drivers/cpufreq/intel_pstate.c

   1 /*
   2  * intel_pstate.c: Native P state management for Intel processors
   3  *
   4  * (C) Copyright 2012 Intel Corporation
   5  * Author: Dirk Brandewie <dirk.j.brandewie@intel.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; version 2
  10  * of the License.
  11  */
  12
  13 #include <linux/kernel.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/module.h>
  16 #include <linux/ktime.h>
  17 #include <linux/hrtimer.h>
  18 #include <linux/tick.h>
  19 #include <linux/slab.h>
  20 #include <linux/sched.h>
  21 #include <linux/list.h>
  22 #include <linux/cpu.h>
  23 #include <linux/cpufreq.h>
  24 #include <linux/sysfs.h>
  25 #include <linux/types.h>
  26 #include <linux/fs.h>
  27 #include <linux/debugfs.h>
  28 #include <linux/acpi.h>
  29 #include <trace/events/power.h>
  30
  31 #include <asm/div64.h>
  32 #include <asm/msr.h>
  33 #include <asm/cpu_device_id.h>
  34
  35 #define SAMPLE_COUNT            3
  36
  37 #define BYT_RATIOS      0x66a
  38
  39 #define FRAC_BITS 8
  40 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  41 #define fp_toint(X) ((X) >> FRAC_BITS)
  42
  43 static inline int32_t mul_fp(int32_t x, int32_t y)
  44 {
  45         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  46 }
  47
  48 static inline int32_t div_fp(int32_t x, int32_t y)
  49 {
  50         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  51 }
  52
  53 struct sample {
  54         int32_t core_pct_busy;
  55         u64 aperf;
  56         u64 mperf;
  57         int freq;
  58 };
  59
  60 struct pstate_data {
  61         int     current_pstate;
  62         int     min_pstate;
  63         int     max_pstate;
  64         int     turbo_pstate;
  65 };
  66
  67 struct _pid {
  68         int setpoint;
  69         int32_t integral;
  70         int32_t p_gain;
  71         int32_t i_gain;
  72         int32_t d_gain;
  73         int deadband;
  74         int32_t last_err;
  75 };
  76
  77 struct cpudata {
  78         int cpu;
  79
  80         char name[64];
  81
  82         struct timer_list timer;
  83
  84         struct pstate_data pstate;
  85         struct _pid pid;
  86
  87         int min_pstate_count;
  88
  89         u64     prev_aperf;
  90         u64     prev_mperf;
  91         int     sample_ptr;
  92         struct sample samples[SAMPLE_COUNT];
  93 };
  94
  95 static struct cpudata **all_cpu_data;
  96 struct pstate_adjust_policy {
  97         int sample_rate_ms;
  98         int deadband;
  99         int setpoint;
 100         int p_gain_pct;
 101         int d_gain_pct;
 102         int i_gain_pct;
 103 };
 104
 105 struct pstate_funcs {
 106         int (*get_max)(void);
 107         int (*get_min)(void);
 108         int (*get_turbo)(void);
 109         void (*set)(int pstate);
 110 };
 111
 112 struct cpu_defaults {
 113         struct pstate_adjust_policy pid_policy;
 114         struct pstate_funcs funcs;
 115 };
 116
 117 static struct pstate_adjust_policy pid_params;
 118 static struct pstate_funcs pstate_funcs;
 119
 120 struct perf_limits {
 121         int no_turbo;
 122         int max_perf_pct;
 123         int min_perf_pct;
 124         int32_t max_perf;
 125         int32_t min_perf;
 126         int max_policy_pct;
 127         int max_sysfs_pct;
 128 };
 129
 130 static struct perf_limits limits = {
 131         .no_turbo = 0,
 132         .max_perf_pct = 100,
 133         .max_perf = int_tofp(1),
 134         .min_perf_pct = 0,
 135         .min_perf = 0,
 136         .max_policy_pct = 100,
 137         .max_sysfs_pct = 100,
 138 };
 139
 140 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 141                         int deadband, int integral) {
 142         pid->setpoint = setpoint;
 143         pid->deadband  = deadband;
 144         pid->integral  = int_tofp(integral);
 145         pid->last_err  = setpoint - busy;
 146 }
 147
 148 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 149 {
 150         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 151 }
 152
 153 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 154 {
 155         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 156 }
 157
