Commit 494bc3cd authored by Catalin Marinas's avatar Catalin Marinas

Merge branch 'will/for-next/perf' into for-next/core

* will/for-next/perf:
  arm64: pmuv3: use arm_pmu ACPI framework
  arm64: pmuv3: handle !PMUv3 when probing
  drivers/perf: arm_pmu: add ACPI framework
  arm64: add function to get a cpu's MADT GICC table
  drivers/perf: arm_pmu: split out platform device probe logic
  drivers/perf: arm_pmu: move irq request/free into probe
  drivers/perf: arm_pmu: split cpu-local irq request/free
  drivers/perf: arm_pmu: rename irq request/free functions
  drivers/perf: arm_pmu: handle no platform_device
  drivers/perf: arm_pmu: simplify cpu_pmu_request_irqs()
  drivers/perf: arm_pmu: factor out pmu registration
  drivers/perf: arm_pmu: fold init into alloc
  drivers/perf: arm_pmu: define armpmu_init_fn
  drivers/perf: arm_pmu: remove pointless PMU disabling
  perf: qcom: Add L3 cache PMU driver
  drivers/perf: arm_pmu: split irq request from enable
  drivers/perf: arm_pmu: manage interrupts per-cpu
  drivers/perf: arm_pmu: rework per-cpu allocation
  MAINTAINERS: Add file patterns for perf device tree bindings
parents d91750f1 f00fa5f4
Qualcomm Datacenter Technologies L3 Cache Performance Monitoring Unit (PMU)
===========================================================================
This driver supports the L3 cache PMUs found in Qualcomm Datacenter Technologies
Centriq SoCs. The L3 cache on these SOCs is composed of multiple slices, shared
by all cores within a socket. Each slice is exposed as a separate uncore perf
PMU with device name l3cache_<socket>_<instance>. User space is responsible
for aggregating across slices.
The driver provides a description of its available events and configuration
options in sysfs, see /sys/devices/l3cache*. Given that these are uncore PMUs
the driver also exposes a "cpumask" sysfs attribute which contains a mask
consisting of one CPU per socket which will be used to handle all the PMU
events on that socket.
The hardware implements 32bit event counters and has a flat 8bit event space
exposed via the "event" format attribute. In addition to the 32bit physical
counters the driver supports virtual 64bit hardware counters by using hardware
counter chaining. This feature is exposed via the "lc" (long counter) format
flag. E.g.:
perf stat -e l3cache_0_0/read-miss,lc/
Given that these are uncore PMUs the driver does not support sampling, therefore
"perf record" will not work. Per-task perf sessions are not supported.
...@@ -976,6 +976,7 @@ F: arch/arm*/include/asm/perf_event.h ...@@ -976,6 +976,7 @@ F: arch/arm*/include/asm/perf_event.h
F: drivers/perf/* F: drivers/perf/*
F: include/linux/perf/arm_pmu.h F: include/linux/perf/arm_pmu.h
F: Documentation/devicetree/bindings/arm/pmu.txt F: Documentation/devicetree/bindings/arm/pmu.txt
F: Documentation/devicetree/bindings/perf/
ARM PORT ARM PORT
M: Russell King <linux@armlinux.org.uk> M: Russell King <linux@armlinux.org.uk>
......
...@@ -85,6 +85,8 @@ static inline bool acpi_has_cpu_in_madt(void) ...@@ -85,6 +85,8 @@ static inline bool acpi_has_cpu_in_madt(void)
return true; return true;
} }
struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu);
static inline void arch_fix_phys_package_id(int num, u32 slot) { } static inline void arch_fix_phys_package_id(int num, u32 slot) { }
void __init acpi_init_cpus(void); void __init acpi_init_cpus(void);
......
...@@ -957,11 +957,26 @@ static int armv8_vulcan_map_event(struct perf_event *event) ...@@ -957,11 +957,26 @@ static int armv8_vulcan_map_event(struct perf_event *event)
ARMV8_PMU_EVTYPE_EVENT); ARMV8_PMU_EVTYPE_EVENT);
} }
struct armv8pmu_probe_info {
struct arm_pmu *pmu;
bool present;
};
static void __armv8pmu_probe_pmu(void *info) static void __armv8pmu_probe_pmu(void *info)
{ {
struct arm_pmu *cpu_pmu = info; struct armv8pmu_probe_info *probe = info;
struct arm_pmu *cpu_pmu = probe->pmu;
u64 dfr0, pmuver;
u32 pmceid[2]; u32 pmceid[2];
dfr0 = read_sysreg(id_aa64dfr0_el1);
pmuver = cpuid_feature_extract_unsigned_field(dfr0,
ID_AA64DFR0_PMUVER_SHIFT);
if (pmuver != 1)
return;
probe->present = true;
/* Read the nb of CNTx counters supported from PMNC */ /* Read the nb of CNTx counters supported from PMNC */
cpu_pmu->num_events = (armv8pmu_pmcr_read() >> ARMV8_PMU_PMCR_N_SHIFT) cpu_pmu->num_events = (armv8pmu_pmcr_read() >> ARMV8_PMU_PMCR_N_SHIFT)
& ARMV8_PMU_PMCR_N_MASK; & ARMV8_PMU_PMCR_N_MASK;
...@@ -979,13 +994,27 @@ static void __armv8pmu_probe_pmu(void *info) ...@@ -979,13 +994,27 @@ static void __armv8pmu_probe_pmu(void *info)
static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu) static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
{ {
return smp_call_function_any(&cpu_pmu->supported_cpus, struct armv8pmu_probe_info probe = {
.pmu = cpu_pmu,
.present = false,
};
int ret;
ret = smp_call_function_any(&cpu_pmu->supported_cpus,
__armv8pmu_probe_pmu, __armv8pmu_probe_pmu,
cpu_pmu, 1); &probe, 1);
if (ret)
return ret;
return probe.present ? 0 : -ENODEV;
} }
static void armv8_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_pmu_init(struct arm_pmu *cpu_pmu)
{ {
int ret = armv8pmu_probe_pmu(cpu_pmu);
if (ret)
return ret;
cpu_pmu->handle_irq = armv8pmu_handle_irq, cpu_pmu->handle_irq = armv8pmu_handle_irq,
cpu_pmu->enable = armv8pmu_enable_event, cpu_pmu->enable = armv8pmu_enable_event,
cpu_pmu->disable = armv8pmu_disable_event, cpu_pmu->disable = armv8pmu_disable_event,
...@@ -997,78 +1026,104 @@ static void armv8_pmu_init(struct arm_pmu *cpu_pmu) ...@@ -997,78 +1026,104 @@ static void armv8_pmu_init(struct arm_pmu *cpu_pmu)
cpu_pmu->reset = armv8pmu_reset, cpu_pmu->reset = armv8pmu_reset,
cpu_pmu->max_period = (1LLU << 32) - 1, cpu_pmu->max_period = (1LLU << 32) - 1,
cpu_pmu->set_event_filter = armv8pmu_set_event_filter; cpu_pmu->set_event_filter = armv8pmu_set_event_filter;
return 0;
} }
static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu) static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_pmuv3"; cpu_pmu->name = "armv8_pmuv3";
cpu_pmu->map_event = armv8_pmuv3_map_event; cpu_pmu->map_event = armv8_pmuv3_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_cortex_a53"; cpu_pmu->name = "armv8_cortex_a53";
cpu_pmu->map_event = armv8_a53_map_event; cpu_pmu->map_event = armv8_a53_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_cortex_a57"; cpu_pmu->name = "armv8_cortex_a57";
cpu_pmu->map_event = armv8_a57_map_event; cpu_pmu->map_event = armv8_a57_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_cortex_a72"; cpu_pmu->name = "armv8_cortex_a72";
cpu_pmu->map_event = armv8_a57_map_event; cpu_pmu->map_event = armv8_a57_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_cavium_thunder"; cpu_pmu->name = "armv8_cavium_thunder";
cpu_pmu->map_event = armv8_thunder_map_event; cpu_pmu->map_event = armv8_thunder_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu) static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu)
{ {
armv8_pmu_init(cpu_pmu); int ret = armv8_pmu_init(cpu_pmu);
if (ret)
return ret;
cpu_pmu->name = "armv8_brcm_vulcan"; cpu_pmu->name = "armv8_brcm_vulcan";
cpu_pmu->map_event = armv8_vulcan_map_event; cpu_pmu->map_event = armv8_vulcan_map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] =
&armv8_pmuv3_events_attr_group; &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] =
&armv8_pmuv3_format_attr_group; &armv8_pmuv3_format_attr_group;
return armv8pmu_probe_pmu(cpu_pmu);
return 0;
} }
static const struct of_device_id armv8_pmu_of_device_ids[] = { static const struct of_device_id armv8_pmu_of_device_ids[] = {
...@@ -1081,24 +1136,9 @@ static const struct of_device_id armv8_pmu_of_device_ids[] = { ...@@ -1081,24 +1136,9 @@ static const struct of_device_id armv8_pmu_of_device_ids[] = {
{}, {},
}; };
/*
* Non DT systems have their micro/arch events probed at run-time.
* A fairly complete list of generic events are provided and ones that
* aren't supported by the current PMU are disabled.
*/
static const struct pmu_probe_info armv8_pmu_probe_table[] = {
PMU_PROBE(0, 0, armv8_pmuv3_init), /* enable all defined counters */
{ /* sentinel value */ }
};
static int armv8_pmu_device_probe(struct platform_device *pdev) static int armv8_pmu_device_probe(struct platform_device *pdev)
{ {
if (acpi_disabled) return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids, NULL);
return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids,
NULL);
return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids,
armv8_pmu_probe_table);
} }
static struct platform_driver armv8_pmu_driver = { static struct platform_driver armv8_pmu_driver = {
...@@ -1109,4 +1149,11 @@ static struct platform_driver armv8_pmu_driver = { ...@@ -1109,4 +1149,11 @@ static struct platform_driver armv8_pmu_driver = {
.probe = armv8_pmu_device_probe, .probe = armv8_pmu_device_probe,
}; };
builtin_platform_driver(armv8_pmu_driver); static int __init armv8_pmu_driver_init(void)
{
if (acpi_disabled)
return platform_driver_register(&armv8_pmu_driver);
else
return arm_pmu_acpi_probe(armv8_pmuv3_init);
}
device_initcall(armv8_pmu_driver_init)
...@@ -521,6 +521,13 @@ static bool bootcpu_valid __initdata; ...@@ -521,6 +521,13 @@ static bool bootcpu_valid __initdata;
static unsigned int cpu_count = 1; static unsigned int cpu_count = 1;
#ifdef CONFIG_ACPI #ifdef CONFIG_ACPI
static struct acpi_madt_generic_interrupt cpu_madt_gicc[NR_CPUS];
struct acpi_madt_generic_interrupt *acpi_cpu_get_madt_gicc(int cpu)
{
return &cpu_madt_gicc[cpu];
}
/* /*
* acpi_map_gic_cpu_interface - parse processor MADT entry * acpi_map_gic_cpu_interface - parse processor MADT entry
* *
...@@ -555,6 +562,7 @@ acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor) ...@@ -555,6 +562,7 @@ acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
return; return;
} }
bootcpu_valid = true; bootcpu_valid = true;
cpu_madt_gicc[0] = *processor;
early_map_cpu_to_node(0, acpi_numa_get_nid(0, hwid)); early_map_cpu_to_node(0, acpi_numa_get_nid(0, hwid));
return; return;
} }
...@@ -565,6 +573,8 @@ acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor) ...@@ -565,6 +573,8 @@ acpi_map_gic_cpu_interface(struct acpi_madt_generic_interrupt *processor)
/* map the logical cpu id to cpu MPIDR */ /* map the logical cpu id to cpu MPIDR */
cpu_logical_map(cpu_count) = hwid; cpu_logical_map(cpu_count) = hwid;
cpu_madt_gicc[cpu_count] = *processor;
/* /*
* Set-up the ACPI parking protocol cpu entries * Set-up the ACPI parking protocol cpu entries
* while initializing the cpu_logical_map to * while initializing the cpu_logical_map to
......
