xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision 161f4089)
1 /*
2  * processor_idle - idle state submodule to the ACPI processor driver
3  *
4  *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
5  *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
6  *  Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
7  *  Copyright (C) 2004  Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
8  *  			- Added processor hotplug support
9  *  Copyright (C) 2005  Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
10  *  			- Added support for C3 on SMP
11  *
12  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13  *
14  *  This program is free software; you can redistribute it and/or modify
15  *  it under the terms of the GNU General Public License as published by
16  *  the Free Software Foundation; either version 2 of the License, or (at
17  *  your option) any later version.
18  *
19  *  This program is distributed in the hope that it will be useful, but
20  *  WITHOUT ANY WARRANTY; without even the implied warranty of
21  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22  *  General Public License for more details.
23  *
24  *  You should have received a copy of the GNU General Public License along
25  *  with this program; if not, write to the Free Software Foundation, Inc.,
26  *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
27  *
28  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
29  */
30 
31 #include <linux/module.h>
32 #include <linux/acpi.h>
33 #include <linux/dmi.h>
34 #include <linux/sched.h>       /* need_resched() */
35 #include <linux/clockchips.h>
36 #include <linux/cpuidle.h>
37 #include <linux/syscore_ops.h>
38 
39 /*
40  * Include the apic definitions for x86 to have the APIC timer related defines
41  * available also for UP (on SMP it gets magically included via linux/smp.h).
42  * asm/acpi.h is not an option, as it would require more include magic. Also
43  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
44  */
45 #ifdef CONFIG_X86
46 #include <asm/apic.h>
47 #endif
48 
49 #include <acpi/acpi_bus.h>
50 #include <acpi/processor.h>
51 
52 #define PREFIX "ACPI: "
53 
54 #define ACPI_PROCESSOR_CLASS            "processor"
55 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
56 ACPI_MODULE_NAME("processor_idle");
57 
58 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
59 module_param(max_cstate, uint, 0000);
60 static unsigned int nocst __read_mostly;
61 module_param(nocst, uint, 0000);
62 static int bm_check_disable __read_mostly;
63 module_param(bm_check_disable, uint, 0000);
64 
65 static unsigned int latency_factor __read_mostly = 2;
66 module_param(latency_factor, uint, 0644);
67 
68 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
69 
70 static DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX],
71 								acpi_cstate);
72 
73 static int disabled_by_idle_boot_param(void)
74 {
75 	return boot_option_idle_override == IDLE_POLL ||
76 		boot_option_idle_override == IDLE_HALT;
77 }
78 
79 /*
80  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
81  * For now disable this. Probably a bug somewhere else.
82  *
83  * To skip this limit, boot/load with a large max_cstate limit.
84  */
85 static int set_max_cstate(const struct dmi_system_id *id)
86 {
87 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
88 		return 0;
89 
90 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
91 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
92 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
93 
94 	max_cstate = (long)id->driver_data;
95 
96 	return 0;
97 }
98 
99 static struct dmi_system_id processor_power_dmi_table[] = {
100 	{ set_max_cstate, "Clevo 5600D", {
101 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
102 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
103 	 (void *)2},
104 	{ set_max_cstate, "Pavilion zv5000", {
105 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
106 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
107 	 (void *)1},
108 	{ set_max_cstate, "Asus L8400B", {
109 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
110 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
111 	 (void *)1},
112 	{},
113 };
114 
115 
116 /*
117  * Callers should disable interrupts before the call and enable
118  * interrupts after return.
119  */
120 static void acpi_safe_halt(void)
121 {
122 	if (!tif_need_resched()) {
123 		safe_halt();
124 		local_irq_disable();
125 	}
126 }
127 
128 #ifdef ARCH_APICTIMER_STOPS_ON_C3
129 
130 /*
131  * Some BIOS implementations switch to C3 in the published C2 state.
132  * This seems to be a common problem on AMD boxen, but other vendors
133  * are affected too. We pick the most conservative approach: we assume
134  * that the local APIC stops in both C2 and C3.
