Loading...
1/*
2 * intel_idle.c - native hardware idle loop for modern Intel processors
3 *
4 * Copyright (c) 2010, Intel Corporation.
5 * Len Brown <len.brown@intel.com>
6 *
7 * This program is free software; you can redistribute it and/or modify it
8 * under the terms and conditions of the GNU General Public License,
9 * version 2, as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along with
17 * this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
21/*
22 * intel_idle is a cpuidle driver that loads on specific Intel processors
23 * in lieu of the legacy ACPI processor_idle driver. The intent is to
24 * make Linux more efficient on these processors, as intel_idle knows
25 * more than ACPI, as well as make Linux more immune to ACPI BIOS bugs.
26 */
27
28/*
29 * Design Assumptions
30 *
31 * All CPUs have same idle states as boot CPU
32 *
33 * Chipset BM_STS (bus master status) bit is a NOP
34 * for preventing entry into deep C-stats
35 */
36
37/*
38 * Known limitations
39 *
40 * The driver currently initializes for_each_online_cpu() upon modprobe.
41 * It it unaware of subsequent processors hot-added to the system.
42 * This means that if you boot with maxcpus=n and later online
43 * processors above n, those processors will use C1 only.
44 *
45 * ACPI has a .suspend hack to turn off deep c-statees during suspend
46 * to avoid complications with the lapic timer workaround.
47 * Have not seen issues with suspend, but may need same workaround here.
48 *
49 * There is currently no kernel-based automatic probing/loading mechanism
50 * if the driver is built as a module.
51 */
52
53/* un-comment DEBUG to enable pr_debug() statements */
54#define DEBUG
55
56#include <linux/kernel.h>
57#include <linux/cpuidle.h>
58#include <linux/clockchips.h>
59#include <linux/hrtimer.h> /* ktime_get_real() */
60#include <trace/events/power.h>
61#include <linux/sched.h>
62#include <linux/notifier.h>
63#include <linux/cpu.h>
64#include <asm/mwait.h>
65#include <asm/msr.h>
66
67#define INTEL_IDLE_VERSION "0.4"
68#define PREFIX "intel_idle: "
69
70static struct cpuidle_driver intel_idle_driver = {
71 .name = "intel_idle",
72 .owner = THIS_MODULE,
73};
74/* intel_idle.max_cstate=0 disables driver */
75static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
76
77static unsigned int mwait_substates;
78
79#define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
80/* Reliable LAPIC Timer States, bit 1 for C1 etc. */
81static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
82
83static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
84static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
85
86static struct cpuidle_state *cpuidle_state_table;
87
88/*
89 * Hardware C-state auto-demotion may not always be optimal.
90 * Indicate which enable bits to clear here.
91 */
92static unsigned long long auto_demotion_disable_flags;
93
94/*
95 * Set this flag for states where the HW flushes the TLB for us
96 * and so we don't need cross-calls to keep it consistent.
97 * If this flag is set, SW flushes the TLB, so even if the
98 * HW doesn't do the flushing, this flag is safe to use.
99 */
100#define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
101
102/*
103 * States are indexed by the cstate number,
104 * which is also the index into the MWAIT hint array.
105 * Thus C0 is a dummy.
106 */
107static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
108 { /* MWAIT C0 */ },
109 { /* MWAIT C1 */
110 .name = "C1-NHM",
111 .desc = "MWAIT 0x00",
112 .driver_data = (void *) 0x00,
113 .flags = CPUIDLE_FLAG_TIME_VALID,
114 .exit_latency = 3,
115 .target_residency = 6,
116 .enter = &intel_idle },
117 { /* MWAIT C2 */
118 .name = "C3-NHM",
119 .desc = "MWAIT 0x10",
120 .driver_data = (void *) 0x10,
121 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
122 .exit_latency = 20,
123 .target_residency = 80,
124 .enter = &intel_idle },
125 { /* MWAIT C3 */
126 .name = "C6-NHM",
127 .desc = "MWAIT 0x20",
128 .driver_data = (void *) 0x20,
129 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
130 .exit_latency = 200,
131 .target_residency = 800,
132 .enter = &intel_idle },
133};
134
135static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
136 { /* MWAIT C0 */ },
137 { /* MWAIT C1 */
138 .name = "C1-SNB",
139 .desc = "MWAIT 0x00",
140 .driver_data = (void *) 0x00,
141 .flags = CPUIDLE_FLAG_TIME_VALID,
142 .exit_latency = 1,
143 .target_residency = 1,
144 .enter = &intel_idle },
145 { /* MWAIT C2 */
146 .name = "C3-SNB",
147 .desc = "MWAIT 0x10",
148 .driver_data = (void *) 0x10,
149 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
150 .exit_latency = 80,
151 .target_residency = 211,
152 .enter = &intel_idle },
153 { /* MWAIT C3 */
154 .name = "C6-SNB",
155 .desc = "MWAIT 0x20",
156 .driver_data = (void *) 0x20,
157 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
158 .exit_latency = 104,
159 .target_residency = 345,
160 .enter = &intel_idle },
161 { /* MWAIT C4 */
162 .name = "C7-SNB",
163 .desc = "MWAIT 0x30",
164 .driver_data = (void *) 0x30,
165 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
166 .exit_latency = 109,
167 .target_residency = 345,
168 .enter = &intel_idle },
169};
170
171static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
172 { /* MWAIT C0 */ },
173 { /* MWAIT C1 */
174 .name = "C1-ATM",
175 .desc = "MWAIT 0x00",
176 .driver_data = (void *) 0x00,
177 .flags = CPUIDLE_FLAG_TIME_VALID,
178 .exit_latency = 1,
179 .target_residency = 4,
180 .enter = &intel_idle },
181 { /* MWAIT C2 */
182 .name = "C2-ATM",
183 .desc = "MWAIT 0x10",
184 .driver_data = (void *) 0x10,
185 .flags = CPUIDLE_FLAG_TIME_VALID,
186 .exit_latency = 20,
187 .target_residency = 80,
188 .enter = &intel_idle },
189 { /* MWAIT C3 */ },
190 { /* MWAIT C4 */
191 .name = "C4-ATM",
192 .desc = "MWAIT 0x30",
193 .driver_data = (void *) 0x30,
194 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
195 .exit_latency = 100,
196 .target_residency = 400,
197 .enter = &intel_idle },
198 { /* MWAIT C5 */ },
199 { /* MWAIT C6 */
200 .name = "C6-ATM",
201 .desc = "MWAIT 0x52",
202 .driver_data = (void *) 0x52,
203 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
204 .exit_latency = 140,
205 .target_residency = 560,
206 .enter = &intel_idle },
207};
208
209/**
210 * intel_idle
211 * @dev: cpuidle_device
212 * @state: cpuidle state
213 *
214 */
215static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
216{
217 unsigned long ecx = 1; /* break on interrupt flag */
218 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
219 unsigned int cstate;
220 ktime_t kt_before, kt_after;
221 s64 usec_delta;
222 int cpu = smp_processor_id();
223
224 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
225
226 local_irq_disable();
227
228 /*
229 * leave_mm() to avoid costly and often unnecessary wakeups
230 * for flushing the user TLB's associated with the active mm.
231 */
232 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
233 leave_mm(cpu);
234
235 if (!(lapic_timer_reliable_states & (1 << (cstate))))
236 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
237
238 kt_before = ktime_get_real();
239
240 stop_critical_timings();
241 if (!need_resched()) {
242
243 __monitor((void *)¤t_thread_info()->flags, 0, 0);
244 smp_mb();
245 if (!need_resched())
246 __mwait(eax, ecx);
247 }
248
249 start_critical_timings();
250
251 kt_after = ktime_get_real();
252 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
253
254 local_irq_enable();
255
256 if (!(lapic_timer_reliable_states & (1 << (cstate))))
257 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
258
259 return usec_delta;
260}
261
262static void __setup_broadcast_timer(void *arg)
263{
264 unsigned long reason = (unsigned long)arg;
265 int cpu = smp_processor_id();
266
267 reason = reason ?
