git.oblomov.eu Git - linux-2.6/blob - drivers/acpi/processor_idle.c

   1 /*
   2  * processor_idle - idle state submodule to the ACPI processor driver
   3  *
   4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   6  *  Copyright (C) 2004, 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 PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
  67 #define C2_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  68 #define C3_OVERHEAD                     4       /* 1us (3.579 ticks per us) */
  69 static void (*pm_idle_save) (void) __read_mostly;
  70 module_param(max_cstate, uint, 0644);
  71
  72 static unsigned int nocst __read_mostly;
  73 module_param(nocst, uint, 0000);
  74
  75 /*
  76  * bm_history -- bit-mask with a bit per jiffy of bus-master activity
  77  * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
  78  * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
  79  * 100 HZ: 0x0000000F: 4 jiffies = 40ms
  80  * reduce history for more aggressive entry into C3
  81  */
  82 static unsigned int bm_history __read_mostly =
  83     (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
  84 module_param(bm_history, uint, 0644);
  85 /* --------------------------------------------------------------------------
  86                                 Power Management
  87    -------------------------------------------------------------------------- */
  88
  89 /*
  90  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  91  * For now disable this. Probably a bug somewhere else.
  92  *
  93  * To skip this limit, boot/load with a large max_cstate limit.
  94  */
  95 static int set_max_cstate(struct dmi_system_id *id)
  96 {
  97         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  98                 return 0;
  99
 100         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
 101                " Override with \"processor.max_cstate=%d\"\n", id->ident,
 102                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
 103
 104         max_cstate = (long)id->driver_data;
 105
 106         return 0;
 107 }
 108
 109 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
 110    callers to only run once -AK */
 111 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 112         { set_max_cstate, "IBM ThinkPad R40e", {
 113           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 114           DMI_MATCH(DMI_BIOS_VERSION,"1SET70WW")}, (void *)1},
 115         { set_max_cstate, "IBM ThinkPad R40e", {
 116           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 117           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW")}, (void *)1},
 118         { set_max_cstate, "IBM ThinkPad R40e", {
 119           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 120           DMI_MATCH(DMI_BIOS_VERSION,"1SET43WW") }, (void*)1},
 121         { set_max_cstate, "IBM ThinkPad R40e", {
 122           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 123           DMI_MATCH(DMI_BIOS_VERSION,"1SET45WW") }, (void*)1},
 124         { set_max_cstate, "IBM ThinkPad R40e", {
 125           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 126           DMI_MATCH(DMI_BIOS_VERSION,"1SET47WW") }, (void*)1},
 127         { set_max_cstate, "IBM ThinkPad R40e", {
 128           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 129           DMI_MATCH(DMI_BIOS_VERSION,"1SET50WW") }, (void*)1},
 130         { set_max_cstate, "IBM ThinkPad R40e", {
 131           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 132           DMI_MATCH(DMI_BIOS_VERSION,"1SET52WW") }, (void*)1},
 133         { set_max_cstate, "IBM ThinkPad R40e", {
 134           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 135           DMI_MATCH(DMI_BIOS_VERSION,"1SET55WW") }, (void*)1},
 136         { set_max_cstate, "IBM ThinkPad R40e", {
 137           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 138           DMI_MATCH(DMI_BIOS_VERSION,"1SET56WW") }, (void*)1},
 139         { set_max_cstate, "IBM ThinkPad R40e", {
 140           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 141           DMI_MATCH(DMI_BIOS_VERSION,"1SET59WW") }, (void*)1},
 142         { set_max_cstate, "IBM ThinkPad R40e", {
 143           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 144           DMI_MATCH(DMI_BIOS_VERSION,"1SET60WW") }, (void*)1},
 145         { set_max_cstate, "IBM ThinkPad R40e", {
 146           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 147           DMI_MATCH(DMI_BIOS_VERSION,"1SET61WW") }, (void*)1},
 148         { set_max_cstate, "IBM ThinkPad R40e", {
 149           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 150           DMI_MATCH(DMI_BIOS_VERSION,"1SET62WW") }, (void*)1},
 151         { set_max_cstate, "IBM ThinkPad R40e", {
 152           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 153           DMI_MATCH(DMI_BIOS_VERSION,"1SET64WW") }, (void*)1},
 154         { set_max_cstate, "IBM ThinkPad R40e", {
 155           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 156           DMI_MATCH(DMI_BIOS_VERSION,"1SET65WW") }, (void*)1},
 157         { set_max_cstate, "IBM ThinkPad R40e", {
 158           DMI_MATCH(DMI_BIOS_VENDOR,"IBM"),
 159           DMI_MATCH(DMI_BIOS_VERSION,"1SET68WW") }, (void*)1},
 160         { set_max_cstate, "Medion 41700", {
 161           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 162           DMI_MATCH(DMI_BIOS_VERSION,"R01-A1J")}, (void *)1},
 163         { set_max_cstate, "Clevo 5600D", {
 164           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 165           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 166          (void *)2},
 167         {},
 168 };
 169
 170 static inline u32 ticks_elapsed(u32 t1, u32 t2)
 171 {
 172         if (t2 >= t1)
 173                 return (t2 - t1);
 174         else if (!(acpi_gbl_FADT.flags & ACPI_FADT_32BIT_TIMER))
 175                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 176         else
 177                 return ((0xFFFFFFFF - t1) + t2);
 178 }
 179
 180 static void
