xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision dff03381)
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 
41 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
42 module_param(max_cstate, uint, 0400);
43 static bool nocst __read_mostly;
44 module_param(nocst, bool, 0400);
45 static bool bm_check_disable __read_mostly;
46 module_param(bm_check_disable, bool, 0400);
47 
48 static unsigned int latency_factor __read_mostly = 2;
49 module_param(latency_factor, uint, 0644);
50 
51 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
52 
53 struct cpuidle_driver acpi_idle_driver = {
54 	.name =		"acpi_idle",
55 	.owner =	THIS_MODULE,
56 };
57 
58 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
59 static
60 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
61 
62 static 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  */
74 static 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 
88 static 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  */
109 static void __cpuidle acpi_safe_halt(void)
110 {
111 	if (!tif_need_resched()) {
112 		safe_halt();
113 		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  */
125 static 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 
148 static void __lapic_timer_propagate_broadcast(void *arg)
149 {
150 	struct acpi_processor *pr = (struct acpi_processor *) arg;
151 
152 	if (pr->power.timer_broadcast_on_state < INT_MAX)
153 		tick_broadcast_enable();
154 	else
155 		tick_broadcast_disable();
156 }
157 
158 static 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 */
165 static 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 
173 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
174 				   struct acpi_processor_cx *cstate) { }
175 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
176 
177 static 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)
186 static 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
208 static void tsc_check_state(int state) { return; }
209 #endif
210 
211 static 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 
275 static 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 
292 static 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 
310 static 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 	else 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 	return;
389 }
390 
391 static int acpi_cst_latency_cmp(const void *a, const void *b)
392 {
393 	const struct acpi_processor_cx *x = a, *y = b;
394 
395 	if (!(x->valid && y->valid))
396 		return 0;
397 	if (x->latency > y->latency)
398 		return 1;
399 	if (x->latency < y->latency)
400 		return -1;
401 	return 0;
402 }
403 static void acpi_cst_latency_swap(void *a, void *b, int n)
404 {
405 	struct acpi_processor_cx *x = a, *y = b;
406 
407 	if (!(x->valid && y->valid))
408 		return;
409 	swap(x->latency, y->latency);
410 }
411 
412 static int acpi_processor_power_verify(struct acpi_processor *pr)
413 {
414 	unsigned int i;
415 	unsigned int working = 0;
416 	unsigned int last_latency = 0;
417 	unsigned int last_type = 0;
418 	bool buggy_latency = false;
419 
420 	pr->power.timer_broadcast_on_state = INT_MAX;
421 
422 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
423 		struct acpi_processor_cx *cx = &pr->power.states[i];
424 
425 		switch (cx->type) {
426 		case ACPI_STATE_C1:
427 			cx->valid = 1;
428 			break;
429 
430 		case ACPI_STATE_C2:
431 			if (!cx->address)
432 				break;
433 			cx->valid = 1;
434 			break;
435 
436 		case ACPI_STATE_C3:
437 			acpi_processor_power_verify_c3(pr, cx);
438 			break;
439 		}
440 		if (!cx->valid)
441 			continue;
442 		if (cx->type >= last_type && cx->latency < last_latency)
443 			buggy_latency = true;
444 		last_latency = cx->latency;
445 		last_type = cx->type;
446 
447 		lapic_timer_check_state(i, pr, cx);
448 		tsc_check_state(cx->type);
449 		working++;
450 	}
451 
452 	if (buggy_latency) {
453 		pr_notice("FW issue: working around C-state latencies out of order\n");
454 		sort(&pr->power.states[1], max_cstate,
455 		     sizeof(struct acpi_processor_cx),
456 		     acpi_cst_latency_cmp,
457 		     acpi_cst_latency_swap);
458 	}
459 
460 	lapic_timer_propagate_broadcast(pr);
461 
462 	return (working);
463 }
464 
465 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
466 {
467 	unsigned int i;
468 	int result;
469 
470 
