Commit | Line | Data |
---|---|---|
93f0822d | 1 | /* |
d1b68485 | 2 | * intel_pstate.c: Native P state management for Intel processors |
93f0822d DB |
3 | * |
4 | * (C) Copyright 2012 Intel Corporation | |
5 | * Author: Dirk Brandewie <dirk.j.brandewie@intel.com> | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; version 2 | |
10 | * of the License. | |
11 | */ | |
12 | ||
13 | #include <linux/kernel.h> | |
14 | #include <linux/kernel_stat.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/ktime.h> | |
17 | #include <linux/hrtimer.h> | |
18 | #include <linux/tick.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/sched.h> | |
21 | #include <linux/list.h> | |
22 | #include <linux/cpu.h> | |
23 | #include <linux/cpufreq.h> | |
24 | #include <linux/sysfs.h> | |
25 | #include <linux/types.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/debugfs.h> | |
fbbcdc07 | 28 | #include <linux/acpi.h> |
93f0822d DB |
29 | #include <trace/events/power.h> |
30 | ||
31 | #include <asm/div64.h> | |
32 | #include <asm/msr.h> | |
33 | #include <asm/cpu_device_id.h> | |
34 | ||
35 | #define SAMPLE_COUNT 3 | |
36 | ||
61d8d2ab DB |
37 | #define BYT_RATIOS 0x66a |
38 | #define BYT_VIDS 0x66b | |
39 | #define BYT_TURBO_RATIOS 0x66c | |
21855ff5 | 40 | #define BYT_TURBO_VIDS 0x66d |
61d8d2ab | 41 | |
19e77c28 | 42 | |
f0fe3cd7 | 43 | #define FRAC_BITS 8 |
93f0822d DB |
44 | #define int_tofp(X) ((int64_t)(X) << FRAC_BITS) |
45 | #define fp_toint(X) ((X) >> FRAC_BITS) | |
f0fe3cd7 | 46 | |
93f0822d DB |
47 | |
48 | static inline int32_t mul_fp(int32_t x, int32_t y) | |
49 | { | |
50 | return ((int64_t)x * (int64_t)y) >> FRAC_BITS; | |
51 | } | |
52 | ||
53 | static inline int32_t div_fp(int32_t x, int32_t y) | |
54 | { | |
55 | return div_s64((int64_t)x << FRAC_BITS, (int64_t)y); | |
56 | } | |
57 | ||
58 | struct sample { | |
d253d2a5 | 59 | int32_t core_pct_busy; |
93f0822d DB |
60 | u64 aperf; |
61 | u64 mperf; | |
62 | int freq; | |
c4ee841f | 63 | ktime_t time; |
93f0822d DB |
64 | }; |
65 | ||
66 | struct pstate_data { | |
67 | int current_pstate; | |
68 | int min_pstate; | |
69 | int max_pstate; | |
70 | int turbo_pstate; | |
71 | }; | |
72 | ||
007bea09 | 73 | struct vid_data { |
21855ff5 DB |
74 | int min; |
75 | int max; | |
76 | int turbo; | |
007bea09 DB |
77 | int32_t ratio; |
78 | }; | |
79 | ||
93f0822d DB |
80 | struct _pid { |
81 | int setpoint; | |
82 | int32_t integral; | |
83 | int32_t p_gain; | |
84 | int32_t i_gain; | |
85 | int32_t d_gain; | |
86 | int deadband; | |
d253d2a5 | 87 | int32_t last_err; |
93f0822d DB |
88 | }; |
89 | ||
90 | struct cpudata { | |
91 | int cpu; | |
92 | ||
93 | char name[64]; | |
94 | ||
95 | struct timer_list timer; | |
96 | ||
93f0822d | 97 | struct pstate_data pstate; |
007bea09 | 98 | struct vid_data vid; |
93f0822d | 99 | struct _pid pid; |
93f0822d | 100 | |
c4ee841f | 101 | ktime_t last_sample_time; |
93f0822d DB |
102 | u64 prev_aperf; |
103 | u64 prev_mperf; | |
d37e2b76 | 104 | struct sample sample; |
93f0822d DB |
105 | }; |
106 | ||
107 | static struct cpudata **all_cpu_data; | |
108 | struct pstate_adjust_policy { | |
109 | int sample_rate_ms; | |
110 | int deadband; | |
111 | int setpoint; | |
112 | int p_gain_pct; | |
113 | int d_gain_pct; | |
114 | int i_gain_pct; | |
115 | }; | |
116 | ||
016c8150 DB |
117 | struct pstate_funcs { |
118 | int (*get_max)(void); | |
119 | int (*get_min)(void); | |
120 | int (*get_turbo)(void); | |
007bea09 DB |
121 | void (*set)(struct cpudata*, int pstate); |
122 | void (*get_vid)(struct cpudata *); | |
93f0822d DB |
123 | }; |
124 | ||
016c8150 DB |
125 | struct cpu_defaults { |
126 | struct pstate_adjust_policy pid_policy; | |
127 | struct pstate_funcs funcs; | |
93f0822d DB |
128 | }; |
129 | ||
016c8150 DB |
130 | static struct pstate_adjust_policy pid_params; |
131 | static struct pstate_funcs pstate_funcs; | |
132 | ||
93f0822d DB |
133 | struct perf_limits { |
134 | int no_turbo; | |
135 | int max_perf_pct; | |
136 | int min_perf_pct; | |
137 | int32_t max_perf; | |
138 | int32_t min_perf; | |
d8f469e9 DB |
139 | int max_policy_pct; |
140 | int max_sysfs_pct; | |
93f0822d DB |
141 | }; |
142 | ||
143 | static struct perf_limits limits = { | |
144 | .no_turbo = 0, | |
145 | .max_perf_pct = 100, | |