268 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
269
270 clockevents_notify(reason, &cpu);
271}
272
273static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
274 unsigned long action, void *hcpu)
275{
276 int hotcpu = (unsigned long)hcpu;
277
278 switch (action & 0xf) {
279 case CPU_ONLINE:
280 smp_call_function_single(hotcpu, __setup_broadcast_timer,
281 (void *)true, 1);
282 break;
283 }
284 return NOTIFY_OK;
285}
286
287static struct notifier_block setup_broadcast_notifier = {
288 .notifier_call = setup_broadcast_cpuhp_notify,
289};
290
291static void auto_demotion_disable(void *dummy)
292{
293 unsigned long long msr_bits;
294
295 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
296 msr_bits &= ~auto_demotion_disable_flags;
297 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
298}
299
300/*
301 * intel_idle_probe()
302 */
303static int intel_idle_probe(void)
304{
305 unsigned int eax, ebx, ecx;
306
307 if (max_cstate == 0) {
308 pr_debug(PREFIX "disabled\n");
309 return -EPERM;
310 }
311
312 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
313 return -ENODEV;
314
315 if (!boot_cpu_has(X86_FEATURE_MWAIT))
316 return -ENODEV;
317
318 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
319 return -ENODEV;
320
321 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
322
323 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
324 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
325 return -ENODEV;
326
327 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
328
329
330 if (boot_cpu_data.x86 != 6) /* family 6 */
331 return -ENODEV;
332
333 switch (boot_cpu_data.x86_model) {
334
335 case 0x1A: /* Core i7, Xeon 5500 series */
336 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
337 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
338 case 0x2E: /* Nehalem-EX Xeon */
339 case 0x2F: /* Westmere-EX Xeon */
340 case 0x25: /* Westmere */
341 case 0x2C: /* Westmere */
342 cpuidle_state_table = nehalem_cstates;
343 auto_demotion_disable_flags =
344 (NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE);
345 break;
346
347 case 0x1C: /* 28 - Atom Processor */
348 cpuidle_state_table = atom_cstates;
349 break;
350
351 case 0x26: /* 38 - Lincroft Atom Processor */
352 cpuidle_state_table = atom_cstates;
353 auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE;
354 break;
355
356 case 0x2A: /* SNB */
357 case 0x2D: /* SNB Xeon */
358 cpuidle_state_table = snb_cstates;
359 break;
360
361 default:
362 pr_debug(PREFIX "does not run on family %d model %d\n",
363 boot_cpu_data.x86, boot_cpu_data.x86_model);
364 return -ENODEV;
365 }
366
367 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
368 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
369 else {
370 smp_call_function(__setup_broadcast_timer, (void *)true, 1);
371 register_cpu_notifier(&setup_broadcast_notifier);
372 }
373
374 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
375 " model 0x%X\n", boot_cpu_data.x86_model);
376
377 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
378 lapic_timer_reliable_states);
379 return 0;
380}
381
382/*
383 * intel_idle_cpuidle_devices_uninit()
384 * unregister, free cpuidle_devices
385 */
386static void intel_idle_cpuidle_devices_uninit(void)
387{
388 int i;
389 struct cpuidle_device *dev;
390
391 for_each_online_cpu(i) {
392 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
393 cpuidle_unregister_device(dev);
394 }
395
396 free_percpu(intel_idle_cpuidle_devices);
397 return;
398}
399/*
400 * intel_idle_cpuidle_devices_init()
401 * allocate, initialize, register cpuidle_devices
402 */
403static int intel_idle_cpuidle_devices_init(void)
404{
405 int i, cstate;
406 struct cpuidle_device *dev;
407
408 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
409 if (intel_idle_cpuidle_devices == NULL)
410 return -ENOMEM;
411
412 for_each_online_cpu(i) {
413 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
414
415 dev->state_count = 1;
416
417 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
418 int num_substates;
419
420 if (cstate > max_cstate) {
421 printk(PREFIX "max_cstate %d reached\n",
422 max_cstate);
423 break;
424 }
425
426 /* does the state exist in CPUID.MWAIT? */
427 num_substates = (mwait_substates >> ((cstate) * 4))
428 & MWAIT_SUBSTATE_MASK;
429 if (num_substates == 0)
430 continue;
431 /* is the state not enabled? */
432 if (cpuidle_state_table[cstate].enter == NULL) {
433 /* does the driver not know about the state? */
434 if (*cpuidle_state_table[cstate].name == '\0')
435 pr_debug(PREFIX "unaware of model 0x%x"
436 " MWAIT %d please"
437 " contact lenb@kernel.org",
438 boot_cpu_data.x86_model, cstate);
439 continue;
440 }
441
442 if ((cstate > 2) &&
443 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
444 mark_tsc_unstable("TSC halts in idle"
445 " states deeper than C2");
446
447 dev->states[dev->state_count] = /* structure copy */
448 cpuidle_state_table[cstate];
449
450 dev->state_count += 1;
451 }
452
453 dev->cpu = i;
454 if (cpuidle_register_device(dev)) {
455 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
456 i);
457 intel_idle_cpuidle_devices_uninit();
458 return -EIO;
459 }
460 }
461 if (auto_demotion_disable_flags)
462 smp_call_function(auto_demotion_disable, NULL, 1);
463
464 return 0;
465}
466
467
468static int __init intel_idle_init(void)
469{
470 int retval;
471
472 /* Do not load intel_idle at all for now if idle= is passed */
473 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
474 return -ENODEV;
475
476 retval = intel_idle_probe();
477 if (retval)
478 return retval;
479
480 retval = cpuidle_register_driver(&intel_idle_driver);
481 if (retval) {
482 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
483 cpuidle_get_driver()->name);
484 return retval;
485 }
486
487 retval = intel_idle_cpuidle_devices_init();
488 if (retval) {
489 cpuidle_unregister_driver(&intel_idle_driver);
490 return retval;
491 }
492
493 return 0;
494}
495
496static void __exit intel_idle_exit(void)
497{
498 intel_idle_cpuidle_devices_uninit();
499 cpuidle_unregister_driver(&intel_idle_driver);
500
501 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
502 smp_call_function(__setup_broadcast_timer, (void *)false, 1);
503 unregister_cpu_notifier(&setup_broadcast_notifier);
504 }
505
506 return;
507}
508
509module_init(intel_idle_init);
510module_exit(intel_idle_exit);
511
512module_param(max_cstate, int, 0444);
513
514MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
515MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
516MODULE_LICENSE("GPL");
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/tsc.h>
59#include <asm/fpu/api.h>
60
61#define INTEL_IDLE_VERSION "0.5.1"
62
63static struct cpuidle_driver intel_idle_driver = {
64 .name = "intel_idle",
65 .owner = THIS_MODULE,
66};
67/* intel_idle.max_cstate=0 disables driver */
68static int max_cstate = CPUIDLE_STATE_MAX - 1;
69static unsigned int disabled_states_mask __read_mostly;
70static unsigned int preferred_states_mask __read_mostly;
71static bool force_irq_on __read_mostly;
72static bool ibrs_off __read_mostly;
73
74static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
75
76static unsigned long auto_demotion_disable_flags;
77
78static enum {
79 C1E_PROMOTION_PRESERVE,
80 C1E_PROMOTION_ENABLE,
81 C1E_PROMOTION_DISABLE
82} c1e_promotion = C1E_PROMOTION_PRESERVE;
83
84struct idle_cpu {
85 struct cpuidle_state *state_table;
86
87 /*
88 * Hardware C-state auto-demotion may not always be optimal.
89 * Indicate which enable bits to clear here.
90 */
91 unsigned long auto_demotion_disable_flags;
92 bool byt_auto_demotion_disable_flag;
93 bool disable_promotion_to_c1e;
94 bool use_acpi;
95};
96
97static const struct idle_cpu *icpu __initdata;
98static struct cpuidle_state *cpuidle_state_table __initdata;
99
100static unsigned int mwait_substates __initdata;
101
102/*
103 * Enable interrupts before entering the C-state. On some platforms and for
104 * some C-states, this may measurably decrease interrupt latency.
105 */
106#define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
107
108/*
109 * Enable this state by default even if the ACPI _CST does not list it.
110 */
111#define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
112
113/*
114 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
115 * above.
116 */
117#define CPUIDLE_FLAG_IBRS BIT(16)
118
119/*
120 * Initialize large xstate for the C6-state entrance.
121 */
122#define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
123
124/*
125 * Ignore the sub-state when matching mwait hints between the ACPI _CST and
126 * custom tables.
127 */
128#define CPUIDLE_FLAG_PARTIAL_HINT_MATCH BIT(18)
129
130/*
131 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
132 * the C-state (top nibble) and sub-state (bottom nibble)
133 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
134 *
135 * We store the hint at the top of our "flags" for each state.
136 */
137#define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
138#define MWAIT2flg(eax) ((eax & 0xFF) << 24)
139
140static __always_inline int __intel_idle(struct cpuidle_device *dev,
141 struct cpuidle_driver *drv,
142 int index, bool irqoff)
143{
144 struct cpuidle_state *state = &drv->states[index];
145 unsigned long eax = flg2MWAIT(state->flags);
146 unsigned long ecx = 1*irqoff; /* break on interrupt flag */
147
148 mwait_idle_with_hints(eax, ecx);
149
150 return index;
151}
152
153/**
154 * intel_idle - Ask the processor to enter the given idle state.
155 * @dev: cpuidle device of the target CPU.
156 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
157 * @index: Target idle state index.
158 *
159 * Use the MWAIT instruction to notify the processor that the CPU represented by
160 * @dev is idle and it can try to enter the idle state corresponding to @index.
161 *
162 * If the local APIC timer is not known to be reliable in the target idle state,
163 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
164 *
165 * Must be called under local_irq_disable().
166 */
167static __cpuidle int intel_idle(struct cpuidle_device *dev,
168 struct cpuidle_driver *drv, int index)
169{
170 return __intel_idle(dev, drv, index, true);
171}
172
173static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
174 struct cpuidle_driver *drv, int index)
175{
176 return __intel_idle(dev, drv, index, false);
177}
178
179static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
180 struct cpuidle_driver *drv, int index)
181{
182 bool smt_active = sched_smt_active();
183 u64 spec_ctrl = spec_ctrl_current();
184 int ret;
185
186 if (smt_active)
187 __update_spec_ctrl(0);
188
189 ret = __intel_idle(dev, drv, index, true);
190
191 if (smt_active)
192 __update_spec_ctrl(spec_ctrl);
193
194 return ret;
195}
196
197static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
198 struct cpuidle_driver *drv, int index)
199{
200 fpu_idle_fpregs();
201 return __intel_idle(dev, drv, index, true);
202}
203
204/**
205 * intel_idle_s2idle - Ask the processor to enter the given idle state.
206 * @dev: cpuidle device of the target CPU.
207 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
208 * @index: Target idle state index.
209 *
210 * Use the MWAIT instruction to notify the processor that the CPU represented by
211 * @dev is idle and it can try to enter the idle state corresponding to @index.
212 *
213 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
214 * scheduler tick and suspended scheduler clock on the target CPU.
215 */
216static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
217 struct cpuidle_driver *drv, int index)
218{
219 unsigned long ecx = 1; /* break on interrupt flag */
220 struct cpuidle_state *state = &drv->states[index];
221 unsigned long eax = flg2MWAIT(state->flags);
222
223 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
224 fpu_idle_fpregs();
225
226 mwait_idle_with_hints(eax, ecx);
227
228 return 0;
229}
230
231/*
232 * States are indexed by the cstate number,
233 * which is also the index into the MWAIT hint array.
234 * Thus C0 is a dummy.
