fsstack/ecryptfs: remove unused get_nlinks param to fsstack_copy_attr_all
[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
8e677ce8 181#define define_one_ro(_name) \
49b015ce 182static struct global_attr _name = \
b9170836
DJ
183__ATTR(_name, 0444, show_##_name, NULL)
184
185define_one_ro(sampling_rate_max);
186define_one_ro(sampling_rate_min);
187
188/* cpufreq_conservative Governor Tunables */
189#define show_one(file_name, object) \
190static ssize_t show_##file_name \
49b015ce 191(struct kobject *kobj, struct attribute *attr, char *buf) \
b9170836
DJ
192{ \
193 return sprintf(buf, "%u\n", dbs_tuners_ins.object); \
194}
195show_one(sampling_rate, sampling_rate);
196show_one(sampling_down_factor, sampling_down_factor);
197show_one(up_threshold, up_threshold);
198show_one(down_threshold, down_threshold);
001893cd 199show_one(ignore_nice_load, ignore_nice);
b9170836
DJ
200show_one(freq_step, freq_step);
201
49b015ce
TR
202/*** delete after deprecation time ***/
203#define DEPRECATION_MSG(file_name) \
204 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
205 "interface is deprecated - " #file_name "\n");
206
207#define show_one_old(file_name) \
208static ssize_t show_##file_name##_old \
209(struct cpufreq_policy *unused, char *buf) \
210{ \
211 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
212 "interface is deprecated - " #file_name "\n"); \
213 return show_##file_name(NULL, NULL, buf); \
214}
215show_one_old(sampling_rate);
216show_one_old(sampling_down_factor);
217show_one_old(up_threshold);
218show_one_old(down_threshold);
219show_one_old(ignore_nice_load);
220show_one_old(freq_step);
221show_one_old(sampling_rate_min);
222show_one_old(sampling_rate_max);
223
224#define define_one_ro_old(object, _name) \
225static struct freq_attr object = \
226__ATTR(_name, 0444, show_##_name##_old, NULL)
227
228define_one_ro_old(sampling_rate_min_old, sampling_rate_min);
229define_one_ro_old(sampling_rate_max_old, sampling_rate_max);
230
231/*** delete after deprecation time ***/
232
233static ssize_t store_sampling_down_factor(struct kobject *a,
234 struct attribute *b,
235 const char *buf, size_t count)
b9170836
DJ
236{
237 unsigned int input;
238 int ret;
9acef487 239 ret = sscanf(buf, "%u", &input);
8e677ce8 240
2c906b31 241 if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1)
b9170836
DJ
242 return -EINVAL;
243
3fc54d37 244 mutex_lock(&dbs_mutex);
b9170836 245 dbs_tuners_ins.sampling_down_factor = input;
3fc54d37 246 mutex_unlock(&dbs_mutex);
b9170836
DJ
247
248 return count;
249}
250
49b015ce
TR
251static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
252 const char *buf, size_t count)
b9170836
DJ
253{
254 unsigned int input;
255 int ret;
9acef487 256 ret = sscanf(buf, "%u", &input);
b9170836 257
8e677ce8 258 if (ret != 1)
b9170836 259 return -EINVAL;
8e677ce8
AC
260
261 mutex_lock(&dbs_mutex);
cef9615a 262 dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate);
3fc54d37 263 mutex_unlock(&dbs_mutex);
b9170836
DJ
264
265 return count;
266}
267
49b015ce
TR
268static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
269 const char *buf, size_t count)
b9170836
DJ
270{
271 unsigned int input;
272 int ret;
9acef487 273 ret = sscanf(buf, "%u", &input);
b9170836 274
3fc54d37 275 mutex_lock(&dbs_mutex);
9acef487 276 if (ret != 1 || input > 100 ||
8e677ce8 277 input <= dbs_tuners_ins.down_threshold) {
3fc54d37 278 mutex_unlock(&dbs_mutex);
b9170836
DJ
279 return -EINVAL;
280 }
281
282 dbs_tuners_ins.up_threshold = input;
3fc54d37 283 mutex_unlock(&dbs_mutex);
b9170836
DJ
284
285 return count;
286}
287
49b015ce
TR
288static ssize_t store_down_threshold(struct kobject *a, struct attribute *b,
289 const char *buf, size_t count)
b9170836
DJ
290{
291 unsigned int input;
292 int ret;
9acef487 293 ret = sscanf(buf, "%u", &input);
b9170836 294
3fc54d37 295 mutex_lock(&dbs_mutex);
8e677ce8
AC
296 /* cannot be lower than 11 otherwise freq will not fall */
297 if (ret != 1 || input < 11 || input > 100 ||
