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, 2005 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 #include <linux/latency.h>
  42 #include <linux/clockchips.h>
  43
  44 /*
  45  * Include the apic definitions for x86 to have the APIC timer related defines
  46  * available also for UP (on SMP it gets magically included via linux/smp.h).
  47  * asm/acpi.h is not an option, as it would require more include magic. Also
  48  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  49  */
  50 #ifdef CONFIG_X86
  51 #include <asm/apic.h>
  52 #endif
  53
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56
  57 #include <acpi/acpi_bus.h>
  58 #include <acpi/processor.h>
  59
  60 #define ACPI_PROCESSOR_COMPONENT        0x01000000
  61 #define ACPI_PROCESSOR_CLASS            "processor"
  62 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  63 ACPI_MODULE_NAME("processor_idle");
  64 #define ACPI_PROCESSOR_FILE_POWER       "power"
  65 #define US_TO_PM_TIMER_TICKS(t)         ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
  66 #define C2_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  67 #define C3_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  68 static void (*pm_idle_save) (void) __read_mostly;
  69 module_param(max_cstate, uint, 0644);
  70
  71 static unsigned int nocst __read_mostly;
  72 module_param(nocst, uint, 0000);
  73
  74 /*
  75  * bm_history -- bit-mask with a bit per jiffy of bus-master activity
  76  * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
  77  * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
  78  * 100 HZ: 0x0000000F: 4 jiffies = 40ms
  79  * reduce history for more aggressive entry into C3
  80  */
  81 static unsigned int bm_history __read_mostly =
  82     (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
  83 module_param(bm_history, uint, 0644);
  84 /* --------------------------------------------------------------------------
  85                                 Power Management
  86    -------------------------------------------------------------------------- */
  87
  88 /*
  89  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  90  * For now disable this. Probably a bug somewhere else.
  91  *
  92  * To skip this limit, boot/load with a large max_cstate limit.
  93  */
  94 static int set_max_cstate(struct dmi_system_id *id)
  95 {
  96         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  97                 return 0;
  98
  99         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
 100                " Override with \"processor.max_cstate=%d\"\n", id->ident,
 101                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
 102
 103         max_cstate = (long)id->driver_data;
 104
 105         return 0;
 106 }
 107
 108 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
 109    callers to only run once -AK */
 110 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 111         { set_max_cstate, "IBM ThinkPad R40e", {
 112           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 113           DMI_MATCH(DMI_BIOS_VERSION,"1SET70WW")}, (void *)1},
 114         { set_max_cstate, "IBM ThinkPad R40e", {
 115           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 116           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW")}, (void *)1},
 117         { set_max_cstate, "IBM ThinkPad R40e", {
 118           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 119           DMI_MATCH(DMI_BIOS_VERSION,"1SET43WW") }, (void*)1},
 120         { set_max_cstate, "IBM ThinkPad R40e", {
 121           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 122           DMI_MATCH(DMI_BIOS_VERSION,"1SET45WW") }, (void*)1},
 123         { set_max_cstate, "IBM ThinkPad R40e", {
 124           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 125           DMI_MATCH(DMI_BIOS_VERSION,"1SET47WW") }, (void*)1},
 126         { set_max_cstate, "IBM ThinkPad R40e", {
 127           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 128           DMI_MATCH(DMI_BIOS_VERSION,"1SET50WW") }, (void*)1},
 129         { set_max_cstate, "IBM ThinkPad R40e", {
 130           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 131           DMI_MATCH(DMI_BIOS_VERSION,"1SET52WW") }, (void*)1},
 132         { set_max_cstate, "IBM ThinkPad R40e", {
 133           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 134           DMI_MATCH(DMI_BIOS_VERSION,"1SET55WW") }, (void*)1},
 135         { set_max_cstate, "IBM ThinkPad R40e", {
 136           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 137           DMI_MATCH(DMI_BIOS_VERSION,"1SET56WW") }, (void*)1},
 138         { set_max_cstate, "IBM ThinkPad R40e", {
 139           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 140           DMI_MATCH(DMI_BIOS_VERSION,"1SET59WW") }, (void*)1},
 141         { set_max_cstate, "IBM ThinkPad R40e", {
 142           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 143           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW") }, (void*)1},
 144         { set_max_cstate, "IBM ThinkPad R40e", {
 145           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 146           DMI_MATCH(DMI_BIOS_VERSION,"1SET61WW") }, (void*)1},
 147         { set_max_cstate, "IBM ThinkPad R40e", {
 148           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 149           DMI_MATCH(DMI_BIOS_VERSION,"1SET62WW") }, (void*)1},
 150         { set_max_cstate, "IBM ThinkPad R40e", {
 151           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 152           DMI_MATCH(DMI_BIOS_VERSION,"1SET64WW") }, (void*)1},
 153         { set_max_cstate, "IBM ThinkPad R40e", {
 154           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 155           DMI_MATCH(DMI_BIOS_VERSION,"1SET65WW") }, (void*)1},
 156         { set_max_cstate, "IBM ThinkPad R40e", {
 157           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 158           DMI_MATCH(DMI_BIOS_VERSION,"1SET68WW") }, (void*)1},
 159         { set_max_cstate, "Medion 41700", {
 160           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 161           DMI_MATCH(DMI_BIOS_VERSION,"R01-A1J")}, (void *)1},
 162         { set_max_cstate, "Clevo 5600D", {
