xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision baa7eb025ab14f3cba2e35c0a8648f9c9f01d24f)
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/kernel.h>
32 #include <linux/module.h>
33 #include <linux/init.h>
34 #include <linux/cpufreq.h>
35 #include <linux/slab.h>
36 #include <linux/acpi.h>
37 #include <linux/dmi.h>
38 #include <linux/moduleparam.h>
39 #include <linux/sched.h>	/* need_resched() */
40 #include <linux/pm_qos_params.h>
41 #include <linux/clockchips.h>
42 #include <linux/cpuidle.h>
43 #include <linux/irqflags.h>
44 
45 /*
46  * Include the apic definitions for x86 to have the APIC timer related defines
47  * available also for UP (on SMP it gets magically included via linux/smp.h).
48  * asm/acpi.h is not an option, as it would require more include magic. Also
49  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
50  */
51 #ifdef CONFIG_X86
52 #include <asm/apic.h>
53 #endif
54 
55 #include <asm/io.h>
56 #include <asm/uaccess.h>
57 
58 #include <acpi/acpi_bus.h>
59 #include <acpi/processor.h>
60 #include <asm/processor.h>
61 
62 #define PREFIX "ACPI: "
63 
64 #define ACPI_PROCESSOR_CLASS            "processor"
65 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
66 ACPI_MODULE_NAME("processor_idle");
67 #define PM_TIMER_TICK_NS		(1000000000ULL/PM_TIMER_FREQUENCY)
68 #define C2_OVERHEAD			1	/* 1us */
69 #define C3_OVERHEAD			1	/* 1us */
70 #define PM_TIMER_TICKS_TO_US(p)		(((p) * 1000)/(PM_TIMER_FREQUENCY/1000))
71 
72 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
73 module_param(max_cstate, uint, 0000);
74 static unsigned int nocst __read_mostly;
75 module_param(nocst, uint, 0000);
76 static int bm_check_disable __read_mostly;
77 module_param(bm_check_disable, uint, 0000);
78 
79 static unsigned int latency_factor __read_mostly = 2;
80 module_param(latency_factor, uint, 0644);
81 
82 /*
83  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
84  * For now disable this. Probably a bug somewhere else.
85  *
86  * To skip this limit, boot/load with a large max_cstate limit.
87  */
88 static int set_max_cstate(const struct dmi_system_id *id)
89 {
90 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
91 		return 0;
92 
93 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
94 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
95 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
96 
97 	max_cstate = (long)id->driver_data;
98 
99 	return 0;
100 }
101 
102 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
103    callers to only run once -AK */
104 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
105 	{ set_max_cstate, "Clevo 5600D", {
106 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
107 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
108 	 (void *)2},
109 	{ set_max_cstate, "Pavilion zv5000", {
110 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
111 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
112 	 (void *)1},
113 	{ set_max_cstate, "Asus L8400B", {
114 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
115 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
116 	 (void *)1},
117 	{},
118 };
119 
120 
121 /*
122  * Callers should disable interrupts before the call and enable
123  * interrupts after return.
124  */
125 static void acpi_safe_halt(void)
126 {
127 	current_thread_info()->status &= ~TS_POLLING;
128 	/*
129 	 * TS_POLLING-cleared state must be visible before we
130 	 * test NEED_RESCHED:
131 	 */
132 	smp_mb();
133 	if (!need_resched()) {
134 		safe_halt();
135 		local_irq_disable();
136 	}
137 	current_thread_info()->status |= TS_POLLING;
138 }
139 
140 #ifdef ARCH_APICTIMER_STOPS_ON_C3
141 
142 /*
143  * Some BIOS implementations switch to C3 in the published C2 state.
144  * This seems to be a common problem on AMD boxen, but other vendors
145  * are affected too. We pick the most conservative approach: we assume
146  * that the local APIC stops in both C2 and C3.