298 input >= dbs_tuners_ins.up_threshold) {
3fc54d37 299 mutex_unlock(&dbs_mutex);
b9170836
DJ
300 return -EINVAL;
301 }
302
303 dbs_tuners_ins.down_threshold = input;
3fc54d37 304 mutex_unlock(&dbs_mutex);
b9170836
DJ
305
306 return count;
307}
308
49b015ce
TR
309static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
310 const char *buf, size_t count)
b9170836
DJ
311{
312 unsigned int input;
313 int ret;
314
315 unsigned int j;
18a7247d
DJ
316
317 ret = sscanf(buf, "%u", &input);
318 if (ret != 1)
b9170836
DJ
319 return -EINVAL;
320
18a7247d 321 if (input > 1)
b9170836 322 input = 1;
18a7247d 323
3fc54d37 324 mutex_lock(&dbs_mutex);
18a7247d 325 if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
3fc54d37 326 mutex_unlock(&dbs_mutex);
b9170836
DJ
327 return count;
328 }
329 dbs_tuners_ins.ignore_nice = input;
330
8e677ce8 331 /* we need to re-evaluate prev_cpu_idle */
dac1c1a5 332 for_each_online_cpu(j) {
8e677ce8 333 struct cpu_dbs_info_s *dbs_info;
245b2e70 334 dbs_info = &per_cpu(cs_cpu_dbs_info, j);
8e677ce8
AC
335 dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
336 &dbs_info->prev_cpu_wall);
337 if (dbs_tuners_ins.ignore_nice)
338 dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
b9170836 339 }
3fc54d37 340 mutex_unlock(&dbs_mutex);
b9170836
DJ
341
342 return count;
343}
344
49b015ce
TR
345static ssize_t store_freq_step(struct kobject *a, struct attribute *b,
346 const char *buf, size_t count)
b9170836
DJ
347{
348 unsigned int input;
349 int ret;
18a7247d 350 ret = sscanf(buf, "%u", &input);
b9170836 351
18a7247d 352 if (ret != 1)
b9170836
DJ
353 return -EINVAL;
354
18a7247d 355 if (input > 100)
b9170836 356 input = 100;
18a7247d 357
b9170836
DJ
358 /* no need to test here if freq_step is zero as the user might actually
359 * want this, they would be crazy though :) */
3fc54d37 360 mutex_lock(&dbs_mutex);
b9170836 361 dbs_tuners_ins.freq_step = input;
3fc54d37 362 mutex_unlock(&dbs_mutex);
b9170836
DJ
363
364 return count;
365}
366
367#define define_one_rw(_name) \
49b015ce 368static struct global_attr _name = \
b9170836
DJ
369__ATTR(_name, 0644, show_##_name, store_##_name)
370
371define_one_rw(sampling_rate);
372define_one_rw(sampling_down_factor);
373define_one_rw(up_threshold);
374define_one_rw(down_threshold);
001893cd 375define_one_rw(ignore_nice_load);
b9170836
DJ
376define_one_rw(freq_step);
377
9acef487 378static struct attribute *dbs_attributes[] = {
b9170836
DJ
379 &sampling_rate_max.attr,
380 &sampling_rate_min.attr,
381 &sampling_rate.attr,
382 &sampling_down_factor.attr,
383 &up_threshold.attr,
384 &down_threshold.attr,
001893cd 385 &ignore_nice_load.attr,
b9170836
DJ
386 &freq_step.attr,
387 NULL
388};
389
390static struct attribute_group dbs_attr_group = {
391 .attrs = dbs_attributes,
392 .name = "conservative",
393};
394
49b015ce
TR
395/*** delete after deprecation time ***/
396
397#define write_one_old(file_name) \
398static ssize_t store_##file_name##_old \
399(struct cpufreq_policy *unused, const char *buf, size_t count) \
400{ \
401 printk_once(KERN_INFO "CPUFREQ: Per core conservative sysfs " \
402 "interface is deprecated - " #file_name "\n"); \
403 return store_##file_name(NULL, NULL, buf, count); \
404}
405write_one_old(sampling_rate);
406write_one_old(sampling_down_factor);
407write_one_old(up_threshold);
408write_one_old(down_threshold);
409write_one_old(ignore_nice_load);
410write_one_old(freq_step);
411
412#define define_one_rw_old(object, _name) \
413static struct freq_attr object = \
414__ATTR(_name, 0644, show_##_name##_old, store_##_name##_old)
415
416define_one_rw_old(sampling_rate_old, sampling_rate);
417define_one_rw_old(sampling_down_factor_old, sampling_down_factor);
418define_one_rw_old(up_threshold_old, up_threshold);
419define_one_rw_old(down_threshold_old, down_threshold);
420define_one_rw_old(ignore_nice_load_old, ignore_nice_load);
421define_one_rw_old(freq_step_old, freq_step);
422
423static struct attribute *dbs_attributes_old[] = {
