These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / bus / arm-cci.c
index 5340604..577cc4b 100644 (file)
@@ -51,13 +51,16 @@ static const struct cci_nb_ports cci400_ports = {
 static const struct of_device_id arm_cci_matches[] = {
 #ifdef CONFIG_ARM_CCI400_COMMON
        {.compatible = "arm,cci-400", .data = CCI400_PORTS_DATA },
+#endif
+#ifdef CONFIG_ARM_CCI500_PMU
+       { .compatible = "arm,cci-500", },
 #endif
        {},
 };
 
-#ifdef CONFIG_ARM_CCI400_PMU
+#ifdef CONFIG_ARM_CCI_PMU
 
-#define DRIVER_NAME            "CCI-400"
+#define DRIVER_NAME            "ARM-CCI"
 #define DRIVER_NAME_PMU                DRIVER_NAME " PMU"
 
 #define CCI_PMCR               0x0100
@@ -77,20 +80,21 @@ static const struct of_device_id arm_cci_matches[] = {
 
 #define CCI_PMU_OVRFLW_FLAG    1
 
-#define CCI_PMU_CNTR_BASE(idx) ((idx) * SZ_4K)
-
-#define CCI_PMU_CNTR_MASK      ((1ULL << 32) -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_EVENT_MASK             0xffUL
-#define CCI_PMU_EVENT_SOURCE(event)    ((event >> 5) & 0x7)
-#define CCI_PMU_EVENT_CODE(event)      (event & 0x1f)
-
-#define CCI_PMU_MAX_HW_EVENTS 5   /* CCI PMU has 4 counters + 1 cycle counter */
+#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_CCI500_PMU
+       CCI_IF_GLOBAL,
+#endif
        CCI_IF_MAX,
 };
 
@@ -100,14 +104,30 @@ struct event_range {
 };
 
 struct cci_pmu_hw_events {
-       struct perf_event *events[CCI_PMU_MAX_HW_EVENTS];
-       unsigned long used_mask[BITS_TO_LONGS(CCI_PMU_MAX_HW_EVENTS)];
+       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;
+       u64 nformat_attrs;
+       u64 nevent_attrs;
+       struct dev_ext_attribute *format_attrs;
+       struct dev_ext_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);
 };
 
 static struct cci_pmu_model cci_pmu_models[];
@@ -116,33 +136,59 @@ struct cci_pmu {
        void __iomem *base;
        struct pmu pmu;
        int nr_irqs;
-       int irqs[CCI_PMU_MAX_HW_EVENTS];
+       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_events;
+       int num_cntrs;
        atomic_t active_events;
        struct mutex reserve_mutex;
+       struct notifier_block cpu_nb;
        cpumask_t cpus;
 };
-static struct cci_pmu *pmu;
 
 #define to_cci_pmu(c)  (container_of(c, struct cci_pmu, pmu))
 
+enum cci_models {
+#ifdef CONFIG_ARM_CCI400_PMU
+       CCI400_R0,
+       CCI400_R1,
+#endif
+#ifdef CONFIG_ARM_CCI500_PMU
+       CCI500_R0,
+#endif
+       CCI_MODEL_MAX
+};
+
+static ssize_t cci_pmu_format_show(struct device *dev,
+                       struct device_attribute *attr, char *buf);
+static ssize_t cci_pmu_event_show(struct device *dev,
+                       struct device_attribute *attr, char *buf);
+
+#define CCI_EXT_ATTR_ENTRY(_name, _func, _config) \
+       { __ATTR(_name, S_IRUGO, _func, NULL), (void *)_config }
+
+#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 CCI_PORT_S0    0
-#define CCI_PORT_S1    1
-#define CCI_PORT_S2    2
-#define CCI_PORT_S3    3
-#define CCI_PORT_S4    4
-#define CCI_PORT_M0    5
-#define CCI_PORT_M1    6
-#define CCI_PORT_M2    7
-
-#define CCI_REV_R0             0
-#define CCI_REV_R1             1
-#define CCI_REV_R1_PX          5
+#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
@@ -150,12 +196,11 @@ static struct cci_pmu *pmu;
  * make use of this event in hardware.
  */
 enum cci400_perf_events {
-       CCI_PMU_CYCLES = 0xff
+       CCI400_PMU_CYCLES = 0xff
 };
 
-#define CCI_PMU_CYCLE_CNTR_IDX         0
-#define CCI_PMU_CNTR0_IDX              1
-#define CCI_PMU_CNTR_LAST(cci_pmu)     (CCI_PMU_CYCLE_CNTR_IDX + cci_pmu->num_events - 1)
+#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
@@ -169,37 +214,173 @@ enum cci400_perf_events {
  * the different revisions and are used to validate the event to be monitored.
  */
 
