projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'pull-fd' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-block.git] / drivers / cpufreq / amd-pstate.c
CommitLineData
ec437d71
HR		 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>
f1375ec1 39#include <linux/amd-pstate.h>
ec437d71
HR		 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>
60e10f89 48#include "amd-pstate-trace.h"
ec437d71 49
ca08e46d
PY		 50#define AMD_PSTATE_TRANSITION_LATENCY 20000
51#define AMD_PSTATE_TRANSITION_DELAY 1000
ec437d71 52
e059c184
HR		 53/*
54 * TODO: We need more time to fine tune processors with shared memory solution
55 * with community together.
56 *
57 * There are some performance drops on the CPU benchmarks which reports from
58 * Suse. We are co-working with them to fine tune the shared memory solution. So
59 * we disable it by default to go acpi-cpufreq on these processors and add a
60 * module parameter to be able to enable it manually for debugging.
61 */
ffa5096a 62static struct cpufreq_driver *current_pstate_driver;
ec437d71 63static struct cpufreq_driver amd_pstate_driver;
ffa5096a 64static struct cpufreq_driver amd_pstate_epp_driver;
36c5014e 65static int cppc_state = AMD_PSTATE_DISABLE;
abd61c08 66struct kobject *amd_pstate_kobj;
36c5014e 67
ffa5096a
PY		 68/*
69 * AMD Energy Preference Performance (EPP)
70 * The EPP is used in the CCLK DPM controller to drive
71 * the frequency that a core is going to operate during
72 * short periods of activity. EPP values will be utilized for
73 * different OS profiles (balanced, performance, power savings)
74 * display strings corresponding to EPP index in the
75 * energy_perf_strings[]
76 * index String
77 *-------------------------------------
78 * 0 default
79 * 1 performance
80 * 2 balance_performance
81 * 3 balance_power
82 * 4 power
83 */
84enum energy_perf_value_index {
85 EPP_INDEX_DEFAULT = 0,
86 EPP_INDEX_PERFORMANCE,
87 EPP_INDEX_BALANCE_PERFORMANCE,
88 EPP_INDEX_BALANCE_POWERSAVE,
89 EPP_INDEX_POWERSAVE,
90};
91
92static const char * const energy_perf_strings[] = {
93 [EPP_INDEX_DEFAULT] = "default",
94 [EPP_INDEX_PERFORMANCE] = "performance",
95 [EPP_INDEX_BALANCE_PERFORMANCE] = "balance_performance",
96 [EPP_INDEX_BALANCE_POWERSAVE] = "balance_power",
97 [EPP_INDEX_POWERSAVE] = "power",
98 NULL
99};
100
101static unsigned int epp_values[] = {
102 [EPP_INDEX_DEFAULT] = 0,
103 [EPP_INDEX_PERFORMANCE] = AMD_CPPC_EPP_PERFORMANCE,
104 [EPP_INDEX_BALANCE_PERFORMANCE] = AMD_CPPC_EPP_BALANCE_PERFORMANCE,
105 [EPP_INDEX_BALANCE_POWERSAVE] = AMD_CPPC_EPP_BALANCE_POWERSAVE,
106 [EPP_INDEX_POWERSAVE] = AMD_CPPC_EPP_POWERSAVE,
107 };
108
36c5014e
WK		 109static inline int get_mode_idx_from_str(const char *str, size_t size)
110{
111 int i;
112
113 for (i=0; i < AMD_PSTATE_MAX; i++) {
114 if (!strncmp(str, amd_pstate_mode_string[i], size))
115 return i;
116 }
117 return -EINVAL;
118}
ec437d71 119
ffa5096a
PY		 120static DEFINE_MUTEX(amd_pstate_limits_lock);
121static DEFINE_MUTEX(amd_pstate_driver_lock);
122
123static s16 amd_pstate_get_epp(struct amd_cpudata *cpudata, u64 cppc_req_cached)
124{
125 u64 epp;
126 int ret;
127
128 if (boot_cpu_has(X86_FEATURE_CPPC)) {
129 if (!cppc_req_cached) {
130 epp = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ,
131 &cppc_req_cached);
132 if (epp)
133 return epp;
134 }
135 epp = (cppc_req_cached >> 24) & 0xFF;
136 } else {
137 ret = cppc_get_epp_perf(cpudata->cpu, &epp);
138 if (ret < 0) {
139 pr_debug("Could not retrieve energy perf value (%d)\n", ret);
140 return -EIO;
141 }
142 }
143
144 return (s16)(epp & 0xff);
145}
146
147static int amd_pstate_get_energy_pref_index(struct amd_cpudata *cpudata)
148{
149 s16 epp;
150 int index = -EINVAL;
151
152 epp = amd_pstate_get_epp(cpudata, 0);
153 if (epp < 0)
154 return epp;
155
156 switch (epp) {
157 case AMD_CPPC_EPP_PERFORMANCE:
158 index = EPP_INDEX_PERFORMANCE;
159 break;
160 case AMD_CPPC_EPP_BALANCE_PERFORMANCE:
161 index = EPP_INDEX_BALANCE_PERFORMANCE;
162 break;
163 case AMD_CPPC_EPP_BALANCE_POWERSAVE:
164 index = EPP_INDEX_BALANCE_POWERSAVE;
165 break;
166 case AMD_CPPC_EPP_POWERSAVE:
167 index = EPP_INDEX_POWERSAVE;
168 break;
169 default:
170 break;
171 }
172
173 return index;
174}
175
176static int amd_pstate_set_epp(struct amd_cpudata *cpudata, u32 epp)
177{
178 int ret;
179 struct cppc_perf_ctrls perf_ctrls;
180
181 if (boot_cpu_has(X86_FEATURE_CPPC)) {
182 u64 value = READ_ONCE(cpudata->cppc_req_cached);
183
184 value &= ~GENMASK_ULL(31, 24);
185 value |= (u64)epp << 24;
186 WRITE_ONCE(cpudata->cppc_req_cached, value);
187
188 ret = wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
189 if (!ret)
190 cpudata->epp_cached = epp;
191 } else {
192 perf_ctrls.energy_perf = epp;
193 ret = cppc_set_epp_perf(cpudata->cpu, &perf_ctrls, 1);
194 if (ret) {
195 pr_debug("failed to set energy perf value (%d)\n", ret);
196 return ret;
197 }
198 cpudata->epp_cached = epp;
199 }
200
201 return ret;
202}
203
204static int amd_pstate_set_energy_pref_index(struct amd_cpudata *cpudata,
