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 static inline int ceiling_fp(int32_t x)
  59 {
  60         int mask, ret;
  61
  62         ret = fp_toint(x);
  63         mask = (1 << FRAC_BITS) - 1;
  64         if (x & mask)
  65                 ret += 1;
  66         return ret;
  67 }
  68
  69 struct sample {
  70         int32_t core_pct_busy;
  71         u64 aperf;
  72         u64 mperf;
  73         int freq;
  74         ktime_t time;
  75 };
  76
  77 struct pstate_data {
  78         int     current_pstate;
  79         int     min_pstate;
  80         int     max_pstate;
  81         int     scaling;
  82         int     turbo_pstate;
  83 };
  84
  85 struct vid_data {
  86         int min;
  87         int max;
  88         int turbo;
  89         int32_t ratio;
  90 };
  91
  92 struct _pid {
  93         int setpoint;
  94         int32_t integral;
  95         int32_t p_gain;
  96         int32_t i_gain;
  97         int32_t d_gain;
  98         int deadband;
  99         int32_t last_err;
 100 };
 101
 102 struct cpudata {
 103         int cpu;
 104
 105         char name[64];
 106
 107         struct timer_list timer;
 108
 109         struct pstate_data pstate;
 110         struct vid_data vid;
 111         struct _pid pid;
 112
 113         ktime_t last_sample_time;
 114         u64     prev_aperf;
 115         u64     prev_mperf;
 116         struct sample sample;
 117 };
 118
 119 static struct cpudata **all_cpu_data;
 120 struct pstate_adjust_policy {
 121         int sample_rate_ms;
 122         int deadband;
 123         int setpoint;
 124         int p_gain_pct;
 125         int d_gain_pct;
 126         int i_gain_pct;
 127 };
 128
 129 struct pstate_funcs {
 130         int (*get_max)(void);
 131         int (*get_min)(void);
 132         int (*get_turbo)(void);
 133         int (*get_scaling)(void);
 134         void (*set)(struct cpudata*, int pstate);
 135         void (*get_vid)(struct cpudata *);
 136 };
 137
 138 struct cpu_defaults {
 139         struct pstate_adjust_policy pid_policy;
 140         struct pstate_funcs funcs;
 141 };
 142
 143 static struct pstate_adjust_policy pid_params;
 144 static struct pstate_funcs pstate_funcs;
 145
 146 struct perf_limits {
 147         int no_turbo;
 148         int turbo_disabled;
 149         int max_perf_pct;
 150         int min_perf_pct;
 151         int32_t max_perf;
 152         int32_t min_perf;
 153         int max_policy_pct;
 154         int max_sysfs_pct;
 155 };
 156
 157 static struct perf_limits limits = {
 158         .no_turbo = 0,
 159         .max_perf_pct = 100,
 160         .max_perf = int_tofp(1),
 161         .min_perf_pct = 0,
 162         .min_perf = 0,
 163         .max_policy_pct = 100,
 164         .max_sysfs_pct = 100,
 165 };
 166
 167 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 168                         int deadband, int integral) {
 169         pid->setpoint = setpoint;
 170         pid->deadband  = deadband;
 171         pid->integral  = int_tofp(integral);
 172         pid->last_err  = setpoint - busy;
 173 }
 174
 175 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 176 {
 177         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 178 }
 179
 180 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 181 {
 182         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 183 }
 184
 185 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 186 {
 187
 188         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 189 }
 190
 191 static signed int pid_calc(struct _pid *pid, int32_t busy)
 192 {
 193         signed int result;
 194         int32_t pterm, dterm, fp_error;
 195         int32_t integral_limit;
 196
 197         fp_error = int_tofp(pid->setpoint) - busy;
 198
 199         if (abs(fp_error) <= int_tofp(pid->deadband))
 200                 return 0;
 201
 202         pterm = mul_fp(pid->p_gain, fp_error);
 203
 204         pid->integral += fp_error;
 205
 206         /* limit the integral term */
 207         integral_limit = int_tofp(30);
 208         if (pid->integral > integral_limit)
 209                 pid->integral = integral_limit;
 210         if (pid->integral < -integral_limit)
 211                 pid->integral = -integral_limit;
 212
 213         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 214         pid->last_err = fp_error;
 215
 216         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 217         result = result + (1 << (FRAC_BITS-1));
 218         return (signed int)fp_toint(result);
 219 }
 220
 221 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 222 {
