1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * processor_idle - idle state submodule to the ACPI processor driver
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
13 #define pr_fmt(fmt) "ACPI: " fmt
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>
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.
34 #define ACPI_PROCESSOR_CLASS "processor"
35 #define _COMPONENT ACPI_PROCESSOR_COMPONENT
36 ACPI_MODULE_NAME("processor_idle");
38 #define ACPI_IDLE_STATE_START (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX) ? 1 : 0)
40 static unsigned int max_cstate __read_mostly = ACPI_PROCESSOR_MAX_POWER;
41 module_param(max_cstate, uint, 0000);
42 static unsigned int nocst __read_mostly;
43 module_param(nocst, uint, 0000);
44 static int bm_check_disable __read_mostly;
45 module_param(bm_check_disable, uint, 0000);
47 static unsigned int latency_factor __read_mostly = 2;
48 module_param(latency_factor, uint, 0644);
50 static DEFINE_PER_CPU(struct cpuidle_device *, acpi_cpuidle_device);
52 struct cpuidle_driver acpi_idle_driver = {
57 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
59 DEFINE_PER_CPU(struct acpi_processor_cx * [CPUIDLE_STATE_MAX], acpi_cstate);
61 static int disabled_by_idle_boot_param(void)
63 return boot_option_idle_override == IDLE_POLL ||
64 boot_option_idle_override == IDLE_HALT;
68 * IBM ThinkPad R40e crashes mysteriously when going into C2 or C3.
69 * For now disable this. Probably a bug somewhere else.
71 * To skip this limit, boot/load with a large max_cstate limit.
73 static int set_max_cstate(const struct dmi_system_id *id)
75 if (max_cstate > ACPI_PROCESSOR_MAX_POWER)
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);
82 max_cstate = (long)id->driver_data;
87 static 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")},
92 { set_max_cstate, "Pavilion zv5000", {
93 DMI_MATCH(DMI_SYS_VENDOR, "Hewlett-Packard"),
94 DMI_MATCH(DMI_PRODUCT_NAME,"Pavilion zv5000 (DS502A#ABA)")},
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")},
105 * Callers should disable interrupts before the call and enable
106 * interrupts after return.
108 static void __cpuidle acpi_safe_halt(void)
110 if (!tif_need_resched()) {
116 #ifdef ARCH_APICTIMER_STOPS_ON_C3
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.
124 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
125 struct acpi_processor_cx *cx)
127 struct acpi_processor_power *pwr = &pr->power;
128 u8 type = local_apic_timer_c2_ok ? ACPI_STATE_C3 : ACPI_STATE_C2;
130 if (cpu_has(&cpu_data(pr->id), X86_FEATURE_ARAT))
133 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E))
134 type = ACPI_STATE_C1;
137 * Check, if one of the previous states already marked the lapic
140 if (pwr->timer_broadcast_on_state < state)
143 if (cx->type >= type)
144 pr->power.timer_broadcast_on_state = state;
147 static void __lapic_timer_propagate_broadcast(void *arg)
149 struct acpi_processor *pr = (struct acpi_processor *) arg;
151 if (pr->power.timer_broadcast_on_state < INT_MAX)
152 tick_broadcast_enable();
154 tick_broadcast_disable();
157 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr)
159 smp_call_function_single(pr->id, __lapic_timer_propagate_broadcast,
163 /* Power(C) State timer broadcast control */
164 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
165 struct acpi_processor_cx *cx,
168 int state = cx - pr->power.states;
170 if (state >= pr->power.timer_broadcast_on_state) {
172 tick_broadcast_enter();
174 tick_broadcast_exit();
180 static void lapic_timer_check_state(int state, struct acpi_processor *pr,
181 struct acpi_processor_cx *cstate) { }
182 static void lapic_timer_propagate_broadcast(struct acpi_processor *pr) { }
183 static void lapic_timer_state_broadcast(struct acpi_processor *pr,
184 struct acpi_processor_cx *cx,
191 #if defined(CONFIG_X86)
192 static void tsc_check_state(int state)
194 switch (boot_cpu_data.x86_vendor) {
195 case X86_VENDOR_HYGON:
197 case X86_VENDOR_INTEL:
198 case X86_VENDOR_CENTAUR:
200 * AMD Fam10h TSC will tick in all
201 * C/P/S0/S1 states when this bit is set.
