]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - drivers/cpufreq/cpufreq_ondemand.c
[CPUFREQ] ondemand,conservative governor store the idle ticks for all cpus
[mv-sheeva.git] / drivers / cpufreq / cpufreq_ondemand.c
index 8d83a21c6477f3e7a3cd223a9bffa4e8b1af1710..f239545ac1b84df92b4e0fa04d5078726d68ea08 100644 (file)
@@ -57,7 +57,6 @@ static unsigned int                           def_sampling_rate;
 #define DEF_SAMPLING_RATE_LATENCY_MULTIPLIER   (1000)
 #define DEF_SAMPLING_DOWN_FACTOR               (10)
 #define TRANSITION_LATENCY_LIMIT               (10 * 1000)
-#define sampling_rate_in_HZ(x)                 (((x * HZ) < (1000 * 1000))?1:((x * HZ) / (1000 * 1000)))
 
 static void do_dbs_timer(void *data);
 
@@ -79,6 +78,8 @@ struct dbs_tuners {
        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 = {
@@ -87,6 +88,15 @@ static struct dbs_tuners dbs_tuners_ins = {
        .sampling_down_factor   = DEF_SAMPLING_DOWN_FACTOR,
 };
 
+static inline unsigned int get_cpu_idle_time(unsigned int cpu)
+{
+       return  kstat_cpu(cpu).cpustat.idle +
+               kstat_cpu(cpu).cpustat.iowait +
+               ( !dbs_tuners_ins.ignore_nice ? 
+                 kstat_cpu(cpu).cpustat.nice :
+                 0);
+}
+
 /************************** sysfs interface ************************/
 static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf)
 {
@@ -116,6 +126,8 @@ show_one(sampling_rate, sampling_rate);
 show_one(sampling_down_factor, sampling_down_factor);
 show_one(up_threshold, up_threshold);
 show_one(down_threshold, down_threshold);
+show_one(ignore_nice, 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)
@@ -194,6 +206,63 @@ static ssize_t store_down_threshold(struct cpufreq_policy *unused,
        return count;
 }
 
+static ssize_t store_ignore_nice(struct cpufreq_policy *policy,
+               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;
+       
+       down(&dbs_sem);
+       if ( input == dbs_tuners_ins.ignore_nice ) { /* nothing to do */
+               up(&dbs_sem);
+               return count;
+       }
+       dbs_tuners_ins.ignore_nice = input;
+
+       /* we need to re-evaluate prev_cpu_idle_up and prev_cpu_idle_down */
+       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;
+       }
+       up(&dbs_sem);
+
+       return count;
+}
+
+static ssize_t store_freq_step(struct cpufreq_policy *policy,
+               const char *buf, size_t count)
+{
+       unsigned int input;
+       int ret;
+
+       ret = sscanf (buf, "%u", &input);
+
+       if ( ret != 1 )
+               return -EINVAL;
+
+       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 :) */
+       down(&dbs_sem);
+       dbs_tuners_ins.freq_step = input;
+       up(&dbs_sem);
+
+       return count;
+}
+
 #define define_one_rw(_name) \
 static struct freq_attr _name = \
 __ATTR(_name, 0644, show_##_name, store_##_name)
@@ -202,6 +271,8 @@ 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);
+define_one_rw(freq_step);
 
 static struct attribute * dbs_attributes[] = {
        &sampling_rate_max.attr,
@@ -210,6 +281,8 @@ static struct attribute * dbs_attributes[] = {
        &sampling_down_factor.attr,
        &up_threshold.attr,
        &down_threshold.attr,
+       &ignore_nice.attr,
+       &freq_step.attr,
        NULL
 };
 
@@ -248,16 +321,14 @@ static void dbs_check_cpu(int cpu)
         *
         * Any frequency increase takes it to the maximum frequency. 
         * Frequency reduction happens at minimum steps of 
-        * 5% of max_frequency 
+        * 5% (default) of max_frequency 
         */
 
        /* Check for frequency increase */
-       total_idle_ticks = kstat_cpu(cpu).cpustat.idle +
-               kstat_cpu(cpu).cpustat.iowait;
+       total_idle_ticks = get_cpu_idle_time(cpu);
        idle_ticks = total_idle_ticks -
                this_dbs_info->prev_cpu_idle_up;
        this_dbs_info->prev_cpu_idle_up = total_idle_ticks;
-       
 
        for_each_cpu_mask(j, policy->cpus) {
                unsigned int tmp_idle_ticks;
@@ -268,8 +339,7 @@ static void dbs_check_cpu(int cpu)
 
                j_dbs_info = &per_cpu(cpu_dbs_info, j);
                /* Check for frequency increase */
-               total_idle_ticks = kstat_cpu(j).cpustat.idle +
-                       kstat_cpu(j).cpustat.iowait;
+               total_idle_ticks = get_cpu_idle_time(j);
                tmp_idle_ticks = total_idle_ticks -
                        j_dbs_info->prev_cpu_idle_up;
                j_dbs_info->prev_cpu_idle_up = total_idle_ticks;
@@ -281,13 +351,23 @@ static void dbs_check_cpu(int cpu)
        /* Scale idle ticks by 100 and compare with up and down ticks */
        idle_ticks *= 100;
        up_idle_ticks = (100 - dbs_tuners_ins.up_threshold) *
-                       sampling_rate_in_HZ(dbs_tuners_ins.sampling_rate);
+                       usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
 
