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