203 if (boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
208 /* TSC could halt in idle, so notify users */
209 if (state > ACPI_STATE_C1)
210 mark_tsc_unstable("TSC halts in idle");
214 static void tsc_check_state(int state) { return; }
217 static int acpi_processor_get_power_info_fadt(struct acpi_processor *pr)
223 /* if info is obtained from pblk/fadt, type equals state */
224 pr->power.states[ACPI_STATE_C2].type = ACPI_STATE_C2;
225 pr->power.states[ACPI_STATE_C3].type = ACPI_STATE_C3;
227 #ifndef CONFIG_HOTPLUG_CPU
229 * Check for P_LVL2_UP flag before entering C2 and above on
232 if ((num_online_cpus() > 1) &&
233 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED))
237 /* determine C2 and C3 address from pblk */
238 pr->power.states[ACPI_STATE_C2].address = pr->pblk + 4;
239 pr->power.states[ACPI_STATE_C3].address = pr->pblk + 5;
241 /* determine latencies from FADT */
242 pr->power.states[ACPI_STATE_C2].latency = acpi_gbl_FADT.c2_latency;
243 pr->power.states[ACPI_STATE_C3].latency = acpi_gbl_FADT.c3_latency;
246 * FADT specified C2 latency must be less than or equal to
249 if (acpi_gbl_FADT.c2_latency > ACPI_PROCESSOR_MAX_C2_LATENCY) {
250 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
251 "C2 latency too large [%d]\n", acpi_gbl_FADT.c2_latency));
253 pr->power.states[ACPI_STATE_C2].address = 0;
257 * FADT supplied C3 latency must be less than or equal to
260 if (acpi_gbl_FADT.c3_latency > ACPI_PROCESSOR_MAX_C3_LATENCY) {
261 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
262 "C3 latency too large [%d]\n", acpi_gbl_FADT.c3_latency));
264 pr->power.states[ACPI_STATE_C3].address = 0;
267 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
268 "lvl2[0x%08x] lvl3[0x%08x]\n",
269 pr->power.states[ACPI_STATE_C2].address,
270 pr->power.states[ACPI_STATE_C3].address));
272 snprintf(pr->power.states[ACPI_STATE_C2].desc,
273 ACPI_CX_DESC_LEN, "ACPI P_LVL2 IOPORT 0x%x",
274 pr->power.states[ACPI_STATE_C2].address);
275 snprintf(pr->power.states[ACPI_STATE_C3].desc,
276 ACPI_CX_DESC_LEN, "ACPI P_LVL3 IOPORT 0x%x",
277 pr->power.states[ACPI_STATE_C3].address);
282 static int acpi_processor_get_power_info_default(struct acpi_processor *pr)
284 if (!pr->power.states[ACPI_STATE_C1].valid) {
285 /* set the first C-State to C1 */
286 /* all processors need to support C1 */
287 pr->power.states[ACPI_STATE_C1].type = ACPI_STATE_C1;
288 pr->power.states[ACPI_STATE_C1].valid = 1;
289 pr->power.states[ACPI_STATE_C1].entry_method = ACPI_CSTATE_HALT;
291 snprintf(pr->power.states[ACPI_STATE_C1].desc,
292 ACPI_CX_DESC_LEN, "ACPI HLT");
294 /* the C0 state only exists as a filler in our array */
295 pr->power.states[ACPI_STATE_C0].valid = 1;
299 static int acpi_processor_get_power_info_cst(struct acpi_processor *pr)
305 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
306 union acpi_object *cst;
313 status = acpi_evaluate_object(pr->handle, "_CST", NULL, &buffer);
314 if (ACPI_FAILURE(status)) {
315 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _CST, giving up\n"));
319 cst = buffer.pointer;
321 /* There must be at least 2 elements */
322 if (!cst || (cst->type != ACPI_TYPE_PACKAGE) || cst->package.count < 2) {
323 pr_err("not enough elements in _CST\n");
328 count = cst->package.elements[0].integer.value;
330 /* Validate number of power states. */
331 if (count < 1 || count != cst->package.count - 1) {
332 pr_err("count given by _CST is not valid\n");
337 /* Tell driver that at least _CST is supported. */
338 pr->flags.has_cst = 1;
340 for (i = 1; i <= count; i++) {
341 union acpi_object *element;
342 union acpi_object *obj;
343 struct acpi_power_register *reg;
344 struct acpi_processor_cx cx;
346 memset(&cx, 0, sizeof(cx));
348 element = &(cst->package.elements[i]);
349 if (element->type != ACPI_TYPE_PACKAGE)
352 if (element->package.count != 4)
355 obj = &(element->package.elements[0]);
357 if (obj->type != ACPI_TYPE_BUFFER)
360 reg = (struct acpi_power_register *)obj->buffer.pointer;
362 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
363 (reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE))
366 /* There should be an easy way to extract an integer... */
367 obj = &(element->package.elements[1]);
368 if (obj->type != ACPI_TYPE_INTEGER)
371 cx.type = obj->integer.value;
373 * Some buggy BIOSes won't list C1 in _CST -
374 * Let acpi_processor_get_power_info_default() handle them later
376 if (i == 1 && cx.type != ACPI_STATE_C1)
379 cx.address = reg->address;
380 cx.index = current_count + 1;
382 cx.entry_method = ACPI_CSTATE_SYSTEMIO;
383 if (reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE) {
384 if (acpi_processor_ffh_cstate_probe
385 (pr->id, &cx, reg) == 0) {
386 cx.entry_method = ACPI_CSTATE_FFH;
387 } else if (cx.type == ACPI_STATE_C1) {
389 * C1 is a special case where FIXED_HARDWARE
390 * can be handled in non-MWAIT way as well.
