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 #define BYT_VIDS        0x66b
  39
  40 #define FRAC_BITS 8
  41 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  42 #define fp_toint(X) ((X) >> FRAC_BITS)
  43
  44 static inline int32_t mul_fp(int32_t x, int32_t y)
  45 {
  46         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  47 }
  48
  49 static inline int32_t div_fp(int32_t x, int32_t y)
  50 {
  51         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  52 }
  53
  54 static u64 energy_divisor;
  55
  56 struct sample {
  57         int32_t core_pct_busy;
  58         u64 aperf;
  59         u64 mperf;
  60         unsigned long long tsc;
  61         int freq;
  62 };
  63
  64 struct pstate_data {
  65         int     current_pstate;
  66         int     min_pstate;
  67         int     max_pstate;
  68         int     turbo_pstate;
  69 };
  70
  71 struct vid_data {
  72         int32_t min;
  73         int32_t max;
  74         int32_t ratio;
  75 };
  76
  77 struct _pid {
  78         int setpoint;
  79         int32_t integral;
  80         int32_t p_gain;
  81         int32_t i_gain;
  82         int32_t d_gain;
  83         int deadband;
  84         int32_t last_err;
  85 };
  86
  87 struct cpudata {
  88         int cpu;
  89
  90         char name[64];
  91
  92         struct timer_list timer;
  93
  94         struct pstate_data pstate;
  95         struct vid_data vid;
  96         struct _pid pid;
  97
  98         u64     prev_aperf;
  99         u64     prev_mperf;
 100         unsigned long long prev_tsc;
 101         int     sample_ptr;
 102         struct sample samples[SAMPLE_COUNT];
 103 };
 104
 105 static struct cpudata **all_cpu_data;
 106 struct pstate_adjust_policy {
 107         int sample_rate_ms;
 108         int deadband;
 109         int setpoint;
 110         int p_gain_pct;
 111         int d_gain_pct;
 112         int i_gain_pct;
 113 };
 114
 115 struct pstate_funcs {
 116         int (*get_max)(void);
 117         int (*get_min)(void);
 118         int (*get_turbo)(void);
 119         void (*set)(struct cpudata*, int pstate);
 120         void (*get_vid)(struct cpudata *);
 121 };
 122
 123 struct cpu_defaults {
 124         struct pstate_adjust_policy pid_policy;
 125         struct pstate_funcs funcs;
 126 };
 127
 128 static struct pstate_adjust_policy pid_params;
 129 static struct pstate_funcs pstate_funcs;
 130
 131 struct perf_limits {
 132         int no_turbo;
 133         int max_perf_pct;
 134         int min_perf_pct;
 135         int32_t max_perf;
 136         int32_t min_perf;
 137         int max_policy_pct;
 138         int max_sysfs_pct;
 139 };
 140
 141 static struct perf_limits limits = {
 142         .no_turbo = 0,
 143         .max_perf_pct = 100,
 144         .max_perf = int_tofp(1),
 145         .min_perf_pct = 0,
 146         .min_perf = 0,
 147         .max_policy_pct = 100,
 148         .max_sysfs_pct = 100,
 149 };
 150
 151 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 152                         int deadband, int integral) {
 153         pid->setpoint = setpoint;
 154         pid->deadband  = deadband;
 155         pid->integral  = int_tofp(integral);
 156         pid->last_err  = setpoint - busy;
 157 }
 158
 159 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 160 {
 161         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 162 }
 163
 164 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 165 {
 166         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 167 }
 168
 169 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 170 {
 171
 172         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 173 }
 174
 175 static signed int pid_calc(struct _pid *pid, int32_t busy)
 176 {
 177         signed int result;
 178         int32_t pterm, dterm, fp_error;
 179         int32_t integral_limit;
 180
 181         fp_error = int_tofp(pid->setpoint) - busy;
 182
 183         if (abs(fp_error) <= int_tofp(pid->deadband))
 184                 return 0;
 185
 186         pterm = mul_fp(pid->p_gain, fp_error);
 187
 188         pid->integral += fp_error;
 189
 190         /* limit the integral term */
 191         integral_limit = int_tofp(30);
 192         if (pid->integral > integral_limit)
 193                 pid->integral = integral_limit;
 194         if (pid->integral < -integral_limit)
 195                 pid->integral = -integral_limit;
