Merge branch 'bkl/procfs' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[linux-2.6-block.git] / drivers / cpufreq / cpufreq_conservative.c
CommitLineData
b9170836
DJ
1/*
2 * drivers/cpufreq/cpufreq_conservative.c
3 *
4 * Copyright (C) 2001 Russell King
5 * (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
6 * Jun Nakajima <jun.nakajima@intel.com>
11a80a9c 7 * (C) 2009 Alexander Clouter <alex@digriz.org.uk>
b9170836
DJ
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 */
13
14#include <linux/kernel.h>
15#include <linux/module.h>
b9170836 16#include <linux/init.h>
b9170836 17#include <linux/cpufreq.h>
138a0128 18#include <linux/cpu.h>
b9170836
DJ
19#include <linux/jiffies.h>
20#include <linux/kernel_stat.h>
3fc54d37 21#include <linux/mutex.h>
8e677ce8
AC
22#include <linux/hrtimer.h>
23#include <linux/tick.h>
24#include <linux/ktime.h>
25#include <linux/sched.h>
26
b9170836
DJ
27/*
28 * dbs is used in this file as a shortform for demandbased switching
29 * It helps to keep variable names smaller, simpler
30 */
31
32#define DEF_FREQUENCY_UP_THRESHOLD (80)
b9170836 33#define DEF_FREQUENCY_DOWN_THRESHOLD (20)
b9170836 34
18a7247d
DJ
35/*
36 * The polling frequency of this governor depends on the capability of
b9170836 37 * the processor. Default polling frequency is 1000 times the transition
18a7247d
DJ
38 * latency of the processor. The governor will work on any processor with
39 * transition latency <= 10mS, using appropriate sampling
b9170836 40 * rate.
8e677ce8
AC
41 * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL)
42 * this governor will not work.
b9170836
DJ
43 * All times here are in uS.
44 */
2c906b31 45#define MIN_SAMPLING_RATE_RATIO (2)
112124ab 46
cef9615a
TR
47static unsigned int min_sampling_rate;
48
112124ab 49#define LATENCY_MULTIPLIER (1000)
cef9615a 50#define MIN_LATENCY_MULTIPLIER (100)
2c906b31
AC
51#define DEF_SAMPLING_DOWN_FACTOR (1)
52#define MAX_SAMPLING_DOWN_FACTOR (10)
1c256245 53#define TRANSITION_LATENCY_LIMIT (10 * 1000 * 1000)
b9170836 54
c4028958 55static void do_dbs_timer(struct work_struct *work);
b9170836
DJ
56
57struct cpu_dbs_info_s {
8e677ce8
AC
58 cputime64_t prev_cpu_idle;
59 cputime64_t prev_cpu_wall;
60 cputime64_t prev_cpu_nice;
18a7247d 61 struct cpufreq_policy *cur_policy;
8e677ce8 62 struct delayed_work work;
18a7247d
DJ
63 unsigned int down_skip;
64 unsigned int requested_freq;
8e677ce8
AC
65 int cpu;
66 unsigned int enable:1;
ee88415c 67 /*
68 * percpu mutex that serializes governor limit change with
69 * do_dbs_timer invocation. We do not want do_dbs_timer to run
70 * when user is changing the governor or limits.
71 */
72 struct mutex timer_mutex;
b9170836 73};
245b2e70 74static DEFINE_PER_CPU(struct cpu_dbs_info_s, cs_cpu_dbs_info);
b9170836
DJ
75
76static unsigned int dbs_enable; /* number of CPUs using this policy */
77
4ec223d0 78/*
7d26e2d5 79 * dbs_mutex protects data in dbs_tuners_ins from concurrent changes on
ee88415c 80 * different CPUs. It protects dbs_enable in governor start/stop.