391 * In that case, save this _CST entry info.
392 * Otherwise, ignore this info and continue.
394 cx.entry_method = ACPI_CSTATE_HALT;
395 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
399 if (cx.type == ACPI_STATE_C1 &&
400 (boot_option_idle_override == IDLE_NOMWAIT)) {
402 * In most cases the C1 space_id obtained from
403 * _CST object is FIXED_HARDWARE access mode.
404 * But when the option of idle=halt is added,
405 * the entry_method type should be changed from
406 * CSTATE_FFH to CSTATE_HALT.
407 * When the option of idle=nomwait is added,
408 * the C1 entry_method type should be
411 cx.entry_method = ACPI_CSTATE_HALT;
412 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI HLT");
415 snprintf(cx.desc, ACPI_CX_DESC_LEN, "ACPI IOPORT 0x%x",
419 if (cx.type == ACPI_STATE_C1) {
423 obj = &(element->package.elements[2]);
424 if (obj->type != ACPI_TYPE_INTEGER)
427 cx.latency = obj->integer.value;
429 obj = &(element->package.elements[3]);
430 if (obj->type != ACPI_TYPE_INTEGER)
434 memcpy(&(pr->power.states[current_count]), &cx, sizeof(cx));
437 * We support total ACPI_PROCESSOR_MAX_POWER - 1
438 * (From 1 through ACPI_PROCESSOR_MAX_POWER - 1)
440 if (current_count >= (ACPI_PROCESSOR_MAX_POWER - 1)) {
441 pr_warn("Limiting number of power states to max (%d)\n",
442 ACPI_PROCESSOR_MAX_POWER);
443 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
448 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %d power states\n",
451 /* Validate number of power states discovered */
452 if (current_count < 2)
456 kfree(buffer.pointer);
461 static void acpi_processor_power_verify_c3(struct acpi_processor *pr,
462 struct acpi_processor_cx *cx)
464 static int bm_check_flag = -1;
465 static int bm_control_flag = -1;
472 * PIIX4 Erratum #18: We don't support C3 when Type-F (fast)
473 * DMA transfers are used by any ISA device to avoid livelock.
474 * Note that we could disable Type-F DMA (as recommended by
475 * the erratum), but this is known to disrupt certain ISA
476 * devices thus we take the conservative approach.
478 else if (errata.piix4.fdma) {
479 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
480 "C3 not supported on PIIX4 with Type-F DMA\n"));
484 /* All the logic here assumes flags.bm_check is same across all CPUs */
485 if (bm_check_flag == -1) {
486 /* Determine whether bm_check is needed based on CPU */
487 acpi_processor_power_init_bm_check(&(pr->flags), pr->id);
488 bm_check_flag = pr->flags.bm_check;
489 bm_control_flag = pr->flags.bm_control;
491 pr->flags.bm_check = bm_check_flag;
492 pr->flags.bm_control = bm_control_flag;
495 if (pr->flags.bm_check) {
496 if (!pr->flags.bm_control) {
497 if (pr->flags.has_cst != 1) {
498 /* bus mastering control is necessary */
499 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
500 "C3 support requires BM control\n"));
503 /* Here we enter C3 without bus mastering */
504 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
505 "C3 support without BM control\n"));
510 * WBINVD should be set in fadt, for C3 state to be
511 * supported on when bm_check is not required.
513 if (!(acpi_gbl_FADT.flags & ACPI_FADT_WBINVD)) {
514 ACPI_DEBUG_PRINT((ACPI_DB_INFO,
515 "Cache invalidation should work properly"
516 " for C3 to be enabled on SMP systems\n"));
522 * Otherwise we've met all of our C3 requirements.
523 * Normalize the C3 latency to expidite policy. Enable
524 * checking of bus mastering status (bm_check) so we can
525 * use this in our C3 policy
530 * On older chipsets, BM_RLD needs to be set
531 * in order for Bus Master activity to wake the
532 * system from C3. Newer chipsets handle DMA
533 * during C3 automatically and BM_RLD is a NOP.
534 * In either case, the proper way to
535 * handle BM_RLD is to set it and leave it set.
537 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_RLD, 1);
542 static int acpi_processor_power_verify(struct acpi_processor *pr)
545 unsigned int working = 0;
547 pr->power.timer_broadcast_on_state = INT_MAX;
549 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
550 struct acpi_processor_cx *cx = &pr->power.states[i];
564 acpi_processor_power_verify_c3(pr, cx);
570 lapic_timer_check_state(i, pr, cx);
571 tsc_check_state(cx->type);
575 lapic_timer_propagate_broadcast(pr);
580 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
586 /* NOTE: the idle thread may not be running while calling
589 /* Zero initialize all the C-states info. */
590 memset(pr->power.states, 0, sizeof(pr->power.states));
592 result = acpi_processor_get_power_info_cst(pr);
593 if (result == -ENODEV)
594 result = acpi_processor_get_power_info_fadt(pr);
599 acpi_processor_get_power_info_default(pr);
601 pr->power.count = acpi_processor_power_verify(pr);
604 * if one state of type C2 or C3 is available, mark this
605 * CPU as being "idle manageable"
607 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER; i++) {
608 if (pr->power.states[i].valid) {
610 if (pr->power.states[i].type >= ACPI_STATE_C2)
619 * acpi_idle_bm_check - checks if bus master activity was detected
621 static int acpi_idle_bm_check(void)
625 if (bm_check_disable)
628 acpi_read_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, &bm_status);
630 acpi_write_bit_register(ACPI_BITREG_BUS_MASTER_STATUS, 1);
632 * PIIX4 Erratum #18: Note that BM_STS doesn't always reflect
633 * the true state of bus mastering activity; forcing us to
634 * manually check the BMIDEA bit of each IDE channel.
