1 // SPDX-License-Identifier: GPL-2.0-only
3 * intel_idle.c - native hardware idle loop for modern Intel processors
5 * Copyright (c) 2013 - 2020, Intel Corporation.
6 * Len Brown <len.brown@intel.com>
7 * Rafael J. Wysocki <rafael.j.wysocki@intel.com>
11 * intel_idle is a cpuidle driver that loads on all Intel CPUs with MWAIT
12 * in lieu of the legacy ACPI processor_idle driver. The intent is to
13 * make Linux more efficient on these processors, as intel_idle knows
14 * more than ACPI, as well as make Linux more immune to ACPI BIOS bugs.
20 * All CPUs have same idle states as boot CPU
22 * Chipset BM_STS (bus master status) bit is a NOP
23 * for preventing entry into deep C-states
25 * CPU will flush caches as needed when entering a C-state via MWAIT
26 * (in contrast to entering ACPI C3, in which case the WBINVD
27 * instruction needs to be executed to flush the caches)
33 * ACPI has a .suspend hack to turn off deep c-statees during suspend
34 * to avoid complications with the lapic timer workaround.
35 * Have not seen issues with suspend, but may need same workaround here.
39 /* un-comment DEBUG to enable pr_debug() statements */
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
44 #include <linux/acpi.h>
45 #include <linux/kernel.h>
46 #include <linux/cpuidle.h>
47 #include <linux/tick.h>
48 #include <trace/events/power.h>
49 #include <linux/sched.h>
50 #include <linux/sched/smt.h>
51 #include <linux/notifier.h>
52 #include <linux/cpu.h>
53 #include <linux/moduleparam.h>
54 #include <asm/cpu_device_id.h>
55 #include <asm/intel-family.h>
56 #include <asm/nospec-branch.h>
57 #include <asm/mwait.h>
59 #include <asm/fpu/api.h>
61 #define INTEL_IDLE_VERSION "0.5.1"
63 static struct cpuidle_driver intel_idle_driver = {
67 /* intel_idle.max_cstate=0 disables driver */
68 static int max_cstate = CPUIDLE_STATE_MAX - 1;
69 static unsigned int disabled_states_mask;
70 static unsigned int preferred_states_mask;
72 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
74 static unsigned long auto_demotion_disable_flags;
77 C1E_PROMOTION_PRESERVE,
80 } c1e_promotion = C1E_PROMOTION_PRESERVE;
83 struct cpuidle_state *state_table;
86 * Hardware C-state auto-demotion may not always be optimal.
87 * Indicate which enable bits to clear here.
89 unsigned long auto_demotion_disable_flags;
90 bool byt_auto_demotion_disable_flag;
91 bool disable_promotion_to_c1e;
95 static const struct idle_cpu *icpu __initdata;
96 static struct cpuidle_state *cpuidle_state_table __initdata;
98 static unsigned int mwait_substates __initdata;
101 * Enable interrupts before entering the C-state. On some platforms and for
102 * some C-states, this may measurably decrease interrupt latency.
104 #define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
107 * Enable this state by default even if the ACPI _CST does not list it.
109 #define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
112 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
115 #define CPUIDLE_FLAG_IBRS BIT(16)
118 * Initialize large xstate for the C6-state entrance.
120 #define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
123 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
124 * the C-state (top nibble) and sub-state (bottom nibble)
125 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
127 * We store the hint at the top of our "flags" for each state.
129 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
130 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
132 static __always_inline int __intel_idle(struct cpuidle_device *dev,
133 struct cpuidle_driver *drv, int index)
135 struct cpuidle_state *state = &drv->states[index];
136 unsigned long eax = flg2MWAIT(state->flags);
137 unsigned long ecx = 1; /* break on interrupt flag */
139 mwait_idle_with_hints(eax, ecx);
145 * intel_idle - Ask the processor to enter the given idle state.
146 * @dev: cpuidle device of the target CPU.
147 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
148 * @index: Target idle state index.
150 * Use the MWAIT instruction to notify the processor that the CPU represented by
151 * @dev is idle and it can try to enter the idle state corresponding to @index.
153 * If the local APIC timer is not known to be reliable in the target idle state,
154 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
156 * Must be called under local_irq_disable().
158 static __cpuidle int intel_idle(struct cpuidle_device *dev,
159 struct cpuidle_driver *drv, int index)
161 return __intel_idle(dev, drv, index);
164 static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
165 struct cpuidle_driver *drv, int index)
169 raw_local_irq_enable();
170 ret = __intel_idle(dev, drv, index);
173 * The lockdep hardirqs state may be changed to 'on' with timer
174 * tick interrupt followed by __do_softirq(). Use local_irq_disable()
175 * to keep the hardirqs state correct.
182 static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
183 struct cpuidle_driver *drv, int index)
185 bool smt_active = sched_smt_active();
186 u64 spec_ctrl = spec_ctrl_current();
190 wrmsrl(MSR_IA32_SPEC_CTRL, 0);
192 ret = __intel_idle(dev, drv, index);
195 wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl);
200 static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
201 struct cpuidle_driver *drv, int index)
204 return __intel_idle(dev, drv, index);
208 * intel_idle_s2idle - Ask the processor to enter the given idle state.
209 * @dev: cpuidle device of the target CPU.
210 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
211 * @index: Target idle state index.
213 * Use the MWAIT instruction to notify the processor that the CPU represented by
214 * @dev is idle and it can try to enter the idle state corresponding to @index.
216 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
217 * scheduler tick and suspended scheduler clock on the target CPU.
219 static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
220 struct cpuidle_driver *drv, int index)
222 unsigned long ecx = 1; /* break on interrupt flag */
223 struct cpuidle_state *state = &drv->states[index];
224 unsigned long eax = flg2MWAIT(state->flags);
226 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
229 mwait_idle_with_hints(eax, ecx);
235 * States are indexed by the cstate number,
236 * which is also the index into the MWAIT hint array.
237 * Thus C0 is a dummy.
