Pileus Git - ~andy/linux/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 US_TO_PM_TIMER_TICKS(t)         ((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
  68 #define PM_TIMER_TICK_NS                (1000000000ULL/PM_TIMER_FREQUENCY)
  69 #define C2_OVERHEAD                     1       /* 1us */
  70 #define C3_OVERHEAD                     1       /* 1us */
  71 #define PM_TIMER_TICKS_TO_US(p)         (((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
  72
  73 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  74 module_param(max_cstate, uint, 0000);
  75 static unsigned int nocst __read_mostly;
  76 module_param(nocst, uint, 0000);
  77
  78 static unsigned int latency_factor __read_mostly = 2;
  79 module_param(latency_factor, uint, 0644);
  80
  81 /*
  82  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  83  * For now disable this. Probably a bug somewhere else.
  84  *
  85  * To skip this limit, boot/load with a large max_cstate limit.
  86  */
  87 static int set_max_cstate(const struct dmi_system_id *id)
  88 {
  89         if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  90                 return 0;
  91
  92         printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
  93                " Override with \"processor.max_cstate=%d\"\n", id->ident,
  94                (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  95
  96         max_cstate = (long)id->driver_data;
  97
  98         return 0;
  99 }
 100
 101 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
 102    callers to only run once -AK */
 103 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
 104         { set_max_cstate, "Clevo 5600D", {
 105           DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
 106           DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
 107          (void *)2},
 108         {},
 109 };
 110
 111 static inline u32 ticks_elapsed(u32 t1, u32 t2)
 112 {
 113         if (t2 >= t1)
 114                 return (t2 - t1);
 115         else if (!(acpi_gbl_FADT.flags & ACPI_FADT_32BIT_TIMER))
 116                 return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 117         else
 118                 return ((0xFFFFFFFF - t1) + t2);
 119 }
 120
 121 static inline u32 ticks_elapsed_in_us(u32 t1, u32 t2)
 122 {
 123         if (t2 >= t1)
 124                 return PM_TIMER_TICKS_TO_US(t2 - t1);
 125         else if (!(acpi_gbl_FADT.flags & ACPI_FADT_32BIT_TIMER))
 126                 return PM_TIMER_TICKS_TO_US(((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
 127         else
 128                 return PM_TIMER_TICKS_TO_US((0xFFFFFFFF - t1) + t2);
 129 }
 130
 131 /*
 132  * Callers should disable interrupts before the call and enable
 133  * interrupts after return.
 134  */
 135 static void acpi_safe_halt(void)
 136 {
 137         current_thread_info()->status &= ~TS_POLLING;
 138         /*
 139          * TS_POLLING-cleared state must be visible before we
 140          * test NEED_RESCHED:
 141          */
 142         smp_mb();
 143         if (!need_resched()) {
 144                 safe_halt();
 145                 local_irq_disable();
 146         }
 147         current_thread_info()->status |= TS_POLLING;
 148 }
 149
 150 #ifdef ARCH_APICTIMER_STOPS_ON_C3
 151
 152 /*
 153  * Some BIOS implementations switch to C3 in the published C2 state.
 154  * This seems to be a common problem on AMD boxen, but other vendors
 155  * are affected too. We pick the most conservative approach: we assume
 156  * that the local APIC stops in both C2 and C3.
 157  */
 158 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 159                                    struct acpi_processor_cx *cx)
 160 {
 161         struct acpi_processor_power *pwr = &pr->power;
 162         u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 163
 164         /*
 165          * Check, if one of the previous states already marked the lapic
 166          * unstable
 167          */
 168         if (pwr->timer_broadcast_on_state < state)
 169                 return;
 170
 171         if (cx->type >= type)
 172                 pr->power.timer_broadcast_on_state = state;
 173 }
 174
 175 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
 176 {
 177         unsigned long reason;
 178
 179         reason = pr->power.timer_broadcast_on_state < INT_MAX ?
