[CPUFREQ] Remove cpufreq_stats sysfs entries on module unload.
[linux-2.6.git] / drivers / cpufreq / cpufreq_conservative.c
index 8fe13ec..33b56e5 100644 (file)
@@ -4,7 +4,7 @@
  *  Copyright (C)  2001 Russell King
  *            (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
  *                      Jun Nakajima <jun.nakajima@intel.com>
- *            (C)  2004 Alexander Clouter <alex-kernel@digriz.org.uk>
+ *            (C)  2009 Alexander Clouter <alex@digriz.org.uk>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
 
 #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/sched.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
 #define DEF_FREQUENCY_UP_THRESHOLD             (80)
 #define DEF_FREQUENCY_DOWN_THRESHOLD           (20)
 
-/* 
- * The polling frequency of this governor depends on the capability of 
+/*
+ * The polling frequency of this governor depends on the capability of
  * the processor. Default polling frequency is 1000 times the transition
- * latency of the processor. The governor will work on any processor with 
- * transition latency <= 10mS, using appropriate sampling 
+ * 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.
  * All times here are in uS.
  */
-static unsigned int                            def_sampling_rate;
 #define MIN_SAMPLING_RATE_RATIO                        (2)
-/* for correct statistics, we need at least 10 ticks between each measure */
-#define MIN_STAT_SAMPLING_RATE                 (MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10))
-#define MIN_SAMPLING_RATE                      (def_sampling_rate / MIN_SAMPLING_RATE_RATIO)
-#define MAX_SAMPLING_RATE                      (500 * def_sampling_rate)
-#define DEF_SAMPLING_RATE_LATENCY_MULTIPLIER   (1000)
+
+static unsigned int min_sampling_rate;
+
+#define LATENCY_MULTIPLIER                     (1000)
+#define MIN_LATENCY_MULTIPLIER                 (100)
 #define DEF_SAMPLING_DOWN_FACTOR               (1)
 #define MAX_SAMPLING_DOWN_FACTOR               (10)
-#define TRANSITION_LATENCY_LIMIT               (10 * 1000)
+#define TRANSITION_LATENCY_LIMIT               (10 * 1000 * 1000)
 
-static void do_dbs_timer(void *data);
+static void do_dbs_timer(struct work_struct *work);
 
 struct cpu_dbs_info_s {
-       struct cpufreq_policy   *cur_policy;
-       unsigned int            prev_cpu_idle_up;
-       unsigned int            prev_cpu_idle_down;
-       unsigned int            enable;
-       unsigned int            down_skip;
-       unsigned int            requested_freq;
+       cputime64_t prev_cpu_idle;
+       cputime64_t prev_cpu_wall;
+       cputime64_t prev_cpu_nice;
+       struct cpufreq_policy *cur_policy;
+       struct delayed_work work;
+       unsigned int down_skip;
+       unsigned int requested_freq;
+       int cpu;
+       unsigned int enable:1;
+       /*
+        * percpu mutex that serializes governor limit change with
+        * do_dbs_timer invocation. We do not want do_dbs_timer to run
+        * when user is changing the governor or limits.
+        */
+       struct mutex timer_mutex;
 };
-static DEFINE_PER_CPU(struct cpu_dbs_info_s, cpu_dbs_info);
+static DEFINE_PER_CPU(struct cpu_dbs_info_s, cs_cpu_dbs_info);
 
 static unsigned int dbs_enable;        /* number of CPUs using this policy */
 
 /*
- * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug
- * lock and dbs_mutex. cpu_hotplug lock should always be held before
- * dbs_mutex. If any function that can potentially take cpu_hotplug lock
- * (like __cpufreq_driver_target()) is being called with dbs_mutex taken, then
- * cpu_hotplug lock should be taken before that. Note that cpu_hotplug lock
- * is recursive for the same process. -Venki
+ * dbs_mutex protects dbs_enable in governor start/stop.
  */
-static DEFINE_MUTEX    (dbs_mutex);
-static DECLARE_WORK    (dbs_work, do_dbs_timer, NULL);
-
-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 = {
-       .up_threshold           = DEF_FREQUENCY_UP_THRESHOLD,
-       .down_threshold         = DEF_FREQUENCY_DOWN_THRESHOLD,
-       .sampling_down_factor   = DEF_SAMPLING_DOWN_FACTOR,
-       .ignore_nice            = 0,
-       .freq_step              = 5,
+static DEFINE_MUTEX(dbs_mutex);
+
+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;
+} dbs_tuners_ins = {
+       .up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
+       .down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD,
+       .sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR,
+       .ignore_nice = 0,
+       .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)
 {
-       return  kstat_cpu(cpu).cpustat.idle +
-               kstat_cpu(cpu).cpustat.iowait +
-               ( dbs_tuners_ins.ignore_nice ?
-                 kstat_cpu(cpu).cpustat.nice :
-                 0);
+       cputime64_t idle_time;
+       cputime64_t cur_wall_time;
+       cputime64_t busy_time;
+
+       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 = (cputime64_t)jiffies_to_usecs(cur_wall_time);
+
+       return (cputime64_t)jiffies_to_usecs(idle_time);
 }
 
