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
  41 #define FRAC_BITS 8
  42 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  43 #define fp_toint(X) ((X) >> FRAC_BITS)
  44
  45
  46 static inline int32_t mul_fp(int32_t x, int32_t y)
  47 {
  48         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  49 }
  50
  51 static inline int32_t div_fp(int32_t x, int32_t y)
  52 {
  53         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  54 }
  55
  56 struct sample {
  57         int32_t core_pct_busy;
  58         u64 aperf;
  59         u64 mperf;
  60         int freq;
  61         ktime_t time;
  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         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         void (*set)(struct cpudata*, int pstate);
 118         void (*get_vid)(struct cpudata *);
 119 };
 120
 121 struct cpu_defaults {
 122         struct pstate_adjust_policy pid_policy;
 123         struct pstate_funcs funcs;
 124 };
 125
 126 static struct pstate_adjust_policy pid_params;
 127 static struct pstate_funcs pstate_funcs;
 128
 129 struct perf_limits {
 130         int no_turbo;
 131         int max_perf_pct;
 132         int min_perf_pct;
 133         int32_t max_perf;
 134         int32_t min_perf;
 135         int max_policy_pct;
 136         int max_sysfs_pct;
 137 };
 138
 139 static struct perf_limits limits = {
 140         .no_turbo = 0,
 141         .max_perf_pct = 100,
 142         .max_perf = int_tofp(1),
 143         .min_perf_pct = 0,
 144         .min_perf = 0,
 145         .max_policy_pct = 100,
 146         .max_sysfs_pct = 100,
 147 };
 148
 149 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 150                         int deadband, int integral) {
 151         pid->setpoint = setpoint;
 152         pid->deadband  = deadband;
 153         pid->integral  = int_tofp(integral);
 154         pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
 155 }
 156
 157 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 158 {
 159         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 160 }
 161
 162 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 163 {
 164         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 165 }
 166
 167 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 168 {
 169
 170         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 171 }
 172
 173 static signed int pid_calc(struct _pid *pid, int32_t busy)
 174 {
 175         signed int result;
 176         int32_t pterm, dterm, fp_error;
 177         int32_t integral_limit;
 178
 179         fp_error = int_tofp(pid->setpoint) - busy;
 180
 181         if (abs(fp_error) <= int_tofp(pid->deadband))
 182                 return 0;
 183
 184         pterm = mul_fp(pid->p_gain, fp_error);
 185
 186         pid->integral += fp_error;
 187
 188         /* limit the integral term */
 189         integral_limit = int_tofp(30);
 190         if (pid->integral > integral_limit)
 191                 pid->integral = integral_limit;
 192         if (pid->integral < -integral_limit)
 193                 pid->integral = -integral_limit;
 194
 195         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 196         pid->last_err = fp_error;
 197
 198         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 199         result = result + (1 << (FRAC_BITS-1));
 200         return (signed int)fp_toint(result);
 201 }
 202
 203 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 204 {
 205         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 206         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 207         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 208
 209         pid_reset(&cpu->pid,
 210                 pid_params.setpoint,
 211                 100,
 212                 pid_params.deadband,
 213                 0);
 214 }
 215
 216 static inline void intel_pstate_reset_all_pid(void)
 217 {
 218         unsigned int cpu;
 219         for_each_online_cpu(cpu) {
 220                 if (all_cpu_data[cpu])
 221                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 222         }
 223 }
 224
 225 /************************** debugfs begin ************************/
 226 static int pid_param_set(void *data, u64 val)
 227 {
 228         *(u32 *)data = val;
 229         intel_pstate_reset_all_pid();
 230         return 0;
 231 }
 232 static int pid_param_get(void *data, u64 *val)
 233 {
 234         *val = *(u32 *)data;
 235         return 0;
 236 }
 237 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 238                         pid_param_set, "%llu\n");
 239
 240 struct pid_param {
 241         char *name;
 242         void *value;
 243 };
 244
 245 static struct pid_param pid_files[] = {
 246         {"sample_rate_ms", &pid_params.sample_rate_ms},
 247         {"d_gain_pct", &pid_params.d_gain_pct},
 248         {"i_gain_pct", &pid_params.i_gain_pct},
 249         {"deadband", &pid_params.deadband},