135  */
136 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
137 				   struct acpi_processor_cx *cx)
138 {
139 	struct acpi_processor_power *pwr = &pr->power;
140 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
141 
142 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
143 		return;
144 
145 	if (amd_e400_c1e_detected)
146 		type = ACPI_STATE_C1;
147 
148 	/*
149 	 * Check, if one of the previous states already marked the lapic
150 	 * unstable
151 	 */
152 	if (pwr->timer_broadcast_on_state < state)
153 		return;
154 
155 	if (cx->type >= type)
156 		pr->power.timer_broadcast_on_state = state;
157 }
158 
159 static void __lapic_timer_propagate_broadcast(void *arg)
160 {
161 	struct acpi_processor *pr = (struct acpi_processor *) arg;
162 	unsigned long reason;
163 
164 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
165 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
166 
167 	clockevents_notify(reason, &pr->id);
168 }
169 
170 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
171 {
172 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
173 				 (void *)pr, 1);
174 }
175 
176 /* Power(C) State timer broadcast control */
177 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
178 				       struct acpi_processor_cx *cx,
179 				       int broadcast)
180 {
181 	int state = cx - pr->power.states;
182 
183 	if (state >= pr->power.timer_broadcast_on_state) {
184 		unsigned long reason;
185 
186 		reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
187 			CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
188 		clockevents_notify(reason, &pr->id);
189 	}
190 }
191 
192 #else
193 
194 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
195 				   struct acpi_processor_cx *cstate) { }
196 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
197 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
198 				       struct acpi_processor_cx *cx,
199 				       int broadcast)
200 {
201 }
202 
203 #endif
204 
205 #ifdef CONFIG_PM_SLEEP
206 static u32 saved_bm_rld;
207 
208 static int acpi_processor_suspend(void)
209 {
210 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
211 	return 0;
212 }
213 
214 static void acpi_processor_resume(void)
215 {
216 	u32 resumed_bm_rld;
217 
218 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
219 	if (resumed_bm_rld == saved_bm_rld)
220 		return;
221 
222 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
223 }
224 
225 static struct syscore_ops acpi_processor_syscore_ops = {
226 	.suspend = acpi_processor_suspend,
227 	.resume = acpi_processor_resume,
228 };
229 
230 void acpi_processor_syscore_init(void)
231 {
232 	register_syscore_ops(&acpi_processor_syscore_ops);
233 }
234 
235 void acpi_processor_syscore_exit(void)
236 {
237 	unregister_syscore_ops(&acpi_processor_syscore_ops);
238 }
239 #endif /* CONFIG_PM_SLEEP */
240 
241 #if defined(CONFIG_X86)
242 static void tsc_check_state(int state)
243 {
244 	switch (boot_cpu_data.x86_vendor) {
245 	case X86_VENDOR_AMD:
246 	case X86_VENDOR_INTEL:
247 		/*
248 		 * AMD Fam10h TSC will tick in all
249 		 * C/P/S0/S1 states when this bit is set.
250 		 */
251 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
252 			return;
253 
254 		/*FALL THROUGH*/
255 	default:
256 		/* TSC could halt in idle, so notify users */
257 		if (state > ACPI_STATE_C1)
258 			mark_tsc_unstable("TSC halts in idle");
259 	}
260 }
261 #else
262 static void tsc_check_state(int state) { return; }
263 #endif
264 
265 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
266 {
267 
268 	if (!pr)
269 		return -EINVAL;
270 
271 	if (!pr->pblk)
272 		return -ENODEV;
273 
274 	/* if info is obtained from pblk/fadt, type equals state */
275 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
276 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
277 
278 #ifndef CONFIG_HOTPLUG_CPU
279 	/*
280 	 * Check for P_LVL2_UP flag before entering C2 and above on
281 	 * an SMP system.
282 	 */
283 	if ((num_online_cpus() > 1) &&
284 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
285 		return -ENODEV;
286 #endif
287 
288 	/* determine C2 and C3 address from pblk */
289 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
290 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
291 
292 	/* determine latencies from FADT */
293 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
294 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
295 
296 	/*
297 	 * FADT specified C2 latency must be less than or equal to
298 	 * 100 microseconds.
299 	 */
300 	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
301 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
302 			"C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
303 		/* invalidate C2 */
304 		pr->power.states[ACPI_STATE_C2].address = 0;
305 	}
306 
307 	/*
308 	 * FADT supplied C3 latency must be less than or equal to
309 	 * 1000 microseconds.
