xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision fd589a8f)
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/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/acpi.h>
38 #include <linux/dmi.h>
39 #include <linux/moduleparam.h>
40 #include <linux/sched.h>	/* need_resched() */
41 #include <linux/pm_qos_params.h>
42 #include <linux/clockchips.h>
43 #include <linux/cpuidle.h>
44 #include <linux/irqflags.h>
45 
46 /*
47  * Include the apic definitions for x86 to have the APIC timer related defines
48  * available also for UP (on SMP it gets magically included via linux/smp.h).
49  * asm/acpi.h is not an option, as it would require more include magic. Also
50  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
51  */
52 #ifdef CONFIG_X86
53 #include <asm/apic.h>
54 #endif
55 
56 #include <asm/io.h>
57 #include <asm/uaccess.h>
58 
59 #include <acpi/acpi_bus.h>
60 #include <acpi/processor.h>
61 #include <asm/processor.h>
62 
63 #define ACPI_PROCESSOR_CLASS            "processor"
64 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
65 ACPI_MODULE_NAME("processor_idle");
66 #define ACPI_PROCESSOR_FILE_POWER	"power"
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 
77 static unsigned int latency_factor __read_mostly = 2;
78 module_param(latency_factor, uint, 0644);
79 
80 static s64 us_to_pm_timer_ticks(s64 t)
81 {
82 	return div64_u64(t * PM_TIMER_FREQUENCY, 1000000);
83 }
84 /*
85  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
86  * For now disable this. Probably a bug somewhere else.
87  *
88  * To skip this limit, boot/load with a large max_cstate limit.
89  */
90 static int set_max_cstate(const struct dmi_system_id *id)
91 {
92 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
93 		return 0;
94 
95 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
96 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
97 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
98 
99 	max_cstate = (long)id->driver_data;
100 
101 	return 0;
102 }
103 
104 /* Actually this shouldn't be __cpuinitdata, would be better to fix the
105    callers to only run once -AK */
106 static struct dmi_system_id __cpuinitdata processor_power_dmi_table[] = {
107 	{ set_max_cstate, "Clevo 5600D", {
108 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
109 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
110 	 (void *)2},
111 	{},
112 };
113 
114 
115 /*
116  * Callers should disable interrupts before the call and enable
117  * interrupts after return.
118  */
119 static void acpi_safe_halt(void)
120 {
121 	current_thread_info()->status &= ~TS_POLLING;
122 	/*
123 	 * TS_POLLING-cleared state must be visible before we
124 	 * test NEED_RESCHED:
125 	 */
126 	smp_mb();
127 	if (!need_resched()) {
128 		safe_halt();
129 		local_irq_disable();
130 	}
131 	current_thread_info()->status |= TS_POLLING;
132 }
133 
134 #ifdef ARCH_APICTIMER_STOPS_ON_C3
135 
136 /*
137  * Some BIOS implementations switch to C3 in the published C2 state.
138  * This seems to be a common problem on AMD boxen, but other vendors
139  * are affected too. We pick the most conservative approach: we assume
140  * that the local APIC stops in both C2 and C3.
141  */
142 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
143 				   struct acpi_processor_cx *cx)
144 {
145 	struct acpi_processor_power *pwr = &pr->power;
146 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
147 
148 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
149 		return;
150 
151 	if (boot_cpu_has(X86_FEATURE_AMDC1E))
152 		type = ACPI_STATE_C1;
153 
154 	/*
155 	 * Check, if one of the previous states already marked the lapic
156 	 * unstable
157 	 */
158 	if (pwr->timer_broadcast_on_state < state)
159 		return;
160 
161 	if (cx->type >= type)
162 		pr->power.timer_broadcast_on_state = state;
163 }
164 
165 static void lapic_timer_propagate_broadcast(void *arg)
166 {
167 	struct acpi_processor *pr = (struct acpi_processor *) arg;
168 	unsigned long reason;
169 
170 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
171 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
172 
173 	clockevents_notify(reason, &pr->id);
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 /*
206  * Suspend / resume control
207  */
208 static int acpi_idle_suspend;
209 static u32 saved_bm_rld;
210 
211 static void acpi_idle_bm_rld_save(void)
212 {
213 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
214 }
215 static void acpi_idle_bm_rld_restore(void)
216 {
217 	u32 resumed_bm_rld;
218 
219 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
220 
221 	if (resumed_bm_rld != saved_bm_rld)
222 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
223 }
224 
225 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
226 {
227 	if (acpi_idle_suspend == 1)
228 		return 0;
229 
230 	acpi_idle_bm_rld_save();
231 	acpi_idle_suspend = 1;
232 	return 0;
233 }
234 
235 int acpi_processor_resume(struct acpi_device * device)
236 {
237 	if (acpi_idle_suspend == 0)
238 		return 0;
239 
240 	acpi_idle_bm_rld_restore();
241 	acpi_idle_suspend = 0;
242 	return 0;
243 }
244 
245 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
246 static void tsc_check_state(int state)
247 {
248 	switch (boot_cpu_data.x86_vendor) {
249 	case X86_VENDOR_AMD:
250 	case X86_VENDOR_INTEL:
251 		/*
252 		 * AMD Fam10h TSC will tick in all
253 		 * C/P/S0/S1 states when this bit is set.
