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/cpuid.h>
55 #include <asm/cpu_device_id.h>
56 #include <asm/intel-family.h>
57 #include <asm/mwait.h>
58 #include <asm/spec-ctrl.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 __read_mostly
;
70 static unsigned int preferred_states_mask __read_mostly
;
71 static bool force_irq_on __read_mostly
;
72 static bool ibrs_off __read_mostly
;
74 static struct cpuidle_device __percpu
*intel_idle_cpuidle_devices
;
76 static unsigned long auto_demotion_disable_flags
;
79 C1E_PROMOTION_PRESERVE
,
82 } c1e_promotion
= C1E_PROMOTION_PRESERVE
;
85 struct cpuidle_state
*state_table
;
88 * Hardware C-state auto-demotion may not always be optimal.
89 * Indicate which enable bits to clear here.
91 unsigned long auto_demotion_disable_flags
;
92 bool byt_auto_demotion_disable_flag
;
93 bool disable_promotion_to_c1e
;
97 static const struct idle_cpu
*icpu __initdata
;
98 static struct cpuidle_state
*cpuidle_state_table __initdata
;
100 static unsigned int mwait_substates __initdata
;
103 * Enable interrupts before entering the C-state. On some platforms and for
104 * some C-states, this may measurably decrease interrupt latency.
106 #define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
109 * Enable this state by default even if the ACPI _CST does not list it.
111 #define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
114 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
117 #define CPUIDLE_FLAG_IBRS BIT(16)
120 * Initialize large xstate for the C6-state entrance.
122 #define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
125 * Ignore the sub-state when matching mwait hints between the ACPI _CST and
128 #define CPUIDLE_FLAG_PARTIAL_HINT_MATCH BIT(18)
131 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
132 * the C-state (top nibble) and sub-state (bottom nibble)
133 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
135 * We store the hint at the top of our "flags" for each state.
137 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
138 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
140 static __always_inline
int __intel_idle(struct cpuidle_device
*dev
,
141 struct cpuidle_driver
*drv
,
142 int index
, bool irqoff
)
144 struct cpuidle_state
*state
= &drv
->states
[index
];
145 unsigned long eax
= flg2MWAIT(state
->flags
);
146 unsigned long ecx
= 1*irqoff
; /* break on interrupt flag */
148 mwait_idle_with_hints(eax
, ecx
);
154 * intel_idle - Ask the processor to enter the given idle state.
155 * @dev: cpuidle device of the target CPU.
156 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
157 * @index: Target idle state index.
159 * Use the MWAIT instruction to notify the processor that the CPU represented by
160 * @dev is idle and it can try to enter the idle state corresponding to @index.
162 * If the local APIC timer is not known to be reliable in the target idle state,
163 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
165 * Must be called under local_irq_disable().
167 static __cpuidle
int intel_idle(struct cpuidle_device
*dev
,
168 struct cpuidle_driver
*drv
, int index
)
170 return __intel_idle(dev
, drv
, index
, true);
173 static __cpuidle
int intel_idle_irq(struct cpuidle_device
*dev
,
174 struct cpuidle_driver
*drv
, int index
)
176 return __intel_idle(dev
, drv
, index
, false);
179 static __cpuidle
int intel_idle_ibrs(struct cpuidle_device
*dev
,
180 struct cpuidle_driver
*drv
, int index
)
182 bool smt_active
= sched_smt_active();
183 u64 spec_ctrl
= spec_ctrl_current();
187 __update_spec_ctrl(0);
189 ret
= __intel_idle(dev
, drv
, index
, true);
192 __update_spec_ctrl(spec_ctrl
);
197 static __cpuidle
int intel_idle_xstate(struct cpuidle_device
*dev
,
198 struct cpuidle_driver
*drv
, int index
)
201 return __intel_idle(dev
, drv
, index
, true);
205 * intel_idle_s2idle - Ask the processor to enter the given idle state.
206 * @dev: cpuidle device of the target CPU.
207 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
208 * @index: Target idle state index.
210 * Use the MWAIT instruction to notify the processor that the CPU represented by
211 * @dev is idle and it can try to enter the idle state corresponding to @index.
213 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
214 * scheduler tick and suspended scheduler clock on the target CPU.
216 static __cpuidle
int intel_idle_s2idle(struct cpuidle_device
*dev
,
217 struct cpuidle_driver
*drv
, int index
)
219 unsigned long ecx
= 1; /* break on interrupt flag */
220 struct cpuidle_state
*state
= &drv
->states
[index
];
221 unsigned long eax
= flg2MWAIT(state
->flags
);
223 if (state
->flags
& CPUIDLE_FLAG_INIT_XSTATE
)
226 mwait_idle_with_hints(eax
, ecx
);
232 * States are indexed by the cstate number,
233 * which is also the index into the MWAIT hint array.
234 * Thus C0 is a dummy.
236 static struct cpuidle_state nehalem_cstates
[] __initdata
= {
239 .desc
= "MWAIT 0x00",
240 .flags
= MWAIT2flg(0x00),
242 .target_residency
= 6,
243 .enter
= &intel_idle
,
244 .enter_s2idle
= intel_idle_s2idle
, },
247 .desc
= "MWAIT 0x01",
248 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
250 .target_residency
= 20,
251 .enter
= &intel_idle
,
252 .enter_s2idle
= intel_idle_s2idle
, },
255 .desc
= "MWAIT 0x10",
256 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
258 .target_residency
= 80,
259 .enter
= &intel_idle
,
260 .enter_s2idle
= intel_idle_s2idle
, },
263 .desc
= "MWAIT 0x20",
264 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
266 .target_residency
= 800,
267 .enter
= &intel_idle
,
268 .enter_s2idle
= intel_idle_s2idle
, },
273 static struct cpuidle_state snb_cstates
[] __initdata
= {
276 .desc
= "MWAIT 0x00",
277 .flags
= MWAIT2flg(0x00),
279 .target_residency
= 2,
280 .enter
= &intel_idle
,
281 .enter_s2idle
= intel_idle_s2idle
, },
284 .desc
= "MWAIT 0x01",
285 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
287 .target_residency
= 20,
288 .enter
= &intel_idle
,
289 .enter_s2idle
= intel_idle_s2idle
, },
292 .desc
= "MWAIT 0x10",
293 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
295 .target_residency
= 211,
296 .enter
= &intel_idle
,
297 .enter_s2idle
= intel_idle_s2idle
, },
300 .desc
= "MWAIT 0x20",
301 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
303 .target_residency
= 345,
304 .enter
= &intel_idle
,
305 .enter_s2idle
= intel_idle_s2idle
, },
308 .desc
= "MWAIT 0x30",
309 .flags
= MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED
,
311 .target_residency
= 345,
312 .enter
= &intel_idle
,
313 .enter_s2idle
= intel_idle_s2idle
, },
318 static struct cpuidle_state byt_cstates
[] __initdata
= {
321 .desc
= "MWAIT 0x00",
322 .flags
= MWAIT2flg(0x00),
324 .target_residency
= 1,
325 .enter
= &intel_idle
,
326 .enter_s2idle
= intel_idle_s2idle
, },
329 .desc
= "MWAIT 0x58",
330 .flags
= MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED
,
332 .target_residency
= 275,
333 .enter
= &intel_idle
,
334 .enter_s2idle
= intel_idle_s2idle
, },
337 .desc
= "MWAIT 0x52",
338 .flags
= MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED
,
340 .target_residency
= 560,
341 .enter
= &intel_idle
,
342 .enter_s2idle
= intel_idle_s2idle
, },
345 .desc
= "MWAIT 0x60",
346 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
347 .exit_latency
= 1200,
348 .target_residency
= 4000,
349 .enter
= &intel_idle
,
350 .enter_s2idle
= intel_idle_s2idle
, },
353 .desc
= "MWAIT 0x64",
354 .flags
= MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED
,
355 .exit_latency
= 10000,
356 .target_residency
= 20000,
357 .enter
= &intel_idle
,
358 .enter_s2idle
= intel_idle_s2idle
, },
363 static struct cpuidle_state cht_cstates
[] __initdata
= {
366 .desc
= "MWAIT 0x00",
367 .flags
= MWAIT2flg(0x00),
369 .target_residency
= 1,
370 .enter
= &intel_idle
,
371 .enter_s2idle
= intel_idle_s2idle
, },
374 .desc
= "MWAIT 0x58",
375 .flags
= MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED
,
377 .target_residency
= 275,
378 .enter
= &intel_idle
,
379 .enter_s2idle
= intel_idle_s2idle
, },
382 .desc
= "MWAIT 0x52",
383 .flags
= MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED
,
385 .target_residency
= 560,
386 .enter
= &intel_idle
,
387 .enter_s2idle
= intel_idle_s2idle
, },
390 .desc
= "MWAIT 0x60",
391 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
392 .exit_latency
= 1200,
393 .target_residency
= 4000,
394 .enter
= &intel_idle
,
395 .enter_s2idle
= intel_idle_s2idle
, },
398 .desc
= "MWAIT 0x64",
399 .flags
= MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED
,
400 .exit_latency
= 10000,
401 .target_residency
= 20000,
402 .enter
= &intel_idle
,
403 .enter_s2idle
= intel_idle_s2idle
, },
408 static struct cpuidle_state ivb_cstates
[] __initdata
= {
411 .desc
= "MWAIT 0x00",
412 .flags
= MWAIT2flg(0x00),
414 .target_residency
= 1,
415 .enter
= &intel_idle
,
416 .enter_s2idle
= intel_idle_s2idle
, },
419 .desc
= "MWAIT 0x01",
420 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
422 .target_residency
= 20,
423 .enter
= &intel_idle
,
424 .enter_s2idle
= intel_idle_s2idle
