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 <linux/vmalloc.h>
30 #include <trace/events/power.h>
31 
32 #include <asm/div64.h>
33 #include <asm/msr.h>
34 #include <asm/cpu_device_id.h>
35 #include <asm/cpufeature.h>
36 
37 #define ATOM_RATIOS		0x66a
38 #define ATOM_VIDS		0x66b
39 #define ATOM_TURBO_RATIOS	0x66c
40 #define ATOM_TURBO_VIDS		0x66d
41 
42 #define FRAC_BITS 8
43 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
44 #define fp_toint(X) ((X) >> FRAC_BITS)
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(s64 x, s64 y)
52 {
53 	return div64_s64((int64_t)x << FRAC_BITS, y);
54 }
55 
56 static inline int ceiling_fp(int32_t x)
57 {
58 	int mask, ret;
59 
60 	ret = fp_toint(x);
61 	mask = (1 << FRAC_BITS) - 1;
62 	if (x & mask)
63 		ret += 1;
64 	return ret;
65 }
66 
67 struct sample {
68 	int32_t core_pct_busy;
69 	u64 aperf;
70 	u64 mperf;
71 	u64 tsc;
72 	int freq;
73 	ktime_t time;
74 };
75 
76 struct pstate_data {
77 	int	current_pstate;
78 	int	min_pstate;
79 	int	max_pstate;
80 	int	max_pstate_physical;
81 	int	scaling;
82 	int	turbo_pstate;
83 };
84 
85 struct vid_data {
86 	int min;
87 	int max;
88 	int turbo;
89 	int32_t ratio;
90 };
91 
92 struct _pid {
93 	int setpoint;
94 	int32_t integral;
95 	int32_t p_gain;
96 	int32_t i_gain;
97 	int32_t d_gain;
98 	int deadband;
99 	int32_t last_err;
100 };
101 
102 struct cpudata {
103 	int cpu;
104 
105 	struct timer_list timer;
106 
107 	struct pstate_data pstate;
108 	struct vid_data vid;
109 	struct _pid pid;
110 
111 	ktime_t last_sample_time;
112 	u64	prev_aperf;
113 	u64	prev_mperf;
114 	u64	prev_tsc;
115 	struct sample sample;
116 };
117 
118 static struct cpudata **all_cpu_data;
119 struct pstate_adjust_policy {
120 	int sample_rate_ms;
121 	int deadband;
122 	int setpoint;
123 	int p_gain_pct;
124 	int d_gain_pct;
125 	int i_gain_pct;
126 };
127 
128 struct pstate_funcs {
129 	int (*get_max)(void);
130 	int (*get_max_physical)(void);
131 	int (*get_min)(void);
132 	int (*get_turbo)(void);
133 	int (*get_scaling)(void);
134 	void (*set)(struct cpudata*, int pstate);
135 	void (*get_vid)(struct cpudata *);
136 };
137 
138 struct cpu_defaults {
139 	struct pstate_adjust_policy pid_policy;
140 	struct pstate_funcs funcs;
141 };
142 
143 static struct pstate_adjust_policy pid_params;
144 static struct pstate_funcs pstate_funcs;
145 static int hwp_active;
146 
147 struct perf_limits {
148 	int no_turbo;
149 	int turbo_disabled;
150 	int max_perf_pct;
151 	int min_perf_pct;
152 	int32_t max_perf;
153 	int32_t min_perf;
154 	int max_policy_pct;
155 	int max_sysfs_pct;
156 	int min_policy_pct;
157 	int min_sysfs_pct;
158 };
159 
160 static struct perf_limits performance_limits = {
161 	.no_turbo = 0,
162 	.turbo_disabled = 0,
163 	.max_perf_pct = 100,
164 	.max_perf = int_tofp(1),
165 	.min_perf_pct = 100,
166 	.min_perf = int_tofp(1),
167 	.max_policy_pct = 100,
168 	.max_sysfs_pct = 100,
169 	.min_policy_pct = 0,
170 	.min_sysfs_pct = 0,
171 };
172 
173 static struct perf_limits powersave_limits = {
174 	.no_turbo = 0,
175 	.turbo_disabled = 0,
176 	.max_perf_pct = 100,
177 	.max_perf = int_tofp(1),
178 	.min_perf_pct = 0,
179 	.min_perf = 0,
180 	.max_policy_pct = 100,
181 	.max_sysfs_pct = 100,
182 	.min_policy_pct = 0,
183 	.min_sysfs_pct = 0,
184 };
185 
186 #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE
187 static struct perf_limits *limits = &performance_limits;
188 #else
189 static struct perf_limits *limits = &powersave_limits;
190 #endif
191 
192 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
193 			     int deadband, int integral) {
194 	pid->setpoint = setpoint;
195 	pid->deadband  = deadband;
196 	pid->integral  = int_tofp(integral);
197 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
198 }
199 
200 static inline void pid_p_gain_set(struct _pid *pid, int percent)
201 {
202 	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
203 }
204 
205 static inline void pid_i_gain_set(struct _pid *pid, int percent)
206 {
207 	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
208 }
209 
210 static inline void pid_d_gain_set(struct _pid *pid, int percent)
211 {
212 	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
213 }
214 
215 static signed int pid_calc(struct _pid *pid, int32_t busy)
216 {
217 	signed int result;
218 	int32_t pterm, dterm, fp_error;
219 	int32_t integral_limit;
220 
221 	fp_error = int_tofp(pid->setpoint) - busy;
222 
223 	if (abs(fp_error) <= int_tofp(pid->deadband))
224 		return 0;
225 
226 	pterm = mul_fp(pid->p_gain, fp_error);
227 
228 	pid->integral += fp_error;
229 
230 	/*
231 	 * We limit the integral here so that it will never
232 	 * get higher than 30.  This prevents it from becoming
233 	 * too large an input over long periods of time and allows
234 	 * it to get factored out sooner.
