drivers/cpufreq/intel_pstate.c

   1 /*
   2  * intel_pstate.c: Native P state management for Intel processors
   3  *
   4  * (C) Copyright 2012 Intel Corporation
   5  * Author: Dirk Brandewie <dirk.j.brandewie@intel.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; version 2
  10  * of the License.
  11  */
  12
  13 #include <linux/kernel.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/module.h>
  16 #include <linux/ktime.h>
  17 #include <linux/hrtimer.h>
  18 #include <linux/tick.h>
  19 #include <linux/slab.h>
  20 #include <linux/sched.h>
  21 #include <linux/list.h>
  22 #include <linux/cpu.h>
  23 #include <linux/cpufreq.h>
  24 #include <linux/sysfs.h>
  25 #include <linux/types.h>
  26 #include <linux/fs.h>
  27 #include <linux/debugfs.h>
  28 #include <linux/acpi.h>
  29 #include <trace/events/power.h>
  30
  31 #include <asm/div64.h>
  32 #include <asm/msr.h>
  33 #include <asm/cpu_device_id.h>
  34
  35 #define SAMPLE_COUNT            3
  36
  37 #define BYT_RATIOS              0x66a
  38 #define BYT_VIDS                0x66b
  39 #define BYT_TURBO_RATIOS        0x66c
  40 #define BYT_TURBO_VIDS          0x66d
  41
  42
  43 #define FRAC_BITS 8
  44 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  45 #define fp_toint(X) ((X) >> FRAC_BITS)
  46
  47
  48 static inline int32_t mul_fp(int32_t x, int32_t y)
  49 {
  50         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  51 }
  52
  53 static inline int32_t div_fp(int32_t x, int32_t y)
  54 {
  55         return div_s64((int64_t)x << FRAC_BITS, (int64_t)y);
  56 }
  57
  58 struct sample {
  59         int32_t core_pct_busy;
  60         u64 aperf;
  61         u64 mperf;
  62         int freq;
  63         ktime_t time;
  64 };
  65
  66 struct pstate_data {
  67         int     current_pstate;
  68         int     min_pstate;
  69         int     max_pstate;
  70         int     turbo_pstate;
  71 };
  72
  73 struct vid_data {
  74         int min;
  75         int max;
  76         int turbo;
  77         int32_t ratio;
  78 };
  79
  80 struct _pid {
  81         int setpoint;
  82         int32_t integral;
  83         int32_t p_gain;
  84         int32_t i_gain;
  85         int32_t d_gain;
  86         int deadband;
  87         int32_t last_err;
  88 };
  89
  90 struct cpudata {
  91         int cpu;
  92
  93         char name[64];
  94
  95         struct timer_list timer;
  96
  97         struct pstate_data pstate;
  98         struct vid_data vid;
  99         struct _pid pid;
 100
 101         ktime_t last_sample_time;
 102         u64     prev_aperf;
 103         u64     prev_mperf;
 104         struct sample sample;
 105 };
 106
 107 static struct cpudata **all_cpu_data;
 108 struct pstate_adjust_policy {
 109         int sample_rate_ms;
 110         int deadband;
 111         int setpoint;
 112         int p_gain_pct;
 113         int d_gain_pct;
 114         int i_gain_pct;
 115 };
 116
 117 struct pstate_funcs {
 118         int (*get_max)(void);
 119         int (*get_min)(void);
 120         int (*get_turbo)(void);
 121         void (*set)(struct cpudata*, int pstate);
 122         void (*get_vid)(struct cpudata *);
 123 };
 124
 125 struct cpu_defaults {
 126         struct pstate_adjust_policy pid_policy;
 127         struct pstate_funcs funcs;
 128 };
 129
 130 static struct pstate_adjust_policy pid_params;
 131 static struct pstate_funcs pstate_funcs;
 132
 133 struct perf_limits {
 134         int no_turbo;
 135         int turbo_disabled;
 136         int max_perf_pct;
 137         int min_perf_pct;
 138         int32_t max_perf;
 139         int32_t min_perf;
 140         int max_policy_pct;
 141         int max_sysfs_pct;
 142 };
 143
 144 static struct perf_limits limits = {
 145         .no_turbo = 0,
 146         .max_perf_pct = 100,
 147         .max_perf = int_tofp(1),
 148         .min_perf_pct = 0,
 149         .min_perf = 0,
 150         .max_policy_pct = 100,
 151         .max_sysfs_pct = 100,
 152 };
 153
 154 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 155                         int deadband, int integral) {
 156         pid->setpoint = setpoint;
 157         pid->deadband  = deadband;
 158         pid->integral  = int_tofp(integral);
 159         pid->last_err  = setpoint - busy;
 160 }
 161
 162 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 163 {
 164         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 165 }
 166
 167 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 168 {
 169         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 170 }
 171
 172 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 173 {
 174
