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/nospec-branch.h>
57#include <asm/mwait.h>
58#include <asm/msr.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;
70static unsigned int preferred_states_mask;
71
72static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
73
74static unsigned long auto_demotion_disable_flags;
75
76static enum {
77 C1E_PROMOTION_PRESERVE,
78 C1E_PROMOTION_ENABLE,
79 C1E_PROMOTION_DISABLE
80} c1e_promotion = C1E_PROMOTION_PRESERVE;
81
82struct idle_cpu {
83 struct cpuidle_state *state_table;
84
85 /*
86 * Hardware C-state auto-demotion may not always be optimal.
87 * Indicate which enable bits to clear here.
88 */
89 unsigned long auto_demotion_disable_flags;
90 bool byt_auto_demotion_disable_flag;
91 bool disable_promotion_to_c1e;
92 bool use_acpi;
93};
94
95static const struct idle_cpu *icpu __initdata;
96static struct cpuidle_state *cpuidle_state_table __initdata;
97
98static unsigned int mwait_substates __initdata;
99
100/*
101 * Enable interrupts before entering the C-state. On some platforms and for
102 * some C-states, this may measurably decrease interrupt latency.
103 */
104#define CPUIDLE_FLAG_IRQ_ENABLE BIT(14)
105
106/*
107 * Enable this state by default even if the ACPI _CST does not list it.
108 */
109#define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
110
111/*
112 * Disable IBRS across idle (when KERNEL_IBRS), is exclusive vs IRQ_ENABLE
113 * above.
114 */
115#define CPUIDLE_FLAG_IBRS BIT(16)
116
117/*
118 * Initialize large xstate for the C6-state entrance.
119 */
120#define CPUIDLE_FLAG_INIT_XSTATE BIT(17)
121
122/*
123 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
124 * the C-state (top nibble) and sub-state (bottom nibble)
125 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
126 *
127 * We store the hint at the top of our "flags" for each state.
128 */
129#define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
130#define MWAIT2flg(eax) ((eax & 0xFF) << 24)
131
132static __always_inline int __intel_idle(struct cpuidle_device *dev,
133 struct cpuidle_driver *drv, int index)
134{
135 struct cpuidle_state *state = &drv->states[index];
136 unsigned long eax = flg2MWAIT(state->flags);
137 unsigned long ecx = 1; /* break on interrupt flag */
138
139 mwait_idle_with_hints(eax, ecx);
140
141 return index;
142}
143
144/**
145 * intel_idle - Ask the processor to enter the given idle state.
146 * @dev: cpuidle device of the target CPU.
147 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
148 * @index: Target idle state index.
149 *
150 * Use the MWAIT instruction to notify the processor that the CPU represented by
151 * @dev is idle and it can try to enter the idle state corresponding to @index.
152 *
153 * If the local APIC timer is not known to be reliable in the target idle state,
154 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
155 *
156 * Must be called under local_irq_disable().
157 */
158static __cpuidle int intel_idle(struct cpuidle_device *dev,
159 struct cpuidle_driver *drv, int index)
160{
161 return __intel_idle(dev, drv, index);
162}
163
164static __cpuidle int intel_idle_irq(struct cpuidle_device *dev,
165 struct cpuidle_driver *drv, int index)
166{
167 int ret;
168
169 raw_local_irq_enable();
170 ret = __intel_idle(dev, drv, index);
171
172 /*
173 * The lockdep hardirqs state may be changed to 'on' with timer
174 * tick interrupt followed by __do_softirq(). Use local_irq_disable()
175 * to keep the hardirqs state correct.
