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 #include <asm/cpu.h> 33 #endif 34 35 #define ACPI_IDLE_STATE_START (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0) 36 37 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER; 38 module_param(max_cstate, uint, 0000); 39 static unsigned int nocst __read_mostly; 40 module_param(nocst, uint, 0000); 41 static int bm_check_disable __read_mostly; 42 module_param(bm_check_disable, uint, 0000); 43 44 static unsigned int latency_factor __read_mostly = 2; 45 module_param(latency_factor, uint, 0644); 46 47 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device); 48 49 struct cpuidle_driver acpi_idle_driver = { 50 .name = "acpi_idle", 51 .owner = THIS_MODULE, 52 }; 53 54 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE 55 static 56 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate); 57 58 static int disabled_by_idle_boot_param(void) 59 { 60 return boot_option_idle_override == IDLE_POLL || 61 boot_option_idle_override == IDLE_HALT; 62 } 63 64 /* 65 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3. 66 * For now disable this. Probably a bug somewhere else. 67 * 68 * To skip this limit, boot/load with a large max_cstate limit. 69 */ 70 static int set_max_cstate(const struct dmi_system_id *id) 71 { 72 if (max_cstate > ACPI_PROCESSOR_MAX_POWER) 73 return 0; 74 75 pr_notice("%s detected - limiting to C%ld max_cstate." 76 " Override with \"processor.max_cstate=%d\"\n", id->ident, 77 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1); 78 79 max_cstate = (long)id->driver_data; 80 81 return 0; 82 } 83 84 static const struct dmi_system_id processor_power_dmi_table[] = { 85 { set_max_cstate, "Clevo 5600D", { 86 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"), 87 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")}, 88 (void *)2}, 89 { set_max_cstate, "Pavilion zv5000", { 90 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"), 91 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")}, 92 (void *)1}, 93 { set_max_cstate, "Asus L8400B", { 94 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."), 95 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")}, 96 (void *)1}, 97 {}, 98 }; 99 100 101 /* 102 * Callers should disable interrupts before the call and enable 103 * interrupts after return. 104 */ 105 static void __cpuidle acpi_safe_halt(void) 106 { 107 if (!tif_need_resched()) { 108 safe_halt(); 109 local_irq_disable(); 110 } 111 } 112 113 #ifdef ARCH_APICTIMER_STOPS_ON_C3 114 115 /* 116 * Some BIOS implementations switch to C3 in the published C2 state. 117 * This seems to be a common problem on AMD boxen, but other vendors 118 * are affected too. We pick the most conservative approach: we assume 119 * that the local APIC stops in both C2 and C3. 120 */ 121 static void lapic_timer_check_state(int state, struct acpi_processor *pr, 122 struct acpi_processor_cx *cx) 123 { 124 struct acpi_processor_power *pwr = &pr->power; 125 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2; 126 127 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT)) 128 return; 129 130 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) 131 type = ACPI_STATE_C1; 132 133 /* 134 * Check, if one of the previous states already marked the lapic 135 * unstable 136 */ 137 if (pwr->timer_broadcast_on_state < state) 138 return; 139 140 if (cx->type >= type) 141 pr->power.timer_broadcast_on_state = state; 142 } 143 144 static void __lapic_timer_propagate_broadcast(void *arg) 145 { 146 struct acpi_processor *pr = (struct acpi_processor *) arg; 147 148 if (pr->power.timer_broadcast_on_state < INT_MAX) 149 tick_broadcast_enable(); 150 else 151 tick_broadcast_disable(); 152 } 153 154 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) 155 { 156 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast, 157 (void *)pr, 1); 158 } 159 160 /* Power(C) State timer broadcast control */ 161 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr, 162 struct acpi_processor_cx *cx) 163 { 164 return cx - pr->power.states >= pr->power.timer_broadcast_on_state; 165 } 166 167 #else 168 169 static void lapic_timer_check_state(int state, struct acpi_processor *pr, 170 struct acpi_processor_cx *cstate) { } 171 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { } 172 173 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr, 174 struct acpi_processor_cx *cx) 175 { 176 return false; 177 } 178 179 #endif 180 181 #if defined(CONFIG_X86) 182 static void tsc_check_state(int state) 183 { 184 switch (boot_cpu_data.x86_vendor) { 185 case X86_VENDOR_HYGON: 186 case X86_VENDOR_AMD: 187 case X86_VENDOR_INTEL: 188 