xref: /linux/drivers/acpi/processor_idle.c (revision 23b0f90ba871f096474e1c27c3d14f455189d2d9)
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 #include "internal.h"
28 
29 /*
30  * Include the apic definitions for x86 to have the APIC timer related defines
31  * available also for UP (on SMP it gets magically included via linux/smp.h).
32  * asm/acpi.h is not an option, as it would require more include magic. Also
33  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
34  */
35 #ifdef CONFIG_X86
36 #include <asm/apic.h>
37 #include <asm/cpu.h>
38 #endif
39 
40 #define ACPI_IDLE_STATE_START	(IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
41 
42 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
43 module_param(max_cstate, uint, 0400);
44 static bool nocst __read_mostly;
45 module_param(nocst, bool, 0400);
46 static bool bm_check_disable __read_mostly;
47 module_param(bm_check_disable, bool, 0400);
48 
49 static unsigned int latency_factor __read_mostly = 2;
50 module_param(latency_factor, uint, 0644);
51 
52 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
53 
54 static struct cpuidle_driver acpi_idle_driver = {
55 	.name =		"acpi_idle",
56 	.owner =	THIS_MODULE,
57 };
58 
59 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
60 void acpi_idle_rescan_dead_smt_siblings(void)
61 {
62 	if (cpuidle_get_driver() == &acpi_idle_driver)
63 		arch_cpu_rescan_dead_smt_siblings();
64 }
65 
66 static
67 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
68 
69 static int disabled_by_idle_boot_param(void)
70 {
71 	return boot_option_idle_override == IDLE_POLL ||
72 		boot_option_idle_override == IDLE_HALT;
73 }
74 
75 /*
76  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
77  * For now disable this. Probably a bug somewhere else.
78  *
79  * To skip this limit, boot/load with a large max_cstate limit.
80  */
81 static int set_max_cstate(const struct dmi_system_id *id)
82 {
83 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
84 		return 0;
85 
86 	pr_notice("%s detected - limiting to C%ld max_cstate."
87 		  " Override with \"processor.max_cstate=%d\"\n", id->ident,
88 		  (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
89 
90 	max_cstate = (long)id->driver_data;
91 
92 	return 0;
93 }
94 
95 static const struct dmi_system_id processor_power_dmi_table[] = {
96 	{ set_max_cstate, "Clevo 5600D", {
97 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
98 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
99 	 (void *)2},
100 	{ set_max_cstate, "Pavilion zv5000", {
101 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
102 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
103 	 (void *)1},
104 	{ set_max_cstate, "Asus L8400B", {
105 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
106 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
107 	 (void *)1},
108 	{},
109 };
110 
111 
112 /*
113  * Callers should disable interrupts before the call and enable
114  * interrupts after return.
115  */
116 static void __cpuidle acpi_safe_halt(void)
117 {
118 	if (!tif_need_resched()) {
119 		raw_safe_halt();
120 		raw_local_irq_disable();
121 	}
122 }
123 
124 #ifdef ARCH_APICTIMER_STOPS_ON_C3
125 
126 /*
127  * Some BIOS implementations switch to C3 in the published C2 state.
128  * This seems to be a common problem on AMD boxen, but other vendors
129  * are affected too. We pick the most conservative approach: we assume
130  * that the local APIC stops in both C2 and C3.
131  */
132 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
133 				   struct acpi_processor_cx *cx)
134 {
135 	struct acpi_processor_power *pwr = &pr->power;
136 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
137 
138 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
139 		return;
140 
141 	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
142 		type = ACPI_STATE_C1;
143 
144 	/*
145 	 * Check, if one of the previous states already marked the lapic
146 	 * unstable
147 	 */
148 	if (pwr->timer_broadcast_on_state < state)
149 		return;
150 
151 	if (cx->type >= type)
152 		pr->power.timer_broadcast_on_state = state;
153 }
154 
155 static void __lapic_timer_propagate_broadcast(void *arg)
156 {
157 	struct acpi_processor *pr = arg;
158 
159 	if (pr->power.timer_broadcast_on_state < INT_MAX)
160 		tick_broadcast_enable();
161 	else
162 		tick_broadcast_disable();
163 }
164 
165 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
166 {
167 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
168 				 (void *)pr, 1);
169 }
170 
171 /* Power(C) State timer broadcast control */
172 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
173 					struct acpi_processor_cx *cx)
174 {
175 	return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
176 }
177 
178 #else
179 
180 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
181 				   struct acpi_processor_cx *cstate) { }
182 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
183 
184 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
185 					struct acpi_processor_cx *cx)
186 {
187 	return false;
188 }
189 
190 #endif
191 
192 #if defined(CONFIG_X86)
193 static void tsc_check_state(int state)
194 {
195 	switch (boot_cpu_data.x86_vendor) {
196 	case X86_VENDOR_HYGON:
197 	case X86_VENDOR_AMD:
198 	case X86_VENDOR_INTEL:
199 	case X86_VENDOR_CENTAUR:
200 	case X86_VENDOR_ZHAOXIN:
201 		/*
202 		 * AMD Fam10h TSC will tick in all
203 		 * C/P/S0/S1 states when this bit is set.