471 	/* NOTE: the idle thread may not be running while calling
472 	 * this function */
473 
474 	/* Zero initialize all the C-states info. */
475 	memset(pr->power.states, 0, sizeof(pr->power.states));
476 
477 	result = acpi_processor_get_power_info_cst(pr);
478 	if (result == -ENODEV)
479 		result = acpi_processor_get_power_info_fadt(pr);
480 
481 	if (result)
482 		return result;
483 
484 	acpi_processor_get_power_info_default(pr);
485 
486 	pr->power.count = acpi_processor_power_verify(pr);
487 
488 	/*
489 	 * if one state of type C2 or C3 is available, mark this
490 	 * CPU as being "idle manageable"
491 	 */
492 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
493 		if (pr->power.states[i].valid) {
494 			pr->power.count = i;
495 			pr->flags.power = 1;
496 		}
497 	}
498 
499 	return 0;
500 }
501 
502 /**
503  * acpi_idle_bm_check - checks if bus master activity was detected
504  */
505 static int acpi_idle_bm_check(void)
506 {
507 	u32 bm_status = 0;
508 
509 	if (bm_check_disable)
510 		return 0;
511 
512 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
513 	if (bm_status)
514 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
515 	/*
516 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
517 	 * the true state of bus mastering activity; forcing us to
518 	 * manually check the BMIDEA bit of each IDE channel.
519 	 */
520 	else if (errata.piix4.bmisx) {
521 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
522 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
523 			bm_status = 1;
524 	}
525 	return bm_status;
526 }
527 
528 static void wait_for_freeze(void)
529 {
530 #ifdef	CONFIG_X86
531 	/* No delay is needed if we are in guest */
532 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
533 		return;
534 #endif
535 	/* Dummy wait op - must do something useless after P_LVL2 read
536 	   because chipsets cannot guarantee that STPCLK# signal
537 	   gets asserted in time to freeze execution properly. */
538 	inl(acpi_gbl_FADT.xpm_timer_block.address);
539 }
540 
541 /**
542  * acpi_idle_do_entry - enter idle state using the appropriate method
543  * @cx: cstate data
544  *
545  * Caller disables interrupt before call and enables interrupt after return.
546  */
547 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
548 {
549 	perf_lopwr_cb(true);
550 
551 	if (cx->entry_method == ACPI_CSTATE_FFH) {
552 		/* Call into architectural FFH based C-state */
553 		acpi_processor_ffh_cstate_enter(cx);
554 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
555 		acpi_safe_halt();
556 	} else {
557 		/* IO port based C-state */
558 		inb(cx->address);
559 		wait_for_freeze();
560 	}
561 
562 	perf_lopwr_cb(false);
563 }
564 
565 /**
566  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
567  * @dev: the target CPU
568  * @index: the index of suggested state
569  */
570 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
571 {
572 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
573 
574 	ACPI_FLUSH_CPU_CACHE();
575 
576 	while (1) {
577 
578 		if (cx->entry_method == ACPI_CSTATE_HALT)
579 			safe_halt();
580 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
581 			inb(cx->address);
582 			wait_for_freeze();
583 		} else
584 			return -ENODEV;
585 
586 #if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU)
587 		cond_wakeup_cpu0();
588 #endif
589 	}
590 
591 	/* Never reached */
592 	return 0;
593 }
594 
595 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
596 {
597 	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
598 		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
599 }
600 
601 static int c3_cpu_count;
602 static DEFINE_RAW_SPINLOCK(c3_lock);
603 
604 /**
605  * acpi_idle_enter_bm - enters C3 with proper BM handling
606  * @drv: cpuidle driver
607  * @pr: Target processor
608  * @cx: Target state context
609  * @index: index of target state
610  */
611 static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
612 			       struct acpi_processor *pr,
613 			       struct acpi_processor_cx *cx,
614 			       int index)
615 {
616 	static struct acpi_processor_cx safe_cx = {
617 		.entry_method = ACPI_CSTATE_HALT,
618 	};
619 
620 	/*
621 	 * disable bus master
622 	 * bm_check implies we need ARB_DIS
623 	 * bm_control implies whether we can do ARB_DIS
624 	 *
625 	 * That leaves a case where bm_check is set and bm_control is not set.
