xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision 930beb5a)
1 /*
2  * processor_idle - idle state submodule to the ACPI processor driver
3  *
4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
6  *  Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
7  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
8  *  			- Added processor hotplug support
9  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
10  *  			- Added support for C3 on SMP
11  *
12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13  *
14  *  This program is free software; you can redistribute it and/or modify
15  *  it under the terms of the GNU General Public License as published by
16  *  the Free Software Foundation; either version 2 of the License, or (at
17  *  your option) any later version.
18  *
19  *  This program is distributed in the hope that it will be useful, but
20  *  WITHOUT ANY WARRANTY; without even the implied warranty of
21  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22  *  General Public License for more details.
23  *
24  *  You should have received a copy of the GNU General Public License along
25  *  with this program; if not, write to the Free Software Foundation, Inc.,
26  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
27  *
28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
29  */
30 
31 #include <linux/module.h>
32 #include <linux/acpi.h>
33 #include <linux/dmi.h>
34 #include <linux/sched.h>       /* need_resched() */
35 #include <linux/clockchips.h>
36 #include <linux/cpuidle.h>
37 #include <linux/syscore_ops.h>
38 
39 /*
40  * Include the apic definitions for x86 to have the APIC timer related defines
41  * available also for UP (on SMP it gets magically included via linux/smp.h).
42  * asm/acpi.h is not an option, as it would require more include magic. Also
43  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
44  */
45 #ifdef CONFIG_X86
46 #include <asm/apic.h>
47 #endif
48 
49 #include <acpi/acpi_bus.h>
50 #include <acpi/processor.h>
51 
52 #define PREFIX "ACPI: "
53 
54 #define ACPI_PROCESSOR_CLASS            "processor"
55 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
56 ACPI_MODULE_NAME("processor_idle");
57 
58 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
59 module_param(max_cstate, uint, 0000);
60 static unsigned int nocst __read_mostly;
61 module_param(nocst, uint, 0000);
62 static int bm_check_disable __read_mostly;
63 module_param(bm_check_disable, uint, 0000);
64 
65 static unsigned int latency_factor __read_mostly = 2;
66 module_param(latency_factor, uint, 0644);
67 
68 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
69 
70 static DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX],
71 								acpi_cstate);
72 
73 static int disabled_by_idle_boot_param(void)
74 {
75 	return boot_option_idle_override == IDLE_POLL ||
76 		boot_option_idle_override == IDLE_HALT;
77 }
78 
79 /*
80  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
81  * For now disable this. Probably a bug somewhere else.
82  *
83  * To skip this limit, boot/load with a large max_cstate limit.
84  */
85 static int set_max_cstate(const struct dmi_system_id *id)
86 {
87 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
88 		return 0;
89 
90 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
91 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
92 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
93 
94 	max_cstate = (long)id->driver_data;
95 
96 	return 0;
97 }
98 
99 static struct dmi_system_id processor_power_dmi_table[] = {
100 	{ set_max_cstate, "Clevo 5600D", {
101 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
102 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
103 	 (void *)2},
104 	{ set_max_cstate, "Pavilion zv5000", {
105 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
106 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
107 	 (void *)1},
108 	{ set_max_cstate, "Asus L8400B", {
109 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
110 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
111 	 (void *)1},
112 	{},
113 };
114 
115 
116 /*
117  * Callers should disable interrupts before the call and enable
118  * interrupts after return.
119  */
120 static void acpi_safe_halt(void)
121 {
122 	if (!tif_need_resched()) {
123 		safe_halt();
124 		local_irq_disable();
125 	}
126 }
127 
128 #ifdef ARCH_APICTIMER_STOPS_ON_C3
129 
130 /*
131  * Some BIOS implementations switch to C3 in the published C2 state.
132  * This seems to be a common problem on AMD boxen, but other vendors
133  * are affected too. We pick the most conservative approach: we assume
134  * that the local APIC stops in both C2 and C3.
