cpufreq_ondemand.c 15.0 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12
/*
 *  drivers/cpufreq/cpufreq_ondemand.c
 *
 *  Copyright (C)  2001 Russell King
 *            (C)  2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
 *                      Jun Nakajima <jun.nakajima@intel.com>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

13 14
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

L
Linus Torvalds 已提交
15
#include <linux/cpufreq.h>
16 17
#include <linux/init.h>
#include <linux/kernel.h>
L
Linus Torvalds 已提交
18
#include <linux/kernel_stat.h>
19 20
#include <linux/kobject.h>
#include <linux/module.h>
21
#include <linux/mutex.h>
22 23
#include <linux/percpu-defs.h>
#include <linux/sysfs.h>
24
#include <linux/tick.h>
25
#include <linux/types.h>
L
Linus Torvalds 已提交
26

27
#include "cpufreq_governor.h"
L
Linus Torvalds 已提交
28

29
/* On-demand governor macors */
30
#define DEF_FREQUENCY_DOWN_DIFFERENTIAL		(10)
L
Linus Torvalds 已提交
31
#define DEF_FREQUENCY_UP_THRESHOLD		(80)
32 33
#define DEF_SAMPLING_DOWN_FACTOR		(1)
#define MAX_SAMPLING_DOWN_FACTOR		(100000)
34 35
#define MICRO_FREQUENCY_DOWN_DIFFERENTIAL	(3)
#define MICRO_FREQUENCY_UP_THRESHOLD		(95)
36
#define MICRO_FREQUENCY_MIN_SAMPLE_RATE		(10000)
37
#define MIN_FREQUENCY_UP_THRESHOLD		(11)
L
Linus Torvalds 已提交
38 39
#define MAX_FREQUENCY_UP_THRESHOLD		(100)

40 41
static struct dbs_data od_dbs_data;
static DEFINE_PER_CPU(struct od_cpu_dbs_info_s, od_cpu_dbs_info);
L
Linus Torvalds 已提交
42

43
static struct od_dbs_tuners od_tuners = {
44
	.up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
45
	.sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR,
46
	.down_differential = DEF_FREQUENCY_DOWN_DIFFERENTIAL,
47
	.ignore_nice = 0,
48
	.powersave_bias = 0,
L
Linus Torvalds 已提交
49 50
};

51
static void ondemand_powersave_bias_init_cpu(int cpu)
52
{
53
	struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu);
54

55 56 57
	dbs_info->freq_table = cpufreq_frequency_get_table(cpu);
	dbs_info->freq_lo = 0;
}
58

59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79
/*
 * 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 (androidlcom) calis 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) andl 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;
80 81
}

82 83 84 85 86
/*
 * 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_jiffies,
 * freq_lo, and freq_lo_jiffies in percpu area for averaging freqs.
 */
87
static unsigned int powersave_bias_target(struct cpufreq_policy *policy,
88
		unsigned int freq_next, unsigned int relation)
89 90 91 92 93
{
	unsigned int freq_req, freq_reduc, freq_avg;
	unsigned int freq_hi, freq_lo;
	unsigned int index = 0;
	unsigned int jiffies_total, jiffies_hi, jiffies_lo;
94
	struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info,
95
						   policy->cpu);
96 97 98 99 100 101 102 103 104 105

	if (!dbs_info->freq_table) {
		dbs_info->freq_lo = 0;
		dbs_info->freq_lo_jiffies = 0;
		return freq_next;
	}

	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next,
			relation, &index);
	freq_req = dbs_info->freq_table[index].frequency;
106
	freq_reduc = freq_req * od_tuners.powersave_bias / 1000;
107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
	freq_avg = freq_req - freq_reduc;

	/* Find freq bounds for freq_avg in freq_table */
	index = 0;
	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
			CPUFREQ_RELATION_H, &index);
	freq_lo = dbs_info->freq_table[index].frequency;
	index = 0;
	cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
			CPUFREQ_RELATION_L, &index);
	freq_hi = dbs_info->freq_table[index].frequency;

