cpufreq_conservative.c 19.8 KB
Newer Older
1 2 3 4 5 6
/*
 *  drivers/cpufreq/cpufreq_conservative.c
 *
 *  Copyright (C)  2001 Russell King
 *            (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
 *                      Jun Nakajima <jun.nakajima@intel.com>
7
 *            (C)  2009 Alexander Clouter <alex@digriz.org.uk>
8 9 10 11 12 13 14 15 16 17
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/cpufreq.h>
A
Andrew Morton 已提交
18
#include <linux/cpu.h>
19 20
#include <linux/jiffies.h>
#include <linux/kernel_stat.h>
21
#include <linux/mutex.h>
22 23 24 25 26
#include <linux/hrtimer.h>
#include <linux/tick.h>
#include <linux/ktime.h>
#include <linux/sched.h>

27 28 29 30 31 32 33 34
/*
 * dbs is used in this file as a shortform for demandbased switching
 * It helps to keep variable names smaller, simpler
 */

#define DEF_FREQUENCY_UP_THRESHOLD		(80)
#define DEF_FREQUENCY_DOWN_THRESHOLD		(20)

35 36
/*
 * The polling frequency of this governor depends on the capability of
37
 * the processor. Default polling frequency is 1000 times the transition
38 39
 * latency of the processor. The governor will work on any processor with
 * transition latency <= 10mS, using appropriate sampling
40
 * rate.
41 42
 * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL)
 * this governor will not work.
43 44
 * All times here are in uS.
 */
45
#define MIN_SAMPLING_RATE_RATIO			(2)
46

47 48
static unsigned int min_sampling_rate;

49
#define LATENCY_MULTIPLIER			(1000)
50
#define MIN_LATENCY_MULTIPLIER			(100)
51 52
#define DEF_SAMPLING_DOWN_FACTOR		(1)
#define MAX_SAMPLING_DOWN_FACTOR		(10)
53
#define TRANSITION_LATENCY_LIMIT		(10 * 1000 * 1000)
54

D
David Howells 已提交
55
static void do_dbs_timer(struct work_struct *work);
56 57

struct cpu_dbs_info_s {
58 59 60
	cputime64_t prev_cpu_idle;
	cputime64_t prev_cpu_wall;
	cputime64_t prev_cpu_nice;
61
	struct cpufreq_policy *cur_policy;
62
	struct delayed_work work;
63 64
	unsigned int down_skip;
	unsigned int requested_freq;
65 66
	int cpu;
	unsigned int enable:1;
67 68 69 70 71 72
	/*
	 * percpu mutex that serializes governor limit change with
	 * do_dbs_timer invocation. We do not want do_dbs_timer to run
	 * when user is changing the governor or limits.
	 */
	struct mutex timer_mutex;
73
};
74
static DEFINE_PER_CPU(struct cpu_dbs_info_s, cs_cpu_dbs_info);
75 76 77

static unsigned int dbs_enable;	/* number of CPUs using this policy */

78
/*
79
 * dbs_mutex protects data in dbs_tuners_ins from concurrent changes on
80
 * different CPUs. It protects dbs_enable in governor start/stop.
81
 */
82
static DEFINE_MUTEX(dbs_mutex);
83

84 85 86
static struct workqueue_struct	*kconservative_wq;

static struct dbs_tuners {
87 88 89 90 91 92
	unsigned int sampling_rate;
	unsigned int sampling_down_factor;
	unsigned int up_threshold;
	unsigned int down_threshold;
	unsigned int ignore_nice;
	unsigned int freq_step;
93
} dbs_tuners_ins = {
94 95 96 97 98
	.up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
	.down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD,
	.sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR,
	.ignore_nice = 0,
	.freq_step = 5,
99 100
};

101 102
static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu,
							cputime64_t *wall)
103
{
104 105 106 107 108 109 110
	cputime64_t idle_time;
	cputime64_t cur_wall_time;
	cputime64_t busy_time;

	cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
	busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user,
			kstat_cpu(cpu).cpustat.system);
111

112 113 114 115
	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq);
	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq);
	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal);
	busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice);
116

