1 /*
2  * intel_pstate.c: Native P state management for Intel processors
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>
28 #include <linux/acpi.h>
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 BYT_RATIOS		0x66a
36 #define BYT_VIDS		0x66b
37 #define BYT_TURBO_RATIOS	0x66c
38 #define BYT_TURBO_VIDS		0x66d
39 
40 
41 #define FRAC_BITS 8
42 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
43 #define fp_toint(X) ((X) >> FRAC_BITS)
44 
45 
46 static inline int32_t mul_fp(int32_t x, int32_t y)
47 {
48 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
49 }
50 
51 static inline int32_t div_fp(int32_t x, int32_t y)
52 {
53 	return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
54 }
55 
56 struct sample {
57 	int32_t core_pct_busy;
58 	u64 aperf;
59 	u64 mperf;
60 	int freq;
61 	ktime_t time;
62 };
63 
64 struct pstate_data {
65 	int	current_pstate;
66 	int	min_pstate;
67 	int	max_pstate;
68 	int	turbo_pstate;
69 };
70 
71 struct vid_data {
72 	int min;
73 	int max;
74 	int turbo;
75 	int32_t ratio;
76 };
77 
78 struct _pid {
79 	int setpoint;
80 	int32_t integral;
81 	int32_t p_gain;
82 	int32_t i_gain;
83 	int32_t d_gain;
84 	int deadband;
85 	int32_t last_err;
86 };
87 
88 struct cpudata {
89 	int cpu;
90 
91 	struct timer_list timer;
92 
93 	struct pstate_data pstate;
94 	struct vid_data vid;
95 	struct _pid pid;
96 
97 	ktime_t last_sample_time;
98 	u64	prev_aperf;
99 	u64	prev_mperf;
100 	struct sample sample;
101 };
102 
103 static struct cpudata **all_cpu_data;
104 struct pstate_adjust_policy {
105 	int sample_rate_ms;
106 	int deadband;
107 	int setpoint;
108 	int p_gain_pct;
109 	int d_gain_pct;
110 	int i_gain_pct;
111 };
112 
113 struct pstate_funcs {
114 	int (*get_max)(void);
115 	int (*get_min)(void);
116 	int (*get_turbo)(void);
117 	void (*set)(struct cpudata*, int pstate);
118 	void (*get_vid)(struct cpudata *);
119 };
120 
121 struct cpu_defaults {
122 	struct pstate_adjust_policy pid_policy;
123 	struct pstate_funcs funcs;
124 };
125 
126 static struct pstate_adjust_policy pid_params;
127 static struct pstate_funcs pstate_funcs;
128 
129 struct perf_limits {
130 	int no_turbo;
131 	int turbo_disabled;
132 	int max_perf_pct;
133 	int min_perf_pct;
134 	int32_t max_perf;
135 	int32_t min_perf;
136 	int max_policy_pct;
137 	int max_sysfs_pct;
138 };
139 
140 static struct perf_limits limits = {
141 	.no_turbo = 0,
142 	.max_perf_pct = 100,
143 	.max_perf = int_tofp(1),
144 	.min_perf_pct = 0,
145 	.min_perf = 0,
146 	.max_policy_pct = 100,
147 	.max_sysfs_pct = 100,
148 };
149 
150 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
151 			int deadband, int integral) {
152 	pid->setpoint = setpoint;
153 	pid->deadband  = deadband;
154 	pid->integral  = int_tofp(integral);
155 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
156 }
157 
158 static inline void pid_p_gain_set(struct _pid *pid, int percent)
159 {
160 	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
161 }
162 
163 static inline void pid_i_gain_set(struct _pid *pid, int percent)
164 {
165 	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
166 }
167 
168 static inline void pid_d_gain_set(struct _pid *pid, int percent)
169 {
170 
171 	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
172 }
173 
174 static signed int pid_calc(struct _pid *pid, int32_t busy)
175 {
176 	signed int result;
177 	int32_t pterm, dterm, fp_error;
178 	int32_t integral_limit;
179 
180 	fp_error = int_tofp(pid->setpoint) - busy;
181 
182 	if (abs(fp_error) <= int_tofp(pid->deadband))
183 		return 0;
184 
185 	pterm = mul_fp(pid->p_gain, fp_error);
186 
187 	pid->integral += fp_error;
188 
189 	/* limit the integral term */
