git.oblomov.eu Git - linux-2.6/blob - 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 #include <linux/sched.h>        /* need_resched() */
  41
  42 #include <asm/io.h>
  43 #include <asm/uaccess.h>
  44
  45 #include <acpi/acpi_bus.h>
  46 #include <acpi/processor.h>
  47
  48 #define ACPI_PROCESSOR_COMPONENT        0x01000000
  49 #define ACPI_PROCESSOR_CLASS            "processor"
  50 #define ACPI_PROCESSOR_DRIVER_NAME      "ACPI Processor Driver"
  51 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  52 ACPI_MODULE_NAME("acpi_processor")
  53 #define ACPI_PROCESSOR_FILE_POWER       "power"
  54 #define US_TO_PM_TIMER_TICKS(t)         ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
  55 #define C2_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  56 #define C3_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  57 static void (*pm_idle_save) (void);
  58 module_param(max_cstate, uint, 0644);
  59
  60 static unsigned int nocst = 0;
  61 module_param(nocst, uint, 0000);
  62
  63 /*
  64  * bm_history -- bit-mask with a bit per jiffy of bus-master activity
  65  * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
  66  * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
  67  * 100 HZ: 0x0000000F: 4 jiffies = 40ms
  68  * reduce history for more aggressive entry into C3
  69  */
  70 static unsigned int bm_history =
  71     (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
  72 module_param(bm_history, uint, 0644);
  73 /* --------------------------------------------------------------------------
  74                                 Power Management
  75    -------------------------------------------------------------------------- */
  76
  77 /*
  78  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  79  * For now disable this. Probably a bug somewhere else.
  80  *
  81  * To skip this limit, boot/load with a large max_cstate limit.
  82  */
  83 static int set_max_cstate(struct dmi_system_id *id)
  84 {
  85         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  86                 return 0;
  87
  88         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  89                " Override with \"processor.max_cstate=%d\"\n", id->ident,
  90                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  91
  92         max_cstate = (long)id->driver_data;
  93
  94         return 0;
  95 }
  96
  97 static struct dmi_system_id __initdata processor_power_dmi_table[] = {
  98         { set_max_cstate, "IBM ThinkPad R40e", {
  99           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 100           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW")}, (void *)1},
 101         { set_max_cstate, "IBM ThinkPad R40e", {
 102           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 103           DMI_MATCH(DMI_BIOS_VERSION,"1SET43WW") }, (void*)1},
 104         { set_max_cstate, "IBM ThinkPad R40e", {
 105           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 106           DMI_MATCH(DMI_BIOS_VERSION,"1SET45WW") }, (void*)1},
 107         { set_max_cstate, "IBM ThinkPad R40e", {
 108           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 109           DMI_MATCH(DMI_BIOS_VERSION,"1SET47WW") }, (void*)1},
 110         { set_max_cstate, "IBM ThinkPad R40e", {
 111           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 112           DMI_MATCH(DMI_BIOS_VERSION,"1SET50WW") }, (void*)1},
 113         { set_max_cstate, "IBM ThinkPad R40e", {
 114           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 115           DMI_MATCH(DMI_BIOS_VERSION,"1SET52WW") }, (void*)1},
 116         { set_max_cstate, "IBM ThinkPad R40e", {
 117           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 118           DMI_MATCH(DMI_BIOS_VERSION,"1SET55WW") }, (void*)1},
 119         { set_max_cstate, "IBM ThinkPad R40e", {
 120           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 121           DMI_MATCH(DMI_BIOS_VERSION,"1SET56WW") }, (void*)1},
 122         { set_max_cstate, "IBM ThinkPad R40e", {
 123           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 124           DMI_MATCH(DMI_BIOS_VERSION,"1SET59WW") }, (void*)1},
 125         { set_max_cstate, "IBM ThinkPad R40e", {
 126           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 127           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW") }, (void*)1},
 128         { set_max_cstate, "IBM ThinkPad R40e", {
 129           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 130           DMI_MATCH(DMI_BIOS_VERSION,"1SET61WW") }, (void*)1},
 131         { set_max_cstate, "IBM ThinkPad R40e", {
 132           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 133           DMI_MATCH(DMI_BIOS_VERSION,"1SET62WW") }, (void*)1},
 134         { set_max_cstate, "IBM ThinkPad R40e", {
 135           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 136           DMI_MATCH(DMI_BIOS_VERSION,"1SET64WW") }, (void*)1},
 137         { set_max_cstate, "IBM ThinkPad R40e", {
 138           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 139           DMI_MATCH(DMI_BIOS_VERSION,"1SET65WW") }, (void*)1},
 140         { set_max_cstate, "IBM ThinkPad R40e", {
