Linux Audio

Check our new training course

Loading...
v6.13.7
   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 <linux/minmax.h>
  23#include <linux/perf_event.h>
  24#include <acpi/processor.h>
  25#include <linux/context_tracking.h>
  26
  27/*
  28 * Include the apic definitions for x86 to have the APIC timer related defines
  29 * available also for UP (on SMP it gets magically included via linux/smp.h).
  30 * asm/acpi.h is not an option, as it would require more include magic. Also
  31 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  32 */
  33#ifdef CONFIG_X86
  34#include <asm/apic.h>
  35#include <asm/cpu.h>
  36#endif
  37
  38#define ACPI_IDLE_STATE_START	(IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
  39
  40static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  41module_param(max_cstate, uint, 0400);
  42static bool nocst __read_mostly;
  43module_param(nocst, bool, 0400);
  44static bool bm_check_disable __read_mostly;
  45module_param(bm_check_disable, bool, 0400);
  46
  47static unsigned int latency_factor __read_mostly = 2;
  48module_param(latency_factor, uint, 0644);
  49
  50static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
  51
  52struct cpuidle_driver acpi_idle_driver = {
  53	.name =		"acpi_idle",
  54	.owner =	THIS_MODULE,
  55};
  56
  57#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
  58static
  59DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
  60
  61static 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 */
  73static 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
  87static 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 */
 108static void __cpuidle acpi_safe_halt(void)
 109{
 110	if (!tif_need_resched()) {
 111		raw_safe_halt();
 112		raw_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 */
 124static 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
 147static void __lapic_timer_propagate_broadcast(void *arg)
 148{
 149	struct acpi_processor *pr = arg;
 150
 151	if (pr->power.timer_broadcast_on_state < INT_MAX)
 152		tick_broadcast_enable();
 153	else
 154		tick_broadcast_disable();
 155}
 156
 157static 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 */
 164static 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
 172static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 173				   struct acpi_processor_cx *cstate) { }
 174static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 175
 176static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 177					struct acpi_processor_cx *cx)
 178{
 179	return false;
 180}
 181
 182#endif
 183
 184#if defined(CONFIG_X86)
 185static void tsc_check_state(int state)
 186{
 187	switch (boot_cpu_data.x86_vendor) {
 188	case X86_VENDOR_HYGON:
 189	case X86_VENDOR_AMD:
 190	case X86_VENDOR_INTEL:
 191	case X86_VENDOR_CENTAUR:
 192	case X86_VENDOR_ZHAOXIN:
 193		/*
 194		 * AMD Fam10h TSC will tick in all
 195		 * C/P/S0/S1 states when this bit is set.
 196		 */
 197		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 198			return;
 199		fallthrough;
 200	default:
 201		/* TSC could halt in idle, so notify users */
 202		if (state > ACPI_STATE_C1)
 203			mark_tsc_unstable("TSC halts in idle");
 204	}
 205}
 206#else
 207static void tsc_check_state(int state) { return; }
 208#endif
 209
 210static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 211{
 212
 213	if (!pr->pblk)
 214		return -ENODEV;
 215
 216	/* if info is obtained from pblk/fadt, type equals state */
 217	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 218	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 219
 220#ifndef CONFIG_HOTPLUG_CPU
 221	/*
 222	 * Check for P_LVL2_UP flag before entering C2 and above on
 223	 * an SMP system.
 224	 */
 225	if ((num_online_cpus() > 1) &&
 226	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 227		return -ENODEV;
 228#endif
 229
 230	/* determine C2 and C3 address from pblk */
 231	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 232	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 233
 234	/* determine latencies from FADT */
 235	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
 236	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
 237
 238	/*
 239	 * FADT specified C2 latency must be less than or equal to
 240	 * 100 microseconds.
 241	 */
 242	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 243		acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n",
 244				  acpi_gbl_FADT.c2_latency);
 245		/* invalidate C2 */
 246		pr->power.states[ACPI_STATE_C2].address = 0;
 247	}
 248
 249	/*
 250	 * FADT supplied C3 latency must be less than or equal to
 251	 * 1000 microseconds.
 252	 */
 253	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 254		acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n",
 255				  acpi_gbl_FADT.c3_latency);
 256		/* invalidate C3 */
 257		pr->power.states[ACPI_STATE_C3].address = 0;
 258	}
 259
 260	acpi_handle_debug(pr->handle, "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
 274static 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
 291static 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
 309static 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	if (errata.piix4.fdma) {
 327		acpi_handle_debug(pr->handle,
 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_handle_debug(pr->handle,
 348						  "C3 support requires BM control\n");
 349				return;
 350			} else {
 351				/* Here we enter C3 without bus mastering */
 352				acpi_handle_debug(pr->handle,
 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_handle_debug(pr->handle,
 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
 388static void acpi_cst_latency_sort(struct acpi_processor_cx *states, size_t length)
 389{
 390	int i, j, k;
 391
 392	for (i = 1; i < length; i++) {
 393		if (!states[i].valid)
 394			continue;
 395
 396		for (j = i - 1, k = i; j >= 0; j--) {
 397			if (!states[j].valid)
 398				continue;
 399
 400			if (states[j].latency > states[k].latency)
 401				swap(states[j].latency, states[k].latency);
 
 402
 403			k = j;
 404		}
 405	}
 406}
 407
 408static int acpi_processor_power_verify(struct acpi_processor *pr)
 409{
 410	unsigned int i;
 411	unsigned int working = 0;
 412	unsigned int last_latency = 0;
 413	unsigned int last_type = 0;
 414	bool buggy_latency = false;
 415
 416	pr->power.timer_broadcast_on_state = INT_MAX;
 417
 418	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 419		struct acpi_processor_cx *cx = &pr->power.states[i];
 420
 421		switch (cx->type) {
 422		case ACPI_STATE_C1:
 423			cx->valid = 1;
 424			break;
 425
 426		case ACPI_STATE_C2:
 427			if (!cx->address)
 428				break;
 429			cx->valid = 1;
 430			break;
 431
 432		case ACPI_STATE_C3:
 433			acpi_processor_power_verify_c3(pr, cx);
 434			break;
 435		}
 436		if (!cx->valid)
 437			continue;
 438		if (cx->type >= last_type && cx->latency < last_latency)
 439			buggy_latency = true;
 440		last_latency = cx->latency;
 441		last_type = cx->type;
 442
 443		lapic_timer_check_state(i, pr, cx);
 444		tsc_check_state(cx->type);
 445		working++;
 446	}
 447
 448	if (buggy_latency) {
 449		pr_notice("FW issue: working around C-state latencies out of order\n");
 450		acpi_cst_latency_sort(&pr->power.states[1], max_cstate);
 
 
 
 451	}
 452
 453	lapic_timer_propagate_broadcast(pr);
 454
 455	return working;
 456}
 457
 458static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 459{
 460	unsigned int i;
 461	int result;
 462
 463
 464	/* NOTE: the idle thread may not be running while calling
 465	 * this function */
 466
 467	/* Zero initialize all the C-states info. */
 468	memset(pr->power.states, 0, sizeof(pr->power.states));
 469
 470	result = acpi_processor_get_power_info_cst(pr);
 471	if (result == -ENODEV)
 472		result = acpi_processor_get_power_info_fadt(pr);
 473
 474	if (result)
 475		return result;
 476
 477	acpi_processor_get_power_info_default(pr);
 478
 479	pr->power.count = acpi_processor_power_verify(pr);
 480
 481	/*
 482	 * if one state of type C2 or C3 is available, mark this
 483	 * CPU as being "idle manageable"
 484	 */
 485	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 486		if (pr->power.states[i].valid) {
 487			pr->power.count = i;
 488			pr->flags.power = 1;
 489		}
 490	}
 491
 492	return 0;
 493}
 494
 495/**
 496 * acpi_idle_bm_check - checks if bus master activity was detected
 497 */
 498static int acpi_idle_bm_check(void)
 499{
 500	u32 bm_status = 0;
 501
 502	if (bm_check_disable)
 503		return 0;
 504
 505	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 506	if (bm_status)
 507		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 508	/*
 509	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 510	 * the true state of bus mastering activity; forcing us to
 511	 * manually check the BMIDEA bit of each IDE channel.
 512	 */
 513	else if (errata.piix4.bmisx) {
 514		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 515		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 516			bm_status = 1;
 517	}
 518	return bm_status;
 519}
 520
 521static __cpuidle void io_idle(unsigned long addr)
 522{
 523	/* IO port based C-state */
 524	inb(addr);
 525
 526#ifdef	CONFIG_X86
 527	/* No delay is needed if we are in guest */
 528	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 529		return;
 530	/*
 531	 * Modern (>=Nehalem) Intel systems use ACPI via intel_idle,
 532	 * not this code.  Assume that any Intel systems using this
 533	 * are ancient and may need the dummy wait.  This also assumes
 534	 * that the motivating chipset issue was Intel-only.
 535	 */
 536	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
 537		return;
 538#endif
 539	/*
 540	 * Dummy wait op - must do something useless after P_LVL2 read
 541	 * because chipsets cannot guarantee that STPCLK# signal gets
 542	 * asserted in time to freeze execution properly
 543	 *
 544	 * This workaround has been in place since the original ACPI
 545	 * implementation was merged, circa 2002.
 546	 *
 547	 * If a profile is pointing to this instruction, please first
 548	 * consider moving your system to a more modern idle
 549	 * mechanism.
 550	 */
 551	inl(acpi_gbl_FADT.xpm_timer_block.address);
 552}
 553
 554/**
 555 * acpi_idle_do_entry - enter idle state using the appropriate method
 556 * @cx: cstate data
 557 *
 558 * Caller disables interrupt before call and enables interrupt after return.