 175         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 176 }
 177
 178 static signed int pid_calc(struct _pid *pid, int32_t busy)
 179 {
 180         signed int result;
 181         int32_t pterm, dterm, fp_error;
 182         int32_t integral_limit;
 183
 184         fp_error = int_tofp(pid->setpoint) - busy;
 185
 186         if (abs(fp_error) <= int_tofp(pid->deadband))
 187                 return 0;
 188
 189         pterm = mul_fp(pid->p_gain, fp_error);
 190
 191         pid->integral += fp_error;
 192
 193         /* limit the integral term */
 194         integral_limit = int_tofp(30);
 195         if (pid->integral > integral_limit)
 196                 pid->integral = integral_limit;
 197         if (pid->integral < -integral_limit)
 198                 pid->integral = -integral_limit;
 199
 200         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 201         pid->last_err = fp_error;
 202
 203         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 204         result = result + (1 << (FRAC_BITS-1));
 205         return (signed int)fp_toint(result);
 206 }
 207
 208 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 209 {
 210         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 211         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 212         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 213
 214         pid_reset(&cpu->pid,
 215                 pid_params.setpoint,
 216                 100,
 217                 pid_params.deadband,
 218                 0);
 219 }
 220
 221 static inline void intel_pstate_reset_all_pid(void)
 222 {
 223         unsigned int cpu;
 224         for_each_online_cpu(cpu) {
 225                 if (all_cpu_data[cpu])
 226                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 227         }
 228 }
 229
 230 /************************** debugfs begin ************************/
 231 static int pid_param_set(void *data, u64 val)
 232 {
 233         *(u32 *)data = val;
 234         intel_pstate_reset_all_pid();
 235         return 0;
 236 }
 237 static int pid_param_get(void *data, u64 *val)
 238 {
 239         *val = *(u32 *)data;
 240         return 0;
 241 }
 242 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 243                         pid_param_set, "%llu\n");
 244
 245 struct pid_param {
 246         char *name;
 247         void *value;
 248 };
 249
 250 static struct pid_param pid_files[] = {
 251         {"sample_rate_ms", &pid_params.sample_rate_ms},
 252         {"d_gain_pct", &pid_params.d_gain_pct},
 253         {"i_gain_pct", &pid_params.i_gain_pct},
 254         {"deadband", &pid_params.deadband},
 255         {"setpoint", &pid_params.setpoint},
 256         {"p_gain_pct", &pid_params.p_gain_pct},
 257         {NULL, NULL}
 258 };
 259
 260 static struct dentry *debugfs_parent;
 261 static void intel_pstate_debug_expose_params(void)
 262 {
 263         int i = 0;
 264
 265         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 266         if (IS_ERR_OR_NULL(debugfs_parent))
 267                 return;
 268         while (pid_files[i].name) {
 269                 debugfs_create_file(pid_files[i].name, 0660,
 270                                 debugfs_parent, pid_files[i].value,
 271                                 &fops_pid_param);
 272                 i++;
 273         }
 274 }
 275
 276 /************************** debugfs end ************************/
 277
 278 /************************** sysfs begin ************************/
 279 #define show_one(file_name, object)                                     \
 280         static ssize_t show_##file_name                                 \
 281         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 282         {                                                               \
 283                 return sprintf(buf, "%u\n", limits.object);             \
 284         }
 285
 286 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 287                                 const char *buf, size_t count)
 288 {
 289         unsigned int input;
 290         int ret;
 291         ret = sscanf(buf, "%u", &input);
 292         if (ret != 1)
 293                 return -EINVAL;
 294         limits.no_turbo = clamp_t(int, input, 0 , 1);
 295         if (limits.turbo_disabled) {
 296                 pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
 297                 limits.no_turbo = limits.turbo_disabled;
 298         }
 299         return count;
 300 }
 301
 302 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 303                                 const char *buf, size_t count)
 304 {
 305         unsigned int input;
 306         int ret;
 307         ret = sscanf(buf, "%u", &input);