254 		 */
255 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
256 			return;
257 
258 		/*FALL THROUGH*/
259 	default:
260 		/* TSC could halt in idle, so notify users */
261 		if (state > ACPI_STATE_C1)
262 			mark_tsc_unstable("TSC halts in idle");
263 	}
264 }
265 #else
266 static void tsc_check_state(int state) { return; }
267 #endif
268 
269 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
270 {
271 
272 	if (!pr)
273 		return -EINVAL;
274 
275 	if (!pr->pblk)
276 		return -ENODEV;
277 
278 	/* if info is obtained from pblk/fadt, type equals state */
279 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
280 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
281 
282 #ifndef CONFIG_HOTPLUG_CPU
283 	/*
284 	 * Check for P_LVL2_UP flag before entering C2 and above on
285 	 * an SMP system.
286 	 */
287 	if ((num_online_cpus() > 1) &&
288 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
289 		return -ENODEV;
290 #endif
291 
292 	/* determine C2 and C3 address from pblk */
293 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
294 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
295 
296 	/* determine latencies from FADT */
297 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
298 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
299 
300 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
301 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
302 			  pr->power.states[ACPI_STATE_C2].address,
303 			  pr->power.states[ACPI_STATE_C3].address));
304 
305 	return 0;
306 }
307 
308 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
309 {
310 	if (!pr->power.states[ACPI_STATE_C1].valid) {
311 		/* set the first C-State to C1 */
312 		/* all processors need to support C1 */
313 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
314 		pr->power.states[ACPI_STATE_C1].valid = 1;
315 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
316 	}
317 	/* the C0 state only exists as a filler in our array */
318 	pr->power.states[ACPI_STATE_C0].valid = 1;
319 	return 0;
320 }
321 
322 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
323 {
324 	acpi_status status = 0;
325 	acpi_integer count;
326 	int current_count;
327 	int i;
328 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
329 	union acpi_object *cst;
330 
331 
332 	if (nocst)
333 		return -ENODEV;
334 
335 	current_count = 0;
336 
337 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
338 	if (ACPI_FAILURE(status)) {
339 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
340 		return -ENODEV;
341 	}
342 
343 	cst = buffer.pointer;
344 
345 	/* There must be at least 2 elements */
346 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
347 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
348 		status = -EFAULT;
349 		goto end;
350 	}
351 
352 	count = cst->package.elements[0].integer.value;
353 
354 	/* Validate number of power states. */
355 	if (count < 1 || count != cst->package.count - 1) {
356 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
357 		status = -EFAULT;
358 		goto end;
359 	}
360 
361 	/* Tell driver that at least _CST is supported. */
362 	pr->flags.has_cst = 1;
363 
364 	for (i = 1; i <= count; i++) {
365 		union acpi_object *element;
366 		union acpi_object *obj;
367 		struct acpi_power_register *reg;
368 		struct acpi_processor_cx cx;
369 
370 		memset(&cx, 0, sizeof(cx));
371 
372 		element = &(cst->package.elements[i]);
373 		if (element->type != ACPI_TYPE_PACKAGE)
374 			continue;
375 
376 		if (element->package.count != 4)
377 			continue;
378 
379 		obj = &(element->package.elements[0]);
380 
381 		if (obj->type != ACPI_TYPE_BUFFER)
382 			continue;
383 
384 		reg = (struct acpi_power_register *)obj->buffer.pointer;
385 
386 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
387 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
388 			continue;
389 
390 		/* There should be an easy way to extract an integer... */
