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 BYT_RATIOS              0x66a
  36 #define BYT_VIDS                0x66b
  37 #define BYT_TURBO_RATIOS        0x66c
  38 #define BYT_TURBO_VIDS          0x66d
  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
  45 static inline int32_t mul_fp(int32_t x, int32_t y)
  46 {
  47         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  48 }
  49
  50 static inline int32_t div_fp(int32_t x, int32_t y)
  51 {
  52         return div_s64((int64_t)x << FRAC_BITS, y);
  53 }
  54
  55 struct sample {
  56         int32_t core_pct_busy;
  57         u64 aperf;
  58         u64 mperf;
  59         int freq;
  60         ktime_t time;
  61 };
  62
  63 struct pstate_data {
  64         int     current_pstate;
  65         int     min_pstate;
  66         int     max_pstate;
  67         int     scaling;
  68         int     turbo_pstate;
  69 };
  70
  71 struct vid_data {
  72         int min;
  73         int max;
  74         int turbo;
  75         int32_t ratio;
  76 };
  77
  78 struct _pid {
  79         int setpoint;
  80         int32_t integral;
  81         int32_t p_gain;
  82         int32_t i_gain;
  83         int32_t d_gain;
  84         int deadband;
  85         int32_t last_err;
  86 };
  87
  88 struct cpudata {
  89         int cpu;
  90
  91         struct timer_list timer;
  92
  93         struct pstate_data pstate;
  94         struct vid_data vid;
  95         struct _pid pid;
  96
  97         ktime_t last_sample_time;
  98         u64     prev_aperf;
  99         u64     prev_mperf;
 100         struct sample sample;
 101 };
 102
 103 static struct cpudata **all_cpu_data;
 104 struct pstate_adjust_policy {
 105         int sample_rate_ms;
 106         int deadband;
 107         int setpoint;
 108         int p_gain_pct;
 109         int d_gain_pct;
 110         int i_gain_pct;
 111 };
 112
 113 struct pstate_funcs {
 114         int (*get_max)(void);
 115         int (*get_min)(void);
 116         int (*get_turbo)(void);
 117         int (*get_scaling)(void);
 118         void (*set)(struct cpudata*, int pstate);
 119         void (*get_vid)(struct cpudata *);
 120 };
 121
 122 struct cpu_defaults {
 123         struct pstate_adjust_policy pid_policy;
 124         struct pstate_funcs funcs;
 125 };
 126
 127 static struct pstate_adjust_policy pid_params;
 128 static struct pstate_funcs pstate_funcs;
 129
 130 struct perf_limits {
 131         int no_turbo;
 132         int turbo_disabled;
 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         .turbo_disabled = 0,
 144         .max_perf_pct = 100,
 145         .max_perf = int_tofp(1),
 146         .min_perf_pct = 0,
 147         .min_perf = 0,
 148         .max_policy_pct = 100,
 149         .max_sysfs_pct = 100,
 150 };
 151
 152 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 153                              int deadband, int integral) {
 154         pid->setpoint = setpoint;
 155         pid->deadband  = deadband;
 156         pid->integral  = int_tofp(integral);
 157         pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
 158 }
 159
 160 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 161 {
 162         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 163 }
 164
 165 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 166 {
 167         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 168 }
 169
 170 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 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         result = result + (1 << (FRAC_BITS-1));
 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, pid_params.setpoint, 100, pid_params.deadband, 0);
 212 }
 213
 214 static inline void intel_pstate_reset_all_pid(void)
 215 {
 216         unsigned int cpu;
 217
 218         for_each_online_cpu(cpu) {
 219                 if (all_cpu_data[cpu])
 220                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 221         }
 222 }
 223
 224 static inline void update_turbo_state(void)
 225 {
 226         u64 misc_en;
 227         struct cpudata *cpu;
 228
 229         cpu = all_cpu_data[0];
 230         rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
 231         limits.turbo_disabled =
 232                 (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
 233                  cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
 234 }
 235