205 int pref_index)
206{
207 int epp = -EINVAL;
208 int ret;
209
210 if (!pref_index) {
211 pr_debug("EPP pref_index is invalid\n");
212 return -EINVAL;
213 }
214
215 if (epp == -EINVAL)
216 epp = epp_values[pref_index];
217
218 if (epp > 0 && cpudata->policy == CPUFREQ_POLICY_PERFORMANCE) {
219 pr_debug("EPP cannot be set under performance policy\n");
220 return -EBUSY;
221 }
222
223 ret = amd_pstate_set_epp(cpudata, epp);
224
225 return ret;
226}
227
e059c184 228static inline int pstate_enable(bool enable)
ec437d71
HR		 229{
230 return wrmsrl_safe(MSR_AMD_CPPC_ENABLE, enable);
231}
232
e059c184
HR		 233static int cppc_enable(bool enable)
234{
235 int cpu, ret = 0;
ffa5096a 236 struct cppc_perf_ctrls perf_ctrls;
e059c184
HR		 237
238 for_each_present_cpu(cpu) {
239 ret = cppc_set_enable(cpu, enable);
240 if (ret)
241 return ret;
ffa5096a
PY		 242
243 /* Enable autonomous mode for EPP */
244 if (cppc_state == AMD_PSTATE_ACTIVE) {
245 /* Set desired perf as zero to allow EPP firmware control */
246 perf_ctrls.desired_perf = 0;
247 ret = cppc_set_perf(cpu, &perf_ctrls);
248 if (ret)
249 return ret;
250 }
e059c184
HR		 251 }
252
253 return ret;
254}
255
256DEFINE_STATIC_CALL(amd_pstate_enable, pstate_enable);
257
258static inline int amd_pstate_enable(bool enable)
259{
260 return static_call(amd_pstate_enable)(enable);
261}
262
263static int pstate_init_perf(struct amd_cpudata *cpudata)
ec437d71
HR		 264{
265 u64 cap1;
bedadcfb 266 u32 highest_perf;
ec437d71
HR		 267
268 int ret = rdmsrl_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
269 &cap1);
270 if (ret)
271 return ret;
272
273 /*
274 * TODO: Introduce AMD specific power feature.
275 *
276 * CPPC entry doesn't indicate the highest performance in some ASICs.
277 */
bedadcfb
PY		 278 highest_perf = amd_get_highest_perf();
279 if (highest_perf > AMD_CPPC_HIGHEST_PERF(cap1))
280 highest_perf = AMD_CPPC_HIGHEST_PERF(cap1);
281
282 WRITE_ONCE(cpudata->highest_perf, highest_perf);
ec437d71
HR		 283
284 WRITE_ONCE(cpudata->nominal_perf, AMD_CPPC_NOMINAL_PERF(cap1));
285 WRITE_ONCE(cpudata->lowest_nonlinear_perf, AMD_CPPC_LOWNONLIN_PERF(cap1));
286 WRITE_ONCE(cpudata->lowest_perf, AMD_CPPC_LOWEST_PERF(cap1));
287
288 return 0;
289}
290
e059c184
HR		 291static int cppc_init_perf(struct amd_cpudata *cpudata)
292{
293 struct cppc_perf_caps cppc_perf;
bedadcfb 294 u32 highest_perf;
e059c184
HR		 295
296 int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
297 if (ret)
298 return ret;
299
bedadcfb
PY		 300 highest_perf = amd_get_highest_perf();
301 if (highest_perf > cppc_perf.highest_perf)
302 highest_perf = cppc_perf.highest_perf;
303
304 WRITE_ONCE(cpudata->highest_perf, highest_perf);
e059c184
HR		 305
306 WRITE_ONCE(cpudata->nominal_perf, cppc_perf.nominal_perf);
307 WRITE_ONCE(cpudata->lowest_nonlinear_perf,
308 cppc_perf.lowest_nonlinear_perf);
309 WRITE_ONCE(cpudata->lowest_perf, cppc_perf.lowest_perf);
310
311 return 0;
312}
313
314DEFINE_STATIC_CALL(amd_pstate_init_perf, pstate_init_perf);
315
316static inline int amd_pstate_init_perf(struct amd_cpudata *cpudata)
317{
318 return static_call(amd_pstate_init_perf)(cpudata);
319}
320
321static void pstate_update_perf(struct amd_cpudata *cpudata, u32 min_perf,
322 u32 des_perf, u32 max_perf, bool fast_switch)
ec437d71
HR		 323{
324 if (fast_switch)
325 wrmsrl(MSR_AMD_CPPC_REQ, READ_ONCE(cpudata->cppc_req_cached));
326 else
327 wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ,
328 READ_ONCE(cpudata->cppc_req_cached));
329}
330
e059c184
HR		 331static void cppc_update_perf(struct amd_cpudata *cpudata,
332 u32 min_perf, u32 des_perf,
333 u32 max_perf, bool fast_switch)
334{
335 struct cppc_perf_ctrls perf_ctrls;
336
337 perf_ctrls.max_perf = max_perf;
338 perf_ctrls.min_perf = min_perf;
339 perf_ctrls.desired_perf = des_perf;
340
341 cppc_set_perf(cpudata->cpu, &perf_ctrls);
342}
343
344DEFINE_STATIC_CALL(amd_pstate_update_perf, pstate_update_perf);
345
346static inline void amd_pstate_update_perf(struct amd_cpudata *cpudata,
347 u32 min_perf, u32 des_perf,
348 u32 max_perf, bool fast_switch)
349{
350 static_call(amd_pstate_update_perf)(cpudata, min_perf, des_perf,
351 max_perf, fast_switch);
352}
353
23c296fb
JS		 354static inline bool amd_pstate_sample(struct amd_cpudata *cpudata)
355{
356 u64 aperf, mperf, tsc;
357 unsigned long flags;
358
359 local_irq_save(flags);
360 rdmsrl(MSR_IA32_APERF, aperf);
361 rdmsrl(MSR_IA32_MPERF, mperf);
362 tsc = rdtsc();
363
364 if (cpudata->prev.mperf == mperf || cpudata->prev.tsc == tsc) {
365 local_irq_restore(flags);
366 return false;
367 }
368
369 local_irq_restore(flags);
370
371 cpudata->cur.aperf = aperf;
372 cpudata->cur.mperf = mperf;
373 cpudata->cur.tsc = tsc;
374 cpudata->cur.aperf -= cpudata->prev.aperf;
375 cpudata->cur.mperf -= cpudata->prev.mperf;
376 cpudata->cur.tsc -= cpudata->prev.tsc;
377
378 cpudata->prev.aperf = aperf;
379 cpudata->prev.mperf = mperf;
380 cpudata->prev.tsc = tsc;
381
