[deliverable/linux.git] / drivers / cpufreq / cpufreq_ondemand.c

/*
 *  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.
 */

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/cpu.h>
#include <linux/percpu-defs.h>
#include <linux/slab.h>
#include <linux/tick.h>

#include "cpufreq_ondemand.h"

/* On-demand governor macros */
#define DEF_FREQUENCY_UP_THRESHOLD		(80)
#define DEF_SAMPLING_DOWN_FACTOR		(1)
#define MAX_SAMPLING_DOWN_FACTOR		(100000)
#define MICRO_FREQUENCY_UP_THRESHOLD		(95)
#define MICRO_FREQUENCY_MIN_SAMPLE_RATE		(10000)
#define MIN_FREQUENCY_UP_THRESHOLD		(11)
#define MAX_FREQUENCY_UP_THRESHOLD		(100)

static struct od_ops od_ops;

static unsigned int default_powersave_bias;

/*
 * Not all CPUs want IO time to be accounted as busy; this depends on how
 * efficient idling at a higher frequency/voltage is.
 * Pavel Machek says this is not so for various generations of AMD and old
 * Intel systems.
 * Mike Chan (android.com) claims this is also not true for ARM.
 * Because of this, whitelist specific known (series) of CPUs by default, and
 * leave all others up to the user.
 */
static int should_io_be_busy(void)
{
#if defined(CONFIG_X86)
	/*
	 * For Intel, Core 2 (model 15) and later have an efficient idle.
	 */
	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
			boot_cpu_data.x86 == 6 &&
			boot_cpu_data.x86_model >= 15)
		return 1;
#endif
	return 0;
}

/*
 * 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_delay_us,
 * freq_lo, and freq_lo_delay_us in percpu area for averaging freqs.
 */
static unsigned int generic_powersave_bias_target(struct cpufreq_policy *policy,
		unsigned int freq_next, unsigned int relation)
{
	unsigned int freq_req, freq_reduc, freq_avg;
	unsigned int freq_hi, freq_lo;
	unsigned int index;
	unsigned int delay_hi_us;
	struct policy_dbs_info *policy_dbs = policy->governor_data;
	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
	struct dbs_data *dbs_data = policy_dbs->dbs_data;
	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	struct cpufreq_frequency_table *freq_table = policy->freq_table;

	if (!freq_table) {
		dbs_info->freq_lo = 0;
		dbs_info->freq_lo_delay_us = 0;
		return freq_next;
	}

	index = cpufreq_frequency_table_target(policy, freq_next, relation);
	freq_req = freq_table[index].frequency;
	freq_reduc = freq_req * od_tuners->powersave_bias / 1000;
	freq_avg = freq_req - freq_reduc;

	/* Find freq bounds for freq_avg in freq_table */
	index = cpufreq_frequency_table_target(policy, freq_avg,
					       CPUFREQ_RELATION_H);
	freq_lo = freq_table[index].frequency;
	index = cpufreq_frequency_table_target(policy, freq_avg,
					       CPUFREQ_RELATION_L);
	freq_hi = 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_delay_us = 0;
		return freq_lo;
	}
	delay_hi_us = (freq_avg - freq_lo) * dbs_data->sampling_rate;
	delay_hi_us += (freq_hi - freq_lo) / 2;
	delay_hi_us /= freq_hi - freq_lo;
	dbs_info->freq_hi_delay_us = delay_hi_us;
	dbs_info->freq_lo = freq_lo;
	dbs_info->freq_lo_delay_us = dbs_data->sampling_rate - delay_hi_us;
	return freq_hi;
}

static void ondemand_powersave_bias_init(struct cpufreq_policy *policy)
{
	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);

	dbs_info->freq_lo = 0;
}

static void dbs_freq_increase(struct cpufreq_policy *policy, unsigned int freq)
{
	struct policy_dbs_info *policy_dbs = policy->governor_data;
	struct dbs_data *dbs_data = policy_dbs->dbs_data;
	struct od_dbs_tuners *od_tuners = dbs_data->tuners;

	if (od_tuners->powersave_bias)
		freq = od_ops.powersave_bias_target(policy, freq,
				CPUFREQ_RELATION_H);
	else if (policy->cur == policy->max)
		return;

	__cpufreq_driver_target(policy, freq, od_tuners->powersave_bias ?
			CPUFREQ_RELATION_L : CPUFREQ_RELATION_H);
}

/*
 * Every sampling_rate, we check, if current idle time is less than 20%
 * (default), then we try to increase frequency. Else, we adjust the frequency
 * proportional to load.
 */
static void od_update(struct cpufreq_policy *policy)
{
	struct policy_dbs_info *policy_dbs = policy->governor_data;
	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
	struct dbs_data *dbs_data = policy_dbs->dbs_data;
	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	unsigned int load = dbs_update(policy);

	dbs_info->freq_lo = 0;

