xref: /linux/drivers/acpi/processor_idle.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
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 
40 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
41 module_param(max_cstate, uint, 0400);
42 static bool nocst __read_mostly;
43 module_param(nocst, bool, 0400);
44 static bool bm_check_disable __read_mostly;
45 module_param(bm_check_disable, bool, 0400);
46 
47 static unsigned int latency_factor __read_mostly = 2;
48 module_param(latency_factor, uint, 0644);
49 
50 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
51 
52 struct cpuidle_driver acpi_idle_driver = {
53 	.name =		"acpi_idle",
54 	.owner =	THIS_MODULE,
55 };
56 
57 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
58 static
59 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
60 
61 static int disabled_by_idle_boot_param(void)
62 {
63 	return boot_option_idle_override == IDLE_POLL ||
64 		boot_option_idle_override == IDLE_HALT;
65 }
66 
67 /*
68  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
69  * For now disable this. Probably a bug somewhere else.
70  *
71  * To skip this limit, boot/load with a large max_cstate limit.
72  */
73 static int set_max_cstate(const struct dmi_system_id *id)
74 {
75 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
76 		return 0;
77 
78 	pr_notice("%s detected - limiting to C%ld max_cstate."
79 		  " Override with \"processor.max_cstate=%d\"\n", id->ident,
80 		  (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
81 
82 	max_cstate = (long)id->driver_data;
83 
84 	return 0;
85 }
86 
87 static const struct dmi_system_id processor_power_dmi_table[] = {
88 	{ set_max_cstate, "Clevo 5600D", {
89 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
90 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
91 	 (void *)2},
92 	{ set_max_cstate, "Pavilion zv5000", {
93 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
94 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
95 	 (void *)1},
96 	{ set_max_cstate, "Asus L8400B", {
97 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
98 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
99 	 (void *)1},
100 	{},
101 };
102 
103 
104 /*
105  * Callers should disable interrupts before the call and enable
106  * interrupts after return.
107  */
108 static void __cpuidle acpi_safe_halt(void)
109 {
110 	if (!tif_need_resched()) {
111 		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  */
124 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
125 				   struct acpi_processor_cx *cx)
126 {
127 	struct acpi_processor_power *pwr = &pr->power;
128 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
129 
130 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
131 		return;
132 
133 	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
134 		type = ACPI_STATE_C1;
135 
136 	/*
137 	 * Check, if one of the previous states already marked the lapic
138 	 * unstable
139 	 */
140 	if (pwr->timer_broadcast_on_state < state)
141 		return;
142 
143 	if (cx->type >= type)
144 		pr->power.timer_broadcast_on_state = state;
145 }
146 
147 static void __lapic_timer_propagate_broadcast(void *arg)
148 {
149 	struct acpi_processor *pr = arg;
150 
151 	if (pr->power.timer_broadcast_on_state < INT_MAX)
152 		tick_broadcast_enable();
153 	else
154 		tick_broadcast_disable();
155 }
156 
157 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
158 {
159 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
160 				 (void *)pr, 1);
161 }
162 
163 /* Power(C) State timer broadcast control */
164 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
165 					struct acpi_processor_cx *cx)
166 {
167 	return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
168 }
169 
170 #else
171 
172 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
173 				   struct acpi_processor_cx *cstate) { }
174 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
175 
176 static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
177 					struct acpi_processor_cx *cx)
178 {
179 	return false;
180 }
181 
182 #endif
183 
184 #if defined(CONFIG_X86)
185 static 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
207 static void tsc_check_state(int state) { return; }
208 #endif
209 
210 static 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 
274 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
275 {
276 	if (!pr->power.states[ACPI_STATE_C1].valid) {
277 		/* set the first C-State to C1 */
278 		/* all processors need to support C1 */
279 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
280 		pr->power.states[ACPI_STATE_C1].valid = 1;
281 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
282 
283 		snprintf(pr->power.states[ACPI_STATE_C1].desc,
284 			 ACPI_CX_DESC_LEN, "ACPI HLT");
285 	}
286 	/* the C0 state only exists as a filler in our array */
287 	pr->power.states[ACPI_STATE_C0].valid = 1;
288 	return 0;
289 }
290 
291 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
292 {
293 	int ret;
294 
295 	if (nocst)
296 		return -ENODEV;
297 
298 	ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
299 	if (ret)
300 		return ret;
301 
302 	if (!pr->power.count)
303 		return -EFAULT;
304 
305 	pr->flags.has_cst = 1;
306 	return 0;
307 }
308 
309 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
310 					   struct acpi_processor_cx *cx)
311 {
312 	static int bm_check_flag = -1;
313 	static int bm_control_flag = -1;
314 
315 
316 	if (!cx->address)
317 		return;
318 
319 	/*
320 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
321 	 * DMA transfers are used by any ISA device to avoid livelock.