 250         {"setpoint", &pid_params.setpoint},
 251         {"p_gain_pct", &pid_params.p_gain_pct},
 252         {NULL, NULL}
 253 };
 254
 255 static struct dentry *debugfs_parent;
 256 static void intel_pstate_debug_expose_params(void)
 257 {
 258         int i = 0;
 259
 260         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 261         if (IS_ERR_OR_NULL(debugfs_parent))
 262                 return;
 263         while (pid_files[i].name) {
 264                 debugfs_create_file(pid_files[i].name, 0660,
 265                                 debugfs_parent, pid_files[i].value,
 266                                 &fops_pid_param);
 267                 i++;
 268         }
 269 }
 270
 271 /************************** debugfs end ************************/
 272
 273 /************************** sysfs begin ************************/
 274 #define show_one(file_name, object)                                     \
 275         static ssize_t show_##file_name                                 \
 276         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 277         {                                                               \
 278                 return sprintf(buf, "%u\n", limits.object);             \
 279         }
 280
 281 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 282                                 const char *buf, size_t count)
 283 {
 284         unsigned int input;
 285         int ret;
 286         ret = sscanf(buf, "%u", &input);
 287         if (ret != 1)
 288                 return -EINVAL;
 289         limits.no_turbo = clamp_t(int, input, 0 , 1);
 290
 291         return count;
 292 }
 293
 294 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 295                                 const char *buf, size_t count)
 296 {
 297         unsigned int input;
 298         int ret;
 299         ret = sscanf(buf, "%u", &input);
 300         if (ret != 1)
 301                 return -EINVAL;
 302
 303         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 304         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 305         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 306         return count;
 307 }
 308
 309 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 310                                 const char *buf, size_t count)
 311 {
 312         unsigned int input;
 313         int ret;
 314         ret = sscanf(buf, "%u", &input);
 315         if (ret != 1)
 316                 return -EINVAL;
 317         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 318         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 319
 320         return count;
 321 }
 322
 323 show_one(no_turbo, no_turbo);
 324 show_one(max_perf_pct, max_perf_pct);
 325 show_one(min_perf_pct, min_perf_pct);
 326
 327 define_one_global_rw(no_turbo);
 328 define_one_global_rw(max_perf_pct);
 329 define_one_global_rw(min_perf_pct);
 330
 331 static struct attribute *intel_pstate_attributes[] = {
 332         &no_turbo.attr,
 333         &max_perf_pct.attr,
 334         &min_perf_pct.attr,
 335         NULL
 336 };
 337
 338 static struct attribute_group intel_pstate_attr_group = {
 339         .attrs = intel_pstate_attributes,
 340 };
 341 static struct kobject *intel_pstate_kobject;
 342
 343 static void intel_pstate_sysfs_expose_params(void)
 344 {
 345         int rc;
 346
 347         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 348                                                 &cpu_subsys.dev_root->kobj);
 349         BUG_ON(!intel_pstate_kobject);
 350         rc = sysfs_create_group(intel_pstate_kobject,
 351                                 &intel_pstate_attr_group);
 352         BUG_ON(rc);
 353 }
 354
 355 /************************** sysfs end ************************/
 356 static int byt_get_min_pstate(void)
 357 {
 358         u64 value;
 359         rdmsrl(BYT_RATIOS, value);
 360         return (value >> 8) & 0x3F;
 361 }
 362
 363 static int byt_get_max_pstate(void)
 364 {
 365         u64 value;
 366         rdmsrl(BYT_RATIOS, value);
 367         return (value >> 16) & 0x3F;
 368 }
 369
 370 static int byt_get_turbo_pstate(void)
 371 {
 372         u64 value;
 373         rdmsrl(BYT_TURBO_RATIOS, value);
 374         return value & 0x3F;
 375 }
 376
 377 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 378 {
 379         u64 val;
 380         int32_t vid_fp;
 381         u32 vid;
 382
 383         val = pstate << 8;
 384         if (limits.no_turbo)
 385                 val |= (u64)1 << 32;
 386
 387         vid_fp = cpudata->vid.min + mul_fp(
 388                 int_tofp(pstate - cpudata->pstate.min_pstate),
 389                 cpudata->vid.ratio);
 390
 391         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 392         vid = fp_toint(vid_fp);
 393
 394         if (pstate > cpudata->pstate.max_pstate)
 395                 vid = cpudata->vid.turbo;
 396
 397         val |= vid;
 398
 399         wrmsrl(MSR_IA32_PERF_CTL, val);
