xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision 76afff43)
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 		raw_safe_halt();
113 		raw_local_irq_disable();
114 	}
115 }
116 
117 #ifdef ARCH_APICTIMER_STOPS_ON_C3
118 
119 /*
120  * Some BIOS implementations switch to C3 in the published C2 state.
121  * This seems to be a common problem on AMD boxen, but other vendors
122  * are affected too. We pick the most conservative approach: we assume
123  * that the local APIC stops in both C2 and C3.
124  */
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 = 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 	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 
389 static int acpi_cst_latency_cmp(const void *a, const void *b)
390 {
391 	const struct acpi_processor_cx *x = a, *y = b;
392 
393 	if (!(x->valid && y->valid))
394 		return 0;
395 	if (x->latency > y->latency)
396 		return 1;
397 	if (x->latency < y->latency)
398 		return -1;
399 	return 0;
400 }
401 static void acpi_cst_latency_swap(void *a, void *b, int n)
402 {
403 	struct acpi_processor_cx *x = a, *y = b;
404 
405 	if (!(x->valid && y->valid))
406 		return;
407 	swap(x->latency, y->latency);
408 }
409 
410 static int acpi_processor_power_verify(struct acpi_processor *pr)
411 {
412 	unsigned int i;
413 	unsigned int working = 0;
414 	unsigned int last_latency = 0;
415 	unsigned int last_type = 0;
416 	bool buggy_latency = false;
417 
418 	pr->power.timer_broadcast_on_state = INT_MAX;
419 
420 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
421 		struct acpi_processor_cx *cx = &pr->power.states[i];
422 
423 		switch (cx->type) {
424 		case ACPI_STATE_C1:
425 			cx->valid = 1;
426 			break;
427 
428 		case ACPI_STATE_C2:
429 			if (!cx->address)
430 				break;
431 			cx->valid = 1;
432 			break;
433 
434 		case ACPI_STATE_C3:
435 			acpi_processor_power_verify_c3(pr, cx);
436 			break;
437 		}
438 		if (!cx->valid)
439 			continue;
440 		if (cx->type >= last_type && cx->latency < last_latency)
441 			buggy_latency = true;
442 		last_latency = cx->latency;
443 		last_type = cx->type;
444 
445 		lapic_timer_check_state(i, pr, cx);
446 		tsc_check_state(cx->type);
447 		working++;
448 	}
449 
450 	if (buggy_latency) {
451 		pr_notice("FW issue: working around C-state latencies out of order\n");
452 		sort(&pr->power.states[1], max_cstate,
453 		     sizeof(struct acpi_processor_cx),
454 		     acpi_cst_latency_cmp,
455 		     acpi_cst_latency_swap);
456 	}
457 
458 	lapic_timer_propagate_broadcast(pr);
459 
460 	return working;
461 }
462 
463 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
464 {
465 	unsigned int i;
466 	int result;
467 
468 
469 	/* NOTE: the idle thread may not be running while calling
470 	 * this function */
471 
472 	/* Zero initialize all the C-states info. */
473 	memset(pr->power.states, 0, sizeof(pr->power.states));
474 
475 	result = acpi_processor_get_power_info_cst(pr);
476 	if (result == -ENODEV)
477 		result = acpi_processor_get_power_info_fadt(pr);
478 
479 	if (result)
480 		return result;
481 
482 	acpi_processor_get_power_info_default(pr);
483 
484 	pr->power.count = acpi_processor_power_verify(pr);
485 
486 	/*
487 	 * if one state of type C2 or C3 is available, mark this
488 	 * CPU as being "idle manageable"
489 	 */
490 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
491 		if (pr->power.states[i].valid) {
492 			pr->power.count = i;
493 			pr->flags.power = 1;
494 		}
495 	}
496 
497 	return 0;
498 }
499 
500 /**
501  * acpi_idle_bm_check - checks if bus master activity was detected
502  */
503 static int acpi_idle_bm_check(void)
504 {
505 	u32 bm_status = 0;
506 
507 	if (bm_check_disable)
508 		return 0;
509 
510 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
511 	if (bm_status)
512 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
513 	/*
514 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
515 	 * the true state of bus mastering activity; forcing us to
516 	 * manually check the BMIDEA bit of each IDE channel.
