drivers/cpufreq/intel_pstate.c

   1 /*
   2  * intel_pstate.c: Native P state management for Intel processors
   3  *
   4  * (C) Copyright 2012 Intel Corporation
   5  * Author: Dirk Brandewie <dirk.j.brandewie@intel.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License
   9  * as published by the Free Software Foundation; version 2
  10  * of the License.
  11  */
  12
  13 #include <linux/kernel.h>
  14 #include <linux/kernel_stat.h>
  15 #include <linux/module.h>
  16 #include <linux/ktime.h>
  17 #include <linux/hrtimer.h>
  18 #include <linux/tick.h>
  19 #include <linux/slab.h>
  20 #include <linux/sched.h>
  21 #include <linux/list.h>
  22 #include <linux/cpu.h>
  23 #include <linux/cpufreq.h>
  24 #include <linux/sysfs.h>
  25 #include <linux/types.h>
  26 #include <linux/fs.h>
  27 #include <linux/debugfs.h>
  28 #include <linux/acpi.h>
  29 #include <trace/events/power.h>
  30
  31 #include <asm/div64.h>
  32 #include <asm/msr.h>
  33 #include <asm/cpu_device_id.h>
  34
  35 #define BYT_RATIOS              0x66a
  36 #define BYT_VIDS                0x66b
  37 #define BYT_TURBO_RATIOS        0x66c
  38 #define BYT_TURBO_VIDS          0x66d
  39
  40
  41 #define FRAC_BITS 8
  42 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  43 #define fp_toint(X) ((X) >> FRAC_BITS)
  44
  45
  46 static inline int32_t mul_fp(int32_t x, int32_t y)
  47 {
  48         return ((int64_t)x * (int64_t)y) >> FRAC_BITS;
  49 }
  50
  51 static inline int32_t div_fp(s64 x, s64 y)
  52 {
  53         return div64_s64((int64_t)x << FRAC_BITS, y);
  54 }
  55
  56 static inline int ceiling_fp(int32_t x)
  57 {
  58         int mask, ret;
  59
  60         ret = fp_toint(x);
  61         mask = (1 << FRAC_BITS) - 1;
  62         if (x & mask)
  63                 ret += 1;
  64         return ret;
  65 }
  66
  67 struct sample {
  68         int32_t core_pct_busy;
  69         u64 aperf;
  70         u64 mperf;
  71         int freq;
  72         ktime_t time;
  73 };
  74
  75 struct pstate_data {
  76         int     current_pstate;
  77         int     min_pstate;
  78         int     max_pstate;
  79         int     scaling;
  80         int     turbo_pstate;
  81 };
  82
  83 struct vid_data {
  84         int min;
  85         int max;
  86         int turbo;
  87         int32_t ratio;
  88 };
  89
  90 struct _pid {
  91         int setpoint;
  92         int32_t integral;
  93         int32_t p_gain;
  94         int32_t i_gain;
  95         int32_t d_gain;
  96         int deadband;
  97         int32_t last_err;
  98 };
  99
 100 struct cpudata {
 101         int cpu;
 102
 103         struct timer_list timer;
 104
 105         struct pstate_data pstate;
 106         struct vid_data vid;
 107         struct _pid pid;
 108
 109         ktime_t last_sample_time;
 110         u64     prev_aperf;
 111         u64     prev_mperf;
 112         struct sample sample;
 113 };
 114
 115 static struct cpudata **all_cpu_data;
 116 struct pstate_adjust_policy {
 117         int sample_rate_ms;
 118         int deadband;
 119         int setpoint;
 120         int p_gain_pct;
 121         int d_gain_pct;
 122         int i_gain_pct;
 123 };
 124
 125 struct pstate_funcs {
 126         int (*get_max)(void);
 127         int (*get_min)(void);
 128         int (*get_turbo)(void);
 129         int (*get_scaling)(void);
 130         void (*set)(struct cpudata*, int pstate);
 131         void (*get_vid)(struct cpudata *);
 132 };
 133
 134 struct cpu_defaults {
 135         struct pstate_adjust_policy pid_policy;
 136         struct pstate_funcs funcs;
 137 };
 138
 139 static struct pstate_adjust_policy pid_params;
 140 static struct pstate_funcs pstate_funcs;
 141
 142 struct perf_limits {
 143         int no_turbo;
 144         int turbo_disabled;
 145         int max_perf_pct;
 146         int min_perf_pct;
 147         int32_t max_perf;
 148         int32_t min_perf;
 149         int max_policy_pct;
 150         int max_sysfs_pct;
 151 };
 152
 153 static struct perf_limits limits = {
 154         .no_turbo = 0,
 155         .turbo_disabled = 0,
 156         .max_perf_pct = 100,
 157         .max_perf = int_tofp(1),
 158         .min_perf_pct = 0,
 159         .min_perf = 0,
 160         .max_policy_pct = 100,
 161         .max_sysfs_pct = 100,
 162 };
 163
 164 static inline void pid_reset(struct _pid *pid, int setpoint, int busy,