-#define CCI_REV_R0_SLAVE_PORT_MIN_EV   0x00
-#define CCI_REV_R0_SLAVE_PORT_MAX_EV   0x13
-#define CCI_REV_R0_MASTER_PORT_MIN_EV  0x14
-#define CCI_REV_R0_MASTER_PORT_MAX_EV  0x1a
+#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 dev_ext_attribute cci400_pmu_format_attrs[] = {
+       CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"),
+       CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-7"),
+};
+
+static struct dev_ext_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),
+};
+
+static struct dev_ext_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),
+};
 
-#define CCI_REV_R1_SLAVE_PORT_MIN_EV   0x00
-#define CCI_REV_R1_SLAVE_PORT_MAX_EV   0x14
-#define CCI_REV_R1_MASTER_PORT_MIN_EV  0x00
-#define CCI_REV_R1_MASTER_PORT_MAX_EV  0x11
+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 pmu_validate_hw_event(unsigned long hw_event)
+static int cci400_get_event_idx(struct cci_pmu *cci_pmu,
+                               struct cci_pmu_hw_events *hw,
+                               unsigned long cci_event)
 {
-       u8 ev_source = CCI_PMU_EVENT_SOURCE(hw_event);
-       u8 ev_code = CCI_PMU_EVENT_CODE(hw_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 & ~CCI_PMU_EVENT_MASK)
+       if (hw_event & ~CCI400_PMU_EVENT_MASK)
                return -ENOENT;
 
+       if (hw_event == CCI400_PMU_CYCLES)
+               return hw_event;
+
        switch (ev_source) {
-       case CCI_PORT_S0:
-       case CCI_PORT_S1:
-       case CCI_PORT_S2:
-       case CCI_PORT_S3:
-       case CCI_PORT_S4:
+       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 CCI_PORT_M0:
-       case CCI_PORT_M1:
-       case CCI_PORT_M2:
+       case CCI400_PORT_M0:
+       case CCI400_PORT_M1:
+       case CCI400_PORT_M2:
                /* Master Interface */
                if_type = CCI_IF_MASTER;
                break;
@@ -207,87 +388,291 @@ static int pmu_validate_hw_event(unsigned long hw_event)
                return -ENOENT;
        }
 
-       if (ev_code >= pmu->model->event_ranges[if_type].min &&
-               ev_code <= pmu->model->event_ranges[if_type].max)
+       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_cci_revision(void)
+static int probe_cci400_revision(void)
 {
        int rev;
        rev = readl_relaxed(cci_ctrl_base + CCI_PID2) & CCI_PID2_REV_MASK;
        rev >>= CCI_PID2_REV_SHIFT;
 
-       if (rev < CCI_REV_R1_PX)
-               return CCI_REV_R0;
+       if (rev < CCI400_R1_PX)
+               return CCI400_R0;
        else
-               return CCI_REV_R1;
+               return CCI400_R1;
 }
 
