xref: /linux/drivers/cpufreq/amd-pstate.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * amd-pstate.c - AMD Processor P-state Frequency Driver
4  *
5  * Copyright (C) 2021 Advanced Micro Devices, Inc. All Rights Reserved.
6  *
7  * Author: Huang Rui <ray.huang@amd.com>
8  *
9  * AMD P-State introduces a new CPU performance scaling design for AMD
10  * processors using the ACPI Collaborative Performance and Power Control (CPPC)
11  * feature which works with the AMD SMU firmware providing a finer grained
12  * frequency control range. It is to replace the legacy ACPI P-States control,
13  * allows a flexible, low-latency interface for the Linux kernel to directly
14  * communicate the performance hints to hardware.
15  *
16  * AMD P-State is supported on recent AMD Zen base CPU series include some of
17  * Zen2 and Zen3 processors. _CPC needs to be present in the ACPI tables of AMD
18  * P-State supported system. And there are two types of hardware implementations
19  * for AMD P-State: 1) Full MSR Solution and 2) Shared Memory Solution.
20  * X86_FEATURE_CPPC CPU feature flag is used to distinguish the different types.
21  */
22 
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 
25 #include <linux/bitfield.h>
26 #include <linux/kernel.h>
27 #include <linux/module.h>
28 #include <linux/init.h>
29 #include <linux/smp.h>
30 #include <linux/sched.h>
31 #include <linux/cpufreq.h>
32 #include <linux/compiler.h>
33 #include <linux/dmi.h>
34 #include <linux/slab.h>
35 #include <linux/acpi.h>
36 #include <linux/io.h>
37 #include <linux/delay.h>
38 #include <linux/uaccess.h>
39 #include <linux/static_call.h>
40 #include <linux/topology.h>
41 
42 #include <acpi/processor.h>
43 #include <acpi/cppc_acpi.h>
44 
45 #include <asm/msr.h>
46 #include <asm/processor.h>
47 #include <asm/cpufeature.h>
48 #include <asm/cpu_device_id.h>
49 
50 #include "amd-pstate.h"
51 #include "amd-pstate-trace.h"
52 
53 #define AMD_PSTATE_TRANSITION_LATENCY	20000
54 #define AMD_PSTATE_TRANSITION_DELAY	1000
55 #define AMD_PSTATE_FAST_CPPC_TRANSITION_DELAY 600
56 
57 #define AMD_CPPC_EPP_PERFORMANCE		0x00
58 #define AMD_CPPC_EPP_BALANCE_PERFORMANCE	0x80
59 #define AMD_CPPC_EPP_BALANCE_POWERSAVE		0xBF
60 #define AMD_CPPC_EPP_POWERSAVE			0xFF
61 
62 static const char * const amd_pstate_mode_string[] = {
63 	[AMD_PSTATE_UNDEFINED]   = "undefined",
64 	[AMD_PSTATE_DISABLE]     = "disable",
65 	[AMD_PSTATE_PASSIVE]     = "passive",
66 	[AMD_PSTATE_ACTIVE]      = "active",
67 	[AMD_PSTATE_GUIDED]      = "guided",
68 };
69 static_assert(ARRAY_SIZE(amd_pstate_mode_string) == AMD_PSTATE_MAX);
70 
71 const char *amd_pstate_get_mode_string(enum amd_pstate_mode mode)
72 {
73 	if (mode < AMD_PSTATE_UNDEFINED || mode >= AMD_PSTATE_MAX)
74 		mode = AMD_PSTATE_UNDEFINED;
75 	return amd_pstate_mode_string[mode];
76 }
77 EXPORT_SYMBOL_GPL(amd_pstate_get_mode_string);
78 
79 struct quirk_entry {
80 	u32 nominal_freq;
81 	u32 lowest_freq;
82 };
83 
84 static struct cpufreq_driver *current_pstate_driver;
85 static struct cpufreq_driver amd_pstate_driver;
86 static struct cpufreq_driver amd_pstate_epp_driver;
87 static int cppc_state = AMD_PSTATE_UNDEFINED;
88 static bool amd_pstate_prefcore = true;
89 static struct quirk_entry *quirks;
90 
91 /*
92  * AMD Energy Preference Performance (EPP)
93  * The EPP is used in the CCLK DPM controller to drive
94  * the frequency that a core is going to operate during
95  * short periods of activity. EPP values will be utilized for
96  * different OS profiles (balanced, performance, power savings)
97  * display strings corresponding to EPP index in the
98  * energy_perf_strings[]
99  *	index		String
100  *-------------------------------------
101  *	0		default
102  *	1		performance
103  *	2		balance_performance
104  *	3		balance_power
105  *	4		power
106  */
107 enum energy_perf_value_index {
108 	EPP_INDEX_DEFAULT = 0,
109 	EPP_INDEX_PERFORMANCE,
110 	EPP_INDEX_BALANCE_PERFORMANCE,
111 	EPP_INDEX_BALANCE_POWERSAVE,
112 	EPP_INDEX_POWERSAVE,
113 	EPP_INDEX_MAX,
114 };
115 
116 static const char * const energy_perf_strings[] = {
117 	[EPP_INDEX_DEFAULT] = "default",
118 	[EPP_INDEX_PERFORMANCE] = "performance",
119 	[EPP_INDEX_BALANCE_PERFORMANCE] = "balance_performance",
120 	[EPP_INDEX_BALANCE_POWERSAVE] = "balance_power",
121 	[EPP_INDEX_POWERSAVE] = "power",
122 };
123 static_assert(ARRAY_SIZE(energy_perf_strings) == EPP_INDEX_MAX);
124 
125 static unsigned int epp_values[] = {
126 	[EPP_INDEX_DEFAULT] = 0,
127 	[EPP_INDEX_PERFORMANCE] = AMD_CPPC_EPP_PERFORMANCE,
128 	[EPP_INDEX_BALANCE_PERFORMANCE] = AMD_CPPC_EPP_BALANCE_PERFORMANCE,
129 	[EPP_INDEX_BALANCE_POWERSAVE] = AMD_CPPC_EPP_BALANCE_POWERSAVE,
130 	[EPP_INDEX_POWERSAVE] = AMD_CPPC_EPP_POWERSAVE,
131 };
132 static_assert(ARRAY_SIZE(epp_values) == EPP_INDEX_MAX);
133 
134 typedef int (*cppc_mode_transition_fn)(int);
135 
136 static struct quirk_entry quirk_amd_7k62 = {
137 	.nominal_freq = 2600,
138 	.lowest_freq = 550,
139 };
140 
141 static inline u8 freq_to_perf(union perf_cached perf, u32 nominal_freq, unsigned int freq_val)
142 {
143 	u32 perf_val = DIV_ROUND_UP_ULL((u64)freq_val * perf.nominal_perf, nominal_freq);
144 
145 	return (u8)clamp(perf_val, perf.lowest_perf, perf.highest_perf);
146 }
147 
148 static inline u32 perf_to_freq(union perf_cached perf, u32 nominal_freq, u8 perf_val)
149 {
150 	return DIV_ROUND_UP_ULL((u64)nominal_freq * perf_val,
151 				perf.nominal_perf);
152 }
153 
154 static int __init dmi_matched_7k62_bios_bug(const struct dmi_system_id *dmi)
155 {
156 	/**
157 	 * match the broken bios for family 17h processor support CPPC V2
158 	 * broken BIOS lack of nominal_freq and lowest_freq capabilities
159 	 * definition in ACPI tables
160 	 */
161 	if (cpu_feature_enabled(X86_FEATURE_ZEN2)) {
162 		quirks = dmi->driver_data;
163 		pr_info("Overriding nominal and lowest frequencies for %s\n", dmi->ident);
164 		return 1;
165 	}
166 
167 	return 0;
168 }
169 
170 static const struct dmi_system_id amd_pstate_quirks_table[] __initconst = {
171 	{
172 		.callback = dmi_matched_7k62_bios_bug,
173 		.ident = "AMD EPYC 7K62",
174 		.matches = {
175 			DMI_MATCH(DMI_BIOS_VERSION, "5.14"),
176 			DMI_MATCH(DMI_BIOS_RELEASE, "12/12/2019"),
177 		},
178 		.driver_data = &quirk_amd_7k62,
179 	},
180 	{}
181 };
182 MODULE_DEVICE_TABLE(dmi, amd_pstate_quirks_table);
183 
184 static inline int get_mode_idx_from_str(const char *str, size_t size)
185 {
186 	int i;
187 
188 	for (i = 0; i < AMD_PSTATE_MAX; i++) {
189 		if (!strncmp(str, amd_pstate_mode_string[i], size))
190 			return i;
191 	}
192 	return -EINVAL;
193 }
194 
195 static DEFINE_MUTEX(amd_pstate_driver_lock);
196 
197 static u8 msr_get_epp(struct amd_cpudata *cpudata)
198 {
199 	u64 value;
200 	int ret;
201 
202 	ret = rdmsrq_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &value);
203 	if (ret < 0) {
204 		pr_debug("Could not retrieve energy perf value (%d)\n", ret);
205 		return ret;
206 	}
207 
208 	return FIELD_GET(AMD_CPPC_EPP_PERF_MASK, value);
209 }
210 
211 DEFINE_STATIC_CALL(amd_pstate_get_epp, msr_get_epp);
212 
213 static inline s16 amd_pstate_get_epp(struct amd_cpudata *cpudata)
