_low, 0x16),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_arvalid, 0x17),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall, 0x18),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall, 0x19),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_w_data_stall, 0x1A),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_w_resp_stall, 0x1B),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_srq_stall, 0x1C),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_s_data_stall, 0x1D),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_rq_stall_ot_limit, 0x1E),
+	CCI_EVENT_EXT_ATTR_ENTRY(si_r_stall_arbit, 0x1F),
+
+	/* Master events */
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_r_data_beat_any, 0x0),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_w_data_beat_any, 0x1),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall, 0x2),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_r_data_stall, 0x3),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall, 0x4),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_w_data_stall, 0x5),
+	CCI_EVENT_EXT_ATTR_ENTRY(mi_w_resp_stall, 0x6),
+
+	/* Global events */
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_0_1, 0x0),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_2_3, 0x1),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_4_5, 0x2),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_6_7, 0x3),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_0_1, 0x4),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_2_3, 0x5),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_4_5, 0x6),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_6_7, 0x7),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_back_invalidation, 0x8),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_alloc_busy, 0x9),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_tt_full, 0xA),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_wrq, 0xB),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_cd_hs, 0xC),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_rq_stall_addr_hazard, 0xD),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_rq_stall_tt_full, 0xE),
+	CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_rq_tzmp1_prot, 0xF),
+	NULL
+};
+
+static ssize_t cci5xx_pmu_global_event_show(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+					struct dev_ext_attribute, attr);
+	/* Global events have single fixed source code */
+	return snprintf(buf, PAGE_SIZE, "event=0x%lx,source=0x%x\n",
+				(unsigned long)eattr->var, CCI5xx_PORT_GLOBAL);
+}
+
+/*
+ * CCI500 provides 8 independent event counters that can count
+ * any of the events available.
+ * CCI500 PMU event source ids
+ *	0x0-0x6 - Slave interfaces
+ *	0x8-0xD - Master interfaces
+ *	0xf     - Global Events
+ *	0x7,0xe - Reserved
+ */
+static int cci500_validate_hw_event(struct cci_pmu *cci_pmu,
+					unsigned long hw_event)
+{
+	u32 ev_source = CCI5xx_PMU_EVENT_SOURCE(hw_event);
+	u32 ev_code = CCI5xx_PMU_EVENT_CODE(hw_event);
+	int if_type;
+
+	if (hw_event & ~CCI5xx_PMU_EVENT_MASK)
+		return -ENOENT;
+
+	switch (ev_source) {
+	case CCI5xx_PORT_S0:
+	case CCI5xx_PORT_S1:
+	case CCI5xx_PORT_S2:
+	case CCI5xx_PORT_S3:
+	case CCI5xx_PORT_S4:
+	case CCI5xx_PORT_S5:
+	case CCI5xx_PORT_S6:
+		if_type = CCI_IF_SLAVE;
+		break;
+	case CCI5xx_PORT_M0:
+	case CCI5xx_PORT_M1:
+	case CCI5xx_PORT_M2:
+	case CCI5xx_PORT_M3:
+	case CCI5xx_PORT_M4:
+	case CCI5xx_PORT_M5:
+		if_type = CCI_IF_MASTER;
+		break;
+	case CCI5xx_PORT_GLOBAL:
+		if_type = CCI_IF_GLOBAL;
+		break;
+	default:
+		return -ENOENT;
+	}
+
+	if (ev_code >= cci_pmu->model->event_ranges[if_type].min &&
+		ev_code <= cci_pmu->model->event_ranges[if_type].max)
+		return hw_event;
+
+	return -ENOENT;
+}
+
+/*
+ * CCI550 provides 8 independent event counters that can count
+ * any of the events available.
+ * CCI550 PMU event source ids
+ *	0x0-0x6 - Slave interfaces
+ *	0x8-0xe - Master interfaces
+ *	0xf     - Global Events
+ *	0x7	- Reserved
+ */
+static int cci550_validate_hw_event(struct cci_pmu *cci_pmu,
+					unsigned long hw_event)
+{
+	u32 ev_source = CCI5xx_PMU_EVENT_SOURCE(hw_event);
+	u32 ev_code = CCI5xx_PMU_EVENT_CODE(hw_event);
+	int if_type;
+
+	if (hw_event & ~CCI5xx_PMU_EVENT_MASK)
+		return -ENOENT;
+
+	switch (ev_source) {
+	case CCI5xx_PORT_S0:
+	case CCI5xx_PORT_S1:
+	case CCI5xx_PORT_S2:
+	case CCI5xx_PORT_S3:
+	case CCI5xx_PORT_S4:
+	case CCI5xx_PORT_S5:
+	case CCI5xx_PORT_S6:
+		if_type = CCI_IF_SLAVE;
+		break;
+	case CCI5xx_PORT_M0:
+	case CCI5xx_PORT_M1:
+	case CCI5xx_PORT_M2:
+	case CCI5xx_PORT_M3:
+	case CCI5xx_PORT_M4:
+	case CCI5xx_PORT_M5:
+	case CCI5xx_PORT_M6:
+		if_type = CCI_IF_MASTER;
+		break;
+	case CCI5xx_PORT_GLOBAL:
+		if_type = CCI_IF_GLOBAL;
+		break;
+	default:
+		return -ENOENT;
+	}
+
+	if (ev_code >= cci_pmu->model->event_ranges[if_type].min &&
+		ev_code <= cci_pmu->model->event_ranges[if_type].max)
+		return hw_event;
+
+	return -ENOENT;
+}
+
+#endif	/* CONFIG_ARM_CCI5xx_PMU */
+
+/*
+ * Program the CCI PMU counters which have PERF_HES_ARCH set
+ * with the event period and mark them ready before we enable
+ * PMU.
+ */
+static void cci_pmu_sync_counters(struct cci_pmu *cci_pmu)
+{
+	int i;
+	struct cci_pmu_hw_events *cci_hw = &cci_pmu->hw_events;
+	DECLARE_BITMAP(mask, HW_CNTRS_MAX);
+
+	bitmap_zero(mask, cci_pmu->num_cntrs);
+	for_each_set_bit(i, cci_pmu->hw_events.used_mask, cci_pmu->num_cntrs) {
+		struct perf_event *event = cci_hw->events[i];
+
+		if (WARN_ON(!event))
+			continue;
+
+		/* Leave the events which are not counting */
+		if (event->hw.state & PERF_HES_STOPPED)
+			continue;
+		if (event->hw.state & PERF_HES_ARCH) {
+			set_bit(i, mask);
+			event->hw.state &= ~PERF_HES_ARCH;
+		}
+	}
+
+	pmu_write_counters(cci_pmu, mask);
+}
+
+/* Should be called with cci_pmu->hw_events->pmu_lock held */
+static void __cci_pmu_enable_nosync(struct cci_pmu *cci_pmu)
+{
+	u32 val;
+
+	/* Enable all the PMU counters. */
+	val = readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) | CCI_PMCR_CEN;
+	writel(val, cci_pmu->ctrl_base + CCI_PMCR);
+}
+
+/* Should be called with cci_pmu->hw_events->pmu_lock held */
+static void __cci_pmu_enable_sync(struct cci_pmu *cci_pmu)
+{
+	cci_pmu_sync_counters(cci_pmu);
+	__cci_pmu_enable_nosync(cci_pmu);
+}
+
+/* Should be called with cci_pmu->hw_events->pmu_lock held */
+static void __cci_pmu_disable(struct cci_pmu *cci_pmu)
+{
+	u32 val;
+
+	/* Disable all the PMU counters. */
+	val = readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) & ~CCI_PMCR_CEN;
+	writel(val, cci_pmu->ctrl_base + CCI_PMCR);
+}
+
+static ssize_t cci_pmu_format_show(struct device *dev,
+			struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+				struct dev_ext_attribute, attr);
+	return snprintf(buf, PAGE_SIZE, "%s\n", (char *)eattr->var);
+}
+
+static ssize_t cci_pmu_event_show(struct device *dev,
+			struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+				struct dev_ext_attribute, attr);
+	/* source parameter is mandatory for normal PMU events */
+	return snprintf(buf, PAGE_SIZE, "source=?,event=0x%lx\n",
+					 (unsigned long)eattr->var);
+}
+
+static int pmu_is_valid_counter(struct cci_pmu *cci_pmu, int idx)
+{
+	return 0 <= idx && idx <= CCI_PMU_CNTR_LAST(cci_pmu);
+}
+
+static u32 pmu_read_register(struct cci_pmu *cci_pmu, int idx, unsigned int offset)
+{
+	return readl_relaxed(cci_pmu->base +
+			     CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
+}
+
+static void pmu_write_register(struct cci_pmu *cci_pmu, u32 value,
+			       int idx, unsigned int offset)
+{
+	writel_relaxed(value, cci_pmu->base +
+		       CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
+}
+
+static void pmu_disable_counter(struct cci_pmu *cci_pmu, int idx)
+{
+	pmu_write_register(cci_pmu, 0, idx, CCI_PMU_CNTR_CTRL);
+}
+
+static void pmu_enable_counter(struct cci_pmu *cci_pmu, int idx)
+{
+	pmu_write_register(cci_pmu, 1, idx, CCI_PMU_CNTR_CTRL);
+}
+
+static bool __maybe_unused
+pmu_counter_is_enabled(struct cci_pmu *cci_pmu, int idx)
+{
+	return (pmu_read_register(cci_pmu, idx, CCI_PMU_CNTR_CTRL) & 0x1) != 0;
+}
+
+static void pmu_set_event(struct cci_pmu *cci_pmu, int idx, unsigned long event)
+{
+	pmu_write_register(cci_pmu, event, idx, CCI_PMU_EVT_SEL);
+}
+
+/*
+ * For all counters on the CCI-PMU, disable any 'enabled' counters,
+ * saving the changed counters in the mask, so that we can restore
+ * it later using pmu_restore_counters. The mask is private to the
+ * caller. We cannot rely on the used_mask maintained by the CCI_PMU
+ * as it only tells us if the counter is assigned to perf_event or not.
+ * The state of the perf_event cannot be locked by the PMU layer, hence
+ * we check the individual counter status (which can be locked by
+ * cci_pm->hw_events->pmu_lock).
+ *
+ * @mask should be initialised to empty by the caller.
+ */
+static void __maybe_unused
+pmu_save_counters(struct cci_pmu *cci_pmu, unsigned long *mask)
+{
+	int i;
+
+	for (i = 0; i < cci_pmu->num_cntrs; i++) {
+		if (pmu_counter_is_enabled(cci_pmu, i)) {
+			set_bit(i, mask);
+			pmu_disable_counter(cci_pmu, i);
+		}
+	}
+}
+
+/*
+ * Restore the status of the counters. Reversal of the pmu_save_counters().
+ * For each counter set in the mask, enable the counter back.
+ */
+static void __maybe_unused
+pmu_restore_counters(struct cci_pmu *cci_pmu, unsigned long *mask)
+{
+	int i;
+
+	for_each_set_bit(i, mask, cci_pmu->num_cntrs)
+		pmu_enable_counter(cci_pmu, i);
+}
+
+/*
+ * Returns the number of programmable counters actually implemented
+ * by the cci
+ */
+static u32 pmu_get_max_counters(struct cci_pmu *cci_pmu)
+{
+	return (readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) &
+		CCI_PMCR_NCNT_MASK) >> CCI_PMCR_NCNT_SHIFT;
+}
+
+static int pmu_get_event_idx(struct cci_pmu_hw_events *hw, struct perf_event *event)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	unsigned long cci_event = event->hw.config_base;
+	int idx;
+
+	if (cci_pmu->model->get_event_idx)
+		return cci_pmu->model->get_event_idx(cci_pmu, hw, cci_event);
+
+	/* Generic code to find an unused idx from the mask */
+	for(idx = 0; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++)
+		if (!test_and_set_bit(idx, hw->used_mask))
+			return idx;
+
+	/* No counters available */
+	return -EAGAIN;
+}
+
+static int pmu_map_event(struct perf_event *event)
+{
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+
+	if (event->attr.type < PERF_TYPE_MAX ||
+			!cci_pmu->model->validate_hw_event)
+		return -ENOENT;
+
+	return	cci_pmu->model->validate_hw_event(cci_pmu, event->attr.config);
+}
+
+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;
+
+	if (unlikely(!pmu_device))
+		return -ENODEV;
+
+	if (cci_pmu->nr_irqs < 1) {
+		dev_err(&pmu_device->dev, "no irqs for CCI PMUs defined\n");
+		return -ENODEV;
+	}
+
+	/*
+	 * Register all available CCI PMU interrupts. In the interrupt handler
+	 * we iterate over the counters checking for interrupt source (the
+	 * overflowing counter) and clear it.
+	 *
+	 * This should allow handling of non-unique interrupt for the counters.
+	 */
+	for (i = 0; i < cci_pmu->nr_irqs; i++) {
+		int err = request_irq(cci_pmu->irqs[i], handler, IRQF_SHARED,
+				"arm-cci-pmu", cci_pmu);
+		if (err) {
+			dev_err(&pmu_device->dev, "unable to request IRQ%d for ARM CCI PMU counters\n",
+				cci_pmu->irqs[i]);
+			return err;
+		}
+
+		set_bit(i, &cci_pmu->active_irqs);
+	}
+
+	return 0;
+}
+
+static void pmu_free_irq(struct cci_pmu *cci_pmu)
+{
+	int i;
+
+	for (i = 0; i < cci_pmu->nr_irqs; i++) {
+		if (!test_and_clear_bit(i, &cci_pmu->active_irqs))
+			continue;
+
+		free_irq(cci_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(cci_pmu, idx, CCI_PMU_CNTR);
+
+	return value;
+}
+
+static void pmu_write_counter(struct cci_pmu *cci_pmu, u32 value, int idx)
+{
+	pmu_write_register(cci_pmu, value, idx, CCI_PMU_CNTR);
+}
+
+static void __pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask)
+{
+	int i;
+	struct cci_pmu_hw_events *cci_hw = &cci_pmu->hw_events;
+
+	for_each_set_bit(i, mask, cci_pmu->num_cntrs) {
+		struct perf_event *event = cci_hw->events[i];
+
+		if (WARN_ON(!event))
+			continue;
+		pmu_write_counter(cci_pmu, local64_read(&event->hw.prev_count), i);
+	}
+}
+
+static void pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask)
+{
+	if (cci_pmu->model->write_counters)
+		cci_pmu->model->write_counters(cci_pmu, mask);
+	else
+		__pmu_write_counters(cci_pmu, mask);
+}
+
+#ifdef CONFIG_ARM_CCI5xx_PMU
+
+/*
+ * CCI-500/CCI-550 has advanced power saving policies, which could gate the
+ * clocks to the PMU counters, which makes the writes to them ineffective.
+ * The only way to write to those counters is when the global counters
+ * are enabled and the particular counter is enabled.
+ *
+ * So we do the following :
+ *
+ * 1) Disable all the PMU counters, saving their current state
+ * 2) Enable the global PMU profiling, now that all counters are
+ *    disabled.
+ *
+ * For each counter to be programmed, repeat steps 3-7:
+ *
+ * 3) Write an invalid event code to the event control register for the
+      counter, so that the counters are not modified.
+ * 4) Enable the counter control for the counter.
+ * 5) Set the counter value
+ * 6) Disable the counter
+ * 7) Restore the event in the target counter
+ *
+ * 8) Disable the global PMU.
+ * 9) Restore the status of the rest of the counters.
+ *
+ * We choose an event which for CCI-5xx is guaranteed not to count.
+ * We use the highest possible event code (0x1f) for the master interface 0.
+ */
+#define CCI5xx_INVALID_EVENT	((CCI5xx_PORT_M0 << CCI5xx_PMU_EVENT_SOURCE_SHIFT) | \
+				 (CCI5xx_PMU_EVENT_CODE_MASK << CCI5xx_PMU_EVENT_CODE_SHIFT))
+static void cci5xx_pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask)
+{
+	int i;
+	DECLARE_BITMAP(saved_mask, HW_CNTRS_MAX);
+
+	bitmap_zero(saved_mask, cci_pmu->num_cntrs);
+	pmu_save_counters(cci_pmu, saved_mask);
+
+	/*
+	 * Now that all the counters are disabled, we can safely turn the PMU on,
+	 * without syncing the status of the counters
+	 */
+	__cci_pmu_enable_nosync(cci_pmu);
+
+	for_each_set_bit(i, mask, cci_pmu->num_cntrs) {
+		struct perf_event *event = cci_pmu->hw_events.events[i];
+
+		if (WARN_ON(!event))
+			continue;
+
+		pmu_set_event(cci_pmu, i, CCI5xx_INVALID_EVENT);
+		pmu_enable_counter(cci_pmu, i);
+		pmu_write_counter(cci_pmu, local64_read(&event->hw.prev_count), i);
+		pmu_disable_counter(cci_pmu, i);
+		pmu_set_event(cci_pmu, i, event->hw.config_base);
+	}
+
+	__cci_pmu_disable(cci_pmu);
+
+	pmu_restore_counters(cci_pmu, saved_mask);
+}
+
+#endif	/* CONFIG_ARM_CCI5xx_PMU */
+
+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);
+}
+
+static 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);
+
+	/*
+	 * CCI PMU uses PERF_HES_ARCH to keep track of the counters, whose
+	 * values needs to be sync-ed with the s/w state before the PMU is
+	 * enabled.
+	 * Mark this counter for sync.
+	 */
+	hwc->state |= PERF_HES_ARCH;
+}
+
+static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
+{
+	unsigned long flags;
+	struct cci_pmu *cci_pmu = dev;
+	struct cci_pmu_hw_events *events = &cci_pmu->hw_events;
+	int idx, handled = IRQ_NONE;
+
+	raw_spin_lock_irqsave(&events->pmu_lock, flags);
+
+	/* Disable the PMU while we walk through the counters */
+	__cci_pmu_disable(cci_pmu);
+	/*
+	 * Iterate over counters and update the corresponding perf events.
+	 * This should work regardless of whether we have per-counter overflow
+	 * interrupt or a combined overflow interrupt.
+	 */
+	for (idx = 0; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++) {
+		struct perf_event *event = events->events[idx];
+
+		if (!event)
+			continue;
+
+		/* Did this counter overflow? */
+		if (!(pmu_read_register(cci_pmu, idx, CCI_PMU_OVRFLW) &
+		      CCI_PMU_OVRFLW_FLAG))
+			continue;
+
+		pmu_write_register(cci_pmu, CCI_PMU_OVRFLW_FLAG, idx,
+							CCI_PMU_OVRFLW);
+
+		pmu_event_update(event);
+		pmu_event_set_period(event);
+		handled = IRQ_HANDLED;
+	}
+
+	/* Enable the PMU and sync possibly overflowed counters */
+	__cci_pmu_enable_sync(cci_pmu);
+	raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+
+	return IRQ_RETVAL(handled);
+}
+
+static int cci_pmu_get_hw(struct cci_pmu *cci_pmu)
+{
+	int ret = pmu_request_irq(cci_pmu, pmu_handle_irq);
+	if (ret) {
+		pmu_free_irq(cci_pmu);
+		return ret;
+	}
+	return 0;
+}
+
+static void cci_pmu_put_hw(struct cci_pmu *cci_pmu)
+{
+	pmu_free_irq(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 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_cntrs);
+	unsigned long flags;
+
+	if (!enabled)
+		return;
+
+	raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
+	__cci_pmu_enable_sync(cci_pmu);
+	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;
+
+	raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
+	__cci_pmu_disable(cci_pmu);
+	raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
+}
+
+/*
+ * Check if the idx represents a non-programmable counter.
+ * All the fixed event counters are mapped before the programmable
+ * counters.
+ */
+static bool pmu_fixed_hw_idx(struct cci_pmu *cci_pmu, int idx)
+{
+	return (idx >= 0) && (idx < cci_pmu->model->fixed_hw_cntrs);
+}
+
+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(&hw_events->pmu_lock, flags);
+
+	/* Configure the counter unless you are counting a fixed event */
+	if (!pmu_fixed_hw_idx(cci_pmu, idx))
+		pmu_set_event(cci_pmu, idx, hwc->config_base);
+
+	pmu_event_set_period(event);
+	pmu_enable_counter(cci_pmu, idx);
+
+	raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
+}
+
+static void cci_pmu_stop(struct perf_event *event, int pmu_flags)
+{
+	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(cci_pmu, idx);
+	pmu_event_update(event);
+	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+}
+
+static int cci_pmu_add(struct perf_event *event, int 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;
+
+	/* If we don't have a space for the counter then finish early. */
+	idx = pmu_get_event_idx(hw_events, event);
+	if (idx < 0)
+		return idx;
+
+	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);
+
+	return 0;
+}
+
+static void cci_pmu_del(struct perf_event *event, int 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;
+
+	cci_pmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[idx] = NULL;
+	clear_bit(idx, hw_events->used_mask);
+
+	perf_event_update_userpage(event);
+}
+
+static int validate_event(struct pmu *cci_pmu,
+			  struct cci_pmu_hw_events *hw_events,
+			  struct perf_event *event)
+{
+	if (is_software_event(event))
+		return 1;
+
+	/*
+	 * Reject groups spanning multiple HW PMUs (e.g. CPU + CCI). The
+	 * core perf code won't check that the pmu->ctx == leader->ctx
+	 * until after pmu->event_init(event).
+	 */
+	if (event->pmu != cci_pmu)
+		return 0;
+
+	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 int validate_group(struct perf_event *event)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+	unsigned long mask[BITS_TO_LONGS(HW_CNTRS_MAX)];
+	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 = mask,
+	};
+	memset(mask, 0, BITS_TO_LONGS(cci_pmu->num_cntrs) * sizeof(unsigned long));
+
+	if (!validate_event(event->pmu, &fake_pmu, leader))
+		return -EINVAL;
+
+	for_each_sibling_event(sibling, leader) {
+		if (!validate_event(event->pmu, &fake_pmu, sibling))
+			return -EINVAL;
+	}
+
+	if (!validate_event(event->pmu, &fake_pmu, event))
+		return -EINVAL;
+
+	return 0;
+}
+
+static int __hw_perf_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int mapping;
+
+	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;
+
+	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;
+
+	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;
+
+	/*
+	 * 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.
+	 */
+	if (event->cpu < 0)
+		return -EINVAL;
+	event->cpu = cci_pmu->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 ssize_t pmu_cpumask_attr_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
+{
+	struct pmu *pmu = dev_get_drvdata(dev);
+	struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(cci_pmu->cpu));
+}
+
+static struct device_attribute pmu_cpumask_attr =
+	__ATTR(cpumask, S_IRUGO, pmu_cpumask_attr_show, NULL);
+
+static struct attribute *pmu_attrs[] = {
+	&pmu_cpumask_attr.attr,
+	NULL,
+};
+
+static struct attribute_group pmu_attr_group = {
+	.attrs = pmu_attrs,
+};
+
+static struct attribute_group pmu_format_attr_group = {
+	.name = "format",
+	.attrs = NULL,		/* Filled in cci_pmu_init_attrs */
+};
+
+static struct attribute_group pmu_event_attr_group = {
+	.name = "events",
+	.attrs = NULL,		/* Filled in cci_pmu_init_attrs */
+};
+
+static const struct attribute_group *pmu_attr_groups[] = {
+	&pmu_attr_group,
+	&pmu_format_attr_group,
+	&pmu_event_attr_group,
+	NULL
+};
+
+static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
+{
+	const struct cci_pmu_model *model = cci_pmu->model;
+	char *name = model->name;
+	u32 num_cntrs;
+
+	if (WARN_ON(model->num_hw_cntrs > NUM_HW_CNTRS_MAX))
+		return -EINVAL;
+	if (WARN_ON(model->fixed_hw_cntrs > FIXED_HW_CNTRS_MAX))
+		return -EINVAL;
+
+	pmu_event_attr_group.attrs = model->event_attrs;
+	pmu_format_attr_group.attrs = model->format_attrs;
+
+	cci_pmu->pmu = (struct pmu) {
+		.module		= THIS_MODULE,
+		.name		= cci_pmu->model->name,
+		.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,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	cci_pmu->plat_device = pdev;
+	num_cntrs = pmu_get_max_counters(cci_pmu);
+	if (num_cntrs > cci_pmu->model->num_hw_cntrs) {
+		dev_warn(&pdev->dev,
+			"PMU implements more counters(%d) than supported by"
+			" the model(%d), truncated.",
+			num_cntrs, cci_pmu->model->num_hw_cntrs);
+		num_cntrs = cci_pmu->model->num_hw_cntrs;
+	}
+	cci_pmu->num_cntrs = num_cntrs + cci_pmu->model->fixed_hw_cntrs;
+
+	return perf_pmu_register(&cci_pmu->pmu, name, -1);
+}
+
+static int cci_pmu_offline_cpu(unsigned int cpu)
+{
+	int target;
+
+	if (!g_cci_pmu || cpu != g_cci_pmu->cpu)
+		return 0;
+
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	perf_pmu_migrate_context(&g_cci_pmu->pmu, cpu, target);
+	g_cci_pmu->cpu = target;
+	return 0;
+}
+
+static __maybe_unused struct cci_pmu_model cci_pmu_models[] = {
+#ifdef CONFIG_ARM_CCI400_PMU
+	[CCI400_R0] = {
+		.name = "CCI_400",
+		.fixed_hw_cntrs = FIXED_HW_CNTRS_CII_4XX, /* Cycle counter */
+		.num_hw_cntrs = NUM_HW_CNTRS_CII_4XX,
+		.cntr_size = SZ_4K,
+		.format_attrs = cci400_pmu_format_attrs,
+		.event_attrs = cci400_r0_pmu_event_attrs,
+		.event_ranges = {
+			[CCI_IF_SLAVE] = {
+				CCI400_R0_SLAVE_PORT_MIN_EV,
+				CCI400_R0_SLAVE_PORT_MAX_EV,
+			},
+			[CCI_IF_MASTER] = {
+				CCI400_R0_MASTER_PORT_MIN_EV,
+				CCI400_R0_MASTER_PORT_MAX_EV,
+			},
+		},
+		.validate_hw_event = cci400_validate_hw_event,
+		.get_event_idx = cci400_get_event_idx,
+	},
+	[CCI400_R1] = {
+		.name = "CCI_400_r1",
+		.fixed_hw_cntrs = FIXED_HW_CNTRS_CII_4XX, /* Cycle counter */
+		.num_hw_cntrs = NUM_HW_CNTRS_CII_4XX,
+		.cntr_size = SZ_4K,
+		.format_attrs = cci400_pmu_format_attrs,
+		.event_attrs = cci400_r1_pmu_event_attrs,
+		.event_ranges = {
+			[CCI_IF_SLAVE] = {
+				CCI400_R1_SLAVE_PORT_MIN_EV,
+				CCI400_R1_SLAVE_PORT_MAX_EV,
+			},
+			[CCI_IF_MASTER] = {
+				CCI400_R1_MASTER_PORT_MIN_EV,
+				CCI400_R1_MASTER_PORT_MAX_EV,
+			},
+		},
+		.validate_hw_event = cci400_validate_hw_event,
+		.get_event_idx = cci400_get_event_idx,
+	},
+#endif
+#ifdef CONFIG_ARM_CCI5xx_PMU
+	[CCI500_R0] = {
+		.name = "CCI_500",
+		.fixed_hw_cntrs = FIXED_HW_CNTRS_CII_5XX,
+		.num_hw_cntrs = NUM_HW_CNTRS_CII_5XX,
+		.cntr_size = SZ_64K,
+		.format_attrs = cci5xx_pmu_format_attrs,
+		.event_attrs = cci5xx_pmu_event_attrs,
+		.event_ranges = {
+			[CCI_IF_SLAVE] = {
+				CCI5xx_SLAVE_PORT_MIN_EV,
+				CCI5xx_SLAVE_PORT_MAX_EV,
+			},
+			[CCI_IF_MASTER] = {
+				CCI5xx_MASTER_PORT_MIN_EV,
+				CCI5xx_MASTER_PORT_MAX_EV,
+			},
+			[CCI_IF_GLOBAL] = {
+				CCI5xx_GLOBAL_PORT_MIN_EV,
+				CCI5xx_GLOBAL_PORT_MAX_EV,
+			},
+		},
+		.validate_hw_event = cci500_validate_hw_event,
+		.write_counters	= cci5xx_pmu_write_counters,
+	},
+	[CCI550_R0] = {
+		.name = "CCI_550",
+		.fixed_hw_cntrs = FIXED_HW_CNTRS_CII_5XX,
+		.num_hw_cntrs = NUM_HW_CNTRS_CII_5XX,
+		.cntr_size = SZ_64K,
+		.format_attrs = cci5xx_pmu_format_attrs,
+		.event_attrs = cci5xx_pmu_event_attrs,
+		.event_ranges = {
+			[CCI_IF_SLAVE] = {
+				CCI5xx_SLAVE_PORT_MIN_EV,
+				CCI5xx_SLAVE_PORT_MAX_EV,
+			},
+			[CCI_IF_MASTER] = {
+				CCI5xx_MASTER_PORT_MIN_EV,
+				CCI5xx_MASTER_PORT_MAX_EV,
+			},
+			[CCI_IF_GLOBAL] = {
+				CCI5xx_GLOBAL_PORT_MIN_EV,
+				CCI5xx_GLOBAL_PORT_MAX_EV,
+			},
+		},
+		.validate_hw_event = cci550_validate_hw_event,
+		.write_counters	= cci5xx_pmu_write_counters,
+	},
+#endif
+};
+
+static const struct of_device_id arm_cci_pmu_matches[] = {
+#ifdef CONFIG_ARM_CCI400_PMU
+	{
+		.compatible = "arm,cci-400-pmu",
+		.data	= NULL,
+	},
+	{
+		.compatible = "arm,cci-400-pmu,r0",
+		.data	= &cci_pmu_models[CCI400_R0],
+	},
+	{
+		.compatible = "arm,cci-400-pmu,r1",
+		.data	= &cci_pmu_models[CCI400_R1],
+	},
+#endif
+#ifdef CONFIG_ARM_CCI5xx_PMU
+	{
+		.compatible = "arm,cci-500-pmu,r0",
+		.data = &cci_pmu_models[CCI500_R0],
+	},
+	{
+		.compatible = "arm,cci-550-pmu,r0",
+		.data = &cci_pmu_models[CCI550_R0],
+	},
+#endif
+	{},
+};
+MODULE_DEVICE_TABLE(of, arm_cci_pmu_matches);
+
+static bool is_duplicate_irq(int irq, int *irqs, int nr_irqs)
+{
+	int i;
+
+	for (i = 0; i < nr_irqs; i++)
+		if (irq == irqs[i])
+			return true;
+
+	return false;
+}
+
+static struct cci_pmu *cci_pmu_alloc(struct device *dev)
+{
+	struct cci_pmu *cci_pmu;
+	const struct cci_pmu_model *model;
+
+	/*
+	 * All allocations are devm_* hence we don't have to free
+	 * them explicitly on an error, as it would end up in driver
+	 * detach.
+	 */
+	cci_pmu = devm_kzalloc(dev, sizeof(*cci_pmu), GFP_KERNEL);
+	if (!cci_pmu)
+		return ERR_PTR(-ENOMEM);
+
+	cci_pmu->ctrl_base = *(void __iomem **)dev->platform_data;
+
+	model = of_device_get_match_data(dev);
+	if (!model) {
+		dev_warn(dev,
+			 "DEPRECATED compatible property, requires secure access to CCI registers");
+		model = probe_cci_model(cci_pmu);
+	}
+	if (!model) {
+		dev_warn(dev, "CCI PMU version not supported\n");
+		return ERR_PTR(-ENODEV);
+	}
+
+	cci_pmu->model = model;
+	cci_pmu->irqs = devm_kcalloc(dev, CCI_PMU_MAX_HW_CNTRS(model),
+					sizeof(*cci_pmu->irqs), GFP_KERNEL);
+	if (!cci_pmu->irqs)
+		return ERR_PTR(-ENOMEM);
+	cci_pmu->hw_events.events = devm_kcalloc(dev,
+					     CCI_PMU_MAX_HW_CNTRS(model),
+					     sizeof(*cci_pmu->hw_events.events),
+					     GFP_KERNEL);
+	if (!cci_pmu->hw_events.events)
+		return ERR_PTR(-ENOMEM);
+	cci_pmu->hw_events.used_mask = devm_kcalloc(dev,
+						BITS_TO_LONGS(CCI_PMU_MAX_HW_CNTRS(model)),
+						sizeof(*cci_pmu->hw_events.used_mask),
+						GFP_KERNEL);
+	if (!cci_pmu->hw_events.used_mask)
+		return ERR_PTR(-ENOMEM);
+
+	return cci_pmu;
+}
+
+static int cci_pmu_probe(struct platform_device *pdev)
+{
+	struct cci_pmu *cci_pmu;
+	int i, ret, irq;
+
+	cci_pmu = cci_pmu_alloc(&pdev->dev);
+	if (IS_ERR(cci_pmu))
+		return PTR_ERR(cci_pmu);
+
+	cci_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(cci_pmu->base))
+		return -ENOMEM;
+
+	/*
+	 * CCI PMU has one overflow interrupt per counter; but some may be tied
+	 * together to a common interrupt.
+	 */
+	cci_pmu->nr_irqs = 0;
+	for (i = 0; i < CCI_PMU_MAX_HW_CNTRS(cci_pmu->model); i++) {
+		irq = platform_get_irq(pdev, i);
+		if (irq < 0)
+			break;
+
+		if (is_duplicate_irq(irq, cci_pmu->irqs, cci_pmu->nr_irqs))
+			continue;
+
+		cci_pmu->irqs[cci_pmu->nr_irqs++] = irq;
+	}
+
+	/*
+	 * Ensure that the device tree has as many interrupts as the number
+	 * of counters.
+	 */
+	if (i < CCI_PMU_MAX_HW_CNTRS(cci_pmu->model)) {
+		dev_warn(&pdev->dev, "In-correct number of interrupts: %d, should be %d\n",
+			i, CCI_PMU_MAX_HW_CNTRS(cci_pmu->model));
+		return -EINVAL;
+	}
+
+	raw_spin_lock_init(&cci_pmu->hw_events.pmu_lock);
+	mutex_init(&cci_pmu->reserve_mutex);
+	atomic_set(&cci_pmu->active_events, 0);
+
+	cci_pmu->cpu = raw_smp_processor_id();
+	g_cci_pmu = cci_pmu;
+	cpuhp_setup_state_nocalls(CPUHP_AP_PERF_ARM_CCI_ONLINE,
+				  "perf/arm/cci:online", NULL,
+				  cci_pmu_offline_cpu);
+
+	ret = cci_pmu_init(cci_pmu, pdev);
+	if (ret)
+		goto error_pmu_init;
+
+	pr_info("ARM %s PMU driver probed", cci_pmu->model->name);
+	return 0;
+
+error_pmu_init:
+	cpuhp_remove_state(CPUHP_AP_PERF_ARM_CCI_ONLINE);
+	g_cci_pmu = NULL;
+	return ret;
+}
+
+static int cci_pmu_remove(struct platform_device *pdev)
+{
+	if (!g_cci_pmu)
+		return 0;
+
+	cpuhp_remove_state(CPUHP_AP_PERF_ARM_CCI_ONLINE);
+	perf_pmu_unregister(&g_cci_pmu->pmu);
+	g_cci_pmu = NULL;
+
+	return 0;
+}
+
+static struct platform_driver cci_pmu_driver = {
+	.driver = {
+		   .name = DRIVER_NAME,
+		   .of_match_table = arm_cci_pmu_matches,
+		   .suppress_bind_attrs = true,
+		  },
+	.probe = cci_pmu_probe,
+	.remove = cci_pmu_remove,
+};
+
+module_platform_driver(cci_pmu_driver);
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("ARM CCI PMU support");
diff --git a/drivers/perf/arm-ccn.c b/drivers/perf/arm-ccn.c
new file mode 100644
index 000000000..a0a71c1df
--- /dev/null
+++ b/drivers/perf/arm-ccn.c
@@ -0,0 +1,1583 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ *
+ * Copyright (C) 2014 ARM Limited
+ */
+
+#include <linux/ctype.h>
+#include <linux/hrtimer.h>
+#include <linux/idr.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/mod_devicetable.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+
+#define CCN_NUM_XP_PORTS 2
+#define CCN_NUM_VCS 4
+#define CCN_NUM_REGIONS	256
+#define CCN_REGION_SIZE	0x10000
+
+#define CCN_ALL_OLY_ID			0xff00
+#define CCN_ALL_OLY_ID__OLY_ID__SHIFT			0
+#define CCN_ALL_OLY_ID__OLY_ID__MASK			0x1f
+#define CCN_ALL_OLY_ID__NODE_ID__SHIFT			8
+#define CCN_ALL_OLY_ID__NODE_ID__MASK			0x3f
+
+#define CCN_MN_ERRINT_STATUS		0x0008
+#define CCN_MN_ERRINT_STATUS__INTREQ__DESSERT		0x11
+#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__ENABLE	0x02
+#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLED	0x20
+#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLE	0x22
+#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_ENABLE	0x04
+#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_DISABLED	0x40
+#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_DISABLE	0x44
+#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__ENABLE	0x08
+#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLED	0x80
+#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLE	0x88
+#define CCN_MN_OLY_COMP_LIST_63_0	0x01e0
+#define CCN_MN_ERR_SIG_VAL_63_0		0x0300
+#define CCN_MN_ERR_SIG_VAL_63_0__DT			(1 << 1)
+
+#define CCN_DT_ACTIVE_DSM		0x0000
+#define CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(n)		((n) * 8)
+#define CCN_DT_ACTIVE_DSM__DSM_ID__MASK			0xff
+#define CCN_DT_CTL			0x0028
+#define CCN_DT_CTL__DT_EN				(1 << 0)
+#define CCN_DT_PMEVCNT(n)		(0x0100 + (n) * 0x8)
+#define CCN_DT_PMCCNTR			0x0140
+#define CCN_DT_PMCCNTRSR		0x0190
+#define CCN_DT_PMOVSR			0x0198
+#define CCN_DT_PMOVSR_CLR		0x01a0
+#define CCN_DT_PMOVSR_CLR__MASK				0x1f
+#define CCN_DT_PMCR			0x01a8
+#define CCN_DT_PMCR__OVFL_INTR_EN			(1 << 6)
+#define CCN_DT_PMCR__PMU_EN				(1 << 0)
+#define CCN_DT_PMSR			0x01b0
+#define CCN_DT_PMSR_REQ			0x01b8
+#define CCN_DT_PMSR_CLR			0x01c0
+
+#define CCN_HNF_PMU_EVENT_SEL		0x0600
+#define CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(n)		((n) * 4)
+#define CCN_HNF_PMU_EVENT_SEL__ID__MASK			0xf
+
+#define CCN_XP_DT_CONFIG		0x0300
+#define CCN_XP_DT_CONFIG__DT_CFG__SHIFT(n)		((n) * 4)
+#define CCN_XP_DT_CONFIG__DT_CFG__MASK			0xf
+#define CCN_XP_DT_CONFIG__DT_CFG__PASS_THROUGH		0x0
+#define CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT_0_OR_1	0x1
+#define CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT(n)		(0x2 + (n))
+#define CCN_XP_DT_CONFIG__DT_CFG__XP_PMU_EVENT(n)	(0x4 + (n))
+#define CCN_XP_DT_CONFIG__DT_CFG__DEVICE_PMU_EVENT(d, n) (0x8 + (d) * 4 + (n))
+#define CCN_XP_DT_INTERFACE_SEL		0x0308
+#define CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(n)	(0 + (n) * 8)
+#define CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__MASK	0x1
+#define CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(n)	(1 + (n) * 8)
+#define CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__MASK	0x1
+#define CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(n)	(2 + (n) * 8)
+#define CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__MASK	0x3
+#define CCN_XP_DT_CMP_VAL_L(n)		(0x0310 + (n) * 0x40)
+#define CCN_XP_DT_CMP_VAL_H(n)		(0x0318 + (n) * 0x40)
+#define CCN_XP_DT_CMP_MASK_L(n)		(0x0320 + (n) * 0x40)
+#define CCN_XP_DT_CMP_MASK_H(n)		(0x0328 + (n) * 0x40)
+#define CCN_XP_DT_CONTROL		0x0370
+#define CCN_XP_DT_CONTROL__DT_ENABLE			(1 << 0)
+#define CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(n)		(12 + (n) * 4)
+#define CCN_XP_DT_CONTROL__WP_ARM_SEL__MASK		0xf
+#define CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS		0xf
+#define CCN_XP_PMU_EVENT_SEL		0x0600
+#define CCN_XP_PMU_EVENT_SEL__ID__SHIFT(n)		((n) * 7)
+#define CCN_XP_PMU_EVENT_SEL__ID__MASK			0x3f
+
+#define CCN_SBAS_PMU_EVENT_SEL		0x0600
+#define CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(n)		((n) * 4)
+#define CCN_SBAS_PMU_EVENT_SEL__ID__MASK		0xf
+
+#define CCN_RNI_PMU_EVENT_SEL		0x0600
+#define CCN_RNI_PMU_EVENT_SEL__ID__SHIFT(n)		((n) * 4)
+#define CCN_RNI_PMU_EVENT_SEL__ID__MASK			0xf
+
+#define CCN_TYPE_MN	0x01
+#define CCN_TYPE_DT	0x02
+#define CCN_TYPE_HNF	0x04
+#define CCN_TYPE_HNI	0x05
+#define CCN_TYPE_XP	0x08
+#define CCN_TYPE_SBSX	0x0c
+#define CCN_TYPE_SBAS	0x10
+#define CCN_TYPE_RNI_1P	0x14
+#define CCN_TYPE_RNI_2P	0x15
+#define CCN_TYPE_RNI_3P	0x16
+#define CCN_TYPE_RND_1P	0x18 /* RN-D = RN-I + DVM */
+#define CCN_TYPE_RND_2P	0x19
+#define CCN_TYPE_RND_3P	0x1a
+#define CCN_TYPE_CYCLES	0xff /* Pseudotype */
+
+#define CCN_EVENT_WATCHPOINT 0xfe /* Pseudoevent */
+
+#define CCN_NUM_PMU_EVENTS		4
+#define CCN_NUM_XP_WATCHPOINTS		2 /* See DT.dbg_id.num_watchpoints */
+#define CCN_NUM_PMU_EVENT_COUNTERS	8 /* See DT.dbg_id.num_pmucntr */
+#define CCN_IDX_PMU_CYCLE_COUNTER	CCN_NUM_PMU_EVENT_COUNTERS
+
+#define CCN_NUM_PREDEFINED_MASKS	4
+#define CCN_IDX_MASK_ANY		(CCN_NUM_PMU_EVENT_COUNTERS + 0)
+#define CCN_IDX_MASK_EXACT		(CCN_NUM_PMU_EVENT_COUNTERS + 1)
+#define CCN_IDX_MASK_ORDER		(CCN_NUM_PMU_EVENT_COUNTERS + 2)
+#define CCN_IDX_MASK_OPCODE		(CCN_NUM_PMU_EVENT_COUNTERS + 3)
+
+struct arm_ccn_component {
+	void __iomem *base;
+	u32 type;
+
+	DECLARE_BITMAP(pmu_events_mask, CCN_NUM_PMU_EVENTS);
+	union {
+		struct {
+			DECLARE_BITMAP(dt_cmp_mask, CCN_NUM_XP_WATCHPOINTS);
+		} xp;
+	};
+};
+
+#define pmu_to_arm_ccn(_pmu) container_of(container_of(_pmu, \
+	struct arm_ccn_dt, pmu), struct arm_ccn, dt)
+
+struct arm_ccn_dt {
+	int id;
+	void __iomem *base;
+
+	spinlock_t config_lock;
+
+	DECLARE_BITMAP(pmu_counters_mask, CCN_NUM_PMU_EVENT_COUNTERS + 1);
+	struct {
+		struct arm_ccn_component *source;
+		struct perf_event *event;
+	} pmu_counters[CCN_NUM_PMU_EVENT_COUNTERS + 1];
+
+	struct {
+	       u64 l, h;
+	} cmp_mask[CCN_NUM_PMU_EVENT_COUNTERS + CCN_NUM_PREDEFINED_MASKS];
+
+	struct hrtimer hrtimer;
+
+	unsigned int cpu;
+	struct hlist_node node;
+
+	struct pmu pmu;
+};
+
+struct arm_ccn {
+	struct device *dev;
+	void __iomem *base;
+	unsigned int irq;
+
+	unsigned sbas_present:1;
+	unsigned sbsx_present:1;
+
+	int num_nodes;
+	struct arm_ccn_component *node;
+
+	int num_xps;
+	struct arm_ccn_component *xp;
+
+	struct arm_ccn_dt dt;
+	int mn_id;
+};
+
+static int arm_ccn_node_to_xp(int node)
+{
+	return node / CCN_NUM_XP_PORTS;
+}
+
+static int arm_ccn_node_to_xp_port(int node)
+{
+	return node % CCN_NUM_XP_PORTS;
+}
+
+
+/*
+ * Bit shifts and masks in these defines must be kept in sync with
+ * arm_ccn_pmu_config_set() and CCN_FORMAT_ATTRs below!
+ */
+#define CCN_CONFIG_NODE(_config)	(((_config) >> 0) & 0xff)
+#define CCN_CONFIG_XP(_config)		(((_config) >> 0) & 0xff)
+#define CCN_CONFIG_TYPE(_config)	(((_config) >> 8) & 0xff)
+#define CCN_CONFIG_EVENT(_config)	(((_config) >> 16) & 0xff)
+#define CCN_CONFIG_PORT(_config)	(((_config) >> 24) & 0x3)
+#define CCN_CONFIG_BUS(_config)		(((_config) >> 24) & 0x3)
+#define CCN_CONFIG_VC(_config)		(((_config) >> 26) & 0x7)
+#define CCN_CONFIG_DIR(_config)		(((_config) >> 29) & 0x1)
+#define CCN_CONFIG_MASK(_config)	(((_config) >> 30) & 0xf)
+
+static void arm_ccn_pmu_config_set(u64 *config, u32 node_xp, u32 type, u32 port)
+{
+	*config &= ~((0xff << 0) | (0xff << 8) | (0x3 << 24));
+	*config |= (node_xp << 0) | (type << 8) | (port << 24);
+}
+
+static ssize_t arm_ccn_pmu_format_show(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *ea = container_of(attr,
+			struct dev_ext_attribute, attr);
+
+	return snprintf(buf, PAGE_SIZE, "%s\n", (char *)ea->var);
+}
+
+#define CCN_FORMAT_ATTR(_name, _config) \
+	struct dev_ext_attribute arm_ccn_pmu_format_attr_##_name = \
+			{ __ATTR(_name, S_IRUGO, arm_ccn_pmu_format_show, \
+			NULL), _config }
+
+static CCN_FORMAT_ATTR(node, "config:0-7");
+static CCN_FORMAT_ATTR(xp, "config:0-7");
+static CCN_FORMAT_ATTR(type, "config:8-15");
+static CCN_FORMAT_ATTR(event, "config:16-23");
+static CCN_FORMAT_ATTR(port, "config:24-25");
+static CCN_FORMAT_ATTR(bus, "config:24-25");
+static CCN_FORMAT_ATTR(vc, "config:26-28");
+static CCN_FORMAT_ATTR(dir, "config:29-29");
+static CCN_FORMAT_ATTR(mask, "config:30-33");
+static CCN_FORMAT_ATTR(cmp_l, "config1:0-62");
+static CCN_FORMAT_ATTR(cmp_h, "config2:0-59");
+
+static struct attribute *arm_ccn_pmu_format_attrs[] = {
+	&arm_ccn_pmu_format_attr_node.attr.attr,
+	&arm_ccn_pmu_format_attr_xp.attr.attr,
+	&arm_ccn_pmu_format_attr_type.attr.attr,
+	&arm_ccn_pmu_format_attr_event.attr.attr,
+	&arm_ccn_pmu_format_attr_port.attr.attr,
+	&arm_ccn_pmu_format_attr_bus.attr.attr,
+	&arm_ccn_pmu_format_attr_vc.attr.attr,
+	&arm_ccn_pmu_format_attr_dir.attr.attr,
+	&arm_ccn_pmu_format_attr_mask.attr.attr,
+	&arm_ccn_pmu_format_attr_cmp_l.attr.attr,
+	&arm_ccn_pmu_format_attr_cmp_h.attr.attr,
+	NULL
+};
+
+static const struct attribute_group arm_ccn_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = arm_ccn_pmu_format_attrs,
+};
+
+
+struct arm_ccn_pmu_event {
+	struct device_attribute attr;
+	u32 type;
+	u32 event;
+	int num_ports;
+	int num_vcs;
+	const char *def;
+	int mask;
+};
+
+#define CCN_EVENT_ATTR(_name) \
+	__ATTR(_name, S_IRUGO, arm_ccn_pmu_event_show, NULL)
+
+/*
+ * Events defined in TRM for MN, HN-I and SBSX are actually watchpoints set on
+ * their ports in XP they are connected to. For the sake of usability they are
+ * explicitly defined here (and translated into a relevant watchpoint in
+ * arm_ccn_pmu_event_init()) so the user can easily request them without deep
+ * knowledge of the flit format.
+ */
+
+#define CCN_EVENT_MN(_name, _def, _mask) { .attr = CCN_EVENT_ATTR(mn_##_name), \
+		.type = CCN_TYPE_MN, .event = CCN_EVENT_WATCHPOINT, \
+		.num_ports = CCN_NUM_XP_PORTS, .num_vcs = CCN_NUM_VCS, \
+		.def = _def, .mask = _mask, }
+
+#define CCN_EVENT_HNI(_name, _def, _mask) { \
+		.attr = CCN_EVENT_ATTR(hni_##_name), .type = CCN_TYPE_HNI, \
+		.event = CCN_EVENT_WATCHPOINT, .num_ports = CCN_NUM_XP_PORTS, \
+		.num_vcs = CCN_NUM_VCS, .def = _def, .mask = _mask, }
+
+#define CCN_EVENT_SBSX(_name, _def, _mask) { \
+		.attr = CCN_EVENT_ATTR(sbsx_##_name), .type = CCN_TYPE_SBSX, \
+		.event = CCN_EVENT_WATCHPOINT, .num_ports = CCN_NUM_XP_PORTS, \
+		.num_vcs = CCN_NUM_VCS, .def = _def, .mask = _mask, }
+
+#define CCN_EVENT_HNF(_name, _event) { .attr = CCN_EVENT_ATTR(hnf_##_name), \
+		.type = CCN_TYPE_HNF, .event = _event, }
+
+#define CCN_EVENT_XP(_name, _event) { .attr = CCN_EVENT_ATTR(xp_##_name), \
+		.type = CCN_TYPE_XP, .event = _event, \
+		.num_ports = CCN_NUM_XP_PORTS, .num_vcs = CCN_NUM_VCS, }
+
+/*
+ * RN-I & RN-D (RN-D = RN-I + DVM) nodes have different type ID depending
+ * on configuration. One of them is picked to represent the whole group,
+ * as they all share the same event types.
+ */
+#define CCN_EVENT_RNI(_name, _event) { .attr = CCN_EVENT_ATTR(rni_##_name), \
+		.type = CCN_TYPE_RNI_3P, .event = _event, }
+
+#define CCN_EVENT_SBAS(_name, _event) { .attr = CCN_EVENT_ATTR(sbas_##_name), \
+		.type = CCN_TYPE_SBAS, .event = _event, }
+
+#define CCN_EVENT_CYCLES(_name) { .attr = CCN_EVENT_ATTR(_name), \
+		.type = CCN_TYPE_CYCLES }
+
+
+static ssize_t arm_ccn_pmu_event_show(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev));
+	struct arm_ccn_pmu_event *event = container_of(attr,
+			struct arm_ccn_pmu_event, attr);
+	ssize_t res;
+
+	res = scnprintf(buf, PAGE_SIZE, "type=0x%x", event->type);
+	if (event->event)
+		res += scnprintf(buf + res, PAGE_SIZE - res, ",event=0x%x",
+				event->event);
+	if (event->def)
+		res += scnprintf(buf + res, PAGE_SIZE - res, ",%s",
+				event->def);
+	if (event->mask)
+		res += scnprintf(buf + res, PAGE_SIZE - res, ",mask=0x%x",
+				event->mask);
+
+	/* Arguments required by an event */
+	switch (event->type) {
+	case CCN_TYPE_CYCLES:
+		break;
+	case CCN_TYPE_XP:
+		res += scnprintf(buf + res, PAGE_SIZE - res,
+				",xp=?,vc=?");
+		if (event->event == CCN_EVENT_WATCHPOINT)
+			res += scnprintf(buf + res, PAGE_SIZE - res,
+					",port=?,dir=?,cmp_l=?,cmp_h=?,mask=?");
+		else
+			res += scnprintf(buf + res, PAGE_SIZE - res,
+					",bus=?");
+
+		break;
+	case CCN_TYPE_MN:
+		res += scnprintf(buf + res, PAGE_SIZE - res, ",node=%d", ccn->mn_id);
+		break;
+	default:
+		res += scnprintf(buf + res, PAGE_SIZE - res, ",node=?");
+		break;
+	}
+
+	res += scnprintf(buf + res, PAGE_SIZE - res, "\n");
+
+	return res;
+}
+
+static umode_t arm_ccn_pmu_events_is_visible(struct kobject *kobj,
+				     struct attribute *attr, int index)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev));
+	struct device_attribute *dev_attr = container_of(attr,
+			struct device_attribute, attr);
+	struct arm_ccn_pmu_event *event = container_of(dev_attr,
+			struct arm_ccn_pmu_event, attr);
+
+	if (event->type == CCN_TYPE_SBAS && !ccn->sbas_present)
+		return 0;
+	if (event->type == CCN_TYPE_SBSX && !ccn->sbsx_present)
+		return 0;
+
+	return attr->mode;
+}
+
+static struct arm_ccn_pmu_event arm_ccn_pmu_events[] = {
+	CCN_EVENT_MN(eobarrier, "dir=1,vc=0,cmp_h=0x1c00", CCN_IDX_MASK_OPCODE),
+	CCN_EVENT_MN(ecbarrier, "dir=1,vc=0,cmp_h=0x1e00", CCN_IDX_MASK_OPCODE),
+	CCN_EVENT_MN(dvmop, "dir=1,vc=0,cmp_h=0x2800", CCN_IDX_MASK_OPCODE),
+	CCN_EVENT_HNI(txdatflits, "dir=1,vc=3", CCN_IDX_MASK_ANY),
+	CCN_EVENT_HNI(rxdatflits, "dir=0,vc=3", CCN_IDX_MASK_ANY),
+	CCN_EVENT_HNI(txreqflits, "dir=1,vc=0", CCN_IDX_MASK_ANY),
+	CCN_EVENT_HNI(rxreqflits, "dir=0,vc=0", CCN_IDX_MASK_ANY),
+	CCN_EVENT_HNI(rxreqflits_order, "dir=0,vc=0,cmp_h=0x8000",
+			CCN_IDX_MASK_ORDER),
+	CCN_EVENT_SBSX(txdatflits, "dir=1,vc=3", CCN_IDX_MASK_ANY),
+	CCN_EVENT_SBSX(rxdatflits, "dir=0,vc=3", CCN_IDX_MASK_ANY),
+	CCN_EVENT_SBSX(txreqflits, "dir=1,vc=0", CCN_IDX_MASK_ANY),
+	CCN_EVENT_SBSX(rxreqflits, "dir=0,vc=0", CCN_IDX_MASK_ANY),
+	CCN_EVENT_SBSX(rxreqflits_order, "dir=0,vc=0,cmp_h=0x8000",
+			CCN_IDX_MASK_ORDER),
+	CCN_EVENT_HNF(cache_miss, 0x1),
+	CCN_EVENT_HNF(l3_sf_cache_access, 0x02),
+	CCN_EVENT_HNF(cache_fill, 0x3),
+	CCN_EVENT_HNF(pocq_retry, 0x4),
+	CCN_EVENT_HNF(pocq_reqs_recvd, 0x5),
+	CCN_EVENT_HNF(sf_hit, 0x6),
+	CCN_EVENT_HNF(sf_evictions, 0x7),
+	CCN_EVENT_HNF(snoops_sent, 0x8),
+	CCN_EVENT_HNF(snoops_broadcast, 0x9),
+	CCN_EVENT_HNF(l3_eviction, 0xa),
+	CCN_EVENT_HNF(l3_fill_invalid_way, 0xb),
+	CCN_EVENT_HNF(mc_retries, 0xc),
+	CCN_EVENT_HNF(mc_reqs, 0xd),
+	CCN_EVENT_HNF(qos_hh_retry, 0xe),
+	CCN_EVENT_RNI(rdata_beats_p0, 0x1),
+	CCN_EVENT_RNI(rdata_beats_p1, 0x2),
+	CCN_EVENT_RNI(rdata_beats_p2, 0x3),
+	CCN_EVENT_RNI(rxdat_flits, 0x4),
+	CCN_EVENT_RNI(txdat_flits, 0x5),
+	CCN_EVENT_RNI(txreq_flits, 0x6),
+	CCN_EVENT_RNI(txreq_flits_retried, 0x7),
+	CCN_EVENT_RNI(rrt_full, 0x8),
+	CCN_EVENT_RNI(wrt_full, 0x9),
+	CCN_EVENT_RNI(txreq_flits_replayed, 0xa),
+	CCN_EVENT_XP(upload_starvation, 0x1),
+	CCN_EVENT_XP(download_starvation, 0x2),
+	CCN_EVENT_XP(respin, 0x3),
+	CCN_EVENT_XP(valid_flit, 0x4),
+	CCN_EVENT_XP(watchpoint, CCN_EVENT_WATCHPOINT),
+	CCN_EVENT_SBAS(rdata_beats_p0, 0x1),
+	CCN_EVENT_SBAS(rxdat_flits, 0x4),
+	CCN_EVENT_SBAS(txdat_flits, 0x5),
+	CCN_EVENT_SBAS(txreq_flits, 0x6),
+	CCN_EVENT_SBAS(txreq_flits_retried, 0x7),
+	CCN_EVENT_SBAS(rrt_full, 0x8),
+	CCN_EVENT_SBAS(wrt_full, 0x9),
+	CCN_EVENT_SBAS(txreq_flits_replayed, 0xa),
+	CCN_EVENT_CYCLES(cycles),
+};
+
+/* Populated in arm_ccn_init() */
+static struct attribute
+		*arm_ccn_pmu_events_attrs[ARRAY_SIZE(arm_ccn_pmu_events) + 1];
+
+static const struct attribute_group arm_ccn_pmu_events_attr_group = {
+	.name = "events",
+	.is_visible = arm_ccn_pmu_events_is_visible,
+	.attrs = arm_ccn_pmu_events_attrs,
+};
+
+
+static u64 *arm_ccn_pmu_get_cmp_mask(struct arm_ccn *ccn, const char *name)
+{
+	unsigned long i;
+
+	if (WARN_ON(!name || !name[0] || !isxdigit(name[0]) || !name[1]))
+		return NULL;
+	i = isdigit(name[0]) ? name[0] - '0' : 0xa + tolower(name[0]) - 'a';
+
+	switch (name[1]) {
+	case 'l':
+		return &ccn->dt.cmp_mask[i].l;
+	case 'h':
+		return &ccn->dt.cmp_mask[i].h;
+	default:
+		return NULL;
+	}
+}
+
+static ssize_t arm_ccn_pmu_cmp_mask_show(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev));
+	u64 *mask = arm_ccn_pmu_get_cmp_mask(ccn, attr->attr.name);
+
+	return mask ? snprintf(buf, PAGE_SIZE, "0x%016llx\n", *mask) : -EINVAL;
+}
+
+static ssize_t arm_ccn_pmu_cmp_mask_store(struct device *dev,
+		struct device_attribute *attr, const char *buf, size_t count)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev));
+	u64 *mask = arm_ccn_pmu_get_cmp_mask(ccn, attr->attr.name);
+	int err = -EINVAL;
+
+	if (mask)
+		err = kstrtoull(buf, 0, mask);
+
+	return err ? err : count;
+}
+
+#define CCN_CMP_MASK_ATTR(_name) \
+	struct device_attribute arm_ccn_pmu_cmp_mask_attr_##_name = \
+			__ATTR(_name, S_IRUGO | S_IWUSR, \
+			arm_ccn_pmu_cmp_mask_show, arm_ccn_pmu_cmp_mask_store)
+
+#define CCN_CMP_MASK_ATTR_RO(_name) \
+	struct device_attribute arm_ccn_pmu_cmp_mask_attr_##_name = \
+			__ATTR(_name, S_IRUGO, arm_ccn_pmu_cmp_mask_show, NULL)
+
+static CCN_CMP_MASK_ATTR(0l);
+static CCN_CMP_MASK_ATTR(0h);
+static CCN_CMP_MASK_ATTR(1l);
+static CCN_CMP_MASK_ATTR(1h);
+static CCN_CMP_MASK_ATTR(2l);
+static CCN_CMP_MASK_ATTR(2h);
+static CCN_CMP_MASK_ATTR(3l);
+static CCN_CMP_MASK_ATTR(3h);
+static CCN_CMP_MASK_ATTR(4l);
+static CCN_CMP_MASK_ATTR(4h);
+static CCN_CMP_MASK_ATTR(5l);
+static CCN_CMP_MASK_ATTR(5h);
+static CCN_CMP_MASK_ATTR(6l);
+static CCN_CMP_MASK_ATTR(6h);
+static CCN_CMP_MASK_ATTR(7l);
+static CCN_CMP_MASK_ATTR(7h);
+static CCN_CMP_MASK_ATTR_RO(8l);
+static CCN_CMP_MASK_ATTR_RO(8h);
+static CCN_CMP_MASK_ATTR_RO(9l);
+static CCN_CMP_MASK_ATTR_RO(9h);
+static CCN_CMP_MASK_ATTR_RO(al);
+static CCN_CMP_MASK_ATTR_RO(ah);
+static CCN_CMP_MASK_ATTR_RO(bl);
+static CCN_CMP_MASK_ATTR_RO(bh);
+
+static struct attribute *arm_ccn_pmu_cmp_mask_attrs[] = {
+	&arm_ccn_pmu_cmp_mask_attr_0l.attr, &arm_ccn_pmu_cmp_mask_attr_0h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_1l.attr, &arm_ccn_pmu_cmp_mask_attr_1h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_2l.attr, &arm_ccn_pmu_cmp_mask_attr_2h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_3l.attr, &arm_ccn_pmu_cmp_mask_attr_3h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_4l.attr, &arm_ccn_pmu_cmp_mask_attr_4h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_5l.attr, &arm_ccn_pmu_cmp_mask_attr_5h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_6l.attr, &arm_ccn_pmu_cmp_mask_attr_6h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_7l.attr, &arm_ccn_pmu_cmp_mask_attr_7h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_8l.attr, &arm_ccn_pmu_cmp_mask_attr_8h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_9l.attr, &arm_ccn_pmu_cmp_mask_attr_9h.attr,
+	&arm_ccn_pmu_cmp_mask_attr_al.attr, &arm_ccn_pmu_cmp_mask_attr_ah.attr,
+	&arm_ccn_pmu_cmp_mask_attr_bl.attr, &arm_ccn_pmu_cmp_mask_attr_bh.attr,
+	NULL
+};
+
+static const struct attribute_group arm_ccn_pmu_cmp_mask_attr_group = {
+	.name = "cmp_mask",
+	.attrs = arm_ccn_pmu_cmp_mask_attrs,
+};
+
+static ssize_t arm_ccn_pmu_cpumask_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(ccn->dt.cpu));
+}
+
+static struct device_attribute arm_ccn_pmu_cpumask_attr =
+		__ATTR(cpumask, S_IRUGO, arm_ccn_pmu_cpumask_show, NULL);
+
+static struct attribute *arm_ccn_pmu_cpumask_attrs[] = {
+	&arm_ccn_pmu_cpumask_attr.attr,
+	NULL,
+};
+
+static const struct attribute_group arm_ccn_pmu_cpumask_attr_group = {
+	.attrs = arm_ccn_pmu_cpumask_attrs,
+};
+
+/*
+ * Default poll period is 10ms, which is way over the top anyway,
+ * as in the worst case scenario (an event every cycle), with 1GHz
+ * clocked bus, the smallest, 32 bit counter will overflow in
+ * more than 4s.
+ */
+static unsigned int arm_ccn_pmu_poll_period_us = 10000;
+module_param_named(pmu_poll_period_us, arm_ccn_pmu_poll_period_us, uint,
+		S_IRUGO | S_IWUSR);
+
+static ktime_t arm_ccn_pmu_timer_period(void)
+{
+	return ns_to_ktime((u64)arm_ccn_pmu_poll_period_us * 1000);
+}
+
+
+static const struct attribute_group *arm_ccn_pmu_attr_groups[] = {
+	&arm_ccn_pmu_events_attr_group,
+	&arm_ccn_pmu_format_attr_group,
+	&arm_ccn_pmu_cmp_mask_attr_group,
+	&arm_ccn_pmu_cpumask_attr_group,
+	NULL
+};
+
+
+static int arm_ccn_pmu_alloc_bit(unsigned long *bitmap, unsigned long size)
+{
+	int bit;
+
+	do {
+		bit = find_first_zero_bit(bitmap, size);
+		if (bit >= size)
+			return -EAGAIN;
+	} while (test_and_set_bit(bit, bitmap));
+
+	return bit;
+}
+
+/* All RN-I and RN-D nodes have identical PMUs */
+static int arm_ccn_pmu_type_eq(u32 a, u32 b)
+{
+	if (a == b)
+		return 1;
+
+	switch (a) {
+	case CCN_TYPE_RNI_1P:
+	case CCN_TYPE_RNI_2P:
+	case CCN_TYPE_RNI_3P:
+	case CCN_TYPE_RND_1P:
+	case CCN_TYPE_RND_2P:
+	case CCN_TYPE_RND_3P:
+		switch (b) {
+		case CCN_TYPE_RNI_1P:
+		case CCN_TYPE_RNI_2P:
+		case CCN_TYPE_RNI_3P:
+		case CCN_TYPE_RND_1P:
+		case CCN_TYPE_RND_2P:
+		case CCN_TYPE_RND_3P:
+			return 1;
+		}
+		break;
+	}
+
+	return 0;
+}
+
+static int arm_ccn_pmu_event_alloc(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	u32 node_xp, type, event_id;
+	struct arm_ccn_component *source;
+	int bit;
+
+	node_xp = CCN_CONFIG_NODE(event->attr.config);
+	type = CCN_CONFIG_TYPE(event->attr.config);
+	event_id = CCN_CONFIG_EVENT(event->attr.config);
+
+	/* Allocate the cycle counter */
+	if (type == CCN_TYPE_CYCLES) {
+		if (test_and_set_bit(CCN_IDX_PMU_CYCLE_COUNTER,
+				ccn->dt.pmu_counters_mask))
+			return -EAGAIN;
+
+		hw->idx = CCN_IDX_PMU_CYCLE_COUNTER;
+		ccn->dt.pmu_counters[CCN_IDX_PMU_CYCLE_COUNTER].event = event;
+
+		return 0;
+	}
+
+	/* Allocate an event counter */
+	hw->idx = arm_ccn_pmu_alloc_bit(ccn->dt.pmu_counters_mask,
+			CCN_NUM_PMU_EVENT_COUNTERS);
+	if (hw->idx < 0) {
+		dev_dbg(ccn->dev, "No more counters available!\n");
+		return -EAGAIN;
+	}
+
+	if (type == CCN_TYPE_XP)
+		source = &ccn->xp[node_xp];
+	else
+		source = &ccn->node[node_xp];
+	ccn->dt.pmu_counters[hw->idx].source = source;
+
+	/* Allocate an event source or a watchpoint */
+	if (type == CCN_TYPE_XP && event_id == CCN_EVENT_WATCHPOINT)
+		bit = arm_ccn_pmu_alloc_bit(source->xp.dt_cmp_mask,
+				CCN_NUM_XP_WATCHPOINTS);
+	else
+		bit = arm_ccn_pmu_alloc_bit(source->pmu_events_mask,
+				CCN_NUM_PMU_EVENTS);
+	if (bit < 0) {
+		dev_dbg(ccn->dev, "No more event sources/watchpoints on node/XP %d!\n",
+				node_xp);
+		clear_bit(hw->idx, ccn->dt.pmu_counters_mask);
+		return -EAGAIN;
+	}
+	hw->config_base = bit;
+
+	ccn->dt.pmu_counters[hw->idx].event = event;
+
+	return 0;
+}
+
+static void arm_ccn_pmu_event_release(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+
+	if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER) {
+		clear_bit(CCN_IDX_PMU_CYCLE_COUNTER, ccn->dt.pmu_counters_mask);
+	} else {
+		struct arm_ccn_component *source =
+				ccn->dt.pmu_counters[hw->idx].source;
+
+		if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP &&
+				CCN_CONFIG_EVENT(event->attr.config) ==
+				CCN_EVENT_WATCHPOINT)
+			clear_bit(hw->config_base, source->xp.dt_cmp_mask);
+		else
+			clear_bit(hw->config_base, source->pmu_events_mask);
+		clear_bit(hw->idx, ccn->dt.pmu_counters_mask);
+	}
+
+	ccn->dt.pmu_counters[hw->idx].source = NULL;
+	ccn->dt.pmu_counters[hw->idx].event = NULL;
+}
+
+static int arm_ccn_pmu_event_init(struct perf_event *event)
+{
+	struct arm_ccn *ccn;
+	struct hw_perf_event *hw = &event->hw;
+	u32 node_xp, type, event_id;
+	int valid;
+	int i;
+	struct perf_event *sibling;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	ccn = pmu_to_arm_ccn(event->pmu);
+
+	if (hw->sample_period) {
+		dev_dbg(ccn->dev, "Sampling not supported!\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (has_branch_stack(event)) {
+		dev_dbg(ccn->dev, "Can't exclude execution levels!\n");
+		return -EINVAL;
+	}
+
+	if (event->cpu < 0) {
+		dev_dbg(ccn->dev, "Can't provide per-task data!\n");
+		return -EOPNOTSUPP;
+	}
+	/*
+	 * Many perf core operations (eg. events rotation) operate on a
+	 * single CPU context. This is obvious for CPU PMUs, where one
+	 * expects the same sets of events being observed on all CPUs,
+	 * but can lead to issues for off-core PMUs, like CCN, where each
+	 * event could be theoretically assigned to a different CPU. To
+	 * mitigate this, we enforce CPU assignment to one, selected
+	 * processor (the one described in the "cpumask" attribute).
+	 */
+	event->cpu = ccn->dt.cpu;
+
+	node_xp = CCN_CONFIG_NODE(event->attr.config);
+	type = CCN_CONFIG_TYPE(event->attr.config);
+	event_id = CCN_CONFIG_EVENT(event->attr.config);
+
+	/* Validate node/xp vs topology */
+	switch (type) {
+	case CCN_TYPE_MN:
+		if (node_xp != ccn->mn_id) {
+			dev_dbg(ccn->dev, "Invalid MN ID %d!\n", node_xp);
+			return -EINVAL;
+		}
+		break;
+	case CCN_TYPE_XP:
+		if (node_xp >= ccn->num_xps) {
+			dev_dbg(ccn->dev, "Invalid XP ID %d!\n", node_xp);
+			return -EINVAL;
+		}
+		break;
+	case CCN_TYPE_CYCLES:
+		break;
+	default:
+		if (node_xp >= ccn->num_nodes) {
+			dev_dbg(ccn->dev, "Invalid node ID %d!\n", node_xp);
+			return -EINVAL;
+		}
+		if (!arm_ccn_pmu_type_eq(type, ccn->node[node_xp].type)) {
+			dev_dbg(ccn->dev, "Invalid type 0x%x for node %d!\n",
+					type, node_xp);
+			return -EINVAL;
+		}
+		break;
+	}
+
+	/* Validate event ID vs available for the type */
+	for (i = 0, valid = 0; i < ARRAY_SIZE(arm_ccn_pmu_events) && !valid;
+			i++) {
+		struct arm_ccn_pmu_event *e = &arm_ccn_pmu_events[i];
+		u32 port = CCN_CONFIG_PORT(event->attr.config);
+		u32 vc = CCN_CONFIG_VC(event->attr.config);
+
+		if (!arm_ccn_pmu_type_eq(type, e->type))
+			continue;
+		if (event_id != e->event)
+			continue;
+		if (e->num_ports && port >= e->num_ports) {
+			dev_dbg(ccn->dev, "Invalid port %d for node/XP %d!\n",
+					port, node_xp);
+			return -EINVAL;
+		}
+		if (e->num_vcs && vc >= e->num_vcs) {
+			dev_dbg(ccn->dev, "Invalid vc %d for node/XP %d!\n",
+					vc, node_xp);
+			return -EINVAL;
+		}
+		valid = 1;
+	}
+	if (!valid) {
+		dev_dbg(ccn->dev, "Invalid event 0x%x for node/XP %d!\n",
+				event_id, node_xp);
+		return -EINVAL;
+	}
+
+	/* Watchpoint-based event for a node is actually set on XP */
+	if (event_id == CCN_EVENT_WATCHPOINT && type != CCN_TYPE_XP) {
+		u32 port;
+
+		type = CCN_TYPE_XP;
+		port = arm_ccn_node_to_xp_port(node_xp);
+		node_xp = arm_ccn_node_to_xp(node_xp);
+
+		arm_ccn_pmu_config_set(&event->attr.config,
+				node_xp, type, port);
+	}
+
+	/*
+	 * We must NOT create groups containing mixed PMUs, although software
+	 * events are acceptable (for example to create a CCN group
+	 * periodically read when a hrtimer aka cpu-clock leader triggers).