239 static struct cpuidle_state nehalem_cstates[] __initdata = {
242 .desc = "MWAIT 0x00",
243 .flags = MWAIT2flg(0x00),
245 .target_residency = 6,
246 .enter = &intel_idle,
247 .enter_s2idle = intel_idle_s2idle, },
250 .desc = "MWAIT 0x01",
251 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
253 .target_residency = 20,
254 .enter = &intel_idle,
255 .enter_s2idle = intel_idle_s2idle, },
258 .desc = "MWAIT 0x10",
259 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
261 .target_residency = 80,
262 .enter = &intel_idle,
263 .enter_s2idle = intel_idle_s2idle, },
266 .desc = "MWAIT 0x20",
267 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
269 .target_residency = 800,
270 .enter = &intel_idle,
271 .enter_s2idle = intel_idle_s2idle, },
276 static struct cpuidle_state snb_cstates[] __initdata = {
279 .desc = "MWAIT 0x00",
280 .flags = MWAIT2flg(0x00),
282 .target_residency = 2,
283 .enter = &intel_idle,
284 .enter_s2idle = intel_idle_s2idle, },
287 .desc = "MWAIT 0x01",
288 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
290 .target_residency = 20,
291 .enter = &intel_idle,
292 .enter_s2idle = intel_idle_s2idle, },
295 .desc = "MWAIT 0x10",
296 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
298 .target_residency = 211,
299 .enter = &intel_idle,
300 .enter_s2idle = intel_idle_s2idle, },
303 .desc = "MWAIT 0x20",
304 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
306 .target_residency = 345,
307 .enter = &intel_idle,
308 .enter_s2idle = intel_idle_s2idle, },
311 .desc = "MWAIT 0x30",
312 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
314 .target_residency = 345,
315 .enter = &intel_idle,
316 .enter_s2idle = intel_idle_s2idle, },
321 static struct cpuidle_state byt_cstates[] __initdata = {
324 .desc = "MWAIT 0x00",
325 .flags = MWAIT2flg(0x00),
327 .target_residency = 1,
328 .enter = &intel_idle,
329 .enter_s2idle = intel_idle_s2idle, },
332 .desc = "MWAIT 0x58",
333 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
335 .target_residency = 275,
336 .enter = &intel_idle,
337 .enter_s2idle = intel_idle_s2idle, },
340 .desc = "MWAIT 0x52",
341 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
343 .target_residency = 560,
344 .enter = &intel_idle,
345 .enter_s2idle = intel_idle_s2idle, },
348 .desc = "MWAIT 0x60",
349 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
350 .exit_latency = 1200,
351 .target_residency = 4000,
352 .enter = &intel_idle,
353 .enter_s2idle = intel_idle_s2idle, },
356 .desc = "MWAIT 0x64",
357 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
358 .exit_latency = 10000,
359 .target_residency = 20000,
360 .enter = &intel_idle,
361 .enter_s2idle = intel_idle_s2idle, },
366 static struct cpuidle_state cht_cstates[] __initdata = {
369 .desc = "MWAIT 0x00",
370 .flags = MWAIT2flg(0x00),
372 .target_residency = 1,
373 .enter = &intel_idle,
374 .enter_s2idle = intel_idle_s2idle, },
377 .desc = "MWAIT 0x58",
378 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
380 .target_residency = 275,
381 .enter = &intel_idle,
382 .enter_s2idle = intel_idle_s2idle, },
385 .desc = "MWAIT 0x52",
386 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
388 .target_residency = 560,
389 .enter = &intel_idle,
390 .enter_s2idle = intel_idle_s2idle, },
393 .desc = "MWAIT 0x60",
394 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
395 .exit_latency = 1200,
396 .target_residency = 4000,
397 .enter = &intel_idle,
398 .enter_s2idle = intel_idle_s2idle, },
401 .desc = "MWAIT 0x64",
402 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
403 .exit_latency = 10000,
404 .target_residency = 20000,
405 .enter = &intel_idle,
406 .enter_s2idle = intel_idle_s2idle, },
411 static struct cpuidle_state ivb_cstates[] __initdata = {
414 .desc = "MWAIT 0x00",
415 .flags = MWAIT2flg(0x00),
417 .target_residency = 1,
418 .enter = &intel_idle,
419 .enter_s2idle = intel_idle_s2idle, },
422 .desc = "MWAIT 0x01",
423 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
425 .target_residency = 20,
426 .enter = &intel_idle,
427 .enter_s2idle = intel_idle_s2idle, },
430 .desc = "MWAIT 0x10",
431 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
433 .target_residency = 156,
434 .enter = &intel_idle,
435 .enter_s2idle = intel_idle_s2idle, },
438 .desc = "MWAIT 0x20",
439 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
441 .target_residency = 300,
442 .enter = &intel_idle,
443 .enter_s2idle = intel_idle_s2idle, },
446 .desc = "MWAIT 0x30",
447 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
449 .target_residency = 300,
450 .enter = &intel_idle,
451 .enter_s2idle = intel_idle_s2idle, },
456 static struct cpuidle_state ivt_cstates[] __initdata = {
459 .desc = "MWAIT 0x00",
460 .flags = MWAIT2flg(0x00),
462 .target_residency = 1,
463 .enter = &intel_idle,
464 .enter_s2idle = intel_idle_s2idle, },
467 .desc = "MWAIT 0x01",
468 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
470 .target_residency = 80,
471 .enter = &intel_idle,
472 .enter_s2idle = intel_idle_s2idle, },
475 .desc = "MWAIT 0x10",
476 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
478 .target_residency = 156,
479 .enter = &intel_idle,
480 .enter_s2idle = intel_idle_s2idle, },
483 .desc = "MWAIT 0x20",
484 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
486 .target_residency = 300,
487 .enter = &intel_idle,
488 .enter_s2idle = intel_idle_s2idle, },
493 static struct cpuidle_state ivt_cstates_4s[] __initdata = {
496 .desc = "MWAIT 0x00",
497 .flags = MWAIT2flg(0x00),
499 .target_residency = 1,
500 .enter = &intel_idle,
501 .enter_s2idle = intel_idle_s2idle, },
504 .desc = "MWAIT 0x01",
505 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
507 .target_residency = 250,
508 .enter = &intel_idle,
509 .enter_s2idle = intel_idle_s2idle, },
512 .desc = "MWAIT 0x10",
513 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
515 .target_residency = 300,
516 .enter = &intel_idle,
517 .enter_s2idle = intel_idle_s2idle, },
520 .desc = "MWAIT 0x20",
521 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
523 .target_residency = 400,
524 .enter = &intel_idle,
525 .enter_s2idle = intel_idle_s2idle, },
530 static struct cpuidle_state ivt_cstates_8s[] __initdata = {
533 .desc = "MWAIT 0x00",
534 .flags = MWAIT2flg(0x00),
536 .target_residency = 1,
537 .enter = &intel_idle,
538 .enter_s2idle = intel_idle_s2idle, },
541 .desc = "MWAIT 0x01",
542 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
544 .target_residency = 500,