636 else if (errata.piix4.bmisx) {
637 if ((inb_p(errata.piix4.bmisx + 0x02) & 0x01)
638 || (inb_p(errata.piix4.bmisx + 0x0A) & 0x01))
645 * acpi_idle_do_entry - enter idle state using the appropriate method
648 * Caller disables interrupt before call and enables interrupt after return.
650 static void __cpuidle acpi_idle_do_entry(struct acpi_processor_cx *cx)
652 if (cx->entry_method == ACPI_CSTATE_FFH) {
653 /* Call into architectural FFH based C-state */
654 acpi_processor_ffh_cstate_enter(cx);
655 } else if (cx->entry_method == ACPI_CSTATE_HALT) {
658 /* IO port based C-state */
660 /* Dummy wait op - must do something useless after P_LVL2 read
661 because chipsets cannot guarantee that STPCLK# signal
662 gets asserted in time to freeze execution properly. */
663 inl(acpi_gbl_FADT.xpm_timer_block.address);
668 * acpi_idle_play_dead - enters an ACPI state for long-term idle (i.e. off-lining)
669 * @dev: the target CPU
670 * @index: the index of suggested state
672 static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
674 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
676 ACPI_FLUSH_CPU_CACHE();
680 if (cx->entry_method == ACPI_CSTATE_HALT)
682 else if (cx->entry_method == ACPI_CSTATE_SYSTEMIO) {
684 /* See comment in acpi_idle_do_entry() */
685 inl(acpi_gbl_FADT.xpm_timer_block.address);
694 static bool acpi_idle_fallback_to_c1(struct acpi_processor *pr)
696 return IS_ENABLED(CONFIG_HOTPLUG_CPU) && !pr->flags.has_cst &&
697 !(acpi_gbl_FADT.flags & ACPI_FADT_C2_MP_SUPPORTED);
700 static int c3_cpu_count;
701 static DEFINE_RAW_SPINLOCK(c3_lock);
704 * acpi_idle_enter_bm - enters C3 with proper BM handling
705 * @pr: Target processor
706 * @cx: Target state context
707 * @timer_bc: Whether or not to change timer mode to broadcast
709 static void acpi_idle_enter_bm(struct acpi_processor *pr,
710 struct acpi_processor_cx *cx, bool timer_bc)
712 acpi_unlazy_tlb(smp_processor_id());
715 * Must be done before busmaster disable as we might need to
719 lapic_timer_state_broadcast(pr, cx, 1);
723 * bm_check implies we need ARB_DIS
724 * bm_control implies whether we can do ARB_DIS
726 * That leaves a case where bm_check is set and bm_control is
727 * not set. In that case we cannot do much, we enter C3
728 * without doing anything.
730 if (pr->flags.bm_control) {
731 raw_spin_lock(&c3_lock);
733 /* Disable bus master arbitration when all CPUs are in C3 */
734 if (c3_cpu_count == num_online_cpus())
735 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 1);
736 raw_spin_unlock(&c3_lock);
739 acpi_idle_do_entry(cx);
741 /* Re-enable bus master arbitration */
742 if (pr->flags.bm_control) {
743 raw_spin_lock(&c3_lock);
744 acpi_write_bit_register(ACPI_BITREG_ARB_DISABLE, 0);
746 raw_spin_unlock(&c3_lock);
750 lapic_timer_state_broadcast(pr, cx, 0);
753 static int acpi_idle_enter(struct cpuidle_device *dev,
754 struct cpuidle_driver *drv, int index)
756 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
757 struct acpi_processor *pr;
759 pr = __this_cpu_read(processors);
763 if (cx->type != ACPI_STATE_C1) {
764 if (acpi_idle_fallback_to_c1(pr) && num_online_cpus() > 1) {
765 index = ACPI_IDLE_STATE_START;
766 cx = per_cpu(acpi_cstate[index], dev->cpu);
767 } else if (cx->type == ACPI_STATE_C3 && pr->flags.bm_check) {
768 if (cx->bm_sts_skip || !acpi_idle_bm_check()) {
769 acpi_idle_enter_bm(pr, cx, true);
771 } else if (drv->safe_state_index >= 0) {
772 index = drv->safe_state_index;
773 cx = per_cpu(acpi_cstate[index], dev->cpu);
781 lapic_timer_state_broadcast(pr, cx, 1);
783 if (cx->type == ACPI_STATE_C3)
784 ACPI_FLUSH_CPU_CACHE();
786 acpi_idle_do_entry(cx);
788 lapic_timer_state_broadcast(pr, cx, 0);
793 static void acpi_idle_enter_s2idle(struct cpuidle_device *dev,
794 struct cpuidle_driver *drv, int index)
796 struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);
798 if (cx->type == ACPI_STATE_C3) {
799 struct acpi_processor *pr = __this_cpu_read(processors);
804 if (pr->flags.bm_check) {
805 acpi_idle_enter_bm(pr, cx, false);
808 ACPI_FLUSH_CPU_CACHE();
811 acpi_idle_do_entry(cx);
814 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
815 struct cpuidle_device *dev)
817 int i, count = ACPI_IDLE_STATE_START;
818 struct acpi_processor_cx *cx;
823 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
824 cx = &pr->power.states[i];
829 per_cpu(acpi_cstate[count], dev->cpu) = cx;
