Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * processor_idle - idle state submodule to the ACPI processor driver
4 *
5 * Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
6 * Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
7 * Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
8 * Copyright (C) 2004 Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
9 * - Added processor hotplug support
10 * Copyright (C) 2005 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
11 * - Added support for C3 on SMP
12 */
13#define pr_fmt(fmt) "ACPI: " fmt
14
15#include <linux/module.h>
16#include <linux/acpi.h>
17#include <linux/dmi.h>
18#include <linux/sched.h> /* need_resched() */
19#include <linux/tick.h>
20#include <linux/cpuidle.h>
21#include <linux/cpu.h>
22#include <acpi/processor.h>
23
24/*
25 * Include the apic definitions for x86 to have the APIC timer related defines
26 * available also for UP (on SMP it gets magically included via linux/smp.h).
27 * asm/acpi.h is not an option, as it would require more include magic. Also
28 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
29 */
30#ifdef CONFIG_X86
31#include <asm/apic.h>
32#endif
33
34#define ACPI_PROCESSOR_CLASS "processor"
35#define _COMPONENT ACPI_PROCESSOR_COMPONENT
36ACPI_MODULE_NAME("processor_idle");
37
38#define ACPI_IDLE_STATE_START (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
39
40static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
41module_param(max_cstate, uint, 0000);
42static unsigned int nocst __read_mostly;
43module_param(nocst, uint, 0000);
44static int bm_check_disable __read_mostly;
45module_param(bm_check_disable, uint, 0000);
46
47static unsigned int latency_factor __read_mostly = 2;
48module_param(latency_factor, uint, 0644);
49
50static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
51
52struct cpuidle_driver acpi_idle_driver = {
53 .name = "acpi_idle",
54 .owner = THIS_MODULE,
55};
56
57#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
58static
59DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
60
61static int disabled_by_idle_boot_param(void)
62{
63 return boot_option_idle_override == IDLE_POLL ||
64 boot_option_idle_override == IDLE_HALT;
65}
66
67/*
68 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
69 * For now disable this. Probably a bug somewhere else.
70 *
71 * To skip this limit, boot/load with a large max_cstate limit.
72 */
73static int set_max_cstate(const struct dmi_system_id *id)
74{
75 if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
76 return 0;
77
78 pr_notice("%s detected - limiting to C%ld max_cstate."
79 " Override with \"processor.max_cstate=%d\"\n", id->ident,
80 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
81
82 max_cstate = (long)id->driver_data;
83
84 return 0;
85}
86
87static const struct dmi_system_id processor_power_dmi_table[] = {
88 { set_max_cstate, "Clevo 5600D", {
89 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
90 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
91 (void *)2},
92 { set_max_cstate, "Pavilion zv5000", {
93 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
94 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
95 (void *)1},
96 { set_max_cstate, "Asus L8400B", {
97 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
98 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
99 (void *)1},
100 {},
101};
102
103
104/*
105 * Callers should disable interrupts before the call and enable
106 * interrupts after return.
107 */
108static void __cpuidle acpi_safe_halt(void)
109{
110 if (!tif_need_resched()) {
111 safe_halt();
112 local_irq_disable();
113 }
114}
115
116#ifdef ARCH_APICTIMER_STOPS_ON_C3
117
118/*
119 * Some BIOS implementations switch to C3 in the published C2 state.
120 * This seems to be a common problem on AMD boxen, but other vendors
121 * are affected too. We pick the most conservative approach: we assume
122 * that the local APIC stops in both C2 and C3.
123 */
124static void lapic_timer_check_state(int state, struct acpi_processor *pr,
125 struct acpi_processor_cx *cx)
126{
127 struct acpi_processor_power *pwr = &pr->power;
128 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
129
130 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
131 return;
132
133 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
134 type = ACPI_STATE_C1;
135
136 /*
137 * Check, if one of the previous states already marked the lapic
138 * unstable
139 */
140 if (pwr->timer_broadcast_on_state < state)
141 return;
142
143 if (cx->type >= type)
144 pr->power.timer_broadcast_on_state = state;
145}
146
147static void __lapic_timer_propagate_broadcast(void *arg)
148{
149 struct acpi_processor *pr = (struct acpi_processor *) arg;
150
151 if (pr->power.timer_broadcast_on_state < INT_MAX)
152 tick_broadcast_enable();
153 else
154 tick_broadcast_disable();
155}
156
157static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
158{
159 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
160 (void *)pr, 1);
161}
162
163/* Power(C) State timer broadcast control */
164static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
165 struct acpi_processor_cx *cx)
166{
167 return cx - pr->power.states >= pr->power.timer_broadcast_on_state;
168}
169
170#else
171
172static void lapic_timer_check_state(int state, struct acpi_processor *pr,
173 struct acpi_processor_cx *cstate) { }
174static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
175
176static bool lapic_timer_needs_broadcast(struct acpi_processor *pr,
177 struct acpi_processor_cx *cx)
178{
179 return false;
180}
181
182#endif
183
184#if defined(CONFIG_X86)
185static void tsc_check_state(int state)
186{
187 switch (boot_cpu_data.x86_vendor) {
188 case X86_VENDOR_HYGON:
189 case X86_VENDOR_AMD:
190 case X86_VENDOR_INTEL:
191 case X86_VENDOR_CENTAUR:
192 case X86_VENDOR_ZHAOXIN:
193 /*
194 * AMD Fam10h TSC will tick in all
195 * C/P/S0/S1 states when this bit is set.
196 */
197 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
198 return;
199 fallthrough;
200 default:
201 /* TSC could halt in idle, so notify users */
202 if (state > ACPI_STATE_C1)
203 mark_tsc_unstable("TSC halts in idle");
204 }
205}
206#else
207static void tsc_check_state(int state) { return; }
208#endif
209
210static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
211{
212
213 if (!pr->pblk)
214 return -ENODEV;
215
216 /* if info is obtained from pblk/fadt, type equals state */
217 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
218 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
219
220#ifndef CONFIG_HOTPLUG_CPU
221 /*
222 * Check for P_LVL2_UP flag before entering C2 and above on
223 * an SMP system.
224 */
225 if ((num_online_cpus() > 1) &&
226 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
227 return -ENODEV;
228#endif
229
230 /* determine C2 and C3 address from pblk */
231 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
232 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
233
234 /* determine latencies from FADT */
235 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
236 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
237
238 /*
239 * FADT specified C2 latency must be less than or equal to
240 * 100 microseconds.
241 */
242 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
243 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
244 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
245 /* invalidate C2 */
246 pr->power.states[ACPI_STATE_C2].address = 0;
247 }
248
249 /*
250 * FADT supplied C3 latency must be less than or equal to
251 * 1000 microseconds.
252 */
253 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
254 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
255 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
256 /* invalidate C3 */
257 pr->power.states[ACPI_STATE_C3].address = 0;
258 }
259
260 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
261 "lvl2[0x%08x] lvl3[0x%08x]\n",
262 pr->power.states[ACPI_STATE_C2].address,
263 pr->power.states[ACPI_STATE_C3].address));
264
265 snprintf(pr->power.states[ACPI_STATE_C2].desc,
266 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
267 pr->power.states[ACPI_STATE_C2].address);
268 snprintf(pr->power.states[ACPI_STATE_C3].desc,
269 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
270 pr->power.states[ACPI_STATE_C3].address);
271
272 return 0;
273}
274
275static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
276{
277 if (!pr->power.states[ACPI_STATE_C1].valid) {
278 /* set the first C-State to C1 */
279 /* all processors need to support C1 */
280 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
281 pr->power.states[ACPI_STATE_C1].valid = 1;
282 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
283
284 snprintf(pr->power.states[ACPI_STATE_C1].desc,
285 ACPI_CX_DESC_LEN, "ACPI HLT");
286 }
287 /* the C0 state only exists as a filler in our array */
288 pr->power.states[ACPI_STATE_C0].valid = 1;
289 return 0;
290}
291
292static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
293{
294 int ret;
295
296 if (nocst)
297 return -ENODEV;
298
299 ret = acpi_processor_evaluate_cst(pr->handle, pr->id, &pr->power);
300 if (ret)
301 return ret;
302
303 if (!pr->power.count)
304 return -EFAULT;
305
306 pr->flags.has_cst = 1;
307 return 0;
308}
309
310static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
311 struct acpi_processor_cx *cx)
312{
313 static int bm_check_flag = -1;
314 static int bm_control_flag = -1;
315
316
317 if (!cx->address)
318 return;
319
320 /*
321 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
322 * DMA transfers are used by any ISA device to avoid livelock.
323 * Note that we could disable Type-F DMA (as recommended by
324 * the erratum), but this is known to disrupt certain ISA
325 * devices thus we take the conservative approach.