 196
 197         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 198         pid->last_err = fp_error;
 199
 200         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 201
 202         return (signed int)fp_toint(result);
 203 }
 204
 205 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 206 {
 207         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 208         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 209         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 210
 211         pid_reset(&cpu->pid,
 212                 pid_params.setpoint,
 213                 100,
 214                 pid_params.deadband,
 215                 0);
 216 }
 217
 218 static inline void intel_pstate_reset_all_pid(void)
 219 {
 220         unsigned int cpu;
 221         for_each_online_cpu(cpu) {
 222                 if (all_cpu_data[cpu])
 223                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 224         }
 225 }
 226
 227 /************************** debugfs begin ************************/
 228 static int pid_param_set(void *data, u64 val)
 229 {
 230         *(u32 *)data = val;
 231         intel_pstate_reset_all_pid();
 232         return 0;
 233 }
 234 static int pid_param_get(void *data, u64 *val)
 235 {
 236         *val = *(u32 *)data;
 237         return 0;
 238 }
 239 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 240                         pid_param_set, "%llu\n");
 241
 242 struct pid_param {
 243         char *name;
 244         void *value;
 245 };
 246
 247 static struct pid_param pid_files[] = {
 248         {"sample_rate_ms", &pid_params.sample_rate_ms},
 249         {"d_gain_pct", &pid_params.d_gain_pct},
 250         {"i_gain_pct", &pid_params.i_gain_pct},
 251         {"deadband", &pid_params.deadband},
 252         {"setpoint", &pid_params.setpoint},
 253         {"p_gain_pct", &pid_params.p_gain_pct},
 254         {NULL, NULL}
 255 };
 256
 257 static struct dentry *debugfs_parent;
 258 static void intel_pstate_debug_expose_params(void)
 259 {
 260         int i = 0;
 261
 262         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 263         if (IS_ERR_OR_NULL(debugfs_parent))
 264                 return;
 265         while (pid_files[i].name) {
 266                 debugfs_create_file(pid_files[i].name, 0660,
 267                                 debugfs_parent, pid_files[i].value,
 268                                 &fops_pid_param);
 269                 i++;
 270         }
 271 }
 272
 273 /************************** debugfs end ************************/
 274
 275 /************************** sysfs begin ************************/
 276 #define show_one(file_name, object)                                     \
 277         static ssize_t show_##file_name                                 \
 278         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 279         {                                                               \
 280                 return sprintf(buf, "%u\n", limits.object);             \
 281         }
 282
 283 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 284                                 const char *buf, size_t count)
 285 {
 286         unsigned int input;
 287         int ret;
 288         ret = sscanf(buf, "%u", &input);
 289         if (ret != 1)
 290                 return -EINVAL;
 291         limits.no_turbo = clamp_t(int, input, 0 , 1);
 292
 293         return count;
 294 }
 295
 296 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 297                                 const char *buf, size_t count)
 298 {
 299         unsigned int input;
 300         int ret;
 301         ret = sscanf(buf, "%u", &input);
 302         if (ret != 1)
 303                 return -EINVAL;
 304
 305         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 306         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 307         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 308         return count;
 309 }
 310
 311 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 312                                 const char *buf, size_t count)
 313 {
 314         unsigned int input;
 315         int ret;
 316         ret = sscanf(buf, "%u", &input);
 317         if (ret != 1)
 318                 return -EINVAL;
 319         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 320         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 321
 322         return count;
