xref: /linux/drivers/acpi/processor_idle.c (revision 2624f124b3b5d550ab2fbef7ee3bc0e1fed09722)
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       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/kernel.h>
32 #include <linux/module.h>
33 #include <linux/init.h>
34 #include <linux/cpufreq.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/acpi.h>
38 #include <linux/dmi.h>
39 #include <linux/moduleparam.h>
40 
41 #include <asm/io.h>
42 #include <asm/uaccess.h>
43 
44 #include <acpi/acpi_bus.h>
45 #include <acpi/processor.h>
46 
47 #define ACPI_PROCESSOR_COMPONENT        0x01000000
48 #define ACPI_PROCESSOR_CLASS            "processor"
49 #define ACPI_PROCESSOR_DRIVER_NAME      "ACPI Processor Driver"
50 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
51 ACPI_MODULE_NAME("acpi_processor")
52 #define ACPI_PROCESSOR_FILE_POWER	"power"
53 #define US_TO_PM_TIMER_TICKS(t)		((t * (PM_TIMER_FREQUENCY/1000)) / 1000)
54 #define C2_OVERHEAD			4	/* 1us (3.579 ticks per us) */
55 #define C3_OVERHEAD			4	/* 1us (3.579 ticks per us) */
56 static void (*pm_idle_save) (void);
57 module_param(max_cstate, uint, 0644);
58 
59 static unsigned int nocst = 0;
60 module_param(nocst, uint, 0000);
61 
62 /*
63  * bm_history -- bit-mask with a bit per jiffy of bus-master activity
64  * 1000 HZ: 0xFFFFFFFF: 32 jiffies = 32ms
65  * 800 HZ: 0xFFFFFFFF: 32 jiffies = 40ms
66  * 100 HZ: 0x0000000F: 4 jiffies = 40ms
67  * reduce history for more aggressive entry into C3
68  */
69 static unsigned int bm_history =
70     (HZ >= 800 ? 0xFFFFFFFF : ((1U << (HZ / 25)) - 1));
71 module_param(bm_history, uint, 0644);
72 /* --------------------------------------------------------------------------
73                                 Power Management
74    -------------------------------------------------------------------------- */
75 
76 /*
77  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
78  * For now disable this. Probably a bug somewhere else.
79  *
80  * To skip this limit, boot/load with a large max_cstate limit.
81  */
82 static int set_max_cstate(struct dmi_system_id *id)
83 {
84 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
85 		return 0;
86 
87 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
88 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
89 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
90 
91 	max_cstate = (long)id->driver_data;
92 
93 	return 0;
94 }
95 
96 static struct dmi_system_id __initdata processor_power_dmi_table[] = {
97 	{set_max_cstate, "IBM ThinkPad R40e", {
98 					       DMI_MATCH(DMI_BIOS_VENDOR,
99 							 "IBM"),
100 					       DMI_MATCH(DMI_BIOS_VERSION,
101 							 "1SET60WW")},
102 	 (void *)1},
103 	{set_max_cstate, "Medion 41700", {
104 					  DMI_MATCH(DMI_BIOS_VENDOR,
105 						    "Phoenix Technologies LTD"),
106 					  DMI_MATCH(DMI_BIOS_VERSION,
107 						    "R01-A1J")}, (void *)1},
108 	{set_max_cstate, "Clevo 5600D", {
109 					 DMI_MATCH(DMI_BIOS_VENDOR,
110 						   "Phoenix Technologies LTD"),
111 					 DMI_MATCH(DMI_BIOS_VERSION,
112 						   "SHE845M0.86C.0013.D.0302131307")},
113 	 (void *)2},
114 	{},
115 };
116 
117 static inline u32 ticks_elapsed(u32 t1, u32 t2)
118 {
119 	if (t2 >= t1)
120 		return (t2 - t1);
121 	else if (!acpi_fadt.tmr_val_ext)
122 		return (((0x00FFFFFF - t1) + t2) & 0x00FFFFFF);
123 	else
124 		return ((0xFFFFFFFF - t1) + t2);
125 }
126 
127 static void
128 acpi_processor_power_activate(struct acpi_processor *pr,
129 			      struct acpi_processor_cx *new)
130 {
131 	struct acpi_processor_cx *old;
132 
133 	if (!pr || !new)
134 		return;
135 
136 	old = pr->power.state;
137 
138 	if (old)
139 		old->promotion.count = 0;
140 	new->demotion.count = 0;
141 
142 	/* Cleanup from old state. */
143 	if (old) {