135  */
136 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
137 				   struct acpi_processor_cx *cx)
138 {
139 	struct acpi_processor_power *pwr = &pr->power;
140 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
141 
142 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
143 		return;
144 
145 	if (amd_e400_c1e_detected)
146 		type = ACPI_STATE_C1;
147 
148 	/*
149 	 * Check, if one of the previous states already marked the lapic
150 	 * unstable
151 	 */
152 	if (pwr->timer_broadcast_on_state < state)
153 		return;
154 
155 	if (cx->type >= type)
156 		pr->power.timer_broadcast_on_state = state;
157 }
158 
159 static void __lapic_timer_propagate_broadcast(void *arg)
160 {
161 	struct acpi_processor *pr = (struct acpi_processor *) arg;
162 	unsigned long reason;
163 
164 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
165 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
166 
167 	clockevents_notify(reason, &pr->id);
168 }
169 
170 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
171 {
172 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
173 				 (void *)pr, 1);
174 }
175 
176 /* Power(C) State timer broadcast control */
177 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
178 				       struct acpi_processor_cx *cx,
179 				       int broadcast)
180 {
181 	int state = cx - pr->power.states;
182 
183 	if (state >= pr->power.timer_broadcast_on_state) {
184 		unsigned long reason;
185 
186 		reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
187 			CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
188 		clockevents_notify(reason, &pr->id);
189 	}
190 }
191 
192 #else
193 
194 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
195 				   struct acpi_processor_cx *cstate) { }
196 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
197 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
198 				       struct acpi_processor_cx *cx,
199 				       int broadcast)
200 {
201 }
202 
203 #endif
204 
205 #ifdef CONFIG_PM_SLEEP
206 static u32 saved_bm_rld;
207 
208 static int acpi_processor_suspend(void)
209 {
210 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
211 	return 0;
212 }
213 
214 static void acpi_processor_resume(void)
215 {
216 	u32 resumed_bm_rld;
217 
218 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
219 	if (resumed_bm_rld == saved_bm_rld)
220 		return;
221 
222 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
223 }
224 
225 static struct syscore_ops acpi_processor_syscore_ops = {
226 	.suspend = acpi_processor_suspend,
227 	.resume = acpi_processor_resume,
228 };
229 
230 void acpi_processor_syscore_init(void)
231 {
232 	register_syscore_ops(&acpi_processor_syscore_ops);
233 }
234 
235 void acpi_processor_syscore_exit(void)
236 {
237 	unregister_syscore_ops(&acpi_processor_syscore_ops);
238 }
239 #endif /* CONFIG_PM_SLEEP */
240 
241 #if defined(CONFIG_X86)
242 static void tsc_check_state(int state)
243 {
244 	switch (boot_cpu_data.x86_vendor) {
245 	case X86_VENDOR_AMD:
246 	case X86_VENDOR_INTEL:
247 		/*
248 		 * AMD Fam10h TSC will tick in all
249 		 * C/P/S0/S1 states when this bit is set.
250 		 */
251 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
252 			return;
253 
254 		/*FALL THROUGH*/
255 	default:
256 		/* TSC could halt in idle, so notify users */
257 		if (state > ACPI_STATE_C1)
258 			mark_tsc_unstable("TSC halts in idle");
259 	}
260 }
261 #else
262 static void tsc_check_state(int state) { return; }
263 #endif
264 
265 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
266 {
267 
268 	if (!pr->pblk)
269 		return -ENODEV;
270 
271 	/* if info is obtained from pblk/fadt, type equals state */
272 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
273 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
274 
275 #ifndef CONFIG_HOTPLUG_CPU
276 	/*
277 	 * Check for P_LVL2_UP flag before entering C2 and above on
278 	 * an SMP system.
279 	 */
280 	if ((num_online_cpus() > 1) &&
281 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
282 		return -ENODEV;
283 #endif
284 
285 	/* determine C2 and C3 address from pblk */
286 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
287 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
288 
289 	/* determine latencies from FADT */
290 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
291 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
292 
293 	/*
294 	 * FADT specified C2 latency must be less than or equal to
295 	 * 100 microseconds.
296 	 */
297 	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
298 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
299 			"C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
300 		/* invalidate C2 */
301 		pr->power.states[ACPI_STATE_C2].address = 0;
302 	}
303 
304 	/*
305 	 * FADT supplied C3 latency must be less than or equal to
306 	 * 1000 microseconds.