235 */
236static struct cpuidle_state nehalem_cstates[] __initdata = {
237 {
238 .name = "C1",
239 .desc = "MWAIT 0x00",
240 .flags = MWAIT2flg(0x00),
241 .exit_latency = 3,
242 .target_residency = 6,
243 .enter = &intel_idle,
244 .enter_s2idle = intel_idle_s2idle, },
245 {
246 .name = "C1E",
247 .desc = "MWAIT 0x01",
248 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
249 .exit_latency = 10,
250 .target_residency = 20,
251 .enter = &intel_idle,
252 .enter_s2idle = intel_idle_s2idle, },
253 {
254 .name = "C3",
255 .desc = "MWAIT 0x10",
256 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
257 .exit_latency = 20,
258 .target_residency = 80,
259 .enter = &intel_idle,
260 .enter_s2idle = intel_idle_s2idle, },
261 {
262 .name = "C6",
263 .desc = "MWAIT 0x20",
264 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
265 .exit_latency = 200,
266 .target_residency = 800,
267 .enter = &intel_idle,
268 .enter_s2idle = intel_idle_s2idle, },
269 {
270 .enter = NULL }
271};
272
273static struct cpuidle_state snb_cstates[] __initdata = {
274 {
275 .name = "C1",
276 .desc = "MWAIT 0x00",
277 .flags = MWAIT2flg(0x00),
278 .exit_latency = 2,
279 .target_residency = 2,
280 .enter = &intel_idle,
281 .enter_s2idle = intel_idle_s2idle, },
282 {
283 .name = "C1E",
284 .desc = "MWAIT 0x01",
285 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
286 .exit_latency = 10,
287 .target_residency = 20,
288 .enter = &intel_idle,
289 .enter_s2idle = intel_idle_s2idle, },
290 {
291 .name = "C3",
292 .desc = "MWAIT 0x10",
293 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
294 .exit_latency = 80,
295 .target_residency = 211,
296 .enter = &intel_idle,
297 .enter_s2idle = intel_idle_s2idle, },
298 {
299 .name = "C6",
300 .desc = "MWAIT 0x20",
301 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
302 .exit_latency = 104,
303 .target_residency = 345,
304 .enter = &intel_idle,
305 .enter_s2idle = intel_idle_s2idle, },
306 {
307 .name = "C7",
308 .desc = "MWAIT 0x30",
309 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
310 .exit_latency = 109,
311 .target_residency = 345,
312 .enter = &intel_idle,
313 .enter_s2idle = intel_idle_s2idle, },
314 {
315 .enter = NULL }
316};
317
318static struct cpuidle_state byt_cstates[] __initdata = {
319 {
320 .name = "C1",
321 .desc = "MWAIT 0x00",
322 .flags = MWAIT2flg(0x00),
323 .exit_latency = 1,
324 .target_residency = 1,
325 .enter = &intel_idle,
326 .enter_s2idle = intel_idle_s2idle, },
327 {
328 .name = "C6N",
329 .desc = "MWAIT 0x58",
330 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
331 .exit_latency = 300,
332 .target_residency = 275,
333 .enter = &intel_idle,
334 .enter_s2idle = intel_idle_s2idle, },
335 {
336 .name = "C6S",
337 .desc = "MWAIT 0x52",
338 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
339 .exit_latency = 500,
340 .target_residency = 560,
341 .enter = &intel_idle,
342 .enter_s2idle = intel_idle_s2idle, },
343 {
344 .name = "C7",
345 .desc = "MWAIT 0x60",
346 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
347 .exit_latency = 1200,
348 .target_residency = 4000,
349 .enter = &intel_idle,
350 .enter_s2idle = intel_idle_s2idle, },
351 {
352 .name = "C7S",
353 .desc = "MWAIT 0x64",
354 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
355 .exit_latency = 10000,
356 .target_residency = 20000,
357 .enter = &intel_idle,
358 .enter_s2idle = intel_idle_s2idle, },
359 {
360 .enter = NULL }
361};
362
363static struct cpuidle_state cht_cstates[] __initdata = {
364 {
365 .name = "C1",
366 .desc = "MWAIT 0x00",
367 .flags = MWAIT2flg(0x00),
368 .exit_latency = 1,
369 .target_residency = 1,
370 .enter = &intel_idle,
371 .enter_s2idle = intel_idle_s2idle, },
372 {
373 .name = "C6N",
374 .desc = "MWAIT 0x58",
375 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
376 .exit_latency = 80,
377 .target_residency = 275,
378 .enter = &intel_idle,
379 .enter_s2idle = intel_idle_s2idle, },
380 {
381 .name = "C6S",
382 .desc = "MWAIT 0x52",
383 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
384 .exit_latency = 200,
385 .target_residency = 560,
386 .enter = &intel_idle,
387 .enter_s2idle = intel_idle_s2idle, },
388 {
389 .name = "C7",
390 .desc = "MWAIT 0x60",
391 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
392 .exit_latency = 1200,
393 .target_residency = 4000,
394 .enter = &intel_idle,
395 .enter_s2idle = intel_idle_s2idle, },
396 {
397 .name = "C7S",
398 .desc = "MWAIT 0x64",
399 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
400 .exit_latency = 10000,
401 .target_residency = 20000,
402 .enter = &intel_idle,
403 .enter_s2idle = intel_idle_s2idle, },
404 {
405 .enter = NULL }
406};
407
408static struct cpuidle_state ivb_cstates[] __initdata = {
409 {
410 .name = "C1",
411 .desc = "MWAIT 0x00",
412 .flags = MWAIT2flg(0x00),
413 .exit_latency = 1,
414 .target_residency = 1,
415 .enter = &intel_idle,
416 .enter_s2idle = intel_idle_s2idle, },
417 {
418 .name = "C1E",
419 .desc = "MWAIT 0x01",
420 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
421 .exit_latency = 10,
422 .target_residency = 20,
423 .enter = &intel_idle,
424 .enter_s2idle = intel_idle_s2idle, },
425 {
426 .name = "C3",
427 .desc = "MWAIT 0x10",
428 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
429 .exit_latency = 59,
430 .target_residency = 156,
431 .enter = &intel_idle,
432 .enter_s2idle = intel_idle_s2idle, },
433 {
434 .name = "C6",
435 .desc = "MWAIT 0x20",
436 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
437 .exit_latency = 80,
438 .target_residency = 300,
439 .enter = &intel_idle,
440 .enter_s2idle = intel_idle_s2idle, },
441 {
442 .name = "C7",
443 .desc = "MWAIT 0x30",
444 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
445 .exit_latency = 87,
446 .target_residency = 300,
447 .enter = &intel_idle,
448 .enter_s2idle = intel_idle_s2idle, },
449 {
450 .enter = NULL }
451};
452
453static struct cpuidle_state ivt_cstates[] __initdata = {
454 {
455 .name = "C1",
456 .desc = "MWAIT 0x00",
457 .flags = MWAIT2flg(0x00),
458 .exit_latency = 1,
459 .target_residency = 1,
460 .enter = &intel_idle,
461 .enter_s2idle = intel_idle_s2idle, },
462 {
463 .name = "C1E",
464 .desc = "MWAIT 0x01",
465 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
466 .exit_latency = 10,
467 .target_residency = 80,
468 .enter = &intel_idle,
469 .enter_s2idle = intel_idle_s2idle, },
470 {
471 .name = "C3",
472 .desc = "MWAIT 0x10",
473 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
474 .exit_latency = 59,
475 .target_residency = 156,
476 .enter = &intel_idle,
477 .enter_s2idle = intel_idle_s2idle, },
478 {
479 .name = "C6",
480 .desc = "MWAIT 0x20",
481 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
482 .exit_latency = 82,
483 .target_residency = 300,
484 .enter = &intel_idle,
485 .enter_s2idle = intel_idle_s2idle, },
486 {
487 .enter = NULL }
488};
489
490static struct cpuidle_state ivt_cstates_4s[] __initdata = {
491 {
492 .name = "C1",
493 .desc = "MWAIT 0x00",
494 .flags = MWAIT2flg(0x00),
495 .exit_latency = 1,
496 .target_residency = 1,
497 .enter = &intel_idle,
498 .enter_s2idle = intel_idle_s2idle, },
499 {
500 .name = "C1E",
501 .desc = "MWAIT 0x01",
502 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
503 .exit_latency = 10,
504 .target_residency = 250,
505 .enter = &intel_idle,
506 .enter_s2idle = intel_idle_s2idle, },
507 {
508 .name = "C3",
509 .desc = "MWAIT 0x10",
510 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
511 .exit_latency = 59,
512 .target_residency = 300,
513 .enter = &intel_idle,
514 .enter_s2idle = intel_idle_s2idle, },
515 {
516 .name = "C6",
517 .desc = "MWAIT 0x20",
518 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
519 .exit_latency = 84,
520 .target_residency = 400,
521 .enter = &intel_idle,
522 .enter_s2idle = intel_idle_s2idle, },
523 {
524 .enter = NULL }
525};
526
527static struct cpuidle_state ivt_cstates_8s[] __initdata = {
528 {
529 .name = "C1",
530 .desc = "MWAIT 0x00",
531 .flags = MWAIT2flg(0x00),
532 .exit_latency = 1,
533 .target_residency = 1,
534 .enter = &intel_idle,
535 .enter_s2idle = intel_idle_s2idle, },
536 {
537 .name = "C1E",
538 .desc = "MWAIT 0x01",
539 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
540 .exit_latency = 10,
541 .target_residency = 500,
542 .enter = &intel_idle,
543 .enter_s2idle = intel_idle_s2idle, },
544 {
545 .name = "C3",
546 .desc = "MWAIT 0x10",
547 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
548 .exit_latency = 59,
549 .target_residency = 600,
550 .enter = &intel_idle,
551 .enter_s2idle = intel_idle_s2idle, },
552 {
553 .name = "C6",
554 .desc = "MWAIT 0x20",
555 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
556 .exit_latency = 88,
557 .target_residency = 700,
558 .enter = &intel_idle,
559 .enter_s2idle = intel_idle_s2idle, },
560 {
561 .enter = NULL }
562};
563
564static struct cpuidle_state hsw_cstates[] __initdata = {
565 {
566 .name = "C1",
567 .desc = "MWAIT 0x00",
568 .flags = MWAIT2flg(0x00),
569 .exit_latency = 2,
570 .target_residency = 2,
571 .enter = &intel_idle,
572 .enter_s2idle = intel_idle_s2idle, },
573 {
574 .name = "C1E",
575 .desc = "MWAIT 0x01",
576 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
577 .exit_latency = 10,
578 .target_residency = 20,
579 .enter = &intel_idle,
580 .enter_s2idle = intel_idle_s2idle, },
581 {
582 .name = "C3",
583 .desc = "MWAIT 0x10",
584 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
585 .exit_latency = 33,
586 .target_residency = 100,
587 .enter = &intel_idle,
588 .enter_s2idle = intel_idle_s2idle, },
589 {
590 .name = "C6",
591 .desc = "MWAIT 0x20",
592 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
593 .exit_latency = 133,
594 .target_residency = 400,
595 .enter = &intel_idle,
596 .enter_s2idle = intel_idle_s2idle, },
597 {
598 .name = "C7s",
599 .desc = "MWAIT 0x32",
600 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
601 .exit_latency = 166,
602 .target_residency = 500,
603 .enter = &intel_idle,
604 .enter_s2idle = intel_idle_s2idle, },
605 {
606 .name = "C8",
607 .desc = "MWAIT 0x40",
608 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
