diff options
Diffstat (limited to 'drivers/perf')
| -rw-r--r-- | drivers/perf/Kconfig | 33 | ||||
| -rw-r--r-- | drivers/perf/Makefile | 2 | ||||
| -rw-r--r-- | drivers/perf/arm-cci.c | 1722 | ||||
| -rw-r--r-- | drivers/perf/arm-ccn.c | 1597 |
4 files changed, 3354 insertions, 0 deletions
diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig index da5724cd89cf..28bb5a029558 100644 --- a/drivers/perf/Kconfig +++ b/drivers/perf/Kconfig @@ -5,6 +5,39 @@ menu "Performance monitor support" depends on PERF_EVENTS +config ARM_CCI_PMU + bool + select ARM_CCI + +config ARM_CCI400_PMU + bool "ARM CCI400 PMU support" + depends on (ARM && CPU_V7) || ARM64 + select ARM_CCI400_COMMON + select ARM_CCI_PMU + help + Support for PMU events monitoring on the ARM CCI-400 (cache coherent + interconnect). CCI-400 supports counting events related to the + connected slave/master interfaces. + +config ARM_CCI5xx_PMU + bool "ARM CCI-500/CCI-550 PMU support" + depends on (ARM && CPU_V7) || ARM64 + select ARM_CCI_PMU + help + Support for PMU events monitoring on the ARM CCI-500/CCI-550 cache + coherent interconnects. Both of them provide 8 independent event counters, + which can count events pertaining to the slave/master interfaces as well + as the internal events to the CCI. + + If unsure, say Y + +config ARM_CCN + tristate "ARM CCN driver support" + depends on ARM || ARM64 + help + PMU (perf) driver supporting the ARM CCN (Cache Coherent Network) + interconnect. + config ARM_PMU depends on ARM || ARM64 bool "ARM PMU framework" diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile index c2f27419bdf0..b3902bd37d53 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -1,4 +1,6 @@ # SPDX-License-Identifier: GPL-2.0 +obj-$(CONFIG_ARM_CCI_PMU) += arm-cci.o +obj-$(CONFIG_ARM_CCN) += arm-ccn.o obj-$(CONFIG_ARM_DSU_PMU) += arm_dsu_pmu.o obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o diff --git a/drivers/perf/arm-cci.c b/drivers/perf/arm-cci.c new file mode 100644 index 000000000000..383b2d3dcbc6 --- /dev/null +++ b/drivers/perf/arm-cci.c @@ -0,0 +1,1722 @@ +// SPDX-License-Identifier: GPL-2.0 +// CCI Cache Coherent Interconnect PMU driver +// Copyright (C) 2013-2018 Arm Ltd. +// Author: Punit Agrawal <punit.agrawal@arm.com>, Suzuki Poulose <suzuki.poulose@arm.com> + +#include <linux/arm-cci.h> +#include <linux/io.h> +#include <linux/interrupt.h> +#include <linux/module.h> +#include <linux/of_address.h> +#include <linux/of_device.h> +#include <linux/of_irq.h> +#include <linux/of_platform.h> +#include <linux/perf_event.h> +#include <linux/platform_device.h> +#include <linux/slab.h> +#include <linux/spinlock.h> + +#define DRIVER_NAME "ARM-CCI PMU" + +#define CCI_PMCR 0x0100 +#define CCI_PID2 0x0fe8 + +#define CCI_PMCR_CEN 0x00000001 +#define CCI_PMCR_NCNT_MASK 0x0000f800 +#define CCI_PMCR_NCNT_SHIFT 11 + +#define CCI_PID2_REV_MASK 0xf0 +#define CCI_PID2_REV_SHIFT 4 + +#define CCI_PMU_EVT_SEL 0x000 +#define CCI_PMU_CNTR 0x004 +#define CCI_PMU_CNTR_CTRL 0x008 +#define CCI_PMU_OVRFLW 0x00c + +#define CCI_PMU_OVRFLW_FLAG 1 + +#define CCI_PMU_CNTR_SIZE(model) ((model)->cntr_size) +#define CCI_PMU_CNTR_BASE(model, idx) ((idx) * CCI_PMU_CNTR_SIZE(model)) +#define CCI_PMU_CNTR_MASK ((1ULL << 32) -1) +#define CCI_PMU_CNTR_LAST(cci_pmu) (cci_pmu->num_cntrs - 1) + +#define CCI_PMU_MAX_HW_CNTRS(model) \ + ((model)->num_hw_cntrs + (model)->fixed_hw_cntrs) + +/* Types of interfaces that can generate events */ +enum { + CCI_IF_SLAVE, + CCI_IF_MASTER, +#ifdef CONFIG_ARM_CCI5xx_PMU + CCI_IF_GLOBAL, +#endif + CCI_IF_MAX, +}; + +struct event_range { + u32 min; + u32 max; +}; + +struct cci_pmu_hw_events { + struct perf_event **events; + unsigned long *used_mask; + raw_spinlock_t pmu_lock; +}; + +struct cci_pmu; +/* + * struct cci_pmu_model: + * @fixed_hw_cntrs - Number of fixed event counters + * @num_hw_cntrs - Maximum number of programmable event counters + * @cntr_size - Size of an event counter mapping + */ +struct cci_pmu_model { + char *name; + u32 fixed_hw_cntrs; + u32 num_hw_cntrs; + u32 cntr_size; + struct attribute **format_attrs; + struct attribute **event_attrs; + struct event_range event_ranges[CCI_IF_MAX]; + int (*validate_hw_event)(struct cci_pmu *, unsigned long); + int (*get_event_idx)(struct cci_pmu *, struct cci_pmu_hw_events *, unsigned long); + void (*write_counters)(struct cci_pmu *, unsigned long *); +}; + +static struct cci_pmu_model cci_pmu_models[]; + +struct cci_pmu { + void __iomem *base; + void __iomem *ctrl_base; + struct pmu pmu; + int cpu; + int nr_irqs; + int *irqs; + unsigned long active_irqs; + const struct cci_pmu_model *model; + struct cci_pmu_hw_events hw_events; + struct platform_device *plat_device; + int num_cntrs; + atomic_t active_events; + struct mutex reserve_mutex; +}; + +#define to_cci_pmu(c) (container_of(c, struct cci_pmu, pmu)) + +static struct cci_pmu *g_cci_pmu; + +enum cci_models { +#ifdef CONFIG_ARM_CCI400_PMU + CCI400_R0, + CCI400_R1, +#endif +#ifdef CONFIG_ARM_CCI5xx_PMU + CCI500_R0, + CCI550_R0, +#endif + CCI_MODEL_MAX +}; + +static void pmu_write_counters(struct cci_pmu *cci_pmu, + unsigned long *mask); +static ssize_t 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) \ + &((struct dev_ext_attribute[]) { \ + { __ATTR(_name, S_IRUGO, _func, NULL), (void *)_config } \ + })[0].attr.attr + +#define CCI_FORMAT_EXT_ATTR_ENTRY(_name, _config) \ + CCI_EXT_ATTR_ENTRY(_name, cci_pmu_format_show, (char *)_config) +#define CCI_EVENT_EXT_ATTR_ENTRY(_name, _config) \ + CCI_EXT_ATTR_ENTRY(_name, cci_pmu_event_show, (unsigned long)_config) + +/* CCI400 PMU Specific definitions */ + +#ifdef CONFIG_ARM_CCI400_PMU + +/* Port ids */ +#define CCI400_PORT_S0 0 +#define CCI400_PORT_S1 1 +#define CCI400_PORT_S2 2 +#define CCI400_PORT_S3 3 +#define CCI400_PORT_S4 4 +#define CCI400_PORT_M0 5 +#define CCI400_PORT_M1 6 +#define CCI400_PORT_M2 7 + +#define CCI400_R1_PX 5 + +/* + * Instead of an event id to monitor CCI cycles, a dedicated counter is + * provided. Use 0xff to represent CCI cycles and hope that no future revisions + * make use of this event in hardware. + */ +enum cci400_perf_events { + CCI400_PMU_CYCLES = 0xff +}; + +#define CCI400_PMU_CYCLE_CNTR_IDX 0 +#define CCI400_PMU_CNTR0_IDX 1 + +/* + * CCI PMU event id is an 8-bit value made of two parts - bits 7:5 for one of 8 + * ports and bits 4:0 are event codes. There are different event codes + * associated with each port type. + * + * Additionally, the range of events associated with the port types changed + * between Rev0 and Rev1. + * + * The constants below define the range of valid codes for each port type for + * the different revisions and are used to validate the event to be monitored. + */ + +#define CCI400_PMU_EVENT_MASK 0xffUL +#define CCI400_PMU_EVENT_SOURCE_SHIFT 5 +#define CCI400_PMU_EVENT_SOURCE_MASK 0x7 +#define CCI400_PMU_EVENT_CODE_SHIFT 0 +#define CCI400_PMU_EVENT_CODE_MASK 0x1f +#define CCI400_PMU_EVENT_SOURCE(event) \ + ((event >> CCI400_PMU_EVENT_SOURCE_SHIFT) & \ + CCI400_PMU_EVENT_SOURCE_MASK) +#define CCI400_PMU_EVENT_CODE(event) \ + ((event >> CCI400_PMU_EVENT_CODE_SHIFT) & CCI400_PMU_EVENT_CODE_MASK) + +#define CCI400_R0_SLAVE_PORT_MIN_EV 0x00 +#define CCI400_R0_SLAVE_PORT_MAX_EV 0x13 +#define CCI400_R0_MASTER_PORT_MIN_EV 0x14 +#define CCI400_R0_MASTER_PORT_MAX_EV 0x1a + +#define CCI400_R1_SLAVE_PORT_MIN_EV 0x00 +#define CCI400_R1_SLAVE_PORT_MAX_EV 0x14 +#define CCI400_R1_MASTER_PORT_MIN_EV 0x00 +#define CCI400_R1_MASTER_PORT_MAX_EV 0x11 + +#define CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(_name, _config) \ + CCI_EXT_ATTR_ENTRY(_name, cci400_pmu_cycle_event_show, \ + (unsigned long)_config) + +static ssize_t cci400_pmu_cycle_event_show(struct device *dev, + struct device_attribute *attr, char *buf); + +static struct attribute *cci400_pmu_format_attrs[] = { + CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"), + CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-7"), + NULL +}; + +static struct attribute *cci400_r0_pmu_event_attrs[] = { + /* Slave events */ + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_any, 0x0), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_device, 0x01), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_normal_or_nonshareable, 0x2), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_inner_or_outershareable, 0x3), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maintenance, 0x4), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_mem_barrier, 0x5), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_sync_barrier, 0x6), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg_sync, 0x8), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_tt_full, 0x9), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_last_hs_snoop, 0xA), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall_rvalids_h_rready_l, 0xB), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_any, 0xC), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_device, 0xD), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_normal_or_nonshareable, 0xE), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_inner_or_outershare_wback_wclean, 0xF), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_unique, 0x10), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_line_unique, 0x11), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_evict, 0x12), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall_tt_full, 0x13), + /* Master events */ + CCI_EVENT_EXT_ATTR_ENTRY(mi_retry_speculative_fetch, 0x14), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_addr_hazard, 0x15), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_id_hazard, 0x16), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_tt_full, 0x17), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_barrier_hazard, 0x18), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_barrier_hazard, 0x19), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_tt_full, 0x1A), + /* Special event for cycles counter */ + CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(cycles, 0xff), + NULL +}; + +static struct attribute *cci400_r1_pmu_event_attrs[] = { + /* Slave events */ + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_any, 0x0), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_device, 0x01), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_normal_or_nonshareable, 0x2), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_inner_or_outershareable, 0x3), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maintenance, 0x4), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_mem_barrier, 0x5), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_sync_barrier, 0x6), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg_sync, 0x8), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_tt_full, 0x9), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_last_hs_snoop, 0xA), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall_rvalids_h_rready_l, 0xB), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_any, 0xC), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_device, 0xD), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_normal_or_nonshareable, 0xE), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_inner_or_outershare_wback_wclean, 0xF), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_unique, 0x10), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_write_line_unique, 0x11), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_evict, 0x12), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall_tt_full, 0x13), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_slave_id_hazard, 0x14), + /* Master events */ + CCI_EVENT_EXT_ATTR_ENTRY(mi_retry_speculative_fetch, 0x0), + CCI_EVENT_EXT_ATTR_ENTRY(mi_stall_cycle_addr_hazard, 0x1), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_master_id_hazard, 0x2), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_hi_prio_rtq_full, 0x3), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_barrier_hazard, 0x4), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_barrier_hazard, 0x5), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_wtq_full, 0x6), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_low_prio_rtq_full, 