xref: /linux/drivers/cpufreq/amd-pstate.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/kernel.h>
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/smp.h>
29 #include <linux/sched.h>
30 #include <linux/cpufreq.h>
31 #include <linux/compiler.h>
32 #include <linux/dmi.h>
33 #include <linux/slab.h>
34 #include <linux/acpi.h>
35 #include <linux/io.h>
36 #include <linux/delay.h>
37 #include <linux/uaccess.h>
38 #include <linux/static_call.h>
39 #include <linux/topology.h>
40 
41 #include <acpi/processor.h>
42 #include <acpi/cppc_acpi.h>
43 
44 #include <asm/msr.h>
45 #include <asm/processor.h>
46 #include <asm/cpufeature.h>
47 #include <asm/cpu_device_id.h>
48 
49 #include "amd-pstate.h"
50 #include "amd-pstate-trace.h"
51 
52 #define AMD_PSTATE_TRANSITION_LATENCY	20000
53 #define AMD_PSTATE_TRANSITION_DELAY	1000
54 #define AMD_PSTATE_FAST_CPPC_TRANSITION_DELAY 600
55 
56 #define AMD_CPPC_EPP_PERFORMANCE		0x00
57 #define AMD_CPPC_EPP_BALANCE_PERFORMANCE	0x80
58 #define AMD_CPPC_EPP_BALANCE_POWERSAVE		0xBF
59 #define AMD_CPPC_EPP_POWERSAVE			0xFF
60 
61 static const char * const amd_pstate_mode_string[] = {
62 	[AMD_PSTATE_UNDEFINED]   = "undefined",
63 	[AMD_PSTATE_DISABLE]     = "disable",
64 	[AMD_PSTATE_PASSIVE]     = "passive",
65 	[AMD_PSTATE_ACTIVE]      = "active",
66 	[AMD_PSTATE_GUIDED]      = "guided",
67 	NULL,
68 };
69 
70 const char *amd_pstate_get_mode_string(enum amd_pstate_mode mode)
71 {
72 	if (mode < 0 || mode >= AMD_PSTATE_MAX)
73 		return NULL;
74 	return amd_pstate_mode_string[mode];
75 }
76 EXPORT_SYMBOL_GPL(amd_pstate_get_mode_string);
77 
78 struct quirk_entry {
79 	u32 nominal_freq;
80 	u32 lowest_freq;
81 };
82 
83 static struct cpufreq_driver *current_pstate_driver;
84 static struct cpufreq_driver amd_pstate_driver;
85 static struct cpufreq_driver amd_pstate_epp_driver;
86 static int cppc_state = AMD_PSTATE_UNDEFINED;
87 static bool cppc_enabled;
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 };
114 
115 static const char * const energy_perf_strings[] = {
116 	[EPP_INDEX_DEFAULT] = "default",
117 	[EPP_INDEX_PERFORMANCE] = "performance",
118 	[EPP_INDEX_BALANCE_PERFORMANCE] = "balance_performance",
119 	[EPP_INDEX_BALANCE_POWERSAVE] = "balance_power",
120 	[EPP_INDEX_POWERSAVE] = "power",
121 	NULL
122 };
123 
124 static unsigned int epp_values[] = {
125 	[EPP_INDEX_DEFAULT] = 0,
126 	[EPP_INDEX_PERFORMANCE] = AMD_CPPC_EPP_PERFORMANCE,
127 	[EPP_INDEX_BALANCE_PERFORMANCE] = AMD_CPPC_EPP_BALANCE_PERFORMANCE,
128 	[EPP_INDEX_BALANCE_POWERSAVE] = AMD_CPPC_EPP_BALANCE_POWERSAVE,
129 	[EPP_INDEX_POWERSAVE] = AMD_CPPC_EPP_POWERSAVE,
130  };
131 
132 typedef int (*cppc_mode_transition_fn)(int);
133 
134 static struct quirk_entry quirk_amd_7k62 = {
135 	.nominal_freq = 2600,
136 	.lowest_freq = 550,
137 };
138 
139 static int __init dmi_matched_7k62_bios_bug(const struct dmi_system_id *dmi)
140 {
141 	/**
142 	 * match the broken bios for family 17h processor support CPPC V2
143 	 * broken BIOS lack of nominal_freq and lowest_freq capabilities
144 	 * definition in ACPI tables
145 	 */
146 	if (cpu_feature_enabled(X86_FEATURE_ZEN2)) {
147 		quirks = dmi->driver_data;
148 		pr_info("Overriding nominal and lowest frequencies for %s\n", dmi->ident);
149 		return 1;
150 	}
151 
152 	return 0;
153 }
154 
155 static const struct dmi_system_id amd_pstate_quirks_table[] __initconst = {
156 	{
157 		.callback = dmi_matched_7k62_bios_bug,
158 		.ident = "AMD EPYC 7K62",
159 		.matches = {
160 			DMI_MATCH(DMI_BIOS_VERSION, "5.14"),
161 			DMI_MATCH(DMI_BIOS_RELEASE, "12/12/2019"),
162 		},
163 		.driver_data = &quirk_amd_7k62,
164 	},
165 	{}
166 };
167 MODULE_DEVICE_TABLE(dmi, amd_pstate_quirks_table);
168 
169 static inline int get_mode_idx_from_str(const char *str, size_t size)
170 {
171 	int i;
172 
173 	for (i=0; i < AMD_PSTATE_MAX; i++) {
174 		if (!strncmp(str, amd_pstate_mode_string[i], size))
175 			return i;
176 	}
177 	return -EINVAL;
178 }
179 
180 static DEFINE_MUTEX(amd_pstate_limits_lock);
181 static DEFINE_MUTEX(amd_pstate_driver_lock);
182 
183 static s16 amd_pstate_get_epp(struct amd_cpudata *cpudata, u64 cppc_req_cached)
184 {
185 	u64 epp;
186 	int ret;
187 
188 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
189 		if (!cppc_req_cached) {
190 			epp = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ,
191 					&cppc_req_cached);
192 			if (epp)
193 				return epp;
194 		}
195 		epp = (cppc_req_cached >> 24) & 0xFF;
196 	} else {
197 		ret = cppc_get_epp_perf(cpudata->cpu, &epp);
198 		if (ret < 0) {
199 			pr_debug("Could not retrieve energy perf value (%d)\n", ret);
200 			return -EIO;
201 		}
202 	}
203 
204 	return (s16)(epp & 0xff);
205 }
206 
207 static int amd_pstate_get_energy_pref_index(struct amd_cpudata *cpudata)
208 {
209 	s16 epp;
210 	int index = -EINVAL;
211 
212 	epp = amd_pstate_get_epp(cpudata, 0);
213 	if (epp < 0)
214 		return epp;
215 
216 	switch (epp) {
217 	case AMD_CPPC_EPP_PERFORMANCE:
218 		index = EPP_INDEX_PERFORMANCE;
219 		break;
220 	case AMD_CPPC_EPP_BALANCE_PERFORMANCE:
221 		index = EPP_INDEX_BALANCE_PERFORMANCE;
222 		break;
223 	case AMD_CPPC_EPP_BALANCE_POWERSAVE:
224 		index = EPP_INDEX_BALANCE_POWERSAVE;
225 		break;
226 	case AMD_CPPC_EPP_POWERSAVE:
227 		index = EPP_INDEX_POWERSAVE;
228 		break;
229 	default:
230 		break;
231 	}
232 
233 	return index;
234 }
235 
236 static void pstate_update_perf(struct amd_cpudata *cpudata, u32 min_perf,
237 			       u32 des_perf, u32 max_perf, bool fast_switch)
238 {
239 	if (fast_switch)
240 		wrmsrl(MSR_AMD_CPPC_REQ, READ_ONCE(cpudata->cppc_req_cached));
241 	else
242 		wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ,
243 			      READ_ONCE(cpudata->cppc_req_cached));
244 }
245 
246 DEFINE_STATIC_CALL(amd_pstate_update_perf, pstate_update_perf);
247 
248 static inline void amd_pstate_update_perf(struct amd_cpudata *cpudata,
249 					  u32 min_perf, u32 des_perf,
250 					  u32 max_perf, bool fast_switch)
251 {
252 	static_call(amd_pstate_update_perf)(cpudata, min_perf, des_perf,
253 					    max_perf, fast_switch);
254 }
255 
256 static int amd_pstate_set_epp(struct amd_cpudata *cpudata, u32 epp)
257 {
258 	int ret;
259 	struct cppc_perf_ctrls perf_ctrls;
260 
261 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
262 		u64 value = READ_ONCE(cpudata->cppc_req_cached);
263 
264 		value &= ~GENMASK_ULL(31, 24);
265 		value |= (u64)epp << 24;
266 		WRITE_ONCE(cpudata->cppc_req_cached, value);
267 
268 		ret = wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
269 		if (!ret)
270 			cpudata->epp_cached = epp;
271 	} else {
272 		amd_pstate_update_perf(cpudata, cpudata->min_limit_perf, 0U,
273 					     cpudata->max_limit_perf, false);
274 
275 		perf_ctrls.energy_perf = epp;
276 		ret = cppc_set_epp_perf(cpudata->cpu, &perf_ctrls, 1);
277 		if (ret) {
278 			pr_debug("failed to set energy perf value (%d)\n", ret);
279 			return ret;
280 		}
281 		cpudata->epp_cached = epp;
282 	}
283 
284 	return ret;
285 }
286 
287 static int amd_pstate_set_energy_pref_index(struct amd_cpudata *cpudata,
288 		int pref_index)
289 {
290 	int epp = -EINVAL;
291 	int ret;
292 
293 	if (!pref_index)
294 		epp = cpudata->epp_default;
295 
296 	if (epp == -EINVAL)
297 		epp = epp_values[pref_index];
298 
299 	if (epp > 0 && cpudata->policy == CPUFREQ_POLICY_PERFORMANCE) {