310 	 */
311 	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
312 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
313 			"C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
314 		/* invalidate C3 */
315 		pr->power.states[ACPI_STATE_C3].address = 0;
316 	}
317 
318 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
319 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
320 			  pr->power.states[ACPI_STATE_C2].address,
321 			  pr->power.states[ACPI_STATE_C3].address));
322 
323 	return 0;
324 }
325 
326 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
327 {
328 	if (!pr->power.states[ACPI_STATE_C1].valid) {
329 		/* set the first C-State to C1 */
330 		/* all processors need to support C1 */
331 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
332 		pr->power.states[ACPI_STATE_C1].valid = 1;
333 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
334 	}
335 	/* the C0 state only exists as a filler in our array */
336 	pr->power.states[ACPI_STATE_C0].valid = 1;
337 	return 0;
338 }
339 
340 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
341 {
342 	acpi_status status = 0;
343 	u64 count;
344 	int current_count;
345 	int i;
346 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
347 	union acpi_object *cst;
348 
349 
350 	if (nocst)
351 		return -ENODEV;
352 
353 	current_count = 0;
354 
355 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
356 	if (ACPI_FAILURE(status)) {
357 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
358 		return -ENODEV;
359 	}
360 
361 	cst = buffer.pointer;
362 
363 	/* There must be at least 2 elements */
364 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
365 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
366 		status = -EFAULT;
367 		goto end;
368 	}
369 
370 	count = cst->package.elements[0].integer.value;
371 
372 	/* Validate number of power states. */
373 	if (count < 1 || count != cst->package.count - 1) {
374 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
375 		status = -EFAULT;
376 		goto end;
377 	}
378 
379 	/* Tell driver that at least _CST is supported. */
380 	pr->flags.has_cst = 1;
381 
382 	for (i = 1; i <= count; i++) {
383 		union acpi_object *element;
384 		union acpi_object *obj;
385 		struct acpi_power_register *reg;
386 		struct acpi_processor_cx cx;
387 
388 		memset(&cx, 0, sizeof(cx));
389 
390 		element = &(cst->package.elements[i]);
391 		if (element->type != ACPI_TYPE_PACKAGE)
392 			continue;
393 
394 		if (element->package.count != 4)
395 			continue;
396 
397 		obj = &(element->package.elements[0]);
398 
399 		if (obj->type != ACPI_TYPE_BUFFER)
400 			continue;
401 
402 		reg = (struct acpi_power_register *)obj->buffer.pointer;
403 
404 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
405 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
406 			continue;
407 
408 		/* There should be an easy way to extract an integer... */
409 		obj = &(element->package.elements[1]);
410 		if (obj->type != ACPI_TYPE_INTEGER)
411 			continue;
412 
413 		cx.type = obj->integer.value;
414 		/*
415 		 * Some buggy BIOSes won't list C1 in _CST -
416 		 * Let acpi_processor_get_power_info_default() handle them later
417 		 */
418 		if (i == 1 && cx.type != ACPI_STATE_C1)
419 			current_count++;
420 
421 		cx.address = reg->address;
422 		cx.index = current_count + 1;
423 
424 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
425 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
426 			if (acpi_processor_ffh_cstate_probe
427 					(pr->id, &cx, reg) == 0) {
428 				cx.entry_method = ACPI_CSTATE_FFH;
429 			} else if (cx.type == ACPI_STATE_C1) {
430 				/*
431 				 * C1 is a special case where FIXED_HARDWARE
432 				 * can be handled in non-MWAIT way as well.
433 				 * In that case, save this _CST entry info.
434 				 * Otherwise, ignore this info and continue.
435 				 */
436 				cx.entry_method = ACPI_CSTATE_HALT;
437 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
438 			} else {
439 				continue;
440 			}
441 			if (cx.type == ACPI_STATE_C1 &&
442 			    (boot_option_idle_override == IDLE_NOMWAIT)) {
443 				/*
444 				 * In most cases the C1 space_id obtained from
445 				 * _CST object is FIXED_HARDWARE access mode.
446 				 * But when the option of idle=halt is added,
447 				 * the entry_method type should be changed from
448 				 * CSTATE_FFH to CSTATE_HALT.
449 				 * When the option of idle=nomwait is added,
450 				 * the C1 entry_method type should be
451 				 * CSTATE_HALT.