190 	integral_limit = int_tofp(30);
191 	if (pid->integral > integral_limit)
192 		pid->integral = integral_limit;
193 	if (pid->integral < -integral_limit)
194 		pid->integral = -integral_limit;
195 
196 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
197 	pid->last_err = fp_error;
198 
199 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
200 	result = result + (1 << (FRAC_BITS-1));
201 	return (signed int)fp_toint(result);
202 }
203 
204 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
205 {
206 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
207 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
208 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
209 
210 	pid_reset(&cpu->pid,
211 		pid_params.setpoint,
212 		100,
213 		pid_params.deadband,
214 		0);
215 }
216 
217 static inline void intel_pstate_reset_all_pid(void)
218 {
219 	unsigned int cpu;
220 	for_each_online_cpu(cpu) {
221 		if (all_cpu_data[cpu])
222 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
223 	}
224 }
225 
226 /************************** debugfs begin ************************/
227 static int pid_param_set(void *data, u64 val)
228 {
229 	*(u32 *)data = val;
230 	intel_pstate_reset_all_pid();
231 	return 0;
232 }
233 static int pid_param_get(void *data, u64 *val)
234 {
235 	*val = *(u32 *)data;
236 	return 0;
237 }
238 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
239 			pid_param_set, "%llu\n");
240 
241 struct pid_param {
242 	char *name;
243 	void *value;
244 };
245 
246 static struct pid_param pid_files[] = {
247 	{"sample_rate_ms", &pid_params.sample_rate_ms},
248 	{"d_gain_pct", &pid_params.d_gain_pct},
249 	{"i_gain_pct", &pid_params.i_gain_pct},
250 	{"deadband", &pid_params.deadband},
251 	{"setpoint", &pid_params.setpoint},
252 	{"p_gain_pct", &pid_params.p_gain_pct},
253 	{NULL, NULL}
254 };
255 
256 static struct dentry *debugfs_parent;
257 static void intel_pstate_debug_expose_params(void)
258 {
259 	int i = 0;
260 
261 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
262 	if (IS_ERR_OR_NULL(debugfs_parent))
263 		return;
264 	while (pid_files[i].name) {
265 		debugfs_create_file(pid_files[i].name, 0660,
266 				debugfs_parent, pid_files[i].value,
267 				&fops_pid_param);
268 		i++;
269 	}
270 }
271 
272 /************************** debugfs end ************************/
273 
274 /************************** sysfs begin ************************/
275 #define show_one(file_name, object)					\
276 	static ssize_t show_##file_name					\
277 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
278 	{								\
279 		return sprintf(buf, "%u\n", limits.object);		\
280 	}
281 
282 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
283 				const char *buf, size_t count)
284 {
285 	unsigned int input;
286 	int ret;
287 	ret = sscanf(buf, "%u", &input);
288 	if (ret != 1)
289 		return -EINVAL;
290 	limits.no_turbo = clamp_t(int, input, 0 , 1);
291 	if (limits.turbo_disabled) {
292 		pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
293 		limits.no_turbo = limits.turbo_disabled;
294 	}
295 	return count;
296 }
297 
298 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
299 				const char *buf, size_t count)
300 {
301 	unsigned int input;
302 	int ret;
303 	ret = sscanf(buf, "%u", &input);
304 	if (ret != 1)
305 		return -EINVAL;
306 
307 	limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
308 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
309 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
310 	return count;
311 }
312 
313 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
314 				const char *buf, size_t count)
315 {
316 	unsigned int input;
317 	int ret;
318 	ret = sscanf(buf, "%u", &input);