609 .exit_latency = 300,
610 .target_residency = 900,
611 .enter = &intel_idle,
612 .enter_s2idle = intel_idle_s2idle, },
613 {
614 .name = "C9",
615 .desc = "MWAIT 0x50",
616 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
617 .exit_latency = 600,
618 .target_residency = 1800,
619 .enter = &intel_idle,
620 .enter_s2idle = intel_idle_s2idle, },
621 {
622 .name = "C10",
623 .desc = "MWAIT 0x60",
624 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
625 .exit_latency = 2600,
626 .target_residency = 7700,
627 .enter = &intel_idle,
628 .enter_s2idle = intel_idle_s2idle, },
629 {
630 .enter = NULL }
631};
632static struct cpuidle_state bdw_cstates[] __initdata = {
633 {
634 .name = "C1",
635 .desc = "MWAIT 0x00",
636 .flags = MWAIT2flg(0x00),
637 .exit_latency = 2,
638 .target_residency = 2,
639 .enter = &intel_idle,
640 .enter_s2idle = intel_idle_s2idle, },
641 {
642 .name = "C1E",
643 .desc = "MWAIT 0x01",
644 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
645 .exit_latency = 10,
646 .target_residency = 20,
647 .enter = &intel_idle,
648 .enter_s2idle = intel_idle_s2idle, },
649 {
650 .name = "C3",
651 .desc = "MWAIT 0x10",
652 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
653 .exit_latency = 40,
654 .target_residency = 100,
655 .enter = &intel_idle,
656 .enter_s2idle = intel_idle_s2idle, },
657 {
658 .name = "C6",
659 .desc = "MWAIT 0x20",
660 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
661 .exit_latency = 133,
662 .target_residency = 400,
663 .enter = &intel_idle,
664 .enter_s2idle = intel_idle_s2idle, },
665 {
666 .name = "C7s",
667 .desc = "MWAIT 0x32",
668 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
669 .exit_latency = 166,
670 .target_residency = 500,
671 .enter = &intel_idle,
672 .enter_s2idle = intel_idle_s2idle, },
673 {
674 .name = "C8",
675 .desc = "MWAIT 0x40",
676 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
677 .exit_latency = 300,
678 .target_residency = 900,
679 .enter = &intel_idle,
680 .enter_s2idle = intel_idle_s2idle, },
681 {
682 .name = "C9",
683 .desc = "MWAIT 0x50",
684 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
685 .exit_latency = 600,
686 .target_residency = 1800,
687 .enter = &intel_idle,
688 .enter_s2idle = intel_idle_s2idle, },
689 {
690 .name = "C10",
691 .desc = "MWAIT 0x60",
692 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
693 .exit_latency = 2600,
694 .target_residency = 7700,
695 .enter = &intel_idle,
696 .enter_s2idle = intel_idle_s2idle, },
697 {
698 .enter = NULL }
699};
700
701static struct cpuidle_state skl_cstates[] __initdata = {
702 {
703 .name = "C1",
704 .desc = "MWAIT 0x00",
705 .flags = MWAIT2flg(0x00),
706 .exit_latency = 2,
707 .target_residency = 2,
708 .enter = &intel_idle,
709 .enter_s2idle = intel_idle_s2idle, },
710 {
711 .name = "C1E",
712 .desc = "MWAIT 0x01",
713 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
714 .exit_latency = 10,
715 .target_residency = 20,
716 .enter = &intel_idle,
717 .enter_s2idle = intel_idle_s2idle, },
718 {
719 .name = "C3",
720 .desc = "MWAIT 0x10",
721 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
722 .exit_latency = 70,
723 .target_residency = 100,
724 .enter = &intel_idle,
725 .enter_s2idle = intel_idle_s2idle, },
726 {
727 .name = "C6",
728 .desc = "MWAIT 0x20",
729 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
730 .exit_latency = 85,
731 .target_residency = 200,
732 .enter = &intel_idle,
733 .enter_s2idle = intel_idle_s2idle, },
734 {
735 .name = "C7s",
736 .desc = "MWAIT 0x33",
737 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
738 .exit_latency = 124,
739 .target_residency = 800,
740 .enter = &intel_idle,
741 .enter_s2idle = intel_idle_s2idle, },
742 {
743 .name = "C8",
744 .desc = "MWAIT 0x40",
745 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
746 .exit_latency = 200,
747 .target_residency = 800,
748 .enter = &intel_idle,
749 .enter_s2idle = intel_idle_s2idle, },
750 {
751 .name = "C9",
752 .desc = "MWAIT 0x50",
753 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
754 .exit_latency = 480,
755 .target_residency = 5000,
756 .enter = &intel_idle,
757 .enter_s2idle = intel_idle_s2idle, },
758 {
759 .name = "C10",
760 .desc = "MWAIT 0x60",
761 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
762 .exit_latency = 890,
763 .target_residency = 5000,
764 .enter = &intel_idle,
765 .enter_s2idle = intel_idle_s2idle, },
766 {
767 .enter = NULL }
768};
769
770static struct cpuidle_state skx_cstates[] __initdata = {
771 {
772 .name = "C1",
773 .desc = "MWAIT 0x00",
774 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
775 .exit_latency = 2,
776 .target_residency = 2,
777 .enter = &intel_idle,
778 .enter_s2idle = intel_idle_s2idle, },
779 {
780 .name = "C1E",
781 .desc = "MWAIT 0x01",
782 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
783 .exit_latency = 10,
784 .target_residency = 20,
785 .enter = &intel_idle,
786 .enter_s2idle = intel_idle_s2idle, },
787 {
788 .name = "C6",
789 .desc = "MWAIT 0x20",
790 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
791 .exit_latency = 133,
792 .target_residency = 600,
793 .enter = &intel_idle,
794 .enter_s2idle = intel_idle_s2idle, },
795 {
796 .enter = NULL }
797};
798
799static struct cpuidle_state icx_cstates[] __initdata = {
800 {
801 .name = "C1",
802 .desc = "MWAIT 0x00",
803 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
804 .exit_latency = 1,
805 .target_residency = 1,
806 .enter = &intel_idle,
807 .enter_s2idle = intel_idle_s2idle, },
808 {
809 .name = "C1E",
810 .desc = "MWAIT 0x01",
811 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
812 .exit_latency = 4,
813 .target_residency = 4,
814 .enter = &intel_idle,
815 .enter_s2idle = intel_idle_s2idle, },
816 {
817 .name = "C6",
818 .desc = "MWAIT 0x20",
819 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
820 .exit_latency = 170,
821 .target_residency = 600,
822 .enter = &intel_idle,
823 .enter_s2idle = intel_idle_s2idle, },
824 {
825 .enter = NULL }
826};
827
828/*
829 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
830 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
831 * But in this case there is effectively no C1, because C1 requests are
832 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
833 * and C1E requests end up with C1, so there is effectively no C1E.
834 *
835 * By default we enable C1E and disable C1 by marking it with
836 * 'CPUIDLE_FLAG_UNUSABLE'.
837 */
838static struct cpuidle_state adl_cstates[] __initdata = {
839 {
840 .name = "C1",
841 .desc = "MWAIT 0x00",
842 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
843 .exit_latency = 1,
844 .target_residency = 1,
845 .enter = &intel_idle,
846 .enter_s2idle = intel_idle_s2idle, },
847 {
848 .name = "C1E",
849 .desc = "MWAIT 0x01",
850 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
851 .exit_latency = 2,
852 .target_residency = 4,
853 .enter = &intel_idle,
854 .enter_s2idle = intel_idle_s2idle, },
855 {
856 .name = "C6",
857 .desc = "MWAIT 0x20",
858 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
859 .exit_latency = 220,
860 .target_residency = 600,
861 .enter = &intel_idle,
862 .enter_s2idle = intel_idle_s2idle, },
863 {
864 .name = "C8",
865 .desc = "MWAIT 0x40",
866 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
867 .exit_latency = 280,
868 .target_residency = 800,
869 .enter = &intel_idle,
870 .enter_s2idle = intel_idle_s2idle, },
871 {
872 .name = "C10",
873 .desc = "MWAIT 0x60",
874 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
875 .exit_latency = 680,
876 .target_residency = 2000,
877 .enter = &intel_idle,
878 .enter_s2idle = intel_idle_s2idle, },
879 {
880 .enter = NULL }
881};
882
883static struct cpuidle_state adl_l_cstates[] __initdata = {
884 {
885 .name = "C1",
886 .desc = "MWAIT 0x00",
887 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
888 .exit_latency = 1,
889 .target_residency = 1,
890 .enter = &intel_idle,
891 .enter_s2idle = intel_idle_s2idle, },
892 {
893 .name = "C1E",
894 .desc = "MWAIT 0x01",
895 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
896 .exit_latency = 2,
897 .target_residency = 4,
898 .enter = &intel_idle,
899 .enter_s2idle = intel_idle_s2idle, },
900 {
901 .name = "C6",
902 .desc = "MWAIT 0x20",
903 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
904 .exit_latency = 170,
905 .target_residency = 500,
906 .enter = &intel_idle,
907 .enter_s2idle = intel_idle_s2idle, },
908 {
909 .name = "C8",
910 .desc = "MWAIT 0x40",
911 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
912 .exit_latency = 200,
913 .target_residency = 600,
914 .enter = &intel_idle,
915 .enter_s2idle = intel_idle_s2idle, },
916 {
917 .name = "C10",
918 .desc = "MWAIT 0x60",
919 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
920 .exit_latency = 230,
921 .target_residency = 700,
922 .enter = &intel_idle,
923 .enter_s2idle = intel_idle_s2idle, },
924 {
925 .enter = NULL }
926};
927
928static struct cpuidle_state mtl_l_cstates[] __initdata = {
929 {
930 .name = "C1E",
931 .desc = "MWAIT 0x01",
932 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
933 .exit_latency = 1,
934 .target_residency = 1,
935 .enter = &intel_idle,
936 .enter_s2idle = intel_idle_s2idle, },
937 {
938 .name = "C6",
939 .desc = "MWAIT 0x20",
940 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
941 .exit_latency = 140,
942 .target_residency = 420,
943 .enter = &intel_idle,
944 .enter_s2idle = intel_idle_s2idle, },
945 {
946 .name = "C10",
947 .desc = "MWAIT 0x60",
948 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
949 .exit_latency = 310,
950 .target_residency = 930,
951 .enter = &intel_idle,
952 .enter_s2idle = intel_idle_s2idle, },
953 {
954 .enter = NULL }
955};
956
957static struct cpuidle_state gmt_cstates[] __initdata = {
958 {
959 .name = "C1",
960 .desc = "MWAIT 0x00",
961 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
962 .exit_latency = 1,
963 .target_residency = 1,
964 .enter = &intel_idle,
965 .enter_s2idle = intel_idle_s2idle, },
966 {
967 .name = "C1E",
968 .desc = "MWAIT 0x01",