626 	 * In that case we cannot do much, we enter C3 without doing anything.
627 	 */
628 	bool dis_bm = pr->flags.bm_control;
629 
630 	/* If we can skip BM, demote to a safe state. */
631 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
632 		dis_bm = false;
633 		index = drv->safe_state_index;
634 		if (index >= 0) {
635 			cx = this_cpu_read(acpi_cstate[index]);
636 		} else {
637 			cx = &safe_cx;
638 			index = -EBUSY;
639 		}
640 	}
641 
642 	if (dis_bm) {
643 		raw_spin_lock(&c3_lock);
644 		c3_cpu_count++;
645 		/* Disable bus master arbitration when all CPUs are in C3 */
646 		if (c3_cpu_count == num_online_cpus())
647 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
648 		raw_spin_unlock(&c3_lock);
649 	}
650 
651 	ct_idle_enter();
652 
653 	acpi_idle_do_entry(cx);
654 
655 	ct_idle_exit();
656 
657 	/* Re-enable bus master arbitration */
658 	if (dis_bm) {
659 		raw_spin_lock(&c3_lock);
660 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
661 		c3_cpu_count--;
662 		raw_spin_unlock(&c3_lock);
663 	}
664 
665 	return index;
666 }
667 
668 static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
669 			   struct cpuidle_driver *drv, int index)
670 {
671 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
672 	struct acpi_processor *pr;
673 
674 	pr = __this_cpu_read(processors);
675 	if (unlikely(!pr))
676 		return -EINVAL;
677 
678 	if (cx->type != ACPI_STATE_C1) {
679 		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
680 			return acpi_idle_enter_bm(drv, pr, cx, index);
681 
682 		/* C2 to C1 demotion. */
683 		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
684 			index = ACPI_IDLE_STATE_START;
685 			cx = per_cpu(acpi_cstate[index], dev->cpu);
686 		}
687 	}
688 
689 	if (cx->type == ACPI_STATE_C3)
690 		ACPI_FLUSH_CPU_CACHE();
691 
692 	acpi_idle_do_entry(cx);
693 
694 	return index;
695 }
696 
697 static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
698 				  struct cpuidle_driver *drv, int index)
699 {
700 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
701 
702 	if (cx->type == ACPI_STATE_C3) {
703 		struct acpi_processor *pr = __this_cpu_read(processors);
704 
705 		if (unlikely(!pr))
706 			return 0;
707 
708 		if (pr->flags.bm_check) {
709 			u8 bm_sts_skip = cx->bm_sts_skip;
710 
711 			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
712 			cx->bm_sts_skip = 1;
713 			acpi_idle_enter_bm(drv, pr, cx, index);
714 			cx->bm_sts_skip = bm_sts_skip;
715 
716 			return 0;
717 		} else {
718 			ACPI_FLUSH_CPU_CACHE();
719 		}
720 	}
721 	acpi_idle_do_entry(cx);
722 
723 	return 0;
724 }
725 
726 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
727 					   struct cpuidle_device *dev)
728 {
729 	int i, count = ACPI_IDLE_STATE_START;
730 	struct acpi_processor_cx *cx;
731 	struct cpuidle_state *state;
732 
733 	if (max_cstate == 0)
734 		max_cstate = 1;
735 
736 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
737 		state = &acpi_idle_driver.states[count];
738 		cx = &pr->power.states[i];
739 
740 		if (!cx->valid)
741 			continue;
742 
743 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
744 
745 		if (lapic_timer_needs_broadcast(pr, cx))
746 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
747 
748 		if (cx->type == ACPI_STATE_C3) {
749 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
750 			if (pr->flags.bm_check)
751 				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
752 		}
753 
754 		count++;
755 		if (count == CPUIDLE_STATE_MAX)
756 			break;
757 	}
758 
759 	if (!count)
760 		return -EINVAL;
761 
762 	return 0;
763 }
764 
765 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
766 {
767 	int i, count;
768 	struct acpi_processor_cx *cx;
769 	struct cpuidle_state *state;
770 	struct cpuidle_driver *drv = &acpi_idle_driver;
771 
772 	if (max_cstate == 0)
773 		max_cstate = 1;
774 
775 	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
