blob: e1314212d8d4e2d3789c2ae4658e504963bc6a90 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * drivers/cpufreq/cpufreq_ondemand.c
3 *
4 * Copyright (C) 2001 Russell King
5 * (C) 2003 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>.
6 * Jun Nakajima <jun.nakajima@intel.com>
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 */
12
13#include <linux/kernel.h>
14#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/cpufreq.h>
Andrew Morton138a01282006-06-23 03:31:19 -070017#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/jiffies.h>
19#include <linux/kernel_stat.h>
akpm@osdl.org3fc54d32006-01-13 15:54:22 -080020#include <linux/mutex.h>
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -070021#include <linux/hrtimer.h>
22#include <linux/tick.h>
23#include <linux/ktime.h>
Thomas Renninger9411b4e2009-02-04 11:54:04 +010024#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26/*
27 * dbs is used in this file as a shortform for demandbased switching
28 * It helps to keep variable names smaller, simpler
29 */
30
venkatesh.pallipadi@intel.come9d95bf2008-08-04 11:59:10 -070031#define DEF_FREQUENCY_DOWN_DIFFERENTIAL (10)
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#define DEF_FREQUENCY_UP_THRESHOLD (80)
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -070033#define MICRO_FREQUENCY_DOWN_DIFFERENTIAL (3)
34#define MICRO_FREQUENCY_UP_THRESHOLD (95)
Thomas Renningercef96152009-04-22 13:48:29 +020035#define MICRO_FREQUENCY_MIN_SAMPLE_RATE (10000)
Dave Jonesc29f1402005-05-31 19:03:50 -070036#define MIN_FREQUENCY_UP_THRESHOLD (11)
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#define MAX_FREQUENCY_UP_THRESHOLD (100)
38
Dave Jones32ee8c32006-02-28 00:43:23 -050039/*
40 * The polling frequency of this governor depends on the capability of
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 * the processor. Default polling frequency is 1000 times the transition
Dave Jones32ee8c32006-02-28 00:43:23 -050042 * latency of the processor. The governor will work on any processor with
43 * transition latency <= 10mS, using appropriate sampling
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * rate.
45 * For CPUs with transition latency > 10mS (mostly drivers with CPUFREQ_ETERNAL)
46 * this governor will not work.
47 * All times here are in uS.
48 */
Dave Jonesdf8b59b2005-09-20 12:39:35 -070049#define MIN_SAMPLING_RATE_RATIO (2)
Thomas Renninger112124a2009-02-04 11:55:12 +010050
Thomas Renningercef96152009-04-22 13:48:29 +020051static unsigned int min_sampling_rate;
52
Thomas Renninger112124a2009-02-04 11:55:12 +010053#define LATENCY_MULTIPLIER (1000)
Thomas Renningercef96152009-04-22 13:48:29 +020054#define MIN_LATENCY_MULTIPLIER (100)
Thomas Renninger1c256242007-10-02 13:28:12 -070055#define TRANSITION_LATENCY_LIMIT (10 * 1000 * 1000)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
David Howellsc4028952006-11-22 14:57:56 +000057static void do_dbs_timer(struct work_struct *work);
Thomas Renninger0e625ac2009-07-24 15:25:06 +020058static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
59 unsigned int event);
60
61#ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
62static
63#endif
64struct cpufreq_governor cpufreq_gov_ondemand = {
65 .name = "ondemand",
66 .governor = cpufreq_governor_dbs,
67 .max_transition_latency = TRANSITION_LATENCY_LIMIT,
68 .owner = THIS_MODULE,
69};
David Howellsc4028952006-11-22 14:57:56 +000070
71/* Sampling types */
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -080072enum {DBS_NORMAL_SAMPLE, DBS_SUB_SAMPLE};
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74struct cpu_dbs_info_s {
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -070075 cputime64_t prev_cpu_idle;
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -070076 cputime64_t prev_cpu_iowait;
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -070077 cputime64_t prev_cpu_wall;
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -070078 cputime64_t prev_cpu_nice;
Dave Jones32ee8c32006-02-28 00:43:23 -050079 struct cpufreq_policy *cur_policy;
Dave Jones2b03f892009-01-18 01:43:44 -050080 struct delayed_work work;
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +040081 struct cpufreq_frequency_table *freq_table;
82 unsigned int freq_lo;
83 unsigned int freq_lo_jiffies;
84 unsigned int freq_hi_jiffies;
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -080085 int cpu;
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -070086 unsigned int sample_type:1;
87 /*
88 * percpu mutex that serializes governor limit change with
89 * do_dbs_timer invocation. We do not want do_dbs_timer to run
90 * when user is changing the governor or limits.
91 */
92 struct mutex timer_mutex;
Linus Torvalds1da177e2005-04-16 15:20:36 -070093};
Tejun Heo245b2e72009-06-24 15:13:48 +090094static DEFINE_PER_CPU(struct cpu_dbs_info_s, od_cpu_dbs_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96static unsigned int dbs_enable; /* number of CPUs using this policy */
97
Venkatesh Pallipadi4ec223d2006-06-21 15:18:34 -070098/*
venkatesh.pallipadi@intel.com7d26e2d2009-07-02 17:08:30 -070099 * dbs_mutex protects data in dbs_tuners_ins from concurrent changes on
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700100 * different CPUs. It protects dbs_enable in governor start/stop.