424 &sampling_rate_max_old.attr,
425 &sampling_rate_min_old.attr,
426 &sampling_rate_old.attr,
427 &sampling_down_factor_old.attr,
428 &up_threshold_old.attr,
429 &down_threshold_old.attr,
430 &ignore_nice_load_old.attr,
431 &freq_step_old.attr,
432 NULL
433};
434
435static struct attribute_group dbs_attr_group_old = {
436 .attrs = dbs_attributes_old,
437 .name = "conservative",
438};
439
440/*** delete after deprecation time ***/
441
b9170836
DJ
442/************************** sysfs end ************************/
443
8e677ce8 444static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
b9170836 445{
8e677ce8 446 unsigned int load = 0;
f068c04b 447 unsigned int freq_target;
b9170836 448
8e677ce8
AC
449 struct cpufreq_policy *policy;
450 unsigned int j;
b9170836 451
08a28e2e
AC
452 policy = this_dbs_info->cur_policy;
453
18a7247d 454 /*
8e677ce8
AC
455 * Every sampling_rate, we check, if current idle time is less
456 * than 20% (default), then we try to increase frequency
457 * Every sampling_rate*sampling_down_factor, we check, if current
458 * idle time is more than 80%, then we try to decrease frequency
b9170836 459 *
18a7247d
DJ
460 * Any frequency increase takes it to the maximum frequency.
461 * Frequency reduction happens at minimum steps of
8e677ce8 462 * 5% (default) of maximum frequency
b9170836
DJ
463 */
464
8e677ce8
AC
465 /* Get Absolute Load */
466 for_each_cpu(j, policy->cpus) {
467 struct cpu_dbs_info_s *j_dbs_info;
468 cputime64_t cur_wall_time, cur_idle_time;
469 unsigned int idle_time, wall_time;
b9170836 470
245b2e70 471 j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
8e677ce8
AC
472
473 cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);
474
475 wall_time = (unsigned int) cputime64_sub(cur_wall_time,
476 j_dbs_info->prev_cpu_wall);
477 j_dbs_info->prev_cpu_wall = cur_wall_time;
08a28e2e 478
8e677ce8
AC
479 idle_time = (unsigned int) cputime64_sub(cur_idle_time,
480 j_dbs_info->prev_cpu_idle);
481 j_dbs_info->prev_cpu_idle = cur_idle_time;
b9170836 482
8e677ce8
AC
483 if (dbs_tuners_ins.ignore_nice) {
484 cputime64_t cur_nice;
485 unsigned long cur_nice_jiffies;
486
487 cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice,
488 j_dbs_info->prev_cpu_nice);
489 /*
490 * Assumption: nice time between sampling periods will
491 * be less than 2^32 jiffies for 32 bit sys
492 */
493 cur_nice_jiffies = (unsigned long)
494 cputime64_to_jiffies64(cur_nice);
495
496 j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
497 idle_time += jiffies_to_usecs(cur_nice_jiffies);
498 }
499
500 if (unlikely(!wall_time || wall_time < idle_time))
501 continue;
502
503 load = 100 * (wall_time - idle_time) / wall_time;
504 }
505
506 /*
507 * break out if we 'cannot' reduce the speed as the user might
508 * want freq_step to be zero
509 */
510 if (dbs_tuners_ins.freq_step == 0)
511 return;
b9170836 512
8e677ce8
AC
513 /* Check for frequency increase */
514 if (load > dbs_tuners_ins.up_threshold) {
a159b827 515 this_dbs_info->down_skip = 0;
790d76fa 516
b9170836 517 /* if we are already at full speed then break out early */
a159b827 518 if (this_dbs_info->requested_freq == policy->max)
b9170836 519 return;
18a7247d 520
f068c04b 521 freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
b9170836
DJ
522
523 /* max freq cannot be less than 100. But who knows.... */
f068c04b
DJ
524 if (unlikely(freq_target == 0))
525 freq_target = 5;
18a7247d 526
f068c04b 527 this_dbs_info->requested_freq += freq_target;
a159b827
AC
528 if (this_dbs_info->requested_freq > policy->max)
529 this_dbs_info->requested_freq = policy->max;
b9170836 530
a159b827 531 __cpufreq_driver_target(policy, this_dbs_info->requested_freq,
b9170836 532 CPUFREQ_RELATION_H);
b9170836
DJ
533 return;
534 }
535
8e677ce8
AC
536 /*
537 * The optimal frequency is the frequency that is the lowest that
538 * can support the current CPU usage without triggering the up
539 * policy. To be safe, we focus 10 points under the threshold.