 165                         int deadband, int integral) {
 166         pid->setpoint = setpoint;
 167         pid->deadband  = deadband;
 168         pid->integral  = int_tofp(integral);
 169         pid->last_err  = int_tofp(setpoint) - int_tofp(busy);
 170 }
 171
 172 static inline void pid_p_gain_set(struct _pid *pid, int percent)
 173 {
 174         pid->p_gain = div_fp(int_tofp(percent), int_tofp(100));
 175 }
 176
 177 static inline void pid_i_gain_set(struct _pid *pid, int percent)
 178 {
 179         pid->i_gain = div_fp(int_tofp(percent), int_tofp(100));
 180 }
 181
 182 static inline void pid_d_gain_set(struct _pid *pid, int percent)
 183 {
 184
 185         pid->d_gain = div_fp(int_tofp(percent), int_tofp(100));
 186 }
 187
 188 static signed int pid_calc(struct _pid *pid, int32_t busy)
 189 {
 190         signed int result;
 191         int32_t pterm, dterm, fp_error;
 192         int32_t integral_limit;
 193
 194         fp_error = int_tofp(pid->setpoint) - busy;
 195
 196         if (abs(fp_error) <= int_tofp(pid->deadband))
 197                 return 0;
 198
 199         pterm = mul_fp(pid->p_gain, fp_error);
 200
 201         pid->integral += fp_error;
 202
 203         /* limit the integral term */
 204         integral_limit = int_tofp(30);
 205         if (pid->integral > integral_limit)
 206                 pid->integral = integral_limit;
 207         if (pid->integral < -integral_limit)
 208                 pid->integral = -integral_limit;
 209
 210         dterm = mul_fp(pid->d_gain, fp_error - pid->last_err);
 211         pid->last_err = fp_error;
 212
 213         result = pterm + mul_fp(pid->integral, pid->i_gain) + dterm;
 214         result = result + (1 << (FRAC_BITS-1));
 215         return (signed int)fp_toint(result);
 216 }
 217
 218 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 219 {
 220         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 221         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 222         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 223
 224         pid_reset(&cpu->pid,
 225                 pid_params.setpoint,
 226                 100,
 227                 pid_params.deadband,
 228                 0);
 229 }
 230
 231 static inline void intel_pstate_reset_all_pid(void)
 232 {
 233         unsigned int cpu;
 234         for_each_online_cpu(cpu) {
 235                 if (all_cpu_data[cpu])
 236                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 237         }
 238 }
 239
 240 static inline void update_turbo_state(void)
 241 {
 242         u64 misc_en;
 243         struct cpudata *cpu;
 244
 245         cpu = all_cpu_data[0];
 246         rdmsrl(MSR_IA32_MISC_ENABLE, misc_en);
 247         limits.turbo_disabled =
 248                 (misc_en & MSR_IA32_MISC_ENABLE_TURBO_DISABLE ||
 249                  cpu->pstate.max_pstate == cpu->pstate.turbo_pstate);
 250 }
 251
 252 /************************** debugfs begin ************************/
 253 static int pid_param_set(void *data, u64 val)
 254 {
 255         *(u32 *)data = val;
 256         intel_pstate_reset_all_pid();
 257         return 0;
 258 }
 259 static int pid_param_get(void *data, u64 *val)
 260 {
 261         *val = *(u32 *)data;
 262         return 0;
 263 }
 264 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 265                         pid_param_set, "%llu\n");
 266
 267 struct pid_param {
 268         char *name;
 269         void *value;
 270 };
 271
 272 static struct pid_param pid_files[] = {
 273         {"sample_rate_ms", &pid_params.sample_rate_ms},
 274         {"d_gain_pct", &pid_params.d_gain_pct},
 275         {"i_gain_pct", &pid_params.i_gain_pct},
 276         {"deadband", &pid_params.deadband},
 277         {"setpoint", &pid_params.setpoint},
 278         {"p_gain_pct", &pid_params.p_gain_pct},
 279         {NULL, NULL}
 280 };
 281
 282 static struct dentry *debugfs_parent;
 283 static void intel_pstate_debug_expose_params(void)
 284 {
 285         int i = 0;
 286
 287         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 288         if (IS_ERR_OR_NULL(debugfs_parent))
 289                 return;
 290         while (pid_files[i].name) {
 291                 debugfs_create_file(pid_files[i].name, 0660,
 292                                 debugfs_parent, pid_files[i].value,