 400 }
 401
 402 static void byt_get_vid(struct cpudata *cpudata)
 403 {
 404         u64 value;
 405
 406
 407         rdmsrl(BYT_VIDS, value);
 408         cpudata->vid.min = int_tofp((value >> 8) & 0x3f);
 409         cpudata->vid.max = int_tofp((value >> 16) & 0x3f);
 410         cpudata->vid.ratio = div_fp(
 411                 cpudata->vid.max - cpudata->vid.min,
 412                 int_tofp(cpudata->pstate.max_pstate -
 413                         cpudata->pstate.min_pstate));
 414
 415         rdmsrl(BYT_TURBO_VIDS, value);
 416         cpudata->vid.turbo = value & 0x7f;
 417 }
 418
 419
 420 static int core_get_min_pstate(void)
 421 {
 422         u64 value;
 423         rdmsrl(MSR_PLATFORM_INFO, value);
 424         return (value >> 40) & 0xFF;
 425 }
 426
 427 static int core_get_max_pstate(void)
 428 {
 429         u64 value;
 430         rdmsrl(MSR_PLATFORM_INFO, value);
 431         return (value >> 8) & 0xFF;
 432 }
 433
 434 static int core_get_turbo_pstate(void)
 435 {
 436         u64 value;
 437         int nont, ret;
 438         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 439         nont = core_get_max_pstate();
 440         ret = ((value) & 255);
 441         if (ret <= nont)
 442                 ret = nont;
 443         return ret;
 444 }
 445
 446 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 447 {
 448         u64 val;
 449
 450         val = pstate << 8;
 451         if (limits.no_turbo)
 452                 val |= (u64)1 << 32;
 453
 454         wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 455 }
 456
 457 static struct cpu_defaults core_params = {
 458         .pid_policy = {
 459                 .sample_rate_ms = 10,
 460                 .deadband = 0,
 461                 .setpoint = 97,
 462                 .p_gain_pct = 20,
 463                 .d_gain_pct = 0,
 464                 .i_gain_pct = 0,
 465         },
 466         .funcs = {
 467                 .get_max = core_get_max_pstate,
 468                 .get_min = core_get_min_pstate,
 469                 .get_turbo = core_get_turbo_pstate,
 470                 .set = core_set_pstate,
 471         },
 472 };
 473
 474 static struct cpu_defaults byt_params = {
 475         .pid_policy = {
 476                 .sample_rate_ms = 10,
 477                 .deadband = 0,
 478                 .setpoint = 97,
 479                 .p_gain_pct = 14,
 480                 .d_gain_pct = 0,
 481                 .i_gain_pct = 4,
 482         },
 483         .funcs = {
 484                 .get_max = byt_get_max_pstate,
 485                 .get_min = byt_get_min_pstate,
 486                 .get_turbo = byt_get_turbo_pstate,
 487                 .set = byt_set_pstate,
 488                 .get_vid = byt_get_vid,
 489         },
 490 };
 491
 492
 493 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 494 {
 495         int max_perf = cpu->pstate.turbo_pstate;
 496         int max_perf_adj;
 497         int min_perf;
 498         if (limits.no_turbo)
 499                 max_perf = cpu->pstate.max_pstate;
 500
 501         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 502         *max = clamp_t(int, max_perf_adj,
 503                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 504
 505         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 506         *min = clamp_t(int, min_perf,
 507                         cpu->pstate.min_pstate, max_perf);
 508 }
 509
 510 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 511 {
 512         int max_perf, min_perf;
 513
 514         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 515
 516         pstate = clamp_t(int, pstate, min_perf, max_perf);
 517
 518         if (pstate == cpu->pstate.current_pstate)
 519                 return;
 520
 521         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 522
 523         cpu->pstate.current_pstate = pstate;
 524
 525         pstate_funcs.set(cpu, pstate);
 526 }
 527
 528 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 529 {
 530         int target;
 531         target = cpu->pstate.current_pstate + steps;
 532
 533         intel_pstate_set_pstate(cpu, target);
 534 }
 535
 536 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 537 {
 538         int target;
 539         target = cpu->pstate.current_pstate - steps;
 540         intel_pstate_set_pstate(cpu, target);
 541 }
 542
 543 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 544 {
 545         cpu->pstate.min_pstate = pstate_funcs.get_min();
 546         cpu->pstate.max_pstate = pstate_funcs.get_max();
 547         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 548
 549         if (pstate_funcs.get_vid)
 550                 pstate_funcs.get_vid(cpu);
 551         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 552 }
 553
 554 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
 555 {
 556         struct sample *sample = &cpu->sample;
 557         int64_t core_pct;
