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 bool lapic_timer_needs_broadcast(struct acpi_processor *pr, 165 struct acpi_processor_cx *cx) 166 { 167 return cx - pr->power.states >= pr->power.timer_broadcast_on_state; 168 } 169 170 #else 171 172 static void lapic_timer_check_state(int state, struct acpi_processor *pr, 173 struct acpi_processor_cx *cstate) { } 174 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { } 175 176 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr, 177 struct acpi_processor_cx *cx) 178 { 179 } 180 181 #endif 182 183 #if defined(CONFIG_X86) 184 static void tsc_check_state(int state) 185 { 186 switch (boot_cpu_data.x86_vendor) { 187 case X86_VENDOR_HYGON: 188 case X86_VENDOR_AMD: 189 case X86_VENDOR_INTEL: 190 case X86_VENDOR_CENTAUR: 191 case X86_VENDOR_ZHAOXIN: 192 /* 193 * AMD Fam10h TSC will tick in all 194 * C/P/S0/S1 states when this bit is set. 195 */ 196 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 197 return; 198 fallthrough; 199 default: 200 /* TSC could halt in idle, so notify users */ 201 if (state > ACPI_STATE_C1) 202 mark_tsc_unstable("TSC halts in idle"); 203 } 204 } 205 #else 206 static void tsc_check_state(int state) { return; } 207 #endif 208 209 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr) 210 { 211 212 if (!pr->pblk) 213 return -ENODEV; 214 215 /* if info is obtained from pblk/fadt, type equals state */ 216 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2; 217 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3; 218 219 #ifndef CONFIG_HOTPLUG_CPU 220 /* 221 * Check for P_LVL2_UP flag before entering C2 and above on 222 * an SMP system. 223 */ 224 if ((num_online_cpus() > 1) && 225 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) 226 return -ENODEV; 227 #endif 228 229 /* determine C2 and C3 address from pblk */ 230 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4; 231 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5; 232 233 /* determine latencies from FADT */ 234 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency; 235 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency; 236 237 /* 238 * FADT specified C2 latency must be less than or equal to 239 * 100 microseconds. 240 */ 241 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) { 242 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 243 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency)); 244 /* invalidate C2 */ 245 pr->power.states[ACPI_STATE_C2].address = 0; 246 } 247 248 /* 249 * FADT supplied C3 latency must be less than or equal to 250 * 1000 microseconds. 251 */ 252 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) { 253 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 254 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency)); 255 /* invalidate C3 */ 256 pr->power.states[ACPI_STATE_C3].address = 0; 257 } 258 259 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 260 "lvl2[0x%08x] lvl3[0x%08x]\n", 261 pr->power.states[ACPI_STATE_C2].address, 262 pr->power.states[ACPI_STATE_C3].address)); 263 264 snprintf(pr->power.states[ACPI_STATE_C2].desc, 265 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x", 266 pr->power.states[ACPI_STATE_C2].address); 267 snprintf(pr->power.states[ACPI_STATE_C3].desc, 268 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x", 269 pr->power.states[ACPI_STATE_C3].address); 270 271 return 0; 272 } 273 274 static int acpi_processor_get_power_info_default(struct acpi_processor *pr) 275 { 276 if (!pr->power.states[ACPI_STATE_C1].valid) { 277 /* set the first C-State to C1 */ 278 /* all processors need to support C1 */ 279 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1; 280 pr->power.states[ACPI_STATE_C1].valid = 1; 281 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT; 282 283 snprintf(pr->power.states[ACPI_STATE_C1].desc, 284 ACPI_CX_DESC_LEN, "ACPI HLT"); 285 } 286 /* the C0 state only exists as a filler in our array */ 287 pr->power.states[ACPI_STATE_C0].valid = 1; 288 return 0; 289 } 290 291 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr) 292 { 293 int ret; 294 295 if (nocst) 296 return -ENODEV; 297 298 ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power); 299 if (ret) 300 return ret; 301 302 if (!pr->power.count) 303 return -EFAULT; 304 305 pr->flags.has_cst = 1; 306 return 0; 307 } 308 309 static void acpi_processor_power_verify_c3(struct acpi_processor *pr, 310 struct acpi_processor_cx *cx) 311 { 312 static int bm_check_flag = -1; 313 static int bm_control_flag = -1; 314 315 316 if (!cx->address) 317 return; 318 319 /* 320 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast) 321 * DMA transfers are used by any ISA device to avoid livelock. 