Commit | Line | Data |
---|---|---|
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 |
47 | static 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 | 55 | static void do_dbs_timer(struct work_struct *work); |
b9170836 DJ |
56 | |
57 | struct 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; | |
b9170836 DJ |
67 | }; |
68 | static DEFINE_PER_CPU(struct cpu_dbs_info_s, cpu_dbs_info); | |
69 | ||
70 | static unsigned int dbs_enable; /* number of CPUs using this policy */ | |
71 | ||
4ec223d0 VP |
72 | /* |
73 | * DEADLOCK ALERT! There is a ordering requirement between cpu_hotplug | |
74 | * lock and dbs_mutex. cpu_hotplug lock should always be held before | |
75 | * dbs_mutex. If any function that can potentially take cpu_hotplug lock | |
76 | * (like __cpufreq_driver_target()) is being called with dbs_mutex taken, then | |
77 | * cpu_hotplug lock should be taken before that. Note that cpu_hotplug lock | |
78 | * is recursive for the same process. -Venki | |
b253d2b2 MD |
79 | * DEADLOCK ALERT! (2) : do_dbs_timer() must not take the dbs_mutex, because it |
80 | * would deadlock with cancel_delayed_work_sync(), which is needed for proper | |
81 | * raceless workqueue teardown. | |
4ec223d0 | 82 | */ |
9acef487 | 83 | static DEFINE_MUTEX(dbs_mutex); |
b9170836 | 84 | |
8e677ce8 AC |
85 | static struct workqueue_struct *kconservative_wq; |
86 | ||
87 | static struct dbs_tuners { | |
18a7247d DJ |
88 | unsigned int sampling_rate; |
89 | unsigned int sampling_down_factor; | |
90 | unsigned int up_threshold; | |
91 | unsigned int down_threshold; | |
92 | unsigned int ignore_nice; | |
93 | unsigned int freq_step; | |
8e677ce8 | 94 | } dbs_tuners_ins = { |
18a7247d DJ |
95 | .up_threshold = DEF_FREQUENCY_UP_THRESHOLD, |
96 | .down_threshold = DEF_FREQUENCY_DOWN_THRESHOLD, | |
97 | .sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR, | |
98 | .ignore_nice = 0, | |
99 | .freq_step = 5, | |
b9170836 DJ |
100 | }; |
101 | ||
8e677ce8 AC |
102 | static inline cputime64_t get_cpu_idle_time_jiffy(unsigned int cpu, |
103 | cputime64_t *wall) | |
dac1c1a5 | 104 | { |
8e677ce8 AC |
105 | cputime64_t idle_time; |
106 | cputime64_t cur_wall_time; | |
107 | cputime64_t busy_time; | |
108 | ||
109 | cur_wall_time = jiffies64_to_cputime64(get_jiffies_64()); | |
110 | busy_time = cputime64_add(kstat_cpu(cpu).cpustat.user, | |
111 | kstat_cpu(cpu).cpustat.system); | |
e08f5f5b | 112 | |
8e677ce8 AC |
113 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.irq); |
114 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.softirq); | |
115 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.steal); | |
116 | busy_time = cputime64_add(busy_time, kstat_cpu(cpu).cpustat.nice); | |
e08f5f5b | 117 | |
8e677ce8 AC |
118 | idle_time = cputime64_sub(cur_wall_time, busy_time); |
119 | if (wall) | |
120 | *wall = cur_wall_time; | |
e08f5f5b | 121 | |
8e677ce8 AC |
122 | return idle_time; |
123 | } | |
124 | ||
125 | static inline cputime64_t get_cpu_idle_time(unsigned int cpu, cputime64_t *wall) | |
126 | { | |
127 | u64 idle_time = get_cpu_idle_time_us(cpu, wall); | |
128 | ||
129 | if (idle_time == -1ULL) | |
130 | return get_cpu_idle_time_jiffy(cpu, wall); | |
131 | ||
132 | return idle_time; | |
dac1c1a5 DJ |
133 | } |
134 | ||
a8d7c3bc EO |
135 | /* keep track of frequency transitions */ |
136 | static int | |
