2 * processor_idle - idle state submodule to the ACPI processor driver
8 * - Added processor hotplug support
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) {
208 case X86_VENDOR_HYGON:
210 case X86_VENDOR_INTEL:
211 case X86_VENDOR_CENTAUR:
213 * AMD Fam10h TSC will tick in all
214 * C/P/S0/S1 states when this bit is set.
216 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
221 /* TSC could halt in idle, so notify users */
222 if (state > ACPI_STATE_C1)
223 mark_tsc_unstable("TSC halts in idle");
227 static void tsc_check_state(int state) { return; }
230 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
236 /* if info is obtained from pblk/fadt, type equals state */
237 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
238 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
240 #ifndef CONFIG_HOTPLUG_CPU
242 * Check for P_LVL2_UP flag before entering C2 and above on
245 if ((num_online_cpus() > 1) &&
246 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
250 /* determine C2 and C3 address from pblk */
251 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
252 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
254 /* determine latencies from FADT */
255 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
256 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
259 * FADT specified C2 latency must be less than or equal to
262 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
263 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
264 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
266 pr->power.states[ACPI_STATE_C2].address = 0;
270 * FADT supplied C3 latency must be less than or equal to
273 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
274 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
275 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
277 pr->power.states[ACPI_STATE_C3].address = 0;
280 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
281 "lvl2[0x%08x] lvl3[0x%08x]\n",
282 pr->power.states[ACPI_STATE_C2].address,
283 pr->power.states[ACPI_STATE_C3].address));
288 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
290 if (!pr->power.states[ACPI_STATE_C1].valid) {
291 /* set the first C-State to C1 */
292 /* all processors need to support C1 */
293 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
294 pr->power.states[ACPI_STATE_C1].valid = 1;
295 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
297 snprintf(pr->power.states[ACPI_STATE_C1].desc,
298 ACPI_CX_DESC_LEN, "ACPI HLT");
300 /* the C0 state only exists as a filler in our array */
301 pr->power.states[ACPI_STATE_C0].valid = 1;
305 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
311 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
312 union acpi_object *cst;
319 status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
320 if (ACPI_FAILURE(status)) {
321 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
325 cst = buffer.pointer;
327 /* There must be at least 2 elements */
328 if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
329 pr_err("not enough elements in _CST\n");
334 count = cst->package.elements[0].integer.value;
336 /* Validate number of power states. */
337 if (count < 1 || count != cst->package.count - 1) {
338 pr_err("count given by _CST is not valid\n");
343 /* Tell driver that at least _CST is supported. */
344 pr->flags.has_cst = 1;
346 for (i = 1; i <= count; i++) {
347 union acpi_object *element;
348 union acpi_object *obj;
349 struct acpi_power_register *reg;
350 struct acpi_processor_cx cx;
352 memset(&cx, 0, sizeof(cx));
354 element = &(cst->package.elements[i]);
355 if (element->type != ACPI_TYPE_PACKAGE)
358 if (element->package.count != 4)
361 obj = &(element->package.elements[0]);
363 if (obj->type != ACPI_TYPE_BUFFER)
366 reg = (struct acpi_power_register *)obj->buffer.pointer;
368 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
369 (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
372 /* There should be an easy way to extract an integer... */
373 obj = &(element->package.elements[1]);
374 if (obj->type != ACPI_TYPE_INTEGER)
377 cx.type = obj->integer.value;
379 * Some buggy BIOSes won't list C1 in _CST -
380 * Let acpi_processor_get_power_info_default() handle them later
382 if (i == 1 && cx.type != ACPI_STATE_C1)
385 cx.address = reg->address;
386 cx.index = current_count + 1;
388 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
389 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
390 if (acpi_processor_ffh_cstate_probe
391 (pr->id, &cx, reg) == 0) {
392 cx.entry_method = ACPI_CSTATE_FFH;
393 } else if (cx.type == ACPI_STATE_C1) {
395 * C1 is a special case where FIXED_HARDWARE
396 * can be handled in non-MWAIT way as well.