832 if (count == CPUIDLE_STATE_MAX)
842 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
845 struct acpi_processor_cx *cx;
846 struct cpuidle_state *state;
847 struct cpuidle_driver *drv = &acpi_idle_driver;
852 if (IS_ENABLED(CONFIG_ARCH_HAS_CPU_RELAX)) {
853 cpuidle_poll_state_init(drv);
859 for (i = 1; i < ACPI_PROCESSOR_MAX_POWER && i <= max_cstate; i++) {
860 cx = &pr->power.states[i];
865 state = &drv->states[count];
866 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d", i);
867 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
868 state->exit_latency = cx->latency;
869 state->target_residency = cx->latency * latency_factor;
870 state->enter = acpi_idle_enter;
873 if (cx->type == ACPI_STATE_C1 || cx->type == ACPI_STATE_C2) {
874 state->enter_dead = acpi_idle_play_dead;
875 drv->safe_state_index = count;
878 * Halt-induced C1 is not good for ->enter_s2idle, because it
879 * re-enables interrupts on exit. Moreover, C1 is generally not
880 * particularly interesting from the suspend-to-idle angle, so
881 * avoid C1 and the situations in which we may need to fall back
884 if (cx->type != ACPI_STATE_C1 && !acpi_idle_fallback_to_c1(pr))
885 state->enter_s2idle = acpi_idle_enter_s2idle;
888 if (count == CPUIDLE_STATE_MAX)
892 drv->state_count = count;
900 static inline void acpi_processor_cstate_first_run_checks(void)
903 static int first_run;
907 dmi_check_system(processor_power_dmi_table);
908 max_cstate = acpi_processor_cstate_check(max_cstate);
909 if (max_cstate < ACPI_C_STATES_MAX)
910 pr_notice("ACPI: processor limited to max C-state %d\n",
914 if (acpi_gbl_FADT.cst_control && !nocst) {
915 status = acpi_os_write_port(acpi_gbl_FADT.smi_command,
916 acpi_gbl_FADT.cst_control, 8);
917 if (ACPI_FAILURE(status))
918 ACPI_EXCEPTION((AE_INFO, status,
919 "Notifying BIOS of _CST ability failed"));
924 static inline int disabled_by_idle_boot_param(void) { return 0; }
925 static inline void acpi_processor_cstate_first_run_checks(void) { }
926 static int acpi_processor_get_cstate_info(struct acpi_processor *pr)
931 static int acpi_processor_setup_cpuidle_cx(struct acpi_processor *pr,
932 struct cpuidle_device *dev)
937 static int acpi_processor_setup_cstates(struct acpi_processor *pr)
942 #endif /* CONFIG_ACPI_PROCESSOR_CSTATE */
944 struct acpi_lpi_states_array {
946 unsigned int composite_states_size;
947 struct acpi_lpi_state *entries;
948 struct acpi_lpi_state *composite_states[ACPI_PROCESSOR_MAX_POWER];
951 static int obj_get_integer(union acpi_object *obj, u32 *value)
953 if (obj->type != ACPI_TYPE_INTEGER)
956 *value = obj->integer.value;
960 static int acpi_processor_evaluate_lpi(acpi_handle handle,
961 struct acpi_lpi_states_array *info)
965 int pkg_count, state_idx = 1, loop;
966 struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
967 union acpi_object *lpi_data;
968 struct acpi_lpi_state *lpi_state;
970 status = acpi_evaluate_object(handle, "_LPI", NULL, &buffer);
971 if (ACPI_FAILURE(status)) {
972 ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No _LPI, giving up\n"));
976 lpi_data = buffer.pointer;
978 /* There must be at least 4 elements = 3 elements + 1 package */
979 if (!lpi_data || lpi_data->type != ACPI_TYPE_PACKAGE ||
980 lpi_data->package.count < 4) {
981 pr_debug("not enough elements in _LPI\n");
986 pkg_count = lpi_data->package.elements[2].integer.value;
988 /* Validate number of power states. */
989 if (pkg_count < 1 || pkg_count != lpi_data->package.count - 3) {
990 pr_debug("count given by _LPI is not valid\n");
995 lpi_state = kcalloc(pkg_count, sizeof(*lpi_state), GFP_KERNEL);
1001 info->size = pkg_count;
1002 info->entries = lpi_state;
1004 /* LPI States start at index 3 */
1005 for (loop = 3; state_idx <= pkg_count; loop++, state_idx++, lpi_state++) {
1006 union acpi_object *element, *pkg_elem, *obj;
1008 element = &lpi_data->package.elements[loop];
1009 if (element->type != ACPI_TYPE_PACKAGE || element->package.count < 7)
1012 pkg_elem = element->package.elements;
1015 if (obj->type == ACPI_TYPE_BUFFER) {
1016 struct acpi_power_register *reg;
1018 reg = (struct acpi_power_register *)obj->buffer.pointer;
1019 if (reg->space_id != ACPI_ADR_SPACE_SYSTEM_IO &&
1020 reg->space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)
1023 lpi_state->address = reg->address;
1024 lpi_state->entry_method =
1025 reg->space_id == ACPI_ADR_SPACE_FIXED_HARDWARE ?
