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 ACPI_PROCESSOR_CLASS            "processor"
  64 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
  65 ACPI_MODULE_NAME("processor_idle");
  66 #define ACPI_PROCESSOR_FILE_POWER       "power"
  67 #define PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
  68 #define C2_OVERHEAD                     1       /* 1us */
  69 #define C3_OVERHEAD                     1       /* 1us */
  70 #define PM_TIMER_TICKS_TO_US(p)         (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
  71
  72 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  73 module_param(max_cstate, uint, 0000);
  74 static unsigned int nocst __read_mostly;
  75 module_param(nocst, uint, 0000);
  76
  77 static unsigned int latency_factor __read_mostly = 2;
  78 module_param(latency_factor, uint, 0644);
  79
  80 static s64 us_to_pm_timer_ticks(s64 t)
  81 {
  82         return div64_u64(t * PM_TIMER_FREQUENCY, 1000000);
  83 }
  84 /*
  85  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  86  * For now disable this. Probably a bug somewhere else.
  87  *
  88  * To skip this limit, boot/load with a large max_cstate limit.
  89  */
  90 static int set_max_cstate(const struct dmi_system_id *id)
  91 {
  92         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  93                 return 0;
  94
  95         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  96                " Override with \"processor.max_cstate=%d\"\n", id->ident,
  97                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  98
  99         max_cstate = (long)id->driver_data;
 100
 101         return 0;
 102 }
 103
 104 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
 105    callers to only run once -AK */
 106 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 107         { set_max_cstate, "Clevo 5600D", {
 108           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 109           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 110          (void *)2},
 111         {},
 112 };
 113
 114
 115 /*
 116  * Callers should disable interrupts before the call and enable
 117  * interrupts after return.
 118  */
 119 static void acpi_safe_halt(void)
 120 {
 121         current_thread_info()->status &= ~TS_POLLING;
 122         /*
 123          * TS_POLLING-cleared state must be visible before we
 124          * test NEED_RESCHED:
 125          */
 126         smp_mb();
 127         if (!need_resched()) {
 128                 safe_halt();
 129                 local_irq_disable();
 130         }
 131         current_thread_info()->status |= TS_POLLING;
 132 }
 133
 134 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 135
 136 /*
 137  * Some BIOS implementations switch to C3 in the published C2 state.
 138  * This seems to be a common problem on AMD boxen, but other vendors
 139  * are affected too. We pick the most conservative approach: we assume
 140  * that the local APIC stops in both C2 and C3.
 141  */
 142 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 143                                    struct acpi_processor_cx *cx)
 144 {
 145         struct acpi_processor_power *pwr = &pr->power;
 146         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 147
 148         /*
 149          * Check, if one of the previous states already marked the lapic
 150          * unstable
 151          */
 152         if (pwr->timer_broadcast_on_state < state)
 153                 return;
 154
 155         if (cx->type >= type)
 156                 pr->power.timer_broadcast_on_state = state;
 157 }
 158
 159 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 160 {
 161         unsigned long reason;
 162
 163         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 164                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 165
 166         clockevents_notify(reason, &pr->id);
 167 }
 168
 169 /* Power(C) State timer broadcast control */
 170 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 171                                        struct acpi_processor_cx *cx,
 172                                        int broadcast)
 173 {
 174         int state = cx - pr->power.states;
 175
 176         if (state >= pr->power.timer_broadcast_on_state) {
 177                 unsigned long reason;
 178
 179                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 180                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 181                 clockevents_notify(reason, &pr->id);
 182         }
 183 }
 184
 185 #else
 186
 187 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 188                                    struct acpi_processor_cx *cstate) { }
 189 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
 190 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 191                                        struct acpi_processor_cx *cx,
 192                                        int broadcast)
 193 {
 194 }
 195
 196 #endif
 197
 198 /*
 199  * Suspend / resume control
 200  */
 201 static int acpi_idle_suspend;
 202
 203 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
 204 {
 205         acpi_idle_suspend = 1;
 206         return 0;
 207 }
 208
 209 int acpi_processor_resume(struct acpi_device * device)
 210 {
 211         acpi_idle_suspend = 0;
 212         return 0;
 213 }
 214
 215 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 216 static int tsc_halts_in_c(int state)
 217 {
 218         switch (boot_cpu_data.x86_vendor) {
 219         case X86_VENDOR_AMD:
 220         case X86_VENDOR_INTEL:
 221                 /*
 222                  * AMD Fam10h TSC will tick in all
 223                  * C/P/S0/S1 states when this bit is set.
