arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.c

   1 /*
   2  * acpi-cpufreq.c - ACPI Processor P-States Driver ($Revision: 1.4 $)
   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 #include <linux/kernel.h>
  29 #include <linux/module.h>
  30 #include <linux/init.h>
  31 #include <linux/smp.h>
  32 #include <linux/sched.h>
  33 #include <linux/cpufreq.h>
  34 #include <linux/compiler.h>
  35 #include <linux/dmi.h>
  36 #include <linux/ftrace.h>
  37
  38 #include <linux/acpi.h>
  39 #include <acpi/processor.h>
  40
  41 #include <asm/io.h>
  42 #include <asm/msr.h>
  43 #include <asm/processor.h>
  44 #include <asm/cpufeature.h>
  45 #include <asm/delay.h>
  46 #include <asm/uaccess.h>
  47
  48 #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "acpi-cpufreq", msg)
  49
  50 MODULE_AUTHOR("Paul Diefenbaugh, Dominik Brodowski");
  51 MODULE_DESCRIPTION("ACPI Processor P-States Driver");
  52 MODULE_LICENSE("GPL");
  53
  54 enum {
  55         UNDEFINED_CAPABLE = 0,
  56         SYSTEM_INTEL_MSR_CAPABLE,
  57         SYSTEM_IO_CAPABLE,
  58 };
  59
  60 #define INTEL_MSR_RANGE         (0xffff)
  61 #define CPUID_6_ECX_APERFMPERF_CAPABILITY       (0x1)
  62
  63 struct acpi_cpufreq_data {
  64         struct acpi_processor_performance *acpi_data;
  65         struct cpufreq_frequency_table *freq_table;
  66         unsigned int max_freq;
  67         unsigned int resume;
  68         unsigned int cpu_feature;
  69 };
  70
  71 static DEFINE_PER_CPU(struct acpi_cpufreq_data *, drv_data);
  72
  73 /* acpi_perf_data is a pointer to percpu data. */
  74 static struct acpi_processor_performance *acpi_perf_data;
  75
  76 static struct cpufreq_driver acpi_cpufreq_driver;
  77
  78 static unsigned int acpi_pstate_strict;
  79
  80 static int check_est_cpu(unsigned int cpuid)
  81 {
  82         struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
  83
  84         if (cpu->x86_vendor != X86_VENDOR_INTEL ||
  85             !cpu_has(cpu, X86_FEATURE_EST))
  86                 return 0;
  87
  88         return 1;
  89 }
  90
  91 static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
  92 {
  93         struct acpi_processor_performance *perf;
  94         int i;
  95
  96         perf = data->acpi_data;
  97
  98         for (i=0; i<perf->state_count; i++) {
  99                 if (value == perf->states[i].status)
 100                         return data->freq_table[i].frequency;
 101         }
 102         return 0;
 103 }
 104
 105 static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
 106 {
 107         int i;
 108         struct acpi_processor_performance *perf;
 109
 110         msr &= INTEL_MSR_RANGE;
 111         perf = data->acpi_data;
 112
 113         for (i=0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 114                 if (msr == perf->states[data->freq_table[i].index].status)
 115                         return data->freq_table[i].frequency;
 116         }
 117         return data->freq_table[0].frequency;
 118 }
 119
 120 static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
 121 {
 122         switch (data->cpu_feature) {
 123         case SYSTEM_INTEL_MSR_CAPABLE:
 124                 return extract_msr(val, data);
 125         case SYSTEM_IO_CAPABLE:
 126                 return extract_io(val, data);
 127         default:
 128                 return 0;
 129         }
 130 }
 131
 132 struct msr_addr {
 133         u32 reg;
 134 };
 135
 136 struct io_addr {
 137         u16 port;
 138         u8 bit_width;
 139 };
 140
 141 typedef union {
 142         struct msr_addr msr;
 143         struct io_addr io;
 144 } drv_addr_union;
 145
 146 struct drv_cmd {
 147         unsigned int type;
 148         cpumask_var_t mask;
 149         drv_addr_union addr;
 150         u32 val;
 151 };
 152
