[CPUFREQ] conservative: fixup governor to function more like ondemand logic

As conservative is based off ondemand the codebases occasionally need to be
resync'd.  This patch, although ugly, does this.
Signed-off-by: Alexander Clouter <alex@digriz.org.uk>
Signed-off-by: Dave Jones <davej@redhat.com>

drivers/cpufreq/cpufreq_conservative.c

@@ -13,22 +13,17 @@
 
 #include <linux/kernel.h>
 #include <linux/module.h>
-#include <linux/smp.h>
 #include <linux/init.h>
-#include <linux/interrupt.h>
-#include <linux/ctype.h>
 #include <linux/cpufreq.h>
-#include <linux/sysctl.h>
-#include <linux/types.h>
-#include <linux/fs.h>
-#include <linux/sysfs.h>
 #include <linux/cpu.h>
-#include <linux/kmod.h>
-#include <linux/workqueue.h>
 #include <linux/jiffies.h>
 #include <linux/kernel_stat.h>
-#include <linux/percpu.h>
 #include <linux/mutex.h>
+#include <linux/hrtimer.h>
+#include <linux/tick.h>
+#include <linux/ktime.h>
+#include <linux/sched.h>
+
 /*
  * dbs is used in this file as a shortform for demandbased switching
  * It helps to keep variable names smaller, simpler
@@ -43,8 +38,8 @@
  * latency of the processor. The governor will work on any processor with
  * transition latency <= 10mS, using appropriate sampling
  * rate.
- * For CPUs with transition latency > 10mS (mostly drivers
- * with CPUFREQ_ETERNAL), this governor will not work.
+ * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL)
+ * this governor will not work.
  * All times here are in uS.
  */
 static unsigned int def_sampling_rate;
@@ -75,12 +70,15 @@ static unsigned int minimum_sampling_rate(void)
 static void do_dbs_timer(struct work_struct *work);
 
 struct cpu_dbs_info_s {
+	cputime64_t prev_cpu_idle;
+	cputime64_t prev_cpu_wall;
+	cputime64_t prev_cpu_nice;
 	struct cpufreq_policy *cur_policy;
-	unsigned int prev_cpu_idle_up;
-	unsigned int prev_cpu_idle_down;
-	unsigned int enable;
+	struct delayed_work work;
 	unsigned int down_skip;
 	unsigned int requested_freq;
+	int cpu;
+	unsigned int enable:1;
 };
 static DEFINE_PER_CPU(struct cpu_dbs_info_s, cpu_dbs_info);
 
@@ -95,18 +93,17 @@ static unsigned int dbs_enable;	/* number of CPUs using this policy */
  * is recursive for the same process. -Venki
  */
 static DEFINE_MUTEX(dbs_mutex);
-static DECLARE_DELAYED_WORK(dbs_work, do_dbs_timer);
 
-struct dbs_tuners {
+static struct workqueue_struct	*kconservative_wq;
+
+static struct dbs_tuners {
 	unsigned int sampling_rate;
 	unsigned int sampling_down_factor;
 	unsigned int up_threshold;
 	unsigned int down_threshold;
 	unsigned int ignore_nice;
 	unsigned int freq_step;
-};
-
-static struct dbs_tuners dbs_tuners_ins = {
+} dbs_tuners_ins = {
 	.up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
 	.down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD,
 	.sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR,
@@ -114,18 +111,37 @@ static struct dbs_tuners dbs_tuners_ins = {
 	.freq_step = 5,
 };
 
-static inline unsigned int get_cpu_idle_time(unsigned int cpu)
+static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu,
+							cputime64_t *wall)
 {
-	unsigned int add_nice = 0, ret;
+	cputime64_t idle_time;
+	cputime64_t cur_wall_time;
+	cputime64_t busy_time;
 
-	if (dbs_tuners_ins.ignore_nice)
-		add_nice = kstat_cpu(cpu).cpustat.nice;
+	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
+	busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user,
+			kstat_cpu(cpu).cpustat.system);
+
+	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq);
+	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq);
+	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal);
+	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice);
+
+	idle_time = cputime64_sub(cur_wall_time, busy_time);
+	if (wall)
+		*wall = cur_wall_time;
+
+	return 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, wall);
 
-	ret = kstat_cpu(cpu).cpustat.idle +
-		kstat_cpu(cpu).cpustat.iowait +
-		add_nice;
+	if (idle_time == -1ULL)
+		return get_cpu_idle_time_jiffy(cpu, wall);
 
-	return ret;
+	return idle_time;
 }
 
 /* keep track of frequency transitions */
@@ -213,6 +229,7 @@ static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused,
 	unsigned int input;
 	int ret;
 	ret = sscanf(buf, "%u", &input);
+
 	if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
 		return -EINVAL;
 