452 				 */
453 				cx.entry_method = ACPI_CSTATE_HALT;
454 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
455 			}
456 		} else {
457 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
458 				 cx.address);
459 		}
460 
461 		if (cx.type == ACPI_STATE_C1) {
462 			cx.valid = 1;
463 		}
464 
465 		obj = &(element->package.elements[2]);
466 		if (obj->type != ACPI_TYPE_INTEGER)
467 			continue;
468 
469 		cx.latency = obj->integer.value;
470 
471 		obj = &(element->package.elements[3]);
472 		if (obj->type != ACPI_TYPE_INTEGER)
473 			continue;
474 
475 		current_count++;
476 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
477 
478 		/*
479 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
480 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
481 		 */
482 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
483 			printk(KERN_WARNING
484 			       "Limiting number of power states to max (%d)\n",
485 			       ACPI_PROCESSOR_MAX_POWER);
486 			printk(KERN_WARNING
487 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
488 			break;
489 		}
490 	}
491 
492 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
493 			  current_count));
494 
495 	/* Validate number of power states discovered */
496 	if (current_count < 2)
497 		status = -EFAULT;
498 
499       end:
500 	kfree(buffer.pointer);
501 
502 	return status;
503 }
504 
505 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
506 					   struct acpi_processor_cx *cx)
507 {
508 	static int bm_check_flag = -1;
509 	static int bm_control_flag = -1;
510 
511 
512 	if (!cx->address)
513 		return;
514 
515 	/*
516 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
517 	 * DMA transfers are used by any ISA device to avoid livelock.
518 	 * Note that we could disable Type-F DMA (as recommended by
519 	 * the erratum), but this is known to disrupt certain ISA
520 	 * devices thus we take the conservative approach.
521 	 */
522 	else if (errata.piix4.fdma) {
523 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
524 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
525 		return;
526 	}
527 
528 	/* All the logic here assumes flags.bm_check is same across all CPUs */
529 	if (bm_check_flag == -1) {
530 		/* Determine whether bm_check is needed based on CPU  */
531 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
532 		bm_check_flag = pr->flags.bm_check;
533 		bm_control_flag = pr->flags.bm_control;
534 	} else {
535 		pr->flags.bm_check = bm_check_flag;
536 		pr->flags.bm_control = bm_control_flag;
537 	}
538 
539 	if (pr->flags.bm_check) {
540 		if (!pr->flags.bm_control) {
541 			if (pr->flags.has_cst != 1) {
542 				/* bus mastering control is necessary */
543 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
544 					"C3 support requires BM control\n"));
545 				return;
546 			} else {
547 				/* Here we enter C3 without bus mastering */
548 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
549 					"C3 support without BM control\n"));
550 			}
551 		}
552 	} else {
553 		/*
554 		 * WBINVD should be set in fadt, for C3 state to be
555 		 * supported on when bm_check is not required.
556 		 */
557 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
558 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
559 					  "Cache invalidation should work properly"
560 					  " for C3 to be enabled on SMP systems\n"));
561 			return;
562 		}
563 	}
564 
565 	/*
566 	 * Otherwise we've met all of our C3 requirements.
567 	 * Normalize the C3 latency to expidite policy.  Enable
568 	 * checking of bus mastering status (bm_check) so we can
569 	 * use this in our C3 policy
570 	 */
571 	cx->valid = 1;
572 
573 	/*
574 	 * On older chipsets, BM_RLD needs to be set
575 	 * in order for Bus Master activity to wake the
576 	 * system from C3.  Newer chipsets handle DMA
577 	 * during C3 automatically and BM_RLD is a NOP.
578 	 * In either case, the proper way to
579 	 * handle BM_RLD is to set it and leave it set.
580 	 */
581 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
582 
583 	return;
584 }
585 
586 static int acpi_processor_power_verify(struct acpi_processor *pr)
587 {
588 	unsigned int i;
589 	unsigned int working = 0;
590 
591 	pr->power.timer_broadcast_on_state = INT_MAX;
592 
593 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
594 		struct acpi_processor_cx *cx = &pr->power.states[i];
595 
596 		switch (cx->type) {
597 		case ACPI_STATE_C1:
598 			cx->valid = 1;
599 			break;
600 
601 		case ACPI_STATE_C2:
602 			if (!cx->address)
603 				break;
604 			cx->valid = 1;
605 			break;
606 
607 		case ACPI_STATE_C3:
608 			acpi_processor_power_verify_c3(pr, cx);
609 			break;
610 		}
611 		if (!cx->valid)
612 			continue;
613 
614 		lapic_timer_check_state(i, pr, cx);
615 		tsc_check_state(cx->type);
616 		working++;
617 	}
618 
619 	lapic_timer_propagate_broadcast(pr);
620 
621 	return (working);
622 }
623 
624 static int acpi_processor_get_power_info(struct acpi_processor *pr)
625 {
626 	unsigned int i;
627 	int result;
628 
629 
630 	/* NOTE: the idle thread may not be running while calling
631 	 * this function */
632 
633 	/* Zero initialize all the C-states info. */
634 	memset(pr->power.states, 0, sizeof(pr->power.states));
635 
636 	result = acpi_processor_get_power_info_cst(pr);
637 	if (result == -ENODEV)
638 		result = acpi_processor_get_power_info_fadt(pr);
639 
640 	if (result)
641 		return result;
642 
643 	acpi_processor_get_power_info_default(pr);
644 
645 	pr->power.count = acpi_processor_power_verify(pr);
646 
647 	/*
648 	 * if one state of type C2 or C3 is available, mark this
649 	 * CPU as being "idle manageable"
650 	 */
651 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
652 		if (pr->power.states[i].valid) {
653 			pr->power.count = i;
654 			if (pr->power.states[i].type >= ACPI_STATE_C2)
655 				pr->flags.power = 1;
656 		}
657 	}
658 
659 	return 0;
660 }
661 
662 /**
663  * acpi_idle_bm_check - checks if bus master activity was detected
664  */
665 static int acpi_idle_bm_check(void)
666 {
667 	u32 bm_status = 0;
668 
669 	if (bm_check_disable)
670 		return 0;
671 
672 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
673 	if (bm_status)
674 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
675 	/*
676 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
677 	 * the true state of bus mastering activity; forcing us to
678 	 * manually check the BMIDEA bit of each IDE channel.