 153 static void do_drv_read(struct drv_cmd *cmd)
 154 {
 155         u32 h;
 156
 157         switch (cmd->type) {
 158         case SYSTEM_INTEL_MSR_CAPABLE:
 159                 rdmsr(cmd->addr.msr.reg, cmd->val, h);
 160                 break;
 161         case SYSTEM_IO_CAPABLE:
 162                 acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
 163                                 &cmd->val,
 164                                 (u32)cmd->addr.io.bit_width);
 165                 break;
 166         default:
 167                 break;
 168         }
 169 }
 170
 171 static void do_drv_write(struct drv_cmd *cmd)
 172 {
 173         u32 lo, hi;
 174
 175         switch (cmd->type) {
 176         case SYSTEM_INTEL_MSR_CAPABLE:
 177                 rdmsr(cmd->addr.msr.reg, lo, hi);
 178                 lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
 179                 wrmsr(cmd->addr.msr.reg, lo, hi);
 180                 break;
 181         case SYSTEM_IO_CAPABLE:
 182                 acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
 183                                 cmd->val,
 184                                 (u32)cmd->addr.io.bit_width);
 185                 break;
 186         default:
 187                 break;
 188         }
 189 }
 190
 191 static void drv_read(struct drv_cmd *cmd)
 192 {
 193         cpumask_t saved_mask = current->cpus_allowed;
 194         cmd->val = 0;
 195
 196         set_cpus_allowed_ptr(current, cmd->mask);
 197         do_drv_read(cmd);
 198         set_cpus_allowed_ptr(current, &saved_mask);
 199 }
 200
 201 static void drv_write(struct drv_cmd *cmd)
 202 {
 203         cpumask_t saved_mask = current->cpus_allowed;
 204         unsigned int i;
 205
 206         for_each_cpu(i, cmd->mask) {
 207                 set_cpus_allowed_ptr(current, cpumask_of(i));
 208                 do_drv_write(cmd);
 209         }
 210
 211         set_cpus_allowed_ptr(current, &saved_mask);
 212         return;
 213 }
 214
 215 static u32 get_cur_val(const struct cpumask *mask)
 216 {
 217         struct acpi_processor_performance *perf;
 218         struct drv_cmd cmd;
 219
 220         if (unlikely(cpumask_empty(mask)))
 221                 return 0;
 222
 223         switch (per_cpu(drv_data, cpumask_first(mask))->cpu_feature) {
 224         case SYSTEM_INTEL_MSR_CAPABLE:
 225                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 226                 cmd.addr.msr.reg = MSR_IA32_PERF_STATUS;
 227                 break;
 228         case SYSTEM_IO_CAPABLE:
 229                 cmd.type = SYSTEM_IO_CAPABLE;
 230                 perf = per_cpu(drv_data, cpumask_first(mask))->acpi_data;
 231                 cmd.addr.io.port = perf->control_register.address;
 232                 cmd.addr.io.bit_width = perf->control_register.bit_width;
 233                 break;
 234         default:
 235                 return 0;
 236         }
 237
 238         drv_read(&cmd);
 239
 240         dprintk("get_cur_val = %u\n", cmd.val);
 241
 242         return cmd.val;
 243 }
 244
 245 struct perf_cur {
 246         union {
 247                 struct {
 248                         u32 lo;
 249                         u32 hi;
 250                 } split;
 251                 u64 whole;
 252         } aperf_cur, mperf_cur;
 253 };
 254
 255
 256 static long read_measured_perf_ctrs(void *_cur)
 257 {
 258         struct perf_cur *cur = _cur;
 259
 260         rdmsr(MSR_IA32_APERF, cur->aperf_cur.split.lo, cur->aperf_cur.split.hi);
 261         rdmsr(MSR_IA32_MPERF, cur->mperf_cur.split.lo, cur->mperf_cur.split.hi);
 262
 263         wrmsr(MSR_IA32_APERF, 0, 0);
 264         wrmsr(MSR_IA32_MPERF, 0, 0);
 265
 266         return 0;
 267 }
 268
 269 /*
 270  * Return the measured active (C0) frequency on this CPU since last call
 271  * to this function.