	/* Find out how long we have to be in hi and lo freqs */
	if (freq_hi == freq_lo) {
		dbs_info->freq_lo = 0;
		dbs_info->freq_lo_jiffies = 0;
		return freq_lo;
	}
125
	jiffies_total = usecs_to_jiffies(od_tuners.sampling_rate);
126 127 128 129 130 131 132 133 134 135 136 137 138 139
	jiffies_hi = (freq_avg - freq_lo) * jiffies_total;
	jiffies_hi += ((freq_hi - freq_lo) / 2);
	jiffies_hi /= (freq_hi - freq_lo);
	jiffies_lo = jiffies_total - jiffies_hi;
	dbs_info->freq_lo = freq_lo;
	dbs_info->freq_lo_jiffies = jiffies_lo;
	dbs_info->freq_hi_jiffies = jiffies_hi;
	return freq_hi;
}

static void ondemand_powersave_bias_init(void)
{
	int i;
	for_each_online_cpu(i) {
140
		ondemand_powersave_bias_init_cpu(i);
141 142 143
	}
}

144 145 146 147 148 149
static void dbs_freq_increase(struct cpufreq_policy *p, unsigned int freq)
{
	if (od_tuners.powersave_bias)
		freq = powersave_bias_target(p, freq, CPUFREQ_RELATION_H);
	else if (p->cur == p->max)
		return;
150

151 152 153 154 155 156 157 158 159 160 161 162 163 164
	__cpufreq_driver_target(p, 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 Every sampling_rate, we look for
 * a the lowest frequency which can sustain the load while keeping idle time
 * over 30%. If such a frequency exist, we try to decrease to this frequency.
 *
 * Any frequency increase takes it to the maximum frequency. Frequency reduction
 * happens at minimum steps of 5% (default) of current frequency
 */
static void od_check_cpu(int cpu, unsigned int load_freq)
L
Linus Torvalds 已提交
165
{
166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
	struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu);
	struct cpufreq_policy *policy = dbs_info->cdbs.cur_policy;

	dbs_info->freq_lo = 0;

	/* Check for frequency increase */
	if (load_freq > od_tuners.up_threshold * policy->cur) {
		/* If switching to max speed, apply sampling_down_factor */
		if (policy->cur < policy->max)
			dbs_info->rate_mult =
				od_tuners.sampling_down_factor;
		dbs_freq_increase(policy, policy->max);
		return;
	}

	/* Check for frequency decrease */
	/* if we cannot reduce the frequency anymore, break out early */
	if (policy->cur == policy->min)
		return;

	/*
	 * 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.
	 */
	if (load_freq < (od_tuners.up_threshold - od_tuners.down_differential) *
			policy->cur) {
		unsigned int freq_next;
		freq_next = load_freq / (od_tuners.up_threshold -
				od_tuners.down_differential);

		/* No longer fully busy, reset rate_mult */
		dbs_info->rate_mult = 1;

		if (freq_next < policy->min)
			freq_next = policy->min;

		if (!od_tuners.powersave_bias) {
			__cpufreq_driver_target(policy, freq_next,
					CPUFREQ_RELATION_L);
		} else {
			int freq = powersave_bias_target(policy, freq_next,
					CPUFREQ_RELATION_L);
			__cpufreq_driver_target(policy, freq,
					CPUFREQ_RELATION_L);
		}
	}
L
Linus Torvalds 已提交
213 214
}

215 216 217 218 219 220
static void od_dbs_timer(struct work_struct *work)
{
	struct od_cpu_dbs_info_s *dbs_info =
		container_of(work, struct od_cpu_dbs_info_s, cdbs.work.work);
	unsigned int cpu = dbs_info->cdbs.cpu;
	int delay, sample_type = dbs_info->sample_type;
L
Linus Torvalds 已提交
221

222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
	mutex_lock(&dbs_info->cdbs.timer_mutex);