776 		cpuidle_poll_state_init(drv);
777 		count = 1;
778 	} else {
779 		count = 0;
780 	}
781 
782 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
783 		cx = &pr->power.states[i];
784 
785 		if (!cx->valid)
786 			continue;
787 
788 		state = &drv->states[count];
789 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
790 		strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
791 		state->exit_latency = cx->latency;
792 		state->target_residency = cx->latency * latency_factor;
793 		state->enter = acpi_idle_enter;
794 
795 		state->flags = 0;
796 		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
797 		    cx->type == ACPI_STATE_C3) {
798 			state->enter_dead = acpi_idle_play_dead;
799 			if (cx->type != ACPI_STATE_C3)
800 				drv->safe_state_index = count;
801 		}
802 		/*
803 		 * Halt-induced C1 is not good for ->enter_s2idle, because it
804 		 * re-enables interrupts on exit.  Moreover, C1 is generally not
805 		 * particularly interesting from the suspend-to-idle angle, so
806 		 * avoid C1 and the situations in which we may need to fall back
807 		 * to it altogether.
808 		 */
809 		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
810 			state->enter_s2idle = acpi_idle_enter_s2idle;
811 
812 		count++;
813 		if (count == CPUIDLE_STATE_MAX)
814 			break;
815 	}
816 
817 	drv->state_count = count;
818 
819 	if (!count)
820 		return -EINVAL;
821 
822 	return 0;
823 }
824 
825 static inline void acpi_processor_cstate_first_run_checks(void)
826 {
827 	static int first_run;
828 
829 	if (first_run)
830 		return;
831 	dmi_check_system(processor_power_dmi_table);
832 	max_cstate = acpi_processor_cstate_check(max_cstate);
833 	if (max_cstate < ACPI_C_STATES_MAX)
834 		pr_notice("processor limited to max C-state %d\n", max_cstate);
835 
836 	first_run++;
837 
838 	if (nocst)
839 		return;
840 
841 	acpi_processor_claim_cst_control();
842 }
843 #else
844 
845 static inline int disabled_by_idle_boot_param(void) { return 0; }
846 static inline void acpi_processor_cstate_first_run_checks(void) { }
847 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
848 {
849 	return -ENODEV;
850 }
851 
852 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
853 					   struct cpuidle_device *dev)
854 {
855 	return -EINVAL;
856 }
857 
858 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
859 {
860 	return -EINVAL;
861 }
862 
863 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
864 
865 struct acpi_lpi_states_array {
866 	unsigned int size;
867 	unsigned int composite_states_size;
868 	struct acpi_lpi_state *entries;
869 	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
870 };
871 
872 static int obj_get_integer(union acpi_object *obj, u32 *value)
873 {
874 	if (obj->type != ACPI_TYPE_INTEGER)
875 		return -EINVAL;
876 
877 	*value = obj->integer.value;
878 	return 0;
879 }
880 
881 static int acpi_processor_evaluate_lpi(acpi_handle handle,
882 				       struct acpi_lpi_states_array *info)
883 {
884 	acpi_status status;
885 	int ret = 0;
886 	int pkg_count, state_idx = 1, loop;
887 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
888 	union acpi_object *lpi_data;
889 	struct acpi_lpi_state *lpi_state;
890 
891 	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
892 	if (ACPI_FAILURE(status)) {
893 		acpi_handle_debug(handle, "No _LPI, giving up\n");
894 		return -ENODEV;
895 	}
896 
897 	lpi_data = buffer.pointer;
898 
899 	/* There must be at least 4 elements = 3 elements + 1 package */
900 	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
901 	    lpi_data->package.count < 4) {
902 		pr_debug("not enough elements in _LPI\n");
903 		ret = -ENODATA;
904 		goto end;
905 	}
906 
907 	pkg_count = lpi_data->package.elements[2].integer.value;
908 
909 	/* Validate number of power states. */
910 	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