326 */
327 else if (errata.piix4.fdma) {
328 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
329 "C3 not supported on PIIX4 with Type-F DMA\n"));
330 return;
331 }
332
333 /* All the logic here assumes flags.bm_check is same across all CPUs */
334 if (bm_check_flag == -1) {
335 /* Determine whether bm_check is needed based on CPU */
336 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
337 bm_check_flag = pr->flags.bm_check;
338 bm_control_flag = pr->flags.bm_control;
339 } else {
340 pr->flags.bm_check = bm_check_flag;
341 pr->flags.bm_control = bm_control_flag;
342 }
343
344 if (pr->flags.bm_check) {
345 if (!pr->flags.bm_control) {
346 if (pr->flags.has_cst != 1) {
347 /* bus mastering control is necessary */
348 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
349 "C3 support requires BM control\n"));
350 return;
351 } else {
352 /* Here we enter C3 without bus mastering */
353 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
354 "C3 support without BM control\n"));
355 }
356 }
357 } else {
358 /*
359 * WBINVD should be set in fadt, for C3 state to be
360 * supported on when bm_check is not required.
361 */
362 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
363 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
364 "Cache invalidation should work properly"
365 " for C3 to be enabled on SMP systems\n"));
366 return;
367 }
368 }
369
370 /*
371 * Otherwise we've met all of our C3 requirements.
372 * Normalize the C3 latency to expidite policy. Enable
373 * checking of bus mastering status (bm_check) so we can
374 * use this in our C3 policy
375 */
376 cx->valid = 1;
377
378 /*
379 * On older chipsets, BM_RLD needs to be set
380 * in order for Bus Master activity to wake the
381 * system from C3. Newer chipsets handle DMA
382 * during C3 automatically and BM_RLD is a NOP.
383 * In either case, the proper way to
384 * handle BM_RLD is to set it and leave it set.
385 */
386 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
387
388 return;
389}
390
391static int acpi_processor_power_verify(struct acpi_processor *pr)
392{
393 unsigned int i;
394 unsigned int working = 0;
395
396 pr->power.timer_broadcast_on_state = INT_MAX;
397
398 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
399 struct acpi_processor_cx *cx = &pr->power.states[i];
400
401 switch (cx->type) {
402 case ACPI_STATE_C1:
403 cx->valid = 1;
404 break;
405
406 case ACPI_STATE_C2:
407 if (!cx->address)
408 break;
409 cx->valid = 1;
410 break;
411
412 case ACPI_STATE_C3:
413 acpi_processor_power_verify_c3(pr, cx);
414 break;
415 }
416 if (!cx->valid)
417 continue;
418
419 lapic_timer_check_state(i, pr, cx);
420 tsc_check_state(cx->type);
421 working++;
422 }
423
424 lapic_timer_propagate_broadcast(pr);
425
426 return (working);
427}
428
429static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
430{
431 unsigned int i;
432 int result;
433
434
435 /* NOTE: the idle thread may not be running while calling
436 * this function */
437
438 /* Zero initialize all the C-states info. */
439 memset(pr->power.states, 0, sizeof(pr->power.states));
440
441 result = acpi_processor_get_power_info_cst(pr);
442 if (result == -ENODEV)
443 result = acpi_processor_get_power_info_fadt(pr);
444
445 if (result)
446 return result;
447
448 acpi_processor_get_power_info_default(pr);
449
450 pr->power.count = acpi_processor_power_verify(pr);
451
452 /*
453 * if one state of type C2 or C3 is available, mark this
454 * CPU as being "idle manageable"
455 */
456 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
457 if (pr->power.states[i].valid) {
458 pr->power.count = i;
459 pr->flags.power = 1;
460 }
461 }
462
463 return 0;
464}
465
466/**
467 * acpi_idle_bm_check - checks if bus master activity was detected
468 */
469static int acpi_idle_bm_check(void)
470{
471 u32 bm_status = 0;
472
473 if (bm_check_disable)
474 return 0;
475
476 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
477 if (bm_status)
478 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
479 /*
480 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
481 * the true state of bus mastering activity; forcing us to
482 * manually check the BMIDEA bit of each IDE channel.
483 */
484 else if (errata.piix4.bmisx) {
485 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
486 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
487 bm_status = 1;
488 }
489 return bm_status;
490}
491
492static void wait_for_freeze(void)
493{
494#ifdef CONFIG_X86
495 /* No delay is needed if we are in guest */
496 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
497 return;
498#endif
499 /* Dummy wait op - must do something useless after P_LVL2 read
500 because chipsets cannot guarantee that STPCLK# signal
501 gets asserted in time to freeze execution properly. */
502 inl(acpi_gbl_FADT.xpm_timer_block.address);
503}
504
505/**
506 * acpi_idle_do_entry - enter idle state using the appropriate method
507 * @cx: cstate data
508 *
509 * Caller disables interrupt before call and enables interrupt after return.
510 */
511static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
512{
513 if (cx->entry_method == ACPI_CSTATE_FFH) {
514 /* Call into architectural FFH based C-state */
515 acpi_processor_ffh_cstate_enter(cx);
516 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
517 acpi_safe_halt();
518 } else {
519 /* IO port based C-state */
520 inb(cx->address);
521 wait_for_freeze();
522 }
523}
524
525/**
526 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
527 * @dev: the target CPU
528 * @index: the index of suggested state
529 */
530static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
531{
532 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
533
534 ACPI_FLUSH_CPU_CACHE();
535
536 while (1) {
537
538 if (cx->entry_method == ACPI_CSTATE_HALT)
539 safe_halt();
540 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
541 inb(cx->address);
542 wait_for_freeze();
543 } else
544 return -ENODEV;
545 }
546
547 /* Never reached */
548 return 0;
549}
550
551static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
552{
553 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
554 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
555}
556
557static int c3_cpu_count;
558static DEFINE_RAW_SPINLOCK(c3_lock);
559
560/**
561 * acpi_idle_enter_bm - enters C3 with proper BM handling
562 * @drv: cpuidle driver
563 * @pr: Target processor
564 * @cx: Target state context
565 * @index: index of target state
566 */
567static int acpi_idle_enter_bm(struct cpuidle_driver *drv,
568 struct acpi_processor *pr,
569 struct acpi_processor_cx *cx,
570 int index)
571{
572 static struct acpi_processor_cx safe_cx = {
573 .entry_method = ACPI_CSTATE_HALT,
574 };
575
576 /*
577 * disable bus master
578 * bm_check implies we need ARB_DIS
579 * bm_control implies whether we can do ARB_DIS
580 *
581 * That leaves a case where bm_check is set and bm_control is not set.
582 * In that case we cannot do much, we enter C3 without doing anything.
583 */
584 bool dis_bm = pr->flags.bm_control;
585
586 /* If we can skip BM, demote to a safe state. */
587 if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
588 dis_bm = false;
589 index = drv->safe_state_index;
590 if (index >= 0) {
591 cx = this_cpu_read(acpi_cstate[index]);
592 } else {
593 cx = &safe_cx;
594 index = -EBUSY;
595 }
596 }
597
598 if (dis_bm) {
599 raw_spin_lock(&c3_lock);
600 c3_cpu_count++;
601 /* Disable bus master arbitration when all CPUs are in C3 */
602 if (c3_cpu_count == num_online_cpus())
603 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
604 raw_spin_unlock(&c3_lock);
605 }
606
607 rcu_idle_enter();
608
609 acpi_idle_do_entry(cx);
610
611 rcu_idle_exit();
612
613 /* Re-enable bus master arbitration */
614 if (dis_bm) {
615 raw_spin_lock(&c3_lock);
616 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
617 c3_cpu_count--;
618 raw_spin_unlock(&c3_lock);
619 }
620
621 return index;
622}
623
624static int acpi_idle_enter(struct cpuidle_device *dev,
625 struct cpuidle_driver *drv, int index)
626{
627 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
628 struct acpi_processor *pr;
629
630 pr = __this_cpu_read(processors);
631 if (unlikely(!pr))
632 return -EINVAL;
633
634 if (cx->type != ACPI_STATE_C1) {
635 if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check)
636 return acpi_idle_enter_bm(drv, pr, cx, index);
637
638 /* C2 to C1 demotion. */
639 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
640 index = ACPI_IDLE_STATE_START;
641 cx = per_cpu(acpi_cstate[index], dev->cpu);
642 }
643 }
644
645 if (cx->type == ACPI_STATE_C3)
646 ACPI_FLUSH_CPU_CACHE();
647
648 acpi_idle_do_entry(cx);
649
650 return index;
651}
652
653static int acpi_idle_enter_s2idle(struct cpuidle_device *dev,
654 struct cpuidle_driver *drv, int index)
655{
656 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
657
658 if (cx->type == ACPI_STATE_C3) {
659 struct acpi_processor *pr = __this_cpu_read(processors);
660
661 if (unlikely(!pr))
662 return 0;
663
664 if (pr->flags.bm_check) {
665 u8 bm_sts_skip = cx->bm_sts_skip;
666
667 /* Don't check BM_STS, do an unconditional ARB_DIS for S2IDLE */
668 cx->bm_sts_skip = 1;
669 acpi_idle_enter_bm(drv, pr, cx, index);
670 cx->bm_sts_skip = bm_sts_skip;
671
672 return 0;
673 } else {
674 ACPI_FLUSH_CPU_CACHE();
675 }
676 }
677 acpi_idle_do_entry(cx);
678
679 return 0;
680}
681
682static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
683 struct cpuidle_device *dev)
684{
685 int i, count = ACPI_IDLE_STATE_START;
686 struct acpi_processor_cx *cx;
687 struct cpuidle_state *state;
688
689 if (max_cstate == 0)
690 max_cstate = 1;
691
692 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
693 state = &acpi_idle_driver.states[count];
694 cx = &pr->power.states[i];
695
696 if (!cx->valid)
697 continue;
698
699 per_cpu(acpi_cstate[count], dev->cpu) = cx;
700
701 if (lapic_timer_needs_broadcast(pr, cx))
702 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
703
704 if (cx->type == ACPI_STATE_C3) {
705 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
706 if (pr->flags.bm_check)
707 state->flags |= CPUIDLE_FLAG_RCU_IDLE;
708 }
709
710 count++;
711 if (count == CPUIDLE_STATE_MAX)
712 break;
713 }
714
715 if (!count)
716 return -EINVAL;
717
718 return 0;
719}
720
721static int acpi_processor_setup_cstates(struct acpi_processor *pr)
722{
723 int i, count;
724 struct acpi_processor_cx *cx;
725 struct cpuidle_state *state;
726 struct cpuidle_driver *drv = &acpi_idle_driver;
727
728 if (max_cstate == 0)
729 max_cstate = 1;
730
731 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
732 cpuidle_poll_state_init(drv);
733 count = 1;
734 } else {
735 count = 0;
736 }
737
738 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
739 cx = &pr->power.states[i];
740
741 if (!cx->valid)
742 continue;
743
744 state = &drv->states[count];
745 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
746 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
747 state->exit_latency = cx->latency;
748 state->target_residency = cx->latency * latency_factor;
749 state->enter = acpi_idle_enter;
750
751 state->flags = 0;
752 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) {
753 state->enter_dead = acpi_idle_play_dead;
754 drv->safe_state_index = count;
755 }
756 /*
757 * Halt-induced C1 is not good for ->enter_s2idle, because it
758 * re-enables interrupts on exit. Moreover, C1 is generally not
759 * particularly interesting from the suspend-to-idle angle, so
760 * avoid C1 and the situations in which we may need to fall back
761 * to it altogether.