 559 */
 560static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
 561{
 562	perf_lopwr_cb(true);
 563
 564	if (cx->entry_method == ACPI_CSTATE_FFH) {
 565		/* Call into architectural FFH based C-state */
 566		acpi_processor_ffh_cstate_enter(cx);
 567	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
 568		acpi_safe_halt();
 569	} else {
 570		io_idle(cx->address);
 571	}
 572
 573	perf_lopwr_cb(false);
 574}
 575
 576/**
 577 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
 578 * @dev: the target CPU
 579 * @index: the index of suggested state
 580 */
 581static void acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 582{
 583	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 584
 585	ACPI_FLUSH_CPU_CACHE();
 586
 587	while (1) {
 588
 589		if (cx->entry_method == ACPI_CSTATE_HALT)
 590			raw_safe_halt();
 591		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 592			io_idle(cx->address);
 593		} else
 594			return;
 595	}
 
 
 
 596}
 597
 598static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
 599{
 600	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
 601		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
 602}
 603
 604static int c3_cpu_count;
 605static DEFINE_RAW_SPINLOCK(c3_lock);
 606
 607/**
 608 * acpi_idle_enter_bm - enters C3 with proper BM handling
 609 * @drv: cpuidle driver
 610 * @pr: Target processor
 611 * @cx: Target state context
 612 * @index: index of target state
 613 */
 614static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
 615			       struct acpi_processor *pr,
 616			       struct acpi_processor_cx *cx,
 617			       int index)
 618{
 619	static struct acpi_processor_cx safe_cx = {
 620		.entry_method = ACPI_CSTATE_HALT,
 621	};
 622
 623	/*
 624	 * disable bus master
 625	 * bm_check implies we need ARB_DIS
 626	 * bm_control implies whether we can do ARB_DIS
 627	 *
 628	 * That leaves a case where bm_check is set and bm_control is not set.
 629	 * In that case we cannot do much, we enter C3 without doing anything.
 630	 */
 631	bool dis_bm = pr->flags.bm_control;
 632
 633	instrumentation_begin();
 634
 635	/* If we can skip BM, demote to a safe state. */
 636	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
 637		dis_bm = false;
 638		index = drv->safe_state_index;
 639		if (index >= 0) {
 640			cx = this_cpu_read(acpi_cstate[index]);
 641		} else {
 642			cx = &safe_cx;
 643			index = -EBUSY;
 644		}
 645	}
 646
 647	if (dis_bm) {
 648		raw_spin_lock(&c3_lock);
 649		c3_cpu_count++;
 650		/* Disable bus master arbitration when all CPUs are in C3 */
 651		if (c3_cpu_count == num_online_cpus())
 652			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 653		raw_spin_unlock(&c3_lock);
 654	}
 655
 656	ct_cpuidle_enter();
 657
 658	acpi_idle_do_entry(cx);
 659
 660	ct_cpuidle_exit();
 661
 662	/* Re-enable bus master arbitration */
 663	if (dis_bm) {
 664		raw_spin_lock(&c3_lock);
 665		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 666		c3_cpu_count--;
 667		raw_spin_unlock(&c3_lock);
 668	}
 669
 670	instrumentation_end();
 671
 672	return index;
 673}
 674
 675static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
 676			   struct cpuidle_driver *drv, int index)
 677{
 678	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 679	struct acpi_processor *pr;
 680
 681	pr = __this_cpu_read(processors);
 682	if (unlikely(!pr))
 683		return -EINVAL;
 684
 685	if (cx->type != ACPI_STATE_C1) {
 686		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
 687			return acpi_idle_enter_bm(drv, pr, cx, index);
 688
 689		/* C2 to C1 demotion. */
 690		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
 691			index = ACPI_IDLE_STATE_START;
 692			cx = per_cpu(acpi_cstate[index], dev->cpu);
 693		}
 694	}
 695
 696	if (cx->type == ACPI_STATE_C3)
 697		ACPI_FLUSH_CPU_CACHE();
 698
 699	acpi_idle_do_entry(cx);
 700
 701	return index;
 702}
 703
 704static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
 705				  struct cpuidle_driver *drv, int index)
 706{
 707	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 708
 709	if (cx->type == ACPI_STATE_C3) {
 710		struct acpi_processor *pr = __this_cpu_read(processors);
 711
 712		if (unlikely(!pr))
 713			return 0;
 714
 715		if (pr->flags.bm_check) {
 716			u8 bm_sts_skip = cx->bm_sts_skip;
 717
 718			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
 719			cx->bm_sts_skip = 1;
 720			acpi_idle_enter_bm(drv, pr, cx, index);
 721			cx->bm_sts_skip = bm_sts_skip;
 722
 723			return 0;
 724		} else {
 725			ACPI_FLUSH_CPU_CACHE();
 726		}
 727	}
 728	acpi_idle_do_entry(cx);
 729
 730	return 0;
 731}
 732
 733static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 734					   struct cpuidle_device *dev)
 735{
 736	int i, count = ACPI_IDLE_STATE_START;
 737	struct acpi_processor_cx *cx;
 738	struct cpuidle_state *state;
 739
 740	if (max_cstate == 0)
 741		max_cstate = 1;
 742
 743	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 744		state = &acpi_idle_driver.states[count];
 745		cx = &pr->power.states[i];
 746
 747		if (!cx->valid)
 748			continue;
 749
 750		per_cpu(acpi_cstate[count], dev->cpu) = cx;
 751
 752		if (lapic_timer_needs_broadcast(pr, cx))
 753			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
 754
 755		if (cx->type == ACPI_STATE_C3) {
 756			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
 757			if (pr->flags.bm_check)
 758				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
 759		}
 760
 761		count++;
 762		if (count == CPUIDLE_STATE_MAX)
 763			break;
 764	}
 765
 766	if (!count)
 767		return -EINVAL;
 768
 769	return 0;
 770}
 771
 772static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 773{
 774	int i, count;
 775	struct acpi_processor_cx *cx;
 776	struct cpuidle_state *state;
 777	struct cpuidle_driver *drv = &acpi_idle_driver;
 778
 779	if (max_cstate == 0)
 780		max_cstate = 1;
 781
 782	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
 783		cpuidle_poll_state_init(drv);
 784		count = 1;
 785	} else {
 786		count = 0;
 787	}
 788
 789	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 790		cx = &pr->power.states[i];
 791
 792		if (!cx->valid)
 793			continue;
 794
 795		state = &drv->states[count];
 796		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
 797		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
 798		state->exit_latency = cx->latency;
 799		state->target_residency = cx->latency * latency_factor;
 800		state->enter = acpi_idle_enter;
 801
 802		state->flags = 0;
 803
 804		state->enter_dead = acpi_idle_play_dead;
 805
 806		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2)
 807			drv->safe_state_index = count;
 808
 809		/*
 810		 * Halt-induced C1 is not good for ->enter_s2idle, because it
 811		 * re-enables interrupts on exit.  Moreover, C1 is generally not
 812		 * particularly interesting from the suspend-to-idle angle, so
 813		 * avoid C1 and the situations in which we may need to fall back
 814		 * to it altogether.
