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 fallthrough; 207 default: 208 /* TSC could halt in idle, so notify users */ 209 if (state > ACPI_STATE_C1) 210 mark_tsc_unstable("TSC halts in idle"); 211 } 212 } 213 #else 214 static void tsc_check_state(int state) { return; } 215 #endif 216 217 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr) 218 { 219 220 if (!pr->pblk) 221 return -ENODEV; 222 223 /* if info is obtained from pblk/fadt, type equals state */ 224 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2; 225 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3; 226 227 #ifndef CONFIG_HOTPLUG_CPU 228 /* 229 * Check for P_LVL2_UP flag before entering C2 and above on 230 * an SMP system. 231 */ 232 if ((num_online_cpus() > 1) && 233 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) 234 return -ENODEV; 235 #endif 236 237 /* determine C2 and C3 address from pblk */ 238 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4; 239 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5; 240 241 /* determine latencies from FADT */ 242 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency; 243 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency; 244 245 /* 246 * FADT specified C2 latency must be less than or equal to 247 * 100 microseconds. 248 */ 249 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) { 250 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 251 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency)); 252 /* invalidate C2 */ 253 pr->power.states[ACPI_STATE_C2].address = 0; 254 } 255 256 /* 257 * FADT supplied C3 latency must be less than or equal to 258 * 1000 microseconds. 259 */ 260 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) { 261 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 262 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency)); 263 /* invalidate C3 */ 264 pr->power.states[ACPI_STATE_C3].address = 0; 265 } 266 267 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 268 "lvl2[0x%08x] lvl3[0x%08x]\n", 269 pr->power.states[ACPI_STATE_C2].address, 270 pr->power.states[ACPI_STATE_C3].address)); 271 272 snprintf(pr->power.states[ACPI_STATE_C2].desc, 273 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x", 274 pr->power.states[ACPI_STATE_C2].address); 275 snprintf(pr->power.states[ACPI_STATE_C3].desc, 276 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x", 277 pr->power.states[ACPI_STATE_C3].address); 278 279 return 0; 280 } 281 282 static int acpi_processor_get_power_info_default(struct acpi_processor *pr) 283 { 284 if (!pr->power.states[ACPI_STATE_C1].valid) { 285 /* set the first C-State to C1 */ 286 /* all processors need to support C1 */ 287 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1; 288 pr->power.states[ACPI_STATE_C1].valid = 1; 289 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT; 290 291 snprintf(pr->power.states[ACPI_STATE_C1].desc, 292 ACPI_CX_DESC_LEN, "ACPI HLT"); 293 } 294 /* the C0 state only exists as a filler in our array */ 295 pr->power.states[ACPI_STATE_C0].valid = 1; 296 return 0; 297 } 298 299 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr) 300 { 301 int ret; 302 303 if (nocst) 304 return -ENODEV; 305 306 ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power); 307 if (ret) 308 return ret; 309 310 if (!pr->power.count) 311 return -EFAULT; 312 313 pr->flags.has_cst = 1; 314 return 0; 315 } 316 317 static void acpi_processor_power_verify_c3(struct acpi_processor *pr, 318 struct acpi_processor_cx *cx) 319 { 320 static int bm_check_flag = -1; 321 static int bm_control_flag = -1; 322 323 324 if (!cx->address) 325 return; 326 327 /* 328 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast) 329 * DMA transfers are used by any ISA device to avoid livelock. 330 * Note that we could disable Type-F DMA (as recommended by 331 * the erratum), but this is known to disrupt certain ISA 332 * devices thus we take the conservative approach. 