cpufreq: introduce hotplug governor
author黄涛 <huangtao@rock-chips.com>
Wed, 22 Feb 2012 06:54:22 +0000 (14:54 +0800)
committer黄涛 <huangtao@rock-chips.com>
Wed, 22 Feb 2012 06:54:22 +0000 (14:54 +0800)
drivers/cpufreq/cpufreq_hotplug.c [new file with mode: 0644]

diff --git a/drivers/cpufreq/cpufreq_hotplug.c b/drivers/cpufreq/cpufreq_hotplug.c
new file mode 100644 (file)
index 0000000..4a1479d
--- /dev/null
@@ -0,0 +1,744 @@
+/*
+ * CPUFreq hotplug governor
+ *
+ * Copyright (C) 2010 Texas Instruments, Inc.
+ *   Mike Turquette <mturquette@ti.com>
+ *   Santosh Shilimkar <santosh.shilimkar@ti.com>
+ *
+ * Based on ondemand governor
+ * Copyright (C)  2001 Russell King
+ *           (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>,
+ *                     Jun Nakajima <jun.nakajima@intel.com>
+ *
+ * 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
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/cpufreq.h>
+#include <linux/cpu.h>
+#include <linux/jiffies.h>
+#include <linux/kernel_stat.h>
+#include <linux/mutex.h>
+#include <linux/hrtimer.h>
+#include <linux/tick.h>
+#include <linux/ktime.h>
+#include <linux/sched.h>
+#include <linux/err.h>
+#include <linux/slab.h>
+
+/* greater than 80% avg load across online CPUs increases frequency */
+#define DEFAULT_UP_FREQ_MIN_LOAD                       (80)
+
+/* Keep 10% of idle under the up threshold when decreasing the frequency */
+#define DEFAULT_FREQ_DOWN_DIFFERENTIAL                 (10)
+
+/* less than 35% avg load across online CPUs decreases frequency */
+#define DEFAULT_DOWN_FREQ_MAX_LOAD                     (35)
+
+/* default sampling period (uSec) is bogus; 10x ondemand's default for x86 */
+#define DEFAULT_SAMPLING_PERIOD                                (100000)
+
+/* default number of sampling periods to average before hotplug-in decision */
+#define DEFAULT_HOTPLUG_IN_SAMPLING_PERIODS            (5)
+
+/* default number of sampling periods to average before hotplug-out decision */
+#define DEFAULT_HOTPLUG_OUT_SAMPLING_PERIODS           (20)
+
+static void do_dbs_timer(struct work_struct *work);
+static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
+               unsigned int event);
+
+#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_HOTPLUG
+static
+#endif
+struct cpufreq_governor cpufreq_gov_hotplug = {
+       .name                   = "hotplug",
+       .governor               = cpufreq_governor_dbs,
+       .owner                  = THIS_MODULE,
+};
+
+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;
+       struct delayed_work work;
+       struct cpufreq_frequency_table *freq_table;
+       int cpu;
+       /*
+        * 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, hp_cpu_dbs_info);
+
+static unsigned int dbs_enable;        /* number of CPUs using this policy */
+
+/*
+ * dbs_mutex protects data in dbs_tuners_ins from concurrent changes on
+ * different CPUs. It protects dbs_enable in governor start/stop.
+ */
+static DEFINE_MUTEX(dbs_mutex);
+
+static struct workqueue_struct *khotplug_wq;
+
+static struct dbs_tuners {
+       unsigned int sampling_rate;
+       unsigned int up_threshold;
+       unsigned int down_differential;
+       unsigned int down_threshold;
+       unsigned int hotplug_in_sampling_periods;
+       unsigned int hotplug_out_sampling_periods;
+       unsigned int hotplug_load_index;
+       unsigned int *hotplug_load_history;
+       unsigned int ignore_nice;
+       unsigned int io_is_busy;
+} dbs_tuners_ins = {
+       .sampling_rate =                DEFAULT_SAMPLING_PERIOD,
+       .up_threshold =                 DEFAULT_UP_FREQ_MIN_LOAD,
+       .down_differential =            DEFAULT_FREQ_DOWN_DIFFERENTIAL,
+       .down_threshold =               DEFAULT_DOWN_FREQ_MAX_LOAD,
+       .hotplug_in_sampling_periods =  DEFAULT_HOTPLUG_IN_SAMPLING_PERIODS,
+       .hotplug_out_sampling_periods = DEFAULT_HOTPLUG_OUT_SAMPLING_PERIODS,
+       .hotplug_load_index =           0,
+       .ignore_nice =                  0,
+       .io_is_busy =                   0,
+};
+
+/*
+ * A corner case exists when switching io_is_busy at run-time: comparing idle
+ * times from a non-io_is_busy period to an io_is_busy period (or vice-versa)
+ * will misrepresent the actual change in system idleness.  We ignore this
+ * corner case: enabling io_is_busy might cause freq increase and disabling
+ * might cause freq decrease, which probably matches the original intent.