@@ -230,11 +247,10 @@ static ssize_t store_sampling_rate(struct cpufreq_policy *unused,
 	int ret;
 	ret = sscanf(buf, "%u", &input);
 
-	mutex_lock(&dbs_mutex);
-	if (ret != 1) {
-		mutex_unlock(&dbs_mutex);
+	if (ret != 1)
 		return -EINVAL;
-	}
+
+	mutex_lock(&dbs_mutex);
 	dbs_tuners_ins.sampling_rate = max(input, minimum_sampling_rate());
 	mutex_unlock(&dbs_mutex);
 
@@ -269,7 +285,9 @@ static ssize_t store_down_threshold(struct cpufreq_policy *unused,
 	ret = sscanf(buf, "%u", &input);
 
 	mutex_lock(&dbs_mutex);
-	if (ret != 1 || input > 100 || input >= dbs_tuners_ins.up_threshold) {
+	/* cannot be lower than 11 otherwise freq will not fall */
+	if (ret != 1 || input < 11 || input > 100 ||
+			input >= dbs_tuners_ins.up_threshold) {
 		mutex_unlock(&dbs_mutex);
 		return -EINVAL;
 	}
@@ -302,12 +320,14 @@ static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy,
 	}
 	dbs_tuners_ins.ignore_nice = input;
 
-	/* we need to re-evaluate prev_cpu_idle_up and prev_cpu_idle_down */
+	/* we need to re-evaluate prev_cpu_idle */
 	for_each_online_cpu(j) {
-		struct cpu_dbs_info_s *j_dbs_info;
-		j_dbs_info = &per_cpu(cpu_dbs_info, j);
-		j_dbs_info->prev_cpu_idle_up = get_cpu_idle_time(j);
-		j_dbs_info->prev_cpu_idle_down = j_dbs_info->prev_cpu_idle_up;
+		struct cpu_dbs_info_s *dbs_info;
+		dbs_info = &per_cpu(cpu_dbs_info, j);
+		dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
+						&dbs_info->prev_cpu_wall);
+		if (dbs_tuners_ins.ignore_nice)
+			dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
 	}
 	mutex_unlock(&dbs_mutex);
 
@@ -319,7 +339,6 @@ static ssize_t store_freq_step(struct cpufreq_policy *policy,
 {
 	unsigned int input;
 	int ret;
-
 	ret = sscanf(buf, "%u", &input);
 
 	if (ret != 1)
@@ -367,55 +386,78 @@ static struct attribute_group dbs_attr_group = {
 
 /************************** sysfs end ************************/
 
-static void dbs_check_cpu(int cpu)
+static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
 {
-	unsigned int idle_ticks, up_idle_ticks, down_idle_ticks;
-	unsigned int tmp_idle_ticks, total_idle_ticks;
+	unsigned int load = 0;
 	unsigned int freq_target;
-	unsigned int freq_down_sampling_rate;
-	struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info, cpu);
-	struct cpufreq_policy *policy;
 
-	if (!this_dbs_info->enable)
-		return;
+	struct cpufreq_policy *policy;
+	unsigned int j;
 
 	policy = this_dbs_info->cur_policy;
 
 	/*
-	 * The default safe range is 20% to 80%
-	 * Every sampling_rate, we check
-	 *	- If current idle time is less than 20%, then we try to
-	 *	  increase frequency
-	 * Every sampling_rate*sampling_down_factor, we check
-	 *	- If current idle time is more than 80%, then we try to
-	 *	  decrease frequency
+	 * Every sampling_rate, we check, if current idle time is less
+	 * than 20% (default), then we try to increase frequency
+	 * Every sampling_rate*sampling_down_factor, we check, if current
+	 * idle time is more than 80%, then we try to decrease frequency
 	 *
 	 * Any frequency increase takes it to the maximum frequency.
 	 * Frequency reduction happens at minimum steps of
-	 * 5% (default) of max_frequency
+	 * 5% (default) of maximum frequency
 	 */
 
-	/* Check for frequency increase */
-	idle_ticks = UINT_MAX;
+	/* Get Absolute Load */
+	for_each_cpu(j, policy->cpus) {
+		struct cpu_dbs_info_s *j_dbs_info;
+		cputime64_t cur_wall_time, cur_idle_time;
+		unsigned int idle_time, wall_time;
 
-	/* Check for frequency increase */
-	total_idle_ticks = get_cpu_idle_time(cpu);
-	tmp_idle_ticks = total_idle_ticks -
-		this_dbs_info->prev_cpu_idle_up;
-	this_dbs_info->prev_cpu_idle_up = total_idle_ticks;
+		j_dbs_info = &per_cpu(cpu_dbs_info, j);
+
+		cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);
 