322 * Note that we could disable Type-F DMA (as recommended by 323 * the erratum), but this is known to disrupt certain ISA 324 * devices thus we take the conservative approach. 325 */ 326 else if (errata.piix4.fdma) { 327 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 328 "C3 not supported on PIIX4 with Type-F DMA\n")); 329 return; 330 } 331 332 /* All the logic here assumes flags.bm_check is same across all CPUs */ 333 if (bm_check_flag == -1) { 334 /* Determine whether bm_check is needed based on CPU */ 335 acpi_processor_power_init_bm_check(&(pr->flags), pr->id); 336 bm_check_flag = pr->flags.bm_check; 337 bm_control_flag = pr->flags.bm_control; 338 } else { 339 pr->flags.bm_check = bm_check_flag; 340 pr->flags.bm_control = bm_control_flag; 341 } 342 343 if (pr->flags.bm_check) { 344 if (!pr->flags.bm_control) { 345 if (pr->flags.has_cst != 1) { 346 /* bus mastering control is necessary */ 347 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 348 "C3 support requires BM control\n")); 349 return; 350 } else { 351 /* Here we enter C3 without bus mastering */ 352 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 353 "C3 support without BM control\n")); 354 } 355 } 356 } else { 357 /* 358 * WBINVD should be set in fadt, for C3 state to be 359 * supported on when bm_check is not required. 360 */ 361 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) { 362 ACPI_DEBUG_PRINT((ACPI_DB_INFO, 363 "Cache invalidation should work properly" 364 " for C3 to be enabled on SMP systems\n")); 365 return; 366 } 367 } 368 369 /* 370 * Otherwise we've met all of our C3 requirements. 371 * Normalize the C3 latency to expidite policy. Enable 372 * checking of bus mastering status (bm_check) so we can 373 * use this in our C3 policy 374 */ 375 cx->valid = 1; 376 377 /* 378 * On older chipsets, BM_RLD needs to be set 379 * in order for Bus Master activity to wake the 380 * system from C3. Newer chipsets handle DMA 381 * during C3 automatically and BM_RLD is a NOP. 382 * In either case, the proper way to 383 * handle BM_RLD is to set it and leave it set. 384 */ 385 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1); 386 387 return; 388 } 389 390 static int acpi_processor_power_verify(struct acpi_processor *pr) 391 { 392 unsigned int i; 393 unsigned int working = 0; 394 395 pr->power.timer_broadcast_on_state = INT_MAX; 396 397 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 398 struct acpi_processor_cx *cx = &pr->power.states[i]; 399 400 switch (cx->type) { 401 case ACPI_STATE_C1: 402 cx->valid = 1; 403 break; 404 405 case ACPI_STATE_C2: 406 if (!cx->address) 407 break; 408 cx->valid = 1; 409 break; 410 411 case ACPI_STATE_C3: 412 acpi_processor_power_verify_c3(pr, cx); 413 break; 414 } 415 if (!cx->valid) 416 continue; 417 418 lapic_timer_check_state(i, pr, cx); 419 tsc_check_state(cx->type); 420 working++; 421 } 422 423 lapic_timer_propagate_broadcast(pr); 424 425 return (working); 426 } 427 428 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 429 { 430 unsigned int i; 431 int result; 432 433 434 /* NOTE: the idle thread may not be running while calling 435 * this function */ 436 437 /* Zero initialize all the C-states info. */ 438 memset(pr->power.states, 0, sizeof(pr->power.states)); 439 440 result = acpi_processor_get_power_info_cst(pr); 441 if (result == -ENODEV) 442 result = acpi_processor_get_power_info_fadt(pr); 443 444 if (result) 445 return result; 446 447 acpi_processor_get_power_info_default(pr); 448 449 pr->power.count = acpi_processor_power_verify(pr); 450 451 /* 452 * if one state of type C2 or C3 is available, mark this 