517 	 */
518 	else if (errata.piix4.bmisx) {
519 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
520 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
521 			bm_status = 1;
522 	}
523 	return bm_status;
524 }
525 
526 static __cpuidle void io_idle(unsigned long addr)
527 {
528 	/* IO port based C-state */
529 	inb(addr);
530 
531 #ifdef	CONFIG_X86
532 	/* No delay is needed if we are in guest */
533 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
534 		return;
535 	/*
536 	 * Modern (>=Nehalem) Intel systems use ACPI via intel_idle,
537 	 * not this code.  Assume that any Intel systems using this
538 	 * are ancient and may need the dummy wait.  This also assumes
539 	 * that the motivating chipset issue was Intel-only.
540 	 */
541 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
542 		return;
543 #endif
544 	/*
545 	 * Dummy wait op - must do something useless after P_LVL2 read
546 	 * because chipsets cannot guarantee that STPCLK# signal gets
547 	 * asserted in time to freeze execution properly
548 	 *
549 	 * This workaround has been in place since the original ACPI
550 	 * implementation was merged, circa 2002.
551 	 *
552 	 * If a profile is pointing to this instruction, please first
553 	 * consider moving your system to a more modern idle
554 	 * mechanism.
555 	 */
556 	inl(acpi_gbl_FADT.xpm_timer_block.address);
557 }
558 
559 /**
560  * acpi_idle_do_entry - enter idle state using the appropriate method
561  * @cx: cstate data
562  *
563  * Caller disables interrupt before call and enables interrupt after return.
564  */
565 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
566 {
567 	perf_lopwr_cb(true);
568 
569 	if (cx->entry_method == ACPI_CSTATE_FFH) {
570 		/* Call into architectural FFH based C-state */
571 		acpi_processor_ffh_cstate_enter(cx);
572 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
573 		acpi_safe_halt();
574 	} else {
575 		io_idle(cx->address);
576 	}
577 
578 	perf_lopwr_cb(false);
579 }
580 
581 /**
582  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
583  * @dev: the target CPU
584  * @index: the index of suggested state
585  */
586 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
587 {
588 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
589 
590 	ACPI_FLUSH_CPU_CACHE();
591 
592 	while (1) {
593 
594 		if (cx->entry_method == ACPI_CSTATE_HALT)
595 			safe_halt();
596 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
597 			io_idle(cx->address);
598 		} else
599 			return -ENODEV;
600 
601 #if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU)
602 		cond_wakeup_cpu0();
603 #endif
604 	}
605 
606 	/* Never reached */
607 	return 0;
608 }
609 
610 static __always_inline bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
611 {
612 	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
613 		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
614 }
615 
616 static int c3_cpu_count;
617 static DEFINE_RAW_SPINLOCK(c3_lock);
618 
619 /**
620  * acpi_idle_enter_bm - enters C3 with proper BM handling
621  * @drv: cpuidle driver
622  * @pr: Target processor
623  * @cx: Target state context
624  * @index: index of target state
625  */
626 static int __cpuidle acpi_idle_enter_bm(struct cpuidle_driver *drv,
627 			       struct acpi_processor *pr,
628 			       struct acpi_processor_cx *cx,
629 			       int index)
630 {
631 	static struct acpi_processor_cx safe_cx = {
632 		.entry_method = ACPI_CSTATE_HALT,
633 	};
634 
635 	/*
636 	 * disable bus master
637 	 * bm_check implies we need ARB_DIS
638 	 * bm_control implies whether we can do ARB_DIS
639 	 *
640 	 * That leaves a case where bm_check is set and bm_control is not set.
641 	 * In that case we cannot do much, we enter C3 without doing anything.