146 | .max_perf = int_tofp(1), | |
147 | .min_perf_pct = 0, | |
148 | .min_perf = 0, | |
d8f469e9 DB |
149 | .max_policy_pct = 100, |
150 | .max_sysfs_pct = 100, | |
93f0822d DB |
151 | }; |
152 | ||
153 | static inline void pid_reset(struct _pid *pid, int setpoint, int busy, | |
154 | int deadband, int integral) { | |
155 | pid->setpoint = setpoint; | |
156 | pid->deadband = deadband; | |
157 | pid->integral = int_tofp(integral); | |
d98d099b | 158 | pid->last_err = int_tofp(setpoint) - int_tofp(busy); |
93f0822d DB |
159 | } |
160 | ||
161 | static inline void pid_p_gain_set(struct _pid *pid, int percent) | |
162 | { | |
163 | pid->p_gain = div_fp(int_tofp(percent), int_tofp(100)); | |
164 | } | |
165 | ||
166 | static inline void pid_i_gain_set(struct _pid *pid, int percent) | |
167 | { | |
168 | pid->i_gain = div_fp(int_tofp(percent), int_tofp(100)); | |
169 | } | |
170 | ||
171 | static inline void pid_d_gain_set(struct _pid *pid, int percent) | |
172 | { | |
173 | ||
174 | pid->d_gain = div_fp(int_tofp(percent), int_tofp(100)); | |
175 | } | |
176 | ||
d253d2a5 | 177 | static signed int pid_calc(struct _pid *pid, int32_t busy) |
93f0822d | 178 | { |
d253d2a5 | 179 | signed int result; |
93f0822d DB |
180 | int32_t pterm, dterm, fp_error; |
181 | int32_t integral_limit; | |
182 | ||
d253d2a5 | 183 | fp_error = int_tofp(pid->setpoint) - busy; |
93f0822d | 184 | |
d253d2a5 | 185 | if (abs(fp_error) <= int_tofp(pid->deadband)) |
93f0822d DB |
186 | return 0; |
187 | ||
188 | pterm = mul_fp(pid->p_gain, fp_error); | |
189 | ||
190 | pid->integral += fp_error; | |
191 | ||
192 | /* limit the integral term */ | |
193 | integral_limit = int_tofp(30); | |
194 | if (pid->integral > integral_limit) | |
195 | pid->integral = integral_limit; | |
196 | if (pid->integral < -integral_limit) | |
197 | pid->integral = -integral_limit; | |
198 | ||
d253d2a5 BS |
199 | dterm = mul_fp(pid->d_gain, fp_error - pid->last_err); |
200 | pid->last_err = fp_error; | |
93f0822d DB |
201 | |
202 | result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm; | |
f0fe3cd7 DB |
203 | if (result >= 0) |
204 | result = result + (1 << (FRAC_BITS-1)); | |
205 | else | |
206 | result = result - (1 << (FRAC_BITS-1)); | |
93f0822d DB |
207 | return (signed int)fp_toint(result); |
208 | } | |
209 | ||
210 | static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu) | |
211 | { | |
016c8150 DB |
212 | pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct); |
213 | pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct); | |
214 | pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct); | |
93f0822d DB |
215 | |
216 | pid_reset(&cpu->pid, | |
016c8150 | 217 | pid_params.setpoint, |
93f0822d | 218 | 100, |
016c8150 | 219 | pid_params.deadband, |
93f0822d DB |
220 | 0); |
221 | } | |
222 | ||
93f0822d DB |
223 | static inline void intel_pstate_reset_all_pid(void) |
224 | { | |
225 | unsigned int cpu; | |
226 | for_each_online_cpu(cpu) { | |
227 | if (all_cpu_data[cpu]) | |
228 | intel_pstate_busy_pid_reset(all_cpu_data[cpu]); | |
229 | } | |
230 | } | |
231 | ||
232 | /************************** debugfs begin ************************/ | |
233 | static int pid_param_set(void *data, u64 val) | |
234 | { | |
235 | *(u32 *)data = val; | |
236 | intel_pstate_reset_all_pid(); | |
237 | return 0; | |
238 | } | |
239 | static int pid_param_get(void *data, u64 *val) | |
240 | { | |
241 | *val = *(u32 *)data; | |
242 | return 0; | |
243 | } | |
244 | DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, | |
245 | pid_param_set, "%llu\n"); | |
246 | ||
247 | struct pid_param { | |
248 | char *name; | |
249 | void *value; | |
250 | }; | |
251 | ||
252 | static struct pid_param pid_files[] = { | |
016c8150 DB |
253 | {"sample_rate_ms", &pid_params.sample_rate_ms}, |
254 | {"d_gain_pct", &pid_params.d_gain_pct}, | |
255 | {"i_gain_pct", &pid_params.i_gain_pct}, | |
256 | {"deadband", &pid_params.deadband}, | |
257 | {"setpoint", &pid_params.setpoint}, | |
258 | {"p_gain_pct", &pid_params.p_gain_pct}, | |
93f0822d DB |
259 | {NULL, NULL} |
260 | }; | |
261 | ||
262 | static struct dentry *debugfs_parent; | |
263 | static void intel_pstate_debug_expose_params(void) | |
264 | { | |
265 | int i = 0; | |
266 | ||
267 | debugfs_parent = debugfs_create_dir("pstate_snb", NULL); | |