 293                                 &fops_pid_param);
 294                 i++;
 295         }
 296 }
 297
 298 /************************** debugfs end ************************/
 299
 300 /************************** sysfs begin ************************/
 301 #define show_one(file_name, object)                                     \
 302         static ssize_t show_##file_name                                 \
 303         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 304         {                                                               \
 305                 return sprintf(buf, "%u\n", limits.object);             \
 306         }
 307
 308 static ssize_t show_no_turbo(struct kobject *kobj,
 309                              struct attribute *attr, char *buf)
 310 {
 311         ssize_t ret;
 312
 313         update_turbo_state();
 314         if (limits.turbo_disabled)
 315                 ret = sprintf(buf, "%u\n", limits.turbo_disabled);
 316         else
 317                 ret = sprintf(buf, "%u\n", limits.no_turbo);
 318
 319         return ret;
 320 }
 321
 322 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 323                                 const char *buf, size_t count)
 324 {
 325         unsigned int input;
 326         int ret;
 327         ret = sscanf(buf, "%u", &input);
 328         if (ret != 1)
 329                 return -EINVAL;
 330
 331         update_turbo_state();
 332         if (limits.turbo_disabled) {
 333                 pr_warn("Turbo disabled by BIOS or unavailable on processor\n");
 334                 return -EPERM;
 335         }
 336         limits.no_turbo = clamp_t(int, input, 0, 1);
 337
 338         return count;
 339 }
 340
 341 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 342                                 const char *buf, size_t count)
 343 {
 344         unsigned int input;
 345         int ret;
 346         ret = sscanf(buf, "%u", &input);
 347         if (ret != 1)
 348                 return -EINVAL;
 349
 350         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 351         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 352         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 353         return count;
 354 }
 355
 356 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 357                                 const char *buf, size_t count)
 358 {
 359         unsigned int input;
 360         int ret;
 361         ret = sscanf(buf, "%u", &input);
 362         if (ret != 1)
 363                 return -EINVAL;
 364         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 365         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 366
 367         return count;
 368 }
 369
 370 show_one(max_perf_pct, max_perf_pct);
 371 show_one(min_perf_pct, min_perf_pct);
 372
 373 define_one_global_rw(no_turbo);
 374 define_one_global_rw(max_perf_pct);
 375 define_one_global_rw(min_perf_pct);
 376
 377 static struct attribute *intel_pstate_attributes[] = {
 378         &no_turbo.attr,
 379         &max_perf_pct.attr,
 380         &min_perf_pct.attr,
 381         NULL
 382 };
 383
 384 static struct attribute_group intel_pstate_attr_group = {
 385         .attrs = intel_pstate_attributes,
 386 };
 387 static struct kobject *intel_pstate_kobject;
 388
 389 static void intel_pstate_sysfs_expose_params(void)
 390 {
 391         int rc;
 392
 393         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 394                                                 &cpu_subsys.dev_root->kobj);
 395         BUG_ON(!intel_pstate_kobject);
 396         rc = sysfs_create_group(intel_pstate_kobject,
 397                                 &intel_pstate_attr_group);
 398         BUG_ON(rc);
 399 }
 400
 401 /************************** sysfs end ************************/
 402 static int byt_get_min_pstate(void)
 403 {
 404         u64 value;
 405         rdmsrl(BYT_RATIOS, value);
 406         return (value >> 8) & 0x7F;
 407 }
 408
 409 static int byt_get_max_pstate(void)
 410 {
 411         u64 value;
 412         rdmsrl(BYT_RATIOS, value);
 413         return (value >> 16) & 0x7F;
 414 }
 415
 416 static int byt_get_turbo_pstate(void)
 417 {
 418         u64 value;
 419         rdmsrl(BYT_TURBO_RATIOS, value);
 420         return value & 0x7F;
 421 }
 422
 423 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 424 {