642 	 */
643 	bool dis_bm = pr->flags.bm_control;
644 
645 	instrumentation_begin();
646 
647 	/* If we can skip BM, demote to a safe state. */
648 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
649 		dis_bm = false;
650 		index = drv->safe_state_index;
651 		if (index >= 0) {
652 			cx = this_cpu_read(acpi_cstate[index]);
653 		} else {
654 			cx = &safe_cx;
655 			index = -EBUSY;
656 		}
657 	}
658 
659 	if (dis_bm) {
660 		raw_spin_lock(&c3_lock);
661 		c3_cpu_count++;
662 		/* Disable bus master arbitration when all CPUs are in C3 */
663 		if (c3_cpu_count == num_online_cpus())
664 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
665 		raw_spin_unlock(&c3_lock);
666 	}
667 
668 	ct_cpuidle_enter();
669 
670 	acpi_idle_do_entry(cx);
671 
672 	ct_cpuidle_exit();
673 
674 	/* Re-enable bus master arbitration */
675 	if (dis_bm) {
676 		raw_spin_lock(&c3_lock);
677 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
678 		c3_cpu_count--;
679 		raw_spin_unlock(&c3_lock);
680 	}
681 
682 	instrumentation_end();
683 
684 	return index;
685 }
686 
687 static int __cpuidle acpi_idle_enter(struct cpuidle_device *dev,
688 			   struct cpuidle_driver *drv, int index)
689 {
690 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
691 	struct acpi_processor *pr;
692 
693 	pr = __this_cpu_read(processors);
694 	if (unlikely(!pr))
695 		return -EINVAL;
696 
697 	if (cx->type != ACPI_STATE_C1) {
698 		if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
699 			return acpi_idle_enter_bm(drv, pr, cx, index);
700 
701 		/* C2 to C1 demotion. */
702 		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
703 			index = ACPI_IDLE_STATE_START;
704 			cx = per_cpu(acpi_cstate[index], dev->cpu);
705 		}
706 	}
707 
708 	if (cx->type == ACPI_STATE_C3)
709 		ACPI_FLUSH_CPU_CACHE();
710 
711 	acpi_idle_do_entry(cx);
712 
713 	return index;
714 }
715 
716 static int __cpuidle acpi_idle_enter_s2idle(struct cpuidle_device *dev,
717 				  struct cpuidle_driver *drv, int index)
718 {
719 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
720 
721 	if (cx->type == ACPI_STATE_C3) {
722 		struct acpi_processor *pr = __this_cpu_read(processors);
723 
724 		if (unlikely(!pr))
725 			return 0;
726 
727 		if (pr->flags.bm_check) {
728 			u8 bm_sts_skip = cx->bm_sts_skip;
729 
730 			/* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
731 			cx->bm_sts_skip = 1;
732 			acpi_idle_enter_bm(drv, pr, cx, index);
733 			cx->bm_sts_skip = bm_sts_skip;
734 
735 			return 0;
736 		} else {
737 			ACPI_FLUSH_CPU_CACHE();
738 		}
739 	}
740 	acpi_idle_do_entry(cx);
741 
742 	return 0;
743 }
744 
745 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
746 					   struct cpuidle_device *dev)
747 {
748 	int i, count = ACPI_IDLE_STATE_START;
749 	struct acpi_processor_cx *cx;
750 	struct cpuidle_state *state;
751 
752 	if (max_cstate == 0)
753 		max_cstate = 1;
754 
755 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
756 		state = &acpi_idle_driver.states[count];
757 		cx = &pr->power.states[i];
758 
759 		if (!cx->valid)
760 			continue;
761 
762 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
763 
764 		if (lapic_timer_needs_broadcast(pr, cx))
765 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
766 
767 		if (cx->type == ACPI_STATE_C3) {
768 			state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
769 			if (pr->flags.bm_check)
770 				state->flags |= CPUIDLE_FLAG_RCU_IDLE;
771 		}
772 
773 		count++;
774 		if (count == CPUIDLE_STATE_MAX)
775 			break;
776 	}
777 
778 	if (!count)
779 		return -EINVAL;
780 
781 	return 0;
782 }
783 
784 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
785 {
786 	int i, count;
787 	struct acpi_processor_cx *cx;
788 	struct cpuidle_state *state;
789 	struct cpuidle_driver *drv = &acpi_idle_driver;
790 
791 	if (max_cstate == 0)
792 		max_cstate = 1;
793 
794 	if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
795 		cpuidle_poll_state_init(drv);
796 		count = 1;
797 	} else {
798 		count = 0;
799 	}
800 
801 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
802 		cx = &pr->power.states[i];
803 
804 		if (!cx->valid)
805 			continue;
806 
807 		state = &drv->states[count];
808 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
809 		strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
810 		state->exit_latency = cx->latency;
811 		state->target_residency = cx->latency * latency_factor;
812 		state->enter = acpi_idle_enter;
813 
814 		state->flags = 0;
815 		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2 ||
816 		    cx->type == ACPI_STATE_C3) {
817 			state->enter_dead = acpi_idle_play_dead;
818 			if (cx->type != ACPI_STATE_C3)
819 				drv->safe_state_index = count;
820 		}
821 		/*
822 		 * Halt-induced C1 is not good for ->enter_s2idle, because it
823 		 * re-enables interrupts on exit.  Moreover, C1 is generally not
824 		 * particularly interesting from the suspend-to-idle angle, so
825 		 * avoid C1 and the situations in which we may need to fall back
826 		 * to it altogether.