1026 ACPI_CSTATE_FFH : ACPI_CSTATE_SYSTEMIO;
1027 } else if (obj->type == ACPI_TYPE_INTEGER) {
1028 lpi_state->entry_method = ACPI_CSTATE_INTEGER;
1029 lpi_state->address = obj->integer.value;
1034 /* elements[7,8] skipped for now i.e. Residency/Usage counter*/
1037 if (obj->type == ACPI_TYPE_STRING)
1038 strlcpy(lpi_state->desc, obj->string.pointer,
1041 lpi_state->index = state_idx;
1042 if (obj_get_integer(pkg_elem + 0, &lpi_state->min_residency)) {
1043 pr_debug("No min. residency found, assuming 10 us\n");
1044 lpi_state->min_residency = 10;
1047 if (obj_get_integer(pkg_elem + 1, &lpi_state->wake_latency)) {
1048 pr_debug("No wakeup residency found, assuming 10 us\n");
1049 lpi_state->wake_latency = 10;
1052 if (obj_get_integer(pkg_elem + 2, &lpi_state->flags))
1053 lpi_state->flags = 0;
1055 if (obj_get_integer(pkg_elem + 3, &lpi_state->arch_flags))
1056 lpi_state->arch_flags = 0;
1058 if (obj_get_integer(pkg_elem + 4, &lpi_state->res_cnt_freq))
1059 lpi_state->res_cnt_freq = 1;
1061 if (obj_get_integer(pkg_elem + 5, &lpi_state->enable_parent_state))
1062 lpi_state->enable_parent_state = 0;
1065 acpi_handle_debug(handle, "Found %d power states\n", state_idx);
1067 kfree(buffer.pointer);
1072 * flat_state_cnt - the number of composite LPI states after the process of flattening
1074 static int flat_state_cnt;
1077 * combine_lpi_states - combine local and parent LPI states to form a composite LPI state
1079 * @local: local LPI state
1080 * @parent: parent LPI state
1081 * @result: composite LPI state
1083 static bool combine_lpi_states(struct acpi_lpi_state *local,
1084 struct acpi_lpi_state *parent,
1085 struct acpi_lpi_state *result)
1087 if (parent->entry_method == ACPI_CSTATE_INTEGER) {
1088 if (!parent->address) /* 0 means autopromotable */
1090 result->address = local->address + parent->address;
1092 result->address = parent->address;
1095 result->min_residency = max(local->min_residency, parent->min_residency);
1096 result->wake_latency = local->wake_latency + parent->wake_latency;
1097 result->enable_parent_state = parent->enable_parent_state;
1098 result->entry_method = local->entry_method;
1100 result->flags = parent->flags;
1101 result->arch_flags = parent->arch_flags;
1102 result->index = parent->index;
1104 strlcpy(result->desc, local->desc, ACPI_CX_DESC_LEN);
1105 strlcat(result->desc, "+", ACPI_CX_DESC_LEN);
1106 strlcat(result->desc, parent->desc, ACPI_CX_DESC_LEN);
1110 #define ACPI_LPI_STATE_FLAGS_ENABLED BIT(0)
1112 static void stash_composite_state(struct acpi_lpi_states_array *curr_level,
1113 struct acpi_lpi_state *t)
1115 curr_level->composite_states[curr_level->composite_states_size++] = t;
1118 static int flatten_lpi_states(struct acpi_processor *pr,
1119 struct acpi_lpi_states_array *curr_level,
1120 struct acpi_lpi_states_array *prev_level)
1122 int i, j, state_count = curr_level->size;
1123 struct acpi_lpi_state *p, *t = curr_level->entries;
1125 curr_level->composite_states_size = 0;
1126 for (j = 0; j < state_count; j++, t++) {
1127 struct acpi_lpi_state *flpi;
1129 if (!(t->flags & ACPI_LPI_STATE_FLAGS_ENABLED))
1132 if (flat_state_cnt >= ACPI_PROCESSOR_MAX_POWER) {
1133 pr_warn("Limiting number of LPI states to max (%d)\n",
1134 ACPI_PROCESSOR_MAX_POWER);
1135 pr_warn("Please increase ACPI_PROCESSOR_MAX_POWER if needed.\n");
1139 flpi = &pr->power.lpi_states[flat_state_cnt];
1141 if (!prev_level) { /* leaf/processor node */
1142 memcpy(flpi, t, sizeof(*t));
1143 stash_composite_state(curr_level, flpi);
1148 for (i = 0; i < prev_level->composite_states_size; i++) {
1149 p = prev_level->composite_states[i];
1150 if (t->index <= p->enable_parent_state &&
1151 combine_lpi_states(p, t, flpi)) {
1152 stash_composite_state(curr_level, flpi);
