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 #define FRAC_BITS 8
41 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
42 #define fp_toint(X) ((X) >> FRAC_BITS)
43 
44 
45 static inline int32_t mul_fp(int32_t x, int32_t y)
46 {
47 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
48 }
49 
50 static inline int32_t div_fp(int32_t x, int32_t y)
51 {
52 	return div_s64((int64_t)x << FRAC_BITS, y);
53 }
54 
55 struct sample {
56 	int32_t core_pct_busy;
57 	u64 aperf;
58 	u64 mperf;
59 	int freq;
60 	ktime_t time;
61 };
62 
63 struct pstate_data {
64 	int	current_pstate;
65 	int	min_pstate;
66 	int	max_pstate;
67 	int	turbo_pstate;
68 };
69 
70 struct vid_data {
71 	int min;
72 	int max;
73 	int turbo;
74 	int32_t ratio;
75 };
76 
77 struct _pid {
78 	int setpoint;
79 	int32_t integral;
80 	int32_t p_gain;
81 	int32_t i_gain;
82 	int32_t d_gain;
83 	int deadband;
84 	int32_t last_err;
85 };
86 
87 struct cpudata {
88 	int cpu;
89 
90 	struct timer_list timer;
91 
92 	struct pstate_data pstate;
93 	struct vid_data vid;
94 	struct _pid pid;
95 
96 	ktime_t last_sample_time;
97 	u64	prev_aperf;
98 	u64	prev_mperf;
99 	struct sample sample;
100 };
101 
102 static struct cpudata **all_cpu_data;
103 struct pstate_adjust_policy {
104 	int sample_rate_ms;
105 	int deadband;
106 	int setpoint;
107 	int p_gain_pct;
108 	int d_gain_pct;
109 	int i_gain_pct;
110 };
111 
112 struct pstate_funcs {
113 	int (*get_max)(void);
114 	int (*get_min)(void);
115 	int (*get_turbo)(void);
116 	void (*set)(struct cpudata*, int pstate);
117 	void (*get_vid)(struct cpudata *);
118 };
119 
120 struct cpu_defaults {
121 	struct pstate_adjust_policy pid_policy;
122 	struct pstate_funcs funcs;
123 };
124 
125 static struct pstate_adjust_policy pid_params;
126 static struct pstate_funcs pstate_funcs;
127 
128 struct perf_limits {
129 	int no_turbo;
130 	int turbo_disabled;
131 	int max_perf_pct;
132 	int min_perf_pct;
133 	int32_t max_perf;
134 	int32_t min_perf;
135 	int max_policy_pct;
136 	int max_sysfs_pct;
137 };
138 
139 static struct perf_limits limits = {
140 	.no_turbo = 0,
141 	.max_perf_pct = 100,
142 	.max_perf = int_tofp(1),
143 	.min_perf_pct = 0,
144 	.min_perf = 0,
145 	.max_policy_pct = 100,
146 	.max_sysfs_pct = 100,
147 };
148 
149 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
150 			     int deadband, int integral) {
151 	pid->setpoint = setpoint;
152 	pid->deadband  = deadband;
153 	pid->integral  = int_tofp(integral);
154 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
155 }
156 
157 static inline void pid_p_gain_set(struct _pid *pid, int percent)
158 {
159 	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
160 }
161 
162 static inline void pid_i_gain_set(struct _pid *pid, int percent)
163 {
164 	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
165 }
166 
167 static inline void pid_d_gain_set(struct _pid *pid, int percent)
168 {
169 	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
170 }
171 
172 static signed int pid_calc(struct _pid *pid, int32_t busy)
173 {
174 	signed int result;
175 	int32_t pterm, dterm, fp_error;
176 	int32_t integral_limit;
177 
178 	fp_error = int_tofp(pid->setpoint) - busy;
179 
180 	if (abs(fp_error) <= int_tofp(pid->deadband))
181 		return 0;
182 
183 	pterm = mul_fp(pid->p_gain, fp_error);
184 
185 	pid->integral += fp_error;
186 
187 	/* limit the integral term */
188 	integral_limit = int_tofp(30);
189 	if (pid->integral > integral_limit)
190 		pid->integral = integral_limit;