307 	 */
308 	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
309 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
310 			"C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
311 		/* invalidate C3 */
312 		pr->power.states[ACPI_STATE_C3].address = 0;
313 	}
314 
315 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
316 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
317 			  pr->power.states[ACPI_STATE_C2].address,
318 			  pr->power.states[ACPI_STATE_C3].address));
319 
320 	return 0;
321 }
322 
323 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
324 {
325 	if (!pr->power.states[ACPI_STATE_C1].valid) {
326 		/* set the first C-State to C1 */
327 		/* all processors need to support C1 */
328 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
329 		pr->power.states[ACPI_STATE_C1].valid = 1;
330 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
331 	}
332 	/* the C0 state only exists as a filler in our array */
333 	pr->power.states[ACPI_STATE_C0].valid = 1;
334 	return 0;
335 }
336 
337 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
338 {
339 	acpi_status status = 0;
340 	u64 count;
341 	int current_count;
342 	int i;
343 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
344 	union acpi_object *cst;
345 
346 
347 	if (nocst)
348 		return -ENODEV;
349 
350 	current_count = 0;
351 
352 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
353 	if (ACPI_FAILURE(status)) {
354 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
355 		return -ENODEV;
356 	}
357 
358 	cst = buffer.pointer;
359 
360 	/* There must be at least 2 elements */
361 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
362 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
363 		status = -EFAULT;
364 		goto end;
365 	}
366 
367 	count = cst->package.elements[0].integer.value;
368 
369 	/* Validate number of power states. */
370 	if (count < 1 || count != cst->package.count - 1) {
371 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
372 		status = -EFAULT;
373 		goto end;
374 	}
375 
376 	/* Tell driver that at least _CST is supported. */
377 	pr->flags.has_cst = 1;
378 
379 	for (i = 1; i <= count; i++) {
380 		union acpi_object *element;
381 		union acpi_object *obj;
382 		struct acpi_power_register *reg;
383 		struct acpi_processor_cx cx;
384 
385 		memset(&cx, 0, sizeof(cx));
386 
387 		element = &(cst->package.elements[i]);
388 		if (element->type != ACPI_TYPE_PACKAGE)
389 			continue;
390 
391 		if (element->package.count != 4)
392 			continue;
393 
394 		obj = &(element->package.elements[0]);
395 
396 		if (obj->type != ACPI_TYPE_BUFFER)
397 			continue;
398 
399 		reg = (struct acpi_power_register *)obj->buffer.pointer;
400 
401 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
402 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
403 			continue;
404 
405 		/* There should be an easy way to extract an integer... */
406 		obj = &(element->package.elements[1]);
407 		if (obj->type != ACPI_TYPE_INTEGER)
408 			continue;
409 
410 		cx.type = obj->integer.value;
411 		/*
412 		 * Some buggy BIOSes won't list C1 in _CST -
413 		 * Let acpi_processor_get_power_info_default() handle them later
414 		 */
415 		if (i == 1 && cx.type != ACPI_STATE_C1)
416 			current_count++;
417 
418 		cx.address = reg->address;
419 		cx.index = current_count + 1;
420 
421 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
422 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
423 			if (acpi_processor_ffh_cstate_probe
424 					(pr->id, &cx, reg) == 0) {
425 				cx.entry_method = ACPI_CSTATE_FFH;
426 			} else if (cx.type == ACPI_STATE_C1) {
427 				/*
428 				 * C1 is a special case where FIXED_HARDWARE
429 				 * can be handled in non-MWAIT way as well.
430 				 * In that case, save this _CST entry info.
431 				 * Otherwise, ignore this info and continue.
432 				 */
433 				cx.entry_method = ACPI_CSTATE_HALT;
434 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
435 			} else {
436 				continue;
437 			}
438 			if (cx.type == ACPI_STATE_C1 &&
439 			    (boot_option_idle_override == IDLE_NOMWAIT)) {
440 				/*
441 				 * In most cases the C1 space_id obtained from
442 				 * _CST object is FIXED_HARDWARE access mode.
443 				 * But when the option of idle=halt is added,
444 				 * the entry_method type should be changed from
445 				 * CSTATE_FFH to CSTATE_HALT.
446 				 * When the option of idle=nomwait is added,
447 				 * the C1 entry_method type should be
448 				 * CSTATE_HALT.