214 {
215 	return static_call(amd_pstate_get_epp)(cpudata);
216 }
217 
218 static u8 shmem_get_epp(struct amd_cpudata *cpudata)
219 {
220 	u64 epp;
221 	int ret;
222 
223 	ret = cppc_get_epp_perf(cpudata->cpu, &epp);
224 	if (ret < 0) {
225 		pr_debug("Could not retrieve energy perf value (%d)\n", ret);
226 		return ret;
227 	}
228 
229 	return FIELD_GET(AMD_CPPC_EPP_PERF_MASK, epp);
230 }
231 
232 static int msr_update_perf(struct cpufreq_policy *policy, u8 min_perf,
233 			   u8 des_perf, u8 max_perf, u8 epp, bool fast_switch)
234 {
235 	struct amd_cpudata *cpudata = policy->driver_data;
236 	u64 value, prev;
237 
238 	value = prev = READ_ONCE(cpudata->cppc_req_cached);
239 
240 	value &= ~(AMD_CPPC_MAX_PERF_MASK | AMD_CPPC_MIN_PERF_MASK |
241 		   AMD_CPPC_DES_PERF_MASK | AMD_CPPC_EPP_PERF_MASK);
242 	value |= FIELD_PREP(AMD_CPPC_MAX_PERF_MASK, max_perf);
243 	value |= FIELD_PREP(AMD_CPPC_DES_PERF_MASK, des_perf);
244 	value |= FIELD_PREP(AMD_CPPC_MIN_PERF_MASK, min_perf);
245 	value |= FIELD_PREP(AMD_CPPC_EPP_PERF_MASK, epp);
246 
247 	if (trace_amd_pstate_epp_perf_enabled()) {
248 		union perf_cached perf = READ_ONCE(cpudata->perf);
249 
250 		trace_amd_pstate_epp_perf(cpudata->cpu,
251 					  perf.highest_perf,
252 					  epp,
253 					  min_perf,
254 					  max_perf,
255 					  policy->boost_enabled,
256 					  value != prev);
257 	}
258 
259 	if (value == prev)
260 		return 0;
261 
262 	if (fast_switch) {
263 		wrmsrq(MSR_AMD_CPPC_REQ, value);
264 		return 0;
265 	} else {
266 		int ret = wrmsrq_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
267 
268 		if (ret)
269 			return ret;
270 	}
271 
272 	WRITE_ONCE(cpudata->cppc_req_cached, value);
273 
274 	return 0;
275 }
276 
277 DEFINE_STATIC_CALL(amd_pstate_update_perf, msr_update_perf);
278 
279 static inline int amd_pstate_update_perf(struct cpufreq_policy *policy,
280 					  u8 min_perf, u8 des_perf,
281 					  u8 max_perf, u8 epp,
282 					  bool fast_switch)
283 {
284 	return static_call(amd_pstate_update_perf)(policy, min_perf, des_perf,
285 						   max_perf, epp, fast_switch);
286 }
287 
288 static int msr_set_epp(struct cpufreq_policy *policy, u8 epp)
289 {
290 	struct amd_cpudata *cpudata = policy->driver_data;
291 	u64 value, prev;
292 	int ret;
293 
294 	value = prev = READ_ONCE(cpudata->cppc_req_cached);
295 	value &= ~AMD_CPPC_EPP_PERF_MASK;
296 	value |= FIELD_PREP(AMD_CPPC_EPP_PERF_MASK, epp);
297 
298 	if (trace_amd_pstate_epp_perf_enabled()) {
299 		union perf_cached perf = cpudata->perf;
300 
301 		trace_amd_pstate_epp_perf(cpudata->cpu, perf.highest_perf,
302 					  epp,
303 					  FIELD_GET(AMD_CPPC_MIN_PERF_MASK,
304 						    cpudata->cppc_req_cached),
305 					  FIELD_GET(AMD_CPPC_MAX_PERF_MASK,
306 						    cpudata->cppc_req_cached),
307 					  policy->boost_enabled,
308 					  value != prev);
309 	}
310 
311 	if (value == prev)
312 		return 0;
313 
314 	ret = wrmsrq_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
315 	if (ret) {
316 		pr_err("failed to set energy perf value (%d)\n", ret);
317 		return ret;
318 	}
319 
320 	/* update both so that msr_update_perf() can effectively check */
321 	WRITE_ONCE(cpudata->cppc_req_cached, value);
322 
323 	return ret;
324 }
325 
326 DEFINE_STATIC_CALL(amd_pstate_set_epp, msr_set_epp);
327 
328 static inline int amd_pstate_set_epp(struct cpufreq_policy *policy, u8 epp)
329 {
330 	return static_call(amd_pstate_set_epp)(policy, epp);
331 }
332 
333 static int shmem_set_epp(struct cpufreq_policy *policy, u8 epp)
334 {
335 	struct amd_cpudata *cpudata = policy->driver_data;
336 	struct cppc_perf_ctrls perf_ctrls;
337 	u8 epp_cached;
338 	u64 value;
339 	int ret;
340 
341 
342 	epp_cached = FIELD_GET(AMD_CPPC_EPP_PERF_MASK, cpudata->cppc_req_cached);
343 	if (trace_amd_pstate_epp_perf_enabled()) {
344 		union perf_cached perf = cpudata->perf;
345 
346 		trace_amd_pstate_epp_perf(cpudata->cpu, perf.highest_perf,
347 					  epp,
348 					  FIELD_GET(AMD_CPPC_MIN_PERF_MASK,
349 						    cpudata->cppc_req_cached),
350 					  FIELD_GET(AMD_CPPC_MAX_PERF_MASK,
351 						    cpudata->cppc_req_cached),
352 					  policy->boost_enabled,
353 					  epp != epp_cached);
354 	}
355 
356 	if (epp == epp_cached)
357 		return 0;
358 
359 	perf_ctrls.energy_perf = epp;
360 	ret = cppc_set_epp_perf(cpudata->cpu, &perf_ctrls, 1);
361 	if (ret) {
362 		pr_debug("failed to set energy perf value (%d)\n", ret);
363 		return ret;
364 	}
365 
366 	value = READ_ONCE(cpudata->cppc_req_cached);
367 	value &= ~AMD_CPPC_EPP_PERF_MASK;
368 	value |= FIELD_PREP(AMD_CPPC_EPP_PERF_MASK, epp);
369 	WRITE_ONCE(cpudata->cppc_req_cached, value);
370 
371 	return ret;
372 }
373 
374 static inline int msr_cppc_enable(struct cpufreq_policy *policy)
375 {
376 	return wrmsrq_safe_on_cpu(policy->cpu, MSR_AMD_CPPC_ENABLE, 1);
377 }
378 
379 static int shmem_cppc_enable(struct cpufreq_policy *policy)
380 {
381 	return cppc_set_enable(policy->cpu, 1);
382 }
383 
384 DEFINE_STATIC_CALL(amd_pstate_cppc_enable, msr_cppc_enable);
385 
386 static inline int amd_pstate_cppc_enable(struct cpufreq_policy *policy)
387 {
388 	return static_call(amd_pstate_cppc_enable)(policy);
389 }
390 
391 static int msr_init_perf(struct amd_cpudata *cpudata)
392 {
393 	union perf_cached perf = READ_ONCE(cpudata->perf);
394 	u64 cap1, numerator, cppc_req;
395 	u8 min_perf;
396 
397 	int ret = rdmsrq_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
398 				     &cap1);
399 	if (ret)
400 		return ret;
401 
402 	ret = amd_get_boost_ratio_numerator(cpudata->cpu, &numerator);
403 	if (ret)
404 		return ret;
405 
406 	ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &cppc_req);
407 	if (ret)
408 		return ret;
409 
410 	WRITE_ONCE(cpudata->cppc_req_cached, cppc_req);
411 	min_perf = FIELD_GET(AMD_CPPC_MIN_PERF_MASK, cppc_req);
412 
413 	/*
414 	 * Clear out the min_perf part to check if the rest of the MSR is 0, if yes, this is an
415 	 * indication that the min_perf value is the one specified through the BIOS option
416 	 */
417 	cppc_req &= ~(AMD_CPPC_MIN_PERF_MASK);
418 
419 	if (!cppc_req)
420 		perf.bios_min_perf = min_perf;
421 
422 	perf.highest_perf = numerator;
423 	perf.max_limit_perf = numerator;
424 	perf.min_limit_perf = FIELD_GET(AMD_CPPC_LOWEST_PERF_MASK, cap1);
425 	perf.nominal_perf = FIELD_GET(AMD_CPPC_NOMINAL_PERF_MASK, cap1);
426 	perf.lowest_nonlinear_perf = FIELD_GET(AMD_CPPC_LOWNONLIN_PERF_MASK, cap1);
427 	perf.lowest_perf = FIELD_GET(AMD_CPPC_LOWEST_PERF_MASK, cap1);
428 	WRITE_ONCE(cpudata->perf, perf);
429 	WRITE_ONCE(cpudata->prefcore_ranking, FIELD_GET(AMD_CPPC_HIGHEST_PERF_MASK, cap1));
430 
431 	return 0;
432 }
433 
434 static int shmem_init_perf(struct amd_cpudata *cpudata)
435 {
436 	struct cppc_perf_caps cppc_perf;
437 	union perf_cached perf = READ_ONCE(cpudata->perf);
438 	u64 numerator;
439 	bool auto_sel;
440 
441 	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
442 	if (ret)
443 		return ret;
444 
445 	ret = amd_get_boost_ratio_numerator(cpudata->cpu, &numerator);
446 	if (ret)
447 		return ret;
448 
449 	perf.highest_perf = numerator;
450 	perf.max_limit_perf = numerator;