191 	if (pid->integral < -integral_limit)
192 		pid->integral = -integral_limit;
193 
194 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
195 	pid->last_err = fp_error;
196 
197 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
198 	result = result + (1 << (FRAC_BITS-1));
199 	return (signed int)fp_toint(result);
200 }
201 
202 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
203 {
204 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
205 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
206 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
207 
208 	pid_reset(&cpu->pid, pid_params.setpoint, 100, pid_params.deadband, 0);
209 }
210 
211 static inline void intel_pstate_reset_all_pid(void)
212 {
213 	unsigned int cpu;
214 
215 	for_each_online_cpu(cpu) {
216 		if (all_cpu_data[cpu])
217 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
218 	}
219 }
220 
221 /************************** debugfs begin ************************/
222 static int pid_param_set(void *data, u64 val)
223 {
224 	*(u32 *)data = val;
225 	intel_pstate_reset_all_pid();
226 	return 0;
227 }
228 
229 static int pid_param_get(void *data, u64 *val)
230 {
231 	*val = *(u32 *)data;
232 	return 0;
233 }
234 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, pid_param_set, "%llu\n");
235 
236 struct pid_param {
237 	char *name;
238 	void *value;
239 };
240 
241 static struct pid_param pid_files[] = {
242 	{"sample_rate_ms", &pid_params.sample_rate_ms},
243 	{"d_gain_pct", &pid_params.d_gain_pct},
244 	{"i_gain_pct", &pid_params.i_gain_pct},
245 	{"deadband", &pid_params.deadband},
246 	{"setpoint", &pid_params.setpoint},
247 	{"p_gain_pct", &pid_params.p_gain_pct},
248 	{NULL, NULL}
249 };
250 
251 static void __init intel_pstate_debug_expose_params(void)
252 {
253 	struct dentry *debugfs_parent;
254 	int i = 0;
255 
256 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
257 	if (IS_ERR_OR_NULL(debugfs_parent))
258 		return;
259 	while (pid_files[i].name) {
260 		debugfs_create_file(pid_files[i].name, 0660,
261 				    debugfs_parent, pid_files[i].value,
262 				    &fops_pid_param);
263 		i++;
264 	}
265 }
266 
267 /************************** debugfs end ************************/
268 
269 /************************** sysfs begin ************************/
270 #define show_one(file_name, object)					\
271 	static ssize_t show_##file_name					\
272 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
273 	{								\
274 		return sprintf(buf, "%u\n", limits.object);		\
275 	}
276 
277 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
278 			      const char *buf, size_t count)
279 {
280 	unsigned int input;
281 	int ret;
282 
283 	ret = sscanf(buf, "%u", &input);
284 	if (ret != 1)
285 		return -EINVAL;
286 	limits.no_turbo = clamp_t(int, input, 0 , 1);
287 	if (limits.turbo_disabled) {
288 		pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
289 		limits.no_turbo = limits.turbo_disabled;
290 	}
291 	return count;
292 }
293 
294 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
295 				  const char *buf, size_t count)
296 {
297 	unsigned int input;
298 	int ret;
299 
300 	ret = sscanf(buf, "%u", &input);
301 	if (ret != 1)
302 		return -EINVAL;
303 
304 	limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
305 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
306 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
307 
308 	return count;
309 }
310 
311 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
312 				  const char *buf, size_t count)
313 {
