kernel-ark/drivers/cpufreq/cpufreq_ondemand.c

611 lines
17 KiB
C
Raw Normal View History

/*
* drivers/cpufreq/cpufreq_ondemand.c
*
* 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>
/*
* 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 MIN_FREQUENCY_UP_THRESHOLD (11)
#define MAX_FREQUENCY_UP_THRESHOLD (100)
/*
* 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
* 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)
#define TRANSITION_LATENCY_LIMIT (10 * 1000)
static void do_dbs_timer(struct work_struct *work);
/* Sampling types */
enum dbs_sample {DBS_NORMAL_SAMPLE, DBS_SUB_SAMPLE};
struct cpu_dbs_info_s {
cputime64_t prev_cpu_idle;
cputime64_t prev_cpu_wall;
struct cpufreq_policy *cur_policy;
struct delayed_work work;
enum dbs_sample sample_type;
unsigned int enable;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
struct cpufreq_frequency_table *freq_table;
unsigned int freq_lo;
unsigned int freq_lo_jiffies;
unsigned int freq_hi_jiffies;
};
static DEFINE_PER_CPU(struct cpu_dbs_info_s, 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
*/
static DEFINE_MUTEX(dbs_mutex);
static struct workqueue_struct *kondemand_wq;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
static struct dbs_tuners {
unsigned int sampling_rate;
unsigned int up_threshold;
unsigned int ignore_nice;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
unsigned int powersave_bias;
} dbs_tuners_ins = {
.up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
.ignore_nice = 0,
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
.powersave_bias = 0,
};
static inline cputime64_t get_cpu_idle_time(unsigned int cpu)
{
cputime64_t retval;
retval = cputime64_add(kstat_cpu(cpu).cpustat.idle,
kstat_cpu(cpu).cpustat.iowait);
if (dbs_tuners_ins.ignore_nice)
retval = cputime64_add(retval, kstat_cpu(cpu).cpustat.nice);
return retval;
}
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
/*
* Find right freq to be set now with powersave_bias on.
* Returns the freq_hi to be used right now and will set freq_hi_jiffies,
* freq_lo, and freq_lo_jiffies in percpu area for averaging freqs.
*/
static unsigned int powersave_bias_target(struct cpufreq_policy *policy,
unsigned int freq_next,
unsigned int relation)
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
{
unsigned int freq_req, freq_reduc, freq_avg;
unsigned int freq_hi, freq_lo;
unsigned int index = 0;
unsigned int jiffies_total, jiffies_hi, jiffies_lo;
struct cpu_dbs_info_s *dbs_info = &per_cpu(cpu_dbs_info, policy->cpu);
if (!dbs_info->freq_table) {
dbs_info->freq_lo = 0;
dbs_info->freq_lo_jiffies = 0;
return freq_next;
}
cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next,
relation, &index);
freq_req = dbs_info->freq_table[index].frequency;
freq_reduc = freq_req * dbs_tuners_ins.powersave_bias / 1000;
freq_avg = freq_req - freq_reduc;
/* Find freq bounds for freq_avg in freq_table */
index = 0;
cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
CPUFREQ_RELATION_H, &index);
freq_lo = dbs_info->freq_table[index].frequency;
index = 0;
cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
CPUFREQ_RELATION_L, &index);
freq_hi = dbs_info->freq_table[index].frequency;
/* Find out how long we have to be in hi and lo freqs */
if (freq_hi == freq_lo) {
dbs_info->freq_lo = 0;
dbs_info->freq_lo_jiffies = 0;
return freq_lo;
}
jiffies_total = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
jiffies_hi = (freq_avg - freq_lo) * jiffies_total;
jiffies_hi += ((freq_hi - freq_lo) / 2);
jiffies_hi /= (freq_hi - freq_lo);
jiffies_lo = jiffies_total - jiffies_hi;
dbs_info->freq_lo = freq_lo;
dbs_info->freq_lo_jiffies = jiffies_lo;
dbs_info->freq_hi_jiffies = jiffies_hi;
return freq_hi;
}
static void ondemand_powersave_bias_init(void)
{
int i;
for_each_online_cpu(i) {
