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 <trace/events/power.h>
  29
  30 #include <asm/div64.h>
  31 #include <asm/msr.h>
  32 #include <asm/cpu_device_id.h>
  33
  34 #define SAMPLE_COUNT            3
  35
  36 #define FRAC_BITS 8
  37 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  38 #define fp_toint(X) ((X) >> FRAC_BITS)
  39
  40 static inline int32_t mul_fp(int32_t x, int32_t y)
  41 {
  42         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  43 }
  44
  45 static inline int32_t div_fp(int32_t x, int32_t y)
  46 {
  47         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  48 }
  49
  50 struct sample {
  51         int core_pct_busy;
  52         u64 aperf;
  53         u64 mperf;
  54         int freq;
  55 };
  56
  57 struct pstate_data {
  58         int     current_pstate;
  59         int     min_pstate;
  60         int     max_pstate;
  61         int     turbo_pstate;
  62 };
  63
  64 struct _pid {
  65         int setpoint;
  66         int32_t integral;
  67         int32_t p_gain;
  68         int32_t i_gain;
  69         int32_t d_gain;
  70         int deadband;
  71         int last_err;
  72 };
  73
  74 struct cpudata {
  75         int cpu;
  76
  77         char name[64];
  78
  79         struct timer_list timer;
  80
  81         struct pstate_adjust_policy *pstate_policy;
  82         struct pstate_data pstate;
  83         struct _pid pid;
  84
  85         int min_pstate_count;
  86
  87         u64     prev_aperf;
  88         u64     prev_mperf;
  89         int     sample_ptr;
  90         struct sample samples[SAMPLE_COUNT];
  91 };
  92
  93 static struct cpudata **all_cpu_data;
  94 struct pstate_adjust_policy {
  95         int sample_rate_ms;
  96         int deadband;
  97         int setpoint;
  98         int p_gain_pct;
  99         int d_gain_pct;
 100         int i_gain_pct;
 101 };
 102
 103 static struct pstate_adjust_policy default_policy = {
 104         .sample_rate_ms = 10,
 105         .deadband = 0,
 106         .setpoint = 97,
 107         .p_gain_pct = 20,
 108         .d_gain_pct = 0,
 109         .i_gain_pct = 0,
 110 };
 111
 112 struct perf_limits {
 113         int no_turbo;
 114         int max_perf_pct;
 115         int min_perf_pct;
 116         int32_t max_perf;
 117         int32_t min_perf;
 118         int max_policy_pct;
 119         int max_sysfs_pct;
 120 };
 121
 122 static struct perf_limits limits = {
 123         .no_turbo = 0,
 124         .max_perf_pct = 100,
 125         .max_perf = int_tofp(1),
 126         .min_perf_pct = 0,
 127         .min_perf = 0,
 128         .max_policy_pct = 100,
 129         .max_sysfs_pct = 100,
 130 };
 131
 132 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 133                         int deadband, int integral) {
 134         pid->setpoint = setpoint;
 135         pid->deadband  = deadband;
 136         pid->integral  = int_tofp(integral);
 137         pid->last_err  = setpoint - busy;
 138 }
 139
 140 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 141 {
 142         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 143 }
 144
 145 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 146 {
 147         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 148 }
 149
 150 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 151 {
 152
 153         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 154 }
 155
 156 static signed int pid_calc(struct _pid *pid, int busy)
 157 {
 158         signed int err, result;
 159         int32_t pterm, dterm, fp_error;
 160         int32_t integral_limit;
 161
 162         err = pid->setpoint - busy;
 163         fp_error = int_tofp(err);
 164
 165         if (abs(err) <= pid->deadband)
 166                 return 0;
 167
 168         pterm = mul_fp(pid->p_gain, fp_error);
 169
 170         pid->integral += fp_error;
 171
 172         /* limit the integral term */
 173         integral_limit = int_tofp(30);
 174         if (pid->integral > integral_limit)
 175                 pid->integral = integral_limit;
 176         if (pid->integral < -integral_limit)
 177                 pid->integral = -integral_limit;
 178
 179         dterm = mul_fp(pid->d_gain, (err - pid->last_err));
 180         pid->last_err = err;
 181
 182         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 183
 184         return (signed int)fp_toint(result);
 185 }
 186
 187 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 188 {
 189         pid_p_gain_set(&cpu->pid, cpu->pstate_policy->p_gain_pct);
 190         pid_d_gain_set(&cpu->pid, cpu->pstate_policy->d_gain_pct);
 191         pid_i_gain_set(&cpu->pid, cpu->pstate_policy->i_gain_pct);
 192
 193         pid_reset(&cpu->pid,
