1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * processor_idle - idle state submodule to the ACPI processor driver 4 * 5 * Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com> 6 * Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com> 7 * Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de> 8 * Copyright (C) 2004 Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> 9 * - Added processor hotplug support 10 * Copyright (C) 2005 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> 11 * - Added support for C3 on SMP 12 */ 13 #define pr_fmt(fmt) "ACPI: " fmt 14 15 #include <linux/module.h> 16 #include <linux/acpi.h> 17 #include <linux/dmi.h> 18 #include <linux/sched.h> /* need_resched() */ 19 #include <linux/tick.h> 20 #include <linux/cpuidle.h> 21 #include <linux/cpu.h> 22 #include <acpi/processor.h> 23 24 /* 25 * Include the apic definitions for x86 to have the APIC timer related defines 26 * available also for UP (on SMP it gets magically included via linux/smp.h). 27 * asm/acpi.h is not an option, as it would require more include magic. Also 28 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera. 29 */ 30 #ifdef CONFIG_X86 31 #include <asm/apic.h> 32 #endif 33 34 #define ACPI_PROCESSOR_CLASS "processor" 35 #define _COMPONENT ACPI_PROCESSOR_COMPONENT 36 ACPI_MODULE_NAME("processor_idle"); 37 38 #define ACPI_IDLE_STATE_START (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0) 39 40 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER; 41 module_param(max_cstate, uint, 0000); 42 static unsigned int nocst __read_mostly; 43 module_param(nocst, uint, 0000); 44 static int bm_check_disable __read_mostly; 45 module_param(bm_check_disable, uint, 0000); 46 47 static unsigned int latency_factor __read_mostly = 2; 48 module_param(latency_factor, uint, 0644); 49 50 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device); 51 52 struct cpuidle_driver acpi_idle_driver = { 53 .name = "acpi_idle", 54 .owner = THIS_MODULE, 55 }; 56 57 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE 58 static 59 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate); 60 61 static int disabled_by_idle_boot_param(void) 62 { 63 return boot_option_idle_override == IDLE_POLL || 64 boot_option_idle_override == IDLE_HALT; 65 } 66 67 /* 68 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3. 69 * For now disable this. Probably a bug somewhere else. 70 * 71 * To skip this limit, boot/load with a large max_cstate limit. 72 */ 73 static int set_max_cstate(const struct dmi_system_id *id) 74 { 75 if (max_cstate > ACPI_PROCESSOR_MAX_POWER) 76 return 0; 77 78 pr_notice("%s detected - limiting to C%ld max_cstate." 79 " Override with \"processor.max_cstate=%d\"\n", id->ident, 80 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1); 81 82 max_cstate = (long)id->driver_data; 83 84 return 0; 85 } 86 87 static const struct dmi_system_id processor_power_dmi_table[] = { 88 { set_max_cstate, "Clevo 5600D", { 89 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"), 90 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")}, 91 (void *)2}, 92 { set_max_cstate, "Pavilion zv5000", { 93 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"), 94 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")}, 95 (void *)1}, 96 { set_max_cstate, "Asus L8400B", { 97 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."), 98 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")}, 99 (void *)1}, 100 {}, 101 }; 102 103 104 /* 105 * Callers should disable interrupts before the call and enable 106 * interrupts after return. 107 */ 108 static void __cpuidle acpi_safe_halt(void) 109 { 110 if (!tif_need_resched()) { 111 safe_halt(); 112 local_irq_disable(); 113 } 114 } 115 116 #ifdef ARCH_APICTIMER_STOPS_ON_C3 117 118 /* 119 * Some BIOS implementations switch to C3 in the published C2 state. 120 * This seems to be a common problem on AMD boxen, but other vendors 121 * are affected too. We pick the most conservative approach: we assume 122 * that the local APIC stops in both C2 and C3. 123 */ 124 static void lapic_timer_check_state(int state, struct acpi_processor *pr, 125 struct acpi_processor_cx *cx) 126 { 127 struct acpi_processor_power *pwr = &pr->power; 128 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2; 129 130 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT)) 131 return; 132 133 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) 134 type = ACPI_STATE_C1; 135 136 /* 137 * Check, if one of the previous states already marked the lapic 138 * unstable 139 */ 140 if (pwr->timer_broadcast_on_state < state) 141 return; 142 143 if (cx->type >= type) 144 pr->power.timer_broadcast_on_state = state; 145 } 146 147 static void __lapic_timer_propagate_broadcast(void *arg) 148 { 149 struct