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 max_perf_pct;
 136         int min_perf_pct;
 137         int32_t max_perf;
 138         int32_t min_perf;
 139         int max_policy_pct;
 140         int max_sysfs_pct;
 141 };
 142
 143 static struct perf_limits limits = {
 144         .no_turbo = 0,
 145         .max_perf_pct = 100,
 146         .max_perf = int_tofp(1),
 147         .min_perf_pct = 0,
 148         .min_perf = 0,
 149         .max_policy_pct = 100,
 150         .max_sysfs_pct = 100,
 151 };
 152
 153 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 154                         int deadband, int integral) {
 155         pid->setpoint = setpoint;
 156         pid->deadband  = deadband;
 157         pid->integral  = int_tofp(integral);
 158         pid->last_err  = setpoint - busy;
 159 }
 160
 161 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 162 {
 163         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 164 }
 165
 166 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 167 {
 168         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 169 }
 170
 171 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 172 {
 173
 174         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 175 }
 176
 177 static signed int pid_calc(struct _pid *pid, int32_t busy)
 178 {
 179         signed int result;
 180         int32_t pterm, dterm, fp_error;
 181         int32_t integral_limit;
 182
 183         fp_error = int_tofp(pid->setpoint) - busy;
 184
 185         if (abs(fp_error) <= int_tofp(pid->deadband))
 186                 return 0;
 187
 188         pterm = mul_fp(pid->p_gain, fp_error);
 189
 190         pid->integral += fp_error;
 191
 192         /* limit the integral term */
 193         integral_limit = int_tofp(30);
 194         if (pid->integral > integral_limit)
 195                 pid->integral = integral_limit;
 196         if (pid->integral < -integral_limit)
 197                 pid->integral = -integral_limit;
 198
 199         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 200         pid->last_err = fp_error;
 201
 202         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 203         if (result >= 0)
 204                 result = result + (1 << (FRAC_BITS-1));
 205         else
 206                 result = result - (1 << (FRAC_BITS-1));
 207         return (signed int)fp_toint(result);
 208 }
 209
 210 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 211 {
 212         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 213         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 214         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 215
 216         pid_reset(&cpu->pid,
 217                 pid_params.setpoint,
 218                 100,
 219                 pid_params.deadband,
 220                 0);
 221 }
 222
 223 static inline void intel_pstate_reset_all_pid(void)
 224 {
 225         unsigned int cpu;
 226         for_each_online_cpu(cpu) {
 227                 if (all_cpu_data[cpu])
 228                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 229         }
 230 }
 231
 232 /************************** debugfs begin ************************/
 233 static int pid_param_set(void *data, u64 val)
 234 {
 235         *(u32 *)data = val;
 236         intel_pstate_reset_all_pid();
 237         return 0;
 238 }
 239 static int pid_param_get(void *data, u64 *val)
 240 {
 241         *val = *(u32 *)data;
 242         return 0;
 243 }
 244 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 245                         pid_param_set, "%llu\n");
 246
 247 struct pid_param {
 248         char *name;
 249         void *value;
 250 };
 251
 252 static struct pid_param pid_files[] = {
 253         {"sample_rate_ms", &pid_params.sample_rate_ms},
 254         {"d_gain_pct", &pid_params.d_gain_pct},
 255         {"i_gain_pct", &pid_params.i_gain_pct},
 256         {"deadband", &pid_params.deadband},
 257         {"setpoint", &pid_params.setpoint},
 258         {"p_gain_pct", &pid_params.p_gain_pct},
 259         {NULL, NULL}
 260 };
 261
 262 static struct dentry *debugfs_parent;
 263 static void intel_pstate_debug_expose_params(void)
 264 {
 265         int i = 0;
 266
 267         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 268         if (IS_ERR_OR_NULL(debugfs_parent))
 269                 return;
 270         while (pid_files[i].name) {
 271                 debugfs_create_file(pid_files[i].name, 0660,
 272                                 debugfs_parent, pid_files[i].value,
 273                                 &fops_pid_param);
 274                 i++;
 275         }
 276 }
 277
 278 /************************** debugfs end ************************/
 279