, },
427 .desc
= "MWAIT 0x10",
428 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
430 .target_residency
= 156,
431 .enter
= &intel_idle
,
432 .enter_s2idle
= intel_idle_s2idle
, },
435 .desc
= "MWAIT 0x20",
436 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
438 .target_residency
= 300,
439 .enter
= &intel_idle
,
440 .enter_s2idle
= intel_idle_s2idle
, },
443 .desc
= "MWAIT 0x30",
444 .flags
= MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED
,
446 .target_residency
= 300,
447 .enter
= &intel_idle
,
448 .enter_s2idle
= intel_idle_s2idle
, },
453 static struct cpuidle_state ivt_cstates
[] __initdata
= {
456 .desc
= "MWAIT 0x00",
457 .flags
= MWAIT2flg(0x00),
459 .target_residency
= 1,
460 .enter
= &intel_idle
,
461 .enter_s2idle
= intel_idle_s2idle
, },
464 .desc
= "MWAIT 0x01",
465 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
467 .target_residency
= 80,
468 .enter
= &intel_idle
,
469 .enter_s2idle
= intel_idle_s2idle
, },
472 .desc
= "MWAIT 0x10",
473 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
475 .target_residency
= 156,
476 .enter
= &intel_idle
,
477 .enter_s2idle
= intel_idle_s2idle
, },
480 .desc
= "MWAIT 0x20",
481 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
483 .target_residency
= 300,
484 .enter
= &intel_idle
,
485 .enter_s2idle
= intel_idle_s2idle
, },
490 static struct cpuidle_state ivt_cstates_4s
[] __initdata
= {
493 .desc
= "MWAIT 0x00",
494 .flags
= MWAIT2flg(0x00),
496 .target_residency
= 1,
497 .enter
= &intel_idle
,
498 .enter_s2idle
= intel_idle_s2idle
, },
501 .desc
= "MWAIT 0x01",
502 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
504 .target_residency
= 250,
505 .enter
= &intel_idle
,
506 .enter_s2idle
= intel_idle_s2idle
, },
509 .desc
= "MWAIT 0x10",
510 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
512 .target_residency
= 300,
513 .enter
= &intel_idle
,
514 .enter_s2idle
= intel_idle_s2idle
, },
517 .desc
= "MWAIT 0x20",
518 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
520 .target_residency
= 400,
521 .enter
= &intel_idle
,
522 .enter_s2idle
= intel_idle_s2idle
, },
527 static struct cpuidle_state ivt_cstates_8s
[] __initdata
= {
530 .desc
= "MWAIT 0x00",
531 .flags
= MWAIT2flg(0x00),
533 .target_residency
= 1,
534 .enter
= &intel_idle
,
535 .enter_s2idle
= intel_idle_s2idle
, },
538 .desc
= "MWAIT 0x01",
539 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
541 .target_residency
= 500,
542 .enter
= &intel_idle
,
543 .enter_s2idle
= intel_idle_s2idle
, },
546 .desc
= "MWAIT 0x10",
547 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
549 .target_residency
= 600,
550 .enter
= &intel_idle
,
551 .enter_s2idle
= intel_idle_s2idle
, },
554 .desc
= "MWAIT 0x20",
555 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
557 .target_residency
= 700,
558 .enter
= &intel_idle
,
559 .enter_s2idle
= intel_idle_s2idle
, },
564 static struct cpuidle_state hsw_cstates
[] __initdata
= {
567 .desc
= "MWAIT 0x00",
568 .flags
= MWAIT2flg(0x00),
570 .target_residency
= 2,
571 .enter
= &intel_idle
,
572 .enter_s2idle
= intel_idle_s2idle
, },
575 .desc
= "MWAIT 0x01",
576 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
578 .target_residency
= 20,
579 .enter
= &intel_idle
,
580 .enter_s2idle
= intel_idle_s2idle
, },
583 .desc
= "MWAIT 0x10",
584 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
586 .target_residency
= 100,
587 .enter
= &intel_idle
,
588 .enter_s2idle
= intel_idle_s2idle
, },
591 .desc
= "MWAIT 0x20",
592 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
594 .target_residency
= 400,
595 .enter
= &intel_idle
,
596 .enter_s2idle
= intel_idle_s2idle
, },
599 .desc
= "MWAIT 0x32",
600 .flags
= MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED
,
602 .target_residency
= 500,
603 .enter
= &intel_idle
,
604 .enter_s2idle
= intel_idle_s2idle
, },
607 .desc
= "MWAIT 0x40",
608 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
610 .target_residency
= 900,
611 .enter
= &intel_idle
,
612 .enter_s2idle
= intel_idle_s2idle
, },
615 .desc
= "MWAIT 0x50",
616 .flags
= MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED
,
618 .target_residency
= 1800,
619 .enter
= &intel_idle
,
620 .enter_s2idle
= intel_idle_s2idle
, },
623 .desc
= "MWAIT 0x60",
624 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
625 .exit_latency
= 2600,
626 .target_residency
= 7700,
627 .enter
= &intel_idle
,
628 .enter_s2idle
= intel_idle_s2idle
, },
632 static struct cpuidle_state bdw_cstates
[] __initdata
= {
635 .desc
= "MWAIT 0x00",
636 .flags
= MWAIT2flg(0x00),
638 .target_residency
= 2,
639 .enter
= &intel_idle
,
640 .enter_s2idle
= intel_idle_s2idle
, },
643 .desc
= "MWAIT 0x01",
644 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
646 .target_residency
= 20,
647 .enter
= &intel_idle
,
648 .enter_s2idle
= intel_idle_s2idle
, },
651 .desc
= "MWAIT 0x10",
652 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
654 .target_residency
= 100,
655 .enter
= &intel_idle
,
656 .enter_s2idle
= intel_idle_s2idle
, },
659 .desc
= "MWAIT 0x20",
660 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
662 .target_residency
= 400,
663 .enter
= &intel_idle
,
664 .enter_s2idle
= intel_idle_s2idle
, },
667 .desc
= "MWAIT 0x32",
668 .flags
= MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED
,
670 .target_residency
= 500,
671 .enter
= &intel_idle
,
672 .enter_s2idle
= intel_idle_s2idle
, },
675 .desc
= "MWAIT 0x40",
676 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
678 .target_residency
= 900,
679 .enter
= &intel_idle
,
680 .enter_s2idle
= intel_idle_s2idle
, },
683 .desc
= "MWAIT 0x50",
684 .flags
= MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED
,
686 .target_residency
= 1800,
687 .enter
= &intel_idle
,
688 .enter_s2idle
= intel_idle_s2idle
, },
691 .desc
= "MWAIT 0x60",
692 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
693 .exit_latency
= 2600,
694 .target_residency
= 7700,
695 .enter
= &intel_idle
,
696 .enter_s2idle
= intel_idle_s2idle
, },
701 static struct cpuidle_state skl_cstates
[] __initdata
= {
704 .desc
= "MWAIT 0x00",
705 .flags
= MWAIT2flg(0x00),
707 .target_residency
= 2,
708 .enter
= &intel_idle
,
709 .enter_s2idle
= intel_idle_s2idle
, },
712 .desc
= "MWAIT 0x01",
713 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
715 .target_residency
= 20,
716 .enter
= &intel_idle
,
717 .enter_s2idle
= intel_idle_s2idle
, },
720 .desc
= "MWAIT 0x10",
721 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
723 .target_residency
= 100,
724 .enter
= &intel_idle
,
725 .enter_s2idle
= intel_idle_s2idle
, },
728 .desc
= "MWAIT 0x20",
729 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
731 .target_residency
= 200,
732 .enter
= &intel_idle
,
733 .enter_s2idle
= intel_idle_s2idle
, },
736 .desc
= "MWAIT 0x33",
737 .flags
= MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
739 .target_residency
= 800,
740 .enter
= &intel_idle
,
741 .enter_s2idle
= intel_idle_s2idle
, },
744 .desc
= "MWAIT 0x40",
745 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
747 .target_residency
= 800,
748 .enter
= &intel_idle
,
749 .enter_s2idle
= intel_idle_s2idle
, },
752 .desc
= "MWAIT 0x50",
753 .flags
= MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
755 .target_residency
= 5000,
756 .enter
= &intel_idle
,
757 .enter_s2idle
= intel_idle_s2idle
, },
760 .desc
= "MWAIT 0x60",
761 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
763 .target_residency
= 5000,
764 .enter
= &intel_idle
,
765 .enter_s2idle
= intel_idle_s2idle
, },
770 static struct cpuidle_state skx_cstates
[] __initdata
= {
773 .desc
= "MWAIT 0x00",
774 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE
,
776 .target_residency
= 2,
777 .enter
= &intel_idle
,
778 .enter_s2idle
= intel_idle_s2idle
, },
781 .desc
= "MWAIT 0x01",
782 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
784 .target_residency
= 20,
785 .enter
= &intel_idle
,
786 .enter_s2idle
= intel_idle_s2idle
, },
789 .desc
= "MWAIT 0x20",
790 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
| CPUIDLE_FLAG_IBRS
,
792 .target_residency
= 600,
793 .enter
= &intel_idle
,
794 .enter_s2idle
= intel_idle_s2idle
, },
799 static struct cpuidle_state icx_cstates
[] __initdata
= {
802 .desc
= "MWAIT 0x00",
803 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE
,
805 .target_residency
= 1,
806 .enter
= &intel_idle
,
807 .enter_s2idle
= intel_idle_s2idle
, },
810 .desc
= "MWAIT 0x01",
811 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
813 .target_residency
= 4,
814 .enter
= &intel_idle
,
815 .enter_s2idle
= intel_idle_s2idle
, },
818 .desc
= "MWAIT 0x20",
819 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
821 .target_residency
= 600,
822 .enter
= &intel_idle
,
823 .enter_s2idle
= intel_idle_s2idle
, },
829 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
830 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
831 * But in this case there is effectively no C1, because C1 requests are
832 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
833 * and C1E requests end up with C1, so there is effectively no C1E.
