Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
4471a34f VK |
13 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
14 | ||
5ff0a268 | 15 | #include <linux/cpu.h> |
4471a34f | 16 | #include <linux/percpu-defs.h> |
4d5dcc42 | 17 | #include <linux/slab.h> |
80800913 | 18 | #include <linux/tick.h> |
4471a34f | 19 | #include "cpufreq_governor.h" |
1da177e4 | 20 | |
06eb09d1 | 21 | /* On-demand governor macros */ |
1da177e4 | 22 | #define DEF_FREQUENCY_UP_THRESHOLD (80) |
3f78a9f7 DN |
23 | #define DEF_SAMPLING_DOWN_FACTOR (1) |
24 | #define MAX_SAMPLING_DOWN_FACTOR (100000) | |
80800913 | 25 | #define MICRO_FREQUENCY_UP_THRESHOLD (95) |
cef9615a | 26 | #define MICRO_FREQUENCY_MIN_SAMPLE_RATE (10000) |
c29f1403 | 27 | #define MIN_FREQUENCY_UP_THRESHOLD (11) |
1da177e4 LT |
28 | #define MAX_FREQUENCY_UP_THRESHOLD (100) |
29 | ||
4471a34f | 30 | static DEFINE_PER_CPU(struct od_cpu_dbs_info_s, od_cpu_dbs_info); |
1da177e4 | 31 | |
fb30809e JS |
32 | static struct od_ops od_ops; |
33 | ||
c2837558 JS |
34 | static unsigned int default_powersave_bias; |
35 | ||
4471a34f | 36 | static void ondemand_powersave_bias_init_cpu(int cpu) |
6b8fcd90 | 37 | { |
4471a34f | 38 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu); |
6b8fcd90 | 39 | |
4471a34f VK |
40 | dbs_info->freq_table = cpufreq_frequency_get_table(cpu); |
41 | dbs_info->freq_lo = 0; | |
42 | } | |
6b8fcd90 | 43 | |
4471a34f VK |
44 | /* |
45 | * Not all CPUs want IO time to be accounted as busy; this depends on how | |
46 | * efficient idling at a higher frequency/voltage is. | |
47 | * Pavel Machek says this is not so for various generations of AMD and old | |
48 | * Intel systems. | |
06eb09d1 | 49 | * Mike Chan (android.com) claims this is also not true for ARM. |
4471a34f VK |
50 | * Because of this, whitelist specific known (series) of CPUs by default, and |
51 | * leave all others up to the user. | |
52 | */ | |
53 | static int should_io_be_busy(void) | |
54 | { | |
55 | #if defined(CONFIG_X86) | |
56 | /* | |
06eb09d1 | 57 | * For Intel, Core 2 (model 15) and later have an efficient idle. |
4471a34f VK |
58 | */ |
59 | if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && | |
60 | boot_cpu_data.x86 == 6 && | |
61 | boot_cpu_data.x86_model >= 15) | |
62 | return 1; | |
63 | #endif | |
64 | return 0; | |
6b8fcd90 AV |
65 | } |
66 | ||
05ca0350 AS |
67 | /* |
68 | * Find right freq to be set now with powersave_bias on. | |
69 | * Returns the freq_hi to be used right now and will set freq_hi_jiffies, | |
70 | * freq_lo, and freq_lo_jiffies in percpu area for averaging freqs. | |
71 | */ | |
fb30809e | 72 | static unsigned int generic_powersave_bias_target(struct cpufreq_policy *policy, |
4471a34f | 73 | unsigned int freq_next, unsigned int relation) |
05ca0350 AS |
74 | { |
75 | unsigned int freq_req, freq_reduc, freq_avg; | |
76 | unsigned int freq_hi, freq_lo; | |
77 | unsigned int index = 0; | |
78 | unsigned int jiffies_total, jiffies_hi, jiffies_lo; | |
4471a34f | 79 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, |
245b2e70 | 80 | policy->cpu); |
4d5dcc42 VK |
81 | struct dbs_data *dbs_data = policy->governor_data; |
82 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; | |
05ca0350 AS |
83 | |
84 | if (!dbs_info->freq_table) { | |
85 | dbs_info->freq_lo = 0; | |