Venkatesh Pallipadi4ec223d2006-06-21 15:18:34 -0700101 */
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700102static DEFINE_MUTEX(dbs_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Venkatesh Pallipadi2f8a8352006-06-28 13:51:19 -0700104static struct workqueue_struct *kondemand_wq;
Andi Kleen6810b542006-05-08 15:17:31 +0200105
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400106static struct dbs_tuners {
Dave Jones32ee8c32006-02-28 00:43:23 -0500107 unsigned int sampling_rate;
Dave Jones32ee8c32006-02-28 00:43:23 -0500108 unsigned int up_threshold;
venkatesh.pallipadi@intel.come9d95bf2008-08-04 11:59:10 -0700109 unsigned int down_differential;
Dave Jones32ee8c32006-02-28 00:43:23 -0500110 unsigned int ignore_nice;
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400111 unsigned int powersave_bias;
Arjan van de Ven19379b12010-05-09 08:26:51 -0700112 unsigned int io_is_busy;
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400113} dbs_tuners_ins = {
Dave Jones32ee8c32006-02-28 00:43:23 -0500114 .up_threshold = DEF_FREQUENCY_UP_THRESHOLD,
venkatesh.pallipadi@intel.come9d95bf2008-08-04 11:59:10 -0700115 .down_differential = DEF_FREQUENCY_DOWN_DIFFERENTIAL,
Eric Piel9cbad612006-03-10 11:35:27 +0200116 .ignore_nice = 0,
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400117 .powersave_bias = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118};
119
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700120static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu,
121 cputime64_t *wall)
Dave Jonesdac1c1a2005-05-31 19:03:49 -0700122{
Venki Pallipadiea487612007-06-20 14:26:24 -0700123 cputime64_t idle_time;
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700124 cputime64_t cur_wall_time;
Venki Pallipadiea487612007-06-20 14:26:24 -0700125 cputime64_t busy_time;
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700126
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700127 cur_wall_time = jiffies64_to_cputime64(get_jiffies_64());
Venki Pallipadiea487612007-06-20 14:26:24 -0700128 busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user,
129 kstat_cpu(cpu).cpustat.system);
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700130
Venki Pallipadiea487612007-06-20 14:26:24 -0700131 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq);
132 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq);
133 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal);
Venkatesh Pallipadi1ca3abd2009-01-23 09:25:02 -0500134 busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice);
Venki Pallipadiea487612007-06-20 14:26:24 -0700135
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700136 idle_time = cputime64_sub(cur_wall_time, busy_time);
137 if (wall)
Pallipadi, Venkatesh54c9a352009-11-11 16:50:29 -0800138 *wall = (cputime64_t)jiffies_to_usecs(cur_wall_time);
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700139
Pallipadi, Venkatesh54c9a352009-11-11 16:50:29 -0800140 return (cputime64_t)jiffies_to_usecs(idle_time);
Dave Jonesdac1c1a2005-05-31 19:03:49 -0700141}
142
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700143static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall)
144{
145 u64 idle_time = get_cpu_idle_time_us(cpu, wall);
146
147 if (idle_time == -1ULL)
148 return get_cpu_idle_time_jiffy(cpu, wall);
149
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700150 return idle_time;
151}
152
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700153static inline cputime64_t get_cpu_iowait_time(unsigned int cpu, cputime64_t *wall)
154{
155 u64 iowait_time = get_cpu_iowait_time_us(cpu, wall);
156
157 if (iowait_time == -1ULL)
158 return 0;
159
160 return iowait_time;
161}
162
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400163/*
164 * Find right freq to be set now with powersave_bias on.
165 * Returns the freq_hi to be used right now and will set freq_hi_jiffies,
166 * freq_lo, and freq_lo_jiffies in percpu area for averaging freqs.
