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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/kernel.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/module.h>
18 #include <linux/ktime.h>
19 #include <linux/hrtimer.h>
20 #include <linux/tick.h>
21 #include <linux/slab.h>
22 #include <linux/sched.h>
23 #include <linux/list.h>
24 #include <linux/cpu.h>
25 #include <linux/cpufreq.h>
26 #include <linux/sysfs.h>
27 #include <linux/types.h>
28 #include <linux/fs.h>
29 #include <linux/debugfs.h>
30 #include <linux/acpi.h>
31 #include <linux/vmalloc.h>
32 #include <trace/events/power.h>
33 
34 #include <asm/div64.h>
35 #include <asm/msr.h>
36 #include <asm/cpu_device_id.h>
37 #include <asm/cpufeature.h>
38 #include <asm/intel-family.h>
39 
40 #define INTEL_CPUFREQ_TRANSITION_LATENCY	20000
41 
42 #define ATOM_RATIOS		0x66a
43 #define ATOM_VIDS		0x66b
44 #define ATOM_TURBO_RATIOS	0x66c
45 #define ATOM_TURBO_VIDS		0x66d
46 
47 #ifdef CONFIG_ACPI
48 #include <acpi/processor.h>
49 #include <acpi/cppc_acpi.h>
50 #endif
51 
52 #define FRAC_BITS 8
53 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
54 #define fp_toint(X) ((X) >> FRAC_BITS)
55 
56 #define EXT_BITS 6
57 #define EXT_FRAC_BITS (EXT_BITS + FRAC_BITS)
58 #define fp_ext_toint(X) ((X) >> EXT_FRAC_BITS)
59 #define int_ext_tofp(X) ((int64_t)(X) << EXT_FRAC_BITS)
60 
61 static inline int32_t mul_fp(int32_t x, int32_t y)
62 {
63 	return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
64 }
65 
66 static inline int32_t div_fp(s64 x, s64 y)
67 {
68 	return div64_s64((int64_t)x << FRAC_BITS, y);
69 }
70 
71 static inline int ceiling_fp(int32_t x)
72 {
73 	int mask, ret;
74 
75 	ret = fp_toint(x);
76 	mask = (1 << FRAC_BITS) - 1;
77 	if (x & mask)
78 		ret += 1;
79 	return ret;
80 }
81 
82 static inline u64 mul_ext_fp(u64 x, u64 y)
83 {
84 	return (x * y) >> EXT_FRAC_BITS;
85 }
86 
87 static inline u64 div_ext_fp(u64 x, u64 y)
88 {
89 	return div64_u64(x << EXT_FRAC_BITS, y);
90 }
91 
92 /**
93  * struct sample -	Store performance sample
94  * @core_avg_perf:	Ratio of APERF/MPERF which is the actual average
95  *			performance during last sample period
96  * @busy_scaled:	Scaled busy value which is used to calculate next
97  *			P state. This can be different than core_avg_perf
98  *			to account for cpu idle period
99  * @aperf:		Difference of actual performance frequency clock count
100  *			read from APERF MSR between last and current sample
101  * @mperf:		Difference of maximum performance frequency clock count
102  *			read from MPERF MSR between last and current sample
103  * @tsc:		Difference of time stamp counter between last and
104  *			current sample
105  * @time:		Current time from scheduler
106  *
107  * This structure is used in the cpudata structure to store performance sample
108  * data for choosing next P State.
109  */
110 struct sample {
111 	int32_t core_avg_perf;
112 	int32_t busy_scaled;
113 	u64 aperf;
114 	u64 mperf;
115 	u64 tsc;
116 	u64 time;
117 };
118 
119 /**
120  * struct pstate_data - Store P state data
121  * @current_pstate:	Current requested P state
122  * @min_pstate:		Min P state possible for this platform
123  * @max_pstate:		Max P state possible for this platform
124  * @max_pstate_physical:This is physical Max P state for a processor
125  *			This can be higher than the max_pstate which can
126  *			be limited by platform thermal design power limits
127  * @scaling:		Scaling factor to  convert frequency to cpufreq
128  *			frequency units
129  * @turbo_pstate:	Max Turbo P state possible for this platform
130  * @max_freq:		@max_pstate frequency in cpufreq units
131  * @turbo_freq:		@turbo_pstate frequency in cpufreq units
132  *
133  * Stores the per cpu model P state limits and current P state.
134  */
135 struct pstate_data {
136 	int	current_pstate;
137 	int	min_pstate;
138 	int	max_pstate;
139 	int	max_pstate_physical;
140 	int	scaling;
141 	int	turbo_pstate;
142 	unsigned int max_freq;
143 	unsigned int turbo_freq;
144 };
145 
146 /**
147  * struct vid_data -	Stores voltage information data
148  * @min:		VID data for this platform corresponding to
149  *			the lowest P state
150  * @max:		VID data corresponding to the highest P State.
151  * @turbo:		VID data for turbo P state
152  * @ratio:		Ratio of (vid max - vid min) /
153  *			(max P state - Min P State)
154  *
155  * Stores the voltage data for DVFS (Dynamic Voltage and Frequency Scaling)
156  * This data is used in Atom platforms, where in addition to target P state,
157  * the voltage data needs to be specified to select next P State.
158  */
159 struct vid_data {
160 	int min;
161 	int max;
162 	int turbo;
163 	int32_t ratio;
164 };
165 
166 /**
167  * struct _pid -	Stores PID data
168  * @setpoint:		Target set point for busyness or performance
169  * @integral:		Storage for accumulated error values
170  * @p_gain:		PID proportional gain
171  * @i_gain:		PID integral gain
172  * @d_gain:		PID derivative gain
173  * @deadband:		PID deadband
174  * @last_err:		Last error storage for integral part of PID calculation
175  *
176  * Stores PID coefficients and last error for PID controller.
177  */
178 struct _pid {
179 	int setpoint;
180 	int32_t integral;
181 	int32_t p_gain;
182 	int32_t i_gain;
183 	int32_t d_gain;
184 	int deadband;
185 	int32_t last_err;
186 };
187 
188 /**
189  * struct perf_limits - Store user and policy limits
190  * @no_turbo:		User requested turbo state from intel_pstate sysfs
191  * @turbo_disabled:	Platform turbo status either from msr
192  *			MSR_IA32_MISC_ENABLE or when maximum available pstate
193  *			matches the maximum turbo pstate
194  * @max_perf_pct:	Effective maximum performance limit in percentage, this
195  *			is minimum of either limits enforced by cpufreq policy
196  *			or limits from user set limits via intel_pstate sysfs
197  * @min_perf_pct:	Effective minimum performance limit in percentage, this
198  *			is maximum of either limits enforced by cpufreq policy
199  *			or limits from user set limits via intel_pstate sysfs
200  * @max_perf:		This is a scaled value between 0 to 255 for max_perf_pct
201  *			This value is used to limit max pstate
202  * @min_perf:		This is a scaled value between 0 to 255 for min_perf_pct
203  *			This value is used to limit min pstate
204  * @max_policy_pct:	The maximum performance in percentage enforced by
205  *			cpufreq setpolicy interface
206  * @max_sysfs_pct:	The maximum performance in percentage enforced by
207  *			intel pstate sysfs interface, unused when per cpu
208  *			controls are enforced
209  * @min_policy_pct:	The minimum performance in percentage enforced by
210  *			cpufreq setpolicy interface
211  * @min_sysfs_pct:	The minimum performance in percentage enforced by
212  *			intel pstate sysfs interface, unused when per cpu
213  *			controls are enforced
214  *
215  * Storage for user and policy defined limits.
216  */
217 struct perf_limits {
218 	int no_turbo;
219 	int turbo_disabled;
220 	int max_perf_pct;
221 	int min_perf_pct;
222 	int32_t max_perf;
223 	int32_t min_perf;
224 	int max_policy_pct;
225 	int max_sysfs_pct;
226 	int min_policy_pct;
227 	int min_sysfs_pct;
228 };
229 
230 /**
231  * struct cpudata -	Per CPU instance data storage
232  * @cpu:		CPU number for this instance data
233  * @policy:		CPUFreq policy value
234  * @update_util:	CPUFreq utility callback information
235  * @update_util_set:	CPUFreq utility callback is set
236  * @iowait_boost:	iowait-related boost fraction
237  * @last_update:	Time of the last update.
238  * @pstate:		Stores P state limits for this CPU
239  * @vid:		Stores VID limits for this CPU
240  * @pid:		Stores PID parameters for this CPU
241  * @last_sample_time:	Last Sample time
242  * @prev_aperf:		Last APERF value read from APERF MSR
243  * @prev_mperf:		Last MPERF value read from MPERF MSR
244  * @prev_tsc:		Last timestamp counter (TSC) value
245  * @prev_cummulative_iowait: IO Wait time difference from last and
246  *			current sample
247  * @sample:		Storage for storing last Sample data
248  * @perf_limits:	Pointer to perf_limit unique to this CPU
249  *			Not all field in the structure are applicable
250  *			when per cpu controls are enforced
251  * @acpi_perf_data:	Stores ACPI perf information read from _PSS
252  * @valid_pss_table:	Set to true for valid ACPI _PSS entries found
253  * @epp_powersave:	Last saved HWP energy performance preference
254  *			(EPP) or energy performance bias (EPB),
255  *			when policy switched to performance
256  * @epp_policy:		Last saved policy used to set EPP/EPB
257  * @epp_default:	Power on default HWP energy performance
258  *			preference/bias
259  * @epp_saved:		Saved EPP/EPB during system suspend or CPU offline
260  *			operation
261  *
262  * This structure stores per CPU instance data for all CPUs.
263  */
264 struct cpudata {
265 	int cpu;
266 
267 	unsigned int policy;
268 	struct update_util_data update_util;
269 	bool   update_util_set;
270 
271 	struct pstate_data pstate;
272 	struct vid_data vid;
273 	struct _pid pid;
274 
275 	u64	last_update;
276 	u64	last_sample_time;
277 	u64	prev_aperf;
278 	u64	prev_mperf;
279 	u64	prev_tsc;
280 	u64	prev_cummulative_iowait;
281 	struct sample sample;
282 	struct perf_limits *perf_limits;
283 #ifdef CONFIG_ACPI
284 	struct acpi_processor_performance acpi_perf_data;
285 	bool valid_pss_table;
286 #endif
287 	unsigned int iowait_boost;
288 	s16 epp_powersave;
289 	s16 epp_policy;
290 	s16 epp_default;
291 	s16 epp_saved;
292 };
293 
294 static struct cpudata **all_cpu_data;
295 
296 /**
297  * struct pstate_adjust_policy - Stores static PID configuration data
298  * @sample_rate_ms:	PID calculation sample rate in ms
299  * @sample_rate_ns:	Sample rate calculation in ns
300  * @deadband:		PID deadband
301  * @setpoint:		PID Setpoint
302  * @p_gain_pct:		PID proportional gain
303  * @i_gain_pct:		PID integral gain
304  * @d_gain_pct:		PID derivative gain
305  *
306  * Stores per CPU model static PID configuration data.
307  */
308 struct pstate_adjust_policy {
309 	int sample_rate_ms;
310 	s64 sample_rate_ns;
311 	int deadband;
312 	int setpoint;
313 	int p_gain_pct;
314 	int d_gain_pct;
315 	int i_gain_pct;
316 };
317 
318 /**
319  * struct pstate_funcs - Per CPU model specific callbacks
320  * @get_max:		Callback to get maximum non turbo effective P state
321  * @get_max_physical:	Callback to get maximum non turbo physical P state
322  * @get_min:		Callback to get minimum P state
323  * @get_turbo:		Callback to get turbo P state
324  * @get_scaling:	Callback to get frequency scaling factor
325  * @get_val:		Callback to convert P state to actual MSR write value
326  * @get_vid:		Callback to get VID data for Atom platforms
327  * @get_target_pstate:	Callback to a function to calculate next P state to use
328  *
329  * Core and Atom CPU models have different way to get P State limits. This
330  * structure is used to store those callbacks.
