SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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         if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 149                 return;
 150
 151         /*
 152          * Check, if one of the previous states already marked the lapic
 153          * unstable
 154          */
 155         if (pwr->timer_broadcast_on_state < state)
 156                 return;
 157
 158         if (cx->type >= type)
 159                 pr->power.timer_broadcast_on_state = state;
 160 }
 161
 162 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 163 {
 164         unsigned long reason;
 165
 166         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 167                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 168
 169         clockevents_notify(reason, &pr->id);
 170 }
 171
 172 /* Power(C) State timer broadcast control */
 173 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 174                                        struct acpi_processor_cx *cx,
 175                                        int broadcast)
 176 {
 177         int state = cx - pr->power.states;
 178
 179         if (state >= pr->power.timer_broadcast_on_state) {
 180                 unsigned long reason;
 181
 182                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 183                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 184                 clockevents_notify(reason, &pr->id);
 185         }
 186 }
 187
 188 #else
 189
 190 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 191                                    struct acpi_processor_cx *cstate) { }
 192 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
 193 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 194                                        struct acpi_processor_cx *cx,
 195                                        int broadcast)
 196 {
 197 }
 198
 199 #endif
 200
 201 /*
 202  * Suspend / resume control
 203  */
 204 static int acpi_idle_suspend;
 205
 206 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
 207 {
 208         acpi_idle_suspend = 1;
 209         return 0;
 210 }
 211
 212 int acpi_processor_resume(struct acpi_device * device)
 213 {
 214         acpi_idle_suspend = 0;
 215         return 0;
 216 }
 217
 218 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 219 static void tsc_check_state(int state)
 220 {
 221         switch (boot_cpu_data.x86_vendor) {
 222         case X86_VENDOR_AMD:
 223         case X86_VENDOR_INTEL:
 224                 /*
 225                  * AMD Fam10h TSC will tick in all
 226                  * C/P/S0/S1 states when this bit is set.
 227                  */
 228                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 229                         return;
 230
 231                 /*FALL THROUGH*/
 232         default:
 233                 /* TSC could halt in idle, so notify users */
 234                 if (state > ACPI_STATE_C1)
 235                         mark_tsc_unstable("TSC halts in idle");
 236         }
 237 }
 238 #else
 239 static void tsc_check_state(int state) { return; }
 240 #endif
 241
 242 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 243 {
 244
 245         if (!pr)
 246                 return -EINVAL;
 247
 248         if (!pr->pblk)
 249                 return -ENODEV;
 250
 251         /* if info is obtained from pblk/fadt, type equals state */
 252         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 253         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 254
 255 #ifndef CONFIG_HOTPLUG_CPU
 256         /*
 257          * Check for P_LVL2_UP flag before entering C2 and above on
 258          * an SMP system.
 259          */
 260         if ((num_online_cpus() > 1) &&
 261             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 262                 return -ENODEV;
 263 #endif
 264
 265         /* determine C2 and C3 address from pblk */
 266         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 267         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 268
 269         /* determine latencies from FADT */
 270         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 271         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 272
 273         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 274                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 275                           pr->power.states[ACPI_STATE_C2].address,
 276                           pr->power.states[ACPI_STATE_C3].address));
 277
 278         return 0;
 279 }
 280
 281 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 282 {
 283         if (!pr->power.states[ACPI_STATE_C1].valid) {
 284                 /* set the first C-State to C1 */
 285                 /* all processors need to support C1 */
 286                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 287                 pr->power.states[ACPI_STATE_C1].valid = 1;
 288                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 289         }
 290         /* the C0 state only exists as a filler in our array */
 291         pr->power.states[ACPI_STATE_C0].valid = 1;
 292         return 0;
 293 }
 294
 295 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 296 {
 297         acpi_status status = 0;
 298         acpi_integer count;
 299         int current_count;
 300         int i;
 301         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 302         union acpi_object *cst;
 303
 304
 305         if (nocst)
 306                 return -ENODEV;
 307
 308         current_count = 0;
 309
 310         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 311         if (ACPI_FAILURE(status)) {
 312                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 313                 return -ENODEV;
 314         }
 315
 316         cst = buffer.pointer;
 317
 318         /* There must be at least 2 elements */
 319         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 320                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 321                 status = -EFAULT;
 322                 goto end;
 323         }
 324
 325         count = cst->package.elements[0].integer.value;
 326
