Merge "msm: rq_stats: Support to know cpu utilization in userspace" into msm-3.4
diff --git a/arch/arm/mach-msm/msm_rq_stats.c b/arch/arm/mach-msm/msm_rq_stats.c
index 2ea7ed3..ea08f4b 100644
--- a/arch/arm/mach-msm/msm_rq_stats.c
+++ b/arch/arm/mach-msm/msm_rq_stats.c
@@ -1,4 +1,4 @@
-/* Copyright (c) 2010-2011, Code Aurora Forum. All rights reserved.
+/* Copyright (c) 2010-2012, Code Aurora Forum. 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
@@ -11,7 +11,7 @@
  *
  */
 /*
- * Qualcomm MSM Runqueue Stats Interface for Userspace
+ * Qualcomm MSM Runqueue Stats and cpu utilization Interface for Userspace
  */
 #include <linux/kernel.h>
 #include <linux/init.h>
@@ -26,12 +26,183 @@
 #include <linux/sched.h>
 #include <linux/spinlock.h>
 #include <linux/rq_stats.h>
+#include <linux/cpufreq.h>
+#include <linux/kernel_stat.h>
+#include <linux/tick.h>
 #include <asm/smp_plat.h>
 
 #define MAX_LONG_SIZE 24
 #define DEFAULT_RQ_POLL_JIFFIES 1
 #define DEFAULT_DEF_TIMER_JIFFIES 5
 
+struct notifier_block freq_transition;
+struct notifier_block cpu_hotplug;
+
+struct cpu_load_data {
+	cputime64_t prev_cpu_idle;
+	cputime64_t prev_cpu_wall;
+	cputime64_t prev_cpu_iowait;
+	unsigned int avg_load_maxfreq;
+	unsigned int samples;
+	unsigned int window_size;
+	unsigned int cur_freq;
+	unsigned int policy_max;
+	cpumask_var_t related_cpus;
+	struct mutex cpu_load_mutex;
+};
+
+static DEFINE_PER_CPU(struct cpu_load_data, cpuload);
+
+static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
+{
+	u64 idle_time;
+	u64 cur_wall_time;
+	u64 busy_time;
+
+	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
+
+	busy_time  = kcpustat_cpu(cpu).cpustat[CPUTIME_USER];
+	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SYSTEM];
+	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_IRQ];
+	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_SOFTIRQ];
+	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_STEAL];
+	busy_time += kcpustat_cpu(cpu).cpustat[CPUTIME_NICE];
+
+	idle_time = cur_wall_time - busy_time;
+	if (wall)
+		*wall = jiffies_to_usecs(cur_wall_time);
+
+	return jiffies_to_usecs(idle_time);
+}
+
+static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
+{
+	u64 idle_time = get_cpu_idle_time_us(cpu, NULL);
+
+	if (idle_time == -1ULL)
+		return get_cpu_idle_time_jiffy(cpu, wall);
+	else
+		idle_time += get_cpu_iowait_time_us(cpu, wall);
+
+	return idle_time;
+}
+
+static inline cputime64_t get_cpu_iowait_time(unsigned int cpu,
+							cputime64_t *wall)
+{
+	u64 iowait_time = get_cpu_iowait_time_us(cpu, wall);
+
+	if (iowait_time == -1ULL)
+		return 0;
+
+	return iowait_time;
+}
+
+static int update_average_load(unsigned int freq, unsigned int cpu)
+{
+
+	struct cpu_load_data *pcpu = &per_cpu(cpuload, cpu);
+	cputime64_t cur_wall_time, cur_idle_time, cur_iowait_time;
+	unsigned int idle_time, wall_time, iowait_time;
+	unsigned int cur_load, load_at_max_freq;
+
+	cur_idle_time = get_cpu_idle_time(cpu, &cur_wall_time);
+	cur_iowait_time = get_cpu_iowait_time(cpu, &cur_wall_time);
+
+	wall_time = (unsigned int) (cur_wall_time - pcpu->prev_cpu_wall);
+	pcpu->prev_cpu_wall = cur_wall_time;
+
+	idle_time = (unsigned int) (cur_idle_time - pcpu->prev_cpu_idle);
+	pcpu->prev_cpu_idle = cur_idle_time;
+
+	iowait_time = (unsigned int) (cur_iowait_time - pcpu->prev_cpu_iowait);
+	pcpu->prev_cpu_iowait = cur_iowait_time;
+
+	if (idle_time >= iowait_time)
+		idle_time -= iowait_time;
+
+	if (unlikely(!wall_time || wall_time < idle_time))
+		return 0;
+
+	cur_load = 100 * (wall_time - idle_time) / wall_time;
+
+	/* Calculate the scaled load across CPU */
+	load_at_max_freq = (cur_load * freq) / pcpu->policy_max;
+
+	if (!pcpu->avg_load_maxfreq) {
+		/* This is the first sample in this window*/
+		pcpu->avg_load_maxfreq = load_at_max_freq;
+		pcpu->window_size = wall_time;
+	} else {
+		/*
+		 * The is already a sample available in this window.
+		 * Compute weighted average with prev entry, so that we get
+		 * the precise weighted load.
+		 */
+		pcpu->avg_load_maxfreq =
+			((pcpu->avg_load_maxfreq * pcpu->window_size) +
+			(load_at_max_freq * wall_time)) /
+			(wall_time + pcpu->window_size);
+
+		pcpu->window_size += wall_time;
+	}
+
+	return 0;
+}
+
+static unsigned int report_load_at_max_freq(void)
+{
+	int cpu;
+	struct cpu_load_data *pcpu;
+	unsigned int total_load = 0;
+
+	for_each_online_cpu(cpu) {
+		pcpu = &per_cpu(cpuload, cpu);
+		mutex_lock(&pcpu->cpu_load_mutex);
+		update_average_load(pcpu->cur_freq, cpu);
+		total_load += pcpu->avg_load_maxfreq;
+		pcpu->avg_load_maxfreq = 0;
+		mutex_unlock(&pcpu->cpu_load_mutex);
+	}
+	return total_load;
+}
+
+static int cpufreq_transition_handler(struct notifier_block *nb,
+			unsigned long val, void *data)
+{
+	struct cpufreq_freqs *freqs = data;
+	struct cpu_load_data *this_cpu = &per_cpu(cpuload, freqs->cpu);
+	int j;
+
+	switch (val) {
+	case CPUFREQ_POSTCHANGE:
+		for_each_cpu(j, this_cpu->related_cpus) {
+			struct cpu_load_data *pcpu = &per_cpu(cpuload, j);
+			mutex_lock(&pcpu->cpu_load_mutex);
+			update_average_load(freqs->old, freqs->cpu);
+			pcpu->cur_freq = freqs->new;
+			mutex_unlock(&pcpu->cpu_load_mutex);
+		}
+		break;
+	}
+	return 0;
+}
+
+static int cpu_hotplug_handler(struct notifier_block *nb,
+			unsigned long val, void *data)
+{
+	unsigned int cpu = (unsigned long)data;
+	struct cpu_load_data *this_cpu = &per_cpu(cpuload, cpu);
+
+	switch (val) {
+	case CPU_ONLINE:
+	case CPU_ONLINE_FROZEN:
+		this_cpu->avg_load_maxfreq = 0;
+	}
+
+	return NOTIFY_OK;
+}
+
 static void def_work_fn(struct work_struct *work)
 {
 	int64_t diff;
@@ -121,7 +292,18 @@
 	__ATTR(def_timer_ms, S_IWUSR | S_IRUSR, show_def_timer_ms,
 			store_def_timer_ms);
 