137 | dbs_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
138 | void *data) | |
139 | { | |
140 | struct cpufreq_freqs *freq = data; | |
141 | struct cpu_dbs_info_s *this_dbs_info = &per_cpu(cpu_dbs_info, | |
142 | freq->cpu); | |
143 | ||
f407a08b AC |
144 | struct cpufreq_policy *policy; |
145 | ||
a8d7c3bc EO |
146 | if (!this_dbs_info->enable) |
147 | return 0; | |
148 | ||
f407a08b AC |
149 | policy = this_dbs_info->cur_policy; |
150 | ||
151 | /* | |
152 | * we only care if our internally tracked freq moves outside | |
153 | * the 'valid' ranges of freqency available to us otherwise | |
154 | * we do not change it | |
155 | */ | |
156 | if (this_dbs_info->requested_freq > policy->max | |
157 | || this_dbs_info->requested_freq < policy->min) | |
158 | this_dbs_info->requested_freq = freq->new; | |
a8d7c3bc EO |
159 | |
160 | return 0; | |
161 | } | |
162 | ||
163 | static struct notifier_block dbs_cpufreq_notifier_block = { | |
164 | .notifier_call = dbs_cpufreq_notifier | |
165 | }; | |
166 | ||
b9170836 DJ |
167 | /************************** sysfs interface ************************/ |
168 | static ssize_t show_sampling_rate_max(struct cpufreq_policy *policy, char *buf) | |
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 | ||
175 | static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf) | |
176 | { | |
cef9615a | 177 | return sprintf(buf, "%u\n", min_sampling_rate); |
b9170836 DJ |
178 | } |
179 | ||
8e677ce8 AC |
180 | #define define_one_ro(_name) \ |
181 | static struct freq_attr _name = \ | |
b9170836 DJ |
182 | __ATTR(_name, 0444, show_##_name, NULL) |
183 | ||
184 | define_one_ro(sampling_rate_max); | |
185 | define_one_ro(sampling_rate_min); | |
186 | ||
187 | /* cpufreq_conservative Governor Tunables */ | |
188 | #define show_one(file_name, object) \ | |
189 | static ssize_t show_##file_name \ | |
190 | (struct cpufreq_policy *unused, char *buf) \ | |
191 | { \ | |
192 | return sprintf(buf, "%u\n", dbs_tuners_ins.object); \ | |
193 | } | |
194 | show_one(sampling_rate, sampling_rate); | |
195 | show_one(sampling_down_factor, sampling_down_factor); | |
196 | show_one(up_threshold, up_threshold); | |
197 | show_one(down_threshold, down_threshold); | |
001893cd | 198 | show_one(ignore_nice_load, ignore_nice); |
b9170836 DJ |
199 | show_one(freq_step, freq_step); |
200 | ||
18a7247d | 201 | static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused, |
b9170836 DJ |
202 | const char *buf, size_t count) |
203 | { | |
204 | unsigned int input; | |
205 | int ret; | |
9acef487 | 206 | ret = sscanf(buf, "%u", &input); |
8e677ce8 | 207 | |
2c906b31 | 208 | if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1) |
b9170836 DJ |
209 | return -EINVAL; |
210 | ||
3fc54d37 | 211 | mutex_lock(&dbs_mutex); |
b9170836 | 212 | dbs_tuners_ins.sampling_down_factor = input; |
3fc54d37 | 213 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
214 | |
215 | return count; | |
216 | } | |
217 | ||
18a7247d | 218 | static ssize_t store_sampling_rate(struct cpufreq_policy *unused, |
b9170836 DJ |
219 | const char *buf, size_t count) |
220 | { | |
221 | unsigned int input; | |
222 | int ret; | |
9acef487 | 223 | ret = sscanf(buf, "%u", &input); |
b9170836 | 224 | |
8e677ce8 | 225 | if (ret != 1) |
b9170836 | 226 | return -EINVAL; |
8e677ce8 AC |
227 | |
228 | mutex_lock(&dbs_mutex); | |