 224                  */
 225                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 226                         return 0;
 227
 228                 /*FALL THROUGH*/
 229         default:
 230                 return state > ACPI_STATE_C1;
 231         }
 232 }
 233 #endif
 234
 235 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 236 {
 237
 238         if (!pr)
 239                 return -EINVAL;
 240
 241         if (!pr->pblk)
 242                 return -ENODEV;
 243
 244         /* if info is obtained from pblk/fadt, type equals state */
 245         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 246         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 247
 248 #ifndef CONFIG_HOTPLUG_CPU
 249         /*
 250          * Check for P_LVL2_UP flag before entering C2 and above on
 251          * an SMP system.
 252          */
 253         if ((num_online_cpus() > 1) &&
 254             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 255                 return -ENODEV;
 256 #endif
 257
 258         /* determine C2 and C3 address from pblk */
 259         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 260         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 261
 262         /* determine latencies from FADT */
 263         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 264         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 265
 266         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 267                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 268                           pr->power.states[ACPI_STATE_C2].address,
 269                           pr->power.states[ACPI_STATE_C3].address));
 270
 271         return 0;
 272 }
 273
 274 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 275 {
 276         if (!pr->power.states[ACPI_STATE_C1].valid) {
 277                 /* set the first C-State to C1 */
 278                 /* all processors need to support C1 */
 279                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 280                 pr->power.states[ACPI_STATE_C1].valid = 1;
 281                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 282         }
 283         /* the C0 state only exists as a filler in our array */
 284         pr->power.states[ACPI_STATE_C0].valid = 1;
 285         return 0;
 286 }
 287
 288 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 289 {
 290         acpi_status status = 0;
 291         acpi_integer count;
 292         int current_count;
 293         int i;
 294         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 295         union acpi_object *cst;
 296
 297
 298         if (nocst)
 299                 return -ENODEV;
 300
 301         current_count = 0;
 302
 303         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 304         if (ACPI_FAILURE(status)) {
 305                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 306                 return -ENODEV;
 307         }
 308
 309         cst = buffer.pointer;
 310
 311         /* There must be at least 2 elements */
 312         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 313                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 314                 status = -EFAULT;
 315                 goto end;
 316         }
 317
 318         count = cst->package.elements[0].integer.value;
 319
 320         /* Validate number of power states. */
 321         if (count < 1 || count != cst->package.count - 1) {
 322                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 323                 status = -EFAULT;
 324                 goto end;
 325         }
 326
 327         /* Tell driver that at least _CST is supported. */
 328         pr->flags.has_cst = 1;
 329
 330         for (i = 1; i <= count; i++) {
 331                 union acpi_object *element;
 332                 union acpi_object *obj;
 333                 struct acpi_power_register *reg;
 334                 struct acpi_processor_cx cx;
 335
 336                 memset(&cx, 0, sizeof(cx));
 337
 338                 element = &(cst->package.elements[i]);
 339                 if (element->type != ACPI_TYPE_PACKAGE)
 340                         continue;
 341
 342                 if (element->package.count != 4)
 343                         continue;
 344
 345                 obj = &(element->package.elements[0]);
 346
 347                 if (obj->type != ACPI_TYPE_BUFFER)
 348                         continue;
 349
 350                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 351
 352                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 353                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 354                         continue;
 355
 356                 /* There should be an easy way to extract an integer... */
 357                 obj = &(element->package.elements[1]);
 358                 if (obj->type != ACPI_TYPE_INTEGER)
 359                         continue;
 360
 361                 cx.type = obj->integer.value;
 362                 /*
 363                  * Some buggy BIOSes won't list C1 in _CST -
 364                  * Let acpi_processor_get_power_info_default() handle them later
 365                  */
 366                 if (i == 1 && cx.type != ACPI_STATE_C1)
 367                         current_count++;
 368
 369                 cx.address = reg->address;
 370                 cx.index = current_count + 1;
 371
 372                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 373                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 374                         if (acpi_processor_ffh_cstate_probe
 375                                         (pr->id, &cx, reg) == 0) {
 376                                 cx.entry_method = ACPI_CSTATE_FFH;
 377                         } else if (cx.type == ACPI_STATE_C1) {
 378                                 /*
 379                                  * C1 is a special case where FIXED_HARDWARE
 380                                  * can be handled in non-MWAIT way as well.