679 	 */
680 	else if (errata.piix4.bmisx) {
681 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
682 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
683 			bm_status = 1;
684 	}
685 	return bm_status;
686 }
687 
688 /**
689  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
690  * @cx: cstate data
691  *
692  * Caller disables interrupt before call and enables interrupt after return.
693  */
694 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
695 {
696 	/* Don't trace irqs off for idle */
697 	stop_critical_timings();
698 	if (cx->entry_method == ACPI_CSTATE_FFH) {
699 		/* Call into architectural FFH based C-state */
700 		acpi_processor_ffh_cstate_enter(cx);
701 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
702 		acpi_safe_halt();
703 	} else {
704 		/* IO port based C-state */
705 		inb(cx->address);
706 		/* Dummy wait op - must do something useless after P_LVL2 read
707 		   because chipsets cannot guarantee that STPCLK# signal
708 		   gets asserted in time to freeze execution properly. */
709 		inl(acpi_gbl_FADT.xpm_timer_block.address);
710 	}
711 	start_critical_timings();
712 }
713 
714 /**
715  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
716  * @dev: the target CPU
717  * @drv: cpuidle driver containing cpuidle state info
718  * @index: index of target state
719  *
720  * This is equivalent to the HALT instruction.
721  */
722 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
723 		struct cpuidle_driver *drv, int index)
724 {
725 	struct acpi_processor *pr;
726 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
727 
728 	pr = __this_cpu_read(processors);
729 
730 	if (unlikely(!pr))
731 		return -EINVAL;
732 
733 	if (cx->entry_method == ACPI_CSTATE_FFH) {
734 		if (current_set_polling_and_test())
735 			return -EINVAL;
736 	}
737 
738 	lapic_timer_state_broadcast(pr, cx, 1);
739 	acpi_idle_do_entry(cx);
740 
741 	lapic_timer_state_broadcast(pr, cx, 0);
742 
743 	return index;
744 }
745 
746 
747 /**
748  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
749  * @dev: the target CPU
750  * @index: the index of suggested state
751  */
752 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
753 {
754 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
755 
756 	ACPI_FLUSH_CPU_CACHE();
757 
758 	while (1) {
759 
760 		if (cx->entry_method == ACPI_CSTATE_HALT)
761 			safe_halt();
762 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
763 			inb(cx->address);
764 			/* See comment in acpi_idle_do_entry() */
765 			inl(acpi_gbl_FADT.xpm_timer_block.address);
766 		} else
767 			return -ENODEV;
768 	}
769 
770 	/* Never reached */
771 	return 0;
772 }
773 
774 /**
775  * acpi_idle_enter_simple - enters an ACPI state without BM handling
776  * @dev: the target CPU
777  * @drv: cpuidle driver with cpuidle state information
778  * @index: the index of suggested state
779  */
780 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
781 		struct cpuidle_driver *drv, int index)
782 {
783 	struct acpi_processor *pr;
784 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
785 
786 	pr = __this_cpu_read(processors);
787 
788 	if (unlikely(!pr))
789 		return -EINVAL;
790 
791 	if (cx->entry_method == ACPI_CSTATE_FFH) {
792 		if (current_set_polling_and_test())
793 			return -EINVAL;
794 	}
795 
796 	/*
797 	 * Must be done before busmaster disable as we might need to
798 	 * access HPET !