235 	 *
236 	 * The value of 30 was chosen through experimentation.
237 	 */
238 	integral_limit = int_tofp(30);
239 	if (pid->integral > integral_limit)
240 		pid->integral = integral_limit;
241 	if (pid->integral < -integral_limit)
242 		pid->integral = -integral_limit;
243 
244 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
245 	pid->last_err = fp_error;
246 
247 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
248 	result = result + (1 << (FRAC_BITS-1));
249 	return (signed int)fp_toint(result);
250 }
251 
252 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
253 {
254 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
255 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
256 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
257 
258 	pid_reset(&cpu->pid, pid_params.setpoint, 100, pid_params.deadband, 0);
259 }
260 
261 static inline void intel_pstate_reset_all_pid(void)
262 {
263 	unsigned int cpu;
264 
265 	for_each_online_cpu(cpu) {
266 		if (all_cpu_data[cpu])
267 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
268 	}
269 }
270 
271 static inline void update_turbo_state(void)
272 {
273 	u64 misc_en;
274 	struct cpudata *cpu;
275 
276 	cpu = all_cpu_data[0];
277 	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
278 	limits->turbo_disabled =
279 		(misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
280 		 cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
281 }
282 
283 static void intel_pstate_hwp_set(void)
284 {
285 	int min, hw_min, max, hw_max, cpu, range, adj_range;
286 	u64 value, cap;
287 
288 	rdmsrl(MSR_HWP_CAPABILITIES, cap);
289 	hw_min = HWP_LOWEST_PERF(cap);
290 	hw_max = HWP_HIGHEST_PERF(cap);
291 	range = hw_max - hw_min;
292 
293 	get_online_cpus();
294 
295 	for_each_online_cpu(cpu) {
296 		rdmsrl_on_cpu(cpu, MSR_HWP_REQUEST, &value);
297 		adj_range = limits->min_perf_pct * range / 100;
298 		min = hw_min + adj_range;
299 		value &= ~HWP_MIN_PERF(~0L);
300 		value |= HWP_MIN_PERF(min);
301 
302 		adj_range = limits->max_perf_pct * range / 100;
303 		max = hw_min + adj_range;
304 		if (limits->no_turbo) {
305 			hw_max = HWP_GUARANTEED_PERF(cap);
306 			if (hw_max < max)
307 				max = hw_max;
308 		}
309 
310 		value &= ~HWP_MAX_PERF(~0L);
311 		value |= HWP_MAX_PERF(max);
312 		wrmsrl_on_cpu(cpu, MSR_HWP_REQUEST, value);
313 	}
314 
315 	put_online_cpus();
316 }
317 
318 /************************** debugfs begin ************************/
319 static int pid_param_set(void *data, u64 val)
320 {
321 	*(u32 *)data = val;
322 	intel_pstate_reset_all_pid();
323 	return 0;
324 }
325 
326 static int pid_param_get(void *data, u64 *val)
327 {
328 	*val = *(u32 *)data;
329 	return 0;
330 }
331 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, pid_param_set, "%llu\n");
332 
333 struct pid_param {
334 	char *name;
335 	void *value;
336 };
337 
338 static struct pid_param pid_files[] = {
339 	{"sample_rate_ms", &pid_params.sample_rate_ms},
340 	{"d_gain_pct", &pid_params.d_gain_pct},
341 	{"i_gain_pct", &pid_params.i_gain_pct},
342 	{"deadband", &pid_params.deadband},
343 	{"setpoint", &pid_params.setpoint},
344 	{"p_gain_pct", &pid_params.p_gain_pct},
345 	{NULL, NULL}
346 };
347 
348 static void __init intel_pstate_debug_expose_params(void)
349 {
350 	struct dentry *debugfs_parent;
351 	int i = 0;
352 
353 	if (hwp_active)
354 		return;
355 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
356 	if (IS_ERR_OR_NULL(debugfs_parent))
357 		return;
358 	while (pid_files[i].name) {
359 		debugfs_create_file(pid_files[i].name, 0660,
360 				    debugfs_parent, pid_files[i].value,
361 				    &fops_pid_param);
362 		i++;
363 	}
364 }
365 
366 /************************** debugfs end ************************/
367 
368 /************************** sysfs begin ************************/
369 #define show_one(file_name, object)					\
370 	static ssize_t show_##file_name					\
371 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
372 	{								\
373 		return sprintf(buf, "%u\n", limits->object);		\
374 	}
375 
376 static ssize_t show_turbo_pct(struct kobject *kobj,
377 				struct attribute *attr, char *buf)
378 {
379 	struct cpudata *cpu;
380 	int total, no_turbo, turbo_pct;
381 	uint32_t turbo_fp;
382 
383 	cpu = all_cpu_data[0];
384 
385 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
386 	no_turbo = cpu->pstate.max_pstate - cpu->pstate.min_pstate + 1;
387 	turbo_fp = div_fp(int_tofp(no_turbo), int_tofp(total));
388 	turbo_pct = 100 - fp_toint(mul_fp(turbo_fp, int_tofp(100)));
389 	return sprintf(buf, "%u\n", turbo_pct);
390 }
391 
392 static ssize_t show_num_pstates(struct kobject *kobj,
393 				struct attribute *attr, char *buf)
394 {
395 	struct cpudata *cpu;
396 	int total;
397 
398 	cpu = all_cpu_data[0];
399 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
400 	return sprintf(buf, "%u\n", total);
401 }
402 
403 static ssize_t show_no_turbo(struct kobject *kobj,
404 			     struct attribute *attr, char *buf)
405 {
406 	ssize_t ret;
407 
408 	update_turbo_state();
409 	if (limits->turbo_disabled)
410 		ret = sprintf(buf, "%u\n", limits->turbo_disabled);
