#include <linux/arm-cci.h>
#include <linux/io.h>
+#include <linux/interrupt.h>
#include <linux/module.h>
-#include <linux/platform_device.h>
#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/of_platform.h>
+#include <linux/perf_event.h>
+#include <linux/platform_device.h>
#include <linux/slab.h>
+#include <linux/spinlock.h>
#include <asm/cacheflush.h>
-#include <asm/irq_regs.h>
-#include <asm/pmu.h>
#include <asm/smp_plat.h>
-#define DRIVER_NAME "CCI"
-
-#define CCI_PORT_CTRL 0x0
-#define CCI_CTRL_STATUS 0xc
-
-#define CCI_ENABLE_SNOOP_REQ 0x1
-#define CCI_ENABLE_DVM_REQ 0x2
-#define CCI_ENABLE_REQ (CCI_ENABLE_SNOOP_REQ | CCI_ENABLE_DVM_REQ)
+static void __iomem *cci_ctrl_base;
+static unsigned long cci_ctrl_phys;
+#ifdef CONFIG_ARM_CCI400_PORT_CTRL
struct cci_nb_ports {
unsigned int nb_ace;
unsigned int nb_ace_lite;
};
-enum cci_ace_port_type {
- ACE_INVALID_PORT = 0x0,
- ACE_PORT,
- ACE_LITE_PORT,
+static const struct cci_nb_ports cci400_ports = {
+ .nb_ace = 2,
+ .nb_ace_lite = 3
};
-struct cci_ace_port {
+#define CCI400_PORTS_DATA (&cci400_ports)
+#else
+#define CCI400_PORTS_DATA (NULL)
+#endif
+
+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_CCI_PMU
+
+#define DRIVER_NAME "ARM-CCI"
+#define DRIVER_NAME_PMU DRIVER_NAME " 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_CCI500_PMU
+ CCI_IF_GLOBAL,
+#endif
+ CCI_IF_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;
+ 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[];
+
+struct cci_pmu {
void __iomem *base;
- unsigned long phys;
- enum cci_ace_port_type type;
- struct device_node *dn;
+ struct pmu pmu;
+ 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;
+ struct notifier_block cpu_nb;
+ cpumask_t cpus;
};
-static struct cci_ace_port *ports;
-static unsigned int nb_cci_ports;
+#define to_cci_pmu(c) (container_of(c, struct cci_pmu, pmu))
-static void __iomem *cci_ctrl_base;
-static unsigned long cci_ctrl_phys;
+enum cci_models {
+#ifdef CONFIG_ARM_CCI400_PMU
+ CCI400_R0,
+ CCI400_R1,
+#endif
+#ifdef CONFIG_ARM_CCI500_PMU
+ CCI500_R0,
+#endif
+ CCI_MODEL_MAX
+};
-#ifdef CONFIG_HW_PERF_EVENTS
+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);
-static void __iomem *cci_pmu_base;
+#define CCI_EXT_ATTR_ENTRY(_name, _func, _config) \
+ { __ATTR(_name, S_IRUGO, _func, NULL), (void *)_config }
-#define CCI400_PMCR 0x0100
+#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)
-#define CCI400_PMU_CYCLE_CNTR_BASE 0x0000
-#define CCI400_PMU_CNTR_BASE(idx) (CCI400_PMU_CYCLE_CNTR_BASE + (idx) * 0x1000)
+/* CCI400 PMU Specific definitions */
-#define CCI400_PMCR_CEN 0x00000001
-#define CCI400_PMCR_RST 0x00000002
-#define CCI400_PMCR_CCR 0x00000004
-#define CCI400_PMCR_CCD 0x00000008
-#define CCI400_PMCR_EX 0x00000010
-#define CCI400_PMCR_DP 0x00000020
-#define CCI400_PMCR_NCNT_MASK 0x0000F800
-#define CCI400_PMCR_NCNT_SHIFT 11
+#ifdef CONFIG_ARM_CCI400_PMU
-#define CCI400_PMU_EVT_SEL 0x000
-#define CCI400_PMU_CNTR 0x004
-#define CCI400_PMU_CNTR_CTRL 0x008
-#define CCI400_PMU_OVERFLOW 0x00C
+/* 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_PMU_OVERFLOW_FLAG 1
+#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
+ CCI400_PMU_CYCLES = 0xff
};
-#define CCI400_PMU_EVENT_MASK 0xff
-#define CCI400_PMU_EVENT_SOURCE(event) ((event >> 5) & 0x7)
-#define CCI400_PMU_EVENT_CODE(event) (event & 0x1f)
+#define CCI400_PMU_CYCLE_CNTR_IDX 0
+#define CCI400_PMU_CNTR0_IDX 1
-#define CCI400_PMU_EVENT_SOURCE_S0 0
-#define CCI400_PMU_EVENT_SOURCE_S4 4
-#define CCI400_PMU_EVENT_SOURCE_M0 5
-#define CCI400_PMU_EVENT_SOURCE_M2 7
+/*
+ * 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_SLAVE_MIN 0x0
-#define CCI400_PMU_EVENT_SLAVE_MAX 0x13
+#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"),
+};
-#define CCI400_PMU_EVENT_MASTER_MIN 0x14
-#define CCI400_PMU_EVENT_MASTER_MAX 0x1A
+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),
+};