acpi_processor *pr = (struct acpi_processor *) arg; 150 151 if (pr->power.timer_broadcast_on_state < INT_MAX) 152 tick_broadcast_enable(); 153 else 154 tick_broadcast_disable(); 155 } 156 157 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) 158 { 159 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast, 160 (void *)pr, 1); 161 } 162 163 /* Power(C) State timer broadcast control */ 164 static void lapic_timer_state_broadcast(struct acpi_processor *pr, 165 struct acpi_processor_cx *cx, 166 int broadcast) 167 { 168 int state = cx - pr->power.states; 169 170 if (state >= pr->power.timer_broadcast_on_state) { 171 if (broadcast) 172 tick_broadcast_enter(); 173 else 174 tick_broadcast_exit(); 175 } 176 } 177 178 #else 179 180 static void lapic_timer_check_state(int state, struct acpi_processor *pr, 181 struct acpi_processor_cx *cstate) { } 182 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { } 183 static void lapic_timer_state_broadcast(struct acpi_processor *pr, 184 struct acpi_processor_cx *cx, 185 int broadcast) 186 { 187 } 188 189 #endif 190 191 #if defined(CONFIG_X86) 192 static void tsc_check_state(int state) 193 { 194 switch (boot_cpu_data.x86_vendor) { 195 case X86_VENDOR_HYGON: 196 case X86_VENDOR_AMD: 197 case X86_VENDOR_INTEL: 198 case X86_VENDOR_CENTAUR: 199 case X86_VENDOR_ZHAOXIN: 200 /* 201 * AMD Fam10h TSC will tick in all 202 * C/P/S0/S1 states when this bit is set. 203 */ 204 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 205 return; 206 207 /*FALL THROUGH*/ 208 default: 209 /* TSC could halt in idle, so notify users */ 210 if (state > ACPI_STATE_C1) 211 mark_tsc_unstable("TSC halts in idle"); 212 } 213 } 214 #else 215 static void tsc_check_state(int state) { return; } 216 #endif 217 218 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr) 219 { 220 221 if (!pr->pblk) 222 return -ENODEV; 223 224 /* if info is obtained from pblk/fadt, type equals state */ 225 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2; 226 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3; 227 228 #ifndef CONFIG_HOTPLUG_CPU 229 /* 230 * Check for P_LVL2_UP flag before entering C2 and above on 231 * an SMP system. 232 */ 233 if ((num_online_cpus() > 1) && 234 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) 235 return -ENODEV; 236 #endif 237 238 /* determine C2 and C3 address from pblk */ 239 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4; 240 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5; 241 242 /* determine latencies from FADT */ 243 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency; 244 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency; 245 246 /* 247 * FADT specified C2 latency must be less than or equal to 248 * 100 microseconds. 249 */ 250 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) { 251 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 252 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency)); 253 /* invalidate C2 */ 254 pr->power.states[ACPI_STATE_C2].address = 0; 255 } 256 257 /* 258 * FADT supplied C3 latency must be less than or equal to 259 * 1000 microseconds. 260 */ 261 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) { 262 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 263 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency)); 264 /* invalidate C3 */ 265 pr->power.states[ACPI_STATE_C3].address = 0; 266 } 267 268 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 269 "lvl2[0x%08x] lvl3[0x%08x]\n", 270 pr->power.states[ACPI_STATE_C2].address, 271 pr->power.states[ACPI_STATE_C3].address)); 272 273 snprintf(pr->power.states[ACPI_STATE_C2].desc, 274 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x", 275 pr->power.states[ACPI_STATE_C2].address); 276 snprintf(pr->power.states[ACPI_STATE_C3].desc, 277 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x", 278 pr->power.states[ACPI_STATE_C3].address); 279 280 return 0; 281 } 282 283 static int acpi_processor_get_power_info_default(struct acpi_processor *pr) 284 { 285 if (!pr->power.states[ACPI_STATE_C1].valid) { 286 /* set the first C-State to C1 */ 287 /* all processors need to support C1 */ 288 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1; 289 pr->power.states[ACPI_STATE_C1].valid = 1; 290 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT; 291 292 snprintf(pr->power.states[ACPI_STATE_C1].desc, 293 ACPI_CX_DESC_LEN, "ACPI HLT"); 294 } 295 /* the C0 state only exists as a filler in our array */ 296 pr->power.states[ACPI_STATE_C0].valid = 1; 297 return 0; 298 } 299 300 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr) 301 { 302 int ret; 303 304 if (nocst) 305 return -ENODEV; 306 307 ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power); 308 if (ret) 309 return ret; 310 311 if (!pr->power.count) 312 return -EFAULT; 313 314 pr->flags.has_cst = 1; 315 return 0; 316 } 317 318 static void acpi_processor_power_verify_c3(struct acpi_processor *pr, 319 struct acpi_processor_cx *cx) 320 { 321 static int bm_check_flag = -1; 322 static int bm_control_flag = -1; 323 324 325 if (!cx->address) 326 return; 327 328 /* 329 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast) 330 * DMA transfers are used by any ISA device to avoid livelock. 