drivers/acpi/processor_idle.c

   1 /*
   2  * processor_idle - idle state submodule to the ACPI processor driver
   3  *
   4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   6  *  Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
   7  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
   8  *                      - Added processor hotplug support
   9  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  10  *                      - Added support for C3 on SMP
  11  *
  12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  13  *
  14  *  This program is free software; you can redistribute it and/or modify
  15  *  it under the terms of the GNU General Public License as published by
  16  *  the Free Software Foundation; either version 2 of the License, or (at
  17  *  your option) any later version.
  18  *
  19  *  This program is distributed in the hope that it will be useful, but
  20  *  WITHOUT ANY WARRANTY; without even the implied warranty of
  21  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  *  General Public License for more details.
  23  *
  24  *  You should have received a copy of the GNU General Public License along
  25  *  with this program; if not, write to the Free Software Foundation, Inc.,
  26  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
  27  *
  28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  29  */
  30
  31 #include <linux/module.h>
  32 #include <linux/acpi.h>
  33 #include <linux/dmi.h>
  34 #include <linux/sched.h>       /* need_resched() */
  35 #include <linux/clockchips.h>
  36 #include <linux/cpuidle.h>
  37
  38 /*
  39  * Include the apic definitions for x86 to have the APIC timer related defines
  40  * available also for UP (on SMP it gets magically included via linux/smp.h).
  41  * asm/acpi.h is not an option, as it would require more include magic. Also
  42  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  43  */
  44 #ifdef CONFIG_X86
  45 #include <asm/apic.h>
  46 #endif
  47
  48 #include <acpi/acpi_bus.h>
  49 #include <acpi/processor.h>
  50
  51 #define PREFIX "ACPI: "
  52
  53 #define ACPI_PROCESSOR_CLASS            "processor"
  54 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  55 ACPI_MODULE_NAME("processor_idle");
  56
  57 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  58 module_param(max_cstate, uint, 0000);
  59 static unsigned int nocst __read_mostly;
  60 module_param(nocst, uint, 0000);
  61 static int bm_check_disable __read_mostly;
  62 module_param(bm_check_disable, uint, 0000);
  63
  64 static unsigned int latency_factor __read_mostly = 2;
  65 module_param(latency_factor, uint, 0644);
  66
  67 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
  68
  69 static struct acpi_processor_cx *acpi_cstate[CPUIDLE_STATE_MAX];
  70
  71 static int disabled_by_idle_boot_param(void)
  72 {
  73         return boot_option_idle_override == IDLE_POLL ||
  74                 boot_option_idle_override == IDLE_HALT;
  75 }
  76
  77 /*
  78  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  79  * For now disable this. Probably a bug somewhere else.
  80  *
  81  * To skip this limit, boot/load with a large max_cstate limit.
  82  */
  83 static int set_max_cstate(const struct dmi_system_id *id)
  84 {
  85         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  86                 return 0;
  87
  88         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  89                " Override with \"processor.max_cstate=%d\"\n", id->ident,
  90                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  91
  92         max_cstate = (long)id->driver_data;
  93
  94         return 0;
  95 }
  96
  97 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
  98    callers to only run once -AK */
  99 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 100         { set_max_cstate, "Clevo 5600D", {
 101           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 102           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 103          (void *)2},
 104         { set_max_cstate, "Pavilion zv5000", {
 105           DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
 106           DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
 107          (void *)1},
 108         { set_max_cstate, "Asus L8400B", {
 109           DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
 110           DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
 111          (void *)1},
 112         {},
 113 };
 114
 115
 116 /*
 117  * Callers should disable interrupts before the call and enable
 118  * interrupts after return.
 119  */
 120 static void acpi_safe_halt(void)
 121 {
 122         current_thread_info()->status &= ~TS_POLLING;
 123         /*
 124          * TS_POLLING-cleared state must be visible before we
 125          * test NEED_RESCHED:
 126          */
 127         smp_mb();
 128         if (!need_resched()) {
 129                 safe_halt();
 130                 local_irq_disable();
 131         }
 132         current_thread_info()->status |= TS_POLLING;
 133 }
 134
 135 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 136
 137 /*
 138  * Some BIOS implementations switch to C3 in the published C2 state.
 139  * This seems to be a common problem on AMD boxen, but other vendors
 140  * are affected too. We pick the most conservative approach: we assume
 141  * that the local APIC stops in both C2 and C3.
 142  */
 143 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 144                                    struct acpi_processor_cx *cx)
 145 {
 146         struct acpi_processor_power *pwr = &pr->power;
 147         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 148
 149         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 150                 return;
 151
 152         if (amd_e400_c1e_detected)
 153                 type = ACPI_STATE_C1;
 154
 155         /*
 156          * Check, if one of the previous states already marked the lapic
 157          * unstable
 158          */
 159         if (pwr->timer_broadcast_on_state < state)
 160                 return;
 161
 162         if (cx->type >= type)
 163                 pr->power.timer_broadcast_on_state = state;
 164 }
 165
 166 static void __lapic_timer_propagate_broadcast(void *arg)
 167 {
 168         struct acpi_processor *pr = (struct acpi_processor *) arg;
 169         unsigned long reason;
 170
 171         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 172                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 173
 174         clockevents_notify(reason, &pr->id);
 175 }
 176
 177 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 178 {
 179         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
 180                                  (void *)pr, 1);
 181 }
 182
 183 /* Power(C) State timer broadcast control */
 184 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 185                                        struct acpi_processor_cx *cx,
 186                                        int broadcast)
 187 {
 188         int state = cx - pr->power.states;
 189
 190         if (state >= pr->power.timer_broadcast_on_state) {
 191                 unsigned long reason;
 192
 193                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 194                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 195                 clockevents_notify(reason, &pr->id);
 196         }
 197 }
 198
 199 #else
 200
 201 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 202                                    struct acpi_processor_cx *cstate) { }
 203 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 204 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 205                                        struct acpi_processor_cx *cx,
 206                                        int broadcast)
 207 {
 208 }
 209
 210 #endif
 211
 212 static u32 saved_bm_rld;
 213
 214 static void acpi_idle_bm_rld_save(void)
 215 {
 216         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
 217 }
 218 static void acpi_idle_bm_rld_restore(void)
 219 {
 220         u32 resumed_bm_rld;
 221
 222         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
 223
 224         if (resumed_bm_rld != saved_bm_rld)
 225                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
 226 }
 227
 228 int acpi_processor_suspend(struct device *dev)
 229 {
 230         acpi_idle_bm_rld_save();
 231         return 0;
 232 }
 233
 234 int acpi_processor_resume(struct device *dev)
 235 {
 236         acpi_idle_bm_rld_restore();
 237         return 0;
 238 }
 239
 240 #if defined(CONFIG_X86)
 241 static void tsc_check_state(int state)
 242 {
 243         switch (boot_cpu_data.x86_vendor) {
 244         case X86_VENDOR_AMD:
 245         case X86_VENDOR_INTEL:
 246                 /*
 247                  * AMD Fam10h TSC will tick in all
 248                  * C/P/S0/S1 states when this bit is set.
 249                  */
 250                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 251                         return;
 252
 253                 /*FALL THROUGH*/
 254         default:
 255                 /* TSC could halt in idle, so notify users */
 256                 if (state > ACPI_STATE_C1)
 257                         mark_tsc_unstable("TSC halts in idle");
 258         }
 259 }
 260 #else
 261 static void tsc_check_state(int state) { return; }
 262 #endif
 263
 264 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 265 {
 266
 267         if (!pr)
 268                 return -EINVAL;
 269
 270         if (!pr->pblk)
 271                 return -ENODEV;
 272
 273         /* if info is obtained from pblk/fadt, type equals state */
 274         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 275         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 276
 277 #ifndef CONFIG_HOTPLUG_CPU
 278         /*
 279          * Check for P_LVL2_UP flag before entering C2 and above on
 280          * an SMP system.
 281          */
 282         if ((num_online_cpus() > 1) &&
 283             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 284                 return -ENODEV;
 285 #endif
 286
 287         /* determine C2 and C3 address from pblk */
 288         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 289         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 290
 291         /* determine latencies from FADT */
 292         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
 293         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
 294
 295         /*
 296          * FADT specified C2 latency must be less than or equal to
 297          * 100 microseconds.
 298          */
 299         if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 300                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 301                         "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
 302                 /* invalidate C2 */
 303                 pr->power.states[ACPI_STATE_C2].address = 0;
 304         }
 305
 306         /*
 307          * FADT supplied C3 latency must be less than or equal to
 308          * 1000 microseconds.
 309          */
 310         if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 311                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 312                         "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
 313                 /* invalidate C3 */
 314                 pr->power.states[ACPI_STATE_C3].address = 0;
 315         }
 316
 317         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 318                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 319                           pr->power.states[ACPI_STATE_C2].address,
 320                           pr->power.states[ACPI_STATE_C3].address));
 321
 322         return 0;
 323 }
 324
 325 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 326 {
 327         if (!pr->power.states[ACPI_STATE_C1].valid) {
 328                 /* set the first C-State to C1 */
 329                 /* all processors need to support C1 */
 330                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 331                 pr->power.states[ACPI_STATE_C1].valid = 1;
 332                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 333         }
 334         /* the C0 state only exists as a filler in our array */
 335         pr->power.states[ACPI_STATE_C0].valid = 1;
 336         return 0;
 337 }
 338
 339 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 340 {
 341         acpi_status status = 0;
 342         u64 count;
 343         int current_count;
 344         int i;
 345         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 346         union acpi_object *cst;
 347
 348
 349         if (nocst)
 350                 return -ENODEV;
 351
 352         current_count = 0;
 353
 354         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 355         if (ACPI_FAILURE(status)) {
 356                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 357                 return -ENODEV;
 358         }
 359
 360         cst = buffer.pointer;
 361
 362         /* There must be at least 2 elements */
 363         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 364                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 365                 status = -EFAULT;
 366                 goto end;
 367         }
 368
 369         count = cst->package.elements[0].integer.value;
 370
 371         /* Validate number of power states. */
 372         if (count < 1 || count != cst->package.count - 1) {
 373                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 374                 status = -EFAULT;
 375                 goto end;
 376         }
 377
 378         /* Tell driver that at least _CST is supported. */
 379         pr->flags.has_cst = 1;
 380
 381         for (i = 1; i <= count; i++) {
 382                 union acpi_object *element;
 383                 union acpi_object *obj;
 384                 struct acpi_power_register *reg;
 385                 struct acpi_processor_cx cx;
 386
 387                 memset(&cx, 0, sizeof(cx));
 388
 389                 element = &(cst->package.elements[i]);
 390                 if (element->type != ACPI_TYPE_PACKAGE)
 391                         continue;
 392
 393                 if (element->package.count != 4)
 394                         continue;
 395
 396                 obj = &(element->package.elements[0]);
 397
 398                 if (obj->type != ACPI_TYPE_BUFFER)
 399                         continue;
 400
 401                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 402
 403                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 404                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 405                         continue;
 406
 407                 /* There should be an easy way to extract an integer... */
 408                 obj = &(element->package.elements[1]);
 409                 if (obj->type != ACPI_TYPE_INTEGER)
 410                         continue;
 411
 412                 cx.type = obj->integer.value;
 413                 /*
 414                  * Some buggy BIOSes won't list C1 in _CST -
 415                  * Let acpi_processor_get_power_info_default() handle them later
 416                  */
 417                 if (i == 1 && cx.type != ACPI_STATE_C1)
 418                         current_count++;
 419
 420                 cx.address = reg->address;
 421                 cx.index = current_count + 1;
 422
 423                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 424                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 425                         if (acpi_processor_ffh_cstate_probe
 426                                         (pr->id, &cx, reg) == 0) {
 427                                 cx.entry_method = ACPI_CSTATE_FFH;
 428                         } else if (cx.type == ACPI_STATE_C1) {
 429                                 /*
 430                                  * C1 is a special case where FIXED_HARDWARE
 431                                  * can be handled in non-MWAIT way as well.
 432                                  * In that case, save this _CST entry info.
 433                                  * Otherwise, ignore this info and continue.
 434                                  */
 435                                 cx.entry_method = ACPI_CSTATE_HALT;
 436                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 437                         } else {
 438                                 continue;
 439                         }
 440                         if (cx.type == ACPI_STATE_C1 &&
 441                             (boot_option_idle_override == IDLE_NOMWAIT)) {
 442                                 /*
 443                                  * In most cases the C1 space_id obtained from
 444                                  * _CST object is FIXED_HARDWARE access mode.
 445                                  * But when the option of idle=halt is added,
 446                                  * the entry_method type should be changed from
 447                                  * CSTATE_FFH to CSTATE_HALT.
 448                                  * When the option of idle=nomwait is added,
 449                                  * the C1 entry_method type should be
 450                                  * CSTATE_HALT.
 451                                  */
 452                                 cx.entry_method = ACPI_CSTATE_HALT;
 453                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 454                         }
 455                 } else {
 456                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 457                                  cx.address);
 458                 }
 459
 460                 if (cx.type == ACPI_STATE_C1) {
 461                         cx.valid = 1;
 462                 }
 463
 464                 obj = &(element->package.elements[2]);
 465                 if (obj->type != ACPI_TYPE_INTEGER)
 466                         continue;
 467
 468                 cx.latency = obj->integer.value;
 469
 470                 obj = &(element->package.elements[3]);
 471                 if (obj->type != ACPI_TYPE_INTEGER)
 472                         continue;
 473
 474                 current_count++;
 475                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 476
 477                 /*
 478                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 479                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 480                  */
 481                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 482                         printk(KERN_WARNING
 483                                "Limiting number of power states to max (%d)\n",
 484                                ACPI_PROCESSOR_MAX_POWER);
 485                         printk(KERN_WARNING
 486                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 487                         break;
 488                 }
 489         }
 490
 491         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 492                           current_count));
 493
 494         /* Validate number of power states discovered */
 495         if (current_count < 2)
 496                 status = -EFAULT;
 497
 498       end:
 499         kfree(buffer.pointer);
 500
 501         return status;
 502 }
 503
 504 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 505                                            struct acpi_processor_cx *cx)
 506 {
 507         static int bm_check_flag = -1;
 508         static int bm_control_flag = -1;
 509
 510
 511         if (!cx->address)
 512                 return;
 513
 514         /*
 515          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 516          * DMA transfers are used by any ISA device to avoid livelock.
 517          * Note that we could disable Type-F DMA (as recommended by
 518          * the erratum), but this is known to disrupt certain ISA
 519          * devices thus we take the conservative approach.
 520          */
 521         else if (errata.piix4.fdma) {
 522                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 523                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 524                 return;
 525         }
 526
 527         /* All the logic here assumes flags.bm_check is same across all CPUs */
 528         if (bm_check_flag == -1) {
 529                 /* Determine whether bm_check is needed based on CPU  */
 530                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 531                 bm_check_flag = pr->flags.bm_check;
 532                 bm_control_flag = pr->flags.bm_control;
 533         } else {
 534                 pr->flags.bm_check = bm_check_flag;
 535                 pr->flags.bm_control = bm_control_flag;
 536         }
 537
 538         if (pr->flags.bm_check) {
 539                 if (!pr->flags.bm_control) {
 540                         if (pr->flags.has_cst != 1) {
 541                                 /* bus mastering control is necessary */
 542                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 543                                         "C3 support requires BM control\n"));
 544                                 return;
 545                         } else {
 546                                 /* Here we enter C3 without bus mastering */
 547                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 548                                         "C3 support without BM control\n"));
 549                         }
 550                 }
 551         } else {
 552                 /*
 553                  * WBINVD should be set in fadt, for C3 state to be
 554                  * supported on when bm_check is not required.
 555                  */
 556                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 557                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 558                                           "Cache invalidation should work properly"
 559                                           " for C3 to be enabled on SMP systems\n"));
 560                         return;
 561                 }
 562         }
 563
 564         /*
 565          * Otherwise we've met all of our C3 requirements.
 566          * Normalize the C3 latency to expidite policy.  Enable
 567          * checking of bus mastering status (bm_check) so we can
 568          * use this in our C3 policy
 569          */
 570         cx->valid = 1;
 571
 572         /*
 573          * On older chipsets, BM_RLD needs to be set
 574          * in order for Bus Master activity to wake the
 575          * system from C3.  Newer chipsets handle DMA
 576          * during C3 automatically and BM_RLD is a NOP.
 577          * In either case, the proper way to
 578          * handle BM_RLD is to set it and leave it set.
 579          */
 580         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 581
 582         return;
 583 }
 584
 585 static int acpi_processor_power_verify(struct acpi_processor *pr)
 586 {
 587         unsigned int i;
 588         unsigned int working = 0;
 589
 590         pr->power.timer_broadcast_on_state = INT_MAX;
 591
 592         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 593                 struct acpi_processor_cx *cx = &pr->power.states[i];
 594
 595                 switch (cx->type) {
 596                 case ACPI_STATE_C1:
 597                         cx->valid = 1;
 598                         break;
 599
 600                 case ACPI_STATE_C2:
 601                         if (!cx->address)
 602                                 break;
 603                         cx->valid = 1;
 604                         break;
 605
 606                 case ACPI_STATE_C3:
 607                         acpi_processor_power_verify_c3(pr, cx);
 608                         break;
 609                 }
 610                 if (!cx->valid)
 611                         continue;
 612
 613                 lapic_timer_check_state(i, pr, cx);
 614                 tsc_check_state(cx->type);
 615                 working++;
 616         }
 617
 618         lapic_timer_propagate_broadcast(pr);
 619
 620         return (working);
 621 }
 622
 623 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 624 {
 625         unsigned int i;
 626         int result;
 627
 628
 629         /* NOTE: the idle thread may not be running while calling
 630          * this function */
 631
 632         /* Zero initialize all the C-states info. */
 633         memset(pr->power.states, 0, sizeof(pr->power.states));
 634
 635         result = acpi_processor_get_power_info_cst(pr);
 636         if (result == -ENODEV)
 637                 result = acpi_processor_get_power_info_fadt(pr);
 638
 639         if (result)
 640                 return result;
 641
 642         acpi_processor_get_power_info_default(pr);
 643
 644         pr->power.count = acpi_processor_power_verify(pr);
 645
 646         /*
 647          * if one state of type C2 or C3 is available, mark this
 648          * CPU as being "idle manageable"
 649          */
 650         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 651                 if (pr->power.states[i].valid) {
 652                         pr->power.count = i;
 653                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 654                                 pr->flags.power = 1;
 655                 }
 656         }
 657
 658         return 0;
 659 }
 660
 661 /**
 662  * acpi_idle_bm_check - checks if bus master activity was detected
 663  */
 664 static int acpi_idle_bm_check(void)
 665 {
 666         u32 bm_status = 0;
 667
 668         if (bm_check_disable)
 669                 return 0;
 670
 671         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 672         if (bm_status)
 673                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 674         /*
 675          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 676          * the true state of bus mastering activity; forcing us to
 677          * manually check the BMIDEA bit of each IDE channel.
 678          */
 679         else if (errata.piix4.bmisx) {
 680                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 681                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 682                         bm_status = 1;
 683         }
 684         return bm_status;
 685 }
 686
 687 /**
 688  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
 689  * @cx: cstate data
 690  *
 691  * Caller disables interrupt before call and enables interrupt after return.
 692  */
 693 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
 694 {
 695         /* Don't trace irqs off for idle */
 696         stop_critical_timings();
 697         if (cx->entry_method == ACPI_CSTATE_FFH) {
 698                 /* Call into architectural FFH based C-state */
 699                 acpi_processor_ffh_cstate_enter(cx);
 700         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 701                 acpi_safe_halt();
 702         } else {
 703                 /* IO port based C-state */
 704                 inb(cx->address);
 705                 /* Dummy wait op - must do something useless after P_LVL2 read
 706                    because chipsets cannot guarantee that STPCLK# signal
 707                    gets asserted in time to freeze execution properly. */
 708                 inl(acpi_gbl_FADT.xpm_timer_block.address);
 709         }
 710         start_critical_timings();
 711 }
 712
 713 /**
 714  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
 715  * @dev: the target CPU
 716  * @drv: cpuidle driver containing cpuidle state info
 717  * @index: index of target state
 718  *
 719  * This is equivalent to the HALT instruction.
 720  */
 721 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
 722                 struct cpuidle_driver *drv, int index)
 723 {
 724         struct acpi_processor *pr;
 725         struct acpi_processor_cx *cx = acpi_cstate[index];
 726
 727         pr = __this_cpu_read(processors);
 728
 729         if (unlikely(!pr))
 730                 return -EINVAL;
 731
 732         lapic_timer_state_broadcast(pr, cx, 1);
 733         acpi_idle_do_entry(cx);
 734
 735         lapic_timer_state_broadcast(pr, cx, 0);
 736
 737         return index;
 738 }
 739
 740
 741 /**
 742  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
 743  * @dev: the target CPU
 744  * @index: the index of suggested state
 745  */
 746 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 747 {
 748         struct acpi_processor_cx *cx = acpi_cstate[index];
 749
 750         ACPI_FLUSH_CPU_CACHE();
 751
 752         while (1) {
 753
 754                 if (cx->entry_method == ACPI_CSTATE_HALT)
 755                         safe_halt();
 756                 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 757                         inb(cx->address);
 758                         /* See comment in acpi_idle_do_entry() */
 759                         inl(acpi_gbl_FADT.xpm_timer_block.address);
 760                 } else
 761                         return -ENODEV;
 762         }
 763
 764         /* Never reached */
 765         return 0;
 766 }
 767
 768 /**
 769  * acpi_idle_enter_simple - enters an ACPI state without BM handling
 770  * @dev: the target CPU
 771  * @drv: cpuidle driver with cpuidle state information
 772  * @index: the index of suggested state
 773  */
 774 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 775                 struct cpuidle_driver *drv, int index)
 776 {
 777         struct acpi_processor *pr;
 778         struct acpi_processor_cx *cx = acpi_cstate[index];
 779
 780         pr = __this_cpu_read(processors);
 781
 782         if (unlikely(!pr))
 783                 return -EINVAL;
 784
 785         if (cx->entry_method != ACPI_CSTATE_FFH) {
 786                 current_thread_info()->status &= ~TS_POLLING;
 787                 /*
 788                  * TS_POLLING-cleared state must be visible before we test
 789                  * NEED_RESCHED:
 790                  */
 791                 smp_mb();
 792
 793                 if (unlikely(need_resched())) {
 794                         current_thread_info()->status |= TS_POLLING;
 795                         return -EINVAL;
 796                 }
 797         }
 798
 799         /*
 800          * Must be done before busmaster disable as we might need to
 801          * access HPET !
 802          */
 803         lapic_timer_state_broadcast(pr, cx, 1);
 804
 805         if (cx->type == ACPI_STATE_C3)
 806                 ACPI_FLUSH_CPU_CACHE();
 807
 808         /* Tell the scheduler that we are going deep-idle: */
 809         sched_clock_idle_sleep_event();
 810         acpi_idle_do_entry(cx);
 811
 812         sched_clock_idle_wakeup_event(0);
 813
 814         if (cx->entry_method != ACPI_CSTATE_FFH)
 815                 current_thread_info()->status |= TS_POLLING;
 816
 817         lapic_timer_state_broadcast(pr, cx, 0);
 818         return index;
 819 }
 820
 821 static int c3_cpu_count;
 822 static DEFINE_RAW_SPINLOCK(c3_lock);
 823
 824 /**
 825  * acpi_idle_enter_bm - enters C3 with proper BM handling
 826  * @dev: the target CPU
 827  * @drv: cpuidle driver containing state data
 828  * @index: the index of suggested state
 829  *
 830  * If BM is detected, the deepest non-C3 idle state is entered instead.
 831  */
 832 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 833                 struct cpuidle_driver *drv, int index)
 834 {
 835         struct acpi_processor *pr;
 836         struct acpi_processor_cx *cx = acpi_cstate[index];
 837
 838         pr = __this_cpu_read(processors);
 839
 840         if (unlikely(!pr))
 841                 return -EINVAL;
 842
 843         if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
 844                 if (drv->safe_state_index >= 0) {
 845                         return drv->states[drv->safe_state_index].enter(dev,
 846                                                 drv, drv->safe_state_index);
 847                 } else {
 848                         acpi_safe_halt();
 849                         return -EBUSY;
 850                 }
 851         }
 852
 853         if (cx->entry_method != ACPI_CSTATE_FFH) {
 854                 current_thread_info()->status &= ~TS_POLLING;
 855                 /*
 856                  * TS_POLLING-cleared state must be visible before we test
 857                  * NEED_RESCHED:
 858                  */
 859                 smp_mb();
 860
 861                 if (unlikely(need_resched())) {
 862                         current_thread_info()->status |= TS_POLLING;
 863                         return -EINVAL;
 864                 }
 865         }
 866
 867         acpi_unlazy_tlb(smp_processor_id());
 868
 869         /* Tell the scheduler that we are going deep-idle: */
 870         sched_clock_idle_sleep_event();
 871         /*
 872          * Must be done before busmaster disable as we might need to
 873          * access HPET !
 874          */
 875         lapic_timer_state_broadcast(pr, cx, 1);
 876
 877         /*
 878          * disable bus master
 879          * bm_check implies we need ARB_DIS
 880          * !bm_check implies we need cache flush
 881          * bm_control implies whether we can do ARB_DIS
 882          *
 883          * That leaves a case where bm_check is set and bm_control is
 884          * not set. In that case we cannot do much, we enter C3
 885          * without doing anything.
 886          */
 887         if (pr->flags.bm_check && pr->flags.bm_control) {
 888                 raw_spin_lock(&c3_lock);
 889                 c3_cpu_count++;
 890                 /* Disable bus master arbitration when all CPUs are in C3 */
 891                 if (c3_cpu_count == num_online_cpus())
 892                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 893                 raw_spin_unlock(&c3_lock);
 894         } else if (!pr->flags.bm_check) {
 895                 ACPI_FLUSH_CPU_CACHE();
 896         }
 897
 898         acpi_idle_do_entry(cx);
 899
 900         /* Re-enable bus master arbitration */
 901         if (pr->flags.bm_check && pr->flags.bm_control) {
 902                 raw_spin_lock(&c3_lock);
 903                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 904                 c3_cpu_count--;
 905                 raw_spin_unlock(&c3_lock);
 906         }
 907
 908         sched_clock_idle_wakeup_event(0);
 909
 910         if (cx->entry_method != ACPI_CSTATE_FFH)
 911                 current_thread_info()->status |= TS_POLLING;
 912
 913         lapic_timer_state_broadcast(pr, cx, 0);
 914         return index;
 915 }
 916
 917 struct cpuidle_driver acpi_idle_driver = {
 918         .name =         "acpi_idle",
 919         .owner =        THIS_MODULE,
 920         .en_core_tk_irqen = 1,
 921 };
 922
 923 /**
 924  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
 925  * device i.e. per-cpu data
 926  *
 927  * @pr: the ACPI processor
 928  * @dev : the cpuidle device
 929  */
 930 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 931                                            struct cpuidle_device *dev)
 932 {
 933         int i, count = CPUIDLE_DRIVER_STATE_START;
 934         struct acpi_processor_cx *cx;
 935
 936         if (!pr->flags.power_setup_done)
 937                 return -EINVAL;
 938
 939         if (pr->flags.power == 0) {
 940                 return -EINVAL;
 941         }
 942
 943         if (!dev)
 944                 return -EINVAL;
 945
 946         dev->cpu = pr->id;
 947
 948         if (max_cstate == 0)
 949                 max_cstate = 1;
 950
 951         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 952                 cx = &pr->power.states[i];
 953
 954                 if (!cx->valid)
 955                         continue;
 956
 957 #ifdef CONFIG_HOTPLUG_CPU
 958                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
 959                     !pr->flags.has_cst &&
 960                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 961                         continue;
 962 #endif
 963                 acpi_cstate[count] = cx;
 964
 965                 count++;
 966                 if (count == CPUIDLE_STATE_MAX)
 967                         break;
 968         }
 969
 970         dev->state_count = count;
 971
 972         if (!count)
 973                 return -EINVAL;
 974
 975         return 0;
 976 }
 977
 978 /**
 979  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
 980  * global state data i.e. idle routines
 981  *
 982  * @pr: the ACPI processor
 983  */
 984 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
 985 {
 986         int i, count = CPUIDLE_DRIVER_STATE_START;
 987         struct acpi_processor_cx *cx;
 988         struct cpuidle_state *state;
 989         struct cpuidle_driver *drv = &acpi_idle_driver;
 990
 991         if (!pr->flags.power_setup_done)
 992                 return -EINVAL;
 993
 994         if (pr->flags.power == 0)
 995                 return -EINVAL;
 996
 997         drv->safe_state_index = -1;
 998         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
 999                 drv->states[i].name[0] = '\0';
1000                 drv->states[i].desc[0] = '\0';
1001         }
1002
1003         if (max_cstate == 0)
1004                 max_cstate = 1;
1005
1006         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1007                 cx = &pr->power.states[i];
1008
1009                 if (!cx->valid)
1010                         continue;
1011
1012 #ifdef CONFIG_HOTPLUG_CPU
1013                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1014                     !pr->flags.has_cst &&
1015                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1016                         continue;
1017 #endif
1018
1019                 state = &drv->states[count];
1020                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1021                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1022                 state->exit_latency = cx->latency;
1023                 state->target_residency = cx->latency * latency_factor;
1024
1025                 state->flags = 0;
1026                 switch (cx->type) {
1027                         case ACPI_STATE_C1:
1028                         if (cx->entry_method == ACPI_CSTATE_FFH)
1029                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1030
1031                         state->enter = acpi_idle_enter_c1;
1032                         state->enter_dead = acpi_idle_play_dead;
1033                         drv->safe_state_index = count;
1034                         break;
1035
1036                         case ACPI_STATE_C2:
1037                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1038                         state->enter = acpi_idle_enter_simple;
1039                         state->enter_dead = acpi_idle_play_dead;
1040                         drv->safe_state_index = count;
1041                         break;
1042
1043                         case ACPI_STATE_C3:
1044                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1045                         state->enter = pr->flags.bm_check ?
1046                                         acpi_idle_enter_bm :
1047                                         acpi_idle_enter_simple;
1048                         break;
1049                 }
1050
1051                 count++;
1052                 if (count == CPUIDLE_STATE_MAX)
1053                         break;
1054         }
1055
1056         drv->state_count = count;
1057
1058         if (!count)
1059                 return -EINVAL;
1060
1061         return 0;
1062 }
1063
1064 int acpi_processor_hotplug(struct acpi_processor *pr)
1065 {
1066         int ret = 0;
1067         struct cpuidle_device *dev;
1068
1069         if (disabled_by_idle_boot_param())
1070                 return 0;
1071
1072         if (!pr)
1073                 return -EINVAL;
1074
1075         if (nocst) {
1076                 return -ENODEV;
1077         }
1078
1079         if (!pr->flags.power_setup_done)
1080                 return -ENODEV;
1081
1082         dev = per_cpu(acpi_cpuidle_device, pr->id);
1083         cpuidle_pause_and_lock();
1084         cpuidle_disable_device(dev);
1085         acpi_processor_get_power_info(pr);
1086         if (pr->flags.power) {
1087                 acpi_processor_setup_cpuidle_cx(pr, dev);
1088                 ret = cpuidle_enable_device(dev);
1089         }
1090         cpuidle_resume_and_unlock();
1091
1092         return ret;
1093 }
1094
1095 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1096 {
1097         int cpu;
1098         struct acpi_processor *_pr;
1099         struct cpuidle_device *dev;
1100
1101         if (disabled_by_idle_boot_param())
1102                 return 0;
1103
1104         if (!pr)
1105                 return -EINVAL;
1106
1107         if (nocst)
1108                 return -ENODEV;
1109
1110         if (!pr->flags.power_setup_done)
1111                 return -ENODEV;
1112
1113         /*
1114          * FIXME:  Design the ACPI notification to make it once per
1115          * system instead of once per-cpu.  This condition is a hack
1116          * to make the code that updates C-States be called once.
1117          */
1118
1119         if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1120
1121                 cpuidle_pause_and_lock();
1122                 /* Protect against cpu-hotplug */
1123                 get_online_cpus();
1124
1125                 /* Disable all cpuidle devices */
1126                 for_each_online_cpu(cpu) {
1127                         _pr = per_cpu(processors, cpu);
1128                         if (!_pr || !_pr->flags.power_setup_done)
1129                                 continue;
1130                         dev = per_cpu(acpi_cpuidle_device, cpu);
1131                         cpuidle_disable_device(dev);
1132                 }
1133
1134                 /* Populate Updated C-state information */
1135                 acpi_processor_get_power_info(pr);
1136                 acpi_processor_setup_cpuidle_states(pr);
1137
1138                 /* Enable all cpuidle devices */
1139                 for_each_online_cpu(cpu) {
1140                         _pr = per_cpu(processors, cpu);
1141                         if (!_pr || !_pr->flags.power_setup_done)
1142                                 continue;
1143                         acpi_processor_get_power_info(_pr);
1144                         if (_pr->flags.power) {
1145                                 dev = per_cpu(acpi_cpuidle_device, cpu);
1146                                 acpi_processor_setup_cpuidle_cx(_pr, dev);
1147                                 cpuidle_enable_device(dev);
1148                         }
1149                 }
1150                 put_online_cpus();
1151                 cpuidle_resume_and_unlock();
1152         }
1153
1154         return 0;
1155 }
1156
1157 static int acpi_processor_registered;
1158
1159 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr)
1160 {
1161         acpi_status status = 0;
1162         int retval;
1163         struct cpuidle_device *dev;
1164         static int first_run;
1165
1166         if (disabled_by_idle_boot_param())
1167                 return 0;
1168
1169         if (!first_run) {
1170                 dmi_check_system(processor_power_dmi_table);
1171                 max_cstate = acpi_processor_cstate_check(max_cstate);
1172                 if (max_cstate < ACPI_C_STATES_MAX)
1173                         printk(KERN_NOTICE
1174                                "ACPI: processor limited to max C-state %d\n",
1175                                max_cstate);
1176                 first_run++;
1177         }
1178
1179         if (!pr)
1180                 return -EINVAL;
1181
1182         if (acpi_gbl_FADT.cst_control && !nocst) {
1183                 status =
1184                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1185                 if (ACPI_FAILURE(status)) {
1186                         ACPI_EXCEPTION((AE_INFO, status,
1187                                         "Notifying BIOS of _CST ability failed"));
1188                 }
1189         }
1190
1191         acpi_processor_get_power_info(pr);
1192         pr->flags.power_setup_done = 1;
1193
1194         /*
1195          * Install the idle handler if processor power management is supported.
1196          * Note that we use previously set idle handler will be used on
1197          * platforms that only support C1.
1198          */
1199         if (pr->flags.power) {
1200                 /* Register acpi_idle_driver if not already registered */
1201                 if (!acpi_processor_registered) {
1202                         acpi_processor_setup_cpuidle_states(pr);
1203                         retval = cpuidle_register_driver(&acpi_idle_driver);
1204                         if (retval)
1205                                 return retval;
1206                         printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1207                                         acpi_idle_driver.name);
1208                 }
1209
1210                 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1211                 if (!dev)
1212                         return -ENOMEM;
1213                 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1214
1215                 acpi_processor_setup_cpuidle_cx(pr, dev);
1216
1217                 /* Register per-cpu cpuidle_device. Cpuidle driver
1218                  * must already be registered before registering device
1219                  */
1220                 retval = cpuidle_register_device(dev);
1221                 if (retval) {
1222                         if (acpi_processor_registered == 0)
1223                                 cpuidle_unregister_driver(&acpi_idle_driver);
1224                         return retval;
1225                 }
1226                 acpi_processor_registered++;
1227         }
1228         return 0;
1229 }
1230
1231 int acpi_processor_power_exit(struct acpi_processor *pr)
1232 {
1233         struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1234
1235         if (disabled_by_idle_boot_param())
1236                 return 0;
1237
1238         if (pr->flags.power) {
1239                 cpuidle_unregister_device(dev);
1240                 acpi_processor_registered--;
1241                 if (acpi_processor_registered == 0)
1242                         cpuidle_unregister_driver(&acpi_idle_driver);
1243         }
1244
1245         pr->flags.power_setup_done = 0;
1246         return 0;
1247 }