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 6
  44 #define int_tofp(X) ((int64_t)(X) << FRAC_BITS)
  45 #define fp_toint(X) ((X) >> FRAC_BITS)
  46 #define FP_ROUNDUP(X) ((X) += 1 << FRAC_BITS)
  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         unsigned long long tsc;
  63         int freq;
  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         u64     prev_aperf;
 102         u64     prev_mperf;
 103         unsigned long long prev_tsc;
 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  = int_tofp(setpoint) - int_tofp(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
 204         return (signed int)fp_toint(result);
 205 }
 206
 207 static inline void intel_pstate_busy_pid_reset(struct cpudata *cpu)
 208 {
 209         pid_p_gain_set(&cpu->pid, pid_params.p_gain_pct);
 210         pid_d_gain_set(&cpu->pid, pid_params.d_gain_pct);
 211         pid_i_gain_set(&cpu->pid, pid_params.i_gain_pct);
 212
 213         pid_reset(&cpu->pid,
 214                 pid_params.setpoint,
 215                 100,
 216                 pid_params.deadband,
 217                 0);
 218 }
 219
 220 static inline void intel_pstate_reset_all_pid(void)
 221 {
 222         unsigned int cpu;
 223         for_each_online_cpu(cpu) {
 224                 if (all_cpu_data[cpu])
 225                         intel_pstate_busy_pid_reset(all_cpu_data[cpu]);
 226         }
 227 }
 228
 229 /************************** debugfs begin ************************/
 230 static int pid_param_set(void *data, u64 val)
 231 {
 232         *(u32 *)data = val;
 233         intel_pstate_reset_all_pid();
 234         return 0;
 235 }
 236 static int pid_param_get(void *data, u64 *val)
 237 {
 238         *val = *(u32 *)data;
 239         return 0;
 240 }
 241 DEFINE_SIMPLE_ATTRIBUTE(fops_pid_param, pid_param_get,
 242                         pid_param_set, "%llu\n");
 243
 244 struct pid_param {
 245         char *name;
 246         void *value;
 247 };
 248
 249 static struct pid_param pid_files[] = {
 250         {"sample_rate_ms", &pid_params.sample_rate_ms},
 251         {"d_gain_pct", &pid_params.d_gain_pct},
 252         {"i_gain_pct", &pid_params.i_gain_pct},
 253         {"deadband", &pid_params.deadband},
 254         {"setpoint", &pid_params.setpoint},
 255         {"p_gain_pct", &pid_params.p_gain_pct},
 256         {NULL, NULL}
 257 };
 258
 259 static struct dentry *debugfs_parent;
 260 static void intel_pstate_debug_expose_params(void)
 261 {
 262         int i = 0;
 263
 264         debugfs_parent = debugfs_create_dir("pstate_snb", NULL);
 265         if (IS_ERR_OR_NULL(debugfs_parent))
 266                 return;
 267         while (pid_files[i].name) {
 268                 debugfs_create_file(pid_files[i].name, 0660,
 269                                 debugfs_parent, pid_files[i].value,
 270                                 &fops_pid_param);
 271                 i++;
 272         }
 273 }
 274
 275 /************************** debugfs end ************************/
 276
 277 /************************** sysfs begin ************************/
 278 #define show_one(file_name, object)                                     \
 279         static ssize_t show_##file_name                                 \
 280         (struct kobject *kobj, struct attribute *attr, char *buf)       \
 281         {                                                               \
 282                 return sprintf(buf, "%u\n", limits.object);             \
 283         }
 284
 285 static ssize_t store_no_turbo(struct kobject *a, struct attribute *b,
 286                                 const char *buf, size_t count)
 287 {
 288         unsigned int input;
 289         int ret;
 290         ret = sscanf(buf, "%u", &input);
 291         if (ret != 1)
 292                 return -EINVAL;
 293         limits.no_turbo = clamp_t(int, input, 0 , 1);
 294
 295         return count;
 296 }
 297
 298 static ssize_t store_max_perf_pct(struct kobject *a, struct attribute *b,
