]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/cpufreq/intel_pstate.c
cpufreq: SPEAr: clk_round_rate() can return a zero upon error
[mirror_ubuntu-artful-kernel.git] / drivers / cpufreq / intel_pstate.c
CommitLineData
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
19e77c28
DB
37#define BYT_RATIOS 0x66a
38
93f0822d
DB
39#define FRAC_BITS 8
40#define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
41#define fp_toint(X) ((X) >> FRAC_BITS)
42
43static inline int32_t mul_fp(int32_t x, int32_t y)
44{
45 return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
46}
47
48static inline int32_t div_fp(int32_t x, int32_t y)
49{
50 return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
51}
52
53struct sample {
d253d2a5 54 int32_t core_pct_busy;
93f0822d
DB
55 u64 aperf;
56 u64 mperf;
57 int freq;
58};
59
60struct pstate_data {
61 int current_pstate;
62 int min_pstate;
63 int max_pstate;
64 int turbo_pstate;
65};
66
67struct _pid {
68 int setpoint;
69 int32_t integral;
70 int32_t p_gain;
71 int32_t i_gain;
72 int32_t d_gain;
73 int deadband;
d253d2a5 74 int32_t last_err;
93f0822d
DB
75};
76
77struct cpudata {
78 int cpu;
79
80 char name[64];
81
82 struct timer_list timer;
83
93f0822d
DB
84 struct pstate_data pstate;
85 struct _pid pid;
93f0822d
DB
86
87 int min_pstate_count;
93f0822d 88
93f0822d
DB
89 u64 prev_aperf;
90 u64 prev_mperf;
91 int sample_ptr;
92 struct sample samples[SAMPLE_COUNT];
93};
94
95static struct cpudata **all_cpu_data;
96struct pstate_adjust_policy {
97 int sample_rate_ms;
98 int deadband;
99 int setpoint;
100 int p_gain_pct;
101 int d_gain_pct;
102 int i_gain_pct;
103};
104
016c8150
DB
105struct pstate_funcs {
106 int (*get_max)(void);
107 int (*get_min)(void);
108 int (*get_turbo)(void);
109 void (*set)(int pstate);
93f0822d
DB
110};
111
016c8150
DB
112struct cpu_defaults {
113 struct pstate_adjust_policy pid_policy;
114 struct pstate_funcs funcs;
93f0822d
DB
115};
116
016c8150
DB
117static struct pstate_adjust_policy pid_params;
118static struct pstate_funcs pstate_funcs;
119
93f0822d
DB
120struct perf_limits {
121 int no_turbo;
122 int max_perf_pct;
123 int min_perf_pct;
124 int32_t max_perf;
125 int32_t min_perf;
d8f469e9
DB
126 int max_policy_pct;
127 int max_sysfs_pct;
93f0822d
DB
128};
129
130static struct perf_limits limits = {
131 .no_turbo = 0,
132 .max_perf_pct = 100,
133 .max_perf = int_tofp(1),
134 .min_perf_pct = 0,
135 .min_perf = 0,
d8f469e9
DB
136 .max_policy_pct = 100,
137 .max_sysfs_pct = 100,
93f0822d
DB
138};
139
140static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
141 int deadband, int integral) {
142 pid->setpoint = setpoint;
143 pid->deadband = deadband;
144 pid->integral = int_tofp(integral);
145 pid->last_err = setpoint - busy;
146}
147
148static inline void pid_p_gain_set(struct _pid *pid, int percent)
149{
150 pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
151}
152
153static inline void pid_i_gain_set(struct _pid *pid, int percent)
154{
155 pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
156}
157
158static inline void pid_d_gain_set(struct _pid *pid, int percent)
159{
160
161 pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
162}
163
d253d2a5 164static signed int pid_calc(struct _pid *pid, int32_t busy)
93f0822d 165{
d253d2a5 166 signed int result;
93f0822d
DB
167 int32_t pterm, dterm, fp_error;
168 int32_t integral_limit;
169
d253d2a5 170 fp_error = int_tofp(pid->setpoint) - busy;
93f0822d 171