 181 acpi_processor_power_activate(struct acpi_processor *pr,
 182                               struct acpi_processor_cx *new)
 183 {
 184         struct acpi_processor_cx *old;
 185
 186         if (!pr || !new)
 187                 return;
 188
 189         old = pr->power.state;
 190
 191         if (old)
 192                 old->promotion.count = 0;
 193         new->demotion.count = 0;
 194
 195         /* Cleanup from old state. */
 196         if (old) {
 197                 switch (old->type) {
 198                 case ACPI_STATE_C3:
 199                         /* Disable bus master reload */
 200                         if (new->type != ACPI_STATE_C3 && pr->flags.bm_check)
 201                                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0);
 202                         break;
 203                 }
 204         }
 205
 206         /* Prepare to use new state. */
 207         switch (new->type) {
 208         case ACPI_STATE_C3:
 209                 /* Enable bus master reload */
 210                 if (old->type != ACPI_STATE_C3 && pr->flags.bm_check)
 211                         acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 212                 break;
 213         }
 214
 215         pr->power.state = new;
 216
 217         return;
 218 }
 219
 220 static void acpi_safe_halt(void)
 221 {
 222         current_thread_info()->status &= ~TS_POLLING;
 223         /*
 224          * TS_POLLING-cleared state must be visible before we
 225          * test NEED_RESCHED:
 226          */
 227         smp_mb();
 228         if (!need_resched())
 229                 safe_halt();
 230         current_thread_info()->status |= TS_POLLING;
 231 }
 232
 233 static atomic_t c3_cpu_count;
 234
 235 /* Common C-state entry for C2, C3, .. */
 236 static void acpi_cstate_enter(struct acpi_processor_cx *cstate)
 237 {
 238         if (cstate->space_id == ACPI_CSTATE_FFH) {
 239                 /* Call into architectural FFH based C-state */
 240                 acpi_processor_ffh_cstate_enter(cstate);
 241         } else {
 242                 int unused;
 243                 /* IO port based C-state */
 244                 inb(cstate->address);
 245                 /* Dummy wait op - must do something useless after P_LVL2 read
 246                    because chipsets cannot guarantee that STPCLK# signal
 247                    gets asserted in time to freeze execution properly. */
 248                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 249         }
 250 }
 251
 252 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 253
 254 /*
 255  * Some BIOS implementations switch to C3 in the published C2 state.
 256  * This seems to be a common problem on AMD boxen, but other vendors
 257  * are affected too. We pick the most conservative approach: we assume
 258  * that the local APIC stops in both C2 and C3.
 259  */
 260 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 261                                    struct acpi_processor_cx *cx)
 262 {
 263         struct acpi_processor_power *pwr = &pr->power;
 264         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 265
 266         /*
 267          * Check, if one of the previous states already marked the lapic
 268          * unstable
 269          */
 270         if (pwr->timer_broadcast_on_state < state)
 271                 return;
 272
 273         if (cx->type >= type)
 274                 pr->power.timer_broadcast_on_state = state;
 275 }
 276
 277 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 278 {
 279 #ifdef CONFIG_GENERIC_CLOCKEVENTS
 280         unsigned long reason;
 281
 282         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 283                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 284
 285         clockevents_notify(reason, &pr->id);
 286 #else
 287         cpumask_t mask = cpumask_of_cpu(pr->id);
 288
 289         if (pr->power.timer_broadcast_on_state < INT_MAX)
 290                 on_each_cpu(switch_APIC_timer_to_ipi, &mask, 1, 1);
 291         else
 292                 on_each_cpu(switch_ipi_to_APIC_timer, &mask, 1, 1);
 293 #endif
 294 }
 295
 296 /* Power(C) State timer broadcast control */
 297 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 298                                        struct acpi_processor_cx *cx,
 299                                        int broadcast)
 300 {
 301 #ifdef CONFIG_GENERIC_CLOCKEVENTS
 302
 303         int state = cx - pr->power.states;
 304
 305         if (state >= pr->power.timer_broadcast_on_state) {
 306                 unsigned long reason;
 307
 308                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 309                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 310                 clockevents_notify(reason, &pr->id);
 311         }
 312 #endif
 313 }
 314
 315 #else
 316
 317 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 318                                    struct acpi_processor_cx *cstate) { }
 319 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
 320 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 321                                        struct acpi_processor_cx *cx,
 322                                        int broadcast)
 323 {
 324 }
 325
 326 #endif
 327
 328 /*
 329  * Suspend / resume control
 330  */
 331 static int acpi_idle_suspend;
 332
 333 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
 334 {
 335         acpi_idle_suspend = 1;
 336         return 0;
 337 }
 338
 339 int acpi_processor_resume(struct acpi_device * device)
 340 {
 341         acpi_idle_suspend = 0;
 342         return 0;
 343 }
 344
 345 static void acpi_processor_idle(void)
 346 {
 347         struct acpi_processor *pr = NULL;
 348         struct acpi_processor_cx *cx = NULL;
 349         struct acpi_processor_cx *next_state = NULL;
 350         int sleep_ticks = 0;
 351         u32 t1, t2 = 0;
 352
 353         /*
 354          * Interrupts must be disabled during bus mastering calculations and
 355          * for C2/C3 transitions.
