drivers/acpi/processor_idle.c

   1 /*
   2  * processor_idle - idle state submodule to the ACPI processor 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) 2004       Dominik Brodowski <linux@brodo.de>
   7  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
   8  *                      - Added processor hotplug support
   9  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  10  *                      - Added support for C3 on SMP
  11  *
  12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  13  *
  14  *  This program is free software; you can redistribute it and/or modify
  15  *  it under the terms of the GNU General Public License as published by
  16  *  the Free Software Foundation; either version 2 of the License, or (at
  17  *  your option) any later version.
  18  *
  19  *  This program is distributed in the hope that it will be useful, but
  20  *  WITHOUT ANY WARRANTY; without even the implied warranty of
  21  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  *  General Public License for more details.
  23  *
  24  *  You should have received a copy of the GNU General Public License along
  25  *  with this program; if not, write to the Free Software Foundation, Inc.,
  26  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
  27  *
  28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  29  */
  30
  31 #include <linux/kernel.h>
  32 #include <linux/module.h>
  33 #include <linux/init.h>
  34 #include <linux/cpufreq.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/acpi.h>
  38 #include <linux/dmi.h>
  39 #include <linux/moduleparam.h>
  40
  41 #include <asm/io.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <acpi/acpi_bus.h>
  45 #include <acpi/processor.h>
  46
  47 #define ACPI_PROCESSOR_COMPONENT        0x01000000
  48 #define ACPI_PROCESSOR_CLASS            "processor"
  49 #define ACPI_PROCESSOR_DRIVER_NAME      "ACPI Processor Driver"
  50 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  51 ACPI_MODULE_NAME                ("acpi_processor")
  52
  53 #define ACPI_PROCESSOR_FILE_POWER       "power"
  54
  55 #define US_TO_PM_TIMER_TICKS(t)         ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
  56 #define C2_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  57 #define C3_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  58
  59 static void (*pm_idle_save)(void);
  60 module_param(max_cstate, uint, 0644);
  61
  62 static unsigned int nocst = 0;
  63 module_param(nocst, uint, 0000);
  64
  65 /*
  66  * bm_history -- bit-mask with a bit per jiffy of bus-master activity
  67  * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
  68  * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
  69  * 100 HZ: 0x0000000F: 4 jiffies = 40ms
  70  * reduce history for more aggressive entry into C3
  71  */
  72 static unsigned int bm_history = (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
  73 module_param(bm_history, uint, 0644);
  74 /* --------------------------------------------------------------------------
  75                                 Power Management
  76    -------------------------------------------------------------------------- */
  77
  78 /*
  79  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  80  * For now disable this. Probably a bug somewhere else.
  81  *
  82  * To skip this limit, boot/load with a large max_cstate limit.
  83  */
  84 static int set_max_cstate(struct dmi_system_id *id)
  85 {
  86         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  87                 return 0;
  88
  89         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  90                 " Override with \"processor.max_cstate=%d\"\n", id->ident,
  91                 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  92
  93         max_cstate = (long)id->driver_data;
  94
  95         return 0;
  96 }
  97
  98
  99 static struct dmi_system_id __initdata processor_power_dmi_table[] = {
 100         { set_max_cstate, "IBM ThinkPad R40e", {
 101           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 102           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW") }, (void*)1},
 103         { set_max_cstate, "Medion 41700", {
 104           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 105           DMI_MATCH(DMI_BIOS_VERSION,"R01-A1J") }, (void*)1},
 106         { set_max_cstate, "Clevo 5600D", {
 107           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 108           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307") },
 109           (void*)2},
 110         {},
 111 };
 112
 113
 114 static inline u32
 115 ticks_elapsed (
 116         u32                     t1,
 117         u32                     t2)
 118 {
 119         if (t2 >= t1)
 120                 return (t2 - t1);
 121         else if (!acpi_fadt.tmr_val_ext)
 122                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 123         else
 124                 return ((0xFFFFFFFF - t1) + t2);
 125 }
 126
 127
 128 static void
 129 acpi_processor_power_activate (
 130         struct acpi_processor   *pr,
 131         struct acpi_processor_cx  *new)
 132 {
 133         struct acpi_processor_cx  *old;
 134
 135         if (!pr || !new)
 136                 return;
 137
 138         old = pr->power.state;
 139
 140         if (old)
