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 SAMPLE_COUNT		3
36 
37 #define BYT_RATIOS		0x66a
38 #define BYT_VIDS		0x66b
39 #define BYT_TURBO_RATIOS	0x66c
40 
41 
42 #define FRAC_BITS 6
43 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
44 #define fp_toint(X) ((X) >> FRAC_BITS)
45 #define FP_ROUNDUP(X) ((X) += 1 << FRAC_BITS)
46 
47 static inline int32_t mul_fp(int32_t x, int32_t y)
48 {
49 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
50 }
51 
52 static inline int32_t div_fp(int32_t x, int32_t y)
53 {
54 	return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
55 }
56 
57 struct sample {
58 	int32_t core_pct_busy;
59 	u64 aperf;
60 	u64 mperf;
61 	unsigned long long tsc;
62 	int freq;
63 };
64 
65 struct pstate_data {
66 	int	current_pstate;
67 	int	min_pstate;
68 	int	max_pstate;
69 	int	turbo_pstate;
70 };
71 
72 struct vid_data {
73 	int32_t min;
74 	int32_t max;
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 	char name[64];
92 
93 	struct timer_list timer;
94 
95 	struct pstate_data pstate;
96 	struct vid_data vid;
97 	struct _pid pid;
98 
99 	u64	prev_aperf;
100 	u64	prev_mperf;
101 	unsigned long long prev_tsc;
102 	struct sample sample;
103 };
104 
105 static struct cpudata **all_cpu_data;
106 struct pstate_adjust_policy {
107 	int sample_rate_ms;
108 	int deadband;
109 	int setpoint;
110 	int p_gain_pct;
111 	int d_gain_pct;
112 	int i_gain_pct;
113 };
114 
115 struct pstate_funcs {
116 	int (*get_max)(void);
117 	int (*get_min)(void);
118 	int (*get_turbo)(void);
119 	void (*set)(struct cpudata*, int pstate);
120 	void (*get_vid)(struct cpudata *);
121 };
122 
123 struct cpu_defaults {
124 	struct pstate_adjust_policy pid_policy;
125 	struct pstate_funcs funcs;
126 };
127 
128 static struct pstate_adjust_policy pid_params;
129 static struct pstate_funcs pstate_funcs;
130 
131 struct perf_limits {
132 	int no_turbo;
133 	int max_perf_pct;
134 	int min_perf_pct;
135 	int32_t max_perf;
136 	int32_t min_perf;
137 	int max_policy_pct;
138 	int max_sysfs_pct;
139 };
140 
141 static struct perf_limits limits = {
142 	.no_turbo = 0,
143 	.max_perf_pct = 100,
144 	.max_perf = int_tofp(1),
145 	.min_perf_pct = 0,
146 	.min_perf = 0,
147 	.max_policy_pct = 100,
148 	.max_sysfs_pct = 100,
149 };
150 
151 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
152 			int deadband, int integral) {
153 	pid->setpoint = setpoint;
154 	pid->deadband  = deadband;
155 	pid->integral  = int_tofp(integral);
156 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
157 }
158 
159 static inline void pid_p_gain_set(struct _pid *pid, int percent)
160 {
161 	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
162 }
163 
164 static inline void pid_i_gain_set(struct _pid *pid, int percent)
165 {
166 	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
167 }
168 
169 static inline void pid_d_gain_set(struct _pid *pid, int percent)
170 {
171 
172 	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
173 }
174 
175 static signed int pid_calc(struct _pid *pid, int32_t busy)
176 {
177 	signed int result;
178 	int32_t pterm, dterm, fp_error;
179 	int32_t integral_limit;
180 
181 	fp_error = int_tofp(pid->setpoint) - busy;
182 
183 	if (abs(fp_error) <= int_tofp(pid->deadband))
184 		return 0;
185 
186 	pterm = mul_fp(pid->p_gain, fp_error);
187 
188 	pid->integral += fp_error;
189 
190 	/* limit the integral term */
191 	integral_limit = int_tofp(30);
192 	if (pid->integral > integral_limit)
193 		pid->integral = integral_limit;
194 	if (pid->integral < -integral_limit)
