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 #if IS_ENABLED(CONFIG_ACPI)
38 #include <acpi/processor.h>
39 #endif
40 
41 #define BYT_RATIOS		0x66a
42 #define BYT_VIDS		0x66b
43 #define BYT_TURBO_RATIOS	0x66c
44 #define BYT_TURBO_VIDS		0x66d
45 
46 #define FRAC_BITS 8
47 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
48 #define fp_toint(X) ((X) >> FRAC_BITS)
49 
50 static inline int32_t mul_fp(int32_t x, int32_t y)
51 {
52 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
53 }
54 
55 static inline int32_t div_fp(s64 x, s64 y)
56 {
57 	return div64_s64((int64_t)x << FRAC_BITS, y);
58 }
59 
60 static inline int ceiling_fp(int32_t x)
61 {
62 	int mask, ret;
63 
64 	ret = fp_toint(x);
65 	mask = (1 << FRAC_BITS) - 1;
66 	if (x & mask)
67 		ret += 1;
68 	return ret;
69 }
70 
71 struct sample {
72 	int32_t core_pct_busy;
73 	u64 aperf;
74 	u64 mperf;
75 	u64 tsc;
76 	int freq;
77 	ktime_t time;
78 };
79 
80 struct pstate_data {
81 	int	current_pstate;
82 	int	min_pstate;
83 	int	max_pstate;
84 	int	max_pstate_physical;
85 	int	scaling;
86 	int	turbo_pstate;
87 };
88 
89 struct vid_data {
90 	int min;
91 	int max;
92 	int turbo;
93 	int32_t ratio;
94 };
95 
96 struct _pid {
97 	int setpoint;
98 	int32_t integral;
99 	int32_t p_gain;
100 	int32_t i_gain;
101 	int32_t d_gain;
102 	int deadband;
103 	int32_t last_err;
104 };
105 
106 struct cpudata {
107 	int cpu;
108 
109 	struct timer_list timer;
110 
111 	struct pstate_data pstate;
112 	struct vid_data vid;
113 	struct _pid pid;
114 
115 	ktime_t last_sample_time;
116 	u64	prev_aperf;
117 	u64	prev_mperf;
118 	u64	prev_tsc;
119 	struct sample sample;
120 #if IS_ENABLED(CONFIG_ACPI)
121 	struct acpi_processor_performance acpi_perf_data;
122 #endif
123 };
124 
125 static struct cpudata **all_cpu_data;
126 struct pstate_adjust_policy {
127 	int sample_rate_ms;
128 	int deadband;
129 	int setpoint;
130 	int p_gain_pct;
131 	int d_gain_pct;
132 	int i_gain_pct;
133 };
134 
135 struct pstate_funcs {
136 	int (*get_max)(void);
137 	int (*get_max_physical)(void);
138 	int (*get_min)(void);
139 	int (*get_turbo)(void);
140 	int (*get_scaling)(void);
141 	void (*set)(struct cpudata*, int pstate);
142 	void (*get_vid)(struct cpudata *);
143 };
144 
145 struct cpu_defaults {
146 	struct pstate_adjust_policy pid_policy;
147 	struct pstate_funcs funcs;
148 };
149 
150 static struct pstate_adjust_policy pid_params;
151 static struct pstate_funcs pstate_funcs;
152 static int hwp_active;
153 static int no_acpi_perf;
154 
155 struct perf_limits {
156 	int no_turbo;
157 	int turbo_disabled;
158 	int max_perf_pct;
159 	int min_perf_pct;
160 	int32_t max_perf;
161 	int32_t min_perf;
162 	int max_policy_pct;
163 	int max_sysfs_pct;
164 	int min_policy_pct;
165 	int min_sysfs_pct;
166 	int max_perf_ctl;
167 	int min_perf_ctl;
168 };
169 
170 static struct perf_limits performance_limits = {
171 	.no_turbo = 0,
172 	.turbo_disabled = 0,
173 	.max_perf_pct = 100,
174 	.max_perf = int_tofp(1),
175 	.min_perf_pct = 100,
176 	.min_perf = int_tofp(1),
177 	.max_policy_pct = 100,
178 	.max_sysfs_pct = 100,
179 	.min_policy_pct = 0,
180 	.min_sysfs_pct = 0,
181 };
182 
183 static struct perf_limits powersave_limits = {
184 	.no_turbo = 0,
185 	.turbo_disabled = 0,
186 	.max_perf_pct = 100,
187 	.max_perf = int_tofp(1),
188 	.min_perf_pct = 0,
189 	.min_perf = 0,
190 	.max_policy_pct = 100,
191 	.max_sysfs_pct = 100,
192 	.min_policy_pct = 0,
193 	.min_sysfs_pct = 0,
194 	.max_perf_ctl = 0,
195 	.min_perf_ctl = 0,
196 };
197 
198 #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE
199 static struct perf_limits *limits = &performance_limits;
200 #else
201 static struct perf_limits *limits = &powersave_limits;
202 #endif
203 
204 #if IS_ENABLED(CONFIG_ACPI)
205 /*
206  * The max target pstate ratio is a 8 bit value in both PLATFORM_INFO MSR and
207  * in TURBO_RATIO_LIMIT MSR, which pstate driver stores in max_pstate and
208  * max_turbo_pstate fields. The PERF_CTL MSR contains 16 bit value for P state
209  * ratio, out of it only high 8 bits are used. For example 0x1700 is setting
210  * target ratio 0x17. The _PSS control value stores in a format which can be
211  * directly written to PERF_CTL MSR. But in intel_pstate driver this shift
212  * occurs during write to PERF_CTL (E.g. for cores core_set_pstate()).
213  * This function converts the _PSS control value to intel pstate driver format
214  * for comparison and assignment.
215  */
216 static int convert_to_native_pstate_format(struct cpudata *cpu, int index)
217 {
218 	return cpu->acpi_perf_data.states[index].control >> 8;
219 }
220 
221 static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
222 {
223 	struct cpudata *cpu;
224 	int ret;
225 	bool turbo_absent = false;
226 	int max_pstate_index;
227 	int min_pss_ctl, max_pss_ctl, turbo_pss_ctl;
228 	int i;
229 
230 	cpu = all_cpu_data[policy->cpu];
231 
232 	pr_debug("intel_pstate: default limits 0x%x 0x%x 0x%x\n",
233 		 cpu->pstate.min_pstate, cpu->pstate.max_pstate,
234 		 cpu->pstate.turbo_pstate);
235 
236 	if (!cpu->acpi_perf_data.shared_cpu_map &&
237 	    zalloc_cpumask_var_node(&cpu->acpi_perf_data.shared_cpu_map,
238 				    GFP_KERNEL, cpu_to_node(policy->cpu))) {
239 		return -ENOMEM;
240 	}
241 
242 	ret = acpi_processor_register_performance(&cpu->acpi_perf_data,
243 						  policy->cpu);
244 	if (ret)
245 		return ret;
246 
247 	/*
248 	 * Check if the control value in _PSS is for PERF_CTL MSR, which should
249 	 * guarantee that the states returned by it map to the states in our
250 	 * list directly.