 static const struct cci_pmu_model *probe_cci_model(struct platform_device *pdev)
 {
        if (platform_has_secure_cci_access())
-               return &cci_pmu_models[probe_cci_revision()];
+               return &cci_pmu_models[probe_cci400_revision()];
        return NULL;
 }
+#else  /* !CONFIG_ARM_CCI400_PMU */
+static inline struct cci_pmu_model *probe_cci_model(struct platform_device *pdev)
+{
+       return NULL;
+}
+#endif /* CONFIG_ARM_CCI400_PMU */
+
+#ifdef CONFIG_ARM_CCI500_PMU
+
+/*
+ * CCI500 provides 8 independent event counters that can count
+ * any of the events available.
+ *
+ * CCI500 PMU event id is an 9-bit value made of two parts.
+ *      bits [8:5] - Source for the event
+ *                   0x0-0x6 - Slave interfaces
+ *                   0x8-0xD - Master interfaces
+ *                   0xf     - Global Events
+ *                   0x7,0xe - Reserved
+ *
+ *      bits [4:0] - Event code (specific to type of interface)
+ */
+
+/* Port ids */
+#define CCI500_PORT_S0                 0x0
+#define CCI500_PORT_S1                 0x1
+#define CCI500_PORT_S2                 0x2
+#define CCI500_PORT_S3                 0x3
+#define CCI500_PORT_S4                 0x4
+#define CCI500_PORT_S5                 0x5
+#define CCI500_PORT_S6                 0x6
+
+#define CCI500_PORT_M0                 0x8
+#define CCI500_PORT_M1                 0x9
+#define CCI500_PORT_M2                 0xa
+#define CCI500_PORT_M3                 0xb
+#define CCI500_PORT_M4                 0xc
+#define CCI500_PORT_M5                 0xd
+
+#define CCI500_PORT_GLOBAL             0xf
+
+#define CCI500_PMU_EVENT_MASK          0x1ffUL
+#define CCI500_PMU_EVENT_SOURCE_SHIFT  0x5
+#define CCI500_PMU_EVENT_SOURCE_MASK   0xf
+#define CCI500_PMU_EVENT_CODE_SHIFT    0x0
+#define CCI500_PMU_EVENT_CODE_MASK     0x1f
+
+#define CCI500_PMU_EVENT_SOURCE(event) \
+       ((event >> CCI500_PMU_EVENT_SOURCE_SHIFT) & CCI500_PMU_EVENT_SOURCE_MASK)
+#define CCI500_PMU_EVENT_CODE(event)   \
+       ((event >> CCI500_PMU_EVENT_CODE_SHIFT) & CCI500_PMU_EVENT_CODE_MASK)
+
+#define CCI500_SLAVE_PORT_MIN_EV       0x00
+#define CCI500_SLAVE_PORT_MAX_EV       0x1f
+#define CCI500_MASTER_PORT_MIN_EV      0x00
+#define CCI500_MASTER_PORT_MAX_EV      0x06
+#define CCI500_GLOBAL_PORT_MIN_EV      0x00
+#define CCI500_GLOBAL_PORT_MAX_EV      0x0f
+
+
+#define CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(_name, _config) \
+       CCI_EXT_ATTR_ENTRY(_name, cci500_pmu_global_event_show, \
+                                       (unsigned long) _config)
+
+static ssize_t cci500_pmu_global_event_show(struct device *dev,
+                               struct device_attribute *attr, char *buf);
+
+static struct dev_ext_attribute cci500_pmu_format_attrs[] = {
+       CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"),
+       CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-8"),
+};
+
+static struct dev_ext_attribute cci500_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 */
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_0_1, 0x0),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_2_3, 0x1),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_4_5, 0x2),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_6_7, 0x3),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_0_1, 0x4),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_2_3, 0x5),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_4_5, 0x6),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_6_7, 0x7),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_back_invalidation, 0x8),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_alloc_busy, 0x9),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_tt_full, 0xA),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_wrq, 0xB),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_cd_hs, 0xC),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_rq_stall_addr_hazard, 0xD),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snopp_rq_stall_tt_full, 0xE),
+       CCI500_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_rq_tzmp1_prot, 0xF),
+};
+
+static ssize_t cci500_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, CCI500_PORT_GLOBAL);
+}
+
+static int cci500_validate_hw_event(struct cci_pmu *cci_pmu,
+                                       unsigned long hw_event)
+{
+       u32 ev_source = CCI500_PMU_EVENT_SOURCE(hw_event);
+       u32 ev_code = CCI500_PMU_EVENT_CODE(hw_event);
+       int if_type;
+
+       if (hw_event & ~CCI500_PMU_EVENT_MASK)
+               return -ENOENT;
+
+       switch (ev_source) {
+       case CCI500_PORT_S0:
+       case CCI500_PORT_S1:
+       case CCI500_PORT_S2:
+       case CCI500_PORT_S3:
+       case CCI500_PORT_S4:
+       case CCI500_PORT_S5:
+       case CCI500_PORT_S6:
+               if_type = CCI_IF_SLAVE;
+               break;
+       case CCI500_PORT_M0:
+       case CCI500_PORT_M1:
+       case CCI500_PORT_M2:
+       case CCI500_PORT_M3:
+       case CCI500_PORT_M4:
+       case CCI500_PORT_M5:
+               if_type = CCI_IF_MASTER;
+               break;
+       case CCI500_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_CCI500_PMU */
+
+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 CCI_PMU_CYCLE_CNTR_IDX <= idx &&
-               idx <= CCI_PMU_CNTR_LAST(cci_pmu);
+       return 0 <= idx && idx <= CCI_PMU_CNTR_LAST(cci_pmu);
 }
 
-static u32 pmu_read_register(int idx, unsigned int offset)
+static u32 pmu_read_register(struct cci_pmu *cci_pmu, int idx, unsigned int offset)
 {
-       return readl_relaxed(pmu->base + CCI_PMU_CNTR_BASE(idx) + offset);
+       return readl_relaxed(cci_pmu->base +
+                            CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
 }
 