...@@ -12,6 +12,10 @@ config ARM_PMU ...@@ -12,6 +12,10 @@ config ARM_PMU
Say y if you want to use CPU performance monitors on ARM-based Say y if you want to use CPU performance monitors on ARM-based
systems. systems.
config ARM_PMU_ACPI
depends on ARM_PMU && ACPI
def_bool y
config QCOM_L2_PMU config QCOM_L2_PMU
bool "Qualcomm Technologies L2-cache PMU" bool "Qualcomm Technologies L2-cache PMU"
depends on ARCH_QCOM && ARM64 && PERF_EVENTS && ACPI depends on ARCH_QCOM && ARM64 && PERF_EVENTS && ACPI
...@@ -21,6 +25,16 @@ config QCOM_L2_PMU ...@@ -21,6 +25,16 @@ config QCOM_L2_PMU
Adds the L2 cache PMU into the perf events subsystem for Adds the L2 cache PMU into the perf events subsystem for
monitoring L2 cache events. monitoring L2 cache events.
config QCOM_L3_PMU
bool "Qualcomm Technologies L3-cache PMU"
depends on ARCH_QCOM && ARM64 && PERF_EVENTS && ACPI
select QCOM_IRQ_COMBINER
help
Provides support for the L3 cache performance monitor unit (PMU)
in Qualcomm Technologies processors.
Adds the L3 cache PMU into the perf events subsystem for
monitoring L3 cache events.
config XGENE_PMU config XGENE_PMU
depends on PERF_EVENTS && ARCH_XGENE depends on PERF_EVENTS && ARCH_XGENE
bool "APM X-Gene SoC PMU" bool "APM X-Gene SoC PMU"
......
obj-$(CONFIG_ARM_PMU) += arm_pmu.o obj-$(CONFIG_ARM_PMU) += arm_pmu.o arm_pmu_platform.o
obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o
obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o obj-$(CONFIG_QCOM_L2_PMU) += qcom_l2_pmu.o
obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o
obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o
...@@ -16,7 +16,6 @@ ...@@ -16,7 +16,6 @@
#include <linux/cpu_pm.h> #include <linux/cpu_pm.h>
#include <linux/export.h> #include <linux/export.h>
#include <linux/kernel.h> #include <linux/kernel.h>
#include <linux/of_device.h>
#include <linux/perf/arm_pmu.h> #include <linux/perf/arm_pmu.h>
#include <linux/platform_device.h> #include <linux/platform_device.h>
#include <linux/slab.h> #include <linux/slab.h>
...@@ -25,7 +24,6 @@ ...@@ -25,7 +24,6 @@
#include <linux/irq.h> #include <linux/irq.h>
#include <linux/irqdesc.h> #include <linux/irqdesc.h>
#include <asm/cputype.h>
#include <asm/irq_regs.h> #include <asm/irq_regs.h>
static int static int
...@@ -235,20 +233,15 @@ armpmu_add(struct perf_event *event, int flags) ...@@ -235,20 +233,15 @@ armpmu_add(struct perf_event *event, int flags)
struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events); struct pmu_hw_events *hw_events = this_cpu_ptr(armpmu->hw_events);
struct hw_perf_event *hwc = &event->hw; struct hw_perf_event *hwc = &event->hw;
int idx; int idx;
int err = 0;
/* An event following a process won't be stopped earlier */ /* An event following a process won't be stopped earlier */
if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus)) if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
return -ENOENT; return -ENOENT;
perf_pmu_disable(event->pmu);
/* If we don't have a space for the counter then finish early. */ /* If we don't have a space for the counter then finish early. */
idx = armpmu->get_event_idx(hw_events, event); idx = armpmu->get_event_idx(hw_events, event);
if (idx < 0) { if (idx < 0)
err = idx; return idx;
goto out;
}
/* /*
* If there is an event in the counter we are going to use then make * If there is an event in the counter we are going to use then make
...@@ -265,9 +258,7 @@ armpmu_add(struct perf_event *event, int flags) ...@@ -265,9 +258,7 @@ armpmu_add(struct perf_event *event, int flags)
/* Propagate our changes to the userspace mapping. */ /* Propagate our changes to the userspace mapping. */
perf_event_update_userpage(event); perf_event_update_userpage(event);
out: return 0;
perf_pmu_enable(event->pmu);
return err;
} }
static int static int
...@@ -323,10 +314,16 @@ validate_group(struct perf_event *event) ...@@ -323,10 +314,16 @@ validate_group(struct perf_event *event)
return 0; return 0;
} }
static struct arm_pmu_platdata *armpmu_get_platdata(struct arm_pmu *armpmu)
{
struct platform_device *pdev = armpmu->plat_device;
return pdev ? dev_get_platdata(&pdev->dev) : NULL;
}
static irqreturn_t armpmu_dispatch_irq(int irq, void *dev) static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
{ {
struct arm_pmu *armpmu; struct arm_pmu *armpmu;
struct platform_device *plat_device;
struct arm_pmu_platdata *plat; struct arm_pmu_platdata *plat;
int ret; int ret;
u64 start_clock, finish_clock; u64 start_clock, finish_clock;
...@@ -338,8 +335,8 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev) ...@@ -338,8 +335,8 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
* dereference. * dereference.
*/ */
armpmu = *(void **)dev; armpmu = *(void **)dev;
plat_device = armpmu->plat_device;
plat = dev_get_platdata(&plat_device->dev); plat = armpmu_get_platdata(armpmu);
start_clock = sched_clock(); start_clock = sched_clock();
if (plat && plat->handle_irq) if (plat && plat->handle_irq)
...@@ -352,37 +349,6 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev) ...@@ -352,37 +349,6 @@ static irqreturn_t armpmu_dispatch_irq(int irq, void *dev)
return ret; return ret;
} }
static void
armpmu_release_hardware(struct arm_pmu *armpmu)
{
armpmu->free_irq(armpmu);
}
static int
armpmu_reserve_hardware(struct arm_pmu *armpmu)
{
int err = armpmu->request_irq(armpmu, armpmu_dispatch_irq);
if (err) {
armpmu_release_hardware(armpmu);
return err;
}
return 0;
}
static void
hw_perf_event_destroy(struct perf_event *event)
{
struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
atomic_t *active_events = &armpmu->active_events;
struct mutex *pmu_reserve_mutex = &armpmu->reserve_mutex;
if (atomic_dec_and_mutex_lock(active_events, pmu_reserve_mutex)) {
armpmu_release_hardware(armpmu);
mutex_unlock(pmu_reserve_mutex);
}
}
static int static int
event_requires_mode_exclusion(struct perf_event_attr *attr) event_requires_mode_exclusion(struct perf_event_attr *attr)
{ {
...@@ -455,8 +421,6 @@ __hw_perf_event_init(struct perf_event *event) ...@@ -455,8 +421,6 @@ __hw_perf_event_init(struct perf_event *event)
static int armpmu_event_init(struct perf_event *event) static int armpmu_event_init(struct perf_event *event)
{ {
struct arm_pmu *armpmu = to_arm_pmu(event->pmu); struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
int err = 0;
atomic_t *active_events = &armpmu->active_events;
/* /*
* Reject CPU-affine events for CPUs that are of a different class to * Reject CPU-affine events for CPUs that are of a different class to
...@@ -476,26 +440,7 @@ static int armpmu_event_init(struct perf_event *event) ...@@ -476,26 +440,7 @@ static int armpmu_event_init(struct perf_event *event)
if (armpmu->map_event(event) == -ENOENT) if (armpmu->map_event(event) == -ENOENT)
return -ENOENT; return -ENOENT;
event->destroy = hw_perf_event_destroy; return __hw_perf_event_init(event);
if (!atomic_inc_not_zero(active_events)) {
mutex_lock(&armpmu->reserve_mutex);
if (atomic_read(active_events) == 0)
err = armpmu_reserve_hardware(armpmu);
if (!err)
atomic_inc(active_events);
mutex_unlock(&armpmu->reserve_mutex);
}
if (err)
return err;
err = __hw_perf_event_init(event);
if (err)
hw_perf_event_destroy(event);
return err;
} }
static void armpmu_enable(struct pmu *pmu) static void armpmu_enable(struct pmu *pmu)
...@@ -553,27 +498,6 @@ static struct attribute_group armpmu_common_attr_group = { ...@@ -553,27 +498,6 @@ static struct attribute_group armpmu_common_attr_group = {
.attrs = armpmu_common_attrs, .attrs = armpmu_common_attrs,
}; };
static void armpmu_init(struct arm_pmu *armpmu)
{
atomic_set(&armpmu->active_events, 0);
mutex_init(&armpmu->reserve_mutex);
armpmu->pmu = (struct pmu) {
.pmu_enable = armpmu_enable,
.pmu_disable = armpmu_disable,
.event_init = armpmu_event_init,
.add = armpmu_add,
.del = armpmu_del,
.start = armpmu_start,
.stop = armpmu_stop,
.read = armpmu_read,
.filter_match = armpmu_filter_match,
.attr_groups = armpmu->attr_groups,
};
armpmu->attr_groups[ARMPMU_ATTR_GROUP_COMMON] =
&armpmu_common_attr_group;
}
/* Set at runtime when we know what CPU type we are. */ /* Set at runtime when we know what CPU type we are. */
static struct arm_pmu *__oprofile_cpu_pmu; static struct arm_pmu *__oprofile_cpu_pmu;
...@@ -601,115 +525,87 @@ int perf_num_counters(void) ...@@ -601,115 +525,87 @@ int perf_num_counters(void)
