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/mwait.h>
57 #include <asm/spec-ctrl.h>
58 #include <asm/fpu/api.h>
60 #define INTEL_IDLE_VERSION "0.5.1"
62 static struct cpuidle_driver intel_idle_driver = {
66 /* intel_idle.max_cstate=0 disables driver */
67 static int max_cstate = CPUIDLE_STATE_MAX - 1;
68 static unsigned int disabled_states_mask __read_mostly;
69 static unsigned int preferred_states_mask __read_mostly;
70 static bool force_irq_on __read_mostly;
71 static bool ibrs_off __read_mostly;
73 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
75 static unsigned long auto_demotion_disable_flags;
78 C1E_PROMOTION_PRESERVE,
81 } c1e_promotion = C1E_PROMOTION_PRESERVE;
84 struct cpuidle_state *state_table;
87 * Hardware C-state auto-demotion may not always be optimal.
88 * Indicate which enable bits to clear here.
90 unsigned long auto_demotion_disable_flags;
91 bool byt_auto_demotion_disable_flag;
92 bool disable_promotion_to_c1e;
96 static const struct idle_cpu *icpu __initdata;
97 static struct cpuidle_state *cpuidle_state_table __initdata;
99 static unsigned int mwait_substates __initdata;
102 * Enable interrupts before entering the C-state. On some platforms and for
103 * some C-states, this may measurably decrease interrupt latency.
105 #define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
108 * Enable this state by default even if the ACPI _CST does not list it.
110 #define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
113 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
116 #define CPUIDLE_FLAG_IBRS BIT(16)
119 * Initialize large xstate for the C6-state entrance.
121 #define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
124 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
125 * the C-state (top nibble) and sub-state (bottom nibble)
126 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
128 * We store the hint at the top of our "flags" for each state.
130 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
131 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
133 static __always_inline int __intel_idle(struct cpuidle_device *dev,
134 struct cpuidle_driver *drv,
135 int index, bool irqoff)
137 struct cpuidle_state *state = &drv->states[index];
138 unsigned long eax = flg2MWAIT(state->flags);
139 unsigned long ecx = 1*irqoff; /* break on interrupt flag */
141 mwait_idle_with_hints(eax, ecx);
147 * intel_idle - Ask the processor to enter the given idle state.
148 * @dev: cpuidle device of the target CPU.
149 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
150 * @index: Target idle state index.
152 * Use the MWAIT instruction to notify the processor that the CPU represented by
153 * @dev is idle and it can try to enter the idle state corresponding to @index.
155 * If the local APIC timer is not known to be reliable in the target idle state,
156 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
158 * Must be called under local_irq_disable().
160 static __cpuidle int intel_idle(struct cpuidle_device *dev,
161 struct cpuidle_driver *drv, int index)
163 return __intel_idle(dev, drv, index, true);
166 static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
167 struct cpuidle_driver *drv, int index)
169 return __intel_idle(dev, drv, index, false);
172 static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
173 struct cpuidle_driver *drv, int index)
175 bool smt_active = sched_smt_active();
176 u64 spec_ctrl = spec_ctrl_current();
180 __update_spec_ctrl(0);
182 ret = __intel_idle(dev, drv, index, true);
185 __update_spec_ctrl(spec_ctrl);
190 static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
191 struct cpuidle_driver *drv, int index)
194 return __intel_idle(dev, drv, index, true);
198 * intel_idle_s2idle - Ask the processor to enter the given idle state.
199 * @dev: cpuidle device of the target CPU.
200 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
201 * @index: Target idle state index.
203 * Use the MWAIT instruction to notify the processor that the CPU represented by
204 * @dev is idle and it can try to enter the idle state corresponding to @index.
206 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
207 * scheduler tick and suspended scheduler clock on the target CPU.
209 static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
210 struct cpuidle_driver *drv, int index)
212 unsigned long ecx = 1; /* break on interrupt flag */
213 struct cpuidle_state *state = &drv->states[index];
214 unsigned long eax = flg2MWAIT(state->flags);
216 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
219 mwait_idle_with_hints(eax, ecx);
225 * States are indexed by the cstate number,
226 * which is also the index into the MWAIT hint array.
227 * Thus C0 is a dummy.
229 static struct cpuidle_state nehalem_cstates[] __initdata = {
232 .desc = "MWAIT 0x00",
233 .flags = MWAIT2flg(0x00),
235 .target_residency = 6,
236 .enter = &intel_idle,
237 .enter_s2idle = intel_idle_s2idle, },
240 .desc = "MWAIT 0x01",
241 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
243 .target_residency = 20,
244 .enter = &intel_idle,
245 .enter_s2idle = intel_idle_s2idle, },
248 .desc = "MWAIT 0x10",
249 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
251 .target_residency = 80,
252 .enter = &intel_idle,
253 .enter_s2idle = intel_idle_s2idle, },
256 .desc = "MWAIT 0x20",
257 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
259 .target_residency = 800,
260 .enter = &intel_idle,
261 .enter_s2idle = intel_idle_s2idle, },
266 static struct cpuidle_state snb_cstates[] __initdata = {
269 .desc = "MWAIT 0x00",
270 .flags = MWAIT2flg(0x00),
272 .target_residency = 2,
273 .enter = &intel_idle,
274 .enter_s2idle = intel_idle_s2idle, },
277 .desc = "MWAIT 0x01",
278 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
280 .target_residency = 20,
281 .enter = &intel_idle,
282 .enter_s2idle = intel_idle_s2idle, },
285 .desc = "MWAIT 0x10",