 163           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 164           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 165          (void *)2},
 166         {},
 167 };
 168
 169 static inline u32 ticks_elapsed(u32 t1, u32 t2)
 170 {
 171         if (t2 >= t1)
 172                 return (t2 - t1);
 173         else if (!(acpi_gbl_FADT.flags & ACPI_FADT_32BIT_TIMER))
 174                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 175         else
 176                 return ((0xFFFFFFFF - t1) + t2);
 177 }
 178
 179 static void
 180 acpi_processor_power_activate(struct acpi_processor *pr,
 181                               struct acpi_processor_cx *new)
 182 {
 183         struct acpi_processor_cx *old;
 184
 185         if (!pr || !new)
 186                 return;
 187
 188         old = pr->power.state;
 189
 190         if (old)
 191                 old->promotion.count = 0;
 192         new->demotion.count = 0;
 193
 194         /* Cleanup from old state. */
 195         if (old) {
 196                 switch (old->type) {
 197                 case ACPI_STATE_C3:
 198                         /* Disable bus master reload */
 199                         if (new->type != ACPI_STATE_C3 && pr->flags.bm_check)
 200                                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0);
 201                         break;
 202                 }
 203         }
 204
 205         /* Prepare to use new state. */
 206         switch (new->type) {
 207         case ACPI_STATE_C3:
 208                 /* Enable bus master reload */
 209                 if (old->type != ACPI_STATE_C3 && pr->flags.bm_check)
 210                         acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 211                 break;
 212         }
 213
 214         pr->power.state = new;
 215
 216         return;
 217 }
 218
 219 static void acpi_safe_halt(void)
 220 {
 221         current_thread_info()->status &= ~TS_POLLING;
 222         /*
 223          * TS_POLLING-cleared state must be visible before we
 224          * test NEED_RESCHED:
 225          */
 226         smp_mb();
 227         if (!need_resched())
 228                 safe_halt();
 229         current_thread_info()->status |= TS_POLLING;
 230 }
 231
 232 static atomic_t c3_cpu_count;
 233
 234 /* Common C-state entry for C2, C3, .. */
 235 static void acpi_cstate_enter(struct acpi_processor_cx *cstate)
 236 {
 237         if (cstate->space_id == ACPI_CSTATE_FFH) {
 238                 /* Call into architectural FFH based C-state */
 239                 acpi_processor_ffh_cstate_enter(cstate);
 240         } else {
 241                 int unused;
 242                 /* IO port based C-state */
 243                 inb(cstate->address);
 244                 /* Dummy wait op - must do something useless after P_LVL2 read
 245                    because chipsets cannot guarantee that STPCLK# signal
 246                    gets asserted in time to freeze execution properly. */
 247                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 248         }
 249 }
 250
 251 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 252
 253 /*
 254  * Some BIOS implementations switch to C3 in the published C2 state.
 255  * This seems to be a common problem on AMD boxen, but other vendors
 256  * are affected too. We pick the most conservative approach: we assume
 257  * that the local APIC stops in both C2 and C3.
 258  */
 259 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 260                                    struct acpi_processor_cx *cx)
 261 {
 262         struct acpi_processor_power *pwr = &pr->power;
 263         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 264
 265         /*
 266          * Check, if one of the previous states already marked the lapic
 267          * unstable
 268          */
 269         if (pwr->timer_broadcast_on_state < state)
 270                 return;
 271
 272         if (cx->type >= type)
 273                 pr->power.timer_broadcast_on_state = state;
 274 }
 275
 276 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 277 {
 278 #ifdef CONFIG_GENERIC_CLOCKEVENTS
 279         unsigned long reason;
 280
 281         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 282                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 283
 284         clockevents_notify(reason, &pr->id);
 285 #else
 286         cpumask_t mask = cpumask_of_cpu(pr->id);
 287
 288         if (pr->power.timer_broadcast_on_state < INT_MAX)
 289                 on_each_cpu(switch_APIC_timer_to_ipi, &mask, 1, 1);
 290         else
 291                 on_each_cpu(switch_ipi_to_APIC_timer, &mask, 1, 1);
 292 #endif
 293 }
 294
 295 /* Power(C) State timer broadcast control */
 296 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 297                                        struct acpi_processor_cx *cx,
 298                                        int broadcast)
 299 {
 300 #ifdef CONFIG_GENERIC_CLOCKEVENTS
 301
 302         int state = cx - pr->power.states;
 303
 304         if (state >= pr->power.timer_broadcast_on_state) {
 305                 unsigned long reason;
 306
 307                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 308                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 309                 clockevents_notify(reason, &pr->id);
 310         }
 311 #endif
 312 }
 313
 314 #else
 315
 316 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 317                                    struct acpi_processor_cx *cstate) { }
 318 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
 319 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 320                                        struct acpi_processor_cx *cx,
 321                                        int broadcast)
 322 {
 323 }
 324
 325 #endif
 326
 327 static void acpi_processor_idle(void)
 328 {
 329         struct acpi_processor *pr = NULL;
 330         struct acpi_processor_cx *cx = NULL;
 331         struct acpi_processor_cx *next_state = NULL;
 332         int sleep_ticks = 0;
 333         u32 t1, t2 = 0;
 334
 335         /*
 336          * Interrupts must be disabled during bus mastering calculations and
 337          * for C2/C3 transitions.