268 | if (IS_ERR_OR_NULL(debugfs_parent)) | |
269 | return; | |
270 | while (pid_files[i].name) { | |
271 | debugfs_create_file(pid_files[i].name, 0660, | |
272 | debugfs_parent, pid_files[i].value, | |
273 | &fops_pid_param); | |
274 | i++; | |
275 | } | |
276 | } | |
277 | ||
278 | /************************** debugfs end ************************/ | |
279 | ||
280 | /************************** sysfs begin ************************/ | |
281 | #define show_one(file_name, object) \ | |
282 | static ssize_t show_##file_name \ | |
283 | (struct kobject *kobj, struct attribute *attr, char *buf) \ | |
284 | { \ | |
285 | return sprintf(buf, "%u\n", limits.object); \ | |
286 | } | |
287 | ||
288 | static ssize_t store_no_turbo(struct kobject *a, struct attribute *b, | |
289 | const char *buf, size_t count) | |
290 | { | |
291 | unsigned int input; | |
292 | int ret; | |
293 | ret = sscanf(buf, "%u", &input); | |
294 | if (ret != 1) | |
295 | return -EINVAL; | |
296 | limits.no_turbo = clamp_t(int, input, 0 , 1); | |
297 | ||
298 | return count; | |
299 | } | |
300 | ||
301 | static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b, | |
302 | const char *buf, size_t count) | |
303 | { | |
304 | unsigned int input; | |
305 | int ret; | |
306 | ret = sscanf(buf, "%u", &input); | |
307 | if (ret != 1) | |
308 | return -EINVAL; | |
309 | ||
d8f469e9 DB |
310 | limits.max_sysfs_pct = clamp_t(int, input, 0 , 100); |
311 | limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct); | |
93f0822d DB |
312 | limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100)); |
313 | return count; | |
314 | } | |
315 | ||
316 | static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b, | |
317 | const char *buf, size_t count) | |
318 | { | |
319 | unsigned int input; | |
320 | int ret; | |
321 | ret = sscanf(buf, "%u", &input); | |
322 | if (ret != 1) | |
323 | return -EINVAL; | |
324 | limits.min_perf_pct = clamp_t(int, input, 0 , 100); | |
325 | limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100)); | |
326 | ||
327 | return count; | |
328 | } | |
329 | ||
330 | show_one(no_turbo, no_turbo); | |
331 | show_one(max_perf_pct, max_perf_pct); | |
332 | show_one(min_perf_pct, min_perf_pct); | |
333 | ||
334 | define_one_global_rw(no_turbo); | |
335 | define_one_global_rw(max_perf_pct); | |
336 | define_one_global_rw(min_perf_pct); | |
337 | ||
338 | static struct attribute *intel_pstate_attributes[] = { | |
339 | &no_turbo.attr, | |
340 | &max_perf_pct.attr, | |
341 | &min_perf_pct.attr, | |
342 | NULL | |
343 | }; | |
344 | ||
345 | static struct attribute_group intel_pstate_attr_group = { | |
346 | .attrs = intel_pstate_attributes, | |
347 | }; | |
348 | static struct kobject *intel_pstate_kobject; | |
349 | ||
350 | static void intel_pstate_sysfs_expose_params(void) | |
351 | { | |
352 | int rc; | |
353 | ||
354 | intel_pstate_kobject = kobject_create_and_add("intel_pstate", | |
355 | &cpu_subsys.dev_root->kobj); | |
356 | BUG_ON(!intel_pstate_kobject); | |
357 | rc = sysfs_create_group(intel_pstate_kobject, | |
358 | &intel_pstate_attr_group); | |
359 | BUG_ON(rc); | |
360 | } | |
361 | ||
362 | /************************** sysfs end ************************/ | |
19e77c28 DB |
363 | static int byt_get_min_pstate(void) |
364 | { | |
365 | u64 value; | |
366 | rdmsrl(BYT_RATIOS, value); | |
21855ff5 | 367 | return (value >> 8) & 0x3F; |
19e77c28 DB |
368 | } |
369 | ||
370 | static int byt_get_max_pstate(void) | |
371 | { | |
372 | u64 value; | |
373 | rdmsrl(BYT_RATIOS, value); | |
21855ff5 | 374 | return (value >> 16) & 0x3F; |
19e77c28 | 375 | } |
93f0822d | 376 | |
61d8d2ab DB |
377 | static int byt_get_turbo_pstate(void) |
378 | { | |
379 | u64 value; | |
380 | rdmsrl(BYT_TURBO_RATIOS, value); | |
381 | return value & 0x3F; | |
382 | } | |
383 | ||
007bea09 DB |
384 | static void byt_set_pstate(struct cpudata *cpudata, int pstate) |
385 | { | |
386 | u64 val; | |
387 | int32_t vid_fp; | |
388 | u32 vid; | |
389 | ||
390 | val = pstate << 8; | |
391 | if (limits.no_turbo) | |
392 | val |= (u64)1 << 32; | |
393 | ||
394 | vid_fp = cpudata->vid.min + mul_fp( | |
395 | int_tofp(pstate - cpudata->pstate.min_pstate), | |
396 | cpudata->vid.ratio); | |
397 | ||