-#define CCI400_PMU_MAX_HW_EVENTS 5 /* CCI PMU has 4 counters + 1 cycle counter */
+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 CCI400_PMU_CYCLE_COUNTER_IDX 0
-#define CCI400_PMU_COUNTER0_IDX 1
-#define CCI400_PMU_COUNTER_LAST(cci_pmu) (CCI400_PMU_CYCLE_COUNTER_IDX + cci_pmu->num_events - 1)
+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;
-static struct perf_event *events[CCI400_PMU_MAX_HW_EVENTS];
-static unsigned long used_mask[BITS_TO_LONGS(CCI400_PMU_MAX_HW_EVENTS)];
-static struct pmu_hw_events cci_hw_events = {
- .events = events,
- .used_mask = used_mask,
-};
+ /* 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;
-static int cci_pmu_validate_hw_event(u8 hw_event)
+ 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 (ev_source <= CCI400_PMU_EVENT_SOURCE_S4 &&
- ev_code <= CCI400_PMU_EVENT_SLAVE_MAX)
- return hw_event;
- else if (CCI400_PMU_EVENT_SOURCE_M0 <= ev_source &&
- ev_source <= CCI400_PMU_EVENT_SOURCE_M2 &&
- CCI400_PMU_EVENT_MASTER_MIN <= ev_code &&
- ev_code <= CCI400_PMU_EVENT_MASTER_MAX)
- return hw_event;
+ if (hw_event & ~CCI400_PMU_EVENT_MASK)
+ return -ENOENT;
- return -EINVAL;
+ 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 inline int cci_pmu_counter_is_valid(struct arm_pmu *cci_pmu, int idx)
+static int probe_cci400_revision(void)
{
- return CCI400_PMU_CYCLE_COUNTER_IDX <= idx &&
- idx <= CCI400_PMU_COUNTER_LAST(cci_pmu);
+ int rev;
+ rev = readl_relaxed(cci_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 inline u32 cci_pmu_read_register(int idx, unsigned int offset)
+static const struct cci_pmu_model *probe_cci_model(struct platform_device *pdev)
+{
+ if (platform_has_secure_cci_access())
+ 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 readl_relaxed(cci_pmu_base + CCI400_PMU_CNTR_BASE(idx) + offset);
+ return NULL;
}
+#endif /* CONFIG_ARM_CCI400_PMU */
+
+#ifdef CONFIG_ARM_CCI500_PMU
-static inline void cci_pmu_write_register(u32 value, int idx, unsigned int offset)
+/*
+ * 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)
{
- return writel_relaxed(value, cci_pmu_base + CCI400_PMU_CNTR_BASE(idx) + offset);
+ 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 inline void cci_pmu_disable_counter(int idx)
+static int cci500_validate_hw_event(struct cci_pmu *cci_pmu,
+ unsigned long hw_event)
{
- cci_pmu_write_register(0, idx, CCI400_PMU_CNTR_CTRL);
+ 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 inline void cci_pmu_enable_counter(int idx)
+static ssize_t cci_pmu_event_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
{
- cci_pmu_write_register(1, idx, CCI400_PMU_CNTR_CTRL);
+ 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 inline void cci_pmu_select_event(int idx, unsigned long event)
+static int pmu_is_valid_counter(struct cci_pmu *cci_pmu, int idx)
{
- event &= CCI400_PMU_EVENT_MASK;
- cci_pmu_write_register(event, idx, CCI400_PMU_EVT_SEL);
+ return 0 <= idx && idx <= CCI_PMU_CNTR_LAST(cci_pmu);
}
-static u32 cci_pmu_get_max_counters(void)
+static u32 pmu_read_register(struct cci_pmu *cci_pmu, int idx, unsigned int offset)
{
- u32 n_cnts = (readl_relaxed(cci_ctrl_base + CCI400_PMCR) &
- CCI400_PMCR_NCNT_MASK) >> CCI400_PMCR_NCNT_SHIFT;
+ return readl_relaxed(cci_pmu->base +
+ CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
+}
- /* add 1 for cycle counter */
- return n_cnts + 1;
+static void pmu_write_register(struct cci_pmu *cci_pmu, u32 value,
+ int idx, unsigned int offset)
+{
+ return writel_relaxed(value, cci_pmu->base +
+ CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset);
}
-static struct pmu_hw_events *cci_pmu_get_hw_events(void)
+static void pmu_disable_counter(struct cci_pmu *cci_pmu, int idx)
{
- return &cci_hw_events;
+ pmu_write_register(cci_pmu, 0, idx, CCI_PMU_CNTR_CTRL);
}
-static int cci_pmu_get_event_idx(struct pmu_hw_events *hw, struct perf_event *event)
+static void pmu_enable_counter(struct cci_pmu *cci_pmu, int idx)