204 		 */
205 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
206 			return;
207 		fallthrough;
208 	default:
209 		/* TSC could halt in idle, so notify users */
210 		if (state > ACPI_STATE_C1)
211 			mark_tsc_unstable("TSC halts in idle");
212 	}
213 }
214 #else
215 static void tsc_check_state(int state) { return; }
216 #endif
217 
218 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
219 {
220 
221 	if (!pr->pblk)
222 		return -ENODEV;
223 
224 	/* if info is obtained from pblk/fadt, type equals state */
225 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
226 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
227 
228 #ifndef CONFIG_HOTPLUG_CPU
229 	/*
230 	 * Check for P_LVL2_UP flag before entering C2 and above on
231 	 * an SMP system.
232 	 */
233 	if ((num_online_cpus() > 1) &&
234 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
235 		return -ENODEV;
236 #endif
237 
238 	/* determine C2 and C3 address from pblk */
239 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
240 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
241 
242 	/* determine latencies from FADT */
243 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
244 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
245 
246 	/*
247 	 * FADT specified C2 latency must be less than or equal to
248 	 * 100 microseconds.
249 	 */
250 	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
251 		acpi_handle_debug(pr->handle, "C2 latency too large [%d]\n",
252 				  acpi_gbl_FADT.c2_latency);
253 		/* invalidate C2 */
254 		pr->power.states[ACPI_STATE_C2].address = 0;
255 	}
256 
257 	/*
258 	 * FADT supplied C3 latency must be less than or equal to
259 	 * 1000 microseconds.
260 	 */
261 	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
262 		acpi_handle_debug(pr->handle, "C3 latency too large [%d]\n",
263 				  acpi_gbl_FADT.c3_latency);
264 		/* invalidate C3 */
265 		pr->power.states[ACPI_STATE_C3].address = 0;
266 	}
267 
268 	acpi_handle_debug(pr->handle, "lvl2[0x%08x] lvl3[0x%08x]\n",
269 			  pr->power.states[ACPI_STATE_C2].address,
270 			  pr->power.states[ACPI_STATE_C3].address);
271 
272 	snprintf(pr->power.states[ACPI_STATE_C2].desc,
273 			 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
274 			 pr->power.states[ACPI_STATE_C2].address);
275 	snprintf(pr->power.states[ACPI_STATE_C3].desc,
276 			 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
277 			 pr->power.states[ACPI_STATE_C3].address);
278 
279 	if (!pr->power.states[ACPI_STATE_C2].address &&
280 	    !pr->power.states[ACPI_STATE_C3].address)
281 		return -ENODEV;
282 
283 	return 0;
284 }
285 
286 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
287 {
288 	if (!pr->power.states[ACPI_STATE_C1].valid) {
289 		/* set the first C-State to C1 */
290 		/* all processors need to support C1 */
291 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
292 		pr->power.states[ACPI_STATE_C1].valid = 1;
293 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
294 
295 		snprintf(pr->power.states[ACPI_STATE_C1].desc,
296 			 ACPI_CX_DESC_LEN, "ACPI HLT");
297 	}
298 	/* the C0 state only exists as a filler in our array */
299 	pr->power.states[ACPI_STATE_C0].valid = 1;
300 	return 0;
301 }
302 
303 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
304 {
305 	int ret;
306 
307 	if (nocst)
308 		return -ENODEV;
309 
310 	ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
311 	if (ret)
312 		return ret;
313 
314 	if (!pr->power.count)
315 		return -EFAULT;
316 
317 	pr->flags.has_cst = 1;
318 	return 0;
319 }
320 
321 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
322 					   struct acpi_processor_cx *cx)
323 {
324 	static int bm_check_flag = -1;
325 	static int bm_control_flag = -1;
326 
327 
328 	if (!cx->address)
329 		return;
330 
331 	/*
332 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
333 	 * DMA transfers are used by any ISA device to avoid livelock.
334 	 * Note that we could disable Type-F DMA (as recommended by
335 	 * the erratum), but this is known to disrupt certain ISA
336 	 * devices thus we take the conservative approach.
337 	 */
338 	if (errata.piix4.fdma) {
339 		acpi_handle_debug(pr->handle,
340 				  "C3 not supported on PIIX4 with Type-F DMA\n");
341 		return;
342 	}
343 
344 	/* All the logic here assumes flags.bm_check is same across all CPUs */
345 	if (bm_check_flag == -1) {
346 		/* Determine whether bm_check is needed based on CPU  */
347 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
348 		bm_check_flag = pr->flags.bm_check;
349 		bm_control_flag = pr->flags.bm_control;
350 	} else {
351 		pr->flags.bm_check = bm_check_flag;
352 		pr->flags.bm_control = bm_control_flag;
353 	}
354 
355 	if (pr->flags.bm_check) {
356 		if (!pr->flags.bm_control) {
357 			if (pr->flags.has_cst != 1) {
358 				/* bus mastering control is necessary */
359 				acpi_handle_debug(pr->handle,
360 						  "C3 support requires BM control\n");
361 				return;
362 			} else {
363 				/* Here we enter C3 without bus mastering */
364 				acpi_handle_debug(pr->handle,
365 						  "C3 support without BM control\n");
366 			}
367 		}
368 	} else {
369 		/*
370 		 * WBINVD should be set in fadt, for C3 state to be
371 		 * supported on when bm_check is not required.