314 	unsigned int input;
315 	int ret;
316 
317 	ret = sscanf(buf, "%u", &input);
318 	if (ret != 1)
319 		return -EINVAL;
320 	limits.min_perf_pct = clamp_t(int, input, 0 , 100);
321 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
322 
323 	return count;
324 }
325 
326 show_one(no_turbo, no_turbo);
327 show_one(max_perf_pct, max_perf_pct);
328 show_one(min_perf_pct, min_perf_pct);
329 
330 define_one_global_rw(no_turbo);
331 define_one_global_rw(max_perf_pct);
332 define_one_global_rw(min_perf_pct);
333 
334 static struct attribute *intel_pstate_attributes[] = {
335 	&no_turbo.attr,
336 	&max_perf_pct.attr,
337 	&min_perf_pct.attr,
338 	NULL
339 };
340 
341 static struct attribute_group intel_pstate_attr_group = {
342 	.attrs = intel_pstate_attributes,
343 };
344 
345 static void __init intel_pstate_sysfs_expose_params(void)
346 {
347 	struct kobject *intel_pstate_kobject;
348 	int rc;
349 
350 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
351 						&cpu_subsys.dev_root->kobj);
352 	BUG_ON(!intel_pstate_kobject);
353 	rc = sysfs_create_group(intel_pstate_kobject, &intel_pstate_attr_group);
354 	BUG_ON(rc);
355 }
356 
357 /************************** sysfs end ************************/
358 static int byt_get_min_pstate(void)
359 {
360 	u64 value;
361 
362 	rdmsrl(BYT_RATIOS, value);
363 	return (value >> 8) & 0x7F;
364 }
365 
366 static int byt_get_max_pstate(void)
367 {
368 	u64 value;
369 
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 
378 	rdmsrl(BYT_TURBO_RATIOS, value);
379 	return value & 0x7F;
380 }
381 
382 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
383 {
384 	u64 val;
385 	int32_t vid_fp;
386 	u32 vid;
387 
388 	val = pstate << 8;
389 	if (limits.no_turbo && !limits.turbo_disabled)
390 		val |= (u64)1 << 32;
391 
392 	vid_fp = cpudata->vid.min + mul_fp(
393 		int_tofp(pstate - cpudata->pstate.min_pstate),
394 		cpudata->vid.ratio);
395 
396 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
397 	vid = fp_toint(vid_fp);
398 
399 	if (pstate > cpudata->pstate.max_pstate)
400 		vid = cpudata->vid.turbo;
401 
402 	val |= vid;
403 
404 	wrmsrl(MSR_IA32_PERF_CTL, val);
405 }
406 
407 static void byt_get_vid(struct cpudata *cpudata)
408 {
409 	u64 value;
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 static int core_get_min_pstate(void)
424 {
425 	u64 value;
426 
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 
435 	rdmsrl(MSR_PLATFORM_INFO, value);
436 	return (value >> 8) & 0xFF;
437 }
438 
439 static int core_get_turbo_pstate(void)
440 {
441 	u64 value;
442 	int nont, ret;
443 
444 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
445 	nont = core_get_max_pstate();
446 	ret = (value) & 255;
447 	if (ret <= nont)
448 		ret = nont;
449 	return ret;
450 }
451 
452 static void core_set_pstate(struct cpudata *cpudata, int pstate)
453 {
454 	u64 val;
455 
456 	val = pstate << 8;
457 	if (limits.no_turbo && !limits.turbo_disabled)
458 		val |= (u64)1 << 32;
459 
460 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
461 }
462 
463 static struct cpu_defaults core_params = {
464 	.pid_policy = {
465 		.sample_rate_ms = 10,
466 		.deadband = 0,
467 		.setpoint = 97,
468 		.p_gain_pct = 20,
469 		.d_gain_pct = 0,
470 		.i_gain_pct = 0,
471 	},
472 	.funcs = {
473 		.get_max = core_get_max_pstate,
474 		.get_min = core_get_min_pstate,
475 		.get_turbo = core_get_turbo_pstate,
476 		.set = core_set_pstate,
477 	},
478 };
479 
480 static struct cpu_defaults byt_params = {
481 	.pid_policy = {
482 		.sample_rate_ms = 10,