331  */
332 struct pstate_funcs {
333 	int (*get_max)(void);
334 	int (*get_max_physical)(void);
335 	int (*get_min)(void);
336 	int (*get_turbo)(void);
337 	int (*get_scaling)(void);
338 	u64 (*get_val)(struct cpudata*, int pstate);
339 	void (*get_vid)(struct cpudata *);
340 	int32_t (*get_target_pstate)(struct cpudata *);
341 };
342 
343 /**
344  * struct cpu_defaults- Per CPU model default config data
345  * @pid_policy:	PID config data
346  * @funcs:		Callback function data
347  */
348 struct cpu_defaults {
349 	struct pstate_adjust_policy pid_policy;
350 	struct pstate_funcs funcs;
351 };
352 
353 static inline int32_t get_target_pstate_use_performance(struct cpudata *cpu);
354 static inline int32_t get_target_pstate_use_cpu_load(struct cpudata *cpu);
355 
356 static struct pstate_adjust_policy pid_params __read_mostly;
357 static struct pstate_funcs pstate_funcs __read_mostly;
358 static int hwp_active __read_mostly;
359 static bool per_cpu_limits __read_mostly;
360 
361 static bool driver_registered __read_mostly;
362 
363 #ifdef CONFIG_ACPI
364 static bool acpi_ppc;
365 #endif
366 
367 static struct perf_limits performance_limits = {
368 	.no_turbo = 0,
369 	.turbo_disabled = 0,
370 	.max_perf_pct = 100,
371 	.max_perf = int_ext_tofp(1),
372 	.min_perf_pct = 100,
373 	.min_perf = int_ext_tofp(1),
374 	.max_policy_pct = 100,
375 	.max_sysfs_pct = 100,
376 	.min_policy_pct = 0,
377 	.min_sysfs_pct = 0,
378 };
379 
380 static struct perf_limits powersave_limits = {
381 	.no_turbo = 0,
382 	.turbo_disabled = 0,
383 	.max_perf_pct = 100,
384 	.max_perf = int_ext_tofp(1),
385 	.min_perf_pct = 0,
386 	.min_perf = 0,
387 	.max_policy_pct = 100,
388 	.max_sysfs_pct = 100,
389 	.min_policy_pct = 0,
390 	.min_sysfs_pct = 0,
391 };
392 
393 #ifdef CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE
394 static struct perf_limits *limits = &performance_limits;
395 #else
396 static struct perf_limits *limits = &powersave_limits;
397 #endif
398 
399 static DEFINE_MUTEX(intel_pstate_driver_lock);
400 static DEFINE_MUTEX(intel_pstate_limits_lock);
401 
402 #ifdef CONFIG_ACPI
403 
404 static bool intel_pstate_get_ppc_enable_status(void)
405 {
406 	if (acpi_gbl_FADT.preferred_profile == PM_ENTERPRISE_SERVER ||
407 	    acpi_gbl_FADT.preferred_profile == PM_PERFORMANCE_SERVER)
408 		return true;
409 
410 	return acpi_ppc;
411 }
412 
413 #ifdef CONFIG_ACPI_CPPC_LIB
414 
415 /* The work item is needed to avoid CPU hotplug locking issues */
416 static void intel_pstste_sched_itmt_work_fn(struct work_struct *work)
417 {
418 	sched_set_itmt_support();
419 }
420 
421 static DECLARE_WORK(sched_itmt_work, intel_pstste_sched_itmt_work_fn);
422 
423 static void intel_pstate_set_itmt_prio(int cpu)
424 {
425 	struct cppc_perf_caps cppc_perf;
426 	static u32 max_highest_perf = 0, min_highest_perf = U32_MAX;
427 	int ret;
428 
429 	ret = cppc_get_perf_caps(cpu, &cppc_perf);
430 	if (ret)
431 		return;
432 
433 	/*
434 	 * The priorities can be set regardless of whether or not
435 	 * sched_set_itmt_support(true) has been called and it is valid to
436 	 * update them at any time after it has been called.
437 	 */
438 	sched_set_itmt_core_prio(cppc_perf.highest_perf, cpu);
439 
440 	if (max_highest_perf <= min_highest_perf) {
441 		if (cppc_perf.highest_perf > max_highest_perf)
442 			max_highest_perf = cppc_perf.highest_perf;
443 
444 		if (cppc_perf.highest_perf < min_highest_perf)
445 			min_highest_perf = cppc_perf.highest_perf;
446 
447 		if (max_highest_perf > min_highest_perf) {
448 			/*
449 			 * This code can be run during CPU online under the
450 			 * CPU hotplug locks, so sched_set_itmt_support()
451 			 * cannot be called from here.  Queue up a work item
452 			 * to invoke it.
453 			 */
454 			schedule_work(&sched_itmt_work);
455 		}
456 	}
457 }
458 #else
459 static void intel_pstate_set_itmt_prio(int cpu)
460 {
461 }
462 #endif
463 
464 static void intel_pstate_init_acpi_perf_limits(struct cpufreq_policy *policy)
465 {
466 	struct cpudata *cpu;
467 	int ret;
468 	int i;
469 
470 	if (hwp_active) {
471 		intel_pstate_set_itmt_prio(policy->cpu);
472 		return;
473 	}
474 
475 	if (!intel_pstate_get_ppc_enable_status())
476 		return;
477 
478 	cpu = all_cpu_data[policy->cpu];
479 
480 	ret = acpi_processor_register_performance(&cpu->acpi_perf_data,
481 						  policy->cpu);
482 	if (ret)
483 		return;
484 
485 	/*
486 	 * Check if the control value in _PSS is for PERF_CTL MSR, which should
487 	 * guarantee that the states returned by it map to the states in our
488 	 * list directly.
489 	 */
490 	if (cpu->acpi_perf_data.control_register.space_id !=
491 						ACPI_ADR_SPACE_FIXED_HARDWARE)
492 		goto err;
493 
494 	/*
495 	 * If there is only one entry _PSS, simply ignore _PSS and continue as
496 	 * usual without taking _PSS into account
497 	 */
498 	if (cpu->acpi_perf_data.state_count < 2)
499 		goto err;
500 
501 	pr_debug("CPU%u - ACPI _PSS perf data\n", policy->cpu);
502 	for (i = 0; i < cpu->acpi_perf_data.state_count; i++) {
503 		pr_debug("     %cP%d: %u MHz, %u mW, 0x%x\n",
504 			 (i == cpu->acpi_perf_data.state ? '*' : ' '), i,
505 			 (u32) cpu->acpi_perf_data.states[i].core_frequency,
506 			 (u32) cpu->acpi_perf_data.states[i].power,
507 			 (u32) cpu->acpi_perf_data.states[i].control);
508 	}
509 
510 	/*
511 	 * The _PSS table doesn't contain whole turbo frequency range.
512 	 * This just contains +1 MHZ above the max non turbo frequency,
513 	 * with control value corresponding to max turbo ratio. But
514 	 * when cpufreq set policy is called, it will call with this
515 	 * max frequency, which will cause a reduced performance as
516 	 * this driver uses real max turbo frequency as the max
517 	 * frequency. So correct this frequency in _PSS table to
518 	 * correct max turbo frequency based on the turbo state.
519 	 * Also need to convert to MHz as _PSS freq is in MHz.
520 	 */
521 	if (!limits->turbo_disabled)
522 		cpu->acpi_perf_data.states[0].core_frequency =
523 					policy->cpuinfo.max_freq / 1000;
524 	cpu->valid_pss_table = true;
525 	pr_debug("_PPC limits will be enforced\n");
526 
527 	return;
528 
529  err:
530 	cpu->valid_pss_table = false;
531 	acpi_processor_unregister_performance(policy->cpu);
532 }
533 
534 static void intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
535 {
536 	struct cpudata *cpu;
537 
538 	cpu = all_cpu_data[policy->cpu];
539 	if (!cpu->valid_pss_table)
540 		return;
541 
542 	acpi_processor_unregister_performance(policy->cpu);
543 }
544 #else
545 static inline void intel_pstate_init_acpi_perf_limits(struct cpufreq_policy *policy)
546 {
547 }
548 
549 static inline void intel_pstate_exit_perf_limits(struct cpufreq_policy *policy)
550 {
551 }
552 #endif
553 
554 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
555 			     int deadband, int integral) {
556 	pid->setpoint = int_tofp(setpoint);
557 	pid->deadband  = int_tofp(deadband);
558 	pid->integral  = int_tofp(integral);
559 	pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
560 }
561 
562 static inline void pid_p_gain_set(struct _pid *pid, int percent)
563 {
564 	pid->p_gain = div_fp(percent, 100);
565 }
566 
567 static inline void pid_i_gain_set(struct _pid *pid, int percent)
568 {
569 	pid->i_gain = div_fp(percent, 100);
570 }
571 
572 static inline void pid_d_gain_set(struct _pid *pid, int percent)
573 {
574 	pid->d_gain = div_fp(percent, 100);
575 }
576 
577 static signed int pid_calc(struct _pid *pid, int32_t busy)
578 {
579 	signed int result;
580 	int32_t pterm, dterm, fp_error;
581 	int32_t integral_limit;
582 
583 	fp_error = pid->setpoint - busy;
584 
585 	if (abs(fp_error) <= pid->deadband)
586 		return 0;
587 
588 	pterm = mul_fp(pid->p_gain, fp_error);
589 
590 	pid->integral += fp_error;
591 
592 	/*
593 	 * We limit the integral here so that it will never
594 	 * get higher than 30.  This prevents it from becoming
595 	 * too large an input over long periods of time and allows
596 	 * it to get factored out sooner.
597 	 *
598 	 * The value of 30 was chosen through experimentation.
599 	 */
600 	integral_limit = int_tofp(30);
601 	if (pid->integral > integral_limit)
602 		pid->integral = integral_limit;
603 	if (pid->integral < -integral_limit)
604 		pid->integral = -integral_limit;
605 
606 	dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
607 	pid->last_err = fp_error;
608 
609 	result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
610 	result = result + (1 << (FRAC_BITS-1));
611 	return (signed int)fp_toint(result);
612 }
613 
614 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
615 {
616 	pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
617 	pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
618 	pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
619 
620 	pid_reset(&cpu->pid, pid_params.setpoint, 100, pid_params.deadband, 0);
621 }
622 
623 static inline void intel_pstate_reset_all_pid(void)
624 {
625 	unsigned int cpu;
626 
627 	for_each_online_cpu(cpu) {
628 		if (all_cpu_data[cpu])
629 			intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
630 	}
631 }
632 
633 static inline void update_turbo_state(void)
634 {
635 	u64 misc_en;
636 	struct cpudata *cpu;
637 
638 	cpu = all_cpu_data[0];
639 	rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
640 	limits->turbo_disabled =
641 		(misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
642 		 cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
643 }
644 
645 static s16 intel_pstate_get_epb(struct cpudata *cpu_data)
646 {
647 	u64 epb;
648 	int ret;
649 
650 	if (!static_cpu_has(X86_FEATURE_EPB))
651 		return -ENXIO;
652 
653 	ret = rdmsrl_on_cpu(cpu_data->cpu, MSR_IA32_ENERGY_PERF_BIAS, &epb);
654 	if (ret)
655 		return (s16)ret;
656 
657 	return (s16)(epb & 0x0f);
658 }
659 
660 static s16 intel_pstate_get_epp(struct cpudata *cpu_data, u64 hwp_req_data)
661 {
662 	s16 epp;
663 
664 	if (static_cpu_has(X86_FEATURE_HWP_EPP)) {
665 		/*
666 		 * When hwp_req_data is 0, means that caller didn't read
667 		 * MSR_HWP_REQUEST, so need to read and get EPP.
668 		 */
669 		if (!hwp_req_data) {
670 			epp = rdmsrl_on_cpu(cpu_data->cpu, MSR_HWP_REQUEST,
671 					    &hwp_req_data);
672 			if (epp)
673 				return epp;
674 		}
675 		epp = (hwp_req_data >> 24) & 0xff;
676 	} else {
677 		/* When there is no EPP present, HWP uses EPB settings */
678 		epp = intel_pstate_get_epb(cpu_data);
679 	}
680 
681 	return epp;
682 }
683 
684 static int intel_pstate_set_epb(int cpu, s16 pref)
685 {
686 	u64 epb;
687 	int ret;
688 
689 	if (!static_cpu_has(X86_FEATURE_EPB))
690 		return -ENXIO;
691 
692 	ret = rdmsrl_on_cpu(cpu, MSR_IA32_ENERGY_PERF_BIAS, &epb);
693 	if (ret)
694 		return ret;
695 
696 	epb = (epb & ~0x0f) | pref;
697 	wrmsrl_on_cpu(cpu, MSR_IA32_ENERGY_PERF_BIAS, epb);
698 
699 	return 0;
700 }
701 
702 /*
703  * EPP/EPB display strings corresponding to EPP index in the
704  * energy_perf_strings[]
705  *	index		String
706  *-------------------------------------
707  *	0		default
708  *	1		performance
709  *	2		balance_performance
710  *	3		balance_power
711  *	4		power
712  */
713 static const char * const energy_perf_strings[] = {
714 	"default",
715 	"performance",
716 	"balance_performance",
717 	"balance_power",
718 	"power",
719 	NULL
720 };
721 
722 static int intel_pstate_get_energy_pref_index(struct cpudata *cpu_data)
723 {
724 	s16 epp;
725 	int index = -EINVAL;
726 
727 	epp = intel_pstate_get_epp(cpu_data, 0);
728 	if (epp < 0)
729 		return epp;
730 
731 	if (static_cpu_has(X86_FEATURE_HWP_EPP)) {
732 		/*
733 		 * Range:
734 		 *	0x00-0x3F	:	Performance
735 		 *	0x40-0x7F	:	Balance performance
736 		 *	0x80-0xBF	:	Balance power
737 		 *	0xC0-0xFF	:	Power
738 		 * The EPP is a 8 bit value, but our ranges restrict the
739 		 * value which can be set. Here only using top two bits
740 		 * effectively.
741 		 */
742 		index = (epp >> 6) + 1;
743 	} else if (static_cpu_has(X86_FEATURE_EPB)) {
744 		/*
745 		 * Range:
746 		 *	0x00-0x03	:	Performance
747 		 *	0x04-0x07	:	Balance performance
748 		 *	0x08-0x0B	:	Balance power
749 		 *	0x0C-0x0F	:	Power
750 		 * The EPB is a 4 bit value, but our ranges restrict the
751 		 * value which can be set. Here only using top two bits
752 		 * effectively.