300 		pr_debug("EPP cannot be set under performance policy\n");
301 		return -EBUSY;
302 	}
303 
304 	ret = amd_pstate_set_epp(cpudata, epp);
305 
306 	return ret;
307 }
308 
309 static inline int pstate_enable(bool enable)
310 {
311 	int ret, cpu;
312 	unsigned long logical_proc_id_mask = 0;
313 
314 	if (enable == cppc_enabled)
315 		return 0;
316 
317 	for_each_present_cpu(cpu) {
318 		unsigned long logical_id = topology_logical_package_id(cpu);
319 
320 		if (test_bit(logical_id, &logical_proc_id_mask))
321 			continue;
322 
323 		set_bit(logical_id, &logical_proc_id_mask);
324 
325 		ret = wrmsrl_safe_on_cpu(cpu, MSR_AMD_CPPC_ENABLE,
326 				enable);
327 		if (ret)
328 			return ret;
329 	}
330 
331 	cppc_enabled = enable;
332 	return 0;
333 }
334 
335 static int cppc_enable(bool enable)
336 {
337 	int cpu, ret = 0;
338 	struct cppc_perf_ctrls perf_ctrls;
339 
340 	if (enable == cppc_enabled)
341 		return 0;
342 
343 	for_each_present_cpu(cpu) {
344 		ret = cppc_set_enable(cpu, enable);
345 		if (ret)
346 			return ret;
347 
348 		/* Enable autonomous mode for EPP */
349 		if (cppc_state == AMD_PSTATE_ACTIVE) {
350 			/* Set desired perf as zero to allow EPP firmware control */
351 			perf_ctrls.desired_perf = 0;
352 			ret = cppc_set_perf(cpu, &perf_ctrls);
353 			if (ret)
354 				return ret;
355 		}
356 	}
357 
358 	cppc_enabled = enable;
359 	return ret;
360 }
361 
362 DEFINE_STATIC_CALL(amd_pstate_enable, pstate_enable);
363 
364 static inline int amd_pstate_enable(bool enable)
365 {
366 	return static_call(amd_pstate_enable)(enable);
367 }
368 
369 static int pstate_init_perf(struct amd_cpudata *cpudata)
370 {
371 	u64 cap1;
372 
373 	int ret = rdmsrl_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
374 				     &cap1);
375 	if (ret)
376 		return ret;
377 
378 	WRITE_ONCE(cpudata->highest_perf, AMD_CPPC_HIGHEST_PERF(cap1));
379 	WRITE_ONCE(cpudata->max_limit_perf, AMD_CPPC_HIGHEST_PERF(cap1));
380 	WRITE_ONCE(cpudata->nominal_perf, AMD_CPPC_NOMINAL_PERF(cap1));
381 	WRITE_ONCE(cpudata->lowest_nonlinear_perf, AMD_CPPC_LOWNONLIN_PERF(cap1));
382 	WRITE_ONCE(cpudata->lowest_perf, AMD_CPPC_LOWEST_PERF(cap1));
383 	WRITE_ONCE(cpudata->prefcore_ranking, AMD_CPPC_HIGHEST_PERF(cap1));
384 	WRITE_ONCE(cpudata->min_limit_perf, AMD_CPPC_LOWEST_PERF(cap1));
385 	return 0;
386 }
387 
388 static int cppc_init_perf(struct amd_cpudata *cpudata)
389 {
390 	struct cppc_perf_caps cppc_perf;
391 
392 	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
393 	if (ret)
394 		return ret;
395 
396 	WRITE_ONCE(cpudata->highest_perf, cppc_perf.highest_perf);
397 	WRITE_ONCE(cpudata->max_limit_perf, cppc_perf.highest_perf);
398 	WRITE_ONCE(cpudata->nominal_perf, cppc_perf.nominal_perf);
399 	WRITE_ONCE(cpudata->lowest_nonlinear_perf,
400 		   cppc_perf.lowest_nonlinear_perf);
401 	WRITE_ONCE(cpudata->lowest_perf, cppc_perf.lowest_perf);
402 	WRITE_ONCE(cpudata->prefcore_ranking, cppc_perf.highest_perf);
403 	WRITE_ONCE(cpudata->min_limit_perf, cppc_perf.lowest_perf);
404 
405 	if (cppc_state == AMD_PSTATE_ACTIVE)
406 		return 0;
407 
408 	ret = cppc_get_auto_sel_caps(cpudata->cpu, &cppc_perf);
409 	if (ret) {
410 		pr_warn("failed to get auto_sel, ret: %d\n", ret);
411 		return 0;
412 	}
413 
414 	ret = cppc_set_auto_sel(cpudata->cpu,
415 			(cppc_state == AMD_PSTATE_PASSIVE) ? 0 : 1);
416 
417 	if (ret)
418 		pr_warn("failed to set auto_sel, ret: %d\n", ret);
419 
420 	return ret;
421 }
422 
423 DEFINE_STATIC_CALL(amd_pstate_init_perf, pstate_init_perf);
424 
425 static inline int amd_pstate_init_perf(struct amd_cpudata *cpudata)
426 {
427 	return static_call(amd_pstate_init_perf)(cpudata);
428 }
429 
430 static void cppc_update_perf(struct amd_cpudata *cpudata,
431 			     u32 min_perf, u32 des_perf,
432 			     u32 max_perf, bool fast_switch)
433 {
434 	struct cppc_perf_ctrls perf_ctrls;
435 
436 	perf_ctrls.max_perf = max_perf;
437 	perf_ctrls.min_perf = min_perf;
438 	perf_ctrls.desired_perf = des_perf;
439 
440 	cppc_set_perf(cpudata->cpu, &perf_ctrls);
441 }
442 
443 static inline bool amd_pstate_sample(struct amd_cpudata *cpudata)
444 {
445 	u64 aperf, mperf, tsc;
446 	unsigned long flags;
447 
448 	local_irq_save(flags);
449 	rdmsrl(MSR_IA32_APERF, aperf);
450 	rdmsrl(MSR_IA32_MPERF, mperf);
451 	tsc = rdtsc();
452 
453 	if (cpudata->prev.mperf == mperf || cpudata->prev.tsc == tsc) {
454 		local_irq_restore(flags);
455 		return false;
456 	}
457 
458 	local_irq_restore(flags);
459 
460 	cpudata->cur.aperf = aperf;
461 	cpudata->cur.mperf = mperf;
462 	cpudata->cur.tsc =  tsc;
463 	cpudata->cur.aperf -= cpudata->prev.aperf;
464 	cpudata->cur.mperf -= cpudata->prev.mperf;
465 	cpudata->cur.tsc -= cpudata->prev.tsc;
466 
467 	cpudata->prev.aperf = aperf;
468 	cpudata->prev.mperf = mperf;
469 	cpudata->prev.tsc = tsc;
470 
471 	cpudata->freq = div64_u64((cpudata->cur.aperf * cpu_khz), cpudata->cur.mperf);
472 
473 	return true;
474 }
475 
476 static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
477 			      u32 des_perf, u32 max_perf, bool fast_switch, int gov_flags)
478 {
479 	unsigned long max_freq;
480 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpudata->cpu);
481 	u64 prev = READ_ONCE(cpudata->cppc_req_cached);
482 	u32 nominal_perf = READ_ONCE(cpudata->nominal_perf);
483 	u64 value = prev;
484 
485 	min_perf = clamp_t(unsigned long, min_perf, cpudata->min_limit_perf,
486 			cpudata->max_limit_perf);
487 	max_perf = clamp_t(unsigned long, max_perf, cpudata->min_limit_perf,
488 			cpudata->max_limit_perf);
489 	des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
490 
491 	max_freq = READ_ONCE(cpudata->max_limit_freq);
492 	policy->cur = div_u64(des_perf * max_freq, max_perf);
493 
494 	if ((cppc_state == AMD_PSTATE_GUIDED) && (gov_flags & CPUFREQ_GOV_DYNAMIC_SWITCHING)) {
495 		min_perf = des_perf;
496 		des_perf = 0;
497 	}
498 
499 	value &= ~AMD_CPPC_MIN_PERF(~0L);
500 	value |= AMD_CPPC_MIN_PERF(min_perf);
501 
502 	value &= ~AMD_CPPC_DES_PERF(~0L);
503 	value |= AMD_CPPC_DES_PERF(des_perf);
504 
505 	/* limit the max perf when core performance boost feature is disabled */
506 	if (!cpudata->boost_supported)
507 		max_perf = min_t(unsigned long, nominal_perf, max_perf);
508 
509 	value &= ~AMD_CPPC_MAX_PERF(~0L);
510 	value |= AMD_CPPC_MAX_PERF(max_perf);
511 
512 	if (trace_amd_pstate_perf_enabled() && amd_pstate_sample(cpudata)) {
513 		trace_amd_pstate_perf(min_perf, des_perf, max_perf, cpudata->freq,
514 			cpudata->cur.mperf, cpudata->cur.aperf, cpudata->cur.tsc,
515 				cpudata->cpu, (value != prev), fast_switch);
516 	}
517 
518 	if (value == prev)
519 		goto cpufreq_policy_put;
520 
521 	WRITE_ONCE(cpudata->cppc_req_cached, value);
522 
523 	amd_pstate_update_perf(cpudata, min_perf, des_perf,
524 			       max_perf, fast_switch);
525 
526 cpufreq_policy_put:
527 	cpufreq_cpu_put(policy);
528 }
529 
530 static int amd_pstate_verify(struct cpufreq_policy_data *policy)
531 {
532 	cpufreq_verify_within_cpu_limits(policy);
533 
534 	return 0;
535 }
536 
537 static int amd_pstate_update_min_max_limit(struct cpufreq_policy *policy)
538 {
539 	u32 max_limit_perf, min_limit_perf, lowest_perf, max_perf;
540 	struct amd_cpudata *cpudata = policy->driver_data;
541 
542 	if (cpudata->boost_supported && !policy->boost_enabled)
543 		max_perf = READ_ONCE(cpudata->nominal_perf);