835 * By default we enable C1E and disable C1 by marking it with
836 * 'CPUIDLE_FLAG_UNUSABLE'.
838 static struct cpuidle_state adl_cstates
[] __initdata
= {
841 .desc
= "MWAIT 0x00",
842 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE
,
844 .target_residency
= 1,
845 .enter
= &intel_idle
,
846 .enter_s2idle
= intel_idle_s2idle
, },
849 .desc
= "MWAIT 0x01",
850 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
852 .target_residency
= 4,
853 .enter
= &intel_idle
,
854 .enter_s2idle
= intel_idle_s2idle
, },
857 .desc
= "MWAIT 0x20",
858 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
860 .target_residency
= 600,
861 .enter
= &intel_idle
,
862 .enter_s2idle
= intel_idle_s2idle
, },
865 .desc
= "MWAIT 0x40",
866 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
868 .target_residency
= 800,
869 .enter
= &intel_idle
,
870 .enter_s2idle
= intel_idle_s2idle
, },
873 .desc
= "MWAIT 0x60",
874 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
876 .target_residency
= 2000,
877 .enter
= &intel_idle
,
878 .enter_s2idle
= intel_idle_s2idle
, },
883 static struct cpuidle_state adl_l_cstates
[] __initdata
= {
886 .desc
= "MWAIT 0x00",
887 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE
,
889 .target_residency
= 1,
890 .enter
= &intel_idle
,
891 .enter_s2idle
= intel_idle_s2idle
, },
894 .desc
= "MWAIT 0x01",
895 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
897 .target_residency
= 4,
898 .enter
= &intel_idle
,
899 .enter_s2idle
= intel_idle_s2idle
, },
902 .desc
= "MWAIT 0x20",
903 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
905 .target_residency
= 500,
906 .enter
= &intel_idle
,
907 .enter_s2idle
= intel_idle_s2idle
, },
910 .desc
= "MWAIT 0x40",
911 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
913 .target_residency
= 600,
914 .enter
= &intel_idle
,
915 .enter_s2idle
= intel_idle_s2idle
, },
918 .desc
= "MWAIT 0x60",
919 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
921 .target_residency
= 700,
922 .enter
= &intel_idle
,
923 .enter_s2idle
= intel_idle_s2idle
, },
928 static struct cpuidle_state mtl_l_cstates
[] __initdata
= {
931 .desc
= "MWAIT 0x01",
932 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
934 .target_residency
= 1,
935 .enter
= &intel_idle
,
936 .enter_s2idle
= intel_idle_s2idle
, },
939 .desc
= "MWAIT 0x20",
940 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
942 .target_residency
= 420,
943 .enter
= &intel_idle
,
944 .enter_s2idle
= intel_idle_s2idle
, },
947 .desc
= "MWAIT 0x60",
948 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
950 .target_residency
= 930,
951 .enter
= &intel_idle
,
952 .enter_s2idle
= intel_idle_s2idle
, },
957 static struct cpuidle_state gmt_cstates
[] __initdata
= {
960 .desc
= "MWAIT 0x00",
961 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE
,
963 .target_residency
= 1,
964 .enter
= &intel_idle
,
965 .enter_s2idle
= intel_idle_s2idle
, },
968 .desc
= "MWAIT 0x01",
969 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
971 .target_residency
= 4,
972 .enter
= &intel_idle
,
973 .enter_s2idle
= intel_idle_s2idle
, },
976 .desc
= "MWAIT 0x20",
977 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
979 .target_residency
= 585,
980 .enter
= &intel_idle
,
981 .enter_s2idle
= intel_idle_s2idle
, },
984 .desc
= "MWAIT 0x40",
985 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
987 .target_residency
= 1040,
988 .enter
= &intel_idle
,
989 .enter_s2idle
= intel_idle_s2idle
, },
992 .desc
= "MWAIT 0x60",
993 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
995 .target_residency
= 1980,
996 .enter
= &intel_idle
,
997 .enter_s2idle
= intel_idle_s2idle
, },
1002 static struct cpuidle_state spr_cstates
[] __initdata
= {
1005 .desc
= "MWAIT 0x00",
1006 .flags
= MWAIT2flg(0x00),
1008 .target_residency
= 1,
1009 .enter
= &intel_idle
,
1010 .enter_s2idle
= intel_idle_s2idle
, },
1013 .desc
= "MWAIT 0x01",
1014 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1016 .target_residency
= 4,
1017 .enter
= &intel_idle
,
1018 .enter_s2idle
= intel_idle_s2idle
, },
1021 .desc
= "MWAIT 0x20",
1022 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
|
1023 CPUIDLE_FLAG_INIT_XSTATE
,
1024 .exit_latency
= 290,
1025 .target_residency
= 800,
1026 .enter
= &intel_idle
,
1027 .enter_s2idle
= intel_idle_s2idle
, },
1032 static struct cpuidle_state gnr_cstates
[] __initdata
= {
1035 .desc
= "MWAIT 0x00",
1036 .flags
= MWAIT2flg(0x00),
1038 .target_residency
= 1,
1039 .enter
= &intel_idle
,
1040 .enter_s2idle
= intel_idle_s2idle
, },
1043 .desc
= "MWAIT 0x01",
1044 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1046 .target_residency
= 4,
1047 .enter
= &intel_idle
,
1048 .enter_s2idle
= intel_idle_s2idle
, },
1051 .desc
= "MWAIT 0x20",
1052 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
|
1053 CPUIDLE_FLAG_INIT_XSTATE
|
1054 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1055 .exit_latency
= 170,
1056 .target_residency
= 650,
1057 .enter
= &intel_idle
,
1058 .enter_s2idle
= intel_idle_s2idle
, },
1061 .desc
= "MWAIT 0x21",
1062 .flags
= MWAIT2flg(0x21) | CPUIDLE_FLAG_TLB_FLUSHED
|
1063 CPUIDLE_FLAG_INIT_XSTATE
|
1064 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1065 .exit_latency
= 210,
1066 .target_residency
= 1000,
1067 .enter
= &intel_idle
,
1068 .enter_s2idle
= intel_idle_s2idle
, },
1073 static struct cpuidle_state gnrd_cstates
[] __initdata
= {
1076 .desc
= "MWAIT 0x00",
1077 .flags
= MWAIT2flg(0x00),
1079 .target_residency
= 1,
1080 .enter
= &intel_idle
,
1081 .enter_s2idle
= intel_idle_s2idle
, },
1084 .desc
= "MWAIT 0x01",
1085 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1087 .target_residency
= 4,
1088 .enter
= &intel_idle
,
1089 .enter_s2idle
= intel_idle_s2idle