753 		 */
754 		index = (epp >> 2) + 1;
755 	}
756 
757 	return index;
758 }
759 
760 static int intel_pstate_set_energy_pref_index(struct cpudata *cpu_data,
761 					      int pref_index)
762 {
763 	int epp = -EINVAL;
764 	int ret;
765 
766 	if (!pref_index)
767 		epp = cpu_data->epp_default;
768 
769 	mutex_lock(&intel_pstate_limits_lock);
770 
771 	if (static_cpu_has(X86_FEATURE_HWP_EPP)) {
772 		u64 value;
773 
774 		ret = rdmsrl_on_cpu(cpu_data->cpu, MSR_HWP_REQUEST, &value);
775 		if (ret)
776 			goto return_pref;
777 
778 		value &= ~GENMASK_ULL(31, 24);
779 
780 		/*
781 		 * If epp is not default, convert from index into
782 		 * energy_perf_strings to epp value, by shifting 6
783 		 * bits left to use only top two bits in epp.
784 		 * The resultant epp need to shifted by 24 bits to
785 		 * epp position in MSR_HWP_REQUEST.
786 		 */
787 		if (epp == -EINVAL)
788 			epp = (pref_index - 1) << 6;
789 
790 		value |= (u64)epp << 24;
791 		ret = wrmsrl_on_cpu(cpu_data->cpu, MSR_HWP_REQUEST, value);
792 	} else {
793 		if (epp == -EINVAL)
794 			epp = (pref_index - 1) << 2;
795 		ret = intel_pstate_set_epb(cpu_data->cpu, epp);
796 	}
797 return_pref:
798 	mutex_unlock(&intel_pstate_limits_lock);
799 
800 	return ret;
801 }
802 
803 static ssize_t show_energy_performance_available_preferences(
804 				struct cpufreq_policy *policy, char *buf)
805 {
806 	int i = 0;
807 	int ret = 0;
808 
809 	while (energy_perf_strings[i] != NULL)
810 		ret += sprintf(&buf[ret], "%s ", energy_perf_strings[i++]);
811 
812 	ret += sprintf(&buf[ret], "\n");
813 
814 	return ret;
815 }
816 
817 cpufreq_freq_attr_ro(energy_performance_available_preferences);
818 
819 static ssize_t store_energy_performance_preference(
820 		struct cpufreq_policy *policy, const char *buf, size_t count)
821 {
822 	struct cpudata *cpu_data = all_cpu_data[policy->cpu];
823 	char str_preference[21];
824 	int ret, i = 0;
825 
826 	ret = sscanf(buf, "%20s", str_preference);
827 	if (ret != 1)
828 		return -EINVAL;
829 
830 	while (energy_perf_strings[i] != NULL) {
831 		if (!strcmp(str_preference, energy_perf_strings[i])) {
832 			intel_pstate_set_energy_pref_index(cpu_data, i);
833 			return count;
834 		}
835 		++i;
836 	}
837 
838 	return -EINVAL;
839 }
840 
841 static ssize_t show_energy_performance_preference(
842 				struct cpufreq_policy *policy, char *buf)
843 {
844 	struct cpudata *cpu_data = all_cpu_data[policy->cpu];
845 	int preference;
846 
847 	preference = intel_pstate_get_energy_pref_index(cpu_data);
848 	if (preference < 0)
849 		return preference;
850 
851 	return  sprintf(buf, "%s\n", energy_perf_strings[preference]);
852 }
853 
854 cpufreq_freq_attr_rw(energy_performance_preference);
855 
856 static struct freq_attr *hwp_cpufreq_attrs[] = {
857 	&energy_performance_preference,
858 	&energy_performance_available_preferences,
859 	NULL,
860 };
861 
862 static void intel_pstate_hwp_set(struct cpufreq_policy *policy)
863 {
864 	int min, hw_min, max, hw_max, cpu, range, adj_range;
865 	struct perf_limits *perf_limits = limits;
866 	u64 value, cap;
867 
868 	for_each_cpu(cpu, policy->cpus) {
869 		int max_perf_pct, min_perf_pct;
870 		struct cpudata *cpu_data = all_cpu_data[cpu];
871 		s16 epp;
872 
873 		if (per_cpu_limits)
874 			perf_limits = all_cpu_data[cpu]->perf_limits;
875 
876 		rdmsrl_on_cpu(cpu, MSR_HWP_CAPABILITIES, &cap);
877 		hw_min = HWP_LOWEST_PERF(cap);
878 		if (limits->no_turbo)
879 			hw_max = HWP_GUARANTEED_PERF(cap);
880 		else
881 			hw_max = HWP_HIGHEST_PERF(cap);
882 		range = hw_max - hw_min;
883 
884 		max_perf_pct = perf_limits->max_perf_pct;
885 		min_perf_pct = perf_limits->min_perf_pct;
886 
887 		rdmsrl_on_cpu(cpu, MSR_HWP_REQUEST, &value);
888 		adj_range = min_perf_pct * range / 100;
889 		min = hw_min + adj_range;
890 		value &= ~HWP_MIN_PERF(~0L);
891 		value |= HWP_MIN_PERF(min);
892 
893 		adj_range = max_perf_pct * range / 100;
894 		max = hw_min + adj_range;
895 
896 		value &= ~HWP_MAX_PERF(~0L);
897 		value |= HWP_MAX_PERF(max);
898 
899 		if (cpu_data->epp_policy == cpu_data->policy)
900 			goto skip_epp;
901 
902 		cpu_data->epp_policy = cpu_data->policy;
903 
904 		if (cpu_data->epp_saved >= 0) {
905 			epp = cpu_data->epp_saved;
906 			cpu_data->epp_saved = -EINVAL;
907 			goto update_epp;
908 		}
909 
910 		if (cpu_data->policy == CPUFREQ_POLICY_PERFORMANCE) {
911 			epp = intel_pstate_get_epp(cpu_data, value);
912 			cpu_data->epp_powersave = epp;
913 			/* If EPP read was failed, then don't try to write */
914 			if (epp < 0)
915 				goto skip_epp;
916 
917 
918 			epp = 0;
919 		} else {
920 			/* skip setting EPP, when saved value is invalid */
921 			if (cpu_data->epp_powersave < 0)
922 				goto skip_epp;
923 
924 			/*
925 			 * No need to restore EPP when it is not zero. This
926 			 * means:
927 			 *  - Policy is not changed
928 			 *  - user has manually changed
929 			 *  - Error reading EPB
930 			 */
931 			epp = intel_pstate_get_epp(cpu_data, value);
932 			if (epp)
933 				goto skip_epp;
934 
935 			epp = cpu_data->epp_powersave;
936 		}
937 update_epp:
938 		if (static_cpu_has(X86_FEATURE_HWP_EPP)) {
939 			value &= ~GENMASK_ULL(31, 24);
940 			value |= (u64)epp << 24;
941 		} else {
942 			intel_pstate_set_epb(cpu, epp);
943 		}
944 skip_epp:
945 		wrmsrl_on_cpu(cpu, MSR_HWP_REQUEST, value);
946 	}
947 }
948 
949 static int intel_pstate_hwp_set_policy(struct cpufreq_policy *policy)
950 {
951 	if (hwp_active)
952 		intel_pstate_hwp_set(policy);
953 
954 	return 0;
955 }
956 
957 static int intel_pstate_hwp_save_state(struct cpufreq_policy *policy)
958 {
959 	struct cpudata *cpu_data = all_cpu_data[policy->cpu];
960 
961 	if (!hwp_active)
962 		return 0;
963 
964 	cpu_data->epp_saved = intel_pstate_get_epp(cpu_data, 0);
965 
966 	return 0;
967 }
968 
969 static int intel_pstate_resume(struct cpufreq_policy *policy)
970 {
971 	int ret;
972 
973 	if (!hwp_active)
974 		return 0;
975 
976 	mutex_lock(&intel_pstate_limits_lock);
977 
978 	all_cpu_data[policy->cpu]->epp_policy = 0;
979 
980 	ret = intel_pstate_hwp_set_policy(policy);
981 
982 	mutex_unlock(&intel_pstate_limits_lock);
983 
984 	return ret;
985 }
986 
987 static void intel_pstate_update_policies(void)
988 {
989 	int cpu;
990 
991 	for_each_possible_cpu(cpu)
992 		cpufreq_update_policy(cpu);
993 }
994 
995 /************************** debugfs begin ************************/
996 static int pid_param_set(void *data, u64 val)
997 {
998 	*(u32 *)data = val;
999 	intel_pstate_reset_all_pid();
1000 	return 0;
1001 }
1002 
1003 static int pid_param_get(void *data, u64 *val)
1004 {
1005 	*val = *(u32 *)data;
1006 	return 0;
1007 }
1008 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, pid_param_set, "%llu\n");
1009 
1010 static struct dentry *debugfs_parent;
1011 
1012 struct pid_param {
1013 	char *name;
1014 	void *value;
1015 	struct dentry *dentry;
1016 };
1017 
1018 static struct pid_param pid_files[] = {
1019 	{"sample_rate_ms", &pid_params.sample_rate_ms, },
1020 	{"d_gain_pct", &pid_params.d_gain_pct, },
1021 	{"i_gain_pct", &pid_params.i_gain_pct, },
1022 	{"deadband", &pid_params.deadband, },
1023 	{"setpoint", &pid_params.setpoint, },
1024 	{"p_gain_pct", &pid_params.p_gain_pct, },
1025 	{NULL, NULL, }
1026 };
1027 
1028 static void intel_pstate_debug_expose_params(void)
1029 {
1030 	int i;
1031 
1032 	debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
1033 	if (IS_ERR_OR_NULL(debugfs_parent))
1034 		return;
1035 
1036 	for (i = 0; pid_files[i].name; i++) {
1037 		struct dentry *dentry;
1038 
1039 		dentry = debugfs_create_file(pid_files[i].name, 0660,
1040 					     debugfs_parent, pid_files[i].value,
1041 					     &fops_pid_param);
1042 		if (!IS_ERR(dentry))
1043 			pid_files[i].dentry = dentry;
1044 	}
1045 }
1046 
1047 static void intel_pstate_debug_hide_params(void)
1048 {
1049 	int i;
1050 
1051 	if (IS_ERR_OR_NULL(debugfs_parent))
1052 		return;
1053 
1054 	for (i = 0; pid_files[i].name; i++) {
1055 		debugfs_remove(pid_files[i].dentry);
1056 		pid_files[i].dentry = NULL;
1057 	}
1058 
1059 	debugfs_remove(debugfs_parent);
1060 	debugfs_parent = NULL;
1061 }
1062 
1063 /************************** debugfs end ************************/
1064 
1065 /************************** sysfs begin ************************/
1066 #define show_one(file_name, object)					\
1067 	static ssize_t show_##file_name					\
1068 	(struct kobject *kobj, struct attribute *attr, char *buf)	\
1069 	{								\
1070 		return sprintf(buf, "%u\n", limits->object);		\
1071 	}
1072 
1073 static ssize_t intel_pstate_show_status(char *buf);
1074 static int intel_pstate_update_status(const char *buf, size_t size);
1075 
1076 static ssize_t show_status(struct kobject *kobj,
1077 			   struct attribute *attr, char *buf)
1078 {
1079 	ssize_t ret;
1080 
1081 	mutex_lock(&intel_pstate_driver_lock);
1082 	ret = intel_pstate_show_status(buf);
1083 	mutex_unlock(&intel_pstate_driver_lock);
1084 
1085 	return ret;
1086 }
1087 
1088 static ssize_t store_status(struct kobject *a, struct attribute *b,
1089 			    const char *buf, size_t count)
1090 {
1091 	char *p = memchr(buf, '\n', count);
1092 	int ret;
1093 
1094 	mutex_lock(&intel_pstate_driver_lock);
1095 	ret = intel_pstate_update_status(buf, p ? p - buf : count);
1096 	mutex_unlock(&intel_pstate_driver_lock);
1097 
1098 	return ret < 0 ? ret : count;
1099 }
1100 
1101 static ssize_t show_turbo_pct(struct kobject *kobj,
1102 				struct attribute *attr, char *buf)
1103 {
1104 	struct cpudata *cpu;
1105 	int total, no_turbo, turbo_pct;
1106 	uint32_t turbo_fp;
1107 
1108 	mutex_lock(&intel_pstate_driver_lock);
1109 
1110 	if (!driver_registered) {
1111 		mutex_unlock(&intel_pstate_driver_lock);
1112 		return -EAGAIN;
1113 	}
1114 
1115 	cpu = all_cpu_data[0];
1116 
1117 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
1118 	no_turbo = cpu->pstate.max_pstate - cpu->pstate.min_pstate + 1;
1119 	turbo_fp = div_fp(no_turbo, total);
1120 	turbo_pct = 100 - fp_toint(mul_fp(turbo_fp, int_tofp(100)));
1121 
1122 	mutex_unlock(&intel_pstate_driver_lock);
1123 
1124 	return sprintf(buf, "%u\n", turbo_pct);
1125 }
1126 
1127 static ssize_t show_num_pstates(struct kobject *kobj,