 141           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 142           DMI_MATCH(DMI_BIOS_VERSION,"1SET68WW") }, (void*)1},
 143         { set_max_cstate, "Medion 41700", {
 144           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 145           DMI_MATCH(DMI_BIOS_VERSION,"R01-A1J")}, (void *)1},
 146         { set_max_cstate, "Clevo 5600D", {
 147           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 148           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 149          (void *)2},
 150         {},
 151 };
 152
 153 static inline u32 ticks_elapsed(u32 t1, u32 t2)
 154 {
 155         if (t2 >= t1)
 156                 return (t2 - t1);
 157         else if (!acpi_fadt.tmr_val_ext)
 158                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 159         else
 160                 return ((0xFFFFFFFF - t1) + t2);
 161 }
 162
 163 static void
 164 acpi_processor_power_activate(struct acpi_processor *pr,
 165                               struct acpi_processor_cx *new)
 166 {
 167         struct acpi_processor_cx *old;
 168
 169         if (!pr || !new)
 170                 return;
 171
 172         old = pr->power.state;
 173
 174         if (old)
 175                 old->promotion.count = 0;
 176         new->demotion.count = 0;
 177
 178         /* Cleanup from old state. */
 179         if (old) {
 180                 switch (old->type) {
 181                 case ACPI_STATE_C3:
 182                         /* Disable bus master reload */
 183                         if (new->type != ACPI_STATE_C3 && pr->flags.bm_check)
 184                                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0,
 185                                                   ACPI_MTX_DO_NOT_LOCK);
 186                         break;
 187                 }
 188         }
 189
 190         /* Prepare to use new state. */
 191         switch (new->type) {
 192         case ACPI_STATE_C3:
 193                 /* Enable bus master reload */
 194                 if (old->type != ACPI_STATE_C3 && pr->flags.bm_check)
 195                         acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1,
 196                                           ACPI_MTX_DO_NOT_LOCK);
 197                 break;
 198         }
 199
 200         pr->power.state = new;
 201
 202         return;
 203 }
 204
 205 static void acpi_safe_halt(void)
 206 {
 207         clear_thread_flag(TIF_POLLING_NRFLAG);
 208         smp_mb__after_clear_bit();
 209         if (!need_resched())
 210                 safe_halt();
 211         set_thread_flag(TIF_POLLING_NRFLAG);
 212 }
 213
 214 static atomic_t c3_cpu_count;
 215
 216 static void acpi_processor_idle(void)
 217 {
 218         struct acpi_processor *pr = NULL;
 219         struct acpi_processor_cx *cx = NULL;
 220         struct acpi_processor_cx *next_state = NULL;
 221         int sleep_ticks = 0;
 222         u32 t1, t2 = 0;
 223
 224         pr = processors[smp_processor_id()];
 225         if (!pr)
 226                 return;
 227
 228         /*
 229          * Interrupts must be disabled during bus mastering calculations and
 230          * for C2/C3 transitions.
 231          */
 232         local_irq_disable();
 233
 234         /*
 235          * Check whether we truly need to go idle, or should
 236          * reschedule:
 237          */
 238         if (unlikely(need_resched())) {
 239                 local_irq_enable();
 240                 return;
 241         }
 242
 243         cx = pr->power.state;
 244         if (!cx) {
 245                 if (pm_idle_save)
 246                         pm_idle_save();
 247                 else
 248                         acpi_safe_halt();
 249                 return;
 250         }
 251
 252         /*
 253          * Check BM Activity
 254          * -----------------
 255          * Check for bus mastering activity (if required), record, and check
 256          * for demotion.
 257          */
 258         if (pr->flags.bm_check) {
 259                 u32 bm_status = 0;
 260                 unsigned long diff = jiffies - pr->power.bm_check_timestamp;
 261
 262                 if (diff > 32)
 263                         diff = 32;
 264
 265                 while (diff) {
 266                         /* if we didn't get called, assume there was busmaster activity */
 267                         diff--;
 268                         if (diff)
 269                                 pr->power.bm_activity |= 0x1;
 270                         pr->power.bm_activity <<= 1;
 271                 }
 272
 273                 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS,
 274                                   &bm_status, ACPI_MTX_DO_NOT_LOCK);
 275                 if (bm_status) {
 276                         pr->power.bm_activity++;
 277                         acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS,
 278                                           1, ACPI_MTX_DO_NOT_LOCK);
 279                 }
 280                 /*
 281                  * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 282                  * the true state of bus mastering activity; forcing us to
 283                  * manually check the BMIDEA bit of each IDE channel.