 223         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 224         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 225         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 226
 227         pid_reset(&cpu->pid,
 228                 pid_params.setpoint,
 229                 100,
 230                 pid_params.deadband,
 231                 0);
 232 }
 233
 234 static inline void intel_pstate_reset_all_pid(void)
 235 {
 236         unsigned int cpu;
 237         for_each_online_cpu(cpu) {
 238                 if (all_cpu_data[cpu])
 239                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 240         }
 241 }
 242
 243 /************************** debugfs begin ************************/
 244 static int pid_param_set(void *data, u64 val)
 245 {
 246         *(u32 *)data = val;
 247         intel_pstate_reset_all_pid();
 248         return 0;
 249 }
 250 static int pid_param_get(void *data, u64 *val)
 251 {
 252         *val = *(u32 *)data;
 253         return 0;
 254 }
 255 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 256                         pid_param_set, "%llu\n");
 257
 258 struct pid_param {
 259         char *name;
 260         void *value;
 261 };
 262
 263 static struct pid_param pid_files[] = {
 264         {"sample_rate_ms", &pid_params.sample_rate_ms},
 265         {"d_gain_pct", &pid_params.d_gain_pct},
 266         {"i_gain_pct", &pid_params.i_gain_pct},
 267         {"deadband", &pid_params.deadband},
 268         {"setpoint", &pid_params.setpoint},
 269         {"p_gain_pct", &pid_params.p_gain_pct},
 270         {NULL, NULL}
 271 };
 272
 273 static struct dentry *debugfs_parent;
 274 static void intel_pstate_debug_expose_params(void)
 275 {
 276         int i = 0;
 277
 278         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 279         if (IS_ERR_OR_NULL(debugfs_parent))
 280                 return;
 281         while (pid_files[i].name) {
 282                 debugfs_create_file(pid_files[i].name, 0660,
 283                                 debugfs_parent, pid_files[i].value,
 284                                 &fops_pid_param);
 285                 i++;
 286         }
 287 }
 288
 289 /************************** debugfs end ************************/
 290
 291 /************************** sysfs begin ************************/
 292 #define show_one(file_name, object)                                     \
 293         static ssize_t show_##file_name                                 \
 294         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 295         {                                                               \
 296                 return sprintf(buf, "%u\n", limits.object);             \
 297         }
 298
 299 static ssize_t store_no_turbo(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.no_turbo = clamp_t(int, input, 0 , 1);
 308         if (limits.turbo_disabled) {
 309                 pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
 310                 limits.no_turbo = limits.turbo_disabled;
 311         }
 312         return count;
 313 }
 314
 315 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 316                                 const char *buf, size_t count)
 317 {
 318         unsigned int input;
 319         int ret;
 320         ret = sscanf(buf, "%u", &input);
 321         if (ret != 1)
 322                 return -EINVAL;
 323
 324         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 325         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 326         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 327         return count;
 328 }
 329
 330 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 331                                 const char *buf, size_t count)
 332 {
 333         unsigned int input;
 334         int ret;
 335         ret = sscanf(buf, "%u", &input);
 336         if (ret != 1)
 337                 return -EINVAL;
 338         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 339         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 340
 341         return count;
 342 }
 343
 344 show_one(no_turbo, no_turbo);
 345 show_one(max_perf_pct, max_perf_pct);
 346 show_one(min_perf_pct, min_perf_pct);
 347
 348 define_one_global_rw(no_turbo);
 349 define_one_global_rw(max_perf_pct);
 350 define_one_global_rw(min_perf_pct);
 351
 352 static struct attribute *intel_pstate_attributes[] = {
 353         &no_turbo.attr,
 354         &max_perf_pct.attr,
 355         &min_perf_pct.attr,
 356         NULL
 357 };
 358
 359 static struct attribute_group intel_pstate_attr_group = {
