drivers/acpi/processor_idle.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * processor_idle - idle state submodule to the ACPI processor driver
   4  *
   5  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   6  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   7  *  Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
   8  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
   9  *                      - Added processor hotplug support
  10  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  11  *                      - Added support for C3 on SMP
  12  */
  13 #define pr_fmt(fmt) "ACPI: " fmt
  14
  15 #include <linux/module.h>
  16 #include <linux/acpi.h>
  17 #include <linux/dmi.h>
  18 #include <linux/sched.h>       /* need_resched() */
  19 #include <linux/sort.h>
  20 #include <linux/tick.h>
  21 #include <linux/cpuidle.h>
  22 #include <linux/cpu.h>
  23 #include <linux/minmax.h>
  24 #include <acpi/processor.h>
  25
  26 /*
  27  * Include the apic definitions for x86 to have the APIC timer related defines
  28  * available also for UP (on SMP it gets magically included via linux/smp.h).
  29  * asm/acpi.h is not an option, as it would require more include magic. Also
  30  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  31  */
  32 #ifdef CONFIG_X86
  33 #include <asm/apic.h>
  34 #include <asm/cpu.h>
  35 #endif
  36
  37 #define ACPI_IDLE_STATE_START   (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
  38
  39 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  40 module_param(max_cstate, uint, 0000);
  41 static unsigned int nocst __read_mostly;
  42 module_param(nocst, uint, 0000);
  43 static int bm_check_disable __read_mostly;
  44 module_param(bm_check_disable, uint, 0000);
  45
  46 static unsigned int latency_factor __read_mostly = 2;
  47 module_param(latency_factor, uint, 0644);
  48
  49 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
  50
  51 struct cpuidle_driver acpi_idle_driver = {
  52         .name =         "acpi_idle",
  53         .owner =        THIS_MODULE,
  54 };
  55
  56 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
  57 static
  58 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
  59
  60 static int disabled_by_idle_boot_param(void)
  61 {
  62         return boot_option_idle_override == IDLE_POLL ||
  63                 boot_option_idle_override == IDLE_HALT;
  64 }
  65
  66 /*
  67  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  68  * For now disable this. Probably a bug somewhere else.
  69  *
  70  * To skip this limit, boot/load with a large max_cstate limit.
  71  */
  72 static int set_max_cstate(const struct dmi_system_id *id)
  73 {
  74         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  75                 return 0;
  76
  77         pr_notice("%s detected - limiting to C%ld max_cstate."
  78                   " Override with \"processor.max_cstate=%d\"\n", id->ident,
  79                   (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  80
  81         max_cstate = (long)id->driver_data;
  82
  83         return 0;
  84 }
  85
  86 static const struct dmi_system_id processor_power_dmi_table[] = {
  87         { set_max_cstate, "Clevo 5600D", {
  88           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
  89           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
  90          (void *)2},
  91         { set_max_cstate, "Pavilion zv5000", {
  92           DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
  93           DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
  94          (void *)1},
  95         { set_max_cstate, "Asus L8400B", {
  96           DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
  97           DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
  98          (void *)1},
  99         /* T40 can not handle C3 idle state */
 100         { set_max_cstate, "IBM ThinkPad T40", {
 101           DMI_MATCH(DMI_SYS_VENDOR, "IBM"),
 102           DMI_MATCH(DMI_PRODUCT_NAME, "23737CU")},
 103          (void *)2},
 104         {},
 105 };
 106
 107
 108 /*
 109  * Callers should disable interrupts before the call and enable
 110  * interrupts after return.
 111  */
 112 static void __cpuidle acpi_safe_halt(void)
 113 {
 114         if (!tif_need_resched()) {
 115                 safe_halt();
 116                 local_irq_disable();
 117         }
 118 }
 119
 120 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 121
 122 /*
 123  * Some BIOS implementations switch to C3 in the published C2 state.
 124  * This seems to be a common problem on AMD boxen, but other vendors
 125  * are affected too. We pick the most conservative approach: we assume
 126  * that the local APIC stops in both C2 and C3.
 127  */
 128 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 129                                    struct acpi_processor_cx *cx)
 130 {
 131         struct acpi_processor_power *pwr = &pr->power;
 132         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 133
 134         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 135                 return;
 136
 137         if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
 138                 type = ACPI_STATE_C1;
 139
 140         /*
 141          * Check, if one of the previous states already marked the lapic
 142          * unstable
 143          */
 144         if (pwr->timer_broadcast_on_state < state)
 145                 return;
 146
 147         if (cx->type >= type)
 148                 pr->power.timer_broadcast_on_state = state;
 149 }
 150
 151 static void __lapic_timer_propagate_broadcast(void *arg)
 152 {
 153         struct acpi_processor *pr = (struct acpi_processor *) arg;
 154
 155         if (pr->power.timer_broadcast_on_state < INT_MAX)
 156                 tick_broadcast_enable();
 157         else
 158                 tick_broadcast_disable();
 159 }
 160
 161 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 162 {
 163         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
 164                                  (void *)pr, 1);
 165 }
 166
 167 /* Power(C) State timer broadcast control */
 168 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 169                                         struct acpi_processor_cx *cx)
 170 {
 171         return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
 172 }
 173
 174 #else
 175
 176 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 177                                    struct acpi_processor_cx *cstate) { }
 178 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 179
 180 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 181                                         struct acpi_processor_cx *cx)
 182 {
 183         return false;
 184 }
 185
 186 #endif
 187
 188 #if defined(CONFIG_X86)
 189 static void tsc_check_state(int state)
 190 {
 191         switch (boot_cpu_data.x86_vendor) {
 192         case X86_VENDOR_HYGON:
 193         case X86_VENDOR_AMD:
 194         case X86_VENDOR_INTEL:
 195         case X86_VENDOR_CENTAUR:
 196         case X86_VENDOR_ZHAOXIN:
 197                 /*
 198                  * AMD Fam10h TSC will tick in all
 199                  * C/P/S0/S1 states when this bit is set.
