2 * processor_idle - idle state submodule to the ACPI processor driver
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
12 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
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.
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.
24 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
26 #define pr_fmt(fmt) "ACPI: " fmt
28 #include <linux/module.h>
29 #include <linux/acpi.h>
30 #include <linux/dmi.h>
31 #include <linux/sched.h> /* need_resched() */
32 #include <linux/tick.h>
33 #include <linux/cpuidle.h>
34 #include <linux/cpu.h>
35 #include <acpi/processor.h>
38 * Include the apic definitions for x86 to have the APIC timer related defines
39 * available also for UP (on SMP it gets magically included via linux/smp.h).
40 * asm/acpi.h is not an option, as it would require more include magic. Also
41 * creating an empty asm-ia64/apic.h would just trade pest vs. cholera.
47 #define ACPI_PROCESSOR_CLASS "processor"
48 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
49 ACPI_MODULE_NAME("processor_idle");
51 #define ACPI_IDLE_STATE_START (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
53 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
54 module_param(max_cstate, uint, 0000);
55 static unsigned int nocst __read_mostly;
56 module_param(nocst, uint, 0000);
57 static int bm_check_disable __read_mostly;
58 module_param(bm_check_disable, uint, 0000);
60 static unsigned int latency_factor __read_mostly = 2;
61 module_param(latency_factor, uint, 0644);
63 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
65 struct cpuidle_driver acpi_idle_driver = {
70 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
72 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
74 static int disabled_by_idle_boot_param(void)
76 return boot_option_idle_override == IDLE_POLL ||
77 boot_option_idle_override == IDLE_HALT;
81 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
82 * For now disable this. Probably a bug somewhere else.
84 * To skip this limit, boot/load with a large max_cstate limit.
86 static int set_max_cstate(const struct dmi_system_id *id)
88 if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
91 pr_notice("%s detected - limiting to C%ld max_cstate."
92 " Override with \"processor.max_cstate=%d\"\n", id->ident,
93 (long)id->driver_data, ACPI_PROCESSOR_MAX_POWER + 1);
95 max_cstate = (long)id->driver_data;
100 static const struct dmi_system_id processor_power_dmi_table[] = {
101 { set_max_cstate, "Clevo 5600D", {
102 DMI_MATCH(DMI_BIOS_VENDOR,"Phoenix Technologies LTD"),
103 DMI_MATCH(DMI_BIOS_VERSION,"SHE845M0.86C.0013.D.0302131307")},
105 { set_max_cstate, "Pavilion zv5000", {
106 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
107 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
109 { set_max_cstate, "Asus L8400B", {
110 DMI_MATCH(DMI_SYS_VENDOR, "ASUSTeK Computer Inc."),
111 DMI_MATCH(DMI_PRODUCT_NAME,"L8400B series Notebook PC")},
118 * Callers should disable interrupts before the call and enable
119 * interrupts after return.
121 static void __cpuidle acpi_safe_halt(void)
123 if (!tif_need_resched()) {
129 #ifdef ARCH_APICTIMER_STOPS_ON_C3
132 * Some BIOS implementations switch to C3 in the published C2 state.
133 * This seems to be a common problem on AMD boxen, but other vendors
134 * are affected too. We pick the most conservative approach: we assume
135 * that the local APIC stops in both C2 and C3.