{
- struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
- struct hw_perf_event *hw_event = &event->hw;
- unsigned long cci_event = hw_event->config_base & CCI400_PMU_EVENT_MASK;
- int idx;
+ pmu_write_register(cci_pmu, 1, idx, CCI_PMU_CNTR_CTRL);
+}
- if (cci_event == CCI400_PMU_CYCLES) {
- if (test_and_set_bit(CCI400_PMU_CYCLE_COUNTER_IDX, hw->used_mask))
- return -EAGAIN;
+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);
+}
- return CCI400_PMU_CYCLE_COUNTER_IDX;
- }
+/*
+ * Returns the number of programmable counters actually implemented
+ * by the cci
+ */
+static u32 pmu_get_max_counters(void)
+{
+ return (readl_relaxed(cci_ctrl_base + CCI_PMCR) &
+ CCI_PMCR_NCNT_MASK) >> CCI_PMCR_NCNT_SHIFT;
+}
- for (idx = CCI400_PMU_COUNTER0_IDX; idx <= CCI400_PMU_COUNTER_LAST(cci_pmu); ++idx) {
+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 cci_pmu_map_event(struct perf_event *event)
+static int pmu_map_event(struct perf_event *event)
{
- int mapping;
- u8 config = event->attr.config & CCI400_PMU_EVENT_MASK;
+ 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;
- /* 0xff is used to represent CCI Cycles */
- if (config == 0xff)
- mapping = config;
- else
- mapping = cci_pmu_validate_hw_event(config);
-
- return mapping;
+ return cci_pmu->model->validate_hw_event(cci_pmu, event->attr.config);
}
-static int cci_pmu_request_irq(struct arm_pmu *cci_pmu, irq_handler_t handler)
+static int pmu_request_irq(struct cci_pmu *cci_pmu, irq_handler_t handler)
{
- int irq, err, i = 0;
+ int i;
struct platform_device *pmu_device = cci_pmu->plat_device;
if (unlikely(!pmu_device))
return -ENODEV;
- /* CCI exports 6 interrupts - 1 nERRORIRQ + 5 nEVNTCNTOVERFLOW (PMU)
- nERRORIRQ will be handled by secure firmware on TC2. So we
- assume that all CCI interrupts listed in the linux device
- tree are PMU interrupts.
+ if (cci_pmu->nr_irqs < 1) {
+ dev_err(&pmu_device->dev, "no irqs for CCI PMUs defined\n");
+ return -ENODEV;
+ }
- The following code should then be able to handle different routing
- of the CCI PMU interrupts.
- */
- while ((irq = platform_get_irq(pmu_device, i)) > 0) {
- err = request_irq(irq, handler, 0, "arm-cci-pmu", cci_pmu);
+ /*
+ * 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",
- irq);
+ cci_pmu->irqs[i]);
return err;
}
- i++;
+
+ set_bit(i, &cci_pmu->active_irqs);
}
return 0;
}
-static irqreturn_t cci_pmu_handle_irq(int irq_num, void *dev)
+static void pmu_free_irq(struct cci_pmu *cci_pmu)
{
- struct arm_pmu *cci_pmu = (struct arm_pmu *)dev;
- struct pmu_hw_events *events = cci_pmu->get_hw_events();
- struct perf_sample_data data;
- struct pt_regs *regs;
- int idx;
+ 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 perf_event *event, u32 value)
+{
+ struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+ struct hw_perf_event *hw_counter = &event->hw;
+ int idx = hw_counter->idx;
+
+ if (unlikely(!pmu_is_valid_counter(cci_pmu, idx)))
+ dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
+ else
+ pmu_write_register(cci_pmu, value, idx, CCI_PMU_CNTR);
+}
+
+static u64 pmu_event_update(struct perf_event *event)
+{
+ struct hw_perf_event *hwc = &event->hw;
+ u64 delta, prev_raw_count, new_raw_count;
- regs = get_irq_regs();
+ 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;
+}
- /* 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 = CCI400_PMU_CYCLE_COUNTER_IDX; idx <= CCI400_PMU_COUNTER_LAST(cci_pmu); idx++) {
+static void pmu_read(struct perf_event *event)
+{
+ pmu_event_update(event);
+}
+
+void pmu_event_set_period(struct perf_event *event)
+{
+ struct hw_perf_event *hwc = &event->hw;
+ /*
+ * The CCI PMU counters have a period of 2^32. To account for the
+ * possiblity of extreme interrupt latency we program for a period of
+ * half that. Hopefully we can handle the interrupt before another 2^31
+ * events occur and the counter overtakes its previous value.