 141                 old->promotion.count = 0;
 142         new->demotion.count = 0;
 143
 144         /* Cleanup from old state. */
 145         if (old) {
 146                 switch (old->type) {
 147                 case ACPI_STATE_C3:
 148                         /* Disable bus master reload */
 149                         if (new->type != ACPI_STATE_C3 && pr->flags.bm_check)
 150                                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0, ACPI_MTX_DO_NOT_LOCK);
 151                         break;
 152                 }
 153         }
 154
 155         /* Prepare to use new state. */
 156         switch (new->type) {
 157         case ACPI_STATE_C3:
 158                 /* Enable bus master reload */
 159                 if (old->type != ACPI_STATE_C3 && pr->flags.bm_check)
 160                         acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1, ACPI_MTX_DO_NOT_LOCK);
 161                 break;
 162         }
 163
 164         pr->power.state = new;
 165
 166         return;
 167 }
 168
 169
 170 static atomic_t         c3_cpu_count;
 171
 172
 173 static void acpi_processor_idle (void)
 174 {
 175         struct acpi_processor   *pr = NULL;
 176         struct acpi_processor_cx *cx = NULL;
 177         struct acpi_processor_cx *next_state = NULL;
 178         int                     sleep_ticks = 0;
 179         u32                     t1, t2 = 0;
 180
 181         pr = processors[raw_smp_processor_id()];
 182         if (!pr)
 183                 return;
 184
 185         /*
 186          * Interrupts must be disabled during bus mastering calculations and
 187          * for C2/C3 transitions.
 188          */
 189         local_irq_disable();
 190
 191         /*
 192          * Check whether we truly need to go idle, or should
 193          * reschedule:
 194          */
 195         if (unlikely(need_resched())) {
 196                 local_irq_enable();
 197                 return;
 198         }
 199
 200         cx = pr->power.state;
 201         if (!cx)
 202                 goto easy_out;
 203
 204         /*
 205          * Check BM Activity
 206          * -----------------
 207          * Check for bus mastering activity (if required), record, and check
 208          * for demotion.
 209          */
 210         if (pr->flags.bm_check) {
 211                 u32             bm_status = 0;
 212                 unsigned long   diff = jiffies - pr->power.bm_check_timestamp;
 213
 214                 if (diff > 32)
 215                         diff = 32;
 216
 217                 while (diff) {
 218                         /* if we didn't get called, assume there was busmaster activity */
 219                         diff--;
 220                         if (diff)
 221                                 pr->power.bm_activity |= 0x1;
 222                         pr->power.bm_activity <<= 1;
 223                 }
 224
 225                 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS,
 226                         &bm_status, ACPI_MTX_DO_NOT_LOCK);
 227                 if (bm_status) {
 228                         pr->power.bm_activity++;
 229                         acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS,
 230                                 1, ACPI_MTX_DO_NOT_LOCK);
 231                 }
 232                 /*
 233                  * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 234                  * the true state of bus mastering activity; forcing us to
 235                  * manually check the BMIDEA bit of each IDE channel.
 236                  */
 237                 else if (errata.piix4.bmisx) {
 238                         if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 239                                 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 240                                 pr->power.bm_activity++;
 241                 }
 242
 243                 pr->power.bm_check_timestamp = jiffies;
 244
 245                 /*
 246                  * Apply bus mastering demotion policy.  Automatically demote
 247                  * to avoid a faulty transition.  Note that the processor
 248                  * won't enter a low-power state during this call (to this
 249                  * funciton) but should upon the next.
 250                  *
 251                  * TBD: A better policy might be to fallback to the demotion
 252                  *      state (use it for this quantum only) istead of
 253                  *      demoting -- and rely on duration as our sole demotion
 254                  *      qualification.  This may, however, introduce DMA
 255                  *      issues (e.g. floppy DMA transfer overrun/underrun).
 256                  */
 257                 if (pr->power.bm_activity & cx->demotion.threshold.bm) {
 258                         local_irq_enable();
 259                         next_state = cx->demotion.state;
 260                         goto end;
 261                 }
 262         }
 263
 264         cx->usage++;
 265
 266         /*
 267          * Sleep:
 268          * ------
 269          * Invoke the current Cx state to put the processor to sleep.
 270          */
 271         switch (cx->type) {
 272
 273         case ACPI_STATE_C1:
 274                 /*
 275                  * Invoke C1.