d253d2a5 172 if (abs(fp_error) <= int_tofp(pid->deadband))
93f0822d
DB
173 return 0;
174
175 pterm = mul_fp(pid->p_gain, fp_error);
176
177 pid->integral += fp_error;
178
179 /* limit the integral term */
180 integral_limit = int_tofp(30);
181 if (pid->integral > integral_limit)
182 pid->integral = integral_limit;
183 if (pid->integral < -integral_limit)
184 pid->integral = -integral_limit;
185
d253d2a5
BS
186 dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
187 pid->last_err = fp_error;
93f0822d
DB
188
189 result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
190
191 return (signed int)fp_toint(result);
192}
193
194static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
195{
016c8150
DB
196 pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
197 pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
198 pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
93f0822d
DB
199
200 pid_reset(&cpu->pid,
016c8150 201 pid_params.setpoint,
93f0822d 202 100,
016c8150 203 pid_params.deadband,
93f0822d
DB
204 0);
205}
206
93f0822d
DB
207static inline void intel_pstate_reset_all_pid(void)
208{
209 unsigned int cpu;
210 for_each_online_cpu(cpu) {
211 if (all_cpu_data[cpu])
212 intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
213 }
214}
215
216/************************** debugfs begin ************************/
217static int pid_param_set(void *data, u64 val)
218{
219 *(u32 *)data = val;
220 intel_pstate_reset_all_pid();
221 return 0;
222}
223static int pid_param_get(void *data, u64 *val)
224{
225 *val = *(u32 *)data;
226 return 0;
227}
228DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
229 pid_param_set, "%llu\n");
230
231struct pid_param {
232 char *name;
233 void *value;
234};
235
236static struct pid_param pid_files[] = {
016c8150
DB
237 {"sample_rate_ms", &pid_params.sample_rate_ms},
238 {"d_gain_pct", &pid_params.d_gain_pct},
239 {"i_gain_pct", &pid_params.i_gain_pct},
240 {"deadband", &pid_params.deadband},
241 {"setpoint", &pid_params.setpoint},
242 {"p_gain_pct", &pid_params.p_gain_pct},
93f0822d
DB
243 {NULL, NULL}
244};
245
246static struct dentry *debugfs_parent;
247static void intel_pstate_debug_expose_params(void)
248{
249 int i = 0;
250
251 debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
252 if (IS_ERR_OR_NULL(debugfs_parent))
253 return;
254 while (pid_files[i].name) {
255 debugfs_create_file(pid_files[i].name, 0660,
256 debugfs_parent, pid_files[i].value,
257 &fops_pid_param);
258 i++;
259 }
260}
261
262/************************** debugfs end ************************/
263
264/************************** sysfs begin ************************/
265#define show_one(file_name, object) \
266 static ssize_t show_##file_name \
267 (struct kobject *kobj, struct attribute *attr, char *buf) \
268 { \
269 return sprintf(buf, "%u\n", limits.object); \
270 }
271
272static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
273 const char *buf, size_t count)
274{
275 unsigned int input;
276 int ret;
277 ret = sscanf(buf, "%u", &input);
278 if (ret != 1)
279 return -EINVAL;
280 limits.no_turbo = clamp_t(int, input, 0 , 1);
281
282 return count;
283}
284
285static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
286 const char *buf, size_t count)
287{
288 unsigned int input;
289 int ret;
290 ret = sscanf(buf, "%u", &input);
291 if (ret != 1)
292 return -EINVAL;
293
d8f469e9
DB
294 limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
295 limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