 360         .attrs = intel_pstate_attributes,
 361 };
 362 static struct kobject *intel_pstate_kobject;
 363
 364 static void intel_pstate_sysfs_expose_params(void)
 365 {
 366         int rc;
 367
 368         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 369                                                 &cpu_subsys.dev_root->kobj);
 370         BUG_ON(!intel_pstate_kobject);
 371         rc = sysfs_create_group(intel_pstate_kobject,
 372                                 &intel_pstate_attr_group);
 373         BUG_ON(rc);
 374 }
 375
 376 /************************** sysfs end ************************/
 377 static int byt_get_min_pstate(void)
 378 {
 379         u64 value;
 380         rdmsrl(BYT_RATIOS, value);
 381         return (value >> 8) & 0x7F;
 382 }
 383
 384 static int byt_get_max_pstate(void)
 385 {
 386         u64 value;
 387         rdmsrl(BYT_RATIOS, value);
 388         return (value >> 16) & 0x7F;
 389 }
 390
 391 static int byt_get_turbo_pstate(void)
 392 {
 393         u64 value;
 394         rdmsrl(BYT_TURBO_RATIOS, value);
 395         return value & 0x7F;
 396 }
 397
 398 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 399 {
 400         u64 val;
 401         int32_t vid_fp;
 402         u32 vid;
 403
 404         val = pstate << 8;
 405         if (limits.no_turbo && !limits.turbo_disabled)
 406                 val |= (u64)1 << 32;
 407
 408         vid_fp = cpudata->vid.min + mul_fp(
 409                 int_tofp(pstate - cpudata->pstate.min_pstate),
 410                 cpudata->vid.ratio);
 411
 412         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 413         vid = ceiling_fp(vid_fp);
 414
 415         if (pstate > cpudata->pstate.max_pstate)
 416                 vid = cpudata->vid.turbo;
 417
 418         val |= vid;
 419
 420         wrmsrl(MSR_IA32_PERF_CTL, val);
 421 }
 422
 423 #define BYT_BCLK_FREQS 5
 424 static int byt_freq_table[BYT_BCLK_FREQS] = { 833, 1000, 1333, 1167, 800};
 425
 426 static int byt_get_scaling(void)
 427 {
 428         u64 value;
 429         int i;
 430
 431         rdmsrl(MSR_FSB_FREQ, value);
 432         i = value & 0x3;
 433
 434         BUG_ON(i > BYT_BCLK_FREQS);
 435
 436         return byt_freq_table[i] * 100;
 437 }
 438
 439 static void byt_get_vid(struct cpudata *cpudata)
 440 {
 441         u64 value;
 442
 443
 444         rdmsrl(BYT_VIDS, value);
 445         cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
 446         cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
 447         cpudata->vid.ratio = div_fp(
 448                 cpudata->vid.max - cpudata->vid.min,
 449                 int_tofp(cpudata->pstate.max_pstate -
 450                         cpudata->pstate.min_pstate));
 451
 452         rdmsrl(BYT_TURBO_VIDS, value);
 453         cpudata->vid.turbo = value & 0x7f;
 454 }
 455
 456
 457 static int core_get_min_pstate(void)
 458 {
 459         u64 value;
 460         rdmsrl(MSR_PLATFORM_INFO, value);
 461         return (value >> 40) & 0xFF;
 462 }
 463
 464 static int core_get_max_pstate(void)
 465 {
 466         u64 value;
 467         rdmsrl(MSR_PLATFORM_INFO, value);
 468         return (value >> 8) & 0xFF;
 469 }
 470
 471 static int core_get_turbo_pstate(void)
 472 {
 473         u64 value;
 474         int nont, ret;
 475         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 476         nont = core_get_max_pstate();
 477         ret = ((value) & 255);
 478         if (ret <= nont)
 479                 ret = nont;
 480         return ret;
 481 }
 482
 483 static inline int core_get_scaling(void)
 484 {
 485         return 100000;
 486 }
 487
 488 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 489 {
 490         u64 val;
 491
 492         val = pstate << 8;
 493         if (limits.no_turbo && !limits.turbo_disabled)
 494                 val |= (u64)1 << 32;
 495
 496         wrmsrl(MSR_IA32_PERF_CTL, val);
 497 }
 498
 499 static struct cpu_defaults core_params = {
 500         .pid_policy = {
 501                 .sample_rate_ms = 10,
 502                 .deadband = 0,
 503                 .setpoint = 97,
 504                 .p_gain_pct = 20,
 505                 .d_gain_pct = 0,
 506                 .i_gain_pct = 0,
 507         },
 508         .funcs = {
 509                 .get_max = core_get_max_pstate,
 510                 .get_min = core_get_min_pstate,
 511                 .get_turbo = core_get_turbo_pstate,
 512                 .get_scaling = core_get_scaling,
 513                 .set = core_set_pstate,
 514         },
 515 };
 516
 517 static struct cpu_defaults byt_params = {