545 .enter = &intel_idle,
546 .enter_s2idle = intel_idle_s2idle, },
549 .desc = "MWAIT 0x10",
550 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
552 .target_residency = 600,
553 .enter = &intel_idle,
554 .enter_s2idle = intel_idle_s2idle, },
557 .desc = "MWAIT 0x20",
558 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
560 .target_residency = 700,
561 .enter = &intel_idle,
562 .enter_s2idle = intel_idle_s2idle, },
567 static struct cpuidle_state hsw_cstates[] __initdata = {
570 .desc = "MWAIT 0x00",
571 .flags = MWAIT2flg(0x00),
573 .target_residency = 2,
574 .enter = &intel_idle,
575 .enter_s2idle = intel_idle_s2idle, },
578 .desc = "MWAIT 0x01",
579 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
581 .target_residency = 20,
582 .enter = &intel_idle,
583 .enter_s2idle = intel_idle_s2idle, },
586 .desc = "MWAIT 0x10",
587 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
589 .target_residency = 100,
590 .enter = &intel_idle,
591 .enter_s2idle = intel_idle_s2idle, },
594 .desc = "MWAIT 0x20",
595 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
597 .target_residency = 400,
598 .enter = &intel_idle,
599 .enter_s2idle = intel_idle_s2idle, },
602 .desc = "MWAIT 0x32",
603 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
605 .target_residency = 500,
606 .enter = &intel_idle,
607 .enter_s2idle = intel_idle_s2idle, },
610 .desc = "MWAIT 0x40",
611 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
613 .target_residency = 900,
614 .enter = &intel_idle,
615 .enter_s2idle = intel_idle_s2idle, },
618 .desc = "MWAIT 0x50",
619 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
621 .target_residency = 1800,
622 .enter = &intel_idle,
623 .enter_s2idle = intel_idle_s2idle, },
626 .desc = "MWAIT 0x60",
627 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
628 .exit_latency = 2600,
629 .target_residency = 7700,
630 .enter = &intel_idle,
631 .enter_s2idle = intel_idle_s2idle, },
635 static struct cpuidle_state bdw_cstates[] __initdata = {
638 .desc = "MWAIT 0x00",
639 .flags = MWAIT2flg(0x00),
641 .target_residency = 2,
642 .enter = &intel_idle,
643 .enter_s2idle = intel_idle_s2idle, },
646 .desc = "MWAIT 0x01",
647 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
649 .target_residency = 20,
650 .enter = &intel_idle,
651 .enter_s2idle = intel_idle_s2idle, },
654 .desc = "MWAIT 0x10",
655 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
657 .target_residency = 100,
658 .enter = &intel_idle,
659 .enter_s2idle = intel_idle_s2idle, },
662 .desc = "MWAIT 0x20",
663 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
665 .target_residency = 400,
666 .enter = &intel_idle,
667 .enter_s2idle = intel_idle_s2idle, },
670 .desc = "MWAIT 0x32",
671 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
673 .target_residency = 500,
674 .enter = &intel_idle,
675 .enter_s2idle = intel_idle_s2idle, },
678 .desc = "MWAIT 0x40",
679 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
681 .target_residency = 900,
682 .enter = &intel_idle,
683 .enter_s2idle = intel_idle_s2idle, },
686 .desc = "MWAIT 0x50",
687 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
689 .target_residency = 1800,
690 .enter = &intel_idle,
691 .enter_s2idle = intel_idle_s2idle, },
694 .desc = "MWAIT 0x60",
695 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
696 .exit_latency = 2600,
697 .target_residency = 7700,
698 .enter = &intel_idle,
699 .enter_s2idle = intel_idle_s2idle, },
704 static struct cpuidle_state skl_cstates[] __initdata = {
707 .desc = "MWAIT 0x00",
708 .flags = MWAIT2flg(0x00),
710 .target_residency = 2,
711 .enter = &intel_idle,
712 .enter_s2idle = intel_idle_s2idle, },
715 .desc = "MWAIT 0x01",
716 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
718 .target_residency = 20,
719 .enter = &intel_idle,
720 .enter_s2idle = intel_idle_s2idle, },
723 .desc = "MWAIT 0x10",
724 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
726 .target_residency = 100,
727 .enter = &intel_idle,
728 .enter_s2idle = intel_idle_s2idle, },
731 .desc = "MWAIT 0x20",
732 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
734 .target_residency = 200,
735 .enter = &intel_idle,
736 .enter_s2idle = intel_idle_s2idle, },
739 .desc = "MWAIT 0x33",
740 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
742 .target_residency = 800,
743 .enter = &intel_idle,
744 .enter_s2idle = intel_idle_s2idle, },
747 .desc = "MWAIT 0x40",
748 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
750 .target_residency = 800,
751 .enter = &intel_idle,
752 .enter_s2idle = intel_idle_s2idle, },
755 .desc = "MWAIT 0x50",
756 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
758 .target_residency = 5000,
759 .enter = &intel_idle,
760 .enter_s2idle = intel_idle_s2idle, },
763 .desc = "MWAIT 0x60",
764 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
766 .target_residency = 5000,
767 .enter = &intel_idle,
768 .enter_s2idle = intel_idle_s2idle, },
773 static struct cpuidle_state skx_cstates[] __initdata = {
776 .desc = "MWAIT 0x00",
777 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
779 .target_residency = 2,
780 .enter = &intel_idle,
781 .enter_s2idle = intel_idle_s2idle, },
784 .desc = "MWAIT 0x01",
785 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
787 .target_residency = 20,
788 .enter = &intel_idle,
789 .enter_s2idle = intel_idle_s2idle, },
792 .desc = "MWAIT 0x20",
793 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
795 .target_residency = 600,
796 .enter = &intel_idle,
797 .enter_s2idle = intel_idle_s2idle, },
802 static struct cpuidle_state icx_cstates[] __initdata = {
805 .desc = "MWAIT 0x00",
806 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
808 .target_residency = 1,
809 .enter = &intel_idle,
810 .enter_s2idle = intel_idle_s2idle, },
813 .desc = "MWAIT 0x01",
814 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
816 .target_residency = 4,
817 .enter = &intel_idle,
818 .enter_s2idle = intel_idle_s2idle, },
821 .desc = "MWAIT 0x20",
822 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
824 .target_residency = 600,
825 .enter = &intel_idle,
826 .enter_s2idle = intel_idle_s2idle, },
832 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
833 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
834 * But in this case there is effectively no C1, because C1 requests are
835 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
836 * and C1E requests end up with C1, so there is effectively no C1E.