 308         if (ret != 1)
 309                 return -EINVAL;
 310
 311         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 312         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 313         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 314         return count;
 315 }
 316
 317 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 318                                 const char *buf, size_t count)
 319 {
 320         unsigned int input;
 321         int ret;
 322         ret = sscanf(buf, "%u", &input);
 323         if (ret != 1)
 324                 return -EINVAL;
 325         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 326         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 327
 328         return count;
 329 }
 330
 331 show_one(no_turbo, no_turbo);
 332 show_one(max_perf_pct, max_perf_pct);
 333 show_one(min_perf_pct, min_perf_pct);
 334
 335 define_one_global_rw(no_turbo);
 336 define_one_global_rw(max_perf_pct);
 337 define_one_global_rw(min_perf_pct);
 338
 339 static struct attribute *intel_pstate_attributes[] = {
 340         &no_turbo.attr,
 341         &max_perf_pct.attr,
 342         &min_perf_pct.attr,
 343         NULL
 344 };
 345
 346 static struct attribute_group intel_pstate_attr_group = {
 347         .attrs = intel_pstate_attributes,
 348 };
 349 static struct kobject *intel_pstate_kobject;
 350
 351 static void intel_pstate_sysfs_expose_params(void)
 352 {
 353         int rc;
 354
 355         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 356                                                 &cpu_subsys.dev_root->kobj);
 357         BUG_ON(!intel_pstate_kobject);
 358         rc = sysfs_create_group(intel_pstate_kobject,
 359                                 &intel_pstate_attr_group);
 360         BUG_ON(rc);
 361 }
 362
 363 /************************** sysfs end ************************/
 364 static int byt_get_min_pstate(void)
 365 {
 366         u64 value;
 367         rdmsrl(BYT_RATIOS, value);
 368         return (value >> 8) & 0x7F;
 369 }
 370
 371 static int byt_get_max_pstate(void)
 372 {
 373         u64 value;
 374         rdmsrl(BYT_RATIOS, value);
 375         return (value >> 16) & 0x7F;
 376 }
 377
 378 static int byt_get_turbo_pstate(void)
 379 {
 380         u64 value;
 381         rdmsrl(BYT_TURBO_RATIOS, value);
 382         return value & 0x7F;
 383 }
 384
 385 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 386 {
 387         u64 val;
 388         int32_t vid_fp;
 389         u32 vid;
 390
 391         val = pstate << 8;
 392         if (limits.no_turbo && !limits.turbo_disabled)
 393                 val |= (u64)1 << 32;
 394
 395         vid_fp = cpudata->vid.min + mul_fp(
 396                 int_tofp(pstate - cpudata->pstate.min_pstate),
 397                 cpudata->vid.ratio);
 398
 399         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 400         vid = fp_toint(vid_fp);
 401
 402         if (pstate > cpudata->pstate.max_pstate)
 403                 vid = cpudata->vid.turbo;
 404
 405         val |= vid;
 406
 407         wrmsrl(MSR_IA32_PERF_CTL, val);
 408 }
 409
 410 static void byt_get_vid(struct cpudata *cpudata)
 411 {
 412         u64 value;
 413
 414
 415         rdmsrl(BYT_VIDS, value);
 416         cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
 417         cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
 418         cpudata->vid.ratio = div_fp(
 419                 cpudata->vid.max - cpudata->vid.min,
 420                 int_tofp(cpudata->pstate.max_pstate -
 421                         cpudata->pstate.min_pstate));
 422
 423         rdmsrl(BYT_TURBO_VIDS, value);
 424         cpudata->vid.turbo = value & 0x7f;
 425 }
 426
 427
 428 static int core_get_min_pstate(void)
 429 {
 430         u64 value;
 431         rdmsrl(MSR_PLATFORM_INFO, value);
 432         return (value >> 40) & 0xFF;
 433 }
 434
 435 static int core_get_max_pstate(void)
 436 {
 437         u64 value;
 438         rdmsrl(MSR_PLATFORM_INFO, value);
 439         return (value >> 8) & 0xFF;
 440 }
 441
 442 static int core_get_turbo_pstate(void)
 443 {
 444         u64 value;
 445         int nont, ret;
 446         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 447         nont = core_get_max_pstate();
 448         ret = ((value) & 255);
 449         if (ret <= nont)
 450                 ret = nont;
 451         return ret;
 452 }
 453
 454 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 455 {
 456         u64 val;
 457
 458         val = pstate << 8;
 459         if (limits.no_turbo && !limits.turbo_disabled)
 460                 val |= (u64)1 << 32;
 461