137 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
138 struct acpi_processor_cx *cx)
140 struct acpi_processor_power *pwr = &pr->power;
141 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
143 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
146 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
147 type = ACPI_STATE_C1;
150 * Check, if one of the previous states already marked the lapic
153 if (pwr->timer_broadcast_on_state < state)
156 if (cx->type >= type)
157 pr->power.timer_broadcast_on_state = state;
160 static void __lapic_timer_propagate_broadcast(void *arg)
162 struct acpi_processor *pr = (struct acpi_processor *) arg;
164 if (pr->power.timer_broadcast_on_state < INT_MAX)
165 tick_broadcast_enable();
167 tick_broadcast_disable();
170 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
172 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
176 /* Power(C) State timer broadcast control */
177 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
178 struct acpi_processor_cx *cx,
181 int state = cx - pr->power.states;
183 if (state >= pr->power.timer_broadcast_on_state) {
185 tick_broadcast_enter();
187 tick_broadcast_exit();
193 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
194 struct acpi_processor_cx *cstate) { }
195 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
196 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
197 struct acpi_processor_cx *cx,
204 #if defined(CONFIG_X86)
205 static void tsc_check_state(int state)
207 switch (boot_cpu_data.x86_vendor) {
209 case X86_VENDOR_INTEL:
210 case X86_VENDOR_CENTAUR:
212 * AMD Fam10h TSC will tick in all
213 * C/P/S0/S1 states when this bit is set.
215 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
220 /* TSC could halt in idle, so notify users */
221 if (state > ACPI_STATE_C1)
222 mark_tsc_unstable("TSC halts in idle");
226 static void tsc_check_state(int state) { return; }
229 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
235 /* if info is obtained from pblk/fadt, type equals state */
236 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
237 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
239 #ifndef CONFIG_HOTPLUG_CPU
241 * Check for P_LVL2_UP flag before entering C2 and above on
244 if ((num_online_cpus() > 1) &&
245 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
249 /* determine C2 and C3 address from pblk */
250 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
251 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
253 /* determine latencies from FADT */
254 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
255 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
258 * FADT specified C2 latency must be less than or equal to
261 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
262 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
263 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
265 pr->power.states[ACPI_STATE_C2].address = 0;
269 * FADT supplied C3 latency must be less than or equal to
272 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
273 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
274 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
276 pr->power.states[ACPI_STATE_C3].address = 0;
279 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
280 "lvl2[0x%08x] lvl3[0x%08x]\n",
281 pr->power.states[ACPI_STATE_C2].address,
282 pr->power.states[ACPI_STATE_C3].address));
287 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
289 if (!pr->power.states[ACPI_STATE_C1].valid) {
290 /* set the first C-State to C1 */
291 /* all processors need to support C1 */
292 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
293 pr->power.states[ACPI_STATE_C1].valid = 1;
294 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
296 /* the C0 state only exists as a filler in our array */
297 pr->power.states[ACPI_STATE_C0].valid = 1;
301 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
307 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
308 union acpi_object *cst;
315 status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
316 if (ACPI_FAILURE(status)) {
317 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
321 cst = buffer.pointer;
323 /* There must be at least 2 elements */
324 if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
325 pr_err("not enough elements in _CST\n");
330 count = cst->package.elements[0].integer.value;
332 /* Validate number of power states. */
333 if (count < 1 || count != cst->package.count - 1) {
334 pr_err("count given by _CST is not valid\n");
339 /* Tell driver that at least _CST is supported. */
340 pr->flags.has_cst = 1;
342 for (i = 1; i <= count; i++) {
343 union acpi_object *element;
344 union acpi_object *obj;
345 struct acpi_power_register *reg;
346 struct acpi_processor_cx cx;
348 memset(&cx, 0, sizeof(cx));
350 element = &(cst->package.elements[i]);
351 if (element->type != ACPI_TYPE_PACKAGE)
354 if (element->package.count != 4)
357 obj = &(element->package.elements[0]);
359 if (obj->type != ACPI_TYPE_BUFFER)
362 reg = (struct acpi_power_register *)obj->buffer.pointer;
364 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
365 (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
368 /* There should be an easy way to extract an integer... */
369 obj = &(element->package.elements[1]);
370 if (obj->type != ACPI_TYPE_INTEGER)
373 cx.type = obj->integer.value;
375 * Some buggy BIOSes won't list C1 in _CST -
376 * Let acpi_processor_get_power_info_default() handle them later
378 if (i == 1 && cx.type != ACPI_STATE_C1)
381 cx.address = reg->address;
382 cx.index = current_count + 1;
384 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
385 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
386 if (acpi_processor_ffh_cstate_probe
387 (pr->id, &cx, reg) == 0) {
388 cx.entry_method = ACPI_CSTATE_FFH;
389 } else if (cx.type == ACPI_STATE_C1) {
391 * C1 is a special case where FIXED_HARDWARE
392 * can be handled in non-MWAIT way as well.