322 	 * Note that we could disable Type-F DMA (as recommended by
323 	 * the erratum), but this is known to disrupt certain ISA
324 	 * devices thus we take the conservative approach.
325 	 */
326 	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 
388 static 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 
408 static 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 
458 static 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  */
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 int 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 -ENODEV;
595 	}
596 
597 	/* Never reached */
598 	return 0;
599 }
600 
601 static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
602 {
603 	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
604 		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
605 }
606 
607 static int c3_cpu_count;
608 static DEFINE_RAW_SPINLOCK(c3_lock);
609 
610 /**
611  * acpi_idle_enter_bm - enters C3 with proper BM handling
612  * @drv: cpuidle driver
613  * @pr: Target processor
614  * @cx: Target state context
615  * @index: index of target state
616  */
617 static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
618 			       struct acpi_processor *pr,
619 			       struct acpi_processor_cx *cx,
620 			       int index)
621 {
622 	static struct acpi_processor_cx safe_cx = {
623 		.entry_method = ACPI_CSTATE_HALT,
624 	};
625 
626 	/*
627 	 * disable bus master
628 	 * bm_check implies we need ARB_DIS
629 	 * bm_control implies whether we can do ARB_DIS
630 	 *
631 	 * That leaves a case where bm_check is set and bm_control is not set.
632 	 * In that case we cannot do much, we enter C3 without doing anything.
633 	 */
634 	bool dis_bm = pr->flags.bm_control;
635 
636 	instrumentation_begin();
637 
638 	/* If we can skip BM, demote to a safe state. */
639 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
640 		dis_bm = false;
641 		index = drv->safe_state_index;
642 		if (index >= 0) {
643 			cx = this_cpu_read(acpi_cstate[index]);
644 		} else {
645 			cx = &safe_cx;
646 			index = -EBUSY;
647 		}
648 	}
649 
650 	if (dis_bm) {
651 		raw_spin_lock(&c3_lock);
652 		c3_cpu_count++;
653 		/* Disable bus master arbitration when all CPUs are in C3 */
654 		if (c3_cpu_count == num_online_cpus())
655 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
656 		raw_spin_unlock(&c3_lock);
657 	}
658 
659 	ct_cpuidle_enter();
660 
661 	acpi_idle_do_entry(cx);
662 
663 	ct_cpuidle_exit();
664 
665 	/* Re-enable bus master arbitration */
666 	if (dis_bm) {
667 		raw_spin_lock(&c3_lock);
668 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
669 		c3_cpu_count--;
670 		raw_spin_unlock(&c3_lock);
671 	}
672 
673 	instrumentation_end();
674 
675 	return index;
676 }
677 
678 static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
679 			   struct cpuidle_driver *drv, int index)
680 {
681 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
682 	struct acpi_processor *pr;
683 
684 	pr = __this_cpu_read(processors);
685 	if (unlikely(!pr))
686 		return -EINVAL;
687 
688 	if (cx->type != ACPI_STATE_C1) {
689 		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
690 			return acpi_idle_enter_bm(drv, pr, cx, index);
691 
692 		/* C2 to C1 demotion. */
693 		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
694 			index = ACPI_IDLE_STATE_START;
695 			cx = per_cpu(acpi_cstate[index], dev->cpu);
696 		}
697 	}
698 
699 	if (cx->type == ACPI_STATE_C3)
700 		ACPI_FLUSH_CPU_CACHE();
701 
702 	acpi_idle_do_entry(cx);
703 
704 	return index;
705 }
706 
707 static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
708 				  struct cpuidle_driver *drv, int index)
709 {
710 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
711 
712 	if (cx->type == ACPI_STATE_C3) {
713 		struct acpi_processor *pr = __this_cpu_read(processors);
714 
715 		if (unlikely(!pr))
716 			return 0;
717 
718 		if (pr->flags.bm_check) {
719 			u8 bm_sts_skip = cx->bm_sts_skip;
720 
721 			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
722 			cx->bm_sts_skip = 1;
723 			acpi_idle_enter_bm(drv, pr, cx, index);
724 			cx->bm_sts_skip = bm_sts_skip;
725 
726 			return 0;
727 		} else {
728 			ACPI_FLUSH_CPU_CACHE();
729 		}
730 	}
731 	acpi_idle_do_entry(cx);
732 
733 	return 0;