} }
EXPORT_SYMBOL_GPL(perf_num_counters); EXPORT_SYMBOL_GPL(perf_num_counters);
static void cpu_pmu_enable_percpu_irq(void *data) void armpmu_free_irq(struct arm_pmu *armpmu, int cpu)
{ {
int irq = *(int *)data; struct pmu_hw_events __percpu *hw_events = armpmu->hw_events;
int irq = per_cpu(hw_events->irq, cpu);
enable_percpu_irq(irq, IRQ_TYPE_NONE); if (!cpumask_test_and_clear_cpu(cpu, &armpmu->active_irqs))
} return;
static void cpu_pmu_disable_percpu_irq(void *data)
{
int irq = *(int *)data;
disable_percpu_irq(irq);
}
static void cpu_pmu_free_irq(struct arm_pmu *cpu_pmu)
{
int i, irq, irqs;
struct platform_device *pmu_device = cpu_pmu->plat_device;
struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
irqs = min(pmu_device->num_resources, num_possible_cpus());
irq = platform_get_irq(pmu_device, 0); if (irq_is_percpu(irq)) {
if (irq > 0 && irq_is_percpu(irq)) {
on_each_cpu_mask(&cpu_pmu->supported_cpus,
cpu_pmu_disable_percpu_irq, &irq, 1);
free_percpu_irq(irq, &hw_events->percpu_pmu); free_percpu_irq(irq, &hw_events->percpu_pmu);
} else { cpumask_clear(&armpmu->active_irqs);
for (i = 0; i < irqs; ++i) { return;
int cpu = i; }
if (cpu_pmu->irq_affinity)
cpu = cpu_pmu->irq_affinity[i];
if (!cpumask_test_and_clear_cpu(cpu, &cpu_pmu->active_irqs))
continue;
irq = platform_get_irq(pmu_device, i);
if (irq > 0)
free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, cpu)); free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, cpu));
}
}
} }
static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler) void armpmu_free_irqs(struct arm_pmu *armpmu)
{ {
int i, err, irq, irqs; int cpu;
struct platform_device *pmu_device = cpu_pmu->plat_device;
struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
if (!pmu_device) for_each_cpu(cpu, &armpmu->supported_cpus)
return -ENODEV; armpmu_free_irq(armpmu, cpu);
}
irqs = min(pmu_device->num_resources, num_possible_cpus()); int armpmu_request_irq(struct arm_pmu *armpmu, int cpu)
if (irqs < 1) { {
pr_warn_once("perf/ARM: No irqs for PMU defined, sampling events not supported\n"); int err = 0;
struct pmu_hw_events __percpu *hw_events = armpmu->hw_events;
const irq_handler_t handler = armpmu_dispatch_irq;
int irq = per_cpu(hw_events->irq, cpu);
if (!irq)
return 0; return 0;
}
irq = platform_get_irq(pmu_device, 0); if (irq_is_percpu(irq) && cpumask_empty(&armpmu->active_irqs)) {
if (irq > 0 && irq_is_percpu(irq)) {
err = request_percpu_irq(irq, handler, "arm-pmu", err = request_percpu_irq(irq, handler, "arm-pmu",
&hw_events->percpu_pmu); &hw_events->percpu_pmu);
if (err) { } else if (irq_is_percpu(irq)) {
pr_err("unable to request IRQ%d for ARM PMU counters\n", int other_cpu = cpumask_first(&armpmu->active_irqs);
irq); int other_irq = per_cpu(hw_events->irq, other_cpu);
return err;
}
on_each_cpu_mask(&cpu_pmu->supported_cpus,
cpu_pmu_enable_percpu_irq, &irq, 1);
} else {
for (i = 0; i < irqs; ++i) {
int cpu = i;
err = 0;
irq = platform_get_irq(pmu_device, i);
if (irq < 0)
continue;
if (cpu_pmu->irq_affinity)
cpu = cpu_pmu->irq_affinity[i];
/* if (irq != other_irq) {
* If we have a single PMU interrupt that we can't shift, pr_warn("mismatched PPIs detected.\n");
* assume that we're running on a uniprocessor machine and err = -EINVAL;
* continue. Otherwise, continue without this interrupt.
*/
if (irq_set_affinity(irq, cpumask_of(cpu)) && irqs > 1) {
pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
irq, cpu);
continue;
} }
} else {
err = request_irq(irq, handler, err = request_irq(irq, handler,
IRQF_NOBALANCING | IRQF_NO_THREAD, "arm-pmu", IRQF_NOBALANCING | IRQF_NO_THREAD, "arm-pmu",
per_cpu_ptr(&hw_events->percpu_pmu, cpu)); per_cpu_ptr(&hw_events->percpu_pmu, cpu));
}
if (err) { if (err) {
pr_err("unable to request IRQ%d for ARM PMU counters\n", pr_err("unable to request IRQ%d for ARM PMU counters\n",
irq); irq);
return err; return err;
} }
cpumask_set_cpu(cpu, &cpu_pmu->active_irqs); cpumask_set_cpu(cpu, &armpmu->active_irqs);
}
}
return 0; return 0;
} }
int armpmu_request_irqs(struct arm_pmu *armpmu)
{
int cpu, err;
for_each_cpu(cpu, &armpmu->supported_cpus) {
err = armpmu_request_irq(armpmu, cpu);
if (err)
break;
}
return err;
}
static int armpmu_get_cpu_irq(struct arm_pmu *pmu, int cpu)
{
struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
return per_cpu(hw_events->irq, cpu);
}
/* /*
* PMU hardware loses all context when a CPU goes offline. * PMU hardware loses all context when a CPU goes offline.
* When a CPU is hotplugged back in, since some hardware registers are * When a CPU is hotplugged back in, since some hardware registers are
...@@ -719,11 +615,42 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler) ...@@ -719,11 +615,42 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node) static int arm_perf_starting_cpu(unsigned int cpu, struct hlist_node *node)
{ {
struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node); struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
int irq;
if (!cpumask_test_cpu(cpu, &pmu->supported_cpus)) if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
return 0; return 0;
if (pmu->reset) if (pmu->reset)
pmu->reset(pmu); pmu->reset(pmu);
irq = armpmu_get_cpu_irq(pmu, cpu);
if (irq) {
if (irq_is_percpu(irq)) {
enable_percpu_irq(irq, IRQ_TYPE_NONE);
return 0;
}
if (irq_force_affinity(irq, cpumask_of(cpu)) &&
num_possible_cpus() > 1) {
pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
irq, cpu);
}
}
return 0;
}
static int arm_perf_teardown_cpu(unsigned int cpu, struct hlist_node *node)
{
struct arm_pmu *pmu = hlist_entry_safe(node, struct arm_pmu, node);
int irq;
if (!cpumask_test_cpu(cpu, &pmu->supported_cpus))
return 0;
irq = armpmu_get_cpu_irq(pmu, cpu);
if (irq && irq_is_percpu(irq))
disable_percpu_irq(irq);
return 0; return 0;
} }
...@@ -828,56 +755,22 @@ static inline void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu) { } ...@@ -828,56 +755,22 @@ static inline void cpu_pm_pmu_unregister(struct arm_pmu *cpu_pmu) { }
static int cpu_pmu_init(struct arm_pmu *cpu_pmu) static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
{ {
int err; int err;
int cpu;
struct pmu_hw_events __percpu *cpu_hw_events;
cpu_hw_events = alloc_percpu(struct pmu_hw_events);
if (!cpu_hw_events)
return -ENOMEM;
err = cpuhp_state_add_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING, err = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_STARTING,
&cpu_pmu->node); &cpu_pmu->node);
if (err) if (err)
goto out_free; goto out;
err = cpu_pm_pmu_register(cpu_pmu); err = cpu_pm_pmu_register(cpu_pmu);
if (err) if (err)
goto out_unregister; goto out_unregister;
for_each_possible_cpu(cpu) {
struct pmu_hw_events *events = per_cpu_ptr(cpu_hw_events, cpu);
raw_spin_lock_init(&events->pmu_lock);
events->percpu_pmu = cpu_pmu;
}
cpu_pmu->hw_events = cpu_hw_events;
cpu_pmu->request_irq = cpu_pmu_request_irq;
cpu_pmu->free_irq = cpu_pmu_free_irq;
/* Ensure the PMU has sane values out of reset. */
if (cpu_pmu->reset)
on_each_cpu_mask(&cpu_pmu->supported_cpus, cpu_pmu->reset,
cpu_pmu, 1);
/* If no interrupts available, set the corresponding capability flag */
if (!platform_get_irq(cpu_pmu->plat_device, 0))
cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
/*
* This is a CPU PMU potentially in a heterogeneous configuration (e.g.
* big.LITTLE). This is not an uncore PMU, and we have taken ctx
* sharing into account (e.g. with our pmu::filter_match callback and
* pmu::event_init group validation).
*/
cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_HETEROGENEOUS_CPUS;
return 0; return 0;
out_unregister: out_unregister:
cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING, cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
&cpu_pmu->node); &cpu_pmu->node);
out_free: out:
free_percpu(cpu_hw_events);
return err; return err;
} }
...@@ -886,177 +779,78 @@ static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu) ...@@ -886,177 +779,78 @@ static void cpu_pmu_destroy(struct arm_pmu *cpu_pmu)
cpu_pm_pmu_unregister(cpu_pmu); cpu_pm_pmu_unregister(cpu_pmu);
cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING, cpuhp_state_remove_instance_nocalls(CPUHP_AP_PERF_ARM_STARTING,
&cpu_pmu->node); &cpu_pmu->node);
free_percpu(cpu_pmu->hw_events);
} }
/* struct arm_pmu *armpmu_alloc(void)
* CPU PMU identification and probing.