195 		pid->integral = -integral_limit;
196 
197 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
198 	pid->last_err = fp_error;
199 
200 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
201 
202 	return (signed int)fp_toint(result);
203 }
204 
205 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
206 {
207 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
208 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
209 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
210 
211 	pid_reset(&cpu->pid,
212 		pid_params.setpoint,
213 		100,
214 		pid_params.deadband,
215 		0);
216 }
217 
218 static inline void intel_pstate_reset_all_pid(void)
219 {
220 	unsigned int cpu;
221 	for_each_online_cpu(cpu) {
222 		if (all_cpu_data[cpu])
223 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
224 	}
225 }
226 
227 /************************** debugfs begin ************************/
228 static int pid_param_set(void *data, u64 val)
229 {
230 	*(u32 *)data = val;
231 	intel_pstate_reset_all_pid();
232 	return 0;
233 }
234 static int pid_param_get(void *data, u64 *val)
235 {
236 	*val = *(u32 *)data;
237 	return 0;
238 }
239 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
240 			pid_param_set, "%llu\n");
241 
242 struct pid_param {
243 	char *name;
244 	void *value;
245 };
246 
247 static struct pid_param pid_files[] = {
248 	{"sample_rate_ms", &pid_params.sample_rate_ms},
249 	{"d_gain_pct", &pid_params.d_gain_pct},
250 	{"i_gain_pct", &pid_params.i_gain_pct},
251 	{"deadband", &pid_params.deadband},
252 	{"setpoint", &pid_params.setpoint},
253 	{"p_gain_pct", &pid_params.p_gain_pct},
254 	{NULL, NULL}
255 };
256 
257 static struct dentry *debugfs_parent;
258 static void intel_pstate_debug_expose_params(void)
259 {
260 	int i = 0;
261 
262 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
263 	if (IS_ERR_OR_NULL(debugfs_parent))
264 		return;
265 	while (pid_files[i].name) {
266 		debugfs_create_file(pid_files[i].name, 0660,
267 				debugfs_parent, pid_files[i].value,
268 				&fops_pid_param);
269 		i++;
270 	}
271 }
272 
273 /************************** debugfs end ************************/
274 
275 /************************** sysfs begin ************************/
276 #define show_one(file_name, object)					\
277 	static ssize_t show_##file_name					\
278 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
279 	{								\
280 		return sprintf(buf, "%u\n", limits.object);		\
281 	}
282 
283 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
284 				const char *buf, size_t count)
285 {
286 	unsigned int input;
287 	int ret;
288 	ret = sscanf(buf, "%u", &input);
289 	if (ret != 1)
290 		return -EINVAL;
291 	limits.no_turbo = clamp_t(int, input, 0 , 1);
292 
293 	return count;
294 }
295 
296 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
297 				const char *buf, size_t count)
298 {
299 	unsigned int input;
300 	int ret;
301 	ret = sscanf(buf, "%u", &input);
302 	if (ret != 1)
303 		return -EINVAL;
304 
305 	limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
306 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
307 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
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 	ret = sscanf(buf, "%u", &input);
317 	if (ret != 1)
318 		return -EINVAL;
319 	limits.min_perf_pct = clamp_t(int, input, 0 , 100);
320 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
321 
322 	return count;
323 }
324 
325 show_one(no_turbo, no_turbo);
326 show_one(max_perf_pct, max_perf_pct);