397 * In that case, save this _CST entry info.
398 * Otherwise, ignore this info and continue.
400 cx.entry_method = ACPI_CSTATE_HALT;
401 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
405 if (cx.type == ACPI_STATE_C1 &&
406 (boot_option_idle_override == IDLE_NOMWAIT)) {
408 * In most cases the C1 space_id obtained from
409 * _CST object is FIXED_HARDWARE access mode.
410 * But when the option of idle=halt is added,
411 * the entry_method type should be changed from
412 * CSTATE_FFH to CSTATE_HALT.
413 * When the option of idle=nomwait is added,
414 * the C1 entry_method type should be
417 cx.entry_method = ACPI_CSTATE_HALT;
418 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
421 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
425 if (cx.type == ACPI_STATE_C1) {
429 obj = &(element->package.elements[2]);
430 if (obj->type != ACPI_TYPE_INTEGER)
433 cx.latency = obj->integer.value;
435 obj = &(element->package.elements[3]);
436 if (obj->type != ACPI_TYPE_INTEGER)
440 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
443 * We support total ACPI_PROCESSOR_MAX_POWER - 1
444 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
446 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
447 pr_warn("Limiting number of power states to max (%d)\n",
448 ACPI_PROCESSOR_MAX_POWER);
449 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
454 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
457 /* Validate number of power states discovered */
458 if (current_count < 2)
462 kfree(buffer.pointer);
467 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
468 struct acpi_processor_cx *cx)
470 static int bm_check_flag = -1;
471 static int bm_control_flag = -1;
478 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
479 * DMA transfers are used by any ISA device to avoid livelock.
480 * Note that we could disable Type-F DMA (as recommended by
481 * the erratum), but this is known to disrupt certain ISA
482 * devices thus we take the conservative approach.
484 else if (errata.piix4.fdma) {
485 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
486 "C3 not supported on PIIX4 with Type-F DMA\n"));
490 /* All the logic here assumes flags.bm_check is same across all CPUs */
491 if (bm_check_flag == -1) {
492 /* Determine whether bm_check is needed based on CPU */
493 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
494 bm_check_flag = pr->flags.bm_check;
495 bm_control_flag = pr->flags.bm_control;
497 pr->flags.bm_check = bm_check_flag;
498 pr->flags.bm_control = bm_control_flag;
501 if (pr->flags.bm_check) {
502 if (!pr->flags.bm_control) {
503 if (pr->flags.has_cst != 1) {
504 /* bus mastering control is necessary */
505 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
506 "C3 support requires BM control\n"));
509 /* Here we enter C3 without bus mastering */
510 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
511 "C3 support without BM control\n"));
516 * WBINVD should be set in fadt, for C3 state to be
517 * supported on when bm_check is not required.
519 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
520 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
521 "Cache invalidation should work properly"
522 " for C3 to be enabled on SMP systems\n"));
528 * Otherwise we've met all of our C3 requirements.
529 * Normalize the C3 latency to expidite policy. Enable
530 * checking of bus mastering status (bm_check) so we can
531 * use this in our C3 policy
536 * On older chipsets, BM_RLD needs to be set
537 * in order for Bus Master activity to wake the
538 * system from C3. Newer chipsets handle DMA
539 * during C3 automatically and BM_RLD is a NOP.
540 * In either case, the proper way to
541 * handle BM_RLD is to set it and leave it set.
