Perf: arm64: Add L1 counters to tracepoints

Create tracepoints framework for collecting L1 counters on
per-pid basis.
Add hooks into scheduler to invoke collection.

Change-Id: I04cbf0c526162e70680573f5f365f8b41d3079ed
Signed-off-by: Sheetal Sahasrabudhe <sheetals@codeaurora.org>
[satyap: trivial merge conflict resolution]
Signed-off-by: Satya Durga Srinivasu Prabhala <satyap@codeaurora.org>

arch/arm64/kernel/Makefile

@@ -31,7 +31,8 @@ arm64-obj-$(CONFIG_COMPAT)		+= sys32.o kuser32.o signal32.o 	\
 arm64-obj-$(CONFIG_FUNCTION_TRACER)	+= ftrace.o entry-ftrace.o
 arm64-obj-$(CONFIG_MODULES)		+= arm64ksyms.o module.o
 arm64-obj-$(CONFIG_PERF_EVENTS)		+= perf_regs.o perf_callchain.o
-arm64-obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event.o perf_debug.o
+arm64-obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event.o perf_debug.o		\
+					   perf_trace_counters.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_debug.c

@@ -23,6 +23,7 @@
  */
 static char *descriptions =
 	" 0 arm64: perf: add debug patch logging framework\n"
+	" 1 Perf: arm64: Add L1 counters to tracepoints\n"
 ;
 
 static ssize_t desc_read(struct file *fp, char __user *buf,

arch/arm64/kernel/perf_trace_counters.c

@@ -0,0 +1,125 @@
+/* Copyright (c) 2013-2014, 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/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, previous_ccnt);
+DEFINE_PER_CPU(u32[NUM_L1_CTRS], previous_l1_cnts);
+DEFINE_PER_CPU(u32, old_pid);
+
+void tracectr_notifier(void *ignore, struct task_struct *prev,
+					struct task_struct *next)
+{
+	int current_pid;
+	u32 cpu = next->on_cpu;
+
+	if (tp_pid_state != 1)
+		return;
+	current_pid = next->pid;
+	if (per_cpu(old_pid, cpu) != -1)
+		trace_sched_switch_with_ctrs(per_cpu(old_pid, cpu),
+				current_pid);
+	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_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;
+
+	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;
+	return 0;
+}
+
+int __exit exit_tracecounters(void)
+{
+	return 0;
+}
+late_initcall(init_tracecounters);

arch/arm64/kernel/perf_trace_counters.h

@@ -0,0 +1,116 @@
+/* Copyright (c) 2013-2014, 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.
+ */
+
+#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 C_ALL (CC | C0 | C1 | C2 | C3)
+#define NUM_L1_CTRS 4
+
+#include <linux/sched.h>
+#include <linux/cpumask.h>
+#include <linux/tracepoint.h>
+
+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, lctr0)
+			__field(u32, lctr1)
+		),
+
+		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;
+
+
+			/* Read PMCNTENSET */
+			asm volatile("mrs %0, pmcntenset_el0"
+							: "=r" (cnten_val));
+			/* Disable all the counters that were enabled */
+			asm volatile("msr pmcntenclr_el0, %0"
+							: : "r" (cnten_val));
+			if (cnten_val & CC) {
+				asm volatile("mrs %0, pmccntr_el0"
+							: "=r" (total_ccnt));
+				/* Read value */
+				__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 */
+					asm volatile("msr pmselr_el0, %0"
+							: : "r" (i));
+					isb();
+					asm volatile("mrs %0, pmxevcntr_el0"
+							: "=r" (total_cnt));
+					/* Read value */
+					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;
+			}
+			/* Enable all the counters that were disabled */
+			asm volatile("msr pmcntenset_el0, %0"
+							: : "r" (cnten_val));
+
+			__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->lctr0 = 0;
+			__entry->lctr1 = 0;
+		),
+
+		TP_printk("prev_pid=%d, next_pid=%d, CCNTR: %u, CTR0: %u, CTR1: %u, CTR2: %u, CTR3: %u, L2CTR0: %u, L2CTR1: %u",
+				__entry->old_pid, __entry->new_pid,
+				__entry->cctr, __entry->ctr0, __entry->ctr1,
+				__entry->ctr2, __entry->ctr3,
+				__entry->lctr0, __entry->lctr1)
+);
+
+#endif
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH ../../arch/arm64/kernel
+#define TRACE_INCLUDE_FILE perf_trace_counters
+#include <trace/define_trace.h>