827 		 */
828 		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
829 			state->enter_s2idle = acpi_idle_enter_s2idle;
830 
831 		count++;
832 		if (count == CPUIDLE_STATE_MAX)
833 			break;
834 	}
835 
836 	drv->state_count = count;
837 
838 	if (!count)
839 		return -EINVAL;
840 
841 	return 0;
842 }
843 
844 static inline void acpi_processor_cstate_first_run_checks(void)
845 {
846 	static int first_run;
847 
848 	if (first_run)
849 		return;
850 	dmi_check_system(processor_power_dmi_table);
851 	max_cstate = acpi_processor_cstate_check(max_cstate);
852 	if (max_cstate < ACPI_C_STATES_MAX)
853 		pr_notice("processor limited to max C-state %d\n", max_cstate);
854 
855 	first_run++;
856 
857 	if (nocst)
858 		return;
859 
860 	acpi_processor_claim_cst_control();
861 }
862 #else
863 
864 static inline int disabled_by_idle_boot_param(void) { return 0; }
865 static inline void acpi_processor_cstate_first_run_checks(void) { }
866 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
867 {
868 	return -ENODEV;
869 }
870 
871 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
872 					   struct cpuidle_device *dev)
873 {
874 	return -EINVAL;
875 }
876 
877 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
878 {
879 	return -EINVAL;
880 }
881 
882 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
883 
884 struct acpi_lpi_states_array {
885 	unsigned int size;
886 	unsigned int composite_states_size;
887 	struct acpi_lpi_state *entries;
888 	struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
889 };
890 
891 static int obj_get_integer(union acpi_object *obj, u32 *value)
892 {
893 	if (obj->type != ACPI_TYPE_INTEGER)
894 		return -EINVAL;
895 
896 	*value = obj->integer.value;
897 	return 0;
898 }
899 
900 static int acpi_processor_evaluate_lpi(acpi_handle handle,
901 				       struct acpi_lpi_states_array *info)
902 {
903 	acpi_status status;
904 	int ret = 0;
905 	int pkg_count, state_idx = 1, loop;
906 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
907 	union acpi_object *lpi_data;
908 	struct acpi_lpi_state *lpi_state;
909 
910 	status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
911 	if (ACPI_FAILURE(status)) {
912 		acpi_handle_debug(handle, "No _LPI, giving up\n");
913 		return -ENODEV;
914 	}
915 
916 	lpi_data = buffer.pointer;
917 
918 	/* There must be at least 4 elements = 3 elements + 1 package */
919 	if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
920 	    lpi_data->package.count < 4) {
921 		pr_debug("not enough elements in _LPI\n");
922 		ret = -ENODATA;
923 		goto end;
924 	}
925 
926 	pkg_count = lpi_data->package.elements[2].integer.value;
927 
928 	/* Validate number of power states. */
929 	if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
930 		pr_debug("count given by _LPI is not valid\n");
931 		ret = -ENODATA;
932 		goto end;
933 	}
934 
935 	lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
936 	if (!lpi_state) {
937 		ret = -ENOMEM;
938 		goto end;
939 	}
940 
941 	info->size = pkg_count;
942 	info->entries = lpi_state;
943 
944 	/* LPI States start at index 3 */
945 	for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
946 		union acpi_object *element, *pkg_elem, *obj;
947 
948 		element = &lpi_data->package.elements[loop];
949 		if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
950 			continue;
951 
952 		pkg_elem = element->package.elements;
953 
954 		obj = pkg_elem + 6;
955 		if (obj->type == ACPI_TYPE_BUFFER) {
956 			struct acpi_power_register *reg;
957 
958 			reg = (struct acpi_power_register *)obj->buffer.pointer;
959 			if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
960 			    reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