331 * Note that we could disable Type-F DMA (as recommended by 332 * the erratum), but this is known to disrupt certain ISA 333 * devices thus we take the conservative approach. 334 */ 335 else if (errata.piix4.fdma) { 336 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 337 "C3 not supported on PIIX4 with Type-F DMA\n")); 338 return; 339 } 340 341 /* All the logic here assumes flags.bm_check is same across all CPUs */ 342 if (bm_check_flag == -1) { 343 /* Determine whether bm_check is needed based on CPU */ 344 acpi_processor_power_init_bm_check(&(pr->flags), pr->id); 345 bm_check_flag = pr->flags.bm_check; 346 bm_control_flag = pr->flags.bm_control; 347 } else { 348 pr->flags.bm_check = bm_check_flag; 349 pr->flags.bm_control = bm_control_flag; 350 } 351 352 if (pr->flags.bm_check) { 353 if (!pr->flags.bm_control) { 354 if (pr->flags.has_cst != 1) { 355 /* bus mastering control is necessary */ 356 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 357 "C3 support requires BM control\n")); 358 return; 359 } else { 360 /* Here we enter C3 without bus mastering */ 361 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 362 "C3 support without BM control\n")); 363 } 364 } 365 } else { 366 /* 367 * WBINVD should be set in fadt, for C3 state to be 368 * supported on when bm_check is not required. 369 */ 370 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) { 371 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 372 "Cache invalidation should work properly" 373 " for C3 to be enabled on SMP systems\n")); 374 return; 375 } 376 } 377 378 /* 379 * Otherwise we've met all of our C3 requirements. 380 * Normalize the C3 latency to expidite policy. Enable 381 * checking of bus mastering status (bm_check) so we can 382 * use this in our C3 policy 383 */ 384 cx->valid = 1; 385 386 /* 387 * On older chipsets, BM_RLD needs to be set 388 * in order for Bus Master activity to wake the 389 * system from C3. Newer chipsets handle DMA 390 * during C3 automatically and BM_RLD is a NOP. 391 * In either case, the proper way to 392 * handle BM_RLD is to set it and leave it set. 393 */ 394 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1); 395 396 return; 397 } 398 399 static int acpi_processor_power_verify(struct acpi_processor *pr) 400 { 401 unsigned int i; 402 unsigned int working = 0; 403 404 pr->power.timer_broadcast_on_state = INT_MAX; 405 406 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 407 struct acpi_processor_cx *cx = &pr->power.states[i]; 408 409 switch (cx->type) { 410 case ACPI_STATE_C1: 411 cx->valid = 1; 412 break; 413 414 case ACPI_STATE_C2: 415 if (!cx->address) 416 break; 417 cx->valid = 1; 418 break; 419 420 case ACPI_STATE_C3: 421 acpi_processor_power_verify_c3(pr, cx); 422 break; 423 } 424 if (!cx->valid) 425 continue; 426 427 lapic_timer_check_state(i, pr, cx); 428 tsc_check_state(cx->type); 429 working++; 430 } 431 432 lapic_timer_propagate_broadcast(pr); 433 434 return (working); 435 } 436 437 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 438 { 439 unsigned int i; 440 int result; 441 442 443 /* NOTE: the idle thread may not be running while calling 444 * this function */ 445 446 /* Zero initialize all the C-states info. */ 447 memset(pr->power.states, 0, sizeof(pr->power.states)); 448 449 result = acpi_processor_get_power_info_cst(pr); 450 if (result == -ENODEV) 451 result = acpi_processor_get_power_info_fadt(pr); 452 453 if (result) 454 return result; 455 456 acpi_processor_get_power_info_default(pr); 457 458 pr->power.count = acpi_processor_power_verify(pr); 459 460 /* 461 * if one state of type C2 or C3 is available, mark this 462 * CPU as being "idle manageable" 463 */ 464 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) { 465 if (pr->power.states[i].valid) { 466 pr->power.count = i; 467 pr->flags.power = 1; 468 } 469 } 470 471 return 0; 472 } 473 474 /** 475 * acpi_idle_bm_check - checks if bus master activity was detected 476 */ 477 static int acpi_idle_bm_check(void) 478 { 479 u32 bm_status = 0; 480 481 if (bm_check_disable) 482 return 0; 483 484 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status); 485 if (bm_status) 486 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1); 487 /* 488 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect 489 * the true state of bus mastering activity; forcing us to 490 * manually check the BMIDEA bit of each IDE channel. 