, },
1092 .desc
= "MWAIT 0x20",
1093 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
|
1094 CPUIDLE_FLAG_INIT_XSTATE
|
1095 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1096 .exit_latency
= 220,
1097 .target_residency
= 650,
1098 .enter
= &intel_idle
,
1099 .enter_s2idle
= intel_idle_s2idle
, },
1102 .desc
= "MWAIT 0x21",
1103 .flags
= MWAIT2flg(0x21) | CPUIDLE_FLAG_TLB_FLUSHED
|
1104 CPUIDLE_FLAG_INIT_XSTATE
|
1105 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1106 .exit_latency
= 240,
1107 .target_residency
= 750,
1108 .enter
= &intel_idle
,
1109 .enter_s2idle
= intel_idle_s2idle
, },
1114 static struct cpuidle_state atom_cstates
[] __initdata
= {
1117 .desc
= "MWAIT 0x00",
1118 .flags
= MWAIT2flg(0x00),
1120 .target_residency
= 20,
1121 .enter
= &intel_idle
,
1122 .enter_s2idle
= intel_idle_s2idle
, },
1125 .desc
= "MWAIT 0x10",
1126 .flags
= MWAIT2flg(0x10),
1128 .target_residency
= 80,
1129 .enter
= &intel_idle
,
1130 .enter_s2idle
= intel_idle_s2idle
, },
1133 .desc
= "MWAIT 0x30",
1134 .flags
= MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED
,
1135 .exit_latency
= 100,
1136 .target_residency
= 400,
1137 .enter
= &intel_idle
,
1138 .enter_s2idle
= intel_idle_s2idle
, },
1141 .desc
= "MWAIT 0x52",
1142 .flags
= MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED
,
1143 .exit_latency
= 140,
1144 .target_residency
= 560,
1145 .enter
= &intel_idle
,
1146 .enter_s2idle
= intel_idle_s2idle
, },
1150 static struct cpuidle_state tangier_cstates
[] __initdata
= {
1153 .desc
= "MWAIT 0x00",
1154 .flags
= MWAIT2flg(0x00),
1156 .target_residency
= 4,
1157 .enter
= &intel_idle
,
1158 .enter_s2idle
= intel_idle_s2idle
, },
1161 .desc
= "MWAIT 0x30",
1162 .flags
= MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED
,
1163 .exit_latency
= 100,
1164 .target_residency
= 400,
1165 .enter
= &intel_idle
,
1166 .enter_s2idle
= intel_idle_s2idle
, },
1169 .desc
= "MWAIT 0x52",
1170 .flags
= MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED
,
1171 .exit_latency
= 140,
1172 .target_residency
= 560,
1173 .enter
= &intel_idle
,
1174 .enter_s2idle
= intel_idle_s2idle
, },
1177 .desc
= "MWAIT 0x60",
1178 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
1179 .exit_latency
= 1200,
1180 .target_residency
= 4000,
1181 .enter
= &intel_idle
,
1182 .enter_s2idle
= intel_idle_s2idle
, },
1185 .desc
= "MWAIT 0x64",
1186 .flags
= MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED
,
1187 .exit_latency
= 10000,
1188 .target_residency
= 20000,
1189 .enter
= &intel_idle
,
1190 .enter_s2idle
= intel_idle_s2idle
, },
1194 static struct cpuidle_state avn_cstates
[] __initdata
= {
1197 .desc
= "MWAIT 0x00",
1198 .flags
= MWAIT2flg(0x00),
1200 .target_residency
= 2,
1201 .enter
= &intel_idle
,
1202 .enter_s2idle
= intel_idle_s2idle
, },
1205 .desc
= "MWAIT 0x51",
1206 .flags
= MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED
,
1208 .target_residency
= 45,
1209 .enter
= &intel_idle
,
1210 .enter_s2idle
= intel_idle_s2idle
, },
1214 static struct cpuidle_state knl_cstates
[] __initdata
= {
1217 .desc
= "MWAIT 0x00",
1218 .flags
= MWAIT2flg(0x00),
1220 .target_residency
= 2,
1221 .enter
= &intel_idle
,
1222 .enter_s2idle
= intel_idle_s2idle
},
1225 .desc
= "MWAIT 0x10",
1226 .flags
= MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED
,
1227 .exit_latency
= 120,
1228 .target_residency
= 500,
1229 .enter
= &intel_idle
,
1230 .enter_s2idle
= intel_idle_s2idle
},
1235 static struct cpuidle_state bxt_cstates
[] __initdata
= {
1238 .desc
= "MWAIT 0x00",
1239 .flags
= MWAIT2flg(0x00),
1241 .target_residency
= 2,
1242 .enter
= &intel_idle
,
1243 .enter_s2idle
= intel_idle_s2idle
, },
1246 .desc
= "MWAIT 0x01",
1247 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1249 .target_residency
= 20,
1250 .enter
= &intel_idle
,
1251 .enter_s2idle
= intel_idle_s2idle
, },
1254 .desc
= "MWAIT 0x20",
1255 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
1256 .exit_latency
= 133,
1257 .target_residency
= 133,
1258 .enter
= &intel_idle
,
1259 .enter_s2idle
= intel_idle_s2idle
, },
1262 .desc
= "MWAIT 0x31",
1263 .flags
= MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED
,
1264 .exit_latency
= 155,
1265 .target_residency
= 155,
1266 .enter
= &intel_idle
,
1267 .enter_s2idle
= intel_idle_s2idle
, },
1270 .desc
= "MWAIT 0x40",
1271 .flags
= MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED
,
1272 .exit_latency
= 1000,
1273 .target_residency
= 1000,
1274 .enter
= &intel_idle
,
1275 .enter_s2idle
= intel_idle_s2idle
, },
1278 .desc
= "MWAIT 0x50",
1279 .flags
= MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED
,
1280 .exit_latency
= 2000,
1281 .target_residency
= 2000,
1282 .enter
= &intel_idle
,
1283 .enter_s2idle
= intel_idle_s2idle
, },
1286 .desc
= "MWAIT 0x60",
1287 .flags
= MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED
,
1288 .exit_latency
= 10000,
1289 .target_residency
= 10000,
1290 .enter
= &intel_idle
,
1291 .enter_s2idle
= intel_idle_s2idle
, },
1296 static struct cpuidle_state dnv_cstates
[] __initdata
= {
1299 .desc
= "MWAIT 0x00",
1300 .flags
= MWAIT2flg(0x00),
1302 .target_residency
= 2,
1303 .enter
= &intel_idle
,
1304 .enter_s2idle
= intel_idle_s2idle
, },
1307 .desc
= "MWAIT 0x01",
1308 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1310 .target_residency
= 20,
1311 .enter
= &intel_idle
,
1312 .enter_s2idle
= intel_idle_s2idle
, },
1315 .desc
= "MWAIT 0x20",
1316 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
1318 .target_residency
= 500,
1319 .enter
= &intel_idle
,
1320 .enter_s2idle
= intel_idle_s2idle
, },
1326 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1327 * C6, and this is indicated in the CPUID mwait leaf.