167 */
Adrian Bunkb5ecf602006-08-13 23:00:08 +0200168static unsigned int powersave_bias_target(struct cpufreq_policy *policy,
169 unsigned int freq_next,
170 unsigned int relation)
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400171{
172 unsigned int freq_req, freq_reduc, freq_avg;
173 unsigned int freq_hi, freq_lo;
174 unsigned int index = 0;
175 unsigned int jiffies_total, jiffies_hi, jiffies_lo;
Tejun Heo245b2e72009-06-24 15:13:48 +0900176 struct cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info,
177 policy->cpu);
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400178
179 if (!dbs_info->freq_table) {
180 dbs_info->freq_lo = 0;
181 dbs_info->freq_lo_jiffies = 0;
182 return freq_next;
183 }
184
185 cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next,
186 relation, &index);
187 freq_req = dbs_info->freq_table[index].frequency;
188 freq_reduc = freq_req * dbs_tuners_ins.powersave_bias / 1000;
189 freq_avg = freq_req - freq_reduc;
190
191 /* Find freq bounds for freq_avg in freq_table */
192 index = 0;
193 cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
194 CPUFREQ_RELATION_H, &index);
195 freq_lo = dbs_info->freq_table[index].frequency;
196 index = 0;
197 cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg,
198 CPUFREQ_RELATION_L, &index);
199 freq_hi = dbs_info->freq_table[index].frequency;
200
201 /* Find out how long we have to be in hi and lo freqs */
202 if (freq_hi == freq_lo) {
203 dbs_info->freq_lo = 0;
204 dbs_info->freq_lo_jiffies = 0;
205 return freq_lo;
206 }
207 jiffies_total = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
208 jiffies_hi = (freq_avg - freq_lo) * jiffies_total;
209 jiffies_hi += ((freq_hi - freq_lo) / 2);
210 jiffies_hi /= (freq_hi - freq_lo);
211 jiffies_lo = jiffies_total - jiffies_hi;
212 dbs_info->freq_lo = freq_lo;
213 dbs_info->freq_lo_jiffies = jiffies_lo;
214 dbs_info->freq_hi_jiffies = jiffies_hi;
215 return freq_hi;
216}
217
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700218static void ondemand_powersave_bias_init_cpu(int cpu)
219{
Tejun Heo384be2b2009-08-14 14:41:02 +0900220 struct cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu);
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700221 dbs_info->freq_table = cpufreq_frequency_get_table(cpu);
222 dbs_info->freq_lo = 0;
223}
224
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400225static void ondemand_powersave_bias_init(void)
226{
227 int i;
228 for_each_online_cpu(i) {
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700229 ondemand_powersave_bias_init_cpu(i);
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400230 }
231}
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233/************************** sysfs interface ************************/
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200234
235static ssize_t show_sampling_rate_max(struct kobject *kobj,
236 struct attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Thomas Renninger4f4d1ad2009-04-22 13:48:31 +0200238 printk_once(KERN_INFO "CPUFREQ: ondemand sampling_rate_max "
239 "sysfs file is deprecated - used by: %s\n", current->comm);
Thomas Renningercef96152009-04-22 13:48:29 +0200240 return sprintf(buf, "%u\n", -1U);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200243static ssize_t show_sampling_rate_min(struct kobject *kobj,
244 struct attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Thomas Renningercef96152009-04-22 13:48:29 +0200246 return sprintf(buf, "%u\n", min_sampling_rate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
Borislav Petkov6dad2a22010-03-31 21:56:46 +0200249define_one_global_ro(sampling_rate_max);
250define_one_global_ro(sampling_rate_min);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252/* cpufreq_ondemand Governor Tunables */
253#define show_one(file_name, object) \
254static ssize_t show_##file_name \
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200255(struct kobject *kobj, struct attribute *attr, char *buf) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{ \
257 return sprintf(buf, "%u\n", dbs_tuners_ins.object); \
258}
259show_one(sampling_rate, sampling_rate);
Arjan van de Ven19379b12010-05-09 08:26:51 -0700260show_one(io_is_busy, io_is_busy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261show_one(up_threshold, up_threshold);
Alexander Clouter001893c2005-12-01 01:09:25 -0800262show_one(ignore_nice_load, ignore_nice);
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400263show_one(powersave_bias, powersave_bias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200265/*** delete after deprecation time ***/
266
267#define DEPRECATION_MSG(file_name) \
268 printk_once(KERN_INFO "CPUFREQ: Per core ondemand sysfs " \
269 "interface is deprecated - " #file_name "\n");
270
271#define show_one_old(file_name) \
272static ssize_t show_##file_name##_old \
273(struct cpufreq_policy *unused, char *buf) \
274{ \
275 printk_once(KERN_INFO "CPUFREQ: Per core ondemand sysfs " \
276 "interface is deprecated - " #file_name "\n"); \
277 return show_##file_name(NULL, NULL, buf); \
278}
279show_one_old(sampling_rate);
280show_one_old(up_threshold);
281show_one_old(ignore_nice_load);
282show_one_old(powersave_bias);
283show_one_old(sampling_rate_min);
284show_one_old(sampling_rate_max);
285
Borislav Petkov6dad2a22010-03-31 21:56:46 +0200286cpufreq_freq_attr_ro_old(sampling_rate_min);
287cpufreq_freq_attr_ro_old(sampling_rate_max);
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200288
