[linux-block.git] / drivers / cpufreq / cpufreq_governor.c

/*
 * drivers/cpufreq/cpufreq_governor.c
 *
 * CPUFREQ governors common code
 *
 * Copyright	(C) 2001 Russell King
 *		(C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
 *		(C) 2003 Jun Nakajima <jun.nakajima@intel.com>
 *		(C) 2009 Alexander Clouter <alex@digriz.org.uk>
 *		(c) 2012 Viresh Kumar <viresh.kumar@linaro.org>
 *
 * 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 <asm/cputime.h>
#include <linux/cpufreq.h>
#include <linux/cpumask.h>
#include <linux/export.h>
#include <linux/kernel_stat.h>
#include <linux/mutex.h>
#include <linux/slab.h>
#include <linux/types.h>
#include <linux/workqueue.h>

#include "cpufreq_governor.h"

static struct attribute_group *get_sysfs_attr(struct dbs_data *dbs_data)
{
	if (have_governor_per_policy())
		return dbs_data->cdata->attr_group_gov_pol;
	else
		return dbs_data->cdata->attr_group_gov_sys;
}

void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
{
	struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);
	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
	struct cpufreq_policy *policy;
	unsigned int max_load = 0;
	unsigned int ignore_nice;
	unsigned int j;

	if (dbs_data->cdata->governor == GOV_ONDEMAND)
		ignore_nice = od_tuners->ignore_nice;
	else
		ignore_nice = cs_tuners->ignore_nice;

	policy = cdbs->cur_policy;

	/* Get Absolute Load (in terms of freq for ondemand gov) */
	for_each_cpu(j, policy->cpus) {
		struct cpu_dbs_common_info *j_cdbs;
		u64 cur_wall_time, cur_idle_time;
		unsigned int idle_time, wall_time;
		unsigned int load;
		int io_busy = 0;

		j_cdbs = dbs_data->cdata->get_cpu_cdbs(j);

		/*
		 * For the purpose of ondemand, waiting for disk IO is
		 * an indication that you're performance critical, and
		 * not that the system is actually idle. So do not add
		 * the iowait time to the cpu idle time.
		 */
		if (dbs_data->cdata->governor == GOV_ONDEMAND)
			io_busy = od_tuners->io_is_busy;
		cur_idle_time = get_cpu_idle_time(j, &cur_wall_time, io_busy);

		wall_time = (unsigned int)
			(cur_wall_time - j_cdbs->prev_cpu_wall);
		j_cdbs->prev_cpu_wall = cur_wall_time;

		idle_time = (unsigned int)
			(cur_idle_time - j_cdbs->prev_cpu_idle);
		j_cdbs->prev_cpu_idle = cur_idle_time;

		if (ignore_nice) {
			u64 cur_nice;
			unsigned long cur_nice_jiffies;

			cur_nice = kcpustat_cpu(j).cpustat[CPUTIME_NICE] -
					 cdbs->prev_cpu_nice;
			/*
			 * Assumption: nice time between sampling periods will
			 * be less than 2^32 jiffies for 32 bit sys
			 */
			cur_nice_jiffies = (unsigned long)
					cputime64_to_jiffies64(cur_nice);

			cdbs->prev_cpu_nice =
				kcpustat_cpu(j).cpustat[CPUTIME_NICE];
			idle_time += jiffies_to_usecs(cur_nice_jiffies);
		}

		if (unlikely(!wall_time || wall_time < idle_time))
			continue;

		load = 100 * (wall_time - idle_time) / wall_time;

		if (dbs_data->cdata->governor == GOV_ONDEMAND) {
			int freq_avg = __cpufreq_driver_getavg(policy, j);
			if (freq_avg <= 0)
				freq_avg = policy->cur;

			load *= freq_avg;
		}

		if (load > max_load)
			max_load = load;
	}

	dbs_data->cdata->gov_check_cpu(cpu, max_load);
}
EXPORT_SYMBOL_GPL(dbs_check_cpu);