	/* Check for frequency increase */
	if (load > dbs_data->up_threshold) {
		/* If switching to max speed, apply sampling_down_factor */
		if (policy->cur < policy->max)
			policy_dbs->rate_mult = dbs_data->sampling_down_factor;
		dbs_freq_increase(policy, policy->max);
	} else {
		/* Calculate the next frequency proportional to load */
		unsigned int freq_next, min_f, max_f;

		min_f = policy->cpuinfo.min_freq;
		max_f = policy->cpuinfo.max_freq;
		freq_next = min_f + load * (max_f - min_f) / 100;

		/* No longer fully busy, reset rate_mult */
		policy_dbs->rate_mult = 1;

		if (od_tuners->powersave_bias)
			freq_next = od_ops.powersave_bias_target(policy,
								 freq_next,
								 CPUFREQ_RELATION_L);

		__cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_C);
	}
}

static unsigned int od_dbs_timer(struct cpufreq_policy *policy)
{
	struct policy_dbs_info *policy_dbs = policy->governor_data;
	struct dbs_data *dbs_data = policy_dbs->dbs_data;
	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
	int sample_type = dbs_info->sample_type;

	/* Common NORMAL_SAMPLE setup */
	dbs_info->sample_type = OD_NORMAL_SAMPLE;
	/*
	 * OD_SUB_SAMPLE doesn't make sense if sample_delay_ns is 0, so ignore
	 * it then.
	 */
	if (sample_type == OD_SUB_SAMPLE && policy_dbs->sample_delay_ns > 0) {
		__cpufreq_driver_target(policy, dbs_info->freq_lo,
					CPUFREQ_RELATION_H);
		return dbs_info->freq_lo_delay_us;
	}

	od_update(policy);

	if (dbs_info->freq_lo) {
		/* Setup timer for SUB_SAMPLE */
		dbs_info->sample_type = OD_SUB_SAMPLE;
		return dbs_info->freq_hi_delay_us;
	}

	return dbs_data->sampling_rate * policy_dbs->rate_mult;
}

/************************** sysfs interface ************************/
static struct dbs_governor od_dbs_gov;

static ssize_t store_io_is_busy(struct gov_attr_set *attr_set, const char *buf,
				size_t count)
{
	struct dbs_data *dbs_data = to_dbs_data(attr_set);
	unsigned int input;
	int ret;

	ret = sscanf(buf, "%u", &input);
	if (ret != 1)
		return -EINVAL;
	dbs_data->io_is_busy = !!input;

	/* we need to re-evaluate prev_cpu_idle */
	gov_update_cpu_data(dbs_data);

	return count;
}

static ssize_t store_up_threshold(struct gov_attr_set *attr_set,
				  const char *buf, size_t count)
{
	struct dbs_data *dbs_data = to_dbs_data(attr_set);
	unsigned int input;
	int ret;
	ret = sscanf(buf, "%u", &input);

	if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD ||
			input < MIN_FREQUENCY_UP_THRESHOLD) {
		return -EINVAL;
	}

	dbs_data->up_threshold = input;
	return count;
}

static ssize_t store_sampling_down_factor(struct gov_attr_set *attr_set,
					  const char *buf, size_t count)
{
	struct dbs_data *dbs_data = to_dbs_data(attr_set);
	struct policy_dbs_info *policy_dbs;
	unsigned int input;
	int ret;
	ret = sscanf(buf, "%u", &input);

	if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
		return -EINVAL;

	dbs_data->sampling_down_factor = input;

	/* Reset down sampling multiplier in case it was active */
	list_for_each_entry(policy_dbs, &attr_set->policy_list, list) {
		/*
		 * Doing this without locking might lead to using different
		 * rate_mult values in od_update() and od_dbs_timer().
		 */
		mutex_lock(&policy_dbs->timer_mutex);
		policy_dbs->rate_mult = 1;
		mutex_unlock(&policy_dbs->timer_mutex);
	}

	return count;
}

static ssize_t store_ignore_nice_load(struct gov_attr_set *attr_set,
				      const char *buf, size_t count)
{
	struct dbs_data *dbs_data = to_dbs_data(attr_set);
	unsigned int input;
	int ret;

	ret = sscanf(buf, "%u", &input);
	if (ret != 1)
		return -EINVAL;

	if (input > 1)
		input = 1;

	if (input == dbs_data->ignore_nice_load) { /* nothing to do */
		return count;
	}
	dbs_data->ignore_nice_load = input;

	/* we need to re-evaluate prev_cpu_idle */
	gov_update_cpu_data(dbs_data);

	return count;
}

static ssize_t store_powersave_bias(struct gov_attr_set *attr_set,
				    const char *buf, size_t count)
{
	struct dbs_data *dbs_data = to_dbs_data(attr_set);
	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	struct policy_dbs_info *policy_dbs;
	unsigned int input;
	int ret;
	ret = sscanf(buf, "%u", &input);

	if (ret != 1)
		return -EINVAL;

	if (input > 1000)
		input = 1000;

	od_tuners->powersave_bias = input;

	list_for_each_entry(policy_dbs, &attr_set->policy_list, list)
		ondemand_powersave_bias_init(policy_dbs->policy);