799 	 */
800 	lapic_timer_state_broadcast(pr, cx, 1);
801 
802 	if (cx->type == ACPI_STATE_C3)
803 		ACPI_FLUSH_CPU_CACHE();
804 
805 	/* Tell the scheduler that we are going deep-idle: */
806 	sched_clock_idle_sleep_event();
807 	acpi_idle_do_entry(cx);
808 
809 	sched_clock_idle_wakeup_event(0);
810 
811 	lapic_timer_state_broadcast(pr, cx, 0);
812 	return index;
813 }
814 
815 static int c3_cpu_count;
816 static DEFINE_RAW_SPINLOCK(c3_lock);
817 
818 /**
819  * acpi_idle_enter_bm - enters C3 with proper BM handling
820  * @dev: the target CPU
821  * @drv: cpuidle driver containing state data
822  * @index: the index of suggested state
823  *
824  * If BM is detected, the deepest non-C3 idle state is entered instead.
825  */
826 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
827 		struct cpuidle_driver *drv, int index)
828 {
829 	struct acpi_processor *pr;
830 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
831 
832 	pr = __this_cpu_read(processors);
833 
834 	if (unlikely(!pr))
835 		return -EINVAL;
836 
837 	if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
838 		if (drv->safe_state_index >= 0) {
839 			return drv->states[drv->safe_state_index].enter(dev,
840 						drv, drv->safe_state_index);
841 		} else {
842 			acpi_safe_halt();
843 			return -EBUSY;
844 		}
845 	}
846 
847 	if (cx->entry_method == ACPI_CSTATE_FFH) {
848 		if (current_set_polling_and_test())
849 			return -EINVAL;
850 	}
851 
852 	acpi_unlazy_tlb(smp_processor_id());
853 
854 	/* Tell the scheduler that we are going deep-idle: */
855 	sched_clock_idle_sleep_event();
856 	/*
857 	 * Must be done before busmaster disable as we might need to
858 	 * access HPET !
859 	 */
860 	lapic_timer_state_broadcast(pr, cx, 1);
861 
862 	/*
863 	 * disable bus master
864 	 * bm_check implies we need ARB_DIS
865 	 * !bm_check implies we need cache flush
866 	 * bm_control implies whether we can do ARB_DIS
867 	 *
868 	 * That leaves a case where bm_check is set and bm_control is
869 	 * not set. In that case we cannot do much, we enter C3
870 	 * without doing anything.
871 	 */
872 	if (pr->flags.bm_check && pr->flags.bm_control) {
873 		raw_spin_lock(&c3_lock);
874 		c3_cpu_count++;
875 		/* Disable bus master arbitration when all CPUs are in C3 */
876 		if (c3_cpu_count == num_online_cpus())
877 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
878 		raw_spin_unlock(&c3_lock);
879 	} else if (!pr->flags.bm_check) {
880 		ACPI_FLUSH_CPU_CACHE();
881 	}
882 
883 	acpi_idle_do_entry(cx);
884 
885 	/* Re-enable bus master arbitration */
886 	if (pr->flags.bm_check && pr->flags.bm_control) {
887 		raw_spin_lock(&c3_lock);
888 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
889 		c3_cpu_count--;
890 		raw_spin_unlock(&c3_lock);
891 	}
892 
893 	sched_clock_idle_wakeup_event(0);
894 
895 	lapic_timer_state_broadcast(pr, cx, 0);
896 	return index;
897 }
898 
899 struct cpuidle_driver acpi_idle_driver = {
900 	.name =		"acpi_idle",
901 	.owner =	THIS_MODULE,
902 };
903 
904 /**
905  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
906  * device i.e. per-cpu data
907  *
908  * @pr: the ACPI processor
909  * @dev : the cpuidle device
910  */
911 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
912 					   struct cpuidle_device *dev)
913 {
914 	int i, count = CPUIDLE_DRIVER_STATE_START;
915 	struct acpi_processor_cx *cx;
916 
917 	if (!pr->flags.power_setup_done)
918 		return -EINVAL;
919 
920 	if (pr->flags.power == 0) {
921 		return -EINVAL;
922 	}
923 
924 	if (!dev)
925 		return -EINVAL;
926 
927 	dev->cpu = pr->id;
928 
929 	if (max_cstate == 0)
930 		max_cstate = 1;
931 
932 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
933 		cx = &pr->power.states[i];
934 
935 		if (!cx->valid)
936 			continue;
937 
938 #ifdef CONFIG_HOTPLUG_CPU
939 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
940 		    !pr->flags.has_cst &&
941 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