176 */
177 local_irq_disable();
178
179 return ret;
180}
181
182static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev,
183 struct cpuidle_driver *drv, int index)
184{
185 bool smt_active = sched_smt_active();
186 u64 spec_ctrl = spec_ctrl_current();
187 int ret;
188
189 if (smt_active)
190 wrmsrl(MSR_IA32_SPEC_CTRL, 0);
191
192 ret = __intel_idle(dev, drv, index);
193
194 if (smt_active)
195 wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl);
196
197 return ret;
198}
199
200static __cpuidle int intel_idle_xstate(struct cpuidle_device *dev,
201 struct cpuidle_driver *drv, int index)
202{
203 fpu_idle_fpregs();
204 return __intel_idle(dev, drv, index);
205}
206
207/**
208 * intel_idle_s2idle - Ask the processor to enter the given idle state.
209 * @dev: cpuidle device of the target CPU.
210 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
211 * @index: Target idle state index.
212 *
213 * Use the MWAIT instruction to notify the processor that the CPU represented by
214 * @dev is idle and it can try to enter the idle state corresponding to @index.
215 *
216 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
217 * scheduler tick and suspended scheduler clock on the target CPU.
218 */
219static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
220 struct cpuidle_driver *drv, int index)
221{
222 unsigned long ecx = 1; /* break on interrupt flag */
223 struct cpuidle_state *state = &drv->states[index];
224 unsigned long eax = flg2MWAIT(state->flags);
225
226 if (state->flags & CPUIDLE_FLAG_INIT_XSTATE)
227 fpu_idle_fpregs();
228
229 mwait_idle_with_hints(eax, ecx);
230
231 return 0;
232}
233
234/*
235 * States are indexed by the cstate number,
236 * which is also the index into the MWAIT hint array.
237 * Thus C0 is a dummy.
238 */
239static struct cpuidle_state nehalem_cstates[] __initdata = {
240 {
241 .name = "C1",
242 .desc = "MWAIT 0x00",
243 .flags = MWAIT2flg(0x00),
244 .exit_latency = 3,
245 .target_residency = 6,
246 .enter = &intel_idle,
247 .enter_s2idle = intel_idle_s2idle, },
248 {
249 .name = "C1E",
250 .desc = "MWAIT 0x01",
251 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
252 .exit_latency = 10,
253 .target_residency = 20,
254 .enter = &intel_idle,
255 .enter_s2idle = intel_idle_s2idle, },
256 {
257 .name = "C3",
258 .desc = "MWAIT 0x10",
259 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
260 .exit_latency = 20,
261 .target_residency = 80,
262 .enter = &intel_idle,
263 .enter_s2idle = intel_idle_s2idle, },
264 {
265 .name = "C6",
266 .desc = "MWAIT 0x20",
267 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
268 .exit_latency = 200,
269 .target_residency = 800,
270 .enter = &intel_idle,
271 .enter_s2idle = intel_idle_s2idle, },
272 {
273 .enter = NULL }
274};
275
276static struct cpuidle_state snb_cstates[] __initdata = {
277 {
278 .name = "C1",
279 .desc = "MWAIT 0x00",
280 .flags = MWAIT2flg(0x00),
281 .exit_latency = 2,
282 .target_residency = 2,
283 .enter = &intel_idle,
284 .enter_s2idle = intel_idle_s2idle, },
285 {
286 .name = "C1E",
287 .desc = "MWAIT 0x01",
288 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
289 .exit_latency = 10,
290 .target_residency = 20,
291 .enter = &intel_idle,
292 .enter_s2idle = intel_idle_s2idle, },
293 {
294 .name = "C3",
295 .desc = "MWAIT 0x10",
296 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
297 .exit_latency = 80,
298 .target_residency = 211,
299 .enter = &intel_idle,
300 .enter_s2idle = intel_idle_s2idle, },
301 {
302 .name = "C6",
303 .desc = "MWAIT 0x20",