 280 /************************** sysfs begin ************************/
 281 #define show_one(file_name, object)                                     \
 282         static ssize_t show_##file_name                                 \
 283         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 284         {                                                               \
 285                 return sprintf(buf, "%u\n", limits.object);             \
 286         }
 287
 288 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 289                                 const char *buf, size_t count)
 290 {
 291         unsigned int input;
 292         int ret;
 293         ret = sscanf(buf, "%u", &input);
 294         if (ret != 1)
 295                 return -EINVAL;
 296         limits.no_turbo = clamp_t(int, input, 0 , 1);
 297
 298         return count;
 299 }
 300
 301 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 302                                 const char *buf, size_t count)
 303 {
 304         unsigned int input;
 305         int ret;
 306         ret = sscanf(buf, "%u", &input);
 307         if (ret != 1)
 308                 return -EINVAL;
 309
 310         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 311         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 312         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 313         return count;
 314 }
 315
 316 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 317                                 const char *buf, size_t count)
 318 {
 319         unsigned int input;
 320         int ret;
 321         ret = sscanf(buf, "%u", &input);
 322         if (ret != 1)
 323                 return -EINVAL;
 324         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 325         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 326
 327         return count;
 328 }
 329
 330 show_one(no_turbo, no_turbo);
 331 show_one(max_perf_pct, max_perf_pct);
 332 show_one(min_perf_pct, min_perf_pct);
 333
 334 define_one_global_rw(no_turbo);
 335 define_one_global_rw(max_perf_pct);
 336 define_one_global_rw(min_perf_pct);
 337
 338 static struct attribute *intel_pstate_attributes[] = {
 339         &no_turbo.attr,
 340         &max_perf_pct.attr,
 341         &min_perf_pct.attr,
 342         NULL
 343 };
 344
 345 static struct attribute_group intel_pstate_attr_group = {
 346         .attrs = intel_pstate_attributes,
 347 };
 348 static struct kobject *intel_pstate_kobject;
 349
 350 static void intel_pstate_sysfs_expose_params(void)
 351 {
 352         int rc;
 353
 354         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 355                                                 &cpu_subsys.dev_root->kobj);
 356         BUG_ON(!intel_pstate_kobject);
 357         rc = sysfs_create_group(intel_pstate_kobject,
 358                                 &intel_pstate_attr_group);
 359         BUG_ON(rc);
 360 }
 361
 362 /************************** sysfs end ************************/
 363 static int byt_get_min_pstate(void)
 364 {
 365         u64 value;
 366         rdmsrl(BYT_RATIOS, value);
 367         return (value >> 8) & 0x3F;
 368 }
 369
 370 static int byt_get_max_pstate(void)
 371 {
 372         u64 value;
 373         rdmsrl(BYT_RATIOS, value);
 374         return (value >> 16) & 0x3F;
 375 }
 376
 377 static int byt_get_turbo_pstate(void)
 378 {
 379         u64 value;
 380         rdmsrl(BYT_TURBO_RATIOS, value);
 381         return value & 0x3F;
 382 }
 383
 384 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 385 {
 386         u64 val;
 387         int32_t vid_fp;
 388         u32 vid;
 389
 390         val = pstate << 8;
 391         if (limits.no_turbo)
 392                 val |= (u64)1 << 32;
 393
 394         vid_fp = cpudata->vid.min + mul_fp(
 395                 int_tofp(pstate - cpudata->pstate.min_pstate),
 396                 cpudata->vid.ratio);
 397
 398         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 399         vid = fp_toint(vid_fp);
 400
 401         if (pstate > cpudata->pstate.max_pstate)
 402                 vid = cpudata->vid.turbo;
 403
 404         val |= vid;
 405
 406         wrmsrl(MSR_IA32_PERF_CTL, val);
 407 }
 408
 409 static void byt_get_vid(struct cpudata *cpudata)
 410 {
 411         u64 value;
 412
 413
 414         rdmsrl(BYT_VIDS, value);
 415         cpudata->vid.min = int_tofp((value >> 8) & 0x3f);
 416         cpudata->vid.max = int_tofp((value >> 16) & 0x3f);
 417         cpudata->vid.ratio = div_fp(
 418                 cpudata->vid.max - cpudata->vid.min,
 419                 int_tofp(cpudata->pstate.max_pstate -
 420                         cpudata->pstate.min_pstate));
 421
 422         rdmsrl(BYT_TURBO_VIDS, value);
 423         cpudata->vid.turbo = value & 0x7f;
 424 }
 425
 426