static inline void __gov_queue_work(int cpu, struct dbs_data *dbs_data,
		unsigned int delay)
{
	struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);

	mod_delayed_work_on(cpu, system_wq, &cdbs->work, delay);
}

void gov_queue_work(struct dbs_data *dbs_data, struct cpufreq_policy *policy,
		unsigned int delay, bool all_cpus)
{
	int i;

	if (!all_cpus) {
		__gov_queue_work(smp_processor_id(), dbs_data, delay);
	} else {
		for_each_cpu(i, policy->cpus)
			__gov_queue_work(i, dbs_data, delay);
	}
}
EXPORT_SYMBOL_GPL(gov_queue_work);

static inline void gov_cancel_work(struct dbs_data *dbs_data,
		struct cpufreq_policy *policy)
{
	struct cpu_dbs_common_info *cdbs;
	int i;

	for_each_cpu(i, policy->cpus) {
		cdbs = dbs_data->cdata->get_cpu_cdbs(i);
		cancel_delayed_work_sync(&cdbs->work);
	}
}

/* Will return if we need to evaluate cpu load again or not */
bool need_load_eval(struct cpu_dbs_common_info *cdbs,
		unsigned int sampling_rate)
{
	if (policy_is_shared(cdbs->cur_policy)) {
		ktime_t time_now = ktime_get();
		s64 delta_us = ktime_us_delta(time_now, cdbs->time_stamp);

		/* Do nothing if we recently have sampled */
		if (delta_us < (s64)(sampling_rate / 2))
			return false;
		else
			cdbs->time_stamp = time_now;
	}

	return true;
}
EXPORT_SYMBOL_GPL(need_load_eval);

static void set_sampling_rate(struct dbs_data *dbs_data,
		unsigned int sampling_rate)
{
	if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
		struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
		cs_tuners->sampling_rate = sampling_rate;
	} else {
		struct od_dbs_tuners *od_tuners = dbs_data->tuners;
		od_tuners->sampling_rate = sampling_rate;
	}
}

int cpufreq_governor_dbs(struct cpufreq_policy *policy,
		struct common_dbs_data *cdata, unsigned int event)
{
	struct dbs_data *dbs_data;
	struct od_cpu_dbs_info_s *od_dbs_info = NULL;
	struct cs_cpu_dbs_info_s *cs_dbs_info = NULL;
	struct od_ops *od_ops = NULL;
	struct od_dbs_tuners *od_tuners = NULL;
	struct cs_dbs_tuners *cs_tuners = NULL;
	struct cpu_dbs_common_info *cpu_cdbs;
	unsigned int sampling_rate, latency, ignore_nice, j, cpu = policy->cpu;
	int io_busy = 0;
	int rc;

	if (have_governor_per_policy())
		dbs_data = policy->governor_data;
	else
		dbs_data = cdata->gdbs_data;

	WARN_ON(!dbs_data && (event != CPUFREQ_GOV_POLICY_INIT));

	switch (event) {
	case CPUFREQ_GOV_POLICY_INIT:
		if (have_governor_per_policy()) {
			WARN_ON(dbs_data);
		} else if (dbs_data) {
			dbs_data->usage_count++;
			policy->governor_data = dbs_data;
			return 0;
		}

		dbs_data = kzalloc(sizeof(*dbs_data), GFP_KERNEL);
		if (!dbs_data) {
			pr_err("%s: POLICY_INIT: kzalloc failed\n", __func__);
			return -ENOMEM;
		}

		dbs_data->cdata = cdata;
		dbs_data->usage_count = 1;
		rc = cdata->init(dbs_data);
		if (rc) {
			pr_err("%s: POLICY_INIT: init() failed\n", __func__);
			kfree(dbs_data);
			return rc;
		}

		if (!have_governor_per_policy())
			WARN_ON(cpufreq_get_global_kobject());

		rc = sysfs_create_group(get_governor_parent_kobj(policy),
				get_sysfs_attr(dbs_data));
		if (rc) {
			cdata->exit(dbs_data);
			kfree(dbs_data);
			return rc;
		}

		policy->governor_data = dbs_data;