 284                  */
 285                 else if (errata.piix4.bmisx) {
 286                         if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 287                             || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 288                                 pr->power.bm_activity++;
 289                 }
 290
 291                 pr->power.bm_check_timestamp = jiffies;
 292
 293                 /*
 294                  * Apply bus mastering demotion policy.  Automatically demote
 295                  * to avoid a faulty transition.  Note that the processor
 296                  * won't enter a low-power state during this call (to this
 297                  * funciton) but should upon the next.
 298                  *
 299                  * TBD: A better policy might be to fallback to the demotion
 300                  *      state (use it for this quantum only) istead of
 301                  *      demoting -- and rely on duration as our sole demotion
 302                  *      qualification.  This may, however, introduce DMA
 303                  *      issues (e.g. floppy DMA transfer overrun/underrun).
 304                  */
 305                 if (pr->power.bm_activity & cx->demotion.threshold.bm) {
 306                         local_irq_enable();
 307                         next_state = cx->demotion.state;
 308                         goto end;
 309                 }
 310         }
 311
 312 #ifdef CONFIG_HOTPLUG_CPU
 313         /*
 314          * Check for P_LVL2_UP flag before entering C2 and above on
 315          * an SMP system. We do it here instead of doing it at _CST/P_LVL
 316          * detection phase, to work cleanly with logical CPU hotplug.
 317          */
 318         if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 319             !pr->flags.has_cst && !acpi_fadt.plvl2_up)
 320                 cx = &pr->power.states[ACPI_STATE_C1];
 321 #endif
 322
 323         cx->usage++;
 324
 325         /*
 326          * Sleep:
 327          * ------
 328          * Invoke the current Cx state to put the processor to sleep.
 329          */
 330         if (cx->type == ACPI_STATE_C2 || cx->type == ACPI_STATE_C3) {
 331                 clear_thread_flag(TIF_POLLING_NRFLAG);
 332                 smp_mb__after_clear_bit();
 333                 if (need_resched()) {
 334                         set_thread_flag(TIF_POLLING_NRFLAG);
 335                         local_irq_enable();
 336                         return;
 337                 }
 338         }
 339
 340         switch (cx->type) {
 341
 342         case ACPI_STATE_C1:
 343                 /*
 344                  * Invoke C1.
 345                  * Use the appropriate idle routine, the one that would
 346                  * be used without acpi C-states.
 347                  */
 348                 if (pm_idle_save)
 349                         pm_idle_save();
 350                 else
 351                         acpi_safe_halt();
 352
 353                 /*
 354                  * TBD: Can't get time duration while in C1, as resumes
 355                  *      go to an ISR rather than here.  Need to instrument
 356                  *      base interrupt handler.
 357                  */
 358                 sleep_ticks = 0xFFFFFFFF;
 359                 break;
 360
 361         case ACPI_STATE_C2:
 362                 /* Get start time (ticks) */
 363                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
 364                 /* Invoke C2 */
 365                 inb(cx->address);
 366                 /* Dummy op - must do something useless after P_LVL2 read */
 367                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 368                 /* Get end time (ticks) */
 369                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 370                 /* Re-enable interrupts */
 371                 local_irq_enable();
 372                 set_thread_flag(TIF_POLLING_NRFLAG);
 373                 /* Compute time (ticks) that we were actually asleep */
 374                 sleep_ticks =
 375                     ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
 376                 break;
 377
 378         case ACPI_STATE_C3:
 379
 380                 if (pr->flags.bm_check) {
 381                         if (atomic_inc_return(&c3_cpu_count) ==
 382                             num_online_cpus()) {
 383                                 /*
 384                                  * All CPUs are trying to go to C3
 385                                  * Disable bus master arbitration
 386                                  */
 387                                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1,
 388                                                   ACPI_MTX_DO_NOT_LOCK);
 389                         }
 390                 } else {
 391                         /* SMP with no shared cache... Invalidate cache  */
 392                         ACPI_FLUSH_CPU_CACHE();
 393                 }
 394
 395                 /* Get start time (ticks) */
 396                 t1 = inl(acpi_fadt.xpm_tmr_blk.address);
 397                 /* Invoke C3 */
 398                 inb(cx->address);
 399                 /* Dummy op - must do something useless after P_LVL3 read */
 400                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 401                 /* Get end time (ticks) */
 402                 t2 = inl(acpi_fadt.xpm_tmr_blk.address);
 403                 if (pr->flags.bm_check) {
 404                         /* Enable bus master arbitration */
 405                         atomic_dec(&c3_cpu_count);
 406                         acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0,
 407                                           ACPI_MTX_DO_NOT_LOCK);
 408                 }
 409
 410                 /* Re-enable interrupts */
 411                 local_irq_enable();
 412                 set_thread_flag(TIF_POLLING_NRFLAG);
 413                 /* Compute time (ticks) that we were actually asleep */
 414                 sleep_ticks =
 415                     ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
 416                 break;
 417
 418         default:
 419                 local_irq_enable();
 420                 return;
 421         }
 422
 423         next_state = pr->power.state;
 424
 425 #ifdef CONFIG_HOTPLUG_CPU
 426         /* Don't do promotion/demotion */
 427         if ((cx->type == ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 428             !pr->flags.has_cst && !acpi_fadt.plvl2_up) {
 429                 next_state = cx;
 430                 goto end;
 431         }
 432 #endif
 433
 434         /*
 435          * Promotion?