942 			continue;
943 #endif
944 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
945 
946 		count++;
947 		if (count == CPUIDLE_STATE_MAX)
948 			break;
949 	}
950 
951 	dev->state_count = count;
952 
953 	if (!count)
954 		return -EINVAL;
955 
956 	return 0;
957 }
958 
959 /**
960  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
961  * global state data i.e. idle routines
962  *
963  * @pr: the ACPI processor
964  */
965 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
966 {
967 	int i, count = CPUIDLE_DRIVER_STATE_START;
968 	struct acpi_processor_cx *cx;
969 	struct cpuidle_state *state;
970 	struct cpuidle_driver *drv = &acpi_idle_driver;
971 
972 	if (!pr->flags.power_setup_done)
973 		return -EINVAL;
974 
975 	if (pr->flags.power == 0)
976 		return -EINVAL;
977 
978 	drv->safe_state_index = -1;
979 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
980 		drv->states[i].name[0] = '\0';
981 		drv->states[i].desc[0] = '\0';
982 	}
983 
984 	if (max_cstate == 0)
985 		max_cstate = 1;
986 
987 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
988 		cx = &pr->power.states[i];
989 
990 		if (!cx->valid)
991 			continue;
992 
993 #ifdef CONFIG_HOTPLUG_CPU
994 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
995 		    !pr->flags.has_cst &&
996 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
997 			continue;
998 #endif
999 
1000 		state = &drv->states[count];
1001 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1002 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1003 		state->exit_latency = cx->latency;
1004 		state->target_residency = cx->latency * latency_factor;
1005 
1006 		state->flags = 0;
1007 		switch (cx->type) {
1008 			case ACPI_STATE_C1:
1009 			if (cx->entry_method == ACPI_CSTATE_FFH)
1010 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1011 
1012 			state->enter = acpi_idle_enter_c1;
1013 			state->enter_dead = acpi_idle_play_dead;
1014 			drv->safe_state_index = count;
1015 			break;
1016 
1017 			case ACPI_STATE_C2:
1018 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1019 			state->enter = acpi_idle_enter_simple;
1020 			state->enter_dead = acpi_idle_play_dead;
1021 			drv->safe_state_index = count;
1022 			break;
1023 
1024 			case ACPI_STATE_C3:
1025 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1026 			state->enter = pr->flags.bm_check ?
1027 					acpi_idle_enter_bm :
1028 					acpi_idle_enter_simple;
1029 			break;
1030 		}
1031 
1032 		count++;
1033 		if (count == CPUIDLE_STATE_MAX)
1034 			break;
1035 	}
1036 
1037 	drv->state_count = count;
1038 
1039 	if (!count)
1040 		return -EINVAL;
1041 
1042 	return 0;
1043 }
1044 
1045 int acpi_processor_hotplug(struct acpi_processor *pr)
1046 {
1047 	int ret = 0;
1048 	struct cpuidle_device *dev;
1049 
1050 	if (disabled_by_idle_boot_param())
1051 		return 0;
1052 
1053 	if (!pr)
1054 		return -EINVAL;
1055 
1056 	if (nocst) {
1057 		return -ENODEV;
1058 	}
1059 
1060 	if (!pr->flags.power_setup_done)
1061 		return -ENODEV;
1062 
1063 	dev = per_cpu(acpi_cpuidle_device, pr->id);
1064 	cpuidle_pause_and_lock();
1065 	cpuidle_disable_device(dev);
1066 	acpi_processor_get_power_info(pr);
1067 	if (pr->flags.power) {
1068 		acpi_processor_setup_cpuidle_cx(pr, dev);
1069 		ret = cpuidle_enable_device(dev);
1070 	}
1071 	cpuidle_resume_and_unlock();
1072 
1073 	return ret;
1074 }
1075 
1076 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1077 {
1078 	int cpu;
1079 	struct acpi_processor *_pr;
1080 	struct cpuidle_device *dev;
1081 
1082 	if (disabled_by_idle_boot_param())
1083 		return 0;
1084 
1085 	if (!pr)
1086 		return -EINVAL;
1087 
1088 	if (nocst)
1089 		return -ENODEV;
1090 
1091 	if (!pr->flags.power_setup_done)
1092 		return -ENODEV;
1093 
1094 	/*
1095 	 * FIXME:  Design the ACPI notification to make it once per
1096 	 * system instead of once per-cpu.  This condition is a hack
1097 	 * to make the code that updates C-States be called once.