 158 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 159 {
 160
 161         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 162 }
 163
 164 static signed int pid_calc(struct _pid *pid, int32_t busy)
 165 {
 166         signed int result;
 167         int32_t pterm, dterm, fp_error;
 168         int32_t integral_limit;
 169
 170         fp_error = int_tofp(pid->setpoint) - busy;
 171
 172         if (abs(fp_error) <= int_tofp(pid->deadband))
 173                 return 0;
 174
 175         pterm = mul_fp(pid->p_gain, fp_error);
 176
 177         pid->integral += fp_error;
 178
 179         /* limit the integral term */
 180         integral_limit = int_tofp(30);
 181         if (pid->integral > integral_limit)
 182                 pid->integral = integral_limit;
 183         if (pid->integral < -integral_limit)
 184                 pid->integral = -integral_limit;
 185
 186         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 187         pid->last_err = fp_error;
 188
 189         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 190
 191         return (signed int)fp_toint(result);
 192 }
 193
 194 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 195 {
 196         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 197         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 198         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 199
 200         pid_reset(&cpu->pid,
 201                 pid_params.setpoint,
 202                 100,
 203                 pid_params.deadband,
 204                 0);
 205 }
 206
 207 static inline void intel_pstate_reset_all_pid(void)
 208 {
 209         unsigned int cpu;
 210         for_each_online_cpu(cpu) {
 211                 if (all_cpu_data[cpu])
 212                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 213         }
 214 }
 215
 216 /************************** debugfs begin ************************/
 217 static int pid_param_set(void *data, u64 val)
 218 {
 219         *(u32 *)data = val;
 220         intel_pstate_reset_all_pid();
 221         return 0;
 222 }
 223 static int pid_param_get(void *data, u64 *val)
 224 {
 225         *val = *(u32 *)data;
 226         return 0;
 227 }
 228 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 229                         pid_param_set, "%llu\n");
 230
 231 struct pid_param {
 232         char *name;
 233         void *value;
 234 };
 235
 236 static struct pid_param pid_files[] = {
 237         {"sample_rate_ms", &pid_params.sample_rate_ms},
 238         {"d_gain_pct", &pid_params.d_gain_pct},
 239         {"i_gain_pct", &pid_params.i_gain_pct},
 240         {"deadband", &pid_params.deadband},
 241         {"setpoint", &pid_params.setpoint},
 242         {"p_gain_pct", &pid_params.p_gain_pct},
 243         {NULL, NULL}
 244 };
 245
 246 static struct dentry *debugfs_parent;
 247 static void intel_pstate_debug_expose_params(void)
 248 {
 249         int i = 0;
 250
 251         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 252         if (IS_ERR_OR_NULL(debugfs_parent))
 253                 return;
 254         while (pid_files[i].name) {
 255                 debugfs_create_file(pid_files[i].name, 0660,
 256                                 debugfs_parent, pid_files[i].value,
 257                                 &fops_pid_param);
 258                 i++;
 259         }
 260 }
 261
 262 /************************** debugfs end ************************/
 263
 264 /************************** sysfs begin ************************/
 265 #define show_one(file_name, object)                                     \
 266         static ssize_t show_##file_name                                 \
 267         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 268         {                                                               \
 269                 return sprintf(buf, "%u\n", limits.object);             \
 270         }
 271
 272 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 273                                 const char *buf, size_t count)
 274 {
 275         unsigned int input;
 276         int ret;
 277         ret = sscanf(buf, "%u", &input);
 278         if (ret != 1)
 279                 return -EINVAL;