 462         wrmsrl(MSR_IA32_PERF_CTL, val);
 463 }
 464
 465 static struct cpu_defaults core_params = {
 466         .pid_policy = {
 467                 .sample_rate_ms = 10,
 468                 .deadband = 0,
 469                 .setpoint = 97,
 470                 .p_gain_pct = 20,
 471                 .d_gain_pct = 0,
 472                 .i_gain_pct = 0,
 473         },
 474         .funcs = {
 475                 .get_max = core_get_max_pstate,
 476                 .get_min = core_get_min_pstate,
 477                 .get_turbo = core_get_turbo_pstate,
 478                 .set = core_set_pstate,
 479         },
 480 };
 481
 482 static struct cpu_defaults byt_params = {
 483         .pid_policy = {
 484                 .sample_rate_ms = 10,
 485                 .deadband = 0,
 486                 .setpoint = 97,
 487                 .p_gain_pct = 14,
 488                 .d_gain_pct = 0,
 489                 .i_gain_pct = 4,
 490         },
 491         .funcs = {
 492                 .get_max = byt_get_max_pstate,
 493                 .get_min = byt_get_min_pstate,
 494                 .get_turbo = byt_get_turbo_pstate,
 495                 .set = byt_set_pstate,
 496                 .get_vid = byt_get_vid,
 497         },
 498 };
 499
 500
 501 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 502 {
 503         int max_perf = cpu->pstate.turbo_pstate;
 504         int max_perf_adj;
 505         int min_perf;
 506         if (limits.no_turbo)
 507                 max_perf = cpu->pstate.max_pstate;
 508
 509         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 510         *max = clamp_t(int, max_perf_adj,
 511                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 512
 513         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 514         *min = clamp_t(int, min_perf,
 515                         cpu->pstate.min_pstate, max_perf);
 516 }
 517
 518 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 519 {
 520         int max_perf, min_perf;
 521
 522         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 523
 524         pstate = clamp_t(int, pstate, min_perf, max_perf);
 525
 526         if (pstate == cpu->pstate.current_pstate)
 527                 return;
 528
 529         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 530
 531         cpu->pstate.current_pstate = pstate;
 532
 533         pstate_funcs.set(cpu, pstate);
 534 }
 535
 536 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 537 {
 538         int target;
 539         target = cpu->pstate.current_pstate + steps;
 540
 541         intel_pstate_set_pstate(cpu, target);
 542 }
 543
 544 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 545 {
 546         int target;
 547         target = cpu->pstate.current_pstate - steps;
 548         intel_pstate_set_pstate(cpu, target);
 549 }
 550
 551 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 552 {
 553         sprintf(cpu->name, "Intel 2nd generation core");
 554
 555         cpu->pstate.min_pstate = pstate_funcs.get_min();
 556         cpu->pstate.max_pstate = pstate_funcs.get_max();
 557         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 558
 559         if (pstate_funcs.get_vid)
 560                 pstate_funcs.get_vid(cpu);
 561         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 562 }
 563
 564 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 565                                         struct sample *sample)
 566 {
 567         int64_t core_pct;
 568         int32_t rem;
 569
 570         core_pct = int_tofp(sample->aperf) * int_tofp(100);
 571         core_pct = div_u64_rem(core_pct, int_tofp(sample->mperf), &rem);
 572
 573         if ((rem << 1) >= int_tofp(sample->mperf))
 574                 core_pct += 1;
 575
 576         sample->freq = fp_toint(
 577                 mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
 578
 579         sample->core_pct_busy = (int32_t)core_pct;
 580 }
 581
 582 static inline void intel_pstate_sample(struct cpudata *cpu)
 583 {
 584         u64 aperf, mperf;
 585
 586         rdmsrl(MSR_IA32_APERF, aperf);
 587         rdmsrl(MSR_IA32_MPERF, mperf);
 588
 589         aperf = aperf >> FRAC_BITS;
 590         mperf = mperf >> FRAC_BITS;
 591
 592         cpu->last_sample_time = cpu->sample.time;
 593         cpu->sample.time = ktime_get();
 594         cpu->sample.aperf = aperf;
 595         cpu->sample.mperf = mperf;
 596         cpu->sample.aperf -= cpu->prev_aperf;
 597         cpu->sample.mperf -= cpu->prev_mperf;
 598
 599         intel_pstate_calc_busy(cpu, &cpu->sample);
 600
 601         cpu->prev_aperf = aperf;
 602         cpu->prev_mperf = mperf;
 603 }
 604
