[linux-2.6-block.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 = 0;
	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;

	if (!dbs_info->freq_table) {
		dbs_info->freq_lo = 0;
		dbs_info->freq_lo_delay_us = 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 * od_tuners->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_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_table = cpufreq_frequency_get_table(policy->cpu);
	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 dbs_data *dbs_data, const char *buf,
		size_t count)
{
	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 dbs_data *dbs_data, const char *buf,
		size_t count)
{
	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 dbs_data *dbs_data,
		const char *buf, size_t count)
{
	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, &dbs_data->policy_dbs_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 dbs_data *dbs_data,
		const char *buf, size_t count)
{
	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 dbs_data *dbs_data, const char *buf,
		size_t count)
{
	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, &dbs_data->policy_dbs_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, bool notify)
{
	struct od_dbs_tuners *tuners;
	u64 idle_time;
	int cpu;

	tuners = kzalloc(sizeof(*tuners), GFP_KERNEL);
	if (!tuners) {
		pr_err("%s: kzalloc failed\n", __func__);
		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, bool notify)
{
	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 = {
		.name = "ondemand",
		.governor = cpufreq_governor_dbs,
		.max_transition_latency	= TRANSITION_LATENCY_LIMIT,
		.owner = THIS_MODULE,
	},
	.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;
	68	unsigned int index = 0;
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;
05ca0350 AS	74
	75	if (!dbs_info->freq_table) {
	76	dbs_info->freq_lo = 0;
07aa4402	77	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	78	return freq_next;
	79	}
	80
	81	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next,
	82	relation, &index);
	83	freq_req = dbs_info->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 */
	88	index = 0;
	89	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
	90	CPUFREQ_RELATION_H, &index);
	91	freq_lo = dbs_info->freq_table[index].frequency;
	92	index = 0;
	93	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
	94	CPUFREQ_RELATION_L, &index);
	95	freq_hi = dbs_info->freq_table[index].frequency;
	96
	97	/* Find out how long we have to be in hi and lo freqs */
	98	if (freq_hi == freq_lo) {
	99	dbs_info->freq_lo = 0;
07aa4402	100	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	101	return freq_lo;
05ca0350 AS	102	}
07aa4402 RW	103	delay_hi_us = (freq_avg - freq_lo) * dbs_data->sampling_rate;
	104	delay_hi_us += (freq_hi - freq_lo) / 2;
	105	delay_hi_us /= freq_hi - freq_lo;
	106	dbs_info->freq_hi_delay_us = delay_hi_us;
05ca0350	107	dbs_info->freq_lo = freq_lo;
07aa4402	108	dbs_info->freq_lo_delay_us = dbs_data->sampling_rate - delay_hi_us;
05ca0350 AS	109	return freq_hi;
	110	}
	111
d1db75ff	112	static void ondemand_powersave_bias_init(struct cpufreq_policy *policy)
05ca0350	113	{
7d5a9956	114	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);
d1db75ff	115
7d5a9956	116	dbs_info->freq_table = cpufreq_frequency_get_table(policy->cpu);
d1db75ff	117	dbs_info->freq_lo = 0;
05ca0350 AS	118	}
05ca0350 AS	119
3a3e9e06	120	static void dbs_freq_increase(struct cpufreq_policy *policy, unsigned int freq)
4471a34f	121	{
bc505475 RW	122	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	123	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42 VK	124	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	125
	126	if (od_tuners->powersave_bias)
3a3e9e06	127	freq = od_ops.powersave_bias_target(policy, freq,
fb30809e	128	CPUFREQ_RELATION_H);
3a3e9e06	129	else if (policy->cur == policy->max)
4471a34f	130	return;
0e625ac1	131
3a3e9e06	132	__cpufreq_driver_target(policy, freq, od_tuners->powersave_bias ?
4471a34f VK	133	CPUFREQ_RELATION_L : CPUFREQ_RELATION_H);
	134	}
	135
	136	/*
	137	* Every sampling_rate, we check, if current idle time is less than 20%
dfa5bb62 SK	138	* (default), then we try to increase frequency. Else, we adjust the frequency
dfa5bb62 SK	139	* proportional to load.