 381                                  * In that case, save this _CST entry info.
 382                                  * Otherwise, ignore this info and continue.
 383                                  */
 384                                 cx.entry_method = ACPI_CSTATE_HALT;
 385                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 386                         } else {
 387                                 continue;
 388                         }
 389                         if (cx.type == ACPI_STATE_C1 &&
 390                                         (idle_halt || idle_nomwait)) {
 391                                 /*
 392                                  * In most cases the C1 space_id obtained from
 393                                  * _CST object is FIXED_HARDWARE access mode.
 394                                  * But when the option of idle=halt is added,
 395                                  * the entry_method type should be changed from
 396                                  * CSTATE_FFH to CSTATE_HALT.
 397                                  * When the option of idle=nomwait is added,
 398                                  * the C1 entry_method type should be
 399                                  * CSTATE_HALT.
 400                                  */
 401                                 cx.entry_method = ACPI_CSTATE_HALT;
 402                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 403                         }
 404                 } else {
 405                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 406                                  cx.address);
 407                 }
 408
 409                 if (cx.type == ACPI_STATE_C1) {
 410                         cx.valid = 1;
 411                 }
 412
 413                 obj = &(element->package.elements[2]);
 414                 if (obj->type != ACPI_TYPE_INTEGER)
 415                         continue;
 416
 417                 cx.latency = obj->integer.value;
 418
 419                 obj = &(element->package.elements[3]);
 420                 if (obj->type != ACPI_TYPE_INTEGER)
 421                         continue;
 422
 423                 cx.power = obj->integer.value;
 424
 425                 current_count++;
 426                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 427
 428                 /*
 429                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 430                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 431                  */
 432                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 433                         printk(KERN_WARNING
 434                                "Limiting number of power states to max (%d)\n",
 435                                ACPI_PROCESSOR_MAX_POWER);
 436                         printk(KERN_WARNING
 437                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 438                         break;
 439                 }
 440         }
 441
 442         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 443                           current_count));
 444
 445         /* Validate number of power states discovered */
 446         if (current_count < 2)
 447                 status = -EFAULT;
 448
 449       end:
 450         kfree(buffer.pointer);
 451
 452         return status;
 453 }
 454
 455 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 456 {
 457
 458         if (!cx->address)
 459                 return;
 460
 461         /*
 462          * C2 latency must be less than or equal to 100
 463          * microseconds.
 464          */
 465         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 466                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 467                                   "latency too large [%d]\n", cx->latency));
 468                 return;
 469         }
 470
 471         /*
 472          * Otherwise we've met all of our C2 requirements.
 473          * Normalize the C2 latency to expidite policy
 474          */
 475         cx->valid = 1;
 476
 477         cx->latency_ticks = cx->latency;
 478
 479         return;
 480 }
 481
 482 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 483                                            struct acpi_processor_cx *cx)
 484 {
 485         static int bm_check_flag;
 486
 487
 488         if (!cx->address)
 489                 return;
 490
 491         /*
 492          * C3 latency must be less than or equal to 1000
 493          * microseconds.
 494          */
 495         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 496                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 497                                   "latency too large [%d]\n", cx->latency));
 498                 return;
 499         }
 500
 501         /*
 502          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 503          * DMA transfers are used by any ISA device to avoid livelock.
 504          * Note that we could disable Type-F DMA (as recommended by
 505          * the erratum), but this is known to disrupt certain ISA
 506          * devices thus we take the conservative approach.
 507          */
 508         else if (errata.piix4.fdma) {
 509                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 510                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 511                 return;
 512         }
 513
 514         /* All the logic here assumes flags.bm_check is same across all CPUs */
 515         if (!bm_check_flag) {
 516                 /* Determine whether bm_check is needed based on CPU  */
 517                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 518                 bm_check_flag = pr->flags.bm_check;
 519         } else {
 520                 pr->flags.bm_check = bm_check_flag;
 521         }
 522
 523         if (pr->flags.bm_check) {
 524                 if (!pr->flags.bm_control) {
 525                         if (pr->flags.has_cst != 1) {
 526                                 /* bus mastering control is necessary */
 527                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 528                                         "C3 support requires BM control\n"));
 529                                 return;
 530                         } else {
 531                                 /* Here we enter C3 without bus mastering */
 532                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 533                                         "C3 support without BM control\n"));
 534                         }
 535                 }
 536         } else {
 537                 /*
 538                  * WBINVD should be set in fadt, for C3 state to be
 539                  * supported on when bm_check is not required.