147  */
148 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
149 				   struct acpi_processor_cx *cx)
150 {
151 	struct acpi_processor_power *pwr = &pr->power;
152 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
153 
154 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
155 		return;
156 
157 	if (c1e_detected)
158 		type = ACPI_STATE_C1;
159 
160 	/*
161 	 * Check, if one of the previous states already marked the lapic
162 	 * unstable
163 	 */
164 	if (pwr->timer_broadcast_on_state < state)
165 		return;
166 
167 	if (cx->type >= type)
168 		pr->power.timer_broadcast_on_state = state;
169 }
170 
171 static void __lapic_timer_propagate_broadcast(void *arg)
172 {
173 	struct acpi_processor *pr = (struct acpi_processor *) arg;
174 	unsigned long reason;
175 
176 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
177 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
178 
179 	clockevents_notify(reason, &pr->id);
180 }
181 
182 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
183 {
184 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
185 				 (void *)pr, 1);
186 }
187 
188 /* Power(C) State timer broadcast control */
189 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
190 				       struct acpi_processor_cx *cx,
191 				       int broadcast)
192 {
193 	int state = cx - pr->power.states;
194 
195 	if (state >= pr->power.timer_broadcast_on_state) {
196 		unsigned long reason;
197 
198 		reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
199 			CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
200 		clockevents_notify(reason, &pr->id);
201 	}
202 }
203 
204 #else
205 
206 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
207 				   struct acpi_processor_cx *cstate) { }
208 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
209 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
210 				       struct acpi_processor_cx *cx,
211 				       int broadcast)
212 {
213 }
214 
215 #endif
216 
217 /*
218  * Suspend / resume control
219  */
220 static int acpi_idle_suspend;
221 static u32 saved_bm_rld;
222 
223 static void acpi_idle_bm_rld_save(void)
224 {
225 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
226 }
227 static void acpi_idle_bm_rld_restore(void)
228 {
229 	u32 resumed_bm_rld;
230 
231 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
232 
233 	if (resumed_bm_rld != saved_bm_rld)
234 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
235 }
236 
237 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
238 {
239 	if (acpi_idle_suspend == 1)
240 		return 0;
241 
242 	acpi_idle_bm_rld_save();
243 	acpi_idle_suspend = 1;
244 	return 0;
245 }
246 
247 int acpi_processor_resume(struct acpi_device * device)
248 {
249 	if (acpi_idle_suspend == 0)
250 		return 0;
251 
252 	acpi_idle_bm_rld_restore();
253 	acpi_idle_suspend = 0;
254 	return 0;
255 }
256 
257 #if defined(CONFIG_X86)
258 static void tsc_check_state(int state)
259 {
260 	switch (boot_cpu_data.x86_vendor) {
261 	case X86_VENDOR_AMD:
262 	case X86_VENDOR_INTEL:
263 		/*
264 		 * AMD Fam10h TSC will tick in all
265 		 * C/P/S0/S1 states when this bit is set.
266 		 */
267 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
268 			return;
269 
270 		/*FALL THROUGH*/
271 	default:
272 		/* TSC could halt in idle, so notify users */
273 		if (state > ACPI_STATE_C1)
274 			mark_tsc_unstable("TSC halts in idle");
275 	}
276 }
277 #else
278 static void tsc_check_state(int state) { return; }
279 #endif
280 
281 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
282 {
283 
284 	if (!pr)
285 		return -EINVAL;
286 
287 	if (!pr->pblk)
288 		return -ENODEV;
289 
290 	/* if info is obtained from pblk/fadt, type equals state */
291 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
292 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
293 
294 #ifndef CONFIG_HOTPLUG_CPU
295 	/*
296 	 * Check for P_LVL2_UP flag before entering C2 and above on
297 	 * an SMP system.
298 	 */
299 	if ((num_online_cpus() > 1) &&
300 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
301 		return -ENODEV;
302 #endif
303 
304 	/* determine C2 and C3 address from pblk */
305 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
306 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
307 
308 	/* determine latencies from FADT */
309 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
310 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
311 
312 	/*
313 	 * FADT specified C2 latency must be less than or equal to
314 	 * 100 microseconds.
315 	 */
316 	if (acpi_gbl_FADT.C2latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
317 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
318 			"C2 latency too large [%d]\n", acpi_gbl_FADT.C2latency));
319 		/* invalidate C2 */
320 		pr->power.states[ACPI_STATE_C2].address = 0;
321 	}
322 
323 	/*
324 	 * FADT supplied C3 latency must be less than or equal to
325 	 * 1000 microseconds.