 605 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 606 {
 607         int sample_time, delay;
 608
 609         sample_time = pid_params.sample_rate_ms;
 610         delay = msecs_to_jiffies(sample_time);
 611         mod_timer_pinned(&cpu->timer, jiffies + delay);
 612 }
 613
 614 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 615 {
 616         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 617         u32 duration_us;
 618         u32 sample_time;
 619
 620         core_busy = cpu->sample.core_pct_busy;
 621         max_pstate = int_tofp(cpu->pstate.max_pstate);
 622         current_pstate = int_tofp(cpu->pstate.current_pstate);
 623         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 624
 625         sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
 626         duration_us = (u32) ktime_us_delta(cpu->sample.time,
 627                                         cpu->last_sample_time);
 628         if (duration_us > sample_time * 3) {
 629                 sample_ratio = div_fp(int_tofp(sample_time),
 630                                 int_tofp(duration_us));
 631                 core_busy = mul_fp(core_busy, sample_ratio);
 632         }
 633
 634         return core_busy;
 635 }
 636
 637 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 638 {
 639         int32_t busy_scaled;
 640         struct _pid *pid;
 641         signed int ctl = 0;
 642         int steps;
 643
 644         pid = &cpu->pid;
 645         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 646
 647         ctl = pid_calc(pid, busy_scaled);
 648
 649         steps = abs(ctl);
 650
 651         if (ctl < 0)
 652                 intel_pstate_pstate_increase(cpu, steps);
 653         else
 654                 intel_pstate_pstate_decrease(cpu, steps);
 655 }
 656
 657 static void intel_pstate_timer_func(unsigned long __data)
 658 {
 659         struct cpudata *cpu = (struct cpudata *) __data;
 660         struct sample *sample;
 661
 662         intel_pstate_sample(cpu);
 663
 664         sample = &cpu->sample;
 665
 666         intel_pstate_adjust_busy_pstate(cpu);
 667
 668         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 669                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 670                         cpu->pstate.current_pstate,
 671                         sample->mperf,
 672                         sample->aperf,
 673                         sample->freq);
 674
 675         intel_pstate_set_sample_time(cpu);
 676 }
 677
 678 #define ICPU(model, policy) \
 679         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 680                         (unsigned long)&policy }
 681
 682 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 683         ICPU(0x2a, core_params),
 684         ICPU(0x2d, core_params),
 685         ICPU(0x37, byt_params),
 686         ICPU(0x3a, core_params),
 687         ICPU(0x3c, core_params),
 688         ICPU(0x3d, core_params),
 689         ICPU(0x3e, core_params),
 690         ICPU(0x3f, core_params),
 691         ICPU(0x45, core_params),
 692         ICPU(0x46, core_params),
 693         ICPU(0x4f, core_params),
 694         ICPU(0x56, core_params),
 695         {}
 696 };
 697 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 698
 699 static int intel_pstate_init_cpu(unsigned int cpunum)
 700 {
 701
 702         const struct x86_cpu_id *id;
 703         struct cpudata *cpu;
 704
 705         id = x86_match_cpu(intel_pstate_cpu_ids);
 706         if (!id)
 707                 return -ENODEV;
 708
 709         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 710         if (!all_cpu_data[cpunum])
 711                 return -ENOMEM;
 712
 713         cpu = all_cpu_data[cpunum];
 714
 715         cpu->cpu = cpunum;
 716         intel_pstate_get_cpu_pstates(cpu);
 717
 718         init_timer_deferrable(&cpu->timer);
 719         cpu->timer.function = intel_pstate_timer_func;
 720         cpu->timer.data =
 721                 (unsigned long)cpu;
 722         cpu->timer.expires = jiffies + HZ/100;
 723         intel_pstate_busy_pid_reset(cpu);
 724         intel_pstate_sample(cpu);
 725
 726         add_timer_on(&cpu->timer, cpunum);
 727
 728         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 729
 730         return 0;
 731 }
 732
 733 static unsigned int intel_pstate_get(unsigned int cpu_num)
 734 {
 735         struct sample *sample;
 736         struct cpudata *cpu;
 737
 738         cpu = all_cpu_data[cpu_num];
 739         if (!cpu)
 740                 return 0;
 741         sample = &cpu->sample;
 742         return sample->freq;
 743 }
 744
 745 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 746 {
 747         struct cpudata *cpu;