251 	 */
252 	if (cpu->acpi_perf_data.control_register.space_id !=
253 						ACPI_ADR_SPACE_FIXED_HARDWARE)
254 		return -EIO;
255 
256 	pr_debug("intel_pstate: CPU%u - ACPI _PSS perf data\n", policy->cpu);
257 	for (i = 0; i < cpu->acpi_perf_data.state_count; i++)
258 		pr_debug("     %cP%d: %u MHz, %u mW, 0x%x\n",
259 			 (i == cpu->acpi_perf_data.state ? '*' : ' '), i,
260 			 (u32) cpu->acpi_perf_data.states[i].core_frequency,
261 			 (u32) cpu->acpi_perf_data.states[i].power,
262 			 (u32) cpu->acpi_perf_data.states[i].control);
263 
264 	/*
265 	 * If there is only one entry _PSS, simply ignore _PSS and continue as
266 	 * usual without taking _PSS into account
267 	 */
268 	if (cpu->acpi_perf_data.state_count < 2)
269 		return 0;
270 
271 	turbo_pss_ctl = convert_to_native_pstate_format(cpu, 0);
272 	min_pss_ctl = convert_to_native_pstate_format(cpu,
273 					cpu->acpi_perf_data.state_count - 1);
274 	/* Check if there is a turbo freq in _PSS */
275 	if (turbo_pss_ctl <= cpu->pstate.max_pstate &&
276 	    turbo_pss_ctl > cpu->pstate.min_pstate) {
277 		pr_debug("intel_pstate: no turbo range exists in _PSS\n");
278 		limits->no_turbo = limits->turbo_disabled = 1;
279 		cpu->pstate.turbo_pstate = cpu->pstate.max_pstate;
280 		turbo_absent = true;
281 	}
282 
283 	/* Check if the max non turbo p state < Intel P state max */
284 	max_pstate_index = turbo_absent ? 0 : 1;
285 	max_pss_ctl = convert_to_native_pstate_format(cpu, max_pstate_index);
286 	if (max_pss_ctl < cpu->pstate.max_pstate &&
287 	    max_pss_ctl > cpu->pstate.min_pstate)
288 		cpu->pstate.max_pstate = max_pss_ctl;
289 
290 	/* check If min perf > Intel P State min */
291 	if (min_pss_ctl > cpu->pstate.min_pstate &&
292 	    min_pss_ctl < cpu->pstate.max_pstate) {
293 		cpu->pstate.min_pstate = min_pss_ctl;
294 		policy->cpuinfo.min_freq = min_pss_ctl * cpu->pstate.scaling;
295 	}
296 
297 	if (turbo_absent)
298 		policy->cpuinfo.max_freq = cpu->pstate.max_pstate *
299 						cpu->pstate.scaling;
300 	else {
301 		policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate *
302 						cpu->pstate.scaling;
303 		/*
304 		 * The _PSS table doesn't contain whole turbo frequency range.
305 		 * This just contains +1 MHZ above the max non turbo frequency,
306 		 * with control value corresponding to max turbo ratio. But
307 		 * when cpufreq set policy is called, it will call with this
308 		 * max frequency, which will cause a reduced performance as
309 		 * this driver uses real max turbo frequency as the max
310 		 * frequeny. So correct this frequency in _PSS table to
311 		 * correct max turbo frequency based on the turbo ratio.
312 		 * Also need to convert to MHz as _PSS freq is in MHz.
313 		 */
314 		cpu->acpi_perf_data.states[0].core_frequency =
315 						turbo_pss_ctl * 100;
316 	}
317 
318 	pr_debug("intel_pstate: Updated limits using _PSS 0x%x 0x%x 0x%x\n",
319 		 cpu->pstate.min_pstate, cpu->pstate.max_pstate,
320 		 cpu->pstate.turbo_pstate);
321 	pr_debug("intel_pstate: policy max_freq=%d Khz min_freq = %d KHz\n",
322 		 policy->cpuinfo.max_freq, policy->cpuinfo.min_freq);
323 
324 	return 0;
325 }
326 
327 static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
328 {
329 	struct cpudata *cpu;
330 
331 	if (!no_acpi_perf)
332 		return 0;
333 
334 	cpu = all_cpu_data[policy->cpu];
335 	acpi_processor_unregister_performance(policy->cpu);
336 	return 0;
337 }
338 
339 #else
340 static int intel_pstate_init_perf_limits(struct cpufreq_policy *policy)
341 {
342 	return 0;
343 }
344 
345 static int intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
346 {
347 	return 0;
348 }
349 #endif
350 
351 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
352 			     int deadband, int integral) {
353 	pid->setpoint = setpoint;
354 	pid->deadband  = deadband;
355 	pid->integral  = int_tofp(integral);
356 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
357 }
358 
359 static inline void pid_p_gain_set(struct _pid *pid, int percent)
360 {
361 	pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
362 }
363 
364 static inline void pid_i_gain_set(struct _pid *pid, int percent)
365 {
366 	pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
367 }
368 
369 static inline void pid_d_gain_set(struct _pid *pid, int percent)
370 {
371 	pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
372 }
373 
374 static signed int pid_calc(struct _pid *pid, int32_t busy)
375 {
376 	signed int result;
377 	int32_t pterm, dterm, fp_error;
378 	int32_t integral_limit;
379 
380 	fp_error = int_tofp(pid->setpoint) - busy;
381 
382 	if (abs(fp_error) <= int_tofp(pid->deadband))
383 		return 0;
384 
385 	pterm = mul_fp(pid->p_gain, fp_error);
386 
387 	pid->integral += fp_error;
388 
389 	/*
390 	 * We limit the integral here so that it will never
391 	 * get higher than 30.  This prevents it from becoming
392 	 * too large an input over long periods of time and allows
393 	 * it to get factored out sooner.
394 	 *
395 	 * The value of 30 was chosen through experimentation.
