xref: /openbmc/linux/drivers/acpi/processor_idle.c (revision a06c488d)
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  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
25  */
26 
27 #include <linux/module.h>
28 #include <linux/acpi.h>
29 #include <linux/dmi.h>
30 #include <linux/sched.h>       /* need_resched() */
31 #include <linux/tick.h>
32 #include <linux/cpuidle.h>
33 #include <linux/syscore_ops.h>
34 #include <acpi/processor.h>
35 
36 /*
37  * Include the apic definitions for x86 to have the APIC timer related defines
38  * available also for UP (on SMP it gets magically included via linux/smp.h).
39  * asm/acpi.h is not an option, as it would require more include magic. Also
40  * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
41  */
42 #ifdef CONFIG_X86
43 #include <asm/apic.h>
44 #endif
45 
46 #define PREFIX "ACPI: "
47 
48 #define ACPI_PROCESSOR_CLASS            "processor"
49 #define _COMPONENT              ACPI_PROCESSOR_COMPONENT
50 ACPI_MODULE_NAME("processor_idle");
51 
52 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
53 module_param(max_cstate, uint, 0000);
54 static unsigned int nocst __read_mostly;
55 module_param(nocst, uint, 0000);
56 static int bm_check_disable __read_mostly;
57 module_param(bm_check_disable, uint, 0000);
58 
59 static unsigned int latency_factor __read_mostly = 2;
60 module_param(latency_factor, uint, 0644);
61 
62 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
63 
64 static DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX],
65 								acpi_cstate);
66 
67 static int disabled_by_idle_boot_param(void)
68 {
69 	return boot_option_idle_override == IDLE_POLL ||
70 		boot_option_idle_override == IDLE_HALT;
71 }
72 
73 /*
74  * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
75  * For now disable this. Probably a bug somewhere else.
76  *
77  * To skip this limit, boot/load with a large max_cstate limit.
78  */
79 static int set_max_cstate(const struct dmi_system_id *id)
80 {
81 	if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
82 		return 0;
83 
84 	printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
85 	       " Override with \"processor.max_cstate=%d\"\n", id->ident,
86 	       (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
87 
88 	max_cstate = (long)id->driver_data;
89 
90 	return 0;
91 }
92 
93 static const struct dmi_system_id processor_power_dmi_table[] = {
94 	{ set_max_cstate, "Clevo 5600D", {
95 	  DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
96 	  DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
97 	 (void *)2},
98 	{ set_max_cstate, "Pavilion zv5000", {
99 	  DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
100 	  DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
101 	 (void *)1},
102 	{ set_max_cstate, "Asus L8400B", {
103 	  DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
104 	  DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
105 	 (void *)1},
106 	{},
107 };
108 
109 
110 /*
111  * Callers should disable interrupts before the call and enable
112  * interrupts after return.
113  */
114 static void acpi_safe_halt(void)
115 {
116 	if (!tif_need_resched()) {
117 		safe_halt();
118 		local_irq_disable();
119 	}
120 }
121 
122 #ifdef ARCH_APICTIMER_STOPS_ON_C3
123 
124 /*
125  * Some BIOS implementations switch to C3 in the published C2 state.
126  * This seems to be a common problem on AMD boxen, but other vendors
127  * are affected too. We pick the most conservative approach: we assume
128  * that the local APIC stops in both C2 and C3.
