Merge "perf: Set the DSU PMU to be readable from any CPU" (0259737f) · Commits · e / devices / android_kernel_fairphone_FP4

arch/arm64/kernel/Makefile

+2 −1

Original line number	Diff line number	Diff line
		@@ -33,7 +33,8 @@ arm64-obj-$(CONFIG_FUNCTION_TRACER) += ftrace.o entry-ftrace.o
		arm64-obj-$(CONFIG_MODULES) += arm64ksyms.o module.o
		arm64-obj-$(CONFIG_ARM64_MODULE_PLTS) += module-plts.o
		arm64-obj-$(CONFIG_PERF_EVENTS) += perf_regs.o perf_callchain.o
		arm64-obj-$(CONFIG_HW_PERF_EVENTS) += perf_event.o
		arm64-obj-$(CONFIG_HW_PERF_EVENTS) += perf_event.o perf_trace_counters.o \
		perf_trace_user.o
		arm64-obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
		arm64-obj-$(CONFIG_CPU_PM) += sleep.o suspend.o
		arm64-obj-$(CONFIG_CPU_IDLE) += cpuidle.o

arch/arm64/kernel/perf_event.c

+126 −7

Original line number	Diff line number	Diff line
		@@ -30,6 +30,8 @@
		#include <linux/perf/arm_pmu.h>
		#include <linux/platform_device.h>

		static DEFINE_PER_CPU(bool, is_hotplugging);

		/*
		* ARMv8 PMUv3 Performance Events handling code.
		* Common event types (some are defined in asm/perf_event.h).
		@@ -870,8 +872,8 @@ static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
		struct perf_event *event = cpuc->events[idx];
		struct hw_perf_event *hwc;

		/* Ignore if we don't have an event. */
		if (!event)
		/* Ignore if we don't have an event */
		if (!event \|\| event->state != PERF_EVENT_STATE_ACTIVE)
		continue;

		/*
		@@ -990,9 +992,6 @@ static int armv8pmu_set_event_filter(struct hw_perf_event *event,
		{
		unsigned long config_base = 0;

		if (attr->exclude_idle)
		return -EPERM;

		/*
		* If we're running in hyp mode, then we are the hypervisor.
		* Therefore we ignore exclude_hv in this configuration, since
		@@ -1140,6 +1139,55 @@ static void __armv8pmu_probe_pmu(void *info)
		pmceid, ARMV8_PMUV3_MAX_COMMON_EVENTS);
		}

		static void armv8pmu_idle_update(struct arm_pmu *cpu_pmu)
		{
		struct pmu_hw_events *hw_events;
		struct perf_event *event;
		int idx;

		if (!cpu_pmu)
		return;

		if (__this_cpu_read(is_hotplugging))
		return;

		hw_events = this_cpu_ptr(cpu_pmu->hw_events);

		if (!hw_events)
		return;

		for (idx = 0; idx < cpu_pmu->num_events; ++idx) {

		if (!test_bit(idx, hw_events->used_mask))
		continue;

		event = hw_events->events[idx];

		if (!event \|\| !event->attr.exclude_idle \|\|
		event->state != PERF_EVENT_STATE_ACTIVE)
		continue;

		cpu_pmu->pmu.read(event);
		}
		}

		struct arm_pmu_and_idle_nb {
		struct arm_pmu *cpu_pmu;
		struct notifier_block perf_cpu_idle_nb;
		};

		static int perf_cpu_idle_notifier(struct notifier_block *nb,
		unsigned long action, void *data)
		{
		struct arm_pmu_and_idle_nb *pmu_nb = container_of(nb,
		struct arm_pmu_and_idle_nb, perf_cpu_idle_nb);

		if (action == IDLE_START)
		armv8pmu_idle_update(pmu_nb->cpu_pmu);

		return NOTIFY_OK;
		}

		static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
		{
		struct armv8pmu_probe_info probe = {
		@@ -1147,6 +1195,15 @@ static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
		.present = false,
		};
		int ret;
		struct arm_pmu_and_idle_nb *pmu_idle_nb;

		pmu_idle_nb = devm_kzalloc(&cpu_pmu->plat_device->dev,
		sizeof(*pmu_idle_nb), GFP_KERNEL);
		if (!pmu_idle_nb)
		return -ENOMEM;

		pmu_idle_nb->cpu_pmu = cpu_pmu;
		pmu_idle_nb->perf_cpu_idle_nb.notifier_call = perf_cpu_idle_notifier;

		ret = smp_call_function_any(&cpu_pmu->supported_cpus,
		__armv8pmu_probe_pmu,
		@@ -1154,7 +1211,13 @@ static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
		if (ret)
		return ret;

		return probe.present ? 0 : -ENODEV;
		if (!probe.present)
		return -ENODEV;

		idle_notifier_register(&pmu_idle_nb->perf_cpu_idle_nb);

		return 0;