 323 }
 324
 325 show_one(no_turbo, no_turbo);
 326 show_one(max_perf_pct, max_perf_pct);
 327 show_one(min_perf_pct, min_perf_pct);
 328
 329 define_one_global_rw(no_turbo);
 330 define_one_global_rw(max_perf_pct);
 331 define_one_global_rw(min_perf_pct);
 332
 333 static struct attribute *intel_pstate_attributes[] = {
 334         &no_turbo.attr,
 335         &max_perf_pct.attr,
 336         &min_perf_pct.attr,
 337         NULL
 338 };
 339
 340 static struct attribute_group intel_pstate_attr_group = {
 341         .attrs = intel_pstate_attributes,
 342 };
 343 static struct kobject *intel_pstate_kobject;
 344
 345 static void intel_pstate_sysfs_expose_params(void)
 346 {
 347         int rc;
 348
 349         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 350                                                 &cpu_subsys.dev_root->kobj);
 351         BUG_ON(!intel_pstate_kobject);
 352         rc = sysfs_create_group(intel_pstate_kobject,
 353                                 &intel_pstate_attr_group);
 354         BUG_ON(rc);
 355 }
 356
 357 /************************** sysfs end ************************/
 358 static int byt_get_min_pstate(void)
 359 {
 360         u64 value;
 361         rdmsrl(BYT_RATIOS, value);
 362         return value & 0xFF;
 363 }
 364
 365 static int byt_get_max_pstate(void)
 366 {
 367         u64 value;
 368         rdmsrl(BYT_RATIOS, value);
 369         return (value >> 16) & 0xFF;
 370 }
 371
 372 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 373 {
 374         u64 val;
 375         int32_t vid_fp;
 376         u32 vid;
 377
 378         val = pstate << 8;
 379         if (limits.no_turbo)
 380                 val |= (u64)1 << 32;
 381
 382         vid_fp = cpudata->vid.min + mul_fp(
 383                 int_tofp(pstate - cpudata->pstate.min_pstate),
 384                 cpudata->vid.ratio);
 385
 386         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 387         vid = fp_toint(vid_fp);
 388
 389         val |= vid;
 390
 391         wrmsrl(MSR_IA32_PERF_CTL, val);
 392 }
 393
 394 static void byt_get_vid(struct cpudata *cpudata)
 395 {
 396         u64 value;
 397
 398         rdmsrl(BYT_VIDS, value);
 399         cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
 400         cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
 401         cpudata->vid.ratio = div_fp(
 402                 cpudata->vid.max - cpudata->vid.min,
 403                 int_tofp(cpudata->pstate.max_pstate -
 404                         cpudata->pstate.min_pstate));
 405 }
 406
 407
 408 static int core_get_min_pstate(void)
 409 {
 410         u64 value;
 411         rdmsrl(MSR_PLATFORM_INFO, value);
 412         return (value >> 40) & 0xFF;
 413 }
 414
 415 static int core_get_max_pstate(void)
 416 {
 417         u64 value;
 418         rdmsrl(MSR_PLATFORM_INFO, value);
 419         return (value >> 8) & 0xFF;
 420 }
 421
 422 static int core_get_turbo_pstate(void)
 423 {
 424         u64 value;
 425         int nont, ret;
 426         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 427         nont = core_get_max_pstate();
 428         ret = ((value) & 255);
 429         if (ret <= nont)
 430                 ret = nont;
 431         return ret;
 432 }
 433
 434 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 435 {
 436         u64 val;
 437
 438         val = pstate << 8;
 439         if (limits.no_turbo)
 440                 val |= (u64)1 << 32;
 441
 442         wrmsrl(MSR_IA32_PERF_CTL, val);
 443 }
 444
 445 static struct cpu_defaults core_params = {
 446         .pid_policy = {
 447                 .sample_rate_ms = 10,
 448                 .deadband = 0,
 449                 .setpoint = 97,
 450                 .p_gain_pct = 20,
 451                 .d_gain_pct = 0,
 452                 .i_gain_pct = 0,
 453         },
 454         .funcs = {
 455                 .get_max = core_get_max_pstate,
 456                 .get_min = core_get_min_pstate,
 457                 .get_turbo = core_get_turbo_pstate,
 458                 .set = core_set_pstate,
 459         },
 460 };
 461
 462 static struct cpu_defaults byt_params = {
 463         .pid_policy = {
 464                 .sample_rate_ms = 10,
 465                 .deadband = 0,
 466                 .setpoint = 97,
 467                 .p_gain_pct = 14,
 468                 .d_gain_pct = 0,
 469                 .i_gain_pct = 4,
 470         },
 471         .funcs = {