 272  * Input: cpu number
 273  * Return: Average CPU frequency in terms of max frequency (zero on error)
 274  *
 275  * We use IA32_MPERF and IA32_APERF MSRs to get the measured performance
 276  * over a period of time, while CPU is in C0 state.
 277  * IA32_MPERF counts at the rate of max advertised frequency
 278  * IA32_APERF counts at the rate of actual CPU frequency
 279  * Only IA32_APERF/IA32_MPERF ratio is architecturally defined and
 280  * no meaning should be associated with absolute values of these MSRs.
 281  */
 282 static unsigned int get_measured_perf(struct cpufreq_policy *policy,
 283                                       unsigned int cpu)
 284 {
 285         struct perf_cur cur;
 286         unsigned int perf_percent;
 287         unsigned int retval;
 288
 289         if (!work_on_cpu(cpu, read_measured_perf_ctrs, &cur))
 290                 return 0;
 291
 292 #ifdef __i386__
 293         /*
 294          * We dont want to do 64 bit divide with 32 bit kernel
 295          * Get an approximate value. Return failure in case we cannot get
 296          * an approximate value.
 297          */
 298         if (unlikely(cur.aperf_cur.split.hi || cur.mperf_cur.split.hi)) {
 299                 int shift_count;
 300                 u32 h;
 301
 302                 h = max_t(u32, cur.aperf_cur.split.hi, cur.mperf_cur.split.hi);
 303                 shift_count = fls(h);
 304
 305                 cur.aperf_cur.whole >>= shift_count;
 306                 cur.mperf_cur.whole >>= shift_count;
 307         }
 308
 309         if (((unsigned long)(-1) / 100) < cur.aperf_cur.split.lo) {
 310                 int shift_count = 7;
 311                 cur.aperf_cur.split.lo >>= shift_count;
 312                 cur.mperf_cur.split.lo >>= shift_count;
 313         }
 314
 315         if (cur.aperf_cur.split.lo && cur.mperf_cur.split.lo)
 316                 perf_percent = (cur.aperf_cur.split.lo * 100) /
 317                                 cur.mperf_cur.split.lo;
 318         else
 319                 perf_percent = 0;
 320
 321 #else
 322         if (unlikely(((unsigned long)(-1) / 100) < cur.aperf_cur.whole)) {
 323                 int shift_count = 7;
 324                 cur.aperf_cur.whole >>= shift_count;
 325                 cur.mperf_cur.whole >>= shift_count;
 326         }
 327
 328         if (cur.aperf_cur.whole && cur.mperf_cur.whole)
 329                 perf_percent = (cur.aperf_cur.whole * 100) /
 330                                 cur.mperf_cur.whole;
 331         else
 332                 perf_percent = 0;
 333
 334 #endif
 335
 336         retval = per_cpu(drv_data, policy->cpu)->max_freq * perf_percent / 100;
 337
 338         return retval;
 339 }
 340
 341 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 342 {
 343         struct acpi_cpufreq_data *data = per_cpu(drv_data, cpu);
 344         unsigned int freq;
 345         unsigned int cached_freq;
 346
 347         dprintk("get_cur_freq_on_cpu (%d)\n", cpu);
 348
 349         if (unlikely(data == NULL ||
 350                      data->acpi_data == NULL || data->freq_table == NULL)) {
 351                 return 0;
 352         }
 353
 354         cached_freq = data->freq_table[data->acpi_data->state].frequency;
 355         freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
 356         if (freq != cached_freq) {
 357                 /*
 358                  * The dreaded BIOS frequency change behind our back.
 359                  * Force set the frequency on next target call.