case X86_VENDOR_CENTAUR: 189 case X86_VENDOR_ZHAOXIN: 190 /* 191 * AMD Fam10h TSC will tick in all 192 * C/P/S0/S1 states when this bit is set. 193 */ 194 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 195 return; 196 fallthrough; 197 default: 198 /* TSC could halt in idle, so notify users */ 199 if (state > ACPI_STATE_C1) 200 mark_tsc_unstable("TSC halts in idle"); 201 } 202 } 203 #else 204 static void tsc_check_state(int state) { return; } 205 #endif 206 207 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr) 208 { 209 210 if (!pr->pblk) 211 return -ENODEV; 212 213 /* if info is obtained from pblk/fadt, type equals state */ 214 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2; 215 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3; 216 217 #ifndef CONFIG_HOTPLUG_CPU 218 /* 219 * Check for P_LVL2_UP flag before entering C2 and above on 220 * an SMP system. 221 */ 222 if ((num_online_cpus() > 1) && 223 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED)) 224 return -ENODEV; 225 #endif 226 227 /* determine C2 and C3 address from pblk */ 228 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4; 229 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5; 230 231 /* determine latencies from FADT */ 232 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency; 233 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency; 234 235 /* 236 * FADT specified C2 latency must be less than or equal to 237 * 100 microseconds. 238 */ 239 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) { 240 acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n", 241 acpi_gbl_FADT.c2_latency); 242 /* invalidate C2 */ 243 pr->power.states[ACPI_STATE_C2].address = 0; 244 } 245 246 /* 247 * FADT supplied C3 latency must be less than or equal to 248 * 1000 microseconds. 249 */ 250 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) { 251 acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n", 252 acpi_gbl_FADT.c3_latency); 253 /* invalidate C3 */ 254 pr->power.states[ACPI_STATE_C3].address = 0; 255 } 256 257 acpi_handle_debug(pr->handle, "lvl2[0x%08x] lvl3[0x%08x]\n", 258 pr->power.states[ACPI_STATE_C2].address, 259 pr->power.states[ACPI_STATE_C3].address); 260 261 snprintf(pr->power.states[ACPI_STATE_C2].desc, 262 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x", 263 pr->power.states[ACPI_STATE_C2].address); 264 snprintf(pr->power.states[ACPI_STATE_C3].desc, 265 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x", 266 pr->power.states[ACPI_STATE_C3].address); 267 268 return 0; 269 } 270 271 static int acpi_processor_get_power_info_default(struct acpi_processor *pr) 272 { 273 if (!pr->power.states[ACPI_STATE_C1].valid) { 274 /* set the first C-State to C1 */ 275 /* all processors need to support C1 */ 276 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1; 277 pr->power.states[ACPI_STATE_C1].valid = 1; 278 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT; 279 280 snprintf(pr->power.states[ACPI_STATE_C1].desc, 281 ACPI_CX_DESC_LEN, "ACPI HLT"); 282 } 283 /* the C0 state only exists as a filler in our array */ 284 pr->power.states[ACPI_STATE_C0].valid = 1; 285 return 0; 286 } 287 288 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr) 289 { 290 int ret; 291 292 if (nocst) 293 return -ENODEV; 294 295 ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power); 296 if (ret) 297 return ret; 298 299 if (!pr->power.count) 300 return -EFAULT; 301 302 pr->flags.has_cst = 1; 303 return 0; 304 } 305 306 static void acpi_processor_power_verify_c3(struct acpi_processor *pr, 307 struct acpi_processor_cx *cx) 308 { 309 static int bm_check_flag = -1; 310 static int bm_control_flag = -1; 311 312 313 if (!cx->address) 314 return; 315 316 /* 317 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast) 318 * DMA transfers are used by any ISA device to avoid livelock. 319 * Note that we could disable Type-F DMA (as recommended by 320 * the erratum), but this is known to disrupt certain ISA 321 * devices thus we take the conservative approach. 322 */ 323 else if (errata.piix4.fdma) { 324 acpi_handle_debug(pr->handle, 325 "C3 not supported on PIIX4 with Type-F DMA\n"); 326 return; 327 } 328 329 /* All the logic here assumes flags.bm_check is same across all CPUs */ 330 if (bm_check_flag == -1) { 331 /* Determine whether bm_check is needed based on CPU */ 332 acpi_processor_power_init_bm_check(&(pr->flags), pr->id); 333 bm_check_flag = pr->flags.bm_check; 334 bm_control_flag = pr->flags.bm_control; 335 } else { 336 pr->flags.bm_check = bm_check_flag; 337 pr->flags.bm_control = bm_control_flag; 338 } 339 340 if (pr->flags.bm_check) { 341 if (!pr->flags.bm_control) { 342 if (pr->flags.has_cst != 1) { 343 /* bus mastering control is necessary */ 344 acpi_handle_debug(pr->handle, 345 "C3 support requires BM control\n"); 346 return; 347 } else { 348 /* Here we enter C3 without bus mastering */ 349 acpi_handle_debug(pr->handle, 350 "C3 support without BM control\n"); 351 } 352 } 353 } else { 354 /* 355 * WBINVD should be set in fadt, for C3 state to be 356 * supported on when bm_check is not required. 