969 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
970 .exit_latency = 2,
971 .target_residency = 4,
972 .enter = &intel_idle,
973 .enter_s2idle = intel_idle_s2idle, },
974 {
975 .name = "C6",
976 .desc = "MWAIT 0x20",
977 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
978 .exit_latency = 195,
979 .target_residency = 585,
980 .enter = &intel_idle,
981 .enter_s2idle = intel_idle_s2idle, },
982 {
983 .name = "C8",
984 .desc = "MWAIT 0x40",
985 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
986 .exit_latency = 260,
987 .target_residency = 1040,
988 .enter = &intel_idle,
989 .enter_s2idle = intel_idle_s2idle, },
990 {
991 .name = "C10",
992 .desc = "MWAIT 0x60",
993 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
994 .exit_latency = 660,
995 .target_residency = 1980,
996 .enter = &intel_idle,
997 .enter_s2idle = intel_idle_s2idle, },
998 {
999 .enter = NULL }
1000};
1001
1002static struct cpuidle_state spr_cstates[] __initdata = {
1003 {
1004 .name = "C1",
1005 .desc = "MWAIT 0x00",
1006 .flags = MWAIT2flg(0x00),
1007 .exit_latency = 1,
1008 .target_residency = 1,
1009 .enter = &intel_idle,
1010 .enter_s2idle = intel_idle_s2idle, },
1011 {
1012 .name = "C1E",
1013 .desc = "MWAIT 0x01",
1014 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1015 .exit_latency = 2,
1016 .target_residency = 4,
1017 .enter = &intel_idle,
1018 .enter_s2idle = intel_idle_s2idle, },
1019 {
1020 .name = "C6",
1021 .desc = "MWAIT 0x20",
1022 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
1023 CPUIDLE_FLAG_INIT_XSTATE,
1024 .exit_latency = 290,
1025 .target_residency = 800,
1026 .enter = &intel_idle,
1027 .enter_s2idle = intel_idle_s2idle, },
1028 {
1029 .enter = NULL }
1030};
1031
1032static struct cpuidle_state gnr_cstates[] __initdata = {
1033 {
1034 .name = "C1",
1035 .desc = "MWAIT 0x00",
1036 .flags = MWAIT2flg(0x00),
1037 .exit_latency = 1,
1038 .target_residency = 1,
1039 .enter = &intel_idle,
1040 .enter_s2idle = intel_idle_s2idle, },
1041 {
1042 .name = "C1E",
1043 .desc = "MWAIT 0x01",
1044 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1045 .exit_latency = 4,
1046 .target_residency = 4,
1047 .enter = &intel_idle,
1048 .enter_s2idle = intel_idle_s2idle, },
1049 {
1050 .name = "C6",
1051 .desc = "MWAIT 0x20",
1052 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
1053 CPUIDLE_FLAG_INIT_XSTATE |
1054 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1055 .exit_latency = 170,
1056 .target_residency = 650,
1057 .enter = &intel_idle,
1058 .enter_s2idle = intel_idle_s2idle, },
1059 {
1060 .name = "C6P",
1061 .desc = "MWAIT 0x21",
1062 .flags = MWAIT2flg(0x21) | CPUIDLE_FLAG_TLB_FLUSHED |
1063 CPUIDLE_FLAG_INIT_XSTATE |
1064 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1065 .exit_latency = 210,
1066 .target_residency = 1000,
1067 .enter = &intel_idle,
1068 .enter_s2idle = intel_idle_s2idle, },
1069 {
1070 .enter = NULL }
1071};
1072
1073static struct cpuidle_state gnrd_cstates[] __initdata = {
1074 {
1075 .name = "C1",
1076 .desc = "MWAIT 0x00",
1077 .flags = MWAIT2flg(0x00),
1078 .exit_latency = 1,
1079 .target_residency = 1,
1080 .enter = &intel_idle,
1081 .enter_s2idle = intel_idle_s2idle, },
1082 {
1083 .name = "C1E",
1084 .desc = "MWAIT 0x01",
1085 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1086 .exit_latency = 4,
1087 .target_residency = 4,
1088 .enter = &intel_idle,
1089 .enter_s2idle = intel_idle_s2idle, },
1090 {
1091 .name = "C6",
1092 .desc = "MWAIT 0x20",
1093 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
1094 CPUIDLE_FLAG_INIT_XSTATE |
1095 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1096 .exit_latency = 220,
1097 .target_residency = 650,
1098 .enter = &intel_idle,
1099 .enter_s2idle = intel_idle_s2idle, },
1100 {
1101 .name = "C6P",
1102 .desc = "MWAIT 0x21",
1103 .flags = MWAIT2flg(0x21) | CPUIDLE_FLAG_TLB_FLUSHED |
1104 CPUIDLE_FLAG_INIT_XSTATE |
1105 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1106 .exit_latency = 240,
1107 .target_residency = 750,
1108 .enter = &intel_idle,
1109 .enter_s2idle = intel_idle_s2idle, },
1110 {
1111 .enter = NULL }
1112};
1113
1114static struct cpuidle_state atom_cstates[] __initdata = {
1115 {
1116 .name = "C1E",
1117 .desc = "MWAIT 0x00",
1118 .flags = MWAIT2flg(0x00),
1119 .exit_latency = 10,
1120 .target_residency = 20,
1121 .enter = &intel_idle,
1122 .enter_s2idle = intel_idle_s2idle, },
1123 {
1124 .name = "C2",
1125 .desc = "MWAIT 0x10",
1126 .flags = MWAIT2flg(0x10),
1127 .exit_latency = 20,
1128 .target_residency = 80,
1129 .enter = &intel_idle,
1130 .enter_s2idle = intel_idle_s2idle, },
1131 {
1132 .name = "C4",
1133 .desc = "MWAIT 0x30",
1134 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1135 .exit_latency = 100,
1136 .target_residency = 400,
1137 .enter = &intel_idle,
1138 .enter_s2idle = intel_idle_s2idle, },
1139 {
1140 .name = "C6",
1141 .desc = "MWAIT 0x52",
1142 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1143 .exit_latency = 140,
1144 .target_residency = 560,
1145 .enter = &intel_idle,
1146 .enter_s2idle = intel_idle_s2idle, },
1147 {
1148 .enter = NULL }
1149};
1150static struct cpuidle_state tangier_cstates[] __initdata = {
1151 {
1152 .name = "C1",
1153 .desc = "MWAIT 0x00",
1154 .flags = MWAIT2flg(0x00),
1155 .exit_latency = 1,
1156 .target_residency = 4,
1157 .enter = &intel_idle,
1158 .enter_s2idle = intel_idle_s2idle, },
1159 {
1160 .name = "C4",
1161 .desc = "MWAIT 0x30",
1162 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1163 .exit_latency = 100,
1164 .target_residency = 400,
1165 .enter = &intel_idle,
1166 .enter_s2idle = intel_idle_s2idle, },
1167 {
1168 .name = "C6",
1169 .desc = "MWAIT 0x52",
1170 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1171 .exit_latency = 140,
1172 .target_residency = 560,
1173 .enter = &intel_idle,
1174 .enter_s2idle = intel_idle_s2idle, },
1175 {
1176 .name = "C7",
1177 .desc = "MWAIT 0x60",
1178 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1179 .exit_latency = 1200,
1180 .target_residency = 4000,
1181 .enter = &intel_idle,
1182 .enter_s2idle = intel_idle_s2idle, },
1183 {
1184 .name = "C9",
1185 .desc = "MWAIT 0x64",
1186 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1187 .exit_latency = 10000,
1188 .target_residency = 20000,
1189 .enter = &intel_idle,
1190 .enter_s2idle = intel_idle_s2idle, },
1191 {
1192 .enter = NULL }
1193};
1194static struct cpuidle_state avn_cstates[] __initdata = {
1195 {
1196 .name = "C1",
1197 .desc = "MWAIT 0x00",
1198 .flags = MWAIT2flg(0x00),
1199 .exit_latency = 2,
1200 .target_residency = 2,
1201 .enter = &intel_idle,
1202 .enter_s2idle = intel_idle_s2idle, },
1203 {
1204 .name = "C6",
1205 .desc = "MWAIT 0x51",
1206 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1207 .exit_latency = 15,
1208 .target_residency = 45,
1209 .enter = &intel_idle,
1210 .enter_s2idle = intel_idle_s2idle, },
1211 {
1212 .enter = NULL }
1213};
1214static struct cpuidle_state knl_cstates[] __initdata = {
1215 {
1216 .name = "C1",
1217 .desc = "MWAIT 0x00",
1218 .flags = MWAIT2flg(0x00),
1219 .exit_latency = 1,
1220 .target_residency = 2,
1221 .enter = &intel_idle,
1222 .enter_s2idle = intel_idle_s2idle },
1223 {
1224 .name = "C6",
1225 .desc = "MWAIT 0x10",
1226 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1227 .exit_latency = 120,
1228 .target_residency = 500,
1229 .enter = &intel_idle,
1230 .enter_s2idle = intel_idle_s2idle },
1231 {
1232 .enter = NULL }
1233};
1234
1235static struct cpuidle_state bxt_cstates[] __initdata = {
1236 {
1237 .name = "C1",
1238 .desc = "MWAIT 0x00",
1239 .flags = MWAIT2flg(0x00),
1240 .exit_latency = 2,
1241 .target_residency = 2,
1242 .enter = &intel_idle,
1243 .enter_s2idle = intel_idle_s2idle, },
1244 {
1245 .name = "C1E",
1246 .desc = "MWAIT 0x01",
1247 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1248 .exit_latency = 10,
1249 .target_residency = 20,
1250 .enter = &intel_idle,
1251 .enter_s2idle = intel_idle_s2idle, },
1252 {
1253 .name = "C6",
1254 .desc = "MWAIT 0x20",
1255 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1256 .exit_latency = 133,
1257 .target_residency = 133,
1258 .enter = &intel_idle,
1259 .enter_s2idle = intel_idle_s2idle, },
1260 {
1261 .name = "C7s",
1262 .desc = "MWAIT 0x31",
1263 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1264 .exit_latency = 155,
1265 .target_residency = 155,
1266 .enter = &intel_idle,
1267 .enter_s2idle = intel_idle_s2idle, },
1268 {
1269 .name = "C8",
1270 .desc = "MWAIT 0x40",
1271 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1272 .exit_latency = 1000,
1273 .target_residency = 1000,
1274 .enter = &intel_idle,
1275 .enter_s2idle = intel_idle_s2idle, },
1276 {
1277 .name = "C9",
1278 .desc = "MWAIT 0x50",
1279 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1280 .exit_latency = 2000,
1281 .target_residency = 2000,
1282 .enter = &intel_idle,
1283 .enter_s2idle = intel_idle_s2idle, },
1284 {
1285 .name = "C10",
1286 .desc = "MWAIT 0x60",
1287 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1288 .exit_latency = 10000,
1289 .target_residency = 10000,
1290 .enter = &intel_idle,
1291 .enter_s2idle = intel_idle_s2idle, },
1292 {
1293 .enter = NULL }
1294};
1295
1296static struct cpuidle_state dnv_cstates[] __initdata = {
1297 {
1298 .name = "C1",
1299 .desc = "MWAIT 0x00",
1300 .flags = MWAIT2flg(0x00),
1301 .exit_latency = 2,
1302 .target_residency = 2,
1303 .enter = &intel_idle,
1304 .enter_s2idle = intel_idle_s2idle, },
1305 {
1306 .name = "C1E",
1307 .desc = "MWAIT 0x01",
1308 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1309 .exit_latency = 10,
1310 .target_residency = 20,
1311 .enter = &intel_idle,
1312 .enter_s2idle = intel_idle_s2idle, },
1313 {
1314 .name = "C6",
1315 .desc = "MWAIT 0x20",
1316 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1317 .exit_latency = 50,
1318 .target_residency = 500,
1319 .enter = &intel_idle,
1320 .enter_s2idle = intel_idle_s2idle, },
1321 {
1322 .enter = NULL }
1323};
1324
1325/*
1326 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1327 * C6, and this is indicated in the CPUID mwait leaf.
