xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision b627b4ed)
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 acpi_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 	/*
152 	 * Check, if one of the previous states already marked the lapic
153 	 * unstable
154 	 */
155 	if (pwr->timer_broadcast_on_state < state)
156 		return;
157 
158 	if (cx->type >= type)
159 		pr->power.timer_broadcast_on_state = state;
160 }
161 
162 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr)
163 {
164 	unsigned long reason;
165 
166 	reason = pr->power.timer_broadcast_on_state < INT_MAX ?
167 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
168 
169 	clockevents_notify(reason, &pr->id);
170 }
171 
172 /* Power(C) State timer broadcast control */
173 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
174 				       struct acpi_processor_cx *cx,
175 				       int broadcast)
176 {
177 	int state = cx - pr->power.states;
178 
179 	if (state >= pr->power.timer_broadcast_on_state) {
180 		unsigned long reason;
181 
182 		reason = broadcast ?  CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
183 			CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
184 		clockevents_notify(reason, &pr->id);
185 	}
186 }
187 
188 #else
189 
190 static void acpi_timer_check_state(int state, struct acpi_processor *pr,
191 				   struct acpi_processor_cx *cstate) { }
192 static void acpi_propagate_timer_broadcast(struct acpi_processor *pr) { }
193 static void acpi_state_timer_broadcast(struct acpi_processor *pr,
194 				       struct acpi_processor_cx *cx,
195 				       int broadcast)
196 {
197 }
198 
199 #endif
200 
201 /*
202  * Suspend / resume control
203  */
204 static int acpi_idle_suspend;
205 
206 int acpi_processor_suspend(struct acpi_device * device, pm_message_t state)
207 {
208 	acpi_idle_suspend = 1;
209 	return 0;
210 }
211 
212 int acpi_processor_resume(struct acpi_device * device)
213 {
214 	acpi_idle_suspend = 0;
215 	return 0;
216 }
217 
218 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
219 static int tsc_halts_in_c(int state)
220 {
221 	switch (boot_cpu_data.x86_vendor) {
222 	case X86_VENDOR_AMD:
223 	case X86_VENDOR_INTEL:
224 		/*
225 		 * AMD Fam10h TSC will tick in all
226 		 * C/P/S0/S1 states when this bit is set.
227 		 */
228 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
229 			return 0;
230 
231 		/*FALL THROUGH*/
232 	default:
233 		return state > ACPI_STATE_C1;
234 	}
235 }
236 #endif
237 
238 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
239 {
240 
241 	if (!pr)
242 		return -EINVAL;
243 
244 	if (!pr->pblk)
245 		return -ENODEV;
246 
247 	/* if info is obtained from pblk/fadt, type equals state */
248 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
249 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
250 
251 #ifndef CONFIG_HOTPLUG_CPU
252 	/*
253 	 * Check for P_LVL2_UP flag before entering C2 and above on
254 	 * an SMP system.
255 	 */
256 	if ((num_online_cpus() > 1) &&
257 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
258 		return -ENODEV;
259 #endif
260 
261 	/* determine C2 and C3 address from pblk */
262 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
263 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
264 
265 	/* determine latencies from FADT */
266 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.C2latency;
267 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.C3latency;
268 
269 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
270 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
271 			  pr->power.states[ACPI_STATE_C2].address,
272 			  pr->power.states[ACPI_STATE_C3].address));
273 
274 	return 0;
275 }
276 
277 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
278 {
279 	if (!pr->power.states[ACPI_STATE_C1].valid) {
280 		/* set the first C-State to C1 */
281 		/* all processors need to support C1 */
282 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
283 		pr->power.states[ACPI_STATE_C1].valid = 1;
284 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
285 	}
286 	/* the C0 state only exists as a filler in our array */
287 	pr->power.states[ACPI_STATE_C0].valid = 1;
288 	return 0;
289 }
290 
291 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
292 {
293 	acpi_status status = 0;
294 	acpi_integer count;
295 	int current_count;
296 	int i;
297 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
298 	union acpi_object *cst;
299 
300 
301 	if (nocst)
302 		return -ENODEV;
303 
304 	current_count = 0;
305 
306 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
307 	if (ACPI_FAILURE(status)) {