 280         limits.no_turbo = clamp_t(int, input, 0 , 1);
 281
 282         return count;
 283 }
 284
 285 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 286                                 const char *buf, size_t count)
 287 {
 288         unsigned int input;
 289         int ret;
 290         ret = sscanf(buf, "%u", &input);
 291         if (ret != 1)
 292                 return -EINVAL;
 293
 294         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 295         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 296         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 297         return count;
 298 }
 299
 300 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 301                                 const char *buf, size_t count)
 302 {
 303         unsigned int input;
 304         int ret;
 305         ret = sscanf(buf, "%u", &input);
 306         if (ret != 1)
 307                 return -EINVAL;
 308         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 309         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 310
 311         return count;
 312 }
 313
 314 show_one(no_turbo, no_turbo);
 315 show_one(max_perf_pct, max_perf_pct);
 316 show_one(min_perf_pct, min_perf_pct);
 317
 318 define_one_global_rw(no_turbo);
 319 define_one_global_rw(max_perf_pct);
 320 define_one_global_rw(min_perf_pct);
 321
 322 static struct attribute *intel_pstate_attributes[] = {
 323         &no_turbo.attr,
 324         &max_perf_pct.attr,
 325         &min_perf_pct.attr,
 326         NULL
 327 };
 328
 329 static struct attribute_group intel_pstate_attr_group = {
 330         .attrs = intel_pstate_attributes,
 331 };
 332 static struct kobject *intel_pstate_kobject;
 333
 334 static void intel_pstate_sysfs_expose_params(void)
 335 {
 336         int rc;
 337
 338         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 339                                                 &cpu_subsys.dev_root->kobj);
 340         BUG_ON(!intel_pstate_kobject);
 341         rc = sysfs_create_group(intel_pstate_kobject,
 342                                 &intel_pstate_attr_group);
 343         BUG_ON(rc);
 344 }
 345
 346 /************************** sysfs end ************************/
 347 static int byt_get_min_pstate(void)
 348 {
 349         u64 value;
 350         rdmsrl(BYT_RATIOS, value);
 351         return value & 0xFF;
 352 }
 353
 354 static int byt_get_max_pstate(void)
 355 {
 356         u64 value;
 357         rdmsrl(BYT_RATIOS, value);
 358         return (value >> 16) & 0xFF;
 359 }
 360
 361 static int core_get_min_pstate(void)
 362 {
 363         u64 value;
 364         rdmsrl(MSR_PLATFORM_INFO, value);
 365         return (value >> 40) & 0xFF;
 366 }
 367
 368 static int core_get_max_pstate(void)
 369 {
 370         u64 value;
 371         rdmsrl(MSR_PLATFORM_INFO, value);
 372         return (value >> 8) & 0xFF;
 373 }
 374
 375 static int core_get_turbo_pstate(void)
 376 {
 377         u64 value;
 378         int nont, ret;
 379         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 380         nont = core_get_max_pstate();
 381         ret = ((value) & 255);
 382         if (ret <= nont)
 383                 ret = nont;
 384         return ret;
 385 }
 386
 387 static void core_set_pstate(int pstate)
 388 {
 389         u64 val;
 390
 391         val = pstate << 8;
 392         if (limits.no_turbo)
 393                 val |= (u64)1 << 32;
 394
 395         wrmsrl(MSR_IA32_PERF_CTL, val);
 396 }
 397
 398 static struct cpu_defaults core_params = {
 399         .pid_policy = {
 400                 .sample_rate_ms = 10,
 401                 .deadband = 0,
 402                 .setpoint = 97,
 403                 .p_gain_pct = 20,
 404                 .d_gain_pct = 0,
 405                 .i_gain_pct = 0,
 406         },
 407         .funcs = {
 408                 .get_max = core_get_max_pstate,
 409                 .get_min = core_get_min_pstate,
 410                 .get_turbo = core_get_turbo_pstate,
 411                 .set = core_set_pstate,
 412         },
 413 };
 414
 415 static struct cpu_defaults byt_params = {
 416         .pid_policy = {
 417                 .sample_rate_ms = 10,
 418                 .deadband = 0,