 194                 cpu->pstate_policy->setpoint,
 195                 100,
 196                 cpu->pstate_policy->deadband,
 197                 0);
 198 }
 199
 200 static inline void intel_pstate_reset_all_pid(void)
 201 {
 202         unsigned int cpu;
 203         for_each_online_cpu(cpu) {
 204                 if (all_cpu_data[cpu])
 205                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 206         }
 207 }
 208
 209 /************************** debugfs begin ************************/
 210 static int pid_param_set(void *data, u64 val)
 211 {
 212         *(u32 *)data = val;
 213         intel_pstate_reset_all_pid();
 214         return 0;
 215 }
 216 static int pid_param_get(void *data, u64 *val)
 217 {
 218         *val = *(u32 *)data;
 219         return 0;
 220 }
 221 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 222                         pid_param_set, "%llu\n");
 223
 224 struct pid_param {
 225         char *name;
 226         void *value;
 227 };
 228
 229 static struct pid_param pid_files[] = {
 230         {"sample_rate_ms", &default_policy.sample_rate_ms},
 231         {"d_gain_pct", &default_policy.d_gain_pct},
 232         {"i_gain_pct", &default_policy.i_gain_pct},
 233         {"deadband", &default_policy.deadband},
 234         {"setpoint", &default_policy.setpoint},
 235         {"p_gain_pct", &default_policy.p_gain_pct},
 236         {NULL, NULL}
 237 };
 238
 239 static struct dentry *debugfs_parent;
 240 static void intel_pstate_debug_expose_params(void)
 241 {
 242         int i = 0;
 243
 244         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 245         if (IS_ERR_OR_NULL(debugfs_parent))
 246                 return;
 247         while (pid_files[i].name) {
 248                 debugfs_create_file(pid_files[i].name, 0660,
 249                                 debugfs_parent, pid_files[i].value,
 250                                 &fops_pid_param);
 251                 i++;
 252         }
 253 }
 254
 255 /************************** debugfs end ************************/
 256
 257 /************************** sysfs begin ************************/
 258 #define show_one(file_name, object)                                     \
 259         static ssize_t show_##file_name                                 \
 260         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 261         {                                                               \
 262                 return sprintf(buf, "%u\n", limits.object);             \
 263         }
 264
 265 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 266                                 const char *buf, size_t count)
 267 {
 268         unsigned int input;
 269         int ret;
 270         ret = sscanf(buf, "%u", &input);
 271         if (ret != 1)
 272                 return -EINVAL;
 273         limits.no_turbo = clamp_t(int, input, 0 , 1);
 274
 275         return count;
 276 }
 277
 278 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 279                                 const char *buf, size_t count)
 280 {
 281         unsigned int input;
 282         int ret;
 283         ret = sscanf(buf, "%u", &input);
 284         if (ret != 1)
 285                 return -EINVAL;
 286
 287         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 288         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 289         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 290         return count;
 291 }
 292
 293 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 294                                 const char *buf, size_t count)
 295 {
 296         unsigned int input;
 297         int ret;
 298         ret = sscanf(buf, "%u", &input);
 299         if (ret != 1)
 300                 return -EINVAL;
 301         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 302         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 303
 304         return count;
 305 }
 306
 307 show_one(no_turbo, no_turbo);
 308 show_one(max_perf_pct, max_perf_pct);
 309 show_one(min_perf_pct, min_perf_pct);
 310
 311 define_one_global_rw(no_turbo);
 312 define_one_global_rw(max_perf_pct);
 313 define_one_global_rw(min_perf_pct);
 314
 315 static struct attribute *intel_pstate_attributes[] = {
 316         &no_turbo.attr,
 317         &max_perf_pct.attr,
 318         &min_perf_pct.attr,
 319         NULL
 320 };
 321
 322 static struct attribute_group intel_pstate_attr_group = {
 323         .attrs = intel_pstate_attributes,
 324 };
 325 static struct kobject *intel_pstate_kobject;
 326
 327 static void intel_pstate_sysfs_expose_params(void)
 328 {
 329         int rc;
 330
 331         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 332                                                 &cpu_subsys.dev_root->kobj);