+	 */
+	if (event->group_leader->pmu != event->pmu &&
+			!is_software_event(event->group_leader))
+		return -EINVAL;
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (sibling->pmu != event->pmu &&
+				!is_software_event(sibling))
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+static u64 arm_ccn_pmu_read_counter(struct arm_ccn *ccn, int idx)
+{
+	u64 res;
+
+	if (idx == CCN_IDX_PMU_CYCLE_COUNTER) {
+#ifdef readq
+		res = readq(ccn->dt.base + CCN_DT_PMCCNTR);
+#else
+		/* 40 bit counter, can do snapshot and read in two parts */
+		writel(0x1, ccn->dt.base + CCN_DT_PMSR_REQ);
+		while (!(readl(ccn->dt.base + CCN_DT_PMSR) & 0x1))
+			;
+		writel(0x1, ccn->dt.base + CCN_DT_PMSR_CLR);
+		res = readl(ccn->dt.base + CCN_DT_PMCCNTRSR + 4) & 0xff;
+		res <<= 32;
+		res |= readl(ccn->dt.base + CCN_DT_PMCCNTRSR);
+#endif
+	} else {
+		res = readl(ccn->dt.base + CCN_DT_PMEVCNT(idx));
+	}
+
+	return res;
+}
+
+static void arm_ccn_pmu_event_update(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	u64 prev_count, new_count, mask;
+
+	do {
+		prev_count = local64_read(&hw->prev_count);
+		new_count = arm_ccn_pmu_read_counter(ccn, hw->idx);
+	} while (local64_xchg(&hw->prev_count, new_count) != prev_count);
+
+	mask = (1LLU << (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER ? 40 : 32)) - 1;
+
+	local64_add((new_count - prev_count) & mask, &event->count);
+}
+
+static void arm_ccn_pmu_xp_dt_config(struct perf_event *event, int enable)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	struct arm_ccn_component *xp;
+	u32 val, dt_cfg;
+
+	/* Nothing to do for cycle counter */
+	if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER)
+		return;
+
+	if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP)
+		xp = &ccn->xp[CCN_CONFIG_XP(event->attr.config)];
+	else
+		xp = &ccn->xp[arm_ccn_node_to_xp(
+				CCN_CONFIG_NODE(event->attr.config))];
+
+	if (enable)
+		dt_cfg = hw->event_base;
+	else
+		dt_cfg = CCN_XP_DT_CONFIG__DT_CFG__PASS_THROUGH;
+
+	spin_lock(&ccn->dt.config_lock);
+
+	val = readl(xp->base + CCN_XP_DT_CONFIG);
+	val &= ~(CCN_XP_DT_CONFIG__DT_CFG__MASK <<
+			CCN_XP_DT_CONFIG__DT_CFG__SHIFT(hw->idx));
+	val |= dt_cfg << CCN_XP_DT_CONFIG__DT_CFG__SHIFT(hw->idx);
+	writel(val, xp->base + CCN_XP_DT_CONFIG);
+
+	spin_unlock(&ccn->dt.config_lock);
+}
+
+static void arm_ccn_pmu_event_start(struct perf_event *event, int flags)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+
+	local64_set(&event->hw.prev_count,
+			arm_ccn_pmu_read_counter(ccn, hw->idx));
+	hw->state = 0;
+
+	/* Set the DT bus input, engaging the counter */
+	arm_ccn_pmu_xp_dt_config(event, 1);
+}
+
+static void arm_ccn_pmu_event_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hw = &event->hw;
+
+	/* Disable counting, setting the DT bus to pass-through mode */
+	arm_ccn_pmu_xp_dt_config(event, 0);
+
+	if (flags & PERF_EF_UPDATE)
+		arm_ccn_pmu_event_update(event);
+
+	hw->state |= PERF_HES_STOPPED;
+}
+
+static void arm_ccn_pmu_xp_watchpoint_config(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	struct arm_ccn_component *source =
+			ccn->dt.pmu_counters[hw->idx].source;
+	unsigned long wp = hw->config_base;
+	u32 val;
+	u64 cmp_l = event->attr.config1;
+	u64 cmp_h = event->attr.config2;
+	u64 mask_l = ccn->dt.cmp_mask[CCN_CONFIG_MASK(event->attr.config)].l;
+	u64 mask_h = ccn->dt.cmp_mask[CCN_CONFIG_MASK(event->attr.config)].h;
+
+	hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT(wp);
+
+	/* Direction (RX/TX), device (port) & virtual channel */
+	val = readl(source->base + CCN_XP_DT_INTERFACE_SEL);
+	val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__MASK <<
+			CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(wp));
+	val |= CCN_CONFIG_DIR(event->attr.config) <<
+			CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(wp);
+	val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__MASK <<
+			CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(wp));
+	val |= CCN_CONFIG_PORT(event->attr.config) <<
+			CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(wp);
+	val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__MASK <<
+			CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(wp));
+	val |= CCN_CONFIG_VC(event->attr.config) <<
+			CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(wp);
+	writel(val, source->base + CCN_XP_DT_INTERFACE_SEL);
+
+	/* Comparison values */
+	writel(cmp_l & 0xffffffff, source->base + CCN_XP_DT_CMP_VAL_L(wp));
+	writel((cmp_l >> 32) & 0x7fffffff,
+			source->base + CCN_XP_DT_CMP_VAL_L(wp) + 4);
+	writel(cmp_h & 0xffffffff, source->base + CCN_XP_DT_CMP_VAL_H(wp));
+	writel((cmp_h >> 32) & 0x0fffffff,
+			source->base + CCN_XP_DT_CMP_VAL_H(wp) + 4);
+
+	/* Mask */
+	writel(mask_l & 0xffffffff, source->base + CCN_XP_DT_CMP_MASK_L(wp));
+	writel((mask_l >> 32) & 0x7fffffff,
+			source->base + CCN_XP_DT_CMP_MASK_L(wp) + 4);
+	writel(mask_h & 0xffffffff, source->base + CCN_XP_DT_CMP_MASK_H(wp));
+	writel((mask_h >> 32) & 0x0fffffff,
+			source->base + CCN_XP_DT_CMP_MASK_H(wp) + 4);
+}
+
+static void arm_ccn_pmu_xp_event_config(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	struct arm_ccn_component *source =
+			ccn->dt.pmu_counters[hw->idx].source;
+	u32 val, id;
+
+	hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__XP_PMU_EVENT(hw->config_base);
+
+	id = (CCN_CONFIG_VC(event->attr.config) << 4) |
+			(CCN_CONFIG_BUS(event->attr.config) << 3) |
+			(CCN_CONFIG_EVENT(event->attr.config) << 0);
+
+	val = readl(source->base + CCN_XP_PMU_EVENT_SEL);
+	val &= ~(CCN_XP_PMU_EVENT_SEL__ID__MASK <<
+			CCN_XP_PMU_EVENT_SEL__ID__SHIFT(hw->config_base));
+	val |= id << CCN_XP_PMU_EVENT_SEL__ID__SHIFT(hw->config_base);
+	writel(val, source->base + CCN_XP_PMU_EVENT_SEL);
+}
+
+static void arm_ccn_pmu_node_event_config(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	struct arm_ccn_component *source =
+			ccn->dt.pmu_counters[hw->idx].source;
+	u32 type = CCN_CONFIG_TYPE(event->attr.config);
+	u32 val, port;
+
+	port = arm_ccn_node_to_xp_port(CCN_CONFIG_NODE(event->attr.config));
+	hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__DEVICE_PMU_EVENT(port,
+			hw->config_base);
+
+	/* These *_event_sel regs should be identical, but let's make sure... */
+	BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL != CCN_SBAS_PMU_EVENT_SEL);
+	BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL != CCN_RNI_PMU_EVENT_SEL);
+	BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(1) !=
+			CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(1));
+	BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(1) !=
+			CCN_RNI_PMU_EVENT_SEL__ID__SHIFT(1));
+	BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL__ID__MASK !=
+			CCN_SBAS_PMU_EVENT_SEL__ID__MASK);
+	BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL__ID__MASK !=
+			CCN_RNI_PMU_EVENT_SEL__ID__MASK);
+	if (WARN_ON(type != CCN_TYPE_HNF && type != CCN_TYPE_SBAS &&
+			!arm_ccn_pmu_type_eq(type, CCN_TYPE_RNI_3P)))
+		return;
+
+	/* Set the event id for the pre-allocated counter */
+	val = readl(source->base + CCN_HNF_PMU_EVENT_SEL);
+	val &= ~(CCN_HNF_PMU_EVENT_SEL__ID__MASK <<
+		CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(hw->config_base));
+	val |= CCN_CONFIG_EVENT(event->attr.config) <<
+		CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(hw->config_base);
+	writel(val, source->base + CCN_HNF_PMU_EVENT_SEL);
+}
+
+static void arm_ccn_pmu_event_config(struct perf_event *event)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	u32 xp, offset, val;
+
+	/* Cycle counter requires no setup */
+	if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER)
+		return;
+
+	if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP)
+		xp = CCN_CONFIG_XP(event->attr.config);
+	else
+		xp = arm_ccn_node_to_xp(CCN_CONFIG_NODE(event->attr.config));
+
+	spin_lock(&ccn->dt.config_lock);
+
+	/* Set the DT bus "distance" register */
+	offset = (hw->idx / 4) * 4;
+	val = readl(ccn->dt.base + CCN_DT_ACTIVE_DSM + offset);
+	val &= ~(CCN_DT_ACTIVE_DSM__DSM_ID__MASK <<
+			CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(hw->idx % 4));
+	val |= xp << CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(hw->idx % 4);
+	writel(val, ccn->dt.base + CCN_DT_ACTIVE_DSM + offset);
+
+	if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP) {
+		if (CCN_CONFIG_EVENT(event->attr.config) ==
+				CCN_EVENT_WATCHPOINT)
+			arm_ccn_pmu_xp_watchpoint_config(event);
+		else
+			arm_ccn_pmu_xp_event_config(event);
+	} else {
+		arm_ccn_pmu_node_event_config(event);
+	}
+
+	spin_unlock(&ccn->dt.config_lock);
+}
+
+static int arm_ccn_pmu_active_counters(struct arm_ccn *ccn)
+{
+	return bitmap_weight(ccn->dt.pmu_counters_mask,
+			     CCN_NUM_PMU_EVENT_COUNTERS + 1);
+}
+
+static int arm_ccn_pmu_event_add(struct perf_event *event, int flags)
+{
+	int err;
+	struct hw_perf_event *hw = &event->hw;
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+
+	err = arm_ccn_pmu_event_alloc(event);
+	if (err)
+		return err;
+
+	/*
+	 * Pin the timer, so that the overflows are handled by the chosen
+	 * event->cpu (this is the same one as presented in "cpumask"
+	 * attribute).
+	 */
+	if (!ccn->irq && arm_ccn_pmu_active_counters(ccn) == 1)
+		hrtimer_start(&ccn->dt.hrtimer, arm_ccn_pmu_timer_period(),
+			      HRTIMER_MODE_REL_PINNED);
+
+	arm_ccn_pmu_event_config(event);
+
+	hw->state = PERF_HES_STOPPED;
+
+	if (flags & PERF_EF_START)
+		arm_ccn_pmu_event_start(event, PERF_EF_UPDATE);
+
+	return 0;
+}
+
+static void arm_ccn_pmu_event_del(struct perf_event *event, int flags)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu);
+
+	arm_ccn_pmu_event_stop(event, PERF_EF_UPDATE);
+
+	arm_ccn_pmu_event_release(event);
+
+	if (!ccn->irq && arm_ccn_pmu_active_counters(ccn) == 0)
+		hrtimer_cancel(&ccn->dt.hrtimer);
+}
+
+static void arm_ccn_pmu_event_read(struct perf_event *event)
+{
+	arm_ccn_pmu_event_update(event);
+}
+
+static void arm_ccn_pmu_enable(struct pmu *pmu)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(pmu);
+
+	u32 val = readl(ccn->dt.base + CCN_DT_PMCR);
+	val |= CCN_DT_PMCR__PMU_EN;
+	writel(val, ccn->dt.base + CCN_DT_PMCR);
+}
+
+static void arm_ccn_pmu_disable(struct pmu *pmu)
+{
+	struct arm_ccn *ccn = pmu_to_arm_ccn(pmu);
+
+	u32 val = readl(ccn->dt.base + CCN_DT_PMCR);
+	val &= ~CCN_DT_PMCR__PMU_EN;
+	writel(val, ccn->dt.base + CCN_DT_PMCR);
+}
+
+static irqreturn_t arm_ccn_pmu_overflow_handler(struct arm_ccn_dt *dt)
+{
+	u32 pmovsr = readl(dt->base + CCN_DT_PMOVSR);
+	int idx;
+
+	if (!pmovsr)
+		return IRQ_NONE;
+
+	writel(pmovsr, dt->base + CCN_DT_PMOVSR_CLR);
+
+	BUILD_BUG_ON(CCN_IDX_PMU_CYCLE_COUNTER != CCN_NUM_PMU_EVENT_COUNTERS);
+
+	for (idx = 0; idx < CCN_NUM_PMU_EVENT_COUNTERS + 1; idx++) {
+		struct perf_event *event = dt->pmu_counters[idx].event;
+		int overflowed = pmovsr & BIT(idx);
+
+		WARN_ON_ONCE(overflowed && !event &&
+				idx != CCN_IDX_PMU_CYCLE_COUNTER);
+
+		if (!event || !overflowed)
+			continue;
+
+		arm_ccn_pmu_event_update(event);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static enum hrtimer_restart arm_ccn_pmu_timer_handler(struct hrtimer *hrtimer)
+{
+	struct arm_ccn_dt *dt = container_of(hrtimer, struct arm_ccn_dt,
+			hrtimer);
+	unsigned long flags;
+
+	local_irq_save(flags);
+	arm_ccn_pmu_overflow_handler(dt);
+	local_irq_restore(flags);
+
+	hrtimer_forward_now(hrtimer, arm_ccn_pmu_timer_period());
+	return HRTIMER_RESTART;
+}
+
+
+static int arm_ccn_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_ccn_dt *dt = hlist_entry_safe(node, struct arm_ccn_dt, node);
+	struct arm_ccn *ccn = container_of(dt, struct arm_ccn, dt);
+	unsigned int target;
+
+	if (cpu != dt->cpu)
+		return 0;
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+	perf_pmu_migrate_context(&dt->pmu, cpu, target);
+	dt->cpu = target;
+	if (ccn->irq)
+		WARN_ON(irq_set_affinity_hint(ccn->irq, cpumask_of(dt->cpu)));
+	return 0;
+}
+
+static DEFINE_IDA(arm_ccn_pmu_ida);
+
+static int arm_ccn_pmu_init(struct arm_ccn *ccn)
+{
+	int i;
+	char *name;
+	int err;
+
+	/* Initialize DT subsystem */
+	ccn->dt.base = ccn->base + CCN_REGION_SIZE;
+	spin_lock_init(&ccn->dt.config_lock);
+	writel(CCN_DT_PMOVSR_CLR__MASK, ccn->dt.base + CCN_DT_PMOVSR_CLR);
+	writel(CCN_DT_CTL__DT_EN, ccn->dt.base + CCN_DT_CTL);
+	writel(CCN_DT_PMCR__OVFL_INTR_EN | CCN_DT_PMCR__PMU_EN,
+			ccn->dt.base + CCN_DT_PMCR);
+	writel(0x1, ccn->dt.base + CCN_DT_PMSR_CLR);
+	for (i = 0; i < ccn->num_xps; i++) {
+		writel(0, ccn->xp[i].base + CCN_XP_DT_CONFIG);
+		writel((CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS <<
+				CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(0)) |
+				(CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS <<
+				CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(1)) |
+				CCN_XP_DT_CONTROL__DT_ENABLE,
+				ccn->xp[i].base + CCN_XP_DT_CONTROL);
+	}
+	ccn->dt.cmp_mask[CCN_IDX_MASK_ANY].l = ~0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_ANY].h = ~0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_EXACT].l = 0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_EXACT].h = 0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_ORDER].l = ~0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_ORDER].h = ~(0x1 << 15);
+	ccn->dt.cmp_mask[CCN_IDX_MASK_OPCODE].l = ~0;
+	ccn->dt.cmp_mask[CCN_IDX_MASK_OPCODE].h = ~(0x1f << 9);
+
+	/* Get a convenient /sys/event_source/devices/ name */
+	ccn->dt.id = ida_simple_get(&arm_ccn_pmu_ida, 0, 0, GFP_KERNEL);
+	if (ccn->dt.id == 0) {
+		name = "ccn";
+	} else {
+		name = devm_kasprintf(ccn->dev, GFP_KERNEL, "ccn_%d",
+				      ccn->dt.id);
+		if (!name) {
+			err = -ENOMEM;
+			goto error_choose_name;
+		}
+	}
+
+	/* Perf driver registration */
+	ccn->dt.pmu = (struct pmu) {
+		.module = THIS_MODULE,
+		.attr_groups = arm_ccn_pmu_attr_groups,
+		.task_ctx_nr = perf_invalid_context,
+		.event_init = arm_ccn_pmu_event_init,
+		.add = arm_ccn_pmu_event_add,
+		.del = arm_ccn_pmu_event_del,
+		.start = arm_ccn_pmu_event_start,
+		.stop = arm_ccn_pmu_event_stop,
+		.read = arm_ccn_pmu_event_read,
+		.pmu_enable = arm_ccn_pmu_enable,
+		.pmu_disable = arm_ccn_pmu_disable,
+		.capabilities = PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	/* No overflow interrupt? Have to use a timer instead. */
+	if (!ccn->irq) {
+		dev_info(ccn->dev, "No access to interrupts, using timer.\n");
+		hrtimer_init(&ccn->dt.hrtimer, CLOCK_MONOTONIC,
+				HRTIMER_MODE_REL);
+		ccn->dt.hrtimer.function = arm_ccn_pmu_timer_handler;
+	}
+
+	/* Pick one CPU which we will use to collect data from CCN... */
+	ccn->dt.cpu = raw_smp_processor_id();
+
+	/* Also make sure that the overflow interrupt is handled by this CPU */
+	if (ccn->irq) {
+		err = irq_set_affinity_hint(ccn->irq, cpumask_of(ccn->dt.cpu));
+		if (err) {
+			dev_err(ccn->dev, "Failed to set interrupt affinity!\n");
+			goto error_set_affinity;
+		}
+	}
+
+	cpuhp_state_add_instance_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE,
+					 &ccn->dt.node);
+
+	err = perf_pmu_register(&ccn->dt.pmu, name, -1);
+	if (err)
+		goto error_pmu_register;
+
+	return 0;
+
+error_pmu_register:
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE,
+					    &ccn->dt.node);
+error_set_affinity:
+error_choose_name:
+	ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id);
+	for (i = 0; i < ccn->num_xps; i++)
+		writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL);
+	writel(0, ccn->dt.base + CCN_DT_PMCR);
+	return err;
+}
+
+static void arm_ccn_pmu_cleanup(struct arm_ccn *ccn)
+{
+	int i;
+
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE,
+					    &ccn->dt.node);
+	if (ccn->irq)
+		irq_set_affinity_hint(ccn->irq, NULL);
+	for (i = 0; i < ccn->num_xps; i++)
+		writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL);
+	writel(0, ccn->dt.base + CCN_DT_PMCR);
+	perf_pmu_unregister(&ccn->dt.pmu);
+	ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id);
+}
+
+static int arm_ccn_for_each_valid_region(struct arm_ccn *ccn,
+		int (*callback)(struct arm_ccn *ccn, int region,
+		void __iomem *base, u32 type, u32 id))
+{
+	int region;
+
+	for (region = 0; region < CCN_NUM_REGIONS; region++) {
+		u32 val, type, id;
+		void __iomem *base;
+		int err;
+
+		val = readl(ccn->base + CCN_MN_OLY_COMP_LIST_63_0 +
+				4 * (region / 32));
+		if (!(val & (1 << (region % 32))))
+			continue;
+
+		base = ccn->base + region * CCN_REGION_SIZE;
+		val = readl(base + CCN_ALL_OLY_ID);
+		type = (val >> CCN_ALL_OLY_ID__OLY_ID__SHIFT) &
+				CCN_ALL_OLY_ID__OLY_ID__MASK;
+		id = (val >> CCN_ALL_OLY_ID__NODE_ID__SHIFT) &
+				CCN_ALL_OLY_ID__NODE_ID__MASK;
+
+		err = callback(ccn, region, base, type, id);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static int arm_ccn_get_nodes_num(struct arm_ccn *ccn, int region,
+		void __iomem *base, u32 type, u32 id)
+{
+
+	if (type == CCN_TYPE_XP && id >= ccn->num_xps)
+		ccn->num_xps = id + 1;
+	else if (id >= ccn->num_nodes)
+		ccn->num_nodes = id + 1;
+
+	return 0;
+}
+
+static int arm_ccn_init_nodes(struct arm_ccn *ccn, int region,
+		void __iomem *base, u32 type, u32 id)
+{
+	struct arm_ccn_component *component;
+
+	dev_dbg(ccn->dev, "Region %d: id=%u, type=0x%02x\n", region, id, type);
+
+	switch (type) {
+	case CCN_TYPE_MN:
+		ccn->mn_id = id;
+		return 0;
+	case CCN_TYPE_DT:
+		return 0;
+	case CCN_TYPE_XP:
+		component = &ccn->xp[id];
+		break;
+	case CCN_TYPE_SBSX:
+		ccn->sbsx_present = 1;
+		component = &ccn->node[id];
+		break;
+	case CCN_TYPE_SBAS:
+		ccn->sbas_present = 1;
+		fallthrough;
+	default:
+		component = &ccn->node[id];
+		break;
+	}
+
+	component->base = base;
+	component->type = type;
+
+	return 0;
+}
+
+
+static irqreturn_t arm_ccn_error_handler(struct arm_ccn *ccn,
+		const u32 *err_sig_val)
+{
+	/* This should be really handled by firmware... */
+	dev_err(ccn->dev, "Error reported in %08x%08x%08x%08x%08x%08x.\n",
+			err_sig_val[5], err_sig_val[4], err_sig_val[3],
+			err_sig_val[2], err_sig_val[1], err_sig_val[0]);
+	dev_err(ccn->dev, "Disabling interrupt generation for all errors.\n");
+	writel(CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLE,
+			ccn->base + CCN_MN_ERRINT_STATUS);
+
+	return IRQ_HANDLED;
+}
+
+
+static irqreturn_t arm_ccn_irq_handler(int irq, void *dev_id)
+{
+	irqreturn_t res = IRQ_NONE;
+	struct arm_ccn *ccn = dev_id;
+	u32 err_sig_val[6];
+	u32 err_or;
+	int i;
+
+	/* PMU overflow is a special case */
+	err_or = err_sig_val[0] = readl(ccn->base + CCN_MN_ERR_SIG_VAL_63_0);
+	if (err_or & CCN_MN_ERR_SIG_VAL_63_0__DT) {
+		err_or &= ~CCN_MN_ERR_SIG_VAL_63_0__DT;
+		res = arm_ccn_pmu_overflow_handler(&ccn->dt);
+	}
+
+	/* Have to read all err_sig_vals to clear them */
+	for (i = 1; i < ARRAY_SIZE(err_sig_val); i++) {
+		err_sig_val[i] = readl(ccn->base +
+				CCN_MN_ERR_SIG_VAL_63_0 + i * 4);
+		err_or |= err_sig_val[i];
+	}
+	if (err_or)
+		res |= arm_ccn_error_handler(ccn, err_sig_val);
+
+	if (res != IRQ_NONE)
+		writel(CCN_MN_ERRINT_STATUS__INTREQ__DESSERT,
+				ccn->base + CCN_MN_ERRINT_STATUS);
+
+	return res;
+}
+
+
+static int arm_ccn_probe(struct platform_device *pdev)
+{
+	struct arm_ccn *ccn;
+	struct resource *res;
+	unsigned int irq;
+	int err;
+
+	ccn = devm_kzalloc(&pdev->dev, sizeof(*ccn), GFP_KERNEL);
+	if (!ccn)
+		return -ENOMEM;
+	ccn->dev = &pdev->dev;
+	platform_set_drvdata(pdev, ccn);
+
+	ccn->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(ccn->base))
+		return PTR_ERR(ccn->base);
+
+	res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
+	if (!res)
+		return -EINVAL;
+	irq = res->start;
+
+	/* Check if we can use the interrupt */
+	writel(CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLE,
+			ccn->base + CCN_MN_ERRINT_STATUS);
+	if (readl(ccn->base + CCN_MN_ERRINT_STATUS) &
+			CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLED) {
+		/* Can set 'disable' bits, so can acknowledge interrupts */
+		writel(CCN_MN_ERRINT_STATUS__PMU_EVENTS__ENABLE,
+				ccn->base + CCN_MN_ERRINT_STATUS);
+		err = devm_request_irq(ccn->dev, irq, arm_ccn_irq_handler,
+				       IRQF_NOBALANCING | IRQF_NO_THREAD,
+				       dev_name(ccn->dev), ccn);
+		if (err)
+			return err;
+
+		ccn->irq = irq;
+	}
+
+
+	/* Build topology */
+
+	err = arm_ccn_for_each_valid_region(ccn, arm_ccn_get_nodes_num);
+	if (err)
+		return err;
+
+	ccn->node = devm_kcalloc(ccn->dev, ccn->num_nodes, sizeof(*ccn->node),
+				 GFP_KERNEL);
+	ccn->xp = devm_kcalloc(ccn->dev, ccn->num_xps, sizeof(*ccn->node),
+			       GFP_KERNEL);
+	if (!ccn->node || !ccn->xp)
+		return -ENOMEM;
+
+	err = arm_ccn_for_each_valid_region(ccn, arm_ccn_init_nodes);
+	if (err)
+		return err;
+
+	return arm_ccn_pmu_init(ccn);
+}
+
+static int arm_ccn_remove(struct platform_device *pdev)
+{
+	struct arm_ccn *ccn = platform_get_drvdata(pdev);
+
+	arm_ccn_pmu_cleanup(ccn);
+
+	return 0;
+}
+
+static const struct of_device_id arm_ccn_match[] = {
+	{ .compatible = "arm,ccn-502", },
+	{ .compatible = "arm,ccn-504", },
+	{ .compatible = "arm,ccn-512", },
+	{},
+};
+MODULE_DEVICE_TABLE(of, arm_ccn_match);
+
+static struct platform_driver arm_ccn_driver = {
+	.driver = {
+		.name = "arm-ccn",
+		.of_match_table = arm_ccn_match,
+		.suppress_bind_attrs = true,
+	},
+	.probe = arm_ccn_probe,
+	.remove = arm_ccn_remove,
+};
+
+static int __init arm_ccn_init(void)
+{
+	int i, ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_CCN_ONLINE,
+				      "perf/arm/ccn:online", NULL,
+				      arm_ccn_pmu_offline_cpu);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < ARRAY_SIZE(arm_ccn_pmu_events); i++)
+		arm_ccn_pmu_events_attrs[i] = &arm_ccn_pmu_events[i].attr.attr;
+
+	ret = platform_driver_register(&arm_ccn_driver);
+	if (ret)
+		cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CCN_ONLINE);
+	return ret;
+}
+
+static void __exit arm_ccn_exit(void)
+{
+	platform_driver_unregister(&arm_ccn_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CCN_ONLINE);
+}
+
+module_init(arm_ccn_init);
+module_exit(arm_ccn_exit);
+
+MODULE_AUTHOR("Pawel Moll <pawel.moll@arm.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c
new file mode 100644
index 000000000..ac4428e8f
--- /dev/null
+++ b/drivers/perf/arm-cmn.c
@@ -0,0 +1,1639 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (C) 2016-2020 Arm Limited
+// CMN-600 Coherent Mesh Network PMU driver
+
+#include <linux/acpi.h>
+#include <linux/bitfield.h>
+#include <linux/bitops.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/sort.h>
+
+/* Common register stuff */
+#define CMN_NODE_INFO			0x0000
+#define CMN_NI_NODE_TYPE		GENMASK_ULL(15, 0)
+#define CMN_NI_NODE_ID			GENMASK_ULL(31, 16)
+#define CMN_NI_LOGICAL_ID		GENMASK_ULL(47, 32)
+
+#define CMN_NODEID_DEVID(reg)		((reg) & 3)
+#define CMN_NODEID_PID(reg)		(((reg) >> 2) & 1)
+#define CMN_NODEID_X(reg, bits)		((reg) >> (3 + (bits)))
+#define CMN_NODEID_Y(reg, bits)		(((reg) >> 3) & ((1U << (bits)) - 1))
+
+#define CMN_CHILD_INFO			0x0080
+#define CMN_CI_CHILD_COUNT		GENMASK_ULL(15, 0)
+#define CMN_CI_CHILD_PTR_OFFSET		GENMASK_ULL(31, 16)
+
+#define CMN_CHILD_NODE_ADDR		GENMASK(27,0)
+#define CMN_CHILD_NODE_EXTERNAL		BIT(31)
+
+#define CMN_ADDR_NODE_PTR		GENMASK(27, 14)
+
+#define CMN_NODE_PTR_DEVID(ptr)		(((ptr) >> 2) & 3)
+#define CMN_NODE_PTR_PID(ptr)		((ptr) & 1)
+#define CMN_NODE_PTR_X(ptr, bits)	((ptr) >> (6 + (bits)))
+#define CMN_NODE_PTR_Y(ptr, bits)	(((ptr) >> 6) & ((1U << (bits)) - 1))
+
+#define CMN_MAX_XPS			(8 * 8)
+
+/* The CFG node has one other useful purpose */
+#define CMN_CFGM_PERIPH_ID_2		0x0010
+#define CMN_CFGM_PID2_REVISION		GENMASK(7, 4)
+
+/* PMU registers occupy the 3rd 4KB page of each node's 16KB space */
+#define CMN_PMU_OFFSET			0x2000
+
+/* For most nodes, this is all there is */
+#define CMN_PMU_EVENT_SEL		0x000
+#define CMN_PMU_EVENTn_ID_SHIFT(n)	((n) * 8)
+
+/* DTMs live in the PMU space of XP registers */
+#define CMN_DTM_WPn(n)			(0x1A0 + (n) * 0x18)
+#define CMN_DTM_WPn_CONFIG(n)		(CMN_DTM_WPn(n) + 0x00)
+#define CMN_DTM_WPn_CONFIG_WP_COMBINE	BIT(6)
+#define CMN_DTM_WPn_CONFIG_WP_EXCLUSIVE	BIT(5)
+#define CMN_DTM_WPn_CONFIG_WP_GRP	BIT(4)
+#define CMN_DTM_WPn_CONFIG_WP_CHN_SEL	GENMASK_ULL(3, 1)
+#define CMN_DTM_WPn_CONFIG_WP_DEV_SEL	BIT(0)
+#define CMN_DTM_WPn_VAL(n)		(CMN_DTM_WPn(n) + 0x08)
+#define CMN_DTM_WPn_MASK(n)		(CMN_DTM_WPn(n) + 0x10)
+
+#define CMN_DTM_PMU_CONFIG		0x210
+#define CMN__PMEVCNT0_INPUT_SEL		GENMASK_ULL(37, 32)
+#define CMN__PMEVCNT0_INPUT_SEL_WP	0x00
+#define CMN__PMEVCNT0_INPUT_SEL_XP	0x04
+#define CMN__PMEVCNT0_INPUT_SEL_DEV	0x10
+#define CMN__PMEVCNT0_GLOBAL_NUM	GENMASK_ULL(18, 16)
+#define CMN__PMEVCNTn_GLOBAL_NUM_SHIFT(n)	((n) * 4)
+#define CMN__PMEVCNT_PAIRED(n)		BIT(4 + (n))
+#define CMN__PMEVCNT23_COMBINED		BIT(2)
+#define CMN__PMEVCNT01_COMBINED		BIT(1)
+#define CMN_DTM_PMU_CONFIG_PMU_EN	BIT(0)
+
+#define CMN_DTM_PMEVCNT			0x220
+
+#define CMN_DTM_PMEVCNTSR		0x240
+
+#define CMN_DTM_NUM_COUNTERS		4
+
+/* The DTC node is where the magic happens */
+#define CMN_DT_DTC_CTL			0x0a00
+#define CMN_DT_DTC_CTL_DT_EN		BIT(0)
+
+/* DTC counters are paired in 64-bit registers on a 16-byte stride. Yuck */
+#define _CMN_DT_CNT_REG(n)		((((n) / 2) * 4 + (n) % 2) * 4)
+#define CMN_DT_PMEVCNT(n)		(CMN_PMU_OFFSET + _CMN_DT_CNT_REG(n))
+#define CMN_DT_PMCCNTR			(CMN_PMU_OFFSET + 0x40)
+
+#define CMN_DT_PMEVCNTSR(n)		(CMN_PMU_OFFSET + 0x50 + _CMN_DT_CNT_REG(n))
+#define CMN_DT_PMCCNTRSR		(CMN_PMU_OFFSET + 0x90)
+
+#define CMN_DT_PMCR			(CMN_PMU_OFFSET + 0x100)
+#define CMN_DT_PMCR_PMU_EN		BIT(0)
+#define CMN_DT_PMCR_CNTR_RST		BIT(5)
+#define CMN_DT_PMCR_OVFL_INTR_EN	BIT(6)
+
+#define CMN_DT_PMOVSR			(CMN_PMU_OFFSET + 0x118)
+#define CMN_DT_PMOVSR_CLR		(CMN_PMU_OFFSET + 0x120)
+
+#define CMN_DT_PMSSR			(CMN_PMU_OFFSET + 0x128)
+#define CMN_DT_PMSSR_SS_STATUS(n)	BIT(n)
+
+#define CMN_DT_PMSRR			(CMN_PMU_OFFSET + 0x130)
+#define CMN_DT_PMSRR_SS_REQ		BIT(0)
+
+#define CMN_DT_NUM_COUNTERS		8
+#define CMN_MAX_DTCS			4
+
+/*
+ * Even in the worst case a DTC counter can't wrap in fewer than 2^42 cycles,
+ * so throwing away one bit to make overflow handling easy is no big deal.
+ */
+#define CMN_COUNTER_INIT		0x80000000
+/* Similarly for the 40-bit cycle counter */
+#define CMN_CC_INIT			0x8000000000ULL
+
+
+/* Event attributes */
+#define CMN_CONFIG_TYPE			GENMASK(15, 0)
+#define CMN_CONFIG_EVENTID		GENMASK(23, 16)
+#define CMN_CONFIG_OCCUPID		GENMASK(27, 24)
+#define CMN_CONFIG_BYNODEID		BIT(31)
+#define CMN_CONFIG_NODEID		GENMASK(47, 32)
+
+#define CMN_EVENT_TYPE(event)		FIELD_GET(CMN_CONFIG_TYPE, (event)->attr.config)
+#define CMN_EVENT_EVENTID(event)	FIELD_GET(CMN_CONFIG_EVENTID, (event)->attr.config)
+#define CMN_EVENT_OCCUPID(event)	FIELD_GET(CMN_CONFIG_OCCUPID, (event)->attr.config)
+#define CMN_EVENT_BYNODEID(event)	FIELD_GET(CMN_CONFIG_BYNODEID, (event)->attr.config)
+#define CMN_EVENT_NODEID(event)		FIELD_GET(CMN_CONFIG_NODEID, (event)->attr.config)
+
+#define CMN_CONFIG_WP_COMBINE		GENMASK(27, 24)
+#define CMN_CONFIG_WP_DEV_SEL		BIT(48)
+#define CMN_CONFIG_WP_CHN_SEL		GENMASK(50, 49)
+#define CMN_CONFIG_WP_GRP		BIT(52)
+#define CMN_CONFIG_WP_EXCLUSIVE		BIT(53)
+#define CMN_CONFIG1_WP_VAL		GENMASK(63, 0)
+#define CMN_CONFIG2_WP_MASK		GENMASK(63, 0)
+
+#define CMN_EVENT_WP_COMBINE(event)	FIELD_GET(CMN_CONFIG_WP_COMBINE, (event)->attr.config)
+#define CMN_EVENT_WP_DEV_SEL(event)	FIELD_GET(CMN_CONFIG_WP_DEV_SEL, (event)->attr.config)
+#define CMN_EVENT_WP_CHN_SEL(event)	FIELD_GET(CMN_CONFIG_WP_CHN_SEL, (event)->attr.config)
+#define CMN_EVENT_WP_GRP(event)		FIELD_GET(CMN_CONFIG_WP_GRP, (event)->attr.config)
+#define CMN_EVENT_WP_EXCLUSIVE(event)	FIELD_GET(CMN_CONFIG_WP_EXCLUSIVE, (event)->attr.config)
+#define CMN_EVENT_WP_VAL(event)		FIELD_GET(CMN_CONFIG1_WP_VAL, (event)->attr.config1)
+#define CMN_EVENT_WP_MASK(event)	FIELD_GET(CMN_CONFIG2_WP_MASK, (event)->attr.config2)
+
+/* Made-up event IDs for watchpoint direction */
+#define CMN_WP_UP			0
+#define CMN_WP_DOWN			2
+
+
+/* r0px probably don't exist in silicon, thankfully */
+enum cmn_revision {
+	CMN600_R1P0,
+	CMN600_R1P1,
+	CMN600_R1P2,
+	CMN600_R1P3,
+	CMN600_R2P0,
+	CMN600_R3P0,
+};
+
+enum cmn_node_type {
+	CMN_TYPE_INVALID,
+	CMN_TYPE_DVM,
+	CMN_TYPE_CFG,
+	CMN_TYPE_DTC,
+	CMN_TYPE_HNI,
+	CMN_TYPE_HNF,
+	CMN_TYPE_XP,
+	CMN_TYPE_SBSX,
+	CMN_TYPE_RNI = 0xa,
+	CMN_TYPE_RND = 0xd,
+	CMN_TYPE_RNSAM = 0xf,
+	CMN_TYPE_CXRA = 0x100,
+	CMN_TYPE_CXHA = 0x101,
+	CMN_TYPE_CXLA = 0x102,
+	/* Not a real node type */
+	CMN_TYPE_WP = 0x7770
+};
+
+struct arm_cmn_node {
+	void __iomem *pmu_base;
+	u16 id, logid;
+	enum cmn_node_type type;
+
+	union {
+		/* Device node */
+		struct {
+			int to_xp;
+			/* DN/HN-F/CXHA */
+			unsigned int occupid_val;
+			unsigned int occupid_count;
+		};
+		/* XP */
+		struct {
+			int dtc;
+			u32 pmu_config_low;
+			union {
+				u8 input_sel[4];
+				__le32 pmu_config_high;
+			};
+			s8 wp_event[4];
+		};
+	};
+
+	union {
+		u8 event[4];
+		__le32 event_sel;
+	};
+};
+
+struct arm_cmn_dtc {
+	void __iomem *base;
+	int irq;
+	int irq_friend;
+	bool cc_active;
+
+	struct perf_event *counters[CMN_DT_NUM_COUNTERS];
+	struct perf_event *cycles;
+};
+
+#define CMN_STATE_DISABLED	BIT(0)
+#define CMN_STATE_TXN		BIT(1)
+
+struct arm_cmn {
+	struct device *dev;
+	void __iomem *base;
+
+	enum cmn_revision rev;
+	u8 mesh_x;
+	u8 mesh_y;
+	u16 num_xps;
+	u16 num_dns;
+	struct arm_cmn_node *xps;
+	struct arm_cmn_node *dns;
+
+	struct arm_cmn_dtc *dtc;
+	unsigned int num_dtcs;
+
+	int cpu;
+	struct hlist_node cpuhp_node;
+
+	unsigned int state;
+	struct pmu pmu;
+};
+
+#define to_cmn(p)	container_of(p, struct arm_cmn, pmu)
+
+static int arm_cmn_hp_state;
+
+struct arm_cmn_hw_event {
+	struct arm_cmn_node *dn;
+	u64 dtm_idx[2];
+	unsigned int dtc_idx;
+	u8 dtcs_used;
+	u8 num_dns;
+};
+
+#define for_each_hw_dn(hw, dn, i) \
+	for (i = 0, dn = hw->dn; i < hw->num_dns; i++, dn++)
+
+static struct arm_cmn_hw_event *to_cmn_hw(struct perf_event *event)
+{
+	BUILD_BUG_ON(sizeof(struct arm_cmn_hw_event) > offsetof(struct hw_perf_event, target));
+	return (struct arm_cmn_hw_event *)&event->hw;
+}
+
+static void arm_cmn_set_index(u64 x[], unsigned int pos, unsigned int val)
+{
+	x[pos / 32] |= (u64)val << ((pos % 32) * 2);
+}
+
+static unsigned int arm_cmn_get_index(u64 x[], unsigned int pos)
+{
+	return (x[pos / 32] >> ((pos % 32) * 2)) & 3;
+}
+
+struct arm_cmn_event_attr {
+	struct device_attribute attr;
+	enum cmn_node_type type;
+	u8 eventid;
+	u8 occupid;
+};
+
+struct arm_cmn_format_attr {
+	struct device_attribute attr;
+	u64 field;
+	int config;
+};
+
+static int arm_cmn_xyidbits(const struct arm_cmn *cmn)
+{
+	return cmn->mesh_x > 4 || cmn->mesh_y > 4 ? 3 : 2;
+}
+
+static void arm_cmn_init_node_to_xp(const struct arm_cmn *cmn,
+				    struct arm_cmn_node *dn)
+{
+	int bits = arm_cmn_xyidbits(cmn);
+	int x = CMN_NODEID_X(dn->id, bits);
+	int y = CMN_NODEID_Y(dn->id, bits);
+	int xp_idx = cmn->mesh_x * y + x;
+
+	dn->to_xp = (cmn->xps + xp_idx) - dn;
+}
+
+static struct arm_cmn_node *arm_cmn_node_to_xp(struct arm_cmn_node *dn)
+{
+	return dn->type == CMN_TYPE_XP ? dn : dn + dn->to_xp;
+}
+
+static struct arm_cmn_node *arm_cmn_node(const struct arm_cmn *cmn,
+					 enum cmn_node_type type)
+{
+	int i;
+
+	for (i = 0; i < cmn->num_dns; i++)
+		if (cmn->dns[i].type == type)
+			return &cmn->dns[i];
+	return NULL;
+}
+
+#define CMN_EVENT_ATTR(_name, _type, _eventid, _occupid)		\
+	(&((struct arm_cmn_event_attr[]) {{				\
+		.attr = __ATTR(_name, 0444, arm_cmn_event_show, NULL),	\
+		.type = _type,						\
+		.eventid = _eventid,					\
+		.occupid = _occupid,					\
+	}})[0].attr.attr)
+
+static bool arm_cmn_is_occup_event(enum cmn_node_type type, unsigned int id)
+{
+	return (type == CMN_TYPE_DVM && id == 0x05) ||
+	       (type == CMN_TYPE_HNF && id == 0x0f);
+}
+
+static ssize_t arm_cmn_event_show(struct device *dev,
+				  struct device_attribute *attr, char *buf)
+{
+	struct arm_cmn_event_attr *eattr;
+
+	eattr = container_of(attr, typeof(*eattr), attr);
+
+	if (eattr->type == CMN_TYPE_DTC)
+		return snprintf(buf, PAGE_SIZE, "type=0x%x\n", eattr->type);
+
+	if (eattr->type == CMN_TYPE_WP)
+		return snprintf(buf, PAGE_SIZE,
+				"type=0x%x,eventid=0x%x,wp_dev_sel=?,wp_chn_sel=?,wp_grp=?,wp_val=?,wp_mask=?\n",
+				eattr->type, eattr->eventid);
+
+	if (arm_cmn_is_occup_event(eattr->type, eattr->eventid))
+		return snprintf(buf, PAGE_SIZE, "type=0x%x,eventid=0x%x,occupid=0x%x\n",
+				eattr->type, eattr->eventid, eattr->occupid);
+
+	return snprintf(buf, PAGE_SIZE, "type=0x%x,eventid=0x%x\n",
+			eattr->type, eattr->eventid);
+}
+
+static umode_t arm_cmn_event_attr_is_visible(struct kobject *kobj,
+					     struct attribute *attr,
+					     int unused)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct arm_cmn *cmn = to_cmn(dev_get_drvdata(dev));
+	struct arm_cmn_event_attr *eattr;
+	enum cmn_node_type type;
+
+	eattr = container_of(attr, typeof(*eattr), attr.attr);
+	type = eattr->type;
+
+	/* Watchpoints aren't nodes */
+	if (type == CMN_TYPE_WP)
+		type = CMN_TYPE_XP;
+
+	/* Revision-specific differences */
+	if (cmn->rev < CMN600_R1P2) {
+		if (type == CMN_TYPE_HNF && eattr->eventid == 0x1b)
+			return 0;
+	}
+
+	if (!arm_cmn_node(cmn, type))
+		return 0;
+
+	return attr->mode;
+}
+
+#define _CMN_EVENT_DVM(_name, _event, _occup)			\
+	CMN_EVENT_ATTR(dn_##_name, CMN_TYPE_DVM, _event, _occup)
+#define CMN_EVENT_DTC(_name)					\
+	CMN_EVENT_ATTR(dtc_##_name, CMN_TYPE_DTC, 0, 0)
+#define _CMN_EVENT_HNF(_name, _event, _occup)			\
+	CMN_EVENT_ATTR(hnf_##_name, CMN_TYPE_HNF, _event, _occup)
+#define CMN_EVENT_HNI(_name, _event)				\
+	CMN_EVENT_ATTR(hni_##_name, CMN_TYPE_HNI, _event, 0)
+#define __CMN_EVENT_XP(_name, _event)				\
+	CMN_EVENT_ATTR(mxp_##_name, CMN_TYPE_XP, _event, 0)
+#define CMN_EVENT_SBSX(_name, _event)				\
+	CMN_EVENT_ATTR(sbsx_##_name, CMN_TYPE_SBSX, _event, 0)
+#define CMN_EVENT_RNID(_name, _event)				\
+	CMN_EVENT_ATTR(rnid_##_name, CMN_TYPE_RNI, _event, 0)
+
+#define CMN_EVENT_DVM(_name, _event)				\
+	_CMN_EVENT_DVM(_name, _event, 0)
+#define CMN_EVENT_HNF(_name, _event)				\
+	_CMN_EVENT_HNF(_name, _event, 0)
+#define _CMN_EVENT_XP(_name, _event)				\
+	__CMN_EVENT_XP(e_##_name, (_event) | (0 << 2)),		\
+	__CMN_EVENT_XP(w_##_name, (_event) | (1 << 2)),		\
+	__CMN_EVENT_XP(n_##_name, (_event) | (2 << 2)),		\
+	__CMN_EVENT_XP(s_##_name, (_event) | (3 << 2)),		\
+	__CMN_EVENT_XP(p0_##_name, (_event) | (4 << 2)),	\
+	__CMN_EVENT_XP(p1_##_name, (_event) | (5 << 2))
+
+/* Good thing there are only 3 fundamental XP events... */
+#define CMN_EVENT_XP(_name, _event)				\
+	_CMN_EVENT_XP(req_##_name, (_event) | (0 << 5)),	\
+	_CMN_EVENT_XP(rsp_##_name, (_event) | (1 << 5)),	\
+	_CMN_EVENT_XP(snp_##_name, (_event) | (2 << 5)),	\
+	_CMN_EVENT_XP(dat_##_name, (_event) | (3 << 5))
+
+
+static struct attribute *arm_cmn_event_attrs[] = {
+	CMN_EVENT_DTC(cycles),
+
+	/*
+	 * DVM node events conflict with HN-I events in the equivalent PMU
+	 * slot, but our lazy short-cut of using the DTM counter index for
+	 * the PMU index as well happens to avoid that by construction.
+	 */
+	CMN_EVENT_DVM(rxreq_dvmop,	0x01),
+	CMN_EVENT_DVM(rxreq_dvmsync,	0x02),
+	CMN_EVENT_DVM(rxreq_dvmop_vmid_filtered, 0x03),
+	CMN_EVENT_DVM(rxreq_retried,	0x04),
+	_CMN_EVENT_DVM(rxreq_trk_occupancy_all, 0x05, 0),
+	_CMN_EVENT_DVM(rxreq_trk_occupancy_dvmop, 0x05, 1),
+	_CMN_EVENT_DVM(rxreq_trk_occupancy_dvmsync, 0x05, 2),
+
+	CMN_EVENT_HNF(cache_miss,	0x01),
+	CMN_EVENT_HNF(slc_sf_cache_access, 0x02),
+	CMN_EVENT_HNF(cache_fill,	0x03),
+	CMN_EVENT_HNF(pocq_retry,	0x04),
+	CMN_EVENT_HNF(pocq_reqs_recvd,	0x05),
+	CMN_EVENT_HNF(sf_hit,		0x06),
+	CMN_EVENT_HNF(sf_evictions,	0x07),
+	CMN_EVENT_HNF(dir_snoops_sent,	0x08),
+	CMN_EVENT_HNF(brd_snoops_sent,	0x09),
+	CMN_EVENT_HNF(slc_eviction,	0x0a),
+	CMN_EVENT_HNF(slc_fill_invalid_way, 0x0b),
+	CMN_EVENT_HNF(mc_retries,	0x0c),
+	CMN_EVENT_HNF(mc_reqs,		0x0d),
+	CMN_EVENT_HNF(qos_hh_retry,	0x0e),
+	_CMN_EVENT_HNF(qos_pocq_occupancy_all, 0x0f, 0),
+	_CMN_EVENT_HNF(qos_pocq_occupancy_read, 0x0f, 1),
+	_CMN_EVENT_HNF(qos_pocq_occupancy_write, 0x0f, 2),
+	_CMN_EVENT_HNF(qos_pocq_occupancy_atomic, 0x0f, 3),
+	_CMN_EVENT_HNF(qos_pocq_occupancy_stash, 0x0f, 4),
+	CMN_EVENT_HNF(pocq_addrhaz,	0x10),
+	CMN_EVENT_HNF(pocq_atomic_addrhaz, 0x11),
+	CMN_EVENT_HNF(ld_st_swp_adq_full, 0x12),
+	CMN_EVENT_HNF(cmp_adq_full,	0x13),
+	CMN_EVENT_HNF(txdat_stall,	0x14),
+	CMN_EVENT_HNF(txrsp_stall,	0x15),
+	CMN_EVENT_HNF(seq_full,		0x16),
+	CMN_EVENT_HNF(seq_hit,		0x17),
+	CMN_EVENT_HNF(snp_sent,		0x18),
+	CMN_EVENT_HNF(sfbi_dir_snp_sent, 0x19),
+	CMN_EVENT_HNF(sfbi_brd_snp_sent, 0x1a),
+	CMN_EVENT_HNF(snp_sent_untrk,	0x1b),
+	CMN_EVENT_HNF(intv_dirty,	0x1c),
+	CMN_EVENT_HNF(stash_snp_sent,	0x1d),
+	CMN_EVENT_HNF(stash_data_pull,	0x1e),
+	CMN_EVENT_HNF(snp_fwded,	0x1f),
+
+	CMN_EVENT_HNI(rrt_rd_occ_cnt_ovfl, 0x20),
+	CMN_EVENT_HNI(rrt_wr_occ_cnt_ovfl, 0x21),
+	CMN_EVENT_HNI(rdt_rd_occ_cnt_ovfl, 0x22),
+	CMN_EVENT_HNI(rdt_wr_occ_cnt_ovfl, 0x23),
+	CMN_EVENT_HNI(wdb_occ_cnt_ovfl,	0x24),
+	CMN_EVENT_HNI(rrt_rd_alloc,	0x25),
+	CMN_EVENT_HNI(rrt_wr_alloc,	0x26),
+	CMN_EVENT_HNI(rdt_rd_alloc,	0x27),
+	CMN_EVENT_HNI(rdt_wr_alloc,	0x28),
+	CMN_EVENT_HNI(wdb_alloc,	0x29),
+	CMN_EVENT_HNI(txrsp_retryack,	0x2a),
+	CMN_EVENT_HNI(arvalid_no_arready, 0x2b),
+	CMN_EVENT_HNI(arready_no_arvalid, 0x2c),
+	CMN_EVENT_HNI(awvalid_no_awready, 0x2d),
+	CMN_EVENT_HNI(awready_no_awvalid, 0x2e),
+	CMN_EVENT_HNI(wvalid_no_wready,	0x2f),
+	CMN_EVENT_HNI(txdat_stall,	0x30),
+	CMN_EVENT_HNI(nonpcie_serialization, 0x31),
+	CMN_EVENT_HNI(pcie_serialization, 0x32),
+
+	CMN_EVENT_XP(txflit_valid,	0x01),
+	CMN_EVENT_XP(txflit_stall,	0x02),
+	CMN_EVENT_XP(partial_dat_flit,	0x03),
+	/* We treat watchpoints as a special made-up class of XP events */
+	CMN_EVENT_ATTR(watchpoint_up, CMN_TYPE_WP, 0, 0),
+	CMN_EVENT_ATTR(watchpoint_down, CMN_TYPE_WP, 2, 0),
+
+	CMN_EVENT_SBSX(rd_req,		0x01),
+	CMN_EVENT_SBSX(wr_req,		0x02),
+	CMN_EVENT_SBSX(cmo_req,		0x03),
+	CMN_EVENT_SBSX(txrsp_retryack,	0x04),
+	CMN_EVENT_SBSX(txdat_flitv,	0x05),
+	CMN_EVENT_SBSX(txrsp_flitv,	0x06),
+	CMN_EVENT_SBSX(rd_req_trkr_occ_cnt_ovfl, 0x11),
+	CMN_EVENT_SBSX(wr_req_trkr_occ_cnt_ovfl, 0x12),
+	CMN_EVENT_SBSX(cmo_req_trkr_occ_cnt_ovfl, 0x13),
+	CMN_EVENT_SBSX(wdb_occ_cnt_ovfl, 0x14),
+	CMN_EVENT_SBSX(rd_axi_trkr_occ_cnt_ovfl, 0x15),
+	CMN_EVENT_SBSX(cmo_axi_trkr_occ_cnt_ovfl, 0x16),
+	CMN_EVENT_SBSX(arvalid_no_arready, 0x21),
+	CMN_EVENT_SBSX(awvalid_no_awready, 0x22),
+	CMN_EVENT_SBSX(wvalid_no_wready, 0x23),
+	CMN_EVENT_SBSX(txdat_stall,	0x24),
+	CMN_EVENT_SBSX(txrsp_stall,	0x25),
+
+	CMN_EVENT_RNID(s0_rdata_beats,	0x01),
+	CMN_EVENT_RNID(s1_rdata_beats,	0x02),
+	CMN_EVENT_RNID(s2_rdata_beats,	0x03),
+	CMN_EVENT_RNID(rxdat_flits,	0x04),
+	CMN_EVENT_RNID(txdat_flits,	0x05),
+	CMN_EVENT_RNID(txreq_flits_total, 0x06),
+	CMN_EVENT_RNID(txreq_flits_retried, 0x07),
+	CMN_EVENT_RNID(rrt_occ_ovfl,	0x08),
+	CMN_EVENT_RNID(wrt_occ_ovfl,	0x09),
+	CMN_EVENT_RNID(txreq_flits_replayed, 0x0a),
+	CMN_EVENT_RNID(wrcancel_sent,	0x0b),
+	CMN_EVENT_RNID(s0_wdata_beats,	0x0c),
+	CMN_EVENT_RNID(s1_wdata_beats,	0x0d),
+	CMN_EVENT_RNID(s2_wdata_beats,	0x0e),
+	CMN_EVENT_RNID(rrt_alloc,	0x0f),
+	CMN_EVENT_RNID(wrt_alloc,	0x10),
+	CMN_EVENT_RNID(rdb_unord,	0x11),
+	CMN_EVENT_RNID(rdb_replay,	0x12),
+	CMN_EVENT_RNID(rdb_hybrid,	0x13),
+	CMN_EVENT_RNID(rdb_ord,		0x14),
+
+	NULL
+};
+
+static const struct attribute_group arm_cmn_event_attrs_group = {
+	.name = "events",
+	.attrs = arm_cmn_event_attrs,
+	.is_visible = arm_cmn_event_attr_is_visible,
+};
+
+static ssize_t arm_cmn_format_show(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct arm_cmn_format_attr *fmt = container_of(attr, typeof(*fmt), attr);
+	int lo = __ffs(fmt->field), hi = __fls(fmt->field);
+
+	if (lo == hi)
+		return snprintf(buf, PAGE_SIZE, "config:%d\n", lo);
+
+	if (!fmt->config)
+		return snprintf(buf, PAGE_SIZE, "config:%d-%d\n", lo, hi);
+
+	return snprintf(buf, PAGE_SIZE, "config%d:%d-%d\n", fmt->config, lo, hi);
+}
+
+#define _CMN_FORMAT_ATTR(_name, _cfg, _fld)				\
+	(&((struct arm_cmn_format_attr[]) {{				\
+		.attr = __ATTR(_name, 0444, arm_cmn_format_show, NULL),	\
+		.config = _cfg,						\
+		.field = _fld,						\
+	}})[0].attr.attr)
+#define CMN_FORMAT_ATTR(_name, _fld)	_CMN_FORMAT_ATTR(_name, 0, _fld)
+
+static struct attribute *arm_cmn_format_attrs[] = {
+	CMN_FORMAT_ATTR(type, CMN_CONFIG_TYPE),
+	CMN_FORMAT_ATTR(eventid, CMN_CONFIG_EVENTID),
+	CMN_FORMAT_ATTR(occupid, CMN_CONFIG_OCCUPID),
+	CMN_FORMAT_ATTR(bynodeid, CMN_CONFIG_BYNODEID),
+	CMN_FORMAT_ATTR(nodeid, CMN_CONFIG_NODEID),
+
+	CMN_FORMAT_ATTR(wp_dev_sel, CMN_CONFIG_WP_DEV_SEL),
+	CMN_FORMAT_ATTR(wp_chn_sel, CMN_CONFIG_WP_CHN_SEL),
+	CMN_FORMAT_ATTR(wp_grp, CMN_CONFIG_WP_GRP),
+	CMN_FORMAT_ATTR(wp_exclusive, CMN_CONFIG_WP_EXCLUSIVE),
+	CMN_FORMAT_ATTR(wp_combine, CMN_CONFIG_WP_COMBINE),
+
+	_CMN_FORMAT_ATTR(wp_val, 1, CMN_CONFIG1_WP_VAL),
+	_CMN_FORMAT_ATTR(wp_mask, 2, CMN_CONFIG2_WP_MASK),
+
+	NULL
+};
+
+static const struct attribute_group arm_cmn_format_attrs_group = {
+	.name = "format",
+	.attrs = arm_cmn_format_attrs,
+};
+
+static ssize_t arm_cmn_cpumask_show(struct device *dev,
+				    struct device_attribute *attr, char *buf)
+{
+	struct arm_cmn *cmn = to_cmn(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(cmn->cpu));
+}
+
+static struct device_attribute arm_cmn_cpumask_attr =
+		__ATTR(cpumask, 0444, arm_cmn_cpumask_show, NULL);
+
+static struct attribute *arm_cmn_cpumask_attrs[] = {
+	&arm_cmn_cpumask_attr.attr,
+	NULL,
+};
+
+static struct attribute_group arm_cmn_cpumask_attr_group = {
+	.attrs = arm_cmn_cpumask_attrs,
+};
+
+static const struct attribute_group *arm_cmn_attr_groups[] = {
+	&arm_cmn_event_attrs_group,
+	&arm_cmn_format_attrs_group,
+	&arm_cmn_cpumask_attr_group,
+	NULL
+};
+
+static int arm_cmn_wp_idx(struct perf_event *event)
+{
+	return CMN_EVENT_EVENTID(event) + CMN_EVENT_WP_GRP(event);
+}
+
+static u32 arm_cmn_wp_config(struct perf_event *event)
+{
+	u32 config;
+	u32 dev = CMN_EVENT_WP_DEV_SEL(event);
+	u32 chn = CMN_EVENT_WP_CHN_SEL(event);
+	u32 grp = CMN_EVENT_WP_GRP(event);
+	u32 exc = CMN_EVENT_WP_EXCLUSIVE(event);
+	u32 combine = CMN_EVENT_WP_COMBINE(event);
+
+	config = FIELD_PREP(CMN_DTM_WPn_CONFIG_WP_DEV_SEL, dev) |
+		 FIELD_PREP(CMN_DTM_WPn_CONFIG_WP_CHN_SEL, chn) |
+		 FIELD_PREP(CMN_DTM_WPn_CONFIG_WP_GRP, grp) |
+		 FIELD_PREP(CMN_DTM_WPn_CONFIG_WP_EXCLUSIVE, exc);
+	if (combine && !grp)
+		config |= CMN_DTM_WPn_CONFIG_WP_COMBINE;
+
+	return config;
+}
+
+static void arm_cmn_set_state(struct arm_cmn *cmn, u32 state)
+{
+	if (!cmn->state)
+		writel_relaxed(0, cmn->dtc[0].base + CMN_DT_PMCR);
+	cmn->state |= state;
+}
+
+static void arm_cmn_clear_state(struct arm_cmn *cmn, u32 state)
+{
+	cmn->state &= ~state;
+	if (!cmn->state)
+		writel_relaxed(CMN_DT_PMCR_PMU_EN | CMN_DT_PMCR_OVFL_INTR_EN,
+			       cmn->dtc[0].base + CMN_DT_PMCR);
+}
+
+static void arm_cmn_pmu_enable(struct pmu *pmu)
+{
+	arm_cmn_clear_state(to_cmn(pmu), CMN_STATE_DISABLED);
+}
+
+static void arm_cmn_pmu_disable(struct pmu *pmu)
+{
+	arm_cmn_set_state(to_cmn(pmu), CMN_STATE_DISABLED);
+}
+
+static u64 arm_cmn_read_dtm(struct arm_cmn *cmn, struct arm_cmn_hw_event *hw,
+			    bool snapshot)
+{
+	struct arm_cmn_node *dn;
+	unsigned int i, offset;
+	u64 count = 0;
+
+	offset = snapshot ? CMN_DTM_PMEVCNTSR : CMN_DTM_PMEVCNT;
+	for_each_hw_dn(hw, dn, i) {
+		struct arm_cmn_node *xp = arm_cmn_node_to_xp(dn);
+		int dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
+		u64 reg = readq_relaxed(xp->pmu_base + offset);
+		u16 dtm_count = reg >> (dtm_idx * 16);
+
+		count += dtm_count;
+	}
+	return count;
+}
+
+static u64 arm_cmn_read_cc(struct arm_cmn_dtc *dtc)
+{
+	u64 val = readq_relaxed(dtc->base + CMN_DT_PMCCNTR);
+
+	writeq_relaxed(CMN_CC_INIT, dtc->base + CMN_DT_PMCCNTR);
+	return (val - CMN_CC_INIT) & ((CMN_CC_INIT << 1) - 1);
+}
+
+static u32 arm_cmn_read_counter(struct arm_cmn_dtc *dtc, int idx)
+{
+	u32 val, pmevcnt = CMN_DT_PMEVCNT(idx);
+
+	val = readl_relaxed(dtc->base + pmevcnt);
+	writel_relaxed(CMN_COUNTER_INIT, dtc->base + pmevcnt);
+	return val - CMN_COUNTER_INIT;
+}
+
+static void arm_cmn_init_counter(struct perf_event *event)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	unsigned int i, pmevcnt = CMN_DT_PMEVCNT(hw->dtc_idx);
+	u64 count;
+
+	for (i = 0; hw->dtcs_used & (1U << i); i++) {
+		writel_relaxed(CMN_COUNTER_INIT, cmn->dtc[i].base + pmevcnt);
+		cmn->dtc[i].counters[hw->dtc_idx] = event;
+	}
+
+	count = arm_cmn_read_dtm(cmn, hw, false);
+	local64_set(&event->hw.prev_count, count);
+}
+
+static void arm_cmn_event_read(struct perf_event *event)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	u64 delta, new, prev;
+	unsigned long flags;
+	unsigned int i;
+
+	if (hw->dtc_idx == CMN_DT_NUM_COUNTERS) {
+		i = __ffs(hw->dtcs_used);
+		delta = arm_cmn_read_cc(cmn->dtc + i);
+		local64_add(delta, &event->count);
+		return;
+	}
+	new = arm_cmn_read_dtm(cmn, hw, false);
+	prev = local64_xchg(&event->hw.prev_count, new);
+
+	delta = new - prev;
+
+	local_irq_save(flags);
+	for (i = 0; hw->dtcs_used & (1U << i); i++) {
+		new = arm_cmn_read_counter(cmn->dtc + i, hw->dtc_idx);
+		delta += new << 16;
+	}
+	local_irq_restore(flags);
+	local64_add(delta, &event->count);
+}
+
+static void arm_cmn_event_start(struct perf_event *event, int flags)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	struct arm_cmn_node *dn;
+	enum cmn_node_type type = CMN_EVENT_TYPE(event);
+	int i;
+
+	if (type == CMN_TYPE_DTC) {
+		i = __ffs(hw->dtcs_used);
+		writeq_relaxed(CMN_CC_INIT, cmn->dtc[i].base + CMN_DT_PMCCNTR);
+		cmn->dtc[i].cc_active = true;
+	} else if (type == CMN_TYPE_WP) {
+		int wp_idx = arm_cmn_wp_idx(event);
+		u64 val = CMN_EVENT_WP_VAL(event);
+		u64 mask = CMN_EVENT_WP_MASK(event);
+
+		for_each_hw_dn(hw, dn, i) {
+			writeq_relaxed(val, dn->pmu_base + CMN_DTM_WPn_VAL(wp_idx));
+			writeq_relaxed(mask, dn->pmu_base + CMN_DTM_WPn_MASK(wp_idx));
+		}
+	} else for_each_hw_dn(hw, dn, i) {
+		int dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
+
+		dn->event[dtm_idx] = CMN_EVENT_EVENTID(event);
+		writel_relaxed(le32_to_cpu(dn->event_sel), dn->pmu_base + CMN_PMU_EVENT_SEL);
+	}
+}
+
+static void arm_cmn_event_stop(struct perf_event *event, int flags)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	struct arm_cmn_node *dn;
+	enum cmn_node_type type = CMN_EVENT_TYPE(event);
+	int i;
+
+	if (type == CMN_TYPE_DTC) {
+		i = __ffs(hw->dtcs_used);
+		cmn->dtc[i].cc_active = false;
+	} else if (type == CMN_TYPE_WP) {
+		int wp_idx = arm_cmn_wp_idx(event);
+
+		for_each_hw_dn(hw, dn, i) {
+			writeq_relaxed(0, dn->pmu_base + CMN_DTM_WPn_MASK(wp_idx));
+			writeq_relaxed(~0ULL, dn->pmu_base + CMN_DTM_WPn_VAL(wp_idx));
+		}
+	} else for_each_hw_dn(hw, dn, i) {
+		int dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
+
+		dn->event[dtm_idx] = 0;
+		writel_relaxed(le32_to_cpu(dn->event_sel), dn->pmu_base + CMN_PMU_EVENT_SEL);
+	}
+
+	arm_cmn_event_read(event);
+}
+
+struct arm_cmn_val {
+	u8 dtm_count[CMN_MAX_XPS];
+	u8 occupid[CMN_MAX_XPS];
+	u8 wp[CMN_MAX_XPS][4];
+	int dtc_count;
+	bool cycles;
+};
+
+static void arm_cmn_val_add_event(struct arm_cmn_val *val, struct perf_event *event)
+{
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	struct arm_cmn_node *dn;
+	enum cmn_node_type type;
+	int i;
+	u8 occupid;
+
+	if (is_software_event(event))
+		return;
+
+	type = CMN_EVENT_TYPE(event);
+	if (type == CMN_TYPE_DTC) {
+		val->cycles = true;
+		return;
+	}
+
+	val->dtc_count++;
+	if (arm_cmn_is_occup_event(type, CMN_EVENT_EVENTID(event)))
+		occupid = CMN_EVENT_OCCUPID(event) + 1;
+	else
+		occupid = 0;
+
+	for_each_hw_dn(hw, dn, i) {
+		int wp_idx, xp = arm_cmn_node_to_xp(dn)->logid;
+
+		val->dtm_count[xp]++;
+		val->occupid[xp] = occupid;
+
+		if (type != CMN_TYPE_WP)
+			continue;
+
+		wp_idx = arm_cmn_wp_idx(event);
+		val->wp[xp][wp_idx] = CMN_EVENT_WP_COMBINE(event) + 1;
+	}
+}
+
+static int arm_cmn_validate_group(struct perf_event *event)
+{
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	struct arm_cmn_node *dn;
+	struct perf_event *sibling, *leader = event->group_leader;
+	enum cmn_node_type type;
+	struct arm_cmn_val val;
+	int i;
+	u8 occupid;
+
+	if (leader == event)
+		return 0;
+
+	if (event->pmu != leader->pmu && !is_software_event(leader))
+		return -EINVAL;
+
+	memset(&val, 0, sizeof(val));
+
+	arm_cmn_val_add_event(&val, leader);
+	for_each_sibling_event(sibling, leader)
+		arm_cmn_val_add_event(&val, sibling);
+
+	type = CMN_EVENT_TYPE(event);
+	if (type == CMN_TYPE_DTC)
+		return val.cycles ? -EINVAL : 0;
+
+	if (val.dtc_count == CMN_DT_NUM_COUNTERS)
+		return -EINVAL;
+
+	if (arm_cmn_is_occup_event(type, CMN_EVENT_EVENTID(event)))
+		occupid = CMN_EVENT_OCCUPID(event) + 1;
+	else
+		occupid = 0;
+
+	for_each_hw_dn(hw, dn, i) {
+		int wp_idx, wp_cmb, xp = arm_cmn_node_to_xp(dn)->logid;
+
+		if (val.dtm_count[xp] == CMN_DTM_NUM_COUNTERS)
+			return -EINVAL;
+
+		if (occupid && val.occupid[xp] && occupid != val.occupid[xp])
+			return -EINVAL;
+
+		if (type != CMN_TYPE_WP)
+			continue;
+
+		wp_idx = arm_cmn_wp_idx(event);
+		if (val.wp[xp][wp_idx])
+			return -EINVAL;
+
+		wp_cmb = val.wp[xp][wp_idx ^ 1];
+		if (wp_cmb && wp_cmb != CMN_EVENT_WP_COMBINE(event) + 1)
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int arm_cmn_event_init(struct perf_event *event)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	enum cmn_node_type type;
+	unsigned int i;
+	bool bynodeid;
+	u16 nodeid, eventid;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EINVAL;
+
+	event->cpu = cmn->cpu;
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	type = CMN_EVENT_TYPE(event);
+	/* DTC events (i.e. cycles) already have everything they need */
+	if (type == CMN_TYPE_DTC)
+		return 0;
+
+	/* For watchpoints we need the actual XP node here */
+	if (type == CMN_TYPE_WP) {
+		type = CMN_TYPE_XP;
+		/* ...and we need a "real" direction */
+		eventid = CMN_EVENT_EVENTID(event);
+		if (eventid != CMN_WP_UP && eventid != CMN_WP_DOWN)
+			return -EINVAL;
+	}
+
+	bynodeid = CMN_EVENT_BYNODEID(event);
+	nodeid = CMN_EVENT_NODEID(event);
+
+	hw->dn = arm_cmn_node(cmn, type);
+	for (i = hw->dn - cmn->dns; i < cmn->num_dns && cmn->dns[i].type == type; i++) {
+		if (!bynodeid) {
+			hw->num_dns++;
+		} else if (cmn->dns[i].id != nodeid) {
+			hw->dn++;
+		} else {
+			hw->num_dns = 1;
+			break;
+		}
+	}
+
+	if (!hw->num_dns) {
+		int bits = arm_cmn_xyidbits(cmn);
+
+		dev_dbg(cmn->dev, "invalid node 0x%x (%d,%d,%d,%d) type 0x%x\n",
+			nodeid, CMN_NODEID_X(nodeid, bits), CMN_NODEID_Y(nodeid, bits),
+			CMN_NODEID_PID(nodeid), CMN_NODEID_DEVID(nodeid), type);
+		return -EINVAL;
+	}
+	/*
+	 * By assuming events count in all DTC domains, we cunningly avoid
+	 * needing to know anything about how XPs are assigned to domains.
+	 */
+	hw->dtcs_used = (1U << cmn->num_dtcs) - 1;
+
+	return arm_cmn_validate_group(event);
+}
+
+static void arm_cmn_event_clear(struct arm_cmn *cmn, struct perf_event *event,
+				int i)
+{
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	enum cmn_node_type type = CMN_EVENT_TYPE(event);
+
+	while (i--) {
+		struct arm_cmn_node *xp = arm_cmn_node_to_xp(hw->dn + i);
+		unsigned int dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
+
+		if (type == CMN_TYPE_WP)
+			hw->dn[i].wp_event[arm_cmn_wp_idx(event)] = -1;
+
+		if (arm_cmn_is_occup_event(type, CMN_EVENT_EVENTID(event)))
+			hw->dn[i].occupid_count--;
+
+		xp->pmu_config_low &= ~CMN__PMEVCNT_PAIRED(dtm_idx);
+		writel_relaxed(xp->pmu_config_low, xp->pmu_base + CMN_DTM_PMU_CONFIG);
+	}
+	memset(hw->dtm_idx, 0, sizeof(hw->dtm_idx));
+
+	for (i = 0; hw->dtcs_used & (1U << i); i++)
+		cmn->dtc[i].counters[hw->dtc_idx] = NULL;
+}
+
+static int arm_cmn_event_add(struct perf_event *event, int flags)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	struct arm_cmn_dtc *dtc = &cmn->dtc[0];
+	struct arm_cmn_node *dn;
+	enum cmn_node_type type = CMN_EVENT_TYPE(event);
+	unsigned int i, dtc_idx, input_sel;
+
+	if (type == CMN_TYPE_DTC) {
+		i = 0;
+		while (cmn->dtc[i].cycles)
+			if (++i == cmn->num_dtcs)
+				return -ENOSPC;
+
+		cmn->dtc[i].cycles = event;
+		hw->dtc_idx = CMN_DT_NUM_COUNTERS;
+		hw->dtcs_used = 1U << i;
+
+		if (flags & PERF_EF_START)
+			arm_cmn_event_start(event, 0);
+		return 0;
+	}
+
+	/* Grab a free global counter first... */
+	dtc_idx = 0;
+	while (dtc->counters[dtc_idx])
+		if (++dtc_idx == CMN_DT_NUM_COUNTERS)
+			return -ENOSPC;
+
+	hw->dtc_idx = dtc_idx;
+
+	/* ...then the local counters to feed it. */
+	for_each_hw_dn(hw, dn, i) {
+		struct arm_cmn_node *xp = arm_cmn_node_to_xp(dn);
+		unsigned int dtm_idx, shift;
+		u64 reg;
+
+		dtm_idx = 0;
+		while (xp->pmu_config_low & CMN__PMEVCNT_PAIRED(dtm_idx))
+			if (++dtm_idx == CMN_DTM_NUM_COUNTERS)
+				goto free_dtms;
+
+		if (type == CMN_TYPE_XP) {
+			input_sel = CMN__PMEVCNT0_INPUT_SEL_XP + dtm_idx;
+		} else if (type == CMN_TYPE_WP) {
+			int tmp, wp_idx = arm_cmn_wp_idx(event);
+			u32 cfg = arm_cmn_wp_config(event);
+
+			if (dn->wp_event[wp_idx] >= 0)
+				goto free_dtms;
+
+			tmp = dn->wp_event[wp_idx ^ 1];
+			if (tmp >= 0 && CMN_EVENT_WP_COMBINE(event) !=
+					CMN_EVENT_WP_COMBINE(dtc->counters[tmp]))
+				goto free_dtms;
+
+			input_sel = CMN__PMEVCNT0_INPUT_SEL_WP + wp_idx;
+			dn->wp_event[wp_idx] = dtc_idx;
+			writel_relaxed(cfg, dn->pmu_base + CMN_DTM_WPn_CONFIG(wp_idx));
+		} else {
+			unsigned int port = CMN_NODEID_PID(dn->id);
+			unsigned int dev = CMN_NODEID_DEVID(dn->id);
+
+			input_sel = CMN__PMEVCNT0_INPUT_SEL_DEV + dtm_idx +
+				    (port << 4) + (dev << 2);
+
+			if (arm_cmn_is_occup_event(type, CMN_EVENT_EVENTID(event))) {
+				int occupid = CMN_EVENT_OCCUPID(event);
+
+				if (dn->occupid_count == 0) {
+					dn->occupid_val = occupid;
+					writel_relaxed(occupid,
+						       dn->pmu_base + CMN_PMU_EVENT_SEL + 4);
+				} else if (dn->occupid_val != occupid) {
+					goto free_dtms;
+				}
+				dn->occupid_count++;
+			}
+		}
+
+		arm_cmn_set_index(hw->dtm_idx, i, dtm_idx);
+
+		xp->input_sel[dtm_idx] = input_sel;
+		shift = CMN__PMEVCNTn_GLOBAL_NUM_SHIFT(dtm_idx);
+		xp->pmu_config_low &= ~(CMN__PMEVCNT0_GLOBAL_NUM << shift);
+		xp->pmu_config_low |= FIELD_PREP(CMN__PMEVCNT0_GLOBAL_NUM, dtc_idx) << shift;
+		xp->pmu_config_low |= CMN__PMEVCNT_PAIRED(dtm_idx);
+		reg = (u64)le32_to_cpu(xp->pmu_config_high) << 32 | xp->pmu_config_low;
+		writeq_relaxed(reg, xp->pmu_base + CMN_DTM_PMU_CONFIG);
+	}
+
+	/* Go go go! */
+	arm_cmn_init_counter(event);
+
+	if (flags & PERF_EF_START)
+		arm_cmn_event_start(event, 0);
+
+	return 0;
+
+free_dtms:
+	arm_cmn_event_clear(cmn, event, i);
+	return -ENOSPC;
+}
+
+static void arm_cmn_event_del(struct perf_event *event, int flags)
+{
+	struct arm_cmn *cmn = to_cmn(event->pmu);
+	struct arm_cmn_hw_event *hw = to_cmn_hw(event);
+	enum cmn_node_type type = CMN_EVENT_TYPE(event);
+
+	arm_cmn_event_stop(event, PERF_EF_UPDATE);
+
+	if (type == CMN_TYPE_DTC)
+		cmn->dtc[__ffs(hw->dtcs_used)].cycles = NULL;
+	else
+		arm_cmn_event_clear(cmn, event, hw->num_dns);
+}
+
+/*
+ * We stop the PMU for both add and read, to avoid skew across DTM counters.