1128 				struct attribute *attr, char *buf)
1129 {
1130 	struct cpudata *cpu;
1131 	int total;
1132 
1133 	mutex_lock(&intel_pstate_driver_lock);
1134 
1135 	if (!driver_registered) {
1136 		mutex_unlock(&intel_pstate_driver_lock);
1137 		return -EAGAIN;
1138 	}
1139 
1140 	cpu = all_cpu_data[0];
1141 	total = cpu->pstate.turbo_pstate - cpu->pstate.min_pstate + 1;
1142 
1143 	mutex_unlock(&intel_pstate_driver_lock);
1144 
1145 	return sprintf(buf, "%u\n", total);
1146 }
1147 
1148 static ssize_t show_no_turbo(struct kobject *kobj,
1149 			     struct attribute *attr, char *buf)
1150 {
1151 	ssize_t ret;
1152 
1153 	mutex_lock(&intel_pstate_driver_lock);
1154 
1155 	if (!driver_registered) {
1156 		mutex_unlock(&intel_pstate_driver_lock);
1157 		return -EAGAIN;
1158 	}
1159 
1160 	update_turbo_state();
1161 	if (limits->turbo_disabled)
1162 		ret = sprintf(buf, "%u\n", limits->turbo_disabled);
1163 	else
1164 		ret = sprintf(buf, "%u\n", limits->no_turbo);
1165 
1166 	mutex_unlock(&intel_pstate_driver_lock);
1167 
1168 	return ret;
1169 }
1170 
1171 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
1172 			      const char *buf, size_t count)
1173 {
1174 	unsigned int input;
1175 	int ret;
1176 
1177 	ret = sscanf(buf, "%u", &input);
1178 	if (ret != 1)
1179 		return -EINVAL;
1180 
1181 	mutex_lock(&intel_pstate_driver_lock);
1182 
1183 	if (!driver_registered) {
1184 		mutex_unlock(&intel_pstate_driver_lock);
1185 		return -EAGAIN;
1186 	}
1187 
1188 	mutex_lock(&intel_pstate_limits_lock);
1189 
1190 	update_turbo_state();
1191 	if (limits->turbo_disabled) {
1192 		pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
1193 		mutex_unlock(&intel_pstate_limits_lock);
1194 		mutex_unlock(&intel_pstate_driver_lock);
1195 		return -EPERM;
1196 	}
1197 
1198 	limits->no_turbo = clamp_t(int, input, 0, 1);
1199 
1200 	mutex_unlock(&intel_pstate_limits_lock);
1201 
1202 	intel_pstate_update_policies();
1203 
1204 	mutex_unlock(&intel_pstate_driver_lock);
1205 
1206 	return count;
1207 }
1208 
1209 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
1210 				  const char *buf, size_t count)
1211 {
1212 	unsigned int input;
1213 	int ret;
1214 
1215 	ret = sscanf(buf, "%u", &input);
1216 	if (ret != 1)
1217 		return -EINVAL;
1218 
1219 	mutex_lock(&intel_pstate_driver_lock);
1220 
1221 	if (!driver_registered) {
1222 		mutex_unlock(&intel_pstate_driver_lock);
1223 		return -EAGAIN;
1224 	}
1225 
1226 	mutex_lock(&intel_pstate_limits_lock);
1227 
1228 	limits->max_sysfs_pct = clamp_t(int, input, 0 , 100);
1229 	limits->max_perf_pct = min(limits->max_policy_pct,
1230 				   limits->max_sysfs_pct);
1231 	limits->max_perf_pct = max(limits->min_policy_pct,
1232 				   limits->max_perf_pct);
1233 	limits->max_perf_pct = max(limits->min_perf_pct,
1234 				   limits->max_perf_pct);
1235 	limits->max_perf = div_ext_fp(limits->max_perf_pct, 100);
1236 
1237 	mutex_unlock(&intel_pstate_limits_lock);
1238 
1239 	intel_pstate_update_policies();
1240 
1241 	mutex_unlock(&intel_pstate_driver_lock);
1242 
1243 	return count;
1244 }
1245 
1246 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
1247 				  const char *buf, size_t count)
1248 {
1249 	unsigned int input;
1250 	int ret;
1251 
1252 	ret = sscanf(buf, "%u", &input);
1253 	if (ret != 1)
1254 		return -EINVAL;
1255 
1256 	mutex_lock(&intel_pstate_driver_lock);
1257 
1258 	if (!driver_registered) {
1259 		mutex_unlock(&intel_pstate_driver_lock);
1260 		return -EAGAIN;
1261 	}
1262 
1263 	mutex_lock(&intel_pstate_limits_lock);
1264 
1265 	limits->min_sysfs_pct = clamp_t(int, input, 0 , 100);
1266 	limits->min_perf_pct = max(limits->min_policy_pct,
1267 				   limits->min_sysfs_pct);
1268 	limits->min_perf_pct = min(limits->max_policy_pct,
1269 				   limits->min_perf_pct);
1270 	limits->min_perf_pct = min(limits->max_perf_pct,
1271 				   limits->min_perf_pct);
1272 	limits->min_perf = div_ext_fp(limits->min_perf_pct, 100);
1273 
1274 	mutex_unlock(&intel_pstate_limits_lock);
1275 
1276 	intel_pstate_update_policies();
1277 
1278 	mutex_unlock(&intel_pstate_driver_lock);
1279 
1280 	return count;
1281 }
1282 
1283 show_one(max_perf_pct, max_perf_pct);
1284 show_one(min_perf_pct, min_perf_pct);
1285 
1286 define_one_global_rw(status);
1287 define_one_global_rw(no_turbo);
1288 define_one_global_rw(max_perf_pct);
1289 define_one_global_rw(min_perf_pct);
1290 define_one_global_ro(turbo_pct);
1291 define_one_global_ro(num_pstates);
1292 
1293 static struct attribute *intel_pstate_attributes[] = {
1294 	&status.attr,
1295 	&no_turbo.attr,
1296 	&turbo_pct.attr,
1297 	&num_pstates.attr,
1298 	NULL
1299 };
1300 
1301 static struct attribute_group intel_pstate_attr_group = {
1302 	.attrs = intel_pstate_attributes,
1303 };
1304 
1305 static void __init intel_pstate_sysfs_expose_params(void)
1306 {
1307 	struct kobject *intel_pstate_kobject;
1308 	int rc;
1309 
1310 	intel_pstate_kobject = kobject_create_and_add("intel_pstate",
1311 						&cpu_subsys.dev_root->kobj);
1312 	if (WARN_ON(!intel_pstate_kobject))
1313 		return;
1314 
1315 	rc = sysfs_create_group(intel_pstate_kobject, &intel_pstate_attr_group);
1316 	if (WARN_ON(rc))
1317 		return;
1318 
1319 	/*
1320 	 * If per cpu limits are enforced there are no global limits, so
1321 	 * return without creating max/min_perf_pct attributes
1322 	 */
1323 	if (per_cpu_limits)
1324 		return;
1325 
1326 	rc = sysfs_create_file(intel_pstate_kobject, &max_perf_pct.attr);
1327 	WARN_ON(rc);
1328 
1329 	rc = sysfs_create_file(intel_pstate_kobject, &min_perf_pct.attr);
1330 	WARN_ON(rc);
1331 
1332 }
1333 /************************** sysfs end ************************/
1334 
1335 static void intel_pstate_hwp_enable(struct cpudata *cpudata)
1336 {
1337 	/* First disable HWP notification interrupt as we don't process them */
1338 	if (static_cpu_has(X86_FEATURE_HWP_NOTIFY))
1339 		wrmsrl_on_cpu(cpudata->cpu, MSR_HWP_INTERRUPT, 0x00);
1340 
1341 	wrmsrl_on_cpu(cpudata->cpu, MSR_PM_ENABLE, 0x1);
1342 	cpudata->epp_policy = 0;
1343 	if (cpudata->epp_default == -EINVAL)
1344 		cpudata->epp_default = intel_pstate_get_epp(cpudata, 0);
1345 }
1346 
1347 #define MSR_IA32_POWER_CTL_BIT_EE	19
1348 
1349 /* Disable energy efficiency optimization */
1350 static void intel_pstate_disable_ee(int cpu)
1351 {
1352 	u64 power_ctl;
1353 	int ret;
1354 
1355 	ret = rdmsrl_on_cpu(cpu, MSR_IA32_POWER_CTL, &power_ctl);
1356 	if (ret)
1357 		return;
1358 
1359 	if (!(power_ctl & BIT(MSR_IA32_POWER_CTL_BIT_EE))) {
1360 		pr_info("Disabling energy efficiency optimization\n");
1361 		power_ctl |= BIT(MSR_IA32_POWER_CTL_BIT_EE);
1362 		wrmsrl_on_cpu(cpu, MSR_IA32_POWER_CTL, power_ctl);
1363 	}
1364 }
1365 
1366 static int atom_get_min_pstate(void)
1367 {
1368 	u64 value;
1369 
1370 	rdmsrl(ATOM_RATIOS, value);
1371 	return (value >> 8) & 0x7F;
1372 }
1373 
1374 static int atom_get_max_pstate(void)
1375 {
1376 	u64 value;
1377 
1378 	rdmsrl(ATOM_RATIOS, value);
1379 	return (value >> 16) & 0x7F;
1380 }
1381 
1382 static int atom_get_turbo_pstate(void)
1383 {
1384 	u64 value;
1385 
1386 	rdmsrl(ATOM_TURBO_RATIOS, value);
1387 	return value & 0x7F;
1388 }
1389 
1390 static u64 atom_get_val(struct cpudata *cpudata, int pstate)
1391 {
1392 	u64 val;
1393 	int32_t vid_fp;
1394 	u32 vid;
1395 
1396 	val = (u64)pstate << 8;
1397 	if (limits->no_turbo && !limits->turbo_disabled)
1398 		val |= (u64)1 << 32;
1399 
1400 	vid_fp = cpudata->vid.min + mul_fp(
1401 		int_tofp(pstate - cpudata->pstate.min_pstate),
1402 		cpudata->vid.ratio);
1403 
1404 	vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
1405 	vid = ceiling_fp(vid_fp);
1406 
1407 	if (pstate > cpudata->pstate.max_pstate)
1408 		vid = cpudata->vid.turbo;
1409 
1410 	return val | vid;
1411 }
1412 
1413 static int silvermont_get_scaling(void)
1414 {
1415 	u64 value;
1416 	int i;
1417 	/* Defined in Table 35-6 from SDM (Sept 2015) */
1418 	static int silvermont_freq_table[] = {
1419 		83300, 100000, 133300, 116700, 80000};
1420 
1421 	rdmsrl(MSR_FSB_FREQ, value);
1422 	i = value & 0x7;
1423 	WARN_ON(i > 4);
1424 
1425 	return silvermont_freq_table[i];
1426 }
1427 
1428 static int airmont_get_scaling(void)
1429 {
1430 	u64 value;
1431 	int i;
1432 	/* Defined in Table 35-10 from SDM (Sept 2015) */
1433 	static int airmont_freq_table[] = {
1434 		83300, 100000, 133300, 116700, 80000,
1435 		93300, 90000, 88900, 87500};
1436 
1437 	rdmsrl(MSR_FSB_FREQ, value);
1438 	i = value & 0xF;
1439 	WARN_ON(i > 8);
1440 
1441 	return airmont_freq_table[i];
1442 }
1443 
1444 static void atom_get_vid(struct cpudata *cpudata)
1445 {
1446 	u64 value;
1447 
1448 	rdmsrl(ATOM_VIDS, value);
1449 	cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
1450 	cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
1451 	cpudata->vid.ratio = div_fp(
1452 		cpudata->vid.max - cpudata->vid.min,
1453 		int_tofp(cpudata->pstate.max_pstate -
1454 			cpudata->pstate.min_pstate));
1455 
1456 	rdmsrl(ATOM_TURBO_VIDS, value);
1457 	cpudata->vid.turbo = value & 0x7f;
1458 }
1459 
1460 static int core_get_min_pstate(void)
1461 {
1462 	u64 value;
1463 
1464 	rdmsrl(MSR_PLATFORM_INFO, value);
1465 	return (value >> 40) & 0xFF;
1466 }
1467 
1468 static int core_get_max_pstate_physical(void)
1469 {
1470 	u64 value;
1471 
1472 	rdmsrl(MSR_PLATFORM_INFO, value);
1473 	return (value >> 8) & 0xFF;
1474 }
1475 
1476 static int core_get_tdp_ratio(u64 plat_info)
1477 {
1478 	/* Check how many TDP levels present */
1479 	if (plat_info & 0x600000000) {
1480 		u64 tdp_ctrl;
1481 		u64 tdp_ratio;
1482 		int tdp_msr;
1483 		int err;
1484 
1485 		/* Get the TDP level (0, 1, 2) to get ratios */
1486 		err = rdmsrl_safe(MSR_CONFIG_TDP_CONTROL, &tdp_ctrl);
1487 		if (err)
1488 			return err;
1489 
1490 		/* TDP MSR are continuous starting at 0x648 */
1491 		tdp_msr = MSR_CONFIG_TDP_NOMINAL + (tdp_ctrl & 0x03);
1492 		err = rdmsrl_safe(tdp_msr, &tdp_ratio);
1493 		if (err)
1494 			return err;
1495 
1496 		/* For level 1 and 2, bits[23:16] contain the ratio */
1497 		if (tdp_ctrl & 0x03)
1498 			tdp_ratio >>= 16;
1499 
1500 		tdp_ratio &= 0xff; /* ratios are only 8 bits long */
1501 		pr_debug("tdp_ratio %x\n", (int)tdp_ratio);