 276                  * Use the appropriate idle routine, the one that would
 277                  * be used without acpi C-states.
 278                  */
 279                 if (pm_idle_save)
 280                         pm_idle_save();
 281                 else
 282                         safe_halt();
 283                 /*
 284                  * TBD: Can't get time duration while in C1, as resumes
 285                  *      go to an ISR rather than here.  Need to instrument
 286                  *      base interrupt handler.
 287                  */
 288                 sleep_ticks = 0xFFFFFFFF;
 289                 break;
 290
 291         case ACPI_STATE_C2:
 292                 /* Get start time (ticks) */
 293                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
 294                 /* Invoke C2 */
 295                 inb(cx->address);
 296                 /* Dummy op - must do something useless after P_LVL2 read */
 297                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 298                 /* Get end time (ticks) */
 299                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 300                 /* Re-enable interrupts */
 301                 local_irq_enable();
 302                 /* Compute time (ticks) that we were actually asleep */
 303                 sleep_ticks = ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
 304                 break;
 305
 306         case ACPI_STATE_C3:
 307
 308                 if (pr->flags.bm_check) {
 309                         if (atomic_inc_return(&c3_cpu_count) ==
 310                                         num_online_cpus()) {
 311                                 /*
 312                                  * All CPUs are trying to go to C3
 313                                  * Disable bus master arbitration
 314                                  */
 315                                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1,
 316                                         ACPI_MTX_DO_NOT_LOCK);
 317                         }
 318                 } else {
 319                         /* SMP with no shared cache... Invalidate cache  */
 320                         ACPI_FLUSH_CPU_CACHE();
 321                 }
 322
 323                 /* Get start time (ticks) */
 324                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
 325                 /* Invoke C3 */
 326                 inb(cx->address);
 327                 /* Dummy op - must do something useless after P_LVL3 read */
 328                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 329                 /* Get end time (ticks) */
 330                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 331                 if (pr->flags.bm_check) {
 332                         /* Enable bus master arbitration */
 333                         atomic_dec(&c3_cpu_count);
 334                         acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0, ACPI_MTX_DO_NOT_LOCK);
 335                 }
 336
 337                 /* Re-enable interrupts */
 338                 local_irq_enable();
 339                 /* Compute time (ticks) that we were actually asleep */
 340                 sleep_ticks = ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
 341                 break;
 342
 343         default:
 344                 local_irq_enable();
 345                 return;
 346         }
 347
 348         next_state = pr->power.state;
 349
 350         /*
 351          * Promotion?
 352          * ----------
 353          * Track the number of longs (time asleep is greater than threshold)
 354          * and promote when the count threshold is reached.  Note that bus
 355          * mastering activity may prevent promotions.
 356          * Do not promote above max_cstate.
 357          */
 358         if (cx->promotion.state &&
 359             ((cx->promotion.state - pr->power.states) <= max_cstate)) {
 360                 if (sleep_ticks > cx->promotion.threshold.ticks) {
 361                         cx->promotion.count++;
 362                         cx->demotion.count = 0;
 363                         if (cx->promotion.count >= cx->promotion.threshold.count) {
 364                                 if (pr->flags.bm_check) {
 365                                         if (!(pr->power.bm_activity & cx->promotion.threshold.bm)) {
 366                                                 next_state = cx->promotion.state;
 367                                                 goto end;
 368                                         }
 369                                 }
 370                                 else {
 371                                         next_state = cx->promotion.state;
 372                                         goto end;
 373                                 }
 374                         }
 375                 }
 376         }
 377
 378         /*
 379          * Demotion?
 380          * ---------
 381          * Track the number of shorts (time asleep is less than time threshold)
 382          * and demote when the usage threshold is reached.
 383          */
 384         if (cx->demotion.state) {
 385                 if (sleep_ticks < cx->demotion.threshold.ticks) {
 386                         cx->demotion.count++;
 387                         cx->promotion.count = 0;
 388                         if (cx->demotion.count >= cx->demotion.threshold.count) {
 389                                 next_state = cx->demotion.state;
 390                                 goto end;
 391                         }
 392                 }
 393         }
 394
 395 end:
 396         /*
 397          * Demote if current state exceeds max_cstate
 398          */
 399         if ((pr->power.state - pr->power.states) > max_cstate) {
 400                 if (cx->demotion.state)
 401                         next_state = cx->demotion.state;
 402         }
 403
 404         /*
 405          * New Cx State?