382 cpudata->freq = div64_u64((cpudata->cur.aperf * cpu_khz), cpudata->cur.mperf);
383
384 return true;
385}
386
ec437d71
HR		 387static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
388 u32 des_perf, u32 max_perf, bool fast_switch)
389{
390 u64 prev = READ_ONCE(cpudata->cppc_req_cached);
391 u64 value = prev;
392
0e9a8638 393 des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
ec437d71
HR		 394 value &= ~AMD_CPPC_MIN_PERF(~0L);
395 value |= AMD_CPPC_MIN_PERF(min_perf);
396
397 value &= ~AMD_CPPC_DES_PERF(~0L);
398 value |= AMD_CPPC_DES_PERF(des_perf);
399
400 value &= ~AMD_CPPC_MAX_PERF(~0L);
401 value |= AMD_CPPC_MAX_PERF(max_perf);
402
23c296fb
JS		 403 if (trace_amd_pstate_perf_enabled() && amd_pstate_sample(cpudata)) {
404 trace_amd_pstate_perf(min_perf, des_perf, max_perf, cpudata->freq,
405 cpudata->cur.mperf, cpudata->cur.aperf, cpudata->cur.tsc,
406 cpudata->cpu, (value != prev), fast_switch);
407 }
60e10f89 408
ec437d71
HR		 409 if (value == prev)
410 return;
411
412 WRITE_ONCE(cpudata->cppc_req_cached, value);
413
414 amd_pstate_update_perf(cpudata, min_perf, des_perf,
415 max_perf, fast_switch);
416}
417
418static int amd_pstate_verify(struct cpufreq_policy_data *policy)
419{
420 cpufreq_verify_within_cpu_limits(policy);
421
422 return 0;
423}
424
425static int amd_pstate_target(struct cpufreq_policy *policy,
426 unsigned int target_freq,
427 unsigned int relation)
428{
429 struct cpufreq_freqs freqs;
430 struct amd_cpudata *cpudata = policy->driver_data;
431 unsigned long max_perf, min_perf, des_perf, cap_perf;
432
433 if (!cpudata->max_freq)
434 return -ENODEV;
435
436 cap_perf = READ_ONCE(cpudata->highest_perf);
b185c505 437 min_perf = READ_ONCE(cpudata->lowest_perf);
ec437d71
HR		 438 max_perf = cap_perf;
439
440 freqs.old = policy->cur;
441 freqs.new = target_freq;
442
443 des_perf = DIV_ROUND_CLOSEST(target_freq * cap_perf,
444 cpudata->max_freq);
445
446 cpufreq_freq_transition_begin(policy, &freqs);
447 amd_pstate_update(cpudata, min_perf, des_perf,
448 max_perf, false);
449 cpufreq_freq_transition_end(policy, &freqs, false);
450
451 return 0;
452}
453
1d215f03
HR		 454static void amd_pstate_adjust_perf(unsigned int cpu,
455 unsigned long _min_perf,
456 unsigned long target_perf,
457 unsigned long capacity)
458{
459 unsigned long max_perf, min_perf, des_perf,
460 cap_perf, lowest_nonlinear_perf;
461 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
462 struct amd_cpudata *cpudata = policy->driver_data;
463
464 cap_perf = READ_ONCE(cpudata->highest_perf);
465 lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf);
466
467 des_perf = cap_perf;
468 if (target_perf < capacity)
469 des_perf = DIV_ROUND_UP(cap_perf * target_perf, capacity);
470
471 min_perf = READ_ONCE(cpudata->highest_perf);
472 if (_min_perf < capacity)
473 min_perf = DIV_ROUND_UP(cap_perf * _min_perf, capacity);
474
475 if (min_perf < lowest_nonlinear_perf)
476 min_perf = lowest_nonlinear_perf;
477
478 max_perf = cap_perf;
479 if (max_perf < min_perf)
480 max_perf = min_perf;
481
1d215f03 482 amd_pstate_update(cpudata, min_perf, des_perf, max_perf, true);
4f3085f8 483 cpufreq_cpu_put(policy);
1d215f03
HR		 484}
485
ec437d71
HR		 486static int amd_get_min_freq(struct amd_cpudata *cpudata)
487{
488 struct cppc_perf_caps cppc_perf;
489
490 int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
491 if (ret)
492 return ret;
493
494 /* Switch to khz */
495 return cppc_perf.lowest_freq * 1000;
496}
497
498static int amd_get_max_freq(struct amd_cpudata *cpudata)
499{
500 struct cppc_perf_caps cppc_perf;
501 u32 max_perf, max_freq, nominal_freq, nominal_perf;
502 u64 boost_ratio;
503
504 int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
505 if (ret)
506 return ret;
507
508 nominal_freq = cppc_perf.nominal_freq;
509 nominal_perf = READ_ONCE(cpudata->nominal_perf);
510 max_perf = READ_ONCE(cpudata->highest_perf);
511
512 boost_ratio = div_u64(max_perf << SCHED_CAPACITY_SHIFT,
513 nominal_perf);
514
515 max_freq = nominal_freq * boost_ratio >> SCHED_CAPACITY_SHIFT;
516
517 /* Switch to khz */
518 return max_freq * 1000;
519}
520
521static int amd_get_nominal_freq(struct amd_cpudata *cpudata)
522{
523 struct cppc_perf_caps cppc_perf;
524
525 int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
526 if (ret)
527 return ret;
528
529 /* Switch to khz */
530 return cppc_perf.nominal_freq * 1000;
531}
532
533static int amd_get_lowest_nonlinear_freq(struct amd_cpudata *cpudata)
534{
535 struct cppc_perf_caps cppc_perf;
536 u32 lowest_nonlinear_freq, lowest_nonlinear_perf,
537 nominal_freq, nominal_perf;
538 u64 lowest_nonlinear_ratio;
539
540 int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
541 if (ret)
542 return ret;
543
544 nominal_freq = cppc_perf.nominal_freq;
545 nominal_perf = READ_ONCE(cpudata->nominal_perf);
546
547 lowest_nonlinear_perf = cppc_perf.lowest_nonlinear_perf;
548
549 lowest_nonlinear_ratio = div_u64(lowest_nonlinear_perf << SCHED_CAPACITY_SHIFT,
550 nominal_perf);
551