0x7), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_mid_prio_rtq_full, 0x8), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn0, 0x9), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn1, 0xA), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn2, 0xB), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall_qvn_vn3, 0xC), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn0, 0xD), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn1, 0xE), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn2, 0xF), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall_qvn_vn3, 0x10), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_unique_or_line_unique_addr_hazard, 0x11), + /* Special event for cycles counter */ + CCI400_CYCLE_EVENT_EXT_ATTR_ENTRY(cycles, 0xff), + NULL +}; + +static ssize_t cci400_pmu_cycle_event_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *eattr = container_of(attr, + struct dev_ext_attribute, attr); + return snprintf(buf, PAGE_SIZE, "config=0x%lx\n", (unsigned long)eattr->var); +} + +static int cci400_get_event_idx(struct cci_pmu *cci_pmu, + struct cci_pmu_hw_events *hw, + unsigned long cci_event) +{ + int idx; + + /* cycles event idx is fixed */ + if (cci_event == CCI400_PMU_CYCLES) { + if (test_and_set_bit(CCI400_PMU_CYCLE_CNTR_IDX, hw->used_mask)) + return -EAGAIN; + + return CCI400_PMU_CYCLE_CNTR_IDX; + } + + for (idx = CCI400_PMU_CNTR0_IDX; idx <= CCI_PMU_CNTR_LAST(cci_pmu); ++idx) + if (!test_and_set_bit(idx, hw->used_mask)) + return idx; + + /* No counters available */ + return -EAGAIN; +} + +static int cci400_validate_hw_event(struct cci_pmu *cci_pmu, unsigned long hw_event) +{ + u8 ev_source = CCI400_PMU_EVENT_SOURCE(hw_event); + u8 ev_code = CCI400_PMU_EVENT_CODE(hw_event); + int if_type; + + if (hw_event & ~CCI400_PMU_EVENT_MASK) + return -ENOENT; + + if (hw_event == CCI400_PMU_CYCLES) + return hw_event; + + switch (ev_source) { + case CCI400_PORT_S0: + case CCI400_PORT_S1: + case CCI400_PORT_S2: + case CCI400_PORT_S3: + case CCI400_PORT_S4: + /* Slave Interface */ + if_type = CCI_IF_SLAVE; + break; + case CCI400_PORT_M0: + case CCI400_PORT_M1: + case CCI400_PORT_M2: + /* Master Interface */ + if_type = CCI_IF_MASTER; + break; + default: + return -ENOENT; + } + + if (ev_code >= cci_pmu->model->event_ranges[if_type].min && + ev_code <= cci_pmu->model->event_ranges[if_type].max) + return hw_event; + + return -ENOENT; +} + +static int probe_cci400_revision(struct cci_pmu *cci_pmu) +{ + int rev; + rev = readl_relaxed(cci_pmu->ctrl_base + CCI_PID2) & CCI_PID2_REV_MASK; + rev >>= CCI_PID2_REV_SHIFT; + + if (rev < CCI400_R1_PX) + return CCI400_R0; + else + return CCI400_R1; +} + +static const struct cci_pmu_model *probe_cci_model(struct cci_pmu *cci_pmu) +{ + if (platform_has_secure_cci_access()) + return &cci_pmu_models[probe_cci400_revision(cci_pmu)]; + return NULL; +} +#else /* !CONFIG_ARM_CCI400_PMU */ +static inline struct cci_pmu_model *probe_cci_model(struct cci_pmu *cci_pmu) +{ + return NULL; +} +#endif /* CONFIG_ARM_CCI400_PMU */ + +#ifdef CONFIG_ARM_CCI5xx_PMU + +/* + * CCI5xx PMU event id is an 9-bit value made of two parts. + * bits [8:5] - Source for the event + * bits [4:0] - Event code (specific to type of interface) + * + * + */ + +/* Port ids */ +#define CCI5xx_PORT_S0 0x0 +#define CCI5xx_PORT_S1 0x1 +#define CCI5xx_PORT_S2 0x2 +#define CCI5xx_PORT_S3 0x3 +#define CCI5xx_PORT_S4 0x4 +#define CCI5xx_PORT_S5 0x5 +#define CCI5xx_PORT_S6 0x6 + +#define CCI5xx_PORT_M0 0x8 +#define CCI5xx_PORT_M1 0x9 +#define CCI5xx_PORT_M2 0xa +#define CCI5xx_PORT_M3 0xb +#define CCI5xx_PORT_M4 0xc +#define CCI5xx_PORT_M5 0xd +#define CCI5xx_PORT_M6 0xe + +#define CCI5xx_PORT_GLOBAL 0xf + +#define CCI5xx_PMU_EVENT_MASK 0x1ffUL +#define CCI5xx_PMU_EVENT_SOURCE_SHIFT 0x5 +#define CCI5xx_PMU_EVENT_SOURCE_MASK 0xf +#define CCI5xx_PMU_EVENT_CODE_SHIFT 0x0 +#define CCI5xx_PMU_EVENT_CODE_MASK 0x1f + +#define CCI5xx_PMU_EVENT_SOURCE(event) \ + ((event >> CCI5xx_PMU_EVENT_SOURCE_SHIFT) & CCI5xx_PMU_EVENT_SOURCE_MASK) +#define CCI5xx_PMU_EVENT_CODE(event) \ + ((event >> CCI5xx_PMU_EVENT_CODE_SHIFT) & CCI5xx_PMU_EVENT_CODE_MASK) + +#define CCI5xx_SLAVE_PORT_MIN_EV 0x00 +#define CCI5xx_SLAVE_PORT_MAX_EV 0x1f +#define CCI5xx_MASTER_PORT_MIN_EV 0x00 +#define CCI5xx_MASTER_PORT_MAX_EV 0x06 +#define CCI5xx_GLOBAL_PORT_MIN_EV 0x00 +#define CCI5xx_GLOBAL_PORT_MAX_EV 0x0f + + +#define CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(_name, _config) \ + CCI_EXT_ATTR_ENTRY(_name, cci5xx_pmu_global_event_show, \ + (unsigned long) _config) + +static ssize_t cci5xx_pmu_global_event_show(struct device *dev, + struct device_attribute *attr, char *buf); + +static struct attribute *cci5xx_pmu_format_attrs[] = { + CCI_FORMAT_EXT_ATTR_ENTRY(event, "config:0-4"), + CCI_FORMAT_EXT_ATTR_ENTRY(source, "config:5-8"), + NULL, +}; + +static struct attribute *cci5xx_pmu_event_attrs[] = { + /* Slave events */ + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_arvalid, 0x0), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_dev, 0x1), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_nonshareable, 0x2), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_shareable_non_alloc, 0x3), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_shareable_alloc, 0x4), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_invalidate, 0x5), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_cache_maint, 0x6), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_dvm_msg, 0x7), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_rval, 0x8), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_hs_rlast_snoop, 0x9), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_hs_awalid, 0xA), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_dev, 0xB), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_non_shareable, 0xC), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wb, 0xD), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wlu, 0xE), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_share_wunique, 0xF), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_evict, 0x10), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_wrevict, 0x11), + CCI_EVENT_EXT_ATTR_ENTRY(si_w_data_beat, 0x12), + CCI_EVENT_EXT_ATTR_ENTRY(si_srq_acvalid, 0x13), + CCI_EVENT_EXT_ATTR_ENTRY(si_srq_read, 0x14), + CCI_EVENT_EXT_ATTR_ENTRY(si_srq_clean, 0x15), + CCI_EVENT_EXT_ATTR_ENTRY(si_srq_data_transfer_low, 0x16), + CCI_EVENT_EXT_ATTR_ENTRY(si_rrq_stall_arvalid, 0x17), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_data_stall, 0x18), + CCI_EVENT_EXT_ATTR_ENTRY(si_wrq_stall, 0x19), + CCI_EVENT_EXT_ATTR_ENTRY(si_w_data_stall, 0x1A), + CCI_EVENT_EXT_ATTR_ENTRY(si_w_resp_stall, 0x1B), + CCI_EVENT_EXT_ATTR_ENTRY(si_srq_stall, 0x1C), + CCI_EVENT_EXT_ATTR_ENTRY(si_s_data_stall, 0x1D), + CCI_EVENT_EXT_ATTR_ENTRY(si_rq_stall_ot_limit, 0x1E), + CCI_EVENT_EXT_ATTR_ENTRY(si_r_stall_arbit, 0x1F), + + /* Master events */ + CCI_EVENT_EXT_ATTR_ENTRY(mi_r_data_beat_any, 0x0), + CCI_EVENT_EXT_ATTR_ENTRY(mi_w_data_beat_any, 0x1), + CCI_EVENT_EXT_ATTR_ENTRY(mi_rrq_stall, 0x2), + CCI_EVENT_EXT_ATTR_ENTRY(mi_r_data_stall, 0x3), + CCI_EVENT_EXT_ATTR_ENTRY(mi_wrq_stall, 0x4), + CCI_EVENT_EXT_ATTR_ENTRY(mi_w_data_stall, 0x5), + CCI_EVENT_EXT_ATTR_ENTRY(mi_w_resp_stall, 0x6), + + /* Global events */ + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_0_1, 0x0), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_2_3, 0x1), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_4_5, 0x2), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_filter_bank_6_7, 0x3), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_0_1, 0x4), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_2_3, 0x5), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_4_5, 0x6), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_access_miss_filter_bank_6_7, 0x7), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_back_invalidation, 0x8), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_alloc_busy, 0x9), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_stall_tt_full, 0xA), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_wrq, 0xB), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_cd_hs, 0xC), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_rq_stall_addr_hazard, 0xD), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_rq_stall_tt_full, 0xE), + CCI5xx_GLOBAL_EVENT_EXT_ATTR_ENTRY(cci_snoop_rq_tzmp1_prot, 0xF), + NULL +}; + +static ssize_t cci5xx_pmu_global_event_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *eattr = container_of(attr, + struct dev_ext_attribute, attr); + /* Global events have single fixed source code */ + return snprintf(buf, PAGE_SIZE, "event=0x%lx,source=0x%x\n", + (unsigned long)eattr->var, CCI5xx_PORT_GLOBAL); +} + +/* + * CCI500 provides 8 independent event counters that can count + * any of the events available. + * CCI500 PMU event source ids + * 0x0-0x6 - Slave interfaces + * 0x8-0xD - Master interfaces + * 0xf - Global Events + * 0x7,0xe - Reserved + */ +static int cci500_validate_hw_event(struct cci_pmu *cci_pmu, + unsigned long hw_event) +{ + u32 ev_source = CCI5xx_PMU_EVENT_SOURCE(hw_event); + u32 ev_code = CCI5xx_PMU_EVENT_CODE(hw_event); + int if_type; + + if (hw_event & ~CCI5xx_PMU_EVENT_MASK) + return -ENOENT; + + switch (ev_source) { + case CCI5xx_PORT_S0: + case CCI5xx_PORT_S1: + case CCI5xx_PORT_S2: + case CCI5xx_PORT_S3: + case CCI5xx_PORT_S4: + case CCI5xx_PORT_S5: + case CCI5xx_PORT_S6: + if_type = CCI_IF_SLAVE; + break; + case CCI5xx_PORT_M0: + case CCI5xx_PORT_M1: + case CCI5xx_PORT_M2: + case CCI5xx_PORT_M3: + case CCI5xx_PORT_M4: + case CCI5xx_PORT_M5: + if_type = CCI_IF_MASTER; + break; + case CCI5xx_PORT_GLOBAL: + if_type = CCI_IF_GLOBAL; + break; + default: + return -ENOENT; + } + + if (ev_code >= cci_pmu->model->event_ranges[if_type].min && + ev_code <= cci_pmu->model->event_ranges[if_type].max) + return hw_event; + + return -ENOENT; +} + +/* + * CCI550 provides 8 independent event counters that can count + * any of the events available. + * CCI550 PMU event source ids + * 0x0-0x6 - Slave interfaces + * 0x8-0xe - Master interfaces + * 0xf - Global Events + * 0x7 - Reserved + */ +static int cci550_validate_hw_event(struct cci_pmu *cci_pmu, + unsigned long hw_event) +{ + u32 ev_source = CCI5xx_PMU_EVENT_SOURCE(hw_event); + u32 ev_code = CCI5xx_PMU_EVENT_CODE(hw_event); + int if_type; + + if (hw_event & ~CCI5xx_PMU_EVENT_MASK) + return -ENOENT; + + switch (ev_source) { + case CCI5xx_PORT_S0: + case CCI5xx_PORT_S1: + case CCI5xx_PORT_S2: + case CCI5xx_PORT_S3: + case CCI5xx_PORT_S4: + case CCI5xx_PORT_S5: + case CCI5xx_PORT_S6: + if_type = CCI_IF_SLAVE; + break; + case CCI5xx_PORT_M0: + case CCI5xx_PORT_M1: + case CCI5xx_PORT_M2: + case CCI5xx_PORT_M3: + case CCI5xx_PORT_M4: + case CCI5xx_PORT_M5: + case CCI5xx_PORT_M6: + if_type = CCI_IF_MASTER; + break; + case CCI5xx_PORT_GLOBAL: + if_type = CCI_IF_GLOBAL; + break; + default: + return -ENOENT; + } + + if (ev_code >= cci_pmu->model->event_ranges[if_type].min && + ev_code <= cci_pmu->model->event_ranges[if_type].max) + return hw_event; + + return -ENOENT; +} + +#endif /* CONFIG_ARM_CCI5xx_PMU */ + +/* + * Program the CCI PMU counters which have PERF_HES_ARCH set + * with the event period and mark them ready before we enable + * PMU. + */ +static void cci_pmu_sync_counters(struct cci_pmu *cci_pmu) +{ + int i; + struct cci_pmu_hw_events *cci_hw = &cci_pmu->hw_events; + + DECLARE_BITMAP(mask, cci_pmu->num_cntrs); + + bitmap_zero(mask, cci_pmu->num_cntrs); + for_each_set_bit(i, cci_pmu->hw_events.used_mask, cci_pmu->num_cntrs) { + struct perf_event *event = cci_hw->events[i]; + + if (WARN_ON(!event)) + continue; + + /* Leave the events which are not counting */ + if (event->hw.state & PERF_HES_STOPPED) + continue; + if (event->hw.state & PERF_HES_ARCH) { + set_bit(i, mask); + event->hw.state &= ~PERF_HES_ARCH; + } + } + + pmu_write_counters(cci_pmu, mask); +} + +/* Should be called with cci_pmu->hw_events->pmu_lock held */ +static void __cci_pmu_enable_nosync(struct cci_pmu *cci_pmu) +{ + u32 val; + + /* Enable all the PMU counters. */ + val = readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) | CCI_PMCR_CEN; + writel(val, cci_pmu->ctrl_base + CCI_PMCR); +} + +/* Should be called with cci_pmu->hw_events->pmu_lock held */ +static void __cci_pmu_enable_sync(struct cci_pmu *cci_pmu) +{ + cci_pmu_sync_counters(cci_pmu); + __cci_pmu_enable_nosync(cci_pmu); +} + +/* Should be called with cci_pmu->hw_events->pmu_lock held */ +static void __cci_pmu_disable(struct cci_pmu *cci_pmu) +{ + u32 val; + + /* Disable all the PMU counters. */ + val = readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) & ~CCI_PMCR_CEN; + writel(val, cci_pmu->ctrl_base + CCI_PMCR); +} + +static ssize_t cci_pmu_format_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *eattr = container_of(attr, + struct dev_ext_attribute, attr); + return snprintf(buf, PAGE_SIZE, "%s\n", (char *)eattr->var); +} + +static ssize_t cci_pmu_event_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *eattr = container_of(attr, + struct dev_ext_attribute, attr); + /* source parameter is mandatory for normal PMU events */ + return snprintf(buf, PAGE_SIZE, "source=?,event=0x%lx\n", + (unsigned long)eattr->var); +} + +static int pmu_is_valid_counter(struct cci_pmu *cci_pmu, int idx) +{ + return 0 <= idx && idx <= CCI_PMU_CNTR_LAST(cci_pmu); +} + +static u32 pmu_read_register(struct cci_pmu *cci_pmu, int idx, unsigned int offset) +{ + return readl_relaxed(cci_pmu->base + + CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset); +} + +static void pmu_write_register(struct cci_pmu *cci_pmu, u32 value, + int idx, unsigned int offset) +{ + writel_relaxed(value, cci_pmu->base + + CCI_PMU_CNTR_BASE(cci_pmu->model, idx) + offset); +} + +static void pmu_disable_counter(struct cci_pmu *cci_pmu, int idx) +{ + pmu_write_register(cci_pmu, 0, idx, CCI_PMU_CNTR_CTRL); +} + +static void pmu_enable_counter(struct cci_pmu *cci_pmu, int idx) +{ + pmu_write_register(cci_pmu, 1, idx, CCI_PMU_CNTR_CTRL); +} + +static bool __maybe_unused +pmu_counter_is_enabled(struct cci_pmu *cci_pmu, int idx) +{ + return (pmu_read_register(cci_pmu, idx, CCI_PMU_CNTR_CTRL) & 0x1) != 0; +} + +static void pmu_set_event(struct cci_pmu *cci_pmu, int idx, unsigned long event) +{ + pmu_write_register(cci_pmu, event, idx, CCI_PMU_EVT_SEL); +} + +/* + * For all counters on the CCI-PMU, disable any 'enabled' counters, + * saving the changed counters in the mask, so that we can restore + * it later using pmu_restore_counters. The mask is private to the + * caller. We cannot rely on the used_mask maintained by the CCI_PMU + * as it only tells us if the counter is assigned to perf_event or not. + * The state of the perf_event cannot be locked by the PMU layer, hence + * we check the individual counter status (which can be locked by + * cci_pm->hw_events->pmu_lock). + * + * @mask should be initialised to empty by the caller. + */ +static void __maybe_unused +pmu_save_counters(struct cci_pmu *cci_pmu, unsigned long *mask) +{ + int i; + + for (i = 0; i < cci_pmu->num_cntrs; i++) { + if (pmu_counter_is_enabled(cci_pmu, i)) { + set_bit(i, mask); + pmu_disable_counter(cci_pmu, i); + } + } +} + +/* + * Restore the status of the counters. Reversal of the pmu_save_counters(). + * For each counter set in the mask, enable the counter back. + */ +static void __maybe_unused +pmu_restore_counters(struct cci_pmu *cci_pmu, unsigned long *mask) +{ + int i; + + for_each_set_bit(i, mask, cci_pmu->num_cntrs) + pmu_enable_counter(cci_pmu, i); +} + +/* + * Returns the number of programmable counters actually implemented + * by the cci + */ +static u32 pmu_get_max_counters(struct cci_pmu *cci_pmu) +{ + return (readl_relaxed(cci_pmu->ctrl_base + CCI_PMCR) & + CCI_PMCR_NCNT_MASK) >> CCI_PMCR_NCNT_SHIFT; +} + +static int pmu_get_event_idx(struct cci_pmu_hw_events *hw, struct perf_event *event) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + unsigned long cci_event = event->hw.config_base; + int idx; + + if (cci_pmu->model->get_event_idx) + return cci_pmu->model->get_event_idx(cci_pmu, hw, cci_event); + + /* Generic code to find an unused idx from the mask */ + for(idx = 0; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++) + if (!test_and_set_bit(idx, hw->used_mask)) + return idx; + + /* No counters available */ + return -EAGAIN; +} + +static int pmu_map_event(struct perf_event *event) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + + if (event->attr.type < PERF_TYPE_MAX || + !cci_pmu->model->validate_hw_event) + return -ENOENT; + + return cci_pmu->model->validate_hw_event(cci_pmu, event->attr.config); +} + +static int pmu_request_irq(struct cci_pmu *cci_pmu, irq_handler_t handler) +{ + int i; + struct platform_device *pmu_device = cci_pmu->plat_device; + + if (unlikely(!pmu_device)) + return -ENODEV; + + if (cci_pmu->nr_irqs < 1) { + dev_err(&pmu_device->dev, "no irqs for CCI PMUs defined\n"); + return -ENODEV; + } + + /* + * Register all available CCI PMU interrupts. In the interrupt handler + * we iterate over the counters checking for interrupt source (the + * overflowing counter) and clear it. + * + * This should allow handling of non-unique interrupt for the counters. + */ + for (i = 0; i < cci_pmu->nr_irqs; i++) { + int err = request_irq(cci_pmu->irqs[i], handler, IRQF_SHARED, + "arm-cci-pmu", cci_pmu); + if (err) { + dev_err(&pmu_device->dev, "unable to request IRQ%d for ARM CCI PMU counters\n", + cci_pmu->irqs[i]); + return err; + } + + set_bit(i, &cci_pmu->active_irqs); + } + + return 0; +} + +static void pmu_free_irq(struct cci_pmu *cci_pmu) +{ + int i; + + for (i = 0; i < cci_pmu->nr_irqs; i++) { + if (!test_and_clear_bit(i, &cci_pmu->active_irqs)) + continue; + + free_irq(cci_pmu->irqs[i], cci_pmu); + } +} + +static u32 pmu_read_counter(struct perf_event *event) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + struct hw_perf_event *hw_counter = &event->hw; + int idx = hw_counter->idx; + u32 value; + + if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) { + dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx); + return 0; + } + value = pmu_read_register(cci_pmu, idx, CCI_PMU_CNTR); + + return value; +} + +static void pmu_write_counter(struct cci_pmu *cci_pmu, u32 value, int idx) +{ + pmu_write_register(cci_pmu, value, idx, CCI_PMU_CNTR); +} + +static void __pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask) +{ + int i; + struct cci_pmu_hw_events *cci_hw = &cci_pmu->hw_events; + + for_each_set_bit(i, mask, cci_pmu->num_cntrs) { + struct perf_event *event = cci_hw->events[i]; + + if (WARN_ON(!event)) + continue; + pmu_write_counter(cci_pmu, local64_read(&event->hw.prev_count), i); + } +} + +static void pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask) +{ + if (cci_pmu->model->write_counters) + cci_pmu->model->write_counters(cci_pmu, mask); + else + __pmu_write_counters(cci_pmu, mask); +} + +#ifdef CONFIG_ARM_CCI5xx_PMU + +/* + * CCI-500/CCI-550 has advanced power saving policies, which could gate the + * clocks to the PMU counters, which makes the writes to them ineffective. + * The only way to write to those counters is when the global counters + * are enabled and the particular counter is enabled. + * + * So we do the following : + * + * 1) Disable all the PMU counters, saving their current state + * 2) Enable the global PMU profiling, now that all counters are + * disabled. + * + * For each counter to be programmed, repeat steps 3-7: + * + * 3) Write an invalid event code to the event control register for the + counter, so that the counters are not modified. + * 4) Enable the counter control for the counter. + * 5) Set the counter value + * 6) Disable the counter + * 7) Restore the event in the target counter + * + * 8) Disable the global PMU. + * 9) Restore the status of the rest of the counters. + * + * We choose an event which for CCI-5xx is guaranteed not to count. + * We use the highest possible event code (0x1f) for the master interface 0. + */ +#define CCI5xx_INVALID_EVENT ((CCI5xx_PORT_M0 << CCI5xx_PMU_EVENT_SOURCE_SHIFT) | \ + (CCI5xx_PMU_EVENT_CODE_MASK << CCI5xx_PMU_EVENT_CODE_SHIFT)) +static void cci5xx_pmu_write_counters(struct cci_pmu *cci_pmu, unsigned long *mask) +{ + int i; + DECLARE_BITMAP(saved_mask, cci_pmu->num_cntrs); + + bitmap_zero(saved_mask, cci_pmu->num_cntrs); + pmu_save_counters(cci_pmu, saved_mask); + + /* + * Now that all the counters are disabled, we can safely turn the PMU on, + * without syncing the status of the counters + */ + __cci_pmu_enable_nosync(cci_pmu); + + for_each_set_bit(i, mask, cci_pmu->num_cntrs) { + struct perf_event *event = cci_pmu->hw_events.events[i]; + + if (WARN_ON(!event)) + continue; + + pmu_set_event(cci_pmu, i, CCI5xx_INVALID_EVENT); + pmu_enable_counter(cci_pmu, i); + pmu_write_counter(cci_pmu, local64_read(&event->hw.prev_count), i); + pmu_disable_counter(cci_pmu, i); + pmu_set_event(cci_pmu, i, event->hw.config_base); + } + + __cci_pmu_disable(cci_pmu); + + pmu_restore_counters(cci_pmu, saved_mask); +} + +#endif /* CONFIG_ARM_CCI5xx_PMU */ + +static u64 pmu_event_update(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + u64 delta, prev_raw_count, new_raw_count; + + do { + prev_raw_count = local64_read(&hwc->prev_count); + new_raw_count = pmu_read_counter(event); + } while (local64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count); + + delta = (new_raw_count - prev_raw_count) & CCI_PMU_CNTR_MASK; + + local64_add(delta, &event->count); + + return new_raw_count; +} + +static void pmu_read(struct perf_event *event) +{ + pmu_event_update(event); +} + +static void pmu_event_set_period(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + /* + * The CCI PMU counters have a period of 2^32. To account for the + * possiblity of extreme interrupt latency we program for a period of + * half that. Hopefully we can handle the interrupt before another 2^31 + * events occur and the counter overtakes its previous value. + */ + u64 val = 1ULL << 31; + local64_set(&hwc->prev_count, val); + + /* + * CCI PMU uses PERF_HES_ARCH to keep track of the counters, whose + * values needs to be sync-ed with the s/w state before the PMU is + * enabled. + * Mark this counter for sync. + */ + hwc->state |= PERF_HES_ARCH; +} + +static irqreturn_t pmu_handle_irq(int irq_num, void *dev) +{ + unsigned long flags; + struct cci_pmu *cci_pmu = dev; + struct cci_pmu_hw_events *events = &cci_pmu->hw_events; + int idx, handled = IRQ_NONE; + + raw_spin_lock_irqsave(&events->pmu_lock, flags); + + /* Disable the PMU while we walk through the counters */ + __cci_pmu_disable(cci_pmu); + /* + * Iterate over counters and update the corresponding perf events. + * This should work regardless of whether we have per-counter overflow + * interrupt or a combined overflow interrupt. + */ + for (idx = 0; idx <= CCI_PMU_CNTR_LAST(cci_pmu); idx++) { + struct perf_event *event = events->events[idx]; + + if (!event) + continue; + + /* Did this counter overflow? */ + if (!