 419                 .setpoint = 97,
 420                 .p_gain_pct = 14,
 421                 .d_gain_pct = 0,
 422                 .i_gain_pct = 4,
 423         },
 424         .funcs = {
 425                 .get_max = byt_get_max_pstate,
 426                 .get_min = byt_get_min_pstate,
 427                 .get_turbo = byt_get_max_pstate,
 428                 .set = core_set_pstate,
 429         },
 430 };
 431
 432
 433 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 434 {
 435         int max_perf = cpu->pstate.turbo_pstate;
 436         int max_perf_adj;
 437         int min_perf;
 438         if (limits.no_turbo)
 439                 max_perf = cpu->pstate.max_pstate;
 440
 441         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 442         *max = clamp_t(int, max_perf_adj,
 443                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 444
 445         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 446         *min = clamp_t(int, min_perf,
 447                         cpu->pstate.min_pstate, max_perf);
 448 }
 449
 450 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 451 {
 452         int max_perf, min_perf;
 453
 454         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 455
 456         pstate = clamp_t(int, pstate, min_perf, max_perf);
 457
 458         if (pstate == cpu->pstate.current_pstate)
 459                 return;
 460
 461         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 462
 463         cpu->pstate.current_pstate = pstate;
 464
 465         pstate_funcs.set(pstate);
 466 }
 467
 468 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 469 {
 470         int target;
 471         target = cpu->pstate.current_pstate + steps;
 472
 473         intel_pstate_set_pstate(cpu, target);
 474 }
 475
 476 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 477 {
 478         int target;
 479         target = cpu->pstate.current_pstate - steps;
 480         intel_pstate_set_pstate(cpu, target);
 481 }
 482
 483 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 484 {
 485         sprintf(cpu->name, "Intel 2nd generation core");
 486
 487         cpu->pstate.min_pstate = pstate_funcs.get_min();
 488         cpu->pstate.max_pstate = pstate_funcs.get_max();
 489         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 490
 491         /*
 492          * goto max pstate so we don't slow up boot if we are built-in if we are
 493          * a module we will take care of it during normal operation
 494          */
 495         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 496 }
 497
 498 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 499                                         struct sample *sample)
 500 {
 501         u64 core_pct;
 502         core_pct = div64_u64(int_tofp(sample->aperf * 100),
 503                              sample->mperf);
 504         sample->freq = fp_toint(cpu->pstate.max_pstate * core_pct * 1000);
 505
 506         sample->core_pct_busy = core_pct;
 507 }
 508
 509 static inline void intel_pstate_sample(struct cpudata *cpu)
 510 {
 511         u64 aperf, mperf;
 512
 513         rdmsrl(MSR_IA32_APERF, aperf);
 514         rdmsrl(MSR_IA32_MPERF, mperf);
 515         cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
 516         cpu->samples[cpu->sample_ptr].aperf = aperf;
 517         cpu->samples[cpu->sample_ptr].mperf = mperf;
 518         cpu->samples[cpu->sample_ptr].aperf -= cpu->prev_aperf;
 519         cpu->samples[cpu->sample_ptr].mperf -= cpu->prev_mperf;
 520
 521         intel_pstate_calc_busy(cpu, &cpu->samples[cpu->sample_ptr]);
 522
 523         cpu->prev_aperf = aperf;
 524         cpu->prev_mperf = mperf;
 525 }
 526
 527 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 528 {
 529         int sample_time, delay;
 530
 531         sample_time = pid_params.sample_rate_ms;
 532         delay = msecs_to_jiffies(sample_time);
 533         mod_timer_pinned(&cpu->timer, jiffies + delay);
 534 }
 535
 536 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 537 {
 538         int32_t core_busy, max_pstate, current_pstate;
 539
 540         core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy;
 541         max_pstate = int_tofp(cpu->pstate.max_pstate);