4471a34f	140	*/
4cccf755	141	static void od_update(struct cpufreq_policy *policy)
1da177e4	142	{
7d5a9956 RW	143	struct policy_dbs_info *policy_dbs = policy->governor_data;
7d5a9956 RW	144	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
bc505475	145	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42	146	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
4cccf755	147	unsigned int load = dbs_update(policy);
4471a34f VK	148
	149	dbs_info->freq_lo = 0;
	150
	151	/* Check for frequency increase */
ff4b1789	152	if (load > dbs_data->up_threshold) {
4471a34f VK	153	/* If switching to max speed, apply sampling_down_factor */
4471a34f VK	154	if (policy->cur < policy->max)
57dc3bcd	155	policy_dbs->rate_mult = dbs_data->sampling_down_factor;
4471a34f	156	dbs_freq_increase(policy, policy->max);
dfa5bb62 SK	157	} else {
dfa5bb62 SK	158	/* Calculate the next frequency proportional to load */
6393d6a1 SK	159	unsigned int freq_next, min_f, max_f;
	160
	161	min_f = policy->cpuinfo.min_freq;
	162	max_f = policy->cpuinfo.max_freq;
	163	freq_next = min_f + load * (max_f - min_f) / 100;
4471a34f VK	164
4471a34f VK	165	/* No longer fully busy, reset rate_mult */
57dc3bcd	166	policy_dbs->rate_mult = 1;
4471a34f	167
a7f35cff RW	168	if (od_tuners->powersave_bias)
	169	freq_next = od_ops.powersave_bias_target(policy,
	170	freq_next,
	171	CPUFREQ_RELATION_L);
	172
6393d6a1	173	__cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_C);
4471a34f	174	}
1da177e4 LT	175	}
1da177e4 LT	176
9be4fd2c	177	static unsigned int od_dbs_timer(struct cpufreq_policy *policy)
4471a34f	178	{
bc505475 RW	179	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	180	struct dbs_data *dbs_data = policy_dbs->dbs_data;
7d5a9956	181	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
6e96c5b3	182	int sample_type = dbs_info->sample_type;
4447266b	183
4471a34f	184	/* Common NORMAL_SAMPLE setup */
43e0ee36	185	dbs_info->sample_type = OD_NORMAL_SAMPLE;
4cccf755 RW	186	/*
	187	* OD_SUB_SAMPLE doesn't make sense if sample_delay_ns is 0, so ignore
	188	* it then.
	189	*/
	190	if (sample_type == OD_SUB_SAMPLE && policy_dbs->sample_delay_ns > 0) {
43e0ee36	191	__cpufreq_driver_target(policy, dbs_info->freq_lo,
42994af6	192	CPUFREQ_RELATION_H);
07aa4402	193	return dbs_info->freq_lo_delay_us;
6e96c5b3 RW	194	}
	195
	196	od_update(policy);
	197
	198	if (dbs_info->freq_lo) {
	199	/* Setup timer for SUB_SAMPLE */
	200	dbs_info->sample_type = OD_SUB_SAMPLE;
07aa4402	201	return dbs_info->freq_hi_delay_us;
4471a34f VK	202	}
4471a34f VK	203
07aa4402	204	return dbs_data->sampling_rate * policy_dbs->rate_mult;
da53d61e FB	205	}
da53d61e FB	206
4471a34f	207	/************************ sysfs interface **********************/
7bdad34d	208	static struct dbs_governor od_dbs_gov;
1da177e4	209
4d5dcc42 VK	210	static ssize_t store_io_is_busy(struct dbs_data dbs_data, const char buf,
4d5dcc42 VK	211	size_t count)
19379b11 AV	212	{
	213	unsigned int input;
	214	int ret;
	215
	216	ret = sscanf(buf, "%u", &input);
	217	if (ret != 1)
	218	return -EINVAL;
8847e038	219	dbs_data->io_is_busy = !!input;
9366d840 SK	220
9366d840 SK	221	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	222	gov_update_cpu_data(dbs_data);
a33cce1c	223
19379b11 AV	224	return count;
	225	}
	226
4d5dcc42 VK	227	static ssize_t store_up_threshold(struct dbs_data dbs_data, const char buf,
4d5dcc42 VK	228	size_t count)
1da177e4 LT	229	{
	230	unsigned int input;
	231	int ret;
ffac80e9	232	ret = sscanf(buf, "%u", &input);
1da177e4	233
32ee8c3e	234	if (ret != 1 \|\| input > MAX_FREQUENCY_UP_THRESHOLD \|\|
