xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision 39b6f3aa)
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 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
100    callers to only run once -AK */
101 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
102 	{ set_max_cstate, "Clevo 5600D", {
103 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
104 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
105 	 (void *)2},
106 	{ set_max_cstate, "Pavilion zv5000", {
107 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
108 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
109 	 (void *)1},
110 	{ set_max_cstate, "Asus L8400B", {
111 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
112 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
113 	 (void *)1},
114 	{},
115 };
116 
117 
118 /*
119  * Callers should disable interrupts before the call and enable
120  * interrupts after return.
121  */
122 static void acpi_safe_halt(void)
123 {
124 	current_thread_info()->status &= ~TS_POLLING;
125 	/*
126 	 * TS_POLLING-cleared state must be visible before we
127 	 * test NEED_RESCHED:
128 	 */
129 	smp_mb();
130 	if (!need_resched()) {
131 		safe_halt();
132 		local_irq_disable();
133 	}
134 	current_thread_info()->status |= TS_POLLING;
135 }
136 
137 #ifdef ARCH_APICTIMER_STOPS_ON_C3
138 
139 /*
140  * Some BIOS implementations switch to C3 in the published C2 state.
141  * This seems to be a common problem on AMD boxen, but other vendors
142  * are affected too. We pick the most conservative approach: we assume
143  * that the local APIC stops in both C2 and C3.
144  */
145 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
146 				   struct acpi_processor_cx *cx)
147 {
148 	struct acpi_processor_power *pwr = &pr->power;
149 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
150 
151 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
152 		return;
153 
154 	if (amd_e400_c1e_detected)
155 		type = ACPI_STATE_C1;
156 
157 	/*
158 	 * Check, if one of the previous states already marked the lapic
159 	 * unstable
160 	 */
161 	if (pwr->timer_broadcast_on_state < state)
162 		return;
163 
164 	if (cx->type >= type)
165 		pr->power.timer_broadcast_on_state = state;
166 }
167 
168 static void __lapic_timer_propagate_broadcast(void *arg)
169 {
170 	struct acpi_processor *pr = (struct acpi_processor *) arg;
171 	unsigned long reason;
172 
173 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
174 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
175 
176 	clockevents_notify(reason, &pr->id);
177 }
178 
179 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
180 {
181 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
182 				 (void *)pr, 1);
183 }
184 
185 /* Power(C) State timer broadcast control */
186 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
187 				       struct acpi_processor_cx *cx,
188 				       int broadcast)
189 {
190 	int state = cx - pr->power.states;
191 
192 	if (state >= pr->power.timer_broadcast_on_state) {
193 		unsigned long reason;
194 
195 		reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
196 			CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
197 		clockevents_notify(reason, &pr->id);
198 	}
199 }
200 
201 #else
202 
203 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
204 				   struct acpi_processor_cx *cstate) { }
205 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
206 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
207 				       struct acpi_processor_cx *cx,
208 				       int broadcast)
209 {
210 }
211 
212 #endif
213 
214 #ifdef CONFIG_PM_SLEEP
215 static u32 saved_bm_rld;
216 
217 int acpi_processor_suspend(void)
218 {
219 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
220 	return 0;
221 }
222 
223 void acpi_processor_resume(void)
224 {
225 	u32 resumed_bm_rld;
226 
227 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
228 	if (resumed_bm_rld == saved_bm_rld)
229 		return;
230 
231 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
232 }
233 
234 static struct syscore_ops acpi_processor_syscore_ops = {
235 	.suspend = acpi_processor_suspend,
236 	.resume = acpi_processor_resume,
237 };
238 
239 void acpi_processor_syscore_init(void)
240 {
241 	register_syscore_ops(&acpi_processor_syscore_ops);
242 }
243 
244 void acpi_processor_syscore_exit(void)
245 {
246 	unregister_syscore_ops(&acpi_processor_syscore_ops);
247 }
248 #endif /* CONFIG_PM_SLEEP */
249 
250 #if defined(CONFIG_X86)
251 static void tsc_check_state(int state)
252 {
253 	switch (boot_cpu_data.x86_vendor) {
254 	case X86_VENDOR_AMD:
255 	case X86_VENDOR_INTEL:
256 		/*
257 		 * AMD Fam10h TSC will tick in all
258 		 * C/P/S0/S1 states when this bit is set.