398 | vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max); | |
399 | vid = fp_toint(vid_fp); | |
400 | ||
21855ff5 DB |
401 | if (pstate > cpudata->pstate.max_pstate) |
402 | vid = cpudata->vid.turbo; | |
403 | ||
007bea09 DB |
404 | val |= vid; |
405 | ||
406 | wrmsrl(MSR_IA32_PERF_CTL, val); | |
407 | } | |
408 | ||
409 | static void byt_get_vid(struct cpudata *cpudata) | |
410 | { | |
411 | u64 value; | |
412 | ||
21855ff5 | 413 | |
007bea09 | 414 | rdmsrl(BYT_VIDS, value); |
21855ff5 DB |
415 | cpudata->vid.min = int_tofp((value >> 8) & 0x3f); |
416 | cpudata->vid.max = int_tofp((value >> 16) & 0x3f); | |
007bea09 DB |
417 | cpudata->vid.ratio = div_fp( |
418 | cpudata->vid.max - cpudata->vid.min, | |
419 | int_tofp(cpudata->pstate.max_pstate - | |
420 | cpudata->pstate.min_pstate)); | |
21855ff5 DB |
421 | |
422 | rdmsrl(BYT_TURBO_VIDS, value); | |
423 | cpudata->vid.turbo = value & 0x7f; | |
007bea09 DB |
424 | } |
425 | ||
426 | ||
016c8150 | 427 | static int core_get_min_pstate(void) |
93f0822d DB |
428 | { |
429 | u64 value; | |
05e99c8c | 430 | rdmsrl(MSR_PLATFORM_INFO, value); |
93f0822d DB |
431 | return (value >> 40) & 0xFF; |
432 | } | |
433 | ||
016c8150 | 434 | static int core_get_max_pstate(void) |
93f0822d DB |
435 | { |
436 | u64 value; | |
05e99c8c | 437 | rdmsrl(MSR_PLATFORM_INFO, value); |
93f0822d DB |
438 | return (value >> 8) & 0xFF; |
439 | } | |
440 | ||
016c8150 | 441 | static int core_get_turbo_pstate(void) |
93f0822d DB |
442 | { |
443 | u64 value; | |
444 | int nont, ret; | |
05e99c8c | 445 | rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value); |
016c8150 | 446 | nont = core_get_max_pstate(); |
93f0822d DB |
447 | ret = ((value) & 255); |
448 | if (ret <= nont) | |
449 | ret = nont; | |
450 | return ret; | |
451 | } | |
452 | ||
007bea09 | 453 | static void core_set_pstate(struct cpudata *cpudata, int pstate) |
016c8150 DB |
454 | { |
455 | u64 val; | |
456 | ||
457 | val = pstate << 8; | |
458 | if (limits.no_turbo) | |
459 | val |= (u64)1 << 32; | |
460 | ||
bb18008f | 461 | wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val); |
016c8150 DB |
462 | } |
463 | ||
464 | static struct cpu_defaults core_params = { | |
465 | .pid_policy = { | |
466 | .sample_rate_ms = 10, | |
467 | .deadband = 0, | |
468 | .setpoint = 97, | |
469 | .p_gain_pct = 20, | |
470 | .d_gain_pct = 0, | |
471 | .i_gain_pct = 0, | |
472 | }, | |
473 | .funcs = { | |
474 | .get_max = core_get_max_pstate, | |
475 | .get_min = core_get_min_pstate, | |
476 | .get_turbo = core_get_turbo_pstate, | |
477 | .set = core_set_pstate, | |
478 | }, | |
479 | }; | |
480 | ||
19e77c28 DB |
481 | static struct cpu_defaults byt_params = { |
482 | .pid_policy = { | |
483 | .sample_rate_ms = 10, | |
484 | .deadband = 0, | |
485 | .setpoint = 97, | |
486 | .p_gain_pct = 14, | |
487 | .d_gain_pct = 0, | |
488 | .i_gain_pct = 4, | |
489 | }, | |
490 | .funcs = { | |
491 | .get_max = byt_get_max_pstate, | |
492 | .get_min = byt_get_min_pstate, | |
61d8d2ab | 493 | .get_turbo = byt_get_turbo_pstate, |
007bea09 DB |
494 | .set = byt_set_pstate, |
495 | .get_vid = byt_get_vid, | |
19e77c28 DB |
496 | }, |
497 | }; | |
498 | ||
499 | ||
93f0822d DB |
500 | static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max) |
501 | { | |
502 | int max_perf = cpu->pstate.turbo_pstate; | |
7244cb62 | 503 | int max_perf_adj; |
93f0822d DB |
504 | int min_perf; |
505 | if (limits.no_turbo) | |
506 | max_perf = cpu->pstate.max_pstate; | |
507 | ||
7244cb62 DB |
508 | max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf)); |
509 | *max = clamp_t(int, max_perf_adj, | |
93f0822d DB |
510 | cpu->pstate.min_pstate, cpu->pstate.turbo_pstate); |
511 | ||
512 | min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf)); | |
513 | *min = clamp_t(int, min_perf, | |
514 | cpu->pstate.min_pstate, max_perf); | |
515 | } | |
516 | ||
517 | static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate) | |
518 | { | |
519 | int max_perf, min_perf; | |
520 | ||
521 | intel_pstate_get_min_max(cpu, &min_perf, &max_perf); | |
522 | ||
523 | pstate = clamp_t(int, pstate, min_perf, max_perf); | |
524 | ||
525 | if (pstate == cpu->pstate.current_pstate) | |
526 | return; | |
527 | ||
93f0822d | 528 | trace_cpu_frequency(pstate * 100000, cpu->cpu); |