286 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
288 .target_residency = 211,
289 .enter = &intel_idle,
290 .enter_s2idle = intel_idle_s2idle, },
293 .desc = "MWAIT 0x20",
294 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
296 .target_residency = 345,
297 .enter = &intel_idle,
298 .enter_s2idle = intel_idle_s2idle, },
301 .desc = "MWAIT 0x30",
302 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
304 .target_residency = 345,
305 .enter = &intel_idle,
306 .enter_s2idle = intel_idle_s2idle, },
311 static struct cpuidle_state byt_cstates[] __initdata = {
314 .desc = "MWAIT 0x00",
315 .flags = MWAIT2flg(0x00),
317 .target_residency = 1,
318 .enter = &intel_idle,
319 .enter_s2idle = intel_idle_s2idle, },
322 .desc = "MWAIT 0x58",
323 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
325 .target_residency = 275,
326 .enter = &intel_idle,
327 .enter_s2idle = intel_idle_s2idle, },
330 .desc = "MWAIT 0x52",
331 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
333 .target_residency = 560,
334 .enter = &intel_idle,
335 .enter_s2idle = intel_idle_s2idle, },
338 .desc = "MWAIT 0x60",
339 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
340 .exit_latency = 1200,
341 .target_residency = 4000,
342 .enter = &intel_idle,
343 .enter_s2idle = intel_idle_s2idle, },
346 .desc = "MWAIT 0x64",
347 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
348 .exit_latency = 10000,
349 .target_residency = 20000,
350 .enter = &intel_idle,
351 .enter_s2idle = intel_idle_s2idle, },
356 static struct cpuidle_state cht_cstates[] __initdata = {
359 .desc = "MWAIT 0x00",
360 .flags = MWAIT2flg(0x00),
362 .target_residency = 1,
363 .enter = &intel_idle,
364 .enter_s2idle = intel_idle_s2idle, },
367 .desc = "MWAIT 0x58",
368 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
370 .target_residency = 275,
371 .enter = &intel_idle,
372 .enter_s2idle = intel_idle_s2idle, },
375 .desc = "MWAIT 0x52",
376 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
378 .target_residency = 560,
379 .enter = &intel_idle,
380 .enter_s2idle = intel_idle_s2idle, },
383 .desc = "MWAIT 0x60",
384 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
385 .exit_latency = 1200,
386 .target_residency = 4000,
387 .enter = &intel_idle,
388 .enter_s2idle = intel_idle_s2idle, },
391 .desc = "MWAIT 0x64",
392 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
393 .exit_latency = 10000,
394 .target_residency = 20000,
395 .enter = &intel_idle,
396 .enter_s2idle = intel_idle_s2idle, },
401 static struct cpuidle_state ivb_cstates[] __initdata = {
404 .desc = "MWAIT 0x00",
405 .flags = MWAIT2flg(0x00),
407 .target_residency = 1,
408 .enter = &intel_idle,
409 .enter_s2idle = intel_idle_s2idle, },
412 .desc = "MWAIT 0x01",
413 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
415 .target_residency = 20,
416 .enter = &intel_idle,
417 .enter_s2idle = intel_idle_s2idle, },
420 .desc = "MWAIT 0x10",
421 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
423 .target_residency = 156,
424 .enter = &intel_idle,
425 .enter_s2idle = intel_idle_s2idle, },
428 .desc = "MWAIT 0x20",
429 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
431 .target_residency = 300,
432 .enter = &intel_idle,
433 .enter_s2idle = intel_idle_s2idle, },
436 .desc = "MWAIT 0x30",
437 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
439 .target_residency = 300,
440 .enter = &intel_idle,
441 .enter_s2idle = intel_idle_s2idle, },
446 static struct cpuidle_state ivt_cstates[] __initdata = {
449 .desc = "MWAIT 0x00",
450 .flags = MWAIT2flg(0x00),
452 .target_residency = 1,
453 .enter = &intel_idle,
454 .enter_s2idle = intel_idle_s2idle, },
457 .desc = "MWAIT 0x01",
458 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
460 .target_residency = 80,
461 .enter = &intel_idle,
462 .enter_s2idle = intel_idle_s2idle, },
465 .desc = "MWAIT 0x10",
466 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
468 .target_residency = 156,
469 .enter = &intel_idle,
470 .enter_s2idle = intel_idle_s2idle, },
473 .desc = "MWAIT 0x20",
474 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
476 .target_residency = 300,
477 .enter = &intel_idle,
478 .enter_s2idle = intel_idle_s2idle, },
483 static struct cpuidle_state ivt_cstates_4s[] __initdata = {
486 .desc = "MWAIT 0x00",
487 .flags = MWAIT2flg(0x00),
489 .target_residency = 1,
490 .enter = &intel_idle,
491 .enter_s2idle = intel_idle_s2idle, },
494 .desc = "MWAIT 0x01",
495 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
497 .target_residency = 250,
498 .enter = &intel_idle,
499 .enter_s2idle = intel_idle_s2idle, },
502 .desc = "MWAIT 0x10",
503 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
505 .target_residency = 300,
506 .enter = &intel_idle,
507 .enter_s2idle = intel_idle_s2idle, },
510 .desc = "MWAIT 0x20",
511 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
513 .target_residency = 400,
514 .enter = &intel_idle,
515 .enter_s2idle = intel_idle_s2idle, },
520 static struct cpuidle_state ivt_cstates_8s[] __initdata = {
523 .desc = "MWAIT 0x00",
524 .flags = MWAIT2flg(0x00),
526 .target_residency = 1,
527 .enter = &intel_idle,
528 .enter_s2idle = intel_idle_s2idle, },
531 .desc = "MWAIT 0x01",
532 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
534 .target_residency = 500,
535 .enter = &intel_idle,
536 .enter_s2idle = intel_idle_s2idle, },
539 .desc = "MWAIT 0x10",
540 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
542 .target_residency = 600,
543 .enter = &intel_idle,
544 .enter_s2idle = intel_idle_s2idle, },
547 .desc = "MWAIT 0x20",
548 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
550 .target_residency = 700,
551 .enter = &intel_idle,
552 .enter_s2idle = intel_idle_s2idle, },
557 static struct cpuidle_state hsw_cstates[] __initdata = {
560 .desc = "MWAIT 0x00",
561 .flags = MWAIT2flg(0x00),
563 .target_residency = 2,
564 .enter = &intel_idle,
565 .enter_s2idle = intel_idle_s2idle, },