911 		pr_debug("count given by _LPI is not valid\n");
912 		ret = -ENODATA;
913 		goto end;
914 	}
915 
916 	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
917 	if (!lpi_state) {
918 		ret = -ENOMEM;
919 		goto end;
920 	}
921 
922 	info->size = pkg_count;
923 	info->entries = lpi_state;
924 
925 	/* LPI States start at index 3 */
926 	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
927 		union acpi_object *element, *pkg_elem, *obj;
928 
929 		element = &lpi_data->package.elements[loop];
930 		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
931 			continue;
932 
933 		pkg_elem = element->package.elements;
934 
935 		obj = pkg_elem + 6;
936 		if (obj->type == ACPI_TYPE_BUFFER) {
937 			struct acpi_power_register *reg;
938 
939 			reg = (struct acpi_power_register *)obj->buffer.pointer;
940 			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
941 			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
942 				continue;
943 
944 			lpi_state->address = reg->address;
945 			lpi_state->entry_method =
946 				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
947 				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
948 		} else if (obj->type == ACPI_TYPE_INTEGER) {
949 			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
950 			lpi_state->address = obj->integer.value;
951 		} else {
952 			continue;
953 		}
954 
955 		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
956 
957 		obj = pkg_elem + 9;
958 		if (obj->type == ACPI_TYPE_STRING)
959 			strlcpy(lpi_state->desc, obj->string.pointer,
960 				ACPI_CX_DESC_LEN);
961 
962 		lpi_state->index = state_idx;
963 		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
964 			pr_debug("No min. residency found, assuming 10 us\n");
965 			lpi_state->min_residency = 10;
966 		}
967 
968 		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
969 			pr_debug("No wakeup residency found, assuming 10 us\n");
970 			lpi_state->wake_latency = 10;
971 		}
972 
973 		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
974 			lpi_state->flags = 0;
975 
976 		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
977 			lpi_state->arch_flags = 0;
978 
979 		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
980 			lpi_state->res_cnt_freq = 1;
981 
982 		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
983 			lpi_state->enable_parent_state = 0;
984 	}
985 
986 	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
987 end:
988 	kfree(buffer.pointer);
989 	return ret;
990 }
991 
992 /*
993  * flat_state_cnt - the number of composite LPI states after the process of flattening
994  */
995 static int flat_state_cnt;
996 
997 /**
998  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
999  *
1000  * @local: local LPI state
1001  * @parent: parent LPI state
1002  * @result: composite LPI state
1003  */
1004 static bool combine_lpi_states(struct acpi_lpi_state *local,
1005 			       struct acpi_lpi_state *parent,
1006 			       struct acpi_lpi_state *result)
1007 {
1008 	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1009 		if (!parent->address) /* 0 means autopromotable */
1010 			return false;
1011 		result->address = local->address + parent->address;
1012 	} else {
1013 		result->address = parent->address;
1014 	}
1015 
1016 	result->min_residency = max(local->min_residency, parent->min_residency);
1017 	result->wake_latency = local->wake_latency + parent->wake_latency;
1018 	result->enable_parent_state = parent->enable_parent_state;
1019 	result->entry_method = local->entry_method;
1020 
1021 	result->flags = parent->flags;
1022 	result->arch_flags = parent->arch_flags;
1023 	result->index = parent->index;
1024 
1025 	strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1026 	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1027 	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1028 	return true;
1029 }
1030 