86 | dbs_info->freq_lo_jiffies = 0; | |
87 | return freq_next; | |
88 | } | |
89 | ||
90 | cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_next, | |
91 | relation, &index); | |
92 | freq_req = dbs_info->freq_table[index].frequency; | |
4d5dcc42 | 93 | freq_reduc = freq_req * od_tuners->powersave_bias / 1000; |
05ca0350 AS |
94 | freq_avg = freq_req - freq_reduc; |
95 | ||
96 | /* Find freq bounds for freq_avg in freq_table */ | |
97 | index = 0; | |
98 | cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg, | |
99 | CPUFREQ_RELATION_H, &index); | |
100 | freq_lo = dbs_info->freq_table[index].frequency; | |
101 | index = 0; | |
102 | cpufreq_frequency_table_target(policy, dbs_info->freq_table, freq_avg, | |
103 | CPUFREQ_RELATION_L, &index); | |
104 | freq_hi = dbs_info->freq_table[index].frequency; | |
105 | ||
106 | /* Find out how long we have to be in hi and lo freqs */ | |
107 | if (freq_hi == freq_lo) { | |
108 | dbs_info->freq_lo = 0; | |
109 | dbs_info->freq_lo_jiffies = 0; | |
110 | return freq_lo; | |
111 | } | |
4d5dcc42 | 112 | jiffies_total = usecs_to_jiffies(od_tuners->sampling_rate); |
05ca0350 AS |
113 | jiffies_hi = (freq_avg - freq_lo) * jiffies_total; |
114 | jiffies_hi += ((freq_hi - freq_lo) / 2); | |
115 | jiffies_hi /= (freq_hi - freq_lo); | |
116 | jiffies_lo = jiffies_total - jiffies_hi; | |
117 | dbs_info->freq_lo = freq_lo; | |
118 | dbs_info->freq_lo_jiffies = jiffies_lo; | |
119 | dbs_info->freq_hi_jiffies = jiffies_hi; | |
120 | return freq_hi; | |
121 | } | |
122 | ||
123 | static void ondemand_powersave_bias_init(void) | |
124 | { | |
125 | int i; | |
126 | for_each_online_cpu(i) { | |
5a75c828 | 127 | ondemand_powersave_bias_init_cpu(i); |
05ca0350 AS |
128 | } |
129 | } | |
130 | ||
3a3e9e06 | 131 | static void dbs_freq_increase(struct cpufreq_policy *policy, unsigned int freq) |
4471a34f | 132 | { |
3a3e9e06 | 133 | struct dbs_data *dbs_data = policy->governor_data; |
4d5dcc42 VK |
134 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
135 | ||
136 | if (od_tuners->powersave_bias) | |
3a3e9e06 | 137 | freq = od_ops.powersave_bias_target(policy, freq, |
fb30809e | 138 | CPUFREQ_RELATION_H); |
3a3e9e06 | 139 | else if (policy->cur == policy->max) |
4471a34f | 140 | return; |
0e625ac1 | 141 | |
3a3e9e06 | 142 | __cpufreq_driver_target(policy, freq, od_tuners->powersave_bias ? |
4471a34f VK |
143 | CPUFREQ_RELATION_L : CPUFREQ_RELATION_H); |
144 | } | |
145 | ||
146 | /* | |
147 | * Every sampling_rate, we check, if current idle time is less than 20% | |
dfa5bb62 SK |
148 | * (default), then we try to increase frequency. Else, we adjust the frequency |
149 | * proportional to load. | |
4471a34f | 150 | */ |
dfa5bb62 | 151 | static void od_check_cpu(int cpu, unsigned int load) |
1da177e4 | 152 | { |
4471a34f | 153 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, cpu); |
44152cb8 | 154 | struct cpufreq_policy *policy = dbs_info->cdbs.shared->policy; |
4d5dcc42 VK |
155 | struct dbs_data *dbs_data = policy->governor_data; |
156 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; | |
4471a34f VK |
157 | |
158 | dbs_info->freq_lo = 0; | |
159 | ||
160 | /* Check for frequency increase */ | |
dfa5bb62 | 161 | if (load > od_tuners->up_threshold) { |
4471a34f VK |