144 		switch (old->type) {
145 		case ACPI_STATE_C3:
146 			/* Disable bus master reload */
147 			if (new->type != ACPI_STATE_C3 && pr->flags.bm_check)
148 				acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 0,
149 						  ACPI_MTX_DO_NOT_LOCK);
150 			break;
151 		}
152 	}
153 
154 	/* Prepare to use new state. */
155 	switch (new->type) {
156 	case ACPI_STATE_C3:
157 		/* Enable bus master reload */
158 		if (old->type != ACPI_STATE_C3 && pr->flags.bm_check)
159 			acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD, 1,
160 					  ACPI_MTX_DO_NOT_LOCK);
161 		break;
162 	}
163 
164 	pr->power.state = new;
165 
166 	return;
167 }
168 
169 static atomic_t c3_cpu_count;
170 
171 static void acpi_processor_idle(void)
172 {
173 	struct acpi_processor *pr = NULL;
174 	struct acpi_processor_cx *cx = NULL;
175 	struct acpi_processor_cx *next_state = NULL;
176 	int sleep_ticks = 0;
177 	u32 t1, t2 = 0;
178 
179 	pr = processors[raw_smp_processor_id()];
180 	if (!pr)
181 		return;
182 
183 	/*
184 	 * Interrupts must be disabled during bus mastering calculations and
185 	 * for C2/C3 transitions.
186 	 */
187 	local_irq_disable();
188 
189 	/*
190 	 * Check whether we truly need to go idle, or should
191 	 * reschedule:
192 	 */
193 	if (unlikely(need_resched())) {
194 		local_irq_enable();
195 		return;
196 	}
197 
198 	cx = pr->power.state;
199 	if (!cx)
200 		goto easy_out;
201 
202 	/*
203 	 * Check BM Activity
204 	 * -----------------
205 	 * Check for bus mastering activity (if required), record, and check
206 	 * for demotion.
207 	 */
208 	if (pr->flags.bm_check) {
209 		u32 bm_status = 0;
210 		unsigned long diff = jiffies - pr->power.bm_check_timestamp;
211 
212 		if (diff > 32)
213 			diff = 32;
214 
215 		while (diff) {
216 			/* if we didn't get called, assume there was busmaster activity */
217 			diff--;
218 			if (diff)
219 				pr->power.bm_activity |= 0x1;
220 			pr->power.bm_activity <<= 1;
221 		}
222 
223 		acpi_get_register(ACPI_BITREG_BUS_MASTER_STATUS,
224 				  &bm_status, ACPI_MTX_DO_NOT_LOCK);
225 		if (bm_status) {
226 			pr->power.bm_activity++;
227 			acpi_set_register(ACPI_BITREG_BUS_MASTER_STATUS,
228 					  1, ACPI_MTX_DO_NOT_LOCK);
229 		}
230 		/*
231 		 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
232 		 * the true state of bus mastering activity; forcing us to
233 		 * manually check the BMIDEA bit of each IDE channel.
234 		 */
235 		else if (errata.piix4.bmisx) {
236 			if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
237 			    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
238 				pr->power.bm_activity++;
239 		}
240 
241 		pr->power.bm_check_timestamp = jiffies;
242 
243 		/*
244 		 * Apply bus mastering demotion policy.  Automatically demote
245 		 * to avoid a faulty transition.  Note that the processor
246 		 * won't enter a low-power state during this call (to this
247 		 * funciton) but should upon the next.
248 		 *
249 		 * TBD: A better policy might be to fallback to the demotion
250 		 *      state (use it for this quantum only) istead of
251 		 *      demoting -- and rely on duration as our sole demotion
252 		 *      qualification.  This may, however, introduce DMA
253 		 *      issues (e.g. floppy DMA transfer overrun/underrun).
254 		 */
255 		if (pr->power.bm_activity & cx->demotion.threshold.bm) {
256 			local_irq_enable();
257 			next_state = cx->demotion.state;
258 			goto end;
259 		}
260 	}
261 
262 	cx->usage++;
263 
264 	/*
265 	 * Sleep:
266 	 * ------
267 	 * Invoke the current Cx state to put the processor to sleep.
268 	 */
269 	switch (cx->type) {
270 
271 	case ACPI_STATE_C1:
272 		/*
273 		 * Invoke C1.
274 		 * Use the appropriate idle routine, the one that would
275 		 * be used without acpi C-states.