372 		 */
373 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
374 			acpi_handle_debug(pr->handle,
375 					  "Cache invalidation should work properly"
376 					  " for C3 to be enabled on SMP systems\n");
377 			return;
378 		}
379 	}
380 
381 	/*
382 	 * Otherwise we've met all of our C3 requirements.
383 	 * Normalize the C3 latency to expidite policy.  Enable
384 	 * checking of bus mastering status (bm_check) so we can
385 	 * use this in our C3 policy
386 	 */
387 	cx->valid = 1;
388 
389 	/*
390 	 * On older chipsets, BM_RLD needs to be set
391 	 * in order for Bus Master activity to wake the
392 	 * system from C3.  Newer chipsets handle DMA
393 	 * during C3 automatically and BM_RLD is a NOP.
394 	 * In either case, the proper way to
395 	 * handle BM_RLD is to set it and leave it set.
396 	 */
397 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
398 }
399 
400 static void acpi_cst_latency_sort(struct acpi_processor_cx *states, size_t length)
401 {
402 	int i, j, k;
403 
404 	for (i = 1; i < length; i++) {
405 		if (!states[i].valid)
406 			continue;
407 
408 		for (j = i - 1, k = i; j >= 0; j--) {
409 			if (!states[j].valid)
410 				continue;
411 
412 			if (states[j].latency > states[k].latency)
413 				swap(states[j].latency, states[k].latency);
414 
415 			k = j;
416 		}
417 	}
418 }
419 
420 static int acpi_processor_power_verify(struct acpi_processor *pr)
421 {
422 	unsigned int i;
423 	unsigned int working = 0;
424 	unsigned int last_latency = 0;
425 	unsigned int last_type = 0;
426 	bool buggy_latency = false;
427 
428 	pr->power.timer_broadcast_on_state = INT_MAX;
429 
430 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
431 		struct acpi_processor_cx *cx = &pr->power.states[i];
432 
433 		switch (cx->type) {
434 		case ACPI_STATE_C1:
435 			cx->valid = 1;
436 			break;
437 
438 		case ACPI_STATE_C2:
439 			if (!cx->address)
440 				break;
441 			cx->valid = 1;
442 			break;
443 
444 		case ACPI_STATE_C3:
445 			acpi_processor_power_verify_c3(pr, cx);
446 			break;
447 		}
448 		if (!cx->valid)
449 			continue;
450 		if (cx->type >= last_type && cx->latency < last_latency)
451 			buggy_latency = true;
452 		last_latency = cx->latency;
453 		last_type = cx->type;
454 
455 		lapic_timer_check_state(i, pr, cx);
456 		tsc_check_state(cx->type);
457 		working++;
458 	}
459 
460 	if (buggy_latency) {
461 		pr_notice("FW issue: working around C-state latencies out of order\n");
462 		acpi_cst_latency_sort(&pr->power.states[1], max_cstate);
463 	}
464 
465 	lapic_timer_propagate_broadcast(pr);
466 
467 	return working;
468 }
469 
470 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
471 {
472 	int result;
473 
474 	/* NOTE: the idle thread may not be running while calling
475 	 * this function */
476 
477 	/* Zero initialize all the C-states info. */
478 	memset(pr->power.states, 0, sizeof(pr->power.states));
479 
480 	result = acpi_processor_get_power_info_cst(pr);
481 	if (result == -ENODEV)
482 		result = acpi_processor_get_power_info_fadt(pr);
483 
484 	if (result)
485 		return result;
486 
487 	acpi_processor_get_power_info_default(pr);
488 
489 	pr->power.count = acpi_processor_power_verify(pr);
490 	pr->flags.power = 1;
491 
492 	return 0;
493 }
494 
495 /**
496  * acpi_idle_bm_check - checks if bus master activity was detected
497  */
498 static 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 
521 static __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  */
560 static 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  */
581 static 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 if (cx->entry_method == ACPI_CSTATE_FFH) {
594 			acpi_processor_ffh_play_dead(cx);
595 		} else
596 			return;
597 	}
598 }
599 
600 static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
601 {
602 	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
603 		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
604 }
605 
606 static int c3_cpu_count;
607 static DEFINE_RAW_SPINLOCK(c3_lock);
608 
609 /**
610  * acpi_idle_enter_bm - enters C3 with proper BM handling
611  * @drv: cpuidle driver
612  * @pr: Target processor
613  * @cx: Target state context
614  * @index: index of target state
615  */
616 static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
617 			       struct acpi_processor *pr,
618 			       struct acpi_processor_cx *cx,
619 			       int index)
620 {
621 	static struct acpi_processor_cx safe_cx = {
622 		.entry_method = ACPI_CSTATE_HALT,
623 	};
624 
625 	/*
626 	 * disable bus master
627 	 * bm_check implies we need ARB_DIS
628 	 * bm_control implies whether we can do ARB_DIS
629 	 *
630 	 * That leaves a case where bm_check is set and bm_control is not set.