	/* Common NORMAL_SAMPLE setup */
	dbs_info->sample_type = OD_NORMAL_SAMPLE;
	if (sample_type == OD_SUB_SAMPLE) {
		delay = dbs_info->freq_lo_jiffies;
		__cpufreq_driver_target(dbs_info->cdbs.cur_policy,
			dbs_info->freq_lo, CPUFREQ_RELATION_H);
	} else {
		dbs_check_cpu(&od_dbs_data, cpu);
		if (dbs_info->freq_lo) {
			/* Setup timer for SUB_SAMPLE */
			dbs_info->sample_type = OD_SUB_SAMPLE;
			delay = dbs_info->freq_hi_jiffies;
		} else {
			delay = delay_for_sampling_rate(dbs_info->rate_mult);
		}
	}

	schedule_delayed_work_on(cpu, &dbs_info->cdbs.work, delay);
	mutex_unlock(&dbs_info->cdbs.timer_mutex);
}

/************************** sysfs interface ************************/

static ssize_t show_sampling_rate_min(struct kobject *kobj,
				      struct attribute *attr, char *buf)
{
	return sprintf(buf, "%u\n", od_dbs_data.min_sampling_rate);
L
Linus Torvalds 已提交
251 252
}

253 254 255 256 257
/**
 * update_sampling_rate - update sampling rate effective immediately if needed.
 * @new_rate: new sampling rate
 *
 * If new rate is smaller than the old, simply updaing
258 259 260 261 262 263 264
 * dbs_tuners_int.sampling_rate might not be appropriate. For example, if the
 * original sampling_rate was 1 second and the requested new sampling rate is 10
 * ms because the user needs immediate reaction from ondemand governor, but not
 * sure if higher frequency will be required or not, then, the governor may
 * change the sampling rate too late; up to 1 second later. Thus, if we are
 * reducing the sampling rate, we need to make the new value effective
 * immediately.
265 266 267 268 269
 */
static void update_sampling_rate(unsigned int new_rate)
{
	int cpu;

270 271
	od_tuners.sampling_rate = new_rate = max(new_rate,
			od_dbs_data.min_sampling_rate);
272 273 274

	for_each_online_cpu(cpu) {
		struct cpufreq_policy *policy;
275
		struct od_cpu_dbs_info_s *dbs_info;
276 277 278 279 280 281 282 283
		unsigned long next_sampling, appointed_at;

		policy = cpufreq_cpu_get(cpu);
		if (!policy)
			continue;
		dbs_info = &per_cpu(od_cpu_dbs_info, policy->cpu);
		cpufreq_cpu_put(policy);

284
		mutex_lock(&dbs_info->cdbs.timer_mutex);
285

286 287
		if (!delayed_work_pending(&dbs_info->cdbs.work)) {
			mutex_unlock(&dbs_info->cdbs.timer_mutex);
288 289 290
			continue;
		}

291 292
		next_sampling = jiffies + usecs_to_jiffies(new_rate);
		appointed_at = dbs_info->cdbs.work.timer.expires;
293 294 295

		if (time_before(next_sampling, appointed_at)) {

296 297 298
			mutex_unlock(&dbs_info->cdbs.timer_mutex);
			cancel_delayed_work_sync(&dbs_info->cdbs.work);
			mutex_lock(&dbs_info->cdbs.timer_mutex);
299

300 301 302
			schedule_delayed_work_on(dbs_info->cdbs.cpu,
					&dbs_info->cdbs.work,
					usecs_to_jiffies(new_rate));
303 304

		}
305
		mutex_unlock(&dbs_info->cdbs.timer_mutex);
306 307 308
	}
}

309 310
static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
				   const char *buf, size_t count)
L
Linus Torvalds 已提交
311 312 313
{
	unsigned int input;
	int ret;
314
	ret = sscanf(buf, "%u", &input);
315 316
	if (ret != 1)
		return -EINVAL;
317
	update_sampling_rate(input);
L
Linus Torvalds 已提交
318 319 320
	return count;
}