289/*** delete after deprecation time ***/
290
291static ssize_t store_sampling_rate(struct kobject *a, struct attribute *b,
292 const char *buf, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
294 unsigned int input;
295 int ret;
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700296 ret = sscanf(buf, "%u", &input);
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700297 if (ret != 1)
298 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800300 mutex_lock(&dbs_mutex);
Thomas Renningercef96152009-04-22 13:48:29 +0200301 dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate);
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800302 mutex_unlock(&dbs_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304 return count;
305}
306
Arjan van de Ven19379b12010-05-09 08:26:51 -0700307static ssize_t store_io_is_busy(struct kobject *a, struct attribute *b,
308 const char *buf, size_t count)
309{
310 unsigned int input;
311 int ret;
312
313 ret = sscanf(buf, "%u", &input);
314 if (ret != 1)
315 return -EINVAL;
316
317 mutex_lock(&dbs_mutex);
318 dbs_tuners_ins.io_is_busy = !!input;
319 mutex_unlock(&dbs_mutex);
320
321 return count;
322}
323
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200324static ssize_t store_up_threshold(struct kobject *a, struct attribute *b,
325 const char *buf, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326{
327 unsigned int input;
328 int ret;
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700329 ret = sscanf(buf, "%u", &input);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Dave Jones32ee8c32006-02-28 00:43:23 -0500331 if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD ||
Dave Jonesc29f1402005-05-31 19:03:50 -0700332 input < MIN_FREQUENCY_UP_THRESHOLD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 return -EINVAL;
334 }
335
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700336 mutex_lock(&dbs_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 dbs_tuners_ins.up_threshold = input;
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800338 mutex_unlock(&dbs_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 return count;
341}
342
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200343static ssize_t store_ignore_nice_load(struct kobject *a, struct attribute *b,
344 const char *buf, size_t count)
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700345{
346 unsigned int input;
347 int ret;
348
349 unsigned int j;
Dave Jones32ee8c32006-02-28 00:43:23 -0500350
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700351 ret = sscanf(buf, "%u", &input);
Dave Jones2b03f892009-01-18 01:43:44 -0500352 if (ret != 1)
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700353 return -EINVAL;
354
Dave Jones2b03f892009-01-18 01:43:44 -0500355 if (input > 1)
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700356 input = 1;
Dave Jones32ee8c32006-02-28 00:43:23 -0500357
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800358 mutex_lock(&dbs_mutex);
Dave Jones2b03f892009-01-18 01:43:44 -0500359 if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800360 mutex_unlock(&dbs_mutex);
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700361 return count;
362 }
363 dbs_tuners_ins.ignore_nice = input;
364
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700365 /* we need to re-evaluate prev_cpu_idle */
Dave Jonesdac1c1a2005-05-31 19:03:49 -0700366 for_each_online_cpu(j) {
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700367 struct cpu_dbs_info_s *dbs_info;
Tejun Heo245b2e72009-06-24 15:13:48 +0900368 dbs_info = &per_cpu(od_cpu_dbs_info, j);
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700369 dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
370 &dbs_info->prev_cpu_wall);
Venkatesh Pallipadi1ca3abd2009-01-23 09:25:02 -0500371 if (dbs_tuners_ins.ignore_nice)
372 dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
373
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700374 }
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800375 mutex_unlock(&dbs_mutex);
Dave Jones3d5ee9e2005-05-31 19:03:47 -0700376
377 return count;
378}
379
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200380static ssize_t store_powersave_bias(struct kobject *a, struct attribute *b,
381 const char *buf, size_t count)
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400382{
383 unsigned int input;
384 int ret;
385 ret = sscanf(buf, "%u", &input);
386
387 if (ret != 1)
388 return -EINVAL;
389
390 if (input > 1000)
391 input = 1000;
392
393 mutex_lock(&dbs_mutex);
394 dbs_tuners_ins.powersave_bias = input;
395 ondemand_powersave_bias_init();
396 mutex_unlock(&dbs_mutex);
397
398 return count;
399}
400
Borislav Petkov6dad2a22010-03-31 21:56:46 +0200401define_one_global_rw(sampling_rate);
Linus Torvalds07d77752010-05-18 08:49:13 -0700402define_one_global_rw(io_is_busy);
Borislav Petkov6dad2a22010-03-31 21:56:46 +0200403define_one_global_rw(up_threshold);
404define_one_global_rw(ignore_nice_load);
405define_one_global_rw(powersave_bias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Dave Jones2b03f892009-01-18 01:43:44 -0500407static struct attribute *dbs_attributes[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 &sampling_rate_max.attr,
409 &sampling_rate_min.attr,
410 &sampling_rate.attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 &up_threshold.attr,
Alexander Clouter001893c2005-12-01 01:09:25 -0800412 &ignore_nice_load.attr,