1159 kfree(curr_level->entries);
1163 static int acpi_processor_get_lpi_info(struct acpi_processor *pr)
1167 acpi_handle handle = pr->handle, pr_ahandle;
1168 struct acpi_device *d = NULL;
1169 struct acpi_lpi_states_array info[2], *tmp, *prev, *curr;
1171 if (!osc_pc_lpi_support_confirmed)
1174 if (!acpi_has_method(handle, "_LPI"))
1180 handle = pr->handle;
1181 ret = acpi_processor_evaluate_lpi(handle, prev);
1184 flatten_lpi_states(pr, prev, NULL);
1186 status = acpi_get_parent(handle, &pr_ahandle);
1187 while (ACPI_SUCCESS(status)) {
1188 acpi_bus_get_device(pr_ahandle, &d);
1189 handle = pr_ahandle;
1191 if (strcmp(acpi_device_hid(d), ACPI_PROCESSOR_CONTAINER_HID))
1194 /* can be optional ? */
1195 if (!acpi_has_method(handle, "_LPI"))
1198 ret = acpi_processor_evaluate_lpi(handle, curr);
1202 /* flatten all the LPI states in this level of hierarchy */
1203 flatten_lpi_states(pr, curr, prev);
1205 tmp = prev, prev = curr, curr = tmp;
1207 status = acpi_get_parent(handle, &pr_ahandle);
1210 pr->power.count = flat_state_cnt;
1211 /* reset the index after flattening */
1212 for (i = 0; i < pr->power.count; i++)
1213 pr->power.lpi_states[i].index = i;
1215 /* Tell driver that _LPI is supported. */
1216 pr->flags.has_lpi = 1;
1217 pr->flags.power = 1;
1222 int __weak acpi_processor_ffh_lpi_probe(unsigned int cpu)
1227 int __weak acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
1233 * acpi_idle_lpi_enter - enters an ACPI any LPI state
1234 * @dev: the target CPU
1235 * @drv: cpuidle driver containing cpuidle state info
1236 * @index: index of target state
1238 * Return: 0 for success or negative value for error
1240 static int acpi_idle_lpi_enter(struct cpuidle_device *dev,
1241 struct cpuidle_driver *drv, int index)
1243 struct acpi_processor *pr;
1244 struct acpi_lpi_state *lpi;
1246 pr = __this_cpu_read(processors);
1251 lpi = &pr->power.lpi_states[index];
1252 if (lpi->entry_method == ACPI_CSTATE_FFH)
1253 return acpi_processor_ffh_lpi_enter(lpi);
1258 static int acpi_processor_setup_lpi_states(struct acpi_processor *pr)
1261 struct acpi_lpi_state *lpi;
1262 struct cpuidle_state *state;
1263 struct cpuidle_driver *drv = &acpi_idle_driver;
1265 if (!pr->flags.has_lpi)
1268 for (i = 0; i < pr->power.count && i < CPUIDLE_STATE_MAX; i++) {
1269 lpi = &pr->power.lpi_states[i];
1271 state = &drv->states[i];
1272 snprintf(state->name, CPUIDLE_NAME_LEN, "LPI-%d", i);
1273 strlcpy(state->desc, lpi->desc, CPUIDLE_DESC_LEN);
1274 state->exit_latency = lpi->wake_latency;
1275 state->target_residency = lpi->min_residency;
1276 if (lpi->arch_flags)
1277 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1278 state->enter = acpi_idle_lpi_enter;
1279 drv->safe_state_index = i;
1282 drv->state_count = i;
1288 * acpi_processor_setup_cpuidle_states- prepares and configures cpuidle
1289 * global state data i.e. idle routines
1291 * @pr: the ACPI processor
1293 static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
1296 struct cpuidle_driver *drv = &acpi_idle_driver;
1298 if (!pr->flags.power_setup_done || !pr->flags.power)
1301 drv->safe_state_index = -1;
1302 for (i = ACPI_IDLE_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
1303 drv->states[i].name[0] = '\0';
1304 drv->states[i].desc[0] = '\0';
1307 if (pr->flags.has_lpi)
1308 return acpi_processor_setup_lpi_states(pr);
1310 return acpi_processor_setup_cstates(pr);
1314 * acpi_processor_setup_cpuidle_dev - prepares and configures CPUIDLE
1315 * device i.e. per-cpu data
1317 * @pr: the ACPI processor
1318 * @dev : the cpuidle device
1320 static int acpi_processor_setup_cpuidle_dev(struct acpi_processor *pr,
1321 struct cpuidle_device *dev)
1323 if (!pr->flags.power_setup_done || !pr->flags.power || !dev)