326 	 */
327 	if (acpi_gbl_FADT.C3latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
328 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
329 			"C3 latency too large [%d]\n", acpi_gbl_FADT.C3latency));
330 		/* invalidate C3 */
331 		pr->power.states[ACPI_STATE_C3].address = 0;
332 	}
333 
334 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
335 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
336 			  pr->power.states[ACPI_STATE_C2].address,
337 			  pr->power.states[ACPI_STATE_C3].address));
338 
339 	return 0;
340 }
341 
342 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
343 {
344 	if (!pr->power.states[ACPI_STATE_C1].valid) {
345 		/* set the first C-State to C1 */
346 		/* all processors need to support C1 */
347 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
348 		pr->power.states[ACPI_STATE_C1].valid = 1;
349 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
350 	}
351 	/* the C0 state only exists as a filler in our array */
352 	pr->power.states[ACPI_STATE_C0].valid = 1;
353 	return 0;
354 }
355 
356 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
357 {
358 	acpi_status status = 0;
359 	u64 count;
360 	int current_count;
361 	int i;
362 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
363 	union acpi_object *cst;
364 
365 
366 	if (nocst)
367 		return -ENODEV;
368 
369 	current_count = 0;
370 
371 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
372 	if (ACPI_FAILURE(status)) {
373 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
374 		return -ENODEV;
375 	}
376 
377 	cst = buffer.pointer;
378 
379 	/* There must be at least 2 elements */
380 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
381 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
382 		status = -EFAULT;
383 		goto end;
384 	}
385 
386 	count = cst->package.elements[0].integer.value;
387 
388 	/* Validate number of power states. */
389 	if (count < 1 || count != cst->package.count - 1) {
390 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
391 		status = -EFAULT;
392 		goto end;
393 	}
394 
395 	/* Tell driver that at least _CST is supported. */
396 	pr->flags.has_cst = 1;
397 
398 	for (i = 1; i <= count; i++) {
399 		union acpi_object *element;
400 		union acpi_object *obj;
401 		struct acpi_power_register *reg;
402 		struct acpi_processor_cx cx;
403 
404 		memset(&cx, 0, sizeof(cx));
405 
406 		element = &(cst->package.elements[i]);
407 		if (element->type != ACPI_TYPE_PACKAGE)
408 			continue;
409 
410 		if (element->package.count != 4)
411 			continue;
412 
413 		obj = &(element->package.elements[0]);
414 
415 		if (obj->type != ACPI_TYPE_BUFFER)
416 			continue;
417 
418 		reg = (struct acpi_power_register *)obj->buffer.pointer;
419 
420 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
421 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
422 			continue;
423 
424 		/* There should be an easy way to extract an integer... */
425 		obj = &(element->package.elements[1]);
426 		if (obj->type != ACPI_TYPE_INTEGER)
427 			continue;
428 
429 		cx.type = obj->integer.value;
430 		/*
431 		 * Some buggy BIOSes won't list C1 in _CST -
432 		 * Let acpi_processor_get_power_info_default() handle them later
433 		 */
434 		if (i == 1 && cx.type != ACPI_STATE_C1)
435 			current_count++;
436 
437 		cx.address = reg->address;
438 		cx.index = current_count + 1;
439 
440 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
441 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
442 			if (acpi_processor_ffh_cstate_probe
443 					(pr->id, &cx, reg) == 0) {
444 				cx.entry_method = ACPI_CSTATE_FFH;
445 			} else if (cx.type == ACPI_STATE_C1) {
446 				/*
447 				 * C1 is a special case where FIXED_HARDWARE
448 				 * can be handled in non-MWAIT way as well.
449 				 * In that case, save this _CST entry info.
450 				 * Otherwise, ignore this info and continue.
451 				 */
452 				cx.entry_method = ACPI_CSTATE_HALT;
453 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
454 			} else {
455 				continue;
456 			}
457 			if (cx.type == ACPI_STATE_C1 &&
458 					(idle_halt || idle_nomwait)) {
459 				/*
460 				 * In most cases the C1 space_id obtained from
461 				 * _CST object is FIXED_HARDWARE access mode.
462 				 * But when the option of idle=halt is added,
463 				 * the entry_method type should be changed from
464 				 * CSTATE_FFH to CSTATE_HALT.