762 */
763 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
764 state->enter_s2idle = acpi_idle_enter_s2idle;
765
766 count++;
767 if (count == CPUIDLE_STATE_MAX)
768 break;
769 }
770
771 drv->state_count = count;
772
773 if (!count)
774 return -EINVAL;
775
776 return 0;
777}
778
779static inline void acpi_processor_cstate_first_run_checks(void)
780{
781 static int first_run;
782
783 if (first_run)
784 return;
785 dmi_check_system(processor_power_dmi_table);
786 max_cstate = acpi_processor_cstate_check(max_cstate);
787 if (max_cstate < ACPI_C_STATES_MAX)
788 pr_notice("ACPI: processor limited to max C-state %d\n",
789 max_cstate);
790 first_run++;
791
792 if (nocst)
793 return;
794
795 acpi_processor_claim_cst_control();
796}
797#else
798
799static inline int disabled_by_idle_boot_param(void) { return 0; }
800static inline void acpi_processor_cstate_first_run_checks(void) { }
801static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
802{
803 return -ENODEV;
804}
805
806static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
807 struct cpuidle_device *dev)
808{
809 return -EINVAL;
810}
811
812static int acpi_processor_setup_cstates(struct acpi_processor *pr)
813{
814 return -EINVAL;
815}
816
817#endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
818
819struct acpi_lpi_states_array {
820 unsigned int size;
821 unsigned int composite_states_size;
822 struct acpi_lpi_state *entries;
823 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
824};
825
826static int obj_get_integer(union acpi_object *obj, u32 *value)
827{
828 if (obj->type != ACPI_TYPE_INTEGER)
829 return -EINVAL;
830
831 *value = obj->integer.value;
832 return 0;
833}
834
835static int acpi_processor_evaluate_lpi(acpi_handle handle,
836 struct acpi_lpi_states_array *info)
837{
838 acpi_status status;
839 int ret = 0;
840 int pkg_count, state_idx = 1, loop;
841 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
842 union acpi_object *lpi_data;
843 struct acpi_lpi_state *lpi_state;
844
845 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
846 if (ACPI_FAILURE(status)) {
847 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n"));
848 return -ENODEV;
849 }
850
851 lpi_data = buffer.pointer;
852
853 /* There must be at least 4 elements = 3 elements + 1 package */
854 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
855 lpi_data->package.count < 4) {
856 pr_debug("not enough elements in _LPI\n");
857 ret = -ENODATA;
858 goto end;
859 }
860
861 pkg_count = lpi_data->package.elements[2].integer.value;
862
863 /* Validate number of power states. */
864 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
865 pr_debug("count given by _LPI is not valid\n");
866 ret = -ENODATA;
867 goto end;
868 }
869
870 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
871 if (!lpi_state) {
872 ret = -ENOMEM;
873 goto end;
874 }
875
876 info->size = pkg_count;
877 info->entries = lpi_state;
878
879 /* LPI States start at index 3 */
880 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
881 union acpi_object *element, *pkg_elem, *obj;
882
883 element = &lpi_data->package.elements[loop];
884 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
885 continue;
886
887 pkg_elem = element->package.elements;
888
889 obj = pkg_elem + 6;
890 if (obj->type == ACPI_TYPE_BUFFER) {
891 struct acpi_power_register *reg;
892
893 reg = (struct acpi_power_register *)obj->buffer.pointer;
894 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
895 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
896 continue;
897
898 lpi_state->address = reg->address;
899 lpi_state->entry_method =
900 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
901 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
902 } else if (obj->type == ACPI_TYPE_INTEGER) {
903 lpi_state->entry_method = ACPI_CSTATE_INTEGER;
904 lpi_state->address = obj->integer.value;
905 } else {
906 continue;
907 }
908
909 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
910
911 obj = pkg_elem + 9;
912 if (obj->type == ACPI_TYPE_STRING)
913 strlcpy(lpi_state->desc, obj->string.pointer,
914 ACPI_CX_DESC_LEN);
915
916 lpi_state->index = state_idx;
917 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
918 pr_debug("No min. residency found, assuming 10 us\n");
919 lpi_state->min_residency = 10;
920 }
921
922 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
923 pr_debug("No wakeup residency found, assuming 10 us\n");
924 lpi_state->wake_latency = 10;
925 }
926
927 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
928 lpi_state->flags = 0;
929
930 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
931 lpi_state->arch_flags = 0;
932
933 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
934 lpi_state->res_cnt_freq = 1;
935
936 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
937 lpi_state->enable_parent_state = 0;
938 }
939
940 acpi_handle_debug(handle, "Found %d power states\n", state_idx);
941end:
942 kfree(buffer.pointer);
943 return ret;
944}
945
946/*
947 * flat_state_cnt - the number of composite LPI states after the process of flattening
948 */
949static int flat_state_cnt;
950
951/**
952 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
953 *
954 * @local: local LPI state
955 * @parent: parent LPI state
956 * @result: composite LPI state
957 */
958static bool combine_lpi_states(struct acpi_lpi_state *local,
959 struct acpi_lpi_state *parent,
960 struct acpi_lpi_state *result)
961{
962 if (parent->entry_method == ACPI_CSTATE_INTEGER) {
963 if (!parent->address) /* 0 means autopromotable */
964 return false;
965 result->address = local->address + parent->address;
966 } else {
967 result->address = parent->address;
968 }
969
970 result->min_residency = max(local->min_residency, parent->min_residency);
971 result->wake_latency = local->wake_latency + parent->wake_latency;
972 result->enable_parent_state = parent->enable_parent_state;
973 result->entry_method = local->entry_method;
974
975 result->flags = parent->flags;
976 result->arch_flags = parent->arch_flags;
977 result->index = parent->index;
978
979 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
980 strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
981 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
982 return true;
983}
984
985#define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0)
986
987static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
988 struct acpi_lpi_state *t)
989{
990 curr_level->composite_states[curr_level->composite_states_size++] = t;
991}
992
993static int flatten_lpi_states(struct acpi_processor *pr,
994 struct acpi_lpi_states_array *curr_level,
995 struct acpi_lpi_states_array *prev_level)
996{
997 int i, j, state_count = curr_level->size;
998 struct acpi_lpi_state *p, *t = curr_level->entries;
999
1000 curr_level->composite_states_size = 0;
1001 for (j = 0; j < state_count; j++, t++) {
1002 struct acpi_lpi_state *flpi;
1003
1004 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1005 continue;
1006
1007 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1008 pr_warn("Limiting number of LPI states to max (%d)\n",
1009 ACPI_PROCESSOR_MAX_POWER);
1010 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1011 break;
1012 }
1013
1014 flpi = &pr->power.lpi_states[flat_state_cnt];
1015
1016 if (!prev_level) { /* leaf/processor node */
1017 memcpy(flpi, t, sizeof(*t));
1018 stash_composite_state(curr_level, flpi);
1019 flat_state_cnt++;
1020 continue;
1021 }
1022
1023 for (i = 0; i < prev_level->composite_states_size; i++) {
1024 p = prev_level->composite_states[i];
