1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * intel_idle.c - native hardware idle loop for modern Intel processors
4 *
5 * Copyright (c) 2013 - 2020, Intel Corporation.
6 * Len Brown <len.brown@intel.com>
7 * Rafael J. Wysocki <rafael.j.wysocki@intel.com>
8 */
9
10/*
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.
15 */
16
17/*
18 * Design Assumptions
19 *
20 * All CPUs have same idle states as boot CPU
21 *
22 * Chipset BM_STS (bus master status) bit is a NOP
23 * for preventing entry into deep C-states
24 *
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)
28 */
29
30/*
31 * Known limitations
32 *
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.
36 *
37 */
38
39/* un-comment DEBUG to enable pr_debug() statements */
40/* #define DEBUG */
41
42#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43
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>
59
60#define INTEL_IDLE_VERSION "0.5.1"
61
62static struct cpuidle_driver intel_idle_driver = {
63 .name = "intel_idle",
64 .owner = THIS_MODULE,
65};
66/* intel_idle.max_cstate=0 disables driver */
67static int max_cstate = CPUIDLE_STATE_MAX - 1;
68static unsigned int disabled_states_mask __read_mostly;
69static unsigned int preferred_states_mask __read_mostly;
70static bool force_irq_on __read_mostly;
71static bool ibrs_off __read_mostly;
72
73static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
74
75static unsigned long auto_demotion_disable_flags;
76
77static enum {
78 C1E_PROMOTION_PRESERVE,
79 C1E_PROMOTION_ENABLE,
80 C1E_PROMOTION_DISABLE
81} c1e_promotion = C1E_PROMOTION_PRESERVE;
82
83struct idle_cpu {
84 struct cpuidle_state *state_table;
85
86 /*
87 * Hardware C-state auto-demotion may not always be optimal.
88 * Indicate which enable bits to clear here.
89 */
90 unsigned long auto_demotion_disable_flags;
91 bool byt_auto_demotion_disable_flag;
92 bool disable_promotion_to_c1e;
93 bool use_acpi;
94};
95
96static const struct idle_cpu *icpu __initdata;
97static struct cpuidle_state *cpuidle_state_table __initdata;
98
99static unsigned int mwait_substates __initdata;
100
101/*
102 * Enable interrupts before entering the C-state. On some platforms and for
103 * some C-states, this may measurably decrease interrupt latency.
104 */
105#define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
106
107/*
108 * Enable this state by default even if the ACPI _CST does not list it.
109 */
110#define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
111
112/*
113 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
114 * above.
115 */
116#define CPUIDLE_FLAG_IBRS BIT(16)
117
118/*
119 * Initialize large xstate for the C6-state entrance.
120 */
121#define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
122
123/*
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.
127 *
128 * We store the hint at the top of our "flags" for each state.
129 */
130#define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
131#define MWAIT2flg(eax) ((eax & 0xFF) << 24)
132
133static __always_inline int __intel_idle(struct cpuidle_device *dev,
134 struct cpuidle_driver *drv, int index)
135{
136 struct cpuidle_state *state = &drv->states[index];
137 unsigned long eax = flg2MWAIT(state->flags);
138 unsigned long ecx = 1; /* break on interrupt flag */
139
140 mwait_idle_with_hints(eax, ecx);
141
142 return index;
143}
144
145/**
146 * intel_idle - Ask the processor to enter the given idle state.
147 * @dev: cpuidle device of the target CPU.
148 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
149 * @index: Target idle state index.
150 *
151 * Use the MWAIT instruction to notify the processor that the CPU represented by
152 * @dev is idle and it can try to enter the idle state corresponding to @index.
153 *
154 * If the local APIC timer is not known to be reliable in the target idle state,
155 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
156 *
157 * Must be called under local_irq_disable().
158 */
159static __cpuidle int intel_idle(struct cpuidle_device *dev,
160 struct cpuidle_driver *drv, int index)
161{
162 return __intel_idle(dev, drv, index);
163}
164
165static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
166 struct cpuidle_driver *drv, int index)
167{
168 int ret;
169
170 raw_local_irq_enable();
171 ret = __intel_idle(dev, drv, index);
172 raw_local_irq_disable();
173
174 return ret;
175}
176
177static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
178 struct cpuidle_driver *drv, int index)
179{
180 bool smt_active = sched_smt_active();
181 u64 spec_ctrl = spec_ctrl_current();
182 int ret;
183
184 if (smt_active)
185 __update_spec_ctrl(val: 0);
186
187 ret = __intel_idle(dev, drv, index);
188
189 if (smt_active)
190 __update_spec_ctrl(val: spec_ctrl);
191
192 return ret;
193}
194
195static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
196 struct cpuidle_driver *drv, int index)
197{
198 fpu_idle_fpregs();
199 return __intel_idle(dev, drv, index);
200}
201
202/**
203 * intel_idle_s2idle - Ask the processor to enter the given idle state.
204 * @dev: cpuidle device of the target CPU.
205 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
206 * @index: Target idle state index.
207 *
208 * Use the MWAIT instruction to notify the processor that the CPU represented by
209 * @dev is idle and it can try to enter the idle state corresponding to @index.
210 *
211 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
212 * scheduler tick and suspended scheduler clock on the target CPU.
213 */
214static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
215 struct cpuidle_driver *drv, int index)
216{
217 unsigned long ecx = 1; /* break on interrupt flag */
218 struct cpuidle_state *state = &drv->states[index];
219 unsigned long eax = flg2MWAIT(state->flags);
220
221 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
222 fpu_idle_fpregs();
223
224 mwait_idle_with_hints(eax, ecx);
225
226 return 0;
227}
228
229/*
230 * States are indexed by the cstate number,
231 * which is also the index into the MWAIT hint array.
232 * Thus C0 is a dummy.
233 */
234static struct cpuidle_state nehalem_cstates[] __initdata = {
235 {
236 .name = "C1",
237 .desc = "MWAIT 0x00",
238 .flags = MWAIT2flg(0x00),
239 .exit_latency = 3,
240 .target_residency = 6,
241 .enter = &intel_idle,
242 .enter_s2idle = intel_idle_s2idle, },
243 {
244 .name = "C1E",
245 .desc = "MWAIT 0x01",
246 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
247 .exit_latency = 10,
248 .target_residency = 20,
249 .enter = &intel_idle,
250 .enter_s2idle = intel_idle_s2idle, },
251 {
252 .name = "C3",
253 .desc = "MWAIT 0x10",
254 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
255 .exit_latency = 20,
256 .target_residency = 80,
257 .enter = &intel_idle,
258 .enter_s2idle = intel_idle_s2idle, },
259 {
260 .name = "C6",
261 .desc = "MWAIT 0x20",
262 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
263 .exit_latency = 200,
264 .target_residency = 800,
265 .enter = &intel_idle,
266 .enter_s2idle = intel_idle_s2idle, },
267 {
268 .enter = NULL }
269};
270
271static struct cpuidle_state snb_cstates[] __initdata = {
272 {
273 .name = "C1",
274 .desc = "MWAIT 0x00",
275 .flags = MWAIT2flg(0x00),
276 .exit_latency = 2,
277 .target_residency = 2,
278 .enter = &intel_idle,
279 .enter_s2idle = intel_idle_s2idle, },
280 {
281 .name = "C1E",
282 .desc = "MWAIT 0x01",
283 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
284 .exit_latency = 10,
285 .target_residency = 20,
286 .enter = &intel_idle,
287 .enter_s2idle = intel_idle_s2idle, },