552 lowest_nonlinear_freq = nominal_freq * lowest_nonlinear_ratio >> SCHED_CAPACITY_SHIFT;
553
554 /* Switch to khz */
555 return lowest_nonlinear_freq * 1000;
556}
557
41271016
HR		 558static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
559{
560 struct amd_cpudata *cpudata = policy->driver_data;
561 int ret;
562
563 if (!cpudata->boost_supported) {
564 pr_err("Boost mode is not supported by this processor or SBIOS\n");
565 return -EINVAL;
566 }
567
568 if (state)
569 policy->cpuinfo.max_freq = cpudata->max_freq;
570 else
571 policy->cpuinfo.max_freq = cpudata->nominal_freq;
572
573 policy->max = policy->cpuinfo.max_freq;
574
575 ret = freq_qos_update_request(&cpudata->req[1],
576 policy->cpuinfo.max_freq);
577 if (ret < 0)
578 return ret;
579
580 return 0;
581}
582
583static void amd_pstate_boost_init(struct amd_cpudata *cpudata)
584{
585 u32 highest_perf, nominal_perf;
586
587 highest_perf = READ_ONCE(cpudata->highest_perf);
588 nominal_perf = READ_ONCE(cpudata->nominal_perf);
589
590 if (highest_perf <= nominal_perf)
591 return;
592
593 cpudata->boost_supported = true;
ffa5096a 594 current_pstate_driver->boost_enabled = true;
41271016
HR		 595}
596
919f4557
WK		 597static void amd_perf_ctl_reset(unsigned int cpu)
598{
599 wrmsrl_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
600}
601
ec437d71
HR		 602static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
603{
604 int min_freq, max_freq, nominal_freq, lowest_nonlinear_freq, ret;
605 struct device *dev;
606 struct amd_cpudata *cpudata;
607
919f4557
WK		 608 /*
609 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
610 * which is ideal for initialization process.
611 */
612 amd_perf_ctl_reset(policy->cpu);
ec437d71
HR		 613 dev = get_cpu_device(policy->cpu);
614 if (!dev)
615 return -ENODEV;
616
617 cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
618 if (!cpudata)
619 return -ENOMEM;
620
621 cpudata->cpu = policy->cpu;
622
623 ret = amd_pstate_init_perf(cpudata);
624 if (ret)
41271016 625 goto free_cpudata1;
ec437d71
HR		 626
627 min_freq = amd_get_min_freq(cpudata);
628 max_freq = amd_get_max_freq(cpudata);
629 nominal_freq = amd_get_nominal_freq(cpudata);
630 lowest_nonlinear_freq = amd_get_lowest_nonlinear_freq(cpudata);
631
632 if (min_freq < 0 || max_freq < 0 || min_freq > max_freq) {
633 dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n",
634 min_freq, max_freq);
635 ret = -EINVAL;
41271016 636 goto free_cpudata1;
ec437d71
HR		 637 }
638
639 policy->cpuinfo.transition_latency = AMD_PSTATE_TRANSITION_LATENCY;
640 policy->transition_delay_us = AMD_PSTATE_TRANSITION_DELAY;
641
642 policy->min = min_freq;
643 policy->max = max_freq;
644
645 policy->cpuinfo.min_freq = min_freq;
646 policy->cpuinfo.max_freq = max_freq;
647
648 /* It will be updated by governor */
649 policy->cur = policy->cpuinfo.min_freq;
650
e059c184
HR		 651 if (boot_cpu_has(X86_FEATURE_CPPC))
652 policy->fast_switch_possible = true;
1d215f03 653
41271016
HR		 654 ret = freq_qos_add_request(&policy->constraints, &cpudata->req[0],
655 FREQ_QOS_MIN, policy->cpuinfo.min_freq);
656 if (ret < 0) {
657 dev_err(dev, "Failed to add min-freq constraint (%d)\n", ret);
658 goto free_cpudata1;
659 }
660
661 ret = freq_qos_add_request(&policy->constraints, &cpudata->req[1],
662 FREQ_QOS_MAX, policy->cpuinfo.max_freq);
663 if (ret < 0) {
664 dev_err(dev, "Failed to add max-freq constraint (%d)\n", ret);
665 goto free_cpudata2;
666 }
667
ec437d71
HR		 668 /* Initial processor data capability frequencies */
669 cpudata->max_freq = max_freq;
670 cpudata->min_freq = min_freq;
671 cpudata->nominal_freq = nominal_freq;
672 cpudata->lowest_nonlinear_freq = lowest_nonlinear_freq;
673
674 policy->driver_data = cpudata;
675
41271016 676 amd_pstate_boost_init(cpudata);
abd61c08
PY		 677 if (!current_pstate_driver->adjust_perf)
678 current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
41271016 679
ec437d71
HR		 680 return 0;
681
41271016
HR		 682free_cpudata2:
683 freq_qos_remove_request(&cpudata->req[0]);
684free_cpudata1:
ec437d71
HR		 685 kfree(cpudata);
686 return ret;
687}
688
689static int amd_pstate_cpu_exit(struct cpufreq_policy *policy)
690{
4f59540c 691 struct amd_cpudata *cpudata = policy->driver_data;
ec437d71 692
41271016
HR		 693 freq_qos_remove_request(&cpudata->req[1]);
694 freq_qos_remove_request(&cpudata->req[0]);
ec437d71
HR		 695 kfree(cpudata);
696
697 return 0;
698}
699
b376471f
JS		 700static int amd_pstate_cpu_resume(struct cpufreq_policy *policy)
701{
702 int ret;
703
704 ret = amd_pstate_enable(true);
705 if (ret)
706 pr_err("failed to enable amd-pstate during resume, return %d\n", ret);
707
708 return ret;
709}
710
711static int amd_pstate_cpu_suspend(struct cpufreq_policy *policy)
712{
713 int ret;
714
715 ret = amd_pstate_enable(false);
716 if (ret)
717 pr_err("failed to disable amd-pstate during suspend, return %d\n", ret);
718
719 return ret;
720}
721
ec4e3326
HR		 722/* Sysfs attributes */
723
724/*
725 * This frequency is to indicate the maximum hardware frequency.