1502 
1503 		return (int)tdp_ratio;
1504 	}
1505 
1506 	return -ENXIO;
1507 }
1508 
1509 static int core_get_max_pstate(void)
1510 {
1511 	u64 tar;
1512 	u64 plat_info;
1513 	int max_pstate;
1514 	int tdp_ratio;
1515 	int err;
1516 
1517 	rdmsrl(MSR_PLATFORM_INFO, plat_info);
1518 	max_pstate = (plat_info >> 8) & 0xFF;
1519 
1520 	tdp_ratio = core_get_tdp_ratio(plat_info);
1521 	if (tdp_ratio <= 0)
1522 		return max_pstate;
1523 
1524 	if (hwp_active) {
1525 		/* Turbo activation ratio is not used on HWP platforms */
1526 		return tdp_ratio;
1527 	}
1528 
1529 	err = rdmsrl_safe(MSR_TURBO_ACTIVATION_RATIO, &tar);
1530 	if (!err) {
1531 		int tar_levels;
1532 
1533 		/* Do some sanity checking for safety */
1534 		tar_levels = tar & 0xff;
1535 		if (tdp_ratio - 1 == tar_levels) {
1536 			max_pstate = tar_levels;
1537 			pr_debug("max_pstate=TAC %x\n", max_pstate);
1538 		}
1539 	}
1540 
1541 	return max_pstate;
1542 }
1543 
1544 static int core_get_turbo_pstate(void)
1545 {
1546 	u64 value;
1547 	int nont, ret;
1548 
1549 	rdmsrl(MSR_TURBO_RATIO_LIMIT, value);
1550 	nont = core_get_max_pstate();
1551 	ret = (value) & 255;
1552 	if (ret <= nont)
1553 		ret = nont;
1554 	return ret;
1555 }
1556 
1557 static inline int core_get_scaling(void)
1558 {
1559 	return 100000;
1560 }
1561 
1562 static u64 core_get_val(struct cpudata *cpudata, int pstate)
1563 {
1564 	u64 val;
1565 
1566 	val = (u64)pstate << 8;
1567 	if (limits->no_turbo && !limits->turbo_disabled)
1568 		val |= (u64)1 << 32;
1569 
1570 	return val;
1571 }
1572 
1573 static int knl_get_turbo_pstate(void)
1574 {
1575 	u64 value;
1576 	int nont, ret;
1577 
1578 	rdmsrl(MSR_TURBO_RATIO_LIMIT, value);
1579 	nont = core_get_max_pstate();
1580 	ret = (((value) >> 8) & 0xFF);
1581 	if (ret <= nont)
1582 		ret = nont;
1583 	return ret;
1584 }
1585 
1586 static struct cpu_defaults core_params = {
1587 	.pid_policy = {
1588 		.sample_rate_ms = 10,
1589 		.deadband = 0,
1590 		.setpoint = 97,
1591 		.p_gain_pct = 20,
1592 		.d_gain_pct = 0,
1593 		.i_gain_pct = 0,
1594 	},
1595 	.funcs = {
1596 		.get_max = core_get_max_pstate,
1597 		.get_max_physical = core_get_max_pstate_physical,
1598 		.get_min = core_get_min_pstate,
1599 		.get_turbo = core_get_turbo_pstate,
1600 		.get_scaling = core_get_scaling,
1601 		.get_val = core_get_val,
1602 		.get_target_pstate = get_target_pstate_use_performance,
1603 	},
1604 };
1605 
1606 static const struct cpu_defaults silvermont_params = {
1607 	.pid_policy = {
1608 		.sample_rate_ms = 10,
1609 		.deadband = 0,
1610 		.setpoint = 60,
1611 		.p_gain_pct = 14,
1612 		.d_gain_pct = 0,
1613 		.i_gain_pct = 4,
1614 	},
1615 	.funcs = {
1616 		.get_max = atom_get_max_pstate,
1617 		.get_max_physical = atom_get_max_pstate,
1618 		.get_min = atom_get_min_pstate,
1619 		.get_turbo = atom_get_turbo_pstate,
1620 		.get_val = atom_get_val,
1621 		.get_scaling = silvermont_get_scaling,
1622 		.get_vid = atom_get_vid,
1623 		.get_target_pstate = get_target_pstate_use_cpu_load,
1624 	},
1625 };
1626 
1627 static const struct cpu_defaults airmont_params = {
1628 	.pid_policy = {
1629 		.sample_rate_ms = 10,
1630 		.deadband = 0,
1631 		.setpoint = 60,
1632 		.p_gain_pct = 14,
1633 		.d_gain_pct = 0,
1634 		.i_gain_pct = 4,
1635 	},
1636 	.funcs = {
1637 		.get_max = atom_get_max_pstate,
1638 		.get_max_physical = atom_get_max_pstate,
1639 		.get_min = atom_get_min_pstate,
1640 		.get_turbo = atom_get_turbo_pstate,
1641 		.get_val = atom_get_val,
1642 		.get_scaling = airmont_get_scaling,
1643 		.get_vid = atom_get_vid,
1644 		.get_target_pstate = get_target_pstate_use_cpu_load,
1645 	},
1646 };
1647 
1648 static const struct cpu_defaults knl_params = {
1649 	.pid_policy = {
1650 		.sample_rate_ms = 10,
1651 		.deadband = 0,
1652 		.setpoint = 97,
1653 		.p_gain_pct = 20,
1654 		.d_gain_pct = 0,
1655 		.i_gain_pct = 0,
1656 	},
1657 	.funcs = {
1658 		.get_max = core_get_max_pstate,
1659 		.get_max_physical = core_get_max_pstate_physical,
1660 		.get_min = core_get_min_pstate,
1661 		.get_turbo = knl_get_turbo_pstate,
1662 		.get_scaling = core_get_scaling,
1663 		.get_val = core_get_val,
1664 		.get_target_pstate = get_target_pstate_use_performance,
1665 	},
1666 };
1667 
1668 static const struct cpu_defaults bxt_params = {
1669 	.pid_policy = {
1670 		.sample_rate_ms = 10,
1671 		.deadband = 0,
1672 		.setpoint = 60,
1673 		.p_gain_pct = 14,
1674 		.d_gain_pct = 0,
1675 		.i_gain_pct = 4,
1676 	},
1677 	.funcs = {
1678 		.get_max = core_get_max_pstate,
1679 		.get_max_physical = core_get_max_pstate_physical,
1680 		.get_min = core_get_min_pstate,
1681 		.get_turbo = core_get_turbo_pstate,
1682 		.get_scaling = core_get_scaling,
1683 		.get_val = core_get_val,
1684 		.get_target_pstate = get_target_pstate_use_cpu_load,
1685 	},
1686 };
1687 
1688 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
1689 {
1690 	int max_perf = cpu->pstate.turbo_pstate;
1691 	int max_perf_adj;
1692 	int min_perf;
1693 	struct perf_limits *perf_limits = limits;
1694 
1695 	if (limits->no_turbo || limits->turbo_disabled)
1696 		max_perf = cpu->pstate.max_pstate;
1697 
1698 	if (per_cpu_limits)
1699 		perf_limits = cpu->perf_limits;
1700 
1701 	/*
1702 	 * performance can be limited by user through sysfs, by cpufreq
1703 	 * policy, or by cpu specific default values determined through
1704 	 * experimentation.
1705 	 */
1706 	max_perf_adj = fp_ext_toint(max_perf * perf_limits->max_perf);
1707 	*max = clamp_t(int, max_perf_adj,
1708 			cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
1709 
1710 	min_perf = fp_ext_toint(max_perf * perf_limits->min_perf);
1711 	*min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
1712 }
1713 
1714 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
1715 {
1716 	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
1717 	cpu->pstate.current_pstate = pstate;
1718 	/*
1719 	 * Generally, there is no guarantee that this code will always run on
1720 	 * the CPU being updated, so force the register update to run on the
1721 	 * right CPU.
1722 	 */
1723 	wrmsrl_on_cpu(cpu->cpu, MSR_IA32_PERF_CTL,
1724 		      pstate_funcs.get_val(cpu, pstate));
1725 }
1726 
1727 static void intel_pstate_set_min_pstate(struct cpudata *cpu)
1728 {
1729 	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
1730 }
1731 
1732 static void intel_pstate_max_within_limits(struct cpudata *cpu)
1733 {
1734 	int min_pstate, max_pstate;
1735 
1736 	update_turbo_state();
1737 	intel_pstate_get_min_max(cpu, &min_pstate, &max_pstate);
1738 	intel_pstate_set_pstate(cpu, max_pstate);
1739 }
1740 
1741 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
1742 {
1743 	cpu->pstate.min_pstate = pstate_funcs.get_min();
1744 	cpu->pstate.max_pstate = pstate_funcs.get_max();
1745 	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical();
1746 	cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
1747 	cpu->pstate.scaling = pstate_funcs.get_scaling();
1748 	cpu->pstate.max_freq = cpu->pstate.max_pstate * cpu->pstate.scaling;
1749 	cpu->pstate.turbo_freq = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
1750 
1751 	if (pstate_funcs.get_vid)
1752 		pstate_funcs.get_vid(cpu);
1753 
1754 	intel_pstate_set_min_pstate(cpu);
1755 }
1756 
1757 static inline void intel_pstate_calc_avg_perf(struct cpudata *cpu)
1758 {
1759 	struct sample *sample = &cpu->sample;
1760 
1761 	sample->core_avg_perf = div_ext_fp(sample->aperf, sample->mperf);
1762 }
1763 
1764 static inline bool intel_pstate_sample(struct cpudata *cpu, u64 time)
1765 {
1766 	u64 aperf, mperf;
1767 	unsigned long flags;
1768 	u64 tsc;
1769 
1770 	local_irq_save(flags);
1771 	rdmsrl(MSR_IA32_APERF, aperf);
1772 	rdmsrl(MSR_IA32_MPERF, mperf);
1773 	tsc = rdtsc();
1774 	if (cpu->prev_mperf == mperf || cpu->prev_tsc == tsc) {
1775 		local_irq_restore(flags);
1776 		return false;
1777 	}
1778 	local_irq_restore(flags);
1779 
1780 	cpu->last_sample_time = cpu->sample.time;
1781 	cpu->sample.time = time;
1782 	cpu->sample.aperf = aperf;
1783 	cpu->sample.mperf = mperf;
1784 	cpu->sample.tsc =  tsc;
1785 	cpu->sample.aperf -= cpu->prev_aperf;
1786 	cpu->sample.mperf -= cpu->prev_mperf;
1787 	cpu->sample.tsc -= cpu->prev_tsc;
1788 
1789 	cpu->prev_aperf = aperf;
1790 	cpu->prev_mperf = mperf;
1791 	cpu->prev_tsc = tsc;
1792 	/*
1793 	 * First time this function is invoked in a given cycle, all of the
1794 	 * previous sample data fields are equal to zero or stale and they must
1795 	 * be populated with meaningful numbers for things to work, so assume
1796 	 * that sample.time will always be reset before setting the utilization
1797 	 * update hook and make the caller skip the sample then.
1798 	 */
1799 	return !!cpu->last_sample_time;
1800 }
1801 
1802 static inline int32_t get_avg_frequency(struct cpudata *cpu)
1803 {
1804 	return mul_ext_fp(cpu->sample.core_avg_perf,
1805 			  cpu->pstate.max_pstate_physical * cpu->pstate.scaling);
1806 }
1807 
1808 static inline int32_t get_avg_pstate(struct cpudata *cpu)
1809 {
1810 	return mul_ext_fp(cpu->pstate.max_pstate_physical,
1811 			  cpu->sample.core_avg_perf);
1812 }
1813 
1814 static inline int32_t get_target_pstate_use_cpu_load(struct cpudata *cpu)
1815 {
1816 	struct sample *sample = &cpu->sample;
1817 	int32_t busy_frac, boost;
1818 	int target, avg_pstate;
1819 
1820 	busy_frac = div_fp(sample->mperf, sample->tsc);
1821 
1822 	boost = cpu->iowait_boost;
1823 	cpu->iowait_boost >>= 1;
1824 
1825 	if (busy_frac < boost)
1826 		busy_frac = boost;
1827 
1828 	sample->busy_scaled = busy_frac * 100;
1829 
1830 	target = limits->no_turbo || limits->turbo_disabled ?
1831 			cpu->pstate.max_pstate : cpu->pstate.turbo_pstate;
1832 	target += target >> 2;
1833 	target = mul_fp(target, busy_frac);
1834 	if (target < cpu->pstate.min_pstate)
1835 		target = cpu->pstate.min_pstate;
1836 
1837 	/*
1838 	 * If the average P-state during the previous cycle was higher than the
1839 	 * current target, add 50% of the difference to the target to reduce
1840 	 * possible performance oscillations and offset possible performance
1841 	 * loss related to moving the workload from one CPU to another within
1842 	 * a package/module.