-	if (tmp_idle_ticks < idle_ticks)
-		idle_ticks = tmp_idle_ticks;
+		wall_time = (unsigned int) cputime64_sub(cur_wall_time,
+				j_dbs_info->prev_cpu_wall);
+		j_dbs_info->prev_cpu_wall = cur_wall_time;
 
-	/* Scale idle ticks by 100 and compare with up and down ticks */
-	idle_ticks *= 100;
-	up_idle_ticks = (100 - dbs_tuners_ins.up_threshold) *
-			usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
+		idle_time = (unsigned int) cputime64_sub(cur_idle_time,
+				j_dbs_info->prev_cpu_idle);
+		j_dbs_info->prev_cpu_idle = cur_idle_time;
 
-	if (idle_ticks < up_idle_ticks) {
+		if (dbs_tuners_ins.ignore_nice) {
+			cputime64_t cur_nice;
+			unsigned long cur_nice_jiffies;
+
+			cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice,
+					 j_dbs_info->prev_cpu_nice);
+			/*
+			 * Assumption: nice time between sampling periods will
+			 * be less than 2^32 jiffies for 32 bit sys
+			 */
+			cur_nice_jiffies = (unsigned long)
+					cputime64_to_jiffies64(cur_nice);
+
+			j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
+			idle_time += jiffies_to_usecs(cur_nice_jiffies);
+		}
+
+		if (unlikely(!wall_time || wall_time < idle_time))
+			continue;
+
+		load = 100 * (wall_time - idle_time) / wall_time;
+	}
+
+	/*
+	 * break out if we 'cannot' reduce the speed as the user might
+	 * want freq_step to be zero
+	 */
+	if (dbs_tuners_ins.freq_step == 0)
+		return;
+
+	/* Check for frequency increase */
+	if (load > dbs_tuners_ins.up_threshold) {
 		this_dbs_info->down_skip = 0;
-		this_dbs_info->prev_cpu_idle_down =
-			this_dbs_info->prev_cpu_idle_up;
 
 		/* if we are already at full speed then break out early */
 		if (this_dbs_info->requested_freq == policy->max)
@@ -436,49 +478,24 @@ static void dbs_check_cpu(int cpu)
 		return;
 	}
 
-	/* Check for frequency decrease */
-	this_dbs_info->down_skip++;
-	if (this_dbs_info->down_skip < dbs_tuners_ins.sampling_down_factor)
-		return;
-
-	/* Check for frequency decrease */
-	total_idle_ticks = this_dbs_info->prev_cpu_idle_up;
-	tmp_idle_ticks = total_idle_ticks -
-		this_dbs_info->prev_cpu_idle_down;
-	this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
-
-	if (tmp_idle_ticks < idle_ticks)
-		idle_ticks = tmp_idle_ticks;
-
-	/* Scale idle ticks by 100 and compare with up and down ticks */
-	idle_ticks *= 100;
-	this_dbs_info->down_skip = 0;
-
-	freq_down_sampling_rate = dbs_tuners_ins.sampling_rate *
-		dbs_tuners_ins.sampling_down_factor;
-	down_idle_ticks = (100 - dbs_tuners_ins.down_threshold) *
-		usecs_to_jiffies(freq_down_sampling_rate);
-
-	if (idle_ticks > down_idle_ticks) {
 	/*
-		 * if we are already at the lowest speed then break out early
-		 * or if we 'cannot' reduce the speed as the user might want
-		 * freq_target to be zero
+	 * The optimal frequency is the frequency that is the lowest that
+	 * can support the current CPU usage without triggering the up
+	 * policy. To be safe, we focus 10 points under the threshold.
 	 */
-		if (this_dbs_info->requested_freq == policy->min
-				|| dbs_tuners_ins.freq_step == 0)
-			return;
-
+	if (load < (dbs_tuners_ins.down_threshold - 10)) {
 		freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
 
-		/* max freq cannot be less than 100. But who knows.... */
-		if (unlikely(freq_target == 0))
-			freq_target = 5;
-
 		this_dbs_info->requested_freq -= freq_target;
 		if (this_dbs_info->requested_freq < policy->min)
 			this_dbs_info->requested_freq = policy->min;
 