 299                                 const char *buf, size_t count)
 300 {
 301         unsigned int input;
 302         int ret;
 303         ret = sscanf(buf, "%u", &input);
 304         if (ret != 1)
 305                 return -EINVAL;
 306
 307         limits.max_sysfs_pct = clamp_t(int, input, 0 , 100);
 308         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 309         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 310         return count;
 311 }
 312
 313 static ssize_t store_min_perf_pct(struct kobject *a, struct attribute *b,
 314                                 const char *buf, size_t count)
 315 {
 316         unsigned int input;
 317         int ret;
 318         ret = sscanf(buf, "%u", &input);
 319         if (ret != 1)
 320                 return -EINVAL;
 321         limits.min_perf_pct = clamp_t(int, input, 0 , 100);
 322         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 323
 324         return count;
 325 }
 326
 327 show_one(no_turbo, no_turbo);
 328 show_one(max_perf_pct, max_perf_pct);
 329 show_one(min_perf_pct, min_perf_pct);
 330
 331 define_one_global_rw(no_turbo);
 332 define_one_global_rw(max_perf_pct);
 333 define_one_global_rw(min_perf_pct);
 334
 335 static struct attribute *intel_pstate_attributes[] = {
 336         &no_turbo.attr,
 337         &max_perf_pct.attr,
 338         &min_perf_pct.attr,
 339         NULL
 340 };
 341
 342 static struct attribute_group intel_pstate_attr_group = {
 343         .attrs = intel_pstate_attributes,
 344 };
 345 static struct kobject *intel_pstate_kobject;
 346
 347 static void intel_pstate_sysfs_expose_params(void)
 348 {
 349         int rc;
 350
 351         intel_pstate_kobject = kobject_create_and_add("intel_pstate",
 352                                                 &cpu_subsys.dev_root->kobj);
 353         BUG_ON(!intel_pstate_kobject);
 354         rc = sysfs_create_group(intel_pstate_kobject,
 355                                 &intel_pstate_attr_group);
 356         BUG_ON(rc);
 357 }
 358
 359 /************************** sysfs end ************************/
 360 static int byt_get_min_pstate(void)
 361 {
 362         u64 value;
 363         rdmsrl(BYT_RATIOS, value);
 364         return (value >> 8) & 0x3F;
 365 }
 366
 367 static int byt_get_max_pstate(void)
 368 {
 369         u64 value;
 370         rdmsrl(BYT_RATIOS, value);
 371         return (value >> 16) & 0x3F;
 372 }
 373
 374 static int byt_get_turbo_pstate(void)
 375 {
 376         u64 value;
 377         rdmsrl(BYT_TURBO_RATIOS, value);
 378         return value & 0x3F;
 379 }
 380
 381 static void byt_set_pstate(struct cpudata *cpudata, int pstate)
 382 {
 383         u64 val;
 384         int32_t vid_fp;
 385         u32 vid;
 386
 387         val = pstate << 8;
 388         if (limits.no_turbo)
 389                 val |= (u64)1 << 32;
 390
 391         vid_fp = cpudata->vid.min + mul_fp(
 392                 int_tofp(pstate - cpudata->pstate.min_pstate),
 393                 cpudata->vid.ratio);
 394
 395         vid_fp = clamp_t(int32_t, vid_fp, cpudata->vid.min, cpudata->vid.max);
 396         vid = fp_toint(vid_fp);
 397
 398         if (pstate > cpudata->pstate.max_pstate)
 399                 vid = cpudata->vid.turbo;
 400
 401         val |= vid;
 402
 403         wrmsrl(MSR_IA32_PERF_CTL, val);
 404 }
 405
 406 static void byt_get_vid(struct cpudata *cpudata)
 407 {
 408         u64 value;
 409
 410
 411         rdmsrl(BYT_VIDS, value);
 412         cpudata->vid.min = int_tofp((value >> 8) & 0x3f);
 413         cpudata->vid.max = int_tofp((value >> 16) & 0x3f);
 414         cpudata->vid.ratio = div_fp(
 415                 cpudata->vid.max - cpudata->vid.min,
 416                 int_tofp(cpudata->pstate.max_pstate -
 417                         cpudata->pstate.min_pstate));
 418
 419         rdmsrl(BYT_TURBO_VIDS, value);
 420         cpudata->vid.turbo = value & 0x7f;
 421 }
 422
 423
 424 static int core_get_min_pstate(void)
 425 {