 327         /* Validate number of power states. */
 328         if (count < 1 || count != cst->package.count - 1) {
 329                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 330                 status = -EFAULT;
 331                 goto end;
 332         }
 333
 334         /* Tell driver that at least _CST is supported. */
 335         pr->flags.has_cst = 1;
 336
 337         for (i = 1; i <= count; i++) {
 338                 union acpi_object *element;
 339                 union acpi_object *obj;
 340                 struct acpi_power_register *reg;
 341                 struct acpi_processor_cx cx;
 342
 343                 memset(&cx, 0, sizeof(cx));
 344
 345                 element = &(cst->package.elements[i]);
 346                 if (element->type != ACPI_TYPE_PACKAGE)
 347                         continue;
 348
 349                 if (element->package.count != 4)
 350                         continue;
 351
 352                 obj = &(element->package.elements[0]);
 353
 354                 if (obj->type != ACPI_TYPE_BUFFER)
 355                         continue;
 356
 357                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 358
 359                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 360                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 361                         continue;
 362
 363                 /* There should be an easy way to extract an integer... */
 364                 obj = &(element->package.elements[1]);
 365                 if (obj->type != ACPI_TYPE_INTEGER)
 366                         continue;
 367
 368                 cx.type = obj->integer.value;
 369                 /*
 370                  * Some buggy BIOSes won't list C1 in _CST -
 371                  * Let acpi_processor_get_power_info_default() handle them later
 372                  */
 373                 if (i == 1 && cx.type != ACPI_STATE_C1)
 374                         current_count++;
 375
 376                 cx.address = reg->address;
 377                 cx.index = current_count + 1;
 378
 379                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 380                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 381                         if (acpi_processor_ffh_cstate_probe
 382                                         (pr->id, &cx, reg) == 0) {
 383                                 cx.entry_method = ACPI_CSTATE_FFH;
 384                         } else if (cx.type == ACPI_STATE_C1) {
 385                                 /*
 386                                  * C1 is a special case where FIXED_HARDWARE
 387                                  * can be handled in non-MWAIT way as well.
 388                                  * In that case, save this _CST entry info.
 389                                  * Otherwise, ignore this info and continue.
 390                                  */
 391                                 cx.entry_method = ACPI_CSTATE_HALT;
 392                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 393                         } else {
 394                                 continue;
 395                         }
 396                         if (cx.type == ACPI_STATE_C1 &&
 397                                         (idle_halt || idle_nomwait)) {
 398                                 /*
 399                                  * In most cases the C1 space_id obtained from
 400                                  * _CST object is FIXED_HARDWARE access mode.
 401                                  * But when the option of idle=halt is added,
 402                                  * the entry_method type should be changed from
 403                                  * CSTATE_FFH to CSTATE_HALT.
 404                                  * When the option of idle=nomwait is added,
 405                                  * the C1 entry_method type should be
 406                                  * CSTATE_HALT.
 407                                  */
 408                                 cx.entry_method = ACPI_CSTATE_HALT;
 409                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 410                         }
 411                 } else {
 412                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 413                                  cx.address);
 414                 }
 415
 416                 if (cx.type == ACPI_STATE_C1) {
 417                         cx.valid = 1;
 418                 }
 419
 420                 obj = &(element->package.elements[2]);
 421                 if (obj->type != ACPI_TYPE_INTEGER)
 422                         continue;
 423
 424                 cx.latency = obj->integer.value;
 425
 426                 obj = &(element->package.elements[3]);
 427                 if (obj->type != ACPI_TYPE_INTEGER)
 428                         continue;
 429
 430                 cx.power = obj->integer.value;
 431
 432                 current_count++;
 433                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 434
 435                 /*
 436                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 437                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 438                  */
 439                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 440                         printk(KERN_WARNING
 441                                "Limiting number of power states to max (%d)\n",
 442                                ACPI_PROCESSOR_MAX_POWER);
 443                         printk(KERN_WARNING
 444                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 445                         break;
 446                 }
 447         }
 448
 449         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 450                           current_count));
 451
 452         /* Validate number of power states discovered */
 453         if (current_count < 2)
 454                 status = -EFAULT;
 455
 456       end:
 457         kfree(buffer.pointer);
 458
 459         return status;
 460 }
 461
 462 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 463 {
 464
 465         if (!cx->address)
 466                 return;
 467
 468         /*
 469          * C2 latency must be less than or equal to 100
 470          * microseconds.