333 */ 334 else if (errata.piix4.fdma) { 335 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 336 "C3 not supported on PIIX4 with Type-F DMA\n")); 337 return; 338 } 339 340 /* All the logic here assumes flags.bm_check is same across all CPUs */ 341 if (bm_check_flag == -1) { 342 /* Determine whether bm_check is needed based on CPU */ 343 acpi_processor_power_init_bm_check(&(pr->flags), pr->id); 344 bm_check_flag = pr->flags.bm_check; 345 bm_control_flag = pr->flags.bm_control; 346 } else { 347 pr->flags.bm_check = bm_check_flag; 348 pr->flags.bm_control = bm_control_flag; 349 } 350 351 if (pr->flags.bm_check) { 352 if (!pr->flags.bm_control) { 353 if (pr->flags.has_cst != 1) { 354 /* bus mastering control is necessary */ 355 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 356 "C3 support requires BM control\n")); 357 return; 358 } else { 359 /* Here we enter C3 without bus mastering */ 360 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 361 "C3 support without BM control\n")); 362 } 363 } 364 } else { 365 /* 366 * WBINVD should be set in fadt, for C3 state to be 367 * supported on when bm_check is not required. 368 */ 369 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) { 370 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 371 "Cache invalidation should work properly" 372 " for C3 to be enabled on SMP systems\n")); 373 return; 374 } 375 } 376 377 /* 378 * Otherwise we've met all of our C3 requirements. 379 * Normalize the C3 latency to expidite policy. Enable 380 * checking of bus mastering status (bm_check) so we can 381 * use this in our C3 policy 382 */ 383 cx->valid = 1; 384 385 /* 386 * On older chipsets, BM_RLD needs to be set 387 * in order for Bus Master activity to wake the 388 * system from C3. Newer chipsets handle DMA 389 * during C3 automatically and BM_RLD is a NOP. 390 * In either case, the proper way to 391 * handle BM_RLD is to set it and leave it set. 392 */ 393 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1); 394 395 return; 396 } 397 398 static int acpi_processor_power_verify(struct acpi_processor *pr) 399 { 400 unsigned int i; 401 unsigned int working = 0; 402 403 pr->power.timer_broadcast_on_state = INT_MAX; 404 405 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 406 struct acpi_processor_cx *cx = &pr->power.states[i]; 407 408 switch (cx->type) { 409 case ACPI_STATE_C1: 410 cx->valid = 1; 411 break; 412 413 case ACPI_STATE_C2: 414 if (!cx->address) 415 break; 416 cx->valid = 1; 417 break; 418 419 case ACPI_STATE_C3: 420 acpi_processor_power_verify_c3(pr, cx); 421 break; 422 } 423 if (!cx->valid) 424 continue; 425 426 lapic_timer_check_state(i, pr, cx); 427 tsc_check_state(cx->type); 428 working++; 429 } 430 431 lapic_timer_propagate_broadcast(pr); 432 433 return (working); 434 } 435 436 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 437 { 438 unsigned int i; 439 int result; 440 441 442 /* NOTE: the idle thread may not be running while calling 443 * this function */ 444 445 /* Zero initialize all the C-states info. */ 446 memset(pr->power.states, 0, sizeof(pr->power.states)); 447 448 result = acpi_processor_get_power_info_cst(pr); 449 if (result == -ENODEV) 450 result = acpi_processor_get_power_info_fadt(pr); 451 452 if (result) 453 return result; 454 455 acpi_processor_get_power_info_default(pr); 456 457 pr->power.count = acpi_processor_power_verify(pr); 458 459 /* 460 * if one state of type C2 or C3 is available, mark this 461 * CPU as being "idle manageable" 462 */ 463 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) { 464 if (pr->power.states[i].valid) { 465 pr->power.count = i; 466 pr->flags.power = 1; 467 } 468 } 469 470 return 0; 471 } 472 473 /** 474 * acpi_idle_bm_check - checks if bus master activity was detected 475 */ 476 static int acpi_idle_bm_check(void) 477 { 478 u32 bm_status = 0; 479 480 if (bm_check_disable) 481 return 0; 482 483 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status); 484 if (bm_status) 485 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1); 486 /* 487 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect 488 * the true state of bus mastering activity; forcing us to 489 * manually check the BMIDEA bit of each IDE channel. 