(pmu_read_register(cci_pmu, idx, CCI_PMU_OVRFLW) & + CCI_PMU_OVRFLW_FLAG)) + continue; + + pmu_write_register(cci_pmu, CCI_PMU_OVRFLW_FLAG, idx, + CCI_PMU_OVRFLW); + + pmu_event_update(event); + pmu_event_set_period(event); + handled = IRQ_HANDLED; + } + + /* Enable the PMU and sync possibly overflowed counters */ + __cci_pmu_enable_sync(cci_pmu); + raw_spin_unlock_irqrestore(&events->pmu_lock, flags); + + return IRQ_RETVAL(handled); +} + +static int cci_pmu_get_hw(struct cci_pmu *cci_pmu) +{ + int ret = pmu_request_irq(cci_pmu, pmu_handle_irq); + if (ret) { + pmu_free_irq(cci_pmu); + return ret; + } + return 0; +} + +static void cci_pmu_put_hw(struct cci_pmu *cci_pmu) +{ + pmu_free_irq(cci_pmu); +} + +static void hw_perf_event_destroy(struct perf_event *event) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + atomic_t *active_events = &cci_pmu->active_events; + struct mutex *reserve_mutex = &cci_pmu->reserve_mutex; + + if (atomic_dec_and_mutex_lock(active_events, reserve_mutex)) { + cci_pmu_put_hw(cci_pmu); + mutex_unlock(reserve_mutex); + } +} + +static void cci_pmu_enable(struct pmu *pmu) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(pmu); + struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events; + int enabled = bitmap_weight(hw_events->used_mask, cci_pmu->num_cntrs); + unsigned long flags; + + if (!enabled) + return; + + raw_spin_lock_irqsave(&hw_events->pmu_lock, flags); + __cci_pmu_enable_sync(cci_pmu); + raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags); + +} + +static void cci_pmu_disable(struct pmu *pmu) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(pmu); + struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events; + unsigned long flags; + + raw_spin_lock_irqsave(&hw_events->pmu_lock, flags); + __cci_pmu_disable(cci_pmu); + raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags); +} + +/* + * Check if the idx represents a non-programmable counter. + * All the fixed event counters are mapped before the programmable + * counters. + */ +static bool pmu_fixed_hw_idx(struct cci_pmu *cci_pmu, int idx) +{ + return (idx >= 0) && (idx < cci_pmu->model->fixed_hw_cntrs); +} + +static void cci_pmu_start(struct perf_event *event, int pmu_flags) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events; + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + unsigned long flags; + + /* + * To handle interrupt latency, we always reprogram the period + * regardlesss of PERF_EF_RELOAD. + */ + if (pmu_flags & PERF_EF_RELOAD) + WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE)); + + hwc->state = 0; + + if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) { + dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx); + return; + } + + raw_spin_lock_irqsave(&hw_events->pmu_lock, flags); + + /* Configure the counter unless you are counting a fixed event */ + if (!pmu_fixed_hw_idx(cci_pmu, idx)) + pmu_set_event(cci_pmu, idx, hwc->config_base); + + pmu_event_set_period(event); + pmu_enable_counter(cci_pmu, idx); + + raw_spin_unlock_irqrestore(&hw_events->pmu_lock, flags); +} + +static void cci_pmu_stop(struct perf_event *event, int pmu_flags) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + if (hwc->state & PERF_HES_STOPPED) + return; + + if (unlikely(!pmu_is_valid_counter(cci_pmu, idx))) { + dev_err(&cci_pmu->plat_device->dev, "Invalid CCI PMU counter %d\n", idx); + return; + } + + /* + * We always reprogram the counter, so ignore PERF_EF_UPDATE. See + * cci_pmu_start() + */ + pmu_disable_counter(cci_pmu, idx); + pmu_event_update(event); + hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE; +} + +static int cci_pmu_add(struct perf_event *event, int flags) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events; + struct hw_perf_event *hwc = &event->hw; + int idx; + int err = 0; + + perf_pmu_disable(event->pmu); + + /* 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; + } + + event->hw.idx = idx; + hw_events->events[idx] = event; + + hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE; + if (flags & PERF_EF_START) + cci_pmu_start(event, PERF_EF_RELOAD); + + /* Propagate our changes to the userspace mapping. */ + perf_event_update_userpage(event); + +out: + perf_pmu_enable(event->pmu); + return err; +} + +static void cci_pmu_del(struct perf_event *event, int flags) +{ + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + struct cci_pmu_hw_events *hw_events = &cci_pmu->hw_events; + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + cci_pmu_stop(event, PERF_EF_UPDATE); + hw_events->events[idx] = NULL; + clear_bit(idx, hw_events->used_mask); + + perf_event_update_userpage(event); +} + +static int validate_event(struct pmu *cci_pmu, + struct cci_pmu_hw_events *hw_events, + struct perf_event *event) +{ + if (is_software_event(event)) + return 1; + + /* + * Reject groups spanning multiple HW PMUs (e.g. CPU + CCI). The + * core perf code won't check that the pmu->ctx == leader->ctx + * until after pmu->event_init(event). + */ + if (event->pmu != cci_pmu) + return 0; + + if (event->state < PERF_EVENT_STATE_OFF) + return 1; + + if (event->state == PERF_EVENT_STATE_OFF && !event->attr.enable_on_exec) + return 1; + + return pmu_get_event_idx(hw_events, event) >= 0; +} + +static int validate_group(struct perf_event *event) +{ + struct perf_event *sibling, *leader = event->group_leader; + struct cci_pmu *cci_pmu = to_cci_pmu(event->pmu); + unsigned long mask[BITS_TO_LONGS(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; + + for_each_sibling_event(sibling, leader) { + if (!validate_event(event->pmu, &fake_pmu, sibling)) + return -EINVAL; + } + + if (!validate_event(event->pmu, &fake_pmu, event)) + return -EINVAL; + + return 0; +} + +static int __hw_perf_event_init(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + int mapping; + + mapping = pmu_map_event(event); + + if (mapping < 0) { + pr_debug("event %x:%llx not supported\n", event->attr.type, + event->attr.config); + return mapping; + } + + /* + * We don't assign an index until we actually place the event onto + * hardware. Use -1 to signify that we haven't decided where to put it + * yet. + */ + hwc->idx = -1; + hwc->config_base = 0; + hwc->config = 0; + hwc->event_base = 0; + + /* + * Store the event encoding into the config_base field. + */ + hwc->config_base |= (unsigned long)mapping; + + /* + * 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; + + 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. + */ + if (event->cpu < 0) + return -EINVAL; + event->cpu = cci_pmu->cpu; + + event->destroy = hw_perf_event_destroy; + if (!atomic_inc_not_zero(active_events)) { + mutex_lock(&cci_pmu->reserve_mutex); + if (atomic_read(active_events) == 0) + err = cci_pmu_get_hw(cci_pmu); + if (!err) + atomic_inc(active_events); + mutex_unlock(&cci_pmu->reserve_mutex); + } + if (err) + return err; + + err = __hw_perf_event_init(event); + if (err) + hw_perf_event_destroy(event); + + return err; +} + +static ssize_t pmu_cpumask_attr_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct pmu *pmu = dev_get_drvdata(dev); + struct cci_pmu *cci_pmu = to_cci_pmu(pmu); + + return cpumap_print_to_pagebuf(true, buf, cpumask_of(cci_pmu->cpu)); +} + +static struct device_attribute pmu_cpumask_attr = + __ATTR(cpumask, S_IRUGO, pmu_cpumask_attr_show, NULL); + +static struct attribute *pmu_attrs[] = { + &pmu_cpumask_attr.attr, + NULL, +}; + +static struct attribute_group pmu_attr_group = { + .attrs = pmu_attrs, +}; + +static struct attribute_group pmu_format_attr_group = { + .name = "format", + .attrs = NULL, /* Filled in cci_pmu_init_attrs */ +}; + +static struct attribute_group pmu_event_attr_group = { + .name = "events", + .attrs = NULL, /* Filled in cci_pmu_init_attrs */ +}; + +static const struct attribute_group *pmu_attr_groups[] = { + &pmu_attr_group, + &pmu_format_attr_group, + &pmu_event_attr_group, + NULL +}; + +static int cci_pmu_init(struct cci_pmu *cci_pmu, struct platform_device *pdev) +{ + const struct cci_pmu_model *model = cci_pmu->model; + char *name = model->name; + u32 num_cntrs; + + pmu_event_attr_group.attrs = model->event_attrs; + pmu_format_attr_group.attrs = model->format_attrs; + + 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(cci_pmu); + if (num_cntrs > cci_pmu->model->num_hw_cntrs) { + dev_warn(&pdev->dev, + "PMU implements more counters(%d) than supported by" + " the model(%d), truncated.", + num_cntrs, cci_pmu->model->num_hw_cntrs); + num_cntrs = cci_pmu->model->num_hw_cntrs; + } + cci_pmu->num_cntrs = num_cntrs + cci_pmu->model->fixed_hw_cntrs; + + return perf_pmu_register(&cci_pmu->pmu, name, -1); +} + +static int cci_pmu_offline_cpu(unsigned int cpu) +{ + int target; + + if (!g_cci_pmu || cpu != g_cci_pmu->cpu) + return 0; + + target = cpumask_any_but(cpu_online_mask, cpu); + if (target >= nr_cpu_ids) + return 0; + + perf_pmu_migrate_context(&g_cci_pmu->pmu, cpu, target); + g_cci_pmu->cpu = target; + return 0; +} + +static 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, + .event_attrs = cci400_r0_pmu_event_attrs, + .event_ranges = { + [CCI_IF_SLAVE] = { + CCI400_R0_SLAVE_PORT_MIN_EV, + CCI400_R0_SLAVE_PORT_MAX_EV, + }, + [CCI_IF_MASTER] = { + CCI400_R0_MASTER_PORT_MIN_EV, + CCI400_R0_MASTER_PORT_MAX_EV, + }, + }, + .validate_hw_event = cci400_validate_hw_event, + .get_event_idx = cci400_get_event_idx, + }, + [CCI400_R1] = { + .name = "CCI_400_r1", + .fixed_hw_cntrs = 1, /* Cycle counter */ + .num_hw_cntrs = 4, + .cntr_size = SZ_4K, + .format_attrs = cci400_pmu_format_attrs, + .event_attrs = cci400_r1_pmu_event_attrs, + .event_ranges = { + [CCI_IF_SLAVE] = { + CCI400_R1_SLAVE_PORT_MIN_EV, + CCI400_R1_SLAVE_PORT_MAX_EV, + }, + [CCI_IF_MASTER] = { + CCI400_R1_MASTER_PORT_MIN_EV, + CCI400_R1_MASTER_PORT_MAX_EV, + }, + }, + .validate_hw_event = cci400_validate_hw_event, + .get_event_idx = cci400_get_event_idx, + }, +#endif +#ifdef CONFIG_ARM_CCI5xx_PMU + [CCI500_R0] = { + .name = "CCI_500", + .fixed_hw_cntrs = 0, + .num_hw_cntrs = 8, + .cntr_size = SZ_64K, + .format_attrs = cci5xx_pmu_format_attrs, + .event_attrs = cci5xx_pmu_event_attrs, + .event_ranges = { + [CCI_IF_SLAVE] = { + CCI5xx_SLAVE_PORT_MIN_EV, + CCI5xx_SLAVE_PORT_MAX_EV, + }, + [CCI_IF_MASTER] = { + CCI5xx_MASTER_PORT_MIN_EV, + CCI5xx_MASTER_PORT_MAX_EV, + }, + [CCI_IF_GLOBAL] = { + CCI5xx_GLOBAL_PORT_MIN_EV, + CCI5xx_GLOBAL_PORT_MAX_EV, + }, + }, + .validate_hw_event = cci500_validate_hw_event, + .write_counters = cci5xx_pmu_write_counters, + }, + [CCI550_R0] = { + .name = "CCI_550", + .fixed_hw_cntrs = 0, + .num_hw_cntrs = 8, + .cntr_size = SZ_64K, + .format_attrs = cci5xx_pmu_format_attrs, + .event_attrs = cci5xx_pmu_event_attrs, + .event_ranges = { + [CCI_IF_SLAVE] = { + CCI5xx_SLAVE_PORT_MIN_EV, + CCI5xx_SLAVE_PORT_MAX_EV, + }, + [CCI_IF_MASTER] = { + CCI5xx_MASTER_PORT_MIN_EV, + CCI5xx_MASTER_PORT_MAX_EV, + }, + [CCI_IF_GLOBAL] = { + CCI5xx_GLOBAL_PORT_MIN_EV, + CCI5xx_GLOBAL_PORT_MAX_EV, + }, + }, + .validate_hw_event = cci550_validate_hw_event, + .write_counters = cci5xx_pmu_write_counters, + }, +#endif +}; + +static const struct of_device_id arm_cci_pmu_matches[] = { +#ifdef CONFIG_ARM_CCI400_PMU + { + .compatible = "arm,cci-400-pmu", + .data = NULL, + }, + { + .compatible = "arm,cci-400-pmu,r0", + .data = &cci_pmu_models[CCI400_R0], + }, + { + .compatible = "arm,cci-400-pmu,r1", + .data = &cci_pmu_models[CCI400_R1], + }, +#endif +#ifdef CONFIG_ARM_CCI5xx_PMU + { + .compatible = "arm,cci-500-pmu,r0", + .data = &cci_pmu_models[CCI500_R0], + }, + { + .compatible = "arm,cci-550-pmu,r0", + .data = &cci_pmu_models[CCI550_R0], + }, +#endif + {}, +}; + +static bool is_duplicate_irq(int irq, int *irqs, int nr_irqs) +{ + int i; + + for (i = 0; i < nr_irqs; i++) + if (irq == irqs[i]) + return true; + + return false; +} + +static struct cci_pmu *cci_pmu_alloc(struct device *dev) +{ + struct cci_pmu *cci_pmu; + const struct cci_pmu_model *model; + + /* + * All allocations are devm_* hence we don't have to free + * them explicitly on an error, as it would end up in driver + * detach. + */ + cci_pmu = devm_kzalloc(dev, sizeof(*cci_pmu), GFP_KERNEL); + if (!cci_pmu) + return ERR_PTR(-ENOMEM); + + cci_pmu->ctrl_base = *(void __iomem **)dev->platform_data; + + model = of_device_get_match_data(dev); + if (!model) { + dev_warn(dev, + "DEPRECATED compatible property, requires secure access to CCI registers"); + model = probe_cci_model(cci_pmu); + } + if (!model) { + dev_warn(dev, "CCI PMU version not supported\n"); + return ERR_PTR(-ENODEV); + } + + cci_pmu->model = model; + cci_pmu->irqs = devm_kcalloc(dev, CCI_PMU_MAX_HW_CNTRS(model), + sizeof(*cci_pmu->irqs), GFP_KERNEL); + if (!cci_pmu->irqs) + return ERR_PTR(-ENOMEM); + cci_pmu->hw_events.events = devm_kcalloc(dev, + CCI_PMU_MAX_HW_CNTRS(model), + sizeof(*cci_pmu->hw_events.events), + GFP_KERNEL); + if (!cci_pmu->hw_events.events) + return ERR_PTR(-ENOMEM); + cci_pmu->hw_events.used_mask = devm_kcalloc(dev, + BITS_TO_LONGS(CCI_PMU_MAX_HW_CNTRS(model)), + sizeof(*cci_pmu->hw_events.used_mask), + GFP_KERNEL); + if (!cci_pmu->hw_events.used_mask) + return ERR_PTR(-ENOMEM); + + return cci_pmu; +} + +static int cci_pmu_probe(struct platform_device *pdev) +{ + struct resource *res; + struct cci_pmu *cci_pmu; + int i, ret, irq; + + cci_pmu = cci_pmu_alloc(&pdev->dev); + if (IS_ERR(cci_pmu)) + return PTR_ERR(cci_pmu); + + 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); + cci_pmu->cpu = get_cpu(); + + ret = cci_pmu_init(cci_pmu, pdev); + if (ret) { + put_cpu(); + return ret; + } + + cpuhp_setup_state_nocalls(CPUHP_AP_PERF_ARM_CCI_ONLINE, + "perf/arm/cci:online", NULL, + cci_pmu_offline_cpu); + put_cpu(); + g_cci_pmu = cci_pmu; + pr_info("ARM %s PMU driver probed", cci_pmu->model->name); + return 0; +} + +static struct platform_driver cci_pmu_driver = { + .driver = { + .name = DRIVER_NAME, + .of_match_table = arm_cci_pmu_matches, + }, + .probe = cci_pmu_probe, +}; + +builtin_platform_driver(cci_pmu_driver); +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("ARM CCI PMU support"); diff --git a/drivers/perf/arm-ccn.c b/drivers/perf/arm-ccn.c new file mode 100644 index 000000000000..65b7e4042ece --- /dev/null +++ b/drivers/perf/arm-ccn.c @@ -0,0 +1,1597 @@ +/* + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * Copyright (C) 2014 ARM Limited + */ + +#include <linux/ctype.h> +#include <linux/hrtimer.h> +#include <linux/idr.h> +#include <linux/interrupt.h> +#include <linux/io.h> +#include <linux/module.h> +#include <linux/perf_event.h> +#include <linux/platform_device.h> +#include <linux/slab.h> + +#define CCN_NUM_XP_PORTS 2 +#define CCN_NUM_VCS 4 +#define CCN_NUM_REGIONS 256 +#define CCN_REGION_SIZE 0x10000 + +#define CCN_ALL_OLY_ID 0xff00 +#define CCN_ALL_OLY_ID__OLY_ID__SHIFT 0 +#define CCN_ALL_OLY_ID__OLY_ID__MASK 0x1f +#define CCN_ALL_OLY_ID__NODE_ID__SHIFT 8 +#define CCN_ALL_OLY_ID__NODE_ID__MASK 0x3f + +#define CCN_MN_ERRINT_STATUS 0x0008 +#define CCN_MN_ERRINT_STATUS__INTREQ__DESSERT 0x11 +#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__ENABLE 0x02 +#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLED 0x20 +#define CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLE 0x22 +#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_ENABLE 0x04 +#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_DISABLED 0x40 +#define CCN_MN_ERRINT_STATUS__CORRECTED_ERRORS_DISABLE 0x44 +#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__ENABLE 0x08 +#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLED 0x80 +#define CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLE 0x88 +#define CCN_MN_OLY_COMP_LIST_63_0 0x01e0 +#define CCN_MN_ERR_SIG_VAL_63_0 0x0300 +#define CCN_MN_ERR_SIG_VAL_63_0__DT (1 << 1) + +#define CCN_DT_ACTIVE_DSM 0x0000 +#define CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(n) ((n) * 8) +#define CCN_DT_ACTIVE_DSM__DSM_ID__MASK 0xff +#define CCN_DT_CTL 0x0028 +#define CCN_DT_CTL__DT_EN (1 << 0) +#define CCN_DT_PMEVCNT(n) (0x0100 + (n) * 0x8) +#define CCN_DT_PMCCNTR 0x0140 +#define CCN_DT_PMCCNTRSR 0x0190 +#define CCN_DT_PMOVSR 0x0198 +#define CCN_DT_PMOVSR_CLR 0x01a0 +#define CCN_DT_PMOVSR_CLR__MASK 0x1f +#define CCN_DT_PMCR 0x01a8 +#define CCN_DT_PMCR__OVFL_INTR_EN (1 << 6) +#define CCN_DT_PMCR__PMU_EN (1 << 0) +#define CCN_DT_PMSR 0x01b0 +#define CCN_DT_PMSR_REQ 0x01b8 +#define CCN_DT_PMSR_CLR 0x01c0 + +#define CCN_HNF_PMU_EVENT_SEL 0x0600 +#define CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(n) ((n) * 4) +#define CCN_HNF_PMU_EVENT_SEL__ID__MASK 0xf + +#define CCN_XP_DT_CONFIG 0x0300 +#define CCN_XP_DT_CONFIG__DT_CFG__SHIFT(n) ((n) * 4) +#define CCN_XP_DT_CONFIG__DT_CFG__MASK 0xf +#define CCN_XP_DT_CONFIG__DT_CFG__PASS_THROUGH 0x0 +#define CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT_0_OR_1 0x1 +#define CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT(n) (0x2 + (n)) +#define CCN_XP_DT_CONFIG__DT_CFG__XP_PMU_EVENT(n) (0x4 + (n)) +#define CCN_XP_DT_CONFIG__DT_CFG__DEVICE_PMU_EVENT(d, n) (0x8 + (d) * 4 + (n)) +#define CCN_XP_DT_INTERFACE_SEL 0x0308 +#define CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(n) (0 + (n) * 8) +#define CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__MASK 0x1 +#define CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(n) (1 + (n) * 8) +#define CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__MASK 0x1 +#define CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(n) (2 + (n) * 8) +#define CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__MASK 0x3 +#define CCN_XP_DT_CMP_VAL_L(n) (0x0310 + (n) * 0x40) +#define CCN_XP_DT_CMP_VAL_H(n) (0x0318 + (n) * 0x40) +#define CCN_XP_DT_CMP_MASK_L(n) (0x0320 + (n) * 0x40) +#define CCN_XP_DT_CMP_MASK_H(n) (0x0328 + (n) * 0x40) +#define CCN_XP_DT_CONTROL 0x0370 +#define CCN_XP_DT_CONTROL__DT_ENABLE (1 << 0) +#define CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(n) (12 + (n) * 4) +#define CCN_XP_DT_CONTROL__WP_ARM_SEL__MASK 0xf +#define CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS 0xf +#define CCN_XP_PMU_EVENT_SEL 0x0600 +#define CCN_XP_PMU_EVENT_SEL__ID__SHIFT(n) ((n) * 7) +#define CCN_XP_PMU_EVENT_SEL__ID__MASK 0x3f + +#define CCN_SBAS_PMU_EVENT_SEL 0x0600 +#define CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(n) ((n) * 4) +#define CCN_SBAS_PMU_EVENT_SEL__ID__MASK 0xf + +#define CCN_RNI_PMU_EVENT_SEL 0x0600 +#define CCN_RNI_PMU_EVENT_SEL__ID__SHIFT(n) ((n) * 4) +#define CCN_RNI_PMU_EVENT_SEL__ID__MASK 0xf + +#define CCN_TYPE_MN 0x01 +#define CCN_TYPE_DT 0x02 +#define CCN_TYPE_HNF 0x04 +#define CCN_TYPE_HNI 0x05 +#define CCN_TYPE_XP 0x08 +#define CCN_TYPE_SBSX 0x0c +#define CCN_TYPE_SBAS 0x10 +#define CCN_TYPE_RNI_1P 0x14 +#define CCN_TYPE_RNI_2P 0x15 +#define CCN_TYPE_RNI_3P 0x16 +#define CCN_TYPE_RND_1P 0x18 /* RN-D = RN-I + DVM */ +#define CCN_TYPE_RND_2P 0x19 +#define CCN_TYPE_RND_3P 0x1a +#define CCN_TYPE_CYCLES 0xff /* Pseudotype */ + +#define CCN_EVENT_WATCHPOINT 0xfe /* Pseudoevent */ + +#define CCN_NUM_PMU_EVENTS 4 +#define CCN_NUM_XP_WATCHPOINTS 2 /* See DT.dbg_id.num_watchpoints */ +#define CCN_NUM_PMU_EVENT_COUNTERS 8 /* See DT.dbg_id.num_pmucntr */ +#define CCN_IDX_PMU_CYCLE_COUNTER CCN_NUM_PMU_EVENT_COUNTERS + +#define CCN_NUM_PREDEFINED_MASKS 4 +#define CCN_IDX_MASK_ANY (CCN_NUM_PMU_EVENT_COUNTERS + 0) +#define CCN_IDX_MASK_EXACT (CCN_NUM_PMU_EVENT_COUNTERS + 1) +#define CCN_IDX_MASK_ORDER (CCN_NUM_PMU_EVENT_COUNTERS + 2) +#define CCN_IDX_MASK_OPCODE (CCN_NUM_PMU_EVENT_COUNTERS + 3) + +struct arm_ccn_component { + void __iomem *base; + u32 type; + + DECLARE_BITMAP(pmu_events_mask, CCN_NUM_PMU_EVENTS); + union { + struct { + DECLARE_BITMAP(dt_cmp_mask, CCN_NUM_XP_WATCHPOINTS); + } xp; + }; +}; + +#define pmu_to_arm_ccn(_pmu) container_of(container_of(_pmu, \ + struct arm_ccn_dt, pmu), struct arm_ccn, dt) + +struct arm_ccn_dt { + int id; + void __iomem *base; + + spinlock_t config_lock; + + DECLARE_BITMAP(pmu_counters_mask, CCN_NUM_PMU_EVENT_COUNTERS + 1); + struct { + struct arm_ccn_component *source; + struct perf_event *event; + } pmu_counters[CCN_NUM_PMU_EVENT_COUNTERS + 1]; + + struct { + u64 l, h; + } cmp_mask[CCN_NUM_PMU_EVENT_COUNTERS + CCN_NUM_PREDEFINED_MASKS]; + + struct hrtimer hrtimer; + + cpumask_t cpu; + struct hlist_node node; + + struct pmu pmu; +}; + +struct arm_ccn { + struct device *dev; + void __iomem *base; + unsigned int irq; + + unsigned sbas_present:1; + unsigned sbsx_present:1; + + int num_nodes; + struct arm_ccn_component *node; + + int num_xps; + struct arm_ccn_component *xp; + + struct arm_ccn_dt dt; + int mn_id; +}; + +static int arm_ccn_node_to_xp(int node) +{ + return node / CCN_NUM_XP_PORTS; +} + +static int arm_ccn_node_to_xp_port(int node) +{ + return node % CCN_NUM_XP_PORTS; +} + + +/* + * Bit shifts and masks in these defines must be kept in sync with + * arm_ccn_pmu_config_set() and CCN_FORMAT_ATTRs below! + */ +#define CCN_CONFIG_NODE(_config) (((_config) >> 0) & 0xff) +#define CCN_CONFIG_XP(_config) (((_config) >> 0) & 0xff) +#define CCN_CONFIG_TYPE(_config) (((_config) >> 8) & 0xff) +#define CCN_CONFIG_EVENT(_config) (((_config) >> 16) & 0xff) +#define CCN_CONFIG_PORT(_config) (((_config) >> 24) & 0x3) +#define CCN_CONFIG_BUS(_config) (((_config) >> 24) & 0x3) +#define CCN_CONFIG_VC(_config) (((_config) >> 26) & 0x7) +#define CCN_CONFIG_DIR(_config) (((_config) >> 29) & 0x1) +#define CCN_CONFIG_MASK(_config) (((_config) >> 30) & 0xf) + +static void arm_ccn_pmu_config_set(u64 *config, u32 node_xp, u32 type, u32 port) +{ + *config &= ~((0xff << 0) | (0xff << 8) | (0x3 << 24)); + *config |= (node_xp << 0) | (type << 8) | (port << 24); +} + +static ssize_t arm_ccn_pmu_format_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *ea = container_of(attr, + struct dev_ext_attribute, attr); + + return snprintf(buf, PAGE_SIZE, "%s\n", (char *)ea->var); +} + +#define CCN_FORMAT_ATTR(_name, _config) \ + struct dev_ext_attribute arm_ccn_pmu_format_attr_##_name = \ + { __ATTR(_name, S_IRUGO, arm_ccn_pmu_format_show, \ + NULL), _config } + +static CCN_FORMAT_ATTR(node, "config:0-7"); +static CCN_FORMAT_ATTR(xp, "config:0-7"); +static CCN_FORMAT_ATTR(type, "config:8-15"); +static CCN_FORMAT_ATTR(event, "config:16-23"); +static CCN_FORMAT_ATTR(port, "config:24-25"); +static CCN_FORMAT_ATTR(bus, "config:24-25"); +static CCN_FORMAT_ATTR(vc, "config:26-28"); +static CCN_FORMAT_ATTR(dir, "config:29-29"); +static CCN_FORMAT_ATTR(mask, "config:30-33"); +static CCN_FORMAT_ATTR(cmp_l, "config1:0-62"); +static CCN_FORMAT_ATTR(cmp_h, "config2:0-59"); + +static struct attribute *arm_ccn_pmu_format_attrs[] = { + &arm_ccn_pmu_format_attr_node.attr.attr, + &arm_ccn_pmu_format_attr_xp.attr.attr, + &arm_ccn_pmu_format_attr_type.attr.attr, + &arm_ccn_pmu_format_attr_event.attr.attr, + &arm_ccn_pmu_format_attr_port.attr.attr, + &arm_ccn_pmu_format_attr_bus.attr.attr, + &arm_ccn_pmu_format_attr_vc.attr.attr, + &arm_ccn_pmu_format_attr_dir.attr.attr, + &arm_ccn_pmu_format_attr_mask.attr.attr, + &arm_ccn_pmu_format_attr_cmp_l.attr.attr, + &arm_ccn_pmu_format_attr_cmp_h.attr.attr, + NULL +}; + +static const struct attribute_group arm_ccn_pmu_format_attr_group = { + .name = "format", + .attrs = arm_ccn_pmu_format_attrs, +}; + + +struct arm_ccn_pmu_event { + struct device_attribute attr; + u32 type; + u32 event; + int num_ports; + int num_vcs; + const char *def; + int mask; +}; + +#define CCN_EVENT_ATTR(_name) \ + __ATTR(_name, S_IRUGO, arm_ccn_pmu_event_show, NULL) + +/* + * Events defined in TRM for MN, HN-I and SBSX are actually watchpoints set on + * their ports in XP they are connected to. For the sake of usability they are + * explicitly defined here (and translated into a relevant watchpoint in + * arm_ccn_pmu_event_init()) so the user can easily request them without deep + * knowledge of the flit format. + */ + +#define CCN_EVENT_MN(_name, _def, _mask) { .attr = CCN_EVENT_ATTR(mn_##_name), \ + .type = CCN_TYPE_MN, .event = CCN_EVENT_WATCHPOINT, \ + .num_ports = CCN_NUM_XP_PORTS, .num_vcs = CCN_NUM_VCS, \ + .def = _def, .mask = _mask, } + +#define CCN_EVENT_HNI(_name, _def, _mask) { \ + .attr = CCN_EVENT_ATTR(hni_##_name), .type = CCN_TYPE_HNI, \ + .event = CCN_EVENT_WATCHPOINT, .num_ports = CCN_NUM_XP_PORTS, \ + .num_vcs = CCN_NUM_VCS, .def = _def, .mask = _mask, } + +#define CCN_EVENT_SBSX(_name, _def, _mask) { \ + .attr = CCN_EVENT_ATTR(sbsx_##_name), .type = CCN_TYPE_SBSX, \ + .event = CCN_EVENT_WATCHPOINT, .num_ports = CCN_NUM_XP_PORTS, \ + .num_vcs = CCN_NUM_VCS, .def = _def, .mask = _mask, } + +#define CCN_EVENT_HNF(_name, _event) { .attr = CCN_EVENT_ATTR(hnf_##_name), \ + .type = CCN_TYPE_HNF, .event = _event, } + +#define CCN_EVENT_XP(_name, _event) { .attr = CCN_EVENT_ATTR(xp_##_name), \ + .type = CCN_TYPE_XP, .event = _event, \ + .num_ports = CCN_NUM_XP_PORTS, .num_vcs = CCN_NUM_VCS, } + +/* + * RN-I & RN-D (RN-D = RN-I + DVM) nodes have different type ID depending + * on configuration. One of them is picked to represent the whole group, + * as they all share the same event types. + */ +#define CCN_EVENT_RNI(_name, _event) { .attr = CCN_EVENT_ATTR(rni_##_name), \ + .type = CCN_TYPE_RNI_3P, .event = _event, } + +#define CCN_EVENT_SBAS(_name, _event) { .attr = CCN_EVENT_ATTR(sbas_##_name), \ + .type = CCN_TYPE_SBAS, .event = _event, } + +#define CCN_EVENT_CYCLES(_name) { .attr = CCN_EVENT_ATTR(_name), \ + .type = CCN_TYPE_CYCLES } + + +static ssize_t arm_ccn_pmu_event_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev)); + struct arm_ccn_pmu_event *event = container_of(attr, + struct arm_ccn_pmu_event, attr); + ssize_t res; + + res = snprintf(buf, PAGE_SIZE, "type=0x%x", event->type); + if (event->event) + res += snprintf(buf + res, PAGE_SIZE - res, ",event=0x%x", + event->event); + if (event->def) + res += snprintf(buf + res, PAGE_SIZE - res, ",%s", + event->def); + if (event->mask) + res += snprintf(buf + res, PAGE_SIZE - res, ",mask=0x%x", + event->mask); + + /* Arguments required by an event */ + switch (event->type) { + case CCN_TYPE_CYCLES: + break; + case CCN_TYPE_XP: + res += snprintf(buf + res, PAGE_SIZE - res, + ",xp=?,vc=?"); + if (event->event == CCN_EVENT_WATCHPOINT) + res += snprintf(buf + res, PAGE_SIZE - res, + ",port=?,dir=?,cmp_l=?,cmp_h=?,mask=?"); + else + res += snprintf(buf + res, PAGE_SIZE - res, + ",bus=?"); + + break; + case CCN_TYPE_MN: + res += snprintf(buf + res, PAGE_SIZE - res, ",node=%d", ccn->mn_id); + break; + default: + res += snprintf(buf + res, PAGE_SIZE - res, ",node=?"); + break; + } + + res += snprintf(buf + res, PAGE_SIZE - res, "\n"); + + return res; +} + +static umode_t arm_ccn_pmu_events_is_visible(struct kobject *kobj, + struct attribute *attr, int index) +{ + struct device *dev = kobj_to_dev(kobj); + struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev)); + struct device_attribute *dev_attr = container_of(attr, + struct device_attribute, attr); + struct arm_ccn_pmu_event *event = container_of(dev_attr, + struct arm_ccn_pmu_event, attr); + + if (event->type == CCN_TYPE_SBAS && !ccn->sbas_present) + return 0; + if (event->type == CCN_TYPE_SBSX && !ccn->sbsx_present) + return 0; + + return attr->mode; +} + +static struct arm_ccn_pmu_event arm_ccn_pmu_events[] = { + CCN_EVENT_MN(eobarrier, "dir=1,vc=0,cmp_h=0x1c00", CCN_IDX_MASK_OPCODE), + CCN_EVENT_MN(ecbarrier, "dir=1,vc=0,cmp_h=0x1e00", CCN_IDX_MASK_OPCODE), + CCN_EVENT_MN(dvmop, "dir=1,vc=0,cmp_h=0x2800", CCN_IDX_MASK_OPCODE), + CCN_EVENT_HNI(txdatflits, "dir=1,vc=3", CCN_IDX_MASK_ANY), + CCN_EVENT_HNI(rxdatflits, "dir=0,vc=3", CCN_IDX_MASK_ANY), + CCN_EVENT_HNI(txreqflits, "dir=1,vc=0", CCN_IDX_MASK_ANY), + CCN_EVENT_HNI(rxreqflits, "dir=0,vc=0", CCN_IDX_MASK_ANY), + CCN_EVENT_HNI(rxreqflits_order, "dir=0,vc=0,cmp_h=0x8000", + CCN_IDX_MASK_ORDER), + CCN_EVENT_SBSX(txdatflits, "dir=1,vc=3", CCN_IDX_MASK_ANY), + CCN_EVENT_SBSX(rxdatflits, "dir=0,vc=3", CCN_IDX_MASK_ANY), + CCN_EVENT_SBSX(txreqflits, "dir=1,vc=0", CCN_IDX_MASK_ANY), + CCN_EVENT_SBSX(rxreqflits, "dir=0,vc=0", CCN_IDX_MASK_ANY), + CCN_EVENT_SBSX(rxreqflits_order, "dir=0,vc=0,cmp_h=0x8000", + CCN_IDX_MASK_ORDER), + CCN_EVENT_HNF(cache_miss, 0x1), + CCN_EVENT_HNF(l3_sf_cache_access, 0x02), + CCN_EVENT_HNF(cache_fill, 0x3), + CCN_EVENT_HNF(pocq_retry, 0x4), + CCN_EVENT_HNF(pocq_reqs_recvd, 0x5), + CCN_EVENT_HNF(sf_hit, 0x6), + CCN_EVENT_HNF(sf_evictions, 0x7), + CCN_EVENT_HNF(snoops_sent, 0x8), + CCN_EVENT_HNF(snoops_broadcast, 0x9), + CCN_EVENT_HNF(l3_eviction, 0xa), + CCN_EVENT_HNF(l3_fill_invalid_way, 0xb), + CCN_EVENT_HNF(mc_retries, 0xc), + CCN_EVENT_HNF(mc_reqs, 0xd), + CCN_EVENT_HNF(qos_hh_retry, 0xe), + CCN_EVENT_RNI(rdata_beats_p0, 0x1), + CCN_EVENT_RNI(rdata_beats_p1, 0x2), + CCN_EVENT_RNI(rdata_beats_p2, 0x3), + CCN_EVENT_RNI(rxdat_flits, 0x4), + CCN_EVENT_RNI(txdat_flits, 0x5), + CCN_EVENT_RNI(txreq_flits, 0x6), + CCN_EVENT_RNI(txreq_flits_retried, 0x7), + CCN_EVENT_RNI(rrt_full, 0x8), + CCN_EVENT_RNI(wrt_full, 0x9), + CCN_EVENT_RNI(txreq_flits_replayed, 0xa), + CCN_EVENT_XP(upload_starvation, 0x1), + CCN_EVENT_XP(download_starvation, 0x2), + CCN_EVENT_XP(respin, 0x3), + CCN_EVENT_XP(valid_flit, 0x4), + CCN_EVENT_XP(watchpoint, CCN_EVENT_WATCHPOINT), + CCN_EVENT_SBAS(rdata_beats_p0, 0x1), + CCN_EVENT_SBAS(rxdat_flits, 0x4), + CCN_EVENT_SBAS(txdat_flits, 0x5), + CCN_EVENT_SBAS(txreq_flits, 0x6), + CCN_EVENT_SBAS(txreq_flits_retried, 0x7), + CCN_EVENT_SBAS(rrt_full, 0x8), + CCN_EVENT_SBAS(wrt_full, 0x9), + CCN_EVENT_SBAS(txreq_flits_replayed, 0xa), + CCN_EVENT_CYCLES(cycles), +}; + +/* Populated in arm_ccn_init() */ +static struct attribute + *arm_ccn_pmu_events_attrs[ARRAY_SIZE(arm_ccn_pmu_events) + 1]; + +static const struct attribute_group arm_ccn_pmu_events_attr_group = { + .name = "events", + .is_visible = arm_ccn_pmu_events_is_visible, + .attrs = arm_ccn_pmu_events_attrs, +}; + + +static u64 *arm_ccn_pmu_get_cmp_mask(struct arm_ccn *ccn, const char *name) +{ + unsigned long i; + + if (WARN_ON(!name || !name[0] || !isxdigit(name[0]) || !name[1])) + return NULL; + i = isdigit(name[0]) ? name[0] - '0' : 0xa + tolower(name[0]) - 'a'; + + switch (name[1]) { + case 'l': + return &ccn->dt.cmp_mask[i].l; + case 'h': + return &ccn->dt.cmp_mask[i].h; + default: + return NULL; + } +} + +static ssize_t arm_ccn_pmu_cmp_mask_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev)); + u64 *mask = arm_ccn_pmu_get_cmp_mask(ccn, attr->attr.name); + + return mask ? snprintf(buf, PAGE_SIZE, "0x%016llx\n", *mask) : -EINVAL; +} + +static ssize_t arm_ccn_pmu_cmp_mask_store(struct device *dev, + struct device_attribute *attr, const char *buf, size_t count) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev)); + u64 *mask = arm_ccn_pmu_get_cmp_mask(ccn, attr->attr.name); + int err = -EINVAL; + + if (mask) + err = kstrtoull(buf, 0, mask); + + return err ? err : count; +} + +#define CCN_CMP_MASK_ATTR(_name) \ + struct device_attribute arm_ccn_pmu_cmp_mask_attr_##_name = \ + __ATTR(_name, S_IRUGO | S_IWUSR, \ + arm_ccn_pmu_cmp_mask_show, arm_ccn_pmu_cmp_mask_store) + +#define CCN_CMP_MASK_ATTR_RO(_name) \ + struct device_attribute arm_ccn_pmu_cmp_mask_attr_##_name = \ + __ATTR(_name, S_IRUGO, arm_ccn_pmu_cmp_mask_show, NULL) + +static CCN_CMP_MASK_ATTR(0l); +static CCN_CMP_MASK_ATTR(0h); +static CCN_CMP_MASK_ATTR(1l); +static CCN_CMP_MASK_ATTR(1h); +static CCN_CMP_MASK_ATTR(2l); +static CCN_CMP_MASK_ATTR(2h); +static CCN_CMP_MASK_ATTR(3l); +static CCN_CMP_MASK_ATTR(3h); +static CCN_CMP_MASK_ATTR(4l); +static CCN_CMP_MASK_ATTR(4h); +static CCN_CMP_MASK_ATTR(5l); +static CCN_CMP_MASK_ATTR(5h); +static CCN_CMP_MASK_ATTR(6l); +static CCN_CMP_MASK_ATTR(6h); +static CCN_CMP_MASK_ATTR(7l); +static CCN_CMP_MASK_ATTR(7h); +static CCN_CMP_MASK_ATTR_RO(8l); +static CCN_CMP_MASK_ATTR_RO(8h); +static CCN_CMP_MASK_ATTR_RO(9l); +static CCN_CMP_MASK_ATTR_RO(9h); +static CCN_CMP_MASK_ATTR_RO(al); +static CCN_CMP_MASK_ATTR_RO(ah); +static CCN_CMP_MASK_ATTR_RO(bl); +static CCN_CMP_MASK_ATTR_RO(bh); + +static struct attribute *arm_ccn_pmu_cmp_mask_attrs[] = { + &arm_ccn_pmu_cmp_mask_attr_0l.attr, &arm_ccn_pmu_cmp_mask_attr_0h.attr, + &arm_ccn_pmu_cmp_mask_attr_1l.attr, &arm_ccn_pmu_cmp_mask_attr_1h.attr, + &arm_ccn_pmu_cmp_mask_attr_2l.attr, &arm_ccn_pmu_cmp_mask_attr_2h.attr, + &arm_ccn_pmu_cmp_mask_attr_3l.attr, &arm_ccn_pmu_cmp_mask_attr_3h.attr, + &arm_ccn_pmu_cmp_mask_attr_4l.attr, &arm_ccn_pmu_cmp_mask_attr_4h.attr, + &arm_ccn_pmu_cmp_mask_attr_5l.attr, &arm_ccn_pmu_cmp_mask_attr_5h.attr, + &arm_ccn_pmu_cmp_mask_attr_6l.attr, &arm_ccn_pmu_cmp_mask_attr_6h.attr, + &arm_ccn_pmu_cmp_mask_attr_7l.attr, &arm_ccn_pmu_cmp_mask_attr_7h.attr, + &arm_ccn_pmu_cmp_mask_attr_8l.attr, &arm_ccn_pmu_cmp_mask_attr_8h.attr, + &arm_ccn_pmu_cmp_mask_attr_9l.attr, &arm_ccn_pmu_cmp_mask_attr_9h.attr, + &arm_ccn_pmu_cmp_mask_attr_al.attr, &arm_ccn_pmu_cmp_mask_attr_ah.attr, + &arm_ccn_pmu_cmp_mask_attr_bl.attr, &arm_ccn_pmu_cmp_mask_attr_bh.attr, + NULL +}; + +static const struct attribute_group arm_ccn_pmu_cmp_mask_attr_group = { + .name = "cmp_mask", + .attrs = arm_ccn_pmu_cmp_mask_attrs, +}; + +static ssize_t arm_ccn_pmu_cpumask_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(dev_get_drvdata(dev)); + + return cpumap_print_to_pagebuf(true, buf, &ccn->dt.cpu); +} + +static struct device_attribute arm_ccn_pmu_cpumask_attr = + __ATTR(cpumask, S_IRUGO, arm_ccn_pmu_cpumask_show, NULL); + +static struct attribute *arm_ccn_pmu_cpumask_attrs[] = { + &arm_ccn_pmu_cpumask_attr.attr, + NULL, +}; + +static const struct attribute_group arm_ccn_pmu_cpumask_attr_group = { + .attrs = arm_ccn_pmu_cpumask_attrs, +}; + +/* + * Default poll period is 10ms, which is way over the top anyway, + * as in the worst case scenario (an event every cycle), with 1GHz + * clocked bus, the smallest, 32 bit counter will overflow in + * more than 4s. + */ +static unsigned int arm_ccn_pmu_poll_period_us = 10000; +module_param_named(pmu_poll_period_us, arm_ccn_pmu_poll_period_us, uint, + S_IRUGO | S_IWUSR); + +static ktime_t arm_ccn_pmu_timer_period(void) +{ + return ns_to_ktime((u64)arm_ccn_pmu_poll_period_us * 1000); +} + + +static const struct attribute_group *arm_ccn_pmu_attr_groups[] = { + &arm_ccn_pmu_events_attr_group, + &arm_ccn_pmu_format_attr_group, + &arm_ccn_pmu_cmp_mask_attr_group, + &arm_ccn_pmu_cpumask_attr_group, + NULL +}; + + +static int arm_ccn_pmu_alloc_bit(unsigned long *bitmap, unsigned long size) +{ + int bit; + + do { + bit = find_first_zero_bit(bitmap, size); + if (bit >= size) + return -EAGAIN; + } while (test_and_set_bit(bit, bitmap)); + + return bit; +} + +/* All RN-I and RN-D nodes have identical PMUs */ +static int arm_ccn_pmu_type_eq(u32 a, u32 b) +{ + if (a == b) + return 1; + + switch (a) { + case CCN_TYPE_RNI_1P: + case CCN_TYPE_RNI_2P: + case CCN_TYPE_RNI_3P: + case CCN_TYPE_RND_1P: + case CCN_TYPE_RND_2P: + case CCN_TYPE_RND_3P: + switch (b) { + case CCN_TYPE_RNI_1P: + case CCN_TYPE_RNI_2P: + case CCN_TYPE_RNI_3P: + case CCN_TYPE_RND_1P: + case CCN_TYPE_RND_2P: + case CCN_TYPE_RND_3P: + return 1; + } + break; + } + + return 0; +} + +static int arm_ccn_pmu_event_alloc(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + u32 node_xp, type, event_id; + struct arm_ccn_component *source; + int bit; + + node_xp = CCN_CONFIG_NODE(event->attr.config); + type = CCN_CONFIG_TYPE(event->attr.config); + event_id = CCN_CONFIG_EVENT(event->attr.config); + + /* Allocate the cycle counter */ + if (type == CCN_TYPE_CYCLES) { + if (test_and_set_bit(CCN_IDX_PMU_CYCLE_COUNTER, + ccn->dt.pmu_counters_mask)) + return -EAGAIN; + + hw->idx = CCN_IDX_PMU_CYCLE_COUNTER; + ccn->dt.pmu_counters[CCN_IDX_PMU_CYCLE_COUNTER].event = event; + + return 0; + } + + /* Allocate an event counter */ + hw->idx = arm_ccn_pmu_alloc_bit(ccn->dt.pmu_counters_mask, + CCN_NUM_PMU_EVENT_COUNTERS); + if (hw->idx < 0) { + dev_dbg(ccn->dev, "No more counters available!\n"); + return -EAGAIN; + } + + if (type == CCN_TYPE_XP) + source = &ccn->xp[node_xp]; + else + source = &ccn->node[node_xp]; + ccn->dt.pmu_counters[hw->idx].source = source; + + /* Allocate an event source or a watchpoint */ + if (type == CCN_TYPE_XP && event_id == CCN_EVENT_WATCHPOINT) + bit = arm_ccn_pmu_alloc_bit(source->xp.dt_cmp_mask, + CCN_NUM_XP_WATCHPOINTS); + else + bit = arm_ccn_pmu_alloc_bit(source->pmu_events_mask, + CCN_NUM_PMU_EVENTS); + if (bit < 0) { + dev_dbg(ccn->dev, "No more event sources/watchpoints on node/XP %d!\n", + node_xp); + clear_bit(hw->idx, ccn->dt.pmu_counters_mask); + return -EAGAIN; + } + hw->config_base = bit; + + ccn->dt.pmu_counters[hw->idx].event = event; + + return 0; +} + +static void arm_ccn_pmu_event_release(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + + if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER) { + clear_bit(CCN_IDX_PMU_CYCLE_COUNTER, ccn->dt.pmu_counters_mask); + } else { + struct arm_ccn_component *source = + ccn->dt.pmu_counters[hw->idx].source; + + if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP && + CCN_CONFIG_EVENT(event->attr.config) == + CCN_EVENT_WATCHPOINT) + clear_bit(hw->config_base, source->xp.dt_cmp_mask); + else + clear_bit(hw->config_base, source->pmu_events_mask); + clear_bit(hw->idx, ccn->dt.pmu_counters_mask); + } + + ccn->dt.pmu_counters[hw->idx].source = NULL; + ccn->dt.pmu_counters[hw->idx].event = NULL; +} + +static int arm_ccn_pmu_event_init(struct perf_event *event) +{ + struct arm_ccn *ccn; + struct hw_perf_event *hw = &event->hw; + u32 node_xp, type, event_id; + int valid; + int i; + struct perf_event *sibling; + + if (event->attr.type != event->pmu->type) + return -ENOENT; + + ccn = pmu_to_arm_ccn(event->pmu); + + if (hw->sample_period) { + dev_warn(ccn->dev, "Sampling not supported!\n"); + return -EOPNOTSUPP; + } + + if (has_branch_stack(event) || event->attr.exclude_user || + event->attr.exclude_kernel || event->attr.exclude_hv || + event->attr.exclude_idle || event->attr.exclude_host || + event->attr.exclude_guest) { + dev_warn(ccn->dev, "Can't exclude execution levels!