391 		obj = &(element->package.elements[1]);
392 		if (obj->type != ACPI_TYPE_INTEGER)
393 			continue;
394 
395 		cx.type = obj->integer.value;
396 		/*
397 		 * Some buggy BIOSes won't list C1 in _CST -
398 		 * Let acpi_processor_get_power_info_default() handle them later
399 		 */
400 		if (i == 1 && cx.type != ACPI_STATE_C1)
401 			current_count++;
402 
403 		cx.address = reg->address;
404 		cx.index = current_count + 1;
405 
406 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
407 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
408 			if (acpi_processor_ffh_cstate_probe
409 					(pr->id, &cx, reg) == 0) {
410 				cx.entry_method = ACPI_CSTATE_FFH;
411 			} else if (cx.type == ACPI_STATE_C1) {
412 				/*
413 				 * C1 is a special case where FIXED_HARDWARE
414 				 * can be handled in non-MWAIT way as well.
415 				 * In that case, save this _CST entry info.
416 				 * Otherwise, ignore this info and continue.
417 				 */
418 				cx.entry_method = ACPI_CSTATE_HALT;
419 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
420 			} else {
421 				continue;
422 			}
423 			if (cx.type == ACPI_STATE_C1 &&
424 					(idle_halt || idle_nomwait)) {
425 				/*
426 				 * In most cases the C1 space_id obtained from
427 				 * _CST object is FIXED_HARDWARE access mode.
428 				 * But when the option of idle=halt is added,
429 				 * the entry_method type should be changed from
430 				 * CSTATE_FFH to CSTATE_HALT.
431 				 * When the option of idle=nomwait is added,
432 				 * the C1 entry_method type should be
433 				 * CSTATE_HALT.
434 				 */
435 				cx.entry_method = ACPI_CSTATE_HALT;
436 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
437 			}
438 		} else {
439 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
440 				 cx.address);
441 		}
442 
443 		if (cx.type == ACPI_STATE_C1) {
444 			cx.valid = 1;
445 		}
446 
447 		obj = &(element->package.elements[2]);
448 		if (obj->type != ACPI_TYPE_INTEGER)
449 			continue;
450 
451 		cx.latency = obj->integer.value;
452 
453 		obj = &(element->package.elements[3]);
454 		if (obj->type != ACPI_TYPE_INTEGER)
455 			continue;
456 
457 		cx.power = obj->integer.value;
458 
459 		current_count++;
460 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
461 
462 		/*
463 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
464 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
465 		 */
466 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
467 			printk(KERN_WARNING
468 			       "Limiting number of power states to max (%d)\n",
469 			       ACPI_PROCESSOR_MAX_POWER);
470 			printk(KERN_WARNING
471 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
472 			break;
473 		}
474 	}
475 
476 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
477 			  current_count));
478 
479 	/* Validate number of power states discovered */
480 	if (current_count < 2)
481 		status = -EFAULT;
482 
483       end:
484 	kfree(buffer.pointer);
485 
486 	return status;
487 }
488 
489 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
490 {
491 
492 	if (!cx->address)
493 		return;
494 
495 	/*
496 	 * C2 latency must be less than or equal to 100
497 	 * microseconds.
498 	 */
499 	else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
500 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
501 				  "latency too large [%d]\n", cx->latency));
502 		return;
503 	}
504 
505 	/*
506 	 * Otherwise we've met all of our C2 requirements.
507 	 * Normalize the C2 latency to expidite policy
508 	 */
509 	cx->valid = 1;
510 
511 	cx->latency_ticks = cx->latency;
512 
513 	return;
514 }
515 
516 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
517 					   struct acpi_processor_cx *cx)
518 {
519 	static int bm_check_flag = -1;
520 	static int bm_control_flag = -1;
521 
522 
523 	if (!cx->address)
524 		return;
525 
526 	/*
527 	 * C3 latency must be less than or equal to 1000
528 	 * microseconds.