		/* policy latency is in nS. Convert it to uS first */
		latency = policy->cpuinfo.transition_latency / 1000;
		if (latency == 0)
			latency = 1;

		/* Bring kernel and HW constraints together */
		dbs_data->min_sampling_rate = max(dbs_data->min_sampling_rate,
				MIN_LATENCY_MULTIPLIER * latency);
		set_sampling_rate(dbs_data, max(dbs_data->min_sampling_rate,
					latency * LATENCY_MULTIPLIER));

		if ((cdata->governor == GOV_CONSERVATIVE) &&
				(!policy->governor->initialized)) {
			struct cs_ops *cs_ops = dbs_data->cdata->gov_ops;

			cpufreq_register_notifier(cs_ops->notifier_block,
					CPUFREQ_TRANSITION_NOTIFIER);
		}

		if (!have_governor_per_policy())
			cdata->gdbs_data = dbs_data;

		return 0;
	case CPUFREQ_GOV_POLICY_EXIT:
		if (!--dbs_data->usage_count) {
			sysfs_remove_group(get_governor_parent_kobj(policy),
					get_sysfs_attr(dbs_data));

			if (!have_governor_per_policy())
				cpufreq_put_global_kobject();

			if ((dbs_data->cdata->governor == GOV_CONSERVATIVE) &&
				(policy->governor->initialized == 1)) {
				struct cs_ops *cs_ops = dbs_data->cdata->gov_ops;

				cpufreq_unregister_notifier(cs_ops->notifier_block,
						CPUFREQ_TRANSITION_NOTIFIER);
			}

			cdata->exit(dbs_data);
			kfree(dbs_data);
			cdata->gdbs_data = NULL;
		}

		policy->governor_data = NULL;
		return 0;
	}

	cpu_cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);

	if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
		cs_tuners = dbs_data->tuners;
		cs_dbs_info = dbs_data->cdata->get_cpu_dbs_info_s(cpu);
		sampling_rate = cs_tuners->sampling_rate;
		ignore_nice = cs_tuners->ignore_nice;
	} else {
		od_tuners = dbs_data->tuners;
		od_dbs_info = dbs_data->cdata->get_cpu_dbs_info_s(cpu);
		sampling_rate = od_tuners->sampling_rate;
		ignore_nice = od_tuners->ignore_nice;
		od_ops = dbs_data->cdata->gov_ops;
		io_busy = od_tuners->io_is_busy;
	}

	switch (event) {
	case CPUFREQ_GOV_START:
		if (!policy->cur)
			return -EINVAL;

		mutex_lock(&dbs_data->mutex);

		for_each_cpu(j, policy->cpus) {
			struct cpu_dbs_common_info *j_cdbs =
				dbs_data->cdata->get_cpu_cdbs(j);

			j_cdbs->cpu = j;
			j_cdbs->cur_policy = policy;
			j_cdbs->prev_cpu_idle = get_cpu_idle_time(j,
					       &j_cdbs->prev_cpu_wall, io_busy);
			if (ignore_nice)
				j_cdbs->prev_cpu_nice =
					kcpustat_cpu(j).cpustat[CPUTIME_NICE];

			mutex_init(&j_cdbs->timer_mutex);
			INIT_DEFERRABLE_WORK(&j_cdbs->work,
					     dbs_data->cdata->gov_dbs_timer);
		}

		/*
		 * conservative does not implement micro like ondemand
		 * governor, thus we are bound to jiffes/HZ
		 */
		if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
			cs_dbs_info->down_skip = 0;
			cs_dbs_info->enable = 1;
			cs_dbs_info->requested_freq = policy->cur;
		} else {
			od_dbs_info->rate_mult = 1;
			od_dbs_info->sample_type = OD_NORMAL_SAMPLE;
			od_ops->powersave_bias_init_cpu(cpu);
		}

		mutex_unlock(&dbs_data->mutex);

