phc-intel/phc-intel.c

   1 /*
   2  * acpi-cpufreq.c - ACPI Processor P-States Driver
   3  *
   4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   6  *  Copyright (C) 2002 - 2004 Dominik Brodowski <linux@brodo.de>
   7  *  Copyright (C) 2006       Denis Sadykov <denis.m.sadykov@intel.com>
   8  *
   9  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  10  *
  11  *  This program is free software; you can redistribute it and/or modify
  12  *  it under the terms of the GNU General Public License as published by
  13  *  the Free Software Foundation; either version 2 of the License, or (at
  14  *  your option) any later version.
  15  *
  16  *  This program is distributed in the hope that it will be useful, but
  17  *  WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  *  General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU General Public License along
  22  *  with this program; if not, write to the Free Software Foundation, Inc.,
  23  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
  24  *
  25  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  26  */
  27
  28 /* This file has been patched with Linux PHC: www.linux-phc.org
  29 * Patch version: linux-phc-0.3.2
  30 */
  31
  32 #include <linux/kernel.h>
  33 #include <linux/module.h>
  34 #include <linux/init.h>
  35 #include <linux/smp.h>
  36 #include <linux/sched.h>
  37 #include <linux/compiler.h>
  38 #include <linux/dmi.h>
  39 #include <linux/slab.h>
  40 #include <linux/cpufreq.h>
  41
  42 #include <linux/acpi.h>
  43 #include <linux/io.h>
  44 #include <linux/delay.h>
  45 #include <linux/uaccess.h>
  46
  47 #include <acpi/processor.h>
  48
  49 #include <asm/msr.h>
  50 #include <asm/processor.h>
  51 #include <asm/cpufeature.h>
  52 #include "../drivers/cpufreq/mperf.h"
  53
  54 MODULE_AUTHOR("Paul Diefenbaugh, Dominik Brodowski");
  55 MODULE_DESCRIPTION("ACPI Processor P-States Driver");
  56 MODULE_LICENSE("GPL");
  57
  58 enum {
  59         UNDEFINED_CAPABLE = 0,
  60         SYSTEM_INTEL_MSR_CAPABLE,
  61         SYSTEM_IO_CAPABLE,
  62 };
  63
  64 #define INTEL_MSR_RANGE         (0xffff)
  65 #define INTEL_MSR_VID_MASK      (0x00ff)
  66 #define INTEL_MSR_FID_MASK      (0xff00)
  67 #define INTEL_MSR_FID_SHIFT     (0x8)
  68 #define PHC_VERSION_STRING      "0.3.2:2"
  69
  70 struct acpi_cpufreq_data {
  71         struct acpi_processor_performance *acpi_data;
  72         struct cpufreq_frequency_table *freq_table;
  73         unsigned int resume;
  74         unsigned int cpu_feature;
  75         acpi_integer *original_controls;
  76 };
  77
  78 static DEFINE_PER_CPU(struct acpi_cpufreq_data *, acfreq_data);
  79
  80 /* acpi_perf_data is a pointer to percpu data. */
  81 static struct acpi_processor_performance __percpu *acpi_perf_data;
  82
  83 static struct cpufreq_driver acpi_cpufreq_driver;
  84
  85 static unsigned int acpi_pstate_strict;
  86
  87 static int check_est_cpu(unsigned int cpuid)
  88 {
  89         struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
  90
  91         return cpu_has(cpu, X86_FEATURE_EST);
  92 }
  93
  94 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
  95 {
  96         struct acpi_processor_performance *perf;
  97         int i;
  98
  99         perf = data->acpi_data;
 100
 101         for (i = 0; i < perf->state_count; i++) {
 102                 if (value == perf->states[i].status)
 103                         return data->freq_table[i].frequency;
 104         }
 105         return 0;
 106 }
 107
 108 static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
 109 {
 110         int i;
 111         u32 fid;
 112         struct acpi_processor_performance *perf;
 113
 114         fid = msr & INTEL_MSR_FID_MASK;
 115         perf = data->acpi_data;
 116
 117         for (i = 0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 118                 if (fid == (perf->states[data->freq_table[i].index].status & INTEL_MSR_FID_MASK))
 119                         return data->freq_table[i].frequency;
 120         }
 121         return data->freq_table[0].frequency;
 122 }
 123
 124 static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
 125 {
 126         switch (data->cpu_feature) {
 127         case SYSTEM_INTEL_MSR_CAPABLE:
 128                 return extract_msr(val, data);
 129         case SYSTEM_IO_CAPABLE:
 130                 return extract_io(val, data);
 131         default:
 132                 return 0;
 133         }
 134 }
 135
 136 struct msr_addr {
 137         u32 reg;
 138 };
 139
 140 struct io_addr {
 141         u16 port;
 142         u8 bit_width;
 143 };
 144
 145 struct drv_cmd {
 146         unsigned int type;
 147         const struct cpumask *mask;
 148         union {
 149                 struct msr_addr msr;
 150                 struct io_addr io;
 151         } addr;
 152         u32 val;
 153 };
 154
 155 /* Called via smp_call_function_single(), on the target CPU */
 156 static void do_drv_read(void *_cmd)
 157 {
 158         struct drv_cmd *cmd = _cmd;
 159         u32 h;
 160
 161         switch (cmd->type) {
 162         case SYSTEM_INTEL_MSR_CAPABLE:
 163                 rdmsr(cmd->addr.msr.reg, cmd->val, h);
 164                 break;
 165         case SYSTEM_IO_CAPABLE:
 166                 acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
 167                                 &cmd->val,
 168                                 (u32)cmd->addr.io.bit_width);
 169                 break;
 170         default:
 171                 break;
 172         }
 173 }
 174
 175 /* Called via smp_call_function_many(), on the target CPUs */