529 	 */
530 	else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
531 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
532 				  "latency too large [%d]\n", cx->latency));
533 		return;
534 	}
535 
536 	/*
537 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
538 	 * DMA transfers are used by any ISA device to avoid livelock.
539 	 * Note that we could disable Type-F DMA (as recommended by
540 	 * the erratum), but this is known to disrupt certain ISA
541 	 * devices thus we take the conservative approach.
542 	 */
543 	else if (errata.piix4.fdma) {
544 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
545 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
546 		return;
547 	}
548 
549 	/* All the logic here assumes flags.bm_check is same across all CPUs */
550 	if (bm_check_flag == -1) {
551 		/* Determine whether bm_check is needed based on CPU  */
552 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
553 		bm_check_flag = pr->flags.bm_check;
554 		bm_control_flag = pr->flags.bm_control;
555 	} else {
556 		pr->flags.bm_check = bm_check_flag;
557 		pr->flags.bm_control = bm_control_flag;
558 	}
559 
560 	if (pr->flags.bm_check) {
561 		if (!pr->flags.bm_control) {
562 			if (pr->flags.has_cst != 1) {
563 				/* bus mastering control is necessary */
564 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
565 					"C3 support requires BM control\n"));
566 				return;
567 			} else {
568 				/* Here we enter C3 without bus mastering */
569 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
570 					"C3 support without BM control\n"));
571 			}
572 		}
573 	} else {
574 		/*
575 		 * WBINVD should be set in fadt, for C3 state to be
576 		 * supported on when bm_check is not required.
577 		 */
578 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
579 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
580 					  "Cache invalidation should work properly"
581 					  " for C3 to be enabled on SMP systems\n"));
582 			return;
583 		}
584 	}
585 
586 	/*
587 	 * Otherwise we've met all of our C3 requirements.
588 	 * Normalize the C3 latency to expidite policy.  Enable
589 	 * checking of bus mastering status (bm_check) so we can
590 	 * use this in our C3 policy
591 	 */
592 	cx->valid = 1;
593 
594 	cx->latency_ticks = cx->latency;
595 	/*
596 	 * On older chipsets, BM_RLD needs to be set
597 	 * in order for Bus Master activity to wake the
598 	 * system from C3.  Newer chipsets handle DMA
599 	 * during C3 automatically and BM_RLD is a NOP.
600 	 * In either case, the proper way to
601 	 * handle BM_RLD is to set it and leave it set.
602 	 */
603 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
604 
605 	return;
606 }
607 
608 static int acpi_processor_power_verify(struct acpi_processor *pr)
609 {
610 	unsigned int i;
611 	unsigned int working = 0;
612 
613 	pr->power.timer_broadcast_on_state = INT_MAX;
614 
615 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
616 		struct acpi_processor_cx *cx = &pr->power.states[i];
617 
618 		switch (cx->type) {
619 		case ACPI_STATE_C1:
620 			cx->valid = 1;
621 			break;
622 
623 		case ACPI_STATE_C2:
624 			acpi_processor_power_verify_c2(cx);
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 	smp_call_function_single(pr->id, lapic_timer_propagate_broadcast,
640 				 pr, 1);
641 
642 	return (working);
643 }
644 
645 static int acpi_processor_get_power_info(struct acpi_processor *pr)
646 {
647 	unsigned int i;
648 	int result;
649 
650 
651 	/* NOTE: the idle thread may not be running while calling
652 	 * this function */
653 
654 	/* Zero initialize all the C-states info. */
655 	memset(pr->power.states, 0, sizeof(pr->power.states));
656 
657 	result = acpi_processor_get_power_info_cst(pr);
658 	if (result == -ENODEV)
659 		result = acpi_processor_get_power_info_fadt(pr);
660 
661 	if (result)
662 		return result;
663 
664 	acpi_processor_get_power_info_default(pr);
665 
666 	pr->power.count = acpi_processor_power_verify(pr);
667 
668 	/*
669 	 * if one state of type C2 or C3 is available, mark this
670 	 * CPU as being "idle manageable"
671 	 */
672 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
673 		if (pr->power.states[i].valid) {
674 			pr->power.count = i;