1025 if (t->index <= p->enable_parent_state &&
1026 combine_lpi_states(p, t, flpi)) {
1027 stash_composite_state(curr_level, flpi);
1028 flat_state_cnt++;
1029 flpi++;
1030 }
1031 }
1032 }
1033
1034 kfree(curr_level->entries);
1035 return 0;
1036}
1037
1038static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1039{
1040 int ret, i;
1041 acpi_status status;
1042 acpi_handle handle = pr->handle, pr_ahandle;
1043 struct acpi_device *d = NULL;
1044 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1045
1046 if (!osc_pc_lpi_support_confirmed)
1047 return -EOPNOTSUPP;
1048
1049 if (!acpi_has_method(handle, "_LPI"))
1050 return -EINVAL;
1051
1052 flat_state_cnt = 0;
1053 prev = &info[0];
1054 curr = &info[1];
1055 handle = pr->handle;
1056 ret = acpi_processor_evaluate_lpi(handle, prev);
1057 if (ret)
1058 return ret;
1059 flatten_lpi_states(pr, prev, NULL);
1060
1061 status = acpi_get_parent(handle, &pr_ahandle);
1062 while (ACPI_SUCCESS(status)) {
1063 acpi_bus_get_device(pr_ahandle, &d);
1064 handle = pr_ahandle;
1065
1066 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1067 break;
1068
1069 /* can be optional ? */
1070 if (!acpi_has_method(handle, "_LPI"))
1071 break;
1072
1073 ret = acpi_processor_evaluate_lpi(handle, curr);
1074 if (ret)
1075 break;
1076
1077 /* flatten all the LPI states in this level of hierarchy */
1078 flatten_lpi_states(pr, curr, prev);
1079
1080 tmp = prev, prev = curr, curr = tmp;
1081
1082 status = acpi_get_parent(handle, &pr_ahandle);
1083 }
1084
1085 pr->power.count = flat_state_cnt;
1086 /* reset the index after flattening */
1087 for (i = 0; i < pr->power.count; i++)
1088 pr->power.lpi_states[i].index = i;
1089
1090 /* Tell driver that _LPI is supported. */
1091 pr->flags.has_lpi = 1;
1092 pr->flags.power = 1;
1093
1094 return 0;
1095}
1096
1097int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1098{
1099 return -ENODEV;
1100}
1101
1102int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1103{
1104 return -ENODEV;
1105}
1106
1107/**
1108 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1109 * @dev: the target CPU
1110 * @drv: cpuidle driver containing cpuidle state info
1111 * @index: index of target state
1112 *
1113 * Return: 0 for success or negative value for error
1114 */
1115static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1116 struct cpuidle_driver *drv, int index)
1117{
1118 struct acpi_processor *pr;
1119 struct acpi_lpi_state *lpi;
1120
1121 pr = __this_cpu_read(processors);
1122
1123 if (unlikely(!pr))
1124 return -EINVAL;
1125
1126 lpi = &pr->power.lpi_states[index];
1127 if (lpi->entry_method == ACPI_CSTATE_FFH)
1128 return acpi_processor_ffh_lpi_enter(lpi);
1129
1130 return -EINVAL;
1131}
1132
1133static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1134{
1135 int i;
1136 struct acpi_lpi_state *lpi;
1137 struct cpuidle_state *state;
1138 struct cpuidle_driver *drv = &acpi_idle_driver;
1139
1140 if (!pr->flags.has_lpi)
1141 return -EOPNOTSUPP;
1142
1143 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1144 lpi = &pr->power.lpi_states[i];
1145
1146 state = &drv->states[i];
1147 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1148 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1149 state->exit_latency = lpi->wake_latency;
1150 state->target_residency = lpi->min_residency;
1151 if (lpi->arch_flags)
1152 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1153 state->enter = acpi_idle_lpi_enter;
1154 drv->safe_state_index = i;
1155 }
1156
1157 drv->state_count = i;
1158
1159 return 0;
1160}
1161
1162/**
1163 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1164 * global state data i.e. idle routines
1165 *
1166 * @pr: the ACPI processor
1167 */
1168static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1169{
1170 int i;
1171 struct cpuidle_driver *drv = &acpi_idle_driver;
1172
1173 if (!pr->flags.power_setup_done || !pr->flags.power)
1174 return -EINVAL;
1175
1176 drv->safe_state_index = -1;
1177 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1178 drv->states[i].name[0] = '\0';
1179 drv->states[i].desc[0] = '\0';
1180 }
1181
1182 if (pr->flags.has_lpi)
1183 return acpi_processor_setup_lpi_states(pr);
1184
1185 return acpi_processor_setup_cstates(pr);
1186}
1187
1188/**
1189 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1190 * device i.e. per-cpu data
1191 *
1192 * @pr: the ACPI processor
1193 * @dev : the cpuidle device
1194 */
1195static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1196 struct cpuidle_device *dev)
1197{
1198 if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1199 return -EINVAL;
1200
1201 dev->cpu = pr->id;
1202 if (pr->flags.has_lpi)
1203 return acpi_processor_ffh_lpi_probe(pr->id);
1204
1205 return acpi_processor_setup_cpuidle_cx(pr, dev);
1206}
1207
1208static int acpi_processor_get_power_info(struct acpi_processor *pr)
1209{
1210 int ret;
1211
1212 ret = acpi_processor_get_lpi_info(pr);
1213 if (ret)
1214 ret = acpi_processor_get_cstate_info(pr);
1215
1216 return ret;
1217}
1218
1219int acpi_processor_hotplug(struct acpi_processor *pr)
1220{
1221 int ret = 0;
1222 struct cpuidle_device *dev;
1223
1224 if (disabled_by_idle_boot_param())
1225 return 0;
1226
1227 if (!pr->flags.power_setup_done)
1228 return -ENODEV;
1229
1230 dev = per_cpu(acpi_cpuidle_device, pr->id);
1231 cpuidle_pause_and_lock();
1232 cpuidle_disable_device(dev);
1233 ret = acpi_processor_get_power_info(pr);
1234 if (!ret && pr->flags.power) {
1235 acpi_processor_setup_cpuidle_dev(pr, dev);
1236 ret = cpuidle_enable_device(dev);
1237 }
1238 cpuidle_resume_and_unlock();
1239
1240 return ret;
1241}
1242
1243int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1244{
1245 int cpu;
1246 struct acpi_processor *_pr;
1247 struct cpuidle_device *dev;
1248
1249 if (disabled_by_idle_boot_param())
1250 return 0;
1251
1252 if (!pr->flags.power_setup_done)
1253 return -ENODEV;
1254
1255 /*
1256 * FIXME: Design the ACPI notification to make it once per
1257 * system instead of once per-cpu. This condition is a hack
1258 * to make the code that updates C-States be called once.
1259 */
1260
1261 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1262
1263 /* Protect against cpu-hotplug */
1264 get_online_cpus();
1265 cpuidle_pause_and_lock();
1266
1267 /* Disable all cpuidle devices */
1268 for_each_online_cpu(cpu) {
1269 _pr = per_cpu(processors, cpu);
1270 if (!_pr || !_pr->flags.power_setup_done)
1271 continue;
1272 dev = per_cpu(acpi_cpuidle_device, cpu);
1273 cpuidle_disable_device(dev);
1274 }
1275
1276 /* Populate Updated C-state information */
1277 acpi_processor_get_power_info(pr);
1278 acpi_processor_setup_cpuidle_states(pr);
1279
1280 /* Enable all cpuidle devices */
1281 for_each_online_cpu(cpu) {
1282 _pr = per_cpu(processors, cpu);
1283 if (!_pr || !_pr->flags.power_setup_done)
1284 continue;
1285 acpi_processor_get_power_info(_pr);
1286 if (_pr->flags.power) {
1287 dev = per_cpu(acpi_cpuidle_device, cpu);
1288 acpi_processor_setup_cpuidle_dev(_pr, dev);
1289 cpuidle_enable_device(dev);
1290 }
1291 }
1292 cpuidle_resume_and_unlock();
1293 put_online_cpus();
1294 }
1295
1296 return 0;
1297}
1298
1299static int acpi_processor_registered;
1300
1301int acpi_processor_power_init(struct acpi_processor *pr)
1302{
1303 int retval;
1304 struct cpuidle_device *dev;
1305
1306 if (disabled_by_idle_boot_param())
1307 return 0;
1308
1309 acpi_processor_cstate_first_run_checks();
1310
1311 if (!acpi_processor_get_power_info(pr))
1312 pr->flags.power_setup_done = 1;
1313
1314 /*
1315 * Install the idle handler if processor power management is supported.