 236 /************************** debugfs begin ************************/
 237 static int pid_param_set(void *data, u64 val)
 238 {
 239         *(u32 *)data = val;
 240         intel_pstate_reset_all_pid();
 241         return 0;
 242 }
 243
 244 static int pid_param_get(void *data, u64 *val)
 245 {
 246         *val = *(u32 *)data;
 247         return 0;
 248 }
 249 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get, pid_param_set, "%llu\n");
 250
 251 struct pid_param {
 252         char *name;
 253         void *value;
 254 };
 255
 256 static struct pid_param pid_files[] = {
 257         {"sample_rate_ms", &pid_params.sample_rate_ms},
 258         {"d_gain_pct", &pid_params.d_gain_pct},
 259         {"i_gain_pct", &pid_params.i_gain_pct},
 260         {"deadband", &pid_params.deadband},
 261         {"setpoint", &pid_params.setpoint},
 262         {"p_gain_pct", &pid_params.p_gain_pct},
 263         {NULL, NULL}
 264 };
 265
 266 static void __init intel_pstate_debug_expose_params(void)
 267 {
 268         struct dentry *debugfs_parent;
 269         int i = 0;
 270
 271         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 272         if (IS_ERR_OR_NULL(debugfs_parent))
 273                 return;
 274         while (pid_files[i].name) {
 275                 debugfs_create_file(pid_files[i].name, 0660,
 276                                     debugfs_parent, pid_files[i].value,
 277                                     &fops_pid_param);
 278                 i++;
 279         }
 280 }
 281
 282 /************************** debugfs end ************************/
 283
 284 /************************** sysfs begin ************************/
 285 #define show_one(file_name, object)                                     \
 286         static ssize_t show_##file_name                                 \
 287         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 288         {                                                               \
 289                 return sprintf(buf, "%u\n", limits.object);             \
 290         }
 291
 292 static ssize_t show_no_turbo(struct kobject *kobj,
 293                              struct attribute *attr, char *buf)
 294 {
 295         ssize_t ret;
 296
 297         update_turbo_state();
 298         if (limits.turbo_disabled)
 299                 ret = sprintf(buf, "%u\n", limits.turbo_disabled);
 300         else
 301                 ret = sprintf(buf, "%u\n", limits.no_turbo);
 302
 303         return ret;
 304 }
 305
 306 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 307                               const char *buf, size_t count)
 308 {
 309         unsigned int input;
 310         int ret;
 311
 312         ret = sscanf(buf, "%u", &input);
 313         if (ret != 1)
 314                 return -EINVAL;
 315
 316         update_turbo_state();
 317         if (limits.turbo_disabled) {
 318                 pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
 319                 return -EPERM;
 320         }
 321         limits.no_turbo = clamp_t(int, input, 0, 1);
 322
 323         return count;
 324 }
 325
 326 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 327                                   const char *buf, size_t count)
 328 {
 329         unsigned int input;
 330         int ret;
 331
 332         ret = sscanf(buf, "%u", &input);
 333         if (ret != 1)
 334                 return -EINVAL;
 335
 336         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 337         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 338         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 339
 340         return count;
 341 }
 342
 343 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 344                                   const char *buf, size_t count)
 345 {
 346         unsigned int input;
 347         int ret;
 348
 349         ret = sscanf(buf, "%u", &input);
 350         if (ret != 1)
 351                 return -EINVAL;
 352         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 353         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 354
 355         return count;
 356 }
 357
 358 show_one(max_perf_pct, max_perf_pct);
 359 show_one(min_perf_pct, min_perf_pct);
 360
 361 define_one_global_rw(no_turbo);
 362 define_one_global_rw(max_perf_pct);
 363 define_one_global_rw(min_perf_pct);
 364
 365 static struct attribute *intel_pstate_attributes[] = {
 366         &no_turbo.attr,