 558         int32_t rem;
 559
 560         core_pct = int_tofp(sample->aperf) * int_tofp(100);
 561         core_pct = div_u64_rem(core_pct, int_tofp(sample->mperf), &rem);
 562
 563         if ((rem << 1) >= int_tofp(sample->mperf))
 564                 core_pct += 1;
 565
 566         sample->freq = fp_toint(
 567                 mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
 568
 569         sample->core_pct_busy = (int32_t)core_pct;
 570 }
 571
 572 static inline void intel_pstate_sample(struct cpudata *cpu)
 573 {
 574         u64 aperf, mperf;
 575
 576         rdmsrl(MSR_IA32_APERF, aperf);
 577         rdmsrl(MSR_IA32_MPERF, mperf);
 578
 579         aperf = aperf >> FRAC_BITS;
 580         mperf = mperf >> FRAC_BITS;
 581
 582         cpu->last_sample_time = cpu->sample.time;
 583         cpu->sample.time = ktime_get();
 584         cpu->sample.aperf = aperf;
 585         cpu->sample.mperf = mperf;
 586         cpu->sample.aperf -= cpu->prev_aperf;
 587         cpu->sample.mperf -= cpu->prev_mperf;
 588
 589         intel_pstate_calc_busy(cpu);
 590
 591         cpu->prev_aperf = aperf;
 592         cpu->prev_mperf = mperf;
 593 }
 594
 595 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 596 {
 597         int sample_time, delay;
 598
 599         sample_time = pid_params.sample_rate_ms;
 600         delay = msecs_to_jiffies(sample_time);
 601         mod_timer_pinned(&cpu->timer, jiffies + delay);
 602 }
 603
 604 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 605 {
 606         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 607         u32 duration_us;
 608         u32 sample_time;
 609
 610         core_busy = cpu->sample.core_pct_busy;
 611         max_pstate = int_tofp(cpu->pstate.max_pstate);
 612         current_pstate = int_tofp(cpu->pstate.current_pstate);
 613         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 614
 615         sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
 616         duration_us = (u32) ktime_us_delta(cpu->sample.time,
 617                                         cpu->last_sample_time);
 618         if (duration_us > sample_time * 3) {
 619                 sample_ratio = div_fp(int_tofp(sample_time),
 620                                 int_tofp(duration_us));
 621                 core_busy = mul_fp(core_busy, sample_ratio);
 622         }
 623
 624         return core_busy;
 625 }
 626
 627 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 628 {
 629         int32_t busy_scaled;
 630         struct _pid *pid;
 631         signed int ctl = 0;
 632         int steps;
 633
 634         pid = &cpu->pid;
 635         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 636
 637         ctl = pid_calc(pid, busy_scaled);
 638
 639         steps = abs(ctl);
 640
 641         if (ctl < 0)
 642                 intel_pstate_pstate_increase(cpu, steps);
 643         else
 644                 intel_pstate_pstate_decrease(cpu, steps);
 645 }
 646
 647 static void intel_pstate_timer_func(unsigned long __data)
 648 {
 649         struct cpudata *cpu = (struct cpudata *) __data;
 650         struct sample *sample;
 651
 652         intel_pstate_sample(cpu);
 653
 654         sample = &cpu->sample;
 655
 656         intel_pstate_adjust_busy_pstate(cpu);
 657
 658         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 659                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 660                         cpu->pstate.current_pstate,
 661                         sample->mperf,
 662                         sample->aperf,
 663                         sample->freq);
 664
 665         intel_pstate_set_sample_time(cpu);
 666 }
 667
 668 #define ICPU(model, policy) \
 669         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 670                         (unsigned long)&policy }
 671
 672 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 673         ICPU(0x2a, core_params),
 674         ICPU(0x2d, core_params),
 675         ICPU(0x37, byt_params),
 676         ICPU(0x3a, core_params),
 677         ICPU(0x3c, core_params),
 678         ICPU(0x3d, core_params),
 679         ICPU(0x3e, core_params),
 680         ICPU(0x3f, core_params),
 681         ICPU(0x45, core_params),
 682         ICPU(0x46, core_params),
 683         ICPU(0x4f, core_params),
 684         ICPU(0x56, core_params),
 685         {}
 686 };
 687 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 688
 689 static int intel_pstate_init_cpu(unsigned int cpunum)
 690 {
 691         struct cpudata *cpu;
 692
 693         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 694         if (!all_cpu_data[cpunum])
 695                 return -ENOMEM;
 696
 697         cpu = all_cpu_data[cpunum];
 698
 699         intel_pstate_get_cpu_pstates(cpu);
 700
 701         cpu->cpu = cpunum;
 702
 703         init_timer_deferrable(&cpu->timer);
 704         cpu->timer.function = intel_pstate_timer_func;
 705         cpu->timer.data =