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400413 &powersave_bias.attr,
Arjan van de Ven19379b12010-05-09 08:26:51 -0700414 &io_is_busy.attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 NULL
416};
417
418static struct attribute_group dbs_attr_group = {
419 .attrs = dbs_attributes,
420 .name = "ondemand",
421};
422
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200423/*** delete after deprecation time ***/
424
425#define write_one_old(file_name) \
426static ssize_t store_##file_name##_old \
427(struct cpufreq_policy *unused, const char *buf, size_t count) \
428{ \
429 printk_once(KERN_INFO "CPUFREQ: Per core ondemand sysfs " \
430 "interface is deprecated - " #file_name "\n"); \
431 return store_##file_name(NULL, NULL, buf, count); \
432}
433write_one_old(sampling_rate);
434write_one_old(up_threshold);
435write_one_old(ignore_nice_load);
436write_one_old(powersave_bias);
437
Borislav Petkov6dad2a22010-03-31 21:56:46 +0200438cpufreq_freq_attr_rw_old(sampling_rate);
439cpufreq_freq_attr_rw_old(up_threshold);
440cpufreq_freq_attr_rw_old(ignore_nice_load);
441cpufreq_freq_attr_rw_old(powersave_bias);
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200442
443static struct attribute *dbs_attributes_old[] = {
444 &sampling_rate_max_old.attr,
445 &sampling_rate_min_old.attr,
446 &sampling_rate_old.attr,
447 &up_threshold_old.attr,
448 &ignore_nice_load_old.attr,
449 &powersave_bias_old.attr,
450 NULL
451};
452
453static struct attribute_group dbs_attr_group_old = {
454 .attrs = dbs_attributes_old,
455 .name = "ondemand",
456};
457
458/*** delete after deprecation time ***/
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460/************************** sysfs end ************************/
461
Venkatesh Pallipadi2f8a8352006-06-28 13:51:19 -0700462static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700464 unsigned int max_load_freq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 struct cpufreq_policy *policy;
467 unsigned int j;
468
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400469 this_dbs_info->freq_lo = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 policy = this_dbs_info->cur_policy;
Venki Pallipadiea487612007-06-20 14:26:24 -0700471
Dave Jones32ee8c32006-02-28 00:43:23 -0500472 /*
Dave Jonesc29f1402005-05-31 19:03:50 -0700473 * Every sampling_rate, we check, if current idle time is less
474 * than 20% (default), then we try to increase frequency
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700475 * Every sampling_rate, we look for a the lowest
Dave Jonesc29f1402005-05-31 19:03:50 -0700476 * frequency which can sustain the load while keeping idle time over
477 * 30%. If such a frequency exist, we try to decrease to this frequency.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 *
Dave Jones32ee8c32006-02-28 00:43:23 -0500479 * Any frequency increase takes it to the maximum frequency.
480 * Frequency reduction happens at minimum steps of
481 * 5% (default) of current frequency
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 */
483
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700484 /* Get Absolute Load - in terms of freq */
485 max_load_freq = 0;
486
Rusty Russell835481d2009-01-04 05:18:06 -0800487 for_each_cpu(j, policy->cpus) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 struct cpu_dbs_info_s *j_dbs_info;
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700489 cputime64_t cur_wall_time, cur_idle_time, cur_iowait_time;
490 unsigned int idle_time, wall_time, iowait_time;
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700491 unsigned int load, load_freq;
492 int freq_avg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Tejun Heo245b2e72009-06-24 15:13:48 +0900494 j_dbs_info = &per_cpu(od_cpu_dbs_info, j);
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700495
496 cur_idle_time = get_cpu_idle_time(j, &cur_wall_time);
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700497 cur_iowait_time = get_cpu_iowait_time(j, &cur_wall_time);
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700498
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700499 wall_time = (unsigned int) cputime64_sub(cur_wall_time,
500 j_dbs_info->prev_cpu_wall);
501 j_dbs_info->prev_cpu_wall = cur_wall_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700503 idle_time = (unsigned int) cputime64_sub(cur_idle_time,
504 j_dbs_info->prev_cpu_idle);
505 j_dbs_info->prev_cpu_idle = cur_idle_time;
506
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700507 iowait_time = (unsigned int) cputime64_sub(cur_iowait_time,
508 j_dbs_info->prev_cpu_iowait);
509 j_dbs_info->prev_cpu_iowait = cur_iowait_time;
510
Venkatesh Pallipadi1ca3abd2009-01-23 09:25:02 -0500511 if (dbs_tuners_ins.ignore_nice) {
512 cputime64_t cur_nice;
513 unsigned long cur_nice_jiffies;
514
515 cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice,
516 j_dbs_info->prev_cpu_nice);
517 /*
518 * Assumption: nice time between sampling periods will
519 * be less than 2^32 jiffies for 32 bit sys
520 */
521 cur_nice_jiffies = (unsigned long)
522 cputime64_to_jiffies64(cur_nice);
523
524 j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice;
525 idle_time += jiffies_to_usecs(cur_nice_jiffies);
526 }
527
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700528 /*
529 * For the purpose of ondemand, waiting for disk IO is an
530 * indication that you're performance critical, and not that
531 * the system is actually idle. So subtract the iowait time
532 * from the cpu idle time.