734 }
735 
736 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
737 					   struct cpuidle_device *dev)
738 {
739 	int i, count = ACPI_IDLE_STATE_START;
740 	struct acpi_processor_cx *cx;
741 	struct cpuidle_state *state;
742 
743 	if (max_cstate == 0)
744 		max_cstate = 1;
745 
746 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
747 		state = &acpi_idle_driver.states[count];
748 		cx = &pr->power.states[i];
749 
750 		if (!cx->valid)
751 			continue;
752 
753 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
754 
755 		if (lapic_timer_needs_broadcast(pr, cx))
756 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
757 
758 		if (cx->type == ACPI_STATE_C3) {
759 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
760 			if (pr->flags.bm_check)
761 				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
762 		}
763 
764 		count++;
765 		if (count == CPUIDLE_STATE_MAX)
766 			break;
767 	}
768 
769 	if (!count)
770 		return -EINVAL;
771 
772 	return 0;
773 }
774 
775 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
776 {
777 	int i, count;
778 	struct acpi_processor_cx *cx;
779 	struct cpuidle_state *state;
780 	struct cpuidle_driver *drv = &acpi_idle_driver;
781 
782 	if (max_cstate == 0)
783 		max_cstate = 1;
784 
785 	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
786 		cpuidle_poll_state_init(drv);
787 		count = 1;
788 	} else {
789 		count = 0;
790 	}
791 
792 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
793 		cx = &pr->power.states[i];
794 
795 		if (!cx->valid)
796 			continue;
797 
798 		state = &drv->states[count];
799 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
800 		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
801 		state->exit_latency = cx->latency;
802 		state->target_residency = cx->latency * latency_factor;
803 		state->enter = acpi_idle_enter;
804 
805 		state->flags = 0;
806 		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
807 		    cx->type == ACPI_STATE_C3) {
808 			state->enter_dead = acpi_idle_play_dead;
809 			if (cx->type != ACPI_STATE_C3)
810 				drv->safe_state_index = count;
811 		}
812 		/*
813 		 * Halt-induced C1 is not good for ->enter_s2idle, because it
814 		 * re-enables interrupts on exit.  Moreover, C1 is generally not
815 		 * particularly interesting from the suspend-to-idle angle, so
816 		 * avoid C1 and the situations in which we may need to fall back
817 		 * to it altogether.
818 		 */
819 		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
820 			state->enter_s2idle = acpi_idle_enter_s2idle;
821 
822 		count++;
823 		if (count == CPUIDLE_STATE_MAX)
824 			break;
825 	}
826 
827 	drv->state_count = count;
828 
829 	if (!count)
830 		return -EINVAL;
831 
832 	return 0;
833 }
834 
835 static inline void acpi_processor_cstate_first_run_checks(void)
836 {
837 	static int first_run;
838 
839 	if (first_run)
840 		return;
841 	dmi_check_system(processor_power_dmi_table);
842 	max_cstate = acpi_processor_cstate_check(max_cstate);
843 	if (max_cstate < ACPI_C_STATES_MAX)
844 		pr_notice("processor limited to max C-state %d\n", max_cstate);
845 
846 	first_run++;
847 
848 	if (nocst)
849 		return;
850 
851 	acpi_processor_claim_cst_control();
852 }
853 #else
854 
855 static inline int disabled_by_idle_boot_param(void) { return 0; }
856 static inline void acpi_processor_cstate_first_run_checks(void) { }
857 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
858 {
859 	return -ENODEV;
860 }
861 
862 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
863 					   struct cpuidle_device *dev)
864 {
865 	return -EINVAL;
866 }
867 
868 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
869 {
870 	return -EINVAL;
871 }
872 
873 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
874 
875 struct acpi_lpi_states_array {
876 	unsigned int size;
877 	unsigned int composite_states_size;
878 	struct acpi_lpi_state *entries;
879 	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
880 };
881 
882 static int obj_get_integer(union acpi_object *obj, u32 *value)
883 {
884 	if (obj->type != ACPI_TYPE_INTEGER)
885 		return -EINVAL;
886 
887 	*value = obj->integer.value;
888 	return 0;
889 }
890 
891 static int acpi_processor_evaluate_lpi(acpi_handle handle,
892 				       struct acpi_lpi_states_array *info)