 427 static int core_get_min_pstate(void)
 428 {
 429         u64 value;
 430         rdmsrl(MSR_PLATFORM_INFO, value);
 431         return (value >> 40) & 0xFF;
 432 }
 433
 434 static int core_get_max_pstate(void)
 435 {
 436         u64 value;
 437         rdmsrl(MSR_PLATFORM_INFO, value);
 438         return (value >> 8) & 0xFF;
 439 }
 440
 441 static int core_get_turbo_pstate(void)
 442 {
 443         u64 value;
 444         int nont, ret;
 445         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 446         nont = core_get_max_pstate();
 447         ret = ((value) & 255);
 448         if (ret <= nont)
 449                 ret = nont;
 450         return ret;
 451 }
 452
 453 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 454 {
 455         u64 val;
 456
 457         val = pstate << 8;
 458         if (limits.no_turbo)
 459                 val |= (u64)1 << 32;
 460
 461         wrmsrl(MSR_IA32_PERF_CTL, val);
 462 }
 463
 464 static struct cpu_defaults core_params = {
 465         .pid_policy = {
 466                 .sample_rate_ms = 10,
 467                 .deadband = 0,
 468                 .setpoint = 97,
 469                 .p_gain_pct = 20,
 470                 .d_gain_pct = 0,
 471                 .i_gain_pct = 0,
 472         },
 473         .funcs = {
 474                 .get_max = core_get_max_pstate,
 475                 .get_min = core_get_min_pstate,
 476                 .get_turbo = core_get_turbo_pstate,
 477                 .set = core_set_pstate,
 478         },
 479 };
 480
 481 static struct cpu_defaults byt_params = {
 482         .pid_policy = {
 483                 .sample_rate_ms = 10,
 484                 .deadband = 0,
 485                 .setpoint = 97,
 486                 .p_gain_pct = 14,
 487                 .d_gain_pct = 0,
 488                 .i_gain_pct = 4,
 489         },
 490         .funcs = {
 491                 .get_max = byt_get_max_pstate,
 492                 .get_min = byt_get_min_pstate,
 493                 .get_turbo = byt_get_turbo_pstate,
 494                 .set = byt_set_pstate,
 495                 .get_vid = byt_get_vid,
 496         },
 497 };
 498
 499
 500 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 501 {
 502         int max_perf = cpu->pstate.turbo_pstate;
 503         int max_perf_adj;
 504         int min_perf;
 505         if (limits.no_turbo)
 506                 max_perf = cpu->pstate.max_pstate;
 507
 508         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 509         *max = clamp_t(int, max_perf_adj,
 510                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 511
 512         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 513         *min = clamp_t(int, min_perf,
 514                         cpu->pstate.min_pstate, max_perf);
 515 }
 516
 517 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 518 {
 519         int max_perf, min_perf;
 520
 521         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 522
 523         pstate = clamp_t(int, pstate, min_perf, max_perf);
 524
 525         if (pstate == cpu->pstate.current_pstate)
 526                 return;
 527
 528         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 529
 530         cpu->pstate.current_pstate = pstate;
 531
 532         pstate_funcs.set(cpu, pstate);
 533 }
 534
 535 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 536 {
 537         int target;
 538         target = cpu->pstate.current_pstate + steps;
 539
 540         intel_pstate_set_pstate(cpu, target);
 541 }
 542
 543 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 544 {
 545         int target;
 546         target = cpu->pstate.current_pstate - steps;
 547         intel_pstate_set_pstate(cpu, target);
 548 }
 549
 550 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 551 {
 552         sprintf(cpu->name, "Intel 2nd generation core");
 553
 554         cpu->pstate.min_pstate = pstate_funcs.get_min();
 555         cpu->pstate.max_pstate = pstate_funcs.get_max();
 556         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 557
 558         if (pstate_funcs.get_vid)
 559                 pstate_funcs.get_vid(cpu);
 560         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 561 }
 562
 563 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 564                                         struct sample *sample)
 565 {
 566         int32_t core_pct;
 567
 568         core_pct = div_fp(int_tofp((sample->aperf)),
 569                         int_tofp((sample->mperf)));
 570         core_pct = mul_fp(core_pct, int_tofp(100));
 571
 572         sample->freq = fp_toint(
 573                 mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
 574
 575         sample->core_pct_busy = core_pct;
 576 }
 577