321 322 323 324 325 326 327 328 329
static ssize_t store_io_is_busy(struct kobject *a, struct attribute *b,
				   const char *buf, size_t count)
{
	unsigned int input;
	int ret;

	ret = sscanf(buf, "%u", &input);
	if (ret != 1)
		return -EINVAL;
330
	od_tuners.io_is_busy = !!input;
331 332 333
	return count;
}

334 335
static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
				  const char *buf, size_t count)
L
Linus Torvalds 已提交
336 337 338
{
	unsigned int input;
	int ret;
339
	ret = sscanf(buf, "%u", &input);
L
Linus Torvalds 已提交
340

341
	if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD ||
342
			input < MIN_FREQUENCY_UP_THRESHOLD) {
L
Linus Torvalds 已提交
343 344
		return -EINVAL;
	}
345
	od_tuners.up_threshold = input;
L
Linus Torvalds 已提交
346 347 348
	return count;
}

349 350 351 352 353 354 355 356 357
static ssize_t store_sampling_down_factor(struct kobject *a,
			struct attribute *b, const char *buf, size_t count)
{
	unsigned int input, j;
	int ret;
	ret = sscanf(buf, "%u", &input);

	if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
		return -EINVAL;
358
	od_tuners.sampling_down_factor = input;
359 360 361

	/* Reset down sampling multiplier in case it was active */
	for_each_online_cpu(j) {
362 363
		struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info,
				j);
364 365 366 367 368
		dbs_info->rate_mult = 1;
	}
	return count;
}

369 370
static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
				      const char *buf, size_t count)
371 372 373 374 375
{
	unsigned int input;
	int ret;

	unsigned int j;
376

377
	ret = sscanf(buf, "%u", &input);
378
	if (ret != 1)
379 380
		return -EINVAL;

381
	if (input > 1)
382
		input = 1;
383

384
	if (input == od_tuners.ignore_nice) { /* nothing to do */
385 386
		return count;
	}
387
	od_tuners.ignore_nice = input;
388

389
	/* we need to re-evaluate prev_cpu_idle */
390
	for_each_online_cpu(j) {
391
		struct od_cpu_dbs_info_s *dbs_info;
392
		dbs_info = &per_cpu(od_cpu_dbs_info, j);
393 394 395 396 397
		dbs_info->cdbs.prev_cpu_idle = get_cpu_idle_time(j,
						&dbs_info->cdbs.prev_cpu_wall);
		if (od_tuners.ignore_nice)
			dbs_info->cdbs.prev_cpu_nice =
				kcpustat_cpu(j).cpustat[CPUTIME_NICE];
398

399 400 401 402
	}
	return count;
}

403 404
static ssize_t store_powersave_bias(struct kobject *a, struct attribute *b,
				    const char *buf, size_t count)
405 406 407 408 409 410 411 412 413 414 415
{
	unsigned int input;
	int ret;
	ret = sscanf(buf, "%u", &input);

	if (ret != 1)
		return -EINVAL;

	if (input > 1000)
		input = 1000;

416
	od_tuners.powersave_bias = input;
417 418 419 420
	ondemand_powersave_bias_init();
	return count;
}

421 422 423 424 425 426 427
show_one(od, sampling_rate, sampling_rate);
show_one(od, io_is_busy, io_is_busy);
show_one(od, up_threshold, up_threshold);
show_one(od, sampling_down_factor, sampling_down_factor);
show_one(od, ignore_nice_load, ignore_nice);
show_one(od, powersave_bias, powersave_bias);

428
define_one_global_rw(sampling_rate);
429
define_one_global_rw(io_is_busy);
430
define_one_global_rw(up_threshold);
431
define_one_global_rw(sampling_down_factor);
432 433
define_one_global_rw(ignore_nice_load);
define_one_global_rw(powersave_bias);
434
define_one_global_ro(sampling_rate_min);
L
Linus Torvalds 已提交
435