 542         current_pstate = int_tofp(cpu->pstate.current_pstate);
 543         return mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 544 }
 545
 546 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 547 {
 548         int32_t busy_scaled;
 549         struct _pid *pid;
 550         signed int ctl = 0;
 551         int steps;
 552
 553         pid = &cpu->pid;
 554         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 555
 556         ctl = pid_calc(pid, busy_scaled);
 557
 558         steps = abs(ctl);
 559         if (ctl < 0)
 560                 intel_pstate_pstate_increase(cpu, steps);
 561         else
 562                 intel_pstate_pstate_decrease(cpu, steps);
 563 }
 564
 565 static void intel_pstate_timer_func(unsigned long __data)
 566 {
 567         struct cpudata *cpu = (struct cpudata *) __data;
 568
 569         intel_pstate_sample(cpu);
 570         intel_pstate_adjust_busy_pstate(cpu);
 571
 572         if (cpu->pstate.current_pstate == cpu->pstate.min_pstate) {
 573                 cpu->min_pstate_count++;
 574                 if (!(cpu->min_pstate_count % 5)) {
 575                         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 576                 }
 577         } else
 578                 cpu->min_pstate_count = 0;
 579
 580         intel_pstate_set_sample_time(cpu);
 581 }
 582
 583 #define ICPU(model, policy) \
 584         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 585                         (unsigned long)&policy }
 586
 587 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 588         ICPU(0x2a, core_params),
 589         ICPU(0x2d, core_params),
 590         ICPU(0x37, byt_params),
 591         ICPU(0x3a, core_params),
 592         ICPU(0x3c, core_params),
 593         ICPU(0x3e, core_params),
 594         ICPU(0x3f, core_params),
 595         ICPU(0x45, core_params),
 596         ICPU(0x46, core_params),
 597         {}
 598 };
 599 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 600
 601 static int intel_pstate_init_cpu(unsigned int cpunum)
 602 {
 603
 604         const struct x86_cpu_id *id;
 605         struct cpudata *cpu;
 606
 607         id = x86_match_cpu(intel_pstate_cpu_ids);
 608         if (!id)
 609                 return -ENODEV;
 610
 611         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 612         if (!all_cpu_data[cpunum])
 613                 return -ENOMEM;
 614
 615         cpu = all_cpu_data[cpunum];
 616
 617         intel_pstate_get_cpu_pstates(cpu);
 618         if (!cpu->pstate.current_pstate) {
 619                 all_cpu_data[cpunum] = NULL;
 620                 kfree(cpu);
 621                 return -ENODATA;
 622         }
 623
 624         cpu->cpu = cpunum;
 625
 626         init_timer_deferrable(&cpu->timer);
 627         cpu->timer.function = intel_pstate_timer_func;
 628         cpu->timer.data =
 629                 (unsigned long)cpu;
 630         cpu->timer.expires = jiffies + HZ/100;
 631         intel_pstate_busy_pid_reset(cpu);
 632         intel_pstate_sample(cpu);
 633         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 634
 635         add_timer_on(&cpu->timer, cpunum);
 636
 637         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 638
 639         return 0;
 640 }
 641
 642 static unsigned int intel_pstate_get(unsigned int cpu_num)
 643 {
 644         struct sample *sample;
 645         struct cpudata *cpu;
 646
 647         cpu = all_cpu_data[cpu_num];
 648         if (!cpu)
 649                 return 0;
 650         sample = &cpu->samples[cpu->sample_ptr];
 651         return sample->freq;
 652 }
 653
 654 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 655 {
 656         struct cpudata *cpu;
 657
 658         cpu = all_cpu_data[policy->cpu];
 659
 660         if (!policy->cpuinfo.max_freq)
 661                 return -ENODEV;
 662
 663         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 664                 limits.min_perf_pct = 100;
 665                 limits.min_perf = int_tofp(1);
 666                 limits.max_perf_pct = 100;
 667                 limits.max_perf = int_tofp(1);
 668                 limits.no_turbo = 0;
 669                 return 0;
 670         }