 540                  */
 541                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 542                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 543                                           "Cache invalidation should work properly"
 544                                           " for C3 to be enabled on SMP systems\n"));
 545                         return;
 546                 }
 547         }
 548
 549         /*
 550          * Otherwise we've met all of our C3 requirements.
 551          * Normalize the C3 latency to expidite policy.  Enable
 552          * checking of bus mastering status (bm_check) so we can
 553          * use this in our C3 policy
 554          */
 555         cx->valid = 1;
 556
 557         cx->latency_ticks = cx->latency;
 558         /*
 559          * On older chipsets, BM_RLD needs to be set
 560          * in order for Bus Master activity to wake the
 561          * system from C3.  Newer chipsets handle DMA
 562          * during C3 automatically and BM_RLD is a NOP.
 563          * In either case, the proper way to
 564          * handle BM_RLD is to set it and leave it set.
 565          */
 566         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 567
 568         return;
 569 }
 570
 571 static int acpi_processor_power_verify(struct acpi_processor *pr)
 572 {
 573         unsigned int i;
 574         unsigned int working = 0;
 575
 576         pr->power.timer_broadcast_on_state = INT_MAX;
 577
 578         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 579                 struct acpi_processor_cx *cx = &pr->power.states[i];
 580
 581                 switch (cx->type) {
 582                 case ACPI_STATE_C1:
 583                         cx->valid = 1;
 584                         break;
 585
 586                 case ACPI_STATE_C2:
 587                         acpi_processor_power_verify_c2(cx);
 588                         if (cx->valid)
 589                                 acpi_timer_check_state(i, pr, cx);
 590                         break;
 591
 592                 case ACPI_STATE_C3:
 593                         acpi_processor_power_verify_c3(pr, cx);
 594                         if (cx->valid)
 595                                 acpi_timer_check_state(i, pr, cx);
 596                         break;
 597                 }
 598
 599                 if (cx->valid)
 600                         working++;
 601         }
 602
 603         acpi_propagate_timer_broadcast(pr);
 604
 605         return (working);
 606 }
 607
 608 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 609 {
 610         unsigned int i;
 611         int result;
 612
 613
 614         /* NOTE: the idle thread may not be running while calling
 615          * this function */
 616
 617         /* Zero initialize all the C-states info. */
 618         memset(pr->power.states, 0, sizeof(pr->power.states));
 619
 620         result = acpi_processor_get_power_info_cst(pr);
 621         if (result == -ENODEV)
 622                 result = acpi_processor_get_power_info_fadt(pr);
 623
 624         if (result)
 625                 return result;
 626
 627         acpi_processor_get_power_info_default(pr);
 628
 629         pr->power.count = acpi_processor_power_verify(pr);
 630
 631         /*
 632          * if one state of type C2 or C3 is available, mark this
 633          * CPU as being "idle manageable"
 634          */
 635         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 636                 if (pr->power.states[i].valid) {
 637                         pr->power.count = i;
 638                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 639                                 pr->flags.power = 1;
 640                 }
 641         }
 642
 643         return 0;
 644 }
 645
 646 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 647 {
 648         struct acpi_processor *pr = seq->private;
 649         unsigned int i;
 650
 651
 652         if (!pr)
 653                 goto end;
 654
 655         seq_printf(seq, "active state:            C%zd\n"
 656                    "max_cstate:              C%d\n"
 657                    "bus master activity:     %08x\n"
 658                    "maximum allowed latency: %d usec\n",
 659                    pr->power.state ? pr->power.state - pr->power.states : 0,
 660                    max_cstate, (unsigned)pr->power.bm_activity,
 661                    pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
 662
 663         seq_puts(seq, "states:\n");
 664
 665         for (i = 1; i <= pr->power.count; i++) {
 666                 seq_printf(seq, "   %cC%d:                  ",