*/
static int probe_current_pmu(struct arm_pmu *pmu,
const struct pmu_probe_info *info)
{ {
int cpu = get_cpu(); struct arm_pmu *pmu;
unsigned int cpuid = read_cpuid_id(); int cpu;
int ret = -ENODEV;
pr_info("probing PMU on CPU %d\n", cpu);
for (; info->init != NULL; info++) {
if ((cpuid & info->mask) != info->cpuid)
continue;
ret = info->init(pmu);
break;
}
put_cpu();
return ret;
}
static int of_pmu_irq_cfg(struct arm_pmu *pmu)
{
int *irqs, i = 0;
bool using_spi = false;
struct platform_device *pdev = pmu->plat_device;
irqs = kcalloc(pdev->num_resources, sizeof(*irqs), GFP_KERNEL);
if (!irqs)
return -ENOMEM;
do {
struct device_node *dn;
int cpu, irq;
/* See if we have an affinity entry */
dn = of_parse_phandle(pdev->dev.of_node, "interrupt-affinity", i);
if (!dn)
break;
/* Check the IRQ type and prohibit a mix of PPIs and SPIs */
irq = platform_get_irq(pdev, i);
if (irq > 0) {
bool spi = !irq_is_percpu(irq);
if (i > 0 && spi != using_spi) {
pr_err("PPI/SPI IRQ type mismatch for %s!\n",
dn->name);
of_node_put(dn);
kfree(irqs);
return -EINVAL;
}
using_spi = spi;
}
/* Now look up the logical CPU number */
for_each_possible_cpu(cpu) {
struct device_node *cpu_dn;
cpu_dn = of_cpu_device_node_get(cpu);
of_node_put(cpu_dn);
if (dn == cpu_dn)
break;
}
if (cpu >= nr_cpu_ids) { pmu = kzalloc(sizeof(*pmu), GFP_KERNEL);
pr_warn("Failed to find logical CPU for %s\n", if (!pmu) {
dn->name); pr_info("failed to allocate PMU device!\n");
of_node_put(dn); goto out;
cpumask_setall(&pmu->supported_cpus);
break;
} }
of_node_put(dn);
/* For SPIs, we need to track the affinity per IRQ */
if (using_spi) {
if (i >= pdev->num_resources)
break;
irqs[i] = cpu; pmu->hw_events = alloc_percpu(struct pmu_hw_events);
if (!pmu->hw_events) {
pr_info("failed to allocate per-cpu PMU data.\n");
goto out_free_pmu;
} }
/* Keep track of the CPUs containing this PMU type */ pmu->pmu = (struct pmu) {
cpumask_set_cpu(cpu, &pmu->supported_cpus); .pmu_enable = armpmu_enable,
i++; .pmu_disable = armpmu_disable,
} while (1); .event_init = armpmu_event_init,
.add = armpmu_add,
.del = armpmu_del,
.start = armpmu_start,
.stop = armpmu_stop,
.read = armpmu_read,
.filter_match = armpmu_filter_match,
.attr_groups = pmu->attr_groups,
/*
* This is a CPU PMU potentially in a heterogeneous
* configuration (e.g. big.LITTLE). This is not an uncore PMU,
* and we have taken ctx sharing into account (e.g. with our
* pmu::filter_match callback and pmu::event_init group
* validation).
*/
.capabilities = PERF_PMU_CAP_HETEROGENEOUS_CPUS,
};
/* If we didn't manage to parse anything, try the interrupt affinity */ pmu->attr_groups[ARMPMU_ATTR_GROUP_COMMON] =
if (cpumask_weight(&pmu->supported_cpus) == 0) { &armpmu_common_attr_group;
int irq = platform_get_irq(pdev, 0);
if (irq > 0 && irq_is_percpu(irq)) { for_each_possible_cpu(cpu) {
/* If using PPIs, check the affinity of the partition */ struct pmu_hw_events *events;
int ret;
ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus); events = per_cpu_ptr(pmu->hw_events, cpu);
if (ret) { raw_spin_lock_init(&events->pmu_lock);
kfree(irqs); events->percpu_pmu = pmu;
return ret;
}
} else {
/* Otherwise default to all CPUs */
cpumask_setall(&pmu->supported_cpus);
}
} }
/* If we matched up the IRQ affinities, use them to route the SPIs */ return pmu;
if (using_spi && i == pdev->num_resources)
pmu->irq_affinity = irqs;
else
kfree(irqs);
return 0; out_free_pmu:
kfree(pmu);
out:
return NULL;
} }
int arm_pmu_device_probe(struct platform_device *pdev, void armpmu_free(struct arm_pmu *pmu)
const struct of_device_id *of_table,
const struct pmu_probe_info *probe_table)
{ {
const struct of_device_id *of_id; free_percpu(pmu->hw_events);
const int (*init_fn)(struct arm_pmu *); kfree(pmu);
struct device_node *node = pdev->dev.of_node; }
struct arm_pmu *pmu;
int ret = -ENODEV;
pmu = kzalloc(sizeof(struct arm_pmu), GFP_KERNEL);
if (!pmu) {
pr_info("failed to allocate PMU device!\n");
return -ENOMEM;
}
armpmu_init(pmu);
pmu->plat_device = pdev;
if (node && (of_id = of_match_node(of_table, pdev->dev.of_node))) {
init_fn = of_id->data;
pmu->secure_access = of_property_read_bool(pdev->dev.of_node,
"secure-reg-access");
/* arm64 systems boot only as non-secure */
if (IS_ENABLED(CONFIG_ARM64) && pmu->secure_access) {
pr_warn("ignoring \"secure-reg-access\" property for arm64\n");
pmu->secure_access = false;
}
ret = of_pmu_irq_cfg(pmu);
if (!ret)
ret = init_fn(pmu);
} else if (probe_table) {
cpumask_setall(&pmu->supported_cpus);
ret = probe_current_pmu(pmu, probe_table);
}
if (ret) {
pr_info("%s: failed to probe PMU!\n", of_node_full_name(node));
goto out_free;
}
int armpmu_register(struct arm_pmu *pmu)
{
int ret;
ret = cpu_pmu_init(pmu); ret = cpu_pmu_init(pmu);
if (ret) if (ret)
goto out_free; return ret;
ret = perf_pmu_register(&pmu->pmu, pmu->name, -1); ret = perf_pmu_register(&pmu->pmu, pmu->name, -1);
if (ret) if (ret)
...@@ -1072,11 +866,6 @@ int arm_pmu_device_probe(struct platform_device *pdev, ...@@ -1072,11 +866,6 @@ int arm_pmu_device_probe(struct platform_device *pdev,
out_destroy: out_destroy:
cpu_pmu_destroy(pmu); cpu_pmu_destroy(pmu);
out_free:
pr_info("%s: failed to register PMU devices!\n",
of_node_full_name(node));
kfree(pmu->irq_affinity);
kfree(pmu);
return ret; return ret;
} }
...@@ -1086,7 +875,8 @@ static int arm_pmu_hp_init(void) ...@@ -1086,7 +875,8 @@ static int arm_pmu_hp_init(void)
ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_STARTING, ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_STARTING,
"perf/arm/pmu:starting", "perf/arm/pmu:starting",
arm_perf_starting_cpu, NULL); arm_perf_starting_cpu,
arm_perf_teardown_cpu);
if (ret) if (ret)
pr_err("CPU hotplug notifier for ARM PMU could not be registered: %d\n", pr_err("CPU hotplug notifier for ARM PMU could not be registered: %d\n",
ret); ret);
......
/*
* ACPI probing code for ARM performance counters.
*
* Copyright (C) 2017 ARM Ltd.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*/
#include <linux/acpi.h>
#include <linux/cpumask.h>
#include <linux/init.h>
#include <linux/percpu.h>
#include <linux/perf/arm_pmu.h>
#include <asm/cputype.h>
static DEFINE_PER_CPU(struct arm_pmu *, probed_pmus);
static DEFINE_PER_CPU(int, pmu_irqs);
static int arm_pmu_acpi_register_irq(int cpu)
{
struct acpi_madt_generic_interrupt *gicc;
int gsi, trigger;
gicc = acpi_cpu_get_madt_gicc(cpu);
if (WARN_ON(!gicc))
return -EINVAL;
gsi = gicc->performance_interrupt;
if (gicc->flags & ACPI_MADT_PERFORMANCE_IRQ_MODE)
trigger = ACPI_EDGE_SENSITIVE;
else
trigger = ACPI_LEVEL_SENSITIVE;
/*
* Helpfully, the MADT GICC doesn't have a polarity flag for the
* "performance interrupt". Luckily, on compliant GICs the polarity is
* a fixed value in HW (for both SPIs and PPIs) that we cannot change
* from SW.
*
* Here we pass in ACPI_ACTIVE_HIGH to keep the core code happy. This
* may not match the real polarity, but that should not matter.
*
* Other interrupt controllers are not supported with ACPI.
*/
return acpi_register_gsi(NULL, gsi, trigger, ACPI_ACTIVE_HIGH);
}
static void arm_pmu_acpi_unregister_irq(int cpu)
{
struct acpi_madt_generic_interrupt *gicc;
int gsi;
gicc = acpi_cpu_get_madt_gicc(cpu);
if (!gicc)
return;
gsi = gicc->performance_interrupt;
acpi_unregister_gsi(gsi);
}
static int arm_pmu_acpi_parse_irqs(void)
{
int irq, cpu, irq_cpu, err;
for_each_possible_cpu(cpu) {
irq = arm_pmu_acpi_register_irq(cpu);
if (irq < 0) {
err = irq;
pr_warn("Unable to parse ACPI PMU IRQ for CPU%d: %d\n",
cpu, err);
goto out_err;
} else if (irq == 0) {
pr_warn("No ACPI PMU IRQ for CPU%d\n", cpu);
}
per_cpu(pmu_irqs, cpu) = irq;
}
return 0;
out_err:
for_each_possible_cpu(cpu) {
irq = per_cpu(pmu_irqs, cpu);
if (!irq)
continue;
arm_pmu_acpi_unregister_irq(cpu);
/*
* Blat all copies of the IRQ so that we only unregister the
* corresponding GSI once (e.g. when we have PPIs).
*/
for_each_possible_cpu(irq_cpu) {
if (per_cpu(pmu_irqs, irq_cpu) == irq)
per_cpu(pmu_irqs, irq_cpu) = 0;
}
}
return err;
}
static struct arm_pmu *arm_pmu_acpi_find_alloc_pmu(void)
{
unsigned long cpuid = read_cpuid_id();
struct arm_pmu *pmu;
int cpu;
for_each_possible_cpu(cpu) {
pmu = per_cpu(probed_pmus, cpu);
if (!pmu || pmu->acpi_cpuid != cpuid)
continue;
return pmu;
}
pmu = armpmu_alloc();
if (!pmu) {
pr_warn("Unable to allocate PMU for CPU%d\n",
smp_processor_id());
return NULL;
}
pmu->acpi_cpuid = cpuid;
return pmu;
}
/*
* This must run before the common arm_pmu hotplug logic, so that we can
* associate a CPU and its interrupt before the common code tries to manage the
* affinity and so on.
*
* Note that hotplug events are serialized, so we cannot race with another CPU
* coming up. The perf core won't open events while a hotplug event is in
* progress.
*/
static int arm_pmu_acpi_cpu_starting(unsigned int cpu)
{
struct arm_pmu *pmu;
struct pmu_hw_events __percpu *hw_events;
int irq;
/* If we've already probed this CPU, we have nothing to do */
if (per_cpu(probed_pmus, cpu))
return 0;
irq = per_cpu(pmu_irqs, cpu);
pmu = arm_pmu_acpi_find_alloc_pmu();
if (!pmu)
return -ENOMEM;
cpumask_set_cpu(cpu, &pmu->supported_cpus);
per_cpu(probed_pmus, cpu) = pmu;
/*
* Log and request the IRQ so the core arm_pmu code can manage it. In
* some situations (e.g. mismatched PPIs), we may fail to request the
* IRQ. However, it may be too late for us to do anything about it.