308 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
309 		return -ENODEV;
310 	}
311 
312 	cst = buffer.pointer;
313 
314 	/* There must be at least 2 elements */
315 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
316 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
317 		status = -EFAULT;
318 		goto end;
319 	}
320 
321 	count = cst->package.elements[0].integer.value;
322 
323 	/* Validate number of power states. */
324 	if (count < 1 || count != cst->package.count - 1) {
325 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
326 		status = -EFAULT;
327 		goto end;
328 	}
329 
330 	/* Tell driver that at least _CST is supported. */
331 	pr->flags.has_cst = 1;
332 
333 	for (i = 1; i <= count; i++) {
334 		union acpi_object *element;
335 		union acpi_object *obj;
336 		struct acpi_power_register *reg;
337 		struct acpi_processor_cx cx;
338 
339 		memset(&cx, 0, sizeof(cx));
340 
341 		element = &(cst->package.elements[i]);
342 		if (element->type != ACPI_TYPE_PACKAGE)
343 			continue;
344 
345 		if (element->package.count != 4)
346 			continue;
347 
348 		obj = &(element->package.elements[0]);
349 
350 		if (obj->type != ACPI_TYPE_BUFFER)
351 			continue;
352 
353 		reg = (struct acpi_power_register *)obj->buffer.pointer;
354 
355 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
356 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
357 			continue;
358 
359 		/* There should be an easy way to extract an integer... */
360 		obj = &(element->package.elements[1]);
361 		if (obj->type != ACPI_TYPE_INTEGER)
362 			continue;
363 
364 		cx.type = obj->integer.value;
365 		/*
366 		 * Some buggy BIOSes won't list C1 in _CST -
367 		 * Let acpi_processor_get_power_info_default() handle them later
368 		 */
369 		if (i == 1 && cx.type != ACPI_STATE_C1)
370 			current_count++;
371 
372 		cx.address = reg->address;
373 		cx.index = current_count + 1;
374 
375 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
376 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
377 			if (acpi_processor_ffh_cstate_probe
378 					(pr->id, &cx, reg) == 0) {
379 				cx.entry_method = ACPI_CSTATE_FFH;
380 			} else if (cx.type == ACPI_STATE_C1) {
381 				/*
382 				 * C1 is a special case where FIXED_HARDWARE
383 				 * can be handled in non-MWAIT way as well.
384 				 * In that case, save this _CST entry info.
385 				 * Otherwise, ignore this info and continue.
386 				 */
387 				cx.entry_method = ACPI_CSTATE_HALT;
388 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
389 			} else {
390 				continue;
391 			}
392 			if (cx.type == ACPI_STATE_C1 &&
393 					(idle_halt || idle_nomwait)) {
394 				/*
395 				 * In most cases the C1 space_id obtained from
396 				 * _CST object is FIXED_HARDWARE access mode.
397 				 * But when the option of idle=halt is added,
398 				 * the entry_method type should be changed from
399 				 * CSTATE_FFH to CSTATE_HALT.
400 				 * When the option of idle=nomwait is added,
401 				 * the C1 entry_method type should be
402 				 * CSTATE_HALT.
403 				 */
404 				cx.entry_method = ACPI_CSTATE_HALT;
405 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
406 			}
407 		} else {
408 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
409 				 cx.address);
410 		}
411 
412 		if (cx.type == ACPI_STATE_C1) {
413 			cx.valid = 1;
414 		}
415 
416 		obj = &(element->package.elements[2]);
417 		if (obj->type != ACPI_TYPE_INTEGER)
418 			continue;
419 
420 		cx.latency = obj->integer.value;
421 
422 		obj = &(element->package.elements[3]);
423 		if (obj->type != ACPI_TYPE_INTEGER)
424 			continue;
425 
426 		cx.power = obj->integer.value;
427 
428 		current_count++;
429 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
430 
431 		/*
432 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
433 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
434 		 */
435 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
436 			printk(KERN_WARNING
437 			       "Limiting number of power states to max (%d)\n",
438 			       ACPI_PROCESSOR_MAX_POWER);
439 			printk(KERN_WARNING
440 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
441 			break;
442 		}
443 	}
444 
445 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
446 			  current_count));
447 
448 	/* Validate number of power states discovered */
449 	if (current_count < 2)
450 		status = -EFAULT;
451 
452       end:
453 	kfree(buffer.pointer);
454 
455 	return status;
456 }
457 
458 static void acpi_processor_power_verify_c2(struct acpi_processor_cx *cx)
459 {
460 
461 	if (!cx->address)
462 		return;
463 
464 	/*
465 	 * C2 latency must be less than or equal to 100
466 	 * microseconds.