327 show_one(min_perf_pct, min_perf_pct);
328 
329 define_one_global_rw(no_turbo);
330 define_one_global_rw(max_perf_pct);
331 define_one_global_rw(min_perf_pct);
332 
333 static struct attribute *intel_pstate_attributes[] = {
334 	&no_turbo.attr,
335 	&max_perf_pct.attr,
336 	&min_perf_pct.attr,
337 	NULL
338 };
339 
340 static struct attribute_group intel_pstate_attr_group = {
341 	.attrs = intel_pstate_attributes,
342 };
343 static struct kobject *intel_pstate_kobject;
344 
345 static void intel_pstate_sysfs_expose_params(void)
346 {
347 	int rc;
348 
349 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
350 						&cpu_subsys.dev_root->kobj);
351 	BUG_ON(!intel_pstate_kobject);
352 	rc = sysfs_create_group(intel_pstate_kobject,
353 				&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 	rdmsrl(BYT_RATIOS, value);
362 	return (value >> 8) & 0xFF;
363 }
364 
365 static int byt_get_max_pstate(void)
366 {
367 	u64 value;
368 	rdmsrl(BYT_RATIOS, value);
369 	return (value >> 16) & 0xFF;
370 }
371 
372 static int byt_get_turbo_pstate(void)
373 {
374 	u64 value;
375 	rdmsrl(BYT_TURBO_RATIOS, value);
376 	return value & 0x3F;
377 }
378 
379 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
380 {
381 	u64 val;
382 	int32_t vid_fp;
383 	u32 vid;
384 
385 	val = pstate << 8;
386 	if (limits.no_turbo)
387 		val |= (u64)1 << 32;
388 
389 	vid_fp = cpudata->vid.min + mul_fp(
390 		int_tofp(pstate - cpudata->pstate.min_pstate),
391 		cpudata->vid.ratio);
392 
393 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
394 	vid = fp_toint(vid_fp);
395 
396 	val |= vid;
397 
398 	wrmsrl(MSR_IA32_PERF_CTL, val);
399 }
400 
401 static void byt_get_vid(struct cpudata *cpudata)
402 {
403 	u64 value;
404 
405 	rdmsrl(BYT_VIDS, value);
406 	cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
407 	cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
408 	cpudata->vid.ratio = div_fp(
409 		cpudata->vid.max - cpudata->vid.min,
410 		int_tofp(cpudata->pstate.max_pstate -
411 			cpudata->pstate.min_pstate));
412 }
413 
414 
415 static int core_get_min_pstate(void)
416 {
417 	u64 value;
418 	rdmsrl(MSR_PLATFORM_INFO, value);
419 	return (value >> 40) & 0xFF;
420 }
421 
422 static int core_get_max_pstate(void)
423 {
424 	u64 value;
425 	rdmsrl(MSR_PLATFORM_INFO, value);
426 	return (value >> 8) & 0xFF;
427 }
428 
429 static int core_get_turbo_pstate(void)
430 {
431 	u64 value;
432 	int nont, ret;
433 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
434 	nont = core_get_max_pstate();
435 	ret = ((value) & 255);
436 	if (ret <= nont)
437 		ret = nont;
438 	return ret;
439 }
440 
441 static void core_set_pstate(struct cpudata *cpudata, int pstate)
442 {
443 	u64 val;
444 
445 	val = pstate << 8;
446 	if (limits.no_turbo)
447 		val |= (u64)1 << 32;
448 
449 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
450 }
451 
452 static struct cpu_defaults core_params = {
453 	.pid_policy = {
454 		.sample_rate_ms = 10,
455 		.deadband = 0,
456 		.setpoint = 97,
457 		.p_gain_pct = 20,
458 		.d_gain_pct = 0,
459 		.i_gain_pct = 0,
460 	},
461 	.funcs = {
462 		.get_max = core_get_max_pstate,
463 		.get_min = core_get_min_pstate,
464 		.get_turbo = core_get_turbo_pstate,
465 		.set = core_set_pstate,
466 	},
467 };
468 
469 static struct cpu_defaults byt_params = {
470 	.pid_policy = {
471 		.sample_rate_ms = 10,
472 		.deadband = 0,
473 		.setpoint = 97,
474 		.p_gain_pct = 14,
475 		.d_gain_pct = 0,
476 		.i_gain_pct = 4,
477 	},
478 	.funcs = {
479 		.get_max = byt_get_max_pstate,
