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

// SPDX-License-Identifier: GPL-2.0-only
/*
 *  drivers/cpufreq/cpufreq_ondemand.c
 *
 *  Copyright (C)  2001 Russell King
 *            (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
 *                      Jun Nakajima <jun.nakajima@intel.com>
 */

#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 <linux/sched/cpufreq.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		(1)
#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_table_find_index_h(policy, freq_avg);
	freq_lo = freq_table[index].frequency;
	index = cpufreq_table_find_index_l(policy, freq_avg);
	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_update(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 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_update().
		 */
		mutex_lock(&policy_dbs->update_mutex);
		policy_dbs->rate_mult = 1;
		mutex_unlock(&policy_dbs->update_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(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);

static struct attribute *od_attributes[] = {
	&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;
	} else {
		dbs_data->up_threshold = DEF_FREQUENCY_UP_THRESHOLD;
	}

	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_update = od_dbs_update,
	.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;
}

core_initcall(cpufreq_gov_dbs_init);
#else
module_init(cpufreq_gov_dbs_init);
#endif
module_exit(cpufreq_gov_dbs_exit);
Commit	Line	Data
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
1da177e4 LT	2	/*
	3	* drivers/cpufreq/cpufreq_ondemand.c
	4	*
	5	* Copyright (C) 2001 Russell King
	6	* (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
	7	* Jun Nakajima <jun.nakajima@intel.com>
1da177e4 LT	8	*/
1da177e4 LT	9
4471a34f VK	10	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
4471a34f VK	11
5ff0a268	12	#include <linux/cpu.h>
4471a34f	13	#include <linux/percpu-defs.h>
4d5dcc42	14	#include <linux/slab.h>
80800913	15	#include <linux/tick.h>
55687da1	16	#include <linux/sched/cpufreq.h>
7d5a9956 RW	17
7d5a9956 RW	18	#include "cpufreq_ondemand.h"
1da177e4	19
06eb09d1	20	/* On-demand governor macros */
1da177e4	21	#define DEF_FREQUENCY_UP_THRESHOLD (80)
3f78a9f7 DN	22	#define DEF_SAMPLING_DOWN_FACTOR (1)
3f78a9f7 DN	23	#define MAX_SAMPLING_DOWN_FACTOR (100000)
80800913	24	#define MICRO_FREQUENCY_UP_THRESHOLD (95)
cef9615a	25	#define MICRO_FREQUENCY_MIN_SAMPLE_RATE (10000)
4dd63b49	26	#define MIN_FREQUENCY_UP_THRESHOLD (1)
1da177e4 LT	27	#define MAX_FREQUENCY_UP_THRESHOLD (100)
1da177e4 LT	28
fb30809e JS	29	static struct od_ops od_ops;
fb30809e JS	30
c2837558 JS	31	static unsigned int default_powersave_bias;
c2837558 JS	32
4471a34f VK	33	/*
	34	* Not all CPUs want IO time to be accounted as busy; this depends on how
	35	* efficient idling at a higher frequency/voltage is.
	36	* Pavel Machek says this is not so for various generations of AMD and old
	37	* Intel systems.
06eb09d1	38	* Mike Chan (android.com) claims this is also not true for ARM.
4471a34f VK	39	* Because of this, whitelist specific known (series) of CPUs by default, and
	40	* leave all others up to the user.
	41	*/
	42	static int should_io_be_busy(void)
	43	{
	44	#if defined(CONFIG_X86)
	45	/*
06eb09d1	46	* For Intel, Core 2 (model 15) and later have an efficient idle.
4471a34f VK	47	*/
	48	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
	49	boot_cpu_data.x86 == 6 &&
	50	boot_cpu_data.x86_model >= 15)
	51	return 1;
	52	#endif
	53	return 0;
6b8fcd90 AV	54	}
6b8fcd90 AV	55
05ca0350 AS	56	/*
05ca0350 AS	57	* Find right freq to be set now with powersave_bias on.
07aa4402 RW	58	* Returns the freq_hi to be used right now and will set freq_hi_delay_us,
07aa4402 RW	59	* freq_lo, and freq_lo_delay_us in percpu area for averaging freqs.