 671         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 672         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 673         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 674
 675         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 676         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 677         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 678         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 679
 680         return 0;
 681 }
 682
 683 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 684 {
 685         cpufreq_verify_within_cpu_limits(policy);
 686
 687         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 688                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 689                 return -EINVAL;
 690
 691         return 0;
 692 }
 693
 694 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 695 {
 696         int cpu = policy->cpu;
 697
 698         del_timer(&all_cpu_data[cpu]->timer);
 699         kfree(all_cpu_data[cpu]);
 700         all_cpu_data[cpu] = NULL;
 701         return 0;
 702 }
 703
 704 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 705 {
 706         struct cpudata *cpu;
 707         int rc;
 708
 709         rc = intel_pstate_init_cpu(policy->cpu);
 710         if (rc)
 711                 return rc;
 712
 713         cpu = all_cpu_data[policy->cpu];
 714
 715         if (!limits.no_turbo &&
 716                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 717                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 718         else
 719                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 720
 721         policy->min = cpu->pstate.min_pstate * 100000;
 722         policy->max = cpu->pstate.turbo_pstate * 100000;
 723
 724         /* cpuinfo and default policy values */
 725         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 726         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 727         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 728         cpumask_set_cpu(policy->cpu, policy->cpus);
 729
 730         return 0;
 731 }
 732
 733 static struct cpufreq_driver intel_pstate_driver = {
 734         .flags          = CPUFREQ_CONST_LOOPS,
 735         .verify         = intel_pstate_verify_policy,
 736         .setpolicy      = intel_pstate_set_policy,
 737         .get            = intel_pstate_get,
 738         .init           = intel_pstate_cpu_init,
 739         .exit           = intel_pstate_cpu_exit,
 740         .name           = "intel_pstate",
 741 };
 742
 743 static int __initdata no_load;
 744
 745 static int intel_pstate_msrs_not_valid(void)
 746 {
 747         /* Check that all the msr's we are using are valid. */
 748         u64 aperf, mperf, tmp;
 749
 750         rdmsrl(MSR_IA32_APERF, aperf);
 751         rdmsrl(MSR_IA32_MPERF, mperf);
 752
 753         if (!pstate_funcs.get_max() ||
 754                 !pstate_funcs.get_min() ||
 755                 !pstate_funcs.get_turbo())
 756                 return -ENODEV;
 757
 758         rdmsrl(MSR_IA32_APERF, tmp);
 759         if (!(tmp - aperf))
 760                 return -ENODEV;
 761
 762         rdmsrl(MSR_IA32_MPERF, tmp);
 763         if (!(tmp - mperf))
 764                 return -ENODEV;
 765
 766         return 0;
 767 }
 768
 769 static void copy_pid_params(struct pstate_adjust_policy *policy)
 770 {
 771         pid_params.sample_rate_ms = policy->sample_rate_ms;
 772         pid_params.p_gain_pct = policy->p_gain_pct;
 773         pid_params.i_gain_pct = policy->i_gain_pct;
 774         pid_params.d_gain_pct = policy->d_gain_pct;
 775         pid_params.deadband = policy->deadband;
 776         pid_params.setpoint = policy->setpoint;
 777 }
 778
 779 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 780 {
 781         pstate_funcs.get_max   = funcs->get_max;
 782         pstate_funcs.get_min   = funcs->get_min;
 783         pstate_funcs.get_turbo = funcs->get_turbo;
 784         pstate_funcs.set       = funcs->set;
 785 }
 786
 787 #if IS_ENABLED(CONFIG_ACPI)
 788 #include <acpi/processor.h>
 789
 790 static bool intel_pstate_no_acpi_pss(void)
 791 {
 792         int i;
 793