396 	 */
397 	integral_limit = int_tofp(30);
398 	if (pid->integral > integral_limit)
399 		pid->integral = integral_limit;
400 	if (pid->integral < -integral_limit)
401 		pid->integral = -integral_limit;
402 
403 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
404 	pid->last_err = fp_error;
405 
406 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
407 	result = result + (1 << (FRAC_BITS-1));
408 	return (signed int)fp_toint(result);
409 }
410 
411 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
412 {
413 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
414 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
415 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
416 
417 	pid_reset(&cpu->pid, pid_params.setpoint, 100, pid_params.deadband, 0);
418 }
419 
420 static inline void intel_pstate_reset_all_pid(void)
421 {
422 	unsigned int cpu;
423 
424 	for_each_online_cpu(cpu) {
425 		if (all_cpu_data[cpu])
426 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
427 	}
428 }
429 
430 static inline void update_turbo_state(void)
431 {
432 	u64 misc_en;
433 	struct cpudata *cpu;
434 
435 	cpu = all_cpu_data[0];
436 	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
437 	limits->turbo_disabled =
438 		(misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
439 		 cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
440 }
441 
442 static void intel_pstate_hwp_set(void)
443 {
444 	int min, hw_min, max, hw_max, cpu, range, adj_range;
445 	u64 value, cap;
446 
447 	rdmsrl(MSR_HWP_CAPABILITIES, cap);
448 	hw_min = HWP_LOWEST_PERF(cap);
449 	hw_max = HWP_HIGHEST_PERF(cap);
450 	range = hw_max - hw_min;
451 
452 	get_online_cpus();
453 
454 	for_each_online_cpu(cpu) {
455 		rdmsrl_on_cpu(cpu, MSR_HWP_REQUEST, &value);
456 		adj_range = limits->min_perf_pct * range / 100;
457 		min = hw_min + adj_range;
458 		value &= ~HWP_MIN_PERF(~0L);
459 		value |= HWP_MIN_PERF(min);
460 
461 		adj_range = limits->max_perf_pct * range / 100;
462 		max = hw_min + adj_range;
463 		if (limits->no_turbo) {
464 			hw_max = HWP_GUARANTEED_PERF(cap);
465 			if (hw_max < max)
466 				max = hw_max;
467 		}
468 
469 		value &= ~HWP_MAX_PERF(~0L);
470 		value |= HWP_MAX_PERF(max);
471 		wrmsrl_on_cpu(cpu, MSR_HWP_REQUEST, value);
472 	}
473 
474 	put_online_cpus();
475 }
476 
477 /************************** debugfs begin ************************/
478 static int pid_param_set(void *data, u64 val)
479 {
480 	*(u32 *)data = val;
481 	intel_pstate_reset_all_pid();
482 	return 0;
483 }
484 
485 static int pid_param_get(void *data, u64 *val)
486 {
487 	*val = *(u32 *)data;
488 	return 0;
489 }
490 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, pid_param_set, "%llu\n");
491 
492 struct pid_param {
493 	char *name;
494 	void *value;
495 };
496 
497 static struct pid_param pid_files[] = {
498 	{"sample_rate_ms", &pid_params.sample_rate_ms},
499 	{"d_gain_pct", &pid_params.d_gain_pct},
500 	{"i_gain_pct", &pid_params.i_gain_pct},
501 	{"deadband", &pid_params.deadband},
502 	{"setpoint", &pid_params.setpoint},
503 	{"p_gain_pct", &pid_params.p_gain_pct},
504 	{NULL, NULL}
505 };
506 
507 static void __init intel_pstate_debug_expose_params(void)
508 {
509 	struct dentry *debugfs_parent;
510 	int i = 0;
511 
512 	if (hwp_active)
513 		return;
514 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
515 	if (IS_ERR_OR_NULL(debugfs_parent))
516 		return;
517 	while (pid_files[i].name) {
518 		debugfs_create_file(pid_files[i].name, 0660,
519 				    debugfs_parent, pid_files[i].value,
520 				    &fops_pid_param);
521 		i++;
522 	}
523 }
524 
525 /************************** debugfs end ************************/
526 
527 /************************** sysfs begin ************************/
528 #define show_one(file_name, object)					\
529 	static ssize_t show_##file_name					\
530 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
531 	{								\
532 		return sprintf(buf, "%u\n", limits->object);		\
533 	}
534 
535 static ssize_t show_turbo_pct(struct kobject *kobj,
536 				struct attribute *attr, char *buf)
537 {
538 	struct cpudata *cpu;
539 	int total, no_turbo, turbo_pct;
540 	uint32_t turbo_fp;
541 
542 	cpu = all_cpu_data[0];
543 
544 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
545 	no_turbo = cpu->pstate.max_pstate - cpu->pstate.min_pstate + 1;
546 	turbo_fp = div_fp(int_tofp(no_turbo), int_tofp(total));
547 	turbo_pct = 100 - fp_toint(mul_fp(turbo_fp, int_tofp(100)));
548 	return sprintf(buf, "%u\n", turbo_pct);
549 }
550 
551 static ssize_t show_num_pstates(struct kobject *kobj,
552 				struct attribute *attr, char *buf)
553 {
554 	struct cpudata *cpu;
555 	int total;
556 
557 	cpu = all_cpu_data[0];
558 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
559 	return sprintf(buf, "%u\n", total);
560 }
561 
562 static ssize_t show_no_turbo(struct kobject *kobj,
563 			     struct attribute *attr, char *buf)
564 {
565 	ssize_t ret;
566 
567 	update_turbo_state();
568 	if (limits->turbo_disabled)
569 		ret = sprintf(buf, "%u\n", limits->turbo_disabled);
570 	else
571 		ret = sprintf(buf, "%u\n", limits->no_turbo);
572 
573 	return ret;
574 }
575 
576 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
577 			      const char *buf, size_t count)
578 {
579 	unsigned int input;
580 	int ret;
581 
582 	ret = sscanf(buf, "%u", &input);
583 	if (ret != 1)
584 		return -EINVAL;
585 
586 	update_turbo_state();
587 	if (limits->turbo_disabled) {
588 		pr_warn("intel_pstate: Turbo disabled by BIOS or unavailable on processor\n");
589 		return -EPERM;
590 	}
591 
592 	limits->no_turbo = clamp_t(int, input, 0, 1);
593 
594 	if (hwp_active)
595 		intel_pstate_hwp_set();
596 
597 	return count;
598 }
599 
600 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
601 				  const char *buf, size_t count)
602 {
603 	unsigned int input;
604 	int ret;
605 
606 	ret = sscanf(buf, "%u", &input);
607 	if (ret != 1)
608 		return -EINVAL;
609 
610 	limits->max_sysfs_pct = clamp_t(int, input, 0 , 100);
611 	limits->max_perf_pct = min(limits->max_policy_pct,