 425         u64 val;
 426         int32_t vid_fp;
 427         u32 vid;
 428
 429         val = pstate << 8;
 430         if (limits.no_turbo && !limits.turbo_disabled)
 431                 val |= (u64)1 << 32;
 432
 433         vid_fp = cpudata->vid.min + mul_fp(
 434                 int_tofp(pstate - cpudata->pstate.min_pstate),
 435                 cpudata->vid.ratio);
 436
 437         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 438         vid = ceiling_fp(vid_fp);
 439
 440         if (pstate > cpudata->pstate.max_pstate)
 441                 vid = cpudata->vid.turbo;
 442
 443         val |= vid;
 444
 445         wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 446 }
 447
 448 #define BYT_BCLK_FREQS 5
 449 static int byt_freq_table[BYT_BCLK_FREQS] = { 833, 1000, 1333, 1167, 800};
 450
 451 static int byt_get_scaling(void)
 452 {
 453         u64 value;
 454         int i;
 455
 456         rdmsrl(MSR_FSB_FREQ, value);
 457         i = value & 0x3;
 458
 459         BUG_ON(i > BYT_BCLK_FREQS);
 460
 461         return byt_freq_table[i] * 100;
 462 }
 463
 464 static void byt_get_vid(struct cpudata *cpudata)
 465 {
 466         u64 value;
 467
 468
 469         rdmsrl(BYT_VIDS, value);
 470         cpudata->vid.min = int_tofp((value >> 8) & 0x7f);
 471         cpudata->vid.max = int_tofp((value >> 16) & 0x7f);
 472         cpudata->vid.ratio = div_fp(
 473                 cpudata->vid.max - cpudata->vid.min,
 474                 int_tofp(cpudata->pstate.max_pstate -
 475                         cpudata->pstate.min_pstate));
 476
 477         rdmsrl(BYT_TURBO_VIDS, value);
 478         cpudata->vid.turbo = value & 0x7f;
 479 }
 480
 481
 482 static int core_get_min_pstate(void)
 483 {
 484         u64 value;
 485         rdmsrl(MSR_PLATFORM_INFO, value);
 486         return (value >> 40) & 0xFF;
 487 }
 488
 489 static int core_get_max_pstate(void)
 490 {
 491         u64 value;
 492         rdmsrl(MSR_PLATFORM_INFO, value);
 493         return (value >> 8) & 0xFF;
 494 }
 495
 496 static int core_get_turbo_pstate(void)
 497 {
 498         u64 value;
 499         int nont, ret;
 500         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 501         nont = core_get_max_pstate();
 502         ret = ((value) & 255);
 503         if (ret <= nont)
 504                 ret = nont;
 505         return ret;
 506 }
 507
 508 static inline int core_get_scaling(void)
 509 {
 510         return 100000;
 511 }
 512
 513 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 514 {
 515         u64 val;
 516
 517         val = pstate << 8;
 518         if (limits.no_turbo && !limits.turbo_disabled)
 519                 val |= (u64)1 << 32;
 520
 521         wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 522 }
 523
 524 static struct cpu_defaults core_params = {
 525         .pid_policy = {
 526                 .sample_rate_ms = 10,
 527                 .deadband = 0,
 528                 .setpoint = 97,
 529                 .p_gain_pct = 20,
 530                 .d_gain_pct = 0,
 531                 .i_gain_pct = 0,
 532         },
 533         .funcs = {
 534                 .get_max = core_get_max_pstate,
 535                 .get_min = core_get_min_pstate,
 536                 .get_turbo = core_get_turbo_pstate,
 537                 .get_scaling = core_get_scaling,
 538                 .set = core_set_pstate,
 539         },
 540 };
 541
 542 static struct cpu_defaults byt_params = {
 543         .pid_policy = {
 544                 .sample_rate_ms = 10,
 545                 .deadband = 0,
 546                 .setpoint = 97,
 547                 .p_gain_pct = 14,
 548                 .d_gain_pct = 0,
 549                 .i_gain_pct = 4,
 550         },
 551         .funcs = {
 552                 .get_max = byt_get_max_pstate,
 553                 .get_min = byt_get_min_pstate,
 554                 .get_turbo = byt_get_turbo_pstate,
 555                 .set = byt_set_pstate,
 556                 .get_scaling = byt_get_scaling,
 557                 .get_vid = byt_get_vid,
 558         },
 559 };
 560
 561
 562 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 563 {
 564         int max_perf = cpu->pstate.turbo_pstate;
 565         int max_perf_adj;
 566         int min_perf;
 567
 568         if (limits.no_turbo || limits.turbo_disabled)