961 				continue;
962 
963 			lpi_state->address = reg->address;
964 			lpi_state->entry_method =
965 				reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
966 				ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
967 		} else if (obj->type == ACPI_TYPE_INTEGER) {
968 			lpi_state->entry_method = ACPI_CSTATE_INTEGER;
969 			lpi_state->address = obj->integer.value;
970 		} else {
971 			continue;
972 		}
973 
974 		/* elements[7,8] skipped for now i.e. Residency/Usage counter*/
975 
976 		obj = pkg_elem + 9;
977 		if (obj->type == ACPI_TYPE_STRING)
978 			strscpy(lpi_state->desc, obj->string.pointer,
979 				ACPI_CX_DESC_LEN);
980 
981 		lpi_state->index = state_idx;
982 		if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
983 			pr_debug("No min. residency found, assuming 10 us\n");
984 			lpi_state->min_residency = 10;
985 		}
986 
987 		if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
988 			pr_debug("No wakeup residency found, assuming 10 us\n");
989 			lpi_state->wake_latency = 10;
990 		}
991 
992 		if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
993 			lpi_state->flags = 0;
994 
995 		if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
996 			lpi_state->arch_flags = 0;
997 
998 		if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
999 			lpi_state->res_cnt_freq = 1;
1000 
1001 		if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
1002 			lpi_state->enable_parent_state = 0;
1003 	}
1004 
1005 	acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1006 end:
1007 	kfree(buffer.pointer);
1008 	return ret;
1009 }
1010 
1011 /*
1012  * flat_state_cnt - the number of composite LPI states after the process of flattening
1013  */
1014 static int flat_state_cnt;
1015 
1016 /**
1017  * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1018  *
1019  * @local: local LPI state
1020  * @parent: parent LPI state
1021  * @result: composite LPI state
1022  */
1023 static bool combine_lpi_states(struct acpi_lpi_state *local,
1024 			       struct acpi_lpi_state *parent,
1025 			       struct acpi_lpi_state *result)
1026 {
1027 	if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1028 		if (!parent->address) /* 0 means autopromotable */
1029 			return false;
1030 		result->address = local->address + parent->address;
1031 	} else {
1032 		result->address = parent->address;
1033 	}
1034 
1035 	result->min_residency = max(local->min_residency, parent->min_residency);
1036 	result->wake_latency = local->wake_latency + parent->wake_latency;
1037 	result->enable_parent_state = parent->enable_parent_state;
1038 	result->entry_method = local->entry_method;
1039 
1040 	result->flags = parent->flags;
1041 	result->arch_flags = parent->arch_flags;
1042 	result->index = parent->index;
1043 
1044 	strscpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1045 	strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1046 	strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1047 	return true;
1048 }
1049 
1050 #define ACPI_LPI_STATE_FLAGS_ENABLED			BIT(0)
1051 
1052 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1053 				  struct acpi_lpi_state *t)
1054 {
1055 	curr_level->composite_states[curr_level->composite_states_size++] = t;
1056 }
1057 
1058 static int flatten_lpi_states(struct acpi_processor *pr,
1059 			      struct acpi_lpi_states_array *curr_level,
1060 			      struct acpi_lpi_states_array *prev_level)
1061 {
1062 	int i, j, state_count = curr_level->size;
1063 	struct acpi_lpi_state *p, *t = curr_level->entries;
1064 
1065 	curr_level->composite_states_size = 0;
1066 	for (j = 0; j < state_count; j++, t++) {
1067 		struct acpi_lpi_state *flpi;
1068 
