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 | { | |
9411b4ef TR |
170 | static int print_once; |
171 | ||
172 | if (!print_once) { | |
173 | printk(KERN_INFO "CPUFREQ: conservative sampling_rate_max " | |
174 | "sysfs file is deprecated - used by: %s\n", | |
175 | current->comm); | |
176 | print_once = 1; | |
177 | } | |
cef9615a | 178 | return sprintf(buf, "%u\n", -1U); |
b9170836 DJ |
179 | } |
180 | ||
181 | static ssize_t show_sampling_rate_min(struct cpufreq_policy *policy, char *buf) | |
182 | { | |
9411b4ef TR |
183 | static int print_once; |
184 | ||
185 | if (!print_once) { | |
186 | printk(KERN_INFO "CPUFREQ: conservative sampling_rate_max " | |
187 | "sysfs file is deprecated - used by: %s\n", current->comm); | |
188 | print_once = 1; | |
189 | } | |
cef9615a | 190 | return sprintf(buf, "%u\n", min_sampling_rate); |
b9170836 DJ |
191 | } |
192 | ||
8e677ce8 AC |
193 | #define define_one_ro(_name) \ |
194 | static struct freq_attr _name = \ | |
b9170836 DJ |
195 | __ATTR(_name, 0444, show_##_name, NULL) |
196 | ||
197 | define_one_ro(sampling_rate_max); | |
198 | define_one_ro(sampling_rate_min); | |
199 | ||
200 | /* cpufreq_conservative Governor Tunables */ | |
201 | #define show_one(file_name, object) \ | |
202 | static ssize_t show_##file_name \ | |
203 | (struct cpufreq_policy *unused, char *buf) \ | |
204 | { \ | |
205 | return sprintf(buf, "%u\n", dbs_tuners_ins.object); \ | |
206 | } | |
207 | show_one(sampling_rate, sampling_rate); | |
208 | show_one(sampling_down_factor, sampling_down_factor); | |
209 | show_one(up_threshold, up_threshold); | |
210 | show_one(down_threshold, down_threshold); | |
001893cd | 211 | show_one(ignore_nice_load, ignore_nice); |
b9170836 DJ |
212 | show_one(freq_step, freq_step); |
213 | ||
18a7247d | 214 | static ssize_t store_sampling_down_factor(struct cpufreq_policy *unused, |
b9170836 DJ |
215 | const char *buf, size_t count) |
216 | { | |
217 | unsigned int input; | |
218 | int ret; | |
9acef487 | 219 | ret = sscanf(buf, "%u", &input); |
8e677ce8 | 220 | |
2c906b31 | 221 | if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1) |
b9170836 DJ |
222 | return -EINVAL; |
223 | ||
3fc54d37 | 224 | mutex_lock(&dbs_mutex); |
b9170836 | 225 | dbs_tuners_ins.sampling_down_factor = input; |
3fc54d37 | 226 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
227 | |
228 | return count; | |
229 | } | |
230 | ||
18a7247d | 231 | static ssize_t store_sampling_rate(struct cpufreq_policy *unused, |
b9170836 DJ |
232 | const char *buf, size_t count) |
233 | { | |
234 | unsigned int input; | |
235 | int ret; | |
9acef487 | 236 | ret = sscanf(buf, "%u", &input); |
b9170836 | 237 | |
8e677ce8 | 238 | if (ret != 1) |
b9170836 | 239 | return -EINVAL; |
8e677ce8 AC |
240 | |
241 | mutex_lock(&dbs_mutex); | |
cef9615a | 242 | dbs_tuners_ins.sampling_rate = max(input, min_sampling_rate); |
3fc54d37 | 243 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
244 | |
245 | return count; | |
246 | } | |
247 | ||
18a7247d | 248 | static ssize_t store_up_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
249 | const char *buf, size_t count) |
250 | { | |
251 | unsigned int input; | |
252 | int ret; | |
9acef487 | 253 | ret = sscanf(buf, "%u", &input); |
b9170836 | 254 | |
3fc54d37 | 255 | mutex_lock(&dbs_mutex); |