05ca0350	60	*/
fb30809e	61	static unsigned int generic_powersave_bias_target(struct cpufreq_policy *policy,
4471a34f	62	unsigned int freq_next, unsigned int relation)
05ca0350 AS	63	{
	64	unsigned int freq_req, freq_reduc, freq_avg;
	65	unsigned int freq_hi, freq_lo;
d218ed77	66	unsigned int index;
07aa4402	67	unsigned int delay_hi_us;
bc505475	68	struct policy_dbs_info *policy_dbs = policy->governor_data;
7d5a9956	69	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
bc505475	70	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42	71	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
34ac5d7a	72	struct cpufreq_frequency_table *freq_table = policy->freq_table;
05ca0350	73
34ac5d7a	74	if (!freq_table) {
05ca0350	75	dbs_info->freq_lo = 0;
07aa4402	76	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	77	return freq_next;
	78	}
	79
d218ed77	80	index = cpufreq_frequency_table_target(policy, freq_next, relation);
34ac5d7a	81	freq_req = freq_table[index].frequency;
4d5dcc42	82	freq_reduc = freq_req * od_tuners->powersave_bias / 1000;
05ca0350 AS	83	freq_avg = freq_req - freq_reduc;
	84
	85	/* Find freq bounds for freq_avg in freq_table */
82577360	86	index = cpufreq_table_find_index_h(policy, freq_avg);
34ac5d7a	87	freq_lo = freq_table[index].frequency;
82577360	88	index = cpufreq_table_find_index_l(policy, freq_avg);
34ac5d7a	89	freq_hi = freq_table[index].frequency;
05ca0350 AS	90
	91	/* Find out how long we have to be in hi and lo freqs */
	92	if (freq_hi == freq_lo) {
	93	dbs_info->freq_lo = 0;
07aa4402	94	dbs_info->freq_lo_delay_us = 0;
05ca0350 AS	95	return freq_lo;
05ca0350 AS	96	}
07aa4402 RW	97	delay_hi_us = (freq_avg - freq_lo) * dbs_data->sampling_rate;
	98	delay_hi_us += (freq_hi - freq_lo) / 2;
	99	delay_hi_us /= freq_hi - freq_lo;
	100	dbs_info->freq_hi_delay_us = delay_hi_us;
05ca0350	101	dbs_info->freq_lo = freq_lo;
07aa4402	102	dbs_info->freq_lo_delay_us = dbs_data->sampling_rate - delay_hi_us;
05ca0350 AS	103	return freq_hi;
	104	}
	105
d1db75ff	106	static void ondemand_powersave_bias_init(struct cpufreq_policy *policy)
05ca0350	107	{
7d5a9956	108	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);
d1db75ff	109
d1db75ff	110	dbs_info->freq_lo = 0;
05ca0350 AS	111	}
05ca0350 AS	112
3a3e9e06	113	static void dbs_freq_increase(struct cpufreq_policy *policy, unsigned int freq)
4471a34f	114	{
bc505475 RW	115	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	116	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42 VK	117	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	118
	119	if (od_tuners->powersave_bias)
3a3e9e06	120	freq = od_ops.powersave_bias_target(policy, freq,
fb30809e	121	CPUFREQ_RELATION_H);
3a3e9e06	122	else if (policy->cur == policy->max)
4471a34f	123	return;
0e625ac1	124
3a3e9e06	125	__cpufreq_driver_target(policy, freq, od_tuners->powersave_bias ?
4471a34f VK	126	CPUFREQ_RELATION_L : CPUFREQ_RELATION_H);
	127	}
	128
	129	/*
	130	* Every sampling_rate, we check, if current idle time is less than 20%
dfa5bb62 SK	131	* (default), then we try to increase frequency. Else, we adjust the frequency
dfa5bb62 SK	132	* proportional to load.