276 		 */
277 		if (pm_idle_save)
278 			pm_idle_save();
279 		else
280 			safe_halt();
281 		/*
282 		 * TBD: Can't get time duration while in C1, as resumes
283 		 *      go to an ISR rather than here.  Need to instrument
284 		 *      base interrupt handler.
285 		 */
286 		sleep_ticks = 0xFFFFFFFF;
287 		break;
288 
289 	case ACPI_STATE_C2:
290 		/* Get start time (ticks) */
291 		t1 = inl(acpi_fadt.xpm_tmr_blk.address);
292 		/* Invoke C2 */
293 		inb(cx->address);
294 		/* Dummy op - must do something useless after P_LVL2 read */
295 		t2 = inl(acpi_fadt.xpm_tmr_blk.address);
296 		/* Get end time (ticks) */
297 		t2 = inl(acpi_fadt.xpm_tmr_blk.address);
298 		/* Re-enable interrupts */
299 		local_irq_enable();
300 		/* Compute time (ticks) that we were actually asleep */
301 		sleep_ticks =
302 		    ticks_elapsed(t1, t2) - cx->latency_ticks - C2_OVERHEAD;
303 		break;
304 
305 	case ACPI_STATE_C3:
306 
307 		if (pr->flags.bm_check) {
308 			if (atomic_inc_return(&c3_cpu_count) ==
309 			    num_online_cpus()) {
310 				/*
311 				 * All CPUs are trying to go to C3
312 				 * Disable bus master arbitration
313 				 */
314 				acpi_set_register(ACPI_BITREG_ARB_DISABLE, 1,
315 						  ACPI_MTX_DO_NOT_LOCK);
316 			}
317 		} else {
318 			/* SMP with no shared cache... Invalidate cache  */
319 			ACPI_FLUSH_CPU_CACHE();
320 		}
321 
322 		/* Get start time (ticks) */
323 		t1 = inl(acpi_fadt.xpm_tmr_blk.address);
324 		/* Invoke C3 */
325 		inb(cx->address);
326 		/* Dummy op - must do something useless after P_LVL3 read */
327 		t2 = inl(acpi_fadt.xpm_tmr_blk.address);
328 		/* Get end time (ticks) */
329 		t2 = inl(acpi_fadt.xpm_tmr_blk.address);
330 		if (pr->flags.bm_check) {
331 			/* Enable bus master arbitration */
332 			atomic_dec(&c3_cpu_count);
333 			acpi_set_register(ACPI_BITREG_ARB_DISABLE, 0,
334 					  ACPI_MTX_DO_NOT_LOCK);
335 		}
336 
337 		/* Re-enable interrupts */
338 		local_irq_enable();
339 		/* Compute time (ticks) that we were actually asleep */
340 		sleep_ticks =
341 		    ticks_elapsed(t1, t2) - cx->latency_ticks - C3_OVERHEAD;
342 		break;
343 
344 	default:
345 		local_irq_enable();
346 		return;
347 	}
348 
349 	next_state = pr->power.state;
350 
351 	/*
352 	 * Promotion?
353 	 * ----------
354 	 * Track the number of longs (time asleep is greater than threshold)
355 	 * and promote when the count threshold is reached.  Note that bus
356 	 * mastering activity may prevent promotions.
357 	 * Do not promote above max_cstate.
358 	 */
359 	if (cx->promotion.state &&
360 	    ((cx->promotion.state - pr->power.states) <= max_cstate)) {
361 		if (sleep_ticks > cx->promotion.threshold.ticks) {
362 			cx->promotion.count++;
363 			cx->demotion.count = 0;
364 			if (cx->promotion.count >=
365 			    cx->promotion.threshold.count) {
366 				if (pr->flags.bm_check) {
367 					if (!
368 					    (pr->power.bm_activity & cx->
369 					     promotion.threshold.bm)) {
370 						next_state =
371 						    cx->promotion.state;
372 						goto end;
373 					}
374 				} else {
375 					next_state = cx->promotion.state;
376 					goto end;
377 				}
378 			}
379 		}
380 	}
381 
382 	/*
383 	 * Demotion?
384 	 * ---------
385 	 * Track the number of shorts (time asleep is less than time threshold)
386 	 * and demote when the usage threshold is reached.