 180                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 181
 182         clockevents_notify(reason, &pr->id);
 183 }
 184
 185 /* Power(C) State timer broadcast control */
 186 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 187                                        struct acpi_processor_cx *cx,
 188                                        int broadcast)
 189 {
 190         int state = cx - pr->power.states;
 191
 192         if (state >= pr->power.timer_broadcast_on_state) {
 193                 unsigned long reason;
 194
 195                 reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
 196                         CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
 197                 clockevents_notify(reason, &pr->id);
 198         }
 199 }
 200
 201 #else
 202
 203 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
 204                                    struct acpi_processor_cx *cstate) { }
 205 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
 206 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
 207                                        struct acpi_processor_cx *cx,
 208                                        int broadcast)
 209 {
 210 }
 211
 212 #endif
 213
 214 /*
 215  * Suspend / resume control
 216  */
 217 static int acpi_idle_suspend;
 218
 219 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
 220 {
 221         acpi_idle_suspend = 1;
 222         return 0;
 223 }
 224
 225 int acpi_processor_resume(struct acpi_device * device)
 226 {
 227         acpi_idle_suspend = 0;
 228         return 0;
 229 }
 230
 231 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 232 static int tsc_halts_in_c(int state)
 233 {
 234         switch (boot_cpu_data.x86_vendor) {
 235         case X86_VENDOR_AMD:
 236         case X86_VENDOR_INTEL:
 237                 /*
 238                  * AMD Fam10h TSC will tick in all
 239                  * C/P/S0/S1 states when this bit is set.
 240                  */
 241                 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 242                         return 0;
 243
 244                 /*FALL THROUGH*/
 245         default:
 246                 return state > ACPI_STATE_C1;
 247         }
 248 }
 249 #endif
 250
 251 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 252 {
 253
 254         if (!pr)
 255                 return -EINVAL;
 256
 257         if (!pr->pblk)
 258                 return -ENODEV;
 259
 260         /* if info is obtained from pblk/fadt, type equals state */
 261         pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 262         pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 263
 264 #ifndef CONFIG_HOTPLUG_CPU
 265         /*
 266          * Check for P_LVL2_UP flag before entering C2 and above on
 267          * an SMP system.
 268          */
 269         if ((num_online_cpus() > 1) &&
 270             !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 271                 return -ENODEV;
 272 #endif
 273
 274         /* determine C2 and C3 address from pblk */
 275         pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 276         pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 277
 278         /* determine latencies from FADT */
 279         pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
 280         pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
 281
 282         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 283                           "lvl2[0x%08x] lvl3[0x%08x]\n",
 284                           pr->power.states[ACPI_STATE_C2].address,
 285                           pr->power.states[ACPI_STATE_C3].address));
 286
 287         return 0;
 288 }
 289
 290 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 291 {
 292         if (!pr->power.states[ACPI_STATE_C1].valid) {
 293                 /* set the first C-State to C1 */
 294                 /* all processors need to support C1 */
 295                 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 296                 pr->power.states[ACPI_STATE_C1].valid = 1;
 297                 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 298         }
 299         /* the C0 state only exists as a filler in our array */
 300         pr->power.states[ACPI_STATE_C0].valid = 1;
 301         return 0;
 302 }
 303
 304 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 305 {
 306         acpi_status status = 0;
 307         acpi_integer count;
 308         int current_count;
 309         int i;
 310         struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 311         union acpi_object *cst;
 312
 313
 314         if (nocst)
 315                 return -ENODEV;
 316
 317         current_count = 0;
 318
 319         status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
 320         if (ACPI_FAILURE(status)) {
 321                 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
 322                 return -ENODEV;
 323         }
 324
 325         cst = buffer.pointer;
 326
 327         /* There must be at least 2 elements */
 328         if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
 329                 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
 330                 status = -EFAULT;
 331                 goto end;
 332         }
 333
 334         count = cst->package.elements[0].integer.value;
 335
 336         /* Validate number of power states. */
 337         if (count < 1 || count != cst->package.count - 1) {
 338                 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
 339                 status = -EFAULT;
 340                 goto end;
 341         }
 342
 343         /* Tell driver that at least _CST is supported. */
 344         pr->flags.has_cst = 1;
 345
 346         for (i = 1; i <= count; i++) {
 347                 union acpi_object *element;
 348                 union acpi_object *obj;
 349                 struct acpi_power_register *reg;
 350                 struct acpi_processor_cx cx;
 351
 352                 memset(&cx, 0, sizeof(cx));
 353
 354                 element = &(cst->package.elements[i]);
 355                 if (element->type != ACPI_TYPE_PACKAGE)
 356                         continue;
 357
 358                 if (element->package.count != 4)
 359                         continue;
 360
 361                 obj = &(element->package.elements[0]);
 362
 363                 if (obj->type != ACPI_TYPE_BUFFER)
 364                         continue;
 365
 366                 reg = (struct acpi_power_register *)obj->buffer.pointer;
 367
 368                 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 369                     (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
 370                         continue;
 371
 372                 /* There should be an easy way to extract an integer... */
 373                 obj = &(element->package.elements[1]);
 374                 if (obj->type != ACPI_TYPE_INTEGER)
 375                         continue;
 376
 377                 cx.type = obj->integer.value;
 378                 /*
 379                  * Some buggy BIOSes won't list C1 in _CST -
 380                  * Let acpi_processor_get_power_info_default() handle them later
 381                  */
 382                 if (i == 1 && cx.type != ACPI_STATE_C1)
 383                         current_count++;
 384
 385                 cx.address = reg->address;
 386                 cx.index = current_count + 1;
 387
 388                 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
 389                 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
 390                         if (acpi_processor_ffh_cstate_probe
 391                                         (pr->id, &cx, reg) == 0) {
 392                                 cx.entry_method = ACPI_CSTATE_FFH;
 393                         } else if (cx.type == ACPI_STATE_C1) {
 394                                 /*
 395                                  * C1 is a special case where FIXED_HARDWARE
 396                                  * can be handled in non-MWAIT way as well.