451 	perf.min_limit_perf = cppc_perf.lowest_perf;
452 	perf.nominal_perf = cppc_perf.nominal_perf;
453 	perf.lowest_nonlinear_perf = cppc_perf.lowest_nonlinear_perf;
454 	perf.lowest_perf = cppc_perf.lowest_perf;
455 	WRITE_ONCE(cpudata->perf, perf);
456 	WRITE_ONCE(cpudata->prefcore_ranking, cppc_perf.highest_perf);
457 
458 	if (cppc_state == AMD_PSTATE_ACTIVE)
459 		return 0;
460 
461 	ret = cppc_get_auto_sel(cpudata->cpu, &auto_sel);
462 	if (ret) {
463 		pr_warn("failed to get auto_sel, ret: %d\n", ret);
464 		return 0;
465 	}
466 
467 	ret = cppc_set_auto_sel(cpudata->cpu,
468 			(cppc_state == AMD_PSTATE_PASSIVE) ? 0 : 1);
469 
470 	if (ret)
471 		pr_warn("failed to set auto_sel, ret: %d\n", ret);
472 
473 	return ret;
474 }
475 
476 DEFINE_STATIC_CALL(amd_pstate_init_perf, msr_init_perf);
477 
478 static inline int amd_pstate_init_perf(struct amd_cpudata *cpudata)
479 {
480 	return static_call(amd_pstate_init_perf)(cpudata);
481 }
482 
483 static int shmem_update_perf(struct cpufreq_policy *policy, u8 min_perf,
484 			     u8 des_perf, u8 max_perf, u8 epp, bool fast_switch)
485 {
486 	struct amd_cpudata *cpudata = policy->driver_data;
487 	struct cppc_perf_ctrls perf_ctrls;
488 	u64 value, prev;
489 	int ret;
490 
491 	if (cppc_state == AMD_PSTATE_ACTIVE) {
492 		int ret = shmem_set_epp(policy, epp);
493 
494 		if (ret)
495 			return ret;
496 	}
497 
498 	value = prev = READ_ONCE(cpudata->cppc_req_cached);
499 
500 	value &= ~(AMD_CPPC_MAX_PERF_MASK | AMD_CPPC_MIN_PERF_MASK |
501 		   AMD_CPPC_DES_PERF_MASK | AMD_CPPC_EPP_PERF_MASK);
502 	value |= FIELD_PREP(AMD_CPPC_MAX_PERF_MASK, max_perf);
503 	value |= FIELD_PREP(AMD_CPPC_DES_PERF_MASK, des_perf);
504 	value |= FIELD_PREP(AMD_CPPC_MIN_PERF_MASK, min_perf);
505 	value |= FIELD_PREP(AMD_CPPC_EPP_PERF_MASK, epp);
506 
507 	if (trace_amd_pstate_epp_perf_enabled()) {
508 		union perf_cached perf = READ_ONCE(cpudata->perf);
509 
510 		trace_amd_pstate_epp_perf(cpudata->cpu,
511 					  perf.highest_perf,
512 					  epp,
513 					  min_perf,
514 					  max_perf,
515 					  policy->boost_enabled,
516 					  value != prev);
517 	}
518 
519 	if (value == prev)
520 		return 0;
521 
522 	perf_ctrls.max_perf = max_perf;
523 	perf_ctrls.min_perf = min_perf;
524 	perf_ctrls.desired_perf = des_perf;
525 
526 	ret = cppc_set_perf(cpudata->cpu, &perf_ctrls);
527 	if (ret)
528 		return ret;
529 
530 	WRITE_ONCE(cpudata->cppc_req_cached, value);
531 
532 	return 0;
533 }
534 
535 static inline bool amd_pstate_sample(struct amd_cpudata *cpudata)
536 {
537 	u64 aperf, mperf, tsc;
538 	unsigned long flags;
539 
540 	local_irq_save(flags);
541 	rdmsrq(MSR_IA32_APERF, aperf);
542 	rdmsrq(MSR_IA32_MPERF, mperf);
543 	tsc = rdtsc();
544 
545 	if (cpudata->prev.mperf == mperf || cpudata->prev.tsc == tsc) {
546 		local_irq_restore(flags);
547 		return false;
548 	}
549 
550 	local_irq_restore(flags);
551 
552 	cpudata->cur.aperf = aperf;
553 	cpudata->cur.mperf = mperf;
554 	cpudata->cur.tsc =  tsc;
555 	cpudata->cur.aperf -= cpudata->prev.aperf;
556 	cpudata->cur.mperf -= cpudata->prev.mperf;
557 	cpudata->cur.tsc -= cpudata->prev.tsc;
558 
559 	cpudata->prev.aperf = aperf;
560 	cpudata->prev.mperf = mperf;
561 	cpudata->prev.tsc = tsc;
562 
563 	cpudata->freq = div64_u64((cpudata->cur.aperf * cpu_khz), cpudata->cur.mperf);
564 
565 	return true;
566 }
567 
568 static void amd_pstate_update(struct amd_cpudata *cpudata, u8 min_perf,
569 			      u8 des_perf, u8 max_perf, bool fast_switch, int gov_flags)
570 {
571 	struct cpufreq_policy *policy __free(put_cpufreq_policy) = cpufreq_cpu_get(cpudata->cpu);
572 	union perf_cached perf = READ_ONCE(cpudata->perf);
573 
574 	if (!policy)
575 		return;
576 
577 	/* limit the max perf when core performance boost feature is disabled */
578 	if (!cpudata->boost_supported)
579 		max_perf = min_t(u8, perf.nominal_perf, max_perf);
580 
581 	des_perf = clamp_t(u8, des_perf, min_perf, max_perf);
582 
583 	policy->cur = perf_to_freq(perf, cpudata->nominal_freq, des_perf);
584 
585 	if ((cppc_state == AMD_PSTATE_GUIDED) && (gov_flags & CPUFREQ_GOV_DYNAMIC_SWITCHING)) {
586 		min_perf = des_perf;
587 		des_perf = 0;
588 	}
589 
590 	if (trace_amd_pstate_perf_enabled() && amd_pstate_sample(cpudata)) {
591 		trace_amd_pstate_perf(min_perf, des_perf, max_perf, cpudata->freq,
592 			cpudata->cur.mperf, cpudata->cur.aperf, cpudata->cur.tsc,
593 				cpudata->cpu, fast_switch);
594 	}
595 
596 	amd_pstate_update_perf(policy, min_perf, des_perf, max_perf, 0, fast_switch);
597 }
598 
599 static int amd_pstate_verify(struct cpufreq_policy_data *policy_data)
600 {
601 	/*
602 	 * Initialize lower frequency limit (i.e.policy->min) with
603 	 * lowest_nonlinear_frequency or the min frequency (if) specified in BIOS,
604 	 * Override the initial value set by cpufreq core and amd-pstate qos_requests.
605 	 */
606 	if (policy_data->min == FREQ_QOS_MIN_DEFAULT_VALUE) {
607 		struct cpufreq_policy *policy __free(put_cpufreq_policy) =
608 					      cpufreq_cpu_get(policy_data->cpu);
609 		struct amd_cpudata *cpudata;
610 		union perf_cached perf;
611 
612 		if (!policy)
613 			return -EINVAL;
614 
615 		cpudata = policy->driver_data;
616 		perf = READ_ONCE(cpudata->perf);
617 
618 		if (perf.bios_min_perf)
619 			policy_data->min = perf_to_freq(perf, cpudata->nominal_freq,
620 							perf.bios_min_perf);
621 		else
622 			policy_data->min = cpudata->lowest_nonlinear_freq;
623 	}
624 
625 	cpufreq_verify_within_cpu_limits(policy_data);
626 
627 	return 0;
628 }
629 
630 static void amd_pstate_update_min_max_limit(struct cpufreq_policy *policy)
631 {
632 	struct amd_cpudata *cpudata = policy->driver_data;
633 	union perf_cached perf = READ_ONCE(cpudata->perf);
634 
635 	perf.max_limit_perf = freq_to_perf(perf, cpudata->nominal_freq, policy->max);
636 	WRITE_ONCE(cpudata->max_limit_freq, policy->max);
637 
638 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE) {
639 		perf.min_limit_perf = min(perf.nominal_perf, perf.max_limit_perf);
640 		WRITE_ONCE(cpudata->min_limit_freq, min(cpudata->nominal_freq, cpudata->max_limit_freq));
641 	} else {
642 		perf.min_limit_perf = freq_to_perf(perf, cpudata->nominal_freq, policy->min);
643 		WRITE_ONCE(cpudata->min_limit_freq, policy->min);
644 	}
645 
646 	WRITE_ONCE(cpudata->perf, perf);
647 }
648 
649 static int amd_pstate_update_freq(struct cpufreq_policy *policy,
650 				  unsigned int target_freq, bool fast_switch)
651 {
652 	struct cpufreq_freqs freqs;
653 	struct amd_cpudata *cpudata;
654 	union perf_cached perf;
655 	u8 des_perf;
656 
657 	cpudata = policy->driver_data;
658 
659 	if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)
660 		amd_pstate_update_min_max_limit(policy);
661 
662 	perf = READ_ONCE(cpudata->perf);
663 
664 	freqs.old = policy->cur;
665 	freqs.new = target_freq;
666 
667 	des_perf = freq_to_perf(perf, cpudata->nominal_freq, target_freq);
668 
669 	WARN_ON(fast_switch && !policy->fast_switch_enabled);
670 	/*
671 	 * If fast_switch is desired, then there aren't any registered
672 	 * transition notifiers. See comment for
673 	 * cpufreq_enable_fast_switch().