+ */
+ u64 val = 1ULL << 31;
+ local64_set(&hwc->prev_count, val);
+ pmu_write_counter(event, val);
+}
+
+static irqreturn_t pmu_handle_irq(int irq_num, void *dev)
+{
+ unsigned long flags;
+ struct 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);
+ /*
+ * 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];
struct hw_perf_event *hw_counter;
hw_counter = &event->hw;
/* Did this counter overflow? */
- if (!(cci_pmu_read_register(idx, CCI400_PMU_OVERFLOW) & CCI400_PMU_OVERFLOW_FLAG))
+ if (!(pmu_read_register(cci_pmu, idx, CCI_PMU_OVRFLW) &
+ CCI_PMU_OVRFLW_FLAG))
continue;
- cci_pmu_write_register(CCI400_PMU_OVERFLOW_FLAG, idx, CCI400_PMU_OVERFLOW);
- armpmu_event_update(event);
- perf_sample_data_init(&data, 0, hw_counter->last_period);
- if (!armpmu_event_set_period(event))
- continue;
+ pmu_write_register(cci_pmu, CCI_PMU_OVRFLW_FLAG, idx,
+ CCI_PMU_OVRFLW);
- if (perf_event_overflow(event, &data, regs))
- cci_pmu->disable(event);
+ pmu_event_update(event);
+ pmu_event_set_period(event);
+ handled = IRQ_HANDLED;
}
+ raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
- irq_work_run();
- return IRQ_HANDLED;
+ return IRQ_RETVAL(handled);
}
-static void cci_pmu_free_irq(struct arm_pmu *cci_pmu)
+static int cci_pmu_get_hw(struct cci_pmu *cci_pmu)
{
- int irq, i = 0;
- struct platform_device *pmu_device = cci_pmu->plat_device;
+ 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;
- while ((irq = platform_get_irq(pmu_device, i)) > 0) {
- free_irq(irq, cci_pmu);
- i++;
+ 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_event(struct perf_event *event)
+static void cci_pmu_enable(struct pmu *pmu)
{
+ struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
+ struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+ int enabled = bitmap_weight(hw_events->used_mask, cci_pmu->num_cntrs);
unsigned long flags;
- struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
- struct pmu_hw_events *events = cci_pmu->get_hw_events();
- struct hw_perf_event *hw_counter = &event->hw;
- int idx = hw_counter->idx;
+ u32 val;
- if (unlikely(!cci_pmu_counter_is_valid(cci_pmu, idx))) {
- dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
+ if (!enabled)
return;
- }
- raw_spin_lock_irqsave(&events->pmu_lock, flags);
+ raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
- /* Configure the event to count, unless you are counting cycles */
- if (idx != CCI400_PMU_CYCLE_COUNTER_IDX)
- cci_pmu_select_event(idx, hw_counter->config_base);
+ /* Enable all the PMU counters. */
+ val = readl_relaxed(cci_ctrl_base + CCI_PMCR) | CCI_PMCR_CEN;
+ writel(val, cci_ctrl_base + CCI_PMCR);
+ raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
- cci_pmu_enable_counter(idx);
+}
- raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+static void cci_pmu_disable(struct pmu *pmu)
+{
+ struct cci_pmu *cci_pmu = to_cci_pmu(pmu);
+ struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+ unsigned long flags;
+ u32 val;
+
+ raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
+
+ /* Disable all the PMU counters. */
+ val = readl_relaxed(cci_ctrl_base + CCI_PMCR) & ~CCI_PMCR_CEN;
+ writel(val, cci_ctrl_base + CCI_PMCR);
+ raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
+}
+
+/*
+ * 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_disable_event(struct perf_event *event)
+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;
- struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
- struct pmu_hw_events *events = cci_pmu->get_hw_events();
- struct hw_perf_event *hw_counter = &event->hw;
- int idx = hw_counter->idx;
- if (unlikely(!cci_pmu_counter_is_valid(cci_pmu, idx))) {
+ /*
+ * To handle interrupt latency, we always reprogram the period
+ * regardlesss of PERF_EF_RELOAD.