 815		 */
 816		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
 817			state->enter_s2idle = acpi_idle_enter_s2idle;
 818
 819		count++;
 820		if (count == CPUIDLE_STATE_MAX)
 821			break;
 822	}
 823
 824	drv->state_count = count;
 825
 826	if (!count)
 827		return -EINVAL;
 828
 829	return 0;
 830}
 831
 832static inline void acpi_processor_cstate_first_run_checks(void)
 833{
 834	static int first_run;
 835
 836	if (first_run)
 837		return;
 838	dmi_check_system(processor_power_dmi_table);
 839	max_cstate = acpi_processor_cstate_check(max_cstate);
 840	if (max_cstate < ACPI_C_STATES_MAX)
 841		pr_notice("processor limited to max C-state %d\n", max_cstate);
 842
 843	first_run++;
 844
 845	if (nocst)
 846		return;
 847
 848	acpi_processor_claim_cst_control();
 849}
 850#else
 851
 852static inline int disabled_by_idle_boot_param(void) { return 0; }
 853static inline void acpi_processor_cstate_first_run_checks(void) { }
 854static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 855{
 856	return -ENODEV;
 857}
 858
 859static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 860					   struct cpuidle_device *dev)
 861{
 862	return -EINVAL;
 863}
 864
 865static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 866{
 867	return -EINVAL;
 868}
 869
 870#endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
 871
 872struct acpi_lpi_states_array {
 873	unsigned int size;
 874	unsigned int composite_states_size;
 875	struct acpi_lpi_state *entries;
 876	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
 877};
 878
 879static int obj_get_integer(union acpi_object *obj, u32 *value)
 880{
 881	if (obj->type != ACPI_TYPE_INTEGER)
 882		return -EINVAL;
 883
 884	*value = obj->integer.value;
 885	return 0;
 886}
 887
 888static int acpi_processor_evaluate_lpi(acpi_handle handle,
 889				       struct acpi_lpi_states_array *info)
 890{
 891	acpi_status status;
 892	int ret = 0;
 893	int pkg_count, state_idx = 1, loop;
 894	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 895	union acpi_object *lpi_data;
 896	struct acpi_lpi_state *lpi_state;
 897
 898	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
 899	if (ACPI_FAILURE(status)) {
 900		acpi_handle_debug(handle, "No _LPI, giving up\n");
 901		return -ENODEV;
 902	}
 903
 904	lpi_data = buffer.pointer;
 905
 906	/* There must be at least 4 elements = 3 elements + 1 package */
 907	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
 908	    lpi_data->package.count < 4) {
 909		pr_debug("not enough elements in _LPI\n");
 910		ret = -ENODATA;
 911		goto end;
 912	}
 913
 914	pkg_count = lpi_data->package.elements[2].integer.value;
 915
 916	/* Validate number of power states. */
 917	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
 918		pr_debug("count given by _LPI is not valid\n");
 919		ret = -ENODATA;
 920		goto end;
 921	}
 922
 923	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
 924	if (!lpi_state) {
 925		ret = -ENOMEM;
 926		goto end;
 927	}
 928
 929	info->size = pkg_count;
 930	info->entries = lpi_state;
 931
 932	/* LPI States start at index 3 */
 933	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
 934		union acpi_object *element, *pkg_elem, *obj;
 935
 936		element = &lpi_data->package.elements[loop];
 937		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
 938			continue;
 939
 940		pkg_elem = element->package.elements;
 941
 942		obj = pkg_elem + 6;
 943		if (obj->type == ACPI_TYPE_BUFFER) {
 944			struct acpi_power_register *reg;
 945
 946			reg = (struct acpi_power_register *)obj->buffer.pointer;
 947			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 948			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
 949				continue;
 950
 951			lpi_state->address = reg->address;
 952			lpi_state->entry_method =
 953				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
 954				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
 955		} else if (obj->type == ACPI_TYPE_INTEGER) {
 956			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
 957			lpi_state->address = obj->integer.value;
 958		} else {
 959			continue;
 960		}
 961
 962		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
 963
 964		obj = pkg_elem + 9;
 965		if (obj->type == ACPI_TYPE_STRING)
 966			strscpy(lpi_state->desc, obj->string.pointer,
 967				ACPI_CX_DESC_LEN);
 968
 969		lpi_state->index = state_idx;
 970		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
 971			pr_debug("No min. residency found, assuming 10 us\n");
 972			lpi_state->min_residency = 10;
 973		}
 974
 975		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
 976			pr_debug("No wakeup residency found, assuming 10 us\n");
 977			lpi_state->wake_latency = 10;
 978		}
 979
 980		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
 981			lpi_state->flags = 0;
 982
 983		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
 984			lpi_state->arch_flags = 0;
 985
 986		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
 987			lpi_state->res_cnt_freq = 1;
 988
 989		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
 990			lpi_state->enable_parent_state = 0;
 991	}
 992
 993	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
 994end:
 995	kfree(buffer.pointer);
 996	return ret;
 997}
 998
 999/*
1000 * flat_state_cnt - the number of composite LPI states after the process of flattening
1001 */
1002static int flat_state_cnt;
1003
1004/**
1005 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1006 *
1007 * @local: local LPI state
1008 * @parent: parent LPI state
1009 * @result: composite LPI state
1010 */
1011static bool combine_lpi_states(struct acpi_lpi_state *local,
1012			       struct acpi_lpi_state *parent,
1013			       struct acpi_lpi_state *result)
1014{
1015	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1016		if (!parent->address) /* 0 means autopromotable */
1017			return false;
1018		result->address = local->address + parent->address;
1019	} else {
1020		result->address = parent->address;
1021	}
1022
1023	result->min_residency = max(local->min_residency, parent->min_residency);
1024	result->wake_latency = local->wake_latency + parent->wake_latency;
1025	result->enable_parent_state = parent->enable_parent_state;
1026	result->entry_method = local->entry_method;
1027
1028	result->flags = parent->flags;
1029	result->arch_flags = parent->arch_flags;
1030	result->index = parent->index;
1031
1032	strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1033	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1034	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1035	return true;
1036}
1037
1038#define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1039
1040static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1041				  struct acpi_lpi_state *t)
1042{
1043	curr_level->composite_states[curr_level->composite_states_size++] = t;
1044}
1045
1046static int flatten_lpi_states(struct acpi_processor *pr,
1047			      struct acpi_lpi_states_array *curr_level,
1048			      struct acpi_lpi_states_array *prev_level)
1049{
1050	int i, j, state_count = curr_level->size;
1051	struct acpi_lpi_state *p, *t = curr_level->entries;
1052
1053	curr_level->composite_states_size = 0;
1054	for (j = 0; j < state_count; j++, t++) {
1055		struct acpi_lpi_state *flpi;
1056
1057		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1058			continue;
1059
1060		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1061			pr_warn("Limiting number of LPI states to max (%d)\n",
1062				ACPI_PROCESSOR_MAX_POWER);
1063			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1064			break;
1065		}
1066
1067		flpi = &pr->power.lpi_states[flat_state_cnt];
1068
1069		if (!prev_level) { /* leaf/processor node */
1070			memcpy(flpi, t, sizeof(*t));
1071			stash_composite_state(curr_level, flpi);
1072			flat_state_cnt++;
1073			continue;
1074		}
1075
1076		for (i = 0; i < prev_level->composite_states_size; i++) {
1077			p = prev_level->composite_states[i];
1078			if (t->index <= p->enable_parent_state &&
1079			    combine_lpi_states(p, t, flpi)) {
1080				stash_composite_state(curr_level, flpi);
1081				flat_state_cnt++;
1082				flpi++;
1083			}
1084		}
1085	}
1086
1087	kfree(curr_level->entries);
1088	return 0;
1089}
1090
1091int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1092{
1093	return -EOPNOTSUPP;
1094}
1095
1096static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1097{
1098	int ret, i;
1099	acpi_status status;
1100	acpi_handle handle = pr->handle, pr_ahandle;
1101	struct acpi_device *d = NULL;
1102	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1103
1104	/* make sure our architecture has support */
1105	ret = acpi_processor_ffh_lpi_probe(pr->id);
1106	if (ret == -EOPNOTSUPP)
1107		return ret;
1108
1109	if (!osc_pc_lpi_support_confirmed)
1110		return -EOPNOTSUPP;
1111
1112	if (!acpi_has_method(handle, "_LPI"))
1113		return -EINVAL;
1114
1115	flat_state_cnt = 0;
1116	prev = &info[0];
1117	curr = &info[1];
1118	handle = pr->handle;
1119	ret = acpi_processor_evaluate_lpi(handle, prev);
1120	if (ret)
1121		return ret;
1122	flatten_lpi_states(pr, prev, NULL);
1123
1124	status = acpi_get_parent(handle, &pr_ahandle);
1125	while (ACPI_SUCCESS(status)) {
1126		d = acpi_fetch_acpi_dev(pr_ahandle);
1127		if (!d)
1128			break;
1129
1130		handle = pr_ahandle;
1131
1132		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1133			break;
1134
1135		/* can be optional ? */
1136		if (!acpi_has_method(handle, "_LPI"))
1137			break;
1138
1139		ret = acpi_processor_evaluate_lpi(handle, curr);
1140		if (ret)
1141			break;
1142
1143		/* flatten all the LPI states in this level of hierarchy */
1144		flatten_lpi_states(pr, curr, prev);
1145
1146		tmp = prev, prev = curr, curr = tmp;
1147
1148		status = acpi_get_parent(handle, &pr_ahandle);
1149	}
1150
1151	pr->power.count = flat_state_cnt;
1152	/* reset the index after flattening */
1153	for (i = 0; i < pr->power.count; i++)
1154		pr->power.lpi_states[i].index = i;
1155
1156	/* Tell driver that _LPI is supported. */
1157	pr->flags.has_lpi = 1;
1158	pr->flags.power = 1;
1159
1160	return 0;
1161}
1162
1163int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1164{
1165	return -ENODEV;
1166}
1167
1168/**
1169 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1170 * @dev: the target CPU
1171 * @drv: cpuidle driver containing cpuidle state info
1172 * @index: index of target state
1173 *
1174 * Return: 0 for success or negative value for error
1175 */
1176static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1177			       struct cpuidle_driver *drv, int index)
1178{
1179	struct acpi_processor *pr;
1180	struct acpi_lpi_state *lpi;
1181
1182	pr = __this_cpu_read(processors);
1183
1184	if (unlikely(!pr))
1185		return -EINVAL;
1186
1187	lpi = &pr->power.lpi_states[index];
1188	if (lpi->entry_method == ACPI_CSTATE_FFH)
1189		return acpi_processor_ffh_lpi_enter(lpi);
1190
1191	return -EINVAL;
1192}
1193
1194static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1195{
1196	int i;
1197	struct acpi_lpi_state *lpi;
1198	struct cpuidle_state *state;
1199	struct cpuidle_driver *drv = &acpi_idle_driver;
1200
1201	if (!pr->flags.has_lpi)
1202		return -EOPNOTSUPP;
1203
1204	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1205		lpi = &pr->power.lpi_states[i];
1206
1207		state = &drv->states[i];
1208		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1209		strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1210		state->exit_latency = lpi->wake_latency;
1211		state->target_residency = lpi->min_residency;
1212		state->flags |= arch_get_idle_state_flags(lpi->arch_flags);
1213		if (i != 0 && lpi->entry_method == ACPI_CSTATE_FFH)
1214			state->flags |= CPUIDLE_FLAG_RCU_IDLE;
1215		state->enter = acpi_idle_lpi_enter;
1216		drv->safe_state_index = i;
1217	}
1218
1219	drv->state_count = i;
1220
1221	return 0;
1222}
1223
1224/**
1225 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1226 * global state data i.e. idle routines
1227 *
1228 * @pr: the ACPI processor
1229 */
1230static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1231{
1232	int i;
1233	struct cpuidle_driver *drv = &acpi_idle_driver;
1234
1235	if (!pr->flags.power_setup_done || !pr->flags.power)
1236		return -EINVAL;
1237
1238	drv->safe_state_index = -1;
1239	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1240		drv->states[i].name[0] = '\0';
1241		drv->states[i].desc[0] = '\0';
1242	}
1243
1244	if (pr->flags.has_lpi)
1245		return acpi_processor_setup_lpi_states(pr);
1246
1247	return acpi_processor_setup_cstates(pr);
1248}
1249
1250/**
1251 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1252 * device i.e. per-cpu data
1253 *
1254 * @pr: the ACPI processor
1255 * @dev : the cpuidle device
1256 */
1257static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1258					    struct cpuidle_device *dev)
1259{
1260	if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1261		return -EINVAL;
1262
1263	dev->cpu = pr->id;
1264	if (pr->flags.has_lpi)
1265		return acpi_processor_ffh_lpi_probe(pr->id);
1266
1267	return acpi_processor_setup_cpuidle_cx(pr, dev);
1268}
1269
1270static int acpi_processor_get_power_info(struct acpi_processor *pr)
1271{
1272	int ret;
1273
1274	ret = acpi_processor_get_lpi_info(pr);
1275	if (ret)
1276		ret = acpi_processor_get_cstate_info(pr);
1277
1278	return ret;
1279}
1280
1281int acpi_processor_hotplug(struct acpi_processor *pr)
1282{
1283	int ret = 0;
1284	struct cpuidle_device *dev;
1285
1286	if (disabled_by_idle_boot_param())
1287		return 0;
1288
1289	if (!pr->flags.power_setup_done)
1290		return -ENODEV;
1291
1292	dev = per_cpu(acpi_cpuidle_device, pr->id);
1293	cpuidle_pause_and_lock();
1294	cpuidle_disable_device(dev);
1295	ret = acpi_processor_get_power_info(pr);
1296	if (!ret && pr->flags.power) {
1297		acpi_processor_setup_cpuidle_dev(pr, dev);
1298		ret = cpuidle_enable_device(dev);
1299	}
1300	cpuidle_resume_and_unlock();
1301
1302	return ret;
1303}
1304
1305int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1306{
1307	int cpu;
1308	struct acpi_processor *_pr;
1309	struct cpuidle_device *dev;
1310
1311	if (disabled_by_idle_boot_param())
1312		return 0;
1313
1314	if (!pr->flags.power_setup_done)
1315		return -ENODEV;
1316
1317	/*
1318	 * FIXME:  Design the ACPI notification to make it once per
1319	 * system instead of once per-cpu.  This condition is a hack
1320	 * to make the code that updates C-States be called once.