 406          * -------------
 407          * If we're going to start using a new Cx state we must clean up
 408          * from the previous and prepare to use the new.
 409          */
 410         if (next_state != pr->power.state)
 411                 acpi_processor_power_activate(pr, next_state);
 412
 413         return;
 414
 415  easy_out:
 416         /* do C1 instead of busy loop */
 417         if (pm_idle_save)
 418                 pm_idle_save();
 419         else
 420                 safe_halt();
 421         return;
 422 }
 423
 424
 425 static int
 426 acpi_processor_set_power_policy (
 427         struct acpi_processor   *pr)
 428 {
 429         unsigned int i;
 430         unsigned int state_is_set = 0;
 431         struct acpi_processor_cx *lower = NULL;
 432         struct acpi_processor_cx *higher = NULL;
 433         struct acpi_processor_cx *cx;
 434
 435         ACPI_FUNCTION_TRACE("acpi_processor_set_power_policy");
 436
 437         if (!pr)
 438                 return_VALUE(-EINVAL);
 439
 440         /*
 441          * This function sets the default Cx state policy (OS idle handler).
 442          * Our scheme is to promote quickly to C2 but more conservatively
 443          * to C3.  We're favoring C2  for its characteristics of low latency
 444          * (quick response), good power savings, and ability to allow bus
 445          * mastering activity.  Note that the Cx state policy is completely
 446          * customizable and can be altered dynamically.
 447          */
 448
 449         /* startup state */
 450         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 451                 cx = &pr->power.states[i];
 452                 if (!cx->valid)
 453                         continue;
 454
 455                 if (!state_is_set)
 456                         pr->power.state = cx;
 457                 state_is_set++;
 458                 break;
 459         }
 460
 461         if (!state_is_set)
 462                 return_VALUE(-ENODEV);
 463
 464         /* demotion */
 465         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 466                 cx = &pr->power.states[i];
 467                 if (!cx->valid)
 468                         continue;
 469
 470                 if (lower) {
 471                         cx->demotion.state = lower;
 472                         cx->demotion.threshold.ticks = cx->latency_ticks;
 473                         cx->demotion.threshold.count = 1;
 474                         if (cx->type == ACPI_STATE_C3)
 475                                 cx->demotion.threshold.bm = bm_history;
 476                 }
 477
 478                 lower = cx;
 479         }
 480
 481         /* promotion */
 482         for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
 483                 cx = &pr->power.states[i];
 484                 if (!cx->valid)
 485                         continue;
 486
 487                 if (higher) {
 488                         cx->promotion.state  = higher;
 489                         cx->promotion.threshold.ticks = cx->latency_ticks;
 490                         if (cx->type >= ACPI_STATE_C2)
 491                                 cx->promotion.threshold.count = 4;
 492                         else
 493                                 cx->promotion.threshold.count = 10;
 494                         if (higher->type == ACPI_STATE_C3)
 495                                 cx->promotion.threshold.bm = bm_history;
 496                 }
 497
 498                 higher = cx;
 499         }
 500
 501         return_VALUE(0);
 502 }
 503
 504
 505 static int acpi_processor_get_power_info_fadt (struct acpi_processor *pr)
 506 {
 507         int i;
 508
 509         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_fadt");
 510
 511         if (!pr)
 512                 return_VALUE(-EINVAL);
 513
 514         if (!pr->pblk)
 515                 return_VALUE(-ENODEV);
 516
 517         for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
 518                 memset(pr->power.states, 0, sizeof(struct acpi_processor_cx));
 519
 520         /* if info is obtained from pblk/fadt, type equals state */
 521         pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 522         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 523         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 524
 525         /* the C0 state only exists as a filler in our array,
 526          * and all processors need to support C1 */
 527         pr->power.states[ACPI_STATE_C0].valid = 1;
 528         pr->power.states[ACPI_STATE_C1].valid = 1;
 529
 530         /* determine C2 and C3 address from pblk */
 531         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 532         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 533
 534         /* determine latencies from FADT */