467 	 */
468 	else if (cx->latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
469 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
470 				  "latency too large [%d]\n", cx->latency));
471 		return;
472 	}
473 
474 	/*
475 	 * Otherwise we've met all of our C2 requirements.
476 	 * Normalize the C2 latency to expidite policy
477 	 */
478 	cx->valid = 1;
479 
480 	cx->latency_ticks = cx->latency;
481 
482 	return;
483 }
484 
485 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
486 					   struct acpi_processor_cx *cx)
487 {
488 	static int bm_check_flag;
489 
490 
491 	if (!cx->address)
492 		return;
493 
494 	/*
495 	 * C3 latency must be less than or equal to 1000
496 	 * microseconds.
497 	 */
498 	else if (cx->latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
499 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
500 				  "latency too large [%d]\n", cx->latency));
501 		return;
502 	}
503 
504 	/*
505 	 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
506 	 * DMA transfers are used by any ISA device to avoid livelock.
507 	 * Note that we could disable Type-F DMA (as recommended by
508 	 * the erratum), but this is known to disrupt certain ISA
509 	 * devices thus we take the conservative approach.
510 	 */
511 	else if (errata.piix4.fdma) {
512 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
513 				  "C3 not supported on PIIX4 with Type-F DMA\n"));
514 		return;
515 	}
516 
517 	/* All the logic here assumes flags.bm_check is same across all CPUs */
518 	if (!bm_check_flag) {
519 		/* Determine whether bm_check is needed based on CPU  */
520 		acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
521 		bm_check_flag = pr->flags.bm_check;
522 	} else {
523 		pr->flags.bm_check = bm_check_flag;
524 	}
525 
526 	if (pr->flags.bm_check) {
527 		if (!pr->flags.bm_control) {
528 			if (pr->flags.has_cst != 1) {
529 				/* bus mastering control is necessary */
530 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
531 					"C3 support requires BM control\n"));
532 				return;
533 			} else {
534 				/* Here we enter C3 without bus mastering */
535 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
536 					"C3 support without BM control\n"));
537 			}
538 		}
539 	} else {
540 		/*
541 		 * WBINVD should be set in fadt, for C3 state to be
542 		 * supported on when bm_check is not required.
543 		 */
544 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
545 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
546 					  "Cache invalidation should work properly"
547 					  " for C3 to be enabled on SMP systems\n"));
548 			return;
549 		}
550 	}
551 
552 	/*
553 	 * Otherwise we've met all of our C3 requirements.
554 	 * Normalize the C3 latency to expidite policy.  Enable
555 	 * checking of bus mastering status (bm_check) so we can
556 	 * use this in our C3 policy
557 	 */
558 	cx->valid = 1;
559 
560 	cx->latency_ticks = cx->latency;
561 	/*
562 	 * On older chipsets, BM_RLD needs to be set
563 	 * in order for Bus Master activity to wake the
564 	 * system from C3.  Newer chipsets handle DMA
565 	 * during C3 automatically and BM_RLD is a NOP.
566 	 * In either case, the proper way to
567 	 * handle BM_RLD is to set it and leave it set.
