cpufreq_conservative.c 16.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
/*
 *  drivers/cpufreq/cpufreq_conservative.c
 *
 *  Copyright (C)  2001 Russell King
 *            (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
 *                      Jun Nakajima <jun.nakajima@intel.com>
 *            (C)  2004 Alexander Clouter <alex-kernel@digriz.org.uk>
 *
 * 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/smp.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/ctype.h>
#include <linux/cpufreq.h>
#include <linux/sysctl.h>
#include <linux/types.h>
#include <linux/fs.h>
#include <linux/sysfs.h>
A
Andrew Morton 已提交
25
#include <linux/cpu.h>
26 27 28 29 30
#include <linux/kmod.h>
#include <linux/workqueue.h>
#include <linux/jiffies.h>
#include <linux/kernel_stat.h>
#include <linux/percpu.h>
31
#include <linux/mutex.h>
32 33 34 35 36 37 38 39
/*
 * 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)

40 41
/*
 * The polling frequency of this governor depends on the capability of
42
 * the processor. Default polling frequency is 1000 times the transition
43 44
 * latency of the processor. The governor will work on any processor with
 * transition latency <= 10mS, using appropriate sampling
45
 * rate.
46 47
 * For CPUs with transition latency > 10mS (mostly drivers
 * with CPUFREQ_ETERNAL), this governor will not work.
48 49
 * All times here are in uS.
 */
50
static unsigned int def_sampling_rate;
51 52
#define MIN_SAMPLING_RATE_RATIO			(2)
/* for correct statistics, we need at least 10 ticks between each measure */
53 54 55 56
#define MIN_STAT_SAMPLING_RATE			\
			(MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10))
#define MIN_SAMPLING_RATE			\
			(def_sampling_rate / MIN_SAMPLING_RATE_RATIO)
57
#define MAX_SAMPLING_RATE			(500 * def_sampling_rate)
58 59 60
#define DEF_SAMPLING_RATE_LATENCY_MULTIPLIER	(1000)
#define DEF_SAMPLING_DOWN_FACTOR		(1)
#define MAX_SAMPLING_DOWN_FACTOR		(10)
61
#define TRANSITION_LATENCY_LIMIT		(10 * 1000 * 1000)
62

D
David Howells 已提交
63
static void do_dbs_timer(struct work_struct *work);
64 65

struct cpu_dbs_info_s {
66 67 68 69 70 71
	struct cpufreq_policy *cur_policy;
	unsigned int prev_cpu_idle_up;
	unsigned int prev_cpu_idle_down;
	unsigned int enable;
	unsigned int down_skip;
	unsigned int requested_freq;
72 73 74 75 76
};
static DEFINE_PER_CPU(struct cpu_dbs_info_s, cpu_dbs_info);

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

77 78 79 80 81 82 83 84
/*
 * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug
 * lock and dbs_mutex. cpu_hotplug lock should always be held before
 * dbs_mutex. If any function that can potentially take cpu_hotplug lock
 * (like __cpufreq_driver_target()) is being called with dbs_mutex taken, then
 * cpu_hotplug lock should be taken before that. Note that cpu_hotplug lock
 * is recursive for the same process. -Venki
 */
85
static DEFINE_MUTEX (dbs_mutex);
D
David Howells 已提交
86
static DECLARE_DELAYED_WORK(dbs_work, do_dbs_timer);
87 88

struct dbs_tuners {
89 90 91 92 93 94
	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;
95 96 97
};

static struct dbs_tuners dbs_tuners_ins = {
98 99 100 101 102
	.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,
103 104
};

105 106
static inline unsigned int get_cpu_idle_time(unsigned int cpu)
{
107 108 109 110 111
	unsigned int add_nice = 0, ret;

	if (dbs_tuners_ins.ignore_nice)
		add_nice = kstat_cpu(cpu).cpustat.nice;