+ */
+ if (pmu_flags & PERF_EF_RELOAD)
+ WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE));
+
+ hwc->state = 0;
+
+ if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) {
dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
return;
}
- raw_spin_lock_irqsave(&events->pmu_lock, flags);
+ raw_spin_lock_irqsave(&hw_events->pmu_lock, flags);
- cci_pmu_disable_counter(idx);
+ /* 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);
- raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+ pmu_event_set_period(event);
+ pmu_enable_counter(cci_pmu, idx);
+
+ raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags);
}
-static void cci_pmu_start(struct arm_pmu *cci_pmu)
+static void cci_pmu_stop(struct perf_event *event, int pmu_flags)
{
- u32 val;
- unsigned long flags;
- struct pmu_hw_events *events = cci_pmu->get_hw_events();
+ struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+ struct hw_perf_event *hwc = &event->hw;
+ int idx = hwc->idx;
- raw_spin_lock_irqsave(&events->pmu_lock, flags);
+ if (hwc->state & PERF_HES_STOPPED)
+ return;
- /* Enable all the PMU counters. */
- val = readl(cci_ctrl_base + CCI400_PMCR) | CCI400_PMCR_CEN;
- writel(val, cci_ctrl_base + CCI400_PMCR);
+ 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_unlock_irqrestore(&events->pmu_lock, flags);
+ /*
+ * 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 void cci_pmu_stop(struct arm_pmu *cci_pmu)
+static int cci_pmu_add(struct perf_event *event, int flags)
{
- u32 val;
- unsigned long flags;
- struct pmu_hw_events *events = cci_pmu->get_hw_events();
+ struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+ struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events;
+ struct hw_perf_event *hwc = &event->hw;
+ int idx;
+ int err = 0;
- raw_spin_lock_irqsave(&events->pmu_lock, flags);
+ perf_pmu_disable(event->pmu);
- /* Disable all the PMU counters. */
- val = readl(cci_ctrl_base + CCI400_PMCR) & ~CCI400_PMCR_CEN;
- writel(val, cci_ctrl_base + CCI400_PMCR);
+ /* If we don't have a space for the counter then finish early. */
+ idx = pmu_get_event_idx(hw_events, event);
+ if (idx < 0) {
+ err = idx;
+ goto out;
+ }
- raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+ event->hw.idx = idx;
+ hw_events->events[idx] = event;
+
+ hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
+ if (flags & PERF_EF_START)
+ cci_pmu_start(event, PERF_EF_RELOAD);
+
+ /* Propagate our changes to the userspace mapping. */
+ perf_event_update_userpage(event);
+
+out:
+ perf_pmu_enable(event->pmu);
+ return err;
}
-static u32 cci_pmu_read_counter(struct perf_event *event)
+static void cci_pmu_del(struct perf_event *event, int flags)
{
- struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
- struct hw_perf_event *hw_counter = &event->hw;
- int idx = hw_counter->idx;
- u32 value;
+ struct 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;
- if (unlikely(!cci_pmu_counter_is_valid(cci_pmu, idx))) {
- dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", 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(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 = mask,
+ };
+ memset(mask, 0, BITS_TO_LONGS(cci_pmu->num_cntrs) * sizeof(unsigned long));
+
+ if (!validate_event(event->pmu, &fake_pmu, leader))
+ return -EINVAL;
+
+ list_for_each_entry(sibling, &leader->sibling_list, group_entry) {
+ if (!validate_event(event->pmu, &fake_pmu, sibling))
+ return -EINVAL;
}
- value = cci_pmu_read_register(idx, CCI400_PMU_CNTR);
- return value;
+ if (!validate_event(event->pmu, &fake_pmu, event))
+ return -EINVAL;
+
+ return 0;
}
-static void cci_pmu_write_counter(struct perf_event *event, u32 value)
+static int
+__hw_perf_event_init(struct perf_event *event)
{
- struct arm_pmu *cci_pmu = to_arm_pmu(event->pmu);
- struct hw_perf_event *hw_counter = &event->hw;
- int idx = hw_counter->idx;
+ struct hw_perf_event *hwc = &event->hw;
+ int mapping;
- if (unlikely(!cci_pmu_counter_is_valid(cci_pmu, idx)))
- dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx);
- else
- cci_pmu_write_register(value, idx, CCI400_PMU_CNTR);
-}
-
-static struct arm_pmu cci_pmu = {
- .name = DRIVER_NAME,
- .max_period = (1LLU << 32) - 1,
- .get_hw_events = cci_pmu_get_hw_events,
- .get_event_idx = cci_pmu_get_event_idx,
- .map_event = cci_pmu_map_event,
- .request_irq = cci_pmu_request_irq,
- .handle_irq = cci_pmu_handle_irq,
- .free_irq = cci_pmu_free_irq,
- .enable = cci_pmu_enable_event,
- .disable = cci_pmu_disable_event,
- .start = cci_pmu_start,
- .stop = cci_pmu_stop,
- .read_counter = cci_pmu_read_counter,
- .write_counter = cci_pmu_write_counter,
+ mapping = pmu_map_event(event);
+
+ if (mapping < 0) {
+ pr_debug("event %x:%llx not supported\n", event->attr.type,
+ event->attr.config);
+ return mapping;
+ }
+
+ /*
+ * We don't assign an index until we actually place the event onto
+ * hardware. Use -1 to signify that we haven't decided where to put it
+ * yet.