 472                 .get_max = byt_get_max_pstate,
 473                 .get_min = byt_get_min_pstate,
 474                 .get_turbo = byt_get_max_pstate,
 475                 .set = byt_set_pstate,
 476                 .get_vid = byt_get_vid,
 477         },
 478 };
 479
 480
 481 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 482 {
 483         int max_perf = cpu->pstate.turbo_pstate;
 484         int max_perf_adj;
 485         int min_perf;
 486         if (limits.no_turbo)
 487                 max_perf = cpu->pstate.max_pstate;
 488
 489         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 490         *max = clamp_t(int, max_perf_adj,
 491                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 492
 493         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 494         *min = clamp_t(int, min_perf,
 495                         cpu->pstate.min_pstate, max_perf);
 496 }
 497
 498 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 499 {
 500         int max_perf, min_perf;
 501
 502         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 503
 504         pstate = clamp_t(int, pstate, min_perf, max_perf);
 505
 506         if (pstate == cpu->pstate.current_pstate)
 507                 return;
 508
 509         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 510
 511         cpu->pstate.current_pstate = pstate;
 512
 513         pstate_funcs.set(cpu, pstate);
 514 }
 515
 516 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 517 {
 518         int target;
 519         target = cpu->pstate.current_pstate + steps;
 520
 521         intel_pstate_set_pstate(cpu, target);
 522 }
 523
 524 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 525 {
 526         int target;
 527         target = cpu->pstate.current_pstate - steps;
 528         intel_pstate_set_pstate(cpu, target);
 529 }
 530
 531 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 532 {
 533         sprintf(cpu->name, "Intel 2nd generation core");
 534
 535         cpu->pstate.min_pstate = pstate_funcs.get_min();
 536         cpu->pstate.max_pstate = pstate_funcs.get_max();
 537         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 538
 539         if (pstate_funcs.get_vid)
 540                 pstate_funcs.get_vid(cpu);
 541
 542         /*
 543          * goto max pstate so we don't slow up boot if we are built-in if we are
 544          * a module we will take care of it during normal operation
 545          */
 546         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 547 }
 548
 549 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 550                                         struct sample *sample)
 551 {
 552         u64 core_pct;
 553         u64 c0_pct;
 554
 555         core_pct = div64_u64(sample->aperf * 100, sample->mperf);
 556
 557         c0_pct = div64_u64(sample->mperf * 100, sample->tsc);
 558         sample->freq = fp_toint(
 559                 mul_fp(int_tofp(cpu->pstate.max_pstate),
 560                         int_tofp(core_pct * 1000)));
 561
 562         sample->core_pct_busy = mul_fp(int_tofp(core_pct),
 563                                 div_fp(int_tofp(c0_pct + 1), int_tofp(100)));
 564 }
 565
 566 static inline void intel_pstate_sample(struct cpudata *cpu)
 567 {
 568         u64 aperf, mperf;
 569         unsigned long long tsc;
 570
 571         rdmsrl(MSR_IA32_APERF, aperf);
 572         rdmsrl(MSR_IA32_MPERF, mperf);
 573         tsc = native_read_tsc();
 574
 575         cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
 576         cpu->samples[cpu->sample_ptr].aperf = aperf;
 577         cpu->samples[cpu->sample_ptr].mperf = mperf;
 578         cpu->samples[cpu->sample_ptr].tsc = tsc;
 579         cpu->samples[cpu->sample_ptr].aperf -= cpu->prev_aperf;
 580         cpu->samples[cpu->sample_ptr].mperf -= cpu->prev_mperf;
 581         cpu->samples[cpu->sample_ptr].tsc -= cpu->prev_tsc;
 582
 583         intel_pstate_calc_busy(cpu, &cpu->samples[cpu->sample_ptr]);
 584
 585         cpu->prev_aperf = aperf;
 586         cpu->prev_mperf = mperf;
 587         cpu->prev_tsc = tsc;
 588 }
 589
 590 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 591 {
 592         int sample_time, delay;
 593
 594         sample_time = pid_params.sample_rate_ms;
 595         delay = msecs_to_jiffies(sample_time);
 596         mod_timer_pinned(&cpu->timer, jiffies + delay);
 597 }
 598