465 				 * When the option of idle=nomwait is added,
466 				 * the C1 entry_method type should be
467 				 * CSTATE_HALT.
468 				 */
469 				cx.entry_method = ACPI_CSTATE_HALT;
470 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
471 			}
472 		} else {
473 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
474 				 cx.address);
475 		}
476 
477 		if (cx.type == ACPI_STATE_C1) {
478 			cx.valid = 1;
479 		}
480 
481 		obj = &(element->package.elements[2]);
482 		if (obj->type != ACPI_TYPE_INTEGER)
483 			continue;
484 
485 		cx.latency = obj->integer.value;
486 
487 		obj = &(element->package.elements[3]);
488 		if (obj->type != ACPI_TYPE_INTEGER)
489 			continue;
490 
491 		cx.power = obj->integer.value;
492 
493 		current_count++;
494 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
495 
496 		/*
497 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
498 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
499 		 */
500 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
501 			printk(KERN_WARNING
502 			       "Limiting number of power states to max (%d)\n",
503 			       ACPI_PROCESSOR_MAX_POWER);
504 			printk(KERN_WARNING
505 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
506 			break;
507 		}
508 	}
509 
510 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
511 			  current_count));
512 
513 	/* Validate number of power states discovered */
514 	if (current_count < 2)
515 		status = -EFAULT;
516 
517       end:
518 	kfree(buffer.pointer);
519 
520 	return status;
521 }
522 
523 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
524 					   struct acpi_processor_cx *cx)
525 {
526 	static int bm_check_flag = -1;
527 	static int bm_control_flag = -1;
528 
529 
530 	if (!cx->address)
531 		return;
532 
533 	/*
534 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
535 	 * DMA transfers are used by any ISA device to avoid livelock.
536 	 * Note that we could disable Type-F DMA (as recommended by
537 	 * the erratum), but this is known to disrupt certain ISA
538 	 * devices thus we take the conservative approach.
539 	 */
540 	else if (errata.piix4.fdma) {
541 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
542 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
543 		return;
544 	}
545 
546 	/* All the logic here assumes flags.bm_check is same across all CPUs */
547 	if (bm_check_flag == -1) {
548 		/* Determine whether bm_check is needed based on CPU  */
549 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
550 		bm_check_flag = pr->flags.bm_check;
551 		bm_control_flag = pr->flags.bm_control;
552 	} else {
553 		pr->flags.bm_check = bm_check_flag;
554 		pr->flags.bm_control = bm_control_flag;
555 	}
556 
557 	if (pr->flags.bm_check) {
558 		if (!pr->flags.bm_control) {
559 			if (pr->flags.has_cst != 1) {
560 				/* bus mastering control is necessary */
561 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
562 					"C3 support requires BM control\n"));
563 				return;
564 			} else {
565 				/* Here we enter C3 without bus mastering */
566 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
567 					"C3 support without BM control\n"));
568 			}
569 		}
570 	} else {
571 		/*
572 		 * WBINVD should be set in fadt, for C3 state to be
573 		 * supported on when bm_check is not required.
574 		 */
575 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
576 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
577 					  "Cache invalidation should work properly"
578 					  " for C3 to be enabled on SMP systems\n"));
579 			return;
580 		}
581 	}
582 
583 	/*
584 	 * Otherwise we've met all of our C3 requirements.
585 	 * Normalize the C3 latency to expidite policy.  Enable
586 	 * checking of bus mastering status (bm_check) so we can
587 	 * use this in our C3 policy
588 	 */
589 	cx->valid = 1;
590 
591 	cx->latency_ticks = cx->latency;
592 	/*
593 	 * On older chipsets, BM_RLD needs to be set
594 	 * in order for Bus Master activity to wake the
595 	 * system from C3.  Newer chipsets handle DMA
596 	 * during C3 automatically and BM_RLD is a NOP.
597 	 * In either case, the proper way to
598 	 * handle BM_RLD is to set it and leave it set.