112
	ret = kstat_cpu(cpu).cpustat.idle +
113
		kstat_cpu(cpu).cpustat.iowait +
114 115 116
		add_nice;

	return ret;
117 118
}

119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
/* keep track of frequency transitions */
static int
dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
		     void *data)
{
	struct cpufreq_freqs *freq = data;
	struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info,
							freq->cpu);

	if (!this_dbs_info->enable)
		return 0;

	this_dbs_info->requested_freq = freq->new;

	return 0;
}

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

140 141 142 143 144 145 146 147 148 149 150
/************************** sysfs interface ************************/
static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf)
{
	return sprintf (buf, "%u\n", MAX_SAMPLING_RATE);
}

static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf)
{
	return sprintf (buf, "%u\n", MIN_SAMPLING_RATE);
}

151 152
#define define_one_ro(_name)				\
static struct freq_attr _name =				\
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168
__ATTR(_name, 0444, show_##_name, NULL)

define_one_ro(sampling_rate_max);
define_one_ro(sampling_rate_min);

/* cpufreq_conservative Governor Tunables */
#define show_one(file_name, object)					\
static ssize_t show_##file_name						\
(struct cpufreq_policy *unused, char *buf)				\
{									\
	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);
169
show_one(ignore_nice_load, ignore_nice);
170 171
show_one(freq_step, freq_step);

172
static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused,
173 174 175 176 177
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;
	ret = sscanf (buf, "%u", &input);
178
	if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
179 180
		return -EINVAL;

181
	mutex_lock(&dbs_mutex);
182
	dbs_tuners_ins.sampling_down_factor = input;
183
	mutex_unlock(&dbs_mutex);
184 185 186 187

	return count;
}

188
static ssize_t store_sampling_rate(struct cpufreq_policy *unused,
189 190 191 192 193 194
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;
	ret = sscanf (buf, "%u", &input);

195
	mutex_lock(&dbs_mutex);
196
	if (ret != 1 || input > MAX_SAMPLING_RATE || input < MIN_SAMPLING_RATE) {
197
		mutex_unlock(&dbs_mutex);
198 199 200 201
		return -EINVAL;
	}

	dbs_tuners_ins.sampling_rate = input;
202
	mutex_unlock(&dbs_mutex);
203 204 205 206

	return count;
}

207
static ssize_t store_up_threshold(struct cpufreq_policy *unused,
208 209 210 211 212 213
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;
	ret = sscanf (buf, "%u", &input);

214
	mutex_lock(&dbs_mutex);
215
	if (ret != 1 || input > 100 || input <= dbs_tuners_ins.down_threshold) {
216
		mutex_unlock(&dbs_mutex);
217 218 219 220
		return -EINVAL;
	}

	dbs_tuners_ins.up_threshold = input;
221
	mutex_unlock(&dbs_mutex);
222 223 224 225

	return count;
}

226
static ssize_t store_down_threshold(struct cpufreq_policy *unused,
227 228 229 230 231 232
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;
	ret = sscanf (buf, "%u", &input);

233
	mutex_lock(&dbs_mutex);
234
	if (ret != 1 || input > 100 || input >= dbs_tuners_ins.up_threshold) {
235
		mutex_unlock(&dbs_mutex);
236 237 238 239
		return -EINVAL;
	}

	dbs_tuners_ins.down_threshold = input;
240
	mutex_unlock(&dbs_mutex);
241 242 243 244

	return count;
}

245
static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy,
246 247 248 249 250 251
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;

	unsigned int j;
252 253 254

	ret = sscanf(buf, "%u", &input);
	if (ret != 1)
255 256
		return -EINVAL;

257
	if (input > 1)
258
		input = 1;
259

260
	mutex_lock(&dbs_mutex);
261
	if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
262
		mutex_unlock(&dbs_mutex);
263 264 265 266 267
		return count;
	}
	dbs_tuners_ins.ignore_nice = input;