 360                  */
 361                 data->resume = 1;
 362         }
 363
 364         dprintk("cur freq = %u\n", freq);
 365
 366         return freq;
 367 }
 368
 369 static unsigned int check_freqs(const cpumask_t *mask, unsigned int freq,
 370                                 struct acpi_cpufreq_data *data)
 371 {
 372         unsigned int cur_freq;
 373         unsigned int i;
 374
 375         for (i=0; i<100; i++) {
 376                 cur_freq = extract_freq(get_cur_val(mask), data);
 377                 if (cur_freq == freq)
 378                         return 1;
 379                 udelay(10);
 380         }
 381         return 0;
 382 }
 383
 384 static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 385                                unsigned int target_freq, unsigned int relation)
 386 {
 387         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
 388         struct acpi_processor_performance *perf;
 389         struct cpufreq_freqs freqs;
 390         struct drv_cmd cmd;
 391         unsigned int next_state = 0; /* Index into freq_table */
 392         unsigned int next_perf_state = 0; /* Index into perf table */
 393         unsigned int i;
 394         int result = 0;
 395         struct power_trace it;
 396
 397         dprintk("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 398
 399         if (unlikely(data == NULL ||
 400              data->acpi_data == NULL || data->freq_table == NULL)) {
 401                 return -ENODEV;
 402         }
 403
 404         if (unlikely(!alloc_cpumask_var(&cmd.mask, GFP_KERNEL)))
 405                 return -ENOMEM;
 406
 407         perf = data->acpi_data;
 408         result = cpufreq_frequency_table_target(policy,
 409                                                 data->freq_table,
 410                                                 target_freq,
 411                                                 relation, &next_state);
 412         if (unlikely(result)) {
 413                 result = -ENODEV;
 414                 goto out;
 415         }
 416
 417         next_perf_state = data->freq_table[next_state].index;
 418         if (perf->state == next_perf_state) {
 419                 if (unlikely(data->resume)) {
 420                         dprintk("Called after resume, resetting to P%d\n",
 421                                 next_perf_state);
 422                         data->resume = 0;
 423                 } else {
 424                         dprintk("Already at target state (P%d)\n",
 425                                 next_perf_state);
 426                         goto out;
 427                 }
 428         }
 429
 430         trace_power_mark(&it, POWER_PSTATE, next_perf_state);
 431
 432         switch (data->cpu_feature) {
 433         case SYSTEM_INTEL_MSR_CAPABLE:
 434                 cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 435                 cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
 436                 cmd.val = (u32) perf->states[next_perf_state].control;
 437                 break;
 438         case SYSTEM_IO_CAPABLE:
 439                 cmd.type = SYSTEM_IO_CAPABLE;
 440                 cmd.addr.io.port = perf->control_register.address;
 441                 cmd.addr.io.bit_width = perf->control_register.bit_width;
 442                 cmd.val = (u32) perf->states[next_perf_state].control;
 443                 break;
 444         default:
 445                 result = -ENODEV;
 446                 goto out;
 447         }
 448
 449         /* cpufreq holds the hotplug lock, so we are safe from here on */
 450         if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
 451                 cpumask_and(cmd.mask, cpu_online_mask, policy->cpus);
 452         else
 453                 cpumask_copy(cmd.mask, cpumask_of(policy->cpu));
 454
 455         freqs.old = perf->states[perf->state].core_frequency * 1000;
 456         freqs.new = data->freq_table[next_state].frequency;
 457         for_each_cpu(i, cmd.mask) {
 458                 freqs.cpu = i;
 459                 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 460         }
 461
 462         drv_write(&cmd);
 463
 464         if (acpi_pstate_strict) {
 465                 if (!check_freqs(cmd.mask, freqs.new, data)) {
 466                         dprintk("acpi_cpufreq_target failed (%d)\n",
 467                                 policy->cpu);
 468                         result = -EAGAIN;
 469                         goto out;
 470                 }
 471         }
 472
 473         for_each_cpu(i, cmd.mask) {
 474                 freqs.cpu = i;
 475                 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 476         }
 477         perf->state = next_perf_state;
 478
 479 out:
 480         free_cpumask_var(cmd.mask);
 481         return result;
 482 }
 483
 484 static int acpi_cpufreq_verify(struct cpufreq_policy *policy)
 485 {
 486         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
 487
 488         dprintk("acpi_cpufreq_verify\n");
 489
 490         return cpufreq_frequency_table_verify(policy, data->freq_table);
 491 }
 492
 493 static unsigned long