259 		 */
260 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
261 			return;
262 
263 		/*FALL THROUGH*/
264 	default:
265 		/* TSC could halt in idle, so notify users */
266 		if (state > ACPI_STATE_C1)
267 			mark_tsc_unstable("TSC halts in idle");
268 	}
269 }
270 #else
271 static void tsc_check_state(int state) { return; }
272 #endif
273 
274 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
275 {
276 
277 	if (!pr)
278 		return -EINVAL;
279 
280 	if (!pr->pblk)
281 		return -ENODEV;
282 
283 	/* if info is obtained from pblk/fadt, type equals state */
284 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
285 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
286 
287 #ifndef CONFIG_HOTPLUG_CPU
288 	/*
289 	 * Check for P_LVL2_UP flag before entering C2 and above on
290 	 * an SMP system.
291 	 */
292 	if ((num_online_cpus() > 1) &&
293 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
294 		return -ENODEV;
295 #endif
296 
297 	/* determine C2 and C3 address from pblk */
298 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
299 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
300 
301 	/* determine latencies from FADT */
302 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
303 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
304 
305 	/*
306 	 * FADT specified C2 latency must be less than or equal to
307 	 * 100 microseconds.
308 	 */
309 	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
310 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
311 			"C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
312 		/* invalidate C2 */
313 		pr->power.states[ACPI_STATE_C2].address = 0;
314 	}
315 
316 	/*
317 	 * FADT supplied C3 latency must be less than or equal to
318 	 * 1000 microseconds.
319 	 */
320 	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
321 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
322 			"C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
323 		/* invalidate C3 */
324 		pr->power.states[ACPI_STATE_C3].address = 0;
325 	}
326 
327 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
328 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
329 			  pr->power.states[ACPI_STATE_C2].address,
330 			  pr->power.states[ACPI_STATE_C3].address));
331 
332 	return 0;
333 }
334 
335 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
336 {
337 	if (!pr->power.states[ACPI_STATE_C1].valid) {
338 		/* set the first C-State to C1 */
339 		/* all processors need to support C1 */
340 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
341 		pr->power.states[ACPI_STATE_C1].valid = 1;
342 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
343 	}
344 	/* the C0 state only exists as a filler in our array */
345 	pr->power.states[ACPI_STATE_C0].valid = 1;
346 	return 0;
347 }
348 
349 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
350 {
351 	acpi_status status = 0;
352 	u64 count;
353 	int current_count;
354 	int i;
355 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
356 	union acpi_object *cst;
357 
358 
359 	if (nocst)
360 		return -ENODEV;
361 
362 	current_count = 0;
363 
364 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
365 	if (ACPI_FAILURE(status)) {
366 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
367 		return -ENODEV;
368 	}
369 
370 	cst = buffer.pointer;
371 
372 	/* There must be at least 2 elements */
373 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
374 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
375 		status = -EFAULT;
376 		goto end;
377 	}
378 
379 	count = cst->package.elements[0].integer.value;
380 
381 	/* Validate number of power states. */
382 	if (count < 1 || count != cst->package.count - 1) {
383 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
384 		status = -EFAULT;
385 		goto end;
386 	}
387 
388 	/* Tell driver that at least _CST is supported. */
389 	pr->flags.has_cst = 1;
390 
391 	for (i = 1; i <= count; i++) {
392 		union acpi_object *element;
393 		union acpi_object *obj;
394 		struct acpi_power_register *reg;
395 		struct acpi_processor_cx cx;
396 
397 		memset(&cx, 0, sizeof(cx));
398 
399 		element = &(cst->package.elements[i]);
400 		if (element->type != ACPI_TYPE_PACKAGE)
401 			continue;
402 
403 		if (element->package.count != 4)
404 			continue;
405 
406 		obj = &(element->package.elements[0]);
407 
408 		if (obj->type != ACPI_TYPE_BUFFER)
409 			continue;
410 
411 		reg = (struct acpi_power_register *)obj->buffer.pointer;
412 
413 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
414 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
415 			continue;
416 
417 		/* There should be an easy way to extract an integer... */
418 		obj = &(element->package.elements[1]);
419 		if (obj->type != ACPI_TYPE_INTEGER)
420 			continue;
421 
422 		cx.type = obj->integer.value;
423 		/*
424 		 * Some buggy BIOSes won't list C1 in _CST -
425 		 * Let acpi_processor_get_power_info_default() handle them later
426 		 */
427 		if (i == 1 && cx.type != ACPI_STATE_C1)
428 			current_count++;
429 
430 		cx.address = reg->address;
431 		cx.index = current_count + 1;
432 
433 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
434 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
435 			if (acpi_processor_ffh_cstate_probe
436 					(pr->id, &cx, reg) == 0) {
437 				cx.entry_method = ACPI_CSTATE_FFH;
438 			} else if (cx.type == ACPI_STATE_C1) {
439 				/*
440 				 * C1 is a special case where FIXED_HARDWARE
441 				 * can be handled in non-MWAIT way as well.
