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
62 static struct cpuidle_driver intel_idle_driver = {
63 .name = "intel_idle",
64 .owner = THIS_MODULE,
65 };
66 /* intel_idle.max_cstate=0 disables driver */
67 static int max_cstate = CPUIDLE_STATE_MAX - 1;
68 static unsigned int disabled_states_mask __read_mostly;
69 static unsigned int preferred_states_mask __read_mostly;
70 static bool force_irq_on __read_mostly;
71 static bool ibrs_off __read_mostly;
72
73 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
74
75 static unsigned long auto_demotion_disable_flags;
76
77 static enum {
78 C1E_PROMOTION_PRESERVE,
79 C1E_PROMOTION_ENABLE,
80 C1E_PROMOTION_DISABLE
81 } c1e_promotion = C1E_PROMOTION_PRESERVE;
82
83 struct 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
96 static const struct idle_cpu *icpu __initdata;
97 static struct cpuidle_state *cpuidle_state_table __initdata;
98
99 static 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 * Ignore the sub-state when matching mwait hints between the ACPI _CST and
125 * custom tables.
126 */
127 #define CPUIDLE_FLAG_PARTIAL_HINT_MATCH BIT(18)
128
129 /*
130 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
131 * the C-state (top nibble) and sub-state (bottom nibble)
132 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
133 *
134 * We store the hint at the top of our "flags" for each state.
135 */
136 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
137 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
138
__intel_idle(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index,bool irqoff)139 static __always_inline int __intel_idle(struct cpuidle_device *dev,
140 struct cpuidle_driver *drv,
141 int index, bool irqoff)
142 {
143 struct cpuidle_state *state = &drv->states[index];
144 unsigned long eax = flg2MWAIT(state->flags);
145 unsigned long ecx = 1*irqoff; /* break on interrupt flag */
146
147 mwait_idle_with_hints(eax, ecx);
148
149 return index;
150 }
151
152 /**
153 * intel_idle - Ask the processor to enter the given idle state.
154 * @dev: cpuidle device of the target CPU.
155 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
156 * @index: Target idle state index.
157 *
158 * Use the MWAIT instruction to notify the processor that the CPU represented by
159 * @dev is idle and it can try to enter the idle state corresponding to @index.
160 *
161 * If the local APIC timer is not known to be reliable in the target idle state,
162 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
163 *
164 * Must be called under local_irq_disable().
165 */
intel_idle(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)166 static __cpuidle int intel_idle(struct cpuidle_device *dev,
167 struct cpuidle_driver *drv, int index)
168 {
169 return __intel_idle(dev, drv, index, true);
170 }
171
intel_idle_irq(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)172 static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
173 struct cpuidle_driver *drv, int index)
174 {
175 return __intel_idle(dev, drv, index, false);
176 }
177
intel_idle_ibrs(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)178 static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
179 struct cpuidle_driver *drv, int index)
180 {
181 bool smt_active = sched_smt_active();
182 u64 spec_ctrl = spec_ctrl_current();
183 int ret;
184
185 if (smt_active)
186 __update_spec_ctrl(0);
187
188 ret = __intel_idle(dev, drv, index, true);
189
190 if (smt_active)
191 __update_spec_ctrl(spec_ctrl);
192
193 return ret;
194 }
195
intel_idle_xstate(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)196 static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
197 struct cpuidle_driver *drv, int index)
198 {
199 fpu_idle_fpregs();
200 return __intel_idle(dev, drv, index, true);
201 }
202
203 /**
204 * intel_idle_s2idle - Ask the processor to enter the given idle state.
205 * @dev: cpuidle device of the target CPU.
206 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
207 * @index: Target idle state index.
208 *
209 * Use the MWAIT instruction to notify the processor that the CPU represented by
210 * @dev is idle and it can try to enter the idle state corresponding to @index.
211 *
212 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
213 * scheduler tick and suspended scheduler clock on the target CPU.
214 */
intel_idle_s2idle(struct cpuidle_device * dev,struct cpuidle_driver * drv,int index)215 static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
216 struct cpuidle_driver *drv, int index)
217 {
218 unsigned long ecx = 1; /* break on interrupt flag */
219 struct cpuidle_state *state = &drv->states[index];
220 unsigned long eax = flg2MWAIT(state->flags);
221
222 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
223 fpu_idle_fpregs();
224
225 mwait_idle_with_hints(eax, ecx);
226
227 return 0;
228 }
229
230 /*
231 * States are indexed by the cstate number,
232 * which is also the index into the MWAIT hint array.
233 * Thus C0 is a dummy.
234 */
235 static struct cpuidle_state nehalem_cstates[] __initdata = {
236 {
237 .name = "C1",
238 .desc = "MWAIT 0x00",
239 .flags = MWAIT2flg(0x00),
240 .exit_latency = 3,
241 .target_residency = 6,
242 .enter = &intel_idle,
243 .enter_s2idle = intel_idle_s2idle, },
244 {
245 .name = "C1E",
246 .desc = "MWAIT 0x01",
247 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
248 .exit_latency = 10,
249 .target_residency = 20,
250 .enter = &intel_idle,
251 .enter_s2idle = intel_idle_s2idle, },
252 {
253 .name = "C3",
254 .desc = "MWAIT 0x10",
255 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
256 .exit_latency = 20,
257 .target_residency = 80,
258 .enter = &intel_idle,
259 .enter_s2idle = intel_idle_s2idle, },
260 {
261 .name = "C6",
262 .desc = "MWAIT 0x20",
263 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
264 .exit_latency = 200,
265 .target_residency = 800,
266 .enter = &intel_idle,
267 .enter_s2idle = intel_idle_s2idle, },
268 {
269 .enter = NULL }
270 };
271
272 static struct cpuidle_state snb_cstates[] __initdata = {
273 {
274 .name = "C1",
275 .desc = "MWAIT 0x00",
276 .flags = MWAIT2flg(0x00),
277 .exit_latency = 2,
278 .target_residency = 2,
279 .enter = &intel_idle,
280 .enter_s2idle = intel_idle_s2idle, },
281 {
282 .name = "C1E",
283 .desc = "MWAIT 0x01",
284 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
285 .exit_latency = 10,
286 .target_residency = 20,
287 .enter = &intel_idle,
288 .enter_s2idle = intel_idle_s2idle, },
289 {
290 .name = "C3",
291 .desc = "MWAIT 0x10",
292 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
293 .exit_latency = 80,
294 .target_residency = 211,
295 .enter = &intel_idle,
296 .enter_s2idle = intel_idle_s2idle, },
297 {
298 .name = "C6",
299 .desc = "MWAIT 0x20",
300 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
301 .exit_latency = 104,
302 .target_residency = 345,
303 .enter = &intel_idle,
304 .enter_s2idle = intel_idle_s2idle, },
305 {
306 .name = "C7",
307 .desc = "MWAIT 0x30",
308 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
309 .exit_latency = 109,
310 .target_residency = 345,
311 .enter = &intel_idle,
312 .enter_s2idle = intel_idle_s2idle, },
313 {
314 .enter = NULL }
315 };
316
317 static struct cpuidle_state byt_cstates[] __initdata = {
318 {
319 .name = "C1",
320 .desc = "MWAIT 0x00",
321 .flags = MWAIT2flg(0x00),
322 .exit_latency = 1,
323 .target_residency = 1,
324 .enter = &intel_idle,
325 .enter_s2idle = intel_idle_s2idle, },
326 {
327 .name = "C6N",
328 .desc = "MWAIT 0x58",
329 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
330 .exit_latency = 300,
331 .target_residency = 275,
332 .enter = &intel_idle,
333 .enter_s2idle = intel_idle_s2idle, },
334 {
335 .name = "C6S",
336 .desc = "MWAIT 0x52",
337 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
338 .exit_latency = 500,
339 .target_residency = 560,
340 .enter = &intel_idle,
341 .enter_s2idle = intel_idle_s2idle, },
342 {
343 .name = "C7",
344 .desc = "MWAIT 0x60",
345 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
346 .exit_latency = 1200,
347 .target_residency = 4000,
348 .enter = &intel_idle,
349 .enter_s2idle = intel_idle_s2idle, },
350 {
351 .name = "C7S",
352 .desc = "MWAIT 0x64",
353 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
354 .exit_latency = 10000,
355 .target_residency = 20000,
356 .enter = &intel_idle,
357 .enter_s2idle = intel_idle_s2idle, },
358 {
359 .enter = NULL }
360 };
361
362 static struct cpuidle_state cht_cstates[] __initdata = {
363 {
364 .name = "C1",
365 .desc = "MWAIT 0x00",