675 			if (pr->power.states[i].type >= ACPI_STATE_C2)
676 				pr->flags.power = 1;
677 		}
678 	}
679 
680 	return 0;
681 }
682 
683 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
684 {
685 	struct acpi_processor *pr = seq->private;
686 	unsigned int i;
687 
688 
689 	if (!pr)
690 		goto end;
691 
692 	seq_printf(seq, "active state:            C%zd\n"
693 		   "max_cstate:              C%d\n"
694 		   "maximum allowed latency: %d usec\n",
695 		   pr->power.state ? pr->power.state - pr->power.states : 0,
696 		   max_cstate, pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
697 
698 	seq_puts(seq, "states:\n");
699 
700 	for (i = 1; i <= pr->power.count; i++) {
701 		seq_printf(seq, "   %cC%d:                  ",
702 			   (&pr->power.states[i] ==
703 			    pr->power.state ? '*' : ' '), i);
704 
705 		if (!pr->power.states[i].valid) {
706 			seq_puts(seq, "<not supported>\n");
707 			continue;
708 		}
709 
710 		switch (pr->power.states[i].type) {
711 		case ACPI_STATE_C1:
712 			seq_printf(seq, "type[C1] ");
713 			break;
714 		case ACPI_STATE_C2:
715 			seq_printf(seq, "type[C2] ");
716 			break;
717 		case ACPI_STATE_C3:
718 			seq_printf(seq, "type[C3] ");
719 			break;
720 		default:
721 			seq_printf(seq, "type[--] ");
722 			break;
723 		}
724 
725 		if (pr->power.states[i].promotion.state)
726 			seq_printf(seq, "promotion[C%zd] ",
727 				   (pr->power.states[i].promotion.state -
728 				    pr->power.states));
729 		else
730 			seq_puts(seq, "promotion[--] ");
731 
732 		if (pr->power.states[i].demotion.state)
733 			seq_printf(seq, "demotion[C%zd] ",
734 				   (pr->power.states[i].demotion.state -
735 				    pr->power.states));
736 		else
737 			seq_puts(seq, "demotion[--] ");
738 
739 		seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
740 			   pr->power.states[i].latency,
741 			   pr->power.states[i].usage,
742 			   (unsigned long long)pr->power.states[i].time);
743 	}
744 
745       end:
746 	return 0;
747 }
748 
749 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
750 {
751 	return single_open(file, acpi_processor_power_seq_show,
752 			   PDE(inode)->data);
753 }
754 
755 static const struct file_operations acpi_processor_power_fops = {
756 	.owner = THIS_MODULE,
757 	.open = acpi_processor_power_open_fs,
758 	.read = seq_read,
759 	.llseek = seq_lseek,
760 	.release = single_release,
761 };
762 
763 
764 /**
765  * acpi_idle_bm_check - checks if bus master activity was detected
766  */
767 static int acpi_idle_bm_check(void)
768 {
769 	u32 bm_status = 0;
770 
771 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
772 	if (bm_status)
773 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
774 	/*
775 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
776 	 * the true state of bus mastering activity; forcing us to
777 	 * manually check the BMIDEA bit of each IDE channel.
778 	 */
779 	else if (errata.piix4.bmisx) {
780 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
781 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
782 			bm_status = 1;
783 	}
784 	return bm_status;
785 }
786 
787 /**
788  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
789  * @cx: cstate data
790  *
791  * Caller disables interrupt before call and enables interrupt after return.
792  */
793 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
794 {
795 	/* Don't trace irqs off for idle */
796 	stop_critical_timings();
797 	if (cx->entry_method == ACPI_CSTATE_FFH) {
798 		/* Call into architectural FFH based C-state */
799 		acpi_processor_ffh_cstate_enter(cx);
800 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
801 		acpi_safe_halt();
802 	} else {
803 		int unused;
804 		/* IO port based C-state */
805 		inb(cx->address);
806 		/* Dummy wait op - must do something useless after P_LVL2 read
807 		   because chipsets cannot guarantee that STPCLK# signal
808 		   gets asserted in time to freeze execution properly. */
809 		unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
810 	}
811 	start_critical_timings();
812 }
813 
814 /**
815  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
816  * @dev: the target CPU
817  * @state: the state data
818  *
819  * This is equivalent to the HALT instruction.