	return count;
}

gov_show_one_common(sampling_rate);
gov_show_one_common(up_threshold);
gov_show_one_common(sampling_down_factor);
gov_show_one_common(ignore_nice_load);
gov_show_one_common(min_sampling_rate);
gov_show_one_common(io_is_busy);
gov_show_one(od, powersave_bias);

gov_attr_rw(sampling_rate);
gov_attr_rw(io_is_busy);
gov_attr_rw(up_threshold);
gov_attr_rw(sampling_down_factor);
gov_attr_rw(ignore_nice_load);
gov_attr_rw(powersave_bias);
gov_attr_ro(min_sampling_rate);

static struct attribute *od_attributes[] = {
	&min_sampling_rate.attr,
	&sampling_rate.attr,
	&up_threshold.attr,
	&sampling_down_factor.attr,
	&ignore_nice_load.attr,
	&powersave_bias.attr,
	&io_is_busy.attr,
	NULL
};

/************************** sysfs end ************************/

static struct policy_dbs_info *od_alloc(void)
{
	struct od_policy_dbs_info *dbs_info;

	dbs_info = kzalloc(sizeof(*dbs_info), GFP_KERNEL);
	return dbs_info ? &dbs_info->policy_dbs : NULL;
}

static void od_free(struct policy_dbs_info *policy_dbs)
{
	kfree(to_dbs_info(policy_dbs));
}

static int od_init(struct dbs_data *dbs_data)
{
	struct od_dbs_tuners *tuners;
	u64 idle_time;
	int cpu;

	tuners = kzalloc(sizeof(*tuners), GFP_KERNEL);
	if (!tuners)
		return -ENOMEM;

	cpu = get_cpu();
	idle_time = get_cpu_idle_time_us(cpu, NULL);
	put_cpu();
	if (idle_time != -1ULL) {
		/* Idle micro accounting is supported. Use finer thresholds */
		dbs_data->up_threshold = MICRO_FREQUENCY_UP_THRESHOLD;
		/*
		 * In nohz/micro accounting case we set the minimum frequency
		 * not depending on HZ, but fixed (very low). The deferred
		 * timer might skip some samples if idle/sleeping as needed.
		*/
		dbs_data->min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE;
	} else {
		dbs_data->up_threshold = DEF_FREQUENCY_UP_THRESHOLD;

		/* For correct statistics, we need 10 ticks for each measure */
		dbs_data->min_sampling_rate = MIN_SAMPLING_RATE_RATIO *
			jiffies_to_usecs(10);
	}

	dbs_data->sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR;
	dbs_data->ignore_nice_load = 0;
	tuners->powersave_bias = default_powersave_bias;
	dbs_data->io_is_busy = should_io_be_busy();

	dbs_data->tuners = tuners;
	return 0;
}

static void od_exit(struct dbs_data *dbs_data)
{
	kfree(dbs_data->tuners);
}

static void od_start(struct cpufreq_policy *policy)
{
	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);

	dbs_info->sample_type = OD_NORMAL_SAMPLE;
	ondemand_powersave_bias_init(policy);
}

static struct od_ops od_ops = {
	.powersave_bias_target = generic_powersave_bias_target,
};

static struct dbs_governor od_dbs_gov = {
	.gov = CPUFREQ_DBS_GOVERNOR_INITIALIZER("ondemand"),
	.kobj_type = { .default_attrs = od_attributes },
	.gov_dbs_timer = od_dbs_timer,
	.alloc = od_alloc,
	.free = od_free,
	.init = od_init,
	.exit = od_exit,
	.start = od_start,
};

#define CPU_FREQ_GOV_ONDEMAND	(&od_dbs_gov.gov)

static void od_set_powersave_bias(unsigned int powersave_bias)
{
	unsigned int cpu;
	cpumask_t done;

	default_powersave_bias = powersave_bias;
	cpumask_clear(&done);

	get_online_cpus();
	for_each_online_cpu(cpu) {
		struct cpufreq_policy *policy;
		struct policy_dbs_info *policy_dbs;
		struct dbs_data *dbs_data;
		struct od_dbs_tuners *od_tuners;

		if (cpumask_test_cpu(cpu, &done))
			continue;

		policy = cpufreq_cpu_get_raw(cpu);
		if (!policy || policy->governor != CPU_FREQ_GOV_ONDEMAND)
			continue;

		policy_dbs = policy->governor_data;
		if (!policy_dbs)
			continue;

		cpumask_or(&done, &done, policy->cpus);

		dbs_data = policy_dbs->dbs_data;
		od_tuners = dbs_data->tuners;
		od_tuners->powersave_bias = default_powersave_bias;
	}
	put_online_cpus();
}

void od_register_powersave_bias_handler(unsigned int (*f)
		(struct cpufreq_policy *, unsigned int, unsigned int),
		unsigned int powersave_bias)
{
	od_ops.powersave_bias_target = f;
	od_set_powersave_bias(powersave_bias);
}
EXPORT_SYMBOL_GPL(od_register_powersave_bias_handler);

void od_unregister_powersave_bias_handler(void)
{
	od_ops.powersave_bias_target = generic_powersave_bias_target;
	od_set_powersave_bias(0);
}
EXPORT_SYMBOL_GPL(od_unregister_powersave_bias_handler);

static int __init cpufreq_gov_dbs_init(void)
{
	return cpufreq_register_governor(CPU_FREQ_GOV_ONDEMAND);
}

static void __exit cpufreq_gov_dbs_exit(void)
{
	cpufreq_unregister_governor(CPU_FREQ_GOV_ONDEMAND);
}