543 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
548 static int acpi_processor_power_verify(struct acpi_processor *pr)
551 unsigned int working = 0;
553 pr->power.timer_broadcast_on_state = INT_MAX;
555 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
556 struct acpi_processor_cx *cx = &pr->power.states[i];
570 acpi_processor_power_verify_c3(pr, cx);
576 lapic_timer_check_state(i, pr, cx);
577 tsc_check_state(cx->type);
581 lapic_timer_propagate_broadcast(pr);
586 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
592 /* NOTE: the idle thread may not be running while calling
595 /* Zero initialize all the C-states info. */
596 memset(pr->power.states, 0, sizeof(pr->power.states));
598 result = acpi_processor_get_power_info_cst(pr);
599 if (result == -ENODEV)
600 result = acpi_processor_get_power_info_fadt(pr);
605 acpi_processor_get_power_info_default(pr);
607 pr->power.count = acpi_processor_power_verify(pr);
610 * if one state of type C2 or C3 is available, mark this
611 * CPU as being "idle manageable"
613 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
614 if (pr->power.states[i].valid) {
616 if (pr->power.states[i].type >= ACPI_STATE_C2)
625 * acpi_idle_bm_check - checks if bus master activity was detected
627 static int acpi_idle_bm_check(void)
631 if (bm_check_disable)
634 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
636 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
638 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
639 * the true state of bus mastering activity; forcing us to
640 * manually check the BMIDEA bit of each IDE channel.
642 else if (errata.piix4.bmisx) {
643 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
644 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
651 * acpi_idle_do_entry - enter idle state using the appropriate method
654 * Caller disables interrupt before call and enables interrupt after return.
656 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
658 if (cx->entry_method == ACPI_CSTATE_FFH) {
659 /* Call into architectural FFH based C-state */
660 acpi_processor_ffh_cstate_enter(cx);
661 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
664 /* IO port based C-state */
666 /* Dummy wait op - must do something useless after P_LVL2 read
667 because chipsets cannot guarantee that STPCLK# signal
668 gets asserted in time to freeze execution properly. */
669 inl(acpi_gbl_FADT.xpm_timer_block.address);
674 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
675 * @dev: the target CPU
676 * @index: the index of suggested state
678 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
680 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
682 ACPI_FLUSH_CPU_CACHE();
686 if (cx->entry_method == ACPI_CSTATE_HALT)
688 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
690 /* See comment in acpi_idle_do_entry() */
691 inl(acpi_gbl_FADT.xpm_timer_block.address);
700 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
702 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
703 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
706 static int c3_cpu_count;
707 static DEFINE_RAW_SPINLOCK(c3_lock);
710 * acpi_idle_enter_bm - enters C3 with proper BM handling
711 * @pr: Target processor
712 * @cx: Target state context
713 * @timer_bc: Whether or not to change timer mode to broadcast
715 static void acpi_idle_enter_bm(struct acpi_processor *pr,
716 struct acpi_processor_cx *cx, bool timer_bc)
718 acpi_unlazy_tlb(smp_processor_id());
721 * Must be done before busmaster disable as we might need to
725 lapic_timer_state_broadcast(pr, cx, 1);
729 * bm_check implies we need ARB_DIS
730 * bm_control implies whether we can do ARB_DIS
732 * That leaves a case where bm_check is set and bm_control is
733 * not set. In that case we cannot do much, we enter C3
734 * without doing anything.
