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 BYT_RATIOS      0x66a
  37
  38 #define FRAC_BITS 8
  39 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  40 #define fp_toint(X) ((X) >> FRAC_BITS)
  41
  42 static inline int32_t mul_fp(int32_t x, int32_t y)
  43 {
  44         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  45 }
  46
  47 static inline int32_t div_fp(int32_t x, int32_t y)
  48 {
  49         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  50 }
  51
  52 struct sample {
  53         int32_t core_pct_busy;
  54         u64 aperf;
  55         u64 mperf;
  56         int freq;
  57 };
  58
  59 struct pstate_data {
  60         int     current_pstate;
  61         int     min_pstate;
  62         int     max_pstate;
  63         int     turbo_pstate;
  64 };
  65
  66 struct _pid {
  67         int setpoint;
  68         int32_t integral;
  69         int32_t p_gain;
  70         int32_t i_gain;
  71         int32_t d_gain;
  72         int deadband;
  73         int32_t last_err;
  74 };
  75
  76 struct cpudata {
  77         int cpu;
  78
  79         char name[64];
  80
  81         struct timer_list timer;
  82
  83         struct pstate_data pstate;
  84         struct _pid pid;
  85
  86         int min_pstate_count;
  87
  88         u64     prev_aperf;
  89         u64     prev_mperf;
  90         int     sample_ptr;
  91         struct sample samples[SAMPLE_COUNT];
  92 };
  93
  94 static struct cpudata **all_cpu_data;
  95 struct pstate_adjust_policy {
  96         int sample_rate_ms;
  97         int deadband;
  98         int setpoint;
  99         int p_gain_pct;
 100         int d_gain_pct;
 101         int i_gain_pct;
 102 };
 103
 104 struct pstate_funcs {
 105         int (*get_max)(void);
 106         int (*get_min)(void);
 107         int (*get_turbo)(void);
 108         void (*set)(int pstate);
 109 };
 110
 111 struct cpu_defaults {
 112         struct pstate_adjust_policy pid_policy;
 113         struct pstate_funcs funcs;
 114 };
 115
 116 static struct pstate_adjust_policy pid_params;
 117 static struct pstate_funcs pstate_funcs;
 118
 119 struct perf_limits {
 120         int no_turbo;
 121         int max_perf_pct;
 122         int min_perf_pct;
 123         int32_t max_perf;
 124         int32_t min_perf;
 125         int max_policy_pct;
 126         int max_sysfs_pct;
 127 };
 128
 129 static struct perf_limits limits = {
 130         .no_turbo = 0,
 131         .max_perf_pct = 100,
 132         .max_perf = int_tofp(1),
 133         .min_perf_pct = 0,
 134         .min_perf = 0,
 135         .max_policy_pct = 100,
 136         .max_sysfs_pct = 100,
 137 };
 138
 139 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 140                         int deadband, int integral) {
 141         pid->setpoint = setpoint;
 142         pid->deadband  = deadband;
 143         pid->integral  = int_tofp(integral);
 144         pid->last_err  = setpoint - busy;
 145 }
 146
 147 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 148 {
 149         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 150 }
 151
 152 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 153 {
 154         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 155 }
 156
 157 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 158 {
 159
 160         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 161 }
 162
 163 static signed int pid_calc(struct _pid *pid, int32_t busy)
 164 {
 165         signed int result;
 166         int32_t pterm, dterm, fp_error;
 167         int32_t integral_limit;
 168
 169         fp_error = int_tofp(pid->setpoint) - busy;
 170
 171         if (abs(fp_error) <= int_tofp(pid->deadband))
 172                 return 0;
 173
 174         pterm = mul_fp(pid->p_gain, fp_error);
 175
 176         pid->integral += fp_error;
 177
 178         /* limit the integral term */
 179         integral_limit = int_tofp(30);
 180         if (pid->integral > integral_limit)
 181                 pid->integral = integral_limit;
 182         if (pid->integral < -integral_limit)
 183                 pid->integral = -integral_limit;
 184
 185         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 186         pid->last_err = fp_error;
 187
 188         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 189