544 	else
545 		max_perf = READ_ONCE(cpudata->highest_perf);
546 
547 	max_limit_perf = div_u64(policy->max * max_perf, policy->cpuinfo.max_freq);
548 	min_limit_perf = div_u64(policy->min * max_perf, policy->cpuinfo.max_freq);
549 
550 	lowest_perf = READ_ONCE(cpudata->lowest_perf);
551 	if (min_limit_perf < lowest_perf)
552 		min_limit_perf = lowest_perf;
553 
554 	if (max_limit_perf < min_limit_perf)
555 		max_limit_perf = min_limit_perf;
556 
557 	WRITE_ONCE(cpudata->max_limit_perf, max_limit_perf);
558 	WRITE_ONCE(cpudata->min_limit_perf, min_limit_perf);
559 	WRITE_ONCE(cpudata->max_limit_freq, policy->max);
560 	WRITE_ONCE(cpudata->min_limit_freq, policy->min);
561 
562 	return 0;
563 }
564 
565 static int amd_pstate_update_freq(struct cpufreq_policy *policy,
566 				  unsigned int target_freq, bool fast_switch)
567 {
568 	struct cpufreq_freqs freqs;
569 	struct amd_cpudata *cpudata = policy->driver_data;
570 	unsigned long max_perf, min_perf, des_perf, cap_perf;
571 
572 	if (!cpudata->max_freq)
573 		return -ENODEV;
574 
575 	if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)
576 		amd_pstate_update_min_max_limit(policy);
577 
578 	cap_perf = READ_ONCE(cpudata->highest_perf);
579 	min_perf = READ_ONCE(cpudata->lowest_perf);
580 	max_perf = cap_perf;
581 
582 	freqs.old = policy->cur;
583 	freqs.new = target_freq;
584 
585 	des_perf = DIV_ROUND_CLOSEST(target_freq * cap_perf,
586 				     cpudata->max_freq);
587 
588 	WARN_ON(fast_switch && !policy->fast_switch_enabled);
589 	/*
590 	 * If fast_switch is desired, then there aren't any registered
591 	 * transition notifiers. See comment for
592 	 * cpufreq_enable_fast_switch().
593 	 */
594 	if (!fast_switch)
595 		cpufreq_freq_transition_begin(policy, &freqs);
596 
597 	amd_pstate_update(cpudata, min_perf, des_perf,
598 			max_perf, fast_switch, policy->governor->flags);
599 
600 	if (!fast_switch)
601 		cpufreq_freq_transition_end(policy, &freqs, false);
602 
603 	return 0;
604 }
605 
606 static int amd_pstate_target(struct cpufreq_policy *policy,
607 			     unsigned int target_freq,
608 			     unsigned int relation)
609 {
610 	return amd_pstate_update_freq(policy, target_freq, false);
611 }
612 
613 static unsigned int amd_pstate_fast_switch(struct cpufreq_policy *policy,
614 				  unsigned int target_freq)
615 {
616 	if (!amd_pstate_update_freq(policy, target_freq, true))
617 		return target_freq;
618 	return policy->cur;
619 }
620 
621 static void amd_pstate_adjust_perf(unsigned int cpu,
622 				   unsigned long _min_perf,
623 				   unsigned long target_perf,
624 				   unsigned long capacity)
625 {
626 	unsigned long max_perf, min_perf, des_perf,
627 		      cap_perf, lowest_nonlinear_perf;
628 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
629 	struct amd_cpudata *cpudata;
630 
631 	if (!policy)
632 		return;
633 
634 	cpudata = policy->driver_data;
635 
636 	if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq)
637 		amd_pstate_update_min_max_limit(policy);
638 
639 
640 	cap_perf = READ_ONCE(cpudata->highest_perf);
641 	lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf);
642 
643 	des_perf = cap_perf;
644 	if (target_perf < capacity)
645 		des_perf = DIV_ROUND_UP(cap_perf * target_perf, capacity);
646 
647 	min_perf = READ_ONCE(cpudata->lowest_perf);
648 	if (_min_perf < capacity)
649 		min_perf = DIV_ROUND_UP(cap_perf * _min_perf, capacity);
650 
651 	if (min_perf < lowest_nonlinear_perf)
652 		min_perf = lowest_nonlinear_perf;
653 
654 	max_perf = cap_perf;
655 	if (max_perf < min_perf)
656 		max_perf = min_perf;
657 
658 	des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
659 
660 	amd_pstate_update(cpudata, min_perf, des_perf, max_perf, true,
661 			policy->governor->flags);
662 	cpufreq_cpu_put(policy);
663 }
664 
665 static int amd_pstate_cpu_boost_update(struct cpufreq_policy *policy, bool on)
666 {
667 	struct amd_cpudata *cpudata = policy->driver_data;
668 	struct cppc_perf_ctrls perf_ctrls;
669 	u32 highest_perf, nominal_perf, nominal_freq, max_freq;
670 	int ret = 0;
671 
672 	highest_perf = READ_ONCE(cpudata->highest_perf);
673 	nominal_perf = READ_ONCE(cpudata->nominal_perf);
674 	nominal_freq = READ_ONCE(cpudata->nominal_freq);
675 	max_freq = READ_ONCE(cpudata->max_freq);
676 
677 	if (boot_cpu_has(X86_FEATURE_CPPC)) {
678 		u64 value = READ_ONCE(cpudata->cppc_req_cached);
679 
680 		value &= ~GENMASK_ULL(7, 0);
681 		value |= on ? highest_perf : nominal_perf;
682 		WRITE_ONCE(cpudata->cppc_req_cached, value);
683 
684 		wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
685 	} else {
686 		perf_ctrls.max_perf = on ? highest_perf : nominal_perf;
687 		ret = cppc_set_perf(cpudata->cpu, &perf_ctrls);
688 		if (ret) {
689 			cpufreq_cpu_release(policy);
690 			pr_debug("Failed to set max perf on CPU:%d. ret:%d\n",
691 				cpudata->cpu, ret);
692 			return ret;
693 		}
694 	}
695 
696 	if (on)
697 		policy->cpuinfo.max_freq = max_freq;
698 	else if (policy->cpuinfo.max_freq > nominal_freq * 1000)
699 		policy->cpuinfo.max_freq = nominal_freq * 1000;
700 
701 	policy->max = policy->cpuinfo.max_freq;
702 
703 	if (cppc_state == AMD_PSTATE_PASSIVE) {
704 		ret = freq_qos_update_request(&cpudata->req[1], policy->cpuinfo.max_freq);
705 		if (ret < 0)
706 			pr_debug("Failed to update freq constraint: CPU%d\n", cpudata->cpu);
707 	}
708 
709 	return ret < 0 ? ret : 0;
710 }
711 
712 static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
713 {
714 	struct amd_cpudata *cpudata = policy->driver_data;
715 	int ret;
716 
717 	if (!cpudata->boost_supported) {
718 		pr_err("Boost mode is not supported by this processor or SBIOS\n");
719 		return -EOPNOTSUPP;
720 	}
721 	mutex_lock(&amd_pstate_driver_lock);
722 	ret = amd_pstate_cpu_boost_update(policy, state);
723 	WRITE_ONCE(cpudata->boost_state, !ret ? state : false);
724 	policy->boost_enabled = !ret ? state : false;
725 	refresh_frequency_limits(policy);
726 	mutex_unlock(&amd_pstate_driver_lock);
727 
728 	return ret;
729 }
730 
731 static int amd_pstate_init_boost_support(struct amd_cpudata *cpudata)
732 {
733 	u64 boost_val;
734 	int ret = -1;
735 
736 	/*
737 	 * If platform has no CPB support or disable it, initialize current driver
738 	 * boost_enabled state to be false, it is not an error for cpufreq core to handle.
739 	 */
740 	if (!cpu_feature_enabled(X86_FEATURE_CPB)) {
741 		pr_debug_once("Boost CPB capabilities not present in the processor\n");
742 		ret = 0;
743 		goto exit_err;
744 	}
745 
746 	/* at least one CPU supports CPB, even if others fail later on to set up */
747 	current_pstate_driver->boost_enabled = true;
748 
749 	ret = rdmsrl_on_cpu(cpudata->cpu, MSR_K7_HWCR, &boost_val);
750 	if (ret) {
751 		pr_err_once("failed to read initial CPU boost state!\n");
752 		ret = -EIO;
753 		goto exit_err;
754 	}
755 
756 	if (!(boost_val & MSR_K7_HWCR_CPB_DIS))
757 		cpudata->boost_supported = true;
758 
759 	return 0;
760 
761 exit_err:
762 	cpudata->boost_supported = false;
763 	return ret;
764 }
765 
766 static void amd_perf_ctl_reset(unsigned int cpu)
767 {
768 	wrmsrl_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
769 }
770 
771 /*
772  * Set amd-pstate preferred core enable can't be done directly from cpufreq callbacks
773  * due to locking, so queue the work for later.