411 	else
412 		ret = sprintf(buf, "%u\n", limits->no_turbo);
413 
414 	return ret;
415 }
416 
417 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
418 			      const char *buf, size_t count)
419 {
420 	unsigned int input;
421 	int ret;
422 
423 	ret = sscanf(buf, "%u", &input);
424 	if (ret != 1)
425 		return -EINVAL;
426 
427 	update_turbo_state();
428 	if (limits->turbo_disabled) {
429 		pr_warn("intel_pstate: Turbo disabled by BIOS or unavailable on processor\n");
430 		return -EPERM;
431 	}
432 
433 	limits->no_turbo = clamp_t(int, input, 0, 1);
434 
435 	if (hwp_active)
436 		intel_pstate_hwp_set();
437 
438 	return count;
439 }
440 
441 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
442 				  const char *buf, size_t count)
443 {
444 	unsigned int input;
445 	int ret;
446 
447 	ret = sscanf(buf, "%u", &input);
448 	if (ret != 1)
449 		return -EINVAL;
450 
451 	limits->max_sysfs_pct = clamp_t(int, input, 0 , 100);
452 	limits->max_perf_pct = min(limits->max_policy_pct,
453 				   limits->max_sysfs_pct);
454 	limits->max_perf_pct = max(limits->min_policy_pct,
455 				   limits->max_perf_pct);
456 	limits->max_perf_pct = max(limits->min_perf_pct,
457 				   limits->max_perf_pct);
458 	limits->max_perf = div_fp(int_tofp(limits->max_perf_pct),
459 				  int_tofp(100));
460 
461 	if (hwp_active)
462 		intel_pstate_hwp_set();
463 	return count;
464 }
465 
466 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
467 				  const char *buf, size_t count)
468 {
469 	unsigned int input;
470 	int ret;
471 
472 	ret = sscanf(buf, "%u", &input);
473 	if (ret != 1)
474 		return -EINVAL;
475 
476 	limits->min_sysfs_pct = clamp_t(int, input, 0 , 100);
477 	limits->min_perf_pct = max(limits->min_policy_pct,
478 				   limits->min_sysfs_pct);
479 	limits->min_perf_pct = min(limits->max_policy_pct,
480 				   limits->min_perf_pct);
481 	limits->min_perf_pct = min(limits->max_perf_pct,
482 				   limits->min_perf_pct);
483 	limits->min_perf = div_fp(int_tofp(limits->min_perf_pct),
484 				  int_tofp(100));
485 
486 	if (hwp_active)
487 		intel_pstate_hwp_set();
488 	return count;
489 }
490 
491 show_one(max_perf_pct, max_perf_pct);
492 show_one(min_perf_pct, min_perf_pct);
493 
494 define_one_global_rw(no_turbo);
495 define_one_global_rw(max_perf_pct);
496 define_one_global_rw(min_perf_pct);
497 define_one_global_ro(turbo_pct);
498 define_one_global_ro(num_pstates);
499 
500 static struct attribute *intel_pstate_attributes[] = {
501 	&no_turbo.attr,
502 	&max_perf_pct.attr,
503 	&min_perf_pct.attr,
504 	&turbo_pct.attr,
505 	&num_pstates.attr,
506 	NULL
507 };
508 
509 static struct attribute_group intel_pstate_attr_group = {
510 	.attrs = intel_pstate_attributes,
511 };
512 
513 static void __init intel_pstate_sysfs_expose_params(void)
514 {
515 	struct kobject *intel_pstate_kobject;
516 	int rc;
517 
518 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
519 						&cpu_subsys.dev_root->kobj);
520 	BUG_ON(!intel_pstate_kobject);
521 	rc = sysfs_create_group(intel_pstate_kobject, &intel_pstate_attr_group);
522 	BUG_ON(rc);
523 }
524 /************************** sysfs end ************************/
525 
526 static void intel_pstate_hwp_enable(struct cpudata *cpudata)
527 {
528 	wrmsrl_on_cpu(cpudata->cpu, MSR_PM_ENABLE, 0x1);
529 }
530 
531 static int atom_get_min_pstate(void)
532 {
533 	u64 value;
534 
535 	rdmsrl(ATOM_RATIOS, value);
536 	return (value >> 8) & 0x7F;
537 }
538 
539 static int atom_get_max_pstate(void)
540 {
541 	u64 value;
542 
543 	rdmsrl(ATOM_RATIOS, value);
544 	return (value >> 16) & 0x7F;
545 }
546 
547 static int atom_get_turbo_pstate(void)
548 {
549 	u64 value;
550 
551 	rdmsrl(ATOM_TURBO_RATIOS, value);
552 	return value & 0x7F;
553 }
554 
555 static void atom_set_pstate(struct cpudata *cpudata, int pstate)
556 {
557 	u64 val;
558 	int32_t vid_fp;
559 	u32 vid;
560 
561 	val = (u64)pstate << 8;
562 	if (limits->no_turbo && !limits->turbo_disabled)
563 		val |= (u64)1 << 32;
564 
565 	vid_fp = cpudata->vid.min + mul_fp(
566 		int_tofp(pstate - cpudata->pstate.min_pstate),
567 		cpudata->vid.ratio);
568 
569 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
570 	vid = ceiling_fp(vid_fp);
571 
572 	if (pstate > cpudata->pstate.max_pstate)
573 		vid = cpudata->vid.turbo;
574 
575 	val |= vid;
576 
577 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
578 }
579 
580 static int silvermont_get_scaling(void)
581 {
582 	u64 value;
583 	int i;
584 	/* Defined in Table 35-6 from SDM (Sept 2015) */
585 	static int silvermont_freq_table[] = {
586 		83300, 100000, 133300, 116700, 80000};
587 
588 	rdmsrl(MSR_FSB_FREQ, value);
589 	i = value & 0x7;
590 	WARN_ON(i > 4);
591 
592 	return silvermont_freq_table[i];
593 }
594 
595 static int airmont_get_scaling(void)
596 {
597 	u64 value;
598 	int i;
599 	/* Defined in Table 35-10 from SDM (Sept 2015) */
600 	static int airmont_freq_table[] = {
601 		83300, 100000, 133300, 116700, 80000,
602 		93300, 90000, 88900, 87500};
603 
604 	rdmsrl(MSR_FSB_FREQ, value);
605 	i = value & 0xF;
606 	WARN_ON(i > 8);
607 
608 	return airmont_freq_table[i];
609 }