1843 	 */
1844 	avg_pstate = get_avg_pstate(cpu);
1845 	if (avg_pstate > target)
1846 		target += (avg_pstate - target) >> 1;
1847 
1848 	return target;
1849 }
1850 
1851 static inline int32_t get_target_pstate_use_performance(struct cpudata *cpu)
1852 {
1853 	int32_t perf_scaled, max_pstate, current_pstate, sample_ratio;
1854 	u64 duration_ns;
1855 
1856 	/*
1857 	 * perf_scaled is the ratio of the average P-state during the last
1858 	 * sampling period to the P-state requested last time (in percent).
1859 	 *
1860 	 * That measures the system's response to the previous P-state
1861 	 * selection.
1862 	 */
1863 	max_pstate = cpu->pstate.max_pstate_physical;
1864 	current_pstate = cpu->pstate.current_pstate;
1865 	perf_scaled = mul_ext_fp(cpu->sample.core_avg_perf,
1866 			       div_fp(100 * max_pstate, current_pstate));
1867 
1868 	/*
1869 	 * Since our utilization update callback will not run unless we are
1870 	 * in C0, check if the actual elapsed time is significantly greater (3x)
1871 	 * than our sample interval.  If it is, then we were idle for a long
1872 	 * enough period of time to adjust our performance metric.
1873 	 */
1874 	duration_ns = cpu->sample.time - cpu->last_sample_time;
1875 	if ((s64)duration_ns > pid_params.sample_rate_ns * 3) {
1876 		sample_ratio = div_fp(pid_params.sample_rate_ns, duration_ns);
1877 		perf_scaled = mul_fp(perf_scaled, sample_ratio);
1878 	} else {
1879 		sample_ratio = div_fp(100 * cpu->sample.mperf, cpu->sample.tsc);
1880 		if (sample_ratio < int_tofp(1))
1881 			perf_scaled = 0;
1882 	}
1883 
1884 	cpu->sample.busy_scaled = perf_scaled;
1885 	return cpu->pstate.current_pstate - pid_calc(&cpu->pid, perf_scaled);
1886 }
1887 
1888 static int intel_pstate_prepare_request(struct cpudata *cpu, int pstate)
1889 {
1890 	int max_perf, min_perf;
1891 
1892 	intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
1893 	pstate = clamp_t(int, pstate, min_perf, max_perf);
1894 	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
1895 	return pstate;
1896 }
1897 
1898 static void intel_pstate_update_pstate(struct cpudata *cpu, int pstate)
1899 {
1900 	pstate = intel_pstate_prepare_request(cpu, pstate);
1901 	if (pstate == cpu->pstate.current_pstate)
1902 		return;
1903 
1904 	cpu->pstate.current_pstate = pstate;
1905 	wrmsrl(MSR_IA32_PERF_CTL, pstate_funcs.get_val(cpu, pstate));
1906 }
1907 
1908 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
1909 {
1910 	int from, target_pstate;
1911 	struct sample *sample;
1912 
1913 	from = cpu->pstate.current_pstate;
1914 
1915 	target_pstate = cpu->policy == CPUFREQ_POLICY_PERFORMANCE ?
1916 		cpu->pstate.turbo_pstate : pstate_funcs.get_target_pstate(cpu);
1917 
1918 	update_turbo_state();
1919 
1920 	intel_pstate_update_pstate(cpu, target_pstate);
1921 
1922 	sample = &cpu->sample;
1923 	trace_pstate_sample(mul_ext_fp(100, sample->core_avg_perf),
1924 		fp_toint(sample->busy_scaled),
1925 		from,
1926 		cpu->pstate.current_pstate,
1927 		sample->mperf,
1928 		sample->aperf,
1929 		sample->tsc,
1930 		get_avg_frequency(cpu),
1931 		fp_toint(cpu->iowait_boost * 100));
1932 }
1933 
1934 static void intel_pstate_update_util(struct update_util_data *data, u64 time,
1935 				     unsigned int flags)
1936 {
1937 	struct cpudata *cpu = container_of(data, struct cpudata, update_util);
1938 	u64 delta_ns;
1939 
1940 	if (pstate_funcs.get_target_pstate == get_target_pstate_use_cpu_load) {
1941 		if (flags & SCHED_CPUFREQ_IOWAIT) {
1942 			cpu->iowait_boost = int_tofp(1);
1943 		} else if (cpu->iowait_boost) {
1944 			/* Clear iowait_boost if the CPU may have been idle. */
1945 			delta_ns = time - cpu->last_update;
1946 			if (delta_ns > TICK_NSEC)
1947 				cpu->iowait_boost = 0;
1948 		}
1949 		cpu->last_update = time;
1950 	}
1951 
1952 	delta_ns = time - cpu->sample.time;
1953 	if ((s64)delta_ns >= pid_params.sample_rate_ns) {
1954 		bool sample_taken = intel_pstate_sample(cpu, time);
1955 
1956 		if (sample_taken) {
1957 			intel_pstate_calc_avg_perf(cpu);
1958 			if (!hwp_active)
1959 				intel_pstate_adjust_busy_pstate(cpu);
1960 		}
1961 	}
1962 }
1963 
1964 #define ICPU(model, policy) \
1965 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
1966 			(unsigned long)&policy }
1967 
1968 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
1969 	ICPU(INTEL_FAM6_SANDYBRIDGE, 		core_params),
1970 	ICPU(INTEL_FAM6_SANDYBRIDGE_X,		core_params),
1971 	ICPU(INTEL_FAM6_ATOM_SILVERMONT1,	silvermont_params),
1972 	ICPU(INTEL_FAM6_IVYBRIDGE,		core_params),
1973 	ICPU(INTEL_FAM6_HASWELL_CORE,		core_params),
1974 	ICPU(INTEL_FAM6_BROADWELL_CORE,		core_params),
1975 	ICPU(INTEL_FAM6_IVYBRIDGE_X,		core_params),
1976 	ICPU(INTEL_FAM6_HASWELL_X,		core_params),
1977 	ICPU(INTEL_FAM6_HASWELL_ULT,		core_params),
1978 	ICPU(INTEL_FAM6_HASWELL_GT3E,		core_params),
1979 	ICPU(INTEL_FAM6_BROADWELL_GT3E,		core_params),
1980 	ICPU(INTEL_FAM6_ATOM_AIRMONT,		airmont_params),
1981 	ICPU(INTEL_FAM6_SKYLAKE_MOBILE,		core_params),
1982 	ICPU(INTEL_FAM6_BROADWELL_X,		core_params),
1983 	ICPU(INTEL_FAM6_SKYLAKE_DESKTOP,	core_params),
1984 	ICPU(INTEL_FAM6_BROADWELL_XEON_D,	core_params),
1985 	ICPU(INTEL_FAM6_XEON_PHI_KNL,		knl_params),
1986 	ICPU(INTEL_FAM6_XEON_PHI_KNM,		knl_params),
1987 	ICPU(INTEL_FAM6_ATOM_GOLDMONT,		bxt_params),
1988 	{}
1989 };
1990 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
1991 
1992 static const struct x86_cpu_id intel_pstate_cpu_oob_ids[] __initconst = {
1993 	ICPU(INTEL_FAM6_BROADWELL_XEON_D, core_params),
1994 	ICPU(INTEL_FAM6_BROADWELL_X, core_params),
1995 	ICPU(INTEL_FAM6_SKYLAKE_X, core_params),
1996 	{}
1997 };
1998 
1999 static const struct x86_cpu_id intel_pstate_cpu_ee_disable_ids[] = {
2000 	ICPU(INTEL_FAM6_KABYLAKE_DESKTOP, core_params),
2001 	{}
2002 };
2003 
2004 static int intel_pstate_init_cpu(unsigned int cpunum)
2005 {
2006 	struct cpudata *cpu;
2007 
2008 	cpu = all_cpu_data[cpunum];
2009 
2010 	if (!cpu) {
2011 		unsigned int size = sizeof(struct cpudata);
2012 
2013 		if (per_cpu_limits)
2014 			size += sizeof(struct perf_limits);
2015 
2016 		cpu = kzalloc(size, GFP_KERNEL);
2017 		if (!cpu)
2018 			return -ENOMEM;
2019 
2020 		all_cpu_data[cpunum] = cpu;
2021 		if (per_cpu_limits)
2022 			cpu->perf_limits = (struct perf_limits *)(cpu + 1);
2023 
2024 		cpu->epp_default = -EINVAL;
2025 		cpu->epp_powersave = -EINVAL;
2026 		cpu->epp_saved = -EINVAL;
2027 	}
2028 
2029 	cpu = all_cpu_data[cpunum];
2030 
2031 	cpu->cpu = cpunum;
2032 
2033 	if (hwp_active) {
2034 		const struct x86_cpu_id *id;
2035 
2036 		id = x86_match_cpu(intel_pstate_cpu_ee_disable_ids);
2037 		if (id)
2038 			intel_pstate_disable_ee(cpunum);
2039 
2040 		intel_pstate_hwp_enable(cpu);
2041 		pid_params.sample_rate_ms = 50;
2042 		pid_params.sample_rate_ns = 50 * NSEC_PER_MSEC;
2043 	}
2044 
2045 	intel_pstate_get_cpu_pstates(cpu);
2046 
2047 	intel_pstate_busy_pid_reset(cpu);
2048 
2049 	pr_debug("controlling: cpu %d\n", cpunum);
2050 
2051 	return 0;
2052 }
2053 
2054 static unsigned int intel_pstate_get(unsigned int cpu_num)
2055 {
2056 	struct cpudata *cpu = all_cpu_data[cpu_num];
2057 
2058 	return cpu ? get_avg_frequency(cpu) : 0;
2059 }
2060 
2061 static void intel_pstate_set_update_util_hook(unsigned int cpu_num)
2062 {
2063 	struct cpudata *cpu = all_cpu_data[cpu_num];
2064 
2065 	if (cpu->update_util_set)
2066 		return;
2067 
2068 	/* Prevent intel_pstate_update_util() from using stale data. */
2069 	cpu->sample.time = 0;
2070 	cpufreq_add_update_util_hook(cpu_num, &cpu->update_util,
2071 				     intel_pstate_update_util);
2072 	cpu->update_util_set = true;
2073 }
2074 
2075 static void intel_pstate_clear_update_util_hook(unsigned int cpu)
2076 {
2077 	struct cpudata *cpu_data = all_cpu_data[cpu];
2078 
2079 	if (!cpu_data->update_util_set)
2080 		return;
2081 
2082 	cpufreq_remove_update_util_hook(cpu);
2083 	cpu_data->update_util_set = false;
2084 	synchronize_sched();
2085 }
2086 
2087 static void intel_pstate_set_performance_limits(struct perf_limits *limits)
2088 {
2089 	limits->no_turbo = 0;
2090 	limits->turbo_disabled = 0;
2091 	limits->max_perf_pct = 100;
2092 	limits->max_perf = int_ext_tofp(1);
2093 	limits->min_perf_pct = 100;
2094 	limits->min_perf = int_ext_tofp(1);
2095 	limits->max_policy_pct = 100;
2096 	limits->max_sysfs_pct = 100;
2097 	limits->min_policy_pct = 0;
2098 	limits->min_sysfs_pct = 0;
2099 }
2100 
2101 static void intel_pstate_update_perf_limits(struct cpufreq_policy *policy,
2102 					    struct perf_limits *limits)
2103 {
2104 
2105 	limits->max_policy_pct = DIV_ROUND_UP(policy->max * 100,
2106 					      policy->cpuinfo.max_freq);
2107 	limits->max_policy_pct = clamp_t(int, limits->max_policy_pct, 0, 100);
2108 	if (policy->max == policy->min) {
2109 		limits->min_policy_pct = limits->max_policy_pct;
2110 	} else {
2111 		limits->min_policy_pct = DIV_ROUND_UP(policy->min * 100,
2112 						      policy->cpuinfo.max_freq);
2113 		limits->min_policy_pct = clamp_t(int, limits->min_policy_pct,
2114 						 0, 100);
2115 	}
2116 
2117 	/* Normalize user input to [min_policy_pct, max_policy_pct] */
2118 	limits->min_perf_pct = max(limits->min_policy_pct,
2119 				   limits->min_sysfs_pct);
2120 	limits->min_perf_pct = min(limits->max_policy_pct,
2121 				   limits->min_perf_pct);
2122 	limits->max_perf_pct = min(limits->max_policy_pct,
2123 				   limits->max_sysfs_pct);
2124 	limits->max_perf_pct = max(limits->min_policy_pct,
2125 				   limits->max_perf_pct);
2126 
2127 	/* Make sure min_perf_pct <= max_perf_pct */
2128 	limits->min_perf_pct = min(limits->max_perf_pct, limits->min_perf_pct);
2129 
2130 	limits->min_perf = div_ext_fp(limits->min_perf_pct, 100);
2131 	limits->max_perf = div_ext_fp(limits->max_perf_pct, 100);
2132 	limits->max_perf = round_up(limits->max_perf, EXT_FRAC_BITS);
2133 	limits->min_perf = round_up(limits->min_perf, EXT_FRAC_BITS);
2134 
2135 	pr_debug("cpu:%d max_perf_pct:%d min_perf_pct:%d\n", policy->cpu,
2136 		 limits->max_perf_pct, limits->min_perf_pct);
2137 }
2138 
2139 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
2140 {
2141 	struct cpudata *cpu;
2142 	struct perf_limits *perf_limits = NULL;
2143 
2144 	if (!policy->cpuinfo.max_freq)
2145 		return -ENODEV;
2146 
2147 	pr_debug("set_policy cpuinfo.max %u policy->max %u\n",
2148 		 policy->cpuinfo.max_freq, policy->max);
2149 
2150 	cpu = all_cpu_data[policy->cpu];
2151 	cpu->policy = policy->policy;
2152 
2153 	if (cpu->pstate.max_pstate_physical > cpu->pstate.max_pstate &&
2154 	    policy->max < policy->cpuinfo.max_freq &&
2155 	    policy->max > cpu->pstate.max_pstate * cpu->pstate.scaling) {
2156 		pr_debug("policy->max > max non turbo frequency\n");
2157 		policy->max = policy->cpuinfo.max_freq;
2158 	}
2159 
2160 	if (per_cpu_limits)
2161 		perf_limits = cpu->perf_limits;
2162 
2163 	mutex_lock(&intel_pstate_limits_lock);
2164 
2165 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
2166 		if (!perf_limits) {
2167 			limits = &performance_limits;
2168 			perf_limits = limits;
2169 		}
2170 		if (policy->max >= policy->cpuinfo.max_freq &&
2171 		    !limits->no_turbo) {
2172 			pr_debug("set performance\n");
2173 			intel_pstate_set_performance_limits(perf_limits);
2174 			goto out;
2175 		}
2176 	} else {
2177 		pr_debug("set powersave\n");
2178 		if (!perf_limits) {
2179 			limits = &powersave_limits;
2180 			perf_limits = limits;
2181 		}
2182 
2183 	}
2184 
2185 	intel_pstate_update_perf_limits(policy, perf_limits);
2186  out:
2187 	if (cpu->policy == CPUFREQ_POLICY_PERFORMANCE) {
2188 		/*
2189 		 * NOHZ_FULL CPUs need this as the governor callback may not
2190 		 * be invoked on them.