 397                                  * In that case, save this _CST entry info.
 398                                  * Otherwise, ignore this info and continue.
 399                                  */
 400                                 cx.entry_method = ACPI_CSTATE_HALT;
 401                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 402                         } else {
 403                                 continue;
 404                         }
 405                         if (cx.type == ACPI_STATE_C1 &&
 406                                         (idle_halt || idle_nomwait)) {
 407                                 /*
 408                                  * In most cases the C1 space_id obtained from
 409                                  * _CST object is FIXED_HARDWARE access mode.
 410                                  * But when the option of idle=halt is added,
 411                                  * the entry_method type should be changed from
 412                                  * CSTATE_FFH to CSTATE_HALT.
 413                                  * When the option of idle=nomwait is added,
 414                                  * the C1 entry_method type should be
 415                                  * CSTATE_HALT.
 416                                  */
 417                                 cx.entry_method = ACPI_CSTATE_HALT;
 418                                 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
 419                         }
 420                 } else {
 421                         snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
 422                                  cx.address);
 423                 }
 424
 425                 if (cx.type == ACPI_STATE_C1) {
 426                         cx.valid = 1;
 427                 }
 428
 429                 obj = &(element->package.elements[2]);
 430                 if (obj->type != ACPI_TYPE_INTEGER)
 431                         continue;
 432
 433                 cx.latency = obj->integer.value;
 434
 435                 obj = &(element->package.elements[3]);
 436                 if (obj->type != ACPI_TYPE_INTEGER)
 437                         continue;
 438
 439                 cx.power = obj->integer.value;
 440
 441                 current_count++;
 442                 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
 443
 444                 /*
 445                  * We support total ACPI_PROCESSOR_MAX_POWER - 1
 446                  * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
 447                  */
 448                 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
 449                         printk(KERN_WARNING
 450                                "Limiting number of power states to max (%d)\n",
 451                                ACPI_PROCESSOR_MAX_POWER);
 452                         printk(KERN_WARNING
 453                                "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
 454                         break;
 455                 }
 456         }
 457
 458         ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
 459                           current_count));
 460
 461         /* Validate number of power states discovered */
 462         if (current_count < 2)
 463                 status = -EFAULT;
 464
 465       end:
 466         kfree(buffer.pointer);
 467
 468         return status;
 469 }
 470
 471 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
 472 {
 473
 474         if (!cx->address)
 475                 return;
 476
 477         /*
 478          * C2 latency must be less than or equal to 100
 479          * microseconds.
 480          */
 481         else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 482                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 483                                   "latency too large [%d]\n", cx->latency));
 484                 return;
 485         }
 486
 487         /*
 488          * Otherwise we've met all of our C2 requirements.
 489          * Normalize the C2 latency to expidite policy
 490          */
 491         cx->valid = 1;
 492
 493         cx->latency_ticks = cx->latency;
 494
 495         return;
 496 }
 497
 498 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 499                                            struct acpi_processor_cx *cx)
 500 {
 501         static int bm_check_flag;
 502
 503
 504         if (!cx->address)
 505                 return;
 506
 507         /*
 508          * C3 latency must be less than or equal to 1000
 509          * microseconds.
 510          */
 511         else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 512                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 513                                   "latency too large [%d]\n", cx->latency));
 514                 return;
 515         }
 516
 517         /*
 518          * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 519          * DMA transfers are used by any ISA device to avoid livelock.
 520          * Note that we could disable Type-F DMA (as recommended by
 521          * the erratum), but this is known to disrupt certain ISA
 522          * devices thus we take the conservative approach.