 471          */
 472         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 473                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 474                                   "latency too large [%d]\n", cx->latency));
 475                 return;
 476         }
 477
 478         /*
 479          * Otherwise we've met all of our C2 requirements.
 480          * Normalize the C2 latency to expidite policy
 481          */
 482         cx->valid = 1;
 483
 484         cx->latency_ticks = cx->latency;
 485
 486         return;
 487 }
 488
 489 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 490                                            struct acpi_processor_cx *cx)
 491 {
 492         static int bm_check_flag;
 493
 494
 495         if (!cx->address)
 496                 return;
 497
 498         /*
 499          * C3 latency must be less than or equal to 1000
 500          * microseconds.
 501          */
 502         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 503                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 504                                   "latency too large [%d]\n", cx->latency));
 505                 return;
 506         }
 507
 508         /*
 509          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 510          * DMA transfers are used by any ISA device to avoid livelock.
 511          * Note that we could disable Type-F DMA (as recommended by
 512          * the erratum), but this is known to disrupt certain ISA
 513          * devices thus we take the conservative approach.
 514          */
 515         else if (errata.piix4.fdma) {
 516                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 517                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 518                 return;
 519         }
 520
 521         /* All the logic here assumes flags.bm_check is same across all CPUs */
 522         if (!bm_check_flag) {
 523                 /* Determine whether bm_check is needed based on CPU  */
 524                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 525                 bm_check_flag = pr->flags.bm_check;
 526         } else {
 527                 pr->flags.bm_check = bm_check_flag;
 528         }
 529
 530         if (pr->flags.bm_check) {
 531                 if (!pr->flags.bm_control) {
 532                         if (pr->flags.has_cst != 1) {
 533                                 /* bus mastering control is necessary */
 534                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 535                                         "C3 support requires BM control\n"));
 536                                 return;
 537                         } else {
 538                                 /* Here we enter C3 without bus mastering */
 539                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 540                                         "C3 support without BM control\n"));
 541                         }
 542                 }
 543         } else {
 544                 /*
 545                  * WBINVD should be set in fadt, for C3 state to be
 546                  * supported on when bm_check is not required.
 547                  */
 548                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 549                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 550                                           "Cache invalidation should work properly"
 551                                           " for C3 to be enabled on SMP systems\n"));
 552                         return;
 553                 }
 554         }
 555
 556         /*
 557          * Otherwise we've met all of our C3 requirements.
 558          * Normalize the C3 latency to expidite policy.  Enable
 559          * checking of bus mastering status (bm_check) so we can
 560          * use this in our C3 policy
 561          */
 562         cx->valid = 1;
 563
 564         cx->latency_ticks = cx->latency;
 565         /*
 566          * On older chipsets, BM_RLD needs to be set
 567          * in order for Bus Master activity to wake the
 568          * system from C3.  Newer chipsets handle DMA
 569          * during C3 automatically and BM_RLD is a NOP.
 570          * In either case, the proper way to
 571          * handle BM_RLD is to set it and leave it set.