 356          */
 357         local_irq_disable();
 358
 359         pr = processors[smp_processor_id()];
 360         if (!pr) {
 361                 local_irq_enable();
 362                 return;
 363         }
 364
 365         /*
 366          * Check whether we truly need to go idle, or should
 367          * reschedule:
 368          */
 369         if (unlikely(need_resched())) {
 370                 local_irq_enable();
 371                 return;
 372         }
 373
 374         cx = pr->power.state;
 375         if (!cx || acpi_idle_suspend) {
 376                 if (pm_idle_save)
 377                         pm_idle_save();
 378                 else
 379                         acpi_safe_halt();
 380                 return;
 381         }
 382
 383         /*
 384          * Check BM Activity
 385          * -----------------
 386          * Check for bus mastering activity (if required), record, and check
 387          * for demotion.
 388          */
 389         if (pr->flags.bm_check) {
 390                 u32 bm_status = 0;
 391                 unsigned long diff = jiffies - pr->power.bm_check_timestamp;
 392
 393                 if (diff > 31)
 394                         diff = 31;
 395
 396                 pr->power.bm_activity <<= diff;
 397
 398                 acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 399                 if (bm_status) {
 400                         pr->power.bm_activity |= 0x1;
 401                         acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 402                 }
 403                 /*
 404                  * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 405                  * the true state of bus mastering activity; forcing us to
 406                  * manually check the BMIDEA bit of each IDE channel.
 407                  */
 408                 else if (errata.piix4.bmisx) {
 409                         if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 410                             || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 411                                 pr->power.bm_activity |= 0x1;
 412                 }
 413
 414                 pr->power.bm_check_timestamp = jiffies;
 415
 416                 /*
 417                  * If bus mastering is or was active this jiffy, demote
 418                  * to avoid a faulty transition.  Note that the processor
 419                  * won't enter a low-power state during this call (to this
 420                  * function) but should upon the next.
 421                  *
 422                  * TBD: A better policy might be to fallback to the demotion
 423                  *      state (use it for this quantum only) istead of
 424                  *      demoting -- and rely on duration as our sole demotion
 425                  *      qualification.  This may, however, introduce DMA
 426                  *      issues (e.g. floppy DMA transfer overrun/underrun).
 427                  */
 428                 if ((pr->power.bm_activity & 0x1) &&
 429                     cx->demotion.threshold.bm) {
 430                         local_irq_enable();
 431                         next_state = cx->demotion.state;
 432                         goto end;
 433                 }
 434         }
 435
 436 #ifdef CONFIG_HOTPLUG_CPU
 437         /*
 438          * Check for P_LVL2_UP flag before entering C2 and above on
 439          * an SMP system. We do it here instead of doing it at _CST/P_LVL
 440          * detection phase, to work cleanly with logical CPU hotplug.
 441          */
 442         if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 443             !pr->flags.has_cst && !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 444                 cx = &pr->power.states[ACPI_STATE_C1];
 445 #endif
 446
 447         /*
 448          * Sleep:
 449          * ------
 450          * Invoke the current Cx state to put the processor to sleep.
 451          */
 452         if (cx->type == ACPI_STATE_C2 || cx->type == ACPI_STATE_C3) {
 453                 current_thread_info()->status &= ~TS_POLLING;
 454                 /*
 455                  * TS_POLLING-cleared state must be visible before we
 456                  * test NEED_RESCHED:
 457                  */
 458                 smp_mb();
 459                 if (need_resched()) {
 460                         current_thread_info()->status |= TS_POLLING;
 461                         local_irq_enable();
 462                         return;
 463                 }
 464         }
 465
 466         switch (cx->type) {
 467
 468         case ACPI_STATE_C1:
 469                 /*
 470                  * Invoke C1.
 471                  * Use the appropriate idle routine, the one that would
 472                  * be used without acpi C-states.
 473                  */
 474                 if (pm_idle_save)
 475                         pm_idle_save();
 476                 else
 477                         acpi_safe_halt();
 478
 479                 /*
 480                  * TBD: Can't get time duration while in C1, as resumes
 481                  *      go to an ISR rather than here.  Need to instrument
 482                  *      base interrupt handler.
 483                  *
 484                  * Note: the TSC better not stop in C1, sched_clock() will
 485                  *       skew otherwise.
 486                  */
 487                 sleep_ticks = 0xFFFFFFFF;
 488                 break;
 489
 490         case ACPI_STATE_C2:
 491                 /* Get start time (ticks) */
 492                 t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 493                 /* Tell the scheduler that we are going deep-idle: */
 494                 sched_clock_idle_sleep_event();
 495                 /* Invoke C2 */
 496                 acpi_state_timer_broadcast(pr, cx, 1);
 497                 acpi_cstate_enter(cx);
 498                 /* Get end time (ticks) */
 499                 t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 500
 501 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86_TSC)
 502                 /* TSC halts in C2, so notify users */
 503                 mark_tsc_unstable("possible TSC halt in C2");
 504 #endif
 505                 /* Compute time (ticks) that we were actually asleep */
 506                 sleep_ticks = ticks_elapsed(t1, t2);
 507
 508                 /* Tell the scheduler how much we idled: */
 509                 sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 510
 511                 /* Re-enable interrupts */
 512                 local_irq_enable();
 513                 /* Do not account our idle-switching overhead: */
 514                 sleep_ticks -= cx->latency_ticks + C2_OVERHEAD;
 515
 516                 current_thread_info()->status |= TS_POLLING;
 517                 acpi_state_timer_broadcast(pr, cx, 0);
 518                 break;
 519
 520         case ACPI_STATE_C3:
 521                 /*
 522                  * disable bus master
 523                  * bm_check implies we need ARB_DIS
 524                  * !bm_check implies we need cache flush
 525                  * bm_control implies whether we can do ARB_DIS
 526                  *
 527                  * That leaves a case where bm_check is set and bm_control is
 528                  * not set. In that case we cannot do much, we enter C3
 529                  * without doing anything.