1327 if (pr->flags.has_lpi)
1328 return acpi_processor_ffh_lpi_probe(pr->id);
1330 return acpi_processor_setup_cpuidle_cx(pr, dev);
1333 static int acpi_processor_get_power_info(struct acpi_processor *pr)
1337 ret = acpi_processor_get_lpi_info(pr);
1339 ret = acpi_processor_get_cstate_info(pr);
1344 int acpi_processor_hotplug(struct acpi_processor *pr)
1347 struct cpuidle_device *dev;
1349 if (disabled_by_idle_boot_param())
1352 if (!pr->flags.power_setup_done)
1355 dev = per_cpu(acpi_cpuidle_device, pr->id);
1356 cpuidle_pause_and_lock();
1357 cpuidle_disable_device(dev);
1358 ret = acpi_processor_get_power_info(pr);
1359 if (!ret && pr->flags.power) {
1360 acpi_processor_setup_cpuidle_dev(pr, dev);
1361 ret = cpuidle_enable_device(dev);
1363 cpuidle_resume_and_unlock();
1368 int acpi_processor_power_state_has_changed(struct acpi_processor *pr)
1371 struct acpi_processor *_pr;
1372 struct cpuidle_device *dev;
1374 if (disabled_by_idle_boot_param())
1377 if (!pr->flags.power_setup_done)
1381 * FIXME: Design the ACPI notification to make it once per
1382 * system instead of once per-cpu. This condition is a hack
1383 * to make the code that updates C-States be called once.
1386 if (pr->id == 0 && cpuidle_get_driver() == &acpi_idle_driver) {
1388 /* Protect against cpu-hotplug */
1390 cpuidle_pause_and_lock();
1392 /* Disable all cpuidle devices */
1393 for_each_online_cpu(cpu) {
1394 _pr = per_cpu(processors, cpu);
1395 if (!_pr || !_pr->flags.power_setup_done)
1397 dev = per_cpu(acpi_cpuidle_device, cpu);
1398 cpuidle_disable_device(dev);
1401 /* Populate Updated C-state information */
1402 acpi_processor_get_power_info(pr);
1403 acpi_processor_setup_cpuidle_states(pr);
1405 /* Enable all cpuidle devices */
1406 for_each_online_cpu(cpu) {
1407 _pr = per_cpu(processors, cpu);
1408 if (!_pr || !_pr->flags.power_setup_done)
1410 acpi_processor_get_power_info(_pr);
1411 if (_pr->flags.power) {
1412 dev = per_cpu(acpi_cpuidle_device, cpu);
1413 acpi_processor_setup_cpuidle_dev(_pr, dev);
1414 cpuidle_enable_device(dev);
1417 cpuidle_resume_and_unlock();
1424 static int acpi_processor_registered;
1426 int acpi_processor_power_init(struct acpi_processor *pr)
1429 struct cpuidle_device *dev;
1431 if (disabled_by_idle_boot_param())
1434 acpi_processor_cstate_first_run_checks();
1436 if (!acpi_processor_get_power_info(pr))
1437 pr->flags.power_setup_done = 1;
1440 * Install the idle handler if processor power management is supported.
1441 * Note that we use previously set idle handler will be used on
1442 * platforms that only support C1.
1444 if (pr->flags.power) {
1445 /* Register acpi_idle_driver if not already registered */
1446 if (!acpi_processor_registered) {
1447 acpi_processor_setup_cpuidle_states(pr);
1448 retval = cpuidle_register_driver(&acpi_idle_driver);
1451 pr_debug("%s registered with cpuidle\n",
1452 acpi_idle_driver.name);
1455 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1458 per_cpu(acpi_cpuidle_device, pr->id) = dev;
1460 acpi_processor_setup_cpuidle_dev(pr, dev);
1462 /* Register per-cpu cpuidle_device. Cpuidle driver
1463 * must already be registered before registering device
1465 retval = cpuidle_register_device(dev);
1467 if (acpi_processor_registered == 0)
1468 cpuidle_unregister_driver(&acpi_idle_driver);
1471 acpi_processor_registered++;
1476 int acpi_processor_power_exit(struct acpi_processor *pr)
1478 struct cpuidle_device *dev = per_cpu(acpi_cpuidle_device, pr->id);
1480 if (disabled_by_idle_boot_param())
1483 if (pr->flags.power) {
1484 cpuidle_unregister_device(dev);
1485 acpi_processor_registered--;
1486 if (acpi_processor_registered == 0)
1487 cpuidle_unregister_driver(&acpi_idle_driver);
1490 pr->flags.power_setup_done = 0;