610 
611 static void atom_get_vid(struct cpudata *cpudata)
612 {
613 	u64 value;
614 
615 	rdmsrl(ATOM_VIDS, value);
616 	cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
617 	cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
618 	cpudata->vid.ratio = div_fp(
619 		cpudata->vid.max - cpudata->vid.min,
620 		int_tofp(cpudata->pstate.max_pstate -
621 			cpudata->pstate.min_pstate));
622 
623 	rdmsrl(ATOM_TURBO_VIDS, value);
624 	cpudata->vid.turbo = value & 0x7f;
625 }
626 
627 static int core_get_min_pstate(void)
628 {
629 	u64 value;
630 
631 	rdmsrl(MSR_PLATFORM_INFO, value);
632 	return (value >> 40) & 0xFF;
633 }
634 
635 static int core_get_max_pstate_physical(void)
636 {
637 	u64 value;
638 
639 	rdmsrl(MSR_PLATFORM_INFO, value);
640 	return (value >> 8) & 0xFF;
641 }
642 
643 static int core_get_max_pstate(void)
644 {
645 	u64 tar;
646 	u64 plat_info;
647 	int max_pstate;
648 	int err;
649 
650 	rdmsrl(MSR_PLATFORM_INFO, plat_info);
651 	max_pstate = (plat_info >> 8) & 0xFF;
652 
653 	err = rdmsrl_safe(MSR_TURBO_ACTIVATION_RATIO, &tar);
654 	if (!err) {
655 		/* Do some sanity checking for safety */
656 		if (plat_info & 0x600000000) {
657 			u64 tdp_ctrl;
658 			u64 tdp_ratio;
659 			int tdp_msr;
660 
661 			err = rdmsrl_safe(MSR_CONFIG_TDP_CONTROL, &tdp_ctrl);
662 			if (err)
663 				goto skip_tar;
664 
665 			tdp_msr = MSR_CONFIG_TDP_NOMINAL + tdp_ctrl;
666 			err = rdmsrl_safe(tdp_msr, &tdp_ratio);
667 			if (err)
668 				goto skip_tar;
669 
670 			if (tdp_ratio - 1 == tar) {
671 				max_pstate = tar;
672 				pr_debug("max_pstate=TAC %x\n", max_pstate);
673 			} else {
674 				goto skip_tar;
675 			}
676 		}
677 	}
678 
679 skip_tar:
680 	return max_pstate;
681 }
682 
683 static int core_get_turbo_pstate(void)
684 {
685 	u64 value;
686 	int nont, ret;
687 
688 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
689 	nont = core_get_max_pstate();
690 	ret = (value) & 255;
691 	if (ret <= nont)
692 		ret = nont;
693 	return ret;
694 }
695 
696 static inline int core_get_scaling(void)
697 {
698 	return 100000;
699 }
700 
701 static void core_set_pstate(struct cpudata *cpudata, int pstate)
702 {
703 	u64 val;
704 
705 	val = (u64)pstate << 8;
706 	if (limits->no_turbo && !limits->turbo_disabled)
707 		val |= (u64)1 << 32;
708 
709 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
710 }
711 
712 static int knl_get_turbo_pstate(void)
713 {
714 	u64 value;
715 	int nont, ret;
716 
717 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
718 	nont = core_get_max_pstate();
719 	ret = (((value) >> 8) & 0xFF);
720 	if (ret <= nont)
721 		ret = nont;
722 	return ret;
723 }
724 
725 static struct cpu_defaults core_params = {
726 	.pid_policy = {
727 		.sample_rate_ms = 10,
728 		.deadband = 0,
729 		.setpoint = 97,
730 		.p_gain_pct = 20,
731 		.d_gain_pct = 0,
732 		.i_gain_pct = 0,
733 	},
734 	.funcs = {
735 		.get_max = core_get_max_pstate,
736 		.get_max_physical = core_get_max_pstate_physical,
737 		.get_min = core_get_min_pstate,
738 		.get_turbo = core_get_turbo_pstate,
739 		.get_scaling = core_get_scaling,
740 		.set = core_set_pstate,
741 	},
742 };
743 
744 static struct cpu_defaults silvermont_params = {
745 	.pid_policy = {
746 		.sample_rate_ms = 10,
747 		.deadband = 0,
748 		.setpoint = 60,
749 		.p_gain_pct = 14,
750 		.d_gain_pct = 0,
751 		.i_gain_pct = 4,
752 	},
753 	.funcs = {
754 		.get_max = atom_get_max_pstate,
755 		.get_max_physical = atom_get_max_pstate,
756 		.get_min = atom_get_min_pstate,
757 		.get_turbo = atom_get_turbo_pstate,
758 		.set = atom_set_pstate,
759 		.get_scaling = silvermont_get_scaling,
760 		.get_vid = atom_get_vid,
761 	},
762 };
763 
764 static struct cpu_defaults airmont_params = {
765 	.pid_policy = {
766 		.sample_rate_ms = 10,
767 		.deadband = 0,
768 		.setpoint = 60,
769 		.p_gain_pct = 14,
770 		.d_gain_pct = 0,
771 		.i_gain_pct = 4,
772 	},
773 	.funcs = {
774 		.get_max = atom_get_max_pstate,
775 		.get_max_physical = atom_get_max_pstate,
776 		.get_min = atom_get_min_pstate,
777 		.get_turbo = atom_get_turbo_pstate,
778 		.set = atom_set_pstate,
779 		.get_scaling = airmont_get_scaling,
780 		.get_vid = atom_get_vid,
781 	},
782 };
783 
784 static struct cpu_defaults knl_params = {
785 	.pid_policy = {
786 		.sample_rate_ms = 10,
787 		.deadband = 0,
788 		.setpoint = 97,
789 		.p_gain_pct = 20,
790 		.d_gain_pct = 0,
791 		.i_gain_pct = 0,
792 	},
793 	.funcs = {
794 		.get_max = core_get_max_pstate,
795 		.get_max_physical = core_get_max_pstate_physical,
796 		.get_min = core_get_min_pstate,
797 		.get_turbo = knl_get_turbo_pstate,
798 		.get_scaling = core_get_scaling,
799 		.set = core_set_pstate,
800 	},
801 };
802 
803 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
804 {
805 	int max_perf = cpu->pstate.turbo_pstate;
806 	int max_perf_adj;
807 	int min_perf;
808 
809 	if (limits->no_turbo || limits->turbo_disabled)
810 		max_perf = cpu->pstate.max_pstate;
811 
812 	/*
813 	 * performance can be limited by user through sysfs, by cpufreq
814 	 * policy, or by cpu specific default values determined through
815 	 * experimentation.