 367         &max_perf_pct.attr,
 368         &min_perf_pct.attr,
 369         NULL
 370 };
 371
 372 static struct attribute_group intel_pstate_attr_group = {
 373         .attrs = intel_pstate_attributes,
 374 };
 375
 376 static void __init intel_pstate_sysfs_expose_params(void)
 377 {
 378         struct kobject *intel_pstate_kobject;
 379         int rc;
 380
 381         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 382                                                 &cpu_subsys.dev_root->kobj);
 383         BUG_ON(!intel_pstate_kobject);
 384         rc = sysfs_create_group(intel_pstate_kobject, &intel_pstate_attr_group);
 385         BUG_ON(rc);
 386 }
 387
 388 /************************** sysfs end ************************/
 389 static int byt_get_min_pstate(void)
 390 {
 391         u64 value;
 392
 393         rdmsrl(BYT_RATIOS, value);
 394         return (value >> 8) & 0x7F;
 395 }
 396
 397 static int byt_get_max_pstate(void)
 398 {
 399         u64 value;
 400
 401         rdmsrl(BYT_RATIOS, value);
 402         return (value >> 16) & 0x7F;
 403 }
 404
 405 static int byt_get_turbo_pstate(void)
 406 {
 407         u64 value;
 408
 409         rdmsrl(BYT_TURBO_RATIOS, value);
 410         return value & 0x7F;
 411 }
 412
 413 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 414 {
 415         u64 val;
 416         int32_t vid_fp;
 417         u32 vid;
 418
 419         val = pstate << 8;
 420         if (limits.no_turbo && !limits.turbo_disabled)
 421                 val |= (u64)1 << 32;
 422
 423         vid_fp = cpudata->vid.min + mul_fp(
 424                 int_tofp(pstate - cpudata->pstate.min_pstate),
 425                 cpudata->vid.ratio);
 426
 427         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 428         vid = fp_toint(vid_fp);
 429
 430         if (pstate > cpudata->pstate.max_pstate)
 431                 vid = cpudata->vid.turbo;
 432
 433         val |= vid;
 434
 435         wrmsrl(MSR_IA32_PERF_CTL, val);
 436 }
 437
 438 #define BYT_BCLK_FREQS 5
 439 static int byt_freq_table[BYT_BCLK_FREQS] = { 833, 1000, 1333, 1167, 800};
 440
 441 static int byt_get_scaling(void)
 442 {
 443         u64 value;
 444         int i;
 445
 446         rdmsrl(MSR_FSB_FREQ, value);
 447         i = value & 0x3;
 448
 449         BUG_ON(i > BYT_BCLK_FREQS);
 450
 451         return byt_freq_table[i] * 100;
 452 }
 453
 454 static void byt_get_vid(struct cpudata *cpudata)
 455 {
 456         u64 value;
 457
 458         rdmsrl(BYT_VIDS, value);
 459         cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
 460         cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
 461         cpudata->vid.ratio = div_fp(
 462                 cpudata->vid.max - cpudata->vid.min,
 463                 int_tofp(cpudata->pstate.max_pstate -
 464                         cpudata->pstate.min_pstate));
 465
 466         rdmsrl(BYT_TURBO_VIDS, value);
 467         cpudata->vid.turbo = value & 0x7f;
 468 }
 469
 470 static int core_get_min_pstate(void)
 471 {
 472         u64 value;
 473
 474         rdmsrl(MSR_PLATFORM_INFO, value);
 475         return (value >> 40) & 0xFF;
 476 }
 477
 478 static int core_get_max_pstate(void)
 479 {
 480         u64 value;
 481
 482         rdmsrl(MSR_PLATFORM_INFO, value);
 483         return (value >> 8) & 0xFF;
 484 }
 485
 486 static int core_get_turbo_pstate(void)
 487 {
 488         u64 value;
 489         int nont, ret;
 490
 491         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 492         nont = core_get_max_pstate();
 493         ret = (value) & 255;
 494         if (ret <= nont)
 495                 ret = nont;
 496         return ret;
 497 }
 498
 499 static inline int core_get_scaling(void)
 500 {
 501         return 100000;
 502 }
 503
 504 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 505 {
 506         u64 val;
 507
 508         val = pstate << 8;
 509         if (limits.no_turbo && !limits.turbo_disabled)
 510                 val |= (u64)1 << 32;
 511
 512         wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 513 }
 514
 515 static struct cpu_defaults core_params = {
 516         .pid_policy = {
 517                 .sample_rate_ms = 10,
 518                 .deadband = 0,
 519                 .setpoint = 97,
 520                 .p_gain_pct = 20,
 521                 .d_gain_pct = 0,
 522                 .i_gain_pct = 0,
 523         },