726 * If boost is not active but supported, the frequency will be larger than the
727 * one in cpuinfo.
728 */
729static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy,
730 char *buf)
731{
732 int max_freq;
4f59540c 733 struct amd_cpudata *cpudata = policy->driver_data;
ec4e3326
HR		 734
735 max_freq = amd_get_max_freq(cpudata);
736 if (max_freq < 0)
737 return max_freq;
738
3ec32b6d 739 return sysfs_emit(buf, "%u\n", max_freq);
ec4e3326
HR		 740}
741
742static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *policy,
743 char *buf)
744{
745 int freq;
4f59540c 746 struct amd_cpudata *cpudata = policy->driver_data;
ec4e3326
HR		 747
748 freq = amd_get_lowest_nonlinear_freq(cpudata);
749 if (freq < 0)
750 return freq;
751
3ec32b6d 752 return sysfs_emit(buf, "%u\n", freq);
ec4e3326
HR		 753}
754
3ad7fde1
HR		 755/*
756 * In some of ASICs, the highest_perf is not the one in the _CPC table, so we
757 * need to expose it to sysfs.
758 */
759static ssize_t show_amd_pstate_highest_perf(struct cpufreq_policy *policy,
760 char *buf)
761{
762 u32 perf;
763 struct amd_cpudata *cpudata = policy->driver_data;
764
765 perf = READ_ONCE(cpudata->highest_perf);
766
3ec32b6d 767 return sysfs_emit(buf, "%u\n", perf);
3ad7fde1
HR		 768}
769
ffa5096a
PY		 770static ssize_t show_energy_performance_available_preferences(
771 struct cpufreq_policy *policy, char *buf)
772{
773 int i = 0;
774 int offset = 0;
775
776 while (energy_perf_strings[i] != NULL)
777 offset += sysfs_emit_at(buf, offset, "%s ", energy_perf_strings[i++]);
778
779 sysfs_emit_at(buf, offset, "\n");
780
781 return offset;
782}
783
784static ssize_t store_energy_performance_preference(
785 struct cpufreq_policy *policy, const char *buf, size_t count)
786{
787 struct amd_cpudata *cpudata = policy->driver_data;
788 char str_preference[21];
789 ssize_t ret;
790
791 ret = sscanf(buf, "%20s", str_preference);
792 if (ret != 1)
793 return -EINVAL;
794
795 ret = match_string(energy_perf_strings, -1, str_preference);
796 if (ret < 0)
797 return -EINVAL;
798
799 mutex_lock(&amd_pstate_limits_lock);
800 ret = amd_pstate_set_energy_pref_index(cpudata, ret);
801 mutex_unlock(&amd_pstate_limits_lock);
802
803 return ret ?: count;
804}
805
806static ssize_t show_energy_performance_preference(
807 struct cpufreq_policy *policy, char *buf)
808{
809 struct amd_cpudata *cpudata = policy->driver_data;
810 int preference;
811
812 preference = amd_pstate_get_energy_pref_index(cpudata);
813 if (preference < 0)
814 return preference;
815
816 return sysfs_emit(buf, "%s\n", energy_perf_strings[preference]);
817}
818
abd61c08
PY		 819static ssize_t amd_pstate_show_status(char *buf)
820{
821 if (!current_pstate_driver)
822 return sysfs_emit(buf, "disable\n");
823
824 return sysfs_emit(buf, "%s\n", amd_pstate_mode_string[cppc_state]);
825}
826
827static void amd_pstate_driver_cleanup(void)
828{
829 current_pstate_driver = NULL;
830}
831
832static int amd_pstate_update_status(const char *buf, size_t size)
833{
dd329e1e 834 int ret = 0;
abd61c08
PY		 835 int mode_idx;
836
837 if (size > 7 || size < 6)
838 return -EINVAL;
839 mode_idx = get_mode_idx_from_str(buf, size);
840
841 switch(mode_idx) {
842 case AMD_PSTATE_DISABLE:
4654e9f9
WK		 843 if (current_pstate_driver) {
844 cpufreq_unregister_driver(current_pstate_driver);
845 amd_pstate_driver_cleanup();
846 }
abd61c08
PY		 847 break;
848 case AMD_PSTATE_PASSIVE:
849 if (current_pstate_driver) {
850 if (current_pstate_driver == &amd_pstate_driver)
851 return 0;
852 cpufreq_unregister_driver(current_pstate_driver);
abd61c08
PY		 853 }
854
4654e9f9
WK		 855 current_pstate_driver = &amd_pstate_driver;
856 cppc_state = AMD_PSTATE_PASSIVE;
abd61c08
PY		 857 ret = cpufreq_register_driver(current_pstate_driver);
858 break;
859 case AMD_PSTATE_ACTIVE:
860 if (current_pstate_driver) {
861 if (current_pstate_driver == &amd_pstate_epp_driver)
862 return 0;
863 cpufreq_unregister_driver(current_pstate_driver);
abd61c08
PY		 864 }
865
4654e9f9
WK		 866 current_pstate_driver = &amd_pstate_epp_driver;
867 cppc_state = AMD_PSTATE_ACTIVE;
abd61c08
PY		 868 ret = cpufreq_register_driver(current_pstate_driver);
869 break;
870 default:
871 ret = -EINVAL;
872 break;
873 }
874
875 return ret;
876}
877
878static ssize_t show_status(struct kobject *kobj,
879 struct kobj_attribute *attr, char *buf)
880{
881 ssize_t ret;
882
883 mutex_lock(&amd_pstate_driver_lock);
884 ret = amd_pstate_show_status(buf);
885 mutex_unlock(&amd_pstate_driver_lock);
886
887 return ret;
888}
889
890static ssize_t store_status(struct kobject *a, struct kobj_attribute *b,
891 const char *buf, size_t count)
892{
893 char *p = memchr(buf, '\n', count);
894 int ret;
895
896 mutex_lock(&amd_pstate_driver_lock);
897 ret = amd_pstate_update_status(buf, p ? p - buf : count);
898 mutex_unlock(&amd_pstate_driver_lock);
899
900 return ret < 0 ? ret : count;
901}
902
ec4e3326
HR		 903cpufreq_freq_attr_ro(amd_pstate_max_freq);
904cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
905
3ad7fde1 906cpufreq_freq_attr_ro(amd_pstate_highest_perf);
ffa5096a
PY		 907cpufreq_freq_attr_rw(energy_performance_preference);
908cpufreq_freq_attr_ro(energy_performance_available_preferences);
abd61c08 909define_one_global_rw(status);
3ad7fde1 910
ec4e3326
HR		 911static struct freq_attr *amd_pstate_attr[] = {
912 &amd_pstate_max_freq,
913 &amd_pstate_lowest_nonlinear_freq,
3ad7fde1 914 &amd_pstate_highest_perf,
ec4e3326
HR		 915 NULL,
916};
917
ffa5096a
PY		 918static struct freq_attr *amd_pstate_epp_attr[] = {
919 &amd_pstate_max_freq,
920 &amd_pstate_lowest_nonlinear_freq,
921 &amd_pstate_highest_perf,
922 &energy_performance_preference,
923 &energy_performance_available_preferences,
924 NULL,
925};
926
abd61c08
PY		 927static struct attribute *pstate_global_attributes[] = {
928 &status.attr,
929 NULL
930};
931
932static const struct attribute_group amd_pstate_global_attr_group = {
933 .attrs = pstate_global_attributes,
934};
935
ffa5096a
PY		 936static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
937{
938 int min_freq, max_freq, nominal_freq, lowest_nonlinear_freq, ret;
939 struct amd_cpudata *cpudata;
940 struct device *dev;
ffa5096a
PY		 941 u64 value;
942
943 /*
944 * Resetting PERF_CTL_MSR will put the CPU in P0 frequency,
945 * which is ideal for initialization process.
946 */
947 amd_perf_ctl_reset(policy->cpu);
948 dev = get_cpu_device(policy->cpu);
949 if (!dev)
7cca9a98 950 return -ENODEV;
ffa5096a
PY		 951
952 cpudata = kzalloc(sizeof(*cpudata), GFP_KERNEL);
953 if (!cpudata)
954 return -ENOMEM;
955
956 cpudata->cpu = policy->cpu;
957 cpudata->epp_policy = 0;
958
7cca9a98
AB		 959 ret = amd_pstate_init_perf(cpudata);
960 if (ret)
ffa5096a
PY		 961 goto free_cpudata1;
962
963 min_freq = amd_get_min_freq(cpudata);
964 max_freq = amd_get_max_freq(cpudata);
965 nominal_freq = amd_get_nominal_freq(cpudata);
966 lowest_nonlinear_freq = amd_get_lowest_nonlinear_freq(cpudata);
967 if (min_freq < 0 || max_freq < 0 || min_freq > max_freq) {
968 dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n",
969 min_freq, max_freq);
970 ret = -EINVAL;
971 goto free_cpudata1;
972 }
973
974 policy->cpuinfo.min_freq = min_freq;
975 policy->cpuinfo.max_freq = max_freq;
976 /* It will be updated by governor */
977 policy->cur = policy->cpuinfo.min_freq;
978
979 /* Initial processor data capability frequencies */
980 cpudata->max_freq = max_freq;
981 cpudata->min_freq = min_freq;
982 cpudata->nominal_freq = nominal_freq;
983 cpudata->lowest_nonlinear_freq = lowest_nonlinear_freq;
984
985 policy->driver_data = cpudata;
986
987 cpudata->epp_cached = amd_pstate_get_epp(cpudata, 0);
988
989 policy->min = policy->cpuinfo.min_freq;
990 policy->max = policy->cpuinfo.max_freq;
991
992 /*
993 * Set the policy to powersave to provide a valid fallback value in case
994 * the default cpufreq governor is neither powersave nor performance.
995 */
996 policy->policy = CPUFREQ_POLICY_POWERSAVE;
997
998 if (boot_cpu_has(X86_FEATURE_CPPC)) {
999 policy->fast_switch_possible = true;
1000 ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &value);
1001 if (ret)
1002 return ret;
1003 WRITE_ONCE(cpudata->cppc_req_cached, value);
1004
1005 ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1, &value);
1006 if (ret)
1007 return ret;
1008 WRITE_ONCE(cpudata->cppc_cap1_cached, value);
1009 }
1010 amd_pstate_boost_init(cpudata);
1011
1012 return 0;
1013
1014free_cpudata1:
1015 kfree(cpudata);
1016 return ret;
1017}
1018
1019static int amd_pstate_epp_cpu_exit(struct cpufreq_policy *policy)
1020{
1021 pr_debug("CPU %d exiting\n", policy->cpu);
1022 policy->fast_switch_possible = false;
1023 return 0;
1024}
1025
1026static void amd_pstate_epp_init(unsigned int cpu)
1027{
1028 struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
1029 struct amd_cpudata *cpudata = policy->driver_data;
1030 u32 max_perf, min_perf;
1031 u64 value;
1032 s16 epp;
1033
1034 max_perf = READ_ONCE(cpudata->highest_perf);
1035 min_perf = READ_ONCE(cpudata->lowest_perf);
1036
1037 value = READ_ONCE(cpudata->cppc_req_cached);
1038
1039 if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1040 min_perf = max_perf;
1041
1042 /* Initial min/max values for CPPC Performance Controls Register */
1043 value &= ~AMD_CPPC_MIN_PERF(~0L);
1044 value |= AMD_CPPC_MIN_PERF(min_perf);
1045
1046 value &= ~AMD_CPPC_MAX_PERF(~0L);
1047 value |= AMD_CPPC_MAX_PERF(max_perf);
1048
1049 /* CPPC EPP feature require to set zero to the desire perf bit */
1050 value &= ~AMD_CPPC_DES_PERF(~0L);
1051 value |= AMD_CPPC_DES_PERF(0);
1052
1053 if (cpudata->epp_policy == cpudata->policy)
1054 goto skip_epp;
1055
1056 cpudata->epp_policy = cpudata->policy;
1057
6e9d1212
WK		 1058 /* Get BIOS pre-defined epp value */
1059 epp = amd_pstate_get_epp(cpudata, value);
1060 if (epp < 0) {
1061 /**
1062 * This return value can only be negative for shared_memory
1063 * systems where EPP register read/write not supported.