816 	 */
817 	max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits->max_perf));
818 	*max = clamp_t(int, max_perf_adj,
819 			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
820 
821 	min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits->min_perf));
822 	*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
823 }
824 
825 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate, bool force)
826 {
827 	int max_perf, min_perf;
828 
829 	if (force) {
830 		update_turbo_state();
831 
832 		intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
833 
834 		pstate = clamp_t(int, pstate, min_perf, max_perf);
835 
836 		if (pstate == cpu->pstate.current_pstate)
837 			return;
838 	}
839 	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
840 
841 	cpu->pstate.current_pstate = pstate;
842 
843 	pstate_funcs.set(cpu, pstate);
844 }
845 
846 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
847 {
848 	cpu->pstate.min_pstate = pstate_funcs.get_min();
849 	cpu->pstate.max_pstate = pstate_funcs.get_max();
850 	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical();
851 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
852 	cpu->pstate.scaling = pstate_funcs.get_scaling();
853 
854 	if (pstate_funcs.get_vid)
855 		pstate_funcs.get_vid(cpu);
856 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate, false);
857 }
858 
859 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
860 {
861 	struct sample *sample = &cpu->sample;
862 	int64_t core_pct;
863 
864 	core_pct = int_tofp(sample->aperf) * int_tofp(100);
865 	core_pct = div64_u64(core_pct, int_tofp(sample->mperf));
866 
867 	sample->freq = fp_toint(
868 		mul_fp(int_tofp(
869 			cpu->pstate.max_pstate_physical *
870 			cpu->pstate.scaling / 100),
871 			core_pct));
872 
873 	sample->core_pct_busy = (int32_t)core_pct;
874 }
875 
876 static inline void intel_pstate_sample(struct cpudata *cpu)
877 {
878 	u64 aperf, mperf;
879 	unsigned long flags;
880 	u64 tsc;
881 
882 	local_irq_save(flags);
883 	rdmsrl(MSR_IA32_APERF, aperf);
884 	rdmsrl(MSR_IA32_MPERF, mperf);
885 	if (cpu->prev_mperf == mperf) {
886 		local_irq_restore(flags);
887 		return;
888 	}
889 
890 	tsc = rdtsc();
891 	local_irq_restore(flags);
892 
893 	cpu->last_sample_time = cpu->sample.time;
894 	cpu->sample.time = ktime_get();
895 	cpu->sample.aperf = aperf;
896 	cpu->sample.mperf = mperf;
897 	cpu->sample.tsc =  tsc;
898 	cpu->sample.aperf -= cpu->prev_aperf;
899 	cpu->sample.mperf -= cpu->prev_mperf;
900 	cpu->sample.tsc -= cpu->prev_tsc;
901 
902 	intel_pstate_calc_busy(cpu);
903 
904 	cpu->prev_aperf = aperf;
905 	cpu->prev_mperf = mperf;
906 	cpu->prev_tsc = tsc;
907 }
908 
909 static inline void intel_hwp_set_sample_time(struct cpudata *cpu)
910 {
911 	int delay;
912 
913 	delay = msecs_to_jiffies(50);
914 	mod_timer_pinned(&cpu->timer, jiffies + delay);
915 }
916 
917 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
918 {
919 	int delay;
920 
921 	delay = msecs_to_jiffies(pid_params.sample_rate_ms);
922 	mod_timer_pinned(&cpu->timer, jiffies + delay);
923 }
924 
925 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
926 {
927 	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
928 	s64 duration_us;
929 	u32 sample_time;
930 
931 	/*
932 	 * core_busy is the ratio of actual performance to max
933 	 * max_pstate is the max non turbo pstate available
934 	 * current_pstate was the pstate that was requested during
935 	 * 	the last sample period.
936 	 *
937 	 * We normalize core_busy, which was our actual percent
938 	 * performance to what we requested during the last sample
939 	 * period. The result will be a percentage of busy at a
940 	 * specified pstate.
941 	 */
942 	core_busy = cpu->sample.core_pct_busy;
943 	max_pstate = int_tofp(cpu->pstate.max_pstate_physical);
944 	current_pstate = int_tofp(cpu->pstate.current_pstate);
945 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
946 
947 	/*
948 	 * Since we have a deferred timer, it will not fire unless
949 	 * we are in C0.  So, determine if the actual elapsed time
950 	 * is significantly greater (3x) than our sample interval.  If it
951 	 * is, then we were idle for a long enough period of time
952 	 * to adjust our busyness.