533 */
534
Arjan van de Ven19379b12010-05-09 08:26:51 -0700535 if (dbs_tuners_ins.io_is_busy && idle_time >= iowait_time)
Arjan van de Ven6b8fcd92010-05-09 08:26:06 -0700536 idle_time -= iowait_time;
537
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700538 if (unlikely(!wall_time || wall_time < idle_time))
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700539 continue;
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700540
541 load = 100 * (wall_time - idle_time) / wall_time;
542
543 freq_avg = __cpufreq_driver_getavg(policy, j);
544 if (freq_avg <= 0)
545 freq_avg = policy->cur;
546
547 load_freq = load * freq_avg;
548 if (load_freq > max_load_freq)
549 max_load_freq = load_freq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 }
551
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700552 /* Check for frequency increase */
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700553 if (max_load_freq > dbs_tuners_ins.up_threshold * policy->cur) {
Dave Jonesc11420a2005-05-31 19:03:48 -0700554 /* if we are already at full speed then break out early */
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400555 if (!dbs_tuners_ins.powersave_bias) {
556 if (policy->cur == policy->max)
557 return;
Dave Jones32ee8c32006-02-28 00:43:23 -0500558
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400559 __cpufreq_driver_target(policy, policy->max,
560 CPUFREQ_RELATION_H);
561 } else {
562 int freq = powersave_bias_target(policy, policy->max,
563 CPUFREQ_RELATION_H);
564 __cpufreq_driver_target(policy, freq,
565 CPUFREQ_RELATION_L);
566 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 return;
568 }
569
570 /* Check for frequency decrease */
Dave Jonesc29f1402005-05-31 19:03:50 -0700571 /* if we cannot reduce the frequency anymore, break out early */
572 if (policy->cur == policy->min)
573 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574
Dave Jonesc29f1402005-05-31 19:03:50 -0700575 /*
576 * The optimal frequency is the frequency that is the lowest that
577 * can support the current CPU usage without triggering the up
578 * policy. To be safe, we focus 10 points under the threshold.
579 */
venkatesh.pallipadi@intel.come9d95bf2008-08-04 11:59:10 -0700580 if (max_load_freq <
581 (dbs_tuners_ins.up_threshold - dbs_tuners_ins.down_differential) *
582 policy->cur) {
venkatesh.pallipadi@intel.comc43aa3b2008-08-04 11:59:08 -0700583 unsigned int freq_next;
venkatesh.pallipadi@intel.come9d95bf2008-08-04 11:59:10 -0700584 freq_next = max_load_freq /
585 (dbs_tuners_ins.up_threshold -
586 dbs_tuners_ins.down_differential);
Venkatesh Pallipadidfde5d62006-10-03 12:38:45 -0700587
Nagananda.Chumbalkar@hp.com1dbf5882009-12-21 23:40:52 +0100588 if (freq_next < policy->min)
589 freq_next = policy->min;
590
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400591 if (!dbs_tuners_ins.powersave_bias) {
592 __cpufreq_driver_target(policy, freq_next,
593 CPUFREQ_RELATION_L);
594 } else {
595 int freq = powersave_bias_target(policy, freq_next,
596 CPUFREQ_RELATION_L);
597 __cpufreq_driver_target(policy, freq,
598 CPUFREQ_RELATION_L);
599 }
Venkatesh Pallipadiccb2fe22006-06-28 13:49:52 -0700600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
David Howellsc4028952006-11-22 14:57:56 +0000603static void do_dbs_timer(struct work_struct *work)
Dave Jones32ee8c32006-02-28 00:43:23 -0500604{
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -0800605 struct cpu_dbs_info_s *dbs_info =
606 container_of(work, struct cpu_dbs_info_s, work.work);
607 unsigned int cpu = dbs_info->cpu;
608 int sample_type = dbs_info->sample_type;
609
Alexey Starikovskiy1ce28d62006-07-31 22:25:20 +0400610 /* We want all CPUs to do sampling nearly on same jiffy */
611 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
David Howellsc4028952006-11-22 14:57:56 +0000612
Alexey Starikovskiy1ce28d62006-07-31 22:25:20 +0400613 delay -= jiffies % delay;
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700614 mutex_lock(&dbs_info->timer_mutex);
Venkatesh Pallipadi56463b72007-02-05 16:12:45 -0800615
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400616 /* Common NORMAL_SAMPLE setup */
David Howellsc4028952006-11-22 14:57:56 +0000617 dbs_info->sample_type = DBS_NORMAL_SAMPLE;
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400618 if (!dbs_tuners_ins.powersave_bias ||
David Howellsc4028952006-11-22 14:57:56 +0000619 sample_type == DBS_NORMAL_SAMPLE) {
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400620 dbs_check_cpu(dbs_info);
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400621 if (dbs_info->freq_lo) {
622 /* Setup timer for SUB_SAMPLE */
David Howellsc4028952006-11-22 14:57:56 +0000623 dbs_info->sample_type = DBS_SUB_SAMPLE;
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400624 delay = dbs_info->freq_hi_jiffies;
625 }
626 } else {
627 __cpufreq_driver_target(dbs_info->cur_policy,
Dave Jones2b03f892009-01-18 01:43:44 -0500628 dbs_info->freq_lo, CPUFREQ_RELATION_H);
Alexey Starikovskiy05ca0352006-07-31 22:28:12 +0400629 }
Alexey Starikovskiy1ce28d62006-07-31 22:25:20 +0400630 queue_delayed_work_on(cpu, kondemand_wq, &dbs_info->work, delay);
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700631 mutex_unlock(&dbs_info->timer_mutex);
Dave Jones32ee8c32006-02-28 00:43:23 -0500632}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -0800634static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635{
Alexey Starikovskiy1ce28d62006-07-31 22:25:20 +0400636 /* We want all CPUs to do sampling nearly on same jiffy */
637 int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate);
638 delay -= jiffies % delay;
Venkatesh Pallipadi2f8a8352006-06-28 13:51:19 -0700639
David Howellsc4028952006-11-22 14:57:56 +0000640 dbs_info->sample_type = DBS_NORMAL_SAMPLE;
Venki Pallipadi28287032007-05-08 00:27:47 -0700641 INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer);
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -0800642 queue_delayed_work_on(dbs_info->cpu, kondemand_wq, &dbs_info->work,
Dave Jones2b03f892009-01-18 01:43:44 -0500643 delay);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
645
Linus Torvalds2cd7cbd2006-07-23 12:05:00 -0700646static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Mathieu Desnoyersb14893a2009-05-17 10:30:45 -0400648 cancel_delayed_work_sync(&dbs_info->work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Arjan van de Ven19379b12010-05-09 08:26:51 -0700651/*
652 * Not all CPUs want IO time to be accounted as busy; this dependson how
653 * efficient idling at a higher frequency/voltage is.