 426         u64 value;
 427         rdmsrl(MSR_PLATFORM_INFO, value);
 428         return (value >> 40) & 0xFF;
 429 }
 430
 431 static int core_get_max_pstate(void)
 432 {
 433         u64 value;
 434         rdmsrl(MSR_PLATFORM_INFO, value);
 435         return (value >> 8) & 0xFF;
 436 }
 437
 438 static int core_get_turbo_pstate(void)
 439 {
 440         u64 value;
 441         int nont, ret;
 442         rdmsrl(MSR_NHM_TURBO_RATIO_LIMIT, value);
 443         nont = core_get_max_pstate();
 444         ret = ((value) & 255);
 445         if (ret <= nont)
 446                 ret = nont;
 447         return ret;
 448 }
 449
 450 static void core_set_pstate(struct cpudata *cpudata, int pstate)
 451 {
 452         u64 val;
 453
 454         val = pstate << 8;
 455         if (limits.no_turbo)
 456                 val |= (u64)1 << 32;
 457
 458         wrmsrl_on_cpu(cpudata->cpu, MSR_IA32_PERF_CTL, val);
 459 }
 460
 461 static struct cpu_defaults core_params = {
 462         .pid_policy = {
 463                 .sample_rate_ms = 10,
 464                 .deadband = 0,
 465                 .setpoint = 97,
 466                 .p_gain_pct = 20,
 467                 .d_gain_pct = 0,
 468                 .i_gain_pct = 0,
 469         },
 470         .funcs = {
 471                 .get_max = core_get_max_pstate,
 472                 .get_min = core_get_min_pstate,
 473                 .get_turbo = core_get_turbo_pstate,
 474                 .set = core_set_pstate,
 475         },
 476 };
 477
 478 static struct cpu_defaults byt_params = {
 479         .pid_policy = {
 480                 .sample_rate_ms = 10,
 481                 .deadband = 0,
 482                 .setpoint = 97,
 483                 .p_gain_pct = 14,
 484                 .d_gain_pct = 0,
 485                 .i_gain_pct = 4,
 486         },
 487         .funcs = {
 488                 .get_max = byt_get_max_pstate,
 489                 .get_min = byt_get_min_pstate,
 490                 .get_turbo = byt_get_turbo_pstate,
 491                 .set = byt_set_pstate,
 492                 .get_vid = byt_get_vid,
 493         },
 494 };
 495
 496
 497 static void intel_pstate_get_min_max(struct cpudata *cpu, int *min, int *max)
 498 {
 499         int max_perf = cpu->pstate.turbo_pstate;
 500         int max_perf_adj;
 501         int min_perf;
 502         if (limits.no_turbo)
 503                 max_perf = cpu->pstate.max_pstate;
 504
 505         max_perf_adj = fp_toint(mul_fp(int_tofp(max_perf), limits.max_perf));
 506         *max = clamp_t(int, max_perf_adj,
 507                         cpu->pstate.min_pstate, cpu->pstate.turbo_pstate);
 508
 509         min_perf = fp_toint(mul_fp(int_tofp(max_perf), limits.min_perf));
 510         *min = clamp_t(int, min_perf,
 511                         cpu->pstate.min_pstate, max_perf);
 512 }
 513
 514 static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
 515 {
 516         int max_perf, min_perf;
 517
 518         intel_pstate_get_min_max(cpu, &min_perf, &max_perf);
 519
 520         pstate = clamp_t(int, pstate, min_perf, max_perf);
 521
 522         if (pstate == cpu->pstate.current_pstate)
 523                 return;
 524
 525         trace_cpu_frequency(pstate * 100000, cpu->cpu);
 526
 527         cpu->pstate.current_pstate = pstate;
 528
 529         pstate_funcs.set(cpu, pstate);
 530 }
 531
 532 static inline void intel_pstate_pstate_increase(struct cpudata *cpu, int steps)
 533 {
 534         int target;
 535         target = cpu->pstate.current_pstate + steps;
 536
 537         intel_pstate_set_pstate(cpu, target);
 538 }
 539
 540 static inline void intel_pstate_pstate_decrease(struct cpudata *cpu, int steps)
 541 {
 542         int target;
 543         target = cpu->pstate.current_pstate - steps;
 544         intel_pstate_set_pstate(cpu, target);
 545 }
 546
 547 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 548 {
 549         sprintf(cpu->name, "Intel 2nd generation core");
 550