 176 static void do_drv_write(void *_cmd)
 177 {
 178         struct drv_cmd *cmd = _cmd;
 179         u32 lo, hi;
 180
 181         switch (cmd->type) {
 182         case SYSTEM_INTEL_MSR_CAPABLE:
 183                 rdmsr(cmd->addr.msr.reg, lo, hi);
 184                 lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
 185                 wrmsr(cmd->addr.msr.reg, lo, hi);
 186                 break;
 187         case SYSTEM_IO_CAPABLE:
 188                 acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
 189                                 cmd->val,
 190                                 (u32)cmd->addr.io.bit_width);
 191                 break;
 192         default:
 193                 break;
 194         }
 195 }
 196
 197 static void drv_read(struct drv_cmd *cmd)
 198 {
 199         int err;
 200         cmd->val = 0;
 201
 202         err = smp_call_function_any(cmd->mask, do_drv_read, cmd, 1);
 203         WARN_ON_ONCE(err);      /* smp_call_function_any() was buggy? */
 204 }
 205
 206 static void drv_write(struct drv_cmd *cmd)
 207 {
 208         int this_cpu;
 209
 210         this_cpu = get_cpu();
 211         if (cpumask_test_cpu(this_cpu, cmd->mask))
 212                 do_drv_write(cmd);
 213         smp_call_function_many(cmd->mask, do_drv_write, cmd, 1);
 214         put_cpu();
 215 }
 216
 217 static u32 get_cur_val(const struct cpumask *mask)
 218 {
 219         struct acpi_processor_performance *perf;
 220         struct drv_cmd cmd;
 221
 222         if (unlikely(cpumask_empty(mask)))
 223                 return 0;
 224
 225         switch (per_cpu(acfreq_data, cpumask_first(mask))->cpu_feature) {
 226         case SYSTEM_INTEL_MSR_CAPABLE:
 227                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 228                 cmd.addr.msr.reg = MSR_IA32_PERF_STATUS;
 229                 break;
 230         case SYSTEM_IO_CAPABLE:
 231                 cmd.type = SYSTEM_IO_CAPABLE;
 232                 perf = per_cpu(acfreq_data, cpumask_first(mask))->acpi_data;
 233                 cmd.addr.io.port = perf->control_register.address;
 234                 cmd.addr.io.bit_width = perf->control_register.bit_width;
 235                 break;
 236         default:
 237                 return 0;
 238         }
 239
 240         cmd.mask = mask;
 241         drv_read(&cmd);
 242
 243         pr_debug("get_cur_val = %u\n", cmd.val);
 244
 245         return cmd.val;
 246 }
 247
 248 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 249 {
 250         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
 251         unsigned int freq;
 252         unsigned int cached_freq;
 253
 254         pr_debug("get_cur_freq_on_cpu (%d)\n", cpu);
 255
 256         if (unlikely(data == NULL ||
 257                      data->acpi_data == NULL || data->freq_table == NULL)) {
 258                 return 0;
 259         }
 260
 261         cached_freq = data->freq_table[data->acpi_data->state].frequency;
 262         freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
 263         if (freq != cached_freq) {
 264                 /*
 265                  * The dreaded BIOS frequency change behind our back.
 266                  * Force set the frequency on next target call.
 267                  */
 268                 data->resume = 1;
 269         }
 270
 271         pr_debug("cur freq = %u\n", freq);
 272
 273         return freq;
 274 }
 275
 276 static unsigned int check_freqs(const struct cpumask *mask, unsigned int freq,
 277                                 struct acpi_cpufreq_data *data)
 278 {
 279         unsigned int cur_freq;
 280         unsigned int i;
 281
 282         for (i = 0; i < 100; i++) {
 283                 cur_freq = extract_freq(get_cur_val(mask), data);
 284                 if (cur_freq == freq)
 285                         return 1;
 286                 udelay(10);
 287         }
 288         return 0;
 289 }
 290
 291 static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 292                                unsigned int target_freq, unsigned int relation)
 293 {
 294         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 295         struct acpi_processor_performance *perf;
 296         struct cpufreq_freqs freqs;
 297         struct drv_cmd cmd;
 298         unsigned int next_state = 0; /* Index into freq_table */
 299         unsigned int next_perf_state = 0; /* Index into perf table */
 300         unsigned int i;
 301         int result = 0;
 302
 303         pr_debug("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 304
 305         if (unlikely(data == NULL ||
 306              data->acpi_data == NULL || data->freq_table == NULL)) {
 307                 return -ENODEV;
 308         }
 309
 310         perf = data->acpi_data;
 311         result = cpufreq_frequency_table_target(policy,
 312                                                 data->freq_table,
 313                                                 target_freq,
 314                                                 relation, &next_state);
 315         if (unlikely(result)) {
 316                 result = -ENODEV;
 317                 goto out;
 318         }
 319
 320         next_perf_state = data->freq_table[next_state].index;
 321         if (perf->state == next_perf_state) {
 322                 if (unlikely(data->resume)) {
 323                         pr_debug("Called after resume, resetting to P%d\n",
 324                                 next_perf_state);
 325                         data->resume = 0;
 326                 } else {
 327                         pr_debug("Already at target state (P%d)\n",