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");

#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
struct cpufreq_governor *cpufreq_default_governor(void)
{
	return CPU_FREQ_GOV_ONDEMAND;
}

fs_initcall(cpufreq_gov_dbs_init);
#else
module_init(cpufreq_gov_dbs_init);
#endif
module_exit(cpufreq_gov_dbs_exit);
Commit	Line	Data
1da177e4 LT	1	/*
	2	* drivers/cpufreq/cpufreq_ondemand.c
	3	*
	4	* Copyright (C) 2001 Russell King
	5	* (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
	6	* Jun Nakajima <jun.nakajima@intel.com>
	7	*
	8	* This program is free software; you can redistribute it and/or modify
	9	* it under the terms of the GNU General Public License version 2 as
	10	* published by the Free Software Foundation.
	11	*/
	12
4471a34f VK	13	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
4471a34f VK	14
5ff0a268	15	#include <linux/cpu.h>
4471a34f	16	#include <linux/percpu-defs.h>
4d5dcc42	17	#include <linux/slab.h>
80800913	18	#include <linux/tick.h>
7d5a9956 RW	19
7d5a9956 RW	20	#include "cpufreq_ondemand.h"
1da177e4	21
06eb09d1	22	/* On-demand governor macros */
1da177e4	23	#define DEF_FREQUENCY_UP_THRESHOLD (80)
3f78a9f7 DN	24	#define DEF_SAMPLING_DOWN_FACTOR (1)
3f78a9f7 DN	25	#define MAX_SAMPLING_DOWN_FACTOR (100000)
80800913	26	#define MICRO_FREQUENCY_UP_THRESHOLD (95)
cef9615a	27	#define MICRO_FREQUENCY_MIN_SAMPLE_RATE (10000)
c29f1403	28	#define MIN_FREQUENCY_UP_THRESHOLD (11)
1da177e4 LT	29	#define MAX_FREQUENCY_UP_THRESHOLD (100)
1da177e4 LT	30
fb30809e JS	31	static struct od_ops od_ops;
fb30809e JS	32
c2837558 JS	33	static unsigned int default_powersave_bias;
c2837558 JS	34
4471a34f VK	35	/*
	36	* Not all CPUs want IO time to be accounted as busy; this depends on how
	37	* efficient idling at a higher frequency/voltage is.
	38	* Pavel Machek says this is not so for various generations of AMD and old
	39	* Intel systems.
06eb09d1	40	* Mike Chan (android.com) claims this is also not true for ARM.
4471a34f VK	41	* Because of this, whitelist specific known (series) of CPUs by default, and
	42	* leave all others up to the user.
	43	*/
	44	static int should_io_be_busy(void)
	45	{
	46	#if defined(CONFIG_X86)
	47	/*
06eb09d1	48	* For Intel, Core 2 (model 15) and later have an efficient idle.
4471a34f VK	49	*/
	50	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
	51	boot_cpu_data.x86 == 6 &&
	52	boot_cpu_data.x86_model >= 15)
	53	return 1;
	54	#endif
	55	return 0;
6b8fcd90 AV	56	}
6b8fcd90 AV	57
05ca0350 AS	58	/*
05ca0350 AS	59	* Find right freq to be set now with powersave_bias on.
07aa4402 RW	60	* Returns the freq_hi to be used right now and will set freq_hi_delay_us,
07aa4402 RW	61	* freq_lo, and freq_lo_delay_us in percpu area for averaging freqs.
05ca0350	62	*/
fb30809e	63	static unsigned int generic_powersave_bias_target(struct cpufreq_policy *policy,
4471a34f	64	unsigned int freq_next, unsigned int relation)
05ca0350 AS	65	{
	66	unsigned int freq_req, freq_reduc, freq_avg;
	67	unsigned int freq_hi, freq_lo;
d218ed77	68	unsigned int index;
07aa4402	69	unsigned int delay_hi_us;
bc505475	70	struct policy_dbs_info *policy_dbs = policy->governor_data;
7d5a9956	71	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
bc505475	72	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42	73	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
34ac5d7a	74	struct cpufreq_frequency_table *freq_table = policy->freq_table;
05ca0350	75
34ac5d7a	76	if (!freq_table) {
05ca0350	77	dbs_info->freq_lo = 0;
07aa4402	78	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	79	return freq_next;
	80	}
	81
d218ed77	82	index = cpufreq_frequency_table_target(policy, freq_next, relation);
34ac5d7a	83	freq_req = freq_table[index].frequency;
4d5dcc42	84	freq_reduc = freq_req * od_tuners->powersave_bias / 1000;
05ca0350 AS	85	freq_avg = freq_req - freq_reduc;