774  */
775 static void amd_pstste_sched_prefcore_workfn(struct work_struct *work)
776 {
777 	sched_set_itmt_support();
778 }
779 static DECLARE_WORK(sched_prefcore_work, amd_pstste_sched_prefcore_workfn);
780 
781 #define CPPC_MAX_PERF	U8_MAX
782 
783 static void amd_pstate_init_prefcore(struct amd_cpudata *cpudata)
784 {
785 	/* user disabled or not detected */
786 	if (!amd_pstate_prefcore)
787 		return;
788 
789 	cpudata->hw_prefcore = true;
790 
791 	/*
792 	 * The priorities can be set regardless of whether or not
793 	 * sched_set_itmt_support(true) has been called and it is valid to
794 	 * update them at any time after it has been called.
795 	 */
796 	sched_set_itmt_core_prio((int)READ_ONCE(cpudata->highest_perf), cpudata->cpu);
797 
798 	schedule_work(&sched_prefcore_work);
799 }
800 
801 static void amd_pstate_update_limits(unsigned int cpu)
802 {
803 	struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
804 	struct amd_cpudata *cpudata;
805 	u32 prev_high = 0, cur_high = 0;
806 	int ret;
807 	bool highest_perf_changed = false;
808 
809 	if (!policy)
810 		return;
811 
812 	cpudata = policy->driver_data;
813 
814 	if (!amd_pstate_prefcore)
815 		return;
816 
817 	mutex_lock(&amd_pstate_driver_lock);
818 	ret = amd_get_highest_perf(cpu, &cur_high);
819 	if (ret)
820 		goto free_cpufreq_put;
821 
822 	prev_high = READ_ONCE(cpudata->prefcore_ranking);
823 	highest_perf_changed = (prev_high != cur_high);
824 	if (highest_perf_changed) {
825 		WRITE_ONCE(cpudata->prefcore_ranking, cur_high);
826 
827 		if (cur_high < CPPC_MAX_PERF)
828 			sched_set_itmt_core_prio((int)cur_high, cpu);
829 	}
830 
831 free_cpufreq_put:
832 	cpufreq_cpu_put(policy);
833 
834 	if (!highest_perf_changed)
835 		cpufreq_update_policy(cpu);
836 
837 	mutex_unlock(&amd_pstate_driver_lock);
838 }
839 
840 /*
841  * Get pstate transition delay time from ACPI tables that firmware set
842  * instead of using hardcode value directly.
843  */
844 static u32 amd_pstate_get_transition_delay_us(unsigned int cpu)
845 {
846 	u32 transition_delay_ns;
847 
848 	transition_delay_ns = cppc_get_transition_latency(cpu);
849 	if (transition_delay_ns == CPUFREQ_ETERNAL) {
850 		if (cpu_feature_enabled(X86_FEATURE_FAST_CPPC))
851 			return AMD_PSTATE_FAST_CPPC_TRANSITION_DELAY;
852 		else
853 			return AMD_PSTATE_TRANSITION_DELAY;
854 	}
855 
856 	return transition_delay_ns / NSEC_PER_USEC;
857 }
858 
859 /*
860  * Get pstate transition latency value from ACPI tables that firmware
861  * set instead of using hardcode value directly.
862  */
863 static u32 amd_pstate_get_transition_latency(unsigned int cpu)
864 {
865 	u32 transition_latency;
866 
867 	transition_latency = cppc_get_transition_latency(cpu);
868 	if (transition_latency  == CPUFREQ_ETERNAL)
869 		return AMD_PSTATE_TRANSITION_LATENCY;
870 
871 	return transition_latency;
872 }
873 
874 /*
875  * amd_pstate_init_freq: Initialize the max_freq, min_freq,
876  *                       nominal_freq and lowest_nonlinear_freq for
877  *                       the @cpudata object.
878  *
879  *  Requires: highest_perf, lowest_perf, nominal_perf and
880  *            lowest_nonlinear_perf members of @cpudata to be
881  *            initialized.
882  *
883  *  Returns 0 on success, non-zero value on failure.
884  */
885 static int amd_pstate_init_freq(struct amd_cpudata *cpudata)
886 {
887 	int ret;
888 	u32 min_freq, max_freq;
889 	u64 numerator;
890 	u32 nominal_perf, nominal_freq;
891 	u32 lowest_nonlinear_perf, lowest_nonlinear_freq;
892 	u32 boost_ratio, lowest_nonlinear_ratio;
893 	struct cppc_perf_caps cppc_perf;
894 
895 	ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
896 	if (ret)
897 		return ret;
898 
899 	if (quirks && quirks->lowest_freq)
900 		min_freq = quirks->lowest_freq * 1000;
901 	else
902 		min_freq = cppc_perf.lowest_freq * 1000;
903 
904 	if (quirks && quirks->nominal_freq)
905 		nominal_freq = quirks->nominal_freq ;
906 	else
907 		nominal_freq = cppc_perf.nominal_freq;
908 
909 	nominal_perf = READ_ONCE(cpudata->nominal_perf);
910 
911 	ret = amd_get_boost_ratio_numerator(cpudata->cpu, &numerator);
912 	if (ret)
913 		return ret;
914 	boost_ratio = div_u64(numerator << SCHED_CAPACITY_SHIFT, nominal_perf);
915 	max_freq = (nominal_freq * boost_ratio >> SCHED_CAPACITY_SHIFT) * 1000;
916 
917 	lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf);
918 	lowest_nonlinear_ratio = div_u64(lowest_nonlinear_perf << SCHED_CAPACITY_SHIFT,
919 					 nominal_perf);
920 	lowest_nonlinear_freq = (nominal_freq * lowest_nonlinear_ratio >> SCHED_CAPACITY_SHIFT) * 1000;
921 
922 	WRITE_ONCE(cpudata->min_freq, min_freq);
923 	WRITE_ONCE(cpudata->lowest_nonlinear_freq, lowest_nonlinear_freq);
924 	WRITE_ONCE(cpudata->nominal_freq, nominal_freq);
925 	WRITE_ONCE(cpudata->max_freq, max_freq);
926 
927 	/**
928 	 * Below values need to be initialized correctly, otherwise driver will fail to load
929 	 * max_freq is calculated according to (nominal_freq * highest_perf)/nominal_perf
930 	 * lowest_nonlinear_freq is a value between [min_freq, nominal_freq]
931 	 * Check _CPC in ACPI table objects if any values are incorrect
932 	 */
933 	if (min_freq <= 0 || max_freq <= 0 || nominal_freq <= 0 || min_freq > max_freq) {
934 		pr_err("min_freq(%d) or max_freq(%d) or nominal_freq(%d) value is incorrect\n",
935 			min_freq, max_freq, nominal_freq * 1000);
936 		return -EINVAL;
937 	}
938 
939 	if (lowest_nonlinear_freq <= min_freq || lowest_nonlinear_freq > nominal_freq * 1000) {
940 		pr_err("lowest_nonlinear_freq(%d) value is out of range [min_freq(%d), nominal_freq(%d)]\n",
941 			lowest_nonlinear_freq, min_freq, nominal_freq * 1000);
942 		return -EINVAL;
943 	}
944 
945 	return 0;
946 }
947 
948 static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
949 {
950 	int min_freq, max_freq, ret;
951 	struct device *dev;
952 	struct amd_cpudata *cpudata;
953 
954 	/*
955 	 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
956 	 * which is ideal for initialization process.