436
static struct attribute *dbs_attributes[] = {
L
Linus Torvalds 已提交
437 438 439
	&sampling_rate_min.attr,
	&sampling_rate.attr,
	&up_threshold.attr,
440
	&sampling_down_factor.attr,
441
	&ignore_nice_load.attr,
442
	&powersave_bias.attr,
443
	&io_is_busy.attr,
L
Linus Torvalds 已提交
444 445 446
	NULL
};

447
static struct attribute_group od_attr_group = {
L
Linus Torvalds 已提交
448 449 450 451 452 453
	.attrs = dbs_attributes,
	.name = "ondemand",
};

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

454
define_get_cpu_dbs_routines(od_cpu_dbs_info);
455

456 457 458 459 460 461
static struct od_ops od_ops = {
	.io_busy = should_io_be_busy,
	.powersave_bias_init_cpu = ondemand_powersave_bias_init_cpu,
	.powersave_bias_target = powersave_bias_target,
	.freq_increase = dbs_freq_increase,
};
462

463 464 465 466 467 468 469 470 471 472
static struct dbs_data od_dbs_data = {
	.governor = GOV_ONDEMAND,
	.attr_group = &od_attr_group,
	.tuners = &od_tuners,
	.get_cpu_cdbs = get_cpu_cdbs,
	.get_cpu_dbs_info_s = get_cpu_dbs_info_s,
	.gov_dbs_timer = od_dbs_timer,
	.gov_check_cpu = od_check_cpu,
	.gov_ops = &od_ops,
};
L
Linus Torvalds 已提交
473

474 475
static int od_cpufreq_governor_dbs(struct cpufreq_policy *policy,
		unsigned int event)
L
Linus Torvalds 已提交
476
{
477
	return cpufreq_governor_dbs(&od_dbs_data, policy, event);
L
Linus Torvalds 已提交
478 479
}

480 481
#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
static
482
#endif
483 484 485 486 487 488
struct cpufreq_governor cpufreq_gov_ondemand = {
	.name			= "ondemand",
	.governor		= od_cpufreq_governor_dbs,
	.max_transition_latency	= TRANSITION_LATENCY_LIMIT,
	.owner			= THIS_MODULE,
};
L
Linus Torvalds 已提交
489 490 491

static int __init cpufreq_gov_dbs_init(void)
{
492 493
	u64 idle_time;
	int cpu = get_cpu();
494

495
	mutex_init(&od_dbs_data.mutex);
496
	idle_time = get_cpu_idle_time_us(cpu, NULL);
497
	put_cpu();
498 499
	if (idle_time != -1ULL) {
		/* Idle micro accounting is supported. Use finer thresholds */
500 501
		od_tuners.up_threshold = MICRO_FREQUENCY_UP_THRESHOLD;
		od_tuners.down_differential = MICRO_FREQUENCY_DOWN_DIFFERENTIAL;
502
		/*
503
		 * In nohz/micro accounting case we set the minimum frequency
504 505 506
		 * not depending on HZ, but fixed (very low). The deferred
		 * timer might skip some samples if idle/sleeping as needed.
		*/
507
		od_dbs_data.min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE;
508 509
	} else {
		/* For correct statistics, we need 10 ticks for each measure */
510 511
		od_dbs_data.min_sampling_rate = MIN_SAMPLING_RATE_RATIO *
			jiffies_to_usecs(10);
512
	}
513

514
	return cpufreq_register_governor(&cpufreq_gov_ondemand);
L
Linus Torvalds 已提交
515 516 517 518
}

static void __exit cpufreq_gov_dbs_exit(void)
{
519
	cpufreq_unregister_governor(&cpufreq_gov_ondemand);
L
Linus Torvalds 已提交
520 521
}

522 523 524
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 "
525
	"Low Latency Frequency Transition capable processors");
526
MODULE_LICENSE("GPL");
L
Linus Torvalds 已提交
527

528 529 530
#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
fs_initcall(cpufreq_gov_dbs_init);
#else
L
Linus Torvalds 已提交
531
module_init(cpufreq_gov_dbs_init);
532
#endif
L
Linus Torvalds 已提交
533
module_exit(cpufreq_gov_dbs_exit);