162 | /* If switching to max speed, apply sampling_down_factor */ |
163 | if (policy->cur < policy->max) | |
164 | dbs_info->rate_mult = | |
4d5dcc42 | 165 | od_tuners->sampling_down_factor; |
4471a34f | 166 | dbs_freq_increase(policy, policy->max); |
dfa5bb62 SK |
167 | } else { |
168 | /* Calculate the next frequency proportional to load */ | |
6393d6a1 SK |
169 | unsigned int freq_next, min_f, max_f; |
170 | ||
171 | min_f = policy->cpuinfo.min_freq; | |
172 | max_f = policy->cpuinfo.max_freq; | |
173 | freq_next = min_f + load * (max_f - min_f) / 100; | |
4471a34f VK |
174 | |
175 | /* No longer fully busy, reset rate_mult */ | |
176 | dbs_info->rate_mult = 1; | |
177 | ||
4d5dcc42 | 178 | if (!od_tuners->powersave_bias) { |
4471a34f | 179 | __cpufreq_driver_target(policy, freq_next, |
6393d6a1 | 180 | CPUFREQ_RELATION_C); |
fb30809e | 181 | return; |
4471a34f | 182 | } |
fb30809e JS |
183 | |
184 | freq_next = od_ops.powersave_bias_target(policy, freq_next, | |
185 | CPUFREQ_RELATION_L); | |
6393d6a1 | 186 | __cpufreq_driver_target(policy, freq_next, CPUFREQ_RELATION_C); |
4471a34f | 187 | } |
1da177e4 LT |
188 | } |
189 | ||
9be4fd2c | 190 | static unsigned int od_dbs_timer(struct cpufreq_policy *policy) |
4471a34f | 191 | { |
affde5d0 | 192 | struct dbs_data *dbs_data = policy->governor_data; |
44152cb8 | 193 | unsigned int cpu = policy->cpu; |
43e0ee36 | 194 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, |
4447266b | 195 | cpu); |
4d5dcc42 | 196 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
43e0ee36 | 197 | int delay = 0, sample_type = dbs_info->sample_type; |
4447266b | 198 | |
4471a34f | 199 | /* Common NORMAL_SAMPLE setup */ |
43e0ee36 | 200 | dbs_info->sample_type = OD_NORMAL_SAMPLE; |
4471a34f | 201 | if (sample_type == OD_SUB_SAMPLE) { |
43e0ee36 VK |
202 | delay = dbs_info->freq_lo_jiffies; |
203 | __cpufreq_driver_target(policy, dbs_info->freq_lo, | |
42994af6 | 204 | CPUFREQ_RELATION_H); |
4471a34f | 205 | } else { |
9d445920 | 206 | dbs_check_cpu(dbs_data, cpu); |
43e0ee36 | 207 | if (dbs_info->freq_lo) { |
4471a34f | 208 | /* Setup timer for SUB_SAMPLE */ |
43e0ee36 VK |
209 | dbs_info->sample_type = OD_SUB_SAMPLE; |
210 | delay = dbs_info->freq_hi_jiffies; | |
4471a34f VK |
211 | } |
212 | } | |
213 | ||
9d445920 VK |
214 | if (!delay) |
215 | delay = delay_for_sampling_rate(od_tuners->sampling_rate | |
43e0ee36 | 216 | * dbs_info->rate_mult); |
9d445920 | 217 | |
43e0ee36 | 218 | return delay; |
da53d61e FB |
219 | } |
220 | ||
4471a34f | 221 | /************************** sysfs interface ************************/ |
7bdad34d | 222 | static struct dbs_governor od_dbs_gov; |
1da177e4 | 223 | |
fd0ef7a0 MH |
224 | /** |
225 | * update_sampling_rate - update sampling rate effective immediately if needed. | |
226 | * @new_rate: new sampling rate | |
227 | * | |
06eb09d1 | 228 | * If new rate is smaller than the old, simply updating |
4471a34f VK |
229 | * dbs_tuners_int.sampling_rate might not be appropriate. For example, if the |
230 | * original sampling_rate was 1 second and the requested new sampling rate is 10 | |
231 | * ms because the user needs immediate reaction from ondemand governor, but not | |
232 | * sure if higher frequency will be required or not, then, the governor may | |