-/************************** sysfs interface ************************/
-static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf)
+static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
 {
-       return sprintf (buf, "%u\n", MAX_SAMPLING_RATE);
+       u64 idle_time = get_cpu_idle_time_us(cpu, wall);
+
+       if (idle_time == -1ULL)
+               return get_cpu_idle_time_jiffy(cpu, wall);
+
+       return idle_time;
 }
 
-static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf)
+/* keep track of frequency transitions */
+static int
+dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
+                    void *data)
 {
-       return sprintf (buf, "%u\n", MIN_SAMPLING_RATE);
+       struct cpufreq_freqs *freq = data;
+       struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cs_cpu_dbs_info,
+                                                       freq->cpu);
+
+       struct cpufreq_policy *policy;
+
+       if (!this_dbs_info->enable)
+               return 0;
+
+       policy = this_dbs_info->cur_policy;
+
+       /*
+        * we only care if our internally tracked freq moves outside
+        * the 'valid' ranges of freqency available to us otherwise
+        * we do not change it
+       */
+       if (this_dbs_info->requested_freq > policy->max
+                       || this_dbs_info->requested_freq < policy->min)
+               this_dbs_info->requested_freq = freq->new;
+
+       return 0;
 }
 
-#define define_one_ro(_name)                                   \
-static struct freq_attr _name =                                \
-__ATTR(_name, 0444, show_##_name, NULL)
+static struct notifier_block dbs_cpufreq_notifier_block = {
+       .notifier_call = dbs_cpufreq_notifier
+};
+
+/************************** sysfs interface ************************/
+static ssize_t show_sampling_rate_min(struct kobject *kobj,
+                                     struct attribute *attr, char *buf)
+{
+       return sprintf(buf, "%u\n", min_sampling_rate);
+}
 
-define_one_ro(sampling_rate_max);
-define_one_ro(sampling_rate_min);
+define_one_global_ro(sampling_rate_min);
 
 /* cpufreq_conservative Governor Tunables */
 #define show_one(file_name, object)                                    \
 static ssize_t show_##file_name                                                \
-(struct cpufreq_policy *unused, char *buf)                             \
+(struct kobject *kobj, struct attribute *attr, char *buf)              \
 {                                                                      \
        return sprintf(buf, "%u\n", dbs_tuners_ins.object);             \
 }
@@ -142,149 +183,125 @@ show_one(down_threshold, down_threshold);
 show_one(ignore_nice_load, ignore_nice);
 show_one(freq_step, freq_step);
 
-static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused, 
-               const char *buf, size_t count)
+static ssize_t store_sampling_down_factor(struct kobject *a,
+                                         struct attribute *b,
+                                         const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
-       ret = sscanf (buf, "%u", &input);
+       ret = sscanf(buf, "%u", &input);
+
        if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
                return -EINVAL;
 
-       mutex_lock(&dbs_mutex);
        dbs_tuners_ins.sampling_down_factor = input;
-       mutex_unlock(&dbs_mutex);
-
        return count;
 }
 
-static ssize_t store_sampling_rate(struct cpufreq_policy *unused, 
-               const char *buf, size_t count)
+static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
+                                  const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
-       ret = sscanf (buf, "%u", &input);
+       ret = sscanf(buf, "%u", &input);
 
-       mutex_lock(&dbs_mutex);
-       if (ret != 1 || input > MAX_SAMPLING_RATE || input < MIN_SAMPLING_RATE) {
-               mutex_unlock(&dbs_mutex);
+       if (ret != 1)
                return -EINVAL;
-       }
-
-       dbs_tuners_ins.sampling_rate = input;
-       mutex_unlock(&dbs_mutex);
 
+       dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate);
        return count;
 }
 
-static ssize_t store_up_threshold(struct cpufreq_policy *unused, 
-               const char *buf, size_t count)
+static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
+                                 const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
-       ret = sscanf (buf, "%u", &input);
+       ret = sscanf(buf, "%u", &input);
 