 436          * ----------
 437          * Track the number of longs (time asleep is greater than threshold)
 438          * and promote when the count threshold is reached.  Note that bus
 439          * mastering activity may prevent promotions.
 440          * Do not promote above max_cstate.
 441          */
 442         if (cx->promotion.state &&
 443             ((cx->promotion.state - pr->power.states) <= max_cstate)) {
 444                 if (sleep_ticks > cx->promotion.threshold.ticks) {
 445                         cx->promotion.count++;
 446                         cx->demotion.count = 0;
 447                         if (cx->promotion.count >=
 448                             cx->promotion.threshold.count) {
 449                                 if (pr->flags.bm_check) {
 450                                         if (!
 451                                             (pr->power.bm_activity & cx->
 452                                              promotion.threshold.bm)) {
 453                                                 next_state =
 454                                                     cx->promotion.state;
 455                                                 goto end;
 456                                         }
 457                                 } else {
 458                                         next_state = cx->promotion.state;
 459                                         goto end;
 460                                 }
 461                         }
 462                 }
 463         }
 464
 465         /*
 466          * Demotion?
 467          * ---------
 468          * Track the number of shorts (time asleep is less than time threshold)
 469          * and demote when the usage threshold is reached.
 470          */
 471         if (cx->demotion.state) {
 472                 if (sleep_ticks < cx->demotion.threshold.ticks) {
 473                         cx->demotion.count++;
 474                         cx->promotion.count = 0;
 475                         if (cx->demotion.count >= cx->demotion.threshold.count) {
 476                                 next_state = cx->demotion.state;
 477                                 goto end;
 478                         }
 479                 }
 480         }
 481
 482       end:
 483         /*
 484          * Demote if current state exceeds max_cstate
 485          */
 486         if ((pr->power.state - pr->power.states) > max_cstate) {
 487                 if (cx->demotion.state)
 488                         next_state = cx->demotion.state;
 489         }
 490
 491         /*
 492          * New Cx State?
 493          * -------------
 494          * If we're going to start using a new Cx state we must clean up
 495          * from the previous and prepare to use the new.
 496          */
 497         if (next_state != pr->power.state)
 498                 acpi_processor_power_activate(pr, next_state);
 499 }
 500
 501 static int acpi_processor_set_power_policy(struct acpi_processor *pr)
 502 {
 503         unsigned int i;
 504         unsigned int state_is_set = 0;
 505         struct acpi_processor_cx *lower = NULL;
 506         struct acpi_processor_cx *higher = NULL;
 507         struct acpi_processor_cx *cx;
 508
 509         ACPI_FUNCTION_TRACE("acpi_processor_set_power_policy");
 510
 511         if (!pr)
 512                 return_VALUE(-EINVAL);
 513
 514         /*
 515          * This function sets the default Cx state policy (OS idle handler).
 516          * Our scheme is to promote quickly to C2 but more conservatively
 517          * to C3.  We're favoring C2  for its characteristics of low latency
 518          * (quick response), good power savings, and ability to allow bus
 519          * mastering activity.  Note that the Cx state policy is completely
 520          * customizable and can be altered dynamically.
 521          */
 522
 523         /* startup state */
 524         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 525                 cx = &pr->power.states[i];
 526                 if (!cx->valid)
 527                         continue;
 528
 529                 if (!state_is_set)
 530                         pr->power.state = cx;
 531                 state_is_set++;
 532                 break;
 533         }
 534
 535         if (!state_is_set)
 536                 return_VALUE(-ENODEV);
 537
 538         /* demotion */
 539         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 540                 cx = &pr->power.states[i];
 541                 if (!cx->valid)
 542                         continue;
 543
 544                 if (lower) {
 545                         cx->demotion.state = lower;
 546                         cx->demotion.threshold.ticks = cx->latency_ticks;
 547                         cx->demotion.threshold.count = 1;
 548                         if (cx->type == ACPI_STATE_C3)
 549                                 cx->demotion.threshold.bm = bm_history;
 550                 }
 551
 552                 lower = cx;
 553         }
 554
 555         /* promotion */
 556         for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
 557                 cx = &pr->power.states[i];
 558                 if (!cx->valid)
 559                         continue;
 560
 561                 if (higher) {
 562                         cx->promotion.state = higher;
 563                         cx->promotion.threshold.ticks = cx->latency_ticks;
 564                         if (cx->type >= ACPI_STATE_C2)
 565                                 cx->promotion.threshold.count = 4;
 566                         else
 567                                 cx->promotion.threshold.count = 10;
 568                         if (higher->type == ACPI_STATE_C3)
 569                                 cx->promotion.threshold.bm = bm_history;
 570                 }
 571
 572                 higher = cx;
 573         }
 574
 575         return_VALUE(0);
 576 }
 577
 578 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 579 {
 580         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_fadt");
 581
 582         if (!pr)
 583                 return_VALUE(-EINVAL);
 584
 585         if (!pr->pblk)
 586                 return_VALUE(-ENODEV);
 587
 588         /* if info is obtained from pblk/fadt, type equals state */
 589         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 590         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 591
 592 #ifndef CONFIG_HOTPLUG_CPU
 593         /*
 594          * Check for P_LVL2_UP flag before entering C2 and above on
 595          * an SMP system.