	/* we need to re-evaluate prev_cpu_idle_up and prev_cpu_idle_down */
268
	for_each_online_cpu(j) {
269 270
		struct cpu_dbs_info_s *j_dbs_info;
		j_dbs_info = &per_cpu(cpu_dbs_info, j);
271
		j_dbs_info->prev_cpu_idle_up = get_cpu_idle_time(j);
272 273
		j_dbs_info->prev_cpu_idle_down = j_dbs_info->prev_cpu_idle_up;
	}
274
	mutex_unlock(&dbs_mutex);
275 276 277 278 279 280 281 282 283 284

	return count;
}

static ssize_t store_freq_step(struct cpufreq_policy *policy,
		const char *buf, size_t count)
{
	unsigned int input;
	int ret;

285
	ret = sscanf(buf, "%u", &input);
286

287
	if (ret != 1)
288 289
		return -EINVAL;

290
	if (input > 100)
291
		input = 100;
292

293 294
	/* no need to test here if freq_step is zero as the user might actually
	 * want this, they would be crazy though :) */
295
	mutex_lock(&dbs_mutex);
296
	dbs_tuners_ins.freq_step = input;
297
	mutex_unlock(&dbs_mutex);
298 299 300 301 302 303 304 305 306 307 308 309

	return count;
}

#define define_one_rw(_name) \
static struct freq_attr _name = \
__ATTR(_name, 0644, show_##_name, store_##_name)

define_one_rw(sampling_rate);
define_one_rw(sampling_down_factor);
define_one_rw(up_threshold);
define_one_rw(down_threshold);
310
define_one_rw(ignore_nice_load);
311 312 313 314 315 316 317 318 319
define_one_rw(freq_step);

static struct attribute * dbs_attributes[] = {
	&sampling_rate_max.attr,
	&sampling_rate_min.attr,
	&sampling_rate.attr,
	&sampling_down_factor.attr,
	&up_threshold.attr,
	&down_threshold.attr,
320
	&ignore_nice_load.attr,
321 322 323 324 325 326 327 328 329 330 331 332 333 334
	&freq_step.attr,
	NULL
};

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

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

static void dbs_check_cpu(int cpu)
{
	unsigned int idle_ticks, up_idle_ticks, down_idle_ticks;
335
	unsigned int tmp_idle_ticks, total_idle_ticks;
336
	unsigned int freq_target;
337
	unsigned int freq_down_sampling_rate;
338
	struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info, cpu);
339 340 341 342 343
	struct cpufreq_policy *policy;

	if (!this_dbs_info->enable)
		return;

344 345
	policy = this_dbs_info->cur_policy;

346 347
	/*
	 * The default safe range is 20% to 80%
348
	 * Every sampling_rate, we check
349 350
	 *	- If current idle time is less than 20%, then we try to
	 *	  increase frequency
351
	 * Every sampling_rate*sampling_down_factor, we check
352 353
	 *	- If current idle time is more than 80%, then we try to
	 *	  decrease frequency
354
	 *
355 356 357
	 * Any frequency increase takes it to the maximum frequency.
	 * Frequency reduction happens at minimum steps of
	 * 5% (default) of max_frequency
358 359 360
	 */

	/* Check for frequency increase */
361
	idle_ticks = UINT_MAX;
362

363 364 365 366 367 368 369 370
	/* Check for frequency increase */
	total_idle_ticks = get_cpu_idle_time(cpu);
	tmp_idle_ticks = total_idle_ticks -
		this_dbs_info->prev_cpu_idle_up;
	this_dbs_info->prev_cpu_idle_up = total_idle_ticks;

	if (tmp_idle_ticks < idle_ticks)
		idle_ticks = tmp_idle_ticks;
371 372 373 374

	/* Scale idle ticks by 100 and compare with up and down ticks */
	idle_ticks *= 100;
	up_idle_ticks = (100 - dbs_tuners_ins.up_threshold) *
375
			usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
376 377

	if (idle_ticks < up_idle_ticks) {
378
		this_dbs_info->down_skip = 0;
379 380
		this_dbs_info->prev_cpu_idle_down =
			this_dbs_info->prev_cpu_idle_up;
381

382
		/* if we are already at full speed then break out early */
383
		if (this_dbs_info->requested_freq == policy->max)
384
			return;
385