cef9615a | 229 | dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate); |
3fc54d37 | 230 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
231 | |
232 | return count; | |
233 | } | |
234 | ||
18a7247d | 235 | static ssize_t store_up_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
236 | const char *buf, size_t count) |
237 | { | |
238 | unsigned int input; | |
239 | int ret; | |
9acef487 | 240 | ret = sscanf(buf, "%u", &input); |
b9170836 | 241 | |
3fc54d37 | 242 | mutex_lock(&dbs_mutex); |
9acef487 | 243 | if (ret != 1 || input > 100 || |
8e677ce8 | 244 | input <= dbs_tuners_ins.down_threshold) { |
3fc54d37 | 245 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
246 | return -EINVAL; |
247 | } | |
248 | ||
249 | dbs_tuners_ins.up_threshold = input; | |
3fc54d37 | 250 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
251 | |
252 | return count; | |
253 | } | |
254 | ||
18a7247d | 255 | static ssize_t store_down_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
256 | const char *buf, size_t count) |
257 | { | |
258 | unsigned int input; | |
259 | int ret; | |
9acef487 | 260 | ret = sscanf(buf, "%u", &input); |
b9170836 | 261 | |
3fc54d37 | 262 | mutex_lock(&dbs_mutex); |
8e677ce8 AC |
263 | /* cannot be lower than 11 otherwise freq will not fall */ |
264 | if (ret != 1 || input < 11 || input > 100 || | |
265 | input >= dbs_tuners_ins.up_threshold) { | |
3fc54d37 | 266 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
267 | return -EINVAL; |
268 | } | |
269 | ||
270 | dbs_tuners_ins.down_threshold = input; | |
3fc54d37 | 271 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
272 | |
273 | return count; | |
274 | } | |
275 | ||
001893cd | 276 | static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy, |
b9170836 DJ |
277 | const char *buf, size_t count) |
278 | { | |
279 | unsigned int input; | |
280 | int ret; | |
281 | ||
282 | unsigned int j; | |
18a7247d DJ |
283 | |
284 | ret = sscanf(buf, "%u", &input); | |
285 | if (ret != 1) | |
b9170836 DJ |
286 | return -EINVAL; |
287 | ||
18a7247d | 288 | if (input > 1) |
b9170836 | 289 | input = 1; |
18a7247d | 290 | |
3fc54d37 | 291 | mutex_lock(&dbs_mutex); |
18a7247d | 292 | if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ |
3fc54d37 | 293 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
294 | return count; |
295 | } | |
296 | dbs_tuners_ins.ignore_nice = input; | |
297 | ||
8e677ce8 | 298 | /* we need to re-evaluate prev_cpu_idle */ |
dac1c1a5 | 299 | for_each_online_cpu(j) { |
8e677ce8 AC |
300 | struct cpu_dbs_info_s *dbs_info; |
301 | dbs_info = &per_cpu(cpu_dbs_info, j); | |
302 | dbs_info->prev_cpu_idle = get_cpu_idle_time(j, | |
303 | &dbs_info->prev_cpu_wall); | |
304 | if (dbs_tuners_ins.ignore_nice) | |
305 | dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
b9170836 | 306 | } |
3fc54d37 | 307 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
308 | |
309 | return count; | |
310 | } | |
311 | ||
312 | static ssize_t store_freq_step(struct cpufreq_policy *policy, | |
313 | const char *buf, size_t count) | |
314 | { | |
315 | unsigned int input; | |
316 | int ret; | |
18a7247d | 317 | ret = sscanf(buf, "%u", &input); |
b9170836 | 318 | |
18a7247d | 319 | if (ret != 1) |
b9170836 DJ |
320 | return -EINVAL; |
321 | ||
18a7247d | 322 | if (input > 100) |
b9170836 | 323 | input = 100; |
18a7247d | 324 | |