480 		.get_min = byt_get_min_pstate,
481 		.get_turbo = byt_get_turbo_pstate,
482 		.set = byt_set_pstate,
483 		.get_vid = byt_get_vid,
484 	},
485 };
486 
487 
488 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
489 {
490 	int max_perf = cpu->pstate.turbo_pstate;
491 	int max_perf_adj;
492 	int min_perf;
493 	if (limits.no_turbo)
494 		max_perf = cpu->pstate.max_pstate;
495 
496 	max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
497 	*max = clamp_t(int, max_perf_adj,
498 			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
499 
500 	min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
501 	*min = clamp_t(int, min_perf,
502 			cpu->pstate.min_pstate, max_perf);
503 }
504 
505 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
506 {
507 	int max_perf, min_perf;
508 
509 	intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
510 
511 	pstate = clamp_t(int, pstate, min_perf, max_perf);
512 
513 	if (pstate == cpu->pstate.current_pstate)
514 		return;
515 
516 	trace_cpu_frequency(pstate * 100000, cpu->cpu);
517 
518 	cpu->pstate.current_pstate = pstate;
519 
520 	pstate_funcs.set(cpu, pstate);
521 }
522 
523 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
524 {
525 	int target;
526 	target = cpu->pstate.current_pstate + steps;
527 
528 	intel_pstate_set_pstate(cpu, target);
529 }
530 
531 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
532 {
533 	int target;
534 	target = cpu->pstate.current_pstate - steps;
535 	intel_pstate_set_pstate(cpu, target);
536 }
537 
538 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
539 {
540 	sprintf(cpu->name, "Intel 2nd generation core");
541 
542 	cpu->pstate.min_pstate = pstate_funcs.get_min();
543 	cpu->pstate.max_pstate = pstate_funcs.get_max();
544 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
545 
546 	if (pstate_funcs.get_vid)
547 		pstate_funcs.get_vid(cpu);
548 
549 	/*
550 	 * goto max pstate so we don't slow up boot if we are built-in if we are
551 	 * a module we will take care of it during normal operation
552 	 */
553 	intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
554 }
555 
556 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
557 					struct sample *sample)
558 {
559 	int32_t core_pct;
560 	int32_t c0_pct;
561 
562 	core_pct = div_fp(int_tofp((sample->aperf)),
563 			int_tofp((sample->mperf)));
564 	core_pct = mul_fp(core_pct, int_tofp(100));
565 	FP_ROUNDUP(core_pct);
566 
567 	c0_pct = div_fp(int_tofp(sample->mperf), int_tofp(sample->tsc));
568 
569 	sample->freq = fp_toint(
570 		mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
571 
572 	sample->core_pct_busy = mul_fp(core_pct, c0_pct);
573 }
574 
575 static inline void intel_pstate_sample(struct cpudata *cpu)
576 {
577 	u64 aperf, mperf;
578 	unsigned long long tsc;
579 
580 	rdmsrl(MSR_IA32_APERF, aperf);
581 	rdmsrl(MSR_IA32_MPERF, mperf);
582 	tsc = native_read_tsc();
583 
584 	aperf = aperf >> FRAC_BITS;
585 	mperf = mperf >> FRAC_BITS;
586 	tsc = tsc >> FRAC_BITS;
587 
588 	cpu->sample.aperf = aperf;
589 	cpu->sample.mperf = mperf;
590 	cpu->sample.tsc = tsc;
591 	cpu->sample.aperf -= cpu->prev_aperf;
592 	cpu->sample.mperf -= cpu->prev_mperf;
593 	cpu->sample.tsc -= cpu->prev_tsc;
594 
595 	intel_pstate_calc_busy(cpu, &cpu->sample);
596 
597 	cpu->prev_aperf = aperf;
598 	cpu->prev_mperf = mperf;
599 	cpu->prev_tsc = tsc;
600 }
601 
602 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
603 {
604 	int sample_time, delay;
605 
606 	sample_time = pid_params.sample_rate_ms;