 190         return (signed int)fp_toint(result);
 191 }
 192
 193 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 194 {
 195         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 196         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 197         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 198
 199         pid_reset(&cpu->pid,
 200                 pid_params.setpoint,
 201                 100,
 202                 pid_params.deadband,
 203                 0);
 204 }
 205
 206 static inline void intel_pstate_reset_all_pid(void)
 207 {
 208         unsigned int cpu;
 209         for_each_online_cpu(cpu) {
 210                 if (all_cpu_data[cpu])
 211                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 212         }
 213 }
 214
 215 /************************** debugfs begin ************************/
 216 static int pid_param_set(void *data, u64 val)
 217 {
 218         *(u32 *)data = val;
 219         intel_pstate_reset_all_pid();
 220         return 0;
 221 }
 222 static int pid_param_get(void *data, u64 *val)
 223 {
 224         *val = *(u32 *)data;
 225         return 0;
 226 }
 227 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 228                         pid_param_set, "%llu\n");
 229
 230 struct pid_param {
 231         char *name;
 232         void *value;
 233 };
 234
 235 static struct pid_param pid_files[] = {
 236         {"sample_rate_ms", &pid_params.sample_rate_ms},
 237         {"d_gain_pct", &pid_params.d_gain_pct},
 238         {"i_gain_pct", &pid_params.i_gain_pct},
 239         {"deadband", &pid_params.deadband},
 240         {"setpoint", &pid_params.setpoint},
 241         {"p_gain_pct", &pid_params.p_gain_pct},
 242         {NULL, NULL}
 243 };
 244
 245 static struct dentry *debugfs_parent;
 246 static void intel_pstate_debug_expose_params(void)
 247 {
 248         int i = 0;
 249
 250         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 251         if (IS_ERR_OR_NULL(debugfs_parent))
 252                 return;
 253         while (pid_files[i].name) {
 254                 debugfs_create_file(pid_files[i].name, 0660,
 255                                 debugfs_parent, pid_files[i].value,
 256                                 &fops_pid_param);
 257                 i++;
 258         }
 259 }
 260
 261 /************************** debugfs end ************************/
 262
 263 /************************** sysfs begin ************************/
 264 #define show_one(file_name, object)                                     \
 265         static ssize_t show_##file_name                                 \
 266         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 267         {                                                               \
 268                 return sprintf(buf, "%u\n", limits.object);             \
 269         }
 270
 271 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 272                                 const char *buf, size_t count)
 273 {
 274         unsigned int input;
 275         int ret;
 276         ret = sscanf(buf, "%u", &input);
 277         if (ret != 1)
 278                 return -EINVAL;
 279         limits.no_turbo = clamp_t(int, input, 0 , 1);
 280
 281         return count;
 282 }
 283
 284 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 285                                 const char *buf, size_t count)
 286 {
 287         unsigned int input;
 288         int ret;
 289         ret = sscanf(buf, "%u", &input);
 290         if (ret != 1)
 291                 return -EINVAL;
 292
 293         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 294         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 295         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 296         return count;
 297 }
 298
 299 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 300                                 const char *buf, size_t count)
 301 {
 302         unsigned int input;
 303         int ret;
 304         ret = sscanf(buf, "%u", &input);
 305         if (ret != 1)
 306                 return -EINVAL;
 307         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 308         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 309
 310         return count;
 311 }
 312
 313 show_one(no_turbo, no_turbo);