 535         pr->power.states[ACPI_STATE_C2].latency = acpi_fadt.plvl2_lat;
 536         pr->power.states[ACPI_STATE_C3].latency = acpi_fadt.plvl3_lat;
 537
 538         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 539                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 540                           pr->power.states[ACPI_STATE_C2].address,
 541                           pr->power.states[ACPI_STATE_C3].address));
 542
 543         return_VALUE(0);
 544 }
 545
 546
 547 static int acpi_processor_get_power_info_default_c1 (struct acpi_processor *pr)
 548 {
 549         int i;
 550
 551         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_default_c1");
 552
 553         for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
 554                 memset(&(pr->power.states[i]), 0,
 555                        sizeof(struct acpi_processor_cx));
 556
 557         /* if info is obtained from pblk/fadt, type equals state */
 558         pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 559         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 560         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 561
 562         /* the C0 state only exists as a filler in our array,
 563          * and all processors need to support C1 */
 564         pr->power.states[ACPI_STATE_C0].valid = 1;
 565         pr->power.states[ACPI_STATE_C1].valid = 1;
 566
 567         return_VALUE(0);
 568 }
 569
 570
 571 static int acpi_processor_get_power_info_cst (struct acpi_processor *pr)
 572 {
 573         acpi_status             status = 0;
 574         acpi_integer            count;
 575         int                     i;
 576         struct acpi_buffer      buffer = {ACPI_ALLOCATE_BUFFER, NULL};
 577         union acpi_object       *cst;
 578
 579         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_cst");
 580
 581         if (nocst)
 582                 return_VALUE(-ENODEV);
 583
 584         pr->power.count = 0;
 585         for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
 586                 memset(&(pr->power.states[i]), 0,
 587                        sizeof(struct acpi_processor_cx));
 588
 589         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 590         if (ACPI_FAILURE(status)) {
 591                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 592                 return_VALUE(-ENODEV);
 593         }
 594
 595         cst = (union acpi_object *) buffer.pointer;
 596
 597         /* There must be at least 2 elements */
 598         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 599                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "not enough elements in _CST\n"));
 600                 status = -EFAULT;
 601                 goto end;
 602         }
 603
 604         count = cst->package.elements[0].integer.value;
 605
 606         /* Validate number of power states. */
 607         if (count < 1 || count != cst->package.count - 1) {
 608                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR, "count given by _CST is not valid\n"));
 609                 status = -EFAULT;
 610                 goto end;
 611         }
 612
 613         /* We support up to ACPI_PROCESSOR_MAX_POWER. */
 614         if (count > ACPI_PROCESSOR_MAX_POWER) {
 615                 printk(KERN_WARNING "Limiting number of power states to max (%d)\n", ACPI_PROCESSOR_MAX_POWER);
 616                 printk(KERN_WARNING "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 617                 count = ACPI_PROCESSOR_MAX_POWER;
 618         }
 619
 620         /* Tell driver that at least _CST is supported. */
 621         pr->flags.has_cst = 1;
 622
 623         for (i = 1; i <= count; i++) {
 624                 union acpi_object *element;
 625                 union acpi_object *obj;
 626                 struct acpi_power_register *reg;
 627                 struct acpi_processor_cx cx;
 628
 629                 memset(&cx, 0, sizeof(cx));
 630
 631                 element = (union acpi_object *) &(cst->package.elements[i]);
 632                 if (element->type != ACPI_TYPE_PACKAGE)
 633                         continue;
 634
 635                 if (element->package.count != 4)
 636                         continue;
 637
 638                 obj = (union acpi_object *) &(element->package.elements[0]);
 639
 640                 if (obj->type != ACPI_TYPE_BUFFER)
 641                         continue;
 642
 643                 reg = (struct acpi_power_register *) obj->buffer.pointer;
 644
 645                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 646                         (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 647                         continue;
 648
 649                 cx.address = (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) ?