442 				 * In that case, save this _CST entry info.
443 				 * Otherwise, ignore this info and continue.
444 				 */
445 				cx.entry_method = ACPI_CSTATE_HALT;
446 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
447 			} else {
448 				continue;
449 			}
450 			if (cx.type == ACPI_STATE_C1 &&
451 			    (boot_option_idle_override == IDLE_NOMWAIT)) {
452 				/*
453 				 * In most cases the C1 space_id obtained from
454 				 * _CST object is FIXED_HARDWARE access mode.
455 				 * But when the option of idle=halt is added,
456 				 * the entry_method type should be changed from
457 				 * CSTATE_FFH to CSTATE_HALT.
458 				 * When the option of idle=nomwait is added,
459 				 * the C1 entry_method type should be
460 				 * CSTATE_HALT.
461 				 */
462 				cx.entry_method = ACPI_CSTATE_HALT;
463 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
464 			}
465 		} else {
466 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
467 				 cx.address);
468 		}
469 
470 		if (cx.type == ACPI_STATE_C1) {
471 			cx.valid = 1;
472 		}
473 
474 		obj = &(element->package.elements[2]);
475 		if (obj->type != ACPI_TYPE_INTEGER)
476 			continue;
477 
478 		cx.latency = obj->integer.value;
479 
480 		obj = &(element->package.elements[3]);
481 		if (obj->type != ACPI_TYPE_INTEGER)
482 			continue;
483 
484 		current_count++;
485 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
486 
487 		/*
488 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
489 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
490 		 */
491 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
492 			printk(KERN_WARNING
493 			       "Limiting number of power states to max (%d)\n",
494 			       ACPI_PROCESSOR_MAX_POWER);
495 			printk(KERN_WARNING
496 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
497 			break;
498 		}
499 	}
500 
501 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
502 			  current_count));
503 
504 	/* Validate number of power states discovered */
505 	if (current_count < 2)
506 		status = -EFAULT;
507 
508       end:
509 	kfree(buffer.pointer);
510 
511 	return status;
512 }
513 
514 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
515 					   struct acpi_processor_cx *cx)
516 {
517 	static int bm_check_flag = -1;
518 	static int bm_control_flag = -1;
519 
520 
521 	if (!cx->address)
522 		return;
523 
524 	/*
525 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
526 	 * DMA transfers are used by any ISA device to avoid livelock.
527 	 * Note that we could disable Type-F DMA (as recommended by
528 	 * the erratum), but this is known to disrupt certain ISA
529 	 * devices thus we take the conservative approach.
530 	 */
531 	else if (errata.piix4.fdma) {
532 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
533 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
534 		return;
535 	}
536 
537 	/* All the logic here assumes flags.bm_check is same across all CPUs */
538 	if (bm_check_flag == -1) {
539 		/* Determine whether bm_check is needed based on CPU  */
540 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
541 		bm_check_flag = pr->flags.bm_check;
542 		bm_control_flag = pr->flags.bm_control;
543 	} else {
544 		pr->flags.bm_check = bm_check_flag;
545 		pr->flags.bm_control = bm_control_flag;
546 	}
547 
548 	if (pr->flags.bm_check) {
549 		if (!pr->flags.bm_control) {
550 			if (pr->flags.has_cst != 1) {
551 				/* bus mastering control is necessary */
552 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
553 					"C3 support requires BM control\n"));
554 				return;
555 			} else {
556 				/* Here we enter C3 without bus mastering */
557 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
558 					"C3 support without BM control\n"));
559 			}
560 		}
561 	} else {
562 		/*
563 		 * WBINVD should be set in fadt, for C3 state to be
564 		 * supported on when bm_check is not required.