 328                                 next_perf_state);
 329                         goto out;
 330                 }
 331         }
 332
 333         switch (data->cpu_feature) {
 334         case SYSTEM_INTEL_MSR_CAPABLE:
 335                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 336                 cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
 337                 cmd.val = (u32) perf->states[next_perf_state].control;
 338                 break;
 339         case SYSTEM_IO_CAPABLE:
 340                 cmd.type = SYSTEM_IO_CAPABLE;
 341                 cmd.addr.io.port = perf->control_register.address;
 342                 cmd.addr.io.bit_width = perf->control_register.bit_width;
 343                 cmd.val = (u32) perf->states[next_perf_state].control;
 344                 break;
 345         default:
 346                 result = -ENODEV;
 347                 goto out;
 348         }
 349
 350         /* cpufreq holds the hotplug lock, so we are safe from here on */
 351         if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
 352                 cmd.mask = policy->cpus;
 353         else
 354                 cmd.mask = cpumask_of(policy->cpu);
 355
 356         freqs.old = perf->states[perf->state].core_frequency * 1000;
 357         freqs.new = data->freq_table[next_state].frequency;
 358         for_each_cpu(i, policy->cpus) {
 359                 freqs.cpu = i;
 360                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 361         }
 362
 363         drv_write(&cmd);
 364
 365         if (acpi_pstate_strict) {
 366                 if (!check_freqs(cmd.mask, freqs.new, data)) {
 367                         pr_debug("acpi_cpufreq_target failed (%d)\n",
 368                                 policy->cpu);
 369                         result = -EAGAIN;
 370                         goto out;
 371                 }
 372         }
 373
 374         for_each_cpu(i, policy->cpus) {
 375                 freqs.cpu = i;
 376                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 377         }
 378         perf->state = next_perf_state;
 379
 380 out:
 381         return result;
 382 }
 383
 384 static int acpi_cpufreq_verify(struct cpufreq_policy *policy)
 385 {
 386         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 387
 388         pr_debug("acpi_cpufreq_verify\n");
 389
 390         return cpufreq_frequency_table_verify(policy, data->freq_table);
 391 }
 392
 393 static unsigned long
 394 acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
 395 {
 396         struct acpi_processor_performance *perf = data->acpi_data;
 397
 398         if (cpu_khz) {
 399                 /* search the closest match to cpu_khz */
 400                 unsigned int i;
 401                 unsigned long freq;
 402                 unsigned long freqn = perf->states[0].core_frequency * 1000;
 403
 404                 for (i = 0; i < (perf->state_count-1); i++) {
 405                         freq = freqn;
 406                         freqn = perf->states[i+1].core_frequency * 1000;
 407                         if ((2 * cpu_khz) > (freqn + freq)) {
 408                                 perf->state = i;
 409                                 return freq;
 410                         }
 411                 }
 412                 perf->state = perf->state_count-1;
 413                 return freqn;
 414         } else {
 415                 /* assume CPU is at P0... */
 416                 perf->state = 0;
 417                 return perf->states[0].core_frequency * 1000;
 418         }
 419 }
 420
 421 static void free_acpi_perf_data(void)
 422 {
 423         unsigned int i;
 424
 425         /* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
 426         for_each_possible_cpu(i)
 427                 free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
 428                                  ->shared_cpu_map);
 429         free_percpu(acpi_perf_data);
 430 }
 431
 432 /*
 433  * acpi_cpufreq_early_init - initialize ACPI P-States library
 434  *
 435  * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
 436  * in order to determine correct frequency and voltage pairings. We can
 437  * do _PDC and _PSD and find out the processor dependency for the
 438  * actual init that will happen later...
 439  */
 440 static int __init acpi_cpufreq_early_init(void)
 441 {
 442         unsigned int i;
 443         pr_debug("acpi_cpufreq_early_init\n");
 444
 445         acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
 446         if (!acpi_perf_data) {
 447                 pr_debug("Memory allocation error for acpi_perf_data.\n");
 448                 return -ENOMEM;
 449         }
 450         for_each_possible_cpu(i) {
 451                 if (!zalloc_cpumask_var_node(
 452                         &per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
 453                         GFP_KERNEL, cpu_to_node(i))) {
 454
 455                         /* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
 456                         free_acpi_perf_data();
 457                         return -ENOMEM;
 458                 }
 459         }
 460
 461         /* Do initialization in ACPI core */
 462         acpi_processor_preregister_performance(acpi_perf_data);
 463         return 0;
 464 }
 465
 466 #ifdef CONFIG_SMP
 467 /*
 468  * Some BIOSes do SW_ANY coordination internally, either set it up in hw
 469  * or do it in BIOS firmware and won't inform about it to OS. If not
 470  * detected, this has a side effect of making CPU run at a different speed
 471  * than OS intended it to run at. Detect it and handle it cleanly.