* The common ARM PMU code will log a warning in this case.
*/
hw_events = pmu->hw_events;
per_cpu(hw_events->irq, cpu) = irq;
armpmu_request_irq(pmu, cpu);
/*
* Ideally, we'd probe the PMU here when we find the first matching
* CPU. We can't do that for several reasons; see the comment in
* arm_pmu_acpi_init().
*
* So for the time being, we're done.
*/
return 0;
}
int arm_pmu_acpi_probe(armpmu_init_fn init_fn)
{
int pmu_idx = 0;
int cpu, ret;
if (acpi_disabled)
return 0;
/*
* Initialise and register the set of PMUs which we know about right
* now. Ideally we'd do this in arm_pmu_acpi_cpu_starting() so that we
* could handle late hotplug, but this may lead to deadlock since we
* might try to register a hotplug notifier instance from within a
* hotplug notifier.
*
* There's also the problem of having access to the right init_fn,
* without tying this too deeply into the "real" PMU driver.
*
* For the moment, as with the platform/DT case, we need at least one
* of a PMU's CPUs to be online at probe time.
*/
for_each_possible_cpu(cpu) {
struct arm_pmu *pmu = per_cpu(probed_pmus, cpu);
char *base_name;
if (!pmu || pmu->name)
continue;
ret = init_fn(pmu);
if (ret == -ENODEV) {
/* PMU not handled by this driver, or not present */
continue;
} else if (ret) {
pr_warn("Unable to initialise PMU for CPU%d\n", cpu);
return ret;
}
base_name = pmu->name;
pmu->name = kasprintf(GFP_KERNEL, "%s_%d", base_name, pmu_idx++);
if (!pmu->name) {
pr_warn("Unable to allocate PMU name for CPU%d\n", cpu);
return -ENOMEM;
}
ret = armpmu_register(pmu);
if (ret) {
pr_warn("Failed to register PMU for CPU%d\n", cpu);
return ret;
}
}
return 0;
}
static int arm_pmu_acpi_init(void)
{
int ret;
if (acpi_disabled)
return 0;
/*
* We can't request IRQs yet, since we don't know the cookie value
* until we know which CPUs share the same logical PMU. We'll handle
* that in arm_pmu_acpi_cpu_starting().
*/
ret = arm_pmu_acpi_parse_irqs();
if (ret)
return ret;
ret = cpuhp_setup_state(CPUHP_AP_PERF_ARM_ACPI_STARTING,
"perf/arm/pmu_acpi:starting",
arm_pmu_acpi_cpu_starting, NULL);
return ret;
}
subsys_initcall(arm_pmu_acpi_init)
/*
* platform_device probing code for ARM performance counters.
*
* Copyright (C) 2009 picoChip Designs, Ltd., Jamie Iles
* Copyright (C) 2010 ARM Ltd., Will Deacon <will.deacon@arm.com>
*/
#define pr_fmt(fmt) "hw perfevents: " fmt
#include <linux/bug.h>
#include <linux/cpumask.h>
#include <linux/device.h>
#include <linux/errno.h>
#include <linux/irq.h>
#include <linux/irqdesc.h>
#include <linux/kconfig.h>
#include <linux/of.h>
#include <linux/of_device.h>
#include <linux/percpu.h>
#include <linux/perf/arm_pmu.h>
#include <linux/platform_device.h>
#include <linux/printk.h>
#include <linux/smp.h>
static int probe_current_pmu(struct arm_pmu *pmu,
const struct pmu_probe_info *info)
{
int cpu = get_cpu();
unsigned int cpuid = read_cpuid_id();
int ret = -ENODEV;
pr_info("probing PMU on CPU %d\n", cpu);
for (; info->init != NULL; info++) {
if ((cpuid & info->mask) != info->cpuid)
continue;
ret = info->init(pmu);
break;
}
put_cpu();
return ret;
}
static int pmu_parse_percpu_irq(struct arm_pmu *pmu, int irq)
{
int cpu, ret;
struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
if (ret)
return ret;
for_each_cpu(cpu, &pmu->supported_cpus)
per_cpu(hw_events->irq, cpu) = irq;
return 0;
}
static bool pmu_has_irq_affinity(struct device_node *node)
{
return !!of_find_property(node, "interrupt-affinity", NULL);
}
static int pmu_parse_irq_affinity(struct device_node *node, int i)
{
struct device_node *dn;
int cpu;
/*
* If we don't have an interrupt-affinity property, we guess irq
* affinity matches our logical CPU order, as we used to assume.
* This is fragile, so we'll warn in pmu_parse_irqs().
*/
if (!pmu_has_irq_affinity(node))
return i;
dn = of_parse_phandle(node, "interrupt-affinity", i);
if (!dn) {
pr_warn("failed to parse interrupt-affinity[%d] for %s\n",
i, node->name);
return -EINVAL;
}
/* Now look up the logical CPU number */
for_each_possible_cpu(cpu) {
struct device_node *cpu_dn;
cpu_dn = of_cpu_device_node_get(cpu);
of_node_put(cpu_dn);
if (dn == cpu_dn)
break;
}
if (cpu >= nr_cpu_ids) {
pr_warn("failed to find logical CPU for %s\n", dn->name);
}
of_node_put(dn);
return cpu;
}
static int pmu_parse_irqs(struct arm_pmu *pmu)
{
int i = 0, num_irqs;
struct platform_device *pdev = pmu->plat_device;
struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
num_irqs = platform_irq_count(pdev);
if (num_irqs < 0) {
pr_err("unable to count PMU IRQs\n");
return num_irqs;
}
/*
* In this case we have no idea which CPUs are covered by the PMU.
* To match our prior behaviour, we assume all CPUs in this case.
*/
if (num_irqs == 0) {
pr_warn("no irqs for PMU, sampling events not supported\n");
pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
cpumask_setall(&pmu->supported_cpus);
return 0;
}
if (num_irqs == 1) {
int irq = platform_get_irq(pdev, 0);
if (irq && irq_is_percpu(irq))
return pmu_parse_percpu_irq(pmu, irq);
}
if (!pmu_has_irq_affinity(pdev->dev.of_node)) {
pr_warn("no interrupt-affinity property for %s, guessing.\n",
of_node_full_name(pdev->dev.of_node));
}
/*
* Some platforms have all PMU IRQs OR'd into a single IRQ, with a
* special platdata function that attempts to demux them.
*/
if (dev_get_platdata(&pdev->dev))
cpumask_setall(&pmu->supported_cpus);
for (i = 0; i < num_irqs; i++) {
int cpu, irq;
irq = platform_get_irq(pdev, i);
if (WARN_ON(irq <= 0))
continue;
if (irq_is_percpu(irq)) {
pr_warn("multiple PPIs or mismatched SPI/PPI detected\n");
return -EINVAL;
}
cpu = pmu_parse_irq_affinity(pdev->dev.of_node, i);
if (cpu < 0)
return cpu;
if (cpu >= nr_cpu_ids)
continue;
if (per_cpu(hw_events->irq, cpu)) {
pr_warn("multiple PMU IRQs for the same CPU detected\n");
return -EINVAL;
}
per_cpu(hw_events->irq, cpu) = irq;
cpumask_set_cpu(cpu, &pmu->supported_cpus);
}
return 0;
}
int arm_pmu_device_probe(struct platform_device *pdev,
const struct of_device_id *of_table,
const struct pmu_probe_info *probe_table)
{
const struct of_device_id *of_id;
armpmu_init_fn init_fn;
struct device_node *node = pdev->dev.of_node;
struct arm_pmu *pmu;
int ret = -ENODEV;
pmu = armpmu_alloc();
if (!pmu)
return -ENOMEM;
pmu->plat_device = pdev;
ret = pmu_parse_irqs(pmu);
if (ret)
goto out_free;
if (node && (of_id = of_match_node(of_table, pdev->dev.of_node))) {
init_fn = of_id->data;
pmu->secure_access = of_property_read_bool(pdev->dev.of_node,
"secure-reg-access");
/* arm64 systems boot only as non-secure */
if (IS_ENABLED(CONFIG_ARM64) && pmu->secure_access) {
pr_warn("ignoring \"secure-reg-access\" property for arm64\n");
pmu->secure_access = false;
}
ret = init_fn(pmu);
} else if (probe_table) {
cpumask_setall(&pmu->supported_cpus);
ret = probe_current_pmu(pmu, probe_table);
}
if (ret) {
pr_info("%s: failed to probe PMU!\n", of_node_full_name(node));
goto out_free;
}
ret = armpmu_request_irqs(pmu);
if (ret)
goto out_free_irqs;
ret = armpmu_register(pmu);
if (ret)
goto out_free;
return 0;
out_free_irqs:
armpmu_free_irqs(pmu);
out_free:
pr_info("%s: failed to register PMU devices!\n",
of_node_full_name(node));
armpmu_free(pmu);
return ret;
}
/*
* Driver for the L3 cache PMUs in Qualcomm Technologies chips.
*
* The driver supports a distributed cache architecture where the overall
* cache for a socket is comprised of multiple slices each with its own PMU.
* Access to each individual PMU is provided even though all CPUs share all
* the slices. User space needs to aggregate to individual counts to provide
* a global picture.
*
* See Documentation/perf/qcom_l3_pmu.txt for more details.
*
* Copyright (c) 2015-2017, The Linux Foundation. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 and
* only 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.
*/
#include <linux/acpi.h>
#include <linux/bitops.h>
#include <linux/interrupt.h>
#include <linux/io.h>
#include <linux/list.h>
#include <linux/module.h>
#include <linux/perf_event.h>
#include <linux/platform_device.h>
/*
* General constants
*/
/* Number of counters on each PMU */
#define L3_NUM_COUNTERS 8
/* Mask for the event type field within perf_event_attr.config and EVTYPE reg */
#define L3_EVTYPE_MASK 0xFF
/*
* Bit position of the 'long counter' flag within perf_event_attr.config.
* Reserve some space between the event type and this flag to allow expansion
* in the event type field.