357 */ 358 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) { 359 acpi_handle_debug(pr->handle, 360 "Cache invalidation should work properly" 361 " for C3 to be enabled on SMP systems\n"); 362 return; 363 } 364 } 365 366 /* 367 * Otherwise we've met all of our C3 requirements. 368 * Normalize the C3 latency to expidite policy. Enable 369 * checking of bus mastering status (bm_check) so we can 370 * use this in our C3 policy 371 */ 372 cx->valid = 1; 373 374 /* 375 * On older chipsets, BM_RLD needs to be set 376 * in order for Bus Master activity to wake the 377 * system from C3. Newer chipsets handle DMA 378 * during C3 automatically and BM_RLD is a NOP. 379 * In either case, the proper way to 380 * handle BM_RLD is to set it and leave it set. 381 */ 382 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1); 383 384 return; 385 } 386 387 static int acpi_processor_power_verify(struct acpi_processor *pr) 388 { 389 unsigned int i; 390 unsigned int working = 0; 391 392 pr->power.timer_broadcast_on_state = INT_MAX; 393 394 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 395 struct acpi_processor_cx *cx = &pr->power.states[i]; 396 397 switch (cx->type) { 398 case ACPI_STATE_C1: 399 cx->valid = 1; 400 break; 401 402 case ACPI_STATE_C2: 403 if (!cx->address) 404 break; 405 cx->valid = 1; 406 break; 407 408 case ACPI_STATE_C3: 409 acpi_processor_power_verify_c3(pr, cx); 410 break; 411 } 412 if (!cx->valid) 413 continue; 414 415 lapic_timer_check_state(i, pr, cx); 416 tsc_check_state(cx->type); 417 working++; 418 } 419 420 lapic_timer_propagate_broadcast(pr); 421 422 return (working); 423 } 424 425 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 426 { 427 unsigned int i; 428 int result; 429 430 431 /* NOTE: the idle thread may not be running while calling 432 * this function */ 433 434 /* Zero initialize all the C-states info. */ 435 memset(pr->power.states, 0, sizeof(pr->power.states)); 436 437 result = acpi_processor_get_power_info_cst(pr); 438 if (result == -ENODEV) 439 result = acpi_processor_get_power_info_fadt(pr); 440 441 if (result) 442 return result; 443 444 acpi_processor_get_power_info_default(pr); 445 446 pr->power.count = acpi_processor_power_verify(pr); 447 448 /* 449 * if one state of type C2 or C3 is available, mark this 450 * CPU as being "idle manageable" 451 */ 452 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) { 453 if (pr->power.states[i].valid) { 454 pr->power.count = i; 455 pr->flags.power = 1; 456 } 457 } 458 459 return 0; 460 } 461 462 /** 463 * acpi_idle_bm_check - checks if bus master activity was detected 464 */ 465 static int acpi_idle_bm_check(void) 466 { 467 u32 bm_status = 0; 468 469 if (bm_check_disable) 470 return 0; 471 472 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status); 473 if (bm_status) 474 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1); 475 /* 476 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect 477 * the true state of bus mastering activity; forcing us to 478 * manually check the BMIDEA bit of each IDE channel. 479 */ 480 else if (errata.piix4.bmisx) { 481 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01) 482 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01)) 483 bm_status = 1; 484 } 485 return bm_status; 486 } 487 488 static void wait_for_freeze(void) 489 { 490 #ifdef CONFIG_X86 491 /* No delay is needed if we are in guest */ 492 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 493 return; 494 #endif 495 /* Dummy wait op - must do something useless after P_LVL2 read 496 because chipsets cannot guarantee that STPCLK# signal 497 gets asserted in time to freeze execution properly. */ 498 inl(acpi_gbl_FADT.xpm_timer_block.address); 499 } 500 501 /** 502 * acpi_idle_do_entry - enter idle state using the appropriate method 503 * @cx: cstate data 504 * 505 * Caller disables interrupt before call and enables interrupt after return. 