4ec223d0 81 */
9acef487 82static DEFINE_MUTEX(dbs_mutex);
b9170836 83
8e677ce8
AC
84static struct workqueue_struct *kconservative_wq;
85
86static struct dbs_tuners {
18a7247d
DJ
87 unsigned int sampling_rate;
88 unsigned int sampling_down_factor;
89 unsigned int up_threshold;
90 unsigned int down_threshold;
91 unsigned int ignore_nice;
92 unsigned int freq_step;
8e677ce8 93} dbs_tuners_ins = {
18a7247d
DJ
94 .up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
95 .down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD,
96 .sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR,
97 .ignore_nice = 0,
98 .freq_step = 5,
b9170836
DJ
99};
100
8e677ce8
AC
101static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu,
102 cputime64_t *wall)
dac1c1a5 103{
8e677ce8
AC
104 cputime64_t idle_time;
105 cputime64_t cur_wall_time;
106 cputime64_t busy_time;
107
108 cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
109 busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user,
110 kstat_cpu(cpu).cpustat.system);
e08f5f5b 111
8e677ce8
AC
112 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq);
113 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq);
114 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal);
115 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice);
e08f5f5b 116
8e677ce8
AC
117 idle_time = cputime64_sub(cur_wall_time, busy_time);
118 if (wall)
54c9a35d 119 *wall = (cputime64_t)jiffies_to_usecs(cur_wall_time);
e08f5f5b 120
54c9a35d 121 return (cputime64_t)jiffies_to_usecs(idle_time);;
8e677ce8
AC
122}
123
124static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
125{
126 u64 idle_time = get_cpu_idle_time_us(cpu, wall);
127
128 if (idle_time == -1ULL)
129 return get_cpu_idle_time_jiffy(cpu, wall);
130
131 return idle_time;
dac1c1a5
DJ
132}
133
a8d7c3bc
EO
134/* keep track of frequency transitions */
135static int
136dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
137 void *data)
138{
139 struct cpufreq_freqs *freq = data;
245b2e70 140 struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cs_cpu_dbs_info,
a8d7c3bc
EO
141 freq->cpu);
142
f407a08b
AC
143 struct cpufreq_policy *policy;
144
a8d7c3bc
EO
145 if (!this_dbs_info->enable)
146 return 0;
147
f407a08b
AC
148 policy = this_dbs_info->cur_policy;
149
150 /*
151 * we only care if our internally tracked freq moves outside
152 * the 'valid' ranges of freqency available to us otherwise
153 * we do not change it
154 */
155 if (this_dbs_info->requested_freq > policy->max
156 || this_dbs_info->requested_freq < policy->min)
157 this_dbs_info->requested_freq = freq->new;
a8d7c3bc
EO
158
159 return 0;
160}
161
162static struct notifier_block dbs_cpufreq_notifier_block = {
163 .notifier_call = dbs_cpufreq_notifier
164};
165
b9170836 166/************************** sysfs interface ************************/
49b015ce
TR
167static ssize_t show_sampling_rate_max(struct kobject *kobj,
168 struct attribute *attr, char *buf)
b9170836 169{
4f4d1ad6
TR
170 printk_once(KERN_INFO "CPUFREQ: conservative sampling_rate_max "
171 "sysfs file is deprecated - used by: %s\n", current->comm);
cef9615a 172 return sprintf(buf, "%u\n", -1U);
b9170836
DJ
173}
174
49b015ce
TR
175static ssize_t show_sampling_rate_min(struct kobject *kobj,
176 struct attribute *attr, char *buf)
b9170836 177{
cef9615a 178 return sprintf(buf, "%u\n", min_sampling_rate);
b9170836
DJ
179}
180
6dad2a29
BP
181define_one_global_ro(sampling_rate_max);
182define_one_global_ro(sampling_rate_min);
b9170836
DJ
183
184/* cpufreq_conservative Governor Tunables */
185#define show_one(file_name, object) \
186static ssize_t show_##file_name \
49b015ce 187(struct kobject *kobj, struct attribute *attr, char *buf) \
b9170836
DJ
188{ \
189 return sprintf(buf, "%u\n", dbs_tuners_ins.object); \
190}
191show_one(sampling_rate, sampling_rate);
192show_one(sampling_down_factor, sampling_down_factor);
193show_one(up_threshold, up_threshold);
194show_one(down_threshold, down_threshold);