957 	 */
958 	amd_perf_ctl_reset(policy->cpu);
959 	dev = get_cpu_device(policy->cpu);
960 	if (!dev)
961 		return -ENODEV;
962 
963 	cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
964 	if (!cpudata)
965 		return -ENOMEM;
966 
967 	cpudata->cpu = policy->cpu;
968 
969 	ret = amd_pstate_init_perf(cpudata);
970 	if (ret)
971 		goto free_cpudata1;
972 
973 	amd_pstate_init_prefcore(cpudata);
974 
975 	ret = amd_pstate_init_freq(cpudata);
976 	if (ret)
977 		goto free_cpudata1;
978 
979 	ret = amd_pstate_init_boost_support(cpudata);
980 	if (ret)
981 		goto free_cpudata1;
982 
983 	min_freq = READ_ONCE(cpudata->min_freq);
984 	max_freq = READ_ONCE(cpudata->max_freq);
985 
986 	policy->cpuinfo.transition_latency = amd_pstate_get_transition_latency(policy->cpu);
987 	policy->transition_delay_us = amd_pstate_get_transition_delay_us(policy->cpu);
988 
989 	policy->min = min_freq;
990 	policy->max = max_freq;
991 
992 	policy->cpuinfo.min_freq = min_freq;
993 	policy->cpuinfo.max_freq = max_freq;
994 
995 	policy->boost_enabled = READ_ONCE(cpudata->boost_supported);
996 
997 	/* It will be updated by governor */
998 	policy->cur = policy->cpuinfo.min_freq;
999 
1000 	if (cpu_feature_enabled(X86_FEATURE_CPPC))
1001 		policy->fast_switch_possible = true;
1002 
1003 	ret = freq_qos_add_request(&policy->constraints, &cpudata->req[0],
1004 				   FREQ_QOS_MIN, policy->cpuinfo.min_freq);
1005 	if (ret < 0) {
1006 		dev_err(dev, "Failed to add min-freq constraint (%d)\n", ret);
1007 		goto free_cpudata1;
1008 	}
1009 
1010 	ret = freq_qos_add_request(&policy->constraints, &cpudata->req[1],
1011 				   FREQ_QOS_MAX, policy->cpuinfo.max_freq);
1012 	if (ret < 0) {
1013 		dev_err(dev, "Failed to add max-freq constraint (%d)\n", ret);
1014 		goto free_cpudata2;
1015 	}
1016 
1017 	cpudata->max_limit_freq = max_freq;
1018 	cpudata->min_limit_freq = min_freq;
1019 
1020 	policy->driver_data = cpudata;
1021 
1022 	if (!current_pstate_driver->adjust_perf)
1023 		current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
1024 
1025 	return 0;
1026 
1027 free_cpudata2:
1028 	freq_qos_remove_request(&cpudata->req[0]);
1029 free_cpudata1:
1030 	kfree(cpudata);
1031 	return ret;
1032 }
1033 
1034 static void amd_pstate_cpu_exit(struct cpufreq_policy *policy)
1035 {
1036 	struct amd_cpudata *cpudata = policy->driver_data;
1037 
1038 	freq_qos_remove_request(&cpudata->req[1]);
1039 	freq_qos_remove_request(&cpudata->req[0]);
1040 	policy->fast_switch_possible = false;
1041 	kfree(cpudata);
1042 }
1043 
1044 static int amd_pstate_cpu_resume(struct cpufreq_policy *policy)
1045 {
1046 	int ret;
1047 
1048 	ret = amd_pstate_enable(true);
1049 	if (ret)
1050 		pr_err("failed to enable amd-pstate during resume, return %d\n", ret);
1051 
1052 	return ret;
1053 }
1054 
1055 static int amd_pstate_cpu_suspend(struct cpufreq_policy *policy)
1056 {
1057 	int ret;
1058 
1059 	ret = amd_pstate_enable(false);
1060 	if (ret)
1061 		pr_err("failed to disable amd-pstate during suspend, return %d\n", ret);
1062 
1063 	return ret;
1064 }
1065 
1066 /* Sysfs attributes */
1067 
1068 /*
1069  * This frequency is to indicate the maximum hardware frequency.
1070  * If boost is not active but supported, the frequency will be larger than the
1071  * one in cpuinfo.
1072  */
1073 static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy,
1074 					char *buf)
1075 {
1076 	int max_freq;
1077 	struct amd_cpudata *cpudata = policy->driver_data;
1078 
1079 	max_freq = READ_ONCE(cpudata->max_freq);
1080 	if (max_freq < 0)
1081 		return max_freq;
1082 
1083 	return sysfs_emit(buf, "%u\n", max_freq);
1084 }
1085 
1086 static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *policy,
1087 						     char *buf)
1088 {
1089 	int freq;
1090 	struct amd_cpudata *cpudata = policy->driver_data;
1091 
1092 	freq = READ_ONCE(cpudata->lowest_nonlinear_freq);
1093 	if (freq < 0)
1094 		return freq;
1095 
1096 	return sysfs_emit(buf, "%u\n", freq);
1097 }
1098 
1099 /*
1100  * In some of ASICs, the highest_perf is not the one in the _CPC table, so we
1101  * need to expose it to sysfs.
1102  */
1103 static ssize_t show_amd_pstate_highest_perf(struct cpufreq_policy *policy,
1104 					    char *buf)
1105 {
1106 	u32 perf;
1107 	struct amd_cpudata *cpudata = policy->driver_data;
1108 
1109 	perf = READ_ONCE(cpudata->highest_perf);
1110 
1111 	return sysfs_emit(buf, "%u\n", perf);
1112 }
1113 
1114 static ssize_t show_amd_pstate_prefcore_ranking(struct cpufreq_policy *policy,
1115 						char *buf)
1116 {
1117 	u32 perf;
1118 	struct amd_cpudata *cpudata = policy->driver_data;
1119 
1120 	perf = READ_ONCE(cpudata->prefcore_ranking);
1121 
1122 	return sysfs_emit(buf, "%u\n", perf);
1123 }
1124 
1125 static ssize_t show_amd_pstate_hw_prefcore(struct cpufreq_policy *policy,
1126 					   char *buf)
1127 {
1128 	bool hw_prefcore;
1129 	struct amd_cpudata *cpudata = policy->driver_data;
1130 
1131 	hw_prefcore = READ_ONCE(cpudata->hw_prefcore);
1132 
1133 	return sysfs_emit(buf, "%s\n", str_enabled_disabled(hw_prefcore));
1134 }
1135 
1136 static ssize_t show_energy_performance_available_preferences(
1137 				struct cpufreq_policy *policy, char *buf)
1138 {
1139 	int i = 0;
1140 	int offset = 0;
1141 	struct amd_cpudata *cpudata = policy->driver_data;
1142 
1143 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1144 		return sysfs_emit_at(buf, offset, "%s\n",
1145 				energy_perf_strings[EPP_INDEX_PERFORMANCE]);
1146 
1147 	while (energy_perf_strings[i] != NULL)
1148 		offset += sysfs_emit_at(buf, offset, "%s ", energy_perf_strings[i++]);
1149 
1150 	offset += sysfs_emit_at(buf, offset, "\n");
1151 
1152 	return offset;
1153 }
1154 
1155 static ssize_t store_energy_performance_preference(
1156 		struct cpufreq_policy *policy, const char *buf, size_t count)
1157 {
1158 	struct amd_cpudata *cpudata = policy->driver_data;
1159 	char str_preference[21];
1160 	ssize_t ret;
1161 
1162 	ret = sscanf(buf, "%20s", str_preference);
1163 	if (ret != 1)
1164 		return -EINVAL;
1165 
1166 	ret = match_string(energy_perf_strings, -1, str_preference);
1167 	if (ret < 0)
1168 		return -EINVAL;
1169 
1170 	mutex_lock(&amd_pstate_limits_lock);
1171 	ret = amd_pstate_set_energy_pref_index(cpudata, ret);
1172 	mutex_unlock(&amd_pstate_limits_lock);
1173 
1174 	return ret ?: count;
1175 }
1176 
1177 static ssize_t show_energy_performance_preference(
1178 				struct cpufreq_policy *policy, char *buf)
1179 {
1180 	struct amd_cpudata *cpudata = policy->driver_data;
1181 	int preference;
1182 
1183 	preference = amd_pstate_get_energy_pref_index(cpudata);
1184 	if (preference < 0)
1185 		return preference;
1186 
1187 	return sysfs_emit(buf, "%s\n", energy_perf_strings[preference]);
1188 }
1189 
1190 static void amd_pstate_driver_cleanup(void)
1191 {
1192 	amd_pstate_enable(false);
1193 	cppc_state = AMD_PSTATE_DISABLE;
1194 	current_pstate_driver = NULL;
1195 }
1196 
1197 static int amd_pstate_register_driver(int mode)
1198 {
1199 	int ret;
1200 
1201 	if (mode == AMD_PSTATE_PASSIVE || mode == AMD_PSTATE_GUIDED)
1202 		current_pstate_driver = &amd_pstate_driver;
1203 	else if (mode == AMD_PSTATE_ACTIVE)
1204 		current_pstate_driver = &amd_pstate_epp_driver;
1205 	else
1206 		return -EINVAL;
1207 
1208 	cppc_state = mode;
1209 
1210 	ret = amd_pstate_enable(true);
1211 	if (ret) {
1212 		pr_err("failed to enable cppc during amd-pstate driver registration, return %d\n",
1213 		       ret);
1214 		amd_pstate_driver_cleanup();
1215 		return ret;
1216 	}
1217 