*/
#define L3_EVENT_LC_BIT 32
/*
* Register offsets
*/
/* Perfmon registers */
#define L3_HML3_PM_CR 0x000
#define L3_HML3_PM_EVCNTR(__cntr) (0x420 + ((__cntr) & 0x7) * 8)
#define L3_HML3_PM_CNTCTL(__cntr) (0x120 + ((__cntr) & 0x7) * 8)
#define L3_HML3_PM_EVTYPE(__cntr) (0x220 + ((__cntr) & 0x7) * 8)
#define L3_HML3_PM_FILTRA 0x300
#define L3_HML3_PM_FILTRB 0x308
#define L3_HML3_PM_FILTRC 0x310
#define L3_HML3_PM_FILTRAM 0x304
#define L3_HML3_PM_FILTRBM 0x30C
#define L3_HML3_PM_FILTRCM 0x314
/* Basic counter registers */
#define L3_M_BC_CR 0x500
#define L3_M_BC_SATROLL_CR 0x504
#define L3_M_BC_CNTENSET 0x508
#define L3_M_BC_CNTENCLR 0x50C
#define L3_M_BC_INTENSET 0x510
#define L3_M_BC_INTENCLR 0x514
#define L3_M_BC_GANG 0x718
#define L3_M_BC_OVSR 0x740
#define L3_M_BC_IRQCTL 0x96C
/*
* Bit field definitions
*/
/* L3_HML3_PM_CR */
#define PM_CR_RESET (0)
/* L3_HML3_PM_XCNTCTL/L3_HML3_PM_CNTCTLx */
#define PMCNT_RESET (0)
/* L3_HML3_PM_EVTYPEx */
#define EVSEL(__val) ((__val) & L3_EVTYPE_MASK)
/* Reset value for all the filter registers */
#define PM_FLTR_RESET (0)
/* L3_M_BC_CR */
#define BC_RESET (1UL << 1)
#define BC_ENABLE (1UL << 0)
/* L3_M_BC_SATROLL_CR */
#define BC_SATROLL_CR_RESET (0)
/* L3_M_BC_CNTENSET */
#define PMCNTENSET(__cntr) (1UL << ((__cntr) & 0x7))
/* L3_M_BC_CNTENCLR */
#define PMCNTENCLR(__cntr) (1UL << ((__cntr) & 0x7))
#define BC_CNTENCLR_RESET (0xFF)
/* L3_M_BC_INTENSET */
#define PMINTENSET(__cntr) (1UL << ((__cntr) & 0x7))
/* L3_M_BC_INTENCLR */
#define PMINTENCLR(__cntr) (1UL << ((__cntr) & 0x7))
#define BC_INTENCLR_RESET (0xFF)
/* L3_M_BC_GANG */
#define GANG_EN(__cntr) (1UL << ((__cntr) & 0x7))
#define BC_GANG_RESET (0)
/* L3_M_BC_OVSR */
#define PMOVSRCLR(__cntr) (1UL << ((__cntr) & 0x7))
#define PMOVSRCLR_RESET (0xFF)
/* L3_M_BC_IRQCTL */
#define PMIRQONMSBEN(__cntr) (1UL << ((__cntr) & 0x7))
#define BC_IRQCTL_RESET (0x0)
/*
* Events
*/
#define L3_EVENT_CYCLES 0x01
#define L3_EVENT_READ_HIT 0x20
#define L3_EVENT_READ_MISS 0x21
#define L3_EVENT_READ_HIT_D 0x22
#define L3_EVENT_READ_MISS_D 0x23
#define L3_EVENT_WRITE_HIT 0x24
#define L3_EVENT_WRITE_MISS 0x25
/*
* Decoding of settings from perf_event_attr
*
* The config format for perf events is:
* - config: bits 0-7: event type
* bit 32: HW counter size requested, 0: 32 bits, 1: 64 bits
*/
static inline u32 get_event_type(struct perf_event *event)
{
return (event->attr.config) & L3_EVTYPE_MASK;
}
static inline bool event_uses_long_counter(struct perf_event *event)
{
return !!(event->attr.config & BIT_ULL(L3_EVENT_LC_BIT));
}
static inline int event_num_counters(struct perf_event *event)
{
return event_uses_long_counter(event) ? 2 : 1;
}
/*
* Main PMU, inherits from the core perf PMU type
*/
struct l3cache_pmu {
struct pmu pmu;
struct hlist_node node;
void __iomem *regs;
struct perf_event *events[L3_NUM_COUNTERS];
unsigned long used_mask[BITS_TO_LONGS(L3_NUM_COUNTERS)];
cpumask_t cpumask;
};
#define to_l3cache_pmu(p) (container_of(p, struct l3cache_pmu, pmu))
/*
* Type used to group hardware counter operations
*
* Used to implement two types of hardware counters, standard (32bits) and
* long (64bits). The hardware supports counter chaining which we use to
* implement long counters. This support is exposed via the 'lc' flag field
* in perf_event_attr.config.
*/
struct l3cache_event_ops {
/* Called to start event monitoring */
void (*start)(struct perf_event *event);
/* Called to stop event monitoring */
void (*stop)(struct perf_event *event, int flags);
/* Called to update the perf_event */
void (*update)(struct perf_event *event);
};
/*
* Implementation of long counter operations
*
* 64bit counters are implemented by chaining two of the 32bit physical
* counters. The PMU only supports chaining of adjacent even/odd pairs
* and for simplicity the driver always configures the odd counter to
* count the overflows of the lower-numbered even counter. Note that since
* the resulting hardware counter is 64bits no IRQs are required to maintain
* the software counter which is also 64bits.
*/
static void qcom_l3_cache__64bit_counter_start(struct perf_event *event)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 evsel = get_event_type(event);
u32 gang;
/* Set the odd counter to count the overflows of the even counter */
gang = readl_relaxed(l3pmu->regs + L3_M_BC_GANG);
gang |= GANG_EN(idx + 1);
writel_relaxed(gang, l3pmu->regs + L3_M_BC_GANG);
/* Initialize the hardware counters and reset prev_count*/
local64_set(&event->hw.prev_count, 0);
writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1));
writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
/*
* Set the event types, the upper half must use zero and the lower
* half the actual event type
*/
writel_relaxed(EVSEL(0), l3pmu->regs + L3_HML3_PM_EVTYPE(idx + 1));
writel_relaxed(EVSEL(evsel), l3pmu->regs + L3_HML3_PM_EVTYPE(idx));
/* Finally, enable the counters */
writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx + 1));
writel_relaxed(PMCNTENSET(idx + 1), l3pmu->regs + L3_M_BC_CNTENSET);
writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx));
writel_relaxed(PMCNTENSET(idx), l3pmu->regs + L3_M_BC_CNTENSET);
}
static void qcom_l3_cache__64bit_counter_stop(struct perf_event *event,
int flags)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 gang = readl_relaxed(l3pmu->regs + L3_M_BC_GANG);
/* Disable the counters */
writel_relaxed(PMCNTENCLR(idx), l3pmu->regs + L3_M_BC_CNTENCLR);
writel_relaxed(PMCNTENCLR(idx + 1), l3pmu->regs + L3_M_BC_CNTENCLR);
/* Disable chaining */
writel_relaxed(gang & ~GANG_EN(idx + 1), l3pmu->regs + L3_M_BC_GANG);
}
static void qcom_l3_cache__64bit_counter_update(struct perf_event *event)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 hi, lo;
u64 prev, new;
do {
prev = local64_read(&event->hw.prev_count);
do {
hi = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1));
lo = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
} while (hi != readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx + 1)));
new = ((u64)hi << 32) | lo;
} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);
local64_add(new - prev, &event->count);
}
static const struct l3cache_event_ops event_ops_long = {
.start = qcom_l3_cache__64bit_counter_start,
.stop = qcom_l3_cache__64bit_counter_stop,
.update = qcom_l3_cache__64bit_counter_update,
};
/*
* Implementation of standard counter operations
*
* 32bit counters use a single physical counter and a hardware feature that
* asserts the overflow IRQ on the toggling of the most significant bit in
* the counter. This feature allows the counters to be left free-running
* without needing the usual reprogramming required to properly handle races
* during concurrent calls to update.
*/
static void qcom_l3_cache__32bit_counter_start(struct perf_event *event)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 evsel = get_event_type(event);
u32 irqctl = readl_relaxed(l3pmu->regs + L3_M_BC_IRQCTL);
/* Set the counter to assert the overflow IRQ on MSB toggling */
writel_relaxed(irqctl | PMIRQONMSBEN(idx), l3pmu->regs + L3_M_BC_IRQCTL);
/* Initialize the hardware counter and reset prev_count*/
local64_set(&event->hw.prev_count, 0);
writel_relaxed(0, l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
/* Set the event type */
writel_relaxed(EVSEL(evsel), l3pmu->regs + L3_HML3_PM_EVTYPE(idx));
/* Enable interrupt generation by this counter */
writel_relaxed(PMINTENSET(idx), l3pmu->regs + L3_M_BC_INTENSET);
/* Finally, enable the counter */
writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(idx));
writel_relaxed(PMCNTENSET(idx), l3pmu->regs + L3_M_BC_CNTENSET);
}
static void qcom_l3_cache__32bit_counter_stop(struct perf_event *event,
int flags)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 irqctl = readl_relaxed(l3pmu->regs + L3_M_BC_IRQCTL);
/* Disable the counter */
writel_relaxed(PMCNTENCLR(idx), l3pmu->regs + L3_M_BC_CNTENCLR);
/* Disable interrupt generation by this counter */
writel_relaxed(PMINTENCLR(idx), l3pmu->regs + L3_M_BC_INTENCLR);
/* Set the counter to not assert the overflow IRQ on MSB toggling */
writel_relaxed(irqctl & ~PMIRQONMSBEN(idx), l3pmu->regs + L3_M_BC_IRQCTL);
}
static void qcom_l3_cache__32bit_counter_update(struct perf_event *event)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
int idx = event->hw.idx;
u32 prev, new;
do {
prev = local64_read(&event->hw.prev_count);
new = readl_relaxed(l3pmu->regs + L3_HML3_PM_EVCNTR(idx));
} while (local64_cmpxchg(&event->hw.prev_count, prev, new) != prev);
local64_add(new - prev, &event->count);
}
static const struct l3cache_event_ops event_ops_std = {
.start = qcom_l3_cache__32bit_counter_start,
.stop = qcom_l3_cache__32bit_counter_stop,
.update = qcom_l3_cache__32bit_counter_update,
};
/* Retrieve the appropriate operations for the given event */
static
const struct l3cache_event_ops *l3cache_event_get_ops(struct perf_event *event)
{
if (event_uses_long_counter(event))
return &event_ops_long;
else
return &event_ops_std;
}
/*
* Top level PMU functions.