953 	 */
954 	sample_time = pid_params.sample_rate_ms  * USEC_PER_MSEC;
955 	duration_us = ktime_us_delta(cpu->sample.time,
956 				     cpu->last_sample_time);
957 	if (duration_us > sample_time * 3) {
958 		sample_ratio = div_fp(int_tofp(sample_time),
959 				      int_tofp(duration_us));
960 		core_busy = mul_fp(core_busy, sample_ratio);
961 	}
962 
963 	return core_busy;
964 }
965 
966 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
967 {
968 	int32_t busy_scaled;
969 	struct _pid *pid;
970 	signed int ctl;
971 	int from;
972 	struct sample *sample;
973 
974 	from = cpu->pstate.current_pstate;
975 
976 	pid = &cpu->pid;
977 	busy_scaled = intel_pstate_get_scaled_busy(cpu);
978 
979 	ctl = pid_calc(pid, busy_scaled);
980 
981 	/* Negative values of ctl increase the pstate and vice versa */
982 	intel_pstate_set_pstate(cpu, cpu->pstate.current_pstate - ctl, true);
983 
984 	sample = &cpu->sample;
985 	trace_pstate_sample(fp_toint(sample->core_pct_busy),
986 		fp_toint(busy_scaled),
987 		from,
988 		cpu->pstate.current_pstate,
989 		sample->mperf,
990 		sample->aperf,
991 		sample->tsc,
992 		sample->freq);
993 }
994 
995 static void intel_hwp_timer_func(unsigned long __data)
996 {
997 	struct cpudata *cpu = (struct cpudata *) __data;
998 
999 	intel_pstate_sample(cpu);
1000 	intel_hwp_set_sample_time(cpu);
1001 }
1002 
1003 static void intel_pstate_timer_func(unsigned long __data)
1004 {
1005 	struct cpudata *cpu = (struct cpudata *) __data;
1006 
1007 	intel_pstate_sample(cpu);
1008 
1009 	intel_pstate_adjust_busy_pstate(cpu);
1010 
1011 	intel_pstate_set_sample_time(cpu);
1012 }
1013 
1014 #define ICPU(model, policy) \
1015 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
1016 			(unsigned long)&policy }
1017 
1018 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
1019 	ICPU(0x2a, core_params),
1020 	ICPU(0x2d, core_params),
1021 	ICPU(0x37, silvermont_params),
1022 	ICPU(0x3a, core_params),
1023 	ICPU(0x3c, core_params),
1024 	ICPU(0x3d, core_params),
1025 	ICPU(0x3e, core_params),
1026 	ICPU(0x3f, core_params),
1027 	ICPU(0x45, core_params),
1028 	ICPU(0x46, core_params),
1029 	ICPU(0x47, core_params),
1030 	ICPU(0x4c, airmont_params),
1031 	ICPU(0x4e, core_params),
1032 	ICPU(0x4f, core_params),
1033 	ICPU(0x5e, core_params),
1034 	ICPU(0x56, core_params),
1035 	ICPU(0x57, knl_params),
1036 	{}
1037 };
1038 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
1039 
1040 static const struct x86_cpu_id intel_pstate_cpu_oob_ids[] = {
1041 	ICPU(0x56, core_params),
1042 	{}
1043 };
1044 
1045 static int intel_pstate_init_cpu(unsigned int cpunum)
1046 {
1047 	struct cpudata *cpu;
1048 
1049 	if (!all_cpu_data[cpunum])
1050 		all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata),
1051 					       GFP_KERNEL);
1052 	if (!all_cpu_data[cpunum])
1053 		return -ENOMEM;
1054 
1055 	cpu = all_cpu_data[cpunum];
1056 
1057 	cpu->cpu = cpunum;
1058 
1059 	if (hwp_active)
1060 		intel_pstate_hwp_enable(cpu);
1061 
1062 	intel_pstate_get_cpu_pstates(cpu);
1063 
1064 	init_timer_deferrable(&cpu->timer);
1065 	cpu->timer.data = (unsigned long)cpu;
1066 	cpu->timer.expires = jiffies + HZ/100;
1067 
1068 	if (!hwp_active)
1069 		cpu->timer.function = intel_pstate_timer_func;
1070 	else
1071 		cpu->timer.function = intel_hwp_timer_func;
1072 
1073 	intel_pstate_busy_pid_reset(cpu);
1074 	intel_pstate_sample(cpu);
1075 
1076 	add_timer_on(&cpu->timer, cpunum);
1077 
1078 	pr_debug("intel_pstate: controlling: cpu %d\n", cpunum);
1079 
1080 	return 0;
1081 }
1082 
1083 static unsigned int intel_pstate_get(unsigned int cpu_num)
1084 {
1085 	struct sample *sample;
1086 	struct cpudata *cpu;
1087 
1088 	cpu = all_cpu_data[cpu_num];
1089 	if (!cpu)
1090 		return 0;
1091 	sample = &cpu->sample;
1092 	return sample->freq;
1093 }
1094 
1095 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
1096 {
1097 	if (!policy->cpuinfo.max_freq)
1098 		return -ENODEV;
1099 
1100 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE &&
1101 	    policy->max >= policy->cpuinfo.max_freq) {
1102 		pr_debug("intel_pstate: set performance\n");
1103 		limits = &performance_limits;
1104 		if (hwp_active)
1105 			intel_pstate_hwp_set();
1106 		return 0;
1107 	}
1108 
1109 	pr_debug("intel_pstate: set powersave\n");
1110 	limits = &powersave_limits;
1111 	limits->min_policy_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
1112 	limits->min_policy_pct = clamp_t(int, limits->min_policy_pct, 0 , 100);
1113 	limits->max_policy_pct = DIV_ROUND_UP(policy->max * 100,
1114 					      policy->cpuinfo.max_freq);
1115 	limits->max_policy_pct = clamp_t(int, limits->max_policy_pct, 0 , 100);
1116 
1117 	/* Normalize user input to [min_policy_pct, max_policy_pct] */