631 	 * In that case we cannot do much, we enter C3 without doing anything.
632 	 */
633 	bool dis_bm = pr->flags.bm_control;
634 
635 	instrumentation_begin();
636 
637 	/* If we can skip BM, demote to a safe state. */
638 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
639 		dis_bm = false;
640 		index = drv->safe_state_index;
641 		if (index >= 0) {
642 			cx = this_cpu_read(acpi_cstate[index]);
643 		} else {
644 			cx = &safe_cx;
645 			index = -EBUSY;
646 		}
647 	}
648 
649 	if (dis_bm) {
650 		raw_spin_lock(&c3_lock);
651 		c3_cpu_count++;
652 		/* Disable bus master arbitration when all CPUs are in C3 */
653 		if (c3_cpu_count == num_online_cpus())
654 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
655 		raw_spin_unlock(&c3_lock);
656 	}
657 
658 	ct_cpuidle_enter();
659 
660 	acpi_idle_do_entry(cx);
661 
662 	ct_cpuidle_exit();
663 
664 	/* Re-enable bus master arbitration */
665 	if (dis_bm) {
666 		raw_spin_lock(&c3_lock);
667 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
668 		c3_cpu_count--;
669 		raw_spin_unlock(&c3_lock);
670 	}
671 
672 	instrumentation_end();
673 
674 	return index;
675 }
676 
677 static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
678 			   struct cpuidle_driver *drv, int index)
679 {
680 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
681 	struct acpi_processor *pr;
682 
683 	pr = __this_cpu_read(processors);
684 	if (unlikely(!pr))
685 		return -EINVAL;
686 
687 	if (cx->type != ACPI_STATE_C1) {
688 		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
689 			return acpi_idle_enter_bm(drv, pr, cx, index);
690 
691 		/* C2 to C1 demotion. */
692 		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
693 			index = ACPI_IDLE_STATE_START;
694 			cx = per_cpu(acpi_cstate[index], dev->cpu);
695 		}
696 	}
697 
698 	if (cx->type == ACPI_STATE_C3)
699 		ACPI_FLUSH_CPU_CACHE();
700 
701 	acpi_idle_do_entry(cx);
702 
703 	return index;
704 }
705 
706 static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
707 				  struct cpuidle_driver *drv, int index)
708 {
709 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
710 
711 	if (cx->type == ACPI_STATE_C3) {
712 		struct acpi_processor *pr = __this_cpu_read(processors);
713 
714 		if (unlikely(!pr))
715 			return 0;
716 
717 		if (pr->flags.bm_check) {
718 			u8 bm_sts_skip = cx->bm_sts_skip;
719 
720 			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
721 			cx->bm_sts_skip = 1;
722 			acpi_idle_enter_bm(drv, pr, cx, index);
723 			cx->bm_sts_skip = bm_sts_skip;
724 
725 			return 0;
726 		} else {
727 			ACPI_FLUSH_CPU_CACHE();
728 		}
729 	}
730 	acpi_idle_do_entry(cx);
731 
732 	return 0;
733 }
734 
735 static void acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
736 					    struct cpuidle_device *dev)
737 {
738 	int i, count = ACPI_IDLE_STATE_START;
739 	struct acpi_processor_cx *cx;
740 
741 	if (max_cstate == 0)
742 		max_cstate = 1;
743 
744 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
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 		count++;
753 		if (count == CPUIDLE_STATE_MAX)
754 			break;
755 	}
756 }
757 
758 static void acpi_processor_setup_cstates(struct acpi_processor *pr)
759 {
760 	int i, count;
761 	struct acpi_processor_cx *cx;
762 	struct cpuidle_state *state;
763 	struct cpuidle_driver *drv = &acpi_idle_driver;
764 
765 	if (max_cstate == 0)
766 		max_cstate = 1;
767 
768 	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
769 		cpuidle_poll_state_init(drv);
770 		count = 1;
771 	} else {
772 		count = 0;
773 	}
774 
775 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
776 		cx = &pr->power.states[i];
777 
778 		if (!cx->valid)
779 			continue;
780 
781 		state = &drv->states[count];
782 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
783 		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
784 		state->exit_latency = cx->latency;
785 		state->target_residency = cx->latency * latency_factor;
786 		state->enter = acpi_idle_enter;
787 
788 		state->flags = 0;
789 
790 		state->enter_dead = acpi_idle_play_dead;
791 
792 		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2)
793 			drv->safe_state_index = count;
794 
795 		/*
796 		 * Halt-induced C1 is not good for ->enter_s2idle, because it
797 		 * re-enables interrupts on exit.  Moreover, C1 is generally not
798 		 * particularly interesting from the suspend-to-idle angle, so
799 		 * avoid C1 and the situations in which we may need to fall back
800 		 * to it altogether.