c29f1403	235	input < MIN_FREQUENCY_UP_THRESHOLD) {
1da177e4 LT	236	return -EINVAL;
1da177e4 LT	237	}
4bd4e428	238
ff4b1789	239	dbs_data->up_threshold = input;
1da177e4 LT	240	return count;
	241	}
	242
4d5dcc42 VK	243	static ssize_t store_sampling_down_factor(struct dbs_data *dbs_data,
4d5dcc42 VK	244	const char *buf, size_t count)
3f78a9f7	245	{
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 */
57dc3bcd RW	257	list_for_each_entry(policy_dbs, &dbs_data->policy_dbs_list, list) {
	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
6c4640c3 VK	270	static ssize_t store_ignore_nice_load(struct dbs_data *dbs_data,
6c4640c3 VK	271	const char *buf, size_t count)
3d5ee9e5 DJ	272	{
	273	unsigned int input;
	274	int ret;
	275
ffac80e9	276	ret = sscanf(buf, "%u", &input);
2b03f891	277	if (ret != 1)
3d5ee9e5 DJ	278	return -EINVAL;
3d5ee9e5 DJ	279
2b03f891	280	if (input > 1)
3d5ee9e5	281	input = 1;
32ee8c3e	282
ff4b1789	283	if (input == dbs_data->ignore_nice_load) { /* nothing to do */
3d5ee9e5 DJ	284	return count;
3d5ee9e5 DJ	285	}
ff4b1789	286	dbs_data->ignore_nice_load = input;
3d5ee9e5	287
ccb2fe20	288	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	289	gov_update_cpu_data(dbs_data);
1ca3abdb	290
3d5ee9e5 DJ	291	return count;
	292	}
	293
4d5dcc42 VK	294	static ssize_t store_powersave_bias(struct dbs_data dbs_data, const char buf,
4d5dcc42 VK	295	size_t count)
05ca0350	296	{
4d5dcc42	297	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
d1db75ff	298	struct policy_dbs_info *policy_dbs;
05ca0350 AS	299	unsigned int input;
	300	int ret;
	301	ret = sscanf(buf, "%u", &input);
	302
	303	if (ret != 1)
	304	return -EINVAL;
	305
	306	if (input > 1000)
	307	input = 1000;
	308
4d5dcc42	309	od_tuners->powersave_bias = input;
d1db75ff RW	310
	311	list_for_each_entry(policy_dbs, &dbs_data->policy_dbs_list, list)
	312	ondemand_powersave_bias_init(policy_dbs->policy);
	313
05ca0350 AS	314	return count;
	315	}
	316
c4435630 VK	317	gov_show_one_common(sampling_rate);
	318	gov_show_one_common(up_threshold);
	319	gov_show_one_common(sampling_down_factor);
	320	gov_show_one_common(ignore_nice_load);
	321	gov_show_one_common(min_sampling_rate);
8847e038	322	gov_show_one_common(io_is_busy);
c4435630 VK	323	gov_show_one(od, powersave_bias);
	324
	325	gov_attr_rw(sampling_rate);
	326	gov_attr_rw(io_is_busy);
	327	gov_attr_rw(up_threshold);
	328	gov_attr_rw(sampling_down_factor);
	329	gov_attr_rw(ignore_nice_load);
	330	gov_attr_rw(powersave_bias);
	331	gov_attr_ro(min_sampling_rate);
	332
	333	static struct attribute *od_attributes[] = {
	334	&min_sampling_rate.attr,
	335	&sampling_rate.attr,
	336	&up_threshold.attr,
	337	&sampling_down_factor.attr,
	338	&ignore_nice_load.attr,
	339	&powersave_bias.attr,
	340	&io_is_busy.attr,
1da177e4 LT	341	NULL
	342	};
	343
1da177e4 LT	344	/************************ sysfs end **********************/
1da177e4 LT	345
7d5a9956 RW	346	static struct policy_dbs_info *od_alloc(void)
	347	{
	348	struct od_policy_dbs_info *dbs_info;
	349
	350	dbs_info = kzalloc(sizeof(*dbs_info), GFP_KERNEL);
	351	return dbs_info ? &dbs_info->policy_dbs : NULL;
	352	}
	353
	354	static void od_free(struct policy_dbs_info *policy_dbs)
	355	{
	356	kfree(to_dbs_info(policy_dbs));
	357	}
	358
8e0484d2	359	static int od_init(struct dbs_data *dbs_data, bool notify)
4d5dcc42 VK	360	{
	361	struct od_dbs_tuners *tuners;
	362	u64 idle_time;
	363	int cpu;
	364
d5b73cd8	365	tuners = kzalloc(sizeof(*tuners), GFP_KERNEL);
4d5dcc42 VK	366	if (!tuners) {
	367	pr_err("%s: kzalloc failed\n", __func__);