838 * By default we enable C1E and disable C1 by marking it with
839 * 'CPUIDLE_FLAG_UNUSABLE'.
841 static struct cpuidle_state adl_cstates[] __initdata = {
844 .desc = "MWAIT 0x00",
845 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
847 .target_residency = 1,
848 .enter = &intel_idle,
849 .enter_s2idle = intel_idle_s2idle, },
852 .desc = "MWAIT 0x01",
853 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
855 .target_residency = 4,
856 .enter = &intel_idle,
857 .enter_s2idle = intel_idle_s2idle, },
860 .desc = "MWAIT 0x20",
861 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
863 .target_residency = 600,
864 .enter = &intel_idle,
865 .enter_s2idle = intel_idle_s2idle, },
868 .desc = "MWAIT 0x40",
869 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
871 .target_residency = 800,
872 .enter = &intel_idle,
873 .enter_s2idle = intel_idle_s2idle, },
876 .desc = "MWAIT 0x60",
877 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
879 .target_residency = 2000,
880 .enter = &intel_idle,
881 .enter_s2idle = intel_idle_s2idle, },
886 static struct cpuidle_state adl_l_cstates[] __initdata = {
889 .desc = "MWAIT 0x00",
890 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
892 .target_residency = 1,
893 .enter = &intel_idle,
894 .enter_s2idle = intel_idle_s2idle, },
897 .desc = "MWAIT 0x01",
898 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
900 .target_residency = 4,
901 .enter = &intel_idle,
902 .enter_s2idle = intel_idle_s2idle, },
905 .desc = "MWAIT 0x20",
906 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
908 .target_residency = 500,
909 .enter = &intel_idle,
910 .enter_s2idle = intel_idle_s2idle, },
913 .desc = "MWAIT 0x40",
914 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
916 .target_residency = 600,
917 .enter = &intel_idle,
918 .enter_s2idle = intel_idle_s2idle, },
921 .desc = "MWAIT 0x60",
922 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
924 .target_residency = 700,
925 .enter = &intel_idle,
926 .enter_s2idle = intel_idle_s2idle, },
931 static struct cpuidle_state adl_n_cstates[] __initdata = {
934 .desc = "MWAIT 0x00",
935 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
937 .target_residency = 1,
938 .enter = &intel_idle,
939 .enter_s2idle = intel_idle_s2idle, },
942 .desc = "MWAIT 0x01",
943 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
945 .target_residency = 4,
946 .enter = &intel_idle,
947 .enter_s2idle = intel_idle_s2idle, },
950 .desc = "MWAIT 0x20",
951 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
953 .target_residency = 585,
954 .enter = &intel_idle,
955 .enter_s2idle = intel_idle_s2idle, },
958 .desc = "MWAIT 0x40",
959 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
961 .target_residency = 1040,
962 .enter = &intel_idle,
963 .enter_s2idle = intel_idle_s2idle, },
966 .desc = "MWAIT 0x60",
967 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
969 .target_residency = 1980,
970 .enter = &intel_idle,
971 .enter_s2idle = intel_idle_s2idle, },
976 static struct cpuidle_state spr_cstates[] __initdata = {
979 .desc = "MWAIT 0x00",
980 .flags = MWAIT2flg(0x00),
982 .target_residency = 1,
983 .enter = &intel_idle,
984 .enter_s2idle = intel_idle_s2idle, },
987 .desc = "MWAIT 0x01",
988 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
990 .target_residency = 4,
991 .enter = &intel_idle,
992 .enter_s2idle = intel_idle_s2idle, },
995 .desc = "MWAIT 0x20",
996 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
997 CPUIDLE_FLAG_INIT_XSTATE,
999 .target_residency = 800,
1000 .enter = &intel_idle,
1001 .enter_s2idle = intel_idle_s2idle, },
1006 static struct cpuidle_state atom_cstates[] __initdata = {
1009 .desc = "MWAIT 0x00",
1010 .flags = MWAIT2flg(0x00),
1012 .target_residency = 20,
1013 .enter = &intel_idle,
1014 .enter_s2idle = intel_idle_s2idle, },
1017 .desc = "MWAIT 0x10",
1018 .flags = MWAIT2flg(0x10),
1020 .target_residency = 80,
1021 .enter = &intel_idle,
1022 .enter_s2idle = intel_idle_s2idle, },
1025 .desc = "MWAIT 0x30",
1026 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1027 .exit_latency = 100,
1028 .target_residency = 400,
1029 .enter = &intel_idle,
1030 .enter_s2idle = intel_idle_s2idle, },
1033 .desc = "MWAIT 0x52",
1034 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1035 .exit_latency = 140,
1036 .target_residency = 560,
1037 .enter = &intel_idle,
1038 .enter_s2idle = intel_idle_s2idle, },
1042 static struct cpuidle_state tangier_cstates[] __initdata = {
1045 .desc = "MWAIT 0x00",
1046 .flags = MWAIT2flg(0x00),
1048 .target_residency = 4,
1049 .enter = &intel_idle,
1050 .enter_s2idle = intel_idle_s2idle, },
1053 .desc = "MWAIT 0x30",
1054 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1055 .exit_latency = 100,
1056 .target_residency = 400,
1057 .enter = &intel_idle,
1058 .enter_s2idle = intel_idle_s2idle, },
1061 .desc = "MWAIT 0x52",
1062 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1063 .exit_latency = 140,
1064 .target_residency = 560,
1065 .enter = &intel_idle,
1066 .enter_s2idle = intel_idle_s2idle, },
1069 .desc = "MWAIT 0x60",
1070 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1071 .exit_latency = 1200,
1072 .target_residency = 4000,
1073 .enter = &intel_idle,
1074 .enter_s2idle = intel_idle_s2idle, },
1077 .desc = "MWAIT 0x64",
1078 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1079 .exit_latency = 10000,
1080 .target_residency = 20000,
1081 .enter = &intel_idle,
1082 .enter_s2idle = intel_idle_s2idle, },
1086 static struct cpuidle_state avn_cstates[] __initdata = {
1089 .desc = "MWAIT 0x00",
1090 .flags = MWAIT2flg(0x00),
1092 .target_residency = 2,
1093 .enter = &intel_idle,
1094 .enter_s2idle = intel_idle_s2idle, },
1097 .desc = "MWAIT 0x51",
1098 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1100 .target_residency = 45,
1101 .enter = &intel_idle,
1102 .enter_s2idle = intel_idle_s2idle, },
1106 static struct cpuidle_state knl_cstates[] __initdata = {
1109 .desc = "MWAIT 0x00",
1110 .flags = MWAIT2flg(0x00),
1112 .target_residency = 2,
1113 .enter = &intel_idle,
1114 .enter_s2idle = intel_idle_s2idle },
1117 .desc = "MWAIT 0x10",
1118 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1119 .exit_latency = 120,
1120 .target_residency = 500,
1121 .enter = &intel_idle,
1122 .enter_s2idle = intel_idle_s2idle },
1127 static struct cpuidle_state bxt_cstates[] __initdata = {
1130 .desc = "MWAIT 0x00",
1131 .flags = MWAIT2flg(0x00),