288 {
289 .name = "C3",
290 .desc = "MWAIT 0x10",
291 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
292 .exit_latency = 80,
293 .target_residency = 211,
294 .enter = &intel_idle,
295 .enter_s2idle = intel_idle_s2idle, },
296 {
297 .name = "C6",
298 .desc = "MWAIT 0x20",
299 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
300 .exit_latency = 104,
301 .target_residency = 345,
302 .enter = &intel_idle,
303 .enter_s2idle = intel_idle_s2idle, },
304 {
305 .name = "C7",
306 .desc = "MWAIT 0x30",
307 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
308 .exit_latency = 109,
309 .target_residency = 345,
310 .enter = &intel_idle,
311 .enter_s2idle = intel_idle_s2idle, },
312 {
313 .enter = NULL }
314};
315
316static struct cpuidle_state byt_cstates[] __initdata = {
317 {
318 .name = "C1",
319 .desc = "MWAIT 0x00",
320 .flags = MWAIT2flg(0x00),
321 .exit_latency = 1,
322 .target_residency = 1,
323 .enter = &intel_idle,
324 .enter_s2idle = intel_idle_s2idle, },
325 {
326 .name = "C6N",
327 .desc = "MWAIT 0x58",
328 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
329 .exit_latency = 300,
330 .target_residency = 275,
331 .enter = &intel_idle,
332 .enter_s2idle = intel_idle_s2idle, },
333 {
334 .name = "C6S",
335 .desc = "MWAIT 0x52",
336 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
337 .exit_latency = 500,
338 .target_residency = 560,
339 .enter = &intel_idle,
340 .enter_s2idle = intel_idle_s2idle, },
341 {
342 .name = "C7",
343 .desc = "MWAIT 0x60",
344 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
345 .exit_latency = 1200,
346 .target_residency = 4000,
347 .enter = &intel_idle,
348 .enter_s2idle = intel_idle_s2idle, },
349 {
350 .name = "C7S",
351 .desc = "MWAIT 0x64",
352 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
353 .exit_latency = 10000,
354 .target_residency = 20000,
355 .enter = &intel_idle,
356 .enter_s2idle = intel_idle_s2idle, },
357 {
358 .enter = NULL }
359};
360
361static struct cpuidle_state cht_cstates[] __initdata = {
362 {
363 .name = "C1",
364 .desc = "MWAIT 0x00",
365 .flags = MWAIT2flg(0x00),
366 .exit_latency = 1,
367 .target_residency = 1,
368 .enter = &intel_idle,
369 .enter_s2idle = intel_idle_s2idle, },
370 {
371 .name = "C6N",
372 .desc = "MWAIT 0x58",
373 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
374 .exit_latency = 80,
375 .target_residency = 275,
376 .enter = &intel_idle,
377 .enter_s2idle = intel_idle_s2idle, },
378 {
379 .name = "C6S",
380 .desc = "MWAIT 0x52",
381 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
382 .exit_latency = 200,
383 .target_residency = 560,
384 .enter = &intel_idle,
385 .enter_s2idle = intel_idle_s2idle, },
386 {
387 .name = "C7",
388 .desc = "MWAIT 0x60",
389 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
390 .exit_latency = 1200,
391 .target_residency = 4000,
392 .enter = &intel_idle,
393 .enter_s2idle = intel_idle_s2idle, },
394 {
395 .name = "C7S",
396 .desc = "MWAIT 0x64",
397 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
398 .exit_latency = 10000,
399 .target_residency = 20000,
400 .enter = &intel_idle,
401 .enter_s2idle = intel_idle_s2idle, },
402 {
403 .enter = NULL }
404};
405
406static struct cpuidle_state ivb_cstates[] __initdata = {
407 {
408 .name = "C1",
409 .desc = "MWAIT 0x00",
410 .flags = MWAIT2flg(0x00),
411 .exit_latency = 1,
412 .target_residency = 1,
413 .enter = &intel_idle,
414 .enter_s2idle = intel_idle_s2idle, },
415 {
416 .name = "C1E",
417 .desc = "MWAIT 0x01",
418 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
419 .exit_latency = 10,
420 .target_residency = 20,
421 .enter = &intel_idle,
422 .enter_s2idle = intel_idle_s2idle, },
423 {
424 .name = "C3",
425 .desc = "MWAIT 0x10",
426 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
427 .exit_latency = 59,
428 .target_residency = 156,
429 .enter = &intel_idle,
430 .enter_s2idle = intel_idle_s2idle, },
431 {
432 .name = "C6",
433 .desc = "MWAIT 0x20",
434 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
435 .exit_latency = 80,
436 .target_residency = 300,
437 .enter = &intel_idle,
438 .enter_s2idle = intel_idle_s2idle, },
439 {
440 .name = "C7",
441 .desc = "MWAIT 0x30",
442 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
443 .exit_latency = 87,
444 .target_residency = 300,
445 .enter = &intel_idle,
446 .enter_s2idle = intel_idle_s2idle, },
447 {
448 .enter = NULL }
449};
450
451static struct cpuidle_state ivt_cstates[] __initdata = {
452 {
453 .name = "C1",
454 .desc = "MWAIT 0x00",
455 .flags = MWAIT2flg(0x00),
456 .exit_latency = 1,
457 .target_residency = 1,
458 .enter = &intel_idle,
459 .enter_s2idle = intel_idle_s2idle, },
460 {
461 .name = "C1E",
462 .desc = "MWAIT 0x01",
463 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
464 .exit_latency = 10,
465 .target_residency = 80,
466 .enter = &intel_idle,
467 .enter_s2idle = intel_idle_s2idle, },
468 {
469 .name = "C3",
470 .desc = "MWAIT 0x10",
471 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
472 .exit_latency = 59,
473 .target_residency = 156,
474 .enter = &intel_idle,
475 .enter_s2idle = intel_idle_s2idle, },
476 {
477 .name = "C6",
478 .desc = "MWAIT 0x20",
479 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
480 .exit_latency = 82,
481 .target_residency = 300,
482 .enter = &intel_idle,
483 .enter_s2idle = intel_idle_s2idle, },
484 {
485 .enter = NULL }
486};
487
488static struct cpuidle_state ivt_cstates_4s[] __initdata = {
489 {
490 .name = "C1",
491 .desc = "MWAIT 0x00",
492 .flags = MWAIT2flg(0x00),
493 .exit_latency = 1,
494 .target_residency = 1,
495 .enter = &intel_idle,
496 .enter_s2idle = intel_idle_s2idle, },
497 {
498 .name = "C1E",
499 .desc = "MWAIT 0x01",
500 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
501 .exit_latency = 10,
502 .target_residency = 250,
503 .enter = &intel_idle,
504 .enter_s2idle = intel_idle_s2idle, },
505 {
506 .name = "C3",
507 .desc = "MWAIT 0x10",
508 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
509 .exit_latency = 59,
510 .target_residency = 300,
511 .enter = &intel_idle,
512 .enter_s2idle = intel_idle_s2idle, },
513 {
514 .name = "C6",
515 .desc = "MWAIT 0x20",
516 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
517 .exit_latency = 84,
518 .target_residency = 400,
519 .enter = &intel_idle,
520 .enter_s2idle = intel_idle_s2idle, },
521 {
522 .enter = NULL }
523};
524
525static struct cpuidle_state ivt_cstates_8s[] __initdata = {
526 {
527 .name = "C1",
528 .desc = "MWAIT 0x00",
529 .flags = MWAIT2flg(0x00),
530 .exit_latency = 1,
531 .target_residency = 1,
532 .enter = &intel_idle,
533 .enter_s2idle = intel_idle_s2idle, },
534 {
535 .name = "C1E",
536 .desc = "MWAIT 0x01",
537 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
538 .exit_latency = 10,
539 .target_residency = 500,
540 .enter = &intel_idle,
541 .enter_s2idle = intel_idle_s2idle, },
542 {
543 .name = "C3",
544 .desc = "MWAIT 0x10",
545 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
546 .exit_latency = 59,
547 .target_residency = 600,
548 .enter = &intel_idle,
549 .enter_s2idle = intel_idle_s2idle, },
550 {
551 .name = "C6",
552 .desc = "MWAIT 0x20",
553 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
554 .exit_latency = 88,
555 .target_residency = 700,
556 .enter = &intel_idle,
557 .enter_s2idle = intel_idle_s2idle, },
558 {
559 .enter = NULL }
560};
561
562static struct cpuidle_state hsw_cstates[] __initdata = {
563 {
564 .name = "C1",
565 .desc = "MWAIT 0x00",
566 .flags = MWAIT2flg(0x00),
567 .exit_latency = 2,
568 .target_residency = 2,