+ */
+static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
+{
+        u64 idle_time;
+        u64 iowait_time;
+
+        /* cpufreq-hotplug always assumes CONFIG_NO_HZ */
+        idle_time = get_cpu_idle_time_us(cpu, wall);
+
+       /* add time spent doing I/O to idle time */
+        if (dbs_tuners_ins.io_is_busy) {
+                iowait_time = get_cpu_iowait_time_us(cpu, wall);
+                /* cpufreq-hotplug always assumes CONFIG_NO_HZ */
+                if (iowait_time != -1ULL && idle_time >= iowait_time)
+                        idle_time -= iowait_time;
+        }
+
+        return idle_time;
+}
+
+/************************** sysfs interface ************************/
+
+/* XXX look at global sysfs macros in cpufreq.h, can those be used here? */
+
+/* cpufreq_hotplug Governor Tunables */
+#define show_one(file_name, object)                                    \
+static ssize_t show_##file_name                                                \
+(struct kobject *kobj, struct attribute *attr, char *buf)              \
+{                                                                      \
+       return sprintf(buf, "%u\n", dbs_tuners_ins.object);             \
+}
+show_one(sampling_rate, sampling_rate);
+show_one(up_threshold, up_threshold);
+show_one(down_differential, down_differential);
+show_one(down_threshold, down_threshold);
+show_one(hotplug_in_sampling_periods, hotplug_in_sampling_periods);
+show_one(hotplug_out_sampling_periods, hotplug_out_sampling_periods);
+show_one(ignore_nice_load, ignore_nice);
+show_one(io_is_busy, io_is_busy);
+
+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);
+       if (ret != 1)
+               return -EINVAL;
+
+       mutex_lock(&dbs_mutex);
+       dbs_tuners_ins.sampling_rate = input;
+       mutex_unlock(&dbs_mutex);
+
+       return 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);
+
+       if (ret != 1 || input <= dbs_tuners_ins.down_threshold) {
+               return -EINVAL;
+       }
+
+       mutex_lock(&dbs_mutex);
+       dbs_tuners_ins.up_threshold = input;
+       mutex_unlock(&dbs_mutex);
+
+       return count;
+}
+
+static ssize_t store_down_differential(struct kobject *a, struct attribute *b,
+                                 const char *buf, size_t count)
+{
+       unsigned int input;
+       int ret;
+       ret = sscanf(buf, "%u", &input);
+
+       if (ret != 1 || input >= dbs_tuners_ins.up_threshold)
+               return -EINVAL;
+
+       mutex_lock(&dbs_mutex);
+       dbs_tuners_ins.down_differential = input;
+       mutex_unlock(&dbs_mutex);
+
+       return 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);
+
+       if (ret != 1 || input >= dbs_tuners_ins.up_threshold) {
+               return -EINVAL;
+       }
+
+       mutex_lock(&dbs_mutex);
+       dbs_tuners_ins.down_threshold = input;
+       mutex_unlock(&dbs_mutex);
+
+       return count;
+}
+
+static ssize_t store_hotplug_in_sampling_periods(struct kobject *a,
+               struct attribute *b, const char *buf, size_t count)
+{
+       unsigned int input;
+       unsigned int *temp;
+       unsigned int max_windows;
+       int ret;
+       ret = sscanf(buf, "%u", &input);
+
+       if (ret != 1)
+               return -EINVAL;
+
+       /* already using this value, bail out */
+       if (input == dbs_tuners_ins.hotplug_in_sampling_periods)
+               return count;
+
+       mutex_lock(&dbs_mutex);
+       ret = count;
+       max_windows = max(dbs_tuners_ins.hotplug_in_sampling_periods,
+                       dbs_tuners_ins.hotplug_out_sampling_periods);
+
+       /* no need to resize array */
+       if (input <= max_windows) {
+               dbs_tuners_ins.hotplug_in_sampling_periods = input;
+               goto out;
+       }
+
+       /* resize array */
+       temp = kmalloc((sizeof(unsigned int) * input), GFP_KERNEL);
+
+       if (!temp || IS_ERR(temp)) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       memcpy(temp, dbs_tuners_ins.hotplug_load_history,
+                       (max_windows * sizeof(unsigned int)));