+ * In theory we could use snapshots to read without stopping, but then it
+ * becomes a lot trickier to deal with overlow and racing against interrupts,
+ * plus it seems they don't work properly on some hardware anyway :(
+ */
+static void arm_cmn_start_txn(struct pmu *pmu, unsigned int flags)
+{
+	arm_cmn_set_state(to_cmn(pmu), CMN_STATE_TXN);
+}
+
+static void arm_cmn_end_txn(struct pmu *pmu)
+{
+	arm_cmn_clear_state(to_cmn(pmu), CMN_STATE_TXN);
+}
+
+static int arm_cmn_commit_txn(struct pmu *pmu)
+{
+	arm_cmn_end_txn(pmu);
+	return 0;
+}
+
+static int arm_cmn_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_cmn *cmn;
+	unsigned int i, target;
+
+	cmn = hlist_entry_safe(node, struct arm_cmn, cpuhp_node);
+	if (cpu != cmn->cpu)
+		return 0;
+
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	perf_pmu_migrate_context(&cmn->pmu, cpu, target);
+	for (i = 0; i < cmn->num_dtcs; i++)
+		irq_set_affinity_hint(cmn->dtc[i].irq, cpumask_of(target));
+	cmn->cpu = target;
+	return 0;
+}
+
+static irqreturn_t arm_cmn_handle_irq(int irq, void *dev_id)
+{
+	struct arm_cmn_dtc *dtc = dev_id;
+	irqreturn_t ret = IRQ_NONE;
+
+	for (;;) {
+		u32 status = readl_relaxed(dtc->base + CMN_DT_PMOVSR);
+		u64 delta;
+		int i;
+
+		for (i = 0; i < CMN_DT_NUM_COUNTERS; i++) {
+			if (status & (1U << i)) {
+				ret = IRQ_HANDLED;
+				if (WARN_ON(!dtc->counters[i]))
+					continue;
+				delta = (u64)arm_cmn_read_counter(dtc, i) << 16;
+				local64_add(delta, &dtc->counters[i]->count);
+			}
+		}
+
+		if (status & (1U << CMN_DT_NUM_COUNTERS)) {
+			ret = IRQ_HANDLED;
+			if (dtc->cc_active && !WARN_ON(!dtc->cycles)) {
+				delta = arm_cmn_read_cc(dtc);
+				local64_add(delta, &dtc->cycles->count);
+			}
+		}
+
+		writel_relaxed(status, dtc->base + CMN_DT_PMOVSR_CLR);
+
+		if (!dtc->irq_friend)
+			return ret;
+		dtc += dtc->irq_friend;
+	}
+}
+
+/* We can reasonably accommodate DTCs of the same CMN sharing IRQs */
+static int arm_cmn_init_irqs(struct arm_cmn *cmn)
+{
+	int i, j, irq, err;
+
+	for (i = 0; i < cmn->num_dtcs; i++) {
+		irq = cmn->dtc[i].irq;
+		for (j = i; j--; ) {
+			if (cmn->dtc[j].irq == irq) {
+				cmn->dtc[j].irq_friend = i - j;
+				goto next;
+			}
+		}
+		err = devm_request_irq(cmn->dev, irq, arm_cmn_handle_irq,
+				       IRQF_NOBALANCING | IRQF_NO_THREAD,
+				       dev_name(cmn->dev), &cmn->dtc[i]);
+		if (err)
+			return err;
+
+		err = irq_set_affinity_hint(irq, cpumask_of(cmn->cpu));
+		if (err)
+			return err;
+	next:
+		; /* isn't C great? */
+	}
+	return 0;
+}
+
+static void arm_cmn_init_dtm(struct arm_cmn_node *xp)
+{
+	int i;
+
+	for (i = 0; i < 4; i++) {
+		xp->wp_event[i] = -1;
+		writeq_relaxed(0, xp->pmu_base + CMN_DTM_WPn_MASK(i));
+		writeq_relaxed(~0ULL, xp->pmu_base + CMN_DTM_WPn_VAL(i));
+	}
+	xp->pmu_config_low = CMN_DTM_PMU_CONFIG_PMU_EN;
+	xp->dtc = -1;
+}
+
+static int arm_cmn_init_dtc(struct arm_cmn *cmn, struct arm_cmn_node *dn, int idx)
+{
+	struct arm_cmn_dtc *dtc = cmn->dtc + idx;
+	struct arm_cmn_node *xp;
+
+	dtc->base = dn->pmu_base - CMN_PMU_OFFSET;
+	dtc->irq = platform_get_irq(to_platform_device(cmn->dev), idx);
+	if (dtc->irq < 0)
+		return dtc->irq;
+
+	writel_relaxed(CMN_DT_DTC_CTL_DT_EN, dtc->base + CMN_DT_DTC_CTL);
+	writel_relaxed(CMN_DT_PMCR_PMU_EN | CMN_DT_PMCR_OVFL_INTR_EN, dtc->base + CMN_DT_PMCR);
+	writeq_relaxed(0, dtc->base + CMN_DT_PMCCNTR);
+	writel_relaxed(0x1ff, dtc->base + CMN_DT_PMOVSR_CLR);
+
+	/* We do at least know that a DTC's XP must be in that DTC's domain */
+	xp = arm_cmn_node_to_xp(dn);
+	xp->dtc = idx;
+
+	return 0;
+}
+
+static int arm_cmn_node_cmp(const void *a, const void *b)
+{
+	const struct arm_cmn_node *dna = a, *dnb = b;
+	int cmp;
+
+	cmp = dna->type - dnb->type;
+	if (!cmp)
+		cmp = dna->logid - dnb->logid;
+	return cmp;
+}
+
+static int arm_cmn_init_dtcs(struct arm_cmn *cmn)
+{
+	struct arm_cmn_node *dn;
+	int dtc_idx = 0;
+
+	cmn->dtc = devm_kcalloc(cmn->dev, cmn->num_dtcs, sizeof(cmn->dtc[0]), GFP_KERNEL);
+	if (!cmn->dtc)
+		return -ENOMEM;
+
+	sort(cmn->dns, cmn->num_dns, sizeof(cmn->dns[0]), arm_cmn_node_cmp, NULL);
+
+	cmn->xps = arm_cmn_node(cmn, CMN_TYPE_XP);
+
+	for (dn = cmn->dns; dn < cmn->dns + cmn->num_dns; dn++) {
+		if (dn->type != CMN_TYPE_XP)
+			arm_cmn_init_node_to_xp(cmn, dn);
+		else if (cmn->num_dtcs == 1)
+			dn->dtc = 0;
+
+		if (dn->type == CMN_TYPE_DTC)
+			arm_cmn_init_dtc(cmn, dn, dtc_idx++);
+
+		/* To the PMU, RN-Ds don't add anything over RN-Is, so smoosh them together */
+		if (dn->type == CMN_TYPE_RND)
+			dn->type = CMN_TYPE_RNI;
+	}
+
+	arm_cmn_set_state(cmn, CMN_STATE_DISABLED);
+
+	return 0;
+}
+
+static void arm_cmn_init_node_info(struct arm_cmn *cmn, u32 offset, struct arm_cmn_node *node)
+{
+	int level;
+	u64 reg = readq_relaxed(cmn->base + offset + CMN_NODE_INFO);
+
+	node->type = FIELD_GET(CMN_NI_NODE_TYPE, reg);
+	node->id = FIELD_GET(CMN_NI_NODE_ID, reg);
+	node->logid = FIELD_GET(CMN_NI_LOGICAL_ID, reg);
+
+	node->pmu_base = cmn->base + offset + CMN_PMU_OFFSET;
+
+	if (node->type == CMN_TYPE_CFG)
+		level = 0;
+	else if (node->type == CMN_TYPE_XP)
+		level = 1;
+	else
+		level = 2;
+
+	dev_dbg(cmn->dev, "node%*c%#06hx%*ctype:%-#6x id:%-4hd off:%#x\n",
+			(level * 2) + 1, ' ', node->id, 5 - (level * 2), ' ',
+			node->type, node->logid, offset);
+}
+
+static int arm_cmn_discover(struct arm_cmn *cmn, unsigned int rgn_offset)
+{
+	void __iomem *cfg_region;
+	struct arm_cmn_node cfg, *dn;
+	u16 child_count, child_poff;
+	u32 xp_offset[CMN_MAX_XPS];
+	u64 reg;
+	int i, j;
+
+	cfg_region = cmn->base + rgn_offset;
+	reg = readl_relaxed(cfg_region + CMN_CFGM_PERIPH_ID_2);
+	cmn->rev = FIELD_GET(CMN_CFGM_PID2_REVISION, reg);
+	dev_dbg(cmn->dev, "periph_id_2 revision: %d\n", cmn->rev);
+
+	arm_cmn_init_node_info(cmn, rgn_offset, &cfg);
+	if (cfg.type != CMN_TYPE_CFG)
+		return -ENODEV;
+
+	reg = readq_relaxed(cfg_region + CMN_CHILD_INFO);
+	child_count = FIELD_GET(CMN_CI_CHILD_COUNT, reg);
+	child_poff = FIELD_GET(CMN_CI_CHILD_PTR_OFFSET, reg);
+
+	cmn->num_xps = child_count;
+	cmn->num_dns = cmn->num_xps;
+
+	/* Pass 1: visit the XPs, enumerate their children */
+	for (i = 0; i < cmn->num_xps; i++) {
+		reg = readq_relaxed(cfg_region + child_poff + i * 8);
+		xp_offset[i] = reg & CMN_CHILD_NODE_ADDR;
+
+		reg = readq_relaxed(cmn->base + xp_offset[i] + CMN_CHILD_INFO);
+		cmn->num_dns += FIELD_GET(CMN_CI_CHILD_COUNT, reg);
+	}
+
+	/* Cheeky +1 to help terminate pointer-based iteration */
+	cmn->dns = devm_kcalloc(cmn->dev, cmn->num_dns + 1,
+				sizeof(*cmn->dns), GFP_KERNEL);
+	if (!cmn->dns)
+		return -ENOMEM;
+
+	/* Pass 2: now we can actually populate the nodes */
+	dn = cmn->dns;
+	for (i = 0; i < cmn->num_xps; i++) {
+		void __iomem *xp_region = cmn->base + xp_offset[i];
+		struct arm_cmn_node *xp = dn++;
+
+		arm_cmn_init_node_info(cmn, xp_offset[i], xp);
+		arm_cmn_init_dtm(xp);
+		/*
+		 * Thanks to the order in which XP logical IDs seem to be
+		 * assigned, we can handily infer the mesh X dimension by
+		 * looking out for the XP at (0,1) without needing to know
+		 * the exact node ID format, which we can later derive.
+		 */
+		if (xp->id == (1 << 3))
+			cmn->mesh_x = xp->logid;
+
+		reg = readq_relaxed(xp_region + CMN_CHILD_INFO);
+		child_count = FIELD_GET(CMN_CI_CHILD_COUNT, reg);
+		child_poff = FIELD_GET(CMN_CI_CHILD_PTR_OFFSET, reg);
+
+		for (j = 0; j < child_count; j++) {
+			reg = readq_relaxed(xp_region + child_poff + j * 8);
+			/*
+			 * Don't even try to touch anything external, since in general
+			 * we haven't a clue how to power up arbitrary CHI requesters.
+			 * As of CMN-600r1 these could only be RN-SAMs or CXLAs,
+			 * neither of which have any PMU events anyway.
+			 * (Actually, CXLAs do seem to have grown some events in r1p2,
+			 * but they don't go to regular XP DTMs, and they depend on
+			 * secure configuration which we can't easily deal with)
+			 */
+			if (reg & CMN_CHILD_NODE_EXTERNAL) {
+				dev_dbg(cmn->dev, "ignoring external node %llx\n", reg);
+				continue;
+			}
+
+			arm_cmn_init_node_info(cmn, reg & CMN_CHILD_NODE_ADDR, dn);
+
+			switch (dn->type) {
+			case CMN_TYPE_DTC:
+				cmn->num_dtcs++;
+				dn++;
+				break;
+			/* These guys have PMU events */
+			case CMN_TYPE_DVM:
+			case CMN_TYPE_HNI:
+			case CMN_TYPE_HNF:
+			case CMN_TYPE_SBSX:
+			case CMN_TYPE_RNI:
+			case CMN_TYPE_RND:
+			case CMN_TYPE_CXRA:
+			case CMN_TYPE_CXHA:
+				dn++;
+				break;
+			/* Nothing to see here */
+			case CMN_TYPE_RNSAM:
+			case CMN_TYPE_CXLA:
+				break;
+			/* Something has gone horribly wrong */
+			default:
+				dev_err(cmn->dev, "invalid device node type: 0x%x\n", dn->type);
+				return -ENODEV;
+			}
+		}
+	}
+
+	/* Correct for any nodes we skipped */
+	cmn->num_dns = dn - cmn->dns;
+
+	/*
+	 * If mesh_x wasn't set during discovery then we never saw
+	 * an XP at (0,1), thus we must have an Nx1 configuration.
+	 */
+	if (!cmn->mesh_x)
+		cmn->mesh_x = cmn->num_xps;
+	cmn->mesh_y = cmn->num_xps / cmn->mesh_x;
+
+	dev_dbg(cmn->dev, "mesh %dx%d, ID width %d\n",
+		cmn->mesh_x, cmn->mesh_y, arm_cmn_xyidbits(cmn));
+
+	return 0;
+}
+
+static int arm_cmn_acpi_probe(struct platform_device *pdev, struct arm_cmn *cmn)
+{
+	struct resource *cfg, *root;
+
+	cfg = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (!cfg)
+		return -EINVAL;
+
+	root = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	if (!root)
+		return -EINVAL;
+
+	if (!resource_contains(cfg, root))
+		swap(cfg, root);
+	/*
+	 * Note that devm_ioremap_resource() is dumb and won't let the platform
+	 * device claim cfg when the ACPI companion device has already claimed
+	 * root within it. But since they *are* already both claimed in the
+	 * appropriate name, we don't really need to do it again here anyway.
+	 */
+	cmn->base = devm_ioremap(cmn->dev, cfg->start, resource_size(cfg));
+	if (!cmn->base)
+		return -ENOMEM;
+
+	return root->start - cfg->start;
+}
+
+static int arm_cmn_of_probe(struct platform_device *pdev, struct arm_cmn *cmn)
+{
+	struct device_node *np = pdev->dev.of_node;
+	u32 rootnode;
+	int ret;
+
+	cmn->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(cmn->base))
+		return PTR_ERR(cmn->base);
+
+	ret = of_property_read_u32(np, "arm,root-node", &rootnode);
+	if (ret)
+		return ret;
+
+	return rootnode;
+}
+
+static int arm_cmn_probe(struct platform_device *pdev)
+{
+	struct arm_cmn *cmn;
+	const char *name;
+	static atomic_t id;
+	int err, rootnode;
+
+	cmn = devm_kzalloc(&pdev->dev, sizeof(*cmn), GFP_KERNEL);
+	if (!cmn)
+		return -ENOMEM;
+
+	cmn->dev = &pdev->dev;
+	platform_set_drvdata(pdev, cmn);
+
+	if (has_acpi_companion(cmn->dev))
+		rootnode = arm_cmn_acpi_probe(pdev, cmn);
+	else
+		rootnode = arm_cmn_of_probe(pdev, cmn);
+	if (rootnode < 0)
+		return rootnode;
+
+	err = arm_cmn_discover(cmn, rootnode);
+	if (err)
+		return err;
+
+	err = arm_cmn_init_dtcs(cmn);
+	if (err)
+		return err;
+
+	err = arm_cmn_init_irqs(cmn);
+	if (err)
+		return err;
+
+	cmn->cpu = raw_smp_processor_id();
+	cmn->pmu = (struct pmu) {
+		.module = THIS_MODULE,
+		.attr_groups = arm_cmn_attr_groups,
+		.capabilities = PERF_PMU_CAP_NO_EXCLUDE,
+		.task_ctx_nr = perf_invalid_context,
+		.pmu_enable = arm_cmn_pmu_enable,
+		.pmu_disable = arm_cmn_pmu_disable,
+		.event_init = arm_cmn_event_init,
+		.add = arm_cmn_event_add,
+		.del = arm_cmn_event_del,
+		.start = arm_cmn_event_start,
+		.stop = arm_cmn_event_stop,
+		.read = arm_cmn_event_read,
+		.start_txn = arm_cmn_start_txn,
+		.commit_txn = arm_cmn_commit_txn,
+		.cancel_txn = arm_cmn_end_txn,
+	};
+
+	name = devm_kasprintf(cmn->dev, GFP_KERNEL, "arm_cmn_%d", atomic_fetch_inc(&id));
+	if (!name)
+		return -ENOMEM;
+
+	err = cpuhp_state_add_instance(arm_cmn_hp_state, &cmn->cpuhp_node);
+	if (err)
+		return err;
+
+	err = perf_pmu_register(&cmn->pmu, name, -1);
+	if (err)
+		cpuhp_state_remove_instance(arm_cmn_hp_state, &cmn->cpuhp_node);
+	return err;
+}
+
+static int arm_cmn_remove(struct platform_device *pdev)
+{
+	struct arm_cmn *cmn = platform_get_drvdata(pdev);
+	int i;
+
+	writel_relaxed(0, cmn->dtc[0].base + CMN_DT_DTC_CTL);
+
+	perf_pmu_unregister(&cmn->pmu);
+	cpuhp_state_remove_instance(arm_cmn_hp_state, &cmn->cpuhp_node);
+
+	for (i = 0; i < cmn->num_dtcs; i++)
+		irq_set_affinity_hint(cmn->dtc[i].irq, NULL);
+
+	return 0;
+}
+
+#ifdef CONFIG_OF
+static const struct of_device_id arm_cmn_of_match[] = {
+	{ .compatible = "arm,cmn-600", },
+	{}
+};
+MODULE_DEVICE_TABLE(of, arm_cmn_of_match);
+#endif
+
+#ifdef CONFIG_ACPI
+static const struct acpi_device_id arm_cmn_acpi_match[] = {
+	{ "ARMHC600", },
+	{}
+};
+MODULE_DEVICE_TABLE(acpi, arm_cmn_acpi_match);
+#endif
+
+static struct platform_driver arm_cmn_driver = {
+	.driver = {
+		.name = "arm-cmn",
+		.of_match_table = of_match_ptr(arm_cmn_of_match),
+		.acpi_match_table = ACPI_PTR(arm_cmn_acpi_match),
+	},
+	.probe = arm_cmn_probe,
+	.remove = arm_cmn_remove,
+};
+
+static int __init arm_cmn_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
+				      "perf/arm/cmn:online", NULL,
+				      arm_cmn_pmu_offline_cpu);
+	if (ret < 0)
+		return ret;
+
+	arm_cmn_hp_state = ret;
+	ret = platform_driver_register(&arm_cmn_driver);
+	if (ret)
+		cpuhp_remove_multi_state(arm_cmn_hp_state);
+	return ret;
+}
+
+static void __exit arm_cmn_exit(void)
+{
+	platform_driver_unregister(&arm_cmn_driver);
+	cpuhp_remove_multi_state(arm_cmn_hp_state);
+}
+
+module_init(arm_cmn_init);
+module_exit(arm_cmn_exit);
+
+MODULE_AUTHOR("Robin Murphy <robin.murphy@arm.com>");
+MODULE_DESCRIPTION("Arm CMN-600 PMU driver");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/arm_dsu_pmu.c b/drivers/perf/arm_dsu_pmu.c
new file mode 100644
index 000000000..1db8eccc9
--- /dev/null
+++ b/drivers/perf/arm_dsu_pmu.c
@@ -0,0 +1,887 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * ARM DynamIQ Shared Unit (DSU) PMU driver
+ *
+ * Copyright (C) ARM Limited, 2017.
+ *
+ * Based on ARM CCI-PMU, ARMv8 PMU-v3 drivers.
+ */
+
+#define PMUNAME		"arm_dsu"
+#define DRVNAME		PMUNAME "_pmu"
+#define pr_fmt(fmt)	DRVNAME ": " fmt
+
+#include <linux/acpi.h>
+#include <linux/bitmap.h>
+#include <linux/bitops.h>
+#include <linux/bug.h>
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of_device.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/smp.h>
+#include <linux/sysfs.h>
+#include <linux/types.h>
+
+#include <asm/arm_dsu_pmu.h>
+#include <asm/local64.h>
+
+/* PMU event codes */
+#define DSU_PMU_EVT_CYCLES		0x11
+#define DSU_PMU_EVT_CHAIN		0x1e
+
+#define DSU_PMU_MAX_COMMON_EVENTS	0x40
+
+#define DSU_PMU_MAX_HW_CNTRS		32
+#define DSU_PMU_HW_COUNTER_MASK		(DSU_PMU_MAX_HW_CNTRS - 1)
+
+#define CLUSTERPMCR_E			BIT(0)
+#define CLUSTERPMCR_P			BIT(1)
+#define CLUSTERPMCR_C			BIT(2)
+#define CLUSTERPMCR_N_SHIFT		11
+#define CLUSTERPMCR_N_MASK		0x1f
+#define CLUSTERPMCR_IDCODE_SHIFT	16
+#define CLUSTERPMCR_IDCODE_MASK		0xff
+#define CLUSTERPMCR_IMP_SHIFT		24
+#define CLUSTERPMCR_IMP_MASK		0xff
+#define CLUSTERPMCR_RES_MASK		0x7e8
+#define CLUSTERPMCR_RES_VAL		0x40
+
+#define DSU_ACTIVE_CPU_MASK		0x0
+#define DSU_ASSOCIATED_CPU_MASK		0x1
+
+/*
+ * We use the index of the counters as they appear in the counter
+ * bit maps in the PMU registers (e.g CLUSTERPMSELR).
+ * i.e,
+ *	counter 0	- Bit 0
+ *	counter 1	- Bit 1
+ *	...
+ *	Cycle counter	- Bit 31
+ */
+#define DSU_PMU_IDX_CYCLE_COUNTER	31
+
+/* All event counters are 32bit, with a 64bit Cycle counter */
+#define DSU_PMU_COUNTER_WIDTH(idx)	\
+	(((idx) == DSU_PMU_IDX_CYCLE_COUNTER) ? 64 : 32)
+
+#define DSU_PMU_COUNTER_MASK(idx)	\
+	GENMASK_ULL((DSU_PMU_COUNTER_WIDTH((idx)) - 1), 0)
+
+#define DSU_EXT_ATTR(_name, _func, _config)		\
+	(&((struct dev_ext_attribute[]) {				\
+		{							\
+			.attr = __ATTR(_name, 0444, _func, NULL),	\
+			.var = (void *)_config				\
+		}							\
+	})[0].attr.attr)
+
+#define DSU_EVENT_ATTR(_name, _config)		\
+	DSU_EXT_ATTR(_name, dsu_pmu_sysfs_event_show, (unsigned long)_config)
+
+#define DSU_FORMAT_ATTR(_name, _config)		\
+	DSU_EXT_ATTR(_name, dsu_pmu_sysfs_format_show, (char *)_config)
+
+#define DSU_CPUMASK_ATTR(_name, _config)	\
+	DSU_EXT_ATTR(_name, dsu_pmu_cpumask_show, (unsigned long)_config)
+
+struct dsu_hw_events {
+	DECLARE_BITMAP(used_mask, DSU_PMU_MAX_HW_CNTRS);
+	struct perf_event	*events[DSU_PMU_MAX_HW_CNTRS];
+};
+
+/*
+ * struct dsu_pmu	- DSU PMU descriptor
+ *
+ * @pmu_lock		: Protects accesses to DSU PMU register from normal vs
+ *			  interrupt handler contexts.
+ * @hw_events		: Holds the event counter state.
+ * @associated_cpus	: CPUs attached to the DSU.
+ * @active_cpu		: CPU to which the PMU is bound for accesses.
+ * @cpuhp_node		: Node for CPU hotplug notifier link.
+ * @num_counters	: Number of event counters implemented by the PMU,
+ *			  excluding the cycle counter.
+ * @irq			: Interrupt line for counter overflow.
+ * @cpmceid_bitmap	: Bitmap for the availability of architected common
+ *			  events (event_code < 0x40).
+ */
+struct dsu_pmu {
+	struct pmu			pmu;
+	struct device			*dev;
+	raw_spinlock_t			pmu_lock;
+	struct dsu_hw_events		hw_events;
+	cpumask_t			associated_cpus;
+	cpumask_t			active_cpu;
+	struct hlist_node		cpuhp_node;
+	s8				num_counters;
+	int				irq;
+	DECLARE_BITMAP(cpmceid_bitmap, DSU_PMU_MAX_COMMON_EVENTS);
+};
+
+static unsigned long dsu_pmu_cpuhp_state;
+
+static inline struct dsu_pmu *to_dsu_pmu(struct pmu *pmu)
+{
+	return container_of(pmu, struct dsu_pmu, pmu);
+}
+
+static ssize_t dsu_pmu_sysfs_event_show(struct device *dev,
+					struct device_attribute *attr,
+					char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+					struct dev_ext_attribute, attr);
+	return snprintf(buf, PAGE_SIZE, "event=0x%lx\n",
+					 (unsigned long)eattr->var);
+}
+
+static ssize_t dsu_pmu_sysfs_format_show(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct dev_ext_attribute *eattr = container_of(attr,
+					struct dev_ext_attribute, attr);
+	return snprintf(buf, PAGE_SIZE, "%s\n", (char *)eattr->var);
+}
+
+static ssize_t dsu_pmu_cpumask_show(struct device *dev,
+				    struct device_attribute *attr,
+				    char *buf)
+{
+	struct pmu *pmu = dev_get_drvdata(dev);
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(pmu);
+	struct dev_ext_attribute *eattr = container_of(attr,
+					struct dev_ext_attribute, attr);
+	unsigned long mask_id = (unsigned long)eattr->var;
+	const cpumask_t *cpumask;
+
+	switch (mask_id) {
+	case DSU_ACTIVE_CPU_MASK:
+		cpumask = &dsu_pmu->active_cpu;
+		break;
+	case DSU_ASSOCIATED_CPU_MASK:
+		cpumask = &dsu_pmu->associated_cpus;
+		break;
+	default:
+		return 0;
+	}
+	return cpumap_print_to_pagebuf(true, buf, cpumask);
+}
+
+static struct attribute *dsu_pmu_format_attrs[] = {
+	DSU_FORMAT_ATTR(event, "config:0-31"),
+	NULL,
+};
+
+static const struct attribute_group dsu_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = dsu_pmu_format_attrs,
+};
+
+static struct attribute *dsu_pmu_event_attrs[] = {
+	DSU_EVENT_ATTR(cycles, 0x11),
+	DSU_EVENT_ATTR(bus_access, 0x19),
+	DSU_EVENT_ATTR(memory_error, 0x1a),
+	DSU_EVENT_ATTR(bus_cycles, 0x1d),
+	DSU_EVENT_ATTR(l3d_cache_allocate, 0x29),
+	DSU_EVENT_ATTR(l3d_cache_refill, 0x2a),
+	DSU_EVENT_ATTR(l3d_cache, 0x2b),
+	DSU_EVENT_ATTR(l3d_cache_wb, 0x2c),
+	NULL,
+};
+
+static umode_t
+dsu_pmu_event_attr_is_visible(struct kobject *kobj, struct attribute *attr,
+				int unused)
+{
+	struct pmu *pmu = dev_get_drvdata(kobj_to_dev(kobj));
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(pmu);
+	struct dev_ext_attribute *eattr = container_of(attr,
+					struct dev_ext_attribute, attr.attr);
+	unsigned long evt = (unsigned long)eattr->var;
+
+	return test_bit(evt, dsu_pmu->cpmceid_bitmap) ? attr->mode : 0;
+}
+
+static const struct attribute_group dsu_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = dsu_pmu_event_attrs,
+	.is_visible = dsu_pmu_event_attr_is_visible,
+};
+
+static struct attribute *dsu_pmu_cpumask_attrs[] = {
+	DSU_CPUMASK_ATTR(cpumask, DSU_ACTIVE_CPU_MASK),
+	DSU_CPUMASK_ATTR(associated_cpus, DSU_ASSOCIATED_CPU_MASK),
+	NULL,
+};
+
+static const struct attribute_group dsu_pmu_cpumask_attr_group = {
+	.attrs = dsu_pmu_cpumask_attrs,
+};
+
+static const struct attribute_group *dsu_pmu_attr_groups[] = {
+	&dsu_pmu_cpumask_attr_group,
+	&dsu_pmu_events_attr_group,
+	&dsu_pmu_format_attr_group,
+	NULL,
+};
+
+static int dsu_pmu_get_online_cpu_any_but(struct dsu_pmu *dsu_pmu, int cpu)
+{
+	struct cpumask online_supported;
+
+	cpumask_and(&online_supported,
+			 &dsu_pmu->associated_cpus, cpu_online_mask);
+	return cpumask_any_but(&online_supported, cpu);
+}
+
+static inline bool dsu_pmu_counter_valid(struct dsu_pmu *dsu_pmu, u32 idx)
+{
+	return (idx < dsu_pmu->num_counters) ||
+	       (idx == DSU_PMU_IDX_CYCLE_COUNTER);
+}
+
+static inline u64 dsu_pmu_read_counter(struct perf_event *event)
+{
+	u64 val;
+	unsigned long flags;
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+	int idx = event->hw.idx;
+
+	if (WARN_ON(!cpumask_test_cpu(smp_processor_id(),
+				 &dsu_pmu->associated_cpus)))
+		return 0;
+
+	if (!dsu_pmu_counter_valid(dsu_pmu, idx)) {
+		dev_err(event->pmu->dev,
+			"Trying reading invalid counter %d\n", idx);
+		return 0;
+	}
+
+	raw_spin_lock_irqsave(&dsu_pmu->pmu_lock, flags);
+	if (idx == DSU_PMU_IDX_CYCLE_COUNTER)
+		val = __dsu_pmu_read_pmccntr();
+	else
+		val = __dsu_pmu_read_counter(idx);
+	raw_spin_unlock_irqrestore(&dsu_pmu->pmu_lock, flags);
+
+	return val;
+}
+
+static void dsu_pmu_write_counter(struct perf_event *event, u64 val)
+{
+	unsigned long flags;
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+	int idx = event->hw.idx;
+
+	if (WARN_ON(!cpumask_test_cpu(smp_processor_id(),
+			 &dsu_pmu->associated_cpus)))
+		return;
+
+	if (!dsu_pmu_counter_valid(dsu_pmu, idx)) {
+		dev_err(event->pmu->dev,
+			"writing to invalid counter %d\n", idx);
+		return;
+	}
+
+	raw_spin_lock_irqsave(&dsu_pmu->pmu_lock, flags);
+	if (idx == DSU_PMU_IDX_CYCLE_COUNTER)
+		__dsu_pmu_write_pmccntr(val);
+	else
+		__dsu_pmu_write_counter(idx, val);
+	raw_spin_unlock_irqrestore(&dsu_pmu->pmu_lock, flags);
+}
+
+static int dsu_pmu_get_event_idx(struct dsu_hw_events *hw_events,
+				 struct perf_event *event)
+{
+	int idx;
+	unsigned long evtype = event->attr.config;
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+	unsigned long *used_mask = hw_events->used_mask;
+
+	if (evtype == DSU_PMU_EVT_CYCLES) {
+		if (test_and_set_bit(DSU_PMU_IDX_CYCLE_COUNTER, used_mask))
+			return -EAGAIN;
+		return DSU_PMU_IDX_CYCLE_COUNTER;
+	}
+
+	idx = find_first_zero_bit(used_mask, dsu_pmu->num_counters);
+	if (idx >= dsu_pmu->num_counters)
+		return -EAGAIN;
+	set_bit(idx, hw_events->used_mask);
+	return idx;
+}
+
+static void dsu_pmu_enable_counter(struct dsu_pmu *dsu_pmu, int idx)
+{
+	__dsu_pmu_counter_interrupt_enable(idx);
+	__dsu_pmu_enable_counter(idx);
+}
+
+static void dsu_pmu_disable_counter(struct dsu_pmu *dsu_pmu, int idx)
+{
+	__dsu_pmu_disable_counter(idx);
+	__dsu_pmu_counter_interrupt_disable(idx);
+}
+
+static inline void dsu_pmu_set_event(struct dsu_pmu *dsu_pmu,
+					struct perf_event *event)
+{
+	int idx = event->hw.idx;
+	unsigned long flags;
+
+	if (!dsu_pmu_counter_valid(dsu_pmu, idx)) {
+		dev_err(event->pmu->dev,
+			"Trying to set invalid counter %d\n", idx);
+		return;
+	}
+
+	raw_spin_lock_irqsave(&dsu_pmu->pmu_lock, flags);
+	__dsu_pmu_set_event(idx, event->hw.config_base);
+	raw_spin_unlock_irqrestore(&dsu_pmu->pmu_lock, flags);
+}
+
+static void dsu_pmu_event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev_count, new_count;
+
+	do {
+		/* We may also be called from the irq handler */
+		prev_count = local64_read(&hwc->prev_count);
+		new_count = dsu_pmu_read_counter(event);
+	} while (local64_cmpxchg(&hwc->prev_count, prev_count, new_count) !=
+			prev_count);
+	delta = (new_count - prev_count) & DSU_PMU_COUNTER_MASK(hwc->idx);
+	local64_add(delta, &event->count);
+}
+
+static void dsu_pmu_read(struct perf_event *event)
+{
+	dsu_pmu_event_update(event);
+}
+
+static inline u32 dsu_pmu_get_reset_overflow(void)
+{
+	return __dsu_pmu_get_reset_overflow();
+}
+
+/**
+ * dsu_pmu_set_event_period: Set the period for the counter.
+ *
+ * All DSU PMU event counters, except the cycle counter are 32bit
+ * counters. To handle cases of extreme interrupt latency, we program
+ * the counter with half of the max count for the counters.
+ */
+static void dsu_pmu_set_event_period(struct perf_event *event)
+{
+	int idx = event->hw.idx;
+	u64 val = DSU_PMU_COUNTER_MASK(idx) >> 1;
+
+	local64_set(&event->hw.prev_count, val);
+	dsu_pmu_write_counter(event, val);
+}
+
+static irqreturn_t dsu_pmu_handle_irq(int irq_num, void *dev)
+{
+	int i;
+	bool handled = false;
+	struct dsu_pmu *dsu_pmu = dev;
+	struct dsu_hw_events *hw_events = &dsu_pmu->hw_events;
+	unsigned long overflow;
+
+	overflow = dsu_pmu_get_reset_overflow();
+	if (!overflow)
+		return IRQ_NONE;
+
+	for_each_set_bit(i, &overflow, DSU_PMU_MAX_HW_CNTRS) {
+		struct perf_event *event = hw_events->events[i];
+
+		if (!event)
+			continue;
+		dsu_pmu_event_update(event);
+		dsu_pmu_set_event_period(event);
+		handled = true;
+	}
+
+	return IRQ_RETVAL(handled);
+}
+
+static void dsu_pmu_start(struct perf_event *event, int pmu_flags)
+{
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+
+	/* We always reprogram the counter */
+	if (pmu_flags & PERF_EF_RELOAD)
+		WARN_ON(!(event->hw.state & PERF_HES_UPTODATE));
+	dsu_pmu_set_event_period(event);
+	if (event->hw.idx != DSU_PMU_IDX_CYCLE_COUNTER)
+		dsu_pmu_set_event(dsu_pmu, event);
+	event->hw.state = 0;
+	dsu_pmu_enable_counter(dsu_pmu, event->hw.idx);
+}
+
+static void dsu_pmu_stop(struct perf_event *event, int pmu_flags)
+{
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+
+	if (event->hw.state & PERF_HES_STOPPED)
+		return;
+	dsu_pmu_disable_counter(dsu_pmu, event->hw.idx);
+	dsu_pmu_event_update(event);
+	event->hw.state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+}
+
+static int dsu_pmu_add(struct perf_event *event, int flags)
+{
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+	struct dsu_hw_events *hw_events = &dsu_pmu->hw_events;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+
+	if (WARN_ON_ONCE(!cpumask_test_cpu(smp_processor_id(),
+					   &dsu_pmu->associated_cpus)))
+		return -ENOENT;
+
+	idx = dsu_pmu_get_event_idx(hw_events, event);
+	if (idx < 0)
+		return idx;
+
+	hwc->idx = idx;
+	hw_events->events[idx] = event;
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+
+	if (flags & PERF_EF_START)
+		dsu_pmu_start(event, PERF_EF_RELOAD);
+
+	perf_event_update_userpage(event);
+	return 0;
+}
+
+static void dsu_pmu_del(struct perf_event *event, int flags)
+{
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+	struct dsu_hw_events *hw_events = &dsu_pmu->hw_events;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	dsu_pmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[idx] = NULL;
+	clear_bit(idx, hw_events->used_mask);
+	perf_event_update_userpage(event);
+}
+
+static void dsu_pmu_enable(struct pmu *pmu)
+{
+	u32 pmcr;
+	unsigned long flags;
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(pmu);
+
+	/* If no counters are added, skip enabling the PMU */
+	if (bitmap_empty(dsu_pmu->hw_events.used_mask, DSU_PMU_MAX_HW_CNTRS))
+		return;
+
+	raw_spin_lock_irqsave(&dsu_pmu->pmu_lock, flags);
+	pmcr = __dsu_pmu_read_pmcr();
+	pmcr |= CLUSTERPMCR_E;
+	__dsu_pmu_write_pmcr(pmcr);
+	raw_spin_unlock_irqrestore(&dsu_pmu->pmu_lock, flags);
+}
+
+static void dsu_pmu_disable(struct pmu *pmu)
+{
+	u32 pmcr;
+	unsigned long flags;
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(pmu);
+
+	raw_spin_lock_irqsave(&dsu_pmu->pmu_lock, flags);
+	pmcr = __dsu_pmu_read_pmcr();
+	pmcr &= ~CLUSTERPMCR_E;
+	__dsu_pmu_write_pmcr(pmcr);
+	raw_spin_unlock_irqrestore(&dsu_pmu->pmu_lock, flags);
+}
+
+static bool dsu_pmu_validate_event(struct pmu *pmu,
+				  struct dsu_hw_events *hw_events,
+				  struct perf_event *event)
+{
+	if (is_software_event(event))
+		return true;
+	/* Reject groups spanning multiple HW PMUs. */
+	if (event->pmu != pmu)
+		return false;
+	return dsu_pmu_get_event_idx(hw_events, event) >= 0;
+}
+
+/*
+ * Make sure the group of events can be scheduled at once
+ * on the PMU.
+ */
+static bool dsu_pmu_validate_group(struct perf_event *event)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	struct dsu_hw_events fake_hw;
+
+	if (event->group_leader == event)
+		return true;
+
+	memset(fake_hw.used_mask, 0, sizeof(fake_hw.used_mask));
+	if (!dsu_pmu_validate_event(event->pmu, &fake_hw, leader))
+		return false;
+	for_each_sibling_event(sibling, leader) {
+		if (!dsu_pmu_validate_event(event->pmu, &fake_hw, sibling))
+			return false;
+	}
+	return dsu_pmu_validate_event(event->pmu, &fake_hw, event);
+}
+
+static int dsu_pmu_event_init(struct perf_event *event)
+{
+	struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/* We don't support sampling */
+	if (is_sampling_event(event)) {
+		dev_dbg(dsu_pmu->pmu.dev, "Can't support sampling events\n");
+		return -EOPNOTSUPP;
+	}
+
+	/* We cannot support task bound events */
+	if (event->cpu < 0 || event->attach_state & PERF_ATTACH_TASK) {
+		dev_dbg(dsu_pmu->pmu.dev, "Can't support per-task counters\n");
+		return -EINVAL;
+	}
+
+	if (has_branch_stack(event)) {
+		dev_dbg(dsu_pmu->pmu.dev, "Can't support filtering\n");
+		return -EINVAL;
+	}
+
+	if (!cpumask_test_cpu(event->cpu, &dsu_pmu->associated_cpus)) {
+		dev_dbg(dsu_pmu->pmu.dev,
+			 "Requested cpu is not associated with the DSU\n");
+		return -EINVAL;
+	}
+	/*
+	 * Choose the current active CPU to read the events. We don't want
+	 * to migrate the event contexts, irq handling etc to the requested
+	 * CPU. As long as the requested CPU is within the same DSU, we
+	 * are fine.
+	 */
+	event->cpu = cpumask_first(&dsu_pmu->active_cpu);
+	if (event->cpu >= nr_cpu_ids)
+		return -EINVAL;
+	if (!dsu_pmu_validate_group(event))
+		return -EINVAL;
+
+	event->hw.config_base = event->attr.config;
+	return 0;
+}
+
+static struct dsu_pmu *dsu_pmu_alloc(struct platform_device *pdev)
+{
+	struct dsu_pmu *dsu_pmu;
+
+	dsu_pmu = devm_kzalloc(&pdev->dev, sizeof(*dsu_pmu), GFP_KERNEL);
+	if (!dsu_pmu)
+		return ERR_PTR(-ENOMEM);
+
+	raw_spin_lock_init(&dsu_pmu->pmu_lock);
+	/*
+	 * Initialise the number of counters to -1, until we probe
+	 * the real number on a connected CPU.
+	 */
+	dsu_pmu->num_counters = -1;
+	return dsu_pmu;
+}
+
+/**
+ * dsu_pmu_dt_get_cpus: Get the list of CPUs in the cluster
+ * from device tree.
+ */
+static int dsu_pmu_dt_get_cpus(struct device *dev, cpumask_t *mask)
+{
+	int i = 0, n, cpu;
+	struct device_node *cpu_node;
+
+	n = of_count_phandle_with_args(dev->of_node, "cpus", NULL);
+	if (n <= 0)
+		return -ENODEV;
+	for (; i < n; i++) {
+		cpu_node = of_parse_phandle(dev->of_node, "cpus", i);
+		if (!cpu_node)
+			break;
+		cpu = of_cpu_node_to_id(cpu_node);
+		of_node_put(cpu_node);
+		/*
+		 * We have to ignore the failures here and continue scanning
+		 * the list to handle cases where the nr_cpus could be capped
+		 * in the running kernel.
+		 */
+		if (cpu < 0)
+			continue;
+		cpumask_set_cpu(cpu, mask);
+	}
+	return 0;
+}
+
+/**
+ * dsu_pmu_acpi_get_cpus: Get the list of CPUs in the cluster
+ * from ACPI.
+ */
+static int dsu_pmu_acpi_get_cpus(struct device *dev, cpumask_t *mask)
+{
+#ifdef CONFIG_ACPI
+	int cpu;
+
+	/*
+	 * A dsu pmu node is inside a cluster parent node along with cpu nodes.
+	 * We need to find out all cpus that have the same parent with this pmu.
+	 */
+	for_each_possible_cpu(cpu) {
+		struct acpi_device *acpi_dev;
+		struct device *cpu_dev = get_cpu_device(cpu);
+
+		if (!cpu_dev)
+			continue;
+
+		acpi_dev = ACPI_COMPANION(cpu_dev);
+		if (acpi_dev &&
+			acpi_dev->parent == ACPI_COMPANION(dev)->parent)
+			cpumask_set_cpu(cpu, mask);
+	}
+#endif
+
+	return 0;
+}
+
+/*
+ * dsu_pmu_probe_pmu: Probe the PMU details on a CPU in the cluster.
+ */
+static void dsu_pmu_probe_pmu(struct dsu_pmu *dsu_pmu)
+{
+	u64 num_counters;
+	u32 cpmceid[2];
+
+	num_counters = (__dsu_pmu_read_pmcr() >> CLUSTERPMCR_N_SHIFT) &
+						CLUSTERPMCR_N_MASK;
+	/* We can only support up to 31 independent counters */
+	if (WARN_ON(num_counters > 31))
+		num_counters = 31;
+	dsu_pmu->num_counters = num_counters;
+	if (!dsu_pmu->num_counters)
+		return;
+	cpmceid[0] = __dsu_pmu_read_pmceid(0);
+	cpmceid[1] = __dsu_pmu_read_pmceid(1);
+	bitmap_from_arr32(dsu_pmu->cpmceid_bitmap, cpmceid,
+			  DSU_PMU_MAX_COMMON_EVENTS);
+}
+
+static void dsu_pmu_set_active_cpu(int cpu, struct dsu_pmu *dsu_pmu)
+{
+	cpumask_set_cpu(cpu, &dsu_pmu->active_cpu);
+	if (irq_set_affinity_hint(dsu_pmu->irq, &dsu_pmu->active_cpu))
+		pr_warn("Failed to set irq affinity to %d\n", cpu);
+}
+
+/*
+ * dsu_pmu_init_pmu: Initialise the DSU PMU configurations if
+ * we haven't done it already.
+ */
+static void dsu_pmu_init_pmu(struct dsu_pmu *dsu_pmu)
+{
+	if (dsu_pmu->num_counters == -1)
+		dsu_pmu_probe_pmu(dsu_pmu);
+	/* Reset the interrupt overflow mask */
+	dsu_pmu_get_reset_overflow();
+}
+
+static int dsu_pmu_device_probe(struct platform_device *pdev)
+{
+	int irq, rc;
+	struct dsu_pmu *dsu_pmu;
+	struct fwnode_handle *fwnode = dev_fwnode(&pdev->dev);
+	char *name;
+	static atomic_t pmu_idx = ATOMIC_INIT(-1);
+
+	dsu_pmu = dsu_pmu_alloc(pdev);
+	if (IS_ERR(dsu_pmu))
+		return PTR_ERR(dsu_pmu);
+
+	if (IS_ERR_OR_NULL(fwnode))
+		return -ENOENT;
+
+	if (is_of_node(fwnode))
+		rc = dsu_pmu_dt_get_cpus(&pdev->dev, &dsu_pmu->associated_cpus);
+	else if (is_acpi_device_node(fwnode))
+		rc = dsu_pmu_acpi_get_cpus(&pdev->dev, &dsu_pmu->associated_cpus);
+	else
+		return -ENOENT;
+
+	if (rc) {
+		dev_warn(&pdev->dev, "Failed to parse the CPUs\n");
+		return rc;
+	}
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return -EINVAL;
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "%s_%d",
+				PMUNAME, atomic_inc_return(&pmu_idx));
+	if (!name)
+		return -ENOMEM;
+	rc = devm_request_irq(&pdev->dev, irq, dsu_pmu_handle_irq,
+			      IRQF_NOBALANCING, name, dsu_pmu);
+	if (rc) {
+		dev_warn(&pdev->dev, "Failed to request IRQ %d\n", irq);
+		return rc;
+	}
+
+	dsu_pmu->irq = irq;
+	platform_set_drvdata(pdev, dsu_pmu);
+	rc = cpuhp_state_add_instance(dsu_pmu_cpuhp_state,
+						&dsu_pmu->cpuhp_node);
+	if (rc)
+		return rc;
+
+	dsu_pmu->pmu = (struct pmu) {
+		.task_ctx_nr	= perf_invalid_context,
+		.module		= THIS_MODULE,
+		.pmu_enable	= dsu_pmu_enable,
+		.pmu_disable	= dsu_pmu_disable,
+		.event_init	= dsu_pmu_event_init,
+		.add		= dsu_pmu_add,
+		.del		= dsu_pmu_del,
+		.start		= dsu_pmu_start,
+		.stop		= dsu_pmu_stop,
+		.read		= dsu_pmu_read,
+
+		.attr_groups	= dsu_pmu_attr_groups,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	rc = perf_pmu_register(&dsu_pmu->pmu, name, -1);
+	if (rc) {
+		cpuhp_state_remove_instance(dsu_pmu_cpuhp_state,
+						 &dsu_pmu->cpuhp_node);
+		irq_set_affinity_hint(dsu_pmu->irq, NULL);
+	}
+
+	return rc;
+}
+
+static int dsu_pmu_device_remove(struct platform_device *pdev)
+{
+	struct dsu_pmu *dsu_pmu = platform_get_drvdata(pdev);
+
+	perf_pmu_unregister(&dsu_pmu->pmu);
+	cpuhp_state_remove_instance(dsu_pmu_cpuhp_state, &dsu_pmu->cpuhp_node);
+	irq_set_affinity_hint(dsu_pmu->irq, NULL);
+
+	return 0;
+}
+
+static const struct of_device_id dsu_pmu_of_match[] = {
+	{ .compatible = "arm,dsu-pmu", },
+	{},
+};
+MODULE_DEVICE_TABLE(of, dsu_pmu_of_match);
+
+#ifdef CONFIG_ACPI
+static const struct acpi_device_id dsu_pmu_acpi_match[] = {
+	{ "ARMHD500", 0},
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, dsu_pmu_acpi_match);
+#endif
+
+static struct platform_driver dsu_pmu_driver = {
+	.driver = {
+		.name	= DRVNAME,
+		.of_match_table = of_match_ptr(dsu_pmu_of_match),
+		.acpi_match_table = ACPI_PTR(dsu_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = dsu_pmu_device_probe,
+	.remove = dsu_pmu_device_remove,
+};
+
+static int dsu_pmu_cpu_online(unsigned int cpu, struct hlist_node *node)
+{
+	struct dsu_pmu *dsu_pmu = hlist_entry_safe(node, struct dsu_pmu,
+						   cpuhp_node);
+
+	if (!cpumask_test_cpu(cpu, &dsu_pmu->associated_cpus))
+		return 0;
+
+	/* If the PMU is already managed, there is nothing to do */
+	if (!cpumask_empty(&dsu_pmu->active_cpu))
+		return 0;
+
+	dsu_pmu_init_pmu(dsu_pmu);
+	dsu_pmu_set_active_cpu(cpu, dsu_pmu);
+
+	return 0;
+}
+
+static int dsu_pmu_cpu_teardown(unsigned int cpu, struct hlist_node *node)
+{
+	int dst;
+	struct dsu_pmu *dsu_pmu = hlist_entry_safe(node, struct dsu_pmu,
+						   cpuhp_node);
+
+	if (!cpumask_test_and_clear_cpu(cpu, &dsu_pmu->active_cpu))
+		return 0;
+
+	dst = dsu_pmu_get_online_cpu_any_but(dsu_pmu, cpu);
+	/* If there are no active CPUs in the DSU, leave IRQ disabled */
+	if (dst >= nr_cpu_ids) {
+		irq_set_affinity_hint(dsu_pmu->irq, NULL);
+		return 0;
+	}
+
+	perf_pmu_migrate_context(&dsu_pmu->pmu, cpu, dst);
+	dsu_pmu_set_active_cpu(dst, dsu_pmu);
+
+	return 0;
+}
+
+static int __init dsu_pmu_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
+					DRVNAME,
+					dsu_pmu_cpu_online,
+					dsu_pmu_cpu_teardown);
+	if (ret < 0)
+		return ret;
+	dsu_pmu_cpuhp_state = ret;
+	ret = platform_driver_register(&dsu_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(dsu_pmu_cpuhp_state);
+
+	return ret;
+}
+
+static void __exit dsu_pmu_exit(void)
+{
+	platform_driver_unregister(&dsu_pmu_driver);
+	cpuhp_remove_multi_state(dsu_pmu_cpuhp_state);
+}
+
+module_init(dsu_pmu_init);
+module_exit(dsu_pmu_exit);
+
+MODULE_DESCRIPTION("Perf driver for ARM DynamIQ Shared Unit");
+MODULE_AUTHOR("Suzuki K Poulose <suzuki.poulose@arm.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
new file mode 100644
index 000000000..7fd11ef5c
--- /dev/null
+++ b/drivers/perf/arm_pmu.c
@@ -0,0 +1,1007 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#undef DEBUG
+
+/*
+ * ARM performance counter 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 sparc64 perf event code, which is in turn based
+ * on the x86 code.
+ */
+#define pr_fmt(fmt) "hw perfevents: " fmt
+
+#include <linux/bitmap.h>
+#include <linux/cpumask.h>
+#include <linux/cpu_pm.h>
+#include <linux/export.h>
+#include <linux/kernel.h>
+#include <linux/perf/arm_pmu.h>
+#include <linux/slab.h>
+#include <linux/sched/clock.h>
+#include <linux/spinlock.h>
+#include <linux/irq.h>
+#include <linux/irqdesc.h>
+
+#include <asm/irq_regs.h>
+
+static int armpmu_count_irq_users(const int irq);
+
+struct pmu_irq_ops {
+	void (*enable_pmuirq)(unsigned int irq);
+	void (*disable_pmuirq)(unsigned int irq);
+	void (*free_pmuirq)(unsigned int irq, int cpu, void __percpu *devid);
+};
+
+static void armpmu_free_pmuirq(unsigned int irq, int cpu, void __percpu *devid)
+{
+	free_irq(irq, per_cpu_ptr(devid, cpu));
+}
+
+static const struct pmu_irq_ops pmuirq_ops = {
+	.enable_pmuirq = enable_irq,
+	.disable_pmuirq = disable_irq_nosync,
+	.free_pmuirq = armpmu_free_pmuirq
+};
+
+static void armpmu_free_pmunmi(unsigned int irq, int cpu, void __percpu *devid)
+{
+	free_nmi(irq, per_cpu_ptr(devid, cpu));
+}
+
+static const struct pmu_irq_ops pmunmi_ops = {
+	.enable_pmuirq = enable_nmi,
+	.disable_pmuirq = disable_nmi_nosync,
+	.free_pmuirq = armpmu_free_pmunmi
+};
+
+static void armpmu_enable_percpu_pmuirq(unsigned int irq)
+{
+	enable_percpu_irq(irq, IRQ_TYPE_NONE);
+}
+
+static void armpmu_free_percpu_pmuirq(unsigned int irq, int cpu,
+				   void __percpu *devid)
+{
+	if (armpmu_count_irq_users(irq) == 1)
+		free_percpu_irq(irq, devid);
+}
+
+static const struct pmu_irq_ops percpu_pmuirq_ops = {
+	.enable_pmuirq = armpmu_enable_percpu_pmuirq,
+	.disable_pmuirq = disable_percpu_irq,
+	.free_pmuirq = armpmu_free_percpu_pmuirq
+};
+
+static void armpmu_enable_percpu_pmunmi(unsigned int irq)
+{
+	if (!prepare_percpu_nmi(irq))
+		enable_percpu_nmi(irq, IRQ_TYPE_NONE);
+}
+
+static void armpmu_disable_percpu_pmunmi(unsigned int irq)
+{
+	disable_percpu_nmi(irq);
+	teardown_percpu_nmi(irq);
+}
+
+static void armpmu_free_percpu_pmunmi(unsigned int irq, int cpu,
+				      void __percpu *devid)
+{
+	if (armpmu_count_irq_users(irq) == 1)
+		free_percpu_nmi(irq, devid);
+}
+
+static const struct pmu_irq_ops percpu_pmunmi_ops = {
+	.enable_pmuirq = armpmu_enable_percpu_pmunmi,
+	.disable_pmuirq = armpmu_disable_percpu_pmunmi,
+	.free_pmuirq = armpmu_free_percpu_pmunmi
+};
+
+static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu);
+static DEFINE_PER_CPU(int, cpu_irq);
+static DEFINE_PER_CPU(const struct pmu_irq_ops *, cpu_irq_ops);
+
+static bool has_nmi;
+
+static inline u64 arm_pmu_event_max_period(struct perf_event *event)
+{
+	if (event->hw.flags & ARMPMU_EVT_64BIT)
+		return GENMASK_ULL(63, 0);
+	else
+		return GENMASK_ULL(31, 0);
+}
+
+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;
+
+	if (!cache_map)
+		return -ENOENT;
+
+	ret = (int)(*cache_map)[cache_type][cache_op][cache_result];
+
+	if (ret == CACHE_OP_UNSUPPORTED)
+		return -ENOENT;
+
+	return ret;
+}
+
+static int
+armpmu_map_hw_event(const unsigned (*event_map)[PERF_COUNT_HW_MAX], u64 config)
+{
+	int mapping;
+
+	if (config >= PERF_COUNT_HW_MAX)
+		return -EINVAL;
+
+	if (!event_map)
+		return -ENOENT;
+
+	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);
+}
+
+int
+armpmu_map_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;
+	int type = 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:
+		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 arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	s64 left = local64_read(&hwc->period_left);
+	s64 period = hwc->sample_period;
+	u64 max_period;
+	int ret = 0;
+
+	max_period = arm_pmu_event_max_period(event);
+	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;
+	}
+
+	/*
+	 * Limit the maximum period to prevent the counter value
+	 * from overtaking the one we are about to program. In
+	 * effect we are reducing max_period to account for
+	 * interrupt latency (and we are being very conservative).
+	 */
+	if (left > (max_period >> 1))
+		left = (max_period >> 1);
+
+	local64_set(&hwc->prev_count, (u64)-left);
+
+	armpmu->write_counter(event, (u64)(-left) & max_period);
+
+	perf_event_update_userpage(event);
+
+	return ret;
+}
+
+u64 armpmu_event_update(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev_raw_count, new_raw_count;
+	u64 max_period = arm_pmu_event_max_period(event);
+
+again:
+	prev_raw_count = local64_read(&hwc->prev_count);
+	new_raw_count = armpmu->read_counter(event);
+
+	if (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+			     new_raw_count) != prev_raw_count)
+		goto again;
+
+	delta = (new_raw_count - prev_raw_count) & 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)
+{
+	armpmu_event_update(event);
+}
+
+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(event);
+		armpmu_event_update(event);
+		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);
+	armpmu->enable(event);
+}
+
+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 = this_cpu_ptr(armpmu->hw_events);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	armpmu_stop(event, PERF_EF_UPDATE);
+	hw_events->events[idx] = NULL;
+	armpmu->clear_event_idx(hw_events, event);
+	perf_event_update_userpage(event);
+	/* Clear the allocated counter */
+	hwc->idx = -1;
+}
+
+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 = this_cpu_ptr(armpmu->hw_events);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+
+	/* An event following a process won't be stopped earlier */
+	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
+		return -ENOENT;
+
+	/* If we don't have a space for the counter then finish early. */
+	idx = armpmu->get_event_idx(hw_events, event);
+	if (idx < 0)
+		return idx;
+
+	/*
+	 * 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(event);
+	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);
+
+	return 0;
+}
+
+static int
+validate_event(struct pmu *pmu, struct pmu_hw_events *hw_events,
+			       struct perf_event *event)
+{
+	struct arm_pmu *armpmu;
+
+	if (is_software_event(event))
+		return 1;
+
+	/*
+	 * Reject groups spanning multiple HW PMUs (e.g. CPU + CCI). The
+	 * core perf code won't check that the pmu->ctx == leader->ctx
+	 * until after pmu->event_init(event).
+	 */
+	if (event->pmu != pmu)
+		return 0;
+
+	if (event->state < PERF_EVENT_STATE_OFF)
+		return 1;
+
+	if (event->state == PERF_EVENT_STATE_OFF && !event->attr.enable_on_exec)
+		return 1;
+
+	armpmu = to_arm_pmu(event->pmu);
+	return armpmu->get_event_idx(hw_events, event) >= 0;
+}
+
+static int
+validate_group(struct perf_event *event)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	struct pmu_hw_events fake_pmu;
+
+	/*
+	 * Initialise the fake PMU. We only need to populate the
+	 * used_mask for the purposes of validation.
+	 */
+	memset(&fake_pmu.used_mask, 0, sizeof(fake_pmu.used_mask));
+
+	if (!validate_event(event->pmu, &fake_pmu, leader))
+		return -EINVAL;
+
+	if (event == leader)
+		return 0;
+
+	for_each_sibling_event(sibling, leader) {
+		if (!validate_event(event->pmu, &fake_pmu, sibling))
+			return -EINVAL;
+	}
+
+	if (!validate_event(event->pmu, &fake_pmu, event))
+		return -EINVAL;
+
+	return 0;
+}
+
+static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
+{
+	struct arm_pmu *armpmu;
+	int ret;
+	u64 start_clock, finish_clock;
+
+	/*
+	 * 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;
+	if (WARN_ON_ONCE(!armpmu))
+		return IRQ_NONE;
+
+	start_clock = sched_clock();
+	ret = armpmu->handle_irq(armpmu);
+	finish_clock = sched_clock();
+
+	perf_sample_event_took(finish_clock - start_clock);
+	return ret;
+}
+
+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;
+
+	hwc->flags = 0;
+	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)) {
+		pr_debug("ARM performance counters do not support "
+			 "mode exclusion\n");
+		return -EOPNOTSUPP;
+	}
+
+	/*
+	 * Store the event encoding into the config_base field.
+	 */
+	hwc->config_base	    |= (unsigned long)mapping;
+
+	if (!is_sampling_event(event)) {
+		/*
+		 * 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  = arm_pmu_event_max_period(event) >> 1;
+		hwc->last_period    = hwc->sample_period;
+		local64_set(&hwc->period_left, hwc->sample_period);
+	}
+
+	return validate_group(event);
+}
+
+static int armpmu_event_init(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+
+	/*
+	 * Reject CPU-affine events for CPUs that are of a different class to
+	 * that which this PMU handles. Process-following events (where
+	 * event->cpu == -1) can be migrated between CPUs, and thus we have to
+	 * reject them later (in armpmu_add) if they're scheduled on a
+	 * different class of CPU.
+	 */
+	if (event->cpu != -1 &&
+		!cpumask_test_cpu(event->cpu, &armpmu->supported_cpus))
+		return -ENOENT;
+
+	/* does not support taken branch sampling */
+	if (has_branch_stack(event))
+		return -EOPNOTSUPP;
+
+	if (armpmu->map_event(event) == -ENOENT)
+		return -ENOENT;
+
+	return __hw_perf_event_init(event);
+}
+
+static void armpmu_enable(struct pmu *pmu)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(pmu);
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
+	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);
+
+	/* For task-bound events we may be called on other CPUs */
+	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
+		return;
+
+	if (enabled)
+		armpmu->start(armpmu);
+}
+
+static void armpmu_disable(struct pmu *pmu)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(pmu);
+
+	/* For task-bound events we may be called on other CPUs */
+	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
+		return;
+
+	armpmu->stop(armpmu);
+}
+
+/*
+ * In heterogeneous systems, events are specific to a particular
+ * microarchitecture, and aren't suitable for another. Thus, only match CPUs of
+ * the same microarchitecture.
+ */
+static int armpmu_filter_match(struct perf_event *event)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
+	unsigned int cpu = smp_processor_id();
+	int ret;
+
+	ret = cpumask_test_cpu(cpu, &armpmu->supported_cpus);
+	if (ret && armpmu->filter_match)
+		return armpmu->filter_match(event);
+
+	return ret;
+}
+
+static ssize_t armpmu_cpumask_show(struct device *dev,
+				   struct device_attribute *attr, char *buf)
+{
+	struct arm_pmu *armpmu = to_arm_pmu(dev_get_drvdata(dev));
+	return cpumap_print_to_pagebuf(true, buf, &armpmu->supported_cpus);
+}
+
+static DEVICE_ATTR(cpus, S_IRUGO, armpmu_cpumask_show, NULL);
+
+static struct attribute *armpmu_common_attrs[] = {
+	&dev_attr_cpus.attr,
+	NULL,
+};
+
+static struct attribute_group armpmu_common_attr_group = {
+	.attrs = armpmu_common_attrs,
+};
+
+/* Set at runtime when we know what CPU type we are. */
+static struct arm_pmu *__oprofile_cpu_pmu;
+
+/*
+ * Despite the names, these two functions are CPU-specific and are used
+ * by the OProfile/perf code.
+ */
+const char *perf_pmu_name(void)
+{
+	if (!__oprofile_cpu_pmu)
+		return NULL;
+
+	return __oprofile_cpu_pmu->name;
+}
+EXPORT_SYMBOL_GPL(perf_pmu_name);
+
+int perf_num_counters(void)
+{
+	int max_events = 0;
+
+	if (__oprofile_cpu_pmu != NULL)
+		max_events = __oprofile_cpu_pmu->num_events;
+
+	return max_events;
+}
+EXPORT_SYMBOL_GPL(perf_num_counters);
+
+static int armpmu_count_irq_users(const int irq)
+{
+	int cpu, count = 0;
+
+	for_each_possible_cpu(cpu) {
+		if (per_cpu(cpu_irq, cpu) == irq)
+			count++;
+	}
+
+	return count;
+}
+
+static const struct pmu_irq_ops *armpmu_find_irq_ops(int irq)
+{
+	const struct pmu_irq_ops *ops = NULL;
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		if (per_cpu(cpu_irq, cpu) != irq)
+			continue;
+
+		ops = per_cpu(cpu_irq_ops, cpu);
+		if (ops)
+			break;
+	}
+
+	return ops;
+}
+
+void armpmu_free_irq(int irq, int cpu)
+{
+	if (per_cpu(cpu_irq, cpu) == 0)
+		return;
+	if (WARN_ON(irq != per_cpu(cpu_irq, cpu)))
+		return;
+
+	per_cpu(cpu_irq_ops, cpu)->free_pmuirq(irq, cpu, &cpu_armpmu);
+
+	per_cpu(cpu_irq, cpu) = 0;
+	per_cpu(cpu_irq_ops, cpu) = NULL;
+}
+
+int armpmu_request_irq(int irq, int cpu)
+{
+	int err = 0;
+	const irq_handler_t handler = armpmu_dispatch_irq;
+	const struct pmu_irq_ops *irq_ops;
+
+	if (!irq)
+		return 0;
+
+	if (!irq_is_percpu_devid(irq)) {
+		unsigned long irq_flags;
+
+		err = irq_force_affinity(irq, cpumask_of(cpu));
+
+		if (err && num_possible_cpus() > 1) {
+			pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
+				irq, cpu);
+			goto err_out;
+		}
+
+		irq_flags = IRQF_PERCPU |
+			    IRQF_NOBALANCING |
+			    IRQF_NO_THREAD;
+
+		irq_set_status_flags(irq, IRQ_NOAUTOEN);
+
+		err = request_nmi(irq, handler, irq_flags, "arm-pmu",
+				  per_cpu_ptr(&cpu_armpmu, cpu));
+
+		/* If cannot get an NMI, get a normal interrupt */
+		if (err) {
+			err = request_irq(irq, handler, irq_flags, "arm-pmu",
+					  per_cpu_ptr(&cpu_armpmu, cpu));
+			irq_ops = &pmuirq_ops;
+		} else {
+			has_nmi = true;
+			irq_ops = &pmunmi_ops;
+		}
+	} else if (armpmu_count_irq_users(irq) == 0) {
+		err = request_percpu_nmi(irq, handler, "arm-pmu", &cpu_armpmu);
+
+		/* If cannot get an NMI, get a normal interrupt */
+		if (err) {
+			err = request_percpu_irq(irq, handler, "arm-pmu",
+						 &cpu_armpmu);
+			irq_ops = &percpu_pmuirq_ops;
+		} else {
+			has_nmi= true;
+			irq_ops = &percpu_pmunmi_ops;
+		}
+	} else {
+		/* Per cpudevid irq was already requested by another CPU */
+		irq_ops = armpmu_find_irq_ops(irq);
+
+		if (WARN_ON(!irq_ops))
+			err = -EINVAL;
+	}
+
+	if (err)
+		goto err_out;
+
+	per_cpu(cpu_irq, cpu) = irq;
+	per_cpu(cpu_irq_ops, cpu) = irq_ops;
+	return 0;
+
+err_out:
+	pr_err("unable to request IRQ%d for ARM PMU counters\n", irq);
+	return err;
+}
+
+static int armpmu_get_cpu_irq(struct arm_pmu *pmu, int cpu)
+{
+	struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
+	return per_cpu(hw_events->irq, cpu);
+}
+
+/*
+ * 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 arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
+	int irq;
+
+	if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
+		return 0;
+	if (pmu->reset)
+		pmu->reset(pmu);
+
+	per_cpu(cpu_armpmu, cpu) = pmu;
+
+	irq = armpmu_get_cpu_irq(pmu, cpu);
+	if (irq)
+		per_cpu(cpu_irq_ops, cpu)->enable_pmuirq(irq);
+
+	return 0;
+}
+
+static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
+	int irq;
+
+	if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
+		return 0;
+
+	irq = armpmu_get_cpu_irq(pmu, cpu);
+	if (irq)
+		per_cpu(cpu_irq_ops, cpu)->disable_pmuirq(irq);
+
+	per_cpu(cpu_armpmu, cpu) = NULL;
+
+	return 0;
+}
+
+#ifdef CONFIG_CPU_PM
+static void cpu_pm_pmu_setup(struct arm_pmu *armpmu, unsigned long cmd)
+{
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
+	struct perf_event *event;
+	int idx;
+
+	for (idx = 0; idx < armpmu->num_events; idx++) {
+		event = hw_events->events[idx];
+		if (!event)
+			continue;
+
+		switch (cmd) {
+		case CPU_PM_ENTER:
+			/*
+			 * Stop and update the counter
+			 */
+			armpmu_stop(event, PERF_EF_UPDATE);
+			break;
+		case CPU_PM_EXIT:
+		case CPU_PM_ENTER_FAILED:
+			 /*
+			  * Restore and enable the counter.
+			  * armpmu_start() indirectly calls
+			  *
+			  * perf_event_update_userpage()
+			  *
+			  * that requires RCU read locking to be functional,
+			  * wrap the call within RCU_NONIDLE to make the
+			  * RCU subsystem aware this cpu is not idle from
+			  * an RCU perspective for the armpmu_start() call
+			  * duration.
+			  */
+			RCU_NONIDLE(armpmu_start(event, PERF_EF_RELOAD));
+			break;
+		default:
+			break;
+		}
+	}
+}
+
+static int cpu_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
+			     void *v)
+{
+	struct arm_pmu *armpmu = container_of(b, struct arm_pmu, cpu_pm_nb);
+	struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
+	int enabled = bitmap_weight(hw_events->used_mask, armpmu->num_events);
+
+	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
+		return NOTIFY_DONE;
+
+	/*
+	 * Always reset the PMU registers on power-up even if
+	 * there are no events running.
+	 */
+	if (cmd == CPU_PM_EXIT && armpmu->reset)
+		armpmu->reset(armpmu);
+
+	if (!enabled)
+		return NOTIFY_OK;
+
+	switch (cmd) {
+	case CPU_PM_ENTER:
+		armpmu->stop(armpmu);
+		cpu_pm_pmu_setup(armpmu, cmd);
+		break;
+	case CPU_PM_EXIT:
+	case CPU_PM_ENTER_FAILED:
+		cpu_pm_pmu_setup(armpmu, cmd);
+		armpmu->start(armpmu);
+		break;
+	default:
+		return NOTIFY_DONE;
+	}
+
+	return NOTIFY_OK;
+}
+
+static int cpu_pm_pmu_register(struct arm_pmu *cpu_pmu)
+{
+	cpu_pmu->cpu_pm_nb.notifier_call = cpu_pm_pmu_notify;
+	return cpu_pm_register_notifier(&cpu_pmu->cpu_pm_nb);
+}
+
+static void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu)
+{
+	cpu_pm_unregister_notifier(&cpu_pmu->cpu_pm_nb);
+}
+#else
+static inline int cpu_pm_pmu_register(struct arm_pmu *cpu_pmu) { return 0; }
+static inline void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu) { }
+#endif
+
+static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
+{
+	int err;
+
+	err = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_STARTING,
+				       &cpu_pmu->node);
+	if (err)
+		goto out;
+
+	err = cpu_pm_pmu_register(cpu_pmu);
+	if (err)
+		goto out_unregister;
+
+	return 0;
+
+out_unregister:
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
+					    &cpu_pmu->node);
+out:
+	return err;
+}
+
+static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu)
+{
+	cpu_pm_pmu_unregister(cpu_pmu);
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
+					    &cpu_pmu->node);
+}
+
+static struct arm_pmu *__armpmu_alloc(gfp_t flags)
+{
+	struct arm_pmu *pmu;
+	int cpu;
+
+	pmu = kzalloc(sizeof(*pmu), flags);
+	if (!pmu) {
+		pr_info("failed to allocate PMU device!\n");
+		goto out;
+	}
+
+	pmu->hw_events = alloc_percpu_gfp(struct pmu_hw_events, flags);
+	if (!pmu->hw_events) {
+		pr_info("failed to allocate per-cpu PMU data.\n");
+		goto out_free_pmu;
+	}
+
+	pmu->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,
+		.filter_match	= armpmu_filter_match,
+		.attr_groups	= pmu->attr_groups,
+		/*
+		 * This is a CPU PMU potentially in a heterogeneous
+		 * configuration (e.g. big.LITTLE). This is not an uncore PMU,
+		 * and we have taken ctx sharing into account (e.g. with our
+		 * pmu::filter_match callback and pmu::event_init group
+		 * validation).