1321	 */
1322
1323	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1324
1325		/* Protect against cpu-hotplug */
1326		cpus_read_lock();
1327		cpuidle_pause_and_lock();
1328
1329		/* Disable all cpuidle devices */
1330		for_each_online_cpu(cpu) {
1331			_pr = per_cpu(processors, cpu);
1332			if (!_pr || !_pr->flags.power_setup_done)
1333				continue;
1334			dev = per_cpu(acpi_cpuidle_device, cpu);
1335			cpuidle_disable_device(dev);
1336		}
1337
1338		/* Populate Updated C-state information */
1339		acpi_processor_get_power_info(pr);
1340		acpi_processor_setup_cpuidle_states(pr);
1341
1342		/* Enable all cpuidle devices */
1343		for_each_online_cpu(cpu) {
1344			_pr = per_cpu(processors, cpu);
1345			if (!_pr || !_pr->flags.power_setup_done)
1346				continue;
1347			acpi_processor_get_power_info(_pr);
1348			if (_pr->flags.power) {
1349				dev = per_cpu(acpi_cpuidle_device, cpu);
1350				acpi_processor_setup_cpuidle_dev(_pr, dev);
1351				cpuidle_enable_device(dev);
1352			}
1353		}
1354		cpuidle_resume_and_unlock();
1355		cpus_read_unlock();
1356	}
1357
1358	return 0;
1359}
1360
1361static int acpi_processor_registered;
1362
1363int acpi_processor_power_init(struct acpi_processor *pr)
1364{
1365	int retval;
1366	struct cpuidle_device *dev;
1367
1368	if (disabled_by_idle_boot_param())
1369		return 0;
1370
1371	acpi_processor_cstate_first_run_checks();
1372
1373	if (!acpi_processor_get_power_info(pr))
1374		pr->flags.power_setup_done = 1;
1375
1376	/*
1377	 * Install the idle handler if processor power management is supported.
1378	 * Note that we use previously set idle handler will be used on
1379	 * platforms that only support C1.
1380	 */
1381	if (pr->flags.power) {
1382		/* Register acpi_idle_driver if not already registered */
1383		if (!acpi_processor_registered) {
1384			acpi_processor_setup_cpuidle_states(pr);
1385			retval = cpuidle_register_driver(&acpi_idle_driver);
1386			if (retval)
1387				return retval;
1388			pr_debug("%s registered with cpuidle\n",
1389				 acpi_idle_driver.name);
1390		}
1391
1392		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1393		if (!dev)
1394			return -ENOMEM;
1395		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1396
1397		acpi_processor_setup_cpuidle_dev(pr, dev);
1398
1399		/* Register per-cpu cpuidle_device. Cpuidle driver
1400		 * must already be registered before registering device
1401		 */
1402		retval = cpuidle_register_device(dev);
1403		if (retval) {
1404			if (acpi_processor_registered == 0)
1405				cpuidle_unregister_driver(&acpi_idle_driver);
1406			return retval;
1407		}
1408		acpi_processor_registered++;
1409	}
1410	return 0;
1411}
1412
1413int acpi_processor_power_exit(struct acpi_processor *pr)
1414{
1415	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1416
1417	if (disabled_by_idle_boot_param())
1418		return 0;
1419
1420	if (pr->flags.power) {
1421		cpuidle_unregister_device(dev);
1422		acpi_processor_registered--;
1423		if (acpi_processor_registered == 0)
1424			cpuidle_unregister_driver(&acpi_idle_driver);
1425
1426		kfree(dev);
1427	}
1428
1429	pr->flags.power_setup_done = 0;
1430	return 0;
1431}
v6.9.4
   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/sort.h>
  20#include <linux/tick.h>
  21#include <linux/cpuidle.h>
  22#include <linux/cpu.h>
  23#include <linux/minmax.h>
  24#include <linux/perf_event.h>
  25#include <acpi/processor.h>
  26#include <linux/context_tracking.h>
  27
  28/*
  29 * Include the apic definitions for x86 to have the APIC timer related defines
  30 * available also for UP (on SMP it gets magically included via linux/smp.h).
  31 * asm/acpi.h is not an option, as it would require more include magic. Also
  32 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
  33 */
  34#ifdef CONFIG_X86
  35#include <asm/apic.h>
  36#include <asm/cpu.h>
  37#endif
  38
  39#define ACPI_IDLE_STATE_START	(IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
  40
  41static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
  42module_param(max_cstate, uint, 0400);
  43static bool nocst __read_mostly;
  44module_param(nocst, bool, 0400);
  45static bool bm_check_disable __read_mostly;
  46module_param(bm_check_disable, bool, 0400);
  47
  48static unsigned int latency_factor __read_mostly = 2;
  49module_param(latency_factor, uint, 0644);
  50
  51static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
  52
  53struct cpuidle_driver acpi_idle_driver = {
  54	.name =		"acpi_idle",
  55	.owner =	THIS_MODULE,
  56};
  57
  58#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
  59static
  60DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
  61
  62static int disabled_by_idle_boot_param(void)
  63{
  64	return boot_option_idle_override == IDLE_POLL ||
  65		boot_option_idle_override == IDLE_HALT;
  66}
  67
  68/*
  69 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
  70 * For now disable this. Probably a bug somewhere else.
  71 *
  72 * To skip this limit, boot/load with a large max_cstate limit.
  73 */
  74static int set_max_cstate(const struct dmi_system_id *id)
  75{
  76	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
  77		return 0;
  78
  79	pr_notice("%s detected - limiting to C%ld max_cstate."
  80		  " Override with \"processor.max_cstate=%d\"\n", id->ident,
  81		  (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
  82
  83	max_cstate = (long)id->driver_data;
  84
  85	return 0;
  86}
  87
  88static const struct dmi_system_id processor_power_dmi_table[] = {
  89	{ set_max_cstate, "Clevo 5600D", {
  90	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
  91	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
  92	 (void *)2},
  93	{ set_max_cstate, "Pavilion zv5000", {
  94	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
  95	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
  96	 (void *)1},
  97	{ set_max_cstate, "Asus L8400B", {
  98	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
  99	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
 100	 (void *)1},
 101	{},
 102};
 103
 104
 105/*
 106 * Callers should disable interrupts before the call and enable
 107 * interrupts after return.
 108 */
 109static void __cpuidle acpi_safe_halt(void)
 110{
 111	if (!tif_need_resched()) {
 112		raw_safe_halt();
 113		raw_local_irq_disable();
 114	}
 115}
 116
 117#ifdef ARCH_APICTIMER_STOPS_ON_C3
 118
 119/*
 120 * Some BIOS implementations switch to C3 in the published C2 state.
 121 * This seems to be a common problem on AMD boxen, but other vendors
 122 * are affected too. We pick the most conservative approach: we assume
 123 * that the local APIC stops in both C2 and C3.