1316 * Note that we use previously set idle handler will be used on
1317 * platforms that only support C1.
1318 */
1319 if (pr->flags.power) {
1320 /* Register acpi_idle_driver if not already registered */
1321 if (!acpi_processor_registered) {
1322 acpi_processor_setup_cpuidle_states(pr);
1323 retval = cpuidle_register_driver(&acpi_idle_driver);
1324 if (retval)
1325 return retval;
1326 pr_debug("%s registered with cpuidle\n",
1327 acpi_idle_driver.name);
1328 }
1329
1330 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1331 if (!dev)
1332 return -ENOMEM;
1333 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1334
1335 acpi_processor_setup_cpuidle_dev(pr, dev);
1336
1337 /* Register per-cpu cpuidle_device. Cpuidle driver
1338 * must already be registered before registering device
1339 */
1340 retval = cpuidle_register_device(dev);
1341 if (retval) {
1342 if (acpi_processor_registered == 0)
1343 cpuidle_unregister_driver(&acpi_idle_driver);
1344 return retval;
1345 }
1346 acpi_processor_registered++;
1347 }
1348 return 0;
1349}
1350
1351int acpi_processor_power_exit(struct acpi_processor *pr)
1352{
1353 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1354
1355 if (disabled_by_idle_boot_param())
1356 return 0;
1357
1358 if (pr->flags.power) {
1359 cpuidle_unregister_device(dev);
1360 acpi_processor_registered--;
1361 if (acpi_processor_registered == 0)
1362 cpuidle_unregister_driver(&acpi_idle_driver);
1363 }
1364
1365 pr->flags.power_setup_done = 0;
1366 return 0;
1367}
1/*
2 * processor_idle - idle state submodule to the ACPI processor driver
3 *
4 * Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
5 * Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
6 * Copyright (C) 2004, 2005 Dominik Brodowski <linux@brodo.de>
7 * Copyright (C) 2004 Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
8 * - Added processor hotplug support
9 * Copyright (C) 2005 Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
10 * - Added support for C3 on SMP
11 *
12 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
13 *
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License as published by
16 * the Free Software Foundation; either version 2 of the License, or (at
17 * your option) any later version.
18 *
19 * This program is distributed in the hope that it will be useful, but
20 * WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
23 *
24 * You should have received a copy of the GNU General Public License along
25 * with this program; if not, write to the Free Software Foundation, Inc.,
26 * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
27 *
28 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
29 */
30
31#include <linux/module.h>
32#include <linux/acpi.h>
33#include <linux/dmi.h>
34#include <linux/sched.h> /* need_resched() */
35#include <linux/clockchips.h>
36#include <linux/cpuidle.h>
37#include <linux/syscore_ops.h>
38#include <acpi/processor.h>
39
40/*
41 * Include the apic definitions for x86 to have the APIC timer related defines
42 * available also for UP (on SMP it gets magically included via linux/smp.h).
43 * asm/acpi.h is not an option, as it would require more include magic. Also
44 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
45 */
46#ifdef CONFIG_X86
47#include <asm/apic.h>
48#endif
49
50#define PREFIX "ACPI: "
51
52#define ACPI_PROCESSOR_CLASS "processor"
53#define _COMPONENT ACPI_PROCESSOR_COMPONENT
54ACPI_MODULE_NAME("processor_idle");
55
56static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
57module_param(max_cstate, uint, 0000);
58static unsigned int nocst __read_mostly;
59module_param(nocst, uint, 0000);
60static int bm_check_disable __read_mostly;
61module_param(bm_check_disable, uint, 0000);
62
63static unsigned int latency_factor __read_mostly = 2;
64module_param(latency_factor, uint, 0644);
65
66static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
67
68static DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX],
69 acpi_cstate);
70
71static int disabled_by_idle_boot_param(void)
72{
73 return boot_option_idle_override == IDLE_POLL ||
74 boot_option_idle_override == IDLE_HALT;
75}
76
77/*
78 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
79 * For now disable this. Probably a bug somewhere else.
80 *
81 * To skip this limit, boot/load with a large max_cstate limit.
82 */
83static int set_max_cstate(const struct dmi_system_id *id)
84{
85 if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
86 return 0;
87
88 printk(KERN_NOTICE PREFIX "%s detected - limiting to C%ld max_cstate."
89 " Override with \"processor.max_cstate=%d\"\n", id->ident,
90 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
91
92 max_cstate = (long)id->driver_data;
93
94 return 0;
95}
96
97static struct dmi_system_id processor_power_dmi_table[] = {
98 { set_max_cstate, "Clevo 5600D", {
99 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
100 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
101 (void *)2},
102 { set_max_cstate, "Pavilion zv5000", {
103 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
104 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
105 (void *)1},
106 { set_max_cstate, "Asus L8400B", {
107 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
108 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
109 (void *)1},
110 {},
111};
112
113
114/*
115 * Callers should disable interrupts before the call and enable
116 * interrupts after return.
117 */
118static void acpi_safe_halt(void)
119{
120 if (!tif_need_resched()) {
121 safe_halt();
122 local_irq_disable();
123 }
124}
125
126#ifdef ARCH_APICTIMER_STOPS_ON_C3
127
128/*
129 * Some BIOS implementations switch to C3 in the published C2 state.
130 * This seems to be a common problem on AMD boxen, but other vendors
131 * are affected too. We pick the most conservative approach: we assume
132 * that the local APIC stops in both C2 and C3.
133 */
134static void lapic_timer_check_state(int state, struct acpi_processor *pr,
135 struct acpi_processor_cx *cx)
136{
137 struct acpi_processor_power *pwr = &pr->power;
138 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
139
140 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
141 return;
142
143 if (amd_e400_c1e_detected)
144 type = ACPI_STATE_C1;
145
146 /*
147 * Check, if one of the previous states already marked the lapic
148 * unstable
149 */
150 if (pwr->timer_broadcast_on_state < state)
151 return;
152
153 if (cx->type >= type)
154 pr->power.timer_broadcast_on_state = state;
155}
156
157static void __lapic_timer_propagate_broadcast(void *arg)
158{
159 struct acpi_processor *pr = (struct acpi_processor *) arg;
160 unsigned long reason;
161
162 reason = pr->power.timer_broadcast_on_state < INT_MAX ?
163 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
164
165 clockevents_notify(reason, &pr->id);
166}
167
168static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
169{
170 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
171 (void *)pr, 1);
172}
173
174/* Power(C) State timer broadcast control */
175static void lapic_timer_state_broadcast(struct acpi_processor *pr,
176 struct acpi_processor_cx *cx,
177 int broadcast)
178{
179 int state = cx - pr->power.states;
180
181 if (state >= pr->power.timer_broadcast_on_state) {
182 unsigned long reason;
183
184 reason = broadcast ? CLOCK_EVT_NOTIFY_BROADCAST_ENTER :
185 CLOCK_EVT_NOTIFY_BROADCAST_EXIT;
186 clockevents_notify(reason, &pr->id);
187 }
188}
189
190#else
191
192static void lapic_timer_check_state(int state, struct acpi_processor *pr,
193 struct acpi_processor_cx *cstate) { }
194static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
195static void lapic_timer_state_broadcast(struct acpi_processor *pr,
196 struct acpi_processor_cx *cx,
197 int broadcast)
198{
199}
200
201#endif
202
203#ifdef CONFIG_PM_SLEEP
204static u32 saved_bm_rld;
205
206static int acpi_processor_suspend(void)
207{
208 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &saved_bm_rld);
209 return 0;
210}
211
212static void acpi_processor_resume(void)
213{
214 u32 resumed_bm_rld = 0;
215
216 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_RLD, &resumed_bm_rld);
217 if (resumed_bm_rld == saved_bm_rld)
218 return;
219
220 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, saved_bm_rld);
221}
222
223static struct syscore_ops acpi_processor_syscore_ops = {
224 .suspend = acpi_processor_suspend,
225 .resume = acpi_processor_resume,
226};
227
228void acpi_processor_syscore_init(void)
229{
230 register_syscore_ops(&acpi_processor_syscore_ops);
231}
232
233void acpi_processor_syscore_exit(void)
234{
235 unregister_syscore_ops(&acpi_processor_syscore_ops);
236}
237#endif /* CONFIG_PM_SLEEP */
238
239#if defined(CONFIG_X86)
240static void tsc_check_state(int state)
241{
242 switch (boot_cpu_data.x86_vendor) {
243 case X86_VENDOR_AMD:
244 case X86_VENDOR_INTEL:
245 /*
246 * AMD Fam10h TSC will tick in all
247 * C/P/S0/S1 states when this bit is set.
248 */
249 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
250 return;
251
252 /*FALL THROUGH*/
253 default:
254 /* TSC could halt in idle, so notify users */
255 if (state > ACPI_STATE_C1)
256 mark_tsc_unstable("TSC halts in idle");
257 }
258}
259#else
260static void tsc_check_state(int state) { return; }
261#endif
262
263static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
264{
265
266 if (!pr->pblk)
267 return -ENODEV;
268
269 /* if info is obtained from pblk/fadt, type equals state */
270 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
271 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
272
273#ifndef CONFIG_HOTPLUG_CPU
274 /*
275 * Check for P_LVL2_UP flag before entering C2 and above on
276 * an SMP system.