9acef487 | 256 | if (ret != 1 || input > 100 || |
8e677ce8 | 257 | input <= dbs_tuners_ins.down_threshold) { |
3fc54d37 | 258 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
259 | return -EINVAL; |
260 | } | |
261 | ||
262 | dbs_tuners_ins.up_threshold = input; | |
3fc54d37 | 263 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
264 | |
265 | return count; | |
266 | } | |
267 | ||
18a7247d | 268 | static ssize_t store_down_threshold(struct cpufreq_policy *unused, |
b9170836 DJ |
269 | const char *buf, size_t count) |
270 | { | |
271 | unsigned int input; | |
272 | int ret; | |
9acef487 | 273 | ret = sscanf(buf, "%u", &input); |
b9170836 | 274 | |
3fc54d37 | 275 | mutex_lock(&dbs_mutex); |
8e677ce8 AC |
276 | /* cannot be lower than 11 otherwise freq will not fall */ |
277 | if (ret != 1 || input < 11 || input > 100 || | |
278 | input >= dbs_tuners_ins.up_threshold) { | |
3fc54d37 | 279 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
280 | return -EINVAL; |
281 | } | |
282 | ||
283 | dbs_tuners_ins.down_threshold = input; | |
3fc54d37 | 284 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
285 | |
286 | return count; | |
287 | } | |
288 | ||
001893cd | 289 | static ssize_t store_ignore_nice_load(struct cpufreq_policy *policy, |
b9170836 DJ |
290 | const char *buf, size_t count) |
291 | { | |
292 | unsigned int input; | |
293 | int ret; | |
294 | ||
295 | unsigned int j; | |
18a7247d DJ |
296 | |
297 | ret = sscanf(buf, "%u", &input); | |
298 | if (ret != 1) | |
b9170836 DJ |
299 | return -EINVAL; |
300 | ||
18a7247d | 301 | if (input > 1) |
b9170836 | 302 | input = 1; |
18a7247d | 303 | |
3fc54d37 | 304 | mutex_lock(&dbs_mutex); |
18a7247d | 305 | if (input == dbs_tuners_ins.ignore_nice) { /* nothing to do */ |
3fc54d37 | 306 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
307 | return count; |
308 | } | |
309 | dbs_tuners_ins.ignore_nice = input; | |
310 | ||
8e677ce8 | 311 | /* we need to re-evaluate prev_cpu_idle */ |
dac1c1a5 | 312 | for_each_online_cpu(j) { |
8e677ce8 AC |
313 | struct cpu_dbs_info_s *dbs_info; |
314 | dbs_info = &per_cpu(cpu_dbs_info, j); | |
315 | dbs_info->prev_cpu_idle = get_cpu_idle_time(j, | |
316 | &dbs_info->prev_cpu_wall); | |
317 | if (dbs_tuners_ins.ignore_nice) | |
318 | dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
b9170836 | 319 | } |
3fc54d37 | 320 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
321 | |
322 | return count; | |
323 | } | |
324 | ||
325 | static ssize_t store_freq_step(struct cpufreq_policy *policy, | |
326 | const char *buf, size_t count) | |
327 | { | |
328 | unsigned int input; | |
329 | int ret; | |
18a7247d | 330 | ret = sscanf(buf, "%u", &input); |
b9170836 | 331 | |
18a7247d | 332 | if (ret != 1) |
b9170836 DJ |
333 | return -EINVAL; |
334 | ||
18a7247d | 335 | if (input > 100) |
b9170836 | 336 | input = 100; |
18a7247d | 337 | |
b9170836 DJ |
338 | /* no need to test here if freq_step is zero as the user might actually |
339 | * want this, they would be crazy though :) */ | |
3fc54d37 | 340 | mutex_lock(&dbs_mutex); |
b9170836 | 341 | dbs_tuners_ins.freq_step = input; |
3fc54d37 | 342 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
343 | |
344 | return count; | |
345 | } | |
346 | ||
347 | #define define_one_rw(_name) \ | |