 667                            (&pr->power.states[i] ==
 668                             pr->power.state ? '*' : ' '), i);
 669
 670                 if (!pr->power.states[i].valid) {
 671                         seq_puts(seq, "<not supported>\n");
 672                         continue;
 673                 }
 674
 675                 switch (pr->power.states[i].type) {
 676                 case ACPI_STATE_C1:
 677                         seq_printf(seq, "type[C1] ");
 678                         break;
 679                 case ACPI_STATE_C2:
 680                         seq_printf(seq, "type[C2] ");
 681                         break;
 682                 case ACPI_STATE_C3:
 683                         seq_printf(seq, "type[C3] ");
 684                         break;
 685                 default:
 686                         seq_printf(seq, "type[--] ");
 687                         break;
 688                 }
 689
 690                 if (pr->power.states[i].promotion.state)
 691                         seq_printf(seq, "promotion[C%zd] ",
 692                                    (pr->power.states[i].promotion.state -
 693                                     pr->power.states));
 694                 else
 695                         seq_puts(seq, "promotion[--] ");
 696
 697                 if (pr->power.states[i].demotion.state)
 698                         seq_printf(seq, "demotion[C%zd] ",
 699                                    (pr->power.states[i].demotion.state -
 700                                     pr->power.states));
 701                 else
 702                         seq_puts(seq, "demotion[--] ");
 703
 704                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
 705                            pr->power.states[i].latency,
 706                            pr->power.states[i].usage,
 707                            (unsigned long long)pr->power.states[i].time);
 708         }
 709
 710       end:
 711         return 0;
 712 }
 713
 714 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
 715 {
 716         return single_open(file, acpi_processor_power_seq_show,
 717                            PDE(inode)->data);
 718 }
 719
 720 static const struct file_operations acpi_processor_power_fops = {
 721         .owner = THIS_MODULE,
 722         .open = acpi_processor_power_open_fs,
 723         .read = seq_read,
 724         .llseek = seq_lseek,
 725         .release = single_release,
 726 };
 727
 728
 729 /**
 730  * acpi_idle_bm_check - checks if bus master activity was detected
 731  */
 732 static int acpi_idle_bm_check(void)
 733 {
 734         u32 bm_status = 0;
 735
 736         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 737         if (bm_status)
 738                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 739         /*
 740          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 741          * the true state of bus mastering activity; forcing us to
 742          * manually check the BMIDEA bit of each IDE channel.
 743          */
 744         else if (errata.piix4.bmisx) {
 745                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 746                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 747                         bm_status = 1;
 748         }
 749         return bm_status;
 750 }
 751
 752 /**
 753  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
 754  * @cx: cstate data
 755  *
 756  * Caller disables interrupt before call and enables interrupt after return.
 757  */
 758 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
 759 {
 760         /* Don't trace irqs off for idle */
 761         stop_critical_timings();
 762         if (cx->entry_method == ACPI_CSTATE_FFH) {
 763                 /* Call into architectural FFH based C-state */
 764                 acpi_processor_ffh_cstate_enter(cx);
 765         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 766                 acpi_safe_halt();
 767         } else {
 768                 int unused;
 769                 /* IO port based C-state */
 770                 inb(cx->address);
 771                 /* Dummy wait op - must do something useless after P_LVL2 read
 772                    because chipsets cannot guarantee that STPCLK# signal
 773                    gets asserted in time to freeze execution properly. */
 774                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 775         }
 776         start_critical_timings();
 777 }
 778
 779 /**
 780  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
 781  * @dev: the target CPU
 782  * @state: the state data
 783  *
 784  * This is equivalent to the HALT instruction.