490 */ 491 else if (errata.piix4.bmisx) { 492 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01) 493 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01)) 494 bm_status = 1; 495 } 496 return bm_status; 497 } 498 499 static void wait_for_freeze(void) 500 { 501 #ifdef CONFIG_X86 502 /* No delay is needed if we are in guest */ 503 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 504 return; 505 #endif 506 /* Dummy wait op - must do something useless after P_LVL2 read 507 because chipsets cannot guarantee that STPCLK# signal 508 gets asserted in time to freeze execution properly. */ 509 inl(acpi_gbl_FADT.xpm_timer_block.address); 510 } 511 512 /** 513 * acpi_idle_do_entry - enter idle state using the appropriate method 514 * @cx: cstate data 515 * 516 * Caller disables interrupt before call and enables interrupt after return. 517 */ 518 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx) 519 { 520 if (cx->entry_method == ACPI_CSTATE_FFH) { 521 /* Call into architectural FFH based C-state */ 522 acpi_processor_ffh_cstate_enter(cx); 523 } else if (cx->entry_method == ACPI_CSTATE_HALT) { 524 acpi_safe_halt(); 525 } else { 526 /* IO port based C-state */ 527 inb(cx->address); 528 wait_for_freeze(); 529 } 530 } 531 532 /** 533 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining) 534 * @dev: the target CPU 535 * @index: the index of suggested state 536 */ 537 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index) 538 { 539 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 540 541 ACPI_FLUSH_CPU_CACHE(); 542 543 while (1) { 544 545 if (cx->entry_method == ACPI_CSTATE_HALT) 546 safe_halt(); 547 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) { 548 inb(cx->address); 549 wait_for_freeze(); 550 } else 551 return -ENODEV; 552 } 553 554 /* Never reached */ 555 return 0; 556 } 557 558 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr) 559 { 560 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst && 561 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED); 562 } 563 564 static int c3_cpu_count; 565 static DEFINE_RAW_SPINLOCK(c3_lock); 566 567 /** 568 * acpi_idle_enter_bm - enters C3 with proper BM handling 569 * @pr: Target processor 570 * @cx: Target state context 571 * @timer_bc: Whether or not to change timer mode to broadcast 572 */ 573 static void acpi_idle_enter_bm(struct acpi_processor *pr, 574 struct acpi_processor_cx *cx, bool timer_bc) 575 { 576 acpi_unlazy_tlb(smp_processor_id()); 577 578 /* 579 * Must be done before busmaster disable as we might need to 580 * access HPET ! 581 */ 582 if (timer_bc) 583 lapic_timer_state_broadcast(pr, cx, 1); 584 585 /* 586 * disable bus master 587 * bm_check implies we need ARB_DIS 588 * bm_control implies whether we can do ARB_DIS 589 * 590 * That leaves a case where bm_check is set and bm_control is 591 * not set. In that case we cannot do much, we enter C3 592 * without doing anything. 593 */ 594 if (pr->flags.bm_control) { 595 raw_spin_lock(&c3_lock); 596 c3_cpu_count++; 597 /* Disable bus master arbitration when all CPUs are in C3 */ 598 if (c3_cpu_count == num_online_cpus()) 599 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1); 600 raw_spin_unlock(&c3_lock); 601 } 602 603 acpi_idle_do_entry(cx); 604 605 /* Re-enable bus master arbitration */ 606 if (pr->flags.bm_control) { 607 raw_spin_lock(&c3_lock); 608 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0); 609 c3_cpu_count--; 610 raw_spin_unlock(&c3_lock); 611 } 612 613 if (timer_bc) 614 lapic_timer_state_broadcast(pr, cx, 0); 615 } 616 617 static int acpi_idle_enter(struct cpuidle_device *dev, 618 struct cpuidle_driver *drv, int index) 619 { 620 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 621 struct acpi_processor *pr; 622 623 pr = __this_cpu_read(processors); 624 if (unlikely(!pr)) 625 return -EINVAL; 626 627 if (cx->type != ACPI_STATE_C1) { 628 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) { 629 index = ACPI_IDLE_STATE_START; 630 cx = per_cpu(acpi_cstate[index], dev->cpu); 631 } else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) { 632 if (cx->bm_sts_skip || !acpi_idle_bm_check()) { 633 acpi_idle_enter_bm(pr, cx, true); 634 return index; 635 } else if (drv->safe_state_index >= 0) { 636 index = drv->safe_state_index; 637 cx = per_cpu(acpi_cstate[index], dev->cpu); 638 } else { 639 acpi_safe_halt(); 640 return -EBUSY; 641 } 642 } 643 } 644 645 lapic_timer_state_broadcast(pr, cx, 1); 646 647 if (cx->type == ACPI_STATE_C3) 648 ACPI_FLUSH_CPU_CACHE(); 649 650 acpi_idle_do_entry(cx); 651 652 lapic_timer_state_broadcast(pr, cx, 0); 653 654 return index; 655 } 656 657 static int acpi_idle_enter_s2idle(struct cpuidle_device *dev, 658 struct cpuidle_driver *drv, int index) 659 { 660 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 661 662 if (cx->type == ACPI_STATE_C3) { 663 struct acpi_processor *pr = __this_cpu_read(processors); 664 665 if (unlikely(!pr)) 666 return 0; 667 668 if (pr->flags.bm_check) { 669 acpi_idle_enter_bm(pr, cx, false); 670 return 0; 671 } else { 672 ACPI_FLUSH_CPU_CACHE(); 673 } 674 } 675 acpi_idle_do_entry(cx); 676 677 return 0; 678 } 679 680 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 681 struct cpuidle_device *dev) 682 { 683 int i, count = ACPI_IDLE_STATE_START; 684 struct acpi_processor_cx *cx; 685 686 if (max_cstate == 0) 687 max_cstate = 1; 688 689 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 690 cx = &pr->power.states[i]; 691 692 if (!cx->valid) 693 continue; 694 695 per_cpu(acpi_cstate[count], dev->cpu) = cx; 696 697 count++; 698 if (count == CPUIDLE_STATE_MAX) 699 break; 700 } 701 702 if (!count) 703 return -EINVAL; 704 705 return 0; 706 } 707 708 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 709 { 710 int i, count; 711 struct acpi_processor_cx *cx; 712 struct cpuidle_state *state; 713 struct cpuidle_driver *drv = &acpi_idle_driver; 714 715 if (max_cstate == 0) 716 max_cstate = 1; 717 718 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) { 719 cpuidle_poll_state_init(drv); 720 count = 1; 721 } else { 722 count = 0; 723 } 724 725 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 726 cx = &pr->power.states[i]; 727 728 if (!cx->valid) 729 continue; 730 731 state = &drv->states[count]; 732 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i); 733 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN); 734 state->exit_latency = cx->latency; 735 state->target_residency = cx->latency * latency_factor; 736 state->enter = acpi_idle_enter; 737 738 state->flags = 0; 739 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { 740 state->enter_dead = acpi_idle_play_dead; 741 drv->safe_state_index = count; 742 } 743 /* 744 * Halt-induced C1 is not good for ->enter_s2idle, because it 745 * re-enables interrupts on exit. Moreover, C1 is generally not 746 * particularly interesting from the suspend-to-idle angle, so 747 * avoid C1 and the situations in which we may need to fall back 748 * to it altogether. 749 */ 750 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr)) 751 state->enter_s2idle = acpi_idle_enter_s2idle; 752 753 count++; 754 if (count == CPUIDLE_STATE_MAX) 755 break; 756 } 757 758 drv->state_count = count; 759 760 if (!count) 761 return -EINVAL; 762 763 return 0; 764 } 765 766 static inline void acpi_processor_cstate_first_run_checks(void) 767 { 768 static int first_run; 769 770 if (first_run) 771 return; 772 dmi_check_system(processor_power_dmi_table); 773 max_cstate = acpi_processor_cstate_check(max_cstate); 774 if (max_cstate < ACPI_C_STATES_MAX) 775 pr_notice("ACPI: processor limited to max C-state %d\n", 776 max_cstate); 777 first_run++; 778 779 if (nocst) 780 return; 781 782 acpi_processor_claim_cst_control(); 783 } 784 #else 785 786 static inline int disabled_by_idle_boot_param(void) { return 0; } 787 static inline void