393 * In that case, save this _CST entry info.
394 * Otherwise, ignore this info and continue.
396 cx.entry_method = ACPI_CSTATE_HALT;
397 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
401 if (cx.type == ACPI_STATE_C1 &&
402 (boot_option_idle_override == IDLE_NOMWAIT)) {
404 * In most cases the C1 space_id obtained from
405 * _CST object is FIXED_HARDWARE access mode.
406 * But when the option of idle=halt is added,
407 * the entry_method type should be changed from
408 * CSTATE_FFH to CSTATE_HALT.
409 * When the option of idle=nomwait is added,
410 * the C1 entry_method type should be
413 cx.entry_method = ACPI_CSTATE_HALT;
414 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
417 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
421 if (cx.type == ACPI_STATE_C1) {
425 obj = &(element->package.elements[2]);
426 if (obj->type != ACPI_TYPE_INTEGER)
429 cx.latency = obj->integer.value;
431 obj = &(element->package.elements[3]);
432 if (obj->type != ACPI_TYPE_INTEGER)
436 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
439 * We support total ACPI_PROCESSOR_MAX_POWER - 1
440 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
442 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
443 pr_warn("Limiting number of power states to max (%d)\n",
444 ACPI_PROCESSOR_MAX_POWER);
445 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
450 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
453 /* Validate number of power states discovered */
454 if (current_count < 2)
458 kfree(buffer.pointer);
463 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
464 struct acpi_processor_cx *cx)
466 static int bm_check_flag = -1;
467 static int bm_control_flag = -1;
474 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
475 * DMA transfers are used by any ISA device to avoid livelock.
476 * Note that we could disable Type-F DMA (as recommended by
477 * the erratum), but this is known to disrupt certain ISA
478 * devices thus we take the conservative approach.
480 else if (errata.piix4.fdma) {
481 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
482 "C3 not supported on PIIX4 with Type-F DMA\n"));
486 /* All the logic here assumes flags.bm_check is same across all CPUs */
487 if (bm_check_flag == -1) {
488 /* Determine whether bm_check is needed based on CPU */
489 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
490 bm_check_flag = pr->flags.bm_check;
491 bm_control_flag = pr->flags.bm_control;
493 pr->flags.bm_check = bm_check_flag;
494 pr->flags.bm_control = bm_control_flag;
497 if (pr->flags.bm_check) {
498 if (!pr->flags.bm_control) {
499 if (pr->flags.has_cst != 1) {
500 /* bus mastering control is necessary */
501 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
502 "C3 support requires BM control\n"));
505 /* Here we enter C3 without bus mastering */
506 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
507 "C3 support without BM control\n"));
512 * WBINVD should be set in fadt, for C3 state to be
513 * supported on when bm_check is not required.
515 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
516 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
517 "Cache invalidation should work properly"
518 " for C3 to be enabled on SMP systems\n"));
524 * Otherwise we've met all of our C3 requirements.
525 * Normalize the C3 latency to expidite policy. Enable
526 * checking of bus mastering status (bm_check) so we can
527 * use this in our C3 policy
532 * On older chipsets, BM_RLD needs to be set
533 * in order for Bus Master activity to wake the
534 * system from C3. Newer chipsets handle DMA
535 * during C3 automatically and BM_RLD is a NOP.
536 * In either case, the proper way to
537 * handle BM_RLD is to set it and leave it set.