	86
	87	/* Find freq bounds for freq_avg in freq_table */
d218ed77 VK	88	index = cpufreq_frequency_table_target(policy, freq_avg,
d218ed77 VK	89	CPUFREQ_RELATION_H);
34ac5d7a	90	freq_lo = freq_table[index].frequency;
d218ed77 VK	91	index = cpufreq_frequency_table_target(policy, freq_avg,
d218ed77 VK	92	CPUFREQ_RELATION_L);
34ac5d7a	93	freq_hi = freq_table[index].frequency;
05ca0350 AS	94
	95	/* Find out how long we have to be in hi and lo freqs */
	96	if (freq_hi == freq_lo) {
	97	dbs_info->freq_lo = 0;
07aa4402	98	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	99	return freq_lo;
05ca0350 AS	100	}
07aa4402 RW	101	delay_hi_us = (freq_avg - freq_lo) * dbs_data->sampling_rate;
	102	delay_hi_us += (freq_hi - freq_lo) / 2;
	103	delay_hi_us /= freq_hi - freq_lo;
	104	dbs_info->freq_hi_delay_us = delay_hi_us;
05ca0350	105	dbs_info->freq_lo = freq_lo;
07aa4402	106	dbs_info->freq_lo_delay_us = dbs_data->sampling_rate - delay_hi_us;
05ca0350 AS	107	return freq_hi;
	108	}
	109
d1db75ff	110	static void ondemand_powersave_bias_init(struct cpufreq_policy *policy)
05ca0350	111	{
7d5a9956	112	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);
d1db75ff	113
d1db75ff	114	dbs_info->freq_lo = 0;
05ca0350 AS	115	}
05ca0350 AS	116
3a3e9e06	117	static void dbs_freq_increase(struct cpufreq_policy *policy, unsigned int freq)
4471a34f	118	{
bc505475 RW	119	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	120	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42 VK	121	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	122
	123	if (od_tuners->powersave_bias)
3a3e9e06	124	freq = od_ops.powersave_bias_target(policy, freq,
fb30809e	125	CPUFREQ_RELATION_H);
3a3e9e06	126	else if (policy->cur == policy->max)
4471a34f	127	return;
0e625ac1	128
3a3e9e06	129	__cpufreq_driver_target(policy, freq, od_tuners->powersave_bias ?
4471a34f VK	130	CPUFREQ_RELATION_L : CPUFREQ_RELATION_H);
	131	}
	132
	133	/*
	134	* Every sampling_rate, we check, if current idle time is less than 20%
dfa5bb62 SK	135	* (default), then we try to increase frequency. Else, we adjust the frequency
dfa5bb62 SK	136	* proportional to load.
4471a34f	137	*/
4cccf755	138	static void od_update(struct cpufreq_policy *policy)
1da177e4	139	{
7d5a9956 RW	140	struct policy_dbs_info *policy_dbs = policy->governor_data;
7d5a9956 RW	141	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
bc505475	142	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42	143	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
4cccf755	144	unsigned int load = dbs_update(policy);
4471a34f VK	145
	146	dbs_info->freq_lo = 0;
	147
	148	/* Check for frequency increase */
ff4b1789	149	if (load > dbs_data->up_threshold) {
4471a34f VK	150	/* If switching to max speed, apply sampling_down_factor */
4471a34f VK	151	if (policy->cur < policy->max)
57dc3bcd	152	policy_dbs->rate_mult = dbs_data->sampling_down_factor;
4471a34f	153	dbs_freq_increase(policy, policy->max);
dfa5bb62 SK	154	} else {
dfa5bb62 SK	155	/* Calculate the next frequency proportional to load */
6393d6a1 SK	156	unsigned int freq_next, min_f, max_f;
	157
	158	min_f = policy->cpuinfo.min_freq;
	159	max_f = policy->cpuinfo.max_freq;
	160	freq_next = min_f + load * (max_f - min_f) / 100;
4471a34f VK	161
4471a34f VK	162	/* No longer fully busy, reset rate_mult */
57dc3bcd	163	policy_dbs->rate_mult = 1;
4471a34f	164
a7f35cff RW	165	if (od_tuners->powersave_bias)
	166	freq_next = od_ops.powersave_bias_target(policy,
	167	freq_next,
	168	CPUFREQ_RELATION_L);
	169
6393d6a1	170	__cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_C);
4471a34f	171	}
1da177e4 LT	172	}
1da177e4 LT	173
9be4fd2c	174	static unsigned int od_dbs_timer(struct cpufreq_policy *policy)
4471a34f	175	{
bc505475 RW	176	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	177	struct dbs_data *dbs_data = policy_dbs->dbs_data;