 523          */
 524         else if (errata.piix4.fdma) {
 525                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 526                                   "C3 not supported on PIIX4 with Type-F DMA\n"));
 527                 return;
 528         }
 529
 530         /* All the logic here assumes flags.bm_check is same across all CPUs */
 531         if (!bm_check_flag) {
 532                 /* Determine whether bm_check is needed based on CPU  */
 533                 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 534                 bm_check_flag = pr->flags.bm_check;
 535         } else {
 536                 pr->flags.bm_check = bm_check_flag;
 537         }
 538
 539         if (pr->flags.bm_check) {
 540                 if (!pr->flags.bm_control) {
 541                         if (pr->flags.has_cst != 1) {
 542                                 /* bus mastering control is necessary */
 543                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 544                                         "C3 support requires BM control\n"));
 545                                 return;
 546                         } else {
 547                                 /* Here we enter C3 without bus mastering */
 548                                 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 549                                         "C3 support without BM control\n"));
 550                         }
 551                 }
 552         } else {
 553                 /*
 554                  * WBINVD should be set in fadt, for C3 state to be
 555                  * supported on when bm_check is not required.
 556                  */
 557                 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 558                         ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 559                                           "Cache invalidation should work properly"
 560                                           " for C3 to be enabled on SMP systems\n"));
 561                         return;
 562                 }
 563         }
 564
 565         /*
 566          * Otherwise we've met all of our C3 requirements.
 567          * Normalize the C3 latency to expidite policy.  Enable
 568          * checking of bus mastering status (bm_check) so we can
 569          * use this in our C3 policy
 570          */
 571         cx->valid = 1;
 572
 573         cx->latency_ticks = cx->latency;
 574         /*
 575          * On older chipsets, BM_RLD needs to be set
 576          * in order for Bus Master activity to wake the
 577          * system from C3.  Newer chipsets handle DMA
 578          * during C3 automatically and BM_RLD is a NOP.
 579          * In either case, the proper way to
 580          * handle BM_RLD is to set it and leave it set.
 581          */
 582         acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 583
 584         return;
 585 }
 586
 587 static int acpi_processor_power_verify(struct acpi_processor *pr)
 588 {
 589         unsigned int i;
 590         unsigned int working = 0;
 591
 592         pr->power.timer_broadcast_on_state = INT_MAX;
 593
 594         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 595                 struct acpi_processor_cx *cx = &pr->power.states[i];
 596
 597                 switch (cx->type) {
 598                 case ACPI_STATE_C1:
 599                         cx->valid = 1;
 600                         break;
 601
 602                 case ACPI_STATE_C2:
 603                         acpi_processor_power_verify_c2(cx);
 604                         if (cx->valid)
 605                                 acpi_timer_check_state(i, pr, cx);
 606                         break;
 607
 608                 case ACPI_STATE_C3:
 609                         acpi_processor_power_verify_c3(pr, cx);
 610                         if (cx->valid)
 611                                 acpi_timer_check_state(i, pr, cx);
 612                         break;
 613                 }
 614
 615                 if (cx->valid)
 616                         working++;
 617         }
 618
 619         acpi_propagate_timer_broadcast(pr);
 620
 621         return (working);
 622 }
 623
 624 static int acpi_processor_get_power_info(struct acpi_processor *pr)
 625 {
 626         unsigned int i;
 627         int result;
 628
 629
 630         /* NOTE: the idle thread may not be running while calling
 631          * this function */
 632
 633         /* Zero initialize all the C-states info. */
 634         memset(pr->power.states, 0, sizeof(pr->power.states));
 635
 636         result = acpi_processor_get_power_info_cst(pr);
 637         if (result == -ENODEV)
 638                 result = acpi_processor_get_power_info_fadt(pr);
 639
 640         if (result)
 641                 return result;
 642
 643         acpi_processor_get_power_info_default(pr);
 644
 645         pr->power.count = acpi_processor_power_verify(pr);
 646
 647         /*
 648          * if one state of type C2 or C3 is available, mark this
 649          * CPU as being "idle manageable"
 650          */
 651         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 652                 if (pr->power.states[i].valid) {
 653                         pr->power.count = i;
 654                         if (pr->power.states[i].type >= ACPI_STATE_C2)
 655                                 pr->flags.power = 1;
 656                 }
 657         }
 658
 659         return 0;
 660 }
 661
 662 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