 338          */
 339         local_irq_disable();
 340
 341         pr = processors[smp_processor_id()];
 342         if (!pr) {
 343                 local_irq_enable();
 344                 return;
 345         }
 346
 347         /*
 348          * Check whether we truly need to go idle, or should
 349          * reschedule:
 350          */
 351         if (unlikely(need_resched())) {
 352                 local_irq_enable();
 353                 return;
 354         }
 355
 356         cx = pr->power.state;
 357         if (!cx) {
 358                 if (pm_idle_save)
 359                         pm_idle_save();
 360                 else
 361                         acpi_safe_halt();
 362                 return;
 363         }
 364
 365         /*
 366          * Check BM Activity
 367          * -----------------
 368          * Check for bus mastering activity (if required), record, and check
 369          * for demotion.
 370          */
 371         if (pr->flags.bm_check) {
 372                 u32 bm_status = 0;
 373                 unsigned long diff = jiffies - pr->power.bm_check_timestamp;
 374
 375                 if (diff > 31)
 376                         diff = 31;
 377
 378                 pr->power.bm_activity <<= diff;
 379
 380                 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 381                 if (bm_status) {
 382                         pr->power.bm_activity |= 0x1;
 383                         acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 384                 }
 385                 /*
 386                  * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 387                  * the true state of bus mastering activity; forcing us to
 388                  * manually check the BMIDEA bit of each IDE channel.
 389                  */
 390                 else if (errata.piix4.bmisx) {
 391                         if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 392                             || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 393                                 pr->power.bm_activity |= 0x1;
 394                 }
 395
 396                 pr->power.bm_check_timestamp = jiffies;
 397
 398                 /*
 399                  * If bus mastering is or was active this jiffy, demote
 400                  * to avoid a faulty transition.  Note that the processor
 401                  * won't enter a low-power state during this call (to this
 402                  * function) but should upon the next.
 403                  *
 404                  * TBD: A better policy might be to fallback to the demotion
 405                  *      state (use it for this quantum only) istead of
 406                  *      demoting -- and rely on duration as our sole demotion
 407                  *      qualification.  This may, however, introduce DMA
 408                  *      issues (e.g. floppy DMA transfer overrun/underrun).
 409                  */
 410                 if ((pr->power.bm_activity & 0x1) &&
 411                     cx->demotion.threshold.bm) {
 412                         local_irq_enable();
 413                         next_state = cx->demotion.state;
 414                         goto end;
 415                 }
 416         }
 417
 418 #ifdef CONFIG_HOTPLUG_CPU
 419         /*
 420          * Check for P_LVL2_UP flag before entering C2 and above on
 421          * an SMP system. We do it here instead of doing it at _CST/P_LVL
 422          * detection phase, to work cleanly with logical CPU hotplug.
 423          */
 424         if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 425             !pr->flags.has_cst && !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 426                 cx = &pr->power.states[ACPI_STATE_C1];
 427 #endif
 428
 429         /*
 430          * Sleep:
 431          * ------
 432          * Invoke the current Cx state to put the processor to sleep.
 433          */
 434         if (cx->type == ACPI_STATE_C2 || cx->type == ACPI_STATE_C3) {
 435                 current_thread_info()->status &= ~TS_POLLING;
 436                 /*
 437                  * TS_POLLING-cleared state must be visible before we
 438                  * test NEED_RESCHED:
 439                  */
 440                 smp_mb();
 441                 if (need_resched()) {
 442                         current_thread_info()->status |= TS_POLLING;
 443                         local_irq_enable();
 444                         return;
 445                 }
 446         }
 447
 448         switch (cx->type) {
 449
 450         case ACPI_STATE_C1:
 451                 /*
 452                  * Invoke C1.
 453                  * Use the appropriate idle routine, the one that would
 454                  * be used without acpi C-states.
 455                  */
 456                 if (pm_idle_save)
 457                         pm_idle_save();
 458                 else
 459                         acpi_safe_halt();
 460
 461                 /*
 462                  * TBD: Can't get time duration while in C1, as resumes
 463                  *      go to an ISR rather than here.  Need to instrument
 464                  *      base interrupt handler.