569 .enter = &intel_idle,
570 .enter_s2idle = intel_idle_s2idle, },
571 {
572 .name = "C1E",
573 .desc = "MWAIT 0x01",
574 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
575 .exit_latency = 10,
576 .target_residency = 20,
577 .enter = &intel_idle,
578 .enter_s2idle = intel_idle_s2idle, },
579 {
580 .name = "C3",
581 .desc = "MWAIT 0x10",
582 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
583 .exit_latency = 33,
584 .target_residency = 100,
585 .enter = &intel_idle,
586 .enter_s2idle = intel_idle_s2idle, },
587 {
588 .name = "C6",
589 .desc = "MWAIT 0x20",
590 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
591 .exit_latency = 133,
592 .target_residency = 400,
593 .enter = &intel_idle,
594 .enter_s2idle = intel_idle_s2idle, },
595 {
596 .name = "C7s",
597 .desc = "MWAIT 0x32",
598 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
599 .exit_latency = 166,
600 .target_residency = 500,
601 .enter = &intel_idle,
602 .enter_s2idle = intel_idle_s2idle, },
603 {
604 .name = "C8",
605 .desc = "MWAIT 0x40",
606 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
607 .exit_latency = 300,
608 .target_residency = 900,
609 .enter = &intel_idle,
610 .enter_s2idle = intel_idle_s2idle, },
611 {
612 .name = "C9",
613 .desc = "MWAIT 0x50",
614 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
615 .exit_latency = 600,
616 .target_residency = 1800,
617 .enter = &intel_idle,
618 .enter_s2idle = intel_idle_s2idle, },
619 {
620 .name = "C10",
621 .desc = "MWAIT 0x60",
622 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
623 .exit_latency = 2600,
624 .target_residency = 7700,
625 .enter = &intel_idle,
626 .enter_s2idle = intel_idle_s2idle, },
627 {
628 .enter = NULL }
629};
630static struct cpuidle_state bdw_cstates[] __initdata = {
631 {
632 .name = "C1",
633 .desc = "MWAIT 0x00",
634 .flags = MWAIT2flg(0x00),
635 .exit_latency = 2,
636 .target_residency = 2,
637 .enter = &intel_idle,
638 .enter_s2idle = intel_idle_s2idle, },
639 {
640 .name = "C1E",
641 .desc = "MWAIT 0x01",
642 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
643 .exit_latency = 10,
644 .target_residency = 20,
645 .enter = &intel_idle,
646 .enter_s2idle = intel_idle_s2idle, },
647 {
648 .name = "C3",
649 .desc = "MWAIT 0x10",
650 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
651 .exit_latency = 40,
652 .target_residency = 100,
653 .enter = &intel_idle,
654 .enter_s2idle = intel_idle_s2idle, },
655 {
656 .name = "C6",
657 .desc = "MWAIT 0x20",
658 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
659 .exit_latency = 133,
660 .target_residency = 400,
661 .enter = &intel_idle,
662 .enter_s2idle = intel_idle_s2idle, },
663 {
664 .name = "C7s",
665 .desc = "MWAIT 0x32",
666 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
667 .exit_latency = 166,
668 .target_residency = 500,
669 .enter = &intel_idle,
670 .enter_s2idle = intel_idle_s2idle, },
671 {
672 .name = "C8",
673 .desc = "MWAIT 0x40",
674 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
675 .exit_latency = 300,
676 .target_residency = 900,
677 .enter = &intel_idle,
678 .enter_s2idle = intel_idle_s2idle, },
679 {
680 .name = "C9",
681 .desc = "MWAIT 0x50",
682 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
683 .exit_latency = 600,
684 .target_residency = 1800,
685 .enter = &intel_idle,
686 .enter_s2idle = intel_idle_s2idle, },
687 {
688 .name = "C10",
689 .desc = "MWAIT 0x60",
690 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
691 .exit_latency = 2600,
692 .target_residency = 7700,
693 .enter = &intel_idle,
694 .enter_s2idle = intel_idle_s2idle, },
695 {
696 .enter = NULL }
697};
698
699static struct cpuidle_state skl_cstates[] __initdata = {
700 {
701 .name = "C1",
702 .desc = "MWAIT 0x00",
703 .flags = MWAIT2flg(0x00),
704 .exit_latency = 2,
705 .target_residency = 2,
706 .enter = &intel_idle,
707 .enter_s2idle = intel_idle_s2idle, },
708 {
709 .name = "C1E",
710 .desc = "MWAIT 0x01",
711 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
712 .exit_latency = 10,
713 .target_residency = 20,
714 .enter = &intel_idle,
715 .enter_s2idle = intel_idle_s2idle, },
716 {
717 .name = "C3",
718 .desc = "MWAIT 0x10",
719 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
720 .exit_latency = 70,
721 .target_residency = 100,
722 .enter = &intel_idle,
723 .enter_s2idle = intel_idle_s2idle, },
724 {
725 .name = "C6",
726 .desc = "MWAIT 0x20",
727 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
728 .exit_latency = 85,
729 .target_residency = 200,
730 .enter = &intel_idle,
731 .enter_s2idle = intel_idle_s2idle, },
732 {
733 .name = "C7s",
734 .desc = "MWAIT 0x33",
735 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
736 .exit_latency = 124,
737 .target_residency = 800,
738 .enter = &intel_idle,
739 .enter_s2idle = intel_idle_s2idle, },
740 {
741 .name = "C8",
742 .desc = "MWAIT 0x40",
743 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
744 .exit_latency = 200,
745 .target_residency = 800,
746 .enter = &intel_idle,
747 .enter_s2idle = intel_idle_s2idle, },
748 {
749 .name = "C9",
750 .desc = "MWAIT 0x50",
751 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
752 .exit_latency = 480,
753 .target_residency = 5000,
754 .enter = &intel_idle,
755 .enter_s2idle = intel_idle_s2idle, },
756 {
757 .name = "C10",
758 .desc = "MWAIT 0x60",
759 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
760 .exit_latency = 890,
761 .target_residency = 5000,
762 .enter = &intel_idle,
763 .enter_s2idle = intel_idle_s2idle, },
764 {
765 .enter = NULL }
766};
767
768static struct cpuidle_state skx_cstates[] __initdata = {
769 {
770 .name = "C1",
771 .desc = "MWAIT 0x00",
772 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
773 .exit_latency = 2,
774 .target_residency = 2,
775 .enter = &intel_idle,
776 .enter_s2idle = intel_idle_s2idle, },
777 {
778 .name = "C1E",
779 .desc = "MWAIT 0x01",
780 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
781 .exit_latency = 10,
782 .target_residency = 20,
783 .enter = &intel_idle,
784 .enter_s2idle = intel_idle_s2idle, },
785 {
786 .name = "C6",
787 .desc = "MWAIT 0x20",
788 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
789 .exit_latency = 133,
790 .target_residency = 600,
791 .enter = &intel_idle,
792 .enter_s2idle = intel_idle_s2idle, },
793 {
794 .enter = NULL }
795};
796
797static struct cpuidle_state icx_cstates[] __initdata = {
798 {
799 .name = "C1",
800 .desc = "MWAIT 0x00",
801 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
802 .exit_latency = 1,
803 .target_residency = 1,
804 .enter = &intel_idle,
805 .enter_s2idle = intel_idle_s2idle, },
806 {
807 .name = "C1E",
808 .desc = "MWAIT 0x01",
809 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
810 .exit_latency = 4,
811 .target_residency = 4,
812 .enter = &intel_idle,
813 .enter_s2idle = intel_idle_s2idle, },
814 {
815 .name = "C6",
816 .desc = "MWAIT 0x20",
817 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
818 .exit_latency = 170,
819 .target_residency = 600,
820 .enter = &intel_idle,
821 .enter_s2idle = intel_idle_s2idle, },
822 {
823 .enter = NULL }
824};
825
826/*
827 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
828 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
829 * But in this case there is effectively no C1, because C1 requests are
830 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
831 * and C1E requests end up with C1, so there is effectively no C1E.
832 *
833 * By default we enable C1E and disable C1 by marking it with
834 * 'CPUIDLE_FLAG_UNUSABLE'.