 124 */
 125static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 126				   struct acpi_processor_cx *cx)
 127{
 128	struct acpi_processor_power *pwr = &pr->power;
 129	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
 130
 131	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
 132		return;
 133
 134	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
 135		type = ACPI_STATE_C1;
 136
 137	/*
 138	 * Check, if one of the previous states already marked the lapic
 139	 * unstable
 140	 */
 141	if (pwr->timer_broadcast_on_state < state)
 142		return;
 143
 144	if (cx->type >= type)
 145		pr->power.timer_broadcast_on_state = state;
 146}
 147
 148static void __lapic_timer_propagate_broadcast(void *arg)
 149{
 150	struct acpi_processor *pr = arg;
 151
 152	if (pr->power.timer_broadcast_on_state < INT_MAX)
 153		tick_broadcast_enable();
 154	else
 155		tick_broadcast_disable();
 156}
 157
 158static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
 159{
 160	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
 161				 (void *)pr, 1);
 162}
 163
 164/* Power(C) State timer broadcast control */
 165static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 166					struct acpi_processor_cx *cx)
 167{
 168	return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
 169}
 170
 171#else
 172
 173static void lapic_timer_check_state(int state, struct acpi_processor *pr,
 174				   struct acpi_processor_cx *cstate) { }
 175static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
 176
 177static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
 178					struct acpi_processor_cx *cx)
 179{
 180	return false;
 181}
 182
 183#endif
 184
 185#if defined(CONFIG_X86)
 186static void tsc_check_state(int state)
 187{
 188	switch (boot_cpu_data.x86_vendor) {
 189	case X86_VENDOR_HYGON:
 190	case X86_VENDOR_AMD:
 191	case X86_VENDOR_INTEL:
 192	case X86_VENDOR_CENTAUR:
 193	case X86_VENDOR_ZHAOXIN:
 194		/*
 195		 * AMD Fam10h TSC will tick in all
 196		 * C/P/S0/S1 states when this bit is set.
 197		 */
 198		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 199			return;
 200		fallthrough;
 201	default:
 202		/* TSC could halt in idle, so notify users */
 203		if (state > ACPI_STATE_C1)
 204			mark_tsc_unstable("TSC halts in idle");
 205	}
 206}
 207#else
 208static void tsc_check_state(int state) { return; }
 209#endif
 210
 211static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
 212{
 213
 214	if (!pr->pblk)
 215		return -ENODEV;
 216
 217	/* if info is obtained from pblk/fadt, type equals state */
 218	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
 219	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
 220
 221#ifndef CONFIG_HOTPLUG_CPU
 222	/*
 223	 * Check for P_LVL2_UP flag before entering C2 and above on
 224	 * an SMP system.
 225	 */
 226	if ((num_online_cpus() > 1) &&
 227	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
 228		return -ENODEV;
 229#endif
 230
 231	/* determine C2 and C3 address from pblk */
 232	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
 233	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
 234
 235	/* determine latencies from FADT */
 236	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
 237	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
 238
 239	/*
 240	 * FADT specified C2 latency must be less than or equal to
 241	 * 100 microseconds.
 242	 */
 243	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
 244		acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n",
 245				  acpi_gbl_FADT.c2_latency);
 246		/* invalidate C2 */
 247		pr->power.states[ACPI_STATE_C2].address = 0;
 248	}
 249
 250	/*
 251	 * FADT supplied C3 latency must be less than or equal to
 252	 * 1000 microseconds.
 253	 */
 254	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
 255		acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n",
 256				  acpi_gbl_FADT.c3_latency);
 257		/* invalidate C3 */
 258		pr->power.states[ACPI_STATE_C3].address = 0;
 259	}
 260
 261	acpi_handle_debug(pr->handle, "lvl2[0x%08x] lvl3[0x%08x]\n",
 262			  pr->power.states[ACPI_STATE_C2].address,
 263			  pr->power.states[ACPI_STATE_C3].address);
 264
 265	snprintf(pr->power.states[ACPI_STATE_C2].desc,
 266			 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
 267			 pr->power.states[ACPI_STATE_C2].address);
 268	snprintf(pr->power.states[ACPI_STATE_C3].desc,
 269			 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
 270			 pr->power.states[ACPI_STATE_C3].address);
 271
 272	return 0;
 273}
 274
 275static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
 276{
 277	if (!pr->power.states[ACPI_STATE_C1].valid) {
 278		/* set the first C-State to C1 */
 279		/* all processors need to support C1 */
 280		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
 281		pr->power.states[ACPI_STATE_C1].valid = 1;
 282		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
 283
 284		snprintf(pr->power.states[ACPI_STATE_C1].desc,
 285			 ACPI_CX_DESC_LEN, "ACPI HLT");
 286	}
 287	/* the C0 state only exists as a filler in our array */
 288	pr->power.states[ACPI_STATE_C0].valid = 1;
 289	return 0;
 290}
 291
 292static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
 293{
 294	int ret;
 295
 296	if (nocst)
 297		return -ENODEV;
 298
 299	ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
 300	if (ret)
 301		return ret;
 302
 303	if (!pr->power.count)
 304		return -EFAULT;
 305
 306	pr->flags.has_cst = 1;
 307	return 0;
 308}
 309
 310static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
 311					   struct acpi_processor_cx *cx)
 312{
 313	static int bm_check_flag = -1;
 314	static int bm_control_flag = -1;
 315
 316
 317	if (!cx->address)
 318		return;
 319
 320	/*
 321	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
 322	 * DMA transfers are used by any ISA device to avoid livelock.
 323	 * Note that we could disable Type-F DMA (as recommended by
 324	 * the erratum), but this is known to disrupt certain ISA
 325	 * devices thus we take the conservative approach.
 326	 */
 327	if (errata.piix4.fdma) {
 328		acpi_handle_debug(pr->handle,
 329				  "C3 not supported on PIIX4 with Type-F DMA\n");
 330		return;
 331	}
 332
 333	/* All the logic here assumes flags.bm_check is same across all CPUs */
 334	if (bm_check_flag == -1) {
 335		/* Determine whether bm_check is needed based on CPU  */
 336		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
 337		bm_check_flag = pr->flags.bm_check;
 338		bm_control_flag = pr->flags.bm_control;
 339	} else {
 340		pr->flags.bm_check = bm_check_flag;
 341		pr->flags.bm_control = bm_control_flag;
 342	}
 343
 344	if (pr->flags.bm_check) {
 345		if (!pr->flags.bm_control) {
 346			if (pr->flags.has_cst != 1) {
 347				/* bus mastering control is necessary */
 348				acpi_handle_debug(pr->handle,
 349						  "C3 support requires BM control\n");
 350				return;
 351			} else {
 352				/* Here we enter C3 without bus mastering */
 353				acpi_handle_debug(pr->handle,
 354						  "C3 support without BM control\n");
 355			}
 356		}
 357	} else {
 358		/*
 359		 * WBINVD should be set in fadt, for C3 state to be
 360		 * supported on when bm_check is not required.
 361		 */
 362		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
 363			acpi_handle_debug(pr->handle,
 364					  "Cache invalidation should work properly"
 365					  " for C3 to be enabled on SMP systems\n");
 366			return;
 367		}
 368	}
 369
 370	/*
 371	 * Otherwise we've met all of our C3 requirements.
 372	 * Normalize the C3 latency to expidite policy.  Enable
 373	 * checking of bus mastering status (bm_check) so we can
 374	 * use this in our C3 policy
 375	 */
 376	cx->valid = 1;
 377
 378	/*
 379	 * On older chipsets, BM_RLD needs to be set
 380	 * in order for Bus Master activity to wake the
 381	 * system from C3.  Newer chipsets handle DMA
 382	 * during C3 automatically and BM_RLD is a NOP.
 383	 * In either case, the proper way to
 384	 * handle BM_RLD is to set it and leave it set.
 385	 */
 386	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
 387}
 388
 389static int acpi_cst_latency_cmp(const void *a, const void *b)
 390{
 391	const struct acpi_processor_cx *x = a, *y = b;
 392
 393	if (!(x->valid && y->valid))
 394		return 0;
 395	if (x->latency > y->latency)
 396		return 1;
 397	if (x->latency < y->latency)
 398		return -1;
 399	return 0;
 400}
 401static void acpi_cst_latency_swap(void *a, void *b, int n)
 402{
 403	struct acpi_processor_cx *x = a, *y = b;
 404
 405	if (!(x->valid && y->valid))
 406		return;
 407	swap(x->latency, y->latency);
 408}
 409
 410static int acpi_processor_power_verify(struct acpi_processor *pr)
 411{
 412	unsigned int i;
 413	unsigned int working = 0;
 414	unsigned int last_latency = 0;
 415	unsigned int last_type = 0;
 416	bool buggy_latency = false;
 417
 418	pr->power.timer_broadcast_on_state = INT_MAX;
 419
 420	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 421		struct acpi_processor_cx *cx = &pr->power.states[i];
 422
 423		switch (cx->type) {
 424		case ACPI_STATE_C1:
 425			cx->valid = 1;
 426			break;
 427
 428		case ACPI_STATE_C2:
 429			if (!cx->address)
 430				break;
 431			cx->valid = 1;
 432			break;
 433
 434		case ACPI_STATE_C3:
 435			acpi_processor_power_verify_c3(pr, cx);
 436			break;
 437		}
 438		if (!cx->valid)
 439			continue;
 440		if (cx->type >= last_type && cx->latency < last_latency)
 441			buggy_latency = true;
 442		last_latency = cx->latency;
 443		last_type = cx->type;
 444
 445		lapic_timer_check_state(i, pr, cx);
 446		tsc_check_state(cx->type);
 447		working++;
 448	}
 449
 450	if (buggy_latency) {
 451		pr_notice("FW issue: working around C-state latencies out of order\n");
 452		sort(&pr->power.states[1], max_cstate,
 453		     sizeof(struct acpi_processor_cx),
 454		     acpi_cst_latency_cmp,
 455		     acpi_cst_latency_swap);
 456	}
 457
 458	lapic_timer_propagate_broadcast(pr);
 459
 460	return working;
 461}
 462
 463static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 464{
 465	unsigned int i;
 466	int result;
 467
 468
 469	/* NOTE: the idle thread may not be running while calling
 470	 * this function */
 471
 472	/* Zero initialize all the C-states info. */
 473	memset(pr->power.states, 0, sizeof(pr->power.states));
 474
 475	result = acpi_processor_get_power_info_cst(pr);
 476	if (result == -ENODEV)
 477		result = acpi_processor_get_power_info_fadt(pr);
 478
 479	if (result)
 480		return result;
 481
 482	acpi_processor_get_power_info_default(pr);
 483
 484	pr->power.count = acpi_processor_power_verify(pr);
 485
 486	/*
 487	 * if one state of type C2 or C3 is available, mark this
 488	 * CPU as being "idle manageable"
 489	 */
 490	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
 491		if (pr->power.states[i].valid) {
 492			pr->power.count = i;
 493			pr->flags.power = 1;
 494		}
 495	}
 496
 497	return 0;
 498}
 499
 500/**
 501 * acpi_idle_bm_check - checks if bus master activity was detected
 502 */
 503static int acpi_idle_bm_check(void)
 504{
 505	u32 bm_status = 0;
 506
 507	if (bm_check_disable)
 508		return 0;
 509
 510	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
 511	if (bm_status)
 512		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
 513	/*
 514	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
 515	 * the true state of bus mastering activity; forcing us to
 516	 * manually check the BMIDEA bit of each IDE channel.
