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 <linux/perf_event.h>
  25 #include <acpi/processor.h>
  26 #include <linux/context_tracking.h>
  27
  28 /*
  29  * Include the apic definitions for x86 to have the APIC timer related defines
  30  * available also for UP (on SMP it gets magically included via linux/smp.h).
  31  * asm/acpi.h is not an option, as it would require more include magic. Also
  32  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  33  */
  34 #ifdef CONFIG_X86
  35 #include <asm/apic.h>
  36 #include <asm/cpu.h>
  37 #endif
  38
  39 #define ACPI_IDLE_STATE_START   (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
  40
  41 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  42 module_param(max_cstate, uint, 0400);
  43 static bool nocst __read_mostly;
  44 module_param(nocst, bool, 0400);
  45 static bool bm_check_disable __read_mostly;
  46 module_param(bm_check_disable, bool, 0400);
  47
  48 static unsigned int latency_factor __read_mostly = 2;
  49 module_param(latency_factor, uint, 0644);
  50
  51 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
  52
  53 struct cpuidle_driver acpi_idle_driver = {
  54         .name =         "acpi_idle",
  55         .owner =        THIS_MODULE,
  56 };
  57
  58 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
  59 static
  60 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
  61
  62 static int disabled_by_idle_boot_param(void)
  63 {
  64         return boot_option_idle_override == IDLE_POLL ||
  65                 boot_option_idle_override == IDLE_HALT;
  66 }
  67
  68 /*
  69  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  70  * For now disable this. Probably a bug somewhere else.
  71  *
  72  * To skip this limit, boot/load with a large max_cstate limit.
  73  */
  74 static int set_max_cstate(const struct dmi_system_id *id)
  75 {
  76         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  77                 return 0;
  78
  79         pr_notice("%s detected - limiting to C%ld max_cstate."
  80                   " Override with \"processor.max_cstate=%d\"\n", id->ident,
  81                   (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  82
  83         max_cstate = (long)id->driver_data;
  84
  85         return 0;
  86 }
  87
  88 static const struct dmi_system_id processor_power_dmi_table[] = {
  89         { set_max_cstate, "Clevo 5600D", {
  90           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
  91           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
  92          (void *)2},
  93         { set_max_cstate, "Pavilion zv5000", {
  94           DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
  95           DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
  96          (void *)1},
  97         { set_max_cstate, "Asus L8400B", {
  98           DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
  99           DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
 100          (void *)1},
 101         {},
 102 };
 103
 104
 105 /*
 106  * Callers should disable interrupts before the call and enable
 107  * interrupts after return.
 108  */
 109 static void __cpuidle acpi_safe_halt(void)
 110 {
 111         if (!tif_need_resched()) {
 112                 safe_halt();
 113                 local_irq_disable();
 114         }
 115 }
 116
 117 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 118
 119 /*
 120  * Some BIOS implementations switch to C3 in the published C2 state.
 121  * This seems to be a common problem on AMD boxen, but other vendors
 122  * are affected too. We pick the most conservative approach: we assume
 123  * that the local APIC stops in both C2 and C3.
 124  */
 125 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 126                                    struct acpi_processor_cx *cx)
 127 {
 128         struct acpi_processor_power *pwr = &pr->power;
 129         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 130
 131         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 132                 return;
 133
 134         if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
 135                 type = ACPI_STATE_C1;
 136
 137         /*
 138          * Check, if one of the previous states already marked the lapic
 139          * unstable
 140          */
 141         if (pwr->timer_broadcast_on_state < state)
 142                 return;
 143
 144         if (cx->type >= type)
 145                 pr->power.timer_broadcast_on_state = state;
 146 }
 147
 148 static void __lapic_timer_propagate_broadcast(void *arg)
 149 {
 150         struct acpi_processor *pr = (struct acpi_processor *) arg;
 151
 152         if (pr->power.timer_broadcast_on_state < INT_MAX)
 153                 tick_broadcast_enable();
 154         else
 155                 tick_broadcast_disable();
 156 }
 157
 158 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 159 {
 160         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
 161                                  (void *)pr, 1);
 162 }
 163
 164 /* Power(C) State timer broadcast control */
 165 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 166                                         struct acpi_processor_cx *cx)
 167 {
 168         return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
 169 }
 170
 171 #else
 172
 173 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 174                                    struct acpi_processor_cx *cstate) { }
 175 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 176
 177 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 178                                         struct acpi_processor_cx *cx)
 179 {
 180         return false;
 181 }
 182
 183 #endif
 184
 185 #if defined(CONFIG_X86)
 186 static void tsc_check_state(int state)
 187 {
 188         switch (boot_cpu_data.x86_vendor) {
 189         case X86_VENDOR_HYGON:
 190         case X86_VENDOR_AMD:
 191         case X86_VENDOR_INTEL:
 192         case X86_VENDOR_CENTAUR:
 193         case X86_VENDOR_ZHAOXIN:
 194                 /*
 195                  * AMD Fam10h TSC will tick in all
 196                  * C/P/S0/S1 states when this bit is set.
 197                  */
 198                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 199                         return;
 200                 fallthrough;
 201         default:
 202                 /* TSC could halt in idle, so notify users */
 203                 if (state > ACPI_STATE_C1)
 204                         mark_tsc_unstable("TSC halts in idle");
 205         }
 206 }
 207 #else
 208 static void tsc_check_state(int state) { return; }
 209 #endif
 210
 211 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 212 {
 213
 214         if (!pr->pblk)
 215                 return -ENODEV;
 216
 217         /* if info is obtained from pblk/fadt, type equals state */
 218         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 219         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 220
 221 #ifndef CONFIG_HOTPLUG_CPU
 222         /*
 223          * Check for P_LVL2_UP flag before entering C2 and above on
 224          * an SMP system.
 225          */
 226         if ((num_online_cpus() > 1) &&
 227             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 228                 return -ENODEV;
 229 #endif
 230
 231         /* determine C2 and C3 address from pblk */
 232         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 233         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 234
 235         /* determine latencies from FADT */
 236         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
 237         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
 238
 239         /*
 240          * FADT specified C2 latency must be less than or equal to
 241          * 100 microseconds.
 242          */
 243         if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 244                 acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n",
 245                                   acpi_gbl_FADT.c2_latency);
 246                 /* invalidate C2 */
 247                 pr->power.states[ACPI_STATE_C2].address = 0;
 248         }
 249
 250         /*
 251          * FADT supplied C3 latency must be less than or equal to
 252          * 1000 microseconds.
 253          */
 254         if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 255                 acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n",
 256                                   acpi_gbl_FADT.c3_latency);
 257                 /* invalidate C3 */
 258                 pr->power.states[ACPI_STATE_C3].address = 0;
 259         }
 260
 261         acpi_handle_debug(pr->handle, "lvl2[0x%08x] lvl3[0x%08x]\n",
 262                           pr->power.states[ACPI_STATE_C2].address,
 263                           pr->power.states[ACPI_STATE_C3].address);
 264
 265         snprintf(pr->power.states[ACPI_STATE_C2].desc,
 266                          ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
 267                          pr->power.states[ACPI_STATE_C2].address);
 268         snprintf(pr->power.states[ACPI_STATE_C3].desc,
 269                          ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
 270                          pr->power.states[ACPI_STATE_C3].address);
 271
 272         return 0;
 273 }
 274
 275 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 276 {
 277         if (!pr->power.states[ACPI_STATE_C1].valid) {
 278                 /* set the first C-State to C1 */
 279                 /* all processors need to support C1 */
 280                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 281                 pr->power.states[ACPI_STATE_C1].valid = 1;
 282                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 283
 284                 snprintf(pr->power.states[ACPI_STATE_C1].desc,
 285                          ACPI_CX_DESC_LEN, "ACPI HLT");
 286         }
 287         /* the C0 state only exists as a filler in our array */
 288         pr->power.states[ACPI_STATE_C0].valid = 1;
 289         return 0;
 290 }
 291
 292 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 293 {
 294         int ret;
 295
 296         if (nocst)
 297                 return -ENODEV;
 298
 299         ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
 300         if (ret)
 301                 return ret;
 302
 303         if (!pr->power.count)
 304                 return -EFAULT;
 305
 306         pr->flags.has_cst = 1;
 307         return 0;
 308 }
 309
 310 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 311                                            struct acpi_processor_cx *cx)
 312 {
 313         static int bm_check_flag = -1;
 314         static int bm_control_flag = -1;
 315
 316
 317         if (!cx->address)
 318                 return;
 319
 320         /*
 321          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 322          * DMA transfers are used by any ISA device to avoid livelock.
 323          * Note that we could disable Type-F DMA (as recommended by
 324          * the erratum), but this is known to disrupt certain ISA
 325          * devices thus we take the conservative approach.
 326          */
 327         else if (errata.piix4.fdma) {
 328                 acpi_handle_debug(pr->handle,
 329                                   "C3 not supported on PIIX4 with Type-F DMA\n");
 330                 return;
 331         }
 332
 333         /* All the logic here assumes flags.bm_check is same across all CPUs */
 334         if (bm_check_flag == -1) {
 335                 /* Determine whether bm_check is needed based on CPU  */
 336                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 337                 bm_check_flag = pr->flags.bm_check;
 338                 bm_control_flag = pr->flags.bm_control;
 339         } else {
 340                 pr->flags.bm_check = bm_check_flag;
 341                 pr->flags.bm_control = bm_control_flag;
 342         }
 343
 344         if (pr->flags.bm_check) {
 345                 if (!pr->flags.bm_control) {
 346                         if (pr->flags.has_cst != 1) {
 347                                 /* bus mastering control is necessary */
 348                                 acpi_handle_debug(pr->handle,
 349                                                   "C3 support requires BM control\n");
 350                                 return;
 351                         } else {
 352                                 /* Here we enter C3 without bus mastering */
 353                                 acpi_handle_debug(pr->handle,
 354                                                   "C3 support without BM control\n");
 355                         }
 356                 }
 357         } else {
 358                 /*
 359                  * WBINVD should be set in fadt, for C3 state to be
 360                  * supported on when bm_check is not required.
 361                  */
 362                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 363                         acpi_handle_debug(pr->handle,
 364                                           "Cache invalidation should work properly"
 365                                           " for C3 to be enabled on SMP systems\n");
 366                         return;
 367                 }
 368         }
 369
 370         /*
 371          * Otherwise we've met all of our C3 requirements.
 372          * Normalize the C3 latency to expidite policy.  Enable
 373          * checking of bus mastering status (bm_check) so we can
 374          * use this in our C3 policy
 375          */
 376         cx->valid = 1;
 377
 378         /*
 379          * On older chipsets, BM_RLD needs to be set
 380          * in order for Bus Master activity to wake the
 381          * system from C3.  Newer chipsets handle DMA
 382          * during C3 automatically and BM_RLD is a NOP.
 383          * In either case, the proper way to
 384          * handle BM_RLD is to set it and leave it set.
 385          */
 386         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 387
 388         return;
 389 }
 390
 391 static int acpi_cst_latency_cmp(const void *a, const void *b)
 392 {
 393         const struct acpi_processor_cx *x = a, *y = b;
 394
 395         if (!(x->valid && y->valid))
 396                 return 0;
 397         if (x->latency > y->latency)
 398                 return 1;
 399         if (x->latency < y->latency)
 400                 return -1;
 401         return 0;
 402 }
 403 static void acpi_cst_latency_swap(void *a, void *b, int n)
 404 {
 405         struct acpi_processor_cx *x = a, *y = b;
 406
 407         if (!(x->valid && y->valid))
 408                 return;
 409         swap(x->latency, y->latency);
 410 }
 411
 412 static int acpi_processor_power_verify(struct acpi_processor *pr)
 413 {
 414         unsigned int i;
 415         unsigned int working = 0;
 416         unsigned int last_latency = 0;
 417         unsigned int last_type = 0;
 418         bool buggy_latency = false;
 419
 420         pr->power.timer_broadcast_on_state = INT_MAX;
 421
 422         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 423                 struct acpi_processor_cx *cx = &pr->power.states[i];
 424
 425                 switch (cx->type) {
 426                 case ACPI_STATE_C1:
 427                         cx->valid = 1;
 428                         break;
 429
 430                 case ACPI_STATE_C2:
 431                         if (!cx->address)
 432                                 break;
 433                         cx->valid = 1;
 434                         break;
 435
 436                 case ACPI_STATE_C3:
 437                         acpi_processor_power_verify_c3(pr, cx);
 438                         break;
 439                 }
 440                 if (!cx->valid)
 441                         continue;
 442                 if (cx->type >= last_type && cx->latency < last_latency)
 443                         buggy_latency = true;
 444                 last_latency = cx->latency;
 445                 last_type = cx->type;
 446
 447                 lapic_timer_check_state(i, pr, cx);
 448                 tsc_check_state(cx->type);
 449                 working++;
 450         }
 451
 452         if (buggy_latency) {
 453                 pr_notice("FW issue: working around C-state latencies out of order\n");
 454                 sort(&pr->power.states[1], max_cstate,
 455                      sizeof(struct acpi_processor_cx),
 456                      acpi_cst_latency_cmp,
 457                      acpi_cst_latency_swap);
 458         }
 459
 460         lapic_timer_propagate_broadcast(pr);
 461
 462         return (working);
 463 }
 464
 465 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 466 {
 467         unsigned int i;
 468         int result;
 469
 470
 471         /* NOTE: the idle thread may not be running while calling
 472          * this function */
 473
 474         /* Zero initialize all the C-states info. */
 475         memset(pr->power.states, 0, sizeof(pr->power.states));
 476
 477         result = acpi_processor_get_power_info_cst(pr);
 478         if (result == -ENODEV)
 479                 result = acpi_processor_get_power_info_fadt(pr);
 480
 481         if (result)
 482                 return result;
 483
 484         acpi_processor_get_power_info_default(pr);
 485
 486         pr->power.count = acpi_processor_power_verify(pr);
 487
 488         /*
 489          * if one state of type C2 or C3 is available, mark this
 490          * CPU as being "idle manageable"
 491          */
 492         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 493                 if (pr->power.states[i].valid) {
 494                         pr->power.count = i;
 495                         pr->flags.power = 1;
 496                 }
 497         }
 498
 499         return 0;
 500 }
 501
 502 /**
 503  * acpi_idle_bm_check - checks if bus master activity was detected
 504  */
 505 static int acpi_idle_bm_check(void)
 506 {
 507         u32 bm_status = 0;
 508
 509         if (bm_check_disable)
 510                 return 0;
 511
 512         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 513         if (bm_status)
 514                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 515         /*
 516          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 517          * the true state of bus mastering activity; forcing us to
 518          * manually check the BMIDEA bit of each IDE channel.
 519          */
 520         else if (errata.piix4.bmisx) {
 521                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 522                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 523                         bm_status = 1;
 524         }
 525         return bm_status;
 526 }
 527
 528 static void wait_for_freeze(void)
 529 {
 530 #ifdef  CONFIG_X86
 531         /* No delay is needed if we are in guest */
 532         if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 533                 return;
 534         /*
 535          * Modern (>=Nehalem) Intel systems use ACPI via intel_idle,
 536          * not this code.  Assume that any Intel systems using this
 537          * are ancient and may need the dummy wait.  This also assumes
 538          * that the motivating chipset issue was Intel-only.
 539          */
 540         if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
 541                 return;
 542 #endif
 543         /*
 544          * Dummy wait op - must do something useless after P_LVL2 read
 545          * because chipsets cannot guarantee that STPCLK# signal gets
 546          * asserted in time to freeze execution properly
 547          *
 548          * This workaround has been in place since the original ACPI
 549          * implementation was merged, circa 2002.
 550          *
 551          * If a profile is pointing to this instruction, please first
 552          * consider moving your system to a more modern idle
 553          * mechanism.
 554          */
 555         inl(acpi_gbl_FADT.xpm_timer_block.address);
 556 }
 557
 558 /**
 559  * acpi_idle_do_entry - enter idle state using the appropriate method
 560  * @cx: cstate data
 561  *
 562  * Caller disables interrupt before call and enables interrupt after return.
 563  */
 564 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
 565 {
 566         perf_lopwr_cb(true);
 567
 568         if (cx->entry_method == ACPI_CSTATE_FFH) {
 569                 /* Call into architectural FFH based C-state */
 570                 acpi_processor_ffh_cstate_enter(cx);
 571         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 572                 acpi_safe_halt();
 573         } else {
 574                 /* IO port based C-state */
 575                 inb(cx->address);
 576                 wait_for_freeze();
 577         }
 578
 579         perf_lopwr_cb(false);
 580 }
 581
 582 /**
 583  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
 584  * @dev: the target CPU
 585  * @index: the index of suggested state
 586  */
 587 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 588 {
 589         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 590
 591         ACPI_FLUSH_CPU_CACHE();
 592
 593         while (1) {
 594
 595                 if (cx->entry_method == ACPI_CSTATE_HALT)
 596                         safe_halt();
 597                 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 598                         inb(cx->address);
 599                         wait_for_freeze();
 600                 } else
 601                         return -ENODEV;
 602
 603 #if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU)
 604                 cond_wakeup_cpu0();
 605 #endif
 606         }
 607
 608         /* Never reached */
 609         return 0;
 610 }
 611
 612 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
 613 {
 614         return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
 615                 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
 616 }
 617
 618 static int c3_cpu_count;
 619 static DEFINE_RAW_SPINLOCK(c3_lock);
 620
 621 /**
 622  * acpi_idle_enter_bm - enters C3 with proper BM handling
 623  * @drv: cpuidle driver
 624  * @pr: Target processor
 625  * @cx: Target state context
 626  * @index: index of target state
 627  */
 628 static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
 629                                struct acpi_processor *pr,
 630                                struct acpi_processor_cx *cx,
 631                                int index)
 632 {
 633         static struct acpi_processor_cx safe_cx = {
 634                 .entry_method = ACPI_CSTATE_HALT,
 635         };
 636
 637         /*
 638          * disable bus master
 639          * bm_check implies we need ARB_DIS
 640          * bm_control implies whether we can do ARB_DIS
 641          *
 642          * That leaves a case where bm_check is set and bm_control is not set.
 643          * In that case we cannot do much, we enter C3 without doing anything.
 644          */
 645         bool dis_bm = pr->flags.bm_control;
 646
 647         /* If we can skip BM, demote to a safe state. */
 648         if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
 649                 dis_bm = false;
 650                 index = drv->safe_state_index;
 651                 if (index >= 0) {
 652                         cx = this_cpu_read(acpi_cstate[index]);
 653                 } else {
 654                         cx = &safe_cx;
 655                         index = -EBUSY;
 656                 }
 657         }
 658
 659         if (dis_bm) {
 660                 raw_spin_lock(&c3_lock);
 661                 c3_cpu_count++;
 662                 /* Disable bus master arbitration when all CPUs are in C3 */
 663                 if (c3_cpu_count == num_online_cpus())
 664                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 665                 raw_spin_unlock(&c3_lock);
 666         }
 667
 668         ct_idle_enter();
 669
 670         acpi_idle_do_entry(cx);
 671
 672         ct_idle_exit();
 673
 674         /* Re-enable bus master arbitration */
 675         if (dis_bm) {
 676                 raw_spin_lock(&c3_lock);
 677                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 678                 c3_cpu_count--;
 679                 raw_spin_unlock(&c3_lock);
 680         }
 681
 682         return index;
 683 }
 684
 685 static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
 686                            struct cpuidle_driver *drv, int index)
 687 {
 688         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 689         struct acpi_processor *pr;
 690
 691         pr = __this_cpu_read(processors);
 692         if (unlikely(!pr))
 693                 return -EINVAL;
 694
 695         if (cx->type != ACPI_STATE_C1) {
 696                 if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
 697                         return acpi_idle_enter_bm(drv, pr, cx, index);
 698
 699                 /* C2 to C1 demotion. */
 700                 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
 701                         index = ACPI_IDLE_STATE_START;
 702                         cx = per_cpu(acpi_cstate[index], dev->cpu);
 703                 }
 704         }
 705
 706         if (cx->type == ACPI_STATE_C3)
 707                 ACPI_FLUSH_CPU_CACHE();
 708
 709         acpi_idle_do_entry(cx);
 710
 711         return index;
 712 }
 713
 714 static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
 715                                   struct cpuidle_driver *drv, int index)
 716 {
 717         struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 718
 719         if (cx->type == ACPI_STATE_C3) {
 720                 struct acpi_processor *pr = __this_cpu_read(processors);
 721
 722                 if (unlikely(!pr))
 723                         return 0;
 724
 725                 if (pr->flags.bm_check) {
 726                         u8 bm_sts_skip = cx->bm_sts_skip;
 727
 728                         /* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
 729                         cx->bm_sts_skip = 1;
 730                         acpi_idle_enter_bm(drv, pr, cx, index);
 731                         cx->bm_sts_skip = bm_sts_skip;
 732
 733                         return 0;
 734                 } else {
 735                         ACPI_FLUSH_CPU_CACHE();
 736                 }
 737         }
 738         acpi_idle_do_entry(cx);
 739
 740         return 0;
 741 }
 742
 743 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 744                                            struct cpuidle_device *dev)
 745 {
 746         int i, count = ACPI_IDLE_STATE_START;
 747         struct acpi_processor_cx *cx;
 748         struct cpuidle_state *state;
 749
 750         if (max_cstate == 0)
 751                 max_cstate = 1;
 752
 753         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 754                 state = &acpi_idle_driver.states[count];
 755                 cx = &pr->power.states[i];
 756
 757                 if (!cx->valid)
 758                         continue;
 759
 760                 per_cpu(acpi_cstate[count], dev->cpu) = cx;
 761
 762                 if (lapic_timer_needs_broadcast(pr, cx))
 763                         state->flags |= CPUIDLE_FLAG_TIMER_STOP;
 764
 765                 if (cx->type == ACPI_STATE_C3) {
 766                         state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
 767                         if (pr->flags.bm_check)
 768                                 state->flags |= CPUIDLE_FLAG_RCU_IDLE;
 769                 }
 770
 771                 count++;
 772                 if (count == CPUIDLE_STATE_MAX)
 773                         break;
 774         }
 775
 776         if (!count)
 777                 return -EINVAL;
 778
 779         return 0;
 780 }
 781
 782 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 783 {
 784         int i, count;
 785         struct acpi_processor_cx *cx;
 786         struct cpuidle_state *state;
 787         struct cpuidle_driver *drv = &acpi_idle_driver;
 788
 789         if (max_cstate == 0)
 790                 max_cstate = 1;
 791
 792         if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
 793                 cpuidle_poll_state_init(drv);
 794                 count = 1;
 795         } else {
 796                 count = 0;
 797         }
 798
 799         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 800                 cx = &pr->power.states[i];
 801
 802                 if (!cx->valid)
 803                         continue;
 804
 805                 state = &drv->states[count];
 806                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
 807                 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
 808                 state->exit_latency = cx->latency;
 809                 state->target_residency = cx->latency * latency_factor;
 810                 state->enter = acpi_idle_enter;
 811
 812                 state->flags = 0;
 813                 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
 814                     cx->type == ACPI_STATE_C3) {
 815                         state->enter_dead = acpi_idle_play_dead;
 816                         if (cx->type != ACPI_STATE_C3)
 817                                 drv->safe_state_index = count;
 818                 }
 819                 /*
 820                  * Halt-induced C1 is not good for ->enter_s2idle, because it
 821                  * re-enables interrupts on exit.  Moreover, C1 is generally not
 822                  * particularly interesting from the suspend-to-idle angle, so
 823                  * avoid C1 and the situations in which we may need to fall back
 824                  * to it altogether.
 825                  */
 826                 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
 827                         state->enter_s2idle = acpi_idle_enter_s2idle;
 828
 829                 count++;
 830                 if (count == CPUIDLE_STATE_MAX)
 831                         break;
 832         }
 833
 834         drv->state_count = count;
 835
 836         if (!count)
 837                 return -EINVAL;
 838
 839         return 0;
 840 }
 841
 842 static inline void acpi_processor_cstate_first_run_checks(void)
 843 {
 844         static int first_run;
 845
 846         if (first_run)
 847                 return;
 848         dmi_check_system(processor_power_dmi_table);
 849         max_cstate = acpi_processor_cstate_check(max_cstate);
 850         if (max_cstate < ACPI_C_STATES_MAX)
 851                 pr_notice("processor limited to max C-state %d\n", max_cstate);
 852
 853         first_run++;
 854
 855         if (nocst)
 856                 return;
 857
 858         acpi_processor_claim_cst_control();
 859 }
 860 #else
 861
 862 static inline int disabled_by_idle_boot_param(void) { return 0; }
 863 static inline void acpi_processor_cstate_first_run_checks(void) { }
 864 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 865 {
 866         return -ENODEV;
 867 }
 868
 869 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 870                                            struct cpuidle_device *dev)
 871 {
 872         return -EINVAL;
 873 }
 874
 875 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 876 {
 877         return -EINVAL;
 878 }
 879
 880 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
 881
 882 struct acpi_lpi_states_array {
 883         unsigned int size;
 884         unsigned int composite_states_size;
 885         struct acpi_lpi_state *entries;
 886         struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
 887 };
 888
 889 static int obj_get_integer(union acpi_object *obj, u32 *value)
 890 {
 891         if (obj->type != ACPI_TYPE_INTEGER)
 892                 return -EINVAL;
 893
 894         *value = obj->integer.value;
 895         return 0;
 896 }
 897
 898 static int acpi_processor_evaluate_lpi(acpi_handle handle,
 899                                        struct acpi_lpi_states_array *info)
 900 {
 901         acpi_status status;
 902         int ret = 0;
 903         int pkg_count, state_idx = 1, loop;
 904         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 905         union acpi_object *lpi_data;
 906         struct acpi_lpi_state *lpi_state;
 907
 908         status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
 909         if (ACPI_FAILURE(status)) {
 910                 acpi_handle_debug(handle, "No _LPI, giving up\n");
 911                 return -ENODEV;
 912         }
 913
 914         lpi_data = buffer.pointer;
 915
 916         /* There must be at least 4 elements = 3 elements + 1 package */
 917         if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
 918             lpi_data->package.count < 4) {
 919                 pr_debug("not enough elements in _LPI\n");
 920                 ret = -ENODATA;
 921                 goto end;
 922         }
 923
 924         pkg_count = lpi_data->package.elements[2].integer.value;
 925
 926         /* Validate number of power states. */
 927         if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
 928                 pr_debug("count given by _LPI is not valid\n");
 929                 ret = -ENODATA;
 930                 goto end;
 931         }
 932
 933         lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
 934         if (!lpi_state) {
 935                 ret = -ENOMEM;
 936                 goto end;
 937         }
 938
 939         info->size = pkg_count;
 940         info->entries = lpi_state;
 941
 942         /* LPI States start at index 3 */
 943         for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
 944                 union acpi_object *element, *pkg_elem, *obj;
 945
 946                 element = &lpi_data->package.elements[loop];
 947                 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
 948                         continue;
 949
 950                 pkg_elem = element->package.elements;
 951
 952                 obj = pkg_elem + 6;
 953                 if (obj->type == ACPI_TYPE_BUFFER) {
 954                         struct acpi_power_register *reg;
 955
 956                         reg = (struct acpi_power_register *)obj->buffer.pointer;
 957                         if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 958                             reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
 959                                 continue;
 960
 961                         lpi_state->address = reg->address;
 962                         lpi_state->entry_method =
 963                                 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
 964                                 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
 965                 } else if (obj->type == ACPI_TYPE_INTEGER) {
 966                         lpi_state->entry_method = ACPI_CSTATE_INTEGER;
 967                         lpi_state->address = obj->integer.value;
 968                 } else {
 969                         continue;
 970                 }
 971
 972                 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
 973
 974                 obj = pkg_elem + 9;
 975                 if (obj->type == ACPI_TYPE_STRING)
 976                         strscpy(lpi_state->desc, obj->string.pointer,
 977                                 ACPI_CX_DESC_LEN);
 978
 979                 lpi_state->index = state_idx;
 980                 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
 981                         pr_debug("No min. residency found, assuming 10 us\n");
 982                         lpi_state->min_residency = 10;
 983                 }
 984
 985                 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
 986                         pr_debug("No wakeup residency found, assuming 10 us\n");
 987                         lpi_state->wake_latency = 10;
 988                 }
 989
 990                 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
 991                         lpi_state->flags = 0;
 992
 993                 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
 994                         lpi_state->arch_flags = 0;
 995
 996                 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
 997                         lpi_state->res_cnt_freq = 1;
 998
 999                 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
1000                         lpi_state->enable_parent_state = 0;
1001         }
1002
1003         acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1004 end:
1005         kfree(buffer.pointer);
1006         return ret;
1007 }
1008
1009 /*
1010  * flat_state_cnt - the number of composite LPI states after the process of flattening
1011  */
1012 static int flat_state_cnt;
1013
1014 /**
1015  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1016  *
1017  * @local: local LPI state
1018  * @parent: parent LPI state
1019  * @result: composite LPI state
1020  */
1021 static bool combine_lpi_states(struct acpi_lpi_state *local,
1022                                struct acpi_lpi_state *parent,
1023                                struct acpi_lpi_state *result)
1024 {
1025         if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1026                 if (!parent->address) /* 0 means autopromotable */
1027                         return false;
1028                 result->address = local->address + parent->address;
1029         } else {
1030                 result->address = parent->address;
1031         }
1032
1033         result->min_residency = max(local->min_residency, parent->min_residency);
1034         result->wake_latency = local->wake_latency + parent->wake_latency;
1035         result->enable_parent_state = parent->enable_parent_state;
1036         result->entry_method = local->entry_method;
1037
1038         result->flags = parent->flags;
1039         result->arch_flags = parent->arch_flags;
1040         result->index = parent->index;
1041
1042         strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1043         strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1044         strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1045         return true;
1046 }
1047
1048 #define ACPI_LPI_STATE_FLAGS_ENABLED                    BIT(0)
1049
1050 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1051                                   struct acpi_lpi_state *t)
1052 {
1053         curr_level->composite_states[curr_level->composite_states_size++] = t;
1054 }
1055
1056 static int flatten_lpi_states(struct acpi_processor *pr,
1057                               struct acpi_lpi_states_array *curr_level,
1058                               struct acpi_lpi_states_array *prev_level)
1059 {
1060         int i, j, state_count = curr_level->size;
1061         struct acpi_lpi_state *p, *t = curr_level->entries;
1062
1063         curr_level->composite_states_size = 0;
1064         for (j = 0; j < state_count; j++, t++) {
1065                 struct acpi_lpi_state *flpi;
1066
1067                 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1068                         continue;
1069
1070                 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1071                         pr_warn("Limiting number of LPI states to max (%d)\n",
1072                                 ACPI_PROCESSOR_MAX_POWER);
1073                         pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1074                         break;
1075                 }
1076
1077                 flpi = &pr->power.lpi_states[flat_state_cnt];
1078
1079                 if (!prev_level) { /* leaf/processor node */
1080                         memcpy(flpi, t, sizeof(*t));
1081                         stash_composite_state(curr_level, flpi);
1082                         flat_state_cnt++;
1083                         continue;
1084                 }
1085
1086                 for (i = 0; i < prev_level->composite_states_size; i++) {
1087                         p = prev_level->composite_states[i];
1088                         if (t->index <= p->enable_parent_state &&
1089                             combine_lpi_states(p, t, flpi)) {
1090                                 stash_composite_state(curr_level, flpi);
1091                                 flat_state_cnt++;
1092                                 flpi++;
1093                         }
1094                 }
1095         }
1096
1097         kfree(curr_level->entries);
1098         return 0;
1099 }
1100
1101 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1102 {
1103         return -EOPNOTSUPP;
1104 }
1105
1106 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1107 {
1108         int ret, i;
1109         acpi_status status;
1110         acpi_handle handle = pr->handle, pr_ahandle;
1111         struct acpi_device *d = NULL;
1112         struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1113
1114         /* make sure our architecture has support */
1115         ret = acpi_processor_ffh_lpi_probe(pr->id);
1116         if (ret == -EOPNOTSUPP)
1117                 return ret;
1118
1119         if (!osc_pc_lpi_support_confirmed)
1120                 return -EOPNOTSUPP;
1121
1122         if (!acpi_has_method(handle, "_LPI"))
1123                 return -EINVAL;
1124
1125         flat_state_cnt = 0;
1126         prev = &info[0];
1127         curr = &info[1];
1128         handle = pr->handle;
1129         ret = acpi_processor_evaluate_lpi(handle, prev);
1130         if (ret)
1131                 return ret;
1132         flatten_lpi_states(pr, prev, NULL);
1133
1134         status = acpi_get_parent(handle, &pr_ahandle);
1135         while (ACPI_SUCCESS(status)) {
1136                 d = acpi_fetch_acpi_dev(pr_ahandle);
1137                 handle = pr_ahandle;
1138
1139                 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1140                         break;
1141
1142                 /* can be optional ? */
1143                 if (!acpi_has_method(handle, "_LPI"))
1144                         break;
1145
1146                 ret = acpi_processor_evaluate_lpi(handle, curr);
1147                 if (ret)
1148                         break;
1149
1150                 /* flatten all the LPI states in this level of hierarchy */
1151                 flatten_lpi_states(pr, curr, prev);
1152
1153                 tmp = prev, prev = curr, curr = tmp;
1154
1155                 status = acpi_get_parent(handle, &pr_ahandle);
1156         }
1157
1158         pr->power.count = flat_state_cnt;
1159         /* reset the index after flattening */
1160         for (i = 0; i < pr->power.count; i++)
1161                 pr->power.lpi_states[i].index = i;
1162
1163         /* Tell driver that _LPI is supported. */
1164         pr->flags.has_lpi = 1;
1165         pr->flags.power = 1;
1166
1167         return 0;
1168 }
1169
1170 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1171 {
1172         return -ENODEV;
1173 }
1174
1175 /**
1176  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1177  * @dev: the target CPU
1178  * @drv: cpuidle driver containing cpuidle state info
1179  * @index: index of target state
1180  *
1181  * Return: 0 for success or negative value for error
1182  */
1183 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1184                                struct cpuidle_driver *drv, int index)
1185 {
1186         struct acpi_processor *pr;
1187         struct acpi_lpi_state *lpi;
1188
1189         pr = __this_cpu_read(processors);
1190
1191         if (unlikely(!pr))
1192                 return -EINVAL;
1193
1194         lpi = &pr->power.lpi_states[index];
1195         if (lpi->entry_method == ACPI_CSTATE_FFH)
1196                 return acpi_processor_ffh_lpi_enter(lpi);
1197
1198         return -EINVAL;
1199 }
1200
1201 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1202 {
1203         int i;
1204         struct acpi_lpi_state *lpi;
1205         struct cpuidle_state *state;
1206         struct cpuidle_driver *drv = &acpi_idle_driver;
1207
1208         if (!pr->flags.has_lpi)
1209                 return -EOPNOTSUPP;
1210
1211         for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1212                 lpi = &pr->power.lpi_states[i];
1213
1214                 state = &drv->states[i];
1215                 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1216                 strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1217                 state->exit_latency = lpi->wake_latency;
1218                 state->target_residency = lpi->min_residency;
1219                 if (lpi->arch_flags)
1220                         state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1221                 state->enter = acpi_idle_lpi_enter;
1222                 drv->safe_state_index = i;
1223         }
1224
1225         drv->state_count = i;
1226
1227         return 0;
1228 }
1229
1230 /**
1231  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1232  * global state data i.e. idle routines
1233  *
1234  * @pr: the ACPI processor
1235  */
1236 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1237 {
1238         int i;
1239         struct cpuidle_driver *drv = &acpi_idle_driver;
1240
1241         if (!pr->flags.power_setup_done || !pr->flags.power)
1242                 return -EINVAL;
1243
1244         drv->safe_state_index = -1;
1245         for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1246                 drv->states[i].name[0] = '\0';
1247                 drv->states[i].desc[0] = '\0';
1248         }
1249
1250         if (pr->flags.has_lpi)
1251                 return acpi_processor_setup_lpi_states(pr);
1252
1253         return acpi_processor_setup_cstates(pr);
1254 }
1255
1256 /**
1257  * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1258  * device i.e. per-cpu data
1259  *
1260  * @pr: the ACPI processor
1261  * @dev : the cpuidle device
1262  */
1263 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1264                                             struct cpuidle_device *dev)
1265 {
1266         if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1267                 return -EINVAL;
1268
1269         dev->cpu = pr->id;
1270         if (pr->flags.has_lpi)
1271                 return acpi_processor_ffh_lpi_probe(pr->id);
1272
1273         return acpi_processor_setup_cpuidle_cx(pr, dev);
1274 }
1275
1276 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1277 {
1278         int ret;
1279
1280         ret = acpi_processor_get_lpi_info(pr);
1281         if (ret)
1282                 ret = acpi_processor_get_cstate_info(pr);
1283
1284         return ret;
1285 }
1286
1287 int acpi_processor_hotplug(struct acpi_processor *pr)
1288 {
1289         int ret = 0;
1290         struct cpuidle_device *dev;
1291
1292         if (disabled_by_idle_boot_param())
1293                 return 0;
1294
1295         if (!pr->flags.power_setup_done)
1296                 return -ENODEV;
1297
1298         dev = per_cpu(acpi_cpuidle_device, pr->id);
1299         cpuidle_pause_and_lock();
1300         cpuidle_disable_device(dev);
1301         ret = acpi_processor_get_power_info(pr);
1302         if (!ret && pr->flags.power) {
1303                 acpi_processor_setup_cpuidle_dev(pr, dev);
1304                 ret = cpuidle_enable_device(dev);
1305         }
1306         cpuidle_resume_and_unlock();
1307
1308         return ret;
1309 }
1310
1311 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1312 {
1313         int cpu;
1314         struct acpi_processor *_pr;
1315         struct cpuidle_device *dev;
1316
1317         if (disabled_by_idle_boot_param())
1318                 return 0;
1319
1320         if (!pr->flags.power_setup_done)
1321                 return -ENODEV;
1322
1323         /*
1324          * FIXME:  Design the ACPI notification to make it once per
1325          * system instead of once per-cpu.  This condition is a hack
1326          * to make the code that updates C-States be called once.
1327          */
1328
1329         if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1330
1331                 /* Protect against cpu-hotplug */
1332                 cpus_read_lock();
1333                 cpuidle_pause_and_lock();
1334
1335                 /* Disable all cpuidle devices */
1336                 for_each_online_cpu(cpu) {
1337                         _pr = per_cpu(processors, cpu);
1338                         if (!_pr || !_pr->flags.power_setup_done)
1339                                 continue;
1340                         dev = per_cpu(acpi_cpuidle_device, cpu);
1341                         cpuidle_disable_device(dev);
1342                 }
1343
1344                 /* Populate Updated C-state information */
1345                 acpi_processor_get_power_info(pr);
1346                 acpi_processor_setup_cpuidle_states(pr);
1347
1348                 /* Enable all cpuidle devices */
1349                 for_each_online_cpu(cpu) {
1350                         _pr = per_cpu(processors, cpu);
1351                         if (!_pr || !_pr->flags.power_setup_done)
1352                                 continue;
1353                         acpi_processor_get_power_info(_pr);
1354                         if (_pr->flags.power) {
1355                                 dev = per_cpu(acpi_cpuidle_device, cpu);
1356                                 acpi_processor_setup_cpuidle_dev(_pr, dev);
1357                                 cpuidle_enable_device(dev);
1358                         }
1359                 }
1360                 cpuidle_resume_and_unlock();
1361                 cpus_read_unlock();
1362         }
1363
1364         return 0;
1365 }
1366
1367 static int acpi_processor_registered;
1368
1369 int acpi_processor_power_init(struct acpi_processor *pr)
1370 {
1371         int retval;
1372         struct cpuidle_device *dev;
1373
1374         if (disabled_by_idle_boot_param())
1375                 return 0;
1376
1377         acpi_processor_cstate_first_run_checks();
1378
1379         if (!acpi_processor_get_power_info(pr))
1380                 pr->flags.power_setup_done = 1;
1381
1382         /*
1383          * Install the idle handler if processor power management is supported.
1384          * Note that we use previously set idle handler will be used on
1385          * platforms that only support C1.
1386          */
1387         if (pr->flags.power) {
1388                 /* Register acpi_idle_driver if not already registered */
1389                 if (!acpi_processor_registered) {
1390                         acpi_processor_setup_cpuidle_states(pr);
1391                         retval = cpuidle_register_driver(&acpi_idle_driver);
1392                         if (retval)
1393                                 return retval;
1394                         pr_debug("%s registered with cpuidle\n",
1395                                  acpi_idle_driver.name);
1396                 }
1397
1398                 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1399                 if (!dev)
1400                         return -ENOMEM;
1401                 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1402
1403                 acpi_processor_setup_cpuidle_dev(pr, dev);
1404
1405                 /* Register per-cpu cpuidle_device. Cpuidle driver
1406                  * must already be registered before registering device
1407                  */
1408                 retval = cpuidle_register_device(dev);
1409                 if (retval) {
1410                         if (acpi_processor_registered == 0)
1411                                 cpuidle_unregister_driver(&acpi_idle_driver);
1412                         return retval;
1413                 }
1414                 acpi_processor_registered++;
1415         }
1416         return 0;
1417 }
1418
1419 int acpi_processor_power_exit(struct acpi_processor *pr)
1420 {
1421         struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1422
1423         if (disabled_by_idle_boot_param())
1424                 return 0;
1425
1426         if (pr->flags.power) {
1427                 cpuidle_unregister_device(dev);
1428                 acpi_processor_registered--;
1429                 if (acpi_processor_registered == 0)
1430                         cpuidle_unregister_driver(&acpi_idle_driver);
1431         }
1432
1433         pr->flags.power_setup_done = 0;
1434         return 0;
1435 }