		/* Initiate timer time stamp */
		cpu_cdbs->time_stamp = ktime_get();

		gov_queue_work(dbs_data, policy,
				delay_for_sampling_rate(sampling_rate), true);
		break;

	case CPUFREQ_GOV_STOP:
		if (dbs_data->cdata->governor == GOV_CONSERVATIVE)
			cs_dbs_info->enable = 0;

		gov_cancel_work(dbs_data, policy);

		mutex_lock(&dbs_data->mutex);
		mutex_destroy(&cpu_cdbs->timer_mutex);
		cpu_cdbs->cur_policy = NULL;

		mutex_unlock(&dbs_data->mutex);

		break;

	case CPUFREQ_GOV_LIMITS:
		mutex_lock(&cpu_cdbs->timer_mutex);
		if (policy->max < cpu_cdbs->cur_policy->cur)
			__cpufreq_driver_target(cpu_cdbs->cur_policy,
					policy->max, CPUFREQ_RELATION_H);
		else if (policy->min > cpu_cdbs->cur_policy->cur)
			__cpufreq_driver_target(cpu_cdbs->cur_policy,
					policy->min, CPUFREQ_RELATION_L);
		dbs_check_cpu(dbs_data, cpu);
		mutex_unlock(&cpu_cdbs->timer_mutex);
		break;
	}
	return 0;
}
EXPORT_SYMBOL_GPL(cpufreq_governor_dbs);
Commit	Line	Data
2aacdfff	1	/*
	2	* drivers/cpufreq/cpufreq_governor.c
	3	*
	4	* CPUFREQ governors common code
	5	*
4471a34f VK	6	* Copyright (C) 2001 Russell King
	7	* (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
	8	* (C) 2003 Jun Nakajima <jun.nakajima@intel.com>
	9	* (C) 2009 Alexander Clouter <alex@digriz.org.uk>
	10	* (c) 2012 Viresh Kumar <viresh.kumar@linaro.org>
	11	*
2aacdfff	12	* This program is free software; you can redistribute it and/or modify
	13	* it under the terms of the GNU General Public License version 2 as
	14	* published by the Free Software Foundation.
	15	*/
	16
4471a34f VK	17	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
4471a34f VK	18
2aacdfff	19	#include <asm/cputime.h>
4471a34f VK	20	#include <linux/cpufreq.h>
4471a34f VK	21	#include <linux/cpumask.h>
2aacdfff	22	#include <linux/export.h>
2aacdfff	23	#include <linux/kernel_stat.h>
4471a34f	24	#include <linux/mutex.h>
4d5dcc42	25	#include <linux/slab.h>
2aacdfff	26	#include <linux/types.h>
4471a34f VK	27	#include <linux/workqueue.h>
	28
	29	#include "cpufreq_governor.h"
	30
4d5dcc42 VK	31	static struct attribute_group get_sysfs_attr(struct dbs_data dbs_data)
	32	{
	33	if (have_governor_per_policy())
	34	return dbs_data->cdata->attr_group_gov_pol;
	35	else
	36	return dbs_data->cdata->attr_group_gov_sys;
	37	}
	38
4471a34f VK	39	void dbs_check_cpu(struct dbs_data *dbs_data, int cpu)
4471a34f VK	40	{
4d5dcc42	41	struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);
4471a34f VK	42	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	43	struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
	44	struct cpufreq_policy *policy;
	45	unsigned int max_load = 0;
	46	unsigned int ignore_nice;
	47	unsigned int j;
	48
4d5dcc42	49	if (dbs_data->cdata->governor == GOV_ONDEMAND)