 663 {
 664         struct acpi_processor *pr = seq->private;
 665         unsigned int i;
 666
 667
 668         if (!pr)
 669                 goto end;
 670
 671         seq_printf(seq, "active state:            C%zd\n"
 672                    "max_cstate:              C%d\n"
 673                    "bus master activity:     %08x\n"
 674                    "maximum allowed latency: %d usec\n",
 675                    pr->power.state ? pr->power.state - pr->power.states : 0,
 676                    max_cstate, (unsigned)pr->power.bm_activity,
 677                    pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
 678
 679         seq_puts(seq, "states:\n");
 680
 681         for (i = 1; i <= pr->power.count; i++) {
 682                 seq_printf(seq, "   %cC%d:                  ",
 683                            (&pr->power.states[i] ==
 684                             pr->power.state ? '*' : ' '), i);
 685
 686                 if (!pr->power.states[i].valid) {
 687                         seq_puts(seq, "<not supported>\n");
 688                         continue;
 689                 }
 690
 691                 switch (pr->power.states[i].type) {
 692                 case ACPI_STATE_C1:
 693                         seq_printf(seq, "type[C1] ");
 694                         break;
 695                 case ACPI_STATE_C2:
 696                         seq_printf(seq, "type[C2] ");
 697                         break;
 698                 case ACPI_STATE_C3:
 699                         seq_printf(seq, "type[C3] ");
 700                         break;
 701                 default:
 702                         seq_printf(seq, "type[--] ");
 703                         break;
 704                 }
 705
 706                 if (pr->power.states[i].promotion.state)
 707                         seq_printf(seq, "promotion[C%zd] ",
 708                                    (pr->power.states[i].promotion.state -
 709                                     pr->power.states));
 710                 else
 711                         seq_puts(seq, "promotion[--] ");
 712
 713                 if (pr->power.states[i].demotion.state)
 714                         seq_printf(seq, "demotion[C%zd] ",
 715                                    (pr->power.states[i].demotion.state -
 716                                     pr->power.states));
 717                 else
 718                         seq_puts(seq, "demotion[--] ");
 719
 720                 seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
 721                            pr->power.states[i].latency,
 722                            pr->power.states[i].usage,
 723                            (unsigned long long)pr->power.states[i].time);
 724         }
 725
 726       end:
 727         return 0;
 728 }
 729
 730 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
 731 {
 732         return single_open(file, acpi_processor_power_seq_show,
 733                            PDE(inode)->data);
 734 }
 735
 736 static const struct file_operations acpi_processor_power_fops = {
 737         .owner = THIS_MODULE,
 738         .open = acpi_processor_power_open_fs,
 739         .read = seq_read,
 740         .llseek = seq_lseek,
 741         .release = single_release,
 742 };
 743
 744
 745 /**
 746  * acpi_idle_bm_check - checks if bus master activity was detected
 747  */
 748 static int acpi_idle_bm_check(void)
 749 {
 750         u32 bm_status = 0;
 751
 752         acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 753         if (bm_status)
 754                 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 755         /*
 756          * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 757          * the true state of bus mastering activity; forcing us to
 758          * manually check the BMIDEA bit of each IDE channel.
 759          */
 760         else if (errata.piix4.bmisx) {
 761                 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 762                     || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 763                         bm_status = 1;
 764         }
 765         return bm_status;
 766 }
 767
 768 /**
 769  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
 770  * @cx: cstate data
 771  *
 772  * Caller disables interrupt before call and enables interrupt after return.
 773  */
 774 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
 775 {
 776         /* Don't trace irqs off for idle */
 777         stop_critical_timings();
 778         if (cx->entry_method == ACPI_CSTATE_FFH) {
 779                 /* Call into architectural FFH based C-state */
 780                 acpi_processor_ffh_cstate_enter(cx);
 781         } else if (cx->entry_method == ACPI_CSTATE_HALT) {
 782                 acpi_safe_halt();
 783         } else {
 784                 int unused;
 785                 /* IO port based C-state */
 786                 inb(cx->address);
 787                 /* Dummy wait op - must do something useless after P_LVL2 read
 788                    because chipsets cannot guarantee that STPCLK# signal
 789                    gets asserted in time to freeze execution properly. */
 790                 unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
 791         }
 792         start_critical_timings();
 793 }
 794
 795 /**
 796  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
 797  * @dev: the target CPU
 798  * @state: the state data
 799  *
 800  * This is equivalent to the HALT instruction.