 748
 749         cpu = all_cpu_data[policy->cpu];
 750
 751         if (!policy->cpuinfo.max_freq)
 752                 return -ENODEV;
 753
 754         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 755                 limits.min_perf_pct = 100;
 756                 limits.min_perf = int_tofp(1);
 757                 limits.max_policy_pct = 100;
 758                 limits.max_perf_pct = 100;
 759                 limits.max_perf = int_tofp(1);
 760                 limits.no_turbo = limits.turbo_disabled;
 761                 return 0;
 762         }
 763         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 764         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 765         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 766
 767         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 768         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 769         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 770         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 771
 772         return 0;
 773 }
 774
 775 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 776 {
 777         cpufreq_verify_within_cpu_limits(policy);
 778
 779         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 780                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 781                 return -EINVAL;
 782
 783         return 0;
 784 }
 785
 786 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 787 {
 788         int cpu = policy->cpu;
 789
 790         del_timer(&all_cpu_data[cpu]->timer);
 791         kfree(all_cpu_data[cpu]);
 792         all_cpu_data[cpu] = NULL;
 793         return 0;
 794 }
 795
 796 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 797 {
 798         struct cpudata *cpu;
 799         int rc;
 800         u64 misc_en;
 801
 802         rc = intel_pstate_init_cpu(policy->cpu);
 803         if (rc)
 804                 return rc;
 805
 806         cpu = all_cpu_data[policy->cpu];
 807
 808         rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
 809         if (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
 810                 cpu->pstate.max_pstate == cpu->pstate.turbo_pstate) {
 811                 limits.turbo_disabled = 1;
 812                 limits.no_turbo = 1;
 813         }
 814         if (limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 815                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 816         else
 817                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 818
 819         policy->min = cpu->pstate.min_pstate * 100000;
 820         policy->max = cpu->pstate.turbo_pstate * 100000;
 821
 822         /* cpuinfo and default policy values */
 823         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 824         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 825         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 826         cpumask_set_cpu(policy->cpu, policy->cpus);
 827
 828         return 0;
 829 }
 830
 831 static struct cpufreq_driver intel_pstate_driver = {
 832         .flags          = CPUFREQ_CONST_LOOPS,
 833         .verify         = intel_pstate_verify_policy,
 834         .setpolicy      = intel_pstate_set_policy,
 835         .get            = intel_pstate_get,
 836         .init           = intel_pstate_cpu_init,
 837         .exit           = intel_pstate_cpu_exit,
 838         .name           = "intel_pstate",
 839 };
 840
 841 static int __initdata no_load;
 842
 843 static int intel_pstate_msrs_not_valid(void)
 844 {
 845         /* Check that all the msr's we are using are valid. */
 846         u64 aperf, mperf, tmp;
 847
 848         rdmsrl(MSR_IA32_APERF, aperf);
 849         rdmsrl(MSR_IA32_MPERF, mperf);
 850
 851         if (!pstate_funcs.get_max() ||
 852                 !pstate_funcs.get_min() ||
 853                 !pstate_funcs.get_turbo())
 854                 return -ENODEV;
 855
 856         rdmsrl(MSR_IA32_APERF, tmp);
 857         if (!(tmp - aperf))
 858                 return -ENODEV;
 859
 860         rdmsrl(MSR_IA32_MPERF, tmp);
 861         if (!(tmp - mperf))
 862                 return -ENODEV;
 863
 864         return 0;
 865 }
 866
 867 static void copy_pid_params(struct pstate_adjust_policy *policy)
 868 {
 869         pid_params.sample_rate_ms = policy->sample_rate_ms;
 870         pid_params.p_gain_pct = policy->p_gain_pct;
 871         pid_params.i_gain_pct = policy->i_gain_pct;
 872         pid_params.d_gain_pct = policy->d_gain_pct;
 873         pid_params.deadband = policy->deadband;
 874         pid_params.setpoint = policy->setpoint;
 875 }
 876
 877 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 878 {