568 .desc = "MWAIT 0x01",
569 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
571 .target_residency = 20,
572 .enter = &intel_idle,
573 .enter_s2idle = intel_idle_s2idle, },
576 .desc = "MWAIT 0x10",
577 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
579 .target_residency = 100,
580 .enter = &intel_idle,
581 .enter_s2idle = intel_idle_s2idle, },
584 .desc = "MWAIT 0x20",
585 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
587 .target_residency = 400,
588 .enter = &intel_idle,
589 .enter_s2idle = intel_idle_s2idle, },
592 .desc = "MWAIT 0x32",
593 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
595 .target_residency = 500,
596 .enter = &intel_idle,
597 .enter_s2idle = intel_idle_s2idle, },
600 .desc = "MWAIT 0x40",
601 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
603 .target_residency = 900,
604 .enter = &intel_idle,
605 .enter_s2idle = intel_idle_s2idle, },
608 .desc = "MWAIT 0x50",
609 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
611 .target_residency = 1800,
612 .enter = &intel_idle,
613 .enter_s2idle = intel_idle_s2idle, },
616 .desc = "MWAIT 0x60",
617 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
618 .exit_latency = 2600,
619 .target_residency = 7700,
620 .enter = &intel_idle,
621 .enter_s2idle = intel_idle_s2idle, },
625 static struct cpuidle_state bdw_cstates[] __initdata = {
628 .desc = "MWAIT 0x00",
629 .flags = MWAIT2flg(0x00),
631 .target_residency = 2,
632 .enter = &intel_idle,
633 .enter_s2idle = intel_idle_s2idle, },
636 .desc = "MWAIT 0x01",
637 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
639 .target_residency = 20,
640 .enter = &intel_idle,
641 .enter_s2idle = intel_idle_s2idle, },
644 .desc = "MWAIT 0x10",
645 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
647 .target_residency = 100,
648 .enter = &intel_idle,
649 .enter_s2idle = intel_idle_s2idle, },
652 .desc = "MWAIT 0x20",
653 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
655 .target_residency = 400,
656 .enter = &intel_idle,
657 .enter_s2idle = intel_idle_s2idle, },
660 .desc = "MWAIT 0x32",
661 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
663 .target_residency = 500,
664 .enter = &intel_idle,
665 .enter_s2idle = intel_idle_s2idle, },
668 .desc = "MWAIT 0x40",
669 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
671 .target_residency = 900,
672 .enter = &intel_idle,
673 .enter_s2idle = intel_idle_s2idle, },
676 .desc = "MWAIT 0x50",
677 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
679 .target_residency = 1800,
680 .enter = &intel_idle,
681 .enter_s2idle = intel_idle_s2idle, },
684 .desc = "MWAIT 0x60",
685 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
686 .exit_latency = 2600,
687 .target_residency = 7700,
688 .enter = &intel_idle,
689 .enter_s2idle = intel_idle_s2idle, },
694 static struct cpuidle_state skl_cstates[] __initdata = {
697 .desc = "MWAIT 0x00",
698 .flags = MWAIT2flg(0x00),
700 .target_residency = 2,
701 .enter = &intel_idle,
702 .enter_s2idle = intel_idle_s2idle, },
705 .desc = "MWAIT 0x01",
706 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
708 .target_residency = 20,
709 .enter = &intel_idle,
710 .enter_s2idle = intel_idle_s2idle, },
713 .desc = "MWAIT 0x10",
714 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
716 .target_residency = 100,
717 .enter = &intel_idle,
718 .enter_s2idle = intel_idle_s2idle, },
721 .desc = "MWAIT 0x20",
722 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
724 .target_residency = 200,
725 .enter = &intel_idle,
726 .enter_s2idle = intel_idle_s2idle, },
729 .desc = "MWAIT 0x33",
730 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
732 .target_residency = 800,
733 .enter = &intel_idle,
734 .enter_s2idle = intel_idle_s2idle, },
737 .desc = "MWAIT 0x40",
738 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
740 .target_residency = 800,
741 .enter = &intel_idle,
742 .enter_s2idle = intel_idle_s2idle, },
745 .desc = "MWAIT 0x50",
746 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
748 .target_residency = 5000,
749 .enter = &intel_idle,
750 .enter_s2idle = intel_idle_s2idle, },
753 .desc = "MWAIT 0x60",
754 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
756 .target_residency = 5000,
757 .enter = &intel_idle,
758 .enter_s2idle = intel_idle_s2idle, },
763 static struct cpuidle_state skx_cstates[] __initdata = {
766 .desc = "MWAIT 0x00",
767 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
769 .target_residency = 2,
770 .enter = &intel_idle,
771 .enter_s2idle = intel_idle_s2idle, },
774 .desc = "MWAIT 0x01",
775 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
777 .target_residency = 20,
778 .enter = &intel_idle,
779 .enter_s2idle = intel_idle_s2idle, },
782 .desc = "MWAIT 0x20",
783 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
785 .target_residency = 600,
786 .enter = &intel_idle,
787 .enter_s2idle = intel_idle_s2idle, },
792 static struct cpuidle_state icx_cstates[] __initdata = {
795 .desc = "MWAIT 0x00",
796 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
798 .target_residency = 1,
799 .enter = &intel_idle,
800 .enter_s2idle = intel_idle_s2idle, },
803 .desc = "MWAIT 0x01",
804 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
806 .target_residency = 4,
807 .enter = &intel_idle,
808 .enter_s2idle = intel_idle_s2idle, },
811 .desc = "MWAIT 0x20",
812 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
814 .target_residency = 600,
815 .enter = &intel_idle,
816 .enter_s2idle = intel_idle_s2idle, },
822 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
823 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
824 * But in this case there is effectively no C1, because C1 requests are
825 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
826 * and C1E requests end up with C1, so there is effectively no C1E.
828 * By default we enable C1E and disable C1 by marking it with
829 * 'CPUIDLE_FLAG_UNUSABLE'.