 596          */
 597         if ((num_online_cpus() > 1) && !acpi_fadt.plvl2_up)
 598                 return_VALUE(-ENODEV);
 599 #endif
 600
 601         /* determine C2 and C3 address from pblk */
 602         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 603         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 604
 605         /* determine latencies from FADT */
 606         pr->power.states[ACPI_STATE_C2].latency = acpi_fadt.plvl2_lat;
 607         pr->power.states[ACPI_STATE_C3].latency = acpi_fadt.plvl3_lat;
 608
 609         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 610                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 611                           pr->power.states[ACPI_STATE_C2].address,
 612                           pr->power.states[ACPI_STATE_C3].address));
 613
 614         return_VALUE(0);
 615 }
 616
 617 static int acpi_processor_get_power_info_default_c1(struct acpi_processor *pr)
 618 {
 619         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_default_c1");
 620
 621         /* Zero initialize all the C-states info. */
 622         memset(pr->power.states, 0, sizeof(pr->power.states));
 623
 624         /* set the first C-State to C1 */
 625         pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 626
 627         /* the C0 state only exists as a filler in our array,
 628          * and all processors need to support C1 */
 629         pr->power.states[ACPI_STATE_C0].valid = 1;
 630         pr->power.states[ACPI_STATE_C1].valid = 1;
 631
 632         return_VALUE(0);
 633 }
 634
 635 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 636 {
 637         acpi_status status = 0;
 638         acpi_integer count;
 639         int current_count;
 640         int i;
 641         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 642         union acpi_object *cst;
 643
 644         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_cst");
 645
 646         if (nocst)
 647                 return_VALUE(-ENODEV);
 648
 649         current_count = 1;
 650
 651         /* Zero initialize C2 onwards and prepare for fresh CST lookup */
 652         for (i = 2; i < ACPI_PROCESSOR_MAX_POWER; i++)
 653                 memset(&(pr->power.states[i]), 0,
 654                                 sizeof(struct acpi_processor_cx));
 655
 656         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 657         if (ACPI_FAILURE(status)) {
 658                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 659                 return_VALUE(-ENODEV);
 660         }
 661
 662         cst = (union acpi_object *)buffer.pointer;
 663
 664         /* There must be at least 2 elements */
 665         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 666                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
 667                                   "not enough elements in _CST\n"));
 668                 status = -EFAULT;
 669                 goto end;
 670         }
 671
 672         count = cst->package.elements[0].integer.value;
 673
 674         /* Validate number of power states. */
 675         if (count < 1 || count != cst->package.count - 1) {
 676                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
 677                                   "count given by _CST is not valid\n"));
 678                 status = -EFAULT;
 679                 goto end;
 680         }
 681
 682         /* Tell driver that at least _CST is supported. */
 683         pr->flags.has_cst = 1;
 684
 685         for (i = 1; i <= count; i++) {
 686                 union acpi_object *element;
 687                 union acpi_object *obj;
 688                 struct acpi_power_register *reg;
 689                 struct acpi_processor_cx cx;
 690
 691                 memset(&cx, 0, sizeof(cx));
 692
 693                 element = (union acpi_object *)&(cst->package.elements[i]);
 694                 if (element->type != ACPI_TYPE_PACKAGE)
 695                         continue;
 696
 697                 if (element->package.count != 4)
 698                         continue;
 699
 700                 obj = (union acpi_object *)&(element->package.elements[0]);
 701
 702                 if (obj->type != ACPI_TYPE_BUFFER)
 703                         continue;
 704
 705                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 706
 707                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 708                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 709                         continue;
 710
 711                 cx.address = (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) ?