1328 */
1329static struct cpuidle_state snr_cstates[] __initdata = {
1330 {
1331 .name = "C1",
1332 .desc = "MWAIT 0x00",
1333 .flags = MWAIT2flg(0x00),
1334 .exit_latency = 2,
1335 .target_residency = 2,
1336 .enter = &intel_idle,
1337 .enter_s2idle = intel_idle_s2idle, },
1338 {
1339 .name = "C1E",
1340 .desc = "MWAIT 0x01",
1341 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1342 .exit_latency = 15,
1343 .target_residency = 25,
1344 .enter = &intel_idle,
1345 .enter_s2idle = intel_idle_s2idle, },
1346 {
1347 .name = "C6",
1348 .desc = "MWAIT 0x20",
1349 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1350 .exit_latency = 130,
1351 .target_residency = 500,
1352 .enter = &intel_idle,
1353 .enter_s2idle = intel_idle_s2idle, },
1354 {
1355 .enter = NULL }
1356};
1357
1358static struct cpuidle_state grr_cstates[] __initdata = {
1359 {
1360 .name = "C1",
1361 .desc = "MWAIT 0x00",
1362 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1363 .exit_latency = 1,
1364 .target_residency = 1,
1365 .enter = &intel_idle,
1366 .enter_s2idle = intel_idle_s2idle, },
1367 {
1368 .name = "C1E",
1369 .desc = "MWAIT 0x01",
1370 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1371 .exit_latency = 2,
1372 .target_residency = 10,
1373 .enter = &intel_idle,
1374 .enter_s2idle = intel_idle_s2idle, },
1375 {
1376 .name = "C6S",
1377 .desc = "MWAIT 0x22",
1378 .flags = MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED,
1379 .exit_latency = 140,
1380 .target_residency = 500,
1381 .enter = &intel_idle,
1382 .enter_s2idle = intel_idle_s2idle, },
1383 {
1384 .enter = NULL }
1385};
1386
1387static struct cpuidle_state srf_cstates[] __initdata = {
1388 {
1389 .name = "C1",
1390 .desc = "MWAIT 0x00",
1391 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1392 .exit_latency = 1,
1393 .target_residency = 1,
1394 .enter = &intel_idle,
1395 .enter_s2idle = intel_idle_s2idle, },
1396 {
1397 .name = "C1E",
1398 .desc = "MWAIT 0x01",
1399 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1400 .exit_latency = 2,
1401 .target_residency = 10,
1402 .enter = &intel_idle,
1403 .enter_s2idle = intel_idle_s2idle, },
1404 {
1405 .name = "C6S",
1406 .desc = "MWAIT 0x22",
1407 .flags = MWAIT2flg(0x22) | CPUIDLE_FLAG_TLB_FLUSHED |
1408 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1409 .exit_latency = 270,
1410 .target_residency = 700,
1411 .enter = &intel_idle,
1412 .enter_s2idle = intel_idle_s2idle, },
1413 {
1414 .name = "C6SP",
1415 .desc = "MWAIT 0x23",
1416 .flags = MWAIT2flg(0x23) | CPUIDLE_FLAG_TLB_FLUSHED |
1417 CPUIDLE_FLAG_PARTIAL_HINT_MATCH,
1418 .exit_latency = 310,
1419 .target_residency = 900,
1420 .enter = &intel_idle,
1421 .enter_s2idle = intel_idle_s2idle, },
1422 {
1423 .enter = NULL }
1424};
1425
1426static const struct idle_cpu idle_cpu_nehalem __initconst = {
1427 .state_table = nehalem_cstates,
1428 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1429 .disable_promotion_to_c1e = true,
1430};
1431
1432static const struct idle_cpu idle_cpu_nhx __initconst = {
1433 .state_table = nehalem_cstates,
1434 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1435 .disable_promotion_to_c1e = true,
1436 .use_acpi = true,
1437};
1438
1439static const struct idle_cpu idle_cpu_atom __initconst = {
1440 .state_table = atom_cstates,
1441};
1442
1443static const struct idle_cpu idle_cpu_tangier __initconst = {
1444 .state_table = tangier_cstates,
1445};
1446
1447static const struct idle_cpu idle_cpu_lincroft __initconst = {
1448 .state_table = atom_cstates,
1449 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1450};
1451
1452static const struct idle_cpu idle_cpu_snb __initconst = {
1453 .state_table = snb_cstates,
1454 .disable_promotion_to_c1e = true,
1455};
1456
1457static const struct idle_cpu idle_cpu_snx __initconst = {
1458 .state_table = snb_cstates,
1459 .disable_promotion_to_c1e = true,
1460 .use_acpi = true,
1461};
1462
1463static const struct idle_cpu idle_cpu_byt __initconst = {
1464 .state_table = byt_cstates,
1465 .disable_promotion_to_c1e = true,
1466 .byt_auto_demotion_disable_flag = true,
1467};
1468
1469static const struct idle_cpu idle_cpu_cht __initconst = {
1470 .state_table = cht_cstates,
1471 .disable_promotion_to_c1e = true,
1472 .byt_auto_demotion_disable_flag = true,
1473};
1474
1475static const struct idle_cpu idle_cpu_ivb __initconst = {
1476 .state_table = ivb_cstates,
1477 .disable_promotion_to_c1e = true,
1478};
1479
1480static const struct idle_cpu idle_cpu_ivt __initconst = {
1481 .state_table = ivt_cstates,
1482 .disable_promotion_to_c1e = true,
1483 .use_acpi = true,
1484};
1485
1486static const struct idle_cpu idle_cpu_hsw __initconst = {
1487 .state_table = hsw_cstates,
1488 .disable_promotion_to_c1e = true,
1489};
1490
1491static const struct idle_cpu idle_cpu_hsx __initconst = {
1492 .state_table = hsw_cstates,
1493 .disable_promotion_to_c1e = true,
1494 .use_acpi = true,
1495};
1496
1497static const struct idle_cpu idle_cpu_bdw __initconst = {
1498 .state_table = bdw_cstates,
1499 .disable_promotion_to_c1e = true,
1500};
1501
1502static const struct idle_cpu idle_cpu_bdx __initconst = {
1503 .state_table = bdw_cstates,
1504 .disable_promotion_to_c1e = true,
1505 .use_acpi = true,
1506};
1507
1508static const struct idle_cpu idle_cpu_skl __initconst = {
1509 .state_table = skl_cstates,
1510 .disable_promotion_to_c1e = true,
1511};
1512
1513static const struct idle_cpu idle_cpu_skx __initconst = {
1514 .state_table = skx_cstates,
1515 .disable_promotion_to_c1e = true,
1516 .use_acpi = true,
1517};
1518
1519static const struct idle_cpu idle_cpu_icx __initconst = {
1520 .state_table = icx_cstates,
1521 .disable_promotion_to_c1e = true,
1522 .use_acpi = true,
1523};
1524
1525static const struct idle_cpu idle_cpu_adl __initconst = {
1526 .state_table = adl_cstates,
1527};
1528
1529static const struct idle_cpu idle_cpu_adl_l __initconst = {
1530 .state_table = adl_l_cstates,
1531};
1532
1533static const struct idle_cpu idle_cpu_mtl_l __initconst = {
1534 .state_table = mtl_l_cstates,
1535};
1536
1537static const struct idle_cpu idle_cpu_gmt __initconst = {
1538 .state_table = gmt_cstates,
1539};
1540
1541static const struct idle_cpu idle_cpu_spr __initconst = {
1542 .state_table = spr_cstates,
1543 .disable_promotion_to_c1e = true,
1544 .use_acpi = true,
1545};
1546
1547static const struct idle_cpu idle_cpu_gnr __initconst = {
1548 .state_table = gnr_cstates,
1549 .disable_promotion_to_c1e = true,
1550 .use_acpi = true,
1551};
1552
1553static const struct idle_cpu idle_cpu_gnrd __initconst = {
1554 .state_table = gnrd_cstates,
1555 .disable_promotion_to_c1e = true,
1556 .use_acpi = true,
1557};
1558
1559static const struct idle_cpu idle_cpu_avn __initconst = {
1560 .state_table = avn_cstates,
1561 .disable_promotion_to_c1e = true,
1562 .use_acpi = true,
1563};
1564
1565static const struct idle_cpu idle_cpu_knl __initconst = {
1566 .state_table = knl_cstates,
1567 .use_acpi = true,
1568};
1569
1570static const struct idle_cpu idle_cpu_bxt __initconst = {
1571 .state_table = bxt_cstates,
1572 .disable_promotion_to_c1e = true,
1573};
1574
1575static const struct idle_cpu idle_cpu_dnv __initconst = {
1576 .state_table = dnv_cstates,
1577 .disable_promotion_to_c1e = true,
1578 .use_acpi = true,
1579};
1580
1581static const struct idle_cpu idle_cpu_tmt __initconst = {
1582 .disable_promotion_to_c1e = true,
1583};
1584
1585static const struct idle_cpu idle_cpu_snr __initconst = {
1586 .state_table = snr_cstates,
1587 .disable_promotion_to_c1e = true,
1588 .use_acpi = true,
1589};
1590
1591static const struct idle_cpu idle_cpu_grr __initconst = {
1592 .state_table = grr_cstates,
1593 .disable_promotion_to_c1e = true,
1594 .use_acpi = true,
1595};
1596
1597static const struct idle_cpu idle_cpu_srf __initconst = {
1598 .state_table = srf_cstates,
1599 .disable_promotion_to_c1e = true,
1600 .use_acpi = true,
1601};
1602
1603static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1604 X86_MATCH_VFM(INTEL_NEHALEM_EP, &idle_cpu_nhx),
1605 X86_MATCH_VFM(INTEL_NEHALEM, &idle_cpu_nehalem),
1606 X86_MATCH_VFM(INTEL_NEHALEM_G, &idle_cpu_nehalem),
1607 X86_MATCH_VFM(INTEL_WESTMERE, &idle_cpu_nehalem),
1608 X86_MATCH_VFM(INTEL_WESTMERE_EP, &idle_cpu_nhx),
1609 X86_MATCH_VFM(INTEL_NEHALEM_EX, &idle_cpu_nhx),
1610 X86_MATCH_VFM(INTEL_ATOM_BONNELL, &idle_cpu_atom),
1611 X86_MATCH_VFM(INTEL_ATOM_BONNELL_MID, &idle_cpu_lincroft),
1612 X86_MATCH_VFM(INTEL_WESTMERE_EX, &idle_cpu_nhx),
1613 X86_MATCH_VFM(INTEL_SANDYBRIDGE, &idle_cpu_snb),
1614 X86_MATCH_VFM(INTEL_SANDYBRIDGE_X, &idle_cpu_snx),
1615 X86_MATCH_VFM(INTEL_ATOM_SALTWELL, &idle_cpu_atom),
1616 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT, &idle_cpu_byt),
1617 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1618 X86_MATCH_VFM(INTEL_ATOM_AIRMONT, &idle_cpu_cht),
1619 X86_MATCH_VFM(INTEL_IVYBRIDGE, &idle_cpu_ivb),
1620 X86_MATCH_VFM(INTEL_IVYBRIDGE_X, &idle_cpu_ivt),
1621 X86_MATCH_VFM(INTEL_HASWELL, &idle_cpu_hsw),
1622 X86_MATCH_VFM(INTEL_HASWELL_X, &idle_cpu_hsx),
1623 X86_MATCH_VFM(INTEL_HASWELL_L, &idle_cpu_hsw),
1624 X86_MATCH_VFM(INTEL_HASWELL_G, &idle_cpu_hsw),
1625 X86_MATCH_VFM(INTEL_ATOM_SILVERMONT_D, &idle_cpu_avn),
1626 X86_MATCH_VFM(INTEL_BROADWELL, &idle_cpu_bdw),
1627 X86_MATCH_VFM(INTEL_BROADWELL_G, &idle_cpu_bdw),
1628 X86_MATCH_VFM(INTEL_BROADWELL_X, &idle_cpu_bdx),
1629 X86_MATCH_VFM(INTEL_BROADWELL_D, &idle_cpu_bdx),
1630 X86_MATCH_VFM(INTEL_SKYLAKE_L, &idle_cpu_skl),
1631 X86_MATCH_VFM(INTEL_SKYLAKE, &idle_cpu_skl),
1632 X86_MATCH_VFM(INTEL_KABYLAKE_L, &idle_cpu_skl),
1633 X86_MATCH_VFM(INTEL_KABYLAKE, &idle_cpu_skl),
1634 X86_MATCH_VFM(INTEL_SKYLAKE_X, &idle_cpu_skx),
1635 X86_MATCH_VFM(INTEL_ICELAKE_X, &idle_cpu_icx),
1636 X86_MATCH_VFM(INTEL_ICELAKE_D, &idle_cpu_icx),
1637 X86_MATCH_VFM(INTEL_ALDERLAKE, &idle_cpu_adl),
1638 X86_MATCH_VFM(INTEL_ALDERLAKE_L, &idle_cpu_adl_l),
1639 X86_MATCH_VFM(INTEL_METEORLAKE_L, &idle_cpu_mtl_l),
1640 X86_MATCH_VFM(INTEL_ATOM_GRACEMONT, &idle_cpu_gmt),
1641 X86_MATCH_VFM(INTEL_SAPPHIRERAPIDS_X, &idle_cpu_spr),
1642 X86_MATCH_VFM(INTEL_EMERALDRAPIDS_X, &idle_cpu_spr),
1643 X86_MATCH_VFM(INTEL_GRANITERAPIDS_X, &idle_cpu_gnr),
1644 X86_MATCH_VFM(INTEL_GRANITERAPIDS_D, &idle_cpu_gnrd),
1645 X86_MATCH_VFM(INTEL_XEON_PHI_KNL, &idle_cpu_knl),
1646 X86_MATCH_VFM(INTEL_XEON_PHI_KNM, &idle_cpu_knl),
1647 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT, &idle_cpu_bxt),
1648 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1649 X86_MATCH_VFM(INTEL_ATOM_GOLDMONT_D, &idle_cpu_dnv),
1650 X86_MATCH_VFM(INTEL_ATOM_TREMONT, &idle_cpu_tmt),
1651 X86_MATCH_VFM(INTEL_ATOM_TREMONT_L, &idle_cpu_tmt),
1652 X86_MATCH_VFM(INTEL_ATOM_TREMONT_D, &idle_cpu_snr),
1653 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT, &idle_cpu_grr),
1654 X86_MATCH_VFM(INTEL_ATOM_CRESTMONT_X, &idle_cpu_srf),
1655 {}
1656};
1657
1658static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1659 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1660 {}
1661};
1662
1663static bool __init intel_idle_max_cstate_reached(int cstate)
1664{
1665 if (cstate + 1 > max_cstate) {
1666 pr_info("max_cstate %d reached\n", max_cstate);
1667 return true;
1668 }
1669 return false;
1670}
1671
1672static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1673{
1674 unsigned long eax = flg2MWAIT(state->flags);
1675
1676 if (boot_cpu_has(X86_FEATURE_ARAT))
1677 return false;
1678
1679 /*
1680 * Switch over to one-shot tick broadcast if the target C-state
1681 * is deeper than C1.