 314 show_one(max_perf_pct, max_perf_pct);
 315 show_one(min_perf_pct, min_perf_pct);
 316
 317 define_one_global_rw(no_turbo);
 318 define_one_global_rw(max_perf_pct);
 319 define_one_global_rw(min_perf_pct);
 320
 321 static struct attribute *intel_pstate_attributes[] = {
 322         &no_turbo.attr,
 323         &max_perf_pct.attr,
 324         &min_perf_pct.attr,
 325         NULL
 326 };
 327
 328 static struct attribute_group intel_pstate_attr_group = {
 329         .attrs = intel_pstate_attributes,
 330 };
 331 static struct kobject *intel_pstate_kobject;
 332
 333 static void intel_pstate_sysfs_expose_params(void)
 334 {
 335         int rc;
 336
 337         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 338                                                 &cpu_subsys.dev_root->kobj);
 339         BUG_ON(!intel_pstate_kobject);
 340         rc = sysfs_create_group(intel_pstate_kobject,
 341                                 &intel_pstate_attr_group);
 342         BUG_ON(rc);
 343 }
 344
 345 /************************** sysfs end ************************/
 346 static int byt_get_min_pstate(void)
 347 {
 348         u64 value;
 349         rdmsrl(BYT_RATIOS, value);
 350         return value & 0xFF;
 351 }
 352
 353 static int byt_get_max_pstate(void)
 354 {
 355         u64 value;
 356         rdmsrl(BYT_RATIOS, value);
 357         return (value >> 16) & 0xFF;
 358 }
 359
 360 static int core_get_min_pstate(void)
 361 {
 362         u64 value;
 363         rdmsrl(MSR_PLATFORM_INFO, value);
 364         return (value >> 40) & 0xFF;
 365 }
 366
 367 static int core_get_max_pstate(void)
 368 {
 369         u64 value;
 370         rdmsrl(MSR_PLATFORM_INFO, value);
 371         return (value >> 8) & 0xFF;
 372 }
 373
 374 static int core_get_turbo_pstate(void)
 375 {
 376         u64 value;
 377         int nont, ret;
 378         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 379         nont = core_get_max_pstate();
 380         ret = ((value) & 255);
 381         if (ret <= nont)
 382                 ret = nont;
 383         return ret;
 384 }
 385
 386 static void core_set_pstate(int pstate)
 387 {
 388         u64 val;
 389
 390         val = pstate << 8;
 391         if (limits.no_turbo)
 392                 val |= (u64)1 << 32;
 393
 394         wrmsrl(MSR_IA32_PERF_CTL, val);
 395 }
 396
 397 static struct cpu_defaults core_params = {
 398         .pid_policy = {
 399                 .sample_rate_ms = 10,
 400                 .deadband = 0,
 401                 .setpoint = 97,
 402                 .p_gain_pct = 20,
 403                 .d_gain_pct = 0,
 404                 .i_gain_pct = 0,
 405         },
 406         .funcs = {
 407                 .get_max = core_get_max_pstate,
 408                 .get_min = core_get_min_pstate,
 409                 .get_turbo = core_get_turbo_pstate,
 410                 .set = core_set_pstate,
 411         },
 412 };
 413
 414 static struct cpu_defaults byt_params = {
 415         .pid_policy = {
 416                 .sample_rate_ms = 10,
 417                 .deadband = 0,
 418                 .setpoint = 97,
 419                 .p_gain_pct = 14,
 420                 .d_gain_pct = 0,
 421                 .i_gain_pct = 4,
 422         },
 423         .funcs = {
 424                 .get_max = byt_get_max_pstate,
 425                 .get_min = byt_get_min_pstate,
 426                 .get_turbo = byt_get_max_pstate,
 427                 .set = core_set_pstate,
 428         },
 429 };
 430
 431
 432 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 433 {
 434         int max_perf = cpu->pstate.turbo_pstate;
 435         int max_perf_adj;
 436         int min_perf;
 437         if (limits.no_turbo)
 438                 max_perf = cpu->pstate.max_pstate;
 439
 440         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 441         *max = clamp_t(int, max_perf_adj,
 442                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 443
 444         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 445         *min = clamp_t(int, min_perf,
 446                         cpu->pstate.min_pstate, max_perf);
 447 }
 448
 449 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 450 {
 451         int max_perf, min_perf;
 452
 453         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 454