 518         .pid_policy = {
 519                 .sample_rate_ms = 10,
 520                 .deadband = 0,
 521                 .setpoint = 97,
 522                 .p_gain_pct = 14,
 523                 .d_gain_pct = 0,
 524                 .i_gain_pct = 4,
 525         },
 526         .funcs = {
 527                 .get_max = byt_get_max_pstate,
 528                 .get_min = byt_get_min_pstate,
 529                 .get_turbo = byt_get_turbo_pstate,
 530                 .set = byt_set_pstate,
 531                 .get_scaling = byt_get_scaling,
 532                 .get_vid = byt_get_vid,
 533         },
 534 };
 535
 536
 537 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 538 {
 539         int max_perf = cpu->pstate.turbo_pstate;
 540         int max_perf_adj;
 541         int min_perf;
 542         if (limits.no_turbo)
 543                 max_perf = cpu->pstate.max_pstate;
 544
 545         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 546         *max = clamp_t(int, max_perf_adj,
 547                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 548
 549         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 550         *min = clamp_t(int, min_perf,
 551                         cpu->pstate.min_pstate, max_perf);
 552 }
 553
 554 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 555 {
 556         int max_perf, min_perf;
 557
 558         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 559
 560         pstate = clamp_t(int, pstate, min_perf, max_perf);
 561
 562         if (pstate == cpu->pstate.current_pstate)
 563                 return;
 564
 565         trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
 566
 567         cpu->pstate.current_pstate = pstate;
 568
 569         pstate_funcs.set(cpu, pstate);
 570 }
 571
 572 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 573 {
 574         int target;
 575         target = cpu->pstate.current_pstate + steps;
 576
 577         intel_pstate_set_pstate(cpu, target);
 578 }
 579
 580 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 581 {
 582         int target;
 583         target = cpu->pstate.current_pstate - steps;
 584         intel_pstate_set_pstate(cpu, target);
 585 }
 586
 587 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 588 {
 589         sprintf(cpu->name, "Intel 2nd generation core");
 590
 591         cpu->pstate.min_pstate = pstate_funcs.get_min();
 592         cpu->pstate.max_pstate = pstate_funcs.get_max();
 593         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 594         cpu->pstate.scaling = pstate_funcs.get_scaling();
 595
 596         if (pstate_funcs.get_vid)
 597                 pstate_funcs.get_vid(cpu);
 598         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 599 }
 600
 601 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 602                                         struct sample *sample)
 603 {
 604         int64_t core_pct;
 605         int32_t rem;
 606
 607         core_pct = int_tofp(sample->aperf) * int_tofp(100);
 608         core_pct = div_u64_rem(core_pct, int_tofp(sample->mperf), &rem);
 609
 610         if ((rem << 1) >= int_tofp(sample->mperf))
 611                 core_pct += 1;
 612
 613         sample->freq = fp_toint(
 614                 mul_fp(int_tofp(
 615                         cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
 616                         core_pct));
 617
 618         sample->core_pct_busy = (int32_t)core_pct;
 619 }
 620
 621 static inline void intel_pstate_sample(struct cpudata *cpu)
 622 {
 623         u64 aperf, mperf;
 624
 625         rdmsrl(MSR_IA32_APERF, aperf);
 626         rdmsrl(MSR_IA32_MPERF, mperf);
 627
 628         aperf = aperf >> FRAC_BITS;
 629         mperf = mperf >> FRAC_BITS;
 630
 631         cpu->last_sample_time = cpu->sample.time;
 632         cpu->sample.time = ktime_get();
 633         cpu->sample.aperf = aperf;
 634         cpu->sample.mperf = mperf;
 635         cpu->sample.aperf -= cpu->prev_aperf;
 636         cpu->sample.mperf -= cpu->prev_mperf;
 637
 638         intel_pstate_calc_busy(cpu, &cpu->sample);
 639
 640         cpu->prev_aperf = aperf;
 641         cpu->prev_mperf = mperf;
 642 }
 643
 644 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 645 {
 646         int sample_time, delay;
 647
 648         sample_time = pid_params.sample_rate_ms;
 649         delay = msecs_to_jiffies(sample_time);
 650         mod_timer_pinned(&cpu->timer, jiffies + delay);
 651 }
 652
 653 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 654 {
 655         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 656         u32 duration_us;
