Perf: arm64: Add Snapshot of perf tracepoints

Snapshot of perf tracepoint framework taken straight of msm-4.4
commit <fec225ae458291> (Merge "qpnp-fg-gen3: prime CC_SOC_SW when
capacity learning begins")
[Fixing coding style issues]
[Switch perf_trace_counters.c to new hotplug notifier method]
[Replace inline assembly with read/write_sysreg()]
[Change perf_trace_user.h to display 7 counters like
perf_trace_counters.h]
[Change perf_trace_user.c to use debugfs_create_dir() instead
of obsolete perf_create_debug_dir()]

It contains the following squashed commit:

Perf: arm64: avoid use of uninitialized variable

Variable "buf" is not initialized. It may contain the data from
stack when "count == 0". Kernel information leak is possible due
to different code path execution.

The commit also includes the squashed patch from Olav as follows:

Perf: arm64: Use proper API to get task cpu

Instead of directly accessing thread_info structure to get the
task cpu use the appropriate wrapper API. This is needed for
subsequent patches refactors the location of task cpu.

Change-Id: I83c3ca3fbc9237c9109735f5fb2398cb4e9f6de9
Signed-off-by: Olav Haugan <ohaugan@codeaurora.org>
Signed-off-by: Patrick Fay <pfay@codeaurora.org>
Signed-off-by: Raghavendra Rao Ananta <rananta@codeaurora.org>
Signed-off-by: Rishabh Bhatnagar <rishabhb@codeaurora.org>

arch/arm64/kernel/Makefile

@@ -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_trace_counters.c

@@ -0,0 +1,171 @@
+// 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

@@ -0,0 +1,103 @@
+/* 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

@@ -0,0 +1,89 @@
+// 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);

arch/arm64/kernel/perf_trace_user.h

@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (c) 2014, 2017-2018, The Linux Foundation. All rights reserved.
+ */
+#if !defined(_PERF_TRACE_USER_H_) || defined(TRACE_HEADER_MULTI_READ)
+#define _PERF_TRACE_USER_H_
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM perf_trace_counters
+
+#include <linux/tracepoint.h>
+
+#define CNTENSET_CC    0x80000000
+#define NUM_L1_CTRS             6
+
+TRACE_EVENT(perf_trace_user,
+	TP_PROTO(char *string, u32 cnten_val),
+	TP_ARGS(string, cnten_val),
+
+	TP_STRUCT__entry(
+		__field(u32, cctr)
+		__field(u32, ctr0)
+		__field(u32, ctr1)
+		__field(u32, ctr2)
+		__field(u32, ctr3)
+		__field(u32, ctr4)
+		__field(u32, ctr5)
+		__string(user_string, string)
+		),
+
+	TP_fast_assign(
+		u32 cnt;
+		u32 l1_cnts[NUM_L1_CTRS];
+		int i;
+
+		if (cnten_val & CNTENSET_CC) {
+			/* Read value */
+			cnt = read_sysreg(pmccntr_el0);
+			__entry->cctr = cnt;
+		} else
+			__entry->cctr = 0;
+		for (i = 0; i < NUM_L1_CTRS; i++) {
+			if (cnten_val & (1 << i)) {
+				/* Select */
+				write_sysreg(i, pmselr_el0);
+				isb();
+				/* Read value */
+				cnt = read_sysreg(pmxevcntr_el0);
+				l1_cnts[i] = cnt;
+			} else {
+				l1_cnts[i] = 0;
+			}
+		}
+
+		__entry->ctr0 = l1_cnts[0];
+		__entry->ctr1 = l1_cnts[1];
+		__entry->ctr2 = l1_cnts[2];
+		__entry->ctr3 = l1_cnts[3];
+		__entry->ctr4 = l1_cnts[4];
+		__entry->ctr5 = l1_cnts[5];
+		__assign_str(user_string, string);
+	),
+
+	TP_printk("CCNTR: %u, CTR0: %u, CTR1: %u, CTR2: %u, CTR3: %u, CTR4: %u, CTR5: %u, MSG=%s",
+			__entry->cctr,
+			__entry->ctr0, __entry->ctr1,
+			__entry->ctr2, __entry->ctr3,
+			__entry->ctr4, __entry->ctr5,
+			__get_str(user_string)
+		)
+);
+
+#endif
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH ../../arch/arm64/kernel
+#define TRACE_INCLUDE_FILE perf_trace_user
+#include <trace/define_trace.h>