233 | * change the sampling rate too late; up to 1 second later. Thus, if we are | |
234 | * reducing the sampling rate, we need to make the new value effective | |
235 | * immediately. | |
fd0ef7a0 | 236 | */ |
4d5dcc42 VK |
237 | static void update_sampling_rate(struct dbs_data *dbs_data, |
238 | unsigned int new_rate) | |
fd0ef7a0 | 239 | { |
4d5dcc42 | 240 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
f08f638b | 241 | struct cpumask cpumask; |
fd0ef7a0 MH |
242 | int cpu; |
243 | ||
4d5dcc42 VK |
244 | od_tuners->sampling_rate = new_rate = max(new_rate, |
245 | dbs_data->min_sampling_rate); | |
fd0ef7a0 | 246 | |
e128c864 VK |
247 | /* |
248 | * Lock governor so that governor start/stop can't execute in parallel. | |
249 | */ | |
2bb8d94f | 250 | mutex_lock(&dbs_data_mutex); |
e128c864 | 251 | |
f08f638b VK |
252 | cpumask_copy(&cpumask, cpu_online_mask); |
253 | ||
254 | for_each_cpu(cpu, &cpumask) { | |
fd0ef7a0 | 255 | struct cpufreq_policy *policy; |
4471a34f | 256 | struct od_cpu_dbs_info_s *dbs_info; |
e128c864 VK |
257 | struct cpu_dbs_info *cdbs; |
258 | struct cpu_common_dbs_info *shared; | |
fd0ef7a0 | 259 | |
e128c864 VK |
260 | dbs_info = &per_cpu(od_cpu_dbs_info, cpu); |
261 | cdbs = &dbs_info->cdbs; | |
262 | shared = cdbs->shared; | |
263 | ||
264 | /* | |
265 | * A valid shared and shared->policy means governor hasn't | |
266 | * stopped or exited yet. | |
267 | */ | |
268 | if (!shared || !shared->policy) | |
fd0ef7a0 | 269 | continue; |
e128c864 VK |
270 | |
271 | policy = shared->policy; | |
272 | ||
f08f638b VK |
273 | /* clear all CPUs of this policy */ |
274 | cpumask_andnot(&cpumask, &cpumask, policy->cpus); | |
275 | ||
e128c864 VK |
276 | /* |
277 | * Update sampling rate for CPUs whose policy is governed by | |
278 | * dbs_data. In case of governor_per_policy, only a single | |
279 | * policy will be governed by dbs_data, otherwise there can be | |
280 | * multiple policies that are governed by the same dbs_data. | |
281 | */ | |
9be4fd2c RW |
282 | if (dbs_data == policy->governor_data) { |
283 | mutex_lock(&shared->timer_mutex); | |
284 | /* | |
285 | * On 32-bit architectures this may race with the | |
286 | * sample_delay_ns read in dbs_update_util_handler(), | |
287 | * but that really doesn't matter. If the read returns | |
288 | * a value that's too big, the sample will be skipped, | |
289 | * but the next invocation of dbs_update_util_handler() | |
290 | * (when the update has been completed) will take a | |
291 | * sample. If the returned value is too small, the | |
292 | * sample will be taken immediately, but that isn't a | |
293 | * problem, as we want the new rate to take effect | |
294 | * immediately anyway. | |
295 | * | |
296 | * If this runs in parallel with dbs_work_handler(), we | |
297 | * may end up overwriting the sample_delay_ns value that | |
298 | * it has just written, but the difference should not be | |
299 | * too big and it will be corrected next time a sample | |
300 | * is taken, so it shouldn't be significant. | |
301 | */ | |
302 | gov_update_sample_delay(shared, new_rate); | |
303 | mutex_unlock(&shared->timer_mutex); | |
fd0ef7a0 | 304 | } |
fd0ef7a0 | 305 | } |
e128c864 | 306 | |
2bb8d94f | 307 | mutex_unlock(&dbs_data_mutex); |
fd0ef7a0 MH |
308 | } |
309 | ||
4d5dcc42 VK |