 530                  */
 531                 if (pr->flags.bm_check && pr->flags.bm_control) {
 532                         if (atomic_inc_return(&c3_cpu_count) ==
 533                             num_online_cpus()) {
 534                                 /*
 535                                  * All CPUs are trying to go to C3
 536                                  * Disable bus master arbitration
 537                                  */
 538                                 acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1);
 539                         }
 540                 } else if (!pr->flags.bm_check) {
 541                         /* SMP with no shared cache... Invalidate cache  */
 542                         ACPI_FLUSH_CPU_CACHE();
 543                 }
 544
 545                 /* Get start time (ticks) */
 546                 t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 547                 /* Invoke C3 */
 548                 acpi_state_timer_broadcast(pr, cx, 1);
 549                 /* Tell the scheduler that we are going deep-idle: */
 550                 sched_clock_idle_sleep_event();
 551                 acpi_cstate_enter(cx);
 552                 /* Get end time (ticks) */
 553                 t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 554                 if (pr->flags.bm_check && pr->flags.bm_control) {
 555                         /* Enable bus master arbitration */
 556                         atomic_dec(&c3_cpu_count);
 557                         acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0);
 558                 }
 559
 560 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86_TSC)
 561                 /* TSC halts in C3, so notify users */
 562                 mark_tsc_unstable("TSC halts in C3");
 563 #endif
 564                 /* Compute time (ticks) that we were actually asleep */
 565                 sleep_ticks = ticks_elapsed(t1, t2);
 566                 /* Tell the scheduler how much we idled: */
 567                 sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 568
 569                 /* Re-enable interrupts */
 570                 local_irq_enable();
 571                 /* Do not account our idle-switching overhead: */
 572                 sleep_ticks -= cx->latency_ticks + C3_OVERHEAD;
 573
 574                 current_thread_info()->status |= TS_POLLING;
 575                 acpi_state_timer_broadcast(pr, cx, 0);
 576                 break;
 577
 578         default:
 579                 local_irq_enable();
 580                 return;
 581         }
 582         cx->usage++;
 583         if ((cx->type != ACPI_STATE_C1) && (sleep_ticks > 0))
 584                 cx->time += sleep_ticks;
 585
 586         next_state = pr->power.state;
 587
 588 #ifdef CONFIG_HOTPLUG_CPU
 589         /* Don't do promotion/demotion */
 590         if ((cx->type == ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 591             !pr->flags.has_cst && !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) {
 592                 next_state = cx;
 593                 goto end;
 594         }
 595 #endif
 596
 597         /*
 598          * Promotion?
 599          * ----------
 600          * Track the number of longs (time asleep is greater than threshold)
 601          * and promote when the count threshold is reached.  Note that bus
 602          * mastering activity may prevent promotions.
 603          * Do not promote above max_cstate.
 604          */
 605         if (cx->promotion.state &&
 606             ((cx->promotion.state - pr->power.states) <= max_cstate)) {
 607                 if (sleep_ticks > cx->promotion.threshold.ticks &&
 608                   cx->promotion.state->latency <= system_latency_constraint()) {
 609                         cx->promotion.count++;
 610                         cx->demotion.count = 0;
 611                         if (cx->promotion.count >=
 612                             cx->promotion.threshold.count) {
 613                                 if (pr->flags.bm_check) {
 614                                         if (!
 615                                             (pr->power.bm_activity & cx->
 616                                              promotion.threshold.bm)) {
 617                                                 next_state =
 618                                                     cx->promotion.state;
 619                                                 goto end;
 620                                         }
 621                                 } else {
 622                                         next_state = cx->promotion.state;
 623                                         goto end;
 624                                 }
 625                         }
 626                 }
 627         }
 628
 629         /*
 630          * Demotion?
 631          * ---------
 632          * Track the number of shorts (time asleep is less than time threshold)
 633          * and demote when the usage threshold is reached.
 634          */
 635         if (cx->demotion.state) {
 636                 if (sleep_ticks < cx->demotion.threshold.ticks) {
 637                         cx->demotion.count++;
 638                         cx->promotion.count = 0;
 639                         if (cx->demotion.count >= cx->demotion.threshold.count) {
 640                                 next_state = cx->demotion.state;
 641                                 goto end;
 642                         }
 643                 }
 644         }
 645
 646       end:
 647         /*
 648          * Demote if current state exceeds max_cstate
 649          * or if the latency of the current state is unacceptable
 650          */
 651         if ((pr->power.state - pr->power.states) > max_cstate ||
 652                 pr->power.state->latency > system_latency_constraint()) {
 653                 if (cx->demotion.state)
 654                         next_state = cx->demotion.state;
 655         }
 656
 657         /*
 658          * New Cx State?