831 static struct cpuidle_state adl_cstates[] __initdata = {
834 .desc = "MWAIT 0x00",
835 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
837 .target_residency = 1,
838 .enter = &intel_idle,
839 .enter_s2idle = intel_idle_s2idle, },
842 .desc = "MWAIT 0x01",
843 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
845 .target_residency = 4,
846 .enter = &intel_idle,
847 .enter_s2idle = intel_idle_s2idle, },
850 .desc = "MWAIT 0x20",
851 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
853 .target_residency = 600,
854 .enter = &intel_idle,
855 .enter_s2idle = intel_idle_s2idle, },
858 .desc = "MWAIT 0x40",
859 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
861 .target_residency = 800,
862 .enter = &intel_idle,
863 .enter_s2idle = intel_idle_s2idle, },
866 .desc = "MWAIT 0x60",
867 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
869 .target_residency = 2000,
870 .enter = &intel_idle,
871 .enter_s2idle = intel_idle_s2idle, },
876 static struct cpuidle_state adl_l_cstates[] __initdata = {
879 .desc = "MWAIT 0x00",
880 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
882 .target_residency = 1,
883 .enter = &intel_idle,
884 .enter_s2idle = intel_idle_s2idle, },
887 .desc = "MWAIT 0x01",
888 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
890 .target_residency = 4,
891 .enter = &intel_idle,
892 .enter_s2idle = intel_idle_s2idle, },
895 .desc = "MWAIT 0x20",
896 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
898 .target_residency = 500,
899 .enter = &intel_idle,
900 .enter_s2idle = intel_idle_s2idle, },
903 .desc = "MWAIT 0x40",
904 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
906 .target_residency = 600,
907 .enter = &intel_idle,
908 .enter_s2idle = intel_idle_s2idle, },
911 .desc = "MWAIT 0x60",
912 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
914 .target_residency = 700,
915 .enter = &intel_idle,
916 .enter_s2idle = intel_idle_s2idle, },
921 static struct cpuidle_state gmt_cstates[] __initdata = {
924 .desc = "MWAIT 0x00",
925 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
927 .target_residency = 1,
928 .enter = &intel_idle,
929 .enter_s2idle = intel_idle_s2idle, },
932 .desc = "MWAIT 0x01",
933 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
935 .target_residency = 4,
936 .enter = &intel_idle,
937 .enter_s2idle = intel_idle_s2idle, },
940 .desc = "MWAIT 0x20",
941 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
943 .target_residency = 585,
944 .enter = &intel_idle,
945 .enter_s2idle = intel_idle_s2idle, },
948 .desc = "MWAIT 0x40",
949 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
951 .target_residency = 1040,
952 .enter = &intel_idle,
953 .enter_s2idle = intel_idle_s2idle, },
956 .desc = "MWAIT 0x60",
957 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
959 .target_residency = 1980,
960 .enter = &intel_idle,
961 .enter_s2idle = intel_idle_s2idle, },
966 static struct cpuidle_state spr_cstates[] __initdata = {
969 .desc = "MWAIT 0x00",
970 .flags = MWAIT2flg(0x00),
972 .target_residency = 1,
973 .enter = &intel_idle,
974 .enter_s2idle = intel_idle_s2idle, },
977 .desc = "MWAIT 0x01",
978 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
980 .target_residency = 4,
981 .enter = &intel_idle,
982 .enter_s2idle = intel_idle_s2idle, },
985 .desc = "MWAIT 0x20",
986 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
987 CPUIDLE_FLAG_INIT_XSTATE,
989 .target_residency = 800,
990 .enter = &intel_idle,
991 .enter_s2idle = intel_idle_s2idle, },
996 static struct cpuidle_state atom_cstates[] __initdata = {
999 .desc = "MWAIT 0x00",
1000 .flags = MWAIT2flg(0x00),
1002 .target_residency = 20,
1003 .enter = &intel_idle,
1004 .enter_s2idle = intel_idle_s2idle, },
1007 .desc = "MWAIT 0x10",
1008 .flags = MWAIT2flg(0x10),
1010 .target_residency = 80,
1011 .enter = &intel_idle,
1012 .enter_s2idle = intel_idle_s2idle, },
1015 .desc = "MWAIT 0x30",
1016 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1017 .exit_latency = 100,
1018 .target_residency = 400,
1019 .enter = &intel_idle,
1020 .enter_s2idle = intel_idle_s2idle, },
1023 .desc = "MWAIT 0x52",
1024 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1025 .exit_latency = 140,
1026 .target_residency = 560,
1027 .enter = &intel_idle,
1028 .enter_s2idle = intel_idle_s2idle, },
1032 static struct cpuidle_state tangier_cstates[] __initdata = {
1035 .desc = "MWAIT 0x00",
1036 .flags = MWAIT2flg(0x00),
1038 .target_residency = 4,
1039 .enter = &intel_idle,
1040 .enter_s2idle = intel_idle_s2idle, },
1043 .desc = "MWAIT 0x30",
1044 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1045 .exit_latency = 100,
1046 .target_residency = 400,
1047 .enter = &intel_idle,
1048 .enter_s2idle = intel_idle_s2idle, },
1051 .desc = "MWAIT 0x52",
1052 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1053 .exit_latency = 140,
1054 .target_residency = 560,
1055 .enter = &intel_idle,
1056 .enter_s2idle = intel_idle_s2idle, },
1059 .desc = "MWAIT 0x60",
1060 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1061 .exit_latency = 1200,
1062 .target_residency = 4000,
1063 .enter = &intel_idle,
1064 .enter_s2idle = intel_idle_s2idle, },
1067 .desc = "MWAIT 0x64",
1068 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1069 .exit_latency = 10000,
1070 .target_residency = 20000,
1071 .enter = &intel_idle,
1072 .enter_s2idle = intel_idle_s2idle, },
1076 static struct cpuidle_state avn_cstates[] __initdata = {
1079 .desc = "MWAIT 0x00",
1080 .flags = MWAIT2flg(0x00),
1082 .target_residency = 2,
1083 .enter = &intel_idle,
1084 .enter_s2idle = intel_idle_s2idle, },
1087 .desc = "MWAIT 0x51",
1088 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1090 .target_residency = 45,
1091 .enter = &intel_idle,
1092 .enter_s2idle = intel_idle_s2idle, },
1096 static struct cpuidle_state knl_cstates[] __initdata = {
1099 .desc = "MWAIT 0x00",
1100 .flags = MWAIT2flg(0x00),
1102 .target_residency = 2,
1103 .enter = &intel_idle,
1104 .enter_s2idle = intel_idle_s2idle },
1107 .desc = "MWAIT 0x10",
1108 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1109 .exit_latency = 120,
1110 .target_residency = 500,
1111 .enter = &intel_idle,
1112 .enter_s2idle = intel_idle_s2idle },
1117 static struct cpuidle_state bxt_cstates[] __initdata = {
1120 .desc = "MWAIT 0x00",
1121 .flags = MWAIT2flg(0x00),
1123 .target_residency = 2,
1124 .enter = &intel_idle,
1125 .enter_s2idle = intel_idle_s2idle, },
1128 .desc = "MWAIT 0x01",
1129 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1131 .target_residency = 20,
1132 .enter = &intel_idle,
1133 .enter_s2idle = intel_idle_s2idle, },
1136 .desc = "MWAIT 0x20",
1137 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1138 .exit_latency = 133,
1139 .target_residency = 133,
1140 .enter = &intel_idle,
1141 .enter_s2idle = intel_idle_s2idle, },
1144 .desc = "MWAIT 0x31",
1145 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1146 .exit_latency = 155,
1147 .target_residency = 155,
1148 .enter = &intel_idle,
1149 .enter_s2idle = intel_idle_s2idle, },
1152 .desc = "MWAIT 0x40",
1153 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1154 .exit_latency = 1000,
1155 .target_residency = 1000,
1156 .enter = &intel_idle,
1157 .enter_s2idle = intel_idle_s2idle, },
1160 .desc = "MWAIT 0x50",
1161 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1162 .exit_latency = 2000,
1163 .target_residency = 2000,
1164 .enter = &intel_idle,
1165 .enter_s2idle = intel_idle_s2idle, },
1168 .desc = "MWAIT 0x60",
1169 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1170 .exit_latency = 10000,
1171 .target_residency = 10000,
1172 .enter = &intel_idle,
1173 .enter_s2idle = intel_idle_s2idle, },
1178 static struct cpuidle_state dnv_cstates[] __initdata = {
1181 .desc = "MWAIT 0x00",
1182 .flags = MWAIT2flg(0x00),
1184 .target_residency = 2,
1185 .enter = &intel_idle,
1186 .enter_s2idle = intel_idle_s2idle, },
1189 .desc = "MWAIT 0x01",
1190 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1192 .target_residency = 20,
1193 .enter = &intel_idle,
1194 .enter_s2idle = intel_idle_s2idle, },
1197 .desc = "MWAIT 0x20",
1198 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1200 .target_residency = 500,
1201 .enter = &intel_idle,
1202 .enter_s2idle = intel_idle_s2idle, },
1208 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1209 * C6, and this is indicated in the CPUID mwait leaf.