 333         BUG_ON(!intel_pstate_kobject);
 334         rc = sysfs_create_group(intel_pstate_kobject,
 335                                 &intel_pstate_attr_group);
 336         BUG_ON(rc);
 337 }
 338
 339 /************************** sysfs end ************************/
 340
 341 static int intel_pstate_min_pstate(void)
 342 {
 343         u64 value;
 344         rdmsrl(MSR_PLATFORM_INFO, value);
 345         return (value >> 40) & 0xFF;
 346 }
 347
 348 static int intel_pstate_max_pstate(void)
 349 {
 350         u64 value;
 351         rdmsrl(MSR_PLATFORM_INFO, value);
 352         return (value >> 8) & 0xFF;
 353 }
 354
 355 static int intel_pstate_turbo_pstate(void)
 356 {
 357         u64 value;
 358         int nont, ret;
 359         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 360         nont = intel_pstate_max_pstate();
 361         ret = ((value) & 255);
 362         if (ret <= nont)
 363                 ret = nont;
 364         return ret;
 365 }
 366
 367 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 368 {
 369         int max_perf = cpu->pstate.turbo_pstate;
 370         int min_perf;
 371         if (limits.no_turbo)
 372                 max_perf = cpu->pstate.max_pstate;
 373
 374         max_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 375         *max = clamp_t(int, max_perf,
 376                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 377
 378         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 379         *min = clamp_t(int, min_perf,
 380                         cpu->pstate.min_pstate, max_perf);
 381 }
 382
 383 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 384 {
 385         int max_perf, min_perf;
 386
 387         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 388
 389         pstate = clamp_t(int, pstate, min_perf, max_perf);
 390
 391         if (pstate == cpu->pstate.current_pstate)
 392                 return;
 393
 394         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 395
 396         cpu->pstate.current_pstate = pstate;
 397         wrmsrl(MSR_IA32_PERF_CTL, pstate << 8);
 398
 399 }
 400
 401 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 402 {
 403         int target;
 404         target = cpu->pstate.current_pstate + steps;
 405
 406         intel_pstate_set_pstate(cpu, target);
 407 }
 408
 409 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 410 {
 411         int target;
 412         target = cpu->pstate.current_pstate - steps;
 413         intel_pstate_set_pstate(cpu, target);
 414 }
 415
 416 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 417 {
 418         sprintf(cpu->name, "Intel 2nd generation core");
 419
 420         cpu->pstate.min_pstate = intel_pstate_min_pstate();
 421         cpu->pstate.max_pstate = intel_pstate_max_pstate();
 422         cpu->pstate.turbo_pstate = intel_pstate_turbo_pstate();
 423
 424         /*
 425          * goto max pstate so we don't slow up boot if we are built-in if we are
 426          * a module we will take care of it during normal operation
 427          */
 428         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 429 }
 430
 431 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 432                                         struct sample *sample)
 433 {
 434         u64 core_pct;
 435         core_pct = div64_u64(sample->aperf * 100, sample->mperf);
 436         sample->freq = cpu->pstate.max_pstate * core_pct * 1000;
 437
 438         sample->core_pct_busy = core_pct;
 439 }
 440
 441 static inline void intel_pstate_sample(struct cpudata *cpu)
 442 {
 443         u64 aperf, mperf;
 444
 445         rdmsrl(MSR_IA32_APERF, aperf);
 446         rdmsrl(MSR_IA32_MPERF, mperf);
 447         cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
 448         cpu->samples[cpu->sample_ptr].aperf = aperf;
 449         cpu->samples[cpu->sample_ptr].mperf = mperf;
 450         cpu->samples[cpu->sample_ptr].aperf -= cpu->prev_aperf;
 451         cpu->samples[cpu->sample_ptr].mperf -= cpu->prev_mperf;
 452
 453         intel_pstate_calc_busy(cpu, &cpu->samples[cpu->sample_ptr]);
 454
 455         cpu->prev_aperf = aperf;
 456         cpu->prev_mperf = mperf;
 457 }
 458
 459 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 460 {
 461         int sample_time, delay;
 462
 463         sample_time = cpu->pstate_policy->sample_rate_ms;
 464         delay = msecs_to_jiffies(sample_time);
 465         mod_timer_pinned(&cpu->timer, jiffies + delay);
 466 }
 467
 468 static inline int intel_pstate_get_scaled_busy(struct cpudata *cpu)
 469 {
 470         int32_t busy_scaled;
 471         int32_t core_busy, max_pstate, current_pstate;
 472