 650                         0 : reg->address;
 651
 652                 /* There should be an easy way to extract an integer... */
 653                 obj = (union acpi_object *) &(element->package.elements[1]);
 654                 if (obj->type != ACPI_TYPE_INTEGER)
 655                         continue;
 656
 657                 cx.type = obj->integer.value;
 658
 659                 if ((cx.type != ACPI_STATE_C1) &&
 660                     (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO))
 661                         continue;
 662
 663                 if ((cx.type < ACPI_STATE_C1) ||
 664                     (cx.type > ACPI_STATE_C3))
 665                         continue;
 666
 667                 obj = (union acpi_object *) &(element->package.elements[2]);
 668                 if (obj->type != ACPI_TYPE_INTEGER)
 669                         continue;
 670
 671                 cx.latency = obj->integer.value;
 672
 673                 obj = (union acpi_object *) &(element->package.elements[3]);
 674                 if (obj->type != ACPI_TYPE_INTEGER)
 675                         continue;
 676
 677                 cx.power = obj->integer.value;
 678
 679                 (pr->power.count)++;
 680                 memcpy(&(pr->power.states[pr->power.count]), &cx, sizeof(cx));
 681         }
 682
 683         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n", pr->power.count));
 684
 685         /* Validate number of power states discovered */
 686         if (pr->power.count < 2)
 687                 status = -ENODEV;
 688
 689 end:
 690         acpi_os_free(buffer.pointer);
 691
 692         return_VALUE(status);
 693 }
 694
 695
 696 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 697 {
 698         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c2");
 699
 700         if (!cx->address)
 701                 return_VOID;
 702
 703         /*
 704          * C2 latency must be less than or equal to 100
 705          * microseconds.
 706          */
 707         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 708                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 709                                   "latency too large [%d]\n",
 710                                   cx->latency));
 711                 return_VOID;
 712         }
 713
 714         /*
 715          * Otherwise we've met all of our C2 requirements.
 716          * Normalize the C2 latency to expidite policy
 717          */
 718         cx->valid = 1;
 719         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 720
 721         return_VOID;
 722 }
 723
 724
 725 static void acpi_processor_power_verify_c3(
 726         struct acpi_processor *pr,
 727         struct acpi_processor_cx *cx)
 728 {
 729         static int bm_check_flag;
 730
 731         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c3");
 732
 733         if (!cx->address)
 734                 return_VOID;
 735
 736         /*
 737          * C3 latency must be less than or equal to 1000
 738          * microseconds.
 739          */
 740         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 741                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 742                                   "latency too large [%d]\n",
 743                                   cx->latency));
 744                 return_VOID;
 745         }
 746
 747         /*
 748          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 749          * DMA transfers are used by any ISA device to avoid livelock.
 750          * Note that we could disable Type-F DMA (as recommended by
 751          * the erratum), but this is known to disrupt certain ISA
 752          * devices thus we take the conservative approach.
 753          */
 754         else if (errata.piix4.fdma) {
 755                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 756                         "C3 not supported on PIIX4 with Type-F DMA\n"));
 757                 return_VOID;
 758         }
 759
 760         /* All the logic here assumes flags.bm_check is same across all CPUs */
 761         if (!bm_check_flag) {
 762                 /* Determine whether bm_check is needed based on CPU  */
 763                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 764                 bm_check_flag = pr->flags.bm_check;
 765         } else {
 766                 pr->flags.bm_check = bm_check_flag;
 767         }
 768
 769         if (pr->flags.bm_check) {
 770                 /* bus mastering control is necessary */
 771                 if (!pr->flags.bm_control) {
 772                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 773                           "C3 support requires bus mastering control\n"));
 774                         return_VOID;
 775                 }
 776         } else {
 777                 /*
 778                  * WBINVD should be set in fadt, for C3 state to be
 779                  * supported on when bm_check is not required.
 780                  */
 781                 if (acpi_fadt.wb_invd != 1) {
 782                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 783                           "Cache invalidation should work properly"
 784                           " for C3 to be enabled on SMP systems\n"));
 785                         return_VOID;
 786                 }
 787                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD,
 788                                 0, ACPI_MTX_DO_NOT_LOCK);
 789         }
 790
 791         /*
 792          * Otherwise we've met all of our C3 requirements.