539 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
544 static int acpi_processor_power_verify(struct acpi_processor *pr)
547 unsigned int working = 0;
549 pr->power.timer_broadcast_on_state = INT_MAX;
551 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
552 struct acpi_processor_cx *cx = &pr->power.states[i];
566 acpi_processor_power_verify_c3(pr, cx);
572 lapic_timer_check_state(i, pr, cx);
573 tsc_check_state(cx->type);
577 lapic_timer_propagate_broadcast(pr);
582 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
588 /* NOTE: the idle thread may not be running while calling
591 /* Zero initialize all the C-states info. */
592 memset(pr->power.states, 0, sizeof(pr->power.states));
594 result = acpi_processor_get_power_info_cst(pr);
595 if (result == -ENODEV)
596 result = acpi_processor_get_power_info_fadt(pr);
601 acpi_processor_get_power_info_default(pr);
603 pr->power.count = acpi_processor_power_verify(pr);
606 * if one state of type C2 or C3 is available, mark this
607 * CPU as being "idle manageable"
609 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
610 if (pr->power.states[i].valid) {
612 if (pr->power.states[i].type >= ACPI_STATE_C2)
621 * acpi_idle_bm_check - checks if bus master activity was detected
623 static int acpi_idle_bm_check(void)
627 if (bm_check_disable)
630 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
632 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
634 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
635 * the true state of bus mastering activity; forcing us to
636 * manually check the BMIDEA bit of each IDE channel.
638 else if (errata.piix4.bmisx) {
639 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
640 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
647 * acpi_idle_do_entry - enter idle state using the appropriate method
650 * Caller disables interrupt before call and enables interrupt after return.
652 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
654 if (cx->entry_method == ACPI_CSTATE_FFH) {
655 /* Call into architectural FFH based C-state */
656 acpi_processor_ffh_cstate_enter(cx);
657 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
660 /* IO port based C-state */
662 /* Dummy wait op - must do something useless after P_LVL2 read
663 because chipsets cannot guarantee that STPCLK# signal
664 gets asserted in time to freeze execution properly. */
665 inl(acpi_gbl_FADT.xpm_timer_block.address);
670 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
671 * @dev: the target CPU
672 * @index: the index of suggested state
674 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
676 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
678 ACPI_FLUSH_CPU_CACHE();
682 if (cx->entry_method == ACPI_CSTATE_HALT)
684 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
686 /* See comment in acpi_idle_do_entry() */
687 inl(acpi_gbl_FADT.xpm_timer_block.address);
696 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
698 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
699 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
702 static int c3_cpu_count;
703 static DEFINE_RAW_SPINLOCK(c3_lock);
706 * acpi_idle_enter_bm - enters C3 with proper BM handling
707 * @pr: Target processor
708 * @cx: Target state context
709 * @timer_bc: Whether or not to change timer mode to broadcast
711 static void acpi_idle_enter_bm(struct acpi_processor *pr,
712 struct acpi_processor_cx *cx, bool timer_bc)
714 acpi_unlazy_tlb(smp_processor_id());
717 * Must be done before busmaster disable as we might need to
721 lapic_timer_state_broadcast(pr, cx, 1);
725 * bm_check implies we need ARB_DIS
726 * bm_control implies whether we can do ARB_DIS
728 * That leaves a case where bm_check is set and bm_control is
729 * not set. In that case we cannot do much, we enter C3
730 * without doing anything.