1211 static struct cpuidle_state snr_cstates[] __initdata = {
1214 .desc = "MWAIT 0x00",
1215 .flags = MWAIT2flg(0x00),
1217 .target_residency = 2,
1218 .enter = &intel_idle,
1219 .enter_s2idle = intel_idle_s2idle, },
1222 .desc = "MWAIT 0x01",
1223 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1225 .target_residency = 25,
1226 .enter = &intel_idle,
1227 .enter_s2idle = intel_idle_s2idle, },
1230 .desc = "MWAIT 0x20",
1231 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1232 .exit_latency = 130,
1233 .target_residency = 500,
1234 .enter = &intel_idle,
1235 .enter_s2idle = intel_idle_s2idle, },
1240 static const struct idle_cpu idle_cpu_nehalem __initconst = {
1241 .state_table = nehalem_cstates,
1242 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1243 .disable_promotion_to_c1e = true,
1246 static const struct idle_cpu idle_cpu_nhx __initconst = {
1247 .state_table = nehalem_cstates,
1248 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1249 .disable_promotion_to_c1e = true,
1253 static const struct idle_cpu idle_cpu_atom __initconst = {
1254 .state_table = atom_cstates,
1257 static const struct idle_cpu idle_cpu_tangier __initconst = {
1258 .state_table = tangier_cstates,
1261 static const struct idle_cpu idle_cpu_lincroft __initconst = {
1262 .state_table = atom_cstates,
1263 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1266 static const struct idle_cpu idle_cpu_snb __initconst = {
1267 .state_table = snb_cstates,
1268 .disable_promotion_to_c1e = true,
1271 static const struct idle_cpu idle_cpu_snx __initconst = {
1272 .state_table = snb_cstates,
1273 .disable_promotion_to_c1e = true,
1277 static const struct idle_cpu idle_cpu_byt __initconst = {
1278 .state_table = byt_cstates,
1279 .disable_promotion_to_c1e = true,
1280 .byt_auto_demotion_disable_flag = true,
1283 static const struct idle_cpu idle_cpu_cht __initconst = {
1284 .state_table = cht_cstates,
1285 .disable_promotion_to_c1e = true,
1286 .byt_auto_demotion_disable_flag = true,
1289 static const struct idle_cpu idle_cpu_ivb __initconst = {
1290 .state_table = ivb_cstates,
1291 .disable_promotion_to_c1e = true,
1294 static const struct idle_cpu idle_cpu_ivt __initconst = {
1295 .state_table = ivt_cstates,
1296 .disable_promotion_to_c1e = true,
1300 static const struct idle_cpu idle_cpu_hsw __initconst = {
1301 .state_table = hsw_cstates,
1302 .disable_promotion_to_c1e = true,
1305 static const struct idle_cpu idle_cpu_hsx __initconst = {
1306 .state_table = hsw_cstates,
1307 .disable_promotion_to_c1e = true,
1311 static const struct idle_cpu idle_cpu_bdw __initconst = {
1312 .state_table = bdw_cstates,
1313 .disable_promotion_to_c1e = true,
1316 static const struct idle_cpu idle_cpu_bdx __initconst = {
1317 .state_table = bdw_cstates,
1318 .disable_promotion_to_c1e = true,
1322 static const struct idle_cpu idle_cpu_skl __initconst = {
1323 .state_table = skl_cstates,
1324 .disable_promotion_to_c1e = true,
1327 static const struct idle_cpu idle_cpu_skx __initconst = {
1328 .state_table = skx_cstates,
1329 .disable_promotion_to_c1e = true,
1333 static const struct idle_cpu idle_cpu_icx __initconst = {
1334 .state_table = icx_cstates,
1335 .disable_promotion_to_c1e = true,
1339 static const struct idle_cpu idle_cpu_adl __initconst = {
1340 .state_table = adl_cstates,
1343 static const struct idle_cpu idle_cpu_adl_l __initconst = {
1344 .state_table = adl_l_cstates,
1347 static const struct idle_cpu idle_cpu_gmt __initconst = {
1348 .state_table = gmt_cstates,
1351 static const struct idle_cpu idle_cpu_spr __initconst = {
1352 .state_table = spr_cstates,
1353 .disable_promotion_to_c1e = true,
1357 static const struct idle_cpu idle_cpu_avn __initconst = {
1358 .state_table = avn_cstates,
1359 .disable_promotion_to_c1e = true,
1363 static const struct idle_cpu idle_cpu_knl __initconst = {
1364 .state_table = knl_cstates,
1368 static const struct idle_cpu idle_cpu_bxt __initconst = {
1369 .state_table = bxt_cstates,
1370 .disable_promotion_to_c1e = true,
1373 static const struct idle_cpu idle_cpu_dnv __initconst = {
1374 .state_table = dnv_cstates,
1375 .disable_promotion_to_c1e = true,
1379 static const struct idle_cpu idle_cpu_snr __initconst = {
1380 .state_table = snr_cstates,
1381 .disable_promotion_to_c1e = true,
1385 static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1386 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EP, &idle_cpu_nhx),
1387 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &idle_cpu_nehalem),
1388 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_G, &idle_cpu_nehalem),
1389 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE, &idle_cpu_nehalem),
1390 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EP, &idle_cpu_nhx),
1391 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EX, &idle_cpu_nhx),
1392 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL, &idle_cpu_atom),
1393 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL_MID, &idle_cpu_lincroft),
1394 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EX, &idle_cpu_nhx),
1395 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE, &idle_cpu_snb),
1396 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &idle_cpu_snx),
1397 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SALTWELL, &idle_cpu_atom),
1398 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT, &idle_cpu_byt),
1399 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1400 X86_MATCH_INTEL_FAM6_MODEL(ATOM_AIRMONT, &idle_cpu_cht),
1401 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE, &idle_cpu_ivb),
1402 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &idle_cpu_ivt),
1403 X86_MATCH_INTEL_FAM6_MODEL(HASWELL, &idle_cpu_hsw),
1404 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &idle_cpu_hsx),
1405 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_L, &idle_cpu_hsw),
1406 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_G, &idle_cpu_hsw),
1407 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_D, &idle_cpu_avn),
1408 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL, &idle_cpu_bdw),
1409 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_G, &idle_cpu_bdw),
1410 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &idle_cpu_bdx),
1411 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &idle_cpu_bdx),
1412 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_L, &idle_cpu_skl),