+       kfree(dbs_tuners_ins.hotplug_load_history);
+
+       /* replace old buffer, old number of sampling periods & old index */
+       dbs_tuners_ins.hotplug_load_history = temp;
+       dbs_tuners_ins.hotplug_in_sampling_periods = input;
+       dbs_tuners_ins.hotplug_load_index = max_windows;
+out:
+       mutex_unlock(&dbs_mutex);
+
+       return ret;
+}
+
+static ssize_t store_hotplug_out_sampling_periods(struct kobject *a,
+               struct attribute *b, const char *buf, size_t count)
+{
+       unsigned int input;
+       unsigned int *temp;
+       unsigned int max_windows;
+       int ret;
+       ret = sscanf(buf, "%u", &input);
+
+       if (ret != 1)
+               return -EINVAL;
+
+       /* already using this value, bail out */
+       if (input == dbs_tuners_ins.hotplug_out_sampling_periods)
+               return count;
+
+       mutex_lock(&dbs_mutex);
+       ret = count;
+       max_windows = max(dbs_tuners_ins.hotplug_in_sampling_periods,
+                       dbs_tuners_ins.hotplug_out_sampling_periods);
+
+       /* no need to resize array */
+       if (input <= max_windows) {
+               dbs_tuners_ins.hotplug_out_sampling_periods = input;
+               goto out;
+       }
+
+       /* resize array */
+       temp = kmalloc((sizeof(unsigned int) * input), GFP_KERNEL);
+
+       if (!temp || IS_ERR(temp)) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       memcpy(temp, dbs_tuners_ins.hotplug_load_history,
+                       (max_windows * sizeof(unsigned int)));
+       kfree(dbs_tuners_ins.hotplug_load_history);
+
+       /* replace old buffer, old number of sampling periods & old index */
+       dbs_tuners_ins.hotplug_load_history = temp;
+       dbs_tuners_ins.hotplug_out_sampling_periods = input;
+       dbs_tuners_ins.hotplug_load_index = max_windows;
+out:
+       mutex_unlock(&dbs_mutex);
+
+       return ret;
+}
+
+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)
+               return -EINVAL;
+
+       if (input > 1)
+               input = 1;
+
+       mutex_lock(&dbs_mutex);
+       if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
+               mutex_unlock(&dbs_mutex);
+               return count;
+       }
+       dbs_tuners_ins.ignore_nice = input;
+
+       /* we need to re-evaluate prev_cpu_idle */
+       for_each_online_cpu(j) {
+               struct cpu_dbs_info_s *dbs_info;
+               dbs_info = &per_cpu(hp_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_io_is_busy(struct kobject *a, struct attribute *b,
+                                  const char *buf, size_t count)
+{
+       unsigned int input;
+       int ret;
+
+       ret = sscanf(buf, "%u", &input);
+       if (ret != 1)
+               return -EINVAL;
+
+       mutex_lock(&dbs_mutex);
+       dbs_tuners_ins.io_is_busy = !!input;
+       mutex_unlock(&dbs_mutex);
+
+       return count;
+}
+
+define_one_global_rw(sampling_rate);
+define_one_global_rw(up_threshold);
+define_one_global_rw(down_differential);
+define_one_global_rw(down_threshold);
+define_one_global_rw(hotplug_in_sampling_periods);
+define_one_global_rw(hotplug_out_sampling_periods);
+define_one_global_rw(ignore_nice_load);
+define_one_global_rw(io_is_busy);
+
+static struct attribute *dbs_attributes[] = {
+       &sampling_rate.attr,
+       &up_threshold.attr,
+       &down_differential.attr,
+       &down_threshold.attr,
+       &hotplug_in_sampling_periods.attr,
+       &hotplug_out_sampling_periods.attr,
+       &ignore_nice_load.attr,
+       &io_is_busy.attr,
+       NULL
+};
+
+static struct attribute_group dbs_attr_group = {
+       .attrs = dbs_attributes,
+       .name = "hotplug",
+};
+
+/************************** sysfs end ************************/
+
+static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
+{
+       /* combined load of all enabled CPUs */
+       unsigned int total_load = 0;
+       /* single largest CPU load percentage*/
+       unsigned int max_load = 0;
+       /* largest CPU load in terms of frequency */
+       unsigned int max_load_freq = 0;
+       /* average load across all enabled CPUs */