 465                  */
 466                 sleep_ticks = 0xFFFFFFFF;
 467                 break;
 468
 469         case ACPI_STATE_C2:
 470                 /* Get start time (ticks) */
 471                 t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 472                 /* Invoke C2 */
 473                 acpi_state_timer_broadcast(pr, cx, 1);
 474                 acpi_cstate_enter(cx);
 475                 /* Get end time (ticks) */
 476                 t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 477
 478 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86_TSC)
 479                 /* TSC halts in C2, so notify users */
 480                 mark_tsc_unstable("possible TSC halt in C2");
 481 #endif
 482                 /* Re-enable interrupts */
 483                 local_irq_enable();
 484                 current_thread_info()->status |= TS_POLLING;
 485                 /* Compute time (ticks) that we were actually asleep */
 486                 sleep_ticks =
 487                     ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
 488                 acpi_state_timer_broadcast(pr, cx, 0);
 489                 break;
 490
 491         case ACPI_STATE_C3:
 492
 493                 if (pr->flags.bm_check) {
 494                         if (atomic_inc_return(&c3_cpu_count) ==
 495                             num_online_cpus()) {
 496                                 /*
 497                                  * All CPUs are trying to go to C3
 498                                  * Disable bus master arbitration
 499                                  */
 500                                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1);
 501                         }
 502                 } else {
 503                         /* SMP with no shared cache... Invalidate cache  */
 504                         ACPI_FLUSH_CPU_CACHE();
 505                 }
 506
 507                 /* Get start time (ticks) */
 508                 t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 509                 /* Invoke C3 */
 510                 acpi_state_timer_broadcast(pr, cx, 1);
 511                 acpi_cstate_enter(cx);
 512                 /* Get end time (ticks) */
 513                 t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 514                 if (pr->flags.bm_check) {
 515                         /* Enable bus master arbitration */
 516                         atomic_dec(&c3_cpu_count);
 517                         acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0);
 518                 }
 519
 520 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86_TSC)
 521                 /* TSC halts in C3, so notify users */
 522                 mark_tsc_unstable("TSC halts in C3");
 523 #endif
 524                 /* Re-enable interrupts */
 525                 local_irq_enable();
 526                 current_thread_info()->status |= TS_POLLING;
 527                 /* Compute time (ticks) that we were actually asleep */
 528                 sleep_ticks =
 529                     ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
 530                 acpi_state_timer_broadcast(pr, cx, 0);
 531                 break;
 532
 533         default:
 534                 local_irq_enable();
 535                 return;
 536         }
 537         cx->usage++;
 538         if ((cx->type != ACPI_STATE_C1) && (sleep_ticks > 0))
 539                 cx->time += sleep_ticks;
 540
 541         next_state = pr->power.state;
 542
 543 #ifdef CONFIG_HOTPLUG_CPU
 544         /* Don't do promotion/demotion */
 545         if ((cx->type == ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 546             !pr->flags.has_cst && !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) {
 547                 next_state = cx;
 548                 goto end;
 549         }
 550 #endif
 551
 552         /*
 553          * Promotion?
 554          * ----------
 555          * Track the number of longs (time asleep is greater than threshold)
 556          * and promote when the count threshold is reached.  Note that bus
 557          * mastering activity may prevent promotions.
 558          * Do not promote above max_cstate.
 559          */
 560         if (cx->promotion.state &&
 561             ((cx->promotion.state - pr->power.states) <= max_cstate)) {
 562                 if (sleep_ticks > cx->promotion.threshold.ticks &&
 563                   cx->promotion.state->latency <= system_latency_constraint()) {
 564                         cx->promotion.count++;
 565                         cx->demotion.count = 0;
 566                         if (cx->promotion.count >=
 567                             cx->promotion.threshold.count) {
 568                                 if (pr->flags.bm_check) {
 569                                         if (!
 570                                             (pr->power.bm_activity & cx->
 571                                              promotion.threshold.bm)) {
 572                                                 next_state =
 573                                                     cx->promotion.state;
 574                                                 goto end;
 575                                         }
 576                                 } else {
 577                                         next_state = cx->promotion.state;
 578                                         goto end;
 579                                 }
 580                         }
 581                 }
 582         }
 583
 584         /*
 585          * Demotion?
 586          * ---------
 587          * Track the number of shorts (time asleep is less than time threshold)
 588          * and demote when the usage threshold is reached.
 589          */
 590         if (cx->demotion.state) {
 591                 if (sleep_ticks < cx->demotion.threshold.ticks) {
 592                         cx->demotion.count++;
 593                         cx->promotion.count = 0;
 594                         if (cx->demotion.count >= cx->demotion.threshold.count) {
 595                                 next_state = cx->demotion.state;
 596                                 goto end;
 597                         }
 598                 }
 599         }
 600
 601       end:
 602         /*
 603          * Demote if current state exceeds max_cstate
 604          * or if the latency of the current state is unacceptable
 605          */
 606         if ((pr->power.state - pr->power.states) > max_cstate ||
 607                 pr->power.state->latency > system_latency_constraint()) {
 608                 if (cx->demotion.state)
 609                         next_state = cx->demotion.state;
 610         }
 611
 612         /*
 613          * New Cx State?
 614          * -------------
 615          * If we're going to start using a new Cx state we must clean up
 616          * from the previous and prepare to use the new.