 794         for_each_possible_cpu(i) {
 795                 acpi_status status;
 796                 union acpi_object *pss;
 797                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 798                 struct acpi_processor *pr = per_cpu(processors, i);
 799
 800                 if (!pr)
 801                         continue;
 802
 803                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 804                 if (ACPI_FAILURE(status))
 805                         continue;
 806
 807                 pss = buffer.pointer;
 808                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 809                         kfree(pss);
 810                         return false;
 811                 }
 812
 813                 kfree(pss);
 814         }
 815
 816         return true;
 817 }
 818
 819 struct hw_vendor_info {
 820         u16  valid;
 821         char oem_id[ACPI_OEM_ID_SIZE];
 822         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 823 };
 824
 825 /* Hardware vendor-specific info that has its own power management modes */
 826 static struct hw_vendor_info vendor_info[] = {
 827         {1, "HP    ", "ProLiant"},
 828         {0, "", ""},
 829 };
 830
 831 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 832 {
 833         struct acpi_table_header hdr;
 834         struct hw_vendor_info *v_info;
 835
 836         if (acpi_disabled
 837             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 838                 return false;
 839
 840         for (v_info = vendor_info; v_info->valid; v_info++) {
 841                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 842                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 843                     && intel_pstate_no_acpi_pss())
 844                         return true;
 845         }
 846
 847         return false;
 848 }
 849 #else /* CONFIG_ACPI not enabled */
 850 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 851 #endif /* CONFIG_ACPI */
 852
 853 static int __init intel_pstate_init(void)
 854 {
 855         int cpu, rc = 0;
 856         const struct x86_cpu_id *id;
 857         struct cpu_defaults *cpu_info;
 858
 859         if (no_load)
 860                 return -ENODEV;
 861
 862         id = x86_match_cpu(intel_pstate_cpu_ids);
 863         if (!id)
 864                 return -ENODEV;
 865
 866         /*
 867          * The Intel pstate driver will be ignored if the platform
 868          * firmware has its own power management modes.
 869          */
 870         if (intel_pstate_platform_pwr_mgmt_exists())
 871                 return -ENODEV;
 872
 873         cpu_info = (struct cpu_defaults *)id->driver_data;
 874
 875         copy_pid_params(&cpu_info->pid_policy);
 876         copy_cpu_funcs(&cpu_info->funcs);
 877
 878         if (intel_pstate_msrs_not_valid())
 879                 return -ENODEV;
 880
 881         pr_info("Intel P-state driver initializing.\n");
 882
 883         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 884         if (!all_cpu_data)
 885                 return -ENOMEM;
 886
 887         rc = cpufreq_register_driver(&intel_pstate_driver);
 888         if (rc)
 889                 goto out;
 890
 891         intel_pstate_debug_expose_params();
 892         intel_pstate_sysfs_expose_params();
 893         return rc;
 894 out:
 895         get_online_cpus();
 896         for_each_online_cpu(cpu) {
 897                 if (all_cpu_data[cpu]) {
 898                         del_timer_sync(&all_cpu_data[cpu]->timer);
 899                         kfree(all_cpu_data[cpu]);
 900                 }
 901         }
 902
 903         put_online_cpus();
 904         vfree(all_cpu_data);
 905         return -ENODEV;
 906 }
 907 device_initcall(intel_pstate_init);
 908
 909 static int __init intel_pstate_setup(char *str)
 910 {
 911         if (!str)
 912                 return -EINVAL;
 913
 914         if (!strcmp(str, "disable"))
 915                 no_load = 1;
 916         return 0;
 917 }
 918 early_param("intel_pstate", intel_pstate_setup);
 919
 920 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
 921 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
 922 MODULE_LICENSE("GPL");