 455         pstate = clamp_t(int, pstate, min_perf, max_perf);
 456
 457         if (pstate == cpu->pstate.current_pstate)
 458                 return;
 459
 460         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 461
 462         cpu->pstate.current_pstate = pstate;
 463
 464         pstate_funcs.set(pstate);
 465 }
 466
 467 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 468 {
 469         int target;
 470         target = cpu->pstate.current_pstate + steps;
 471
 472         intel_pstate_set_pstate(cpu, target);
 473 }
 474
 475 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 476 {
 477         int target;
 478         target = cpu->pstate.current_pstate - steps;
 479         intel_pstate_set_pstate(cpu, target);
 480 }
 481
 482 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 483 {
 484         sprintf(cpu->name, "Intel 2nd generation core");
 485
 486         cpu->pstate.min_pstate = pstate_funcs.get_min();
 487         cpu->pstate.max_pstate = pstate_funcs.get_max();
 488         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 489
 490         /*
 491          * goto max pstate so we don't slow up boot if we are built-in if we are
 492          * a module we will take care of it during normal operation
 493          */
 494         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 495 }
 496
 497 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 498                                         struct sample *sample)
 499 {
 500         u64 core_pct;
 501         core_pct = div64_u64(int_tofp(sample->aperf * 100),
 502                              sample->mperf);
 503         sample->freq = fp_toint(cpu->pstate.max_pstate * core_pct * 1000);
 504
 505         sample->core_pct_busy = core_pct;
 506 }
 507
 508 static inline void intel_pstate_sample(struct cpudata *cpu)
 509 {
 510         u64 aperf, mperf;
 511
 512         rdmsrl(MSR_IA32_APERF, aperf);
 513         rdmsrl(MSR_IA32_MPERF, mperf);
 514         cpu->sample_ptr = (cpu->sample_ptr + 1) % SAMPLE_COUNT;
 515         cpu->samples[cpu->sample_ptr].aperf = aperf;
 516         cpu->samples[cpu->sample_ptr].mperf = mperf;
 517         cpu->samples[cpu->sample_ptr].aperf -= cpu->prev_aperf;
 518         cpu->samples[cpu->sample_ptr].mperf -= cpu->prev_mperf;
 519
 520         intel_pstate_calc_busy(cpu, &cpu->samples[cpu->sample_ptr]);
 521
 522         cpu->prev_aperf = aperf;
 523         cpu->prev_mperf = mperf;
 524 }
 525
 526 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 527 {
 528         int sample_time, delay;
 529
 530         sample_time = pid_params.sample_rate_ms;
 531         delay = msecs_to_jiffies(sample_time);
 532         mod_timer_pinned(&cpu->timer, jiffies + delay);
 533 }
 534
 535 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 536 {
 537         int32_t core_busy, max_pstate, current_pstate;
 538
 539         core_busy = cpu->samples[cpu->sample_ptr].core_pct_busy;
 540         max_pstate = int_tofp(cpu->pstate.max_pstate);
 541         current_pstate = int_tofp(cpu->pstate.current_pstate);
 542         return mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 543 }
 544
 545 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 546 {
 547         int32_t busy_scaled;
 548         struct _pid *pid;
 549         signed int ctl = 0;
 550         int steps;
 551
 552         pid = &cpu->pid;
 553         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 554
 555         ctl = pid_calc(pid, busy_scaled);
 556
 557         steps = abs(ctl);
 558         if (ctl < 0)
 559                 intel_pstate_pstate_increase(cpu, steps);
 560         else
 561                 intel_pstate_pstate_decrease(cpu, steps);
 562 }
 563
 564 static void intel_pstate_timer_func(unsigned long __data)
 565 {
 566         struct cpudata *cpu = (struct cpudata *) __data;
 567
 568         intel_pstate_sample(cpu);
 569         intel_pstate_adjust_busy_pstate(cpu);
 570
 571         if (cpu->pstate.current_pstate == cpu->pstate.min_pstate) {
 572                 cpu->min_pstate_count++;
 573                 if (!(cpu->min_pstate_count % 5)) {
 574                         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 575                 }
 576         } else
 577                 cpu->min_pstate_count = 0;
 578
 579         intel_pstate_set_sample_time(cpu);