 617          */
 618         if (next_state != pr->power.state)
 619                 acpi_processor_power_activate(pr, next_state);
 620 }
 621
 622 static int acpi_processor_set_power_policy(struct acpi_processor *pr)
 623 {
 624         unsigned int i;
 625         unsigned int state_is_set = 0;
 626         struct acpi_processor_cx *lower = NULL;
 627         struct acpi_processor_cx *higher = NULL;
 628         struct acpi_processor_cx *cx;
 629
 630
 631         if (!pr)
 632                 return -EINVAL;
 633
 634         /*
 635          * This function sets the default Cx state policy (OS idle handler).
 636          * Our scheme is to promote quickly to C2 but more conservatively
 637          * to C3.  We're favoring C2  for its characteristics of low latency
 638          * (quick response), good power savings, and ability to allow bus
 639          * mastering activity.  Note that the Cx state policy is completely
 640          * customizable and can be altered dynamically.
 641          */
 642
 643         /* startup state */
 644         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 645                 cx = &pr->power.states[i];
 646                 if (!cx->valid)
 647                         continue;
 648
 649                 if (!state_is_set)
 650                         pr->power.state = cx;
 651                 state_is_set++;
 652                 break;
 653         }
 654
 655         if (!state_is_set)
 656                 return -ENODEV;
 657
 658         /* demotion */
 659         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 660                 cx = &pr->power.states[i];
 661                 if (!cx->valid)
 662                         continue;
 663
 664                 if (lower) {
 665                         cx->demotion.state = lower;
 666                         cx->demotion.threshold.ticks = cx->latency_ticks;
 667                         cx->demotion.threshold.count = 1;
 668                         if (cx->type == ACPI_STATE_C3)
 669                                 cx->demotion.threshold.bm = bm_history;
 670                 }
 671
 672                 lower = cx;
 673         }
 674
 675         /* promotion */
 676         for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
 677                 cx = &pr->power.states[i];
 678                 if (!cx->valid)
 679                         continue;
 680
 681                 if (higher) {
 682                         cx->promotion.state = higher;
 683                         cx->promotion.threshold.ticks = cx->latency_ticks;
 684                         if (cx->type >= ACPI_STATE_C2)
 685                                 cx->promotion.threshold.count = 4;
 686                         else
 687                                 cx->promotion.threshold.count = 10;
 688                         if (higher->type == ACPI_STATE_C3)
 689                                 cx->promotion.threshold.bm = bm_history;
 690                 }
 691
 692                 higher = cx;
 693         }
 694
 695         return 0;
 696 }
 697
 698 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 699 {
 700
 701         if (!pr)
 702                 return -EINVAL;
 703
 704         if (!pr->pblk)
 705                 return -ENODEV;
 706
 707         /* if info is obtained from pblk/fadt, type equals state */
 708         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 709         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 710
 711 #ifndef CONFIG_HOTPLUG_CPU
 712         /*
 713          * Check for P_LVL2_UP flag before entering C2 and above on
 714          * an SMP system.
 715          */
 716         if ((num_online_cpus() > 1) &&
 717             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 718                 return -ENODEV;
 719 #endif
 720
 721         /* determine C2 and C3 address from pblk */
 722         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 723         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 724
 725         /* determine latencies from FADT */
 726         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 727         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 728
 729         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 730                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 731                           pr->power.states[ACPI_STATE_C2].address,
 732                           pr->power.states[ACPI_STATE_C3].address));
 733
 734         return 0;
 735 }
 736
 737 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 738 {
 739         if (!pr->power.states[ACPI_STATE_C1].valid) {
 740                 /* set the first C-State to C1 */
 741                 /* all processors need to support C1 */
 742                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 743                 pr->power.states[ACPI_STATE_C1].valid = 1;
 744         }
 745         /* the C0 state only exists as a filler in our array */
 746         pr->power.states[ACPI_STATE_C0].valid = 1;
 747         return 0;
 748 }
 749
 750 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 751 {
 752         acpi_status status = 0;
 753         acpi_integer count;
 754         int current_count;
 755         int i;
 756         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 757         union acpi_object *cst;
 758
 759
 760         if (nocst)
 761                 return -ENODEV;
 762
 763         current_count = 0;
 764
 765         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 766         if (ACPI_FAILURE(status)) {
 767                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 768                 return -ENODEV;
 769         }
 770
 771         cst = buffer.pointer;
 772
 773         /* There must be at least 2 elements */
 774         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 775                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 776                 status = -EFAULT;
 777                 goto end;
 778         }
 779
 780         count = cst->package.elements[0].integer.value;
 781
 782         /* Validate number of power states. */
 783         if (count < 1 || count != cst->package.count - 1) {
 784                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 785                 status = -EFAULT;
 786                 goto end;
 787         }
 788
 789         /* Tell driver that at least _CST is supported. */
 790         pr->flags.has_cst = 1;
 791
 792         for (i = 1; i <= count; i++) {
 793                 union acpi_object *element;
 794                 union acpi_object *obj;
 795                 struct acpi_power_register *reg;
 796                 struct acpi_processor_cx cx;
 797
 798                 memset(&cx, 0, sizeof(cx));
 799
 800                 element = &(cst->package.elements[i]);
 801                 if (element->type != ACPI_TYPE_PACKAGE)
 802                         continue;
 803
 804                 if (element->package.count != 4)
 805                         continue;
 806
 807                 obj = &(element->package.elements[0]);
 808
 809                 if (obj->type != ACPI_TYPE_BUFFER)
 810                         continue;
 811
 812                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 813
 814                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 815                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 816                         continue;
 817
 818                 /* There should be an easy way to extract an integer... */
 819                 obj = &(element->package.elements[1]);
 820                 if (obj->type != ACPI_TYPE_INTEGER)
 821                         continue;
 822
 823                 cx.type = obj->integer.value;
 824                 /*
 825                  * Some buggy BIOSes won't list C1 in _CST -
 826                  * Let acpi_processor_get_power_info_default() handle them later
 827                  */
 828                 if (i == 1 && cx.type != ACPI_STATE_C1)
 829                         current_count++;
 830
 831                 cx.address = reg->address;
 832                 cx.index = current_count + 1;
 833
 834                 cx.space_id = ACPI_CSTATE_SYSTEMIO;
 835                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 836                         if (acpi_processor_ffh_cstate_probe
 837                                         (pr->id, &cx, reg) == 0) {
 838                                 cx.space_id = ACPI_CSTATE_FFH;
 839                         } else if (cx.type != ACPI_STATE_C1) {
 840                                 /*
 841                                  * C1 is a special case where FIXED_HARDWARE
 842                                  * can be handled in non-MWAIT way as well.