483 		.deadband = 0,
484 		.setpoint = 97,
485 		.p_gain_pct = 14,
486 		.d_gain_pct = 0,
487 		.i_gain_pct = 4,
488 	},
489 	.funcs = {
490 		.get_max = byt_get_max_pstate,
491 		.get_min = byt_get_min_pstate,
492 		.get_turbo = byt_get_turbo_pstate,
493 		.set = byt_set_pstate,
494 		.get_vid = byt_get_vid,
495 	},
496 };
497 
498 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
499 {
500 	int max_perf = cpu->pstate.turbo_pstate;
501 	int max_perf_adj;
502 	int min_perf;
503 
504 	if (limits.no_turbo)
505 		max_perf = cpu->pstate.max_pstate;
506 
507 	max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
508 	*max = clamp_t(int, max_perf_adj,
509 			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
510 
511 	min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
512 	*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
513 }
514 
515 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
516 {
517 	int max_perf, min_perf;
518 
519 	intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
520 
521 	pstate = clamp_t(int, pstate, min_perf, max_perf);
522 
523 	if (pstate == cpu->pstate.current_pstate)
524 		return;
525 
526 	trace_cpu_frequency(pstate * 100000, cpu->cpu);
527 
528 	cpu->pstate.current_pstate = pstate;
529 
530 	pstate_funcs.set(cpu, pstate);
531 }
532 
533 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
534 {
535 	cpu->pstate.min_pstate = pstate_funcs.get_min();
536 	cpu->pstate.max_pstate = pstate_funcs.get_max();
537 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
538 
539 	if (pstate_funcs.get_vid)
540 		pstate_funcs.get_vid(cpu);
541 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
542 }
543 
544 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
545 {
546 	struct sample *sample = &cpu->sample;
547 	int64_t core_pct;
548 
549 	core_pct = int_tofp(sample->aperf) * int_tofp(100);
550 	core_pct = div64_u64(core_pct, int_tofp(sample->mperf));
551 
552 	sample->freq = fp_toint(
553 		mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
554 
555 	sample->core_pct_busy = (int32_t)core_pct;
556 }
557 
558 static inline void intel_pstate_sample(struct cpudata *cpu)
559 {
560 	u64 aperf, mperf;
561 	unsigned long flags;
562 
563 	local_irq_save(flags);
564 	rdmsrl(MSR_IA32_APERF, aperf);
565 	rdmsrl(MSR_IA32_MPERF, mperf);
566 	local_irq_restore(flags);
567 
568 	cpu->last_sample_time = cpu->sample.time;
569 	cpu->sample.time = ktime_get();
570 	cpu->sample.aperf = aperf;
571 	cpu->sample.mperf = mperf;
572 	cpu->sample.aperf -= cpu->prev_aperf;
573 	cpu->sample.mperf -= cpu->prev_mperf;
574 
575 	intel_pstate_calc_busy(cpu);
576 
577 	cpu->prev_aperf = aperf;
578 	cpu->prev_mperf = mperf;
579 }
580 
581 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
582 {
583 	int delay;
584 
585 	delay = msecs_to_jiffies(pid_params.sample_rate_ms);
586 	mod_timer_pinned(&cpu->timer, jiffies + delay);
587 }
588 
589 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
590 {
591 	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
592 	u32 duration_us;
593 	u32 sample_time;
594 
595 	core_busy = cpu->sample.core_pct_busy;
596 	max_pstate = int_tofp(cpu->pstate.max_pstate);
597 	current_pstate = int_tofp(cpu->pstate.current_pstate);
598 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
599 
600 	sample_time = pid_params.sample_rate_ms  * USEC_PER_MSEC;
601 	duration_us = (u32) ktime_us_delta(cpu->sample.time,
602 					   cpu->last_sample_time);
603 	if (duration_us > sample_time * 3) {