 578 static inline void intel_pstate_sample(struct cpudata *cpu)
 579 {
 580         u64 aperf, mperf;
 581
 582         rdmsrl(MSR_IA32_APERF, aperf);
 583         rdmsrl(MSR_IA32_MPERF, mperf);
 584
 585         aperf = aperf >> FRAC_BITS;
 586         mperf = mperf >> FRAC_BITS;
 587
 588         cpu->last_sample_time = cpu->sample.time;
 589         cpu->sample.time = ktime_get();
 590         cpu->sample.aperf = aperf;
 591         cpu->sample.mperf = mperf;
 592         cpu->sample.aperf -= cpu->prev_aperf;
 593         cpu->sample.mperf -= cpu->prev_mperf;
 594
 595         intel_pstate_calc_busy(cpu, &cpu->sample);
 596
 597         cpu->prev_aperf = aperf;
 598         cpu->prev_mperf = mperf;
 599 }
 600
 601 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 602 {
 603         int sample_time, delay;
 604
 605         sample_time = pid_params.sample_rate_ms;
 606         delay = msecs_to_jiffies(sample_time);
 607         mod_timer_pinned(&cpu->timer, jiffies + delay);
 608 }
 609
 610 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 611 {
 612         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 613         u32 duration_us;
 614         u32 sample_time;
 615
 616         core_busy = cpu->sample.core_pct_busy;
 617         max_pstate = int_tofp(cpu->pstate.max_pstate);
 618         current_pstate = int_tofp(cpu->pstate.current_pstate);
 619         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 620
 621         sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
 622         duration_us = (u32) ktime_us_delta(cpu->sample.time,
 623                                         cpu->last_sample_time);
 624         if (duration_us > sample_time * 3) {
 625                 sample_ratio = div_fp(int_tofp(sample_time),
 626                                 int_tofp(duration_us));
 627                 core_busy = mul_fp(core_busy, sample_ratio);
 628         }
 629
 630         return core_busy;
 631 }
 632
 633 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 634 {
 635         int32_t busy_scaled;
 636         struct _pid *pid;
 637         signed int ctl = 0;
 638         int steps;
 639
 640         pid = &cpu->pid;
 641         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 642
 643         ctl = pid_calc(pid, busy_scaled);
 644
 645         steps = abs(ctl);
 646
 647         if (ctl < 0)
 648                 intel_pstate_pstate_increase(cpu, steps);
 649         else
 650                 intel_pstate_pstate_decrease(cpu, steps);
 651 }
 652
 653 static void intel_pstate_timer_func(unsigned long __data)
 654 {
 655         struct cpudata *cpu = (struct cpudata *) __data;
 656         struct sample *sample;
 657
 658         intel_pstate_sample(cpu);
 659
 660         sample = &cpu->sample;
 661
 662         intel_pstate_adjust_busy_pstate(cpu);
 663
 664         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 665                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 666                         cpu->pstate.current_pstate,
 667                         sample->mperf,
 668                         sample->aperf,
 669                         sample->freq);
 670
 671         intel_pstate_set_sample_time(cpu);
 672 }
 673
 674 #define ICPU(model, policy) \
 675         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 676                         (unsigned long)&policy }
 677
 678 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 679         ICPU(0x2a, core_params),
 680         ICPU(0x2d, core_params),
 681         ICPU(0x37, byt_params),
 682         ICPU(0x3a, core_params),
 683         ICPU(0x3c, core_params),
 684         ICPU(0x3e, core_params),
 685         ICPU(0x3f, core_params),
 686         ICPU(0x45, core_params),
 687         ICPU(0x46, core_params),
 688         {}
 689 };
 690 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 691
 692 static int intel_pstate_init_cpu(unsigned int cpunum)
 693 {
 694
 695         const struct x86_cpu_id *id;
 696         struct cpudata *cpu;
 697
 698         id = x86_match_cpu(intel_pstate_cpu_ids);
 699         if (!id)
 700                 return -ENODEV;
 701
 702         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 703         if (!all_cpu_data[cpunum])
 704                 return -ENOMEM;
 705
 706         cpu = all_cpu_data[cpunum];
 707
 708         intel_pstate_get_cpu_pstates(cpu);
 709
 710         cpu->cpu = cpunum;
 711
 712         init_timer_deferrable(&cpu->timer);
 713         cpu->timer.function = intel_pstate_timer_func;
 714         cpu->timer.data =
 715                 (unsigned long)cpu;
 716         cpu->timer.expires = jiffies + HZ/100;
 717         intel_pstate_busy_pid_reset(cpu);
 718         intel_pstate_sample(cpu);
 719
 720         add_timer_on(&cpu->timer, cpunum);