835 */
836static struct cpuidle_state adl_cstates[] __initdata = {
837 {
838 .name = "C1",
839 .desc = "MWAIT 0x00",
840 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
841 .exit_latency = 1,
842 .target_residency = 1,
843 .enter = &intel_idle,
844 .enter_s2idle = intel_idle_s2idle, },
845 {
846 .name = "C1E",
847 .desc = "MWAIT 0x01",
848 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
849 .exit_latency = 2,
850 .target_residency = 4,
851 .enter = &intel_idle,
852 .enter_s2idle = intel_idle_s2idle, },
853 {
854 .name = "C6",
855 .desc = "MWAIT 0x20",
856 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
857 .exit_latency = 220,
858 .target_residency = 600,
859 .enter = &intel_idle,
860 .enter_s2idle = intel_idle_s2idle, },
861 {
862 .name = "C8",
863 .desc = "MWAIT 0x40",
864 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
865 .exit_latency = 280,
866 .target_residency = 800,
867 .enter = &intel_idle,
868 .enter_s2idle = intel_idle_s2idle, },
869 {
870 .name = "C10",
871 .desc = "MWAIT 0x60",
872 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
873 .exit_latency = 680,
874 .target_residency = 2000,
875 .enter = &intel_idle,
876 .enter_s2idle = intel_idle_s2idle, },
877 {
878 .enter = NULL }
879};
880
881static struct cpuidle_state adl_l_cstates[] __initdata = {
882 {
883 .name = "C1",
884 .desc = "MWAIT 0x00",
885 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
886 .exit_latency = 1,
887 .target_residency = 1,
888 .enter = &intel_idle,
889 .enter_s2idle = intel_idle_s2idle, },
890 {
891 .name = "C1E",
892 .desc = "MWAIT 0x01",
893 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
894 .exit_latency = 2,
895 .target_residency = 4,
896 .enter = &intel_idle,
897 .enter_s2idle = intel_idle_s2idle, },
898 {
899 .name = "C6",
900 .desc = "MWAIT 0x20",
901 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
902 .exit_latency = 170,
903 .target_residency = 500,
904 .enter = &intel_idle,
905 .enter_s2idle = intel_idle_s2idle, },
906 {
907 .name = "C8",
908 .desc = "MWAIT 0x40",
909 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
910 .exit_latency = 200,
911 .target_residency = 600,
912 .enter = &intel_idle,
913 .enter_s2idle = intel_idle_s2idle, },
914 {
915 .name = "C10",
916 .desc = "MWAIT 0x60",
917 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
918 .exit_latency = 230,
919 .target_residency = 700,
920 .enter = &intel_idle,
921 .enter_s2idle = intel_idle_s2idle, },
922 {
923 .enter = NULL }
924};
925
926static struct cpuidle_state gmt_cstates[] __initdata = {
927 {
928 .name = "C1",
929 .desc = "MWAIT 0x00",
930 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
931 .exit_latency = 1,
932 .target_residency = 1,
933 .enter = &intel_idle,
934 .enter_s2idle = intel_idle_s2idle, },
935 {
936 .name = "C1E",
937 .desc = "MWAIT 0x01",
938 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
939 .exit_latency = 2,
940 .target_residency = 4,
941 .enter = &intel_idle,
942 .enter_s2idle = intel_idle_s2idle, },
943 {
944 .name = "C6",
945 .desc = "MWAIT 0x20",
946 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
947 .exit_latency = 195,
948 .target_residency = 585,
949 .enter = &intel_idle,
950 .enter_s2idle = intel_idle_s2idle, },
951 {
952 .name = "C8",
953 .desc = "MWAIT 0x40",
954 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
955 .exit_latency = 260,
956 .target_residency = 1040,
957 .enter = &intel_idle,
958 .enter_s2idle = intel_idle_s2idle, },
959 {
960 .name = "C10",
961 .desc = "MWAIT 0x60",
962 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
963 .exit_latency = 660,
964 .target_residency = 1980,
965 .enter = &intel_idle,
966 .enter_s2idle = intel_idle_s2idle, },
967 {
968 .enter = NULL }
969};
970
971static struct cpuidle_state spr_cstates[] __initdata = {
972 {
973 .name = "C1",
974 .desc = "MWAIT 0x00",
975 .flags = MWAIT2flg(0x00),
976 .exit_latency = 1,
977 .target_residency = 1,
978 .enter = &intel_idle,
979 .enter_s2idle = intel_idle_s2idle, },
980 {
981 .name = "C1E",
982 .desc = "MWAIT 0x01",
983 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
984 .exit_latency = 2,
985 .target_residency = 4,
986 .enter = &intel_idle,
987 .enter_s2idle = intel_idle_s2idle, },
988 {
989 .name = "C6",
990 .desc = "MWAIT 0x20",
991 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
992 CPUIDLE_FLAG_INIT_XSTATE,
993 .exit_latency = 290,
994 .target_residency = 800,
995 .enter = &intel_idle,
996 .enter_s2idle = intel_idle_s2idle, },
997 {
998 .enter = NULL }
999};
1000
1001static struct cpuidle_state atom_cstates[] __initdata = {
1002 {
1003 .name = "C1E",
1004 .desc = "MWAIT 0x00",
1005 .flags = MWAIT2flg(0x00),
1006 .exit_latency = 10,
1007 .target_residency = 20,
1008 .enter = &intel_idle,
1009 .enter_s2idle = intel_idle_s2idle, },
1010 {
1011 .name = "C2",
1012 .desc = "MWAIT 0x10",
1013 .flags = MWAIT2flg(0x10),
1014 .exit_latency = 20,
1015 .target_residency = 80,
1016 .enter = &intel_idle,
1017 .enter_s2idle = intel_idle_s2idle, },
1018 {
1019 .name = "C4",
1020 .desc = "MWAIT 0x30",
1021 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1022 .exit_latency = 100,
1023 .target_residency = 400,
1024 .enter = &intel_idle,
1025 .enter_s2idle = intel_idle_s2idle, },
1026 {
1027 .name = "C6",
1028 .desc = "MWAIT 0x52",
1029 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1030 .exit_latency = 140,
1031 .target_residency = 560,
1032 .enter = &intel_idle,
1033 .enter_s2idle = intel_idle_s2idle, },
1034 {
1035 .enter = NULL }
1036};
1037static struct cpuidle_state tangier_cstates[] __initdata = {
1038 {
1039 .name = "C1",
1040 .desc = "MWAIT 0x00",
1041 .flags = MWAIT2flg(0x00),
1042 .exit_latency = 1,
1043 .target_residency = 4,
1044 .enter = &intel_idle,
1045 .enter_s2idle = intel_idle_s2idle, },
1046 {
1047 .name = "C4",
1048 .desc = "MWAIT 0x30",
1049 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1050 .exit_latency = 100,
1051 .target_residency = 400,
1052 .enter = &intel_idle,
1053 .enter_s2idle = intel_idle_s2idle, },
1054 {
1055 .name = "C6",
1056 .desc = "MWAIT 0x52",
1057 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1058 .exit_latency = 140,
1059 .target_residency = 560,
1060 .enter = &intel_idle,
1061 .enter_s2idle = intel_idle_s2idle, },
1062 {
1063 .name = "C7",
1064 .desc = "MWAIT 0x60",
1065 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1066 .exit_latency = 1200,
1067 .target_residency = 4000,
1068 .enter = &intel_idle,
1069 .enter_s2idle = intel_idle_s2idle, },
1070 {
1071 .name = "C9",
1072 .desc = "MWAIT 0x64",
1073 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1074 .exit_latency = 10000,
1075 .target_residency = 20000,
1076 .enter = &intel_idle,
1077 .enter_s2idle = intel_idle_s2idle, },
1078 {
1079 .enter = NULL }
1080};
1081static struct cpuidle_state avn_cstates[] __initdata = {
1082 {
1083 .name = "C1",
1084 .desc = "MWAIT 0x00",
1085 .flags = MWAIT2flg(0x00),
1086 .exit_latency = 2,
1087 .target_residency = 2,
1088 .enter = &intel_idle,
1089 .enter_s2idle = intel_idle_s2idle, },
1090 {
1091 .name = "C6",
1092 .desc = "MWAIT 0x51",
1093 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1094 .exit_latency = 15,
1095 .target_residency = 45,
1096 .enter = &intel_idle,
1097 .enter_s2idle = intel_idle_s2idle, },
1098 {
1099 .enter = NULL }
1100};
1101static struct cpuidle_state knl_cstates[] __initdata = {
1102 {
1103 .name = "C1",
1104 .desc = "MWAIT 0x00",
1105 .flags = MWAIT2flg(0x00),
1106 .exit_latency = 1,
1107 .target_residency = 2,
1108 .enter = &intel_idle,
1109 .enter_s2idle = intel_idle_s2idle },
1110 {
1111 .name = "C6",
1112 .desc = "MWAIT 0x10",
1113 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1114 .exit_latency = 120,
1115 .target_residency = 500,
1116 .enter = &intel_idle,
1117 .enter_s2idle = intel_idle_s2idle },
1118 {
1119 .enter = NULL }
1120};
1121
1122static struct cpuidle_state bxt_cstates[] __initdata = {
1123 {
1124 .name = "C1",
1125 .desc = "MWAIT 0x00",
1126 .flags = MWAIT2flg(0x00),
1127 .exit_latency = 2,
1128 .target_residency = 2,
1129 .enter = &intel_idle,
1130 .enter_s2idle = intel_idle_s2idle, },
1131 {
1132 .name = "C1E",
1133 .desc = "MWAIT 0x01",
1134 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1135 .exit_latency = 10,
1136 .target_residency = 20,
1137 .enter = &intel_idle,
1138 .enter_s2idle = intel_idle_s2idle, },
1139 {