 659          * -------------
 660          * If we're going to start using a new Cx state we must clean up
 661          * from the previous and prepare to use the new.
 662          */
 663         if (next_state != pr->power.state)
 664                 acpi_processor_power_activate(pr, next_state);
 665 }
 666
 667 static int acpi_processor_set_power_policy(struct acpi_processor *pr)
 668 {
 669         unsigned int i;
 670         unsigned int state_is_set = 0;
 671         struct acpi_processor_cx *lower = NULL;
 672         struct acpi_processor_cx *higher = NULL;
 673         struct acpi_processor_cx *cx;
 674
 675
 676         if (!pr)
 677                 return -EINVAL;
 678
 679         /*
 680          * This function sets the default Cx state policy (OS idle handler).
 681          * Our scheme is to promote quickly to C2 but more conservatively
 682          * to C3.  We're favoring C2  for its characteristics of low latency
 683          * (quick response), good power savings, and ability to allow bus
 684          * mastering activity.  Note that the Cx state policy is completely
 685          * customizable and can be altered dynamically.
 686          */
 687
 688         /* startup state */
 689         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 690                 cx = &pr->power.states[i];
 691                 if (!cx->valid)
 692                         continue;
 693
 694                 if (!state_is_set)
 695                         pr->power.state = cx;
 696                 state_is_set++;
 697                 break;
 698         }
 699
 700         if (!state_is_set)
 701                 return -ENODEV;
 702
 703         /* demotion */
 704         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 705                 cx = &pr->power.states[i];
 706                 if (!cx->valid)
 707                         continue;
 708
 709                 if (lower) {
 710                         cx->demotion.state = lower;
 711                         cx->demotion.threshold.ticks = cx->latency_ticks;
 712                         cx->demotion.threshold.count = 1;
 713                         if (cx->type == ACPI_STATE_C3)
 714                                 cx->demotion.threshold.bm = bm_history;
 715                 }
 716
 717                 lower = cx;
 718         }
 719
 720         /* promotion */
 721         for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
 722                 cx = &pr->power.states[i];
 723                 if (!cx->valid)
 724                         continue;
 725
 726                 if (higher) {
 727                         cx->promotion.state = higher;
 728                         cx->promotion.threshold.ticks = cx->latency_ticks;
 729                         if (cx->type >= ACPI_STATE_C2)
 730                                 cx->promotion.threshold.count = 4;
 731                         else
 732                                 cx->promotion.threshold.count = 10;
 733                         if (higher->type == ACPI_STATE_C3)
 734                                 cx->promotion.threshold.bm = bm_history;
 735                 }
 736
 737                 higher = cx;
 738         }
 739
 740         return 0;
 741 }
 742
 743 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 744 {
 745
 746         if (!pr)
 747                 return -EINVAL;
 748
 749         if (!pr->pblk)
 750                 return -ENODEV;
 751
 752         /* if info is obtained from pblk/fadt, type equals state */
 753         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 754         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 755
 756 #ifndef CONFIG_HOTPLUG_CPU
 757         /*
 758          * Check for P_LVL2_UP flag before entering C2 and above on
 759          * an SMP system.
 760          */
 761         if ((num_online_cpus() > 1) &&
 762             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 763                 return -ENODEV;
 764 #endif
 765
 766         /* determine C2 and C3 address from pblk */
 767         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 768         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 769
 770         /* determine latencies from FADT */
 771         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 772         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 773
 774         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 775                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 776                           pr->power.states[ACPI_STATE_C2].address,
 777                           pr->power.states[ACPI_STATE_C3].address));
 778
 779         return 0;
 780 }
 781
 782 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 783 {
 784         if (!pr->power.states[ACPI_STATE_C1].valid) {
 785                 /* set the first C-State to C1 */
 786                 /* all processors need to support C1 */
 787                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 788                 pr->power.states[ACPI_STATE_C1].valid = 1;
 789         }
 790         /* the C0 state only exists as a filler in our array */
 791         pr->power.states[ACPI_STATE_C0].valid = 1;
 792         return 0;
 793 }
 794
 795 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 796 {
 797         acpi_status status = 0;
 798         acpi_integer count;
 799         int current_count;
 800         int i;
 801         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 802         union acpi_object *cst;
 803
 804
 805         if (nocst)
 806                 return -ENODEV;
 807
 808         current_count = 0;
 809
 810         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 811         if (ACPI_FAILURE(status)) {
 812                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 813                 return -ENODEV;
 814         }
 815
 816         cst = buffer.pointer;
 817
 818         /* There must be at least 2 elements */
 819         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 820                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 821                 status = -EFAULT;
 822                 goto end;
 823         }
 824
 825         count = cst->package.elements[0].integer.value;
 826
 827         /* Validate number of power states. */
 828         if (count < 1 || count != cst->package.count - 1) {
 829                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 830                 status = -EFAULT;
 831                 goto end;
 832         }
 833
 834         /* Tell driver that at least _CST is supported. */
 835         pr->flags.has_cst = 1;
 836
 837         for (i = 1; i <= count; i++) {
 838                 union acpi_object *element;
 839                 union acpi_object *obj;
 840                 struct acpi_power_register *reg;
 841                 struct acpi_processor_cx cx;
 842
 843                 memset(&cx, 0, sizeof(cx));
 844
 845                 element = &(cst->package.elements[i]);
 846                 if (element->type != ACPI_TYPE_PACKAGE)
 847                         continue;
 848
 849                 if (element->package.count != 4)
 850                         continue;
 851