 524         .funcs = {
 525                 .get_max = core_get_max_pstate,
 526                 .get_min = core_get_min_pstate,
 527                 .get_turbo = core_get_turbo_pstate,
 528                 .get_scaling = core_get_scaling,
 529                 .set = core_set_pstate,
 530         },
 531 };
 532
 533 static struct cpu_defaults byt_params = {
 534         .pid_policy = {
 535                 .sample_rate_ms = 10,
 536                 .deadband = 0,
 537                 .setpoint = 97,
 538                 .p_gain_pct = 14,
 539                 .d_gain_pct = 0,
 540                 .i_gain_pct = 4,
 541         },
 542         .funcs = {
 543                 .get_max = byt_get_max_pstate,
 544                 .get_min = byt_get_min_pstate,
 545                 .get_turbo = byt_get_turbo_pstate,
 546                 .set = byt_set_pstate,
 547                 .get_scaling = byt_get_scaling,
 548                 .get_vid = byt_get_vid,
 549         },
 550 };
 551
 552 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 553 {
 554         int max_perf = cpu->pstate.turbo_pstate;
 555         int max_perf_adj;
 556         int min_perf;
 557
 558         if (limits.no_turbo || limits.turbo_disabled)
 559                 max_perf = cpu->pstate.max_pstate;
 560
 561         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 562         *max = clamp_t(int, max_perf_adj,
 563                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 564
 565         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 566         *min = clamp_t(int, min_perf, cpu->pstate.min_pstate, max_perf);
 567 }
 568
 569 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 570 {
 571         int max_perf, min_perf;
 572
 573         update_turbo_state();
 574
 575         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 576
 577         pstate = clamp_t(int, pstate, min_perf, max_perf);
 578
 579         if (pstate == cpu->pstate.current_pstate)
 580                 return;
 581
 582         trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
 583
 584         cpu->pstate.current_pstate = pstate;
 585
 586         pstate_funcs.set(cpu, pstate);
 587 }
 588
 589 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 590 {
 591         cpu->pstate.min_pstate = pstate_funcs.get_min();
 592         cpu->pstate.max_pstate = pstate_funcs.get_max();
 593         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 594         cpu->pstate.scaling = pstate_funcs.get_scaling();
 595
 596         if (pstate_funcs.get_vid)
 597                 pstate_funcs.get_vid(cpu);
 598         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 599 }
 600
 601 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
 602 {
 603         struct sample *sample = &cpu->sample;
 604         int64_t core_pct;
 605
 606         core_pct = int_tofp(sample->aperf) * int_tofp(100);
 607         core_pct = div64_u64(core_pct, int_tofp(sample->mperf));
 608
 609         sample->freq = fp_toint(
 610                 mul_fp(int_tofp(
 611                         cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
 612                         core_pct));
 613
 614         sample->core_pct_busy = (int32_t)core_pct;
 615 }
 616
 617 static inline void intel_pstate_sample(struct cpudata *cpu)
 618 {
 619         u64 aperf, mperf;
 620         unsigned long flags;
 621
 622         local_irq_save(flags);
 623         rdmsrl(MSR_IA32_APERF, aperf);
 624         rdmsrl(MSR_IA32_MPERF, mperf);
 625         local_irq_restore(flags);
 626
 627         cpu->last_sample_time = cpu->sample.time;
 628         cpu->sample.time = ktime_get();
 629         cpu->sample.aperf = aperf;
 630         cpu->sample.mperf = mperf;
 631         cpu->sample.aperf -= cpu->prev_aperf;
 632         cpu->sample.mperf -= cpu->prev_mperf;
 633
 634         intel_pstate_calc_busy(cpu);
 635
 636         cpu->prev_aperf = aperf;
 637         cpu->prev_mperf = mperf;
 638 }
 639
 640 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 641 {
 642         int delay;
 643
 644         delay = msecs_to_jiffies(pid_params.sample_rate_ms);
 645         mod_timer_pinned(&cpu->timer, jiffies + delay);
 646 }
 647
 648 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 649 {
 650         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 651         u32 duration_us;
 652         u32 sample_time;
 653
 654         core_busy = cpu->sample.core_pct_busy;