1140 .name = "C6",
1141 .desc = "MWAIT 0x20",
1142 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1143 .exit_latency = 133,
1144 .target_residency = 133,
1145 .enter = &intel_idle,
1146 .enter_s2idle = intel_idle_s2idle, },
1147 {
1148 .name = "C7s",
1149 .desc = "MWAIT 0x31",
1150 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1151 .exit_latency = 155,
1152 .target_residency = 155,
1153 .enter = &intel_idle,
1154 .enter_s2idle = intel_idle_s2idle, },
1155 {
1156 .name = "C8",
1157 .desc = "MWAIT 0x40",
1158 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1159 .exit_latency = 1000,
1160 .target_residency = 1000,
1161 .enter = &intel_idle,
1162 .enter_s2idle = intel_idle_s2idle, },
1163 {
1164 .name = "C9",
1165 .desc = "MWAIT 0x50",
1166 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1167 .exit_latency = 2000,
1168 .target_residency = 2000,
1169 .enter = &intel_idle,
1170 .enter_s2idle = intel_idle_s2idle, },
1171 {
1172 .name = "C10",
1173 .desc = "MWAIT 0x60",
1174 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1175 .exit_latency = 10000,
1176 .target_residency = 10000,
1177 .enter = &intel_idle,
1178 .enter_s2idle = intel_idle_s2idle, },
1179 {
1180 .enter = NULL }
1181};
1182
1183static struct cpuidle_state dnv_cstates[] __initdata = {
1184 {
1185 .name = "C1",
1186 .desc = "MWAIT 0x00",
1187 .flags = MWAIT2flg(0x00),
1188 .exit_latency = 2,
1189 .target_residency = 2,
1190 .enter = &intel_idle,
1191 .enter_s2idle = intel_idle_s2idle, },
1192 {
1193 .name = "C1E",
1194 .desc = "MWAIT 0x01",
1195 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1196 .exit_latency = 10,
1197 .target_residency = 20,
1198 .enter = &intel_idle,
1199 .enter_s2idle = intel_idle_s2idle, },
1200 {
1201 .name = "C6",
1202 .desc = "MWAIT 0x20",
1203 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1204 .exit_latency = 50,
1205 .target_residency = 500,
1206 .enter = &intel_idle,
1207 .enter_s2idle = intel_idle_s2idle, },
1208 {
1209 .enter = NULL }
1210};
1211
1212/*
1213 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1214 * C6, and this is indicated in the CPUID mwait leaf.
1215 */
1216static struct cpuidle_state snr_cstates[] __initdata = {
1217 {
1218 .name = "C1",
1219 .desc = "MWAIT 0x00",
1220 .flags = MWAIT2flg(0x00),
1221 .exit_latency = 2,
1222 .target_residency = 2,
1223 .enter = &intel_idle,
1224 .enter_s2idle = intel_idle_s2idle, },
1225 {
1226 .name = "C1E",
1227 .desc = "MWAIT 0x01",
1228 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1229 .exit_latency = 15,
1230 .target_residency = 25,
1231 .enter = &intel_idle,
1232 .enter_s2idle = intel_idle_s2idle, },
1233 {
1234 .name = "C6",
1235 .desc = "MWAIT 0x20",
1236 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1237 .exit_latency = 130,
1238 .target_residency = 500,
1239 .enter = &intel_idle,
1240 .enter_s2idle = intel_idle_s2idle, },
1241 {
1242 .enter = NULL }
1243};
1244
1245static const struct idle_cpu idle_cpu_nehalem __initconst = {
1246 .state_table = nehalem_cstates,
1247 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1248 .disable_promotion_to_c1e = true,
1249};
1250
1251static const struct idle_cpu idle_cpu_nhx __initconst = {
1252 .state_table = nehalem_cstates,
1253 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1254 .disable_promotion_to_c1e = true,
1255 .use_acpi = true,
1256};
1257
1258static const struct idle_cpu idle_cpu_atom __initconst = {
1259 .state_table = atom_cstates,
1260};
1261
1262static const struct idle_cpu idle_cpu_tangier __initconst = {
1263 .state_table = tangier_cstates,
1264};
1265
1266static const struct idle_cpu idle_cpu_lincroft __initconst = {
1267 .state_table = atom_cstates,
1268 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1269};
1270
1271static const struct idle_cpu idle_cpu_snb __initconst = {
1272 .state_table = snb_cstates,
1273 .disable_promotion_to_c1e = true,
1274};
1275
1276static const struct idle_cpu idle_cpu_snx __initconst = {
1277 .state_table = snb_cstates,
1278 .disable_promotion_to_c1e = true,
1279 .use_acpi = true,
1280};
1281
1282static const struct idle_cpu idle_cpu_byt __initconst = {
1283 .state_table = byt_cstates,
1284 .disable_promotion_to_c1e = true,
1285 .byt_auto_demotion_disable_flag = true,
1286};
1287
1288static const struct idle_cpu idle_cpu_cht __initconst = {
1289 .state_table = cht_cstates,
1290 .disable_promotion_to_c1e = true,
1291 .byt_auto_demotion_disable_flag = true,
1292};
1293
1294static const struct idle_cpu idle_cpu_ivb __initconst = {
1295 .state_table = ivb_cstates,
1296 .disable_promotion_to_c1e = true,
1297};
1298
1299static const struct idle_cpu idle_cpu_ivt __initconst = {
1300 .state_table = ivt_cstates,
1301 .disable_promotion_to_c1e = true,
1302 .use_acpi = true,
1303};
1304
1305static const struct idle_cpu idle_cpu_hsw __initconst = {
1306 .state_table = hsw_cstates,
1307 .disable_promotion_to_c1e = true,
1308};
1309
1310static const struct idle_cpu idle_cpu_hsx __initconst = {
1311 .state_table = hsw_cstates,
1312 .disable_promotion_to_c1e = true,
1313 .use_acpi = true,
1314};
1315
1316static const struct idle_cpu idle_cpu_bdw __initconst = {
1317 .state_table = bdw_cstates,
1318 .disable_promotion_to_c1e = true,
1319};
1320
1321static const struct idle_cpu idle_cpu_bdx __initconst = {
1322 .state_table = bdw_cstates,
1323 .disable_promotion_to_c1e = true,
1324 .use_acpi = true,
1325};
1326
1327static const struct idle_cpu idle_cpu_skl __initconst = {
1328 .state_table = skl_cstates,
1329 .disable_promotion_to_c1e = true,
1330};
1331
1332static const struct idle_cpu idle_cpu_skx __initconst = {
1333 .state_table = skx_cstates,
1334 .disable_promotion_to_c1e = true,
1335 .use_acpi = true,
1336};
1337
1338static const struct idle_cpu idle_cpu_icx __initconst = {
1339 .state_table = icx_cstates,
1340 .disable_promotion_to_c1e = true,
1341 .use_acpi = true,
1342};
1343
1344static const struct idle_cpu idle_cpu_adl __initconst = {
1345 .state_table = adl_cstates,
1346};
1347
1348static const struct idle_cpu idle_cpu_adl_l __initconst = {
1349 .state_table = adl_l_cstates,
1350};
1351
1352static const struct idle_cpu idle_cpu_gmt __initconst = {
1353 .state_table = gmt_cstates,
1354};
1355
1356static const struct idle_cpu idle_cpu_spr __initconst = {
1357 .state_table = spr_cstates,
1358 .disable_promotion_to_c1e = true,
1359 .use_acpi = true,
1360};
1361
1362static const struct idle_cpu idle_cpu_avn __initconst = {
1363 .state_table = avn_cstates,
1364 .disable_promotion_to_c1e = true,
1365 .use_acpi = true,
1366};
1367
1368static const struct idle_cpu idle_cpu_knl __initconst = {
1369 .state_table = knl_cstates,
1370 .use_acpi = true,
1371};
1372
1373static const struct idle_cpu idle_cpu_bxt __initconst = {
1374 .state_table = bxt_cstates,
1375 .disable_promotion_to_c1e = true,
1376};
1377
1378static const struct idle_cpu idle_cpu_dnv __initconst = {
1379 .state_table = dnv_cstates,
1380 .disable_promotion_to_c1e = true,
1381 .use_acpi = true,
1382};
1383
1384static const struct idle_cpu idle_cpu_snr __initconst = {
1385 .state_table = snr_cstates,
1386 .disable_promotion_to_c1e = true,
1387 .use_acpi = true,
1388};
1389
1390static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1391 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EP, &idle_cpu_nhx),
1392 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &idle_cpu_nehalem),
1393 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_G, &idle_cpu_nehalem),
1394 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE, &idle_cpu_nehalem),
1395 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EP, &idle_cpu_nhx),
1396 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EX, &idle_cpu_nhx),
1397 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL, &idle_cpu_atom),
1398 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL_MID, &idle_cpu_lincroft),
1399 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EX, &idle_cpu_nhx),
1400 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE, &idle_cpu_snb),
1401 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &idle_cpu_snx),
1402 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SALTWELL, &idle_cpu_atom),
1403 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT, &idle_cpu_byt),
1404 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1405 X86_MATCH_INTEL_FAM6_MODEL(ATOM_AIRMONT, &idle_cpu_cht),