 551         cpu->pstate.min_pstate = pstate_funcs.get_min();
 552         cpu->pstate.max_pstate = pstate_funcs.get_max();
 553         cpu->pstate.turbo_pstate = pstate_funcs.get_turbo();
 554
 555         if (pstate_funcs.get_vid)
 556                 pstate_funcs.get_vid(cpu);
 557
 558         /*
 559          * goto max pstate so we don't slow up boot if we are built-in if we are
 560          * a module we will take care of it during normal operation
 561          */
 562         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 563 }
 564
 565 static inline void intel_pstate_calc_busy(struct cpudata *cpu,
 566                                         struct sample *sample)
 567 {
 568         int32_t core_pct;
 569         int32_t c0_pct;
 570
 571         core_pct = div_fp(int_tofp((sample->aperf)),
 572                         int_tofp((sample->mperf)));
 573         core_pct = mul_fp(core_pct, int_tofp(100));
 574         FP_ROUNDUP(core_pct);
 575
 576         c0_pct = div_fp(int_tofp(sample->mperf), int_tofp(sample->tsc));
 577
 578         sample->freq = fp_toint(
 579                 mul_fp(int_tofp(cpu->pstate.max_pstate * 1000), core_pct));
 580
 581         sample->core_pct_busy = mul_fp(core_pct, c0_pct);
 582 }
 583
 584 static inline void intel_pstate_sample(struct cpudata *cpu)
 585 {
 586         u64 aperf, mperf;
 587         unsigned long long tsc;
 588
 589         rdmsrl(MSR_IA32_APERF, aperf);
 590         rdmsrl(MSR_IA32_MPERF, mperf);
 591         tsc = native_read_tsc();
 592
 593         aperf = aperf >> FRAC_BITS;
 594         mperf = mperf >> FRAC_BITS;
 595         tsc = tsc >> FRAC_BITS;
 596
 597         cpu->sample.aperf = aperf;
 598         cpu->sample.mperf = mperf;
 599         cpu->sample.tsc = tsc;
 600         cpu->sample.aperf -= cpu->prev_aperf;
 601         cpu->sample.mperf -= cpu->prev_mperf;
 602         cpu->sample.tsc -= cpu->prev_tsc;
 603
 604         intel_pstate_calc_busy(cpu, &cpu->sample);
 605
 606         cpu->prev_aperf = aperf;
 607         cpu->prev_mperf = mperf;
 608         cpu->prev_tsc = tsc;
 609 }
 610
 611 static inline void intel_pstate_set_sample_time(struct cpudata *cpu)
 612 {
 613         int sample_time, delay;
 614
 615         sample_time = pid_params.sample_rate_ms;
 616         delay = msecs_to_jiffies(sample_time);
 617         mod_timer_pinned(&cpu->timer, jiffies + delay);
 618 }
 619
 620 static inline int32_t intel_pstate_get_scaled_busy(struct cpudata *cpu)
 621 {
 622         int32_t core_busy, max_pstate, current_pstate;
 623
 624         core_busy = cpu->sample.core_pct_busy;
 625         max_pstate = int_tofp(cpu->pstate.max_pstate);
 626         current_pstate = int_tofp(cpu->pstate.current_pstate);
 627         core_busy = mul_fp(core_busy, div_fp(max_pstate, current_pstate));
 628         return FP_ROUNDUP(core_busy);
 629 }
 630
 631 static inline void intel_pstate_adjust_busy_pstate(struct cpudata *cpu)
 632 {
 633         int32_t busy_scaled;
 634         struct _pid *pid;
 635         signed int ctl = 0;
 636         int steps;
 637
 638         pid = &cpu->pid;
 639         busy_scaled = intel_pstate_get_scaled_busy(cpu);
 640
 641         ctl = pid_calc(pid, busy_scaled);
 642
 643         steps = abs(ctl);
 644
 645         if (ctl < 0)
 646                 intel_pstate_pstate_increase(cpu, steps);
 647         else
 648                 intel_pstate_pstate_decrease(cpu, steps);
 649 }
 650
 651 static void intel_pstate_timer_func(unsigned long __data)
 652 {
 653         struct cpudata *cpu = (struct cpudata *) __data;
 654         struct sample *sample;
 655
 656         intel_pstate_sample(cpu);
 657
 658         sample = &cpu->sample;
 659
 660         intel_pstate_adjust_busy_pstate(cpu);
 661
 662         trace_pstate_sample(fp_toint(sample->core_pct_busy),
 663                         fp_toint(intel_pstate_get_scaled_busy(cpu)),
 664                         cpu->pstate.current_pstate,
 665                         sample->mperf,