struct cpu_dbs_info_s *dbs_info = &per_cpu(cpu_dbs_info, i);
dbs_info->freq_table = cpufreq_frequency_get_table(i);
dbs_info->freq_lo = 0;
}
}
/************************** sysfs interface ************************/
static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf)
{
return sprintf (buf, "%u\n", MAX_SAMPLING_RATE);
}
static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf)
{
return sprintf (buf, "%u\n", MIN_SAMPLING_RATE);
}
#define define_one_ro(_name) \
static struct freq_attr _name = \
__ATTR(_name, 0444, show_##_name, NULL)
define_one_ro(sampling_rate_max);
define_one_ro(sampling_rate_min);
/* cpufreq_ondemand Governor Tunables */
#define show_one(file_name, object) \
static ssize_t show_##file_name \
(struct cpufreq_policy *unused, 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(ignore_nice_load, ignore_nice);
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
show_one(powersave_bias, powersave_bias);
static ssize_t store_sampling_rate(struct cpufreq_policy *unused,
const char *buf, size_t count)
{
unsigned int input;
int ret;
ret = sscanf(buf, "%u", &input);
mutex_lock(&dbs_mutex);
if (ret != 1 || input > MAX_SAMPLING_RATE
|| input < MIN_SAMPLING_RATE) {
mutex_unlock(&dbs_mutex);
return -EINVAL;
}
dbs_tuners_ins.sampling_rate = input;
mutex_unlock(&dbs_mutex);
return count;
}
static ssize_t store_up_threshold(struct cpufreq_policy *unused,
const char *buf, size_t count)
{
unsigned int input;
int ret;
ret = sscanf(buf, "%u", &input);
mutex_lock(&dbs_mutex);
if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD ||
input < MIN_FREQUENCY_UP_THRESHOLD) {
mutex_unlock(&dbs_mutex);
return -EINVAL;
}
dbs_tuners_ins.up_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)
{
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(cpu_dbs_info, j);
dbs_info->prev_cpu_idle = get_cpu_idle_time(j);
dbs_info->prev_cpu_wall = get_jiffies_64();
}
mutex_unlock(&dbs_mutex);
return count;
}
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
static ssize_t store_powersave_bias(struct cpufreq_policy *unused,
const char *buf, size_t count)
{
unsigned int input;
int ret;
ret = sscanf(buf, "%u", &input);
if (ret != 1)
return -EINVAL;
if (input > 1000)
input = 1000;
mutex_lock(&dbs_mutex);
dbs_tuners_ins.powersave_bias = input;
ondemand_powersave_bias_init();
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(up_threshold);
define_one_rw(ignore_nice_load);
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
define_one_rw(powersave_bias);
static struct attribute * dbs_attributes[] = {
&sampling_rate_max.attr,
&sampling_rate_min.attr,
&sampling_rate.attr,
&up_threshold.attr,
&ignore_nice_load.attr,
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
&powersave_bias.attr,
NULL
};
static struct attribute_group dbs_attr_group = {
.attrs = dbs_attributes,
.name = "ondemand",
};
/************************** sysfs end ************************/
static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
{
unsigned int idle_ticks, total_ticks;
unsigned int load;
cputime64_t cur_jiffies;
struct cpufreq_policy *policy;
unsigned int j;
if (!this_dbs_info->enable)
return;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
this_dbs_info->freq_lo = 0;
policy = this_dbs_info->cur_policy;
cur_jiffies = jiffies64_to_cputime64(get_jiffies_64());
total_ticks = (unsigned int) cputime64_sub(cur_jiffies,
this_dbs_info->prev_cpu_wall);
this_dbs_info->prev_cpu_wall = cur_jiffies;
if (!total_ticks)
return;
/*
* Every sampling_rate, we check, if current idle time is less
* than 20% (default), then we try to increase frequency
* Every sampling_rate, we look for a the lowest
* frequency which can sustain the load while keeping idle time over
* 30%. If such a frequency exist, we try to decrease to this frequency.
*
* Any frequency increase takes it to the maximum frequency.