 793          * Normalize the C3 latency to expidite policy.  Enable
 794          * checking of bus mastering status (bm_check) so we can
 795          * use this in our C3 policy
 796          */
 797         cx->valid = 1;
 798         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 799
 800         return_VOID;
 801 }
 802
 803
 804 static int acpi_processor_power_verify(struct acpi_processor *pr)
 805 {
 806         unsigned int i;
 807         unsigned int working = 0;
 808
 809         for (i=1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 810                 struct acpi_processor_cx *cx = &pr->power.states[i];
 811
 812                 switch (cx->type) {
 813                 case ACPI_STATE_C1:
 814                         cx->valid = 1;
 815                         break;
 816
 817                 case ACPI_STATE_C2:
 818                         acpi_processor_power_verify_c2(cx);
 819                         break;
 820
 821                 case ACPI_STATE_C3:
 822                         acpi_processor_power_verify_c3(pr, cx);
 823                         break;
 824                 }
 825
 826                 if (cx->valid)
 827                         working++;
 828         }
 829
 830         return (working);
 831 }
 832
 833 static int acpi_processor_get_power_info (
 834         struct acpi_processor   *pr)
 835 {
 836         unsigned int i;
 837         int result;
 838
 839         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info");
 840
 841         /* NOTE: the idle thread may not be running while calling
 842          * this function */
 843
 844         result = acpi_processor_get_power_info_cst(pr);
 845         if ((result) || (acpi_processor_power_verify(pr) < 2)) {
 846                 result = acpi_processor_get_power_info_fadt(pr);
 847                 if ((result) || (acpi_processor_power_verify(pr) < 2))
 848                         result = acpi_processor_get_power_info_default_c1(pr);
 849         }
 850
 851         /*
 852          * Set Default Policy
 853          * ------------------
 854          * Now that we know which states are supported, set the default
 855          * policy.  Note that this policy can be changed dynamically
 856          * (e.g. encourage deeper sleeps to conserve battery life when
 857          * not on AC).
 858          */
 859         result = acpi_processor_set_power_policy(pr);
 860         if (result)
 861                 return_VALUE(result);
 862
 863         /*
 864          * if one state of type C2 or C3 is available, mark this
 865          * CPU as being "idle manageable"
 866          */
 867         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 868                 if (pr->power.states[i].valid) {
 869                         pr->power.count = i;
 870                         pr->flags.power = 1;
 871                 }
 872         }
 873
 874         return_VALUE(0);
 875 }
 876
 877 int acpi_processor_cst_has_changed (struct acpi_processor *pr)
 878 {
 879         int                     result = 0;
 880
 881         ACPI_FUNCTION_TRACE("acpi_processor_cst_has_changed");
 882
 883         if (!pr)
 884                 return_VALUE(-EINVAL);
 885
 886         if ( nocst) {
 887                 return_VALUE(-ENODEV);
 888         }
 889
 890         if (!pr->flags.power_setup_done)
 891                 return_VALUE(-ENODEV);
 892
 893         /* Fall back to the default idle loop */
 894         pm_idle = pm_idle_save;
 895         synchronize_sched();  /* Relies on interrupts forcing exit from idle. */
 896
 897         pr->flags.power = 0;
 898         result = acpi_processor_get_power_info(pr);
 899         if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
 900                 pm_idle = acpi_processor_idle;
 901
 902         return_VALUE(result);
 903 }
 904
 905 /* proc interface */
 906
 907 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 908 {
 909         struct acpi_processor   *pr = (struct acpi_processor *)seq->private;
 910         unsigned int            i;
 911
 912         ACPI_FUNCTION_TRACE("acpi_processor_power_seq_show");
 913
 914         if (!pr)
 915                 goto end;
 916
 917         seq_printf(seq, "active state:            C%zd\n"
 918                         "max_cstate:              C%d\n"
 919                         "bus master activity:     %08x\n",
 920                         pr->power.state ? pr->power.state - pr->power.states : 0,
 921                         max_cstate,
 922                         (unsigned)pr->power.bm_activity);
 923
 924         seq_puts(seq, "states:\n");
 925
 926         for (i = 1; i <= pr->power.count; i++) {
 927                 seq_printf(seq, "   %cC%d:                  ",
 928                         (&pr->power.states[i] == pr->power.state?'*':' '), i);
 929
 930                 if (!pr->power.states[i].valid) {
 931                         seq_puts(seq, "<not supported>\n");
 932                         continue;
 933                 }
 934
 935                 switch (pr->power.states[i].type) {
 936                 case ACPI_STATE_C1:
 937                         seq_printf(seq, "type[C1] ");
 938                         break;
 939                 case ACPI_STATE_C2:
 940                         seq_printf(seq, "type[C2] ");
 941                         break;
 942                 case ACPI_STATE_C3:
 943                         seq_printf(seq, "type[C3] ");
 944                         break;
 945                 default:
 946                         seq_printf(seq, "type[--] ");
 947                         break;
 948                 }
 949
 950                 if (pr->power.states[i].promotion.state)
 951                         seq_printf(seq, "promotion[C%zd] ",
 952                                 (pr->power.states[i].promotion.state -
 953                                  pr->power.states));
 954                 else
 955                         seq_puts(seq, "promotion[--] ");
 956
 957                 if (pr->power.states[i].demotion.state)
 958                         seq_printf(seq, "demotion[C%zd] ",
 959                                 (pr->power.states[i].demotion.state -
 960                                  pr->power.states));
 961                 else
 962                         seq_puts(seq, "demotion[--] ");
 963
 964                 seq_printf(seq, "latency[%03d] usage[%08d]\n",
 965                         pr->power.states[i].latency,
 966                         pr->power.states[i].usage);
 967         }
 968
 969 end:
 970         return_VALUE(0);
 971 }
 972
 973 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
 974 {
 975         return single_open(file, acpi_processor_power_seq_show,
 976                                                 PDE(inode)->data);
 977 }
 978
 979 static struct file_operations acpi_processor_power_fops = {
 980         .open           = acpi_processor_power_open_fs,
 981         .read           = seq_read,
 982         .llseek         = seq_lseek,
 983         .release        = single_release,
 984 };
 985
 986 int acpi_processor_power_init(struct acpi_processor *pr, struct acpi_device *device)
 987 {
 988         acpi_status             status = 0;
 989         static int              first_run = 0;
 990         struct proc_dir_entry   *entry = NULL;
 991         unsigned int i;
 992
 993         ACPI_FUNCTION_TRACE("acpi_processor_power_init");
 994
 995         if (!first_run) {
 996                 dmi_check_system(processor_power_dmi_table);
 997                 if (max_cstate < ACPI_C_STATES_MAX)
 998                         printk(KERN_NOTICE "ACPI: processor limited to max C-state %d\n", max_cstate);
 999                 first_run++;
1000         }
1001
1002         if (!pr)
1003                 return_VALUE(-EINVAL);
1004
1005         if (acpi_fadt.cst_cnt && !nocst) {
1006                 status = acpi_os_write_port(acpi_fadt.smi_cmd, acpi_fadt.cst_cnt, 8);
1007                 if (ACPI_FAILURE(status)) {
1008                         ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1009                                           "Notifying BIOS of _CST ability failed\n"));
1010                 }
1011         }
1012
1013         acpi_processor_power_init_pdc(&(pr->power), pr->id);
1014         acpi_processor_set_pdc(pr, pr->power.pdc);
1015         acpi_processor_get_power_info(pr);
1016
1017         /*
1018          * Install the idle handler if processor power management is supported.
1019          * Note that we use previously set idle handler will be used on
1020          * platforms that only support C1.
1021          */
1022         if ((pr->flags.power) && (!boot_option_idle_override)) {
1023                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1024                 for (i = 1; i <= pr->power.count; i++)
1025                         if (pr->power.states[i].valid)
1026                                 printk(" C%d[C%d]", i, pr->power.states[i].type);
1027                 printk(")\n");
1028
1029                 if (pr->id == 0) {
1030                         pm_idle_save = pm_idle;
1031                         pm_idle = acpi_processor_idle;
1032                 }
1033         }
1034
1035         /* 'power' [R] */
1036         entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1037                 S_IRUGO, acpi_device_dir(device));
1038         if (!entry)
1039                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1040                         "Unable to create '%s' fs entry\n",
1041                         ACPI_PROCESSOR_FILE_POWER));
1042         else {
1043                 entry->proc_fops = &acpi_processor_power_fops;
1044                 entry->data = acpi_driver_data(device);
1045                 entry->owner = THIS_MODULE;
1046         }
1047
1048         pr->flags.power_setup_done = 1;
1049
1050         return_VALUE(0);
1051 }
1052
1053 int acpi_processor_power_exit(struct acpi_processor *pr, struct acpi_device *device)
1054 {
1055         ACPI_FUNCTION_TRACE("acpi_processor_power_exit");
1056
1057         pr->flags.power_setup_done = 0;
1058
1059         if (acpi_device_dir(device))
1060                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,acpi_device_dir(device));
1061
1062         /* Unregister the idle handler when processor #0 is removed. */
1063         if (pr->id == 0) {
1064                 pm_idle = pm_idle_save;
1065
1066                 /*
1067                  * We are about to unload the current idle thread pm callback
1068                  * (pm_idle), Wait for all processors to update cached/local
1069                  * copies of pm_idle before proceeding.
1070                  */
1071                 cpu_idle_wait();
1072         }
1073
1074         return_VALUE(0);
1075 }