 712                     0 : reg->address;
 713
 714                 /* There should be an easy way to extract an integer... */
 715                 obj = (union acpi_object *)&(element->package.elements[1]);
 716                 if (obj->type != ACPI_TYPE_INTEGER)
 717                         continue;
 718
 719                 cx.type = obj->integer.value;
 720
 721                 if ((cx.type != ACPI_STATE_C1) &&
 722                     (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO))
 723                         continue;
 724
 725                 if ((cx.type < ACPI_STATE_C2) || (cx.type > ACPI_STATE_C3))
 726                         continue;
 727
 728                 obj = (union acpi_object *)&(element->package.elements[2]);
 729                 if (obj->type != ACPI_TYPE_INTEGER)
 730                         continue;
 731
 732                 cx.latency = obj->integer.value;
 733
 734                 obj = (union acpi_object *)&(element->package.elements[3]);
 735                 if (obj->type != ACPI_TYPE_INTEGER)
 736                         continue;
 737
 738                 cx.power = obj->integer.value;
 739
 740                 current_count++;
 741                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 742
 743                 /*
 744                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 745                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 746                  */
 747                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 748                         printk(KERN_WARNING
 749                                "Limiting number of power states to max (%d)\n",
 750                                ACPI_PROCESSOR_MAX_POWER);
 751                         printk(KERN_WARNING
 752                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 753                         break;
 754                 }
 755         }
 756
 757         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 758                           current_count));
 759
 760         /* Validate number of power states discovered */
 761         if (current_count < 2)
 762                 status = -EFAULT;
 763
 764       end:
 765         acpi_os_free(buffer.pointer);
 766
 767         return_VALUE(status);
 768 }
 769
 770 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 771 {
 772         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c2");
 773
 774         if (!cx->address)
 775                 return_VOID;
 776
 777         /*
 778          * C2 latency must be less than or equal to 100
 779          * microseconds.
 780          */
 781         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 782                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 783                                   "latency too large [%d]\n", cx->latency));
 784                 return_VOID;
 785         }
 786
 787         /*
 788          * Otherwise we've met all of our C2 requirements.
 789          * Normalize the C2 latency to expidite policy
 790          */
 791         cx->valid = 1;
 792         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 793
 794         return_VOID;
 795 }
 796
 797 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 798                                            struct acpi_processor_cx *cx)
 799 {
 800         static int bm_check_flag;
 801
 802         ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c3");
 803
 804         if (!cx->address)
 805                 return_VOID;
 806
 807         /*
 808          * C3 latency must be less than or equal to 1000
 809          * microseconds.
 810          */
 811         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 812                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 813                                   "latency too large [%d]\n", cx->latency));
 814                 return_VOID;
 815         }
 816
 817         /*
 818          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 819          * DMA transfers are used by any ISA device to avoid livelock.
 820          * Note that we could disable Type-F DMA (as recommended by
 821          * the erratum), but this is known to disrupt certain ISA
 822          * devices thus we take the conservative approach.
 823          */
 824         else if (errata.piix4.fdma) {
 825                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 826                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 827                 return_VOID;
 828         }
 829
 830         /* All the logic here assumes flags.bm_check is same across all CPUs */
 831         if (!bm_check_flag) {
 832                 /* Determine whether bm_check is needed based on CPU  */
 833                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 834                 bm_check_flag = pr->flags.bm_check;
 835         } else {
 836                 pr->flags.bm_check = bm_check_flag;
 837         }
 838
 839         if (pr->flags.bm_check) {
 840                 /* bus mastering control is necessary */
 841                 if (!pr->flags.bm_control) {
 842                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 843                                           "C3 support requires bus mastering control\n"));
 844                         return_VOID;
 845                 }
 846         } else {
 847                 /*
 848                  * WBINVD should be set in fadt, for C3 state to be
 849                  * supported on when bm_check is not required.
 850                  */
 851                 if (acpi_fadt.wb_invd != 1) {
 852                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 853                                           "Cache invalidation should work properly"
 854                                           " for C3 to be enabled on SMP systems\n"));
 855                         return_VOID;
 856                 }
 857                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD,
 858                                   0, ACPI_MTX_DO_NOT_LOCK);
 859         }
 860
 861         /*
 862          * Otherwise we've met all of our C3 requirements.