acpi_processor_cstate_first_run_checks(void) { } 788 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 789 { 790 return -ENODEV; 791 } 792 793 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 794 struct cpuidle_device *dev) 795 { 796 return -EINVAL; 797 } 798 799 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 800 { 801 return -EINVAL; 802 } 803 804 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */ 805 806 struct acpi_lpi_states_array { 807 unsigned int size; 808 unsigned int composite_states_size; 809 struct acpi_lpi_state *entries; 810 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER]; 811 }; 812 813 static int obj_get_integer(union acpi_object *obj, u32 *value) 814 { 815 if (obj->type != ACPI_TYPE_INTEGER) 816 return -EINVAL; 817 818 *value = obj->integer.value; 819 return 0; 820 } 821 822 static int acpi_processor_evaluate_lpi(acpi_handle handle, 823 struct acpi_lpi_states_array *info) 824 { 825 acpi_status status; 826 int ret = 0; 827 int pkg_count, state_idx = 1, loop; 828 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; 829 union acpi_object *lpi_data; 830 struct acpi_lpi_state *lpi_state; 831 832 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer); 833 if (ACPI_FAILURE(status)) { 834 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n")); 835 return -ENODEV; 836 } 837 838 lpi_data = buffer.pointer; 839 840 /* There must be at least 4 elements = 3 elements + 1 package */ 841 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE || 842 lpi_data->package.count < 4) { 843 pr_debug("not enough elements in _LPI\n"); 844 ret = -ENODATA; 845 goto end; 846 } 847 848 pkg_count = lpi_data->package.elements[2].integer.value; 849 850 /* Validate number of power states. */ 851 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) { 852 pr_debug("count given by _LPI is not valid\n"); 853 ret = -ENODATA; 854 goto end; 855 } 856 857 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL); 858 if (!lpi_state) { 859 ret = -ENOMEM; 860 goto end; 861 } 862 863 info->size = pkg_count; 864 info->entries = lpi_state; 865 866 /* LPI States start at index 3 */ 867 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) { 868 union acpi_object *element, *pkg_elem, *obj; 869 870 element = &lpi_data->package.elements[loop]; 871 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7) 872 continue; 873 874 pkg_elem = element->package.elements; 875 876 obj = pkg_elem + 6; 877 if (obj->type == ACPI_TYPE_BUFFER) { 878 struct acpi_power_register *reg; 879 880 reg = (struct acpi_power_register *)obj->buffer.pointer; 881 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO && 882 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) 883 continue; 884 885 lpi_state->address = reg->address; 886 lpi_state->entry_method = 887 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ? 888 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO; 889 } else if (obj->type == ACPI_TYPE_INTEGER) { 890 lpi_state->entry_method = ACPI_CSTATE_INTEGER; 891 lpi_state->address = obj->integer.value; 892 } else { 893 continue; 894 } 895 896 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/ 897 898 obj = pkg_elem + 9; 899 if (obj->type == ACPI_TYPE_STRING) 900 strlcpy(lpi_state->desc, obj->string.pointer, 901 ACPI_CX_DESC_LEN); 902 903 lpi_state->index = state_idx; 904 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) { 905 pr_debug("No min. residency found, assuming 10 us\n"); 906 lpi_state->min_residency = 10; 907 } 908 909 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) { 910 pr_debug("No wakeup residency found, assuming 10 us\n"); 911 lpi_state->wake_latency = 10; 912 } 913 914 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags)) 915 lpi_state->flags = 0; 916 917 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags)) 918 lpi_state->arch_flags = 0; 919 920 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq)) 921 lpi_state->res_cnt_freq = 1; 922 923 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state)) 