-static void pmu_write_register(u32 value, int idx, unsigned int offset)
+static void pmu_write_register(struct cci_pmu *cci_pmu, u32 value,
+                              int idx, unsigned int offset)
 {
-       return writel_relaxed(value, pmu->base + CCI_PMU_CNTR_BASE(idx) + offset);
+       return writel_relaxed(value, cci_pmu->base +
+                             CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
 }
 
-static void pmu_disable_counter(int idx)
+static void pmu_disable_counter(struct cci_pmu *cci_pmu, int idx)
 {
-       pmu_write_register(0, idx, CCI_PMU_CNTR_CTRL);
+       pmu_write_register(cci_pmu, 0, idx, CCI_PMU_CNTR_CTRL);
 }
 
-static void pmu_enable_counter(int idx)
+static void pmu_enable_counter(struct cci_pmu *cci_pmu, int idx)
 {
-       pmu_write_register(1, idx, CCI_PMU_CNTR_CTRL);
+       pmu_write_register(cci_pmu, 1, idx, CCI_PMU_CNTR_CTRL);
 }
 
-static void pmu_set_event(int idx, unsigned long event)
+static void pmu_set_event(struct cci_pmu *cci_pmu, int idx, unsigned long event)
 {
-       pmu_write_register(event, idx, CCI_PMU_EVT_SEL);
+       pmu_write_register(cci_pmu, event, idx, CCI_PMU_EVT_SEL);
 }
 
+/*
+ * Returns the number of programmable counters actually implemented
+ * by the cci
+ */
 static u32 pmu_get_max_counters(void)
 {
-       u32 n_cnts = (readl_relaxed(cci_ctrl_base + CCI_PMCR) &
-                     CCI_PMCR_NCNT_MASK) >> CCI_PMCR_NCNT_SHIFT;
-
-       /* add 1 for cycle counter */
-       return n_cnts + 1;
+       return (readl_relaxed(cci_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);
-       struct hw_perf_event *hw_event = &event->hw;
-       unsigned long cci_event = hw_event->config_base;
+       unsigned long cci_event = event->hw.config_base;
        int idx;
 
-       if (cci_event == CCI_PMU_CYCLES) {
-               if (test_and_set_bit(CCI_PMU_CYCLE_CNTR_IDX, hw->used_mask))
-                       return -EAGAIN;
+       if (cci_pmu->model->get_event_idx)
+               return cci_pmu->model->get_event_idx(cci_pmu, hw, cci_event);
 
-               return CCI_PMU_CYCLE_CNTR_IDX;
-       }
-
-       for (idx = CCI_PMU_CNTR0_IDX; idx <= CCI_PMU_CNTR_LAST(cci_pmu); ++idx)
+       /* 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;
 
@@ -297,18 +682,13 @@ static int pmu_get_event_idx(struct cci_pmu_hw_events *hw, struct perf_event *ev
 
 static int pmu_map_event(struct perf_event *event)
 {
-       int mapping;
-       unsigned long config = event->attr.config;
+       struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
 
-       if (event->attr.type < PERF_TYPE_MAX)
+       if (event->attr.type < PERF_TYPE_MAX ||
+                       !cci_pmu->model->validate_hw_event)
                return -ENOENT;
 
-       if (config == CCI_PMU_CYCLES)
-               mapping = config;
-       else
-               mapping = pmu_validate_hw_event(config);
-
-       return mapping;
+       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)
@@ -319,7 +699,7 @@ static int pmu_request_irq(struct cci_pmu *cci_pmu, irq_handler_t handler)
        if (unlikely(!pmu_device))
                return -ENODEV;
 
-       if (pmu->nr_irqs < 1) {
+       if (cci_pmu->nr_irqs < 1) {
                dev_err(&pmu_device->dev, "no irqs for CCI PMUs defined\n");
                return -ENODEV;
        }
@@ -331,16 +711,16 @@ static int pmu_request_irq(struct cci_pmu *cci_pmu, irq_handler_t handler)
         *
         * This should allow handling of non-unique interrupt for the counters.
         */
-       for (i = 0; i < pmu->nr_irqs; i++) {
-               int err = request_irq(pmu->irqs[i], handler, IRQF_SHARED,
+       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",
-                               pmu->irqs[i]);
+                               cci_pmu->irqs[i]);
                        return err;
                }
 