 852                 obj = &(element->package.elements[0]);
 853
 854                 if (obj->type != ACPI_TYPE_BUFFER)
 855                         continue;
 856
 857                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 858
 859                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 860                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 861                         continue;
 862
 863                 /* There should be an easy way to extract an integer... */
 864                 obj = &(element->package.elements[1]);
 865                 if (obj->type != ACPI_TYPE_INTEGER)
 866                         continue;
 867
 868                 cx.type = obj->integer.value;
 869                 /*
 870                  * Some buggy BIOSes won't list C1 in _CST -
 871                  * Let acpi_processor_get_power_info_default() handle them later
 872                  */
 873                 if (i == 1 && cx.type != ACPI_STATE_C1)
 874                         current_count++;
 875
 876                 cx.address = reg->address;
 877                 cx.index = current_count + 1;
 878
 879                 cx.space_id = ACPI_CSTATE_SYSTEMIO;
 880                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 881                         if (acpi_processor_ffh_cstate_probe
 882                                         (pr->id, &cx, reg) == 0) {
 883                                 cx.space_id = ACPI_CSTATE_FFH;
 884                         } else if (cx.type != ACPI_STATE_C1) {
 885                                 /*
 886                                  * C1 is a special case where FIXED_HARDWARE
 887                                  * can be handled in non-MWAIT way as well.
 888                                  * In that case, save this _CST entry info.
 889                                  * That is, we retain space_id of SYSTEM_IO for
 890                                  * halt based C1.
 891                                  * Otherwise, ignore this info and continue.
 892                                  */
 893                                 continue;
 894                         }
 895                 }
 896
 897                 obj = &(element->package.elements[2]);
 898                 if (obj->type != ACPI_TYPE_INTEGER)
 899                         continue;
 900
 901                 cx.latency = obj->integer.value;
 902
 903                 obj = &(element->package.elements[3]);
 904                 if (obj->type != ACPI_TYPE_INTEGER)
 905                         continue;
 906
 907                 cx.power = obj->integer.value;
 908
 909                 current_count++;
 910                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 911
 912                 /*
 913                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 914                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 915                  */
 916                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 917                         printk(KERN_WARNING
 918                                "Limiting number of power states to max (%d)\n",
 919                                ACPI_PROCESSOR_MAX_POWER);
 920                         printk(KERN_WARNING
 921                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 922                         break;
 923                 }
 924         }
 925
 926         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 927                           current_count));
 928
 929         /* Validate number of power states discovered */
 930         if (current_count < 2)
 931                 status = -EFAULT;
 932
 933       end:
 934         kfree(buffer.pointer);
 935
 936         return status;
 937 }
 938
 939 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 940 {
 941
 942         if (!cx->address)
 943                 return;
 944
 945         /*
 946          * C2 latency must be less than or equal to 100
 947          * microseconds.
 948          */
 949         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 950                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 951                                   "latency too large [%d]\n", cx->latency));
 952                 return;
 953         }
 954
 955         /*
 956          * Otherwise we've met all of our C2 requirements.
 957          * Normalize the C2 latency to expidite policy
 958          */
 959         cx->valid = 1;
 960         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
 961
 962         return;
 963 }
 964
 965 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 966                                            struct acpi_processor_cx *cx)
 967 {
 968         static int bm_check_flag;
 969
 970
 971         if (!cx->address)
 972                 return;
 973
 974         /*
 975          * C3 latency must be less than or equal to 1000
 976          * microseconds.
 977          */
 978         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 979                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 980                                   "latency too large [%d]\n", cx->latency));
 981                 return;
 982         }
 983
 984         /*
 985          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 986          * DMA transfers are used by any ISA device to avoid livelock.
 987          * Note that we could disable Type-F DMA (as recommended by
 988          * the erratum), but this is known to disrupt certain ISA
 989          * devices thus we take the conservative approach.
 990          */
 991         else if (errata.piix4.fdma) {
 992                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 993                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 994                 return;
 995         }
 996
 997         /* All the logic here assumes flags.bm_check is same across all CPUs */
 998         if (!bm_check_flag) {
 999                 /* Determine whether bm_check is needed based on CPU  */
1000                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
1001                 bm_check_flag = pr->flags.bm_check;
1002         } else {
1003                 pr->flags.bm_check = bm_check_flag;
1004         }
1005
1006         if (pr->flags.bm_check) {
1007                 if (!pr->flags.bm_control) {
1008                         if (pr->flags.has_cst != 1) {
1009                                 /* bus mastering control is necessary */
1010                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1011                                         "C3 support requires BM control\n"));
1012                                 return;
1013                         } else {
1014                                 /* Here we enter C3 without bus mastering */
1015                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1016                                         "C3 support without BM control\n"));
1017                         }
1018                 }
1019         } else {
1020                 /*
1021                  * WBINVD should be set in fadt, for C3 state to be
1022                  * supported on when bm_check is not required.