366 .flags = MWAIT2flg(0x00),
367 .exit_latency = 1,
368 .target_residency = 1,
369 .enter = &intel_idle,
370 .enter_s2idle = intel_idle_s2idle, },
371 {
372 .name = "C6N",
373 .desc = "MWAIT 0x58",
374 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
375 .exit_latency = 80,
376 .target_residency = 275,
377 .enter = &intel_idle,
378 .enter_s2idle = intel_idle_s2idle, },
379 {
380 .name = "C6S",
381 .desc = "MWAIT 0x52",
382 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
383 .exit_latency = 200,
384 .target_residency = 560,
385 .enter = &intel_idle,
386 .enter_s2idle = intel_idle_s2idle, },
387 {
388 .name = "C7",
389 .desc = "MWAIT 0x60",
390 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
391 .exit_latency = 1200,
392 .target_residency = 4000,
393 .enter = &intel_idle,
394 .enter_s2idle = intel_idle_s2idle, },
395 {
396 .name = "C7S",
397 .desc = "MWAIT 0x64",
398 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
399 .exit_latency = 10000,
400 .target_residency = 20000,
401 .enter = &intel_idle,
402 .enter_s2idle = intel_idle_s2idle, },
403 {
404 .enter = NULL }
405 };
406
407 static struct cpuidle_state ivb_cstates[] __initdata = {
408 {
409 .name = "C1",
410 .desc = "MWAIT 0x00",
411 .flags = MWAIT2flg(0x00),
412 .exit_latency = 1,
413 .target_residency = 1,
414 .enter = &intel_idle,
415 .enter_s2idle = intel_idle_s2idle, },
416 {
417 .name = "C1E",
418 .desc = "MWAIT 0x01",
419 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
420 .exit_latency = 10,
421 .target_residency = 20,
422 .enter = &intel_idle,
423 .enter_s2idle = intel_idle_s2idle, },
424 {
425 .name = "C3",
426 .desc = "MWAIT 0x10",
427 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
428 .exit_latency = 59,
429 .target_residency = 156,
430 .enter = &intel_idle,
431 .enter_s2idle = intel_idle_s2idle, },
432 {
433 .name = "C6",
434 .desc = "MWAIT 0x20",
435 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
436 .exit_latency = 80,
437 .target_residency = 300,
438 .enter = &intel_idle,
439 .enter_s2idle = intel_idle_s2idle, },
440 {
441 .name = "C7",
442 .desc = "MWAIT 0x30",
443 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
444 .exit_latency = 87,
445 .target_residency = 300,
446 .enter = &intel_idle,
447 .enter_s2idle = intel_idle_s2idle, },
448 {
449 .enter = NULL }
450 };
451
452 static struct cpuidle_state ivt_cstates[] __initdata = {
453 {
454 .name = "C1",
455 .desc = "MWAIT 0x00",
456 .flags = MWAIT2flg(0x00),
457 .exit_latency = 1,
458 .target_residency = 1,
459 .enter = &intel_idle,
460 .enter_s2idle = intel_idle_s2idle, },
461 {
462 .name = "C1E",
463 .desc = "MWAIT 0x01",
464 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
465 .exit_latency = 10,
466 .target_residency = 80,
467 .enter = &intel_idle,
468 .enter_s2idle = intel_idle_s2idle, },
469 {
470 .name = "C3",
471 .desc = "MWAIT 0x10",
472 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
473 .exit_latency = 59,
474 .target_residency = 156,
475 .enter = &intel_idle,
476 .enter_s2idle = intel_idle_s2idle, },
477 {
478 .name = "C6",
479 .desc = "MWAIT 0x20",
480 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
481 .exit_latency = 82,
482 .target_residency = 300,
483 .enter = &intel_idle,
484 .enter_s2idle = intel_idle_s2idle, },
485 {
486 .enter = NULL }
487 };
488
489 static struct cpuidle_state ivt_cstates_4s[] __initdata = {
490 {
491 .name = "C1",
492 .desc = "MWAIT 0x00",
493 .flags = MWAIT2flg(0x00),
494 .exit_latency = 1,
495 .target_residency = 1,
496 .enter = &intel_idle,
497 .enter_s2idle = intel_idle_s2idle, },
498 {
499 .name = "C1E",
500 .desc = "MWAIT 0x01",
501 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
502 .exit_latency = 10,
503 .target_residency = 250,
504 .enter = &intel_idle,
505 .enter_s2idle = intel_idle_s2idle, },
506 {
507 .name = "C3",
508 .desc = "MWAIT 0x10",
509 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
510 .exit_latency = 59,
511 .target_residency = 300,
512 .enter = &intel_idle,
513 .enter_s2idle = intel_idle_s2idle, },
514 {
515 .name = "C6",
516 .desc = "MWAIT 0x20",
517 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
518 .exit_latency = 84,
519 .target_residency = 400,
520 .enter = &intel_idle,
521 .enter_s2idle = intel_idle_s2idle, },
522 {
523 .enter = NULL }
524 };
525
526 static struct cpuidle_state ivt_cstates_8s[] __initdata = {
527 {
528 .name = "C1",
529 .desc = "MWAIT 0x00",
530 .flags = MWAIT2flg(0x00),
531 .exit_latency = 1,
532 .target_residency = 1,
533 .enter = &intel_idle,
534 .enter_s2idle = intel_idle_s2idle, },
535 {
536 .name = "C1E",
537 .desc = "MWAIT 0x01",
538 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
539 .exit_latency = 10,
540 .target_residency = 500,
541 .enter = &intel_idle,
542 .enter_s2idle = intel_idle_s2idle, },
543 {
544 .name = "C3",
545 .desc = "MWAIT 0x10",
546 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
547 .exit_latency = 59,
548 .target_residency = 600,
549 .enter = &intel_idle,
550 .enter_s2idle = intel_idle_s2idle, },
551 {
552 .name = "C6",
553 .desc = "MWAIT 0x20",
554 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
555 .exit_latency = 88,
556 .target_residency = 700,
557 .enter = &intel_idle,
558 .enter_s2idle = intel_idle_s2idle, },
559 {
560 .enter = NULL }
561 };
562
563 static struct cpuidle_state hsw_cstates[] __initdata = {
564 {
565 .name = "C1",
566 .desc = "MWAIT 0x00",
567 .flags = MWAIT2flg(0x00),
568 .exit_latency = 2,
569 .target_residency = 2,
570 .enter = &intel_idle,
571 .enter_s2idle = intel_idle_s2idle, },
572 {
573 .name = "C1E",
574 .desc = "MWAIT 0x01",
575 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
576 .exit_latency = 10,
577 .target_residency = 20,
578 .enter = &intel_idle,
579 .enter_s2idle = intel_idle_s2idle, },
580 {
581 .name = "C3",
582 .desc = "MWAIT 0x10",
583 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
584 .exit_latency = 33,
585 .target_residency = 100,
586 .enter = &intel_idle,
587 .enter_s2idle = intel_idle_s2idle, },
588 {
589 .name = "C6",
590 .desc = "MWAIT 0x20",
591 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
592 .exit_latency = 133,
593 .target_residency = 400,
594 .enter = &intel_idle,
595 .enter_s2idle = intel_idle_s2idle, },
596 {
597 .name = "C7s",
598 .desc = "MWAIT 0x32",
599 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
600 .exit_latency = 166,
601 .target_residency = 500,
602 .enter = &intel_idle,
603 .enter_s2idle = intel_idle_s2idle, },
604 {
605 .name = "C8",
606 .desc = "MWAIT 0x40",
607 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
608 .exit_latency = 300,
609 .target_residency = 900,
610 .enter = &intel_idle,
611 .enter_s2idle = intel_idle_s2idle, },
612 {
613 .name = "C9",
614 .desc = "MWAIT 0x50",
615 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
616 .exit_latency = 600,
617 .target_residency = 1800,
618 .enter = &intel_idle,
619 .enter_s2idle = intel_idle_s2idle, },
620 {
621 .name = "C10",
622 .desc = "MWAIT 0x60",
623 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
624 .exit_latency = 2600,
625 .target_residency = 7700,
626 .enter = &intel_idle,
627 .enter_s2idle = intel_idle_s2idle, },
628 {
629 .enter = NULL }
630 };
631 static struct cpuidle_state bdw_cstates[] __initdata = {
632 {
633 .name = "C1",
634 .desc = "MWAIT 0x00",
635 .flags = MWAIT2flg(0x00),
636 .exit_latency = 2,
637 .target_residency = 2,
638 .enter = &intel_idle,
639 .enter_s2idle = intel_idle_s2idle, },
640 {
641 .name = "C1E",
642 .desc = "MWAIT 0x01",
643 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
644 .exit_latency = 10,
645 .target_residency = 20,
646 .enter = &intel_idle,
647 .enter_s2idle = intel_idle_s2idle, },
648 {
649 .name = "C3",
650 .desc = "MWAIT 0x10",
651 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
652 .exit_latency = 40,
653 .target_residency = 100,
654 .enter = &intel_idle,
655 .enter_s2idle = intel_idle_s2idle, },
656 {
657 .name = "C6",
658 .desc = "MWAIT 0x20",
659 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
660 .exit_latency = 133,
661 .target_residency = 400,
662 .enter = &intel_idle,
663 .enter_s2idle = intel_idle_s2idle, },
664 {
665 .name = "C7s",
666 .desc = "MWAIT 0x32",
667 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
668 .exit_latency = 166,
669 .target_residency = 500,
670 .enter = &intel_idle,
671 .enter_s2idle = intel_idle_s2idle, },
672 {
673 .name = "C8",
674 .desc = "MWAIT 0x40",
675 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
676 .exit_latency = 300,