+ */
+ hwc->idx = -1;
+ hwc->config_base = 0;
+ hwc->config = 0;
+ hwc->event_base = 0;
+
+ /*
+ * Store the event encoding into the config_base field.
+ */
+ hwc->config_base |= (unsigned long)mapping;
+
+ /*
+ * Limit the sample_period to half of the counter width. That way, the
+ * new counter value is far less likely to overtake the previous one
+ * unless you have some serious IRQ latency issues.
+ */
+ hwc->sample_period = CCI_PMU_CNTR_MASK >> 1;
+ hwc->last_period = hwc->sample_period;
+ local64_set(&hwc->period_left, hwc->sample_period);
+
+ if (event->group_leader != event) {
+ if (validate_group(event) != 0)
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static int cci_pmu_event_init(struct perf_event *event)
+{
+ struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu);
+ atomic_t *active_events = &cci_pmu->active_events;
+ int err = 0;
+ int cpu;
+
+ if (event->attr.type != event->pmu->type)
+ return -ENOENT;
+
+ /* Shared by all CPUs, no meaningful state to sample */
+ if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
+ return -EOPNOTSUPP;
+
+ /* We have no filtering of any kind */
+ if (event->attr.exclude_user ||
+ event->attr.exclude_kernel ||
+ event->attr.exclude_hv ||
+ event->attr.exclude_idle ||
+ event->attr.exclude_host ||
+ event->attr.exclude_guest)
+ return -EINVAL;
+
+ /*
+ * Following the example set by other "uncore" PMUs, we accept any CPU
+ * and rewrite its affinity dynamically rather than having perf core
+ * handle cpu == -1 and pid == -1 for this case.
+ *
+ * The perf core will pin online CPUs for the duration of this call and
+ * the event being installed into its context, so the PMU's CPU can't
+ * change under our feet.
+ */
+ cpu = cpumask_first(&cci_pmu->cpus);
+ if (event->cpu < 0 || cpu < 0)
+ return -EINVAL;
+ event->cpu = cpu;
+
+ event->destroy = hw_perf_event_destroy;
+ if (!atomic_inc_not_zero(active_events)) {
+ mutex_lock(&cci_pmu->reserve_mutex);
+ if (atomic_read(active_events) == 0)
+ err = cci_pmu_get_hw(cci_pmu);
+ if (!err)
+ atomic_inc(active_events);
+ mutex_unlock(&cci_pmu->reserve_mutex);
+ }
+ if (err)
+ return err;
+
+ err = __hw_perf_event_init(event);
+ if (err)
+ hw_perf_event_destroy(event);
+
+ return err;
+}
+
+static 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(&cci_pmu->cpus));
+ buf[n++] = '\n';
+ buf[n] = '\0';
+ return n;
+}
+
+static struct dev_ext_attribute pmu_cpumask_attr = {
+ __ATTR(cpumask, S_IRUGO, pmu_cpumask_attr_show, NULL),
+ NULL, /* Populated in cci_pmu_init */
};
-static int cci_pmu_probe(struct platform_device *pdev)
+static struct attribute *pmu_attrs[] = {
+ &pmu_cpumask_attr.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 struct attribute **alloc_attrs(struct platform_device *pdev,
+ int n, struct dev_ext_attribute *source)
{
- struct resource *res;
+ 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;
+}
- res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
- cci_pmu_base = devm_ioremap_resource(&pdev->dev, res);
- if (IS_ERR(cci_pmu_base))
- return PTR_ERR(cci_pmu_base);
+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;
- cci_pmu.plat_device = pdev;
- cci_pmu.num_events = cci_pmu_get_max_counters();
- raw_spin_lock_init(&cci_hw_events.pmu_lock);
- cpumask_setall(&cci_pmu.valid_cpus);
+ /*
+ * All allocations below are managed, hence doesn't need to be
+ * free'd explicitly in case of an error.
+ */
- return armpmu_register(&cci_pmu, -1);
+ 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,
+ .pmu_enable = cci_pmu_enable,
+ .pmu_disable = cci_pmu_disable,
+ .event_init = cci_pmu_event_init,
+ .add = cci_pmu_add,
+ .del = cci_pmu_del,
+ .start = cci_pmu_start,
+ .stop = cci_pmu_stop,
+ .read = pmu_read,
+ .attr_groups = pmu_attr_groups,
+ };
+
+ cci_pmu->plat_device = pdev;
+ 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);
+}
+
+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, &cci_pmu->cpus))
+ break;
+ target = cpumask_any_but(cpu_online_mask, cpu);
+ if (target < 0) // UP, last CPU
+ break;
+ /*
+ * TODO: migrate context once core races on event->ctx have
+ * been fixed.