387 	 */
388 	if (cx->demotion.state) {
389 		if (sleep_ticks < cx->demotion.threshold.ticks) {
390 			cx->demotion.count++;
391 			cx->promotion.count = 0;
392 			if (cx->demotion.count >= cx->demotion.threshold.count) {
393 				next_state = cx->demotion.state;
394 				goto end;
395 			}
396 		}
397 	}
398 
399       end:
400 	/*
401 	 * Demote if current state exceeds max_cstate
402 	 */
403 	if ((pr->power.state - pr->power.states) > max_cstate) {
404 		if (cx->demotion.state)
405 			next_state = cx->demotion.state;
406 	}
407 
408 	/*
409 	 * New Cx State?
410 	 * -------------
411 	 * If we're going to start using a new Cx state we must clean up
412 	 * from the previous and prepare to use the new.
413 	 */
414 	if (next_state != pr->power.state)
415 		acpi_processor_power_activate(pr, next_state);
416 
417 	return;
418 
419       easy_out:
420 	/* do C1 instead of busy loop */
421 	if (pm_idle_save)
422 		pm_idle_save();
423 	else
424 		safe_halt();
425 	return;
426 }
427 
428 static int acpi_processor_set_power_policy(struct acpi_processor *pr)
429 {
430 	unsigned int i;
431 	unsigned int state_is_set = 0;
432 	struct acpi_processor_cx *lower = NULL;
433 	struct acpi_processor_cx *higher = NULL;
434 	struct acpi_processor_cx *cx;
435 
436 	ACPI_FUNCTION_TRACE("acpi_processor_set_power_policy");
437 
438 	if (!pr)
439 		return_VALUE(-EINVAL);
440 
441 	/*
442 	 * This function sets the default Cx state policy (OS idle handler).
443 	 * Our scheme is to promote quickly to C2 but more conservatively
444 	 * to C3.  We're favoring C2  for its characteristics of low latency
445 	 * (quick response), good power savings, and ability to allow bus
446 	 * mastering activity.  Note that the Cx state policy is completely
447 	 * customizable and can be altered dynamically.
448 	 */
449 
450 	/* startup state */
451 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
452 		cx = &pr->power.states[i];
453 		if (!cx->valid)
454 			continue;
455 
456 		if (!state_is_set)
457 			pr->power.state = cx;
458 		state_is_set++;
459 		break;
460 	}
461 
462 	if (!state_is_set)
463 		return_VALUE(-ENODEV);
464 
465 	/* demotion */
466 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
467 		cx = &pr->power.states[i];
468 		if (!cx->valid)
469 			continue;
470 
471 		if (lower) {
472 			cx->demotion.state = lower;
473 			cx->demotion.threshold.ticks = cx->latency_ticks;
474 			cx->demotion.threshold.count = 1;
475 			if (cx->type == ACPI_STATE_C3)
476 				cx->demotion.threshold.bm = bm_history;
477 		}
478 
479 		lower = cx;
480 	}
481 
482 	/* promotion */
483 	for (i = (ACPI_PROCESSOR_MAX_POWER - 1); i > 0; i--) {
484 		cx = &pr->power.states[i];
485 		if (!cx->valid)
486 			continue;
487 
488 		if (higher) {
489 			cx->promotion.state = higher;
490 			cx->promotion.threshold.ticks = cx->latency_ticks;
491 			if (cx->type >= ACPI_STATE_C2)
492 				cx->promotion.threshold.count = 4;
493 			else
494 				cx->promotion.threshold.count = 10;
495 			if (higher->type == ACPI_STATE_C3)
496 				cx->promotion.threshold.bm = bm_history;
497 		}
498 
499 		higher = cx;
500 	}
501 
502 	return_VALUE(0);
503 }
504 
505 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
506 {
507 	int i;
508 
509 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_fadt");
510 
511 	if (!pr)
512 		return_VALUE(-EINVAL);
513 
514 	if (!pr->pblk)
515 		return_VALUE(-ENODEV);
516 
517 	for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
518 		memset(pr->power.states, 0, sizeof(struct acpi_processor_cx));
519 
520 	/* if info is obtained from pblk/fadt, type equals state */
521 	pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
522 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
523 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
524 
525 	/* the C0 state only exists as a filler in our array,
526 	 * and all processors need to support C1 */
527 	pr->power.states[ACPI_STATE_C0].valid = 1;
528 	pr->power.states[ACPI_STATE_C1].valid = 1;
529 
530 	/* determine C2 and C3 address from pblk */