 863          * Normalize the C3 latency to expidite policy.  Enable
 864          * checking of bus mastering status (bm_check) so we can
 865          * use this in our C3 policy
 866          */
 867         cx->valid = 1;
 868         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 869
 870         return_VOID;
 871 }
 872
 873 static int acpi_processor_power_verify(struct acpi_processor *pr)
 874 {
 875         unsigned int i;
 876         unsigned int working = 0;
 877
 878 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 879         struct cpuinfo_x86 *c = cpu_data + pr->id;
 880         cpumask_t mask = cpumask_of_cpu(pr->id);
 881
 882         if (c->x86_vendor == X86_VENDOR_INTEL) {
 883                 on_each_cpu(switch_ipi_to_APIC_timer, &mask, 1, 1);
 884         }
 885 #endif
 886
 887         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 888                 struct acpi_processor_cx *cx = &pr->power.states[i];
 889
 890                 switch (cx->type) {
 891                 case ACPI_STATE_C1:
 892                         cx->valid = 1;
 893                         break;
 894
 895                 case ACPI_STATE_C2:
 896                         acpi_processor_power_verify_c2(cx);
 897                         break;
 898
 899                 case ACPI_STATE_C3:
 900                         acpi_processor_power_verify_c3(pr, cx);
 901 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 902                         if (c->x86_vendor == X86_VENDOR_INTEL) {
 903                                 on_each_cpu(switch_APIC_timer_to_ipi,
 904                                                 &mask, 1, 1);
 905                         }
 906 #endif
 907                         break;
 908                 }
 909
 910                 if (cx->valid)
 911                         working++;
 912         }
 913
 914         return (working);
 915 }
 916
 917 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 918 {
 919         unsigned int i;
 920         int result;
 921
 922         ACPI_FUNCTION_TRACE("acpi_processor_get_power_info");
 923
 924         /* NOTE: the idle thread may not be running while calling
 925          * this function */
 926
 927         /* Adding C1 state */
 928         acpi_processor_get_power_info_default_c1(pr);
 929         result = acpi_processor_get_power_info_cst(pr);
 930         if (result == -ENODEV)
 931                 acpi_processor_get_power_info_fadt(pr);
 932
 933         pr->power.count = acpi_processor_power_verify(pr);
 934
 935         /*
 936          * Set Default Policy
 937          * ------------------
 938          * Now that we know which states are supported, set the default
 939          * policy.  Note that this policy can be changed dynamically
 940          * (e.g. encourage deeper sleeps to conserve battery life when
 941          * not on AC).
 942          */
 943         result = acpi_processor_set_power_policy(pr);
 944         if (result)
 945                 return_VALUE(result);
 946
 947         /*
 948          * if one state of type C2 or C3 is available, mark this
 949          * CPU as being "idle manageable"
 950          */
 951         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 952                 if (pr->power.states[i].valid) {
 953                         pr->power.count = i;
 954                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 955                                 pr->flags.power = 1;
 956                 }
 957         }
 958
 959         return_VALUE(0);
 960 }
 961
 962 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
 963 {
 964         int result = 0;
 965
 966         ACPI_FUNCTION_TRACE("acpi_processor_cst_has_changed");
 967
 968         if (!pr)
 969                 return_VALUE(-EINVAL);
 970
 971         if (nocst) {
 972                 return_VALUE(-ENODEV);
 973         }
 974
 975         if (!pr->flags.power_setup_done)
 976                 return_VALUE(-ENODEV);
 977
 978         /* Fall back to the default idle loop */
 979         pm_idle = pm_idle_save;
 980         synchronize_sched();    /* Relies on interrupts forcing exit from idle. */
 981
 982         pr->flags.power = 0;
 983         result = acpi_processor_get_power_info(pr);
 984         if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
 985                 pm_idle = acpi_processor_idle;
 986
 987         return_VALUE(result);
 988 }
 989
 990 /* proc interface */
 991
 992 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 993 {
 994         struct acpi_processor *pr = (struct acpi_processor *)seq->private;
 995         unsigned int i;
 996
 997         ACPI_FUNCTION_TRACE("acpi_processor_power_seq_show");
 998
 999         if (!pr)
1000                 goto end;
1001
1002         seq_printf(seq, "active state:            C%zd\n"
1003                    "max_cstate:              C%d\n"
1004                    "bus master activity:     %08x\n",
1005                    pr->power.state ? pr->power.state - pr->power.states : 0,
1006                    max_cstate, (unsigned)pr->power.bm_activity);
1007
1008         seq_puts(seq, "states:\n");
1009
1010         for (i = 1; i <= pr->power.count; i++) {
1011                 seq_printf(seq, "   %cC%d:                  ",
1012                            (&pr->power.states[i] ==
1013                             pr->power.state ? '*' : ' '), i);
1014
1015                 if (!pr->power.states[i].valid) {
1016                         seq_puts(seq, "<not supported>\n");
1017                         continue;
1018                 }
1019
1020                 switch (pr->power.states[i].type) {
1021                 case ACPI_STATE_C1:
1022                         seq_printf(seq, "type[C1] ");