654 * Pavel Machek says this is not so for various generations of AMD and old
655 * Intel systems.
656 * Mike Chan (androidlcom) calis this is also not true for ARM.
657 * Because of this, whitelist specific known (series) of CPUs by default, and
658 * leave all others up to the user.
659 */
660static int should_io_be_busy(void)
661{
662#if defined(CONFIG_X86)
663 /*
664 * For Intel, Core 2 (model 15) andl later have an efficient idle.
665 */
666 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
667 boot_cpu_data.x86 == 6 &&
668 boot_cpu_data.x86_model >= 15)
669 return 1;
670#endif
671 return 0;
672}
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674static int cpufreq_governor_dbs(struct cpufreq_policy *policy,
675 unsigned int event)
676{
677 unsigned int cpu = policy->cpu;
678 struct cpu_dbs_info_s *this_dbs_info;
679 unsigned int j;
Jeff Garzik914f7c32006-10-20 14:31:00 -0700680 int rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Tejun Heo245b2e72009-06-24 15:13:48 +0900682 this_dbs_info = &per_cpu(od_cpu_dbs_info, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
684 switch (event) {
685 case CPUFREQ_GOV_START:
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700686 if ((!cpu_online(cpu)) || (!policy->cur))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return -EINVAL;
688
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800689 mutex_lock(&dbs_mutex);
Jeff Garzik914f7c32006-10-20 14:31:00 -0700690
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200691 rc = sysfs_create_group(&policy->kobj, &dbs_attr_group_old);
Jeff Garzik914f7c32006-10-20 14:31:00 -0700692 if (rc) {
Jeff Garzik914f7c32006-10-20 14:31:00 -0700693 mutex_unlock(&dbs_mutex);
694 return rc;
695 }
696
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700697 dbs_enable++;
Rusty Russell835481d2009-01-04 05:18:06 -0800698 for_each_cpu(j, policy->cpus) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 struct cpu_dbs_info_s *j_dbs_info;
Tejun Heo245b2e72009-06-24 15:13:48 +0900700 j_dbs_info = &per_cpu(od_cpu_dbs_info, j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 j_dbs_info->cur_policy = policy;
Dave Jones32ee8c32006-02-28 00:43:23 -0500702
venkatesh.pallipadi@intel.com34305022008-08-04 11:59:09 -0700703 j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j,
704 &j_dbs_info->prev_cpu_wall);
Venkatesh Pallipadi1ca3abd2009-01-23 09:25:02 -0500705 if (dbs_tuners_ins.ignore_nice) {
706 j_dbs_info->prev_cpu_nice =
707 kstat_cpu(j).cpustat.nice;
708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Venkatesh Pallipadi529af7a2007-02-05 16:12:44 -0800710 this_dbs_info->cpu = cpu;
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700711 ondemand_powersave_bias_init_cpu(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 /*
713 * Start the timerschedule work, when this governor
714 * is used for first time
715 */
716 if (dbs_enable == 1) {
717 unsigned int latency;
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200718
719 rc = sysfs_create_group(cpufreq_global_kobject,
720 &dbs_attr_group);
721 if (rc) {
722 mutex_unlock(&dbs_mutex);
723 return rc;
724 }
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 /* policy latency is in nS. Convert it to uS first */
Dave Jonesdf8b59b2005-09-20 12:39:35 -0700727 latency = policy->cpuinfo.transition_latency / 1000;
728 if (latency == 0)
729 latency = 1;
Thomas Renningercef96152009-04-22 13:48:29 +0200730 /* Bring kernel and HW constraints together */
731 min_sampling_rate = max(min_sampling_rate,
732 MIN_LATENCY_MULTIPLIER * latency);
733 dbs_tuners_ins.sampling_rate =
734 max(min_sampling_rate,
735 latency * LATENCY_MULTIPLIER);
Arjan van de Ven19379b12010-05-09 08:26:51 -0700736 dbs_tuners_ins.io_is_busy = should_io_be_busy();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800738 mutex_unlock(&dbs_mutex);
venkatesh.pallipadi@intel.com7d26e2d2009-07-02 17:08:30 -0700739
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200740 mutex_init(&this_dbs_info->timer_mutex);
venkatesh.pallipadi@intel.com7d26e2d2009-07-02 17:08:30 -0700741 dbs_timer_init(this_dbs_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 break;
743
744 case CPUFREQ_GOV_STOP:
Linus Torvalds2cd7cbd2006-07-23 12:05:00 -0700745 dbs_timer_exit(this_dbs_info);
venkatesh.pallipadi@intel.com7d26e2d2009-07-02 17:08:30 -0700746
747 mutex_lock(&dbs_mutex);