612 				   limits->max_sysfs_pct);
613 	limits->max_perf_pct = max(limits->min_policy_pct,
614 				   limits->max_perf_pct);
615 	limits->max_perf_pct = max(limits->min_perf_pct,
616 				   limits->max_perf_pct);
617 	limits->max_perf = div_fp(int_tofp(limits->max_perf_pct),
618 				  int_tofp(100));
619 
620 	if (hwp_active)
621 		intel_pstate_hwp_set();
622 	return count;
623 }
624 
625 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
626 				  const char *buf, size_t count)
627 {
628 	unsigned int input;
629 	int ret;
630 
631 	ret = sscanf(buf, "%u", &input);
632 	if (ret != 1)
633 		return -EINVAL;
634 
635 	limits->min_sysfs_pct = clamp_t(int, input, 0 , 100);
636 	limits->min_perf_pct = max(limits->min_policy_pct,
637 				   limits->min_sysfs_pct);
638 	limits->min_perf_pct = min(limits->max_policy_pct,
639 				   limits->min_perf_pct);
640 	limits->min_perf_pct = min(limits->max_perf_pct,
641 				   limits->min_perf_pct);
642 	limits->min_perf = div_fp(int_tofp(limits->min_perf_pct),
643 				  int_tofp(100));
644 
645 	if (hwp_active)
646 		intel_pstate_hwp_set();
647 	return count;
648 }
649 
650 show_one(max_perf_pct, max_perf_pct);
651 show_one(min_perf_pct, min_perf_pct);
652 
653 define_one_global_rw(no_turbo);
654 define_one_global_rw(max_perf_pct);
655 define_one_global_rw(min_perf_pct);
656 define_one_global_ro(turbo_pct);
657 define_one_global_ro(num_pstates);
658 
659 static struct attribute *intel_pstate_attributes[] = {
660 	&no_turbo.attr,
661 	&max_perf_pct.attr,
662 	&min_perf_pct.attr,
663 	&turbo_pct.attr,
664 	&num_pstates.attr,
665 	NULL
666 };
667 
668 static struct attribute_group intel_pstate_attr_group = {
669 	.attrs = intel_pstate_attributes,
670 };
671 
672 static void __init intel_pstate_sysfs_expose_params(void)
673 {
674 	struct kobject *intel_pstate_kobject;
675 	int rc;
676 
677 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
678 						&cpu_subsys.dev_root->kobj);
679 	BUG_ON(!intel_pstate_kobject);
680 	rc = sysfs_create_group(intel_pstate_kobject, &intel_pstate_attr_group);
681 	BUG_ON(rc);
682 }
683 /************************** sysfs end ************************/
684 
685 static void intel_pstate_hwp_enable(struct cpudata *cpudata)
686 {
687 	wrmsrl_on_cpu(cpudata->cpu, MSR_PM_ENABLE, 0x1);
688 }
689 
690 static int byt_get_min_pstate(void)
691 {
692 	u64 value;
693 
694 	rdmsrl(BYT_RATIOS, value);
695 	return (value >> 8) & 0x7F;
696 }
697 
698 static int byt_get_max_pstate(void)
699 {
700 	u64 value;
701 
702 	rdmsrl(BYT_RATIOS, value);
703 	return (value >> 16) & 0x7F;
704 }
705 
706 static int byt_get_turbo_pstate(void)
707 {
708 	u64 value;
709 
710 	rdmsrl(BYT_TURBO_RATIOS, value);
711 	return value & 0x7F;
712 }
713 
714 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
715 {
716 	u64 val;
717 	int32_t vid_fp;
718 	u32 vid;
719 
720 	val = (u64)pstate << 8;
721 	if (limits->no_turbo && !limits->turbo_disabled)
722 		val |= (u64)1 << 32;
723 
724 	vid_fp = cpudata->vid.min + mul_fp(
725 		int_tofp(pstate - cpudata->pstate.min_pstate),
726 		cpudata->vid.ratio);
727 
728 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
729 	vid = ceiling_fp(vid_fp);
730 
731 	if (pstate > cpudata->pstate.max_pstate)
732 		vid = cpudata->vid.turbo;
733 
734 	val |= vid;
735 
736 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
737 }
738 
739 #define BYT_BCLK_FREQS 5
740 static int byt_freq_table[BYT_BCLK_FREQS] = { 833, 1000, 1333, 1167, 800};
741 
742 static int byt_get_scaling(void)
743 {
744 	u64 value;
745 	int i;
746 
747 	rdmsrl(MSR_FSB_FREQ, value);
748 	i = value & 0x3;
749 
750 	BUG_ON(i > BYT_BCLK_FREQS);
751 
752 	return byt_freq_table[i] * 100;
753 }
754 
755 static void byt_get_vid(struct cpudata *cpudata)
756 {
757 	u64 value;
758 
759 	rdmsrl(BYT_VIDS, value);
760 	cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
761 	cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
762 	cpudata->vid.ratio = div_fp(
763 		cpudata->vid.max - cpudata->vid.min,
764 		int_tofp(cpudata->pstate.max_pstate -
765 			cpudata->pstate.min_pstate));
766 
767 	rdmsrl(BYT_TURBO_VIDS, value);
768 	cpudata->vid.turbo = value & 0x7f;
769 }
770 
771 static int core_get_min_pstate(void)
772 {
773 	u64 value;
774 
775 	rdmsrl(MSR_PLATFORM_INFO, value);
776 	return (value >> 40) & 0xFF;
777 }
778 
779 static int core_get_max_pstate_physical(void)
780 {
781 	u64 value;
782 
783 	rdmsrl(MSR_PLATFORM_INFO, value);
784 	return (value >> 8) & 0xFF;
785 }
786 
787 static int core_get_max_pstate(void)
788 {
789 	u64 tar;
790 	u64 plat_info;
791 	int max_pstate;
792 	int err;
793 
794 	rdmsrl(MSR_PLATFORM_INFO, plat_info);
795 	max_pstate = (plat_info >> 8) & 0xFF;
796 
797 	err = rdmsrl_safe(MSR_TURBO_ACTIVATION_RATIO, &tar);
798 	if (!err) {
799 		/* Do some sanity checking for safety */
800 		if (plat_info & 0x600000000) {
801 			u64 tdp_ctrl;
802 			u64 tdp_ratio;
803 			int tdp_msr;
804 
805 			err = rdmsrl_safe(MSR_CONFIG_TDP_CONTROL, &tdp_ctrl);
806 			if (err)
807 				goto skip_tar;
808 
809 			tdp_msr = MSR_CONFIG_TDP_NOMINAL + tdp_ctrl;
810 			err = rdmsrl_safe(tdp_msr, &tdp_ratio);
811 			if (err)
812 				goto skip_tar;
813 
814 			if (tdp_ratio - 1 == tar) {
815 				max_pstate = tar;
816 				pr_debug("max_pstate=TAC %x\n", max_pstate);
817 			} else {
818 				goto skip_tar;
819 			}
820 		}
821 	}
822 
823 skip_tar:
824 	return max_pstate;
825 }
826 
827 static int core_get_turbo_pstate(void)
828 {
829 	u64 value;
830 	int nont, ret;
831 
832 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
833 	nont = core_get_max_pstate();
834 	ret = (value) & 255;
835 	if (ret <= nont)
836 		ret = nont;
837 	return ret;
838 }
839 
840 static inline int core_get_scaling(void)
841 {
842 	return 100000;
843 }
844 
845 static void core_set_pstate(struct cpudata *cpudata, int pstate)
846 {
847 	u64 val;
848 
849 	val = (u64)pstate << 8;
850 	if (limits->no_turbo && !limits->turbo_disabled)
851 		val |= (u64)1 << 32;
852 
853 	wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