*/
static inline void qcom_l3_cache__init(struct l3cache_pmu *l3pmu)
{
int i;
writel_relaxed(BC_RESET, l3pmu->regs + L3_M_BC_CR);
/*
* Use writel for the first programming command to ensure the basic
* counter unit is stopped before proceeding
*/
writel(BC_SATROLL_CR_RESET, l3pmu->regs + L3_M_BC_SATROLL_CR);
writel_relaxed(BC_CNTENCLR_RESET, l3pmu->regs + L3_M_BC_CNTENCLR);
writel_relaxed(BC_INTENCLR_RESET, l3pmu->regs + L3_M_BC_INTENCLR);
writel_relaxed(PMOVSRCLR_RESET, l3pmu->regs + L3_M_BC_OVSR);
writel_relaxed(BC_GANG_RESET, l3pmu->regs + L3_M_BC_GANG);
writel_relaxed(BC_IRQCTL_RESET, l3pmu->regs + L3_M_BC_IRQCTL);
writel_relaxed(PM_CR_RESET, l3pmu->regs + L3_HML3_PM_CR);
for (i = 0; i < L3_NUM_COUNTERS; ++i) {
writel_relaxed(PMCNT_RESET, l3pmu->regs + L3_HML3_PM_CNTCTL(i));
writel_relaxed(EVSEL(0), l3pmu->regs + L3_HML3_PM_EVTYPE(i));
}
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRA);
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRAM);
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRB);
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRBM);
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRC);
writel_relaxed(PM_FLTR_RESET, l3pmu->regs + L3_HML3_PM_FILTRCM);
/*
* Use writel here to ensure all programming commands are done
* before proceeding
*/
writel(BC_ENABLE, l3pmu->regs + L3_M_BC_CR);
}
static irqreturn_t qcom_l3_cache__handle_irq(int irq_num, void *data)
{
struct l3cache_pmu *l3pmu = data;
/* Read the overflow status register */
long status = readl_relaxed(l3pmu->regs + L3_M_BC_OVSR);
int idx;
if (status == 0)
return IRQ_NONE;
/* Clear the bits we read on the overflow status register */
writel_relaxed(status, l3pmu->regs + L3_M_BC_OVSR);
for_each_set_bit(idx, &status, L3_NUM_COUNTERS) {
struct perf_event *event;
const struct l3cache_event_ops *ops;
event = l3pmu->events[idx];
if (!event)
continue;
/*
* Since the IRQ is not enabled for events using long counters
* we should never see one of those here, however, be consistent
* and use the ops indirections like in the other operations.
*/
ops = l3cache_event_get_ops(event);
ops->update(event);
}
return IRQ_HANDLED;
}
/*
* Implementation of abstract pmu functionality required by
* the core perf events code.
*/
static void qcom_l3_cache__pmu_enable(struct pmu *pmu)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(pmu);
/* Ensure the other programming commands are observed before enabling */
wmb();
writel_relaxed(BC_ENABLE, l3pmu->regs + L3_M_BC_CR);
}
static void qcom_l3_cache__pmu_disable(struct pmu *pmu)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(pmu);
writel_relaxed(0, l3pmu->regs + L3_M_BC_CR);
/* Ensure the basic counter unit is stopped before proceeding */
wmb();
}
/*
* We must NOT create groups containing events from multiple hardware PMUs,
* although mixing different software and hardware PMUs is allowed.
*/
static bool qcom_l3_cache__validate_event_group(struct perf_event *event)
{
struct perf_event *leader = event->group_leader;
struct perf_event *sibling;
int counters = 0;
if (leader->pmu != event->pmu && !is_software_event(leader))
return false;
counters = event_num_counters(event);
counters += event_num_counters(leader);
list_for_each_entry(sibling, &leader->sibling_list, group_entry) {
if (is_software_event(sibling))
continue;
if (sibling->pmu != event->pmu)
return false;
counters += event_num_counters(sibling);
}
/*
* If the group requires more counters than the HW has, it
* cannot ever be scheduled.
*/
return counters <= L3_NUM_COUNTERS;
}
static int qcom_l3_cache__event_init(struct perf_event *event)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
struct hw_perf_event *hwc = &event->hw;
/*
* Is the event for this PMU?
*/
if (event->attr.type != event->pmu->type)
return -ENOENT;
/*
* There are no per-counter mode filters in the PMU.
*/
if (event->attr.exclude_user || event->attr.exclude_kernel ||
event->attr.exclude_hv || event->attr.exclude_idle)
return -EINVAL;
/*
* Sampling not supported since these events are not core-attributable.
*/
if (hwc->sample_period)
return -EINVAL;
/*
* Task mode not available, we run the counters as socket counters,
* not attributable to any CPU and therefore cannot attribute per-task.
*/
if (event->cpu < 0)
return -EINVAL;
/* Validate the group */
if (!qcom_l3_cache__validate_event_group(event))
return -EINVAL;
hwc->idx = -1;
/*
* Many perf core operations (eg. events rotation) operate on a
* single CPU context. This is obvious for CPU PMUs, where one
* expects the same sets of events being observed on all CPUs,
* but can lead to issues for off-core PMUs, like this one, where
* each event could be theoretically assigned to a different CPU.
* To mitigate this, we enforce CPU assignment to one designated
* processor (the one described in the "cpumask" attribute exported
* by the PMU device). perf user space tools honor this and avoid
* opening more than one copy of the events.
*/
event->cpu = cpumask_first(&l3pmu->cpumask);
return 0;
}
static void qcom_l3_cache__event_start(struct perf_event *event, int flags)
{
struct hw_perf_event *hwc = &event->hw;
const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
hwc->state = 0;
ops->start(event);
}
static void qcom_l3_cache__event_stop(struct perf_event *event, int flags)
{
struct hw_perf_event *hwc = &event->hw;
const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
if (hwc->state & PERF_HES_STOPPED)
return;
ops->stop(event, flags);
if (flags & PERF_EF_UPDATE)
ops->update(event);
hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
}
static int qcom_l3_cache__event_add(struct perf_event *event, int flags)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
struct hw_perf_event *hwc = &event->hw;
int order = event_uses_long_counter(event) ? 1 : 0;
int idx;
/*
* Try to allocate a counter.
*/
idx = bitmap_find_free_region(l3pmu->used_mask, L3_NUM_COUNTERS, order);
if (idx < 0)
/* The counters are all in use. */
return -EAGAIN;
hwc->idx = idx;
hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
l3pmu->events[idx] = event;
if (flags & PERF_EF_START)
qcom_l3_cache__event_start(event, 0);
/* Propagate changes to the userspace mapping. */
perf_event_update_userpage(event);
return 0;
}
static void qcom_l3_cache__event_del(struct perf_event *event, int flags)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(event->pmu);
struct hw_perf_event *hwc = &event->hw;
int order = event_uses_long_counter(event) ? 1 : 0;
/* Stop and clean up */
qcom_l3_cache__event_stop(event, flags | PERF_EF_UPDATE);
l3pmu->events[hwc->idx] = NULL;
bitmap_release_region(l3pmu->used_mask, hwc->idx, order);
/* Propagate changes to the userspace mapping. */
perf_event_update_userpage(event);
}
static void qcom_l3_cache__event_read(struct perf_event *event)
{
const struct l3cache_event_ops *ops = l3cache_event_get_ops(event);
ops->update(event);
}
/*
* Add sysfs attributes
*
* We export:
* - formats, used by perf user space and other tools to configure events
* - events, used by perf user space and other tools to create events
* symbolically, e.g.:
* perf stat -a -e l3cache_0_0/event=read-miss/ ls
* perf stat -a -e l3cache_0_0/event=0x21/ ls
* - cpumask, used by perf user space and other tools to know on which CPUs
* to open the events
*/
/* formats */
static ssize_t l3cache_pmu_format_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct dev_ext_attribute *eattr;
eattr = container_of(attr, struct dev_ext_attribute, attr);
return sprintf(buf, "%s\n", (char *) eattr->var);
}
#define L3CACHE_PMU_FORMAT_ATTR(_name, _config) \
(&((struct dev_ext_attribute[]) { \
{ .attr = __ATTR(_name, 0444, l3cache_pmu_format_show, NULL), \
.var = (void *) _config, } \
})[0].attr.attr)
static struct attribute *qcom_l3_cache_pmu_formats[] = {
L3CACHE_PMU_FORMAT_ATTR(event, "config:0-7"),
L3CACHE_PMU_FORMAT_ATTR(lc, "config:" __stringify(L3_EVENT_LC_BIT)),
NULL,
};
static struct attribute_group qcom_l3_cache_pmu_format_group = {
.name = "format",
.attrs = qcom_l3_cache_pmu_formats,
};
/* events */
static ssize_t l3cache_pmu_event_show(struct device *dev,
struct device_attribute *attr, char *page)
{
struct perf_pmu_events_attr *pmu_attr;
pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
return sprintf(page, "event=0x%02llx\n", pmu_attr->id);
}
#define L3CACHE_EVENT_ATTR(_name, _id) \
(&((struct perf_pmu_events_attr[]) { \
{ .attr = __ATTR(_name, 0444, l3cache_pmu_event_show, NULL), \
.id = _id, } \
})[0].attr.attr)
static struct attribute *qcom_l3_cache_pmu_events[] = {
L3CACHE_EVENT_ATTR(cycles, L3_EVENT_CYCLES),
L3CACHE_EVENT_ATTR(read-hit, L3_EVENT_READ_HIT),
L3CACHE_EVENT_ATTR(read-miss, L3_EVENT_READ_MISS),
L3CACHE_EVENT_ATTR(read-hit-d-side, L3_EVENT_READ_HIT_D),
L3CACHE_EVENT_ATTR(read-miss-d-side, L3_EVENT_READ_MISS_D),
L3CACHE_EVENT_ATTR(write-hit, L3_EVENT_WRITE_HIT),
L3CACHE_EVENT_ATTR(write-miss, L3_EVENT_WRITE_MISS),
NULL
};
static struct attribute_group qcom_l3_cache_pmu_events_group = {
.name = "events",
.attrs = qcom_l3_cache_pmu_events,
};
/* cpumask */
static ssize_t qcom_l3_cache_pmu_cpumask_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct l3cache_pmu *l3pmu = to_l3cache_pmu(dev_get_drvdata(dev));
return cpumap_print_to_pagebuf(true, buf, &l3pmu->cpumask);
}
static DEVICE_ATTR(cpumask, 0444, qcom_l3_cache_pmu_cpumask_show, NULL);
static struct attribute *qcom_l3_cache_pmu_cpumask_attrs[] = {
&dev_attr_cpumask.attr,
NULL,
};
static struct attribute_group qcom_l3_cache_pmu_cpumask_attr_group = {
.attrs = qcom_l3_cache_pmu_cpumask_attrs,
};
/*
* Per PMU device attribute groups
*/
static const struct attribute_group *qcom_l3_cache_pmu_attr_grps[] = {
&qcom_l3_cache_pmu_format_group,
&qcom_l3_cache_pmu_events_group,
&qcom_l3_cache_pmu_cpumask_attr_group,
NULL,
};
/*
* Probing functions and data.