736 if (pr->flags.bm_control) {
737 raw_spin_lock(&c3_lock);
739 /* Disable bus master arbitration when all CPUs are in C3 */
740 if (c3_cpu_count == num_online_cpus())
741 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
742 raw_spin_unlock(&c3_lock);
745 acpi_idle_do_entry(cx);
747 /* Re-enable bus master arbitration */
748 if (pr->flags.bm_control) {
749 raw_spin_lock(&c3_lock);
750 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
752 raw_spin_unlock(&c3_lock);
756 lapic_timer_state_broadcast(pr, cx, 0);
759 static int acpi_idle_enter(struct cpuidle_device *dev,
760 struct cpuidle_driver *drv, int index)
762 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
763 struct acpi_processor *pr;
765 pr = __this_cpu_read(processors);
769 if (cx->type != ACPI_STATE_C1) {
770 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
771 index = ACPI_IDLE_STATE_START;
772 cx = per_cpu(acpi_cstate[index], dev->cpu);
773 } else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) {
774 if (cx->bm_sts_skip || !acpi_idle_bm_check()) {
775 acpi_idle_enter_bm(pr, cx, true);
777 } else if (drv->safe_state_index >= 0) {
778 index = drv->safe_state_index;
779 cx = per_cpu(acpi_cstate[index], dev->cpu);
787 lapic_timer_state_broadcast(pr, cx, 1);
789 if (cx->type == ACPI_STATE_C3)
790 ACPI_FLUSH_CPU_CACHE();
792 acpi_idle_do_entry(cx);
794 lapic_timer_state_broadcast(pr, cx, 0);
799 static void acpi_idle_enter_s2idle(struct cpuidle_device *dev,
800 struct cpuidle_driver *drv, int index)
802 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
804 if (cx->type == ACPI_STATE_C3) {
805 struct acpi_processor *pr = __this_cpu_read(processors);
810 if (pr->flags.bm_check) {
811 acpi_idle_enter_bm(pr, cx, false);
814 ACPI_FLUSH_CPU_CACHE();
817 acpi_idle_do_entry(cx);
820 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
821 struct cpuidle_device *dev)
823 int i, count = ACPI_IDLE_STATE_START;
824 struct acpi_processor_cx *cx;
829 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
830 cx = &pr->power.states[i];
835 per_cpu(acpi_cstate[count], dev->cpu) = cx;
838 if (count == CPUIDLE_STATE_MAX)
848 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
851 struct acpi_processor_cx *cx;
852 struct cpuidle_state *state;
853 struct cpuidle_driver *drv = &acpi_idle_driver;
858 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
859 cpuidle_poll_state_init(drv);
865 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
866 cx = &pr->power.states[i];
871 state = &drv->states[count];
872 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
873 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
874 state->exit_latency = cx->latency;
875 state->target_residency = cx->latency * latency_factor;
876 state->enter = acpi_idle_enter;
879 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) {
880 state->enter_dead = acpi_idle_play_dead;
881 drv->safe_state_index = count;
884 * Halt-induced C1 is not good for ->enter_s2idle, because it
885 * re-enables interrupts on exit. Moreover, C1 is generally not
886 * particularly interesting from the suspend-to-idle angle, so
887 * avoid C1 and the situations in which we may need to fall back
890 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
891 state->enter_s2idle = acpi_idle_enter_s2idle;
894 if (count == CPUIDLE_STATE_MAX)
898 drv->state_count = count;
906 static inline void acpi_processor_cstate_first_run_checks(void)
909 static int first_run;
913 dmi_check_system(processor_power_dmi_table);
914 max_cstate = acpi_processor_cstate_check(max_cstate);
915 if (max_cstate < ACPI_C_STATES_MAX)
916 pr_notice("ACPI: processor limited to max C-state %d\n",
920 if (acpi_gbl_FADT.cst_control && !nocst) {
921 status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
922 acpi_gbl_FADT.cst_control, 8);
923 if (ACPI_FAILURE(status))
924 ACPI_EXCEPTION((AE_INFO, status,
925 "Notifying BIOS of _CST ability failed"));
930 static inline int disabled_by_idle_boot_param(void) { return 0; }
931 static inline void acpi_processor_cstate_first_run_checks(void) { }
932 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
937 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
938 struct cpuidle_device *dev)
943 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
948 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
950 struct acpi_lpi_states_array {
952 unsigned int composite_states_size;
953 struct acpi_lpi_state *entries;
954 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
957 static int obj_get_integer(union acpi_object *obj, u32 *value)
959 if (obj->type != ACPI_TYPE_INTEGER)
962 *value = obj->integer.value;
966 static int acpi_processor_evaluate_lpi(acpi_handle handle,
967 struct acpi_lpi_states_array *info)
971 int pkg_count, state_idx = 1, loop;
972 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
973 union acpi_object *lpi_data;
974 struct acpi_lpi_state *lpi_state;