 569                 max_perf = cpu->pstate.max_pstate;
 570
 571         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 572         *max = clamp_t(int, max_perf_adj,
 573                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 574
 575         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 576         *min = clamp_t(int, min_perf,
 577                         cpu->pstate.min_pstate, max_perf);
 578 }
 579
 580 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 581 {
 582         int max_perf, min_perf;
 583
 584         update_turbo_state();
 585
 586         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 587
 588         pstate = clamp_t(int, pstate, min_perf, max_perf);
 589
 590         if (pstate == cpu->pstate.current_pstate)
 591                 return;
 592
 593         trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
 594
 595         cpu->pstate.current_pstate = pstate;
 596
 597         pstate_funcs.set(cpu, pstate);
 598 }
 599
 600 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 601 {
 602         int target;
 603         target = cpu->pstate.current_pstate + steps;
 604
 605         intel_pstate_set_pstate(cpu, target);
 606 }
 607
 608 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 609 {
 610         int target;
 611         target = cpu->pstate.current_pstate - steps;
 612         intel_pstate_set_pstate(cpu, target);
 613 }
 614
 615 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 616 {
 617         cpu->pstate.min_pstate = pstate_funcs.get_min();
 618         cpu->pstate.max_pstate = pstate_funcs.get_max();
 619         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 620         cpu->pstate.scaling = pstate_funcs.get_scaling();
 621
 622         if (pstate_funcs.get_vid)
 623                 pstate_funcs.get_vid(cpu);
 624         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 625 }
 626
 627 static inline void intel_pstate_calc_busy(struct cpudata *cpu)
 628 {
 629         struct sample *sample = &cpu->sample;
 630         int64_t core_pct;
 631         int32_t rem;
 632
 633         core_pct = int_tofp(sample->aperf) * int_tofp(100);
 634         core_pct = div_u64_rem(core_pct, int_tofp(sample->mperf), &rem);
 635
 636         if ((rem << 1) >= int_tofp(sample->mperf))
 637                 core_pct += 1;
 638
 639         sample->freq = fp_toint(
 640                 mul_fp(int_tofp(
 641                         cpu->pstate.max_pstate * cpu->pstate.scaling / 100),
 642                         core_pct));
 643
 644         sample->core_pct_busy = (int32_t)core_pct;
 645 }
 646
 647 static inline void intel_pstate_sample(struct cpudata *cpu)
 648 {
 649         u64 aperf, mperf;
 650
 651         rdmsrl(MSR_IA32_APERF, aperf);
 652         rdmsrl(MSR_IA32_MPERF, mperf);
 653
 654         aperf = aperf >> FRAC_BITS;
 655         mperf = mperf >> FRAC_BITS;
 656
 657         cpu->last_sample_time = cpu->sample.time;
 658         cpu->sample.time = ktime_get();
 659         cpu->sample.aperf = aperf;
 660         cpu->sample.mperf = mperf;
 661         cpu->sample.aperf -= cpu->prev_aperf;
 662         cpu->sample.mperf -= cpu->prev_mperf;
 663
 664         intel_pstate_calc_busy(cpu);
 665
 666         cpu->prev_aperf = aperf;
 667         cpu->prev_mperf = mperf;
 668 }
 669
 670 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 671 {
 672         int sample_time, delay;
 673
 674         sample_time = pid_params.sample_rate_ms;
 675         delay = msecs_to_jiffies(sample_time);
 676         mod_timer_pinned(&cpu->timer, jiffies + delay);
 677 }
 678
 679 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 680 {
 681         int32_t core_busy, max_pstate, current_pstate, sample_ratio;
 682         s64 duration_us;
 683         u32 sample_time;
 684
 685         core_busy = cpu->sample.core_pct_busy;
 686         max_pstate = int_tofp(cpu->pstate.max_pstate);
 687         current_pstate = int_tofp(cpu->pstate.current_pstate);
 688         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 689
 690         sample_time = (pid_params.sample_rate_ms  * USEC_PER_MSEC);
 691         duration_us = ktime_us_delta(cpu->sample.time,
 692                                      cpu->last_sample_time);
 693         if (duration_us > sample_time * 3) {
 694                 sample_ratio = div_fp(int_tofp(sample_time),