854 }
855 
856 static int knl_get_turbo_pstate(void)
857 {
858 	u64 value;
859 	int nont, ret;
860 
861 	rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
862 	nont = core_get_max_pstate();
863 	ret = (((value) >> 8) & 0xFF);
864 	if (ret <= nont)
865 		ret = nont;
866 	return ret;
867 }
868 
869 static struct cpu_defaults core_params = {
870 	.pid_policy = {
871 		.sample_rate_ms = 10,
872 		.deadband = 0,
873 		.setpoint = 97,
874 		.p_gain_pct = 20,
875 		.d_gain_pct = 0,
876 		.i_gain_pct = 0,
877 	},
878 	.funcs = {
879 		.get_max = core_get_max_pstate,
880 		.get_max_physical = core_get_max_pstate_physical,
881 		.get_min = core_get_min_pstate,
882 		.get_turbo = core_get_turbo_pstate,
883 		.get_scaling = core_get_scaling,
884 		.set = core_set_pstate,
885 	},
886 };
887 
888 static struct cpu_defaults byt_params = {
889 	.pid_policy = {
890 		.sample_rate_ms = 10,
891 		.deadband = 0,
892 		.setpoint = 60,
893 		.p_gain_pct = 14,
894 		.d_gain_pct = 0,
895 		.i_gain_pct = 4,
896 	},
897 	.funcs = {
898 		.get_max = byt_get_max_pstate,
899 		.get_max_physical = byt_get_max_pstate,
900 		.get_min = byt_get_min_pstate,
901 		.get_turbo = byt_get_turbo_pstate,
902 		.set = byt_set_pstate,
903 		.get_scaling = byt_get_scaling,
904 		.get_vid = byt_get_vid,
905 	},
906 };
907 
908 static struct cpu_defaults knl_params = {
909 	.pid_policy = {
910 		.sample_rate_ms = 10,
911 		.deadband = 0,
912 		.setpoint = 97,
913 		.p_gain_pct = 20,
914 		.d_gain_pct = 0,
915 		.i_gain_pct = 0,
916 	},
917 	.funcs = {
918 		.get_max = core_get_max_pstate,
919 		.get_max_physical = core_get_max_pstate_physical,
920 		.get_min = core_get_min_pstate,
921 		.get_turbo = knl_get_turbo_pstate,
922 		.get_scaling = core_get_scaling,
923 		.set = core_set_pstate,
924 	},
925 };
926 
927 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
928 {
929 	int max_perf = cpu->pstate.turbo_pstate;
930 	int max_perf_adj;
931 	int min_perf;
932 
933 	if (limits->no_turbo || limits->turbo_disabled)
934 		max_perf = cpu->pstate.max_pstate;
935 
936 	/*
937 	 * performance can be limited by user through sysfs, by cpufreq
938 	 * policy, or by cpu specific default values determined through
939 	 * experimentation.
940 	 */
941 	if (limits->max_perf_ctl && limits->max_sysfs_pct >=
942 						limits->max_policy_pct) {
943 		*max = limits->max_perf_ctl;
944 	} else {
945 		max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf),
946 					limits->max_perf));
947 		*max = clamp_t(int, max_perf_adj, cpu->pstate.min_pstate,
948 			       cpu->pstate.turbo_pstate);
949 	}
950 
951 	if (limits->min_perf_ctl) {
952 		*min = limits->min_perf_ctl;
953 	} else {
954 		min_perf = fp_toint(mul_fp(int_tofp(max_perf),
955 				    limits->min_perf));
956 		*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
957 	}
958 }
959 
960 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate, bool force)
961 {
962 	int max_perf, min_perf;
963 
964 	if (force) {
965 		update_turbo_state();
966 
967 		intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
968 
969 		pstate = clamp_t(int, pstate, min_perf, max_perf);
970 
971 		if (pstate == cpu->pstate.current_pstate)
972 			return;
973 	}
974 	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
975 
976 	cpu->pstate.current_pstate = pstate;
977 
978 	pstate_funcs.set(cpu, pstate);
979 }
980 
981 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
982 {
983 	cpu->pstate.min_pstate = pstate_funcs.get_min();
984 	cpu->pstate.max_pstate = pstate_funcs.get_max();
985 	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical();
986 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
987 	cpu->pstate.scaling = pstate_funcs.get_scaling();
988 
989 	if (pstate_funcs.get_vid)
990 		pstate_funcs.get_vid(cpu);
991 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate, false);
992 }
993 
994 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
995 {
996 	struct sample *sample = &cpu->sample;
997 	int64_t core_pct;
998 
999 	core_pct = int_tofp(sample->aperf) * int_tofp(100);
1000 	core_pct = div64_u64(core_pct, int_tofp(sample->mperf));
1001 
1002 	sample->freq = fp_toint(
1003 		mul_fp(int_tofp(
1004 			cpu->pstate.max_pstate_physical *
1005 			cpu->pstate.scaling / 100),
1006 			core_pct));
1007 
1008 	sample->core_pct_busy = (int32_t)core_pct;
1009 }
1010 
1011 static inline void intel_pstate_sample(struct cpudata *cpu)
1012 {
1013 	u64 aperf, mperf;
1014 	unsigned long flags;
1015 	u64 tsc;
1016 
1017 	local_irq_save(flags);
1018 	rdmsrl(MSR_IA32_APERF, aperf);
1019 	rdmsrl(MSR_IA32_MPERF, mperf);
1020 	if (cpu->prev_mperf == mperf) {
1021 		local_irq_restore(flags);
1022 		return;
1023 	}
1024 
1025 	tsc = rdtsc();
1026 	local_irq_restore(flags);
1027 
1028 	cpu->last_sample_time = cpu->sample.time;
1029 	cpu->sample.time = ktime_get();
1030 	cpu->sample.aperf = aperf;
1031 	cpu->sample.mperf = mperf;
1032 	cpu->sample.tsc =  tsc;
1033 	cpu->sample.aperf -= cpu->prev_aperf;
1034 	cpu->sample.mperf -= cpu->prev_mperf;
1035 	cpu->sample.tsc -= cpu->prev_tsc;
1036 
1037 	intel_pstate_calc_busy(cpu);
1038 
1039 	cpu->prev_aperf = aperf;
1040 	cpu->prev_mperf = mperf;
1041 	cpu->prev_tsc = tsc;
1042 }
1043 
1044 static inline void intel_hwp_set_sample_time(struct cpudata *cpu)
1045 {
1046 	int delay;
1047 
1048 	delay = msecs_to_jiffies(50);
1049 	mod_timer_pinned(&cpu->timer, jiffies + delay);
1050 }
1051 
1052 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
1053 {
1054 	int delay;
1055 
1056 	delay = msecs_to_jiffies(pid_params.sample_rate_ms);
1057 	mod_timer_pinned(&cpu->timer, jiffies + delay);
1058 }
1059 
1060 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
1061 {
1062 	int32_t core_busy, max_pstate, current_pstate, sample_ratio;
1063 	s64 duration_us;
1064 	u32 sample_time;
1065 
1066 	/*
1067 	 * core_busy is the ratio of actual performance to max
1068 	 * max_pstate is the max non turbo pstate available
1069 	 * current_pstate was the pstate that was requested during
1070 	 * 	the last sample period.