 473         core_busy = int_tofp(cpu->samples[cpu->sample_ptr].core_pct_busy);
 474         max_pstate = int_tofp(cpu->pstate.max_pstate);
 475         current_pstate = int_tofp(cpu->pstate.current_pstate);
 476         busy_scaled = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 477
 478         return fp_toint(busy_scaled);
 479 }
 480
 481 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 482 {
 483         int busy_scaled;
 484         struct _pid *pid;
 485         signed int ctl = 0;
 486         int steps;
 487
 488         pid = &cpu->pid;
 489         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 490
 491         ctl = pid_calc(pid, busy_scaled);
 492
 493         steps = abs(ctl);
 494         if (ctl < 0)
 495                 intel_pstate_pstate_increase(cpu, steps);
 496         else
 497                 intel_pstate_pstate_decrease(cpu, steps);
 498 }
 499
 500 static void intel_pstate_timer_func(unsigned long __data)
 501 {
 502         struct cpudata *cpu = (struct cpudata *) __data;
 503
 504         intel_pstate_sample(cpu);
 505         intel_pstate_adjust_busy_pstate(cpu);
 506
 507         if (cpu->pstate.current_pstate == cpu->pstate.min_pstate) {
 508                 cpu->min_pstate_count++;
 509                 if (!(cpu->min_pstate_count % 5)) {
 510                         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 511                 }
 512         } else
 513                 cpu->min_pstate_count = 0;
 514
 515         intel_pstate_set_sample_time(cpu);
 516 }
 517
 518 #define ICPU(model, policy) \
 519         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&policy }
 520
 521 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 522         ICPU(0x2a, default_policy),
 523         ICPU(0x2d, default_policy),
 524         ICPU(0x3a, default_policy),
 525         ICPU(0x3c, default_policy),
 526         ICPU(0x3e, default_policy),
 527         ICPU(0x3f, default_policy),
 528         ICPU(0x45, default_policy),
 529         ICPU(0x46, default_policy),
 530         {}
 531 };
 532 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 533
 534 static int intel_pstate_init_cpu(unsigned int cpunum)
 535 {
 536
 537         const struct x86_cpu_id *id;
 538         struct cpudata *cpu;
 539
 540         id = x86_match_cpu(intel_pstate_cpu_ids);
 541         if (!id)
 542                 return -ENODEV;
 543
 544         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 545         if (!all_cpu_data[cpunum])
 546                 return -ENOMEM;
 547
 548         cpu = all_cpu_data[cpunum];
 549
 550         intel_pstate_get_cpu_pstates(cpu);
 551
 552         cpu->cpu = cpunum;
 553         cpu->pstate_policy =
 554                 (struct pstate_adjust_policy *)id->driver_data;
 555         init_timer_deferrable(&cpu->timer);
 556         cpu->timer.function = intel_pstate_timer_func;
 557         cpu->timer.data =
 558                 (unsigned long)cpu;
 559         cpu->timer.expires = jiffies + HZ/100;
 560         intel_pstate_busy_pid_reset(cpu);
 561         intel_pstate_sample(cpu);
 562         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 563
 564         add_timer_on(&cpu->timer, cpunum);
 565
 566         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 567
 568         return 0;
 569 }
 570
 571 static unsigned int intel_pstate_get(unsigned int cpu_num)
 572 {
 573         struct sample *sample;
 574         struct cpudata *cpu;
 575
 576         cpu = all_cpu_data[cpu_num];
 577         if (!cpu)
 578                 return 0;
 579         sample = &cpu->samples[cpu->sample_ptr];
 580         return sample->freq;
 581 }
 582
 583 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 584 {
 585         struct cpudata *cpu;
 586
 587         cpu = all_cpu_data[policy->cpu];
 588
 589         if (!policy->cpuinfo.max_freq)
 590                 return -ENODEV;
 591
 592         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 593                 limits.min_perf_pct = 100;
 594                 limits.min_perf = int_tofp(1);
 595                 limits.max_perf_pct = 100;
 596                 limits.max_perf = int_tofp(1);
 597                 limits.no_turbo = 0;
 598                 return 0;
 599         }
 600         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 601         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 602         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 603
 604         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 605         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 606         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 607         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 608
 609         return 0;
 610 }
 611