277 */
278 if ((num_online_cpus() > 1) &&
279 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
280 return -ENODEV;
281#endif
282
283 /* determine C2 and C3 address from pblk */
284 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
285 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
286
287 /* determine latencies from FADT */
288 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
289 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
290
291 /*
292 * FADT specified C2 latency must be less than or equal to
293 * 100 microseconds.
294 */
295 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
296 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
297 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
298 /* invalidate C2 */
299 pr->power.states[ACPI_STATE_C2].address = 0;
300 }
301
302 /*
303 * FADT supplied C3 latency must be less than or equal to
304 * 1000 microseconds.
305 */
306 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
307 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
308 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
309 /* invalidate C3 */
310 pr->power.states[ACPI_STATE_C3].address = 0;
311 }
312
313 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
314 "lvl2[0x%08x] lvl3[0x%08x]\n",
315 pr->power.states[ACPI_STATE_C2].address,
316 pr->power.states[ACPI_STATE_C3].address));
317
318 return 0;
319}
320
321static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
322{
323 if (!pr->power.states[ACPI_STATE_C1].valid) {
324 /* set the first C-State to C1 */
325 /* all processors need to support C1 */
326 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
327 pr->power.states[ACPI_STATE_C1].valid = 1;
328 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
329 }
330 /* the C0 state only exists as a filler in our array */
331 pr->power.states[ACPI_STATE_C0].valid = 1;
332 return 0;
333}
334
335static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
336{
337 acpi_status status = 0;
338 u64 count;
339 int current_count;
340 int i;
341 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
342 union acpi_object *cst;
343
344
345 if (nocst)
346 return -ENODEV;
347
348 current_count = 0;
349
350 status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
351 if (ACPI_FAILURE(status)) {
352 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
353 return -ENODEV;
354 }
355
356 cst = buffer.pointer;
357
358 /* There must be at least 2 elements */
359 if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
360 printk(KERN_ERR PREFIX "not enough elements in _CST\n");
361 status = -EFAULT;
362 goto end;
363 }
364
365 count = cst->package.elements[0].integer.value;
366
367 /* Validate number of power states. */
368 if (count < 1 || count != cst->package.count - 1) {
369 printk(KERN_ERR PREFIX "count given by _CST is not valid\n");
370 status = -EFAULT;
371 goto end;
372 }
373
374 /* Tell driver that at least _CST is supported. */
375 pr->flags.has_cst = 1;
376
377 for (i = 1; i <= count; i++) {
378 union acpi_object *element;
379 union acpi_object *obj;
380 struct acpi_power_register *reg;
381 struct acpi_processor_cx cx;
382
383 memset(&cx, 0, sizeof(cx));
384
385 element = &(cst->package.elements[i]);
386 if (element->type != ACPI_TYPE_PACKAGE)
387 continue;
388
389 if (element->package.count != 4)
390 continue;
391
392 obj = &(element->package.elements[0]);
393
394 if (obj->type != ACPI_TYPE_BUFFER)
395 continue;
396
397 reg = (struct acpi_power_register *)obj->buffer.pointer;
398
399 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
400 (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
401 continue;
402
403 /* There should be an easy way to extract an integer... */
404 obj = &(element->package.elements[1]);
405 if (obj->type != ACPI_TYPE_INTEGER)
406 continue;
407
408 cx.type = obj->integer.value;
409 /*
410 * Some buggy BIOSes won't list C1 in _CST -
411 * Let acpi_processor_get_power_info_default() handle them later
412 */
413 if (i == 1 && cx.type != ACPI_STATE_C1)
414 current_count++;
415
416 cx.address = reg->address;
417 cx.index = current_count + 1;
418
419 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
420 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
421 if (acpi_processor_ffh_cstate_probe
422 (pr->id, &cx, reg) == 0) {
423 cx.entry_method = ACPI_CSTATE_FFH;
424 } else if (cx.type == ACPI_STATE_C1) {
425 /*
426 * C1 is a special case where FIXED_HARDWARE
427 * can be handled in non-MWAIT way as well.
428 * In that case, save this _CST entry info.
429 * Otherwise, ignore this info and continue.
430 */
431 cx.entry_method = ACPI_CSTATE_HALT;
432 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
433 } else {
434 continue;
435 }
436 if (cx.type == ACPI_STATE_C1 &&
437 (boot_option_idle_override == IDLE_NOMWAIT)) {
438 /*
439 * In most cases the C1 space_id obtained from
440 * _CST object is FIXED_HARDWARE access mode.
441 * But when the option of idle=halt is added,
442 * the entry_method type should be changed from
443 * CSTATE_FFH to CSTATE_HALT.
444 * When the option of idle=nomwait is added,
445 * the C1 entry_method type should be
446 * CSTATE_HALT.
447 */
448 cx.entry_method = ACPI_CSTATE_HALT;
449 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
450 }
451 } else {
452 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
453 cx.address);
454 }
455
456 if (cx.type == ACPI_STATE_C1) {
457 cx.valid = 1;
458 }
459
460 obj = &(element->package.elements[2]);
461 if (obj->type != ACPI_TYPE_INTEGER)
462 continue;
463
464 cx.latency = obj->integer.value;
465
466 obj = &(element->package.elements[3]);
467 if (obj->type != ACPI_TYPE_INTEGER)
468 continue;
469
470 current_count++;
471 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
472
473 /*
474 * We support total ACPI_PROCESSOR_MAX_POWER - 1
475 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
476 */
477 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
478 printk(KERN_WARNING
479 "Limiting number of power states to max (%d)\n",
480 ACPI_PROCESSOR_MAX_POWER);
481 printk(KERN_WARNING
482 "Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
483 break;
484 }
485 }
486
487 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
488 current_count));
489
490 /* Validate number of power states discovered */
491 if (current_count < 2)
492 status = -EFAULT;
493
494 end:
495 kfree(buffer.pointer);
496
497 return status;
498}
499
500static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
501 struct acpi_processor_cx *cx)
502{
503 static int bm_check_flag = -1;
504 static int bm_control_flag = -1;
505
506
507 if (!cx->address)
508 return;
509
510 /*
511 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
512 * DMA transfers are used by any ISA device to avoid livelock.
513 * Note that we could disable Type-F DMA (as recommended by
514 * the erratum), but this is known to disrupt certain ISA
515 * devices thus we take the conservative approach.
516 */
517 else if (errata.piix4.fdma) {
518 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
519 "C3 not supported on PIIX4 with Type-F DMA\n"));
520 return;
521 }
522
523 /* All the logic here assumes flags.bm_check is same across all CPUs */
524 if (bm_check_flag == -1) {
525 /* Determine whether bm_check is needed based on CPU */
526 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
527 bm_check_flag = pr->flags.bm_check;
528 bm_control_flag = pr->flags.bm_control;
529 } else {
530 pr->flags.bm_check = bm_check_flag;
531 pr->flags.bm_control = bm_control_flag;
532 }
533
534 if (pr->flags.bm_check) {
535 if (!pr->flags.bm_control) {
536 if (pr->flags.has_cst != 1) {
537 /* bus mastering control is necessary */
538 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
539 "C3 support requires BM control\n"));
540 return;
541 } else {
542 /* Here we enter C3 without bus mastering */
543 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
544 "C3 support without BM control\n"));
545 }
546 }
547 } else {
548 /*
549 * WBINVD should be set in fadt, for C3 state to be
550 * supported on when bm_check is not required.
551 */
552 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
553 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
554 "Cache invalidation should work properly"
555 " for C3 to be enabled on SMP systems\n"));
556 return;
557 }
558 }
559
560 /*
561 * Otherwise we've met all of our C3 requirements.
562 * Normalize the C3 latency to expidite policy. Enable
563 * checking of bus mastering status (bm_check) so we can
564 * use this in our C3 policy
565 */
566 cx->valid = 1;
567
568 /*
569 * On older chipsets, BM_RLD needs to be set
570 * in order for Bus Master activity to wake the
571 * system from C3. Newer chipsets handle DMA
572 * during C3 automatically and BM_RLD is a NOP.
573 * In either case, the proper way to
574 * handle BM_RLD is to set it and leave it set.