677 .target_residency = 900,
678 .enter = &intel_idle,
679 .enter_s2idle = intel_idle_s2idle, },
680 {
681 .name = "C9",
682 .desc = "MWAIT 0x50",
683 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
684 .exit_latency = 600,
685 .target_residency = 1800,
686 .enter = &intel_idle,
687 .enter_s2idle = intel_idle_s2idle, },
688 {
689 .name = "C10",
690 .desc = "MWAIT 0x60",
691 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
692 .exit_latency = 2600,
693 .target_residency = 7700,
694 .enter = &intel_idle,
695 .enter_s2idle = intel_idle_s2idle, },
696 {
697 .enter = NULL }
698 };
699
700 static struct cpuidle_state skl_cstates[] __initdata = {
701 {
702 .name = "C1",
703 .desc = "MWAIT 0x00",
704 .flags = MWAIT2flg(0x00),
705 .exit_latency = 2,
706 .target_residency = 2,
707 .enter = &intel_idle,
708 .enter_s2idle = intel_idle_s2idle, },
709 {
710 .name = "C1E",
711 .desc = "MWAIT 0x01",
712 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
713 .exit_latency = 10,
714 .target_residency = 20,
715 .enter = &intel_idle,
716 .enter_s2idle = intel_idle_s2idle, },
717 {
718 .name = "C3",
719 .desc = "MWAIT 0x10",
720 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
721 .exit_latency = 70,
722 .target_residency = 100,
723 .enter = &intel_idle,
724 .enter_s2idle = intel_idle_s2idle, },
725 {
726 .name = "C6",
727 .desc = "MWAIT 0x20",
728 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
729 .exit_latency = 85,
730 .target_residency = 200,
731 .enter = &intel_idle,
732 .enter_s2idle = intel_idle_s2idle, },
733 {
734 .name = "C7s",
735 .desc = "MWAIT 0x33",
736 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
737 .exit_latency = 124,
738 .target_residency = 800,
739 .enter = &intel_idle,
740 .enter_s2idle = intel_idle_s2idle, },
741 {
742 .name = "C8",
743 .desc = "MWAIT 0x40",
744 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
745 .exit_latency = 200,
746 .target_residency = 800,
747 .enter = &intel_idle,
748 .enter_s2idle = intel_idle_s2idle, },
749 {
750 .name = "C9",
751 .desc = "MWAIT 0x50",
752 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
753 .exit_latency = 480,
754 .target_residency = 5000,
755 .enter = &intel_idle,
756 .enter_s2idle = intel_idle_s2idle, },
757 {
758 .name = "C10",
759 .desc = "MWAIT 0x60",
760 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
761 .exit_latency = 890,
762 .target_residency = 5000,
763 .enter = &intel_idle,
764 .enter_s2idle = intel_idle_s2idle, },
765 {
766 .enter = NULL }
767 };
768
769 static struct cpuidle_state skx_cstates[] __initdata = {
770 {
771 .name = "C1",
772 .desc = "MWAIT 0x00",
773 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
774 .exit_latency = 2,
775 .target_residency = 2,
776 .enter = &intel_idle,
777 .enter_s2idle = intel_idle_s2idle, },
778 {
779 .name = "C1E",
780 .desc = "MWAIT 0x01",
781 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
782 .exit_latency = 10,
783 .target_residency = 20,
784 .enter = &intel_idle,
785 .enter_s2idle = intel_idle_s2idle, },
786 {
787 .name = "C6",
788 .desc = "MWAIT 0x20",
789 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
790 .exit_latency = 133,
791 .target_residency = 600,
792 .enter = &intel_idle,
793 .enter_s2idle = intel_idle_s2idle, },
794 {
795 .enter = NULL }
796 };
797
798 static struct cpuidle_state icx_cstates[] __initdata = {
799 {
800 .name = "C1",
801 .desc = "MWAIT 0x00",
802 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
803 .exit_latency = 1,
804 .target_residency = 1,
805 .enter = &intel_idle,
806 .enter_s2idle = intel_idle_s2idle, },
807 {
808 .name = "C1E",
809 .desc = "MWAIT 0x01",
810 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
811 .exit_latency = 4,
812 .target_residency = 4,
813 .enter = &intel_idle,
814 .enter_s2idle = intel_idle_s2idle, },
815 {
816 .name = "C6",
817 .desc = "MWAIT 0x20",
818 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
819 .exit_latency = 170,
820 .target_residency = 600,
821 .enter = &intel_idle,
822 .enter_s2idle = intel_idle_s2idle, },
823 {
824 .enter = NULL }
825 };
826
827 /*
828 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
829 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
830 * But in this case there is effectively no C1, because C1 requests are
831 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
832 * and C1E requests end up with C1, so there is effectively no C1E.
833 *
834 * By default we enable C1E and disable C1 by marking it with
835 * 'CPUIDLE_FLAG_UNUSABLE'.
836 */
837 static struct cpuidle_state adl_cstates[] __initdata = {
838 {
839 .name = "C1",
840 .desc = "MWAIT 0x00",
841 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
842 .exit_latency = 1,
843 .target_residency = 1,
844 .enter = &intel_idle,
845 .enter_s2idle = intel_idle_s2idle, },
846 {
847 .name = "C1E",
848 .desc = "MWAIT 0x01",
849 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
850 .exit_latency = 2,
851 .target_residency = 4,
852 .enter = &intel_idle,
853 .enter_s2idle = intel_idle_s2idle, },
854 {
855 .name = "C6",
856 .desc = "MWAIT 0x20",
857 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
858 .exit_latency = 220,
859 .target_residency = 600,
860 .enter = &intel_idle,
861 .enter_s2idle = intel_idle_s2idle, },
862 {
863 .name = "C8",
864 .desc = "MWAIT 0x40",
865 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
866 .exit_latency = 280,
867 .target_residency = 800,
868 .enter = &intel_idle,
869 .enter_s2idle = intel_idle_s2idle, },
870 {
871 .name = "C10",
872 .desc = "MWAIT 0x60",
873 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
874 .exit_latency = 680,
875 .target_residency = 2000,
876 .enter = &intel_idle,
877 .enter_s2idle = intel_idle_s2idle, },
878 {
879 .enter = NULL }
880 };
881
882 static struct cpuidle_state adl_l_cstates[] __initdata = {
883 {
884 .name = "C1",
885 .desc = "MWAIT 0x00",
886 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
887 .exit_latency = 1,
888 .target_residency = 1,
889 .enter = &intel_idle,
890 .enter_s2idle = intel_idle_s2idle, },
891 {
892 .name = "C1E",
893 .desc = "MWAIT 0x01",
894 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
895 .exit_latency = 2,
896 .target_residency = 4,
897 .enter = &intel_idle,
898 .enter_s2idle = intel_idle_s2idle, },
899 {
900 .name = "C6",
901 .desc = "MWAIT 0x20",
902 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
903 .exit_latency = 170,
904 .target_residency = 500,
905 .enter = &intel_idle,
906 .enter_s2idle = intel_idle_s2idle, },
907 {
908 .name = "C8",
909 .desc = "MWAIT 0x40",
910 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
911 .exit_latency = 200,
912 .target_residency = 600,
913 .enter = &intel_idle,
914 .enter_s2idle = intel_idle_s2idle, },
915 {
916 .name = "C10",
917 .desc = "MWAIT 0x60",
918 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
919 .exit_latency = 230,
920 .target_residency = 700,
921 .enter = &intel_idle,
922 .enter_s2idle = intel_idle_s2idle, },
923 {
924 .enter = NULL }
925 };
926
927 static struct cpuidle_state mtl_l_cstates[] __initdata = {
928 {
929 .name = "C1E",
930 .desc = "MWAIT 0x01",
931 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
932 .exit_latency = 1,
933 .target_residency = 1,
934 .enter = &intel_idle,
935 .enter_s2idle = intel_idle_s2idle, },
936 {
937 .name = "C6",
938 .desc = "MWAIT 0x20",
939 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
940 .exit_latency = 140,
941 .target_residency = 420,
942 .enter = &intel_idle,
943 .enter_s2idle = intel_idle_s2idle, },
944 {
945 .name = "C10",
946 .desc = "MWAIT 0x60",
947 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
948 .exit_latency = 310,
949 .target_residency = 930,
950 .enter = &intel_idle,
951 .enter_s2idle = intel_idle_s2idle, },
952 {
953 .enter = NULL }
954 };
955
956 static struct cpuidle_state gmt_cstates[] __initdata = {
957 {
958 .name = "C1",
959 .desc = "MWAIT 0x00",
960 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
961 .exit_latency = 1,
962 .target_residency = 1,
963 .enter = &intel_idle,
964 .enter_s2idle = intel_idle_s2idle, },
965 {
966 .name = "C1E",
967 .desc = "MWAIT 0x01",
968 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
969 .exit_latency = 2,
970 .target_residency = 4,
971 .enter = &intel_idle,
972 .enter_s2idle = intel_idle_s2idle, },
973 {
974 .name = "C6",
975 .desc = "MWAIT 0x20",
976 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
977 .exit_latency = 195,
978 .target_residency = 585,
979 .enter = &intel_idle,
980 .enter_s2idle = intel_idle_s2idle, },