599 	 */
600 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
601 
602 	return;
603 }
604 
605 static int acpi_processor_power_verify(struct acpi_processor *pr)
606 {
607 	unsigned int i;
608 	unsigned int working = 0;
609 
610 	pr->power.timer_broadcast_on_state = INT_MAX;
611 
612 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
613 		struct acpi_processor_cx *cx = &pr->power.states[i];
614 
615 		switch (cx->type) {
616 		case ACPI_STATE_C1:
617 			cx->valid = 1;
618 			break;
619 
620 		case ACPI_STATE_C2:
621 			if (!cx->address)
622 				break;
623 			cx->valid = 1;
624 			cx->latency_ticks = cx->latency; /* Normalize latency */
625 			break;
626 
627 		case ACPI_STATE_C3:
628 			acpi_processor_power_verify_c3(pr, cx);
629 			break;
630 		}
631 		if (!cx->valid)
632 			continue;
633 
634 		lapic_timer_check_state(i, pr, cx);
635 		tsc_check_state(cx->type);
636 		working++;
637 	}
638 
639 	lapic_timer_propagate_broadcast(pr);
640 
641 	return (working);
642 }
643 
644 static int acpi_processor_get_power_info(struct acpi_processor *pr)
645 {
646 	unsigned int i;
647 	int result;
648 
649 
650 	/* NOTE: the idle thread may not be running while calling
651 	 * this function */
652 
653 	/* Zero initialize all the C-states info. */
654 	memset(pr->power.states, 0, sizeof(pr->power.states));
655 
656 	result = acpi_processor_get_power_info_cst(pr);
657 	if (result == -ENODEV)
658 		result = acpi_processor_get_power_info_fadt(pr);
659 
660 	if (result)
661 		return result;
662 
663 	acpi_processor_get_power_info_default(pr);
664 
665 	pr->power.count = acpi_processor_power_verify(pr);
666 
667 	/*
668 	 * if one state of type C2 or C3 is available, mark this
669 	 * CPU as being "idle manageable"
670 	 */
671 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
672 		if (pr->power.states[i].valid) {
673 			pr->power.count = i;
674 			if (pr->power.states[i].type >= ACPI_STATE_C2)
675 				pr->flags.power = 1;
676 		}
677 	}
678 
679 	return 0;
680 }
681 
682 /**
683  * acpi_idle_bm_check - checks if bus master activity was detected
684  */
685 static int acpi_idle_bm_check(void)
686 {
687 	u32 bm_status = 0;
688 
689 	if (bm_check_disable)
690 		return 0;
691 
692 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
693 	if (bm_status)
694 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
695 	/*
696 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
697 	 * the true state of bus mastering activity; forcing us to
698 	 * manually check the BMIDEA bit of each IDE channel.
699 	 */
700 	else if (errata.piix4.bmisx) {
701 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
702 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
703 			bm_status = 1;
704 	}
705 	return bm_status;
706 }
707 
708 /**
709  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
710  * @cx: cstate data
711  *
712  * Caller disables interrupt before call and enables interrupt after return.
713  */
714 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
715 {
716 	/* Don't trace irqs off for idle */
717 	stop_critical_timings();
718 	if (cx->entry_method == ACPI_CSTATE_FFH) {
719 		/* Call into architectural FFH based C-state */
720 		acpi_processor_ffh_cstate_enter(cx);
721 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
722 		acpi_safe_halt();
723 	} else {
724 		/* IO port based C-state */
725 		inb(cx->address);
726 		/* Dummy wait op - must do something useless after P_LVL2 read
727 		   because chipsets cannot guarantee that STPCLK# signal
728 		   gets asserted in time to freeze execution properly. */
729 		inl(acpi_gbl_FADT.xpm_timer_block.address);
730 	}
731 	start_critical_timings();
732 }
733 
734 /**
735  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
736  * @dev: the target CPU
737  * @state: the state data
738  *
739  * This is equivalent to the HALT instruction.