 695                                 int_tofp(duration_us));
 696                 core_busy = mul_fp(core_busy, sample_ratio);
 697         }
 698
 699         return core_busy;
 700 }
 701
 702 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 703 {
 704         int32_t busy_scaled;
 705         struct _pid *pid;
 706         signed int ctl = 0;
 707         int steps;
 708
 709         pid = &cpu->pid;
 710         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 711
 712         ctl = pid_calc(pid, busy_scaled);
 713
 714         steps = abs(ctl);
 715
 716         if (ctl < 0)
 717                 intel_pstate_pstate_increase(cpu, steps);
 718         else
 719                 intel_pstate_pstate_decrease(cpu, steps);
 720 }
 721
 722 static void intel_pstate_timer_func(unsigned long __data)
 723 {
 724         struct cpudata *cpu = (struct cpudata *) __data;
 725         struct sample *sample;
 726
 727         intel_pstate_sample(cpu);
 728
 729         sample = &cpu->sample;
 730
 731         intel_pstate_adjust_busy_pstate(cpu);
 732
 733         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 734                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 735                         cpu->pstate.current_pstate,
 736                         sample->mperf,
 737                         sample->aperf,
 738                         sample->freq);
 739
 740         intel_pstate_set_sample_time(cpu);
 741 }
 742
 743 #define ICPU(model, policy) \
 744         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 745                         (unsigned long)&policy }
 746
 747 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 748         ICPU(0x2a, core_params),
 749         ICPU(0x2d, core_params),
 750         ICPU(0x37, byt_params),
 751         ICPU(0x3a, core_params),
 752         ICPU(0x3c, core_params),
 753         ICPU(0x3d, core_params),
 754         ICPU(0x3e, core_params),
 755         ICPU(0x3f, core_params),
 756         ICPU(0x45, core_params),
 757         ICPU(0x46, core_params),
 758         ICPU(0x4c, byt_params),
 759         ICPU(0x4f, core_params),
 760         ICPU(0x56, core_params),
 761         {}
 762 };
 763 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 764
 765 static int intel_pstate_init_cpu(unsigned int cpunum)
 766 {
 767         struct cpudata *cpu;
 768
 769         if (!all_cpu_data[cpunum])
 770                 all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata),
 771                                                GFP_KERNEL);
 772         if (!all_cpu_data[cpunum])
 773                 return -ENOMEM;
 774
 775         cpu = all_cpu_data[cpunum];
 776
 777         cpu->cpu = cpunum;
 778         intel_pstate_get_cpu_pstates(cpu);
 779
 780         init_timer_deferrable(&cpu->timer);
 781         cpu->timer.function = intel_pstate_timer_func;
 782         cpu->timer.data =
 783                 (unsigned long)cpu;
 784         cpu->timer.expires = jiffies + HZ/100;
 785         intel_pstate_busy_pid_reset(cpu);
 786         intel_pstate_sample(cpu);
 787
 788         add_timer_on(&cpu->timer, cpunum);
 789
 790         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 791
 792         return 0;
 793 }
 794
 795 static unsigned int intel_pstate_get(unsigned int cpu_num)
 796 {
 797         struct sample *sample;
 798         struct cpudata *cpu;
 799
 800         cpu = all_cpu_data[cpu_num];
 801         if (!cpu)
 802                 return 0;
 803         sample = &cpu->sample;
 804         return sample->freq;
 805 }
 806
 807 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 808 {
 809         struct cpudata *cpu;
 810
 811         cpu = all_cpu_data[policy->cpu];
 812
 813         if (!policy->cpuinfo.max_freq)
 814                 return -ENODEV;
 815
 816         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 817                 limits.min_perf_pct = 100;
 818                 limits.min_perf = int_tofp(1);
 819                 limits.max_policy_pct = 100;
 820                 limits.max_perf_pct = 100;
 821                 limits.max_perf = int_tofp(1);
 822                 limits.no_turbo = 0;
 823                 return 0;
 824         }
 825         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 826         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 827         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 828