 666                         sample->aperf,
 667                         sample->freq);
 668
 669         intel_pstate_set_sample_time(cpu);
 670 }
 671
 672 #define ICPU(model, policy) \
 673         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_APERFMPERF,\
 674                         (unsigned long)&policy }
 675
 676 static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
 677         ICPU(0x2a, core_params),
 678         ICPU(0x2d, core_params),
 679         ICPU(0x37, byt_params),
 680         ICPU(0x3a, core_params),
 681         ICPU(0x3c, core_params),
 682         ICPU(0x3e, core_params),
 683         ICPU(0x3f, core_params),
 684         ICPU(0x45, core_params),
 685         ICPU(0x46, core_params),
 686         {}
 687 };
 688 MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
 689
 690 static int intel_pstate_init_cpu(unsigned int cpunum)
 691 {
 692
 693         const struct x86_cpu_id *id;
 694         struct cpudata *cpu;
 695
 696         id = x86_match_cpu(intel_pstate_cpu_ids);
 697         if (!id)
 698                 return -ENODEV;
 699
 700         all_cpu_data[cpunum] = kzalloc(sizeof(struct cpudata), GFP_KERNEL);
 701         if (!all_cpu_data[cpunum])
 702                 return -ENOMEM;
 703
 704         cpu = all_cpu_data[cpunum];
 705
 706         intel_pstate_get_cpu_pstates(cpu);
 707         if (!cpu->pstate.current_pstate) {
 708                 all_cpu_data[cpunum] = NULL;
 709                 kfree(cpu);
 710                 return -ENODATA;
 711         }
 712
 713         cpu->cpu = cpunum;
 714
 715         init_timer_deferrable(&cpu->timer);
 716         cpu->timer.function = intel_pstate_timer_func;
 717         cpu->timer.data =
 718                 (unsigned long)cpu;
 719         cpu->timer.expires = jiffies + HZ/100;
 720         intel_pstate_busy_pid_reset(cpu);
 721         intel_pstate_sample(cpu);
 722         intel_pstate_set_pstate(cpu, cpu->pstate.max_pstate);
 723
 724         add_timer_on(&cpu->timer, cpunum);
 725
 726         pr_info("Intel pstate controlling: cpu %d\n", cpunum);
 727
 728         return 0;
 729 }
 730
 731 static unsigned int intel_pstate_get(unsigned int cpu_num)
 732 {
 733         struct sample *sample;
 734         struct cpudata *cpu;
 735
 736         cpu = all_cpu_data[cpu_num];
 737         if (!cpu)
 738                 return 0;
 739         sample = &cpu->sample;
 740         return sample->freq;
 741 }
 742
 743 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 744 {
 745         struct cpudata *cpu;
 746
 747         cpu = all_cpu_data[policy->cpu];
 748
 749         if (!policy->cpuinfo.max_freq)
 750                 return -ENODEV;
 751
 752         if (policy->policy == CPUFREQ_POLICY_PERFORMANCE) {
 753                 limits.min_perf_pct = 100;
 754                 limits.min_perf = int_tofp(1);
 755                 limits.max_perf_pct = 100;
 756                 limits.max_perf = int_tofp(1);
 757                 limits.no_turbo = 0;
 758                 return 0;
 759         }
 760         limits.min_perf_pct = (policy->min * 100) / policy->cpuinfo.max_freq;
 761         limits.min_perf_pct = clamp_t(int, limits.min_perf_pct, 0 , 100);
 762         limits.min_perf = div_fp(int_tofp(limits.min_perf_pct), int_tofp(100));
 763
 764         limits.max_policy_pct = policy->max * 100 / policy->cpuinfo.max_freq;
 765         limits.max_policy_pct = clamp_t(int, limits.max_policy_pct, 0 , 100);
 766         limits.max_perf_pct = min(limits.max_policy_pct, limits.max_sysfs_pct);
 767         limits.max_perf = div_fp(int_tofp(limits.max_perf_pct), int_tofp(100));
 768
 769         return 0;
 770 }
 771
 772 static int intel_pstate_verify_policy(struct cpufreq_policy *policy)
 773 {
 774         cpufreq_verify_within_cpu_limits(policy);
 775
 776         if ((policy->policy != CPUFREQ_POLICY_POWERSAVE) &&
 777                 (policy->policy != CPUFREQ_POLICY_PERFORMANCE))
 778                 return -EINVAL;
 779
 780         return 0;
 781 }
 782