319 	if (ret != 1)
320 		return -EINVAL;
321 	limits.min_perf_pct = clamp_t(int, input, 0 , 100);
322 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
323 
324 	return count;
325 }
326 
327 show_one(no_turbo, no_turbo);
328 show_one(max_perf_pct, max_perf_pct);
329 show_one(min_perf_pct, min_perf_pct);
330 
331 define_one_global_rw(no_turbo);
332 define_one_global_rw(max_perf_pct);
333 define_one_global_rw(min_perf_pct);
334 
335 static struct attribute *intel_pstate_attributes[] = {
336 	&no_turbo.attr,
337 	&max_perf_pct.attr,
338 	&min_perf_pct.attr,
339 	NULL
340 };
341 
342 static struct attribute_group intel_pstate_attr_group = {
343 	.attrs = intel_pstate_attributes,
344 };
345 static struct kobject *intel_pstate_kobject;
346 
347 static void intel_pstate_sysfs_expose_params(void)
348 {
349 	int rc;
350 
351 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
352 						&cpu_subsys.dev_root->kobj);
353 	BUG_ON(!intel_pstate_kobject);
354 	rc = sysfs_create_group(intel_pstate_kobject,
355 				&intel_pstate_attr_group);
356 	BUG_ON(rc);
357 }
358 
359 /************************** sysfs end ************************/
360 static int byt_get_min_pstate(void)
361 {
362 	u64 value;
363 	rdmsrl(BYT_RATIOS, value);
364 	return (value >> 8) & 0x7F;
365 }
366 
367 static int byt_get_max_pstate(void)
368 {
369 	u64 value;
370 	rdmsrl(BYT_RATIOS, value);
371 	return (value >> 16) & 0x7F;
372 }
373 
374 static int byt_get_turbo_pstate(void)
375 {
376 	u64 value;
377 	rdmsrl(BYT_TURBO_RATIOS, value);
378 	return value & 0x7F;
379 }
380 
381 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
382 {
383 	u64 val;
384 	int32_t vid_fp;
385 	u32 vid;
386 
387 	val = pstate << 8;
388 	if (limits.no_turbo && !limits.turbo_disabled)
389 		val |= (u64)1 << 32;
390 
391 	vid_fp = cpudata->vid.min + mul_fp(
392 		int_tofp(pstate - cpudata->pstate.min_pstate),
393 		cpudata->vid.ratio);
394 
395 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
396 	vid = fp_toint(vid_fp);
397 
398 	if (pstate > cpudata->pstate.max_pstate)
399 		vid = cpudata->vid.turbo;
400 
401 	val |= vid;
402 
403 	wrmsrl(MSR_IA32_PERF_CTL, val);
404 }
405 
406 static void byt_get_vid(struct cpudata *cpudata)
407 {
408 	u64 value;
409 
410 
411 	rdmsrl(BYT_VIDS, value);
412 	cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
413 	cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
414 	cpudata->vid.ratio = div_fp(
415 		cpudata->vid.max - cpudata->vid.min,
416 		int_tofp(cpudata->pstate.max_pstate -
417 			cpudata->pstate.min_pstate));
418 
419 	rdmsrl(BYT_TURBO_VIDS, value);
420 	cpudata->vid.turbo = value & 0x7f;
421 }
422 
423 
424 static int core_get_min_pstate(void)
425 {
426 	u64 value;
427 	rdmsrl(MSR_PLATFORM_INFO, value);
428 	return (value >> 40) & 0xFF;
429 }
430 
431 static int core_get_max_pstate(void)
432 {
433 	u64 value;
434 	rdmsrl(MSR_PLATFORM_INFO, value);
435 	return (value >> 8) & 0xFF;
436 }
437 
438 static int core_get_turbo_pstate(void)
439 {
440 	u64 value;
441 	int nont, ret;
442 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
443 	nont = core_get_max_pstate();
444 	ret = ((value) & 255);
445 	if (ret <= nont)
446 		ret = nont;
447 	return ret;
448 }
449 
450 static void core_set_pstate(struct cpudata *cpudata, int pstate)
451 {
452 	u64 val;
453 
454 	val = pstate << 8;
455 	if (limits.no_turbo && !limits.turbo_disabled)
456 		val |= (u64)1 << 32;
457 
458 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
459 }
460 
461 static struct cpu_defaults core_params = {
462 	.pid_policy = {
463 		.sample_rate_ms = 10,
464 		.deadband = 0,
465 		.setpoint = 97,
466 		.p_gain_pct = 20,
467 		.d_gain_pct = 0,