1023                         break;
1024                 case ACPI_STATE_C2:
1025                         seq_printf(seq, "type[C2] ");
1026                         break;
1027                 case ACPI_STATE_C3:
1028                         seq_printf(seq, "type[C3] ");
1029                         break;
1030                 default:
1031                         seq_printf(seq, "type[--] ");
1032                         break;
1033                 }
1034
1035                 if (pr->power.states[i].promotion.state)
1036                         seq_printf(seq, "promotion[C%zd] ",
1037                                    (pr->power.states[i].promotion.state -
1038                                     pr->power.states));
1039                 else
1040                         seq_puts(seq, "promotion[--] ");
1041
1042                 if (pr->power.states[i].demotion.state)
1043                         seq_printf(seq, "demotion[C%zd] ",
1044                                    (pr->power.states[i].demotion.state -
1045                                     pr->power.states));
1046                 else
1047                         seq_puts(seq, "demotion[--] ");
1048
1049                 seq_printf(seq, "latency[%03d] usage[%08d]\n",
1050                            pr->power.states[i].latency,
1051                            pr->power.states[i].usage);
1052         }
1053
1054       end:
1055         return_VALUE(0);
1056 }
1057
1058 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
1059 {
1060         return single_open(file, acpi_processor_power_seq_show,
1061                            PDE(inode)->data);
1062 }
1063
1064 static struct file_operations acpi_processor_power_fops = {
1065         .open = acpi_processor_power_open_fs,
1066         .read = seq_read,
1067         .llseek = seq_lseek,
1068         .release = single_release,
1069 };
1070
1071 int acpi_processor_power_init(struct acpi_processor *pr,
1072                               struct acpi_device *device)
1073 {
1074         acpi_status status = 0;
1075         static int first_run = 0;
1076         struct proc_dir_entry *entry = NULL;
1077         unsigned int i;
1078
1079         ACPI_FUNCTION_TRACE("acpi_processor_power_init");
1080
1081         if (!first_run) {
1082                 dmi_check_system(processor_power_dmi_table);
1083                 if (max_cstate < ACPI_C_STATES_MAX)
1084                         printk(KERN_NOTICE
1085                                "ACPI: processor limited to max C-state %d\n",
1086                                max_cstate);
1087                 first_run++;
1088         }
1089
1090         if (!pr)
1091                 return_VALUE(-EINVAL);
1092
1093         if (acpi_fadt.cst_cnt && !nocst) {
1094                 status =
1095                     acpi_os_write_port(acpi_fadt.smi_cmd, acpi_fadt.cst_cnt, 8);
1096                 if (ACPI_FAILURE(status)) {
1097                         ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1098                                           "Notifying BIOS of _CST ability failed\n"));
1099                 }
1100         }
1101
1102         acpi_processor_get_power_info(pr);
1103
1104         /*
1105          * Install the idle handler if processor power management is supported.
1106          * Note that we use previously set idle handler will be used on
1107          * platforms that only support C1.
1108          */
1109         if ((pr->flags.power) && (!boot_option_idle_override)) {
1110                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1111                 for (i = 1; i <= pr->power.count; i++)
1112                         if (pr->power.states[i].valid)
1113                                 printk(" C%d[C%d]", i,
1114                                        pr->power.states[i].type);
1115                 printk(")\n");
1116
1117                 if (pr->id == 0) {
1118                         pm_idle_save = pm_idle;
1119                         pm_idle = acpi_processor_idle;
1120                 }
1121         }
1122
1123         /* 'power' [R] */
1124         entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1125                                   S_IRUGO, acpi_device_dir(device));
1126         if (!entry)
1127                 ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1128                                   "Unable to create '%s' fs entry\n",
1129                                   ACPI_PROCESSOR_FILE_POWER));
1130         else {
1131                 entry->proc_fops = &acpi_processor_power_fops;
1132                 entry->data = acpi_driver_data(device);
1133                 entry->owner = THIS_MODULE;
1134         }
1135
1136         pr->flags.power_setup_done = 1;
1137
1138         return_VALUE(0);
1139 }
1140
1141 int acpi_processor_power_exit(struct acpi_processor *pr,
1142                               struct acpi_device *device)
1143 {
1144         ACPI_FUNCTION_TRACE("acpi_processor_power_exit");
1145
1146         pr->flags.power_setup_done = 0;
1147
1148         if (acpi_device_dir(device))
1149                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1150                                   acpi_device_dir(device));
1151
1152         /* Unregister the idle handler when processor #0 is removed. */
1153         if (pr->id == 0) {
1154                 pm_idle = pm_idle_save;
1155
1156                 /*
1157                  * We are about to unload the current idle thread pm callback
1158                  * (pm_idle), Wait for all processors to update cached/local
1159                  * copies of pm_idle before proceeding.
1160                  */
1161                 cpu_idle_wait();
1162         }
1163
1164         return_VALUE(0);
1165 }