+		 */
+		.capabilities	= PERF_PMU_CAP_HETEROGENEOUS_CPUS,
+	};
+
+	pmu->attr_groups[ARMPMU_ATTR_GROUP_COMMON] =
+		&armpmu_common_attr_group;
+
+	for_each_possible_cpu(cpu) {
+		struct pmu_hw_events *events;
+
+		events = per_cpu_ptr(pmu->hw_events, cpu);
+		raw_spin_lock_init(&events->pmu_lock);
+		events->percpu_pmu = pmu;
+	}
+
+	return pmu;
+
+out_free_pmu:
+	kfree(pmu);
+out:
+	return NULL;
+}
+
+struct arm_pmu *armpmu_alloc(void)
+{
+	return __armpmu_alloc(GFP_KERNEL);
+}
+
+struct arm_pmu *armpmu_alloc_atomic(void)
+{
+	return __armpmu_alloc(GFP_ATOMIC);
+}
+
+
+void armpmu_free(struct arm_pmu *pmu)
+{
+	free_percpu(pmu->hw_events);
+	kfree(pmu);
+}
+
+int armpmu_register(struct arm_pmu *pmu)
+{
+	int ret;
+
+	ret = cpu_pmu_init(pmu);
+	if (ret)
+		return ret;
+
+	if (!pmu->set_event_filter)
+		pmu->pmu.capabilities |= PERF_PMU_CAP_NO_EXCLUDE;
+
+	ret = perf_pmu_register(&pmu->pmu, pmu->name, -1);
+	if (ret)
+		goto out_destroy;
+
+	if (!__oprofile_cpu_pmu)
+		__oprofile_cpu_pmu = pmu;
+
+	pr_info("enabled with %s PMU driver, %d counters available%s\n",
+		pmu->name, pmu->num_events,
+		has_nmi ? ", using NMIs" : "");
+
+	return 0;
+
+out_destroy:
+	cpu_pmu_destroy(pmu);
+	return ret;
+}
+
+static int arm_pmu_hp_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_STARTING,
+				      "perf/arm/pmu:starting",
+				      arm_perf_starting_cpu,
+				      arm_perf_teardown_cpu);
+	if (ret)
+		pr_err("CPU hotplug notifier for ARM PMU could not be registered: %d\n",
+		       ret);
+	return ret;
+}
+subsys_initcall(arm_pmu_hp_init);
diff --git a/drivers/perf/arm_pmu_acpi.c b/drivers/perf/arm_pmu_acpi.c
new file mode 100644
index 000000000..f5c7a845c
--- /dev/null
+++ b/drivers/perf/arm_pmu_acpi.c
@@ -0,0 +1,358 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * ACPI probing code for ARM performance counters.
+ *
+ * Copyright (C) 2017 ARM Ltd.
+ */
+
+#include <linux/acpi.h>
+#include <linux/cpumask.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/irqdesc.h>
+#include <linux/percpu.h>
+#include <linux/perf/arm_pmu.h>
+
+#include <asm/cputype.h>
+
+static DEFINE_PER_CPU(struct arm_pmu *, probed_pmus);
+static DEFINE_PER_CPU(int, pmu_irqs);
+
+static int arm_pmu_acpi_register_irq(int cpu)
+{
+	struct acpi_madt_generic_interrupt *gicc;
+	int gsi, trigger;
+
+	gicc = acpi_cpu_get_madt_gicc(cpu);
+
+	gsi = gicc->performance_interrupt;
+
+	/*
+	 * Per the ACPI spec, the MADT cannot describe a PMU that doesn't
+	 * have an interrupt. QEMU advertises this by using a GSI of zero,
+	 * which is not known to be valid on any hardware despite being
+	 * valid per the spec. Take the pragmatic approach and reject a
+	 * GSI of zero for now.
+	 */
+	if (!gsi)
+		return 0;
+
+	if (gicc->flags & ACPI_MADT_PERFORMANCE_IRQ_MODE)
+		trigger = ACPI_EDGE_SENSITIVE;
+	else
+		trigger = ACPI_LEVEL_SENSITIVE;
+
+	/*
+	 * Helpfully, the MADT GICC doesn't have a polarity flag for the
+	 * "performance interrupt". Luckily, on compliant GICs the polarity is
+	 * a fixed value in HW (for both SPIs and PPIs) that we cannot change
+	 * from SW.
+	 *
+	 * Here we pass in ACPI_ACTIVE_HIGH to keep the core code happy. This
+	 * may not match the real polarity, but that should not matter.
+	 *
+	 * Other interrupt controllers are not supported with ACPI.
+	 */
+	return acpi_register_gsi(NULL, gsi, trigger, ACPI_ACTIVE_HIGH);
+}
+
+static void arm_pmu_acpi_unregister_irq(int cpu)
+{
+	struct acpi_madt_generic_interrupt *gicc;
+	int gsi;
+
+	gicc = acpi_cpu_get_madt_gicc(cpu);
+
+	gsi = gicc->performance_interrupt;
+	if (gsi)
+		acpi_unregister_gsi(gsi);
+}
+
+#if IS_ENABLED(CONFIG_ARM_SPE_PMU)
+static struct resource spe_resources[] = {
+	{
+		/* irq */
+		.flags          = IORESOURCE_IRQ,
+	}
+};
+
+static struct platform_device spe_dev = {
+	.name = ARMV8_SPE_PDEV_NAME,
+	.id = -1,
+	.resource = spe_resources,
+	.num_resources = ARRAY_SIZE(spe_resources)
+};
+
+/*
+ * For lack of a better place, hook the normal PMU MADT walk
+ * and create a SPE device if we detect a recent MADT with
+ * a homogeneous PPI mapping.
+ */
+static void arm_spe_acpi_register_device(void)
+{
+	int cpu, hetid, irq, ret;
+	bool first = true;
+	u16 gsi = 0;
+
+	/*
+	 * Sanity check all the GICC tables for the same interrupt number.
+	 * For now, we only support homogeneous ACPI/SPE machines.
+	 */
+	for_each_possible_cpu(cpu) {
+		struct acpi_madt_generic_interrupt *gicc;
+
+		gicc = acpi_cpu_get_madt_gicc(cpu);
+		if (gicc->header.length < ACPI_MADT_GICC_SPE)
+			return;
+
+		if (first) {
+			gsi = gicc->spe_interrupt;
+			if (!gsi)
+				return;
+			hetid = find_acpi_cpu_topology_hetero_id(cpu);
+			first = false;
+		} else if ((gsi != gicc->spe_interrupt) ||
+			   (hetid != find_acpi_cpu_topology_hetero_id(cpu))) {
+			pr_warn("ACPI: SPE must be homogeneous\n");
+			return;
+		}
+	}
+
+	irq = acpi_register_gsi(NULL, gsi, ACPI_LEVEL_SENSITIVE,
+				ACPI_ACTIVE_HIGH);
+	if (irq < 0) {
+		pr_warn("ACPI: SPE Unable to register interrupt: %d\n", gsi);
+		return;
+	}
+
+	spe_resources[0].start = irq;
+	ret = platform_device_register(&spe_dev);
+	if (ret < 0) {
+		pr_warn("ACPI: SPE: Unable to register device\n");
+		acpi_unregister_gsi(gsi);
+	}
+}
+#else
+static inline void arm_spe_acpi_register_device(void)
+{
+}
+#endif /* CONFIG_ARM_SPE_PMU */
+
+static int arm_pmu_acpi_parse_irqs(void)
+{
+	int irq, cpu, irq_cpu, err;
+
+	for_each_possible_cpu(cpu) {
+		irq = arm_pmu_acpi_register_irq(cpu);
+		if (irq < 0) {
+			err = irq;
+			pr_warn("Unable to parse ACPI PMU IRQ for CPU%d: %d\n",
+				cpu, err);
+			goto out_err;
+		} else if (irq == 0) {
+			pr_warn("No ACPI PMU IRQ for CPU%d\n", cpu);
+		}
+
+		/*
+		 * Log and request the IRQ so the core arm_pmu code can manage
+		 * it. We'll have to sanity-check IRQs later when we associate
+		 * them with their PMUs.
+		 */
+		per_cpu(pmu_irqs, cpu) = irq;
+		armpmu_request_irq(irq, cpu);
+	}
+
+	return 0;
+
+out_err:
+	for_each_possible_cpu(cpu) {
+		irq = per_cpu(pmu_irqs, cpu);
+		if (!irq)
+			continue;
+
+		arm_pmu_acpi_unregister_irq(cpu);
+
+		/*
+		 * Blat all copies of the IRQ so that we only unregister the
+		 * corresponding GSI once (e.g. when we have PPIs).
+		 */
+		for_each_possible_cpu(irq_cpu) {
+			if (per_cpu(pmu_irqs, irq_cpu) == irq)
+				per_cpu(pmu_irqs, irq_cpu) = 0;
+		}
+	}
+
+	return err;
+}
+
+static struct arm_pmu *arm_pmu_acpi_find_alloc_pmu(void)
+{
+	unsigned long cpuid = read_cpuid_id();
+	struct arm_pmu *pmu;
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		pmu = per_cpu(probed_pmus, cpu);
+		if (!pmu || pmu->acpi_cpuid != cpuid)
+			continue;
+
+		return pmu;
+	}
+
+	pmu = armpmu_alloc_atomic();
+	if (!pmu) {
+		pr_warn("Unable to allocate PMU for CPU%d\n",
+			smp_processor_id());
+		return NULL;
+	}
+
+	pmu->acpi_cpuid = cpuid;
+
+	return pmu;
+}
+
+/*
+ * Check whether the new IRQ is compatible with those already associated with
+ * the PMU (e.g. we don't have mismatched PPIs).
+ */
+static bool pmu_irq_matches(struct arm_pmu *pmu, int irq)
+{
+	struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
+	int cpu;
+
+	if (!irq)
+		return true;
+
+	for_each_cpu(cpu, &pmu->supported_cpus) {
+		int other_irq = per_cpu(hw_events->irq, cpu);
+		if (!other_irq)
+			continue;
+
+		if (irq == other_irq)
+			continue;
+		if (!irq_is_percpu_devid(irq) && !irq_is_percpu_devid(other_irq))
+			continue;
+
+		pr_warn("mismatched PPIs detected\n");
+		return false;
+	}
+
+	return true;
+}
+
+/*
+ * This must run before the common arm_pmu hotplug logic, so that we can
+ * associate a CPU and its interrupt before the common code tries to manage the
+ * affinity and so on.
+ *
+ * Note that hotplug events are serialized, so we cannot race with another CPU
+ * coming up. The perf core won't open events while a hotplug event is in
+ * progress.
+ */
+static int arm_pmu_acpi_cpu_starting(unsigned int cpu)
+{
+	struct arm_pmu *pmu;
+	struct pmu_hw_events __percpu *hw_events;
+	int irq;
+
+	/* If we've already probed this CPU, we have nothing to do */
+	if (per_cpu(probed_pmus, cpu))
+		return 0;
+
+	irq = per_cpu(pmu_irqs, cpu);
+
+	pmu = arm_pmu_acpi_find_alloc_pmu();
+	if (!pmu)
+		return -ENOMEM;
+
+	per_cpu(probed_pmus, cpu) = pmu;
+
+	if (pmu_irq_matches(pmu, irq)) {
+		hw_events = pmu->hw_events;
+		per_cpu(hw_events->irq, cpu) = irq;
+	}
+
+	cpumask_set_cpu(cpu, &pmu->supported_cpus);
+
+	/*
+	 * Ideally, we'd probe the PMU here when we find the first matching
+	 * CPU. We can't do that for several reasons; see the comment in
+	 * arm_pmu_acpi_init().
+	 *
+	 * So for the time being, we're done.
+	 */
+	return 0;
+}
+
+int arm_pmu_acpi_probe(armpmu_init_fn init_fn)
+{
+	int pmu_idx = 0;
+	int cpu, ret;
+
+	/*
+	 * Initialise and register the set of PMUs which we know about right
+	 * now. Ideally we'd do this in arm_pmu_acpi_cpu_starting() so that we
+	 * could handle late hotplug, but this may lead to deadlock since we
+	 * might try to register a hotplug notifier instance from within a
+	 * hotplug notifier.
+	 *
+	 * There's also the problem of having access to the right init_fn,
+	 * without tying this too deeply into the "real" PMU driver.
+	 *
+	 * For the moment, as with the platform/DT case, we need at least one
+	 * of a PMU's CPUs to be online at probe time.
+	 */
+	for_each_possible_cpu(cpu) {
+		struct arm_pmu *pmu = per_cpu(probed_pmus, cpu);
+		char *base_name;
+
+		if (!pmu || pmu->name)
+			continue;
+
+		ret = init_fn(pmu);
+		if (ret == -ENODEV) {
+			/* PMU not handled by this driver, or not present */
+			continue;
+		} else if (ret) {
+			pr_warn("Unable to initialise PMU for CPU%d\n", cpu);
+			return ret;
+		}
+
+		base_name = pmu->name;
+		pmu->name = kasprintf(GFP_KERNEL, "%s_%d", base_name, pmu_idx++);
+		if (!pmu->name) {
+			pr_warn("Unable to allocate PMU name for CPU%d\n", cpu);
+			return -ENOMEM;
+		}
+
+		ret = armpmu_register(pmu);
+		if (ret) {
+			pr_warn("Failed to register PMU for CPU%d\n", cpu);
+			kfree(pmu->name);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static int arm_pmu_acpi_init(void)
+{
+	int ret;
+
+	if (acpi_disabled)
+		return 0;
+
+	arm_spe_acpi_register_device();
+
+	ret = arm_pmu_acpi_parse_irqs();
+	if (ret)
+		return ret;
+
+	ret = cpuhp_setup_state(CPUHP_AP_PERF_ARM_ACPI_STARTING,
+				"perf/arm/pmu_acpi:starting",
+				arm_pmu_acpi_cpu_starting, NULL);
+
+	return ret;
+}
+subsys_initcall(arm_pmu_acpi_init)
diff --git a/drivers/perf/arm_pmu_platform.c b/drivers/perf/arm_pmu_platform.c
new file mode 100644
index 000000000..2e1f3680d
--- /dev/null
+++ b/drivers/perf/arm_pmu_platform.c
@@ -0,0 +1,248 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * platform_device probing code for ARM performance counters.
+ *
+ * Copyright (C) 2009 picoChip Designs, Ltd., Jamie Iles
+ * Copyright (C) 2010 ARM Ltd., Will Deacon <will.deacon@arm.com>
+ */
+#define pr_fmt(fmt) "hw perfevents: " fmt
+#define dev_fmt pr_fmt
+
+#include <linux/bug.h>
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/errno.h>
+#include <linux/irq.h>
+#include <linux/irqdesc.h>
+#include <linux/kconfig.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/percpu.h>
+#include <linux/perf/arm_pmu.h>
+#include <linux/platform_device.h>
+#include <linux/printk.h>
+#include <linux/smp.h>
+
+static int probe_current_pmu(struct arm_pmu *pmu,
+			     const struct pmu_probe_info *info)
+{
+	int cpu = get_cpu();
+	unsigned int cpuid = read_cpuid_id();
+	int ret = -ENODEV;
+
+	pr_info("probing PMU on CPU %d\n", cpu);
+
+	for (; info->init != NULL; info++) {
+		if ((cpuid & info->mask) != info->cpuid)
+			continue;
+		ret = info->init(pmu);
+		break;
+	}
+
+	put_cpu();
+	return ret;
+}
+
+static int pmu_parse_percpu_irq(struct arm_pmu *pmu, int irq)
+{
+	int cpu, ret;
+	struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
+
+	ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
+	if (ret)
+		return ret;
+
+	for_each_cpu(cpu, &pmu->supported_cpus)
+		per_cpu(hw_events->irq, cpu) = irq;
+
+	return 0;
+}
+
+static bool pmu_has_irq_affinity(struct device_node *node)
+{
+	return !!of_find_property(node, "interrupt-affinity", NULL);
+}
+
+static int pmu_parse_irq_affinity(struct device_node *node, int i)
+{
+	struct device_node *dn;
+	int cpu;
+
+	/*
+	 * If we don't have an interrupt-affinity property, we guess irq
+	 * affinity matches our logical CPU order, as we used to assume.
+	 * This is fragile, so we'll warn in pmu_parse_irqs().
+	 */
+	if (!pmu_has_irq_affinity(node))
+		return i;
+
+	dn = of_parse_phandle(node, "interrupt-affinity", i);
+	if (!dn) {
+		pr_warn("failed to parse interrupt-affinity[%d] for %pOFn\n",
+			i, node);
+		return -EINVAL;
+	}
+
+	cpu = of_cpu_node_to_id(dn);
+	if (cpu < 0) {
+		pr_warn("failed to find logical CPU for %pOFn\n", dn);
+		cpu = nr_cpu_ids;
+	}
+
+	of_node_put(dn);
+
+	return cpu;
+}
+
+static int pmu_parse_irqs(struct arm_pmu *pmu)
+{
+	int i = 0, num_irqs;
+	struct platform_device *pdev = pmu->plat_device;
+	struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
+
+	num_irqs = platform_irq_count(pdev);
+	if (num_irqs < 0)
+		return dev_err_probe(&pdev->dev, num_irqs, "unable to count PMU IRQs\n");
+
+	/*
+	 * In this case we have no idea which CPUs are covered by the PMU.
+	 * To match our prior behaviour, we assume all CPUs in this case.
+	 */
+	if (num_irqs == 0) {
+		pr_warn("no irqs for PMU, sampling events not supported\n");
+		pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
+		cpumask_setall(&pmu->supported_cpus);
+		return 0;
+	}
+
+	if (num_irqs == 1) {
+		int irq = platform_get_irq(pdev, 0);
+		if ((irq > 0) && irq_is_percpu_devid(irq))
+			return pmu_parse_percpu_irq(pmu, irq);
+	}
+
+	if (nr_cpu_ids != 1 && !pmu_has_irq_affinity(pdev->dev.of_node)) {
+		pr_warn("no interrupt-affinity property for %pOF, guessing.\n",
+			pdev->dev.of_node);
+	}
+
+	for (i = 0; i < num_irqs; i++) {
+		int cpu, irq;
+
+		irq = platform_get_irq(pdev, i);
+		if (WARN_ON(irq <= 0))
+			continue;
+
+		if (irq_is_percpu_devid(irq)) {
+			pr_warn("multiple PPIs or mismatched SPI/PPI detected\n");
+			return -EINVAL;
+		}
+
+		cpu = pmu_parse_irq_affinity(pdev->dev.of_node, i);
+		if (cpu < 0)
+			return cpu;
+		if (cpu >= nr_cpu_ids)
+			continue;
+
+		if (per_cpu(hw_events->irq, cpu)) {
+			pr_warn("multiple PMU IRQs for the same CPU detected\n");
+			return -EINVAL;
+		}
+
+		per_cpu(hw_events->irq, cpu) = irq;
+		cpumask_set_cpu(cpu, &pmu->supported_cpus);
+	}
+
+	return 0;
+}
+
+static int armpmu_request_irqs(struct arm_pmu *armpmu)
+{
+	struct pmu_hw_events __percpu *hw_events = armpmu->hw_events;
+	int cpu, err = 0;
+
+	for_each_cpu(cpu, &armpmu->supported_cpus) {
+		int irq = per_cpu(hw_events->irq, cpu);
+		if (!irq)
+			continue;
+
+		err = armpmu_request_irq(irq, cpu);
+		if (err)
+			break;
+	}
+
+	return err;
+}
+
+static void armpmu_free_irqs(struct arm_pmu *armpmu)
+{
+	int cpu;
+	struct pmu_hw_events __percpu *hw_events = armpmu->hw_events;
+
+	for_each_cpu(cpu, &armpmu->supported_cpus) {
+		int irq = per_cpu(hw_events->irq, cpu);
+
+		armpmu_free_irq(irq, cpu);
+	}
+}
+
+int arm_pmu_device_probe(struct platform_device *pdev,
+			 const struct of_device_id *of_table,
+			 const struct pmu_probe_info *probe_table)
+{
+	const struct of_device_id *of_id;
+	armpmu_init_fn init_fn;
+	struct device_node *node = pdev->dev.of_node;
+	struct arm_pmu *pmu;
+	int ret = -ENODEV;
+
+	pmu = armpmu_alloc();
+	if (!pmu)
+		return -ENOMEM;
+
+	pmu->plat_device = pdev;
+
+	ret = pmu_parse_irqs(pmu);
+	if (ret)
+		goto out_free;
+
+	if (node && (of_id = of_match_node(of_table, pdev->dev.of_node))) {
+		init_fn = of_id->data;
+
+		pmu->secure_access = of_property_read_bool(pdev->dev.of_node,
+							   "secure-reg-access");
+
+		/* arm64 systems boot only as non-secure */
+		if (IS_ENABLED(CONFIG_ARM64) && pmu->secure_access) {
+			pr_warn("ignoring \"secure-reg-access\" property for arm64\n");
+			pmu->secure_access = false;
+		}
+
+		ret = init_fn(pmu);
+	} else if (probe_table) {
+		cpumask_setall(&pmu->supported_cpus);
+		ret = probe_current_pmu(pmu, probe_table);
+	}
+
+	if (ret) {
+		pr_info("%pOF: failed to probe PMU!\n", node);
+		goto out_free;
+	}
+
+	ret = armpmu_request_irqs(pmu);
+	if (ret)
+		goto out_free_irqs;
+
+	ret = armpmu_register(pmu);
+	if (ret)
+		goto out_free_irqs;
+
+	return 0;
+
+out_free_irqs:
+	armpmu_free_irqs(pmu);
+out_free:
+	pr_info("%pOF: failed to register PMU devices!\n", node);
+	armpmu_free(pmu);
+	return ret;
+}
diff --git a/drivers/perf/arm_smmuv3_pmu.c b/drivers/perf/arm_smmuv3_pmu.c
new file mode 100644
index 000000000..6ebe72b86
--- /dev/null
+++ b/drivers/perf/arm_smmuv3_pmu.c
@@ -0,0 +1,945 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * This driver adds support for perf events to use the Performance
+ * Monitor Counter Groups (PMCG) associated with an SMMUv3 node
+ * to monitor that node.
+ *
+ * SMMUv3 PMCG devices are named as smmuv3_pmcg_<phys_addr_page> where
+ * <phys_addr_page> is the physical page address of the SMMU PMCG wrapped
+ * to 4K boundary. For example, the PMCG at 0xff88840000 is named
+ * smmuv3_pmcg_ff88840
+ *
+ * Filtering by stream id is done by specifying filtering parameters
+ * with the event. options are:
+ *   filter_enable    - 0 = no filtering, 1 = filtering enabled
+ *   filter_span      - 0 = exact match, 1 = pattern match
+ *   filter_stream_id - pattern to filter against
+ *
+ * To match a partial StreamID where the X most-significant bits must match
+ * but the Y least-significant bits might differ, STREAMID is programmed
+ * with a value that contains:
+ *  STREAMID[Y - 1] == 0.
+ *  STREAMID[Y - 2:0] == 1 (where Y > 1).
+ * The remainder of implemented bits of STREAMID (X bits, from bit Y upwards)
+ * contain a value to match from the corresponding bits of event StreamID.
+ *
+ * Example: perf stat -e smmuv3_pmcg_ff88840/transaction,filter_enable=1,
+ *                    filter_span=1,filter_stream_id=0x42/ -a netperf
+ * Applies filter pattern 0x42 to transaction events, which means events
+ * matching stream ids 0x42 and 0x43 are counted. Further filtering
+ * information is available in the SMMU documentation.
+ *
+ * SMMU events are not attributable to a CPU, so task mode and sampling
+ * are not supported.
+ */
+
+#include <linux/acpi.h>
+#include <linux/acpi_iort.h>
+#include <linux/bitfield.h>
+#include <linux/bitops.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/msi.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/smp.h>
+#include <linux/sysfs.h>
+#include <linux/types.h>
+
+#define SMMU_PMCG_EVCNTR0               0x0
+#define SMMU_PMCG_EVCNTR(n, stride)     (SMMU_PMCG_EVCNTR0 + (n) * (stride))
+#define SMMU_PMCG_EVTYPER0              0x400
+#define SMMU_PMCG_EVTYPER(n)            (SMMU_PMCG_EVTYPER0 + (n) * 4)
+#define SMMU_PMCG_SID_SPAN_SHIFT        29
+#define SMMU_PMCG_SMR0                  0xA00
+#define SMMU_PMCG_SMR(n)                (SMMU_PMCG_SMR0 + (n) * 4)
+#define SMMU_PMCG_CNTENSET0             0xC00
+#define SMMU_PMCG_CNTENCLR0             0xC20
+#define SMMU_PMCG_INTENSET0             0xC40
+#define SMMU_PMCG_INTENCLR0             0xC60
+#define SMMU_PMCG_OVSCLR0               0xC80
+#define SMMU_PMCG_OVSSET0               0xCC0
+#define SMMU_PMCG_CFGR                  0xE00
+#define SMMU_PMCG_CFGR_SID_FILTER_TYPE  BIT(23)
+#define SMMU_PMCG_CFGR_MSI              BIT(21)
+#define SMMU_PMCG_CFGR_RELOC_CTRS       BIT(20)
+#define SMMU_PMCG_CFGR_SIZE             GENMASK(13, 8)
+#define SMMU_PMCG_CFGR_NCTR             GENMASK(5, 0)
+#define SMMU_PMCG_CR                    0xE04
+#define SMMU_PMCG_CR_ENABLE             BIT(0)
+#define SMMU_PMCG_CEID0                 0xE20
+#define SMMU_PMCG_CEID1                 0xE28
+#define SMMU_PMCG_IRQ_CTRL              0xE50
+#define SMMU_PMCG_IRQ_CTRL_IRQEN        BIT(0)
+#define SMMU_PMCG_IRQ_CFG0              0xE58
+#define SMMU_PMCG_IRQ_CFG1              0xE60
+#define SMMU_PMCG_IRQ_CFG2              0xE64
+
+/* MSI config fields */
+#define MSI_CFG0_ADDR_MASK              GENMASK_ULL(51, 2)
+#define MSI_CFG2_MEMATTR_DEVICE_nGnRE   0x1
+
+#define SMMU_PMCG_DEFAULT_FILTER_SPAN   1
+#define SMMU_PMCG_DEFAULT_FILTER_SID    GENMASK(31, 0)
+
+#define SMMU_PMCG_MAX_COUNTERS          64
+#define SMMU_PMCG_ARCH_MAX_EVENTS       128
+
+#define SMMU_PMCG_PA_SHIFT              12
+
+#define SMMU_PMCG_EVCNTR_RDONLY         BIT(0)
+#define SMMU_PMCG_HARDEN_DISABLE        BIT(1)
+
+static int cpuhp_state_num;
+
+struct smmu_pmu {
+	struct hlist_node node;
+	struct perf_event *events[SMMU_PMCG_MAX_COUNTERS];
+	DECLARE_BITMAP(used_counters, SMMU_PMCG_MAX_COUNTERS);
+	DECLARE_BITMAP(supported_events, SMMU_PMCG_ARCH_MAX_EVENTS);
+	unsigned int irq;
+	unsigned int on_cpu;
+	struct pmu pmu;
+	unsigned int num_counters;
+	struct device *dev;
+	void __iomem *reg_base;
+	void __iomem *reloc_base;
+	u64 counter_mask;
+	u32 options;
+	bool global_filter;
+};
+
+#define to_smmu_pmu(p) (container_of(p, struct smmu_pmu, pmu))
+
+#define SMMU_PMU_EVENT_ATTR_EXTRACTOR(_name, _config, _start, _end)        \
+	static inline u32 get_##_name(struct perf_event *event)            \
+	{                                                                  \
+		return FIELD_GET(GENMASK_ULL(_end, _start),                \
+				 event->attr._config);                     \
+	}                                                                  \
+
+SMMU_PMU_EVENT_ATTR_EXTRACTOR(event, config, 0, 15);
+SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_stream_id, config1, 0, 31);
+SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_span, config1, 32, 32);
+SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_enable, config1, 33, 33);
+
+static inline void smmu_pmu_enable(struct pmu *pmu)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
+
+	writel(SMMU_PMCG_IRQ_CTRL_IRQEN,
+	       smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL);
+	writel(SMMU_PMCG_CR_ENABLE, smmu_pmu->reg_base + SMMU_PMCG_CR);
+}
+
+static int smmu_pmu_apply_event_filter(struct smmu_pmu *smmu_pmu,
+				       struct perf_event *event, int idx);
+
+static inline void smmu_pmu_enable_quirk_hip08_09(struct pmu *pmu)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
+	unsigned int idx;
+
+	for_each_set_bit(idx, smmu_pmu->used_counters, smmu_pmu->num_counters)
+		smmu_pmu_apply_event_filter(smmu_pmu, smmu_pmu->events[idx], idx);
+
+	smmu_pmu_enable(pmu);
+}
+
+static inline void smmu_pmu_disable(struct pmu *pmu)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
+
+	writel(0, smmu_pmu->reg_base + SMMU_PMCG_CR);
+	writel(0, smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL);
+}
+
+static inline void smmu_pmu_disable_quirk_hip08_09(struct pmu *pmu)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
+	unsigned int idx;
+
+	/*
+	 * The global disable of PMU sometimes fail to stop the counting.
+	 * Harden this by writing an invalid event type to each used counter
+	 * to forcibly stop counting.
+	 */
+	for_each_set_bit(idx, smmu_pmu->used_counters, smmu_pmu->num_counters)
+		writel(0xffff, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx));
+
+	smmu_pmu_disable(pmu);
+}
+
+static inline void smmu_pmu_counter_set_value(struct smmu_pmu *smmu_pmu,
+					      u32 idx, u64 value)
+{
+	if (smmu_pmu->counter_mask & BIT(32))
+		writeq(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, 8));
+	else
+		writel(value, smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, 4));
+}
+
+static inline u64 smmu_pmu_counter_get_value(struct smmu_pmu *smmu_pmu, u32 idx)
+{
+	u64 value;
+
+	if (smmu_pmu->counter_mask & BIT(32))
+		value = readq(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, 8));
+	else
+		value = readl(smmu_pmu->reloc_base + SMMU_PMCG_EVCNTR(idx, 4));
+
+	return value;
+}
+
+static inline void smmu_pmu_counter_enable(struct smmu_pmu *smmu_pmu, u32 idx)
+{
+	writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENSET0);
+}
+
+static inline void smmu_pmu_counter_disable(struct smmu_pmu *smmu_pmu, u32 idx)
+{
+	writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0);
+}
+
+static inline void smmu_pmu_interrupt_enable(struct smmu_pmu *smmu_pmu, u32 idx)
+{
+	writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENSET0);
+}
+
+static inline void smmu_pmu_interrupt_disable(struct smmu_pmu *smmu_pmu,
+					      u32 idx)
+{
+	writeq(BIT(idx), smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0);
+}
+
+static inline void smmu_pmu_set_evtyper(struct smmu_pmu *smmu_pmu, u32 idx,
+					u32 val)
+{
+	writel(val, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx));
+}
+
+static inline void smmu_pmu_set_smr(struct smmu_pmu *smmu_pmu, u32 idx, u32 val)
+{
+	writel(val, smmu_pmu->reg_base + SMMU_PMCG_SMR(idx));
+}
+
+static void smmu_pmu_event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	u64 delta, prev, now;
+	u32 idx = hwc->idx;
+
+	do {
+		prev = local64_read(&hwc->prev_count);
+		now = smmu_pmu_counter_get_value(smmu_pmu, idx);
+	} while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev);
+
+	/* handle overflow. */
+	delta = now - prev;
+	delta &= smmu_pmu->counter_mask;
+
+	local64_add(delta, &event->count);
+}
+
+static void smmu_pmu_set_period(struct smmu_pmu *smmu_pmu,
+				struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+	u64 new;
+
+	if (smmu_pmu->options & SMMU_PMCG_EVCNTR_RDONLY) {
+		/*
+		 * On platforms that require this quirk, if the counter starts
+		 * at < half_counter value and wraps, the current logic of
+		 * handling the overflow may not work. It is expected that,
+		 * those platforms will have full 64 counter bits implemented
+		 * so that such a possibility is remote(eg: HiSilicon HIP08).
+		 */
+		new = smmu_pmu_counter_get_value(smmu_pmu, idx);
+	} else {
+		/*
+		 * We limit the max period to half the max counter value
+		 * of the counter size, so that even in the case of extreme
+		 * interrupt latency the counter will (hopefully) not wrap
+		 * past its initial value.
+		 */
+		new = smmu_pmu->counter_mask >> 1;
+		smmu_pmu_counter_set_value(smmu_pmu, idx, new);
+	}
+
+	local64_set(&hwc->prev_count, new);
+}
+
+static void smmu_pmu_set_event_filter(struct perf_event *event,
+				      int idx, u32 span, u32 sid)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	u32 evtyper;
+
+	evtyper = get_event(event) | span << SMMU_PMCG_SID_SPAN_SHIFT;
+	smmu_pmu_set_evtyper(smmu_pmu, idx, evtyper);
+	smmu_pmu_set_smr(smmu_pmu, idx, sid);
+}
+
+static bool smmu_pmu_check_global_filter(struct perf_event *curr,
+					 struct perf_event *new)
+{
+	if (get_filter_enable(new) != get_filter_enable(curr))
+		return false;
+
+	if (!get_filter_enable(new))
+		return true;
+
+	return get_filter_span(new) == get_filter_span(curr) &&
+	       get_filter_stream_id(new) == get_filter_stream_id(curr);
+}
+
+static int smmu_pmu_apply_event_filter(struct smmu_pmu *smmu_pmu,
+				       struct perf_event *event, int idx)
+{
+	u32 span, sid;
+	unsigned int cur_idx, num_ctrs = smmu_pmu->num_counters;
+	bool filter_en = !!get_filter_enable(event);
+
+	span = filter_en ? get_filter_span(event) :
+			   SMMU_PMCG_DEFAULT_FILTER_SPAN;
+	sid = filter_en ? get_filter_stream_id(event) :
+			   SMMU_PMCG_DEFAULT_FILTER_SID;
+
+	cur_idx = find_first_bit(smmu_pmu->used_counters, num_ctrs);
+	/*
+	 * Per-counter filtering, or scheduling the first globally-filtered
+	 * event into an empty PMU so idx == 0 and it works out equivalent.
+	 */
+	if (!smmu_pmu->global_filter || cur_idx == num_ctrs) {
+		smmu_pmu_set_event_filter(event, idx, span, sid);
+		return 0;
+	}
+
+	/* Otherwise, must match whatever's currently scheduled */
+	if (smmu_pmu_check_global_filter(smmu_pmu->events[cur_idx], event)) {
+		smmu_pmu_set_evtyper(smmu_pmu, idx, get_event(event));
+		return 0;
+	}
+
+	return -EAGAIN;
+}
+
+static int smmu_pmu_get_event_idx(struct smmu_pmu *smmu_pmu,
+				  struct perf_event *event)
+{
+	int idx, err;
+	unsigned int num_ctrs = smmu_pmu->num_counters;
+
+	idx = find_first_zero_bit(smmu_pmu->used_counters, num_ctrs);
+	if (idx == num_ctrs)
+		/* The counters are all in use. */
+		return -EAGAIN;
+
+	err = smmu_pmu_apply_event_filter(smmu_pmu, event, idx);
+	if (err)
+		return err;
+
+	set_bit(idx, smmu_pmu->used_counters);
+
+	return idx;
+}
+
+static bool smmu_pmu_events_compatible(struct perf_event *curr,
+				       struct perf_event *new)
+{
+	if (new->pmu != curr->pmu)
+		return false;
+
+	if (to_smmu_pmu(new->pmu)->global_filter &&
+	    !smmu_pmu_check_global_filter(curr, new))
+		return false;
+
+	return true;
+}
+
+/*
+ * Implementation of abstract pmu functionality required by
+ * the core perf events code.
+ */
+
+static int smmu_pmu_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	struct device *dev = smmu_pmu->dev;
+	struct perf_event *sibling;
+	int group_num_events = 1;
+	u16 event_id;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	if (hwc->sample_period) {
+		dev_dbg(dev, "Sampling not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (event->cpu < 0) {
+		dev_dbg(dev, "Per-task mode not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	/* Verify specified event is supported on this PMU */
+	event_id = get_event(event);
+	if (event_id < SMMU_PMCG_ARCH_MAX_EVENTS &&
+	    (!test_bit(event_id, smmu_pmu->supported_events))) {
+		dev_dbg(dev, "Invalid event %d for this PMU\n", event_id);
+		return -EINVAL;
+	}
+
+	/* Don't allow groups with mixed PMUs, except for s/w events */
+	if (!is_software_event(event->group_leader)) {
+		if (!smmu_pmu_events_compatible(event->group_leader, event))
+			return -EINVAL;
+
+		if (++group_num_events > smmu_pmu->num_counters)
+			return -EINVAL;
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (is_software_event(sibling))
+			continue;
+
+		if (!smmu_pmu_events_compatible(sibling, event))
+			return -EINVAL;
+
+		if (++group_num_events > smmu_pmu->num_counters)
+			return -EINVAL;
+	}
+
+	hwc->idx = -1;
+
+	/*
+	 * Ensure all events are on the same cpu so all events are in the
+	 * same cpu context, to avoid races on pmu_enable etc.
+	 */
+	event->cpu = smmu_pmu->on_cpu;
+
+	return 0;
+}
+
+static void smmu_pmu_event_start(struct perf_event *event, int flags)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	hwc->state = 0;
+
+	smmu_pmu_set_period(smmu_pmu, hwc);
+
+	smmu_pmu_counter_enable(smmu_pmu, idx);
+}
+
+static void smmu_pmu_event_stop(struct perf_event *event, int flags)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	if (hwc->state & PERF_HES_STOPPED)
+		return;
+
+	smmu_pmu_counter_disable(smmu_pmu, idx);
+	/* As the counter gets updated on _start, ignore PERF_EF_UPDATE */
+	smmu_pmu_event_update(event);
+	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+}
+
+static int smmu_pmu_event_add(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+
+	idx = smmu_pmu_get_event_idx(smmu_pmu, event);
+	if (idx < 0)
+		return idx;
+
+	hwc->idx = idx;
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	smmu_pmu->events[idx] = event;
+	local64_set(&hwc->prev_count, 0);
+
+	smmu_pmu_interrupt_enable(smmu_pmu, idx);
+
+	if (flags & PERF_EF_START)
+		smmu_pmu_event_start(event, flags);
+
+	/* Propagate changes to the userspace mapping. */
+	perf_event_update_userpage(event);
+
+	return 0;
+}
+
+static void smmu_pmu_event_del(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(event->pmu);
+	int idx = hwc->idx;
+
+	smmu_pmu_event_stop(event, flags | PERF_EF_UPDATE);
+	smmu_pmu_interrupt_disable(smmu_pmu, idx);
+	smmu_pmu->events[idx] = NULL;
+	clear_bit(idx, smmu_pmu->used_counters);
+
+	perf_event_update_userpage(event);
+}
+
+static void smmu_pmu_event_read(struct perf_event *event)
+{
+	smmu_pmu_event_update(event);
+}
+
+/* cpumask */
+
+static ssize_t smmu_pmu_cpumask_show(struct device *dev,
+				     struct device_attribute *attr,
+				     char *buf)
+{
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(smmu_pmu->on_cpu));
+}
+
+static struct device_attribute smmu_pmu_cpumask_attr =
+		__ATTR(cpumask, 0444, smmu_pmu_cpumask_show, NULL);
+
+static struct attribute *smmu_pmu_cpumask_attrs[] = {
+	&smmu_pmu_cpumask_attr.attr,
+	NULL
+};
+
+static struct attribute_group smmu_pmu_cpumask_group = {
+	.attrs = smmu_pmu_cpumask_attrs,
+};
+
+/* Events */
+
+static ssize_t smmu_pmu_event_show(struct device *dev,
+				   struct device_attribute *attr, char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+
+	return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define SMMU_EVENT_ATTR(name, config) \
+	PMU_EVENT_ATTR(name, smmu_event_attr_##name, \
+		       config, smmu_pmu_event_show)
+SMMU_EVENT_ATTR(cycles, 0);
+SMMU_EVENT_ATTR(transaction, 1);
+SMMU_EVENT_ATTR(tlb_miss, 2);
+SMMU_EVENT_ATTR(config_cache_miss, 3);
+SMMU_EVENT_ATTR(trans_table_walk_access, 4);
+SMMU_EVENT_ATTR(config_struct_access, 5);
+SMMU_EVENT_ATTR(pcie_ats_trans_rq, 6);
+SMMU_EVENT_ATTR(pcie_ats_trans_passed, 7);
+
+static struct attribute *smmu_pmu_events[] = {
+	&smmu_event_attr_cycles.attr.attr,
+	&smmu_event_attr_transaction.attr.attr,
+	&smmu_event_attr_tlb_miss.attr.attr,
+	&smmu_event_attr_config_cache_miss.attr.attr,
+	&smmu_event_attr_trans_table_walk_access.attr.attr,
+	&smmu_event_attr_config_struct_access.attr.attr,
+	&smmu_event_attr_pcie_ats_trans_rq.attr.attr,
+	&smmu_event_attr_pcie_ats_trans_passed.attr.attr,
+	NULL
+};
+
+static umode_t smmu_pmu_event_is_visible(struct kobject *kobj,
+					 struct attribute *attr, int unused)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct smmu_pmu *smmu_pmu = to_smmu_pmu(dev_get_drvdata(dev));
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr);
+
+	if (test_bit(pmu_attr->id, smmu_pmu->supported_events))
+		return attr->mode;
+
+	return 0;
+}
+
+static struct attribute_group smmu_pmu_events_group = {
+	.name = "events",
+	.attrs = smmu_pmu_events,
+	.is_visible = smmu_pmu_event_is_visible,
+};
+
+/* Formats */
+PMU_FORMAT_ATTR(event,		   "config:0-15");
+PMU_FORMAT_ATTR(filter_stream_id,  "config1:0-31");
+PMU_FORMAT_ATTR(filter_span,	   "config1:32");
+PMU_FORMAT_ATTR(filter_enable,	   "config1:33");
+
+static struct attribute *smmu_pmu_formats[] = {
+	&format_attr_event.attr,
+	&format_attr_filter_stream_id.attr,
+	&format_attr_filter_span.attr,
+	&format_attr_filter_enable.attr,
+	NULL
+};
+
+static struct attribute_group smmu_pmu_format_group = {
+	.name = "format",
+	.attrs = smmu_pmu_formats,
+};
+
+static const struct attribute_group *smmu_pmu_attr_grps[] = {
+	&smmu_pmu_cpumask_group,
+	&smmu_pmu_events_group,
+	&smmu_pmu_format_group,
+	NULL
+};
+
+/*
+ * Generic device handlers
+ */
+
+static int smmu_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct smmu_pmu *smmu_pmu;
+	unsigned int target;
+
+	smmu_pmu = hlist_entry_safe(node, struct smmu_pmu, node);
+	if (cpu != smmu_pmu->on_cpu)
+		return 0;
+
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	perf_pmu_migrate_context(&smmu_pmu->pmu, cpu, target);
+	smmu_pmu->on_cpu = target;
+	WARN_ON(irq_set_affinity_hint(smmu_pmu->irq, cpumask_of(target)));
+
+	return 0;
+}
+
+static irqreturn_t smmu_pmu_handle_irq(int irq_num, void *data)
+{
+	struct smmu_pmu *smmu_pmu = data;
+	u64 ovsr;
+	unsigned int idx;
+
+	ovsr = readq(smmu_pmu->reloc_base + SMMU_PMCG_OVSSET0);
+	if (!ovsr)
+		return IRQ_NONE;
+
+	writeq(ovsr, smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0);
+
+	for_each_set_bit(idx, (unsigned long *)&ovsr, smmu_pmu->num_counters) {
+		struct perf_event *event = smmu_pmu->events[idx];
+		struct hw_perf_event *hwc;
+
+		if (WARN_ON_ONCE(!event))
+			continue;
+
+		smmu_pmu_event_update(event);
+		hwc = &event->hw;
+
+		smmu_pmu_set_period(smmu_pmu, hwc);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static void smmu_pmu_free_msis(void *data)
+{
+	struct device *dev = data;
+
+	platform_msi_domain_free_irqs(dev);
+}
+
+static void smmu_pmu_write_msi_msg(struct msi_desc *desc, struct msi_msg *msg)
+{
+	phys_addr_t doorbell;
+	struct device *dev = msi_desc_to_dev(desc);
+	struct smmu_pmu *pmu = dev_get_drvdata(dev);
+
+	doorbell = (((u64)msg->address_hi) << 32) | msg->address_lo;
+	doorbell &= MSI_CFG0_ADDR_MASK;
+
+	writeq_relaxed(doorbell, pmu->reg_base + SMMU_PMCG_IRQ_CFG0);
+	writel_relaxed(msg->data, pmu->reg_base + SMMU_PMCG_IRQ_CFG1);
+	writel_relaxed(MSI_CFG2_MEMATTR_DEVICE_nGnRE,
+		       pmu->reg_base + SMMU_PMCG_IRQ_CFG2);
+}
+
+static void smmu_pmu_setup_msi(struct smmu_pmu *pmu)
+{
+	struct msi_desc *desc;
+	struct device *dev = pmu->dev;
+	int ret;
+
+	/* Clear MSI address reg */
+	writeq_relaxed(0, pmu->reg_base + SMMU_PMCG_IRQ_CFG0);
+
+	/* MSI supported or not */
+	if (!(readl(pmu->reg_base + SMMU_PMCG_CFGR) & SMMU_PMCG_CFGR_MSI))
+		return;
+
+	ret = platform_msi_domain_alloc_irqs(dev, 1, smmu_pmu_write_msi_msg);
+	if (ret) {
+		dev_warn(dev, "failed to allocate MSIs\n");
+		return;
+	}
+
+	desc = first_msi_entry(dev);
+	if (desc)
+		pmu->irq = desc->irq;
+
+	/* Add callback to free MSIs on teardown */
+	devm_add_action(dev, smmu_pmu_free_msis, dev);
+}
+
+static int smmu_pmu_setup_irq(struct smmu_pmu *pmu)
+{
+	unsigned long flags = IRQF_NOBALANCING | IRQF_SHARED | IRQF_NO_THREAD;
+	int irq, ret = -ENXIO;
+
+	smmu_pmu_setup_msi(pmu);
+
+	irq = pmu->irq;
+	if (irq)
+		ret = devm_request_irq(pmu->dev, irq, smmu_pmu_handle_irq,
+				       flags, "smmuv3-pmu", pmu);
+	return ret;
+}
+
+static void smmu_pmu_reset(struct smmu_pmu *smmu_pmu)
+{
+	u64 counter_present_mask = GENMASK_ULL(smmu_pmu->num_counters - 1, 0);
+
+	smmu_pmu_disable(&smmu_pmu->pmu);
+
+	/* Disable counter and interrupt */
+	writeq_relaxed(counter_present_mask,
+		       smmu_pmu->reg_base + SMMU_PMCG_CNTENCLR0);
+	writeq_relaxed(counter_present_mask,
+		       smmu_pmu->reg_base + SMMU_PMCG_INTENCLR0);
+	writeq_relaxed(counter_present_mask,
+		       smmu_pmu->reloc_base + SMMU_PMCG_OVSCLR0);
+}
+
+static void smmu_pmu_get_acpi_options(struct smmu_pmu *smmu_pmu)
+{
+	u32 model;
+
+	model = *(u32 *)dev_get_platdata(smmu_pmu->dev);
+
+	switch (model) {
+	case IORT_SMMU_V3_PMCG_HISI_HIP08:
+		/* HiSilicon Erratum 162001800 */
+		smmu_pmu->options |= SMMU_PMCG_EVCNTR_RDONLY | SMMU_PMCG_HARDEN_DISABLE;
+		break;
+	case IORT_SMMU_V3_PMCG_HISI_HIP09:
+		smmu_pmu->options |= SMMU_PMCG_HARDEN_DISABLE;
+		break;
+	}
+
+	dev_notice(smmu_pmu->dev, "option mask 0x%x\n", smmu_pmu->options);
+}
+
+static int smmu_pmu_probe(struct platform_device *pdev)
+{
+	struct smmu_pmu *smmu_pmu;
+	struct resource *res_0;
+	u32 cfgr, reg_size;
+	u64 ceid_64[2];
+	int irq, err;
+	char *name;
+	struct device *dev = &pdev->dev;
+
+	smmu_pmu = devm_kzalloc(dev, sizeof(*smmu_pmu), GFP_KERNEL);
+	if (!smmu_pmu)
+		return -ENOMEM;
+
+	smmu_pmu->dev = dev;
+	platform_set_drvdata(pdev, smmu_pmu);
+
+	smmu_pmu->pmu = (struct pmu) {
+		.module		= THIS_MODULE,
+		.task_ctx_nr    = perf_invalid_context,
+		.pmu_enable	= smmu_pmu_enable,
+		.pmu_disable	= smmu_pmu_disable,
+		.event_init	= smmu_pmu_event_init,
+		.add		= smmu_pmu_event_add,
+		.del		= smmu_pmu_event_del,
+		.start		= smmu_pmu_event_start,
+		.stop		= smmu_pmu_event_stop,
+		.read		= smmu_pmu_event_read,
+		.attr_groups	= smmu_pmu_attr_grps,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	smmu_pmu->reg_base = devm_platform_get_and_ioremap_resource(pdev, 0, &res_0);
+	if (IS_ERR(smmu_pmu->reg_base))
+		return PTR_ERR(smmu_pmu->reg_base);
+
+	cfgr = readl_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CFGR);
+
+	/* Determine if page 1 is present */
+	if (cfgr & SMMU_PMCG_CFGR_RELOC_CTRS) {
+		smmu_pmu->reloc_base = devm_platform_ioremap_resource(pdev, 1);
+		if (IS_ERR(smmu_pmu->reloc_base))
+			return PTR_ERR(smmu_pmu->reloc_base);
+	} else {
+		smmu_pmu->reloc_base = smmu_pmu->reg_base;
+	}
+
+	irq = platform_get_irq_optional(pdev, 0);
+	if (irq > 0)
+		smmu_pmu->irq = irq;
+
+	ceid_64[0] = readq_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CEID0);
+	ceid_64[1] = readq_relaxed(smmu_pmu->reg_base + SMMU_PMCG_CEID1);
+	bitmap_from_arr32(smmu_pmu->supported_events, (u32 *)ceid_64,
+			  SMMU_PMCG_ARCH_MAX_EVENTS);
+
+	smmu_pmu->num_counters = FIELD_GET(SMMU_PMCG_CFGR_NCTR, cfgr) + 1;
+
+	smmu_pmu->global_filter = !!(cfgr & SMMU_PMCG_CFGR_SID_FILTER_TYPE);
+
+	reg_size = FIELD_GET(SMMU_PMCG_CFGR_SIZE, cfgr);
+	smmu_pmu->counter_mask = GENMASK_ULL(reg_size, 0);
+
+	smmu_pmu_reset(smmu_pmu);
+
+	err = smmu_pmu_setup_irq(smmu_pmu);
+	if (err) {
+		dev_err(dev, "Setup irq failed, PMU @%pa\n", &res_0->start);
+		return err;
+	}
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "smmuv3_pmcg_%llx",
+			      (res_0->start) >> SMMU_PMCG_PA_SHIFT);
+	if (!name) {
+		dev_err(dev, "Create name failed, PMU @%pa\n", &res_0->start);
+		return -EINVAL;
+	}
+
+	smmu_pmu_get_acpi_options(smmu_pmu);
+
+	/*
+	 * For platforms suffer this quirk, the PMU disable sometimes fails to
+	 * stop the counters. This will leads to inaccurate or error counting.
+	 * Forcibly disable the counters with these quirk handler.
+	 */
+	if (smmu_pmu->options & SMMU_PMCG_HARDEN_DISABLE) {
+		smmu_pmu->pmu.pmu_enable = smmu_pmu_enable_quirk_hip08_09;
+		smmu_pmu->pmu.pmu_disable = smmu_pmu_disable_quirk_hip08_09;
+	}
+
+	/* Pick one CPU to be the preferred one to use */
+	smmu_pmu->on_cpu = raw_smp_processor_id();
+	WARN_ON(irq_set_affinity_hint(smmu_pmu->irq,
+				      cpumask_of(smmu_pmu->on_cpu)));
+
+	err = cpuhp_state_add_instance_nocalls(cpuhp_state_num,
+					       &smmu_pmu->node);
+	if (err) {
+		dev_err(dev, "Error %d registering hotplug, PMU @%pa\n",
+			err, &res_0->start);
+		goto out_clear_affinity;
+	}
+
+	err = perf_pmu_register(&smmu_pmu->pmu, name, -1);
+	if (err) {
+		dev_err(dev, "Error %d registering PMU @%pa\n",
+			err, &res_0->start);
+		goto out_unregister;
+	}
+
+	dev_info(dev, "Registered PMU @ %pa using %d counters with %s filter settings\n",
+		 &res_0->start, smmu_pmu->num_counters,
+		 smmu_pmu->global_filter ? "Global(Counter0)" :
+		 "Individual");
+
+	return 0;
+
+out_unregister:
+	cpuhp_state_remove_instance_nocalls(cpuhp_state_num, &smmu_pmu->node);
+out_clear_affinity:
+	irq_set_affinity_hint(smmu_pmu->irq, NULL);
+	return err;
+}
+
+static int smmu_pmu_remove(struct platform_device *pdev)
+{
+	struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev);
+
+	perf_pmu_unregister(&smmu_pmu->pmu);
+	cpuhp_state_remove_instance_nocalls(cpuhp_state_num, &smmu_pmu->node);
+	irq_set_affinity_hint(smmu_pmu->irq, NULL);
+
+	return 0;
+}
+
+static void smmu_pmu_shutdown(struct platform_device *pdev)
+{
+	struct smmu_pmu *smmu_pmu = platform_get_drvdata(pdev);
+
+	smmu_pmu_disable(&smmu_pmu->pmu);
+}
+
+static struct platform_driver smmu_pmu_driver = {
+	.driver = {
+		.name = "arm-smmu-v3-pmcg",
+		.suppress_bind_attrs = true,
+	},
+	.probe = smmu_pmu_probe,
+	.remove = smmu_pmu_remove,
+	.shutdown = smmu_pmu_shutdown,
+};
+
+static int __init arm_smmu_pmu_init(void)
+{
+	int ret;
+
+	cpuhp_state_num = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
+						  "perf/arm/pmcg:online",
+						  NULL,
+						  smmu_pmu_offline_cpu);
+	if (cpuhp_state_num < 0)
+		return cpuhp_state_num;
+
+	ret = platform_driver_register(&smmu_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(cpuhp_state_num);
+
+	return ret;
+}
+module_init(arm_smmu_pmu_init);
+
+static void __exit arm_smmu_pmu_exit(void)
+{
+	platform_driver_unregister(&smmu_pmu_driver);
+	cpuhp_remove_multi_state(cpuhp_state_num);
+}
+
+module_exit(arm_smmu_pmu_exit);
+
+MODULE_DESCRIPTION("PMU driver for ARM SMMUv3 Performance Monitors Extension");
+MODULE_AUTHOR("Neil Leeder <nleeder@codeaurora.org>");
+MODULE_AUTHOR("Shameer Kolothum <shameerali.kolothum.thodi@huawei.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c
new file mode 100644
index 000000000..6fbfcab49
--- /dev/null
+++ b/drivers/perf/arm_spe_pmu.c
@@ -0,0 +1,1282 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Perf support for the Statistical Profiling Extension, introduced as
+ * part of ARMv8.2.
+ *
+ * Copyright (C) 2016 ARM Limited
+ *
+ * Author: Will Deacon <will.deacon@arm.com>
+ */
+
+#define PMUNAME					"arm_spe"
+#define DRVNAME					PMUNAME "_pmu"
+#define pr_fmt(fmt)				DRVNAME ": " fmt
+
+#include <linux/bitops.h>
+#include <linux/bug.h>
+#include <linux/capability.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/perf_event.h>
+#include <linux/perf/arm_pmu.h>
+#include <linux/platform_device.h>
+#include <linux/printk.h>
+#include <linux/slab.h>
+#include <linux/smp.h>
+#include <linux/vmalloc.h>
+
+#include <asm/barrier.h>
+#include <asm/cpufeature.h>
+#include <asm/mmu.h>
+#include <asm/sysreg.h>
+
+/*
+ * Cache if the event is allowed to trace Context information.
+ * This allows us to perform the check, i.e, perfmon_capable(),
+ * in the context of the event owner, once, during the event_init().
+ */
+#define SPE_PMU_HW_FLAGS_CX			BIT(0)
+
+static void set_spe_event_has_cx(struct perf_event *event)
+{
+	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
+		event->hw.flags |= SPE_PMU_HW_FLAGS_CX;
+}
+
+static bool get_spe_event_has_cx(struct perf_event *event)
+{
+	return !!(event->hw.flags & SPE_PMU_HW_FLAGS_CX);
+}
+
+#define ARM_SPE_BUF_PAD_BYTE			0
+
+struct arm_spe_pmu_buf {
+	int					nr_pages;
+	bool					snapshot;
+	void					*base;
+};
+
+struct arm_spe_pmu {
+	struct pmu				pmu;
+	struct platform_device			*pdev;
+	cpumask_t				supported_cpus;
+	struct hlist_node			hotplug_node;
+
+	int					irq; /* PPI */
+
+	u16					min_period;
+	u16					counter_sz;
+
+#define SPE_PMU_FEAT_FILT_EVT			(1UL << 0)
+#define SPE_PMU_FEAT_FILT_TYP			(1UL << 1)
+#define SPE_PMU_FEAT_FILT_LAT			(1UL << 2)
+#define SPE_PMU_FEAT_ARCH_INST			(1UL << 3)
+#define SPE_PMU_FEAT_LDS			(1UL << 4)
+#define SPE_PMU_FEAT_ERND			(1UL << 5)
+#define SPE_PMU_FEAT_DEV_PROBED			(1UL << 63)
+	u64					features;
+
+	u16					max_record_sz;
+	u16					align;
+	struct perf_output_handle __percpu	*handle;
+};
+
+#define to_spe_pmu(p) (container_of(p, struct arm_spe_pmu, pmu))
+
+/* Convert a free-running index from perf into an SPE buffer offset */
+#define PERF_IDX2OFF(idx, buf)	((idx) % ((buf)->nr_pages << PAGE_SHIFT))
+
+/* Keep track of our dynamic hotplug state */
+static enum cpuhp_state arm_spe_pmu_online;
+
+enum arm_spe_pmu_buf_fault_action {
+	SPE_PMU_BUF_FAULT_ACT_SPURIOUS,
+	SPE_PMU_BUF_FAULT_ACT_FATAL,
+	SPE_PMU_BUF_FAULT_ACT_OK,
+};
+
+/* This sysfs gunk was really good fun to write. */
+enum arm_spe_pmu_capabilities {
+	SPE_PMU_CAP_ARCH_INST = 0,
+	SPE_PMU_CAP_ERND,
+	SPE_PMU_CAP_FEAT_MAX,
+	SPE_PMU_CAP_CNT_SZ = SPE_PMU_CAP_FEAT_MAX,
+	SPE_PMU_CAP_MIN_IVAL,
+};
+
+static int arm_spe_pmu_feat_caps[SPE_PMU_CAP_FEAT_MAX] = {
+	[SPE_PMU_CAP_ARCH_INST]	= SPE_PMU_FEAT_ARCH_INST,
+	[SPE_PMU_CAP_ERND]	= SPE_PMU_FEAT_ERND,
+};
+
+static u32 arm_spe_pmu_cap_get(struct arm_spe_pmu *spe_pmu, int cap)
+{
+	if (cap < SPE_PMU_CAP_FEAT_MAX)
+		return !!(spe_pmu->features & arm_spe_pmu_feat_caps[cap]);
+
+	switch (cap) {
+	case SPE_PMU_CAP_CNT_SZ:
+		return spe_pmu->counter_sz;
+	case SPE_PMU_CAP_MIN_IVAL:
+		return spe_pmu->min_period;
+	default:
+		WARN(1, "unknown cap %d\n", cap);
+	}
+
+	return 0;
+}
+
+static ssize_t arm_spe_pmu_cap_show(struct device *dev,
+				    struct device_attribute *attr,
+				    char *buf)
+{
+	struct arm_spe_pmu *spe_pmu = dev_get_drvdata(dev);
+	struct dev_ext_attribute *ea =
+		container_of(attr, struct dev_ext_attribute, attr);
+	int cap = (long)ea->var;
+
+	return snprintf(buf, PAGE_SIZE, "%u\n",
+		arm_spe_pmu_cap_get(spe_pmu, cap));
+}
+
+#define SPE_EXT_ATTR_ENTRY(_name, _func, _var)				\
+	&((struct dev_ext_attribute[]) {				\
+		{ __ATTR(_name, S_IRUGO, _func, NULL), (void *)_var }	\
+	})[0].attr.attr
+
+#define SPE_CAP_EXT_ATTR_ENTRY(_name, _var)				\
+	SPE_EXT_ATTR_ENTRY(_name, arm_spe_pmu_cap_show, _var)
+
+static struct attribute *arm_spe_pmu_cap_attr[] = {
+	SPE_CAP_EXT_ATTR_ENTRY(arch_inst, SPE_PMU_CAP_ARCH_INST),
+	SPE_CAP_EXT_ATTR_ENTRY(ernd, SPE_PMU_CAP_ERND),
+	SPE_CAP_EXT_ATTR_ENTRY(count_size, SPE_PMU_CAP_CNT_SZ),
+	SPE_CAP_EXT_ATTR_ENTRY(min_interval, SPE_PMU_CAP_MIN_IVAL),
+	NULL,
+};
+
+static struct attribute_group arm_spe_pmu_cap_group = {
+	.name	= "caps",
+	.attrs	= arm_spe_pmu_cap_attr,
+};
+
+/* User ABI */
+#define ATTR_CFG_FLD_ts_enable_CFG		config	/* PMSCR_EL1.TS */
+#define ATTR_CFG_FLD_ts_enable_LO		0
+#define ATTR_CFG_FLD_ts_enable_HI		0
+#define ATTR_CFG_FLD_pa_enable_CFG		config	/* PMSCR_EL1.PA */
+#define ATTR_CFG_FLD_pa_enable_LO		1
+#define ATTR_CFG_FLD_pa_enable_HI		1
+#define ATTR_CFG_FLD_pct_enable_CFG		config	/* PMSCR_EL1.PCT */
+#define ATTR_CFG_FLD_pct_enable_LO		2
+#define ATTR_CFG_FLD_pct_enable_HI		2
+#define ATTR_CFG_FLD_jitter_CFG			config	/* PMSIRR_EL1.RND */
+#define ATTR_CFG_FLD_jitter_LO			16
+#define ATTR_CFG_FLD_jitter_HI			16
+#define ATTR_CFG_FLD_branch_filter_CFG		config	/* PMSFCR_EL1.B */
+#define ATTR_CFG_FLD_branch_filter_LO		32
+#define ATTR_CFG_FLD_branch_filter_HI		32
+#define ATTR_CFG_FLD_load_filter_CFG		config	/* PMSFCR_EL1.LD */
+#define ATTR_CFG_FLD_load_filter_LO		33
+#define ATTR_CFG_FLD_load_filter_HI		33
+#define ATTR_CFG_FLD_store_filter_CFG		config	/* PMSFCR_EL1.ST */
+#define ATTR_CFG_FLD_store_filter_LO		34
+#define ATTR_CFG_FLD_store_filter_HI		34
+
+#define ATTR_CFG_FLD_event_filter_CFG		config1	/* PMSEVFR_EL1 */
+#define ATTR_CFG_FLD_event_filter_LO		0
+#define ATTR_CFG_FLD_event_filter_HI		63
+
+#define ATTR_CFG_FLD_min_latency_CFG		config2	/* PMSLATFR_EL1.MINLAT */
+#define ATTR_CFG_FLD_min_latency_LO		0
+#define ATTR_CFG_FLD_min_latency_HI		11
+
+/* Why does everything I do descend into this? */
+#define __GEN_PMU_FORMAT_ATTR(cfg, lo, hi)				\
+	(lo) == (hi) ? #cfg ":" #lo "\n" : #cfg ":" #lo "-" #hi
+
+#define _GEN_PMU_FORMAT_ATTR(cfg, lo, hi)				\
+	__GEN_PMU_FORMAT_ATTR(cfg, lo, hi)
+
+#define GEN_PMU_FORMAT_ATTR(name)					\
+	PMU_FORMAT_ATTR(name,						\
+	_GEN_PMU_FORMAT_ATTR(ATTR_CFG_FLD_##name##_CFG,			\
+			     ATTR_CFG_FLD_##name##_LO,			\
+			     ATTR_CFG_FLD_##name##_HI))
+
+#define _ATTR_CFG_GET_FLD(attr, cfg, lo, hi)				\
+	((((attr)->cfg) >> lo) & GENMASK(hi - lo, 0))
+
+#define ATTR_CFG_GET_FLD(attr, name)					\
+	_ATTR_CFG_GET_FLD(attr,						\
+			  ATTR_CFG_FLD_##name##_CFG,			\
+			  ATTR_CFG_FLD_##name##_LO,			\
+			  ATTR_CFG_FLD_##name##_HI)
+
+GEN_PMU_FORMAT_ATTR(ts_enable);
+GEN_PMU_FORMAT_ATTR(pa_enable);
+GEN_PMU_FORMAT_ATTR(pct_enable);
+GEN_PMU_FORMAT_ATTR(jitter);
+GEN_PMU_FORMAT_ATTR(branch_filter);
+GEN_PMU_FORMAT_ATTR(load_filter);
+GEN_PMU_FORMAT_ATTR(store_filter);
+GEN_PMU_FORMAT_ATTR(event_filter);
+GEN_PMU_FORMAT_ATTR(min_latency);
+
+static struct attribute *arm_spe_pmu_formats_attr[] = {
+	&format_attr_ts_enable.attr,
+	&format_attr_pa_enable.attr,
+	&format_attr_pct_enable.attr,
+	&format_attr_jitter.attr,
+	&format_attr_branch_filter.attr,
+	&format_attr_load_filter.attr,
+	&format_attr_store_filter.attr,
+	&format_attr_event_filter.attr,
+	&format_attr_min_latency.attr,
+	NULL,
+};
+
+static struct attribute_group arm_spe_pmu_format_group = {
+	.name	= "format",
+	.attrs	= arm_spe_pmu_formats_attr,
+};
+
+static ssize_t arm_spe_pmu_get_attr_cpumask(struct device *dev,
+					    struct device_attribute *attr,
+					    char *buf)
+{
+	struct arm_spe_pmu *spe_pmu = dev_get_drvdata(dev);
+
+	return cpumap_print_to_pagebuf(true, buf, &spe_pmu->supported_cpus);
+}
+static DEVICE_ATTR(cpumask, S_IRUGO, arm_spe_pmu_get_attr_cpumask, NULL);
+
+static struct attribute *arm_spe_pmu_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group arm_spe_pmu_group = {
+	.attrs	= arm_spe_pmu_attrs,
+};
+
+static const struct attribute_group *arm_spe_pmu_attr_groups[] = {
+	&arm_spe_pmu_group,
+	&arm_spe_pmu_cap_group,
+	&arm_spe_pmu_format_group,
+	NULL,
+};
+
+/* Convert between user ABI and register values */
+static u64 arm_spe_event_to_pmscr(struct perf_event *event)
+{
+	struct perf_event_attr *attr = &event->attr;
+	u64 reg = 0;
+
+	reg |= ATTR_CFG_GET_FLD(attr, ts_enable) << SYS_PMSCR_EL1_TS_SHIFT;
+	reg |= ATTR_CFG_GET_FLD(attr, pa_enable) << SYS_PMSCR_EL1_PA_SHIFT;
+	reg |= ATTR_CFG_GET_FLD(attr, pct_enable) << SYS_PMSCR_EL1_PCT_SHIFT;
+
+	if (!attr->exclude_user)
+		reg |= BIT(SYS_PMSCR_EL1_E0SPE_SHIFT);
+
+	if (!attr->exclude_kernel)
+		reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);
+
+	if (get_spe_event_has_cx(event))
+		reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);
+
+	return reg;
+}
+
+static void arm_spe_event_sanitise_period(struct perf_event *event)
+{
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
+	u64 period = event->hw.sample_period;
+	u64 max_period = SYS_PMSIRR_EL1_INTERVAL_MASK
+			 << SYS_PMSIRR_EL1_INTERVAL_SHIFT;
+
+	if (period < spe_pmu->min_period)
+		period = spe_pmu->min_period;
+	else if (period > max_period)
+		period = max_period;
+	else
+		period &= max_period;
+
+	event->hw.sample_period = period;
+}
+
+static u64 arm_spe_event_to_pmsirr(struct perf_event *event)
+{
+	struct perf_event_attr *attr = &event->attr;
+	u64 reg = 0;
+
+	arm_spe_event_sanitise_period(event);
+
+	reg |= ATTR_CFG_GET_FLD(attr, jitter) << SYS_PMSIRR_EL1_RND_SHIFT;
+	reg |= event->hw.sample_period;
+
+	return reg;
+}
+
+static u64 arm_spe_event_to_pmsfcr(struct perf_event *event)
+{
+	struct perf_event_attr *attr = &event->attr;
+	u64 reg = 0;
+
+	reg |= ATTR_CFG_GET_FLD(attr, load_filter) << SYS_PMSFCR_EL1_LD_SHIFT;
+	reg |= ATTR_CFG_GET_FLD(attr, store_filter) << SYS_PMSFCR_EL1_ST_SHIFT;
+	reg |= ATTR_CFG_GET_FLD(attr, branch_filter) << SYS_PMSFCR_EL1_B_SHIFT;
+
+	if (reg)
+		reg |= BIT(SYS_PMSFCR_EL1_FT_SHIFT);
+
+	if (ATTR_CFG_GET_FLD(attr, event_filter))
+		reg |= BIT(SYS_PMSFCR_EL1_FE_SHIFT);
+
+	if (ATTR_CFG_GET_FLD(attr, min_latency))
+		reg |= BIT(SYS_PMSFCR_EL1_FL_SHIFT);
+
+	return reg;
+}
+
+static u64 arm_spe_event_to_pmsevfr(struct perf_event *event)
+{
+	struct perf_event_attr *attr = &event->attr;
+	return ATTR_CFG_GET_FLD(attr, event_filter);
+}
+
+static u64 arm_spe_event_to_pmslatfr(struct perf_event *event)
+{
+	struct perf_event_attr *attr = &event->attr;
+	return ATTR_CFG_GET_FLD(attr, min_latency)
+	       << SYS_PMSLATFR_EL1_MINLAT_SHIFT;
+}
+
+static void arm_spe_pmu_pad_buf(struct perf_output_handle *handle, int len)
+{
+	struct arm_spe_pmu_buf *buf = perf_get_aux(handle);
+	u64 head = PERF_IDX2OFF(handle->head, buf);
+
+	memset(buf->base + head, ARM_SPE_BUF_PAD_BYTE, len);
+	if (!buf->snapshot)
+		perf_aux_output_skip(handle, len);
+}
+
+static u64 arm_spe_pmu_next_snapshot_off(struct perf_output_handle *handle)
+{
+	struct arm_spe_pmu_buf *buf = perf_get_aux(handle);
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu);
+	u64 head = PERF_IDX2OFF(handle->head, buf);
+	u64 limit = buf->nr_pages * PAGE_SIZE;
+
+	/*
+	 * The trace format isn't parseable in reverse, so clamp
+	 * the limit to half of the buffer size in snapshot mode
+	 * so that the worst case is half a buffer of records, as
+	 * opposed to a single record.
+	 */
+	if (head < limit >> 1)
+		limit >>= 1;
+
+	/*
+	 * If we're within max_record_sz of the limit, we must
+	 * pad, move the head index and recompute the limit.
+	 */
+	if (limit - head < spe_pmu->max_record_sz) {
+		arm_spe_pmu_pad_buf(handle, limit - head);
+		handle->head = PERF_IDX2OFF(limit, buf);
+		limit = ((buf->nr_pages * PAGE_SIZE) >> 1) + handle->head;
+	}
+
+	return limit;
+}
+
+static u64 __arm_spe_pmu_next_off(struct perf_output_handle *handle)
+{
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu);
+	struct arm_spe_pmu_buf *buf = perf_get_aux(handle);
+	const u64 bufsize = buf->nr_pages * PAGE_SIZE;
+	u64 limit = bufsize;
+	u64 head, tail, wakeup;
+
+	/*
+	 * The head can be misaligned for two reasons:
+	 *
+	 * 1. The hardware left PMBPTR pointing to the first byte after
+	 *    a record when generating a buffer management event.
+	 *
+	 * 2. We used perf_aux_output_skip to consume handle->size bytes
+	 *    and CIRC_SPACE was used to compute the size, which always
+	 *    leaves one entry free.
+	 *
+	 * Deal with this by padding to the next alignment boundary and
+	 * moving the head index. If we run out of buffer space, we'll
+	 * reduce handle->size to zero and end up reporting truncation.
+	 */
+	head = PERF_IDX2OFF(handle->head, buf);
+	if (!IS_ALIGNED(head, spe_pmu->align)) {
+		unsigned long delta = roundup(head, spe_pmu->align) - head;
+
+		delta = min(delta, handle->size);
+		arm_spe_pmu_pad_buf(handle, delta);
+		head = PERF_IDX2OFF(handle->head, buf);
+	}
+
+	/* If we've run out of free space, then nothing more to do */
+	if (!handle->size)
+		goto no_space;
+
+	/* Compute the tail and wakeup indices now that we've aligned head */
+	tail = PERF_IDX2OFF(handle->head + handle->size, buf);
+	wakeup = PERF_IDX2OFF(handle->wakeup, buf);
+
+	/*
+	 * Avoid clobbering unconsumed data. We know we have space, so
+	 * if we see head == tail we know that the buffer is empty. If
+	 * head > tail, then there's nothing to clobber prior to
+	 * wrapping.
+	 */
+	if (head < tail)
+		limit = round_down(tail, PAGE_SIZE);
+
+	/*
+	 * Wakeup may be arbitrarily far into the future. If it's not in
+	 * the current generation, either we'll wrap before hitting it,
+	 * or it's in the past and has been handled already.
+	 *
+	 * If there's a wakeup before we wrap, arrange to be woken up by
+	 * the page boundary following it. Keep the tail boundary if
+	 * that's lower.
+	 */
+	if (handle->wakeup < (handle->head + handle->size) && head <= wakeup)
+		limit = min(limit, round_up(wakeup, PAGE_SIZE));
+
+	if (limit > head)
+		return limit;
+
+	arm_spe_pmu_pad_buf(handle, handle->size);
+no_space:
+	perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED);
+	perf_aux_output_end(handle, 0);
+	return 0;
+}
+
+static u64 arm_spe_pmu_next_off(struct perf_output_handle *handle)
+{
+	struct arm_spe_pmu_buf *buf = perf_get_aux(handle);
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(handle->event->pmu);
+	u64 limit = __arm_spe_pmu_next_off(handle);
+	u64 head = PERF_IDX2OFF(handle->head, buf);
+
+	/*
+	 * If the head has come too close to the end of the buffer,
+	 * then pad to the end and recompute the limit.
