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