893 {
894 	acpi_status status;
895 	int ret = 0;
896 	int pkg_count, state_idx = 1, loop;
897 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
898 	union acpi_object *lpi_data;
899 	struct acpi_lpi_state *lpi_state;
900 
901 	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
902 	if (ACPI_FAILURE(status)) {
903 		acpi_handle_debug(handle, "No _LPI, giving up\n");
904 		return -ENODEV;
905 	}
906 
907 	lpi_data = buffer.pointer;
908 
909 	/* There must be at least 4 elements = 3 elements + 1 package */
910 	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
911 	    lpi_data->package.count < 4) {
912 		pr_debug("not enough elements in _LPI\n");
913 		ret = -ENODATA;
914 		goto end;
915 	}
916 
917 	pkg_count = lpi_data->package.elements[2].integer.value;
918 
919 	/* Validate number of power states. */
920 	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
921 		pr_debug("count given by _LPI is not valid\n");
922 		ret = -ENODATA;
923 		goto end;
924 	}
925 
926 	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
927 	if (!lpi_state) {
928 		ret = -ENOMEM;
929 		goto end;
930 	}
931 
932 	info->size = pkg_count;
933 	info->entries = lpi_state;
934 
935 	/* LPI States start at index 3 */
936 	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
937 		union acpi_object *element, *pkg_elem, *obj;
938 
939 		element = &lpi_data->package.elements[loop];
940 		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
941 			continue;
942 
943 		pkg_elem = element->package.elements;
944 
945 		obj = pkg_elem + 6;
946 		if (obj->type == ACPI_TYPE_BUFFER) {
947 			struct acpi_power_register *reg;
948 
949 			reg = (struct acpi_power_register *)obj->buffer.pointer;
950 			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
951 			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
952 				continue;
953 
954 			lpi_state->address = reg->address;
955 			lpi_state->entry_method =
956 				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
957 				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
958 		} else if (obj->type == ACPI_TYPE_INTEGER) {
959 			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
960 			lpi_state->address = obj->integer.value;
961 		} else {
962 			continue;
963 		}
964 
965 		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
966 
967 		obj = pkg_elem + 9;
968 		if (obj->type == ACPI_TYPE_STRING)
969 			strscpy(lpi_state->desc, obj->string.pointer,
970 				ACPI_CX_DESC_LEN);
971 
972 		lpi_state->index = state_idx;
973 		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
974 			pr_debug("No min. residency found, assuming 10 us\n");
975 			lpi_state->min_residency = 10;
976 		}
977 
978 		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
979 			pr_debug("No wakeup residency found, assuming 10 us\n");
980 			lpi_state->wake_latency = 10;
981 		}
982 
983 		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
984 			lpi_state->flags = 0;
985 
986 		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
987 			lpi_state->arch_flags = 0;
988 
989 		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
990 			lpi_state->res_cnt_freq = 1;
991 
992 		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
993 			lpi_state->enable_parent_state = 0;
994 	}
995 
996 	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
997 end:
998 	kfree(buffer.pointer);
999 	return ret;
1000 }
1001 
1002 /*
1003  * flat_state_cnt - the number of composite LPI states after the process of flattening
1004  */
1005 static int flat_state_cnt;
1006 
1007 /**
1008  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1009  *
1010  * @local: local LPI state
1011  * @parent: parent LPI state
1012  * @result: composite LPI state
1013  */
1014 static bool combine_lpi_states(struct acpi_lpi_state *local,
1015 			       struct acpi_lpi_state *parent,
1016 			       struct acpi_lpi_state *result)
1017 {
1018 	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1019 		if (!parent->address) /* 0 means autopromotable */
1020 			return false;
1021 		result->address = local->address + parent->address;
1022 	} else {
1023 		result->address = parent->address;
1024 	}
1025 
1026 	result->min_residency = max(local->min_residency, parent->min_residency);