 657         u32 sample_time;
 658
 659         core_busy = cpu->sample.core_pct_busy;
 660         max_pstate = int_tofp(cpu->pstate.max_pstate);
 661         current_pstate = int_tofp(cpu->pstate.current_pstate);
 662         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 663
 664         sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
 665         duration_us = (u32) ktime_us_delta(cpu->sample.time,
 666                                         cpu->last_sample_time);
 667         if (duration_us > sample_time * 3) {
 668                 sample_ratio = div_fp(int_tofp(sample_time),
 669                                 int_tofp(duration_us));
 670                 core_busy = mul_fp(core_busy, sample_ratio);
 671         }
 672
 673         return core_busy;
 674 }
 675
 676 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 677 {
 678         int32_t busy_scaled;
 679         struct _pid *pid;
 680         signed int ctl = 0;
 681         int steps;
 682
 683         pid = &cpu->pid;
 684         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 685
 686         ctl = pid_calc(pid, busy_scaled);
 687
 688         steps = abs(ctl);
 689
 690         if (ctl < 0)
 691                 intel_pstate_pstate_increase(cpu, steps);
 692         else
 693                 intel_pstate_pstate_decrease(cpu, steps);
 694 }
 695
 696 static void intel_pstate_timer_func(unsigned long __data)
 697 {
 698         struct cpudata *cpu = (struct cpudata *) __data;
 699         struct sample *sample;
 700
 701         intel_pstate_sample(cpu);
 702
 703         sample = &cpu->sample;
 704
 705         intel_pstate_adjust_busy_pstate(cpu);
 706
 707         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 708                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 709                         cpu->pstate.current_pstate,
 710                         sample->mperf,
 711                         sample->aperf,
 712                         sample->freq);
 713
 714         intel_pstate_set_sample_time(cpu);
 715 }
 716
 717 #define ICPU(model, policy) \
 718         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 719                         (unsigned long)&policy }
 720
 721 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 722         ICPU(0x2a, core_params),
 723         ICPU(0x2d, core_params),
 724         ICPU(0x37, byt_params),
 725         ICPU(0x3a, core_params),
 726         ICPU(0x3c, core_params),
 727         ICPU(0x3d, core_params),
 728         ICPU(0x3e, core_params),
 729         ICPU(0x3f, core_params),
 730         ICPU(0x45, core_params),
 731         ICPU(0x46, core_params),
 732         ICPU(0x4c, byt_params),
 733         ICPU(0x4f, core_params),
 734         ICPU(0x56, core_params),
 735         {}
 736 };
 737 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 738
 739 static int intel_pstate_init_cpu(unsigned int cpunum)
 740 {
 741
 742         const struct x86_cpu_id *id;
 743         struct cpudata *cpu;
 744
 745         id = x86_match_cpu(intel_pstate_cpu_ids);
 746         if (!id)
 747                 return -ENODEV;
 748
 749         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 750         if (!all_cpu_data[cpunum])
 751                 return -ENOMEM;
 752
 753         cpu = all_cpu_data[cpunum];
 754
 755         cpu->cpu = cpunum;
 756         intel_pstate_get_cpu_pstates(cpu);
 757
 758         init_timer_deferrable(&cpu->timer);
 759         cpu->timer.function = intel_pstate_timer_func;
 760         cpu->timer.data =
 761                 (unsigned long)cpu;
 762         cpu->timer.expires = jiffies + HZ/100;
 763         intel_pstate_busy_pid_reset(cpu);
 764         intel_pstate_sample(cpu);
 765
 766         add_timer_on(&cpu->timer, cpunum);
 767
 768         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 769
 770         return 0;
 771 }
 772
 773 static unsigned int intel_pstate_get(unsigned int cpu_num)
 774 {
 775         struct sample *sample;
 776         struct cpudata *cpu;
 777
 778         cpu = all_cpu_data[cpu_num];
 779         if (!cpu)
 780                 return 0;
 781         sample = &cpu->sample;
 782         return sample->freq;
 783 }
 784
 785 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 786 {
 787         struct cpudata *cpu;
 788
 789         cpu = all_cpu_data[policy->cpu];
 790
 791         if (!policy->cpuinfo.max_freq)
 792                 return -ENODEV;
 793
 794         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 795                 limits.min_perf_pct = 100;