 655         max_pstate = int_tofp(cpu->pstate.max_pstate);
 656         current_pstate = int_tofp(cpu->pstate.current_pstate);
 657         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 658
 659         sample_time = pid_params.sample_rate_ms  * USEC_PER_MSEC;
 660         duration_us = (u32) ktime_us_delta(cpu->sample.time,
 661                                            cpu->last_sample_time);
 662         if (duration_us > sample_time * 3) {
 663                 sample_ratio = div_fp(int_tofp(sample_time),
 664                                       int_tofp(duration_us));
 665                 core_busy = mul_fp(core_busy, sample_ratio);
 666         }
 667
 668         return core_busy;
 669 }
 670
 671 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 672 {
 673         int32_t busy_scaled;
 674         struct _pid *pid;
 675         signed int ctl;
 676
 677         pid = &cpu->pid;
 678         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 679
 680         ctl = pid_calc(pid, busy_scaled);
 681
 682         /* Negative values of ctl increase the pstate and vice versa */
 683         intel_pstate_set_pstate(cpu, cpu->pstate.current_pstate - ctl);
 684 }
 685
 686 static void intel_pstate_timer_func(unsigned long __data)
 687 {
 688         struct cpudata *cpu = (struct cpudata *) __data;
 689         struct sample *sample;
 690
 691         intel_pstate_sample(cpu);
 692
 693         sample = &cpu->sample;
 694
 695         intel_pstate_adjust_busy_pstate(cpu);
 696
 697         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 698                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 699                         cpu->pstate.current_pstate,
 700                         sample->mperf,
 701                         sample->aperf,
 702                         sample->freq);
 703
 704         intel_pstate_set_sample_time(cpu);
 705 }
 706
 707 #define ICPU(model, policy) \
 708         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 709                         (unsigned long)&policy }
 710
 711 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 712         ICPU(0x2a, core_params),
 713         ICPU(0x2d, core_params),
 714         ICPU(0x37, byt_params),
 715         ICPU(0x3a, core_params),
 716         ICPU(0x3c, core_params),
 717         ICPU(0x3d, core_params),
 718         ICPU(0x3e, core_params),
 719         ICPU(0x3f, core_params),
 720         ICPU(0x45, core_params),
 721         ICPU(0x46, core_params),
 722         ICPU(0x4c, byt_params),
 723         ICPU(0x4f, core_params),
 724         ICPU(0x56, core_params),
 725         {}
 726 };
 727 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 728
 729 static int intel_pstate_init_cpu(unsigned int cpunum)
 730 {
 731         struct cpudata *cpu;
 732
 733         if (!all_cpu_data[cpunum])
 734                 all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata),
 735                                                GFP_KERNEL);
 736         if (!all_cpu_data[cpunum])
 737                 return -ENOMEM;
 738
 739         cpu = all_cpu_data[cpunum];
 740
 741         cpu->cpu = cpunum;
 742         intel_pstate_get_cpu_pstates(cpu);
 743
 744         init_timer_deferrable(&cpu->timer);
 745         cpu->timer.function = intel_pstate_timer_func;
 746         cpu->timer.data = (unsigned long)cpu;
 747         cpu->timer.expires = jiffies + HZ/100;
 748         intel_pstate_busy_pid_reset(cpu);
 749         intel_pstate_sample(cpu);
 750
 751         add_timer_on(&cpu->timer, cpunum);
 752
 753         pr_debug("Intel pstate controlling: cpu %d\n", cpunum);
 754
 755         return 0;
 756 }
 757
 758 static unsigned int intel_pstate_get(unsigned int cpu_num)
 759 {
 760         struct sample *sample;
 761         struct cpudata *cpu;
 762
 763         cpu = all_cpu_data[cpu_num];
 764         if (!cpu)
 765                 return 0;
 766         sample = &cpu->sample;
 767         return sample->freq;
 768 }
 769
 770 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 771 {
 772         if (!policy->cpuinfo.max_freq)
 773                 return -ENODEV;
 774
 775         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 776                 limits.min_perf_pct = 100;
 777                 limits.min_perf = int_tofp(1);
 778                 limits.max_policy_pct = 100;
 779                 limits.max_perf_pct = 100;