924 lpi_state->enable_parent_state = 0; 925 } 926 927 acpi_handle_debug(handle, "Found %d power states\n", state_idx); 928 end: 929 kfree(buffer.pointer); 930 return ret; 931 } 932 933 /* 934 * flat_state_cnt - the number of composite LPI states after the process of flattening 935 */ 936 static int flat_state_cnt; 937 938 /** 939 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state 940 * 941 * @local: local LPI state 942 * @parent: parent LPI state 943 * @result: composite LPI state 944 */ 945 static bool combine_lpi_states(struct acpi_lpi_state *local, 946 struct acpi_lpi_state *parent, 947 struct acpi_lpi_state *result) 948 { 949 if (parent->entry_method == ACPI_CSTATE_INTEGER) { 950 if (!parent->address) /* 0 means autopromotable */ 951 return false; 952 result->address = local->address + parent->address; 953 } else { 954 result->address = parent->address; 955 } 956 957 result->min_residency = max(local->min_residency, parent->min_residency); 958 result->wake_latency = local->wake_latency + parent->wake_latency; 959 result->enable_parent_state = parent->enable_parent_state; 960 result->entry_method = local->entry_method; 961 962 result->flags = parent->flags; 963 result->arch_flags = parent->arch_flags; 964 result->index = parent->index; 965 966 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN); 967 strlcat(result->desc, "+", ACPI_CX_DESC_LEN); 968 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN); 969 return true; 970 } 971 972 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0) 973 974 static void stash_composite_state(struct acpi_lpi_states_array *curr_level, 975 struct acpi_lpi_state *t) 976 { 977 curr_level->composite_states[curr_level->composite_states_size++] = t; 978 } 979 980 static int flatten_lpi_states(struct acpi_processor *pr, 981 struct acpi_lpi_states_array *curr_level, 982 struct acpi_lpi_states_array *prev_level) 983 { 984 int i, j, state_count = curr_level->size; 985 struct acpi_lpi_state *p, *t = curr_level->entries; 986 987 curr_level->composite_states_size = 0; 988 for (j = 0; j < state_count; j++, t++) { 989 struct acpi_lpi_state *flpi; 990 991 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED)) 992 continue; 993 994 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) { 995 pr_warn("Limiting number of LPI states to max (%d)\n", 996 ACPI_PROCESSOR_MAX_POWER); 997 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n"); 998 break; 999 } 1000 1001 flpi = &pr->power.lpi_states[flat_state_cnt]; 1002 1003 if (!prev_level) { /* leaf/processor node */ 1004 memcpy(flpi, t, sizeof(*t)); 1005 stash_composite_state(curr_level, flpi); 1006 flat_state_cnt++; 1007 continue; 1008 } 1009 1010 for (i = 0; i < prev_level->composite_states_size; i++) { 1011 p = prev_level->composite_states[i]; 1012 if (t->index <= p->enable_parent_state && 1013 combine_lpi_states(p, t, flpi)) { 1014 stash_composite_state(curr_level, flpi); 1015 flat_state_cnt++; 1016 flpi++; 1017 } 1018 } 1019 } 1020 1021 kfree(curr_level->entries); 1022 return 0; 1023 } 1024 1025 static int acpi_processor_get_lpi_info(struct acpi_processor *pr) 1026 { 1027 int ret, i; 1028 acpi_status status; 1029 acpi_handle handle = pr->handle, pr_ahandle; 1030 struct acpi_device *d = NULL; 1031 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr; 1032 1033 if (!osc_pc_lpi_support_confirmed) 1034 return -EOPNOTSUPP; 1035 1036 if (!acpi_has_method(handle, "_LPI")) 1037 return -EINVAL; 1038 1039 flat_state_cnt = 0; 1040 prev = &info[0]; 1041 curr = &info[1]; 1042 handle = pr->handle; 1043 ret = acpi_processor_evaluate_lpi(handle, prev); 1044 if (ret) 1045 return ret; 1046 flatten_lpi_states(pr, prev, NULL); 1047 1048 status = acpi_get_parent(handle, &pr_ahandle); 1049 while (ACPI_SUCCESS(status)) { 1050 acpi_bus_get_device(pr_ahandle, &d); 1051 handle = pr_ahandle; 1052 1053 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID)) 1054 break; 1055 1056 /* can be optional ? */ 1057 if (!acpi_has_method(handle, "_LPI")) 