565 		 */
566 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
567 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
568 					  "Cache invalidation should work properly"
569 					  " for C3 to be enabled on SMP systems\n"));
570 			return;
571 		}
572 	}
573 
574 	/*
575 	 * Otherwise we've met all of our C3 requirements.
576 	 * Normalize the C3 latency to expidite policy.  Enable
577 	 * checking of bus mastering status (bm_check) so we can
578 	 * use this in our C3 policy
579 	 */
580 	cx->valid = 1;
581 
582 	/*
583 	 * On older chipsets, BM_RLD needs to be set
584 	 * in order for Bus Master activity to wake the
585 	 * system from C3.  Newer chipsets handle DMA
586 	 * during C3 automatically and BM_RLD is a NOP.
587 	 * In either case, the proper way to
588 	 * handle BM_RLD is to set it and leave it set.
589 	 */
590 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
591 
592 	return;
593 }
594 
595 static int acpi_processor_power_verify(struct acpi_processor *pr)
596 {
597 	unsigned int i;
598 	unsigned int working = 0;
599 
600 	pr->power.timer_broadcast_on_state = INT_MAX;
601 
602 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
603 		struct acpi_processor_cx *cx = &pr->power.states[i];
604 
605 		switch (cx->type) {
606 		case ACPI_STATE_C1:
607 			cx->valid = 1;
608 			break;
609 
610 		case ACPI_STATE_C2:
611 			if (!cx->address)
612 				break;
613 			cx->valid = 1;
614 			break;
615 
616 		case ACPI_STATE_C3:
617 			acpi_processor_power_verify_c3(pr, cx);
618 			break;
619 		}
620 		if (!cx->valid)
621 			continue;
622 
623 		lapic_timer_check_state(i, pr, cx);
624 		tsc_check_state(cx->type);
625 		working++;
626 	}
627 
628 	lapic_timer_propagate_broadcast(pr);
629 
630 	return (working);
631 }
632 
633 static int acpi_processor_get_power_info(struct acpi_processor *pr)
634 {
635 	unsigned int i;
636 	int result;
637 
638 
639 	/* NOTE: the idle thread may not be running while calling
640 	 * this function */
641 
642 	/* Zero initialize all the C-states info. */
643 	memset(pr->power.states, 0, sizeof(pr->power.states));
644 
645 	result = acpi_processor_get_power_info_cst(pr);
646 	if (result == -ENODEV)
647 		result = acpi_processor_get_power_info_fadt(pr);
648 
649 	if (result)
650 		return result;
651 
652 	acpi_processor_get_power_info_default(pr);
653 
654 	pr->power.count = acpi_processor_power_verify(pr);
655 
656 	/*
657 	 * if one state of type C2 or C3 is available, mark this
658 	 * CPU as being "idle manageable"
659 	 */
660 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
661 		if (pr->power.states[i].valid) {
662 			pr->power.count = i;
663 			if (pr->power.states[i].type >= ACPI_STATE_C2)
664 				pr->flags.power = 1;
665 		}
666 	}
667 
668 	return 0;
669 }
670 
671 /**
672  * acpi_idle_bm_check - checks if bus master activity was detected
673  */
674 static int acpi_idle_bm_check(void)
675 {
676 	u32 bm_status = 0;
677 
678 	if (bm_check_disable)
679 		return 0;
680 
681 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
682 	if (bm_status)
683 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
684 	/*
685 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
686 	 * the true state of bus mastering activity; forcing us to
687 	 * manually check the BMIDEA bit of each IDE channel.
688 	 */
689 	else if (errata.piix4.bmisx) {
690 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
691 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
692 			bm_status = 1;
693 	}
694 	return bm_status;
695 }
696 
697 /**
698  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
699  * @cx: cstate data
700  *
701  * Caller disables interrupt before call and enables interrupt after return.