 572          */
 573         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 574
 575         return;
 576 }
 577
 578 static int acpi_processor_power_verify(struct acpi_processor *pr)
 579 {
 580         unsigned int i;
 581         unsigned int working = 0;
 582
 583         pr->power.timer_broadcast_on_state = INT_MAX;
 584
 585         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 586                 struct acpi_processor_cx *cx = &pr->power.states[i];
 587
 588                 switch (cx->type) {
 589                 case ACPI_STATE_C1:
 590                         cx->valid = 1;
 591                         break;
 592
 593                 case ACPI_STATE_C2:
 594                         acpi_processor_power_verify_c2(cx);
 595                         if (cx->valid)
 596                                 acpi_timer_check_state(i, pr, cx);
 597                         break;
 598
 599                 case ACPI_STATE_C3:
 600                         acpi_processor_power_verify_c3(pr, cx);
 601                         if (cx->valid)
 602                                 acpi_timer_check_state(i, pr, cx);
 603                         break;
 604                 }
 605                 if (cx->valid)
 606                         tsc_check_state(cx->type);
 607
 608                 if (cx->valid)
 609                         working++;
 610         }
 611
 612         acpi_propagate_timer_broadcast(pr);
 613
 614         return (working);
 615 }
 616
 617 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 618 {
 619         unsigned int i;
 620         int result;
 621
 622
 623         /* NOTE: the idle thread may not be running while calling
 624          * this function */
 625
 626         /* Zero initialize all the C-states info. */
 627         memset(pr->power.states, 0, sizeof(pr->power.states));
 628
 629         result = acpi_processor_get_power_info_cst(pr);
 630         if (result == -ENODEV)
 631                 result = acpi_processor_get_power_info_fadt(pr);
 632
 633         if (result)
 634                 return result;
 635
 636         acpi_processor_get_power_info_default(pr);
 637
 638         pr->power.count = acpi_processor_power_verify(pr);
 639
 640         /*
 641          * if one state of type C2 or C3 is available, mark this
 642          * CPU as being "idle manageable"
 643          */
 644         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 645                 if (pr->power.states[i].valid) {
 646                         pr->power.count = i;
 647                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 648                                 pr->flags.power = 1;
 649                 }
 650         }
 651
 652         return 0;
 653 }
 654
 655 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 656 {
 657         struct acpi_processor *pr = seq->private;
 658         unsigned int i;
 659
 660
 661         if (!pr)
 662                 goto end;
 663
 664         seq_printf(seq, "active state:            C%zd\n"
 665                    "max_cstate:              C%d\n"
 666                    "maximum allowed latency: %d usec\n",
 667                    pr->power.state ? pr->power.state - pr->power.states : 0,
 668                    max_cstate, pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
 669
 670         seq_puts(seq, "states:\n");
 671
 672         for (i = 1; i <= pr->power.count; i++) {
 673                 seq_printf(seq, "   %cC%d:                  ",
 674                            (&pr->power.states[i] ==
 675                             pr->power.state ? '*' : ' '), i);
 676
 677                 if (!pr->power.states[i].valid) {
 678                         seq_puts(seq, "<not supported>\n");
 679                         continue;
 680                 }
 681
 682                 switch (pr->power.states[i].type) {
 683                 case ACPI_STATE_C1:
 684                         seq_printf(seq, "type[C1] ");
 685                         break;
 686                 case ACPI_STATE_C2:
 687                         seq_printf(seq, "type[C2] ");
 688                         break;
 689                 case ACPI_STATE_C3:
 690                         seq_printf(seq, "type[C3] ");
 691                         break;
 692                 default:
 693                         seq_printf(seq, "type[--] ");
 694                         break;
 695                 }
 696
 697                 if (pr->power.states[i].promotion.state)
 698                         seq_printf(seq, "promotion[C%zd] ",
 699                                    (pr->power.states[i].promotion.state -
 700                                     pr->power.states));
 701                 else
 702                         seq_puts(seq, "promotion[--] ");
 703
 704                 if (pr->power.states[i].demotion.state)
 705                         seq_printf(seq, "demotion[C%zd] ",
 706                                    (pr->power.states[i].demotion.state -
 707                                     pr->power.states));
 708                 else
 709                         seq_puts(seq, "demotion[--] ");
 710
 711                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
 712                            pr->power.states[i].latency,
 713                            pr->power.states[i].usage,
 714                            (unsigned long long)pr->power.states[i].time);
 715         }
 716
 717       end:
 718         return 0;
 719 }
 720
 721 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
 722 {
 723         return single_open(file, acpi_processor_power_seq_show,
 724                            PDE(inode)->data);
 725 }
 726
 727 static const struct file_operations acpi_processor_power_fops = {
 728         .owner = THIS_MODULE,
 729         .open = acpi_processor_power_open_fs,
 730         .read = seq_read,
 731         .llseek = seq_lseek,
 732         .release = single_release,
 733 };
 734
 735
 736 /**
 737  * acpi_idle_bm_check - checks if bus master activity was detected
 738  */
 739 static int acpi_idle_bm_check(void)
 740 {
 741         u32 bm_status = 0;
 742
 743         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 744         if (bm_status)
 745                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 746         /*
 747          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 748          * the true state of bus mastering activity; forcing us to
 749          * manually check the BMIDEA bit of each IDE channel.