+	 */
+	if (limit && (limit - head < spe_pmu->max_record_sz)) {
+		arm_spe_pmu_pad_buf(handle, limit - head);
+		limit = __arm_spe_pmu_next_off(handle);
+	}
+
+	return limit;
+}
+
+static void arm_spe_perf_aux_output_begin(struct perf_output_handle *handle,
+					  struct perf_event *event)
+{
+	u64 base, limit;
+	struct arm_spe_pmu_buf *buf;
+
+	/* Start a new aux session */
+	buf = perf_aux_output_begin(handle, event);
+	if (!buf) {
+		event->hw.state |= PERF_HES_STOPPED;
+		/*
+		 * We still need to clear the limit pointer, since the
+		 * profiler might only be disabled by virtue of a fault.
+		 */
+		limit = 0;
+		goto out_write_limit;
+	}
+
+	limit = buf->snapshot ? arm_spe_pmu_next_snapshot_off(handle)
+			      : arm_spe_pmu_next_off(handle);
+	if (limit)
+		limit |= BIT(SYS_PMBLIMITR_EL1_E_SHIFT);
+
+	limit += (u64)buf->base;
+	base = (u64)buf->base + PERF_IDX2OFF(handle->head, buf);
+	write_sysreg_s(base, SYS_PMBPTR_EL1);
+
+out_write_limit:
+	write_sysreg_s(limit, SYS_PMBLIMITR_EL1);
+}
+
+static void arm_spe_perf_aux_output_end(struct perf_output_handle *handle)
+{
+	struct arm_spe_pmu_buf *buf = perf_get_aux(handle);
+	u64 offset, size;
+
+	offset = read_sysreg_s(SYS_PMBPTR_EL1) - (u64)buf->base;
+	size = offset - PERF_IDX2OFF(handle->head, buf);
+
+	if (buf->snapshot)
+		handle->head = offset;
+
+	perf_aux_output_end(handle, size);
+}
+
+static void arm_spe_pmu_disable_and_drain_local(void)
+{
+	/* Disable profiling at EL0 and EL1 */
+	write_sysreg_s(0, SYS_PMSCR_EL1);
+	isb();
+
+	/* Drain any buffered data */
+	psb_csync();
+	dsb(nsh);
+
+	/* Disable the profiling buffer */
+	write_sysreg_s(0, SYS_PMBLIMITR_EL1);
+	isb();
+}
+
+/* IRQ handling */
+static enum arm_spe_pmu_buf_fault_action
+arm_spe_pmu_buf_get_fault_act(struct perf_output_handle *handle)
+{
+	const char *err_str;
+	u64 pmbsr;
+	enum arm_spe_pmu_buf_fault_action ret;
+
+	/*
+	 * Ensure new profiling data is visible to the CPU and any external
+	 * aborts have been resolved.
+	 */
+	psb_csync();
+	dsb(nsh);
+
+	/* Ensure hardware updates to PMBPTR_EL1 are visible */
+	isb();
+
+	/* Service required? */
+	pmbsr = read_sysreg_s(SYS_PMBSR_EL1);
+	if (!(pmbsr & BIT(SYS_PMBSR_EL1_S_SHIFT)))
+		return SPE_PMU_BUF_FAULT_ACT_SPURIOUS;
+
+	/*
+	 * If we've lost data, disable profiling and also set the PARTIAL
+	 * flag to indicate that the last record is corrupted.
+	 */
+	if (pmbsr & BIT(SYS_PMBSR_EL1_DL_SHIFT))
+		perf_aux_output_flag(handle, PERF_AUX_FLAG_TRUNCATED |
+					     PERF_AUX_FLAG_PARTIAL);
+
+	/* Report collisions to userspace so that it can up the period */
+	if (pmbsr & BIT(SYS_PMBSR_EL1_COLL_SHIFT))
+		perf_aux_output_flag(handle, PERF_AUX_FLAG_COLLISION);
+
+	/* We only expect buffer management events */
+	switch (pmbsr & (SYS_PMBSR_EL1_EC_MASK << SYS_PMBSR_EL1_EC_SHIFT)) {
+	case SYS_PMBSR_EL1_EC_BUF:
+		/* Handled below */
+		break;
+	case SYS_PMBSR_EL1_EC_FAULT_S1:
+	case SYS_PMBSR_EL1_EC_FAULT_S2:
+		err_str = "Unexpected buffer fault";
+		goto out_err;
+	default:
+		err_str = "Unknown error code";
+		goto out_err;
+	}
+
+	/* Buffer management event */
+	switch (pmbsr &
+		(SYS_PMBSR_EL1_BUF_BSC_MASK << SYS_PMBSR_EL1_BUF_BSC_SHIFT)) {
+	case SYS_PMBSR_EL1_BUF_BSC_FULL:
+		ret = SPE_PMU_BUF_FAULT_ACT_OK;
+		goto out_stop;
+	default:
+		err_str = "Unknown buffer status code";
+	}
+
+out_err:
+	pr_err_ratelimited("%s on CPU %d [PMBSR=0x%016llx, PMBPTR=0x%016llx, PMBLIMITR=0x%016llx]\n",
+			   err_str, smp_processor_id(), pmbsr,
+			   read_sysreg_s(SYS_PMBPTR_EL1),
+			   read_sysreg_s(SYS_PMBLIMITR_EL1));
+	ret = SPE_PMU_BUF_FAULT_ACT_FATAL;
+
+out_stop:
+	arm_spe_perf_aux_output_end(handle);
+	return ret;
+}
+
+static irqreturn_t arm_spe_pmu_irq_handler(int irq, void *dev)
+{
+	struct perf_output_handle *handle = dev;
+	struct perf_event *event = handle->event;
+	enum arm_spe_pmu_buf_fault_action act;
+
+	if (!perf_get_aux(handle))
+		return IRQ_NONE;
+
+	act = arm_spe_pmu_buf_get_fault_act(handle);
+	if (act == SPE_PMU_BUF_FAULT_ACT_SPURIOUS)
+		return IRQ_NONE;
+
+	/*
+	 * Ensure perf callbacks have completed, which may disable the
+	 * profiling buffer in response to a TRUNCATION flag.
+	 */
+	irq_work_run();
+
+	switch (act) {
+	case SPE_PMU_BUF_FAULT_ACT_FATAL:
+		/*
+		 * If a fatal exception occurred then leaving the profiling
+		 * buffer enabled is a recipe waiting to happen. Since
+		 * fatal faults don't always imply truncation, make sure
+		 * that the profiling buffer is disabled explicitly before
+		 * clearing the syndrome register.
+		 */
+		arm_spe_pmu_disable_and_drain_local();
+		break;
+	case SPE_PMU_BUF_FAULT_ACT_OK:
+		/*
+		 * We handled the fault (the buffer was full), so resume
+		 * profiling as long as we didn't detect truncation.
+		 * PMBPTR might be misaligned, but we'll burn that bridge
+		 * when we get to it.
+		 */
+		if (!(handle->aux_flags & PERF_AUX_FLAG_TRUNCATED)) {
+			arm_spe_perf_aux_output_begin(handle, event);
+			isb();
+		}
+		break;
+	case SPE_PMU_BUF_FAULT_ACT_SPURIOUS:
+		/* We've seen you before, but GCC has the memory of a sieve. */
+		break;
+	}
+
+	/* The buffer pointers are now sane, so resume profiling. */
+	write_sysreg_s(0, SYS_PMBSR_EL1);
+	return IRQ_HANDLED;
+}
+
+/* Perf callbacks */
+static int arm_spe_pmu_event_init(struct perf_event *event)
+{
+	u64 reg;
+	struct perf_event_attr *attr = &event->attr;
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
+
+	/* This is, of course, deeply driver-specific */
+	if (attr->type != event->pmu->type)
+		return -ENOENT;
+
+	if (event->cpu >= 0 &&
+	    !cpumask_test_cpu(event->cpu, &spe_pmu->supported_cpus))
+		return -ENOENT;
+
+	if (arm_spe_event_to_pmsevfr(event) & SYS_PMSEVFR_EL1_RES0)
+		return -EOPNOTSUPP;
+
+	if (attr->exclude_idle)
+		return -EOPNOTSUPP;
+
+	/*
+	 * Feedback-directed frequency throttling doesn't work when we
+	 * have a buffer of samples. We'd need to manually count the
+	 * samples in the buffer when it fills up and adjust the event
+	 * count to reflect that. Instead, just force the user to specify
+	 * a sample period.
+	 */
+	if (attr->freq)
+		return -EINVAL;
+
+	reg = arm_spe_event_to_pmsfcr(event);
+	if ((reg & BIT(SYS_PMSFCR_EL1_FE_SHIFT)) &&
+	    !(spe_pmu->features & SPE_PMU_FEAT_FILT_EVT))
+		return -EOPNOTSUPP;
+
+	if ((reg & BIT(SYS_PMSFCR_EL1_FT_SHIFT)) &&
+	    !(spe_pmu->features & SPE_PMU_FEAT_FILT_TYP))
+		return -EOPNOTSUPP;
+
+	if ((reg & BIT(SYS_PMSFCR_EL1_FL_SHIFT)) &&
+	    !(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT))
+		return -EOPNOTSUPP;
+
+	set_spe_event_has_cx(event);
+	reg = arm_spe_event_to_pmscr(event);
+	if (!perfmon_capable() &&
+	    (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
+		    BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
+		return -EACCES;
+
+	return 0;
+}
+
+static void arm_spe_pmu_start(struct perf_event *event, int flags)
+{
+	u64 reg;
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_output_handle *handle = this_cpu_ptr(spe_pmu->handle);
+
+	hwc->state = 0;
+	arm_spe_perf_aux_output_begin(handle, event);
+	if (hwc->state)
+		return;
+
+	reg = arm_spe_event_to_pmsfcr(event);
+	write_sysreg_s(reg, SYS_PMSFCR_EL1);
+
+	reg = arm_spe_event_to_pmsevfr(event);
+	write_sysreg_s(reg, SYS_PMSEVFR_EL1);
+
+	reg = arm_spe_event_to_pmslatfr(event);
+	write_sysreg_s(reg, SYS_PMSLATFR_EL1);
+
+	if (flags & PERF_EF_RELOAD) {
+		reg = arm_spe_event_to_pmsirr(event);
+		write_sysreg_s(reg, SYS_PMSIRR_EL1);
+		isb();
+		reg = local64_read(&hwc->period_left);
+		write_sysreg_s(reg, SYS_PMSICR_EL1);
+	}
+
+	reg = arm_spe_event_to_pmscr(event);
+	isb();
+	write_sysreg_s(reg, SYS_PMSCR_EL1);
+}
+
+static void arm_spe_pmu_stop(struct perf_event *event, int flags)
+{
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_output_handle *handle = this_cpu_ptr(spe_pmu->handle);
+
+	/* If we're already stopped, then nothing to do */
+	if (hwc->state & PERF_HES_STOPPED)
+		return;
+
+	/* Stop all trace generation */
+	arm_spe_pmu_disable_and_drain_local();
+
+	if (flags & PERF_EF_UPDATE) {
+		/*
+		 * If there's a fault pending then ensure we contain it
+		 * to this buffer, since we might be on the context-switch
+		 * path.
+		 */
+		if (perf_get_aux(handle)) {
+			enum arm_spe_pmu_buf_fault_action act;
+
+			act = arm_spe_pmu_buf_get_fault_act(handle);
+			if (act == SPE_PMU_BUF_FAULT_ACT_SPURIOUS)
+				arm_spe_perf_aux_output_end(handle);
+			else
+				write_sysreg_s(0, SYS_PMBSR_EL1);
+		}
+
+		/*
+		 * This may also contain ECOUNT, but nobody else should
+		 * be looking at period_left, since we forbid frequency
+		 * based sampling.
+		 */
+		local64_set(&hwc->period_left, read_sysreg_s(SYS_PMSICR_EL1));
+		hwc->state |= PERF_HES_UPTODATE;
+	}
+
+	hwc->state |= PERF_HES_STOPPED;
+}
+
+static int arm_spe_pmu_add(struct perf_event *event, int flags)
+{
+	int ret = 0;
+	struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int cpu = event->cpu == -1 ? smp_processor_id() : event->cpu;
+
+	if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus))
+		return -ENOENT;
+
+	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	if (flags & PERF_EF_START) {
+		arm_spe_pmu_start(event, PERF_EF_RELOAD);
+		if (hwc->state & PERF_HES_STOPPED)
+			ret = -EINVAL;
+	}
+
+	return ret;
+}
+
+static void arm_spe_pmu_del(struct perf_event *event, int flags)
+{
+	arm_spe_pmu_stop(event, PERF_EF_UPDATE);
+}
+
+static void arm_spe_pmu_read(struct perf_event *event)
+{
+}
+
+static void *arm_spe_pmu_setup_aux(struct perf_event *event, void **pages,
+				   int nr_pages, bool snapshot)
+{
+	int i, cpu = event->cpu;
+	struct page **pglist;
+	struct arm_spe_pmu_buf *buf;
+
+	/* We need at least two pages for this to work. */
+	if (nr_pages < 2)
+		return NULL;
+
+	/*
+	 * We require an even number of pages for snapshot mode, so that
+	 * we can effectively treat the buffer as consisting of two equal
+	 * parts and give userspace a fighting chance of getting some
+	 * useful data out of it.
+	 */
+	if (snapshot && (nr_pages & 1))
+		return NULL;
+
+	if (cpu == -1)
+		cpu = raw_smp_processor_id();
+
+	buf = kzalloc_node(sizeof(*buf), GFP_KERNEL, cpu_to_node(cpu));
+	if (!buf)
+		return NULL;
+
+	pglist = kcalloc(nr_pages, sizeof(*pglist), GFP_KERNEL);
+	if (!pglist)
+		goto out_free_buf;
+
+	for (i = 0; i < nr_pages; ++i)
+		pglist[i] = virt_to_page(pages[i]);
+
+	buf->base = vmap(pglist, nr_pages, VM_MAP, PAGE_KERNEL);
+	if (!buf->base)
+		goto out_free_pglist;
+
+	buf->nr_pages	= nr_pages;
+	buf->snapshot	= snapshot;
+
+	kfree(pglist);
+	return buf;
+
+out_free_pglist:
+	kfree(pglist);
+out_free_buf:
+	kfree(buf);
+	return NULL;
+}
+
+static void arm_spe_pmu_free_aux(void *aux)
+{
+	struct arm_spe_pmu_buf *buf = aux;
+
+	vunmap(buf->base);
+	kfree(buf);
+}
+
+/* Initialisation and teardown functions */
+static int arm_spe_pmu_perf_init(struct arm_spe_pmu *spe_pmu)
+{
+	static atomic_t pmu_idx = ATOMIC_INIT(-1);
+
+	int idx;
+	char *name;
+	struct device *dev = &spe_pmu->pdev->dev;
+
+	spe_pmu->pmu = (struct pmu) {
+		.module = THIS_MODULE,
+		.capabilities	= PERF_PMU_CAP_EXCLUSIVE | PERF_PMU_CAP_ITRACE,
+		.attr_groups	= arm_spe_pmu_attr_groups,
+		/*
+		 * We hitch a ride on the software context here, so that
+		 * we can support per-task profiling (which is not possible
+		 * with the invalid context as it doesn't get sched callbacks).
+		 * This requires that userspace either uses a dummy event for
+		 * perf_event_open, since the aux buffer is not setup until
+		 * a subsequent mmap, or creates the profiling event in a
+		 * disabled state and explicitly PERF_EVENT_IOC_ENABLEs it
+		 * once the buffer has been created.
+		 */
+		.task_ctx_nr	= perf_sw_context,
+		.event_init	= arm_spe_pmu_event_init,
+		.add		= arm_spe_pmu_add,
+		.del		= arm_spe_pmu_del,
+		.start		= arm_spe_pmu_start,
+		.stop		= arm_spe_pmu_stop,
+		.read		= arm_spe_pmu_read,
+		.setup_aux	= arm_spe_pmu_setup_aux,
+		.free_aux	= arm_spe_pmu_free_aux,
+	};
+
+	idx = atomic_inc_return(&pmu_idx);
+	name = devm_kasprintf(dev, GFP_KERNEL, "%s_%d", PMUNAME, idx);
+	if (!name) {
+		dev_err(dev, "failed to allocate name for pmu %d\n", idx);
+		return -ENOMEM;
+	}
+
+	return perf_pmu_register(&spe_pmu->pmu, name, -1);
+}
+
+static void arm_spe_pmu_perf_destroy(struct arm_spe_pmu *spe_pmu)
+{
+	perf_pmu_unregister(&spe_pmu->pmu);
+}
+
+static void __arm_spe_pmu_dev_probe(void *info)
+{
+	int fld;
+	u64 reg;
+	struct arm_spe_pmu *spe_pmu = info;
+	struct device *dev = &spe_pmu->pdev->dev;
+
+	fld = cpuid_feature_extract_unsigned_field(read_cpuid(ID_AA64DFR0_EL1),
+						   ID_AA64DFR0_PMSVER_SHIFT);
+	if (!fld) {
+		dev_err(dev,
+			"unsupported ID_AA64DFR0_EL1.PMSVer [%d] on CPU %d\n",
+			fld, smp_processor_id());
+		return;
+	}
+
+	/* Read PMBIDR first to determine whether or not we have access */
+	reg = read_sysreg_s(SYS_PMBIDR_EL1);
+	if (reg & BIT(SYS_PMBIDR_EL1_P_SHIFT)) {
+		dev_err(dev,
+			"profiling buffer owned by higher exception level\n");
+		return;
+	}
+
+	/* Minimum alignment. If it's out-of-range, then fail the probe */
+	fld = reg >> SYS_PMBIDR_EL1_ALIGN_SHIFT & SYS_PMBIDR_EL1_ALIGN_MASK;
+	spe_pmu->align = 1 << fld;
+	if (spe_pmu->align > SZ_2K) {
+		dev_err(dev, "unsupported PMBIDR.Align [%d] on CPU %d\n",
+			fld, smp_processor_id());
+		return;
+	}
+
+	/* It's now safe to read PMSIDR and figure out what we've got */
+	reg = read_sysreg_s(SYS_PMSIDR_EL1);
+	if (reg & BIT(SYS_PMSIDR_EL1_FE_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_FILT_EVT;
+
+	if (reg & BIT(SYS_PMSIDR_EL1_FT_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_FILT_TYP;
+
+	if (reg & BIT(SYS_PMSIDR_EL1_FL_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_FILT_LAT;
+
+	if (reg & BIT(SYS_PMSIDR_EL1_ARCHINST_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_ARCH_INST;
+
+	if (reg & BIT(SYS_PMSIDR_EL1_LDS_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_LDS;
+
+	if (reg & BIT(SYS_PMSIDR_EL1_ERND_SHIFT))
+		spe_pmu->features |= SPE_PMU_FEAT_ERND;
+
+	/* This field has a spaced out encoding, so just use a look-up */
+	fld = reg >> SYS_PMSIDR_EL1_INTERVAL_SHIFT & SYS_PMSIDR_EL1_INTERVAL_MASK;
+	switch (fld) {
+	case 0:
+		spe_pmu->min_period = 256;
+		break;
+	case 2:
+		spe_pmu->min_period = 512;
+		break;
+	case 3:
+		spe_pmu->min_period = 768;
+		break;
+	case 4:
+		spe_pmu->min_period = 1024;
+		break;
+	case 5:
+		spe_pmu->min_period = 1536;
+		break;
+	case 6:
+		spe_pmu->min_period = 2048;
+		break;
+	case 7:
+		spe_pmu->min_period = 3072;
+		break;
+	default:
+		dev_warn(dev, "unknown PMSIDR_EL1.Interval [%d]; assuming 8\n",
+			 fld);
+		fallthrough;
+	case 8:
+		spe_pmu->min_period = 4096;
+	}
+
+	/* Maximum record size. If it's out-of-range, then fail the probe */
+	fld = reg >> SYS_PMSIDR_EL1_MAXSIZE_SHIFT & SYS_PMSIDR_EL1_MAXSIZE_MASK;
+	spe_pmu->max_record_sz = 1 << fld;
+	if (spe_pmu->max_record_sz > SZ_2K || spe_pmu->max_record_sz < 16) {
+		dev_err(dev, "unsupported PMSIDR_EL1.MaxSize [%d] on CPU %d\n",
+			fld, smp_processor_id());
+		return;
+	}
+
+	fld = reg >> SYS_PMSIDR_EL1_COUNTSIZE_SHIFT & SYS_PMSIDR_EL1_COUNTSIZE_MASK;
+	switch (fld) {
+	default:
+		dev_warn(dev, "unknown PMSIDR_EL1.CountSize [%d]; assuming 2\n",
+			 fld);
+		fallthrough;
+	case 2:
+		spe_pmu->counter_sz = 12;
+	}
+
+	dev_info(dev,
+		 "probed for CPUs %*pbl [max_record_sz %u, align %u, features 0x%llx]\n",
+		 cpumask_pr_args(&spe_pmu->supported_cpus),
+		 spe_pmu->max_record_sz, spe_pmu->align, spe_pmu->features);
+
+	spe_pmu->features |= SPE_PMU_FEAT_DEV_PROBED;
+	return;
+}
+
+static void __arm_spe_pmu_reset_local(void)
+{
+	/*
+	 * This is probably overkill, as we have no idea where we're
+	 * draining any buffered data to...
+	 */
+	arm_spe_pmu_disable_and_drain_local();
+
+	/* Reset the buffer base pointer */
+	write_sysreg_s(0, SYS_PMBPTR_EL1);
+	isb();
+
+	/* Clear any pending management interrupts */
+	write_sysreg_s(0, SYS_PMBSR_EL1);
+	isb();
+}
+
+static void __arm_spe_pmu_setup_one(void *info)
+{
+	struct arm_spe_pmu *spe_pmu = info;
+
+	__arm_spe_pmu_reset_local();
+	enable_percpu_irq(spe_pmu->irq, IRQ_TYPE_NONE);
+}
+
+static void __arm_spe_pmu_stop_one(void *info)
+{
+	struct arm_spe_pmu *spe_pmu = info;
+
+	disable_percpu_irq(spe_pmu->irq);
+	__arm_spe_pmu_reset_local();
+}
+
+static int arm_spe_pmu_cpu_startup(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_spe_pmu *spe_pmu;
+
+	spe_pmu = hlist_entry_safe(node, struct arm_spe_pmu, hotplug_node);
+	if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus))
+		return 0;
+
+	__arm_spe_pmu_setup_one(spe_pmu);
+	return 0;
+}
+
+static int arm_spe_pmu_cpu_teardown(unsigned int cpu, struct hlist_node *node)
+{
+	struct arm_spe_pmu *spe_pmu;
+
+	spe_pmu = hlist_entry_safe(node, struct arm_spe_pmu, hotplug_node);
+	if (!cpumask_test_cpu(cpu, &spe_pmu->supported_cpus))
+		return 0;
+
+	__arm_spe_pmu_stop_one(spe_pmu);
+	return 0;
+}
+
+static int arm_spe_pmu_dev_init(struct arm_spe_pmu *spe_pmu)
+{
+	int ret;
+	cpumask_t *mask = &spe_pmu->supported_cpus;
+
+	/* Make sure we probe the hardware on a relevant CPU */
+	ret = smp_call_function_any(mask,  __arm_spe_pmu_dev_probe, spe_pmu, 1);
+	if (ret || !(spe_pmu->features & SPE_PMU_FEAT_DEV_PROBED))
+		return -ENXIO;
+
+	/* Request our PPIs (note that the IRQ is still disabled) */
+	ret = request_percpu_irq(spe_pmu->irq, arm_spe_pmu_irq_handler, DRVNAME,
+				 spe_pmu->handle);
+	if (ret)
+		return ret;
+
+	/*
+	 * Register our hotplug notifier now so we don't miss any events.
+	 * This will enable the IRQ for any supported CPUs that are already
+	 * up.
+	 */
+	ret = cpuhp_state_add_instance(arm_spe_pmu_online,
+				       &spe_pmu->hotplug_node);
+	if (ret)
+		free_percpu_irq(spe_pmu->irq, spe_pmu->handle);
+
+	return ret;
+}
+
+static void arm_spe_pmu_dev_teardown(struct arm_spe_pmu *spe_pmu)
+{
+	cpuhp_state_remove_instance(arm_spe_pmu_online, &spe_pmu->hotplug_node);
+	free_percpu_irq(spe_pmu->irq, spe_pmu->handle);
+}
+
+/* Driver and device probing */
+static int arm_spe_pmu_irq_probe(struct arm_spe_pmu *spe_pmu)
+{
+	struct platform_device *pdev = spe_pmu->pdev;
+	int irq = platform_get_irq(pdev, 0);
+
+	if (irq < 0)
+		return -ENXIO;
+
+	if (!irq_is_percpu(irq)) {
+		dev_err(&pdev->dev, "expected PPI but got SPI (%d)\n", irq);
+		return -EINVAL;
+	}
+
+	if (irq_get_percpu_devid_partition(irq, &spe_pmu->supported_cpus)) {
+		dev_err(&pdev->dev, "failed to get PPI partition (%d)\n", irq);
+		return -EINVAL;
+	}
+
+	spe_pmu->irq = irq;
+	return 0;
+}
+
+static const struct of_device_id arm_spe_pmu_of_match[] = {
+	{ .compatible = "arm,statistical-profiling-extension-v1", .data = (void *)1 },
+	{ /* Sentinel */ },
+};
+MODULE_DEVICE_TABLE(of, arm_spe_pmu_of_match);
+
+static const struct platform_device_id arm_spe_match[] = {
+	{ ARMV8_SPE_PDEV_NAME, 0},
+	{ }
+};
+MODULE_DEVICE_TABLE(platform, arm_spe_match);
+
+static int arm_spe_pmu_device_probe(struct platform_device *pdev)
+{
+	int ret;
+	struct arm_spe_pmu *spe_pmu;
+	struct device *dev = &pdev->dev;
+
+	/*
+	 * If kernelspace is unmapped when running at EL0, then the SPE
+	 * buffer will fault and prematurely terminate the AUX session.
+	 */
+	if (arm64_kernel_unmapped_at_el0()) {
+		dev_warn_once(dev, "profiling buffer inaccessible. Try passing \"kpti=off\" on the kernel command line\n");
+		return -EPERM;
+	}
+
+	spe_pmu = devm_kzalloc(dev, sizeof(*spe_pmu), GFP_KERNEL);
+	if (!spe_pmu) {
+		dev_err(dev, "failed to allocate spe_pmu\n");
+		return -ENOMEM;
+	}
+
+	spe_pmu->handle = alloc_percpu(typeof(*spe_pmu->handle));
+	if (!spe_pmu->handle)
+		return -ENOMEM;
+
+	spe_pmu->pdev = pdev;
+	platform_set_drvdata(pdev, spe_pmu);
+
+	ret = arm_spe_pmu_irq_probe(spe_pmu);
+	if (ret)
+		goto out_free_handle;
+
+	ret = arm_spe_pmu_dev_init(spe_pmu);
+	if (ret)
+		goto out_free_handle;
+
+	ret = arm_spe_pmu_perf_init(spe_pmu);
+	if (ret)
+		goto out_teardown_dev;
+
+	return 0;
+
+out_teardown_dev:
+	arm_spe_pmu_dev_teardown(spe_pmu);
+out_free_handle:
+	free_percpu(spe_pmu->handle);
+	return ret;
+}
+
+static int arm_spe_pmu_device_remove(struct platform_device *pdev)
+{
+	struct arm_spe_pmu *spe_pmu = platform_get_drvdata(pdev);
+
+	arm_spe_pmu_perf_destroy(spe_pmu);
+	arm_spe_pmu_dev_teardown(spe_pmu);
+	free_percpu(spe_pmu->handle);
+	return 0;
+}
+
+static struct platform_driver arm_spe_pmu_driver = {
+	.id_table = arm_spe_match,
+	.driver	= {
+		.name		= DRVNAME,
+		.of_match_table	= of_match_ptr(arm_spe_pmu_of_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe	= arm_spe_pmu_device_probe,
+	.remove	= arm_spe_pmu_device_remove,
+};
+
+static int __init arm_spe_pmu_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, DRVNAME,
+				      arm_spe_pmu_cpu_startup,
+				      arm_spe_pmu_cpu_teardown);
+	if (ret < 0)
+		return ret;
+	arm_spe_pmu_online = ret;
+
+	ret = platform_driver_register(&arm_spe_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(arm_spe_pmu_online);
+
+	return ret;
+}
+
+static void __exit arm_spe_pmu_exit(void)
+{
+	platform_driver_unregister(&arm_spe_pmu_driver);
+	cpuhp_remove_multi_state(arm_spe_pmu_online);
+}
+
+module_init(arm_spe_pmu_init);
+module_exit(arm_spe_pmu_exit);
+
+MODULE_DESCRIPTION("Perf driver for the ARMv8.2 Statistical Profiling Extension");
+MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/fsl_imx8_ddr_perf.c b/drivers/perf/fsl_imx8_ddr_perf.c
new file mode 100644
index 000000000..8dfb67530
--- /dev/null
+++ b/drivers/perf/fsl_imx8_ddr_perf.c
@@ -0,0 +1,713 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright 2017 NXP
+ * Copyright 2016 Freescale Semiconductor, Inc.
+ */
+
+#include <linux/bitfield.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/of_irq.h>
+#include <linux/perf_event.h>
+#include <linux/slab.h>
+
+#define COUNTER_CNTL		0x0
+#define COUNTER_READ		0x20
+
+#define COUNTER_DPCR1		0x30
+
+#define CNTL_OVER		0x1
+#define CNTL_CLEAR		0x2
+#define CNTL_EN			0x4
+#define CNTL_EN_MASK		0xFFFFFFFB
+#define CNTL_CLEAR_MASK		0xFFFFFFFD
+#define CNTL_OVER_MASK		0xFFFFFFFE
+
+#define CNTL_CSV_SHIFT		24
+#define CNTL_CSV_MASK		(0xFFU << CNTL_CSV_SHIFT)
+
+#define EVENT_CYCLES_ID		0
+#define EVENT_CYCLES_COUNTER	0
+#define NUM_COUNTERS		4
+
+#define AXI_MASKING_REVERT	0xffff0000	/* AXI_MASKING(MSB 16bits) + AXI_ID(LSB 16bits) */
+
+#define to_ddr_pmu(p)		container_of(p, struct ddr_pmu, pmu)
+
+#define DDR_PERF_DEV_NAME	"imx8_ddr"
+#define DDR_CPUHP_CB_NAME	DDR_PERF_DEV_NAME "_perf_pmu"
+
+static DEFINE_IDA(ddr_ida);
+
+/* DDR Perf hardware feature */
+#define DDR_CAP_AXI_ID_FILTER			0x1     /* support AXI ID filter */
+#define DDR_CAP_AXI_ID_FILTER_ENHANCED		0x3     /* support enhanced AXI ID filter */
+
+struct fsl_ddr_devtype_data {
+	unsigned int quirks;    /* quirks needed for different DDR Perf core */
+};
+
+static const struct fsl_ddr_devtype_data imx8_devtype_data;
+
+static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
+	.quirks = DDR_CAP_AXI_ID_FILTER,
+};
+
+static const struct fsl_ddr_devtype_data imx8mp_devtype_data = {
+	.quirks = DDR_CAP_AXI_ID_FILTER_ENHANCED,
+};
+
+static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
+	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
+	{ .compatible = "fsl,imx8m-ddr-pmu", .data = &imx8m_devtype_data},
+	{ .compatible = "fsl,imx8mp-ddr-pmu", .data = &imx8mp_devtype_data},
+	{ /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, imx_ddr_pmu_dt_ids);
+
+struct ddr_pmu {
+	struct pmu pmu;
+	void __iomem *base;
+	unsigned int cpu;
+	struct	hlist_node node;
+	struct	device *dev;
+	struct perf_event *events[NUM_COUNTERS];
+	int active_events;
+	enum cpuhp_state cpuhp_state;
+	const struct fsl_ddr_devtype_data *devtype_data;
+	int irq;
+	int id;
+	int active_counter;
+};
+
+enum ddr_perf_filter_capabilities {
+	PERF_CAP_AXI_ID_FILTER = 0,
+	PERF_CAP_AXI_ID_FILTER_ENHANCED,
+	PERF_CAP_AXI_ID_FEAT_MAX,
+};
+
+static u32 ddr_perf_filter_cap_get(struct ddr_pmu *pmu, int cap)
+{
+	u32 quirks = pmu->devtype_data->quirks;
+
+	switch (cap) {
+	case PERF_CAP_AXI_ID_FILTER:
+		return !!(quirks & DDR_CAP_AXI_ID_FILTER);
+	case PERF_CAP_AXI_ID_FILTER_ENHANCED:
+		quirks &= DDR_CAP_AXI_ID_FILTER_ENHANCED;
+		return quirks == DDR_CAP_AXI_ID_FILTER_ENHANCED;
+	default:
+		WARN(1, "unknown filter cap %d\n", cap);
+	}
+
+	return 0;
+}
+
+static ssize_t ddr_perf_filter_cap_show(struct device *dev,
+					struct device_attribute *attr,
+					char *buf)
+{
+	struct ddr_pmu *pmu = dev_get_drvdata(dev);
+	struct dev_ext_attribute *ea =
+		container_of(attr, struct dev_ext_attribute, attr);
+	int cap = (long)ea->var;
+
+	return snprintf(buf, PAGE_SIZE, "%u\n",
+			ddr_perf_filter_cap_get(pmu, cap));
+}
+
+#define PERF_EXT_ATTR_ENTRY(_name, _func, _var)				\
+	(&((struct dev_ext_attribute) {					\
+		__ATTR(_name, 0444, _func, NULL), (void *)_var		\
+	}).attr.attr)
+
+#define PERF_FILTER_EXT_ATTR_ENTRY(_name, _var)				\
+	PERF_EXT_ATTR_ENTRY(_name, ddr_perf_filter_cap_show, _var)
+
+static struct attribute *ddr_perf_filter_cap_attr[] = {
+	PERF_FILTER_EXT_ATTR_ENTRY(filter, PERF_CAP_AXI_ID_FILTER),
+	PERF_FILTER_EXT_ATTR_ENTRY(enhanced_filter, PERF_CAP_AXI_ID_FILTER_ENHANCED),
+	NULL,
+};
+
+static struct attribute_group ddr_perf_filter_cap_attr_group = {
+	.name = "caps",
+	.attrs = ddr_perf_filter_cap_attr,
+};
+
+static ssize_t ddr_perf_cpumask_show(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	struct ddr_pmu *pmu = dev_get_drvdata(dev);
+
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(pmu->cpu));
+}
+
+static struct device_attribute ddr_perf_cpumask_attr =
+	__ATTR(cpumask, 0444, ddr_perf_cpumask_show, NULL);
+
+static struct attribute *ddr_perf_cpumask_attrs[] = {
+	&ddr_perf_cpumask_attr.attr,
+	NULL,
+};
+
+static struct attribute_group ddr_perf_cpumask_attr_group = {
+	.attrs = ddr_perf_cpumask_attrs,
+};
+
+static ssize_t
+ddr_pmu_event_show(struct device *dev, struct device_attribute *attr,
+		   char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+	return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define IMX8_DDR_PMU_EVENT_ATTR(_name, _id)				\
+	(&((struct perf_pmu_events_attr[]) {				\
+		{ .attr = __ATTR(_name, 0444, ddr_pmu_event_show, NULL),\
+		  .id = _id, }						\
+	})[0].attr.attr)
+
+static struct attribute *ddr_perf_events_attrs[] = {
+	IMX8_DDR_PMU_EVENT_ATTR(cycles, EVENT_CYCLES_ID),
+	IMX8_DDR_PMU_EVENT_ATTR(selfresh, 0x01),
+	IMX8_DDR_PMU_EVENT_ATTR(read-accesses, 0x04),
+	IMX8_DDR_PMU_EVENT_ATTR(write-accesses, 0x05),
+	IMX8_DDR_PMU_EVENT_ATTR(read-queue-depth, 0x08),
+	IMX8_DDR_PMU_EVENT_ATTR(write-queue-depth, 0x09),
+	IMX8_DDR_PMU_EVENT_ATTR(lp-read-credit-cnt, 0x10),
+	IMX8_DDR_PMU_EVENT_ATTR(hp-read-credit-cnt, 0x11),
+	IMX8_DDR_PMU_EVENT_ATTR(write-credit-cnt, 0x12),
+	IMX8_DDR_PMU_EVENT_ATTR(read-command, 0x20),
+	IMX8_DDR_PMU_EVENT_ATTR(write-command, 0x21),
+	IMX8_DDR_PMU_EVENT_ATTR(read-modify-write-command, 0x22),
+	IMX8_DDR_PMU_EVENT_ATTR(hp-read, 0x23),
+	IMX8_DDR_PMU_EVENT_ATTR(hp-req-nocredit, 0x24),
+	IMX8_DDR_PMU_EVENT_ATTR(hp-xact-credit, 0x25),
+	IMX8_DDR_PMU_EVENT_ATTR(lp-req-nocredit, 0x26),
+	IMX8_DDR_PMU_EVENT_ATTR(lp-xact-credit, 0x27),
+	IMX8_DDR_PMU_EVENT_ATTR(wr-xact-credit, 0x29),
+	IMX8_DDR_PMU_EVENT_ATTR(read-cycles, 0x2a),
+	IMX8_DDR_PMU_EVENT_ATTR(write-cycles, 0x2b),
+	IMX8_DDR_PMU_EVENT_ATTR(read-write-transition, 0x30),
+	IMX8_DDR_PMU_EVENT_ATTR(precharge, 0x31),
+	IMX8_DDR_PMU_EVENT_ATTR(activate, 0x32),
+	IMX8_DDR_PMU_EVENT_ATTR(load-mode, 0x33),
+	IMX8_DDR_PMU_EVENT_ATTR(perf-mwr, 0x34),
+	IMX8_DDR_PMU_EVENT_ATTR(read, 0x35),
+	IMX8_DDR_PMU_EVENT_ATTR(read-activate, 0x36),
+	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
+	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
+	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
+	IMX8_DDR_PMU_EVENT_ATTR(axid-read, 0x41),
+	IMX8_DDR_PMU_EVENT_ATTR(axid-write, 0x42),
+	NULL,
+};
+
+static struct attribute_group ddr_perf_events_attr_group = {
+	.name = "events",
+	.attrs = ddr_perf_events_attrs,
+};
+
+PMU_FORMAT_ATTR(event, "config:0-7");
+PMU_FORMAT_ATTR(axi_id, "config1:0-15");
+PMU_FORMAT_ATTR(axi_mask, "config1:16-31");
+
+static struct attribute *ddr_perf_format_attrs[] = {
+	&format_attr_event.attr,
+	&format_attr_axi_id.attr,
+	&format_attr_axi_mask.attr,
+	NULL,
+};
+
+static struct attribute_group ddr_perf_format_attr_group = {
+	.name = "format",
+	.attrs = ddr_perf_format_attrs,
+};
+
+static const struct attribute_group *attr_groups[] = {
+	&ddr_perf_events_attr_group,
+	&ddr_perf_format_attr_group,
+	&ddr_perf_cpumask_attr_group,
+	&ddr_perf_filter_cap_attr_group,
+	NULL,
+};
+
+static bool ddr_perf_is_filtered(struct perf_event *event)
+{
+	return event->attr.config == 0x41 || event->attr.config == 0x42;
+}
+
+static u32 ddr_perf_filter_val(struct perf_event *event)
+{
+	return event->attr.config1;
+}
+
+static bool ddr_perf_filters_compatible(struct perf_event *a,
+					struct perf_event *b)
+{
+	if (!ddr_perf_is_filtered(a))
+		return true;
+	if (!ddr_perf_is_filtered(b))
+		return true;
+	return ddr_perf_filter_val(a) == ddr_perf_filter_val(b);
+}
+
+static bool ddr_perf_is_enhanced_filtered(struct perf_event *event)
+{
+	unsigned int filt;
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+
+	filt = pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER_ENHANCED;
+	return (filt == DDR_CAP_AXI_ID_FILTER_ENHANCED) &&
+		ddr_perf_is_filtered(event);
+}
+
+static u32 ddr_perf_alloc_counter(struct ddr_pmu *pmu, int event)
+{
+	int i;
+
+	/*
+	 * Always map cycle event to counter 0
+	 * Cycles counter is dedicated for cycle event
+	 * can't used for the other events
+	 */
+	if (event == EVENT_CYCLES_ID) {
+		if (pmu->events[EVENT_CYCLES_COUNTER] == NULL)
+			return EVENT_CYCLES_COUNTER;
+		else
+			return -ENOENT;
+	}
+
+	for (i = 1; i < NUM_COUNTERS; i++) {
+		if (pmu->events[i] == NULL)
+			return i;
+	}
+
+	return -ENOENT;
+}
+
+static void ddr_perf_free_counter(struct ddr_pmu *pmu, int counter)
+{
+	pmu->events[counter] = NULL;
+}
+
+static u32 ddr_perf_read_counter(struct ddr_pmu *pmu, int counter)
+{
+	struct perf_event *event = pmu->events[counter];
+	void __iomem *base = pmu->base;
+
+	/*
+	 * return bytes instead of bursts from ddr transaction for
+	 * axid-read and axid-write event if PMU core supports enhanced
+	 * filter.
+	 */
+	base += ddr_perf_is_enhanced_filtered(event) ? COUNTER_DPCR1 :
+						       COUNTER_READ;
+	return readl_relaxed(base + counter * 4);
+}
+
+static int ddr_perf_event_init(struct perf_event *event)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	struct perf_event *sibling;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EOPNOTSUPP;
+
+	if (event->cpu < 0) {
+		dev_warn(pmu->dev, "Can't provide per-task data!\n");
+		return -EOPNOTSUPP;
+	}
+
+	/*
+	 * We must NOT create groups containing mixed PMUs, although software
+	 * events are acceptable (for example to create a CCN group
+	 * periodically read when a hrtimer aka cpu-clock leader triggers).
+	 */
+	if (event->group_leader->pmu != event->pmu &&
+			!is_software_event(event->group_leader))
+		return -EINVAL;
+
+	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
+		if (!ddr_perf_filters_compatible(event, event->group_leader))
+			return -EINVAL;
+		for_each_sibling_event(sibling, event->group_leader) {
+			if (!ddr_perf_filters_compatible(event, sibling))
+				return -EINVAL;
+		}
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (sibling->pmu != event->pmu &&
+				!is_software_event(sibling))
+			return -EINVAL;
+	}
+
+	event->cpu = pmu->cpu;
+	hwc->idx = -1;
+
+	return 0;
+}
+
+
+static void ddr_perf_event_update(struct perf_event *event)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev_raw_count, new_raw_count;
+	int counter = hwc->idx;
+
+	do {
+		prev_raw_count = local64_read(&hwc->prev_count);
+		new_raw_count = ddr_perf_read_counter(pmu, counter);
+	} while (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+			new_raw_count) != prev_raw_count);
+
+	delta = (new_raw_count - prev_raw_count) & 0xFFFFFFFF;
+
+	local64_add(delta, &event->count);
+}
+
+static void ddr_perf_counter_enable(struct ddr_pmu *pmu, int config,
+				  int counter, bool enable)
+{
+	u8 reg = counter * 4 + COUNTER_CNTL;
+	int val;
+
+	if (enable) {
+		/*
+		 * cycle counter is special which should firstly write 0 then
+		 * write 1 into CLEAR bit to clear it. Other counters only
+		 * need write 0 into CLEAR bit and it turns out to be 1 by
+		 * hardware. Below enable flow is harmless for all counters.
+		 */
+		writel(0, pmu->base + reg);
+		val = CNTL_EN | CNTL_CLEAR;
+		val |= FIELD_PREP(CNTL_CSV_MASK, config);
+		writel(val, pmu->base + reg);
+	} else {
+		/* Disable counter */
+		val = readl_relaxed(pmu->base + reg) & CNTL_EN_MASK;
+		writel(val, pmu->base + reg);
+	}
+}
+
+static void ddr_perf_event_start(struct perf_event *event, int flags)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int counter = hwc->idx;
+
+	local64_set(&hwc->prev_count, 0);
+
+	ddr_perf_counter_enable(pmu, event->attr.config, counter, true);
+
+	if (!pmu->active_counter++)
+		ddr_perf_counter_enable(pmu, EVENT_CYCLES_ID,
+			EVENT_CYCLES_COUNTER, true);
+
+	hwc->state = 0;
+}
+
+static int ddr_perf_event_add(struct perf_event *event, int flags)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int counter;
+	int cfg = event->attr.config;
+	int cfg1 = event->attr.config1;
+
+	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
+		int i;
+
+		for (i = 1; i < NUM_COUNTERS; i++) {
+			if (pmu->events[i] &&
+			    !ddr_perf_filters_compatible(event, pmu->events[i]))
+				return -EINVAL;
+		}
+
+		if (ddr_perf_is_filtered(event)) {
+			/* revert axi id masking(axi_mask) value */
+			cfg1 ^= AXI_MASKING_REVERT;
+			writel(cfg1, pmu->base + COUNTER_DPCR1);
+		}
+	}
+
+	counter = ddr_perf_alloc_counter(pmu, cfg);
+	if (counter < 0) {
+		dev_dbg(pmu->dev, "There are not enough counters\n");
+		return -EOPNOTSUPP;
+	}
+
+	pmu->events[counter] = event;
+	pmu->active_events++;
+	hwc->idx = counter;
+
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (flags & PERF_EF_START)
+		ddr_perf_event_start(event, flags);
+
+	return 0;
+}
+
+static void ddr_perf_event_stop(struct perf_event *event, int flags)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int counter = hwc->idx;
+
+	ddr_perf_counter_enable(pmu, event->attr.config, counter, false);
+	ddr_perf_event_update(event);
+
+	if (!--pmu->active_counter)
+		ddr_perf_counter_enable(pmu, EVENT_CYCLES_ID,
+			EVENT_CYCLES_COUNTER, false);
+
+	hwc->state |= PERF_HES_STOPPED;
+}
+
+static void ddr_perf_event_del(struct perf_event *event, int flags)
+{
+	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int counter = hwc->idx;
+
+	ddr_perf_event_stop(event, PERF_EF_UPDATE);
+
+	ddr_perf_free_counter(pmu, counter);
+	pmu->active_events--;
+	hwc->idx = -1;
+}
+
+static void ddr_perf_pmu_enable(struct pmu *pmu)
+{
+}
+
+static void ddr_perf_pmu_disable(struct pmu *pmu)
+{
+}
+
+static int ddr_perf_init(struct ddr_pmu *pmu, void __iomem *base,
+			 struct device *dev)
+{
+	*pmu = (struct ddr_pmu) {
+		.pmu = (struct pmu) {
+			.module	      = THIS_MODULE,
+			.capabilities = PERF_PMU_CAP_NO_EXCLUDE,
+			.task_ctx_nr = perf_invalid_context,
+			.attr_groups = attr_groups,
+			.event_init  = ddr_perf_event_init,
+			.add	     = ddr_perf_event_add,
+			.del	     = ddr_perf_event_del,
+			.start	     = ddr_perf_event_start,
+			.stop	     = ddr_perf_event_stop,
+			.read	     = ddr_perf_event_update,
+			.pmu_enable  = ddr_perf_pmu_enable,
+			.pmu_disable = ddr_perf_pmu_disable,
+		},
+		.base = base,
+		.dev = dev,
+	};
+
+	pmu->id = ida_simple_get(&ddr_ida, 0, 0, GFP_KERNEL);
+	return pmu->id;
+}
+
+static irqreturn_t ddr_perf_irq_handler(int irq, void *p)
+{
+	int i;
+	struct ddr_pmu *pmu = (struct ddr_pmu *) p;
+	struct perf_event *event, *cycle_event = NULL;
+
+	/* all counter will stop if cycle counter disabled */
+	ddr_perf_counter_enable(pmu,
+			      EVENT_CYCLES_ID,
+			      EVENT_CYCLES_COUNTER,
+			      false);
+	/*
+	 * When the cycle counter overflows, all counters are stopped,
+	 * and an IRQ is raised. If any other counter overflows, it
+	 * continues counting, and no IRQ is raised.
+	 *
+	 * Cycles occur at least 4 times as often as other events, so we
+	 * can update all events on a cycle counter overflow and not
+	 * lose events.
+	 *
+	 */
+	for (i = 0; i < NUM_COUNTERS; i++) {
+
+		if (!pmu->events[i])
+			continue;
+
+		event = pmu->events[i];
+
+		ddr_perf_event_update(event);
+
+		if (event->hw.idx == EVENT_CYCLES_COUNTER)
+			cycle_event = event;
+	}
+
+	ddr_perf_counter_enable(pmu,
+			      EVENT_CYCLES_ID,
+			      EVENT_CYCLES_COUNTER,
+			      true);
+	if (cycle_event)
+		ddr_perf_event_update(cycle_event);
+
+	return IRQ_HANDLED;
+}
+
+static int ddr_perf_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct ddr_pmu *pmu = hlist_entry_safe(node, struct ddr_pmu, node);
+	int target;
+
+	if (cpu != pmu->cpu)
+		return 0;
+
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	perf_pmu_migrate_context(&pmu->pmu, cpu, target);
+	pmu->cpu = target;
+
+	WARN_ON(irq_set_affinity_hint(pmu->irq, cpumask_of(pmu->cpu)));
+
+	return 0;
+}
+
+static int ddr_perf_probe(struct platform_device *pdev)
+{
+	struct ddr_pmu *pmu;
+	struct device_node *np;
+	void __iomem *base;
+	char *name;
+	int num;
+	int ret;
+	int irq;
+
+	base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(base))
+		return PTR_ERR(base);
+
+	np = pdev->dev.of_node;
+
+	pmu = devm_kzalloc(&pdev->dev, sizeof(*pmu), GFP_KERNEL);
+	if (!pmu)
+		return -ENOMEM;
+
+	num = ddr_perf_init(pmu, base, &pdev->dev);
+
+	platform_set_drvdata(pdev, pmu);
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, DDR_PERF_DEV_NAME "%d",
+			      num);
+	if (!name) {
+		ret = -ENOMEM;
+		goto cpuhp_state_err;
+	}
+
+	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
+
+	pmu->cpu = raw_smp_processor_id();
+	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
+				      DDR_CPUHP_CB_NAME,
+				      NULL,
+				      ddr_perf_offline_cpu);
+
+	if (ret < 0) {
+		dev_err(&pdev->dev, "cpuhp_setup_state_multi failed\n");
+		goto cpuhp_state_err;
+	}
+
+	pmu->cpuhp_state = ret;
+
+	/* Register the pmu instance for cpu hotplug */
+	ret = cpuhp_state_add_instance_nocalls(pmu->cpuhp_state, &pmu->node);
+	if (ret) {
+		dev_err(&pdev->dev, "Error %d registering hotplug\n", ret);
+		goto cpuhp_instance_err;
+	}
+
+	/* Request irq */
+	irq = of_irq_get(np, 0);
+	if (irq < 0) {
+		dev_err(&pdev->dev, "Failed to get irq: %d", irq);
+		ret = irq;
+		goto ddr_perf_err;
+	}
+
+	ret = devm_request_irq(&pdev->dev, irq,
+					ddr_perf_irq_handler,
+					IRQF_NOBALANCING | IRQF_NO_THREAD,
+					DDR_CPUHP_CB_NAME,
+					pmu);
+	if (ret < 0) {
+		dev_err(&pdev->dev, "Request irq failed: %d", ret);
+		goto ddr_perf_err;
+	}
+
+	pmu->irq = irq;
+	ret = irq_set_affinity_hint(pmu->irq, cpumask_of(pmu->cpu));
+	if (ret) {
+		dev_err(pmu->dev, "Failed to set interrupt affinity!\n");
+		goto ddr_perf_err;
+	}
+
+	ret = perf_pmu_register(&pmu->pmu, name, -1);
+	if (ret)
+		goto ddr_perf_err;
+
+	return 0;
+
+ddr_perf_err:
+	cpuhp_state_remove_instance_nocalls(pmu->cpuhp_state, &pmu->node);
+cpuhp_instance_err:
+	cpuhp_remove_multi_state(pmu->cpuhp_state);
+cpuhp_state_err:
+	ida_simple_remove(&ddr_ida, pmu->id);
+	dev_warn(&pdev->dev, "i.MX8 DDR Perf PMU failed (%d), disabled\n", ret);
+	return ret;
+}
+
+static int ddr_perf_remove(struct platform_device *pdev)
+{
+	struct ddr_pmu *pmu = platform_get_drvdata(pdev);
+
+	cpuhp_state_remove_instance_nocalls(pmu->cpuhp_state, &pmu->node);
+	cpuhp_remove_multi_state(pmu->cpuhp_state);
+	irq_set_affinity_hint(pmu->irq, NULL);
+
+	perf_pmu_unregister(&pmu->pmu);
+
+	ida_simple_remove(&ddr_ida, pmu->id);
+	return 0;
+}
+
+static struct platform_driver imx_ddr_pmu_driver = {
+	.driver         = {
+		.name   = "imx-ddr-pmu",
+		.of_match_table = imx_ddr_pmu_dt_ids,
+		.suppress_bind_attrs = true,
+	},
+	.probe          = ddr_perf_probe,
+	.remove         = ddr_perf_remove,
+};
+
+module_platform_driver(imx_ddr_pmu_driver);
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/hisilicon/Kconfig b/drivers/perf/hisilicon/Kconfig
new file mode 100644
index 000000000..c5d1b7019
--- /dev/null
+++ b/drivers/perf/hisilicon/Kconfig
@@ -0,0 +1,7 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config HISI_PMU
+	tristate "HiSilicon SoC PMU drivers"
+	depends on ARM64 && ACPI
+	  help
+	  Support for HiSilicon SoC L3 Cache performance monitor, Hydra Home
+	  Agent performance monitor and DDR Controller performance monitor.
diff --git a/drivers/perf/hisilicon/Makefile b/drivers/perf/hisilicon/Makefile
new file mode 100644
index 000000000..e83770618
--- /dev/null
+++ b/drivers/perf/hisilicon/Makefile
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+obj-$(CONFIG_HISI_PMU) += hisi_uncore_pmu.o hisi_uncore_l3c_pmu.o \
+			  hisi_uncore_hha_pmu.o hisi_uncore_ddrc_pmu.o
diff --git a/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
new file mode 100644
index 000000000..5e3645c96
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_ddrc_pmu.c
@@ -0,0 +1,460 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * HiSilicon SoC DDRC uncore Hardware event counters support
+ *
+ * Copyright (C) 2017 Hisilicon Limited
+ * Author: Shaokun Zhang <zhangshaokun@hisilicon.com>
+ *         Anurup M <anurup.m@huawei.com>
+ *
+ * This code is based on the uncore PMUs like arm-cci and arm-ccn.
+ */
+#include <linux/acpi.h>
+#include <linux/bug.h>
+#include <linux/cpuhotplug.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/list.h>
+#include <linux/platform_device.h>
+#include <linux/smp.h>
+
+#include "hisi_uncore_pmu.h"
+
+/* DDRC register definition */
+#define DDRC_PERF_CTRL		0x010
+#define DDRC_FLUX_WR		0x380
+#define DDRC_FLUX_RD		0x384
+#define DDRC_FLUX_WCMD          0x388
+#define DDRC_FLUX_RCMD          0x38c
+#define DDRC_PRE_CMD            0x3c0
+#define DDRC_ACT_CMD            0x3c4
+#define DDRC_RNK_CHG            0x3cc
+#define DDRC_RW_CHG             0x3d0
+#define DDRC_EVENT_CTRL         0x6C0
+#define DDRC_INT_MASK		0x6c8
+#define DDRC_INT_STATUS		0x6cc
+#define DDRC_INT_CLEAR		0x6d0
+
+/* DDRC has 8-counters */
+#define DDRC_NR_COUNTERS	0x8
+#define DDRC_PERF_CTRL_EN	0x2
+
+/*
+ * For DDRC PMU, there are eight-events and every event has been mapped
+ * to fixed-purpose counters which register offset is not consistent.
+ * Therefore there is no write event type and we assume that event
+ * code (0 to 7) is equal to counter index in PMU driver.
+ */
+#define GET_DDRC_EVENTID(hwc)	(hwc->config_base & 0x7)
+
+static const u32 ddrc_reg_off[] = {
+	DDRC_FLUX_WR, DDRC_FLUX_RD, DDRC_FLUX_WCMD, DDRC_FLUX_RCMD,
+	DDRC_PRE_CMD, DDRC_ACT_CMD, DDRC_RNK_CHG, DDRC_RW_CHG
+};
+
+/*
+ * Select the counter register offset using the counter index.
+ * In DDRC there are no programmable counter, the count
+ * is readed form the statistics counter register itself.
+ */
+static u32 hisi_ddrc_pmu_get_counter_offset(int cntr_idx)
+{
+	return ddrc_reg_off[cntr_idx];
+}
+
+static u64 hisi_ddrc_pmu_read_counter(struct hisi_pmu *ddrc_pmu,
+				      struct hw_perf_event *hwc)
+{
+	/* Use event code as counter index */
+	u32 idx = GET_DDRC_EVENTID(hwc);
+
+	if (!hisi_uncore_pmu_counter_valid(ddrc_pmu, idx)) {
+		dev_err(ddrc_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return 0;
+	}
+
+	return readl(ddrc_pmu->base + hisi_ddrc_pmu_get_counter_offset(idx));
+}
+
+static void hisi_ddrc_pmu_write_counter(struct hisi_pmu *ddrc_pmu,
+					struct hw_perf_event *hwc, u64 val)
+{
+	u32 idx = GET_DDRC_EVENTID(hwc);
+
+	if (!hisi_uncore_pmu_counter_valid(ddrc_pmu, idx)) {
+		dev_err(ddrc_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return;
+	}
+
+	writel((u32)val,
+	       ddrc_pmu->base + hisi_ddrc_pmu_get_counter_offset(idx));
+}
+
+/*
+ * For DDRC PMU, event has been mapped to fixed-purpose counter by hardware,
+ * so there is no need to write event type.
+ */
+static void hisi_ddrc_pmu_write_evtype(struct hisi_pmu *hha_pmu, int idx,
+				       u32 type)
+{
+}
+
+static void hisi_ddrc_pmu_start_counters(struct hisi_pmu *ddrc_pmu)
+{
+	u32 val;
+
+	/* Set perf_enable in DDRC_PERF_CTRL to start event counting */
+	val = readl(ddrc_pmu->base + DDRC_PERF_CTRL);
+	val |= DDRC_PERF_CTRL_EN;
+	writel(val, ddrc_pmu->base + DDRC_PERF_CTRL);
+}
+
+static void hisi_ddrc_pmu_stop_counters(struct hisi_pmu *ddrc_pmu)
+{
+	u32 val;
+
+	/* Clear perf_enable in DDRC_PERF_CTRL to stop event counting */
+	val = readl(ddrc_pmu->base + DDRC_PERF_CTRL);
+	val &= ~DDRC_PERF_CTRL_EN;
+	writel(val, ddrc_pmu->base + DDRC_PERF_CTRL);
+}
+
+static void hisi_ddrc_pmu_enable_counter(struct hisi_pmu *ddrc_pmu,
+					 struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Set counter index(event code) in DDRC_EVENT_CTRL register */
+	val = readl(ddrc_pmu->base + DDRC_EVENT_CTRL);
+	val |= (1 << GET_DDRC_EVENTID(hwc));
+	writel(val, ddrc_pmu->base + DDRC_EVENT_CTRL);
+}
+
+static void hisi_ddrc_pmu_disable_counter(struct hisi_pmu *ddrc_pmu,
+					  struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Clear counter index(event code) in DDRC_EVENT_CTRL register */
+	val = readl(ddrc_pmu->base + DDRC_EVENT_CTRL);
+	val &= ~(1 << GET_DDRC_EVENTID(hwc));
+	writel(val, ddrc_pmu->base + DDRC_EVENT_CTRL);
+}
+
+static int hisi_ddrc_pmu_get_event_idx(struct perf_event *event)
+{
+	struct hisi_pmu *ddrc_pmu = to_hisi_pmu(event->pmu);
+	unsigned long *used_mask = ddrc_pmu->pmu_events.used_mask;
+	struct hw_perf_event *hwc = &event->hw;
+	/* For DDRC PMU, we use event code as counter index */
+	int idx = GET_DDRC_EVENTID(hwc);
+
+	if (test_bit(idx, used_mask))
+		return -EAGAIN;
+
+	set_bit(idx, used_mask);
+
+	return idx;
+}
+
+static void hisi_ddrc_pmu_enable_counter_int(struct hisi_pmu *ddrc_pmu,
+					     struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Write 0 to enable interrupt */
+	val = readl(ddrc_pmu->base + DDRC_INT_MASK);
+	val &= ~(1 << GET_DDRC_EVENTID(hwc));
+	writel(val, ddrc_pmu->base + DDRC_INT_MASK);
+}
+
+static void hisi_ddrc_pmu_disable_counter_int(struct hisi_pmu *ddrc_pmu,
+					      struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Write 1 to mask interrupt */
+	val = readl(ddrc_pmu->base + DDRC_INT_MASK);
+	val |= (1 << GET_DDRC_EVENTID(hwc));
+	writel(val, ddrc_pmu->base + DDRC_INT_MASK);
+}
+
+static irqreturn_t hisi_ddrc_pmu_isr(int irq, void *dev_id)
+{
+	struct hisi_pmu *ddrc_pmu = dev_id;
+	struct perf_event *event;
+	unsigned long overflown;
+	int idx;
+
+	/* Read the DDRC_INT_STATUS register */
+	overflown = readl(ddrc_pmu->base + DDRC_INT_STATUS);
+	if (!overflown)
+		return IRQ_NONE;
+
+	/*
+	 * Find the counter index which overflowed if the bit was set
+	 * and handle it
+	 */
+	for_each_set_bit(idx, &overflown, DDRC_NR_COUNTERS) {
+		/* Write 1 to clear the IRQ status flag */
+		writel((1 << idx), ddrc_pmu->base + DDRC_INT_CLEAR);
+
+		/* Get the corresponding event struct */
+		event = ddrc_pmu->pmu_events.hw_events[idx];
+		if (!event)
+			continue;
+
+		hisi_uncore_pmu_event_update(event);
+		hisi_uncore_pmu_set_event_period(event);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static int hisi_ddrc_pmu_init_irq(struct hisi_pmu *ddrc_pmu,
+				  struct platform_device *pdev)
+{
+	int irq, ret;
+
+	/* Read and init IRQ */
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	ret = devm_request_irq(&pdev->dev, irq, hisi_ddrc_pmu_isr,
+			       IRQF_NOBALANCING | IRQF_NO_THREAD,
+			       dev_name(&pdev->dev), ddrc_pmu);
+	if (ret < 0) {
+		dev_err(&pdev->dev,
+			"Fail to request IRQ:%d ret:%d\n", irq, ret);
+		return ret;
+	}
+
+	ddrc_pmu->irq = irq;
+
+	return 0;
+}
+
+static const struct acpi_device_id hisi_ddrc_pmu_acpi_match[] = {
+	{ "HISI0233", },
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, hisi_ddrc_pmu_acpi_match);
+
+static int hisi_ddrc_pmu_init_data(struct platform_device *pdev,
+				   struct hisi_pmu *ddrc_pmu)
+{
+	/*
+	 * Use the SCCL_ID and DDRC channel ID to identify the
+	 * DDRC PMU, while SCCL_ID is in MPIDR[aff2].
+	 */
+	if (device_property_read_u32(&pdev->dev, "hisilicon,ch-id",
+				     &ddrc_pmu->index_id)) {
+		dev_err(&pdev->dev, "Can not read ddrc channel-id!\n");
+		return -EINVAL;
+	}
+
+	if (device_property_read_u32(&pdev->dev, "hisilicon,scl-id",
+				     &ddrc_pmu->sccl_id)) {
+		dev_err(&pdev->dev, "Can not read ddrc sccl-id!\n");
+		return -EINVAL;
+	}
+	/* DDRC PMUs only share the same SCCL */
+	ddrc_pmu->ccl_id = -1;
+
+	ddrc_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(ddrc_pmu->base)) {
+		dev_err(&pdev->dev, "ioremap failed for ddrc_pmu resource\n");
+		return PTR_ERR(ddrc_pmu->base);
+	}
+
+	return 0;
+}
+
+static struct attribute *hisi_ddrc_pmu_format_attr[] = {
+	HISI_PMU_FORMAT_ATTR(event, "config:0-4"),
+	NULL,
+};
+
+static const struct attribute_group hisi_ddrc_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_ddrc_pmu_format_attr,
+};
+
+static struct attribute *hisi_ddrc_pmu_events_attr[] = {
+	HISI_PMU_EVENT_ATTR(flux_wr,		0x00),
+	HISI_PMU_EVENT_ATTR(flux_rd,		0x01),
+	HISI_PMU_EVENT_ATTR(flux_wcmd,		0x02),
+	HISI_PMU_EVENT_ATTR(flux_rcmd,		0x03),
+	HISI_PMU_EVENT_ATTR(pre_cmd,		0x04),
+	HISI_PMU_EVENT_ATTR(act_cmd,		0x05),
+	HISI_PMU_EVENT_ATTR(rnk_chg,		0x06),
+	HISI_PMU_EVENT_ATTR(rw_chg,		0x07),
+	NULL,
+};
+
+static const struct attribute_group hisi_ddrc_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_ddrc_pmu_events_attr,
+};
+
+static DEVICE_ATTR(cpumask, 0444, hisi_cpumask_sysfs_show, NULL);
+
+static struct attribute *hisi_ddrc_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static const struct attribute_group hisi_ddrc_pmu_cpumask_attr_group = {
+	.attrs = hisi_ddrc_pmu_cpumask_attrs,
+};
+
+static const struct attribute_group *hisi_ddrc_pmu_attr_groups[] = {
+	&hisi_ddrc_pmu_format_group,
+	&hisi_ddrc_pmu_events_group,
+	&hisi_ddrc_pmu_cpumask_attr_group,
+	NULL,
+};
+
+static const struct hisi_uncore_ops hisi_uncore_ddrc_ops = {
+	.write_evtype           = hisi_ddrc_pmu_write_evtype,
+	.get_event_idx		= hisi_ddrc_pmu_get_event_idx,
+	.start_counters		= hisi_ddrc_pmu_start_counters,
+	.stop_counters		= hisi_ddrc_pmu_stop_counters,
+	.enable_counter		= hisi_ddrc_pmu_enable_counter,
+	.disable_counter	= hisi_ddrc_pmu_disable_counter,
+	.enable_counter_int	= hisi_ddrc_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_ddrc_pmu_disable_counter_int,
+	.write_counter		= hisi_ddrc_pmu_write_counter,
+	.read_counter		= hisi_ddrc_pmu_read_counter,
+};
+
+static int hisi_ddrc_pmu_dev_probe(struct platform_device *pdev,
+				   struct hisi_pmu *ddrc_pmu)
+{
+	int ret;
+
+	ret = hisi_ddrc_pmu_init_data(pdev, ddrc_pmu);
+	if (ret)
+		return ret;
+
+	ret = hisi_ddrc_pmu_init_irq(ddrc_pmu, pdev);
+	if (ret)
+		return ret;
+
+	ddrc_pmu->num_counters = DDRC_NR_COUNTERS;
+	ddrc_pmu->counter_bits = 32;
+	ddrc_pmu->ops = &hisi_uncore_ddrc_ops;
+	ddrc_pmu->dev = &pdev->dev;
+	ddrc_pmu->on_cpu = -1;
+	ddrc_pmu->check_event = 7;
+
+	return 0;
+}
+
+static int hisi_ddrc_pmu_probe(struct platform_device *pdev)
+{
+	struct hisi_pmu *ddrc_pmu;
+	char *name;
+	int ret;
+
+	ddrc_pmu = devm_kzalloc(&pdev->dev, sizeof(*ddrc_pmu), GFP_KERNEL);
+	if (!ddrc_pmu)
+		return -ENOMEM;
+
+	platform_set_drvdata(pdev, ddrc_pmu);
+
+	ret = hisi_ddrc_pmu_dev_probe(pdev, ddrc_pmu);
+	if (ret)
+		return ret;
+
+	ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE,
+				       &ddrc_pmu->node);
+	if (ret) {
+		dev_err(&pdev->dev, "Error %d registering hotplug;\n", ret);
+		return ret;
+	}
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "hisi_sccl%u_ddrc%u",
+			      ddrc_pmu->sccl_id, ddrc_pmu->index_id);
+	ddrc_pmu->pmu = (struct pmu) {
+		.name		= name,
+		.module		= THIS_MODULE,
+		.task_ctx_nr	= perf_invalid_context,
+		.event_init	= hisi_uncore_pmu_event_init,
+		.pmu_enable	= hisi_uncore_pmu_enable,
+		.pmu_disable	= hisi_uncore_pmu_disable,
+		.add		= hisi_uncore_pmu_add,
+		.del		= hisi_uncore_pmu_del,
+		.start		= hisi_uncore_pmu_start,
+		.stop		= hisi_uncore_pmu_stop,
+		.read		= hisi_uncore_pmu_read,
+		.attr_groups	= hisi_ddrc_pmu_attr_groups,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	ret = perf_pmu_register(&ddrc_pmu->pmu, name, -1);
+	if (ret) {
+		dev_err(ddrc_pmu->dev, "DDRC PMU register failed!\n");
+		cpuhp_state_remove_instance_nocalls(
+			CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE, &ddrc_pmu->node);
+		irq_set_affinity_hint(ddrc_pmu->irq, NULL);
+	}
+
+	return ret;
+}
+
+static int hisi_ddrc_pmu_remove(struct platform_device *pdev)
+{
+	struct hisi_pmu *ddrc_pmu = platform_get_drvdata(pdev);
+
+	perf_pmu_unregister(&ddrc_pmu->pmu);
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE,
+					    &ddrc_pmu->node);
+	irq_set_affinity_hint(ddrc_pmu->irq, NULL);
+
+	return 0;
+}
+
+static struct platform_driver hisi_ddrc_pmu_driver = {
+	.driver = {
+		.name = "hisi_ddrc_pmu",
+		.acpi_match_table = ACPI_PTR(hisi_ddrc_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = hisi_ddrc_pmu_probe,
+	.remove = hisi_ddrc_pmu_remove,
+};
+
+static int __init hisi_ddrc_pmu_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE,
+				      "AP_PERF_ARM_HISI_DDRC_ONLINE",
+				      hisi_uncore_pmu_online_cpu,
+				      hisi_uncore_pmu_offline_cpu);
+	if (ret) {
+		pr_err("DDRC PMU: setup hotplug, ret = %d\n", ret);
+		return ret;
+	}
+
+	ret = platform_driver_register(&hisi_ddrc_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE);
+
+	return ret;
+}
+module_init(hisi_ddrc_pmu_module_init);
+
+static void __exit hisi_ddrc_pmu_module_exit(void)
+{
+	platform_driver_unregister(&hisi_ddrc_pmu_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_DDRC_ONLINE);
+
+}
+module_exit(hisi_ddrc_pmu_module_exit);
+
+MODULE_DESCRIPTION("HiSilicon SoC DDRC uncore PMU driver");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Shaokun Zhang <zhangshaokun@hisilicon.com>");
+MODULE_AUTHOR("Anurup M <anurup.m@huawei.com>");
diff --git a/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
new file mode 100644
index 000000000..5eb816802
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_hha_pmu.c
@@ -0,0 +1,471 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * HiSilicon SoC HHA uncore Hardware event counters support
+ *
+ * Copyright (C) 2017 Hisilicon Limited
+ * Author: Shaokun Zhang <zhangshaokun@hisilicon.com>
+ *         Anurup M <anurup.m@huawei.com>
+ *
+ * This code is based on the uncore PMUs like arm-cci and arm-ccn.
+ */
+#include <linux/acpi.h>
+#include <linux/bug.h>
+#include <linux/cpuhotplug.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/list.h>
+#include <linux/platform_device.h>
+#include <linux/smp.h>
+
+#include "hisi_uncore_pmu.h"
+
+/* HHA register definition */
+#define HHA_INT_MASK		0x0804
+#define HHA_INT_STATUS		0x0808
+#define HHA_INT_CLEAR		0x080C
+#define HHA_PERF_CTRL		0x1E00
+#define HHA_EVENT_CTRL		0x1E04
+#define HHA_EVENT_TYPE0		0x1E80
+/*
+ * Each counter is 48-bits and [48:63] are reserved
+ * which are Read-As-Zero and Writes-Ignored.
+ */
+#define HHA_CNT0_LOWER		0x1F00
+
+/* HHA has 16-counters */
+#define HHA_NR_COUNTERS		0x10
+
+#define HHA_PERF_CTRL_EN	0x1
+#define HHA_EVTYPE_NONE		0xff
+
+/*
+ * Select the counter register offset using the counter index
+ * each counter is 48-bits.
+ */
+static u32 hisi_hha_pmu_get_counter_offset(int cntr_idx)
+{
+	return (HHA_CNT0_LOWER + (cntr_idx * 8));
+}
+
+static u64 hisi_hha_pmu_read_counter(struct hisi_pmu *hha_pmu,
+				     struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+
+	if (!hisi_uncore_pmu_counter_valid(hha_pmu, idx)) {
+		dev_err(hha_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return 0;
+	}
+
+	/* Read 64 bits and like L3C, top 16 bits are RAZ */
+	return readq(hha_pmu->base + hisi_hha_pmu_get_counter_offset(idx));
+}
+
+static void hisi_hha_pmu_write_counter(struct hisi_pmu *hha_pmu,
+				       struct hw_perf_event *hwc, u64 val)
+{
+	u32 idx = hwc->idx;
+
+	if (!hisi_uncore_pmu_counter_valid(hha_pmu, idx)) {
+		dev_err(hha_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return;
+	}
+
+	/* Write 64 bits and like L3C, top 16 bits are WI */
+	writeq(val, hha_pmu->base + hisi_hha_pmu_get_counter_offset(idx));
+}
+
+static void hisi_hha_pmu_write_evtype(struct hisi_pmu *hha_pmu, int idx,
+				      u32 type)
+{
+	u32 reg, reg_idx, shift, val;
+
+	/*
+	 * Select the appropriate event select register(HHA_EVENT_TYPEx).
+	 * There are 4 event select registers for the 16 hardware counters.
+	 * Event code is 8-bits and for the first 4 hardware counters,
+	 * HHA_EVENT_TYPE0 is chosen. For the next 4 hardware counters,
+	 * HHA_EVENT_TYPE1 is chosen and so on.
+	 */
+	reg = HHA_EVENT_TYPE0 + 4 * (idx / 4);
+	reg_idx = idx % 4;
+	shift = 8 * reg_idx;
+
+	/* Write event code to HHA_EVENT_TYPEx register */
+	val = readl(hha_pmu->base + reg);
+	val &= ~(HHA_EVTYPE_NONE << shift);
+	val |= (type << shift);
+	writel(val, hha_pmu->base + reg);
+}
+
+static void hisi_hha_pmu_start_counters(struct hisi_pmu *hha_pmu)
+{
+	u32 val;
+
+	/*
+	 * Set perf_enable bit in HHA_PERF_CTRL to start event
+	 * counting for all enabled counters.