 801  */
 802 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
 803                               struct cpuidle_state *state)
 804 {
 805         u32 t1, t2;
 806         struct acpi_processor *pr;
 807         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 808
 809         pr = __get_cpu_var(processors);
 810
 811         if (unlikely(!pr))
 812                 return 0;
 813
 814         local_irq_disable();
 815
 816         /* Do not access any ACPI IO ports in suspend path */
 817         if (acpi_idle_suspend) {
 818                 acpi_safe_halt();
 819                 local_irq_enable();
 820                 return 0;
 821         }
 822
 823         t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 824         acpi_idle_do_entry(cx);
 825         t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 826
 827         local_irq_enable();
 828         cx->usage++;
 829
 830         return ticks_elapsed_in_us(t1, t2);
 831 }
 832
 833 /**
 834  * acpi_idle_enter_simple - enters an ACPI state without BM handling
 835  * @dev: the target CPU
 836  * @state: the state data
 837  */
 838 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
 839                                   struct cpuidle_state *state)
 840 {
 841         struct acpi_processor *pr;
 842         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 843         u32 t1, t2;
 844         int sleep_ticks = 0;
 845
 846         pr = __get_cpu_var(processors);
 847
 848         if (unlikely(!pr))
 849                 return 0;
 850
 851         if (acpi_idle_suspend)
 852                 return(acpi_idle_enter_c1(dev, state));
 853
 854         local_irq_disable();
 855         current_thread_info()->status &= ~TS_POLLING;
 856         /*
 857          * TS_POLLING-cleared state must be visible before we test
 858          * NEED_RESCHED:
 859          */
 860         smp_mb();
 861
 862         if (unlikely(need_resched())) {
 863                 current_thread_info()->status |= TS_POLLING;
 864                 local_irq_enable();
 865                 return 0;
 866         }
 867
 868         /*
 869          * Must be done before busmaster disable as we might need to
 870          * access HPET !
 871          */
 872         acpi_state_timer_broadcast(pr, cx, 1);
 873
 874         if (cx->type == ACPI_STATE_C3)
 875                 ACPI_FLUSH_CPU_CACHE();
 876
 877         t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 878         /* Tell the scheduler that we are going deep-idle: */
 879         sched_clock_idle_sleep_event();
 880         acpi_idle_do_entry(cx);
 881         t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 882
 883 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 884         /* TSC could halt in idle, so notify users */
 885         if (tsc_halts_in_c(cx->type))
 886                 mark_tsc_unstable("TSC halts in idle");;
 887 #endif
 888         sleep_ticks = ticks_elapsed(t1, t2);
 889
 890         /* Tell the scheduler how much we idled: */
 891         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
 892
 893         local_irq_enable();
 894         current_thread_info()->status |= TS_POLLING;
 895
 896         cx->usage++;
 897
 898         acpi_state_timer_broadcast(pr, cx, 0);
 899         cx->time += sleep_ticks;
 900         return ticks_elapsed_in_us(t1, t2);
 901 }
 902
 903 static int c3_cpu_count;
 904 static DEFINE_SPINLOCK(c3_lock);
 905
 906 /**
 907  * acpi_idle_enter_bm - enters C3 with proper BM handling
 908  * @dev: the target CPU
 909  * @state: the state data
 910  *
 911  * If BM is detected, the deepest non-C3 idle state is entered instead.
 912  */
 913 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
 914                               struct cpuidle_state *state)
 915 {
 916         struct acpi_processor *pr;
 917         struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
 918         u32 t1, t2;
 919         int sleep_ticks = 0;
 920
 921         pr = __get_cpu_var(processors);
 922
 923         if (unlikely(!pr))
 924                 return 0;
 925
 926         if (acpi_idle_suspend)
 927                 return(acpi_idle_enter_c1(dev, state));
 928
 929         if (acpi_idle_bm_check()) {
 930                 if (dev->safe_state) {
 931                         dev->last_state = dev->safe_state;
 932                         return dev->safe_state->enter(dev, dev->safe_state);
 933                 } else {
 934                         local_irq_disable();
 935                         acpi_safe_halt();
 936                         local_irq_enable();
 937                         return 0;
 938                 }
 939         }
 940
 941         local_irq_disable();
 942         current_thread_info()->status &= ~TS_POLLING;
 943         /*
 944          * TS_POLLING-cleared state must be visible before we test
 945          * NEED_RESCHED:
 946          */
 947         smp_mb();
 948
 949         if (unlikely(need_resched())) {
 950                 current_thread_info()->status |= TS_POLLING;
 951                 local_irq_enable();
 952                 return 0;
 953         }
 954
 955         acpi_unlazy_tlb(smp_processor_id());
 956
 957         /* Tell the scheduler that we are going deep-idle: */
 958         sched_clock_idle_sleep_event();
 959         /*
 960          * Must be done before busmaster disable as we might need to
 961          * access HPET !