1027 	result->wake_latency = local->wake_latency + parent->wake_latency;
1028 	result->enable_parent_state = parent->enable_parent_state;
1029 	result->entry_method = local->entry_method;
1030 
1031 	result->flags = parent->flags;
1032 	result->arch_flags = parent->arch_flags;
1033 	result->index = parent->index;
1034 
1035 	strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1036 	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1037 	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1038 	return true;
1039 }
1040 
1041 #define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1042 
1043 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1044 				  struct acpi_lpi_state *t)
1045 {
1046 	curr_level->composite_states[curr_level->composite_states_size++] = t;
1047 }
1048 
1049 static int flatten_lpi_states(struct acpi_processor *pr,
1050 			      struct acpi_lpi_states_array *curr_level,
1051 			      struct acpi_lpi_states_array *prev_level)
1052 {
1053 	int i, j, state_count = curr_level->size;
1054 	struct acpi_lpi_state *p, *t = curr_level->entries;
1055 
1056 	curr_level->composite_states_size = 0;
1057 	for (j = 0; j < state_count; j++, t++) {
1058 		struct acpi_lpi_state *flpi;
1059 
1060 		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1061 			continue;
1062 
1063 		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1064 			pr_warn("Limiting number of LPI states to max (%d)\n",
1065 				ACPI_PROCESSOR_MAX_POWER);
1066 			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1067 			break;
1068 		}
1069 
1070 		flpi = &pr->power.lpi_states[flat_state_cnt];
1071 
1072 		if (!prev_level) { /* leaf/processor node */
1073 			memcpy(flpi, t, sizeof(*t));
1074 			stash_composite_state(curr_level, flpi);
1075 			flat_state_cnt++;
1076 			continue;
1077 		}
1078 
1079 		for (i = 0; i < prev_level->composite_states_size; i++) {
1080 			p = prev_level->composite_states[i];
1081 			if (t->index <= p->enable_parent_state &&
1082 			    combine_lpi_states(p, t, flpi)) {
1083 				stash_composite_state(curr_level, flpi);
1084 				flat_state_cnt++;
1085 				flpi++;
1086 			}
1087 		}
1088 	}
1089 
1090 	kfree(curr_level->entries);
1091 	return 0;
1092 }
1093 
1094 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1095 {
1096 	return -EOPNOTSUPP;
1097 }
1098 
1099 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1100 {
1101 	int ret, i;
1102 	acpi_status status;
1103 	acpi_handle handle = pr->handle, pr_ahandle;
1104 	struct acpi_device *d = NULL;
1105 	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1106 
1107 	/* make sure our architecture has support */
1108 	ret = acpi_processor_ffh_lpi_probe(pr->id);
1109 	if (ret == -EOPNOTSUPP)
1110 		return ret;
1111 
1112 	if (!osc_pc_lpi_support_confirmed)
1113 		return -EOPNOTSUPP;
1114 
1115 	if (!acpi_has_method(handle, "_LPI"))
1116 		return -EINVAL;
1117 
1118 	flat_state_cnt = 0;
1119 	prev = &info[0];
1120 	curr = &info[1];
1121 	handle = pr->handle;
1122 	ret = acpi_processor_evaluate_lpi(handle, prev);
1123 	if (ret)
1124 		return ret;
1125 	flatten_lpi_states(pr, prev, NULL);
1126 
1127 	status = acpi_get_parent(handle, &pr_ahandle);
1128 	while (ACPI_SUCCESS(status)) {
1129 		d = acpi_fetch_acpi_dev(pr_ahandle);
1130 		if (!d)
1131 			break;
1132 
1133 		handle = pr_ahandle;
1134 
1135 		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1136 			break;
1137 
1138 		/* can be optional ? */
1139 		if (!acpi_has_method(handle, "_LPI"))
1140 			break;
1141 
1142 		ret = acpi_processor_evaluate_lpi(handle, curr);
1143 		if (ret)
1144 			break;
1145 
1146 		/* flatten all the LPI states in this level of hierarchy */
1147 		flatten_lpi_states(pr, curr, prev);
1148 
1149 		tmp = prev, prev = curr, curr = tmp;
1150 
1151 		status = acpi_get_parent(handle, &pr_ahandle);
1152 	}
1153 
1154 	pr->power.count = flat_state_cnt;
1155 	/* reset the index after flattening */
1156 	for (i = 0; i < pr->power.count; i++)
1157 		pr->power.lpi_states[i].index = i;
1158 
1159 	/* Tell driver that _LPI is supported. */
1160 	pr->flags.has_lpi = 1;
1161 	pr->flags.power = 1;
1162 
1163 	return 0;
1164 }
1165 
1166 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1167 {
1168 	return -ENODEV;
1169 }
1170 
1171 /**
1172  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1173  * @dev: the target CPU