1218 	ret = cpufreq_register_driver(current_pstate_driver);
1219 	if (ret) {
1220 		amd_pstate_driver_cleanup();
1221 		return ret;
1222 	}
1223 
1224 	return 0;
1225 }
1226 
1227 static int amd_pstate_unregister_driver(int dummy)
1228 {
1229 	cpufreq_unregister_driver(current_pstate_driver);
1230 	amd_pstate_driver_cleanup();
1231 	return 0;
1232 }
1233 
1234 static int amd_pstate_change_mode_without_dvr_change(int mode)
1235 {
1236 	int cpu = 0;
1237 
1238 	cppc_state = mode;
1239 
1240 	if (cpu_feature_enabled(X86_FEATURE_CPPC) || cppc_state == AMD_PSTATE_ACTIVE)
1241 		return 0;
1242 
1243 	for_each_present_cpu(cpu) {
1244 		cppc_set_auto_sel(cpu, (cppc_state == AMD_PSTATE_PASSIVE) ? 0 : 1);
1245 	}
1246 
1247 	return 0;
1248 }
1249 
1250 static int amd_pstate_change_driver_mode(int mode)
1251 {
1252 	int ret;
1253 
1254 	ret = amd_pstate_unregister_driver(0);
1255 	if (ret)
1256 		return ret;
1257 
1258 	ret = amd_pstate_register_driver(mode);
1259 	if (ret)
1260 		return ret;
1261 
1262 	return 0;
1263 }
1264 
1265 static cppc_mode_transition_fn mode_state_machine[AMD_PSTATE_MAX][AMD_PSTATE_MAX] = {
1266 	[AMD_PSTATE_DISABLE]         = {
1267 		[AMD_PSTATE_DISABLE]     = NULL,
1268 		[AMD_PSTATE_PASSIVE]     = amd_pstate_register_driver,
1269 		[AMD_PSTATE_ACTIVE]      = amd_pstate_register_driver,
1270 		[AMD_PSTATE_GUIDED]      = amd_pstate_register_driver,
1271 	},
1272 	[AMD_PSTATE_PASSIVE]         = {
1273 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1274 		[AMD_PSTATE_PASSIVE]     = NULL,
1275 		[AMD_PSTATE_ACTIVE]      = amd_pstate_change_driver_mode,
1276 		[AMD_PSTATE_GUIDED]      = amd_pstate_change_mode_without_dvr_change,
1277 	},
1278 	[AMD_PSTATE_ACTIVE]          = {
1279 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1280 		[AMD_PSTATE_PASSIVE]     = amd_pstate_change_driver_mode,
1281 		[AMD_PSTATE_ACTIVE]      = NULL,
1282 		[AMD_PSTATE_GUIDED]      = amd_pstate_change_driver_mode,
1283 	},
1284 	[AMD_PSTATE_GUIDED]          = {
1285 		[AMD_PSTATE_DISABLE]     = amd_pstate_unregister_driver,
1286 		[AMD_PSTATE_PASSIVE]     = amd_pstate_change_mode_without_dvr_change,
1287 		[AMD_PSTATE_ACTIVE]      = amd_pstate_change_driver_mode,
1288 		[AMD_PSTATE_GUIDED]      = NULL,
1289 	},
1290 };
1291 
1292 static ssize_t amd_pstate_show_status(char *buf)
1293 {
1294 	if (!current_pstate_driver)
1295 		return sysfs_emit(buf, "disable\n");
1296 
1297 	return sysfs_emit(buf, "%s\n", amd_pstate_mode_string[cppc_state]);
1298 }
1299 
1300 int amd_pstate_update_status(const char *buf, size_t size)
1301 {
1302 	int mode_idx;
1303 
1304 	if (size > strlen("passive") || size < strlen("active"))
1305 		return -EINVAL;
1306 
1307 	mode_idx = get_mode_idx_from_str(buf, size);
1308 
1309 	if (mode_idx < 0 || mode_idx >= AMD_PSTATE_MAX)
1310 		return -EINVAL;
1311 
1312 	if (mode_state_machine[cppc_state][mode_idx])
1313 		return mode_state_machine[cppc_state][mode_idx](mode_idx);
1314 
1315 	return 0;
1316 }
1317 EXPORT_SYMBOL_GPL(amd_pstate_update_status);
1318 
1319 static ssize_t status_show(struct device *dev,
1320 			   struct device_attribute *attr, char *buf)
1321 {
1322 	ssize_t ret;
1323 
1324 	mutex_lock(&amd_pstate_driver_lock);
1325 	ret = amd_pstate_show_status(buf);
1326 	mutex_unlock(&amd_pstate_driver_lock);
1327 
1328 	return ret;
1329 }
1330 
1331 static ssize_t status_store(struct device *a, struct device_attribute *b,
1332 			    const char *buf, size_t count)
1333 {
1334 	char *p = memchr(buf, '\n', count);
1335 	int ret;
1336 
1337 	mutex_lock(&amd_pstate_driver_lock);
1338 	ret = amd_pstate_update_status(buf, p ? p - buf : count);
1339 	mutex_unlock(&amd_pstate_driver_lock);
1340 
1341 	return ret < 0 ? ret : count;
1342 }
1343 
1344 static ssize_t prefcore_show(struct device *dev,
1345 			     struct device_attribute *attr, char *buf)
1346 {
1347 	return sysfs_emit(buf, "%s\n", str_enabled_disabled(amd_pstate_prefcore));
1348 }
1349 
1350 cpufreq_freq_attr_ro(amd_pstate_max_freq);
1351 cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
1352 
1353 cpufreq_freq_attr_ro(amd_pstate_highest_perf);
1354 cpufreq_freq_attr_ro(amd_pstate_prefcore_ranking);
1355 cpufreq_freq_attr_ro(amd_pstate_hw_prefcore);
1356 cpufreq_freq_attr_rw(energy_performance_preference);
1357 cpufreq_freq_attr_ro(energy_performance_available_preferences);
1358 static DEVICE_ATTR_RW(status);
1359 static DEVICE_ATTR_RO(prefcore);
1360 
1361 static struct freq_attr *amd_pstate_attr[] = {
1362 	&amd_pstate_max_freq,
1363 	&amd_pstate_lowest_nonlinear_freq,
1364 	&amd_pstate_highest_perf,
1365 	&amd_pstate_prefcore_ranking,
1366 	&amd_pstate_hw_prefcore,
1367 	NULL,
1368 };
1369 
1370 static struct freq_attr *amd_pstate_epp_attr[] = {
1371 	&amd_pstate_max_freq,
1372 	&amd_pstate_lowest_nonlinear_freq,
1373 	&amd_pstate_highest_perf,
1374 	&amd_pstate_prefcore_ranking,
1375 	&amd_pstate_hw_prefcore,
1376 	&energy_performance_preference,
1377 	&energy_performance_available_preferences,
1378 	NULL,
1379 };
1380 
1381 static struct attribute *pstate_global_attributes[] = {
1382 	&dev_attr_status.attr,
1383 	&dev_attr_prefcore.attr,
1384 	NULL
1385 };
1386 
1387 static const struct attribute_group amd_pstate_global_attr_group = {
1388 	.name = "amd_pstate",
1389 	.attrs = pstate_global_attributes,
1390 };
1391 
1392 static bool amd_pstate_acpi_pm_profile_server(void)
1393 {
1394 	switch (acpi_gbl_FADT.preferred_profile) {
1395 	case PM_ENTERPRISE_SERVER:
1396 	case PM_SOHO_SERVER:
1397 	case PM_PERFORMANCE_SERVER:
1398 		return true;
1399 	}
1400 	return false;
1401 }
1402 
1403 static bool amd_pstate_acpi_pm_profile_undefined(void)
1404 {
1405 	if (acpi_gbl_FADT.preferred_profile == PM_UNSPECIFIED)
1406 		return true;
1407 	if (acpi_gbl_FADT.preferred_profile >= NR_PM_PROFILES)
1408 		return true;
1409 	return false;
1410 }
1411 
1412 static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
1413 {
1414 	int min_freq, max_freq, ret;
1415 	struct amd_cpudata *cpudata;
1416 	struct device *dev;
1417 	u64 value;
1418 
1419 	/*
1420 	 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
1421 	 * which is ideal for initialization process.
1422 	 */
1423 	amd_perf_ctl_reset(policy->cpu);
1424 	dev = get_cpu_device(policy->cpu);
1425 	if (!dev)
1426 		return -ENODEV;
1427 
1428 	cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
1429 	if (!cpudata)
1430 		return -ENOMEM;
1431 
1432 	cpudata->cpu = policy->cpu;
1433 	cpudata->epp_policy = 0;
1434 
1435 	ret = amd_pstate_init_perf(cpudata);
1436 	if (ret)
1437 		goto free_cpudata1;
1438 
1439 	amd_pstate_init_prefcore(cpudata);
1440 
1441 	ret = amd_pstate_init_freq(cpudata);
1442 	if (ret)
1443 		goto free_cpudata1;
1444 
1445 	ret = amd_pstate_init_boost_support(cpudata);
1446 	if (ret)
1447 		goto free_cpudata1;
1448 
1449 	min_freq = READ_ONCE(cpudata->min_freq);
1450 	max_freq = READ_ONCE(cpudata->max_freq);
1451 
1452 	policy->cpuinfo.min_freq = min_freq;
1453 	policy->cpuinfo.max_freq = max_freq;
1454 	/* It will be updated by governor */
1455 	policy->cur = policy->cpuinfo.min_freq;
1456 
1457 	policy->driver_data = cpudata;
1458 
1459 	cpudata->epp_cached = cpudata->epp_default = amd_pstate_get_epp(cpudata, 0);
1460 
1461 	policy->min = policy->cpuinfo.min_freq;
1462 	policy->max = policy->cpuinfo.max_freq;
1463 
1464 	policy->boost_enabled = READ_ONCE(cpudata->boost_supported);
1465 
1466 	/*
1467 	 * Set the policy to provide a valid fallback value in case
1468 	 * the default cpufreq governor is neither powersave nor performance.