1413 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE, &idle_cpu_skl),
1414 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE_L, &idle_cpu_skl),
1415 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE, &idle_cpu_skl),
1416 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &idle_cpu_skx),
1417 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &idle_cpu_icx),
1418 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &idle_cpu_icx),
1419 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE, &idle_cpu_adl),
1420 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_L, &idle_cpu_adl_l),
1421 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GRACEMONT, &idle_cpu_gmt),
1422 X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &idle_cpu_spr),
1423 X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &idle_cpu_spr),
1424 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &idle_cpu_knl),
1425 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &idle_cpu_knl),
1426 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT, &idle_cpu_bxt),
1427 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1428 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_D, &idle_cpu_dnv),
1429 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, &idle_cpu_snr),
1433 static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1434 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1438 static bool __init intel_idle_max_cstate_reached(int cstate)
1440 if (cstate + 1 > max_cstate) {
1441 pr_info("max_cstate %d reached\n", max_cstate);
1447 static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1449 unsigned long eax = flg2MWAIT(state->flags);
1451 if (boot_cpu_has(X86_FEATURE_ARAT))
1455 * Switch over to one-shot tick broadcast if the target C-state
1456 * is deeper than C1.
1458 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1461 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1462 #include <acpi/processor.h>
1464 static bool no_acpi __read_mostly;
1465 module_param(no_acpi, bool, 0444);
1466 MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1468 static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1469 module_param_named(use_acpi, force_use_acpi, bool, 0444);
1470 MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1472 static struct acpi_processor_power acpi_state_table __initdata;
1475 * intel_idle_cst_usable - Check if the _CST information can be used.
1477 * Check if all of the C-states listed by _CST in the max_cstate range are
1478 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1480 static bool __init intel_idle_cst_usable(void)
1484 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1485 acpi_state_table.count);
1487 for (cstate = 1; cstate < limit; cstate++) {
1488 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1490 if (cx->entry_method != ACPI_CSTATE_FFH)
1497 static bool __init intel_idle_acpi_cst_extract(void)
1502 pr_debug("Not allowed to use ACPI _CST\n");
1506 for_each_possible_cpu(cpu) {
1507 struct acpi_processor *pr = per_cpu(processors, cpu);
1512 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1515 acpi_state_table.count++;
1517 if (!intel_idle_cst_usable())
1520 if (!acpi_processor_claim_cst_control())
1526 acpi_state_table.count = 0;
1527 pr_debug("ACPI _CST not found or not usable\n");
1531 static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1533 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1536 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1537 * the interesting states are ACPI_CSTATE_FFH.
1539 for (cstate = 1; cstate < limit; cstate++) {
1540 struct acpi_processor_cx *cx;
1541 struct cpuidle_state *state;
1543 if (intel_idle_max_cstate_reached(cstate - 1))
1546 cx = &acpi_state_table.states[cstate];
1548 state = &drv->states[drv->state_count++];
1550 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1551 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1552 state->exit_latency = cx->latency;
1554 * For C1-type C-states use the same number for both the exit
1555 * latency and target residency, because that is the case for
1556 * C1 in the majority of the static C-states tables above.
1557 * For the other types of C-states, however, set the target
1558 * residency to 3 times the exit latency which should lead to
1559 * a reasonable balance between energy-efficiency and
1560 * performance in the majority of interesting cases.
1562 state->target_residency = cx->latency;
1563 if (cx->type > ACPI_STATE_C1)
1564 state->target_residency *= 3;
1566 state->flags = MWAIT2flg(cx->address);
1567 if (cx->type > ACPI_STATE_C2)
1568 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1570 if (disabled_states_mask & BIT(cstate))
1571 state->flags |= CPUIDLE_FLAG_OFF;
1573 if (intel_idle_state_needs_timer_stop(state))
1574 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1576 state->enter = intel_idle;
1577 state->enter_s2idle = intel_idle_s2idle;
1581 static bool __init intel_idle_off_by_default(u32 mwait_hint)
1586 * If there are no _CST C-states, do not disable any C-states by
1589 if (!acpi_state_table.count)
1592 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1594 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1595 * the interesting states are ACPI_CSTATE_FFH.
1597 for (cstate = 1; cstate < limit; cstate++) {
1598 if (acpi_state_table.states[cstate].address == mwait_hint)
1603 #else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1604 #define force_use_acpi (false)
1606 static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1607 static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1608 static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
1609 #endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1612 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1614 * Tune IVT multi-socket targets.
1615 * Assumption: num_sockets == (max_package_num + 1).
1617 static void __init ivt_idle_state_table_update(void)
1619 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1620 int cpu, package_num, num_sockets = 1;
1622 for_each_online_cpu(cpu) {
1623 package_num = topology_physical_package_id(cpu);
1624 if (package_num + 1 > num_sockets) {
1625 num_sockets = package_num + 1;
1627 if (num_sockets > 4) {
1628 cpuidle_state_table = ivt_cstates_8s;
1634 if (num_sockets > 2)
1635 cpuidle_state_table = ivt_cstates_4s;
1637 /* else, 1 and 2 socket systems use default ivt_cstates */
1641 * irtl_2_usec - IRTL to microseconds conversion.