35363e94 | 529 | |
93f0822d | 530 | cpu->pstate.current_pstate = pstate; |
93f0822d | 531 | |
007bea09 | 532 | pstate_funcs.set(cpu, pstate); |
93f0822d DB |
533 | } |
534 | ||
535 | static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps) | |
536 | { | |
537 | int target; | |
538 | target = cpu->pstate.current_pstate + steps; | |
539 | ||
540 | intel_pstate_set_pstate(cpu, target); | |
541 | } | |
542 | ||
543 | static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps) | |
544 | { | |
545 | int target; | |
546 | target = cpu->pstate.current_pstate - steps; | |
547 | intel_pstate_set_pstate(cpu, target); | |
548 | } | |
549 | ||
550 | static void intel_pstate_get_cpu_pstates(struct cpudata *cpu) | |
551 | { | |
552 | sprintf(cpu->name, "Intel 2nd generation core"); | |
553 | ||
016c8150 DB |
554 | cpu->pstate.min_pstate = pstate_funcs.get_min(); |
555 | cpu->pstate.max_pstate = pstate_funcs.get_max(); | |
556 | cpu->pstate.turbo_pstate = pstate_funcs.get_turbo(); | |
93f0822d | 557 | |
007bea09 DB |
558 | if (pstate_funcs.get_vid) |
559 | pstate_funcs.get_vid(cpu); | |
d40a63c4 | 560 | intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate); |
93f0822d DB |
561 | } |
562 | ||
563 | static inline void intel_pstate_calc_busy(struct cpudata *cpu, | |
564 | struct sample *sample) | |
565 | { | |
e66c1768 | 566 | int32_t core_pct; |
93f0822d | 567 | |
e66c1768 DB |
568 | core_pct = div_fp(int_tofp((sample->aperf)), |
569 | int_tofp((sample->mperf))); | |
570 | core_pct = mul_fp(core_pct, int_tofp(100)); | |
e66c1768 | 571 | |
fcb6a15c | 572 | sample->freq = fp_toint( |
e66c1768 | 573 | mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct)); |
fcb6a15c | 574 | |
adacdf3f | 575 | sample->core_pct_busy = core_pct; |
93f0822d DB |
576 | } |
577 | ||
578 | static inline void intel_pstate_sample(struct cpudata *cpu) | |
579 | { | |
93f0822d DB |
580 | u64 aperf, mperf; |
581 | ||
93f0822d DB |
582 | rdmsrl(MSR_IA32_APERF, aperf); |
583 | rdmsrl(MSR_IA32_MPERF, mperf); | |
b69880f9 | 584 | |
e66c1768 DB |
585 | aperf = aperf >> FRAC_BITS; |
586 | mperf = mperf >> FRAC_BITS; | |
e66c1768 | 587 | |
c4ee841f DB |
588 | cpu->last_sample_time = cpu->sample.time; |
589 | cpu->sample.time = ktime_get(); | |
d37e2b76 DB |
590 | cpu->sample.aperf = aperf; |
591 | cpu->sample.mperf = mperf; | |
d37e2b76 DB |
592 | cpu->sample.aperf -= cpu->prev_aperf; |
593 | cpu->sample.mperf -= cpu->prev_mperf; | |
1abc4b20 | 594 | |
d37e2b76 | 595 | intel_pstate_calc_busy(cpu, &cpu->sample); |
93f0822d | 596 | |
93f0822d DB |
597 | cpu->prev_aperf = aperf; |
598 | cpu->prev_mperf = mperf; | |
599 | } | |
600 | ||
601 | static inline void intel_pstate_set_sample_time(struct cpudata *cpu) | |
602 | { | |
603 | int sample_time, delay; | |
604 | ||
016c8150 | 605 | sample_time = pid_params.sample_rate_ms; |
93f0822d | 606 | delay = msecs_to_jiffies(sample_time); |
93f0822d DB |
607 | mod_timer_pinned(&cpu->timer, jiffies + delay); |
608 | } | |
609 | ||
d253d2a5 | 610 | static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu) |
93f0822d | 611 | { |
c4ee841f DB |
612 | int32_t core_busy, max_pstate, current_pstate, sample_ratio; |
613 | u32 duration_us; | |
614 | u32 sample_time; | |
93f0822d | 615 | |
d37e2b76 | 616 | core_busy = cpu->sample.core_pct_busy; |
2134ed4d | 617 | max_pstate = int_tofp(cpu->pstate.max_pstate); |
93f0822d | 618 | current_pstate = int_tofp(cpu->pstate.current_pstate); |
e66c1768 | 619 | core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate)); |
c4ee841f DB |
620 | |
621 | sample_time = (pid_params.sample_rate_ms * USEC_PER_MSEC); | |
622 | duration_us = (u32) ktime_us_delta(cpu->sample.time, | |
623 | cpu->last_sample_time); | |
624 | if (duration_us > sample_time * 3) { | |
625 | sample_ratio = div_fp(int_tofp(sample_time), | |
626 | int_tofp(duration_us)); | |
627 | core_busy = mul_fp(core_busy, sample_ratio); | |
628 | } | |
629 | ||
f0fe3cd7 | 630 | return core_busy; |
93f0822d DB |
631 | } |
632 | ||
633 | static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu) | |
634 | { | |
d253d2a5 | 635 | int32_t busy_scaled; |
93f0822d DB |
636 | struct _pid *pid; |
637 | signed int ctl = 0; | |
638 | int steps; | |
639 | ||
640 | pid = &cpu->pid; | |