310 | static ssize_t store_sampling_rate(struct dbs_data *dbs_data, const char *buf, |
311 | size_t count) | |
1da177e4 LT |
312 | { |
313 | unsigned int input; | |
314 | int ret; | |
ffac80e9 | 315 | ret = sscanf(buf, "%u", &input); |
5a75c828 | 316 | if (ret != 1) |
317 | return -EINVAL; | |
4d5dcc42 VK |
318 | |
319 | update_sampling_rate(dbs_data, input); | |
1da177e4 LT |
320 | return count; |
321 | } | |
322 | ||
4d5dcc42 VK |
323 | static ssize_t store_io_is_busy(struct dbs_data *dbs_data, const char *buf, |
324 | size_t count) | |
19379b11 | 325 | { |
4d5dcc42 | 326 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
19379b11 AV |
327 | unsigned int input; |
328 | int ret; | |
9366d840 | 329 | unsigned int j; |
19379b11 AV |
330 | |
331 | ret = sscanf(buf, "%u", &input); | |
332 | if (ret != 1) | |
333 | return -EINVAL; | |
4d5dcc42 | 334 | od_tuners->io_is_busy = !!input; |
9366d840 SK |
335 | |
336 | /* we need to re-evaluate prev_cpu_idle */ | |
337 | for_each_online_cpu(j) { | |
338 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, | |
339 | j); | |
340 | dbs_info->cdbs.prev_cpu_idle = get_cpu_idle_time(j, | |
341 | &dbs_info->cdbs.prev_cpu_wall, od_tuners->io_is_busy); | |
342 | } | |
19379b11 AV |
343 | return count; |
344 | } | |
345 | ||
4d5dcc42 VK |
346 | static ssize_t store_up_threshold(struct dbs_data *dbs_data, const char *buf, |
347 | size_t count) | |
1da177e4 | 348 | { |
4d5dcc42 | 349 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
1da177e4 LT |
350 | unsigned int input; |
351 | int ret; | |
ffac80e9 | 352 | ret = sscanf(buf, "%u", &input); |
1da177e4 | 353 | |
32ee8c3e | 354 | if (ret != 1 || input > MAX_FREQUENCY_UP_THRESHOLD || |
c29f1403 | 355 | input < MIN_FREQUENCY_UP_THRESHOLD) { |
1da177e4 LT |
356 | return -EINVAL; |
357 | } | |
4bd4e428 | 358 | |
4d5dcc42 | 359 | od_tuners->up_threshold = input; |
1da177e4 LT |
360 | return count; |
361 | } | |
362 | ||
4d5dcc42 VK |
363 | static ssize_t store_sampling_down_factor(struct dbs_data *dbs_data, |
364 | const char *buf, size_t count) | |
3f78a9f7 | 365 | { |
4d5dcc42 | 366 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
3f78a9f7 DN |
367 | unsigned int input, j; |
368 | int ret; | |
369 | ret = sscanf(buf, "%u", &input); | |
370 | ||
371 | if (ret != 1 || input > MAX_SAMPLING_DOWN_FACTOR || input < 1) | |
372 | return -EINVAL; | |
4d5dcc42 | 373 | od_tuners->sampling_down_factor = input; |
3f78a9f7 DN |
374 | |
375 | /* Reset down sampling multiplier in case it was active */ | |
376 | for_each_online_cpu(j) { | |
4471a34f VK |
377 | struct od_cpu_dbs_info_s *dbs_info = &per_cpu(od_cpu_dbs_info, |
378 | j); | |
3f78a9f7 DN |
379 | dbs_info->rate_mult = 1; |
380 | } | |
3f78a9f7 DN |
381 | return count; |
382 | } | |
383 | ||
6c4640c3 VK |
384 | static ssize_t store_ignore_nice_load(struct dbs_data *dbs_data, |
385 | const char *buf, size_t count) | |
3d5ee9e5 | 386 | { |
4d5dcc42 | 387 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
3d5ee9e5 DJ |
388 | unsigned int input; |
389 | int ret; | |
390 | ||
391 | unsigned int j; | |
32ee8c3e | 392 | |
ffac80e9 | 393 | ret = sscanf(buf, "%u", &input); |
2b03f891 | 394 | if (ret != 1) |
3d5ee9e5 DJ |
395 | return -EINVAL; |
396 | ||
2b03f891 | 397 | if (input > 1) |
3d5ee9e5 | 398 | input = 1; |