304 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
305 .exit_latency = 104,
306 .target_residency = 345,
307 .enter = &intel_idle,
308 .enter_s2idle = intel_idle_s2idle, },
309 {
310 .name = "C7",
311 .desc = "MWAIT 0x30",
312 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
313 .exit_latency = 109,
314 .target_residency = 345,
315 .enter = &intel_idle,
316 .enter_s2idle = intel_idle_s2idle, },
317 {
318 .enter = NULL }
319};
320
321static struct cpuidle_state byt_cstates[] __initdata = {
322 {
323 .name = "C1",
324 .desc = "MWAIT 0x00",
325 .flags = MWAIT2flg(0x00),
326 .exit_latency = 1,
327 .target_residency = 1,
328 .enter = &intel_idle,
329 .enter_s2idle = intel_idle_s2idle, },
330 {
331 .name = "C6N",
332 .desc = "MWAIT 0x58",
333 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
334 .exit_latency = 300,
335 .target_residency = 275,
336 .enter = &intel_idle,
337 .enter_s2idle = intel_idle_s2idle, },
338 {
339 .name = "C6S",
340 .desc = "MWAIT 0x52",
341 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
342 .exit_latency = 500,
343 .target_residency = 560,
344 .enter = &intel_idle,
345 .enter_s2idle = intel_idle_s2idle, },
346 {
347 .name = "C7",
348 .desc = "MWAIT 0x60",
349 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
350 .exit_latency = 1200,
351 .target_residency = 4000,
352 .enter = &intel_idle,
353 .enter_s2idle = intel_idle_s2idle, },
354 {
355 .name = "C7S",
356 .desc = "MWAIT 0x64",
357 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
358 .exit_latency = 10000,
359 .target_residency = 20000,
360 .enter = &intel_idle,
361 .enter_s2idle = intel_idle_s2idle, },
362 {
363 .enter = NULL }
364};
365
366static struct cpuidle_state cht_cstates[] __initdata = {
367 {
368 .name = "C1",
369 .desc = "MWAIT 0x00",
370 .flags = MWAIT2flg(0x00),
371 .exit_latency = 1,
372 .target_residency = 1,
373 .enter = &intel_idle,
374 .enter_s2idle = intel_idle_s2idle, },
375 {
376 .name = "C6N",
377 .desc = "MWAIT 0x58",
378 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
379 .exit_latency = 80,
380 .target_residency = 275,
381 .enter = &intel_idle,
382 .enter_s2idle = intel_idle_s2idle, },
383 {
384 .name = "C6S",
385 .desc = "MWAIT 0x52",
386 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
387 .exit_latency = 200,
388 .target_residency = 560,
389 .enter = &intel_idle,
390 .enter_s2idle = intel_idle_s2idle, },
391 {
392 .name = "C7",
393 .desc = "MWAIT 0x60",
394 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
395 .exit_latency = 1200,
396 .target_residency = 4000,
397 .enter = &intel_idle,
398 .enter_s2idle = intel_idle_s2idle, },
399 {
400 .name = "C7S",
401 .desc = "MWAIT 0x64",
402 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
403 .exit_latency = 10000,
404 .target_residency = 20000,
405 .enter = &intel_idle,
406 .enter_s2idle = intel_idle_s2idle, },
407 {
408 .enter = NULL }
409};
410
411static struct cpuidle_state ivb_cstates[] __initdata = {
412 {
413 .name = "C1",
414 .desc = "MWAIT 0x00",
415 .flags = MWAIT2flg(0x00),
416 .exit_latency = 1,
417 .target_residency = 1,
418 .enter = &intel_idle,
419 .enter_s2idle = intel_idle_s2idle, },
420 {
421 .name = "C1E",
422 .desc = "MWAIT 0x01",
423 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
424 .exit_latency = 10,
425 .target_residency = 20,
426 .enter = &intel_idle,
427 .enter_s2idle = intel_idle_s2idle, },
428 {
429 .name = "C3",
430 .desc = "MWAIT 0x10",
431 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
432 .exit_latency = 59,
433 .target_residency = 156,
434 .enter = &intel_idle,