820  */
821 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
822 			      struct cpuidle_state *state)
823 {
824 	ktime_t  kt1, kt2;
825 	s64 idle_time;
826 	struct acpi_processor *pr;
827 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
828 
829 	pr = __get_cpu_var(processors);
830 
831 	if (unlikely(!pr))
832 		return 0;
833 
834 	local_irq_disable();
835 
836 	/* Do not access any ACPI IO ports in suspend path */
837 	if (acpi_idle_suspend) {
838 		local_irq_enable();
839 		cpu_relax();
840 		return 0;
841 	}
842 
843 	lapic_timer_state_broadcast(pr, cx, 1);
844 	kt1 = ktime_get_real();
845 	acpi_idle_do_entry(cx);
846 	kt2 = ktime_get_real();
847 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
848 
849 	local_irq_enable();
850 	cx->usage++;
851 	lapic_timer_state_broadcast(pr, cx, 0);
852 
853 	return idle_time;
854 }
855 
856 /**
857  * acpi_idle_enter_simple - enters an ACPI state without BM handling
858  * @dev: the target CPU
859  * @state: the state data
860  */
861 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
862 				  struct cpuidle_state *state)
863 {
864 	struct acpi_processor *pr;
865 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
866 	ktime_t  kt1, kt2;
867 	s64 idle_time;
868 	s64 sleep_ticks = 0;
869 
870 	pr = __get_cpu_var(processors);
871 
872 	if (unlikely(!pr))
873 		return 0;
874 
875 	if (acpi_idle_suspend)
876 		return(acpi_idle_enter_c1(dev, state));
877 
878 	local_irq_disable();
879 	current_thread_info()->status &= ~TS_POLLING;
880 	/*
881 	 * TS_POLLING-cleared state must be visible before we test
882 	 * NEED_RESCHED:
883 	 */
884 	smp_mb();
885 
886 	if (unlikely(need_resched())) {
887 		current_thread_info()->status |= TS_POLLING;
888 		local_irq_enable();
889 		return 0;
890 	}
891 
892 	/*
893 	 * Must be done before busmaster disable as we might need to
894 	 * access HPET !
895 	 */
896 	lapic_timer_state_broadcast(pr, cx, 1);
897 
898 	if (cx->type == ACPI_STATE_C3)
899 		ACPI_FLUSH_CPU_CACHE();
900 
901 	kt1 = ktime_get_real();
902 	/* Tell the scheduler that we are going deep-idle: */
903 	sched_clock_idle_sleep_event();
904 	acpi_idle_do_entry(cx);
905 	kt2 = ktime_get_real();
906 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
907 
908 	sleep_ticks = us_to_pm_timer_ticks(idle_time);
909 
910 	/* Tell the scheduler how much we idled: */
911 	sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
912 
913 	local_irq_enable();
914 	current_thread_info()->status |= TS_POLLING;
915 
916 	cx->usage++;
917 
918 	lapic_timer_state_broadcast(pr, cx, 0);
919 	cx->time += sleep_ticks;
920 	return idle_time;
921 }
922 
923 static int c3_cpu_count;
924 static DEFINE_SPINLOCK(c3_lock);
925 
926 /**
927  * acpi_idle_enter_bm - enters C3 with proper BM handling
928  * @dev: the target CPU
929  * @state: the state data
930  *
931  * If BM is detected, the deepest non-C3 idle state is entered instead.