b9170836 DJ |
325 | /* no need to test here if freq_step is zero as the user might actually |
326 | * want this, they would be crazy though :) */ | |
3fc54d37 | 327 | mutex_lock(&dbs_mutex); |
b9170836 | 328 | dbs_tuners_ins.freq_step = input; |
3fc54d37 | 329 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
330 | |
331 | return count; | |
332 | } | |
333 | ||
334 | #define define_one_rw(_name) \ | |
335 | static struct freq_attr _name = \ | |
336 | __ATTR(_name, 0644, show_##_name, store_##_name) | |
337 | ||
338 | define_one_rw(sampling_rate); | |
339 | define_one_rw(sampling_down_factor); | |
340 | define_one_rw(up_threshold); | |
341 | define_one_rw(down_threshold); | |
001893cd | 342 | define_one_rw(ignore_nice_load); |
b9170836 DJ |
343 | define_one_rw(freq_step); |
344 | ||
9acef487 | 345 | static struct attribute *dbs_attributes[] = { |
b9170836 DJ |
346 | &sampling_rate_max.attr, |
347 | &sampling_rate_min.attr, | |
348 | &sampling_rate.attr, | |
349 | &sampling_down_factor.attr, | |
350 | &up_threshold.attr, | |
351 | &down_threshold.attr, | |
001893cd | 352 | &ignore_nice_load.attr, |
b9170836 DJ |
353 | &freq_step.attr, |
354 | NULL | |
355 | }; | |
356 | ||
357 | static struct attribute_group dbs_attr_group = { | |
358 | .attrs = dbs_attributes, | |
359 | .name = "conservative", | |
360 | }; | |
361 | ||
362 | /************************** sysfs end ************************/ | |
363 | ||
8e677ce8 | 364 | static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info) |
b9170836 | 365 | { |
8e677ce8 | 366 | unsigned int load = 0; |
f068c04b | 367 | unsigned int freq_target; |
b9170836 | 368 | |
8e677ce8 AC |
369 | struct cpufreq_policy *policy; |
370 | unsigned int j; | |
b9170836 | 371 | |
08a28e2e AC |
372 | policy = this_dbs_info->cur_policy; |
373 | ||
18a7247d | 374 | /* |
8e677ce8 AC |
375 | * Every sampling_rate, we check, if current idle time is less |
376 | * than 20% (default), then we try to increase frequency | |
377 | * Every sampling_rate*sampling_down_factor, we check, if current | |
378 | * idle time is more than 80%, then we try to decrease frequency | |
b9170836 | 379 | * |
18a7247d DJ |
380 | * Any frequency increase takes it to the maximum frequency. |
381 | * Frequency reduction happens at minimum steps of | |
8e677ce8 | 382 | * 5% (default) of maximum frequency |
b9170836 DJ |
383 | */ |
384 | ||
8e677ce8 AC |
385 | /* Get Absolute Load */ |
386 | for_each_cpu(j, policy->cpus) { | |
387 | struct cpu_dbs_info_s *j_dbs_info; | |
388 | cputime64_t cur_wall_time, cur_idle_time; | |
389 | unsigned int idle_time, wall_time; | |
b9170836 | 390 | |
8e677ce8 AC |
391 | j_dbs_info = &per_cpu(cpu_dbs_info, j); |
392 | ||
393 | cur_idle_time = get_cpu_idle_time(j, &cur_wall_time); | |
394 | ||
395 | wall_time = (unsigned int) cputime64_sub(cur_wall_time, | |
396 | j_dbs_info->prev_cpu_wall); | |
397 | j_dbs_info->prev_cpu_wall = cur_wall_time; | |
08a28e2e | 398 | |
8e677ce8 AC |
399 | idle_time = (unsigned int) cputime64_sub(cur_idle_time, |
400 | j_dbs_info->prev_cpu_idle); | |
401 | j_dbs_info->prev_cpu_idle = cur_idle_time; | |
b9170836 | 402 | |
8e677ce8 AC |
403 | if (dbs_tuners_ins.ignore_nice) { |
404 | cputime64_t cur_nice; | |
405 | unsigned long cur_nice_jiffies; | |
406 | ||
407 | cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice, | |
408 | j_dbs_info->prev_cpu_nice); | |