117 118
	idle_time = cputime64_sub(cur_wall_time, busy_time);
	if (wall)
119
		*wall = (cputime64_t)jiffies_to_usecs(cur_wall_time);
120

121
	return (cputime64_t)jiffies_to_usecs(idle_time);;
122 123 124 125 126 127 128 129 130 131
}

static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
{
	u64 idle_time = get_cpu_idle_time_us(cpu, wall);

	if (idle_time == -1ULL)
		return get_cpu_idle_time_jiffy(cpu, wall);

	return idle_time;
132 133
}

134 135 136 137 138 139
/* keep track of frequency transitions */
static int
dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
		     void *data)
{
	struct cpufreq_freqs *freq = data;
140
	struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cs_cpu_dbs_info,
141 142
							freq->cpu);

143 144
	struct cpufreq_policy *policy;

145 146 147
	if (!this_dbs_info->enable)
		return 0;

148 149 150 151 152 153 154 155 156 157
	policy = this_dbs_info->cur_policy;

	/*
	 * we only care if our internally tracked freq moves outside
	 * the 'valid' ranges of freqency available to us otherwise
	 * we do not change it
	*/
	if (this_dbs_info->requested_freq > policy->max
			|| this_dbs_info->requested_freq < policy->min)
		this_dbs_info->requested_freq = freq->new;
158 159 160 161 162 163 164 165

	return 0;
}

static struct notifier_block dbs_cpufreq_notifier_block = {
	.notifier_call = dbs_cpufreq_notifier
};

166
/************************** sysfs interface ************************/
167 168
static ssize_t show_sampling_rate_max(struct kobject *kobj,
				      struct attribute *attr, char *buf)
169
{
170 171
	printk_once(KERN_INFO "CPUFREQ: conservative sampling_rate_max "
		    "sysfs file is deprecated - used by: %s\n", current->comm);
172
	return sprintf(buf, "%u\n", -1U);
173 174
}

175 176
static ssize_t show_sampling_rate_min(struct kobject *kobj,
				      struct attribute *attr, char *buf)
177
{
178
	return sprintf(buf, "%u\n", min_sampling_rate);
179 180
}

181 182
define_one_global_ro(sampling_rate_max);
define_one_global_ro(sampling_rate_min);
183 184 185 186

/* cpufreq_conservative Governor Tunables */
#define show_one(file_name, object)					\
static ssize_t show_##file_name						\
187
(struct kobject *kobj, struct attribute *attr, char *buf)		\
188 189 190 191 192 193 194
{									\
	return sprintf(buf, "%u\n", dbs_tuners_ins.object);		\
}
show_one(sampling_rate, sampling_rate);
show_one(sampling_down_factor, sampling_down_factor);
show_one(up_threshold, up_threshold);
show_one(down_threshold, down_threshold);
195
show_one(ignore_nice_load, ignore_nice);
196 197
show_one(freq_step, freq_step);

198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
/*** delete after deprecation time ***/
#define DEPRECATION_MSG(file_name)					\
	printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs "	\
		"interface is deprecated - " #file_name "\n");

#define show_one_old(file_name)						\
static ssize_t show_##file_name##_old					\
(struct cpufreq_policy *unused, char *buf)				\
{									\
	printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs "	\
		"interface is deprecated - " #file_name "\n");		\
	return show_##file_name(NULL, NULL, buf);			\
}
show_one_old(sampling_rate);
show_one_old(sampling_down_factor);
show_one_old(up_threshold);
show_one_old(down_threshold);
show_one_old(ignore_nice_load);
show_one_old(freq_step);
show_one_old(sampling_rate_min);
show_one_old(sampling_rate_max);

220 221
cpufreq_freq_attr_ro_old(sampling_rate_min);
cpufreq_freq_attr_ro_old(sampling_rate_max);
222 223 224 225 226 227

/*** delete after deprecation time ***/

static ssize_t store_sampling_down_factor(struct kobject *a,
					  struct attribute *b,
					  const char *buf, size_t count)
228 229 230
{
	unsigned int input;
	int ret;
231
	ret = sscanf(buf, "%u", &input);
232

233
	if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
234 235
		return -EINVAL;