732 if (pr->flags.bm_control) {
733 raw_spin_lock(&c3_lock);
735 /* Disable bus master arbitration when all CPUs are in C3 */
736 if (c3_cpu_count == num_online_cpus())
737 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
738 raw_spin_unlock(&c3_lock);
741 acpi_idle_do_entry(cx);
743 /* Re-enable bus master arbitration */
744 if (pr->flags.bm_control) {
745 raw_spin_lock(&c3_lock);
746 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
748 raw_spin_unlock(&c3_lock);
752 lapic_timer_state_broadcast(pr, cx, 0);
755 static int acpi_idle_enter(struct cpuidle_device *dev,
756 struct cpuidle_driver *drv, int index)
758 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
759 struct acpi_processor *pr;
761 pr = __this_cpu_read(processors);
765 if (cx->type != ACPI_STATE_C1) {
766 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
767 index = ACPI_IDLE_STATE_START;
768 cx = per_cpu(acpi_cstate[index], dev->cpu);
769 } else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) {
770 if (cx->bm_sts_skip || !acpi_idle_bm_check()) {
771 acpi_idle_enter_bm(pr, cx, true);
773 } else if (drv->safe_state_index >= 0) {
774 index = drv->safe_state_index;
775 cx = per_cpu(acpi_cstate[index], dev->cpu);
783 lapic_timer_state_broadcast(pr, cx, 1);
785 if (cx->type == ACPI_STATE_C3)
786 ACPI_FLUSH_CPU_CACHE();
788 acpi_idle_do_entry(cx);
790 lapic_timer_state_broadcast(pr, cx, 0);
795 static void acpi_idle_enter_s2idle(struct cpuidle_device *dev,
796 struct cpuidle_driver *drv, int index)
798 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
800 if (cx->type == ACPI_STATE_C3) {
801 struct acpi_processor *pr = __this_cpu_read(processors);
806 if (pr->flags.bm_check) {
807 acpi_idle_enter_bm(pr, cx, false);
810 ACPI_FLUSH_CPU_CACHE();
813 acpi_idle_do_entry(cx);
816 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
817 struct cpuidle_device *dev)
819 int i, count = ACPI_IDLE_STATE_START;
820 struct acpi_processor_cx *cx;
825 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
826 cx = &pr->power.states[i];
831 per_cpu(acpi_cstate[count], dev->cpu) = cx;
834 if (count == CPUIDLE_STATE_MAX)
844 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
847 struct acpi_processor_cx *cx;
848 struct cpuidle_state *state;
849 struct cpuidle_driver *drv = &acpi_idle_driver;
854 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
855 cpuidle_poll_state_init(drv);
861 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
862 cx = &pr->power.states[i];
867 state = &drv->states[count];
868 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
869 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
870 state->exit_latency = cx->latency;
871 state->target_residency = cx->latency * latency_factor;
872 state->enter = acpi_idle_enter;
875 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) {
876 state->enter_dead = acpi_idle_play_dead;
877 drv->safe_state_index = count;
880 * Halt-induced C1 is not good for ->enter_s2idle, because it
881 * re-enables interrupts on exit. Moreover, C1 is generally not
882 * particularly interesting from the suspend-to-idle angle, so
883 * avoid C1 and the situations in which we may need to fall back
886 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
887 state->enter_s2idle = acpi_idle_enter_s2idle;
890 if (count == CPUIDLE_STATE_MAX)
894 drv->state_count = count;
902 static inline void acpi_processor_cstate_first_run_checks(void)
905 static int first_run;
909 dmi_check_system(processor_power_dmi_table);
910 max_cstate = acpi_processor_cstate_check(max_cstate);
911 if (max_cstate < ACPI_C_STATES_MAX)
912 pr_notice("ACPI: processor limited to max C-state %d\n",
916 if (acpi_gbl_FADT.cst_control && !nocst) {
917 status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
918 acpi_gbl_FADT.cst_control, 8);
919 if (ACPI_FAILURE(status))
920 ACPI_EXCEPTION((AE_INFO, status,
921 "Notifying BIOS of _CST ability failed"));
926 static inline int disabled_by_idle_boot_param(void) { return 0; }
927 static inline void acpi_processor_cstate_first_run_checks(void) { }
928 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
933 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
934 struct cpuidle_device *dev)
939 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
944 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
946 struct acpi_lpi_states_array {
948 unsigned int composite_states_size;
949 struct acpi_lpi_state *entries;
950 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
953 static int obj_get_integer(union acpi_object *obj, u32 *value)
955 if (obj->type != ACPI_TYPE_INTEGER)
958 *value = obj->integer.value;
962 static int acpi_processor_evaluate_lpi(acpi_handle handle,
963 struct acpi_lpi_states_array *info)
967 int pkg_count, state_idx = 1, loop;
968 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
969 union acpi_object *lpi_data;
970 struct acpi_lpi_state *lpi_state;