-       mutex_lock(&dbs_mutex);
-       if (ret != 1 || input > 100 || input <= dbs_tuners_ins.down_threshold) {
-               mutex_unlock(&dbs_mutex);
+       if (ret != 1 || input > 100 ||
+                       input <= dbs_tuners_ins.down_threshold)
                return -EINVAL;
-       }
 
        dbs_tuners_ins.up_threshold = input;
-       mutex_unlock(&dbs_mutex);
-
        return count;
 }
 
-static ssize_t store_down_threshold(struct cpufreq_policy *unused, 
-               const char *buf, size_t count)
+static ssize_t store_down_threshold(struct kobject *a, struct attribute *b,
+                                   const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
-       ret = sscanf (buf, "%u", &input);
+       ret = sscanf(buf, "%u", &input);
 
-       mutex_lock(&dbs_mutex);
-       if (ret != 1 || input > 100 || input >= dbs_tuners_ins.up_threshold) {
-               mutex_unlock(&dbs_mutex);
+       /* cannot be lower than 11 otherwise freq will not fall */
+       if (ret != 1 || input < 11 || input > 100 ||
+                       input >= dbs_tuners_ins.up_threshold)
                return -EINVAL;
-       }
 
        dbs_tuners_ins.down_threshold = input;
-       mutex_unlock(&dbs_mutex);
-
        return count;
 }
 
-static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy,
-               const char *buf, size_t count)
+static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
+                                     const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
 
        unsigned int j;
-       
-       ret = sscanf (buf, "%u", &input);
-       if ( ret != 1 )
+
+       ret = sscanf(buf, "%u", &input);
+       if (ret != 1)
                return -EINVAL;
 
-       if ( input > 1 )
+       if (input > 1)
                input = 1;
-       
-       mutex_lock(&dbs_mutex);
-       if ( input == dbs_tuners_ins.ignore_nice ) { /* nothing to do */
-               mutex_unlock(&dbs_mutex);
+
+       if (input == dbs_tuners_ins.ignore_nice) /* nothing to do */
                return count;
-       }
+
        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(cs_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);
-
        return count;
 }
 
-static ssize_t store_freq_step(struct cpufreq_policy *policy,
-               const char *buf, size_t count)
+static ssize_t store_freq_step(struct kobject *a, struct attribute *b,
+                              const char *buf, size_t count)
 {
        unsigned int input;
        int ret;
+       ret = sscanf(buf, "%u", &input);
 
-       ret = sscanf (buf, "%u", &input);
-
-       if ( ret != 1 )
+       if (ret != 1)
                return -EINVAL;
 
-       if ( input > 100 )
+       if (input > 100)
                input = 100;
-       
+
        /* no need to test here if freq_step is zero as the user might actually
         * want this, they would be crazy though :) */
-       mutex_lock(&dbs_mutex);
        dbs_tuners_ins.freq_step = input;
-       mutex_unlock(&dbs_mutex);
-
        return count;
 }
 
-#define define_one_rw(_name) \
-static struct freq_attr _name = \
-__ATTR(_name, 0644, show_##_name, store_##_name)
-
-define_one_rw(sampling_rate);
-define_one_rw(sampling_down_factor);
-define_one_rw(up_threshold);
-define_one_rw(down_threshold);
-define_one_rw(ignore_nice_load);
-define_one_rw(freq_step);
+define_one_global_rw(sampling_rate);
+define_one_global_rw(sampling_down_factor);
+define_one_global_rw(up_threshold);
+define_one_global_rw(down_threshold);
+define_one_global_rw(ignore_nice_load);
+define_one_global_rw(freq_step);
 
-static struct attribute * dbs_attributes[] = {
-       &sampling_rate_max.attr,
+static struct attribute *dbs_attributes[] = {
        &sampling_rate_min.attr,
        &sampling_rate.attr,
        &sampling_down_factor.attr,
@@ -302,67 +319,94 @@ 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 freq_step;
-       unsigned int freq_down_sampling_rate;
-       struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info, cpu);
-       struct cpufreq_policy *policy;
+       unsigned int load = 0;
+       unsigned int max_load = 0;
+       unsigned int freq_target;
 
-       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 
+        * Any frequency increase takes it to the maximum frequency.
+        * Frequency reduction happens at minimum steps of
+        * 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(cs_cpu_dbs_info, j);
 