435 .enter_s2idle = intel_idle_s2idle, },
436 {
437 .name = "C6",
438 .desc = "MWAIT 0x20",
439 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
440 .exit_latency = 80,
441 .target_residency = 300,
442 .enter = &intel_idle,
443 .enter_s2idle = intel_idle_s2idle, },
444 {
445 .name = "C7",
446 .desc = "MWAIT 0x30",
447 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
448 .exit_latency = 87,
449 .target_residency = 300,
450 .enter = &intel_idle,
451 .enter_s2idle = intel_idle_s2idle, },
452 {
453 .enter = NULL }
454};
455
456static struct cpuidle_state ivt_cstates[] __initdata = {
457 {
458 .name = "C1",
459 .desc = "MWAIT 0x00",
460 .flags = MWAIT2flg(0x00),
461 .exit_latency = 1,
462 .target_residency = 1,
463 .enter = &intel_idle,
464 .enter_s2idle = intel_idle_s2idle, },
465 {
466 .name = "C1E",
467 .desc = "MWAIT 0x01",
468 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
469 .exit_latency = 10,
470 .target_residency = 80,
471 .enter = &intel_idle,
472 .enter_s2idle = intel_idle_s2idle, },
473 {
474 .name = "C3",
475 .desc = "MWAIT 0x10",
476 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
477 .exit_latency = 59,
478 .target_residency = 156,
479 .enter = &intel_idle,
480 .enter_s2idle = intel_idle_s2idle, },
481 {
482 .name = "C6",
483 .desc = "MWAIT 0x20",
484 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
485 .exit_latency = 82,
486 .target_residency = 300,
487 .enter = &intel_idle,
488 .enter_s2idle = intel_idle_s2idle, },
489 {
490 .enter = NULL }
491};
492
493static struct cpuidle_state ivt_cstates_4s[] __initdata = {
494 {
495 .name = "C1",
496 .desc = "MWAIT 0x00",
497 .flags = MWAIT2flg(0x00),
498 .exit_latency = 1,
499 .target_residency = 1,
500 .enter = &intel_idle,
501 .enter_s2idle = intel_idle_s2idle, },
502 {
503 .name = "C1E",
504 .desc = "MWAIT 0x01",
505 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
506 .exit_latency = 10,
507 .target_residency = 250,
508 .enter = &intel_idle,
509 .enter_s2idle = intel_idle_s2idle, },
510 {
511 .name = "C3",
512 .desc = "MWAIT 0x10",
513 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
514 .exit_latency = 59,
515 .target_residency = 300,
516 .enter = &intel_idle,
517 .enter_s2idle = intel_idle_s2idle, },
518 {
519 .name = "C6",
520 .desc = "MWAIT 0x20",
521 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
522 .exit_latency = 84,
523 .target_residency = 400,
524 .enter = &intel_idle,
525 .enter_s2idle = intel_idle_s2idle, },
526 {
527 .enter = NULL }
528};
529
530static struct cpuidle_state ivt_cstates_8s[] __initdata = {
531 {
532 .name = "C1",
533 .desc = "MWAIT 0x00",
534 .flags = MWAIT2flg(0x00),
535 .exit_latency = 1,
536 .target_residency = 1,
537 .enter = &intel_idle,
538 .enter_s2idle = intel_idle_s2idle, },
539 {
540 .name = "C1E",
541 .desc = "MWAIT 0x01",
542 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
543 .exit_latency = 10,
544 .target_residency = 500,
545 .enter = &intel_idle,
546 .enter_s2idle = intel_idle_s2idle, },
547 {
548 .name = "C3",
549 .desc = "MWAIT 0x10",
550 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
551 .exit_latency = 59,
552 .target_residency = 600,
553 .enter = &intel_idle,
554 .enter_s2idle = intel_idle_s2idle, },
555 {
556 .name = "C6",
557 .desc = "MWAIT 0x20",
558 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
559 .exit_latency = 88,
560 .target_residency = 700,
561 .enter = &intel_idle,
562 .enter_s2idle = intel_idle_s2idle, },
563 {
564 .enter = NULL }
565};
566
567static struct cpuidle_state hsw_cstates[] __initdata = {
568 {