981 {
982 .name = "C8",
983 .desc = "MWAIT 0x40",
984 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
985 .exit_latency = 260,
986 .target_residency = 1040,
987 .enter = &intel_idle,
988 .enter_s2idle = intel_idle_s2idle, },
989 {
990 .name = "C10",
991 .desc = "MWAIT 0x60",
992 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
993 .exit_latency = 660,
994 .target_residency = 1980,
995 .enter = &intel_idle,
996 .enter_s2idle = intel_idle_s2idle, },
997 {
998 .enter = NULL }
999 };
1000
1001 static struct cpuidle_state spr_cstates[] __initdata = {
1002 {
1003 .name = "C1",
1004 .desc = "MWAIT 0x00",
1005 .flags = MWAIT2flg(0x00),
1006 .exit_latency = 1,
1007 .target_residency = 1,
1008 .enter = &intel_idle,
1009 .enter_s2idle = intel_idle_s2idle, },
1010 {
1011 .name = "C1E",
1012 .desc = "MWAIT 0x01",
1013 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1014 .exit_latency = 2,
1015 .target_residency = 4,
1016 .enter = &intel_idle,
1017 .enter_s2idle = intel_idle_s2idle, },
1018 {
1019 .name = "C6",
1020 .desc = "MWAIT 0x20",
1021 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
1022 CPUIDLE_FLAG_INIT_XSTATE,
1023 .exit_latency = 290,
1024 .target_residency = 800,
1025 .enter = &intel_idle,
1026 .enter_s2idle = intel_idle_s2idle, },
1027 {
1028 .enter = NULL }
1029 };
1030
1031 static struct cpuidle_state gnr_cstates[] __initdata = {
1032 {
1033 .name = "C1",
1034 .desc = "MWAIT 0x00",
1035 .flags = MWAIT2flg(0x00),
1036 .exit_latency = 1,
1037 .target_residency = 1,
1038 .enter = &intel_idle,
1039 .enter_s2idle = intel_idle_s2idle, },
1040 {
1041 .name = "C1E",
1042 .desc = "MWAIT 0x01",
1043 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1044 .exit_latency = 4,
1045 .target_residency = 4,
1046 .enter = &intel_idle,
1047 .enter_s2idle = intel_idle_s2idle, },
1048 {
1049 .name = "C6",
1050 .desc = "MWAIT 0x20",
1051 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
1052 CPUIDLE_FLAG_INIT_XSTATE |
1053 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1054 .exit_latency = 170,
1055 .target_residency = 650,
1056 .enter = &intel_idle,
1057 .enter_s2idle = intel_idle_s2idle, },
1058 {
1059 .name = "C6P",
1060 .desc = "MWAIT 0x21",
1061 .flags = MWAIT2flg(0x21) | CPUIDLE_FLAG_TLB_FLUSHED |
1062 CPUIDLE_FLAG_INIT_XSTATE |
1063 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1064 .exit_latency = 210,
1065 .target_residency = 1000,
1066 .enter = &intel_idle,
1067 .enter_s2idle = intel_idle_s2idle, },
1068 {
1069 .enter = NULL }
1070 };
1071
1072 static struct cpuidle_state atom_cstates[] __initdata = {
1073 {
1074 .name = "C1E",
1075 .desc = "MWAIT 0x00",
1076 .flags = MWAIT2flg(0x00),
1077 .exit_latency = 10,
1078 .target_residency = 20,
1079 .enter = &intel_idle,
1080 .enter_s2idle = intel_idle_s2idle, },
1081 {
1082 .name = "C2",
1083 .desc = "MWAIT 0x10",
1084 .flags = MWAIT2flg(0x10),
1085 .exit_latency = 20,
1086 .target_residency = 80,
1087 .enter = &intel_idle,
1088 .enter_s2idle = intel_idle_s2idle, },
1089 {
1090 .name = "C4",
1091 .desc = "MWAIT 0x30",
1092 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1093 .exit_latency = 100,
1094 .target_residency = 400,
1095 .enter = &intel_idle,
1096 .enter_s2idle = intel_idle_s2idle, },
1097 {
1098 .name = "C6",
1099 .desc = "MWAIT 0x52",
1100 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1101 .exit_latency = 140,
1102 .target_residency = 560,
1103 .enter = &intel_idle,
1104 .enter_s2idle = intel_idle_s2idle, },
1105 {
1106 .enter = NULL }
1107 };
1108 static struct cpuidle_state tangier_cstates[] __initdata = {
1109 {
1110 .name = "C1",
1111 .desc = "MWAIT 0x00",
1112 .flags = MWAIT2flg(0x00),
1113 .exit_latency = 1,
1114 .target_residency = 4,
1115 .enter = &intel_idle,
1116 .enter_s2idle = intel_idle_s2idle, },
1117 {
1118 .name = "C4",
1119 .desc = "MWAIT 0x30",
1120 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1121 .exit_latency = 100,
1122 .target_residency = 400,
1123 .enter = &intel_idle,
1124 .enter_s2idle = intel_idle_s2idle, },
1125 {
1126 .name = "C6",
1127 .desc = "MWAIT 0x52",
1128 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1129 .exit_latency = 140,
1130 .target_residency = 560,
1131 .enter = &intel_idle,
1132 .enter_s2idle = intel_idle_s2idle, },
1133 {
1134 .name = "C7",
1135 .desc = "MWAIT 0x60",
1136 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1137 .exit_latency = 1200,
1138 .target_residency = 4000,
1139 .enter = &intel_idle,
1140 .enter_s2idle = intel_idle_s2idle, },
1141 {
1142 .name = "C9",
1143 .desc = "MWAIT 0x64",
1144 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1145 .exit_latency = 10000,
1146 .target_residency = 20000,
1147 .enter = &intel_idle,
1148 .enter_s2idle = intel_idle_s2idle, },
1149 {
1150 .enter = NULL }
1151 };
1152 static struct cpuidle_state avn_cstates[] __initdata = {
1153 {
1154 .name = "C1",
1155 .desc = "MWAIT 0x00",
1156 .flags = MWAIT2flg(0x00),
1157 .exit_latency = 2,
1158 .target_residency = 2,
1159 .enter = &intel_idle,
1160 .enter_s2idle = intel_idle_s2idle, },
1161 {
1162 .name = "C6",
1163 .desc = "MWAIT 0x51",
1164 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1165 .exit_latency = 15,
1166 .target_residency = 45,
1167 .enter = &intel_idle,
1168 .enter_s2idle = intel_idle_s2idle, },
1169 {
1170 .enter = NULL }
1171 };
1172 static struct cpuidle_state knl_cstates[] __initdata = {
1173 {
1174 .name = "C1",
1175 .desc = "MWAIT 0x00",
1176 .flags = MWAIT2flg(0x00),
1177 .exit_latency = 1,
1178 .target_residency = 2,
1179 .enter = &intel_idle,
1180 .enter_s2idle = intel_idle_s2idle },
1181 {
1182 .name = "C6",
1183 .desc = "MWAIT 0x10",
1184 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1185 .exit_latency = 120,
1186 .target_residency = 500,
1187 .enter = &intel_idle,
1188 .enter_s2idle = intel_idle_s2idle },
1189 {
1190 .enter = NULL }
1191 };
1192
1193 static struct cpuidle_state bxt_cstates[] __initdata = {
1194 {
1195 .name = "C1",
1196 .desc = "MWAIT 0x00",
1197 .flags = MWAIT2flg(0x00),
1198 .exit_latency = 2,
1199 .target_residency = 2,
1200 .enter = &intel_idle,
1201 .enter_s2idle = intel_idle_s2idle, },
1202 {
1203 .name = "C1E",
1204 .desc = "MWAIT 0x01",
1205 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1206 .exit_latency = 10,
1207 .target_residency = 20,
1208 .enter = &intel_idle,
1209 .enter_s2idle = intel_idle_s2idle, },
1210 {
1211 .name = "C6",
1212 .desc = "MWAIT 0x20",
1213 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1214 .exit_latency = 133,
1215 .target_residency = 133,
1216 .enter = &intel_idle,
1217 .enter_s2idle = intel_idle_s2idle, },
1218 {
1219 .name = "C7s",
1220 .desc = "MWAIT 0x31",
1221 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1222 .exit_latency = 155,
1223 .target_residency = 155,
1224 .enter = &intel_idle,
1225 .enter_s2idle = intel_idle_s2idle, },
1226 {
1227 .name = "C8",
1228 .desc = "MWAIT 0x40",
1229 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1230 .exit_latency = 1000,
1231 .target_residency = 1000,
1232 .enter = &intel_idle,
1233 .enter_s2idle = intel_idle_s2idle, },
1234 {
1235 .name = "C9",
1236 .desc = "MWAIT 0x50",
1237 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1238 .exit_latency = 2000,
1239 .target_residency = 2000,
1240 .enter = &intel_idle,
1241 .enter_s2idle = intel_idle_s2idle, },
1242 {
1243 .name = "C10",
1244 .desc = "MWAIT 0x60",
1245 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1246 .exit_latency = 10000,
1247 .target_residency = 10000,
1248 .enter = &intel_idle,
1249 .enter_s2idle = intel_idle_s2idle, },
1250 {
1251 .enter = NULL }
1252 };
1253
1254 static struct cpuidle_state dnv_cstates[] __initdata = {
1255 {
1256 .name = "C1",
1257 .desc = "MWAIT 0x00",
1258 .flags = MWAIT2flg(0x00),
1259 .exit_latency = 2,
1260 .target_residency = 2,
1261 .enter = &intel_idle,
1262 .enter_s2idle = intel_idle_s2idle, },
1263 {
1264 .name = "C1E",
1265 .desc = "MWAIT 0x01",
1266 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1267 .exit_latency = 10,
1268 .target_residency = 20,
1269 .enter = &intel_idle,
1270 .enter_s2idle = intel_idle_s2idle, },
1271 {
1272 .name = "C6",
1273 .desc = "MWAIT 0x20",
1274 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1275 .exit_latency = 50,
1276 .target_residency = 500,
1277 .enter = &intel_idle,
1278 .enter_s2idle = intel_idle_s2idle, },
1279 {
1280 .enter = NULL }
1281 };
1282
1283 /*
1284 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1285 * C6, and this is indicated in the CPUID mwait leaf.