129  */
130 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
131 				   struct acpi_processor_cx *cx)
132 {
133 	struct acpi_processor_power *pwr = &pr->power;
134 	u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
135 
136 	if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
137 		return;
138 
139 	if (amd_e400_c1e_detected)
140 		type = ACPI_STATE_C1;
141 
142 	/*
143 	 * Check, if one of the previous states already marked the lapic
144 	 * unstable
145 	 */
146 	if (pwr->timer_broadcast_on_state < state)
147 		return;
148 
149 	if (cx->type >= type)
150 		pr->power.timer_broadcast_on_state = state;
151 }
152 
153 static void __lapic_timer_propagate_broadcast(void *arg)
154 {
155 	struct acpi_processor *pr = (struct acpi_processor *) arg;
156 
157 	if (pr->power.timer_broadcast_on_state < INT_MAX)
158 		tick_broadcast_enable();
159 	else
160 		tick_broadcast_disable();
161 }
162 
163 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
164 {
165 	smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
166 				 (void *)pr, 1);
167 }
168 
169 /* Power(C) State timer broadcast control */
170 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
171 				       struct acpi_processor_cx *cx,
172 				       int broadcast)
173 {
174 	int state = cx - pr->power.states;
175 
176 	if (state >= pr->power.timer_broadcast_on_state) {
177 		if (broadcast)
178 			tick_broadcast_enter();
179 		else
180 			tick_broadcast_exit();
181 	}
182 }
183 
184 #else
185 
186 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
187 				   struct acpi_processor_cx *cstate) { }
188 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
189 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
190 				       struct acpi_processor_cx *cx,
191 				       int broadcast)
192 {
193 }
194 
195 #endif
196 
197 #ifdef CONFIG_PM_SLEEP
198 static u32 saved_bm_rld;
199 
200 static int acpi_processor_suspend(void)
201 {
202 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
203 	return 0;
204 }
205 
206 static void acpi_processor_resume(void)
207 {
208 	u32 resumed_bm_rld = 0;
209 
210 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
211 	if (resumed_bm_rld == saved_bm_rld)
212 		return;
213 
214 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
215 }
216 
217 static struct syscore_ops acpi_processor_syscore_ops = {
218 	.suspend = acpi_processor_suspend,
219 	.resume = acpi_processor_resume,
220 };
221 
222 void acpi_processor_syscore_init(void)
223 {
224 	register_syscore_ops(&acpi_processor_syscore_ops);
225 }
226 
227 void acpi_processor_syscore_exit(void)
228 {
229 	unregister_syscore_ops(&acpi_processor_syscore_ops);
230 }
231 #endif /* CONFIG_PM_SLEEP */
232 
233 #if defined(CONFIG_X86)
234 static void tsc_check_state(int state)
235 {
236 	switch (boot_cpu_data.x86_vendor) {
237 	case X86_VENDOR_AMD:
238 	case X86_VENDOR_INTEL:
239 		/*
240 		 * AMD Fam10h TSC will tick in all
241 		 * C/P/S0/S1 states when this bit is set.
242 		 */
243 		if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
244 			return;
245 
246 		/*FALL THROUGH*/
247 	default:
248 		/* TSC could halt in idle, so notify users */
249 		if (state > ACPI_STATE_C1)
250 			mark_tsc_unstable("TSC halts in idle");
251 	}
252 }
253 #else
254 static void tsc_check_state(int state) { return; }
255 #endif
256 
257 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
258 {
259 
260 	if (!pr->pblk)
261 		return -ENODEV;
262 
263 	/* if info is obtained from pblk/fadt, type equals state */
264 	pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
265 	pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
266 
267 #ifndef CONFIG_HOTPLUG_CPU
268 	/*
269 	 * Check for P_LVL2_UP flag before entering C2 and above on
270 	 * an SMP system.
271 	 */
272 	if ((num_online_cpus() > 1) &&
273 	    !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
274 		return -ENODEV;
275 #endif
276 
277 	/* determine C2 and C3 address from pblk */
278 	pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
279 	pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
280 
281 	/* determine latencies from FADT */
282 	pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
283 	pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
284 
285 	/*
286 	 * FADT specified C2 latency must be less than or equal to
287 	 * 100 microseconds.
288 	 */
289 	if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
290 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
291 			"C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
292 		/* invalidate C2 */
293 		pr->power.states[ACPI_STATE_C2].address = 0;
294 	}
295 
296 	/*
297 	 * FADT supplied C3 latency must be less than or equal to
298 	 * 1000 microseconds.