1174  * @drv: cpuidle driver containing cpuidle state info
1175  * @index: index of target state
1176  *
1177  * Return: 0 for success or negative value for error
1178  */
1179 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1180 			       struct cpuidle_driver *drv, int index)
1181 {
1182 	struct acpi_processor *pr;
1183 	struct acpi_lpi_state *lpi;
1184 
1185 	pr = __this_cpu_read(processors);
1186 
1187 	if (unlikely(!pr))
1188 		return -EINVAL;
1189 
1190 	lpi = &pr->power.lpi_states[index];
1191 	if (lpi->entry_method == ACPI_CSTATE_FFH)
1192 		return acpi_processor_ffh_lpi_enter(lpi);
1193 
1194 	return -EINVAL;
1195 }
1196 
1197 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1198 {
1199 	int i;
1200 	struct acpi_lpi_state *lpi;
1201 	struct cpuidle_state *state;
1202 	struct cpuidle_driver *drv = &acpi_idle_driver;
1203 
1204 	if (!pr->flags.has_lpi)
1205 		return -EOPNOTSUPP;
1206 
1207 	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1208 		lpi = &pr->power.lpi_states[i];
1209 
1210 		state = &drv->states[i];
1211 		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1212 		strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1213 		state->exit_latency = lpi->wake_latency;
1214 		state->target_residency = lpi->min_residency;
1215 		state->flags |= arch_get_idle_state_flags(lpi->arch_flags);
1216 		if (i != 0 && lpi->entry_method == ACPI_CSTATE_FFH)
1217 			state->flags |= CPUIDLE_FLAG_RCU_IDLE;
1218 		state->enter = acpi_idle_lpi_enter;
1219 		drv->safe_state_index = i;
1220 	}
1221 
1222 	drv->state_count = i;
1223 
1224 	return 0;
1225 }
1226 
1227 /**
1228  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1229  * global state data i.e. idle routines
1230  *
1231  * @pr: the ACPI processor
1232  */
1233 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1234 {
1235 	int i;
1236 	struct cpuidle_driver *drv = &acpi_idle_driver;
1237 
1238 	if (!pr->flags.power_setup_done || !pr->flags.power)
1239 		return -EINVAL;
1240 
1241 	drv->safe_state_index = -1;
1242 	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1243 		drv->states[i].name[0] = '\0';
1244 		drv->states[i].desc[0] = '\0';
1245 	}
1246 
1247 	if (pr->flags.has_lpi)
1248 		return acpi_processor_setup_lpi_states(pr);
1249 
1250 	return acpi_processor_setup_cstates(pr);
1251 }
1252 
1253 /**
1254  * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1255  * device i.e. per-cpu data
1256  *
1257  * @pr: the ACPI processor
1258  * @dev : the cpuidle device
1259  */
1260 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1261 					    struct cpuidle_device *dev)
1262 {
1263 	if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1264 		return -EINVAL;
1265 
1266 	dev->cpu = pr->id;
1267 	if (pr->flags.has_lpi)
1268 		return acpi_processor_ffh_lpi_probe(pr->id);
1269 
1270 	return acpi_processor_setup_cpuidle_cx(pr, dev);
1271 }
1272 
1273 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1274 {
1275 	int ret;
1276 
1277 	ret = acpi_processor_get_lpi_info(pr);
1278 	if (ret)
1279 		ret = acpi_processor_get_cstate_info(pr);
1280 
1281 	return ret;
1282 }
1283 
1284 int acpi_processor_hotplug(struct acpi_processor *pr)
1285 {
1286 	int ret = 0;
1287 	struct cpuidle_device *dev;
1288 
1289 	if (disabled_by_idle_boot_param())
1290 		return 0;
1291 
1292 	if (!pr->flags.power_setup_done)
1293 		return -ENODEV;
1294 
1295 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1296 	cpuidle_pause_and_lock();
1297 	cpuidle_disable_device(dev);
1298 	ret = acpi_processor_get_power_info(pr);
1299 	if (!ret && pr->flags.power) {
1300 		acpi_processor_setup_cpuidle_dev(pr, dev);
1301 		ret = cpuidle_enable_device(dev);
1302 	}
1303 	cpuidle_resume_and_unlock();
1304 
1305 	return ret;
1306 }
1307 
1308 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1309 {
1310 	int cpu;
1311 	struct acpi_processor *_pr;
1312 	struct cpuidle_device *dev;
1313 
1314 	if (disabled_by_idle_boot_param())
1315 		return 0;
1316 
1317 	if (!pr->flags.power_setup_done)
1318 		return -ENODEV;
1319 
1320 	/*
1321 	 * FIXME:  Design the ACPI notification to make it once per
1322 	 * system instead of once per-cpu.  This condition is a hack
1323 	 * to make the code that updates C-States be called once.