32ee8c3e | 399 | |
6c4640c3 | 400 | if (input == od_tuners->ignore_nice_load) { /* nothing to do */ |
3d5ee9e5 DJ |
401 | return count; |
402 | } | |
6c4640c3 | 403 | od_tuners->ignore_nice_load = input; |
3d5ee9e5 | 404 | |
ccb2fe20 | 405 | /* we need to re-evaluate prev_cpu_idle */ |
dac1c1a5 | 406 | for_each_online_cpu(j) { |
4471a34f | 407 | struct od_cpu_dbs_info_s *dbs_info; |
245b2e70 | 408 | dbs_info = &per_cpu(od_cpu_dbs_info, j); |
4471a34f | 409 | dbs_info->cdbs.prev_cpu_idle = get_cpu_idle_time(j, |
9366d840 | 410 | &dbs_info->cdbs.prev_cpu_wall, od_tuners->io_is_busy); |
6c4640c3 | 411 | if (od_tuners->ignore_nice_load) |
4471a34f VK |
412 | dbs_info->cdbs.prev_cpu_nice = |
413 | kcpustat_cpu(j).cpustat[CPUTIME_NICE]; | |
1ca3abdb | 414 | |
3d5ee9e5 | 415 | } |
3d5ee9e5 DJ |
416 | return count; |
417 | } | |
418 | ||
4d5dcc42 VK |
419 | static ssize_t store_powersave_bias(struct dbs_data *dbs_data, const char *buf, |
420 | size_t count) | |
05ca0350 | 421 | { |
4d5dcc42 | 422 | struct od_dbs_tuners *od_tuners = dbs_data->tuners; |
05ca0350 AS |
423 | unsigned int input; |
424 | int ret; | |
425 | ret = sscanf(buf, "%u", &input); | |
426 | ||
427 | if (ret != 1) | |
428 | return -EINVAL; | |
429 | ||
430 | if (input > 1000) | |
431 | input = 1000; | |
432 | ||
4d5dcc42 | 433 | od_tuners->powersave_bias = input; |
05ca0350 | 434 | ondemand_powersave_bias_init(); |
05ca0350 AS |
435 | return count; |
436 | } | |
437 | ||
4d5dcc42 VK |
438 | show_store_one(od, sampling_rate); |
439 | show_store_one(od, io_is_busy); | |
440 | show_store_one(od, up_threshold); | |
441 | show_store_one(od, sampling_down_factor); | |
6c4640c3 | 442 | show_store_one(od, ignore_nice_load); |
4d5dcc42 VK |
443 | show_store_one(od, powersave_bias); |
444 | declare_show_sampling_rate_min(od); | |
445 | ||
446 | gov_sys_pol_attr_rw(sampling_rate); | |
447 | gov_sys_pol_attr_rw(io_is_busy); | |
448 | gov_sys_pol_attr_rw(up_threshold); | |
449 | gov_sys_pol_attr_rw(sampling_down_factor); | |
6c4640c3 | 450 | gov_sys_pol_attr_rw(ignore_nice_load); |
4d5dcc42 VK |
451 | gov_sys_pol_attr_rw(powersave_bias); |
452 | gov_sys_pol_attr_ro(sampling_rate_min); | |
453 | ||
454 | static struct attribute *dbs_attributes_gov_sys[] = { | |
455 | &sampling_rate_min_gov_sys.attr, | |
456 | &sampling_rate_gov_sys.attr, | |
457 | &up_threshold_gov_sys.attr, | |
458 | &sampling_down_factor_gov_sys.attr, | |
6c4640c3 | 459 | &ignore_nice_load_gov_sys.attr, |
4d5dcc42 VK |
460 | &powersave_bias_gov_sys.attr, |
461 | &io_is_busy_gov_sys.attr, | |
1da177e4 LT |
462 | NULL |
463 | }; | |
464 | ||
4d5dcc42 VK |
465 | static struct attribute_group od_attr_group_gov_sys = { |
466 | .attrs = dbs_attributes_gov_sys, | |
467 | .name = "ondemand", | |
468 | }; | |
469 | ||
470 | static struct attribute *dbs_attributes_gov_pol[] = { | |
471 | &sampling_rate_min_gov_pol.attr, | |
472 | &sampling_rate_gov_pol.attr, | |
473 | &up_threshold_gov_pol.attr, | |
474 | &sampling_down_factor_gov_pol.attr, | |
6c4640c3 | 475 | &ignore_nice_load_gov_pol.attr, |
4d5dcc42 VK |
476 | &powersave_bias_gov_pol.attr, |
477 | &io_is_busy_gov_pol.attr, | |
478 | NULL | |
479 | }; | |
480 | ||
481 | static struct attribute_group od_attr_group_gov_pol = { | |
482 | .attrs = dbs_attributes_gov_pol, | |
1da177e4 LT |