1682 */
1683 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1684}
1685
1686#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1687#include <acpi/processor.h>
1688
1689static bool no_acpi __read_mostly;
1690module_param(no_acpi, bool, 0444);
1691MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1692
1693static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1694module_param_named(use_acpi, force_use_acpi, bool, 0444);
1695MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1696
1697static struct acpi_processor_power acpi_state_table __initdata;
1698
1699/**
1700 * intel_idle_cst_usable - Check if the _CST information can be used.
1701 *
1702 * Check if all of the C-states listed by _CST in the max_cstate range are
1703 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1704 */
1705static bool __init intel_idle_cst_usable(void)
1706{
1707 int cstate, limit;
1708
1709 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1710 acpi_state_table.count);
1711
1712 for (cstate = 1; cstate < limit; cstate++) {
1713 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1714
1715 if (cx->entry_method != ACPI_CSTATE_FFH)
1716 return false;
1717 }
1718
1719 return true;
1720}
1721
1722static bool __init intel_idle_acpi_cst_extract(void)
1723{
1724 unsigned int cpu;
1725
1726 if (no_acpi) {
1727 pr_debug("Not allowed to use ACPI _CST\n");
1728 return false;
1729 }
1730
1731 for_each_possible_cpu(cpu) {
1732 struct acpi_processor *pr = per_cpu(processors, cpu);
1733
1734 if (!pr)
1735 continue;
1736
1737 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1738 continue;
1739
1740 acpi_state_table.count++;
1741
1742 if (!intel_idle_cst_usable())
1743 continue;
1744
1745 if (!acpi_processor_claim_cst_control())
1746 break;
1747
1748 return true;
1749 }
1750
1751 acpi_state_table.count = 0;
1752 pr_debug("ACPI _CST not found or not usable\n");
1753 return false;
1754}
1755
1756static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1757{
1758 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1759
1760 /*
1761 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1762 * the interesting states are ACPI_CSTATE_FFH.
1763 */
1764 for (cstate = 1; cstate < limit; cstate++) {
1765 struct acpi_processor_cx *cx;
1766 struct cpuidle_state *state;
1767
1768 if (intel_idle_max_cstate_reached(cstate - 1))
1769 break;
1770
1771 cx = &acpi_state_table.states[cstate];
1772
1773 state = &drv->states[drv->state_count++];
1774
1775 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1776 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1777 state->exit_latency = cx->latency;
1778 /*
1779 * For C1-type C-states use the same number for both the exit
1780 * latency and target residency, because that is the case for
1781 * C1 in the majority of the static C-states tables above.
1782 * For the other types of C-states, however, set the target
1783 * residency to 3 times the exit latency which should lead to
1784 * a reasonable balance between energy-efficiency and
1785 * performance in the majority of interesting cases.
1786 */
1787 state->target_residency = cx->latency;
1788 if (cx->type > ACPI_STATE_C1)
1789 state->target_residency *= 3;
1790
1791 state->flags = MWAIT2flg(cx->address);
1792 if (cx->type > ACPI_STATE_C2)
1793 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1794
1795 if (disabled_states_mask & BIT(cstate))
1796 state->flags |= CPUIDLE_FLAG_OFF;
1797
1798 if (intel_idle_state_needs_timer_stop(state))
1799 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1800
1801 if (cx->type > ACPI_STATE_C1 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1802 mark_tsc_unstable("TSC halts in idle");
1803
1804 state->enter = intel_idle;
1805 state->enter_s2idle = intel_idle_s2idle;
1806 }
1807}
1808
1809static bool __init intel_idle_off_by_default(unsigned int flags, u32 mwait_hint)
1810{
1811 int cstate, limit;
1812
1813 /*
1814 * If there are no _CST C-states, do not disable any C-states by
1815 * default.
1816 */
1817 if (!acpi_state_table.count)
1818 return false;
1819
1820 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1821 /*
1822 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1823 * the interesting states are ACPI_CSTATE_FFH.
1824 */
1825 for (cstate = 1; cstate < limit; cstate++) {
1826 u32 acpi_hint = acpi_state_table.states[cstate].address;
1827 u32 table_hint = mwait_hint;
1828
1829 if (flags & CPUIDLE_FLAG_PARTIAL_HINT_MATCH) {
1830 acpi_hint &= ~MWAIT_SUBSTATE_MASK;
1831 table_hint &= ~MWAIT_SUBSTATE_MASK;
1832 }
1833
1834 if (acpi_hint == table_hint)
1835 return false;
1836 }
1837 return true;
1838}
1839#else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1840#define force_use_acpi (false)
1841
1842static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1843static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1844static inline bool intel_idle_off_by_default(unsigned int flags, u32 mwait_hint)
1845{
1846 return false;
1847}
1848#endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1849
1850/**
1851 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1852 *
1853 * Tune IVT multi-socket targets.
1854 * Assumption: num_sockets == (max_package_num + 1).
1855 */
1856static void __init ivt_idle_state_table_update(void)
1857{
1858 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1859 int cpu, package_num, num_sockets = 1;
1860
1861 for_each_online_cpu(cpu) {
1862 package_num = topology_physical_package_id(cpu);
1863 if (package_num + 1 > num_sockets) {
1864 num_sockets = package_num + 1;
1865
1866 if (num_sockets > 4) {
1867 cpuidle_state_table = ivt_cstates_8s;
1868 return;
1869 }
1870 }
1871 }
1872
1873 if (num_sockets > 2)
1874 cpuidle_state_table = ivt_cstates_4s;
1875
1876 /* else, 1 and 2 socket systems use default ivt_cstates */
1877}
1878
1879/**
1880 * irtl_2_usec - IRTL to microseconds conversion.
1881 * @irtl: IRTL MSR value.
1882 *
1883 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1884 */
1885static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1886{
1887 static const unsigned int irtl_ns_units[] __initconst = {
1888 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1889 };
1890 unsigned long long ns;
1891
1892 if (!irtl)
1893 return 0;
1894
1895 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1896
1897 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1898}
1899
1900/**
1901 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1902 *
1903 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1904 * definitive maximum latency and use the same value for target_residency.
1905 */
1906static void __init bxt_idle_state_table_update(void)
1907{
1908 unsigned long long msr;
1909 unsigned int usec;
1910
1911 rdmsrl(MSR_PKGC6_IRTL, msr);
1912 usec = irtl_2_usec(msr);
1913 if (usec) {
1914 bxt_cstates[2].exit_latency = usec;
1915 bxt_cstates[2].target_residency = usec;
1916 }
1917
1918 rdmsrl(MSR_PKGC7_IRTL, msr);
1919 usec = irtl_2_usec(msr);
1920 if (usec) {
1921 bxt_cstates[3].exit_latency = usec;
1922 bxt_cstates[3].target_residency = usec;
1923 }
1924
1925 rdmsrl(MSR_PKGC8_IRTL, msr);
1926 usec = irtl_2_usec(msr);
1927 if (usec) {
1928 bxt_cstates[4].exit_latency = usec;
1929 bxt_cstates[4].target_residency = usec;
1930 }
1931
1932 rdmsrl(MSR_PKGC9_IRTL, msr);
1933 usec = irtl_2_usec(msr);
1934 if (usec) {
1935 bxt_cstates[5].exit_latency = usec;
1936 bxt_cstates[5].target_residency = usec;
1937 }
1938
1939 rdmsrl(MSR_PKGC10_IRTL, msr);
1940 usec = irtl_2_usec(msr);
1941 if (usec) {
1942 bxt_cstates[6].exit_latency = usec;
1943 bxt_cstates[6].target_residency = usec;
1944 }
1945
1946}
1947
1948/**
1949 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1950 *
1951 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1952 */
1953static void __init sklh_idle_state_table_update(void)
1954{
1955 unsigned long long msr;
1956 unsigned int eax, ebx, ecx, edx;
1957
1958
1959 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1960 if (max_cstate <= 7)
1961 return;
1962
1963 /* if PC10 not present in CPUID.MWAIT.EDX */
1964 if ((mwait_substates & (0xF << 28)) == 0)
1965 return;
1966
1967 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1968
1969 /* PC10 is not enabled in PKG C-state limit */
1970 if ((msr & 0xF) != 8)
1971 return;
1972
1973 ecx = 0;
1974 cpuid(7, &eax, &ebx, &ecx, &edx);
1975
1976 /* if SGX is present */
1977 if (ebx & (1 << 2)) {
1978
1979 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1980
1981 /* if SGX is enabled */
1982 if (msr & (1 << 18))
1983 return;
1984 }
1985
1986 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1987 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1988}
1989
1990/**
1991 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1992 * idle states table.
1993 */
1994static void __init skx_idle_state_table_update(void)
1995{
1996 unsigned long long msr;
1997
1998 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1999
2000 /*
2001 * 000b: C0/C1 (no package C-state support)
2002 * 001b: C2
2003 * 010b: C6 (non-retention)
2004 * 011b: C6 (retention)
2005 * 111b: No Package C state limits.
2006 */
2007 if ((msr & 0x7) < 2) {
2008 /*
2009 * Uses the CC6 + PC0 latency and 3 times of
2010 * latency for target_residency if the PC6
2011 * is disabled in BIOS. This is consistent
2012 * with how intel_idle driver uses _CST
2013 * to set the target_residency.
2014 */
2015 skx_cstates[2].exit_latency = 92;
2016 skx_cstates[2].target_residency = 276;
2017 }
2018}
2019
2020/**
2021 * adl_idle_state_table_update - Adjust AlderLake idle states table.
2022 */
2023static void __init adl_idle_state_table_update(void)
2024{
2025 /* Check if user prefers C1 over C1E. */
2026 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
2027 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
2028 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
2029
2030 /* Disable C1E by clearing the "C1E promotion" bit. */
2031 c1e_promotion = C1E_PROMOTION_DISABLE;
2032 return;
2033 }
2034
2035 /* Make sure C1E is enabled by default */
2036 c1e_promotion = C1E_PROMOTION_ENABLE;
2037}
2038
2039/**
2040 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
2041 */
2042static void __init spr_idle_state_table_update(void)
2043{
2044 unsigned long long msr;
2045
2046 /*
2047 * By default, the C6 state assumes the worst-case scenario of package
2048 * C6. However, if PC6 is disabled, we update the numbers to match
2049 * core C6.