972 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
973 if (ACPI_FAILURE(status)) {
974 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n"));
978 lpi_data = buffer.pointer;
980 /* There must be at least 4 elements = 3 elements + 1 package */
981 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
982 lpi_data->package.count < 4) {
983 pr_debug("not enough elements in _LPI\n");
988 pkg_count = lpi_data->package.elements[2].integer.value;
990 /* Validate number of power states. */
991 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
992 pr_debug("count given by _LPI is not valid\n");
997 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
1003 info->size = pkg_count;
1004 info->entries = lpi_state;
1006 /* LPI States start at index 3 */
1007 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
1008 union acpi_object *element, *pkg_elem, *obj;
1010 element = &lpi_data->package.elements[loop];
1011 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
1014 pkg_elem = element->package.elements;
1017 if (obj->type == ACPI_TYPE_BUFFER) {
1018 struct acpi_power_register *reg;
1020 reg = (struct acpi_power_register *)obj->buffer.pointer;
1021 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
1022 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
1025 lpi_state->address = reg->address;
1026 lpi_state->entry_method =
1027 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
1028 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
1029 } else if (obj->type == ACPI_TYPE_INTEGER) {
1030 lpi_state->entry_method = ACPI_CSTATE_INTEGER;
1031 lpi_state->address = obj->integer.value;
1036 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
1039 if (obj->type == ACPI_TYPE_STRING)
1040 strlcpy(lpi_state->desc, obj->string.pointer,
1043 lpi_state->index = state_idx;
1044 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
1045 pr_debug("No min. residency found, assuming 10 us\n");
1046 lpi_state->min_residency = 10;
1049 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
1050 pr_debug("No wakeup residency found, assuming 10 us\n");
1051 lpi_state->wake_latency = 10;
1054 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
1055 lpi_state->flags = 0;
1057 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
1058 lpi_state->arch_flags = 0;
1060 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
1061 lpi_state->res_cnt_freq = 1;
1063 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
1064 lpi_state->enable_parent_state = 0;
1067 acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1069 kfree(buffer.pointer);
1074 * flat_state_cnt - the number of composite LPI states after the process of flattening
1076 static int flat_state_cnt;
1079 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1081 * @local: local LPI state
1082 * @parent: parent LPI state
1083 * @result: composite LPI state
1085 static bool combine_lpi_states(struct acpi_lpi_state *local,
1086 struct acpi_lpi_state *parent,
1087 struct acpi_lpi_state *result)
1089 if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1090 if (!parent->address) /* 0 means autopromotable */
1092 result->address = local->address + parent->address;
1094 result->address = parent->address;
1097 result->min_residency = max(local->min_residency, parent->min_residency);
1098 result->wake_latency = local->wake_latency + parent->wake_latency;
1099 result->enable_parent_state = parent->enable_parent_state;
1100 result->entry_method = local->entry_method;
1102 result->flags = parent->flags;
1103 result->arch_flags = parent->arch_flags;
1104 result->index = parent->index;
1106 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1107 strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1108 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1112 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0)
1114 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1115 struct acpi_lpi_state *t)
1117 curr_level->composite_states[curr_level->composite_states_size++] = t;
1120 static int flatten_lpi_states(struct acpi_processor *pr,
1121 struct acpi_lpi_states_array *curr_level,
1122 struct acpi_lpi_states_array *prev_level)
1124 int i, j, state_count = curr_level->size;
1125 struct acpi_lpi_state *p, *t = curr_level->entries;
1127 curr_level->composite_states_size = 0;
1128 for (j = 0; j < state_count; j++, t++) {
1129 struct acpi_lpi_state *flpi;
1131 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1134 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1135 pr_warn("Limiting number of LPI states to max (%d)\n",
1136 ACPI_PROCESSOR_MAX_POWER);
1137 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1141 flpi = &pr->power.lpi_states[flat_state_cnt];
1143 if (!prev_level) { /* leaf/processor node */
1144 memcpy(flpi, t, sizeof(*t));
1145 stash_composite_state(curr_level, flpi);
1150 for (i = 0; i < prev_level->composite_states_size; i++) {