1031 #define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1032 
1033 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1034 				  struct acpi_lpi_state *t)
1035 {
1036 	curr_level->composite_states[curr_level->composite_states_size++] = t;
1037 }
1038 
1039 static int flatten_lpi_states(struct acpi_processor *pr,
1040 			      struct acpi_lpi_states_array *curr_level,
1041 			      struct acpi_lpi_states_array *prev_level)
1042 {
1043 	int i, j, state_count = curr_level->size;
1044 	struct acpi_lpi_state *p, *t = curr_level->entries;
1045 
1046 	curr_level->composite_states_size = 0;
1047 	for (j = 0; j < state_count; j++, t++) {
1048 		struct acpi_lpi_state *flpi;
1049 
1050 		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1051 			continue;
1052 
1053 		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1054 			pr_warn("Limiting number of LPI states to max (%d)\n",
1055 				ACPI_PROCESSOR_MAX_POWER);
1056 			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1057 			break;
1058 		}
1059 
1060 		flpi = &pr->power.lpi_states[flat_state_cnt];
1061 
1062 		if (!prev_level) { /* leaf/processor node */
1063 			memcpy(flpi, t, sizeof(*t));
1064 			stash_composite_state(curr_level, flpi);
1065 			flat_state_cnt++;
1066 			continue;
1067 		}
1068 
1069 		for (i = 0; i < prev_level->composite_states_size; i++) {
1070 			p = prev_level->composite_states[i];
1071 			if (t->index <= p->enable_parent_state &&
1072 			    combine_lpi_states(p, t, flpi)) {
1073 				stash_composite_state(curr_level, flpi);
1074 				flat_state_cnt++;
1075 				flpi++;
1076 			}
1077 		}
1078 	}
1079 
1080 	kfree(curr_level->entries);
1081 	return 0;
1082 }
1083 
1084 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1085 {
1086 	return -EOPNOTSUPP;
1087 }
1088 
1089 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1090 {
1091 	int ret, i;
1092 	acpi_status status;
1093 	acpi_handle handle = pr->handle, pr_ahandle;
1094 	struct acpi_device *d = NULL;
1095 	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1096 
1097 	/* make sure our architecture has support */
1098 	ret = acpi_processor_ffh_lpi_probe(pr->id);
1099 	if (ret == -EOPNOTSUPP)
1100 		return ret;
1101 
1102 	if (!osc_pc_lpi_support_confirmed)
1103 		return -EOPNOTSUPP;
1104 
1105 	if (!acpi_has_method(handle, "_LPI"))
1106 		return -EINVAL;
1107 
1108 	flat_state_cnt = 0;
1109 	prev = &info[0];
1110 	curr = &info[1];
1111 	handle = pr->handle;
1112 	ret = acpi_processor_evaluate_lpi(handle, prev);
1113 	if (ret)
1114 		return ret;
1115 	flatten_lpi_states(pr, prev, NULL);
1116 
1117 	status = acpi_get_parent(handle, &pr_ahandle);
1118 	while (ACPI_SUCCESS(status)) {
1119 		d = acpi_fetch_acpi_dev(pr_ahandle);
1120 		handle = pr_ahandle;
1121 
1122 		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1123 			break;
1124 
1125 		/* can be optional ? */
1126 		if (!acpi_has_method(handle, "_LPI"))
1127 			break;
1128 
1129 		ret = acpi_processor_evaluate_lpi(handle, curr);
1130 		if (ret)
1131 			break;
1132 
1133 		/* flatten all the LPI states in this level of hierarchy */
1134 		flatten_lpi_states(pr, curr, prev);
1135 
1136 		tmp = prev, prev = curr, curr = tmp;
1137 
1138 		status = acpi_get_parent(handle, &pr_ahandle);
1139 	}
1140 
1141 	pr->power.count = flat_state_cnt;
1142 	/* reset the index after flattening */
1143 	for (i = 0; i < pr->power.count; i++)
1144 		pr->power.lpi_states[i].index = i;
1145 
1146 	/* Tell driver that _LPI is supported. */
1147 	pr->flags.has_lpi = 1;
1148 	pr->flags.power = 1;
1149 
1150 	return 0;
1151 }
1152 
1153 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1154 {
1155 	return -ENODEV;
1156 }
1157 
1158 /**
1159  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1160  * @dev: the target CPU
1161  * @drv: cpuidle driver containing cpuidle state info
1162  * @index: index of target state
1163  *
1164  * Return: 0 for success or negative value for error