575 */
576 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
577
578 return;
579}
580
581static int acpi_processor_power_verify(struct acpi_processor *pr)
582{
583 unsigned int i;
584 unsigned int working = 0;
585
586 pr->power.timer_broadcast_on_state = INT_MAX;
587
588 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
589 struct acpi_processor_cx *cx = &pr->power.states[i];
590
591 switch (cx->type) {
592 case ACPI_STATE_C1:
593 cx->valid = 1;
594 break;
595
596 case ACPI_STATE_C2:
597 if (!cx->address)
598 break;
599 cx->valid = 1;
600 break;
601
602 case ACPI_STATE_C3:
603 acpi_processor_power_verify_c3(pr, cx);
604 break;
605 }
606 if (!cx->valid)
607 continue;
608
609 lapic_timer_check_state(i, pr, cx);
610 tsc_check_state(cx->type);
611 working++;
612 }
613
614 lapic_timer_propagate_broadcast(pr);
615
616 return (working);
617}
618
619static int acpi_processor_get_power_info(struct acpi_processor *pr)
620{
621 unsigned int i;
622 int result;
623
624
625 /* NOTE: the idle thread may not be running while calling
626 * this function */
627
628 /* Zero initialize all the C-states info. */
629 memset(pr->power.states, 0, sizeof(pr->power.states));
630
631 result = acpi_processor_get_power_info_cst(pr);
632 if (result == -ENODEV)
633 result = acpi_processor_get_power_info_fadt(pr);
634
635 if (result)
636 return result;
637
638 acpi_processor_get_power_info_default(pr);
639
640 pr->power.count = acpi_processor_power_verify(pr);
641
642 /*
643 * if one state of type C2 or C3 is available, mark this
644 * CPU as being "idle manageable"
645 */
646 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
647 if (pr->power.states[i].valid) {
648 pr->power.count = i;
649 if (pr->power.states[i].type >= ACPI_STATE_C2)
650 pr->flags.power = 1;
651 }
652 }
653
654 return 0;
655}
656
657/**
658 * acpi_idle_bm_check - checks if bus master activity was detected
659 */
660static int acpi_idle_bm_check(void)
661{
662 u32 bm_status = 0;
663
664 if (bm_check_disable)
665 return 0;
666
667 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
668 if (bm_status)
669 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
670 /*
671 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
672 * the true state of bus mastering activity; forcing us to
673 * manually check the BMIDEA bit of each IDE channel.
674 */
675 else if (errata.piix4.bmisx) {
676 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
677 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
678 bm_status = 1;
679 }
680 return bm_status;
681}
682
683/**
684 * acpi_idle_do_entry - a helper function that does C2 and C3 type entry
685 * @cx: cstate data
686 *
687 * Caller disables interrupt before call and enables interrupt after return.
688 */
689static inline void acpi_idle_do_entry(struct acpi_processor_cx *cx)
690{
691 /* Don't trace irqs off for idle */
692 stop_critical_timings();
693 if (cx->entry_method == ACPI_CSTATE_FFH) {
694 /* Call into architectural FFH based C-state */
695 acpi_processor_ffh_cstate_enter(cx);
696 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
697 acpi_safe_halt();
698 } else {
699 /* IO port based C-state */
700 inb(cx->address);
701 /* Dummy wait op - must do something useless after P_LVL2 read
702 because chipsets cannot guarantee that STPCLK# signal
703 gets asserted in time to freeze execution properly. */
704 inl(acpi_gbl_FADT.xpm_timer_block.address);
705 }
706 start_critical_timings();
707}
708
709/**
710 * acpi_idle_enter_c1 - enters an ACPI C1 state-type
711 * @dev: the target CPU
712 * @drv: cpuidle driver containing cpuidle state info
713 * @index: index of target state
714 *
715 * This is equivalent to the HALT instruction.
716 */
717static int acpi_idle_enter_c1(struct cpuidle_device *dev,
718 struct cpuidle_driver *drv, int index)
719{
720 struct acpi_processor *pr;
721 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
722
723 pr = __this_cpu_read(processors);
724
725 if (unlikely(!pr))
726 return -EINVAL;
727
728 lapic_timer_state_broadcast(pr, cx, 1);
729 acpi_idle_do_entry(cx);
730
731 lapic_timer_state_broadcast(pr, cx, 0);
732
733 return index;
734}
735
736
737/**
738 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
739 * @dev: the target CPU
740 * @index: the index of suggested state
741 */
742static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
743{
744 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
745
746 ACPI_FLUSH_CPU_CACHE();
747
748 while (1) {
749
750 if (cx->entry_method == ACPI_CSTATE_HALT)
751 safe_halt();
752 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
753 inb(cx->address);
754 /* See comment in acpi_idle_do_entry() */
755 inl(acpi_gbl_FADT.xpm_timer_block.address);
756 } else
757 return -ENODEV;
758 }
759
760 /* Never reached */
761 return 0;
762}
763
764/**
765 * acpi_idle_enter_simple - enters an ACPI state without BM handling
766 * @dev: the target CPU
767 * @drv: cpuidle driver with cpuidle state information
768 * @index: the index of suggested state
769 */
770static int acpi_idle_enter_simple(struct cpuidle_device *dev,
771 struct cpuidle_driver *drv, int index)
772{
773 struct acpi_processor *pr;
774 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
775
776 pr = __this_cpu_read(processors);
777
778 if (unlikely(!pr))
779 return -EINVAL;
780
781#ifdef CONFIG_HOTPLUG_CPU
782 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
783 !pr->flags.has_cst &&
784 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
785 return acpi_idle_enter_c1(dev, drv, CPUIDLE_DRIVER_STATE_START);
786#endif
787
788 /*
789 * Must be done before busmaster disable as we might need to
790 * access HPET !
791 */
792 lapic_timer_state_broadcast(pr, cx, 1);
793
794 if (cx->type == ACPI_STATE_C3)
795 ACPI_FLUSH_CPU_CACHE();
796
797 /* Tell the scheduler that we are going deep-idle: */
798 sched_clock_idle_sleep_event();
799 acpi_idle_do_entry(cx);
800
801 sched_clock_idle_wakeup_event(0);
802
803 lapic_timer_state_broadcast(pr, cx, 0);
804 return index;
805}
806
807static int c3_cpu_count;
808static DEFINE_RAW_SPINLOCK(c3_lock);
809
810/**
811 * acpi_idle_enter_bm - enters C3 with proper BM handling
812 * @dev: the target CPU
813 * @drv: cpuidle driver containing state data
814 * @index: the index of suggested state
815 *
816 * If BM is detected, the deepest non-C3 idle state is entered instead.
817 */
818static int acpi_idle_enter_bm(struct cpuidle_device *dev,
819 struct cpuidle_driver *drv, int index)
820{
821 struct acpi_processor *pr;
822 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
823
824 pr = __this_cpu_read(processors);
825
826 if (unlikely(!pr))
827 return -EINVAL;
828
829#ifdef CONFIG_HOTPLUG_CPU
830 if ((cx->type != ACPI_STATE_C1) && (num_online_cpus() > 1) &&
831 !pr->flags.has_cst &&
832 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
833 return acpi_idle_enter_c1(dev, drv, CPUIDLE_DRIVER_STATE_START);
834#endif
835
836 if (!cx->bm_sts_skip && acpi_idle_bm_check()) {
837 if (drv->safe_state_index >= 0) {
838 return drv->states[drv->safe_state_index].enter(dev,
839 drv, drv->safe_state_index);
840 } else {
841 acpi_safe_halt();
842 return -EBUSY;
843 }
844 }
845
846 acpi_unlazy_tlb(smp_processor_id());
847
848 /* Tell the scheduler that we are going deep-idle: */
849 sched_clock_idle_sleep_event();
850 /*
851 * Must be done before busmaster disable as we might need to
852 * access HPET !
853 */
854 lapic_timer_state_broadcast(pr, cx, 1);
855
856 /*
857 * disable bus master
858 * bm_check implies we need ARB_DIS
859 * !bm_check implies we need cache flush
860 * bm_control implies whether we can do ARB_DIS
861 *
862 * That leaves a case where bm_check is set and bm_control is
863 * not set. In that case we cannot do much, we enter C3
864 * without doing anything.