 796                 limits.min_perf = int_tofp(1);
 797                 limits.max_policy_pct = 100;
 798                 limits.max_perf_pct = 100;
 799                 limits.max_perf = int_tofp(1);
 800                 limits.no_turbo = limits.turbo_disabled;
 801                 return 0;
 802         }
 803         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 804         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 805         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 806
 807         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 808         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 809         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 810         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 811
 812         return 0;
 813 }
 814
 815 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 816 {
 817         cpufreq_verify_within_cpu_limits(policy);
 818
 819         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 820                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 821                 return -EINVAL;
 822
 823         return 0;
 824 }
 825
 826 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 827 {
 828         int cpu = policy->cpu;
 829
 830         del_timer(&all_cpu_data[cpu]->timer);
 831         kfree(all_cpu_data[cpu]);
 832         all_cpu_data[cpu] = NULL;
 833         return 0;
 834 }
 835
 836 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 837 {
 838         struct cpudata *cpu;
 839         int rc;
 840         u64 misc_en;
 841
 842         rc = intel_pstate_init_cpu(policy->cpu);
 843         if (rc)
 844                 return rc;
 845
 846         cpu = all_cpu_data[policy->cpu];
 847
 848         rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
 849         if (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
 850                 cpu->pstate.max_pstate == cpu->pstate.turbo_pstate) {
 851                 limits.turbo_disabled = 1;
 852                 limits.no_turbo = 1;
 853         }
 854         if (limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 855                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 856         else
 857                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 858
 859         policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
 860         policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
 861
 862         /* cpuinfo and default policy values */
 863         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
 864         policy->cpuinfo.max_freq =
 865                 cpu->pstate.turbo_pstate * cpu->pstate.scaling;
 866         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 867         cpumask_set_cpu(policy->cpu, policy->cpus);
 868
 869         return 0;
 870 }
 871
 872 static struct cpufreq_driver intel_pstate_driver = {
 873         .flags          = CPUFREQ_CONST_LOOPS,
 874         .verify         = intel_pstate_verify_policy,
 875         .setpolicy      = intel_pstate_set_policy,
 876         .get            = intel_pstate_get,
 877         .init           = intel_pstate_cpu_init,
 878         .exit           = intel_pstate_cpu_exit,
 879         .name           = "intel_pstate",
 880 };
 881
 882 static int __initdata no_load;
 883
 884 static int intel_pstate_msrs_not_valid(void)
 885 {
 886         /* Check that all the msr's we are using are valid. */
 887         u64 aperf, mperf, tmp;
 888
 889         rdmsrl(MSR_IA32_APERF, aperf);
 890         rdmsrl(MSR_IA32_MPERF, mperf);
 891
 892         if (!pstate_funcs.get_max() ||
 893                 !pstate_funcs.get_min() ||
 894                 !pstate_funcs.get_turbo())
 895                 return -ENODEV;
 896
 897         rdmsrl(MSR_IA32_APERF, tmp);
 898         if (!(tmp - aperf))
 899                 return -ENODEV;
 900
 901         rdmsrl(MSR_IA32_MPERF, tmp);
 902         if (!(tmp - mperf))
 903                 return -ENODEV;
 904
 905         return 0;
 906 }
 907
 908 static void copy_pid_params(struct pstate_adjust_policy *policy)
 909 {
 910         pid_params.sample_rate_ms = policy->sample_rate_ms;
 911         pid_params.p_gain_pct = policy->p_gain_pct;
 912         pid_params.i_gain_pct = policy->i_gain_pct;
 913         pid_params.d_gain_pct = policy->d_gain_pct;
 914         pid_params.deadband = policy->deadband;
 915         pid_params.setpoint = policy->setpoint;
 916 }
 917
 918 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 919 {
 920         pstate_funcs.get_max   = funcs->get_max;
 921         pstate_funcs.get_min   = funcs->get_min;