 780                 limits.max_perf = int_tofp(1);
 781                 limits.no_turbo = 0;
 782                 return 0;
 783         }
 784         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 785         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 786         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 787
 788         limits.max_policy_pct = (policy->max * 100) / policy->cpuinfo.max_freq;
 789         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 790         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 791         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 792
 793         return 0;
 794 }
 795
 796 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 797 {
 798         cpufreq_verify_within_cpu_limits(policy);
 799
 800         if (policy->policy != CPUFREQ_POLICY_POWERSAVE &&
 801             policy->policy != CPUFREQ_POLICY_PERFORMANCE)
 802                 return -EINVAL;
 803
 804         return 0;
 805 }
 806
 807 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
 808 {
 809         int cpu_num = policy->cpu;
 810         struct cpudata *cpu = all_cpu_data[cpu_num];
 811
 812         pr_info("intel_pstate CPU %d exiting\n", cpu_num);
 813
 814         del_timer_sync(&all_cpu_data[cpu_num]->timer);
 815         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 816 }
 817
 818 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 819 {
 820         struct cpudata *cpu;
 821         int rc;
 822
 823         rc = intel_pstate_init_cpu(policy->cpu);
 824         if (rc)
 825                 return rc;
 826
 827         cpu = all_cpu_data[policy->cpu];
 828
 829         if (limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 830                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 831         else
 832                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 833
 834         policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
 835         policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
 836
 837         /* cpuinfo and default policy values */
 838         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
 839         policy->cpuinfo.max_freq =
 840                 cpu->pstate.turbo_pstate * cpu->pstate.scaling;
 841         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 842         cpumask_set_cpu(policy->cpu, policy->cpus);
 843
 844         return 0;
 845 }
 846
 847 static struct cpufreq_driver intel_pstate_driver = {
 848         .flags          = CPUFREQ_CONST_LOOPS,
 849         .verify         = intel_pstate_verify_policy,
 850         .setpolicy      = intel_pstate_set_policy,
 851         .get            = intel_pstate_get,
 852         .init           = intel_pstate_cpu_init,
 853         .stop_cpu       = intel_pstate_stop_cpu,
 854         .name           = "intel_pstate",
 855 };
 856
 857 static int __initdata no_load;
 858
 859 static int intel_pstate_msrs_not_valid(void)
 860 {
 861         /* Check that all the msr's we are using are valid. */
 862         u64 aperf, mperf, tmp;
 863
 864         rdmsrl(MSR_IA32_APERF, aperf);
 865         rdmsrl(MSR_IA32_MPERF, mperf);
 866
 867         if (!pstate_funcs.get_max() ||
 868             !pstate_funcs.get_min() ||
 869             !pstate_funcs.get_turbo())
 870                 return -ENODEV;
 871
 872         rdmsrl(MSR_IA32_APERF, tmp);
 873         if (!(tmp - aperf))
 874                 return -ENODEV;
 875
 876         rdmsrl(MSR_IA32_MPERF, tmp);
 877         if (!(tmp - mperf))
 878                 return -ENODEV;
 879
 880         return 0;
 881 }
 882
 883 static void copy_pid_params(struct pstate_adjust_policy *policy)
 884 {
 885         pid_params.sample_rate_ms = policy->sample_rate_ms;
 886         pid_params.p_gain_pct = policy->p_gain_pct;
 887         pid_params.i_gain_pct = policy->i_gain_pct;
 888         pid_params.d_gain_pct = policy->d_gain_pct;
 889         pid_params.deadband = policy->deadband;
 890         pid_params.setpoint = policy->setpoint;
 891 }
 892
 893 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 894 {
 895         pstate_funcs.get_max   = funcs->get_max;
 896         pstate_funcs.get_min   = funcs->get_min;
 897         pstate_funcs.get_turbo = funcs->get_turbo;
 898         pstate_funcs.get_scaling = funcs->get_scaling;
 899         pstate_funcs.set       = funcs->set;
 900         pstate_funcs.get_vid   = funcs->get_vid;