		}

		static int armv8_pmu_init(struct arm_pmu *cpu_pmu)
		@@ -1319,9 +1382,65 @@ static const struct of_device_id armv8_pmu_of_device_ids[] = {
		{},
		};

		#ifdef CONFIG_HOTPLUG_CPU
		static int perf_event_hotplug_coming_up(unsigned int cpu)
		{
		per_cpu(is_hotplugging, cpu) = false;
		return 0;
		}

		static int perf_event_hotplug_going_down(unsigned int cpu)
		{
		per_cpu(is_hotplugging, cpu) = true;
		return 0;
		}

		static int perf_event_cpu_hp_init(void)
		{
		int ret;

		ret = cpuhp_setup_state_nocalls(CPUHP_AP_NOTIFY_PERF_ONLINE,
		"PERF_EVENT/CPUHP_AP_NOTIFY_PERF_ONLINE",
		perf_event_hotplug_coming_up,
		perf_event_hotplug_going_down);
		if (ret)
		pr_err("CPU hotplug notifier for perf_event.c could not be registered: %d\n",
		ret);

		return ret;
		}
		#else
		static int perf_event_cpu_hp_init(void) { return 0; }
		#endif

		/*
		* 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)
		{
		return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids, NULL);
		int ret, cpu;

		for_each_possible_cpu(cpu)
		per_cpu(is_hotplugging, cpu) = false;

		ret = perf_event_cpu_hp_init();
		if (ret)
		return ret;

		/* set to true so armv8pmu_idle_update doesn't try to load
		* hw_events before arm_pmu_device_probe has initialized it.
		*/
		ret = arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids,
		(acpi_disabled ? NULL : armv8_pmu_probe_table));

		return ret;
		}

		static struct platform_driver armv8_pmu_driver = {

arch/arm64/kernel/perf_trace_counters.c

0 → 100644

+171 −0

Original line number	Diff line number	Diff line
		// SPDX-License-Identifier: GPL-2.0
		/*
		* Copyright (c) 2013-2014, 2017-2018, The Linux Foundation. All rights reserved.
		*/
		#include <linux/uaccess.h>
		#include <linux/debugfs.h>
		#include <linux/cpu.h>
		#include <linux/tracepoint.h>
		#include <trace/events/sched.h>
		#define CREATE_TRACE_POINTS
		#include "perf_trace_counters.h"

		static unsigned int tp_pid_state;

		DEFINE_PER_CPU(u32, cntenset_val);
		DEFINE_PER_CPU(u32, previous_ccnt);
		DEFINE_PER_CPU(u32[NUM_L1_CTRS], previous_l1_cnts);
		DEFINE_PER_CPU(u32, old_pid);
		DEFINE_PER_CPU(u32, hotplug_flag);

		#define USE_CPUHP_STATE CPUHP_AP_ONLINE

		static int tracectr_cpu_hotplug_coming_up(unsigned int cpu)
		{
		per_cpu(hotplug_flag, cpu) = 1;

		return 0;
		}

		static void setup_prev_cnts(u32 cpu, u32 cnten_val)
		{
		int i;

		if (cnten_val & CC)
		per_cpu(previous_ccnt, cpu) =
		read_sysreg(pmccntr_el0);

		for (i = 0; i < NUM_L1_CTRS; i++) {
		if (cnten_val & (1 << i)) {
		/* Select */
		write_sysreg(i, pmselr_el0);
		isb();
		/* Read value */
		per_cpu(previous_l1_cnts[i], cpu) =
		read_sysreg(pmxevcntr_el0);
		}
		}
		}

		void tracectr_notifier(void *ignore, bool preempt,
		struct task_struct prev, struct task_struct next)
		{
		u32 cnten_val;
		int current_pid;
		u32 cpu = task_cpu(next);

		if (tp_pid_state != 1)
		return;
		current_pid = next->pid;
		if (per_cpu(old_pid, cpu) != -1) {
		cnten_val = read_sysreg(pmcntenset_el0);
		per_cpu(cntenset_val, cpu) = cnten_val;
		/* Disable all the counters that were enabled */
		write_sysreg(cnten_val, pmcntenclr_el0);

		if (per_cpu(hotplug_flag, cpu) == 1) {
		per_cpu(hotplug_flag, cpu) = 0;
		setup_prev_cnts(cpu, cnten_val);
		} else {
		trace_sched_switch_with_ctrs(per_cpu(old_pid, cpu),
		current_pid);
		}