1329 static struct cpuidle_state snr_cstates
[] __initdata
= {
1332 .desc
= "MWAIT 0x00",
1333 .flags
= MWAIT2flg(0x00),
1335 .target_residency
= 2,
1336 .enter
= &intel_idle
,
1337 .enter_s2idle
= intel_idle_s2idle
, },
1340 .desc
= "MWAIT 0x01",
1341 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1343 .target_residency
= 25,
1344 .enter
= &intel_idle
,
1345 .enter_s2idle
= intel_idle_s2idle
, },
1348 .desc
= "MWAIT 0x20",
1349 .flags
= MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED
,
1350 .exit_latency
= 130,
1351 .target_residency
= 500,
1352 .enter
= &intel_idle
,
1353 .enter_s2idle
= intel_idle_s2idle
, },
1358 static struct cpuidle_state grr_cstates
[] __initdata
= {
1361 .desc
= "MWAIT 0x00",
1362 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1364 .target_residency
= 1,
1365 .enter
= &intel_idle
,
1366 .enter_s2idle
= intel_idle_s2idle
, },
1369 .desc
= "MWAIT 0x01",
1370 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1372 .target_residency
= 10,
1373 .enter
= &intel_idle
,
1374 .enter_s2idle
= intel_idle_s2idle
, },
1377 .desc
= "MWAIT 0x22",
1378 .flags
= MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED
,
1379 .exit_latency
= 140,
1380 .target_residency
= 500,
1381 .enter
= &intel_idle
,
1382 .enter_s2idle
= intel_idle_s2idle
, },
1387 static struct cpuidle_state srf_cstates
[] __initdata
= {
1390 .desc
= "MWAIT 0x00",
1391 .flags
= MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1393 .target_residency
= 1,
1394 .enter
= &intel_idle
,
1395 .enter_s2idle
= intel_idle_s2idle
, },
1398 .desc
= "MWAIT 0x01",
1399 .flags
= MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE
,
1401 .target_residency
= 10,
1402 .enter
= &intel_idle
,
1403 .enter_s2idle
= intel_idle_s2idle
, },
1406 .desc
= "MWAIT 0x22",
1407 .flags
= MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED
|
1408 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1409 .exit_latency
= 270,
1410 .target_residency
= 700,
1411 .enter
= &intel_idle
,
1412 .enter_s2idle
= intel_idle_s2idle
, },
1415 .desc
= "MWAIT 0x23",
1416 .flags
= MWAIT2flg(0x23) | CPUIDLE_FLAG_TLB_FLUSHED
|
1417 CPUIDLE_FLAG_PARTIAL_HINT_MATCH
,
1418 .exit_latency
= 310,
1419 .target_residency
= 900,
1420 .enter
= &intel_idle
,
1421 .enter_s2idle
= intel_idle_s2idle
, },
1426 static const struct idle_cpu idle_cpu_nehalem __initconst
= {
1427 .state_table
= nehalem_cstates
,
1428 .auto_demotion_disable_flags
= NHM_C1_AUTO_DEMOTE
| NHM_C3_AUTO_DEMOTE
,
1429 .disable_promotion_to_c1e
= true,
1432 static const struct idle_cpu idle_cpu_nhx __initconst
= {
1433 .state_table
= nehalem_cstates
,
1434 .auto_demotion_disable_flags
= NHM_C1_AUTO_DEMOTE
| NHM_C3_AUTO_DEMOTE
,
1435 .disable_promotion_to_c1e
= true,
1439 static const struct idle_cpu idle_cpu_atom __initconst
= {
1440 .state_table
= atom_cstates
,
1443 static const struct idle_cpu idle_cpu_tangier __initconst
= {
1444 .state_table
= tangier_cstates
,
1447 static const struct idle_cpu idle_cpu_lincroft __initconst
= {
1448 .state_table
= atom_cstates
,
1449 .auto_demotion_disable_flags
= ATM_LNC_C6_AUTO_DEMOTE
,
1452 static const struct idle_cpu idle_cpu_snb __initconst
= {
1453 .state_table
= snb_cstates
,
1454 .disable_promotion_to_c1e
= true,
1457 static const struct idle_cpu idle_cpu_snx __initconst
= {
1458 .state_table
= snb_cstates
,
1459 .disable_promotion_to_c1e
= true,
1463 static const struct idle_cpu idle_cpu_byt __initconst
= {
1464 .state_table
= byt_cstates
,
1465 .disable_promotion_to_c1e
= true,
1466 .byt_auto_demotion_disable_flag
= true,
1469 static const struct idle_cpu idle_cpu_cht __initconst
= {
1470 .state_table
= cht_cstates
,
1471 .disable_promotion_to_c1e
= true,
1472 .byt_auto_demotion_disable_flag
= true,
1475 static const struct idle_cpu idle_cpu_ivb __initconst
= {
1476 .state_table
= ivb_cstates
,
1477 .disable_promotion_to_c1e
= true,
1480 static const struct idle_cpu idle_cpu_ivt __initconst
= {
1481 .state_table
= ivt_cstates
,
1482 .disable_promotion_to_c1e
= true,
1486 static const struct idle_cpu idle_cpu_hsw __initconst
= {
1487 .state_table
= hsw_cstates
,
1488 .disable_promotion_to_c1e
= true,
1491 static const struct idle_cpu idle_cpu_hsx __initconst
= {
1492 .state_table
= hsw_cstates
,
1493 .disable_promotion_to_c1e
= true,
1497 static const struct idle_cpu idle_cpu_bdw __initconst
= {
1498 .state_table
= bdw_cstates
,
1499 .disable_promotion_to_c1e
= true,
1502 static const struct idle_cpu idle_cpu_bdx __initconst
= {
1503 .state_table
= bdw_cstates
,
1504 .disable_promotion_to_c1e
= true,
1508 static const struct idle_cpu idle_cpu_skl __initconst
= {
1509 .state_table
= skl_cstates
,
1510 .disable_promotion_to_c1e
= true,
1513 static const struct idle_cpu idle_cpu_skx __initconst
= {
1514 .state_table
= skx_cstates
,
1515 .disable_promotion_to_c1e
= true,
1519 static const struct idle_cpu idle_cpu_icx __initconst
= {
1520 .state_table
= icx_cstates
,
1521 .disable_promotion_to_c1e
= true,
1525 static const struct idle_cpu idle_cpu_adl __initconst
= {
1526 .state_table
= adl_cstates
,
1529 static const struct idle_cpu idle_cpu_adl_l __initconst
= {
1530 .state_table
= adl_l_cstates
,
1533 static const struct idle_cpu idle_cpu_mtl_l __initconst
= {
1534 .state_table
= mtl_l_cstates
,
1537 static const struct idle_cpu idle_cpu_gmt __initconst
= {
1538 .state_table
= gmt_cstates
,
1541 static const struct idle_cpu idle_cpu_spr __initconst
= {
1542 .state_table
= spr_cstates
,
1543 .disable_promotion_to_c1e
= true,
1547 static const struct idle_cpu idle_cpu_gnr __initconst
= {
1548 .state_table
= gnr_cstates
,
1549 .disable_promotion_to_c1e
= true,
1553 static const struct idle_cpu idle_cpu_gnrd __initconst
= {
1554 .state_table
= gnrd_cstates
,
1555 .disable_promotion_to_c1e
= true,
1559 static const struct idle_cpu idle_cpu_avn __initconst
= {
1560 .state_table
= avn_cstates
,
1561 .disable_promotion_to_c1e
= true,
1565 static const struct idle_cpu idle_cpu_knl __initconst
= {
1566 .state_table
= knl_cstates
,
1570 static const struct idle_cpu idle_cpu_bxt __initconst
= {
1571 .state_table
= bxt_cstates
,
1572 .disable_promotion_to_c1e
= true,
1575 static const struct idle_cpu idle_cpu_dnv __initconst
= {
1576 .state_table
= dnv_cstates
,
1577 .disable_promotion_to_c1e
= true,
1581 static const struct idle_cpu idle_cpu_tmt __initconst
= {
1582 .disable_promotion_to_c1e
= true,
1585 static const struct idle_cpu idle_cpu_snr __initconst
= {
1586 .state_table
= snr_cstates
,
1587 .disable_promotion_to_c1e
= true,
1591 static const struct idle_cpu idle_cpu_grr __initconst
= {
1592 .state_table
= grr_cstates
,
1593 .disable_promotion_to_c1e
= true,
1597 static const struct idle_cpu idle_cpu_srf __initconst
= {
1598 .state_table
= srf_cstates
,
1599 .disable_promotion_to_c1e
= true,
1603 static const struct x86_cpu_id intel_idle_ids
[] __initconst
= {
1604 X86_MATCH_VFM(INTEL_NEHALEM_EP
, &idle_cpu_nhx
),
1605 X86_MATCH_VFM(INTEL_NEHALEM
, &idle_cpu_nehalem
),
1606 X86_MATCH_VFM(INTEL_NEHALEM_G
, &idle_cpu_nehalem
),
1607 X86_MATCH_VFM(INTEL_WESTMERE
, &idle_cpu_nehalem
),
1608 X86_MATCH_VFM(INTEL_WESTMERE_EP
, &idle_cpu_nhx
),
1609 X86_MATCH_VFM(INTEL_NEHALEM_EX
, &idle_cpu_nhx
),
1610 X86_MATCH_VFM(INTEL_ATOM_BONNELL
, &idle_cpu_atom
),
1611 X86_MATCH_VFM(INTEL_ATOM_BONNELL_MID
, &idle_cpu_lincroft
),
1612 X86_MATCH_VFM(INTEL_WESTMERE_EX
, &idle_cpu_nhx
),
1613 X86_MATCH_VFM(INTEL_SANDYBRIDGE
, &idle_cpu_snb
),
1614 X86_MATCH_VFM(INTEL_SANDYBRIDGE_X
, &idle_cpu_snx
),
1615 X86_MATCH_VFM(INTEL_ATOM_SALTWELL
, &idle_cpu_atom
),
1616 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT
, &idle_cpu_byt
),
1617 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_MID
, &idle_cpu_tangier
),
1618 X86_MATCH_VFM(INTEL_ATOM_AIRMONT
, &idle_cpu_cht
),
1619 X86_MATCH_VFM(INTEL_IVYBRIDGE
, &idle_cpu_ivb
),
1620 X86_MATCH_VFM(INTEL_IVYBRIDGE_X
, &idle_cpu_ivt
),
1621 X86_MATCH_VFM(INTEL_HASWELL
, &idle_cpu_hsw
),
1622 X86_MATCH_VFM(INTEL_HASWELL_X
, &idle_cpu_hsx
),
1623 X86_MATCH_VFM(INTEL_HASWELL_L
, &idle_cpu_hsw
),
1624 X86_MATCH_VFM(INTEL_HASWELL_G
, &idle_cpu_hsw
),
1625 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_D
, &idle_cpu_avn
),
1626 X86_MATCH_VFM(INTEL_BROADWELL
, &idle_cpu_bdw
),
1627 X86_MATCH_VFM(INTEL_BROADWELL_G
, &idle_cpu_bdw
),
1628 X86_MATCH_VFM(INTEL_BROADWELL_X
, &idle_cpu_bdx
),
1629 X86_MATCH_VFM(INTEL_BROADWELL_D
, &idle_cpu_bdx
),
1630 X86_MATCH_VFM(INTEL_SKYLAKE_L
, &idle_cpu_skl
),
1631 X86_MATCH_VFM(INTEL_SKYLAKE
, &idle_cpu_skl
),
1632 X86_MATCH_VFM(INTEL_KABYLAKE_L
, &idle_cpu_skl
),
1633 X86_MATCH_VFM(INTEL_KABYLAKE
, &idle_cpu_skl
),
1634 X86_MATCH_VFM(INTEL_SKYLAKE_X
, &idle_cpu_skx
),
1635 X86_MATCH_VFM(INTEL_ICELAKE_X
, &idle_cpu_icx
),
1636 X86_MATCH_VFM(INTEL_ICELAKE_D
, &idle_cpu_icx
),
1637 X86_MATCH_VFM(INTEL_ALDERLAKE
, &idle_cpu_adl
),
1638 X86_MATCH_VFM(INTEL_ALDERLAKE_L
, &idle_cpu_adl_l
),
1639 X86_MATCH_VFM(INTEL_METEORLAKE_L
, &idle_cpu_mtl_l
),
1640 X86_MATCH_VFM(INTEL_ATOM_GRACEMONT
, &idle_cpu_gmt
),
1641 X86_MATCH_VFM(INTEL_SAPPHIRERAPIDS_X
, &idle_cpu_spr
),
1642 X86_MATCH_VFM(INTEL_EMERALDRAPIDS_X
, &idle_cpu_spr
),
1643 X86_MATCH_VFM(INTEL_GRANITERAPIDS_X
, &idle_cpu_gnr
),
1644 X86_MATCH_VFM(INTEL_GRANITERAPIDS_D
, &idle_cpu_gnrd
),
1645 X86_MATCH_VFM(INTEL_XEON_PHI_KNL
, &idle_cpu_knl
),
1646 X86_MATCH_VFM(INTEL_XEON_PHI_KNM
, &idle_cpu_knl
),
1647 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT
, &idle_cpu_bxt
),
1648 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_PLUS
, &idle_cpu_bxt
),
1649 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_D
, &idle_cpu_dnv
),
1650 X86_MATCH_VFM(INTEL_ATOM_TREMONT
, &idle_cpu_tmt
),
1651 X86_MATCH_VFM(INTEL_ATOM_TREMONT_L
, &idle_cpu_tmt
),
1652 X86_MATCH_VFM(INTEL_ATOM_TREMONT_D
, &idle_cpu_snr
),
1653 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT
, &idle_cpu_grr
),
1654 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT_X
, &idle_cpu_srf
),
1655 X86_MATCH_VFM(INTEL_ATOM_DARKMONT_X
, &idle_cpu_srf
),
1659 static const struct x86_cpu_id intel_mwait_ids
[] __initconst
= {
1660 X86_MATCH_VENDOR_FAM_FEATURE(INTEL
, 6, X86_FEATURE_MWAIT
, NULL
),
1664 static bool __init
intel_idle_max_cstate_reached(int cstate
)
1666 if (cstate
+ 1 > max_cstate
) {
1667 pr_info("max_cstate %d reached\n", max_cstate
);
1673 static bool __init
intel_idle_state_needs_timer_stop(struct cpuidle_state
*state
)
1675 unsigned long eax
= flg2MWAIT(state
->flags
);
1677 if (boot_cpu_has(X86_FEATURE_ARAT
))
1681 * Switch over to one-shot tick broadcast if the target C-state
1682 * is deeper than C1.
1684 return !!((eax
>> MWAIT_SUBSTATE_SIZE
) & MWAIT_CSTATE_MASK
);
1687 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1688 #include <acpi/processor.h>
1690 static bool no_acpi __read_mostly
;
1691 module_param(no_acpi
, bool, 0444);
1692 MODULE_PARM_DESC(no_acpi
, "Do not use ACPI _CST for building the idle states list");
1694 static bool force_use_acpi __read_mostly
; /* No effect if no_acpi is set. */
1695 module_param_named(use_acpi
, force_use_acpi
, bool, 0444);
1696 MODULE_PARM_DESC(use_acpi
, "Use ACPI _CST for building the idle states list");
1698 static struct acpi_processor_power acpi_state_table __initdata
;
1701 * intel_idle_cst_usable - Check if the _CST information can be used.
1703 * Check if all of the C-states listed by _CST in the max_cstate range are
1704 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1706 static bool __init
intel_idle_cst_usable(void)
1710 limit
= min_t(int, min_t(int, CPUIDLE_STATE_MAX
, max_cstate
+ 1),
1711 acpi_state_table
.count
);
1713 for (cstate
= 1; cstate
< limit
; cstate
++) {
1714 struct acpi_processor_cx
*cx
= &acpi_state_table
.states
[cstate
];
1716 if (cx
->entry_method
!= ACPI_CSTATE_FFH
)
1723 static bool __init
intel_idle_acpi_cst_extract(void)
1728 pr_debug("Not allowed to use ACPI _CST\n");
1732 for_each_possible_cpu(cpu
) {
1733 struct acpi_processor
*pr
= per_cpu(processors
, cpu
);
1738 if (acpi_processor_evaluate_cst(pr
->handle
, cpu
, &acpi_state_table
))
1741 acpi_state_table
.count
++;
1743 if (!intel_idle_cst_usable())
1746 if (!acpi_processor_claim_cst_control())
1752 acpi_state_table
.count
= 0;
1753 pr_debug("ACPI _CST not found or not usable\n");
1757 static void __init
intel_idle_init_cstates_acpi(struct cpuidle_driver
*drv
)
1759 int cstate
, limit
= min_t(int, CPUIDLE_STATE_MAX
, acpi_state_table
.count
);
1762 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1763 * the interesting states are ACPI_CSTATE_FFH.
1765 for (cstate
= 1; cstate
< limit
; cstate
++) {
1766 struct acpi_processor_cx
*cx
;
1767 struct cpuidle_state
*state
;
1769 if (intel_idle_max_cstate_reached(cstate
- 1))
1772 cx
= &acpi_state_table
.states
[cstate
];
1774 state
= &drv
->states
[drv
->state_count
++];
1776 snprintf(state
->name
, CPUIDLE_NAME_LEN
, "C%d_ACPI", cstate
);
1777 strscpy(state
->desc
, cx
->desc
, CPUIDLE_DESC_LEN
);
1778 state
->exit_latency
= cx
->latency
;
1780 * For C1-type C-states use the same number for both the exit
1781 * latency and target residency, because that is the case for
1782 * C1 in the majority of the static C-states tables above.
1783 * For the other types of C-states, however, set the target
1784 * residency to 3 times the exit latency which should lead to
1785 * a reasonable balance between energy-efficiency and
1786 * performance in the majority of interesting cases.
1788 state
->target_residency
= cx
->latency
;
1789 if (cx
->type
> ACPI_STATE_C1
)
1790 state
->target_residency
*= 3;
1792 state
->flags
= MWAIT2flg(cx
->address
);
1793 if (cx
->type
> ACPI_STATE_C2
)
1794 state
->flags
|= CPUIDLE_FLAG_TLB_FLUSHED
;
1796 if (disabled_states_mask
& BIT(cstate
))
1797 state
->flags
|= CPUIDLE_FLAG_OFF
;
1799 if (intel_idle_state_needs_timer_stop(state
))
1800 state
->flags
|= CPUIDLE_FLAG_TIMER_STOP
;
1802 state
->enter
= intel_idle
;
1803 state
->enter_s2idle
= intel_idle_s2idle
;
1807 static bool __init
intel_idle_off_by_default(unsigned int flags
, u32 mwait_hint
)
1812 * If there are no _CST C-states, do not disable any C-states by
1815 if (!acpi_state_table
.count
)
1818 limit
= min_t(int, CPUIDLE_STATE_MAX
, acpi_state_table
.count
);
1820 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1821 * the interesting states are ACPI_CSTATE_FFH.
1823 for (cstate
= 1; cstate
< limit
; cstate
++) {
1824 u32 acpi_hint
= acpi_state_table
.states
[cstate
].address
;
1825 u32 table_hint
= mwait_hint
;
1827 if (flags
& CPUIDLE_FLAG_PARTIAL_HINT_MATCH
) {
1828 acpi_hint
&= ~MWAIT_SUBSTATE_MASK
;
1829 table_hint
&= ~MWAIT_SUBSTATE_MASK
;
1832 if (acpi_hint
== table_hint
)
1837 #else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1838 #define force_use_acpi (false)
1840 static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1841 static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver
*drv
) { }
1842 static inline bool intel_idle_off_by_default(unsigned int flags
, u32 mwait_hint
)
1846 #endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1849 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1851 * Tune IVT multi-socket targets.
1852 * Assumption: num_sockets == (max_package_num + 1).