2191 		 */
2192 		intel_pstate_clear_update_util_hook(policy->cpu);
2193 		intel_pstate_max_within_limits(cpu);
2194 	}
2195 
2196 	intel_pstate_set_update_util_hook(policy->cpu);
2197 
2198 	intel_pstate_hwp_set_policy(policy);
2199 
2200 	mutex_unlock(&intel_pstate_limits_lock);
2201 
2202 	return 0;
2203 }
2204 
2205 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
2206 {
2207 	struct cpudata *cpu = all_cpu_data[policy->cpu];
2208 	struct perf_limits *perf_limits;
2209 
2210 	if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
2211 		perf_limits = &performance_limits;
2212 	else
2213 		perf_limits = &powersave_limits;
2214 
2215 	update_turbo_state();
2216 	policy->cpuinfo.max_freq = perf_limits->turbo_disabled ||
2217 					perf_limits->no_turbo ?
2218 					cpu->pstate.max_freq :
2219 					cpu->pstate.turbo_freq;
2220 
2221 	cpufreq_verify_within_cpu_limits(policy);
2222 
2223 	if (policy->policy != CPUFREQ_POLICY_POWERSAVE &&
2224 	    policy->policy != CPUFREQ_POLICY_PERFORMANCE)
2225 		return -EINVAL;
2226 
2227 	/* When per-CPU limits are used, sysfs limits are not used */
2228 	if (!per_cpu_limits) {
2229 		unsigned int max_freq, min_freq;
2230 
2231 		max_freq = policy->cpuinfo.max_freq *
2232 						limits->max_sysfs_pct / 100;
2233 		min_freq = policy->cpuinfo.max_freq *
2234 						limits->min_sysfs_pct / 100;
2235 		cpufreq_verify_within_limits(policy, min_freq, max_freq);
2236 	}
2237 
2238 	return 0;
2239 }
2240 
2241 static void intel_cpufreq_stop_cpu(struct cpufreq_policy *policy)
2242 {
2243 	intel_pstate_set_min_pstate(all_cpu_data[policy->cpu]);
2244 }
2245 
2246 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
2247 {
2248 	pr_debug("CPU %d exiting\n", policy->cpu);
2249 
2250 	intel_pstate_clear_update_util_hook(policy->cpu);
2251 	if (hwp_active)
2252 		intel_pstate_hwp_save_state(policy);
2253 	else
2254 		intel_cpufreq_stop_cpu(policy);
2255 }
2256 
2257 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
2258 {
2259 	intel_pstate_exit_perf_limits(policy);
2260 
2261 	policy->fast_switch_possible = false;
2262 
2263 	return 0;
2264 }
2265 
2266 static int __intel_pstate_cpu_init(struct cpufreq_policy *policy)
2267 {
2268 	struct cpudata *cpu;
2269 	int rc;
2270 
2271 	rc = intel_pstate_init_cpu(policy->cpu);
2272 	if (rc)
2273 		return rc;
2274 
2275 	cpu = all_cpu_data[policy->cpu];
2276 
2277 	/*
2278 	 * We need sane value in the cpu->perf_limits, so inherit from global
2279 	 * perf_limits limits, which are seeded with values based on the
2280 	 * CONFIG_CPU_FREQ_DEFAULT_GOV_*, during boot up.
2281 	 */
2282 	if (per_cpu_limits)
2283 		memcpy(cpu->perf_limits, limits, sizeof(struct perf_limits));
2284 
2285 	policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
2286 	policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
2287 
2288 	/* cpuinfo and default policy values */
2289 	policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
2290 	update_turbo_state();
2291 	policy->cpuinfo.max_freq = limits->turbo_disabled ?
2292 			cpu->pstate.max_pstate : cpu->pstate.turbo_pstate;
2293 	policy->cpuinfo.max_freq *= cpu->pstate.scaling;
2294 
2295 	intel_pstate_init_acpi_perf_limits(policy);
2296 	cpumask_set_cpu(policy->cpu, policy->cpus);
2297 
2298 	policy->fast_switch_possible = true;
2299 
2300 	return 0;
2301 }
2302 
2303 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
2304 {
2305 	int ret = __intel_pstate_cpu_init(policy);
2306 
2307 	if (ret)
2308 		return ret;
2309 
2310 	policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
2311 	if (limits->min_perf_pct == 100 && limits->max_perf_pct == 100)
2312 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
2313 	else
2314 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
2315 
2316 	return 0;
2317 }
2318 
2319 static struct cpufreq_driver intel_pstate = {
2320 	.flags		= CPUFREQ_CONST_LOOPS,
2321 	.verify		= intel_pstate_verify_policy,
2322 	.setpolicy	= intel_pstate_set_policy,
2323 	.suspend	= intel_pstate_hwp_save_state,
2324 	.resume		= intel_pstate_resume,
2325 	.get		= intel_pstate_get,
2326 	.init		= intel_pstate_cpu_init,
2327 	.exit		= intel_pstate_cpu_exit,
2328 	.stop_cpu	= intel_pstate_stop_cpu,
2329 	.name		= "intel_pstate",
2330 };
2331 
2332 static int intel_cpufreq_verify_policy(struct cpufreq_policy *policy)
2333 {
2334 	struct cpudata *cpu = all_cpu_data[policy->cpu];
2335 	struct perf_limits *perf_limits = limits;
2336 
2337 	update_turbo_state();
2338 	policy->cpuinfo.max_freq = limits->turbo_disabled ?
2339 			cpu->pstate.max_freq : cpu->pstate.turbo_freq;
2340 
2341 	cpufreq_verify_within_cpu_limits(policy);
2342 
2343 	if (per_cpu_limits)
2344 		perf_limits = cpu->perf_limits;
2345 
2346 	mutex_lock(&intel_pstate_limits_lock);
2347 
2348 	intel_pstate_update_perf_limits(policy, perf_limits);
2349 
2350 	mutex_unlock(&intel_pstate_limits_lock);
2351 
2352 	return 0;
2353 }
2354 
2355 static unsigned int intel_cpufreq_turbo_update(struct cpudata *cpu,
2356 					       struct cpufreq_policy *policy,
2357 					       unsigned int target_freq)
2358 {
2359 	unsigned int max_freq;
2360 
2361 	update_turbo_state();
2362 
2363 	max_freq = limits->no_turbo || limits->turbo_disabled ?
2364 			cpu->pstate.max_freq : cpu->pstate.turbo_freq;
2365 	policy->cpuinfo.max_freq = max_freq;
2366 	if (policy->max > max_freq)
2367 		policy->max = max_freq;
2368 
2369 	if (target_freq > max_freq)
2370 		target_freq = max_freq;
2371 
2372 	return target_freq;
2373 }
2374 
2375 static int intel_cpufreq_target(struct cpufreq_policy *policy,
2376 				unsigned int target_freq,
2377 				unsigned int relation)
2378 {
2379 	struct cpudata *cpu = all_cpu_data[policy->cpu];
2380 	struct cpufreq_freqs freqs;
2381 	int target_pstate;
2382 
2383 	freqs.old = policy->cur;
2384 	freqs.new = intel_cpufreq_turbo_update(cpu, policy, target_freq);
2385 
2386 	cpufreq_freq_transition_begin(policy, &freqs);
2387 	switch (relation) {
2388 	case CPUFREQ_RELATION_L:
2389 		target_pstate = DIV_ROUND_UP(freqs.new, cpu->pstate.scaling);
2390 		break;
2391 	case CPUFREQ_RELATION_H:
2392 		target_pstate = freqs.new / cpu->pstate.scaling;
2393 		break;
2394 	default:
2395 		target_pstate = DIV_ROUND_CLOSEST(freqs.new, cpu->pstate.scaling);
2396 		break;
2397 	}
2398 	target_pstate = intel_pstate_prepare_request(cpu, target_pstate);
2399 	if (target_pstate != cpu->pstate.current_pstate) {
2400 		cpu->pstate.current_pstate = target_pstate;
2401 		wrmsrl_on_cpu(policy->cpu, MSR_IA32_PERF_CTL,
2402 			      pstate_funcs.get_val(cpu, target_pstate));
2403 	}
2404 	cpufreq_freq_transition_end(policy, &freqs, false);
2405 
2406 	return 0;
2407 }
2408 
2409 static unsigned int intel_cpufreq_fast_switch(struct cpufreq_policy *policy,
2410 					      unsigned int target_freq)
2411 {
2412 	struct cpudata *cpu = all_cpu_data[policy->cpu];
2413 	int target_pstate;
2414 
2415 	target_freq = intel_cpufreq_turbo_update(cpu, policy, target_freq);
2416 	target_pstate = DIV_ROUND_UP(target_freq, cpu->pstate.scaling);
2417 	intel_pstate_update_pstate(cpu, target_pstate);
2418 	return target_freq;
2419 }
2420 
2421 static int intel_cpufreq_cpu_init(struct cpufreq_policy *policy)
2422 {
2423 	int ret = __intel_pstate_cpu_init(policy);
2424 
2425 	if (ret)
2426 		return ret;
2427 
2428 	policy->cpuinfo.transition_latency = INTEL_CPUFREQ_TRANSITION_LATENCY;
2429 	/* This reflects the intel_pstate_get_cpu_pstates() setting. */
2430 	policy->cur = policy->cpuinfo.min_freq;
2431 
2432 	return 0;
2433 }
2434 
2435 static struct cpufreq_driver intel_cpufreq = {
2436 	.flags		= CPUFREQ_CONST_LOOPS,
2437 	.verify		= intel_cpufreq_verify_policy,
2438 	.target		= intel_cpufreq_target,
2439 	.fast_switch	= intel_cpufreq_fast_switch,
2440 	.init		= intel_cpufreq_cpu_init,
2441 	.exit		= intel_pstate_cpu_exit,
2442 	.stop_cpu	= intel_cpufreq_stop_cpu,
2443 	.name		= "intel_cpufreq",
2444 };
2445 
2446 static struct cpufreq_driver *intel_pstate_driver = &intel_pstate;
2447 
2448 static void intel_pstate_driver_cleanup(void)
2449 {
2450 	unsigned int cpu;
2451 
2452 	get_online_cpus();
2453 	for_each_online_cpu(cpu) {
2454 		if (all_cpu_data[cpu]) {
2455 			if (intel_pstate_driver == &intel_pstate)
2456 				intel_pstate_clear_update_util_hook(cpu);
2457 
2458 			kfree(all_cpu_data[cpu]);
2459 			all_cpu_data[cpu] = NULL;
2460 		}
2461 	}
2462 	put_online_cpus();
2463 }
2464 
2465 static int intel_pstate_register_driver(void)
2466 {
2467 	int ret;
2468 
2469 	ret = cpufreq_register_driver(intel_pstate_driver);
2470 	if (ret) {
2471 		intel_pstate_driver_cleanup();
2472 		return ret;
2473 	}
2474 
2475 	mutex_lock(&intel_pstate_limits_lock);
2476 	driver_registered = true;
2477 	mutex_unlock(&intel_pstate_limits_lock);
2478 
2479 	if (intel_pstate_driver == &intel_pstate && !hwp_active &&
2480 	    pstate_funcs.get_target_pstate != get_target_pstate_use_cpu_load)
2481 		intel_pstate_debug_expose_params();
2482 
2483 	return 0;
2484 }
2485 
2486 static int intel_pstate_unregister_driver(void)
2487 {
2488 	if (hwp_active)
2489 		return -EBUSY;
2490 
2491 	if (intel_pstate_driver == &intel_pstate && !hwp_active &&
2492 	    pstate_funcs.get_target_pstate != get_target_pstate_use_cpu_load)
2493 		intel_pstate_debug_hide_params();
2494 
2495 	mutex_lock(&intel_pstate_limits_lock);
2496 	driver_registered = false;
2497 	mutex_unlock(&intel_pstate_limits_lock);
2498 
2499 	cpufreq_unregister_driver(intel_pstate_driver);
2500 	intel_pstate_driver_cleanup();
2501 
2502 	return 0;
2503 }
2504 
2505 static ssize_t intel_pstate_show_status(char *buf)
2506 {
2507 	if (!driver_registered)
2508 		return sprintf(buf, "off\n");
2509 
2510 	return sprintf(buf, "%s\n", intel_pstate_driver == &intel_pstate ?