* Frequency reduction happens at minimum steps of
* 5% (default) of current frequency
*/
/* Get Idle Time */
idle_ticks = UINT_MAX;
for_each_cpu_mask(j, policy->cpus) {
cputime64_t total_idle_ticks;
unsigned int tmp_idle_ticks;
struct cpu_dbs_info_s *j_dbs_info;
j_dbs_info = &per_cpu(cpu_dbs_info, j);
total_idle_ticks = get_cpu_idle_time(j);
tmp_idle_ticks = (unsigned int) cputime64_sub(total_idle_ticks,
j_dbs_info->prev_cpu_idle);
j_dbs_info->prev_cpu_idle = total_idle_ticks;
if (tmp_idle_ticks < idle_ticks)
idle_ticks = tmp_idle_ticks;
}
load = (100 * (total_ticks - idle_ticks)) / total_ticks;
/* Check for frequency increase */
if (load > dbs_tuners_ins.up_threshold) {
/* if we are already at full speed then break out early */
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
if (!dbs_tuners_ins.powersave_bias) {
if (policy->cur == policy->max)
return;
__cpufreq_driver_target(policy, policy->max,
CPUFREQ_RELATION_H);
} else {
int freq = powersave_bias_target(policy, policy->max,
CPUFREQ_RELATION_H);
__cpufreq_driver_target(policy, freq,
CPUFREQ_RELATION_L);
}
return;
}
/* Check for frequency decrease */
/* if we cannot reduce the frequency anymore, break out early */
if (policy->cur == policy->min)
return;
/*
* 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 (load < (dbs_tuners_ins.up_threshold - 10)) {
unsigned int freq_next, freq_cur;
freq_cur = cpufreq_driver_getavg(policy);
if (!freq_cur)
freq_cur = policy->cur;
freq_next = (freq_cur * load) /
(dbs_tuners_ins.up_threshold - 10);
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
if (!dbs_tuners_ins.powersave_bias) {
__cpufreq_driver_target(policy, freq_next,
CPUFREQ_RELATION_L);
} else {
int freq = powersave_bias_target(policy, freq_next,
CPUFREQ_RELATION_L);
__cpufreq_driver_target(policy, freq,
CPUFREQ_RELATION_L);
}
}
}
static void do_dbs_timer(struct work_struct *work)
{
unsigned int cpu = smp_processor_id();
struct cpu_dbs_info_s *dbs_info = &per_cpu(cpu_dbs_info, cpu);
enum dbs_sample sample_type = dbs_info->sample_type;
/* We want all CPUs to do sampling nearly on same jiffy */
int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
/* Permit rescheduling of this work item */
work_release(work);
delay -= jiffies % delay;
if (!dbs_info->enable)
return;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
/* Common NORMAL_SAMPLE setup */
dbs_info->sample_type = DBS_NORMAL_SAMPLE;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
if (!dbs_tuners_ins.powersave_bias ||
sample_type == DBS_NORMAL_SAMPLE) {
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
lock_cpu_hotplug();
dbs_check_cpu(dbs_info);
unlock_cpu_hotplug();
if (dbs_info->freq_lo) {
/* Setup timer for SUB_SAMPLE */
dbs_info->sample_type = DBS_SUB_SAMPLE;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
delay = dbs_info->freq_hi_jiffies;
}
} else {
__cpufreq_driver_target(dbs_info->cur_policy,
dbs_info->freq_lo,
CPUFREQ_RELATION_H);
}
queue_delayed_work_on(cpu, kondemand_wq, &dbs_info->work, delay);
}
static inline void dbs_timer_init(unsigned int cpu)
{
struct cpu_dbs_info_s *dbs_info = &per_cpu(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;
[CPUFREQ][2/2] ondemand: updated add powersave_bias tunable ondemand selects the minimum frequency that can retire a workload with negligible idle time -- ideally resulting in the highest performance/power efficiency with negligible performance impact. But on some systems and some workloads, this algorithm is more performance biased than necessary, and de-tuning it a bit to allow some performance impact can save measurable power. This patch adds a "powersave_bias" tunable to ondemand to allow it to reduce its target frequency by a specified percent. By default, the powersave_bias is 0 and has no effect. powersave_bias is in units of 0.1%, so it has an effective range of 1 through 1000, resulting