1133 .target_residency = 2,
1134 .enter = &intel_idle,
1135 .enter_s2idle = intel_idle_s2idle, },
1138 .desc = "MWAIT 0x01",
1139 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1141 .target_residency = 20,
1142 .enter = &intel_idle,
1143 .enter_s2idle = intel_idle_s2idle, },
1146 .desc = "MWAIT 0x20",
1147 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1148 .exit_latency = 133,
1149 .target_residency = 133,
1150 .enter = &intel_idle,
1151 .enter_s2idle = intel_idle_s2idle, },
1154 .desc = "MWAIT 0x31",
1155 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1156 .exit_latency = 155,
1157 .target_residency = 155,
1158 .enter = &intel_idle,
1159 .enter_s2idle = intel_idle_s2idle, },
1162 .desc = "MWAIT 0x40",
1163 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1164 .exit_latency = 1000,
1165 .target_residency = 1000,
1166 .enter = &intel_idle,
1167 .enter_s2idle = intel_idle_s2idle, },
1170 .desc = "MWAIT 0x50",
1171 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1172 .exit_latency = 2000,
1173 .target_residency = 2000,
1174 .enter = &intel_idle,
1175 .enter_s2idle = intel_idle_s2idle, },
1178 .desc = "MWAIT 0x60",
1179 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1180 .exit_latency = 10000,
1181 .target_residency = 10000,
1182 .enter = &intel_idle,
1183 .enter_s2idle = intel_idle_s2idle, },
1188 static struct cpuidle_state dnv_cstates[] __initdata = {
1191 .desc = "MWAIT 0x00",
1192 .flags = MWAIT2flg(0x00),
1194 .target_residency = 2,
1195 .enter = &intel_idle,
1196 .enter_s2idle = intel_idle_s2idle, },
1199 .desc = "MWAIT 0x01",
1200 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1202 .target_residency = 20,
1203 .enter = &intel_idle,
1204 .enter_s2idle = intel_idle_s2idle, },
1207 .desc = "MWAIT 0x20",
1208 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1210 .target_residency = 500,
1211 .enter = &intel_idle,
1212 .enter_s2idle = intel_idle_s2idle, },
1218 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1219 * C6, and this is indicated in the CPUID mwait leaf.
1221 static struct cpuidle_state snr_cstates[] __initdata = {
1224 .desc = "MWAIT 0x00",
1225 .flags = MWAIT2flg(0x00),
1227 .target_residency = 2,
1228 .enter = &intel_idle,
1229 .enter_s2idle = intel_idle_s2idle, },
1232 .desc = "MWAIT 0x01",
1233 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1235 .target_residency = 25,
1236 .enter = &intel_idle,
1237 .enter_s2idle = intel_idle_s2idle, },
1240 .desc = "MWAIT 0x20",
1241 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1242 .exit_latency = 130,
1243 .target_residency = 500,
1244 .enter = &intel_idle,
1245 .enter_s2idle = intel_idle_s2idle, },
1250 static const struct idle_cpu idle_cpu_nehalem __initconst = {
1251 .state_table = nehalem_cstates,
1252 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1253 .disable_promotion_to_c1e = true,
1256 static const struct idle_cpu idle_cpu_nhx __initconst = {
1257 .state_table = nehalem_cstates,
1258 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1259 .disable_promotion_to_c1e = true,
1263 static const struct idle_cpu idle_cpu_atom __initconst = {
1264 .state_table = atom_cstates,
1267 static const struct idle_cpu idle_cpu_tangier __initconst = {
1268 .state_table = tangier_cstates,
1271 static const struct idle_cpu idle_cpu_lincroft __initconst = {
1272 .state_table = atom_cstates,
1273 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1276 static const struct idle_cpu idle_cpu_snb __initconst = {
1277 .state_table = snb_cstates,
1278 .disable_promotion_to_c1e = true,
1281 static const struct idle_cpu idle_cpu_snx __initconst = {
1282 .state_table = snb_cstates,
1283 .disable_promotion_to_c1e = true,
1287 static const struct idle_cpu idle_cpu_byt __initconst = {
1288 .state_table = byt_cstates,
1289 .disable_promotion_to_c1e = true,
1290 .byt_auto_demotion_disable_flag = true,
1293 static const struct idle_cpu idle_cpu_cht __initconst = {
1294 .state_table = cht_cstates,
1295 .disable_promotion_to_c1e = true,
1296 .byt_auto_demotion_disable_flag = true,
1299 static const struct idle_cpu idle_cpu_ivb __initconst = {
1300 .state_table = ivb_cstates,
1301 .disable_promotion_to_c1e = true,
1304 static const struct idle_cpu idle_cpu_ivt __initconst = {
1305 .state_table = ivt_cstates,
1306 .disable_promotion_to_c1e = true,
1310 static const struct idle_cpu idle_cpu_hsw __initconst = {
1311 .state_table = hsw_cstates,
1312 .disable_promotion_to_c1e = true,
1315 static const struct idle_cpu idle_cpu_hsx __initconst = {
1316 .state_table = hsw_cstates,
1317 .disable_promotion_to_c1e = true,
1321 static const struct idle_cpu idle_cpu_bdw __initconst = {
1322 .state_table = bdw_cstates,
1323 .disable_promotion_to_c1e = true,
1326 static const struct idle_cpu idle_cpu_bdx __initconst = {
1327 .state_table = bdw_cstates,
1328 .disable_promotion_to_c1e = true,
1332 static const struct idle_cpu idle_cpu_skl __initconst = {
1333 .state_table = skl_cstates,
1334 .disable_promotion_to_c1e = true,
1337 static const struct idle_cpu idle_cpu_skx __initconst = {
1338 .state_table = skx_cstates,
1339 .disable_promotion_to_c1e = true,
1343 static const struct idle_cpu idle_cpu_icx __initconst = {
1344 .state_table = icx_cstates,
1345 .disable_promotion_to_c1e = true,
1349 static const struct idle_cpu idle_cpu_adl __initconst = {
1350 .state_table = adl_cstates,
1353 static const struct idle_cpu idle_cpu_adl_l __initconst = {
1354 .state_table = adl_l_cstates,
1357 static const struct idle_cpu idle_cpu_adl_n __initconst = {
1358 .state_table = adl_n_cstates,
1361 static const struct idle_cpu idle_cpu_spr __initconst = {
1362 .state_table = spr_cstates,
1363 .disable_promotion_to_c1e = true,
1367 static const struct idle_cpu idle_cpu_avn __initconst = {
1368 .state_table = avn_cstates,
1369 .disable_promotion_to_c1e = true,
1373 static const struct idle_cpu idle_cpu_knl __initconst = {
1374 .state_table = knl_cstates,
1378 static const struct idle_cpu idle_cpu_bxt __initconst = {
1379 .state_table = bxt_cstates,
1380 .disable_promotion_to_c1e = true,
1383 static const struct idle_cpu idle_cpu_dnv __initconst = {
1384 .state_table = dnv_cstates,
1385 .disable_promotion_to_c1e = true,
1389 static const struct idle_cpu idle_cpu_snr __initconst = {
1390 .state_table = snr_cstates,
1391 .disable_promotion_to_c1e = true,
1395 static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1396 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EP, &idle_cpu_nhx),