449 				 */
450 				cx.entry_method = ACPI_CSTATE_HALT;
451 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
452 			}
453 		} else {
454 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
455 				 cx.address);
456 		}
457 
458 		if (cx.type == ACPI_STATE_C1) {
459 			cx.valid = 1;
460 		}
461 
462 		obj = &(element->package.elements[2]);
463 		if (obj->type != ACPI_TYPE_INTEGER)
464 			continue;
465 
466 		cx.latency = obj->integer.value;
467 
468 		obj = &(element->package.elements[3]);
469 		if (obj->type != ACPI_TYPE_INTEGER)
470 			continue;
471 
472 		current_count++;
473 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
474 
475 		/*
476 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
477 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
478 		 */
479 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
480 			printk(KERN_WARNING
481 			       "Limiting number of power states to max (%d)\n",
482 			       ACPI_PROCESSOR_MAX_POWER);
483 			printk(KERN_WARNING
484 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
485 			break;
486 		}
487 	}
488 
489 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
490 			  current_count));
491 
492 	/* Validate number of power states discovered */
493 	if (current_count < 2)
494 		status = -EFAULT;
495 
496       end:
497 	kfree(buffer.pointer);
498 
499 	return status;
500 }
501 
502 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
503 					   struct acpi_processor_cx *cx)
504 {
505 	static int bm_check_flag = -1;
506 	static int bm_control_flag = -1;
507 
508 
509 	if (!cx->address)
510 		return;
511 
512 	/*
513 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
514 	 * DMA transfers are used by any ISA device to avoid livelock.
515 	 * Note that we could disable Type-F DMA (as recommended by
516 	 * the erratum), but this is known to disrupt certain ISA
517 	 * devices thus we take the conservative approach.
518 	 */
519 	else if (errata.piix4.fdma) {
520 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
521 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
522 		return;
523 	}
524 
525 	/* All the logic here assumes flags.bm_check is same across all CPUs */
526 	if (bm_check_flag == -1) {
527 		/* Determine whether bm_check is needed based on CPU  */
528 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
529 		bm_check_flag = pr->flags.bm_check;
530 		bm_control_flag = pr->flags.bm_control;
531 	} else {
532 		pr->flags.bm_check = bm_check_flag;
533 		pr->flags.bm_control = bm_control_flag;
534 	}
535 
536 	if (pr->flags.bm_check) {
537 		if (!pr->flags.bm_control) {
538 			if (pr->flags.has_cst != 1) {
539 				/* bus mastering control is necessary */
540 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
541 					"C3 support requires BM control\n"));
542 				return;
543 			} else {
544 				/* Here we enter C3 without bus mastering */
545 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
546 					"C3 support without BM control\n"));
547 			}
548 		}
549 	} else {
550 		/*
551 		 * WBINVD should be set in fadt, for C3 state to be
552 		 * supported on when bm_check is not required.
553 		 */
554 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
555 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
556 					  "Cache invalidation should work properly"
557 					  " for C3 to be enabled on SMP systems\n"));
558 			return;
559 		}
560 	}
561 
562 	/*
563 	 * Otherwise we've met all of our C3 requirements.
564 	 * Normalize the C3 latency to expidite policy.  Enable
565 	 * checking of bus mastering status (bm_check) so we can
566 	 * use this in our C3 policy
567 	 */
568 	cx->valid = 1;
569 
570 	/*
571 	 * On older chipsets, BM_RLD needs to be set
572 	 * in order for Bus Master activity to wake the
573 	 * system from C3.  Newer chipsets handle DMA
574 	 * during C3 automatically and BM_RLD is a NOP.
575 	 * In either case, the proper way to
576 	 * handle BM_RLD is to set it and leave it set.
577 	 */
578 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
579 
580 	return;
581 }
582 
583 static int acpi_processor_power_verify(struct acpi_processor *pr)
584 {
585 	unsigned int i;
586 	unsigned int working = 0;
587 
588 	pr->power.timer_broadcast_on_state = INT_MAX;
589 
590 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
591 		struct acpi_processor_cx *cx = &pr->power.states[i];
592 
593 		switch (cx->type) {
594 		case ACPI_STATE_C1:
595 			cx->valid = 1;
596 			break;
597 
598 		case ACPI_STATE_C2:
599 			if (!cx->address)
600 				break;
601 			cx->valid = 1;
602 			break;
603 
604 		case ACPI_STATE_C3:
605 			acpi_processor_power_verify_c3(pr, cx);
606 			break;
607 		}
608 		if (!cx->valid)
609 			continue;
610 
611 		lapic_timer_check_state(i, pr, cx);
612 		tsc_check_state(cx->type);
613 		working++;
614 	}
615 
616 	lapic_timer_propagate_broadcast(pr);
617 
618 	return (working);
619 }
620 
621 static int acpi_processor_get_power_info(struct acpi_processor *pr)
622 {
623 	unsigned int i;
624 	int result;
625 
626 
627 	/* NOTE: the idle thread may not be running while calling
628 	 * this function */
629 
630 	/* Zero initialize all the C-states info. */
631 	memset(pr->power.states, 0, sizeof(pr->power.states));
632 
633 	result = acpi_processor_get_power_info_cst(pr);
634 	if (result == -ENODEV)
635 		result = acpi_processor_get_power_info_fadt(pr);
636 
637 	if (result)
638 		return result;
639 
640 	acpi_processor_get_power_info_default(pr);
641 
642 	pr->power.count = acpi_processor_power_verify(pr);
643 
644 	/*
645 	 * if one state of type C2 or C3 is available, mark this
646 	 * CPU as being "idle manageable"
647 	 */
648 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
649 		if (pr->power.states[i].valid) {
650 			pr->power.count = i;
651 			if (pr->power.states[i].type >= ACPI_STATE_C2)
652 				pr->flags.power = 1;
653 		}
654 	}
655 
656 	return 0;
657 }
658 
659 /**
660  * acpi_idle_bm_check - checks if bus master activity was detected
661  */
662 static int acpi_idle_bm_check(void)
663 {
664 	u32 bm_status = 0;
665 
666 	if (bm_check_disable)
667 		return 0;
668 
669 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
670 	if (bm_status)
671 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
672 	/*
673 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
674 	 * the true state of bus mastering activity; forcing us to
675 	 * manually check the BMIDEA bit of each IDE channel.