607 	delay = msecs_to_jiffies(sample_time);
608 	mod_timer_pinned(&cpu->timer, jiffies + delay);
609 }
610 
611 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
612 {
613 	int32_t core_busy, max_pstate, current_pstate;
614 
615 	core_busy = cpu->sample.core_pct_busy;
616 	max_pstate = int_tofp(cpu->pstate.max_pstate);
617 	current_pstate = int_tofp(cpu->pstate.current_pstate);
618 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
619 	return FP_ROUNDUP(core_busy);
620 }
621 
622 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
623 {
624 	int32_t busy_scaled;
625 	struct _pid *pid;
626 	signed int ctl = 0;
627 	int steps;
628 
629 	pid = &cpu->pid;
630 	busy_scaled = intel_pstate_get_scaled_busy(cpu);
631 
632 	ctl = pid_calc(pid, busy_scaled);
633 
634 	steps = abs(ctl);
635 
636 	if (ctl < 0)
637 		intel_pstate_pstate_increase(cpu, steps);
638 	else
639 		intel_pstate_pstate_decrease(cpu, steps);
640 }
641 
642 static void intel_pstate_timer_func(unsigned long __data)
643 {
644 	struct cpudata *cpu = (struct cpudata *) __data;
645 	struct sample *sample;
646 
647 	intel_pstate_sample(cpu);
648 
649 	sample = &cpu->sample;
650 
651 	intel_pstate_adjust_busy_pstate(cpu);
652 
653 	trace_pstate_sample(fp_toint(sample->core_pct_busy),
654 			fp_toint(intel_pstate_get_scaled_busy(cpu)),
655 			cpu->pstate.current_pstate,
656 			sample->mperf,
657 			sample->aperf,
658 			sample->freq);
659 
660 	intel_pstate_set_sample_time(cpu);
661 }
662 
663 #define ICPU(model, policy) \
664 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
665 			(unsigned long)&policy }
666 
667 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
668 	ICPU(0x2a, core_params),
669 	ICPU(0x2d, core_params),
670 	ICPU(0x37, byt_params),
671 	ICPU(0x3a, core_params),
672 	ICPU(0x3c, core_params),
673 	ICPU(0x3e, core_params),
674 	ICPU(0x3f, core_params),
675 	ICPU(0x45, core_params),
676 	ICPU(0x46, core_params),
677 	{}
678 };
679 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
680 
681 static int intel_pstate_init_cpu(unsigned int cpunum)
682 {
683 
684 	const struct x86_cpu_id *id;
685 	struct cpudata *cpu;
686 
687 	id = x86_match_cpu(intel_pstate_cpu_ids);
688 	if (!id)
689 		return -ENODEV;
690 
691 	all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
692 	if (!all_cpu_data[cpunum])
693 		return -ENOMEM;
694 
695 	cpu = all_cpu_data[cpunum];
696 
697 	intel_pstate_get_cpu_pstates(cpu);
698 	if (!cpu->pstate.current_pstate) {
699 		all_cpu_data[cpunum] = NULL;
700 		kfree(cpu);
701 		return -ENODATA;
702 	}
703 
704 	cpu->cpu = cpunum;
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 	intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
714 
715 	add_timer_on(&cpu->timer, cpunum);
716 
717 	pr_info("Intel pstate controlling: cpu %d\n", cpunum);
718 
719 	return 0;
720 }
721 
722 static unsigned int intel_pstate_get(unsigned int cpu_num)
723 {
724 	struct sample *sample;
725 	struct cpudata *cpu;
726 
727 	cpu = all_cpu_data[cpu_num];
728 	if (!cpu)
729 		return 0;
730 	sample = &cpu->sample;
731 	return sample->freq;
732 }
733 
734 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
735 {
736 	struct cpudata *cpu;
737 
738 	cpu = all_cpu_data[policy->cpu];
739 
740 	if (!policy->cpuinfo.max_freq)
741 		return -ENODEV;
742 
743 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
744 		limits.min_perf_pct = 100;
745 		limits.min_perf = int_tofp(1);
746 		limits.max_perf_pct = 100;
747 		limits.max_perf = int_tofp(1);
748 		limits.no_turbo = 0;
749 		return 0;