506 */ 507 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx) 508 { 509 if (cx->entry_method == ACPI_CSTATE_FFH) { 510 /* Call into architectural FFH based C-state */ 511 acpi_processor_ffh_cstate_enter(cx); 512 } else if (cx->entry_method == ACPI_CSTATE_HALT) { 513 acpi_safe_halt(); 514 } else { 515 /* IO port based C-state */ 516 inb(cx->address); 517 wait_for_freeze(); 518 } 519 } 520 521 /** 522 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining) 523 * @dev: the target CPU 524 * @index: the index of suggested state 525 */ 526 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index) 527 { 528 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 529 530 ACPI_FLUSH_CPU_CACHE(); 531 532 while (1) { 533 534 if (cx->entry_method == ACPI_CSTATE_HALT) 535 safe_halt(); 536 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) { 537 inb(cx->address); 538 wait_for_freeze(); 539 } else 540 return -ENODEV; 541 542 #if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU) 543 cond_wakeup_cpu0(); 544 #endif 545 } 546 547 /* Never reached */ 548 return 0; 549 } 550 551 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr) 552 { 553 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst && 554 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED); 555 } 556 557 static int c3_cpu_count; 558 static DEFINE_RAW_SPINLOCK(c3_lock); 559 560 /** 561 * acpi_idle_enter_bm - enters C3 with proper BM handling 562 * @drv: cpuidle driver 563 * @pr: Target processor 564 * @cx: Target state context 565 * @index: index of target state 566 */ 567 static int acpi_idle_enter_bm(struct cpuidle_driver *drv, 568 struct acpi_processor *pr, 569 struct acpi_processor_cx *cx, 570 int index) 571 { 572 static struct acpi_processor_cx safe_cx = { 573 .entry_method = ACPI_CSTATE_HALT, 574 }; 575 576 /* 577 * disable bus master 578 * bm_check implies we need ARB_DIS 579 * bm_control implies whether we can do ARB_DIS 580 * 581 * That leaves a case where bm_check is set and bm_control is not set. 582 * In that case we cannot do much, we enter C3 without doing anything. 583 */ 584 bool dis_bm = pr->flags.bm_control; 585 586 /* If we can skip BM, demote to a safe state. */ 587 if (!cx->bm_sts_skip && acpi_idle_bm_check()) { 588 dis_bm = false; 589 index = drv->safe_state_index; 590 if (index >= 0) { 591 cx = this_cpu_read(acpi_cstate[index]); 592 } else { 593 cx = &safe_cx; 594 index = -EBUSY; 595 } 596 } 597 598 if (dis_bm) { 599 raw_spin_lock(&c3_lock); 600 c3_cpu_count++; 601 /* Disable bus master arbitration when all CPUs are in C3 */ 602 if (c3_cpu_count == num_online_cpus()) 603 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1); 604 raw_spin_unlock(&c3_lock); 605 } 606 607 rcu_idle_enter(); 608 609 acpi_idle_do_entry(cx); 610 611 rcu_idle_exit(); 612 613 /* Re-enable bus master arbitration */ 614 if (dis_bm) { 615 raw_spin_lock(&c3_lock); 616 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0); 617 c3_cpu_count--; 618 raw_spin_unlock(&c3_lock); 619 } 620 621 return index; 622 } 623 624 static int acpi_idle_enter(struct cpuidle_device *dev, 625 struct cpuidle_driver *drv, int index) 626 { 627 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 628 struct acpi_processor *pr; 629 630 pr = __this_cpu_read(processors); 631 if (unlikely(!pr)) 632 return -EINVAL; 633 634 if (cx->type != ACPI_STATE_C1) { 635 if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) 636 return acpi_idle_enter_bm(drv, pr, cx, index); 637 638 /* C2 to C1 demotion. */ 639 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) { 640 index = ACPI_IDLE_STATE_START; 641 cx = per_cpu(acpi_cstate[index], dev->cpu); 642 } 643 } 644 645 if (cx->type == ACPI_STATE_C3) 646 ACPI_FLUSH_CPU_CACHE(); 647 648 acpi_idle_do_entry(cx); 649 650 return index; 651 } 652 653 static int acpi_idle_enter_s2idle(struct cpuidle_device *dev, 654 struct cpuidle_driver *drv, int index) 655 { 656 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu); 657 658 if (cx->type == ACPI_STATE_C3) { 659 struct acpi_processor *pr = __this_cpu_read(processors); 660 661 if (unlikely(!pr)) 662 return 0; 663 664 if (pr->flags.bm_check) { 665 u8 bm_sts_skip = cx->bm_sts_skip; 666 667 /* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */ 668 cx->bm_sts_skip = 1; 669 acpi_idle_enter_bm(drv, pr, cx, index); 670 cx->bm_sts_skip = bm_sts_skip; 671 672 return 0; 673 } else { 674 ACPI_FLUSH_CPU_CACHE(); 675 } 676 } 677 acpi_idle_do_entry(cx); 678 679 return 0; 680 } 681 682 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 683 struct cpuidle_device *dev) 684 { 685 int i, count = ACPI_IDLE_STATE_START; 686 struct acpi_processor_cx *cx; 687 struct cpuidle_state *state; 688 689 if (max_cstate == 0) 690 max_cstate = 1; 691 692 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 693 state = &acpi_idle_driver.states[count]; 694 cx = &pr->power.states[i]; 695 696 if (!cx->valid) 697 