531 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
532 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
533 
534 	/* determine latencies from FADT */
535 	pr->power.states[ACPI_STATE_C2].latency = acpi_fadt.plvl2_lat;
536 	pr->power.states[ACPI_STATE_C3].latency = acpi_fadt.plvl3_lat;
537 
538 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
539 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
540 			  pr->power.states[ACPI_STATE_C2].address,
541 			  pr->power.states[ACPI_STATE_C3].address));
542 
543 	return_VALUE(0);
544 }
545 
546 static int acpi_processor_get_power_info_default_c1(struct acpi_processor *pr)
547 {
548 	int i;
549 
550 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_default_c1");
551 
552 	for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
553 		memset(&(pr->power.states[i]), 0,
554 		       sizeof(struct acpi_processor_cx));
555 
556 	/* if info is obtained from pblk/fadt, type equals state */
557 	pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
558 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
559 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
560 
561 	/* the C0 state only exists as a filler in our array,
562 	 * and all processors need to support C1 */
563 	pr->power.states[ACPI_STATE_C0].valid = 1;
564 	pr->power.states[ACPI_STATE_C1].valid = 1;
565 
566 	return_VALUE(0);
567 }
568 
569 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
570 {
571 	acpi_status status = 0;
572 	acpi_integer count;
573 	int i;
574 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
575 	union acpi_object *cst;
576 
577 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_info_cst");
578 
579 	if (nocst)
580 		return_VALUE(-ENODEV);
581 
582 	pr->power.count = 0;
583 	for (i = 0; i < ACPI_PROCESSOR_MAX_POWER; i++)
584 		memset(&(pr->power.states[i]), 0,
585 		       sizeof(struct acpi_processor_cx));
586 
587 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
588 	if (ACPI_FAILURE(status)) {
589 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
590 		return_VALUE(-ENODEV);
591 	}
592 
593 	cst = (union acpi_object *)buffer.pointer;
594 
595 	/* There must be at least 2 elements */
596 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
597 		ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
598 				  "not enough elements in _CST\n"));
599 		status = -EFAULT;
600 		goto end;
601 	}
602 
603 	count = cst->package.elements[0].integer.value;
604 
605 	/* Validate number of power states. */
606 	if (count < 1 || count != cst->package.count - 1) {
607 		ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
608 				  "count given by _CST is not valid\n"));
609 		status = -EFAULT;
610 		goto end;
611 	}
612 
613 	/* We support up to ACPI_PROCESSOR_MAX_POWER. */
614 	if (count > ACPI_PROCESSOR_MAX_POWER) {
615 		printk(KERN_WARNING
616 		       "Limiting number of power states to max (%d)\n",
617 		       ACPI_PROCESSOR_MAX_POWER);
618 		printk(KERN_WARNING
619 		       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
620 		count = ACPI_PROCESSOR_MAX_POWER;
621 	}
622 
623 	/* Tell driver that at least _CST is supported. */
624 	pr->flags.has_cst = 1;
625 
626 	for (i = 1; i <= count; i++) {
627 		union acpi_object *element;
628 		union acpi_object *obj;
629 		struct acpi_power_register *reg;
630 		struct acpi_processor_cx cx;
631 
632 		memset(&cx, 0, sizeof(cx));
633 
634 		element = (union acpi_object *)&(cst->package.elements[i]);
635 		if (element->type != ACPI_TYPE_PACKAGE)
636 			continue;
637 
638 		if (element->package.count != 4)
639 			continue;
640 
641 		obj = (union acpi_object *)&(element->package.elements[0]);
642 
643 		if (obj->type != ACPI_TYPE_BUFFER)
644 			continue;
645 
646 		reg = (struct acpi_power_register *)obj->buffer.pointer;
647 
648 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
649 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
650 			continue;
651 
652 		cx.address = (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) ?