-               set_bit(i, &pmu->active_irqs);
+               set_bit(i, &cci_pmu->active_irqs);
        }
 
        return 0;
@@ -350,11 +730,11 @@ static void pmu_free_irq(struct cci_pmu *cci_pmu)
 {
        int i;
 
-       for (i = 0; i < pmu->nr_irqs; i++) {
-               if (!test_and_clear_bit(i, &pmu->active_irqs))
+       for (i = 0; i < cci_pmu->nr_irqs; i++) {
+               if (!test_and_clear_bit(i, &cci_pmu->active_irqs))
                        continue;
 
-               free_irq(pmu->irqs[i], cci_pmu);
+               free_irq(cci_pmu->irqs[i], cci_pmu);
        }
 }
 
@@ -369,7 +749,7 @@ static u32 pmu_read_counter(struct perf_event *event)
                dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
                return 0;
        }
-       value = pmu_read_register(idx, CCI_PMU_CNTR);
+       value = pmu_read_register(cci_pmu, idx, CCI_PMU_CNTR);
 
        return value;
 }
@@ -383,7 +763,7 @@ static void pmu_write_counter(struct perf_event *event, 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);
        else
-               pmu_write_register(value, idx, CCI_PMU_CNTR);
+               pmu_write_register(cci_pmu, value, idx, CCI_PMU_CNTR);
 }
 
 static u64 pmu_event_update(struct perf_event *event)
@@ -427,7 +807,7 @@ 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 = &pmu->hw_events;
+       struct cci_pmu_hw_events *events = &cci_pmu->hw_events;
        int idx, handled = IRQ_NONE;
 
        raw_spin_lock_irqsave(&events->pmu_lock, flags);
@@ -436,7 +816,7 @@ static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
         * This should work regardless of whether we have per-counter overflow
         * interrupt or a combined overflow interrupt.
         */
-       for (idx = CCI_PMU_CYCLE_CNTR_IDX; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++) {
+       for (idx = 0; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++) {
                struct perf_event *event = events->events[idx];
                struct hw_perf_event *hw_counter;
 
@@ -446,11 +826,12 @@ static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
                hw_counter = &event->hw;
 
                /* Did this counter overflow? */
-               if (!(pmu_read_register(idx, CCI_PMU_OVRFLW) &
+               if (!(pmu_read_register(cci_pmu, idx, CCI_PMU_OVRFLW) &
                      CCI_PMU_OVRFLW_FLAG))
                        continue;
 
-               pmu_write_register(CCI_PMU_OVRFLW_FLAG, idx, CCI_PMU_OVRFLW);
+               pmu_write_register(cci_pmu, CCI_PMU_OVRFLW_FLAG, idx,
+                                                       CCI_PMU_OVRFLW);
 
                pmu_event_update(event);
                pmu_event_set_period(event);
@@ -492,7 +873,7 @@ static void cci_pmu_enable(struct pmu *pmu)
 {
        struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
        struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
-       int enabled = bitmap_weight(hw_events->used_mask, cci_pmu->num_events);
+       int enabled = bitmap_weight(hw_events->used_mask, cci_pmu->num_cntrs);
        unsigned long flags;
        u32 val;
 
@@ -523,6 +904,16 @@ static void cci_pmu_disable(struct pmu *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);
@@ -547,12 +938,12 @@ static void cci_pmu_start(struct perf_event *event, int pmu_flags)
 
        raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
 
-       /* Configure the event to count, unless you are counting cycles */
-       if (idx != CCI_PMU_CYCLE_CNTR_IDX)
-               pmu_set_event(idx, hwc->config_base);
+       /* 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(idx);
+       pmu_enable_counter(cci_pmu, idx);
 
        raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
 }
@@ -575,7 +966,7 @@ static void cci_pmu_stop(struct perf_event *event, int pmu_flags)
         * We always reprogram the counter, so ignore PERF_EF_UPDATE. See
         * cci_pmu_start()
         */
-       pmu_disable_counter(idx);
+       pmu_disable_counter(cci_pmu, idx);
        pmu_event_update(event);
        hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
 }
@@ -655,13 +1046,16 @@ 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(cci_pmu->num_cntrs)];
        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 = { 0 },
+               .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;
@@ -779,20 +1173,27 @@ static int cci_pmu_event_init(struct perf_event *event)
        return err;
 }
 