348 | static struct freq_attr _name = \ | |
349 | __ATTR(_name, 0644, show_##_name, store_##_name) | |
350 | ||
351 | define_one_rw(sampling_rate); | |
352 | define_one_rw(sampling_down_factor); | |
353 | define_one_rw(up_threshold); | |
354 | define_one_rw(down_threshold); | |
001893cd | 355 | define_one_rw(ignore_nice_load); |
b9170836 DJ |
356 | define_one_rw(freq_step); |
357 | ||
9acef487 | 358 | static struct attribute *dbs_attributes[] = { |
b9170836 DJ |
359 | &sampling_rate_max.attr, |
360 | &sampling_rate_min.attr, | |
361 | &sampling_rate.attr, | |
362 | &sampling_down_factor.attr, | |
363 | &up_threshold.attr, | |
364 | &down_threshold.attr, | |
001893cd | 365 | &ignore_nice_load.attr, |
b9170836 DJ |
366 | &freq_step.attr, |
367 | NULL | |
368 | }; | |
369 | ||
370 | static struct attribute_group dbs_attr_group = { | |
371 | .attrs = dbs_attributes, | |
372 | .name = "conservative", | |
373 | }; | |
374 | ||
375 | /************************** sysfs end ************************/ | |
376 | ||
8e677ce8 | 377 | static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info) |
b9170836 | 378 | { |
8e677ce8 | 379 | unsigned int load = 0; |
f068c04b | 380 | unsigned int freq_target; |
b9170836 | 381 | |
8e677ce8 AC |
382 | struct cpufreq_policy *policy; |
383 | unsigned int j; | |
b9170836 | 384 | |
08a28e2e AC |
385 | policy = this_dbs_info->cur_policy; |
386 | ||
18a7247d | 387 | /* |
8e677ce8 AC |
388 | * Every sampling_rate, we check, if current idle time is less |
389 | * than 20% (default), then we try to increase frequency | |
390 | * Every sampling_rate*sampling_down_factor, we check, if current | |
391 | * idle time is more than 80%, then we try to decrease frequency | |
b9170836 | 392 | * |
18a7247d DJ |
393 | * Any frequency increase takes it to the maximum frequency. |
394 | * Frequency reduction happens at minimum steps of | |
8e677ce8 | 395 | * 5% (default) of maximum frequency |
b9170836 DJ |
396 | */ |
397 | ||
8e677ce8 AC |
398 | /* Get Absolute Load */ |
399 | for_each_cpu(j, policy->cpus) { | |
400 | struct cpu_dbs_info_s *j_dbs_info; | |
401 | cputime64_t cur_wall_time, cur_idle_time; | |
402 | unsigned int idle_time, wall_time; | |
b9170836 | 403 | |
8e677ce8 AC |
404 | j_dbs_info = &per_cpu(cpu_dbs_info, j); |
405 | ||
406 | cur_idle_time = get_cpu_idle_time(j, &cur_wall_time); | |
407 | ||
408 | wall_time = (unsigned int) cputime64_sub(cur_wall_time, | |
409 | j_dbs_info->prev_cpu_wall); | |
410 | j_dbs_info->prev_cpu_wall = cur_wall_time; | |
08a28e2e | 411 | |
8e677ce8 AC |
412 | idle_time = (unsigned int) cputime64_sub(cur_idle_time, |
413 | j_dbs_info->prev_cpu_idle); | |
414 | j_dbs_info->prev_cpu_idle = cur_idle_time; | |
b9170836 | 415 | |
8e677ce8 AC |
416 | if (dbs_tuners_ins.ignore_nice) { |
417 | cputime64_t cur_nice; | |
418 | unsigned long cur_nice_jiffies; | |
419 | ||
420 | cur_nice = cputime64_sub(kstat_cpu(j).cpustat.nice, | |
421 | j_dbs_info->prev_cpu_nice); | |
422 | /* | |
423 | * Assumption: nice time between sampling periods will | |
424 | * be less than 2^32 jiffies for 32 bit sys | |
425 | */ | |
426 | cur_nice_jiffies = (unsigned long) | |
427 | cputime64_to_jiffies64(cur_nice); | |
428 | ||
429 | j_dbs_info->prev_cpu_nice = kstat_cpu(j).cpustat.nice; | |