continue; 698 699 per_cpu(acpi_cstate[count], dev->cpu) = cx; 700 701 if (lapic_timer_needs_broadcast(pr, cx)) 702 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 703 704 if (cx->type == ACPI_STATE_C3) { 705 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED; 706 if (pr->flags.bm_check) 707 state->flags |= CPUIDLE_FLAG_RCU_IDLE; 708 } 709 710 count++; 711 if (count == CPUIDLE_STATE_MAX) 712 break; 713 } 714 715 if (!count) 716 return -EINVAL; 717 718 return 0; 719 } 720 721 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 722 { 723 int i, count; 724 struct acpi_processor_cx *cx; 725 struct cpuidle_state *state; 726 struct cpuidle_driver *drv = &acpi_idle_driver; 727 728 if (max_cstate == 0) 729 max_cstate = 1; 730 731 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) { 732 cpuidle_poll_state_init(drv); 733 count = 1; 734 } else { 735 count = 0; 736 } 737 738 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) { 739 cx = &pr->power.states[i]; 740 741 if (!cx->valid) 742 continue; 743 744 state = &drv->states[count]; 745 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i); 746 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN); 747 state->exit_latency = cx->latency; 748 state->target_residency = cx->latency * latency_factor; 749 state->enter = acpi_idle_enter; 750 751 state->flags = 0; 752 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) { 753 state->enter_dead = acpi_idle_play_dead; 754 drv->safe_state_index = count; 755 } 756 /* 757 * Halt-induced C1 is not good for ->enter_s2idle, because it 758 * re-enables interrupts on exit. Moreover, C1 is generally not 759 * particularly interesting from the suspend-to-idle angle, so 760 * avoid C1 and the situations in which we may need to fall back 761 * to it altogether. 762 */ 763 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr)) 764 state->enter_s2idle = acpi_idle_enter_s2idle; 765 766 count++; 767 if (count == CPUIDLE_STATE_MAX) 768 break; 769 } 770 771 drv->state_count = count; 772 773 if (!count) 774 return -EINVAL; 775 776 return 0; 777 } 778 779 static inline void acpi_processor_cstate_first_run_checks(void) 780 { 781 static int first_run; 782 783 if (first_run) 784 return; 785 dmi_check_system(processor_power_dmi_table); 786 max_cstate = acpi_processor_cstate_check(max_cstate); 787 if (max_cstate < ACPI_C_STATES_MAX) 788 pr_notice("processor limited to max C-state %d\n", max_cstate); 789 790 first_run++; 791 792 if (nocst) 793 return; 794 795 acpi_processor_claim_cst_control(); 796 } 797 #else 798 799 static inline int disabled_by_idle_boot_param(void) { return 0; } 800 static inline void acpi_processor_cstate_first_run_checks(void) { } 801 static int acpi_processor_get_cstate_info(struct acpi_processor *pr) 802 { 803 return -ENODEV; 804 } 805 806 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr, 807 struct cpuidle_device *dev) 808 { 809 return -EINVAL; 810 } 811 812 static int acpi_processor_setup_cstates(struct acpi_processor *pr) 813 { 814 return -EINVAL; 815 } 816 817 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */ 818 819 struct acpi_lpi_states_array { 820 unsigned int size; 821 unsigned int composite_states_size; 822 struct acpi_lpi_state *entries; 823 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER]; 824 }; 825 826 static int obj_get_integer(union acpi_object *obj, u32 *value) 827 { 828 if (obj->type != ACPI_TYPE_INTEGER) 829 return -EINVAL; 830 831 *value = obj->integer.value; 832 return 0; 833 } 834 835 static int acpi_processor_evaluate_lpi(acpi_handle handle, 836 struct acpi_lpi_states_array *info) 837 { 838 acpi_status status; 839 int ret = 0; 840 int pkg_count, state_idx = 1, loop; 841 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL }; 842 union acpi_object *lpi_data; 843 struct acpi_lpi_state *lpi_state; 844 845 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer); 846 if (ACPI_FAILURE(status)) { 847 acpi_handle_debug(handle, "No _LPI, giving up\n"); 848 return -ENODEV; 849 } 850 851 lpi_data = buffer.pointer; 852 853 /* There must be at least 4 elements = 3 elements + 1 package */ 854 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE || 855 lpi_data->package.count < 4) { 856 pr_debug("not enough elements in _LPI\n"); 857 ret = -ENODATA; 858 goto end; 859 } 860 861 pkg_count = lpi_data->package.elements[2].integer.value; 862 863 /* Validate number of power states. */ 864 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) { 865 pr_debug("count given by _LPI is not valid\n"); 866 ret = -ENODATA; 867 goto end; 868 } 869 870 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL); 871 if (!lpi_state) { 872 