453 * CPU as being "idle manageable" 454 */ 455 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) { 456 if (pr->power.states[i].valid) { 457 pr->power.count = i; 458 pr->flags.power = 1; 459 } 460 } 461 462 return 0; 463 } 464 465 /** 466 * acpi_idle_bm_check - checks if bus master activity was detected 467 */ 468 static int acpi_idle_bm_check(void) 469 { 470 u32 bm_status = 0; 471 472 if (bm_check_disable) 473 return 0; 474 475 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status); 476 if (bm_status) 477 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1); 478 /* 479 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect 480 * the true state of bus mastering activity; forcing us to 481 * manually check the BMIDEA bit of each IDE channel. 482 */ 483 else if (errata.piix4.bmisx) { 484 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01) 485 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01)) 486 bm_status = 1; 487 } 488 return bm_status; 489 } 490 491 static void wait_for_freeze(void) 492 { 493 #ifdef CONFIG_X86 494 /* No delay is needed if we are in guest */ 495 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 496 return; 497 #endif 498 /* Dummy wait op - must do something useless after P_LVL2 read 499 because chipsets cannot guarantee that STPCLK# signal 500 gets asserted in time to freeze execution properly. */ 501 inl(acpi_gbl_FADT.xpm_timer_block.address); 502 } 503 504 /** 505 * acpi_idle_do_entry - enter idle state using the appropriate method 506 * @cx: cstate data 507 * 508 * Caller disables interrupt before call and enables interrupt after return. 509 */ 510 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx) 511 { 512 if (cx->entry_method == ACPI_CSTATE_FFH) { 513 /* Call into architectural FFH based C-state */ 514 acpi_processor_ffh_cstate_enter(cx); 515 } else if (cx->entry_method == ACPI_CSTATE_HALT) { 516 acpi_safe_halt(); 517 } else { 518 /* IO port based C-state */ 519 inb(cx->address); 520 wait_for_freeze(); 521 } 522 } 523 524 /** 525 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining) 526 * @dev: the target CPU 527 * @index: the index of suggested state 528 */ 529 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index) 530 { 531 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 532 533 ACPI_FLUSH_CPU_CACHE(); 534 535 while (1) { 536 537 if (cx->entry_method == ACPI_CSTATE_HALT) 538 safe_halt(); 539 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) { 540 inb(cx->address); 541 wait_for_freeze(); 542 } else 543 return -ENODEV; 544 } 545 546 /* Never reached */ 547 return 0; 548 } 549 550 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr) 551 { 552 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst && 553 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED); 554 } 555 556 static int c3_cpu_count; 557 static DEFINE_RAW_SPINLOCK(c3_lock); 558 559 /** 560 * acpi_idle_enter_bm - enters C3 with proper BM handling 561 * @drv: cpuidle driver 562 * @pr: Target processor 563 * @cx: Target state context 564 * @index: index of target state 565 */ 566 static int acpi_idle_enter_bm(struct cpuidle_driver *drv, 567 struct acpi_processor *pr, 568 struct acpi_processor_cx *cx, 569 int index) 570 { 571 static struct acpi_processor_cx safe_cx = { 572 .entry_method = ACPI_CSTATE_HALT, 573 }; 574 575 /* 576 * disable bus master 577 * bm_check implies we need ARB_DIS 578 * bm_control implies whether we can do ARB_DIS 579 * 580 * That leaves a case where bm_check is set and bm_control is not set. 581 * In that case we cannot do much, we enter C3 without doing anything. 