 517	 */
 518	else if (errata.piix4.bmisx) {
 519		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
 520		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
 521			bm_status = 1;
 522	}
 523	return bm_status;
 524}
 525
 526static __cpuidle void io_idle(unsigned long addr)
 527{
 528	/* IO port based C-state */
 529	inb(addr);
 530
 531#ifdef	CONFIG_X86
 532	/* No delay is needed if we are in guest */
 533	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 534		return;
 535	/*
 536	 * Modern (>=Nehalem) Intel systems use ACPI via intel_idle,
 537	 * not this code.  Assume that any Intel systems using this
 538	 * are ancient and may need the dummy wait.  This also assumes
 539	 * that the motivating chipset issue was Intel-only.
 540	 */
 541	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
 542		return;
 543#endif
 544	/*
 545	 * Dummy wait op - must do something useless after P_LVL2 read
 546	 * because chipsets cannot guarantee that STPCLK# signal gets
 547	 * asserted in time to freeze execution properly
 548	 *
 549	 * This workaround has been in place since the original ACPI
 550	 * implementation was merged, circa 2002.
 551	 *
 552	 * If a profile is pointing to this instruction, please first
 553	 * consider moving your system to a more modern idle
 554	 * mechanism.
 555	 */
 556	inl(acpi_gbl_FADT.xpm_timer_block.address);
 557}
 558
 559/**
 560 * acpi_idle_do_entry - enter idle state using the appropriate method
 561 * @cx: cstate data
 562 *
 563 * Caller disables interrupt before call and enables interrupt after return.
 564 */
 565static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
 566{
 567	perf_lopwr_cb(true);
 568
 569	if (cx->entry_method == ACPI_CSTATE_FFH) {
 570		/* Call into architectural FFH based C-state */
 571		acpi_processor_ffh_cstate_enter(cx);
 572	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
 573		acpi_safe_halt();
 574	} else {
 575		io_idle(cx->address);
 576	}
 577
 578	perf_lopwr_cb(false);
 579}
 580
 581/**
 582 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
 583 * @dev: the target CPU
 584 * @index: the index of suggested state
 585 */
 586static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 587{
 588	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 589
 590	ACPI_FLUSH_CPU_CACHE();
 591
 592	while (1) {
 593
 594		if (cx->entry_method == ACPI_CSTATE_HALT)
 595			raw_safe_halt();
 596		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
 597			io_idle(cx->address);
 598		} else
 599			return -ENODEV;
 600	}
 601
 602	/* Never reached */
 603	return 0;
 604}
 605
 606static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
 607{
 608	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
 609		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
 610}
 611
 612static int c3_cpu_count;
 613static DEFINE_RAW_SPINLOCK(c3_lock);
 614
 615/**
 616 * acpi_idle_enter_bm - enters C3 with proper BM handling
 617 * @drv: cpuidle driver
 618 * @pr: Target processor
 619 * @cx: Target state context
 620 * @index: index of target state
 621 */
 622static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
 623			       struct acpi_processor *pr,
 624			       struct acpi_processor_cx *cx,
 625			       int index)
 626{
 627	static struct acpi_processor_cx safe_cx = {
 628		.entry_method = ACPI_CSTATE_HALT,
 629	};
 630
 631	/*
 632	 * disable bus master
 633	 * bm_check implies we need ARB_DIS
 634	 * bm_control implies whether we can do ARB_DIS
 635	 *
 636	 * That leaves a case where bm_check is set and bm_control is not set.
 637	 * In that case we cannot do much, we enter C3 without doing anything.
 638	 */
 639	bool dis_bm = pr->flags.bm_control;
 640
 641	instrumentation_begin();
 642
 643	/* If we can skip BM, demote to a safe state. */
 644	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
 645		dis_bm = false;
 646		index = drv->safe_state_index;
 647		if (index >= 0) {
 648			cx = this_cpu_read(acpi_cstate[index]);
 649		} else {
 650			cx = &safe_cx;
 651			index = -EBUSY;
 652		}
 653	}
 654
 655	if (dis_bm) {
 656		raw_spin_lock(&c3_lock);
 657		c3_cpu_count++;
 658		/* Disable bus master arbitration when all CPUs are in C3 */
 659		if (c3_cpu_count == num_online_cpus())
 660			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
 661		raw_spin_unlock(&c3_lock);
 662	}
 663
 664	ct_cpuidle_enter();
 665
 666	acpi_idle_do_entry(cx);
 667
 668	ct_cpuidle_exit();
 669
 670	/* Re-enable bus master arbitration */
 671	if (dis_bm) {
 672		raw_spin_lock(&c3_lock);
 673		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
 674		c3_cpu_count--;
 675		raw_spin_unlock(&c3_lock);
 676	}
 677
 678	instrumentation_end();
 679
 680	return index;
 681}
 682
 683static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
 684			   struct cpuidle_driver *drv, int index)
 685{
 686	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 687	struct acpi_processor *pr;
 688
 689	pr = __this_cpu_read(processors);
 690	if (unlikely(!pr))
 691		return -EINVAL;
 692
 693	if (cx->type != ACPI_STATE_C1) {
 694		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
 695			return acpi_idle_enter_bm(drv, pr, cx, index);
 696
 697		/* C2 to C1 demotion. */
 698		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
 699			index = ACPI_IDLE_STATE_START;
 700			cx = per_cpu(acpi_cstate[index], dev->cpu);
 701		}
 702	}
 703
 704	if (cx->type == ACPI_STATE_C3)
 705		ACPI_FLUSH_CPU_CACHE();
 706
 707	acpi_idle_do_entry(cx);
 708
 709	return index;
 710}
 711
 712static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
 713				  struct cpuidle_driver *drv, int index)
 714{
 715	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
 716
 717	if (cx->type == ACPI_STATE_C3) {
 718		struct acpi_processor *pr = __this_cpu_read(processors);
 719
 720		if (unlikely(!pr))
 721			return 0;
 722
 723		if (pr->flags.bm_check) {
 724			u8 bm_sts_skip = cx->bm_sts_skip;
 725
 726			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
 727			cx->bm_sts_skip = 1;
 728			acpi_idle_enter_bm(drv, pr, cx, index);
 729			cx->bm_sts_skip = bm_sts_skip;
 730
 731			return 0;
 732		} else {
 733			ACPI_FLUSH_CPU_CACHE();
 734		}
 735	}
 736	acpi_idle_do_entry(cx);
 737
 738	return 0;
 739}
 740
 741static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 742					   struct cpuidle_device *dev)
 743{
 744	int i, count = ACPI_IDLE_STATE_START;
 745	struct acpi_processor_cx *cx;
 746	struct cpuidle_state *state;
 747
 748	if (max_cstate == 0)
 749		max_cstate = 1;
 750
 751	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 752		state = &acpi_idle_driver.states[count];
 753		cx = &pr->power.states[i];
 754
 755		if (!cx->valid)
 756			continue;
 757
 758		per_cpu(acpi_cstate[count], dev->cpu) = cx;
 759
 760		if (lapic_timer_needs_broadcast(pr, cx))
 761			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
 762
 763		if (cx->type == ACPI_STATE_C3) {
 764			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
 765			if (pr->flags.bm_check)
 766				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
 767		}
 768
 769		count++;
 770		if (count == CPUIDLE_STATE_MAX)
 771			break;
 772	}
 773
 774	if (!count)
 775		return -EINVAL;
 776
 777	return 0;
 778}
 779
 780static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 781{
 782	int i, count;
 783	struct acpi_processor_cx *cx;
 784	struct cpuidle_state *state;
 785	struct cpuidle_driver *drv = &acpi_idle_driver;
 786
 787	if (max_cstate == 0)
 788		max_cstate = 1;
 789
 790	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
 791		cpuidle_poll_state_init(drv);
 792		count = 1;
 793	} else {
 794		count = 0;
 795	}
 796
 797	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
 798		cx = &pr->power.states[i];
 799
 800		if (!cx->valid)
 801			continue;
 802
 803		state = &drv->states[count];
 804		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
 805		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
 806		state->exit_latency = cx->latency;
 807		state->target_residency = cx->latency * latency_factor;
 808		state->enter = acpi_idle_enter;
 809
 810		state->flags = 0;
 811		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
 812		    cx->type == ACPI_STATE_C3) {
 813			state->enter_dead = acpi_idle_play_dead;
 814			if (cx->type != ACPI_STATE_C3)
 815				drv->safe_state_index = count;
 816		}
 817		/*
 818		 * Halt-induced C1 is not good for ->enter_s2idle, because it
 819		 * re-enables interrupts on exit.  Moreover, C1 is generally not
 820		 * particularly interesting from the suspend-to-idle angle, so
 821		 * avoid C1 and the situations in which we may need to fall back
 822		 * to it altogether.