491 */ 492 else if (errata.piix4.bmisx) { 493 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01) 494 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01)) 495 bm_status = 1; 496 } 497 return bm_status; 498 } 499 500 static void wait_for_freeze(void) 501 { 502 #ifdef CONFIG_X86 503 /* No delay is needed if we are in guest */ 504 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 505 return; 506 #endif 507 /* Dummy wait op - must do something useless after P_LVL2 read 508 because chipsets cannot guarantee that STPCLK# signal 509 gets asserted in time to freeze execution properly. */ 510 inl(acpi_gbl_FADT.xpm_timer_block.address); 511 } 512 513 /** 514 * acpi_idle_do_entry - enter idle state using the appropriate method 515 * @cx: cstate data 516 * 517 * Caller disables interrupt before call and enables interrupt after return. 518 */ 519 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx) 520 { 521 if (cx->entry_method == ACPI_CSTATE_FFH) { 522 /* Call into architectural FFH based C-state */ 523 acpi_processor_ffh_cstate_enter(cx); 524 } else if (cx->entry_method == ACPI_CSTATE_HALT) { 525 acpi_safe_halt(); 526 } else { 527 /* IO port based C-state */ 528 inb(cx->address); 529 wait_for_freeze(); 530 } 531 } 532 533 /** 534 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining) 535 * @dev: the target CPU 536 * @index: the index of suggested state 537 */ 538 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index) 539 { 540 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 541 542 ACPI_FLUSH_CPU_CACHE(); 543 544 while (1) { 545 546 if (cx->entry_method == ACPI_CSTATE_HALT) 547 safe_halt(); 548 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) { 549 inb(cx->address); 550 wait_for_freeze(); 551 } else 552 return -ENODEV; 553 } 554 555 /* Never reached */ 556 return 0; 557 } 558 559 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr) 560 { 561 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst && 562 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED); 563 } 564 565 static int c3_cpu_count; 566 static DEFINE_RAW_SPINLOCK(c3_lock); 567 568 /** 569 * acpi_idle_enter_bm - enters C3 with proper BM handling 570 * @pr: Target processor 571 * @cx: Target state context 572 * @timer_bc: Whether or not to change timer mode to broadcast 573 */ 574 static void acpi_idle_enter_bm(struct acpi_processor *pr, 575 struct acpi_processor_cx *cx, bool timer_bc) 576 { 577 acpi_unlazy_tlb(smp_processor_id()); 578 579 /* 580 * Must be done before busmaster disable as we might need to 581 * access HPET ! 582 */ 583 if (timer_bc) 584 lapic_timer_state_broadcast(pr, cx, 1); 585 586 /* 587 * disable bus master 588 * bm_check implies we need ARB_DIS 589 * bm_control implies whether we can do ARB_DIS 590 * 591 * That leaves a case where bm_check is set and bm_control is 592 * not set. In that case we cannot do much, we enter C3 593 * without doing anything. 594 */ 595 if (pr->flags.bm_control) { 596 raw_spin_lock(&c3_lock); 597 c3_cpu_count++; 598 /* Disable bus master arbitration when all CPUs are in C3 */ 599 if (c3_cpu_count == num_online_cpus()) 600 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1); 601 raw_spin_unlock(&c3_lock); 602 } 603 604 acpi_idle_do_entry(cx); 605 606 /* Re-enable bus master arbitration */ 607 if (pr->flags.bm_control) { 608 raw_spin_lock(&c3_lock); 609 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0); 610 c3_cpu_count--; 611 raw_spin_unlock(&c3_lock); 612 } 613 614 if (timer_bc) 615 lapic_timer_state_broadcast(pr, cx, 0); 616 } 617 618 static int acpi_idle_enter(struct cpuidle_device *dev, 619 struct cpuidle_driver *drv, int index) 620 { 621 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 622 struct acpi_processor *pr; 623 624 pr = __this_cpu_read(processors); 625 if (unlikely(!pr)) 626 return -EINVAL; 627 628 if (cx->type != ACPI_STATE_C1) { 629 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) { 630 index = ACPI_IDLE_STATE_START; 631 cx = per_cpu(acpi_cstate[index], dev->cpu); 632 } else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) { 633 if (cx->bm_sts_skip || !acpi_idle_bm_check()) { 634 acpi_idle_enter_bm(pr, cx, true); 635 return index; 636 } else if (drv->safe_state_index >= 0) { 637 index = drv->safe_state_index; 638 cx = per_cpu(acpi_cstate[index], dev->cpu); 639 } else { 640 acpi_safe_halt(); 641 return -EBUSY; 642 } 643 } 644 } 645 646 lapic_timer_state_broadcast(pr, cx, 1); 647 648 if (cx->type == ACPI_STATE_C3) 649 ACPI_FLUSH_CPU_CACHE(); 650 651 acpi_idle_do_entry(cx); 652 653 lapic_timer_state_broadcast(pr, cx, 0); 654 655 return index; 656 } 657 658 static void acpi_idle_enter_s2idle(struct cpuidle_device *dev, 659 struct