569 .name = "C1",
570 .desc = "MWAIT 0x00",
571 .flags = MWAIT2flg(0x00),
572 .exit_latency = 2,
573 .target_residency = 2,
574 .enter = &intel_idle,
575 .enter_s2idle = intel_idle_s2idle, },
576 {
577 .name = "C1E",
578 .desc = "MWAIT 0x01",
579 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
580 .exit_latency = 10,
581 .target_residency = 20,
582 .enter = &intel_idle,
583 .enter_s2idle = intel_idle_s2idle, },
584 {
585 .name = "C3",
586 .desc = "MWAIT 0x10",
587 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
588 .exit_latency = 33,
589 .target_residency = 100,
590 .enter = &intel_idle,
591 .enter_s2idle = intel_idle_s2idle, },
592 {
593 .name = "C6",
594 .desc = "MWAIT 0x20",
595 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
596 .exit_latency = 133,
597 .target_residency = 400,
598 .enter = &intel_idle,
599 .enter_s2idle = intel_idle_s2idle, },
600 {
601 .name = "C7s",
602 .desc = "MWAIT 0x32",
603 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
604 .exit_latency = 166,
605 .target_residency = 500,
606 .enter = &intel_idle,
607 .enter_s2idle = intel_idle_s2idle, },
608 {
609 .name = "C8",
610 .desc = "MWAIT 0x40",
611 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
612 .exit_latency = 300,
613 .target_residency = 900,
614 .enter = &intel_idle,
615 .enter_s2idle = intel_idle_s2idle, },
616 {
617 .name = "C9",
618 .desc = "MWAIT 0x50",
619 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
620 .exit_latency = 600,
621 .target_residency = 1800,
622 .enter = &intel_idle,
623 .enter_s2idle = intel_idle_s2idle, },
624 {
625 .name = "C10",
626 .desc = "MWAIT 0x60",
627 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
628 .exit_latency = 2600,
629 .target_residency = 7700,
630 .enter = &intel_idle,
631 .enter_s2idle = intel_idle_s2idle, },
632 {
633 .enter = NULL }
634};
635static struct cpuidle_state bdw_cstates[] __initdata = {
636 {
637 .name = "C1",
638 .desc = "MWAIT 0x00",
639 .flags = MWAIT2flg(0x00),
640 .exit_latency = 2,
641 .target_residency = 2,
642 .enter = &intel_idle,
643 .enter_s2idle = intel_idle_s2idle, },
644 {
645 .name = "C1E",
646 .desc = "MWAIT 0x01",
647 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
648 .exit_latency = 10,
649 .target_residency = 20,
650 .enter = &intel_idle,
651 .enter_s2idle = intel_idle_s2idle, },
652 {
653 .name = "C3",
654 .desc = "MWAIT 0x10",
655 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
656 .exit_latency = 40,
657 .target_residency = 100,
658 .enter = &intel_idle,
659 .enter_s2idle = intel_idle_s2idle, },
660 {
661 .name = "C6",
662 .desc = "MWAIT 0x20",
663 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
664 .exit_latency = 133,
665 .target_residency = 400,
666 .enter = &intel_idle,
667 .enter_s2idle = intel_idle_s2idle, },
668 {
669 .name = "C7s",
670 .desc = "MWAIT 0x32",
671 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
672 .exit_latency = 166,
673 .target_residency = 500,
674 .enter = &intel_idle,
675 .enter_s2idle = intel_idle_s2idle, },
676 {
677 .name = "C8",
678 .desc = "MWAIT 0x40",
679 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
680 .exit_latency = 300,
681 .target_residency = 900,
682 .enter = &intel_idle,
683 .enter_s2idle = intel_idle_s2idle, },
684 {
685 .name = "C9",
686 .desc = "MWAIT 0x50",
687 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
688 .exit_latency = 600,
689 .target_residency = 1800,
690 .enter = &intel_idle,
691 .enter_s2idle = intel_idle_s2idle, },
692 {
693 .name = "C10",
694 .desc = "MWAIT 0x60",