 823		 */
 824		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
 825			state->enter_s2idle = acpi_idle_enter_s2idle;
 826
 827		count++;
 828		if (count == CPUIDLE_STATE_MAX)
 829			break;
 830	}
 831
 832	drv->state_count = count;
 833
 834	if (!count)
 835		return -EINVAL;
 836
 837	return 0;
 838}
 839
 840static inline void acpi_processor_cstate_first_run_checks(void)
 841{
 842	static int first_run;
 843
 844	if (first_run)
 845		return;
 846	dmi_check_system(processor_power_dmi_table);
 847	max_cstate = acpi_processor_cstate_check(max_cstate);
 848	if (max_cstate < ACPI_C_STATES_MAX)
 849		pr_notice("processor limited to max C-state %d\n", max_cstate);
 850
 851	first_run++;
 852
 853	if (nocst)
 854		return;
 855
 856	acpi_processor_claim_cst_control();
 857}
 858#else
 859
 860static inline int disabled_by_idle_boot_param(void) { return 0; }
 861static inline void acpi_processor_cstate_first_run_checks(void) { }
 862static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
 863{
 864	return -ENODEV;
 865}
 866
 867static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
 868					   struct cpuidle_device *dev)
 869{
 870	return -EINVAL;
 871}
 872
 873static int acpi_processor_setup_cstates(struct acpi_processor *pr)
 874{
 875	return -EINVAL;
 876}
 877
 878#endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
 879
 880struct acpi_lpi_states_array {
 881	unsigned int size;
 882	unsigned int composite_states_size;
 883	struct acpi_lpi_state *entries;
 884	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
 885};
 886
 887static int obj_get_integer(union acpi_object *obj, u32 *value)
 888{
 889	if (obj->type != ACPI_TYPE_INTEGER)
 890		return -EINVAL;
 891
 892	*value = obj->integer.value;
 893	return 0;
 894}
 895
 896static int acpi_processor_evaluate_lpi(acpi_handle handle,
 897				       struct acpi_lpi_states_array *info)
 898{
 899	acpi_status status;
 900	int ret = 0;
 901	int pkg_count, state_idx = 1, loop;
 902	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
 903	union acpi_object *lpi_data;
 904	struct acpi_lpi_state *lpi_state;
 905
 906	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
 907	if (ACPI_FAILURE(status)) {
 908		acpi_handle_debug(handle, "No _LPI, giving up\n");
 909		return -ENODEV;
 910	}
 911
 912	lpi_data = buffer.pointer;
 913
 914	/* There must be at least 4 elements = 3 elements + 1 package */
 915	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
 916	    lpi_data->package.count < 4) {
 917		pr_debug("not enough elements in _LPI\n");
 918		ret = -ENODATA;
 919		goto end;
 920	}
 921
 922	pkg_count = lpi_data->package.elements[2].integer.value;
 923
 924	/* Validate number of power states. */
 925	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
 926		pr_debug("count given by _LPI is not valid\n");
 927		ret = -ENODATA;
 928		goto end;
 929	}
 930
 931	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
 932	if (!lpi_state) {
 933		ret = -ENOMEM;
 934		goto end;
 935	}
 936
 937	info->size = pkg_count;
 938	info->entries = lpi_state;
 939
 940	/* LPI States start at index 3 */
 941	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
 942		union acpi_object *element, *pkg_elem, *obj;
 943
 944		element = &lpi_data->package.elements[loop];
 945		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
 946			continue;
 947
 948		pkg_elem = element->package.elements;
 949
 950		obj = pkg_elem + 6;
 951		if (obj->type == ACPI_TYPE_BUFFER) {
 952			struct acpi_power_register *reg;
 953
 954			reg = (struct acpi_power_register *)obj->buffer.pointer;
 955			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
 956			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
 957				continue;
 958
 959			lpi_state->address = reg->address;
 960			lpi_state->entry_method =
 961				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
 962				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
 963		} else if (obj->type == ACPI_TYPE_INTEGER) {
 964			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
 965			lpi_state->address = obj->integer.value;
 966		} else {
 967			continue;
 968		}
 969
 970		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
 971
 972		obj = pkg_elem + 9;
 973		if (obj->type == ACPI_TYPE_STRING)
 974			strscpy(lpi_state->desc, obj->string.pointer,
 975				ACPI_CX_DESC_LEN);
 976
 977		lpi_state->index = state_idx;
 978		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
 979			pr_debug("No min. residency found, assuming 10 us\n");
 980			lpi_state->min_residency = 10;
 981		}
 982
 983		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
 984			pr_debug("No wakeup residency found, assuming 10 us\n");
 985			lpi_state->wake_latency = 10;
 986		}
 987
 988		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
 989			lpi_state->flags = 0;
 990
 991		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
 992			lpi_state->arch_flags = 0;
 993
 994		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
 995			lpi_state->res_cnt_freq = 1;
 996
 997		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
 998			lpi_state->enable_parent_state = 0;
 999	}
1000
1001	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1002end:
1003	kfree(buffer.pointer);
1004	return ret;
1005}
1006
1007/*
1008 * flat_state_cnt - the number of composite LPI states after the process of flattening
1009 */
1010static int flat_state_cnt;
1011
1012/**
1013 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1014 *
1015 * @local: local LPI state
1016 * @parent: parent LPI state
1017 * @result: composite LPI state
1018 */
1019static bool combine_lpi_states(struct acpi_lpi_state *local,
1020			       struct acpi_lpi_state *parent,
1021			       struct acpi_lpi_state *result)
1022{
1023	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1024		if (!parent->address) /* 0 means autopromotable */
1025			return false;
1026		result->address = local->address + parent->address;
1027	} else {
1028		result->address = parent->address;
1029	}
1030
1031	result->min_residency = max(local->min_residency, parent->min_residency);
1032	result->wake_latency = local->wake_latency + parent->wake_latency;
1033	result->enable_parent_state = parent->enable_parent_state;
1034	result->entry_method = local->entry_method;
1035
1036	result->flags = parent->flags;
1037	result->arch_flags = parent->arch_flags;
1038	result->index = parent->index;
1039
1040	strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1041	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1042	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1043	return true;
1044}
1045
1046#define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1047
1048static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1049				  struct acpi_lpi_state *t)
1050{
1051	curr_level->composite_states[curr_level->composite_states_size++] = t;
1052}
1053
1054static int flatten_lpi_states(struct acpi_processor *pr,
1055			      struct acpi_lpi_states_array *curr_level,
1056			      struct acpi_lpi_states_array *prev_level)
1057{
1058	int i, j, state_count = curr_level->size;
1059	struct acpi_lpi_state *p, *t = curr_level->entries;
1060
1061	curr_level->composite_states_size = 0;
1062	for (j = 0; j < state_count; j++, t++) {
1063		struct acpi_lpi_state *flpi;
1064
1065		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1066			continue;
1067
1068		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1069			pr_warn("Limiting number of LPI states to max (%d)\n",
1070				ACPI_PROCESSOR_MAX_POWER);
1071			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1072			break;
1073		}
1074
1075		flpi = &pr->power.lpi_states[flat_state_cnt];
1076
1077		if (!prev_level) { /* leaf/processor node */
1078			memcpy(flpi, t, sizeof(*t));
1079			stash_composite_state(curr_level, flpi);
1080			flat_state_cnt++;
1081			continue;
1082		}
1083
1084		for (i = 0; i < prev_level->composite_states_size; i++) {
1085			p = prev_level->composite_states[i];
1086			if (t->index <= p->enable_parent_state &&
1087			    combine_lpi_states(p, t, flpi)) {
1088				stash_composite_state(curr_level, flpi);
1089				flat_state_cnt++;
1090				flpi++;
1091			}
1092		}
1093	}
1094
1095	kfree(curr_level->entries);
1096	return 0;
1097}
1098
1099int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1100{
1101	return -EOPNOTSUPP;
1102}
1103
1104static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1105{
1106	int ret, i;
1107	acpi_status status;
1108	acpi_handle handle = pr->handle, pr_ahandle;
1109	struct acpi_device *d = NULL;
1110	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1111
1112	/* make sure our architecture has support */
1113	ret = acpi_processor_ffh_lpi_probe(pr->id);
1114	if (ret == -EOPNOTSUPP)
1115		return ret;
1116
1117	if (!osc_pc_lpi_support_confirmed)