1469 	 */
1470 	if (amd_pstate_acpi_pm_profile_server() ||
1471 	    amd_pstate_acpi_pm_profile_undefined())
1472 		policy->policy = CPUFREQ_POLICY_PERFORMANCE;
1473 	else
1474 		policy->policy = CPUFREQ_POLICY_POWERSAVE;
1475 
1476 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1477 		ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &value);
1478 		if (ret)
1479 			return ret;
1480 		WRITE_ONCE(cpudata->cppc_req_cached, value);
1481 
1482 		ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1, &value);
1483 		if (ret)
1484 			return ret;
1485 		WRITE_ONCE(cpudata->cppc_cap1_cached, value);
1486 	}
1487 
1488 	return 0;
1489 
1490 free_cpudata1:
1491 	kfree(cpudata);
1492 	return ret;
1493 }
1494 
1495 static void amd_pstate_epp_cpu_exit(struct cpufreq_policy *policy)
1496 {
1497 	struct amd_cpudata *cpudata = policy->driver_data;
1498 
1499 	if (cpudata) {
1500 		kfree(cpudata);
1501 		policy->driver_data = NULL;
1502 	}
1503 
1504 	pr_debug("CPU %d exiting\n", policy->cpu);
1505 }
1506 
1507 static int amd_pstate_epp_update_limit(struct cpufreq_policy *policy)
1508 {
1509 	struct amd_cpudata *cpudata = policy->driver_data;
1510 	u32 max_perf, min_perf, min_limit_perf, max_limit_perf;
1511 	u64 value;
1512 	s16 epp;
1513 
1514 	if (cpudata->boost_supported && !policy->boost_enabled)
1515 		max_perf = READ_ONCE(cpudata->nominal_perf);
1516 	else
1517 		max_perf = READ_ONCE(cpudata->highest_perf);
1518 	min_perf = READ_ONCE(cpudata->lowest_perf);
1519 	max_limit_perf = div_u64(policy->max * max_perf, policy->cpuinfo.max_freq);
1520 	min_limit_perf = div_u64(policy->min * max_perf, policy->cpuinfo.max_freq);
1521 
1522 	if (min_limit_perf < min_perf)
1523 		min_limit_perf = min_perf;
1524 
1525 	if (max_limit_perf < min_limit_perf)
1526 		max_limit_perf = min_limit_perf;
1527 
1528 	WRITE_ONCE(cpudata->max_limit_perf, max_limit_perf);
1529 	WRITE_ONCE(cpudata->min_limit_perf, min_limit_perf);
1530 
1531 	max_perf = clamp_t(unsigned long, max_perf, cpudata->min_limit_perf,
1532 			cpudata->max_limit_perf);
1533 	min_perf = clamp_t(unsigned long, min_perf, cpudata->min_limit_perf,
1534 			cpudata->max_limit_perf);
1535 	value = READ_ONCE(cpudata->cppc_req_cached);
1536 
1537 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1538 		min_perf = max_perf;
1539 
1540 	/* Initial min/max values for CPPC Performance Controls Register */
1541 	value &= ~AMD_CPPC_MIN_PERF(~0L);
1542 	value |= AMD_CPPC_MIN_PERF(min_perf);
1543 
1544 	value &= ~AMD_CPPC_MAX_PERF(~0L);
1545 	value |= AMD_CPPC_MAX_PERF(max_perf);
1546 
1547 	/* CPPC EPP feature require to set zero to the desire perf bit */
1548 	value &= ~AMD_CPPC_DES_PERF(~0L);
1549 	value |= AMD_CPPC_DES_PERF(0);
1550 
1551 	cpudata->epp_policy = cpudata->policy;
1552 
1553 	/* Get BIOS pre-defined epp value */
1554 	epp = amd_pstate_get_epp(cpudata, value);
1555 	if (epp < 0) {
1556 		/**
1557 		 * This return value can only be negative for shared_memory
1558 		 * systems where EPP register read/write not supported.
1559 		 */
1560 		return epp;
1561 	}
1562 
1563 	if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1564 		epp = 0;
1565 
1566 	/* Set initial EPP value */
1567 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1568 		value &= ~GENMASK_ULL(31, 24);
1569 		value |= (u64)epp << 24;
1570 	}
1571 
1572 	WRITE_ONCE(cpudata->cppc_req_cached, value);
1573 	return amd_pstate_set_epp(cpudata, epp);
1574 }
1575 
1576 static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy)
1577 {
1578 	struct amd_cpudata *cpudata = policy->driver_data;
1579 	int ret;
1580 
1581 	if (!policy->cpuinfo.max_freq)
1582 		return -ENODEV;
1583 
1584 	pr_debug("set_policy: cpuinfo.max %u policy->max %u\n",
1585 				policy->cpuinfo.max_freq, policy->max);
1586 
1587 	cpudata->policy = policy->policy;
1588 
1589 	ret = amd_pstate_epp_update_limit(policy);
1590 	if (ret)
1591 		return ret;
1592 
1593 	/*
1594 	 * policy->cur is never updated with the amd_pstate_epp driver, but it
1595 	 * is used as a stale frequency value. So, keep it within limits.
1596 	 */
1597 	policy->cur = policy->min;
1598 
1599 	return 0;
1600 }
1601 
1602 static void amd_pstate_epp_reenable(struct amd_cpudata *cpudata)
1603 {
1604 	struct cppc_perf_ctrls perf_ctrls;
1605 	u64 value, max_perf;
1606 	int ret;
1607 
1608 	ret = amd_pstate_enable(true);
1609 	if (ret)
1610 		pr_err("failed to enable amd pstate during resume, return %d\n", ret);
1611 
1612 	value = READ_ONCE(cpudata->cppc_req_cached);
1613 	max_perf = READ_ONCE(cpudata->highest_perf);
1614 
1615 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1616 		wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
1617 	} else {
1618 		perf_ctrls.max_perf = max_perf;
1619 		perf_ctrls.energy_perf = AMD_CPPC_ENERGY_PERF_PREF(cpudata->epp_cached);
1620 		cppc_set_perf(cpudata->cpu, &perf_ctrls);
1621 	}
1622 }
1623 
1624 static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy)
1625 {
1626 	struct amd_cpudata *cpudata = policy->driver_data;
1627 
1628 	pr_debug("AMD CPU Core %d going online\n", cpudata->cpu);
1629 
1630 	if (cppc_state == AMD_PSTATE_ACTIVE) {
1631 		amd_pstate_epp_reenable(cpudata);
1632 		cpudata->suspended = false;
1633 	}
1634 
1635 	return 0;
1636 }
1637 
1638 static void amd_pstate_epp_offline(struct cpufreq_policy *policy)
1639 {
1640 	struct amd_cpudata *cpudata = policy->driver_data;
1641 	struct cppc_perf_ctrls perf_ctrls;
1642 	int min_perf;
1643 	u64 value;
1644 
1645 	min_perf = READ_ONCE(cpudata->lowest_perf);
1646 	value = READ_ONCE(cpudata->cppc_req_cached);
1647 
1648 	mutex_lock(&amd_pstate_limits_lock);
1649 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1650 		cpudata->epp_policy = CPUFREQ_POLICY_UNKNOWN;
1651 
1652 		/* Set max perf same as min perf */
1653 		value &= ~AMD_CPPC_MAX_PERF(~0L);
1654 		value |= AMD_CPPC_MAX_PERF(min_perf);
1655 		value &= ~AMD_CPPC_MIN_PERF(~0L);
1656 		value |= AMD_CPPC_MIN_PERF(min_perf);
1657 		wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
1658 	} else {
1659 		perf_ctrls.desired_perf = 0;
1660 		perf_ctrls.max_perf = min_perf;
1661 		perf_ctrls.energy_perf = AMD_CPPC_ENERGY_PERF_PREF(HWP_EPP_BALANCE_POWERSAVE);
1662 		cppc_set_perf(cpudata->cpu, &perf_ctrls);
1663 	}
1664 	mutex_unlock(&amd_pstate_limits_lock);
1665 }
1666 
1667 static int amd_pstate_epp_cpu_offline(struct cpufreq_policy *policy)
1668 {
1669 	struct amd_cpudata *cpudata = policy->driver_data;
1670 
1671 	pr_debug("AMD CPU Core %d going offline\n", cpudata->cpu);
1672 
1673 	if (cpudata->suspended)
1674 		return 0;
1675 
1676 	if (cppc_state == AMD_PSTATE_ACTIVE)
1677 		amd_pstate_epp_offline(policy);
1678 
1679 	return 0;
1680 }
1681 
1682 static int amd_pstate_epp_verify_policy(struct cpufreq_policy_data *policy)
1683 {
1684 	cpufreq_verify_within_cpu_limits(policy);
1685 	pr_debug("policy_max =%d, policy_min=%d\n", policy->max, policy->min);
1686 	return 0;
1687 }
1688 
1689 static int amd_pstate_epp_suspend(struct cpufreq_policy *policy)
1690 {
1691 	struct amd_cpudata *cpudata = policy->driver_data;
1692 	int ret;
1693 
1694 	/* avoid suspending when EPP is not enabled */
1695 	if (cppc_state != AMD_PSTATE_ACTIVE)
1696 		return 0;
1697 
1698 	/* set this flag to avoid setting core offline*/
1699 	cpudata->suspended = true;
1700 
1701 	/* disable CPPC in lowlevel firmware */
1702 	ret = amd_pstate_enable(false);
1703 	if (ret)
1704 		pr_err("failed to suspend, return %d\n", ret);
1705 
1706 	return 0;
1707 }
1708 
1709 static int amd_pstate_epp_resume(struct cpufreq_policy *policy)
1710 {
1711 	struct amd_cpudata *cpudata = policy->driver_data;
1712 
1713 	if (cpudata->suspended) {
1714 		mutex_lock(&amd_pstate_limits_lock);
1715 
1716 		/* enable amd pstate from suspend state*/
1717 		amd_pstate_epp_reenable(cpudata);
1718 
1719 		mutex_unlock(&amd_pstate_limits_lock);
1720 
1721 		cpudata->suspended = false;
1722 	}
1723 
1724 	return 0;
1725 }
1726 
1727 static struct cpufreq_driver amd_pstate_driver = {
1728 	.flags		= CPUFREQ_CONST_LOOPS | CPUFREQ_NEED_UPDATE_LIMITS,
1729 	.verify		= amd_pstate_verify,
1730 	.target		= amd_pstate_target,
1731 	.fast_switch    = amd_pstate_fast_switch,
1732 	.init		= amd_pstate_cpu_init,
1733 	.exit		= amd_pstate_cpu_exit,
1734 	.suspend	= amd_pstate_cpu_suspend,
1735 	.resume		= amd_pstate_cpu_resume,
1736 	.set_boost	= amd_pstate_set_boost,
1737 	.update_limits	= amd_pstate_update_limits,
1738 	.name		= "amd-pstate",
1739 	.attr		= amd_pstate_attr,
1740 };
1741 
1742 static struct cpufreq_driver amd_pstate_epp_driver = {
1743 	.flags		= CPUFREQ_CONST_LOOPS,
1744 	.verify		= amd_pstate_epp_verify_policy,
1745 	.setpolicy	= amd_pstate_epp_set_policy,
1746 	.init		= amd_pstate_epp_cpu_init,
1747 	.exit		= amd_pstate_epp_cpu_exit,
1748 	.offline	= amd_pstate_epp_cpu_offline,
1749 	.online		= amd_pstate_epp_cpu_online,
1750 	.suspend	= amd_pstate_epp_suspend,
1751 	.resume		= amd_pstate_epp_resume,
1752 	.update_limits	= amd_pstate_update_limits,
1753 	.set_boost	= amd_pstate_set_boost,
1754 	.name		= "amd-pstate-epp",
1755 	.attr		= amd_pstate_epp_attr,
1756 };
1757 
1758 static int __init amd_pstate_set_driver(int mode_idx)
1759 {
1760 	if (mode_idx >= AMD_PSTATE_DISABLE && mode_idx < AMD_PSTATE_MAX) {
1761 		cppc_state = mode_idx;
1762 		if (cppc_state == AMD_PSTATE_DISABLE)
1763 			pr_info("driver is explicitly disabled\n");
1764 
1765 		if (cppc_state == AMD_PSTATE_ACTIVE)
1766 			current_pstate_driver = &amd_pstate_epp_driver;
1767 
1768 		if (cppc_state == AMD_PSTATE_PASSIVE || cppc_state == AMD_PSTATE_GUIDED)
1769 			current_pstate_driver = &amd_pstate_driver;
1770 
1771 		return 0;
1772 	}
1773 
1774 	return -EINVAL;
1775 }
1776 
1777 /**
1778  * CPPC function is not supported for family ID 17H with model_ID ranging from 0x10 to 0x2F.