4471a34f	133	*/
4cccf755	134	static void od_update(struct cpufreq_policy *policy)
1da177e4	135	{
7d5a9956 RW	136	struct policy_dbs_info *policy_dbs = policy->governor_data;
7d5a9956 RW	137	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
bc505475	138	struct dbs_data *dbs_data = policy_dbs->dbs_data;
4d5dcc42	139	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
4cccf755	140	unsigned int load = dbs_update(policy);
4471a34f VK	141
	142	dbs_info->freq_lo = 0;
	143
	144	/* Check for frequency increase */
ff4b1789	145	if (load > dbs_data->up_threshold) {
4471a34f VK	146	/* If switching to max speed, apply sampling_down_factor */
4471a34f VK	147	if (policy->cur < policy->max)
57dc3bcd	148	policy_dbs->rate_mult = dbs_data->sampling_down_factor;
4471a34f	149	dbs_freq_increase(policy, policy->max);
dfa5bb62 SK	150	} else {
dfa5bb62 SK	151	/* Calculate the next frequency proportional to load */
6393d6a1 SK	152	unsigned int freq_next, min_f, max_f;
	153
	154	min_f = policy->cpuinfo.min_freq;
	155	max_f = policy->cpuinfo.max_freq;
	156	freq_next = min_f + load * (max_f - min_f) / 100;
4471a34f VK	157
4471a34f VK	158	/* No longer fully busy, reset rate_mult */
57dc3bcd	159	policy_dbs->rate_mult = 1;
4471a34f	160
a7f35cff RW	161	if (od_tuners->powersave_bias)
	162	freq_next = od_ops.powersave_bias_target(policy,
	163	freq_next,
	164	CPUFREQ_RELATION_L);
	165
6393d6a1	166	__cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_C);
4471a34f	167	}
1da177e4 LT	168	}
1da177e4 LT	169
26f0dbc9	170	static unsigned int od_dbs_update(struct cpufreq_policy *policy)
4471a34f	171	{
bc505475 RW	172	struct policy_dbs_info *policy_dbs = policy->governor_data;
bc505475 RW	173	struct dbs_data *dbs_data = policy_dbs->dbs_data;
7d5a9956	174	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy_dbs);
6e96c5b3	175	int sample_type = dbs_info->sample_type;
4447266b	176
4471a34f	177	/* Common NORMAL_SAMPLE setup */
43e0ee36	178	dbs_info->sample_type = OD_NORMAL_SAMPLE;
4cccf755 RW	179	/*
	180	* OD_SUB_SAMPLE doesn't make sense if sample_delay_ns is 0, so ignore
	181	* it then.
	182	*/
	183	if (sample_type == OD_SUB_SAMPLE && policy_dbs->sample_delay_ns > 0) {
43e0ee36	184	__cpufreq_driver_target(policy, dbs_info->freq_lo,
42994af6	185	CPUFREQ_RELATION_H);
07aa4402	186	return dbs_info->freq_lo_delay_us;
6e96c5b3 RW	187	}
	188
	189	od_update(policy);
	190
	191	if (dbs_info->freq_lo) {
26f0dbc9	192	/* Setup SUB_SAMPLE */
6e96c5b3	193	dbs_info->sample_type = OD_SUB_SAMPLE;
07aa4402	194	return dbs_info->freq_hi_delay_us;
4471a34f VK	195	}
4471a34f VK	196
07aa4402	197	return dbs_data->sampling_rate * policy_dbs->rate_mult;
da53d61e FB	198	}
da53d61e FB	199
4471a34f	200	/************************ sysfs interface **********************/
7bdad34d	201	static struct dbs_governor od_dbs_gov;
1da177e4	202
0dd3c1d6 RW	203	static ssize_t store_io_is_busy(struct gov_attr_set attr_set, const char buf,
0dd3c1d6 RW	204	size_t count)
19379b11	205	{
0dd3c1d6	206	struct dbs_data *dbs_data = to_dbs_data(attr_set);
19379b11 AV	207	unsigned int input;
	208	int ret;
	209
	210	ret = sscanf(buf, "%u", &input);
	211	if (ret != 1)
	212	return -EINVAL;