750 	}
751 	limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
752 	limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
753 	limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
754 
755 	limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
756 	limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
757 	limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
758 	limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
759 
760 	return 0;
761 }
762 
763 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
764 {
765 	cpufreq_verify_within_cpu_limits(policy);
766 
767 	if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
768 		(policy->policy != CPUFREQ_POLICY_PERFORMANCE))
769 		return -EINVAL;
770 
771 	return 0;
772 }
773 
774 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
775 {
776 	int cpu_num = policy->cpu;
777 	struct cpudata *cpu = all_cpu_data[cpu_num];
778 
779 	pr_info("intel_pstate CPU %d exiting\n", cpu_num);
780 
781 	del_timer_sync(&all_cpu_data[cpu_num]->timer);
782 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
783 	kfree(all_cpu_data[cpu_num]);
784 	all_cpu_data[cpu_num] = NULL;
785 }
786 
787 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
788 {
789 	struct cpudata *cpu;
790 	int rc;
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 	if (!limits.no_turbo &&
799 		limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
800 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
801 	else
802 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
803 
804 	policy->min = cpu->pstate.min_pstate * 100000;
805 	policy->max = cpu->pstate.turbo_pstate * 100000;
806 
807 	/* cpuinfo and default policy values */
808 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
809 	policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
810 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
811 	cpumask_set_cpu(policy->cpu, policy->cpus);
812 
813 	return 0;
814 }
815 
816 static struct cpufreq_driver intel_pstate_driver = {
817 	.flags		= CPUFREQ_CONST_LOOPS,
818 	.verify		= intel_pstate_verify_policy,
819 	.setpolicy	= intel_pstate_set_policy,
820 	.get		= intel_pstate_get,
821 	.init		= intel_pstate_cpu_init,
822 	.stop_cpu	= intel_pstate_stop_cpu,
823 	.name		= "intel_pstate",
824 };
825 
826 static int __initdata no_load;
827 
828 static int intel_pstate_msrs_not_valid(void)
829 {
830 	/* Check that all the msr's we are using are valid. */
831 	u64 aperf, mperf, tmp;
832 
833 	rdmsrl(MSR_IA32_APERF, aperf);
834 	rdmsrl(MSR_IA32_MPERF, mperf);
835 
836 	if (!pstate_funcs.get_max() ||
837 		!pstate_funcs.get_min() ||
838 		!pstate_funcs.get_turbo())
839 		return -ENODEV;
840 
841 	rdmsrl(MSR_IA32_APERF, tmp);
842 	if (!(tmp - aperf))
843 		return -ENODEV;
844 
845 	rdmsrl(MSR_IA32_MPERF, tmp);
846 	if (!(tmp - mperf))
847 		return -ENODEV;
848 
849 	return 0;
850 }
851 
852 static void copy_pid_params(struct pstate_adjust_policy *policy)
853 {
854 	pid_params.sample_rate_ms = policy->sample_rate_ms;
855 	pid_params.p_gain_pct = policy->p_gain_pct;
856 	pid_params.i_gain_pct = policy->i_gain_pct;
857 	pid_params.d_gain_pct = policy->d_gain_pct;
858 	pid_params.deadband = policy->deadband;
859 	pid_params.setpoint = policy->setpoint;
860 }
861 
862 static void copy_cpu_funcs(struct pstate_funcs *funcs)
863 {
864 	pstate_funcs.get_max   = funcs->get_max;
865 	pstate_funcs.get_min   = funcs->get_min;
866 	pstate_funcs.get_turbo = funcs->get_turbo;
867 	pstate_funcs.set       = funcs->set;