 922         pstate_funcs.get_turbo = funcs->get_turbo;
 923         pstate_funcs.get_scaling = funcs->get_scaling;
 924         pstate_funcs.set       = funcs->set;
 925         pstate_funcs.get_vid   = funcs->get_vid;
 926 }
 927
 928 #if IS_ENABLED(CONFIG_ACPI)
 929 #include <acpi/processor.h>
 930
 931 static bool intel_pstate_no_acpi_pss(void)
 932 {
 933         int i;
 934
 935         for_each_possible_cpu(i) {
 936                 acpi_status status;
 937                 union acpi_object *pss;
 938                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 939                 struct acpi_processor *pr = per_cpu(processors, i);
 940
 941                 if (!pr)
 942                         continue;
 943
 944                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 945                 if (ACPI_FAILURE(status))
 946                         continue;
 947
 948                 pss = buffer.pointer;
 949                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 950                         kfree(pss);
 951                         return false;
 952                 }
 953
 954                 kfree(pss);
 955         }
 956
 957         return true;
 958 }
 959
 960 struct hw_vendor_info {
 961         u16  valid;
 962         char oem_id[ACPI_OEM_ID_SIZE];
 963         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 964 };
 965
 966 /* Hardware vendor-specific info that has its own power management modes */
 967 static struct hw_vendor_info vendor_info[] = {
 968         {1, "HP    ", "ProLiant"},
 969         {0, "", ""},
 970 };
 971
 972 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 973 {
 974         struct acpi_table_header hdr;
 975         struct hw_vendor_info *v_info;
 976
 977         if (acpi_disabled
 978             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 979                 return false;
 980
 981         for (v_info = vendor_info; v_info->valid; v_info++) {
 982                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 983                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 984                     && intel_pstate_no_acpi_pss())
 985                         return true;
 986         }
 987
 988         return false;
 989 }
 990 #else /* CONFIG_ACPI not enabled */
 991 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 992 #endif /* CONFIG_ACPI */
 993
 994 static int __init intel_pstate_init(void)
 995 {
 996         int cpu, rc = 0;
 997         const struct x86_cpu_id *id;
 998         struct cpu_defaults *cpu_info;
 999
1000         if (no_load)
1001                 return -ENODEV;
1002
1003         id = x86_match_cpu(intel_pstate_cpu_ids);
1004         if (!id)
1005                 return -ENODEV;
1006
1007         /*
1008          * The Intel pstate driver will be ignored if the platform
1009          * firmware has its own power management modes.
1010          */
1011         if (intel_pstate_platform_pwr_mgmt_exists())
1012                 return -ENODEV;
1013
1014         cpu_info = (struct cpu_defaults *)id->driver_data;
1015
1016         copy_pid_params(&cpu_info->pid_policy);
1017         copy_cpu_funcs(&cpu_info->funcs);
1018
1019         if (intel_pstate_msrs_not_valid())
1020                 return -ENODEV;
1021
1022         pr_info("Intel P-state driver initializing.\n");
1023
1024         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
1025         if (!all_cpu_data)
1026                 return -ENOMEM;
1027
1028         rc = cpufreq_register_driver(&intel_pstate_driver);
1029         if (rc)
1030                 goto out;
1031
1032         intel_pstate_debug_expose_params();
1033         intel_pstate_sysfs_expose_params();
1034
1035         return rc;
1036 out:
1037         get_online_cpus();
1038         for_each_online_cpu(cpu) {
1039                 if (all_cpu_data[cpu]) {
1040                         del_timer_sync(&all_cpu_data[cpu]->timer);
1041                         kfree(all_cpu_data[cpu]);
1042                 }
1043         }
1044
1045         put_online_cpus();
1046         vfree(all_cpu_data);
1047         return -ENODEV;
1048 }
1049 device_initcall(intel_pstate_init);
1050
1051 static int __init intel_pstate_setup(char *str)
1052 {
1053         if (!str)
1054                 return -EINVAL;
1055
1056         if (!strcmp(str, "disable"))
1057                 no_load = 1;
1058         return 0;
1059 }
1060 early_param("intel_pstate", intel_pstate_setup);
1061
1062 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1063 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1064 MODULE_LICENSE("GPL");