 706                 (unsigned long)cpu;
 707         cpu->timer.expires = jiffies + HZ/100;
 708         intel_pstate_busy_pid_reset(cpu);
 709         intel_pstate_sample(cpu);
 710
 711         add_timer_on(&cpu->timer, cpunum);
 712
 713         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 714
 715         return 0;
 716 }
 717
 718 static unsigned int intel_pstate_get(unsigned int cpu_num)
 719 {
 720         struct sample *sample;
 721         struct cpudata *cpu;
 722
 723         cpu = all_cpu_data[cpu_num];
 724         if (!cpu)
 725                 return 0;
 726         sample = &cpu->sample;
 727         return sample->freq;
 728 }
 729
 730 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 731 {
 732         struct cpudata *cpu;
 733
 734         cpu = all_cpu_data[policy->cpu];
 735
 736         if (!policy->cpuinfo.max_freq)
 737                 return -ENODEV;
 738
 739         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 740                 limits.min_perf_pct = 100;
 741                 limits.min_perf = int_tofp(1);
 742                 limits.max_perf_pct = 100;
 743                 limits.max_perf = int_tofp(1);
 744                 limits.no_turbo = 0;
 745                 return 0;
 746         }
 747         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 748         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 749         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 750
 751         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 752         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 753         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 754         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 755
 756         return 0;
 757 }
 758
 759 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 760 {
 761         cpufreq_verify_within_cpu_limits(policy);
 762
 763         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 764                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 765                 return -EINVAL;
 766
 767         return 0;
 768 }
 769
 770 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
 771 {
 772         int cpu_num = policy->cpu;
 773         struct cpudata *cpu = all_cpu_data[cpu_num];
 774
 775         pr_info("intel_pstate CPU %d exiting\n", cpu_num);
 776
 777         del_timer_sync(&all_cpu_data[cpu_num]->timer);
 778         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 779         kfree(all_cpu_data[cpu_num]);
 780         all_cpu_data[cpu_num] = NULL;
 781 }
 782
 783 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 784 {
 785         struct cpudata *cpu;
 786         int rc;
 787
 788         rc = intel_pstate_init_cpu(policy->cpu);
 789         if (rc)
 790                 return rc;
 791
 792         cpu = all_cpu_data[policy->cpu];
 793
 794         if (!limits.no_turbo &&
 795                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 796                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 797         else
 798                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 799
 800         policy->min = cpu->pstate.min_pstate * 100000;
 801         policy->max = cpu->pstate.turbo_pstate * 100000;
 802
 803         /* cpuinfo and default policy values */
 804         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 805         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 806         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 807         cpumask_set_cpu(policy->cpu, policy->cpus);
 808
 809         return 0;
 810 }
 811
 812 static struct cpufreq_driver intel_pstate_driver = {
 813         .flags          = CPUFREQ_CONST_LOOPS,
 814         .verify         = intel_pstate_verify_policy,
 815         .setpolicy      = intel_pstate_set_policy,
 816         .get            = intel_pstate_get,
 817         .init           = intel_pstate_cpu_init,
 818         .stop_cpu       = intel_pstate_stop_cpu,
 819         .name           = "intel_pstate",
 820 };
 821
 822 static int __initdata no_load;
 823
 824 static int intel_pstate_msrs_not_valid(void)
 825 {
 826         /* Check that all the msr's we are using are valid. */
 827         u64 aperf, mperf, tmp;
 828
 829         rdmsrl(MSR_IA32_APERF, aperf);
 830         rdmsrl(MSR_IA32_MPERF, mperf);
 831
 832         if (!pstate_funcs.get_max() ||
 833                 !pstate_funcs.get_min() ||
 834                 !pstate_funcs.get_turbo())
 835                 return -ENODEV;
 836
 837         rdmsrl(MSR_IA32_APERF, tmp);
 838         if (!(tmp - aperf))
 839                 return -ENODEV;
 840
 841         rdmsrl(MSR_IA32_MPERF, tmp);
 842         if (!(tmp - mperf))
 843                 return -ENODEV;
 844
 845         return 0;
 846 }
 847
 848 static void copy_pid_params(struct pstate_adjust_policy *policy)
 849 {
 850         pid_params.sample_rate_ms = policy->sample_rate_ms;
 851         pid_params.p_gain_pct = policy->p_gain_pct;