868 	pstate_funcs.get_vid   = funcs->get_vid;
869 }
870 
871 #if IS_ENABLED(CONFIG_ACPI)
872 #include <acpi/processor.h>
873 
874 static bool intel_pstate_no_acpi_pss(void)
875 {
876 	int i;
877 
878 	for_each_possible_cpu(i) {
879 		acpi_status status;
880 		union acpi_object *pss;
881 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
882 		struct acpi_processor *pr = per_cpu(processors, i);
883 
884 		if (!pr)
885 			continue;
886 
887 		status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
888 		if (ACPI_FAILURE(status))
889 			continue;
890 
891 		pss = buffer.pointer;
892 		if (pss && pss->type == ACPI_TYPE_PACKAGE) {
893 			kfree(pss);
894 			return false;
895 		}
896 
897 		kfree(pss);
898 	}
899 
900 	return true;
901 }
902 
903 struct hw_vendor_info {
904 	u16  valid;
905 	char oem_id[ACPI_OEM_ID_SIZE];
906 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
907 };
908 
909 /* Hardware vendor-specific info that has its own power management modes */
910 static struct hw_vendor_info vendor_info[] = {
911 	{1, "HP    ", "ProLiant"},
912 	{0, "", ""},
913 };
914 
915 static bool intel_pstate_platform_pwr_mgmt_exists(void)
916 {
917 	struct acpi_table_header hdr;
918 	struct hw_vendor_info *v_info;
919 
920 	if (acpi_disabled
921 	    || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
922 		return false;
923 
924 	for (v_info = vendor_info; v_info->valid; v_info++) {
925 		if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
926 		    && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
927 		    && intel_pstate_no_acpi_pss())
928 			return true;
929 	}
930 
931 	return false;
932 }
933 #else /* CONFIG_ACPI not enabled */
934 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
935 #endif /* CONFIG_ACPI */
936 
937 static int __init intel_pstate_init(void)
938 {
939 	int cpu, rc = 0;
940 	const struct x86_cpu_id *id;
941 	struct cpu_defaults *cpu_info;
942 
943 	if (no_load)
944 		return -ENODEV;
945 
946 	id = x86_match_cpu(intel_pstate_cpu_ids);
947 	if (!id)
948 		return -ENODEV;
949 
950 	/*
951 	 * The Intel pstate driver will be ignored if the platform
952 	 * firmware has its own power management modes.
953 	 */
954 	if (intel_pstate_platform_pwr_mgmt_exists())
955 		return -ENODEV;
956 
957 	cpu_info = (struct cpu_defaults *)id->driver_data;
958 
959 	copy_pid_params(&cpu_info->pid_policy);
960 	copy_cpu_funcs(&cpu_info->funcs);
961 
962 	if (intel_pstate_msrs_not_valid())
963 		return -ENODEV;
964 
965 	pr_info("Intel P-state driver initializing.\n");
966 
967 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
968 	if (!all_cpu_data)
969 		return -ENOMEM;
970 
971 	rc = cpufreq_register_driver(&intel_pstate_driver);
972 	if (rc)
973 		goto out;
974 
975 	intel_pstate_debug_expose_params();
976 	intel_pstate_sysfs_expose_params();
977 
978 	return rc;
979 out:
980 	get_online_cpus();
981 	for_each_online_cpu(cpu) {
982 		if (all_cpu_data[cpu]) {
983 			del_timer_sync(&all_cpu_data[cpu]->timer);
984 			kfree(all_cpu_data[cpu]);
985 		}
986 	}
987 
988 	put_online_cpus();
989 	vfree(all_cpu_data);
990 	return -ENODEV;
991 }
992 device_initcall(intel_pstate_init);
993 
994 static int __init intel_pstate_setup(char *str)
995 {
996 	if (!str)
997 		return -EINVAL;
998 
999 	if (!strcmp(str, "disable"))
1000 		no_load = 1;
1001 	return 0;
1002 }
1003 early_param("intel_pstate", intel_pstate_setup);
1004 
1005 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1006 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1007 MODULE_LICENSE("GPL");
1008