 721
 722         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 723
 724         return 0;
 725 }
 726
 727 static unsigned int intel_pstate_get(unsigned int cpu_num)
 728 {
 729         struct sample *sample;
 730         struct cpudata *cpu;
 731
 732         cpu = all_cpu_data[cpu_num];
 733         if (!cpu)
 734                 return 0;
 735         sample = &cpu->sample;
 736         return sample->freq;
 737 }
 738
 739 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 740 {
 741         struct cpudata *cpu;
 742
 743         cpu = all_cpu_data[policy->cpu];
 744
 745         if (!policy->cpuinfo.max_freq)
 746                 return -ENODEV;
 747
 748         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 749                 limits.min_perf_pct = 100;
 750                 limits.min_perf = int_tofp(1);
 751                 limits.max_perf_pct = 100;
 752                 limits.max_perf = int_tofp(1);
 753                 limits.no_turbo = 0;
 754                 return 0;
 755         }
 756         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 757         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 758         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 759
 760         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 761         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 762         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 763         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 764
 765         return 0;
 766 }
 767
 768 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 769 {
 770         cpufreq_verify_within_cpu_limits(policy);
 771
 772         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 773                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 774                 return -EINVAL;
 775
 776         return 0;
 777 }
 778
 779 static int intel_pstate_cpu_exit(struct cpufreq_policy *policy)
 780 {
 781         int cpu = policy->cpu;
 782
 783         del_timer(&all_cpu_data[cpu]->timer);
 784         kfree(all_cpu_data[cpu]);
 785         all_cpu_data[cpu] = NULL;
 786         return 0;
 787 }
 788
 789 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 790 {
 791         struct cpudata *cpu;
 792         int rc;
 793
 794         rc = intel_pstate_init_cpu(policy->cpu);
 795         if (rc)
 796                 return rc;
 797
 798         cpu = all_cpu_data[policy->cpu];
 799
 800         if (!limits.no_turbo &&
 801                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 802                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 803         else
 804                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 805
 806         policy->min = cpu->pstate.min_pstate * 100000;
 807         policy->max = cpu->pstate.turbo_pstate * 100000;
 808
 809         /* cpuinfo and default policy values */
 810         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 811         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 812         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 813         cpumask_set_cpu(policy->cpu, policy->cpus);
 814
 815         return 0;
 816 }
 817
 818 static struct cpufreq_driver intel_pstate_driver = {
 819         .flags          = CPUFREQ_CONST_LOOPS,
 820         .verify         = intel_pstate_verify_policy,
 821         .setpolicy      = intel_pstate_set_policy,
 822         .get            = intel_pstate_get,
 823         .init           = intel_pstate_cpu_init,
 824         .exit           = intel_pstate_cpu_exit,
 825         .name           = "intel_pstate",
 826 };
 827
 828 static int __initdata no_load;
 829
 830 static int intel_pstate_msrs_not_valid(void)
 831 {
 832         /* Check that all the msr's we are using are valid. */
 833         u64 aperf, mperf, tmp;
 834
 835         rdmsrl(MSR_IA32_APERF, aperf);
 836         rdmsrl(MSR_IA32_MPERF, mperf);
 837
 838         if (!pstate_funcs.get_max() ||
 839                 !pstate_funcs.get_min() ||
 840                 !pstate_funcs.get_turbo())
 841                 return -ENODEV;
 842
 843         rdmsrl(MSR_IA32_APERF, tmp);
 844         if (!(tmp - aperf))
 845                 return -ENODEV;
 846
 847         rdmsrl(MSR_IA32_MPERF, tmp);
 848         if (!(tmp - mperf))
 849                 return -ENODEV;
 850
 851         return 0;
 852 }
 853
 854 static void copy_pid_params(struct pstate_adjust_policy *policy)
 855 {
 856         pid_params.sample_rate_ms = policy->sample_rate_ms;
 857         pid_params.p_gain_pct = policy->p_gain_pct;
 858         pid_params.i_gain_pct = policy->i_gain_pct;
 859         pid_params.d_gain_pct = policy->d_gain_pct;
 860         pid_params.deadband = policy->deadband;