8847e038	213	dbs_data->io_is_busy = !!input;
9366d840 SK	214
9366d840 SK	215	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	216	gov_update_cpu_data(dbs_data);
a33cce1c	217
19379b11 AV	218	return count;
	219	}
	220
0dd3c1d6 RW	221	static ssize_t store_up_threshold(struct gov_attr_set *attr_set,
0dd3c1d6 RW	222	const char *buf, size_t count)
1da177e4	223	{
0dd3c1d6	224	struct dbs_data *dbs_data = to_dbs_data(attr_set);
1da177e4 LT	225	unsigned int input;
1da177e4 LT	226	int ret;
ffac80e9	227	ret = sscanf(buf, "%u", &input);
1da177e4	228
32ee8c3e	229	if (ret != 1 \|\| input > MAX_FREQUENCY_UP_THRESHOLD \|\|
c29f1403	230	input < MIN_FREQUENCY_UP_THRESHOLD) {
1da177e4 LT	231	return -EINVAL;
1da177e4 LT	232	}
4bd4e428	233
ff4b1789	234	dbs_data->up_threshold = input;
1da177e4 LT	235	return count;
	236	}
	237
0dd3c1d6 RW	238	static ssize_t store_sampling_down_factor(struct gov_attr_set *attr_set,
0dd3c1d6 RW	239	const char *buf, size_t count)
3f78a9f7	240	{
0dd3c1d6	241	struct dbs_data *dbs_data = to_dbs_data(attr_set);
57dc3bcd RW	242	struct policy_dbs_info *policy_dbs;
57dc3bcd RW	243	unsigned int input;
3f78a9f7 DN	244	int ret;
	245	ret = sscanf(buf, "%u", &input);
	246
	247	if (ret != 1 \|\| input > MAX_SAMPLING_DOWN_FACTOR \|\| input < 1)
	248	return -EINVAL;
57dc3bcd	249
ff4b1789	250	dbs_data->sampling_down_factor = input;
3f78a9f7 DN	251
3f78a9f7 DN	252	/* Reset down sampling multiplier in case it was active */
0dd3c1d6	253	list_for_each_entry(policy_dbs, &attr_set->policy_list, list) {
57dc3bcd RW	254	/*
57dc3bcd RW	255	* Doing this without locking might lead to using different
26f0dbc9	256	* rate_mult values in od_update() and od_dbs_update().
57dc3bcd	257	*/
26f0dbc9	258	mutex_lock(&policy_dbs->update_mutex);
57dc3bcd	259	policy_dbs->rate_mult = 1;
26f0dbc9	260	mutex_unlock(&policy_dbs->update_mutex);
3f78a9f7	261	}
57dc3bcd	262
3f78a9f7 DN	263	return count;
	264	}
	265
0dd3c1d6 RW	266	static ssize_t store_ignore_nice_load(struct gov_attr_set *attr_set,
0dd3c1d6 RW	267	const char *buf, size_t count)
3d5ee9e5	268	{
0dd3c1d6	269	struct dbs_data *dbs_data = to_dbs_data(attr_set);
3d5ee9e5 DJ	270	unsigned int input;
	271	int ret;
	272
ffac80e9	273	ret = sscanf(buf, "%u", &input);
2b03f891	274	if (ret != 1)
3d5ee9e5 DJ	275	return -EINVAL;
3d5ee9e5 DJ	276
2b03f891	277	if (input > 1)
3d5ee9e5	278	input = 1;
32ee8c3e	279
ff4b1789	280	if (input == dbs_data->ignore_nice_load) { /* nothing to do */
3d5ee9e5 DJ	281	return count;
3d5ee9e5 DJ	282	}
ff4b1789	283	dbs_data->ignore_nice_load = input;
3d5ee9e5	284
ccb2fe20	285	/* we need to re-evaluate prev_cpu_idle */
8c8f77fd	286	gov_update_cpu_data(dbs_data);
1ca3abdb	287
3d5ee9e5 DJ	288	return count;
	289	}
	290
0dd3c1d6 RW	291	static ssize_t store_powersave_bias(struct gov_attr_set *attr_set,
0dd3c1d6 RW	292	const char *buf, size_t count)
05ca0350	293	{
0dd3c1d6	294	struct dbs_data *dbs_data = to_dbs_data(attr_set);
4d5dcc42	295	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
d1db75ff	296	struct policy_dbs_info *policy_dbs;
05ca0350 AS	297	unsigned int input;
	298	int ret;
	299	ret = sscanf(buf, "%u", &input);
	300
	301	if (ret != 1)
	302	return -EINVAL;
	303
	304	if (input > 1000)