299 	 */
300 	if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
301 		ACPI_DEBUG_PRINT((ACPI_DB_INFO,
302 			"C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
303 		/* invalidate C3 */
304 		pr->power.states[ACPI_STATE_C3].address = 0;
305 	}
306 
307 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
308 			  "lvl2[0x%08x] lvl3[0x%08x]\n",
309 			  pr->power.states[ACPI_STATE_C2].address,
310 			  pr->power.states[ACPI_STATE_C3].address));
311 
312 	return 0;
313 }
314 
315 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
316 {
317 	if (!pr->power.states[ACPI_STATE_C1].valid) {
318 		/* set the first C-State to C1 */
319 		/* all processors need to support C1 */
320 		pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
321 		pr->power.states[ACPI_STATE_C1].valid = 1;
322 		pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
323 	}
324 	/* the C0 state only exists as a filler in our array */
325 	pr->power.states[ACPI_STATE_C0].valid = 1;
326 	return 0;
327 }
328 
329 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
330 {
331 	acpi_status status;
332 	u64 count;
333 	int current_count;
334 	int i, ret = 0;
335 	struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
336 	union acpi_object *cst;
337 
338 
339 	if (nocst)
340 		return -ENODEV;
341 
342 	current_count = 0;
343 
344 	status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
345 	if (ACPI_FAILURE(status)) {
346 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
347 		return -ENODEV;
348 	}
349 
350 	cst = buffer.pointer;
351 
352 	/* There must be at least 2 elements */
353 	if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
354 		printk(KERN_ERR PREFIX "not enough elements in _CST\n");
355 		ret = -EFAULT;
356 		goto end;
357 	}
358 
359 	count = cst->package.elements[0].integer.value;
360 
361 	/* Validate number of power states. */
362 	if (count < 1 || count != cst->package.count - 1) {
363 		printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
364 		ret = -EFAULT;
365 		goto end;
366 	}
367 
368 	/* Tell driver that at least _CST is supported. */
369 	pr->flags.has_cst = 1;
370 
371 	for (i = 1; i <= count; i++) {
372 		union acpi_object *element;
373 		union acpi_object *obj;
374 		struct acpi_power_register *reg;
375 		struct acpi_processor_cx cx;
376 
377 		memset(&cx, 0, sizeof(cx));
378 
379 		element = &(cst->package.elements[i]);
380 		if (element->type != ACPI_TYPE_PACKAGE)
381 			continue;
382 
383 		if (element->package.count != 4)
384 			continue;
385 
386 		obj = &(element->package.elements[0]);
387 
388 		if (obj->type != ACPI_TYPE_BUFFER)
389 			continue;
390 
391 		reg = (struct acpi_power_register *)obj->buffer.pointer;
392 
393 		if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
394 		    (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
395 			continue;
396 
397 		/* There should be an easy way to extract an integer... */
398 		obj = &(element->package.elements[1]);
399 		if (obj->type != ACPI_TYPE_INTEGER)
400 			continue;
401 
402 		cx.type = obj->integer.value;
403 		/*
404 		 * Some buggy BIOSes won't list C1 in _CST -
405 		 * Let acpi_processor_get_power_info_default() handle them later
406 		 */
407 		if (i == 1 && cx.type != ACPI_STATE_C1)
408 			current_count++;
409 
410 		cx.address = reg->address;
411 		cx.index = current_count + 1;
412 
413 		cx.entry_method = ACPI_CSTATE_SYSTEMIO;
414 		if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
415 			if (acpi_processor_ffh_cstate_probe
416 					(pr->id, &cx, reg) == 0) {
417 				cx.entry_method = ACPI_CSTATE_FFH;
418 			} else if (cx.type == ACPI_STATE_C1) {
419 				/*
420 				 * C1 is a special case where FIXED_HARDWARE
421 				 * can be handled in non-MWAIT way as well.
422 				 * In that case, save this _CST entry info.
423 				 * Otherwise, ignore this info and continue.
424 				 */
425 				cx.entry_method = ACPI_CSTATE_HALT;
426 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
427 			} else {
428 				continue;
429 			}
430 			if (cx.type == ACPI_STATE_C1 &&
431 			    (boot_option_idle_override == IDLE_NOMWAIT)) {
432 				/*
433 				 * In most cases the C1 space_id obtained from
434 				 * _CST object is FIXED_HARDWARE access mode.
435 				 * But when the option of idle=halt is added,
436 				 * the entry_method type should be changed from
437 				 * CSTATE_FFH to CSTATE_HALT.
438 				 * When the option of idle=nomwait is added,
439 				 * the C1 entry_method type should be
440 				 * CSTATE_HALT.