1064 */
1065 goto skip_epp;
ffa5096a 1066 }
6e9d1212
WK		 1067
1068 if (cpudata->policy == CPUFREQ_POLICY_PERFORMANCE)
1069 epp = 0;
1070
ffa5096a
PY		 1071 /* Set initial EPP value */
1072 if (boot_cpu_has(X86_FEATURE_CPPC)) {
1073 value &= ~GENMASK_ULL(31, 24);
1074 value |= (u64)epp << 24;
1075 }
1076
6e9d1212 1077 WRITE_ONCE(cpudata->cppc_req_cached, value);
7cca9a98 1078 amd_pstate_set_epp(cpudata, epp);
ffa5096a 1079skip_epp:
ffa5096a
PY		 1080 cpufreq_cpu_put(policy);
1081}
1082
1083static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy)
1084{
1085 struct amd_cpudata *cpudata = policy->driver_data;
1086
1087 if (!policy->cpuinfo.max_freq)
1088 return -ENODEV;
1089
1090 pr_debug("set_policy: cpuinfo.max %u policy->max %u\n",
1091 policy->cpuinfo.max_freq, policy->max);
1092
1093 cpudata->policy = policy->policy;
1094
1095 amd_pstate_epp_init(policy->cpu);
1096
1097 return 0;
1098}
1099
d4da12f8
PY		 1100static void amd_pstate_epp_reenable(struct amd_cpudata *cpudata)
1101{
1102 struct cppc_perf_ctrls perf_ctrls;
1103 u64 value, max_perf;
1104 int ret;
1105
1106 ret = amd_pstate_enable(true);
1107 if (ret)
1108 pr_err("failed to enable amd pstate during resume, return %d\n", ret);
1109
1110 value = READ_ONCE(cpudata->cppc_req_cached);
1111 max_perf = READ_ONCE(cpudata->highest_perf);
1112
1113 if (boot_cpu_has(X86_FEATURE_CPPC)) {
1114 wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
1115 } else {
1116 perf_ctrls.max_perf = max_perf;
1117 perf_ctrls.energy_perf = AMD_CPPC_ENERGY_PERF_PREF(cpudata->epp_cached);
1118 cppc_set_perf(cpudata->cpu, &perf_ctrls);
1119 }
1120}
1121
1122static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy)
1123{
1124 struct amd_cpudata *cpudata = policy->driver_data;
1125
1126 pr_debug("AMD CPU Core %d going online\n", cpudata->cpu);
1127
1128 if (cppc_state == AMD_PSTATE_ACTIVE) {
1129 amd_pstate_epp_reenable(cpudata);
1130 cpudata->suspended = false;
1131 }
1132
1133 return 0;
1134}
1135
1136static void amd_pstate_epp_offline(struct cpufreq_policy *policy)
1137{
1138 struct amd_cpudata *cpudata = policy->driver_data;
1139 struct cppc_perf_ctrls perf_ctrls;
1140 int min_perf;
1141 u64 value;
1142
1143 min_perf = READ_ONCE(cpudata->lowest_perf);
1144 value = READ_ONCE(cpudata->cppc_req_cached);
1145
1146 mutex_lock(&amd_pstate_limits_lock);
1147 if (boot_cpu_has(X86_FEATURE_CPPC)) {
1148 cpudata->epp_policy = CPUFREQ_POLICY_UNKNOWN;
1149
1150 /* Set max perf same as min perf */
1151 value &= ~AMD_CPPC_MAX_PERF(~0L);
1152 value |= AMD_CPPC_MAX_PERF(min_perf);
1153 value &= ~AMD_CPPC_MIN_PERF(~0L);
1154 value |= AMD_CPPC_MIN_PERF(min_perf);
1155 wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
1156 } else {
1157 perf_ctrls.desired_perf = 0;
1158 perf_ctrls.max_perf = min_perf;
1159 perf_ctrls.energy_perf = AMD_CPPC_ENERGY_PERF_PREF(HWP_EPP_BALANCE_POWERSAVE);
1160 cppc_set_perf(cpudata->cpu, &perf_ctrls);
1161 }
1162 mutex_unlock(&amd_pstate_limits_lock);
1163}
1164
1165static int amd_pstate_epp_cpu_offline(struct cpufreq_policy *policy)
1166{
1167 struct amd_cpudata *cpudata = policy->driver_data;
1168
1169 pr_debug("AMD CPU Core %d going offline\n", cpudata->cpu);
1170
1171 if (cpudata->suspended)
1172 return 0;
1173
1174 if (cppc_state == AMD_PSTATE_ACTIVE)
1175 amd_pstate_epp_offline(policy);
1176
1177 return 0;
1178}
1179
ffa5096a
PY		 1180static int amd_pstate_epp_verify_policy(struct cpufreq_policy_data *policy)
1181{
1182 cpufreq_verify_within_cpu_limits(policy);
1183 pr_debug("policy_max =%d, policy_min=%d\n", policy->max, policy->min);
1184 return 0;
1185}
1186
50ddd2f7
PY		 1187static int amd_pstate_epp_suspend(struct cpufreq_policy *policy)
1188{
1189 struct amd_cpudata *cpudata = policy->driver_data;
1190 int ret;
1191
1192 /* avoid suspending when EPP is not enabled */
1193 if (cppc_state != AMD_PSTATE_ACTIVE)
1194 return 0;
1195
1196 /* set this flag to avoid setting core offline*/
1197 cpudata->suspended = true;
1198
1199 /* disable CPPC in lowlevel firmware */
1200 ret = amd_pstate_enable(false);
1201 if (ret)
1202 pr_err("failed to suspend, return %d\n", ret);
1203
1204 return 0;
1205}
1206
1207static int amd_pstate_epp_resume(struct cpufreq_policy *policy)
1208{
1209 struct amd_cpudata *cpudata = policy->driver_data;
1210
1211 if (cpudata->suspended) {
1212 mutex_lock(&amd_pstate_limits_lock);
1213
1214 /* enable amd pstate from suspend state*/
1215 amd_pstate_epp_reenable(cpudata);
1216
1217 mutex_unlock(&amd_pstate_limits_lock);
1218
1219 cpudata->suspended = false;
1220 }
1221
1222 return 0;
1223}
1224
ec437d71
HR		 1225static struct cpufreq_driver amd_pstate_driver = {