	368	return -ENOMEM;
	369	}
	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
8e0484d2	400	static void od_exit(struct dbs_data *dbs_data, bool notify)
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 = {
af926185 RW	418	.gov = {
af926185 RW	419	.name = "ondemand",
906a6e5a	420	.governor = cpufreq_governor_dbs,
af926185 RW	421	.max_transition_latency = TRANSITION_LATENCY_LIMIT,
	422	.owner = THIS_MODULE,
	423	},
c4435630	424	.kobj_type = { .default_attrs = od_attributes },
4471a34f	425	.gov_dbs_timer = od_dbs_timer,
7d5a9956 RW	426	.alloc = od_alloc,
7d5a9956 RW	427	.free = od_free,
4d5dcc42 VK	428	.init = od_init,
4d5dcc42 VK	429	.exit = od_exit,
702c9e54	430	.start = od_start,
4471a34f	431	};
1da177e4	432
7bdad34d	433	#define CPU_FREQ_GOV_ONDEMAND (&od_dbs_gov.gov)
af926185	434
fb30809e JS	435	static void od_set_powersave_bias(unsigned int powersave_bias)
fb30809e JS	436	{
fb30809e JS	437	unsigned int cpu;
	438	cpumask_t done;
	439
c2837558	440	default_powersave_bias = powersave_bias;
fb30809e JS	441	cpumask_clear(&done);
	442
	443	get_online_cpus();
	444	for_each_online_cpu(cpu) {
8c8f77fd	445	struct cpufreq_policy *policy;
e40e7b25	446	struct policy_dbs_info *policy_dbs;
8c8f77fd RW	447	struct dbs_data *dbs_data;
8c8f77fd RW	448	struct od_dbs_tuners *od_tuners;
44152cb8	449
fb30809e JS	450	if (cpumask_test_cpu(cpu, &done))
	451	continue;
	452
8c8f77fd RW	453	policy = cpufreq_cpu_get_raw(cpu);
	454	if (!policy \|\| policy->governor != CPU_FREQ_GOV_ONDEMAND)
	455	continue;
	456
	457	policy_dbs = policy->governor_data;
e40e7b25	458	if (!policy_dbs)
c2837558	459	continue;
fb30809e JS	460
fb30809e JS	461	cpumask_or(&done, &done, policy->cpus);
c2837558	462
bc505475	463	dbs_data = policy_dbs->dbs_data;
c2837558 JS	464	od_tuners = dbs_data->tuners;
c2837558 JS	465	od_tuners->powersave_bias = default_powersave_bias;
fb30809e JS	466	}
	467	put_online_cpus();
	468	}
	469
	470	void od_register_powersave_bias_handler(unsigned int (*f)
	471	(struct cpufreq_policy *, unsigned int, unsigned int),
	472	unsigned int powersave_bias)
	473	{
	474	od_ops.powersave_bias_target = f;
	475	od_set_powersave_bias(powersave_bias);
	476	}
	477	EXPORT_SYMBOL_GPL(od_register_powersave_bias_handler);
	478
	479	void od_unregister_powersave_bias_handler(void)
	480	{
	481	od_ops.powersave_bias_target = generic_powersave_bias_target;
	482	od_set_powersave_bias(0);
	483	}
	484	EXPORT_SYMBOL_GPL(od_unregister_powersave_bias_handler);
	485
1da177e4 LT	486	static int __init cpufreq_gov_dbs_init(void)
1da177e4 LT	487	{
af926185	488	return cpufreq_register_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	489	}
	490
	491	static void __exit cpufreq_gov_dbs_exit(void)
	492	{
af926185	493	cpufreq_unregister_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	494	}
1da177e4 LT	495
ffac80e9 VP	496	MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>");
	497	MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>");
	498	MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
2b03f891	499	"Low Latency Frequency Transition capable processors");
ffac80e9	500	MODULE_LICENSE("GPL");
1da177e4	501
6915719b	502	#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
de1df26b RW	503	struct cpufreq_governor *cpufreq_default_governor(void)
de1df26b RW	504	{
af926185	505	return CPU_FREQ_GOV_ONDEMAND;
de1df26b RW	506	}
de1df26b RW	507
6915719b JW	508	fs_initcall(cpufreq_gov_dbs_init);
6915719b JW	509	#else
1da177e4	510	module_init(cpufreq_gov_dbs_init);
6915719b	511	#endif
1da177e4	512	module_exit(cpufreq_gov_dbs_exit);