 962          */
 963         acpi_state_timer_broadcast(pr, cx, 1);
 964
 965         /*
 966          * disable bus master
 967          * bm_check implies we need ARB_DIS
 968          * !bm_check implies we need cache flush
 969          * bm_control implies whether we can do ARB_DIS
 970          *
 971          * That leaves a case where bm_check is set and bm_control is
 972          * not set. In that case we cannot do much, we enter C3
 973          * without doing anything.
 974          */
 975         if (pr->flags.bm_check && pr->flags.bm_control) {
 976                 spin_lock(&c3_lock);
 977                 c3_cpu_count++;
 978                 /* Disable bus master arbitration when all CPUs are in C3 */
 979                 if (c3_cpu_count == num_online_cpus())
 980                         acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 981                 spin_unlock(&c3_lock);
 982         } else if (!pr->flags.bm_check) {
 983                 ACPI_FLUSH_CPU_CACHE();
 984         }
 985
 986         t1 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 987         acpi_idle_do_entry(cx);
 988         t2 = inl(acpi_gbl_FADT.xpm_timer_block.address);
 989
 990         /* Re-enable bus master arbitration */
 991         if (pr->flags.bm_check && pr->flags.bm_control) {
 992                 spin_lock(&c3_lock);
 993                 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 994                 c3_cpu_count--;
 995                 spin_unlock(&c3_lock);
 996         }
 997
 998 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
 999         /* TSC could halt in idle, so notify users */
1000         if (tsc_halts_in_c(ACPI_STATE_C3))
1001                 mark_tsc_unstable("TSC halts in idle");
1002 #endif
1003         sleep_ticks = ticks_elapsed(t1, t2);
1004         /* Tell the scheduler how much we idled: */
1005         sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
1006
1007         local_irq_enable();
1008         current_thread_info()->status |= TS_POLLING;
1009
1010         cx->usage++;
1011
1012         acpi_state_timer_broadcast(pr, cx, 0);
1013         cx->time += sleep_ticks;
1014         return ticks_elapsed_in_us(t1, t2);
1015 }
1016
1017 struct cpuidle_driver acpi_idle_driver = {
1018         .name =         "acpi_idle",
1019         .owner =        THIS_MODULE,
1020 };
1021
1022 /**
1023  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1024  * @pr: the ACPI processor
1025  */
1026 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1027 {
1028         int i, count = CPUIDLE_DRIVER_STATE_START;
1029         struct acpi_processor_cx *cx;
1030         struct cpuidle_state *state;
1031         struct cpuidle_device *dev = &pr->power.dev;
1032
1033         if (!pr->flags.power_setup_done)
1034                 return -EINVAL;
1035
1036         if (pr->flags.power == 0) {
1037                 return -EINVAL;
1038         }
1039
1040         dev->cpu = pr->id;
1041         for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1042                 dev->states[i].name[0] = '\0';
1043                 dev->states[i].desc[0] = '\0';
1044         }
1045
1046         for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1047                 cx = &pr->power.states[i];
1048                 state = &dev->states[count];
1049
1050                 if (!cx->valid)
1051                         continue;
1052
1053 #ifdef CONFIG_HOTPLUG_CPU
1054                 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1055                     !pr->flags.has_cst &&
1056                     !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1057                         continue;
1058 #endif
1059                 cpuidle_set_statedata(state, cx);
1060
1061                 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1062                 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1063                 state->exit_latency = cx->latency;
1064                 state->target_residency = cx->latency * latency_factor;
1065                 state->power_usage = cx->power;
1066
1067                 state->flags = 0;
1068                 switch (cx->type) {
1069                         case ACPI_STATE_C1:
1070                         state->flags |= CPUIDLE_FLAG_SHALLOW;
1071                         if (cx->entry_method == ACPI_CSTATE_FFH)
1072                                 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1073
1074                         state->enter = acpi_idle_enter_c1;
1075                         dev->safe_state = state;
1076                         break;
1077
1078                         case ACPI_STATE_C2:
1079                         state->flags |= CPUIDLE_FLAG_BALANCED;
1080                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1081                         state->enter = acpi_idle_enter_simple;
1082                         dev->safe_state = state;
1083                         break;
1084
1085                         case ACPI_STATE_C3:
1086                         state->flags |= CPUIDLE_FLAG_DEEP;
1087                         state->flags |= CPUIDLE_FLAG_TIME_VALID;
1088                         state->flags |= CPUIDLE_FLAG_CHECK_BM;
1089                         state->enter = pr->flags.bm_check ?