409 | /* | |
410 | * Assumption: nice time between sampling periods will | |
411 | * be less than 2^32 jiffies for 32 bit sys | |
412 | */ | |
413 | cur_nice_jiffies = (unsigned long) | |
414 | cputime64_to_jiffies64(cur_nice); | |
415 | ||
416 | j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
417 | idle_time += jiffies_to_usecs(cur_nice_jiffies); | |
418 | } | |
419 | ||
420 | if (unlikely(!wall_time || wall_time < idle_time)) | |
421 | continue; | |
422 | ||
423 | load = 100 * (wall_time - idle_time) / wall_time; | |
424 | } | |
425 | ||
426 | /* | |
427 | * break out if we 'cannot' reduce the speed as the user might | |
428 | * want freq_step to be zero | |
429 | */ | |
430 | if (dbs_tuners_ins.freq_step == 0) | |
431 | return; | |
b9170836 | 432 | |
8e677ce8 AC |
433 | /* Check for frequency increase */ |
434 | if (load > dbs_tuners_ins.up_threshold) { | |
a159b827 | 435 | this_dbs_info->down_skip = 0; |
790d76fa | 436 | |
b9170836 | 437 | /* if we are already at full speed then break out early */ |
a159b827 | 438 | if (this_dbs_info->requested_freq == policy->max) |
b9170836 | 439 | return; |
18a7247d | 440 | |
f068c04b | 441 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 DJ |
442 | |
443 | /* max freq cannot be less than 100. But who knows.... */ | |
f068c04b DJ |
444 | if (unlikely(freq_target == 0)) |
445 | freq_target = 5; | |
18a7247d | 446 | |
f068c04b | 447 | this_dbs_info->requested_freq += freq_target; |
a159b827 AC |
448 | if (this_dbs_info->requested_freq > policy->max) |
449 | this_dbs_info->requested_freq = policy->max; | |
b9170836 | 450 | |
a159b827 | 451 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
b9170836 | 452 | CPUFREQ_RELATION_H); |
b9170836 DJ |
453 | return; |
454 | } | |
455 | ||
8e677ce8 AC |
456 | /* |
457 | * The optimal frequency is the frequency that is the lowest that | |
458 | * can support the current CPU usage without triggering the up | |
459 | * policy. To be safe, we focus 10 points under the threshold. | |
460 | */ | |
461 | if (load < (dbs_tuners_ins.down_threshold - 10)) { | |
f068c04b | 462 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 | 463 | |
f068c04b | 464 | this_dbs_info->requested_freq -= freq_target; |
a159b827 AC |
465 | if (this_dbs_info->requested_freq < policy->min) |
466 | this_dbs_info->requested_freq = policy->min; | |
b9170836 | 467 | |
8e677ce8 AC |
468 | /* |
469 | * if we cannot reduce the frequency anymore, break out early | |
470 | */ | |
471 | if (policy->cur == policy->min) | |
472 | return; | |
473 | ||
a159b827 | 474 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
2c906b31 | 475 | CPUFREQ_RELATION_H); |
b9170836 DJ |
476 | return; |
477 | } | |
478 | } | |
479 | ||
c4028958 | 480 | static void do_dbs_timer(struct work_struct *work) |
18a7247d | 481 | { |
8e677ce8 AC |
482 | struct cpu_dbs_info_s *dbs_info = |
483 | container_of(work, struct cpu_dbs_info_s, work.work); | |
484 | unsigned int cpu = dbs_info->cpu; | |
485 | ||
486 | /* We want all CPUs to do sampling nearly on same jiffy */ | |
487 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
488 | ||
489 | delay -= jiffies % delay; | |
490 | ||
491 | if (lock_policy_rwsem_write(cpu) < 0) | |
492 | return; | |
493 | ||
494 | if (!dbs_info->enable) { | |
495 | unlock_policy_rwsem_write(cpu); | |
496 | return; | |
497 | } | |
498 | ||