430 | idle_time += jiffies_to_usecs(cur_nice_jiffies); | |
431 | } | |
432 | ||
433 | if (unlikely(!wall_time || wall_time < idle_time)) | |
434 | continue; | |
435 | ||
436 | load = 100 * (wall_time - idle_time) / wall_time; | |
437 | } | |
438 | ||
439 | /* | |
440 | * break out if we 'cannot' reduce the speed as the user might | |
441 | * want freq_step to be zero | |
442 | */ | |
443 | if (dbs_tuners_ins.freq_step == 0) | |
444 | return; | |
b9170836 | 445 | |
8e677ce8 AC |
446 | /* Check for frequency increase */ |
447 | if (load > dbs_tuners_ins.up_threshold) { | |
a159b827 | 448 | this_dbs_info->down_skip = 0; |
790d76fa | 449 | |
b9170836 | 450 | /* if we are already at full speed then break out early */ |
a159b827 | 451 | if (this_dbs_info->requested_freq == policy->max) |
b9170836 | 452 | return; |
18a7247d | 453 | |
f068c04b | 454 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 DJ |
455 | |
456 | /* max freq cannot be less than 100. But who knows.... */ | |
f068c04b DJ |
457 | if (unlikely(freq_target == 0)) |
458 | freq_target = 5; | |
18a7247d | 459 | |
f068c04b | 460 | this_dbs_info->requested_freq += freq_target; |
a159b827 AC |
461 | if (this_dbs_info->requested_freq > policy->max) |
462 | this_dbs_info->requested_freq = policy->max; | |
b9170836 | 463 | |
a159b827 | 464 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
b9170836 | 465 | CPUFREQ_RELATION_H); |
b9170836 DJ |
466 | return; |
467 | } | |
468 | ||
8e677ce8 AC |
469 | /* |
470 | * The optimal frequency is the frequency that is the lowest that | |
471 | * can support the current CPU usage without triggering the up | |
472 | * policy. To be safe, we focus 10 points under the threshold. | |
473 | */ | |
474 | if (load < (dbs_tuners_ins.down_threshold - 10)) { | |
f068c04b | 475 | freq_target = (dbs_tuners_ins.freq_step * policy->max) / 100; |
b9170836 | 476 | |
f068c04b | 477 | this_dbs_info->requested_freq -= freq_target; |
a159b827 AC |
478 | if (this_dbs_info->requested_freq < policy->min) |
479 | this_dbs_info->requested_freq = policy->min; | |
b9170836 | 480 | |
8e677ce8 AC |
481 | /* |
482 | * if we cannot reduce the frequency anymore, break out early | |
483 | */ | |
484 | if (policy->cur == policy->min) | |
485 | return; | |
486 | ||
a159b827 | 487 | __cpufreq_driver_target(policy, this_dbs_info->requested_freq, |
2c906b31 | 488 | CPUFREQ_RELATION_H); |
b9170836 DJ |
489 | return; |
490 | } | |
491 | } | |
492 | ||
c4028958 | 493 | static void do_dbs_timer(struct work_struct *work) |
18a7247d | 494 | { |
8e677ce8 AC |
495 | struct cpu_dbs_info_s *dbs_info = |
496 | container_of(work, struct cpu_dbs_info_s, work.work); | |
497 | unsigned int cpu = dbs_info->cpu; | |
498 | ||
499 | /* We want all CPUs to do sampling nearly on same jiffy */ | |
500 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
501 | ||
502 | delay -= jiffies % delay; | |
503 | ||
504 | if (lock_policy_rwsem_write(cpu) < 0) | |
505 | return; | |
506 | ||
507 | if (!dbs_info->enable) { | |
508 | unlock_policy_rwsem_write(cpu); | |
509 | return; | |
510 | } | |
511 | ||
512 | dbs_check_cpu(dbs_info); | |
513 | ||
514 | queue_delayed_work_on(cpu, kconservative_wq, &dbs_info->work, delay); | |
515 | unlock_policy_rwsem_write(cpu); | |
18a7247d | 516 | } |
b9170836 | 517 | |