540 */
541 if (load < (dbs_tuners_ins.down_threshold - 10)) {
f068c04b 542 freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100;
b9170836 543
f068c04b 544 this_dbs_info->requested_freq -= freq_target;
a159b827
AC
545 if (this_dbs_info->requested_freq < policy->min)
546 this_dbs_info->requested_freq = policy->min;
b9170836 547
8e677ce8
AC
548 /*
549 * if we cannot reduce the frequency anymore, break out early
550 */
551 if (policy->cur == policy->min)
552 return;
553
a159b827 554 __cpufreq_driver_target(policy, this_dbs_info->requested_freq,
2c906b31 555 CPUFREQ_RELATION_H);
b9170836
DJ
556 return;
557 }
558}
559
c4028958 560static void do_dbs_timer(struct work_struct *work)
18a7247d 561{
8e677ce8
AC
562 struct cpu_dbs_info_s *dbs_info =
563 container_of(work, struct cpu_dbs_info_s, work.work);
564 unsigned int cpu = dbs_info->cpu;
565
566 /* We want all CPUs to do sampling nearly on same jiffy */
567 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
568
569 delay -= jiffies % delay;
570
ee88415c 571 mutex_lock(&dbs_info->timer_mutex);
8e677ce8
AC
572
573 dbs_check_cpu(dbs_info);
574
575 queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay);
ee88415c 576 mutex_unlock(&dbs_info->timer_mutex);
18a7247d 577}
b9170836 578
8e677ce8 579static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
b9170836 580{
8e677ce8
AC
581 /* We want all CPUs to do sampling nearly on same jiffy */
582 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
583 delay -= jiffies % delay;
584
585 dbs_info->enable = 1;
586 INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
587 queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work,
588 delay);
b9170836
DJ
589}
590
8e677ce8 591static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
b9170836 592{
8e677ce8 593 dbs_info->enable = 0;
b253d2b2 594 cancel_delayed_work_sync(&dbs_info->work);
b9170836
DJ
595}
596
597static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
598 unsigned int event)
599{
600 unsigned int cpu = policy->cpu;
601 struct cpu_dbs_info_s *this_dbs_info;
602 unsigned int j;
914f7c31 603 int rc;
b9170836 604
245b2e70 605 this_dbs_info = &per_cpu(cs_cpu_dbs_info, cpu);
b9170836
DJ
606
607 switch (event) {
608 case CPUFREQ_GOV_START:
18a7247d 609 if ((!cpu_online(cpu)) || (!policy->cur))
b9170836
DJ
610 return -EINVAL;
611
3fc54d37 612 mutex_lock(&dbs_mutex);
914f7c31 613
49b015ce 614 rc = sysfs_create_group(&policy->kobj, &dbs_attr_group_old);
914f7c31
JG
615 if (rc) {
616 mutex_unlock(&dbs_mutex);
617 return rc;
618 }
619
835481d9 620 for_each_cpu(j, policy->cpus) {
b9170836 621 struct cpu_dbs_info_s *j_dbs_info;
245b2e70 622 j_dbs_info = &per_cpu(cs_cpu_dbs_info, j);
b9170836 623 j_dbs_info->cur_policy = policy;
18a7247d 624
8e677ce8
AC
625 j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
626 &j_dbs_info->prev_cpu_wall);
627 if (dbs_tuners_ins.ignore_nice) {
628 j_dbs_info->prev_cpu_nice =
629 kstat_cpu(j).cpustat.nice;
630 }
b9170836 631 }
a159b827
AC
632 this_dbs_info->down_skip = 0;
633 this_dbs_info->requested_freq = policy->cur;
914f7c31 634
ee88415c 635 mutex_init(&this_dbs_info->timer_mutex);
b9170836
DJ
636 dbs_enable++;
637 /*
638 * Start the timerschedule work, when this governor
639 * is used for first time
640 */
641 if (dbs_enable == 1) {
642 unsigned int latency;
643 /* policy latency is in nS. Convert it to uS first */
2c906b31
AC
644 latency = policy->cpuinfo.transition_latency / 1000;
645 if (latency == 0)
646 latency = 1;
b9170836 647
49b015ce
TR
648 rc = sysfs_create_group(cpufreq_global_kobject,