1406 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE, &idle_cpu_ivb),
1407 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &idle_cpu_ivt),
1408 X86_MATCH_INTEL_FAM6_MODEL(HASWELL, &idle_cpu_hsw),
1409 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &idle_cpu_hsx),
1410 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_L, &idle_cpu_hsw),
1411 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_G, &idle_cpu_hsw),
1412 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_D, &idle_cpu_avn),
1413 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL, &idle_cpu_bdw),
1414 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_G, &idle_cpu_bdw),
1415 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &idle_cpu_bdx),
1416 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &idle_cpu_bdx),
1417 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_L, &idle_cpu_skl),
1418 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE, &idle_cpu_skl),
1419 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE_L, &idle_cpu_skl),
1420 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE, &idle_cpu_skl),
1421 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &idle_cpu_skx),
1422 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &idle_cpu_icx),
1423 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &idle_cpu_icx),
1424 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE, &idle_cpu_adl),
1425 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_L, &idle_cpu_adl_l),
1426 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GRACEMONT, &idle_cpu_gmt),
1427 X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &idle_cpu_spr),
1428 X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &idle_cpu_spr),
1429 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &idle_cpu_knl),
1430 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &idle_cpu_knl),
1431 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT, &idle_cpu_bxt),
1432 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1433 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_D, &idle_cpu_dnv),
1434 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, &idle_cpu_snr),
1435 {}
1436};
1437
1438static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1439 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1440 {}
1441};
1442
1443static bool __init intel_idle_max_cstate_reached(int cstate)
1444{
1445 if (cstate + 1 > max_cstate) {
1446 pr_info("max_cstate %d reached\n", max_cstate);
1447 return true;
1448 }
1449 return false;
1450}
1451
1452static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1453{
1454 unsigned long eax = flg2MWAIT(state->flags);
1455
1456 if (boot_cpu_has(X86_FEATURE_ARAT))
1457 return false;
1458
1459 /*
1460 * Switch over to one-shot tick broadcast if the target C-state
1461 * is deeper than C1.
1462 */
1463 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1464}
1465
1466#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1467#include <acpi/processor.h>
1468
1469static bool no_acpi __read_mostly;
1470module_param(no_acpi, bool, 0444);
1471MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1472
1473static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1474module_param_named(use_acpi, force_use_acpi, bool, 0444);
1475MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1476
1477static struct acpi_processor_power acpi_state_table __initdata;
1478
1479/**
1480 * intel_idle_cst_usable - Check if the _CST information can be used.
1481 *
1482 * Check if all of the C-states listed by _CST in the max_cstate range are
1483 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1484 */
1485static bool __init intel_idle_cst_usable(void)
1486{
1487 int cstate, limit;
1488
1489 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1490 acpi_state_table.count);
1491
1492 for (cstate = 1; cstate < limit; cstate++) {
1493 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1494
1495 if (cx->entry_method != ACPI_CSTATE_FFH)
1496 return false;
1497 }
1498
1499 return true;
1500}
1501
1502static bool __init intel_idle_acpi_cst_extract(void)
1503{
1504 unsigned int cpu;
1505
1506 if (no_acpi) {
1507 pr_debug("Not allowed to use ACPI _CST\n");
1508 return false;
1509 }
1510
1511 for_each_possible_cpu(cpu) {
1512 struct acpi_processor *pr = per_cpu(processors, cpu);
1513
1514 if (!pr)
1515 continue;
1516
1517 if (acpi_processor_evaluate_cst(handle: pr->handle, cpu, info: &acpi_state_table))
1518 continue;
1519
1520 acpi_state_table.count++;
1521
1522 if (!intel_idle_cst_usable())
1523 continue;
1524
1525 if (!acpi_processor_claim_cst_control())
1526 break;
1527
1528 return true;
1529 }
1530
1531 acpi_state_table.count = 0;
1532 pr_debug("ACPI _CST not found or not usable\n");
1533 return false;
1534}
1535
1536static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1537{
1538 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1539
1540 /*
1541 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1542 * the interesting states are ACPI_CSTATE_FFH.
1543 */
1544 for (cstate = 1; cstate < limit; cstate++) {
1545 struct acpi_processor_cx *cx;
1546 struct cpuidle_state *state;
1547
1548 if (intel_idle_max_cstate_reached(cstate: cstate - 1))
1549 break;
1550
1551 cx = &acpi_state_table.states[cstate];
1552
1553 state = &drv->states[drv->state_count++];
1554
1555 snprintf(buf: state->name, CPUIDLE_NAME_LEN, fmt: "C%d_ACPI", cstate);
1556 strscpy(p: state->desc, q: cx->desc, CPUIDLE_DESC_LEN);
1557 state->exit_latency = cx->latency;
1558 /*
1559 * For C1-type C-states use the same number for both the exit
1560 * latency and target residency, because that is the case for
1561 * C1 in the majority of the static C-states tables above.
1562 * For the other types of C-states, however, set the target
1563 * residency to 3 times the exit latency which should lead to
1564 * a reasonable balance between energy-efficiency and
1565 * performance in the majority of interesting cases.
1566 */
1567 state->target_residency = cx->latency;
1568 if (cx->type > ACPI_STATE_C1)
1569 state->target_residency *= 3;
1570
1571 state->flags = MWAIT2flg(cx->address);
1572 if (cx->type > ACPI_STATE_C2)
1573 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1574
1575 if (disabled_states_mask & BIT(cstate))
1576 state->flags |= CPUIDLE_FLAG_OFF;
1577
1578 if (intel_idle_state_needs_timer_stop(state))
1579 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1580
1581 state->enter = intel_idle;
1582 state->enter_s2idle = intel_idle_s2idle;
1583 }
1584}
1585
1586static bool __init intel_idle_off_by_default(u32 mwait_hint)
1587{
1588 int cstate, limit;
1589
1590 /*
1591 * If there are no _CST C-states, do not disable any C-states by
1592 * default.
1593 */
1594 if (!acpi_state_table.count)
1595 return false;
1596
1597 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1598 /*
1599 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1600 * the interesting states are ACPI_CSTATE_FFH.
1601 */
1602 for (cstate = 1; cstate < limit; cstate++) {
1603 if (acpi_state_table.states[cstate].address == mwait_hint)
1604 return false;
1605 }
1606 return true;
1607}
1608#else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1609#define force_use_acpi (false)
1610
1611static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1612static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1613static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
1614#endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1615
1616/**
1617 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1618 *
1619 * Tune IVT multi-socket targets.
1620 * Assumption: num_sockets == (max_package_num + 1).
1621 */
1622static void __init ivt_idle_state_table_update(void)
1623{
1624 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1625 int cpu, package_num, num_sockets = 1;
1626
1627 for_each_online_cpu(cpu) {
1628 package_num = topology_physical_package_id(cpu);
1629 if (package_num + 1 > num_sockets) {
1630 num_sockets = package_num + 1;
1631
1632 if (num_sockets > 4) {
1633 cpuidle_state_table = ivt_cstates_8s;
1634 return;
1635 }
1636 }
1637 }
1638
1639 if (num_sockets > 2)
1640 cpuidle_state_table = ivt_cstates_4s;
1641
1642 /* else, 1 and 2 socket systems use default ivt_cstates */
1643}
1644
1645/**
1646 * irtl_2_usec - IRTL to microseconds conversion.