+	 */
+	val = readl(hha_pmu->base + HHA_PERF_CTRL);
+	val |= HHA_PERF_CTRL_EN;
+	writel(val, hha_pmu->base + HHA_PERF_CTRL);
+}
+
+static void hisi_hha_pmu_stop_counters(struct hisi_pmu *hha_pmu)
+{
+	u32 val;
+
+	/*
+	 * Clear perf_enable bit in HHA_PERF_CTRL to stop event
+	 * counting for all enabled counters.
+	 */
+	val = readl(hha_pmu->base + HHA_PERF_CTRL);
+	val &= ~(HHA_PERF_CTRL_EN);
+	writel(val, hha_pmu->base + HHA_PERF_CTRL);
+}
+
+static void hisi_hha_pmu_enable_counter(struct hisi_pmu *hha_pmu,
+					struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Enable counter index in HHA_EVENT_CTRL register */
+	val = readl(hha_pmu->base + HHA_EVENT_CTRL);
+	val |= (1 << hwc->idx);
+	writel(val, hha_pmu->base + HHA_EVENT_CTRL);
+}
+
+static void hisi_hha_pmu_disable_counter(struct hisi_pmu *hha_pmu,
+					 struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Clear counter index in HHA_EVENT_CTRL register */
+	val = readl(hha_pmu->base + HHA_EVENT_CTRL);
+	val &= ~(1 << hwc->idx);
+	writel(val, hha_pmu->base + HHA_EVENT_CTRL);
+}
+
+static void hisi_hha_pmu_enable_counter_int(struct hisi_pmu *hha_pmu,
+					    struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Write 0 to enable interrupt */
+	val = readl(hha_pmu->base + HHA_INT_MASK);
+	val &= ~(1 << hwc->idx);
+	writel(val, hha_pmu->base + HHA_INT_MASK);
+}
+
+static void hisi_hha_pmu_disable_counter_int(struct hisi_pmu *hha_pmu,
+					     struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Write 1 to mask interrupt */
+	val = readl(hha_pmu->base + HHA_INT_MASK);
+	val |= (1 << hwc->idx);
+	writel(val, hha_pmu->base + HHA_INT_MASK);
+}
+
+static irqreturn_t hisi_hha_pmu_isr(int irq, void *dev_id)
+{
+	struct hisi_pmu *hha_pmu = dev_id;
+	struct perf_event *event;
+	unsigned long overflown;
+	int idx;
+
+	/* Read HHA_INT_STATUS register */
+	overflown = readl(hha_pmu->base + HHA_INT_STATUS);
+	if (!overflown)
+		return IRQ_NONE;
+
+	/*
+	 * Find the counter index which overflowed if the bit was set
+	 * and handle it
+	 */
+	for_each_set_bit(idx, &overflown, HHA_NR_COUNTERS) {
+		/* Write 1 to clear the IRQ status flag */
+		writel((1 << idx), hha_pmu->base + HHA_INT_CLEAR);
+
+		/* Get the corresponding event struct */
+		event = hha_pmu->pmu_events.hw_events[idx];
+		if (!event)
+			continue;
+
+		hisi_uncore_pmu_event_update(event);
+		hisi_uncore_pmu_set_event_period(event);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static int hisi_hha_pmu_init_irq(struct hisi_pmu *hha_pmu,
+				 struct platform_device *pdev)
+{
+	int irq, ret;
+
+	/* Read and init IRQ */
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	ret = devm_request_irq(&pdev->dev, irq, hisi_hha_pmu_isr,
+			      IRQF_NOBALANCING | IRQF_NO_THREAD,
+			      dev_name(&pdev->dev), hha_pmu);
+	if (ret < 0) {
+		dev_err(&pdev->dev,
+			"Fail to request IRQ:%d ret:%d\n", irq, ret);
+		return ret;
+	}
+
+	hha_pmu->irq = irq;
+
+	return 0;
+}
+
+static const struct acpi_device_id hisi_hha_pmu_acpi_match[] = {
+	{ "HISI0243", },
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, hisi_hha_pmu_acpi_match);
+
+static int hisi_hha_pmu_init_data(struct platform_device *pdev,
+				  struct hisi_pmu *hha_pmu)
+{
+	unsigned long long id;
+	acpi_status status;
+
+	status = acpi_evaluate_integer(ACPI_HANDLE(&pdev->dev),
+				       "_UID", NULL, &id);
+	if (ACPI_FAILURE(status))
+		return -EINVAL;
+
+	hha_pmu->index_id = id;
+
+	/*
+	 * Use SCCL_ID and UID to identify the HHA PMU, while
+	 * SCCL_ID is in MPIDR[aff2].
+	 */
+	if (device_property_read_u32(&pdev->dev, "hisilicon,scl-id",
+				     &hha_pmu->sccl_id)) {
+		dev_err(&pdev->dev, "Can not read hha sccl-id!\n");
+		return -EINVAL;
+	}
+	/* HHA PMUs only share the same SCCL */
+	hha_pmu->ccl_id = -1;
+
+	hha_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(hha_pmu->base)) {
+		dev_err(&pdev->dev, "ioremap failed for hha_pmu resource\n");
+		return PTR_ERR(hha_pmu->base);
+	}
+
+	return 0;
+}
+
+static struct attribute *hisi_hha_pmu_format_attr[] = {
+	HISI_PMU_FORMAT_ATTR(event, "config:0-7"),
+	NULL,
+};
+
+static const struct attribute_group hisi_hha_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_hha_pmu_format_attr,
+};
+
+static struct attribute *hisi_hha_pmu_events_attr[] = {
+	HISI_PMU_EVENT_ATTR(rx_ops_num,		0x00),
+	HISI_PMU_EVENT_ATTR(rx_outer,		0x01),
+	HISI_PMU_EVENT_ATTR(rx_sccl,		0x02),
+	HISI_PMU_EVENT_ATTR(rx_ccix,		0x03),
+	HISI_PMU_EVENT_ATTR(rx_wbi,		0x04),
+	HISI_PMU_EVENT_ATTR(rx_wbip,		0x05),
+	HISI_PMU_EVENT_ATTR(rx_wtistash,	0x11),
+	HISI_PMU_EVENT_ATTR(rd_ddr_64b,		0x1c),
+	HISI_PMU_EVENT_ATTR(wr_ddr_64b,		0x1d),
+	HISI_PMU_EVENT_ATTR(rd_ddr_128b,	0x1e),
+	HISI_PMU_EVENT_ATTR(wr_ddr_128b,	0x1f),
+	HISI_PMU_EVENT_ATTR(spill_num,		0x20),
+	HISI_PMU_EVENT_ATTR(spill_success,	0x21),
+	HISI_PMU_EVENT_ATTR(bi_num,		0x23),
+	HISI_PMU_EVENT_ATTR(mediated_num,	0x32),
+	HISI_PMU_EVENT_ATTR(tx_snp_num,		0x33),
+	HISI_PMU_EVENT_ATTR(tx_snp_outer,	0x34),
+	HISI_PMU_EVENT_ATTR(tx_snp_ccix,	0x35),
+	HISI_PMU_EVENT_ATTR(rx_snprspdata,	0x38),
+	HISI_PMU_EVENT_ATTR(rx_snprsp_outer,	0x3c),
+	HISI_PMU_EVENT_ATTR(sdir-lookup,	0x40),
+	HISI_PMU_EVENT_ATTR(edir-lookup,	0x41),
+	HISI_PMU_EVENT_ATTR(sdir-hit,		0x42),
+	HISI_PMU_EVENT_ATTR(edir-hit,		0x43),
+	HISI_PMU_EVENT_ATTR(sdir-home-migrate,	0x4c),
+	HISI_PMU_EVENT_ATTR(edir-home-migrate,  0x4d),
+	NULL,
+};
+
+static const struct attribute_group hisi_hha_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_hha_pmu_events_attr,
+};
+
+static DEVICE_ATTR(cpumask, 0444, hisi_cpumask_sysfs_show, NULL);
+
+static struct attribute *hisi_hha_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static const struct attribute_group hisi_hha_pmu_cpumask_attr_group = {
+	.attrs = hisi_hha_pmu_cpumask_attrs,
+};
+
+static const struct attribute_group *hisi_hha_pmu_attr_groups[] = {
+	&hisi_hha_pmu_format_group,
+	&hisi_hha_pmu_events_group,
+	&hisi_hha_pmu_cpumask_attr_group,
+	NULL,
+};
+
+static const struct hisi_uncore_ops hisi_uncore_hha_ops = {
+	.write_evtype		= hisi_hha_pmu_write_evtype,
+	.get_event_idx		= hisi_uncore_pmu_get_event_idx,
+	.start_counters		= hisi_hha_pmu_start_counters,
+	.stop_counters		= hisi_hha_pmu_stop_counters,
+	.enable_counter		= hisi_hha_pmu_enable_counter,
+	.disable_counter	= hisi_hha_pmu_disable_counter,
+	.enable_counter_int	= hisi_hha_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_hha_pmu_disable_counter_int,
+	.write_counter		= hisi_hha_pmu_write_counter,
+	.read_counter		= hisi_hha_pmu_read_counter,
+};
+
+static int hisi_hha_pmu_dev_probe(struct platform_device *pdev,
+				  struct hisi_pmu *hha_pmu)
+{
+	int ret;
+
+	ret = hisi_hha_pmu_init_data(pdev, hha_pmu);
+	if (ret)
+		return ret;
+
+	ret = hisi_hha_pmu_init_irq(hha_pmu, pdev);
+	if (ret)
+		return ret;
+
+	hha_pmu->num_counters = HHA_NR_COUNTERS;
+	hha_pmu->counter_bits = 48;
+	hha_pmu->ops = &hisi_uncore_hha_ops;
+	hha_pmu->dev = &pdev->dev;
+	hha_pmu->on_cpu = -1;
+	hha_pmu->check_event = 0x65;
+
+	return 0;
+}
+
+static int hisi_hha_pmu_probe(struct platform_device *pdev)
+{
+	struct hisi_pmu *hha_pmu;
+	char *name;
+	int ret;
+
+	hha_pmu = devm_kzalloc(&pdev->dev, sizeof(*hha_pmu), GFP_KERNEL);
+	if (!hha_pmu)
+		return -ENOMEM;
+
+	platform_set_drvdata(pdev, hha_pmu);
+
+	ret = hisi_hha_pmu_dev_probe(pdev, hha_pmu);
+	if (ret)
+		return ret;
+
+	ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE,
+				       &hha_pmu->node);
+	if (ret) {
+		dev_err(&pdev->dev, "Error %d registering hotplug\n", ret);
+		return ret;
+	}
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "hisi_sccl%u_hha%u",
+			      hha_pmu->sccl_id, hha_pmu->index_id);
+	hha_pmu->pmu = (struct pmu) {
+		.name		= name,
+		.module		= THIS_MODULE,
+		.task_ctx_nr	= perf_invalid_context,
+		.event_init	= hisi_uncore_pmu_event_init,
+		.pmu_enable	= hisi_uncore_pmu_enable,
+		.pmu_disable	= hisi_uncore_pmu_disable,
+		.add		= hisi_uncore_pmu_add,
+		.del		= hisi_uncore_pmu_del,
+		.start		= hisi_uncore_pmu_start,
+		.stop		= hisi_uncore_pmu_stop,
+		.read		= hisi_uncore_pmu_read,
+		.attr_groups	= hisi_hha_pmu_attr_groups,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	ret = perf_pmu_register(&hha_pmu->pmu, name, -1);
+	if (ret) {
+		dev_err(hha_pmu->dev, "HHA PMU register failed!\n");
+		cpuhp_state_remove_instance_nocalls(
+			CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE, &hha_pmu->node);
+		irq_set_affinity_hint(hha_pmu->irq, NULL);
+	}
+
+	return ret;
+}
+
+static int hisi_hha_pmu_remove(struct platform_device *pdev)
+{
+	struct hisi_pmu *hha_pmu = platform_get_drvdata(pdev);
+
+	perf_pmu_unregister(&hha_pmu->pmu);
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE,
+					    &hha_pmu->node);
+	irq_set_affinity_hint(hha_pmu->irq, NULL);
+
+	return 0;
+}
+
+static struct platform_driver hisi_hha_pmu_driver = {
+	.driver = {
+		.name = "hisi_hha_pmu",
+		.acpi_match_table = ACPI_PTR(hisi_hha_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = hisi_hha_pmu_probe,
+	.remove = hisi_hha_pmu_remove,
+};
+
+static int __init hisi_hha_pmu_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE,
+				      "AP_PERF_ARM_HISI_HHA_ONLINE",
+				      hisi_uncore_pmu_online_cpu,
+				      hisi_uncore_pmu_offline_cpu);
+	if (ret) {
+		pr_err("HHA PMU: Error setup hotplug, ret = %d;\n", ret);
+		return ret;
+	}
+
+	ret = platform_driver_register(&hisi_hha_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE);
+
+	return ret;
+}
+module_init(hisi_hha_pmu_module_init);
+
+static void __exit hisi_hha_pmu_module_exit(void)
+{
+	platform_driver_unregister(&hisi_hha_pmu_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_HHA_ONLINE);
+}
+module_exit(hisi_hha_pmu_module_exit);
+
+MODULE_DESCRIPTION("HiSilicon SoC HHA uncore PMU driver");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Shaokun Zhang <zhangshaokun@hisilicon.com>");
+MODULE_AUTHOR("Anurup M <anurup.m@huawei.com>");
diff --git a/drivers/perf/hisilicon/hisi_uncore_l3c_pmu.c b/drivers/perf/hisilicon/hisi_uncore_l3c_pmu.c
new file mode 100644
index 000000000..3e8b5eab5
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_l3c_pmu.c
@@ -0,0 +1,461 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * HiSilicon SoC L3C uncore Hardware event counters support
+ *
+ * Copyright (C) 2017 Hisilicon Limited
+ * Author: Anurup M <anurup.m@huawei.com>
+ *         Shaokun Zhang <zhangshaokun@hisilicon.com>
+ *
+ * This code is based on the uncore PMUs like arm-cci and arm-ccn.
+ */
+#include <linux/acpi.h>
+#include <linux/bug.h>
+#include <linux/cpuhotplug.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/list.h>
+#include <linux/platform_device.h>
+#include <linux/smp.h>
+
+#include "hisi_uncore_pmu.h"
+
+/* L3C register definition */
+#define L3C_PERF_CTRL		0x0408
+#define L3C_INT_MASK		0x0800
+#define L3C_INT_STATUS		0x0808
+#define L3C_INT_CLEAR		0x080c
+#define L3C_EVENT_CTRL	        0x1c00
+#define L3C_EVENT_TYPE0		0x1d00
+/*
+ * Each counter is 48-bits and [48:63] are reserved
+ * which are Read-As-Zero and Writes-Ignored.
+ */
+#define L3C_CNTR0_LOWER		0x1e00
+
+/* L3C has 8-counters */
+#define L3C_NR_COUNTERS		0x8
+
+#define L3C_PERF_CTRL_EN	0x10000
+#define L3C_EVTYPE_NONE		0xff
+
+/*
+ * Select the counter register offset using the counter index
+ */
+static u32 hisi_l3c_pmu_get_counter_offset(int cntr_idx)
+{
+	return (L3C_CNTR0_LOWER + (cntr_idx * 8));
+}
+
+static u64 hisi_l3c_pmu_read_counter(struct hisi_pmu *l3c_pmu,
+				     struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+
+	if (!hisi_uncore_pmu_counter_valid(l3c_pmu, idx)) {
+		dev_err(l3c_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return 0;
+	}
+
+	/* Read 64-bits and the upper 16 bits are RAZ */
+	return readq(l3c_pmu->base + hisi_l3c_pmu_get_counter_offset(idx));
+}
+
+static void hisi_l3c_pmu_write_counter(struct hisi_pmu *l3c_pmu,
+				       struct hw_perf_event *hwc, u64 val)
+{
+	u32 idx = hwc->idx;
+
+	if (!hisi_uncore_pmu_counter_valid(l3c_pmu, idx)) {
+		dev_err(l3c_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return;
+	}
+
+	/* Write 64-bits and the upper 16 bits are WI */
+	writeq(val, l3c_pmu->base + hisi_l3c_pmu_get_counter_offset(idx));
+}
+
+static void hisi_l3c_pmu_write_evtype(struct hisi_pmu *l3c_pmu, int idx,
+				      u32 type)
+{
+	u32 reg, reg_idx, shift, val;
+
+	/*
+	 * Select the appropriate event select register(L3C_EVENT_TYPE0/1).
+	 * There are 2 event select registers for the 8 hardware counters.
+	 * Event code is 8-bits and for the former 4 hardware counters,
+	 * L3C_EVENT_TYPE0 is chosen. For the latter 4 hardware counters,
+	 * L3C_EVENT_TYPE1 is chosen.
+	 */
+	reg = L3C_EVENT_TYPE0 + (idx / 4) * 4;
+	reg_idx = idx % 4;
+	shift = 8 * reg_idx;
+
+	/* Write event code to L3C_EVENT_TYPEx Register */
+	val = readl(l3c_pmu->base + reg);
+	val &= ~(L3C_EVTYPE_NONE << shift);
+	val |= (type << shift);
+	writel(val, l3c_pmu->base + reg);
+}
+
+static void hisi_l3c_pmu_start_counters(struct hisi_pmu *l3c_pmu)
+{
+	u32 val;
+
+	/*
+	 * Set perf_enable bit in L3C_PERF_CTRL register to start counting
+	 * for all enabled counters.
+	 */
+	val = readl(l3c_pmu->base + L3C_PERF_CTRL);
+	val |= L3C_PERF_CTRL_EN;
+	writel(val, l3c_pmu->base + L3C_PERF_CTRL);
+}
+
+static void hisi_l3c_pmu_stop_counters(struct hisi_pmu *l3c_pmu)
+{
+	u32 val;
+
+	/*
+	 * Clear perf_enable bit in L3C_PERF_CTRL register to stop counting
+	 * for all enabled counters.
+	 */
+	val = readl(l3c_pmu->base + L3C_PERF_CTRL);
+	val &= ~(L3C_PERF_CTRL_EN);
+	writel(val, l3c_pmu->base + L3C_PERF_CTRL);
+}
+
+static void hisi_l3c_pmu_enable_counter(struct hisi_pmu *l3c_pmu,
+					struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Enable counter index in L3C_EVENT_CTRL register */
+	val = readl(l3c_pmu->base + L3C_EVENT_CTRL);
+	val |= (1 << hwc->idx);
+	writel(val, l3c_pmu->base + L3C_EVENT_CTRL);
+}
+
+static void hisi_l3c_pmu_disable_counter(struct hisi_pmu *l3c_pmu,
+					 struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	/* Clear counter index in L3C_EVENT_CTRL register */
+	val = readl(l3c_pmu->base + L3C_EVENT_CTRL);
+	val &= ~(1 << hwc->idx);
+	writel(val, l3c_pmu->base + L3C_EVENT_CTRL);
+}
+
+static void hisi_l3c_pmu_enable_counter_int(struct hisi_pmu *l3c_pmu,
+					    struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(l3c_pmu->base + L3C_INT_MASK);
+	/* Write 0 to enable interrupt */
+	val &= ~(1 << hwc->idx);
+	writel(val, l3c_pmu->base + L3C_INT_MASK);
+}
+
+static void hisi_l3c_pmu_disable_counter_int(struct hisi_pmu *l3c_pmu,
+					     struct hw_perf_event *hwc)
+{
+	u32 val;
+
+	val = readl(l3c_pmu->base + L3C_INT_MASK);
+	/* Write 1 to mask interrupt */
+	val |= (1 << hwc->idx);
+	writel(val, l3c_pmu->base + L3C_INT_MASK);
+}
+
+static irqreturn_t hisi_l3c_pmu_isr(int irq, void *dev_id)
+{
+	struct hisi_pmu *l3c_pmu = dev_id;
+	struct perf_event *event;
+	unsigned long overflown;
+	int idx;
+
+	/* Read L3C_INT_STATUS register */
+	overflown = readl(l3c_pmu->base + L3C_INT_STATUS);
+	if (!overflown)
+		return IRQ_NONE;
+
+	/*
+	 * Find the counter index which overflowed if the bit was set
+	 * and handle it.
+	 */
+	for_each_set_bit(idx, &overflown, L3C_NR_COUNTERS) {
+		/* Write 1 to clear the IRQ status flag */
+		writel((1 << idx), l3c_pmu->base + L3C_INT_CLEAR);
+
+		/* Get the corresponding event struct */
+		event = l3c_pmu->pmu_events.hw_events[idx];
+		if (!event)
+			continue;
+
+		hisi_uncore_pmu_event_update(event);
+		hisi_uncore_pmu_set_event_period(event);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static int hisi_l3c_pmu_init_irq(struct hisi_pmu *l3c_pmu,
+				 struct platform_device *pdev)
+{
+	int irq, ret;
+
+	/* Read and init IRQ */
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	ret = devm_request_irq(&pdev->dev, irq, hisi_l3c_pmu_isr,
+			       IRQF_NOBALANCING | IRQF_NO_THREAD,
+			       dev_name(&pdev->dev), l3c_pmu);
+	if (ret < 0) {
+		dev_err(&pdev->dev,
+			"Fail to request IRQ:%d ret:%d\n", irq, ret);
+		return ret;
+	}
+
+	l3c_pmu->irq = irq;
+
+	return 0;
+}
+
+static const struct acpi_device_id hisi_l3c_pmu_acpi_match[] = {
+	{ "HISI0213", },
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, hisi_l3c_pmu_acpi_match);
+
+static int hisi_l3c_pmu_init_data(struct platform_device *pdev,
+				  struct hisi_pmu *l3c_pmu)
+{
+	unsigned long long id;
+	acpi_status status;
+
+	status = acpi_evaluate_integer(ACPI_HANDLE(&pdev->dev),
+				       "_UID", NULL, &id);
+	if (ACPI_FAILURE(status))
+		return -EINVAL;
+
+	l3c_pmu->index_id = id;
+
+	/*
+	 * Use the SCCL_ID and CCL_ID to identify the L3C PMU, while
+	 * SCCL_ID is in MPIDR[aff2] and CCL_ID is in MPIDR[aff1].
+	 */
+	if (device_property_read_u32(&pdev->dev, "hisilicon,scl-id",
+				     &l3c_pmu->sccl_id)) {
+		dev_err(&pdev->dev, "Can not read l3c sccl-id!\n");
+		return -EINVAL;
+	}
+
+	if (device_property_read_u32(&pdev->dev, "hisilicon,ccl-id",
+				     &l3c_pmu->ccl_id)) {
+		dev_err(&pdev->dev, "Can not read l3c ccl-id!\n");
+		return -EINVAL;
+	}
+
+	l3c_pmu->base = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(l3c_pmu->base)) {
+		dev_err(&pdev->dev, "ioremap failed for l3c_pmu resource\n");
+		return PTR_ERR(l3c_pmu->base);
+	}
+
+	return 0;
+}
+
+static struct attribute *hisi_l3c_pmu_format_attr[] = {
+	HISI_PMU_FORMAT_ATTR(event, "config:0-7"),
+	NULL,
+};
+
+static const struct attribute_group hisi_l3c_pmu_format_group = {
+	.name = "format",
+	.attrs = hisi_l3c_pmu_format_attr,
+};
+
+static struct attribute *hisi_l3c_pmu_events_attr[] = {
+	HISI_PMU_EVENT_ATTR(rd_cpipe,		0x00),
+	HISI_PMU_EVENT_ATTR(wr_cpipe,		0x01),
+	HISI_PMU_EVENT_ATTR(rd_hit_cpipe,	0x02),
+	HISI_PMU_EVENT_ATTR(wr_hit_cpipe,	0x03),
+	HISI_PMU_EVENT_ATTR(victim_num,		0x04),
+	HISI_PMU_EVENT_ATTR(rd_spipe,		0x20),
+	HISI_PMU_EVENT_ATTR(wr_spipe,		0x21),
+	HISI_PMU_EVENT_ATTR(rd_hit_spipe,	0x22),
+	HISI_PMU_EVENT_ATTR(wr_hit_spipe,	0x23),
+	HISI_PMU_EVENT_ATTR(back_invalid,	0x29),
+	HISI_PMU_EVENT_ATTR(retry_cpu,		0x40),
+	HISI_PMU_EVENT_ATTR(retry_ring,		0x41),
+	HISI_PMU_EVENT_ATTR(prefetch_drop,	0x42),
+	NULL,
+};
+
+static const struct attribute_group hisi_l3c_pmu_events_group = {
+	.name = "events",
+	.attrs = hisi_l3c_pmu_events_attr,
+};
+
+static DEVICE_ATTR(cpumask, 0444, hisi_cpumask_sysfs_show, NULL);
+
+static struct attribute *hisi_l3c_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static const struct attribute_group hisi_l3c_pmu_cpumask_attr_group = {
+	.attrs = hisi_l3c_pmu_cpumask_attrs,
+};
+
+static const struct attribute_group *hisi_l3c_pmu_attr_groups[] = {
+	&hisi_l3c_pmu_format_group,
+	&hisi_l3c_pmu_events_group,
+	&hisi_l3c_pmu_cpumask_attr_group,
+	NULL,
+};
+
+static const struct hisi_uncore_ops hisi_uncore_l3c_ops = {
+	.write_evtype		= hisi_l3c_pmu_write_evtype,
+	.get_event_idx		= hisi_uncore_pmu_get_event_idx,
+	.start_counters		= hisi_l3c_pmu_start_counters,
+	.stop_counters		= hisi_l3c_pmu_stop_counters,
+	.enable_counter		= hisi_l3c_pmu_enable_counter,
+	.disable_counter	= hisi_l3c_pmu_disable_counter,
+	.enable_counter_int	= hisi_l3c_pmu_enable_counter_int,
+	.disable_counter_int	= hisi_l3c_pmu_disable_counter_int,
+	.write_counter		= hisi_l3c_pmu_write_counter,
+	.read_counter		= hisi_l3c_pmu_read_counter,
+};
+
+static int hisi_l3c_pmu_dev_probe(struct platform_device *pdev,
+				  struct hisi_pmu *l3c_pmu)
+{
+	int ret;
+
+	ret = hisi_l3c_pmu_init_data(pdev, l3c_pmu);
+	if (ret)
+		return ret;
+
+	ret = hisi_l3c_pmu_init_irq(l3c_pmu, pdev);
+	if (ret)
+		return ret;
+
+	l3c_pmu->num_counters = L3C_NR_COUNTERS;
+	l3c_pmu->counter_bits = 48;
+	l3c_pmu->ops = &hisi_uncore_l3c_ops;
+	l3c_pmu->dev = &pdev->dev;
+	l3c_pmu->on_cpu = -1;
+	l3c_pmu->check_event = 0x59;
+
+	return 0;
+}
+
+static int hisi_l3c_pmu_probe(struct platform_device *pdev)
+{
+	struct hisi_pmu *l3c_pmu;
+	char *name;
+	int ret;
+
+	l3c_pmu = devm_kzalloc(&pdev->dev, sizeof(*l3c_pmu), GFP_KERNEL);
+	if (!l3c_pmu)
+		return -ENOMEM;
+
+	platform_set_drvdata(pdev, l3c_pmu);
+
+	ret = hisi_l3c_pmu_dev_probe(pdev, l3c_pmu);
+	if (ret)
+		return ret;
+
+	ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_HISI_L3_ONLINE,
+				       &l3c_pmu->node);
+	if (ret) {
+		dev_err(&pdev->dev, "Error %d registering hotplug\n", ret);
+		return ret;
+	}
+
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "hisi_sccl%u_l3c%u",
+			      l3c_pmu->sccl_id, l3c_pmu->index_id);
+	l3c_pmu->pmu = (struct pmu) {
+		.name		= name,
+		.module		= THIS_MODULE,
+		.task_ctx_nr	= perf_invalid_context,
+		.event_init	= hisi_uncore_pmu_event_init,
+		.pmu_enable	= hisi_uncore_pmu_enable,
+		.pmu_disable	= hisi_uncore_pmu_disable,
+		.add		= hisi_uncore_pmu_add,
+		.del		= hisi_uncore_pmu_del,
+		.start		= hisi_uncore_pmu_start,
+		.stop		= hisi_uncore_pmu_stop,
+		.read		= hisi_uncore_pmu_read,
+		.attr_groups	= hisi_l3c_pmu_attr_groups,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	ret = perf_pmu_register(&l3c_pmu->pmu, name, -1);
+	if (ret) {
+		dev_err(l3c_pmu->dev, "L3C PMU register failed!\n");
+		cpuhp_state_remove_instance_nocalls(
+			CPUHP_AP_PERF_ARM_HISI_L3_ONLINE, &l3c_pmu->node);
+		irq_set_affinity_hint(l3c_pmu->irq, NULL);
+	}
+
+	return ret;
+}
+
+static int hisi_l3c_pmu_remove(struct platform_device *pdev)
+{
+	struct hisi_pmu *l3c_pmu = platform_get_drvdata(pdev);
+
+	perf_pmu_unregister(&l3c_pmu->pmu);
+	cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_HISI_L3_ONLINE,
+					    &l3c_pmu->node);
+	irq_set_affinity_hint(l3c_pmu->irq, NULL);
+
+	return 0;
+}
+
+static struct platform_driver hisi_l3c_pmu_driver = {
+	.driver = {
+		.name = "hisi_l3c_pmu",
+		.acpi_match_table = ACPI_PTR(hisi_l3c_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = hisi_l3c_pmu_probe,
+	.remove = hisi_l3c_pmu_remove,
+};
+
+static int __init hisi_l3c_pmu_module_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_HISI_L3_ONLINE,
+				      "AP_PERF_ARM_HISI_L3_ONLINE",
+				      hisi_uncore_pmu_online_cpu,
+				      hisi_uncore_pmu_offline_cpu);
+	if (ret) {
+		pr_err("L3C PMU: Error setup hotplug, ret = %d\n", ret);
+		return ret;
+	}
+
+	ret = platform_driver_register(&hisi_l3c_pmu_driver);
+	if (ret)
+		cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_L3_ONLINE);
+
+	return ret;
+}
+module_init(hisi_l3c_pmu_module_init);
+
+static void __exit hisi_l3c_pmu_module_exit(void)
+{
+	platform_driver_unregister(&hisi_l3c_pmu_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_HISI_L3_ONLINE);
+}
+module_exit(hisi_l3c_pmu_module_exit);
+
+MODULE_DESCRIPTION("HiSilicon SoC L3C uncore PMU driver");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Anurup M <anurup.m@huawei.com>");
+MODULE_AUTHOR("Shaokun Zhang <zhangshaokun@hisilicon.com>");
diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.c b/drivers/perf/hisilicon/hisi_uncore_pmu.c
new file mode 100644
index 000000000..97aff877a
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_pmu.c
@@ -0,0 +1,471 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * HiSilicon SoC Hardware event counters support
+ *
+ * Copyright (C) 2017 Hisilicon Limited
+ * Author: Anurup M <anurup.m@huawei.com>
+ *         Shaokun Zhang <zhangshaokun@hisilicon.com>
+ *
+ * This code is based on the uncore PMUs like arm-cci and arm-ccn.
+ */
+#include <linux/bitmap.h>
+#include <linux/bitops.h>
+#include <linux/bug.h>
+#include <linux/err.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+
+#include <asm/cputype.h>
+#include <asm/local64.h>
+
+#include "hisi_uncore_pmu.h"
+
+#define HISI_GET_EVENTID(ev) (ev->hw.config_base & 0xff)
+#define HISI_MAX_PERIOD(nr) (BIT_ULL(nr) - 1)
+
+/*
+ * PMU format attributes
+ */
+ssize_t hisi_format_sysfs_show(struct device *dev,
+			       struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+
+	return sprintf(buf, "%s\n", (char *)eattr->var);
+}
+EXPORT_SYMBOL_GPL(hisi_format_sysfs_show);
+
+/*
+ * PMU event attributes
+ */
+ssize_t hisi_event_sysfs_show(struct device *dev,
+			      struct device_attribute *attr, char *page)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+
+	return sprintf(page, "config=0x%lx\n", (unsigned long)eattr->var);
+}
+EXPORT_SYMBOL_GPL(hisi_event_sysfs_show);
+
+/*
+ * sysfs cpumask attributes. For uncore PMU, we only have a single CPU to show
+ */
+ssize_t hisi_cpumask_sysfs_show(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(dev_get_drvdata(dev));
+
+	return sprintf(buf, "%d\n", hisi_pmu->on_cpu);
+}
+EXPORT_SYMBOL_GPL(hisi_cpumask_sysfs_show);
+
+static bool hisi_validate_event_group(struct perf_event *event)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	/* Include count for the event */
+	int counters = 1;
+
+	if (!is_software_event(leader)) {
+		/*
+		 * We must NOT create groups containing mixed PMUs, although
+		 * software events are acceptable
+		 */
+		if (leader->pmu != event->pmu)
+			return false;
+
+		/* Increment counter for the leader */
+		if (leader != event)
+			counters++;
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (is_software_event(sibling))
+			continue;
+		if (sibling->pmu != event->pmu)
+			return false;
+		/* Increment counter for each sibling */
+		counters++;
+	}
+
+	/* The group can not count events more than the counters in the HW */
+	return counters <= hisi_pmu->num_counters;
+}
+
+int hisi_uncore_pmu_counter_valid(struct hisi_pmu *hisi_pmu, int idx)
+{
+	return idx >= 0 && idx < hisi_pmu->num_counters;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_counter_valid);
+
+int hisi_uncore_pmu_get_event_idx(struct perf_event *event)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	unsigned long *used_mask = hisi_pmu->pmu_events.used_mask;
+	u32 num_counters = hisi_pmu->num_counters;
+	int idx;
+
+	idx = find_first_zero_bit(used_mask, num_counters);
+	if (idx == num_counters)
+		return -EAGAIN;
+
+	set_bit(idx, used_mask);
+
+	return idx;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_get_event_idx);
+
+static void hisi_uncore_pmu_clear_event_idx(struct hisi_pmu *hisi_pmu, int idx)
+{
+	if (!hisi_uncore_pmu_counter_valid(hisi_pmu, idx)) {
+		dev_err(hisi_pmu->dev, "Unsupported event index:%d!\n", idx);
+		return;
+	}
+
+	clear_bit(idx, hisi_pmu->pmu_events.used_mask);
+}
+
+int hisi_uncore_pmu_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct hisi_pmu *hisi_pmu;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * We do not support sampling as the counters are all
+	 * shared by all CPU cores in a CPU die(SCCL). Also we
+	 * do not support attach to a task(per-process mode)
+	 */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EOPNOTSUPP;
+
+	/*
+	 *  The uncore counters not specific to any CPU, so cannot
+	 *  support per-task
+	 */
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	/*
+	 * Validate if the events in group does not exceed the
+	 * available counters in hardware.
+	 */
+	if (!hisi_validate_event_group(event))
+		return -EINVAL;
+
+	hisi_pmu = to_hisi_pmu(event->pmu);
+	if (event->attr.config > hisi_pmu->check_event)
+		return -EINVAL;
+
+	if (hisi_pmu->on_cpu == -1)
+		return -EINVAL;
+	/*
+	 * 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	= event->attr.config;
+
+	/* Enforce to use the same CPU for all events in this PMU */
+	event->cpu = hisi_pmu->on_cpu;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_event_init);
+
+/*
+ * Set the counter to count the event that we're interested in,
+ * and enable interrupt and counter.
+ */
+static void hisi_uncore_pmu_enable_event(struct perf_event *event)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_pmu->ops->write_evtype(hisi_pmu, hwc->idx,
+				    HISI_GET_EVENTID(event));
+
+	hisi_pmu->ops->enable_counter_int(hisi_pmu, hwc);
+	hisi_pmu->ops->enable_counter(hisi_pmu, hwc);
+}
+
+/*
+ * Disable counter and interrupt.
+ */
+static void hisi_uncore_pmu_disable_event(struct perf_event *event)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_pmu->ops->disable_counter(hisi_pmu, hwc);
+	hisi_pmu->ops->disable_counter_int(hisi_pmu, hwc);
+}
+
+void hisi_uncore_pmu_set_event_period(struct perf_event *event)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	/*
+	 * The HiSilicon PMU counters support 32 bits or 48 bits, depending on
+	 * the PMU. We reduce it to 2^(counter_bits - 1) to account for the
+	 * extreme interrupt latency. So we could hopefully handle the overflow
+	 * interrupt before another 2^(counter_bits - 1) events occur and the
+	 * counter overtakes its previous value.
+	 */
+	u64 val = BIT_ULL(hisi_pmu->counter_bits - 1);
+
+	local64_set(&hwc->prev_count, val);
+	/* Write start value to the hardware event counter */
+	hisi_pmu->ops->write_counter(hisi_pmu, hwc, val);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_set_event_period);
+
+void hisi_uncore_pmu_event_update(struct perf_event *event)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev_raw_count, new_raw_count;
+
+	do {
+		/* Read the count from the counter register */
+		new_raw_count = hisi_pmu->ops->read_counter(hisi_pmu, hwc);
+		prev_raw_count = local64_read(&hwc->prev_count);
+	} while (local64_cmpxchg(&hwc->prev_count, prev_raw_count,
+				 new_raw_count) != prev_raw_count);
+	/*
+	 * compute the delta
+	 */
+	delta = (new_raw_count - prev_raw_count) &
+		HISI_MAX_PERIOD(hisi_pmu->counter_bits);
+	local64_add(delta, &event->count);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_event_update);
+
+void hisi_uncore_pmu_start(struct perf_event *event, int flags)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED)))
+		return;
+
+	WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+	hwc->state = 0;
+	hisi_uncore_pmu_set_event_period(event);
+
+	if (flags & PERF_EF_RELOAD) {
+		u64 prev_raw_count =  local64_read(&hwc->prev_count);
+
+		hisi_pmu->ops->write_counter(hisi_pmu, hwc, prev_raw_count);
+	}
+
+	hisi_uncore_pmu_enable_event(event);
+	perf_event_update_userpage(event);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_start);
+
+void hisi_uncore_pmu_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_uncore_pmu_disable_event(event);
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	/* Read hardware counter and update the perf counter statistics */
+	hisi_uncore_pmu_event_update(event);
+	hwc->state |= PERF_HES_UPTODATE;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_stop);
+
+int hisi_uncore_pmu_add(struct perf_event *event, int flags)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+
+	/* Get an available counter index for counting */
+	idx = hisi_pmu->ops->get_event_idx(event);
+	if (idx < 0)
+		return idx;
+
+	event->hw.idx = idx;
+	hisi_pmu->pmu_events.hw_events[idx] = event;
+
+	if (flags & PERF_EF_START)
+		hisi_uncore_pmu_start(event, PERF_EF_RELOAD);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_add);
+
+void hisi_uncore_pmu_del(struct perf_event *event, int flags)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	hisi_uncore_pmu_stop(event, PERF_EF_UPDATE);
+	hisi_uncore_pmu_clear_event_idx(hisi_pmu, hwc->idx);
+	perf_event_update_userpage(event);
+	hisi_pmu->pmu_events.hw_events[hwc->idx] = NULL;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_del);
+
+void hisi_uncore_pmu_read(struct perf_event *event)
+{
+	/* Read hardware counter and update the perf counter statistics */
+	hisi_uncore_pmu_event_update(event);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_read);
+
+void hisi_uncore_pmu_enable(struct pmu *pmu)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(pmu);
+	int enabled = bitmap_weight(hisi_pmu->pmu_events.used_mask,
+				    hisi_pmu->num_counters);
+
+	if (!enabled)
+		return;
+
+	hisi_pmu->ops->start_counters(hisi_pmu);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_enable);
+
+void hisi_uncore_pmu_disable(struct pmu *pmu)
+{
+	struct hisi_pmu *hisi_pmu = to_hisi_pmu(pmu);
+
+	hisi_pmu->ops->stop_counters(hisi_pmu);
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_disable);
+
+
+/*
+ * The Super CPU Cluster (SCCL) and CPU Cluster (CCL) IDs can be
+ * determined from the MPIDR_EL1, but the encoding varies by CPU:
+ *
+ * - For MT variants of TSV110:
+ *   SCCL is Aff2[7:3], CCL is Aff2[2:0]
+ *
+ * - For other MT parts:
+ *   SCCL is Aff3[7:0], CCL is Aff2[7:0]
+ *
+ * - For non-MT parts:
+ *   SCCL is Aff2[7:0], CCL is Aff1[7:0]
+ */
+static void hisi_read_sccl_and_ccl_id(int *scclp, int *cclp)
+{
+	u64 mpidr = read_cpuid_mpidr();
+	int aff3 = MPIDR_AFFINITY_LEVEL(mpidr, 3);
+	int aff2 = MPIDR_AFFINITY_LEVEL(mpidr, 2);
+	int aff1 = MPIDR_AFFINITY_LEVEL(mpidr, 1);
+	bool mt = mpidr & MPIDR_MT_BITMASK;
+	int sccl, ccl;
+
+	if (mt && read_cpuid_part_number() == HISI_CPU_PART_TSV110) {
+		sccl = aff2 >> 3;
+		ccl = aff2 & 0x7;
+	} else if (mt) {
+		sccl = aff3;
+		ccl = aff2;
+	} else {
+		sccl = aff2;
+		ccl = aff1;
+	}
+
+	if (scclp)
+		*scclp = sccl;
+	if (cclp)
+		*cclp = ccl;
+}
+
+/*
+ * Check whether the CPU is associated with this uncore PMU
+ */
+static bool hisi_pmu_cpu_is_associated_pmu(struct hisi_pmu *hisi_pmu)
+{
+	int sccl_id, ccl_id;
+
+	if (hisi_pmu->ccl_id == -1) {
+		/* If CCL_ID is -1, the PMU only shares the same SCCL */
+		hisi_read_sccl_and_ccl_id(&sccl_id, NULL);
+
+		return sccl_id == hisi_pmu->sccl_id;
+	}
+
+	hisi_read_sccl_and_ccl_id(&sccl_id, &ccl_id);
+
+	return sccl_id == hisi_pmu->sccl_id && ccl_id == hisi_pmu->ccl_id;
+}
+
+int hisi_uncore_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct hisi_pmu *hisi_pmu = hlist_entry_safe(node, struct hisi_pmu,
+						     node);
+
+	if (!hisi_pmu_cpu_is_associated_pmu(hisi_pmu))
+		return 0;
+
+	cpumask_set_cpu(cpu, &hisi_pmu->associated_cpus);
+
+	/* If another CPU is already managing this PMU, simply return. */
+	if (hisi_pmu->on_cpu != -1)
+		return 0;
+
+	/* Use this CPU in cpumask for event counting */
+	hisi_pmu->on_cpu = cpu;
+
+	/* Overflow interrupt also should use the same CPU */
+	WARN_ON(irq_set_affinity_hint(hisi_pmu->irq, cpumask_of(cpu)));
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_online_cpu);
+
+int hisi_uncore_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct hisi_pmu *hisi_pmu = hlist_entry_safe(node, struct hisi_pmu,
+						     node);
+	cpumask_t pmu_online_cpus;
+	unsigned int target;
+
+	if (!cpumask_test_and_clear_cpu(cpu, &hisi_pmu->associated_cpus))
+		return 0;
+
+	/* Nothing to do if this CPU doesn't own the PMU */
+	if (hisi_pmu->on_cpu != cpu)
+		return 0;
+
+	/* Give up ownership of the PMU */
+	hisi_pmu->on_cpu = -1;
+
+	/* Choose a new CPU to migrate ownership of the PMU to */
+	cpumask_and(&pmu_online_cpus, &hisi_pmu->associated_cpus,
+		    cpu_online_mask);
+	target = cpumask_any_but(&pmu_online_cpus, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	perf_pmu_migrate_context(&hisi_pmu->pmu, cpu, target);
+	/* Use this CPU for event counting */
+	hisi_pmu->on_cpu = target;
+	WARN_ON(irq_set_affinity_hint(hisi_pmu->irq, cpumask_of(target)));
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(hisi_uncore_pmu_offline_cpu);
+
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.h b/drivers/perf/hisilicon/hisi_uncore_pmu.h
new file mode 100644
index 000000000..b59ec2216
--- /dev/null
+++ b/drivers/perf/hisilicon/hisi_uncore_pmu.h
@@ -0,0 +1,100 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * HiSilicon SoC Hardware event counters support
+ *
+ * Copyright (C) 2017 Hisilicon Limited
+ * Author: Anurup M <anurup.m@huawei.com>
+ *         Shaokun Zhang <zhangshaokun@hisilicon.com>
+ *
+ * This code is based on the uncore PMUs like arm-cci and arm-ccn.
+ */
+#ifndef __HISI_UNCORE_PMU_H__
+#define __HISI_UNCORE_PMU_H__
+
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/perf_event.h>
+#include <linux/types.h>
+
+#undef pr_fmt
+#define pr_fmt(fmt)     "hisi_pmu: " fmt
+
+#define HISI_MAX_COUNTERS 0x10
+#define to_hisi_pmu(p)	(container_of(p, struct hisi_pmu, pmu))
+
+#define HISI_PMU_ATTR(_name, _func, _config)				\
+	(&((struct dev_ext_attribute[]) {				\
+		{ __ATTR(_name, 0444, _func, NULL), (void *)_config }   \
+	})[0].attr.attr)
+
+#define HISI_PMU_FORMAT_ATTR(_name, _config)		\
+	HISI_PMU_ATTR(_name, hisi_format_sysfs_show, (void *)_config)
+#define HISI_PMU_EVENT_ATTR(_name, _config)		\
+	HISI_PMU_ATTR(_name, hisi_event_sysfs_show, (unsigned long)_config)
+
+struct hisi_pmu;
+
+struct hisi_uncore_ops {
+	void (*write_evtype)(struct hisi_pmu *, int, u32);
+	int (*get_event_idx)(struct perf_event *);
+	u64 (*read_counter)(struct hisi_pmu *, struct hw_perf_event *);
+	void (*write_counter)(struct hisi_pmu *, struct hw_perf_event *, u64);
+	void (*enable_counter)(struct hisi_pmu *, struct hw_perf_event *);
+	void (*disable_counter)(struct hisi_pmu *, struct hw_perf_event *);
+	void (*enable_counter_int)(struct hisi_pmu *, struct hw_perf_event *);
+	void (*disable_counter_int)(struct hisi_pmu *, struct hw_perf_event *);
+	void (*start_counters)(struct hisi_pmu *);
+	void (*stop_counters)(struct hisi_pmu *);
+};
+
+struct hisi_pmu_hwevents {
+	struct perf_event *hw_events[HISI_MAX_COUNTERS];
+	DECLARE_BITMAP(used_mask, HISI_MAX_COUNTERS);
+};
+
+/* Generic pmu struct for different pmu types */
+struct hisi_pmu {
+	struct pmu pmu;
+	const struct hisi_uncore_ops *ops;
+	struct hisi_pmu_hwevents pmu_events;
+	/* associated_cpus: All CPUs associated with the PMU */
+	cpumask_t associated_cpus;
+	/* CPU used for counting */
+	int on_cpu;
+	int irq;
+	struct device *dev;
+	struct hlist_node node;
+	int sccl_id;
+	int ccl_id;
+	void __iomem *base;
+	/* the ID of the PMU modules */
+	u32 index_id;
+	int num_counters;
+	int counter_bits;
+	/* check event code range */
+	int check_event;
+};
+
+int hisi_uncore_pmu_counter_valid(struct hisi_pmu *hisi_pmu, int idx);
+int hisi_uncore_pmu_get_event_idx(struct perf_event *event);
+void hisi_uncore_pmu_read(struct perf_event *event);
+int hisi_uncore_pmu_add(struct perf_event *event, int flags);
+void hisi_uncore_pmu_del(struct perf_event *event, int flags);
+void hisi_uncore_pmu_start(struct perf_event *event, int flags);
+void hisi_uncore_pmu_stop(struct perf_event *event, int flags);
+void hisi_uncore_pmu_set_event_period(struct perf_event *event);
+void hisi_uncore_pmu_event_update(struct perf_event *event);
+int hisi_uncore_pmu_event_init(struct perf_event *event);
+void hisi_uncore_pmu_enable(struct pmu *pmu);
+void hisi_uncore_pmu_disable(struct pmu *pmu);
+ssize_t hisi_event_sysfs_show(struct device *dev,
+			      struct device_attribute *attr, char *buf);
+ssize_t hisi_format_sysfs_show(struct device *dev,
+			       struct device_attribute *attr, char *buf);
+ssize_t hisi_cpumask_sysfs_show(struct device *dev,
+				struct device_attribute *attr, char *buf);
+int hisi_uncore_pmu_online_cpu(unsigned int cpu, struct hlist_node *node);
+int hisi_uncore_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node);
+#endif /* __HISI_UNCORE_PMU_H__ */
diff --git a/drivers/perf/qcom_l2_pmu.c b/drivers/perf/qcom_l2_pmu.c
new file mode 100644
index 000000000..d58810f53
--- /dev/null
+++ b/drivers/perf/qcom_l2_pmu.c
@@ -0,0 +1,1008 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2015-2017 The Linux Foundation. All rights reserved.
+ */
+#include <linux/acpi.h>
+#include <linux/bitops.h>
+#include <linux/bug.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/device.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/percpu.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/smp.h>
+#include <linux/spinlock.h>
+#include <linux/sysfs.h>
+#include <linux/types.h>
+
+#include <asm/barrier.h>
+#include <asm/local64.h>
+#include <asm/sysreg.h>
+#include <soc/qcom/kryo-l2-accessors.h>
+
+#define MAX_L2_CTRS             9
+
+#define L2PMCR_NUM_EV_SHIFT     11
+#define L2PMCR_NUM_EV_MASK      0x1F
+
+#define L2PMCR                  0x400
+#define L2PMCNTENCLR            0x403
+#define L2PMCNTENSET            0x404
+#define L2PMINTENCLR            0x405
+#define L2PMINTENSET            0x406
+#define L2PMOVSCLR              0x407
+#define L2PMOVSSET              0x408
+#define L2PMCCNTCR              0x409
+#define L2PMCCNTR               0x40A
+#define L2PMCCNTSR              0x40C
+#define L2PMRESR                0x410
+#define IA_L2PMXEVCNTCR_BASE    0x420
+#define IA_L2PMXEVCNTR_BASE     0x421
+#define IA_L2PMXEVFILTER_BASE   0x423
+#define IA_L2PMXEVTYPER_BASE    0x424
+
+#define IA_L2_REG_OFFSET        0x10
+
+#define L2PMXEVFILTER_SUFILTER_ALL      0x000E0000
+#define L2PMXEVFILTER_ORGFILTER_IDINDEP 0x00000004
+#define L2PMXEVFILTER_ORGFILTER_ALL     0x00000003
+
+#define L2EVTYPER_REG_SHIFT     3
+
+#define L2PMRESR_GROUP_BITS     8
+#define L2PMRESR_GROUP_MASK     GENMASK(7, 0)
+
+#define L2CYCLE_CTR_BIT         31
+#define L2CYCLE_CTR_RAW_CODE    0xFE
+
+#define L2PMCR_RESET_ALL        0x6
+#define L2PMCR_COUNTERS_ENABLE  0x1
+#define L2PMCR_COUNTERS_DISABLE 0x0
+
+#define L2PMRESR_EN             BIT_ULL(63)
+
+#define L2_EVT_MASK             0x00000FFF
+#define L2_EVT_CODE_MASK        0x00000FF0
+#define L2_EVT_GRP_MASK         0x0000000F
+#define L2_EVT_CODE_SHIFT       4
+#define L2_EVT_GRP_SHIFT        0
+
+#define L2_EVT_CODE(event)   (((event) & L2_EVT_CODE_MASK) >> L2_EVT_CODE_SHIFT)
+#define L2_EVT_GROUP(event)  (((event) & L2_EVT_GRP_MASK) >> L2_EVT_GRP_SHIFT)
+
+#define L2_EVT_GROUP_MAX        7
+
+#define L2_COUNTER_RELOAD       BIT_ULL(31)
+#define L2_CYCLE_COUNTER_RELOAD BIT_ULL(63)
+
+
+#define reg_idx(reg, i)         (((i) * IA_L2_REG_OFFSET) + reg##_BASE)
+
+/*
+ * Events
+ */
+#define L2_EVENT_CYCLES                    0xfe
+#define L2_EVENT_DCACHE_OPS                0x400
+#define L2_EVENT_ICACHE_OPS                0x401
+#define L2_EVENT_TLBI                      0x402
+#define L2_EVENT_BARRIERS                  0x403
+#define L2_EVENT_TOTAL_READS               0x405
+#define L2_EVENT_TOTAL_WRITES              0x406
+#define L2_EVENT_TOTAL_REQUESTS            0x407
+#define L2_EVENT_LDREX                     0x420
+#define L2_EVENT_STREX                     0x421
+#define L2_EVENT_CLREX                     0x422
+
+
+
+struct cluster_pmu;
+
+/*
+ * Aggregate PMU. Implements the core pmu functions and manages
+ * the hardware PMUs.
+ */
+struct l2cache_pmu {
+	struct hlist_node node;
+	u32 num_pmus;
+	struct pmu pmu;
+	int num_counters;
+	cpumask_t cpumask;
+	struct platform_device *pdev;
+	struct cluster_pmu * __percpu *pmu_cluster;
+	struct list_head clusters;
+};
+
+/*
+ * The cache is made up of one or more clusters, each cluster has its own PMU.
+ * Each cluster is associated with one or more CPUs.
+ * This structure represents one of the hardware PMUs.
+ *
+ * Events can be envisioned as a 2-dimensional array. Each column represents
+ * a group of events. There are 8 groups. Only one entry from each
+ * group can be in use at a time.
+ *
+ * Events are specified as 0xCCG, where CC is 2 hex digits specifying
+ * the code (array row) and G specifies the group (column).
+ *
+ * In addition there is a cycle counter event specified by L2CYCLE_CTR_RAW_CODE
+ * which is outside the above scheme.
+ */
+struct cluster_pmu {
+	struct list_head next;
+	struct perf_event *events[MAX_L2_CTRS];
+	struct l2cache_pmu *l2cache_pmu;
+	DECLARE_BITMAP(used_counters, MAX_L2_CTRS);
+	DECLARE_BITMAP(used_groups, L2_EVT_GROUP_MAX + 1);
+	int irq;
+	int cluster_id;
+	/* The CPU that is used for collecting events on this cluster */
+	int on_cpu;
+	/* All the CPUs associated with this cluster */
+	cpumask_t cluster_cpus;
+	spinlock_t pmu_lock;
+};
+
+#define to_l2cache_pmu(p) (container_of(p, struct l2cache_pmu, pmu))
+
+static u32 l2_cycle_ctr_idx;
+static u32 l2_counter_present_mask;
+
+static inline u32 idx_to_reg_bit(u32 idx)
+{
+	if (idx == l2_cycle_ctr_idx)
+		return BIT(L2CYCLE_CTR_BIT);
+
+	return BIT(idx);
+}
+
+static inline struct cluster_pmu *get_cluster_pmu(
+	struct l2cache_pmu *l2cache_pmu, int cpu)
+{
+	return *per_cpu_ptr(l2cache_pmu->pmu_cluster, cpu);
+}
+
+static void cluster_pmu_reset(void)
+{
+	/* Reset all counters */
+	kryo_l2_set_indirect_reg(L2PMCR, L2PMCR_RESET_ALL);
+	kryo_l2_set_indirect_reg(L2PMCNTENCLR, l2_counter_present_mask);
+	kryo_l2_set_indirect_reg(L2PMINTENCLR, l2_counter_present_mask);
+	kryo_l2_set_indirect_reg(L2PMOVSCLR, l2_counter_present_mask);
+}
+
+static inline void cluster_pmu_enable(void)
+{
+	kryo_l2_set_indirect_reg(L2PMCR, L2PMCR_COUNTERS_ENABLE);
+}
+
+static inline void cluster_pmu_disable(void)
+{
+	kryo_l2_set_indirect_reg(L2PMCR, L2PMCR_COUNTERS_DISABLE);
+}
+
+static inline void cluster_pmu_counter_set_value(u32 idx, u64 value)
+{
+	if (idx == l2_cycle_ctr_idx)
+		kryo_l2_set_indirect_reg(L2PMCCNTR, value);
+	else
+		kryo_l2_set_indirect_reg(reg_idx(IA_L2PMXEVCNTR, idx), value);
+}
+
+static inline u64 cluster_pmu_counter_get_value(u32 idx)
+{
+	u64 value;
+
+	if (idx == l2_cycle_ctr_idx)
+		value = kryo_l2_get_indirect_reg(L2PMCCNTR);
+	else
+		value = kryo_l2_get_indirect_reg(reg_idx(IA_L2PMXEVCNTR, idx));
+
+	return value;
+}
+
+static inline void cluster_pmu_counter_enable(u32 idx)
+{
+	kryo_l2_set_indirect_reg(L2PMCNTENSET, idx_to_reg_bit(idx));
+}
+
+static inline void cluster_pmu_counter_disable(u32 idx)
+{
+	kryo_l2_set_indirect_reg(L2PMCNTENCLR, idx_to_reg_bit(idx));
+}
+
+static inline void cluster_pmu_counter_enable_interrupt(u32 idx)
+{
+	kryo_l2_set_indirect_reg(L2PMINTENSET, idx_to_reg_bit(idx));
+}
+
+static inline void cluster_pmu_counter_disable_interrupt(u32 idx)
+{
+	kryo_l2_set_indirect_reg(L2PMINTENCLR, idx_to_reg_bit(idx));
+}
+
+static inline void cluster_pmu_set_evccntcr(u32 val)
+{
+	kryo_l2_set_indirect_reg(L2PMCCNTCR, val);
+}
+
+static inline void cluster_pmu_set_evcntcr(u32 ctr, u32 val)
+{
+	kryo_l2_set_indirect_reg(reg_idx(IA_L2PMXEVCNTCR, ctr), val);
+}
+
+static inline void cluster_pmu_set_evtyper(u32 ctr, u32 val)
+{
+	kryo_l2_set_indirect_reg(reg_idx(IA_L2PMXEVTYPER, ctr), val);
+}
+
+static void cluster_pmu_set_resr(struct cluster_pmu *cluster,
+			       u32 event_group, u32 event_cc)
+{
+	u64 field;
+	u64 resr_val;
+	u32 shift;
+	unsigned long flags;
+
+	shift = L2PMRESR_GROUP_BITS * event_group;
+	field = ((u64)(event_cc & L2PMRESR_GROUP_MASK) << shift);
+
+	spin_lock_irqsave(&cluster->pmu_lock, flags);
+
+	resr_val = kryo_l2_get_indirect_reg(L2PMRESR);
+	resr_val &= ~(L2PMRESR_GROUP_MASK << shift);
+	resr_val |= field;
+	resr_val |= L2PMRESR_EN;
+	kryo_l2_set_indirect_reg(L2PMRESR, resr_val);
+
+	spin_unlock_irqrestore(&cluster->pmu_lock, flags);
+}
+
+/*
+ * Hardware allows filtering of events based on the originating
+ * CPU. Turn this off by setting filter bits to allow events from
+ * all CPUS, subunits and ID independent events in this cluster.
+ */
+static inline void cluster_pmu_set_evfilter_sys_mode(u32 ctr)
+{
+	u32 val =  L2PMXEVFILTER_SUFILTER_ALL |
+		   L2PMXEVFILTER_ORGFILTER_IDINDEP |
+		   L2PMXEVFILTER_ORGFILTER_ALL;
+
+	kryo_l2_set_indirect_reg(reg_idx(IA_L2PMXEVFILTER, ctr), val);
+}
+
+static inline u32 cluster_pmu_getreset_ovsr(void)
+{
+	u32 result = kryo_l2_get_indirect_reg(L2PMOVSSET);
+
+	kryo_l2_set_indirect_reg(L2PMOVSCLR, result);
+	return result;
+}
+
+static inline bool cluster_pmu_has_overflowed(u32 ovsr)
+{
+	return !!(ovsr & l2_counter_present_mask);
+}
+
+static inline bool cluster_pmu_counter_has_overflowed(u32 ovsr, u32 idx)
+{
+	return !!(ovsr & idx_to_reg_bit(idx));
+}
+
+static void l2_cache_event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 delta, prev, now;
+	u32 idx = hwc->idx;
+
+	do {
+		prev = local64_read(&hwc->prev_count);
+		now = cluster_pmu_counter_get_value(idx);
+	} while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev);
+
+	/*
+	 * The cycle counter is 64-bit, but all other counters are
+	 * 32-bit, and we must handle 32-bit overflow explicitly.
+	 */
+	delta = now - prev;
+	if (idx != l2_cycle_ctr_idx)
+		delta &= 0xffffffff;
+
+	local64_add(delta, &event->count);
+}
+
+static void l2_cache_cluster_set_period(struct cluster_pmu *cluster,
+				       struct hw_perf_event *hwc)
+{
+	u32 idx = hwc->idx;
+	u64 new;
+
+	/*
+	 * We limit the max period to half the max counter value so
+	 * that even in the case of extreme interrupt latency the
+	 * counter will (hopefully) not wrap past its initial value.
+	 */
+	if (idx == l2_cycle_ctr_idx)
+		new = L2_CYCLE_COUNTER_RELOAD;
+	else
+		new = L2_COUNTER_RELOAD;
+
+	local64_set(&hwc->prev_count, new);
+	cluster_pmu_counter_set_value(idx, new);
+}
+
+static int l2_cache_get_event_idx(struct cluster_pmu *cluster,
+				   struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+	int num_ctrs = cluster->l2cache_pmu->num_counters - 1;
+	unsigned int group;
+
+	if (hwc->config_base == L2CYCLE_CTR_RAW_CODE) {
+		if (test_and_set_bit(l2_cycle_ctr_idx, cluster->used_counters))
+			return -EAGAIN;
+
+		return l2_cycle_ctr_idx;
+	}
+
+	idx = find_first_zero_bit(cluster->used_counters, num_ctrs);
+	if (idx == num_ctrs)
+		/* The counters are all in use. */
+		return -EAGAIN;
+
+	/*
+	 * Check for column exclusion: event column already in use by another
+	 * event. This is for events which are not in the same group.
+	 * Conflicting events in the same group are detected in event_init.
+	 */
+	group = L2_EVT_GROUP(hwc->config_base);
+	if (test_bit(group, cluster->used_groups))
+		return -EAGAIN;
+
+	set_bit(idx, cluster->used_counters);
+	set_bit(group, cluster->used_groups);
+
+	return idx;
+}
+
+static void l2_cache_clear_event_idx(struct cluster_pmu *cluster,
+				      struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	clear_bit(idx, cluster->used_counters);
+	if (hwc->config_base != L2CYCLE_CTR_RAW_CODE)
+		clear_bit(L2_EVT_GROUP(hwc->config_base), cluster->used_groups);
+}
+
+static irqreturn_t l2_cache_handle_irq(int irq_num, void *data)
+{
+	struct cluster_pmu *cluster = data;
+	int num_counters = cluster->l2cache_pmu->num_counters;
+	u32 ovsr;
+	int idx;
+
+	ovsr = cluster_pmu_getreset_ovsr();
+	if (!cluster_pmu_has_overflowed(ovsr))
+		return IRQ_NONE;
+
+	for_each_set_bit(idx, cluster->used_counters, num_counters) {
+		struct perf_event *event = cluster->events[idx];
+		struct hw_perf_event *hwc;
+
+		if (WARN_ON_ONCE(!event))
+			continue;
+
+		if (!cluster_pmu_counter_has_overflowed(ovsr, idx))
+			continue;
+
+		l2_cache_event_update(event);
+		hwc = &event->hw;
+
+		l2_cache_cluster_set_period(cluster, hwc);
+	}
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Implementation of abstract pmu functionality required by
+ * the core perf events code.
+ */
+
+static void l2_cache_pmu_enable(struct pmu *pmu)
+{
+	/*
+	 * Although there is only one PMU (per socket) controlling multiple
+	 * physical PMUs (per cluster), because we do not support per-task mode
+	 * each event is associated with a CPU. Each event has pmu_enable
+	 * called on its CPU, so here it is only necessary to enable the
+	 * counters for the current CPU.
+	 */
+
+	cluster_pmu_enable();
+}
+
+static void l2_cache_pmu_disable(struct pmu *pmu)
+{
+	cluster_pmu_disable();
+}
+
+static int l2_cache_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct cluster_pmu *cluster;
+	struct perf_event *sibling;
+	struct l2cache_pmu *l2cache_pmu;
+
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	l2cache_pmu = to_l2cache_pmu(event->pmu);
+
+	if (hwc->sample_period) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+				    "Sampling not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (event->cpu < 0) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+				    "Per-task mode not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (((L2_EVT_GROUP(event->attr.config) > L2_EVT_GROUP_MAX) ||
+	     ((event->attr.config & ~L2_EVT_MASK) != 0)) &&
+	    (event->attr.config != L2CYCLE_CTR_RAW_CODE)) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+				    "Invalid config %llx\n",
+				    event->attr.config);
+		return -EINVAL;
+	}
+
+	/* Don't allow groups with mixed PMUs, except for s/w events */
+	if (event->group_leader->pmu != event->pmu &&
+	    !is_software_event(event->group_leader)) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+			 "Can't create mixed PMU group\n");
+		return -EINVAL;
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (sibling->pmu != event->pmu &&
+		    !is_software_event(sibling)) {
+			dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+				 "Can't create mixed PMU group\n");
+			return -EINVAL;
+		}
+	}
+
+	cluster = get_cluster_pmu(l2cache_pmu, event->cpu);
+	if (!cluster) {
+		/* CPU has not been initialised */
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+			"CPU%d not associated with L2 cluster\n", event->cpu);
+		return -EINVAL;
+	}
+
+	/* Ensure all events in a group are on the same cpu */
+	if ((event->group_leader != event) &&
+	    (cluster->on_cpu != event->group_leader->cpu)) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+			 "Can't create group on CPUs %d and %d",
+			 event->cpu, event->group_leader->cpu);
+		return -EINVAL;
+	}
+
+	if ((event != event->group_leader) &&
+	    !is_software_event(event->group_leader) &&
+	    (L2_EVT_GROUP(event->group_leader->attr.config) ==
+	     L2_EVT_GROUP(event->attr.config))) {
+		dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+			 "Column exclusion: conflicting events %llx %llx\n",
+		       event->group_leader->attr.config,
+		       event->attr.config);
+		return -EINVAL;
+	}
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if ((sibling != event) &&
+		    !is_software_event(sibling) &&
+		    (L2_EVT_GROUP(sibling->attr.config) ==
+		     L2_EVT_GROUP(event->attr.config))) {
+			dev_dbg_ratelimited(&l2cache_pmu->pdev->dev,
+			     "Column exclusion: conflicting events %llx %llx\n",
+					    sibling->attr.config,
+					    event->attr.config);
+			return -EINVAL;
+		}
+	}
+
+	hwc->idx = -1;
+	hwc->config_base = event->attr.config;
+
+	/*
+	 * Ensure all events are on the same cpu so all events are in the
+	 * same cpu context, to avoid races on pmu_enable etc.
+	 */
+	event->cpu = cluster->on_cpu;
+
+	return 0;
+}
+
+static void l2_cache_event_start(struct perf_event *event, int flags)
+{
+	struct cluster_pmu *cluster;
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+	u32 config;
+	u32 event_cc, event_group;
+
+	hwc->state = 0;
+
+	cluster = get_cluster_pmu(to_l2cache_pmu(event->pmu), event->cpu);
+
+	l2_cache_cluster_set_period(cluster, hwc);
+
+	if (hwc->config_base == L2CYCLE_CTR_RAW_CODE) {
+		cluster_pmu_set_evccntcr(0);
+	} else {
+		config = hwc->config_base;
+		event_cc    = L2_EVT_CODE(config);
+		event_group = L2_EVT_GROUP(config);
+
+		cluster_pmu_set_evcntcr(idx, 0);
+		cluster_pmu_set_evtyper(idx, event_group);
+		cluster_pmu_set_resr(cluster, event_group, event_cc);
+		cluster_pmu_set_evfilter_sys_mode(idx);
+	}
+
+	cluster_pmu_counter_enable_interrupt(idx);
+	cluster_pmu_counter_enable(idx);
+}
+
+static void l2_cache_event_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int idx = hwc->idx;
+
+	if (hwc->state & PERF_HES_STOPPED)
+		return;
+
+	cluster_pmu_counter_disable_interrupt(idx);
+	cluster_pmu_counter_disable(idx);
+
+	if (flags & PERF_EF_UPDATE)
+		l2_cache_event_update(event);
+	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+}
+
+static int l2_cache_event_add(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	int idx;
+	int err = 0;
+	struct cluster_pmu *cluster;
+
+	cluster = get_cluster_pmu(to_l2cache_pmu(event->pmu), event->cpu);
+
+	idx = l2_cache_get_event_idx(cluster, event);
+	if (idx < 0)
+		return idx;
+
+	hwc->idx = idx;
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	cluster->events[idx] = event;
+	local64_set(&hwc->prev_count, 0);
+
+	if (flags & PERF_EF_START)
+		l2_cache_event_start(event, flags);
+
+	/* Propagate changes to the userspace mapping. */
+	perf_event_update_userpage(event);
+
+	return err;
+}
+
+static void l2_cache_event_del(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct cluster_pmu *cluster;
+	int idx = hwc->idx;
+
+	cluster = get_cluster_pmu(to_l2cache_pmu(event->pmu), event->cpu);
+
+	l2_cache_event_stop(event, flags | PERF_EF_UPDATE);
+	cluster->events[idx] = NULL;
+	l2_cache_clear_event_idx(cluster, event);
+
+	perf_event_update_userpage(event);
+}
+
+static void l2_cache_event_read(struct perf_event *event)
+{
+	l2_cache_event_update(event);
+}
+
+static ssize_t l2_cache_pmu_cpumask_show(struct device *dev,
+					 struct device_attribute *attr,
+					 char *buf)
+{
+	struct l2cache_pmu *l2cache_pmu = to_l2cache_pmu(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, &l2cache_pmu->cpumask);
+}
+
+static struct device_attribute l2_cache_pmu_cpumask_attr =
+		__ATTR(cpumask, S_IRUGO, l2_cache_pmu_cpumask_show, NULL);
+
+static struct attribute *l2_cache_pmu_cpumask_attrs[] = {
+	&l2_cache_pmu_cpumask_attr.attr,
+	NULL,
+};
+
+static struct attribute_group l2_cache_pmu_cpumask_group = {
+	.attrs = l2_cache_pmu_cpumask_attrs,
+};
+
+/* CCG format for perf RAW codes. */
+PMU_FORMAT_ATTR(l2_code,   "config:4-11");
+PMU_FORMAT_ATTR(l2_group,  "config:0-3");
+PMU_FORMAT_ATTR(event,     "config:0-11");
+
+static struct attribute *l2_cache_pmu_formats[] = {
+	&format_attr_l2_code.attr,
+	&format_attr_l2_group.attr,
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute_group l2_cache_pmu_format_group = {
+	.name = "format",
+	.attrs = l2_cache_pmu_formats,
+};
+
+static ssize_t l2cache_pmu_event_show(struct device *dev,
+				      struct device_attribute *attr, char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+	return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define L2CACHE_EVENT_ATTR(_name, _id)					     \
+	(&((struct perf_pmu_events_attr[]) {				     \
+		{ .attr = __ATTR(_name, 0444, l2cache_pmu_event_show, NULL), \
+		  .id = _id, }						     \
+	})[0].attr.attr)
+
+static struct attribute *l2_cache_pmu_events[] = {
+	L2CACHE_EVENT_ATTR(cycles, L2_EVENT_CYCLES),
+	L2CACHE_EVENT_ATTR(dcache-ops, L2_EVENT_DCACHE_OPS),
+	L2CACHE_EVENT_ATTR(icache-ops, L2_EVENT_ICACHE_OPS),
+	L2CACHE_EVENT_ATTR(tlbi, L2_EVENT_TLBI),
+	L2CACHE_EVENT_ATTR(barriers, L2_EVENT_BARRIERS),
+	L2CACHE_EVENT_ATTR(total-reads, L2_EVENT_TOTAL_READS),
+	L2CACHE_EVENT_ATTR(total-writes, L2_EVENT_TOTAL_WRITES),
+	L2CACHE_EVENT_ATTR(total-requests, L2_EVENT_TOTAL_REQUESTS),
+	L2CACHE_EVENT_ATTR(ldrex, L2_EVENT_LDREX),
+	L2CACHE_EVENT_ATTR(strex, L2_EVENT_STREX),
+	L2CACHE_EVENT_ATTR(clrex, L2_EVENT_CLREX),
+	NULL
+};
+
+static struct attribute_group l2_cache_pmu_events_group = {
+	.name = "events",
+	.attrs = l2_cache_pmu_events,
+};
+
+static const struct attribute_group *l2_cache_pmu_attr_grps[] = {
+	&l2_cache_pmu_format_group,
+	&l2_cache_pmu_cpumask_group,
+	&l2_cache_pmu_events_group,
+	NULL,
+};
+
+/*
+ * Generic device handlers
+ */
+
+static const struct acpi_device_id l2_cache_pmu_acpi_match[] = {
+	{ "QCOM8130", },
+	{ }
+};
+
+static int get_num_counters(void)
+{
+	int val;
+
+	val = kryo_l2_get_indirect_reg(L2PMCR);
+
+	/*
+	 * Read number of counters from L2PMCR and add 1
+	 * for the cycle counter.
+	 */
+	return ((val >> L2PMCR_NUM_EV_SHIFT) & L2PMCR_NUM_EV_MASK) + 1;
+}
+
+static struct cluster_pmu *l2_cache_associate_cpu_with_cluster(
+	struct l2cache_pmu *l2cache_pmu, int cpu)
+{
+	u64 mpidr;
+	int cpu_cluster_id;
+	struct cluster_pmu *cluster;
+
+	/*
+	 * This assumes that the cluster_id is in MPIDR[aff1] for
+	 * single-threaded cores, and MPIDR[aff2] for multi-threaded
+	 * cores. This logic will have to be updated if this changes.
+	 */
+	mpidr = read_cpuid_mpidr();
+	if (mpidr & MPIDR_MT_BITMASK)
+		cpu_cluster_id = MPIDR_AFFINITY_LEVEL(mpidr, 2);
+	else
+		cpu_cluster_id = MPIDR_AFFINITY_LEVEL(mpidr, 1);
+
+	list_for_each_entry(cluster, &l2cache_pmu->clusters, next) {
+		if (cluster->cluster_id != cpu_cluster_id)
+			continue;
+
+		dev_info(&l2cache_pmu->pdev->dev,
+			 "CPU%d associated with cluster %d\n", cpu,
+			 cluster->cluster_id);
+		cpumask_set_cpu(cpu, &cluster->cluster_cpus);
+		*per_cpu_ptr(l2cache_pmu->pmu_cluster, cpu) = cluster;
+		return cluster;
+	}
+
+	return NULL;
+}
+
+static int l2cache_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct cluster_pmu *cluster;
+	struct l2cache_pmu *l2cache_pmu;
+
+	l2cache_pmu = hlist_entry_safe(node, struct l2cache_pmu, node);
+	cluster = get_cluster_pmu(l2cache_pmu, cpu);
+	if (!cluster) {
+		/* First time this CPU has come online */
+		cluster = l2_cache_associate_cpu_with_cluster(l2cache_pmu, cpu);
+		if (!cluster) {
+			/* Only if broken firmware doesn't list every cluster */
+			WARN_ONCE(1, "No L2 cache cluster for CPU%d\n", cpu);
+			return 0;
+		}
+	}
+
+	/* If another CPU is managing this cluster, we're done */
+	if (cluster->on_cpu != -1)
+		return 0;
+
+	/*
+	 * All CPUs on this cluster were down, use this one.