 200                  */
 201                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 202                         return;
 203                 fallthrough;
 204         default:
 205                 /* TSC could halt in idle, so notify users */
 206                 if (state > ACPI_STATE_C1)
 207                         mark_tsc_unstable("TSC halts in idle");
 208         }
 209 }
 210 #else
 211 static void tsc_check_state(int state) { return; }
 212 #endif
 213
 214 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 215 {
 216
 217         if (!pr->pblk)
 218                 return -ENODEV;
 219
 220         /* if info is obtained from pblk/fadt, type equals state */
 221         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 222         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 223
 224 #ifndef CONFIG_HOTPLUG_CPU
 225         /*
 226          * Check for P_LVL2_UP flag before entering C2 and above on
 227          * an SMP system.
 228          */
 229         if ((num_online_cpus() > 1) &&
 230             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 231                 return -ENODEV;
 232 #endif
 233
 234         /* determine C2 and C3 address from pblk */
 235         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 236         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 237
 238         /* determine latencies from FADT */
 239         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
 240         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
 241
 242         /*
 243          * FADT specified C2 latency must be less than or equal to
 244          * 100 microseconds.
 245          */
 246         if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 247                 acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n",
 248                                   acpi_gbl_FADT.c2_latency);
 249                 /* invalidate C2 */
 250                 pr->power.states[ACPI_STATE_C2].address = 0;
 251         }
 252
 253         /*
 254          * FADT supplied C3 latency must be less than or equal to
 255          * 1000 microseconds.
 256          */
 257         if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 258                 acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n",
 259                                   acpi_gbl_FADT.c3_latency);
 260                 /* invalidate C3 */
 261                 pr->power.states[ACPI_STATE_C3].address = 0;
 262         }
 263
 264         acpi_handle_debug(pr->handle, "lvl2[0x%08x] lvl3[0x%08x]\n",
 265                           pr->power.states[ACPI_STATE_C2].address,
 266                           pr->power.states[ACPI_STATE_C3].address);
 267
 268         snprintf(pr->power.states[ACPI_STATE_C2].desc,
 269                          ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
 270                          pr->power.states[ACPI_STATE_C2].address);
 271         snprintf(pr->power.states[ACPI_STATE_C3].desc,
 272                          ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
 273                          pr->power.states[ACPI_STATE_C3].address);
 274
 275         return 0;
 276 }
 277
 278 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 279 {
 280         if (!pr->power.states[ACPI_STATE_C1].valid) {
 281                 /* set the first C-State to C1 */
 282                 /* all processors need to support C1 */
 283                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 284                 pr->power.states[ACPI_STATE_C1].valid = 1;
 285                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 286
 287                 snprintf(pr->power.states[ACPI_STATE_C1].desc,
 288                          ACPI_CX_DESC_LEN, "ACPI HLT");
 289         }
 290         /* the C0 state only exists as a filler in our array */
 291         pr->power.states[ACPI_STATE_C0].valid = 1;
 292         return 0;
 293 }
 294
 295 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 296 {
 297         int ret;
 298
 299         if (nocst)
 300                 return -ENODEV;
 301
 302         ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
 303         if (ret)
 304                 return ret;
 305
 306         if (!pr->power.count)
 307                 return -EFAULT;
 308
 309         pr->flags.has_cst = 1;
 310         return 0;
 311 }
 312
 313 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 314                                            struct acpi_processor_cx *cx)
 315 {
 316         static int bm_check_flag = -1;
 317         static int bm_control_flag = -1;
 318
 319
 320         if (!cx->address)
 321                 return;
 322
 323         /*
 324          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 325          * DMA transfers are used by any ISA device to avoid livelock.
 326          * Note that we could disable Type-F DMA (as recommended by
 327          * the erratum), but this is known to disrupt certain ISA
 328          * devices thus we take the conservative approach.
 329          */
 330         else if (errata.piix4.fdma) {
 331                 acpi_handle_debug(pr->handle,
 332                                   "C3 not supported on PIIX4 with Type-F DMA\n");
 333                 return;
 334         }
 335
 336         /* All the logic here assumes flags.bm_check is same across all CPUs */
 337         if (bm_check_flag == -1) {
 338                 /* Determine whether bm_check is needed based on CPU  */
 339                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 340                 bm_check_flag = pr->flags.bm_check;
 341                 bm_control_flag = pr->flags.bm_control;
 342         } else {
 343                 pr->flags.bm_check = bm_check_flag;
 344                 pr->flags.bm_control = bm_control_flag;
 345         }
 346
 347         if (pr->flags.bm_check) {
 348                 if (!pr->flags.bm_control) {
 349                         if (pr->flags.has_cst != 1) {
 350                                 /* bus mastering control is necessary */
 351                                 acpi_handle_debug(pr->handle,
 352                                                   "C3 support requires BM control\n");
 353                                 return;
 354                         } else {
 355                                 /* Here we enter C3 without bus mastering */
 356                                 acpi_handle_debug(pr->handle,
 357                                                   "C3 support without BM control\n");
 358                         }
 359                 }
 360         } else {
 361                 /*
 362                  * WBINVD should be set in fadt, for C3 state to be
 363                  * supported on when bm_check is not required.