801 		 */
802 		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
803 			state->enter_s2idle = acpi_idle_enter_s2idle;
804 
805 		if (lapic_timer_needs_broadcast(pr, cx))
806 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
807 
808 		if (cx->type == ACPI_STATE_C3) {
809 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
810 			if (pr->flags.bm_check)
811 				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
812 		}
813 
814 		count++;
815 		if (count == CPUIDLE_STATE_MAX)
816 			break;
817 	}
818 
819 	drv->state_count = count;
820 }
821 
822 static inline void acpi_processor_cstate_first_run_checks(void)
823 {
824 	static int first_run;
825 
826 	if (first_run)
827 		return;
828 	dmi_check_system(processor_power_dmi_table);
829 	max_cstate = acpi_processor_cstate_check(max_cstate);
830 	if (max_cstate < ACPI_C_STATES_MAX)
831 		pr_notice("processor limited to max C-state %d\n", max_cstate);
832 
833 	first_run++;
834 
835 	if (nocst)
836 		return;
837 
838 	acpi_processor_claim_cst_control();
839 }
840 #else
841 
842 static inline int disabled_by_idle_boot_param(void) { return 0; }
843 static inline void acpi_processor_cstate_first_run_checks(void) { }
844 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
845 {
846 	return -ENODEV;
847 }
848 
849 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
850 					   struct cpuidle_device *dev)
851 {
852 	return -EINVAL;
853 }
854 
855 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
856 {
857 	return -EINVAL;
858 }
859 
860 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
861 
862 struct acpi_lpi_states_array {
863 	unsigned int size;
864 	unsigned int composite_states_size;
865 	struct acpi_lpi_state *entries;
866 	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
867 };
868 
869 static int obj_get_integer(union acpi_object *obj, u32 *value)
870 {
871 	if (obj->type != ACPI_TYPE_INTEGER)
872 		return -EINVAL;
873 
874 	*value = obj->integer.value;
875 	return 0;
876 }
877 
878 static int acpi_processor_evaluate_lpi(acpi_handle handle,
879 				       struct acpi_lpi_states_array *info)
880 {
881 	acpi_status status;
882 	int ret = 0;
883 	int pkg_count, state_idx = 1, loop;
884 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
885 	union acpi_object *lpi_data;
886 	struct acpi_lpi_state *lpi_state;
887 
888 	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
889 	if (ACPI_FAILURE(status)) {
890 		acpi_handle_debug(handle, "No _LPI, giving up\n");
891 		return -ENODEV;
892 	}
893 
894 	lpi_data = buffer.pointer;
895 
896 	/* There must be at least 4 elements = 3 elements + 1 package */
897 	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
898 	    lpi_data->package.count < 4) {
899 		pr_debug("not enough elements in _LPI\n");
900 		ret = -ENODATA;
901 		goto end;
902 	}
903 
904 	pkg_count = lpi_data->package.elements[2].integer.value;
905 
906 	/* Validate number of power states. */
907 	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
908 		pr_debug("count given by _LPI is not valid\n");
909 		ret = -ENODATA;
910 		goto end;
911 	}
912 
913 	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
914 	if (!lpi_state) {
915 		ret = -ENOMEM;
916 		goto end;
917 	}
918 
919 	info->size = pkg_count;
920 	info->entries = lpi_state;
921 
922 	/* LPI States start at index 3 */
923 	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
924 		union acpi_object *element, *pkg_elem, *obj;
925 
926 		element = &lpi_data->package.elements[loop];
927 		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
928 			continue;
929 
930 		pkg_elem = element->package.elements;
931 
932 		obj = pkg_elem + 6;
933 		if (obj->type == ACPI_TYPE_BUFFER) {
934 			struct acpi_power_register *reg;
935 
936 			reg = (struct acpi_power_register *)obj->buffer.pointer;
937 			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
938 			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
939 				continue;
940 
941 			lpi_state->address = reg->address;
942 			lpi_state->entry_method =
943 				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
944 				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
945 		} else if (obj->type == ACPI_TYPE_INTEGER) {
946 			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
947 			lpi_state->address = obj->integer.value;
948 		} else {
949 			pr_debug("Entry method of state-%d is invalid, disable it.\n",
950 				 state_idx);
951 			continue;
952 		}
953 
954 		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
955 
956 		obj = pkg_elem + 9;
957 		if (obj->type == ACPI_TYPE_STRING)
958 			strscpy(lpi_state->desc, obj->string.pointer,
959 				ACPI_CX_DESC_LEN);
960 
961 		lpi_state->index = state_idx;
962 		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
963 			pr_debug("No min. residency found, assuming 10 us\n");
964 			lpi_state->min_residency = 10;
965 		}
966 
967 		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
968 			pr_debug("No wakeup residency found, assuming 10 us\n");
969 			lpi_state->wake_latency = 10;