702  */
703 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
704 {
705 	/* Don't trace irqs off for idle */
706 	stop_critical_timings();
707 	if (cx->entry_method == ACPI_CSTATE_FFH) {
708 		/* Call into architectural FFH based C-state */
709 		acpi_processor_ffh_cstate_enter(cx);
710 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
711 		acpi_safe_halt();
712 	} else {
713 		/* IO port based C-state */
714 		inb(cx->address);
715 		/* Dummy wait op - must do something useless after P_LVL2 read
716 		   because chipsets cannot guarantee that STPCLK# signal
717 		   gets asserted in time to freeze execution properly. */
718 		inl(acpi_gbl_FADT.xpm_timer_block.address);
719 	}
720 	start_critical_timings();
721 }
722 
723 /**
724  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
725  * @dev: the target CPU
726  * @drv: cpuidle driver containing cpuidle state info
727  * @index: index of target state
728  *
729  * This is equivalent to the HALT instruction.
730  */
731 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
732 		struct cpuidle_driver *drv, int index)
733 {
734 	struct acpi_processor *pr;
735 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
736 
737 	pr = __this_cpu_read(processors);
738 
739 	if (unlikely(!pr))
740 		return -EINVAL;
741 
742 	lapic_timer_state_broadcast(pr, cx, 1);
743 	acpi_idle_do_entry(cx);
744 
745 	lapic_timer_state_broadcast(pr, cx, 0);
746 
747 	return index;
748 }
749 
750 
751 /**
752  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
753  * @dev: the target CPU
754  * @index: the index of suggested state
755  */
756 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
757 {
758 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
759 
760 	ACPI_FLUSH_CPU_CACHE();
761 
762 	while (1) {
763 
764 		if (cx->entry_method == ACPI_CSTATE_HALT)
765 			safe_halt();
766 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
767 			inb(cx->address);
768 			/* See comment in acpi_idle_do_entry() */
769 			inl(acpi_gbl_FADT.xpm_timer_block.address);
770 		} else
771 			return -ENODEV;
772 	}
773 
774 	/* Never reached */
775 	return 0;
776 }
777 
778 /**
779  * acpi_idle_enter_simple - enters an ACPI state without BM handling
780  * @dev: the target CPU
781  * @drv: cpuidle driver with cpuidle state information
782  * @index: the index of suggested state
783  */
784 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
785 		struct cpuidle_driver *drv, int index)
786 {
787 	struct acpi_processor *pr;
788 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
789 
790 	pr = __this_cpu_read(processors);
791 
792 	if (unlikely(!pr))
793 		return -EINVAL;
794 
795 	if (cx->entry_method != ACPI_CSTATE_FFH) {
796 		current_thread_info()->status &= ~TS_POLLING;
797 		/*
798 		 * TS_POLLING-cleared state must be visible before we test
799 		 * NEED_RESCHED:
800 		 */
801 		smp_mb();
802 
803 		if (unlikely(need_resched())) {
804 			current_thread_info()->status |= TS_POLLING;
805 			return -EINVAL;
806 		}
807 	}
808 
809 	/*
810 	 * Must be done before busmaster disable as we might need to
811 	 * access HPET !
812 	 */
813 	lapic_timer_state_broadcast(pr, cx, 1);
814 
815 	if (cx->type == ACPI_STATE_C3)
816 		ACPI_FLUSH_CPU_CACHE();
817 
818 	/* Tell the scheduler that we are going deep-idle: */
819 	sched_clock_idle_sleep_event();
820 	acpi_idle_do_entry(cx);
821 
822 	sched_clock_idle_wakeup_event(0);
823 
824 	if (cx->entry_method != ACPI_CSTATE_FFH)
825 		current_thread_info()->status |= TS_POLLING;
826 
827 	lapic_timer_state_broadcast(pr, cx, 0);
828 	return index;
829 }
830 
831 static int c3_cpu_count;
832 static DEFINE_RAW_SPINLOCK(c3_lock);
833 
834 /**
835  * acpi_idle_enter_bm - enters C3 with proper BM handling
836  * @dev: the target CPU
837  * @drv: cpuidle driver containing state data
838  * @index: the index of suggested state
839  *
840  * If BM is detected, the deepest non-C3 idle state is entered instead.