740  */
741 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
742 			      struct cpuidle_state *state)
743 {
744 	ktime_t  kt1, kt2;
745 	s64 idle_time;
746 	struct acpi_processor *pr;
747 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
748 
749 	pr = __get_cpu_var(processors);
750 
751 	if (unlikely(!pr))
752 		return 0;
753 
754 	local_irq_disable();
755 
756 	/* Do not access any ACPI IO ports in suspend path */
757 	if (acpi_idle_suspend) {
758 		local_irq_enable();
759 		cpu_relax();
760 		return 0;
761 	}
762 
763 	lapic_timer_state_broadcast(pr, cx, 1);
764 	kt1 = ktime_get_real();
765 	acpi_idle_do_entry(cx);
766 	kt2 = ktime_get_real();
767 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
768 
769 	local_irq_enable();
770 	cx->usage++;
771 	lapic_timer_state_broadcast(pr, cx, 0);
772 
773 	return idle_time;
774 }
775 
776 /**
777  * acpi_idle_enter_simple - enters an ACPI state without BM handling
778  * @dev: the target CPU
779  * @state: the state data
780  */
781 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
782 				  struct cpuidle_state *state)
783 {
784 	struct acpi_processor *pr;
785 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
786 	ktime_t  kt1, kt2;
787 	s64 idle_time_ns;
788 	s64 idle_time;
789 
790 	pr = __get_cpu_var(processors);
791 
792 	if (unlikely(!pr))
793 		return 0;
794 
795 	if (acpi_idle_suspend)
796 		return(acpi_idle_enter_c1(dev, state));
797 
798 	local_irq_disable();
799 
800 	if (cx->entry_method != ACPI_CSTATE_FFH) {
801 		current_thread_info()->status &= ~TS_POLLING;
802 		/*
803 		 * TS_POLLING-cleared state must be visible before we test
804 		 * NEED_RESCHED:
805 		 */
806 		smp_mb();
807 
808 		if (unlikely(need_resched())) {
809 			current_thread_info()->status |= TS_POLLING;
810 			local_irq_enable();
811 			return 0;
812 		}
813 	}
814 
815 	/*
816 	 * Must be done before busmaster disable as we might need to
817 	 * access HPET !
818 	 */
819 	lapic_timer_state_broadcast(pr, cx, 1);
820 
821 	if (cx->type == ACPI_STATE_C3)
822 		ACPI_FLUSH_CPU_CACHE();
823 
824 	kt1 = ktime_get_real();
825 	/* Tell the scheduler that we are going deep-idle: */
826 	sched_clock_idle_sleep_event();
827 	acpi_idle_do_entry(cx);
828 	kt2 = ktime_get_real();
829 	idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
830 	idle_time = idle_time_ns;
831 	do_div(idle_time, NSEC_PER_USEC);
832 
833 	/* Tell the scheduler how much we idled: */
834 	sched_clock_idle_wakeup_event(idle_time_ns);
835 
836 	local_irq_enable();
837 	if (cx->entry_method != ACPI_CSTATE_FFH)
838 		current_thread_info()->status |= TS_POLLING;
839 
840 	cx->usage++;
841 
842 	lapic_timer_state_broadcast(pr, cx, 0);
843 	cx->time += idle_time;
844 	return idle_time;
845 }
846 
847 static int c3_cpu_count;
848 static DEFINE_SPINLOCK(c3_lock);
849 
850 /**
851  * acpi_idle_enter_bm - enters C3 with proper BM handling
852  * @dev: the target CPU
853  * @state: the state data
854  *
855  * If BM is detected, the deepest non-C3 idle state is entered instead.
856  */
857 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
858 			      struct cpuidle_state *state)
859 {
860 	struct acpi_processor *pr;
861 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
862 	ktime_t  kt1, kt2;
863 	s64 idle_time_ns;
864 	s64 idle_time;
865 
866 
867 	pr = __get_cpu_var(processors);
868 
869 	if (unlikely(!pr))
870 		return 0;
871 
872 	if (acpi_idle_suspend)
873 		return(acpi_idle_enter_c1(dev, state));
874 
875 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
876 		if (dev->safe_state) {
877 			dev->last_state = dev->safe_state;
878 			return dev->safe_state->enter(dev, dev->safe_state);
879 		} else {
880 			local_irq_disable();
881 			acpi_safe_halt();
882 			local_irq_enable();
883 			return 0;
884 		}
885 	}
886 
887 	local_irq_disable();
888 
889 	if (cx->entry_method != ACPI_CSTATE_FFH) {
890 		current_thread_info()->status &= ~TS_POLLING;
891 		/*
892 		 * TS_POLLING-cleared state must be visible before we test
893 		 * NEED_RESCHED:
894 		 */
895 		smp_mb();
896 
897 		if (unlikely(need_resched())) {
898 			current_thread_info()->status |= TS_POLLING;
899 			local_irq_enable();
900 			return 0;
901 		}
902 	}
903 
904 	acpi_unlazy_tlb(smp_processor_id());
905 
906 	/* Tell the scheduler that we are going deep-idle: */
907 	sched_clock_idle_sleep_event();
908 	/*
909 	 * Must be done before busmaster disable as we might need to
910 	 * access HPET !