in 0.1% to 100% impact. In practice, users will not be able to detect a difference between 0.1% increments, but 1.0% increments turned out to be too large. Also, the max value of 1000 (100%) would simply peg the system in its deepest power saving P-state, unless the processor really has a hardware P-state at 0Hz:-) For example, If ondemand requests 2.0GHz based on utilization, and powersave_bias=100, this code will knock 10% off the target and seek a target of 1.8GHz instead of 2.0GHz until the next sampling. If 1.8 is an exact match with an hardware frequency we use it, otherwise we average our time between the frequency next higher than 1.8 and next lower than 1.8. Note that a user or administrative program can change powersave_bias at run-time depending on how they expect the system to be used. Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi at intel.com> Signed-off-by: Alexey Starikovskiy <alexey.y.starikovskiy at intel.com> Signed-off-by: Dave Jones <davej@redhat.com>
2006-07-31 18:28:12 +00:00
ondemand_powersave_bias_init();
INIT_DELAYED_WORK_NAR(&dbs_info->work, do_dbs_timer);
dbs_info->sample_type = DBS_NORMAL_SAMPLE;
queue_delayed_work_on(cpu, kondemand_wq, &dbs_info->work, delay);
}
static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
{
dbs_info->enable = 0;
cancel_delayed_work(&dbs_info->work);
flush_workqueue(kondemand_wq);
}
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 j;
int rc;
this_dbs_info = &per_cpu(cpu_dbs_info, cpu);
switch (event) {
case CPUFREQ_GOV_START:
if ((!cpu_online(cpu)) || (!policy->cur))
return -EINVAL;
if (policy->cpuinfo.transition_latency >
(TRANSITION_LATENCY_LIMIT * 1000)) {
printk(KERN_WARNING "ondemand governor failed to load "
"due to too long transition latency\n");
return -EINVAL;
}
if (this_dbs_info->enable) /* Already enabled */
break;
mutex_lock(&dbs_mutex);
dbs_enable++;
if (dbs_enable == 1) {
kondemand_wq = create_workqueue("kondemand");
if (!kondemand_wq) {
printk(KERN_ERR
"Creation of kondemand failed\n");
dbs_enable--;
mutex_unlock(&dbs_mutex);
return -ENOSPC;
}
}
rc = sysfs_create_group(&policy->kobj, &dbs_attr_group);
if (rc) {
if (dbs_enable == 1)
destroy_workqueue(kondemand_wq);
dbs_enable--;
mutex_unlock(&dbs_mutex);
return rc;
}
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->cur_policy = policy;
j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j);
j_dbs_info->prev_cpu_wall = get_jiffies_64();
}
this_dbs_info->enable = 1;
/*
* Start the timerschedule work, when this governor
* is used for first time
*/
if (dbs_enable == 1) {
unsigned int latency;
/* policy latency is in nS. Convert it to uS first */
latency = policy->cpuinfo.transition_latency / 1000;
if (latency == 0)
latency = 1;
def_sampling_rate = 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(policy->cpu);
mutex_unlock(&dbs_mutex);
break;
case CPUFREQ_GOV_STOP:
mutex_lock(&dbs_mutex);
dbs_timer_exit(this_dbs_info);
sysfs_remove_group(&policy->kobj, &dbs_attr_group);
dbs_enable--;
if (dbs_enable == 0)
destroy_workqueue(kondemand_wq);
mutex_unlock(&dbs_mutex);
break;
case CPUFREQ_GOV_LIMITS:
mutex_lock(&dbs_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(&dbs_mutex);
break;
}
return 0;
}
static struct cpufreq_governor cpufreq_gov_dbs = {
.name = "ondemand",
.governor = cpufreq_governor_dbs,
.owner = THIS_MODULE,
};
static int __init cpufreq_gov_dbs_init(void)
{
return cpufreq_register_governor(&cpufreq_gov_dbs);
}
static void __exit cpufreq_gov_dbs_exit(void)
{
cpufreq_unregister_governor(&cpufreq_gov_dbs);
}
MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>");
MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>");
MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
"Low Latency Frequency Transition capable processors");
MODULE_LICENSE("GPL");
module_init(cpufreq_gov_dbs_init);
module_exit(cpufreq_gov_dbs_exit);