970 		}
971 
972 		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
973 			lpi_state->flags = 0;
974 
975 		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
976 			lpi_state->arch_flags = 0;
977 
978 		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
979 			lpi_state->res_cnt_freq = 1;
980 
981 		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
982 			lpi_state->enable_parent_state = 0;
983 	}
984 
985 	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
986 end:
987 	kfree(buffer.pointer);
988 	return ret;
989 }
990 
991 /**
992  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
993  *
994  * @local: local LPI state
995  * @parent: parent LPI state
996  * @result: composite LPI state
997  */
998 static bool combine_lpi_states(struct acpi_lpi_state *local,
999 			       struct acpi_lpi_state *parent,
1000 			       struct acpi_lpi_state *result)
1001 {
1002 	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1003 		if (!parent->address) /* 0 means autopromotable */
1004 			return false;
1005 		result->address = local->address + parent->address;
1006 	} else {
1007 		result->address = parent->address;
1008 	}
1009 
1010 	result->min_residency = max(local->min_residency, parent->min_residency);
1011 	result->wake_latency = local->wake_latency + parent->wake_latency;
1012 	result->enable_parent_state = parent->enable_parent_state;
1013 	result->entry_method = local->entry_method;
1014 
1015 	result->flags = parent->flags;
1016 	result->arch_flags = parent->arch_flags;
1017 	result->index = parent->index;
1018 
1019 	strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1020 	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1021 	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1022 	return true;
1023 }
1024 
1025 #define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1026 
1027 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1028 				  struct acpi_lpi_state *t)
1029 {
1030 	curr_level->composite_states[curr_level->composite_states_size++] = t;
1031 }
1032 
1033 static unsigned int flatten_lpi_states(struct acpi_processor *pr,
1034 				       unsigned int flat_state_cnt,
1035 				       struct acpi_lpi_states_array *curr_level,
1036 				       struct acpi_lpi_states_array *prev_level)
1037 {
1038 	int i, j, state_count = curr_level->size;
1039 	struct acpi_lpi_state *p, *t = curr_level->entries;
1040 
1041 	curr_level->composite_states_size = 0;
1042 	for (j = 0; j < state_count; j++, t++) {
1043 		struct acpi_lpi_state *flpi;
1044 
1045 		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1046 			continue;
1047 
1048 		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1049 			pr_warn("Limiting number of LPI states to max (%d)\n",
1050 				ACPI_PROCESSOR_MAX_POWER);
1051 			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1052 			break;
1053 		}
1054 
1055 		flpi = &pr->power.lpi_states[flat_state_cnt];
1056 
1057 		if (!prev_level) { /* leaf/processor node */
1058 			memcpy(flpi, t, sizeof(*t));
1059 			stash_composite_state(curr_level, flpi);
1060 			flat_state_cnt++;
1061 			continue;
1062 		}
1063 
1064 		for (i = 0; i < prev_level->composite_states_size; i++) {
1065 			p = prev_level->composite_states[i];
1066 			if (t->index <= p->enable_parent_state &&
1067 			    combine_lpi_states(p, t, flpi)) {
1068 				stash_composite_state(curr_level, flpi);
1069 				flat_state_cnt++;
1070 				flpi++;
1071 			}
1072 		}
1073 	}
1074 
1075 	kfree(curr_level->entries);
1076 	return flat_state_cnt;
1077 }
1078 
1079 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1080 {
1081 	return -EOPNOTSUPP;
1082 }
1083 
1084 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1085 {
1086 	int ret, i;
1087 	acpi_status status;
1088 	acpi_handle handle = pr->handle, pr_ahandle;
1089 	struct acpi_device *d = NULL;
1090 	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1091 	unsigned int state_count;
1092 
1093 	/* make sure our architecture has support */
1094 	ret = acpi_processor_ffh_lpi_probe(pr->id);
1095 	if (ret == -EOPNOTSUPP)
1096 		return ret;
1097 
1098 	if (!osc_pc_lpi_support_confirmed)
1099 		return -EOPNOTSUPP;
1100 
1101 	if (!acpi_has_method(handle, "_LPI"))
1102 		return -EINVAL;
1103 
1104 	prev = &info[0];
1105 	curr = &info[1];
1106 	handle = pr->handle;
1107 	ret = acpi_processor_evaluate_lpi(handle, prev);
1108 	if (ret)
1109 		return ret;
1110 	state_count = flatten_lpi_states(pr, 0, prev, NULL);
1111 
1112 	status = acpi_get_parent(handle, &pr_ahandle);
1113 	while (ACPI_SUCCESS(status)) {
1114 		d = acpi_fetch_acpi_dev(pr_ahandle);
1115 		if (!d)
1116 			break;
1117 
1118 		handle = pr_ahandle;
1119 
1120 		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1121 			break;
1122 
1123 		/* can be optional ? */
1124 		if (!acpi_has_method(handle, "_LPI"))
1125 			break;
1126 
1127 		ret = acpi_processor_evaluate_lpi(handle, curr);
1128 		if (ret)
1129 			break;
1130 
1131 		/* flatten all the LPI states in this level of hierarchy */