1286 */
1287 static struct cpuidle_state snr_cstates[] __initdata = {
1288 {
1289 .name = "C1",
1290 .desc = "MWAIT 0x00",
1291 .flags = MWAIT2flg(0x00),
1292 .exit_latency = 2,
1293 .target_residency = 2,
1294 .enter = &intel_idle,
1295 .enter_s2idle = intel_idle_s2idle, },
1296 {
1297 .name = "C1E",
1298 .desc = "MWAIT 0x01",
1299 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1300 .exit_latency = 15,
1301 .target_residency = 25,
1302 .enter = &intel_idle,
1303 .enter_s2idle = intel_idle_s2idle, },
1304 {
1305 .name = "C6",
1306 .desc = "MWAIT 0x20",
1307 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1308 .exit_latency = 130,
1309 .target_residency = 500,
1310 .enter = &intel_idle,
1311 .enter_s2idle = intel_idle_s2idle, },
1312 {
1313 .enter = NULL }
1314 };
1315
1316 static struct cpuidle_state grr_cstates[] __initdata = {
1317 {
1318 .name = "C1",
1319 .desc = "MWAIT 0x00",
1320 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1321 .exit_latency = 1,
1322 .target_residency = 1,
1323 .enter = &intel_idle,
1324 .enter_s2idle = intel_idle_s2idle, },
1325 {
1326 .name = "C1E",
1327 .desc = "MWAIT 0x01",
1328 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1329 .exit_latency = 2,
1330 .target_residency = 10,
1331 .enter = &intel_idle,
1332 .enter_s2idle = intel_idle_s2idle, },
1333 {
1334 .name = "C6S",
1335 .desc = "MWAIT 0x22",
1336 .flags = MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED,
1337 .exit_latency = 140,
1338 .target_residency = 500,
1339 .enter = &intel_idle,
1340 .enter_s2idle = intel_idle_s2idle, },
1341 {
1342 .enter = NULL }
1343 };
1344
1345 static struct cpuidle_state srf_cstates[] __initdata = {
1346 {
1347 .name = "C1",
1348 .desc = "MWAIT 0x00",
1349 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1350 .exit_latency = 1,
1351 .target_residency = 1,
1352 .enter = &intel_idle,
1353 .enter_s2idle = intel_idle_s2idle, },
1354 {
1355 .name = "C1E",
1356 .desc = "MWAIT 0x01",
1357 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1358 .exit_latency = 2,
1359 .target_residency = 10,
1360 .enter = &intel_idle,
1361 .enter_s2idle = intel_idle_s2idle, },
1362 {
1363 .name = "C6S",
1364 .desc = "MWAIT 0x22",
1365 .flags = MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED |
1366 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1367 .exit_latency = 270,
1368 .target_residency = 700,
1369 .enter = &intel_idle,
1370 .enter_s2idle = intel_idle_s2idle, },
1371 {
1372 .name = "C6SP",
1373 .desc = "MWAIT 0x23",
1374 .flags = MWAIT2flg(0x23) | CPUIDLE_FLAG_TLB_FLUSHED |
1375 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1376 .exit_latency = 310,
1377 .target_residency = 900,
1378 .enter = &intel_idle,
1379 .enter_s2idle = intel_idle_s2idle, },
1380 {
1381 .enter = NULL }
1382 };
1383
1384 static const struct idle_cpu idle_cpu_nehalem __initconst = {
1385 .state_table = nehalem_cstates,
1386 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1387 .disable_promotion_to_c1e = true,
1388 };
1389
1390 static const struct idle_cpu idle_cpu_nhx __initconst = {
1391 .state_table = nehalem_cstates,
1392 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1393 .disable_promotion_to_c1e = true,
1394 .use_acpi = true,
1395 };
1396
1397 static const struct idle_cpu idle_cpu_atom __initconst = {
1398 .state_table = atom_cstates,
1399 };
1400
1401 static const struct idle_cpu idle_cpu_tangier __initconst = {
1402 .state_table = tangier_cstates,
1403 };
1404
1405 static const struct idle_cpu idle_cpu_lincroft __initconst = {
1406 .state_table = atom_cstates,
1407 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1408 };
1409
1410 static const struct idle_cpu idle_cpu_snb __initconst = {
1411 .state_table = snb_cstates,
1412 .disable_promotion_to_c1e = true,
1413 };
1414
1415 static const struct idle_cpu idle_cpu_snx __initconst = {
1416 .state_table = snb_cstates,
1417 .disable_promotion_to_c1e = true,
1418 .use_acpi = true,
1419 };
1420
1421 static const struct idle_cpu idle_cpu_byt __initconst = {
1422 .state_table = byt_cstates,
1423 .disable_promotion_to_c1e = true,
1424 .byt_auto_demotion_disable_flag = true,
1425 };
1426
1427 static const struct idle_cpu idle_cpu_cht __initconst = {
1428 .state_table = cht_cstates,
1429 .disable_promotion_to_c1e = true,
1430 .byt_auto_demotion_disable_flag = true,
1431 };
1432
1433 static const struct idle_cpu idle_cpu_ivb __initconst = {
1434 .state_table = ivb_cstates,
1435 .disable_promotion_to_c1e = true,
1436 };
1437
1438 static const struct idle_cpu idle_cpu_ivt __initconst = {
1439 .state_table = ivt_cstates,
1440 .disable_promotion_to_c1e = true,
1441 .use_acpi = true,
1442 };
1443
1444 static const struct idle_cpu idle_cpu_hsw __initconst = {
1445 .state_table = hsw_cstates,
1446 .disable_promotion_to_c1e = true,
1447 };
1448
1449 static const struct idle_cpu idle_cpu_hsx __initconst = {
1450 .state_table = hsw_cstates,
1451 .disable_promotion_to_c1e = true,
1452 .use_acpi = true,
1453 };
1454
1455 static const struct idle_cpu idle_cpu_bdw __initconst = {
1456 .state_table = bdw_cstates,
1457 .disable_promotion_to_c1e = true,
1458 };
1459
1460 static const struct idle_cpu idle_cpu_bdx __initconst = {
1461 .state_table = bdw_cstates,
1462 .disable_promotion_to_c1e = true,
1463 .use_acpi = true,
1464 };
1465
1466 static const struct idle_cpu idle_cpu_skl __initconst = {
1467 .state_table = skl_cstates,
1468 .disable_promotion_to_c1e = true,
1469 };
1470
1471 static const struct idle_cpu idle_cpu_skx __initconst = {
1472 .state_table = skx_cstates,
1473 .disable_promotion_to_c1e = true,
1474 .use_acpi = true,
1475 };
1476
1477 static const struct idle_cpu idle_cpu_icx __initconst = {
1478 .state_table = icx_cstates,
1479 .disable_promotion_to_c1e = true,
1480 .use_acpi = true,
1481 };
1482
1483 static const struct idle_cpu idle_cpu_adl __initconst = {
1484 .state_table = adl_cstates,
1485 };
1486
1487 static const struct idle_cpu idle_cpu_adl_l __initconst = {
1488 .state_table = adl_l_cstates,
1489 };
1490
1491 static const struct idle_cpu idle_cpu_mtl_l __initconst = {
1492 .state_table = mtl_l_cstates,
1493 };
1494
1495 static const struct idle_cpu idle_cpu_gmt __initconst = {
1496 .state_table = gmt_cstates,
1497 };
1498
1499 static const struct idle_cpu idle_cpu_spr __initconst = {
1500 .state_table = spr_cstates,
1501 .disable_promotion_to_c1e = true,
1502 .use_acpi = true,
1503 };
1504
1505 static const struct idle_cpu idle_cpu_gnr __initconst = {
1506 .state_table = gnr_cstates,
1507 .disable_promotion_to_c1e = true,
1508 .use_acpi = true,
1509 };
1510
1511 static const struct idle_cpu idle_cpu_avn __initconst = {
1512 .state_table = avn_cstates,
1513 .disable_promotion_to_c1e = true,
1514 .use_acpi = true,
1515 };
1516
1517 static const struct idle_cpu idle_cpu_knl __initconst = {
1518 .state_table = knl_cstates,
1519 .use_acpi = true,
1520 };
1521
1522 static const struct idle_cpu idle_cpu_bxt __initconst = {
1523 .state_table = bxt_cstates,
1524 .disable_promotion_to_c1e = true,
1525 };
1526
1527 static const struct idle_cpu idle_cpu_dnv __initconst = {
1528 .state_table = dnv_cstates,
1529 .disable_promotion_to_c1e = true,
1530 .use_acpi = true,
1531 };
1532
1533 static const struct idle_cpu idle_cpu_tmt __initconst = {
1534 .disable_promotion_to_c1e = true,
1535 };
1536
1537 static const struct idle_cpu idle_cpu_snr __initconst = {
1538 .state_table = snr_cstates,
1539 .disable_promotion_to_c1e = true,
1540 .use_acpi = true,
1541 };
1542
1543 static const struct idle_cpu idle_cpu_grr __initconst = {
1544 .state_table = grr_cstates,
1545 .disable_promotion_to_c1e = true,
1546 .use_acpi = true,
1547 };
1548
1549 static const struct idle_cpu idle_cpu_srf __initconst = {
1550 .state_table = srf_cstates,
1551 .disable_promotion_to_c1e = true,
1552 .use_acpi = true,
1553 };
1554
1555 static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1556 X86_MATCH_VFM(INTEL_NEHALEM_EP, &idle_cpu_nhx),