676 	 */
677 	else if (errata.piix4.bmisx) {
678 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
679 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
680 			bm_status = 1;
681 	}
682 	return bm_status;
683 }
684 
685 /**
686  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
687  * @cx: cstate data
688  *
689  * Caller disables interrupt before call and enables interrupt after return.
690  */
691 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
692 {
693 	/* Don't trace irqs off for idle */
694 	stop_critical_timings();
695 	if (cx->entry_method == ACPI_CSTATE_FFH) {
696 		/* Call into architectural FFH based C-state */
697 		acpi_processor_ffh_cstate_enter(cx);
698 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
699 		acpi_safe_halt();
700 	} else {
701 		/* IO port based C-state */
702 		inb(cx->address);
703 		/* Dummy wait op - must do something useless after P_LVL2 read
704 		   because chipsets cannot guarantee that STPCLK# signal
705 		   gets asserted in time to freeze execution properly. */
706 		inl(acpi_gbl_FADT.xpm_timer_block.address);
707 	}
708 	start_critical_timings();
709 }
710 
711 /**
712  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
713  * @dev: the target CPU
714  * @drv: cpuidle driver containing cpuidle state info
715  * @index: index of target state
716  *
717  * This is equivalent to the HALT instruction.
718  */
719 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
720 		struct cpuidle_driver *drv, int index)
721 {
722 	struct acpi_processor *pr;
723 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
724 
725 	pr = __this_cpu_read(processors);
726 
727 	if (unlikely(!pr))
728 		return -EINVAL;
729 
730 	if (cx->entry_method == ACPI_CSTATE_FFH) {
731 		if (current_set_polling_and_test())
732 			return -EINVAL;
733 	}
734 
735 	lapic_timer_state_broadcast(pr, cx, 1);
736 	acpi_idle_do_entry(cx);
737 
738 	lapic_timer_state_broadcast(pr, cx, 0);
739 
740 	return index;
741 }
742 
743 
744 /**
745  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
746  * @dev: the target CPU
747  * @index: the index of suggested state
748  */
749 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
750 {
751 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
752 
753 	ACPI_FLUSH_CPU_CACHE();
754 
755 	while (1) {
756 
757 		if (cx->entry_method == ACPI_CSTATE_HALT)
758 			safe_halt();
759 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
760 			inb(cx->address);
761 			/* See comment in acpi_idle_do_entry() */
762 			inl(acpi_gbl_FADT.xpm_timer_block.address);
763 		} else
764 			return -ENODEV;
765 	}
766 
767 	/* Never reached */
768 	return 0;
769 }
770 
771 /**
772  * acpi_idle_enter_simple - enters an ACPI state without BM handling
773  * @dev: the target CPU
774  * @drv: cpuidle driver with cpuidle state information
775  * @index: the index of suggested state
776  */
777 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
778 		struct cpuidle_driver *drv, int index)
779 {
780 	struct acpi_processor *pr;
781 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
782 
783 	pr = __this_cpu_read(processors);
784 
785 	if (unlikely(!pr))
786 		return -EINVAL;
787 
788 	if (cx->entry_method == ACPI_CSTATE_FFH) {
789 		if (current_set_polling_and_test())
790 			return -EINVAL;
791 	}
792 
793 	/*
794 	 * Must be done before busmaster disable as we might need to
795 	 * access HPET !