499 | dbs_check_cpu(dbs_info); | |
500 | ||
501 | queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay); | |
502 | unlock_policy_rwsem_write(cpu); | |
18a7247d | 503 | } |
b9170836 | 504 | |
8e677ce8 | 505 | static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 506 | { |
8e677ce8 AC |
507 | /* We want all CPUs to do sampling nearly on same jiffy */ |
508 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
509 | delay -= jiffies % delay; | |
510 | ||
511 | dbs_info->enable = 1; | |
512 | INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer); | |
513 | queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work, | |
514 | delay); | |
b9170836 DJ |
515 | } |
516 | ||
8e677ce8 | 517 | static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 518 | { |
8e677ce8 | 519 | dbs_info->enable = 0; |
b253d2b2 | 520 | cancel_delayed_work_sync(&dbs_info->work); |
b9170836 DJ |
521 | } |
522 | ||
523 | static int cpufreq_governor_dbs(struct cpufreq_policy *policy, | |
524 | unsigned int event) | |
525 | { | |
526 | unsigned int cpu = policy->cpu; | |
527 | struct cpu_dbs_info_s *this_dbs_info; | |
528 | unsigned int j; | |
914f7c31 | 529 | int rc; |
b9170836 DJ |
530 | |
531 | this_dbs_info = &per_cpu(cpu_dbs_info, cpu); | |
532 | ||
533 | switch (event) { | |
534 | case CPUFREQ_GOV_START: | |
18a7247d | 535 | if ((!cpu_online(cpu)) || (!policy->cur)) |
b9170836 DJ |
536 | return -EINVAL; |
537 | ||
b9170836 DJ |
538 | if (this_dbs_info->enable) /* Already enabled */ |
539 | break; | |
18a7247d | 540 | |
3fc54d37 | 541 | mutex_lock(&dbs_mutex); |
914f7c31 JG |
542 | |
543 | rc = sysfs_create_group(&policy->kobj, &dbs_attr_group); | |
544 | if (rc) { | |
545 | mutex_unlock(&dbs_mutex); | |
546 | return rc; | |
547 | } | |
548 | ||
835481d9 | 549 | for_each_cpu(j, policy->cpus) { |
b9170836 DJ |
550 | struct cpu_dbs_info_s *j_dbs_info; |
551 | j_dbs_info = &per_cpu(cpu_dbs_info, j); | |
552 | j_dbs_info->cur_policy = policy; | |
18a7247d | 553 | |
8e677ce8 AC |
554 | j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j, |
555 | &j_dbs_info->prev_cpu_wall); | |
556 | if (dbs_tuners_ins.ignore_nice) { | |
557 | j_dbs_info->prev_cpu_nice = | |
558 | kstat_cpu(j).cpustat.nice; | |
559 | } | |
b9170836 | 560 | } |
a159b827 AC |
561 | this_dbs_info->down_skip = 0; |
562 | this_dbs_info->requested_freq = policy->cur; | |
914f7c31 | 563 | |
b9170836 DJ |
564 | dbs_enable++; |
565 | /* | |
566 | * Start the timerschedule work, when this governor | |
567 | * is used for first time | |
568 | */ | |
569 | if (dbs_enable == 1) { | |
570 | unsigned int latency; | |
571 | /* policy latency is in nS. Convert it to uS first */ | |
2c906b31 AC |
572 | latency = policy->cpuinfo.transition_latency / 1000; |
573 | if (latency == 0) | |
574 | latency = 1; | |
b9170836 | 575 | |
cef9615a TR |
576 | /* |
577 | * conservative does not implement micro like ondemand | |
578 | * governor, thus we are bound to jiffes/HZ | |
579 | */ | |
580 | min_sampling_rate = | |
581 | MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10); | |
582 | /* Bring kernel and HW constraints together */ | |
583 | min_sampling_rate = max(min_sampling_rate, | |
584 | MIN_LATENCY_MULTIPLIER * latency); | |
585 | dbs_tuners_ins.sampling_rate = | |
586 | max(min_sampling_rate, | |
587 | latency * LATENCY_MULTIPLIER); | |
b9170836 | 588 | |
a8d7c3bc EO |