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200748 sysfs_remove_group(&policy->kobj, &dbs_attr_group_old);
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700749 mutex_destroy(&this_dbs_info->timer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 dbs_enable--;
akpm@osdl.org3fc54d32006-01-13 15:54:22 -0800751 mutex_unlock(&dbs_mutex);
Thomas Renninger0e625ac2009-07-24 15:25:06 +0200752 if (!dbs_enable)
753 sysfs_remove_group(cpufreq_global_kobject,
754 &dbs_attr_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
756 break;
757
758 case CPUFREQ_GOV_LIMITS:
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700759 mutex_lock(&this_dbs_info->timer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (policy->max < this_dbs_info->cur_policy->cur)
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700761 __cpufreq_driver_target(this_dbs_info->cur_policy,
Dave Jones2b03f892009-01-18 01:43:44 -0500762 policy->max, CPUFREQ_RELATION_H);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 else if (policy->min > this_dbs_info->cur_policy->cur)
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700764 __cpufreq_driver_target(this_dbs_info->cur_policy,
Dave Jones2b03f892009-01-18 01:43:44 -0500765 policy->min, CPUFREQ_RELATION_L);
venkatesh.pallipadi@intel.com5a75c822009-07-02 17:08:32 -0700766 mutex_unlock(&this_dbs_info->timer_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 break;
768 }
769 return 0;
770}
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772static int __init cpufreq_gov_dbs_init(void)
773{
Akinobu Mita888a7942008-07-14 12:00:45 +0900774 int err;
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700775 cputime64_t wall;
Andrea Righi4f6e6b92008-09-18 10:43:40 +0000776 u64 idle_time;
777 int cpu = get_cpu();
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700778
Andrea Righi4f6e6b92008-09-18 10:43:40 +0000779 idle_time = get_cpu_idle_time_us(cpu, &wall);
780 put_cpu();
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700781 if (idle_time != -1ULL) {
782 /* Idle micro accounting is supported. Use finer thresholds */
783 dbs_tuners_ins.up_threshold = MICRO_FREQUENCY_UP_THRESHOLD;
784 dbs_tuners_ins.down_differential =
785 MICRO_FREQUENCY_DOWN_DIFFERENTIAL;
Thomas Renningercef96152009-04-22 13:48:29 +0200786 /*
787 * In no_hz/micro accounting case we set the minimum frequency
788 * not depending on HZ, but fixed (very low). The deferred
789 * timer might skip some samples if idle/sleeping as needed.
790 */
791 min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE;
792 } else {
793 /* For correct statistics, we need 10 ticks for each measure */
794 min_sampling_rate =
795 MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10);
venkatesh.pallipadi@intel.com80800912008-08-04 11:59:12 -0700796 }
Akinobu Mita888a7942008-07-14 12:00:45 +0900797
Venkatesh Pallipadi56463b72007-02-05 16:12:45 -0800798 kondemand_wq = create_workqueue("kondemand");
799 if (!kondemand_wq) {
800 printk(KERN_ERR "Creation of kondemand failed\n");
801 return -EFAULT;
802 }
Akinobu Mita888a7942008-07-14 12:00:45 +0900803 err = cpufreq_register_governor(&cpufreq_gov_ondemand);
804 if (err)
805 destroy_workqueue(kondemand_wq);
806
807 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
810static void __exit cpufreq_gov_dbs_exit(void)
811{
Thomas Renninger1c256242007-10-02 13:28:12 -0700812 cpufreq_unregister_governor(&cpufreq_gov_ondemand);
Venkatesh Pallipadi56463b72007-02-05 16:12:45 -0800813 destroy_workqueue(kondemand_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814}
815
816
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700817MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>");
818MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>");
819MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for "
Dave Jones2b03f892009-01-18 01:43:44 -0500820 "Low Latency Frequency Transition capable processors");
Venkatesh Pallipadiffac80e2006-06-28 13:52:18 -0700821MODULE_LICENSE("GPL");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Johannes Weiner69157192008-01-17 15:21:08 -0800823#ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND
824fs_initcall(cpufreq_gov_dbs_init);
825#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826module_init(cpufreq_gov_dbs_init);
Johannes Weiner69157192008-01-17 15:21:08 -0800827#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828module_exit(cpufreq_gov_dbs_exit);