 472  */
 473 static int bios_with_sw_any_bug;
 474
 475 static int sw_any_bug_found(const struct dmi_system_id *d)
 476 {
 477         bios_with_sw_any_bug = 1;
 478         return 0;
 479 }
 480
 481 static const struct dmi_system_id sw_any_bug_dmi_table[] = {
 482         {
 483                 .callback = sw_any_bug_found,
 484                 .ident = "Supermicro Server X6DLP",
 485                 .matches = {
 486                         DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
 487                         DMI_MATCH(DMI_BIOS_VERSION, "080010"),
 488                         DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
 489                 },
 490         },
 491         { }
 492 };
 493
 494 static int acpi_cpufreq_blacklist(struct cpuinfo_x86 *c)
 495 {
 496         /* Intel Xeon Processor 7100 Series Specification Update
 497          * http://www.intel.com/Assets/PDF/specupdate/314554.pdf
 498          * AL30: A Machine Check Exception (MCE) Occurring during an
 499          * Enhanced Intel SpeedStep Technology Ratio Change May Cause
 500          * Both Processor Cores to Lock Up. */
 501         if (c->x86_vendor == X86_VENDOR_INTEL) {
 502                 if ((c->x86 == 15) &&
 503                     (c->x86_model == 6) &&
 504                     (c->x86_mask == 8)) {
 505                         printk(KERN_INFO "acpi-cpufreq: Intel(R) "
 506                             "Xeon(R) 7100 Errata AL30, processors may "
 507                             "lock up on frequency changes: disabling "
 508                             "acpi-cpufreq.\n");
 509                         return -ENODEV;
 510                     }
 511                 }
 512         return 0;
 513 }
 514 #endif
 515
 516 static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 517 {
 518         unsigned int i;
 519         unsigned int valid_states = 0;
 520         unsigned int cpu = policy->cpu;
 521         struct acpi_cpufreq_data *data;
 522         unsigned int result = 0;
 523         struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
 524         struct acpi_processor_performance *perf;
 525 #ifdef CONFIG_SMP
 526         static int blacklisted;
 527 #endif
 528
 529         pr_debug("acpi_cpufreq_cpu_init\n");
 530
 531 #ifdef CONFIG_SMP
 532         if (blacklisted)
 533                 return blacklisted;
 534         blacklisted = acpi_cpufreq_blacklist(c);
 535         if (blacklisted)
 536                 return blacklisted;
 537 #endif
 538
 539         data = kzalloc(sizeof(struct acpi_cpufreq_data), GFP_KERNEL);
 540         if (!data)
 541                 return -ENOMEM;
 542
 543         data->acpi_data = per_cpu_ptr(acpi_perf_data, cpu);
 544         per_cpu(acfreq_data, cpu) = data;
 545
 546         if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
 547                 acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
 548
 549         result = acpi_processor_register_performance(data->acpi_data, cpu);
 550         if (result)
 551                 goto err_free;
 552
 553         perf = data->acpi_data;
 554         policy->shared_type = perf->shared_type;
 555
 556         /*
 557          * Will let policy->cpus know about dependency only when software
 558          * coordination is required.
 559          */
 560         if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
 561             policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
 562                 cpumask_copy(policy->cpus, perf->shared_cpu_map);
 563         }
 564         cpumask_copy(policy->related_cpus, perf->shared_cpu_map);
 565
 566 #ifdef CONFIG_SMP
 567         dmi_check_system(sw_any_bug_dmi_table);
 568         if (bios_with_sw_any_bug && cpumask_weight(policy->cpus) == 1) {
 569                 policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
 570                 cpumask_copy(policy->cpus, cpu_core_mask(cpu));
 571         }
 572 #endif
 573
 574         /* capability check */
 575         if (perf->state_count <= 1) {
 576                 pr_debug("No P-States\n");
 577                 result = -ENODEV;
 578                 goto err_unreg;
 579         }
 580
 581         if (perf->control_register.space_id != perf->status_register.space_id) {
 582                 result = -ENODEV;
 583                 goto err_unreg;
 584         }
 585
 586         switch (perf->control_register.space_id) {
 587         case ACPI_ADR_SPACE_SYSTEM_IO:
 588                 pr_debug("SYSTEM IO addr space\n");
 589                 data->cpu_feature = SYSTEM_IO_CAPABLE;
 590                 break;
 591         case ACPI_ADR_SPACE_FIXED_HARDWARE:
 592                 pr_debug("HARDWARE addr space\n");
 593                 if (!check_est_cpu(cpu)) {
 594                         result = -ENODEV;
 595                         goto err_unreg;
 596                 }
 597                 data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
 598                 break;
 599         default:
 600                 pr_debug("Unknown addr space %d\n",
 601                         (u32) (perf->control_register.space_id));
 602                 result = -ENODEV;
 603                 goto err_unreg;
 604         }
 605
 606         data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) *
 607                     (perf->state_count+1), GFP_KERNEL);
 608         if (!data->freq_table) {
 609                 result = -ENOMEM;
 610                 goto err_unreg;
 611         }
 612
 613         /* detect transition latency */
 614         policy->cpuinfo.transition_latency = 0;
 615         for (i = 0; i < perf->state_count; i++) {
 616                 if ((perf->states[i].transition_latency * 1000) >
 617                     policy->cpuinfo.transition_latency)
 618                         policy->cpuinfo.transition_latency =
 619                             perf->states[i].transition_latency * 1000;
 620         }
 621
 622         /* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
 623         if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
 624             policy->cpuinfo.transition_latency > 20 * 1000) {
 625                 policy->cpuinfo.transition_latency = 20 * 1000;
 626                 printk_once(KERN_INFO
 627                             "P-state transition latency capped at 20 uS\n");
 628         }
 629
 630         /* table init */
 631         for (i = 0; i < perf->state_count; i++) {
 632                 if (i > 0 && perf->states[i].core_frequency >=
 633                     data->freq_table[valid_states-1].frequency / 1000)
 634                         continue;
 635
 636                 data->freq_table[valid_states].index = i;
 637                 data->freq_table[valid_states].frequency =
 638                     perf->states[i].core_frequency * 1000;
 639                 valid_states++;
 640         }
 641         data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
 642         perf->state = 0;
 643
 644         result = cpufreq_frequency_table_cpuinfo(policy, data->freq_table);
 645         if (result)
 646                 goto err_freqfree;
 647
 648         if (perf->states[0].core_frequency * 1000 != policy->cpuinfo.max_freq)
 649                 printk(KERN_WARNING FW_WARN "P-state 0 is not max freq\n");
 650
 651         switch (perf->control_register.space_id) {
 652         case ACPI_ADR_SPACE_SYSTEM_IO:
 653                 /* Current speed is unknown and not detectable by IO port */
 654                 policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
 655                 break;
 656         case ACPI_ADR_SPACE_FIXED_HARDWARE:
 657                 acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
 658                 policy->cur = get_cur_freq_on_cpu(cpu);
 659                 break;
 660         default:
 661                 break;
 662         }
 663
 664         /* notify BIOS that we exist */
 665         acpi_processor_notify_smm(THIS_MODULE);
 666
 667         /* Check for APERF/MPERF support in hardware */
 668         if (cpu_has(c, X86_FEATURE_APERFMPERF))
 669                 acpi_cpufreq_driver.getavg = cpufreq_get_measured_perf;
 670
 671         pr_debug("CPU%u - ACPI performance management activated.\n", cpu);
 672         for (i = 0; i < perf->state_count; i++)
 673                 pr_debug("     %cP%d: %d MHz, %d mW, %d uS\n",
 674                         (i == perf->state ? '*' : ' '), i,
 675                         (u32) perf->states[i].core_frequency,
 676                         (u32) perf->states[i].power,
 677                         (u32) perf->states[i].transition_latency);
 678
 679         cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
 680
 681         /*
 682          * the first call to ->target() should result in us actually
 683          * writing something to the appropriate registers.