1071 	 *
1072 	 * We normalize core_busy, which was our actual percent
1073 	 * performance to what we requested during the last sample
1074 	 * period. The result will be a percentage of busy at a
1075 	 * specified pstate.
1076 	 */
1077 	core_busy = cpu->sample.core_pct_busy;
1078 	max_pstate = int_tofp(cpu->pstate.max_pstate_physical);
1079 	current_pstate = int_tofp(cpu->pstate.current_pstate);
1080 	core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
1081 
1082 	/*
1083 	 * Since we have a deferred timer, it will not fire unless
1084 	 * we are in C0.  So, determine if the actual elapsed time
1085 	 * is significantly greater (3x) than our sample interval.  If it
1086 	 * is, then we were idle for a long enough period of time
1087 	 * to adjust our busyness.
1088 	 */
1089 	sample_time = pid_params.sample_rate_ms  * USEC_PER_MSEC;
1090 	duration_us = ktime_us_delta(cpu->sample.time,
1091 				     cpu->last_sample_time);
1092 	if (duration_us > sample_time * 3) {
1093 		sample_ratio = div_fp(int_tofp(sample_time),
1094 				      int_tofp(duration_us));
1095 		core_busy = mul_fp(core_busy, sample_ratio);
1096 	}
1097 
1098 	return core_busy;
1099 }
1100 
1101 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
1102 {
1103 	int32_t busy_scaled;
1104 	struct _pid *pid;
1105 	signed int ctl;
1106 	int from;
1107 	struct sample *sample;
1108 
1109 	from = cpu->pstate.current_pstate;
1110 
1111 	pid = &cpu->pid;
1112 	busy_scaled = intel_pstate_get_scaled_busy(cpu);
1113 
1114 	ctl = pid_calc(pid, busy_scaled);
1115 
1116 	/* Negative values of ctl increase the pstate and vice versa */
1117 	intel_pstate_set_pstate(cpu, cpu->pstate.current_pstate - ctl, true);
1118 
1119 	sample = &cpu->sample;
1120 	trace_pstate_sample(fp_toint(sample->core_pct_busy),
1121 		fp_toint(busy_scaled),
1122 		from,
1123 		cpu->pstate.current_pstate,
1124 		sample->mperf,
1125 		sample->aperf,
1126 		sample->tsc,
1127 		sample->freq);
1128 }
1129 
1130 static void intel_hwp_timer_func(unsigned long __data)
1131 {
1132 	struct cpudata *cpu = (struct cpudata *) __data;
1133 
1134 	intel_pstate_sample(cpu);
1135 	intel_hwp_set_sample_time(cpu);
1136 }
1137 
1138 static void intel_pstate_timer_func(unsigned long __data)
1139 {
1140 	struct cpudata *cpu = (struct cpudata *) __data;
1141 
1142 	intel_pstate_sample(cpu);
1143 
1144 	intel_pstate_adjust_busy_pstate(cpu);
1145 
1146 	intel_pstate_set_sample_time(cpu);
1147 }
1148 
1149 #define ICPU(model, policy) \
1150 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
1151 			(unsigned long)&policy }
1152 
1153 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
1154 	ICPU(0x2a, core_params),
1155 	ICPU(0x2d, core_params),
1156 	ICPU(0x37, byt_params),
1157 	ICPU(0x3a, core_params),
1158 	ICPU(0x3c, core_params),
1159 	ICPU(0x3d, core_params),
1160 	ICPU(0x3e, core_params),
1161 	ICPU(0x3f, core_params),
1162 	ICPU(0x45, core_params),
1163 	ICPU(0x46, core_params),
1164 	ICPU(0x47, core_params),
1165 	ICPU(0x4c, byt_params),
1166 	ICPU(0x4e, core_params),
1167 	ICPU(0x4f, core_params),
1168 	ICPU(0x5e, core_params),
1169 	ICPU(0x56, core_params),
1170 	ICPU(0x57, knl_params),
1171 	{}
1172 };
1173 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
1174 
1175 static const struct x86_cpu_id intel_pstate_cpu_oob_ids[] = {
1176 	ICPU(0x56, core_params),
1177 	{}
1178 };
1179 
1180 static int intel_pstate_init_cpu(unsigned int cpunum)
1181 {
1182 	struct cpudata *cpu;
1183 
1184 	if (!all_cpu_data[cpunum])
1185 		all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata),
1186 					       GFP_KERNEL);
1187 	if (!all_cpu_data[cpunum])
1188 		return -ENOMEM;
1189 
1190 	cpu = all_cpu_data[cpunum];
1191 
1192 	cpu->cpu = cpunum;
1193 
1194 	if (hwp_active)
1195 		intel_pstate_hwp_enable(cpu);
1196 
1197 	intel_pstate_get_cpu_pstates(cpu);
1198 
1199 	init_timer_deferrable(&cpu->timer);
1200 	cpu->timer.data = (unsigned long)cpu;
1201 	cpu->timer.expires = jiffies + HZ/100;
1202 
1203 	if (!hwp_active)
1204 		cpu->timer.function = intel_pstate_timer_func;
1205 	else
1206 		cpu->timer.function = intel_hwp_timer_func;
1207 
1208 	intel_pstate_busy_pid_reset(cpu);
1209 	intel_pstate_sample(cpu);
1210 
1211 	add_timer_on(&cpu->timer, cpunum);
1212 
1213 	pr_debug("intel_pstate: controlling: cpu %d\n", cpunum);
1214 
1215 	return 0;
1216 }
1217 
1218 static unsigned int intel_pstate_get(unsigned int cpu_num)
1219 {
1220 	struct sample *sample;
1221 	struct cpudata *cpu;
1222 
1223 	cpu = all_cpu_data[cpu_num];
1224 	if (!cpu)
1225 		return 0;
1226 	sample = &cpu->sample;
1227 	return sample->freq;
1228 }
1229 
1230 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
1231 {
1232 #if IS_ENABLED(CONFIG_ACPI)
1233 	struct cpudata *cpu;
1234 	int i;
1235 #endif
1236 	pr_debug("intel_pstate: %s max %u policy->max %u\n", __func__,
1237 		 policy->cpuinfo.max_freq, policy->max);
1238 	if (!policy->cpuinfo.max_freq)
1239 		return -ENODEV;
1240 
1241 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE &&
1242 	    policy->max >= policy->cpuinfo.max_freq) {
1243 		pr_debug("intel_pstate: set performance\n");
1244 		limits = &performance_limits;
1245 		return 0;
1246 	}
1247 
1248 	pr_debug("intel_pstate: set powersave\n");
1249 	limits = &powersave_limits;
1250 	limits->min_policy_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