+static ssize_t show_cpu_normalized_load(struct kobject *kobj,
+		struct kobj_attribute *attr, char *buf)
+{
+	return snprintf(buf, MAX_LONG_SIZE, "%u\n", report_load_at_max_freq());
+}
+
+static struct kobj_attribute cpu_normalized_load_attr =
+	__ATTR(cpu_normalized_load, S_IWUSR | S_IRUSR, show_cpu_normalized_load,
+			NULL);
+
 static struct attribute *rq_attrs[] = {
+	&cpu_normalized_load_attr.attr,
 	&def_timer_ms_attr.attr,
 	&run_queue_avg_attr.attr,
 	&run_queue_poll_ms_attr.attr,
@@ -157,7 +339,8 @@
 static int __init msm_rq_stats_init(void)
 {
 	int ret;
-
+	int i;
+	struct cpufreq_policy cpu_policy;
 	/* Bail out if this is not an SMP Target */
 	if (!is_smp()) {
 		rq_info.init = 0;
@@ -175,6 +358,20 @@
 	ret = init_rq_attribs();
 
 	rq_info.init = 1;
+
+	for_each_possible_cpu(i) {
+		struct cpu_load_data *pcpu = &per_cpu(cpuload, i);
+		mutex_init(&pcpu->cpu_load_mutex);
+		cpufreq_get_policy(&cpu_policy, i);
+		pcpu->policy_max = cpu_policy.cpuinfo.max_freq;
+		cpumask_copy(pcpu->related_cpus, cpu_policy.cpus);
+	}
+	freq_transition.notifier_call = cpufreq_transition_handler;
+	cpu_hotplug.notifier_call = cpu_hotplug_handler;
+	cpufreq_register_notifier(&freq_transition,
+					CPUFREQ_TRANSITION_NOTIFIER);
+	register_hotcpu_notifier(&cpu_hotplug);
+
 	return ret;
 }
 late_initcall(msm_rq_stats_init);