 750          */
 751         else if (errata.piix4.bmisx) {
 752                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 753                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 754                         bm_status = 1;
 755         }
 756         return bm_status;
 757 }
 758
 759 /**
 760  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
 761  * @cx: cstate data
 762  *
 763  * Caller disables interrupt before call and enables interrupt after return.
 764  */
 765 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
 766 {
 767         /* Don't trace irqs off for idle */
 768         stop_critical_timings();
 769         if (cx->entry_method == ACPI_CSTATE_FFH) {
 770                 /* Call into architectural FFH based C-state */
 771                 acpi_processor_ffh_cstate_enter(cx);
 772         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 773                 acpi_safe_halt();
 774         } else {
 775                 int unused;
 776                 /* IO port based C-state */
 777                 inb(cx->address);
 778                 /* Dummy wait op - must do something useless after P_LVL2 read
 779                    because chipsets cannot guarantee that STPCLK# signal
 780                    gets asserted in time to freeze execution properly. */
 781                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 782         }
 783         start_critical_timings();
 784 }
 785
 786 /**
 787  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
 788  * @dev: the target CPU
 789  * @state: the state data
 790  *
 791  * This is equivalent to the HALT instruction.
 792  */
 793 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
 794                               struct cpuidle_state *state)
 795 {
 796         ktime_t  kt1, kt2;
 797         s64 idle_time;
 798         struct acpi_processor *pr;
 799         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 800
 801         pr = __get_cpu_var(processors);
 802
 803         if (unlikely(!pr))
 804                 return 0;
 805
 806         local_irq_disable();
 807
 808         /* Do not access any ACPI IO ports in suspend path */
 809         if (acpi_idle_suspend) {
 810                 acpi_safe_halt();
 811                 local_irq_enable();
 812                 return 0;
 813         }
 814
 815         kt1 = ktime_get_real();
 816         acpi_idle_do_entry(cx);
 817         kt2 = ktime_get_real();
 818         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 819
 820         local_irq_enable();
 821         cx->usage++;
 822
 823         return idle_time;
 824 }
 825
 826 /**
 827  * acpi_idle_enter_simple - enters an ACPI state without BM handling
 828  * @dev: the target CPU
 829  * @state: the state data
 830  */
 831 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 832                                   struct cpuidle_state *state)
 833 {
 834         struct acpi_processor *pr;
 835         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 836         ktime_t  kt1, kt2;
 837         s64 idle_time;
 838         s64 sleep_ticks = 0;
 839
 840         pr = __get_cpu_var(processors);
 841
 842         if (unlikely(!pr))
 843                 return 0;
 844
 845         if (acpi_idle_suspend)
 846                 return(acpi_idle_enter_c1(dev, state));
 847
 848         local_irq_disable();
 849         current_thread_info()->status &= ~TS_POLLING;
 850         /*
 851          * TS_POLLING-cleared state must be visible before we test
 852          * NEED_RESCHED:
 853          */
 854         smp_mb();
 855
 856         if (unlikely(need_resched())) {
 857                 current_thread_info()->status |= TS_POLLING;
 858                 local_irq_enable();
 859                 return 0;
 860         }
 861
 862         /*
 863          * Must be done before busmaster disable as we might need to
 864          * access HPET !
 865          */
 866         acpi_state_timer_broadcast(pr, cx, 1);
 867
 868         if (cx->type == ACPI_STATE_C3)
 869                 ACPI_FLUSH_CPU_CACHE();
 870
 871         kt1 = ktime_get_real();
 872         /* Tell the scheduler that we are going deep-idle: */
 873         sched_clock_idle_sleep_event();
 874         acpi_idle_do_entry(cx);
 875         kt2 = ktime_get_real();
 876         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 877
 878         sleep_ticks = us_to_pm_timer_ticks(idle_time);
 879
 880         /* Tell the scheduler how much we idled: */
 881         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 882
 883         local_irq_enable();
 884         current_thread_info()->status |= TS_POLLING;
 885
 886         cx->usage++;
 887
 888         acpi_state_timer_broadcast(pr, cx, 0);
 889         cx->time += sleep_ticks;
 890         return idle_time;
 891 }
 892
 893 static int c3_cpu_count;
 894 static DEFINE_SPINLOCK(c3_lock);
 895
 896 /**
 897  * acpi_idle_enter_bm - enters C3 with proper BM handling
 898  * @dev: the target CPU
 899  * @state: the state data
 900  *
 901  * If BM is detected, the deepest non-C3 idle state is entered instead.