1069 		if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1070 			continue;
1071 
1072 		if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1073 			pr_warn("Limiting number of LPI states to max (%d)\n",
1074 				ACPI_PROCESSOR_MAX_POWER);
1075 			pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1076 			break;
1077 		}
1078 
1079 		flpi = &pr->power.lpi_states[flat_state_cnt];
1080 
1081 		if (!prev_level) { /* leaf/processor node */
1082 			memcpy(flpi, t, sizeof(*t));
1083 			stash_composite_state(curr_level, flpi);
1084 			flat_state_cnt++;
1085 			continue;
1086 		}
1087 
1088 		for (i = 0; i < prev_level->composite_states_size; i++) {
1089 			p = prev_level->composite_states[i];
1090 			if (t->index <= p->enable_parent_state &&
1091 			    combine_lpi_states(p, t, flpi)) {
1092 				stash_composite_state(curr_level, flpi);
1093 				flat_state_cnt++;
1094 				flpi++;
1095 			}
1096 		}
1097 	}
1098 
1099 	kfree(curr_level->entries);
1100 	return 0;
1101 }
1102 
1103 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1104 {
1105 	return -EOPNOTSUPP;
1106 }
1107 
1108 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1109 {
1110 	int ret, i;
1111 	acpi_status status;
1112 	acpi_handle handle = pr->handle, pr_ahandle;
1113 	struct acpi_device *d = NULL;
1114 	struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1115 
1116 	/* make sure our architecture has support */
1117 	ret = acpi_processor_ffh_lpi_probe(pr->id);
1118 	if (ret == -EOPNOTSUPP)
1119 		return ret;
1120 
1121 	if (!osc_pc_lpi_support_confirmed)
1122 		return -EOPNOTSUPP;
1123 
1124 	if (!acpi_has_method(handle, "_LPI"))
1125 		return -EINVAL;
1126 
1127 	flat_state_cnt = 0;
1128 	prev = &info[0];
1129 	curr = &info[1];
1130 	handle = pr->handle;
1131 	ret = acpi_processor_evaluate_lpi(handle, prev);
1132 	if (ret)
1133 		return ret;
1134 	flatten_lpi_states(pr, prev, NULL);
1135 
1136 	status = acpi_get_parent(handle, &pr_ahandle);
1137 	while (ACPI_SUCCESS(status)) {
1138 		d = acpi_fetch_acpi_dev(pr_ahandle);
1139 		if (!d)
1140 			break;
1141 
1142 		handle = pr_ahandle;
1143 
1144 		if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1145 			break;
1146 
1147 		/* can be optional ? */
1148 		if (!acpi_has_method(handle, "_LPI"))
1149 			break;
1150 
1151 		ret = acpi_processor_evaluate_lpi(handle, curr);
1152 		if (ret)
1153 			break;
1154 
1155 		/* flatten all the LPI states in this level of hierarchy */
1156 		flatten_lpi_states(pr, curr, prev);
1157 
1158 		tmp = prev, prev = curr, curr = tmp;
1159 
1160 		status = acpi_get_parent(handle, &pr_ahandle);
1161 	}
1162 
1163 	pr->power.count = flat_state_cnt;
1164 	/* reset the index after flattening */
1165 	for (i = 0; i < pr->power.count; i++)
1166 		pr->power.lpi_states[i].index = i;
1167 
1168 	/* Tell driver that _LPI is supported. */
1169 	pr->flags.has_lpi = 1;
1170 	pr->flags.power = 1;
1171 
1172 	return 0;
1173 }
1174 
1175 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1176 {
1177 	return -ENODEV;
1178 }
1179 
1180 /**
1181  * acpi_idle_lpi_enter - enters an ACPI any LPI state
1182  * @dev: the target CPU
1183  * @drv: cpuidle driver containing cpuidle state info
1184  * @index: index of target state
1185  *
1186  * Return: 0 for success or negative value for error
1187  */
1188 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1189 			       struct cpuidle_driver *drv, int index)
1190 {
1191 	struct acpi_processor *pr;
1192 	struct acpi_lpi_state *lpi;
1193 
1194 	pr = __this_cpu_read(processors);
1195 
1196 	if (unlikely(!pr))
1197 		return -EINVAL;
1198 
1199 	lpi = &pr->power.lpi_states[index];
1200 	if (lpi->entry_method == ACPI_CSTATE_FFH)
1201 		return acpi_processor_ffh_lpi_enter(lpi);