 843                                  * In that case, save this _CST entry info.
 844                                  * That is, we retain space_id of SYSTEM_IO for
 845                                  * halt based C1.
 846                                  * Otherwise, ignore this info and continue.
 847                                  */
 848                                 continue;
 849                         }
 850                 }
 851
 852                 obj = &(element->package.elements[2]);
 853                 if (obj->type != ACPI_TYPE_INTEGER)
 854                         continue;
 855
 856                 cx.latency = obj->integer.value;
 857
 858                 obj = &(element->package.elements[3]);
 859                 if (obj->type != ACPI_TYPE_INTEGER)
 860                         continue;
 861
 862                 cx.power = obj->integer.value;
 863
 864                 current_count++;
 865                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 866
 867                 /*
 868                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 869                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 870                  */
 871                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 872                         printk(KERN_WARNING
 873                                "Limiting number of power states to max (%d)\n",
 874                                ACPI_PROCESSOR_MAX_POWER);
 875                         printk(KERN_WARNING
 876                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 877                         break;
 878                 }
 879         }
 880
 881         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 882                           current_count));
 883
 884         /* Validate number of power states discovered */
 885         if (current_count < 2)
 886                 status = -EFAULT;
 887
 888       end:
 889         kfree(buffer.pointer);
 890
 891         return status;
 892 }
 893
 894 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 895 {
 896
 897         if (!cx->address)
 898                 return;
 899
 900         /*
 901          * C2 latency must be less than or equal to 100
 902          * microseconds.
 903          */
 904         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 905                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 906                                   "latency too large [%d]\n", cx->latency));
 907                 return;
 908         }
 909
 910         /*
 911          * Otherwise we've met all of our C2 requirements.
 912          * Normalize the C2 latency to expidite policy
 913          */
 914         cx->valid = 1;
 915         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 916
 917         return;
 918 }
 919
 920 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 921                                            struct acpi_processor_cx *cx)
 922 {
 923         static int bm_check_flag;
 924
 925
 926         if (!cx->address)
 927                 return;
 928
 929         /*
 930          * C3 latency must be less than or equal to 1000
 931          * microseconds.
 932          */
 933         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 934                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 935                                   "latency too large [%d]\n", cx->latency));
 936                 return;
 937         }
 938
 939         /*
 940          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 941          * DMA transfers are used by any ISA device to avoid livelock.
 942          * Note that we could disable Type-F DMA (as recommended by
 943          * the erratum), but this is known to disrupt certain ISA
 944          * devices thus we take the conservative approach.
 945          */
 946         else if (errata.piix4.fdma) {
 947                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 948                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 949                 return;
 950         }
 951
 952         /* All the logic here assumes flags.bm_check is same across all CPUs */
 953         if (!bm_check_flag) {
 954                 /* Determine whether bm_check is needed based on CPU  */
 955                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 956                 bm_check_flag = pr->flags.bm_check;
 957         } else {
 958                 pr->flags.bm_check = bm_check_flag;
 959         }
 960
 961         if (pr->flags.bm_check) {
 962                 /* bus mastering control is necessary */
 963                 if (!pr->flags.bm_control) {
 964                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 965                                           "C3 support requires bus mastering control\n"));
 966                         return;
 967                 }
 968         } else {
 969                 /*
 970                  * WBINVD should be set in fadt, for C3 state to be
 971                  * supported on when bm_check is not required.
 972                  */
 973                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 974                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 975                                           "Cache invalidation should work properly"
 976                                           " for C3 to be enabled on SMP systems\n"));
 977                         return;
 978                 }
 979                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0);
 980         }
 981
 982         /*
 983          * Otherwise we've met all of our C3 requirements.