1779  * show the debug message that helps to check if the CPU has CPPC support for loading issue.
1780  */
1781 static bool amd_cppc_supported(void)
1782 {
1783 	struct cpuinfo_x86 *c = &cpu_data(0);
1784 	bool warn = false;
1785 
1786 	if ((boot_cpu_data.x86 == 0x17) && (boot_cpu_data.x86_model < 0x30)) {
1787 		pr_debug_once("CPPC feature is not supported by the processor\n");
1788 		return false;
1789 	}
1790 
1791 	/*
1792 	 * If the CPPC feature is disabled in the BIOS for processors
1793 	 * that support MSR-based CPPC, the AMD Pstate driver may not
1794 	 * function correctly.
1795 	 *
1796 	 * For such processors, check the CPPC flag and display a
1797 	 * warning message if the platform supports CPPC.
1798 	 *
1799 	 * Note: The code check below will not abort the driver
1800 	 * registration process because of the code is added for
1801 	 * debugging purposes. Besides, it may still be possible for
1802 	 * the driver to work using the shared-memory mechanism.
1803 	 */
1804 	if (!cpu_feature_enabled(X86_FEATURE_CPPC)) {
1805 		if (cpu_feature_enabled(X86_FEATURE_ZEN2)) {
1806 			switch (c->x86_model) {
1807 			case 0x60 ... 0x6F:
1808 			case 0x80 ... 0xAF:
1809 				warn = true;
1810 				break;
1811 			}
1812 		} else if (cpu_feature_enabled(X86_FEATURE_ZEN3) ||
1813 			   cpu_feature_enabled(X86_FEATURE_ZEN4)) {
1814 			switch (c->x86_model) {
1815 			case 0x10 ... 0x1F:
1816 			case 0x40 ... 0xAF:
1817 				warn = true;
1818 				break;
1819 			}
1820 		} else if (cpu_feature_enabled(X86_FEATURE_ZEN5)) {
1821 			warn = true;
1822 		}
1823 	}
1824 
1825 	if (warn)
1826 		pr_warn_once("The CPPC feature is supported but currently disabled by the BIOS.\n"
1827 					"Please enable it if your BIOS has the CPPC option.\n");
1828 	return true;
1829 }
1830 
1831 static int __init amd_pstate_init(void)
1832 {
1833 	struct device *dev_root;
1834 	int ret;
1835 
1836 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
1837 		return -ENODEV;
1838 
1839 	/* show debug message only if CPPC is not supported */
1840 	if (!amd_cppc_supported())
1841 		return -EOPNOTSUPP;
1842 
1843 	/* show warning message when BIOS broken or ACPI disabled */
1844 	if (!acpi_cpc_valid()) {
1845 		pr_warn_once("the _CPC object is not present in SBIOS or ACPI disabled\n");
1846 		return -ENODEV;
1847 	}
1848 
1849 	/* don't keep reloading if cpufreq_driver exists */
1850 	if (cpufreq_get_current_driver())
1851 		return -EEXIST;
1852 
1853 	quirks = NULL;
1854 
1855 	/* check if this machine need CPPC quirks */
1856 	dmi_check_system(amd_pstate_quirks_table);
1857 
1858 	/*
1859 	* determine the driver mode from the command line or kernel config.
1860 	* If no command line input is provided, cppc_state will be AMD_PSTATE_UNDEFINED.
1861 	* command line options will override the kernel config settings.
1862 	*/
1863 
1864 	if (cppc_state == AMD_PSTATE_UNDEFINED) {
1865 		/* Disable on the following configs by default:
1866 		 * 1. Undefined platforms
1867 		 * 2. Server platforms
1868 		 */
1869 		if (amd_pstate_acpi_pm_profile_undefined() ||
1870 		    amd_pstate_acpi_pm_profile_server()) {
1871 			pr_info("driver load is disabled, boot with specific mode to enable this\n");
1872 			return -ENODEV;
1873 		}
1874 		/* get driver mode from kernel config option [1:4] */
1875 		cppc_state = CONFIG_X86_AMD_PSTATE_DEFAULT_MODE;
1876 	}
1877 
1878 	switch (cppc_state) {
1879 	case AMD_PSTATE_DISABLE:
1880 		pr_info("driver load is disabled, boot with specific mode to enable this\n");
1881 		return -ENODEV;
1882 	case AMD_PSTATE_PASSIVE:
1883 	case AMD_PSTATE_ACTIVE:
1884 	case AMD_PSTATE_GUIDED:
1885 		ret = amd_pstate_set_driver(cppc_state);
1886 		if (ret)
1887 			return ret;
1888 		break;
1889 	default:
1890 		return -EINVAL;
1891 	}
1892 
1893 	/* capability check */
1894 	if (cpu_feature_enabled(X86_FEATURE_CPPC)) {
1895 		pr_debug("AMD CPPC MSR based functionality is supported\n");
1896 		if (cppc_state != AMD_PSTATE_ACTIVE)
1897 			current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
1898 	} else {
1899 		pr_debug("AMD CPPC shared memory based functionality is supported\n");
1900 		static_call_update(amd_pstate_enable, cppc_enable);
1901 		static_call_update(amd_pstate_init_perf, cppc_init_perf);
1902 		static_call_update(amd_pstate_update_perf, cppc_update_perf);
1903 	}
1904 
1905 	if (amd_pstate_prefcore) {
1906 		ret = amd_detect_prefcore(&amd_pstate_prefcore);
1907 		if (ret)
1908 			return ret;
1909 	}
1910 
1911 	/* enable amd pstate feature */
1912 	ret = amd_pstate_enable(true);
1913 	if (ret) {
1914 		pr_err("failed to enable driver mode(%d)\n", cppc_state);
1915 		return ret;
1916 	}
1917 
1918 	ret = cpufreq_register_driver(current_pstate_driver);
1919 	if (ret) {
1920 		pr_err("failed to register with return %d\n", ret);
1921 		goto disable_driver;
1922 	}
1923 
1924 	dev_root = bus_get_dev_root(&cpu_subsys);
1925 	if (dev_root) {
1926 		ret = sysfs_create_group(&dev_root->kobj, &amd_pstate_global_attr_group);
1927 		put_device(dev_root);
1928 		if (ret) {
1929 			pr_err("sysfs attribute export failed with error %d.\n", ret);
1930 			goto global_attr_free;
1931 		}
1932 	}
1933 
1934 	return ret;
1935 
1936 global_attr_free:
1937 	cpufreq_unregister_driver(current_pstate_driver);
1938 disable_driver:
1939 	amd_pstate_enable(false);
1940 	return ret;
1941 }
1942 device_initcall(amd_pstate_init);
1943 
1944 static int __init amd_pstate_param(char *str)
1945 {
1946 	size_t size;
1947 	int mode_idx;
1948 
1949 	if (!str)
1950 		return -EINVAL;
1951 
1952 	size = strlen(str);
1953 	mode_idx = get_mode_idx_from_str(str, size);
1954 
1955 	return amd_pstate_set_driver(mode_idx);
1956 }
1957 
1958 static int __init amd_prefcore_param(char *str)
1959 {
1960 	if (!strcmp(str, "disable"))
1961 		amd_pstate_prefcore = false;
1962 
1963 	return 0;
1964 }
1965 
1966 early_param("amd_pstate", amd_pstate_param);
1967 early_param("amd_prefcore", amd_prefcore_param);
1968 
1969 MODULE_AUTHOR("Huang Rui <ray.huang@amd.com>");
1970 MODULE_DESCRIPTION("AMD Processor P-state Frequency Driver");
1971