1202 
1203 	return -EINVAL;
1204 }
1205 
1206 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1207 {
1208 	int i;
1209 	struct acpi_lpi_state *lpi;
1210 	struct cpuidle_state *state;
1211 	struct cpuidle_driver *drv = &acpi_idle_driver;
1212 
1213 	if (!pr->flags.has_lpi)
1214 		return -EOPNOTSUPP;
1215 
1216 	for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1217 		lpi = &pr->power.lpi_states[i];
1218 
1219 		state = &drv->states[i];
1220 		snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1221 		strscpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1222 		state->exit_latency = lpi->wake_latency;
1223 		state->target_residency = lpi->min_residency;
1224 		if (lpi->arch_flags)
1225 			state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1226 		if (i != 0 && lpi->entry_method == ACPI_CSTATE_FFH)
1227 			state->flags |= CPUIDLE_FLAG_RCU_IDLE;
1228 		state->enter = acpi_idle_lpi_enter;
1229 		drv->safe_state_index = i;
1230 	}
1231 
1232 	drv->state_count = i;
1233 
1234 	return 0;
1235 }
1236 
1237 /**
1238  * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1239  * global state data i.e. idle routines
1240  *
1241  * @pr: the ACPI processor
1242  */
1243 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1244 {
1245 	int i;
1246 	struct cpuidle_driver *drv = &acpi_idle_driver;
1247 
1248 	if (!pr->flags.power_setup_done || !pr->flags.power)
1249 		return -EINVAL;
1250 
1251 	drv->safe_state_index = -1;
1252 	for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1253 		drv->states[i].name[0] = '\0';
1254 		drv->states[i].desc[0] = '\0';
1255 	}
1256 
1257 	if (pr->flags.has_lpi)
1258 		return acpi_processor_setup_lpi_states(pr);
1259 
1260 	return acpi_processor_setup_cstates(pr);
1261 }
1262 
1263 /**
1264  * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1265  * device i.e. per-cpu data
1266  *
1267  * @pr: the ACPI processor
1268  * @dev : the cpuidle device
1269  */
1270 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1271 					    struct cpuidle_device *dev)
1272 {
1273 	if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1274 		return -EINVAL;
1275 
1276 	dev->cpu = pr->id;
1277 	if (pr->flags.has_lpi)
1278 		return acpi_processor_ffh_lpi_probe(pr->id);
1279 
1280 	return acpi_processor_setup_cpuidle_cx(pr, dev);
1281 }
1282 
1283 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1284 {
1285 	int ret;
1286 
1287 	ret = acpi_processor_get_lpi_info(pr);
1288 	if (ret)
1289 		ret = acpi_processor_get_cstate_info(pr);
1290 
1291 	return ret;
1292 }
1293 
1294 int acpi_processor_hotplug(struct acpi_processor *pr)
1295 {
1296 	int ret = 0;
1297 	struct cpuidle_device *dev;
1298 
1299 	if (disabled_by_idle_boot_param())
1300 		return 0;
1301 
1302 	if (!pr->flags.power_setup_done)
1303 		return -ENODEV;
1304 
1305 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1306 	cpuidle_pause_and_lock();
1307 	cpuidle_disable_device(dev);
1308 	ret = acpi_processor_get_power_info(pr);
1309 	if (!ret && pr->flags.power) {
1310 		acpi_processor_setup_cpuidle_dev(pr, dev);
1311 		ret = cpuidle_enable_device(dev);
1312 	}
1313 	cpuidle_resume_and_unlock();
1314 
1315 	return ret;
1316 }
1317 
1318 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1319 {
1320 	int cpu;
1321 	struct acpi_processor *_pr;
1322 	struct cpuidle_device *dev;
1323 
1324 	if (disabled_by_idle_boot_param())
1325 		return 0;
1326 
1327 	if (!pr->flags.power_setup_done)
1328 		return -ENODEV;
1329 
1330 	/*
1331 	 * FIXME:  Design the ACPI notification to make it once per
1332 	 * system instead of once per-cpu.  This condition is a hack
1333 	 * to make the code that updates C-States be called once.