		/* Enable all the counters that were disabled */
		write_sysreg(cnten_val, pmcntenset_el0);
		}
		per_cpu(old_pid, cpu) = current_pid;
		}

		static void enable_tp_pid(void)
		{
		if (tp_pid_state == 0) {
		tp_pid_state = 1;
		register_trace_sched_switch(tracectr_notifier, NULL);
		}
		}

		static void disable_tp_pid(void)
		{
		if (tp_pid_state == 1) {
		tp_pid_state = 0;
		unregister_trace_sched_switch(tracectr_notifier, NULL);
		}
		}

		static ssize_t read_enabled_perftp_file_bool(struct file *file,
		char __user user_buf, size_t count, loff_t ppos)
		{
		char buf[2];

		buf[1] = '\n';
		if (tp_pid_state == 0)
		buf[0] = '0';
		else
		buf[0] = '1';
		return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
		}

		static ssize_t write_enabled_perftp_file_bool(struct file *file,
		const char __user user_buf, size_t count, loff_t ppos)
		{
		char buf[32];
		size_t buf_size;

		buf[0] = 0;
		buf_size = min(count, (sizeof(buf)-1));
		if (copy_from_user(buf, user_buf, buf_size))
		return -EFAULT;
		switch (buf[0]) {
		case 'y':
		case 'Y':
		case '1':
		enable_tp_pid();
		break;
		case 'n':
		case 'N':
		case '0':
		disable_tp_pid();
		break;
		}

		return count;
		}

		static const struct file_operations fops_perftp = {
		.read = read_enabled_perftp_file_bool,
		.write = write_enabled_perftp_file_bool,
		.llseek = default_llseek,
		};

		int __init init_tracecounters(void)
		{
		struct dentry *dir;
		struct dentry *file;
		unsigned int value = 1;
		int cpu, rc;

		dir = debugfs_create_dir("perf_debug_tp", NULL);
		if (!dir)
		return -ENOMEM;
		file = debugfs_create_file("enabled", 0660, dir,
		&value, &fops_perftp);
		if (!file) {
		debugfs_remove(dir);
		return -ENOMEM;
		}
		for_each_possible_cpu(cpu)
		per_cpu(old_pid, cpu) = -1;
		rc = cpuhp_setup_state_nocalls(USE_CPUHP_STATE,
		"tracectr_cpu_hotplug",
		tracectr_cpu_hotplug_coming_up,
		NULL);
		return 0;
		}

		int __exit exit_tracecounters(void)
		{
		cpuhp_remove_state_nocalls(USE_CPUHP_STATE);
		return 0;
		}
		late_initcall(init_tracecounters);

arch/arm64/kernel/perf_trace_counters.h

0 → 100644

+103 −0

Original line number	Diff line number	Diff line
		/* SPDX-License-Identifier: GPL-2.0 */
		/*
		* Copyright (c) 2013-2014, 2017-2018, The Linux Foundation. All rights reserved.
		*/

		#undef TRACE_SYSTEM
		#define TRACE_SYSTEM perf_trace_counters

		#if !defined(_PERF_TRACE_COUNTERS_H_) \|\| defined(TRACE_HEADER_MULTI_READ)
		#define _PERF_TRACE_COUNTERS_H_

		/* Ctr index for PMCNTENSET/CLR */
		#define CC 0x80000000
		#define C0 0x1
		#define C1 0x2
		#define C2 0x4
		#define C3 0x8
		#define C4 0x10
		#define C5 0x20
		#define C_ALL (CC \| C0 \| C1 \| C2 \| C3 \| C4 \| C5)
		#define NUM_L1_CTRS 6

		#include <linux/sched.h>
		#include <linux/cpumask.h>
		#include <linux/tracepoint.h>

		DECLARE_PER_CPU(u32, cntenset_val);
		DECLARE_PER_CPU(u32, previous_ccnt);
		DECLARE_PER_CPU(u32[NUM_L1_CTRS], previous_l1_cnts);
		TRACE_EVENT(sched_switch_with_ctrs,

		TP_PROTO(pid_t prev, pid_t next),

		TP_ARGS(prev, next),

		TP_STRUCT__entry(
		__field(pid_t, old_pid)
		__field(pid_t, new_pid)
		__field(u32, cctr)
		__field(u32, ctr0)
		__field(u32, ctr1)
		__field(u32, ctr2)
		__field(u32, ctr3)
		__field(u32, ctr4)
		__field(u32, ctr5)
		),

		TP_fast_assign(
		u32 cpu = smp_processor_id();
		u32 i;
		u32 cnten_val;
		u32 total_ccnt = 0;
		u32 total_cnt = 0;
		u32 delta_l1_cnts[NUM_L1_CTRS];