 580 }
 581
 582 #define ICPU(model, policy) \
 583         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&policy }
 584
 585 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 586         ICPU(0x2a, core_params),
 587         ICPU(0x2d, core_params),
 588         ICPU(0x37, byt_params),
 589         ICPU(0x3a, core_params),
 590         ICPU(0x3c, core_params),
 591         ICPU(0x3e, core_params),
 592         ICPU(0x3f, core_params),
 593         ICPU(0x45, core_params),
 594         ICPU(0x46, core_params),
 595         {}
 596 };
 597 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 598
 599 static int intel_pstate_init_cpu(unsigned int cpunum)
 600 {
 601
 602         const struct x86_cpu_id *id;
 603         struct cpudata *cpu;
 604
 605         id = x86_match_cpu(intel_pstate_cpu_ids);
 606         if (!id)
 607                 return -ENODEV;
 608
 609         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 610         if (!all_cpu_data[cpunum])
 611                 return -ENOMEM;
 612
 613         cpu = all_cpu_data[cpunum];
 614
 615         intel_pstate_get_cpu_pstates(cpu);
 616
 617         cpu->cpu = cpunum;
 618
 619         init_timer_deferrable(&cpu->timer);
 620         cpu->timer.function = intel_pstate_timer_func;
 621         cpu->timer.data =
 622                 (unsigned long)cpu;
 623         cpu->timer.expires = jiffies + HZ/100;
 624         intel_pstate_busy_pid_reset(cpu);
 625         intel_pstate_sample(cpu);
 626         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 627
 628         add_timer_on(&cpu->timer, cpunum);
 629
 630         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 631
 632         return 0;
 633 }
 634
 635 static unsigned int intel_pstate_get(unsigned int cpu_num)
 636 {
 637         struct sample *sample;
 638         struct cpudata *cpu;
 639
 640         cpu = all_cpu_data[cpu_num];
 641         if (!cpu)
 642                 return 0;
 643         sample = &cpu->samples[cpu->sample_ptr];
 644         return sample->freq;
 645 }
 646
 647 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 648 {
 649         struct cpudata *cpu;
 650
 651         cpu = all_cpu_data[policy->cpu];
 652
 653         if (!policy->cpuinfo.max_freq)
 654                 return -ENODEV;
 655
 656         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 657                 limits.min_perf_pct = 100;
 658                 limits.min_perf = int_tofp(1);
 659                 limits.max_perf_pct = 100;
 660                 limits.max_perf = int_tofp(1);
 661                 limits.no_turbo = 0;
 662                 return 0;
 663         }
 664         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 665         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 666         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 667
 668         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 669         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 670         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 671         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 672
 673         return 0;
 674 }
 675
 676 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 677 {
 678         cpufreq_verify_within_cpu_limits(policy);
 679
 680         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 681                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 682                 return -EINVAL;
 683
 684         return 0;
 685 }
 686
 687 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 688 {
 689         int cpu = policy->cpu;
 690
 691         del_timer(&all_cpu_data[cpu]->timer);
 692         kfree(all_cpu_data[cpu]);
 693         all_cpu_data[cpu] = NULL;
 694         return 0;
 695 }
 696
 697 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 698 {
 699         struct cpudata *cpu;
 700         int rc;
 701
 702         rc = intel_pstate_init_cpu(policy->cpu);
 703         if (rc)
 704                 return rc;
 705
 706         cpu = all_cpu_data[policy->cpu];
 707
 708         if (!limits.no_turbo &&
 709                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 710                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 711         else
 712                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 713