589 | cpufreq_register_notifier( |
590 | &dbs_cpufreq_notifier_block, | |
591 | CPUFREQ_TRANSITION_NOTIFIER); | |
b9170836 | 592 | } |
8e677ce8 | 593 | dbs_timer_init(this_dbs_info); |
18a7247d | 594 | |
3fc54d37 | 595 | mutex_unlock(&dbs_mutex); |
8e677ce8 | 596 | |
b9170836 DJ |
597 | break; |
598 | ||
599 | case CPUFREQ_GOV_STOP: | |
3fc54d37 | 600 | mutex_lock(&dbs_mutex); |
8e677ce8 | 601 | dbs_timer_exit(this_dbs_info); |
b9170836 DJ |
602 | sysfs_remove_group(&policy->kobj, &dbs_attr_group); |
603 | dbs_enable--; | |
8e677ce8 | 604 | |
b9170836 DJ |
605 | /* |
606 | * Stop the timerschedule work, when this governor | |
607 | * is used for first time | |
608 | */ | |
8e677ce8 | 609 | if (dbs_enable == 0) |
a8d7c3bc EO |
610 | cpufreq_unregister_notifier( |
611 | &dbs_cpufreq_notifier_block, | |
612 | CPUFREQ_TRANSITION_NOTIFIER); | |
a8d7c3bc | 613 | |
3fc54d37 | 614 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
615 | |
616 | break; | |
617 | ||
618 | case CPUFREQ_GOV_LIMITS: | |
3fc54d37 | 619 | mutex_lock(&dbs_mutex); |
b9170836 DJ |
620 | if (policy->max < this_dbs_info->cur_policy->cur) |
621 | __cpufreq_driver_target( | |
622 | this_dbs_info->cur_policy, | |
18a7247d | 623 | policy->max, CPUFREQ_RELATION_H); |
b9170836 DJ |
624 | else if (policy->min > this_dbs_info->cur_policy->cur) |
625 | __cpufreq_driver_target( | |
626 | this_dbs_info->cur_policy, | |
18a7247d | 627 | policy->min, CPUFREQ_RELATION_L); |
3fc54d37 | 628 | mutex_unlock(&dbs_mutex); |
8e677ce8 | 629 | |
b9170836 DJ |
630 | break; |
631 | } | |
632 | return 0; | |
633 | } | |
634 | ||
c4d14bc0 SW |
635 | #ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
636 | static | |
637 | #endif | |
1c256245 TR |
638 | struct cpufreq_governor cpufreq_gov_conservative = { |
639 | .name = "conservative", | |
640 | .governor = cpufreq_governor_dbs, | |
641 | .max_transition_latency = TRANSITION_LATENCY_LIMIT, | |
642 | .owner = THIS_MODULE, | |
b9170836 DJ |
643 | }; |
644 | ||
645 | static int __init cpufreq_gov_dbs_init(void) | |
646 | { | |
8e677ce8 AC |
647 | int err; |
648 | ||
649 | kconservative_wq = create_workqueue("kconservative"); | |
650 | if (!kconservative_wq) { | |
651 | printk(KERN_ERR "Creation of kconservative failed\n"); | |
652 | return -EFAULT; | |
653 | } | |
654 | ||
655 | err = cpufreq_register_governor(&cpufreq_gov_conservative); | |
656 | if (err) | |
657 | destroy_workqueue(kconservative_wq); | |
658 | ||
659 | return err; | |
b9170836 DJ |
660 | } |
661 | ||
662 | static void __exit cpufreq_gov_dbs_exit(void) | |
663 | { | |
1c256245 | 664 | cpufreq_unregister_governor(&cpufreq_gov_conservative); |
8e677ce8 | 665 | destroy_workqueue(kconservative_wq); |
b9170836 DJ |
666 | } |
667 | ||
668 | ||
11a80a9c | 669 | MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>"); |
9acef487 | 670 | MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for " |
b9170836 DJ |
671 | "Low Latency Frequency Transition capable processors " |
672 | "optimised for use in a battery environment"); | |
9acef487 | 673 | MODULE_LICENSE("GPL"); |
b9170836 | 674 | |
6915719b JW |
675 | #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
676 | fs_initcall(cpufreq_gov_dbs_init); | |
677 | #else | |
b9170836 | 678 | module_init(cpufreq_gov_dbs_init); |
6915719b | 679 | #endif |
b9170836 | 680 | module_exit(cpufreq_gov_dbs_exit); |