 785  */
 786 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
 787                               struct cpuidle_state *state)
 788 {
 789         ktime_t  kt1, kt2;
 790         s64 idle_time;
 791         struct acpi_processor *pr;
 792         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 793
 794         pr = __get_cpu_var(processors);
 795
 796         if (unlikely(!pr))
 797                 return 0;
 798
 799         local_irq_disable();
 800
 801         /* Do not access any ACPI IO ports in suspend path */
 802         if (acpi_idle_suspend) {
 803                 acpi_safe_halt();
 804                 local_irq_enable();
 805                 return 0;
 806         }
 807
 808         kt1 = ktime_get_real();
 809         acpi_idle_do_entry(cx);
 810         kt2 = ktime_get_real();
 811         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 812
 813         local_irq_enable();
 814         cx->usage++;
 815
 816         return idle_time;
 817 }
 818
 819 /**
 820  * acpi_idle_enter_simple - enters an ACPI state without BM handling
 821  * @dev: the target CPU
 822  * @state: the state data
 823  */
 824 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 825                                   struct cpuidle_state *state)
 826 {
 827         struct acpi_processor *pr;
 828         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 829         ktime_t  kt1, kt2;
 830         s64 idle_time;
 831         s64 sleep_ticks = 0;
 832
 833         pr = __get_cpu_var(processors);
 834
 835         if (unlikely(!pr))
 836                 return 0;
 837
 838         if (acpi_idle_suspend)
 839                 return(acpi_idle_enter_c1(dev, state));
 840
 841         local_irq_disable();
 842         current_thread_info()->status &= ~TS_POLLING;
 843         /*
 844          * TS_POLLING-cleared state must be visible before we test
 845          * NEED_RESCHED:
 846          */
 847         smp_mb();
 848
 849         if (unlikely(need_resched())) {
 850                 current_thread_info()->status |= TS_POLLING;
 851                 local_irq_enable();
 852                 return 0;
 853         }
 854
 855         /*
 856          * Must be done before busmaster disable as we might need to
 857          * access HPET !
 858          */
 859         acpi_state_timer_broadcast(pr, cx, 1);
 860
 861         if (cx->type == ACPI_STATE_C3)
 862                 ACPI_FLUSH_CPU_CACHE();
 863
 864         kt1 = ktime_get_real();
 865         /* Tell the scheduler that we are going deep-idle: */
 866         sched_clock_idle_sleep_event();
 867         acpi_idle_do_entry(cx);
 868         kt2 = ktime_get_real();
 869         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 870
 871 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 872         /* TSC could halt in idle, so notify users */
 873         if (tsc_halts_in_c(cx->type))
 874                 mark_tsc_unstable("TSC halts in idle");;
 875 #endif
 876         sleep_ticks = us_to_pm_timer_ticks(idle_time);
 877
 878         /* Tell the scheduler how much we idled: */
 879         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 880
 881         local_irq_enable();
 882         current_thread_info()->status |= TS_POLLING;
 883
 884         cx->usage++;
 885
 886         acpi_state_timer_broadcast(pr, cx, 0);
 887         cx->time += sleep_ticks;
 888         return idle_time;
 889 }
 890
 891 static int c3_cpu_count;
 892 static DEFINE_SPINLOCK(c3_lock);
 893
 894 /**
 895  * acpi_idle_enter_bm - enters C3 with proper BM handling
 896  * @dev: the target CPU
 897  * @state: the state data
 898  *
 899  * If BM is detected, the deepest non-C3 idle state is entered instead.
 900  */
 901 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 902                               struct cpuidle_state *state)
 903 {
 904         struct acpi_processor *pr;
 905         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 906         ktime_t  kt1, kt2;
 907         s64 idle_time;
 908         s64 sleep_ticks = 0;
 909
 910
 911         pr = __get_cpu_var(processors);
 912
 913         if (unlikely(!pr))
 914                 return 0;
 915
 916         if (acpi_idle_suspend)
 917                 return(acpi_idle_enter_c1(dev, state));
 918
 919         if (acpi_idle_bm_check()) {
 920                 if (dev->safe_state) {
 921                         dev->last_state = dev->safe_state;
 922                         return dev->safe_state->enter(dev, dev->safe_state);
 923                 } else {
 924                         local_irq_disable();
 925                         acpi_safe_halt();
 926                         local_irq_enable();
 927                         return 0;
 928                 }
 929         }
 930
 931         local_irq_disable();
 932         current_thread_info()->status &= ~TS_POLLING;
 933         /*
 934          * TS_POLLING-cleared state must be visible before we test
 935          * NEED_RESCHED:
 936          */
 937         smp_mb();
 938
 939         if (unlikely(need_resched())) {
 940                 current_thread_info()->status |= TS_POLLING;
 941                 local_irq_enable();
 942                 return 0;
 943         }
 944
 945         acpi_unlazy_tlb(smp_processor_id());
 946
 947         /* Tell the scheduler that we are going deep-idle: */
 948         sched_clock_idle_sleep_event();
 949         /*
 950          * Must be done before busmaster disable as we might need to
 951          * access HPET !