1151 p = prev_level->composite_states[i];
1152 if (t->index <= p->enable_parent_state &&
1153 combine_lpi_states(p, t, flpi)) {
1154 stash_composite_state(curr_level, flpi);
1161 kfree(curr_level->entries);
1165 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1169 acpi_handle handle = pr->handle, pr_ahandle;
1170 struct acpi_device *d = NULL;
1171 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1173 if (!osc_pc_lpi_support_confirmed)
1176 if (!acpi_has_method(handle, "_LPI"))
1182 handle = pr->handle;
1183 ret = acpi_processor_evaluate_lpi(handle, prev);
1186 flatten_lpi_states(pr, prev, NULL);
1188 status = acpi_get_parent(handle, &pr_ahandle);
1189 while (ACPI_SUCCESS(status)) {
1190 acpi_bus_get_device(pr_ahandle, &d);
1191 handle = pr_ahandle;
1193 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1196 /* can be optional ? */
1197 if (!acpi_has_method(handle, "_LPI"))
1200 ret = acpi_processor_evaluate_lpi(handle, curr);
1204 /* flatten all the LPI states in this level of hierarchy */
1205 flatten_lpi_states(pr, curr, prev);
1207 tmp = prev, prev = curr, curr = tmp;
1209 status = acpi_get_parent(handle, &pr_ahandle);
1212 pr->power.count = flat_state_cnt;
1213 /* reset the index after flattening */
1214 for (i = 0; i < pr->power.count; i++)
1215 pr->power.lpi_states[i].index = i;
1217 /* Tell driver that _LPI is supported. */
1218 pr->flags.has_lpi = 1;
1219 pr->flags.power = 1;
1224 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1229 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1235 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1236 * @dev: the target CPU
1237 * @drv: cpuidle driver containing cpuidle state info
1238 * @index: index of target state
1240 * Return: 0 for success or negative value for error
1242 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1243 struct cpuidle_driver *drv, int index)
1245 struct acpi_processor *pr;
1246 struct acpi_lpi_state *lpi;
1248 pr = __this_cpu_read(processors);
1253 lpi = &pr->power.lpi_states[index];
1254 if (lpi->entry_method == ACPI_CSTATE_FFH)
1255 return acpi_processor_ffh_lpi_enter(lpi);
1260 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1263 struct acpi_lpi_state *lpi;
1264 struct cpuidle_state *state;
1265 struct cpuidle_driver *drv = &acpi_idle_driver;
1267 if (!pr->flags.has_lpi)
1270 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1271 lpi = &pr->power.lpi_states[i];
1273 state = &drv->states[i];
1274 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1275 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1276 state->exit_latency = lpi->wake_latency;
1277 state->target_residency = lpi->min_residency;
1278 if (lpi->arch_flags)
1279 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1280 state->enter = acpi_idle_lpi_enter;
1281 drv->safe_state_index = i;
1284 drv->state_count = i;
1290 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1291 * global state data i.e. idle routines
1293 * @pr: the ACPI processor
1295 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1298 struct cpuidle_driver *drv = &acpi_idle_driver;
1300 if (!pr->flags.power_setup_done || !pr->flags.power)
1303 drv->safe_state_index = -1;
1304 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1305 drv->states[i].name[0] = '\0';
1306 drv->states[i].desc[0] = '\0';
1309 if (pr->flags.has_lpi)
1310 return acpi_processor_setup_lpi_states(pr);
1312 return acpi_processor_setup_cstates(pr);
1316 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1317 * device i.e. per-cpu data
1319 * @pr: the ACPI processor
1320 * @dev : the cpuidle device
1322 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1323 struct cpuidle_device *dev)
1325 if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1329 if (pr->flags.has_lpi)
1330 return acpi_processor_ffh_lpi_probe(pr->id);
1332 return acpi_processor_setup_cpuidle_cx(pr, dev);
1335 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1339 ret = acpi_processor_get_lpi_info(pr);
1341 ret = acpi_processor_get_cstate_info(pr);
1346 int acpi_processor_hotplug(struct acpi_processor *pr)
1349 struct cpuidle_device *dev;
1351 if (disabled_by_idle_boot_param())
1354 if (!pr->flags.power_setup_done)
1357 dev = per_cpu(acpi_cpuidle_device, pr->id);
1358 cpuidle_pause_and_lock();
1359 cpuidle_disable_device(dev);
1360 ret = acpi_processor_get_power_info(pr);
1361 if (!ret && pr->flags.power) {
1362 acpi_processor_setup_cpuidle_dev(pr, dev);
1363 ret = cpuidle_enable_device(dev);
1365 cpuidle_resume_and_unlock();
1370 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1373 struct acpi_processor *_pr;
1374 struct cpuidle_device *dev;
1376 if (disabled_by_idle_boot_param())
1379 if (!pr->flags.power_setup_done)
1383 * FIXME: Design the ACPI notification to make it once per
1384 * system instead of once per-cpu. This condition is a hack
1385 * to make the code that updates C-States be called once.