653 		    0 : reg->address;
654 
655 		/* There should be an easy way to extract an integer... */
656 		obj = (union acpi_object *)&(element->package.elements[1]);
657 		if (obj->type != ACPI_TYPE_INTEGER)
658 			continue;
659 
660 		cx.type = obj->integer.value;
661 
662 		if ((cx.type != ACPI_STATE_C1) &&
663 		    (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO))
664 			continue;
665 
666 		if ((cx.type < ACPI_STATE_C1) || (cx.type > ACPI_STATE_C3))
667 			continue;
668 
669 		obj = (union acpi_object *)&(element->package.elements[2]);
670 		if (obj->type != ACPI_TYPE_INTEGER)
671 			continue;
672 
673 		cx.latency = obj->integer.value;
674 
675 		obj = (union acpi_object *)&(element->package.elements[3]);
676 		if (obj->type != ACPI_TYPE_INTEGER)
677 			continue;
678 
679 		cx.power = obj->integer.value;
680 
681 		(pr->power.count)++;
682 		memcpy(&(pr->power.states[pr->power.count]), &cx, sizeof(cx));
683 	}
684 
685 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
686 			  pr->power.count));
687 
688 	/* Validate number of power states discovered */
689 	if (pr->power.count < 2)
690 		status = -ENODEV;
691 
692       end:
693 	acpi_os_free(buffer.pointer);
694 
695 	return_VALUE(status);
696 }
697 
698 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
699 {
700 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c2");
701 
702 	if (!cx->address)
703 		return_VOID;
704 
705 	/*
706 	 * C2 latency must be less than or equal to 100
707 	 * microseconds.
708 	 */
709 	else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
710 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
711 				  "latency too large [%d]\n", cx->latency));
712 		return_VOID;
713 	}
714 
715 	/*
716 	 * Otherwise we've met all of our C2 requirements.
717 	 * Normalize the C2 latency to expidite policy
718 	 */
719 	cx->valid = 1;
720 	cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
721 
722 	return_VOID;
723 }
724 
725 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
726 					   struct acpi_processor_cx *cx)
727 {
728 	static int bm_check_flag;
729 
730 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_verify_c3");
731 
732 	if (!cx->address)
733 		return_VOID;
734 
735 	/*
736 	 * C3 latency must be less than or equal to 1000
737 	 * microseconds.
738 	 */
739 	else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
740 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
741 				  "latency too large [%d]\n", cx->latency));
742 		return_VOID;
743 	}
744 
745 	/*
746 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
747 	 * DMA transfers are used by any ISA device to avoid livelock.
748 	 * Note that we could disable Type-F DMA (as recommended by
749 	 * the erratum), but this is known to disrupt certain ISA
750 	 * devices thus we take the conservative approach.
751 	 */
752 	else if (errata.piix4.fdma) {
753 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
754 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
755 		return_VOID;
756 	}
757 
758 	/* All the logic here assumes flags.bm_check is same across all CPUs */
759 	if (!bm_check_flag) {
760 		/* Determine whether bm_check is needed based on CPU  */
761 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
762 		bm_check_flag = pr->flags.bm_check;
763 	} else {
764 		pr->flags.bm_check = bm_check_flag;
765 	}
766 
767 	if (pr->flags.bm_check) {
768 		/* bus mastering control is necessary */
769 		if (!pr->flags.bm_control) {
770 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
771 					  "C3 support requires bus mastering control\n"));
772 			return_VOID;
773 		}
774 	} else {
775 		/*
776 		 * WBINVD should be set in fadt, for C3 state to be
777 		 * supported on when bm_check is not required.
778 		 */
779 		if (acpi_fadt.wb_invd != 1) {
780 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
781 					  "Cache invalidation should work properly"
782 					  " for C3 to be enabled on SMP systems\n"));
783 			return_VOID;
784 		}
785 		acpi_set_register(ACPI_BITREG_BUS_MASTER_RLD,
786 				  0, ACPI_MTX_DO_NOT_LOCK);
787 	}
788 
789 	/*
790 	 * Otherwise we've met all of our C3 requirements.