 599 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 600 {
 601         int32_t core_busy, max_pstate, current_pstate;
 602
 603         core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy;
 604         max_pstate = int_tofp(cpu->pstate.max_pstate);
 605         current_pstate = int_tofp(cpu->pstate.current_pstate);
 606         return mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 607 }
 608
 609 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 610 {
 611         int32_t busy_scaled;
 612         struct _pid *pid;
 613         signed int ctl = 0;
 614         int steps;
 615
 616         pid = &cpu->pid;
 617         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 618
 619         ctl = pid_calc(pid, busy_scaled);
 620
 621         steps = abs(ctl);
 622
 623         if (ctl < 0)
 624                 intel_pstate_pstate_increase(cpu, steps);
 625         else
 626                 intel_pstate_pstate_decrease(cpu, steps);
 627 }
 628
 629 static void intel_pstate_timer_func(unsigned long __data)
 630 {
 631         struct cpudata *cpu = (struct cpudata *) __data;
 632         struct sample *sample;
 633         u64 energy;
 634
 635         intel_pstate_sample(cpu);
 636
 637         sample = &cpu->samples[cpu->sample_ptr];
 638         rdmsrl(MSR_PKG_ENERGY_STATUS, energy);
 639
 640         intel_pstate_adjust_busy_pstate(cpu);
 641
 642         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 643                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 644                         cpu->pstate.current_pstate,
 645                         sample->mperf,
 646                         sample->aperf,
 647                         div64_u64(energy, energy_divisor),
 648                         sample->freq);
 649
 650         intel_pstate_set_sample_time(cpu);
 651 }
 652
 653 #define ICPU(model, policy) \
 654         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 655                         (unsigned long)&policy }
 656
 657 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 658         ICPU(0x2a, core_params),
 659         ICPU(0x2d, core_params),
 660         ICPU(0x37, byt_params),
 661         ICPU(0x3a, core_params),
 662         ICPU(0x3c, core_params),
 663         ICPU(0x3e, core_params),
 664         ICPU(0x3f, core_params),
 665         ICPU(0x45, core_params),
 666         ICPU(0x46, core_params),
 667         {}
 668 };
 669 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 670
 671 static int intel_pstate_init_cpu(unsigned int cpunum)
 672 {
 673
 674         const struct x86_cpu_id *id;
 675         struct cpudata *cpu;
 676
 677         id = x86_match_cpu(intel_pstate_cpu_ids);
 678         if (!id)
 679                 return -ENODEV;
 680
 681         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 682         if (!all_cpu_data[cpunum])
 683                 return -ENOMEM;
 684
 685         cpu = all_cpu_data[cpunum];
 686
 687         intel_pstate_get_cpu_pstates(cpu);
 688         if (!cpu->pstate.current_pstate) {
 689                 all_cpu_data[cpunum] = NULL;
 690                 kfree(cpu);
 691                 return -ENODATA;
 692         }
 693
 694         cpu->cpu = cpunum;
 695
 696         init_timer_deferrable(&cpu->timer);
 697         cpu->timer.function = intel_pstate_timer_func;
 698         cpu->timer.data =
 699                 (unsigned long)cpu;
 700         cpu->timer.expires = jiffies + HZ/100;
 701         intel_pstate_busy_pid_reset(cpu);
 702         intel_pstate_sample(cpu);
 703         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 704
 705         add_timer_on(&cpu->timer, cpunum);
 706
 707         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 708
 709         return 0;
 710 }
 711
 712 static unsigned int intel_pstate_get(unsigned int cpu_num)
 713 {
 714         struct sample *sample;
 715         struct cpudata *cpu;
 716
 717         cpu = all_cpu_data[cpu_num];
 718         if (!cpu)
 719                 return 0;
 720         sample = &cpu->samples[cpu->sample_ptr];
 721         return sample->freq;
 722 }
 723
 724 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 725 {
 726         struct cpudata *cpu;
 727
 728         cpu = all_cpu_data[policy->cpu];
 729
 730         if (!policy->cpuinfo.max_freq)
 731                 return -ENODEV;
 732
 733         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 734                 limits.min_perf_pct = 100;