841  */
842 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
843 		struct cpuidle_driver *drv, int index)
844 {
845 	struct acpi_processor *pr;
846 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
847 
848 	pr = __this_cpu_read(processors);
849 
850 	if (unlikely(!pr))
851 		return -EINVAL;
852 
853 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
854 		if (drv->safe_state_index >= 0) {
855 			return drv->states[drv->safe_state_index].enter(dev,
856 						drv, drv->safe_state_index);
857 		} else {
858 			acpi_safe_halt();
859 			return -EBUSY;
860 		}
861 	}
862 
863 	if (cx->entry_method != ACPI_CSTATE_FFH) {
864 		current_thread_info()->status &= ~TS_POLLING;
865 		/*
866 		 * TS_POLLING-cleared state must be visible before we test
867 		 * NEED_RESCHED:
868 		 */
869 		smp_mb();
870 
871 		if (unlikely(need_resched())) {
872 			current_thread_info()->status |= TS_POLLING;
873 			return -EINVAL;
874 		}
875 	}
876 
877 	acpi_unlazy_tlb(smp_processor_id());
878 
879 	/* Tell the scheduler that we are going deep-idle: */
880 	sched_clock_idle_sleep_event();
881 	/*
882 	 * Must be done before busmaster disable as we might need to
883 	 * access HPET !
884 	 */
885 	lapic_timer_state_broadcast(pr, cx, 1);
886 
887 	/*
888 	 * disable bus master
889 	 * bm_check implies we need ARB_DIS
890 	 * !bm_check implies we need cache flush
891 	 * bm_control implies whether we can do ARB_DIS
892 	 *
893 	 * That leaves a case where bm_check is set and bm_control is
894 	 * not set. In that case we cannot do much, we enter C3
895 	 * without doing anything.
896 	 */
897 	if (pr->flags.bm_check && pr->flags.bm_control) {
898 		raw_spin_lock(&c3_lock);
899 		c3_cpu_count++;
900 		/* Disable bus master arbitration when all CPUs are in C3 */
901 		if (c3_cpu_count == num_online_cpus())
902 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
903 		raw_spin_unlock(&c3_lock);
904 	} else if (!pr->flags.bm_check) {
905 		ACPI_FLUSH_CPU_CACHE();
906 	}
907 
908 	acpi_idle_do_entry(cx);
909 
910 	/* Re-enable bus master arbitration */
911 	if (pr->flags.bm_check && pr->flags.bm_control) {
912 		raw_spin_lock(&c3_lock);
913 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
914 		c3_cpu_count--;
915 		raw_spin_unlock(&c3_lock);
916 	}
917 
918 	sched_clock_idle_wakeup_event(0);
919 
920 	if (cx->entry_method != ACPI_CSTATE_FFH)
921 		current_thread_info()->status |= TS_POLLING;
922 
923 	lapic_timer_state_broadcast(pr, cx, 0);
924 	return index;
925 }
926 
927 struct cpuidle_driver acpi_idle_driver = {
928 	.name =		"acpi_idle",
929 	.owner =	THIS_MODULE,
930 };
931 
932 /**
933  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
934  * device i.e. per-cpu data
935  *
936  * @pr: the ACPI processor
937  * @dev : the cpuidle device
938  */
939 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
940 					   struct cpuidle_device *dev)
941 {
942 	int i, count = CPUIDLE_DRIVER_STATE_START;
943 	struct acpi_processor_cx *cx;
944 
945 	if (!pr->flags.power_setup_done)
946 		return -EINVAL;
947 
948 	if (pr->flags.power == 0) {
949 		return -EINVAL;
950 	}
951 
952 	if (!dev)
953 		return -EINVAL;
954 
955 	dev->cpu = pr->id;
956 
957 	if (max_cstate == 0)
958 		max_cstate = 1;
959 
960 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
961 		cx = &pr->power.states[i];
962 
963 		if (!cx->valid)
964 			continue;
965 
966 #ifdef CONFIG_HOTPLUG_CPU
967 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
968 		    !pr->flags.has_cst &&
969 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
970 			continue;
971 #endif
972 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
973 
974 		count++;
975 		if (count == CPUIDLE_STATE_MAX)
976 			break;
977 	}
978 
979 	dev->state_count = count;
980 