+		/*
+		 * if we cannot reduce the frequency anymore, break out early
+		 */
+		if (policy->cur == policy->min)
+			return;
+
 		__cpufreq_driver_target(policy, this_dbs_info->requested_freq,
 				CPUFREQ_RELATION_H);
 		return;
@@ -487,27 +504,45 @@ static void dbs_check_cpu(int cpu)
 
 static void do_dbs_timer(struct work_struct *work)
 {
-	int i;
-	mutex_lock(&dbs_mutex);
-	for_each_online_cpu(i)
-		dbs_check_cpu(i);
-	schedule_delayed_work(&dbs_work,
-			usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
-	mutex_unlock(&dbs_mutex);
+	struct cpu_dbs_info_s *dbs_info =
+		container_of(work, struct cpu_dbs_info_s, work.work);
+	unsigned int cpu = dbs_info->cpu;
+
+	/* We want all CPUs to do sampling nearly on same jiffy */
+	int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
+
+	delay -= jiffies % delay;
+
+	if (lock_policy_rwsem_write(cpu) < 0)
+		return;
+
+	if (!dbs_info->enable) {
+		unlock_policy_rwsem_write(cpu);
+		return;
+	}
+
+	dbs_check_cpu(dbs_info);
+
+	queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay);
+	unlock_policy_rwsem_write(cpu);
 }
 
-static inline void dbs_timer_init(void)
+static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
 {
-	init_timer_deferrable(&dbs_work.timer);
-	schedule_delayed_work(&dbs_work,
-			usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
-	return;
+	/* We want all CPUs to do sampling nearly on same jiffy */
+	int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
+	delay -= jiffies % delay;
+
+	dbs_info->enable = 1;
+	INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
+	queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work,
+				delay);
 }
 
-static inline void dbs_timer_exit(void)
+static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
 {
-	cancel_delayed_work(&dbs_work);
-	return;
+	dbs_info->enable = 0;
+	cancel_delayed_work(&dbs_info->work);
 }
 
 static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
@@ -541,11 +576,13 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
 			j_dbs_info = &per_cpu(cpu_dbs_info, j);
 			j_dbs_info->cur_policy = policy;
 
-			j_dbs_info->prev_cpu_idle_up = get_cpu_idle_time(cpu);
-			j_dbs_info->prev_cpu_idle_down
-				= j_dbs_info->prev_cpu_idle_up;
+			j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
+						&j_dbs_info->prev_cpu_wall);
+			if (dbs_tuners_ins.ignore_nice) {
+				j_dbs_info->prev_cpu_nice =
+						kstat_cpu(j).cpustat.nice;
+			}
 		}
-		this_dbs_info->enable = 1;
 		this_dbs_info->down_skip = 0;
 		this_dbs_info->requested_freq = policy->cur;
 
@@ -567,30 +604,30 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
 
 			dbs_tuners_ins.sampling_rate = def_sampling_rate;
 
-			dbs_timer_init();
 			cpufreq_register_notifier(
 					&dbs_cpufreq_notifier_block,
 					CPUFREQ_TRANSITION_NOTIFIER);
 		}
+		dbs_timer_init(this_dbs_info);
 
 		mutex_unlock(&dbs_mutex);
+
 		break;
 
 	case CPUFREQ_GOV_STOP:
 		mutex_lock(&dbs_mutex);
-		this_dbs_info->enable = 0;
+		dbs_timer_exit(this_dbs_info);
 		sysfs_remove_group(&policy->kobj, &dbs_attr_group);
 		dbs_enable--;
+
 		/*
 		 * Stop the timerschedule work, when this governor
 		 * is used for first time
 		 */
-		if (dbs_enable == 0) {
-			dbs_timer_exit();
+		if (dbs_enable == 0)
 			cpufreq_unregister_notifier(
 					&dbs_cpufreq_notifier_block,
 					CPUFREQ_TRANSITION_NOTIFIER);
-		}
 
 		mutex_unlock(&dbs_mutex);
 
@@ -607,6 +644,7 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
 					this_dbs_info->cur_policy,
 					policy->min, CPUFREQ_RELATION_L);
 		mutex_unlock(&dbs_mutex);
+
 		break;
 	}
 	return 0;
@@ -624,15 +662,25 @@ struct cpufreq_governor cpufreq_gov_conservative = {
 
 static int __init cpufreq_gov_dbs_init(void)
 {
-	return cpufreq_register_governor(&cpufreq_gov_conservative);
+	int err;
+
+	kconservative_wq = create_workqueue("kconservative");
+	if (!kconservative_wq) {
+		printk(KERN_ERR "Creation of kconservative failed\n");
+		return -EFAULT;
+	}
+
+	err = cpufreq_register_governor(&cpufreq_gov_conservative);
+	if (err)
+		destroy_workqueue(kconservative_wq);
+
+	return err;
 }
 
 static void __exit cpufreq_gov_dbs_exit(void)
 {
-	/* Make sure that the scheduled work is indeed not running */
-	flush_scheduled_work();
-
 	cpufreq_unregister_governor(&cpufreq_gov_conservative);
+	destroy_workqueue(kconservative_wq);
 }