582 */ 583 bool dis_bm = pr->flags.bm_control; 584 585 /* If we can skip BM, demote to a safe state. */ 586 if (!cx->bm_sts_skip && acpi_idle_bm_check()) { 587 dis_bm = false; 588 index = drv->safe_state_index; 589 if (index >= 0) { 590 cx = this_cpu_read(acpi_cstate[index]); 591 } else { 592 cx = &safe_cx; 593 index = -EBUSY; 594 } 595 } 596 597 if (dis_bm) { 598 raw_spin_lock(&c3_lock); 599 c3_cpu_count++; 600 /* Disable bus master arbitration when all CPUs are in C3 */ 601 if (c3_cpu_count == num_online_cpus()) 602 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1); 603 raw_spin_unlock(&c3_lock); 604 } 605 606 rcu_idle_enter(); 607 608 acpi_idle_do_entry(cx); 609 610 rcu_idle_exit(); 611 612 /* Re-enable bus master arbitration */ 613 if (dis_bm) { 614 raw_spin_lock(&c3_lock); 615 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0); 616 c3_cpu_count--; 617 raw_spin_unlock(&c3_lock); 618 } 619 620 return index; 621 } 622 623 static int acpi_idle_enter(struct cpuidle_device *dev, 624 struct cpuidle_driver *drv, int index) 625 { 626 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 627 struct acpi_processor *pr; 628 629 pr = __this_cpu_read(processors); 630 if (unlikely(!pr)) 631 return -EINVAL; 632 633 if (cx->type != ACPI_STATE_C1) { 634 if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) 635 return acpi_idle_enter_bm(drv, pr, cx, index); 636 637 /* C2 to C1 demotion. */ 638 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) { 639 index = ACPI_IDLE_STATE_START; 640 cx = per_cpu(acpi_cstate[index], dev->cpu); 641 } 642 } 643 644 if (cx->type == ACPI_STATE_C3) 645 ACPI_FLUSH_CPU_CACHE(); 646 647 acpi_idle_do_entry(cx); 648 649 return index; 650 } 651 652 static int acpi_idle_enter_s2idle(struct cpuidle_device *dev, 653 struct cpuidle_driver *drv, int index) 654 { 655 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 656 657 if (cx->type == ACPI_STATE_C3) { 658 struct acpi_processor *pr = __this_cpu_read(processors); 659 660 if (unlikely(!pr)) 661 return 0; 662 663 if (pr->flags.bm_check) { 664 u8 bm_sts_skip = cx->bm_sts_skip; 665 666 /* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */ 667 cx->bm_sts_skip = 1; 668 acpi_idle_enter_bm(drv, pr, cx, index); 669 cx->bm_sts_skip = bm_sts_skip; 670 671 return 0; 672 } else { 673 ACPI_FLUSH_CPU_CACHE(); 674 } 675 } 676 acpi_idle_do_entry(cx); 677 678 return 0; 679 } 680 681 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 682 struct cpuidle_device *dev) 683 { 684 int i, count = ACPI_IDLE_STATE_START; 685 struct acpi_processor_cx *cx; 686 struct cpuidle_state *state; 687 688 if (max_cstate == 0) 689 max_cstate = 1; 690 691 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 692 state = &acpi_idle_driver.states[count]; 693 cx = &pr->power.states[i]; 694 695 if (!cx->valid) 696 continue; 697 698 per_cpu(acpi_cstate[count], dev->cpu) = cx; 699 700 if (lapic_timer_needs_broadcast(pr, cx)) 701 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 702 703 if (cx->type == ACPI_STATE_C3) { 704 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED; 705 if (pr->flags.bm_check) 706 state->flags |= CPUIDLE_FLAG_RCU_IDLE; 707 } 708 709 count++; 710 if (count == CPUIDLE_STATE_MAX) 711 break; 712 } 713 714 if (!count) 715 return -EINVAL; 716 717 return 0; 718 } 719 720 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 721 { 722 int i, count; 723 struct acpi_processor_cx *cx; 724 struct cpuidle_state *state; 725 struct cpuidle_driver *drv = &acpi_idle_driver; 726 727 if (max_cstate == 0) 728 max_cstate = 1; 729 730 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) { 731 cpuidle_poll_state_init(drv); 732 count = 1; 733 } else { 734 count = 0; 735 } 736 737 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 738 cx = &pr->power.states[i]; 739 740 if (!cx->valid) 741 continue; 742 743 state = &drv->states[count]; 744 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i); 745 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN); 746 state->exit_latency = cx->latency; 747 state->target_residency = cx->latency * latency_factor; 748 state->enter = acpi_idle_enter; 749 750 state->flags = 0; 751 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { 752 state->enter_dead = acpi_idle_play_dead; 753 drv->safe_state_index = count; 754 } 755 /* 756 * Halt-induced C1 is not good for ->enter_s2idle, because it 757 * re-enables interrupts on exit. Moreover, C1 is generally not 758 * particularly interesting from the suspend-to-idle angle, so 759 * avoid C1 and the situations in which we may need to fall back 760 * to it altogether. 