1854 static void __init
ivt_idle_state_table_update(void)
1856 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1857 int cpu
, package_num
, num_sockets
= 1;
1859 for_each_online_cpu(cpu
) {
1860 package_num
= topology_physical_package_id(cpu
);
1861 if (package_num
+ 1 > num_sockets
) {
1862 num_sockets
= package_num
+ 1;
1864 if (num_sockets
> 4) {
1865 cpuidle_state_table
= ivt_cstates_8s
;
1871 if (num_sockets
> 2)
1872 cpuidle_state_table
= ivt_cstates_4s
;
1874 /* else, 1 and 2 socket systems use default ivt_cstates */
1878 * irtl_2_usec - IRTL to microseconds conversion.
1879 * @irtl: IRTL MSR value.
1881 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1883 static unsigned long long __init
irtl_2_usec(unsigned long long irtl
)
1885 static const unsigned int irtl_ns_units
[] __initconst
= {
1886 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1888 unsigned long long ns
;
1893 ns
= irtl_ns_units
[(irtl
>> 10) & 0x7];
1895 return div_u64((irtl
& 0x3FF) * ns
, NSEC_PER_USEC
);
1899 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1901 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1902 * definitive maximum latency and use the same value for target_residency.
1904 static void __init
bxt_idle_state_table_update(void)
1906 unsigned long long msr
;
1909 rdmsrl(MSR_PKGC6_IRTL
, msr
);
1910 usec
= irtl_2_usec(msr
);
1912 bxt_cstates
[2].exit_latency
= usec
;
1913 bxt_cstates
[2].target_residency
= usec
;
1916 rdmsrl(MSR_PKGC7_IRTL
, msr
);
1917 usec
= irtl_2_usec(msr
);
1919 bxt_cstates
[3].exit_latency
= usec
;
1920 bxt_cstates
[3].target_residency
= usec
;
1923 rdmsrl(MSR_PKGC8_IRTL
, msr
);
1924 usec
= irtl_2_usec(msr
);
1926 bxt_cstates
[4].exit_latency
= usec
;
1927 bxt_cstates
[4].target_residency
= usec
;
1930 rdmsrl(MSR_PKGC9_IRTL
, msr
);
1931 usec
= irtl_2_usec(msr
);
1933 bxt_cstates
[5].exit_latency
= usec
;
1934 bxt_cstates
[5].target_residency
= usec
;
1937 rdmsrl(MSR_PKGC10_IRTL
, msr
);
1938 usec
= irtl_2_usec(msr
);
1940 bxt_cstates
[6].exit_latency
= usec
;
1941 bxt_cstates
[6].target_residency
= usec
;
1947 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1949 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1951 static void __init
sklh_idle_state_table_update(void)
1953 unsigned long long msr
;
1954 unsigned int eax
, ebx
, ecx
, edx
;
1957 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1958 if (max_cstate
<= 7)
1961 /* if PC10 not present in CPUID.MWAIT.EDX */
1962 if ((mwait_substates
& (0xF << 28)) == 0)
1965 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL
, msr
);
1967 /* PC10 is not enabled in PKG C-state limit */
1968 if ((msr
& 0xF) != 8)
1972 cpuid(7, &eax
, &ebx
, &ecx
, &edx
);
1974 /* if SGX is present */
1975 if (ebx
& (1 << 2)) {
1977 rdmsrl(MSR_IA32_FEAT_CTL
, msr
);
1979 /* if SGX is enabled */
1980 if (msr
& (1 << 18))
1984 skl_cstates
[5].flags
|= CPUIDLE_FLAG_UNUSABLE
; /* C8-SKL */
1985 skl_cstates
[6].flags
|= CPUIDLE_FLAG_UNUSABLE
; /* C9-SKL */
1989 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1990 * idle states table.
1992 static void __init
skx_idle_state_table_update(void)
1994 unsigned long long msr
;
1996 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL
, msr
);
1999 * 000b: C0/C1 (no package C-state support)
2001 * 010b: C6 (non-retention)
2002 * 011b: C6 (retention)
2003 * 111b: No Package C state limits.
2005 if ((msr
& 0x7) < 2) {
2007 * Uses the CC6 + PC0 latency and 3 times of
2008 * latency for target_residency if the PC6
2009 * is disabled in BIOS. This is consistent
2010 * with how intel_idle driver uses _CST
2011 * to set the target_residency.
2013 skx_cstates
[2].exit_latency
= 92;
2014 skx_cstates
[2].target_residency
= 276;
2019 * adl_idle_state_table_update - Adjust AlderLake idle states table.
2021 static void __init
adl_idle_state_table_update(void)
2023 /* Check if user prefers C1 over C1E. */
2024 if (preferred_states_mask
& BIT(1) && !(preferred_states_mask
& BIT(2))) {
2025 cpuidle_state_table
[0].flags
&= ~CPUIDLE_FLAG_UNUSABLE
;
2026 cpuidle_state_table
[1].flags
|= CPUIDLE_FLAG_UNUSABLE
;
2028 /* Disable C1E by clearing the "C1E promotion" bit. */
2029 c1e_promotion
= C1E_PROMOTION_DISABLE
;
2033 /* Make sure C1E is enabled by default */
2034 c1e_promotion
= C1E_PROMOTION_ENABLE
;
2038 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
2040 static void __init
spr_idle_state_table_update(void)
2042 unsigned long long msr
;
2045 * By default, the C6 state assumes the worst-case scenario of package
2046 * C6. However, if PC6 is disabled, we update the numbers to match
2049 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL
, msr
);
2051 /* Limit value 2 and above allow for PC6. */
2052 if ((msr
& 0x7) < 2) {
2053 spr_cstates
[2].exit_latency
= 190;
2054 spr_cstates
[2].target_residency
= 600;
2058 static bool __init
intel_idle_verify_cstate(unsigned int mwait_hint
)
2060 unsigned int mwait_cstate
= (MWAIT_HINT2CSTATE(mwait_hint
) + 1) &
2062 unsigned int num_substates
= (mwait_substates
>> mwait_cstate
* 4) &
2063 MWAIT_SUBSTATE_MASK
;
2065 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
2066 if (num_substates
== 0)
2069 if (mwait_cstate
> 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC
))
2070 mark_tsc_unstable("TSC halts in idle states deeper than C2");
2075 static void state_update_enter_method(struct cpuidle_state
*state
, int cstate
)
2077 if (state
->flags
& CPUIDLE_FLAG_INIT_XSTATE
) {
2079 * Combining with XSTATE with IBRS or IRQ_ENABLE flags
2080 * is not currently supported but this driver.
2082 WARN_ON_ONCE(state
->flags
& CPUIDLE_FLAG_IBRS
);
2083 WARN_ON_ONCE(state
->flags
& CPUIDLE_FLAG_IRQ_ENABLE
);
2084 state
->enter
= intel_idle_xstate
;
2088 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS
) &&
2089 ((state
->flags
& CPUIDLE_FLAG_IBRS
) || ibrs_off
)) {
2091 * IBRS mitigation requires that C-states are entered
2092 * with interrupts disabled.
2094 if (ibrs_off
&& (state
->flags
& CPUIDLE_FLAG_IRQ_ENABLE
))
2095 state
->flags
&= ~CPUIDLE_FLAG_IRQ_ENABLE
;
2096 WARN_ON_ONCE(state
->flags
& CPUIDLE_FLAG_IRQ_ENABLE
);
2097 state
->enter
= intel_idle_ibrs
;
2101 if (state
->flags
& CPUIDLE_FLAG_IRQ_ENABLE
) {
2102 state
->enter
= intel_idle_irq
;
2107 pr_info("forced intel_idle_irq for state %d\n", cstate
);
2108 state
->enter
= intel_idle_irq
;
2112 static void __init
intel_idle_init_cstates_icpu(struct cpuidle_driver
*drv
)
2116 switch (boot_cpu_data
.x86_vfm
) {
2117 case INTEL_IVYBRIDGE_X
:
2118 ivt_idle_state_table_update();
2120 case INTEL_ATOM_GOLDMONT
:
2121 case INTEL_ATOM_GOLDMONT_PLUS
:
2122 bxt_idle_state_table_update();
2125 sklh_idle_state_table_update();
2127 case INTEL_SKYLAKE_X
:
2128 skx_idle_state_table_update();
2130 case INTEL_SAPPHIRERAPIDS_X
:
2131 case INTEL_EMERALDRAPIDS_X
:
2132 spr_idle_state_table_update();
2134 case INTEL_ALDERLAKE
:
2135 case INTEL_ALDERLAKE_L
:
2136 case INTEL_ATOM_GRACEMONT
:
2137 adl_idle_state_table_update();
2141 for (cstate
= 0; cstate
< CPUIDLE_STATE_MAX
; ++cstate
) {
2142 struct cpuidle_state
*state
;
2143 unsigned int mwait_hint
;
2145 if (intel_idle_max_cstate_reached(cstate
))
2148 if (!cpuidle_state_table
[cstate
].enter
&&
2149 !cpuidle_state_table
[cstate
].enter_s2idle
)
2152 /* If marked as unusable, skip this state. */
2153 if (cpuidle_state_table
[cstate
].flags
& CPUIDLE_FLAG_UNUSABLE
) {
2154 pr_debug("state %s is disabled\n",
2155 cpuidle_state_table
[cstate
].name
);
2159 mwait_hint
= flg2MWAIT(cpuidle_state_table
[cstate
].flags
);
2160 if (!intel_idle_verify_cstate(mwait_hint
))
2163 /* Structure copy. */
2164 drv
->states
[drv
->state_count
] = cpuidle_state_table
[cstate
];
2165 state
= &drv
->states
[drv
->state_count
];
2167 state_update_enter_method(state
, cstate
);
2170 if ((disabled_states_mask
& BIT(drv
->state_count
)) ||
2171 ((icpu
->use_acpi
|| force_use_acpi
) &&
2172 intel_idle_off_by_default(state
->flags
, mwait_hint
) &&
2173 !(state
->flags
& CPUIDLE_FLAG_ALWAYS_ENABLE
)))
2174 state
->flags
|= CPUIDLE_FLAG_OFF
;
2176 if (intel_idle_state_needs_timer_stop(state
))
2177 state
->flags
|= CPUIDLE_FLAG_TIMER_STOP
;
2182 if (icpu
->byt_auto_demotion_disable_flag
) {
2183 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG
, 0);
2184 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG
, 0);
2189 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
2190 * @drv: cpuidle driver structure to initialize.