568 	 */
569 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
570 
571 	return;
572 }
573 
574 static int acpi_processor_power_verify(struct acpi_processor *pr)
575 {
576 	unsigned int i;
577 	unsigned int working = 0;
578 
579 	pr->power.timer_broadcast_on_state = INT_MAX;
580 
581 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
582 		struct acpi_processor_cx *cx = &pr->power.states[i];
583 
584 		switch (cx->type) {
585 		case ACPI_STATE_C1:
586 			cx->valid = 1;
587 			break;
588 
589 		case ACPI_STATE_C2:
590 			acpi_processor_power_verify_c2(cx);
591 			if (cx->valid)
592 				acpi_timer_check_state(i, pr, cx);
593 			break;
594 
595 		case ACPI_STATE_C3:
596 			acpi_processor_power_verify_c3(pr, cx);
597 			if (cx->valid)
598 				acpi_timer_check_state(i, pr, cx);
599 			break;
600 		}
601 
602 		if (cx->valid)
603 			working++;
604 	}
605 
606 	acpi_propagate_timer_broadcast(pr);
607 
608 	return (working);
609 }
610 
611 static int acpi_processor_get_power_info(struct acpi_processor *pr)
612 {
613 	unsigned int i;
614 	int result;
615 
616 
617 	/* NOTE: the idle thread may not be running while calling
618 	 * this function */
619 
620 	/* Zero initialize all the C-states info. */
621 	memset(pr->power.states, 0, sizeof(pr->power.states));
622 
623 	result = acpi_processor_get_power_info_cst(pr);
624 	if (result == -ENODEV)
625 		result = acpi_processor_get_power_info_fadt(pr);
626 
627 	if (result)
628 		return result;
629 
630 	acpi_processor_get_power_info_default(pr);
631 
632 	pr->power.count = acpi_processor_power_verify(pr);
633 
634 	/*
635 	 * if one state of type C2 or C3 is available, mark this
636 	 * CPU as being "idle manageable"
637 	 */
638 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
639 		if (pr->power.states[i].valid) {
640 			pr->power.count = i;
641 			if (pr->power.states[i].type >= ACPI_STATE_C2)
642 				pr->flags.power = 1;
643 		}
644 	}
645 
646 	return 0;
647 }
648 
649 static int acpi_processor_power_seq_show(struct seq_file *seq, void *offset)
650 {
651 	struct acpi_processor *pr = seq->private;
652 	unsigned int i;
653 
654 
655 	if (!pr)
656 		goto end;
657 
658 	seq_printf(seq, "active state:            C%zd\n"
659 		   "max_cstate:              C%d\n"
660 		   "bus master activity:     %08x\n"
661 		   "maximum allowed latency: %d usec\n",
662 		   pr->power.state ? pr->power.state - pr->power.states : 0,
663 		   max_cstate, (unsigned)pr->power.bm_activity,
664 		   pm_qos_requirement(PM_QOS_CPU_DMA_LATENCY));
665 
666 	seq_puts(seq, "states:\n");
667 
668 	for (i = 1; i <= pr->power.count; i++) {
669 		seq_printf(seq, "   %cC%d:                  ",
670 			   (&pr->power.states[i] ==
671 			    pr->power.state ? '*' : ' '), i);
672 
673 		if (!pr->power.states[i].valid) {
674 			seq_puts(seq, "<not supported>\n");
675 			continue;
676 		}
677 
678 		switch (pr->power.states[i].type) {
679 		case ACPI_STATE_C1:
680 			seq_printf(seq, "type[C1] ");
681 			break;
682 		case ACPI_STATE_C2:
683 			seq_printf(seq, "type[C2] ");
684 			break;
685 		case ACPI_STATE_C3:
686 			seq_printf(seq, "type[C3] ");
687 			break;
688 		default:
689 			seq_printf(seq, "type[--] ");
690 			break;
691 		}
692 
693 		if (pr->power.states[i].promotion.state)
694 			seq_printf(seq, "promotion[C%zd] ",
695 				   (pr->power.states[i].promotion.state -
696 				    pr->power.states));
697 		else
698 			seq_puts(seq, "promotion[--] ");
699 
700 		if (pr->power.states[i].demotion.state)
701 			seq_printf(seq, "demotion[C%zd] ",
702 				   (pr->power.states[i].demotion.state -
703 				    pr->power.states));
704 		else
705 			seq_puts(seq, "demotion[--] ");
706 
707 		seq_printf(seq, "latency[%03d] usage[%08d] duration[%020llu]\n",
708 			   pr->power.states[i].latency,
709 			   pr->power.states[i].usage,
710 			   (unsigned long long)pr->power.states[i].time);
711 	}
712 
713       end:
714 	return 0;
715 }
716 
717 static int acpi_processor_power_open_fs(struct inode *inode, struct file *file)
718 {
719 	return single_open(file, acpi_processor_power_seq_show,
720 			   PDE(inode)->data);
721 }
722 
723 static const struct file_operations acpi_processor_power_fops = {
724 	.owner = THIS_MODULE,
725 	.open = acpi_processor_power_open_fs,
726 	.read = seq_read,
727 	.llseek = seq_lseek,
728 	.release = single_release,
729 };
730 
731 
732 /**
733  * acpi_idle_bm_check - checks if bus master activity was detected
734  */
735 static int acpi_idle_bm_check(void)
736 {
737 	u32 bm_status = 0;
738 
739 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
740 	if (bm_status)
741 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
742 	/*
743 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
744 	 * the true state of bus mastering activity; forcing us to
745 	 * manually check the BMIDEA bit of each IDE channel.