+ */
+ cpumask_set_cpu(target, &cci_pmu->cpus);
+ default:
+ break;
+ }
+
+ return NOTIFY_OK;
+}
+
+static struct cci_pmu_model cci_pmu_models[] = {
+#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] = {
+ 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 = 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] = {
+ 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_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[] = {
- {.compatible = "arm,cci-400-pmu"},
+#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_CCI500_PMU
+ {
+ .compatible = "arm,cci-500-pmu,r0",
+ .data = &cci_pmu_models[CCI500_R0],
+ },
+#endif
{},
};
-static struct platform_driver cci_pmu_platform_driver = {
+static inline const struct cci_pmu_model *get_cci_model(struct platform_device *pdev)
+{
+ const struct of_device_id *match = of_match_node(arm_cci_pmu_matches,
+ pdev->dev.of_node);
+ if (!match)
+ return NULL;
+ if (match->data)
+ return match->data;
+
+ dev_warn(&pdev->dev, "DEPRECATED compatible property,"
+ "requires secure access to CCI registers");
+ return probe_cci_model(pdev);
+}
+
+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 platform_device *pdev)
+{
+ 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 ERR_PTR(-ENODEV);
+ }
+
+ 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);
+
+ res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+ cci_pmu->base = devm_ioremap_resource(&pdev->dev, res);
+ 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);
+ 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);
+ if (ret)
+ return 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", cci_pmu->model->name);
+ return 0;
+}
+
+static int cci_platform_probe(struct platform_device *pdev)
+{
+ if (!cci_probed())
+ return -ENODEV;
+
+ return of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);
+}
+
+static struct platform_driver cci_pmu_driver = {
.driver = {
- .name = DRIVER_NAME,
+ .name = DRIVER_NAME_PMU,
.of_match_table = arm_cci_pmu_matches,
},
.probe = cci_pmu_probe,
};
-static int __init cci_pmu_init(void)
+static struct platform_driver cci_platform_driver = {
+ .driver = {
+ .name = DRIVER_NAME,
+ .of_match_table = arm_cci_matches,
+ },
+ .probe = cci_platform_probe,
+};
+
+static int __init cci_platform_init(void)
{
- if (platform_driver_register(&cci_pmu_platform_driver))
- WARN(1, "unable to register CCI platform driver\n");
- return 0;
+ int ret;
+
+ ret = platform_driver_register(&cci_pmu_driver);
+ if (ret)
+ return ret;
+
+ return platform_driver_register(&cci_platform_driver);
}
-#else
+#else /* !CONFIG_ARM_CCI_PMU */
-static int __init cci_pmu_init(void)
+static int __init cci_platform_init(void)
{
return 0;
}
-#endif /* CONFIG_HW_PERF_EVENTS */
+#endif /* CONFIG_ARM_CCI_PMU */
+
+#ifdef CONFIG_ARM_CCI400_PORT_CTRL
+
+#define CCI_PORT_CTRL 0x0
+#define CCI_CTRL_STATUS 0xc
+
+#define CCI_ENABLE_SNOOP_REQ 0x1
+#define CCI_ENABLE_DVM_REQ 0x2
+#define CCI_ENABLE_REQ (CCI_ENABLE_SNOOP_REQ | CCI_ENABLE_DVM_REQ)
+
+enum cci_ace_port_type {
+ ACE_INVALID_PORT = 0x0,
+ ACE_PORT,
+ ACE_LITE_PORT,
+};
+
+struct cci_ace_port {
+ void __iomem *base;
+ unsigned long phys;
+ enum cci_ace_port_type type;
+ struct device_node *dn;
+};
+
+static struct cci_ace_port *ports;
+static unsigned int nb_cci_ports;
struct cpu_port {
u64 mpidr;
}
EXPORT_SYMBOL_GPL(cci_ace_get_port);
-static void __init cci_ace_init_ports(void)
+static void cci_ace_init_ports(void)
{
- int port, ac, cpu;
- u64 hwid;
- const u32 *cell;
- struct device_node *cpun, *cpus;
-
- cpus = of_find_node_by_path("/cpus");
- if (WARN(!cpus, "Missing cpus node, bailing out\n"))
- return;
-
- if (WARN_ON(of_property_read_u32(cpus, "#address-cells", &ac)))
- ac = of_n_addr_cells(cpus);
+ int port, cpu;
+ struct device_node *cpun;
/*
* Port index look-up speeds up the function disabling ports by CPU,
* The stashed index array is initialized for all possible CPUs
* at probe time.