1251 	limits->min_policy_pct = clamp_t(int, limits->min_policy_pct, 0 , 100);
1252 	limits->max_policy_pct = (policy->max * 100) / policy->cpuinfo.max_freq;
1253 	limits->max_policy_pct = clamp_t(int, limits->max_policy_pct, 0 , 100);
1254 
1255 	/* Normalize user input to [min_policy_pct, max_policy_pct] */
1256 	limits->min_perf_pct = max(limits->min_policy_pct,
1257 				   limits->min_sysfs_pct);
1258 	limits->min_perf_pct = min(limits->max_policy_pct,
1259 				   limits->min_perf_pct);
1260 	limits->max_perf_pct = min(limits->max_policy_pct,
1261 				   limits->max_sysfs_pct);
1262 	limits->max_perf_pct = max(limits->min_policy_pct,
1263 				   limits->max_perf_pct);
1264 
1265 	/* Make sure min_perf_pct <= max_perf_pct */
1266 	limits->min_perf_pct = min(limits->max_perf_pct, limits->min_perf_pct);
1267 
1268 	limits->min_perf = div_fp(int_tofp(limits->min_perf_pct),
1269 				  int_tofp(100));
1270 	limits->max_perf = div_fp(int_tofp(limits->max_perf_pct),
1271 				  int_tofp(100));
1272 
1273 #if IS_ENABLED(CONFIG_ACPI)
1274 	cpu = all_cpu_data[policy->cpu];
1275 	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
1276 		int control;
1277 
1278 		control = convert_to_native_pstate_format(cpu, i);
1279 		if (control * cpu->pstate.scaling == policy->max)
1280 			limits->max_perf_ctl = control;
1281 		if (control * cpu->pstate.scaling == policy->min)
1282 			limits->min_perf_ctl = control;
1283 	}
1284 
1285 	pr_debug("intel_pstate: max %u policy_max %u perf_ctl [0x%x-0x%x]\n",
1286 		 policy->cpuinfo.max_freq, policy->max, limits->min_perf_ctl,
1287 		 limits->max_perf_ctl);
1288 #endif
1289 
1290 	if (hwp_active)
1291 		intel_pstate_hwp_set();
1292 
1293 	return 0;
1294 }
1295 
1296 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
1297 {
1298 	cpufreq_verify_within_cpu_limits(policy);
1299 
1300 	if (policy->policy != CPUFREQ_POLICY_POWERSAVE &&
1301 	    policy->policy != CPUFREQ_POLICY_PERFORMANCE)
1302 		return -EINVAL;
1303 
1304 	return 0;
1305 }
1306 
1307 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
1308 {
1309 	int cpu_num = policy->cpu;
1310 	struct cpudata *cpu = all_cpu_data[cpu_num];
1311 
1312 	pr_debug("intel_pstate: CPU %d exiting\n", cpu_num);
1313 
1314 	del_timer_sync(&all_cpu_data[cpu_num]->timer);
1315 	if (hwp_active)
1316 		return;
1317 
1318 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate, false);
1319 }
1320 
1321 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
1322 {
1323 	struct cpudata *cpu;
1324 	int rc;
1325 
1326 	rc = intel_pstate_init_cpu(policy->cpu);
1327 	if (rc)
1328 		return rc;
1329 
1330 	cpu = all_cpu_data[policy->cpu];
1331 
1332 	if (limits->min_perf_pct == 100 && limits->max_perf_pct == 100)
1333 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
1334 	else
1335 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
1336 
1337 	policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
1338 	policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
1339 
1340 	/* cpuinfo and default policy values */
1341 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
1342 	policy->cpuinfo.max_freq =
1343 		cpu->pstate.turbo_pstate * cpu->pstate.scaling;
1344 	if (!no_acpi_perf)
1345 		intel_pstate_init_perf_limits(policy);
1346 	/*
1347 	 * If there is no acpi perf data or error, we ignore and use Intel P
1348 	 * state calculated limits, So this is not fatal error.
1349 	 */
1350 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
1351 	cpumask_set_cpu(policy->cpu, policy->cpus);
1352 
1353 	return 0;
1354 }
1355 
1356 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
1357 {
1358 	return intel_pstate_exit_perf_limits(policy);
1359 }
1360 
1361 static struct cpufreq_driver intel_pstate_driver = {
1362 	.flags		= CPUFREQ_CONST_LOOPS,
1363 	.verify		= intel_pstate_verify_policy,
1364 	.setpolicy	= intel_pstate_set_policy,
1365 	.get		= intel_pstate_get,
1366 	.init		= intel_pstate_cpu_init,
1367 	.exit		= intel_pstate_cpu_exit,
1368 	.stop_cpu	= intel_pstate_stop_cpu,
1369 	.name		= "intel_pstate",
1370 };
1371 
1372 static int __initdata no_load;
1373 static int __initdata no_hwp;
1374 static int __initdata hwp_only;
1375 static unsigned int force_load;
1376 
1377 static int intel_pstate_msrs_not_valid(void)
1378 {
1379 	if (!pstate_funcs.get_max() ||
1380 	    !pstate_funcs.get_min() ||
1381 	    !pstate_funcs.get_turbo())
1382 		return -ENODEV;
1383 
1384 	return 0;
1385 }
1386 
1387 static void copy_pid_params(struct pstate_adjust_policy *policy)
1388 {
1389 	pid_params.sample_rate_ms = policy->sample_rate_ms;
1390 	pid_params.p_gain_pct = policy->p_gain_pct;
1391 	pid_params.i_gain_pct = policy->i_gain_pct;
1392 	pid_params.d_gain_pct = policy->d_gain_pct;
1393 	pid_params.deadband = policy->deadband;
1394 	pid_params.setpoint = policy->setpoint;
1395 }
1396 
1397 static void copy_cpu_funcs(struct pstate_funcs *funcs)
1398 {
1399 	pstate_funcs.get_max   = funcs->get_max;
1400 	pstate_funcs.get_max_physical = funcs->get_max_physical;
1401 	pstate_funcs.get_min   = funcs->get_min;
1402 	pstate_funcs.get_turbo = funcs->get_turbo;
1403 	pstate_funcs.get_scaling = funcs->get_scaling;
1404 	pstate_funcs.set       = funcs->set;
1405 	pstate_funcs.get_vid   = funcs->get_vid;
1406 }
1407 
1408 #if IS_ENABLED(CONFIG_ACPI)
1409 
1410 static bool intel_pstate_no_acpi_pss(void)
1411 {
1412 	int i;
1413 