 714         policy->min = cpu->pstate.min_pstate * 100000;
 715         policy->max = cpu->pstate.turbo_pstate * 100000;
 716
 717         /* cpuinfo and default policy values */
 718         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 719         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 720         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 721         cpumask_set_cpu(policy->cpu, policy->cpus);
 722
 723         return 0;
 724 }
 725
 726 static struct cpufreq_driver intel_pstate_driver = {
 727         .flags          = CPUFREQ_CONST_LOOPS,
 728         .verify         = intel_pstate_verify_policy,
 729         .setpolicy      = intel_pstate_set_policy,
 730         .get            = intel_pstate_get,
 731         .init           = intel_pstate_cpu_init,
 732         .exit           = intel_pstate_cpu_exit,
 733         .name           = "intel_pstate",
 734 };
 735
 736 static int __initdata no_load;
 737
 738 static int intel_pstate_msrs_not_valid(void)
 739 {
 740         /* Check that all the msr's we are using are valid. */
 741         u64 aperf, mperf, tmp;
 742
 743         rdmsrl(MSR_IA32_APERF, aperf);
 744         rdmsrl(MSR_IA32_MPERF, mperf);
 745
 746         if (!pstate_funcs.get_max() ||
 747                 !pstate_funcs.get_min() ||
 748                 !pstate_funcs.get_turbo())
 749                 return -ENODEV;
 750
 751         rdmsrl(MSR_IA32_APERF, tmp);
 752         if (!(tmp - aperf))
 753                 return -ENODEV;
 754
 755         rdmsrl(MSR_IA32_MPERF, tmp);
 756         if (!(tmp - mperf))
 757                 return -ENODEV;
 758
 759         return 0;
 760 }
 761
 762 void copy_pid_params(struct pstate_adjust_policy *policy)
 763 {
 764         pid_params.sample_rate_ms = policy->sample_rate_ms;
 765         pid_params.p_gain_pct = policy->p_gain_pct;
 766         pid_params.i_gain_pct = policy->i_gain_pct;
 767         pid_params.d_gain_pct = policy->d_gain_pct;
 768         pid_params.deadband = policy->deadband;
 769         pid_params.setpoint = policy->setpoint;
 770 }
 771
 772 void copy_cpu_funcs(struct pstate_funcs *funcs)
 773 {
 774         pstate_funcs.get_max   = funcs->get_max;
 775         pstate_funcs.get_min   = funcs->get_min;
 776         pstate_funcs.get_turbo = funcs->get_turbo;
 777         pstate_funcs.set       = funcs->set;
 778 }
 779
 780 static int __init intel_pstate_init(void)
 781 {
 782         int cpu, rc = 0;
 783         const struct x86_cpu_id *id;
 784         struct cpu_defaults *cpu_info;
 785
 786         if (no_load)
 787                 return -ENODEV;
 788
 789         id = x86_match_cpu(intel_pstate_cpu_ids);
 790         if (!id)
 791                 return -ENODEV;
 792
 793         cpu_info = (struct cpu_defaults *)id->driver_data;
 794
 795         copy_pid_params(&cpu_info->pid_policy);
 796         copy_cpu_funcs(&cpu_info->funcs);
 797
 798         if (intel_pstate_msrs_not_valid())
 799                 return -ENODEV;
 800
 801         pr_info("Intel P-state driver initializing.\n");
 802
 803         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 804         if (!all_cpu_data)
 805                 return -ENOMEM;
 806
 807         rc = cpufreq_register_driver(&intel_pstate_driver);
 808         if (rc)
 809                 goto out;
 810
 811         intel_pstate_debug_expose_params();
 812         intel_pstate_sysfs_expose_params();
 813         return rc;
 814 out:
 815         get_online_cpus();
 816         for_each_online_cpu(cpu) {
 817                 if (all_cpu_data[cpu]) {
 818                         del_timer_sync(&all_cpu_data[cpu]->timer);
 819                         kfree(all_cpu_data[cpu]);
 820                 }
 821         }
 822
 823         put_online_cpus();
 824         vfree(all_cpu_data);
 825         return -ENODEV;
 826 }
 827 device_initcall(intel_pstate_init);
 828
 829 static int __init intel_pstate_setup(char *str)
 830 {
 831         if (!str)
 832                 return -EINVAL;
 833
 834         if (!strcmp(str, "disable"))
 835                 no_load = 1;
 836         return 0;
 837 }
 838 early_param("intel_pstate", intel_pstate_setup);
 839
 840 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
 841 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
 842 MODULE_LICENSE("GPL");