 735                 limits.min_perf = int_tofp(1);
 736                 limits.max_perf_pct = 100;
 737                 limits.max_perf = int_tofp(1);
 738                 limits.no_turbo = 0;
 739                 return 0;
 740         }
 741         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 742         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 743         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 744
 745         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 746         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 747         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 748         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 749
 750         return 0;
 751 }
 752
 753 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 754 {
 755         cpufreq_verify_within_cpu_limits(policy);
 756
 757         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 758                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 759                 return -EINVAL;
 760
 761         return 0;
 762 }
 763
 764 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 765 {
 766         int cpu = policy->cpu;
 767
 768         del_timer(&all_cpu_data[cpu]->timer);
 769         kfree(all_cpu_data[cpu]);
 770         all_cpu_data[cpu] = NULL;
 771         return 0;
 772 }
 773
 774 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 775 {
 776         struct cpudata *cpu;
 777         int rc;
 778
 779         rc = intel_pstate_init_cpu(policy->cpu);
 780         if (rc)
 781                 return rc;
 782
 783         cpu = all_cpu_data[policy->cpu];
 784
 785         if (!limits.no_turbo &&
 786                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 787                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 788         else
 789                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 790
 791         policy->min = cpu->pstate.min_pstate * 100000;
 792         policy->max = cpu->pstate.turbo_pstate * 100000;
 793
 794         /* cpuinfo and default policy values */
 795         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 796         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 797         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 798         cpumask_set_cpu(policy->cpu, policy->cpus);
 799
 800         return 0;
 801 }
 802
 803 static struct cpufreq_driver intel_pstate_driver = {
 804         .flags          = CPUFREQ_CONST_LOOPS,
 805         .verify         = intel_pstate_verify_policy,
 806         .setpolicy      = intel_pstate_set_policy,
 807         .get            = intel_pstate_get,
 808         .init           = intel_pstate_cpu_init,
 809         .exit           = intel_pstate_cpu_exit,
 810         .name           = "intel_pstate",
 811 };
 812
 813 static int __initdata no_load;
 814
 815 static int intel_pstate_msrs_not_valid(void)
 816 {
 817         /* Check that all the msr's we are using are valid. */
 818         u64 aperf, mperf, tmp;
 819
 820         rdmsrl(MSR_IA32_APERF, aperf);
 821         rdmsrl(MSR_IA32_MPERF, mperf);
 822
 823         if (!pstate_funcs.get_max() ||
 824                 !pstate_funcs.get_min() ||
 825                 !pstate_funcs.get_turbo())
 826                 return -ENODEV;
 827
 828         rdmsrl(MSR_IA32_APERF, tmp);
 829         if (!(tmp - aperf))
 830                 return -ENODEV;
 831
 832         rdmsrl(MSR_IA32_MPERF, tmp);
 833         if (!(tmp - mperf))
 834                 return -ENODEV;
 835
 836         return 0;
 837 }
 838
 839 static void copy_pid_params(struct pstate_adjust_policy *policy)
 840 {
 841         pid_params.sample_rate_ms = policy->sample_rate_ms;
 842         pid_params.p_gain_pct = policy->p_gain_pct;
 843         pid_params.i_gain_pct = policy->i_gain_pct;
 844         pid_params.d_gain_pct = policy->d_gain_pct;
 845         pid_params.deadband = policy->deadband;
 846         pid_params.setpoint = policy->setpoint;
 847 }
 848
 849 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 850 {
 851         pstate_funcs.get_max   = funcs->get_max;
 852         pstate_funcs.get_min   = funcs->get_min;
 853         pstate_funcs.get_turbo = funcs->get_turbo;
 854         pstate_funcs.set       = funcs->set;
 855         pstate_funcs.get_vid   = funcs->get_vid;
 856 }
 857
 858 #if IS_ENABLED(CONFIG_ACPI)
 859 #include <acpi/processor.h>
 860
 861 static bool intel_pstate_no_acpi_pss(void)