932  */
933 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
934 			      struct cpuidle_state *state)
935 {
936 	struct acpi_processor *pr;
937 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
938 	ktime_t  kt1, kt2;
939 	s64 idle_time;
940 	s64 sleep_ticks = 0;
941 
942 
943 	pr = __get_cpu_var(processors);
944 
945 	if (unlikely(!pr))
946 		return 0;
947 
948 	if (acpi_idle_suspend)
949 		return(acpi_idle_enter_c1(dev, state));
950 
951 	if (acpi_idle_bm_check()) {
952 		if (dev->safe_state) {
953 			dev->last_state = dev->safe_state;
954 			return dev->safe_state->enter(dev, dev->safe_state);
955 		} else {
956 			local_irq_disable();
957 			acpi_safe_halt();
958 			local_irq_enable();
959 			return 0;
960 		}
961 	}
962 
963 	local_irq_disable();
964 	current_thread_info()->status &= ~TS_POLLING;
965 	/*
966 	 * TS_POLLING-cleared state must be visible before we test
967 	 * NEED_RESCHED:
968 	 */
969 	smp_mb();
970 
971 	if (unlikely(need_resched())) {
972 		current_thread_info()->status |= TS_POLLING;
973 		local_irq_enable();
974 		return 0;
975 	}
976 
977 	acpi_unlazy_tlb(smp_processor_id());
978 
979 	/* Tell the scheduler that we are going deep-idle: */
980 	sched_clock_idle_sleep_event();
981 	/*
982 	 * Must be done before busmaster disable as we might need to
983 	 * access HPET !
984 	 */
985 	lapic_timer_state_broadcast(pr, cx, 1);
986 
987 	kt1 = ktime_get_real();
988 	/*
989 	 * disable bus master
990 	 * bm_check implies we need ARB_DIS
991 	 * !bm_check implies we need cache flush
992 	 * bm_control implies whether we can do ARB_DIS
993 	 *
994 	 * That leaves a case where bm_check is set and bm_control is
995 	 * not set. In that case we cannot do much, we enter C3
996 	 * without doing anything.
997 	 */
998 	if (pr->flags.bm_check && pr->flags.bm_control) {
999 		spin_lock(&c3_lock);
1000 		c3_cpu_count++;
1001 		/* Disable bus master arbitration when all CPUs are in C3 */
1002 		if (c3_cpu_count == num_online_cpus())
1003 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
1004 		spin_unlock(&c3_lock);
1005 	} else if (!pr->flags.bm_check) {
1006 		ACPI_FLUSH_CPU_CACHE();
1007 	}
1008 
1009 	acpi_idle_do_entry(cx);
1010 
1011 	/* Re-enable bus master arbitration */
1012 	if (pr->flags.bm_check && pr->flags.bm_control) {
1013 		spin_lock(&c3_lock);
1014 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
1015 		c3_cpu_count--;
1016 		spin_unlock(&c3_lock);
1017 	}
1018 	kt2 = ktime_get_real();
1019 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
1020 
1021 	sleep_ticks = us_to_pm_timer_ticks(idle_time);
1022 	/* Tell the scheduler how much we idled: */
1023 	sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
1024 
1025 	local_irq_enable();
1026 	current_thread_info()->status |= TS_POLLING;
1027 
1028 	cx->usage++;
1029 
1030 	lapic_timer_state_broadcast(pr, cx, 0);
1031 	cx->time += sleep_ticks;
1032 	return idle_time;
1033 }
1034 
1035 struct cpuidle_driver acpi_idle_driver = {
1036 	.name =		"acpi_idle",
1037 	.owner =	THIS_MODULE,
1038 };
1039 
1040 /**
1041  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1042  * @pr: the ACPI processor
1043  */
1044 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1045 {
1046 	int i, count = CPUIDLE_DRIVER_STATE_START;
1047 	struct acpi_processor_cx *cx;
1048 	struct cpuidle_state *state;
1049 	struct cpuidle_device *dev = &pr->power.dev;
1050 
1051 	if (!pr->flags.power_setup_done)
1052 		return -EINVAL;
1053 
1054 	if (pr->flags.power == 0) {
1055 		return -EINVAL;
1056 	}
1057 
1058 	dev->cpu = pr->id;
1059 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1060 		dev->states[i].name[0] = '\0';
1061 		dev->states[i].desc[0] = '\0';
1062 	}
1063 
1064 	if (max_cstate == 0)
1065 		max_cstate = 1;
1066 
1067 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1068 		cx = &pr->power.states[i];
1069 		state = &dev->states[count];
1070 
1071 		if (!cx->valid)
1072 			continue;
1073 
1074 #ifdef CONFIG_HOTPLUG_CPU
1075 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1076 		    !pr->flags.has_cst &&
1077 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1078 			continue;
1079 #endif
1080 		cpuidle_set_statedata(state, cx);
1081 
1082 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1083 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1084 		state->exit_latency = cx->latency;
1085 		state->target_residency = cx->latency * latency_factor;
1086 		state->power_usage = cx->power;
1087 
1088 		state->flags = 0;
1089 		switch (cx->type) {
1090 			case ACPI_STATE_C1:
1091 			state->flags |= CPUIDLE_FLAG_SHALLOW;
1092 			if (cx->entry_method == ACPI_CSTATE_FFH)
1093 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1094 
1095 			state->enter = acpi_idle_enter_c1;
1096 			dev->safe_state = state;
1097 			break;
1098 
1099 			case ACPI_STATE_C2:
1100 			state->flags |= CPUIDLE_FLAG_BALANCED;
1101 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1102 			state->enter = acpi_idle_enter_simple;
1103 			dev->safe_state = state;
1104 			break;
1105 
1106 			case ACPI_STATE_C3:
1107 			state->flags |= CPUIDLE_FLAG_DEEP;
1108 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1109 			state->flags |= CPUIDLE_FLAG_CHECK_BM;
1110 			state->enter = pr->flags.bm_check ?