1397 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &idle_cpu_nehalem),
1398 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_G, &idle_cpu_nehalem),
1399 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE, &idle_cpu_nehalem),
1400 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EP, &idle_cpu_nhx),
1401 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EX, &idle_cpu_nhx),
1402 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL, &idle_cpu_atom),
1403 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL_MID, &idle_cpu_lincroft),
1404 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EX, &idle_cpu_nhx),
1405 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE, &idle_cpu_snb),
1406 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &idle_cpu_snx),
1407 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SALTWELL, &idle_cpu_atom),
1408 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT, &idle_cpu_byt),
1409 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1410 X86_MATCH_INTEL_FAM6_MODEL(ATOM_AIRMONT, &idle_cpu_cht),
1411 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE, &idle_cpu_ivb),
1412 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &idle_cpu_ivt),
1413 X86_MATCH_INTEL_FAM6_MODEL(HASWELL, &idle_cpu_hsw),
1414 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &idle_cpu_hsx),
1415 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_L, &idle_cpu_hsw),
1416 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_G, &idle_cpu_hsw),
1417 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_D, &idle_cpu_avn),
1418 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL, &idle_cpu_bdw),
1419 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_G, &idle_cpu_bdw),
1420 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &idle_cpu_bdx),
1421 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &idle_cpu_bdx),
1422 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_L, &idle_cpu_skl),
1423 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE, &idle_cpu_skl),
1424 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE_L, &idle_cpu_skl),
1425 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE, &idle_cpu_skl),
1426 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &idle_cpu_skx),
1427 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &idle_cpu_icx),
1428 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &idle_cpu_icx),
1429 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE, &idle_cpu_adl),
1430 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_L, &idle_cpu_adl_l),
1431 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_N, &idle_cpu_adl_n),
1432 X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &idle_cpu_spr),
1433 X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &idle_cpu_spr),
1434 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &idle_cpu_knl),
1435 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &idle_cpu_knl),
1436 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT, &idle_cpu_bxt),
1437 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1438 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_D, &idle_cpu_dnv),
1439 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, &idle_cpu_snr),
1443 static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1444 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1448 static bool __init intel_idle_max_cstate_reached(int cstate)
1450 if (cstate + 1 > max_cstate) {
1451 pr_info("max_cstate %d reached\n", max_cstate);
1457 static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1459 unsigned long eax = flg2MWAIT(state->flags);
1461 if (boot_cpu_has(X86_FEATURE_ARAT))
1465 * Switch over to one-shot tick broadcast if the target C-state
1466 * is deeper than C1.
1468 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1471 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1472 #include <acpi/processor.h>
1474 static bool no_acpi __read_mostly;
1475 module_param(no_acpi, bool, 0444);
1476 MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1478 static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1479 module_param_named(use_acpi, force_use_acpi, bool, 0444);
1480 MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1482 static struct acpi_processor_power acpi_state_table __initdata;
1485 * intel_idle_cst_usable - Check if the _CST information can be used.
1487 * Check if all of the C-states listed by _CST in the max_cstate range are
1488 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1490 static bool __init intel_idle_cst_usable(void)
1494 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1495 acpi_state_table.count);
1497 for (cstate = 1; cstate < limit; cstate++) {
1498 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1500 if (cx->entry_method != ACPI_CSTATE_FFH)
1507 static bool __init intel_idle_acpi_cst_extract(void)
1512 pr_debug("Not allowed to use ACPI _CST\n");
1516 for_each_possible_cpu(cpu) {
1517 struct acpi_processor *pr = per_cpu(processors, cpu);
1522 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1525 acpi_state_table.count++;
1527 if (!intel_idle_cst_usable())
1530 if (!acpi_processor_claim_cst_control())
1536 acpi_state_table.count = 0;
1537 pr_debug("ACPI _CST not found or not usable\n");
1541 static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1543 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1546 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1547 * the interesting states are ACPI_CSTATE_FFH.
1549 for (cstate = 1; cstate < limit; cstate++) {
1550 struct acpi_processor_cx *cx;
1551 struct cpuidle_state *state;
1553 if (intel_idle_max_cstate_reached(cstate - 1))
1556 cx = &acpi_state_table.states[cstate];
1558 state = &drv->states[drv->state_count++];
1560 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1561 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1562 state->exit_latency = cx->latency;
1564 * For C1-type C-states use the same number for both the exit
1565 * latency and target residency, because that is the case for
1566 * C1 in the majority of the static C-states tables above.
1567 * For the other types of C-states, however, set the target
1568 * residency to 3 times the exit latency which should lead to
1569 * a reasonable balance between energy-efficiency and
1570 * performance in the majority of interesting cases.