 829         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 830         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 831         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 832         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 833
 834         return 0;
 835 }
 836
 837 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 838 {
 839         cpufreq_verify_within_cpu_limits(policy);
 840
 841         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 842                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 843                 return -EINVAL;
 844
 845         return 0;
 846 }
 847
 848 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
 849 {
 850         int cpu_num = policy->cpu;
 851         struct cpudata *cpu = all_cpu_data[cpu_num];
 852
 853         pr_info("intel_pstate CPU %d exiting\n", cpu_num);
 854
 855         del_timer_sync(&all_cpu_data[cpu_num]->timer);
 856         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 857 }
 858
 859 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 860 {
 861         struct cpudata *cpu;
 862         int rc;
 863
 864         rc = intel_pstate_init_cpu(policy->cpu);
 865         if (rc)
 866                 return rc;
 867
 868         cpu = all_cpu_data[policy->cpu];
 869
 870         if (limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 871                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 872         else
 873                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 874
 875         policy->min = cpu->pstate.min_pstate * cpu->pstate.scaling;
 876         policy->max = cpu->pstate.turbo_pstate * cpu->pstate.scaling;
 877
 878         /* cpuinfo and default policy values */
 879         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * cpu->pstate.scaling;
 880         update_turbo_state();
 881         policy->cpuinfo.max_freq = limits.turbo_disabled ?
 882                         cpu->pstate.max_pstate : cpu->pstate.turbo_pstate;
 883         policy->cpuinfo.max_freq *= cpu->pstate.scaling;
 884
 885         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 886         cpumask_set_cpu(policy->cpu, policy->cpus);
 887
 888         return 0;
 889 }
 890
 891 static struct cpufreq_driver intel_pstate_driver = {
 892         .flags          = CPUFREQ_CONST_LOOPS,
 893         .verify         = intel_pstate_verify_policy,
 894         .setpolicy      = intel_pstate_set_policy,
 895         .get            = intel_pstate_get,
 896         .init           = intel_pstate_cpu_init,
 897         .stop_cpu       = intel_pstate_stop_cpu,
 898         .name           = "intel_pstate",
 899 };
 900
 901 static int __initdata no_load;
 902
 903 static int intel_pstate_msrs_not_valid(void)
 904 {
 905         /* Check that all the msr's we are using are valid. */
 906         u64 aperf, mperf, tmp;
 907
 908         rdmsrl(MSR_IA32_APERF, aperf);
 909         rdmsrl(MSR_IA32_MPERF, mperf);
 910
 911         if (!pstate_funcs.get_max() ||
 912                 !pstate_funcs.get_min() ||
 913                 !pstate_funcs.get_turbo())
 914                 return -ENODEV;
 915
 916         rdmsrl(MSR_IA32_APERF, tmp);
 917         if (!(tmp - aperf))
 918                 return -ENODEV;
 919
 920         rdmsrl(MSR_IA32_MPERF, tmp);
 921         if (!(tmp - mperf))
 922                 return -ENODEV;
 923
 924         return 0;
 925 }
 926
 927 static void copy_pid_params(struct pstate_adjust_policy *policy)
 928 {
 929         pid_params.sample_rate_ms = policy->sample_rate_ms;
 930         pid_params.p_gain_pct = policy->p_gain_pct;
 931         pid_params.i_gain_pct = policy->i_gain_pct;
 932         pid_params.d_gain_pct = policy->d_gain_pct;
 933         pid_params.deadband = policy->deadband;
 934         pid_params.setpoint = policy->setpoint;
 935 }
 936
 937 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 938 {
 939         pstate_funcs.get_max   = funcs->get_max;
 940         pstate_funcs.get_min   = funcs->get_min;
 941         pstate_funcs.get_turbo = funcs->get_turbo;
 942         pstate_funcs.get_scaling = funcs->get_scaling;
 943         pstate_funcs.set       = funcs->set;
 944         pstate_funcs.get_vid   = funcs->get_vid;
 945 }
 946
 947 #if IS_ENABLED(CONFIG_ACPI)