641 | busy_scaled = intel_pstate_get_scaled_busy(cpu); | |
642 | ||
643 | ctl = pid_calc(pid, busy_scaled); | |
644 | ||
645 | steps = abs(ctl); | |
b69880f9 | 646 | |
93f0822d DB |
647 | if (ctl < 0) |
648 | intel_pstate_pstate_increase(cpu, steps); | |
649 | else | |
650 | intel_pstate_pstate_decrease(cpu, steps); | |
651 | } | |
652 | ||
93f0822d DB |
653 | static void intel_pstate_timer_func(unsigned long __data) |
654 | { | |
655 | struct cpudata *cpu = (struct cpudata *) __data; | |
b69880f9 | 656 | struct sample *sample; |
93f0822d DB |
657 | |
658 | intel_pstate_sample(cpu); | |
b69880f9 | 659 | |
d37e2b76 | 660 | sample = &cpu->sample; |
b69880f9 | 661 | |
ca182aee | 662 | intel_pstate_adjust_busy_pstate(cpu); |
b69880f9 DB |
663 | |
664 | trace_pstate_sample(fp_toint(sample->core_pct_busy), | |
665 | fp_toint(intel_pstate_get_scaled_busy(cpu)), | |
666 | cpu->pstate.current_pstate, | |
667 | sample->mperf, | |
668 | sample->aperf, | |
b69880f9 DB |
669 | sample->freq); |
670 | ||
93f0822d DB |
671 | intel_pstate_set_sample_time(cpu); |
672 | } | |
673 | ||
674 | #define ICPU(model, policy) \ | |
6cbd7ee1 DB |
675 | { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\ |
676 | (unsigned long)&policy } | |
93f0822d DB |
677 | |
678 | static const struct x86_cpu_id intel_pstate_cpu_ids[] = { | |
016c8150 DB |
679 | ICPU(0x2a, core_params), |
680 | ICPU(0x2d, core_params), | |
19e77c28 | 681 | ICPU(0x37, byt_params), |
016c8150 DB |
682 | ICPU(0x3a, core_params), |
683 | ICPU(0x3c, core_params), | |
684 | ICPU(0x3e, core_params), | |
685 | ICPU(0x3f, core_params), | |
686 | ICPU(0x45, core_params), | |
687 | ICPU(0x46, core_params), | |
93f0822d DB |
688 | {} |
689 | }; | |
690 | MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids); | |
691 | ||
692 | static int intel_pstate_init_cpu(unsigned int cpunum) | |
693 | { | |
694 | ||
695 | const struct x86_cpu_id *id; | |
696 | struct cpudata *cpu; | |
697 | ||
698 | id = x86_match_cpu(intel_pstate_cpu_ids); | |
699 | if (!id) | |
700 | return -ENODEV; | |
701 | ||
702 | all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL); | |
703 | if (!all_cpu_data[cpunum]) | |
704 | return -ENOMEM; | |
705 | ||
706 | cpu = all_cpu_data[cpunum]; | |
707 | ||
708 | intel_pstate_get_cpu_pstates(cpu); | |
709 | ||
710 | cpu->cpu = cpunum; | |
016c8150 | 711 | |
93f0822d DB |
712 | init_timer_deferrable(&cpu->timer); |
713 | cpu->timer.function = intel_pstate_timer_func; | |
714 | cpu->timer.data = | |
715 | (unsigned long)cpu; | |
716 | cpu->timer.expires = jiffies + HZ/100; | |
717 | intel_pstate_busy_pid_reset(cpu); | |
93f0822d | 718 | intel_pstate_sample(cpu); |
93f0822d DB |
719 | |
720 | add_timer_on(&cpu->timer, cpunum); | |
721 | ||
722 | pr_info("Intel pstate controlling: cpu %d\n", cpunum); | |
723 | ||
724 | return 0; | |
725 | } | |
726 | ||
727 | static unsigned int intel_pstate_get(unsigned int cpu_num) | |
728 | { | |
729 | struct sample *sample; | |
730 | struct cpudata *cpu; | |
731 | ||
732 | cpu = all_cpu_data[cpu_num]; | |
733 | if (!cpu) | |
734 | return 0; | |
d37e2b76 | 735 | sample = &cpu->sample; |
93f0822d DB |
736 | return sample->freq; |
737 | } | |
738 | ||
739 | static int intel_pstate_set_policy(struct cpufreq_policy *policy) | |
740 | { | |
741 | struct cpudata *cpu; | |
93f0822d DB |
742 | |
743 | cpu = all_cpu_data[policy->cpu]; | |
744 | ||
d3929b83 DB |
745 | if (!policy->cpuinfo.max_freq) |
746 | return -ENODEV; | |
747 | ||
93f0822d DB |
748 | if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) { |
749 | limits.min_perf_pct = 100; | |
750 | limits.min_perf = int_tofp(1); | |
751 | limits.max_perf_pct = 100; | |
752 | limits.max_perf = int_tofp(1); | |
753 | limits.no_turbo = 0; | |
d1b68485 | 754 | return 0; |
93f0822d | 755 | } |
d1b68485 SP |
756 | limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq; |
757 | limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100); | |
758 | limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100)); | |
759 | ||
d8f469e9 DB |
760 | limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq; |
761 | limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100); | |
762 | limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct); | |
d1b68485 | 763 | limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100)); |