001893cd 195show_one(ignore_nice_load, ignore_nice);
b9170836
DJ
196show_one(freq_step, freq_step);
197
49b015ce
TR
198/*** delete after deprecation time ***/
199#define DEPRECATION_MSG(file_name) \
200 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
201 "interface is deprecated - " #file_name "\n");
202
203#define show_one_old(file_name) \
204static ssize_t show_##file_name##_old \
205(struct cpufreq_policy *unused, char *buf) \
206{ \
207 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
208 "interface is deprecated - " #file_name "\n"); \
209 return show_##file_name(NULL, NULL, buf); \
210}
211show_one_old(sampling_rate);
212show_one_old(sampling_down_factor);
213show_one_old(up_threshold);
214show_one_old(down_threshold);
215show_one_old(ignore_nice_load);
216show_one_old(freq_step);
217show_one_old(sampling_rate_min);
218show_one_old(sampling_rate_max);
219
6dad2a29
BP
220cpufreq_freq_attr_ro_old(sampling_rate_min);
221cpufreq_freq_attr_ro_old(sampling_rate_max);
49b015ce
TR
222
223/*** delete after deprecation time ***/
224
225static ssize_t store_sampling_down_factor(struct kobject *a,
226 struct attribute *b,
227 const char *buf, size_t count)
b9170836
DJ
228{
229 unsigned int input;
230 int ret;
9acef487 231 ret = sscanf(buf, "%u", &input);
8e677ce8 232
2c906b31 233 if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
b9170836
DJ
234 return -EINVAL;
235
3fc54d37 236 mutex_lock(&dbs_mutex);
b9170836 237 dbs_tuners_ins.sampling_down_factor = input;
3fc54d37 238 mutex_unlock(&dbs_mutex);
b9170836
DJ
239
240 return count;
241}
242
49b015ce
TR
243static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
244 const char *buf, size_t count)
b9170836
DJ
245{
246 unsigned int input;
247 int ret;
9acef487 248 ret = sscanf(buf, "%u", &input);
b9170836 249
8e677ce8 250 if (ret != 1)
b9170836 251 return -EINVAL;
8e677ce8
AC
252
253 mutex_lock(&dbs_mutex);
cef9615a 254 dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate);
3fc54d37 255 mutex_unlock(&dbs_mutex);
b9170836
DJ
256
257 return count;
258}
259
49b015ce
TR
260static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
261 const char *buf, size_t count)
b9170836
DJ
262{
263 unsigned int input;
264 int ret;
9acef487 265 ret = sscanf(buf, "%u", &input);
b9170836 266
3fc54d37 267 mutex_lock(&dbs_mutex);
9acef487 268 if (ret != 1 || input > 100 ||
8e677ce8 269 input <= dbs_tuners_ins.down_threshold) {
3fc54d37 270 mutex_unlock(&dbs_mutex);
b9170836
DJ
271 return -EINVAL;
272 }
273
274 dbs_tuners_ins.up_threshold = input;
3fc54d37 275 mutex_unlock(&dbs_mutex);
b9170836
DJ
276
277 return count;
278}
279
49b015ce
TR
280static ssize_t store_down_threshold(struct kobject *a, struct attribute *b,
281 const char *buf, size_t count)
b9170836
DJ
282{
283 unsigned int input;
284 int ret;
9acef487 285 ret = sscanf(buf, "%u", &input);
b9170836 286
3fc54d37 287 mutex_lock(&dbs_mutex);
8e677ce8
AC
288 /* cannot be lower than 11 otherwise freq will not fall */
289 if (ret != 1 || input < 11 || input > 100 ||
290 input >= dbs_tuners_ins.up_threshold) {
3fc54d37 291 mutex_unlock(&dbs_mutex);
b9170836
DJ
292 return -EINVAL;
293 }
294
295 dbs_tuners_ins.down_threshold = input;
3fc54d37 296 mutex_unlock(&dbs_mutex);
b9170836
DJ
297
298 return count;
299}
300
49b015ce
TR
301static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
302 const char *buf, size_t count)
b9170836
DJ
303{
304 unsigned int input;
305 int ret;
306
307 unsigned int j;
18a7247d
DJ
308
309 ret = sscanf(buf, "%u", &input);
310 if (ret != 1)
b9170836
DJ
311 return -EINVAL;
312
18a7247d 313 if (input > 1)
b9170836 314 input = 1;
18a7247d 315
3fc54d37 316 mutex_lock(&dbs_mutex);
18a7247d 317 if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
3fc54d37 318 mutex_unlock(&dbs_mutex);
b9170836
DJ
319 return count;
320 }
321 dbs_tuners_ins.ignore_nice = input;
322
8e677ce8 323 /* we need to re-evaluate prev_cpu_idle */
dac1c1a5 324 for_each_online_cpu(j) {