1388 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1390 /* Protect against cpu-hotplug */
1392 cpuidle_pause_and_lock();
1394 /* Disable all cpuidle devices */
1395 for_each_online_cpu(cpu) {
1396 _pr = per_cpu(processors, cpu);
1397 if (!_pr || !_pr->flags.power_setup_done)
1399 dev = per_cpu(acpi_cpuidle_device, cpu);
1400 cpuidle_disable_device(dev);
1403 /* Populate Updated C-state information */
1404 acpi_processor_get_power_info(pr);
1405 acpi_processor_setup_cpuidle_states(pr);
1407 /* Enable all cpuidle devices */
1408 for_each_online_cpu(cpu) {
1409 _pr = per_cpu(processors, cpu);
1410 if (!_pr || !_pr->flags.power_setup_done)
1412 acpi_processor_get_power_info(_pr);
1413 if (_pr->flags.power) {
1414 dev = per_cpu(acpi_cpuidle_device, cpu);
1415 acpi_processor_setup_cpuidle_dev(_pr, dev);
1416 cpuidle_enable_device(dev);
1419 cpuidle_resume_and_unlock();
1426 static int acpi_processor_registered;
1428 int acpi_processor_power_init(struct acpi_processor *pr)
1431 struct cpuidle_device *dev;
1433 if (disabled_by_idle_boot_param())
1436 acpi_processor_cstate_first_run_checks();
1438 if (!acpi_processor_get_power_info(pr))
1439 pr->flags.power_setup_done = 1;
1442 * Install the idle handler if processor power management is supported.
1443 * Note that we use previously set idle handler will be used on
1444 * platforms that only support C1.
1446 if (pr->flags.power) {
1447 /* Register acpi_idle_driver if not already registered */
1448 if (!acpi_processor_registered) {
1449 acpi_processor_setup_cpuidle_states(pr);
1450 retval = cpuidle_register_driver(&acpi_idle_driver);
1453 pr_debug("%s registered with cpuidle\n",
1454 acpi_idle_driver.name);
1457 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1460 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1462 acpi_processor_setup_cpuidle_dev(pr, dev);
1464 /* Register per-cpu cpuidle_device. Cpuidle driver
1465 * must already be registered before registering device
1467 retval = cpuidle_register_device(dev);
1469 if (acpi_processor_registered == 0)
1470 cpuidle_unregister_driver(&acpi_idle_driver);
1473 acpi_processor_registered++;
1478 int acpi_processor_power_exit(struct acpi_processor *pr)
1480 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1482 if (disabled_by_idle_boot_param())
1485 if (pr->flags.power) {
1486 cpuidle_unregister_device(dev);
1487 acpi_processor_registered--;
1488 if (acpi_processor_registered == 0)
1489 cpuidle_unregister_driver(&acpi_idle_driver);
1492 pr->flags.power_setup_done = 0;