 984          * Normalize the C3 latency to expidite policy.  Enable
 985          * checking of bus mastering status (bm_check) so we can
 986          * use this in our C3 policy
 987          */
 988         cx->valid = 1;
 989         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 990
 991         return;
 992 }
 993
 994 static int acpi_processor_power_verify(struct acpi_processor *pr)
 995 {
 996         unsigned int i;
 997         unsigned int working = 0;
 998
 999         pr->power.timer_broadcast_on_state = INT_MAX;
1000
1001         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
1002                 struct acpi_processor_cx *cx = &pr->power.states[i];
1003
1004                 switch (cx->type) {
1005                 case ACPI_STATE_C1:
1006                         cx->valid = 1;
1007                         break;
1008
1009                 case ACPI_STATE_C2:
1010                         acpi_processor_power_verify_c2(cx);
1011                         if (cx->valid)
1012                                 acpi_timer_check_state(i, pr, cx);
1013                         break;
1014
1015                 case ACPI_STATE_C3:
1016                         acpi_processor_power_verify_c3(pr, cx);
1017                         if (cx->valid)
1018                                 acpi_timer_check_state(i, pr, cx);
1019                         break;
1020                 }
1021
1022                 if (cx->valid)
1023                         working++;
1024         }
1025
1026         acpi_propagate_timer_broadcast(pr);
1027
1028         return (working);
1029 }
1030
1031 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1032 {
1033         unsigned int i;
1034         int result;
1035
1036
1037         /* NOTE: the idle thread may not be running while calling
1038          * this function */
1039
1040         /* Zero initialize all the C-states info. */
1041         memset(pr->power.states, 0, sizeof(pr->power.states));
1042
1043         result = acpi_processor_get_power_info_cst(pr);
1044         if (result == -ENODEV)
1045                 result = acpi_processor_get_power_info_fadt(pr);
1046
1047         if (result)
1048                 return result;
1049
1050         acpi_processor_get_power_info_default(pr);
1051
1052         pr->power.count = acpi_processor_power_verify(pr);
1053
1054         /*
1055          * Set Default Policy
1056          * ------------------
1057          * Now that we know which states are supported, set the default
1058          * policy.  Note that this policy can be changed dynamically
1059          * (e.g. encourage deeper sleeps to conserve battery life when
1060          * not on AC).
1061          */
1062         result = acpi_processor_set_power_policy(pr);
1063         if (result)
1064                 return result;
1065
1066         /*
1067          * if one state of type C2 or C3 is available, mark this
1068          * CPU as being "idle manageable"
1069          */
1070         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
1071                 if (pr->power.states[i].valid) {
1072                         pr->power.count = i;
1073                         if (pr->power.states[i].type >= ACPI_STATE_C2)
1074                                 pr->flags.power = 1;
1075                 }
1076         }
1077
1078         return 0;
1079 }
1080
1081 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1082 {
1083         int result = 0;
1084
1085
1086         if (!pr)
1087                 return -EINVAL;
1088
1089         if (nocst) {
1090                 return -ENODEV;
1091         }
1092
1093         if (!pr->flags.power_setup_done)
1094                 return -ENODEV;
1095
1096         /* Fall back to the default idle loop */
1097         pm_idle = pm_idle_save;
1098         synchronize_sched();    /* Relies on interrupts forcing exit from idle. */
1099
1100         pr->flags.power = 0;
1101         result = acpi_processor_get_power_info(pr);
1102         if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
1103                 pm_idle = acpi_processor_idle;
1104
1105         return result;
1106 }
1107
1108 /* proc interface */
1109
1110 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
1111 {
1112         struct acpi_processor *pr = seq->private;
1113         unsigned int i;
1114
1115
1116         if (!pr)
1117                 goto end;
1118
1119         seq_printf(seq, "active state:            C%zd\n"
1120                    "max_cstate:              C%d\n"
1121                    "bus master activity:     %08x\n"
1122                    "maximum allowed latency: %d usec\n",
1123                    pr->power.state ? pr->power.state - pr->power.states : 0,
1124                    max_cstate, (unsigned)pr->power.bm_activity,
1125                    system_latency_constraint());
1126
1127         seq_puts(seq, "states:\n");
1128
1129         for (i = 1; i <= pr->power.count; i++) {
1130                 seq_printf(seq, "   %cC%d:                  ",
1131                            (&pr->power.states[i] ==
1132                             pr->power.state ? '*' : ' '), i);
1133
1134                 if (!pr->power.states[i].valid) {
1135                         seq_puts(seq, "<not supported>\n");
1136                         continue;
1137                 }
1138
1139                 switch (pr->power.states[i].type) {
1140                 case ACPI_STATE_C1:
1141                         seq_printf(seq, "type[C1] ");
1142                         break;
1143                 case ACPI_STATE_C2:
1144                         seq_printf(seq, "type[C2] ");
1145                         break;
1146                 case ACPI_STATE_C3:
1147                         seq_printf(seq, "type[C3] ");
1148                         break;
1149                 default:
1150                         seq_printf(seq, "type[--] ");
1151                         break;
1152                 }
1153
1154                 if (pr->power.states[i].promotion.state)
1155                         seq_printf(seq, "promotion[C%zd] ",
1156                                    (pr->power.states[i].promotion.state -
1157                                     pr->power.states));
1158                 else
1159                         seq_puts(seq, "promotion[--] ");
1160
1161                 if (pr->power.states[i].demotion.state)
1162                         seq_printf(seq, "demotion[C%zd] ",
1163                                    (pr->power.states[i].demotion.state -
1164                                     pr->power.states));
1165                 else
1166                         seq_puts(seq, "demotion[--] ");
1167
1168                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
1169                            pr->power.states[i].latency,
1170                            pr->power.states[i].usage,
1171                            (unsigned long long)pr->power.states[i].time);
1172         }
1173
1174       end:
1175         return 0;
1176 }
1177
1178 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
1179 {
1180         return single_open(file, acpi_processor_power_seq_show,
1181                            PDE(inode)->data);
1182 }
1183
1184 static const struct file_operations acpi_processor_power_fops = {
1185         .open = acpi_processor_power_open_fs,
1186         .read = seq_read,
1187         .llseek = seq_lseek,
1188         .release = single_release,
1189 };
1190
1191 #ifdef CONFIG_SMP
1192 static void smp_callback(void *v)
1193 {
1194         /* we already woke the CPU up, nothing more to do */
1195 }
1196
1197 /*
1198  * This function gets called when a part of the kernel has a new latency
1199  * requirement.  This means we need to get all processors out of their C-state,
1200  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
1201  * wakes them all right up.