1557 X86_MATCH_VFM(INTEL_NEHALEM, &idle_cpu_nehalem),
1558 X86_MATCH_VFM(INTEL_NEHALEM_G, &idle_cpu_nehalem),
1559 X86_MATCH_VFM(INTEL_WESTMERE, &idle_cpu_nehalem),
1560 X86_MATCH_VFM(INTEL_WESTMERE_EP, &idle_cpu_nhx),
1561 X86_MATCH_VFM(INTEL_NEHALEM_EX, &idle_cpu_nhx),
1562 X86_MATCH_VFM(INTEL_ATOM_BONNELL, &idle_cpu_atom),
1563 X86_MATCH_VFM(INTEL_ATOM_BONNELL_MID, &idle_cpu_lincroft),
1564 X86_MATCH_VFM(INTEL_WESTMERE_EX, &idle_cpu_nhx),
1565 X86_MATCH_VFM(INTEL_SANDYBRIDGE, &idle_cpu_snb),
1566 X86_MATCH_VFM(INTEL_SANDYBRIDGE_X, &idle_cpu_snx),
1567 X86_MATCH_VFM(INTEL_ATOM_SALTWELL, &idle_cpu_atom),
1568 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT, &idle_cpu_byt),
1569 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1570 X86_MATCH_VFM(INTEL_ATOM_AIRMONT, &idle_cpu_cht),
1571 X86_MATCH_VFM(INTEL_IVYBRIDGE, &idle_cpu_ivb),
1572 X86_MATCH_VFM(INTEL_IVYBRIDGE_X, &idle_cpu_ivt),
1573 X86_MATCH_VFM(INTEL_HASWELL, &idle_cpu_hsw),
1574 X86_MATCH_VFM(INTEL_HASWELL_X, &idle_cpu_hsx),
1575 X86_MATCH_VFM(INTEL_HASWELL_L, &idle_cpu_hsw),
1576 X86_MATCH_VFM(INTEL_HASWELL_G, &idle_cpu_hsw),
1577 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_D, &idle_cpu_avn),
1578 X86_MATCH_VFM(INTEL_BROADWELL, &idle_cpu_bdw),
1579 X86_MATCH_VFM(INTEL_BROADWELL_G, &idle_cpu_bdw),
1580 X86_MATCH_VFM(INTEL_BROADWELL_X, &idle_cpu_bdx),
1581 X86_MATCH_VFM(INTEL_BROADWELL_D, &idle_cpu_bdx),
1582 X86_MATCH_VFM(INTEL_SKYLAKE_L, &idle_cpu_skl),
1583 X86_MATCH_VFM(INTEL_SKYLAKE, &idle_cpu_skl),
1584 X86_MATCH_VFM(INTEL_KABYLAKE_L, &idle_cpu_skl),
1585 X86_MATCH_VFM(INTEL_KABYLAKE, &idle_cpu_skl),
1586 X86_MATCH_VFM(INTEL_SKYLAKE_X, &idle_cpu_skx),
1587 X86_MATCH_VFM(INTEL_ICELAKE_X, &idle_cpu_icx),
1588 X86_MATCH_VFM(INTEL_ICELAKE_D, &idle_cpu_icx),
1589 X86_MATCH_VFM(INTEL_ALDERLAKE, &idle_cpu_adl),
1590 X86_MATCH_VFM(INTEL_ALDERLAKE_L, &idle_cpu_adl_l),
1591 X86_MATCH_VFM(INTEL_METEORLAKE_L, &idle_cpu_mtl_l),
1592 X86_MATCH_VFM(INTEL_ATOM_GRACEMONT, &idle_cpu_gmt),
1593 X86_MATCH_VFM(INTEL_SAPPHIRERAPIDS_X, &idle_cpu_spr),
1594 X86_MATCH_VFM(INTEL_EMERALDRAPIDS_X, &idle_cpu_spr),
1595 X86_MATCH_VFM(INTEL_GRANITERAPIDS_X, &idle_cpu_gnr),
1596 X86_MATCH_VFM(INTEL_XEON_PHI_KNL, &idle_cpu_knl),
1597 X86_MATCH_VFM(INTEL_XEON_PHI_KNM, &idle_cpu_knl),
1598 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT, &idle_cpu_bxt),
1599 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1600 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_D, &idle_cpu_dnv),
1601 X86_MATCH_VFM(INTEL_ATOM_TREMONT, &idle_cpu_tmt),
1602 X86_MATCH_VFM(INTEL_ATOM_TREMONT_L, &idle_cpu_tmt),
1603 X86_MATCH_VFM(INTEL_ATOM_TREMONT_D, &idle_cpu_snr),
1604 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT, &idle_cpu_grr),
1605 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT_X, &idle_cpu_srf),
1606 {}
1607 };
1608
1609 static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1610 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1611 {}
1612 };
1613
intel_idle_max_cstate_reached(int cstate)1614 static bool __init intel_idle_max_cstate_reached(int cstate)
1615 {
1616 if (cstate + 1 > max_cstate) {
1617 pr_info("max_cstate %d reached\n", max_cstate);
1618 return true;
1619 }
1620 return false;
1621 }
1622
intel_idle_state_needs_timer_stop(struct cpuidle_state * state)1623 static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1624 {
1625 unsigned long eax = flg2MWAIT(state->flags);
1626
1627 if (boot_cpu_has(X86_FEATURE_ARAT))
1628 return false;
1629
1630 /*
1631 * Switch over to one-shot tick broadcast if the target C-state
1632 * is deeper than C1.
1633 */
1634 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1635 }
1636
1637 #ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1638 #include <acpi/processor.h>
1639
1640 static bool no_acpi __read_mostly;
1641 module_param(no_acpi, bool, 0444);
1642 MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1643
1644 static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1645 module_param_named(use_acpi, force_use_acpi, bool, 0444);
1646 MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1647
1648 static struct acpi_processor_power acpi_state_table __initdata;
1649
1650 /**
1651 * intel_idle_cst_usable - Check if the _CST information can be used.
1652 *
1653 * Check if all of the C-states listed by _CST in the max_cstate range are
1654 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1655 */
intel_idle_cst_usable(void)1656 static bool __init intel_idle_cst_usable(void)
1657 {
1658 int cstate, limit;
1659
1660 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1661 acpi_state_table.count);
1662
1663 for (cstate = 1; cstate < limit; cstate++) {
1664 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1665
1666 if (cx->entry_method != ACPI_CSTATE_FFH)
1667 return false;
1668 }
1669
1670 return true;
1671 }
1672
intel_idle_acpi_cst_extract(void)1673 static bool __init intel_idle_acpi_cst_extract(void)
1674 {
1675 unsigned int cpu;
1676
1677 if (no_acpi) {
1678 pr_debug("Not allowed to use ACPI _CST\n");
1679 return false;
1680 }
1681
1682 for_each_possible_cpu(cpu) {
1683 struct acpi_processor *pr = per_cpu(processors, cpu);
1684
1685 if (!pr)
1686 continue;
1687
1688 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1689 continue;
1690
1691 acpi_state_table.count++;
1692
1693 if (!intel_idle_cst_usable())
1694 continue;
1695
1696 if (!acpi_processor_claim_cst_control())
1697 break;
1698
1699 return true;
1700 }
1701
1702 acpi_state_table.count = 0;
1703 pr_debug("ACPI _CST not found or not usable\n");
1704 return false;
1705 }
1706
intel_idle_init_cstates_acpi(struct cpuidle_driver * drv)1707 static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1708 {
1709 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1710
1711 /*
1712 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1713 * the interesting states are ACPI_CSTATE_FFH.
1714 */
1715 for (cstate = 1; cstate < limit; cstate++) {
1716 struct acpi_processor_cx *cx;
1717 struct cpuidle_state *state;
1718
1719 if (intel_idle_max_cstate_reached(cstate - 1))
1720 break;
1721
1722 cx = &acpi_state_table.states[cstate];
1723
1724 state = &drv->states[drv->state_count++];
1725
1726 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1727 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1728 state->exit_latency = cx->latency;
1729 /*
1730 * For C1-type C-states use the same number for both the exit
1731 * latency and target residency, because that is the case for
1732 * C1 in the majority of the static C-states tables above.
1733 * For the other types of C-states, however, set the target
1734 * residency to 3 times the exit latency which should lead to
1735 * a reasonable balance between energy-efficiency and
1736 * performance in the majority of interesting cases.
1737 */
1738 state->target_residency = cx->latency;
1739 if (cx->type > ACPI_STATE_C1)
1740 state->target_residency *= 3;
1741
1742 state->flags = MWAIT2flg(cx->address);
1743 if (cx->type > ACPI_STATE_C2)
1744 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1745
1746 if (disabled_states_mask & BIT(cstate))
1747 state->flags |= CPUIDLE_FLAG_OFF;
1748
1749 if (intel_idle_state_needs_timer_stop(state))
1750 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1751
1752 state->enter = intel_idle;
1753 state->enter_s2idle = intel_idle_s2idle;
1754 }
1755 }
1756
intel_idle_off_by_default(unsigned int flags,u32 mwait_hint)1757 static bool __init intel_idle_off_by_default(unsigned int flags, u32 mwait_hint)
1758 {
1759 int cstate, limit;
1760
1761 /*
1762 * If there are no _CST C-states, do not disable any C-states by
1763 * default.
1764 */
1765 if (!acpi_state_table.count)
1766 return false;
1767
1768 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1769 /*
1770 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1771 * the interesting states are ACPI_CSTATE_FFH.