 948 #include <acpi/processor.h>
 949
 950 static bool intel_pstate_no_acpi_pss(void)
 951 {
 952         int i;
 953
 954         for_each_possible_cpu(i) {
 955                 acpi_status status;
 956                 union acpi_object *pss;
 957                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 958                 struct acpi_processor *pr = per_cpu(processors, i);
 959
 960                 if (!pr)
 961                         continue;
 962
 963                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 964                 if (ACPI_FAILURE(status))
 965                         continue;
 966
 967                 pss = buffer.pointer;
 968                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 969                         kfree(pss);
 970                         return false;
 971                 }
 972
 973                 kfree(pss);
 974         }
 975
 976         return true;
 977 }
 978
 979 struct hw_vendor_info {
 980         u16  valid;
 981         char oem_id[ACPI_OEM_ID_SIZE];
 982         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 983 };
 984
 985 /* Hardware vendor-specific info that has its own power management modes */
 986 static struct hw_vendor_info vendor_info[] = {
 987         {1, "HP    ", "ProLiant"},
 988         {0, "", ""},
 989 };
 990
 991 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 992 {
 993         struct acpi_table_header hdr;
 994         struct hw_vendor_info *v_info;
 995
 996         if (acpi_disabled
 997             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 998                 return false;
 999
1000         for (v_info = vendor_info; v_info->valid; v_info++) {
1001                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
1002                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
1003                     && intel_pstate_no_acpi_pss())
1004                         return true;
1005         }
1006
1007         return false;
1008 }
1009 #else /* CONFIG_ACPI not enabled */
1010 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
1011 #endif /* CONFIG_ACPI */
1012
1013 static int __init intel_pstate_init(void)
1014 {
1015         int cpu, rc = 0;
1016         const struct x86_cpu_id *id;
1017         struct cpu_defaults *cpu_info;
1018
1019         if (no_load)
1020                 return -ENODEV;
1021
1022         id = x86_match_cpu(intel_pstate_cpu_ids);
1023         if (!id)
1024                 return -ENODEV;
1025
1026         /*
1027          * The Intel pstate driver will be ignored if the platform
1028          * firmware has its own power management modes.
1029          */
1030         if (intel_pstate_platform_pwr_mgmt_exists())
1031                 return -ENODEV;
1032
1033         cpu_info = (struct cpu_defaults *)id->driver_data;
1034
1035         copy_pid_params(&cpu_info->pid_policy);
1036         copy_cpu_funcs(&cpu_info->funcs);
1037
1038         if (intel_pstate_msrs_not_valid())
1039                 return -ENODEV;
1040
1041         pr_info("Intel P-state driver initializing.\n");
1042
1043         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
1044         if (!all_cpu_data)
1045                 return -ENOMEM;
1046
1047         rc = cpufreq_register_driver(&intel_pstate_driver);
1048         if (rc)
1049                 goto out;
1050
1051         intel_pstate_debug_expose_params();
1052         intel_pstate_sysfs_expose_params();
1053
1054         return rc;
1055 out:
1056         get_online_cpus();
1057         for_each_online_cpu(cpu) {
1058                 if (all_cpu_data[cpu]) {
1059                         del_timer_sync(&all_cpu_data[cpu]->timer);
1060                         kfree(all_cpu_data[cpu]);
1061                 }
1062         }
1063
1064         put_online_cpus();
1065         vfree(all_cpu_data);
1066         return -ENODEV;
1067 }
1068 device_initcall(intel_pstate_init);
1069
1070 static int __init intel_pstate_setup(char *str)
1071 {
1072         if (!str)
1073                 return -EINVAL;
1074
1075         if (!strcmp(str, "disable"))
1076                 no_load = 1;
1077         return 0;
1078 }
1079 early_param("intel_pstate", intel_pstate_setup);
1080
1081 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1082 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1083 MODULE_LICENSE("GPL");