	305	input = 1000;
	306
4d5dcc42	307	od_tuners->powersave_bias = input;
d1db75ff	308
0dd3c1d6	309	list_for_each_entry(policy_dbs, &attr_set->policy_list, list)
d1db75ff RW	310	ondemand_powersave_bias_init(policy_dbs->policy);
d1db75ff RW	311
05ca0350 AS	312	return count;
	313	}
	314
c4435630 VK	315	gov_show_one_common(sampling_rate);
	316	gov_show_one_common(up_threshold);
	317	gov_show_one_common(sampling_down_factor);
	318	gov_show_one_common(ignore_nice_load);
8847e038	319	gov_show_one_common(io_is_busy);
c4435630 VK	320	gov_show_one(od, powersave_bias);
	321
	322	gov_attr_rw(sampling_rate);
	323	gov_attr_rw(io_is_busy);
	324	gov_attr_rw(up_threshold);
	325	gov_attr_rw(sampling_down_factor);
	326	gov_attr_rw(ignore_nice_load);
	327	gov_attr_rw(powersave_bias);
c4435630 VK	328
c4435630 VK	329	static struct attribute *od_attributes[] = {
c4435630 VK	330	&sampling_rate.attr,
	331	&up_threshold.attr,
	332	&sampling_down_factor.attr,
	333	&ignore_nice_load.attr,
	334	&powersave_bias.attr,
	335	&io_is_busy.attr,
1da177e4 LT	336	NULL
	337	};
	338
1da177e4 LT	339	/************************ sysfs end **********************/
1da177e4 LT	340
7d5a9956 RW	341	static struct policy_dbs_info *od_alloc(void)
	342	{
	343	struct od_policy_dbs_info *dbs_info;
	344
	345	dbs_info = kzalloc(sizeof(*dbs_info), GFP_KERNEL);
	346	return dbs_info ? &dbs_info->policy_dbs : NULL;
	347	}
	348
	349	static void od_free(struct policy_dbs_info *policy_dbs)
	350	{
	351	kfree(to_dbs_info(policy_dbs));
	352	}
	353
9a15fb2c	354	static int od_init(struct dbs_data *dbs_data)
4d5dcc42 VK	355	{
	356	struct od_dbs_tuners *tuners;
	357	u64 idle_time;
	358	int cpu;
	359
d5b73cd8	360	tuners = kzalloc(sizeof(*tuners), GFP_KERNEL);
a69d6b29	361	if (!tuners)
4d5dcc42	362	return -ENOMEM;
4d5dcc42 VK	363
	364	cpu = get_cpu();
	365	idle_time = get_cpu_idle_time_us(cpu, NULL);
	366	put_cpu();
	367	if (idle_time != -1ULL) {
	368	/* Idle micro accounting is supported. Use finer thresholds */
ff4b1789	369	dbs_data->up_threshold = MICRO_FREQUENCY_UP_THRESHOLD;
4d5dcc42	370	} else {
ff4b1789	371	dbs_data->up_threshold = DEF_FREQUENCY_UP_THRESHOLD;
4d5dcc42 VK	372	}
4d5dcc42 VK	373
ff4b1789 VK	374	dbs_data->sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR;
ff4b1789 VK	375	dbs_data->ignore_nice_load = 0;
c2837558	376	tuners->powersave_bias = default_powersave_bias;
8847e038	377	dbs_data->io_is_busy = should_io_be_busy();
4d5dcc42 VK	378
4d5dcc42 VK	379	dbs_data->tuners = tuners;
4d5dcc42 VK	380	return 0;
	381	}
	382
9a15fb2c	383	static void od_exit(struct dbs_data *dbs_data)
4d5dcc42 VK	384	{
	385	kfree(dbs_data->tuners);
	386	}
	387
702c9e54 RW	388	static void od_start(struct cpufreq_policy *policy)
702c9e54 RW	389	{
7d5a9956	390	struct od_policy_dbs_info *dbs_info = to_dbs_info(policy->governor_data);
702c9e54 RW	391
702c9e54 RW	392	dbs_info->sample_type = OD_NORMAL_SAMPLE;
d1db75ff	393	ondemand_powersave_bias_init(policy);
702c9e54 RW	394	}
702c9e54 RW	395
4471a34f	396	static struct od_ops od_ops = {
fb30809e	397	.powersave_bias_target = generic_powersave_bias_target,
4471a34f	398	};
2f8a835c	399
7bdad34d	400	static struct dbs_governor od_dbs_gov = {