+       unsigned int avg_load = 0;
+       /* average load across multiple sampling periods for hotplug events */
+       unsigned int hotplug_in_avg_load = 0;
+       unsigned int hotplug_out_avg_load = 0;
+       /* number of sampling periods averaged for hotplug decisions */
+       unsigned int periods;
+
+       struct cpufreq_policy *policy;
+       unsigned int i, j;
+
+       policy = this_dbs_info->cur_policy;
+
+       /*
+        * cpu load accounting
+        * get highest load, total load and average load across all CPUs
+        */
+       for_each_cpu(j, policy->cpus) {
+               unsigned int load;
+               unsigned int idle_time, wall_time;
+               cputime64_t cur_wall_time, cur_idle_time;
+               struct cpu_dbs_info_s *j_dbs_info;
+
+               j_dbs_info = &per_cpu(hp_cpu_dbs_info, j);
+
+               /* update both cur_idle_time and cur_wall_time */
+               cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);
+
+               /* how much wall time has passed since last iteration? */
+               wall_time = (unsigned int) cputime64_sub(cur_wall_time,
+                               j_dbs_info->prev_cpu_wall);
+               j_dbs_info->prev_cpu_wall = cur_wall_time;
+
+               /* how much idle time has passed since last iteration? */
+               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 (unlikely(!wall_time || wall_time < idle_time))
+                       continue;
+
+               /* load is the percentage of time not spent in idle */
+               load = 100 * (wall_time - idle_time) / wall_time;
+
+               /* keep track of combined load across all CPUs */
+               total_load += load;
+
+               /* keep track of highest single load across all CPUs */
+               if (load > max_load)
+                       max_load = load;
+       }
+
+       /* use the max load in the OPP freq change policy */
+       max_load_freq = max_load * policy->cur;
+
+       /* calculate the average load across all related CPUs */
+       avg_load = total_load / num_online_cpus();
+
+
+       /*
+        * hotplug load accounting
+        * average load over multiple sampling periods
+        */
+
+       /* how many sampling periods do we use for hotplug decisions? */
+       periods = max(dbs_tuners_ins.hotplug_in_sampling_periods,
+                       dbs_tuners_ins.hotplug_out_sampling_periods);
+
+       /* store avg_load in the circular buffer */
+       dbs_tuners_ins.hotplug_load_history[dbs_tuners_ins.hotplug_load_index]
+               = avg_load;
+
+       /* compute average load across in & out sampling periods */
+       for (i = 0, j = dbs_tuners_ins.hotplug_load_index;
+                       i < periods; i++, j--) {
+               if (i < dbs_tuners_ins.hotplug_in_sampling_periods)
+                       hotplug_in_avg_load +=
+                               dbs_tuners_ins.hotplug_load_history[j];
+               if (i < dbs_tuners_ins.hotplug_out_sampling_periods)
+                       hotplug_out_avg_load +=
+                               dbs_tuners_ins.hotplug_load_history[j];
+
+               if (j == 0)
+                       j = periods;
+       }
+
+       hotplug_in_avg_load = hotplug_in_avg_load /
+               dbs_tuners_ins.hotplug_in_sampling_periods;
+
+       hotplug_out_avg_load = hotplug_out_avg_load /
+               dbs_tuners_ins.hotplug_out_sampling_periods;
+
+       /* return to first element if we're at the circular buffer's end */
+       if (++dbs_tuners_ins.hotplug_load_index == periods)
+               dbs_tuners_ins.hotplug_load_index = 0;
+
+       /* check if auxiliary CPU is needed based on avg_load */
+       if (avg_load > dbs_tuners_ins.up_threshold) {
+               /* should we enable auxillary CPUs? */
+               if (num_online_cpus() < 2 && hotplug_in_avg_load >
+                               dbs_tuners_ins.up_threshold) {
+                       /* hotplug with cpufreq is nasty
+                        * a call to cpufreq_governor_dbs may cause a lockup.