+	 * Reset to put it into sane state.
+	 */
+	cluster->on_cpu = cpu;
+	cpumask_set_cpu(cpu, &l2cache_pmu->cpumask);
+	cluster_pmu_reset();
+
+	WARN_ON(irq_set_affinity(cluster->irq, cpumask_of(cpu)));
+	enable_irq(cluster->irq);
+
+	return 0;
+}
+
+static int l2cache_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct cluster_pmu *cluster;
+	struct l2cache_pmu *l2cache_pmu;
+	cpumask_t cluster_online_cpus;
+	unsigned int target;
+
+	l2cache_pmu = hlist_entry_safe(node, struct l2cache_pmu, node);
+	cluster = get_cluster_pmu(l2cache_pmu, cpu);
+	if (!cluster)
+		return 0;
+
+	/* If this CPU is not managing the cluster, we're done */
+	if (cluster->on_cpu != cpu)
+		return 0;
+
+	/* Give up ownership of cluster */
+	cpumask_clear_cpu(cpu, &l2cache_pmu->cpumask);
+	cluster->on_cpu = -1;
+
+	/* Any other CPU for this cluster which is still online */
+	cpumask_and(&cluster_online_cpus, &cluster->cluster_cpus,
+		    cpu_online_mask);
+	target = cpumask_any_but(&cluster_online_cpus, cpu);
+	if (target >= nr_cpu_ids) {
+		disable_irq(cluster->irq);
+		return 0;
+	}
+
+	perf_pmu_migrate_context(&l2cache_pmu->pmu, cpu, target);
+	cluster->on_cpu = target;
+	cpumask_set_cpu(target, &l2cache_pmu->cpumask);
+	WARN_ON(irq_set_affinity(cluster->irq, cpumask_of(target)));
+
+	return 0;
+}
+
+static int l2_cache_pmu_probe_cluster(struct device *dev, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev->parent);
+	struct platform_device *sdev = to_platform_device(dev);
+	struct l2cache_pmu *l2cache_pmu = data;
+	struct cluster_pmu *cluster;
+	struct acpi_device *device;
+	unsigned long fw_cluster_id;
+	int err;
+	int irq;
+
+	if (acpi_bus_get_device(ACPI_HANDLE(dev), &device))
+		return -ENODEV;
+
+	if (kstrtoul(device->pnp.unique_id, 10, &fw_cluster_id) < 0) {
+		dev_err(&pdev->dev, "unable to read ACPI uid\n");
+		return -ENODEV;
+	}
+
+	cluster = devm_kzalloc(&pdev->dev, sizeof(*cluster), GFP_KERNEL);
+	if (!cluster)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&cluster->next);
+	list_add(&cluster->next, &l2cache_pmu->clusters);
+	cluster->cluster_id = fw_cluster_id;
+
+	irq = platform_get_irq(sdev, 0);
+	if (irq < 0)
+		return irq;
+	irq_set_status_flags(irq, IRQ_NOAUTOEN);
+	cluster->irq = irq;
+
+	cluster->l2cache_pmu = l2cache_pmu;
+	cluster->on_cpu = -1;
+
+	err = devm_request_irq(&pdev->dev, irq, l2_cache_handle_irq,
+			       IRQF_NOBALANCING | IRQF_NO_THREAD,
+			       "l2-cache-pmu", cluster);
+	if (err) {
+		dev_err(&pdev->dev,
+			"Unable to request IRQ%d for L2 PMU counters\n", irq);
+		return err;
+	}
+
+	dev_info(&pdev->dev,
+		"Registered L2 cache PMU cluster %ld\n", fw_cluster_id);
+
+	spin_lock_init(&cluster->pmu_lock);
+
+	l2cache_pmu->num_pmus++;
+
+	return 0;
+}
+
+static int l2_cache_pmu_probe(struct platform_device *pdev)
+{
+	int err;
+	struct l2cache_pmu *l2cache_pmu;
+
+	l2cache_pmu =
+		devm_kzalloc(&pdev->dev, sizeof(*l2cache_pmu), GFP_KERNEL);
+	if (!l2cache_pmu)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&l2cache_pmu->clusters);
+
+	platform_set_drvdata(pdev, l2cache_pmu);
+	l2cache_pmu->pmu = (struct pmu) {
+		/* suffix is instance id for future use with multiple sockets */
+		.name		= "l2cache_0",
+		.task_ctx_nr    = perf_invalid_context,
+		.pmu_enable	= l2_cache_pmu_enable,
+		.pmu_disable	= l2_cache_pmu_disable,
+		.event_init	= l2_cache_event_init,
+		.add		= l2_cache_event_add,
+		.del		= l2_cache_event_del,
+		.start		= l2_cache_event_start,
+		.stop		= l2_cache_event_stop,
+		.read		= l2_cache_event_read,
+		.attr_groups	= l2_cache_pmu_attr_grps,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	l2cache_pmu->num_counters = get_num_counters();
+	l2cache_pmu->pdev = pdev;
+	l2cache_pmu->pmu_cluster = devm_alloc_percpu(&pdev->dev,
+						     struct cluster_pmu *);
+	if (!l2cache_pmu->pmu_cluster)
+		return -ENOMEM;
+
+	l2_cycle_ctr_idx = l2cache_pmu->num_counters - 1;
+	l2_counter_present_mask = GENMASK(l2cache_pmu->num_counters - 2, 0) |
+		BIT(L2CYCLE_CTR_BIT);
+
+	cpumask_clear(&l2cache_pmu->cpumask);
+
+	/* Read cluster info and initialize each cluster */
+	err = device_for_each_child(&pdev->dev, l2cache_pmu,
+				    l2_cache_pmu_probe_cluster);
+	if (err)
+		return err;
+
+	if (l2cache_pmu->num_pmus == 0) {
+		dev_err(&pdev->dev, "No hardware L2 cache PMUs found\n");
+		return -ENODEV;
+	}
+
+	err = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
+				       &l2cache_pmu->node);
+	if (err) {
+		dev_err(&pdev->dev, "Error %d registering hotplug", err);
+		return err;
+	}
+
+	err = perf_pmu_register(&l2cache_pmu->pmu, l2cache_pmu->pmu.name, -1);
+	if (err) {
+		dev_err(&pdev->dev, "Error %d registering L2 cache PMU\n", err);
+		goto out_unregister;
+	}
+
+	dev_info(&pdev->dev, "Registered L2 cache PMU using %d HW PMUs\n",
+		 l2cache_pmu->num_pmus);
+
+	return err;
+
+out_unregister:
+	cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
+				    &l2cache_pmu->node);
+	return err;
+}
+
+static int l2_cache_pmu_remove(struct platform_device *pdev)
+{
+	struct l2cache_pmu *l2cache_pmu =
+		to_l2cache_pmu(platform_get_drvdata(pdev));
+
+	perf_pmu_unregister(&l2cache_pmu->pmu);
+	cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
+				    &l2cache_pmu->node);
+	return 0;
+}
+
+static struct platform_driver l2_cache_pmu_driver = {
+	.driver = {
+		.name = "qcom-l2cache-pmu",
+		.acpi_match_table = ACPI_PTR(l2_cache_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = l2_cache_pmu_probe,
+	.remove = l2_cache_pmu_remove,
+};
+
+static int __init register_l2_cache_pmu_driver(void)
+{
+	int err;
+
+	err = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
+				      "AP_PERF_ARM_QCOM_L2_ONLINE",
+				      l2cache_pmu_online_cpu,
+				      l2cache_pmu_offline_cpu);
+	if (err)
+		return err;
+
+	return platform_driver_register(&l2_cache_pmu_driver);
+}
+device_initcall(register_l2_cache_pmu_driver);
diff --git a/drivers/perf/qcom_l3_pmu.c b/drivers/perf/qcom_l3_pmu.c
new file mode 100644
index 000000000..9ddb577c5
--- /dev/null
+++ b/drivers/perf/qcom_l3_pmu.c
@@ -0,0 +1,836 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Driver for the L3 cache PMUs in Qualcomm Technologies chips.
+ *
+ * The driver supports a distributed cache architecture where the overall
+ * cache for a socket is comprised of multiple slices each with its own PMU.
+ * Access to each individual PMU is provided even though all CPUs share all
+ * the slices. User space needs to aggregate to individual counts to provide
+ * a global picture.
+ *
+ * See Documentation/admin-guide/perf/qcom_l3_pmu.rst for more details.
+ *
+ * Copyright (c) 2015-2017, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/acpi.h>
+#include <linux/bitops.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+
+/*
+ * General constants
+ */
+
+/* Number of counters on each PMU */
+#define L3_NUM_COUNTERS  8
+/* Mask for the event type field within perf_event_attr.config and EVTYPE reg */
+#define L3_EVTYPE_MASK   0xFF
+/*
+ * Bit position of the 'long counter' flag within perf_event_attr.config.
+ * Reserve some space between the event type and this flag to allow expansion
+ * in the event type field.
+ */
+#define L3_EVENT_LC_BIT  32
+
+/*
+ * Register offsets
+ */
+
+/* Perfmon registers */
+#define L3_HML3_PM_CR       0x000
+#define L3_HML3_PM_EVCNTR(__cntr) (0x420 + ((__cntr) & 0x7) * 8)
+#define L3_HML3_PM_CNTCTL(__cntr) (0x120 + ((__cntr) & 0x7) * 8)
+#define L3_HML3_PM_EVTYPE(__cntr) (0x220 + ((__cntr) & 0x7) * 8)
+#define L3_HML3_PM_FILTRA   0x300
+#define L3_HML3_PM_FILTRB   0x308
+#define L3_HML3_PM_FILTRC   0x310
+#define L3_HML3_PM_FILTRAM  0x304
+#define L3_HML3_PM_FILTRBM  0x30C
+#define L3_HML3_PM_FILTRCM  0x314
+
+/* Basic counter registers */
+#define L3_M_BC_CR         0x500
+#define L3_M_BC_SATROLL_CR 0x504
+#define L3_M_BC_CNTENSET   0x508
+#define L3_M_BC_CNTENCLR   0x50C
+#define L3_M_BC_INTENSET   0x510
+#define L3_M_BC_INTENCLR   0x514
+#define L3_M_BC_GANG       0x718
+#define L3_M_BC_OVSR       0x740
+#define L3_M_BC_IRQCTL     0x96C
+
+/*
+ * Bit field definitions
+ */
+
+/* L3_HML3_PM_CR */
+#define PM_CR_RESET           (0)
+
+/* L3_HML3_PM_XCNTCTL/L3_HML3_PM_CNTCTLx */
+#define PMCNT_RESET           (0)
+
+/* L3_HML3_PM_EVTYPEx */
+#define EVSEL(__val)          ((__val) & L3_EVTYPE_MASK)
+
+/* Reset value for all the filter registers */
+#define PM_FLTR_RESET         (0)
+
+/* L3_M_BC_CR */
+#define BC_RESET              (1UL << 1)
+#define BC_ENABLE             (1UL << 0)
+
+/* L3_M_BC_SATROLL_CR */
+#define BC_SATROLL_CR_RESET   (0)
+
+/* L3_M_BC_CNTENSET */
+#define PMCNTENSET(__cntr)    (1UL << ((__cntr) & 0x7))
+
+/* L3_M_BC_CNTENCLR */
+#define PMCNTENCLR(__cntr)    (1UL << ((__cntr) & 0x7))
+#define BC_CNTENCLR_RESET     (0xFF)
+
+/* L3_M_BC_INTENSET */
+#define PMINTENSET(__cntr)    (1UL << ((__cntr) & 0x7))
+
+/* L3_M_BC_INTENCLR */
+#define PMINTENCLR(__cntr)    (1UL << ((__cntr) & 0x7))
+#define BC_INTENCLR_RESET     (0xFF)
+
+/* L3_M_BC_GANG */
+#define GANG_EN(__cntr)       (1UL << ((__cntr) & 0x7))
+#define BC_GANG_RESET         (0)
+
+/* L3_M_BC_OVSR */
+#define PMOVSRCLR(__cntr)     (1UL << ((__cntr) & 0x7))
+#define PMOVSRCLR_RESET       (0xFF)
+
+/* L3_M_BC_IRQCTL */
+#define PMIRQONMSBEN(__cntr)  (1UL << ((__cntr) & 0x7))
+#define BC_IRQCTL_RESET       (0x0)
+
+/*
+ * Events
+ */
+
+#define L3_EVENT_CYCLES		0x01
+#define L3_EVENT_READ_HIT		0x20
+#define L3_EVENT_READ_MISS		0x21
+#define L3_EVENT_READ_HIT_D		0x22
+#define L3_EVENT_READ_MISS_D		0x23
+#define L3_EVENT_WRITE_HIT		0x24
+#define L3_EVENT_WRITE_MISS		0x25
+
+/*
+ * Decoding of settings from perf_event_attr
+ *
+ * The config format for perf events is:
+ * - config: bits 0-7: event type
+ *           bit  32:  HW counter size requested, 0: 32 bits, 1: 64 bits
+ */
+
+static inline u32 get_event_type(struct perf_event *event)
+{
+	return (event->attr.config) & L3_EVTYPE_MASK;
+}
+
+static inline bool event_uses_long_counter(struct perf_event *event)
+{
+	return !!(event->attr.config & BIT_ULL(L3_EVENT_LC_BIT));
+}
+
+static inline int event_num_counters(struct perf_event *event)
+{
+	return event_uses_long_counter(event) ? 2 : 1;
+}
+
+/*
+ * Main PMU, inherits from the core perf PMU type
+ */
+struct l3cache_pmu {
+	struct pmu		pmu;
+	struct hlist_node	node;
+	void __iomem		*regs;
+	struct perf_event	*events[L3_NUM_COUNTERS];
+	unsigned long		used_mask[BITS_TO_LONGS(L3_NUM_COUNTERS)];
+	cpumask_t		cpumask;
+};
+
+#define to_l3cache_pmu(p) (container_of(p, struct l3cache_pmu, pmu))
+
+/*
+ * Type used to group hardware counter operations
+ *
+ * Used to implement two types of hardware counters, standard (32bits) and
+ * long (64bits). The hardware supports counter chaining which we use to
+ * implement long counters. This support is exposed via the 'lc' flag field
+ * in perf_event_attr.config.
+ */
+struct l3cache_event_ops {
+	/* Called to start event monitoring */
+	void (*start)(struct perf_event *event);
+	/* Called to stop event monitoring */
+	void (*stop)(struct perf_event *event, int flags);
+	/* Called to update the perf_event */
+	void (*update)(struct perf_event *event);
+};
+
+/*
+ * Implementation of long counter operations
+ *
+ * 64bit counters are implemented by chaining two of the 32bit physical
+ * counters. The PMU only supports chaining of adjacent even/odd pairs
+ * and for simplicity the driver always configures the odd counter to
+ * count the overflows of the lower-numbered even counter. Note that since
+ * the resulting hardware counter is 64bits no IRQs are required to maintain
+ * the software counter which is also 64bits.
+ */
+
+static void qcom_l3_cache__64bit_counter_start(struct perf_event *event)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 evsel = get_event_type(event);
+	u32 gang;
+
+	/* Set the odd counter to count the overflows of the even counter */
+	gang = readl_relaxed(l3pmu->regs + L3_M_BC_GANG);
+	gang |= GANG_EN(idx + 1);
+	writel_relaxed(gang, l3pmu->regs + L3_M_BC_GANG);
+
+	/* Initialize the hardware counters and reset prev_count*/
+	local64_set(&event->hw.prev_count, 0);
+	writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1));
+	writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
+
+	/*
+	 * Set the event types, the upper half must use zero and the lower
+	 * half the actual event type
+	 */
+	writel_relaxed(EVSEL(0), l3pmu->regs + L3_HML3_PM_EVTYPE(idx + 1));
+	writel_relaxed(EVSEL(evsel), l3pmu->regs + L3_HML3_PM_EVTYPE(idx));
+
+	/* Finally, enable the counters */
+	writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx + 1));
+	writel_relaxed(PMCNTENSET(idx + 1), l3pmu->regs + L3_M_BC_CNTENSET);
+	writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx));
+	writel_relaxed(PMCNTENSET(idx), l3pmu->regs + L3_M_BC_CNTENSET);
+}
+
+static void qcom_l3_cache__64bit_counter_stop(struct perf_event *event,
+					      int flags)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 gang = readl_relaxed(l3pmu->regs + L3_M_BC_GANG);
+
+	/* Disable the counters */
+	writel_relaxed(PMCNTENCLR(idx), l3pmu->regs + L3_M_BC_CNTENCLR);
+	writel_relaxed(PMCNTENCLR(idx + 1), l3pmu->regs + L3_M_BC_CNTENCLR);
+
+	/* Disable chaining */
+	writel_relaxed(gang & ~GANG_EN(idx + 1), l3pmu->regs + L3_M_BC_GANG);
+}
+
+static void qcom_l3_cache__64bit_counter_update(struct perf_event *event)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 hi, lo;
+	u64 prev, new;
+
+	do {
+		prev = local64_read(&event->hw.prev_count);
+		do {
+			hi = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1));
+			lo = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
+		} while (hi != readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1)));
+		new = ((u64)hi << 32) | lo;
+	} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);
+
+	local64_add(new - prev, &event->count);
+}
+
+static const struct l3cache_event_ops event_ops_long = {
+	.start = qcom_l3_cache__64bit_counter_start,
+	.stop = qcom_l3_cache__64bit_counter_stop,
+	.update = qcom_l3_cache__64bit_counter_update,
+};
+
+/*
+ * Implementation of standard counter operations
+ *
+ * 32bit counters use a single physical counter and a hardware feature that
+ * asserts the overflow IRQ on the toggling of the most significant bit in
+ * the counter. This feature allows the counters to be left free-running
+ * without needing the usual reprogramming required to properly handle races
+ * during concurrent calls to update.
+ */
+
+static void qcom_l3_cache__32bit_counter_start(struct perf_event *event)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 evsel = get_event_type(event);
+	u32 irqctl = readl_relaxed(l3pmu->regs + L3_M_BC_IRQCTL);
+
+	/* Set the counter to assert the overflow IRQ on MSB toggling */
+	writel_relaxed(irqctl | PMIRQONMSBEN(idx), l3pmu->regs + L3_M_BC_IRQCTL);
+
+	/* Initialize the hardware counter and reset prev_count*/
+	local64_set(&event->hw.prev_count, 0);
+	writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
+
+	/* Set the event type */
+	writel_relaxed(EVSEL(evsel), l3pmu->regs + L3_HML3_PM_EVTYPE(idx));
+
+	/* Enable interrupt generation by this counter */
+	writel_relaxed(PMINTENSET(idx), l3pmu->regs + L3_M_BC_INTENSET);
+
+	/* Finally, enable the counter */
+	writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx));
+	writel_relaxed(PMCNTENSET(idx), l3pmu->regs + L3_M_BC_CNTENSET);
+}
+
+static void qcom_l3_cache__32bit_counter_stop(struct perf_event *event,
+					      int flags)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 irqctl = readl_relaxed(l3pmu->regs + L3_M_BC_IRQCTL);
+
+	/* Disable the counter */
+	writel_relaxed(PMCNTENCLR(idx), l3pmu->regs + L3_M_BC_CNTENCLR);
+
+	/* Disable interrupt generation by this counter */
+	writel_relaxed(PMINTENCLR(idx), l3pmu->regs + L3_M_BC_INTENCLR);
+
+	/* Set the counter to not assert the overflow IRQ on MSB toggling */
+	writel_relaxed(irqctl & ~PMIRQONMSBEN(idx), l3pmu->regs + L3_M_BC_IRQCTL);
+}
+
+static void qcom_l3_cache__32bit_counter_update(struct perf_event *event)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	int idx = event->hw.idx;
+	u32 prev, new;
+
+	do {
+		prev = local64_read(&event->hw.prev_count);
+		new = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
+	} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);
+
+	local64_add(new - prev, &event->count);
+}
+
+static const struct l3cache_event_ops event_ops_std = {
+	.start = qcom_l3_cache__32bit_counter_start,
+	.stop = qcom_l3_cache__32bit_counter_stop,
+	.update = qcom_l3_cache__32bit_counter_update,
+};
+
+/* Retrieve the appropriate operations for the given event */
+static
+const struct l3cache_event_ops *l3cache_event_get_ops(struct perf_event *event)
+{
+	if (event_uses_long_counter(event))
+		return &event_ops_long;
+	else
+		return &event_ops_std;
+}
+
+/*
+ * Top level PMU functions.
+ */
+
+static inline void qcom_l3_cache__init(struct l3cache_pmu *l3pmu)
+{
+	int i;
+
+	writel_relaxed(BC_RESET, l3pmu->regs + L3_M_BC_CR);
+
+	/*
+	 * Use writel for the first programming command to ensure the basic
+	 * counter unit is stopped before proceeding
+	 */
+	writel(BC_SATROLL_CR_RESET, l3pmu->regs + L3_M_BC_SATROLL_CR);
+
+	writel_relaxed(BC_CNTENCLR_RESET, l3pmu->regs + L3_M_BC_CNTENCLR);
+	writel_relaxed(BC_INTENCLR_RESET, l3pmu->regs + L3_M_BC_INTENCLR);
+	writel_relaxed(PMOVSRCLR_RESET, l3pmu->regs + L3_M_BC_OVSR);
+	writel_relaxed(BC_GANG_RESET, l3pmu->regs + L3_M_BC_GANG);
+	writel_relaxed(BC_IRQCTL_RESET, l3pmu->regs + L3_M_BC_IRQCTL);
+	writel_relaxed(PM_CR_RESET, l3pmu->regs + L3_HML3_PM_CR);
+
+	for (i = 0; i < L3_NUM_COUNTERS; ++i) {
+		writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(i));
+		writel_relaxed(EVSEL(0), l3pmu->regs + L3_HML3_PM_EVTYPE(i));
+	}
+
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRA);
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRAM);
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRB);
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRBM);
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRC);
+	writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRCM);
+
+	/*
+	 * Use writel here to ensure all programming commands are done
+	 *  before proceeding
+	 */
+	writel(BC_ENABLE, l3pmu->regs + L3_M_BC_CR);
+}
+
+static irqreturn_t qcom_l3_cache__handle_irq(int irq_num, void *data)
+{
+	struct l3cache_pmu *l3pmu = data;
+	/* Read the overflow status register */
+	long status = readl_relaxed(l3pmu->regs + L3_M_BC_OVSR);
+	int idx;
+
+	if (status == 0)
+		return IRQ_NONE;
+
+	/* Clear the bits we read on the overflow status register */
+	writel_relaxed(status, l3pmu->regs + L3_M_BC_OVSR);
+
+	for_each_set_bit(idx, &status, L3_NUM_COUNTERS) {
+		struct perf_event *event;
+		const struct l3cache_event_ops *ops;
+
+		event = l3pmu->events[idx];
+		if (!event)
+			continue;
+
+		/*
+		 * Since the IRQ is not enabled for events using long counters
+		 * we should never see one of those here, however, be consistent
+		 * and use the ops indirections like in the other operations.
+		 */
+
+		ops = l3cache_event_get_ops(event);
+		ops->update(event);
+	}
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Implementation of abstract pmu functionality required by
+ * the core perf events code.
+ */
+
+static void qcom_l3_cache__pmu_enable(struct pmu *pmu)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(pmu);
+
+	/* Ensure the other programming commands are observed before enabling */
+	wmb();
+
+	writel_relaxed(BC_ENABLE, l3pmu->regs + L3_M_BC_CR);
+}
+
+static void qcom_l3_cache__pmu_disable(struct pmu *pmu)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(pmu);
+
+	writel_relaxed(0, l3pmu->regs + L3_M_BC_CR);
+
+	/* Ensure the basic counter unit is stopped before proceeding */
+	wmb();
+}
+
+/*
+ * We must NOT create groups containing events from multiple hardware PMUs,
+ * although mixing different software and hardware PMUs is allowed.
+ */
+static bool qcom_l3_cache__validate_event_group(struct perf_event *event)
+{
+	struct perf_event *leader = event->group_leader;
+	struct perf_event *sibling;
+	int counters = 0;
+
+	if (leader->pmu != event->pmu && !is_software_event(leader))
+		return false;
+
+	counters = event_num_counters(event);
+	counters += event_num_counters(leader);
+
+	for_each_sibling_event(sibling, leader) {
+		if (is_software_event(sibling))
+			continue;
+		if (sibling->pmu != event->pmu)
+			return false;
+		counters += event_num_counters(sibling);
+	}
+
+	/*
+	 * If the group requires more counters than the HW has, it
+	 * cannot ever be scheduled.
+	 */
+	return counters <= L3_NUM_COUNTERS;
+}
+
+static int qcom_l3_cache__event_init(struct perf_event *event)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+
+	/*
+	 * Is the event for this PMU?
+	 */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * Sampling not supported since these events are not core-attributable.
+	 */
+	if (hwc->sample_period)
+		return -EINVAL;
+
+	/*
+	 * Task mode not available, we run the counters as socket counters,
+	 * not attributable to any CPU and therefore cannot attribute per-task.
+	 */
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	/* Validate the group */
+	if (!qcom_l3_cache__validate_event_group(event))
+		return -EINVAL;
+
+	hwc->idx = -1;
+
+	/*
+	 * Many perf core operations (eg. events rotation) operate on a
+	 * single CPU context. This is obvious for CPU PMUs, where one
+	 * expects the same sets of events being observed on all CPUs,
+	 * but can lead to issues for off-core PMUs, like this one, where
+	 * each event could be theoretically assigned to a different CPU.
+	 * To mitigate this, we enforce CPU assignment to one designated
+	 * processor (the one described in the "cpumask" attribute exported
+	 * by the PMU device). perf user space tools honor this and avoid
+	 * opening more than one copy of the events.
+	 */
+	event->cpu = cpumask_first(&l3pmu->cpumask);
+
+	return 0;
+}
+
+static void qcom_l3_cache__event_start(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
+
+	hwc->state = 0;
+	ops->start(event);
+}
+
+static void qcom_l3_cache__event_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
+
+	if (hwc->state & PERF_HES_STOPPED)
+		return;
+
+	ops->stop(event, flags);
+	if (flags & PERF_EF_UPDATE)
+		ops->update(event);
+	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
+}
+
+static int qcom_l3_cache__event_add(struct perf_event *event, int flags)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int order = event_uses_long_counter(event) ? 1 : 0;
+	int idx;
+
+	/*
+	 * Try to allocate a counter.
+	 */
+	idx = bitmap_find_free_region(l3pmu->used_mask, L3_NUM_COUNTERS, order);
+	if (idx < 0)
+		/* The counters are all in use. */
+		return -EAGAIN;
+
+	hwc->idx = idx;
+	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+	l3pmu->events[idx] = event;
+
+	if (flags & PERF_EF_START)
+		qcom_l3_cache__event_start(event, 0);
+
+	/* Propagate changes to the userspace mapping. */
+	perf_event_update_userpage(event);
+
+	return 0;
+}
+
+static void qcom_l3_cache__event_del(struct perf_event *event, int flags)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	int order = event_uses_long_counter(event) ? 1 : 0;
+
+	/* Stop and clean up */
+	qcom_l3_cache__event_stop(event,  flags | PERF_EF_UPDATE);
+	l3pmu->events[hwc->idx] = NULL;
+	bitmap_release_region(l3pmu->used_mask, hwc->idx, order);
+
+	/* Propagate changes to the userspace mapping. */
+	perf_event_update_userpage(event);
+}
+
+static void qcom_l3_cache__event_read(struct perf_event *event)
+{
+	const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
+
+	ops->update(event);
+}
+
+/*
+ * Add sysfs attributes
+ *
+ * We export:
+ * - formats, used by perf user space and other tools to configure events
+ * - events, used by perf user space and other tools to create events
+ *   symbolically, e.g.:
+ *     perf stat -a -e l3cache_0_0/event=read-miss/ ls
+ *     perf stat -a -e l3cache_0_0/event=0x21/ ls
+ * - cpumask, used by perf user space and other tools to know on which CPUs
+ *   to open the events
+ */
+
+/* formats */
+
+static ssize_t l3cache_pmu_format_show(struct device *dev,
+				       struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+	return sprintf(buf, "%s\n", (char *) eattr->var);
+}
+
+#define L3CACHE_PMU_FORMAT_ATTR(_name, _config)				      \
+	(&((struct dev_ext_attribute[]) {				      \
+		{ .attr = __ATTR(_name, 0444, l3cache_pmu_format_show, NULL), \
+		  .var = (void *) _config, }				      \
+	})[0].attr.attr)
+
+static struct attribute *qcom_l3_cache_pmu_formats[] = {
+	L3CACHE_PMU_FORMAT_ATTR(event, "config:0-7"),
+	L3CACHE_PMU_FORMAT_ATTR(lc, "config:" __stringify(L3_EVENT_LC_BIT)),
+	NULL,
+};
+
+static struct attribute_group qcom_l3_cache_pmu_format_group = {
+	.name = "format",
+	.attrs = qcom_l3_cache_pmu_formats,
+};
+
+/* events */
+
+static ssize_t l3cache_pmu_event_show(struct device *dev,
+				     struct device_attribute *attr, char *page)
+{
+	struct perf_pmu_events_attr *pmu_attr;
+
+	pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
+	return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
+}
+
+#define L3CACHE_EVENT_ATTR(_name, _id)					     \
+	(&((struct perf_pmu_events_attr[]) {				     \
+		{ .attr = __ATTR(_name, 0444, l3cache_pmu_event_show, NULL), \
+		  .id = _id, }						     \
+	})[0].attr.attr)
+
+static struct attribute *qcom_l3_cache_pmu_events[] = {
+	L3CACHE_EVENT_ATTR(cycles, L3_EVENT_CYCLES),
+	L3CACHE_EVENT_ATTR(read-hit, L3_EVENT_READ_HIT),
+	L3CACHE_EVENT_ATTR(read-miss, L3_EVENT_READ_MISS),
+	L3CACHE_EVENT_ATTR(read-hit-d-side, L3_EVENT_READ_HIT_D),
+	L3CACHE_EVENT_ATTR(read-miss-d-side, L3_EVENT_READ_MISS_D),
+	L3CACHE_EVENT_ATTR(write-hit, L3_EVENT_WRITE_HIT),
+	L3CACHE_EVENT_ATTR(write-miss, L3_EVENT_WRITE_MISS),
+	NULL
+};
+
+static struct attribute_group qcom_l3_cache_pmu_events_group = {
+	.name = "events",
+	.attrs = qcom_l3_cache_pmu_events,
+};
+
+/* cpumask */
+
+static ssize_t qcom_l3_cache_pmu_cpumask_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
+{
+	struct l3cache_pmu *l3pmu = to_l3cache_pmu(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, &l3pmu->cpumask);
+}
+
+static DEVICE_ATTR(cpumask, 0444, qcom_l3_cache_pmu_cpumask_show, NULL);
+
+static struct attribute *qcom_l3_cache_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group qcom_l3_cache_pmu_cpumask_attr_group = {
+	.attrs = qcom_l3_cache_pmu_cpumask_attrs,
+};
+
+/*
+ * Per PMU device attribute groups
+ */
+static const struct attribute_group *qcom_l3_cache_pmu_attr_grps[] = {
+	&qcom_l3_cache_pmu_format_group,
+	&qcom_l3_cache_pmu_events_group,
+	&qcom_l3_cache_pmu_cpumask_attr_group,
+	NULL,
+};
+
+/*
+ * Probing functions and data.
+ */
+
+static int qcom_l3_cache_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct l3cache_pmu *l3pmu = hlist_entry_safe(node, struct l3cache_pmu, node);
+
+	/* If there is not a CPU/PMU association pick this CPU */
+	if (cpumask_empty(&l3pmu->cpumask))
+		cpumask_set_cpu(cpu, &l3pmu->cpumask);
+
+	return 0;
+}
+
+static int qcom_l3_cache_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct l3cache_pmu *l3pmu = hlist_entry_safe(node, struct l3cache_pmu, node);
+	unsigned int target;
+
+	if (!cpumask_test_and_clear_cpu(cpu, &l3pmu->cpumask))
+		return 0;
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+	perf_pmu_migrate_context(&l3pmu->pmu, cpu, target);
+	cpumask_set_cpu(target, &l3pmu->cpumask);
+	return 0;
+}
+
+static int qcom_l3_cache_pmu_probe(struct platform_device *pdev)
+{
+	struct l3cache_pmu *l3pmu;
+	struct acpi_device *acpi_dev;
+	struct resource *memrc;
+	int ret;
+	char *name;
+
+	/* Initialize the PMU data structures */
+
+	acpi_dev = ACPI_COMPANION(&pdev->dev);
+	if (!acpi_dev)
+		return -ENODEV;
+
+	l3pmu = devm_kzalloc(&pdev->dev, sizeof(*l3pmu), GFP_KERNEL);
+	name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "l3cache_%s_%s",
+		      acpi_dev->parent->pnp.unique_id, acpi_dev->pnp.unique_id);
+	if (!l3pmu || !name)
+		return -ENOMEM;
+
+	l3pmu->pmu = (struct pmu) {
+		.task_ctx_nr	= perf_invalid_context,
+
+		.pmu_enable	= qcom_l3_cache__pmu_enable,
+		.pmu_disable	= qcom_l3_cache__pmu_disable,
+		.event_init	= qcom_l3_cache__event_init,
+		.add		= qcom_l3_cache__event_add,
+		.del		= qcom_l3_cache__event_del,
+		.start		= qcom_l3_cache__event_start,
+		.stop		= qcom_l3_cache__event_stop,
+		.read		= qcom_l3_cache__event_read,
+
+		.attr_groups	= qcom_l3_cache_pmu_attr_grps,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	memrc = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	l3pmu->regs = devm_ioremap_resource(&pdev->dev, memrc);
+	if (IS_ERR(l3pmu->regs)) {
+		dev_err(&pdev->dev, "Can't map PMU @%pa\n", &memrc->start);
+		return PTR_ERR(l3pmu->regs);
+	}
+
+	qcom_l3_cache__init(l3pmu);
+
+	ret = platform_get_irq(pdev, 0);
+	if (ret <= 0)
+		return ret;
+
+	ret = devm_request_irq(&pdev->dev, ret, qcom_l3_cache__handle_irq, 0,
+			       name, l3pmu);
+	if (ret) {
+		dev_err(&pdev->dev, "Request for IRQ failed for slice @%pa\n",
+			&memrc->start);
+		return ret;
+	}
+
+	/* Add this instance to the list used by the offline callback */
+	ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE, &l3pmu->node);
+	if (ret) {
+		dev_err(&pdev->dev, "Error %d registering hotplug", ret);
+		return ret;
+	}
+
+	ret = perf_pmu_register(&l3pmu->pmu, name, -1);
+	if (ret < 0) {
+		dev_err(&pdev->dev, "Failed to register L3 cache PMU (%d)\n", ret);
+		return ret;
+	}
+
+	dev_info(&pdev->dev, "Registered %s, type: %d\n", name, l3pmu->pmu.type);
+
+	return 0;
+}
+
+static const struct acpi_device_id qcom_l3_cache_pmu_acpi_match[] = {
+	{ "QCOM8081", },
+	{ }
+};
+MODULE_DEVICE_TABLE(acpi, qcom_l3_cache_pmu_acpi_match);
+
+static struct platform_driver qcom_l3_cache_pmu_driver = {
+	.driver = {
+		.name = "qcom-l3cache-pmu",
+		.acpi_match_table = ACPI_PTR(qcom_l3_cache_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = qcom_l3_cache_pmu_probe,
+};
+
+static int __init register_qcom_l3_cache_pmu_driver(void)
+{
+	int ret;
+
+	/* Install a hook to update the reader CPU in case it goes offline */
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE,
+				      "perf/qcom/l3cache:online",
+				      qcom_l3_cache_pmu_online_cpu,
+				      qcom_l3_cache_pmu_offline_cpu);
+	if (ret)
+		return ret;
+
+	return platform_driver_register(&qcom_l3_cache_pmu_driver);
+}
+device_initcall(register_qcom_l3_cache_pmu_driver);
diff --git a/drivers/perf/thunderx2_pmu.c b/drivers/perf/thunderx2_pmu.c
new file mode 100644
index 000000000..e116815fa
--- /dev/null
+++ b/drivers/perf/thunderx2_pmu.c
@@ -0,0 +1,1058 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * CAVIUM THUNDERX2 SoC PMU UNCORE
+ * Copyright (C) 2018 Cavium Inc.
+ * Author: Ganapatrao Kulkarni <gkulkarni@cavium.com>
+ */
+
+#include <linux/acpi.h>
+#include <linux/cpuhotplug.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+
+/* Each ThunderX2(TX2) Socket has a L3C and DMC UNCORE PMU device.
+ * Each UNCORE PMU device consists of 4 independent programmable counters.
+ * Counters are 32 bit and do not support overflow interrupt,
+ * they need to be sampled before overflow(i.e, at every 2 seconds).
+ */
+
+#define TX2_PMU_DMC_L3C_MAX_COUNTERS	4
+#define TX2_PMU_CCPI2_MAX_COUNTERS	8
+#define TX2_PMU_MAX_COUNTERS		TX2_PMU_CCPI2_MAX_COUNTERS
+
+
+#define TX2_PMU_DMC_CHANNELS		8
+#define TX2_PMU_L3_TILES		16
+
+#define TX2_PMU_HRTIMER_INTERVAL	(2 * NSEC_PER_SEC)
+#define GET_EVENTID(ev, mask)		((ev->hw.config) & mask)
+#define GET_COUNTERID(ev, mask)		((ev->hw.idx) & mask)
+ /* 1 byte per counter(4 counters).
+  * Event id is encoded in bits [5:1] of a byte,
+  */
+#define DMC_EVENT_CFG(idx, val)		((val) << (((idx) * 8) + 1))
+
+/* bits[3:0] to select counters, are indexed from 8 to 15. */
+#define CCPI2_COUNTER_OFFSET		8
+
+#define L3C_COUNTER_CTL			0xA8
+#define L3C_COUNTER_DATA		0xAC
+#define DMC_COUNTER_CTL			0x234
+#define DMC_COUNTER_DATA		0x240
+
+#define CCPI2_PERF_CTL			0x108
+#define CCPI2_COUNTER_CTL		0x10C
+#define CCPI2_COUNTER_SEL		0x12c
+#define CCPI2_COUNTER_DATA_L		0x130
+#define CCPI2_COUNTER_DATA_H		0x134
+
+/* L3C event IDs */
+#define L3_EVENT_READ_REQ		0xD
+#define L3_EVENT_WRITEBACK_REQ		0xE
+#define L3_EVENT_INV_N_WRITE_REQ	0xF
+#define L3_EVENT_INV_REQ		0x10
+#define L3_EVENT_EVICT_REQ		0x13
+#define L3_EVENT_INV_N_WRITE_HIT	0x14
+#define L3_EVENT_INV_HIT		0x15
+#define L3_EVENT_READ_HIT		0x17
+#define L3_EVENT_MAX			0x18
+
+/* DMC event IDs */
+#define DMC_EVENT_COUNT_CYCLES		0x1
+#define DMC_EVENT_WRITE_TXNS		0xB
+#define DMC_EVENT_DATA_TRANSFERS	0xD
+#define DMC_EVENT_READ_TXNS		0xF
+#define DMC_EVENT_MAX			0x10
+
+#define CCPI2_EVENT_REQ_PKT_SENT	0x3D
+#define CCPI2_EVENT_SNOOP_PKT_SENT	0x65
+#define CCPI2_EVENT_DATA_PKT_SENT	0x105
+#define CCPI2_EVENT_GIC_PKT_SENT	0x12D
+#define CCPI2_EVENT_MAX			0x200
+
+#define CCPI2_PERF_CTL_ENABLE		BIT(0)
+#define CCPI2_PERF_CTL_START		BIT(1)
+#define CCPI2_PERF_CTL_RESET		BIT(4)
+#define CCPI2_EVENT_LEVEL_RISING_EDGE	BIT(10)
+#define CCPI2_EVENT_TYPE_EDGE_SENSITIVE	BIT(11)
+
+enum tx2_uncore_type {
+	PMU_TYPE_L3C,
+	PMU_TYPE_DMC,
+	PMU_TYPE_CCPI2,
+	PMU_TYPE_INVALID,
+};
+
+/*
+ * Each socket has 3 uncore devices associated with a PMU. The DMC and
+ * L3C have 4 32-bit counters and the CCPI2 has 8 64-bit counters.
+ */
+struct tx2_uncore_pmu {
+	struct hlist_node hpnode;
+	struct list_head  entry;
+	struct pmu pmu;
+	char *name;
+	int node;
+	int cpu;
+	u32 max_counters;
+	u32 counters_mask;
+	u32 prorate_factor;
+	u32 max_events;
+	u32 events_mask;
+	u64 hrtimer_interval;
+	void __iomem *base;
+	DECLARE_BITMAP(active_counters, TX2_PMU_MAX_COUNTERS);
+	struct perf_event *events[TX2_PMU_MAX_COUNTERS];
+	struct device *dev;
+	struct hrtimer hrtimer;
+	const struct attribute_group **attr_groups;
+	enum tx2_uncore_type type;
+	enum hrtimer_restart (*hrtimer_callback)(struct hrtimer *cb);
+	void (*init_cntr_base)(struct perf_event *event,
+			struct tx2_uncore_pmu *tx2_pmu);
+	void (*stop_event)(struct perf_event *event);
+	void (*start_event)(struct perf_event *event, int flags);
+};
+
+static LIST_HEAD(tx2_pmus);
+
+static inline struct tx2_uncore_pmu *pmu_to_tx2_pmu(struct pmu *pmu)
+{
+	return container_of(pmu, struct tx2_uncore_pmu, pmu);
+}
+
+#define TX2_PMU_FORMAT_ATTR(_var, _name, _format)			\
+static ssize_t								\
+__tx2_pmu_##_var##_show(struct device *dev,				\
+			       struct device_attribute *attr,		\
+			       char *page)				\
+{									\
+	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			\
+	return sprintf(page, _format "\n");				\
+}									\
+									\
+static struct device_attribute format_attr_##_var =			\
+	__ATTR(_name, 0444, __tx2_pmu_##_var##_show, NULL)
+
+TX2_PMU_FORMAT_ATTR(event, event, "config:0-4");
+TX2_PMU_FORMAT_ATTR(event_ccpi2, event, "config:0-9");
+
+static struct attribute *l3c_pmu_format_attrs[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute *dmc_pmu_format_attrs[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute *ccpi2_pmu_format_attrs[] = {
+	&format_attr_event_ccpi2.attr,
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = l3c_pmu_format_attrs,
+};
+
+static const struct attribute_group dmc_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = dmc_pmu_format_attrs,
+};
+
+static const struct attribute_group ccpi2_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = ccpi2_pmu_format_attrs,
+};
+
+/*
+ * sysfs event attributes
+ */
+static ssize_t tx2_pmu_event_show(struct device *dev,
+				    struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+	return sprintf(buf, "event=0x%lx\n", (unsigned long) eattr->var);
+}
+
+#define TX2_EVENT_ATTR(name, config) \
+	PMU_EVENT_ATTR(name, tx2_pmu_event_attr_##name, \
+			config, tx2_pmu_event_show)
+
+TX2_EVENT_ATTR(read_request, L3_EVENT_READ_REQ);
+TX2_EVENT_ATTR(writeback_request, L3_EVENT_WRITEBACK_REQ);
+TX2_EVENT_ATTR(inv_nwrite_request, L3_EVENT_INV_N_WRITE_REQ);
+TX2_EVENT_ATTR(inv_request, L3_EVENT_INV_REQ);
+TX2_EVENT_ATTR(evict_request, L3_EVENT_EVICT_REQ);
+TX2_EVENT_ATTR(inv_nwrite_hit, L3_EVENT_INV_N_WRITE_HIT);
+TX2_EVENT_ATTR(inv_hit, L3_EVENT_INV_HIT);
+TX2_EVENT_ATTR(read_hit, L3_EVENT_READ_HIT);
+
+static struct attribute *l3c_pmu_events_attrs[] = {
+	&tx2_pmu_event_attr_read_request.attr.attr,
+	&tx2_pmu_event_attr_writeback_request.attr.attr,
+	&tx2_pmu_event_attr_inv_nwrite_request.attr.attr,
+	&tx2_pmu_event_attr_inv_request.attr.attr,
+	&tx2_pmu_event_attr_evict_request.attr.attr,
+	&tx2_pmu_event_attr_inv_nwrite_hit.attr.attr,
+	&tx2_pmu_event_attr_inv_hit.attr.attr,
+	&tx2_pmu_event_attr_read_hit.attr.attr,
+	NULL,
+};
+
+TX2_EVENT_ATTR(cnt_cycles, DMC_EVENT_COUNT_CYCLES);
+TX2_EVENT_ATTR(write_txns, DMC_EVENT_WRITE_TXNS);
+TX2_EVENT_ATTR(data_transfers, DMC_EVENT_DATA_TRANSFERS);
+TX2_EVENT_ATTR(read_txns, DMC_EVENT_READ_TXNS);
+
+static struct attribute *dmc_pmu_events_attrs[] = {
+	&tx2_pmu_event_attr_cnt_cycles.attr.attr,
+	&tx2_pmu_event_attr_write_txns.attr.attr,
+	&tx2_pmu_event_attr_data_transfers.attr.attr,
+	&tx2_pmu_event_attr_read_txns.attr.attr,
+	NULL,
+};
+
+TX2_EVENT_ATTR(req_pktsent, CCPI2_EVENT_REQ_PKT_SENT);
+TX2_EVENT_ATTR(snoop_pktsent, CCPI2_EVENT_SNOOP_PKT_SENT);
+TX2_EVENT_ATTR(data_pktsent, CCPI2_EVENT_DATA_PKT_SENT);
+TX2_EVENT_ATTR(gic_pktsent, CCPI2_EVENT_GIC_PKT_SENT);
+
+static struct attribute *ccpi2_pmu_events_attrs[] = {
+	&tx2_pmu_event_attr_req_pktsent.attr.attr,
+	&tx2_pmu_event_attr_snoop_pktsent.attr.attr,
+	&tx2_pmu_event_attr_data_pktsent.attr.attr,
+	&tx2_pmu_event_attr_gic_pktsent.attr.attr,
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = l3c_pmu_events_attrs,
+};
+
+static const struct attribute_group dmc_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = dmc_pmu_events_attrs,
+};
+
+static const struct attribute_group ccpi2_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = ccpi2_pmu_events_attrs,
+};
+
+/*
+ * sysfs cpumask attributes
+ */
+static ssize_t cpumask_show(struct device *dev, struct device_attribute *attr,
+		char *buf)
+{
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	tx2_pmu = pmu_to_tx2_pmu(dev_get_drvdata(dev));
+	return cpumap_print_to_pagebuf(true, buf, cpumask_of(tx2_pmu->cpu));
+}
+static DEVICE_ATTR_RO(cpumask);
+
+static struct attribute *tx2_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static const struct attribute_group pmu_cpumask_attr_group = {
+	.attrs = tx2_pmu_cpumask_attrs,
+};
+
+/*
+ * Per PMU device attribute groups
+ */
+static const struct attribute_group *l3c_pmu_attr_groups[] = {
+	&l3c_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&l3c_pmu_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *dmc_pmu_attr_groups[] = {
+	&dmc_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&dmc_pmu_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *ccpi2_pmu_attr_groups[] = {
+	&ccpi2_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&ccpi2_pmu_events_attr_group,
+	NULL
+};
+
+static inline u32 reg_readl(unsigned long addr)
+{
+	return readl((void __iomem *)addr);
+}
+
+static inline void reg_writel(u32 val, unsigned long addr)
+{
+	writel(val, (void __iomem *)addr);
+}
+
+static int alloc_counter(struct tx2_uncore_pmu *tx2_pmu)
+{
+	int counter;
+
+	counter = find_first_zero_bit(tx2_pmu->active_counters,
+				tx2_pmu->max_counters);
+	if (counter == tx2_pmu->max_counters)
+		return -ENOSPC;
+
+	set_bit(counter, tx2_pmu->active_counters);
+	return counter;
+}
+
+static inline void free_counter(struct tx2_uncore_pmu *tx2_pmu, int counter)
+{
+	clear_bit(counter, tx2_pmu->active_counters);
+}
+
+static void init_cntr_base_l3c(struct perf_event *event,
+		struct tx2_uncore_pmu *tx2_pmu)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u32 cmask;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	cmask = tx2_pmu->counters_mask;
+
+	/* counter ctrl/data reg offset at 8 */
+	hwc->config_base = (unsigned long)tx2_pmu->base
+		+ L3C_COUNTER_CTL + (8 * GET_COUNTERID(event, cmask));
+	hwc->event_base =  (unsigned long)tx2_pmu->base
+		+ L3C_COUNTER_DATA + (8 * GET_COUNTERID(event, cmask));
+}
+
+static void init_cntr_base_dmc(struct perf_event *event,
+		struct tx2_uncore_pmu *tx2_pmu)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u32 cmask;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	cmask = tx2_pmu->counters_mask;
+
+	hwc->config_base = (unsigned long)tx2_pmu->base
+		+ DMC_COUNTER_CTL;
+	/* counter data reg offset at 0xc */
+	hwc->event_base = (unsigned long)tx2_pmu->base
+		+ DMC_COUNTER_DATA + (0xc * GET_COUNTERID(event, cmask));
+}
+
+static void init_cntr_base_ccpi2(struct perf_event *event,
+		struct tx2_uncore_pmu *tx2_pmu)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u32 cmask;
+
+	cmask = tx2_pmu->counters_mask;
+
+	hwc->config_base = (unsigned long)tx2_pmu->base
+		+ CCPI2_COUNTER_CTL + (4 * GET_COUNTERID(event, cmask));
+	hwc->event_base =  (unsigned long)tx2_pmu->base;
+}
+
+static void uncore_start_event_l3c(struct perf_event *event, int flags)
+{
+	u32 val, emask;
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	emask = tx2_pmu->events_mask;
+
+	/* event id encoded in bits [07:03] */
+	val = GET_EVENTID(event, emask) << 3;
+	reg_writel(val, hwc->config_base);
+	local64_set(&hwc->prev_count, 0);
+	reg_writel(0, hwc->event_base);
+}
+
+static inline void uncore_stop_event_l3c(struct perf_event *event)
+{
+	reg_writel(0, event->hw.config_base);
+}
+
+static void uncore_start_event_dmc(struct perf_event *event, int flags)
+{
+	u32 val, cmask, emask;
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+	int idx, event_id;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	cmask = tx2_pmu->counters_mask;
+	emask = tx2_pmu->events_mask;
+
+	idx = GET_COUNTERID(event, cmask);
+	event_id = GET_EVENTID(event, emask);
+
+	/* enable and start counters.
+	 * 8 bits for each counter, bits[05:01] of a counter to set event type.
+	 */
+	val = reg_readl(hwc->config_base);
+	val &= ~DMC_EVENT_CFG(idx, 0x1f);
+	val |= DMC_EVENT_CFG(idx, event_id);
+	reg_writel(val, hwc->config_base);
+	local64_set(&hwc->prev_count, 0);
+	reg_writel(0, hwc->event_base);
+}
+
+static void uncore_stop_event_dmc(struct perf_event *event)
+{
+	u32 val, cmask;
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+	int idx;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	cmask = tx2_pmu->counters_mask;
+	idx = GET_COUNTERID(event, cmask);
+
+	/* clear event type(bits[05:01]) to stop counter */
+	val = reg_readl(hwc->config_base);
+	val &= ~DMC_EVENT_CFG(idx, 0x1f);
+	reg_writel(val, hwc->config_base);
+}
+
+static void uncore_start_event_ccpi2(struct perf_event *event, int flags)
+{
+	u32 emask;
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	emask = tx2_pmu->events_mask;
+
+	/* Bit [09:00] to set event id.
+	 * Bits [10], set level to rising edge.
+	 * Bits [11], set type to edge sensitive.
+	 */
+	reg_writel((CCPI2_EVENT_TYPE_EDGE_SENSITIVE |
+			CCPI2_EVENT_LEVEL_RISING_EDGE |
+			GET_EVENTID(event, emask)), hwc->config_base);
+
+	/* reset[4], enable[0] and start[1] counters */
+	reg_writel(CCPI2_PERF_CTL_RESET |
+			CCPI2_PERF_CTL_START |
+			CCPI2_PERF_CTL_ENABLE,
+			hwc->event_base + CCPI2_PERF_CTL);
+	local64_set(&event->hw.prev_count, 0ULL);
+}
+
+static void uncore_stop_event_ccpi2(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	/* disable and stop counter */
+	reg_writel(0, hwc->event_base + CCPI2_PERF_CTL);
+}
+
+static void tx2_uncore_event_update(struct perf_event *event)
+{
+	u64 prev, delta, new = 0;
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+	enum tx2_uncore_type type;
+	u32 prorate_factor;
+	u32 cmask, emask;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	type = tx2_pmu->type;
+	cmask = tx2_pmu->counters_mask;
+	emask = tx2_pmu->events_mask;
+	prorate_factor = tx2_pmu->prorate_factor;
+	if (type == PMU_TYPE_CCPI2) {
+		reg_writel(CCPI2_COUNTER_OFFSET +
+				GET_COUNTERID(event, cmask),
+				hwc->event_base + CCPI2_COUNTER_SEL);
+		new = reg_readl(hwc->event_base + CCPI2_COUNTER_DATA_H);
+		new = (new << 32) +
+			reg_readl(hwc->event_base + CCPI2_COUNTER_DATA_L);
+		prev = local64_xchg(&hwc->prev_count, new);
+		delta = new - prev;
+	} else {
+		new = reg_readl(hwc->event_base);
+		prev = local64_xchg(&hwc->prev_count, new);
+		/* handles rollover of 32 bit counter */
+		delta = (u32)(((1UL << 32) - prev) + new);
+	}
+
+	/* DMC event data_transfers granularity is 16 Bytes, convert it to 64 */
+	if (type == PMU_TYPE_DMC &&
+			GET_EVENTID(event, emask) == DMC_EVENT_DATA_TRANSFERS)
+		delta = delta/4;
+
+	/* L3C and DMC has 16 and 8 interleave channels respectively.
+	 * The sampled value is for channel 0 and multiplied with
+	 * prorate_factor to get the count for a device.
+	 */
+	local64_add(delta * prorate_factor, &event->count);
+}
+
+static enum tx2_uncore_type get_tx2_pmu_type(struct acpi_device *adev)
+{
+	int i = 0;
+	struct acpi_tx2_pmu_device {
+		__u8 id[ACPI_ID_LEN];
+		enum tx2_uncore_type type;
+	} devices[] = {
+		{"CAV901D", PMU_TYPE_L3C},
+		{"CAV901F", PMU_TYPE_DMC},
+		{"CAV901E", PMU_TYPE_CCPI2},
+		{"", PMU_TYPE_INVALID}
+	};
+
+	while (devices[i].type != PMU_TYPE_INVALID) {
+		if (!strcmp(acpi_device_hid(adev), devices[i].id))
+			break;
+		i++;
+	}
+
+	return devices[i].type;
+}
+
+static bool tx2_uncore_validate_event(struct pmu *pmu,
+				  struct perf_event *event, int *counters)
+{
+	if (is_software_event(event))
+		return true;
+	/* Reject groups spanning multiple HW PMUs. */
+	if (event->pmu != pmu)
+		return false;
+
+	*counters = *counters + 1;
+	return true;
+}
+
+/*
+ * Make sure the group of events can be scheduled at once
+ * on the PMU.
+ */
+static bool tx2_uncore_validate_event_group(struct perf_event *event,
+		int max_counters)
+{
+	struct perf_event *sibling, *leader = event->group_leader;
+	int counters = 0;
+
+	if (event->group_leader == event)
+		return true;
+
+	if (!tx2_uncore_validate_event(event->pmu, leader, &counters))
+		return false;
+
+	for_each_sibling_event(sibling, leader) {
+		if (!tx2_uncore_validate_event(event->pmu, sibling, &counters))
+			return false;
+	}
+
+	if (!tx2_uncore_validate_event(event->pmu, event, &counters))
+		return false;
+
+	/*
+	 * If the group requires more counters than the HW has,
+	 * it cannot ever be scheduled.
+	 */
+	return counters <= max_counters;
+}
+
+
+static int tx2_uncore_event_init(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	/* Test the event attr type check for PMU enumeration */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * SOC PMU counters are shared across all cores.
+	 * Therefore, it does not support per-process mode.
+	 * Also, it does not support event sampling mode.
+	 */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EINVAL;
+
+	if (event->cpu < 0)
+		return -EINVAL;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	if (tx2_pmu->cpu >= nr_cpu_ids)
+		return -EINVAL;
+	event->cpu = tx2_pmu->cpu;
+
+	if (event->attr.config >= tx2_pmu->max_events)
+		return -EINVAL;
+
+	/* store event id */
+	hwc->config = event->attr.config;
+
+	/* Validate the group */
+	if (!tx2_uncore_validate_event_group(event, tx2_pmu->max_counters))
+		return -EINVAL;
+
+	return 0;
+}
+
+static void tx2_uncore_event_start(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	hwc->state = 0;
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+
+	tx2_pmu->start_event(event, flags);
+	perf_event_update_userpage(event);
+
+	/* No hrtimer needed for CCPI2, 64-bit counters */
+	if (!tx2_pmu->hrtimer_callback)
+		return;
+
+	/* Start timer for first event */
+	if (bitmap_weight(tx2_pmu->active_counters,
+				tx2_pmu->max_counters) == 1) {
+		hrtimer_start(&tx2_pmu->hrtimer,
+			ns_to_ktime(tx2_pmu->hrtimer_interval),
+			HRTIMER_MODE_REL_PINNED);
+	}
+}
+
+static void tx2_uncore_event_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	if (hwc->state & PERF_HES_UPTODATE)
+		return;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	tx2_pmu->stop_event(event);
+	WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+	hwc->state |= PERF_HES_STOPPED;
+	if (flags & PERF_EF_UPDATE) {
+		tx2_uncore_event_update(event);
+		hwc->state |= PERF_HES_UPTODATE;
+	}
+}
+
+static int tx2_uncore_event_add(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+
+	/* Allocate a free counter */
+	hwc->idx  = alloc_counter(tx2_pmu);
+	if (hwc->idx < 0)
+		return -EAGAIN;
+
+	tx2_pmu->events[hwc->idx] = event;
+	/* set counter control and data registers base address */
+	tx2_pmu->init_cntr_base(event, tx2_pmu);
+
+	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+	if (flags & PERF_EF_START)
+		tx2_uncore_event_start(event, flags);
+
+	return 0;
+}
+
+static void tx2_uncore_event_del(struct perf_event *event, int flags)
+{
+	struct tx2_uncore_pmu *tx2_pmu = pmu_to_tx2_pmu(event->pmu);
+	struct hw_perf_event *hwc = &event->hw;
+	u32 cmask;
+
+	cmask = tx2_pmu->counters_mask;
+	tx2_uncore_event_stop(event, PERF_EF_UPDATE);
+
+	/* clear the assigned counter */
+	free_counter(tx2_pmu, GET_COUNTERID(event, cmask));
+
+	perf_event_update_userpage(event);
+	tx2_pmu->events[hwc->idx] = NULL;
+	hwc->idx = -1;
+
+	if (!tx2_pmu->hrtimer_callback)
+		return;
+
+	if (bitmap_empty(tx2_pmu->active_counters, tx2_pmu->max_counters))
+		hrtimer_cancel(&tx2_pmu->hrtimer);
+}
+
+static void tx2_uncore_event_read(struct perf_event *event)
+{
+	tx2_uncore_event_update(event);
+}
+
+static enum hrtimer_restart tx2_hrtimer_callback(struct hrtimer *timer)
+{
+	struct tx2_uncore_pmu *tx2_pmu;
+	int max_counters, idx;
+
+	tx2_pmu = container_of(timer, struct tx2_uncore_pmu, hrtimer);
+	max_counters = tx2_pmu->max_counters;
+
+	if (bitmap_empty(tx2_pmu->active_counters, max_counters))
+		return HRTIMER_NORESTART;
+
+	for_each_set_bit(idx, tx2_pmu->active_counters, max_counters) {
+		struct perf_event *event = tx2_pmu->events[idx];
+
+		tx2_uncore_event_update(event);
+	}
+	hrtimer_forward_now(timer, ns_to_ktime(tx2_pmu->hrtimer_interval));
+	return HRTIMER_RESTART;
+}
+
+static int tx2_uncore_pmu_register(
+		struct tx2_uncore_pmu *tx2_pmu)
+{
+	struct device *dev = tx2_pmu->dev;
+	char *name = tx2_pmu->name;
+
+	/* Perf event registration */
+	tx2_pmu->pmu = (struct pmu) {
+		.module         = THIS_MODULE,
+		.attr_groups	= tx2_pmu->attr_groups,
+		.task_ctx_nr	= perf_invalid_context,
+		.event_init	= tx2_uncore_event_init,
+		.add		= tx2_uncore_event_add,
+		.del		= tx2_uncore_event_del,
+		.start		= tx2_uncore_event_start,
+		.stop		= tx2_uncore_event_stop,
+		.read		= tx2_uncore_event_read,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	tx2_pmu->pmu.name = devm_kasprintf(dev, GFP_KERNEL,
+			"%s", name);
+
+	return perf_pmu_register(&tx2_pmu->pmu, tx2_pmu->pmu.name, -1);
+}
+
+static int tx2_uncore_pmu_add_dev(struct tx2_uncore_pmu *tx2_pmu)
+{
+	int ret, cpu;
+
+	cpu = cpumask_any_and(cpumask_of_node(tx2_pmu->node),
+			cpu_online_mask);
+
+	tx2_pmu->cpu = cpu;
+
+	if (tx2_pmu->hrtimer_callback) {
+		hrtimer_init(&tx2_pmu->hrtimer,
+				CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+		tx2_pmu->hrtimer.function = tx2_pmu->hrtimer_callback;
+	}
+
+	ret = tx2_uncore_pmu_register(tx2_pmu);
+	if (ret) {
+		dev_err(tx2_pmu->dev, "%s PMU: Failed to init driver\n",
+				tx2_pmu->name);
+		return -ENODEV;
+	}
+
+	/* register hotplug callback for the pmu */
+	ret = cpuhp_state_add_instance(
+			CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE,
+			&tx2_pmu->hpnode);
+	if (ret) {
+		dev_err(tx2_pmu->dev, "Error %d registering hotplug", ret);
+		return ret;
+	}
+
+	/* Add to list */
+	list_add(&tx2_pmu->entry, &tx2_pmus);
+
+	dev_dbg(tx2_pmu->dev, "%s PMU UNCORE registered\n",
+			tx2_pmu->pmu.name);
+	return ret;
+}
+
+static struct tx2_uncore_pmu *tx2_uncore_pmu_init_dev(struct device *dev,
+		acpi_handle handle, struct acpi_device *adev, u32 type)
+{
+	struct tx2_uncore_pmu *tx2_pmu;
+	void __iomem *base;
+	struct resource res;
+	struct resource_entry *rentry;
+	struct list_head list;
+	int ret;
+
+	INIT_LIST_HEAD(&list);
+	ret = acpi_dev_get_resources(adev, &list, NULL, NULL);
+	if (ret <= 0) {
+		dev_err(dev, "failed to parse _CRS method, error %d\n", ret);
+		return NULL;
+	}
+
+	list_for_each_entry(rentry, &list, node) {
+		if (resource_type(rentry->res) == IORESOURCE_MEM) {
+			res = *rentry->res;
+			rentry = NULL;
+			break;
+		}
+	}
+	acpi_dev_free_resource_list(&list);
+
+	if (rentry) {
+		dev_err(dev, "PMU type %d: Fail to find resource\n", type);
+		return NULL;
+	}
+
+	base = devm_ioremap_resource(dev, &res);
+	if (IS_ERR(base)) {
+		dev_err(dev, "PMU type %d: Fail to map resource\n", type);
+		return NULL;
+	}
+
+	tx2_pmu = devm_kzalloc(dev, sizeof(*tx2_pmu), GFP_KERNEL);
+	if (!tx2_pmu)
+		return NULL;
+
+	tx2_pmu->dev = dev;
+	tx2_pmu->type = type;
+	tx2_pmu->base = base;
+	tx2_pmu->node = dev_to_node(dev);
+	INIT_LIST_HEAD(&tx2_pmu->entry);
+
+	switch (tx2_pmu->type) {
+	case PMU_TYPE_L3C:
+		tx2_pmu->max_counters = TX2_PMU_DMC_L3C_MAX_COUNTERS;
+		tx2_pmu->counters_mask = 0x3;
+		tx2_pmu->prorate_factor = TX2_PMU_L3_TILES;
+		tx2_pmu->max_events = L3_EVENT_MAX;
+		tx2_pmu->events_mask = 0x1f;
+		tx2_pmu->hrtimer_interval = TX2_PMU_HRTIMER_INTERVAL;
+		tx2_pmu->hrtimer_callback = tx2_hrtimer_callback;
+		tx2_pmu->attr_groups = l3c_pmu_attr_groups;
+		tx2_pmu->name = devm_kasprintf(dev, GFP_KERNEL,
+				"uncore_l3c_%d", tx2_pmu->node);
+		tx2_pmu->init_cntr_base = init_cntr_base_l3c;
+		tx2_pmu->start_event = uncore_start_event_l3c;
+		tx2_pmu->stop_event = uncore_stop_event_l3c;
+		break;
+	case PMU_TYPE_DMC:
+		tx2_pmu->max_counters = TX2_PMU_DMC_L3C_MAX_COUNTERS;
+		tx2_pmu->counters_mask = 0x3;
+		tx2_pmu->prorate_factor = TX2_PMU_DMC_CHANNELS;
+		tx2_pmu->max_events = DMC_EVENT_MAX;
+		tx2_pmu->events_mask = 0x1f;
+		tx2_pmu->hrtimer_interval = TX2_PMU_HRTIMER_INTERVAL;
+		tx2_pmu->hrtimer_callback = tx2_hrtimer_callback;
+		tx2_pmu->attr_groups = dmc_pmu_attr_groups;
+		tx2_pmu->name = devm_kasprintf(dev, GFP_KERNEL,
+				"uncore_dmc_%d", tx2_pmu->node);
+		tx2_pmu->init_cntr_base = init_cntr_base_dmc;
+		tx2_pmu->start_event = uncore_start_event_dmc;
+		tx2_pmu->stop_event = uncore_stop_event_dmc;
+		break;
+	case PMU_TYPE_CCPI2:
+		/* CCPI2 has 8 counters */
+		tx2_pmu->max_counters = TX2_PMU_CCPI2_MAX_COUNTERS;
+		tx2_pmu->counters_mask = 0x7;
+		tx2_pmu->prorate_factor = 1;
+		tx2_pmu->max_events = CCPI2_EVENT_MAX;
+		tx2_pmu->events_mask = 0x1ff;
+		tx2_pmu->attr_groups = ccpi2_pmu_attr_groups;
+		tx2_pmu->name = devm_kasprintf(dev, GFP_KERNEL,
+				"uncore_ccpi2_%d", tx2_pmu->node);
+		tx2_pmu->init_cntr_base = init_cntr_base_ccpi2;
+		tx2_pmu->start_event = uncore_start_event_ccpi2;
+		tx2_pmu->stop_event = uncore_stop_event_ccpi2;
+		tx2_pmu->hrtimer_callback = NULL;
+		break;
+	case PMU_TYPE_INVALID:
+		devm_kfree(dev, tx2_pmu);
+		return NULL;
+	}
+
+	return tx2_pmu;
+}
+
+static acpi_status tx2_uncore_pmu_add(acpi_handle handle, u32 level,
+				    void *data, void **return_value)
+{
+	struct tx2_uncore_pmu *tx2_pmu;
+	struct acpi_device *adev;
+	enum tx2_uncore_type type;
+
+	if (acpi_bus_get_device(handle, &adev))
+		return AE_OK;
+	if (acpi_bus_get_status(adev) || !adev->status.present)
+		return AE_OK;
+
+	type = get_tx2_pmu_type(adev);
+	if (type == PMU_TYPE_INVALID)
+		return AE_OK;
+
+	tx2_pmu = tx2_uncore_pmu_init_dev((struct device *)data,
+			handle, adev, type);
+
+	if (!tx2_pmu)
+		return AE_ERROR;
+
+	if (tx2_uncore_pmu_add_dev(tx2_pmu)) {
+		/* Can't add the PMU device, abort */
+		return AE_ERROR;
+	}
+	return AE_OK;
+}
+
+static int tx2_uncore_pmu_online_cpu(unsigned int cpu,
+		struct hlist_node *hpnode)
+{
+	struct tx2_uncore_pmu *tx2_pmu;
+
+	tx2_pmu = hlist_entry_safe(hpnode,
+			struct tx2_uncore_pmu, hpnode);
+
+	/* Pick this CPU, If there is no CPU/PMU association and both are
+	 * from same node.