674 	 */
675 	if (!fast_switch)
676 		cpufreq_freq_transition_begin(policy, &freqs);
677 
678 	amd_pstate_update(cpudata, perf.min_limit_perf, des_perf,
679 			  perf.max_limit_perf, fast_switch,
680 			  policy->governor->flags);
681 
682 	if (!fast_switch)
683 		cpufreq_freq_transition_end(policy, &freqs, false);
684 
685 	return 0;
686 }
687 
688 static int amd_pstate_target(struct cpufreq_policy *policy,
689 			     unsigned int target_freq,
690 			     unsigned int relation)
691 {
692 	return amd_pstate_update_freq(policy, target_freq, false);
693 }
694 
695 static unsigned int amd_pstate_fast_switch(struct cpufreq_policy *policy,
696 				  unsigned int target_freq)
697 {
698 	if (!amd_pstate_update_freq(policy, target_freq, true))
699 		return target_freq;
700 	return policy->cur;
701 }
702 
703 static void amd_pstate_adjust_perf(unsigned int cpu,
704 				   unsigned long _min_perf,
705 				   unsigned long target_perf,
706 				   unsigned long capacity)
707 {
708 	u8 max_perf, min_perf, des_perf, cap_perf;
709 	struct cpufreq_policy *policy __free(put_cpufreq_policy) = cpufreq_cpu_get(cpu);
710 	struct amd_cpudata *cpudata;
711 	union perf_cached perf;
712 
713 	if (!policy)
714 		return;
715 
716 	cpudata = policy->driver_data;
717 
718 	if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)
719 		amd_pstate_update_min_max_limit(policy);
720 
721 	perf = READ_ONCE(cpudata->perf);
722 	cap_perf = perf.highest_perf;
723 
724 	des_perf = cap_perf;
725 	if (target_perf < capacity)
726 		des_perf = DIV_ROUND_UP(cap_perf * target_perf, capacity);
727 
728 	if (_min_perf < capacity)
729 		min_perf = DIV_ROUND_UP(cap_perf * _min_perf, capacity);
730 	else
731 		min_perf = cap_perf;
732 
733 	if (min_perf < perf.min_limit_perf)
734 		min_perf = perf.min_limit_perf;
735 
736 	max_perf = perf.max_limit_perf;
737 	if (max_perf < min_perf)
738 		max_perf = min_perf;
739 
740 	amd_pstate_update(cpudata, min_perf, des_perf, max_perf, true,
741 			policy->governor->flags);
742 }
743 
744 static int amd_pstate_cpu_boost_update(struct cpufreq_policy *policy, bool on)
745 {
746 	struct amd_cpudata *cpudata = policy->driver_data;
747 	union perf_cached perf = READ_ONCE(cpudata->perf);
748 	u32 nominal_freq, max_freq;
749 	int ret = 0;
750 
751 	nominal_freq = READ_ONCE(cpudata->nominal_freq);
752 	max_freq = perf_to_freq(perf, cpudata->nominal_freq, perf.highest_perf);
753 
754 	if (on)
755 		policy->cpuinfo.max_freq = max_freq;
756 	else if (policy->cpuinfo.max_freq > nominal_freq)
757 		policy->cpuinfo.max_freq = nominal_freq;
758 
759 	policy->max = policy->cpuinfo.max_freq;
760 
761 	if (cppc_state == AMD_PSTATE_PASSIVE) {
762 		ret = freq_qos_update_request(&cpudata->req[1], policy->cpuinfo.max_freq);
763 		if (ret < 0)
764 			pr_debug("Failed to update freq constraint: CPU%d\n", cpudata->cpu);
765 	}
766 
767 	return ret < 0 ? ret : 0;
768 }
769 
770 static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
771 {
772 	struct amd_cpudata *cpudata = policy->driver_data;
773 	int ret;
774 
775 	if (!cpudata->boost_supported) {
776 		pr_err("Boost mode is not supported by this processor or SBIOS\n");
777 		return -EOPNOTSUPP;
778 	}
779 
780 	ret = amd_pstate_cpu_boost_update(policy, state);
781 	refresh_frequency_limits(policy);
782 
783 	return ret;
784 }
785 
786 static int amd_pstate_init_boost_support(struct amd_cpudata *cpudata)
787 {
788 	u64 boost_val;
789 	int ret = -1;
790 
791 	/*
792 	 * If platform has no CPB support or disable it, initialize current driver
793 	 * boost_enabled state to be false, it is not an error for cpufreq core to handle.
794 	 */
795 	if (!cpu_feature_enabled(X86_FEATURE_CPB)) {
796 		pr_debug_once("Boost CPB capabilities not present in the processor\n");
797 		ret = 0;
798 		goto exit_err;
799 	}
800 
801 	ret = rdmsrq_on_cpu(cpudata->cpu, MSR_K7_HWCR, &boost_val);
802 	if (ret) {
803 		pr_err_once("failed to read initial CPU boost state!\n");
804 		ret = -EIO;
805 		goto exit_err;
806 	}
807 
808 	if (!(boost_val & MSR_K7_HWCR_CPB_DIS))
809 		cpudata->boost_supported = true;
810 
811 	return 0;
812 
813 exit_err:
814 	cpudata->boost_supported = false;
815 	return ret;
816 }
817 
818 static void amd_perf_ctl_reset(unsigned int cpu)
819 {
820 	wrmsrq_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
821 }
822 
823 #define CPPC_MAX_PERF	U8_MAX
824 
825 static void amd_pstate_init_prefcore(struct amd_cpudata *cpudata)
826 {
827 	/* user disabled or not detected */
828 	if (!amd_pstate_prefcore)
829 		return;
830 
831 	/* should use amd-hfi instead */
832 	if (cpu_feature_enabled(X86_FEATURE_AMD_WORKLOAD_CLASS) &&
833 	    IS_ENABLED(CONFIG_AMD_HFI)) {
834 		amd_pstate_prefcore = false;
835 		return;
836 	}
837 
838 	cpudata->hw_prefcore = true;
839 
840 	/* Priorities must be initialized before ITMT support can be toggled on. */
841 	sched_set_itmt_core_prio((int)READ_ONCE(cpudata->prefcore_ranking), cpudata->cpu);
842 }
843 
844 static void amd_pstate_update_limits(struct cpufreq_policy *policy)
845 {
846 	struct amd_cpudata *cpudata;
847 	u32 prev_high = 0, cur_high = 0;
848 	bool highest_perf_changed = false;
849 	unsigned int cpu = policy->cpu;
850 
851 	if (!amd_pstate_prefcore)
852 		return;
853 
854 	if (amd_get_highest_perf(cpu, &cur_high))
855 		return;
856 
857 	cpudata = policy->driver_data;
858 
859 	prev_high = READ_ONCE(cpudata->prefcore_ranking);
860 	highest_perf_changed = (prev_high != cur_high);
861 	if (highest_perf_changed) {
862 		WRITE_ONCE(cpudata->prefcore_ranking, cur_high);
863 
864 		if (cur_high < CPPC_MAX_PERF) {
865 			sched_set_itmt_core_prio((int)cur_high, cpu);
866 			sched_update_asym_prefer_cpu(cpu, prev_high, cur_high);
867 		}
868 	}
869 }
870 
871 /*
872  * Get pstate transition delay time from ACPI tables that firmware set
873  * instead of using hardcode value directly.
874  */
875 static u32 amd_pstate_get_transition_delay_us(unsigned int cpu)
876 {
877 	int transition_delay_ns;
878 
879 	transition_delay_ns = cppc_get_transition_latency(cpu);
880 	if (transition_delay_ns < 0) {
881 		if (cpu_feature_enabled(X86_FEATURE_AMD_FAST_CPPC))
882 			return AMD_PSTATE_FAST_CPPC_TRANSITION_DELAY;
883 		else
884 			return AMD_PSTATE_TRANSITION_DELAY;
885 	}
886 
887 	return transition_delay_ns / NSEC_PER_USEC;
888 }
889 
890 /*
891  * Get pstate transition latency value from ACPI tables that firmware
892  * set instead of using hardcode value directly.
893  */
894 static u32 amd_pstate_get_transition_latency(unsigned int cpu)
895 {
896 	int transition_latency;
897 
898 	transition_latency = cppc_get_transition_latency(cpu);
899 	if (transition_latency < 0)
900 		return AMD_PSTATE_TRANSITION_LATENCY;
901 
902 	return transition_latency;
903 }
904 
905 /*
906  * amd_pstate_init_freq: Initialize the nominal_freq and lowest_nonlinear_freq
907  *			 for the @cpudata object.
908  *
909  * Requires: all perf members of @cpudata to be initialized.
910  *
911  * Returns 0 on success, non-zero value on failure.