 952          */
 953         acpi_state_timer_broadcast(pr, cx, 1);
 954
 955         /*
 956          * disable bus master
 957          * bm_check implies we need ARB_DIS
 958          * !bm_check implies we need cache flush
 959          * bm_control implies whether we can do ARB_DIS
 960          *
 961          * That leaves a case where bm_check is set and bm_control is
 962          * not set. In that case we cannot do much, we enter C3
 963          * without doing anything.
 964          */
 965         if (pr->flags.bm_check && pr->flags.bm_control) {
 966                 spin_lock(&c3_lock);
 967                 c3_cpu_count++;
 968                 /* Disable bus master arbitration when all CPUs are in C3 */
 969                 if (c3_cpu_count == num_online_cpus())
 970                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 971                 spin_unlock(&c3_lock);
 972         } else if (!pr->flags.bm_check) {
 973                 ACPI_FLUSH_CPU_CACHE();
 974         }
 975
 976         kt1 = ktime_get_real();
 977         acpi_idle_do_entry(cx);
 978         kt2 = ktime_get_real();
 979         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 980
 981         /* Re-enable bus master arbitration */
 982         if (pr->flags.bm_check && pr->flags.bm_control) {
 983                 spin_lock(&c3_lock);
 984                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 985                 c3_cpu_count--;
 986                 spin_unlock(&c3_lock);
 987         }
 988
 989 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 990         /* TSC could halt in idle, so notify users */
 991         if (tsc_halts_in_c(ACPI_STATE_C3))
 992                 mark_tsc_unstable("TSC halts in idle");
 993 #endif
 994         sleep_ticks = us_to_pm_timer_ticks(idle_time);
 995         /* Tell the scheduler how much we idled: */
 996         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 997
 998         local_irq_enable();
 999         current_thread_info()->status |= TS_POLLING;
1000
1001         cx->usage++;
1002
1003         acpi_state_timer_broadcast(pr, cx, 0);
1004         cx->time += sleep_ticks;
1005         return idle_time;
1006 }
1007
1008 struct cpuidle_driver acpi_idle_driver = {
1009         .name =         "acpi_idle",
1010         .owner =        THIS_MODULE,
1011 };
1012
1013 /**
1014  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1015  * @pr: the ACPI processor
1016  */
1017 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1018 {
1019         int i, count = CPUIDLE_DRIVER_STATE_START;
1020         struct acpi_processor_cx *cx;
1021         struct cpuidle_state *state;
1022         struct cpuidle_device *dev = &pr->power.dev;
1023
1024         if (!pr->flags.power_setup_done)
1025                 return -EINVAL;
1026
1027         if (pr->flags.power == 0) {
1028                 return -EINVAL;
1029         }
1030
1031         dev->cpu = pr->id;
1032         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1033                 dev->states[i].name[0] = '\0';
1034                 dev->states[i].desc[0] = '\0';
1035         }
1036
1037         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1038                 cx = &pr->power.states[i];
1039                 state = &dev->states[count];
1040
1041                 if (!cx->valid)
1042                         continue;
1043
1044 #ifdef CONFIG_HOTPLUG_CPU
1045                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1046                     !pr->flags.has_cst &&
1047                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1048                         continue;
1049 #endif
1050                 cpuidle_set_statedata(state, cx);
1051
1052                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1053                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1054                 state->exit_latency = cx->latency;
1055                 state->target_residency = cx->latency * latency_factor;
1056                 state->power_usage = cx->power;
1057
1058                 state->flags = 0;
1059                 switch (cx->type) {
1060                         case ACPI_STATE_C1:
1061                         state->flags |= CPUIDLE_FLAG_SHALLOW;
1062                         if (cx->entry_method == ACPI_CSTATE_FFH)
1063                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1064
1065                         state->enter = acpi_idle_enter_c1;
1066                         dev->safe_state = state;
1067                         break;
1068
1069                         case ACPI_STATE_C2:
1070                         state->flags |= CPUIDLE_FLAG_BALANCED;
1071                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1072                         state->enter = acpi_idle_enter_simple;
1073                         dev->safe_state = state;
1074                         break;
1075
1076                         case ACPI_STATE_C3:
1077                         state->flags |= CPUIDLE_FLAG_DEEP;
1078                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1079                         state->flags |= CPUIDLE_FLAG_CHECK_BM;
1080                         state->enter = pr->flags.bm_check ?