441 				 */
442 				cx.entry_method = ACPI_CSTATE_HALT;
443 				snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
444 			}
445 		} else {
446 			snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
447 				 cx.address);
448 		}
449 
450 		if (cx.type == ACPI_STATE_C1) {
451 			cx.valid = 1;
452 		}
453 
454 		obj = &(element->package.elements[2]);
455 		if (obj->type != ACPI_TYPE_INTEGER)
456 			continue;
457 
458 		cx.latency = obj->integer.value;
459 
460 		obj = &(element->package.elements[3]);
461 		if (obj->type != ACPI_TYPE_INTEGER)
462 			continue;
463 
464 		current_count++;
465 		memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
466 
467 		/*
468 		 * We support total ACPI_PROCESSOR_MAX_POWER - 1
469 		 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
470 		 */
471 		if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
472 			printk(KERN_WARNING
473 			       "Limiting number of power states to max (%d)\n",
474 			       ACPI_PROCESSOR_MAX_POWER);
475 			printk(KERN_WARNING
476 			       "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
477 			break;
478 		}
479 	}
480 
481 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
482 			  current_count));
483 
484 	/* Validate number of power states discovered */
485 	if (current_count < 2)
486 		ret = -EFAULT;
487 
488       end:
489 	kfree(buffer.pointer);
490 
491 	return ret;
492 }
493 
494 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
495 					   struct acpi_processor_cx *cx)
496 {
497 	static int bm_check_flag = -1;
498 	static int bm_control_flag = -1;
499 
500 
501 	if (!cx->address)
502 		return;
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 == -1) {
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 		bm_control_flag = pr->flags.bm_control;
523 	} else {
524 		pr->flags.bm_check = bm_check_flag;
525 		pr->flags.bm_control = bm_control_flag;
526 	}
527 
528 	if (pr->flags.bm_check) {
529 		if (!pr->flags.bm_control) {
530 			if (pr->flags.has_cst != 1) {
531 				/* bus mastering control is necessary */
532 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
533 					"C3 support requires BM control\n"));
534 				return;
535 			} else {
536 				/* Here we enter C3 without bus mastering */
537 				ACPI_DEBUG_PRINT((ACPI_DB_INFO,
538 					"C3 support without BM control\n"));
539 			}
540 		}
541 	} else {
542 		/*
543 		 * WBINVD should be set in fadt, for C3 state to be
544 		 * supported on when bm_check is not required.
545 		 */
546 		if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
547 			ACPI_DEBUG_PRINT((ACPI_DB_INFO,
548 					  "Cache invalidation should work properly"
549 					  " for C3 to be enabled on SMP systems\n"));
550 			return;
551 		}
552 	}
553 
554 	/*
555 	 * Otherwise we've met all of our C3 requirements.
556 	 * Normalize the C3 latency to expidite policy.  Enable
557 	 * checking of bus mastering status (bm_check) so we can
558 	 * use this in our C3 policy
559 	 */
560 	cx->valid = 1;
561 
562 	/*
563 	 * On older chipsets, BM_RLD needs to be set
564 	 * in order for Bus Master activity to wake the
565 	 * system from C3.  Newer chipsets handle DMA
566 	 * during C3 automatically and BM_RLD is a NOP.
567 	 * In either case, the proper way to
568 	 * handle BM_RLD is to set it and leave it set.