1572 state->target_residency = cx->latency;
1573 if (cx->type > ACPI_STATE_C1)
1574 state->target_residency *= 3;
1576 state->flags = MWAIT2flg(cx->address);
1577 if (cx->type > ACPI_STATE_C2)
1578 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1580 if (disabled_states_mask & BIT(cstate))
1581 state->flags |= CPUIDLE_FLAG_OFF;
1583 if (intel_idle_state_needs_timer_stop(state))
1584 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1586 state->enter = intel_idle;
1587 state->enter_s2idle = intel_idle_s2idle;
1591 static bool __init intel_idle_off_by_default(u32 mwait_hint)
1596 * If there are no _CST C-states, do not disable any C-states by
1599 if (!acpi_state_table.count)
1602 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1604 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1605 * the interesting states are ACPI_CSTATE_FFH.
1607 for (cstate = 1; cstate < limit; cstate++) {
1608 if (acpi_state_table.states[cstate].address == mwait_hint)
1613 #else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1614 #define force_use_acpi (false)
1616 static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1617 static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1618 static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
1619 #endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1622 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1624 * Tune IVT multi-socket targets.
1625 * Assumption: num_sockets == (max_package_num + 1).
1627 static void __init ivt_idle_state_table_update(void)
1629 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1630 int cpu, package_num, num_sockets = 1;
1632 for_each_online_cpu(cpu) {
1633 package_num = topology_physical_package_id(cpu);
1634 if (package_num + 1 > num_sockets) {
1635 num_sockets = package_num + 1;
1637 if (num_sockets > 4) {
1638 cpuidle_state_table = ivt_cstates_8s;
1644 if (num_sockets > 2)
1645 cpuidle_state_table = ivt_cstates_4s;
1647 /* else, 1 and 2 socket systems use default ivt_cstates */
1651 * irtl_2_usec - IRTL to microseconds conversion.
1652 * @irtl: IRTL MSR value.
1654 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1656 static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1658 static const unsigned int irtl_ns_units[] __initconst = {
1659 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1661 unsigned long long ns;
1666 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1668 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1672 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1674 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1675 * definitive maximum latency and use the same value for target_residency.
1677 static void __init bxt_idle_state_table_update(void)
1679 unsigned long long msr;
1682 rdmsrl(MSR_PKGC6_IRTL, msr);
1683 usec = irtl_2_usec(msr);
1685 bxt_cstates[2].exit_latency = usec;
1686 bxt_cstates[2].target_residency = usec;
1689 rdmsrl(MSR_PKGC7_IRTL, msr);
1690 usec = irtl_2_usec(msr);
1692 bxt_cstates[3].exit_latency = usec;
1693 bxt_cstates[3].target_residency = usec;
1696 rdmsrl(MSR_PKGC8_IRTL, msr);
1697 usec = irtl_2_usec(msr);
1699 bxt_cstates[4].exit_latency = usec;
1700 bxt_cstates[4].target_residency = usec;
1703 rdmsrl(MSR_PKGC9_IRTL, msr);
1704 usec = irtl_2_usec(msr);
1706 bxt_cstates[5].exit_latency = usec;
1707 bxt_cstates[5].target_residency = usec;
1710 rdmsrl(MSR_PKGC10_IRTL, msr);
1711 usec = irtl_2_usec(msr);
1713 bxt_cstates[6].exit_latency = usec;
1714 bxt_cstates[6].target_residency = usec;
1720 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1722 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1724 static void __init sklh_idle_state_table_update(void)
1726 unsigned long long msr;
1727 unsigned int eax, ebx, ecx, edx;
1730 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1731 if (max_cstate <= 7)
1734 /* if PC10 not present in CPUID.MWAIT.EDX */
1735 if ((mwait_substates & (0xF << 28)) == 0)
1738 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1740 /* PC10 is not enabled in PKG C-state limit */
1741 if ((msr & 0xF) != 8)
1745 cpuid(7, &eax, &ebx, &ecx, &edx);
1747 /* if SGX is present */
1748 if (ebx & (1 << 2)) {
1750 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1752 /* if SGX is enabled */
1753 if (msr & (1 << 18))
1757 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1758 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1762 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1763 * idle states table.
1765 static void __init skx_idle_state_table_update(void)
1767 unsigned long long msr;
1769 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1772 * 000b: C0/C1 (no package C-state support)
1774 * 010b: C6 (non-retention)
1775 * 011b: C6 (retention)
1776 * 111b: No Package C state limits.
1778 if ((msr & 0x7) < 2) {
1780 * Uses the CC6 + PC0 latency and 3 times of
1781 * latency for target_residency if the PC6
1782 * is disabled in BIOS. This is consistent
1783 * with how intel_idle driver uses _CST
1784 * to set the target_residency.
1786 skx_cstates[2].exit_latency = 92;
1787 skx_cstates[2].target_residency = 276;
1792 * adl_idle_state_table_update - Adjust AlderLake idle states table.