4471a34f VK	50	ignore_nice = od_tuners->ignore_nice;
	51	else
	52	ignore_nice = cs_tuners->ignore_nice;
	53
	54	policy = cdbs->cur_policy;
	55
	56	/* Get Absolute Load (in terms of freq for ondemand gov) */
	57	for_each_cpu(j, policy->cpus) {
	58	struct cpu_dbs_common_info *j_cdbs;
9366d840 SK	59	u64 cur_wall_time, cur_idle_time;
9366d840 SK	60	unsigned int idle_time, wall_time;
4471a34f	61	unsigned int load;
9366d840	62	int io_busy = 0;
4471a34f	63
4d5dcc42	64	j_cdbs = dbs_data->cdata->get_cpu_cdbs(j);
4471a34f	65
9366d840 SK	66	/*
	67	* For the purpose of ondemand, waiting for disk IO is
	68	* an indication that you're performance critical, and
	69	* not that the system is actually idle. So do not add
	70	* the iowait time to the cpu idle time.
	71	*/
	72	if (dbs_data->cdata->governor == GOV_ONDEMAND)
	73	io_busy = od_tuners->io_is_busy;
	74	cur_idle_time = get_cpu_idle_time(j, &cur_wall_time, io_busy);
4471a34f VK	75
	76	wall_time = (unsigned int)
	77	(cur_wall_time - j_cdbs->prev_cpu_wall);
	78	j_cdbs->prev_cpu_wall = cur_wall_time;
	79
	80	idle_time = (unsigned int)
	81	(cur_idle_time - j_cdbs->prev_cpu_idle);
	82	j_cdbs->prev_cpu_idle = cur_idle_time;
	83
	84	if (ignore_nice) {
	85	u64 cur_nice;
	86	unsigned long cur_nice_jiffies;
	87
	88	cur_nice = kcpustat_cpu(j).cpustat[CPUTIME_NICE] -
	89	cdbs->prev_cpu_nice;
	90	/*
	91	* Assumption: nice time between sampling periods will
	92	* be less than 2^32 jiffies for 32 bit sys
	93	*/
	94	cur_nice_jiffies = (unsigned long)
	95	cputime64_to_jiffies64(cur_nice);
	96
	97	cdbs->prev_cpu_nice =
	98	kcpustat_cpu(j).cpustat[CPUTIME_NICE];
	99	idle_time += jiffies_to_usecs(cur_nice_jiffies);
	100	}
	101
4471a34f VK	102	if (unlikely(!wall_time \|\| wall_time < idle_time))
	103	continue;
	104
	105	load = 100 * (wall_time - idle_time) / wall_time;
	106
4d5dcc42	107	if (dbs_data->cdata->governor == GOV_ONDEMAND) {
4471a34f VK	108	int freq_avg = __cpufreq_driver_getavg(policy, j);
	109	if (freq_avg <= 0)
	110	freq_avg = policy->cur;
	111
	112	load *= freq_avg;
	113	}
	114
	115	if (load > max_load)
	116	max_load = load;
	117	}
	118
4d5dcc42	119	dbs_data->cdata->gov_check_cpu(cpu, max_load);
4471a34f VK	120	}
	121	EXPORT_SYMBOL_GPL(dbs_check_cpu);
	122
031299b3 VK	123	static inline void __gov_queue_work(int cpu, struct dbs_data *dbs_data,
031299b3 VK	124	unsigned int delay)
4471a34f	125	{
4d5dcc42	126	struct cpu_dbs_common_info *cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);
4471a34f	127
031299b3	128	mod_delayed_work_on(cpu, system_wq, &cdbs->work, delay);
4471a34f VK	129	}
4471a34f VK	130
031299b3 VK	131	void gov_queue_work(struct dbs_data dbs_data, struct cpufreq_policy policy,
031299b3 VK	132	unsigned int delay, bool all_cpus)
4471a34f	133	{