2192 static void __init
intel_idle_cpuidle_driver_init(struct cpuidle_driver
*drv
)
2194 cpuidle_poll_state_init(drv
);
2196 if (disabled_states_mask
& BIT(0))
2197 drv
->states
[0].flags
|= CPUIDLE_FLAG_OFF
;
2199 drv
->state_count
= 1;
2201 if (icpu
&& icpu
->state_table
)
2202 intel_idle_init_cstates_icpu(drv
);
2204 intel_idle_init_cstates_acpi(drv
);
2207 static void auto_demotion_disable(void)
2209 unsigned long long msr_bits
;
2211 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL
, msr_bits
);
2212 msr_bits
&= ~auto_demotion_disable_flags
;
2213 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL
, msr_bits
);
2216 static void c1e_promotion_enable(void)
2218 unsigned long long msr_bits
;
2220 rdmsrl(MSR_IA32_POWER_CTL
, msr_bits
);
2222 wrmsrl(MSR_IA32_POWER_CTL
, msr_bits
);
2225 static void c1e_promotion_disable(void)
2227 unsigned long long msr_bits
;
2229 rdmsrl(MSR_IA32_POWER_CTL
, msr_bits
);
2231 wrmsrl(MSR_IA32_POWER_CTL
, msr_bits
);
2235 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
2236 * @cpu: CPU to initialize.
2238 * Register a cpuidle device object for @cpu and update its MSRs in accordance
2239 * with the processor model flags.
2241 static int intel_idle_cpu_init(unsigned int cpu
)
2243 struct cpuidle_device
*dev
;
2245 dev
= per_cpu_ptr(intel_idle_cpuidle_devices
, cpu
);
2248 if (cpuidle_register_device(dev
)) {
2249 pr_debug("cpuidle_register_device %d failed!\n", cpu
);
2253 if (auto_demotion_disable_flags
)
2254 auto_demotion_disable();
2256 if (c1e_promotion
== C1E_PROMOTION_ENABLE
)
2257 c1e_promotion_enable();
2258 else if (c1e_promotion
== C1E_PROMOTION_DISABLE
)
2259 c1e_promotion_disable();
2264 static int intel_idle_cpu_online(unsigned int cpu
)
2266 struct cpuidle_device
*dev
;
2268 if (!boot_cpu_has(X86_FEATURE_ARAT
))
2269 tick_broadcast_enable();
2272 * Some systems can hotplug a cpu at runtime after
2273 * the kernel has booted, we have to initialize the
2274 * driver in this case
2276 dev
= per_cpu_ptr(intel_idle_cpuidle_devices
, cpu
);
2277 if (!dev
->registered
)
2278 return intel_idle_cpu_init(cpu
);
2284 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2286 static void __init
intel_idle_cpuidle_devices_uninit(void)
2290 for_each_online_cpu(i
)
2291 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices
, i
));
2294 static int __init
intel_idle_init(void)
2296 const struct x86_cpu_id
*id
;
2297 unsigned int eax
, ebx
, ecx
;
2300 /* Do not load intel_idle at all for now if idle= is passed */
2301 if (boot_option_idle_override
!= IDLE_NO_OVERRIDE
)
2304 if (max_cstate
== 0) {
2305 pr_debug("disabled\n");
2309 id
= x86_match_cpu(intel_idle_ids
);
2311 if (!boot_cpu_has(X86_FEATURE_MWAIT
)) {
2312 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2316 id
= x86_match_cpu(intel_mwait_ids
);
2321 cpuid(CPUID_LEAF_MWAIT
, &eax
, &ebx
, &ecx
, &mwait_substates
);
2323 if (!(ecx
& CPUID5_ECX_EXTENSIONS_SUPPORTED
) ||
2324 !(ecx
& CPUID5_ECX_INTERRUPT_BREAK
) ||
2328 pr_debug("MWAIT substates: 0x%x\n", mwait_substates
);
2330 icpu
= (const struct idle_cpu
*)id
->driver_data
;
2332 if (icpu
->state_table
)
2333 cpuidle_state_table
= icpu
->state_table
;
2334 else if (!intel_idle_acpi_cst_extract())
2337 auto_demotion_disable_flags
= icpu
->auto_demotion_disable_flags
;
2338 if (icpu
->disable_promotion_to_c1e
)
2339 c1e_promotion
= C1E_PROMOTION_DISABLE
;
2340 if (icpu
->use_acpi
|| force_use_acpi
)
2341 intel_idle_acpi_cst_extract();
2342 } else if (!intel_idle_acpi_cst_extract()) {
2346 pr_debug("v" INTEL_IDLE_VERSION
" model 0x%X\n",
2347 boot_cpu_data
.x86_model
);
2349 intel_idle_cpuidle_devices
= alloc_percpu(struct cpuidle_device
);
2350 if (!intel_idle_cpuidle_devices
)
2353 intel_idle_cpuidle_driver_init(&intel_idle_driver
);
2355 retval
= cpuidle_register_driver(&intel_idle_driver
);
2357 struct cpuidle_driver
*drv
= cpuidle_get_driver();
2358 printk(KERN_DEBUG
pr_fmt("intel_idle yielding to %s\n"),
2359 drv
? drv
->name
: "none");
2360 goto init_driver_fail
;
2363 retval
= cpuhp_setup_state(CPUHP_AP_ONLINE_DYN
, "idle/intel:online",
2364 intel_idle_cpu_online
, NULL
);
2368 pr_debug("Local APIC timer is reliable in %s\n",
2369 boot_cpu_has(X86_FEATURE_ARAT
) ? "all C-states" : "C1");
2374 intel_idle_cpuidle_devices_uninit();
2375 cpuidle_unregister_driver(&intel_idle_driver
);
2377 free_percpu(intel_idle_cpuidle_devices
);
2381 device_initcall(intel_idle_init
);
2384 * We are not really modular, but we used to support that. Meaning we also
2385 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2386 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2387 * is the easiest way (currently) to continue doing that.
2389 module_param(max_cstate
, int, 0444);
2391 * The positions of the bits that are set in this number are the indices of the
2392 * idle states to be disabled by default (as reflected by the names of the
2393 * corresponding idle state directories in sysfs, "state0", "state1" ...
2394 * "state<i>" ..., where <i> is the index of the given state).
2396 module_param_named(states_off
, disabled_states_mask
, uint
, 0444);
2397 MODULE_PARM_DESC(states_off
, "Mask of disabled idle states");
2399 * Some platforms come with mutually exclusive C-states, so that if one is
2400 * enabled, the other C-states must not be used. Example: C1 and C1E on
2401 * Sapphire Rapids platform. This parameter allows for selecting the
2402 * preferred C-states among the groups of mutually exclusive C-states - the
2403 * selected C-states will be registered, the other C-states from the mutually
2404 * exclusive group won't be registered. If the platform has no mutually
2405 * exclusive C-states, this parameter has no effect.
2407 module_param_named(preferred_cstates
, preferred_states_mask
, uint
, 0444);
2408 MODULE_PARM_DESC(preferred_cstates
, "Mask of preferred idle states");
2410 * Debugging option that forces the driver to enter all C-states with
2411 * interrupts enabled. Does not apply to C-states with
2412 * 'CPUIDLE_FLAG_INIT_XSTATE' and 'CPUIDLE_FLAG_IBRS' flags.
2414 module_param(force_irq_on
, bool, 0444);
2416 * Force the disabling of IBRS when X86_FEATURE_KERNEL_IBRS is on and
2417 * CPUIDLE_FLAG_IRQ_ENABLE isn't set.
2419 module_param(ibrs_off
, bool, 0444);
2420 MODULE_PARM_DESC(ibrs_off
, "Disable IBRS when idle");