1324 	 */
1325 
1326 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1327 
1328 		/* Protect against cpu-hotplug */
1329 		cpus_read_lock();
1330 		cpuidle_pause_and_lock();
1331 
1332 		/* Disable all cpuidle devices */
1333 		for_each_online_cpu(cpu) {
1334 			_pr = per_cpu(processors, cpu);
1335 			if (!_pr || !_pr->flags.power_setup_done)
1336 				continue;
1337 			dev = per_cpu(acpi_cpuidle_device, cpu);
1338 			cpuidle_disable_device(dev);
1339 		}
1340 
1341 		/* Populate Updated C-state information */
1342 		acpi_processor_get_power_info(pr);
1343 		acpi_processor_setup_cpuidle_states(pr);
1344 
1345 		/* Enable all cpuidle devices */
1346 		for_each_online_cpu(cpu) {
1347 			_pr = per_cpu(processors, cpu);
1348 			if (!_pr || !_pr->flags.power_setup_done)
1349 				continue;
1350 			acpi_processor_get_power_info(_pr);
1351 			if (_pr->flags.power) {
1352 				dev = per_cpu(acpi_cpuidle_device, cpu);
1353 				acpi_processor_setup_cpuidle_dev(_pr, dev);
1354 				cpuidle_enable_device(dev);
1355 			}
1356 		}
1357 		cpuidle_resume_and_unlock();
1358 		cpus_read_unlock();
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 static int acpi_processor_registered;
1365 
1366 int acpi_processor_power_init(struct acpi_processor *pr)
1367 {
1368 	int retval;
1369 	struct cpuidle_device *dev;
1370 
1371 	if (disabled_by_idle_boot_param())
1372 		return 0;
1373 
1374 	acpi_processor_cstate_first_run_checks();
1375 
1376 	if (!acpi_processor_get_power_info(pr))
1377 		pr->flags.power_setup_done = 1;
1378 
1379 	/*
1380 	 * Install the idle handler if processor power management is supported.
1381 	 * Note that we use previously set idle handler will be used on
1382 	 * platforms that only support C1.
1383 	 */
1384 	if (pr->flags.power) {
1385 		/* Register acpi_idle_driver if not already registered */
1386 		if (!acpi_processor_registered) {
1387 			acpi_processor_setup_cpuidle_states(pr);
1388 			retval = cpuidle_register_driver(&acpi_idle_driver);
1389 			if (retval)
1390 				return retval;
1391 			pr_debug("%s registered with cpuidle\n",
1392 				 acpi_idle_driver.name);
1393 		}
1394 
1395 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1396 		if (!dev)
1397 			return -ENOMEM;
1398 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1399 
1400 		acpi_processor_setup_cpuidle_dev(pr, dev);
1401 
1402 		/* Register per-cpu cpuidle_device. Cpuidle driver
1403 		 * must already be registered before registering device
1404 		 */
1405 		retval = cpuidle_register_device(dev);
1406 		if (retval) {
1407 			if (acpi_processor_registered == 0)
1408 				cpuidle_unregister_driver(&acpi_idle_driver);
1409 			return retval;
1410 		}
1411 		acpi_processor_registered++;
1412 	}
1413 	return 0;
1414 }
1415 
1416 int acpi_processor_power_exit(struct acpi_processor *pr)
1417 {
1418 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1419 
1420 	if (disabled_by_idle_boot_param())
1421 		return 0;
1422 
1423 	if (pr->flags.power) {
1424 		cpuidle_unregister_device(dev);
1425 		acpi_processor_registered--;
1426 		if (acpi_processor_registered == 0)
1427 			cpuidle_unregister_driver(&acpi_idle_driver);
1428 
1429 		kfree(dev);
1430 	}
1431 
1432 	pr->flags.power_setup_done = 0;
1433 	return 0;
1434 }
1435