746 	 */
747 	else if (errata.piix4.bmisx) {
748 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
749 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
750 			bm_status = 1;
751 	}
752 	return bm_status;
753 }
754 
755 /**
756  * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
757  * @cx: cstate data
758  *
759  * Caller disables interrupt before call and enables interrupt after return.
760  */
761 static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
762 {
763 	/* Don't trace irqs off for idle */
764 	stop_critical_timings();
765 	if (cx->entry_method == ACPI_CSTATE_FFH) {
766 		/* Call into architectural FFH based C-state */
767 		acpi_processor_ffh_cstate_enter(cx);
768 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
769 		acpi_safe_halt();
770 	} else {
771 		int unused;
772 		/* IO port based C-state */
773 		inb(cx->address);
774 		/* Dummy wait op - must do something useless after P_LVL2 read
775 		   because chipsets cannot guarantee that STPCLK# signal
776 		   gets asserted in time to freeze execution properly. */
777 		unused = inl(acpi_gbl_FADT.xpm_timer_block.address);
778 	}
779 	start_critical_timings();
780 }
781 
782 /**
783  * acpi_idle_enter_c1 - enters an ACPI C1 state-type
784  * @dev: the target CPU
785  * @state: the state data
786  *
787  * This is equivalent to the HALT instruction.
788  */
789 static int acpi_idle_enter_c1(struct cpuidle_device *dev,
790 			      struct cpuidle_state *state)
791 {
792 	ktime_t  kt1, kt2;
793 	s64 idle_time;
794 	struct acpi_processor *pr;
795 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
796 
797 	pr = __get_cpu_var(processors);
798 
799 	if (unlikely(!pr))
800 		return 0;
801 
802 	local_irq_disable();
803 
804 	/* Do not access any ACPI IO ports in suspend path */
805 	if (acpi_idle_suspend) {
806 		acpi_safe_halt();
807 		local_irq_enable();
808 		return 0;
809 	}
810 
811 	kt1 = ktime_get_real();
812 	acpi_idle_do_entry(cx);
813 	kt2 = ktime_get_real();
814 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
815 
816 	local_irq_enable();
817 	cx->usage++;
818 
819 	return idle_time;
820 }
821 
822 /**
823  * acpi_idle_enter_simple - enters an ACPI state without BM handling
824  * @dev: the target CPU
825  * @state: the state data
826  */
827 static int acpi_idle_enter_simple(struct cpuidle_device *dev,
828 				  struct cpuidle_state *state)
829 {
830 	struct acpi_processor *pr;
831 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
832 	ktime_t  kt1, kt2;
833 	s64 idle_time;
834 	s64 sleep_ticks = 0;
835 
836 	pr = __get_cpu_var(processors);
837 
838 	if (unlikely(!pr))
839 		return 0;
840 
841 	if (acpi_idle_suspend)
842 		return(acpi_idle_enter_c1(dev, state));
843 
844 	local_irq_disable();
845 	current_thread_info()->status &= ~TS_POLLING;
846 	/*
847 	 * TS_POLLING-cleared state must be visible before we test
848 	 * NEED_RESCHED:
849 	 */
850 	smp_mb();
851 
852 	if (unlikely(need_resched())) {
853 		current_thread_info()->status |= TS_POLLING;
854 		local_irq_enable();
855 		return 0;
856 	}
857 
858 	/*
859 	 * Must be done before busmaster disable as we might need to
860 	 * access HPET !