7d5a9956	178	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
6e96c5b3	179	int sample_type = dbs_info->sample_type;
4447266b	180
4471a34f	181	/* Common NORMAL_SAMPLE setup */
43e0ee36	182	dbs_info->sample_type = OD_NORMAL_SAMPLE;
4cccf755 RW	183	/*
	184	* OD_SUB_SAMPLE doesn't make sense if sample_delay_ns is 0, so ignore
	185	* it then.
	186	*/
	187	if (sample_type == OD_SUB_SAMPLE && policy_dbs->sample_delay_ns > 0) {
43e0ee36	188	__cpufreq_driver_target(policy, dbs_info->freq_lo,
42994af6	189	CPUFREQ_RELATION_H);
07aa4402	190	return dbs_info->freq_lo_delay_us;
6e96c5b3 RW	191	}
	192
	193	od_update(policy);
	194
	195	if (dbs_info->freq_lo) {
	196	/* Setup timer for SUB_SAMPLE */
	197	dbs_info->sample_type = OD_SUB_SAMPLE;
07aa4402	198	return dbs_info->freq_hi_delay_us;
4471a34f VK	199	}
4471a34f VK	200
07aa4402	201	return dbs_data->sampling_rate * policy_dbs->rate_mult;
da53d61e FB	202	}
da53d61e FB	203
4471a34f	204	/************************ sysfs interface **********************/
7bdad34d	205	static struct dbs_governor od_dbs_gov;
1da177e4	206
0dd3c1d6 RW	207	static ssize_t store_io_is_busy(struct gov_attr_set attr_set, const char buf,
0dd3c1d6 RW	208	size_t count)
19379b11	209	{
0dd3c1d6	210	struct dbs_data *dbs_data = to_dbs_data(attr_set);
19379b11 AV	211	unsigned int input;
	212	int ret;
	213
	214	ret = sscanf(buf, "%u", &input);
	215	if (ret != 1)
	216	return -EINVAL;
8847e038	217	dbs_data->io_is_busy = !!input;
9366d840 SK	218
9366d840 SK	219	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	220	gov_update_cpu_data(dbs_data);
a33cce1c	221
19379b11 AV	222	return count;
	223	}
	224
0dd3c1d6 RW	225	static ssize_t store_up_threshold(struct gov_attr_set *attr_set,
0dd3c1d6 RW	226	const char *buf, size_t count)
1da177e4	227	{
0dd3c1d6	228	struct dbs_data *dbs_data = to_dbs_data(attr_set);
1da177e4 LT	229	unsigned int input;
1da177e4 LT	230	int ret;
ffac80e9	231	ret = sscanf(buf, "%u", &input);
1da177e4	232
32ee8c3e	233	if (ret != 1 \|\| input > MAX_FREQUENCY_UP_THRESHOLD \|\|
c29f1403	234	input < MIN_FREQUENCY_UP_THRESHOLD) {
1da177e4 LT	235	return -EINVAL;
1da177e4 LT	236	}
4bd4e428	237
ff4b1789	238	dbs_data->up_threshold = input;
1da177e4 LT	239	return count;
	240	}
	241
0dd3c1d6 RW	242	static ssize_t store_sampling_down_factor(struct gov_attr_set *attr_set,
0dd3c1d6 RW	243	const char *buf, size_t count)
3f78a9f7	244	{
0dd3c1d6	245	struct dbs_data *dbs_data = to_dbs_data(attr_set);
57dc3bcd RW	246	struct policy_dbs_info *policy_dbs;
57dc3bcd RW	247	unsigned int input;
3f78a9f7 DN	248	int ret;
	249	ret = sscanf(buf, "%u", &input);
	250
	251	if (ret != 1 \|\| input > MAX_SAMPLING_DOWN_FACTOR \|\| input < 1)
	252	return -EINVAL;
57dc3bcd	253
ff4b1789	254	dbs_data->sampling_down_factor = input;
3f78a9f7 DN	255
3f78a9f7 DN	256	/* Reset down sampling multiplier in case it was active */
0dd3c1d6	257	list_for_each_entry(policy_dbs, &attr_set->policy_list, list) {
57dc3bcd RW	258	/*
	259	* Doing this without locking might lead to using different
	260	* rate_mult values in od_update() and od_dbs_timer().
	261	*/
	262	mutex_lock(&policy_dbs->timer_mutex);
	263	policy_dbs->rate_mult = 1;
	264	mutex_unlock(&policy_dbs->timer_mutex);
3f78a9f7	265	}
57dc3bcd	266
3f78a9f7 DN	267	return count;
	268	}
	269
0dd3c1d6 RW	270	static ssize_t store_ignore_nice_load(struct gov_attr_set *attr_set,
0dd3c1d6 RW	271	const char *buf, size_t count)
3d5ee9e5	272	{
0dd3c1d6	273	struct dbs_data *dbs_data = to_dbs_data(attr_set);
3d5ee9e5 DJ	274	unsigned int input;
	275	int ret;
	276
ffac80e9	277	ret = sscanf(buf, "%u", &input);
2b03f891	278	if (ret != 1)
3d5ee9e5 DJ	279	return -EINVAL;
3d5ee9e5 DJ	280
2b03f891	281	if (input > 1)
3d5ee9e5	282	input = 1;
32ee8c3e	283
ff4b1789	284	if (input == dbs_data->ignore_nice_load) { /* nothing to do */