93f0822d
DB
296 limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
297 return count;
298}
299
300static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
301 const char *buf, size_t count)
302{
303 unsigned int input;
304 int ret;
305 ret = sscanf(buf, "%u", &input);
306 if (ret != 1)
307 return -EINVAL;
308 limits.min_perf_pct = clamp_t(int, input, 0 , 100);
309 limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
310
311 return count;
312}
313
314show_one(no_turbo, no_turbo);
315show_one(max_perf_pct, max_perf_pct);
316show_one(min_perf_pct, min_perf_pct);
317
318define_one_global_rw(no_turbo);
319define_one_global_rw(max_perf_pct);
320define_one_global_rw(min_perf_pct);
321
322static struct attribute *intel_pstate_attributes[] = {
323 &no_turbo.attr,
324 &max_perf_pct.attr,
325 &min_perf_pct.attr,
326 NULL
327};
328
329static struct attribute_group intel_pstate_attr_group = {
330 .attrs = intel_pstate_attributes,
331};
332static struct kobject *intel_pstate_kobject;
333
334static void intel_pstate_sysfs_expose_params(void)
335{
336 int rc;
337
338 intel_pstate_kobject = kobject_create_and_add("intel_pstate",
339 &cpu_subsys.dev_root->kobj);
340 BUG_ON(!intel_pstate_kobject);
341 rc = sysfs_create_group(intel_pstate_kobject,
342 &intel_pstate_attr_group);
343 BUG_ON(rc);
344}
345
346/************************** sysfs end ************************/
19e77c28
DB
347static int byt_get_min_pstate(void)
348{
349 u64 value;
350 rdmsrl(BYT_RATIOS, value);
351 return value & 0xFF;
352}
353
354static int byt_get_max_pstate(void)
355{
356 u64 value;
357 rdmsrl(BYT_RATIOS, value);
358 return (value >> 16) & 0xFF;
359}
93f0822d 360
016c8150 361static int core_get_min_pstate(void)
93f0822d
DB
362{
363 u64 value;
05e99c8c 364 rdmsrl(MSR_PLATFORM_INFO, value);
93f0822d
DB
365 return (value >> 40) & 0xFF;
366}
367
016c8150 368static int core_get_max_pstate(void)
93f0822d
DB
369{
370 u64 value;
05e99c8c 371 rdmsrl(MSR_PLATFORM_INFO, value);
93f0822d
DB
372 return (value >> 8) & 0xFF;
373}
374
016c8150 375static int core_get_turbo_pstate(void)
93f0822d
DB
376{
377 u64 value;
378 int nont, ret;
05e99c8c 379 rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
016c8150 380 nont = core_get_max_pstate();
93f0822d
DB
381 ret = ((value) & 255);
382 if (ret <= nont)
383 ret = nont;
384 return ret;
385}
386
016c8150
DB
387static void core_set_pstate(int pstate)
388{
389 u64 val;
390
391 val = pstate << 8;
392 if (limits.no_turbo)
393 val |= (u64)1 << 32;
394
395 wrmsrl(MSR_IA32_PERF_CTL, val);
396}
397
398static struct cpu_defaults core_params = {
399 .pid_policy = {
400 .sample_rate_ms = 10,
401 .deadband = 0,
402 .setpoint = 97,
403 .p_gain_pct = 20,
404 .d_gain_pct = 0,
405 .i_gain_pct = 0,
406 },
407 .funcs = {
408 .get_max = core_get_max_pstate,
409 .get_min = core_get_min_pstate,
410 .get_turbo = core_get_turbo_pstate,
411 .set = core_set_pstate,
412 },
413};
414
19e77c28
DB
415static struct cpu_defaults byt_params = {
416 .pid_policy = {
417 .sample_rate_ms = 10,
418 .deadband = 0,
419 .setpoint = 97,
420 .p_gain_pct = 14,
421 .d_gain_pct = 0,
422 .i_gain_pct = 4,
423 },
424 .funcs = {
425 .get_max = byt_get_max_pstate,
426 .get_min = byt_get_min_pstate,
427 .get_turbo = byt_get_max_pstate,
428 .set = core_set_pstate,
429 },
430};
431
432
93f0822d
DB
433static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
434{
435 int max_perf = cpu->pstate.turbo_pstate;
7244cb62 436 int max_perf_adj;