 861         pid_params.setpoint = policy->setpoint;
 862 }
 863
 864 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 865 {
 866         pstate_funcs.get_max   = funcs->get_max;
 867         pstate_funcs.get_min   = funcs->get_min;
 868         pstate_funcs.get_turbo = funcs->get_turbo;
 869         pstate_funcs.set       = funcs->set;
 870         pstate_funcs.get_vid   = funcs->get_vid;
 871 }
 872
 873 #if IS_ENABLED(CONFIG_ACPI)
 874 #include <acpi/processor.h>
 875
 876 static bool intel_pstate_no_acpi_pss(void)
 877 {
 878         int i;
 879
 880         for_each_possible_cpu(i) {
 881                 acpi_status status;
 882                 union acpi_object *pss;
 883                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 884                 struct acpi_processor *pr = per_cpu(processors, i);
 885
 886                 if (!pr)
 887                         continue;
 888
 889                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 890                 if (ACPI_FAILURE(status))
 891                         continue;
 892
 893                 pss = buffer.pointer;
 894                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 895                         kfree(pss);
 896                         return false;
 897                 }
 898
 899                 kfree(pss);
 900         }
 901
 902         return true;
 903 }
 904
 905 struct hw_vendor_info {
 906         u16  valid;
 907         char oem_id[ACPI_OEM_ID_SIZE];
 908         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 909 };
 910
 911 /* Hardware vendor-specific info that has its own power management modes */
 912 static struct hw_vendor_info vendor_info[] = {
 913         {1, "HP    ", "ProLiant"},
 914         {0, "", ""},
 915 };
 916
 917 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 918 {
 919         struct acpi_table_header hdr;
 920         struct hw_vendor_info *v_info;
 921
 922         if (acpi_disabled
 923             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 924                 return false;
 925
 926         for (v_info = vendor_info; v_info->valid; v_info++) {
 927                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 928                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 929                     && intel_pstate_no_acpi_pss())
 930                         return true;
 931         }
 932
 933         return false;
 934 }
 935 #else /* CONFIG_ACPI not enabled */
 936 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 937 #endif /* CONFIG_ACPI */
 938
 939 static int __init intel_pstate_init(void)
 940 {
 941         int cpu, rc = 0;
 942         const struct x86_cpu_id *id;
 943         struct cpu_defaults *cpu_info;
 944
 945         if (no_load)
 946                 return -ENODEV;
 947
 948         id = x86_match_cpu(intel_pstate_cpu_ids);
 949         if (!id)
 950                 return -ENODEV;
 951
 952         /*
 953          * The Intel pstate driver will be ignored if the platform
 954          * firmware has its own power management modes.
 955          */
 956         if (intel_pstate_platform_pwr_mgmt_exists())
 957                 return -ENODEV;
 958
 959         cpu_info = (struct cpu_defaults *)id->driver_data;
 960
 961         copy_pid_params(&cpu_info->pid_policy);
 962         copy_cpu_funcs(&cpu_info->funcs);
 963
 964         if (intel_pstate_msrs_not_valid())
 965                 return -ENODEV;
 966
 967         pr_info("Intel P-state driver initializing.\n");
 968
 969         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 970         if (!all_cpu_data)
 971                 return -ENOMEM;
 972
 973         rc = cpufreq_register_driver(&intel_pstate_driver);
 974         if (rc)
 975                 goto out;
 976
 977         intel_pstate_debug_expose_params();
 978         intel_pstate_sysfs_expose_params();
 979
 980         return rc;
 981 out:
 982         get_online_cpus();
 983         for_each_online_cpu(cpu) {
 984                 if (all_cpu_data[cpu]) {
 985                         del_timer_sync(&all_cpu_data[cpu]->timer);
 986                         kfree(all_cpu_data[cpu]);
 987                 }
 988         }
 989
 990         put_online_cpus();
 991         vfree(all_cpu_data);
 992         return -ENODEV;
 993 }
 994 device_initcall(intel_pstate_init);
 995
 996 static int __init intel_pstate_setup(char *str)
 997 {
 998         if (!str)
 999                 return -EINVAL;
1000
1001         if (!strcmp(str, "disable"))
1002                 no_load = 1;
1003         return 0;
1004 }
1005 early_param("intel_pstate", intel_pstate_setup);
1006
1007 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1008 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1009 MODULE_LICENSE("GPL");