cpuidle_driver *drv, int index) 660 { 661 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 662 663 if (cx->type == ACPI_STATE_C3) { 664 struct acpi_processor *pr = __this_cpu_read(processors); 665 666 if (unlikely(!pr)) 667 return; 668 669 if (pr->flags.bm_check) { 670 acpi_idle_enter_bm(pr, cx, false); 671 return; 672 } else { 673 ACPI_FLUSH_CPU_CACHE(); 674 } 675 } 676 acpi_idle_do_entry(cx); 677 } 678 679 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 680 struct cpuidle_device *dev) 681 { 682 int i, count = ACPI_IDLE_STATE_START; 683 struct acpi_processor_cx *cx; 684 685 if (max_cstate == 0) 686 max_cstate = 1; 687 688 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 689 cx = &pr->power.states[i]; 690 691 if (!cx->valid) 692 continue; 693 694 per_cpu(acpi_cstate[count], dev->cpu) = cx; 695 696 count++; 697 if (count == CPUIDLE_STATE_MAX) 698 break; 699 } 700 701 if (!count) 702 return -EINVAL; 703 704 return 0; 705 } 706 707 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 708 { 709 int i, count; 710 struct acpi_processor_cx *cx; 711 struct cpuidle_state *state; 712 struct cpuidle_driver *drv = &acpi_idle_driver; 713 714 if (max_cstate == 0) 715 max_cstate = 1; 716 717 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) { 718 cpuidle_poll_state_init(drv); 719 count = 1; 720 } else { 721 count = 0; 722 } 723 724 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 725 cx = &pr->power.states[i]; 726 727 if (!cx->valid) 728 continue; 729 730 state = &drv->states[count]; 731 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i); 732 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN); 733 state->exit_latency = cx->latency; 734 state->target_residency = cx->latency * latency_factor; 735 state->enter = acpi_idle_enter; 736 737 state->flags = 0; 738 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { 739 state->enter_dead = acpi_idle_play_dead; 740 drv->safe_state_index = count; 741 } 742 /* 743 * Halt-induced C1 is not good for ->enter_s2idle, because it 744 * re-enables interrupts on exit. Moreover, C1 is generally not 745 * particularly interesting from the suspend-to-idle angle, so 746 * avoid C1 and the situations in which we may need to fall back 747 * to it altogether. 748 */ 749 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr)) 750 state->enter_s2idle = acpi_idle_enter_s2idle; 751 752 count++; 753 if (count == CPUIDLE_STATE_MAX) 754 break; 755 } 756 757 drv->state_count = count; 758 759 if (!count) 760 return -EINVAL; 761 762 return 0; 763 } 764 765 static inline void acpi_processor_cstate_first_run_checks(void) 766 { 767 static int first_run; 768 769 if (first_run) 770 return; 771 dmi_check_system(processor_power_dmi_table); 772 max_cstate = acpi_processor_cstate_check(max_cstate); 773 if (max_cstate < ACPI_C_STATES_MAX) 774 pr_notice("ACPI: processor limited to max C-state %d\n", 775 max_cstate); 776 first_run++; 777 778 if (nocst) 779 return; 780 781 acpi_processor_claim_cst_control(); 782 } 783 #else 784 785 static inline int disabled_by_idle_boot_param(void) { return 0; } 786 static inline void acpi_processor_cstate_first_run_checks(void) { } 787 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 788 { 789 return -ENODEV; 790 } 791 792 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 793 struct cpuidle_device *dev) 794 { 795 return -EINVAL; 796 } 797 798 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 799 { 800 return -EINVAL; 801 } 802 803 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */ 804 805 struct acpi_lpi_states_array { 806 unsigned int size; 807 unsigned int composite_states_size; 808 struct acpi_lpi_state *entries; 809 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER]; 810 }; 811 812 static int obj_get_integer(union acpi_object *obj, u32 *value) 813 { 814 if (obj->type != ACPI_TYPE_INTEGER) 815 return -EINVAL; 816 817 *value = obj->integer.value; 818 return 0; 819 } 820 821 static int acpi_processor_evaluate_lpi(acpi_handle handle, 822 struct acpi_lpi_states_array *info) 823 { 824 acpi_status status; 825 int ret = 0; 826 int pkg_count, state_idx = 1, loop; 827 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; 828 union acpi_object *lpi_data; 829 struct acpi_lpi_state *lpi_state; 830 831 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer); 832 if (ACPI_FAILURE(status)) { 833 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n")); 834 return -ENODEV; 835 } 836 837 lpi_data = buffer.pointer; 838 839 /* There must be at least 4 elements = 3 elements + 1 package */ 840 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE || 841 lpi_data->package.count < 4) { 842 pr_debug("not enough elements in _LPI\n"); 843 ret = -ENODATA; 844 goto end; 845 } 846 847 pkg_count = lpi_data->package.elements[2].integer.value; 