 612 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 613 {
 614         cpufreq_verify_within_limits(policy,
 615                                 policy->cpuinfo.min_freq,
 616                                 policy->cpuinfo.max_freq);
 617
 618         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 619                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 620                 return -EINVAL;
 621
 622         return 0;
 623 }
 624
 625 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 626 {
 627         int cpu = policy->cpu;
 628
 629         del_timer(&all_cpu_data[cpu]->timer);
 630         kfree(all_cpu_data[cpu]);
 631         all_cpu_data[cpu] = NULL;
 632         return 0;
 633 }
 634
 635 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 636 {
 637         int rc, min_pstate, max_pstate;
 638         struct cpudata *cpu;
 639
 640         rc = intel_pstate_init_cpu(policy->cpu);
 641         if (rc)
 642                 return rc;
 643
 644         cpu = all_cpu_data[policy->cpu];
 645
 646         if (!limits.no_turbo &&
 647                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 648                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 649         else
 650                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 651
 652         intel_pstate_get_min_max(cpu, &min_pstate, &max_pstate);
 653         policy->min = min_pstate * 100000;
 654         policy->max = max_pstate * 100000;
 655
 656         /* cpuinfo and default policy values */
 657         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 658         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 659         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 660         cpumask_set_cpu(policy->cpu, policy->cpus);
 661
 662         return 0;
 663 }
 664
 665 static struct cpufreq_driver intel_pstate_driver = {
 666         .flags          = CPUFREQ_CONST_LOOPS,
 667         .verify         = intel_pstate_verify_policy,
 668         .setpolicy      = intel_pstate_set_policy,
 669         .get            = intel_pstate_get,
 670         .init           = intel_pstate_cpu_init,
 671         .exit           = intel_pstate_cpu_exit,
 672         .name           = "intel_pstate",
 673 };
 674
 675 static int __initdata no_load;
 676
 677 static int intel_pstate_msrs_not_valid(void)
 678 {
 679         /* Check that all the msr's we are using are valid. */
 680         u64 aperf, mperf, tmp;
 681
 682         rdmsrl(MSR_IA32_APERF, aperf);
 683         rdmsrl(MSR_IA32_MPERF, mperf);
 684
 685         if (!intel_pstate_min_pstate() ||
 686                 !intel_pstate_max_pstate() ||
 687                 !intel_pstate_turbo_pstate())
 688                 return -ENODEV;
 689
 690         rdmsrl(MSR_IA32_APERF, tmp);
 691         if (!(tmp - aperf))
 692                 return -ENODEV;
 693
 694         rdmsrl(MSR_IA32_MPERF, tmp);
 695         if (!(tmp - mperf))
 696                 return -ENODEV;
 697
 698         return 0;
 699 }
 700 static int __init intel_pstate_init(void)
 701 {
 702         int cpu, rc = 0;
 703         const struct x86_cpu_id *id;
 704
 705         if (no_load)
 706                 return -ENODEV;
 707
 708         id = x86_match_cpu(intel_pstate_cpu_ids);
 709         if (!id)
 710                 return -ENODEV;
 711
 712         if (intel_pstate_msrs_not_valid())
 713                 return -ENODEV;
 714
 715         pr_info("Intel P-state driver initializing.\n");
 716
 717         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 718         if (!all_cpu_data)
 719                 return -ENOMEM;
 720
 721         rc = cpufreq_register_driver(&intel_pstate_driver);
 722         if (rc)
 723                 goto out;
 724
 725         intel_pstate_debug_expose_params();
 726         intel_pstate_sysfs_expose_params();
 727         return rc;
 728 out:
 729         get_online_cpus();
 730         for_each_online_cpu(cpu) {
 731                 if (all_cpu_data[cpu]) {
 732                         del_timer_sync(&all_cpu_data[cpu]->timer);
 733                         kfree(all_cpu_data[cpu]);
 734                 }
 735         }
 736
 737         put_online_cpus();
 738         vfree(all_cpu_data);
 739         return -ENODEV;
 740 }
 741 device_initcall(intel_pstate_init);
 742
 743 static int __init intel_pstate_setup(char *str)
 744 {
 745         if (!str)
 746                 return -EINVAL;
 747
 748         if (!strcmp(str, "disable"))
 749                 no_load = 1;
 750         return 0;
 751 }
 752 early_param("intel_pstate", intel_pstate_setup);
 753
 754 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
 755 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
 756 MODULE_LICENSE("GPL");