761 */ 762 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr)) 763 state->enter_s2idle = acpi_idle_enter_s2idle; 764 765 count++; 766 if (count == CPUIDLE_STATE_MAX) 767 break; 768 } 769 770 drv->state_count = count; 771 772 if (!count) 773 return -EINVAL; 774 775 return 0; 776 } 777 778 static inline void acpi_processor_cstate_first_run_checks(void) 779 { 780 static int first_run; 781 782 if (first_run) 783 return; 784 dmi_check_system(processor_power_dmi_table); 785 max_cstate = acpi_processor_cstate_check(max_cstate); 786 if (max_cstate < ACPI_C_STATES_MAX) 787 pr_notice("ACPI: processor limited to max C-state %d\n", 788 max_cstate); 789 first_run++; 790 791 if (nocst) 792 return; 793 794 acpi_processor_claim_cst_control(); 795 } 796 #else 797 798 static inline int disabled_by_idle_boot_param(void) { return 0; } 799 static inline void acpi_processor_cstate_first_run_checks(void) { } 800 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 801 { 802 return -ENODEV; 803 } 804 805 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 806 struct cpuidle_device *dev) 807 { 808 return -EINVAL; 809 } 810 811 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 812 { 813 return -EINVAL; 814 } 815 816 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */ 817 818 struct acpi_lpi_states_array { 819 unsigned int size; 820 unsigned int composite_states_size; 821 struct acpi_lpi_state *entries; 822 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER]; 823 }; 824 825 static int obj_get_integer(union acpi_object *obj, u32 *value) 826 { 827 if (obj->type != ACPI_TYPE_INTEGER) 828 return -EINVAL; 829 830 *value = obj->integer.value; 831 return 0; 832 } 833 834 static int acpi_processor_evaluate_lpi(acpi_handle handle, 835 struct acpi_lpi_states_array *info) 836 { 837 acpi_status status; 838 int ret = 0; 839 int pkg_count, state_idx = 1, loop; 840 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; 841 union acpi_object *lpi_data; 842 struct acpi_lpi_state *lpi_state; 843 844 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer); 845 if (ACPI_FAILURE(status)) { 846 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n")); 847 return -ENODEV; 848 } 849 850 lpi_data = buffer.pointer; 851 852 /* There must be at least 4 elements = 3 elements + 1 package */ 853 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE || 854 lpi_data->package.count < 4) { 855 pr_debug("not enough elements in _LPI\n"); 856 ret = -ENODATA; 857 goto end; 858 } 859 860 pkg_count = lpi_data->package.elements[2].integer.value; 861 862 /* Validate number of power states. */ 863 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) { 864 pr_debug("count given by _LPI is not valid\n"); 865 ret = -ENODATA; 866 goto end; 867 } 868 869 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL); 870 if (!lpi_state) { 871 ret = -ENOMEM; 872 goto end; 873 } 874 875 info->size = pkg_count; 876 info->entries = lpi_state; 877 878 /* LPI States start at index 3 */ 879 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) { 880 union acpi_object *element, *pkg_elem, *obj; 881 882 element = &lpi_data->package.elements[loop]; 883 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7) 884 continue; 885 886 pkg_elem = element->package.elements; 887 888 obj = pkg_elem + 6; 889 if (obj->type == ACPI_TYPE_BUFFER) { 890 struct acpi_power_register *reg; 891 892 reg = (struct acpi_power_register *)obj->buffer.pointer; 893 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO && 894 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) 895 continue; 896 897 lpi_state->address = reg->address; 898 lpi_state->entry_method = 899 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ? 