 684          */
 685         data->resume = 1;
 686
 687         return result;
 688
 689 err_freqfree:
 690         kfree(data->freq_table);
 691 err_unreg:
 692         acpi_processor_unregister_performance(perf, cpu);
 693 err_free:
 694         kfree(data);
 695         per_cpu(acfreq_data, cpu) = NULL;
 696
 697         return result;
 698 }
 699
 700 static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
 701 {
 702         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 703
 704         pr_debug("acpi_cpufreq_cpu_exit\n");
 705
 706         if (data) {
 707                 cpufreq_frequency_table_put_attr(policy->cpu);
 708                 per_cpu(acfreq_data, policy->cpu) = NULL;
 709                 acpi_processor_unregister_performance(data->acpi_data,
 710                                                       policy->cpu);
 711                 if (data->original_controls)
 712                         kfree(data->original_controls);
 713                 kfree(data->freq_table);
 714                 kfree(data);
 715         }
 716
 717         return 0;
 718 }
 719
 720 static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
 721 {
 722         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 723
 724         pr_debug("acpi_cpufreq_resume\n");
 725
 726         data->resume = 1;
 727
 728         return 0;
 729 }
 730
 731
 732 /* sysfs interface to change operating points voltages */
 733
 734 static unsigned int extract_fid_from_control(unsigned int control)
 735 {
 736         return ((control & INTEL_MSR_FID_MASK) >> INTEL_MSR_FID_SHIFT);
 737 }
 738
 739 static unsigned int extract_vid_from_control(unsigned int control)
 740 {
 741         return (control & INTEL_MSR_VID_MASK);
 742 }
 743
 744
 745 static bool check_cpu_control_capability(struct acpi_cpufreq_data *data) {
 746  /* check if the cpu we are running on is capable of setting new control data
 747   *
 748   */
 749         if (unlikely(data == NULL ||
 750                      data->acpi_data == NULL ||
 751                      data->freq_table == NULL ||
 752                      data->cpu_feature != SYSTEM_INTEL_MSR_CAPABLE)) {
 753                 return false;
 754         } else {
 755                 return true;
 756         };
 757 }
 758
 759
 760 static ssize_t check_origial_table (struct acpi_cpufreq_data *data)
 761 {
 762
 763         struct acpi_processor_performance *acpi_data;
 764         struct cpufreq_frequency_table *freq_table;
 765         unsigned int state_index;
 766
 767         acpi_data = data->acpi_data;
 768         freq_table = data->freq_table;
 769
 770         if (data->original_controls == NULL) {
 771                 // Backup original control values
 772                 data->original_controls = kcalloc(acpi_data->state_count,
 773                                                   sizeof(acpi_integer), GFP_KERNEL);
 774                 if (data->original_controls == NULL) {
 775                         printk("failed to allocate memory for original control values\n");
 776                         return -ENOMEM;
 777                 }
 778                 for (state_index = 0; state_index < acpi_data->state_count; state_index++) {
 779                         data->original_controls[state_index] = acpi_data->states[state_index].control;
 780                 }
 781         }
 782         return 0;
 783 }
 784
 785 static ssize_t show_freq_attr_vids(struct cpufreq_policy *policy, char *buf)
 786  /* display phc's voltage id's
 787   *
 788   */
 789 {
 790         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 791         struct acpi_processor_performance *acpi_data;
 792         struct cpufreq_frequency_table *freq_table;
 793         unsigned int i;
 794         unsigned int vid;
 795         ssize_t count = 0;
 796
 797         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 798
 799         acpi_data = data->acpi_data;
 800         freq_table = data->freq_table;
 801
 802         for (i = 0; freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 803                 vid = extract_vid_from_control(acpi_data->states[freq_table[i].index].control);
 804                 count += sprintf(&buf[count], "%u ", vid);
 805         }
 806         count += sprintf(&buf[count], "\n");
 807
 808         return count;
 809 }
 810
 811 static ssize_t show_freq_attr_default_vids(struct cpufreq_policy *policy, char *buf)
 812  /* display acpi's default voltage id's
 813   *
 814   */
 815 {
 816         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 817         struct cpufreq_frequency_table *freq_table;
 818         unsigned int i;
 819         unsigned int vid;
 820         ssize_t count = 0;
 821         ssize_t retval;
 822
 823         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 824
 825         retval = check_origial_table(data);
 826         if (0 != retval)
 827                 return retval;
 828
 829         freq_table = data->freq_table;
 830
 831         for (i = 0; freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 832                 vid = extract_vid_from_control(data->original_controls[freq_table[i].index]);
 833                 count += sprintf(&buf[count], "%u ", vid);