 879         pstate_funcs.get_max   = funcs->get_max;
 880         pstate_funcs.get_min   = funcs->get_min;
 881         pstate_funcs.get_turbo = funcs->get_turbo;
 882         pstate_funcs.set       = funcs->set;
 883         pstate_funcs.get_vid   = funcs->get_vid;
 884 }
 885
 886 #if IS_ENABLED(CONFIG_ACPI)
 887 #include <acpi/processor.h>
 888
 889 static bool intel_pstate_no_acpi_pss(void)
 890 {
 891         int i;
 892
 893         for_each_possible_cpu(i) {
 894                 acpi_status status;
 895                 union acpi_object *pss;
 896                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 897                 struct acpi_processor *pr = per_cpu(processors, i);
 898
 899                 if (!pr)
 900                         continue;
 901
 902                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 903                 if (ACPI_FAILURE(status))
 904                         continue;
 905
 906                 pss = buffer.pointer;
 907                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 908                         kfree(pss);
 909                         return false;
 910                 }
 911
 912                 kfree(pss);
 913         }
 914
 915         return true;
 916 }
 917
 918 struct hw_vendor_info {
 919         u16  valid;
 920         char oem_id[ACPI_OEM_ID_SIZE];
 921         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 922 };
 923
 924 /* Hardware vendor-specific info that has its own power management modes */
 925 static struct hw_vendor_info vendor_info[] = {
 926         {1, "HP    ", "ProLiant"},
 927         {0, "", ""},
 928 };
 929
 930 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 931 {
 932         struct acpi_table_header hdr;
 933         struct hw_vendor_info *v_info;
 934
 935         if (acpi_disabled
 936             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 937                 return false;
 938
 939         for (v_info = vendor_info; v_info->valid; v_info++) {
 940                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 941                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 942                     && intel_pstate_no_acpi_pss())
 943                         return true;
 944         }
 945
 946         return false;
 947 }
 948 #else /* CONFIG_ACPI not enabled */
 949 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 950 #endif /* CONFIG_ACPI */
 951
 952 static int __init intel_pstate_init(void)
 953 {
 954         int cpu, rc = 0;
 955         const struct x86_cpu_id *id;
 956         struct cpu_defaults *cpu_info;
 957
 958         if (no_load)
 959                 return -ENODEV;
 960
 961         id = x86_match_cpu(intel_pstate_cpu_ids);
 962         if (!id)
 963                 return -ENODEV;
 964
 965         /*
 966          * The Intel pstate driver will be ignored if the platform
 967          * firmware has its own power management modes.
 968          */
 969         if (intel_pstate_platform_pwr_mgmt_exists())
 970                 return -ENODEV;
 971
 972         cpu_info = (struct cpu_defaults *)id->driver_data;
 973
 974         copy_pid_params(&cpu_info->pid_policy);
 975         copy_cpu_funcs(&cpu_info->funcs);
 976
 977         if (intel_pstate_msrs_not_valid())
 978                 return -ENODEV;
 979
 980         pr_info("Intel P-state driver initializing.\n");
 981
 982         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 983         if (!all_cpu_data)
 984                 return -ENOMEM;
 985
 986         rc = cpufreq_register_driver(&intel_pstate_driver);
 987         if (rc)
 988                 goto out;
 989
 990         intel_pstate_debug_expose_params();
 991         intel_pstate_sysfs_expose_params();
 992
 993         return rc;
 994 out:
 995         get_online_cpus();
 996         for_each_online_cpu(cpu) {
 997                 if (all_cpu_data[cpu]) {
 998                         del_timer_sync(&all_cpu_data[cpu]->timer);
 999                         kfree(all_cpu_data[cpu]);
1000                 }
1001         }
1002
1003         put_online_cpus();
1004         vfree(all_cpu_data);
1005         return -ENODEV;
1006 }
1007 device_initcall(intel_pstate_init);
1008
1009 static int __init intel_pstate_setup(char *str)
1010 {
1011         if (!str)
1012                 return -EINVAL;
1013
1014         if (!strcmp(str, "disable"))
1015                 no_load = 1;
1016         return 0;
1017 }
1018 early_param("intel_pstate", intel_pstate_setup);
1019
1020 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1021 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1022 MODULE_LICENSE("GPL");