1023                  */
1024                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
1025                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
1026                                           "Cache invalidation should work properly"
1027                                           " for C3 to be enabled on SMP systems\n"));
1028                         return;
1029                 }
1030                 acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0);
1031         }
1032
1033         /*
1034          * Otherwise we've met all of our C3 requirements.
1035          * Normalize the C3 latency to expidite policy.  Enable
1036          * checking of bus mastering status (bm_check) so we can
1037          * use this in our C3 policy
1038          */
1039         cx->valid = 1;
1040         cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
1041
1042         return;
1043 }
1044
1045 static int acpi_processor_power_verify(struct acpi_processor *pr)
1046 {
1047         unsigned int i;
1048         unsigned int working = 0;
1049
1050         pr->power.timer_broadcast_on_state = INT_MAX;
1051
1052         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
1053                 struct acpi_processor_cx *cx = &pr->power.states[i];
1054
1055                 switch (cx->type) {
1056                 case ACPI_STATE_C1:
1057                         cx->valid = 1;
1058                         break;
1059
1060                 case ACPI_STATE_C2:
1061                         acpi_processor_power_verify_c2(cx);
1062                         if (cx->valid)
1063                                 acpi_timer_check_state(i, pr, cx);
1064                         break;
1065
1066                 case ACPI_STATE_C3:
1067                         acpi_processor_power_verify_c3(pr, cx);
1068                         if (cx->valid)
1069                                 acpi_timer_check_state(i, pr, cx);
1070                         break;
1071                 }
1072
1073                 if (cx->valid)
1074                         working++;
1075         }
1076
1077         acpi_propagate_timer_broadcast(pr);
1078
1079         return (working);
1080 }
1081
1082 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1083 {
1084         unsigned int i;
1085         int result;
1086
1087
1088         /* NOTE: the idle thread may not be running while calling
1089          * this function */
1090
1091         /* Zero initialize all the C-states info. */
1092         memset(pr->power.states, 0, sizeof(pr->power.states));
1093
1094         result = acpi_processor_get_power_info_cst(pr);
1095         if (result == -ENODEV)
1096                 result = acpi_processor_get_power_info_fadt(pr);
1097
1098         if (result)
1099                 return result;
1100
1101         acpi_processor_get_power_info_default(pr);
1102
1103         pr->power.count = acpi_processor_power_verify(pr);
1104
1105         /*
1106          * Set Default Policy
1107          * ------------------
1108          * Now that we know which states are supported, set the default
1109          * policy.  Note that this policy can be changed dynamically
1110          * (e.g. encourage deeper sleeps to conserve battery life when
1111          * not on AC).
1112          */
1113         result = acpi_processor_set_power_policy(pr);
1114         if (result)
1115                 return result;
1116
1117         /*
1118          * if one state of type C2 or C3 is available, mark this
1119          * CPU as being "idle manageable"
1120          */
1121         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
1122                 if (pr->power.states[i].valid) {
1123                         pr->power.count = i;
1124                         if (pr->power.states[i].type >= ACPI_STATE_C2)
1125                                 pr->flags.power = 1;
1126                 }
1127         }
1128
1129         return 0;
1130 }
1131
1132 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1133 {
1134         int result = 0;
1135
1136
1137         if (!pr)
1138                 return -EINVAL;
1139
1140         if (nocst) {
1141                 return -ENODEV;
1142         }
1143
1144         if (!pr->flags.power_setup_done)
1145                 return -ENODEV;
1146
1147         /* Fall back to the default idle loop */
1148         pm_idle = pm_idle_save;
1149         synchronize_sched();    /* Relies on interrupts forcing exit from idle. */
1150
1151         pr->flags.power = 0;
1152         result = acpi_processor_get_power_info(pr);
1153         if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
1154                 pm_idle = acpi_processor_idle;
1155
1156         return result;
1157 }
1158
1159 /* proc interface */
1160
1161 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
1162 {
1163         struct acpi_processor *pr = seq->private;
1164         unsigned int i;
1165
1166
1167         if (!pr)
1168                 goto end;
1169
1170         seq_printf(seq, "active state:            C%zd\n"
1171                    "max_cstate:              C%d\n"
1172                    "bus master activity:     %08x\n"
1173                    "maximum allowed latency: %d usec\n",
1174                    pr->power.state ? pr->power.state - pr->power.states : 0,
1175                    max_cstate, (unsigned)pr->power.bm_activity,
1176                    system_latency_constraint());
1177
1178         seq_puts(seq, "states:\n");
1179
1180         for (i = 1; i <= pr->power.count; i++) {
1181                 seq_printf(seq, "   %cC%d:                  ",
1182                            (&pr->power.states[i] ==
1183                             pr->power.state ? '*' : ' '), i);
1184
1185                 if (!pr->power.states[i].valid) {
1186                         seq_puts(seq, "<not supported>\n");
1187                         continue;
1188                 }
1189
1190                 switch (pr->power.states[i].type) {
1191                 case ACPI_STATE_C1:
1192                         seq_printf(seq, "type[C1] ");
1193                         break;
1194                 case ACPI_STATE_C2:
1195                         seq_printf(seq, "type[C2] ");
1196                         break;
1197                 case ACPI_STATE_C3:
1198                         seq_printf(seq, "type[C3] ");
1199                         break;
1200                 default:
1201                         seq_printf(seq, "type[--] ");
1202                         break;
1203                 }
1204
1205                 if (pr->power.states[i].promotion.state)
1206                         seq_printf(seq, "promotion[C%zd] ",
1207                                    (pr->power.states[i].promotion.state -
1208                                     pr->power.states));
1209                 else
1210                         seq_puts(seq, "promotion[--] ");
1211
1212                 if (pr->power.states[i].demotion.state)
1213                         seq_printf(seq, "demotion[C%zd] ",
1214                                    (pr->power.states[i].demotion.state -
1215                                     pr->power.states));
1216                 else
1217                         seq_puts(seq, "demotion[--] ");
1218
1219                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
1220                            pr->power.states[i].latency,
1221                            pr->power.states[i].usage,
1222                            (unsigned long long)pr->power.states[i].time);
1223         }
1224
1225       end:
1226         return 0;
1227 }
1228
1229 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
1230 {
1231         return single_open(file, acpi_processor_power_seq_show,
1232                            PDE(inode)->data);
1233 }
1234
1235 static const struct file_operations acpi_processor_power_fops = {
1236         .open = acpi_processor_power_open_fs,
1237         .read = seq_read,
1238         .llseek = seq_lseek,
1239         .release = single_release,
1240 };
1241
1242 #ifdef CONFIG_SMP
1243 static void smp_callback(void *v)
1244 {
1245         /* we already woke the CPU up, nothing more to do */
1246 }
1247
1248 /*
1249  * This function gets called when a part of the kernel has a new latency
1250  * requirement.  This means we need to get all processors out of their C-state,
1251  * and then recalculate a new suitable C-state. Just do a cross-cpu IPI; that
1252  * wakes them all right up.
1253  */
1254 static int acpi_processor_latency_notify(struct notifier_block *b,
1255                 unsigned long l, void *v)
1256 {
1257         smp_call_function(smp_callback, NULL, 0, 1);
1258         return NOTIFY_OK;
1259 }
1260
1261 static struct notifier_block acpi_processor_latency_notifier = {
1262         .notifier_call = acpi_processor_latency_notify,
1263 };
1264 #endif
1265
1266 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1267                               struct acpi_device *device)
1268 {
1269         acpi_status status = 0;
1270         static int first_run;
1271         struct proc_dir_entry *entry = NULL;
1272         unsigned int i;
1273
1274
1275         if (!first_run) {
1276                 dmi_check_system(processor_power_dmi_table);
1277                 if (max_cstate < ACPI_C_STATES_MAX)
1278                         printk(KERN_NOTICE
1279                                "ACPI: processor limited to max C-state %d\n",
1280                                max_cstate);
1281                 first_run++;
1282 #ifdef CONFIG_SMP
1283                 register_latency_notifier(&acpi_processor_latency_notifier);
1284 #endif
1285         }
1286
1287         if (!pr)
1288                 return -EINVAL;
1289
1290         if (acpi_gbl_FADT.cst_control && !nocst) {
1291                 status =
1292                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1293                 if (ACPI_FAILURE(status)) {
1294                         ACPI_EXCEPTION((AE_INFO, status,
1295                                         "Notifying BIOS of _CST ability failed"));
1296                 }
1297         }
1298
1299         acpi_processor_get_power_info(pr);
1300
1301         /*
1302          * Install the idle handler if processor power management is supported.
1303          * Note that we use previously set idle handler will be used on
1304          * platforms that only support C1.
1305          */
1306         if ((pr->flags.power) && (!boot_option_idle_override)) {
1307                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1308                 for (i = 1; i <= pr->power.count; i++)
1309                         if (pr->power.states[i].valid)
1310                                 printk(" C%d[C%d]", i,
1311                                        pr->power.states[i].type);
1312                 printk(")\n");
1313
1314                 if (pr->id == 0) {
1315                         pm_idle_save = pm_idle;
1316                         pm_idle = acpi_processor_idle;
1317                 }
1318         }
1319
1320         /* 'power' [R] */
1321         entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1322                                   S_IRUGO, acpi_device_dir(device));
1323         if (!entry)
1324                 return -EIO;
1325         else {
1326                 entry->proc_fops = &acpi_processor_power_fops;
1327                 entry->data = acpi_driver_data(device);
1328                 entry->owner = THIS_MODULE;
1329         }
1330
1331         pr->flags.power_setup_done = 1;
1332
1333         return 0;
1334 }
1335
1336 int acpi_processor_power_exit(struct acpi_processor *pr,
1337                               struct acpi_device *device)
1338 {
1339
1340         pr->flags.power_setup_done = 0;
1341
1342         if (acpi_device_dir(device))
1343                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1344                                   acpi_device_dir(device));
1345
1346         /* Unregister the idle handler when processor #0 is removed. */
1347         if (pr->id == 0) {
1348                 pm_idle = pm_idle_save;
1349
1350                 /*
1351                  * We are about to unload the current idle thread pm callback
1352                  * (pm_idle), Wait for all processors to update cached/local
1353                  * copies of pm_idle before proceeding.
1354                  */
1355                 cpu_idle_wait();
1356 #ifdef CONFIG_SMP
1357                 unregister_latency_notifier(&acpi_processor_latency_notifier);
1358 #endif
1359         }
1360
1361         return 0;
1362 }