386
		freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
387 388

		/* max freq cannot be less than 100. But who knows.... */
389 390
		if (unlikely(freq_target == 0))
			freq_target = 5;
391

392
		this_dbs_info->requested_freq += freq_target;
393 394
		if (this_dbs_info->requested_freq > policy->max)
			this_dbs_info->requested_freq = policy->max;
395

396
		__cpufreq_driver_target(policy, this_dbs_info->requested_freq,
397 398 399 400 401
			CPUFREQ_RELATION_H);
		return;
	}

	/* Check for frequency decrease */
402 403
	this_dbs_info->down_skip++;
	if (this_dbs_info->down_skip < dbs_tuners_ins.sampling_down_factor)
404 405
		return;

406 407 408 409 410
	/* Check for frequency decrease */
	total_idle_ticks = this_dbs_info->prev_cpu_idle_up;
	tmp_idle_ticks = total_idle_ticks -
		this_dbs_info->prev_cpu_idle_down;
	this_dbs_info->prev_cpu_idle_down = total_idle_ticks;
411

412 413
	if (tmp_idle_ticks < idle_ticks)
		idle_ticks = tmp_idle_ticks;
414 415 416

	/* Scale idle ticks by 100 and compare with up and down ticks */
	idle_ticks *= 100;
417
	this_dbs_info->down_skip = 0;
418 419 420 421

	freq_down_sampling_rate = dbs_tuners_ins.sampling_rate *
		dbs_tuners_ins.sampling_down_factor;
	down_idle_ticks = (100 - dbs_tuners_ins.down_threshold) *
422
		usecs_to_jiffies(freq_down_sampling_rate);
423

424
	if (idle_ticks > down_idle_ticks) {
425 426
		/*
		 * if we are already at the lowest speed then break out early
427
		 * or if we 'cannot' reduce the speed as the user might want
428
		 * freq_target to be zero
429
		 */
430
		if (this_dbs_info->requested_freq == policy->min
431 432 433
				|| dbs_tuners_ins.freq_step == 0)
			return;

434
		freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
435 436

		/* max freq cannot be less than 100. But who knows.... */
437 438
		if (unlikely(freq_target == 0))
			freq_target = 5;
439

440
		this_dbs_info->requested_freq -= freq_target;
441 442
		if (this_dbs_info->requested_freq < policy->min)
			this_dbs_info->requested_freq = policy->min;
443

444
		__cpufreq_driver_target(policy, this_dbs_info->requested_freq,
445
				CPUFREQ_RELATION_H);
446 447 448 449
		return;
	}
}

D
David Howells 已提交
450
static void do_dbs_timer(struct work_struct *work)
451
{
452
	int i;
453
	mutex_lock(&dbs_mutex);
454 455
	for_each_online_cpu(i)
		dbs_check_cpu(i);
456
	schedule_delayed_work(&dbs_work,
457
			usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
458
	mutex_unlock(&dbs_mutex);
459
}
460 461 462

static inline void dbs_timer_init(void)
{
463
	init_timer_deferrable(&dbs_work.timer);
464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
	schedule_delayed_work(&dbs_work,
			usecs_to_jiffies(dbs_tuners_ins.sampling_rate));
	return;
}

static inline void dbs_timer_exit(void)
{
	cancel_delayed_work(&dbs_work);
	return;
}

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
481
	int rc;
482 483 484 485 486

	this_dbs_info = &per_cpu(cpu_dbs_info, cpu);

	switch (event) {
	case CPUFREQ_GOV_START:
487
		if ((!cpu_online(cpu)) || (!policy->cur))
488 489 490 491
			return -EINVAL;

		if (this_dbs_info->enable) /* Already enabled */
			break;
492

493
		mutex_lock(&dbs_mutex);
J
Jeff Garzik 已提交
494 495 496 497 498 499 500

		rc = sysfs_create_group(&policy->kobj, &dbs_attr_group);
		if (rc) {
			mutex_unlock(&dbs_mutex);
			return rc;
		}