        if (idle_ticks < up_idle_ticks) {
+               down_skip[cpu] = 0;
+               for_each_cpu_mask(j, policy->cpus) {
+                       struct cpu_dbs_info_s *j_dbs_info;
+
+                       j_dbs_info = &per_cpu(cpu_dbs_info, j);
+                       j_dbs_info->prev_cpu_idle_down = 
+                                       j_dbs_info->prev_cpu_idle_up;
+               }
+               /* if we are already at full speed then break out early */
+               if (policy->cur == policy->max)
+                       return;
+               
                __cpufreq_driver_target(policy, policy->max, 
                        CPUFREQ_RELATION_H);
-               down_skip[cpu] = 0;
-               this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
                return;
        }
 
@@ -296,8 +376,7 @@ static void dbs_check_cpu(int cpu)
        if (down_skip[cpu] < dbs_tuners_ins.sampling_down_factor)
                return;
 
-       total_idle_ticks = kstat_cpu(cpu).cpustat.idle +
-               kstat_cpu(cpu).cpustat.iowait;
+       total_idle_ticks = this_dbs_info->prev_cpu_idle_up;
        idle_ticks = total_idle_ticks -
                this_dbs_info->prev_cpu_idle_down;
        this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
@@ -310,9 +389,8 @@ static void dbs_check_cpu(int cpu)
                        continue;
 
                j_dbs_info = &per_cpu(cpu_dbs_info, j);
-               /* Check for frequency increase */
-               total_idle_ticks = kstat_cpu(j).cpustat.idle +
-                       kstat_cpu(j).cpustat.iowait;
+               /* Check for frequency decrease */
+               total_idle_ticks = j_dbs_info->prev_cpu_idle_up;
                tmp_idle_ticks = total_idle_ticks -
                        j_dbs_info->prev_cpu_idle_down;
                j_dbs_info->prev_cpu_idle_down = total_idle_ticks;
@@ -328,17 +406,23 @@ static void dbs_check_cpu(int cpu)
        freq_down_sampling_rate = dbs_tuners_ins.sampling_rate *
                dbs_tuners_ins.sampling_down_factor;
        down_idle_ticks = (100 - dbs_tuners_ins.down_threshold) *
-                       sampling_rate_in_HZ(freq_down_sampling_rate);
+               usecs_to_jiffies(freq_down_sampling_rate);
 
        if (idle_ticks > down_idle_ticks ) {
-               freq_down_step = (5 * policy->max) / 100;
+               /* 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 (policy->cur == policy->min || dbs_tuners_ins.freq_step == 0)
+                       return;
+
+               freq_down_step = (dbs_tuners_ins.freq_step * policy->max) / 100;
 
                /* max freq cannot be less than 100. But who knows.... */
                if (unlikely(freq_down_step == 0))
                        freq_down_step = 5;
 
                __cpufreq_driver_target(policy,
-                       policy->cur - freq_down_step, 
+                       policy->cur - freq_down_step,
                        CPUFREQ_RELATION_H);
                return;
        }
@@ -348,11 +432,10 @@ static void do_dbs_timer(void *data)
 { 
        int i;
        down(&dbs_sem);
-       for (i = 0; i < NR_CPUS; i++)
-               if (cpu_online(i))
-                       dbs_check_cpu(i);
+       for_each_online_cpu(i)
+               dbs_check_cpu(i);
        schedule_delayed_work(&dbs_work, 
-                       sampling_rate_in_HZ(dbs_tuners_ins.sampling_rate));
+                       usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
        up(&dbs_sem);
 } 
 
@@ -360,7 +443,7 @@ static inline void dbs_timer_init(void)
 {
        INIT_WORK(&dbs_work, do_dbs_timer, NULL);
        schedule_delayed_work(&dbs_work,
-                       sampling_rate_in_HZ(dbs_tuners_ins.sampling_rate));
+                       usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
        return;
 }
 
@@ -397,12 +480,9 @@ 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 = 
-                               kstat_cpu(j).cpustat.idle +
-                               kstat_cpu(j).cpustat.iowait;
-                       j_dbs_info->prev_cpu_idle_down = 
-                               kstat_cpu(j).cpustat.idle +
-                               kstat_cpu(j).cpustat.iowait;
+                       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;
                }
                this_dbs_info->enable = 1;
                sysfs_create_group(&policy->kobj, &dbs_attr_group);
@@ -422,6 +502,8 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
                        def_sampling_rate = (latency / 1000) *
                                        DEF_SAMPLING_RATE_LATENCY_MULTIPLIER;
                        dbs_tuners_ins.sampling_rate = def_sampling_rate;
+                       dbs_tuners_ins.ignore_nice = 0;
+                       dbs_tuners_ins.freq_step = 5;
 
                        dbs_timer_init();
                }
@@ -461,12 +543,11 @@ static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
        return 0;
 }
 
-struct cpufreq_governor cpufreq_gov_dbs = {
+static struct cpufreq_governor cpufreq_gov_dbs = {
        .name           = "ondemand",
        .governor       = cpufreq_governor_dbs,
        .owner          = THIS_MODULE,
 };
-EXPORT_SYMBOL(cpufreq_gov_dbs);
 
 static int __init cpufreq_gov_dbs_init(void)
 {