1794 static void __init adl_idle_state_table_update(void)
1796 /* Check if user prefers C1 over C1E. */
1797 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
1798 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
1799 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
1801 /* Disable C1E by clearing the "C1E promotion" bit. */
1802 c1e_promotion = C1E_PROMOTION_DISABLE;
1806 /* Make sure C1E is enabled by default */
1807 c1e_promotion = C1E_PROMOTION_ENABLE;
1811 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
1813 static void __init spr_idle_state_table_update(void)
1815 unsigned long long msr;
1818 * By default, the C6 state assumes the worst-case scenario of package
1819 * C6. However, if PC6 is disabled, we update the numbers to match
1822 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1824 /* Limit value 2 and above allow for PC6. */
1825 if ((msr & 0x7) < 2) {
1826 spr_cstates[2].exit_latency = 190;
1827 spr_cstates[2].target_residency = 600;
1831 static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
1833 unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
1834 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
1835 MWAIT_SUBSTATE_MASK;
1837 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
1838 if (num_substates == 0)
1841 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1842 mark_tsc_unstable("TSC halts in idle states deeper than C2");
1847 static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
1851 switch (boot_cpu_data.x86_model) {
1852 case INTEL_FAM6_IVYBRIDGE_X:
1853 ivt_idle_state_table_update();
1855 case INTEL_FAM6_ATOM_GOLDMONT:
1856 case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
1857 bxt_idle_state_table_update();
1859 case INTEL_FAM6_SKYLAKE:
1860 sklh_idle_state_table_update();
1862 case INTEL_FAM6_SKYLAKE_X:
1863 skx_idle_state_table_update();
1865 case INTEL_FAM6_SAPPHIRERAPIDS_X:
1866 case INTEL_FAM6_EMERALDRAPIDS_X:
1867 spr_idle_state_table_update();
1869 case INTEL_FAM6_ALDERLAKE:
1870 case INTEL_FAM6_ALDERLAKE_L:
1871 case INTEL_FAM6_ALDERLAKE_N:
1872 adl_idle_state_table_update();
1876 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1877 unsigned int mwait_hint;
1879 if (intel_idle_max_cstate_reached(cstate))
1882 if (!cpuidle_state_table[cstate].enter &&
1883 !cpuidle_state_table[cstate].enter_s2idle)
1886 /* If marked as unusable, skip this state. */
1887 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
1888 pr_debug("state %s is disabled\n",
1889 cpuidle_state_table[cstate].name);
1893 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1894 if (!intel_idle_verify_cstate(mwait_hint))
1897 /* Structure copy. */
1898 drv->states[drv->state_count] = cpuidle_state_table[cstate];
1900 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IRQ_ENABLE)
1901 drv->states[drv->state_count].enter = intel_idle_irq;
1903 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
1904 cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IBRS) {
1905 WARN_ON_ONCE(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IRQ_ENABLE);
1906 drv->states[drv->state_count].enter = intel_idle_ibrs;
1909 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_INIT_XSTATE)
1910 drv->states[drv->state_count].enter = intel_idle_xstate;
1912 if ((disabled_states_mask & BIT(drv->state_count)) ||
1913 ((icpu->use_acpi || force_use_acpi) &&
1914 intel_idle_off_by_default(mwait_hint) &&
1915 !(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
1916 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;
1918 if (intel_idle_state_needs_timer_stop(&drv->states[drv->state_count]))
1919 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_TIMER_STOP;
1924 if (icpu->byt_auto_demotion_disable_flag) {
1925 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
1926 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
1931 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1932 * @drv: cpuidle driver structure to initialize.
1934 static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
1936 cpuidle_poll_state_init(drv);
1938 if (disabled_states_mask & BIT(0))
1939 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
1941 drv->state_count = 1;
1944 intel_idle_init_cstates_icpu(drv);
1946 intel_idle_init_cstates_acpi(drv);
1949 static void auto_demotion_disable(void)
1951 unsigned long long msr_bits;
1953 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1954 msr_bits &= ~auto_demotion_disable_flags;
1955 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1958 static void c1e_promotion_enable(void)
1960 unsigned long long msr_bits;
1962 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1964 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1967 static void c1e_promotion_disable(void)
1969 unsigned long long msr_bits;
1971 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1973 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1977 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
1978 * @cpu: CPU to initialize.
1980 * Register a cpuidle device object for @cpu and update its MSRs in accordance
1981 * with the processor model flags.
1983 static int intel_idle_cpu_init(unsigned int cpu)
1985 struct cpuidle_device *dev;
1987 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1990 if (cpuidle_register_device(dev)) {
1991 pr_debug("cpuidle_register_device %d failed!\n", cpu);
1995 if (auto_demotion_disable_flags)
1996 auto_demotion_disable();
1998 if (c1e_promotion == C1E_PROMOTION_ENABLE)
1999 c1e_promotion_enable();
2000 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
2001 c1e_promotion_disable();
2006 static int intel_idle_cpu_online(unsigned int cpu)
2008 struct cpuidle_device *dev;
2010 if (!boot_cpu_has(X86_FEATURE_ARAT))
2011 tick_broadcast_enable();
2014 * Some systems can hotplug a cpu at runtime after
2015 * the kernel has booted, we have to initialize the
2016 * driver in this case
2018 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2019 if (!dev->registered)
2020 return intel_idle_cpu_init(cpu);
2026 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2028 static void __init intel_idle_cpuidle_devices_uninit(void)
2032 for_each_online_cpu(i)
2033 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2036 static int __init intel_idle_init(void)
2038 const struct x86_cpu_id *id;
2039 unsigned int eax, ebx, ecx;
2042 /* Do not load intel_idle at all for now if idle= is passed */
2043 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2046 if (max_cstate == 0) {
2047 pr_debug("disabled\n");
2051 id = x86_match_cpu(intel_idle_ids);
2053 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2054 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2058 id = x86_match_cpu(intel_mwait_ids);
2063 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2066 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
2068 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2069 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2073 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2075 icpu = (const struct idle_cpu *)id->driver_data;
2077 cpuidle_state_table = icpu->state_table;
2078 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2079 if (icpu->disable_promotion_to_c1e)
2080 c1e_promotion = C1E_PROMOTION_DISABLE;
2081 if (icpu->use_acpi || force_use_acpi)
2082 intel_idle_acpi_cst_extract();
2083 } else if (!intel_idle_acpi_cst_extract()) {
2087 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2088 boot_cpu_data.x86_model);
2090 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2091 if (!intel_idle_cpuidle_devices)
2094 intel_idle_cpuidle_driver_init(&intel_idle_driver);
2096 retval = cpuidle_register_driver(&intel_idle_driver);
2098 struct cpuidle_driver *drv = cpuidle_get_driver();
2099 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2100 drv ? drv->name : "none");
2101 goto init_driver_fail;
2104 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
2105 intel_idle_cpu_online, NULL);
2109 pr_debug("Local APIC timer is reliable in %s\n",
2110 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2115 intel_idle_cpuidle_devices_uninit();
2116 cpuidle_unregister_driver(&intel_idle_driver);
2118 free_percpu(intel_idle_cpuidle_devices);
2122 device_initcall(intel_idle_init);
2125 * We are not really modular, but we used to support that. Meaning we also
2126 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2127 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2128 * is the easiest way (currently) to continue doing that.
2130 module_param(max_cstate, int, 0444);
2132 * The positions of the bits that are set in this number are the indices of the
2133 * idle states to be disabled by default (as reflected by the names of the
2134 * corresponding idle state directories in sysfs, "state0", "state1" ...
2135 * "state<i>" ..., where <i> is the index of the given state).
2137 module_param_named(states_off, disabled_states_mask, uint, 0444);
2138 MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2140 * Some platforms come with mutually exclusive C-states, so that if one is
2141 * enabled, the other C-states must not be used. Example: C1 and C1E on
2142 * Sapphire Rapids platform. This parameter allows for selecting the
2143 * preferred C-states among the groups of mutually exclusive C-states - the
2144 * selected C-states will be registered, the other C-states from the mutually
2145 * exclusive group won't be registered. If the platform has no mutually
2146 * exclusive C-states, this parameter has no effect.
2148 module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2149 MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");