976 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
977 if (ACPI_FAILURE(status)) {
978 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n"));
982 lpi_data = buffer.pointer;
984 /* There must be at least 4 elements = 3 elements + 1 package */
985 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
986 lpi_data->package.count < 4) {
987 pr_debug("not enough elements in _LPI\n");
992 pkg_count = lpi_data->package.elements[2].integer.value;
994 /* Validate number of power states. */
995 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
996 pr_debug("count given by _LPI is not valid\n");
1001 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
1007 info->size = pkg_count;
1008 info->entries = lpi_state;
1010 /* LPI States start at index 3 */
1011 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
1012 union acpi_object *element, *pkg_elem, *obj;
1014 element = &lpi_data->package.elements[loop];
1015 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
1018 pkg_elem = element->package.elements;
1021 if (obj->type == ACPI_TYPE_BUFFER) {
1022 struct acpi_power_register *reg;
1024 reg = (struct acpi_power_register *)obj->buffer.pointer;
1025 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
1026 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
1029 lpi_state->address = reg->address;
1030 lpi_state->entry_method =
1031 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
1032 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
1033 } else if (obj->type == ACPI_TYPE_INTEGER) {
1034 lpi_state->entry_method = ACPI_CSTATE_INTEGER;
1035 lpi_state->address = obj->integer.value;
1040 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
1043 if (obj->type == ACPI_TYPE_STRING)
1044 strlcpy(lpi_state->desc, obj->string.pointer,
1047 lpi_state->index = state_idx;
1048 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
1049 pr_debug("No min. residency found, assuming 10 us\n");
1050 lpi_state->min_residency = 10;
1053 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
1054 pr_debug("No wakeup residency found, assuming 10 us\n");
1055 lpi_state->wake_latency = 10;
1058 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
1059 lpi_state->flags = 0;
1061 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
1062 lpi_state->arch_flags = 0;
1064 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
1065 lpi_state->res_cnt_freq = 1;
1067 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
1068 lpi_state->enable_parent_state = 0;
1071 acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1073 kfree(buffer.pointer);
1078 * flat_state_cnt - the number of composite LPI states after the process of flattening
1080 static int flat_state_cnt;
1083 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1085 * @local: local LPI state
1086 * @parent: parent LPI state
1087 * @result: composite LPI state
1089 static bool combine_lpi_states(struct acpi_lpi_state *local,
1090 struct acpi_lpi_state *parent,
1091 struct acpi_lpi_state *result)
1093 if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1094 if (!parent->address) /* 0 means autopromotable */
1096 result->address = local->address + parent->address;
1098 result->address = parent->address;
1101 result->min_residency = max(local->min_residency, parent->min_residency);
1102 result->wake_latency = local->wake_latency + parent->wake_latency;
1103 result->enable_parent_state = parent->enable_parent_state;
1104 result->entry_method = local->entry_method;
1106 result->flags = parent->flags;
1107 result->arch_flags = parent->arch_flags;
1108 result->index = parent->index;
1110 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1111 strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1112 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1116 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0)
1118 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1119 struct acpi_lpi_state *t)
1121 curr_level->composite_states[curr_level->composite_states_size++] = t;
1124 static int flatten_lpi_states(struct acpi_processor *pr,
1125 struct acpi_lpi_states_array *curr_level,
1126 struct acpi_lpi_states_array *prev_level)
1128 int i, j, state_count = curr_level->size;
1129 struct acpi_lpi_state *p, *t = curr_level->entries;
1131 curr_level->composite_states_size = 0;
1132 for (j = 0; j < state_count; j++, t++) {
1133 struct acpi_lpi_state *flpi;
1135 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1138 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1139 pr_warn("Limiting number of LPI states to max (%d)\n",
1140 ACPI_PROCESSOR_MAX_POWER);
1141 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1145 flpi = &pr->power.lpi_states[flat_state_cnt];
1147 if (!prev_level) { /* leaf/processor node */
1148 memcpy(flpi, t, sizeof(*t));
1149 stash_composite_state(curr_level, flpi);
1154 for (i = 0; i < prev_level->composite_states_size; i++) {