031299b3 VK	134	int i;
	135
	136	if (!all_cpus) {
	137	__gov_queue_work(smp_processor_id(), dbs_data, delay);
	138	} else {
	139	for_each_cpu(i, policy->cpus)
	140	__gov_queue_work(i, dbs_data, delay);
	141	}
	142	}
	143	EXPORT_SYMBOL_GPL(gov_queue_work);
	144
	145	static inline void gov_cancel_work(struct dbs_data *dbs_data,
	146	struct cpufreq_policy *policy)
	147	{
	148	struct cpu_dbs_common_info *cdbs;
	149	int i;
58ddcead	150
031299b3 VK	151	for_each_cpu(i, policy->cpus) {
	152	cdbs = dbs_data->cdata->get_cpu_cdbs(i);
	153	cancel_delayed_work_sync(&cdbs->work);
	154	}
4471a34f VK	155	}
4471a34f VK	156
4447266b VK	157	/* Will return if we need to evaluate cpu load again or not */
	158	bool need_load_eval(struct cpu_dbs_common_info *cdbs,
	159	unsigned int sampling_rate)
	160	{
	161	if (policy_is_shared(cdbs->cur_policy)) {
	162	ktime_t time_now = ktime_get();
	163	s64 delta_us = ktime_us_delta(time_now, cdbs->time_stamp);
	164
	165	/* Do nothing if we recently have sampled */
	166	if (delta_us < (s64)(sampling_rate / 2))
	167	return false;
	168	else
	169	cdbs->time_stamp = time_now;
	170	}
	171
	172	return true;
	173	}
	174	EXPORT_SYMBOL_GPL(need_load_eval);
	175
4d5dcc42 VK	176	static void set_sampling_rate(struct dbs_data *dbs_data,
	177	unsigned int sampling_rate)
	178	{
	179	if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
	180	struct cs_dbs_tuners *cs_tuners = dbs_data->tuners;
	181	cs_tuners->sampling_rate = sampling_rate;
	182	} else {
	183	struct od_dbs_tuners *od_tuners = dbs_data->tuners;
	184	od_tuners->sampling_rate = sampling_rate;
	185	}
	186	}
	187
	188	int cpufreq_governor_dbs(struct cpufreq_policy *policy,
	189	struct common_dbs_data *cdata, unsigned int event)
4471a34f	190	{
4d5dcc42	191	struct dbs_data *dbs_data;
4471a34f VK	192	struct od_cpu_dbs_info_s *od_dbs_info = NULL;
4471a34f VK	193	struct cs_cpu_dbs_info_s *cs_dbs_info = NULL;
8eeed095	194	struct od_ops *od_ops = NULL;
4d5dcc42 VK	195	struct od_dbs_tuners *od_tuners = NULL;
4d5dcc42 VK	196	struct cs_dbs_tuners *cs_tuners = NULL;
4471a34f	197	struct cpu_dbs_common_info *cpu_cdbs;
4d5dcc42	198	unsigned int sampling_rate, latency, ignore_nice, j, cpu = policy->cpu;
9366d840	199	int io_busy = 0;
4471a34f VK	200	int rc;
4471a34f VK	201
4d5dcc42 VK	202	if (have_governor_per_policy())
	203	dbs_data = policy->governor_data;
	204	else
	205	dbs_data = cdata->gdbs_data;
	206
	207	WARN_ON(!dbs_data && (event != CPUFREQ_GOV_POLICY_INIT));
	208
	209	switch (event) {
	210	case CPUFREQ_GOV_POLICY_INIT:
	211	if (have_governor_per_policy()) {
	212	WARN_ON(dbs_data);
	213	} else if (dbs_data) {
a97c98ad	214	dbs_data->usage_count++;
4d5dcc42 VK	215	policy->governor_data = dbs_data;
	216	return 0;
	217	}
	218
	219	dbs_data = kzalloc(sizeof(*dbs_data), GFP_KERNEL);