+                        * wq is not running here so its safe.
+                        */
+                       mutex_unlock(&this_dbs_info->timer_mutex);
+                       cpu_up(1);
+                       mutex_lock(&this_dbs_info->timer_mutex);
+                       goto out;
+               }
+       }
+
+       /* check for frequency increase based on max_load */
+       if (max_load > dbs_tuners_ins.up_threshold) {
+               /* increase to highest frequency supported */
+               if (policy->cur < policy->max)
+                       __cpufreq_driver_target(policy, policy->max,
+                                       CPUFREQ_RELATION_H);
+
+               goto out;
+       }
+
+       /* check for frequency decrease */
+       if (avg_load < dbs_tuners_ins.down_threshold) {
+               /* are we at the minimum frequency already? */
+               if (policy->cur == policy->min) {
+                       /* should we disable auxillary CPUs? */
+                       if (num_online_cpus() > 1 && hotplug_out_avg_load <
+                                       dbs_tuners_ins.down_threshold) {
+                               mutex_unlock(&this_dbs_info->timer_mutex);
+                               cpu_down(1);
+                               mutex_lock(&this_dbs_info->timer_mutex);
+                       }
+                       goto out;
+               }
+       }
+
+       /*
+        * go down to the lowest frequency which can sustain the load by
+        * keeping 30% of idle in order to not cross the up_threshold
+        */
+       if ((max_load_freq <
+           (dbs_tuners_ins.up_threshold - dbs_tuners_ins.down_differential) *
+            policy->cur) && (policy->cur > policy->min)) {
+               unsigned int freq_next;
+               freq_next = max_load_freq /
+                               (dbs_tuners_ins.up_threshold -
+                                dbs_tuners_ins.down_differential);
+
+               if (freq_next < policy->min)
+                       freq_next = policy->min;
+
+                __cpufreq_driver_target(policy, freq_next,
+                                        CPUFREQ_RELATION_L);
+       }
+out:
+       return;
+}
+
+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 related CPUs to do sampling nearly on same jiffy */
+       int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
+
+       mutex_lock(&dbs_info->timer_mutex);
+       dbs_check_cpu(dbs_info);
+       queue_delayed_work_on(cpu, khotplug_wq, &dbs_info->work, delay);
+       mutex_unlock(&dbs_info->timer_mutex);
+}
+
+static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
+{
+       /* We want all related CPUs to do sampling nearly on same jiffy */
+       int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
+       delay -= jiffies % delay;
+
+       INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
+       queue_delayed_work_on(dbs_info->cpu, khotplug_wq, &dbs_info->work,
+               delay);
+}
+
+static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
+{
+       cancel_delayed_work_sync(&dbs_info->work);
+}
+
+static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
+                                  unsigned int event)
+{
+       unsigned int cpu = policy->cpu;
+       struct cpu_dbs_info_s *this_dbs_info;
+       unsigned int i, j, max_periods;
+       int rc;
+
+       this_dbs_info = &per_cpu(hp_cpu_dbs_info, cpu);
+
+       switch (event) {
+       case CPUFREQ_GOV_START:
+               if ((!cpu_online(cpu)) || (!policy->cur))
+                       return -EINVAL;
+
+               mutex_lock(&dbs_mutex);
+               dbs_enable++;
+               for_each_cpu(j, policy->cpus) {
+                       struct cpu_dbs_info_s *j_dbs_info;
+                       j_dbs_info = &per_cpu(hp_cpu_dbs_info, j);
+                       j_dbs_info->cur_policy = policy;
+