861 	 */
862 	acpi_state_timer_broadcast(pr, cx, 1);
863 
864 	if (cx->type == ACPI_STATE_C3)
865 		ACPI_FLUSH_CPU_CACHE();
866 
867 	kt1 = ktime_get_real();
868 	/* Tell the scheduler that we are going deep-idle: */
869 	sched_clock_idle_sleep_event();
870 	acpi_idle_do_entry(cx);
871 	kt2 = ktime_get_real();
872 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
873 
874 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
875 	/* TSC could halt in idle, so notify users */
876 	if (tsc_halts_in_c(cx->type))
877 		mark_tsc_unstable("TSC halts in idle");;
878 #endif
879 	sleep_ticks = us_to_pm_timer_ticks(idle_time);
880 
881 	/* Tell the scheduler how much we idled: */
882 	sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
883 
884 	local_irq_enable();
885 	current_thread_info()->status |= TS_POLLING;
886 
887 	cx->usage++;
888 
889 	acpi_state_timer_broadcast(pr, cx, 0);
890 	cx->time += sleep_ticks;
891 	return idle_time;
892 }
893 
894 static int c3_cpu_count;
895 static DEFINE_SPINLOCK(c3_lock);
896 
897 /**
898  * acpi_idle_enter_bm - enters C3 with proper BM handling
899  * @dev: the target CPU
900  * @state: the state data
901  *
902  * If BM is detected, the deepest non-C3 idle state is entered instead.
903  */
904 static int acpi_idle_enter_bm(struct cpuidle_device *dev,
905 			      struct cpuidle_state *state)
906 {
907 	struct acpi_processor *pr;
908 	struct acpi_processor_cx *cx = cpuidle_get_statedata(state);
909 	ktime_t  kt1, kt2;
910 	s64 idle_time;
911 	s64 sleep_ticks = 0;
912 
913 
914 	pr = __get_cpu_var(processors);
915 
916 	if (unlikely(!pr))
917 		return 0;
918 
919 	if (acpi_idle_suspend)
920 		return(acpi_idle_enter_c1(dev, state));
921 
922 	if (acpi_idle_bm_check()) {
923 		if (dev->safe_state) {
924 			dev->last_state = dev->safe_state;
925 			return dev->safe_state->enter(dev, dev->safe_state);
926 		} else {
927 			local_irq_disable();
928 			acpi_safe_halt();
929 			local_irq_enable();
930 			return 0;
931 		}
932 	}
933 
934 	local_irq_disable();
935 	current_thread_info()->status &= ~TS_POLLING;
936 	/*
937 	 * TS_POLLING-cleared state must be visible before we test
938 	 * NEED_RESCHED:
939 	 */
940 	smp_mb();
941 
942 	if (unlikely(need_resched())) {
943 		current_thread_info()->status |= TS_POLLING;
944 		local_irq_enable();
945 		return 0;
946 	}
947 
948 	acpi_unlazy_tlb(smp_processor_id());
949 
950 	/* Tell the scheduler that we are going deep-idle: */
951 	sched_clock_idle_sleep_event();
952 	/*
953 	 * Must be done before busmaster disable as we might need to
954 	 * access HPET !
955 	 */
956 	acpi_state_timer_broadcast(pr, cx, 1);
957 
958 	/*
959 	 * disable bus master
960 	 * bm_check implies we need ARB_DIS
961 	 * !bm_check implies we need cache flush
962 	 * bm_control implies whether we can do ARB_DIS
963 	 *
964 	 * That leaves a case where bm_check is set and bm_control is
965 	 * not set. In that case we cannot do much, we enter C3
966 	 * without doing anything.