981 	if (!count)
982 		return -EINVAL;
983 
984 	return 0;
985 }
986 
987 /**
988  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
989  * global state data i.e. idle routines
990  *
991  * @pr: the ACPI processor
992  */
993 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
994 {
995 	int i, count = CPUIDLE_DRIVER_STATE_START;
996 	struct acpi_processor_cx *cx;
997 	struct cpuidle_state *state;
998 	struct cpuidle_driver *drv = &acpi_idle_driver;
999 
1000 	if (!pr->flags.power_setup_done)
1001 		return -EINVAL;
1002 
1003 	if (pr->flags.power == 0)
1004 		return -EINVAL;
1005 
1006 	drv->safe_state_index = -1;
1007 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1008 		drv->states[i].name[0] = '\0';
1009 		drv->states[i].desc[0] = '\0';
1010 	}
1011 
1012 	if (max_cstate == 0)
1013 		max_cstate = 1;
1014 
1015 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1016 		cx = &pr->power.states[i];
1017 
1018 		if (!cx->valid)
1019 			continue;
1020 
1021 #ifdef CONFIG_HOTPLUG_CPU
1022 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1023 		    !pr->flags.has_cst &&
1024 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1025 			continue;
1026 #endif
1027 
1028 		state = &drv->states[count];
1029 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1030 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1031 		state->exit_latency = cx->latency;
1032 		state->target_residency = cx->latency * latency_factor;
1033 
1034 		state->flags = 0;
1035 		switch (cx->type) {
1036 			case ACPI_STATE_C1:
1037 			if (cx->entry_method == ACPI_CSTATE_FFH)
1038 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1039 
1040 			state->enter = acpi_idle_enter_c1;
1041 			state->enter_dead = acpi_idle_play_dead;
1042 			drv->safe_state_index = count;
1043 			break;
1044 
1045 			case ACPI_STATE_C2:
1046 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1047 			state->enter = acpi_idle_enter_simple;
1048 			state->enter_dead = acpi_idle_play_dead;
1049 			drv->safe_state_index = count;
1050 			break;
1051 
1052 			case ACPI_STATE_C3:
1053 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1054 			state->enter = pr->flags.bm_check ?
1055 					acpi_idle_enter_bm :
1056 					acpi_idle_enter_simple;
1057 			break;
1058 		}
1059 
1060 		count++;
1061 		if (count == CPUIDLE_STATE_MAX)
1062 			break;
1063 	}
1064 
1065 	drv->state_count = count;
1066 
1067 	if (!count)
1068 		return -EINVAL;
1069 
1070 	return 0;
1071 }
1072 
1073 int acpi_processor_hotplug(struct acpi_processor *pr)
1074 {
1075 	int ret = 0;
1076 	struct cpuidle_device *dev;
1077 
1078 	if (disabled_by_idle_boot_param())
1079 		return 0;
1080 
1081 	if (!pr)
1082 		return -EINVAL;
1083 
1084 	if (nocst) {
1085 		return -ENODEV;
1086 	}
1087 
1088 	if (!pr->flags.power_setup_done)
1089 		return -ENODEV;
1090 
1091 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1092 	cpuidle_pause_and_lock();
1093 	cpuidle_disable_device(dev);
1094 	acpi_processor_get_power_info(pr);
1095 	if (pr->flags.power) {
1096 		acpi_processor_setup_cpuidle_cx(pr, dev);
1097 		ret = cpuidle_enable_device(dev);
1098 	}
1099 	cpuidle_resume_and_unlock();
1100 
1101 	return ret;
1102 }
1103 
1104 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1105 {
1106 	int cpu;
1107 	struct acpi_processor *_pr;
1108 	struct cpuidle_device *dev;
1109 
1110 	if (disabled_by_idle_boot_param())
1111 		return 0;
1112 
1113 	if (!pr)
1114 		return -EINVAL;
1115 
1116 	if (nocst)
1117 		return -ENODEV;
1118 
1119 	if (!pr->flags.power_setup_done)
1120 		return -ENODEV;
1121 
1122 	/*
1123 	 * FIXME:  Design the ACPI notification to make it once per
1124 	 * system instead of once per-cpu.  This condition is a hack
1125 	 * to make the code that updates C-States be called once.