1132 		state_count = flatten_lpi_states(pr, state_count, curr, prev);
1133 
1134 		tmp = prev, prev = curr, curr = tmp;
1135 
1136 		status = acpi_get_parent(handle, &pr_ahandle);
1137 	}
1138 
1139 	/* reset the index after flattening */
1140 	for (i = 0; i < state_count; i++)
1141 		pr->power.lpi_states[i].index = i;
1142 
1143 	pr->power.count = state_count;
1144 
1145 	/* Tell driver that _LPI is supported. */
1146 	pr->flags.has_lpi = 1;
1147 	pr->flags.power = 1;
1148 
1149 	return 0;
1150 }
1151 
1152 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1153 {
1154 	return -ENODEV;
1155 }
1156 
1157 /**
1158  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1159  * @dev: the target CPU
1160  * @drv: cpuidle driver containing cpuidle state info
1161  * @index: index of target state
1162  *
1163  * Return: 0 for success or negative value for error
1164  */
1165 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1166 			       struct cpuidle_driver *drv, int index)
1167 {
1168 	struct acpi_processor *pr;
1169 	struct acpi_lpi_state *lpi;
1170 
1171 	pr = __this_cpu_read(processors);
1172 
1173 	if (unlikely(!pr))
1174 		return -EINVAL;
1175 
1176 	lpi = &pr->power.lpi_states[index];
1177 	if (lpi->entry_method == ACPI_CSTATE_FFH)
1178 		return acpi_processor_ffh_lpi_enter(lpi);
1179 
1180 	return -EINVAL;
1181 }
1182 
1183 static void acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1184 {
1185 	int i;
1186 	struct acpi_lpi_state *lpi;
1187 	struct cpuidle_state *state;
1188 	struct cpuidle_driver *drv = &acpi_idle_driver;
1189 
1190 	if (!pr->flags.has_lpi)
1191 		return;
1192 
1193 	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1194 		lpi = &pr->power.lpi_states[i];
1195 
1196 		state = &drv->states[i];
1197 		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1198 		strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1199 		state->exit_latency = lpi->wake_latency;
1200 		state->target_residency = lpi->min_residency;
1201 		state->flags |= arch_get_idle_state_flags(lpi->arch_flags);
1202 		if (i != 0 && lpi->entry_method == ACPI_CSTATE_FFH)
1203 			state->flags |= CPUIDLE_FLAG_RCU_IDLE;
1204 		state->enter = acpi_idle_lpi_enter;
1205 		drv->safe_state_index = i;
1206 	}
1207 
1208 	drv->state_count = i;
1209 }
1210 
1211 /**
1212  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1213  * global state data i.e. idle routines
1214  *
1215  * @pr: the ACPI processor
1216  */
1217 static void acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1218 {
1219 	int i;
1220 	struct cpuidle_driver *drv = &acpi_idle_driver;
1221 
1222 	if (!pr->flags.power_setup_done || !pr->flags.power)
1223 		return;
1224 
1225 	drv->safe_state_index = -1;
1226 	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1227 		drv->states[i].name[0] = '\0';
1228 		drv->states[i].desc[0] = '\0';
1229 	}
1230 
1231 	if (pr->flags.has_lpi) {
1232 		acpi_processor_setup_lpi_states(pr);
1233 		return;
1234 	}
1235 
1236 	acpi_processor_setup_cstates(pr);
1237 }
1238 
1239 /**
1240  * acpi_processor_setup_cpuidle_dev - configures CPUIDLE
1241  * device i.e. per-cpu data
1242  *
1243  * @pr: the ACPI processor
1244  * @dev : the cpuidle device
1245  */
1246 static void acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1247 					     struct cpuidle_device *dev)
1248 {
1249 	if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1250 		return;
1251 
1252 	dev->cpu = pr->id;
1253 	if (!pr->flags.has_lpi)
1254 		acpi_processor_setup_cpuidle_cx(pr, dev);
1255 }
1256 
1257 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1258 {
1259 	int ret;
1260 
1261 	ret = acpi_processor_get_lpi_info(pr);
1262 	if (ret)
1263 		return acpi_processor_get_cstate_info(pr);
1264 
1265 	if (pr->flags.has_lpi) {
1266 		ret = acpi_processor_ffh_lpi_probe(pr->id);
1267 		if (ret)
1268 			pr_err("CPU%u: Invalid FFH LPI data\n", pr->id);
1269 	}
1270 
1271 	return ret;
1272 }
1273 
1274 int acpi_processor_hotplug(struct acpi_processor *pr)
1275 {
1276 	int ret = 0;
1277 	struct cpuidle_device *dev;
1278 
1279 	if (disabled_by_idle_boot_param())
1280 		return 0;
1281 
1282 	if (!pr->flags.power_setup_done)
1283 		return -ENODEV;
1284 
1285 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1286 	cpuidle_pause_and_lock();
1287 	cpuidle_disable_device(dev);
1288 	ret = acpi_processor_get_power_info(pr);
1289 	if (!ret && pr->flags.power) {
1290 		acpi_processor_setup_cpuidle_dev(pr, dev);
1291 		ret = cpuidle_enable_device(dev);
1292 	}
1293 	cpuidle_resume_and_unlock();
1294 
1295 	return ret;
1296 }
1297 
1298 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1299 {
1300 	int cpu;
1301 	struct acpi_processor *_pr;
1302 	struct cpuidle_device *dev;
1303 
1304 	if (disabled_by_idle_boot_param())
1305 		return 0;
1306 
1307 	if (!pr->flags.power_setup_done)
1308 		return -ENODEV;
1309 
1310 	/*
1311 	 * FIXME:  Design the ACPI notification to make it once per
1312 	 * system instead of once per-cpu.  This condition is a hack
1313 	 * to make the code that updates C-States be called once.