3d5ee9e5 DJ	285	return count;
3d5ee9e5 DJ	286	}
ff4b1789	287	dbs_data->ignore_nice_load = input;
3d5ee9e5	288
ccb2fe20	289	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	290	gov_update_cpu_data(dbs_data);
1ca3abdb	291
3d5ee9e5 DJ	292	return count;
	293	}
	294
0dd3c1d6 RW	295	static ssize_t store_powersave_bias(struct gov_attr_set *attr_set,
0dd3c1d6 RW	296	const char *buf, size_t count)
05ca0350	297	{
0dd3c1d6	298	struct dbs_data *dbs_data = to_dbs_data(attr_set);
4d5dcc42	299	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
d1db75ff	300	struct policy_dbs_info *policy_dbs;
05ca0350 AS	301	unsigned int input;
	302	int ret;
	303	ret = sscanf(buf, "%u", &input);
	304
	305	if (ret != 1)
	306	return -EINVAL;
	307
	308	if (input > 1000)
	309	input = 1000;
	310
4d5dcc42	311	od_tuners->powersave_bias = input;
d1db75ff	312
0dd3c1d6	313	list_for_each_entry(policy_dbs, &attr_set->policy_list, list)
d1db75ff RW	314	ondemand_powersave_bias_init(policy_dbs->policy);
d1db75ff RW	315
05ca0350 AS	316	return count;
	317	}
	318
c4435630 VK	319	gov_show_one_common(sampling_rate);
	320	gov_show_one_common(up_threshold);
	321	gov_show_one_common(sampling_down_factor);
	322	gov_show_one_common(ignore_nice_load);
	323	gov_show_one_common(min_sampling_rate);
8847e038	324	gov_show_one_common(io_is_busy);
c4435630 VK	325	gov_show_one(od, powersave_bias);
	326
	327	gov_attr_rw(sampling_rate);
	328	gov_attr_rw(io_is_busy);
	329	gov_attr_rw(up_threshold);
	330	gov_attr_rw(sampling_down_factor);
	331	gov_attr_rw(ignore_nice_load);
	332	gov_attr_rw(powersave_bias);
	333	gov_attr_ro(min_sampling_rate);
	334
	335	static struct attribute *od_attributes[] = {
	336	&min_sampling_rate.attr,
	337	&sampling_rate.attr,
	338	&up_threshold.attr,
	339	&sampling_down_factor.attr,
	340	&ignore_nice_load.attr,
	341	&powersave_bias.attr,
	342	&io_is_busy.attr,
1da177e4 LT	343	NULL
	344	};
	345
1da177e4 LT	346	/************************ sysfs end **********************/
1da177e4 LT	347
7d5a9956 RW	348	static struct policy_dbs_info *od_alloc(void)
	349	{
	350	struct od_policy_dbs_info *dbs_info;
	351
	352	dbs_info = kzalloc(sizeof(*dbs_info), GFP_KERNEL);
	353	return dbs_info ? &dbs_info->policy_dbs : NULL;
	354	}
	355
	356	static void od_free(struct policy_dbs_info *policy_dbs)
	357	{
	358	kfree(to_dbs_info(policy_dbs));
	359	}
	360
9a15fb2c	361	static int od_init(struct dbs_data *dbs_data)
4d5dcc42 VK	362	{
	363	struct od_dbs_tuners *tuners;
	364	u64 idle_time;
	365	int cpu;
	366
d5b73cd8	367	tuners = kzalloc(sizeof(*tuners), GFP_KERNEL);
a69d6b29	368	if (!tuners)
4d5dcc42	369	return -ENOMEM;
4d5dcc42 VK	370
	371	cpu = get_cpu();
	372	idle_time = get_cpu_idle_time_us(cpu, NULL);
	373	put_cpu();
	374	if (idle_time != -1ULL) {
	375	/* Idle micro accounting is supported. Use finer thresholds */
ff4b1789	376	dbs_data->up_threshold = MICRO_FREQUENCY_UP_THRESHOLD;
4d5dcc42 VK	377	/*
	378	* In nohz/micro accounting case we set the minimum frequency
	379	* not depending on HZ, but fixed (very low). The deferred
	380	* timer might skip some samples if idle/sleeping as needed.
	381	*/
	382	dbs_data->min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE;
	383	} else {
ff4b1789	384	dbs_data->up_threshold = DEF_FREQUENCY_UP_THRESHOLD;
4d5dcc42 VK	385
	386	/* For correct statistics, we need 10 ticks for each measure */
	387	dbs_data->min_sampling_rate = MIN_SAMPLING_RATE_RATIO *
	388	jiffies_to_usecs(10);
	389	}
	390
ff4b1789 VK	391	dbs_data->sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR;
ff4b1789 VK	392	dbs_data->ignore_nice_load = 0;
c2837558	393	tuners->powersave_bias = default_powersave_bias;
8847e038	394	dbs_data->io_is_busy = should_io_be_busy();
4d5dcc42 VK	395
4d5dcc42 VK	396	dbs_data->tuners = tuners;
4d5dcc42 VK	397	return 0;
	398	}
	399