*/
- for_each_child_of_node(cpus, cpun) {
- if (of_node_cmp(cpun->type, "cpu"))
- continue;
- cell = of_get_property(cpun, "reg", NULL);
- if (WARN(!cell, "%s: missing reg property\n", cpun->full_name))
- continue;
-
- hwid = of_read_number(cell, ac);
- cpu = get_logical_index(hwid & MPIDR_HWID_BITMASK);
+ for_each_possible_cpu(cpu) {
+ /* too early to use cpu->of_node */
+ cpun = of_get_cpu_node(cpu, NULL);
- if (cpu < 0 || !cpu_possible(cpu))
+ if (WARN(!cpun, "Missing cpu device node\n"))
continue;
+
port = __cci_ace_get_port(cpun, ACE_PORT);
if (port < 0)
continue;
asmlinkage void __naked cci_enable_port_for_self(void)
{
asm volatile ("\n"
-
+" .arch armv7-a\n"
" mrc p15, 0, r0, c0, c0, 5 @ get MPIDR value \n"
" and r0, r0, #"__stringify(MPIDR_HWID_BITMASK)" \n"
" adr r1, 5f \n"
}
EXPORT_SYMBOL_GPL(__cci_control_port_by_index);
-static const struct cci_nb_ports cci400_ports = {
- .nb_ace = 2,
- .nb_ace_lite = 3
-};
-
-static const struct of_device_id arm_cci_matches[] = {
- {.compatible = "arm,cci-400", .data = &cci400_ports },
- {},
-};
-
static const struct of_device_id arm_cci_ctrl_if_matches[] = {
{.compatible = "arm,cci-400-ctrl-if", },
{},
};
-static int __init cci_probe(void)
+static int cci_probe_ports(struct device_node *np)
{
struct cci_nb_ports const *cci_config;
int ret, i, nb_ace = 0, nb_ace_lite = 0;
- struct device_node *np, *cp;
+ struct device_node *cp;
struct resource res;
const char *match_str;
bool is_ace;
- np = of_find_matching_node(NULL, arm_cci_matches);
- if (!np)
- return -ENODEV;
cci_config = of_match_node(arm_cci_matches, np)->data;
if (!cci_config)
nb_cci_ports = cci_config->nb_ace + cci_config->nb_ace_lite;
- ports = kcalloc(sizeof(*ports), nb_cci_ports, GFP_KERNEL);
+ ports = kcalloc(nb_cci_ports, sizeof(*ports), GFP_KERNEL);
if (!ports)
return -ENOMEM;
- ret = of_address_to_resource(np, 0, &res);
- if (!ret) {
- cci_ctrl_base = ioremap(res.start, resource_size(&res));
- cci_ctrl_phys = res.start;
- }
- if (ret || !cci_ctrl_base) {
- WARN(1, "unable to ioremap CCI ctrl\n");
- ret = -ENXIO;
- goto memalloc_err;
- }
-
for_each_child_of_node(np, cp) {
if (!of_match_node(arm_cci_ctrl_if_matches, cp))
continue;
sync_cache_w(&cpu_port);
__sync_cache_range_w(ports, sizeof(*ports) * nb_cci_ports);
pr_info("ARM CCI driver probed\n");
+
return 0;
+}
+#else /* !CONFIG_ARM_CCI400_PORT_CTRL */
+static inline int cci_probe_ports(struct device_node *np)
+{
+ return 0;
+}
+#endif /* CONFIG_ARM_CCI400_PORT_CTRL */
-memalloc_err:
+static int cci_probe(void)
+{
+ int ret;
+ struct device_node *np;
+ struct resource res;
+
+ np = of_find_matching_node(NULL, arm_cci_matches);
+ if(!np || !of_device_is_available(np))
+ return -ENODEV;
+
+ ret = of_address_to_resource(np, 0, &res);
+ if (!ret) {
+ cci_ctrl_base = ioremap(res.start, resource_size(&res));
+ cci_ctrl_phys = res.start;
+ }
+ if (ret || !cci_ctrl_base) {
+ WARN(1, "unable to ioremap CCI ctrl\n");
+ return -ENXIO;
+ }
- kfree(ports);
- return ret;
+ return cci_probe_ports(np);
}
static int cci_init_status = -EAGAIN;
static DEFINE_MUTEX(cci_probing);
-static int __init cci_init(void)
+static int cci_init(void)
{
if (cci_init_status != -EAGAIN)
return cci_init_status;
* has been initialized, if not it calls the init function that probes
* the driver and updates the return value.
*/
-bool __init cci_probed(void)
+bool cci_probed(void)
{
return cci_init() == 0;
}
EXPORT_SYMBOL_GPL(cci_probed);
early_initcall(cci_init);
-core_initcall(cci_pmu_init);
+core_initcall(cci_platform_init);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("ARM CCI support");