 783 static void intel_pstate_stop_cpu(struct cpufreq_policy *policy)
 784 {
 785         int cpu_num = policy->cpu;
 786         struct cpudata *cpu = all_cpu_data[cpu_num];
 787
 788         pr_info("intel_pstate CPU %d exiting\n", cpu_num);
 789
 790         del_timer_sync(&all_cpu_data[cpu_num]->timer);
 791         intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
 792         kfree(all_cpu_data[cpu_num]);
 793         all_cpu_data[cpu_num] = NULL;
 794 }
 795
 796 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 797 {
 798         struct cpudata *cpu;
 799         int rc;
 800
 801         rc = intel_pstate_init_cpu(policy->cpu);
 802         if (rc)
 803                 return rc;
 804
 805         cpu = all_cpu_data[policy->cpu];
 806
 807         if (!limits.no_turbo &&
 808                 limits.min_perf_pct == 100 && limits.max_perf_pct == 100)
 809                 policy->policy = CPUFREQ_POLICY_PERFORMANCE;
 810         else
 811                 policy->policy = CPUFREQ_POLICY_POWERSAVE;
 812
 813         policy->min = cpu->pstate.min_pstate * 100000;
 814         policy->max = cpu->pstate.turbo_pstate * 100000;
 815
 816         /* cpuinfo and default policy values */
 817         policy->cpuinfo.min_freq = cpu->pstate.min_pstate * 100000;
 818         policy->cpuinfo.max_freq = cpu->pstate.turbo_pstate * 100000;
 819         policy->cpuinfo.transition_latency = CPUFREQ_ETERNAL;
 820         cpumask_set_cpu(policy->cpu, policy->cpus);
 821
 822         return 0;
 823 }
 824
 825 static struct cpufreq_driver intel_pstate_driver = {
 826         .flags          = CPUFREQ_CONST_LOOPS,
 827         .verify         = intel_pstate_verify_policy,
 828         .setpolicy      = intel_pstate_set_policy,
 829         .get            = intel_pstate_get,
 830         .init           = intel_pstate_cpu_init,
 831         .stop_cpu       = intel_pstate_stop_cpu,
 832         .name           = "intel_pstate",
 833 };
 834
 835 static int __initdata no_load;
 836
 837 static int intel_pstate_msrs_not_valid(void)
 838 {
 839         /* Check that all the msr's we are using are valid. */
 840         u64 aperf, mperf, tmp;
 841
 842         rdmsrl(MSR_IA32_APERF, aperf);
 843         rdmsrl(MSR_IA32_MPERF, mperf);
 844
 845         if (!pstate_funcs.get_max() ||
 846                 !pstate_funcs.get_min() ||
 847                 !pstate_funcs.get_turbo())
 848                 return -ENODEV;
 849
 850         rdmsrl(MSR_IA32_APERF, tmp);
 851         if (!(tmp - aperf))
 852                 return -ENODEV;
 853
 854         rdmsrl(MSR_IA32_MPERF, tmp);
 855         if (!(tmp - mperf))
 856                 return -ENODEV;
 857
 858         return 0;
 859 }
 860
 861 static void copy_pid_params(struct pstate_adjust_policy *policy)
 862 {
 863         pid_params.sample_rate_ms = policy->sample_rate_ms;
 864         pid_params.p_gain_pct = policy->p_gain_pct;
 865         pid_params.i_gain_pct = policy->i_gain_pct;
 866         pid_params.d_gain_pct = policy->d_gain_pct;
 867         pid_params.deadband = policy->deadband;
 868         pid_params.setpoint = policy->setpoint;
 869 }
 870
 871 static void copy_cpu_funcs(struct pstate_funcs *funcs)
 872 {
 873         pstate_funcs.get_max   = funcs->get_max;
 874         pstate_funcs.get_min   = funcs->get_min;
 875         pstate_funcs.get_turbo = funcs->get_turbo;
 876         pstate_funcs.set       = funcs->set;
 877         pstate_funcs.get_vid   = funcs->get_vid;
 878 }
 879
 880 #if IS_ENABLED(CONFIG_ACPI)
 881 #include <acpi/processor.h>
 882
 883 static bool intel_pstate_no_acpi_pss(void)
 884 {
 885         int i;
 886
 887         for_each_possible_cpu(i) {
 888                 acpi_status status;
 889                 union acpi_object *pss;
 890                 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 891                 struct acpi_processor *pr = per_cpu(processors, i);
 892
 893                 if (!pr)
 894                         continue;
 895