93f0822d DB |
764 | |
765 | return 0; | |
766 | } | |
767 | ||
768 | static int intel_pstate_verify_policy(struct cpufreq_policy *policy) | |
769 | { | |
be49e346 | 770 | cpufreq_verify_within_cpu_limits(policy); |
93f0822d DB |
771 | |
772 | if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) && | |
773 | (policy->policy != CPUFREQ_POLICY_PERFORMANCE)) | |
774 | return -EINVAL; | |
775 | ||
776 | return 0; | |
777 | } | |
778 | ||
bb18008f | 779 | static void intel_pstate_stop_cpu(struct cpufreq_policy *policy) |
93f0822d | 780 | { |
bb18008f DB |
781 | int cpu_num = policy->cpu; |
782 | struct cpudata *cpu = all_cpu_data[cpu_num]; | |
93f0822d | 783 | |
bb18008f DB |
784 | pr_info("intel_pstate CPU %d exiting\n", cpu_num); |
785 | ||
c2294a2f | 786 | del_timer_sync(&all_cpu_data[cpu_num]->timer); |
bb18008f DB |
787 | intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate); |
788 | kfree(all_cpu_data[cpu_num]); | |
789 | all_cpu_data[cpu_num] = NULL; | |
93f0822d DB |
790 | } |
791 | ||
2760984f | 792 | static int intel_pstate_cpu_init(struct cpufreq_policy *policy) |
93f0822d | 793 | { |
93f0822d | 794 | struct cpudata *cpu; |
52e0a509 | 795 | int rc; |
93f0822d DB |
796 | |
797 | rc = intel_pstate_init_cpu(policy->cpu); | |
798 | if (rc) | |
799 | return rc; | |
800 | ||
801 | cpu = all_cpu_data[policy->cpu]; | |
802 | ||
803 | if (!limits.no_turbo && | |
804 | limits.min_perf_pct == 100 && limits.max_perf_pct == 100) | |
805 | policy->policy = CPUFREQ_POLICY_PERFORMANCE; | |
806 | else | |
807 | policy->policy = CPUFREQ_POLICY_POWERSAVE; | |
808 | ||
52e0a509 DB |
809 | policy->min = cpu->pstate.min_pstate * 100000; |
810 | policy->max = cpu->pstate.turbo_pstate * 100000; | |
93f0822d DB |
811 | |
812 | /* cpuinfo and default policy values */ | |
813 | policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000; | |
814 | policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000; | |
815 | policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL; | |
816 | cpumask_set_cpu(policy->cpu, policy->cpus); | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
821 | static struct cpufreq_driver intel_pstate_driver = { | |
822 | .flags = CPUFREQ_CONST_LOOPS, | |
823 | .verify = intel_pstate_verify_policy, | |
824 | .setpolicy = intel_pstate_set_policy, | |
825 | .get = intel_pstate_get, | |
826 | .init = intel_pstate_cpu_init, | |
bb18008f | 827 | .stop_cpu = intel_pstate_stop_cpu, |
93f0822d | 828 | .name = "intel_pstate", |
93f0822d DB |
829 | }; |
830 | ||
6be26498 DB |
831 | static int __initdata no_load; |
832 | ||
b563b4e3 DB |
833 | static int intel_pstate_msrs_not_valid(void) |
834 | { | |
835 | /* Check that all the msr's we are using are valid. */ | |
836 | u64 aperf, mperf, tmp; | |
837 | ||
838 | rdmsrl(MSR_IA32_APERF, aperf); | |
839 | rdmsrl(MSR_IA32_MPERF, mperf); | |
840 | ||
016c8150 DB |
841 | if (!pstate_funcs.get_max() || |
842 | !pstate_funcs.get_min() || | |
843 | !pstate_funcs.get_turbo()) | |
b563b4e3 DB |
844 | return -ENODEV; |
845 | ||
846 | rdmsrl(MSR_IA32_APERF, tmp); | |
847 | if (!(tmp - aperf)) | |
848 | return -ENODEV; | |
849 | ||
850 | rdmsrl(MSR_IA32_MPERF, tmp); | |
851 | if (!(tmp - mperf)) | |
852 | return -ENODEV; | |
853 | ||
854 | return 0; | |
855 | } | |
016c8150 | 856 | |
e0a261a2 | 857 | static void copy_pid_params(struct pstate_adjust_policy *policy) |
016c8150 DB |
858 | { |
859 | pid_params.sample_rate_ms = policy->sample_rate_ms; | |
860 | pid_params.p_gain_pct = policy->p_gain_pct; | |
861 | pid_params.i_gain_pct = policy->i_gain_pct; | |
862 | pid_params.d_gain_pct = policy->d_gain_pct; | |
863 | pid_params.deadband = policy->deadband; | |
864 | pid_params.setpoint = policy->setpoint; | |
865 | } | |
866 | ||
e0a261a2 | 867 | static void copy_cpu_funcs(struct pstate_funcs *funcs) |
016c8150 DB |
868 | { |
869 | pstate_funcs.get_max = funcs->get_max; | |
870 | pstate_funcs.get_min = funcs->get_min; | |
871 | pstate_funcs.get_turbo = funcs->get_turbo; | |
872 | pstate_funcs.set = funcs->set; | |
007bea09 | 873 | pstate_funcs.get_vid = funcs->get_vid; |
016c8150 DB |
874 | } |
875 | ||
fbbcdc07 AH |
876 | #if IS_ENABLED(CONFIG_ACPI) |
877 | #include <acpi/processor.h> | |
878 | ||