468 		.i_gain_pct = 0,
469 	},
470 	.funcs = {
471 		.get_max = core_get_max_pstate,
472 		.get_min = core_get_min_pstate,
473 		.get_turbo = core_get_turbo_pstate,
474 		.set = core_set_pstate,
475 	},
476 };
477 
478 static struct cpu_defaults byt_params = {
479 	.pid_policy = {
480 		.sample_rate_ms = 10,
481 		.deadband = 0,
482 		.setpoint = 97,
483 		.p_gain_pct = 14,
484 		.d_gain_pct = 0,
485 		.i_gain_pct = 4,
486 	},
487 	.funcs = {
488 		.get_max = byt_get_max_pstate,
489 		.get_min = byt_get_min_pstate,
490 		.get_turbo = byt_get_turbo_pstate,
491 		.set = byt_set_pstate,
492 		.get_vid = byt_get_vid,
493 	},
494 };
495 
496 
497 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
498 {
499 	int max_perf = cpu->pstate.turbo_pstate;
500 	int max_perf_adj;
501 	int min_perf;
502 	if (limits.no_turbo)
503 		max_perf = cpu->pstate.max_pstate;
504 
505 	max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
506 	*max = clamp_t(int, max_perf_adj,
507 			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
508 
509 	min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
510 	*min = clamp_t(int, min_perf,
511 			cpu->pstate.min_pstate, max_perf);
512 }
513 
514 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
515 {
516 	int max_perf, min_perf;
517 
518 	intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
519 
520 	pstate = clamp_t(int, pstate, min_perf, max_perf);
521 
522 	if (pstate == cpu->pstate.current_pstate)
523 		return;
524 
525 	trace_cpu_frequency(pstate * 100000, cpu->cpu);
526 
527 	cpu->pstate.current_pstate = pstate;
528 
529 	pstate_funcs.set(cpu, pstate);
530 }
531 
532 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
533 {
534 	int target;
535 	target = cpu->pstate.current_pstate + steps;
536 
537 	intel_pstate_set_pstate(cpu, target);
538 }
539 
540 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
541 {
542 	int target;
543 	target = cpu->pstate.current_pstate - steps;
544 	intel_pstate_set_pstate(cpu, target);
545 }
546 
547 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
548 {
549 	cpu->pstate.min_pstate = pstate_funcs.get_min();
550 	cpu->pstate.max_pstate = pstate_funcs.get_max();
551 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
552 
553 	if (pstate_funcs.get_vid)
554 		pstate_funcs.get_vid(cpu);
555 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
556 }
557 
558 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
559 {
560 	struct sample *sample = &cpu->sample;
561 	int64_t core_pct;
562 	int32_t rem;
563 
564 	core_pct = int_tofp(sample->aperf) * int_tofp(100);
565 	core_pct = div_u64_rem(core_pct, int_tofp(sample->mperf), &rem);
566 
567 	if ((rem << 1) >= int_tofp(sample->mperf))
568 		core_pct += 1;
569 
570 	sample->freq = fp_toint(
571 		mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
572 
573 	sample->core_pct_busy = (int32_t)core_pct;
574 }
575 
576 static inline void intel_pstate_sample(struct cpudata *cpu)
577 {
578 	u64 aperf, mperf;
579 
580 	rdmsrl(MSR_IA32_APERF, aperf);
581 	rdmsrl(MSR_IA32_MPERF, mperf);
582 
583 	aperf = aperf >> FRAC_BITS;
584 	mperf = mperf >> FRAC_BITS;
585 
586 	cpu->last_sample_time = cpu->sample.time;
587 	cpu->sample.time = ktime_get();
588 	cpu->sample.aperf = aperf;
589 	cpu->sample.mperf = mperf;
590 	cpu->sample.aperf -= cpu->prev_aperf;
591 	cpu->sample.mperf -= cpu->prev_mperf;
592 
593 	intel_pstate_calc_busy(cpu);
594 
595 	cpu->prev_aperf = aperf;
596 	cpu->prev_mperf = mperf;
597 }
598 
599 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
600 {
601 	int sample_time, delay;
602 
603 	sample_time = pid_params.sample_rate_ms;
604 	delay = msecs_to_jiffies(sample_time);