848 849 /* Validate number of power states. */ 850 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) { 851 pr_debug("count given by _LPI is not valid\n"); 852 ret = -ENODATA; 853 goto end; 854 } 855 856 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL); 857 if (!lpi_state) { 858 ret = -ENOMEM; 859 goto end; 860 } 861 862 info->size = pkg_count; 863 info->entries = lpi_state; 864 865 /* LPI States start at index 3 */ 866 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) { 867 union acpi_object *element, *pkg_elem, *obj; 868 869 element = &lpi_data->package.elements[loop]; 870 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7) 871 continue; 872 873 pkg_elem = element->package.elements; 874 875 obj = pkg_elem + 6; 876 if (obj->type == ACPI_TYPE_BUFFER) { 877 struct acpi_power_register *reg; 878 879 reg = (struct acpi_power_register *)obj->buffer.pointer; 880 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO && 881 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) 882 continue; 883 884 lpi_state->address = reg->address; 885 lpi_state->entry_method = 886 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ? 887 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO; 888 } else if (obj->type == ACPI_TYPE_INTEGER) { 889 lpi_state->entry_method = ACPI_CSTATE_INTEGER; 890 lpi_state->address = obj->integer.value; 891 } else { 892 continue; 893 } 894 895 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/ 896 897 obj = pkg_elem + 9; 898 if (obj->type == ACPI_TYPE_STRING) 899 strlcpy(lpi_state->desc, obj->string.pointer, 900 ACPI_CX_DESC_LEN); 901 902 lpi_state->index = state_idx; 903 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) { 904 pr_debug("No min. residency found, assuming 10 us\n"); 905 lpi_state->min_residency = 10; 906 } 907 908 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) { 909 pr_debug("No wakeup residency found, assuming 10 us\n"); 910 lpi_state->wake_latency = 10; 911 } 912 913 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags)) 914 lpi_state->flags = 0; 915 916 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags)) 917 lpi_state->arch_flags = 0; 918 919 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq)) 920 lpi_state->res_cnt_freq = 1; 921 922 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state)) 923 lpi_state->enable_parent_state = 0; 924 } 925 926 acpi_handle_debug(handle, "Found %d power states\n", state_idx); 927 end: 928 kfree(buffer.pointer); 929 return ret; 930 } 931 932 /* 933 * flat_state_cnt - the number of composite LPI states after the process of flattening 934 */ 935 static int flat_state_cnt; 936 937 /** 938 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state 939 * 940 * @local: local LPI state 941 * @parent: parent LPI state 942 * @result: composite LPI state 943 */ 944 static bool combine_lpi_states(struct acpi_lpi_state *local, 945 struct acpi_lpi_state *parent, 946 struct acpi_lpi_state *result) 947 { 948 if (parent->entry_method == ACPI_CSTATE_INTEGER) { 949 if (!parent->address) /* 0 means autopromotable */ 950 return false; 951 result->address = local->address + parent->address; 952 } else { 953 result->address = parent->address; 954 } 955 956 result->min_residency = max(local->min_residency, parent->min_residency); 957 result->wake_latency = local->wake_latency + parent->wake_latency; 958 result->enable_parent_state = parent->enable_parent_state; 959 result->entry_method = local->entry_method; 960 961 result->flags = parent->flags; 962 result->arch_flags = parent->arch_flags; 963 result->index = parent->index; 964 965 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN); 966 strlcat(result->desc, "+", ACPI_CX_DESC_LEN); 967 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN); 968 return true; 969 } 970 971 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0) 972 973 static void stash_composite_state(struct acpi_lpi_states_array *curr_level, 974 struct acpi_lpi_state *t) 975 { 976 curr_level->composite_states[curr_level->composite_states_size++] = t; 977 } 978 979 static int flatten_lpi_states(struct acpi_processor *pr, 980 struct acpi_lpi_states_array *curr_level, 981 struct acpi_lpi_states_array *prev_level) 982 { 983 int i, j, state_count = curr_level->size; 984 struct acpi_lpi_state *p, *t = curr_level->entries; 985 986 curr_level->composite_states_size = 0; 987 for (j = 0; j < state_count; j++, t++) { 988 struct acpi_lpi_state *flpi; 989 990 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED)) 991 continue; 992 993 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) { 994 pr_warn("Limiting number of LPI states to max (%d)\n", 995 ACPI_PROCESSOR_MAX_POWER); 996 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n"); 997 break; 998 } 