 494 acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
 495 {
 496         struct acpi_processor_performance *perf = data->acpi_data;
 497
 498         if (cpu_khz) {
 499                 /* search the closest match to cpu_khz */
 500                 unsigned int i;
 501                 unsigned long freq;
 502                 unsigned long freqn = perf->states[0].core_frequency * 1000;
 503
 504                 for (i=0; i<(perf->state_count-1); i++) {
 505                         freq = freqn;
 506                         freqn = perf->states[i+1].core_frequency * 1000;
 507                         if ((2 * cpu_khz) > (freqn + freq)) {
 508                                 perf->state = i;
 509                                 return freq;
 510                         }
 511                 }
 512                 perf->state = perf->state_count-1;
 513                 return freqn;
 514         } else {
 515                 /* assume CPU is at P0... */
 516                 perf->state = 0;
 517                 return perf->states[0].core_frequency * 1000;
 518         }
 519 }
 520
 521 static void free_acpi_perf_data(void)
 522 {
 523         unsigned int i;
 524
 525         /* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
 526         for_each_possible_cpu(i)
 527                 free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
 528                                  ->shared_cpu_map);
 529         free_percpu(acpi_perf_data);
 530 }
 531
 532 /*
 533  * acpi_cpufreq_early_init - initialize ACPI P-States library
 534  *
 535  * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
 536  * in order to determine correct frequency and voltage pairings. We can
 537  * do _PDC and _PSD and find out the processor dependency for the
 538  * actual init that will happen later...
 539  */
 540 static int __init acpi_cpufreq_early_init(void)
 541 {
 542         unsigned int i;
 543         dprintk("acpi_cpufreq_early_init\n");
 544
 545         acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
 546         if (!acpi_perf_data) {
 547                 dprintk("Memory allocation error for acpi_perf_data.\n");
 548                 return -ENOMEM;
 549         }
 550         for_each_possible_cpu(i) {
 551                 if (!alloc_cpumask_var_node(
 552                         &per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
 553                         GFP_KERNEL, cpu_to_node(i))) {
 554
 555                         /* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
 556                         free_acpi_perf_data();
 557                         return -ENOMEM;
 558                 }
 559         }
 560
 561         /* Do initialization in ACPI core */
 562         acpi_processor_preregister_performance(acpi_perf_data);
 563         return 0;
 564 }
 565
 566 #ifdef CONFIG_SMP
 567 /*
 568  * Some BIOSes do SW_ANY coordination internally, either set it up in hw
 569  * or do it in BIOS firmware and won't inform about it to OS. If not
 570  * detected, this has a side effect of making CPU run at a different speed
 571  * than OS intended it to run at. Detect it and handle it cleanly.
 572  */
 573 static int bios_with_sw_any_bug;
 574
 575 static int sw_any_bug_found(const struct dmi_system_id *d)
 576 {
 577         bios_with_sw_any_bug = 1;
 578         return 0;
 579 }
 580
 581 static const struct dmi_system_id sw_any_bug_dmi_table[] = {
 582         {
 583                 .callback = sw_any_bug_found,
 584                 .ident = "Supermicro Server X6DLP",
 585                 .matches = {
 586                         DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
 587                         DMI_MATCH(DMI_BIOS_VERSION, "080010"),
 588                         DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
 589                 },
 590         },
 591         { }
 592 };
 593 #endif
 594
 595 static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 596 {
 597         unsigned int i;
 598         unsigned int valid_states = 0;
 599         unsigned int cpu = policy->cpu;
 600         struct acpi_cpufreq_data *data;
 601         unsigned int result = 0;
 602         struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
 603         struct acpi_processor_performance *perf;
 604
 605         dprintk("acpi_cpufreq_cpu_init\n");
 606
 607         data = kzalloc(sizeof(struct acpi_cpufreq_data), GFP_KERNEL);
 608         if (!data)
 609                 return -ENOMEM;
 610
 611         data->acpi_data = percpu_ptr(acpi_perf_data, cpu);
 612         per_cpu(drv_data, cpu) = data;
 613
 614         if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
 615                 acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
 616
 617         result = acpi_processor_register_performance(data->acpi_data, cpu);
 618         if (result)
 619                 goto err_free;
 620
 621         perf = data->acpi_data;
 622         policy->shared_type = perf->shared_type;
 623
 624         /*
 625          * Will let policy->cpus know about dependency only when software
 626          * coordination is required.