ret = -ENOMEM; 873 goto end; 874 } 875 876 info->size = pkg_count; 877 info->entries = lpi_state; 878 879 /* LPI States start at index 3 */ 880 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) { 881 union acpi_object *element, *pkg_elem, *obj; 882 883 element = &lpi_data->package.elements[loop]; 884 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7) 885 continue; 886 887 pkg_elem = element->package.elements; 888 889 obj = pkg_elem + 6; 890 if (obj->type == ACPI_TYPE_BUFFER) { 891 struct acpi_power_register *reg; 892 893 reg = (struct acpi_power_register *)obj->buffer.pointer; 894 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO && 895 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) 896 continue; 897 898 lpi_state->address = reg->address; 899 lpi_state->entry_method = 900 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ? 901 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO; 902 } else if (obj->type == ACPI_TYPE_INTEGER) { 903 lpi_state->entry_method = ACPI_CSTATE_INTEGER; 904 lpi_state->address = obj->integer.value; 905 } else { 906 continue; 907 } 908 909 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/ 910 911 obj = pkg_elem + 9; 912 if (obj->type == ACPI_TYPE_STRING) 913 strlcpy(lpi_state->desc, obj->string.pointer, 914 ACPI_CX_DESC_LEN); 915 916 lpi_state->index = state_idx; 917 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) { 918 pr_debug("No min. residency found, assuming 10 us\n"); 919 lpi_state->min_residency = 10; 920 } 921 922 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) { 923 pr_debug("No wakeup residency found, assuming 10 us\n"); 924 lpi_state->wake_latency = 10; 925 } 926 927 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags)) 928 lpi_state->flags = 0; 929 930 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags)) 931 lpi_state->arch_flags = 0; 932 933 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq)) 934 lpi_state->res_cnt_freq = 1; 935 936 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state)) 937 lpi_state->enable_parent_state = 0; 938 } 939 940 acpi_handle_debug(handle, "Found %d power states\n", state_idx); 941 end: 942 kfree(buffer.pointer); 943 return ret; 944 } 945 946 /* 947 * flat_state_cnt - the number of composite LPI states after the process of flattening 948 */ 949 static int flat_state_cnt; 950 951 /** 952 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state 953 * 954 * @local: local LPI state 955 * @parent: parent LPI state 956 * @result: composite LPI state 957 */ 958 static bool combine_lpi_states(struct acpi_lpi_state *local, 959 struct acpi_lpi_state *parent, 960 struct acpi_lpi_state *result) 961 { 962 if (parent->entry_method == ACPI_CSTATE_INTEGER) { 963 if (!parent->address) /* 0 means autopromotable */ 964 return false; 965 result->address = local->address + parent->address; 966 } else { 967 result->address = parent->address; 968 } 969 970 result->min_residency = max(local->min_residency, parent->min_residency); 971 result->wake_latency = local->wake_latency + parent->wake_latency; 972 result->enable_parent_state = parent->enable_parent_state; 973 result->entry_method = local->entry_method; 974 975 result->flags = parent->flags; 976 result->arch_flags = parent->arch_flags; 977 result->index = parent->index; 978 979 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN); 980 strlcat(result->desc, "+", ACPI_CX_DESC_LEN); 981 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN); 982 return true; 983 } 984 985 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0) 986 987 static void stash_composite_state(struct acpi_lpi_states_array *curr_level, 988 struct acpi_lpi_state *t) 989 { 990 curr_level->composite_states[curr_level->composite_states_size++] = t; 991 } 992 993 static int flatten_lpi_states(struct acpi_processor *pr, 994 struct acpi_lpi_states_array *curr_level, 995 struct acpi_lpi_states_array *prev_level) 996 { 997 int i, j, state_count = curr_level->size; 998 struct acpi_lpi_state *p, *t = curr_level->entries; 999 1000 curr_level->composite_states_size = 0; 1001 for (j = 0; j < state_count; j++, t++) { 1002 struct acpi_lpi_state *flpi; 1003 1004 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED)) 1005 continue; 1006 1007 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) { 1008 pr_warn("Limiting number of LPI states to max (%d)\n", 1009 ACPI_PROCESSOR_MAX_POWER); 1010 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n"); 1011 break; 1012 } 1013 1014 flpi = &pr->power.lpi_states[flat_state_cnt]; 1015 1016 if (!prev_level) { /* leaf/processor node */ 1017 memcpy(flpi, t, sizeof(*t)); 1018 