236
	mutex_lock(&dbs_mutex);
237
	dbs_tuners_ins.sampling_down_factor = input;
238
	mutex_unlock(&dbs_mutex);
239 240 241 242

	return count;
}

243 244
static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
				   const char *buf, size_t count)
245 246 247
{
	unsigned int input;
	int ret;
248
	ret = sscanf(buf, "%u", &input);
249

250
	if (ret != 1)
251
		return -EINVAL;
252 253

	mutex_lock(&dbs_mutex);
254
	dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate);
255
	mutex_unlock(&dbs_mutex);
256 257 258 259

	return count;
}

260 261
static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
				  const char *buf, size_t count)
262 263 264
{
	unsigned int input;
	int ret;
265
	ret = sscanf(buf, "%u", &input);
266

267
	mutex_lock(&dbs_mutex);
268
	if (ret != 1 || input > 100 ||
269
			input <= dbs_tuners_ins.down_threshold) {
270
		mutex_unlock(&dbs_mutex);
271 272 273 274
		return -EINVAL;
	}

	dbs_tuners_ins.up_threshold = input;
275
	mutex_unlock(&dbs_mutex);
276 277 278 279

	return count;
}

280 281
static ssize_t store_down_threshold(struct kobject *a, struct attribute *b,
				    const char *buf, size_t count)
282 283 284
{
	unsigned int input;
	int ret;
285
	ret = sscanf(buf, "%u", &input);
286

287
	mutex_lock(&dbs_mutex);
288 289 290
	/* cannot be lower than 11 otherwise freq will not fall */
	if (ret != 1 || input < 11 || input > 100 ||
			input >= dbs_tuners_ins.up_threshold) {
291
		mutex_unlock(&dbs_mutex);
292 293 294 295
		return -EINVAL;
	}

	dbs_tuners_ins.down_threshold = input;
296
	mutex_unlock(&dbs_mutex);
297 298 299 300

	return count;
}

301 302
static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
				      const char *buf, size_t count)
303 304 305 306 307
{
	unsigned int input;
	int ret;

	unsigned int j;
308 309 310

	ret = sscanf(buf, "%u", &input);
	if (ret != 1)
311 312
		return -EINVAL;

313
	if (input > 1)
314
		input = 1;
315

316
	mutex_lock(&dbs_mutex);
317
	if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
318
		mutex_unlock(&dbs_mutex);
319 320 321 322
		return count;
	}
	dbs_tuners_ins.ignore_nice = input;

323
	/* we need to re-evaluate prev_cpu_idle */
324
	for_each_online_cpu(j) {
325
		struct cpu_dbs_info_s *dbs_info;
326
		dbs_info = &per_cpu(cs_cpu_dbs_info, j);
327 328 329 330
		dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
						&dbs_info->prev_cpu_wall);
		if (dbs_tuners_ins.ignore_nice)
			dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
331
	}
332
	mutex_unlock(&dbs_mutex);
333 334 335 336

	return count;
}

337 338
static ssize_t store_freq_step(struct kobject *a, struct attribute *b,
			       const char *buf, size_t count)
339 340 341
{
	unsigned int input;
	int ret;
342
	ret = sscanf(buf, "%u", &input);
343

344
	if (ret != 1)
345 346
		return -EINVAL;

347
	if (input > 100)
348
		input = 100;
349

350 351
	/* no need to test here if freq_step is zero as the user might actually
	 * want this, they would be crazy though :) */
352
	mutex_lock(&dbs_mutex);
353
	dbs_tuners_ins.freq_step = input;
354
	mutex_unlock(&dbs_mutex);
355 356 357 358

	return count;
}

359 360 361 362 363 364
define_one_global_rw(sampling_rate);
define_one_global_rw(sampling_down_factor);
define_one_global_rw(up_threshold);
define_one_global_rw(down_threshold);
define_one_global_rw(ignore_nice_load);
define_one_global_rw(freq_step);
365

366
static struct attribute *dbs_attributes[] = {
367 368 369 370 371 372
	&sampling_rate_max.attr,
	&sampling_rate_min.attr,
	&sampling_rate.attr,
	&sampling_down_factor.attr,
	&up_threshold.attr,
	&down_threshold.attr,
373
	&ignore_nice_load.attr,
374 375 376 377 378 379 380 381 382
	&freq_step.attr,
	NULL
};

static struct attribute_group dbs_attr_group = {
	.attrs = dbs_attributes,
	.name = "conservative",
};