604 		sample_ratio = div_fp(int_tofp(sample_time),
605 				      int_tofp(duration_us));
606 		core_busy = mul_fp(core_busy, sample_ratio);
607 	}
608 
609 	return core_busy;
610 }
611 
612 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
613 {
614 	int32_t busy_scaled;
615 	struct _pid *pid;
616 	signed int ctl;
617 
618 	pid = &cpu->pid;
619 	busy_scaled = intel_pstate_get_scaled_busy(cpu);
620 
621 	ctl = pid_calc(pid, busy_scaled);
622 
623 	/* Negative values of ctl increase the pstate and vice versa */
624 	intel_pstate_set_pstate(cpu, cpu->pstate.current_pstate - ctl);
625 }
626 
627 static void intel_pstate_timer_func(unsigned long __data)
628 {
629 	struct cpudata *cpu = (struct cpudata *) __data;
630 	struct sample *sample;
631 
632 	intel_pstate_sample(cpu);
633 
634 	sample = &cpu->sample;
635 
636 	intel_pstate_adjust_busy_pstate(cpu);
637 
638 	trace_pstate_sample(fp_toint(sample->core_pct_busy),
639 			fp_toint(intel_pstate_get_scaled_busy(cpu)),
640 			cpu->pstate.current_pstate,
641 			sample->mperf,
642 			sample->aperf,
643 			sample->freq);
644 
645 	intel_pstate_set_sample_time(cpu);
646 }
647 
648 #define ICPU(model, policy) \
649 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
650 			(unsigned long)&policy }
651 
652 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
653 	ICPU(0x2a, core_params),
654 	ICPU(0x2d, core_params),
655 	ICPU(0x37, byt_params),
656 	ICPU(0x3a, core_params),
657 	ICPU(0x3c, core_params),
658 	ICPU(0x3d, core_params),
659 	ICPU(0x3e, core_params),
660 	ICPU(0x3f, core_params),
661 	ICPU(0x45, core_params),
662 	ICPU(0x46, core_params),
663 	ICPU(0x4f, core_params),
664 	ICPU(0x56, core_params),
665 	{}
666 };
667 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
668 
669 static int intel_pstate_init_cpu(unsigned int cpunum)
670 {
671 	struct cpudata *cpu;
672 
673 	all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
674 	if (!all_cpu_data[cpunum])
675 		return -ENOMEM;
676 
677 	cpu = all_cpu_data[cpunum];
678 
679 	cpu->cpu = cpunum;
680 	intel_pstate_get_cpu_pstates(cpu);
681 
682 	init_timer_deferrable(&cpu->timer);
683 	cpu->timer.function = intel_pstate_timer_func;
684 	cpu->timer.data = (unsigned long)cpu;
685 	cpu->timer.expires = jiffies + HZ/100;
686 	intel_pstate_busy_pid_reset(cpu);
687 	intel_pstate_sample(cpu);
688 
689 	add_timer_on(&cpu->timer, cpunum);
690 
691 	pr_info("Intel pstate controlling: cpu %d\n", cpunum);
692 
693 	return 0;
694 }
695 
696 static unsigned int intel_pstate_get(unsigned int cpu_num)
697 {
698 	struct sample *sample;
699 	struct cpudata *cpu;
700 
701 	cpu = all_cpu_data[cpu_num];
702 	if (!cpu)
703 		return 0;
704 	sample = &cpu->sample;
705 	return sample->freq;
706 }
707 
708 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
709 {
710 	struct cpudata *cpu;
711 
712 	cpu = all_cpu_data[policy->cpu];
713 
714 	if (!policy->cpuinfo.max_freq)
715 		return -ENODEV;
716 
717 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
718 		limits.min_perf_pct = 100;
719 		limits.min_perf = int_tofp(1);
720 		limits.max_perf_pct = 100;
721 		limits.max_perf = int_tofp(1);
722 		limits.no_turbo = limits.turbo_disabled;
723 		return 0;
724 	}
725 	limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
726 	limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
727 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
728 
729 	limits.max_policy_pct = (policy->max * 100) / policy->cpuinfo.max_freq;
730 	limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