 627          */
 628         if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
 629             policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
 630                 cpumask_copy(policy->cpus, perf->shared_cpu_map);
 631         }
 632         cpumask_copy(policy->related_cpus, perf->shared_cpu_map);
 633
 634 #ifdef CONFIG_SMP
 635         dmi_check_system(sw_any_bug_dmi_table);
 636         if (bios_with_sw_any_bug && cpumask_weight(policy->cpus) == 1) {
 637                 policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
 638                 cpumask_copy(policy->cpus, cpu_core_mask(cpu));
 639         }
 640 #endif
 641
 642         /* capability check */
 643         if (perf->state_count <= 1) {
 644                 dprintk("No P-States\n");
 645                 result = -ENODEV;
 646                 goto err_unreg;
 647         }
 648
 649         if (perf->control_register.space_id != perf->status_register.space_id) {
 650                 result = -ENODEV;
 651                 goto err_unreg;
 652         }
 653
 654         switch (perf->control_register.space_id) {
 655         case ACPI_ADR_SPACE_SYSTEM_IO:
 656                 dprintk("SYSTEM IO addr space\n");
 657                 data->cpu_feature = SYSTEM_IO_CAPABLE;
 658                 break;
 659         case ACPI_ADR_SPACE_FIXED_HARDWARE:
 660                 dprintk("HARDWARE addr space\n");
 661                 if (!check_est_cpu(cpu)) {
 662                         result = -ENODEV;
 663                         goto err_unreg;
 664                 }
 665                 data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
 666                 break;
 667         default:
 668                 dprintk("Unknown addr space %d\n",
 669                         (u32) (perf->control_register.space_id));
 670                 result = -ENODEV;
 671                 goto err_unreg;
 672         }
 673
 674         data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) *
 675                     (perf->state_count+1), GFP_KERNEL);
 676         if (!data->freq_table) {
 677                 result = -ENOMEM;
 678                 goto err_unreg;
 679         }
 680
 681         /* detect transition latency */
 682         policy->cpuinfo.transition_latency = 0;
 683         for (i=0; i<perf->state_count; i++) {
 684                 if ((perf->states[i].transition_latency * 1000) >
 685                     policy->cpuinfo.transition_latency)
 686                         policy->cpuinfo.transition_latency =
 687                             perf->states[i].transition_latency * 1000;
 688         }
 689
 690         data->max_freq = perf->states[0].core_frequency * 1000;
 691         /* table init */
 692         for (i=0; i<perf->state_count; i++) {
 693                 if (i>0 && perf->states[i].core_frequency >=
 694                     data->freq_table[valid_states-1].frequency / 1000)
 695                         continue;
 696
 697                 data->freq_table[valid_states].index = i;
 698                 data->freq_table[valid_states].frequency =
 699                     perf->states[i].core_frequency * 1000;
 700                 valid_states++;
 701         }
 702         data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
 703         perf->state = 0;
 704
 705         result = cpufreq_frequency_table_cpuinfo(policy, data->freq_table);
 706         if (result)
 707                 goto err_freqfree;
 708
 709         switch (perf->control_register.space_id) {
 710         case ACPI_ADR_SPACE_SYSTEM_IO:
 711                 /* Current speed is unknown and not detectable by IO port */
 712                 policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
 713                 break;
 714         case ACPI_ADR_SPACE_FIXED_HARDWARE:
 715                 acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
 716                 policy->cur = get_cur_freq_on_cpu(cpu);
 717                 break;
 718         default:
 719                 break;
 720         }
 721
 722         /* notify BIOS that we exist */
 723         acpi_processor_notify_smm(THIS_MODULE);
 724
 725         /* Check for APERF/MPERF support in hardware */
 726         if (c->x86_vendor == X86_VENDOR_INTEL && c->cpuid_level >= 6) {
 727                 unsigned int ecx;
 728                 ecx = cpuid_ecx(6);
 729                 if (ecx & CPUID_6_ECX_APERFMPERF_CAPABILITY)
 730                         acpi_cpufreq_driver.getavg = get_measured_perf;
 731         }
 732
 733         dprintk("CPU%u - ACPI performance management activated.\n", cpu);
 734         for (i = 0; i < perf->state_count; i++)
 735                 dprintk("     %cP%d: %d MHz, %d mW, %d uS\n",
 736                         (i == perf->state ? '*' : ' '), i,
 737                         (u32) perf->states[i].core_frequency,
 738                         (u32) perf->states[i].power,
 739                         (u32) perf->states[i].transition_latency);
 740
 741         cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
 742
 743         /*
 744          * the first call to ->target() should result in us actually
 745          * writing something to the appropriate registers.