	220	if (!dbs_data) {
	221	pr_err("%s: POLICY_INIT: kzalloc failed\n", __func__);
	222	return -ENOMEM;
	223	}
	224
	225	dbs_data->cdata = cdata;
a97c98ad	226	dbs_data->usage_count = 1;
4d5dcc42 VK	227	rc = cdata->init(dbs_data);
	228	if (rc) {
	229	pr_err("%s: POLICY_INIT: init() failed\n", __func__);
	230	kfree(dbs_data);
	231	return rc;
	232	}
	233
2361be23 VK	234	if (!have_governor_per_policy())
	235	WARN_ON(cpufreq_get_global_kobject());
	236
4d5dcc42 VK	237	rc = sysfs_create_group(get_governor_parent_kobj(policy),
	238	get_sysfs_attr(dbs_data));
	239	if (rc) {
	240	cdata->exit(dbs_data);
	241	kfree(dbs_data);
	242	return rc;
	243	}
	244
	245	policy->governor_data = dbs_data;
	246
	247	/* policy latency is in nS. Convert it to uS first */
	248	latency = policy->cpuinfo.transition_latency / 1000;
	249	if (latency == 0)
	250	latency = 1;
	251
	252	/* Bring kernel and HW constraints together */
	253	dbs_data->min_sampling_rate = max(dbs_data->min_sampling_rate,
	254	MIN_LATENCY_MULTIPLIER * latency);
	255	set_sampling_rate(dbs_data, max(dbs_data->min_sampling_rate,
	256	latency * LATENCY_MULTIPLIER));
	257
a97c98ad VK	258	if ((cdata->governor == GOV_CONSERVATIVE) &&
a97c98ad VK	259	(!policy->governor->initialized)) {
4d5dcc42 VK	260	struct cs_ops *cs_ops = dbs_data->cdata->gov_ops;
	261
	262	cpufreq_register_notifier(cs_ops->notifier_block,
	263	CPUFREQ_TRANSITION_NOTIFIER);
	264	}
	265
	266	if (!have_governor_per_policy())
	267	cdata->gdbs_data = dbs_data;
	268
	269	return 0;
	270	case CPUFREQ_GOV_POLICY_EXIT:
a97c98ad	271	if (!--dbs_data->usage_count) {
4d5dcc42 VK	272	sysfs_remove_group(get_governor_parent_kobj(policy),
	273	get_sysfs_attr(dbs_data));
	274
2361be23 VK	275	if (!have_governor_per_policy())
	276	cpufreq_put_global_kobject();
	277
a97c98ad VK	278	if ((dbs_data->cdata->governor == GOV_CONSERVATIVE) &&
a97c98ad VK	279	(policy->governor->initialized == 1)) {
4d5dcc42 VK	280	struct cs_ops *cs_ops = dbs_data->cdata->gov_ops;
	281
	282	cpufreq_unregister_notifier(cs_ops->notifier_block,
	283	CPUFREQ_TRANSITION_NOTIFIER);
	284	}
	285
	286	cdata->exit(dbs_data);
	287	kfree(dbs_data);
	288	cdata->gdbs_data = NULL;
	289	}
4471a34f	290
4d5dcc42 VK	291	policy->governor_data = NULL;
	292	return 0;
	293	}
	294
	295	cpu_cdbs = dbs_data->cdata->get_cpu_cdbs(cpu);
	296
	297	if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
	298	cs_tuners = dbs_data->tuners;
	299	cs_dbs_info = dbs_data->cdata->get_cpu_dbs_info_s(cpu);
	300	sampling_rate = cs_tuners->sampling_rate;
4471a34f VK	301	ignore_nice = cs_tuners->ignore_nice;
4471a34f VK	302	} else {
4d5dcc42 VK	303	od_tuners = dbs_data->tuners;
	304	od_dbs_info = dbs_data->cdata->get_cpu_dbs_info_s(cpu);