912  */
913 static int amd_pstate_init_freq(struct amd_cpudata *cpudata)
914 {
915 	u32 min_freq, max_freq, nominal_freq, lowest_nonlinear_freq;
916 	struct cppc_perf_caps cppc_perf;
917 	union perf_cached perf;
918 	int ret;
919 
920 	ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
921 	if (ret)
922 		return ret;
923 	perf = READ_ONCE(cpudata->perf);
924 
925 	if (quirks && quirks->nominal_freq)
926 		nominal_freq = quirks->nominal_freq;
927 	else
928 		nominal_freq = cppc_perf.nominal_freq;
929 	nominal_freq *= 1000;
930 
931 	if (quirks && quirks->lowest_freq) {
932 		min_freq = quirks->lowest_freq;
933 		perf.lowest_perf = freq_to_perf(perf, nominal_freq, min_freq);
934 		WRITE_ONCE(cpudata->perf, perf);
935 	} else
936 		min_freq = cppc_perf.lowest_freq;
937 
938 	min_freq *= 1000;
939 
940 	WRITE_ONCE(cpudata->nominal_freq, nominal_freq);
941 
942 	max_freq = perf_to_freq(perf, nominal_freq, perf.highest_perf);
943 	lowest_nonlinear_freq = perf_to_freq(perf, nominal_freq, perf.lowest_nonlinear_perf);
944 	WRITE_ONCE(cpudata->lowest_nonlinear_freq, lowest_nonlinear_freq);
945 
946 	/**
947 	 * Below values need to be initialized correctly, otherwise driver will fail to load
948 	 * max_freq is calculated according to (nominal_freq * highest_perf)/nominal_perf
949 	 * lowest_nonlinear_freq is a value between [min_freq, nominal_freq]
950 	 * Check _CPC in ACPI table objects if any values are incorrect
951 	 */
952 	if (min_freq <= 0 || max_freq <= 0 || nominal_freq <= 0 || min_freq > max_freq) {
953 		pr_err("min_freq(%d) or max_freq(%d) or nominal_freq(%d) value is incorrect\n",
954 			min_freq, max_freq, nominal_freq);
955 		return -EINVAL;
956 	}
957 
958 	if (lowest_nonlinear_freq <= min_freq || lowest_nonlinear_freq > nominal_freq) {
959 		pr_err("lowest_nonlinear_freq(%d) value is out of range [min_freq(%d), nominal_freq(%d)]\n",
960 			lowest_nonlinear_freq, min_freq, nominal_freq);
961 		return -EINVAL;
962 	}
963 
964 	return 0;
965 }
966 
967 static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
968 {
969 	struct amd_cpudata *cpudata;
970 	union perf_cached perf;
971 	struct device *dev;
972 	int ret;
973 
974 	/*
975 	 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
976 	 * which is ideal for initialization process.
977 	 */
978 	amd_perf_ctl_reset(policy->cpu);
979 	dev = get_cpu_device(policy->cpu);
980 	if (!dev)
981 		return -ENODEV;
982 
983 	cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
984 	if (!cpudata)
985 		return -ENOMEM;
986 
987 	cpudata->cpu = policy->cpu;
988 
989 	ret = amd_pstate_init_perf(cpudata);
990 	if (ret)
991 		goto free_cpudata1;
992 
993 	amd_pstate_init_prefcore(cpudata);
994 
995 	ret = amd_pstate_init_freq(cpudata);
996 	if (ret)
997 		goto free_cpudata1;
998 
999 	ret = amd_pstate_init_boost_support(cpudata);
1000 	if (ret)
1001 		goto free_cpudata1;
1002 
1003 	policy->cpuinfo.transition_latency = amd_pstate_get_transition_latency(policy->cpu);
1004 	policy->transition_delay_us = amd_pstate_get_transition_delay_us(policy->cpu);
1005 
1006 	perf = READ_ONCE(cpudata->perf);
1007 
1008 	policy->cpuinfo.min_freq = policy->min = perf_to_freq(perf,
1009 							      cpudata->nominal_freq,
1010 							      perf.lowest_perf);
1011 	policy->cpuinfo.max_freq = policy->max = perf_to_freq(perf,
1012 							      cpudata->nominal_freq,
1013 							      perf.highest_perf);
1014 
1015 	ret = amd_pstate_cppc_enable(policy);
1016 	if (ret)
1017 		goto free_cpudata1;
1018 
1019 	policy->boost_supported = READ_ONCE(cpudata->boost_supported);
1020 
1021 	/* It will be updated by governor */
1022 	policy->cur = policy->cpuinfo.min_freq;
1023 
1024 	if (cpu_feature_enabled(X86_FEATURE_CPPC))
1025 		policy->fast_switch_possible = true;
1026 
1027 	ret = freq_qos_add_request(&policy->constraints, &cpudata->req[0],
1028 				   FREQ_QOS_MIN, FREQ_QOS_MIN_DEFAULT_VALUE);
1029 	if (ret < 0) {
1030 		dev_err(dev, "Failed to add min-freq constraint (%d)\n", ret);
1031 		goto free_cpudata1;
1032 	}
1033 
1034 	ret = freq_qos_add_request(&policy->constraints, &cpudata->req[1],
1035 				   FREQ_QOS_MAX, policy->cpuinfo.max_freq);
1036 	if (ret < 0) {
1037 		dev_err(dev, "Failed to add max-freq constraint (%d)\n", ret);
1038 		goto free_cpudata2;
1039 	}
1040 
1041 	policy->driver_data = cpudata;
1042 
1043 	if (!current_pstate_driver->adjust_perf)
1044 		current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
1045 
1046 	return 0;
1047 
1048 free_cpudata2:
1049 	freq_qos_remove_request(&cpudata->req[0]);
1050 free_cpudata1:
1051 	pr_warn("Failed to initialize CPU %d: %d\n", policy->cpu, ret);
1052 	kfree(cpudata);
1053 	return ret;
1054 }
1055 
1056 static void amd_pstate_cpu_exit(struct cpufreq_policy *policy)
1057 {
1058 	struct amd_cpudata *cpudata = policy->driver_data;
1059 	union perf_cached perf = READ_ONCE(cpudata->perf);
1060 
1061 	/* Reset CPPC_REQ MSR to the BIOS value */
1062 	amd_pstate_update_perf(policy, perf.bios_min_perf, 0U, 0U, 0U, false);
1063 
1064 	freq_qos_remove_request(&cpudata->req[1]);
1065 	freq_qos_remove_request(&cpudata->req[0]);
1066 	policy->fast_switch_possible = false;
1067 	kfree(cpudata);
1068 }
1069 
1070 /* Sysfs attributes */
1071 
1072 /*
1073  * This frequency is to indicate the maximum hardware frequency.
1074  * If boost is not active but supported, the frequency will be larger than the
1075  * one in cpuinfo.
1076  */
1077 static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy,
1078 					char *buf)
1079 {
1080 	struct amd_cpudata *cpudata;
1081 	union perf_cached perf;
1082 
1083 	cpudata = policy->driver_data;
1084 	perf = READ_ONCE(cpudata->perf);
1085 
1086 	return sysfs_emit(buf, "%u\n",
1087 			  perf_to_freq(perf, cpudata->nominal_freq, perf.highest_perf));
1088 }
1089 
1090 static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *policy,
1091 						     char *buf)
1092 {
1093 	struct amd_cpudata *cpudata;
1094 	union perf_cached perf;
1095 
1096 	cpudata = policy->driver_data;
1097 	perf = READ_ONCE(cpudata->perf);
1098 
1099 	return sysfs_emit(buf, "%u\n",
1100 			  perf_to_freq(perf, cpudata->nominal_freq, perf.lowest_nonlinear_perf));
1101 }
1102 
1103 /*
1104  * In some of ASICs, the highest_perf is not the one in the _CPC table, so we
1105  * need to expose it to sysfs.