796 	 */
797 	lapic_timer_state_broadcast(pr, cx, 1);
798 
799 	if (cx->type == ACPI_STATE_C3)
800 		ACPI_FLUSH_CPU_CACHE();
801 
802 	/* Tell the scheduler that we are going deep-idle: */
803 	sched_clock_idle_sleep_event();
804 	acpi_idle_do_entry(cx);
805 
806 	sched_clock_idle_wakeup_event(0);
807 
808 	lapic_timer_state_broadcast(pr, cx, 0);
809 	return index;
810 }
811 
812 static int c3_cpu_count;
813 static DEFINE_RAW_SPINLOCK(c3_lock);
814 
815 /**
816  * acpi_idle_enter_bm - enters C3 with proper BM handling
817  * @dev: the target CPU
818  * @drv: cpuidle driver containing state data
819  * @index: the index of suggested state
820  *
821  * If BM is detected, the deepest non-C3 idle state is entered instead.
822  */
823 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
824 		struct cpuidle_driver *drv, int index)
825 {
826 	struct acpi_processor *pr;
827 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
828 
829 	pr = __this_cpu_read(processors);
830 
831 	if (unlikely(!pr))
832 		return -EINVAL;
833 
834 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
835 		if (drv->safe_state_index >= 0) {
836 			return drv->states[drv->safe_state_index].enter(dev,
837 						drv, drv->safe_state_index);
838 		} else {
839 			acpi_safe_halt();
840 			return -EBUSY;
841 		}
842 	}
843 
844 	if (cx->entry_method == ACPI_CSTATE_FFH) {
845 		if (current_set_polling_and_test())
846 			return -EINVAL;
847 	}
848 
849 	acpi_unlazy_tlb(smp_processor_id());
850 
851 	/* Tell the scheduler that we are going deep-idle: */
852 	sched_clock_idle_sleep_event();
853 	/*
854 	 * Must be done before busmaster disable as we might need to
855 	 * access HPET !
856 	 */
857 	lapic_timer_state_broadcast(pr, cx, 1);
858 
859 	/*
860 	 * disable bus master
861 	 * bm_check implies we need ARB_DIS
862 	 * !bm_check implies we need cache flush
863 	 * bm_control implies whether we can do ARB_DIS
864 	 *
865 	 * That leaves a case where bm_check is set and bm_control is
866 	 * not set. In that case we cannot do much, we enter C3
867 	 * without doing anything.
868 	 */
869 	if (pr->flags.bm_check && pr->flags.bm_control) {
870 		raw_spin_lock(&c3_lock);
871 		c3_cpu_count++;
872 		/* Disable bus master arbitration when all CPUs are in C3 */
873 		if (c3_cpu_count == num_online_cpus())
874 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
875 		raw_spin_unlock(&c3_lock);
876 	} else if (!pr->flags.bm_check) {
877 		ACPI_FLUSH_CPU_CACHE();
878 	}
879 
880 	acpi_idle_do_entry(cx);
881 
882 	/* Re-enable bus master arbitration */
883 	if (pr->flags.bm_check && pr->flags.bm_control) {
884 		raw_spin_lock(&c3_lock);
885 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
886 		c3_cpu_count--;
887 		raw_spin_unlock(&c3_lock);
888 	}
889 
890 	sched_clock_idle_wakeup_event(0);
891 
892 	lapic_timer_state_broadcast(pr, cx, 0);
893 	return index;
894 }
895 
896 struct cpuidle_driver acpi_idle_driver = {
897 	.name =		"acpi_idle",
898 	.owner =	THIS_MODULE,
899 };
900 
901 /**
902  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
903  * device i.e. per-cpu data
904  *
905  * @pr: the ACPI processor
906  * @dev : the cpuidle device
907  */
908 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
909 					   struct cpuidle_device *dev)
910 {
911 	int i, count = CPUIDLE_DRIVER_STATE_START;
912 	struct acpi_processor_cx *cx;
913 
914 	if (!pr->flags.power_setup_done)
915 		return -EINVAL;
916 
917 	if (pr->flags.power == 0) {
918 		return -EINVAL;
919 	}
920 
921 	if (!dev)
922 		return -EINVAL;
923 
924 	dev->cpu = pr->id;
925 
926 	if (max_cstate == 0)
927 		max_cstate = 1;
928 
929 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
930 		cx = &pr->power.states[i];
931 
932 		if (!cx->valid)
933 			continue;
934 
935 #ifdef CONFIG_HOTPLUG_CPU
936 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
937 		    !pr->flags.has_cst &&