569 	 */
570 	acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
571 
572 	return;
573 }
574 
575 static int acpi_processor_power_verify(struct acpi_processor *pr)
576 {
577 	unsigned int i;
578 	unsigned int working = 0;
579 
580 	pr->power.timer_broadcast_on_state = INT_MAX;
581 
582 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
583 		struct acpi_processor_cx *cx = &pr->power.states[i];
584 
585 		switch (cx->type) {
586 		case ACPI_STATE_C1:
587 			cx->valid = 1;
588 			break;
589 
590 		case ACPI_STATE_C2:
591 			if (!cx->address)
592 				break;
593 			cx->valid = 1;
594 			break;
595 
596 		case ACPI_STATE_C3:
597 			acpi_processor_power_verify_c3(pr, cx);
598 			break;
599 		}
600 		if (!cx->valid)
601 			continue;
602 
603 		lapic_timer_check_state(i, pr, cx);
604 		tsc_check_state(cx->type);
605 		working++;
606 	}
607 
608 	lapic_timer_propagate_broadcast(pr);
609 
610 	return (working);
611 }
612 
613 static int acpi_processor_get_power_info(struct acpi_processor *pr)
614 {
615 	unsigned int i;
616 	int result;
617 
618 
619 	/* NOTE: the idle thread may not be running while calling
620 	 * this function */
621 
622 	/* Zero initialize all the C-states info. */
623 	memset(pr->power.states, 0, sizeof(pr->power.states));
624 
625 	result = acpi_processor_get_power_info_cst(pr);
626 	if (result == -ENODEV)
627 		result = acpi_processor_get_power_info_fadt(pr);
628 
629 	if (result)
630 		return result;
631 
632 	acpi_processor_get_power_info_default(pr);
633 
634 	pr->power.count = acpi_processor_power_verify(pr);
635 
636 	/*
637 	 * if one state of type C2 or C3 is available, mark this
638 	 * CPU as being "idle manageable"
639 	 */
640 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
641 		if (pr->power.states[i].valid) {
642 			pr->power.count = i;
643 			if (pr->power.states[i].type >= ACPI_STATE_C2)
644 				pr->flags.power = 1;
645 		}
646 	}
647 
648 	return 0;
649 }
650 
651 /**
652  * acpi_idle_bm_check - checks if bus master activity was detected
653  */
654 static int acpi_idle_bm_check(void)
655 {
656 	u32 bm_status = 0;
657 
658 	if (bm_check_disable)
659 		return 0;
660 
661 	acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
662 	if (bm_status)
663 		acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
664 	/*
665 	 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
666 	 * the true state of bus mastering activity; forcing us to
667 	 * manually check the BMIDEA bit of each IDE channel.
668 	 */
669 	else if (errata.piix4.bmisx) {
670 		if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
671 		    || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
672 			bm_status = 1;
673 	}
674 	return bm_status;
675 }
676 
677 /**
678  * acpi_idle_do_entry - enter idle state using the appropriate method
679  * @cx: cstate data
680  *
681  * Caller disables interrupt before call and enables interrupt after return.
682  */
683 static void acpi_idle_do_entry(struct acpi_processor_cx *cx)
684 {
685 	if (cx->entry_method == ACPI_CSTATE_FFH) {
686 		/* Call into architectural FFH based C-state */
687 		acpi_processor_ffh_cstate_enter(cx);
688 	} else if (cx->entry_method == ACPI_CSTATE_HALT) {
689 		acpi_safe_halt();
690 	} else {
691 		/* IO port based C-state */
692 		inb(cx->address);
693 		/* Dummy wait op - must do something useless after P_LVL2 read
694 		   because chipsets cannot guarantee that STPCLK# signal
695 		   gets asserted in time to freeze execution properly. */
696 		inl(acpi_gbl_FADT.xpm_timer_block.address);
697 	}
698 }
699 
700 /**
701  * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
702  * @dev: the target CPU
703  * @index: the index of suggested state
704  */
705 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
706 {
707 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
708 
709 	ACPI_FLUSH_CPU_CACHE();
710 
711 	while (1) {
712 
713 		if (cx->entry_method == ACPI_CSTATE_HALT)
714 			safe_halt();
715 		else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
716 			inb(cx->address);
717 			/* See comment in acpi_idle_do_entry() */
718 			inl(acpi_gbl_FADT.xpm_timer_block.address);
719 		} else
720 			return -ENODEV;
721 	}
722 
723 	/* Never reached */
724 	return 0;
725 }
726 
727 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
728 {
729 	return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
730 		!(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
731 }
732 
733 static int c3_cpu_count;
734 static DEFINE_RAW_SPINLOCK(c3_lock);
735 
736 /**
737  * acpi_idle_enter_bm - enters C3 with proper BM handling
738  * @pr: Target processor
739  * @cx: Target state context
740  * @timer_bc: Whether or not to change timer mode to broadcast
741  */
742 static void acpi_idle_enter_bm(struct acpi_processor *pr,
743 			       struct acpi_processor_cx *cx, bool timer_bc)
744 {
745 	acpi_unlazy_tlb(smp_processor_id());
746 
747 	/*
748 	 * Must be done before busmaster disable as we might need to
749 	 * access HPET !