1111 					acpi_idle_enter_bm :
1112 					acpi_idle_enter_simple;
1113 			break;
1114 		}
1115 
1116 		count++;
1117 		if (count == CPUIDLE_STATE_MAX)
1118 			break;
1119 	}
1120 
1121 	dev->state_count = count;
1122 
1123 	if (!count)
1124 		return -EINVAL;
1125 
1126 	return 0;
1127 }
1128 
1129 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1130 {
1131 	int ret = 0;
1132 
1133 	if (boot_option_idle_override)
1134 		return 0;
1135 
1136 	if (!pr)
1137 		return -EINVAL;
1138 
1139 	if (nocst) {
1140 		return -ENODEV;
1141 	}
1142 
1143 	if (!pr->flags.power_setup_done)
1144 		return -ENODEV;
1145 
1146 	cpuidle_pause_and_lock();
1147 	cpuidle_disable_device(&pr->power.dev);
1148 	acpi_processor_get_power_info(pr);
1149 	if (pr->flags.power) {
1150 		acpi_processor_setup_cpuidle(pr);
1151 		ret = cpuidle_enable_device(&pr->power.dev);
1152 	}
1153 	cpuidle_resume_and_unlock();
1154 
1155 	return ret;
1156 }
1157 
1158 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1159 			      struct acpi_device *device)
1160 {
1161 	acpi_status status = 0;
1162 	static int first_run;
1163 	struct proc_dir_entry *entry = NULL;
1164 	unsigned int i;
1165 
1166 	if (boot_option_idle_override)
1167 		return 0;
1168 
1169 	if (!first_run) {
1170 		if (idle_halt) {
1171 			/*
1172 			 * When the boot option of "idle=halt" is added, halt
1173 			 * is used for CPU IDLE.
1174 			 * In such case C2/C3 is meaningless. So the max_cstate
1175 			 * is set to one.
1176 			 */
1177 			max_cstate = 1;
1178 		}
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 		acpi_processor_setup_cpuidle(pr);
1210 		if (cpuidle_register_device(&pr->power.dev))
1211 			return -EIO;
1212 
1213 		printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1214 		for (i = 1; i <= pr->power.count; i++)
1215 			if (pr->power.states[i].valid)
1216 				printk(" C%d[C%d]", i,
1217 				       pr->power.states[i].type);
1218 		printk(")\n");
1219 	}
1220 
1221 	/* 'power' [R] */
1222 	entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1223 				 S_IRUGO, acpi_device_dir(device),
1224 				 &acpi_processor_power_fops,
1225 				 acpi_driver_data(device));
1226 	if (!entry)
1227 		return -EIO;
1228 	return 0;
1229 }
1230 
1231 int acpi_processor_power_exit(struct acpi_processor *pr,
1232 			      struct acpi_device *device)
1233 {
1234 	if (boot_option_idle_override)
1235 		return 0;
1236 
1237 	cpuidle_unregister_device(&pr->power.dev);
1238 	pr->flags.power_setup_done = 0;
1239 
1240 	if (acpi_device_dir(device))
1241 		remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1242 				  acpi_device_dir(device));
1243 
1244 	return 0;
1245 }
1246