 901 }
 902
 903 #if IS_ENABLED(CONFIG_ACPI)
 904 #include <acpi/processor.h>
 905
 906 static bool intel_pstate_no_acpi_pss(void)
 907 {
 908         int i;
 909
 910         for_each_possible_cpu(i) {
 911                 acpi_status status;
 912                 union acpi_object *pss;
 913                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 914                 struct acpi_processor *pr = per_cpu(processors, i);
 915
 916                 if (!pr)
 917                         continue;
 918
 919                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 920                 if (ACPI_FAILURE(status))
 921                         continue;
 922
 923                 pss = buffer.pointer;
 924                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 925                         kfree(pss);
 926                         return false;
 927                 }
 928
 929                 kfree(pss);
 930         }
 931
 932         return true;
 933 }
 934
 935 struct hw_vendor_info {
 936         u16  valid;
 937         char oem_id[ACPI_OEM_ID_SIZE];
 938         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 939 };
 940
 941 /* Hardware vendor-specific info that has its own power management modes */
 942 static struct hw_vendor_info vendor_info[] = {
 943         {1, "HP    ", "ProLiant"},
 944         {0, "", ""},
 945 };
 946
 947 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 948 {
 949         struct acpi_table_header hdr;
 950         struct hw_vendor_info *v_info;
 951
 952         if (acpi_disabled ||
 953             ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 954                 return false;
 955
 956         for (v_info = vendor_info; v_info->valid; v_info++) {
 957                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE) &&
 958                     !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) &&
 959                     intel_pstate_no_acpi_pss())
 960                         return true;
 961         }
 962
 963         return false;
 964 }
 965 #else /* CONFIG_ACPI not enabled */
 966 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 967 #endif /* CONFIG_ACPI */
 968
 969 static int __init intel_pstate_init(void)
 970 {
 971         int cpu, rc = 0;
 972         const struct x86_cpu_id *id;
 973         struct cpu_defaults *cpu_info;
 974
 975         if (no_load)
 976                 return -ENODEV;
 977
 978         id = x86_match_cpu(intel_pstate_cpu_ids);
 979         if (!id)
 980                 return -ENODEV;
 981
 982         /*
 983          * The Intel pstate driver will be ignored if the platform
 984          * firmware has its own power management modes.
 985          */
 986         if (intel_pstate_platform_pwr_mgmt_exists())
 987                 return -ENODEV;
 988
 989         cpu_info = (struct cpu_defaults *)id->driver_data;
 990
 991         copy_pid_params(&cpu_info->pid_policy);
 992         copy_cpu_funcs(&cpu_info->funcs);
 993
 994         if (intel_pstate_msrs_not_valid())
 995                 return -ENODEV;
 996
 997         pr_info("Intel P-state driver initializing.\n");
 998
 999         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
1000         if (!all_cpu_data)
1001                 return -ENOMEM;
1002
1003         rc = cpufreq_register_driver(&intel_pstate_driver);
1004         if (rc)
1005                 goto out;
1006
1007         intel_pstate_debug_expose_params();
1008         intel_pstate_sysfs_expose_params();
1009
1010         return rc;
1011 out:
1012         get_online_cpus();
1013         for_each_online_cpu(cpu) {
1014                 if (all_cpu_data[cpu]) {
1015                         del_timer_sync(&all_cpu_data[cpu]->timer);
1016                         kfree(all_cpu_data[cpu]);
1017                 }
1018         }
1019
1020         put_online_cpus();
1021         vfree(all_cpu_data);
1022         return -ENODEV;
1023 }
1024 device_initcall(intel_pstate_init);
1025
1026 static int __init intel_pstate_setup(char *str)
1027 {
1028         if (!str)
1029                 return -EINVAL;
1030
1031         if (!strcmp(str, "disable"))
1032                 no_load = 1;
1033         return 0;
1034 }
1035 early_param("intel_pstate", intel_pstate_setup);
1036
1037 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1038 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1039 MODULE_LICENSE("GPL");