483 | .name = "ondemand", |
484 | }; | |
485 | ||
486 | /************************** sysfs end ************************/ | |
487 | ||
8e0484d2 | 488 | static int od_init(struct dbs_data *dbs_data, bool notify) |
4d5dcc42 VK |
489 | { |
490 | struct od_dbs_tuners *tuners; | |
491 | u64 idle_time; | |
492 | int cpu; | |
493 | ||
d5b73cd8 | 494 | tuners = kzalloc(sizeof(*tuners), GFP_KERNEL); |
4d5dcc42 VK |
495 | if (!tuners) { |
496 | pr_err("%s: kzalloc failed\n", __func__); | |
497 | return -ENOMEM; | |
498 | } | |
499 | ||
500 | cpu = get_cpu(); | |
501 | idle_time = get_cpu_idle_time_us(cpu, NULL); | |
502 | put_cpu(); | |
503 | if (idle_time != -1ULL) { | |
504 | /* Idle micro accounting is supported. Use finer thresholds */ | |
505 | tuners->up_threshold = MICRO_FREQUENCY_UP_THRESHOLD; | |
4d5dcc42 VK |
506 | /* |
507 | * In nohz/micro accounting case we set the minimum frequency | |
508 | * not depending on HZ, but fixed (very low). The deferred | |
509 | * timer might skip some samples if idle/sleeping as needed. | |
510 | */ | |
511 | dbs_data->min_sampling_rate = MICRO_FREQUENCY_MIN_SAMPLE_RATE; | |
512 | } else { | |
513 | tuners->up_threshold = DEF_FREQUENCY_UP_THRESHOLD; | |
4d5dcc42 VK |
514 | |
515 | /* For correct statistics, we need 10 ticks for each measure */ | |
516 | dbs_data->min_sampling_rate = MIN_SAMPLING_RATE_RATIO * | |
517 | jiffies_to_usecs(10); | |
518 | } | |
519 | ||
520 | tuners->sampling_down_factor = DEF_SAMPLING_DOWN_FACTOR; | |
6c4640c3 | 521 | tuners->ignore_nice_load = 0; |
c2837558 | 522 | tuners->powersave_bias = default_powersave_bias; |
4d5dcc42 VK |
523 | tuners->io_is_busy = should_io_be_busy(); |
524 | ||
525 | dbs_data->tuners = tuners; | |
4d5dcc42 VK |
526 | return 0; |
527 | } | |
528 | ||
8e0484d2 | 529 | static void od_exit(struct dbs_data *dbs_data, bool notify) |
4d5dcc42 VK |
530 | { |
531 | kfree(dbs_data->tuners); | |
532 | } | |
533 | ||
4471a34f | 534 | define_get_cpu_dbs_routines(od_cpu_dbs_info); |
6b8fcd90 | 535 | |
4471a34f | 536 | static struct od_ops od_ops = { |
4471a34f | 537 | .powersave_bias_init_cpu = ondemand_powersave_bias_init_cpu, |
fb30809e | 538 | .powersave_bias_target = generic_powersave_bias_target, |
4471a34f VK |
539 | .freq_increase = dbs_freq_increase, |
540 | }; | |
2f8a835c | 541 | |
af926185 RW |
542 | static int od_cpufreq_governor_dbs(struct cpufreq_policy *policy, |
543 | unsigned int event); | |
544 | ||
7bdad34d | 545 | static struct dbs_governor od_dbs_gov = { |
af926185 RW |
546 | .gov = { |
547 | .name = "ondemand", | |
548 | .governor = od_cpufreq_governor_dbs, | |
549 | .max_transition_latency = TRANSITION_LATENCY_LIMIT, | |
550 | .owner = THIS_MODULE, | |
551 | }, | |
4471a34f | 552 | .governor = GOV_ONDEMAND, |
4d5dcc42 VK |
553 | .attr_group_gov_sys = &od_attr_group_gov_sys, |
554 | .attr_group_gov_pol = &od_attr_group_gov_pol, | |
4471a34f VK |
555 | .get_cpu_cdbs = get_cpu_cdbs, |
556 | .get_cpu_dbs_info_s = get_cpu_dbs_info_s, | |
557 | .gov_dbs_timer = od_dbs_timer, | |
558 | .gov_check_cpu = od_check_cpu, | |
559 | .gov_ops = &od_ops, | |
4d5dcc42 VK |
560 | .init = od_init, |
561 | .exit = od_exit, | |
4471a34f | 562 | }; |
1da177e4 | 563 | |
7bdad34d | 564 | #define CPU_FREQ_GOV_ONDEMAND (&od_dbs_gov.gov) |
af926185 | 565 | |