1118		return -EOPNOTSUPP;
1119
1120	if (!acpi_has_method(handle, "_LPI"))
1121		return -EINVAL;
1122
1123	flat_state_cnt = 0;
1124	prev = &info[0];
1125	curr = &info[1];
1126	handle = pr->handle;
1127	ret = acpi_processor_evaluate_lpi(handle, prev);
1128	if (ret)
1129		return ret;
1130	flatten_lpi_states(pr, prev, NULL);
1131
1132	status = acpi_get_parent(handle, &pr_ahandle);
1133	while (ACPI_SUCCESS(status)) {
1134		d = acpi_fetch_acpi_dev(pr_ahandle);
1135		if (!d)
1136			break;
1137
1138		handle = pr_ahandle;
1139
1140		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1141			break;
1142
1143		/* can be optional ? */
1144		if (!acpi_has_method(handle, "_LPI"))
1145			break;
1146
1147		ret = acpi_processor_evaluate_lpi(handle, curr);
1148		if (ret)
1149			break;
1150
1151		/* flatten all the LPI states in this level of hierarchy */
1152		flatten_lpi_states(pr, curr, prev);
1153
1154		tmp = prev, prev = curr, curr = tmp;
1155
1156		status = acpi_get_parent(handle, &pr_ahandle);
1157	}
1158
1159	pr->power.count = flat_state_cnt;
1160	/* reset the index after flattening */
1161	for (i = 0; i < pr->power.count; i++)
1162		pr->power.lpi_states[i].index = i;
1163
1164	/* Tell driver that _LPI is supported. */
1165	pr->flags.has_lpi = 1;
1166	pr->flags.power = 1;
1167
1168	return 0;
1169}
1170
1171int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1172{
1173	return -ENODEV;
1174}
1175
1176/**
1177 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1178 * @dev: the target CPU
1179 * @drv: cpuidle driver containing cpuidle state info
1180 * @index: index of target state
1181 *
1182 * Return: 0 for success or negative value for error
1183 */
1184static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1185			       struct cpuidle_driver *drv, int index)
1186{
1187	struct acpi_processor *pr;
1188	struct acpi_lpi_state *lpi;
1189
1190	pr = __this_cpu_read(processors);
1191
1192	if (unlikely(!pr))
1193		return -EINVAL;
1194
1195	lpi = &pr->power.lpi_states[index];
1196	if (lpi->entry_method == ACPI_CSTATE_FFH)
1197		return acpi_processor_ffh_lpi_enter(lpi);
1198
1199	return -EINVAL;
1200}
1201
1202static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1203{
1204	int i;
1205	struct acpi_lpi_state *lpi;
1206	struct cpuidle_state *state;
1207	struct cpuidle_driver *drv = &acpi_idle_driver;
1208
1209	if (!pr->flags.has_lpi)
1210		return -EOPNOTSUPP;
1211
1212	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1213		lpi = &pr->power.lpi_states[i];
1214
1215		state = &drv->states[i];
1216		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1217		strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1218		state->exit_latency = lpi->wake_latency;
1219		state->target_residency = lpi->min_residency;
1220		state->flags |= arch_get_idle_state_flags(lpi->arch_flags);
1221		if (i != 0 && lpi->entry_method == ACPI_CSTATE_FFH)
1222			state->flags |= CPUIDLE_FLAG_RCU_IDLE;
1223		state->enter = acpi_idle_lpi_enter;
1224		drv->safe_state_index = i;
1225	}
1226
1227	drv->state_count = i;
1228
1229	return 0;
1230}
1231
1232/**
1233 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1234 * global state data i.e. idle routines
1235 *
1236 * @pr: the ACPI processor
1237 */
1238static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1239{
1240	int i;
1241	struct cpuidle_driver *drv = &acpi_idle_driver;
1242
1243	if (!pr->flags.power_setup_done || !pr->flags.power)
1244		return -EINVAL;
1245
1246	drv->safe_state_index = -1;
1247	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1248		drv->states[i].name[0] = '\0';
1249		drv->states[i].desc[0] = '\0';
1250	}
1251
1252	if (pr->flags.has_lpi)
1253		return acpi_processor_setup_lpi_states(pr);
1254
1255	return acpi_processor_setup_cstates(pr);
1256}
1257
1258/**
1259 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1260 * device i.e. per-cpu data
1261 *
1262 * @pr: the ACPI processor
1263 * @dev : the cpuidle device
1264 */
1265static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1266					    struct cpuidle_device *dev)
1267{
1268	if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1269		return -EINVAL;
1270
1271	dev->cpu = pr->id;
1272	if (pr->flags.has_lpi)
1273		return acpi_processor_ffh_lpi_probe(pr->id);
1274
1275	return acpi_processor_setup_cpuidle_cx(pr, dev);
1276}
1277
1278static int acpi_processor_get_power_info(struct acpi_processor *pr)
1279{
1280	int ret;
1281
1282	ret = acpi_processor_get_lpi_info(pr);
1283	if (ret)
1284		ret = acpi_processor_get_cstate_info(pr);
1285
1286	return ret;
1287}
1288
1289int acpi_processor_hotplug(struct acpi_processor *pr)
1290{
1291	int ret = 0;
1292	struct cpuidle_device *dev;
1293
1294	if (disabled_by_idle_boot_param())
1295		return 0;
1296
1297	if (!pr->flags.power_setup_done)
1298		return -ENODEV;
1299
1300	dev = per_cpu(acpi_cpuidle_device, pr->id);
1301	cpuidle_pause_and_lock();
1302	cpuidle_disable_device(dev);
1303	ret = acpi_processor_get_power_info(pr);
1304	if (!ret && pr->flags.power) {
1305		acpi_processor_setup_cpuidle_dev(pr, dev);
1306		ret = cpuidle_enable_device(dev);
1307	}
1308	cpuidle_resume_and_unlock();
1309
1310	return ret;
1311}
1312
1313int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1314{
1315	int cpu;
1316	struct acpi_processor *_pr;
1317	struct cpuidle_device *dev;
1318
1319	if (disabled_by_idle_boot_param())
1320		return 0;
1321
1322	if (!pr->flags.power_setup_done)
1323		return -ENODEV;
1324
1325	/*
1326	 * FIXME:  Design the ACPI notification to make it once per
1327	 * system instead of once per-cpu.  This condition is a hack
1328	 * to make the code that updates C-States be called once.
1329	 */
1330
1331	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1332
1333		/* Protect against cpu-hotplug */
1334		cpus_read_lock();
1335		cpuidle_pause_and_lock();
1336
1337		/* Disable all cpuidle devices */
1338		for_each_online_cpu(cpu) {
1339			_pr = per_cpu(processors, cpu);
1340			if (!_pr || !_pr->flags.power_setup_done)
1341				continue;
1342			dev = per_cpu(acpi_cpuidle_device, cpu);
1343			cpuidle_disable_device(dev);
1344		}
1345
1346		/* Populate Updated C-state information */
1347		acpi_processor_get_power_info(pr);
1348		acpi_processor_setup_cpuidle_states(pr);
1349
1350		/* Enable all cpuidle devices */
1351		for_each_online_cpu(cpu) {
1352			_pr = per_cpu(processors, cpu);
1353			if (!_pr || !_pr->flags.power_setup_done)
1354				continue;
1355			acpi_processor_get_power_info(_pr);
1356			if (_pr->flags.power) {
1357				dev = per_cpu(acpi_cpuidle_device, cpu);
1358				acpi_processor_setup_cpuidle_dev(_pr, dev);
1359				cpuidle_enable_device(dev);
1360			}
1361		}
1362		cpuidle_resume_and_unlock();
1363		cpus_read_unlock();
1364	}
1365
1366	return 0;
1367}
1368
1369static int acpi_processor_registered;
1370
1371int acpi_processor_power_init(struct acpi_processor *pr)
1372{
1373	int retval;
1374	struct cpuidle_device *dev;
1375
1376	if (disabled_by_idle_boot_param())
1377		return 0;
1378
1379	acpi_processor_cstate_first_run_checks();
1380
1381	if (!acpi_processor_get_power_info(pr))
1382		pr->flags.power_setup_done = 1;
1383
1384	/*
1385	 * Install the idle handler if processor power management is supported.
1386	 * Note that we use previously set idle handler will be used on
1387	 * platforms that only support C1.
1388	 */
1389	if (pr->flags.power) {
1390		/* Register acpi_idle_driver if not already registered */
1391		if (!acpi_processor_registered) {
1392			acpi_processor_setup_cpuidle_states(pr);
1393			retval = cpuidle_register_driver(&acpi_idle_driver);
1394			if (retval)
1395				return retval;
1396			pr_debug("%s registered with cpuidle\n",
1397				 acpi_idle_driver.name);
1398		}
1399
1400		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1401		if (!dev)
1402			return -ENOMEM;
1403		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1404
1405		acpi_processor_setup_cpuidle_dev(pr, dev);
1406
1407		/* Register per-cpu cpuidle_device. Cpuidle driver
1408		 * must already be registered before registering device
1409		 */
1410		retval = cpuidle_register_device(dev);
1411		if (retval) {
1412			if (acpi_processor_registered == 0)
1413				cpuidle_unregister_driver(&acpi_idle_driver);
1414			return retval;
1415		}
1416		acpi_processor_registered++;
1417	}
1418	return 0;
1419}
1420
1421int acpi_processor_power_exit(struct acpi_processor *pr)
1422{
1423	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1424
1425	if (disabled_by_idle_boot_param())
1426		return 0;
1427
1428	if (pr->flags.power) {
1429		cpuidle_unregister_device(dev);
1430		acpi_processor_registered--;
1431		if (acpi_processor_registered == 0)
1432			cpuidle_unregister_driver(&acpi_idle_driver);
1433
1434		kfree(dev);
1435	}
1436
1437	pr->flags.power_setup_done = 0;
1438	return 0;
1439}