1772 */
1773 for (cstate = 1; cstate < limit; cstate++) {
1774 u32 acpi_hint = acpi_state_table.states[cstate].address;
1775 u32 table_hint = mwait_hint;
1776
1777 if (flags & CPUIDLE_FLAG_PARTIAL_HINT_MATCH) {
1778 acpi_hint &= ~MWAIT_SUBSTATE_MASK;
1779 table_hint &= ~MWAIT_SUBSTATE_MASK;
1780 }
1781
1782 if (acpi_hint == table_hint)
1783 return false;
1784 }
1785 return true;
1786 }
1787 #else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1788 #define force_use_acpi (false)
1789
intel_idle_acpi_cst_extract(void)1790 static inline bool intel_idle_acpi_cst_extract(void) { return false; }
intel_idle_init_cstates_acpi(struct cpuidle_driver * drv)1791 static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
intel_idle_off_by_default(unsigned int flags,u32 mwait_hint)1792 static inline bool intel_idle_off_by_default(unsigned int flags, u32 mwait_hint)
1793 {
1794 return false;
1795 }
1796 #endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1797
1798 /**
1799 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1800 *
1801 * Tune IVT multi-socket targets.
1802 * Assumption: num_sockets == (max_package_num + 1).
1803 */
ivt_idle_state_table_update(void)1804 static void __init ivt_idle_state_table_update(void)
1805 {
1806 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1807 int cpu, package_num, num_sockets = 1;
1808
1809 for_each_online_cpu(cpu) {
1810 package_num = topology_physical_package_id(cpu);
1811 if (package_num + 1 > num_sockets) {
1812 num_sockets = package_num + 1;
1813
1814 if (num_sockets > 4) {
1815 cpuidle_state_table = ivt_cstates_8s;
1816 return;
1817 }
1818 }
1819 }
1820
1821 if (num_sockets > 2)
1822 cpuidle_state_table = ivt_cstates_4s;
1823
1824 /* else, 1 and 2 socket systems use default ivt_cstates */
1825 }
1826
1827 /**
1828 * irtl_2_usec - IRTL to microseconds conversion.
1829 * @irtl: IRTL MSR value.
1830 *
1831 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1832 */
irtl_2_usec(unsigned long long irtl)1833 static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1834 {
1835 static const unsigned int irtl_ns_units[] __initconst = {
1836 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1837 };
1838 unsigned long long ns;
1839
1840 if (!irtl)
1841 return 0;
1842
1843 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1844
1845 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1846 }
1847
1848 /**
1849 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1850 *
1851 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1852 * definitive maximum latency and use the same value for target_residency.
1853 */
bxt_idle_state_table_update(void)1854 static void __init bxt_idle_state_table_update(void)
1855 {
1856 unsigned long long msr;
1857 unsigned int usec;
1858
1859 rdmsrl(MSR_PKGC6_IRTL, msr);
1860 usec = irtl_2_usec(msr);
1861 if (usec) {
1862 bxt_cstates[2].exit_latency = usec;
1863 bxt_cstates[2].target_residency = usec;
1864 }
1865
1866 rdmsrl(MSR_PKGC7_IRTL, msr);
1867 usec = irtl_2_usec(msr);
1868 if (usec) {
1869 bxt_cstates[3].exit_latency = usec;
1870 bxt_cstates[3].target_residency = usec;
1871 }
1872
1873 rdmsrl(MSR_PKGC8_IRTL, msr);
1874 usec = irtl_2_usec(msr);
1875 if (usec) {
1876 bxt_cstates[4].exit_latency = usec;
1877 bxt_cstates[4].target_residency = usec;
1878 }
1879
1880 rdmsrl(MSR_PKGC9_IRTL, msr);
1881 usec = irtl_2_usec(msr);
1882 if (usec) {
1883 bxt_cstates[5].exit_latency = usec;
1884 bxt_cstates[5].target_residency = usec;
1885 }
1886
1887 rdmsrl(MSR_PKGC10_IRTL, msr);
1888 usec = irtl_2_usec(msr);
1889 if (usec) {
1890 bxt_cstates[6].exit_latency = usec;
1891 bxt_cstates[6].target_residency = usec;
1892 }
1893
1894 }
1895
1896 /**
1897 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1898 *
1899 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1900 */
sklh_idle_state_table_update(void)1901 static void __init sklh_idle_state_table_update(void)
1902 {
1903 unsigned long long msr;
1904 unsigned int eax, ebx, ecx, edx;
1905
1906
1907 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1908 if (max_cstate <= 7)
1909 return;
1910
1911 /* if PC10 not present in CPUID.MWAIT.EDX */
1912 if ((mwait_substates & (0xF << 28)) == 0)
1913 return;
1914
1915 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1916
1917 /* PC10 is not enabled in PKG C-state limit */
1918 if ((msr & 0xF) != 8)
1919 return;
1920
1921 ecx = 0;
1922 cpuid(7, &eax, &ebx, &ecx, &edx);
1923
1924 /* if SGX is present */
1925 if (ebx & (1 << 2)) {
1926
1927 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1928
1929 /* if SGX is enabled */
1930 if (msr & (1 << 18))
1931 return;
1932 }
1933
1934 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1935 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1936 }
1937
1938 /**
1939 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1940 * idle states table.
1941 */
skx_idle_state_table_update(void)1942 static void __init skx_idle_state_table_update(void)
1943 {
1944 unsigned long long msr;
1945
1946 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1947
1948 /*
1949 * 000b: C0/C1 (no package C-state support)
1950 * 001b: C2
1951 * 010b: C6 (non-retention)
1952 * 011b: C6 (retention)
1953 * 111b: No Package C state limits.
1954 */
1955 if ((msr & 0x7) < 2) {
1956 /*
1957 * Uses the CC6 + PC0 latency and 3 times of
1958 * latency for target_residency if the PC6
1959 * is disabled in BIOS. This is consistent
1960 * with how intel_idle driver uses _CST
1961 * to set the target_residency.
1962 */
1963 skx_cstates[2].exit_latency = 92;
1964 skx_cstates[2].target_residency = 276;
1965 }
1966 }
1967
1968 /**
1969 * adl_idle_state_table_update - Adjust AlderLake idle states table.
1970 */
adl_idle_state_table_update(void)1971 static void __init adl_idle_state_table_update(void)
1972 {
1973 /* Check if user prefers C1 over C1E. */
1974 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
1975 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
1976 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
1977
1978 /* Disable C1E by clearing the "C1E promotion" bit. */
1979 c1e_promotion = C1E_PROMOTION_DISABLE;
1980 return;
1981 }
1982
1983 /* Make sure C1E is enabled by default */
1984 c1e_promotion = C1E_PROMOTION_ENABLE;
1985 }
1986
1987 /**
1988 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
1989 */
spr_idle_state_table_update(void)1990 static void __init spr_idle_state_table_update(void)
1991 {
1992 unsigned long long msr;
1993
1994 /*
1995 * By default, the C6 state assumes the worst-case scenario of package
1996 * C6. However, if PC6 is disabled, we update the numbers to match
1997 * core C6.
1998 */
1999 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
2000
2001 /* Limit value 2 and above allow for PC6. */
2002 if ((msr & 0x7) < 2) {
2003 spr_cstates[2].exit_latency = 190;
2004 spr_cstates[2].target_residency = 600;
2005 }
2006 }
2007
intel_idle_verify_cstate(unsigned int mwait_hint)2008 static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
2009 {
2010 unsigned int mwait_cstate = (MWAIT_HINT2CSTATE(mwait_hint) + 1) &
2011 MWAIT_CSTATE_MASK;
2012 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
2013 MWAIT_SUBSTATE_MASK;
2014
2015 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
2016 if (num_substates == 0)
2017 return false;
2018
2019 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
2020 mark_tsc_unstable("TSC halts in idle states deeper than C2");
2021
2022 return true;
2023 }
2024
state_update_enter_method(struct cpuidle_state * state,int cstate)2025 static void state_update_enter_method(struct cpuidle_state *state, int cstate)
2026 {
2027 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE) {
2028 /*
2029 * Combining with XSTATE with IBRS or IRQ_ENABLE flags
2030 * is not currently supported but this driver.
2031 */
2032 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IBRS);
2033 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
2034 state->enter = intel_idle_xstate;
2035 return;
2036 }
2037
2038 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
2039 ((state->flags & CPUIDLE_FLAG_IBRS) || ibrs_off)) {
2040 /*
2041 * IBRS mitigation requires that C-states are entered
2042 * with interrupts disabled.