8e677ce8 325 struct cpu_dbs_info_s *dbs_info;
245b2e70 326 dbs_info = &per_cpu(cs_cpu_dbs_info, j);
8e677ce8
AC
327 dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
328 &dbs_info->prev_cpu_wall);
329 if (dbs_tuners_ins.ignore_nice)
330 dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
b9170836 331 }
3fc54d37 332 mutex_unlock(&dbs_mutex);
b9170836
DJ
333
334 return count;
335}
336
49b015ce
TR
337static ssize_t store_freq_step(struct kobject *a, struct attribute *b,
338 const char *buf, size_t count)
b9170836
DJ
339{
340 unsigned int input;
341 int ret;
18a7247d 342 ret = sscanf(buf, "%u", &input);
b9170836 343
18a7247d 344 if (ret != 1)
b9170836
DJ
345 return -EINVAL;
346
18a7247d 347 if (input > 100)
b9170836 348 input = 100;
18a7247d 349
b9170836
DJ
350 /* no need to test here if freq_step is zero as the user might actually
351 * want this, they would be crazy though :) */
3fc54d37 352 mutex_lock(&dbs_mutex);
b9170836 353 dbs_tuners_ins.freq_step = input;
3fc54d37 354 mutex_unlock(&dbs_mutex);
b9170836
DJ
355
356 return count;
357}
358
6dad2a29
BP
359define_one_global_rw(sampling_rate);
360define_one_global_rw(sampling_down_factor);
361define_one_global_rw(up_threshold);
362define_one_global_rw(down_threshold);
363define_one_global_rw(ignore_nice_load);
364define_one_global_rw(freq_step);
b9170836 365
9acef487 366static struct attribute *dbs_attributes[] = {
b9170836
DJ
367 &sampling_rate_max.attr,
368 &sampling_rate_min.attr,
369 &sampling_rate.attr,
370 &sampling_down_factor.attr,
371 &up_threshold.attr,
372 &down_threshold.attr,
001893cd 373 &ignore_nice_load.attr,
b9170836
DJ
374 &freq_step.attr,
375 NULL
376};
377
378static struct attribute_group dbs_attr_group = {
379 .attrs = dbs_attributes,
380 .name = "conservative",
381};
382
49b015ce
TR
383/*** delete after deprecation time ***/
384
385#define write_one_old(file_name) \
386static ssize_t store_##file_name##_old \
387(struct cpufreq_policy *unused, const char *buf, size_t count) \
388{ \
389 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
390 "interface is deprecated - " #file_name "\n"); \
391 return store_##file_name(NULL, NULL, buf, count); \
392}
393write_one_old(sampling_rate);
394write_one_old(sampling_down_factor);
395write_one_old(up_threshold);
396write_one_old(down_threshold);
397write_one_old(ignore_nice_load);
398write_one_old(freq_step);
399
6dad2a29
BP
400cpufreq_freq_attr_rw_old(sampling_rate);
401cpufreq_freq_attr_rw_old(sampling_down_factor);
402cpufreq_freq_attr_rw_old(up_threshold);
403cpufreq_freq_attr_rw_old(down_threshold);
404cpufreq_freq_attr_rw_old(ignore_nice_load);
405cpufreq_freq_attr_rw_old(freq_step);
49b015ce
TR
406
407static struct attribute *dbs_attributes_old[] = {
408 &sampling_rate_max_old.attr,
409 &sampling_rate_min_old.attr,
410 &sampling_rate_old.attr,
411 &sampling_down_factor_old.attr,
412 &up_threshold_old.attr,
413 &down_threshold_old.attr,
414 &ignore_nice_load_old.attr,
415 &freq_step_old.attr,
416 NULL
417};
418
419static struct attribute_group dbs_attr_group_old = {
420 .attrs = dbs_attributes_old,
421 .name = "conservative",
422};
423
424/*** delete after deprecation time ***/
425
b9170836
DJ
426/************************** sysfs end ************************/
427
8e677ce8 428static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
b9170836 429{
8e677ce8 430 unsigned int load = 0;
fd187aaf 431 unsigned int max_load = 0;
f068c04b 432 unsigned int freq_target;
b9170836 433
8e677ce8
AC
434 struct cpufreq_policy *policy;
435 unsigned int j;
b9170836 436
08a28e2e
AC
437 policy = this_dbs_info->cur_policy;
438
18a7247d 439 /*
8e677ce8
AC
440 * Every sampling_rate, we check, if current idle time is less
441 * than 20% (default), then we try to increase frequency
442 * Every sampling_rate*sampling_down_factor, we check, if current
443 * idle time is more than 80%, then we try to decrease frequency
b9170836 444 *
18a7247d
DJ
445 * Any frequency increase takes it to the maximum frequency.