1155 p = prev_level->composite_states[i];
1156 if (t->index <= p->enable_parent_state &&
1157 combine_lpi_states(p, t, flpi)) {
1158 stash_composite_state(curr_level, flpi);
1165 kfree(curr_level->entries);
1169 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1173 acpi_handle handle = pr->handle, pr_ahandle;
1174 struct acpi_device *d = NULL;
1175 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1177 if (!osc_pc_lpi_support_confirmed)
1180 if (!acpi_has_method(handle, "_LPI"))
1186 handle = pr->handle;
1187 ret = acpi_processor_evaluate_lpi(handle, prev);
1190 flatten_lpi_states(pr, prev, NULL);
1192 status = acpi_get_parent(handle, &pr_ahandle);
1193 while (ACPI_SUCCESS(status)) {
1194 acpi_bus_get_device(pr_ahandle, &d);
1195 handle = pr_ahandle;
1197 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1200 /* can be optional ? */
1201 if (!acpi_has_method(handle, "_LPI"))
1204 ret = acpi_processor_evaluate_lpi(handle, curr);
1208 /* flatten all the LPI states in this level of hierarchy */
1209 flatten_lpi_states(pr, curr, prev);
1211 tmp = prev, prev = curr, curr = tmp;
1213 status = acpi_get_parent(handle, &pr_ahandle);
1216 pr->power.count = flat_state_cnt;
1217 /* reset the index after flattening */
1218 for (i = 0; i < pr->power.count; i++)
1219 pr->power.lpi_states[i].index = i;
1221 /* Tell driver that _LPI is supported. */
1222 pr->flags.has_lpi = 1;
1223 pr->flags.power = 1;
1228 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1233 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1239 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1240 * @dev: the target CPU
1241 * @drv: cpuidle driver containing cpuidle state info
1242 * @index: index of target state
1244 * Return: 0 for success or negative value for error
1246 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1247 struct cpuidle_driver *drv, int index)
1249 struct acpi_processor *pr;
1250 struct acpi_lpi_state *lpi;
1252 pr = __this_cpu_read(processors);
1257 lpi = &pr->power.lpi_states[index];
1258 if (lpi->entry_method == ACPI_CSTATE_FFH)
1259 return acpi_processor_ffh_lpi_enter(lpi);
1264 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1267 struct acpi_lpi_state *lpi;
1268 struct cpuidle_state *state;
1269 struct cpuidle_driver *drv = &acpi_idle_driver;
1271 if (!pr->flags.has_lpi)
1274 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1275 lpi = &pr->power.lpi_states[i];
1277 state = &drv->states[i];
1278 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1279 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1280 state->exit_latency = lpi->wake_latency;
1281 state->target_residency = lpi->min_residency;
1282 if (lpi->arch_flags)
1283 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1284 state->enter = acpi_idle_lpi_enter;
1285 drv->safe_state_index = i;
1288 drv->state_count = i;
1294 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1295 * global state data i.e. idle routines
1297 * @pr: the ACPI processor
1299 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1302 struct cpuidle_driver *drv = &acpi_idle_driver;
1304 if (!pr->flags.power_setup_done || !pr->flags.power)
1307 drv->safe_state_index = -1;
1308 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1309 drv->states[i].name[0] = '\0';
1310 drv->states[i].desc[0] = '\0';
1313 if (pr->flags.has_lpi)
1314 return acpi_processor_setup_lpi_states(pr);
1316 return acpi_processor_setup_cstates(pr);
1320 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1321 * device i.e. per-cpu data
1323 * @pr: the ACPI processor
1324 * @dev : the cpuidle device
1326 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1327 struct cpuidle_device *dev)
1329 if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1333 if (pr->flags.has_lpi)
1334 return acpi_processor_ffh_lpi_probe(pr->id);
1336 return acpi_processor_setup_cpuidle_cx(pr, dev);
1339 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1343 ret = acpi_processor_get_lpi_info(pr);
1345 ret = acpi_processor_get_cstate_info(pr);
1350 int acpi_processor_hotplug(struct acpi_processor *pr)
1353 struct cpuidle_device *dev;
1355 if (disabled_by_idle_boot_param())
1358 if (!pr->flags.power_setup_done)
1361 dev = per_cpu(acpi_cpuidle_device, pr->id);
1362 cpuidle_pause_and_lock();
1363 cpuidle_disable_device(dev);
1364 ret = acpi_processor_get_power_info(pr);
1365 if (!ret && pr->flags.power) {
1366 acpi_processor_setup_cpuidle_dev(pr, dev);
1367 ret = cpuidle_enable_device(dev);
1369 cpuidle_resume_and_unlock();
1374 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1377 struct acpi_processor *_pr;
1378 struct cpuidle_device *dev;
1380 if (disabled_by_idle_boot_param())
1383 if (!pr->flags.power_setup_done)
1387 * FIXME: Design the ACPI notification to make it once per
1388 * system instead of once per-cpu. This condition is a hack
1389 * to make the code that updates C-States be called once.