93f0822d
DB
437 int min_perf;
438 if (limits.no_turbo)
439 max_perf = cpu->pstate.max_pstate;
440
7244cb62
DB
441 max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
442 *max = clamp_t(int, max_perf_adj,
93f0822d
DB
443 cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
444
445 min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
446 *min = clamp_t(int, min_perf,
447 cpu->pstate.min_pstate, max_perf);
448}
449
450static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
451{
452 int max_perf, min_perf;
453
454 intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
455
456 pstate = clamp_t(int, pstate, min_perf, max_perf);
457
458 if (pstate == cpu->pstate.current_pstate)
459 return;
460
93f0822d 461 trace_cpu_frequency(pstate * 100000, cpu->cpu);
35363e94 462
93f0822d 463 cpu->pstate.current_pstate = pstate;
93f0822d 464
016c8150 465 pstate_funcs.set(pstate);
93f0822d
DB
466}
467
468static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
469{
470 int target;
471 target = cpu->pstate.current_pstate + steps;
472
473 intel_pstate_set_pstate(cpu, target);
474}
475
476static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
477{
478 int target;
479 target = cpu->pstate.current_pstate - steps;
480 intel_pstate_set_pstate(cpu, target);
481}
482
483static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
484{
485 sprintf(cpu->name, "Intel 2nd generation core");
486
016c8150
DB
487 cpu->pstate.min_pstate = pstate_funcs.get_min();
488 cpu->pstate.max_pstate = pstate_funcs.get_max();
489 cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
93f0822d
DB
490
491 /*
492 * goto max pstate so we don't slow up boot if we are built-in if we are
493 * a module we will take care of it during normal operation
494 */
495 intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
496}
497
498static inline void intel_pstate_calc_busy(struct cpudata *cpu,
499 struct sample *sample)
500{
501 u64 core_pct;
d253d2a5
BS
502 core_pct = div64_u64(int_tofp(sample->aperf * 100),
503 sample->mperf);
504 sample->freq = fp_toint(cpu->pstate.max_pstate * core_pct * 1000);
93f0822d 505
1abc4b20 506 sample->core_pct_busy = core_pct;
93f0822d
DB
507}
508
509static inline void intel_pstate_sample(struct cpudata *cpu)
510{
93f0822d
DB
511 u64 aperf, mperf;
512
93f0822d
DB
513 rdmsrl(MSR_IA32_APERF, aperf);
514 rdmsrl(MSR_IA32_MPERF, mperf);
1abc4b20
DB
515 cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
516 cpu->samples[cpu->sample_ptr].aperf = aperf;
517 cpu->samples[cpu->sample_ptr].mperf = mperf;
518 cpu->samples[cpu->sample_ptr].aperf -= cpu->prev_aperf;
519 cpu->samples[cpu->sample_ptr].mperf -= cpu->prev_mperf;
520
521 intel_pstate_calc_busy(cpu, &cpu->samples[cpu->sample_ptr]);
93f0822d 522
93f0822d
DB
523 cpu->prev_aperf = aperf;
524 cpu->prev_mperf = mperf;
525}
526
527static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
528{
529 int sample_time, delay;
530
016c8150 531 sample_time = pid_params.sample_rate_ms;
93f0822d 532 delay = msecs_to_jiffies(sample_time);
93f0822d
DB
533 mod_timer_pinned(&cpu->timer, jiffies + delay);
534}
535
d253d2a5 536static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
93f0822d 537{
2134ed4d 538 int32_t core_busy, max_pstate, current_pstate;
93f0822d 539
d253d2a5 540 core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy;
2134ed4d 541 max_pstate = int_tofp(cpu->pstate.max_pstate);
93f0822d 542 current_pstate = int_tofp(cpu->pstate.current_pstate);