731 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
732 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
733 
734 	return 0;
735 }
736 
737 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
738 {
739 	cpufreq_verify_within_cpu_limits(policy);
740 
741 	if (policy->policy != CPUFREQ_POLICY_POWERSAVE &&
742 	    policy->policy != CPUFREQ_POLICY_PERFORMANCE)
743 		return -EINVAL;
744 
745 	return 0;
746 }
747 
748 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
749 {
750 	int cpu_num = policy->cpu;
751 	struct cpudata *cpu = all_cpu_data[cpu_num];
752 
753 	pr_info("intel_pstate CPU %d exiting\n", cpu_num);
754 
755 	del_timer_sync(&all_cpu_data[cpu_num]->timer);
756 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
757 	kfree(all_cpu_data[cpu_num]);
758 	all_cpu_data[cpu_num] = NULL;
759 }
760 
761 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
762 {
763 	struct cpudata *cpu;
764 	int rc;
765 	u64 misc_en;
766 
767 	rc = intel_pstate_init_cpu(policy->cpu);
768 	if (rc)
769 		return rc;
770 
771 	cpu = all_cpu_data[policy->cpu];
772 
773 	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
774 	if (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
775 	    cpu->pstate.max_pstate == cpu->pstate.turbo_pstate) {
776 		limits.turbo_disabled = 1;
777 		limits.no_turbo = 1;
778 	}
779 	if (limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
780 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
781 	else
782 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
783 
784 	policy->min = cpu->pstate.min_pstate * 100000;
785 	policy->max = cpu->pstate.turbo_pstate * 100000;
786 
787 	/* cpuinfo and default policy values */
788 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
789 	policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
790 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
791 	cpumask_set_cpu(policy->cpu, policy->cpus);
792 
793 	return 0;
794 }
795 
796 static struct cpufreq_driver intel_pstate_driver = {
797 	.flags		= CPUFREQ_CONST_LOOPS,
798 	.verify		= intel_pstate_verify_policy,
799 	.setpolicy	= intel_pstate_set_policy,
800 	.get		= intel_pstate_get,
801 	.init		= intel_pstate_cpu_init,
802 	.stop_cpu	= intel_pstate_stop_cpu,
803 	.name		= "intel_pstate",
804 };
805 
806 static int __initdata no_load;
807 
808 static int intel_pstate_msrs_not_valid(void)
809 {
810 	/* Check that all the msr's we are using are valid. */
811 	u64 aperf, mperf, tmp;
812 
813 	rdmsrl(MSR_IA32_APERF, aperf);
814 	rdmsrl(MSR_IA32_MPERF, mperf);
815 
816 	if (!pstate_funcs.get_max() ||
817 	    !pstate_funcs.get_min() ||
818 	    !pstate_funcs.get_turbo())
819 		return -ENODEV;
820 
821 	rdmsrl(MSR_IA32_APERF, tmp);
822 	if (!(tmp - aperf))
823 		return -ENODEV;
824 
825 	rdmsrl(MSR_IA32_MPERF, tmp);
826 	if (!(tmp - mperf))
827 		return -ENODEV;
828 
829 	return 0;
830 }
831 
832 static void copy_pid_params(struct pstate_adjust_policy *policy)
833 {
834 	pid_params.sample_rate_ms = policy->sample_rate_ms;
835 	pid_params.p_gain_pct = policy->p_gain_pct;
836 	pid_params.i_gain_pct = policy->i_gain_pct;
837 	pid_params.d_gain_pct = policy->d_gain_pct;
838 	pid_params.deadband = policy->deadband;
839 	pid_params.setpoint = policy->setpoint;
840 }
841 
842 static void copy_cpu_funcs(struct pstate_funcs *funcs)
843 {
844 	pstate_funcs.get_max   = funcs->get_max;
845 	pstate_funcs.get_min   = funcs->get_min;
846 	pstate_funcs.get_turbo = funcs->get_turbo;
847 	pstate_funcs.set       = funcs->set;
848 	pstate_funcs.get_vid   = funcs->get_vid;