1647 * @irtl: IRTL MSR value.
1648 *
1649 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1650 */
1651static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1652{
1653 static const unsigned int irtl_ns_units[] __initconst = {
1654 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1655 };
1656 unsigned long long ns;
1657
1658 if (!irtl)
1659 return 0;
1660
1661 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1662
1663 return div_u64(dividend: (irtl & 0x3FF) * ns, NSEC_PER_USEC);
1664}
1665
1666/**
1667 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1668 *
1669 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1670 * definitive maximum latency and use the same value for target_residency.
1671 */
1672static void __init bxt_idle_state_table_update(void)
1673{
1674 unsigned long long msr;
1675 unsigned int usec;
1676
1677 rdmsrl(MSR_PKGC6_IRTL, msr);
1678 usec = irtl_2_usec(irtl: msr);
1679 if (usec) {
1680 bxt_cstates[2].exit_latency = usec;
1681 bxt_cstates[2].target_residency = usec;
1682 }
1683
1684 rdmsrl(MSR_PKGC7_IRTL, msr);
1685 usec = irtl_2_usec(irtl: msr);
1686 if (usec) {
1687 bxt_cstates[3].exit_latency = usec;
1688 bxt_cstates[3].target_residency = usec;
1689 }
1690
1691 rdmsrl(MSR_PKGC8_IRTL, msr);
1692 usec = irtl_2_usec(irtl: msr);
1693 if (usec) {
1694 bxt_cstates[4].exit_latency = usec;
1695 bxt_cstates[4].target_residency = usec;
1696 }
1697
1698 rdmsrl(MSR_PKGC9_IRTL, msr);
1699 usec = irtl_2_usec(irtl: msr);
1700 if (usec) {
1701 bxt_cstates[5].exit_latency = usec;
1702 bxt_cstates[5].target_residency = usec;
1703 }
1704
1705 rdmsrl(MSR_PKGC10_IRTL, msr);
1706 usec = irtl_2_usec(irtl: msr);
1707 if (usec) {
1708 bxt_cstates[6].exit_latency = usec;
1709 bxt_cstates[6].target_residency = usec;
1710 }
1711
1712}
1713
1714/**
1715 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1716 *
1717 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1718 */
1719static void __init sklh_idle_state_table_update(void)
1720{
1721 unsigned long long msr;
1722 unsigned int eax, ebx, ecx, edx;
1723
1724
1725 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1726 if (max_cstate <= 7)
1727 return;
1728
1729 /* if PC10 not present in CPUID.MWAIT.EDX */
1730 if ((mwait_substates & (0xF << 28)) == 0)
1731 return;
1732
1733 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1734
1735 /* PC10 is not enabled in PKG C-state limit */
1736 if ((msr & 0xF) != 8)
1737 return;
1738
1739 ecx = 0;
1740 cpuid(op: 7, eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx);
1741
1742 /* if SGX is present */
1743 if (ebx & (1 << 2)) {
1744
1745 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1746
1747 /* if SGX is enabled */
1748 if (msr & (1 << 18))
1749 return;
1750 }
1751
1752 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1753 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1754}
1755
1756/**
1757 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1758 * idle states table.
1759 */
1760static void __init skx_idle_state_table_update(void)
1761{
1762 unsigned long long msr;
1763
1764 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1765
1766 /*
1767 * 000b: C0/C1 (no package C-state support)
1768 * 001b: C2
1769 * 010b: C6 (non-retention)
1770 * 011b: C6 (retention)
1771 * 111b: No Package C state limits.
1772 */
1773 if ((msr & 0x7) < 2) {
1774 /*
1775 * Uses the CC6 + PC0 latency and 3 times of
1776 * latency for target_residency if the PC6
1777 * is disabled in BIOS. This is consistent
1778 * with how intel_idle driver uses _CST
1779 * to set the target_residency.
1780 */
1781 skx_cstates[2].exit_latency = 92;
1782 skx_cstates[2].target_residency = 276;
1783 }
1784}
1785
1786/**
1787 * adl_idle_state_table_update - Adjust AlderLake idle states table.
1788 */
1789static void __init adl_idle_state_table_update(void)
1790{
1791 /* Check if user prefers C1 over C1E. */
1792 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
1793 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
1794 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
1795
1796 /* Disable C1E by clearing the "C1E promotion" bit. */
1797 c1e_promotion = C1E_PROMOTION_DISABLE;
1798 return;
1799 }
1800
1801 /* Make sure C1E is enabled by default */
1802 c1e_promotion = C1E_PROMOTION_ENABLE;
1803}
1804
1805/**
1806 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
1807 */
1808static void __init spr_idle_state_table_update(void)
1809{
1810 unsigned long long msr;
1811
1812 /*
1813 * By default, the C6 state assumes the worst-case scenario of package
1814 * C6. However, if PC6 is disabled, we update the numbers to match
1815 * core C6.
1816 */
1817 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1818
1819 /* Limit value 2 and above allow for PC6. */
1820 if ((msr & 0x7) < 2) {
1821 spr_cstates[2].exit_latency = 190;
1822 spr_cstates[2].target_residency = 600;
1823 }
1824}
1825
1826static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
1827{
1828 unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
1829 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
1830 MWAIT_SUBSTATE_MASK;
1831
1832 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
1833 if (num_substates == 0)
1834 return false;
1835
1836 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1837 mark_tsc_unstable(reason: "TSC halts in idle states deeper than C2");
1838
1839 return true;
1840}
1841
1842static void state_update_enter_method(struct cpuidle_state *state, int cstate)
1843{
1844 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE) {
1845 /*
1846 * Combining with XSTATE with IBRS or IRQ_ENABLE flags
1847 * is not currently supported but this driver.
1848 */
1849 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IBRS);
1850 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
1851 state->enter = intel_idle_xstate;
1852 return;
1853 }
1854
1855 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
1856 ((state->flags & CPUIDLE_FLAG_IBRS) || ibrs_off)) {
1857 /*
1858 * IBRS mitigation requires that C-states are entered
1859 * with interrupts disabled.
1860 */
1861 if (ibrs_off && (state->flags & CPUIDLE_FLAG_IRQ_ENABLE))
1862 state->flags &= ~CPUIDLE_FLAG_IRQ_ENABLE;
1863 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
1864 state->enter = intel_idle_ibrs;
1865 return;
1866 }
1867
1868 if (state->flags & CPUIDLE_FLAG_IRQ_ENABLE) {
1869 state->enter = intel_idle_irq;
1870 return;
1871 }
1872
1873 if (force_irq_on) {
1874 pr_info("forced intel_idle_irq for state %d\n", cstate);
1875 state->enter = intel_idle_irq;
1876 }
1877}
1878
1879static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
1880{
1881 int cstate;
1882
1883 switch (boot_cpu_data.x86_model) {
1884 case INTEL_FAM6_IVYBRIDGE_X:
1885 ivt_idle_state_table_update();
1886 break;
1887 case INTEL_FAM6_ATOM_GOLDMONT:
1888 case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
1889 bxt_idle_state_table_update();
1890 break;
1891 case INTEL_FAM6_SKYLAKE:
1892 sklh_idle_state_table_update();
1893 break;
1894 case INTEL_FAM6_SKYLAKE_X:
1895 skx_idle_state_table_update();
1896 break;
1897 case INTEL_FAM6_SAPPHIRERAPIDS_X:
1898 case INTEL_FAM6_EMERALDRAPIDS_X:
1899 spr_idle_state_table_update();
1900 break;
1901 case INTEL_FAM6_ALDERLAKE:
1902 case INTEL_FAM6_ALDERLAKE_L:
1903 case INTEL_FAM6_ATOM_GRACEMONT:
1904 adl_idle_state_table_update();
1905 break;
1906 }
1907
1908 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1909 struct cpuidle_state *state;
1910 unsigned int mwait_hint;
1911
1912 if (intel_idle_max_cstate_reached(cstate))
1913 break;
1914
1915 if (!cpuidle_state_table[cstate].enter &&
1916 !cpuidle_state_table[cstate].enter_s2idle)
1917 break;
1918
1919 /* If marked as unusable, skip this state. */
1920 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
1921 pr_debug("state %s is disabled\n",
1922 cpuidle_state_table[cstate].name);
1923 continue;
1924 }
1925
1926 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1927 if (!intel_idle_verify_cstate(mwait_hint))
1928 continue;
1929
1930 /* Structure copy. */
1931 drv->states[drv->state_count] = cpuidle_state_table[cstate];
1932 state = &drv->states[drv->state_count];
1933
1934 state_update_enter_method(state, cstate);
1935
1936
1937 if ((disabled_states_mask & BIT(drv->state_count)) ||
1938 ((icpu->use_acpi || force_use_acpi) &&
1939 intel_idle_off_by_default(mwait_hint) &&
1940 !(state->flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
1941 state->flags |= CPUIDLE_FLAG_OFF;
1942
1943 if (intel_idle_state_needs_timer_stop(state))
1944 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1945
1946 drv->state_count++;
1947 }
1948
1949 if (icpu->byt_auto_demotion_disable_flag) {
1950 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, val: 0);
1951 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, val: 0);
1952 }
1953}
1954
1955/**
1956 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1957 * @drv: cpuidle driver structure to initialize.