1081                                         acpi_idle_enter_bm :
1082                                         acpi_idle_enter_simple;
1083                         break;
1084                 }
1085
1086                 count++;
1087                 if (count == CPUIDLE_STATE_MAX)
1088                         break;
1089         }
1090
1091         dev->state_count = count;
1092
1093         if (!count)
1094                 return -EINVAL;
1095
1096         return 0;
1097 }
1098
1099 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1100 {
1101         int ret = 0;
1102
1103         if (boot_option_idle_override)
1104                 return 0;
1105
1106         if (!pr)
1107                 return -EINVAL;
1108
1109         if (nocst) {
1110                 return -ENODEV;
1111         }
1112
1113         if (!pr->flags.power_setup_done)
1114                 return -ENODEV;
1115
1116         cpuidle_pause_and_lock();
1117         cpuidle_disable_device(&pr->power.dev);
1118         acpi_processor_get_power_info(pr);
1119         if (pr->flags.power) {
1120                 acpi_processor_setup_cpuidle(pr);
1121                 ret = cpuidle_enable_device(&pr->power.dev);
1122         }
1123         cpuidle_resume_and_unlock();
1124
1125         return ret;
1126 }
1127
1128 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1129                               struct acpi_device *device)
1130 {
1131         acpi_status status = 0;
1132         static int first_run;
1133         struct proc_dir_entry *entry = NULL;
1134         unsigned int i;
1135
1136         if (boot_option_idle_override)
1137                 return 0;
1138
1139         if (!first_run) {
1140                 if (idle_halt) {
1141                         /*
1142                          * When the boot option of "idle=halt" is added, halt
1143                          * is used for CPU IDLE.
1144                          * In such case C2/C3 is meaningless. So the max_cstate
1145                          * is set to one.
1146                          */
1147                         max_cstate = 1;
1148                 }
1149                 dmi_check_system(processor_power_dmi_table);
1150                 max_cstate = acpi_processor_cstate_check(max_cstate);
1151                 if (max_cstate < ACPI_C_STATES_MAX)
1152                         printk(KERN_NOTICE
1153                                "ACPI: processor limited to max C-state %d\n",
1154                                max_cstate);
1155                 first_run++;
1156         }
1157
1158         if (!pr)
1159                 return -EINVAL;
1160
1161         if (acpi_gbl_FADT.cst_control && !nocst) {
1162                 status =
1163                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1164                 if (ACPI_FAILURE(status)) {
1165                         ACPI_EXCEPTION((AE_INFO, status,
1166                                         "Notifying BIOS of _CST ability failed"));
1167                 }
1168         }
1169
1170         acpi_processor_get_power_info(pr);
1171         pr->flags.power_setup_done = 1;
1172
1173         /*
1174          * Install the idle handler if processor power management is supported.
1175          * Note that we use previously set idle handler will be used on
1176          * platforms that only support C1.
1177          */
1178         if (pr->flags.power) {
1179                 acpi_processor_setup_cpuidle(pr);
1180                 if (cpuidle_register_device(&pr->power.dev))
1181                         return -EIO;
1182
1183                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1184                 for (i = 1; i <= pr->power.count; i++)
1185                         if (pr->power.states[i].valid)
1186                                 printk(" C%d[C%d]", i,
1187                                        pr->power.states[i].type);
1188                 printk(")\n");
1189         }
1190
1191         /* 'power' [R] */
1192         entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1193                                  S_IRUGO, acpi_device_dir(device),
1194                                  &acpi_processor_power_fops,
1195                                  acpi_driver_data(device));
1196         if (!entry)
1197                 return -EIO;
1198         return 0;
1199 }
1200
1201 int acpi_processor_power_exit(struct acpi_processor *pr,
1202                               struct acpi_device *device)
1203 {
1204         if (boot_option_idle_override)
1205                 return 0;
1206
1207         cpuidle_unregister_device(&pr->power.dev);
1208         pr->flags.power_setup_done = 0;
1209
1210         if (acpi_device_dir(device))
1211                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1212                                   acpi_device_dir(device));
1213
1214         return 0;
1215 }