446 * Frequency reduction happens at minimum steps of
8e677ce8 447 * 5% (default) of maximum frequency
b9170836
DJ
448 */
449
8e677ce8
AC
450 /* Get Absolute Load */
451 for_each_cpu(j, policy->cpus) {
452 struct cpu_dbs_info_s *j_dbs_info;
453 cputime64_t cur_wall_time, cur_idle_time;
454 unsigned int idle_time, wall_time;
b9170836 455
245b2e70 456 j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
8e677ce8
AC
457
458 cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);
459
460 wall_time = (unsigned int) cputime64_sub(cur_wall_time,
461 j_dbs_info->prev_cpu_wall);
462 j_dbs_info->prev_cpu_wall = cur_wall_time;
08a28e2e 463
8e677ce8
AC
464 idle_time = (unsigned int) cputime64_sub(cur_idle_time,
465 j_dbs_info->prev_cpu_idle);
466 j_dbs_info->prev_cpu_idle = cur_idle_time;
b9170836 467
8e677ce8
AC
468 if (dbs_tuners_ins.ignore_nice) {
469 cputime64_t cur_nice;
470 unsigned long cur_nice_jiffies;
471
472 cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice,
473 j_dbs_info->prev_cpu_nice);
474 /*
475 * Assumption: nice time between sampling periods will
476 * be less than 2^32 jiffies for 32 bit sys
477 */
478 cur_nice_jiffies = (unsigned long)
479 cputime64_to_jiffies64(cur_nice);
480
481 j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
482 idle_time += jiffies_to_usecs(cur_nice_jiffies);
483 }
484
485 if (unlikely(!wall_time || wall_time < idle_time))
486 continue;
487
488 load = 100 * (wall_time - idle_time) / wall_time;
fd187aaf
DB
489
490 if (load > max_load)
491 max_load = load;
8e677ce8
AC
492 }
493
494 /*
495 * break out if we 'cannot' reduce the speed as the user might
496 * want freq_step to be zero
497 */
498 if (dbs_tuners_ins.freq_step == 0)
499 return;
b9170836 500
8e677ce8 501 /* Check for frequency increase */
fd187aaf 502 if (max_load > dbs_tuners_ins.up_threshold) {
a159b827 503 this_dbs_info->down_skip = 0;
790d76fa 504
b9170836 505 /* if we are already at full speed then break out early */
a159b827 506 if (this_dbs_info->requested_freq == policy->max)
b9170836 507 return;
18a7247d 508
f068c04b 509 freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
b9170836
DJ
510
511 /* max freq cannot be less than 100. But who knows.... */
f068c04b
DJ
512 if (unlikely(freq_target == 0))
513 freq_target = 5;
18a7247d 514
f068c04b 515 this_dbs_info->requested_freq += freq_target;
a159b827
AC
516 if (this_dbs_info->requested_freq > policy->max)
517 this_dbs_info->requested_freq = policy->max;
b9170836 518
a159b827 519 __cpufreq_driver_target(policy, this_dbs_info->requested_freq,
b9170836 520 CPUFREQ_RELATION_H);
b9170836
DJ
521 return;
522 }
523
8e677ce8
AC
524 /*
525 * The optimal frequency is the frequency that is the lowest that
526 * can support the current CPU usage without triggering the up
527 * policy. To be safe, we focus 10 points under the threshold.