1090                                         acpi_idle_enter_bm :
1091                                         acpi_idle_enter_simple;
1092                         break;
1093                 }
1094
1095                 count++;
1096                 if (count == CPUIDLE_STATE_MAX)
1097                         break;
1098         }
1099
1100         dev->state_count = count;
1101
1102         if (!count)
1103                 return -EINVAL;
1104
1105         return 0;
1106 }
1107
1108 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1109 {
1110         int ret = 0;
1111
1112         if (boot_option_idle_override)
1113                 return 0;
1114
1115         if (!pr)
1116                 return -EINVAL;
1117
1118         if (nocst) {
1119                 return -ENODEV;
1120         }
1121
1122         if (!pr->flags.power_setup_done)
1123                 return -ENODEV;
1124
1125         cpuidle_pause_and_lock();
1126         cpuidle_disable_device(&pr->power.dev);
1127         acpi_processor_get_power_info(pr);
1128         if (pr->flags.power) {
1129                 acpi_processor_setup_cpuidle(pr);
1130                 ret = cpuidle_enable_device(&pr->power.dev);
1131         }
1132         cpuidle_resume_and_unlock();
1133
1134         return ret;
1135 }
1136
1137 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1138                               struct acpi_device *device)
1139 {
1140         acpi_status status = 0;
1141         static int first_run;
1142         struct proc_dir_entry *entry = NULL;
1143         unsigned int i;
1144
1145         if (boot_option_idle_override)
1146                 return 0;
1147
1148         if (!first_run) {
1149                 if (idle_halt) {
1150                         /*
1151                          * When the boot option of "idle=halt" is added, halt
1152                          * is used for CPU IDLE.
1153                          * In such case C2/C3 is meaningless. So the max_cstate
1154                          * is set to one.
1155                          */
1156                         max_cstate = 1;
1157                 }
1158                 dmi_check_system(processor_power_dmi_table);
1159                 max_cstate = acpi_processor_cstate_check(max_cstate);
1160                 if (max_cstate < ACPI_C_STATES_MAX)
1161                         printk(KERN_NOTICE
1162                                "ACPI: processor limited to max C-state %d\n",
1163                                max_cstate);
1164                 first_run++;
1165         }
1166
1167         if (!pr)
1168                 return -EINVAL;
1169
1170         if (acpi_gbl_FADT.cst_control && !nocst) {
1171                 status =
1172                     acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1173                 if (ACPI_FAILURE(status)) {
1174                         ACPI_EXCEPTION((AE_INFO, status,
1175                                         "Notifying BIOS of _CST ability failed"));
1176                 }
1177         }
1178
1179         acpi_processor_get_power_info(pr);
1180         pr->flags.power_setup_done = 1;
1181
1182         /*
1183          * Install the idle handler if processor power management is supported.
1184          * Note that we use previously set idle handler will be used on
1185          * platforms that only support C1.
1186          */
1187         if (pr->flags.power) {
1188                 acpi_processor_setup_cpuidle(pr);
1189                 if (cpuidle_register_device(&pr->power.dev))
1190                         return -EIO;
1191
1192                 printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1193                 for (i = 1; i <= pr->power.count; i++)
1194                         if (pr->power.states[i].valid)
1195                                 printk(" C%d[C%d]", i,
1196                                        pr->power.states[i].type);
1197                 printk(")\n");
1198         }
1199
1200         /* 'power' [R] */
1201         entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1202                                  S_IRUGO, acpi_device_dir(device),
1203                                  &acpi_processor_power_fops,
1204                                  acpi_driver_data(device));
1205         if (!entry)
1206                 return -EIO;
1207         return 0;
1208 }
1209
1210 int acpi_processor_power_exit(struct acpi_processor *pr,
1211                               struct acpi_device *device)
1212 {
1213         if (boot_option_idle_override)
1214                 return 0;
1215
1216         cpuidle_unregister_device(&pr->power.dev);
1217         pr->flags.power_setup_done = 0;
1218
1219         if (acpi_device_dir(device))
1220                 remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1221                                   acpi_device_dir(device));
1222
1223         return 0;
1224 }