1642 * @irtl: IRTL MSR value.
1644 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1646 static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1648 static const unsigned int irtl_ns_units[] __initconst = {
1649 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1651 unsigned long long ns;
1656 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1658 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1662 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1664 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1665 * definitive maximum latency and use the same value for target_residency.
1667 static void __init bxt_idle_state_table_update(void)
1669 unsigned long long msr;
1672 rdmsrl(MSR_PKGC6_IRTL, msr);
1673 usec = irtl_2_usec(msr);
1675 bxt_cstates[2].exit_latency = usec;
1676 bxt_cstates[2].target_residency = usec;
1679 rdmsrl(MSR_PKGC7_IRTL, msr);
1680 usec = irtl_2_usec(msr);
1682 bxt_cstates[3].exit_latency = usec;
1683 bxt_cstates[3].target_residency = usec;
1686 rdmsrl(MSR_PKGC8_IRTL, msr);
1687 usec = irtl_2_usec(msr);
1689 bxt_cstates[4].exit_latency = usec;
1690 bxt_cstates[4].target_residency = usec;
1693 rdmsrl(MSR_PKGC9_IRTL, msr);
1694 usec = irtl_2_usec(msr);
1696 bxt_cstates[5].exit_latency = usec;
1697 bxt_cstates[5].target_residency = usec;
1700 rdmsrl(MSR_PKGC10_IRTL, msr);
1701 usec = irtl_2_usec(msr);
1703 bxt_cstates[6].exit_latency = usec;
1704 bxt_cstates[6].target_residency = usec;
1710 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1712 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1714 static void __init sklh_idle_state_table_update(void)
1716 unsigned long long msr;
1717 unsigned int eax, ebx, ecx, edx;
1720 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1721 if (max_cstate <= 7)
1724 /* if PC10 not present in CPUID.MWAIT.EDX */
1725 if ((mwait_substates & (0xF << 28)) == 0)
1728 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1730 /* PC10 is not enabled in PKG C-state limit */
1731 if ((msr & 0xF) != 8)
1735 cpuid(7, &eax, &ebx, &ecx, &edx);
1737 /* if SGX is present */
1738 if (ebx & (1 << 2)) {
1740 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1742 /* if SGX is enabled */
1743 if (msr & (1 << 18))
1747 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1748 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1752 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1753 * idle states table.
1755 static void __init skx_idle_state_table_update(void)
1757 unsigned long long msr;
1759 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1762 * 000b: C0/C1 (no package C-state support)
1764 * 010b: C6 (non-retention)
1765 * 011b: C6 (retention)
1766 * 111b: No Package C state limits.
1768 if ((msr & 0x7) < 2) {
1770 * Uses the CC6 + PC0 latency and 3 times of
1771 * latency for target_residency if the PC6
1772 * is disabled in BIOS. This is consistent
1773 * with how intel_idle driver uses _CST
1774 * to set the target_residency.
1776 skx_cstates[2].exit_latency = 92;
1777 skx_cstates[2].target_residency = 276;
1782 * adl_idle_state_table_update - Adjust AlderLake idle states table.
1784 static void __init adl_idle_state_table_update(void)
1786 /* Check if user prefers C1 over C1E. */
1787 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
1788 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
1789 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
1791 /* Disable C1E by clearing the "C1E promotion" bit. */
1792 c1e_promotion = C1E_PROMOTION_DISABLE;
1796 /* Make sure C1E is enabled by default */
1797 c1e_promotion = C1E_PROMOTION_ENABLE;
1801 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
1803 static void __init spr_idle_state_table_update(void)
1805 unsigned long long msr;
1808 * By default, the C6 state assumes the worst-case scenario of package
1809 * C6. However, if PC6 is disabled, we update the numbers to match
1812 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1814 /* Limit value 2 and above allow for PC6. */
1815 if ((msr & 0x7) < 2) {
1816 spr_cstates[2].exit_latency = 190;
1817 spr_cstates[2].target_residency = 600;
1821 static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
1823 unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
1824 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
1825 MWAIT_SUBSTATE_MASK;
1827 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
1828 if (num_substates == 0)
1831 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1832 mark_tsc_unstable("TSC halts in idle states deeper than C2");
1837 static void state_update_enter_method(struct cpuidle_state *state, int cstate)
1839 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE) {
1841 * Combining with XSTATE with IBRS or IRQ_ENABLE flags
1842 * is not currently supported but this driver.
1844 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IBRS);
1845 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
1846 state->enter = intel_idle_xstate;
1850 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
1851 ((state->flags & CPUIDLE_FLAG_IBRS) || ibrs_off)) {
1853 * IBRS mitigation requires that C-states are entered
1854 * with interrupts disabled.
1856 if (ibrs_off && (state->flags & CPUIDLE_FLAG_IRQ_ENABLE))
1857 state->flags &= ~CPUIDLE_FLAG_IRQ_ENABLE;
1858 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
1859 state->enter = intel_idle_ibrs;
1863 if (state->flags & CPUIDLE_FLAG_IRQ_ENABLE) {
1864 state->enter = intel_idle_irq;
1869 pr_info("forced intel_idle_irq for state %d\n", cstate);
1870 state->enter = intel_idle_irq;
1874 static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
1878 switch (boot_cpu_data.x86_model) {
1879 case INTEL_FAM6_IVYBRIDGE_X:
1880 ivt_idle_state_table_update();
1882 case INTEL_FAM6_ATOM_GOLDMONT:
1883 case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
1884 bxt_idle_state_table_update();
1886 case INTEL_FAM6_SKYLAKE:
1887 sklh_idle_state_table_update();
1889 case INTEL_FAM6_SKYLAKE_X:
1890 skx_idle_state_table_update();
1892 case INTEL_FAM6_SAPPHIRERAPIDS_X:
1893 case INTEL_FAM6_EMERALDRAPIDS_X:
1894 spr_idle_state_table_update();
1896 case INTEL_FAM6_ALDERLAKE:
1897 case INTEL_FAM6_ALDERLAKE_L:
1898 case INTEL_FAM6_ATOM_GRACEMONT:
1899 adl_idle_state_table_update();
1903 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1904 struct cpuidle_state *state;
1905 unsigned int mwait_hint;
1907 if (intel_idle_max_cstate_reached(cstate))
1910 if (!cpuidle_state_table[cstate].enter &&
1911 !cpuidle_state_table[cstate].enter_s2idle)
1914 /* If marked as unusable, skip this state. */
1915 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
1916 pr_debug("state %s is disabled\n",
1917 cpuidle_state_table[cstate].name);
1921 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1922 if (!intel_idle_verify_cstate(mwait_hint))
1925 /* Structure copy. */
1926 drv->states[drv->state_count] = cpuidle_state_table[cstate];
1927 state = &drv->states[drv->state_count];
1929 state_update_enter_method(state, cstate);
1932 if ((disabled_states_mask & BIT(drv->state_count)) ||
1933 ((icpu->use_acpi || force_use_acpi) &&
1934 intel_idle_off_by_default(mwait_hint) &&
1935 !(state->flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
1936 state->flags |= CPUIDLE_FLAG_OFF;
1938 if (intel_idle_state_needs_timer_stop(state))
1939 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1944 if (icpu->byt_auto_demotion_disable_flag) {
1945 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
1946 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
1951 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1952 * @drv: cpuidle driver structure to initialize.