383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
/*** delete after deprecation time ***/

#define write_one_old(file_name)					\
static ssize_t store_##file_name##_old					\
(struct cpufreq_policy *unused, const char *buf, size_t count)		\
{									\
	printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs "	\
		"interface is deprecated - " #file_name "\n");	\
	return store_##file_name(NULL, NULL, buf, count);		\
}
write_one_old(sampling_rate);
write_one_old(sampling_down_factor);
write_one_old(up_threshold);
write_one_old(down_threshold);
write_one_old(ignore_nice_load);
write_one_old(freq_step);

400 401 402 403 404 405
cpufreq_freq_attr_rw_old(sampling_rate);
cpufreq_freq_attr_rw_old(sampling_down_factor);
cpufreq_freq_attr_rw_old(up_threshold);
cpufreq_freq_attr_rw_old(down_threshold);
cpufreq_freq_attr_rw_old(ignore_nice_load);
cpufreq_freq_attr_rw_old(freq_step);
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425

static struct attribute *dbs_attributes_old[] = {
	&sampling_rate_max_old.attr,
	&sampling_rate_min_old.attr,
	&sampling_rate_old.attr,
	&sampling_down_factor_old.attr,
	&up_threshold_old.attr,
	&down_threshold_old.attr,
	&ignore_nice_load_old.attr,
	&freq_step_old.attr,
	NULL
};

static struct attribute_group dbs_attr_group_old = {
	.attrs = dbs_attributes_old,
	.name = "conservative",
};

/*** delete after deprecation time ***/

426 427
/************************** sysfs end ************************/

428
static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
429
{
430
	unsigned int load = 0;
431
	unsigned int max_load = 0;
432
	unsigned int freq_target;
433

434 435
	struct cpufreq_policy *policy;
	unsigned int j;
436

437 438
	policy = this_dbs_info->cur_policy;

439
	/*
440 441 442 443
	 * Every sampling_rate, we check, if current idle time is less
	 * than 20% (default), then we try to increase frequency
	 * Every sampling_rate*sampling_down_factor, we check, if current
	 * idle time is more than 80%, then we try to decrease frequency
444
	 *
445 446
	 * Any frequency increase takes it to the maximum frequency.
	 * Frequency reduction happens at minimum steps of
447
	 * 5% (default) of maximum frequency
448 449
	 */

450 451 452 453 454
	/* Get Absolute Load */
	for_each_cpu(j, policy->cpus) {
		struct cpu_dbs_info_s *j_dbs_info;
		cputime64_t cur_wall_time, cur_idle_time;
		unsigned int idle_time, wall_time;
455

456
		j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
457 458 459 460 461 462

		cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);

		wall_time = (unsigned int) cputime64_sub(cur_wall_time,
				j_dbs_info->prev_cpu_wall);
		j_dbs_info->prev_cpu_wall = cur_wall_time;
463

464 465 466
		idle_time = (unsigned int) cputime64_sub(cur_idle_time,
				j_dbs_info->prev_cpu_idle);
		j_dbs_info->prev_cpu_idle = cur_idle_time;
467

468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
		if (dbs_tuners_ins.ignore_nice) {
			cputime64_t cur_nice;
			unsigned long cur_nice_jiffies;

			cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice,
					 j_dbs_info->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);

			j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.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;
489 490 491

		if (load > max_load)
			max_load = load;
492 493 494 495 496 497 498 499
	}

	/*
	 * break out if we 'cannot' reduce the speed as the user might
	 * want freq_step to be zero
	 */
	if (dbs_tuners_ins.freq_step == 0)
		return;
500

501
	/* Check for frequency increase */
502
	if (max_load > dbs_tuners_ins.up_threshold) {
503
		this_dbs_info->down_skip = 0;
504

505
		/* if we are already at full speed then break out early */
506
		if (this_dbs_info->requested_freq == policy->max)
507
			return;
508

509
		freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
510 511

		/* max freq cannot be less than 100. But who knows.... */
512 513
		if (unlikely(freq_target == 0))
			freq_target = 5;
514