e788892b	401	.gov = CPUFREQ_DBS_GOVERNOR_INITIALIZER("ondemand"),
c4435630	402	.kobj_type = { .default_attrs = od_attributes },
26f0dbc9	403	.gov_dbs_update = od_dbs_update,
7d5a9956 RW	404	.alloc = od_alloc,
7d5a9956 RW	405	.free = od_free,
4d5dcc42 VK	406	.init = od_init,
4d5dcc42 VK	407	.exit = od_exit,
702c9e54	408	.start = od_start,
4471a34f	409	};
1da177e4	410
7bdad34d	411	#define CPU_FREQ_GOV_ONDEMAND (&od_dbs_gov.gov)
af926185	412
fb30809e JS	413	static void od_set_powersave_bias(unsigned int powersave_bias)
fb30809e JS	414	{
fb30809e JS	415	unsigned int cpu;
	416	cpumask_t done;
	417
c2837558	418	default_powersave_bias = powersave_bias;
fb30809e JS	419	cpumask_clear(&done);
	420
	421	get_online_cpus();
	422	for_each_online_cpu(cpu) {
8c8f77fd	423	struct cpufreq_policy *policy;
e40e7b25	424	struct policy_dbs_info *policy_dbs;
8c8f77fd RW	425	struct dbs_data *dbs_data;
8c8f77fd RW	426	struct od_dbs_tuners *od_tuners;
44152cb8	427
fb30809e JS	428	if (cpumask_test_cpu(cpu, &done))
	429	continue;
	430
8c8f77fd RW	431	policy = cpufreq_cpu_get_raw(cpu);
	432	if (!policy \|\| policy->governor != CPU_FREQ_GOV_ONDEMAND)
	433	continue;
	434
	435	policy_dbs = policy->governor_data;
e40e7b25	436	if (!policy_dbs)
c2837558	437	continue;
fb30809e JS	438
fb30809e JS	439	cpumask_or(&done, &done, policy->cpus);
c2837558	440
bc505475	441	dbs_data = policy_dbs->dbs_data;
c2837558 JS	442	od_tuners = dbs_data->tuners;
c2837558 JS	443	od_tuners->powersave_bias = default_powersave_bias;
fb30809e JS	444	}
	445	put_online_cpus();
	446	}
	447
	448	void od_register_powersave_bias_handler(unsigned int (*f)
	449	(struct cpufreq_policy *, unsigned int, unsigned int),
	450	unsigned int powersave_bias)
	451	{
	452	od_ops.powersave_bias_target = f;
	453	od_set_powersave_bias(powersave_bias);
	454	}
	455	EXPORT_SYMBOL_GPL(od_register_powersave_bias_handler);
	456
	457	void od_unregister_powersave_bias_handler(void)
	458	{
	459	od_ops.powersave_bias_target = generic_powersave_bias_target;
	460	od_set_powersave_bias(0);
	461	}
	462	EXPORT_SYMBOL_GPL(od_unregister_powersave_bias_handler);
	463
1da177e4 LT	464	static int __init cpufreq_gov_dbs_init(void)
1da177e4 LT	465	{
af926185	466	return cpufreq_register_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	467	}
	468
	469	static void __exit cpufreq_gov_dbs_exit(void)
	470	{
af926185	471	cpufreq_unregister_governor(CPU_FREQ_GOV_ONDEMAND);
1da177e4 LT	472	}
1da177e4 LT	473
ffac80e9 VP	474	MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>");
	475	MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>");
	476	MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
2b03f891	477	"Low Latency Frequency Transition capable processors");
ffac80e9	478	MODULE_LICENSE("GPL");
1da177e4	479
6915719b	480	#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
de1df26b RW	481	struct cpufreq_governor *cpufreq_default_governor(void)
de1df26b RW	482	{
af926185	483	return CPU_FREQ_GOV_ONDEMAND;
de1df26b RW	484	}
de1df26b RW	485
3f6ec871	486	core_initcall(cpufreq_gov_dbs_init);
6915719b	487	#else
1da177e4	488	module_init(cpufreq_gov_dbs_init);
6915719b	489	#endif
1da177e4	490	module_exit(cpufreq_gov_dbs_exit);