+	 */
+	if ((tx2_pmu->cpu >= nr_cpu_ids) &&
+		(tx2_pmu->node == cpu_to_node(cpu)))
+		tx2_pmu->cpu = cpu;
+
+	return 0;
+}
+
+static int tx2_uncore_pmu_offline_cpu(unsigned int cpu,
+		struct hlist_node *hpnode)
+{
+	int new_cpu;
+	struct tx2_uncore_pmu *tx2_pmu;
+	struct cpumask cpu_online_mask_temp;
+
+	tx2_pmu = hlist_entry_safe(hpnode,
+			struct tx2_uncore_pmu, hpnode);
+
+	if (cpu != tx2_pmu->cpu)
+		return 0;
+
+	if (tx2_pmu->hrtimer_callback)
+		hrtimer_cancel(&tx2_pmu->hrtimer);
+
+	cpumask_copy(&cpu_online_mask_temp, cpu_online_mask);
+	cpumask_clear_cpu(cpu, &cpu_online_mask_temp);
+	new_cpu = cpumask_any_and(
+			cpumask_of_node(tx2_pmu->node),
+			&cpu_online_mask_temp);
+
+	tx2_pmu->cpu = new_cpu;
+	if (new_cpu >= nr_cpu_ids)
+		return 0;
+	perf_pmu_migrate_context(&tx2_pmu->pmu, cpu, new_cpu);
+
+	return 0;
+}
+
+static const struct acpi_device_id tx2_uncore_acpi_match[] = {
+	{"CAV901C", 0},
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, tx2_uncore_acpi_match);
+
+static int tx2_uncore_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	acpi_handle handle;
+	acpi_status status;
+
+	set_dev_node(dev, acpi_get_node(ACPI_HANDLE(dev)));
+
+	if (!has_acpi_companion(dev))
+		return -ENODEV;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -EINVAL;
+
+	/* Walk through the tree for all PMU UNCORE devices */
+	status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 1,
+				     tx2_uncore_pmu_add,
+				     NULL, dev, NULL);
+	if (ACPI_FAILURE(status)) {
+		dev_err(dev, "failed to probe PMU devices\n");
+		return_ACPI_STATUS(status);
+	}
+
+	dev_info(dev, "node%d: pmu uncore registered\n", dev_to_node(dev));
+	return 0;
+}
+
+static int tx2_uncore_remove(struct platform_device *pdev)
+{
+	struct tx2_uncore_pmu *tx2_pmu, *temp;
+	struct device *dev = &pdev->dev;
+
+	if (!list_empty(&tx2_pmus)) {
+		list_for_each_entry_safe(tx2_pmu, temp, &tx2_pmus, entry) {
+			if (tx2_pmu->node == dev_to_node(dev)) {
+				cpuhp_state_remove_instance_nocalls(
+					CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE,
+					&tx2_pmu->hpnode);
+				perf_pmu_unregister(&tx2_pmu->pmu);
+				list_del(&tx2_pmu->entry);
+			}
+		}
+	}
+	return 0;
+}
+
+static struct platform_driver tx2_uncore_driver = {
+	.driver = {
+		.name		= "tx2-uncore-pmu",
+		.acpi_match_table = ACPI_PTR(tx2_uncore_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+	.probe = tx2_uncore_probe,
+	.remove = tx2_uncore_remove,
+};
+
+static int __init tx2_uncore_driver_init(void)
+{
+	int ret;
+
+	ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE,
+				      "perf/tx2/uncore:online",
+				      tx2_uncore_pmu_online_cpu,
+				      tx2_uncore_pmu_offline_cpu);
+	if (ret) {
+		pr_err("TX2 PMU: setup hotplug failed(%d)\n", ret);
+		return ret;
+	}
+	ret = platform_driver_register(&tx2_uncore_driver);
+	if (ret)
+		cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE);
+
+	return ret;
+}
+module_init(tx2_uncore_driver_init);
+
+static void __exit tx2_uncore_driver_exit(void)
+{
+	platform_driver_unregister(&tx2_uncore_driver);
+	cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CAVIUM_TX2_UNCORE_ONLINE);
+}
+module_exit(tx2_uncore_driver_exit);
+
+MODULE_DESCRIPTION("ThunderX2 UNCORE PMU driver");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Ganapatrao Kulkarni <gkulkarni@cavium.com>");
diff --git a/drivers/perf/xgene_pmu.c b/drivers/perf/xgene_pmu.c
new file mode 100644
index 000000000..633cf07ba
--- /dev/null
+++ b/drivers/perf/xgene_pmu.c
@@ -0,0 +1,1984 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * APM X-Gene SoC PMU (Performance Monitor Unit)
+ *
+ * Copyright (c) 2016, Applied Micro Circuits Corporation
+ * Author: Hoan Tran <hotran@apm.com>
+ *         Tai Nguyen <ttnguyen@apm.com>
+ */
+
+#include <linux/acpi.h>
+#include <linux/clk.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/mfd/syscon.h>
+#include <linux/module.h>
+#include <linux/of_address.h>
+#include <linux/of_fdt.h>
+#include <linux/of_irq.h>
+#include <linux/of_platform.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/slab.h>
+
+#define CSW_CSWCR                       0x0000
+#define  CSW_CSWCR_DUALMCB_MASK         BIT(0)
+#define  CSW_CSWCR_MCB0_ROUTING(x)	(((x) & 0x0C) >> 2)
+#define  CSW_CSWCR_MCB1_ROUTING(x)	(((x) & 0x30) >> 4)
+#define MCBADDRMR                       0x0000
+#define  MCBADDRMR_DUALMCU_MODE_MASK    BIT(2)
+
+#define PCPPMU_INTSTATUS_REG	0x000
+#define PCPPMU_INTMASK_REG	0x004
+#define  PCPPMU_INTMASK		0x0000000F
+#define  PCPPMU_INTENMASK	0xFFFFFFFF
+#define  PCPPMU_INTCLRMASK	0xFFFFFFF0
+#define  PCPPMU_INT_MCU		BIT(0)
+#define  PCPPMU_INT_MCB		BIT(1)
+#define  PCPPMU_INT_L3C		BIT(2)
+#define  PCPPMU_INT_IOB		BIT(3)
+
+#define  PCPPMU_V3_INTMASK	0x00FF33FF
+#define  PCPPMU_V3_INTENMASK	0xFFFFFFFF
+#define  PCPPMU_V3_INTCLRMASK	0xFF00CC00
+#define  PCPPMU_V3_INT_MCU	0x000000FF
+#define  PCPPMU_V3_INT_MCB	0x00000300
+#define  PCPPMU_V3_INT_L3C	0x00FF0000
+#define  PCPPMU_V3_INT_IOB	0x00003000
+
+#define PMU_MAX_COUNTERS	4
+#define PMU_CNT_MAX_PERIOD	0xFFFFFFFFULL
+#define PMU_V3_CNT_MAX_PERIOD	0xFFFFFFFFFFFFFFFFULL
+#define PMU_OVERFLOW_MASK	0xF
+#define PMU_PMCR_E		BIT(0)
+#define PMU_PMCR_P		BIT(1)
+
+#define PMU_PMEVCNTR0		0x000
+#define PMU_PMEVCNTR1		0x004
+#define PMU_PMEVCNTR2		0x008
+#define PMU_PMEVCNTR3		0x00C
+#define PMU_PMEVTYPER0		0x400
+#define PMU_PMEVTYPER1		0x404
+#define PMU_PMEVTYPER2		0x408
+#define PMU_PMEVTYPER3		0x40C
+#define PMU_PMAMR0		0xA00
+#define PMU_PMAMR1		0xA04
+#define PMU_PMCNTENSET		0xC00
+#define PMU_PMCNTENCLR		0xC20
+#define PMU_PMINTENSET		0xC40
+#define PMU_PMINTENCLR		0xC60
+#define PMU_PMOVSR		0xC80
+#define PMU_PMCR		0xE04
+
+/* PMU registers for V3 */
+#define PMU_PMOVSCLR		0xC80
+#define PMU_PMOVSSET		0xCC0
+
+#define to_pmu_dev(p)     container_of(p, struct xgene_pmu_dev, pmu)
+#define GET_CNTR(ev)      (ev->hw.idx)
+#define GET_EVENTID(ev)   (ev->hw.config & 0xFFULL)
+#define GET_AGENTID(ev)   (ev->hw.config_base & 0xFFFFFFFFUL)
+#define GET_AGENT1ID(ev)  ((ev->hw.config_base >> 32) & 0xFFFFFFFFUL)
+
+struct hw_pmu_info {
+	u32 type;
+	u32 enable_mask;
+	void __iomem *csr;
+};
+
+struct xgene_pmu_dev {
+	struct hw_pmu_info *inf;
+	struct xgene_pmu *parent;
+	struct pmu pmu;
+	u8 max_counters;
+	DECLARE_BITMAP(cntr_assign_mask, PMU_MAX_COUNTERS);
+	u64 max_period;
+	const struct attribute_group **attr_groups;
+	struct perf_event *pmu_counter_event[PMU_MAX_COUNTERS];
+};
+
+struct xgene_pmu_ops {
+	void (*mask_int)(struct xgene_pmu *pmu);
+	void (*unmask_int)(struct xgene_pmu *pmu);
+	u64 (*read_counter)(struct xgene_pmu_dev *pmu, int idx);
+	void (*write_counter)(struct xgene_pmu_dev *pmu, int idx, u64 val);
+	void (*write_evttype)(struct xgene_pmu_dev *pmu_dev, int idx, u32 val);
+	void (*write_agentmsk)(struct xgene_pmu_dev *pmu_dev, u32 val);
+	void (*write_agent1msk)(struct xgene_pmu_dev *pmu_dev, u32 val);
+	void (*enable_counter)(struct xgene_pmu_dev *pmu_dev, int idx);
+	void (*disable_counter)(struct xgene_pmu_dev *pmu_dev, int idx);
+	void (*enable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx);
+	void (*disable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx);
+	void (*reset_counters)(struct xgene_pmu_dev *pmu_dev);
+	void (*start_counters)(struct xgene_pmu_dev *pmu_dev);
+	void (*stop_counters)(struct xgene_pmu_dev *pmu_dev);
+};
+
+struct xgene_pmu {
+	struct device *dev;
+	struct hlist_node node;
+	int version;
+	void __iomem *pcppmu_csr;
+	u32 mcb_active_mask;
+	u32 mc_active_mask;
+	u32 l3c_active_mask;
+	cpumask_t cpu;
+	int irq;
+	raw_spinlock_t lock;
+	const struct xgene_pmu_ops *ops;
+	struct list_head l3cpmus;
+	struct list_head iobpmus;
+	struct list_head mcbpmus;
+	struct list_head mcpmus;
+};
+
+struct xgene_pmu_dev_ctx {
+	char *name;
+	struct list_head next;
+	struct xgene_pmu_dev *pmu_dev;
+	struct hw_pmu_info inf;
+};
+
+struct xgene_pmu_data {
+	int id;
+	u32 data;
+};
+
+enum xgene_pmu_version {
+	PCP_PMU_V1 = 1,
+	PCP_PMU_V2,
+	PCP_PMU_V3,
+};
+
+enum xgene_pmu_dev_type {
+	PMU_TYPE_L3C = 0,
+	PMU_TYPE_IOB,
+	PMU_TYPE_IOB_SLOW,
+	PMU_TYPE_MCB,
+	PMU_TYPE_MC,
+};
+
+/*
+ * sysfs format attributes
+ */
+static ssize_t xgene_pmu_format_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+	return sprintf(buf, "%s\n", (char *) eattr->var);
+}
+
+#define XGENE_PMU_FORMAT_ATTR(_name, _config)		\
+	(&((struct dev_ext_attribute[]) {		\
+		{ .attr = __ATTR(_name, S_IRUGO, xgene_pmu_format_show, NULL), \
+		  .var = (void *) _config, }		\
+	})[0].attr.attr)
+
+static struct attribute *l3c_pmu_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-7"),
+	XGENE_PMU_FORMAT_ATTR(l3c_agentid, "config1:0-9"),
+	NULL,
+};
+
+static struct attribute *iob_pmu_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-7"),
+	XGENE_PMU_FORMAT_ATTR(iob_agentid, "config1:0-63"),
+	NULL,
+};
+
+static struct attribute *mcb_pmu_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-5"),
+	XGENE_PMU_FORMAT_ATTR(mcb_agentid, "config1:0-9"),
+	NULL,
+};
+
+static struct attribute *mc_pmu_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-28"),
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = l3c_pmu_format_attrs,
+};
+
+static const struct attribute_group iob_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = iob_pmu_format_attrs,
+};
+
+static const struct attribute_group mcb_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = mcb_pmu_format_attrs,
+};
+
+static const struct attribute_group mc_pmu_format_attr_group = {
+	.name = "format",
+	.attrs = mc_pmu_format_attrs,
+};
+
+static struct attribute *l3c_pmu_v3_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-39"),
+	NULL,
+};
+
+static struct attribute *iob_pmu_v3_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-47"),
+	NULL,
+};
+
+static struct attribute *iob_slow_pmu_v3_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(iob_slow_eventid, "config:0-16"),
+	NULL,
+};
+
+static struct attribute *mcb_pmu_v3_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-35"),
+	NULL,
+};
+
+static struct attribute *mc_pmu_v3_format_attrs[] = {
+	XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-44"),
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_v3_format_attr_group = {
+	.name = "format",
+	.attrs = l3c_pmu_v3_format_attrs,
+};
+
+static const struct attribute_group iob_pmu_v3_format_attr_group = {
+	.name = "format",
+	.attrs = iob_pmu_v3_format_attrs,
+};
+
+static const struct attribute_group iob_slow_pmu_v3_format_attr_group = {
+	.name = "format",
+	.attrs = iob_slow_pmu_v3_format_attrs,
+};
+
+static const struct attribute_group mcb_pmu_v3_format_attr_group = {
+	.name = "format",
+	.attrs = mcb_pmu_v3_format_attrs,
+};
+
+static const struct attribute_group mc_pmu_v3_format_attr_group = {
+	.name = "format",
+	.attrs = mc_pmu_v3_format_attrs,
+};
+
+/*
+ * sysfs event attributes
+ */
+static ssize_t xgene_pmu_event_show(struct device *dev,
+				    struct device_attribute *attr, char *buf)
+{
+	struct dev_ext_attribute *eattr;
+
+	eattr = container_of(attr, struct dev_ext_attribute, attr);
+	return sprintf(buf, "config=0x%lx\n", (unsigned long) eattr->var);
+}
+
+#define XGENE_PMU_EVENT_ATTR(_name, _config)		\
+	(&((struct dev_ext_attribute[]) {		\
+		{ .attr = __ATTR(_name, S_IRUGO, xgene_pmu_event_show, NULL), \
+		  .var = (void *) _config, }		\
+	 })[0].attr.attr)
+
+static struct attribute *l3c_pmu_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
+	XGENE_PMU_EVENT_ATTR(read-hit,				0x02),
+	XGENE_PMU_EVENT_ATTR(read-miss,				0x03),
+	XGENE_PMU_EVENT_ATTR(write-need-replacement,		0x06),
+	XGENE_PMU_EVENT_ATTR(write-not-need-replacement,	0x07),
+	XGENE_PMU_EVENT_ATTR(tq-full,				0x08),
+	XGENE_PMU_EVENT_ATTR(ackq-full,				0x09),
+	XGENE_PMU_EVENT_ATTR(wdb-full,				0x0a),
+	XGENE_PMU_EVENT_ATTR(bank-fifo-full,			0x0b),
+	XGENE_PMU_EVENT_ATTR(odb-full,				0x0c),
+	XGENE_PMU_EVENT_ATTR(wbq-full,				0x0d),
+	XGENE_PMU_EVENT_ATTR(bank-conflict-fifo-issue,		0x0e),
+	XGENE_PMU_EVENT_ATTR(bank-fifo-issue,			0x0f),
+	NULL,
+};
+
+static struct attribute *iob_pmu_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
+	XGENE_PMU_EVENT_ATTR(axi0-read,				0x02),
+	XGENE_PMU_EVENT_ATTR(axi0-read-partial,			0x03),
+	XGENE_PMU_EVENT_ATTR(axi1-read,				0x04),
+	XGENE_PMU_EVENT_ATTR(axi1-read-partial,			0x05),
+	XGENE_PMU_EVENT_ATTR(csw-read-block,			0x06),
+	XGENE_PMU_EVENT_ATTR(csw-read-partial,			0x07),
+	XGENE_PMU_EVENT_ATTR(axi0-write,			0x10),
+	XGENE_PMU_EVENT_ATTR(axi0-write-partial,		0x11),
+	XGENE_PMU_EVENT_ATTR(axi1-write,			0x13),
+	XGENE_PMU_EVENT_ATTR(axi1-write-partial,		0x14),
+	XGENE_PMU_EVENT_ATTR(csw-inbound-dirty,			0x16),
+	NULL,
+};
+
+static struct attribute *mcb_pmu_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
+	XGENE_PMU_EVENT_ATTR(csw-read,				0x02),
+	XGENE_PMU_EVENT_ATTR(csw-write-request,			0x03),
+	XGENE_PMU_EVENT_ATTR(mcb-csw-stall,			0x04),
+	XGENE_PMU_EVENT_ATTR(cancel-read-gack,			0x05),
+	NULL,
+};
+
+static struct attribute *mc_pmu_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
+	XGENE_PMU_EVENT_ATTR(act-cmd-sent,			0x02),
+	XGENE_PMU_EVENT_ATTR(pre-cmd-sent,			0x03),
+	XGENE_PMU_EVENT_ATTR(rd-cmd-sent,			0x04),
+	XGENE_PMU_EVENT_ATTR(rda-cmd-sent,			0x05),
+	XGENE_PMU_EVENT_ATTR(wr-cmd-sent,			0x06),
+	XGENE_PMU_EVENT_ATTR(wra-cmd-sent,			0x07),
+	XGENE_PMU_EVENT_ATTR(pde-cmd-sent,			0x08),
+	XGENE_PMU_EVENT_ATTR(sre-cmd-sent,			0x09),
+	XGENE_PMU_EVENT_ATTR(prea-cmd-sent,			0x0a),
+	XGENE_PMU_EVENT_ATTR(ref-cmd-sent,			0x0b),
+	XGENE_PMU_EVENT_ATTR(rd-rda-cmd-sent,			0x0c),
+	XGENE_PMU_EVENT_ATTR(wr-wra-cmd-sent,			0x0d),
+	XGENE_PMU_EVENT_ATTR(in-rd-collision,			0x0e),
+	XGENE_PMU_EVENT_ATTR(in-wr-collision,			0x0f),
+	XGENE_PMU_EVENT_ATTR(collision-queue-not-empty,		0x10),
+	XGENE_PMU_EVENT_ATTR(collision-queue-full,		0x11),
+	XGENE_PMU_EVENT_ATTR(mcu-request,			0x12),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-request,			0x13),
+	XGENE_PMU_EVENT_ATTR(mcu-hp-rd-request,			0x14),
+	XGENE_PMU_EVENT_ATTR(mcu-wr-request,			0x15),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-all,		0x16),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-cancel,		0x17),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-response,			0x18),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-all,	0x19),
+	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-cancel,	0x1a),
+	XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-all,		0x1b),
+	XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-cancel,		0x1c),
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = l3c_pmu_events_attrs,
+};
+
+static const struct attribute_group iob_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = iob_pmu_events_attrs,
+};
+
+static const struct attribute_group mcb_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = mcb_pmu_events_attrs,
+};
+
+static const struct attribute_group mc_pmu_events_attr_group = {
+	.name = "events",
+	.attrs = mc_pmu_events_attrs,
+};
+
+static struct attribute *l3c_pmu_v3_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(read-hit,				0x01),
+	XGENE_PMU_EVENT_ATTR(read-miss,				0x02),
+	XGENE_PMU_EVENT_ATTR(index-flush-eviction,		0x03),
+	XGENE_PMU_EVENT_ATTR(write-caused-replacement,		0x04),
+	XGENE_PMU_EVENT_ATTR(write-not-caused-replacement,	0x05),
+	XGENE_PMU_EVENT_ATTR(clean-eviction,			0x06),
+	XGENE_PMU_EVENT_ATTR(dirty-eviction,			0x07),
+	XGENE_PMU_EVENT_ATTR(read,				0x08),
+	XGENE_PMU_EVENT_ATTR(write,				0x09),
+	XGENE_PMU_EVENT_ATTR(request,				0x0a),
+	XGENE_PMU_EVENT_ATTR(tq-bank-conflict-issue-stall,	0x0b),
+	XGENE_PMU_EVENT_ATTR(tq-full,				0x0c),
+	XGENE_PMU_EVENT_ATTR(ackq-full,				0x0d),
+	XGENE_PMU_EVENT_ATTR(wdb-full,				0x0e),
+	XGENE_PMU_EVENT_ATTR(odb-full,				0x10),
+	XGENE_PMU_EVENT_ATTR(wbq-full,				0x11),
+	XGENE_PMU_EVENT_ATTR(input-req-async-fifo-stall,	0x12),
+	XGENE_PMU_EVENT_ATTR(output-req-async-fifo-stall,	0x13),
+	XGENE_PMU_EVENT_ATTR(output-data-async-fifo-stall,	0x14),
+	XGENE_PMU_EVENT_ATTR(total-insertion,			0x15),
+	XGENE_PMU_EVENT_ATTR(sip-insertions-r-set,		0x16),
+	XGENE_PMU_EVENT_ATTR(sip-insertions-r-clear,		0x17),
+	XGENE_PMU_EVENT_ATTR(dip-insertions-r-set,		0x18),
+	XGENE_PMU_EVENT_ATTR(dip-insertions-r-clear,		0x19),
+	XGENE_PMU_EVENT_ATTR(dip-insertions-force-r-set,	0x1a),
+	XGENE_PMU_EVENT_ATTR(egression,				0x1b),
+	XGENE_PMU_EVENT_ATTR(replacement,			0x1c),
+	XGENE_PMU_EVENT_ATTR(old-replacement,			0x1d),
+	XGENE_PMU_EVENT_ATTR(young-replacement,			0x1e),
+	XGENE_PMU_EVENT_ATTR(r-set-replacement,			0x1f),
+	XGENE_PMU_EVENT_ATTR(r-clear-replacement,		0x20),
+	XGENE_PMU_EVENT_ATTR(old-r-replacement,			0x21),
+	XGENE_PMU_EVENT_ATTR(old-nr-replacement,		0x22),
+	XGENE_PMU_EVENT_ATTR(young-r-replacement,		0x23),
+	XGENE_PMU_EVENT_ATTR(young-nr-replacement,		0x24),
+	XGENE_PMU_EVENT_ATTR(bloomfilter-clearing,		0x25),
+	XGENE_PMU_EVENT_ATTR(generation-flip,			0x26),
+	XGENE_PMU_EVENT_ATTR(vcc-droop-detected,		0x27),
+	NULL,
+};
+
+static struct attribute *iob_fast_pmu_v3_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-all,		0x01),
+	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-rd,		0x02),
+	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-wr,		0x03),
+	XGENE_PMU_EVENT_ATTR(pa-all-cp-req,			0x04),
+	XGENE_PMU_EVENT_ATTR(pa-cp-blk-req,			0x05),
+	XGENE_PMU_EVENT_ATTR(pa-cp-ptl-req,			0x06),
+	XGENE_PMU_EVENT_ATTR(pa-cp-rd-req,			0x07),
+	XGENE_PMU_EVENT_ATTR(pa-cp-wr-req,			0x08),
+	XGENE_PMU_EVENT_ATTR(ba-all-req,			0x09),
+	XGENE_PMU_EVENT_ATTR(ba-rd-req,				0x0a),
+	XGENE_PMU_EVENT_ATTR(ba-wr-req,				0x0b),
+	XGENE_PMU_EVENT_ATTR(pa-rd-shared-req-issued,		0x10),
+	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-req-issued,	0x11),
+	XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-stashable, 0x12),
+	XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-nonstashable, 0x13),
+	XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-stashable,	0x14),
+	XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-nonstashable, 0x15),
+	XGENE_PMU_EVENT_ATTR(pa-ptl-wr-req,			0x16),
+	XGENE_PMU_EVENT_ATTR(pa-ptl-rd-req,			0x17),
+	XGENE_PMU_EVENT_ATTR(pa-wr-back-clean-data,		0x18),
+	XGENE_PMU_EVENT_ATTR(pa-wr-back-cancelled-on-SS,	0x1b),
+	XGENE_PMU_EVENT_ATTR(pa-barrier-occurrence,		0x1c),
+	XGENE_PMU_EVENT_ATTR(pa-barrier-cycles,			0x1d),
+	XGENE_PMU_EVENT_ATTR(pa-total-cp-snoops,		0x20),
+	XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop,		0x21),
+	XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop-hit,		0x22),
+	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop,		0x23),
+	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop-hit,		0x24),
+	XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop,		0x25),
+	XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop-hit,	0x26),
+	XGENE_PMU_EVENT_ATTR(pa-req-buffer-full,		0x28),
+	XGENE_PMU_EVENT_ATTR(cswlf-outbound-req-fifo-full,	0x29),
+	XGENE_PMU_EVENT_ATTR(cswlf-inbound-snoop-fifo-backpressure, 0x2a),
+	XGENE_PMU_EVENT_ATTR(cswlf-outbound-lack-fifo-full,	0x2b),
+	XGENE_PMU_EVENT_ATTR(cswlf-inbound-gack-fifo-backpressure, 0x2c),
+	XGENE_PMU_EVENT_ATTR(cswlf-outbound-data-fifo-full,	0x2d),
+	XGENE_PMU_EVENT_ATTR(cswlf-inbound-data-fifo-backpressure, 0x2e),
+	XGENE_PMU_EVENT_ATTR(cswlf-inbound-req-backpressure,	0x2f),
+	NULL,
+};
+
+static struct attribute *iob_slow_pmu_v3_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(pa-axi0-rd-req,			0x01),
+	XGENE_PMU_EVENT_ATTR(pa-axi0-wr-req,			0x02),
+	XGENE_PMU_EVENT_ATTR(pa-axi1-rd-req,			0x03),
+	XGENE_PMU_EVENT_ATTR(pa-axi1-wr-req,			0x04),
+	XGENE_PMU_EVENT_ATTR(ba-all-axi-req,			0x07),
+	XGENE_PMU_EVENT_ATTR(ba-axi-rd-req,			0x08),
+	XGENE_PMU_EVENT_ATTR(ba-axi-wr-req,			0x09),
+	XGENE_PMU_EVENT_ATTR(ba-free-list-empty,		0x10),
+	NULL,
+};
+
+static struct attribute *mcb_pmu_v3_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(req-receive,			0x01),
+	XGENE_PMU_EVENT_ATTR(rd-req-recv,			0x02),
+	XGENE_PMU_EVENT_ATTR(rd-req-recv-2,			0x03),
+	XGENE_PMU_EVENT_ATTR(wr-req-recv,			0x04),
+	XGENE_PMU_EVENT_ATTR(wr-req-recv-2,			0x05),
+	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu,		0x06),
+	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu-2,		0x07),
+	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu,		0x08),
+	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu-2,		0x09),
+	XGENE_PMU_EVENT_ATTR(glbl-ack-recv-for-rd-sent-to-spec-mcu, 0x0a),
+	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-for-rd-sent-to-spec-mcu, 0x0b),
+	XGENE_PMU_EVENT_ATTR(glbl-ack-nogo-recv-for-rd-sent-to-spec-mcu, 0x0c),
+	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req,	0x0d),
+	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req-2,	0x0e),
+	XGENE_PMU_EVENT_ATTR(wr-req-sent-to-mcu,		0x0f),
+	XGENE_PMU_EVENT_ATTR(gack-recv,				0x10),
+	XGENE_PMU_EVENT_ATTR(rd-gack-recv,			0x11),
+	XGENE_PMU_EVENT_ATTR(wr-gack-recv,			0x12),
+	XGENE_PMU_EVENT_ATTR(cancel-rd-gack,			0x13),
+	XGENE_PMU_EVENT_ATTR(cancel-wr-gack,			0x14),
+	XGENE_PMU_EVENT_ATTR(mcb-csw-req-stall,			0x15),
+	XGENE_PMU_EVENT_ATTR(mcu-req-intf-blocked,		0x16),
+	XGENE_PMU_EVENT_ATTR(mcb-mcu-rd-intf-stall,		0x17),
+	XGENE_PMU_EVENT_ATTR(csw-rd-intf-blocked,		0x18),
+	XGENE_PMU_EVENT_ATTR(csw-local-ack-intf-blocked,	0x19),
+	XGENE_PMU_EVENT_ATTR(mcu-req-table-full,		0x1a),
+	XGENE_PMU_EVENT_ATTR(mcu-stat-table-full,		0x1b),
+	XGENE_PMU_EVENT_ATTR(mcu-wr-table-full,			0x1c),
+	XGENE_PMU_EVENT_ATTR(mcu-rdreceipt-resp,		0x1d),
+	XGENE_PMU_EVENT_ATTR(mcu-wrcomplete-resp,		0x1e),
+	XGENE_PMU_EVENT_ATTR(mcu-retryack-resp,			0x1f),
+	XGENE_PMU_EVENT_ATTR(mcu-pcrdgrant-resp,		0x20),
+	XGENE_PMU_EVENT_ATTR(mcu-req-from-lastload,		0x21),
+	XGENE_PMU_EVENT_ATTR(mcu-req-from-bypass,		0x22),
+	XGENE_PMU_EVENT_ATTR(volt-droop-detect,			0x23),
+	NULL,
+};
+
+static struct attribute *mc_pmu_v3_events_attrs[] = {
+	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
+	XGENE_PMU_EVENT_ATTR(act-sent,				0x01),
+	XGENE_PMU_EVENT_ATTR(pre-sent,				0x02),
+	XGENE_PMU_EVENT_ATTR(rd-sent,				0x03),
+	XGENE_PMU_EVENT_ATTR(rda-sent,				0x04),
+	XGENE_PMU_EVENT_ATTR(wr-sent,				0x05),
+	XGENE_PMU_EVENT_ATTR(wra-sent,				0x06),
+	XGENE_PMU_EVENT_ATTR(pd-entry-vld,			0x07),
+	XGENE_PMU_EVENT_ATTR(sref-entry-vld,			0x08),
+	XGENE_PMU_EVENT_ATTR(prea-sent,				0x09),
+	XGENE_PMU_EVENT_ATTR(ref-sent,				0x0a),
+	XGENE_PMU_EVENT_ATTR(rd-rda-sent,			0x0b),
+	XGENE_PMU_EVENT_ATTR(wr-wra-sent,			0x0c),
+	XGENE_PMU_EVENT_ATTR(raw-hazard,			0x0d),
+	XGENE_PMU_EVENT_ATTR(war-hazard,			0x0e),
+	XGENE_PMU_EVENT_ATTR(waw-hazard,			0x0f),
+	XGENE_PMU_EVENT_ATTR(rar-hazard,			0x10),
+	XGENE_PMU_EVENT_ATTR(raw-war-waw-hazard,		0x11),
+	XGENE_PMU_EVENT_ATTR(hprd-lprd-wr-req-vld,		0x12),
+	XGENE_PMU_EVENT_ATTR(lprd-req-vld,			0x13),
+	XGENE_PMU_EVENT_ATTR(hprd-req-vld,			0x14),
+	XGENE_PMU_EVENT_ATTR(hprd-lprd-req-vld,			0x15),
+	XGENE_PMU_EVENT_ATTR(wr-req-vld,			0x16),
+	XGENE_PMU_EVENT_ATTR(partial-wr-req-vld,		0x17),
+	XGENE_PMU_EVENT_ATTR(rd-retry,				0x18),
+	XGENE_PMU_EVENT_ATTR(wr-retry,				0x19),
+	XGENE_PMU_EVENT_ATTR(retry-gnt,				0x1a),
+	XGENE_PMU_EVENT_ATTR(rank-change,			0x1b),
+	XGENE_PMU_EVENT_ATTR(dir-change,			0x1c),
+	XGENE_PMU_EVENT_ATTR(rank-dir-change,			0x1d),
+	XGENE_PMU_EVENT_ATTR(rank-active,			0x1e),
+	XGENE_PMU_EVENT_ATTR(rank-idle,				0x1f),
+	XGENE_PMU_EVENT_ATTR(rank-pd,				0x20),
+	XGENE_PMU_EVENT_ATTR(rank-sref,				0x21),
+	XGENE_PMU_EVENT_ATTR(queue-fill-gt-thresh,		0x22),
+	XGENE_PMU_EVENT_ATTR(queue-rds-gt-thresh,		0x23),
+	XGENE_PMU_EVENT_ATTR(queue-wrs-gt-thresh,		0x24),
+	XGENE_PMU_EVENT_ATTR(phy-updt-complt,			0x25),
+	XGENE_PMU_EVENT_ATTR(tz-fail,				0x26),
+	XGENE_PMU_EVENT_ATTR(dram-errc,				0x27),
+	XGENE_PMU_EVENT_ATTR(dram-errd,				0x28),
+	XGENE_PMU_EVENT_ATTR(rd-enq,				0x29),
+	XGENE_PMU_EVENT_ATTR(wr-enq,				0x2a),
+	XGENE_PMU_EVENT_ATTR(tmac-limit-reached,		0x2b),
+	XGENE_PMU_EVENT_ATTR(tmaw-tracker-full,			0x2c),
+	NULL,
+};
+
+static const struct attribute_group l3c_pmu_v3_events_attr_group = {
+	.name = "events",
+	.attrs = l3c_pmu_v3_events_attrs,
+};
+
+static const struct attribute_group iob_fast_pmu_v3_events_attr_group = {
+	.name = "events",
+	.attrs = iob_fast_pmu_v3_events_attrs,
+};
+
+static const struct attribute_group iob_slow_pmu_v3_events_attr_group = {
+	.name = "events",
+	.attrs = iob_slow_pmu_v3_events_attrs,
+};
+
+static const struct attribute_group mcb_pmu_v3_events_attr_group = {
+	.name = "events",
+	.attrs = mcb_pmu_v3_events_attrs,
+};
+
+static const struct attribute_group mc_pmu_v3_events_attr_group = {
+	.name = "events",
+	.attrs = mc_pmu_v3_events_attrs,
+};
+
+/*
+ * sysfs cpumask attributes
+ */
+static ssize_t xgene_pmu_cpumask_show(struct device *dev,
+				      struct device_attribute *attr, char *buf)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(dev_get_drvdata(dev));
+
+	return cpumap_print_to_pagebuf(true, buf, &pmu_dev->parent->cpu);
+}
+
+static DEVICE_ATTR(cpumask, S_IRUGO, xgene_pmu_cpumask_show, NULL);
+
+static struct attribute *xgene_pmu_cpumask_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static const struct attribute_group pmu_cpumask_attr_group = {
+	.attrs = xgene_pmu_cpumask_attrs,
+};
+
+/*
+ * Per PMU device attribute groups of PMU v1 and v2
+ */
+static const struct attribute_group *l3c_pmu_attr_groups[] = {
+	&l3c_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&l3c_pmu_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *iob_pmu_attr_groups[] = {
+	&iob_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&iob_pmu_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *mcb_pmu_attr_groups[] = {
+	&mcb_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&mcb_pmu_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *mc_pmu_attr_groups[] = {
+	&mc_pmu_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&mc_pmu_events_attr_group,
+	NULL
+};
+
+/*
+ * Per PMU device attribute groups of PMU v3
+ */
+static const struct attribute_group *l3c_pmu_v3_attr_groups[] = {
+	&l3c_pmu_v3_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&l3c_pmu_v3_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *iob_fast_pmu_v3_attr_groups[] = {
+	&iob_pmu_v3_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&iob_fast_pmu_v3_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *iob_slow_pmu_v3_attr_groups[] = {
+	&iob_slow_pmu_v3_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&iob_slow_pmu_v3_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *mcb_pmu_v3_attr_groups[] = {
+	&mcb_pmu_v3_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&mcb_pmu_v3_events_attr_group,
+	NULL
+};
+
+static const struct attribute_group *mc_pmu_v3_attr_groups[] = {
+	&mc_pmu_v3_format_attr_group,
+	&pmu_cpumask_attr_group,
+	&mc_pmu_v3_events_attr_group,
+	NULL
+};
+
+static int get_next_avail_cntr(struct xgene_pmu_dev *pmu_dev)
+{
+	int cntr;
+
+	cntr = find_first_zero_bit(pmu_dev->cntr_assign_mask,
+				pmu_dev->max_counters);
+	if (cntr == pmu_dev->max_counters)
+		return -ENOSPC;
+	set_bit(cntr, pmu_dev->cntr_assign_mask);
+
+	return cntr;
+}
+
+static void clear_avail_cntr(struct xgene_pmu_dev *pmu_dev, int cntr)
+{
+	clear_bit(cntr, pmu_dev->cntr_assign_mask);
+}
+
+static inline void xgene_pmu_mask_int(struct xgene_pmu *xgene_pmu)
+{
+	writel(PCPPMU_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
+}
+
+static inline void xgene_pmu_v3_mask_int(struct xgene_pmu *xgene_pmu)
+{
+	writel(PCPPMU_V3_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
+}
+
+static inline void xgene_pmu_unmask_int(struct xgene_pmu *xgene_pmu)
+{
+	writel(PCPPMU_INTCLRMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
+}
+
+static inline void xgene_pmu_v3_unmask_int(struct xgene_pmu *xgene_pmu)
+{
+	writel(PCPPMU_V3_INTCLRMASK,
+	       xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
+}
+
+static inline u64 xgene_pmu_read_counter32(struct xgene_pmu_dev *pmu_dev,
+					   int idx)
+{
+	return readl(pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx));
+}
+
+static inline u64 xgene_pmu_read_counter64(struct xgene_pmu_dev *pmu_dev,
+					   int idx)
+{
+	u32 lo, hi;
+
+	/*
+	 * v3 has 64-bit counter registers composed by 2 32-bit registers
+	 * This can be a problem if the counter increases and carries
+	 * out of bit [31] between 2 reads. The extra reads would help
+	 * to prevent this issue.
+	 */
+	do {
+		hi = xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1);
+		lo = xgene_pmu_read_counter32(pmu_dev, 2 * idx);
+	} while (hi != xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1));
+
+	return (((u64)hi << 32) | lo);
+}
+
+static inline void
+xgene_pmu_write_counter32(struct xgene_pmu_dev *pmu_dev, int idx, u64 val)
+{
+	writel(val, pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx));
+}
+
+static inline void
+xgene_pmu_write_counter64(struct xgene_pmu_dev *pmu_dev, int idx, u64 val)
+{
+	u32 cnt_lo, cnt_hi;
+
+	cnt_hi = upper_32_bits(val);
+	cnt_lo = lower_32_bits(val);
+
+	/* v3 has 64-bit counter registers composed by 2 32-bit registers */
+	xgene_pmu_write_counter32(pmu_dev, 2 * idx, cnt_lo);
+	xgene_pmu_write_counter32(pmu_dev, 2 * idx + 1, cnt_hi);
+}
+
+static inline void
+xgene_pmu_write_evttype(struct xgene_pmu_dev *pmu_dev, int idx, u32 val)
+{
+	writel(val, pmu_dev->inf->csr + PMU_PMEVTYPER0 + (4 * idx));
+}
+
+static inline void
+xgene_pmu_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val)
+{
+	writel(val, pmu_dev->inf->csr + PMU_PMAMR0);
+}
+
+static inline void
+xgene_pmu_v3_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val) { }
+
+static inline void
+xgene_pmu_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val)
+{
+	writel(val, pmu_dev->inf->csr + PMU_PMAMR1);
+}
+
+static inline void
+xgene_pmu_v3_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val) { }
+
+static inline void
+xgene_pmu_enable_counter(struct xgene_pmu_dev *pmu_dev, int idx)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMCNTENSET);
+	val |= 1 << idx;
+	writel(val, pmu_dev->inf->csr + PMU_PMCNTENSET);
+}
+
+static inline void
+xgene_pmu_disable_counter(struct xgene_pmu_dev *pmu_dev, int idx)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMCNTENCLR);
+	val |= 1 << idx;
+	writel(val, pmu_dev->inf->csr + PMU_PMCNTENCLR);
+}
+
+static inline void
+xgene_pmu_enable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMINTENSET);
+	val |= 1 << idx;
+	writel(val, pmu_dev->inf->csr + PMU_PMINTENSET);
+}
+
+static inline void
+xgene_pmu_disable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMINTENCLR);
+	val |= 1 << idx;
+	writel(val, pmu_dev->inf->csr + PMU_PMINTENCLR);
+}
+
+static inline void xgene_pmu_reset_counters(struct xgene_pmu_dev *pmu_dev)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMCR);
+	val |= PMU_PMCR_P;
+	writel(val, pmu_dev->inf->csr + PMU_PMCR);
+}
+
+static inline void xgene_pmu_start_counters(struct xgene_pmu_dev *pmu_dev)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMCR);
+	val |= PMU_PMCR_E;
+	writel(val, pmu_dev->inf->csr + PMU_PMCR);
+}
+
+static inline void xgene_pmu_stop_counters(struct xgene_pmu_dev *pmu_dev)
+{
+	u32 val;
+
+	val = readl(pmu_dev->inf->csr + PMU_PMCR);
+	val &= ~PMU_PMCR_E;
+	writel(val, pmu_dev->inf->csr + PMU_PMCR);
+}
+
+static void xgene_perf_pmu_enable(struct pmu *pmu)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+	int enabled = bitmap_weight(pmu_dev->cntr_assign_mask,
+			pmu_dev->max_counters);
+
+	if (!enabled)
+		return;
+
+	xgene_pmu->ops->start_counters(pmu_dev);
+}
+
+static void xgene_perf_pmu_disable(struct pmu *pmu)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+
+	xgene_pmu->ops->stop_counters(pmu_dev);
+}
+
+static int xgene_perf_event_init(struct perf_event *event)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+	struct perf_event *sibling;
+
+	/* Test the event attr type check for PMU enumeration */
+	if (event->attr.type != event->pmu->type)
+		return -ENOENT;
+
+	/*
+	 * SOC PMU counters are shared across all cores.
+	 * Therefore, it does not support per-process mode.
+	 * Also, it does not support event sampling mode.
+	 */
+	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+		return -EINVAL;
+
+	if (event->cpu < 0)
+		return -EINVAL;
+	/*
+	 * Many perf core operations (eg. events rotation) operate on a
+	 * single CPU context. This is obvious for CPU PMUs, where one
+	 * expects the same sets of events being observed on all CPUs,
+	 * but can lead to issues for off-core PMUs, where each
+	 * event could be theoretically assigned to a different CPU. To
+	 * mitigate this, we enforce CPU assignment to one, selected
+	 * processor (the one described in the "cpumask" attribute).
+	 */
+	event->cpu = cpumask_first(&pmu_dev->parent->cpu);
+
+	hw->config = event->attr.config;
+	/*
+	 * Each bit of the config1 field represents an agent from which the
+	 * request of the event come. The event is counted only if it's caused
+	 * by a request of an agent has the bit cleared.
+	 * By default, the event is counted for all agents.
+	 */
+	hw->config_base = event->attr.config1;
+
+	/*
+	 * We must NOT create groups containing mixed PMUs, although software
+	 * events are acceptable
+	 */
+	if (event->group_leader->pmu != event->pmu &&
+			!is_software_event(event->group_leader))
+		return -EINVAL;
+
+	for_each_sibling_event(sibling, event->group_leader) {
+		if (sibling->pmu != event->pmu &&
+				!is_software_event(sibling))
+			return -EINVAL;
+	}
+
+	return 0;
+}
+
+static void xgene_perf_enable_event(struct perf_event *event)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+
+	xgene_pmu->ops->write_evttype(pmu_dev, GET_CNTR(event),
+				      GET_EVENTID(event));
+	xgene_pmu->ops->write_agentmsk(pmu_dev, ~((u32)GET_AGENTID(event)));
+	if (pmu_dev->inf->type == PMU_TYPE_IOB)
+		xgene_pmu->ops->write_agent1msk(pmu_dev,
+						~((u32)GET_AGENT1ID(event)));
+
+	xgene_pmu->ops->enable_counter(pmu_dev, GET_CNTR(event));
+	xgene_pmu->ops->enable_counter_int(pmu_dev, GET_CNTR(event));
+}
+
+static void xgene_perf_disable_event(struct perf_event *event)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+
+	xgene_pmu->ops->disable_counter(pmu_dev, GET_CNTR(event));
+	xgene_pmu->ops->disable_counter_int(pmu_dev, GET_CNTR(event));
+}
+
+static void xgene_perf_event_set_period(struct perf_event *event)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+	struct hw_perf_event *hw = &event->hw;
+	/*
+	 * For 32 bit counter, it has a period of 2^32. To account for the
+	 * possibility 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.
+	 * For 64 bit counter, we don't expect it overflow.
+	 */
+	u64 val = 1ULL << 31;
+
+	local64_set(&hw->prev_count, val);
+	xgene_pmu->ops->write_counter(pmu_dev, hw->idx, val);
+}
+
+static void xgene_perf_event_update(struct perf_event *event)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+	struct hw_perf_event *hw = &event->hw;
+	u64 delta, prev_raw_count, new_raw_count;
+
+again:
+	prev_raw_count = local64_read(&hw->prev_count);
+	new_raw_count = xgene_pmu->ops->read_counter(pmu_dev, GET_CNTR(event));
+
+	if (local64_cmpxchg(&hw->prev_count, prev_raw_count,
+			    new_raw_count) != prev_raw_count)
+		goto again;
+
+	delta = (new_raw_count - prev_raw_count) & pmu_dev->max_period;
+
+	local64_add(delta, &event->count);
+}
+
+static void xgene_perf_read(struct perf_event *event)
+{
+	xgene_perf_event_update(event);
+}
+
+static void xgene_perf_start(struct perf_event *event, int flags)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+	struct hw_perf_event *hw = &event->hw;
+
+	if (WARN_ON_ONCE(!(hw->state & PERF_HES_STOPPED)))
+		return;
+
+	WARN_ON_ONCE(!(hw->state & PERF_HES_UPTODATE));
+	hw->state = 0;
+
+	xgene_perf_event_set_period(event);
+
+	if (flags & PERF_EF_RELOAD) {
+		u64 prev_raw_count =  local64_read(&hw->prev_count);
+
+		xgene_pmu->ops->write_counter(pmu_dev, GET_CNTR(event),
+					      prev_raw_count);
+	}
+
+	xgene_perf_enable_event(event);
+	perf_event_update_userpage(event);
+}
+
+static void xgene_perf_stop(struct perf_event *event, int flags)
+{
+	struct hw_perf_event *hw = &event->hw;
+
+	if (hw->state & PERF_HES_UPTODATE)
+		return;
+
+	xgene_perf_disable_event(event);
+	WARN_ON_ONCE(hw->state & PERF_HES_STOPPED);
+	hw->state |= PERF_HES_STOPPED;
+
+	if (hw->state & PERF_HES_UPTODATE)
+		return;
+
+	xgene_perf_read(event);
+	hw->state |= PERF_HES_UPTODATE;
+}
+
+static int xgene_perf_add(struct perf_event *event, int flags)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+
+	hw->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	/* Allocate an event counter */
+	hw->idx = get_next_avail_cntr(pmu_dev);
+	if (hw->idx < 0)
+		return -EAGAIN;
+
+	/* Update counter event pointer for Interrupt handler */
+	pmu_dev->pmu_counter_event[hw->idx] = event;
+
+	if (flags & PERF_EF_START)
+		xgene_perf_start(event, PERF_EF_RELOAD);
+
+	return 0;
+}
+
+static void xgene_perf_del(struct perf_event *event, int flags)
+{
+	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
+	struct hw_perf_event *hw = &event->hw;
+
+	xgene_perf_stop(event, PERF_EF_UPDATE);
+
+	/* clear the assigned counter */
+	clear_avail_cntr(pmu_dev, GET_CNTR(event));
+
+	perf_event_update_userpage(event);
+	pmu_dev->pmu_counter_event[hw->idx] = NULL;
+}
+
+static int xgene_init_perf(struct xgene_pmu_dev *pmu_dev, char *name)
+{
+	struct xgene_pmu *xgene_pmu;
+
+	if (pmu_dev->parent->version == PCP_PMU_V3)
+		pmu_dev->max_period = PMU_V3_CNT_MAX_PERIOD;
+	else
+		pmu_dev->max_period = PMU_CNT_MAX_PERIOD;
+	/* First version PMU supports only single event counter */
+	xgene_pmu = pmu_dev->parent;
+	if (xgene_pmu->version == PCP_PMU_V1)
+		pmu_dev->max_counters = 1;
+	else
+		pmu_dev->max_counters = PMU_MAX_COUNTERS;
+
+	/* Perf driver registration */
+	pmu_dev->pmu = (struct pmu) {
+		.attr_groups	= pmu_dev->attr_groups,
+		.task_ctx_nr	= perf_invalid_context,
+		.pmu_enable	= xgene_perf_pmu_enable,
+		.pmu_disable	= xgene_perf_pmu_disable,
+		.event_init	= xgene_perf_event_init,
+		.add		= xgene_perf_add,
+		.del		= xgene_perf_del,
+		.start		= xgene_perf_start,
+		.stop		= xgene_perf_stop,
+		.read		= xgene_perf_read,
+		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
+	};
+
+	/* Hardware counter init */
+	xgene_pmu->ops->stop_counters(pmu_dev);
+	xgene_pmu->ops->reset_counters(pmu_dev);
+
+	return perf_pmu_register(&pmu_dev->pmu, name, -1);
+}
+
+static int
+xgene_pmu_dev_add(struct xgene_pmu *xgene_pmu, struct xgene_pmu_dev_ctx *ctx)
+{
+	struct device *dev = xgene_pmu->dev;
+	struct xgene_pmu_dev *pmu;
+
+	pmu = devm_kzalloc(dev, sizeof(*pmu), GFP_KERNEL);
+	if (!pmu)
+		return -ENOMEM;
+	pmu->parent = xgene_pmu;
+	pmu->inf = &ctx->inf;
+	ctx->pmu_dev = pmu;
+
+	switch (pmu->inf->type) {
+	case PMU_TYPE_L3C:
+		if (!(xgene_pmu->l3c_active_mask & pmu->inf->enable_mask))
+			return -ENODEV;
+		if (xgene_pmu->version == PCP_PMU_V3)
+			pmu->attr_groups = l3c_pmu_v3_attr_groups;
+		else
+			pmu->attr_groups = l3c_pmu_attr_groups;
+		break;
+	case PMU_TYPE_IOB:
+		if (xgene_pmu->version == PCP_PMU_V3)
+			pmu->attr_groups = iob_fast_pmu_v3_attr_groups;
+		else
+			pmu->attr_groups = iob_pmu_attr_groups;
+		break;
+	case PMU_TYPE_IOB_SLOW:
+		if (xgene_pmu->version == PCP_PMU_V3)
+			pmu->attr_groups = iob_slow_pmu_v3_attr_groups;
+		break;
+	case PMU_TYPE_MCB:
+		if (!(xgene_pmu->mcb_active_mask & pmu->inf->enable_mask))
+			return -ENODEV;
+		if (xgene_pmu->version == PCP_PMU_V3)
+			pmu->attr_groups = mcb_pmu_v3_attr_groups;
+		else
+			pmu->attr_groups = mcb_pmu_attr_groups;
+		break;
+	case PMU_TYPE_MC:
+		if (!(xgene_pmu->mc_active_mask & pmu->inf->enable_mask))
+			return -ENODEV;
+		if (xgene_pmu->version == PCP_PMU_V3)
+			pmu->attr_groups = mc_pmu_v3_attr_groups;
+		else
+			pmu->attr_groups = mc_pmu_attr_groups;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (xgene_init_perf(pmu, ctx->name)) {
+		dev_err(dev, "%s PMU: Failed to init perf driver\n", ctx->name);
+		return -ENODEV;
+	}
+
+	dev_info(dev, "%s PMU registered\n", ctx->name);
+
+	return 0;
+}
+
+static void _xgene_pmu_isr(int irq, struct xgene_pmu_dev *pmu_dev)
+{
+	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
+	void __iomem *csr = pmu_dev->inf->csr;
+	u32 pmovsr;
+	int idx;
+
+	xgene_pmu->ops->stop_counters(pmu_dev);
+
+	if (xgene_pmu->version == PCP_PMU_V3)
+		pmovsr = readl(csr + PMU_PMOVSSET) & PMU_OVERFLOW_MASK;
+	else
+		pmovsr = readl(csr + PMU_PMOVSR) & PMU_OVERFLOW_MASK;
+
+	if (!pmovsr)
+		goto out;
+
+	/* Clear interrupt flag */
+	if (xgene_pmu->version == PCP_PMU_V1)
+		writel(0x0, csr + PMU_PMOVSR);
+	else if (xgene_pmu->version == PCP_PMU_V2)
+		writel(pmovsr, csr + PMU_PMOVSR);
+	else
+		writel(pmovsr, csr + PMU_PMOVSCLR);
+
+	for (idx = 0; idx < PMU_MAX_COUNTERS; idx++) {
+		struct perf_event *event = pmu_dev->pmu_counter_event[idx];
+		int overflowed = pmovsr & BIT(idx);
+
+		/* Ignore if we don't have an event. */
+		if (!event || !overflowed)
+			continue;
+		xgene_perf_event_update(event);
+		xgene_perf_event_set_period(event);
+	}
+
+out:
+	xgene_pmu->ops->start_counters(pmu_dev);
+}
+
+static irqreturn_t xgene_pmu_isr(int irq, void *dev_id)
+{
+	u32 intr_mcu, intr_mcb, intr_l3c, intr_iob;
+	struct xgene_pmu_dev_ctx *ctx;
+	struct xgene_pmu *xgene_pmu = dev_id;
+	unsigned long flags;
+	u32 val;
+
+	raw_spin_lock_irqsave(&xgene_pmu->lock, flags);
+
+	/* Get Interrupt PMU source */
+	val = readl(xgene_pmu->pcppmu_csr + PCPPMU_INTSTATUS_REG);
+	if (xgene_pmu->version == PCP_PMU_V3) {
+		intr_mcu = PCPPMU_V3_INT_MCU;
+		intr_mcb = PCPPMU_V3_INT_MCB;
+		intr_l3c = PCPPMU_V3_INT_L3C;
+		intr_iob = PCPPMU_V3_INT_IOB;
+	} else {
+		intr_mcu = PCPPMU_INT_MCU;
+		intr_mcb = PCPPMU_INT_MCB;
+		intr_l3c = PCPPMU_INT_L3C;
+		intr_iob = PCPPMU_INT_IOB;
+	}
+	if (val & intr_mcu) {
+		list_for_each_entry(ctx, &xgene_pmu->mcpmus, next) {
+			_xgene_pmu_isr(irq, ctx->pmu_dev);
+		}
+	}
+	if (val & intr_mcb) {
+		list_for_each_entry(ctx, &xgene_pmu->mcbpmus, next) {
+			_xgene_pmu_isr(irq, ctx->pmu_dev);
+		}
+	}
+	if (val & intr_l3c) {
+		list_for_each_entry(ctx, &xgene_pmu->l3cpmus, next) {
+			_xgene_pmu_isr(irq, ctx->pmu_dev);
+		}
+	}
+	if (val & intr_iob) {
+		list_for_each_entry(ctx, &xgene_pmu->iobpmus, next) {
+			_xgene_pmu_isr(irq, ctx->pmu_dev);
+		}
+	}
+
+	raw_spin_unlock_irqrestore(&xgene_pmu->lock, flags);
+
+	return IRQ_HANDLED;
+}
+
+static int acpi_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
+					     struct platform_device *pdev)
+{
+	void __iomem *csw_csr, *mcba_csr, *mcbb_csr;
+	unsigned int reg;
+
+	csw_csr = devm_platform_ioremap_resource(pdev, 1);
+	if (IS_ERR(csw_csr)) {
+		dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n");
+		return PTR_ERR(csw_csr);
+	}
+
+	mcba_csr = devm_platform_ioremap_resource(pdev, 2);
+	if (IS_ERR(mcba_csr)) {
+		dev_err(&pdev->dev, "ioremap failed for MCBA CSR resource\n");
+		return PTR_ERR(mcba_csr);
+	}
+
+	mcbb_csr = devm_platform_ioremap_resource(pdev, 3);
+	if (IS_ERR(mcbb_csr)) {
+		dev_err(&pdev->dev, "ioremap failed for MCBB CSR resource\n");
+		return PTR_ERR(mcbb_csr);
+	}
+
+	xgene_pmu->l3c_active_mask = 0x1;
+
+	reg = readl(csw_csr + CSW_CSWCR);
+	if (reg & CSW_CSWCR_DUALMCB_MASK) {
+		/* Dual MCB active */
+		xgene_pmu->mcb_active_mask = 0x3;
+		/* Probe all active MC(s) */
+		reg = readl(mcbb_csr + CSW_CSWCR);
+		xgene_pmu->mc_active_mask =
+			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5;
+	} else {
+		/* Single MCB active */
+		xgene_pmu->mcb_active_mask = 0x1;
+		/* Probe all active MC(s) */
+		reg = readl(mcba_csr + CSW_CSWCR);
+		xgene_pmu->mc_active_mask =
+			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1;
+	}
+
+	return 0;
+}
+
+static int acpi_pmu_v3_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
+						struct platform_device *pdev)
+{
+	void __iomem *csw_csr;
+	unsigned int reg;
+	u32 mcb0routing;
+	u32 mcb1routing;
+
+	csw_csr = devm_platform_ioremap_resource(pdev, 1);
+	if (IS_ERR(csw_csr)) {
+		dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n");
+		return PTR_ERR(csw_csr);
+	}
+
+	reg = readl(csw_csr + CSW_CSWCR);
+	mcb0routing = CSW_CSWCR_MCB0_ROUTING(reg);
+	mcb1routing = CSW_CSWCR_MCB1_ROUTING(reg);
+	if (reg & CSW_CSWCR_DUALMCB_MASK) {
+		/* Dual MCB active */
+		xgene_pmu->mcb_active_mask = 0x3;
+		/* Probe all active L3C(s), maximum is 8 */
+		xgene_pmu->l3c_active_mask = 0xFF;
+		/* Probe all active MC(s), maximum is 8 */
+		if ((mcb0routing == 0x2) && (mcb1routing == 0x2))
+			xgene_pmu->mc_active_mask = 0xFF;
+		else if ((mcb0routing == 0x1) && (mcb1routing == 0x1))
+			xgene_pmu->mc_active_mask =  0x33;
+		else
+			xgene_pmu->mc_active_mask =  0x11;
+	} else {
+		/* Single MCB active */
+		xgene_pmu->mcb_active_mask = 0x1;
+		/* Probe all active L3C(s), maximum is 4 */
+		xgene_pmu->l3c_active_mask = 0x0F;
+		/* Probe all active MC(s), maximum is 4 */
+		if (mcb0routing == 0x2)
+			xgene_pmu->mc_active_mask = 0x0F;
+		else if (mcb0routing == 0x1)
+			xgene_pmu->mc_active_mask =  0x03;
+		else
+			xgene_pmu->mc_active_mask =  0x01;
+	}
+
+	return 0;
+}
+
+static int fdt_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
+					    struct platform_device *pdev)
+{
+	struct regmap *csw_map, *mcba_map, *mcbb_map;
+	struct device_node *np = pdev->dev.of_node;
+	unsigned int reg;
+
+	csw_map = syscon_regmap_lookup_by_phandle(np, "regmap-csw");
+	if (IS_ERR(csw_map)) {
+		dev_err(&pdev->dev, "unable to get syscon regmap csw\n");
+		return PTR_ERR(csw_map);
+	}
+
+	mcba_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcba");
+	if (IS_ERR(mcba_map)) {
+		dev_err(&pdev->dev, "unable to get syscon regmap mcba\n");
+		return PTR_ERR(mcba_map);
+	}
+
+	mcbb_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcbb");
+	if (IS_ERR(mcbb_map)) {
+		dev_err(&pdev->dev, "unable to get syscon regmap mcbb\n");
+		return PTR_ERR(mcbb_map);
+	}
+
+	xgene_pmu->l3c_active_mask = 0x1;
+	if (regmap_read(csw_map, CSW_CSWCR, &reg))
+		return -EINVAL;
+
+	if (reg & CSW_CSWCR_DUALMCB_MASK) {
+		/* Dual MCB active */
+		xgene_pmu->mcb_active_mask = 0x3;
+		/* Probe all active MC(s) */
+		if (regmap_read(mcbb_map, MCBADDRMR, &reg))
+			return 0;
+		xgene_pmu->mc_active_mask =
+			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5;
+	} else {
+		/* Single MCB active */
+		xgene_pmu->mcb_active_mask = 0x1;
+		/* Probe all active MC(s) */
+		if (regmap_read(mcba_map, MCBADDRMR, &reg))
+			return 0;
+		xgene_pmu->mc_active_mask =
+			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1;
+	}
+
+	return 0;
+}
+
+static int xgene_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
+					      struct platform_device *pdev)
+{
+	if (has_acpi_companion(&pdev->dev)) {
+		if (xgene_pmu->version == PCP_PMU_V3)
+			return acpi_pmu_v3_probe_active_mcb_mcu_l3c(xgene_pmu,
+								    pdev);
+		else
+			return acpi_pmu_probe_active_mcb_mcu_l3c(xgene_pmu,
+								 pdev);
+	}
+	return fdt_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev);
+}
+
+static char *xgene_pmu_dev_name(struct device *dev, u32 type, int id)
+{
+	switch (type) {
+	case PMU_TYPE_L3C:
+		return devm_kasprintf(dev, GFP_KERNEL, "l3c%d", id);
+	case PMU_TYPE_IOB:
+		return devm_kasprintf(dev, GFP_KERNEL, "iob%d", id);
+	case PMU_TYPE_IOB_SLOW:
+		return devm_kasprintf(dev, GFP_KERNEL, "iob_slow%d", id);
+	case PMU_TYPE_MCB:
+		return devm_kasprintf(dev, GFP_KERNEL, "mcb%d", id);
+	case PMU_TYPE_MC:
+		return devm_kasprintf(dev, GFP_KERNEL, "mc%d", id);
+	default:
+		return devm_kasprintf(dev, GFP_KERNEL, "unknown");
+	}
+}
+
+#if defined(CONFIG_ACPI)
+static struct
+xgene_pmu_dev_ctx *acpi_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu,
+				       struct acpi_device *adev, u32 type)
+{
+	struct device *dev = xgene_pmu->dev;
+	struct list_head resource_list;
+	struct xgene_pmu_dev_ctx *ctx;
+	const union acpi_object *obj;
+	struct hw_pmu_info *inf;
+	void __iomem *dev_csr;
+	struct resource res;
+	struct resource_entry *rentry;
+	int enable_bit;
+	int rc;
+
+	ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return NULL;
+
+	INIT_LIST_HEAD(&resource_list);
+	rc = acpi_dev_get_resources(adev, &resource_list, NULL, NULL);
+	if (rc <= 0) {
+		dev_err(dev, "PMU type %d: No resources found\n", type);
+		return NULL;
+	}
+
+	list_for_each_entry(rentry, &resource_list, node) {
+		if (resource_type(rentry->res) == IORESOURCE_MEM) {
+			res = *rentry->res;
+			rentry = NULL;
+			break;
+		}
+	}
+	acpi_dev_free_resource_list(&resource_list);
+
+	if (rentry) {
+		dev_err(dev, "PMU type %d: No memory resource found\n", type);
+		return NULL;
+	}
+
+	dev_csr = devm_ioremap_resource(dev, &res);
+	if (IS_ERR(dev_csr)) {
+		dev_err(dev, "PMU type %d: Fail to map resource\n", type);
+		return NULL;
+	}
+
+	/* A PMU device node without enable-bit-index is always enabled */
+	rc = acpi_dev_get_property(adev, "enable-bit-index",
+				   ACPI_TYPE_INTEGER, &obj);
+	if (rc < 0)
+		enable_bit = 0;
+	else
+		enable_bit = (int) obj->integer.value;
+
+	ctx->name = xgene_pmu_dev_name(dev, type, enable_bit);
+	if (!ctx->name) {
+		dev_err(dev, "PMU type %d: Fail to get device name\n", type);
+		return NULL;
+	}
+	inf = &ctx->inf;
+	inf->type = type;
+	inf->csr = dev_csr;
+	inf->enable_mask = 1 << enable_bit;
+
+	return ctx;
+}
+
+static const struct acpi_device_id xgene_pmu_acpi_type_match[] = {
+	{"APMC0D5D", PMU_TYPE_L3C},
+	{"APMC0D5E", PMU_TYPE_IOB},
+	{"APMC0D5F", PMU_TYPE_MCB},
+	{"APMC0D60", PMU_TYPE_MC},
+	{"APMC0D84", PMU_TYPE_L3C},
+	{"APMC0D85", PMU_TYPE_IOB},
+	{"APMC0D86", PMU_TYPE_IOB_SLOW},
+	{"APMC0D87", PMU_TYPE_MCB},
+	{"APMC0D88", PMU_TYPE_MC},
+	{},
+};
+
+static const struct acpi_device_id *xgene_pmu_acpi_match_type(
+					const struct acpi_device_id *ids,
+					struct acpi_device *adev)
+{
+	const struct acpi_device_id *match_id = NULL;
+	const struct acpi_device_id *id;
+
+	for (id = ids; id->id[0] || id->cls; id++) {
+		if (!acpi_match_device_ids(adev, id))
+			match_id = id;
+		else if (match_id)
+			break;
+	}
+
+	return match_id;
+}
+
+static acpi_status acpi_pmu_dev_add(acpi_handle handle, u32 level,
+				    void *data, void **return_value)
+{
+	const struct acpi_device_id *acpi_id;
+	struct xgene_pmu *xgene_pmu = data;
+	struct xgene_pmu_dev_ctx *ctx;
+	struct acpi_device *adev;
+
+	if (acpi_bus_get_device(handle, &adev))
+		return AE_OK;
+	if (acpi_bus_get_status(adev) || !adev->status.present)
+		return AE_OK;
+
+	acpi_id = xgene_pmu_acpi_match_type(xgene_pmu_acpi_type_match, adev);
+	if (!acpi_id)
+		return AE_OK;
+
+	ctx = acpi_get_pmu_hw_inf(xgene_pmu, adev, (u32)acpi_id->driver_data);
+	if (!ctx)
+		return AE_OK;
+
+	if (xgene_pmu_dev_add(xgene_pmu, ctx)) {
+		/* Can't add the PMU device, skip it */
+		devm_kfree(xgene_pmu->dev, ctx);
+		return AE_OK;
+	}
+
+	switch (ctx->inf.type) {
+	case PMU_TYPE_L3C:
+		list_add(&ctx->next, &xgene_pmu->l3cpmus);
+		break;
+	case PMU_TYPE_IOB:
+		list_add(&ctx->next, &xgene_pmu->iobpmus);
+		break;
+	case PMU_TYPE_IOB_SLOW:
+		list_add(&ctx->next, &xgene_pmu->iobpmus);
+		break;
+	case PMU_TYPE_MCB:
+		list_add(&ctx->next, &xgene_pmu->mcbpmus);
+		break;
+	case PMU_TYPE_MC:
+		list_add(&ctx->next, &xgene_pmu->mcpmus);
+		break;
+	}
+	return AE_OK;
+}
+
+static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
+				  struct platform_device *pdev)
+{
+	struct device *dev = xgene_pmu->dev;
+	acpi_handle handle;
+	acpi_status status;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -EINVAL;
+
+	status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 1,
+				     acpi_pmu_dev_add, NULL, xgene_pmu, NULL);
+	if (ACPI_FAILURE(status)) {
+		dev_err(dev, "failed to probe PMU devices\n");
+		return -ENODEV;
+	}
+
+	return 0;
+}
+#else
+static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
+				  struct platform_device *pdev)
+{
+	return 0;
+}
+#endif
+
+static struct
+xgene_pmu_dev_ctx *fdt_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu,
+				      struct device_node *np, u32 type)
+{
+	struct device *dev = xgene_pmu->dev;
+	struct xgene_pmu_dev_ctx *ctx;
+	struct hw_pmu_info *inf;
+	void __iomem *dev_csr;
+	struct resource res;
+	int enable_bit;
+
+	ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return NULL;
+
+	if (of_address_to_resource(np, 0, &res) < 0) {
+		dev_err(dev, "PMU type %d: No resource address found\n", type);
+		return NULL;
+	}
+
+	dev_csr = devm_ioremap_resource(dev, &res);
+	if (IS_ERR(dev_csr)) {
+		dev_err(dev, "PMU type %d: Fail to map resource\n", type);
+		return NULL;
+	}
+
+	/* A PMU device node without enable-bit-index is always enabled */
+	if (of_property_read_u32(np, "enable-bit-index", &enable_bit))
+		enable_bit = 0;
+
+	ctx->name = xgene_pmu_dev_name(dev, type, enable_bit);
+	if (!ctx->name) {
+		dev_err(dev, "PMU type %d: Fail to get device name\n", type);
+		return NULL;
+	}
+
+	inf = &ctx->inf;
+	inf->type = type;
+	inf->csr = dev_csr;
+	inf->enable_mask = 1 << enable_bit;
+
+	return ctx;
+}
+
+static int fdt_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
+				 struct platform_device *pdev)
+{
+	struct xgene_pmu_dev_ctx *ctx;
+	struct device_node *np;
+
+	for_each_child_of_node(pdev->dev.of_node, np) {
+		if (!of_device_is_available(np))
+			continue;
+
+		if (of_device_is_compatible(np, "apm,xgene-pmu-l3c"))
+			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_L3C);
+		else if (of_device_is_compatible(np, "apm,xgene-pmu-iob"))
+			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_IOB);
+		else if (of_device_is_compatible(np, "apm,xgene-pmu-mcb"))
+			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MCB);
+		else if (of_device_is_compatible(np, "apm,xgene-pmu-mc"))
+			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MC);
+		else
+			ctx = NULL;
+
+		if (!ctx)
+			continue;
+
+		if (xgene_pmu_dev_add(xgene_pmu, ctx)) {
+			/* Can't add the PMU device, skip it */
+			devm_kfree(xgene_pmu->dev, ctx);
+			continue;
+		}
+
+		switch (ctx->inf.type) {
+		case PMU_TYPE_L3C:
+			list_add(&ctx->next, &xgene_pmu->l3cpmus);
+			break;
+		case PMU_TYPE_IOB:
+			list_add(&ctx->next, &xgene_pmu->iobpmus);
+			break;
+		case PMU_TYPE_IOB_SLOW:
+			list_add(&ctx->next, &xgene_pmu->iobpmus);
+			break;
+		case PMU_TYPE_MCB:
+			list_add(&ctx->next, &xgene_pmu->mcbpmus);
+			break;
+		case PMU_TYPE_MC:
+			list_add(&ctx->next, &xgene_pmu->mcpmus);
+			break;
+		}
+	}
+
+	return 0;
+}
+
+static int xgene_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
+				   struct platform_device *pdev)
+{
+	if (has_acpi_companion(&pdev->dev))
+		return acpi_pmu_probe_pmu_dev(xgene_pmu, pdev);
+	return fdt_pmu_probe_pmu_dev(xgene_pmu, pdev);
+}
+
+static const struct xgene_pmu_data xgene_pmu_data = {
+	.id   = PCP_PMU_V1,
+};
+
+static const struct xgene_pmu_data xgene_pmu_v2_data = {
+	.id   = PCP_PMU_V2,
+};
+
+static const struct xgene_pmu_ops xgene_pmu_ops = {
+	.mask_int = xgene_pmu_mask_int,
+	.unmask_int = xgene_pmu_unmask_int,
+	.read_counter = xgene_pmu_read_counter32,
+	.write_counter = xgene_pmu_write_counter32,
+	.write_evttype = xgene_pmu_write_evttype,
+	.write_agentmsk = xgene_pmu_write_agentmsk,
+	.write_agent1msk = xgene_pmu_write_agent1msk,
+	.enable_counter = xgene_pmu_enable_counter,
+	.disable_counter = xgene_pmu_disable_counter,
+	.enable_counter_int = xgene_pmu_enable_counter_int,
+	.disable_counter_int = xgene_pmu_disable_counter_int,
+	.reset_counters = xgene_pmu_reset_counters,
+	.start_counters = xgene_pmu_start_counters,
+	.stop_counters = xgene_pmu_stop_counters,
+};
+
+static const struct xgene_pmu_ops xgene_pmu_v3_ops = {
+	.mask_int = xgene_pmu_v3_mask_int,
+	.unmask_int = xgene_pmu_v3_unmask_int,
+	.read_counter = xgene_pmu_read_counter64,
+	.write_counter = xgene_pmu_write_counter64,
+	.write_evttype = xgene_pmu_write_evttype,
+	.write_agentmsk = xgene_pmu_v3_write_agentmsk,
+	.write_agent1msk = xgene_pmu_v3_write_agent1msk,
+	.enable_counter = xgene_pmu_enable_counter,
+	.disable_counter = xgene_pmu_disable_counter,
+	.enable_counter_int = xgene_pmu_enable_counter_int,
+	.disable_counter_int = xgene_pmu_disable_counter_int,
+	.reset_counters = xgene_pmu_reset_counters,
+	.start_counters = xgene_pmu_start_counters,
+	.stop_counters = xgene_pmu_stop_counters,
+};
+
+static const struct of_device_id xgene_pmu_of_match[] = {
+	{ .compatible	= "apm,xgene-pmu",	.data = &xgene_pmu_data },
+	{ .compatible	= "apm,xgene-pmu-v2",	.data = &xgene_pmu_v2_data },
+	{},
+};
+MODULE_DEVICE_TABLE(of, xgene_pmu_of_match);
+#ifdef CONFIG_ACPI
+static const struct acpi_device_id xgene_pmu_acpi_match[] = {
+	{"APMC0D5B", PCP_PMU_V1},
+	{"APMC0D5C", PCP_PMU_V2},
+	{"APMC0D83", PCP_PMU_V3},
+	{},
+};
+MODULE_DEVICE_TABLE(acpi, xgene_pmu_acpi_match);
+#endif
+
+static int xgene_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct xgene_pmu *xgene_pmu = hlist_entry_safe(node, struct xgene_pmu,
+						       node);
+
+	if (cpumask_empty(&xgene_pmu->cpu))
+		cpumask_set_cpu(cpu, &xgene_pmu->cpu);
+
+	/* Overflow interrupt also should use the same CPU */
+	WARN_ON(irq_set_affinity(xgene_pmu->irq, &xgene_pmu->cpu));
+
+	return 0;
+}
+
+static int xgene_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
+{
+	struct xgene_pmu *xgene_pmu = hlist_entry_safe(node, struct xgene_pmu,
+						       node);
+	struct xgene_pmu_dev_ctx *ctx;
+	unsigned int target;
+
+	if (!cpumask_test_and_clear_cpu(cpu, &xgene_pmu->cpu))
+		return 0;
+	target = cpumask_any_but(cpu_online_mask, cpu);
+	if (target >= nr_cpu_ids)
+		return 0;
+
+	list_for_each_entry(ctx, &xgene_pmu->mcpmus, next) {
+		perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target);
+	}
+	list_for_each_entry(ctx, &xgene_pmu->mcbpmus, next) {
+		perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target);
+	}
+	list_for_each_entry(ctx, &xgene_pmu->l3cpmus, next) {
+		perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target);
+	}
+	list_for_each_entry(ctx, &xgene_pmu->iobpmus, next) {
+		perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target);
+	}
+
+	cpumask_set_cpu(target, &xgene_pmu->cpu);
+	/* Overflow interrupt also should use the same CPU */
+	WARN_ON(irq_set_affinity(xgene_pmu->irq, &xgene_pmu->cpu));
+
+	return 0;
+}
+
+static int xgene_pmu_probe(struct platform_device *pdev)
+{
+	const struct xgene_pmu_data *dev_data;
+	const struct of_device_id *of_id;
+	struct xgene_pmu *xgene_pmu;
+	struct resource *res;
+	int irq, rc;
+	int version;
+
+	/* Install a hook to update the reader CPU in case it goes offline */
+	rc = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE,
+				      "CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE",
+				      xgene_pmu_online_cpu,
+				      xgene_pmu_offline_cpu);
+	if (rc)
+		return rc;
+
+	xgene_pmu = devm_kzalloc(&pdev->dev, sizeof(*xgene_pmu), GFP_KERNEL);
+	if (!xgene_pmu)
+		return -ENOMEM;
+	xgene_pmu->dev = &pdev->dev;
+	platform_set_drvdata(pdev, xgene_pmu);
+
+	version = -EINVAL;
+	of_id = of_match_device(xgene_pmu_of_match, &pdev->dev);
+	if (of_id) {
+		dev_data = (const struct xgene_pmu_data *) of_id->data;
+		version = dev_data->id;
+	}
+
+#ifdef CONFIG_ACPI
+	if (ACPI_COMPANION(&pdev->dev)) {
+		const struct acpi_device_id *acpi_id;
+
+		acpi_id = acpi_match_device(xgene_pmu_acpi_match, &pdev->dev);
+		if (acpi_id)
+			version = (int) acpi_id->driver_data;
+	}
+#endif
+	if (version < 0)
+		return -ENODEV;
+
+	if (version == PCP_PMU_V3)
+		xgene_pmu->ops = &xgene_pmu_v3_ops;
+	else
+		xgene_pmu->ops = &xgene_pmu_ops;
+
+	INIT_LIST_HEAD(&xgene_pmu->l3cpmus);
+	INIT_LIST_HEAD(&xgene_pmu->iobpmus);
+	INIT_LIST_HEAD(&xgene_pmu->mcbpmus);
+	INIT_LIST_HEAD(&xgene_pmu->mcpmus);
+
+	xgene_pmu->version = version;
+	dev_info(&pdev->dev, "X-Gene PMU version %d\n", xgene_pmu->version);
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	xgene_pmu->pcppmu_csr = devm_ioremap_resource(&pdev->dev, res);
+	if (IS_ERR(xgene_pmu->pcppmu_csr)) {
+		dev_err(&pdev->dev, "ioremap failed for PCP PMU resource\n");
+		return PTR_ERR(xgene_pmu->pcppmu_csr);
+	}
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return -EINVAL;
+
+	rc = devm_request_irq(&pdev->dev, irq, xgene_pmu_isr,
+				IRQF_NOBALANCING | IRQF_NO_THREAD,
+				dev_name(&pdev->dev), xgene_pmu);
+	if (rc) {
+		dev_err(&pdev->dev, "Could not request IRQ %d\n", irq);
+		return rc;
+	}
+
+	xgene_pmu->irq = irq;
+
+	raw_spin_lock_init(&xgene_pmu->lock);
+
+	/* Check for active MCBs and MCUs */
+	rc = xgene_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev);
+	if (rc) {
+		dev_warn(&pdev->dev, "Unknown MCB/MCU active status\n");
+		xgene_pmu->mcb_active_mask = 0x1;
+		xgene_pmu->mc_active_mask = 0x1;
+	}
+
+	/* Add this instance to the list used by the hotplug callback */
+	rc = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE,
+				      &xgene_pmu->node);
+	if (rc) {
+		dev_err(&pdev->dev, "Error %d registering hotplug", rc);
+		return rc;
+	}
+
+	/* Walk through the tree for all PMU perf devices */
+	rc = xgene_pmu_probe_pmu_dev(xgene_pmu, pdev);
+	if (rc) {
+		dev_err(&pdev->dev, "No PMU perf devices found!\n");
+		goto out_unregister;
+	}
+
+	/* Enable interrupt */
+	xgene_pmu->ops->unmask_int(xgene_pmu);
+
+	return 0;
+
+out_unregister:
+	cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE,
+				    &xgene_pmu->node);
+	return rc;
+}
+
+static void
+xgene_pmu_dev_cleanup(struct xgene_pmu *xgene_pmu, struct list_head *pmus)
+{
+	struct xgene_pmu_dev_ctx *ctx;
+
+	list_for_each_entry(ctx, pmus, next) {
+		perf_pmu_unregister(&ctx->pmu_dev->pmu);
+	}
+}
+
+static int xgene_pmu_remove(struct platform_device *pdev)
+{
+	struct xgene_pmu *xgene_pmu = dev_get_drvdata(&pdev->dev);
+
+	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->l3cpmus);
+	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->iobpmus);
+	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcbpmus);
+	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcpmus);
+	cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE,
+				    &xgene_pmu->node);
+
+	return 0;
+}
+
+static struct platform_driver xgene_pmu_driver = {
+	.probe = xgene_pmu_probe,
+	.remove = xgene_pmu_remove,
+	.driver = {
+		.name		= "xgene-pmu",
+		.of_match_table = xgene_pmu_of_match,
+		.acpi_match_table = ACPI_PTR(xgene_pmu_acpi_match),
+		.suppress_bind_attrs = true,
+	},
+};
+
+builtin_platform_driver(xgene_pmu_driver);