515
		this_dbs_info->requested_freq += freq_target;
516 517
		if (this_dbs_info->requested_freq > policy->max)
			this_dbs_info->requested_freq = policy->max;
518

519
		__cpufreq_driver_target(policy, this_dbs_info->requested_freq,
520 521 522 523
			CPUFREQ_RELATION_H);
		return;
	}

524 525 526 527 528
	/*
	 * The optimal frequency is the frequency that is the lowest that
	 * can support the current CPU usage without triggering the up
	 * policy. To be safe, we focus 10 points under the threshold.
	 */
529
	if (max_load < (dbs_tuners_ins.down_threshold - 10)) {
530
		freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
531

532
		this_dbs_info->requested_freq -= freq_target;
533 534
		if (this_dbs_info->requested_freq < policy->min)
			this_dbs_info->requested_freq = policy->min;
535

536 537 538 539 540 541
		/*
		 * if we cannot reduce the frequency anymore, break out early
		 */
		if (policy->cur == policy->min)
			return;

542
		__cpufreq_driver_target(policy, this_dbs_info->requested_freq,
543
				CPUFREQ_RELATION_H);
544 545 546 547
		return;
	}
}

D
David Howells 已提交
548
static void do_dbs_timer(struct work_struct *work)
549
{
550 551 552 553 554 555 556 557 558
	struct cpu_dbs_info_s *dbs_info =
		container_of(work, struct cpu_dbs_info_s, work.work);
	unsigned int cpu = dbs_info->cpu;

	/* We want all CPUs to do sampling nearly on same jiffy */
	int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);

	delay -= jiffies % delay;

559
	mutex_lock(&dbs_info->timer_mutex);
560 561 562 563

	dbs_check_cpu(dbs_info);

	queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay);
564
	mutex_unlock(&dbs_info->timer_mutex);
565
}
566

567
static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
568
{
569 570 571 572 573 574 575 576
	/* We want all CPUs to do sampling nearly on same jiffy */
	int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
	delay -= jiffies % delay;

	dbs_info->enable = 1;
	INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
	queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work,
				delay);
577 578
}

579
static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
580
{
581
	dbs_info->enable = 0;
582
	cancel_delayed_work_sync(&dbs_info->work);
583 584 585 586 587 588 589 590
}

static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
				   unsigned int event)
{
	unsigned int cpu = policy->cpu;
	struct cpu_dbs_info_s *this_dbs_info;
	unsigned int j;
J
Jeff Garzik 已提交
591
	int rc;
592

593
	this_dbs_info = &per_cpu(cs_cpu_dbs_info, cpu);
594 595 596

	switch (event) {
	case CPUFREQ_GOV_START:
597
		if ((!cpu_online(cpu)) || (!policy->cur))
598 599
			return -EINVAL;

600
		mutex_lock(&dbs_mutex);
J
Jeff Garzik 已提交
601

602
		rc = sysfs_create_group(&policy->kobj, &dbs_attr_group_old);
J
Jeff Garzik 已提交
603 604 605 606 607
		if (rc) {
			mutex_unlock(&dbs_mutex);
			return rc;
		}

608
		for_each_cpu(j, policy->cpus) {
609
			struct cpu_dbs_info_s *j_dbs_info;
610
			j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
611
			j_dbs_info->cur_policy = policy;
612

613 614 615 616 617 618
			j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
						&j_dbs_info->prev_cpu_wall);
			if (dbs_tuners_ins.ignore_nice) {
				j_dbs_info->prev_cpu_nice =
						kstat_cpu(j).cpustat.nice;
			}
619
		}
620 621
		this_dbs_info->down_skip = 0;
		this_dbs_info->requested_freq = policy->cur;
J
Jeff Garzik 已提交
622

623
		mutex_init(&this_dbs_info->timer_mutex);
624 625 626 627 628 629 630 631
		dbs_enable++;
		/*
		 * Start the timerschedule work, when this governor
		 * is used for first time
		 */
		if (dbs_enable == 1) {
			unsigned int latency;
			/* policy latency is in nS. Convert it to uS first */
632 633 634
			latency = policy->cpuinfo.transition_latency / 1000;
			if (latency == 0)
				latency = 1;
635