791 	 * Normalize the C3 latency to expidite policy.  Enable
792 	 * checking of bus mastering status (bm_check) so we can
793 	 * use this in our C3 policy
794 	 */
795 	cx->valid = 1;
796 	cx->latency_ticks = US_TO_PM_TIMER_TICKS(cx->latency);
797 
798 	return_VOID;
799 }
800 
801 static int acpi_processor_power_verify(struct acpi_processor *pr)
802 {
803 	unsigned int i;
804 	unsigned int working = 0;
805 
806 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
807 		struct acpi_processor_cx *cx = &pr->power.states[i];
808 
809 		switch (cx->type) {
810 		case ACPI_STATE_C1:
811 			cx->valid = 1;
812 			break;
813 
814 		case ACPI_STATE_C2:
815 			acpi_processor_power_verify_c2(cx);
816 			break;
817 
818 		case ACPI_STATE_C3:
819 			acpi_processor_power_verify_c3(pr, cx);
820 			break;
821 		}
822 
823 		if (cx->valid)
824 			working++;
825 	}
826 
827 	return (working);
828 }
829 
830 static int acpi_processor_get_power_info(struct acpi_processor *pr)
831 {
832 	unsigned int i;
833 	int result;
834 
835 	ACPI_FUNCTION_TRACE("acpi_processor_get_power_info");
836 
837 	/* NOTE: the idle thread may not be running while calling
838 	 * this function */
839 
840 	result = acpi_processor_get_power_info_cst(pr);
841 	if ((result) || (acpi_processor_power_verify(pr) < 2)) {
842 		result = acpi_processor_get_power_info_fadt(pr);
843 		if ((result) || (acpi_processor_power_verify(pr) < 2))
844 			result = acpi_processor_get_power_info_default_c1(pr);
845 	}
846 
847 	/*
848 	 * Set Default Policy
849 	 * ------------------
850 	 * Now that we know which states are supported, set the default
851 	 * policy.  Note that this policy can be changed dynamically
852 	 * (e.g. encourage deeper sleeps to conserve battery life when
853 	 * not on AC).
854 	 */
855 	result = acpi_processor_set_power_policy(pr);
856 	if (result)
857 		return_VALUE(result);
858 
859 	/*
860 	 * if one state of type C2 or C3 is available, mark this
861 	 * CPU as being "idle manageable"
862 	 */
863 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
864 		if (pr->power.states[i].valid) {
865 			pr->power.count = i;
866 			pr->flags.power = 1;
867 		}
868 	}
869 
870 	return_VALUE(0);
871 }
872 
873 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
874 {
875 	int result = 0;
876 
877 	ACPI_FUNCTION_TRACE("acpi_processor_cst_has_changed");
878 
879 	if (!pr)
880 		return_VALUE(-EINVAL);
881 
882 	if (nocst) {
883 		return_VALUE(-ENODEV);
884 	}
885 
886 	if (!pr->flags.power_setup_done)
887 		return_VALUE(-ENODEV);
888 
889 	/* Fall back to the default idle loop */
890 	pm_idle = pm_idle_save;
891 	synchronize_sched();	/* Relies on interrupts forcing exit from idle. */
892 
893 	pr->flags.power = 0;
894 	result = acpi_processor_get_power_info(pr);
895 	if ((pr->flags.power == 1) && (pr->flags.power_setup_done))
896 		pm_idle = acpi_processor_idle;
897 
898 	return_VALUE(result);
899 }
900 
901 /* proc interface */
902 
903 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
904 {
905 	struct acpi_processor *pr = (struct acpi_processor *)seq->private;
906 	unsigned int i;
907 
908 	ACPI_FUNCTION_TRACE("acpi_processor_power_seq_show");
909 
910 	if (!pr)
911 		goto end;
912 
913 	seq_printf(seq, "active state:            C%zd\n"
914 		   "max_cstate:              C%d\n"
915 		   "bus master activity:     %08x\n",
916 		   pr->power.state ? pr->power.state - pr->power.states : 0,
917 		   max_cstate, (unsigned)pr->power.bm_activity);
918 
919 	seq_puts(seq, "states:\n");
920 
921 	for (i = 1; i <= pr->power.count; i++) {
922 		seq_printf(seq, "   %cC%d:                  ",
923 			   (&pr->power.states[i] ==
924 			    pr->power.state ? '*' : ' '), i);
925 
926 		if (!pr->power.states[i].valid) {
927 			seq_puts(seq, "<not supported>\n");
928 			continue;
929 		}
930 
931 		switch (pr->power.states[i].type) {
932 		case ACPI_STATE_C1:
933 			seq_printf(seq, "type[C1] ");
934 			break;
935 		case ACPI_STATE_C2:
936 			seq_printf(seq, "type[C2] ");
937 			break;
938 		case ACPI_STATE_C3:
939 			seq_printf(seq, "type[C3] ");
940 			break;
941 		default:
942 			seq_printf(seq, "type[--] ");
943 			break;
944 		}
945 
946 		if (pr->power.states[i].promotion.state)
947 			seq_printf(seq, "promotion[C%zd] ",
948 				   (pr->power.states[i].promotion.state -
949 				    pr->power.states));
950 		else
951 			seq_puts(seq, "promotion[--] ");
952 
953 		if (pr->power.states[i].demotion.state)
954 			seq_printf(seq, "demotion[C%zd] ",
955 				   (pr->power.states[i].demotion.state -
956 				    pr->power.states));
957 		else
958 			seq_puts(seq, "demotion[--] ");
959 
960 		seq_printf(seq, "latency[%03d] usage[%08d]\n",
961 			   pr->power.states[i].latency,
962 			   pr->power.states[i].usage);
963 	}
964 
965       end:
966 	return_VALUE(0);
967 }
968 
969 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
970 {
971 	return single_open(file, acpi_processor_power_seq_show,
972 			   PDE(inode)->data);
973 }
974 
975 static struct file_operations acpi_processor_power_fops = {
976 	.open = acpi_processor_power_open_fs,
977 	.read = seq_read,
978 	.llseek = seq_lseek,
979 	.release = single_release,
980 };
981 
982 int acpi_processor_power_init(struct acpi_processor *pr,
983 			      struct acpi_device *device)
984 {
985 	acpi_status status = 0;
986 	static int first_run = 0;
987 	struct proc_dir_entry *entry = NULL;
988 	unsigned int i;
989 
990 	ACPI_FUNCTION_TRACE("acpi_processor_power_init");
991 
992 	if (!first_run) {
993 		dmi_check_system(processor_power_dmi_table);
994 		if (max_cstate < ACPI_C_STATES_MAX)
995 			printk(KERN_NOTICE
996 			       "ACPI: processor limited to max C-state %d\n",
997 			       max_cstate);
998 		first_run++;
999 	}
1000 
1001 	if (!pr)
1002 		return_VALUE(-EINVAL);
1003 
1004 	if (acpi_fadt.cst_cnt && !nocst) {
1005 		status =
1006 		    acpi_os_write_port(acpi_fadt.smi_cmd, acpi_fadt.cst_cnt, 8);
1007 		if (ACPI_FAILURE(status)) {
1008 			ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1009 					  "Notifying BIOS of _CST ability failed\n"));
1010 		}
1011 	}
1012 
1013 	acpi_processor_power_init_pdc(&(pr->power), pr->id);
1014 	acpi_processor_set_pdc(pr, pr->power.pdc);
1015 	acpi_processor_get_power_info(pr);
1016 
1017 	/*
1018 	 * Install the idle handler if processor power management is supported.
1019 	 * Note that we use previously set idle handler will be used on
1020 	 * platforms that only support C1.
1021 	 */
1022 	if ((pr->flags.power) && (!boot_option_idle_override)) {
1023 		printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1024 		for (i = 1; i <= pr->power.count; i++)
1025 			if (pr->power.states[i].valid)
1026 				printk(" C%d[C%d]", i,
1027 				       pr->power.states[i].type);
1028 		printk(")\n");
1029 
1030 		if (pr->id == 0) {
1031 			pm_idle_save = pm_idle;
1032 			pm_idle = acpi_processor_idle;
1033 		}
1034 	}
1035 
1036 	/* 'power' [R] */
1037 	entry = create_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1038 				  S_IRUGO, acpi_device_dir(device));
1039 	if (!entry)
1040 		ACPI_DEBUG_PRINT((ACPI_DB_ERROR,
1041 				  "Unable to create '%s' fs entry\n",
1042 				  ACPI_PROCESSOR_FILE_POWER));
1043 	else {
1044 		entry->proc_fops = &acpi_processor_power_fops;
1045 		entry->data = acpi_driver_data(device);
1046 		entry->owner = THIS_MODULE;
1047 	}
1048 
1049 	pr->flags.power_setup_done = 1;
1050 
1051 	return_VALUE(0);
1052 }
1053 
1054 int acpi_processor_power_exit(struct acpi_processor *pr,
1055 			      struct acpi_device *device)
1056 {
1057 	ACPI_FUNCTION_TRACE("acpi_processor_power_exit");
1058 
1059 	pr->flags.power_setup_done = 0;
1060 
1061 	if (acpi_device_dir(device))
1062 		remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1063 				  acpi_device_dir(device));
1064 
1065 	/* Unregister the idle handler when processor #0 is removed. */
1066 	if (pr->id == 0) {
1067 		pm_idle = pm_idle_save;
1068 
1069 		/*
1070 		 * We are about to unload the current idle thread pm callback
1071 		 * (pm_idle), Wait for all processors to update cached/local
1072 		 * copies of pm_idle before proceeding.
1073 		 */
1074 		cpu_idle_wait();
1075 	}
1076 
1077 	return_VALUE(0);
1078 }
1079