8e677ce8 | 518 | static inline void dbs_timer_init(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 519 | { |
8e677ce8 AC |
520 | /* We want all CPUs to do sampling nearly on same jiffy */ |
521 | int delay = usecs_to_jiffies(dbs_tuners_ins.sampling_rate); | |
522 | delay -= jiffies % delay; | |
523 | ||
524 | dbs_info->enable = 1; | |
525 | INIT_DELAYED_WORK_DEFERRABLE(&dbs_info->work, do_dbs_timer); | |
526 | queue_delayed_work_on(dbs_info->cpu, kconservative_wq, &dbs_info->work, | |
527 | delay); | |
b9170836 DJ |
528 | } |
529 | ||
8e677ce8 | 530 | static inline void dbs_timer_exit(struct cpu_dbs_info_s *dbs_info) |
b9170836 | 531 | { |
8e677ce8 | 532 | dbs_info->enable = 0; |
b253d2b2 | 533 | cancel_delayed_work_sync(&dbs_info->work); |
b9170836 DJ |
534 | } |
535 | ||
536 | static int cpufreq_governor_dbs(struct cpufreq_policy *policy, | |
537 | unsigned int event) | |
538 | { | |
539 | unsigned int cpu = policy->cpu; | |
540 | struct cpu_dbs_info_s *this_dbs_info; | |
541 | unsigned int j; | |
914f7c31 | 542 | int rc; |
b9170836 DJ |
543 | |
544 | this_dbs_info = &per_cpu(cpu_dbs_info, cpu); | |
545 | ||
546 | switch (event) { | |
547 | case CPUFREQ_GOV_START: | |
18a7247d | 548 | if ((!cpu_online(cpu)) || (!policy->cur)) |
b9170836 DJ |
549 | return -EINVAL; |
550 | ||
b9170836 DJ |
551 | if (this_dbs_info->enable) /* Already enabled */ |
552 | break; | |
18a7247d | 553 | |
3fc54d37 | 554 | mutex_lock(&dbs_mutex); |
914f7c31 JG |
555 | |
556 | rc = sysfs_create_group(&policy->kobj, &dbs_attr_group); | |
557 | if (rc) { | |
558 | mutex_unlock(&dbs_mutex); | |
559 | return rc; | |
560 | } | |
561 | ||
835481d9 | 562 | for_each_cpu(j, policy->cpus) { |
b9170836 DJ |
563 | struct cpu_dbs_info_s *j_dbs_info; |
564 | j_dbs_info = &per_cpu(cpu_dbs_info, j); | |
565 | j_dbs_info->cur_policy = policy; | |
18a7247d | 566 | |
8e677ce8 AC |
567 | j_dbs_info->prev_cpu_idle = get_cpu_idle_time(j, |
568 | &j_dbs_info->prev_cpu_wall); | |
569 | if (dbs_tuners_ins.ignore_nice) { | |
570 | j_dbs_info->prev_cpu_nice = | |
571 | kstat_cpu(j).cpustat.nice; | |
572 | } | |
b9170836 | 573 | } |
a159b827 AC |
574 | this_dbs_info->down_skip = 0; |
575 | this_dbs_info->requested_freq = policy->cur; | |
914f7c31 | 576 | |
b9170836 DJ |
577 | dbs_enable++; |
578 | /* | |
579 | * Start the timerschedule work, when this governor | |
580 | * is used for first time | |
581 | */ | |
582 | if (dbs_enable == 1) { | |
583 | unsigned int latency; | |
584 | /* policy latency is in nS. Convert it to uS first */ | |
2c906b31 AC |
585 | latency = policy->cpuinfo.transition_latency / 1000; |
586 | if (latency == 0) | |
587 | latency = 1; | |
b9170836 | 588 | |
cef9615a TR |
589 | /* |
590 | * conservative does not implement micro like ondemand | |
591 | * governor, thus we are bound to jiffes/HZ | |
592 | */ | |
593 | min_sampling_rate = | |
594 | MIN_SAMPLING_RATE_RATIO * jiffies_to_usecs(10); | |
595 | /* Bring kernel and HW constraints together */ | |
596 | min_sampling_rate = max(min_sampling_rate, | |
597 | MIN_LATENCY_MULTIPLIER * latency); | |
598 | dbs_tuners_ins.sampling_rate = | |
599 | max(min_sampling_rate, | |
600 | latency * LATENCY_MULTIPLIER); | |
b9170836 | 601 | |
a8d7c3bc EO |
602 | cpufreq_register_notifier( |
603 | &dbs_cpufreq_notifier_block, | |