d253d2a5 543 return mul_fp(core_busy, div_fp(max_pstate, current_pstate));
93f0822d
DB
544}
545
546static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
547{
d253d2a5 548 int32_t busy_scaled;
93f0822d
DB
549 struct _pid *pid;
550 signed int ctl = 0;
551 int steps;
552
553 pid = &cpu->pid;
554 busy_scaled = intel_pstate_get_scaled_busy(cpu);
555
556 ctl = pid_calc(pid, busy_scaled);
557
558 steps = abs(ctl);
559 if (ctl < 0)
560 intel_pstate_pstate_increase(cpu, steps);
561 else
562 intel_pstate_pstate_decrease(cpu, steps);
563}
564
93f0822d
DB
565static void intel_pstate_timer_func(unsigned long __data)
566{
567 struct cpudata *cpu = (struct cpudata *) __data;
568
569 intel_pstate_sample(cpu);
ca182aee 570 intel_pstate_adjust_busy_pstate(cpu);
93f0822d 571
93f0822d
DB
572 if (cpu->pstate.current_pstate == cpu->pstate.min_pstate) {
573 cpu->min_pstate_count++;
574 if (!(cpu->min_pstate_count % 5)) {
575 intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
93f0822d
DB
576 }
577 } else
578 cpu->min_pstate_count = 0;
ca182aee 579
93f0822d
DB
580 intel_pstate_set_sample_time(cpu);
581}
582
583#define ICPU(model, policy) \
584 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&policy }
585
586static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
016c8150
DB
587 ICPU(0x2a, core_params),
588 ICPU(0x2d, core_params),
19e77c28 589 ICPU(0x37, byt_params),
016c8150
DB
590 ICPU(0x3a, core_params),
591 ICPU(0x3c, core_params),
592 ICPU(0x3e, core_params),
593 ICPU(0x3f, core_params),
594 ICPU(0x45, core_params),
595 ICPU(0x46, core_params),
93f0822d
DB
596 {}
597};
598MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
599
600static int intel_pstate_init_cpu(unsigned int cpunum)
601{
602
603 const struct x86_cpu_id *id;
604 struct cpudata *cpu;
605
606 id = x86_match_cpu(intel_pstate_cpu_ids);
607 if (!id)
608 return -ENODEV;
609
610 all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
611 if (!all_cpu_data[cpunum])
612 return -ENOMEM;
613
614 cpu = all_cpu_data[cpunum];
615
616 intel_pstate_get_cpu_pstates(cpu);
617
618 cpu->cpu = cpunum;
016c8150 619
93f0822d
DB
620 init_timer_deferrable(&cpu->timer);
621 cpu->timer.function = intel_pstate_timer_func;
622 cpu->timer.data =
623 (unsigned long)cpu;
624 cpu->timer.expires = jiffies + HZ/100;
625 intel_pstate_busy_pid_reset(cpu);
93f0822d
DB
626 intel_pstate_sample(cpu);
627 intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
628
629 add_timer_on(&cpu->timer, cpunum);
630
631 pr_info("Intel pstate controlling: cpu %d\n", cpunum);
632
633 return 0;
634}
635
636static unsigned int intel_pstate_get(unsigned int cpu_num)
637{
638 struct sample *sample;
639 struct cpudata *cpu;
640
641 cpu = all_cpu_data[cpu_num];
642 if (!cpu)
643 return 0;
644 sample = &cpu->samples[cpu->sample_ptr];
645 return sample->freq;
646}
647
648static int intel_pstate_set_policy(struct cpufreq_policy *policy)
649{
650 struct cpudata *cpu;
93f0822d
DB
651
652 cpu = all_cpu_data[policy->cpu];
653
d3929b83
DB
654 if (!policy->cpuinfo.max_freq)
655 return -ENODEV;
656
93f0822d
DB
657 if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
658 limits.min_perf_pct = 100;
659 limits.min_perf = int_tofp(1);
660 limits.max_perf_pct = 100;
661 limits.max_perf = int_tofp(1);
662 limits.no_turbo = 0;
d1b68485 663 return 0;
93f0822d 664 }
d1b68485
SP
665 limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
666 limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