1958 */
1959static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
1960{
1961 cpuidle_poll_state_init(drv);
1962
1963 if (disabled_states_mask & BIT(0))
1964 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
1965
1966 drv->state_count = 1;
1967
1968 if (icpu)
1969 intel_idle_init_cstates_icpu(drv);
1970 else
1971 intel_idle_init_cstates_acpi(drv);
1972}
1973
1974static void auto_demotion_disable(void)
1975{
1976 unsigned long long msr_bits;
1977
1978 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1979 msr_bits &= ~auto_demotion_disable_flags;
1980 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, val: msr_bits);
1981}
1982
1983static void c1e_promotion_enable(void)
1984{
1985 unsigned long long msr_bits;
1986
1987 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1988 msr_bits |= 0x2;
1989 wrmsrl(MSR_IA32_POWER_CTL, val: msr_bits);
1990}
1991
1992static void c1e_promotion_disable(void)
1993{
1994 unsigned long long msr_bits;
1995
1996 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1997 msr_bits &= ~0x2;
1998 wrmsrl(MSR_IA32_POWER_CTL, val: msr_bits);
1999}
2000
2001/**
2002 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
2003 * @cpu: CPU to initialize.
2004 *
2005 * Register a cpuidle device object for @cpu and update its MSRs in accordance
2006 * with the processor model flags.
2007 */
2008static int intel_idle_cpu_init(unsigned int cpu)
2009{
2010 struct cpuidle_device *dev;
2011
2012 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2013 dev->cpu = cpu;
2014
2015 if (cpuidle_register_device(dev)) {
2016 pr_debug("cpuidle_register_device %d failed!\n", cpu);
2017 return -EIO;
2018 }
2019
2020 if (auto_demotion_disable_flags)
2021 auto_demotion_disable();
2022
2023 if (c1e_promotion == C1E_PROMOTION_ENABLE)
2024 c1e_promotion_enable();
2025 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
2026 c1e_promotion_disable();
2027
2028 return 0;
2029}
2030
2031static int intel_idle_cpu_online(unsigned int cpu)
2032{
2033 struct cpuidle_device *dev;
2034
2035 if (!boot_cpu_has(X86_FEATURE_ARAT))
2036 tick_broadcast_enable();
2037
2038 /*
2039 * Some systems can hotplug a cpu at runtime after
2040 * the kernel has booted, we have to initialize the
2041 * driver in this case
2042 */
2043 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2044 if (!dev->registered)
2045 return intel_idle_cpu_init(cpu);
2046
2047 return 0;
2048}
2049
2050/**
2051 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2052 */
2053static void __init intel_idle_cpuidle_devices_uninit(void)
2054{
2055 int i;
2056
2057 for_each_online_cpu(i)
2058 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2059}
2060
2061static int __init intel_idle_init(void)
2062{
2063 const struct x86_cpu_id *id;
2064 unsigned int eax, ebx, ecx;
2065 int retval;
2066
2067 /* Do not load intel_idle at all for now if idle= is passed */
2068 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2069 return -ENODEV;
2070
2071 if (max_cstate == 0) {
2072 pr_debug("disabled\n");
2073 return -EPERM;
2074 }
2075
2076 id = x86_match_cpu(match: intel_idle_ids);
2077 if (id) {
2078 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2079 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2080 return -ENODEV;
2081 }
2082 } else {
2083 id = x86_match_cpu(match: intel_mwait_ids);
2084 if (!id)
2085 return -ENODEV;
2086 }
2087
2088 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2089 return -ENODEV;
2090
2091 cpuid(CPUID_MWAIT_LEAF, eax: &eax, ebx: &ebx, ecx: &ecx, edx: &mwait_substates);
2092
2093 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2094 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2095 !mwait_substates)
2096 return -ENODEV;
2097
2098 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2099
2100 icpu = (const struct idle_cpu *)id->driver_data;
2101 if (icpu) {
2102 cpuidle_state_table = icpu->state_table;
2103 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2104 if (icpu->disable_promotion_to_c1e)
2105 c1e_promotion = C1E_PROMOTION_DISABLE;
2106 if (icpu->use_acpi || force_use_acpi)
2107 intel_idle_acpi_cst_extract();
2108 } else if (!intel_idle_acpi_cst_extract()) {
2109 return -ENODEV;
2110 }
2111
2112 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2113 boot_cpu_data.x86_model);
2114
2115 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2116 if (!intel_idle_cpuidle_devices)
2117 return -ENOMEM;
2118
2119 intel_idle_cpuidle_driver_init(drv: &intel_idle_driver);
2120
2121 retval = cpuidle_register_driver(drv: &intel_idle_driver);
2122 if (retval) {
2123 struct cpuidle_driver *drv = cpuidle_get_driver();
2124 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2125 drv ? drv->name : "none");
2126 goto init_driver_fail;
2127 }
2128
2129 retval = cpuhp_setup_state(state: CPUHP_AP_ONLINE_DYN, name: "idle/intel:online",
2130 startup: intel_idle_cpu_online, NULL);
2131 if (retval < 0)
2132 goto hp_setup_fail;
2133
2134 pr_debug("Local APIC timer is reliable in %s\n",
2135 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2136
2137 return 0;
2138
2139hp_setup_fail:
2140 intel_idle_cpuidle_devices_uninit();
2141 cpuidle_unregister_driver(drv: &intel_idle_driver);
2142init_driver_fail:
2143 free_percpu(pdata: intel_idle_cpuidle_devices);
2144 return retval;
2145
2146}
2147device_initcall(intel_idle_init);
2148
2149/*
2150 * We are not really modular, but we used to support that. Meaning we also
2151 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2152 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2153 * is the easiest way (currently) to continue doing that.
2154 */
2155module_param(max_cstate, int, 0444);
2156/*
2157 * The positions of the bits that are set in this number are the indices of the
2158 * idle states to be disabled by default (as reflected by the names of the
2159 * corresponding idle state directories in sysfs, "state0", "state1" ...
2160 * "state<i>" ..., where <i> is the index of the given state).
2161 */
2162module_param_named(states_off, disabled_states_mask, uint, 0444);
2163MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2164/*
2165 * Some platforms come with mutually exclusive C-states, so that if one is
2166 * enabled, the other C-states must not be used. Example: C1 and C1E on
2167 * Sapphire Rapids platform. This parameter allows for selecting the
2168 * preferred C-states among the groups of mutually exclusive C-states - the
2169 * selected C-states will be registered, the other C-states from the mutually
2170 * exclusive group won't be registered. If the platform has no mutually
2171 * exclusive C-states, this parameter has no effect.
2172 */
2173module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2174MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");
2175/*
2176 * Debugging option that forces the driver to enter all C-states with
2177 * interrupts enabled. Does not apply to C-states with
2178 * 'CPUIDLE_FLAG_INIT_XSTATE' and 'CPUIDLE_FLAG_IBRS' flags.
2179 */
2180module_param(force_irq_on, bool, 0444);
2181/*
2182 * Force the disabling of IBRS when X86_FEATURE_KERNEL_IBRS is on and
2183 * CPUIDLE_FLAG_IRQ_ENABLE isn't set.
2184 */
2185module_param(ibrs_off, bool, 0444);
2186MODULE_PARM_DESC(ibrs_off, "Disable IBRS when idle");
2187

source code of linux/drivers/idle/intel_idle.c