 862 {
 863         int i;
 864
 865         for_each_possible_cpu(i) {
 866                 acpi_status status;
 867                 union acpi_object *pss;
 868                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 869                 struct acpi_processor *pr = per_cpu(processors, i);
 870
 871                 if (!pr)
 872                         continue;
 873
 874                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 875                 if (ACPI_FAILURE(status))
 876                         continue;
 877
 878                 pss = buffer.pointer;
 879                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 880                         kfree(pss);
 881                         return false;
 882                 }
 883
 884                 kfree(pss);
 885         }
 886
 887         return true;
 888 }
 889
 890 struct hw_vendor_info {
 891         u16  valid;
 892         char oem_id[ACPI_OEM_ID_SIZE];
 893         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 894 };
 895
 896 /* Hardware vendor-specific info that has its own power management modes */
 897 static struct hw_vendor_info vendor_info[] = {
 898         {1, "HP    ", "ProLiant"},
 899         {0, "", ""},
 900 };
 901
 902 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 903 {
 904         struct acpi_table_header hdr;
 905         struct hw_vendor_info *v_info;
 906
 907         if (acpi_disabled
 908             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 909                 return false;
 910
 911         for (v_info = vendor_info; v_info->valid; v_info++) {
 912                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 913                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 914                     && intel_pstate_no_acpi_pss())
 915                         return true;
 916         }
 917
 918         return false;
 919 }
 920 #else /* CONFIG_ACPI not enabled */
 921 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 922 #endif /* CONFIG_ACPI */
 923
 924 static int __init intel_pstate_init(void)
 925 {
 926         int cpu, rc = 0;
 927         const struct x86_cpu_id *id;
 928         struct cpu_defaults *cpu_info;
 929         u64 units;
 930
 931         if (no_load)
 932                 return -ENODEV;
 933
 934         id = x86_match_cpu(intel_pstate_cpu_ids);
 935         if (!id)
 936                 return -ENODEV;
 937
 938         /*
 939          * The Intel pstate driver will be ignored if the platform
 940          * firmware has its own power management modes.
 941          */
 942         if (intel_pstate_platform_pwr_mgmt_exists())
 943                 return -ENODEV;
 944
 945         cpu_info = (struct cpu_defaults *)id->driver_data;
 946
 947         copy_pid_params(&cpu_info->pid_policy);
 948         copy_cpu_funcs(&cpu_info->funcs);
 949
 950         if (intel_pstate_msrs_not_valid())
 951                 return -ENODEV;
 952
 953         pr_info("Intel P-state driver initializing.\n");
 954
 955         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 956         if (!all_cpu_data)
 957                 return -ENOMEM;
 958
 959         rc = cpufreq_register_driver(&intel_pstate_driver);
 960         if (rc)
 961                 goto out;
 962
 963         rdmsrl(MSR_RAPL_POWER_UNIT, units);
 964         energy_divisor = 1 << ((units >> 8) & 0x1f); /* bits{12:8} */
 965
 966         intel_pstate_debug_expose_params();
 967         intel_pstate_sysfs_expose_params();
 968
 969         return rc;
 970 out:
 971         get_online_cpus();
 972         for_each_online_cpu(cpu) {
 973                 if (all_cpu_data[cpu]) {
 974                         del_timer_sync(&all_cpu_data[cpu]->timer);
 975                         kfree(all_cpu_data[cpu]);
 976                 }
 977         }
 978
 979         put_online_cpus();
 980         vfree(all_cpu_data);
 981         return -ENODEV;
 982 }
 983 device_initcall(intel_pstate_init);
 984
 985 static int __init intel_pstate_setup(char *str)
 986 {
 987         if (!str)
 988                 return -EINVAL;
 989
 990         if (!strcmp(str, "disable"))
 991                 no_load = 1;
 992         return 0;
 993 }
 994 early_param("intel_pstate", intel_pstate_setup);
 995
 996 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
 997 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
 998 MODULE_LICENSE("GPL");