2043 */
2044 if (ibrs_off && (state->flags & CPUIDLE_FLAG_IRQ_ENABLE))
2045 state->flags &= ~CPUIDLE_FLAG_IRQ_ENABLE;
2046 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
2047 state->enter = intel_idle_ibrs;
2048 return;
2049 }
2050
2051 if (state->flags & CPUIDLE_FLAG_IRQ_ENABLE) {
2052 state->enter = intel_idle_irq;
2053 return;
2054 }
2055
2056 if (force_irq_on) {
2057 pr_info("forced intel_idle_irq for state %d\n", cstate);
2058 state->enter = intel_idle_irq;
2059 }
2060 }
2061
intel_idle_init_cstates_icpu(struct cpuidle_driver * drv)2062 static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
2063 {
2064 int cstate;
2065
2066 switch (boot_cpu_data.x86_vfm) {
2067 case INTEL_IVYBRIDGE_X:
2068 ivt_idle_state_table_update();
2069 break;
2070 case INTEL_ATOM_GOLDMONT:
2071 case INTEL_ATOM_GOLDMONT_PLUS:
2072 bxt_idle_state_table_update();
2073 break;
2074 case INTEL_SKYLAKE:
2075 sklh_idle_state_table_update();
2076 break;
2077 case INTEL_SKYLAKE_X:
2078 skx_idle_state_table_update();
2079 break;
2080 case INTEL_SAPPHIRERAPIDS_X:
2081 case INTEL_EMERALDRAPIDS_X:
2082 spr_idle_state_table_update();
2083 break;
2084 case INTEL_ALDERLAKE:
2085 case INTEL_ALDERLAKE_L:
2086 case INTEL_ATOM_GRACEMONT:
2087 adl_idle_state_table_update();
2088 break;
2089 }
2090
2091 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
2092 struct cpuidle_state *state;
2093 unsigned int mwait_hint;
2094
2095 if (intel_idle_max_cstate_reached(cstate))
2096 break;
2097
2098 if (!cpuidle_state_table[cstate].enter &&
2099 !cpuidle_state_table[cstate].enter_s2idle)
2100 break;
2101
2102 /* If marked as unusable, skip this state. */
2103 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
2104 pr_debug("state %s is disabled\n",
2105 cpuidle_state_table[cstate].name);
2106 continue;
2107 }
2108
2109 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
2110 if (!intel_idle_verify_cstate(mwait_hint))
2111 continue;
2112
2113 /* Structure copy. */
2114 drv->states[drv->state_count] = cpuidle_state_table[cstate];
2115 state = &drv->states[drv->state_count];
2116
2117 state_update_enter_method(state, cstate);
2118
2119
2120 if ((disabled_states_mask & BIT(drv->state_count)) ||
2121 ((icpu->use_acpi || force_use_acpi) &&
2122 intel_idle_off_by_default(state->flags, mwait_hint) &&
2123 !(state->flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
2124 state->flags |= CPUIDLE_FLAG_OFF;
2125
2126 if (intel_idle_state_needs_timer_stop(state))
2127 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
2128
2129 drv->state_count++;
2130 }
2131
2132 if (icpu->byt_auto_demotion_disable_flag) {
2133 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
2134 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
2135 }
2136 }
2137
2138 /**
2139 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
2140 * @drv: cpuidle driver structure to initialize.
2141 */
intel_idle_cpuidle_driver_init(struct cpuidle_driver * drv)2142 static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
2143 {
2144 cpuidle_poll_state_init(drv);
2145
2146 if (disabled_states_mask & BIT(0))
2147 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
2148
2149 drv->state_count = 1;
2150
2151 if (icpu && icpu->state_table)
2152 intel_idle_init_cstates_icpu(drv);
2153 else
2154 intel_idle_init_cstates_acpi(drv);
2155 }
2156
auto_demotion_disable(void)2157 static void auto_demotion_disable(void)
2158 {
2159 unsigned long long msr_bits;
2160
2161 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
2162 msr_bits &= ~auto_demotion_disable_flags;
2163 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
2164 }
2165
c1e_promotion_enable(void)2166 static void c1e_promotion_enable(void)
2167 {
2168 unsigned long long msr_bits;
2169
2170 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
2171 msr_bits |= 0x2;
2172 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
2173 }
2174
c1e_promotion_disable(void)2175 static void c1e_promotion_disable(void)
2176 {
2177 unsigned long long msr_bits;
2178
2179 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
2180 msr_bits &= ~0x2;
2181 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
2182 }
2183
2184 /**
2185 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
2186 * @cpu: CPU to initialize.
2187 *
2188 * Register a cpuidle device object for @cpu and update its MSRs in accordance
2189 * with the processor model flags.
2190 */
intel_idle_cpu_init(unsigned int cpu)2191 static int intel_idle_cpu_init(unsigned int cpu)
2192 {
2193 struct cpuidle_device *dev;
2194
2195 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2196 dev->cpu = cpu;
2197
2198 if (cpuidle_register_device(dev)) {
2199 pr_debug("cpuidle_register_device %d failed!\n", cpu);
2200 return -EIO;
2201 }
2202
2203 if (auto_demotion_disable_flags)
2204 auto_demotion_disable();
2205
2206 if (c1e_promotion == C1E_PROMOTION_ENABLE)
2207 c1e_promotion_enable();
2208 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
2209 c1e_promotion_disable();
2210
2211 return 0;
2212 }
2213
intel_idle_cpu_online(unsigned int cpu)2214 static int intel_idle_cpu_online(unsigned int cpu)
2215 {
2216 struct cpuidle_device *dev;
2217
2218 if (!boot_cpu_has(X86_FEATURE_ARAT))
2219 tick_broadcast_enable();
2220
2221 /*
2222 * Some systems can hotplug a cpu at runtime after
2223 * the kernel has booted, we have to initialize the
2224 * driver in this case
2225 */
2226 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2227 if (!dev->registered)
2228 return intel_idle_cpu_init(cpu);
2229
2230 return 0;
2231 }
2232
2233 /**
2234 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2235 */
intel_idle_cpuidle_devices_uninit(void)2236 static void __init intel_idle_cpuidle_devices_uninit(void)
2237 {
2238 int i;
2239
2240 for_each_online_cpu(i)
2241 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2242 }
2243
intel_idle_init(void)2244 static int __init intel_idle_init(void)
2245 {
2246 const struct x86_cpu_id *id;
2247 unsigned int eax, ebx, ecx;
2248 int retval;
2249
2250 /* Do not load intel_idle at all for now if idle= is passed */
2251 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2252 return -ENODEV;
2253
2254 if (max_cstate == 0) {
2255 pr_debug("disabled\n");
2256 return -EPERM;
2257 }
2258
2259 id = x86_match_cpu(intel_idle_ids);
2260 if (id) {
2261 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2262 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2263 return -ENODEV;
2264 }
2265 } else {
2266 id = x86_match_cpu(intel_mwait_ids);
2267 if (!id)
2268 return -ENODEV;
2269 }
2270
2271 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2272 return -ENODEV;
2273
2274 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
2275
2276 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2277 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2278 !mwait_substates)
2279 return -ENODEV;
2280
2281 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2282
2283 icpu = (const struct idle_cpu *)id->driver_data;
2284 if (icpu) {
2285 if (icpu->state_table)
2286 cpuidle_state_table = icpu->state_table;
2287 else if (!intel_idle_acpi_cst_extract())
2288 return -ENODEV;
2289
2290 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2291 if (icpu->disable_promotion_to_c1e)
2292 c1e_promotion = C1E_PROMOTION_DISABLE;
2293 if (icpu->use_acpi || force_use_acpi)
2294 intel_idle_acpi_cst_extract();
2295 } else if (!intel_idle_acpi_cst_extract()) {
2296 return -ENODEV;
2297 }
2298
2299 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2300 boot_cpu_data.x86_model);
2301
2302 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2303 if (!intel_idle_cpuidle_devices)
2304 return -ENOMEM;
2305
2306 intel_idle_cpuidle_driver_init(&intel_idle_driver);
2307
2308 retval = cpuidle_register_driver(&intel_idle_driver);
2309 if (retval) {
2310 struct cpuidle_driver *drv = cpuidle_get_driver();
2311 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2312 drv ? drv->name : "none");
2313 goto init_driver_fail;
2314 }
2315
2316 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
2317 intel_idle_cpu_online, NULL);
2318 if (retval < 0)
2319 goto hp_setup_fail;
2320
2321 pr_debug("Local APIC timer is reliable in %s\n",
2322 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2323
2324 return 0;
2325
2326 hp_setup_fail:
2327 intel_idle_cpuidle_devices_uninit();
2328 cpuidle_unregister_driver(&intel_idle_driver);
2329 init_driver_fail:
2330 free_percpu(intel_idle_cpuidle_devices);
2331 return retval;
2332
2333 }
2334 device_initcall(intel_idle_init);
2335
2336 /*
2337 * We are not really modular, but we used to support that. Meaning we also
2338 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2339 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2340 * is the easiest way (currently) to continue doing that.
2341 */
2342 module_param(max_cstate, int, 0444);
2343 /*
2344 * The positions of the bits that are set in this number are the indices of the
2345 * idle states to be disabled by default (as reflected by the names of the
2346 * corresponding idle state directories in sysfs, "state0", "state1" ...
2347 * "state<i>" ..., where <i> is the index of the given state).
2348 */
2349 module_param_named(states_off, disabled_states_mask, uint, 0444);
2350 MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2351 /*
2352 * Some platforms come with mutually exclusive C-states, so that if one is
2353 * enabled, the other C-states must not be used. Example: C1 and C1E on
2354 * Sapphire Rapids platform. This parameter allows for selecting the
2355 * preferred C-states among the groups of mutually exclusive C-states - the
2356 * selected C-states will be registered, the other C-states from the mutually
2357 * exclusive group won't be registered. If the platform has no mutually
2358 * exclusive C-states, this parameter has no effect.
2359 */
2360 module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2361 MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");
2362 /*
2363 * Debugging option that forces the driver to enter all C-states with
2364 * interrupts enabled. Does not apply to C-states with
2365 * 'CPUIDLE_FLAG_INIT_XSTATE' and 'CPUIDLE_FLAG_IBRS' flags.
2366 */
2367 module_param(force_irq_on, bool, 0444);
2368 /*
2369 * Force the disabling of IBRS when X86_FEATURE_KERNEL_IBRS is on and
2370 * CPUIDLE_FLAG_IRQ_ENABLE isn't set.
2371 */
2372 module_param(ibrs_off, bool, 0444);
2373 MODULE_PARM_DESC(ibrs_off, "Disable IBRS when idle");
2374