1106  */
1107 static ssize_t show_amd_pstate_highest_perf(struct cpufreq_policy *policy,
1108 					    char *buf)
1109 {
1110 	struct amd_cpudata *cpudata;
1111 
1112 	cpudata = policy->driver_data;
1113 
1114 	return sysfs_emit(buf, "%u\n", cpudata->perf.highest_perf);
1115 }
1116 
1117 static ssize_t show_amd_pstate_prefcore_ranking(struct cpufreq_policy *policy,
1118 						char *buf)
1119 {
1120 	u8 perf;
1121 	struct amd_cpudata *cpudata = policy->driver_data;
1122 
1123 	perf = READ_ONCE(cpudata->prefcore_ranking);
1124 
1125 	return sysfs_emit(buf, "%u\n", perf);
1126 }
1127 
1128 static ssize_t show_amd_pstate_hw_prefcore(struct cpufreq_policy *policy,
1129 					   char *buf)
1130 {
1131 	bool hw_prefcore;
1132 	struct amd_cpudata *cpudata = policy->driver_data;
1133 
1134 	hw_prefcore = READ_ONCE(cpudata->hw_prefcore);
1135 
1136 	return sysfs_emit(buf, "%s\n", str_enabled_disabled(hw_prefcore));
1137 }
1138 
1139 static ssize_t show_energy_performance_available_preferences(
1140 				struct cpufreq_policy *policy, char *buf)
1141 {
1142 	int offset = 0, i;
1143 	struct amd_cpudata *cpudata = policy->driver_data;
1144 
1145 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1146 		return sysfs_emit_at(buf, offset, "%s\n",
1147 				energy_perf_strings[EPP_INDEX_PERFORMANCE]);
1148 
1149 	for (i = 0; i < ARRAY_SIZE(energy_perf_strings); i++)
1150 		offset += sysfs_emit_at(buf, offset, "%s ", energy_perf_strings[i]);
1151 
1152 	offset += sysfs_emit_at(buf, offset, "\n");
1153 
1154 	return offset;
1155 }
1156 
1157 static ssize_t store_energy_performance_preference(
1158 		struct cpufreq_policy *policy, const char *buf, size_t count)
1159 {
1160 	struct amd_cpudata *cpudata = policy->driver_data;
1161 	ssize_t ret;
1162 	u8 epp;
1163 
1164 	ret = sysfs_match_string(energy_perf_strings, buf);
1165 	if (ret < 0)
1166 		return -EINVAL;
1167 
1168 	if (!ret)
1169 		epp = cpudata->epp_default;
1170 	else
1171 		epp = epp_values[ret];
1172 
1173 	if (epp > 0 && policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
1174 		pr_debug("EPP cannot be set under performance policy\n");
1175 		return -EBUSY;
1176 	}
1177 
1178 	ret = amd_pstate_set_epp(policy, epp);
1179 
1180 	return ret ? ret : count;
1181 }
1182 
1183 static ssize_t show_energy_performance_preference(
1184 				struct cpufreq_policy *policy, char *buf)
1185 {
1186 	struct amd_cpudata *cpudata = policy->driver_data;
1187 	u8 preference, epp;
1188 
1189 	epp = FIELD_GET(AMD_CPPC_EPP_PERF_MASK, cpudata->cppc_req_cached);
1190 
1191 	switch (epp) {
1192 	case AMD_CPPC_EPP_PERFORMANCE:
1193 		preference = EPP_INDEX_PERFORMANCE;
1194 		break;
1195 	case AMD_CPPC_EPP_BALANCE_PERFORMANCE:
1196 		preference = EPP_INDEX_BALANCE_PERFORMANCE;
1197 		break;
1198 	case AMD_CPPC_EPP_BALANCE_POWERSAVE:
1199 		preference = EPP_INDEX_BALANCE_POWERSAVE;
1200 		break;
1201 	case AMD_CPPC_EPP_POWERSAVE:
1202 		preference = EPP_INDEX_POWERSAVE;
1203 		break;
1204 	default:
1205 		return -EINVAL;
1206 	}
1207 
1208 	return sysfs_emit(buf, "%s\n", energy_perf_strings[preference]);
1209 }
1210 
1211 static void amd_pstate_driver_cleanup(void)
1212 {
1213 	if (amd_pstate_prefcore)
1214 		sched_clear_itmt_support();
1215 
1216 	cppc_state = AMD_PSTATE_DISABLE;
1217 	current_pstate_driver = NULL;
1218 }
1219 
1220 static int amd_pstate_set_driver(int mode_idx)
1221 {
1222 	if (mode_idx >= AMD_PSTATE_DISABLE && mode_idx < AMD_PSTATE_MAX) {
1223 		cppc_state = mode_idx;
1224 		if (cppc_state == AMD_PSTATE_DISABLE)
1225 			pr_info("driver is explicitly disabled\n");
1226 
1227 		if (cppc_state == AMD_PSTATE_ACTIVE)
1228 			current_pstate_driver = &amd_pstate_epp_driver;
1229 
1230 		if (cppc_state == AMD_PSTATE_PASSIVE || cppc_state == AMD_PSTATE_GUIDED)
1231 			current_pstate_driver = &amd_pstate_driver;
1232 
1233 		return 0;
1234 	}
1235 
1236 	return -EINVAL;
1237 }
1238 
1239 static int amd_pstate_register_driver(int mode)
1240 {
1241 	int ret;
1242 
1243 	ret = amd_pstate_set_driver(mode);
1244 	if (ret)
1245 		return ret;
1246 
1247 	cppc_state = mode;
1248 
1249 	/* at least one CPU supports CPB */
1250 	current_pstate_driver->boost_enabled = cpu_feature_enabled(X86_FEATURE_CPB);
1251 
1252 	ret = cpufreq_register_driver(current_pstate_driver);
1253 	if (ret) {
1254 		amd_pstate_driver_cleanup();
1255 		return ret;
1256 	}
1257 
1258 	/* Enable ITMT support once all CPUs have initialized their asym priorities. */
1259 	if (amd_pstate_prefcore)
1260 		sched_set_itmt_support();
1261 
1262 	return 0;
1263 }
1264 
1265 static int amd_pstate_unregister_driver(int dummy)
1266 {
1267 	cpufreq_unregister_driver(current_pstate_driver);
1268 	amd_pstate_driver_cleanup();
1269 	return 0;
1270 }
1271 
1272 static int amd_pstate_change_mode_without_dvr_change(int mode)
1273 {
1274 	int cpu = 0;
1275 
1276 	cppc_state = mode;
1277 
1278 	if (cpu_feature_enabled(X86_FEATURE_CPPC) || cppc_state == AMD_PSTATE_ACTIVE)
1279 		return 0;
1280 
1281 	for_each_online_cpu(cpu) {
1282 		cppc_set_auto_sel(cpu, (cppc_state == AMD_PSTATE_PASSIVE) ? 0 : 1);
1283 	}
1284 
1285 	return 0;
1286 }
1287 
1288 static int amd_pstate_change_driver_mode(int mode)
1289 {
1290 	int ret;
1291 
1292 	ret = amd_pstate_unregister_driver(0);
1293 	if (ret)
1294 		return ret;
1295 
1296 	ret = amd_pstate_register_driver(mode);
1297 	if (ret)
1298 		return ret;
1299 
1300 	return 0;
1301 }
1302 
1303 static cppc_mode_transition_fn mode_state_machine[AMD_PSTATE_MAX][AMD_PSTATE_MAX] = {
1304 	[AMD_PSTATE_DISABLE]         = {
1305 		[AMD_PSTATE_DISABLE]     = NULL,
1306 		[AMD_PSTATE_PASSIVE]     = amd_pstate_register_driver,
1307 		[AMD_PSTATE_ACTIVE]      = amd_pstate_register_driver,
1308 		[AMD_PSTATE_GUIDED]      = amd_pstate_register_driver,
1309 	},
1310 	[AMD_PSTATE_PASSIVE]         = {
1311 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1312 		[AMD_PSTATE_PASSIVE]     = NULL,
1313 		[AMD_PSTATE_ACTIVE]      = amd_pstate_change_driver_mode,
1314 		[AMD_PSTATE_GUIDED]      = amd_pstate_change_mode_without_dvr_change,
1315 	},
1316 	[AMD_PSTATE_ACTIVE]          = {
1317 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1318 		[AMD_PSTATE_PASSIVE]     = amd_pstate_change_driver_mode,
1319 		[AMD_PSTATE_ACTIVE]      = NULL,
1320 		[AMD_PSTATE_GUIDED]      = amd_pstate_change_driver_mode,
1321 	},
1322 	[AMD_PSTATE_GUIDED]          = {
1323 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1324 		[AMD_PSTATE_PASSIVE]     = amd_pstate_change_mode_without_dvr_change,
1325 		[AMD_PSTATE_ACTIVE]      = amd_pstate_change_driver_mode,
1326 		[AMD_PSTATE_GUIDED]      = NULL,
1327 	},
1328 };
1329 
1330 static ssize_t amd_pstate_show_status(char *buf)
1331 {
1332 	if (!current_pstate_driver)
1333 		return sysfs_emit(buf, "disable\n");
1334 
1335 	return sysfs_emit(buf, "%s\n", amd_pstate_mode_string[cppc_state]);
1336 }
1337 
1338 int amd_pstate_get_status(void)
1339 {
1340 	return cppc_state;
1341 }
1342 EXPORT_SYMBOL_GPL(amd_pstate_get_status);
1343 
1344 int amd_pstate_update_status(const char *buf, size_t size)
1345 {
1346 	int mode_idx;
1347 
1348 	if (size > strlen("passive") || size < strlen("active"))
1349 		return -EINVAL;
1350 
1351 	mode_idx = get_mode_idx_from_str(buf, size);
1352 	if (mode_idx < 0)
1353 		return mode_idx;
1354 
1355 	if (mode_state_machine[cppc_state][mode_idx]) {
1356 		guard(mutex)(&amd_pstate_driver_lock);
1357 		return mode_state_machine[cppc_state][mode_idx](mode_idx);
1358 	}
1359 
1360 	return 0;
1361 }
1362 EXPORT_SYMBOL_GPL(amd_pstate_update_status);
1363 
1364 static ssize_t status_show(struct device *dev,
1365 			   struct device_attribute *attr, char *buf)
1366 {
1367 
1368 	guard(mutex)(&amd_pstate_driver_lock);
1369 
1370 	return amd_pstate_show_status(buf);
1371 }
1372 
1373 static ssize_t status_store(struct device *a, struct device_attribute *b,
1374 			    const char *buf, size_t count)
1375 {
1376 	char *p = memchr(buf, '\n', count);
1377 	int ret;
1378 
1379 	ret = amd_pstate_update_status(buf, p ? p - buf : count);
1380 
1381 	return ret < 0 ? ret : count;
1382 }
1383 
1384 static ssize_t prefcore_show(struct device *dev,
1385 			     struct device_attribute *attr, char *buf)
1386 {
1387 	return sysfs_emit(buf, "%s\n", str_enabled_disabled(amd_pstate_prefcore));
1388 }
1389 
1390 cpufreq_freq_attr_ro(amd_pstate_max_freq);
1391 cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
1392 
1393 cpufreq_freq_attr_ro(amd_pstate_highest_perf);
1394 cpufreq_freq_attr_ro(amd_pstate_prefcore_ranking);
1395 cpufreq_freq_attr_ro(amd_pstate_hw_prefcore);
1396 cpufreq_freq_attr_rw(energy_performance_preference);
1397 cpufreq_freq_attr_ro(energy_performance_available_preferences);
1398 static DEVICE_ATTR_RW(status);
1399 static DEVICE_ATTR_RO(prefcore);
1400 
1401 static struct freq_attr *amd_pstate_attr[] = {
1402 	&amd_pstate_max_freq,
1403 	&amd_pstate_lowest_nonlinear_freq,
1404 	&amd_pstate_highest_perf,
1405 	&amd_pstate_prefcore_ranking,
1406 	&amd_pstate_hw_prefcore,
1407 	NULL,
1408 };
1409 
1410 static struct freq_attr *amd_pstate_epp_attr[] = {
1411 	&amd_pstate_max_freq,
1412 	&amd_pstate_lowest_nonlinear_freq,
1413 	&amd_pstate_highest_perf,
1414 	&amd_pstate_prefcore_ranking,
1415 	&amd_pstate_hw_prefcore,
1416 	&energy_performance_preference,
1417 	&energy_performance_available_preferences,
1418 	NULL,
1419 };
1420 
1421 static struct attribute *pstate_global_attributes[] = {
1422 	&dev_attr_status.attr,
1423 	&dev_attr_prefcore.attr,
1424 	NULL
1425 };
1426 
1427 static const struct attribute_group amd_pstate_global_attr_group = {
1428 	.name = "amd_pstate",
1429 	.attrs = pstate_global_attributes,
1430 };
1431 
1432 static bool amd_pstate_acpi_pm_profile_server(void)
1433 {
1434 	switch (acpi_gbl_FADT.preferred_profile) {
1435 	case PM_ENTERPRISE_SERVER:
1436 	case PM_SOHO_SERVER:
1437 	case PM_PERFORMANCE_SERVER:
1438 		return true;
1439 	}
1440 	return false;
1441 }
1442 
1443 static bool amd_pstate_acpi_pm_profile_undefined(void)
1444 {
1445 	if (acpi_gbl_FADT.preferred_profile == PM_UNSPECIFIED)
1446 		return true;
1447 	if (acpi_gbl_FADT.preferred_profile >= NR_PM_PROFILES)
1448 		return true;
1449 	return false;
1450 }
1451 
1452 static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
1453 {
1454 	struct amd_cpudata *cpudata;
1455 	union perf_cached perf;
1456 	struct device *dev;
1457 	int ret;
1458 
1459 	/*
1460 	 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
1461 	 * which is ideal for initialization process.