865 */
866 if (pr->flags.bm_check && pr->flags.bm_control) {
867 raw_spin_lock(&c3_lock);
868 c3_cpu_count++;
869 /* Disable bus master arbitration when all CPUs are in C3 */
870 if (c3_cpu_count == num_online_cpus())
871 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
872 raw_spin_unlock(&c3_lock);
873 } else if (!pr->flags.bm_check) {
874 ACPI_FLUSH_CPU_CACHE();
875 }
876
877 acpi_idle_do_entry(cx);
878
879 /* Re-enable bus master arbitration */
880 if (pr->flags.bm_check && pr->flags.bm_control) {
881 raw_spin_lock(&c3_lock);
882 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
883 c3_cpu_count--;
884 raw_spin_unlock(&c3_lock);
885 }
886
887 sched_clock_idle_wakeup_event(0);
888
889 lapic_timer_state_broadcast(pr, cx, 0);
890 return index;
891}
892
893struct cpuidle_driver acpi_idle_driver = {
894 .name = "acpi_idle",
895 .owner = THIS_MODULE,
896};
897
898/**
899 * acpi_processor_setup_cpuidle_cx - prepares and configures CPUIDLE
900 * device i.e. per-cpu data
901 *
902 * @pr: the ACPI processor
903 * @dev : the cpuidle device
904 */
905static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
906 struct cpuidle_device *dev)
907{
908 int i, count = CPUIDLE_DRIVER_STATE_START;
909 struct acpi_processor_cx *cx;
910
911 if (!pr->flags.power_setup_done)
912 return -EINVAL;
913
914 if (pr->flags.power == 0) {
915 return -EINVAL;
916 }
917
918 if (!dev)
919 return -EINVAL;
920
921 dev->cpu = pr->id;
922
923 if (max_cstate == 0)
924 max_cstate = 1;
925
926 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
927 cx = &pr->power.states[i];
928
929 if (!cx->valid)
930 continue;
931
932 per_cpu(acpi_cstate[count], dev->cpu) = cx;
933
934 count++;
935 if (count == CPUIDLE_STATE_MAX)
936 break;
937 }
938
939 if (!count)
940 return -EINVAL;
941
942 return 0;
943}
944
945/**
946 * acpi_processor_setup_cpuidle states- prepares and configures cpuidle
947 * global state data i.e. idle routines
948 *
949 * @pr: the ACPI processor
950 */
951static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
952{
953 int i, count = CPUIDLE_DRIVER_STATE_START;
954 struct acpi_processor_cx *cx;
955 struct cpuidle_state *state;
956 struct cpuidle_driver *drv = &acpi_idle_driver;
957
958 if (!pr->flags.power_setup_done)
959 return -EINVAL;
960
961 if (pr->flags.power == 0)
962 return -EINVAL;
963
964 drv->safe_state_index = -1;
965 for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
966 drv->states[i].name[0] = '\0';
967 drv->states[i].desc[0] = '\0';
968 }
969
970 if (max_cstate == 0)
971 max_cstate = 1;
972
973 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
974 cx = &pr->power.states[i];
975
976 if (!cx->valid)
977 continue;
978
979 state = &drv->states[count];
980 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
981 strncpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
982 state->exit_latency = cx->latency;
983 state->target_residency = cx->latency * latency_factor;
984
985 state->flags = 0;
986 switch (cx->type) {
987 case ACPI_STATE_C1:
988 if (cx->entry_method == ACPI_CSTATE_FFH)
989 state->flags |= CPUIDLE_FLAG_TIME_VALID;
990
991 state->enter = acpi_idle_enter_c1;
992 state->enter_dead = acpi_idle_play_dead;
993 drv->safe_state_index = count;
994 break;
995
996 case ACPI_STATE_C2:
997 state->flags |= CPUIDLE_FLAG_TIME_VALID;
998 state->enter = acpi_idle_enter_simple;
999 state->enter_dead = acpi_idle_play_dead;
1000 drv->safe_state_index = count;
1001 break;
1002
1003 case ACPI_STATE_C3:
1004 state->flags |= CPUIDLE_FLAG_TIME_VALID;
1005 state->enter = pr->flags.bm_check ?
1006 acpi_idle_enter_bm :
1007 acpi_idle_enter_simple;
1008 break;
1009 }
1010
1011 count++;
1012 if (count == CPUIDLE_STATE_MAX)
1013 break;
1014 }
1015
1016 drv->state_count = count;
1017
1018 if (!count)
1019 return -EINVAL;
1020
1021 return 0;
1022}
1023
1024int acpi_processor_hotplug(struct acpi_processor *pr)
1025{
1026 int ret = 0;
1027 struct cpuidle_device *dev;
1028
1029 if (disabled_by_idle_boot_param())
1030 return 0;
1031
1032 if (nocst)
1033 return -ENODEV;
1034
1035 if (!pr->flags.power_setup_done)
1036 return -ENODEV;
1037
1038 dev = per_cpu(acpi_cpuidle_device, pr->id);
1039 cpuidle_pause_and_lock();
1040 cpuidle_disable_device(dev);
1041 acpi_processor_get_power_info(pr);
1042 if (pr->flags.power) {
1043 acpi_processor_setup_cpuidle_cx(pr, dev);
1044 ret = cpuidle_enable_device(dev);
1045 }
1046 cpuidle_resume_and_unlock();
1047
1048 return ret;
1049}
1050
1051int acpi_processor_cst_has_changed(struct acpi_processor *pr)
1052{
1053 int cpu;
1054 struct acpi_processor *_pr;
1055 struct cpuidle_device *dev;
1056
1057 if (disabled_by_idle_boot_param())
1058 return 0;
1059
1060 if (nocst)
1061 return -ENODEV;
1062
1063 if (!pr->flags.power_setup_done)
1064 return -ENODEV;
1065
1066 /*
1067 * FIXME: Design the ACPI notification to make it once per
1068 * system instead of once per-cpu. This condition is a hack
1069 * to make the code that updates C-States be called once.
1070 */
1071
1072 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1073
1074 cpuidle_pause_and_lock();
1075 /* Protect against cpu-hotplug */
1076 get_online_cpus();
1077
1078 /* Disable all cpuidle devices */
1079 for_each_online_cpu(cpu) {
1080 _pr = per_cpu(processors, cpu);
1081 if (!_pr || !_pr->flags.power_setup_done)
1082 continue;
1083 dev = per_cpu(acpi_cpuidle_device, cpu);
1084 cpuidle_disable_device(dev);
1085 }
1086
1087 /* Populate Updated C-state information */
1088 acpi_processor_get_power_info(pr);
1089 acpi_processor_setup_cpuidle_states(pr);
1090
1091 /* Enable all cpuidle devices */
1092 for_each_online_cpu(cpu) {
1093 _pr = per_cpu(processors, cpu);
1094 if (!_pr || !_pr->flags.power_setup_done)
1095 continue;
1096 acpi_processor_get_power_info(_pr);
1097 if (_pr->flags.power) {
1098 dev = per_cpu(acpi_cpuidle_device, cpu);
1099 acpi_processor_setup_cpuidle_cx(_pr, dev);
1100 cpuidle_enable_device(dev);
1101 }
1102 }
1103 put_online_cpus();
1104 cpuidle_resume_and_unlock();
1105 }
1106
1107 return 0;
1108}
1109
1110static int acpi_processor_registered;
1111
1112int acpi_processor_power_init(struct acpi_processor *pr)
1113{
1114 acpi_status status = 0;
1115 int retval;
1116 struct cpuidle_device *dev;
1117 static int first_run;
1118
1119 if (disabled_by_idle_boot_param())
1120 return 0;
1121
1122 if (!first_run) {
1123 dmi_check_system(processor_power_dmi_table);
1124 max_cstate = acpi_processor_cstate_check(max_cstate);
1125 if (max_cstate < ACPI_C_STATES_MAX)
1126 printk(KERN_NOTICE
1127 "ACPI: processor limited to max C-state %d\n",
1128 max_cstate);
1129 first_run++;
1130 }
1131
1132 if (acpi_gbl_FADT.cst_control && !nocst) {
1133 status =
1134 acpi_os_write_port(acpi_gbl_FADT.smi_command, acpi_gbl_FADT.cst_control, 8);
1135 if (ACPI_FAILURE(status)) {
1136 ACPI_EXCEPTION((AE_INFO, status,
1137 "Notifying BIOS of _CST ability failed"));
1138 }
1139 }
1140
1141 acpi_processor_get_power_info(pr);
1142 pr->flags.power_setup_done = 1;
1143
1144 /*
1145 * Install the idle handler if processor power management is supported.
1146 * Note that we use previously set idle handler will be used on
1147 * platforms that only support C1.
1148 */
1149 if (pr->flags.power) {
1150 /* Register acpi_idle_driver if not already registered */
1151 if (!acpi_processor_registered) {
1152 acpi_processor_setup_cpuidle_states(pr);
1153 retval = cpuidle_register_driver(&acpi_idle_driver);
1154 if (retval)
1155 return retval;
1156 printk(KERN_DEBUG "ACPI: %s registered with cpuidle\n",
1157 acpi_idle_driver.name);
1158 }
1159
1160 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1161 if (!dev)
1162 return -ENOMEM;
1163 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1164
1165 acpi_processor_setup_cpuidle_cx(pr, dev);
1166
1167 /* Register per-cpu cpuidle_device. Cpuidle driver
1168 * must already be registered before registering device
1169 */
1170 retval = cpuidle_register_device(dev);
1171 if (retval) {
1172 if (acpi_processor_registered == 0)
1173 cpuidle_unregister_driver(&acpi_idle_driver);
1174 return retval;
1175 }
1176 acpi_processor_registered++;
1177 }
1178 return 0;
1179}
1180
1181int acpi_processor_power_exit(struct acpi_processor *pr)
1182{
1183 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1184
1185 if (disabled_by_idle_boot_param())
1186 return 0;
1187
1188 if (pr->flags.power) {
1189 cpuidle_unregister_device(dev);
1190 acpi_processor_registered--;
1191 if (acpi_processor_registered == 0)
1192 cpuidle_unregister_driver(&acpi_idle_driver);
1193 }
1194
1195 pr->flags.power_setup_done = 0;
1196 return 0;
1197}