\n"); + return -EINVAL; + } + + if (event->cpu < 0) { + dev_warn(ccn->dev, "Can't provide per-task data!\n"); + return -EOPNOTSUPP; + } + /* + * Many perf core operations (eg. events rotation) operate on a + * single CPU context. This is obvious for CPU PMUs, where one + * expects the same sets of events being observed on all CPUs, + * but can lead to issues for off-core PMUs, like CCN, where each + * event could be theoretically assigned to a different CPU. To + * mitigate this, we enforce CPU assignment to one, selected + * processor (the one described in the "cpumask" attribute). + */ + event->cpu = cpumask_first(&ccn->dt.cpu); + + node_xp = CCN_CONFIG_NODE(event->attr.config); + type = CCN_CONFIG_TYPE(event->attr.config); + event_id = CCN_CONFIG_EVENT(event->attr.config); + + /* Validate node/xp vs topology */ + switch (type) { + case CCN_TYPE_MN: + if (node_xp != ccn->mn_id) { + dev_warn(ccn->dev, "Invalid MN ID %d!\n", node_xp); + return -EINVAL; + } + break; + case CCN_TYPE_XP: + if (node_xp >= ccn->num_xps) { + dev_warn(ccn->dev, "Invalid XP ID %d!\n", node_xp); + return -EINVAL; + } + break; + case CCN_TYPE_CYCLES: + break; + default: + if (node_xp >= ccn->num_nodes) { + dev_warn(ccn->dev, "Invalid node ID %d!\n", node_xp); + return -EINVAL; + } + if (!arm_ccn_pmu_type_eq(type, ccn->node[node_xp].type)) { + dev_warn(ccn->dev, "Invalid type 0x%x for node %d!\n", + type, node_xp); + return -EINVAL; + } + break; + } + + /* Validate event ID vs available for the type */ + for (i = 0, valid = 0; i < ARRAY_SIZE(arm_ccn_pmu_events) && !valid; + i++) { + struct arm_ccn_pmu_event *e = &arm_ccn_pmu_events[i]; + u32 port = CCN_CONFIG_PORT(event->attr.config); + u32 vc = CCN_CONFIG_VC(event->attr.config); + + if (!arm_ccn_pmu_type_eq(type, e->type)) + continue; + if (event_id != e->event) + continue; + if (e->num_ports && port >= e->num_ports) { + dev_warn(ccn->dev, "Invalid port %d for node/XP %d!\n", + port, node_xp); + return -EINVAL; + } + if (e->num_vcs && vc >= e->num_vcs) { + dev_warn(ccn->dev, "Invalid vc %d for node/XP %d!\n", + vc, node_xp); + return -EINVAL; + } + valid = 1; + } + if (!valid) { + dev_warn(ccn->dev, "Invalid event 0x%x for node/XP %d!\n", + event_id, node_xp); + return -EINVAL; + } + + /* Watchpoint-based event for a node is actually set on XP */ + if (event_id == CCN_EVENT_WATCHPOINT && type != CCN_TYPE_XP) { + u32 port; + + type = CCN_TYPE_XP; + port = arm_ccn_node_to_xp_port(node_xp); + node_xp = arm_ccn_node_to_xp(node_xp); + + arm_ccn_pmu_config_set(&event->attr.config, + node_xp, type, port); + } + + /* + * We must NOT create groups containing mixed PMUs, although software + * events are acceptable (for example to create a CCN group + * periodically read when a hrtimer aka cpu-clock leader triggers). + */ + if (event->group_leader->pmu != event->pmu && + !is_software_event(event->group_leader)) + return -EINVAL; + + for_each_sibling_event(sibling, event->group_leader) { + if (sibling->pmu != event->pmu && + !is_software_event(sibling)) + return -EINVAL; + } + + return 0; +} + +static u64 arm_ccn_pmu_read_counter(struct arm_ccn *ccn, int idx) +{ + u64 res; + + if (idx == CCN_IDX_PMU_CYCLE_COUNTER) { +#ifdef readq + res = readq(ccn->dt.base + CCN_DT_PMCCNTR); +#else + /* 40 bit counter, can do snapshot and read in two parts */ + writel(0x1, ccn->dt.base + CCN_DT_PMSR_REQ); + while (!(readl(ccn->dt.base + CCN_DT_PMSR) & 0x1)) + ; + writel(0x1, ccn->dt.base + CCN_DT_PMSR_CLR); + res = readl(ccn->dt.base + CCN_DT_PMCCNTRSR + 4) & 0xff; + res <<= 32; + res |= readl(ccn->dt.base + CCN_DT_PMCCNTRSR); +#endif + } else { + res = readl(ccn->dt.base + CCN_DT_PMEVCNT(idx)); + } + + return res; +} + +static void arm_ccn_pmu_event_update(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + u64 prev_count, new_count, mask; + + do { + prev_count = local64_read(&hw->prev_count); + new_count = arm_ccn_pmu_read_counter(ccn, hw->idx); + } while (local64_xchg(&hw->prev_count, new_count) != prev_count); + + mask = (1LLU << (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER ? 40 : 32)) - 1; + + local64_add((new_count - prev_count) & mask, &event->count); +} + +static void arm_ccn_pmu_xp_dt_config(struct perf_event *event, int enable) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + struct arm_ccn_component *xp; + u32 val, dt_cfg; + + /* Nothing to do for cycle counter */ + if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER) + return; + + if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP) + xp = &ccn->xp[CCN_CONFIG_XP(event->attr.config)]; + else + xp = &ccn->xp[arm_ccn_node_to_xp( + CCN_CONFIG_NODE(event->attr.config))]; + + if (enable) + dt_cfg = hw->event_base; + else + dt_cfg = CCN_XP_DT_CONFIG__DT_CFG__PASS_THROUGH; + + spin_lock(&ccn->dt.config_lock); + + val = readl(xp->base + CCN_XP_DT_CONFIG); + val &= ~(CCN_XP_DT_CONFIG__DT_CFG__MASK << + CCN_XP_DT_CONFIG__DT_CFG__SHIFT(hw->idx)); + val |= dt_cfg << CCN_XP_DT_CONFIG__DT_CFG__SHIFT(hw->idx); + writel(val, xp->base + CCN_XP_DT_CONFIG); + + spin_unlock(&ccn->dt.config_lock); +} + +static void arm_ccn_pmu_event_start(struct perf_event *event, int flags) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + + local64_set(&event->hw.prev_count, + arm_ccn_pmu_read_counter(ccn, hw->idx)); + hw->state = 0; + + /* Set the DT bus input, engaging the counter */ + arm_ccn_pmu_xp_dt_config(event, 1); +} + +static void arm_ccn_pmu_event_stop(struct perf_event *event, int flags) +{ + struct hw_perf_event *hw = &event->hw; + + /* Disable counting, setting the DT bus to pass-through mode */ + arm_ccn_pmu_xp_dt_config(event, 0); + + if (flags & PERF_EF_UPDATE) + arm_ccn_pmu_event_update(event); + + hw->state |= PERF_HES_STOPPED; +} + +static void arm_ccn_pmu_xp_watchpoint_config(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + struct arm_ccn_component *source = + ccn->dt.pmu_counters[hw->idx].source; + unsigned long wp = hw->config_base; + u32 val; + u64 cmp_l = event->attr.config1; + u64 cmp_h = event->attr.config2; + u64 mask_l = ccn->dt.cmp_mask[CCN_CONFIG_MASK(event->attr.config)].l; + u64 mask_h = ccn->dt.cmp_mask[CCN_CONFIG_MASK(event->attr.config)].h; + + hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__WATCHPOINT(wp); + + /* Direction (RX/TX), device (port) & virtual channel */ + val = readl(source->base + CCN_XP_DT_INTERFACE_SEL); + val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__MASK << + CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(wp)); + val |= CCN_CONFIG_DIR(event->attr.config) << + CCN_XP_DT_INTERFACE_SEL__DT_IO_SEL__SHIFT(wp); + val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__MASK << + CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(wp)); + val |= CCN_CONFIG_PORT(event->attr.config) << + CCN_XP_DT_INTERFACE_SEL__DT_DEV_SEL__SHIFT(wp); + val &= ~(CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__MASK << + CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(wp)); + val |= CCN_CONFIG_VC(event->attr.config) << + CCN_XP_DT_INTERFACE_SEL__DT_VC_SEL__SHIFT(wp); + writel(val, source->base + CCN_XP_DT_INTERFACE_SEL); + + /* Comparison values */ + writel(cmp_l & 0xffffffff, source->base + CCN_XP_DT_CMP_VAL_L(wp)); + writel((cmp_l >> 32) & 0x7fffffff, + source->base + CCN_XP_DT_CMP_VAL_L(wp) + 4); + writel(cmp_h & 0xffffffff, source->base + CCN_XP_DT_CMP_VAL_H(wp)); + writel((cmp_h >> 32) & 0x0fffffff, + source->base + CCN_XP_DT_CMP_VAL_H(wp) + 4); + + /* Mask */ + writel(mask_l & 0xffffffff, source->base + CCN_XP_DT_CMP_MASK_L(wp)); + writel((mask_l >> 32) & 0x7fffffff, + source->base + CCN_XP_DT_CMP_MASK_L(wp) + 4); + writel(mask_h & 0xffffffff, source->base + CCN_XP_DT_CMP_MASK_H(wp)); + writel((mask_h >> 32) & 0x0fffffff, + source->base + CCN_XP_DT_CMP_MASK_H(wp) + 4); +} + +static void arm_ccn_pmu_xp_event_config(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + struct arm_ccn_component *source = + ccn->dt.pmu_counters[hw->idx].source; + u32 val, id; + + hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__XP_PMU_EVENT(hw->config_base); + + id = (CCN_CONFIG_VC(event->attr.config) << 4) | + (CCN_CONFIG_BUS(event->attr.config) << 3) | + (CCN_CONFIG_EVENT(event->attr.config) << 0); + + val = readl(source->base + CCN_XP_PMU_EVENT_SEL); + val &= ~(CCN_XP_PMU_EVENT_SEL__ID__MASK << + CCN_XP_PMU_EVENT_SEL__ID__SHIFT(hw->config_base)); + val |= id << CCN_XP_PMU_EVENT_SEL__ID__SHIFT(hw->config_base); + writel(val, source->base + CCN_XP_PMU_EVENT_SEL); +} + +static void arm_ccn_pmu_node_event_config(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + struct arm_ccn_component *source = + ccn->dt.pmu_counters[hw->idx].source; + u32 type = CCN_CONFIG_TYPE(event->attr.config); + u32 val, port; + + port = arm_ccn_node_to_xp_port(CCN_CONFIG_NODE(event->attr.config)); + hw->event_base = CCN_XP_DT_CONFIG__DT_CFG__DEVICE_PMU_EVENT(port, + hw->config_base); + + /* These *_event_sel regs should be identical, but let's make sure... */ + BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL != CCN_SBAS_PMU_EVENT_SEL); + BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL != CCN_RNI_PMU_EVENT_SEL); + BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(1) != + CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(1)); + BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL__ID__SHIFT(1) != + CCN_RNI_PMU_EVENT_SEL__ID__SHIFT(1)); + BUILD_BUG_ON(CCN_HNF_PMU_EVENT_SEL__ID__MASK != + CCN_SBAS_PMU_EVENT_SEL__ID__MASK); + BUILD_BUG_ON(CCN_SBAS_PMU_EVENT_SEL__ID__MASK != + CCN_RNI_PMU_EVENT_SEL__ID__MASK); + if (WARN_ON(type != CCN_TYPE_HNF && type != CCN_TYPE_SBAS && + !arm_ccn_pmu_type_eq(type, CCN_TYPE_RNI_3P))) + return; + + /* Set the event id for the pre-allocated counter */ + val = readl(source->base + CCN_HNF_PMU_EVENT_SEL); + val &= ~(CCN_HNF_PMU_EVENT_SEL__ID__MASK << + CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(hw->config_base)); + val |= CCN_CONFIG_EVENT(event->attr.config) << + CCN_HNF_PMU_EVENT_SEL__ID__SHIFT(hw->config_base); + writel(val, source->base + CCN_HNF_PMU_EVENT_SEL); +} + +static void arm_ccn_pmu_event_config(struct perf_event *event) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + struct hw_perf_event *hw = &event->hw; + u32 xp, offset, val; + + /* Cycle counter requires no setup */ + if (hw->idx == CCN_IDX_PMU_CYCLE_COUNTER) + return; + + if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP) + xp = CCN_CONFIG_XP(event->attr.config); + else + xp = arm_ccn_node_to_xp(CCN_CONFIG_NODE(event->attr.config)); + + spin_lock(&ccn->dt.config_lock); + + /* Set the DT bus "distance" register */ + offset = (hw->idx / 4) * 4; + val = readl(ccn->dt.base + CCN_DT_ACTIVE_DSM + offset); + val &= ~(CCN_DT_ACTIVE_DSM__DSM_ID__MASK << + CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(hw->idx % 4)); + val |= xp << CCN_DT_ACTIVE_DSM__DSM_ID__SHIFT(hw->idx % 4); + writel(val, ccn->dt.base + CCN_DT_ACTIVE_DSM + offset); + + if (CCN_CONFIG_TYPE(event->attr.config) == CCN_TYPE_XP) { + if (CCN_CONFIG_EVENT(event->attr.config) == + CCN_EVENT_WATCHPOINT) + arm_ccn_pmu_xp_watchpoint_config(event); + else + arm_ccn_pmu_xp_event_config(event); + } else { + arm_ccn_pmu_node_event_config(event); + } + + spin_unlock(&ccn->dt.config_lock); +} + +static int arm_ccn_pmu_active_counters(struct arm_ccn *ccn) +{ + return bitmap_weight(ccn->dt.pmu_counters_mask, + CCN_NUM_PMU_EVENT_COUNTERS + 1); +} + +static int arm_ccn_pmu_event_add(struct perf_event *event, int flags) +{ + int err; + struct hw_perf_event *hw = &event->hw; + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + + err = arm_ccn_pmu_event_alloc(event); + if (err) + return err; + + /* + * Pin the timer, so that the overflows are handled by the chosen + * event->cpu (this is the same one as presented in "cpumask" + * attribute). + */ + if (!ccn->irq && arm_ccn_pmu_active_counters(ccn) == 1) + hrtimer_start(&ccn->dt.hrtimer, arm_ccn_pmu_timer_period(), + HRTIMER_MODE_REL_PINNED); + + arm_ccn_pmu_event_config(event); + + hw->state = PERF_HES_STOPPED; + + if (flags & PERF_EF_START) + arm_ccn_pmu_event_start(event, PERF_EF_UPDATE); + + return 0; +} + +static void arm_ccn_pmu_event_del(struct