-static ssize_t pmu_attr_cpumask_show(struct device *dev,
+static ssize_t pmu_cpumask_attr_show(struct device *dev,
                                     struct device_attribute *attr, char *buf)
 {
+       struct dev_ext_attribute *eattr = container_of(attr,
+                                       struct dev_ext_attribute, attr);
+       struct cci_pmu *cci_pmu = eattr->var;
+
        int n = scnprintf(buf, PAGE_SIZE - 1, "%*pbl",
-                         cpumask_pr_args(&pmu->cpus));
+                         cpumask_pr_args(&cci_pmu->cpus));
        buf[n++] = '\n';
        buf[n] = '\0';
        return n;
 }
 
-static DEVICE_ATTR(cpumask, S_IRUGO, pmu_attr_cpumask_show, NULL);
+static struct dev_ext_attribute pmu_cpumask_attr = {
+       __ATTR(cpumask, S_IRUGO, pmu_cpumask_attr_show, NULL),
+       NULL,           /* Populated in cci_pmu_init */
+};
 
 static struct attribute *pmu_attrs[] = {
-       &dev_attr_cpumask.attr,
+       &pmu_cpumask_attr.attr.attr,
        NULL,
 };
 
@@ -800,14 +1201,78 @@ 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 struct attribute **alloc_attrs(struct platform_device *pdev,
+                               int n, struct dev_ext_attribute *source)
+{
+       int i;
+       struct attribute **attrs;
+
+       /* Alloc n + 1 (for terminating NULL) */
+       attrs  = devm_kcalloc(&pdev->dev, n + 1, sizeof(struct attribute *),
+                                                               GFP_KERNEL);
+       if (!attrs)
+               return attrs;
+       for(i = 0; i < n; i++)
+               attrs[i] = &source[i].attr.attr;
+       return attrs;
+}
+
+static int cci_pmu_init_attrs(struct cci_pmu *cci_pmu, struct platform_device *pdev)
+{
+       const struct cci_pmu_model *model = cci_pmu->model;
+       struct attribute **attrs;
+
+       /*
+        * All allocations below are managed, hence doesn't need to be
+        * free'd explicitly in case of an error.
+        */
+
+       if (model->nevent_attrs) {
+               attrs = alloc_attrs(pdev, model->nevent_attrs,
+                                               model->event_attrs);
+               if (!attrs)
+                       return -ENOMEM;
+               pmu_event_attr_group.attrs = attrs;
+       }
+       if (model->nformat_attrs) {
+               attrs = alloc_attrs(pdev, model->nformat_attrs,
+                                                model->format_attrs);
+               if (!attrs)
+                       return -ENOMEM;
+               pmu_format_attr_group.attrs = attrs;
+       }
+       pmu_cpumask_attr.var = cci_pmu;
+
+       return 0;
+}
+
 static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
 {
        char *name = cci_pmu->model->name;
+       u32 num_cntrs;
+       int rc;
+
+       rc = cci_pmu_init_attrs(cci_pmu, pdev);
+       if (rc)
+               return rc;
+
        cci_pmu->pmu = (struct pmu) {
                .name           = cci_pmu->model->name,
                .task_ctx_nr    = perf_invalid_context,
@@ -823,7 +1288,15 @@ static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
        };
 
        cci_pmu->plat_device = pdev;
-       cci_pmu->num_events = pmu_get_max_counters();
+       num_cntrs = pmu_get_max_counters();
+       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);
 }
@@ -831,12 +1304,14 @@ static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev)
 static int cci_pmu_cpu_notifier(struct notifier_block *self,
                                unsigned long action, void *hcpu)
 {
+       struct cci_pmu *cci_pmu = container_of(self,
+                                       struct cci_pmu, cpu_nb);
        unsigned int cpu = (long)hcpu;
        unsigned int target;
 
        switch (action & ~CPU_TASKS_FROZEN) {
        case CPU_DOWN_PREPARE:
-               if (!cpumask_test_and_clear_cpu(cpu, &pmu->cpus))
+               if (!cpumask_test_and_clear_cpu(cpu, &cci_pmu->cpus))
                        break;
                target = cpumask_any_but(cpu_online_mask, cpu);
                if (target < 0) // UP, last CPU
@@ -845,7 +1320,7 @@ static int cci_pmu_cpu_notifier(struct notifier_block *self,
                 * TODO: migrate context once core races on event->ctx have
                 * been fixed.
                 */
-               cpumask_set_cpu(target, &pmu->cpus);
+               cpumask_set_cpu(target, &cci_pmu->cpus);
        default:
                break;
        }
@@ -853,57 +1328,103 @@ static int cci_pmu_cpu_notifier(struct notifier_block *self,
        return NOTIFY_OK;
 }
 