1462 	 */
1463 	amd_perf_ctl_reset(policy->cpu);
1464 	dev = get_cpu_device(policy->cpu);
1465 	if (!dev)
1466 		return -ENODEV;
1467 
1468 	cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
1469 	if (!cpudata)
1470 		return -ENOMEM;
1471 
1472 	cpudata->cpu = policy->cpu;
1473 
1474 	ret = amd_pstate_init_perf(cpudata);
1475 	if (ret)
1476 		goto free_cpudata1;
1477 
1478 	amd_pstate_init_prefcore(cpudata);
1479 
1480 	ret = amd_pstate_init_freq(cpudata);
1481 	if (ret)
1482 		goto free_cpudata1;
1483 
1484 	ret = amd_pstate_init_boost_support(cpudata);
1485 	if (ret)
1486 		goto free_cpudata1;
1487 
1488 	perf = READ_ONCE(cpudata->perf);
1489 
1490 	policy->cpuinfo.min_freq = policy->min = perf_to_freq(perf,
1491 							      cpudata->nominal_freq,
1492 							      perf.lowest_perf);
1493 	policy->cpuinfo.max_freq = policy->max = perf_to_freq(perf,
1494 							      cpudata->nominal_freq,
1495 							      perf.highest_perf);
1496 	policy->driver_data = cpudata;
1497 
1498 	ret = amd_pstate_cppc_enable(policy);
1499 	if (ret)
1500 		goto free_cpudata1;
1501 
1502 	/* It will be updated by governor */
1503 	policy->cur = policy->cpuinfo.min_freq;
1504 
1505 
1506 	policy->boost_supported = READ_ONCE(cpudata->boost_supported);
1507 
1508 	/*
1509 	 * Set the policy to provide a valid fallback value in case
1510 	 * the default cpufreq governor is neither powersave nor performance.
1511 	 */
1512 	if (amd_pstate_acpi_pm_profile_server() ||
1513 	    amd_pstate_acpi_pm_profile_undefined()) {
1514 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
1515 		cpudata->epp_default = amd_pstate_get_epp(cpudata);
1516 	} else {
1517 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
1518 		cpudata->epp_default = AMD_CPPC_EPP_BALANCE_PERFORMANCE;
1519 	}
1520 
1521 	ret = amd_pstate_set_epp(policy, cpudata->epp_default);
1522 	if (ret)
1523 		return ret;
1524 
1525 	current_pstate_driver->adjust_perf = NULL;
1526 
1527 	return 0;
1528 
1529 free_cpudata1:
1530 	pr_warn("Failed to initialize CPU %d: %d\n", policy->cpu, ret);
1531 	kfree(cpudata);
1532 	return ret;
1533 }
1534 
1535 static void amd_pstate_epp_cpu_exit(struct cpufreq_policy *policy)
1536 {
1537 	struct amd_cpudata *cpudata = policy->driver_data;
1538 
1539 	if (cpudata) {
1540 		union perf_cached perf = READ_ONCE(cpudata->perf);
1541 
1542 		/* Reset CPPC_REQ MSR to the BIOS value */
1543 		amd_pstate_update_perf(policy, perf.bios_min_perf, 0U, 0U, 0U, false);
1544 
1545 		kfree(cpudata);
1546 		policy->driver_data = NULL;
1547 	}
1548 
1549 	pr_debug("CPU %d exiting\n", policy->cpu);
1550 }
1551 
1552 static int amd_pstate_epp_update_limit(struct cpufreq_policy *policy, bool policy_change)
1553 {
1554 	struct amd_cpudata *cpudata = policy->driver_data;
1555 	union perf_cached perf;
1556 	u8 epp;
1557 
1558 	if (policy_change ||
1559 	    policy->min != cpudata->min_limit_freq ||
1560 	    policy->max != cpudata->max_limit_freq)
1561 		amd_pstate_update_min_max_limit(policy);
1562 
1563 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1564 		epp = 0;
1565 	else
1566 		epp = FIELD_GET(AMD_CPPC_EPP_PERF_MASK, cpudata->cppc_req_cached);
1567 
1568 	perf = READ_ONCE(cpudata->perf);
1569 
1570 	return amd_pstate_update_perf(policy, perf.min_limit_perf, 0U,
1571 				      perf.max_limit_perf, epp, false);
1572 }
1573 
1574 static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy)
1575 {
1576 	struct amd_cpudata *cpudata = policy->driver_data;
1577 	int ret;
1578 
1579 	if (!policy->cpuinfo.max_freq)
1580 		return -ENODEV;
1581 
1582 	cpudata->policy = policy->policy;
1583 
1584 	ret = amd_pstate_epp_update_limit(policy, true);
1585 	if (ret)
1586 		return ret;
1587 
1588 	/*
1589 	 * policy->cur is never updated with the amd_pstate_epp driver, but it
1590 	 * is used as a stale frequency value. So, keep it within limits.
1591 	 */
1592 	policy->cur = policy->min;
1593 
1594 	return 0;
1595 }
1596 
1597 static int amd_pstate_cpu_online(struct cpufreq_policy *policy)
1598 {
1599 	return amd_pstate_cppc_enable(policy);
1600 }
1601 
1602 static int amd_pstate_cpu_offline(struct cpufreq_policy *policy)
1603 {
1604 	struct amd_cpudata *cpudata = policy->driver_data;
1605 	union perf_cached perf = READ_ONCE(cpudata->perf);
1606 
1607 	/*
1608 	 * Reset CPPC_REQ MSR to the BIOS value, this will allow us to retain the BIOS specified
1609 	 * min_perf value across kexec reboots. If this CPU is just onlined normally after this, the
1610 	 * limits, epp and desired perf will get reset to the cached values in cpudata struct
1611 	 */
1612 	return amd_pstate_update_perf(policy, perf.bios_min_perf,
1613 				     FIELD_GET(AMD_CPPC_DES_PERF_MASK, cpudata->cppc_req_cached),
1614 				     FIELD_GET(AMD_CPPC_MAX_PERF_MASK, cpudata->cppc_req_cached),
1615 				     FIELD_GET(AMD_CPPC_EPP_PERF_MASK, cpudata->cppc_req_cached),
1616 				     false);
1617 }
1618 
1619 static int amd_pstate_suspend(struct cpufreq_policy *policy)
1620 {
1621 	struct amd_cpudata *cpudata = policy->driver_data;
1622 	union perf_cached perf = READ_ONCE(cpudata->perf);
1623 	int ret;
1624 
1625 	/*
1626 	 * Reset CPPC_REQ MSR to the BIOS value, this will allow us to retain the BIOS specified
1627 	 * min_perf value across kexec reboots. If this CPU is just resumed back without kexec,
1628 	 * the limits, epp and desired perf will get reset to the cached values in cpudata struct
1629 	 */
1630 	ret = amd_pstate_update_perf(policy, perf.bios_min_perf,
1631 				     FIELD_GET(AMD_CPPC_DES_PERF_MASK, cpudata->cppc_req_cached),
1632 				     FIELD_GET(AMD_CPPC_MAX_PERF_MASK, cpudata->cppc_req_cached),
1633 				     FIELD_GET(AMD_CPPC_EPP_PERF_MASK, cpudata->cppc_req_cached),
1634 				     false);
1635 	if (ret)
1636 		return ret;
1637 
1638 	/* set this flag to avoid setting core offline*/
1639 	cpudata->suspended = true;
1640 
1641 	return 0;
1642 }
1643 
1644 static int amd_pstate_resume(struct cpufreq_policy *policy)
1645 {
1646 	struct amd_cpudata *cpudata = policy->driver_data;
1647 	union perf_cached perf = READ_ONCE(cpudata->perf);
1648 	int cur_perf = freq_to_perf(perf, cpudata->nominal_freq, policy->cur);
1649 
1650 	/* Set CPPC_REQ to last sane value until the governor updates it */
1651 	return amd_pstate_update_perf(policy, perf.min_limit_perf, cur_perf, perf.max_limit_perf,
1652 				      0U, false);
1653 }
1654 
1655 static int amd_pstate_epp_resume(struct cpufreq_policy *policy)
1656 {
1657 	struct amd_cpudata *cpudata = policy->driver_data;
1658 
1659 	if (cpudata->suspended) {
1660 		int ret;
1661 
1662 		/* enable amd pstate from suspend state*/
1663 		ret = amd_pstate_epp_update_limit(policy, false);
1664 		if (ret)
1665 			return ret;
1666 
1667 		cpudata->suspended = false;
1668 	}
1669 
1670 	return 0;
1671 }
1672 
1673 static struct cpufreq_driver amd_pstate_driver = {
1674 	.flags		= CPUFREQ_CONST_LOOPS | CPUFREQ_NEED_UPDATE_LIMITS,
1675 	.verify		= amd_pstate_verify,
1676 	.target		= amd_pstate_target,
1677 	.fast_switch    = amd_pstate_fast_switch,
1678 	.init		= amd_pstate_cpu_init,
1679 	.exit		= amd_pstate_cpu_exit,
1680 	.online		= amd_pstate_cpu_online,
1681 	.offline	= amd_pstate_cpu_offline,
1682 	.suspend	= amd_pstate_suspend,
1683 	.resume		= amd_pstate_resume,
1684 	.set_boost	= amd_pstate_set_boost,
1685 	.update_limits	= amd_pstate_update_limits,
1686 	.name		= "amd-pstate",
1687 	.attr		= amd_pstate_attr,
1688 };
1689 
1690 static struct cpufreq_driver amd_pstate_epp_driver = {
1691 	.flags		= CPUFREQ_CONST_LOOPS,
1692 	.verify		= amd_pstate_verify,
1693 	.setpolicy	= amd_pstate_epp_set_policy,
1694 	.init		= amd_pstate_epp_cpu_init,
1695 	.exit		= amd_pstate_epp_cpu_exit,
1696 	.offline	= amd_pstate_cpu_offline,
1697 	.online		= amd_pstate_cpu_online,
1698 	.suspend	= amd_pstate_suspend,
1699 	.resume		= amd_pstate_epp_resume,
1700 	.update_limits	= amd_pstate_update_limits,
1701 	.set_boost	= amd_pstate_set_boost,
1702 	.name		= "amd-pstate-epp",
1703 	.attr		= amd_pstate_epp_attr,
1704 };
1705 
1706 /*
1707  * CPPC function is not supported for family ID 17H with model_ID ranging from 0x10 to 0x2F.