perf_event *event, int flags) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(event->pmu); + + arm_ccn_pmu_event_stop(event, PERF_EF_UPDATE); + + arm_ccn_pmu_event_release(event); + + if (!ccn->irq && arm_ccn_pmu_active_counters(ccn) == 0) + hrtimer_cancel(&ccn->dt.hrtimer); +} + +static void arm_ccn_pmu_event_read(struct perf_event *event) +{ + arm_ccn_pmu_event_update(event); +} + +static void arm_ccn_pmu_enable(struct pmu *pmu) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(pmu); + + u32 val = readl(ccn->dt.base + CCN_DT_PMCR); + val |= CCN_DT_PMCR__PMU_EN; + writel(val, ccn->dt.base + CCN_DT_PMCR); +} + +static void arm_ccn_pmu_disable(struct pmu *pmu) +{ + struct arm_ccn *ccn = pmu_to_arm_ccn(pmu); + + u32 val = readl(ccn->dt.base + CCN_DT_PMCR); + val &= ~CCN_DT_PMCR__PMU_EN; + writel(val, ccn->dt.base + CCN_DT_PMCR); +} + +static irqreturn_t arm_ccn_pmu_overflow_handler(struct arm_ccn_dt *dt) +{ + u32 pmovsr = readl(dt->base + CCN_DT_PMOVSR); + int idx; + + if (!pmovsr) + return IRQ_NONE; + + writel(pmovsr, dt->base + CCN_DT_PMOVSR_CLR); + + BUILD_BUG_ON(CCN_IDX_PMU_CYCLE_COUNTER != CCN_NUM_PMU_EVENT_COUNTERS); + + for (idx = 0; idx < CCN_NUM_PMU_EVENT_COUNTERS + 1; idx++) { + struct perf_event *event = dt->pmu_counters[idx].event; + int overflowed = pmovsr & BIT(idx); + + WARN_ON_ONCE(overflowed && !event && + idx != CCN_IDX_PMU_CYCLE_COUNTER); + + if (!event || !overflowed) + continue; + + arm_ccn_pmu_event_update(event); + } + + return IRQ_HANDLED; +} + +static enum hrtimer_restart arm_ccn_pmu_timer_handler(struct hrtimer *hrtimer) +{ + struct arm_ccn_dt *dt = container_of(hrtimer, struct arm_ccn_dt, + hrtimer); + unsigned long flags; + + local_irq_save(flags); + arm_ccn_pmu_overflow_handler(dt); + local_irq_restore(flags); + + hrtimer_forward_now(hrtimer, arm_ccn_pmu_timer_period()); + return HRTIMER_RESTART; +} + + +static int arm_ccn_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) +{ + struct arm_ccn_dt *dt = hlist_entry_safe(node, struct arm_ccn_dt, node); + struct arm_ccn *ccn = container_of(dt, struct arm_ccn, dt); + unsigned int target; + + if (!cpumask_test_and_clear_cpu(cpu, &dt->cpu)) + return 0; + target = cpumask_any_but(cpu_online_mask, cpu); + if (target >= nr_cpu_ids) + return 0; + perf_pmu_migrate_context(&dt->pmu, cpu, target); + cpumask_set_cpu(target, &dt->cpu); + if (ccn->irq) + WARN_ON(irq_set_affinity_hint(ccn->irq, &dt->cpu) != 0); + return 0; +} + +static DEFINE_IDA(arm_ccn_pmu_ida); + +static int arm_ccn_pmu_init(struct arm_ccn *ccn) +{ + int i; + char *name; + int err; + + /* Initialize DT subsystem */ + ccn->dt.base = ccn->base + CCN_REGION_SIZE; + spin_lock_init(&ccn->dt.config_lock); + writel(CCN_DT_PMOVSR_CLR__MASK, ccn->dt.base + CCN_DT_PMOVSR_CLR); + writel(CCN_DT_CTL__DT_EN, ccn->dt.base + CCN_DT_CTL); + writel(CCN_DT_PMCR__OVFL_INTR_EN | CCN_DT_PMCR__PMU_EN, + ccn->dt.base + CCN_DT_PMCR); + writel(0x1, ccn->dt.base + CCN_DT_PMSR_CLR); + for (i = 0; i < ccn->num_xps; i++) { + writel(0, ccn->xp[i].base + CCN_XP_DT_CONFIG); + writel((CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS << + CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(0)) | + (CCN_XP_DT_CONTROL__WP_ARM_SEL__ALWAYS << + CCN_XP_DT_CONTROL__WP_ARM_SEL__SHIFT(1)) | + CCN_XP_DT_CONTROL__DT_ENABLE, + ccn->xp[i].base + CCN_XP_DT_CONTROL); + } + ccn->dt.cmp_mask[CCN_IDX_MASK_ANY].l = ~0; + ccn->dt.cmp_mask[CCN_IDX_MASK_ANY].h = ~0; + ccn->dt.cmp_mask[CCN_IDX_MASK_EXACT].l = 0; + ccn->dt.cmp_mask[CCN_IDX_MASK_EXACT].h = 0; + ccn->dt.cmp_mask[CCN_IDX_MASK_ORDER].l = ~0; + ccn->dt.cmp_mask[CCN_IDX_MASK_ORDER].h = ~(0x1 << 15); + ccn->dt.cmp_mask[CCN_IDX_MASK_OPCODE].l = ~0; + ccn->dt.cmp_mask[CCN_IDX_MASK_OPCODE].h = ~(0x1f << 9); + + /* Get a convenient /sys/event_source/devices/ name */ + ccn->dt.id = ida_simple_get(&arm_ccn_pmu_ida, 0, 0, GFP_KERNEL); + if (ccn->dt.id == 0) { + name = "ccn"; + } else { + name = devm_kasprintf(ccn->dev, GFP_KERNEL, "ccn_%d", + ccn->dt.id); + if (!name) { + err = -ENOMEM; + goto error_choose_name; + } + } + + /* Perf driver registration */ + ccn->dt.pmu = (struct pmu) { + .module = THIS_MODULE, + .attr_groups = arm_ccn_pmu_attr_groups, + .task_ctx_nr = perf_invalid_context, + .event_init = arm_ccn_pmu_event_init, + .add = arm_ccn_pmu_event_add, + .del = arm_ccn_pmu_event_del, + .start = arm_ccn_pmu_event_start, + .stop = arm_ccn_pmu_event_stop, + .read = arm_ccn_pmu_event_read, + .pmu_enable = arm_ccn_pmu_enable, + .pmu_disable = arm_ccn_pmu_disable, + }; + + /* No overflow interrupt? Have to use a timer instead. */ + if (!ccn->irq) { + dev_info(ccn->dev, "No access to interrupts, using timer.\n"); + hrtimer_init(&ccn->dt.hrtimer, CLOCK_MONOTONIC, + HRTIMER_MODE_REL); + ccn->dt.hrtimer.function = arm_ccn_pmu_timer_handler; + } + + /* Pick one CPU which we will use to collect data from CCN... */ + cpumask_set_cpu(get_cpu(), &ccn->dt.cpu); + + /* Also make sure that the overflow interrupt is handled by this CPU */ + if (ccn->irq) { + err = irq_set_affinity_hint(ccn->irq, &ccn->dt.cpu); + if (err) { + dev_err(ccn->dev, "Failed to set interrupt affinity!\n"); + goto error_set_affinity; + } + } + + err = perf_pmu_register(&ccn->dt.pmu, name, -1); + if (err) + goto error_pmu_register; + + cpuhp_state_add_instance_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE, + &ccn->dt.node); + put_cpu(); + return 0; + +error_pmu_register: +error_set_affinity: + put_cpu(); +error_choose_name: + ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id); + for (i = 0; i < ccn->num_xps; i++) + writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL); + writel(0, ccn->dt.base + CCN_DT_PMCR); + return err; +} + +static void arm_ccn_pmu_cleanup(struct arm_ccn *ccn) +{ + int i; + + cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_CCN_ONLINE, + &ccn->dt.node); + if (ccn->irq) + irq_set_affinity_hint(ccn->irq, NULL); + for (i = 0; i < ccn->num_xps; i++) + writel(0, ccn->xp[i].base + CCN_XP_DT_CONTROL); + writel(0, ccn->dt.base + CCN_DT_PMCR); + perf_pmu_unregister(&ccn->dt.pmu); + ida_simple_remove(&arm_ccn_pmu_ida, ccn->dt.id); +} + +static int arm_ccn_for_each_valid_region(struct arm_ccn *ccn, + int (*callback)(struct arm_ccn *ccn, int region, + void __iomem *base, u32 type, u32 id)) +{ + int region; + + for (region = 0; region < CCN_NUM_REGIONS; region++) { + u32 val, type, id; + void __iomem *base; + int err; + + val = readl(ccn->base + CCN_MN_OLY_COMP_LIST_63_0 + + 4 * (region / 32)); + if (!(val & (1 << (region % 32)))) + continue; + + base = ccn->base + region * CCN_REGION_SIZE; + val = readl(base + CCN_ALL_OLY_ID); + type = (val >> CCN_ALL_OLY_ID__OLY_ID__SHIFT) & + CCN_ALL_OLY_ID__OLY_ID__MASK; + id = (val >> CCN_ALL_OLY_ID__NODE_ID__SHIFT) & + CCN_ALL_OLY_ID__NODE_ID__MASK; + + err = callback(ccn, region, base, type, id); + if (err) + return err; + } + + return 0; +} + +static int arm_ccn_get_nodes_num(struct arm_ccn *ccn, int region, + void __iomem *base, u32 type, u32 id) +{ + + if (type == CCN_TYPE_XP && id >= ccn->num_xps) + ccn->num_xps = id + 1; + else if (id >= ccn->num_nodes) + ccn->num_nodes = id + 1; + + return 0; +} + +static int arm_ccn_init_nodes(struct arm_ccn *ccn, int region, + void __iomem *base, u32 type, u32 id) +{ + struct arm_ccn_component *component; + + dev_dbg(ccn->dev, "Region %d: id=%u, type=0x%02x\n", region, id, type); + + switch (type) { + case CCN_TYPE_MN: + ccn->mn_id = id; + return 0; + case CCN_TYPE_DT: + return 0; + case CCN_TYPE_XP: + component = &ccn->xp[id]; + break; + case CCN_TYPE_SBSX: + ccn->sbsx_present = 1; + component = &ccn->node[id]; + break; + case CCN_TYPE_SBAS: + ccn->sbas_present = 1; + /* Fall-through */ + default: + component = &ccn->node[id]; + break; + } + + component->base = base; + component->type = type; + + return 0; +} + + +static irqreturn_t arm_ccn_error_handler(struct arm_ccn *ccn, + const u32 *err_sig_val) +{ + /* This should be really handled by firmware... */ + dev_err(ccn->dev, "Error reported in %08x%08x%08x%08x%08x%08x.\n", + err_sig_val[5], err_sig_val[4], err_sig_val[3], + err_sig_val[2], err_sig_val[1], err_sig_val[0]); + dev_err(ccn->dev, "Disabling interrupt generation for all errors.\n"); + writel(CCN_MN_ERRINT_STATUS__ALL_ERRORS__DISABLE, + ccn->base + CCN_MN_ERRINT_STATUS); + + return IRQ_HANDLED; +} + + +static irqreturn_t arm_ccn_irq_handler(int irq, void *dev_id) +{ + irqreturn_t res = IRQ_NONE; + struct arm_ccn *ccn = dev_id; + u32 err_sig_val[6]; + u32 err_or; + int i; + + /* PMU overflow is a special case */ + err_or = err_sig_val[0] = readl(ccn->base + CCN_MN_ERR_SIG_VAL_63_0); + if (err_or & CCN_MN_ERR_SIG_VAL_63_0__DT) { + err_or &= ~CCN_MN_ERR_SIG_VAL_63_0__DT; + res = arm_ccn_pmu_overflow_handler(&ccn->dt); + } + + /* Have to read all err_sig_vals to clear them */ + for (i = 1; i < ARRAY_SIZE(err_sig_val); i++) { + err_sig_val[i] = readl(ccn->base + + CCN_MN_ERR_SIG_VAL_63_0 + i * 4); + err_or |= err_sig_val[i]; + } + if (err_or) + res |= arm_ccn_error_handler(ccn, err_sig_val); + + if (res != IRQ_NONE) + writel(CCN_MN_ERRINT_STATUS__INTREQ__DESSERT, + ccn->base + CCN_MN_ERRINT_STATUS); + + return res; +} + + +static int arm_ccn_probe(struct platform_device *pdev) +{ + struct arm_ccn *ccn; + struct resource *res; + unsigned int irq; + int err; + + ccn = devm_kzalloc(&pdev->dev, sizeof(*ccn), GFP_KERNEL); + if (!ccn) + return -ENOMEM; + ccn->dev = &pdev->dev; + platform_set_drvdata(pdev, ccn); + + res = platform_get_resource(pdev, IORESOURCE_MEM, 0); + if (!res) + return -EINVAL; + + if (!devm_request_mem_region(ccn->dev, res->start, + resource_size(res), pdev->name)) + return -EBUSY; + + ccn->base = devm_ioremap(ccn->dev, res->start, + resource_size(res)); + if (!ccn->base) + return -EFAULT; + + res = platform_get_resource(pdev, IORESOURCE_IRQ, 0); + if (!res) + return -EINVAL; + irq = res->start; + + /* Check if we can use the interrupt */ + writel(CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLE, + ccn->base + CCN_MN_ERRINT_STATUS); + if (readl(ccn->base + CCN_MN_ERRINT_STATUS) & + CCN_MN_ERRINT_STATUS__PMU_EVENTS__DISABLED) { + /* Can set 'disable' bits, so can acknowledge interrupts */ + writel(CCN_MN_ERRINT_STATUS__PMU_EVENTS__ENABLE, + ccn->base + CCN_MN_ERRINT_STATUS); + err = devm_request_irq(ccn->dev, irq, arm_ccn_irq_handler, + IRQF_NOBALANCING | IRQF_NO_THREAD, + dev_name(ccn->dev), ccn); + if (err) + return err; + + ccn->irq = irq; + } + + + /* Build topology */ + + err = arm_ccn_for_each_valid_region(ccn, arm_ccn_get_nodes_num); + if (err) + return err; + + ccn->node = devm_kcalloc(ccn->dev, ccn->num_nodes, sizeof(*ccn->node), + GFP_KERNEL); + ccn->xp = devm_kcalloc(ccn->dev, ccn->num_xps, sizeof(*ccn->node), + GFP_KERNEL); + if (!ccn->node || !ccn->xp) + return -ENOMEM; + + err = arm_ccn_for_each_valid_region(ccn, arm_ccn_init_nodes); + if (err) + return err; + + return arm_ccn_pmu_init(ccn); +} + +static int arm_ccn_remove(struct platform_device *pdev) +{ + struct arm_ccn *ccn = platform_get_drvdata(pdev); + + arm_ccn_pmu_cleanup(ccn); + + return 0; +} + +static const struct of_device_id arm_ccn_match[] = { + { .compatible = "arm,ccn-502", }, + { .compatible = "arm,ccn-504", }, + {}, +}; +MODULE_DEVICE_TABLE(of, arm_ccn_match); + +static struct platform_driver arm_ccn_driver = { + .driver = { + .name = "arm-ccn", + .of_match_table = arm_ccn_match, + }, + .probe = arm_ccn_probe, + .remove = arm_ccn_remove, +}; + +static int __init arm_ccn_init(void) +{ + int i, ret; + + ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_CCN_ONLINE, + "perf/arm/ccn:online", NULL, + arm_ccn_pmu_offline_cpu); + if (ret) + return ret; + + for (i = 0; i < ARRAY_SIZE(arm_ccn_pmu_events); i++) + arm_ccn_pmu_events_attrs[i] = &arm_ccn_pmu_events[i].attr.attr; + + ret = platform_driver_register(&arm_ccn_driver); + if (ret) + cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CCN_ONLINE); + return ret; +} + +static void __exit arm_ccn_exit(void) +{ + platform_driver_unregister(&arm_ccn_driver); + cpuhp_remove_multi_state(CPUHP_AP_PERF_ARM_CCN_ONLINE); +} + +module_init(arm_ccn_init); +module_exit(arm_ccn_exit); + +MODULE_AUTHOR("Pawel Moll <pawel.moll@arm.com>"); +MODULE_LICENSE("GPL"); |