 896                 status = acpi_evaluate_object(pr->handle, "_PSS", NULL, &buffer);
 897                 if (ACPI_FAILURE(status))
 898                         continue;
 899
 900                 pss = buffer.pointer;
 901                 if (pss && pss->type == ACPI_TYPE_PACKAGE) {
 902                         kfree(pss);
 903                         return false;
 904                 }
 905
 906                 kfree(pss);
 907         }
 908
 909         return true;
 910 }
 911
 912 struct hw_vendor_info {
 913         u16  valid;
 914         char oem_id[ACPI_OEM_ID_SIZE];
 915         char oem_table_id[ACPI_OEM_TABLE_ID_SIZE];
 916 };
 917
 918 /* Hardware vendor-specific info that has its own power management modes */
 919 static struct hw_vendor_info vendor_info[] = {
 920         {1, "HP    ", "ProLiant"},
 921         {0, "", ""},
 922 };
 923
 924 static bool intel_pstate_platform_pwr_mgmt_exists(void)
 925 {
 926         struct acpi_table_header hdr;
 927         struct hw_vendor_info *v_info;
 928
 929         if (acpi_disabled
 930             || ACPI_FAILURE(acpi_get_table_header(ACPI_SIG_FADT, 0, &hdr)))
 931                 return false;
 932
 933         for (v_info = vendor_info; v_info->valid; v_info++) {
 934                 if (!strncmp(hdr.oem_id, v_info->oem_id, ACPI_OEM_ID_SIZE)
 935                     && !strncmp(hdr.oem_table_id, v_info->oem_table_id, ACPI_OEM_TABLE_ID_SIZE)
 936                     && intel_pstate_no_acpi_pss())
 937                         return true;
 938         }
 939
 940         return false;
 941 }
 942 #else /* CONFIG_ACPI not enabled */
 943 static inline bool intel_pstate_platform_pwr_mgmt_exists(void) { return false; }
 944 #endif /* CONFIG_ACPI */
 945
 946 static int __init intel_pstate_init(void)
 947 {
 948         int cpu, rc = 0;
 949         const struct x86_cpu_id *id;
 950         struct cpu_defaults *cpu_info;
 951
 952         if (no_load)
 953                 return -ENODEV;
 954
 955         id = x86_match_cpu(intel_pstate_cpu_ids);
 956         if (!id)
 957                 return -ENODEV;
 958
 959         /*
 960          * The Intel pstate driver will be ignored if the platform
 961          * firmware has its own power management modes.
 962          */
 963         if (intel_pstate_platform_pwr_mgmt_exists())
 964                 return -ENODEV;
 965
 966         cpu_info = (struct cpu_defaults *)id->driver_data;
 967
 968         copy_pid_params(&cpu_info->pid_policy);
 969         copy_cpu_funcs(&cpu_info->funcs);
 970
 971         if (intel_pstate_msrs_not_valid())
 972                 return -ENODEV;
 973
 974         pr_info("Intel P-state driver initializing.\n");
 975
 976         all_cpu_data = vzalloc(sizeof(void *) * num_possible_cpus());
 977         if (!all_cpu_data)
 978                 return -ENOMEM;
 979
 980         rc = cpufreq_register_driver(&intel_pstate_driver);
 981         if (rc)
 982                 goto out;
 983
 984         intel_pstate_debug_expose_params();
 985         intel_pstate_sysfs_expose_params();
 986
 987         return rc;
 988 out:
 989         get_online_cpus();
 990         for_each_online_cpu(cpu) {
 991                 if (all_cpu_data[cpu]) {
 992                         del_timer_sync(&all_cpu_data[cpu]->timer);
 993                         kfree(all_cpu_data[cpu]);
 994                 }
 995         }
 996
 997         put_online_cpus();
 998         vfree(all_cpu_data);
 999         return -ENODEV;
1000 }
1001 device_initcall(intel_pstate_init);
1002
1003 static int __init intel_pstate_setup(char *str)
1004 {
1005         if (!str)
1006                 return -EINVAL;
1007
1008         if (!strcmp(str, "disable"))
1009                 no_load = 1;
1010         return 0;
1011 }
1012 early_param("intel_pstate", intel_pstate_setup);
1013
1014 MODULE_AUTHOR("Dirk Brandewie <dirk.j.brandewie@intel.com>");
1015 MODULE_DESCRIPTION("'intel_pstate' - P state driver Intel Core processors");
1016 MODULE_LICENSE("GPL");