750 	 */
751 	if (timer_bc)
752 		lapic_timer_state_broadcast(pr, cx, 1);
753 
754 	/*
755 	 * disable bus master
756 	 * bm_check implies we need ARB_DIS
757 	 * bm_control implies whether we can do ARB_DIS
758 	 *
759 	 * That leaves a case where bm_check is set and bm_control is
760 	 * not set. In that case we cannot do much, we enter C3
761 	 * without doing anything.
762 	 */
763 	if (pr->flags.bm_control) {
764 		raw_spin_lock(&c3_lock);
765 		c3_cpu_count++;
766 		/* Disable bus master arbitration when all CPUs are in C3 */
767 		if (c3_cpu_count == num_online_cpus())
768 			acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
769 		raw_spin_unlock(&c3_lock);
770 	}
771 
772 	acpi_idle_do_entry(cx);
773 
774 	/* Re-enable bus master arbitration */
775 	if (pr->flags.bm_control) {
776 		raw_spin_lock(&c3_lock);
777 		acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
778 		c3_cpu_count--;
779 		raw_spin_unlock(&c3_lock);
780 	}
781 
782 	if (timer_bc)
783 		lapic_timer_state_broadcast(pr, cx, 0);
784 }
785 
786 static int acpi_idle_enter(struct cpuidle_device *dev,
787 			   struct cpuidle_driver *drv, int index)
788 {
789 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
790 	struct acpi_processor *pr;
791 
792 	pr = __this_cpu_read(processors);
793 	if (unlikely(!pr))
794 		return -EINVAL;
795 
796 	if (cx->type != ACPI_STATE_C1) {
797 		if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
798 			index = CPUIDLE_DRIVER_STATE_START;
799 			cx = per_cpu(acpi_cstate[index], dev->cpu);
800 		} else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) {
801 			if (cx->bm_sts_skip || !acpi_idle_bm_check()) {
802 				acpi_idle_enter_bm(pr, cx, true);
803 				return index;
804 			} else if (drv->safe_state_index >= 0) {
805 				index = drv->safe_state_index;
806 				cx = per_cpu(acpi_cstate[index], dev->cpu);
807 			} else {
808 				acpi_safe_halt();
809 				return -EBUSY;
810 			}
811 		}
812 	}
813 
814 	lapic_timer_state_broadcast(pr, cx, 1);
815 
816 	if (cx->type == ACPI_STATE_C3)
817 		ACPI_FLUSH_CPU_CACHE();
818 
819 	acpi_idle_do_entry(cx);
820 
821 	lapic_timer_state_broadcast(pr, cx, 0);
822 
823 	return index;
824 }
825 
826 static void acpi_idle_enter_freeze(struct cpuidle_device *dev,
827 				   struct cpuidle_driver *drv, int index)
828 {
829 	struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
830 
831 	if (cx->type == ACPI_STATE_C3) {
832 		struct acpi_processor *pr = __this_cpu_read(processors);
833 
834 		if (unlikely(!pr))
835 			return;
836 
837 		if (pr->flags.bm_check) {
838 			acpi_idle_enter_bm(pr, cx, false);
839 			return;
840 		} else {
841 			ACPI_FLUSH_CPU_CACHE();
842 		}
843 	}
844 	acpi_idle_do_entry(cx);
845 }
846 
847 struct cpuidle_driver acpi_idle_driver = {
848 	.name =		"acpi_idle",
849 	.owner =	THIS_MODULE,
850 };
851 
852 /**
853  * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
854  * device i.e. per-cpu data
855  *
856  * @pr: the ACPI processor
857  * @dev : the cpuidle device
858  */
859 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
860 					   struct cpuidle_device *dev)
861 {
862 	int i, count = CPUIDLE_DRIVER_STATE_START;
863 	struct acpi_processor_cx *cx;
864 
865 	if (!pr->flags.power_setup_done)
866 		return -EINVAL;
867 
868 	if (pr->flags.power == 0) {
869 		return -EINVAL;
870 	}
871 
872 	if (!dev)
873 		return -EINVAL;
874 
875 	dev->cpu = pr->id;
876 
877 	if (max_cstate == 0)
878 		max_cstate = 1;
879 
880 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