900 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO; 901 } else if (obj->type == ACPI_TYPE_INTEGER) { 902 lpi_state->entry_method = ACPI_CSTATE_INTEGER; 903 lpi_state->address = obj->integer.value; 904 } else { 905 continue; 906 } 907 908 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/ 909 910 obj = pkg_elem + 9; 911 if (obj->type == ACPI_TYPE_STRING) 912 strlcpy(lpi_state->desc, obj->string.pointer, 913 ACPI_CX_DESC_LEN); 914 915 lpi_state->index = state_idx; 916 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) { 917 pr_debug("No min. residency found, assuming 10 us\n"); 918 lpi_state->min_residency = 10; 919 } 920 921 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) { 922 pr_debug("No wakeup residency found, assuming 10 us\n"); 923 lpi_state->wake_latency = 10; 924 } 925 926 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags)) 927 lpi_state->flags = 0; 928 929 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags)) 930 lpi_state->arch_flags = 0; 931 932 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq)) 933 lpi_state->res_cnt_freq = 1; 934 935 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state)) 936 lpi_state->enable_parent_state = 0; 937 } 938 939 acpi_handle_debug(handle, "Found %d power states\n", state_idx); 940 end: 941 kfree(buffer.pointer); 942 return ret; 943 } 944 945 /* 946 * flat_state_cnt - the number of composite LPI states after the process of flattening 947 */ 948 static int flat_state_cnt; 949 950 /** 951 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state 952 * 953 * @local: local LPI state 954 * @parent: parent LPI state 955 * @result: composite LPI state 956 */ 957 static bool combine_lpi_states(struct acpi_lpi_state *local, 958 struct acpi_lpi_state *parent, 959 struct acpi_lpi_state *result) 960 { 961 if (parent->entry_method == ACPI_CSTATE_INTEGER) { 962 if (!parent->address) /* 0 means autopromotable */ 963 return false; 964 result->address = local->address + parent->address; 965 } else { 966 result->address = parent->address; 967 } 968 969 result->min_residency = max(local->min_residency, parent->min_residency); 970 result->wake_latency = local->wake_latency + parent->wake_latency; 971 result->enable_parent_state = parent->enable_parent_state; 972 result->entry_method = local->entry_method; 973 974 result->flags = parent->flags; 975 result->arch_flags = parent->arch_flags; 976 result->index = parent->index; 977 978 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN); 979 strlcat(result->desc, "+", ACPI_CX_DESC_LEN); 980 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN); 981 return true; 982 } 983 984 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0) 985 986 static void stash_composite_state(struct acpi_lpi_states_array *curr_level, 987 struct acpi_lpi_state *t) 988 { 989 curr_level->composite_states[curr_level->composite_states_size++] = t; 990 } 991 992 static int flatten_lpi_states(struct acpi_processor *pr, 993 struct acpi_lpi_states_array *curr_level, 994 struct acpi_lpi_states_array *prev_level) 995 { 996 int i, j, state_count = curr_level->size; 997 struct acpi_lpi_state *p, *t = curr_level->entries; 998 999 curr_level->composite_states_size = 0; 1000 for (j = 0; j < state_count; j++, t++) { 1001 struct acpi_lpi_state *flpi; 1002 1003 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED)) 1004 continue; 1005 1006 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) { 1007 pr_warn("Limiting number of LPI states to max (%d)\n", 1008 ACPI_PROCESSOR_MAX_POWER); 1009 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n"); 1010 break; 1011 } 1012 1013 flpi = &pr->power.lpi_states[flat_state_cnt]; 1014 1015 if (!prev_level) { /* leaf/processor node */ 1016 memcpy(flpi, t, sizeof(*t)); 1017 stash_composite_state(curr_level, flpi); 1018 flat_state_cnt++; 1019 continue; 1020 } 1021 1022 for (i = 0; i < prev_level->composite_states_size; i++) { 1023 p = prev_level->composite_states[i]; 1024 if (t->index <= p->enable_parent_state && 1025 combine_lpi_states(p, t, flpi)) { 1026 stash_composite_state(curr_level, flpi); 1027 flat_state_cnt++; 1028 flpi++; 1029 } 1030 } 1031 } 1032 1033 kfree(curr_level->entries); 1034 return 0; 1035 } 1036 1037 static int acpi_processor_get_lpi_info(struct acpi_processor *pr) 1038 { 1039 int ret, i; 1040 acpi_status status; 1041 acpi_handle handle = pr->handle, pr_ahandle; 1042 struct acpi_device *d = NULL; 1043 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr; 1044 1045 if (!osc_pc_lpi_support_confirmed) 1046 return -EOPNOTSUPP; 1047 1048 if (!acpi_has_method(handle, "_LPI")) 1049 return -EINVAL; 1050 1051 flat_state_cnt = 0; 1052 prev = &info[0]; 1053 curr = &info[1]; 1054 handle = pr->handle; 1055 ret = acpi_processor_evaluate_lpi(handle, prev); 1056 if (ret) 1057 return ret; 1058 flatten_lpi_states(pr, prev, NULL); 1059 1060 status = acpi_get_parent(handle, &pr_ahandle); 1061 while (ACPI_SUCCESS(status)) { 1062 acpi_bus_get_device(pr_ahandle, &d); 1063 handle = pr_ahandle; 1064 1065 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID)) 1066 break; 1067 1068 /* can be optional ? */ 1069 if (!acpi_has_method(handle, "_LPI")) 1070 break; 1071 1072 ret = acpi_processor_evaluate_lpi(handle, curr); 1073 if (ret) 1074 break; 1075 1076 /* flatten all the LPI states in this level of hierarchy */ 1077 flatten_lpi_states(pr, curr, prev); 1078 1079 tmp = prev, prev = curr, curr = tmp; 1080 1081 status = acpi_get_parent(handle, &pr_ahandle); 1082 } 1083 1084 pr->power.count = flat_state_cnt; 1085 /* reset the index after flattening */ 1086 for (i = 0; i < pr->power.count; i++) 1087 pr->power.lpi_states[i].index = i; 1088 1089 /* Tell driver that _LPI is supported. */ 1090 pr->flags.has_lpi = 1; 1091 pr->flags.power = 1; 1092 1093 return 0; 1094 } 1095 1096 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu) 1097 { 1098 return -ENODEV; 1099 } 1100 1101 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi) 1102 { 1103 return -ENODEV; 1104 } 1105 1106 /** 1107 * acpi_idle_lpi_enter - enters an ACPI any LPI state 1108 * @dev: the target CPU 1109 * @drv: cpuidle driver containing cpuidle state info 1110 * @index: index of target state 1111 * 1112 * Return: 0 for success or negative value for error 1113 */ 1114 static int acpi_idle_lpi_enter(struct cpuidle_device *dev, 1115 struct cpuidle_driver *drv, int index) 1116 { 1117 struct acpi_processor *pr; 1118 struct acpi_lpi_state *lpi; 1119 1120 pr = __this_cpu_read(processors); 1121 1122 if (unlikely(!pr)) 1123 return -EINVAL; 1124 1125 lpi = &pr->power.lpi_states[index]; 1126 if (lpi->entry_method == ACPI_CSTATE_FFH) 1127 return acpi_processor_ffh_lpi_enter(lpi); 1128 1129 return -EINVAL; 1130 } 1131 1132 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr) 1133 { 1134 int i; 1135 struct acpi_lpi_state *lpi; 1136 struct cpuidle_state *state; 1137 struct cpuidle_driver *drv = &acpi_idle_driver; 1138 1139 if (!pr->flags.has_lpi) 1140 return -EOPNOTSUPP; 1141 1142 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) { 1143 lpi = &pr->power.lpi_states[i]; 1144 1145 state = &drv->states[i]; 1146 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i); 1147 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN); 1148 state->exit_latency = lpi->wake_latency; 1149 state->target_residency = lpi->min_residency; 1150 if (lpi->arch_flags) 1151 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 1152 state->enter = acpi_idle_lpi_enter; 1153 drv->safe_state_index = i; 1154 } 1155 1156 drv->state_count = i; 1157 1158 return 0; 1159 } 1160 1161 /** 1162 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle 1163 * global state data i.e. idle routines 1164 * 1165 * @pr: the ACPI processor 1166 */ 1167 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr) 1168 { 1169 int i; 1170 struct cpuidle_driver *drv = &acpi_idle_driver; 1171 1172 if (!pr->flags.power_setup_done || !pr->flags.power) 1173 return -EINVAL; 1174 1175 drv->safe_state_index = -1; 1176 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) { 1177 drv->states[i].name[0] = '\0'; 1178 drv->states[i].desc[0] = '\0'; 1179 } 1180 1181 if (pr->flags.has_lpi) 1182 return acpi_processor_setup_lpi_states(pr); 1183 1184 return acpi_processor_setup_cstates(pr); 1185 } 1186 1187 /** 1188 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE 1189 * device i.e. per-cpu data 1190 * 1191 * @pr: the ACPI processor 1192 * @dev : the cpuidle device 1193 */ 1194 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr, 1195 struct cpuidle_device *dev) 1196 { 1197 if (!pr->flags.power_setup_done || !pr->flags.power || !dev) 1198 return -EINVAL; 1199 1200 dev->cpu = pr->id; 1201 if (pr->flags.has_lpi) 1202 return acpi_processor_ffh_lpi_probe(pr->id); 1203 1204 return acpi_processor_setup_cpuidle_cx(pr, dev); 1205 } 1206 1207 static int acpi_processor_get_power_info(struct acpi_processor *pr) 1208 { 1209 int ret; 