1165  */
1166 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1167 			       struct cpuidle_driver *drv, int index)
1168 {
1169 	struct acpi_processor *pr;
1170 	struct acpi_lpi_state *lpi;
1171 
1172 	pr = __this_cpu_read(processors);
1173 
1174 	if (unlikely(!pr))
1175 		return -EINVAL;
1176 
1177 	lpi = &pr->power.lpi_states[index];
1178 	if (lpi->entry_method == ACPI_CSTATE_FFH)
1179 		return acpi_processor_ffh_lpi_enter(lpi);
1180 
1181 	return -EINVAL;
1182 }
1183 
1184 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1185 {
1186 	int i;
1187 	struct acpi_lpi_state *lpi;
1188 	struct cpuidle_state *state;
1189 	struct cpuidle_driver *drv = &acpi_idle_driver;
1190 
1191 	if (!pr->flags.has_lpi)
1192 		return -EOPNOTSUPP;
1193 
1194 	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1195 		lpi = &pr->power.lpi_states[i];
1196 
1197 		state = &drv->states[i];
1198 		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1199 		strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1200 		state->exit_latency = lpi->wake_latency;
1201 		state->target_residency = lpi->min_residency;
1202 		if (lpi->arch_flags)
1203 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1204 		state->enter = acpi_idle_lpi_enter;
1205 		drv->safe_state_index = i;
1206 	}
1207 
1208 	drv->state_count = i;
1209 
1210 	return 0;
1211 }
1212 
1213 /**
1214  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1215  * global state data i.e. idle routines
1216  *
1217  * @pr: the ACPI processor
1218  */
1219 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1220 {
1221 	int i;
1222 	struct cpuidle_driver *drv = &acpi_idle_driver;
1223 
1224 	if (!pr->flags.power_setup_done || !pr->flags.power)
1225 		return -EINVAL;
1226 
1227 	drv->safe_state_index = -1;
1228 	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1229 		drv->states[i].name[0] = '\0';
1230 		drv->states[i].desc[0] = '\0';
1231 	}
1232 
1233 	if (pr->flags.has_lpi)
1234 		return acpi_processor_setup_lpi_states(pr);
1235 
1236 	return acpi_processor_setup_cstates(pr);
1237 }
1238 
1239 /**
1240  * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1241  * device i.e. per-cpu data
1242  *
1243  * @pr: the ACPI processor
1244  * @dev : the cpuidle device
1245  */
1246 static int 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 -EINVAL;
1251 
1252 	dev->cpu = pr->id;
1253 	if (pr->flags.has_lpi)
1254 		return acpi_processor_ffh_lpi_probe(pr->id);
1255 
1256 	return acpi_processor_setup_cpuidle_cx(pr, dev);
1257 }
1258 
1259 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1260 {
1261 	int ret;
1262 
1263 	ret = acpi_processor_get_lpi_info(pr);
1264 	if (ret)
1265 		ret = acpi_processor_get_cstate_info(pr);
1266 
1267 	return ret;
1268 }
1269 
1270 int acpi_processor_hotplug(struct acpi_processor *pr)
1271 {
1272 	int ret = 0;
1273 	struct cpuidle_device *dev;
1274 
1275 	if (disabled_by_idle_boot_param())
1276 		return 0;
1277 
1278 	if (!pr->flags.power_setup_done)
1279 		return -ENODEV;
1280 
1281 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1282 	cpuidle_pause_and_lock();
1283 	cpuidle_disable_device(dev);
1284 	ret = acpi_processor_get_power_info(pr);
1285 	if (!ret && pr->flags.power) {
1286 		acpi_processor_setup_cpuidle_dev(pr, dev);
1287 		ret = cpuidle_enable_device(dev);
1288 	}
1289 	cpuidle_resume_and_unlock();
1290 
1291 	return ret;
1292 }
1293 
1294 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1295 {
1296 	int cpu;
1297 	struct acpi_processor *_pr;
1298 	struct cpuidle_device *dev;
1299 
1300 	if (disabled_by_idle_boot_param())
1301 		return 0;
1302 
1303 	if (!pr->flags.power_setup_done)
1304 		return -ENODEV;
1305 
1306 	/*
1307 	 * FIXME:  Design the ACPI notification to make it once per
1308 	 * system instead of once per-cpu.  This condition is a hack
1309 	 * to make the code that updates C-States be called once.