1058 break; 1059 1060 ret = acpi_processor_evaluate_lpi(handle, curr); 1061 if (ret) 1062 break; 1063 1064 /* flatten all the LPI states in this level of hierarchy */ 1065 flatten_lpi_states(pr, curr, prev); 1066 1067 tmp = prev, prev = curr, curr = tmp; 1068 1069 status = acpi_get_parent(handle, &pr_ahandle); 1070 } 1071 1072 pr->power.count = flat_state_cnt; 1073 /* reset the index after flattening */ 1074 for (i = 0; i < pr->power.count; i++) 1075 pr->power.lpi_states[i].index = i; 1076 1077 /* Tell driver that _LPI is supported. */ 1078 pr->flags.has_lpi = 1; 1079 pr->flags.power = 1; 1080 1081 return 0; 1082 } 1083 1084 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu) 1085 { 1086 return -ENODEV; 1087 } 1088 1089 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi) 1090 { 1091 return -ENODEV; 1092 } 1093 1094 /** 1095 * acpi_idle_lpi_enter - enters an ACPI any LPI state 1096 * @dev: the target CPU 1097 * @drv: cpuidle driver containing cpuidle state info 1098 * @index: index of target state 1099 * 1100 * Return: 0 for success or negative value for error 1101 */ 1102 static int acpi_idle_lpi_enter(struct cpuidle_device *dev, 1103 struct cpuidle_driver *drv, int index) 1104 { 1105 struct acpi_processor *pr; 1106 struct acpi_lpi_state *lpi; 1107 1108 pr = __this_cpu_read(processors); 1109 1110 if (unlikely(!pr)) 1111 return -EINVAL; 1112 1113 lpi = &pr->power.lpi_states[index]; 1114 if (lpi->entry_method == ACPI_CSTATE_FFH) 1115 return acpi_processor_ffh_lpi_enter(lpi); 1116 1117 return -EINVAL; 1118 } 1119 1120 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr) 1121 { 1122 int i; 1123 struct acpi_lpi_state *lpi; 1124 struct cpuidle_state *state; 1125 struct cpuidle_driver *drv = &acpi_idle_driver; 1126 1127 if (!pr->flags.has_lpi) 1128 return -EOPNOTSUPP; 1129 1130 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) { 1131 lpi = &pr->power.lpi_states[i]; 1132 1133 state = &drv->states[i]; 1134 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i); 1135 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN); 1136 state->exit_latency = lpi->wake_latency; 1137 state->target_residency = lpi->min_residency; 1138 if (lpi->arch_flags) 1139 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 1140 state->enter = acpi_idle_lpi_enter; 1141 drv->safe_state_index = i; 1142 } 1143 1144 drv->state_count = i; 1145 1146 return 0; 1147 } 1148 1149 /** 1150 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle 1151 * global state data i.e. idle routines 1152 * 1153 * @pr: the ACPI processor 1154 */ 1155 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr) 1156 { 1157 int i; 1158 struct cpuidle_driver *drv = &acpi_idle_driver; 1159 1160 if (!pr->flags.power_setup_done || !pr->flags.power) 1161 return -EINVAL; 1162 1163 drv->safe_state_index = -1; 1164 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) { 1165 drv->states[i].name[0] = '\0'; 1166 drv->states[i].desc[0] = '\0'; 1167 } 1168 1169 if (pr->flags.has_lpi) 1170 return acpi_processor_setup_lpi_states(pr); 1171 1172 return acpi_processor_setup_cstates(pr); 1173 } 1174 1175 /** 1176 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE 1177 * device i.e. per-cpu data 1178 * 1179 * @pr: the ACPI processor 1180 * @dev : the cpuidle device 1181 */ 1182 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr, 1183 struct cpuidle_device *dev) 1184 { 1185 if (!pr->flags.power_setup_done || !pr->flags.power || !dev) 1186 return -EINVAL; 1187 1188 dev->cpu = pr->id; 1189 if (pr->flags.has_lpi) 1190 return acpi_processor_ffh_lpi_probe(pr->id); 1191 1192 return acpi_processor_setup_cpuidle_cx(pr, dev); 1193 } 1194 1195 static int acpi_processor_get_power_info(struct acpi_processor *pr) 1196 { 1197 int ret; 1198 1199 ret = acpi_processor_get_lpi_info(pr); 1200 if (ret) 1201 ret = acpi_processor_get_cstate_info(pr); 1202 1203 return ret; 1204 } 1205 1206 int acpi_processor_hotplug(struct acpi_processor *pr) 1207 { 1208 int ret = 0; 1209 struct cpuidle_device *dev; 1210 