1126 	 */
1127 
1128 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1129 
1130 		cpuidle_pause_and_lock();
1131 		/* Protect against cpu-hotplug */
1132 		get_online_cpus();
1133 
1134 		/* Disable all cpuidle devices */
1135 		for_each_online_cpu(cpu) {
1136 			_pr = per_cpu(processors, cpu);
1137 			if (!_pr || !_pr->flags.power_setup_done)
1138 				continue;
1139 			dev = per_cpu(acpi_cpuidle_device, cpu);
1140 			cpuidle_disable_device(dev);
1141 		}
1142 
1143 		/* Populate Updated C-state information */
1144 		acpi_processor_get_power_info(pr);
1145 		acpi_processor_setup_cpuidle_states(pr);
1146 
1147 		/* Enable all cpuidle devices */
1148 		for_each_online_cpu(cpu) {
1149 			_pr = per_cpu(processors, cpu);
1150 			if (!_pr || !_pr->flags.power_setup_done)
1151 				continue;
1152 			acpi_processor_get_power_info(_pr);
1153 			if (_pr->flags.power) {
1154 				dev = per_cpu(acpi_cpuidle_device, cpu);
1155 				acpi_processor_setup_cpuidle_cx(_pr, dev);
1156 				cpuidle_enable_device(dev);
1157 			}
1158 		}
1159 		put_online_cpus();
1160 		cpuidle_resume_and_unlock();
1161 	}
1162 
1163 	return 0;
1164 }
1165 
1166 static int acpi_processor_registered;
1167 
1168 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr)
1169 {
1170 	acpi_status status = 0;
1171 	int retval;
1172 	struct cpuidle_device *dev;
1173 	static int first_run;
1174 
1175 	if (disabled_by_idle_boot_param())
1176 		return 0;
1177 
1178 	if (!first_run) {
1179 		dmi_check_system(processor_power_dmi_table);
1180 		max_cstate = acpi_processor_cstate_check(max_cstate);
1181 		if (max_cstate < ACPI_C_STATES_MAX)
1182 			printk(KERN_NOTICE
1183 			       "ACPI: processor limited to max C-state %d\n",
1184 			       max_cstate);
1185 		first_run++;
1186 	}
1187 
1188 	if (!pr)
1189 		return -EINVAL;
1190 
1191 	if (acpi_gbl_FADT.cst_control && !nocst) {
1192 		status =
1193 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1194 		if (ACPI_FAILURE(status)) {
1195 			ACPI_EXCEPTION((AE_INFO, status,
1196 					"Notifying BIOS of _CST ability failed"));
1197 		}
1198 	}
1199 
1200 	acpi_processor_get_power_info(pr);
1201 	pr->flags.power_setup_done = 1;
1202 
1203 	/*
1204 	 * Install the idle handler if processor power management is supported.
1205 	 * Note that we use previously set idle handler will be used on
1206 	 * platforms that only support C1.
1207 	 */
1208 	if (pr->flags.power) {
1209 		/* Register acpi_idle_driver if not already registered */
1210 		if (!acpi_processor_registered) {
1211 			acpi_processor_setup_cpuidle_states(pr);
1212 			retval = cpuidle_register_driver(&acpi_idle_driver);
1213 			if (retval)
1214 				return retval;
1215 			printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1216 					acpi_idle_driver.name);
1217 		}
1218 
1219 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1220 		if (!dev)
1221 			return -ENOMEM;
1222 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1223 
1224 		acpi_processor_setup_cpuidle_cx(pr, dev);
1225 
1226 		/* Register per-cpu cpuidle_device. Cpuidle driver
1227 		 * must already be registered before registering device
1228 		 */
1229 		retval = cpuidle_register_device(dev);
1230 		if (retval) {
1231 			if (acpi_processor_registered == 0)
1232 				cpuidle_unregister_driver(&acpi_idle_driver);
1233 			return retval;
1234 		}
1235 		acpi_processor_registered++;
1236 	}
1237 	return 0;
1238 }
1239 
1240 int acpi_processor_power_exit(struct acpi_processor *pr)
1241 {
1242 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1243 
1244 	if (disabled_by_idle_boot_param())
1245 		return 0;
1246 
1247 	if (pr->flags.power) {
1248 		cpuidle_unregister_device(dev);
1249 		acpi_processor_registered--;
1250 		if (acpi_processor_registered == 0)
1251 			cpuidle_unregister_driver(&acpi_idle_driver);
1252 	}
1253 
1254 	pr->flags.power_setup_done = 0;
1255 	return 0;
1256 }
1257