2511 					"active" : "passive");
2512 }
2513 
2514 static int intel_pstate_update_status(const char *buf, size_t size)
2515 {
2516 	int ret;
2517 
2518 	if (size == 3 && !strncmp(buf, "off", size))
2519 		return driver_registered ?
2520 			intel_pstate_unregister_driver() : -EINVAL;
2521 
2522 	if (size == 6 && !strncmp(buf, "active", size)) {
2523 		if (driver_registered) {
2524 			if (intel_pstate_driver == &intel_pstate)
2525 				return 0;
2526 
2527 			ret = intel_pstate_unregister_driver();
2528 			if (ret)
2529 				return ret;
2530 		}
2531 
2532 		intel_pstate_driver = &intel_pstate;
2533 		return intel_pstate_register_driver();
2534 	}
2535 
2536 	if (size == 7 && !strncmp(buf, "passive", size)) {
2537 		if (driver_registered) {
2538 			if (intel_pstate_driver != &intel_pstate)
2539 				return 0;
2540 
2541 			ret = intel_pstate_unregister_driver();
2542 			if (ret)
2543 				return ret;
2544 		}
2545 
2546 		intel_pstate_driver = &intel_cpufreq;
2547 		return intel_pstate_register_driver();
2548 	}
2549 
2550 	return -EINVAL;
2551 }
2552 
2553 static int no_load __initdata;
2554 static int no_hwp __initdata;
2555 static int hwp_only __initdata;
2556 static unsigned int force_load __initdata;
2557 
2558 static int __init intel_pstate_msrs_not_valid(void)
2559 {
2560 	if (!pstate_funcs.get_max() ||
2561 	    !pstate_funcs.get_min() ||
2562 	    !pstate_funcs.get_turbo())
2563 		return -ENODEV;
2564 
2565 	return 0;
2566 }
2567 
2568 static void __init copy_pid_params(struct pstate_adjust_policy *policy)
2569 {
2570 	pid_params.sample_rate_ms = policy->sample_rate_ms;
2571 	pid_params.sample_rate_ns = pid_params.sample_rate_ms * NSEC_PER_MSEC;
2572 	pid_params.p_gain_pct = policy->p_gain_pct;
2573 	pid_params.i_gain_pct = policy->i_gain_pct;
2574 	pid_params.d_gain_pct = policy->d_gain_pct;
2575 	pid_params.deadband = policy->deadband;
2576 	pid_params.setpoint = policy->setpoint;
2577 }
2578 
2579 #ifdef CONFIG_ACPI
2580 static void intel_pstate_use_acpi_profile(void)
2581 {
2582 	if (acpi_gbl_FADT.preferred_profile == PM_MOBILE)
2583 		pstate_funcs.get_target_pstate =
2584 				get_target_pstate_use_cpu_load;
2585 }
2586 #else
2587 static void intel_pstate_use_acpi_profile(void)
2588 {
2589 }
2590 #endif
2591 
2592 static void __init copy_cpu_funcs(struct pstate_funcs *funcs)
2593 {
2594 	pstate_funcs.get_max   = funcs->get_max;
2595 	pstate_funcs.get_max_physical = funcs->get_max_physical;
2596 	pstate_funcs.get_min   = funcs->get_min;
2597 	pstate_funcs.get_turbo = funcs->get_turbo;
2598 	pstate_funcs.get_scaling = funcs->get_scaling;
2599 	pstate_funcs.get_val   = funcs->get_val;
2600 	pstate_funcs.get_vid   = funcs->get_vid;
2601 	pstate_funcs.get_target_pstate = funcs->get_target_pstate;
2602 
2603 	intel_pstate_use_acpi_profile();
2604 }
2605 
2606 #ifdef CONFIG_ACPI
2607 
2608 static bool __init intel_pstate_no_acpi_pss(void)
2609 {
2610 	int i;
2611 
2612 	for_each_possible_cpu(i) {
2613 		acpi_status status;
2614 		union acpi_object *pss;
2615 		struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
2616 		struct acpi_processor *pr = per_cpu(processors, i);
2617 
2618 		if (!pr)
2619 			continue;
2620 
2621 		status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
2622 		if (ACPI_FAILURE(status))
2623 			continue;
2624 
2625 		pss = buffer.pointer;
2626 		if (pss && pss->type == ACPI_TYPE_PACKAGE) {
2627 			kfree(pss);
2628 			return false;
2629 		}
2630 
2631 		kfree(pss);
2632 	}
2633 
2634 	return true;
2635 }
2636 
2637 static bool __init intel_pstate_has_acpi_ppc(void)
2638 {
2639 	int i;
2640 
2641 	for_each_possible_cpu(i) {
2642 		struct acpi_processor *pr = per_cpu(processors, i);
2643 
2644 		if (!pr)
2645 			continue;
2646 		if (acpi_has_method(pr->handle, "_PPC"))
2647 			return true;
2648 	}
2649 	return false;
2650 }
2651 
2652 enum {
2653 	PSS,
2654 	PPC,
2655 };
2656 
2657 struct hw_vendor_info {
2658 	u16  valid;
2659 	char oem_id[ACPI_OEM_ID_SIZE];
2660 	char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
2661 	int  oem_pwr_table;
2662 };
2663 
2664 /* Hardware vendor-specific info that has its own power management modes */
2665 static struct hw_vendor_info vendor_info[] __initdata = {
2666 	{1, "HP    ", "ProLiant", PSS},
2667 	{1, "ORACLE", "X4-2    ", PPC},
2668 	{1, "ORACLE", "X4-2L   ", PPC},
2669 	{1, "ORACLE", "X4-2B   ", PPC},
2670 	{1, "ORACLE", "X3-2    ", PPC},
2671 	{1, "ORACLE", "X3-2L   ", PPC},
2672 	{1, "ORACLE", "X3-2B   ", PPC},
2673 	{1, "ORACLE", "X4470M2 ", PPC},
2674 	{1, "ORACLE", "X4270M3 ", PPC},
2675 	{1, "ORACLE", "X4270M2 ", PPC},
2676 	{1, "ORACLE", "X4170M2 ", PPC},
2677 	{1, "ORACLE", "X4170 M3", PPC},
2678 	{1, "ORACLE", "X4275 M3", PPC},
2679 	{1, "ORACLE", "X6-2    ", PPC},
2680 	{1, "ORACLE", "Sudbury ", PPC},
2681 	{0, "", ""},
2682 };
2683 
2684 static bool __init intel_pstate_platform_pwr_mgmt_exists(void)
2685 {
2686 	struct acpi_table_header hdr;
2687 	struct hw_vendor_info *v_info;
2688 	const struct x86_cpu_id *id;
2689 	u64 misc_pwr;
2690 
2691 	id = x86_match_cpu(intel_pstate_cpu_oob_ids);
2692 	if (id) {
2693 		rdmsrl(MSR_MISC_PWR_MGMT, misc_pwr);
2694 		if ( misc_pwr & (1 << 8))
2695 			return true;
2696 	}
2697 
2698 	if (acpi_disabled ||
2699 	    ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
2700 		return false;
2701 
2702 	for (v_info = vendor_info; v_info->valid; v_info++) {
2703 		if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) &&
2704 			!strncmp(hdr.oem_table_id, v_info->oem_table_id,
2705 						ACPI_OEM_TABLE_ID_SIZE))
2706 			switch (v_info->oem_pwr_table) {
2707 			case PSS:
2708 				return intel_pstate_no_acpi_pss();
2709 			case PPC:
2710 				return intel_pstate_has_acpi_ppc() &&
2711 					(!force_load);
2712 			}
2713 	}
2714 
2715 	return false;
2716 }
2717 
2718 static void intel_pstate_request_control_from_smm(void)
2719 {
2720 	/*
2721 	 * It may be unsafe to request P-states control from SMM if _PPC support
2722 	 * has not been enabled.
2723 	 */
2724 	if (acpi_ppc)
2725 		acpi_processor_pstate_control();
2726 }
2727 #else /* CONFIG_ACPI not enabled */
2728 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
2729 static inline bool intel_pstate_has_acpi_ppc(void) { return false; }
2730 static inline void intel_pstate_request_control_from_smm(void) {}
2731 #endif /* CONFIG_ACPI */
2732 
2733 static const struct x86_cpu_id hwp_support_ids[] __initconst = {
2734 	{ X86_VENDOR_INTEL, 6, X86_MODEL_ANY, X86_FEATURE_HWP },
2735 	{}
2736 };
2737 
2738 static int __init intel_pstate_init(void)
2739 {
2740 	const struct x86_cpu_id *id;
2741 	struct cpu_defaults *cpu_def;
2742 	int rc = 0;
2743 
2744 	if (no_load)
2745 		return -ENODEV;
2746 
2747 	if (x86_match_cpu(hwp_support_ids) && !no_hwp) {
2748 		copy_cpu_funcs(&core_params.funcs);
2749 		hwp_active++;
2750 		intel_pstate.attr = hwp_cpufreq_attrs;
2751 		goto hwp_cpu_matched;
2752 	}
2753 
2754 	id = x86_match_cpu(intel_pstate_cpu_ids);
2755 	if (!id)
2756 		return -ENODEV;
2757 
2758 	cpu_def = (struct cpu_defaults *)id->driver_data;
2759 
2760 	copy_pid_params(&cpu_def->pid_policy);
2761 	copy_cpu_funcs(&cpu_def->funcs);
2762 
2763 	if (intel_pstate_msrs_not_valid())
2764 		return -ENODEV;
2765 
2766 hwp_cpu_matched:
2767 	/*
2768 	 * The Intel pstate driver will be ignored if the platform
2769 	 * firmware has its own power management modes.
2770 	 */
2771 	if (intel_pstate_platform_pwr_mgmt_exists())
2772 		return -ENODEV;
2773 
2774 	if (!hwp_active && hwp_only)
2775 		return -ENOTSUPP;
2776 
2777 	pr_info("Intel P-state driver initializing\n");
2778 
2779 	all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
2780 	if (!all_cpu_data)
2781 		return -ENOMEM;
2782 
2783 	intel_pstate_request_control_from_smm();
2784 
2785 	intel_pstate_sysfs_expose_params();
2786 
2787 	mutex_lock(&intel_pstate_driver_lock);
2788 	rc = intel_pstate_register_driver();
2789 	mutex_unlock(&intel_pstate_driver_lock);
2790 	if (rc)
2791 		return rc;
2792 
2793 	if (hwp_active)
2794 		pr_info("HWP enabled\n");
2795 
2796 	return 0;
2797 }
2798 device_initcall(intel_pstate_init);
2799 
2800 static int __init intel_pstate_setup(char *str)
2801 {
2802 	if (!str)
2803 		return -EINVAL;
2804 
2805 	if (!strcmp(str, "disable")) {
2806 		no_load = 1;
2807 	} else if (!strcmp(str, "passive")) {
2808 		pr_info("Passive mode enabled\n");
2809 		intel_pstate_driver = &intel_cpufreq;
2810 		no_hwp = 1;
2811 	}
2812 	if (!strcmp(str, "no_hwp")) {
2813 		pr_info("HWP disabled\n");
2814 		no_hwp = 1;
2815 	}
2816 	if (!strcmp(str, "force"))
2817 		force_load = 1;
2818 	if (!strcmp(str, "hwp_only"))
2819 		hwp_only = 1;
2820 	if (!strcmp(str, "per_cpu_perf_limits"))
2821 		per_cpu_limits = true;
2822 
2823 #ifdef CONFIG_ACPI
2824 	if (!strcmp(str, "support_acpi_ppc"))
2825 		acpi_ppc = true;
2826 #endif
2827 
2828 	return 0;
2829 }
2830 early_param("intel_pstate", intel_pstate_setup);
2831 
2832 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
2833 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
2834 MODULE_LICENSE("GPL");
2835