2050 */
2051 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
2052
2053 /* Limit value 2 and above allow for PC6. */
2054 if ((msr & 0x7) < 2) {
2055 spr_cstates[2].exit_latency = 190;
2056 spr_cstates[2].target_residency = 600;
2057 }
2058}
2059
2060static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
2061{
2062 unsigned int mwait_cstate = (MWAIT_HINT2CSTATE(mwait_hint) + 1) &
2063 MWAIT_CSTATE_MASK;
2064 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
2065 MWAIT_SUBSTATE_MASK;
2066
2067 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
2068 if (num_substates == 0)
2069 return false;
2070
2071 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
2072 mark_tsc_unstable("TSC halts in idle states deeper than C2");
2073
2074 return true;
2075}
2076
2077static void state_update_enter_method(struct cpuidle_state *state, int cstate)
2078{
2079 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE) {
2080 /*
2081 * Combining with XSTATE with IBRS or IRQ_ENABLE flags
2082 * is not currently supported but this driver.
2083 */
2084 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IBRS);
2085 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
2086 state->enter = intel_idle_xstate;
2087 return;
2088 }
2089
2090 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
2091 ((state->flags & CPUIDLE_FLAG_IBRS) || ibrs_off)) {
2092 /*
2093 * IBRS mitigation requires that C-states are entered
2094 * with interrupts disabled.
2095 */
2096 if (ibrs_off && (state->flags & CPUIDLE_FLAG_IRQ_ENABLE))
2097 state->flags &= ~CPUIDLE_FLAG_IRQ_ENABLE;
2098 WARN_ON_ONCE(state->flags & CPUIDLE_FLAG_IRQ_ENABLE);
2099 state->enter = intel_idle_ibrs;
2100 return;
2101 }
2102
2103 if (state->flags & CPUIDLE_FLAG_IRQ_ENABLE) {
2104 state->enter = intel_idle_irq;
2105 return;
2106 }
2107
2108 if (force_irq_on) {
2109 pr_info("forced intel_idle_irq for state %d\n", cstate);
2110 state->enter = intel_idle_irq;
2111 }
2112}
2113
2114static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
2115{
2116 int cstate;
2117
2118 switch (boot_cpu_data.x86_vfm) {
2119 case INTEL_IVYBRIDGE_X:
2120 ivt_idle_state_table_update();
2121 break;
2122 case INTEL_ATOM_GOLDMONT:
2123 case INTEL_ATOM_GOLDMONT_PLUS:
2124 bxt_idle_state_table_update();
2125 break;
2126 case INTEL_SKYLAKE:
2127 sklh_idle_state_table_update();
2128 break;
2129 case INTEL_SKYLAKE_X:
2130 skx_idle_state_table_update();
2131 break;
2132 case INTEL_SAPPHIRERAPIDS_X:
2133 case INTEL_EMERALDRAPIDS_X:
2134 spr_idle_state_table_update();
2135 break;
2136 case INTEL_ALDERLAKE:
2137 case INTEL_ALDERLAKE_L:
2138 case INTEL_ATOM_GRACEMONT:
2139 adl_idle_state_table_update();
2140 break;
2141 }
2142
2143 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
2144 struct cpuidle_state *state;
2145 unsigned int mwait_hint;
2146
2147 if (intel_idle_max_cstate_reached(cstate))
2148 break;
2149
2150 if (!cpuidle_state_table[cstate].enter &&
2151 !cpuidle_state_table[cstate].enter_s2idle)
2152 break;
2153
2154 /* If marked as unusable, skip this state. */
2155 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
2156 pr_debug("state %s is disabled\n",
2157 cpuidle_state_table[cstate].name);
2158 continue;
2159 }
2160
2161 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
2162 if (!intel_idle_verify_cstate(mwait_hint))
2163 continue;
2164
2165 /* Structure copy. */
2166 drv->states[drv->state_count] = cpuidle_state_table[cstate];
2167 state = &drv->states[drv->state_count];
2168
2169 state_update_enter_method(state, cstate);
2170
2171
2172 if ((disabled_states_mask & BIT(drv->state_count)) ||
2173 ((icpu->use_acpi || force_use_acpi) &&
2174 intel_idle_off_by_default(state->flags, mwait_hint) &&
2175 !(state->flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
2176 state->flags |= CPUIDLE_FLAG_OFF;
2177
2178 if (intel_idle_state_needs_timer_stop(state))
2179 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
2180
2181 drv->state_count++;
2182 }
2183
2184 if (icpu->byt_auto_demotion_disable_flag) {
2185 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
2186 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
2187 }
2188}
2189
2190/**
2191 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
2192 * @drv: cpuidle driver structure to initialize.
2193 */
2194static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
2195{
2196 cpuidle_poll_state_init(drv);
2197
2198 if (disabled_states_mask & BIT(0))
2199 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
2200
2201 drv->state_count = 1;
2202
2203 if (icpu && icpu->state_table)
2204 intel_idle_init_cstates_icpu(drv);
2205 else
2206 intel_idle_init_cstates_acpi(drv);
2207}
2208
2209static void auto_demotion_disable(void)
2210{
2211 unsigned long long msr_bits;
2212
2213 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
2214 msr_bits &= ~auto_demotion_disable_flags;
2215 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
2216}
2217
2218static void c1e_promotion_enable(void)
2219{
2220 unsigned long long msr_bits;
2221
2222 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
2223 msr_bits |= 0x2;
2224 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
2225}
2226
2227static void c1e_promotion_disable(void)
2228{
2229 unsigned long long msr_bits;
2230
2231 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
2232 msr_bits &= ~0x2;
2233 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
2234}
2235
2236/**
2237 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
2238 * @cpu: CPU to initialize.
2239 *
2240 * Register a cpuidle device object for @cpu and update its MSRs in accordance
2241 * with the processor model flags.
2242 */
2243static int intel_idle_cpu_init(unsigned int cpu)
2244{
2245 struct cpuidle_device *dev;
2246
2247 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2248 dev->cpu = cpu;
2249
2250 if (cpuidle_register_device(dev)) {
2251 pr_debug("cpuidle_register_device %d failed!\n", cpu);
2252 return -EIO;
2253 }
2254
2255 if (auto_demotion_disable_flags)
2256 auto_demotion_disable();
2257
2258 if (c1e_promotion == C1E_PROMOTION_ENABLE)
2259 c1e_promotion_enable();
2260 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
2261 c1e_promotion_disable();
2262
2263 return 0;
2264}
2265
2266static int intel_idle_cpu_online(unsigned int cpu)
2267{
2268 struct cpuidle_device *dev;
2269
2270 if (!boot_cpu_has(X86_FEATURE_ARAT))
2271 tick_broadcast_enable();
2272
2273 /*
2274 * Some systems can hotplug a cpu at runtime after
2275 * the kernel has booted, we have to initialize the
2276 * driver in this case
2277 */
2278 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2279 if (!dev->registered)
2280 return intel_idle_cpu_init(cpu);
2281
2282 return 0;
2283}
2284
2285/**
2286 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2287 */
2288static void __init intel_idle_cpuidle_devices_uninit(void)
2289{
2290 int i;
2291
2292 for_each_online_cpu(i)
2293 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2294}
2295
2296static int __init intel_idle_init(void)
2297{
2298 const struct x86_cpu_id *id;
2299 unsigned int eax, ebx, ecx;
2300 int retval;
2301
2302 /* Do not load intel_idle at all for now if idle= is passed */
2303 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2304 return -ENODEV;
2305
2306 if (max_cstate == 0) {
2307 pr_debug("disabled\n");
2308 return -EPERM;
2309 }
2310
2311 id = x86_match_cpu(intel_idle_ids);
2312 if (id) {
2313 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2314 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2315 return -ENODEV;
2316 }
2317 } else {
2318 id = x86_match_cpu(intel_mwait_ids);
2319 if (!id)
2320 return -ENODEV;
2321 }
2322
2323 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2324 return -ENODEV;
2325
2326 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
2327
2328 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2329 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2330 !mwait_substates)
2331 return -ENODEV;
2332
2333 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2334
2335 icpu = (const struct idle_cpu *)id->driver_data;
2336 if (icpu) {
2337 if (icpu->state_table)
2338 cpuidle_state_table = icpu->state_table;
2339 else if (!intel_idle_acpi_cst_extract())
2340 return -ENODEV;
2341
2342 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2343 if (icpu->disable_promotion_to_c1e)
2344 c1e_promotion = C1E_PROMOTION_DISABLE;
2345 if (icpu->use_acpi || force_use_acpi)
2346 intel_idle_acpi_cst_extract();
2347 } else if (!intel_idle_acpi_cst_extract()) {
2348 return -ENODEV;
2349 }
2350
2351 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2352 boot_cpu_data.x86_model);
2353
2354 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2355 if (!intel_idle_cpuidle_devices)
2356 return -ENOMEM;
2357
2358 intel_idle_cpuidle_driver_init(&intel_idle_driver);
2359
2360 retval = cpuidle_register_driver(&intel_idle_driver);
2361 if (retval) {
2362 struct cpuidle_driver *drv = cpuidle_get_driver();
2363 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2364 drv ? drv->name : "none");
2365 goto init_driver_fail;
2366 }
2367
2368 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
2369 intel_idle_cpu_online, NULL);
2370 if (retval < 0)
2371 goto hp_setup_fail;
2372
2373 pr_debug("Local APIC timer is reliable in %s\n",
2374 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2375
2376 return 0;
2377
2378hp_setup_fail:
2379 intel_idle_cpuidle_devices_uninit();
2380 cpuidle_unregister_driver(&intel_idle_driver);
2381init_driver_fail:
2382 free_percpu(intel_idle_cpuidle_devices);
2383 return retval;
2384
2385}
2386device_initcall(intel_idle_init);
2387
2388/*
2389 * We are not really modular, but we used to support that. Meaning we also
2390 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2391 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2392 * is the easiest way (currently) to continue doing that.
2393 */
2394module_param(max_cstate, int, 0444);
2395/*
2396 * The positions of the bits that are set in this number are the indices of the
2397 * idle states to be disabled by default (as reflected by the names of the
2398 * corresponding idle state directories in sysfs, "state0", "state1" ...
2399 * "state<i>" ..., where <i> is the index of the given state).
2400 */
2401module_param_named(states_off, disabled_states_mask, uint, 0444);
2402MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2403/*
2404 * Some platforms come with mutually exclusive C-states, so that if one is
2405 * enabled, the other C-states must not be used. Example: C1 and C1E on
2406 * Sapphire Rapids platform. This parameter allows for selecting the
2407 * preferred C-states among the groups of mutually exclusive C-states - the
2408 * selected C-states will be registered, the other C-states from the mutually
2409 * exclusive group won't be registered. If the platform has no mutually
2410 * exclusive C-states, this parameter has no effect.
2411 */
2412module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2413MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");
2414/*
2415 * Debugging option that forces the driver to enter all C-states with
2416 * interrupts enabled. Does not apply to C-states with
2417 * 'CPUIDLE_FLAG_INIT_XSTATE' and 'CPUIDLE_FLAG_IBRS' flags.
2418 */
2419module_param(force_irq_on, bool, 0444);
2420/*
2421 * Force the disabling of IBRS when X86_FEATURE_KERNEL_IBRS is on and
2422 * CPUIDLE_FLAG_IRQ_ENABLE isn't set.
2423 */
2424module_param(ibrs_off, bool, 0444);
2425MODULE_PARM_DESC(ibrs_off, "Disable IBRS when idle");