1414 	for_each_possible_cpu(i) {
1415 		acpi_status status;
1416 		union acpi_object *pss;
1417 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
1418 		struct acpi_processor *pr = per_cpu(processors, i);
1419 
1420 		if (!pr)
1421 			continue;
1422 
1423 		status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
1424 		if (ACPI_FAILURE(status))
1425 			continue;
1426 
1427 		pss = buffer.pointer;
1428 		if (pss && pss->type == ACPI_TYPE_PACKAGE) {
1429 			kfree(pss);
1430 			return false;
1431 		}
1432 
1433 		kfree(pss);
1434 	}
1435 
1436 	return true;
1437 }
1438 
1439 static bool intel_pstate_has_acpi_ppc(void)
1440 {
1441 	int i;
1442 
1443 	for_each_possible_cpu(i) {
1444 		struct acpi_processor *pr = per_cpu(processors, i);
1445 
1446 		if (!pr)
1447 			continue;
1448 		if (acpi_has_method(pr->handle, "_PPC"))
1449 			return true;
1450 	}
1451 	return false;
1452 }
1453 
1454 enum {
1455 	PSS,
1456 	PPC,
1457 };
1458 
1459 struct hw_vendor_info {
1460 	u16  valid;
1461 	char oem_id[ACPI_OEM_ID_SIZE];
1462 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
1463 	int  oem_pwr_table;
1464 };
1465 
1466 /* Hardware vendor-specific info that has its own power management modes */
1467 static struct hw_vendor_info vendor_info[] = {
1468 	{1, "HP    ", "ProLiant", PSS},
1469 	{1, "ORACLE", "X4-2    ", PPC},
1470 	{1, "ORACLE", "X4-2L   ", PPC},
1471 	{1, "ORACLE", "X4-2B   ", PPC},
1472 	{1, "ORACLE", "X3-2    ", PPC},
1473 	{1, "ORACLE", "X3-2L   ", PPC},
1474 	{1, "ORACLE", "X3-2B   ", PPC},
1475 	{1, "ORACLE", "X4470M2 ", PPC},
1476 	{1, "ORACLE", "X4270M3 ", PPC},
1477 	{1, "ORACLE", "X4270M2 ", PPC},
1478 	{1, "ORACLE", "X4170M2 ", PPC},
1479 	{1, "ORACLE", "X4170 M3", PPC},
1480 	{1, "ORACLE", "X4275 M3", PPC},
1481 	{1, "ORACLE", "X6-2    ", PPC},
1482 	{1, "ORACLE", "Sudbury ", PPC},
1483 	{0, "", ""},
1484 };
1485 
1486 static bool intel_pstate_platform_pwr_mgmt_exists(void)
1487 {
1488 	struct acpi_table_header hdr;
1489 	struct hw_vendor_info *v_info;
1490 	const struct x86_cpu_id *id;
1491 	u64 misc_pwr;
1492 
1493 	id = x86_match_cpu(intel_pstate_cpu_oob_ids);
1494 	if (id) {
1495 		rdmsrl(MSR_MISC_PWR_MGMT, misc_pwr);
1496 		if ( misc_pwr & (1 << 8))
1497 			return true;
1498 	}
1499 
1500 	if (acpi_disabled ||
1501 	    ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
1502 		return false;
1503 
1504 	for (v_info = vendor_info; v_info->valid; v_info++) {
1505 		if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) &&
1506 			!strncmp(hdr.oem_table_id, v_info->oem_table_id,
1507 						ACPI_OEM_TABLE_ID_SIZE))
1508 			switch (v_info->oem_pwr_table) {
1509 			case PSS:
1510 				return intel_pstate_no_acpi_pss();
1511 			case PPC:
1512 				return intel_pstate_has_acpi_ppc() &&
1513 					(!force_load);
1514 			}
1515 	}
1516 
1517 	return false;
1518 }
1519 #else /* CONFIG_ACPI not enabled */
1520 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
1521 static inline bool intel_pstate_has_acpi_ppc(void) { return false; }
1522 #endif /* CONFIG_ACPI */
1523 
1524 static int __init intel_pstate_init(void)
1525 {
1526 	int cpu, rc = 0;
1527 	const struct x86_cpu_id *id;
1528 	struct cpu_defaults *cpu_def;
1529 
1530 	if (no_load)
1531 		return -ENODEV;
1532 
1533 	id = x86_match_cpu(intel_pstate_cpu_ids);
1534 	if (!id)
1535 		return -ENODEV;
1536 
1537 	/*
1538 	 * The Intel pstate driver will be ignored if the platform
1539 	 * firmware has its own power management modes.
1540 	 */
1541 	if (intel_pstate_platform_pwr_mgmt_exists())
1542 		return -ENODEV;
1543 
1544 	cpu_def = (struct cpu_defaults *)id->driver_data;
1545 
1546 	copy_pid_params(&cpu_def->pid_policy);
1547 	copy_cpu_funcs(&cpu_def->funcs);
1548 
1549 	if (intel_pstate_msrs_not_valid())
1550 		return -ENODEV;
1551 
1552 	pr_info("Intel P-state driver initializing.\n");
1553 
1554 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
1555 	if (!all_cpu_data)
1556 		return -ENOMEM;
1557 
1558 	if (static_cpu_has_safe(X86_FEATURE_HWP) && !no_hwp) {
1559 		pr_info("intel_pstate: HWP enabled\n");
1560 		hwp_active++;
1561 	}
1562 
1563 	if (!hwp_active && hwp_only)
1564 		goto out;
1565 
1566 	rc = cpufreq_register_driver(&intel_pstate_driver);
1567 	if (rc)
1568 		goto out;
1569 
1570 	intel_pstate_debug_expose_params();
1571 	intel_pstate_sysfs_expose_params();
1572 
1573 	return rc;
1574 out:
1575 	get_online_cpus();
1576 	for_each_online_cpu(cpu) {
1577 		if (all_cpu_data[cpu]) {
1578 			del_timer_sync(&all_cpu_data[cpu]->timer);
1579 			kfree(all_cpu_data[cpu]);
1580 		}
1581 	}
1582 
1583 	put_online_cpus();
1584 	vfree(all_cpu_data);
1585 	return -ENODEV;
1586 }
1587 device_initcall(intel_pstate_init);
1588 
1589 static int __init intel_pstate_setup(char *str)
1590 {
1591 	if (!str)
1592 		return -EINVAL;
1593 
1594 	if (!strcmp(str, "disable"))
1595 		no_load = 1;
1596 	if (!strcmp(str, "no_hwp")) {
1597 		pr_info("intel_pstate: HWP disabled\n");
1598 		no_hwp = 1;
1599 	}
1600 	if (!strcmp(str, "force"))
1601 		force_load = 1;
1602 	if (!strcmp(str, "hwp_only"))
1603 		hwp_only = 1;
1604 	if (!strcmp(str, "no_acpi"))
1605 		no_acpi_perf = 1;
1606 
1607 	return 0;
1608 }
1609 early_param("intel_pstate", intel_pstate_setup);
1610 
1611 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1612 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1613 MODULE_LICENSE("GPL");
1614