1202  */
1203 static int acpi_processor_latency_notify(struct notifier_block *b,
1204                 unsigned long l, void *v)
1205 {
1206         smp_call_function(smp_callback, NULL, 0, 1);
1207         return NOTIFY_OK;
1208 }
1209
1210 static struct notifier_block acpi_processor_latency_notifier = {
1211         .notifier_call = acpi_processor_latency_notify,
1212 };
1213 #endif
1214
1215 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1216                               struct acpi_device *device)
1217 {
1218         acpi_status status = 0;
1219         static int first_run;
1220         struct proc_dir_entry *entry = NULL;
1221         unsigned int i;
1222
1223
1224         if (!first_run) {
1225                 dmi_check_system(processor_power_dmi_table);
1226                 if (max_cstate < ACPI_C_STATES_MAX)
1227                         printk(KERN_NOTICE
1228                                "ACPI: processor limited to max C-state %d\n",
1229                                max_cstate);
1230                 first_run++;
1231 #ifdef CONFIG_SMP
1232                 register_latency_notifier(&acpi_processor_latency_notifier);
1233 #endif
1234         }
1235
1236         if (!pr)
1237                 return -EINVAL;
1238
1239         if (acpi_gbl_FADT.cst_control && !nocst) {
1240                 status =
1241                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1242                 if (ACPI_FAILURE(status)) {
1243                         ACPI_EXCEPTION((AE_INFO, status,
1244                                         "Notifying BIOS of _CST ability failed"));
1245                 }
1246         }
1247
1248         acpi_processor_get_power_info(pr);
1249
1250         /*
1251          * Install the idle handler if processor power management is supported.
1252          * Note that we use previously set idle handler will be used on
1253          * platforms that only support C1.
1254          */
1255         if ((pr->flags.power) && (!boot_option_idle_override)) {
1256                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1257                 for (i = 1; i <= pr->power.count; i++)
1258                         if (pr->power.states[i].valid)
1259                                 printk(" C%d[C%d]", i,
1260                                        pr->power.states[i].type);
1261                 printk(")\n");
1262
1263                 if (pr->id == 0) {
1264                         pm_idle_save = pm_idle;
1265                         pm_idle = acpi_processor_idle;
1266                 }
1267         }
1268
1269         /* 'power' [R] */
1270         entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1271                                   S_IRUGO, acpi_device_dir(device));
1272         if (!entry)
1273                 return -EIO;
1274         else {
1275                 entry->proc_fops = &acpi_processor_power_fops;
1276                 entry->data = acpi_driver_data(device);
1277                 entry->owner = THIS_MODULE;
1278         }
1279
1280         pr->flags.power_setup_done = 1;
1281
1282         return 0;
1283 }
1284
1285 int acpi_processor_power_exit(struct acpi_processor *pr,
1286                               struct acpi_device *device)
1287 {
1288
1289         pr->flags.power_setup_done = 0;
1290
1291         if (acpi_device_dir(device))
1292                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1293                                   acpi_device_dir(device));
1294
1295         /* Unregister the idle handler when processor #0 is removed. */
1296         if (pr->id == 0) {
1297                 pm_idle = pm_idle_save;
1298
1299                 /*
1300                  * We are about to unload the current idle thread pm callback
1301                  * (pm_idle), Wait for all processors to update cached/local
1302                  * copies of pm_idle before proceeding.
1303                  */
1304                 cpu_idle_wait();
1305 #ifdef CONFIG_SMP
1306                 unregister_latency_notifier(&acpi_processor_latency_notifier);
1307 #endif
1308         }
1309
1310         return 0;
1311 }