		__entry->old_pid = prev;
		__entry->new_pid = next;

		cnten_val = per_cpu(cntenset_val, cpu);

		if (cnten_val & CC) {
		/* Read value */
		total_ccnt = read_sysreg(pmccntr_el0);
		__entry->cctr = total_ccnt -
		per_cpu(previous_ccnt, cpu);
		per_cpu(previous_ccnt, cpu) = total_ccnt;
		}
		for (i = 0; i < NUM_L1_CTRS; i++) {
		if (cnten_val & (1 << i)) {
		/* Select */
		write_sysreg(i, pmselr_el0);
		isb();
		/* Read value */
		total_cnt = read_sysreg(pmxevcntr_el0);
		delta_l1_cnts[i] = total_cnt -
		per_cpu(previous_l1_cnts[i], cpu);
		per_cpu(previous_l1_cnts[i], cpu) =
		total_cnt;
		} else
		delta_l1_cnts[i] = 0;
		}

		__entry->ctr0 = delta_l1_cnts[0];
		__entry->ctr1 = delta_l1_cnts[1];
		__entry->ctr2 = delta_l1_cnts[2];
		__entry->ctr3 = delta_l1_cnts[3];
		__entry->ctr4 = delta_l1_cnts[4];
		__entry->ctr5 = delta_l1_cnts[5];
		),

		TP_printk("prev_pid=%d, next_pid=%d, CCNTR: %u, CTR0: %u, CTR1: %u, CTR2: %u, CTR3: %u, CTR4: %u, CTR5: %u",
		__entry->old_pid, __entry->new_pid,
		__entry->cctr,
		__entry->ctr0, __entry->ctr1,
		__entry->ctr2, __entry->ctr3,
		__entry->ctr4, __entry->ctr5)
		);

		#endif
		#undef TRACE_INCLUDE_PATH
		#define TRACE_INCLUDE_PATH ../../arch/arm64/kernel
		#define TRACE_INCLUDE_FILE perf_trace_counters
		#include <trace/define_trace.h>

arch/arm64/kernel/perf_trace_user.c

0 → 100644

+89 −0

Original line number	Diff line number	Diff line
		// SPDX-License-Identifier: GPL-2.0
		/*
		* Copyright (c) 2014,2017-2018, The Linux Foundation. All rights reserved.
		*/
		#include <linux/perf_event.h>
		#include <linux/types.h>
		#include <linux/tracepoint.h>
		#include <linux/fs.h>
		#include <linux/debugfs.h>
		#include <linux/preempt.h>
		#include <linux/stat.h>
		#include <asm/uaccess.h>

		#define CREATE_TRACE_POINTS
		#include "perf_trace_user.h"

		#undef TRACE_SYSTEM
		#define TRACE_SYSTEM perf_trace_counters

		#define TRACE_USER_MAX_BUF_SIZE 100

		static ssize_t perf_trace_write(struct file *file,
		const char __user *user_string_in,
		size_t len, loff_t *ppos)
		{
		u32 cnten_val;
		int rc;
		char buf[TRACE_USER_MAX_BUF_SIZE + 1];
		ssize_t length;

		if (len == 0)
		return 0;

		length = len > TRACE_USER_MAX_BUF_SIZE ? TRACE_USER_MAX_BUF_SIZE : len;

		rc = copy_from_user(buf, user_string_in, length);
		if (rc) {
		pr_err("%s copy_from_user failed, rc=%d\n", __func__, rc);
		return -EFAULT;
		}

		/* Remove any trailing newline and make sure string is terminated */
		if (buf[length - 1] == '\n')
		buf[length - 1] = '\0';
		else
		buf[length] = '\0';

		/*
		* Disable preemption to ensure that all the performance counter
		* accesses happen on the same cpu
		*/
		preempt_disable();
		/* stop counters, call the trace function, restart them */

		cnten_val = read_sysreg(pmcntenset_el0);
		/* Disable all the counters that were enabled */
		write_sysreg(cnten_val, pmcntenclr_el0);

		trace_perf_trace_user(buf, cnten_val);

		/* Enable all the counters that were disabled */
		write_sysreg(cnten_val, pmcntenset_el0);
		preempt_enable();

		return length;
		}

		static const struct file_operations perf_trace_fops = {
		.write = perf_trace_write
		};

		static int __init init_perf_trace(void)
		{
		struct dentry *dir;
		struct dentry *file;
		unsigned int value = 1;

		dir = debugfs_create_dir("msm_perf", NULL);
		if (!dir)
		return -ENOMEM;
		file = debugfs_create_file("trace_marker", 0220, dir,
		&value, &perf_trace_fops);
		if (!file)
		return -ENOMEM;

		return 0;
		}

		late_initcall(init_perf_trace);