-       if (tmp_idle_ticks < idle_ticks)
-               idle_ticks = tmp_idle_ticks;
+               cur_idle_time = get_cpu_idle_time(j, &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);
+               wall_time = (unsigned int) cputime64_sub(cur_wall_time,
+                               j_dbs_info->prev_cpu_wall);
+               j_dbs_info->prev_cpu_wall = cur_wall_time;
 
-       if (idle_ticks < up_idle_ticks) {
+               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 (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;
+
+               if (load > max_load)
+                       max_load = load;
+       }
+
+       /*
+        * 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 (max_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)
                        return;
-               
-               freq_step = (dbs_tuners_ins.freq_step * policy->max) / 100;
+
+               freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
 
                /* max freq cannot be less than 100. But who knows.... */
-               if (unlikely(freq_step == 0))
-                       freq_step = 5;
-               
-               this_dbs_info->requested_freq += freq_step;
+               if (unlikely(freq_target == 0))
+                       freq_target = 5;
+
+               this_dbs_info->requested_freq += freq_target;
                if (this_dbs_info->requested_freq > policy->max)
                        this_dbs_info->requested_freq = policy->max;
 
@@ -371,80 +415,64 @@ 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;
+       /*
+        * 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 (max_load < (dbs_tuners_ins.down_threshold - 10)) {
+               freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
 
-       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);
+               this_dbs_info->requested_freq -= freq_target;
+               if (this_dbs_info->requested_freq < policy->min)
+                       this_dbs_info->requested_freq = policy->min;
 
-       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_step to be zero
+                * if we cannot reduce the frequency anymore, break out early
                 */
-               if (this_dbs_info->requested_freq == policy->min
-                               || dbs_tuners_ins.freq_step == 0)
+               if (policy->cur == policy->min)
                        return;
 
-               freq_step = (dbs_tuners_ins.freq_step * policy->max) / 100;
-
-               /* max freq cannot be less than 100. But who knows.... */
-               if (unlikely(freq_step == 0))
-                       freq_step = 5;
-
-               this_dbs_info->requested_freq -= freq_step;
-               if (this_dbs_info->requested_freq < policy->min)
-                       this_dbs_info->requested_freq = policy->min;
-
                __cpufreq_driver_target(policy, this_dbs_info->requested_freq,
                                CPUFREQ_RELATION_H);
                return;
        }
 }
 
-static void do_dbs_timer(void *data)
-{ 
-       int i;
-       lock_cpu_hotplug();
-       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);
-       unlock_cpu_hotplug();
-} 
-
-static inline void dbs_timer_init(void)
+static void do_dbs_timer(struct work_struct *work)
+{
+       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;
+
+       mutex_lock(&dbs_info->timer_mutex);
+
+       dbs_check_cpu(dbs_info);
+
+       schedule_delayed_work_on(cpu, &dbs_info->work, delay);
+       mutex_unlock(&dbs_info->timer_mutex);
+}
+
+static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
 {
-       INIT_WORK(&dbs_work, do_dbs_timer, NULL);
-       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);
+       schedule_delayed_work_on(dbs_info->cpu, &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_sync(&dbs_info->work);
 }
 
 static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
@@ -455,41 +483,31 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
        unsigned int j;
        int rc;
 
-       this_dbs_info = &per_cpu(cpu_dbs_info, cpu);
+       this_dbs_info = &per_cpu(cs_cpu_dbs_info, cpu);
 
        switch (event) {
        case CPUFREQ_GOV_START:
-               if ((!cpu_online(cpu)) || 
-                   (!policy->cur))
+               if ((!cpu_online(cpu)) || (!policy->cur))
                        return -EINVAL;
 
-               if (policy->cpuinfo.transition_latency >
-                               (TRANSITION_LATENCY_LIMIT * 1000))
-                       return -EINVAL;
-               if (this_dbs_info->enable) /* Already enabled */
-                       break;
-                
                mutex_lock(&dbs_mutex);
 
-               rc = sysfs_create_group(&policy->kobj, &dbs_attr_group);
-               if (rc) {
-                       mutex_unlock(&dbs_mutex);
-                       return rc;
-               }
-
-               for_each_cpu_mask(j, policy->cpus) {
+               for_each_cpu(j, policy->cpus) {
                        struct cpu_dbs_info_s *j_dbs_info;
-                       j_dbs_info = &per_cpu(cpu_dbs_info, j);
+                       j_dbs_info = &per_cpu(cs_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;
 