 834         }
 835         count += sprintf(&buf[count], "\n");
 836
 837         return count;
 838 }
 839
 840 static ssize_t show_freq_attr_fids(struct cpufreq_policy *policy, char *buf)
 841  /* display phc's frequeny id's
 842   *
 843   */
 844 {
 845         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 846         struct acpi_processor_performance *acpi_data;
 847         struct cpufreq_frequency_table *freq_table;
 848         unsigned int i;
 849         unsigned int fid;
 850         ssize_t count = 0;
 851
 852         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 853
 854         acpi_data = data->acpi_data;
 855         freq_table = data->freq_table;
 856
 857         for (i = 0; freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 858                 fid = extract_fid_from_control(acpi_data->states[freq_table[i].index].control);
 859                 count += sprintf(&buf[count], "%u ", fid);
 860         }
 861         count += sprintf(&buf[count], "\n");
 862
 863         return count;
 864 }
 865
 866 static ssize_t show_freq_attr_controls(struct cpufreq_policy *policy, char *buf)
 867  /* display phc's controls for the cpu (frequency id's and related voltage id's)
 868   *
 869   */
 870 {
 871         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 872         struct acpi_processor_performance *acpi_data;
 873         struct cpufreq_frequency_table *freq_table;
 874         unsigned int i;
 875         unsigned int fid;
 876         unsigned int vid;
 877         ssize_t count = 0;
 878
 879         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 880
 881         acpi_data = data->acpi_data;
 882         freq_table = data->freq_table;
 883
 884         for (i = 0; freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 885                 fid = extract_fid_from_control(acpi_data->states[freq_table[i].index].control);
 886                 vid = extract_vid_from_control(acpi_data->states[freq_table[i].index].control);
 887                 count += sprintf(&buf[count], "%u:%u ", fid, vid);
 888         }
 889         count += sprintf(&buf[count], "\n");
 890
 891         return count;
 892 }
 893
 894 static ssize_t show_freq_attr_default_controls(struct cpufreq_policy *policy, char *buf)
 895  /* display acpi's default controls for the cpu (frequency id's and related voltage id's)
 896   *
 897   */
 898 {
 899         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 900         struct cpufreq_frequency_table *freq_table;
 901         unsigned int i;
 902         unsigned int fid;
 903         unsigned int vid;
 904         ssize_t count = 0;
 905         ssize_t retval;
 906
 907         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 908
 909         retval = check_origial_table(data);
 910         if (0 != retval)
 911                 return retval;
 912
 913         freq_table = data->freq_table;
 914
 915         for (i = 0; freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 916                 fid = extract_fid_from_control(data->original_controls[freq_table[i].index]);
 917                 vid = extract_vid_from_control(data->original_controls[freq_table[i].index]);
 918                 count += sprintf(&buf[count], "%u:%u ", fid, vid);
 919         }
 920         count += sprintf(&buf[count], "\n");
 921
 922         return count;
 923 }
 924
 925
 926 static ssize_t store_freq_attr_vids(struct cpufreq_policy *policy, const char *buf, size_t count)
 927  /* store the voltage id's for the related frequency
 928   * We are going to do some sanity checks here to prevent users
 929   * from setting higher voltages than the default one.
 930   */
 931 {
 932         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 933         struct acpi_processor_performance *acpi_data;
 934         struct cpufreq_frequency_table *freq_table;
 935         unsigned int freq_index;
 936         unsigned int state_index;
 937         unsigned int new_vid;
 938         unsigned int original_vid;
 939         unsigned int new_control;
 940         unsigned int original_control;
 941         const char *curr_buf = buf;
 942         char *next_buf;
 943         ssize_t retval;
 944
 945         if (!check_cpu_control_capability(data)) return -ENODEV; //check if CPU is capable of changing controls
 946
 947         retval = check_origial_table(data);
 948         if (0 != retval)
 949                 return retval;
 950
 951         acpi_data = data->acpi_data;
 952         freq_table = data->freq_table;
 953
 954         /* for each value taken from the sysfs interfalce (phc_vids) get entrys and convert them to unsigned long integers*/
 955         for (freq_index = 0; freq_table[freq_index].frequency != CPUFREQ_TABLE_END; freq_index++) {
 956                 new_vid = simple_strtoul(curr_buf, &next_buf, 10);
 957                 if (next_buf == curr_buf) {
 958                         if ((curr_buf - buf == count - 1) && (*curr_buf == '\n')) {   //end of line?
 959                                 curr_buf++;
 960                                 break;
 961                         }
 962                         //if we didn't got end of line but there is nothing more to read something went wrong...