-static struct notifier_block cci_pmu_cpu_nb = {
-       .notifier_call  = cci_pmu_cpu_notifier,
-       /*
-        * to migrate uncore events, our notifier should be executed
-        * before perf core's notifier.
-        */
-       .priority       = CPU_PRI_PERF + 1,
-};
-
 static struct cci_pmu_model cci_pmu_models[] = {
-       [CCI_REV_R0] = {
+#ifdef CONFIG_ARM_CCI400_PMU
+       [CCI400_R0] = {
                .name = "CCI_400",
+               .fixed_hw_cntrs = 1,    /* Cycle counter */
+               .num_hw_cntrs = 4,
+               .cntr_size = SZ_4K,
+               .format_attrs = cci400_pmu_format_attrs,
+               .nformat_attrs = ARRAY_SIZE(cci400_pmu_format_attrs),
+               .event_attrs = cci400_r0_pmu_event_attrs,
+               .nevent_attrs = ARRAY_SIZE(cci400_r0_pmu_event_attrs),
                .event_ranges = {
                        [CCI_IF_SLAVE] = {
-                               CCI_REV_R0_SLAVE_PORT_MIN_EV,
-                               CCI_REV_R0_SLAVE_PORT_MAX_EV,
+                               CCI400_R0_SLAVE_PORT_MIN_EV,
+                               CCI400_R0_SLAVE_PORT_MAX_EV,
                        },
                        [CCI_IF_MASTER] = {
-                               CCI_REV_R0_MASTER_PORT_MIN_EV,
-                               CCI_REV_R0_MASTER_PORT_MAX_EV,
+                               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,
        },
-       [CCI_REV_R1] = {
+       [CCI400_R1] = {
                .name = "CCI_400_r1",
+               .fixed_hw_cntrs = 1,    /* Cycle counter */
+               .num_hw_cntrs = 4,
+               .cntr_size = SZ_4K,
+               .format_attrs = cci400_pmu_format_attrs,
+               .nformat_attrs = ARRAY_SIZE(cci400_pmu_format_attrs),
+               .event_attrs = cci400_r1_pmu_event_attrs,
+               .nevent_attrs = ARRAY_SIZE(cci400_r1_pmu_event_attrs),
                .event_ranges = {
                        [CCI_IF_SLAVE] = {
-                               CCI_REV_R1_SLAVE_PORT_MIN_EV,
-                               CCI_REV_R1_SLAVE_PORT_MAX_EV,
+                               CCI400_R1_SLAVE_PORT_MIN_EV,
+                               CCI400_R1_SLAVE_PORT_MAX_EV,
                        },
                        [CCI_IF_MASTER] = {
-                               CCI_REV_R1_MASTER_PORT_MIN_EV,
-                               CCI_REV_R1_MASTER_PORT_MAX_EV,
+                               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_CCI500_PMU
+       [CCI500_R0] = {
+               .name = "CCI_500",
+               .fixed_hw_cntrs = 0,
+               .num_hw_cntrs = 8,
+               .cntr_size = SZ_64K,
+               .format_attrs = cci500_pmu_format_attrs,
+               .nformat_attrs = ARRAY_SIZE(cci500_pmu_format_attrs),
+               .event_attrs = cci500_pmu_event_attrs,
+               .nevent_attrs = ARRAY_SIZE(cci500_pmu_event_attrs),
+               .event_ranges = {
+                       [CCI_IF_SLAVE] = {
+                               CCI500_SLAVE_PORT_MIN_EV,
+                               CCI500_SLAVE_PORT_MAX_EV,
+                       },
+                       [CCI_IF_MASTER] = {
+                               CCI500_MASTER_PORT_MIN_EV,
+                               CCI500_MASTER_PORT_MAX_EV,
+                       },
+                       [CCI_IF_GLOBAL] = {
+                               CCI500_GLOBAL_PORT_MIN_EV,
+                               CCI500_GLOBAL_PORT_MAX_EV,
+                       },
+               },
+               .validate_hw_event = cci500_validate_hw_event,
+       },
+#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[CCI_REV_R0],
+               .data   = &cci_pmu_models[CCI400_R0],
        },
        {
                .compatible = "arm,cci-400-pmu,r1",
-               .data   = &cci_pmu_models[CCI_REV_R1],
+               .data   = &cci_pmu_models[CCI400_R1],
+       },
+#endif
+#ifdef CONFIG_ARM_CCI500_PMU
+       {
+               .compatible = "arm,cci-500-pmu,r0",
+               .data = &cci_pmu_models[CCI500_R0],
        },
+#endif
        {},
 };
 