967 	 */
968 	if (pr->flags.bm_check && pr->flags.bm_control) {
969 		spin_lock(&c3_lock);
970 		c3_cpu_count++;
971 		/* Disable bus master arbitration when all CPUs are in C3 */
972 		if (c3_cpu_count == num_online_cpus())
973 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
974 		spin_unlock(&c3_lock);
975 	} else if (!pr->flags.bm_check) {
976 		ACPI_FLUSH_CPU_CACHE();
977 	}
978 
979 	kt1 = ktime_get_real();
980 	acpi_idle_do_entry(cx);
981 	kt2 = ktime_get_real();
982 	idle_time =  ktime_to_us(ktime_sub(kt2, kt1));
983 
984 	/* Re-enable bus master arbitration */
985 	if (pr->flags.bm_check && pr->flags.bm_control) {
986 		spin_lock(&c3_lock);
987 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
988 		c3_cpu_count--;
989 		spin_unlock(&c3_lock);
990 	}
991 
992 #if defined (CONFIG_GENERIC_TIME) && defined (CONFIG_X86)
993 	/* TSC could halt in idle, so notify users */
994 	if (tsc_halts_in_c(ACPI_STATE_C3))
995 		mark_tsc_unstable("TSC halts in idle");
996 #endif
997 	sleep_ticks = us_to_pm_timer_ticks(idle_time);
998 	/* Tell the scheduler how much we idled: */
999 	sched_clock_idle_wakeup_event(sleep_ticks*PM_TIMER_TICK_NS);
1000 
1001 	local_irq_enable();
1002 	current_thread_info()->status |= TS_POLLING;
1003 
1004 	cx->usage++;
1005 
1006 	acpi_state_timer_broadcast(pr, cx, 0);
1007 	cx->time += sleep_ticks;
1008 	return idle_time;
1009 }
1010 
1011 struct cpuidle_driver acpi_idle_driver = {
1012 	.name =		"acpi_idle",
1013 	.owner =	THIS_MODULE,
1014 };
1015 
1016 /**
1017  * acpi_processor_setup_cpuidle - prepares and configures CPUIDLE
1018  * @pr: the ACPI processor
1019  */
1020 static int acpi_processor_setup_cpuidle(struct acpi_processor *pr)
1021 {
1022 	int i, count = CPUIDLE_DRIVER_STATE_START;
1023 	struct acpi_processor_cx *cx;
1024 	struct cpuidle_state *state;
1025 	struct cpuidle_device *dev = &pr->power.dev;
1026 
1027 	if (!pr->flags.power_setup_done)
1028 		return -EINVAL;
1029 
1030 	if (pr->flags.power == 0) {
1031 		return -EINVAL;
1032 	}
1033 
1034 	dev->cpu = pr->id;
1035 	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
1036 		dev->states[i].name[0] = '\0';
1037 		dev->states[i].desc[0] = '\0';
1038 	}
1039 
1040 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
1041 		cx = &pr->power.states[i];
1042 		state = &dev->states[count];
1043 
1044 		if (!cx->valid)
1045 			continue;
1046 
1047 #ifdef CONFIG_HOTPLUG_CPU
1048 		if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
1049 		    !pr->flags.has_cst &&
1050 		    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
1051 			continue;
1052 #endif
1053 		cpuidle_set_statedata(state, cx);
1054 
1055 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
1056 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1057 		state->exit_latency = cx->latency;
1058 		state->target_residency = cx->latency * latency_factor;
1059 		state->power_usage = cx->power;
1060 
1061 		state->flags = 0;
1062 		switch (cx->type) {
1063 			case ACPI_STATE_C1:
1064 			state->flags |= CPUIDLE_FLAG_SHALLOW;
1065 			if (cx->entry_method == ACPI_CSTATE_FFH)
1066 				state->flags |= CPUIDLE_FLAG_TIME_VALID;
1067 
1068 			state->enter = acpi_idle_enter_c1;
1069 			dev->safe_state = state;
1070 			break;
1071 
1072 			case ACPI_STATE_C2:
1073 			state->flags |= CPUIDLE_FLAG_BALANCED;
1074 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1075 			state->enter = acpi_idle_enter_simple;
1076 			dev->safe_state = state;
1077 			break;
1078 
1079 			case ACPI_STATE_C3:
1080 			state->flags |= CPUIDLE_FLAG_DEEP;
1081 			state->flags |= CPUIDLE_FLAG_TIME_VALID;
1082 			state->flags |= CPUIDLE_FLAG_CHECK_BM;
1083 			state->enter = pr->flags.bm_check ?