+               mutex_init(&this_dbs_info->timer_mutex);
                dbs_enable++;
                /*
                 * Start the timerschedule work, when this governor
@@ -502,77 +520,106 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
                        if (latency == 0)
                                latency = 1;
 
-                       def_sampling_rate = 10 * latency *
-                                       DEF_SAMPLING_RATE_LATENCY_MULTIPLIER;
-
-                       if (def_sampling_rate < MIN_STAT_SAMPLING_RATE)
-                               def_sampling_rate = MIN_STAT_SAMPLING_RATE;
-
-                       dbs_tuners_ins.sampling_rate = def_sampling_rate;
-
-                       dbs_timer_init();
+                       rc = sysfs_create_group(cpufreq_global_kobject,
+                                               &dbs_attr_group);
+                       if (rc) {
+                               mutex_unlock(&dbs_mutex);
+                               return rc;
+                       }
+
+                       /*
+                        * conservative does not implement micro like ondemand
+                        * governor, thus we are bound to jiffes/HZ
+                        */
+                       min_sampling_rate =
+                               MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10);
+                       /* Bring kernel and HW constraints together */
+                       min_sampling_rate = max(min_sampling_rate,
+                                       MIN_LATENCY_MULTIPLIER * latency);
+                       dbs_tuners_ins.sampling_rate =
+                               max(min_sampling_rate,
+                                   latency * LATENCY_MULTIPLIER);
+
+                       cpufreq_register_notifier(
+                                       &dbs_cpufreq_notifier_block,
+                                       CPUFREQ_TRANSITION_NOTIFIER);
                }
-               
                mutex_unlock(&dbs_mutex);
+
+               dbs_timer_init(this_dbs_info);
+
                break;
 
        case CPUFREQ_GOV_STOP:
+               dbs_timer_exit(this_dbs_info);
+
                mutex_lock(&dbs_mutex);
-               this_dbs_info->enable = 0;
-               sysfs_remove_group(&policy->kobj, &dbs_attr_group);
                dbs_enable--;
+               mutex_destroy(&this_dbs_info->timer_mutex);
+
                /*
                 * 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);
+               if (!dbs_enable)
+                       sysfs_remove_group(cpufreq_global_kobject,
+                                          &dbs_attr_group);
 
                break;
 
        case CPUFREQ_GOV_LIMITS:
-               mutex_lock(&dbs_mutex);
+               mutex_lock(&this_dbs_info->timer_mutex);
                if (policy->max < this_dbs_info->cur_policy->cur)
                        __cpufreq_driver_target(
                                        this_dbs_info->cur_policy,
-                                       policy->max, CPUFREQ_RELATION_H);
+                                       policy->max, CPUFREQ_RELATION_H);
                else if (policy->min > this_dbs_info->cur_policy->cur)
                        __cpufreq_driver_target(
                                        this_dbs_info->cur_policy,
-                                       policy->min, CPUFREQ_RELATION_L);
-               mutex_unlock(&dbs_mutex);
+                                       policy->min, CPUFREQ_RELATION_L);
+               mutex_unlock(&this_dbs_info->timer_mutex);
+
                break;
        }
        return 0;
 }
 
-static struct cpufreq_governor cpufreq_gov_dbs = {
-       .name           = "conservative",
-       .governor       = cpufreq_governor_dbs,
-       .owner          = THIS_MODULE,
+#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
+static
+#endif
+struct cpufreq_governor cpufreq_gov_conservative = {
+       .name                   = "conservative",
+       .governor               = cpufreq_governor_dbs,
+       .max_transition_latency = TRANSITION_LATENCY_LIMIT,
+       .owner                  = THIS_MODULE,
 };
 
 static int __init cpufreq_gov_dbs_init(void)
 {
-       return cpufreq_register_governor(&cpufreq_gov_dbs);
+       return cpufreq_register_governor(&cpufreq_gov_conservative);
 }
 
 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_dbs);
+       cpufreq_unregister_governor(&cpufreq_gov_conservative);
 }
 
 
-MODULE_AUTHOR ("Alexander Clouter <alex-kernel@digriz.org.uk>");
-MODULE_DESCRIPTION ("'cpufreq_conservative' - A dynamic cpufreq governor for "
+MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>");
+MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for "
                "Low Latency Frequency Transition capable processors "
                "optimised for use in a battery environment");
-MODULE_LICENSE ("GPL");
+MODULE_LICENSE("GPL");
 
+#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
+fs_initcall(cpufreq_gov_dbs_init);
+#else
 module_init(cpufreq_gov_dbs_init);
+#endif
 module_exit(cpufreq_gov_dbs_exit);