667 limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
668
d8f469e9
DB
669 limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
670 limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
671 limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
d1b68485 672 limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
93f0822d
DB
673
674 return 0;
675}
676
677static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
678{
be49e346 679 cpufreq_verify_within_cpu_limits(policy);
93f0822d
DB
680
681 if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
682 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
683 return -EINVAL;
684
685 return 0;
686}
687
2760984f 688static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
93f0822d
DB
689{
690 int cpu = policy->cpu;
691
692 del_timer(&all_cpu_data[cpu]->timer);
693 kfree(all_cpu_data[cpu]);
694 all_cpu_data[cpu] = NULL;
695 return 0;
696}
697
2760984f 698static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
93f0822d 699{
93f0822d 700 struct cpudata *cpu;
52e0a509 701 int rc;
93f0822d
DB
702
703 rc = intel_pstate_init_cpu(policy->cpu);
704 if (rc)
705 return rc;
706
707 cpu = all_cpu_data[policy->cpu];
708
709 if (!limits.no_turbo &&
710 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
711 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
712 else
713 policy->policy = CPUFREQ_POLICY_POWERSAVE;
714
52e0a509
DB
715 policy->min = cpu->pstate.min_pstate * 100000;
716 policy->max = cpu->pstate.turbo_pstate * 100000;
93f0822d
DB
717
718 /* cpuinfo and default policy values */
719 policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
720 policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
721 policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
722 cpumask_set_cpu(policy->cpu, policy->cpus);
723
724 return 0;
725}
726
727static struct cpufreq_driver intel_pstate_driver = {
728 .flags = CPUFREQ_CONST_LOOPS,
729 .verify = intel_pstate_verify_policy,
730 .setpolicy = intel_pstate_set_policy,
731 .get = intel_pstate_get,
732 .init = intel_pstate_cpu_init,
733 .exit = intel_pstate_cpu_exit,
734 .name = "intel_pstate",
93f0822d
DB
735};
736
6be26498
DB
737static int __initdata no_load;
738
b563b4e3
DB
739static int intel_pstate_msrs_not_valid(void)
740{
741 /* Check that all the msr's we are using are valid. */
742 u64 aperf, mperf, tmp;
743
744 rdmsrl(MSR_IA32_APERF, aperf);
745 rdmsrl(MSR_IA32_MPERF, mperf);
746
016c8150
DB
747 if (!pstate_funcs.get_max() ||
748 !pstate_funcs.get_min() ||
749 !pstate_funcs.get_turbo())
b563b4e3
DB
750 return -ENODEV;
751
752 rdmsrl(MSR_IA32_APERF, tmp);
753 if (!(tmp - aperf))
754 return -ENODEV;
755
756 rdmsrl(MSR_IA32_MPERF, tmp);
757 if (!(tmp - mperf))
758 return -ENODEV;
759
760 return 0;
761}
016c8150 762
e0a261a2 763static void copy_pid_params(struct pstate_adjust_policy *policy)
016c8150
DB
764{
765 pid_params.sample_rate_ms = policy->sample_rate_ms;
766 pid_params.p_gain_pct = policy->p_gain_pct;
767 pid_params.i_gain_pct = policy->i_gain_pct;
768 pid_params.d_gain_pct = policy->d_gain_pct;
769 pid_params.deadband = policy->deadband;
770 pid_params.setpoint = policy->setpoint;
771}
772
e0a261a2 773static void copy_cpu_funcs(struct pstate_funcs *funcs)
016c8150
DB
774{
775 pstate_funcs.get_max = funcs->get_max;
776 pstate_funcs.get_min = funcs->get_min;
777 pstate_funcs.get_turbo = funcs->get_turbo;
778 pstate_funcs.set = funcs->set;
779}
780
fbbcdc07
AH
781#if IS_ENABLED(CONFIG_ACPI)