1084 					acpi_idle_enter_bm :
1085 					acpi_idle_enter_simple;
1086 			break;
1087 		}
1088 
1089 		count++;
1090 		if (count == CPUIDLE_STATE_MAX)
1091 			break;
1092 	}
1093 
1094 	dev->state_count = count;
1095 
1096 	if (!count)
1097 		return -EINVAL;
1098 
1099 	return 0;
1100 }
1101 
1102 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1103 {
1104 	int ret = 0;
1105 
1106 	if (boot_option_idle_override)
1107 		return 0;
1108 
1109 	if (!pr)
1110 		return -EINVAL;
1111 
1112 	if (nocst) {
1113 		return -ENODEV;
1114 	}
1115 
1116 	if (!pr->flags.power_setup_done)
1117 		return -ENODEV;
1118 
1119 	cpuidle_pause_and_lock();
1120 	cpuidle_disable_device(&pr->power.dev);
1121 	acpi_processor_get_power_info(pr);
1122 	if (pr->flags.power) {
1123 		acpi_processor_setup_cpuidle(pr);
1124 		ret = cpuidle_enable_device(&pr->power.dev);
1125 	}
1126 	cpuidle_resume_and_unlock();
1127 
1128 	return ret;
1129 }
1130 
1131 int __cpuinit acpi_processor_power_init(struct acpi_processor *pr,
1132 			      struct acpi_device *device)
1133 {
1134 	acpi_status status = 0;
1135 	static int first_run;
1136 	struct proc_dir_entry *entry = NULL;
1137 	unsigned int i;
1138 
1139 	if (boot_option_idle_override)
1140 		return 0;
1141 
1142 	if (!first_run) {
1143 		if (idle_halt) {
1144 			/*
1145 			 * When the boot option of "idle=halt" is added, halt
1146 			 * is used for CPU IDLE.
1147 			 * In such case C2/C3 is meaningless. So the max_cstate
1148 			 * is set to one.
1149 			 */
1150 			max_cstate = 1;
1151 		}
1152 		dmi_check_system(processor_power_dmi_table);
1153 		max_cstate = acpi_processor_cstate_check(max_cstate);
1154 		if (max_cstate < ACPI_C_STATES_MAX)
1155 			printk(KERN_NOTICE
1156 			       "ACPI: processor limited to max C-state %d\n",
1157 			       max_cstate);
1158 		first_run++;
1159 	}
1160 
1161 	if (!pr)
1162 		return -EINVAL;
1163 
1164 	if (acpi_gbl_FADT.cst_control && !nocst) {
1165 		status =
1166 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1167 		if (ACPI_FAILURE(status)) {
1168 			ACPI_EXCEPTION((AE_INFO, status,
1169 					"Notifying BIOS of _CST ability failed"));
1170 		}
1171 	}
1172 
1173 	acpi_processor_get_power_info(pr);
1174 	pr->flags.power_setup_done = 1;
1175 
1176 	/*
1177 	 * Install the idle handler if processor power management is supported.
1178 	 * Note that we use previously set idle handler will be used on
1179 	 * platforms that only support C1.
1180 	 */
1181 	if (pr->flags.power) {
1182 		acpi_processor_setup_cpuidle(pr);
1183 		if (cpuidle_register_device(&pr->power.dev))
1184 			return -EIO;
1185 
1186 		printk(KERN_INFO PREFIX "CPU%d (power states:", pr->id);
1187 		for (i = 1; i <= pr->power.count; i++)
1188 			if (pr->power.states[i].valid)
1189 				printk(" C%d[C%d]", i,
1190 				       pr->power.states[i].type);
1191 		printk(")\n");
1192 	}
1193 
1194 	/* 'power' [R] */
1195 	entry = proc_create_data(ACPI_PROCESSOR_FILE_POWER,
1196 				 S_IRUGO, acpi_device_dir(device),
1197 				 &acpi_processor_power_fops,
1198 				 acpi_driver_data(device));
1199 	if (!entry)
1200 		return -EIO;
1201 	return 0;
1202 }
1203 
1204 int acpi_processor_power_exit(struct acpi_processor *pr,
1205 			      struct acpi_device *device)
1206 {
1207 	if (boot_option_idle_override)
1208 		return 0;
1209 
1210 	cpuidle_unregister_device(&pr->power.dev);
1211 	pr->flags.power_setup_done = 0;
1212 
1213 	if (acpi_device_dir(device))
1214 		remove_proc_entry(ACPI_PROCESSOR_FILE_POWER,
1215 				  acpi_device_dir(device));
1216 
1217 	return 0;
1218 }
1219