stash_composite_state(curr_level, flpi); 1019 flat_state_cnt++; 1020 continue; 1021 } 1022 1023 for (i = 0; i < prev_level->composite_states_size; i++) { 1024 p = prev_level->composite_states[i]; 1025 if (t->index <= p->enable_parent_state && 1026 combine_lpi_states(p, t, flpi)) { 1027 stash_composite_state(curr_level, flpi); 1028 flat_state_cnt++; 1029 flpi++; 1030 } 1031 } 1032 } 1033 1034 kfree(curr_level->entries); 1035 return 0; 1036 } 1037 1038 static int acpi_processor_get_lpi_info(struct acpi_processor *pr) 1039 { 1040 int ret, i; 1041 acpi_status status; 1042 acpi_handle handle = pr->handle, pr_ahandle; 1043 struct acpi_device *d = NULL; 1044 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr; 1045 1046 if (!osc_pc_lpi_support_confirmed) 1047 return -EOPNOTSUPP; 1048 1049 if (!acpi_has_method(handle, "_LPI")) 1050 return -EINVAL; 1051 1052 flat_state_cnt = 0; 1053 prev = &info[0]; 1054 curr = &info[1]; 1055 handle = pr->handle; 1056 ret = acpi_processor_evaluate_lpi(handle, prev); 1057 if (ret) 1058 return ret; 1059 flatten_lpi_states(pr, prev, NULL); 1060 1061 status = acpi_get_parent(handle, &pr_ahandle); 1062 while (ACPI_SUCCESS(status)) { 1063 acpi_bus_get_device(pr_ahandle, &d); 1064 handle = pr_ahandle; 1065 1066 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID)) 1067 break; 1068 1069 /* can be optional ? */ 1070 if (!acpi_has_method(handle, "_LPI")) 1071 break; 1072 1073 ret = acpi_processor_evaluate_lpi(handle, curr); 1074 if (ret) 1075 break; 1076 1077 /* flatten all the LPI states in this level of hierarchy */ 1078 flatten_lpi_states(pr, curr, prev); 1079 1080 tmp = prev, prev = curr, curr = tmp; 1081 1082 status = acpi_get_parent(handle, &pr_ahandle); 1083 } 1084 1085 pr->power.count = flat_state_cnt; 1086 /* reset the index after flattening */ 1087 for (i = 0; i < pr->power.count; i++) 1088 pr->power.lpi_states[i].index = i; 1089 1090 /* Tell driver that _LPI is supported. */ 1091 pr->flags.has_lpi = 1; 1092 pr->flags.power = 1; 1093 1094 return 0; 1095 } 1096 1097 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu) 1098 { 1099 return -ENODEV; 1100 } 1101 1102 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi) 1103 { 1104 return -ENODEV; 1105 } 1106 1107 /** 1108 * acpi_idle_lpi_enter - enters an ACPI any LPI state 1109 * @dev: the target CPU 1110 * @drv: cpuidle driver containing cpuidle state info 1111 * @index: index of target state 1112 * 1113 * Return: 0 for success or negative value for error 1114 */ 1115 static int acpi_idle_lpi_enter(struct cpuidle_device *dev, 1116 struct cpuidle_driver *drv, int index) 1117 { 1118 struct acpi_processor *pr; 1119 struct acpi_lpi_state *lpi; 1120 1121 pr = __this_cpu_read(processors); 1122 1123 if (unlikely(!pr)) 1124 return -EINVAL; 1125 1126 lpi = &pr->power.lpi_states[index]; 1127 if (lpi->entry_method == ACPI_CSTATE_FFH) 1128 return acpi_processor_ffh_lpi_enter(lpi); 1129 1130 return -EINVAL; 1131 } 1132 1133 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr) 1134 { 1135 int i; 1136 struct acpi_lpi_state *lpi; 1137 struct cpuidle_state *state; 1138 struct cpuidle_driver *drv = &acpi_idle_driver; 1139 1140 if (!pr->flags.has_lpi) 1141 return -EOPNOTSUPP; 1142 1143 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) { 1144 lpi = &pr->power.lpi_states[i]; 1145 1146 state = &drv->states[i]; 1147 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i); 1148 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN); 1149 state->exit_latency = lpi->wake_latency; 1150 state->target_residency = lpi->min_residency; 1151 if (lpi->arch_flags) 1152 state->flags |= CPUIDLE_FLAG_TIMER_STOP; 1153 state->enter = acpi_idle_lpi_enter; 1154 drv->safe_state_index = i; 1155 } 1156 1157 drv->state_count = i; 1158 1159 return 0; 1160 } 1161 1162 /** 1163 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle 1164 * global state data i.e. idle routines 1165 * 1166 * @pr: the ACPI processor 1167 */ 1168 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr) 1169 { 1170 int i; 1171 struct cpuidle_driver *drv = &acpi_idle_driver; 1172 1173 if (!pr->flags.power_setup_done || !pr->flags.power) 1174 return -EINVAL; 1175 1176 drv->safe_state_index = -1; 1177 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) { 1178 drv->states[i].name[0] = '\0'; 1179 drv->states[i].desc[0] = '\0'; 1180 } 1181 1182 if (pr->flags.has_lpi) 1183 return acpi_processor_setup_lpi_states(pr); 1184 1185 return acpi_processor_setup_cstates(pr); 1186 } 1187 1188 /** 1189 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE 1190 * device i.e. per-cpu data 1191 * 1192 * @pr: the ACPI processor 1193 * @dev : the cpuidle