	305	sampling_rate = od_tuners->sampling_rate;
4471a34f	306	ignore_nice = od_tuners->ignore_nice;
4d5dcc42	307	od_ops = dbs_data->cdata->gov_ops;
9366d840	308	io_busy = od_tuners->io_is_busy;
4471a34f VK	309	}
	310
	311	switch (event) {
	312	case CPUFREQ_GOV_START:
3361b7b1	313	if (!policy->cur)
4471a34f VK	314	return -EINVAL;
	315
	316	mutex_lock(&dbs_data->mutex);
	317
4471a34f	318	for_each_cpu(j, policy->cpus) {
8eeed095	319	struct cpu_dbs_common_info *j_cdbs =
4d5dcc42	320	dbs_data->cdata->get_cpu_cdbs(j);
4471a34f	321
09dca5ae	322	j_cdbs->cpu = j;
4471a34f VK	323	j_cdbs->cur_policy = policy;
4471a34f VK	324	j_cdbs->prev_cpu_idle = get_cpu_idle_time(j,
9366d840	325	&j_cdbs->prev_cpu_wall, io_busy);
4471a34f VK	326	if (ignore_nice)
	327	j_cdbs->prev_cpu_nice =
	328	kcpustat_cpu(j).cpustat[CPUTIME_NICE];
2abfa876 RA	329
	330	mutex_init(&j_cdbs->timer_mutex);
	331	INIT_DEFERRABLE_WORK(&j_cdbs->work,
4d5dcc42	332	dbs_data->cdata->gov_dbs_timer);
4471a34f VK	333	}
4471a34f VK	334
4471a34f VK	335	/*
	336	* conservative does not implement micro like ondemand
	337	* governor, thus we are bound to jiffes/HZ
	338	*/
4d5dcc42	339	if (dbs_data->cdata->governor == GOV_CONSERVATIVE) {
8eeed095 VK	340	cs_dbs_info->down_skip = 0;
	341	cs_dbs_info->enable = 1;
	342	cs_dbs_info->requested_freq = policy->cur;
4471a34f	343	} else {
8eeed095 VK	344	od_dbs_info->rate_mult = 1;
	345	od_dbs_info->sample_type = OD_NORMAL_SAMPLE;
	346	od_ops->powersave_bias_init_cpu(cpu);
4471a34f VK	347	}
4471a34f VK	348
4471a34f VK	349	mutex_unlock(&dbs_data->mutex);
4471a34f VK	350
58ddcead FB	351	/* Initiate timer time stamp */
58ddcead FB	352	cpu_cdbs->time_stamp = ktime_get();
da53d61e	353
031299b3 VK	354	gov_queue_work(dbs_data, policy,
031299b3 VK	355	delay_for_sampling_rate(sampling_rate), true);
4471a34f VK	356	break;
	357
	358	case CPUFREQ_GOV_STOP:
4d5dcc42	359	if (dbs_data->cdata->governor == GOV_CONSERVATIVE)
4471a34f VK	360	cs_dbs_info->enable = 0;
4471a34f VK	361
031299b3	362	gov_cancel_work(dbs_data, policy);
4471a34f VK	363
	364	mutex_lock(&dbs_data->mutex);
	365	mutex_destroy(&cpu_cdbs->timer_mutex);
419e1721	366	cpu_cdbs->cur_policy = NULL;
8eeed095	367
4471a34f VK	368	mutex_unlock(&dbs_data->mutex);
	369
	370	break;
	371
	372	case CPUFREQ_GOV_LIMITS:
	373	mutex_lock(&cpu_cdbs->timer_mutex);
	374	if (policy->max < cpu_cdbs->cur_policy->cur)
	375	__cpufreq_driver_target(cpu_cdbs->cur_policy,
	376	policy->max, CPUFREQ_RELATION_H);
	377	else if (policy->min > cpu_cdbs->cur_policy->cur)
	378	__cpufreq_driver_target(cpu_cdbs->cur_policy,
	379	policy->min, CPUFREQ_RELATION_L);
	380	dbs_check_cpu(dbs_data, cpu);
	381	mutex_unlock(&cpu_cdbs->timer_mutex);
	382	break;
	383	}
	384	return 0;
	385	}
	386	EXPORT_SYMBOL_GPL(cpufreq_governor_dbs);