1118 	limits->min_perf_pct = max(limits->min_policy_pct,
1119 				   limits->min_sysfs_pct);
1120 	limits->min_perf_pct = min(limits->max_policy_pct,
1121 				   limits->min_perf_pct);
1122 	limits->max_perf_pct = min(limits->max_policy_pct,
1123 				   limits->max_sysfs_pct);
1124 	limits->max_perf_pct = max(limits->min_policy_pct,
1125 				   limits->max_perf_pct);
1126 	limits->max_perf = round_up(limits->max_perf, FRAC_BITS);
1127 
1128 	/* Make sure min_perf_pct <= max_perf_pct */
1129 	limits->min_perf_pct = min(limits->max_perf_pct, limits->min_perf_pct);
1130 
1131 	limits->min_perf = div_fp(int_tofp(limits->min_perf_pct),
1132 				  int_tofp(100));
1133 	limits->max_perf = div_fp(int_tofp(limits->max_perf_pct),
1134 				  int_tofp(100));
1135 
1136 	if (hwp_active)
1137 		intel_pstate_hwp_set();
1138 
1139 	return 0;
1140 }
1141 
1142 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
1143 {
1144 	cpufreq_verify_within_cpu_limits(policy);
1145 
1146 	if (policy->policy != CPUFREQ_POLICY_POWERSAVE &&
1147 	    policy->policy != CPUFREQ_POLICY_PERFORMANCE)
1148 		return -EINVAL;
1149 
1150 	return 0;
1151 }
1152 
1153 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
1154 {
1155 	int cpu_num = policy->cpu;
1156 	struct cpudata *cpu = all_cpu_data[cpu_num];
1157 
1158 	pr_debug("intel_pstate: CPU %d exiting\n", cpu_num);
1159 
1160 	del_timer_sync(&all_cpu_data[cpu_num]->timer);
1161 	if (hwp_active)
1162 		return;
1163 
1164 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate, false);
1165 }
1166 
1167 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
1168 {
1169 	struct cpudata *cpu;
1170 	int rc;
1171 
1172 	rc = intel_pstate_init_cpu(policy->cpu);
1173 	if (rc)
1174 		return rc;
1175 
1176 	cpu = all_cpu_data[policy->cpu];
1177 
1178 	if (limits->min_perf_pct == 100 && limits->max_perf_pct == 100)
1179 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
1180 	else
1181 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
1182 
1183 	policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
1184 	policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
1185 
1186 	/* cpuinfo and default policy values */
1187 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
1188 	policy->cpuinfo.max_freq =
1189 		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
1190 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
1191 	cpumask_set_cpu(policy->cpu, policy->cpus);
1192 
1193 	return 0;
1194 }
1195 
1196 static struct cpufreq_driver intel_pstate_driver = {
1197 	.flags		= CPUFREQ_CONST_LOOPS,
1198 	.verify		= intel_pstate_verify_policy,
1199 	.setpolicy	= intel_pstate_set_policy,
1200 	.get		= intel_pstate_get,
1201 	.init		= intel_pstate_cpu_init,
1202 	.stop_cpu	= intel_pstate_stop_cpu,
1203 	.name		= "intel_pstate",
1204 };
1205 
1206 static int __initdata no_load;
1207 static int __initdata no_hwp;
1208 static int __initdata hwp_only;
1209 static unsigned int force_load;
1210 
1211 static int intel_pstate_msrs_not_valid(void)
1212 {
1213 	if (!pstate_funcs.get_max() ||
1214 	    !pstate_funcs.get_min() ||
1215 	    !pstate_funcs.get_turbo())
1216 		return -ENODEV;
1217 
1218 	return 0;
1219 }
1220 
1221 static void copy_pid_params(struct pstate_adjust_policy *policy)
1222 {
1223 	pid_params.sample_rate_ms = policy->sample_rate_ms;
1224 	pid_params.p_gain_pct = policy->p_gain_pct;
1225 	pid_params.i_gain_pct = policy->i_gain_pct;
1226 	pid_params.d_gain_pct = policy->d_gain_pct;
1227 	pid_params.deadband = policy->deadband;
1228 	pid_params.setpoint = policy->setpoint;
1229 }
1230 
1231 static void copy_cpu_funcs(struct pstate_funcs *funcs)
1232 {
1233 	pstate_funcs.get_max   = funcs->get_max;
1234 	pstate_funcs.get_max_physical = funcs->get_max_physical;
1235 	pstate_funcs.get_min   = funcs->get_min;
1236 	pstate_funcs.get_turbo = funcs->get_turbo;
1237 	pstate_funcs.get_scaling = funcs->get_scaling;
1238 	pstate_funcs.set       = funcs->set;
1239 	pstate_funcs.get_vid   = funcs->get_vid;
1240 }
1241 
1242 #if IS_ENABLED(CONFIG_ACPI)
1243 #include <acpi/processor.h>
1244 
1245 static bool intel_pstate_no_acpi_pss(void)
1246 {
1247 	int i;
1248 
1249 	for_each_possible_cpu(i) {
1250 		acpi_status status;
1251 		union acpi_object *pss;
1252 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
1253 		struct acpi_processor *pr = per_cpu(processors, i);
1254 
1255 		if (!pr)
1256 			continue;
1257 
1258 		status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
1259 		if (ACPI_FAILURE(status))
1260 			continue;
1261 
1262 		pss = buffer.pointer;
1263 		if (pss && pss->type == ACPI_TYPE_PACKAGE) {
1264 			kfree(pss);
1265 			return false;
1266 		}
1267 
1268 		kfree(pss);
1269 	}
1270 
1271 	return true;
1272 }