1708  * show the debug message that helps to check if the CPU has CPPC support for loading issue.
1709  */
1710 static bool amd_cppc_supported(void)
1711 {
1712 	struct cpuinfo_x86 *c = &cpu_data(0);
1713 	bool warn = false;
1714 
1715 	if ((boot_cpu_data.x86 == 0x17) && (boot_cpu_data.x86_model < 0x30)) {
1716 		pr_debug_once("CPPC feature is not supported by the processor\n");
1717 		return false;
1718 	}
1719 
1720 	/*
1721 	 * If the CPPC feature is disabled in the BIOS for processors
1722 	 * that support MSR-based CPPC, the AMD Pstate driver may not
1723 	 * function correctly.
1724 	 *
1725 	 * For such processors, check the CPPC flag and display a
1726 	 * warning message if the platform supports CPPC.
1727 	 *
1728 	 * Note: The code check below will not abort the driver
1729 	 * registration process because of the code is added for
1730 	 * debugging purposes. Besides, it may still be possible for
1731 	 * the driver to work using the shared-memory mechanism.
1732 	 */
1733 	if (!cpu_feature_enabled(X86_FEATURE_CPPC)) {
1734 		if (cpu_feature_enabled(X86_FEATURE_ZEN2)) {
1735 			switch (c->x86_model) {
1736 			case 0x60 ... 0x6F:
1737 			case 0x80 ... 0xAF:
1738 				warn = true;
1739 				break;
1740 			}
1741 		} else if (cpu_feature_enabled(X86_FEATURE_ZEN3) ||
1742 			   cpu_feature_enabled(X86_FEATURE_ZEN4)) {
1743 			switch (c->x86_model) {
1744 			case 0x10 ... 0x1F:
1745 			case 0x40 ... 0xAF:
1746 				warn = true;
1747 				break;
1748 			}
1749 		} else if (cpu_feature_enabled(X86_FEATURE_ZEN5)) {
1750 			warn = true;
1751 		}
1752 	}
1753 
1754 	if (warn)
1755 		pr_warn_once("The CPPC feature is supported but currently disabled by the BIOS.\n"
1756 					"Please enable it if your BIOS has the CPPC option.\n");
1757 	return true;
1758 }
1759 
1760 static int __init amd_pstate_init(void)
1761 {
1762 	struct device *dev_root;
1763 	int ret;
1764 
1765 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
1766 		return -ENODEV;
1767 
1768 	/* show debug message only if CPPC is not supported */
1769 	if (!amd_cppc_supported())
1770 		return -EOPNOTSUPP;
1771 
1772 	/* show warning message when BIOS broken or ACPI disabled */
1773 	if (!acpi_cpc_valid()) {
1774 		pr_warn_once("the _CPC object is not present in SBIOS or ACPI disabled\n");
1775 		return -ENODEV;
1776 	}
1777 
1778 	/* don't keep reloading if cpufreq_driver exists */
1779 	if (cpufreq_get_current_driver())
1780 		return -EEXIST;
1781 
1782 	quirks = NULL;
1783 
1784 	/* check if this machine need CPPC quirks */
1785 	dmi_check_system(amd_pstate_quirks_table);
1786 
1787 	/*
1788 	* determine the driver mode from the command line or kernel config.
1789 	* If no command line input is provided, cppc_state will be AMD_PSTATE_UNDEFINED.
1790 	* command line options will override the kernel config settings.
1791 	*/
1792 
1793 	if (cppc_state == AMD_PSTATE_UNDEFINED) {
1794 		/* Disable on the following configs by default:
1795 		 * 1. Undefined platforms
1796 		 * 2. Server platforms with CPUs older than Family 0x1A.
1797 		 */
1798 		if (amd_pstate_acpi_pm_profile_undefined() ||
1799 		    (amd_pstate_acpi_pm_profile_server() && boot_cpu_data.x86 < 0x1A)) {
1800 			pr_info("driver load is disabled, boot with specific mode to enable this\n");
1801 			return -ENODEV;
1802 		}
1803 		/* get driver mode from kernel config option [1:4] */
1804 		cppc_state = CONFIG_X86_AMD_PSTATE_DEFAULT_MODE;
1805 	}
1806 
1807 	if (cppc_state == AMD_PSTATE_DISABLE) {
1808 		pr_info("driver load is disabled, boot with specific mode to enable this\n");
1809 		return -ENODEV;
1810 	}
1811 
1812 	/* capability check */
1813 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1814 		pr_debug("AMD CPPC MSR based functionality is supported\n");
1815 	} else {
1816 		pr_debug("AMD CPPC shared memory based functionality is supported\n");
1817 		static_call_update(amd_pstate_cppc_enable, shmem_cppc_enable);
1818 		static_call_update(amd_pstate_init_perf, shmem_init_perf);
1819 		static_call_update(amd_pstate_update_perf, shmem_update_perf);
1820 		static_call_update(amd_pstate_get_epp, shmem_get_epp);
1821 		static_call_update(amd_pstate_set_epp, shmem_set_epp);
1822 	}
1823 
1824 	if (amd_pstate_prefcore) {
1825 		ret = amd_detect_prefcore(&amd_pstate_prefcore);
1826 		if (ret)
1827 			return ret;
1828 	}
1829 
1830 	ret = amd_pstate_register_driver(cppc_state);
1831 	if (ret) {
1832 		pr_err("failed to register with return %d\n", ret);
1833 		return ret;
1834 	}
1835 
1836 	dev_root = bus_get_dev_root(&cpu_subsys);
1837 	if (dev_root) {
1838 		ret = sysfs_create_group(&dev_root->kobj, &amd_pstate_global_attr_group);
1839 		put_device(dev_root);
1840 		if (ret) {
1841 			pr_err("sysfs attribute export failed with error %d.\n", ret);
1842 			goto global_attr_free;
1843 		}
1844 	}
1845 
1846 	return ret;
1847 
1848 global_attr_free:
1849 	cpufreq_unregister_driver(current_pstate_driver);
1850 	return ret;
1851 }
1852 device_initcall(amd_pstate_init);
1853 
1854 static int __init amd_pstate_param(char *str)
1855 {
1856 	size_t size;
1857 	int mode_idx;
1858 
1859 	if (!str)
1860 		return -EINVAL;
1861 
1862 	size = strlen(str);
1863 	mode_idx = get_mode_idx_from_str(str, size);
1864 
1865 	return amd_pstate_set_driver(mode_idx);
1866 }
1867 
1868 static int __init amd_prefcore_param(char *str)
1869 {
1870 	if (!strcmp(str, "disable"))
1871 		amd_pstate_prefcore = false;
1872 
1873 	return 0;
1874 }
1875 
1876 early_param("amd_pstate", amd_pstate_param);
1877 early_param("amd_prefcore", amd_prefcore_param);
1878 
1879 MODULE_AUTHOR("Huang Rui <ray.huang@amd.com>");
1880 MODULE_DESCRIPTION("AMD Processor P-state Frequency Driver");
1881