 364                  */
 365                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 366                         acpi_handle_debug(pr->handle,
 367                                           "Cache invalidation should work properly"
 368                                           " for C3 to be enabled on SMP systems\n");
 369                         return;
 370                 }
 371         }
 372
 373         /*
 374          * Otherwise we've met all of our C3 requirements.
 375          * Normalize the C3 latency to expidite policy.  Enable
 376          * checking of bus mastering status (bm_check) so we can
 377          * use this in our C3 policy
 378          */
 379         cx->valid = 1;
 380
 381         /*
 382          * On older chipsets, BM_RLD needs to be set
 383          * in order for Bus Master activity to wake the
 384          * system from C3.  Newer chipsets handle DMA
 385          * during C3 automatically and BM_RLD is a NOP.
 386          * In either case, the proper way to
 387          * handle BM_RLD is to set it and leave it set.
 388          */
 389         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 390
 391         return;
 392 }
 393
 394 static int acpi_cst_latency_cmp(const void *a, const void *b)
 395 {
 396         const struct acpi_processor_cx *x = a, *y = b;
 397
 398         if (!(x->valid && y->valid))
 399                 return 0;
 400         if (x->latency > y->latency)
 401                 return 1;
 402         if (x->latency < y->latency)
 403                 return -1;
 404         return 0;
 405 }
 406 static void acpi_cst_latency_swap(void *a, void *b, int n)
 407 {
 408         struct acpi_processor_cx *x = a, *y = b;
 409
 410         if (!(x->valid && y->valid))
 411                 return;
 412         swap(x->latency, y->latency);
 413 }
 414
 415 static int acpi_processor_power_verify(struct acpi_processor *pr)
 416 {
 417         unsigned int i;
 418         unsigned int working = 0;
 419         unsigned int last_latency = 0;
 420         unsigned int last_type = 0;
 421         bool buggy_latency = false;
 422
 423         pr->power.timer_broadcast_on_state = INT_MAX;
 424
 425         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 426                 struct acpi_processor_cx *cx = &pr->power.states[i];
 427
 428                 switch (cx->type) {
 429                 case ACPI_STATE_C1:
 430                         cx->valid = 1;
 431                         break;
 432
 433                 case ACPI_STATE_C2:
 434                         if (!cx->address)
 435                                 break;
 436                         cx->valid = 1;
 437                         break;
 438
 439                 case ACPI_STATE_C3:
 440                         acpi_processor_power_verify_c3(pr, cx);
 441                         break;
 442                 }
 443                 if (!cx->valid)
 444                         continue;
 445                 if (cx->type >= last_type && cx->latency < last_latency)
 446                         buggy_latency = true;
 447                 last_latency = cx->latency;
 448                 last_type = cx->type;
 449
 450                 lapic_timer_check_state(i, pr, cx);
 451                 tsc_check_state(cx->type);
 452                 working++;
 453         }
 454
 455         if (buggy_latency) {
 456                 pr_notice("FW issue: working around C-state latencies out of order\n");
 457                 sort(&pr->power.states[1], max_cstate,
 458                      sizeof(struct acpi_processor_cx),
 459                      acpi_cst_latency_cmp,
 460                      acpi_cst_latency_swap);
 461         }
 462
 463         lapic_timer_propagate_broadcast(pr);
 464
 465         return (working);
 466 }
 467
 468 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 469 {
 470         unsigned int i;
 471         int result;
 472
 473
 474         /* NOTE: the idle thread may not be running while calling
 475          * this function */
 476
 477         /* Zero initialize all the C-states info. */
 478         memset(pr->power.states, 0, sizeof(pr->power.states));
 479
 480         result = acpi_processor_get_power_info_cst(pr);
 481         if (result == -ENODEV)
 482                 result = acpi_processor_get_power_info_fadt(pr);
 483
 484         if (result)
 485                 return result;
 486
 487         acpi_processor_get_power_info_default(pr);
 488
 489         pr->power.count = acpi_processor_power_verify(pr);
 490
 491         /*
 492          * if one state of type C2 or C3 is available, mark this
 493          * CPU as being "idle manageable"
 494          */
 495         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 496                 if (pr->power.states[i].valid) {
 497                         pr->power.count = i;
 498                         pr->flags.power = 1;
 499                 }
 500         }
 501
 502         return 0;
 503 }
 504
 505 /**
 506  * acpi_idle_bm_check - checks if bus master activity was detected
 507  */
 508 static int acpi_idle_bm_check(void)
 509 {
 510         u32 bm_status = 0;
 511
 512         if (bm_check_disable)
 513                 return 0;
 514
 515         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 516         if (bm_status)
 517                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 518         /*
 519          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 520          * the true state of bus mastering activity; forcing us to
 521          * manually check the BMIDEA bit of each IDE channel.