1954 static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
1956 cpuidle_poll_state_init(drv);
1958 if (disabled_states_mask & BIT(0))
1959 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
1961 drv->state_count = 1;
1964 intel_idle_init_cstates_icpu(drv);
1966 intel_idle_init_cstates_acpi(drv);
1969 static void auto_demotion_disable(void)
1971 unsigned long long msr_bits;
1973 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1974 msr_bits &= ~auto_demotion_disable_flags;
1975 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1978 static void c1e_promotion_enable(void)
1980 unsigned long long msr_bits;
1982 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1984 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1987 static void c1e_promotion_disable(void)
1989 unsigned long long msr_bits;
1991 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1993 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1997 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
1998 * @cpu: CPU to initialize.
2000 * Register a cpuidle device object for @cpu and update its MSRs in accordance
2001 * with the processor model flags.
2003 static int intel_idle_cpu_init(unsigned int cpu)
2005 struct cpuidle_device *dev;
2007 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2010 if (cpuidle_register_device(dev)) {
2011 pr_debug("cpuidle_register_device %d failed!\n", cpu);
2015 if (auto_demotion_disable_flags)
2016 auto_demotion_disable();
2018 if (c1e_promotion == C1E_PROMOTION_ENABLE)
2019 c1e_promotion_enable();
2020 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
2021 c1e_promotion_disable();
2026 static int intel_idle_cpu_online(unsigned int cpu)
2028 struct cpuidle_device *dev;
2030 if (!boot_cpu_has(X86_FEATURE_ARAT))
2031 tick_broadcast_enable();
2034 * Some systems can hotplug a cpu at runtime after
2035 * the kernel has booted, we have to initialize the
2036 * driver in this case
2038 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2039 if (!dev->registered)
2040 return intel_idle_cpu_init(cpu);
2046 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2048 static void __init intel_idle_cpuidle_devices_uninit(void)
2052 for_each_online_cpu(i)
2053 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2056 static int __init intel_idle_init(void)
2058 const struct x86_cpu_id *id;
2059 unsigned int eax, ebx, ecx;
2062 /* Do not load intel_idle at all for now if idle= is passed */
2063 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2066 if (max_cstate == 0) {
2067 pr_debug("disabled\n");
2071 id = x86_match_cpu(intel_idle_ids);
2073 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2074 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2078 id = x86_match_cpu(intel_mwait_ids);
2083 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2086 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
2088 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2089 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2093 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2095 icpu = (const struct idle_cpu *)id->driver_data;
2097 cpuidle_state_table = icpu->state_table;
2098 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2099 if (icpu->disable_promotion_to_c1e)
2100 c1e_promotion = C1E_PROMOTION_DISABLE;
2101 if (icpu->use_acpi || force_use_acpi)
2102 intel_idle_acpi_cst_extract();
2103 } else if (!intel_idle_acpi_cst_extract()) {
2107 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2108 boot_cpu_data.x86_model);
2110 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2111 if (!intel_idle_cpuidle_devices)
2114 intel_idle_cpuidle_driver_init(&intel_idle_driver);
2116 retval = cpuidle_register_driver(&intel_idle_driver);
2118 struct cpuidle_driver *drv = cpuidle_get_driver();
2119 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2120 drv ? drv->name : "none");
2121 goto init_driver_fail;
2124 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
2125 intel_idle_cpu_online, NULL);
2129 pr_debug("Local APIC timer is reliable in %s\n",
2130 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2135 intel_idle_cpuidle_devices_uninit();
2136 cpuidle_unregister_driver(&intel_idle_driver);
2138 free_percpu(intel_idle_cpuidle_devices);
2142 device_initcall(intel_idle_init);
2145 * We are not really modular, but we used to support that. Meaning we also
2146 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2147 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2148 * is the easiest way (currently) to continue doing that.
2150 module_param(max_cstate, int, 0444);
2152 * The positions of the bits that are set in this number are the indices of the
2153 * idle states to be disabled by default (as reflected by the names of the
2154 * corresponding idle state directories in sysfs, "state0", "state1" ...
2155 * "state<i>" ..., where <i> is the index of the given state).
2157 module_param_named(states_off, disabled_states_mask, uint, 0444);
2158 MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2160 * Some platforms come with mutually exclusive C-states, so that if one is
2161 * enabled, the other C-states must not be used. Example: C1 and C1E on
2162 * Sapphire Rapids platform. This parameter allows for selecting the
2163 * preferred C-states among the groups of mutually exclusive C-states - the
2164 * selected C-states will be registered, the other C-states from the mutually
2165 * exclusive group won't be registered. If the platform has no mutually
2166 * exclusive C-states, this parameter has no effect.
2168 module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2169 MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");
2171 * Debugging option that forces the driver to enter all C-states with
2172 * interrupts enabled. Does not apply to C-states with
2173 * 'CPUIDLE_FLAG_INIT_XSTATE' and 'CPUIDLE_FLAG_IBRS' flags.
2175 module_param(force_irq_on, bool, 0444);
2177 * Force the disabling of IBRS when X86_FEATURE_KERNEL_IBRS is on and
2178 * CPUIDLE_FLAG_IRQ_ENABLE isn't set.
2180 module_param(ibrs_off, bool, 0444);
2181 MODULE_PARM_DESC(ibrs_off, "Disable IBRS when idle");