1314 	 */
1315 
1316 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1317 
1318 		/* Protect against cpu-hotplug */
1319 		cpus_read_lock();
1320 		cpuidle_pause_and_lock();
1321 
1322 		/* Disable all cpuidle devices */
1323 		for_each_online_cpu(cpu) {
1324 			_pr = per_cpu(processors, cpu);
1325 			if (!_pr || !_pr->flags.power_setup_done)
1326 				continue;
1327 			dev = per_cpu(acpi_cpuidle_device, cpu);
1328 			cpuidle_disable_device(dev);
1329 		}
1330 
1331 		/* Populate Updated C-state information */
1332 		acpi_processor_get_power_info(pr);
1333 		acpi_processor_setup_cpuidle_states(pr);
1334 
1335 		/* Enable all cpuidle devices */
1336 		for_each_online_cpu(cpu) {
1337 			_pr = per_cpu(processors, cpu);
1338 			if (!_pr || !_pr->flags.power_setup_done)
1339 				continue;
1340 			acpi_processor_get_power_info(_pr);
1341 			if (_pr->flags.power) {
1342 				dev = per_cpu(acpi_cpuidle_device, cpu);
1343 				acpi_processor_setup_cpuidle_dev(_pr, dev);
1344 				cpuidle_enable_device(dev);
1345 			}
1346 		}
1347 		cpuidle_resume_and_unlock();
1348 		cpus_read_unlock();
1349 	}
1350 
1351 	return 0;
1352 }
1353 
1354 void acpi_processor_register_idle_driver(void)
1355 {
1356 	struct acpi_processor *pr;
1357 	int ret = -ENODEV;
1358 	int cpu;
1359 
1360 	/*
1361 	 * ACPI idle driver is used by all possible CPUs.
1362 	 * Use the processor power info of one in them to set up idle states.
1363 	 * Note that the existing idle handler will be used on platforms that
1364 	 * only support C1.
1365 	 */
1366 	for_each_possible_cpu(cpu) {
1367 		pr = per_cpu(processors, cpu);
1368 		if (!pr)
1369 			continue;
1370 
1371 		acpi_processor_cstate_first_run_checks();
1372 		ret = acpi_processor_get_power_info(pr);
1373 		if (!ret) {
1374 			pr->flags.power_setup_done = 1;
1375 			acpi_processor_setup_cpuidle_states(pr);
1376 			break;
1377 		}
1378 	}
1379 
1380 	if (ret) {
1381 		pr_debug("No ACPI power information from any CPUs.\n");
1382 		return;
1383 	}
1384 
1385 	ret = cpuidle_register_driver(&acpi_idle_driver);
1386 	if (ret) {
1387 		pr_debug("register %s failed.\n", acpi_idle_driver.name);
1388 		return;
1389 	}
1390 	pr_debug("%s registered with cpuidle.\n", acpi_idle_driver.name);
1391 }
1392 
1393 void acpi_processor_unregister_idle_driver(void)
1394 {
1395 	cpuidle_unregister_driver(&acpi_idle_driver);
1396 }
1397 
1398 void acpi_processor_power_init(struct acpi_processor *pr)
1399 {
1400 	struct cpuidle_device *dev;
1401 
1402 	/*
1403 	 * The code below only works if the current cpuidle driver is the ACPI
1404 	 * idle driver.
1405 	 */
1406 	if (cpuidle_get_driver() != &acpi_idle_driver)
1407 		return;
1408 
1409 	if (disabled_by_idle_boot_param())
1410 		return;
1411 
1412 	acpi_processor_cstate_first_run_checks();
1413 
1414 	if (!acpi_processor_get_power_info(pr))
1415 		pr->flags.power_setup_done = 1;
1416 
1417 	if (!pr->flags.power)
1418 		return;
1419 
1420 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1421 	if (!dev)
1422 		return;
1423 
1424 	per_cpu(acpi_cpuidle_device, pr->id) = dev;
1425 
1426 	acpi_processor_setup_cpuidle_dev(pr, dev);
1427 
1428 	/*
1429 	 * Register a cpuidle device for this CPU.  The cpuidle driver using
1430 	 * this device is expected to be registered.
1431 	 */
1432 	if (cpuidle_register_device(dev)) {
1433 		per_cpu(acpi_cpuidle_device, pr->id) = NULL;
1434 		kfree(dev);
1435 	}
1436 }
1437 
1438 void acpi_processor_power_exit(struct acpi_processor *pr)
1439 {
1440 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1441 
1442 	if (disabled_by_idle_boot_param())
1443 		return;
1444 
1445 	if (pr->flags.power) {
1446 		cpuidle_unregister_device(dev);
1447 		kfree(dev);
1448 	}
1449 
1450 	pr->flags.power_setup_done = 0;
1451 }
1452 
1453 MODULE_IMPORT_NS("ACPI_PROCESSOR_IDLE");
1454