605 	mod_timer_pinned(&cpu->timer, jiffies + delay);
606 }
607 
608 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
609 {
610 	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
611 	u32 duration_us;
612 	u32 sample_time;
613 
614 	core_busy = cpu->sample.core_pct_busy;
615 	max_pstate = int_tofp(cpu->pstate.max_pstate);
616 	current_pstate = int_tofp(cpu->pstate.current_pstate);
617 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
618 
619 	sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
620 	duration_us = (u32) ktime_us_delta(cpu->sample.time,
621 					cpu->last_sample_time);
622 	if (duration_us > sample_time * 3) {
623 		sample_ratio = div_fp(int_tofp(sample_time),
624 				int_tofp(duration_us));
625 		core_busy = mul_fp(core_busy, sample_ratio);
626 	}
627 
628 	return core_busy;
629 }
630 
631 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
632 {
633 	int32_t busy_scaled;
634 	struct _pid *pid;
635 	signed int ctl = 0;
636 	int steps;
637 
638 	pid = &cpu->pid;
639 	busy_scaled = intel_pstate_get_scaled_busy(cpu);
640 
641 	ctl = pid_calc(pid, busy_scaled);
642 
643 	steps = abs(ctl);
644 
645 	if (ctl < 0)
646 		intel_pstate_pstate_increase(cpu, steps);
647 	else
648 		intel_pstate_pstate_decrease(cpu, steps);
649 }
650 
651 static void intel_pstate_timer_func(unsigned long __data)
652 {
653 	struct cpudata *cpu = (struct cpudata *) __data;
654 	struct sample *sample;
655 
656 	intel_pstate_sample(cpu);
657 
658 	sample = &cpu->sample;
659 
660 	intel_pstate_adjust_busy_pstate(cpu);
661 
662 	trace_pstate_sample(fp_toint(sample->core_pct_busy),
663 			fp_toint(intel_pstate_get_scaled_busy(cpu)),
664 			cpu->pstate.current_pstate,
665 			sample->mperf,
666 			sample->aperf,
667 			sample->freq);
668 
669 	intel_pstate_set_sample_time(cpu);
670 }
671 
672 #define ICPU(model, policy) \
673 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
674 			(unsigned long)&policy }
675 
676 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
677 	ICPU(0x2a, core_params),
678 	ICPU(0x2d, core_params),
679 	ICPU(0x37, byt_params),
680 	ICPU(0x3a, core_params),
681 	ICPU(0x3c, core_params),
682 	ICPU(0x3d, core_params),
683 	ICPU(0x3e, core_params),
684 	ICPU(0x3f, core_params),
685 	ICPU(0x45, core_params),
686 	ICPU(0x46, core_params),
687 	ICPU(0x4f, core_params),
688 	ICPU(0x56, core_params),
689 	{}
690 };
691 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
692 
693 static int intel_pstate_init_cpu(unsigned int cpunum)
694 {
695 	struct cpudata *cpu;
696 
697 	all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
698 	if (!all_cpu_data[cpunum])
699 		return -ENOMEM;
700 
701 	cpu = all_cpu_data[cpunum];
702 
703 	cpu->cpu = cpunum;
704 	intel_pstate_get_cpu_pstates(cpu);
705 
706 	init_timer_deferrable(&cpu->timer);
707 	cpu->timer.function = intel_pstate_timer_func;
708 	cpu->timer.data =
709 		(unsigned long)cpu;
710 	cpu->timer.expires = jiffies + HZ/100;
711 	intel_pstate_busy_pid_reset(cpu);
712 	intel_pstate_sample(cpu);
713 
714 	add_timer_on(&cpu->timer, cpunum);
715 
716 	pr_info("Intel pstate controlling: cpu %d\n", cpunum);
717 
718 	return 0;
719 }
720 
721 static unsigned int intel_pstate_get(unsigned int cpu_num)
722 {
723 	struct sample *sample;
724 	struct cpudata *cpu;
725 
726 	cpu = all_cpu_data[cpu_num];
727 	if (!cpu)
728 		return 0;
729 	sample = &cpu->sample;
730 	return sample->freq;
731 }
732 
733 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
734 {
735 	struct cpudata *cpu;
736 
737 	cpu = all_cpu_data[policy->cpu];
738 
739 	if (!policy->cpuinfo.max_freq)
740 		return -ENODEV;
741 
742 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
743 		limits.min_perf_pct = 100;