881 		cx = &pr->power.states[i];
882 
883 		if (!cx->valid)
884 			continue;
885 
886 		per_cpu(acpi_cstate[count], dev->cpu) = cx;
887 
888 		count++;
889 		if (count == CPUIDLE_STATE_MAX)
890 			break;
891 	}
892 
893 	if (!count)
894 		return -EINVAL;
895 
896 	return 0;
897 }
898 
899 /**
900  * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
901  * global state data i.e. idle routines
902  *
903  * @pr: the ACPI processor
904  */
905 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
906 {
907 	int i, count = CPUIDLE_DRIVER_STATE_START;
908 	struct acpi_processor_cx *cx;
909 	struct cpuidle_state *state;
910 	struct cpuidle_driver *drv = &acpi_idle_driver;
911 
912 	if (!pr->flags.power_setup_done)
913 		return -EINVAL;
914 
915 	if (pr->flags.power == 0)
916 		return -EINVAL;
917 
918 	drv->safe_state_index = -1;
919 	for (i = CPUIDLE_DRIVER_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
920 		drv->states[i].name[0] = '\0';
921 		drv->states[i].desc[0] = '\0';
922 	}
923 
924 	if (max_cstate == 0)
925 		max_cstate = 1;
926 
927 	for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
928 		cx = &pr->power.states[i];
929 
930 		if (!cx->valid)
931 			continue;
932 
933 		state = &drv->states[count];
934 		snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
935 		strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
936 		state->exit_latency = cx->latency;
937 		state->target_residency = cx->latency * latency_factor;
938 		state->enter = acpi_idle_enter;
939 
940 		state->flags = 0;
941 		if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) {
942 			state->enter_dead = acpi_idle_play_dead;
943 			drv->safe_state_index = count;
944 		}
945 		/*
946 		 * Halt-induced C1 is not good for ->enter_freeze, because it
947 		 * re-enables interrupts on exit.  Moreover, C1 is generally not
948 		 * particularly interesting from the suspend-to-idle angle, so
949 		 * avoid C1 and the situations in which we may need to fall back
950 		 * to it altogether.
951 		 */
952 		if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
953 			state->enter_freeze = acpi_idle_enter_freeze;
954 
955 		count++;
956 		if (count == CPUIDLE_STATE_MAX)
957 			break;
958 	}
959 
960 	drv->state_count = count;
961 
962 	if (!count)
963 		return -EINVAL;
964 
965 	return 0;
966 }
967 
968 int acpi_processor_hotplug(struct acpi_processor *pr)
969 {
970 	int ret = 0;
971 	struct cpuidle_device *dev;
972 
973 	if (disabled_by_idle_boot_param())
974 		return 0;
975 
976 	if (nocst)
977 		return -ENODEV;
978 
979 	if (!pr->flags.power_setup_done)
980 		return -ENODEV;
981 
982 	dev = per_cpu(acpi_cpuidle_device, pr->id);
983 	cpuidle_pause_and_lock();
984 	cpuidle_disable_device(dev);
985 	acpi_processor_get_power_info(pr);
986 	if (pr->flags.power) {
987 		acpi_processor_setup_cpuidle_cx(pr, dev);
988 		ret = cpuidle_enable_device(dev);
989 	}
990 	cpuidle_resume_and_unlock();
991 
992 	return ret;
993 }
994 
995 int acpi_processor_cst_has_changed(struct acpi_processor *pr)
996 {
997 	int cpu;
998 	struct acpi_processor *_pr;
999 	struct cpuidle_device *dev;
1000 
1001 	if (disabled_by_idle_boot_param())
1002 		return 0;
1003 
1004 	if (nocst)
1005 		return -ENODEV;
1006 
1007 	if (!pr->flags.power_setup_done)
1008 		return -ENODEV;
1009 
1010 	/*
1011 	 * FIXME:  Design the ACPI notification to make it once per
1012 	 * system instead of once per-cpu.  This condition is a hack
1013 	 * to make the code that updates C-States be called once.