 963                         printk("failed to parse vid value at %i (%s)\n", freq_index, curr_buf);
 964                         return -EINVAL;
 965                 }
 966
 967                 state_index = freq_table[freq_index].index;
 968                 original_control = data->original_controls[state_index];
 969                 original_vid = original_control & INTEL_MSR_VID_MASK;
 970
 971                 /* before we store the values we do some checks to prevent
 972                  * users to set up values higher than the default one
 973                  */
 974                 if (new_vid <= original_vid) {
 975                         new_control = (original_control & ~INTEL_MSR_VID_MASK) | new_vid;
 976                         printk("setting control at %i to %x (default is %x)\n",
 977                                 freq_index, new_control, original_control);
 978                         acpi_data->states[state_index].control = new_control;
 979
 980                 } else {
 981                         printk("skipping vid at %i, %u is greater than default %u\n",
 982                                freq_index, new_vid, original_vid);
 983                 }
 984
 985                 curr_buf = next_buf;
 986                 /* jump over value seperators (space or comma).
 987                  * There could be more than one space or comma character
 988                  * to separate two values so we better do it using a loop.
 989                  */
 990                 while ((curr_buf - buf < count) && ((*curr_buf == ' ') || (*curr_buf == ','))) {
 991                         curr_buf++;
 992                 }
 993         }
 994
 995         /* set new voltage for current frequency */
 996         data->resume = 1;
 997         acpi_cpufreq_target(policy, get_cur_freq_on_cpu(policy->cpu), CPUFREQ_RELATION_L);
 998
 999         return curr_buf - buf;
1000 }
1001
1002 static ssize_t store_freq_attr_controls(struct cpufreq_policy *policy, const char *buf, size_t count)
1003  /* store the controls (frequency id's and related voltage id's)
1004   * We are going to do some sanity checks here to prevent users
1005   * from setting higher voltages than the default one.
1006   */
1007 {
1008         struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
1009         struct acpi_processor_performance *acpi_data;
1010         struct cpufreq_frequency_table *freq_table;
1011         const char   *curr_buf;
1012         unsigned int  op_count;
1013         unsigned int  state_index;
1014         int           isok;
1015         char         *next_buf;
1016         ssize_t       retval;
1017         unsigned int  new_vid;
1018         unsigned int  original_vid;
1019         unsigned int  new_fid;
1020         unsigned int  old_fid;
1021         unsigned int  original_control;
1022         unsigned int  old_control;
1023         unsigned int  new_control;
1024         int           found;
1025
1026         if (!check_cpu_control_capability(data)) return -ENODEV;
1027
1028         retval = check_origial_table(data);
1029         if (0 != retval)
1030                 return retval;
1031
1032         acpi_data = data->acpi_data;
1033         freq_table = data->freq_table;
1034
1035         op_count = 0;
1036         curr_buf = buf;
1037         next_buf = NULL;
1038         isok     = 1;
1039
1040         while ( (isok) && (curr_buf != NULL) )
1041         {
1042                 op_count++;
1043                 // Parse fid
1044                 new_fid = simple_strtoul(curr_buf, &next_buf, 10);
1045                 if ((next_buf != curr_buf) && (next_buf != NULL))
1046                 {
1047                         // Parse separator between frequency and voltage
1048                         curr_buf = next_buf;
1049                         next_buf = NULL;
1050                         if (*curr_buf==':')
1051                         {
1052                                 curr_buf++;
1053                                 // Parse vid
1054                                 new_vid = simple_strtoul(curr_buf, &next_buf, 10);
1055                                 if ((next_buf != curr_buf) && (next_buf != NULL))
1056                                 {
1057                                         found = 0;
1058                                         for (state_index = 0; state_index < acpi_data->state_count; state_index++) {
1059                                                 old_control = acpi_data->states[state_index].control;
1060                                                 old_fid = extract_fid_from_control(old_control);
1061                                                 if (new_fid == old_fid)
1062                                                 {
1063                                                         found = 1;
1064                                                         original_control = data->original_controls[state_index];
1065                                                         original_vid = extract_vid_from_control(original_control);
1066                                                         if (new_vid <= original_vid)
1067                                                         {
1068                                                                 new_control = (original_control & ~INTEL_MSR_VID_MASK) | new_vid;
1069                                                                 printk("setting control at %i to %x (default is %x)\n",
1070                                                                         state_index, new_control, original_control);
1071                                                                 acpi_data->states[state_index].control = new_control;
1072
1073                                                         } else {
1074                                                                 printk("skipping vid at %i, %u is greater than default %u\n",
1075                                                                        state_index, new_vid, original_vid);
1076                                                         }
1077                                                 }
1078                                         }
1079
1080                                         if (found == 0)
1081                                         {
1082                                                 printk("operating point # %u not found (FID = %u)\n", op_count, new_fid);
1083                                                 isok = 0;