636 637 638 639 640 641 642
			rc = sysfs_create_group(cpufreq_global_kobject,
						&dbs_attr_group);
			if (rc) {
				mutex_unlock(&dbs_mutex);
				return rc;
			}

643 644 645 646 647 648 649 650 651 652 653 654
			/*
			 * conservative does not implement micro like ondemand
			 * governor, thus we are bound to jiffes/HZ
			 */
			min_sampling_rate =
				MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10);
			/* Bring kernel and HW constraints together */
			min_sampling_rate = max(min_sampling_rate,
					MIN_LATENCY_MULTIPLIER * latency);
			dbs_tuners_ins.sampling_rate =
				max(min_sampling_rate,
				    latency * LATENCY_MULTIPLIER);
655

656 657 658
			cpufreq_register_notifier(
					&dbs_cpufreq_notifier_block,
					CPUFREQ_TRANSITION_NOTIFIER);
659
		}
660
		mutex_unlock(&dbs_mutex);
661

662 663
		dbs_timer_init(this_dbs_info);

664 665 666
		break;

	case CPUFREQ_GOV_STOP:
667
		dbs_timer_exit(this_dbs_info);
668 669

		mutex_lock(&dbs_mutex);
670
		sysfs_remove_group(&policy->kobj, &dbs_attr_group_old);
671
		dbs_enable--;
672
		mutex_destroy(&this_dbs_info->timer_mutex);
673

674 675 676 677
		/*
		 * Stop the timerschedule work, when this governor
		 * is used for first time
		 */
678
		if (dbs_enable == 0)
679 680 681 682
			cpufreq_unregister_notifier(
					&dbs_cpufreq_notifier_block,
					CPUFREQ_TRANSITION_NOTIFIER);

683
		mutex_unlock(&dbs_mutex);
684 685 686
		if (!dbs_enable)
			sysfs_remove_group(cpufreq_global_kobject,
					   &dbs_attr_group);
687 688 689 690

		break;

	case CPUFREQ_GOV_LIMITS:
691
		mutex_lock(&this_dbs_info->timer_mutex);
692 693 694
		if (policy->max < this_dbs_info->cur_policy->cur)
			__cpufreq_driver_target(
					this_dbs_info->cur_policy,
695
					policy->max, CPUFREQ_RELATION_H);
696 697 698
		else if (policy->min > this_dbs_info->cur_policy->cur)
			__cpufreq_driver_target(
					this_dbs_info->cur_policy,
699
					policy->min, CPUFREQ_RELATION_L);
700
		mutex_unlock(&this_dbs_info->timer_mutex);
701

702 703 704 705 706
		break;
	}
	return 0;
}

707 708 709
#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
static
#endif
710 711 712 713 714
struct cpufreq_governor cpufreq_gov_conservative = {
	.name			= "conservative",
	.governor		= cpufreq_governor_dbs,
	.max_transition_latency	= TRANSITION_LATENCY_LIMIT,
	.owner			= THIS_MODULE,
715 716 717 718
};

static int __init cpufreq_gov_dbs_init(void)
{
719 720 721 722 723 724 725 726 727 728 729 730 731
	int err;

	kconservative_wq = create_workqueue("kconservative");
	if (!kconservative_wq) {
		printk(KERN_ERR "Creation of kconservative failed\n");
		return -EFAULT;
	}

	err = cpufreq_register_governor(&cpufreq_gov_conservative);
	if (err)
		destroy_workqueue(kconservative_wq);

	return err;
732 733 734 735
}

static void __exit cpufreq_gov_dbs_exit(void)
{
736
	cpufreq_unregister_governor(&cpufreq_gov_conservative);
737
	destroy_workqueue(kconservative_wq);
738 739 740
}


741
MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>");
742
MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for "
743 744
		"Low Latency Frequency Transition capable processors "
		"optimised for use in a battery environment");
745
MODULE_LICENSE("GPL");
746

747 748 749
#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
fs_initcall(cpufreq_gov_dbs_init);
#else
750
module_init(cpufreq_gov_dbs_init);
751
#endif
752
module_exit(cpufreq_gov_dbs_exit);