 522          */
 523         else if (errata.piix4.bmisx) {
 524                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 525                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 526                         bm_status = 1;
 527         }
 528         return bm_status;
 529 }
 530
 531 static void wait_for_freeze(void)
 532 {
 533 #ifdef  CONFIG_X86
 534         /* No delay is needed if we are in guest */
 535         if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 536                 return;
 537 #endif
 538         /* Dummy wait op - must do something useless after P_LVL2 read
 539            because chipsets cannot guarantee that STPCLK# signal
 540            gets asserted in time to freeze execution properly. */
 541         inl(acpi_gbl_FADT.xpm_timer_block.address);
 542 }
 543
 544 /**
 545  * acpi_idle_do_entry - enter idle state using the appropriate method
 546  * @cx: cstate data
 547  *
 548  * Caller disables interrupt before call and enables interrupt after return.
 549  */
 550 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
 551 {
 552         if (cx->entry_method == ACPI_CSTATE_FFH) {
 553                 /* Call into architectural FFH based C-state */
 554                 acpi_processor_ffh_cstate_enter(cx);
 555         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 556                 acpi_safe_halt();
 557         } else {
 558                 /* IO port based C-state */
 559                 inb(cx->address);
 560                 wait_for_freeze();
 561         }
 562 }
 563
 564 /**
 565  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
 566  * @dev: the target CPU
 567  * @index: the index of suggested state
 568  */
 569 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 570 {
 571         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 572
 573         if (cx->type == ACPI_STATE_C3)
 574                 ACPI_FLUSH_CPU_CACHE();
 575
 576         while (1) {
 577
 578                 if (cx->entry_method == ACPI_CSTATE_HALT)
 579                         safe_halt();
 580                 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 581                         inb(cx->address);
 582                         wait_for_freeze();
 583                 } else
 584                         return -ENODEV;
 585
 586 #if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU)
 587                 cond_wakeup_cpu0();
 588 #endif
 589         }
 590
 591         /* Never reached */
 592         return 0;
 593 }
 594
 595 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
 596 {
 597         return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
 598                 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
 599 }
 600
 601 static int c3_cpu_count;
 602 static DEFINE_RAW_SPINLOCK(c3_lock);
 603
 604 /**
 605  * acpi_idle_enter_bm - enters C3 with proper BM handling
 606  * @drv: cpuidle driver
 607  * @pr: Target processor
 608  * @cx: Target state context
 609  * @index: index of target state
 610  */
 611 static int acpi_idle_enter_bm(struct cpuidle_driver *drv,
 612                                struct acpi_processor *pr,
 613                                struct acpi_processor_cx *cx,
 614                                int index)
 615 {
 616         static struct acpi_processor_cx safe_cx = {
 617                 .entry_method = ACPI_CSTATE_HALT,
 618         };
 619
 620         /*
 621          * disable bus master
 622          * bm_check implies we need ARB_DIS
 623          * bm_control implies whether we can do ARB_DIS
 624          *
 625          * That leaves a case where bm_check is set and bm_control is not set.
 626          * In that case we cannot do much, we enter C3 without doing anything.
 627          */
 628         bool dis_bm = pr->flags.bm_control;
 629
 630         /* If we can skip BM, demote to a safe state. */
 631         if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
 632                 dis_bm = false;
 633                 index = drv->safe_state_index;
 634                 if (index >= 0) {
 635                         cx = this_cpu_read(acpi_cstate[index]);
 636                 } else {
 637                         cx = &safe_cx;
 638                         index = -EBUSY;
 639                 }
 640         }
 641
 642         if (dis_bm) {
 643                 raw_spin_lock(&c3_lock);
 644                 c3_cpu_count++;
 645                 /* Disable bus master arbitration when all CPUs are in C3 */
 646                 if (c3_cpu_count == num_online_cpus())
 647                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 648                 raw_spin_unlock(&c3_lock);
 649         }
 650
 651         rcu_idle_enter();
 652
 653         acpi_idle_do_entry(cx);
 654
 655         rcu_idle_exit();
 656
 657         /* Re-enable bus master arbitration */
 658         if (dis_bm) {
 659                 raw_spin_lock(&c3_lock);
 660                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 661                 c3_cpu_count--;
 662                 raw_spin_unlock(&c3_lock);
 663         }
 664
 665         return index;
 666 }
 667
 668 static int acpi_idle_enter(struct cpuidle_device *dev,
 669                            struct cpuidle_driver *drv, int index)
 670 {
 671         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 672         struct acpi_processor *pr;
 673
 674         pr = __this_cpu_read(processors);
 675         if (unlikely(!pr))
 676                 return -EINVAL;
 677
 678         if (cx->type != ACPI_STATE_C1) {
 679                 if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
 680                         return acpi_idle_enter_bm(drv, pr, cx, index);
 681
 682                 /* C2 to C1 demotion. */
 683                 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
 684                         index = ACPI_IDLE_STATE_START;
 685                         cx = per_cpu(acpi_cstate[index], dev->cpu);
 686                 }
 687         }
 688
 689         if (cx->type == ACPI_STATE_C3)
 690                 ACPI_FLUSH_CPU_CACHE();
 691
 692         acpi_idle_do_entry(cx);
 693
 694         return index;
 695 }
 696
 697 static int acpi_idle_enter_s2idle(struct cpuidle_device *dev,
 698                                   struct cpuidle_driver *drv, int index)
 699 {
 700         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 701
 702         if (cx->type == ACPI_STATE_C3) {
 703                 struct acpi_processor *pr = __this_cpu_read(processors);
 704
 705                 if (unlikely(!pr))
 706                         return 0;
 707
 708                 if (pr->flags.bm_check) {
 709                         u8 bm_sts_skip = cx->bm_sts_skip;
 710
 711                         /* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
 712                         cx->bm_sts_skip = 1;
 713                         acpi_idle_enter_bm(drv, pr, cx, index);
 714                         cx->bm_sts_skip = bm_sts_skip;
 715
 716                         return 0;
 717                 } else {
 718                         ACPI_FLUSH_CPU_CACHE();
 719                 }
 720         }
 721         acpi_idle_do_entry(cx);
 722
 723         return 0;
 724 }
 725
 726 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 727                                            struct cpuidle_device *dev)
 728 {
 729         int i, count = ACPI_IDLE_STATE_START;
 730         struct acpi_processor_cx *cx;
 731         struct cpuidle_state *state;
 732
 733         if (max_cstate == 0)
 734                 max_cstate = 1;
 735
 736         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 737                 state = &acpi_idle_driver.states[count];
 738                 cx = &pr->power.states[i];
 739
 740                 if (!cx->valid)
 741                         continue;
 742
 743                 per_cpu(acpi_cstate[count], dev->cpu) = cx;
 744
 745                 if (lapic_timer_needs_broadcast(pr, cx))
 746                         state->flags |= CPUIDLE_FLAG_TIMER_STOP;
 747
 748                 if (cx->type == ACPI_STATE_C3) {
 749                         state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
 750                         if (pr->flags.bm_check)
 751                                 state->flags |= CPUIDLE_FLAG_RCU_IDLE;
 752                 }
 753
 754                 count++;
 755                 if (count == CPUIDLE_STATE_MAX)
 756                         break;
 757         }
 758
 759         if (!count)
 760                 return -EINVAL;
 761
 762         return 0;
 763 }
 764
 765 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 766 {
 767         int i, count;
 768         struct acpi_processor_cx *cx;
 769         struct cpuidle_state *state;
 770         struct cpuidle_driver *drv = &acpi_idle_driver;
 771
 772         if (max_cstate == 0)
 773                 max_cstate = 1;
 774
 775         if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
 776                 cpuidle_poll_state_init(drv);
 777                 count = 1;
 778         } else {
 779                 count = 0;
 780         }
 781
 782         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 783                 cx = &pr->power.states[i];
 784
 785                 if (!cx->valid)
 786                         continue;
 787
 788                 state = &drv->states[count];
 789                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
 790                 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
 791                 state->exit_latency = cx->latency;
 792                 state->target_residency = cx->latency * latency_factor;
 793                 state->enter = acpi_idle_enter;
 794
 795                 state->flags = 0;
 796                 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
 797                     cx->type == ACPI_STATE_C3) {
 798                         state->enter_dead = acpi_idle_play_dead;
 799                         drv->safe_state_index = count;
 800                 }
 801                 /*
 802                  * Halt-induced C1 is not good for ->enter_s2idle, because it
 803                  * re-enables interrupts on exit.  Moreover, C1 is generally not
 804                  * particularly interesting from the suspend-to-idle angle, so
 805                  * avoid C1 and the situations in which we may need to fall back
 806                  * to it altogether.
 807                  */
 808                 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
 809                         state->enter_s2idle = acpi_idle_enter_s2idle;
 810
 811                 count++;
 812                 if (count == CPUIDLE_STATE_MAX)
 813                         break;
 814         }
 815
 816         drv->state_count = count;
 817
 818         if (!count)
 819                 return -EINVAL;
 820
 821         return 0;
 822 }
 823
 824 static inline void acpi_processor_cstate_first_run_checks(void)
 825 {
 826         static int first_run;
 827
 828         if (first_run)
 829                 return;
 830         dmi_check_system(processor_power_dmi_table);
 831         max_cstate = acpi_processor_cstate_check(max_cstate);
 832         if (max_cstate < ACPI_C_STATES_MAX)
 833                 pr_notice("processor limited to max C-state %d\n", max_cstate);
 834
 835         first_run++;
 836
 837         if (nocst)
 838                 return;
 839
 840         acpi_processor_claim_cst_control();
 841 }
 842 #else
 843
 844 static inline int disabled_by_idle_boot_param(void) { return 0; }
 845 static inline void acpi_processor_cstate_first_run_checks(void) { }
 846 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 847 {
 848         return -ENODEV;
 849 }
 850
 851 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 852                                            struct cpuidle_device *dev)
 853 {
 854         return -EINVAL;
 855 }
 856
 857 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 858 {
 859         return -EINVAL;
 860 }
 861
 862 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
 863
 864 struct acpi_lpi_states_array {
 865         unsigned int size;
 866         unsigned int composite_states_size;
 867         struct acpi_lpi_state *entries;
 868         struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
 869 };
 870
 871 static int obj_get_integer(union acpi_object *obj, u32 *value)
 872 {
 873         if (obj->type != ACPI_TYPE_INTEGER)
 874                 return -EINVAL;
 875
 876         *value = obj->integer.value;
 877         return 0;
 878 }
 879
 880 static int acpi_processor_evaluate_lpi(acpi_handle handle,
 881                                        struct acpi_lpi_states_array *info)
 882 {
 883         acpi_status status;
 884         int ret = 0;
 885         int pkg_count, state_idx = 1, loop;
 886         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 887         union acpi_object *lpi_data;
 888         struct acpi_lpi_state *lpi_state;
 889
 890         status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
 891         if (ACPI_FAILURE(status)) {
 892                 acpi_handle_debug(handle, "No _LPI, giving up\n");
 893                 return -ENODEV;
 894         }
 895
 896         lpi_data = buffer.pointer;
 897
 898         /* There must be at least 4 elements = 3 elements + 1 package */
 899         if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
 900             lpi_data->package.count < 4) {
 901                 pr_debug("not enough elements in _LPI\n");
 902                 ret = -ENODATA;
 903                 goto end;
 904         }
 905
 906         pkg_count = lpi_data->package.elements[2].integer.value;
 907
 908         /* Validate number of power states. */
 909         if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
 910                 pr_debug("count given by _LPI is not valid\n");
 911                 ret = -ENODATA;
 912                 goto end;
 913         }
 914
 915         lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
 916         if (!lpi_state) {
 917                 ret = -ENOMEM;
 918                 goto end;
 919         }
 920
 921         info->size = pkg_count;
 922         info->entries = lpi_state;
 923
 924         /* LPI States start at index 3 */
 925         for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
 926                 union acpi_object *element, *pkg_elem, *obj;
 927
 928                 element = &lpi_data->package.elements[loop];
 929                 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
 930                         continue;
 931
 932                 pkg_elem = element->package.elements;
 933
 934                 obj = pkg_elem + 6;
 935                 if (obj->type == ACPI_TYPE_BUFFER) {
 936                         struct acpi_power_register *reg;
 937
 938                         reg = (struct acpi_power_register *)obj->buffer.pointer;
 939                         if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 940                             reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
 941                                 continue;
 942
 943                         lpi_state->address = reg->address;
 944                         lpi_state->entry_method =
 945                                 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
 946                                 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
 947                 } else if (obj->type == ACPI_TYPE_INTEGER) {
 948                         lpi_state->entry_method = ACPI_CSTATE_INTEGER;
 949                         lpi_state->address = obj->integer.value;
 950                 } else {
 951                         continue;
 952                 }
 953
 954                 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
 955
 956                 obj = pkg_elem + 9;
 957                 if (obj->type == ACPI_TYPE_STRING)
 958                         strlcpy(lpi_state->desc, obj->string.pointer,
 959                                 ACPI_CX_DESC_LEN);
 960
 961                 lpi_state->index = state_idx;
 962                 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
 963                         pr_debug("No min. residency found, assuming 10 us\n");
 964                         lpi_state->min_residency = 10;
 965                 }
 966
 967                 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
 968                         pr_debug("No wakeup residency found, assuming 10 us\n");
 969                         lpi_state->wake_latency = 10;
 970                 }
 971
 972                 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
 973                         lpi_state->flags = 0;
 974
 975                 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
 976                         lpi_state->arch_flags = 0;
 977
 978                 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
 979                         lpi_state->res_cnt_freq = 1;
 980
 981                 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
 982                         lpi_state->enable_parent_state = 0;
 983         }
 984
 985         acpi_handle_debug(handle, "Found %d power states\n", state_idx);
 986 end:
 987         kfree(buffer.pointer);
 988         return ret;
 989 }
 990
 991 /*
 992  * flat_state_cnt - the number of composite LPI states after the process of flattening
 993  */
 994 static int flat_state_cnt;
 995
 996 /**
 997  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
 998  *
 999  * @local: local LPI state
1000  * @parent: parent LPI state
1001  * @result: composite LPI state
1002  */
1003 static bool combine_lpi_states(struct acpi_lpi_state *local,
1004                                struct acpi_lpi_state *parent,
1005                                struct acpi_lpi_state *result)
1006 {
1007         if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1008                 if (!parent->address) /* 0 means autopromotable */
1009                         return false;
1010                 result->address = local->address + parent->address;
1011         } else {
1012                 result->address = parent->address;
1013         }
1014
1015         result->min_residency = max(local->min_residency, parent->min_residency);
1016         result->wake_latency = local->wake_latency + parent->wake_latency;
1017         result->enable_parent_state = parent->enable_parent_state;
1018         result->entry_method = local->entry_method;
1019
1020         result->flags = parent->flags;
1021         result->arch_flags = parent->arch_flags;
1022         result->index = parent->index;
1023
1024         strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1025         strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1026         strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1027         return true;
1028 }
1029
1030 #define ACPI_LPI_STATE_FLAGS_ENABLED                    BIT(0)
1031
1032 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1033                                   struct acpi_lpi_state *t)
1034 {
1035         curr_level->composite_states[curr_level->composite_states_size++] = t;
1036 }
1037
1038 static int flatten_lpi_states(struct acpi_processor *pr,
1039                               struct acpi_lpi_states_array *curr_level,
1040                               struct acpi_lpi_states_array *prev_level)
1041 {
1042         int i, j, state_count = curr_level->size;
1043         struct acpi_lpi_state *p, *t = curr_level->entries;
1044
1045         curr_level->composite_states_size = 0;
1046         for (j = 0; j < state_count; j++, t++) {
1047                 struct acpi_lpi_state *flpi;
1048
1049                 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1050                         continue;
1051
1052                 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1053                         pr_warn("Limiting number of LPI states to max (%d)\n",
1054                                 ACPI_PROCESSOR_MAX_POWER);
1055                         pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1056                         break;
1057                 }
1058
1059                 flpi = &pr->power.lpi_states[flat_state_cnt];
1060
1061                 if (!prev_level) { /* leaf/processor node */
1062                         memcpy(flpi, t, sizeof(*t));
1063                         stash_composite_state(curr_level, flpi);
1064                         flat_state_cnt++;
1065                         continue;
1066                 }
1067
1068                 for (i = 0; i < prev_level->composite_states_size; i++) {
1069                         p = prev_level->composite_states[i];
1070                         if (t->index <= p->enable_parent_state &&
1071                             combine_lpi_states(p, t, flpi)) {
1072                                 stash_composite_state(curr_level, flpi);
1073                                 flat_state_cnt++;
1074                                 flpi++;
1075                         }
1076                 }
1077         }
1078
1079         kfree(curr_level->entries);
1080         return 0;
1081 }
1082
1083 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1084 {
1085         int ret, i;
1086         acpi_status status;
1087         acpi_handle handle = pr->handle, pr_ahandle;
1088         struct acpi_device *d = NULL;
1089         struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1090
1091         if (!osc_pc_lpi_support_confirmed)
1092                 return -EOPNOTSUPP;
1093
1094         if (!acpi_has_method(handle, "_LPI"))
1095                 return -EINVAL;
1096
1097         flat_state_cnt = 0;
1098         prev = &info[0];
1099         curr = &info[1];
1100         handle = pr->handle;
1101         ret = acpi_processor_evaluate_lpi(handle, prev);
1102         if (ret)
1103                 return ret;
1104         flatten_lpi_states(pr, prev, NULL);
1105
1106         status = acpi_get_parent(handle, &pr_ahandle);
1107         while (ACPI_SUCCESS(status)) {
1108                 d = acpi_fetch_acpi_dev(pr_ahandle);
1109                 handle = pr_ahandle;
1110
1111                 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1112                         break;
1113
1114                 /* can be optional ? */
1115                 if (!acpi_has_method(handle, "_LPI"))
1116                         break;
1117
1118                 ret = acpi_processor_evaluate_lpi(handle, curr);
1119                 if (ret)
1120                         break;
1121
1122                 /* flatten all the LPI states in this level of hierarchy */
1123                 flatten_lpi_states(pr, curr, prev);
1124
1125                 tmp = prev, prev = curr, curr = tmp;
1126
1127                 status = acpi_get_parent(handle, &pr_ahandle);
1128         }
1129
1130         pr->power.count = flat_state_cnt;
1131         /* reset the index after flattening */
1132         for (i = 0; i < pr->power.count; i++)
1133                 pr->power.lpi_states[i].index = i;
1134
1135         /* Tell driver that _LPI is supported. */
1136         pr->flags.has_lpi = 1;
1137         pr->flags.power = 1;
1138
1139         return 0;
1140 }
1141
1142 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1143 {
1144         return -ENODEV;
1145 }
1146
1147 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1148 {
1149         return -ENODEV;
1150 }
1151
1152 /**
1153  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1154  * @dev: the target CPU
1155  * @drv: cpuidle driver containing cpuidle state info
1156  * @index: index of target state
1157  *
1158  * Return: 0 for success or negative value for error
1159  */
1160 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1161                                struct cpuidle_driver *drv, int index)
1162 {
1163         struct acpi_processor *pr;
1164         struct acpi_lpi_state *lpi;
1165
1166         pr = __this_cpu_read(processors);
1167
1168         if (unlikely(!pr))
1169                 return -EINVAL;
1170
1171         lpi = &pr->power.lpi_states[index];
1172         if (lpi->entry_method == ACPI_CSTATE_FFH)
1173                 return acpi_processor_ffh_lpi_enter(lpi);
1174
1175         return -EINVAL;
1176 }
1177
1178 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1179 {
1180         int i;
1181         struct acpi_lpi_state *lpi;
1182         struct cpuidle_state *state;
1183         struct cpuidle_driver *drv = &acpi_idle_driver;
1184
1185         if (!pr->flags.has_lpi)
1186                 return -EOPNOTSUPP;
1187
1188         for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1189                 lpi = &pr->power.lpi_states[i];
1190
1191                 state = &drv->states[i];
1192                 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1193                 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1194                 state->exit_latency = lpi->wake_latency;
1195                 state->target_residency = lpi->min_residency;
1196                 if (lpi->arch_flags)
1197                         state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1198                 state->enter = acpi_idle_lpi_enter;
1199                 drv->safe_state_index = i;
1200         }
1201
1202         drv->state_count = i;
1203
1204         return 0;
1205 }
1206
1207 /**
1208  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1209  * global state data i.e. idle routines
1210  *
1211  * @pr: the ACPI processor
1212  */
1213 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1214 {
1215         int i;
1216         struct cpuidle_driver *drv = &acpi_idle_driver;
1217
1218         if (!pr->flags.power_setup_done || !pr->flags.power)
1219                 return -EINVAL;
1220
1221         drv->safe_state_index = -1;
1222         for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1223                 drv->states[i].name[0] = '\0';
1224                 drv->states[i].desc[0] = '\0';
1225         }
1226
1227         if (pr->flags.has_lpi)
1228                 return acpi_processor_setup_lpi_states(pr);
1229
1230         return acpi_processor_setup_cstates(pr);
1231 }
1232
1233 /**
1234  * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1235  * device i.e. per-cpu data
1236  *
1237  * @pr: the ACPI processor
1238  * @dev : the cpuidle device
1239  */
1240 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1241                                             struct cpuidle_device *dev)
1242 {
1243         if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1244                 return -EINVAL;
1245
1246         dev->cpu = pr->id;
1247         if (pr->flags.has_lpi)
1248                 return acpi_processor_ffh_lpi_probe(pr->id);
1249
1250         return acpi_processor_setup_cpuidle_cx(pr, dev);
1251 }
1252
1253 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1254 {
1255         int ret;
1256
1257         ret = acpi_processor_get_lpi_info(pr);
1258         if (ret)
1259                 ret = acpi_processor_get_cstate_info(pr);
1260
1261         return ret;
1262 }
1263
1264 int acpi_processor_hotplug(struct acpi_processor *pr)
1265 {
1266         int ret = 0;
1267         struct cpuidle_device *dev;
1268
1269         if (disabled_by_idle_boot_param())
1270                 return 0;
1271
1272         if (!pr->flags.power_setup_done)
1273                 return -ENODEV;
1274
1275         dev = per_cpu(acpi_cpuidle_device, pr->id);
1276         cpuidle_pause_and_lock();
1277         cpuidle_disable_device(dev);
1278         ret = acpi_processor_get_power_info(pr);
1279         if (!ret && pr->flags.power) {
1280                 acpi_processor_setup_cpuidle_dev(pr, dev);
1281                 ret = cpuidle_enable_device(dev);
1282         }
1283         cpuidle_resume_and_unlock();
1284
1285         return ret;
1286 }
1287
1288 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1289 {
1290         int cpu;
1291         struct acpi_processor *_pr;
1292         struct cpuidle_device *dev;
1293
1294         if (disabled_by_idle_boot_param())
1295                 return 0;
1296
1297         if (!pr->flags.power_setup_done)
1298                 return -ENODEV;
1299
1300         /*
1301          * FIXME:  Design the ACPI notification to make it once per
1302          * system instead of once per-cpu.  This condition is a hack
1303          * to make the code that updates C-States be called once.
1304          */
1305
1306         if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1307
1308                 /* Protect against cpu-hotplug */
1309                 cpus_read_lock();
1310                 cpuidle_pause_and_lock();
1311
1312                 /* Disable all cpuidle devices */
1313                 for_each_online_cpu(cpu) {
1314                         _pr = per_cpu(processors, cpu);
1315                         if (!_pr || !_pr->flags.power_setup_done)
1316                                 continue;
1317                         dev = per_cpu(acpi_cpuidle_device, cpu);
1318                         cpuidle_disable_device(dev);
1319                 }
1320
1321                 /* Populate Updated C-state information */
1322                 acpi_processor_get_power_info(pr);
1323                 acpi_processor_setup_cpuidle_states(pr);
1324
1325                 /* Enable all cpuidle devices */
1326                 for_each_online_cpu(cpu) {
1327                         _pr = per_cpu(processors, cpu);
1328                         if (!_pr || !_pr->flags.power_setup_done)
1329                                 continue;
1330                         acpi_processor_get_power_info(_pr);
1331                         if (_pr->flags.power) {
1332                                 dev = per_cpu(acpi_cpuidle_device, cpu);
1333                                 acpi_processor_setup_cpuidle_dev(_pr, dev);
1334                                 cpuidle_enable_device(dev);
1335                         }
1336                 }
1337                 cpuidle_resume_and_unlock();
1338                 cpus_read_unlock();
1339         }
1340
1341         return 0;
1342 }
1343
1344 static int acpi_processor_registered;
1345
1346 int acpi_processor_power_init(struct acpi_processor *pr)
1347 {
1348         int retval;
1349         struct cpuidle_device *dev;
1350
1351         if (disabled_by_idle_boot_param())
1352                 return 0;
1353
1354         acpi_processor_cstate_first_run_checks();
1355
1356         if (!acpi_processor_get_power_info(pr))
1357                 pr->flags.power_setup_done = 1;
1358
1359         /*
1360          * Install the idle handler if processor power management is supported.
1361          * Note that we use previously set idle handler will be used on
1362          * platforms that only support C1.
1363          */
1364         if (pr->flags.power) {
1365                 /* Register acpi_idle_driver if not already registered */
1366                 if (!acpi_processor_registered) {
1367                         acpi_processor_setup_cpuidle_states(pr);
1368                         retval = cpuidle_register_driver(&acpi_idle_driver);
1369                         if (retval)
1370                                 return retval;
1371                         pr_debug("%s registered with cpuidle\n",
1372                                  acpi_idle_driver.name);
1373                 }
1374
1375                 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1376                 if (!dev)
1377                         return -ENOMEM;
1378                 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1379
1380                 acpi_processor_setup_cpuidle_dev(pr, dev);
1381
1382                 /* Register per-cpu cpuidle_device. Cpuidle driver
1383                  * must already be registered before registering device
1384                  */
1385                 retval = cpuidle_register_device(dev);
1386                 if (retval) {
1387                         if (acpi_processor_registered == 0)
1388                                 cpuidle_unregister_driver(&acpi_idle_driver);
1389                         return retval;
1390                 }
1391                 acpi_processor_registered++;
1392         }
1393         return 0;
1394 }
1395
1396 int acpi_processor_power_exit(struct acpi_processor *pr)
1397 {
1398         struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1399
1400         if (disabled_by_idle_boot_param())
1401                 return 0;
1402
1403         if (pr->flags.power) {
1404                 cpuidle_unregister_device(dev);
1405                 acpi_processor_registered--;
1406                 if (acpi_processor_registered == 0)
1407                         cpuidle_unregister_driver(&acpi_idle_driver);
1408         }
1409
1410         pr->flags.power_setup_done = 0;
1411         return 0;
1412 }