 902  */
 903 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 904                               struct cpuidle_state *state)
 905 {
 906         struct acpi_processor *pr;
 907         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 908         ktime_t  kt1, kt2;
 909         s64 idle_time;
 910         s64 sleep_ticks = 0;
 911
 912
 913         pr = __get_cpu_var(processors);
 914
 915         if (unlikely(!pr))
 916                 return 0;
 917
 918         if (acpi_idle_suspend)
 919                 return(acpi_idle_enter_c1(dev, state));
 920
 921         if (acpi_idle_bm_check()) {
 922                 if (dev->safe_state) {
 923                         dev->last_state = dev->safe_state;
 924                         return dev->safe_state->enter(dev, dev->safe_state);
 925                 } else {
 926                         local_irq_disable();
 927                         acpi_safe_halt();
 928                         local_irq_enable();
 929                         return 0;
 930                 }
 931         }
 932
 933         local_irq_disable();
 934         current_thread_info()->status &= ~TS_POLLING;
 935         /*
 936          * TS_POLLING-cleared state must be visible before we test
 937          * NEED_RESCHED:
 938          */
 939         smp_mb();
 940
 941         if (unlikely(need_resched())) {
 942                 current_thread_info()->status |= TS_POLLING;
 943                 local_irq_enable();
 944                 return 0;
 945         }
 946
 947         acpi_unlazy_tlb(smp_processor_id());
 948
 949         /* Tell the scheduler that we are going deep-idle: */
 950         sched_clock_idle_sleep_event();
 951         /*
 952          * Must be done before busmaster disable as we might need to
 953          * access HPET !
 954          */
 955         acpi_state_timer_broadcast(pr, cx, 1);
 956
 957         kt1 = ktime_get_real();
 958         /*
 959          * disable bus master
 960          * bm_check implies we need ARB_DIS
 961          * !bm_check implies we need cache flush
 962          * bm_control implies whether we can do ARB_DIS
 963          *
 964          * That leaves a case where bm_check is set and bm_control is
 965          * not set. In that case we cannot do much, we enter C3
 966          * without doing anything.
 967          */
 968         if (pr->flags.bm_check && pr->flags.bm_control) {
 969                 spin_lock(&c3_lock);
 970                 c3_cpu_count++;
 971                 /* Disable bus master arbitration when all CPUs are in C3 */
 972                 if (c3_cpu_count == num_online_cpus())
 973                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 974                 spin_unlock(&c3_lock);
 975         } else if (!pr->flags.bm_check) {
 976                 ACPI_FLUSH_CPU_CACHE();
 977         }
 978
 979         acpi_idle_do_entry(cx);
 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         kt2 = ktime_get_real();
 989         idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
 990
 991         sleep_ticks = us_to_pm_timer_ticks(idle_time);
 992         /* Tell the scheduler how much we idled: */
 993         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 994
 995         local_irq_enable();
 996         current_thread_info()->status |= TS_POLLING;
 997
 998         cx->usage++;
 999
1000         acpi_state_timer_broadcast(pr, cx, 0);
1001         cx->time += sleep_ticks;
1002         return idle_time;
1003 }
1004
1005 struct cpuidle_driver acpi_idle_driver = {
1006         .name =         "acpi_idle",
1007         .owner =        THIS_MODULE,
1008 };
1009
1010 /**
1011  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1012  * @pr: the ACPI processor
1013  */
1014 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1015 {
1016         int i, count = CPUIDLE_DRIVER_STATE_START;
1017         struct acpi_processor_cx *cx;
1018         struct cpuidle_state *state;
1019         struct cpuidle_device *dev = &pr->power.dev;
1020
1021         if (!pr->flags.power_setup_done)
1022                 return -EINVAL;
1023
1024         if (pr->flags.power == 0) {
1025                 return -EINVAL;
1026         }
1027
1028         dev->cpu = pr->id;
1029         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1030                 dev->states[i].name[0] = '\0';
1031                 dev->states[i].desc[0] = '\0';
1032         }
1033
1034         if (max_cstate == 0)
1035                 max_cstate = 1;
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 }