*/
static int qcom_l3_cache_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
{
struct l3cache_pmu *l3pmu = hlist_entry_safe(node, struct l3cache_pmu, node);
/* If there is not a CPU/PMU association pick this CPU */
if (cpumask_empty(&l3pmu->cpumask))
cpumask_set_cpu(cpu, &l3pmu->cpumask);
return 0;
}
static int qcom_l3_cache_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
{
struct l3cache_pmu *l3pmu = hlist_entry_safe(node, struct l3cache_pmu, node);
unsigned int target;
if (!cpumask_test_and_clear_cpu(cpu, &l3pmu->cpumask))
return 0;
target = cpumask_any_but(cpu_online_mask, cpu);
if (target >= nr_cpu_ids)
return 0;
perf_pmu_migrate_context(&l3pmu->pmu, cpu, target);
cpumask_set_cpu(target, &l3pmu->cpumask);
return 0;
}
static int qcom_l3_cache_pmu_probe(struct platform_device *pdev)
{
struct l3cache_pmu *l3pmu;
struct acpi_device *acpi_dev;
struct resource *memrc;
int ret;
char *name;
/* Initialize the PMU data structures */
acpi_dev = ACPI_COMPANION(&pdev->dev);
if (!acpi_dev)
return -ENODEV;
l3pmu = devm_kzalloc(&pdev->dev, sizeof(*l3pmu), GFP_KERNEL);
name = devm_kasprintf(&pdev->dev, GFP_KERNEL, "l3cache_%s_%s",
acpi_dev->parent->pnp.unique_id, acpi_dev->pnp.unique_id);
if (!l3pmu || !name)
return -ENOMEM;
l3pmu->pmu = (struct pmu) {
.task_ctx_nr = perf_invalid_context,
.pmu_enable = qcom_l3_cache__pmu_enable,
.pmu_disable = qcom_l3_cache__pmu_disable,
.event_init = qcom_l3_cache__event_init,
.add = qcom_l3_cache__event_add,
.del = qcom_l3_cache__event_del,
.start = qcom_l3_cache__event_start,
.stop = qcom_l3_cache__event_stop,
.read = qcom_l3_cache__event_read,
.attr_groups = qcom_l3_cache_pmu_attr_grps,
};
memrc = platform_get_resource(pdev, IORESOURCE_MEM, 0);
l3pmu->regs = devm_ioremap_resource(&pdev->dev, memrc);
if (IS_ERR(l3pmu->regs)) {
dev_err(&pdev->dev, "Can't map PMU @%pa\n", &memrc->start);
return PTR_ERR(l3pmu->regs);
}
qcom_l3_cache__init(l3pmu);
ret = platform_get_irq(pdev, 0);
if (ret <= 0)
return ret;
ret = devm_request_irq(&pdev->dev, ret, qcom_l3_cache__handle_irq, 0,
name, l3pmu);
if (ret) {
dev_err(&pdev->dev, "Request for IRQ failed for slice @%pa\n",
&memrc->start);
return ret;
}
/* Add this instance to the list used by the offline callback */
ret = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE, &l3pmu->node);
if (ret) {
dev_err(&pdev->dev, "Error %d registering hotplug", ret);
return ret;
}
ret = perf_pmu_register(&l3pmu->pmu, name, -1);
if (ret < 0) {
dev_err(&pdev->dev, "Failed to register L3 cache PMU (%d)\n", ret);
return ret;
}
dev_info(&pdev->dev, "Registered %s, type: %d\n", name, l3pmu->pmu.type);
return 0;
}
static const struct acpi_device_id qcom_l3_cache_pmu_acpi_match[] = {
{ "QCOM8081", },
{ }
};
MODULE_DEVICE_TABLE(acpi, qcom_l3_cache_pmu_acpi_match);
static struct platform_driver qcom_l3_cache_pmu_driver = {
.driver = {
.name = "qcom-l3cache-pmu",
.acpi_match_table = ACPI_PTR(qcom_l3_cache_pmu_acpi_match),
},
.probe = qcom_l3_cache_pmu_probe,
};
static int __init register_qcom_l3_cache_pmu_driver(void)
{
int ret;
/* Install a hook to update the reader CPU in case it goes offline */
ret = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE,
"perf/qcom/l3cache:online",
qcom_l3_cache_pmu_online_cpu,
qcom_l3_cache_pmu_offline_cpu);
if (ret)
return ret;
return platform_driver_register(&qcom_l3_cache_pmu_driver);
}
device_initcall(register_qcom_l3_cache_pmu_driver);
...@@ -94,6 +94,7 @@ enum cpuhp_state { ...@@ -94,6 +94,7 @@ enum cpuhp_state {
CPUHP_AP_ARM_VFP_STARTING, CPUHP_AP_ARM_VFP_STARTING,
CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING, CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING, CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
CPUHP_AP_PERF_ARM_ACPI_STARTING,
CPUHP_AP_PERF_ARM_STARTING, CPUHP_AP_PERF_ARM_STARTING,
CPUHP_AP_ARM_L2X0_STARTING, CPUHP_AP_ARM_L2X0_STARTING,
CPUHP_AP_ARM_ARCH_TIMER_STARTING, CPUHP_AP_ARM_ARCH_TIMER_STARTING,
...@@ -137,6 +138,7 @@ enum cpuhp_state { ...@@ -137,6 +138,7 @@ enum cpuhp_state {
CPUHP_AP_PERF_ARM_CCN_ONLINE, CPUHP_AP_PERF_ARM_CCN_ONLINE,
CPUHP_AP_PERF_ARM_L2X0_ONLINE, CPUHP_AP_PERF_ARM_L2X0_ONLINE,
CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE, CPUHP_AP_PERF_ARM_QCOM_L2_ONLINE,
CPUHP_AP_PERF_ARM_QCOM_L3_ONLINE,
CPUHP_AP_WORKQUEUE_ONLINE, CPUHP_AP_WORKQUEUE_ONLINE,
CPUHP_AP_RCUTREE_ONLINE, CPUHP_AP_RCUTREE_ONLINE,
CPUHP_AP_ONLINE_DYN, CPUHP_AP_ONLINE_DYN,
......
...@@ -75,6 +75,8 @@ struct pmu_hw_events { ...@@ -75,6 +75,8 @@ struct pmu_hw_events {
* already have to allocate this struct per cpu. * already have to allocate this struct per cpu.
*/ */
struct arm_pmu *percpu_pmu; struct arm_pmu *percpu_pmu;
int irq;
}; };
enum armpmu_attr_groups { enum armpmu_attr_groups {
...@@ -88,7 +90,6 @@ struct arm_pmu { ...@@ -88,7 +90,6 @@ struct arm_pmu {
struct pmu pmu; struct pmu pmu;
cpumask_t active_irqs; cpumask_t active_irqs;
cpumask_t supported_cpus; cpumask_t supported_cpus;
int *irq_affinity;
char *name; char *name;
irqreturn_t (*handle_irq)(int irq_num, void *dev); irqreturn_t (*handle_irq)(int irq_num, void *dev);
void (*enable)(struct perf_event *event); void (*enable)(struct perf_event *event);
...@@ -104,12 +105,8 @@ struct arm_pmu { ...@@ -104,12 +105,8 @@ struct arm_pmu {
void (*start)(struct arm_pmu *); void (*start)(struct arm_pmu *);
void (*stop)(struct arm_pmu *); void (*stop)(struct arm_pmu *);
void (*reset)(void *); void (*reset)(void *);
int (*request_irq)(struct arm_pmu *, irq_handler_t handler);
void (*free_irq)(struct arm_pmu *);
int (*map_event)(struct perf_event *event); int (*map_event)(struct perf_event *event);
int num_events; int num_events;
atomic_t active_events;
struct mutex reserve_mutex;
u64 max_period; u64 max_period;
bool secure_access; /* 32-bit ARM only */ bool secure_access; /* 32-bit ARM only */
#define ARMV8_PMUV3_MAX_COMMON_EVENTS 0x40 #define ARMV8_PMUV3_MAX_COMMON_EVENTS 0x40
...@@ -120,6 +117,9 @@ struct arm_pmu { ...@@ -120,6 +117,9 @@ struct arm_pmu {
struct notifier_block cpu_pm_nb; struct notifier_block cpu_pm_nb;
/* the attr_groups array must be NULL-terminated */ /* the attr_groups array must be NULL-terminated */
const struct attribute_group *attr_groups[ARMPMU_NR_ATTR_GROUPS + 1]; const struct attribute_group *attr_groups[ARMPMU_NR_ATTR_GROUPS + 1];
/* Only to be used by ACPI probing code */
unsigned long acpi_cpuid;
}; };
#define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu)) #define to_arm_pmu(p) (container_of(p, struct arm_pmu, pmu))
...@@ -135,10 +135,12 @@ int armpmu_map_event(struct perf_event *event, ...@@ -135,10 +135,12 @@ int armpmu_map_event(struct perf_event *event,
[PERF_COUNT_HW_CACHE_RESULT_MAX], [PERF_COUNT_HW_CACHE_RESULT_MAX],
u32 raw_event_mask); u32 raw_event_mask);
typedef int (*armpmu_init_fn)(struct arm_pmu *);
struct pmu_probe_info { struct pmu_probe_info {
unsigned int cpuid; unsigned int cpuid;
unsigned int mask; unsigned int mask;
int (*init)(struct arm_pmu *); armpmu_init_fn init;
}; };
#define PMU_PROBE(_cpuid, _mask, _fn) \ #define PMU_PROBE(_cpuid, _mask, _fn) \
...@@ -160,6 +162,21 @@ int arm_pmu_device_probe(struct platform_device *pdev, ...@@ -160,6 +162,21 @@ int arm_pmu_device_probe(struct platform_device *pdev,
const struct of_device_id *of_table, const struct of_device_id *of_table,
const struct pmu_probe_info *probe_table); const struct pmu_probe_info *probe_table);
#ifdef CONFIG_ACPI
int arm_pmu_acpi_probe(armpmu_init_fn init_fn);
#else
static inline int arm_pmu_acpi_probe(armpmu_init_fn init_fn) { return 0; }
#endif
/* Internal functions only for core arm_pmu code */
struct arm_pmu *armpmu_alloc(void);
void armpmu_free(struct arm_pmu *pmu);
int armpmu_register(struct arm_pmu *pmu);
int armpmu_request_irqs(struct arm_pmu *armpmu);
void armpmu_free_irqs(struct arm_pmu *armpmu);
int armpmu_request_irq(struct arm_pmu *armpmu, int cpu);
void armpmu_free_irq(struct arm_pmu *armpmu, int cpu);
#define ARMV8_PMU_PDEV_NAME "armv8-pmu" #define ARMV8_PMU_PDEV_NAME "armv8-pmu"
#endif /* CONFIG_ARM_PMU */ #endif /* CONFIG_ARM_PMU */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment