drivers/acpi/processor_idle.c

   1 /*
   2  * processor_idle - idle state submodule to the ACPI processor driver
   3  *
   4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   6  *  Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
   7  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
   8  *                      - Added processor hotplug support
   9  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
  10  *                      - Added support for C3 on SMP
  11  *
  12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  13  *
  14  *  This program is free software; you can redistribute it and/or modify
  15  *  it under the terms of the GNU General Public License as published by
  16  *  the Free Software Foundation; either version 2 of the License, or (at
  17  *  your option) any later version.
  18  *
  19  *  This program is distributed in the hope that it will be useful, but
  20  *  WITHOUT ANY WARRANTY; without even the implied warranty of
  21  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  *  General Public License for more details.
  23  *
  24  *  You should have received a copy of the GNU General Public License along
  25  *  with this program; if not, write to the Free Software Foundation, Inc.,
  26  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
  27  *
  28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  29  */
  30
  31 #include <linux/kernel.h>
  32 #include <linux/module.h>
  33 #include <linux/init.h>
  34 #include <linux/cpufreq.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/acpi.h>
  38 #include <linux/dmi.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/sched.h>        /* need_resched() */
  41 #include <linux/pm_qos_params.h>
  42 #include <linux/clockchips.h>
  43 #include <linux/cpuidle.h>
  44 #include <linux/irqflags.h>
  45
  46 /*
  47  * Include the apic definitions for x86 to have the APIC timer related defines
  48  * available also for UP (on SMP it gets magically included via linux/smp.h).
  49  * asm/acpi.h is not an option, as it would require more include magic. Also
  50  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  51  */
  52 #ifdef CONFIG_X86
  53 #include <asm/apic.h>
  54 #endif
  55
  56 #include <asm/io.h>
  57 #include <asm/uaccess.h>
  58
  59 #include <acpi/acpi_bus.h>
  60 #include <acpi/processor.h>
  61 #include <asm/processor.h>
  62
  63 #define PREFIX "ACPI: "
  64
  65 #define ACPI_PROCESSOR_CLASS            "processor"
  66 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  67 ACPI_MODULE_NAME("processor_idle");
  68 #define ACPI_PROCESSOR_FILE_POWER       "power"
  69 #define PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
  70 #define C2_OVERHEAD                     1       /* 1us */
  71 #define C3_OVERHEAD                     1       /* 1us */
  72 #define PM_TIMER_TICKS_TO_US(p)         (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
  73
  74 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  75 module_param(max_cstate, uint, 0000);
  76 static unsigned int nocst __read_mostly;
  77 module_param(nocst, uint, 0000);
  78
  79 static unsigned int latency_factor __read_mostly = 2;
  80 module_param(latency_factor, uint, 0644);
  81
  82 static s64 us_to_pm_timer_ticks(s64 t)
  83 {
  84         return div64_u64(t * PM_TIMER_FREQUENCY, 1000000);
  85 }
  86 /*
  87  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  88  * For now disable this. Probably a bug somewhere else.
  89  *
  90  * To skip this limit, boot/load with a large max_cstate limit.
  91  */
  92 static int set_max_cstate(const struct dmi_system_id *id)
  93 {
  94         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  95                 return 0;
  96
  97         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  98                " Override with \"processor.max_cstate=%d\"\n", id->ident,
  99                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
 100
 101         max_cstate = (long)id->driver_data;
 102
 103         return 0;
 104 }
 105
 106 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
 107    callers to only run once -AK */
 108 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 109         { set_max_cstate, "Clevo 5600D", {
 110           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 111           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 112          (void *)2},
 113         {},
 114 };
 115
 116
 117 /*
 118  * Callers should disable interrupts before the call and enable
 119  * interrupts after return.
 120  */
 121 static void acpi_safe_halt(void)
 122 {
 123         current_thread_info()->status &= ~TS_POLLING;
 124         /*
 125          * TS_POLLING-cleared state must be visible before we
 126          * test NEED_RESCHED:
 127          */
 128         smp_mb();
 129         if (!need_resched()) {
 130                 safe_halt();
 131                 local_irq_disable();
 132         }
 133         current_thread_info()->status |= TS_POLLING;
 134 }
 135
 136 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 137
 138 /*
 139  * Some BIOS implementations switch to C3 in the published C2 state.
 140  * This seems to be a common problem on AMD boxen, but other vendors
 141  * are affected too. We pick the most conservative approach: we assume
 142  * that the local APIC stops in both C2 and C3.
 143  */
 144 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 145                                    struct acpi_processor_cx *cx)
 146 {
 147         struct acpi_processor_power *pwr = &pr->power;
 148         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 149
 150         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 151                 return;
 152
 153         if (boot_cpu_has(X86_FEATURE_AMDC1E))
 154                 type = ACPI_STATE_C1;
 155
 156         /*
 157          * Check, if one of the previous states already marked the lapic
 158          * unstable
 159          */
 160         if (pwr->timer_broadcast_on_state < state)
 161                 return;
 162
 163         if (cx->type >= type)
 164                 pr->power.timer_broadcast_on_state = state;
 165 }
 166
 167 static void __lapic_timer_propagate_broadcast(void *arg)
 168 {
 169         struct acpi_processor *pr = (struct acpi_processor *) arg;
 170         unsigned long reason;
 171
 172         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 173                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 174
 175         clockevents_notify(reason, &pr->id);
 176 }
 177
 178 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 179 {
 180         smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
 181                                  (void *)pr, 1);
 182 }
 183
 184 /* Power(C) State timer broadcast control */
 185 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 186                                        struct acpi_processor_cx *cx,
 187                                        int broadcast)
 188 {
 189         int state = cx - pr->power.states;
 190
 191         if (state >= pr->power.timer_broadcast_on_state) {
 192                 unsigned long reason;
 193
 194                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 195                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 196                 clockevents_notify(reason, &pr->id);
 197         }
 198 }
 199
 200 #else
 201
 202 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 203                                    struct acpi_processor_cx *cstate) { }
 204 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 205 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
 206                                        struct acpi_processor_cx *cx,
 207                                        int broadcast)
 208 {
 209 }
 210
 211 #endif
 212
 213 /*
 214  * Suspend / resume control
 215  */
 216 static int acpi_idle_suspend;
 217 static u32 saved_bm_rld;
 218
 219 static void acpi_idle_bm_rld_save(void)
 220 {
 221         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
 222 }
 223 static void acpi_idle_bm_rld_restore(void)
 224 {
 225         u32 resumed_bm_rld;
 226
 227         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
 228
 229         if (resumed_bm_rld != saved_bm_rld)
 230                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
 231 }
 232
 233 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
 234 {
 235         if (acpi_idle_suspend == 1)
 236                 return 0;
 237
 238         acpi_idle_bm_rld_save();
 239         acpi_idle_suspend = 1;
 240         return 0;
 241 }
 242
 243 int acpi_processor_resume(struct acpi_device * device)
 244 {
 245         if (acpi_idle_suspend == 0)
 246                 return 0;
 247
 248         acpi_idle_bm_rld_restore();
 249         acpi_idle_suspend = 0;
 250         return 0;
 251 }
 252
 253 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 254 static void tsc_check_state(int state)
 255 {
 256         switch (boot_cpu_data.x86_vendor) {
 257         case X86_VENDOR_AMD:
 258         case X86_VENDOR_INTEL:
 259                 /*
 260                  * AMD Fam10h TSC will tick in all
 261                  * C/P/S0/S1 states when this bit is set.
 262                  */
 263                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 264                         return;
 265
 266                 /*FALL THROUGH*/
 267         default:
 268                 /* TSC could halt in idle, so notify users */
 269                 if (state > ACPI_STATE_C1)
 270                         mark_tsc_unstable("TSC halts in idle");
 271         }
 272 }
 273 #else
 274 static void tsc_check_state(int state) { return; }
 275 #endif
 276
 277 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 278 {
 279
 280         if (!pr)
 281                 return -EINVAL;
 282
 283         if (!pr->pblk)
 284                 return -ENODEV;
 285
 286         /* if info is obtained from pblk/fadt, type equals state */
 287         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 288         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 289
 290 #ifndef CONFIG_HOTPLUG_CPU
 291         /*
 292          * Check for P_LVL2_UP flag before entering C2 and above on
 293          * an SMP system.
 294          */
 295         if ((num_online_cpus() > 1) &&
 296             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 297                 return -ENODEV;
 298 #endif
 299
 300         /* determine C2 and C3 address from pblk */
 301         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 302         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 303
 304         /* determine latencies from FADT */
 305         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 306         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 307
 308         /*
 309          * FADT specified C2 latency must be less than or equal to
 310          * 100 microseconds.
 311          */
 312         if (acpi_gbl_FADT.C2latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 313                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 314                         "C2 latency too large [%d]\n", acpi_gbl_FADT.C2latency));
 315                 /* invalidate C2 */
 316                 pr->power.states[ACPI_STATE_C2].address = 0;
 317         }
 318
 319         /*
 320          * FADT supplied C3 latency must be less than or equal to
 321          * 1000 microseconds.
 322          */
 323         if (acpi_gbl_FADT.C3latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 324                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 325                         "C3 latency too large [%d]\n", acpi_gbl_FADT.C3latency));
 326                 /* invalidate C3 */
 327                 pr->power.states[ACPI_STATE_C3].address = 0;
 328         }
 329
 330         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 331                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 332                           pr->power.states[ACPI_STATE_C2].address,
 333                           pr->power.states[ACPI_STATE_C3].address));
 334
 335         return 0;
 336 }
 337
 338 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 339 {
 340         if (!pr->power.states[ACPI_STATE_C1].valid) {
 341                 /* set the first C-State to C1 */
 342                 /* all processors need to support C1 */
 343                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 344                 pr->power.states[ACPI_STATE_C1].valid = 1;
 345                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 346         }
 347         /* the C0 state only exists as a filler in our array */
 348         pr->power.states[ACPI_STATE_C0].valid = 1;
 349         return 0;
 350 }
 351
 352 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 353 {
 354         acpi_status status = 0;
 355         acpi_integer count;
 356         int current_count;
 357         int i;
 358         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 359         union acpi_object *cst;
 360
 361
 362         if (nocst)
 363                 return -ENODEV;
 364
 365         current_count = 0;
 366
 367         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 368         if (ACPI_FAILURE(status)) {
 369                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 370                 return -ENODEV;
 371         }
 372
 373         cst = buffer.pointer;
 374
 375         /* There must be at least 2 elements */
 376         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 377                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 378                 status = -EFAULT;
 379                 goto end;
 380         }
 381
 382         count = cst->package.elements[0].integer.value;
 383
 384         /* Validate number of power states. */
 385         if (count < 1 || count != cst->package.count - 1) {
 386                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 387                 status = -EFAULT;
 388                 goto end;
 389         }
 390
 391         /* Tell driver that at least _CST is supported. */
 392         pr->flags.has_cst = 1;
 393
 394         for (i = 1; i <= count; i++) {
 395                 union acpi_object *element;
 396                 union acpi_object *obj;
 397                 struct acpi_power_register *reg;
 398                 struct acpi_processor_cx cx;
 399
 400                 memset(&cx, 0, sizeof(cx));
 401
 402                 element = &(cst->package.elements[i]);
 403                 if (element->type != ACPI_TYPE_PACKAGE)
 404                         continue;
 405
 406                 if (element->package.count != 4)
 407                         continue;
 408
 409                 obj = &(element->package.elements[0]);
 410
 411                 if (obj->type != ACPI_TYPE_BUFFER)
 412                         continue;
 413
 414                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 415
 416                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 417                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 418                         continue;
 419
 420                 /* There should be an easy way to extract an integer... */
 421                 obj = &(element->package.elements[1]);
 422                 if (obj->type != ACPI_TYPE_INTEGER)
 423                         continue;
 424
 425                 cx.type = obj->integer.value;
 426                 /*
 427                  * Some buggy BIOSes won't list C1 in _CST -
 428                  * Let acpi_processor_get_power_info_default() handle them later
 429                  */
 430                 if (i == 1 && cx.type != ACPI_STATE_C1)
 431                         current_count++;
 432
 433                 cx.address = reg->address;
 434                 cx.index = current_count + 1;
 435
 436                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 437                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 438                         if (acpi_processor_ffh_cstate_probe
 439                                         (pr->id, &cx, reg) == 0) {
 440                                 cx.entry_method = ACPI_CSTATE_FFH;
 441                         } else if (cx.type == ACPI_STATE_C1) {
 442                                 /*
 443                                  * C1 is a special case where FIXED_HARDWARE
 444                                  * can be handled in non-MWAIT way as well.
 445                                  * In that case, save this _CST entry info.
 446                                  * Otherwise, ignore this info and continue.
 447                                  */
 448                                 cx.entry_method = ACPI_CSTATE_HALT;
 449                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 450                         } else {
 451                                 continue;
 452                         }
 453                         if (cx.type == ACPI_STATE_C1 &&
 454                                         (idle_halt || idle_nomwait)) {
 455                                 /*
 456                                  * In most cases the C1 space_id obtained from
 457                                  * _CST object is FIXED_HARDWARE access mode.
 458                                  * But when the option of idle=halt is added,
 459                                  * the entry_method type should be changed from
 460                                  * CSTATE_FFH to CSTATE_HALT.
 461                                  * When the option of idle=nomwait is added,
 462                                  * the C1 entry_method type should be
 463                                  * CSTATE_HALT.
 464                                  */
 465                                 cx.entry_method = ACPI_CSTATE_HALT;
 466                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 467                         }
 468                 } else {
 469                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 470                                  cx.address);
 471                 }
 472
 473                 if (cx.type == ACPI_STATE_C1) {
 474                         cx.valid = 1;
 475                 }
 476
 477                 obj = &(element->package.elements[2]);
 478                 if (obj->type != ACPI_TYPE_INTEGER)
 479                         continue;
 480
 481                 cx.latency = obj->integer.value;
 482
 483                 obj = &(element->package.elements[3]);
 484                 if (obj->type != ACPI_TYPE_INTEGER)
 485                         continue;
 486
 487                 cx.power = obj->integer.value;
 488
 489                 current_count++;
 490                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 491
 492                 /*
 493                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 494                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 495                  */
 496                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 497                         printk(KERN_WARNING
 498                                "Limiting number of power states to max (%d)\n",
 499                                ACPI_PROCESSOR_MAX_POWER);
 500                         printk(KERN_WARNING
 501                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 502                         break;
 503                 }
 504         }
 505
 506         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 507                           current_count));
 508
 509         /* Validate number of power states discovered */
 510         if (current_count < 2)
 511                 status = -EFAULT;
 512
 513       end:
 514         kfree(buffer.pointer);
 515
 516         return status;
 517 }
 518
 519 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 520                                            struct acpi_processor_cx *cx)
 521 {
 522         static int bm_check_flag = -1;
 523         static int bm_control_flag = -1;
 524
 525
 526         if (!cx->address)
 527                 return;
 528
 529         /*
 530          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 531          * DMA transfers are used by any ISA device to avoid livelock.
 532          * Note that we could disable Type-F DMA (as recommended by
 533          * the erratum), but this is known to disrupt certain ISA
 534          * devices thus we take the conservative approach.
 535          */
 536         else if (errata.piix4.fdma) {
 537                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 538                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 539                 return;
 540         }
 541
 542         /* All the logic here assumes flags.bm_check is same across all CPUs */
 543         if (bm_check_flag == -1) {
 544                 /* Determine whether bm_check is needed based on CPU  */
 545                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 546                 bm_check_flag = pr->flags.bm_check;
 547                 bm_control_flag = pr->flags.bm_control;
 548         } else {
 549                 pr->flags.bm_check = bm_check_flag;
 550                 pr->flags.bm_control = bm_control_flag;
 551         }
 552
 553         if (pr->flags.bm_check) {
 554                 if (!pr->flags.bm_control) {
 555                         if (pr->flags.has_cst != 1) {
 556                                 /* bus mastering control is necessary */
 557                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 558                                         "C3 support requires BM control\n"));
 559                                 return;
 560                         } else {
 561                                 /* Here we enter C3 without bus mastering */
 562                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 563                                         "C3 support without BM control\n"));
 564                         }
 565                 }
 566         } else {
 567                 /*
 568                  * WBINVD should be set in fadt, for C3 state to be
 569                  * supported on when bm_check is not required.
 570                  */
 571                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 572                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 573                                           "Cache invalidation should work properly"
 574                                           " for C3 to be enabled on SMP systems\n"));
 575                         return;
 576                 }
 577         }
 578
 579         /*
 580          * Otherwise we've met all of our C3 requirements.
 581          * Normalize the C3 latency to expidite policy.  Enable
 582          * checking of bus mastering status (bm_check) so we can
 583          * use this in our C3 policy
 584          */
 585         cx->valid = 1;
 586
 587         cx->latency_ticks = cx->latency;
 588         /*
 589          * On older chipsets, BM_RLD needs to be set
 590          * in order for Bus Master activity to wake the
 591          * system from C3.  Newer chipsets handle DMA
 592          * during C3 automatically and BM_RLD is a NOP.
 593          * In either case, the proper way to
 594          * handle BM_RLD is to set it and leave it set.
 595          */
 596         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 597
 598         return;
 599 }
 600
 601 static int acpi_processor_power_verify(struct acpi_processor *pr)
 602 {
 603         unsigned int i;
 604         unsigned int working = 0;
 605
 606         pr->power.timer_broadcast_on_state = INT_MAX;
 607
 608         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 609                 struct acpi_processor_cx *cx = &pr->power.states[i];
 610
 611                 switch (cx->type) {
 612                 case ACPI_STATE_C1:
 613                         cx->valid = 1;
 614                         break;
 615
 616                 case ACPI_STATE_C2:
 617                         if (!cx->address)
 618                                 break;
 619                         cx->valid = 1;
 620                         cx->latency_ticks = cx->latency; /* Normalize latency */
 621                         break;
 622
 623                 case ACPI_STATE_C3:
 624                         acpi_processor_power_verify_c3(pr, cx);
 625                         break;
 626                 }
 627                 if (!cx->valid)
 628                         continue;
 629
 630                 lapic_timer_check_state(i, pr, cx);
 631                 tsc_check_state(cx->type);
 632                 working++;
 633         }
 634
 635         lapic_timer_propagate_broadcast(pr);
 636
 637         return (working);
 638 }
 639
 640 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 641 {
 642         unsigned int i;
 643         int result;
 644
 645
 646         /* NOTE: the idle thread may not be running while calling
 647          * this function */
 648
 649         /* Zero initialize all the C-states info. */
 650         memset(pr->power.states, 0, sizeof(pr->power.states));
 651
 652         result = acpi_processor_get_power_info_cst(pr);
 653         if (result == -ENODEV)
 654                 result = acpi_processor_get_power_info_fadt(pr);
 655
 656         if (result)
 657                 return result;
 658
 659         acpi_processor_get_power_info_default(pr);
 660
 661         pr->power.count = acpi_processor_power_verify(pr);
 662
 663         /*
 664          * if one state of type C2 or C3 is available, mark this
 665          * CPU as being "idle manageable"
 666          */
 667         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 668                 if (pr->power.states[i].valid) {
 669                         pr->power.count = i;
 670                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 671                                 pr->flags.power = 1;
 672                 }
 673         }
 674
 675         return 0;
 676 }
 677
 678 #ifdef CONFIG_ACPI_PROCFS
 679 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 680 {
 681         struct acpi_processor *pr = seq->private;
 682         unsigned int i;
 683
 684
 685         if (!pr)
 686                 goto end;
 687
 688         seq_printf(seq, "active state:            C%zd\n"
 689                    "max_cstate:              C%d\n"
 690                    "maximum allowed latency: %d usec\n",
 691                    pr->power.state ? pr->power.state - pr->power.states : 0,
 692                    max_cstate, pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
 693
 694         seq_puts(seq, "states:\n");
 695
 696         for (i = 1; i <= pr->power.count; i++) {
 697                 seq_printf(seq, "   %cC%d:                  ",
 698                            (&pr->power.states[i] ==
 699                             pr->power.state ? '*' : ' '), i);
 700
 701                 if (!pr->power.states[i].valid) {
 702                         seq_puts(seq, "<not supported>\n");
 703                         continue;
 704                 }
 705
 706                 switch (pr->power.states[i].type) {
 707                 case ACPI_STATE_C1:
 708                         seq_printf(seq, "type[C1] ");
 709                         break;
 710                 case ACPI_STATE_C2:
 711                         seq_printf(seq, "type[C2] ");
 712                         break;
 713                 case ACPI_STATE_C3:
 714                         seq_printf(seq, "type[C3] ");
 715                         break;
 716                 default:
 717                         seq_printf(seq, "type[--] ");
 718                         break;
 719                 }
 720
 721                 if (pr->power.states[i].promotion.state)
 722                         seq_printf(seq, "promotion[C%zd] ",
 723                                    (pr->power.states[i].promotion.state -
 724                                     pr->power.states));
 725                 else
 726                         seq_puts(seq, "promotion[--] ");
 727
 728                 if (pr->power.states[i].demotion.state)
 729                         seq_printf(seq, "demotion[C%zd] ",
 730                                    (pr->power.states[i].demotion.state -
 731                                     pr->power.states));
 732                 else
 733                         seq_puts(seq, "demotion[--] ");
 734
 735                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
 736                            pr->power.states[i].latency,
 737                            pr->power.states[i].usage,
 738                            (unsigned long long)pr->power.states[i].time);
 739         }
 740
 741       end:
 742         return 0;
 743 }
 744
 745 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
 746 {
 747         return single_open(file, acpi_processor_power_seq_show,
 748                            PDE(inode)->data);
 749 }
 750
 751 static const struct file_operations acpi_processor_power_fops = {
 752         .owner = THIS_MODULE,
 753         .open = acpi_processor_power_open_fs,
 754         .read = seq_read,
 755         .llseek = seq_lseek,
 756         .release = single_release,
 757 };
 758 #endif
 759
 760 /**
 761  * acpi_idle_bm_check - checks if bus master activity was detected
 762  */
 763 static int acpi_idle_bm_check(void)
 764 {
 765         u32 bm_status = 0;
 766
 767         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 768         if (bm_status)
 769                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 770         /*
 771          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 772          * the true state of bus mastering activity; forcing us to
 773          * manually check the BMIDEA bit of each IDE channel.
 774          */
 775         else if (errata.piix4.bmisx) {
 776                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 777                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 778                         bm_status = 1;
 779         }
 780         return bm_status;
 781 }
 782
 783 /**
 784  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
 785  * @cx: cstate data
 786  *
 787  * Caller disables interrupt before call and enables interrupt after return.
 788  */
 789 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
 790 {
 791         /* Don't trace irqs off for idle */
 792         stop_critical_timings();
 793         if (cx->entry_method == ACPI_CSTATE_FFH) {
 794                 /* Call into architectural FFH based C-state */
 795                 acpi_processor_ffh_cstate_enter(cx);
 796         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 797                 acpi_safe_halt();
 798         } else {
 799                 int unused;
 800                 /* IO port based C-state */
 801                 inb(cx->address);
 802                 /* Dummy wait op - must do something useless after P_LVL2 read
 803                    because chipsets cannot guarantee that STPCLK# signal
 804                    gets asserted in time to freeze execution properly. */
 805                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 806         }
 807         start_critical_timings();
 808 }
 809
 810 /**
 811  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
 812  * @dev: the target CPU
 813  * @state: the state data
 814  *
 815  * This is equivalent to the HALT instruction.
 816  */
 817 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
 818                               struct cpuidle_state *state)
 819 {
 820         ktime_t  kt1, kt2;
 821         s64 idle_time;
 822         struct acpi_processor *pr;
 823         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 824
 825         pr = __get_cpu_var(processors);
 826
 827         if (unlikely(!pr))
 828                 return 0;
 829
 830         local_irq_disable();
 831
 832         /* Do not access any ACPI IO ports in suspend path */
 833         if (acpi_idle_suspend) {
 834                 local_irq_enable();
 835                 cpu_relax();
 836                 return 0;
 837         }
 838
 839         lapic_timer_state_broadcast(pr, cx, 1);
 840         kt1 = ktime_get_real();
 841         acpi_idle_do_entry(cx);
 842         kt2 = ktime_get_real();
 843         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 844
 845         local_irq_enable();
 846         cx->usage++;
 847         lapic_timer_state_broadcast(pr, cx, 0);
 848
 849         return idle_time;
 850 }
 851
 852 /**
 853  * acpi_idle_enter_simple - enters an ACPI state without BM handling
 854  * @dev: the target CPU
 855  * @state: the state data
 856  */
 857 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 858                                   struct cpuidle_state *state)
 859 {
 860         struct acpi_processor *pr;
 861         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 862         ktime_t  kt1, kt2;
 863         s64 idle_time;
 864         s64 sleep_ticks = 0;
 865
 866         pr = __get_cpu_var(processors);
 867
 868         if (unlikely(!pr))
 869                 return 0;
 870
 871         if (acpi_idle_suspend)
 872                 return(acpi_idle_enter_c1(dev, state));
 873
 874         local_irq_disable();
 875         current_thread_info()->status &= ~TS_POLLING;
 876         /*
 877          * TS_POLLING-cleared state must be visible before we test
 878          * NEED_RESCHED:
 879          */
 880         smp_mb();
 881
 882         if (unlikely(need_resched())) {
 883                 current_thread_info()->status |= TS_POLLING;
 884                 local_irq_enable();
 885                 return 0;
 886         }
 887
 888         /*
 889          * Must be done before busmaster disable as we might need to
 890          * access HPET !
 891          */
 892         lapic_timer_state_broadcast(pr, cx, 1);
 893
 894         if (cx->type == ACPI_STATE_C3)
 895                 ACPI_FLUSH_CPU_CACHE();
 896
 897         kt1 = ktime_get_real();
 898         /* Tell the scheduler that we are going deep-idle: */
 899         sched_clock_idle_sleep_event();
 900         acpi_idle_do_entry(cx);
 901         kt2 = ktime_get_real();
 902         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 903
 904         sleep_ticks = us_to_pm_timer_ticks(idle_time);
 905
 906         /* Tell the scheduler how much we idled: */
 907         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 908
 909         local_irq_enable();
 910         current_thread_info()->status |= TS_POLLING;
 911
 912         cx->usage++;
 913
 914         lapic_timer_state_broadcast(pr, cx, 0);
 915         cx->time += sleep_ticks;
 916         return idle_time;
 917 }
 918
 919 static int c3_cpu_count;
 920 static DEFINE_SPINLOCK(c3_lock);
 921
 922 /**
 923  * acpi_idle_enter_bm - enters C3 with proper BM handling
 924  * @dev: the target CPU
 925  * @state: the state data
 926  *
 927  * If BM is detected, the deepest non-C3 idle state is entered instead.
 928  */
 929 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 930                               struct cpuidle_state *state)
 931 {
 932         struct acpi_processor *pr;
 933         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 934         ktime_t  kt1, kt2;
 935         s64 idle_time;
 936         s64 sleep_ticks = 0;
 937
 938
 939         pr = __get_cpu_var(processors);
 940
 941         if (unlikely(!pr))
 942                 return 0;
 943
 944         if (acpi_idle_suspend)
 945                 return(acpi_idle_enter_c1(dev, state));
 946
 947         if (acpi_idle_bm_check()) {
 948                 if (dev->safe_state) {
 949                         dev->last_state = dev->safe_state;
 950                         return dev->safe_state->enter(dev, dev->safe_state);
 951                 } else {
 952                         local_irq_disable();
 953                         acpi_safe_halt();
 954                         local_irq_enable();
 955                         return 0;
 956                 }
 957         }
 958
 959         local_irq_disable();
 960         current_thread_info()->status &= ~TS_POLLING;
 961         /*
 962          * TS_POLLING-cleared state must be visible before we test
 963          * NEED_RESCHED:
 964          */
 965         smp_mb();
 966
 967         if (unlikely(need_resched())) {
 968                 current_thread_info()->status |= TS_POLLING;
 969                 local_irq_enable();
 970                 return 0;
 971         }
 972
 973         acpi_unlazy_tlb(smp_processor_id());
 974
 975         /* Tell the scheduler that we are going deep-idle: */
 976         sched_clock_idle_sleep_event();
 977         /*
 978          * Must be done before busmaster disable as we might need to
 979          * access HPET !
 980          */
 981         lapic_timer_state_broadcast(pr, cx, 1);
 982
 983         kt1 = ktime_get_real();
 984         /*
 985          * disable bus master
 986          * bm_check implies we need ARB_DIS
 987          * !bm_check implies we need cache flush
 988          * bm_control implies whether we can do ARB_DIS
 989          *
 990          * That leaves a case where bm_check is set and bm_control is
 991          * not set. In that case we cannot do much, we enter C3
 992          * without doing anything.
 993          */
 994         if (pr->flags.bm_check && pr->flags.bm_control) {
 995                 spin_lock(&c3_lock);
 996                 c3_cpu_count++;
 997                 /* Disable bus master arbitration when all CPUs are in C3 */
 998                 if (c3_cpu_count == num_online_cpus())
 999                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
1000                 spin_unlock(&c3_lock);
1001         } else if (!pr->flags.bm_check) {
1002                 ACPI_FLUSH_CPU_CACHE();
1003         }
1004
1005         acpi_idle_do_entry(cx);
1006
1007         /* Re-enable bus master arbitration */
1008         if (pr->flags.bm_check && pr->flags.bm_control) {
1009                 spin_lock(&c3_lock);
1010                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
1011                 c3_cpu_count--;
1012                 spin_unlock(&c3_lock);
1013         }
1014         kt2 = ktime_get_real();
1015         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
1016
1017         sleep_ticks = us_to_pm_timer_ticks(idle_time);
1018         /* Tell the scheduler how much we idled: */
1019         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
1020
1021         local_irq_enable();
1022         current_thread_info()->status |= TS_POLLING;
1023
1024         cx->usage++;
1025
1026         lapic_timer_state_broadcast(pr, cx, 0);
1027         cx->time += sleep_ticks;
1028         return idle_time;
1029 }
1030
1031 struct cpuidle_driver acpi_idle_driver = {
1032         .name =         "acpi_idle",
1033         .owner =        THIS_MODULE,
1034 };
1035
1036 /**
1037  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1038  * @pr: the ACPI processor
1039  */
1040 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1041 {
1042         int i, count = CPUIDLE_DRIVER_STATE_START;
1043         struct acpi_processor_cx *cx;
1044         struct cpuidle_state *state;
1045         struct cpuidle_device *dev = &pr->power.dev;
1046
1047         if (!pr->flags.power_setup_done)
1048                 return -EINVAL;
1049
1050         if (pr->flags.power == 0) {
1051                 return -EINVAL;
1052         }
1053
1054         dev->cpu = pr->id;
1055         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1056                 dev->states[i].name[0] = '\0';
1057                 dev->states[i].desc[0] = '\0';
1058         }
1059
1060         if (max_cstate == 0)
1061                 max_cstate = 1;
1062
1063         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1064                 cx = &pr->power.states[i];
1065                 state = &dev->states[count];
1066
1067                 if (!cx->valid)
1068                         continue;
1069
1070 #ifdef CONFIG_HOTPLUG_CPU
1071                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1072                     !pr->flags.has_cst &&
1073                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1074                         continue;
1075 #endif
1076                 cpuidle_set_statedata(state, cx);
1077
1078                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1079                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1080                 state->exit_latency = cx->latency;
1081                 state->target_residency = cx->latency * latency_factor;
1082                 state->power_usage = cx->power;
1083
1084                 state->flags = 0;
1085                 switch (cx->type) {
1086                         case ACPI_STATE_C1:
1087                         state->flags |= CPUIDLE_FLAG_SHALLOW;
1088                         if (cx->entry_method == ACPI_CSTATE_FFH)
1089                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1090
1091                         state->enter = acpi_idle_enter_c1;
1092                         dev->safe_state = state;
1093                         break;
1094
1095                         case ACPI_STATE_C2:
1096                         state->flags |= CPUIDLE_FLAG_BALANCED;
1097                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1098                         state->enter = acpi_idle_enter_simple;
1099                         dev->safe_state = state;
1100                         break;
1101
1102                         case ACPI_STATE_C3:
1103                         state->flags |= CPUIDLE_FLAG_DEEP;
1104                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1105                         state->flags |= CPUIDLE_FLAG_CHECK_BM;
1106                         state->enter = pr->flags.bm_check ?
1107                                         acpi_idle_enter_bm :
1108                                         acpi_idle_enter_simple;
1109                         break;
1110                 }
1111
1112                 count++;
1113                 if (count == CPUIDLE_STATE_MAX)
1114                         break;
1115         }
1116
1117         dev->state_count = count;
1118
1119         if (!count)
1120                 return -EINVAL;
1121
1122         return 0;
1123 }
1124
1125 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1126 {
1127         int ret = 0;
1128
1129         if (boot_option_idle_override)
1130                 return 0;
1131
1132         if (!pr)
1133                 return -EINVAL;
1134
1135         if (nocst) {
1136                 return -ENODEV;
1137         }
1138
1139         if (!pr->flags.power_setup_done)
1140                 return -ENODEV;
1141
1142         cpuidle_pause_and_lock();
1143         cpuidle_disable_device(&pr->power.dev);
1144         acpi_processor_get_power_info(pr);
1145         if (pr->flags.power) {
1146                 acpi_processor_setup_cpuidle(pr);
1147                 ret = cpuidle_enable_device(&pr->power.dev);
1148         }
1149         cpuidle_resume_and_unlock();
1150
1151         return ret;
1152 }
1153
1154 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1155                               struct acpi_device *device)
1156 {
1157         acpi_status status = 0;
1158         static int first_run;
1159 #ifdef CONFIG_ACPI_PROCFS
1160         struct proc_dir_entry *entry = NULL;
1161 #endif
1162
1163         if (boot_option_idle_override)
1164                 return 0;
1165
1166         if (!first_run) {
1167                 if (idle_halt) {
1168                         /*
1169                          * When the boot option of "idle=halt" is added, halt
1170                          * is used for CPU IDLE.
1171                          * In such case C2/C3 is meaningless. So the max_cstate
1172                          * is set to one.
1173                          */
1174                         max_cstate = 1;
1175                 }
1176                 dmi_check_system(processor_power_dmi_table);
1177                 max_cstate = acpi_processor_cstate_check(max_cstate);
1178                 if (max_cstate < ACPI_C_STATES_MAX)
1179                         printk(KERN_NOTICE
1180                                "ACPI: processor limited to max C-state %d\n",
1181                                max_cstate);
1182                 first_run++;
1183         }
1184
1185         if (!pr)
1186                 return -EINVAL;
1187
1188         if (acpi_gbl_FADT.cst_control && !nocst) {
1189                 status =
1190                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1191                 if (ACPI_FAILURE(status)) {
1192                         ACPI_EXCEPTION((AE_INFO, status,
1193                                         "Notifying BIOS of _CST ability failed"));
1194                 }
1195         }
1196
1197         acpi_processor_get_power_info(pr);
1198         pr->flags.power_setup_done = 1;
1199
1200         /*
1201          * Install the idle handler if processor power management is supported.
1202          * Note that we use previously set idle handler will be used on
1203          * platforms that only support C1.
1204          */
1205         if (pr->flags.power) {
1206                 acpi_processor_setup_cpuidle(pr);
1207                 if (cpuidle_register_device(&pr->power.dev))
1208                         return -EIO;
1209         }
1210 #ifdef CONFIG_ACPI_PROCFS
1211         /* 'power' [R] */
1212         entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1213                                  S_IRUGO, acpi_device_dir(device),
1214                                  &acpi_processor_power_fops,
1215                                  acpi_driver_data(device));
1216         if (!entry)
1217                 return -EIO;
1218 #endif
1219         return 0;
1220 }
1221
1222 int acpi_processor_power_exit(struct acpi_processor *pr,
1223                               struct acpi_device *device)
1224 {
1225         if (boot_option_idle_override)
1226                 return 0;
1227
1228         cpuidle_unregister_device(&pr->power.dev);
1229         pr->flags.power_setup_done = 0;
1230
1231 #ifdef CONFIG_ACPI_PROCFS
1232         if (acpi_device_dir(device))
1233                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1234                                   acpi_device_dir(device));
1235 #endif
1236
1237         return 0;
1238 }