999 1000 flpi = &pr->power.lpi_states[flat_state_cnt]; 1001 1002 if (!prev_level) { /* leaf/processor node */ 1003 memcpy(flpi, t, sizeof(*t)); 1004 stash_composite_state(curr_level, flpi); 1005 flat_state_cnt++; 1006 continue; 1007 } 1008 1009 for (i = 0; i < prev_level->composite_states_size; i++) { 1010 p = prev_level->composite_states[i]; 1011 if (t->index <= p->enable_parent_state && 1012 combine_lpi_states(p, t, flpi)) { 1013 stash_composite_state(curr_level, flpi); 1014 flat_state_cnt++; 1015 flpi++; 1016 } 1017 } 1018 } 1019 1020 kfree(curr_level->entries); 1021 return 0; 1022 } 1023 1024 static int acpi_processor_get_lpi_info(struct acpi_processor *pr) 1025 { 1026 int ret, i; 1027 acpi_status status; 1028 acpi_handle handle = pr->handle, pr_ahandle; 1029 struct acpi_device *d = NULL; 1030 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr; 1031 1032 if (!osc_pc_lpi_support_confirmed) 1033 return -EOPNOTSUPP; 1034 1035 if (!acpi_has_method(handle, "_LPI")) 1036 return -EINVAL; 1037 1038 flat_state_cnt = 0; 1039 prev = &info[0]; 1040 curr = &info[1]; 1041 handle = pr->handle; 1042 ret = acpi_processor_evaluate_lpi(handle, prev); 1043 if (ret) 1044 return ret; 1045 flatten_lpi_states(pr, prev, NULL); 1046 1047 status = acpi_get_parent(handle, &pr_ahandle); 1048 while (ACPI_SUCCESS(status)) { 1049 acpi_bus_get_device(pr_ahandle, &d); 1050 handle = pr_ahandle; 1051 1052 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID)) 1053 break; 1054 1055 /* can be optional ? */ 1056 if (!acpi_has_method(handle, "_LPI")) 1057 break; 1058 1059 ret = acpi_processor_evaluate_lpi(handle, curr); 1060 if (ret) 1061 break; 1062 1063 /* flatten all the LPI states in this level of hierarchy */ 1064 flatten_lpi_states(pr, curr, prev); 1065 1066 tmp = prev, prev = curr, curr = tmp; 1067 1068 status = acpi_get_parent(handle, &pr_ahandle); 1069 } 1070 1071 pr->power.count = flat_state_cnt; 1072 /* reset the index after flattening */ 1073 for (i = 0; i < pr->power.count; i++) 1074 pr->power.lpi_states[i].index = i; 1075 1076 /* Tell driver that _LPI is supported. */ 1077 pr->flags.has_lpi = 1; 1078 pr->flags.power = 1; 1079 1080 return 0; 1081 } 1082 1083 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu) 1084 { 1085 return -ENODEV; 1086 } 1087 1088 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi) 1089 { 1090 return -ENODEV; 1091 } 1092 1093 /** 1094 * acpi_idle_lpi_enter - enters an ACPI any LPI state 1095 * @dev: the target CPU 1096 * @drv: cpuidle driver containing cpuidle state info 1097 * @index: index of target state 1098 * 1099 * Return: 0 for success or negative value for error 1100 */ 1101 static int acpi_idle_lpi_enter(struct cpuidle_device *dev, 1102 struct cpuidle_driver *drv, int index) 1103 { 1104 struct acpi_processor *pr; 1105 struct acpi_lpi_state *lpi; 1106 1107 pr = __this_cpu_read(processors); 1108 1109 if (unlikely(!pr)) 1110 return -EINVAL; 1111 1112 lpi = &pr->power.lpi_states[index]; 1113 if (lpi->entry_method == ACPI_CSTATE_FFH) 1114 return acpi_processor_ffh_lpi_enter(lpi); 1115 1116 return -EINVAL; 1117 } 1118 1119 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr) 1120 { 1121 int i; 1122 struct acpi_lpi_state *lpi; 1123 struct cpuidle_state *state; 1124 struct cpuidle_driver *drv = &acpi_idle_driver; 1125 1126 if (!pr->flags.has_lpi) 1127 return -EOPNOTSUPP; 1128 1129 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) { 1130 lpi = &pr->power.lpi_states[i]; 1131 1132 state = &drv->states[i]; 1133 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i); 1134 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN); 1135 state->exit_latency = lpi->wake_latency; 1136 state->target_residency = lpi->min_residency; 1137 if (lpi->arch_flags) 1138 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 1139 state->enter = acpi_idle_lpi_enter; 1140 drv->safe_state_index = i; 1141 } 1142 1143 drv->state_count = i; 1144 1145 return 0; 1146 } 1147 1148 /** 1149 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle 1150 * global state data i.e. idle routines 1151 * 1152 * @pr: the ACPI processor 1153 */ 1154 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr) 1155 { 1156 int i; 1157 struct cpuidle_driver *drv = &acpi_idle_driver; 1158 1159 if (!pr->flags.power_setup_done || !pr->flags.power) 1160 return -EINVAL; 1161 1162 drv->safe_state_index = -1; 1163 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) { 1164 drv->states[i].name[0] = '\0'; 1165 drv->states[i].desc[0] = '\0'; 1166 } 1167 1168 if (pr->flags.has_lpi) 1169 return acpi_processor_setup_lpi_states(pr); 1170 1171 return acpi_processor_setup_cstates(pr); 1172 } 1173 1174 /** 1175 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE 1176 * device i.e. per-cpu data 1177 * 1178 * @pr: the ACPI