1014 	 */
1015 
1016 	if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1017 
1018 		/* Protect against cpu-hotplug */
1019 		get_online_cpus();
1020 		cpuidle_pause_and_lock();
1021 
1022 		/* Disable all cpuidle devices */
1023 		for_each_online_cpu(cpu) {
1024 			_pr = per_cpu(processors, cpu);
1025 			if (!_pr || !_pr->flags.power_setup_done)
1026 				continue;
1027 			dev = per_cpu(acpi_cpuidle_device, cpu);
1028 			cpuidle_disable_device(dev);
1029 		}
1030 
1031 		/* Populate Updated C-state information */
1032 		acpi_processor_get_power_info(pr);
1033 		acpi_processor_setup_cpuidle_states(pr);
1034 
1035 		/* Enable all cpuidle devices */
1036 		for_each_online_cpu(cpu) {
1037 			_pr = per_cpu(processors, cpu);
1038 			if (!_pr || !_pr->flags.power_setup_done)
1039 				continue;
1040 			acpi_processor_get_power_info(_pr);
1041 			if (_pr->flags.power) {
1042 				dev = per_cpu(acpi_cpuidle_device, cpu);
1043 				acpi_processor_setup_cpuidle_cx(_pr, dev);
1044 				cpuidle_enable_device(dev);
1045 			}
1046 		}
1047 		cpuidle_resume_and_unlock();
1048 		put_online_cpus();
1049 	}
1050 
1051 	return 0;
1052 }
1053 
1054 static int acpi_processor_registered;
1055 
1056 int acpi_processor_power_init(struct acpi_processor *pr)
1057 {
1058 	acpi_status status;
1059 	int retval;
1060 	struct cpuidle_device *dev;
1061 	static int first_run;
1062 
1063 	if (disabled_by_idle_boot_param())
1064 		return 0;
1065 
1066 	if (!first_run) {
1067 		dmi_check_system(processor_power_dmi_table);
1068 		max_cstate = acpi_processor_cstate_check(max_cstate);
1069 		if (max_cstate < ACPI_C_STATES_MAX)
1070 			printk(KERN_NOTICE
1071 			       "ACPI: processor limited to max C-state %d\n",
1072 			       max_cstate);
1073 		first_run++;
1074 	}
1075 
1076 	if (acpi_gbl_FADT.cst_control && !nocst) {
1077 		status =
1078 		    acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1079 		if (ACPI_FAILURE(status)) {
1080 			ACPI_EXCEPTION((AE_INFO, status,
1081 					"Notifying BIOS of _CST ability failed"));
1082 		}
1083 	}
1084 
1085 	acpi_processor_get_power_info(pr);
1086 	pr->flags.power_setup_done = 1;
1087 
1088 	/*
1089 	 * Install the idle handler if processor power management is supported.
1090 	 * Note that we use previously set idle handler will be used on
1091 	 * platforms that only support C1.
1092 	 */
1093 	if (pr->flags.power) {
1094 		/* Register acpi_idle_driver if not already registered */
1095 		if (!acpi_processor_registered) {
1096 			acpi_processor_setup_cpuidle_states(pr);
1097 			retval = cpuidle_register_driver(&acpi_idle_driver);
1098 			if (retval)
1099 				return retval;
1100 			printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1101 					acpi_idle_driver.name);
1102 		}
1103 
1104 		dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1105 		if (!dev)
1106 			return -ENOMEM;
1107 		per_cpu(acpi_cpuidle_device, pr->id) = dev;
1108 
1109 		acpi_processor_setup_cpuidle_cx(pr, dev);
1110 
1111 		/* Register per-cpu cpuidle_device. Cpuidle driver
1112 		 * must already be registered before registering device
1113 		 */
1114 		retval = cpuidle_register_device(dev);
1115 		if (retval) {
1116 			if (acpi_processor_registered == 0)
1117 				cpuidle_unregister_driver(&acpi_idle_driver);
1118 			return retval;
1119 		}
1120 		acpi_processor_registered++;
1121 	}
1122 	return 0;
1123 }
1124 
1125 int acpi_processor_power_exit(struct acpi_processor *pr)
1126 {
1127 	struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1128 
1129 	if (disabled_by_idle_boot_param())
1130 		return 0;
1131 
1132 	if (pr->flags.power) {
1133 		cpuidle_unregister_device(dev);
1134 		acpi_processor_registered--;
1135 		if (acpi_processor_registered == 0)
1136 			cpuidle_unregister_driver(&acpi_idle_driver);
1137 	}
1138 
1139 	pr->flags.power_setup_done = 0;
1140 	return 0;
1141 }
1142