1273 
1274 static bool intel_pstate_has_acpi_ppc(void)
1275 {
1276 	int i;
1277 
1278 	for_each_possible_cpu(i) {
1279 		struct acpi_processor *pr = per_cpu(processors, i);
1280 
1281 		if (!pr)
1282 			continue;
1283 		if (acpi_has_method(pr->handle, "_PPC"))
1284 			return true;
1285 	}
1286 	return false;
1287 }
1288 
1289 enum {
1290 	PSS,
1291 	PPC,
1292 };
1293 
1294 struct hw_vendor_info {
1295 	u16  valid;
1296 	char oem_id[ACPI_OEM_ID_SIZE];
1297 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
1298 	int  oem_pwr_table;
1299 };
1300 
1301 /* Hardware vendor-specific info that has its own power management modes */
1302 static struct hw_vendor_info vendor_info[] = {
1303 	{1, "HP    ", "ProLiant", PSS},
1304 	{1, "ORACLE", "X4-2    ", PPC},
1305 	{1, "ORACLE", "X4-2L   ", PPC},
1306 	{1, "ORACLE", "X4-2B   ", PPC},
1307 	{1, "ORACLE", "X3-2    ", PPC},
1308 	{1, "ORACLE", "X3-2L   ", PPC},
1309 	{1, "ORACLE", "X3-2B   ", PPC},
1310 	{1, "ORACLE", "X4470M2 ", PPC},
1311 	{1, "ORACLE", "X4270M3 ", PPC},
1312 	{1, "ORACLE", "X4270M2 ", PPC},
1313 	{1, "ORACLE", "X4170M2 ", PPC},
1314 	{1, "ORACLE", "X4170 M3", PPC},
1315 	{1, "ORACLE", "X4275 M3", PPC},
1316 	{1, "ORACLE", "X6-2    ", PPC},
1317 	{1, "ORACLE", "Sudbury ", PPC},
1318 	{0, "", ""},
1319 };
1320 
1321 static bool intel_pstate_platform_pwr_mgmt_exists(void)
1322 {
1323 	struct acpi_table_header hdr;
1324 	struct hw_vendor_info *v_info;
1325 	const struct x86_cpu_id *id;
1326 	u64 misc_pwr;
1327 
1328 	id = x86_match_cpu(intel_pstate_cpu_oob_ids);
1329 	if (id) {
1330 		rdmsrl(MSR_MISC_PWR_MGMT, misc_pwr);
1331 		if ( misc_pwr & (1 << 8))
1332 			return true;
1333 	}
1334 
1335 	if (acpi_disabled ||
1336 	    ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
1337 		return false;
1338 
1339 	for (v_info = vendor_info; v_info->valid; v_info++) {
1340 		if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) &&
1341 			!strncmp(hdr.oem_table_id, v_info->oem_table_id,
1342 						ACPI_OEM_TABLE_ID_SIZE))
1343 			switch (v_info->oem_pwr_table) {
1344 			case PSS:
1345 				return intel_pstate_no_acpi_pss();
1346 			case PPC:
1347 				return intel_pstate_has_acpi_ppc() &&
1348 					(!force_load);
1349 			}
1350 	}
1351 
1352 	return false;
1353 }
1354 #else /* CONFIG_ACPI not enabled */
1355 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
1356 static inline bool intel_pstate_has_acpi_ppc(void) { return false; }
1357 #endif /* CONFIG_ACPI */
1358 
1359 static int __init intel_pstate_init(void)
1360 {
1361 	int cpu, rc = 0;
1362 	const struct x86_cpu_id *id;
1363 	struct cpu_defaults *cpu_def;
1364 
1365 	if (no_load)
1366 		return -ENODEV;
1367 
1368 	id = x86_match_cpu(intel_pstate_cpu_ids);
1369 	if (!id)
1370 		return -ENODEV;
1371 
1372 	/*
1373 	 * The Intel pstate driver will be ignored if the platform
1374 	 * firmware has its own power management modes.
1375 	 */
1376 	if (intel_pstate_platform_pwr_mgmt_exists())
1377 		return -ENODEV;
1378 
1379 	cpu_def = (struct cpu_defaults *)id->driver_data;
1380 
1381 	copy_pid_params(&cpu_def->pid_policy);
1382 	copy_cpu_funcs(&cpu_def->funcs);
1383 
1384 	if (intel_pstate_msrs_not_valid())
1385 		return -ENODEV;
1386 
1387 	pr_info("Intel P-state driver initializing.\n");
1388 
1389 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
1390 	if (!all_cpu_data)
1391 		return -ENOMEM;
1392 
1393 	if (static_cpu_has_safe(X86_FEATURE_HWP) && !no_hwp) {
1394 		pr_info("intel_pstate: HWP enabled\n");
1395 		hwp_active++;
1396 	}
1397 
1398 	if (!hwp_active && hwp_only)
1399 		goto out;
1400 
1401 	rc = cpufreq_register_driver(&intel_pstate_driver);
1402 	if (rc)
1403 		goto out;
1404 
1405 	intel_pstate_debug_expose_params();
1406 	intel_pstate_sysfs_expose_params();
1407 
1408 	return rc;
1409 out:
1410 	get_online_cpus();
1411 	for_each_online_cpu(cpu) {
1412 		if (all_cpu_data[cpu]) {
1413 			del_timer_sync(&all_cpu_data[cpu]->timer);
1414 			kfree(all_cpu_data[cpu]);
1415 		}
1416 	}
1417 
1418 	put_online_cpus();
1419 	vfree(all_cpu_data);
1420 	return -ENODEV;
1421 }
1422 device_initcall(intel_pstate_init);
1423 
1424 static int __init intel_pstate_setup(char *str)
1425 {
1426 	if (!str)
1427 		return -EINVAL;
1428 
1429 	if (!strcmp(str, "disable"))
1430 		no_load = 1;
1431 	if (!strcmp(str, "no_hwp")) {
1432 		pr_info("intel_pstate: HWP disabled\n");
1433 		no_hwp = 1;
1434 	}
1435 	if (!strcmp(str, "force"))
1436 		force_load = 1;
1437 	if (!strcmp(str, "hwp_only"))
1438 		hwp_only = 1;
1439 	return 0;
1440 }
1441 early_param("intel_pstate", intel_pstate_setup);
1442 
1443 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1444 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1445 MODULE_LICENSE("GPL");
1446