1226 .flags = CPUFREQ_CONST_LOOPS | CPUFREQ_NEED_UPDATE_LIMITS,
1227 .verify = amd_pstate_verify,
1228 .target = amd_pstate_target,
1229 .init = amd_pstate_cpu_init,
1230 .exit = amd_pstate_cpu_exit,
b376471f
JS		 1231 .suspend = amd_pstate_cpu_suspend,
1232 .resume = amd_pstate_cpu_resume,
41271016 1233 .set_boost = amd_pstate_set_boost,
ec437d71 1234 .name = "amd-pstate",
d8bee41d 1235 .attr = amd_pstate_attr,
ec437d71
HR		 1236};
1237
ffa5096a
PY		 1238static struct cpufreq_driver amd_pstate_epp_driver = {
1239 .flags = CPUFREQ_CONST_LOOPS,
1240 .verify = amd_pstate_epp_verify_policy,
1241 .setpolicy = amd_pstate_epp_set_policy,
1242 .init = amd_pstate_epp_cpu_init,
1243 .exit = amd_pstate_epp_cpu_exit,
d4da12f8
PY		 1244 .offline = amd_pstate_epp_cpu_offline,
1245 .online = amd_pstate_epp_cpu_online,
50ddd2f7
PY		 1246 .suspend = amd_pstate_epp_suspend,
1247 .resume = amd_pstate_epp_resume,
ffa5096a
PY		 1248 .name = "amd_pstate_epp",
1249 .attr = amd_pstate_epp_attr,
1250};
1251
ec437d71
HR		 1252static int __init amd_pstate_init(void)
1253{
1254 int ret;
1255
1256 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
1257 return -ENODEV;
202e683d
PY		 1258 /*
1259 * by default the pstate driver is disabled to load
1260 * enable the amd_pstate passive mode driver explicitly
36c5014e 1261 * with amd_pstate=passive or other modes in kernel command line
202e683d 1262 */
36c5014e 1263 if (cppc_state == AMD_PSTATE_DISABLE) {
7af78020 1264 pr_info("driver load is disabled, boot with specific mode to enable this\n");
202e683d
PY		 1265 return -ENODEV;
1266 }
ec437d71
HR		 1267
1268 if (!acpi_cpc_valid()) {
a2a9d185 1269 pr_warn_once("the _CPC object is not present in SBIOS or ACPI disabled\n");
ec437d71
HR		 1270 return -ENODEV;
1271 }
1272
1273 /* don't keep reloading if cpufreq_driver exists */
1274 if (cpufreq_get_current_driver())
1275 return -EEXIST;
1276
1277 /* capability check */
e059c184
HR		 1278 if (boot_cpu_has(X86_FEATURE_CPPC)) {
1279 pr_debug("AMD CPPC MSR based functionality is supported\n");
ffa5096a
PY		 1280 if (cppc_state == AMD_PSTATE_PASSIVE)
1281 current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
202e683d
PY		 1282 } else {
1283 pr_debug("AMD CPPC shared memory based functionality is supported\n");
e059c184
HR		 1284 static_call_update(amd_pstate_enable, cppc_enable);
1285 static_call_update(amd_pstate_init_perf, cppc_init_perf);
1286 static_call_update(amd_pstate_update_perf, cppc_update_perf);
ec437d71
HR		 1287 }
1288
1289 /* enable amd pstate feature */
1290 ret = amd_pstate_enable(true);
1291 if (ret) {
ffa5096a 1292 pr_err("failed to enable with return %d\n", ret);
ec437d71
HR		 1293 return ret;
1294 }
1295
ffa5096a 1296 ret = cpufreq_register_driver(current_pstate_driver);
ec437d71 1297 if (ret)
ffa5096a 1298 pr_err("failed to register with return %d\n", ret);
ec437d71 1299
abd61c08
PY		 1300 amd_pstate_kobj = kobject_create_and_add("amd_pstate", &cpu_subsys.dev_root->kobj);
1301 if (!amd_pstate_kobj) {
1302 ret = -EINVAL;
1303 pr_err("global sysfs registration failed.\n");
1304 goto kobject_free;
1305 }
1306
1307 ret = sysfs_create_group(amd_pstate_kobj, &amd_pstate_global_attr_group);
1308 if (ret) {
1309 pr_err("sysfs attribute export failed with error %d.\n", ret);
1310 goto global_attr_free;
1311 }
1312
1313 return ret;
1314
1315global_attr_free:
1316 kobject_put(amd_pstate_kobj);
1317kobject_free:
1318 cpufreq_unregister_driver(current_pstate_driver);
ec437d71
HR		 1319 return ret;
1320}
456ca88d 1321device_initcall(amd_pstate_init);
ec437d71 1322
202e683d
PY		 1323static int __init amd_pstate_param(char *str)
1324{
36c5014e
WK		 1325 size_t size;
1326 int mode_idx;
1327
202e683d
PY		 1328 if (!str)
1329 return -EINVAL;
1330
36c5014e
WK		 1331 size = strlen(str);
1332 mode_idx = get_mode_idx_from_str(str, size);
202e683d 1333
36c5014e
WK		 1334 if (mode_idx >= AMD_PSTATE_DISABLE && mode_idx < AMD_PSTATE_MAX) {
1335 cppc_state = mode_idx;
1336 if (cppc_state == AMD_PSTATE_DISABLE)
1337 pr_info("driver is explicitly disabled\n");
1338
ffa5096a
PY		 1339 if (cppc_state == AMD_PSTATE_ACTIVE)
1340 current_pstate_driver = &amd_pstate_epp_driver;
1341
1342 if (cppc_state == AMD_PSTATE_PASSIVE)
1343 current_pstate_driver = &amd_pstate_driver;
1344
36c5014e
WK		 1345 return 0;
1346 }
1347
1348 return -EINVAL;
202e683d
PY		 1349}
1350early_param("amd_pstate", amd_pstate_param);
1351
ec437d71
HR		 1352MODULE_AUTHOR("Huang Rui <ray.huang@amd.com>");
1353MODULE_DESCRIPTION("AMD Processor P-state Frequency Driver");
Linux 6.x block layer and io_uring tree(s)