processor 1179 * @dev : the cpuidle device 1180 */ 1181 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr, 1182 struct cpuidle_device *dev) 1183 { 1184 if (!pr->flags.power_setup_done || !pr->flags.power || !dev) 1185 return -EINVAL; 1186 1187 dev->cpu = pr->id; 1188 if (pr->flags.has_lpi) 1189 return acpi_processor_ffh_lpi_probe(pr->id); 1190 1191 return acpi_processor_setup_cpuidle_cx(pr, dev); 1192 } 1193 1194 static int acpi_processor_get_power_info(struct acpi_processor *pr) 1195 { 1196 int ret; 1197 1198 ret = acpi_processor_get_lpi_info(pr); 1199 if (ret) 1200 ret = acpi_processor_get_cstate_info(pr); 1201 1202 return ret; 1203 } 1204 1205 int acpi_processor_hotplug(struct acpi_processor *pr) 1206 { 1207 int ret = 0; 1208 struct cpuidle_device *dev; 1209 1210 if (disabled_by_idle_boot_param()) 1211 return 0; 1212 1213 if (!pr->flags.power_setup_done) 1214 return -ENODEV; 1215 1216 dev = per_cpu(acpi_cpuidle_device, pr->id); 1217 cpuidle_pause_and_lock(); 1218 cpuidle_disable_device(dev); 1219 ret = acpi_processor_get_power_info(pr); 1220 if (!ret && pr->flags.power) { 1221 acpi_processor_setup_cpuidle_dev(pr, dev); 1222 ret = cpuidle_enable_device(dev); 1223 } 1224 cpuidle_resume_and_unlock(); 1225 1226 return ret; 1227 } 1228 1229 int acpi_processor_power_state_has_changed(struct acpi_processor *pr) 1230 { 1231 int cpu; 1232 struct acpi_processor *_pr; 1233 struct cpuidle_device *dev; 1234 1235 if (disabled_by_idle_boot_param()) 1236 return 0; 1237 1238 if (!pr->flags.power_setup_done) 1239 return -ENODEV; 1240 1241 /* 1242 * FIXME: Design the ACPI notification to make it once per 1243 * system instead of once per-cpu. This condition is a hack 1244 * to make the code that updates C-States be called once. 1245 */ 1246 1247 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) { 1248 1249 /* Protect against cpu-hotplug */ 1250 get_online_cpus(); 1251 cpuidle_pause_and_lock(); 1252 1253 /* Disable all cpuidle devices */ 1254 for_each_online_cpu(cpu) { 1255 _pr = per_cpu(processors, cpu); 1256 if (!_pr || !_pr->flags.power_setup_done) 1257 continue; 1258 dev = per_cpu(acpi_cpuidle_device, cpu); 1259 cpuidle_disable_device(dev); 1260 } 1261 1262 /* Populate Updated C-state information */ 1263 acpi_processor_get_power_info(pr); 1264 acpi_processor_setup_cpuidle_states(pr); 1265 1266 /* Enable all cpuidle devices */ 1267 for_each_online_cpu(cpu) { 1268 _pr = per_cpu(processors, cpu); 1269 if (!_pr || !_pr->flags.power_setup_done) 1270 continue; 1271 acpi_processor_get_power_info(_pr); 1272 if (_pr->flags.power) { 1273 dev = per_cpu(acpi_cpuidle_device, cpu); 1274 acpi_processor_setup_cpuidle_dev(_pr, dev); 1275 cpuidle_enable_device(dev); 1276 } 1277 } 1278 cpuidle_resume_and_unlock(); 1279 put_online_cpus(); 1280 } 1281 1282 return 0; 1283 } 1284 1285 static int acpi_processor_registered; 1286 1287 int acpi_processor_power_init(struct acpi_processor *pr) 1288 { 1289 int retval; 1290 struct cpuidle_device *dev; 1291 1292 if (disabled_by_idle_boot_param()) 1293 return 0; 1294 1295 acpi_processor_cstate_first_run_checks(); 1296 1297 if (!acpi_processor_get_power_info(pr)) 1298 pr->flags.power_setup_done = 1; 1299 1300 /* 1301 * Install the idle handler if processor power management is supported. 1302 * Note that we use previously set idle handler will be used on 1303 * platforms that only support C1. 1304 */ 1305 if (pr->flags.power) { 1306 /* Register acpi_idle_driver if not already registered */ 1307 if (!acpi_processor_registered) { 1308 acpi_processor_setup_cpuidle_states(pr); 1309 retval = cpuidle_register_driver(&acpi_idle_driver); 1310 if (retval) 1311 return retval; 1312 pr_debug("%s registered with cpuidle\n", 1313 acpi_idle_driver.name); 1314 } 1315 1316 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1317 if (!dev) 1318 return -ENOMEM; 1319 per_cpu(acpi_cpuidle_device, pr->id) = dev; 1320 1321 acpi_processor_setup_cpuidle_dev(pr, dev); 1322 1323 /* Register per-cpu cpuidle_device. Cpuidle driver 1324 * must already be registered before registering device 1325 */ 1326 retval = cpuidle_register_device(dev); 1327 if (retval) { 1328 if (acpi_processor_registered == 0) 1329 cpuidle_unregister_driver(&acpi_idle_driver); 1330 return retval; 1331 } 1332 acpi_processor_registered++; 1333 } 1334 return 0; 1335 } 1336 1337 int acpi_processor_power_exit(struct acpi_processor *pr) 1338 { 1339 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id); 1340 1341 if (disabled_by_idle_boot_param()) 1342 return 0; 1343 1344 if (pr->flags.power) { 1345 cpuidle_unregister_device(dev); 1346 acpi_processor_registered--; 1347 if (acpi_processor_registered == 0) 1348 cpuidle_unregister_driver(&acpi_idle_driver); 1349 } 1350 1351 pr->flags.power_setup_done = 0; 1352 return 0; 1353 } 1354