1392 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1394 /* Protect against cpu-hotplug */
1396 cpuidle_pause_and_lock();
1398 /* Disable all cpuidle devices */
1399 for_each_online_cpu(cpu) {
1400 _pr = per_cpu(processors, cpu);
1401 if (!_pr || !_pr->flags.power_setup_done)
1403 dev = per_cpu(acpi_cpuidle_device, cpu);
1404 cpuidle_disable_device(dev);
1407 /* Populate Updated C-state information */
1408 acpi_processor_get_power_info(pr);
1409 acpi_processor_setup_cpuidle_states(pr);
1411 /* Enable all cpuidle devices */
1412 for_each_online_cpu(cpu) {
1413 _pr = per_cpu(processors, cpu);
1414 if (!_pr || !_pr->flags.power_setup_done)
1416 acpi_processor_get_power_info(_pr);
1417 if (_pr->flags.power) {
1418 dev = per_cpu(acpi_cpuidle_device, cpu);
1419 acpi_processor_setup_cpuidle_dev(_pr, dev);
1420 cpuidle_enable_device(dev);
1423 cpuidle_resume_and_unlock();
1430 static int acpi_processor_registered;
1432 int acpi_processor_power_init(struct acpi_processor *pr)
1435 struct cpuidle_device *dev;
1437 if (disabled_by_idle_boot_param())
1440 acpi_processor_cstate_first_run_checks();
1442 if (!acpi_processor_get_power_info(pr))
1443 pr->flags.power_setup_done = 1;
1446 * Install the idle handler if processor power management is supported.
1447 * Note that we use previously set idle handler will be used on
1448 * platforms that only support C1.
1450 if (pr->flags.power) {
1451 /* Register acpi_idle_driver if not already registered */
1452 if (!acpi_processor_registered) {
1453 acpi_processor_setup_cpuidle_states(pr);
1454 retval = cpuidle_register_driver(&acpi_idle_driver);
1457 pr_debug("%s registered with cpuidle\n",
1458 acpi_idle_driver.name);
1461 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1464 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1466 acpi_processor_setup_cpuidle_dev(pr, dev);
1468 /* Register per-cpu cpuidle_device. Cpuidle driver
1469 * must already be registered before registering device
1471 retval = cpuidle_register_device(dev);
1473 if (acpi_processor_registered == 0)
1474 cpuidle_unregister_driver(&acpi_idle_driver);
1477 acpi_processor_registered++;
1482 int acpi_processor_power_exit(struct acpi_processor *pr)
1484 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1486 if (disabled_by_idle_boot_param())
1489 if (pr->flags.power) {
1490 cpuidle_unregister_device(dev);
1491 acpi_processor_registered--;
1492 if (acpi_processor_registered == 0)
1493 cpuidle_unregister_driver(&acpi_idle_driver);
1496 pr->flags.power_setup_done = 0;