501
		for_each_cpu_mask_nr(j, policy->cpus) {
502 503 504
			struct cpu_dbs_info_s *j_dbs_info;
			j_dbs_info = &per_cpu(cpu_dbs_info, j);
			j_dbs_info->cur_policy = policy;
505

506
			j_dbs_info->prev_cpu_idle_up = get_cpu_idle_time(cpu);
507 508 509 510
			j_dbs_info->prev_cpu_idle_down
				= j_dbs_info->prev_cpu_idle_up;
		}
		this_dbs_info->enable = 1;
511 512
		this_dbs_info->down_skip = 0;
		this_dbs_info->requested_freq = policy->cur;
J
Jeff Garzik 已提交
513

514 515 516 517 518 519 520 521
		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 */
522 523 524
			latency = policy->cpuinfo.transition_latency / 1000;
			if (latency == 0)
				latency = 1;
525

526
			def_sampling_rate = 10 * latency *
527
					DEF_SAMPLING_RATE_LATENCY_MULTIPLIER;
528 529 530 531

			if (def_sampling_rate < MIN_STAT_SAMPLING_RATE)
				def_sampling_rate = MIN_STAT_SAMPLING_RATE;

532 533 534
			dbs_tuners_ins.sampling_rate = def_sampling_rate;

			dbs_timer_init();
535 536 537
			cpufreq_register_notifier(
					&dbs_cpufreq_notifier_block,
					CPUFREQ_TRANSITION_NOTIFIER);
538
		}
539

540
		mutex_unlock(&dbs_mutex);
541 542 543
		break;

	case CPUFREQ_GOV_STOP:
544
		mutex_lock(&dbs_mutex);
545 546 547 548 549 550 551
		this_dbs_info->enable = 0;
		sysfs_remove_group(&policy->kobj, &dbs_attr_group);
		dbs_enable--;
		/*
		 * Stop the timerschedule work, when this governor
		 * is used for first time
		 */
552
		if (dbs_enable == 0) {
553
			dbs_timer_exit();
554 555 556 557 558
			cpufreq_unregister_notifier(
					&dbs_cpufreq_notifier_block,
					CPUFREQ_TRANSITION_NOTIFIER);
		}

559
		mutex_unlock(&dbs_mutex);
560 561 562 563

		break;

	case CPUFREQ_GOV_LIMITS:
564
		mutex_lock(&dbs_mutex);
565 566 567
		if (policy->max < this_dbs_info->cur_policy->cur)
			__cpufreq_driver_target(
					this_dbs_info->cur_policy,
568
					policy->max, CPUFREQ_RELATION_H);
569 570 571
		else if (policy->min > this_dbs_info->cur_policy->cur)
			__cpufreq_driver_target(
					this_dbs_info->cur_policy,
572
					policy->min, CPUFREQ_RELATION_L);
573
		mutex_unlock(&dbs_mutex);
574 575 576 577 578
		break;
	}
	return 0;
}

579 580 581
#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
static
#endif
582 583 584 585 586
struct cpufreq_governor cpufreq_gov_conservative = {
	.name			= "conservative",
	.governor		= cpufreq_governor_dbs,
	.max_transition_latency	= TRANSITION_LATENCY_LIMIT,
	.owner			= THIS_MODULE,
587 588 589 590
};

static int __init cpufreq_gov_dbs_init(void)
{
591
	return cpufreq_register_governor(&cpufreq_gov_conservative);
592 593 594 595 596 597 598
}

static void __exit cpufreq_gov_dbs_exit(void)
{
	/* Make sure that the scheduled work is indeed not running */
	flush_scheduled_work();

599
	cpufreq_unregister_governor(&cpufreq_gov_conservative);
600 601 602 603 604 605 606 607 608
}


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

609 610 611
#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
fs_initcall(cpufreq_gov_dbs_init);
#else
612
module_init(cpufreq_gov_dbs_init);
613
#endif
614
module_exit(cpufreq_gov_dbs_exit);