de1df26b RW |
566 | static int od_cpufreq_governor_dbs(struct cpufreq_policy *policy, |
567 | unsigned int event) | |
568 | { | |
7bdad34d | 569 | return cpufreq_governor_dbs(policy, &od_dbs_gov, event); |
de1df26b RW |
570 | } |
571 | ||
fb30809e JS |
572 | static void od_set_powersave_bias(unsigned int powersave_bias) |
573 | { | |
574 | struct cpufreq_policy *policy; | |
575 | struct dbs_data *dbs_data; | |
576 | struct od_dbs_tuners *od_tuners; | |
577 | unsigned int cpu; | |
578 | cpumask_t done; | |
579 | ||
c2837558 | 580 | default_powersave_bias = powersave_bias; |
fb30809e JS |
581 | cpumask_clear(&done); |
582 | ||
583 | get_online_cpus(); | |
584 | for_each_online_cpu(cpu) { | |
44152cb8 VK |
585 | struct cpu_common_dbs_info *shared; |
586 | ||
fb30809e JS |
587 | if (cpumask_test_cpu(cpu, &done)) |
588 | continue; | |
589 | ||
44152cb8 VK |
590 | shared = per_cpu(od_cpu_dbs_info, cpu).cdbs.shared; |
591 | if (!shared) | |
c2837558 | 592 | continue; |
fb30809e | 593 | |
44152cb8 | 594 | policy = shared->policy; |
fb30809e | 595 | cpumask_or(&done, &done, policy->cpus); |
c2837558 | 596 | |
af926185 | 597 | if (policy->governor != CPU_FREQ_GOV_ONDEMAND) |
c2837558 JS |
598 | continue; |
599 | ||
600 | dbs_data = policy->governor_data; | |
601 | od_tuners = dbs_data->tuners; | |
602 | od_tuners->powersave_bias = default_powersave_bias; | |
fb30809e JS |
603 | } |
604 | put_online_cpus(); | |
605 | } | |
606 | ||
607 | void od_register_powersave_bias_handler(unsigned int (*f) | |
608 | (struct cpufreq_policy *, unsigned int, unsigned int), | |
609 | unsigned int powersave_bias) | |
610 | { | |
611 | od_ops.powersave_bias_target = f; | |
612 | od_set_powersave_bias(powersave_bias); | |
613 | } | |
614 | EXPORT_SYMBOL_GPL(od_register_powersave_bias_handler); | |
615 | ||
616 | void od_unregister_powersave_bias_handler(void) | |
617 | { | |
618 | od_ops.powersave_bias_target = generic_powersave_bias_target; | |
619 | od_set_powersave_bias(0); | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(od_unregister_powersave_bias_handler); | |
622 | ||
1da177e4 LT |
623 | static int __init cpufreq_gov_dbs_init(void) |
624 | { | |
af926185 | 625 | return cpufreq_register_governor(CPU_FREQ_GOV_ONDEMAND); |
1da177e4 LT |
626 | } |
627 | ||
628 | static void __exit cpufreq_gov_dbs_exit(void) | |
629 | { | |
af926185 | 630 | cpufreq_unregister_governor(CPU_FREQ_GOV_ONDEMAND); |
1da177e4 LT |
631 | } |
632 | ||
ffac80e9 VP |
633 | MODULE_AUTHOR("Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>"); |
634 | MODULE_AUTHOR("Alexey Starikovskiy <alexey.y.starikovskiy@intel.com>"); | |
635 | MODULE_DESCRIPTION("'cpufreq_ondemand' - A dynamic cpufreq governor for " | |
2b03f891 | 636 | "Low Latency Frequency Transition capable processors"); |
ffac80e9 | 637 | MODULE_LICENSE("GPL"); |
1da177e4 | 638 | |
6915719b | 639 | #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND |
de1df26b RW |
640 | struct cpufreq_governor *cpufreq_default_governor(void) |
641 | { | |
af926185 | 642 | return CPU_FREQ_GOV_ONDEMAND; |
de1df26b RW |
643 | } |
644 | ||
6915719b JW |
645 | fs_initcall(cpufreq_gov_dbs_init); |
646 | #else | |
1da177e4 | 647 | module_init(cpufreq_gov_dbs_init); |
6915719b | 648 | #endif |
1da177e4 | 649 | module_exit(cpufreq_gov_dbs_exit); |