849 }
850 
851 #if IS_ENABLED(CONFIG_ACPI)
852 #include <acpi/processor.h>
853 
854 static bool intel_pstate_no_acpi_pss(void)
855 {
856 	int i;
857 
858 	for_each_possible_cpu(i) {
859 		acpi_status status;
860 		union acpi_object *pss;
861 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
862 		struct acpi_processor *pr = per_cpu(processors, i);
863 
864 		if (!pr)
865 			continue;
866 
867 		status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
868 		if (ACPI_FAILURE(status))
869 			continue;
870 
871 		pss = buffer.pointer;
872 		if (pss && pss->type == ACPI_TYPE_PACKAGE) {
873 			kfree(pss);
874 			return false;
875 		}
876 
877 		kfree(pss);
878 	}
879 
880 	return true;
881 }
882 
883 struct hw_vendor_info {
884 	u16  valid;
885 	char oem_id[ACPI_OEM_ID_SIZE];
886 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
887 };
888 
889 /* Hardware vendor-specific info that has its own power management modes */
890 static struct hw_vendor_info vendor_info[] = {
891 	{1, "HP    ", "ProLiant"},
892 	{0, "", ""},
893 };
894 
895 static bool intel_pstate_platform_pwr_mgmt_exists(void)
896 {
897 	struct acpi_table_header hdr;
898 	struct hw_vendor_info *v_info;
899 
900 	if (acpi_disabled ||
901 	    ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
902 		return false;
903 
904 	for (v_info = vendor_info; v_info->valid; v_info++) {
905 		if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) &&
906 		    !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) &&
907 		    intel_pstate_no_acpi_pss())
908 			return true;
909 	}
910 
911 	return false;
912 }
913 #else /* CONFIG_ACPI not enabled */
914 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
915 #endif /* CONFIG_ACPI */
916 
917 static int __init intel_pstate_init(void)
918 {
919 	int cpu, rc = 0;
920 	const struct x86_cpu_id *id;
921 	struct cpu_defaults *cpu_info;
922 
923 	if (no_load)
924 		return -ENODEV;
925 
926 	id = x86_match_cpu(intel_pstate_cpu_ids);
927 	if (!id)
928 		return -ENODEV;
929 
930 	/*
931 	 * The Intel pstate driver will be ignored if the platform
932 	 * firmware has its own power management modes.
933 	 */
934 	if (intel_pstate_platform_pwr_mgmt_exists())
935 		return -ENODEV;
936 
937 	cpu_info = (struct cpu_defaults *)id->driver_data;
938 
939 	copy_pid_params(&cpu_info->pid_policy);
940 	copy_cpu_funcs(&cpu_info->funcs);
941 
942 	if (intel_pstate_msrs_not_valid())
943 		return -ENODEV;
944 
945 	pr_info("Intel P-state driver initializing.\n");
946 
947 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
948 	if (!all_cpu_data)
949 		return -ENOMEM;
950 
951 	rc = cpufreq_register_driver(&intel_pstate_driver);
952 	if (rc)
953 		goto out;
954 
955 	intel_pstate_debug_expose_params();
956 	intel_pstate_sysfs_expose_params();
957 
958 	return rc;
959 out:
960 	get_online_cpus();
961 	for_each_online_cpu(cpu) {
962 		if (all_cpu_data[cpu]) {
963 			del_timer_sync(&all_cpu_data[cpu]->timer);
964 			kfree(all_cpu_data[cpu]);
965 		}
966 	}
967 
968 	put_online_cpus();
969 	vfree(all_cpu_data);
970 	return -ENODEV;
971 }
972 device_initcall(intel_pstate_init);
973 
974 static int __init intel_pstate_setup(char *str)
975 {
976 	if (!str)
977 		return -EINVAL;
978 
979 	if (!strcmp(str, "disable"))
980 		no_load = 1;
981 	return 0;
982 }
983 early_param("intel_pstate", intel_pstate_setup);
984 
985 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
986 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
987 MODULE_LICENSE("GPL");
988