911 	 */
912 	lapic_timer_state_broadcast(pr, cx, 1);
913 
914 	kt1 = ktime_get_real();
915 	/*
916 	 * disable bus master
917 	 * bm_check implies we need ARB_DIS
918 	 * !bm_check implies we need cache flush
919 	 * bm_control implies whether we can do ARB_DIS
920 	 *
921 	 * That leaves a case where bm_check is set and bm_control is
922 	 * not set. In that case we cannot do much, we enter C3
923 	 * without doing anything.
924 	 */
925 	if (pr->flags.bm_check && pr->flags.bm_control) {
926 		spin_lock(&c3_lock);
927 		c3_cpu_count++;
928 		/* Disable bus master arbitration when all CPUs are in C3 */
929 		if (c3_cpu_count == num_online_cpus())
930 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
931 		spin_unlock(&c3_lock);
932 	} else if (!pr->flags.bm_check) {
933 		ACPI_FLUSH_CPU_CACHE();
934 	}
935 
936 	acpi_idle_do_entry(cx);
937 
938 	/* Re-enable bus master arbitration */
939 	if (pr->flags.bm_check && pr->flags.bm_control) {
940 		spin_lock(&c3_lock);
941 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
942 		c3_cpu_count--;
943 		spin_unlock(&c3_lock);
944 	}
945 	kt2 = ktime_get_real();
946 	idle_time_ns = ktime_to_ns(ktime_sub(kt2, kt1));
947 	idle_time = idle_time_ns;
948 	do_div(idle_time, NSEC_PER_USEC);
949 
950 	/* Tell the scheduler how much we idled: */
951 	sched_clock_idle_wakeup_event(idle_time_ns);
952 
953 	local_irq_enable();
954 	if (cx->entry_method != ACPI_CSTATE_FFH)
955 		current_thread_info()->status |= TS_POLLING;
956 
957 	cx->usage++;
958 
959 	lapic_timer_state_broadcast(pr, cx, 0);
960 	cx->time += idle_time;
961 	return idle_time;
962 }
963 
964 struct cpuidle_driver acpi_idle_driver = {
965 	.name =		"acpi_idle",
966 	.owner =	THIS_MODULE,
967 };
968 
969 /**
970  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
971  * @pr: the ACPI processor
972  */
973 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
974 {
975 	int i, count = CPUIDLE_DRIVER_STATE_START;
976 	struct acpi_processor_cx *cx;
977 	struct cpuidle_state *state;
978 	struct cpuidle_device *dev = &pr->power.dev;
979 
980 	if (!pr->flags.power_setup_done)
981 		return -EINVAL;
982 
983 	if (pr->flags.power == 0) {
984 		return -EINVAL;
985 	}
986 
987 	dev->cpu = pr->id;
988 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
989 		dev->states[i].name[0] = '\0';
990 		dev->states[i].desc[0] = '\0';
991 	}
992 
993 	if (max_cstate == 0)
994 		max_cstate = 1;
995 
996 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
997 		cx = &pr->power.states[i];
998 		state = &dev->states[count];
999 
1000 		if (!cx->valid)
1001 			continue;
1002 
1003 #ifdef CONFIG_HOTPLUG_CPU
1004 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1005 		    !pr->flags.has_cst &&
1006 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1007 			continue;
1008 #endif
1009 		cpuidle_set_statedata(state, cx);
1010 
1011 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1012 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1013 		state->exit_latency = cx->latency;
1014 		state->target_residency = cx->latency * latency_factor;
1015 
1016 		state->flags = 0;
1017 		switch (cx->type) {
1018 			case ACPI_STATE_C1:
1019 			state->flags |= CPUIDLE_FLAG_SHALLOW;
1020 			if (cx->entry_method == ACPI_CSTATE_FFH)
1021 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1022 
1023 			state->enter = acpi_idle_enter_c1;
1024 			dev->safe_state = state;
1025 			break;
1026 
1027 			case ACPI_STATE_C2:
1028 			state->flags |= CPUIDLE_FLAG_BALANCED;
1029 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1030 			state->enter = acpi_idle_enter_simple;
1031 			dev->safe_state = state;
1032 			break;
1033 
1034 			case ACPI_STATE_C3:
1035 			state->flags |= CPUIDLE_FLAG_DEEP;
1036 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1037 			state->flags |= CPUIDLE_FLAG_CHECK_BM;
1038 			state->enter = pr->flags.bm_check ?