9a15fb2c	400	static void od_exit(struct dbs_data *dbs_data)
4d5dcc42 VK	401	{
	402	kfree(dbs_data->tuners);
	403	}
	404
702c9e54 RW	405	static void od_start(struct cpufreq_policy *policy)
702c9e54 RW	406	{
7d5a9956	407	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);
702c9e54 RW	408
702c9e54 RW	409	dbs_info->sample_type = OD_NORMAL_SAMPLE;
d1db75ff	410	ondemand_powersave_bias_init(policy);
702c9e54 RW	411	}
702c9e54 RW	412
4471a34f	413	static struct od_ops od_ops = {
fb30809e	414	.powersave_bias_target = generic_powersave_bias_target,
4471a34f	415	};
2f8a835c	416
7bdad34d	417	static struct dbs_governor od_dbs_gov = {
e788892b	418	.gov = CPUFREQ_DBS_GOVERNOR_INITIALIZER("ondemand"),
c4435630	419	.kobj_type = { .default_attrs = od_attributes },
4471a34f	420	.gov_dbs_timer = od_dbs_timer,
7d5a9956 RW	421	.alloc = od_alloc,
7d5a9956 RW	422	.free = od_free,
4d5dcc42 VK	423	.init = od_init,
4d5dcc42 VK	424	.exit = od_exit,
702c9e54	425	.start = od_start,
4471a34f	426	};
1da177e4	427
7bdad34d	428	#define CPU_FREQ_GOV_ONDEMAND (&od_dbs_gov.gov)
af926185	429
fb30809e JS	430	static void od_set_powersave_bias(unsigned int powersave_bias)
fb30809e JS	431	{
fb30809e JS	432	unsigned int cpu;
	433	cpumask_t done;
	434
c2837558	435	default_powersave_bias = powersave_bias;
fb30809e JS	436	cpumask_clear(&done);
	437
	438	get_online_cpus();
	439	for_each_online_cpu(cpu) {
8c8f77fd	440	struct cpufreq_policy *policy;
e40e7b25	441	struct policy_dbs_info *policy_dbs;
8c8f77fd RW	442	struct dbs_data *dbs_data;
8c8f77fd RW	443	struct od_dbs_tuners *od_tuners;
44152cb8	444
fb30809e JS	445	if (cpumask_test_cpu(cpu, &done))
	446	continue;
	447
8c8f77fd RW	448	policy = cpufreq_cpu_get_raw(cpu);
	449	if (!policy \|\| policy->governor != CPU_FREQ_GOV_ONDEMAND)
	450	continue;
	451
	452	policy_dbs = policy->governor_data;
e40e7b25	453	if (!policy_dbs)
c2837558	454	continue;
fb30809e JS	455
fb30809e JS	456	cpumask_or(&done, &done, policy->cpus);
c2837558	457
bc505475	458	dbs_data = policy_dbs->dbs_data;
c2837558 JS	459	od_tuners = dbs_data->tuners;
c2837558 JS	460	od_tuners->powersave_bias = default_powersave_bias;
fb30809e JS	461	}
	462	put_online_cpus();
	463	}
	464
	465	void od_register_powersave_bias_handler(unsigned int (*f)
	466	(struct cpufreq_policy *, unsigned int, unsigned int),
	467	unsigned int powersave_bias)
	468	{
	469	od_ops.powersave_bias_target = f;
	470	od_set_powersave_bias(powersave_bias);
	471	}
	472	EXPORT_SYMBOL_GPL(od_register_powersave_bias_handler);
	473
	474	void od_unregister_powersave_bias_handler(void)
	475	{
	476	od_ops.powersave_bias_target = generic_powersave_bias_target;
	477	od_set_powersave_bias(0);
	478	}
	479	EXPORT_SYMBOL_GPL(od_unregister_powersave_bias_handler);
	480
1da177e4 LT	481	static int __init cpufreq_gov_dbs_init(void)
1da177e4 LT	482	{
af926185	483	return cpufreq_register_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	484	}
	485
	486	static void __exit cpufreq_gov_dbs_exit(void)
	487	{
af926185	488	cpufreq_unregister_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	489	}
1da177e4 LT	490
ffac80e9 VP	491	MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>");
	492	MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>");
	493	MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
2b03f891	494	"Low Latency Frequency Transition capable processors");
ffac80e9	495	MODULE_LICENSE("GPL");
1da177e4	496
6915719b	497	#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
de1df26b RW	498	struct cpufreq_governor *cpufreq_default_governor(void)
de1df26b RW	499	{
af926185	500	return CPU_FREQ_GOV_ONDEMAND;
de1df26b RW	501	}
de1df26b RW	502
6915719b JW	503	fs_initcall(cpufreq_gov_dbs_init);
6915719b JW	504	#else
1da177e4	505	module_init(cpufreq_gov_dbs_init);
6915719b	506	#endif
1da177e4	507	module_exit(cpufreq_gov_dbs_exit);