1098 	 */
1099 
1100 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1101 
1102 		cpuidle_pause_and_lock();
1103 		/* Protect against cpu-hotplug */
1104 		get_online_cpus();
1105 
1106 		/* Disable all cpuidle devices */
1107 		for_each_online_cpu(cpu) {
1108 			_pr = per_cpu(processors, cpu);
1109 			if (!_pr || !_pr->flags.power_setup_done)
1110 				continue;
1111 			dev = per_cpu(acpi_cpuidle_device, cpu);
1112 			cpuidle_disable_device(dev);
1113 		}
1114 
1115 		/* Populate Updated C-state information */
1116 		acpi_processor_get_power_info(pr);
1117 		acpi_processor_setup_cpuidle_states(pr);
1118 
1119 		/* Enable all cpuidle devices */
1120 		for_each_online_cpu(cpu) {
1121 			_pr = per_cpu(processors, cpu);
1122 			if (!_pr || !_pr->flags.power_setup_done)
1123 				continue;
1124 			acpi_processor_get_power_info(_pr);
1125 			if (_pr->flags.power) {
1126 				dev = per_cpu(acpi_cpuidle_device, cpu);
1127 				acpi_processor_setup_cpuidle_cx(_pr, dev);
1128 				cpuidle_enable_device(dev);
1129 			}
1130 		}
1131 		put_online_cpus();
1132 		cpuidle_resume_and_unlock();
1133 	}
1134 
1135 	return 0;
1136 }
1137 
1138 static int acpi_processor_registered;
1139 
1140 int acpi_processor_power_init(struct acpi_processor *pr)
1141 {
1142 	acpi_status status = 0;
1143 	int retval;
1144 	struct cpuidle_device *dev;
1145 	static int first_run;
1146 
1147 	if (disabled_by_idle_boot_param())
1148 		return 0;
1149 
1150 	if (!first_run) {
1151 		dmi_check_system(processor_power_dmi_table);
1152 		max_cstate = acpi_processor_cstate_check(max_cstate);
1153 		if (max_cstate < ACPI_C_STATES_MAX)
1154 			printk(KERN_NOTICE
1155 			       "ACPI: processor limited to max C-state %d\n",
1156 			       max_cstate);
1157 		first_run++;
1158 	}
1159 
1160 	if (!pr)
1161 		return -EINVAL;
1162 
1163 	if (acpi_gbl_FADT.cst_control && !nocst) {
1164 		status =
1165 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1166 		if (ACPI_FAILURE(status)) {
1167 			ACPI_EXCEPTION((AE_INFO, status,
1168 					"Notifying BIOS of _CST ability failed"));
1169 		}
1170 	}
1171 
1172 	acpi_processor_get_power_info(pr);
1173 	pr->flags.power_setup_done = 1;
1174 
1175 	/*
1176 	 * Install the idle handler if processor power management is supported.
1177 	 * Note that we use previously set idle handler will be used on
1178 	 * platforms that only support C1.
1179 	 */
1180 	if (pr->flags.power) {
1181 		/* Register acpi_idle_driver if not already registered */
1182 		if (!acpi_processor_registered) {
1183 			acpi_processor_setup_cpuidle_states(pr);
1184 			retval = cpuidle_register_driver(&acpi_idle_driver);
1185 			if (retval)
1186 				return retval;
1187 			printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1188 					acpi_idle_driver.name);
1189 		}
1190 
1191 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1192 		if (!dev)
1193 			return -ENOMEM;
1194 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1195 
1196 		acpi_processor_setup_cpuidle_cx(pr, dev);
1197 
1198 		/* Register per-cpu cpuidle_device. Cpuidle driver
1199 		 * must already be registered before registering device
1200 		 */
1201 		retval = cpuidle_register_device(dev);
1202 		if (retval) {
1203 			if (acpi_processor_registered == 0)
1204 				cpuidle_unregister_driver(&acpi_idle_driver);
1205 			return retval;
1206 		}
1207 		acpi_processor_registered++;
1208 	}
1209 	return 0;
1210 }
1211 
1212 int acpi_processor_power_exit(struct acpi_processor *pr)
1213 {
1214 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1215 
1216 	if (disabled_by_idle_boot_param())
1217 		return 0;
1218 
1219 	if (pr->flags.power) {
1220 		cpuidle_unregister_device(dev);
1221 		acpi_processor_registered--;
1222 		if (acpi_processor_registered == 0)
1223 			cpuidle_unregister_driver(&acpi_idle_driver);
1224 	}
1225 
1226 	pr->flags.power_setup_done = 0;
1227 	return 0;
1228 }
1229