938 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
939 			continue;
940 #endif
941 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
942 
943 		count++;
944 		if (count == CPUIDLE_STATE_MAX)
945 			break;
946 	}
947 
948 	dev->state_count = count;
949 
950 	if (!count)
951 		return -EINVAL;
952 
953 	return 0;
954 }
955 
956 /**
957  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
958  * global state data i.e. idle routines
959  *
960  * @pr: the ACPI processor
961  */
962 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
963 {
964 	int i, count = CPUIDLE_DRIVER_STATE_START;
965 	struct acpi_processor_cx *cx;
966 	struct cpuidle_state *state;
967 	struct cpuidle_driver *drv = &acpi_idle_driver;
968 
969 	if (!pr->flags.power_setup_done)
970 		return -EINVAL;
971 
972 	if (pr->flags.power == 0)
973 		return -EINVAL;
974 
975 	drv->safe_state_index = -1;
976 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
977 		drv->states[i].name[0] = '\0';
978 		drv->states[i].desc[0] = '\0';
979 	}
980 
981 	if (max_cstate == 0)
982 		max_cstate = 1;
983 
984 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
985 		cx = &pr->power.states[i];
986 
987 		if (!cx->valid)
988 			continue;
989 
990 #ifdef CONFIG_HOTPLUG_CPU
991 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
992 		    !pr->flags.has_cst &&
993 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
994 			continue;
995 #endif
996 
997 		state = &drv->states[count];
998 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
999 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1000 		state->exit_latency = cx->latency;
1001 		state->target_residency = cx->latency * latency_factor;
1002 
1003 		state->flags = 0;
1004 		switch (cx->type) {
1005 			case ACPI_STATE_C1:
1006 			if (cx->entry_method == ACPI_CSTATE_FFH)
1007 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1008 
1009 			state->enter = acpi_idle_enter_c1;
1010 			state->enter_dead = acpi_idle_play_dead;
1011 			drv->safe_state_index = count;
1012 			break;
1013 
1014 			case ACPI_STATE_C2:
1015 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1016 			state->enter = acpi_idle_enter_simple;
1017 			state->enter_dead = acpi_idle_play_dead;
1018 			drv->safe_state_index = count;
1019 			break;
1020 
1021 			case ACPI_STATE_C3:
1022 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1023 			state->enter = pr->flags.bm_check ?
1024 					acpi_idle_enter_bm :
1025 					acpi_idle_enter_simple;
1026 			break;
1027 		}
1028 
1029 		count++;
1030 		if (count == CPUIDLE_STATE_MAX)
1031 			break;
1032 	}
1033 
1034 	drv->state_count = count;
1035 
1036 	if (!count)
1037 		return -EINVAL;
1038 
1039 	return 0;
1040 }
1041 
1042 int acpi_processor_hotplug(struct acpi_processor *pr)
1043 {
1044 	int ret = 0;
1045 	struct cpuidle_device *dev;
1046 
1047 	if (disabled_by_idle_boot_param())
1048 		return 0;
1049 
1050 	if (nocst)
1051 		return -ENODEV;
1052 
1053 	if (!pr->flags.power_setup_done)
1054 		return -ENODEV;
1055 
1056 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1057 	cpuidle_pause_and_lock();
1058 	cpuidle_disable_device(dev);
1059 	acpi_processor_get_power_info(pr);
1060 	if (pr->flags.power) {
1061 		acpi_processor_setup_cpuidle_cx(pr, dev);
1062 		ret = cpuidle_enable_device(dev);
1063 	}
1064 	cpuidle_resume_and_unlock();
1065 
1066 	return ret;
1067 }
1068 
1069 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1070 {
1071 	int cpu;
1072 	struct acpi_processor *_pr;
1073 	struct cpuidle_device *dev;
1074 
1075 	if (disabled_by_idle_boot_param())
1076 		return 0;
1077 
1078 	if (nocst)
1079 		return -ENODEV;
1080 
1081 	if (!pr->flags.power_setup_done)
1082 		return -ENODEV;
1083 
1084 	/*
1085 	 * FIXME:  Design the ACPI notification to make it once per
1086 	 * system instead of once per-cpu.  This condition is a hack
1087 	 * to make the code that updates C-States be called once.