device 1194 */ 1195 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr, 1196 struct cpuidle_device *dev) 1197 { 1198 if (!pr->flags.power_setup_done || !pr->flags.power || !dev) 1199 return -EINVAL; 1200 1201 dev->cpu = pr->id; 1202 if (pr->flags.has_lpi) 1203 return acpi_processor_ffh_lpi_probe(pr->id); 1204 1205 return acpi_processor_setup_cpuidle_cx(pr, dev); 1206 } 1207 1208 static int acpi_processor_get_power_info(struct acpi_processor *pr) 1209 { 1210 int ret; 1211 1212 ret = acpi_processor_get_lpi_info(pr); 1213 if (ret) 1214 ret = acpi_processor_get_cstate_info(pr); 1215 1216 return ret; 1217 } 1218 1219 int acpi_processor_hotplug(struct acpi_processor *pr) 1220 { 1221 int ret = 0; 1222 struct cpuidle_device *dev; 1223 1224 if (disabled_by_idle_boot_param()) 1225 return 0; 1226 1227 if (!pr->flags.power_setup_done) 1228 return -ENODEV; 1229 1230 dev = per_cpu(acpi_cpuidle_device, pr->id); 1231 cpuidle_pause_and_lock(); 1232 cpuidle_disable_device(dev); 1233 ret = acpi_processor_get_power_info(pr); 1234 if (!ret && pr->flags.power) { 1235 acpi_processor_setup_cpuidle_dev(pr, dev); 1236 ret = cpuidle_enable_device(dev); 1237 } 1238 cpuidle_resume_and_unlock(); 1239 1240 return ret; 1241 } 1242 1243 int acpi_processor_power_state_has_changed(struct acpi_processor *pr) 1244 { 1245 int cpu; 1246 struct acpi_processor *_pr; 1247 struct cpuidle_device *dev; 1248 1249 if (disabled_by_idle_boot_param()) 1250 return 0; 1251 1252 if (!pr->flags.power_setup_done) 1253 return -ENODEV; 1254 1255 /* 1256 * FIXME: Design the ACPI notification to make it once per 1257 * system instead of once per-cpu. This condition is a hack 1258 * to make the code that updates C-States be called once. 1259 */ 1260 1261 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) { 1262 1263 /* Protect against cpu-hotplug */ 1264 get_online_cpus(); 1265 cpuidle_pause_and_lock(); 1266 1267 /* Disable 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 dev = per_cpu(acpi_cpuidle_device, cpu); 1273 cpuidle_disable_device(dev); 1274 } 1275 1276 /* Populate Updated C-state information */ 1277 acpi_processor_get_power_info(pr); 1278 acpi_processor_setup_cpuidle_states(pr); 1279 1280 /* Enable all cpuidle devices */ 1281 for_each_online_cpu(cpu) { 1282 _pr = per_cpu(processors, cpu); 1283 if (!_pr || !_pr->flags.power_setup_done) 1284 continue; 1285 acpi_processor_get_power_info(_pr); 1286 if (_pr->flags.power) { 1287 dev = per_cpu(acpi_cpuidle_device, cpu); 1288 acpi_processor_setup_cpuidle_dev(_pr, dev); 1289 cpuidle_enable_device(dev); 1290 } 1291 } 1292 cpuidle_resume_and_unlock(); 1293 put_online_cpus(); 1294 } 1295 1296 return 0; 1297 } 1298 1299 static int acpi_processor_registered; 1300 1301 int acpi_processor_power_init(struct acpi_processor *pr) 1302 { 1303 int retval; 1304 struct cpuidle_device *dev; 1305 1306 if (disabled_by_idle_boot_param()) 1307 return 0; 1308 1309 acpi_processor_cstate_first_run_checks(); 1310 1311 if (!acpi_processor_get_power_info(pr)) 1312 pr->flags.power_setup_done = 1; 1313 1314 /* 1315 * Install the idle handler if processor power management is supported. 1316 * Note that we use previously set idle handler will be used on 1317 * platforms that only support C1. 1318 */ 1319 if (pr->flags.power) { 1320 /* Register acpi_idle_driver if not already registered */ 1321 if (!acpi_processor_registered) { 1322 acpi_processor_setup_cpuidle_states(pr); 1323 retval = cpuidle_register_driver(&acpi_idle_driver); 1324 if (retval) 1325 return retval; 1326 pr_debug("%s registered with cpuidle\n", 1327 acpi_idle_driver.name); 1328 } 1329 1330 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 1331 if (!dev) 1332 return -ENOMEM; 1333 per_cpu(acpi_cpuidle_device, pr->id) = dev; 1334 1335 acpi_processor_setup_cpuidle_dev(pr, dev); 1336 1337 /* Register per-cpu cpuidle_device. Cpuidle driver 1338 * must already be registered before registering device 1339 */ 1340 retval = cpuidle_register_device(dev); 1341 if (retval) { 1342 if (acpi_processor_registered == 0) 1343 cpuidle_unregister_driver(&acpi_idle_driver); 1344 return retval; 1345 } 1346 acpi_processor_registered++; 1347 } 1348 return 0; 1349 } 1350 1351 int acpi_processor_power_exit(struct acpi_processor *pr) 1352 { 1353 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id); 1354 1355 if (disabled_by_idle_boot_param()) 1356 return 0; 1357 1358 if (pr->flags.power) { 1359 cpuidle_unregister_device(dev); 1360 acpi_processor_registered--; 1361 if (acpi_processor_registered == 0) 1362 cpuidle_unregister_driver(&acpi_idle_driver); 1363 } 1364 1365 pr->flags.power_setup_done = 0; 1366 return 0; 1367 } 1368