@@ -932,68 +1453,114 @@ static bool is_duplicate_irq(int irq, int *irqs, int nr_irqs)
        return false;
 }
 
-static int cci_pmu_probe(struct platform_device *pdev)
+static struct cci_pmu *cci_pmu_alloc(struct platform_device *pdev)
 {
-       struct resource *res;
-       int i, ret, irq;
+       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.
+        */
        model = get_cci_model(pdev);
        if (!model) {
                dev_warn(&pdev->dev, "CCI PMU version not supported\n");
-               return -ENODEV;
+               return ERR_PTR(-ENODEV);
        }
 
-       pmu = devm_kzalloc(&pdev->dev, sizeof(*pmu), GFP_KERNEL);
-       if (!pmu)
-               return -ENOMEM;
+       cci_pmu = devm_kzalloc(&pdev->dev, sizeof(*cci_pmu), GFP_KERNEL);
+       if (!cci_pmu)
+               return ERR_PTR(-ENOMEM);
+
+       cci_pmu->model = model;
+       cci_pmu->irqs = devm_kcalloc(&pdev->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(&pdev->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(&pdev->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 resource *res;
+       struct cci_pmu *cci_pmu;
+       int i, ret, irq;
+
+       cci_pmu = cci_pmu_alloc(pdev);
+       if (IS_ERR(cci_pmu))
+               return PTR_ERR(cci_pmu);
 
-       pmu->model = model;
        res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
-       pmu->base = devm_ioremap_resource(&pdev->dev, res);
-       if (IS_ERR(pmu->base))
+       cci_pmu->base = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(cci_pmu->base))
                return -ENOMEM;
 
        /*
-        * CCI PMU has 5 overflow signals - one per counter; but some may be tied
+        * CCI PMU has one overflow interrupt per counter; but some may be tied
         * together to a common interrupt.
         */
-       pmu->nr_irqs = 0;
-       for (i = 0; i < CCI_PMU_MAX_HW_EVENTS; i++) {
+       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, pmu->irqs, pmu->nr_irqs))
+               if (is_duplicate_irq(irq, cci_pmu->irqs, cci_pmu->nr_irqs))
                        continue;
 
-               pmu->irqs[pmu->nr_irqs++] = irq;
+               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_EVENTS) {
+       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_EVENTS);
+                       i, CCI_PMU_MAX_HW_CNTRS(cci_pmu->model));
                return -EINVAL;
        }
 
-       raw_spin_lock_init(&pmu->hw_events.pmu_lock);
-       mutex_init(&pmu->reserve_mutex);
-       atomic_set(&pmu->active_events, 0);
-       cpumask_set_cpu(smp_processor_id(), &pmu->cpus);
+       raw_spin_lock_init(&cci_pmu->hw_events.pmu_lock);
+       mutex_init(&cci_pmu->reserve_mutex);
+       atomic_set(&cci_pmu->active_events, 0);
+       cpumask_set_cpu(smp_processor_id(), &cci_pmu->cpus);
+
+       cci_pmu->cpu_nb = (struct notifier_block) {
+               .notifier_call  = cci_pmu_cpu_notifier,
+               /*
+                * to migrate uncore events, our notifier should be executed
+                * before perf core's notifier.
+                */
+               .priority       = CPU_PRI_PERF + 1,
+       };
 
-       ret = register_cpu_notifier(&cci_pmu_cpu_nb);
+       ret = register_cpu_notifier(&cci_pmu->cpu_nb);
        if (ret)
                return ret;
 
-       ret = cci_pmu_init(pmu, pdev);
-       if (ret)
+       ret = cci_pmu_init(cci_pmu, pdev);
+       if (ret) {
+               unregister_cpu_notifier(&cci_pmu->cpu_nb);
                return ret;
+       }
 
-       pr_info("ARM %s PMU driver probed", pmu->model->name);
+       pr_info("ARM %s PMU driver probed", cci_pmu->model->name);
        return 0;
 }
 
@@ -1032,14 +1599,14 @@ static int __init cci_platform_init(void)
        return platform_driver_register(&cci_platform_driver);
 }
 
-#else /* !CONFIG_ARM_CCI400_PMU */
+#else /* !CONFIG_ARM_CCI_PMU */
 
 static int __init cci_platform_init(void)
 {
        return 0;
 }
 
-#endif /* CONFIG_ARM_CCI400_PMU */
+#endif /* CONFIG_ARM_CCI_PMU */
 
 #ifdef CONFIG_ARM_CCI400_PORT_CTRL