879 | static bool intel_pstate_no_acpi_pss(void) | |
880 | { | |
881 | int i; | |
882 | ||
883 | for_each_possible_cpu(i) { | |
884 | acpi_status status; | |
885 | union acpi_object *pss; | |
886 | struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; | |
887 | struct acpi_processor *pr = per_cpu(processors, i); | |
888 | ||
889 | if (!pr) | |
890 | continue; | |
891 | ||
892 | status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer); | |
893 | if (ACPI_FAILURE(status)) | |
894 | continue; | |
895 | ||
896 | pss = buffer.pointer; | |
897 | if (pss && pss->type == ACPI_TYPE_PACKAGE) { | |
898 | kfree(pss); | |
899 | return false; | |
900 | } | |
901 | ||
902 | kfree(pss); | |
903 | } | |
904 | ||
905 | return true; | |
906 | } | |
907 | ||
908 | struct hw_vendor_info { | |
909 | u16 valid; | |
910 | char oem_id[ACPI_OEM_ID_SIZE]; | |
911 | char oem_table_id[ACPI_OEM_TABLE_ID_SIZE]; | |
912 | }; | |
913 | ||
914 | /* Hardware vendor-specific info that has its own power management modes */ | |
915 | static struct hw_vendor_info vendor_info[] = { | |
916 | {1, "HP ", "ProLiant"}, | |
917 | {0, "", ""}, | |
918 | }; | |
919 | ||
920 | static bool intel_pstate_platform_pwr_mgmt_exists(void) | |
921 | { | |
922 | struct acpi_table_header hdr; | |
923 | struct hw_vendor_info *v_info; | |
924 | ||
925 | if (acpi_disabled | |
926 | || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr))) | |
927 | return false; | |
928 | ||
929 | for (v_info = vendor_info; v_info->valid; v_info++) { | |
930 | if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) | |
931 | && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) | |
932 | && intel_pstate_no_acpi_pss()) | |
933 | return true; | |
934 | } | |
935 | ||
936 | return false; | |
937 | } | |
938 | #else /* CONFIG_ACPI not enabled */ | |
939 | static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; } | |
940 | #endif /* CONFIG_ACPI */ | |
941 | ||
93f0822d DB |
942 | static int __init intel_pstate_init(void) |
943 | { | |
907cc908 | 944 | int cpu, rc = 0; |
93f0822d | 945 | const struct x86_cpu_id *id; |
016c8150 | 946 | struct cpu_defaults *cpu_info; |
93f0822d | 947 | |
6be26498 DB |
948 | if (no_load) |
949 | return -ENODEV; | |
950 | ||
93f0822d DB |
951 | id = x86_match_cpu(intel_pstate_cpu_ids); |
952 | if (!id) | |
953 | return -ENODEV; | |
954 | ||
fbbcdc07 AH |
955 | /* |
956 | * The Intel pstate driver will be ignored if the platform | |
957 | * firmware has its own power management modes. | |
958 | */ | |
959 | if (intel_pstate_platform_pwr_mgmt_exists()) | |
960 | return -ENODEV; | |
961 | ||
016c8150 DB |
962 | cpu_info = (struct cpu_defaults *)id->driver_data; |
963 | ||
964 | copy_pid_params(&cpu_info->pid_policy); | |
965 | copy_cpu_funcs(&cpu_info->funcs); | |
966 | ||
b563b4e3 DB |
967 | if (intel_pstate_msrs_not_valid()) |
968 | return -ENODEV; | |
969 | ||
93f0822d DB |
970 | pr_info("Intel P-state driver initializing.\n"); |
971 | ||
b57ffac5 | 972 | all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus()); |
93f0822d DB |
973 | if (!all_cpu_data) |
974 | return -ENOMEM; | |
93f0822d DB |
975 | |
976 | rc = cpufreq_register_driver(&intel_pstate_driver); | |
977 | if (rc) | |
978 | goto out; | |
979 | ||
980 | intel_pstate_debug_expose_params(); | |
981 | intel_pstate_sysfs_expose_params(); | |
b69880f9 | 982 | |
93f0822d DB |
983 | return rc; |
984 | out: | |
907cc908 DB |
985 | get_online_cpus(); |
986 | for_each_online_cpu(cpu) { | |
987 | if (all_cpu_data[cpu]) { | |
988 | del_timer_sync(&all_cpu_data[cpu]->timer); | |
989 | kfree(all_cpu_data[cpu]); | |
990 | } | |
991 | } | |
992 | ||
993 | put_online_cpus(); | |
994 | vfree(all_cpu_data); | |
93f0822d DB |
995 | return -ENODEV; |
996 | } | |
997 | device_initcall(intel_pstate_init); | |
998 | ||
6be26498 DB |
999 | static int __init intel_pstate_setup(char *str) |
1000 | { | |
1001 | if (!str) | |
1002 | return -EINVAL; | |
1003 | ||
1004 | if (!strcmp(str, "disable")) | |
1005 | no_load = 1; | |
1006 | return 0; | |
1007 | } | |
1008 | early_param("intel_pstate", intel_pstate_setup); | |
1009 | ||
93f0822d DB |
1010 | MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>"); |
1011 | MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors"); | |
1012 | MODULE_LICENSE("GPL"); |