744 		limits.min_perf = int_tofp(1);
745 		limits.max_perf_pct = 100;
746 		limits.max_perf = int_tofp(1);
747 		limits.no_turbo = limits.turbo_disabled;
748 		return 0;
749 	}
750 	limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
751 	limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
752 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
753 
754 	limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
755 	limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
756 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
757 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
758 
759 	return 0;
760 }
761 
762 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
763 {
764 	cpufreq_verify_within_cpu_limits(policy);
765 
766 	if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
767 		(policy->policy != CPUFREQ_POLICY_PERFORMANCE))
768 		return -EINVAL;
769 
770 	return 0;
771 }
772 
773 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
774 {
775 	int cpu_num = policy->cpu;
776 	struct cpudata *cpu = all_cpu_data[cpu_num];
777 
778 	pr_info("intel_pstate CPU %d exiting\n", cpu_num);
779 
780 	del_timer_sync(&all_cpu_data[cpu_num]->timer);
781 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
782 	kfree(all_cpu_data[cpu_num]);
783 	all_cpu_data[cpu_num] = NULL;
784 }
785 
786 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
787 {
788 	struct cpudata *cpu;
789 	int rc;
790 	u64 misc_en;
791 
792 	rc = intel_pstate_init_cpu(policy->cpu);
793 	if (rc)
794 		return rc;
795 
796 	cpu = all_cpu_data[policy->cpu];
797 
798 	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
799 	if (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
800 		cpu->pstate.max_pstate == cpu->pstate.turbo_pstate) {
801 		limits.turbo_disabled = 1;
802 		limits.no_turbo = 1;
803 	}
804 	if (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 
809 	policy->min = cpu->pstate.min_pstate * 100000;
810 	policy->max = cpu->pstate.turbo_pstate * 100000;
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,
827 	.stop_cpu	= intel_pstate_stop_cpu,
828 	.name		= "intel_pstate",
829 };
830 
831 static int __initdata no_load;
832 
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 
841 	if (!pstate_funcs.get_max() ||
842 		!pstate_funcs.get_min() ||
843 		!pstate_funcs.get_turbo())
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 }
856 
857 static void copy_pid_params(struct pstate_adjust_policy *policy)
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 
867 static void copy_cpu_funcs(struct pstate_funcs *funcs)
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;
873 	pstate_funcs.get_vid   = funcs->get_vid;
874 }
875 
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 
942 static int __init intel_pstate_init(void)
943 {
944 	int cpu, rc = 0;
945 	const struct x86_cpu_id *id;
946 	struct cpu_defaults *cpu_info;
947 
948 	if (no_load)
949 		return -ENODEV;
950 
951 	id = x86_match_cpu(intel_pstate_cpu_ids);
952 	if (!id)
953 		return -ENODEV;
954 
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 
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 
967 	if (intel_pstate_msrs_not_valid())
968 		return -ENODEV;
969 
970 	pr_info("Intel P-state driver initializing.\n");
971 
972 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
973 	if (!all_cpu_data)
974 		return -ENOMEM;
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();
982 
983 	return rc;
984 out:
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);
995 	return -ENODEV;
996 }
997 device_initcall(intel_pstate_init);
998 
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 
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");
1013