+                       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;
+                       }
+
+                       max_periods = max(DEFAULT_HOTPLUG_IN_SAMPLING_PERIODS,
+                                       DEFAULT_HOTPLUG_OUT_SAMPLING_PERIODS);
+                       dbs_tuners_ins.hotplug_load_history = kmalloc(
+                                       (sizeof(unsigned int) * max_periods),
+                                       GFP_KERNEL);
+                       if (!dbs_tuners_ins.hotplug_load_history) {
+                               WARN_ON(1);
+                               return -ENOMEM;
+                       }
+                       for (i = 0; i < max_periods; i++)
+                               dbs_tuners_ins.hotplug_load_history[i] = 50;
+               }
+               this_dbs_info->cpu = cpu;
+               this_dbs_info->freq_table = cpufreq_frequency_get_table(cpu);
+               /*
+                * Start the timerschedule work, when this governor
+                * is used for first time
+                */
+               if (dbs_enable == 1) {
+                       rc = sysfs_create_group(cpufreq_global_kobject,
+                                               &dbs_attr_group);
+                       if (rc) {
+                               mutex_unlock(&dbs_mutex);
+                               return rc;
+                       }
+               }
+               mutex_unlock(&dbs_mutex);
+
+               mutex_init(&this_dbs_info->timer_mutex);
+               dbs_timer_init(this_dbs_info);
+               break;
+
+       case CPUFREQ_GOV_STOP:
+               dbs_timer_exit(this_dbs_info);
+
+               mutex_lock(&dbs_mutex);
+               mutex_destroy(&this_dbs_info->timer_mutex);
+               dbs_enable--;
+               mutex_unlock(&dbs_mutex);
+               if (!dbs_enable)
+                       sysfs_remove_group(cpufreq_global_kobject,
+                                          &dbs_attr_group);
+               kfree(dbs_tuners_ins.hotplug_load_history);
+               /*
+                * XXX BIG CAVEAT: Stopping the governor with CPU1 offline
+                * will result in it remaining offline until the user onlines
+                * it again.  It is up to the user to do this (for now).
+                */
+               break;
+
+       case CPUFREQ_GOV_LIMITS:
+               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);
+               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(&this_dbs_info->timer_mutex);
+               break;
+       }
+       return 0;
+}
+
+static int __init cpufreq_gov_dbs_init(void)
+{
+       int err;
+       cputime64_t wall;
+       u64 idle_time;
+       int cpu = get_cpu();
+
+       idle_time = get_cpu_idle_time_us(cpu, &wall);
+       put_cpu();
+       if (idle_time != -1ULL) {
+               dbs_tuners_ins.up_threshold = DEFAULT_UP_FREQ_MIN_LOAD;
+       } else {
+               pr_err("cpufreq-hotplug: %s: assumes CONFIG_NO_HZ\n",
+                               __func__);
+               return -EINVAL;
+       }
+
+       khotplug_wq = create_workqueue("khotplug");
+       if (!khotplug_wq) {
+               pr_err("Creation of khotplug failed\n");
+               return -EFAULT;
+       }
+       err = cpufreq_register_governor(&cpufreq_gov_hotplug);
+       if (err)
+               destroy_workqueue(khotplug_wq);
+
+       return err;
+}
+
+static void __exit cpufreq_gov_dbs_exit(void)
+{
+       cpufreq_unregister_governor(&cpufreq_gov_hotplug);
+       destroy_workqueue(khotplug_wq);
+}
+
+MODULE_AUTHOR("Mike Turquette <mturquette@ti.com>");
+MODULE_DESCRIPTION("'cpufreq_hotplug' - cpufreq governor for dynamic frequency scaling and CPU hotplugging");
+MODULE_LICENSE("GPL");
+
+#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_HOTPLUG
+fs_initcall(cpufreq_gov_dbs_init);
+#else
+module_init(cpufreq_gov_dbs_init);
+#endif
+module_exit(cpufreq_gov_dbs_exit);