649 &dbs_attr_group);
650 if (rc) {
651 mutex_unlock(&dbs_mutex);
652 return rc;
653 }
654
cef9615a
TR
655 /*
656 * conservative does not implement micro like ondemand
657 * governor, thus we are bound to jiffes/HZ
658 */
659 min_sampling_rate =
660 MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10);
661 /* Bring kernel and HW constraints together */
662 min_sampling_rate = max(min_sampling_rate,
663 MIN_LATENCY_MULTIPLIER * latency);
664 dbs_tuners_ins.sampling_rate =
665 max(min_sampling_rate,
666 latency * LATENCY_MULTIPLIER);
b9170836 667
a8d7c3bc
EO
668 cpufreq_register_notifier(
669 &dbs_cpufreq_notifier_block,
670 CPUFREQ_TRANSITION_NOTIFIER);
b9170836 671 }
3fc54d37 672 mutex_unlock(&dbs_mutex);
8e677ce8 673
7d26e2d5 674 dbs_timer_init(this_dbs_info);
675
b9170836
DJ
676 break;
677
678 case CPUFREQ_GOV_STOP:
8e677ce8 679 dbs_timer_exit(this_dbs_info);
7d26e2d5 680
681 mutex_lock(&dbs_mutex);
49b015ce 682 sysfs_remove_group(&policy->kobj, &dbs_attr_group_old);
b9170836 683 dbs_enable--;
ee88415c 684 mutex_destroy(&this_dbs_info->timer_mutex);
8e677ce8 685
b9170836
DJ
686 /*
687 * Stop the timerschedule work, when this governor
688 * is used for first time
689 */
8e677ce8 690 if (dbs_enable == 0)
a8d7c3bc
EO
691 cpufreq_unregister_notifier(
692 &dbs_cpufreq_notifier_block,
693 CPUFREQ_TRANSITION_NOTIFIER);
a8d7c3bc 694
3fc54d37 695 mutex_unlock(&dbs_mutex);
49b015ce
TR
696 if (!dbs_enable)
697 sysfs_remove_group(cpufreq_global_kobject,
698 &dbs_attr_group);
b9170836
DJ
699
700 break;
701
702 case CPUFREQ_GOV_LIMITS:
ee88415c 703 mutex_lock(&this_dbs_info->timer_mutex);
b9170836
DJ
704 if (policy->max < this_dbs_info->cur_policy->cur)
705 __cpufreq_driver_target(
706 this_dbs_info->cur_policy,
18a7247d 707 policy->max, CPUFREQ_RELATION_H);
b9170836
DJ
708 else if (policy->min > this_dbs_info->cur_policy->cur)
709 __cpufreq_driver_target(
710 this_dbs_info->cur_policy,
18a7247d 711 policy->min, CPUFREQ_RELATION_L);
ee88415c 712 mutex_unlock(&this_dbs_info->timer_mutex);
8e677ce8 713
b9170836
DJ
714 break;
715 }
716 return 0;
717}
718
c4d14bc0
SW
719#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
720static
721#endif
1c256245
TR
722struct cpufreq_governor cpufreq_gov_conservative = {
723 .name = "conservative",
724 .governor = cpufreq_governor_dbs,
725 .max_transition_latency = TRANSITION_LATENCY_LIMIT,
726 .owner = THIS_MODULE,
b9170836
DJ
727};
728
729static int __init cpufreq_gov_dbs_init(void)
730{
8e677ce8
AC
731 int err;
732
733 kconservative_wq = create_workqueue("kconservative");
734 if (!kconservative_wq) {
735 printk(KERN_ERR "Creation of kconservative failed\n");
736 return -EFAULT;
737 }
738
739 err = cpufreq_register_governor(&cpufreq_gov_conservative);
740 if (err)
741 destroy_workqueue(kconservative_wq);
742
743 return err;
b9170836
DJ
744}
745
746static void __exit cpufreq_gov_dbs_exit(void)
747{
1c256245 748 cpufreq_unregister_governor(&cpufreq_gov_conservative);
8e677ce8 749 destroy_workqueue(kconservative_wq);
b9170836
DJ
750}
751
752
11a80a9c 753MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>");
9acef487 754MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for "
b9170836
DJ
755 "Low Latency Frequency Transition capable processors "
756 "optimised for use in a battery environment");
9acef487 757MODULE_LICENSE("GPL");
b9170836 758
6915719b
JW
759#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE
760fs_initcall(cpufreq_gov_dbs_init);
761#else
b9170836 762module_init(cpufreq_gov_dbs_init);
6915719b 763#endif
b9170836 764module_exit(cpufreq_gov_dbs_exit);