1310 	 */
1311 
1312 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1313 
1314 		/* Protect against cpu-hotplug */
1315 		cpus_read_lock();
1316 		cpuidle_pause_and_lock();
1317 
1318 		/* Disable all cpuidle devices */
1319 		for_each_online_cpu(cpu) {
1320 			_pr = per_cpu(processors, cpu);
1321 			if (!_pr || !_pr->flags.power_setup_done)
1322 				continue;
1323 			dev = per_cpu(acpi_cpuidle_device, cpu);
1324 			cpuidle_disable_device(dev);
1325 		}
1326 
1327 		/* Populate Updated C-state information */
1328 		acpi_processor_get_power_info(pr);
1329 		acpi_processor_setup_cpuidle_states(pr);
1330 
1331 		/* Enable all cpuidle devices */
1332 		for_each_online_cpu(cpu) {
1333 			_pr = per_cpu(processors, cpu);
1334 			if (!_pr || !_pr->flags.power_setup_done)
1335 				continue;
1336 			acpi_processor_get_power_info(_pr);
1337 			if (_pr->flags.power) {
1338 				dev = per_cpu(acpi_cpuidle_device, cpu);
1339 				acpi_processor_setup_cpuidle_dev(_pr, dev);
1340 				cpuidle_enable_device(dev);
1341 			}
1342 		}
1343 		cpuidle_resume_and_unlock();
1344 		cpus_read_unlock();
1345 	}
1346 
1347 	return 0;
1348 }
1349 
1350 static int acpi_processor_registered;
1351 
1352 int acpi_processor_power_init(struct acpi_processor *pr)
1353 {
1354 	int retval;
1355 	struct cpuidle_device *dev;
1356 
1357 	if (disabled_by_idle_boot_param())
1358 		return 0;
1359 
1360 	acpi_processor_cstate_first_run_checks();
1361 
1362 	if (!acpi_processor_get_power_info(pr))
1363 		pr->flags.power_setup_done = 1;
1364 
1365 	/*
1366 	 * Install the idle handler if processor power management is supported.
1367 	 * Note that we use previously set idle handler will be used on
1368 	 * platforms that only support C1.
1369 	 */
1370 	if (pr->flags.power) {
1371 		/* Register acpi_idle_driver if not already registered */
1372 		if (!acpi_processor_registered) {
1373 			acpi_processor_setup_cpuidle_states(pr);
1374 			retval = cpuidle_register_driver(&acpi_idle_driver);
1375 			if (retval)
1376 				return retval;
1377 			pr_debug("%s registered with cpuidle\n",
1378 				 acpi_idle_driver.name);
1379 		}
1380 
1381 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1382 		if (!dev)
1383 			return -ENOMEM;
1384 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1385 
1386 		acpi_processor_setup_cpuidle_dev(pr, dev);
1387 
1388 		/* Register per-cpu cpuidle_device. Cpuidle driver
1389 		 * must already be registered before registering device
1390 		 */
1391 		retval = cpuidle_register_device(dev);
1392 		if (retval) {
1393 			if (acpi_processor_registered == 0)
1394 				cpuidle_unregister_driver(&acpi_idle_driver);
1395 			return retval;
1396 		}
1397 		acpi_processor_registered++;
1398 	}
1399 	return 0;
1400 }
1401 
1402 int acpi_processor_power_exit(struct acpi_processor *pr)
1403 {
1404 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1405 
1406 	if (disabled_by_idle_boot_param())
1407 		return 0;
1408 
1409 	if (pr->flags.power) {
1410 		cpuidle_unregister_device(dev);
1411 		acpi_processor_registered--;
1412 		if (acpi_processor_registered == 0)
1413 			cpuidle_unregister_driver(&acpi_idle_driver);
1414 	}
1415 
1416 	pr->flags.power_setup_done = 0;
1417 	return 0;
1418 }
1419