782#include <acpi/processor.h>
783
784static bool intel_pstate_no_acpi_pss(void)
785{
786 int i;
787
788 for_each_possible_cpu(i) {
789 acpi_status status;
790 union acpi_object *pss;
791 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
792 struct acpi_processor *pr = per_cpu(processors, i);
793
794 if (!pr)
795 continue;
796
797 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
798 if (ACPI_FAILURE(status))
799 continue;
800
801 pss = buffer.pointer;
802 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
803 kfree(pss);
804 return false;
805 }
806
807 kfree(pss);
808 }
809
810 return true;
811}
812
813struct hw_vendor_info {
814 u16 valid;
815 char oem_id[ACPI_OEM_ID_SIZE];
816 char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
817};
818
819/* Hardware vendor-specific info that has its own power management modes */
820static struct hw_vendor_info vendor_info[] = {
821 {1, "HP ", "ProLiant"},
822 {0, "", ""},
823};
824
825static bool intel_pstate_platform_pwr_mgmt_exists(void)
826{
827 struct acpi_table_header hdr;
828 struct hw_vendor_info *v_info;
829
830 if (acpi_disabled
831 || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
832 return false;
833
834 for (v_info = vendor_info; v_info->valid; v_info++) {
835 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
836 && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
837 && intel_pstate_no_acpi_pss())
838 return true;
839 }
840
841 return false;
842}
843#else /* CONFIG_ACPI not enabled */
844static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
845#endif /* CONFIG_ACPI */
846
93f0822d
DB
847static int __init intel_pstate_init(void)
848{
907cc908 849 int cpu, rc = 0;
93f0822d 850 const struct x86_cpu_id *id;
016c8150 851 struct cpu_defaults *cpu_info;
93f0822d 852
6be26498
DB
853 if (no_load)
854 return -ENODEV;
855
93f0822d
DB
856 id = x86_match_cpu(intel_pstate_cpu_ids);
857 if (!id)
858 return -ENODEV;
859
fbbcdc07
AH
860 /*
861 * The Intel pstate driver will be ignored if the platform
862 * firmware has its own power management modes.
863 */
864 if (intel_pstate_platform_pwr_mgmt_exists())
865 return -ENODEV;
866
016c8150
DB
867 cpu_info = (struct cpu_defaults *)id->driver_data;
868
869 copy_pid_params(&cpu_info->pid_policy);
870 copy_cpu_funcs(&cpu_info->funcs);
871
b563b4e3
DB
872 if (intel_pstate_msrs_not_valid())
873 return -ENODEV;
874
93f0822d
DB
875 pr_info("Intel P-state driver initializing.\n");
876
b57ffac5 877 all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
93f0822d
DB
878 if (!all_cpu_data)
879 return -ENOMEM;
93f0822d
DB
880
881 rc = cpufreq_register_driver(&intel_pstate_driver);
882 if (rc)
883 goto out;
884
885 intel_pstate_debug_expose_params();
886 intel_pstate_sysfs_expose_params();
887 return rc;
888out:
907cc908
DB
889 get_online_cpus();
890 for_each_online_cpu(cpu) {
891 if (all_cpu_data[cpu]) {
892 del_timer_sync(&all_cpu_data[cpu]->timer);
893 kfree(all_cpu_data[cpu]);
894 }
895 }
896
897 put_online_cpus();
898 vfree(all_cpu_data);
93f0822d
DB
899 return -ENODEV;
900}
901device_initcall(intel_pstate_init);
902
6be26498
DB
903static int __init intel_pstate_setup(char *str)
904{
905 if (!str)
906 return -EINVAL;
907
908 if (!strcmp(str, "disable"))
909 no_load = 1;
910 return 0;
911}
912early_param("intel_pstate", intel_pstate_setup);
913
93f0822d
DB
914MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
915MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
916MODULE_LICENSE("GPL");