1334 	 */
1335 
1336 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1337 
1338 		/* Protect against cpu-hotplug */
1339 		cpus_read_lock();
1340 		cpuidle_pause_and_lock();
1341 
1342 		/* Disable all cpuidle devices */
1343 		for_each_online_cpu(cpu) {
1344 			_pr = per_cpu(processors, cpu);
1345 			if (!_pr || !_pr->flags.power_setup_done)
1346 				continue;
1347 			dev = per_cpu(acpi_cpuidle_device, cpu);
1348 			cpuidle_disable_device(dev);
1349 		}
1350 
1351 		/* Populate Updated C-state information */
1352 		acpi_processor_get_power_info(pr);
1353 		acpi_processor_setup_cpuidle_states(pr);
1354 
1355 		/* Enable all cpuidle devices */
1356 		for_each_online_cpu(cpu) {
1357 			_pr = per_cpu(processors, cpu);
1358 			if (!_pr || !_pr->flags.power_setup_done)
1359 				continue;
1360 			acpi_processor_get_power_info(_pr);
1361 			if (_pr->flags.power) {
1362 				dev = per_cpu(acpi_cpuidle_device, cpu);
1363 				acpi_processor_setup_cpuidle_dev(_pr, dev);
1364 				cpuidle_enable_device(dev);
1365 			}
1366 		}
1367 		cpuidle_resume_and_unlock();
1368 		cpus_read_unlock();
1369 	}
1370 
1371 	return 0;
1372 }
1373 
1374 static int acpi_processor_registered;
1375 
1376 int acpi_processor_power_init(struct acpi_processor *pr)
1377 {
1378 	int retval;
1379 	struct cpuidle_device *dev;
1380 
1381 	if (disabled_by_idle_boot_param())
1382 		return 0;
1383 
1384 	acpi_processor_cstate_first_run_checks();
1385 
1386 	if (!acpi_processor_get_power_info(pr))
1387 		pr->flags.power_setup_done = 1;
1388 
1389 	/*
1390 	 * Install the idle handler if processor power management is supported.
1391 	 * Note that we use previously set idle handler will be used on
1392 	 * platforms that only support C1.
1393 	 */
1394 	if (pr->flags.power) {
1395 		/* Register acpi_idle_driver if not already registered */
1396 		if (!acpi_processor_registered) {
1397 			acpi_processor_setup_cpuidle_states(pr);
1398 			retval = cpuidle_register_driver(&acpi_idle_driver);
1399 			if (retval)
1400 				return retval;
1401 			pr_debug("%s registered with cpuidle\n",
1402 				 acpi_idle_driver.name);
1403 		}
1404 
1405 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1406 		if (!dev)
1407 			return -ENOMEM;
1408 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1409 
1410 		acpi_processor_setup_cpuidle_dev(pr, dev);
1411 
1412 		/* Register per-cpu cpuidle_device. Cpuidle driver
1413 		 * must already be registered before registering device
1414 		 */
1415 		retval = cpuidle_register_device(dev);
1416 		if (retval) {
1417 			if (acpi_processor_registered == 0)
1418 				cpuidle_unregister_driver(&acpi_idle_driver);
1419 			return retval;
1420 		}
1421 		acpi_processor_registered++;
1422 	}
1423 	return 0;
1424 }
1425 
1426 int acpi_processor_power_exit(struct acpi_processor *pr)
1427 {
1428 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1429 
1430 	if (disabled_by_idle_boot_param())
1431 		return 0;
1432 
1433 	if (pr->flags.power) {
1434 		cpuidle_unregister_device(dev);
1435 		acpi_processor_registered--;
1436 		if (acpi_processor_registered == 0)
1437 			cpuidle_unregister_driver(&acpi_idle_driver);
1438 	}
1439 
1440 	pr->flags.power_setup_done = 0;
1441 	return 0;
1442 }
1443