1210 1211 ret = acpi_processor_get_lpi_info(pr); 1212 if (ret) 1213 ret = acpi_processor_get_cstate_info(pr); 1214 1215 return ret; 1216 } 1217 1218 int acpi_processor_hotplug(struct acpi_processor *pr) 1219 { 1220 int ret = 0; 1221 struct cpuidle_device *dev; 1222 1223 if (disabled_by_idle_boot_param()) 1224 return 0; 1225 1226 if (!pr->flags.power_setup_done) 1227 return -ENODEV; 1228 1229 dev = per_cpu(acpi_cpuidle_device, pr->id); 1230 cpuidle_pause_and_lock(); 1231 cpuidle_disable_device(dev); 1232 ret = acpi_processor_get_power_info(pr); 1233 if (!ret && pr->flags.power) { 1234 acpi_processor_setup_cpuidle_dev(pr, dev); 1235 ret = cpuidle_enable_device(dev); 1236 } 1237 cpuidle_resume_and_unlock(); 1238 1239 return ret; 1240 } 1241 1242 int acpi_processor_power_state_has_changed(struct acpi_processor *pr) 1243 { 1244 int cpu; 1245 struct acpi_processor *_pr; 1246 struct cpuidle_device *dev; 1247 1248 if (disabled_by_idle_boot_param()) 1249 return 0; 1250 1251 if (!pr->flags.power_setup_done) 1252 return -ENODEV; 1253 1254 /* 1255 * FIXME: Design the ACPI notification to make it once per 1256 * system instead of once per-cpu. This condition is a hack 1257 * to make the code that updates C-States be called once. 1258 */ 1259 1260 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) { 1261 1262 /* Protect against cpu-hotplug */ 1263 get_online_cpus(); 1264 cpuidle_pause_and_lock(); 1265 1266 /* Disable 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 dev = per_cpu(acpi_cpuidle_device, cpu); 1272 cpuidle_disable_device(dev); 1273 } 1274 1275 /* Populate Updated C-state information */ 1276 acpi_processor_get_power_info(pr); 1277 acpi_processor_setup_cpuidle_states(pr); 1278 1279 /* Enable all cpuidle devices */ 1280 for_each_online_cpu(cpu) { 1281 _pr = per_cpu(processors, cpu); 1282 if (!_pr || !_pr->flags.power_setup_done) 1283 continue; 1284 acpi_processor_get_power_info(_pr); 1285 if (_pr->flags.power) { 1286 dev = per_cpu(acpi_cpuidle_device, cpu); 1287 acpi_processor_setup_cpuidle_dev(_pr, dev); 1288 cpuidle_enable_device(dev); 1289 } 1290 } 1291 cpuidle_resume_and_unlock(); 1292 put_online_cpus(); 1293 } 1294 1295 return 0; 1296 } 1297 1298 static int acpi_processor_registered; 1299 1300 int acpi_processor_power_init(struct acpi_processor *pr) 1301 { 1302 int retval; 1303 struct cpuidle_device *dev; 1304 1305 if (disabled_by_idle_boot_param()) 1306 return 0; 1307 1308 acpi_processor_cstate_first_run_checks(); 1309 1310 if (!acpi_processor_get_power_info(pr)) 1311 pr->flags.power_setup_done = 1; 1312 1313 /* 1314 * Install the idle handler if processor power management is supported. 1315 * Note that we use previously set idle handler will be used on 1316 * platforms that only support C1. 1317 */ 1318 if (pr->flags.power) { 1319 /* Register acpi_idle_driver if not already registered */ 1320 if (!acpi_processor_registered) { 1321 acpi_processor_setup_cpuidle_states(pr); 1322 retval = cpuidle_register_driver(&acpi_idle_driver); 1323 if (retval) 1324 return retval; 1325 pr_debug("%s registered with cpuidle\n", 1326 acpi_idle_driver.name); 1327 } 1328 1329 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1330 if (!dev) 1331 return -ENOMEM; 1332 per_cpu(acpi_cpuidle_device, pr->id) = dev; 1333 1334 acpi_processor_setup_cpuidle_dev(pr, dev); 1335 1336 /* Register per-cpu cpuidle_device. Cpuidle driver 1337 * must already be registered before registering device 1338 */ 1339 retval = cpuidle_register_device(dev); 1340 if (retval) { 1341 if (acpi_processor_registered == 0) 1342 cpuidle_unregister_driver(&acpi_idle_driver); 1343 return retval; 1344 } 1345 acpi_processor_registered++; 1346 } 1347 return 0; 1348 } 1349 1350 int acpi_processor_power_exit(struct acpi_processor *pr) 1351 { 1352 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id); 1353 1354 if (disabled_by_idle_boot_param()) 1355 return 0; 1356 1357 if (pr->flags.power) { 1358 cpuidle_unregister_device(dev); 1359 acpi_processor_registered--; 1360 if (acpi_processor_registered == 0) 1361 cpuidle_unregister_driver(&acpi_idle_driver); 1362 } 1363 1364 pr->flags.power_setup_done = 0; 1365 return 0; 1366 } 1367