 852         pid_params.i_gain_pct = policy->i_gain_pct;
 853         pid_params.d_gain_pct = policy->d_gain_pct;
 854         pid_params.deadband = policy->deadband;
 855         pid_params.setpoint = policy->setpoint;
 856 }
 857
 858 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 859 {
 860         pstate_funcs.get_max   = funcs->get_max;
 861         pstate_funcs.get_min   = funcs->get_min;
 862         pstate_funcs.get_turbo = funcs->get_turbo;
 863         pstate_funcs.set       = funcs->set;
 864         pstate_funcs.get_vid   = funcs->get_vid;
 865 }
 866
 867 #if IS_ENABLED(CONFIG_ACPI)
 868 #include <acpi/processor.h>
 869
 870 static bool intel_pstate_no_acpi_pss(void)
 871 {
 872         int i;
 873
 874         for_each_possible_cpu(i) {
 875                 acpi_status status;
 876                 union acpi_object *pss;
 877                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 878                 struct acpi_processor *pr = per_cpu(processors, i);
 879
 880                 if (!pr)
 881                         continue;
 882
 883                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 884                 if (ACPI_FAILURE(status))
 885                         continue;
 886
 887                 pss = buffer.pointer;
 888                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 889                         kfree(pss);
 890                         return false;
 891                 }
 892
 893                 kfree(pss);
 894         }
 895
 896         return true;
 897 }
 898
 899 struct hw_vendor_info {
 900         u16  valid;
 901         char oem_id[ACPI_OEM_ID_SIZE];
 902         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 903 };
 904
 905 /* Hardware vendor-specific info that has its own power management modes */
 906 static struct hw_vendor_info vendor_info[] = {
 907         {1, "HP    ", "ProLiant"},
 908         {0, "", ""},
 909 };
 910
 911 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 912 {
 913         struct acpi_table_header hdr;
 914         struct hw_vendor_info *v_info;
 915
 916         if (acpi_disabled
 917             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 918                 return false;
 919
 920         for (v_info = vendor_info; v_info->valid; v_info++) {
 921                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 922                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 923                     && intel_pstate_no_acpi_pss())
 924                         return true;
 925         }
 926
 927         return false;
 928 }
 929 #else /* CONFIG_ACPI not enabled */
 930 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 931 #endif /* CONFIG_ACPI */
 932
 933 static int __init intel_pstate_init(void)
 934 {
 935         int cpu, rc = 0;
 936         const struct x86_cpu_id *id;
 937         struct cpu_defaults *cpu_info;
 938
 939         if (no_load)
 940                 return -ENODEV;
 941
 942         id = x86_match_cpu(intel_pstate_cpu_ids);
 943         if (!id)
 944                 return -ENODEV;
 945
 946         /*
 947          * The Intel pstate driver will be ignored if the platform
 948          * firmware has its own power management modes.
 949          */
 950         if (intel_pstate_platform_pwr_mgmt_exists())
 951                 return -ENODEV;
 952
 953         cpu_info = (struct cpu_defaults *)id->driver_data;
 954
 955         copy_pid_params(&cpu_info->pid_policy);
 956         copy_cpu_funcs(&cpu_info->funcs);
 957
 958         if (intel_pstate_msrs_not_valid())
 959                 return -ENODEV;
 960
 961         pr_info("Intel P-state driver initializing.\n");
 962
 963         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 964         if (!all_cpu_data)
 965                 return -ENOMEM;
 966
 967         rc = cpufreq_register_driver(&intel_pstate_driver);
 968         if (rc)
 969                 goto out;
 970
 971         intel_pstate_debug_expose_params();
 972         intel_pstate_sysfs_expose_params();
 973
 974         return rc;
 975 out:
 976         get_online_cpus();
 977         for_each_online_cpu(cpu) {
 978                 if (all_cpu_data[cpu]) {
 979                         del_timer_sync(&all_cpu_data[cpu]->timer);
 980                         kfree(all_cpu_data[cpu]);
 981                 }
 982         }
 983
 984         put_online_cpus();
 985         vfree(all_cpu_data);
 986         return -ENODEV;
 987 }
 988 device_initcall(intel_pstate_init);
 989
 990 static int __init intel_pstate_setup(char *str)
 991 {
 992         if (!str)
 993                 return -EINVAL;
 994
 995         if (!strcmp(str, "disable"))
 996                 no_load = 1;
 997         return 0;
 998 }
 999 early_param("intel_pstate", intel_pstate_setup);
1000
1001 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1002 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1003 MODULE_LICENSE("GPL");