1088 	 */
1089 
1090 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1091 
1092 		cpuidle_pause_and_lock();
1093 		/* Protect against cpu-hotplug */
1094 		get_online_cpus();
1095 
1096 		/* Disable all cpuidle devices */
1097 		for_each_online_cpu(cpu) {
1098 			_pr = per_cpu(processors, cpu);
1099 			if (!_pr || !_pr->flags.power_setup_done)
1100 				continue;
1101 			dev = per_cpu(acpi_cpuidle_device, cpu);
1102 			cpuidle_disable_device(dev);
1103 		}
1104 
1105 		/* Populate Updated C-state information */
1106 		acpi_processor_get_power_info(pr);
1107 		acpi_processor_setup_cpuidle_states(pr);
1108 
1109 		/* Enable all cpuidle devices */
1110 		for_each_online_cpu(cpu) {
1111 			_pr = per_cpu(processors, cpu);
1112 			if (!_pr || !_pr->flags.power_setup_done)
1113 				continue;
1114 			acpi_processor_get_power_info(_pr);
1115 			if (_pr->flags.power) {
1116 				dev = per_cpu(acpi_cpuidle_device, cpu);
1117 				acpi_processor_setup_cpuidle_cx(_pr, dev);
1118 				cpuidle_enable_device(dev);
1119 			}
1120 		}
1121 		put_online_cpus();
1122 		cpuidle_resume_and_unlock();
1123 	}
1124 
1125 	return 0;
1126 }
1127 
1128 static int acpi_processor_registered;
1129 
1130 int acpi_processor_power_init(struct acpi_processor *pr)
1131 {
1132 	acpi_status status = 0;
1133 	int retval;
1134 	struct cpuidle_device *dev;
1135 	static int first_run;
1136 
1137 	if (disabled_by_idle_boot_param())
1138 		return 0;
1139 
1140 	if (!first_run) {
1141 		dmi_check_system(processor_power_dmi_table);
1142 		max_cstate = acpi_processor_cstate_check(max_cstate);
1143 		if (max_cstate < ACPI_C_STATES_MAX)
1144 			printk(KERN_NOTICE
1145 			       "ACPI: processor limited to max C-state %d\n",
1146 			       max_cstate);
1147 		first_run++;
1148 	}
1149 
1150 	if (acpi_gbl_FADT.cst_control && !nocst) {
1151 		status =
1152 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1153 		if (ACPI_FAILURE(status)) {
1154 			ACPI_EXCEPTION((AE_INFO, status,
1155 					"Notifying BIOS of _CST ability failed"));
1156 		}
1157 	}
1158 
1159 	acpi_processor_get_power_info(pr);
1160 	pr->flags.power_setup_done = 1;
1161 
1162 	/*
1163 	 * Install the idle handler if processor power management is supported.
1164 	 * Note that we use previously set idle handler will be used on
1165 	 * platforms that only support C1.
1166 	 */
1167 	if (pr->flags.power) {
1168 		/* Register acpi_idle_driver if not already registered */
1169 		if (!acpi_processor_registered) {
1170 			acpi_processor_setup_cpuidle_states(pr);
1171 			retval = cpuidle_register_driver(&acpi_idle_driver);
1172 			if (retval)
1173 				return retval;
1174 			printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1175 					acpi_idle_driver.name);
1176 		}
1177 
1178 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1179 		if (!dev)
1180 			return -ENOMEM;
1181 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1182 
1183 		acpi_processor_setup_cpuidle_cx(pr, dev);
1184 
1185 		/* Register per-cpu cpuidle_device. Cpuidle driver
1186 		 * must already be registered before registering device
1187 		 */
1188 		retval = cpuidle_register_device(dev);
1189 		if (retval) {
1190 			if (acpi_processor_registered == 0)
1191 				cpuidle_unregister_driver(&acpi_idle_driver);
1192 			return retval;
1193 		}
1194 		acpi_processor_registered++;
1195 	}
1196 	return 0;
1197 }
1198 
1199 int acpi_processor_power_exit(struct acpi_processor *pr)
1200 {
1201 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1202 
1203 	if (disabled_by_idle_boot_param())
1204 		return 0;
1205 
1206 	if (pr->flags.power) {
1207 		cpuidle_unregister_device(dev);
1208 		acpi_processor_registered--;
1209 		if (acpi_processor_registered == 0)
1210 			cpuidle_unregister_driver(&acpi_idle_driver);
1211 	}
1212 
1213 	pr->flags.power_setup_done = 0;
1214 	return 0;
1215 }
1216