1039 					acpi_idle_enter_bm :
1040 					acpi_idle_enter_simple;
1041 			break;
1042 		}
1043 
1044 		count++;
1045 		if (count == CPUIDLE_STATE_MAX)
1046 			break;
1047 	}
1048 
1049 	dev->state_count = count;
1050 
1051 	if (!count)
1052 		return -EINVAL;
1053 
1054 	return 0;
1055 }
1056 
1057 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1058 {
1059 	int ret = 0;
1060 
1061 	if (boot_option_idle_override)
1062 		return 0;
1063 
1064 	if (!pr)
1065 		return -EINVAL;
1066 
1067 	if (nocst) {
1068 		return -ENODEV;
1069 	}
1070 
1071 	if (!pr->flags.power_setup_done)
1072 		return -ENODEV;
1073 
1074 	cpuidle_pause_and_lock();
1075 	cpuidle_disable_device(&pr->power.dev);
1076 	acpi_processor_get_power_info(pr);
1077 	if (pr->flags.power) {
1078 		acpi_processor_setup_cpuidle(pr);
1079 		ret = cpuidle_enable_device(&pr->power.dev);
1080 	}
1081 	cpuidle_resume_and_unlock();
1082 
1083 	return ret;
1084 }
1085 
1086 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1087 			      struct acpi_device *device)
1088 {
1089 	acpi_status status = 0;
1090 	static int first_run;
1091 
1092 	if (boot_option_idle_override)
1093 		return 0;
1094 
1095 	if (!first_run) {
1096 		if (idle_halt) {
1097 			/*
1098 			 * When the boot option of "idle=halt" is added, halt
1099 			 * is used for CPU IDLE.
1100 			 * In such case C2/C3 is meaningless. So the max_cstate
1101 			 * is set to one.
1102 			 */
1103 			max_cstate = 1;
1104 		}
1105 		dmi_check_system(processor_power_dmi_table);
1106 		max_cstate = acpi_processor_cstate_check(max_cstate);
1107 		if (max_cstate < ACPI_C_STATES_MAX)
1108 			printk(KERN_NOTICE
1109 			       "ACPI: processor limited to max C-state %d\n",
1110 			       max_cstate);
1111 		first_run++;
1112 	}
1113 
1114 	if (!pr)
1115 		return -EINVAL;
1116 
1117 	if (acpi_gbl_FADT.cst_control && !nocst) {
1118 		status =
1119 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1120 		if (ACPI_FAILURE(status)) {
1121 			ACPI_EXCEPTION((AE_INFO, status,
1122 					"Notifying BIOS of _CST ability failed"));
1123 		}
1124 	}
1125 
1126 	acpi_processor_get_power_info(pr);
1127 	pr->flags.power_setup_done = 1;
1128 
1129 	/*
1130 	 * Install the idle handler if processor power management is supported.
1131 	 * Note that we use previously set idle handler will be used on
1132 	 * platforms that only support C1.
1133 	 */
1134 	if (pr->flags.power) {
1135 		acpi_processor_setup_cpuidle(pr);
1136 		if (cpuidle_register_device(&pr->power.dev))
1137 			return -EIO;
1138 	}
1139 	return 0;
1140 }
1141 
1142 int acpi_processor_power_exit(struct acpi_processor *pr,
1143 			      struct acpi_device *device)
1144 {
1145 	if (boot_option_idle_override)
1146 		return 0;
1147 
1148 	cpuidle_unregister_device(&pr->power.dev);
1149 	pr->flags.power_setup_done = 0;
1150 
1151 	return 0;
1152 }
1153