604 | CPUFREQ_TRANSITION_NOTIFIER); | |
b9170836 | 605 | } |
8e677ce8 | 606 | dbs_timer_init(this_dbs_info); |
18a7247d | 607 | |
3fc54d37 | 608 | mutex_unlock(&dbs_mutex); |
8e677ce8 | 609 | |
b9170836 DJ |
610 | break; |
611 | ||
612 | case CPUFREQ_GOV_STOP: | |
3fc54d37 | 613 | mutex_lock(&dbs_mutex); |
8e677ce8 | 614 | dbs_timer_exit(this_dbs_info); |
b9170836 DJ |
615 | sysfs_remove_group(&policy->kobj, &dbs_attr_group); |
616 | dbs_enable--; | |
8e677ce8 | 617 | |
b9170836 DJ |
618 | /* |
619 | * Stop the timerschedule work, when this governor | |
620 | * is used for first time | |
621 | */ | |
8e677ce8 | 622 | if (dbs_enable == 0) |
a8d7c3bc EO |
623 | cpufreq_unregister_notifier( |
624 | &dbs_cpufreq_notifier_block, | |
625 | CPUFREQ_TRANSITION_NOTIFIER); | |
a8d7c3bc | 626 | |
3fc54d37 | 627 | mutex_unlock(&dbs_mutex); |
b9170836 DJ |
628 | |
629 | break; | |
630 | ||
631 | case CPUFREQ_GOV_LIMITS: | |
3fc54d37 | 632 | mutex_lock(&dbs_mutex); |
b9170836 DJ |
633 | if (policy->max < this_dbs_info->cur_policy->cur) |
634 | __cpufreq_driver_target( | |
635 | this_dbs_info->cur_policy, | |
18a7247d | 636 | policy->max, CPUFREQ_RELATION_H); |
b9170836 DJ |
637 | else if (policy->min > this_dbs_info->cur_policy->cur) |
638 | __cpufreq_driver_target( | |
639 | this_dbs_info->cur_policy, | |
18a7247d | 640 | policy->min, CPUFREQ_RELATION_L); |
3fc54d37 | 641 | mutex_unlock(&dbs_mutex); |
8e677ce8 | 642 | |
b9170836 DJ |
643 | break; |
644 | } | |
645 | return 0; | |
646 | } | |
647 | ||
c4d14bc0 SW |
648 | #ifndef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
649 | static | |
650 | #endif | |
1c256245 TR |
651 | struct cpufreq_governor cpufreq_gov_conservative = { |
652 | .name = "conservative", | |
653 | .governor = cpufreq_governor_dbs, | |
654 | .max_transition_latency = TRANSITION_LATENCY_LIMIT, | |
655 | .owner = THIS_MODULE, | |
b9170836 DJ |
656 | }; |
657 | ||
658 | static int __init cpufreq_gov_dbs_init(void) | |
659 | { | |
8e677ce8 AC |
660 | int err; |
661 | ||
662 | kconservative_wq = create_workqueue("kconservative"); | |
663 | if (!kconservative_wq) { | |
664 | printk(KERN_ERR "Creation of kconservative failed\n"); | |
665 | return -EFAULT; | |
666 | } | |
667 | ||
668 | err = cpufreq_register_governor(&cpufreq_gov_conservative); | |
669 | if (err) | |
670 | destroy_workqueue(kconservative_wq); | |
671 | ||
672 | return err; | |
b9170836 DJ |
673 | } |
674 | ||
675 | static void __exit cpufreq_gov_dbs_exit(void) | |
676 | { | |
1c256245 | 677 | cpufreq_unregister_governor(&cpufreq_gov_conservative); |
8e677ce8 | 678 | destroy_workqueue(kconservative_wq); |
b9170836 DJ |
679 | } |
680 | ||
681 | ||
11a80a9c | 682 | MODULE_AUTHOR("Alexander Clouter <alex@digriz.org.uk>"); |
9acef487 | 683 | MODULE_DESCRIPTION("'cpufreq_conservative' - A dynamic cpufreq governor for " |
b9170836 DJ |
684 | "Low Latency Frequency Transition capable processors " |
685 | "optimised for use in a battery environment"); | |
9acef487 | 686 | MODULE_LICENSE("GPL"); |
b9170836 | 687 | |
6915719b JW |
688 | #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE |
689 | fs_initcall(cpufreq_gov_dbs_init); | |
690 | #else | |
b9170836 | 691 | module_init(cpufreq_gov_dbs_init); |
6915719b | 692 | #endif |
b9170836 | 693 | module_exit(cpufreq_gov_dbs_exit); |