528 */
fd187aaf 529 if (max_load < (dbs_tuners_ins.down_threshold - 10)) {
f068c04b 530 freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
b9170836 531
f068c04b 532 this_dbs_info->requested_freq -= freq_target;
a159b827
AC
533 if (this_dbs_info->requested_freq < policy->min)
534 this_dbs_info->requested_freq = policy->min;
b9170836 535
8e677ce8
AC
536 /*
537 * if we cannot reduce the frequency anymore, break out early
538 */
539 if (policy->cur == policy->min)
540 return;
541
a159b827 542 __cpufreq_driver_target(policy, this_dbs_info->requested_freq,
2c906b31 543 CPUFREQ_RELATION_H);
b9170836
DJ
544 return;
545 }
546}
547
c4028958 548static void do_dbs_timer(struct work_struct *work)
18a7247d 549{
8e677ce8
AC
550 struct cpu_dbs_info_s *dbs_info =
551 container_of(work, struct cpu_dbs_info_s, work.work);
552 unsigned int cpu = dbs_info->cpu;
553
554 /* We want all CPUs to do sampling nearly on same jiffy */
555 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
556
557 delay -= jiffies % delay;
558
ee88415c 559 mutex_lock(&dbs_info->timer_mutex);
8e677ce8
AC
560
561 dbs_check_cpu(dbs_info);
562
563 queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay);
ee88415c 564 mutex_unlock(&dbs_info->timer_mutex);
18a7247d 565}
b9170836 566
8e677ce8 567static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
b9170836 568{
8e677ce8
AC
569 /* We want all CPUs to do sampling nearly on same jiffy */
570 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
571 delay -= jiffies % delay;
572
573 dbs_info->enable = 1;
574 INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
575 queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work,
576 delay);
b9170836
DJ
577}
578
8e677ce8 579static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
b9170836 580{
8e677ce8 581 dbs_info->enable = 0;
b253d2b2 582 cancel_delayed_work_sync(&dbs_info->work);
b9170836
DJ
583}
584
585static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
586 unsigned int event)
587{
588 unsigned int cpu = policy->cpu;
589 struct cpu_dbs_info_s *this_dbs_info;
590 unsigned int j;
914f7c31 591 int rc;
b9170836 592
245b2e70 593 this_dbs_info = &per_cpu(cs_cpu_dbs_info, cpu);
b9170836
DJ
594
595 switch (event) {
596 case CPUFREQ_GOV_START:
18a7247d 597 if ((!cpu_online(cpu)) || (!policy->cur))
b9170836
DJ
598 return -EINVAL;
599
3fc54d37 600 mutex_lock(&dbs_mutex);
914f7c31 601
49b015ce 602 rc = sysfs_create_group(&policy->kobj, &dbs_attr_group_old);
914f7c31
JG
603 if (rc) {
604 mutex_unlock(&dbs_mutex);
605 return rc;
606 }
607
835481d9 608 for_each_cpu(j, policy->cpus) {
b9170836 609 struct cpu_dbs_info_s *j_dbs_info;
245b2e70 610 j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
b9170836 611 j_dbs_info->cur_policy = policy;
18a7247d 612
8e677ce8
AC
613 j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
614 &j_dbs_info->prev_cpu_wall);
615 if (dbs_tuners_ins.ignore_nice) {
616 j_dbs_info->prev_cpu_nice =
617 kstat_cpu(j).cpustat.nice;
618 }
b9170836 619 }
a159b827
AC
620 this_dbs_info->down_skip = 0;
621 this_dbs_info->requested_freq = policy->cur;
914f7c31 622
ee88415c 623 mutex_init(&this_dbs_info->timer_mutex);
b9170836
DJ
624 dbs_enable++;
625 /*
626 * Start the timerschedule work, when this governor
627 * is used for first time
628 */
629 if (dbs_enable == 1) {
630 unsigned int latency;
631 /* policy latency is in nS. Convert it to uS first */
2c906b31
AC
632 latency = policy->cpuinfo.transition_latency / 1000;
633 if (latency == 0)
634 latency = 1;
b9170836 635
49b015ce
TR
636 rc = sysfs_create_group(cpufreq_global_kobject,