 746          */
 747         data->resume = 1;
 748
 749         return result;
 750
 751 err_freqfree:
 752         kfree(data->freq_table);
 753 err_unreg:
 754         acpi_processor_unregister_performance(perf, cpu);
 755 err_free:
 756         kfree(data);
 757         per_cpu(drv_data, cpu) = NULL;
 758
 759         return result;
 760 }
 761
 762 static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
 763 {
 764         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
 765
 766         dprintk("acpi_cpufreq_cpu_exit\n");
 767
 768         if (data) {
 769                 cpufreq_frequency_table_put_attr(policy->cpu);
 770                 per_cpu(drv_data, policy->cpu) = NULL;
 771                 acpi_processor_unregister_performance(data->acpi_data,
 772                                                       policy->cpu);
 773                 kfree(data);
 774         }
 775
 776         return 0;
 777 }
 778
 779 static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
 780 {
 781         struct acpi_cpufreq_data *data = per_cpu(drv_data, policy->cpu);
 782
 783         dprintk("acpi_cpufreq_resume\n");
 784
 785         data->resume = 1;
 786
 787         return 0;
 788 }
 789
 790 static struct freq_attr *acpi_cpufreq_attr[] = {
 791         &cpufreq_freq_attr_scaling_available_freqs,
 792         NULL,
 793 };
 794
 795 static struct cpufreq_driver acpi_cpufreq_driver = {
 796         .verify = acpi_cpufreq_verify,
 797         .target = acpi_cpufreq_target,
 798         .init = acpi_cpufreq_cpu_init,
 799         .exit = acpi_cpufreq_cpu_exit,
 800         .resume = acpi_cpufreq_resume,
 801         .name = "acpi-cpufreq",
 802         .owner = THIS_MODULE,
 803         .attr = acpi_cpufreq_attr,
 804 };
 805
 806 static int __init acpi_cpufreq_init(void)
 807 {
 808         int ret;
 809
 810         if (acpi_disabled)
 811                 return 0;
 812
 813         dprintk("acpi_cpufreq_init\n");
 814
 815         ret = acpi_cpufreq_early_init();
 816         if (ret)
 817                 return ret;
 818
 819         ret = cpufreq_register_driver(&acpi_cpufreq_driver);
 820         if (ret)
 821                 free_acpi_perf_data();
 822
 823         return ret;
 824 }
 825
 826 static void __exit acpi_cpufreq_exit(void)
 827 {
 828         dprintk("acpi_cpufreq_exit\n");
 829
 830         cpufreq_unregister_driver(&acpi_cpufreq_driver);
 831
 832         free_percpu(acpi_perf_data);
 833 }
 834
 835 module_param(acpi_pstate_strict, uint, 0644);
 836 MODULE_PARM_DESC(acpi_pstate_strict,
 837         "value 0 or non-zero. non-zero -> strict ACPI checks are "
 838         "performed during frequency changes.");
 839
 840 late_initcall(acpi_cpufreq_init);
 841 module_exit(acpi_cpufreq_exit);
 842
 843 MODULE_ALIAS("acpi");