1211 if (disabled_by_idle_boot_param()) 1212 return 0; 1213 1214 if (!pr->flags.power_setup_done) 1215 return -ENODEV; 1216 1217 dev = per_cpu(acpi_cpuidle_device, pr->id); 1218 cpuidle_pause_and_lock(); 1219 cpuidle_disable_device(dev); 1220 ret = acpi_processor_get_power_info(pr); 1221 if (!ret && pr->flags.power) { 1222 acpi_processor_setup_cpuidle_dev(pr, dev); 1223 ret = cpuidle_enable_device(dev); 1224 } 1225 cpuidle_resume_and_unlock(); 1226 1227 return ret; 1228 } 1229 1230 int acpi_processor_power_state_has_changed(struct acpi_processor *pr) 1231 { 1232 int cpu; 1233 struct acpi_processor *_pr; 1234 struct cpuidle_device *dev; 1235 1236 if (disabled_by_idle_boot_param()) 1237 return 0; 1238 1239 if (!pr->flags.power_setup_done) 1240 return -ENODEV; 1241 1242 /* 1243 * FIXME: Design the ACPI notification to make it once per 1244 * system instead of once per-cpu. This condition is a hack 1245 * to make the code that updates C-States be called once. 1246 */ 1247 1248 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) { 1249 1250 /* Protect against cpu-hotplug */ 1251 get_online_cpus(); 1252 cpuidle_pause_and_lock(); 1253 1254 /* Disable all cpuidle devices */ 1255 for_each_online_cpu(cpu) { 1256 _pr = per_cpu(processors, cpu); 1257 if (!_pr || !_pr->flags.power_setup_done) 1258 continue; 1259 dev = per_cpu(acpi_cpuidle_device, cpu); 1260 cpuidle_disable_device(dev); 1261 } 1262 1263 /* Populate Updated C-state information */ 1264 acpi_processor_get_power_info(pr); 1265 acpi_processor_setup_cpuidle_states(pr); 1266 1267 /* Enable all cpuidle devices */ 1268 for_each_online_cpu(cpu) { 1269 _pr = per_cpu(processors, cpu); 1270 if (!_pr || !_pr->flags.power_setup_done) 1271 continue; 1272 acpi_processor_get_power_info(_pr); 1273 if (_pr->flags.power) { 1274 dev = per_cpu(acpi_cpuidle_device, cpu); 1275 acpi_processor_setup_cpuidle_dev(_pr, dev); 1276 cpuidle_enable_device(dev); 1277 } 1278 } 1279 cpuidle_resume_and_unlock(); 1280 put_online_cpus(); 1281 } 1282 1283 return 0; 1284 } 1285 1286 static int acpi_processor_registered; 1287 1288 int acpi_processor_power_init(struct acpi_processor *pr) 1289 { 1290 int retval; 1291 struct cpuidle_device *dev; 1292 1293 if (disabled_by_idle_boot_param()) 1294 return 0; 1295 1296 acpi_processor_cstate_first_run_checks(); 1297 1298 if (!acpi_processor_get_power_info(pr)) 1299 pr->flags.power_setup_done = 1; 1300 1301 /* 1302 * Install the idle handler if processor power management is supported. 1303 * Note that we use previously set idle handler will be used on 1304 * platforms that only support C1. 1305 */ 1306 if (pr->flags.power) { 1307 /* Register acpi_idle_driver if not already registered */ 1308 if (!acpi_processor_registered) { 1309 acpi_processor_setup_cpuidle_states(pr); 1310 retval = cpuidle_register_driver(&acpi_idle_driver); 1311 if (retval) 1312 return retval; 1313 pr_debug("%s registered with cpuidle\n", 1314 acpi_idle_driver.name); 1315 } 1316 1317 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1318 if (!dev) 1319 return -ENOMEM; 1320 per_cpu(acpi_cpuidle_device, pr->id) = dev; 1321 1322 acpi_processor_setup_cpuidle_dev(pr, dev); 1323 1324 /* Register per-cpu cpuidle_device. Cpuidle driver 1325 * must already be registered before registering device 1326 */ 1327 retval = cpuidle_register_device(dev); 1328 if (retval) { 1329 if (acpi_processor_registered == 0) 1330 cpuidle_unregister_driver(&acpi_idle_driver); 1331 return retval; 1332 } 1333 acpi_processor_registered++; 1334 } 1335 return 0; 1336 } 1337 1338 int acpi_processor_power_exit(struct acpi_processor *pr) 1339 { 1340 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id); 1341 1342 if (disabled_by_idle_boot_param()) 1343 return 0; 1344 1345 if (pr->flags.power) { 1346 cpuidle_unregister_device(dev); 1347 acpi_processor_registered--; 1348 if (acpi_processor_registered == 0) 1349 cpuidle_unregister_driver(&acpi_idle_driver); 1350 } 1351 1352 pr->flags.power_setup_done = 0; 1353 return 0; 1354 } 1355