637 &dbs_attr_group);
638 if (rc) {
639 mutex_unlock(&dbs_mutex);
640 return rc;
641 }
642
cef9615a
TR
643 /*
644 * conservative does not implement micro like ondemand
645 * governor, thus we are bound to jiffes/HZ
646 */
647 min_sampling_rate =
648 MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10);
649 /* Bring kernel and HW constraints together */
650 min_sampling_rate = max(min_sampling_rate,
651 MIN_LATENCY_MULTIPLIER * latency);
652 dbs_tuners_ins.sampling_rate =
653 max(min_sampling_rate,
654 latency * LATENCY_MULTIPLIER);
b9170836 655
a8d7c3bc
EO
656 cpufreq_register_notifier(
657 &dbs_cpufreq_notifier_block,
658 CPUFREQ_TRANSITION_NOTIFIER);
b9170836 659 }
3fc54d37 660 mutex_unlock(&dbs_mutex);
8e677ce8 661
7d26e2d5 662 dbs_timer_init(this_dbs_info);
663
b9170836
DJ
664 break;
665
666 case CPUFREQ_GOV_STOP:
8e677ce8 667 dbs_timer_exit(this_dbs_info);
7d26e2d5 668
669 mutex_lock(&dbs_mutex);
49b015ce 670 sysfs_remove_group(&policy->kobj, &dbs_attr_group_old);
b9170836 671 dbs_enable--;
ee88415c 672 mutex_destroy(&this_dbs_info->timer_mutex);
8e677ce8 673
b9170836
DJ
674 /*
675 * Stop the timerschedule work, when this governor
676 * is used for first time
677 */
8e677ce8 678 if (dbs_enable == 0)
a8d7c3bc
EO
679 cpufreq_unregister_notifier(
680 &dbs_cpufreq_notifier_block,
681 CPUFREQ_TRANSITION_NOTIFIER);
a8d7c3bc 682
3fc54d37 683 mutex_unlock(&dbs_mutex);
49b015ce
TR
684 if (!dbs_enable)
685 sysfs_remove_group(cpufreq_global_kobject,
686 &dbs_attr_group);
b9170836
DJ
687
688 break;
689
690 case CPUFREQ_GOV_LIMITS:
ee88415c 691 mutex_lock(&this_dbs_info->timer_mutex);
b9170836
DJ
692 if (policy->max < this_dbs_info->cur_policy->cur)
693 __cpufreq_driver_target(
694 this_dbs_info->cur_policy,
18a7247d 695 policy->max, CPUFREQ_RELATION_H);
b9170836
DJ
696 else if (policy->min > this_dbs_info->cur_policy->cur)
697 __cpufreq_driver_target(
698 this_dbs_info->cur_policy,
18a7247d 699 policy->min, CPUFREQ_RELATION_L);
ee88415c 700 mutex_unlock(&this_dbs_info->timer_mutex);
8e677ce8 701
b9170836
DJ
702 break;
703 }
704 return 0;
705}
706
c4d14bc0
SW
707#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
708static
709#endif
1c256245
TR
710struct cpufreq_governor cpufreq_gov_conservative = {
711 .name = "conservative",
712 .governor = cpufreq_governor_dbs,
713 .max_transition_latency = TRANSITION_LATENCY_LIMIT,
714 .owner = THIS_MODULE,
b9170836
DJ
715};
716
717static int __init cpufreq_gov_dbs_init(void)
718{
8e677ce8
AC
719 int err;
720
721 kconservative_wq = create_workqueue("kconservative");
722 if (!kconservative_wq) {
723 printk(KERN_ERR "Creation of kconservative failed\n");
724 return -EFAULT;
725 }
726
727 err = cpufreq_register_governor(&cpufreq_gov_conservative);
728 if (err)
729 destroy_workqueue(kconservative_wq);
730
731 return err;
b9170836
DJ
732}
733
734static void __exit cpufreq_gov_dbs_exit(void)
735{
1c256245 736 cpufreq_unregister_governor(&cpufreq_gov_conservative);
8e677ce8 737 destroy_workqueue(kconservative_wq);
b9170836
DJ
738}
739
740
11a80a9c 741MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>");
9acef487 742MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for "
b9170836
DJ
743 "Low Latency Frequency Transition capable processors "
744 "optimised for use in a battery environment");
9acef487 745MODULE_LICENSE("GPL");
b9170836 746
6915719b
JW
747#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
748fs_initcall(cpufreq_gov_dbs_init);
749#else
b9170836 750module_init(cpufreq_gov_dbs_init);
6915719b 751#endif
b9170836 752module_exit(cpufreq_gov_dbs_exit);