1084                                         }
1085
1086                                         // Parse seprator before next operating point, if any
1087                                         curr_buf = next_buf;
1088                                         next_buf = NULL;
1089                                         if ((*curr_buf == ',') || (*curr_buf == ' '))
1090                                                 curr_buf++;
1091                                         else
1092                                                 curr_buf = NULL;
1093                                 }
1094                                 else
1095                                 {
1096                                         printk("failed to parse VID of operating point # %u (%s)\n", op_count, curr_buf);
1097                                         isok = 0;
1098                                 }
1099                         }
1100                         else
1101                         {
1102                                 printk("failed to parse operating point # %u (%s)\n", op_count, curr_buf);
1103                                 isok = 0;
1104                         }
1105                 }
1106                 else
1107                 {
1108                         printk("failed to parse FID of operating point # %u (%s)\n", op_count, curr_buf);
1109                         isok = 0;
1110                 }
1111         }
1112
1113         if (isok)
1114         {
1115                 retval = count;
1116                 /* set new voltage at current frequency */
1117                 data->resume = 1;
1118                 acpi_cpufreq_target(policy, get_cur_freq_on_cpu(policy->cpu), CPUFREQ_RELATION_L);
1119         }
1120         else
1121         {
1122                 retval = -EINVAL;
1123         }
1124
1125         return retval;
1126 }
1127
1128 static ssize_t show_freq_attr_phc_version(struct cpufreq_policy *policy, char *buf)
1129  /* print out the phc version string set at the beginning of that file
1130   */
1131 {
1132         ssize_t count = 0;
1133         count += sprintf(&buf[count], "%s\n", PHC_VERSION_STRING);
1134         return count;
1135 }
1136
1137
1138
1139 static struct freq_attr cpufreq_freq_attr_phc_version =
1140 {
1141         /*display phc's version string*/
1142        .attr = { .name = "phc_version", .mode = 0444/*, .owner = THIS_MODULE */},
1143        .show = show_freq_attr_phc_version,
1144        .store = NULL,
1145 };
1146
1147 static struct freq_attr cpufreq_freq_attr_vids =
1148 {
1149         /*display phc's voltage id's for the cpu*/
1150        .attr = { .name = "phc_vids", .mode = 0644/*, .owner = THIS_MODULE */},
1151        .show = show_freq_attr_vids,
1152        .store = store_freq_attr_vids,
1153 };
1154
1155 static struct freq_attr cpufreq_freq_attr_default_vids =
1156 {
1157         /*display acpi's default frequency id's for the cpu*/
1158        .attr = { .name = "phc_default_vids", .mode = 0444/*, .owner = THIS_MODULE */},
1159        .show = show_freq_attr_default_vids,
1160        .store = NULL,
1161 };
1162
1163 static struct freq_attr cpufreq_freq_attr_fids =
1164 {
1165         /*display phc's default frequency id's for the cpu*/
1166        .attr = { .name = "phc_fids", .mode = 0444/*, .owner = THIS_MODULE */},
1167        .show = show_freq_attr_fids,
1168        .store = NULL,
1169 };
1170
1171 static struct freq_attr cpufreq_freq_attr_controls =
1172 {
1173         /*display phc's current voltage/frequency controls for the cpu*/
1174        .attr = { .name = "phc_controls", .mode = 0644/*, .owner = THIS_MODULE */},
1175        .show = show_freq_attr_controls,
1176        .store = store_freq_attr_controls,
1177 };
1178
1179 static struct freq_attr cpufreq_freq_attr_default_controls =
1180 {
1181         /*display acpi's default voltage/frequency controls for the cpu*/
1182        .attr = { .name = "phc_default_controls", .mode = 0444/*, .owner = THIS_MODULE */},
1183        .show = show_freq_attr_default_controls,
1184        .store = NULL,
1185 };
1186
1187 static struct freq_attr *acpi_cpufreq_attr[] = {
1188         &cpufreq_freq_attr_scaling_available_freqs,
1189         &cpufreq_freq_attr_phc_version,
1190         &cpufreq_freq_attr_vids,
1191         &cpufreq_freq_attr_default_vids,
1192         &cpufreq_freq_attr_fids,
1193         &cpufreq_freq_attr_controls,
1194         &cpufreq_freq_attr_default_controls,
1195         NULL,
1196 };
1197
1198 static struct cpufreq_driver acpi_cpufreq_driver = {
1199         .verify         = acpi_cpufreq_verify,
1200         .target         = acpi_cpufreq_target,
1201         .bios_limit     = acpi_processor_get_bios_limit,
1202         .init           = acpi_cpufreq_cpu_init,
1203         .exit           = acpi_cpufreq_cpu_exit,
1204         .resume         = acpi_cpufreq_resume,
1205         .name           = "acpi-cpufreq",
1206         .owner          = THIS_MODULE,
1207         .attr           = acpi_cpufreq_attr,
1208 };
1209
1210 static int __init acpi_cpufreq_init(void)
1211 {
1212         int ret;
1213
1214         if (acpi_disabled)
1215                 return 0;
1216
1217         pr_debug("acpi_cpufreq_init\n");
1218
1219         ret = acpi_cpufreq_early_init();
1220         if (ret)
1221                 return ret;
1222
1223         ret = cpufreq_register_driver(&acpi_cpufreq_driver);
1224         if (ret)
1225                 free_acpi_perf_data();
1226
1227         return ret;
1228 }
1229
1230 static void __exit acpi_cpufreq_exit(void)
1231 {
1232         pr_debug("acpi_cpufreq_exit\n");
1233
1234         cpufreq_unregister_driver(&acpi_cpufreq_driver);
1235
1236         free_acpi_perf_data();
1237 }
1238
1239 module_param(acpi_pstate_strict, uint, 0644);
1240 MODULE_PARM_DESC(acpi_pstate_strict,
1241         "value 0 or non-zero. non-zero -> strict ACPI checks are "
1242         "performed during frequency changes.");
1243
1244 static int param_set_phc_controls(const char *controls, struct kernel_param *kp)
1245 {
1246         int cpu;
1247
1248         for_each_possible_cpu(cpu) {
1249                 store_freq_attr_controls(cpufreq_cpu_get(cpu), controls, 0);
1250         }
1251         return 0;
1252 }
1253
1254 module_param_call(phc_controls, param_set_phc_controls, NULL, NULL, 0644);
1255 __MODULE_PARM_TYPE(phc_controls, "string");
1256 MODULE_PARM_DESC(phc_controls, "Set initial phc_controls");
1257
1258 late_initcall(acpi_cpufreq_init);
1259 module_exit(acpi_cpufreq_exit);
1260
1261 MODULE_ALIAS("acpi");