695 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
696 .exit_latency = 2600,
697 .target_residency = 7700,
698 .enter = &intel_idle,
699 .enter_s2idle = intel_idle_s2idle, },
700 {
701 .enter = NULL }
702};
703
704static struct cpuidle_state skl_cstates[] __initdata = {
705 {
706 .name = "C1",
707 .desc = "MWAIT 0x00",
708 .flags = MWAIT2flg(0x00),
709 .exit_latency = 2,
710 .target_residency = 2,
711 .enter = &intel_idle,
712 .enter_s2idle = intel_idle_s2idle, },
713 {
714 .name = "C1E",
715 .desc = "MWAIT 0x01",
716 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
717 .exit_latency = 10,
718 .target_residency = 20,
719 .enter = &intel_idle,
720 .enter_s2idle = intel_idle_s2idle, },
721 {
722 .name = "C3",
723 .desc = "MWAIT 0x10",
724 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
725 .exit_latency = 70,
726 .target_residency = 100,
727 .enter = &intel_idle,
728 .enter_s2idle = intel_idle_s2idle, },
729 {
730 .name = "C6",
731 .desc = "MWAIT 0x20",
732 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
733 .exit_latency = 85,
734 .target_residency = 200,
735 .enter = &intel_idle,
736 .enter_s2idle = intel_idle_s2idle, },
737 {
738 .name = "C7s",
739 .desc = "MWAIT 0x33",
740 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
741 .exit_latency = 124,
742 .target_residency = 800,
743 .enter = &intel_idle,
744 .enter_s2idle = intel_idle_s2idle, },
745 {
746 .name = "C8",
747 .desc = "MWAIT 0x40",
748 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
749 .exit_latency = 200,
750 .target_residency = 800,
751 .enter = &intel_idle,
752 .enter_s2idle = intel_idle_s2idle, },
753 {
754 .name = "C9",
755 .desc = "MWAIT 0x50",
756 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
757 .exit_latency = 480,
758 .target_residency = 5000,
759 .enter = &intel_idle,
760 .enter_s2idle = intel_idle_s2idle, },
761 {
762 .name = "C10",
763 .desc = "MWAIT 0x60",
764 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
765 .exit_latency = 890,
766 .target_residency = 5000,
767 .enter = &intel_idle,
768 .enter_s2idle = intel_idle_s2idle, },
769 {
770 .enter = NULL }
771};
772
773static struct cpuidle_state skx_cstates[] __initdata = {
774 {
775 .name = "C1",
776 .desc = "MWAIT 0x00",
777 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
778 .exit_latency = 2,
779 .target_residency = 2,
780 .enter = &intel_idle,
781 .enter_s2idle = intel_idle_s2idle, },
782 {
783 .name = "C1E",
784 .desc = "MWAIT 0x01",
785 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
786 .exit_latency = 10,
787 .target_residency = 20,
788 .enter = &intel_idle,
789 .enter_s2idle = intel_idle_s2idle, },
790 {
791 .name = "C6",
792 .desc = "MWAIT 0x20",
793 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED | CPUIDLE_FLAG_IBRS,
794 .exit_latency = 133,
795 .target_residency = 600,
796 .enter = &intel_idle,
797 .enter_s2idle = intel_idle_s2idle, },
798 {
799 .enter = NULL }
800};
801
802static struct cpuidle_state icx_cstates[] __initdata = {
803 {
804 .name = "C1",
805 .desc = "MWAIT 0x00",
806 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_IRQ_ENABLE,
807 .exit_latency = 1,
808 .target_residency = 1,
809 .enter = &intel_idle,
810 .enter_s2idle = intel_idle_s2idle, },
811 {
812 .name = "C1E",
813 .desc = "MWAIT 0x01",
814 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
815 .exit_latency = 4,
816 .target_residency = 4,
817 .enter = &intel_idle,
818 .enter_s2idle = intel_idle_s2idle, },
819 {
820 .name = "C6",
821 .desc = "MWAIT 0x20",
822 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
823 .exit_latency = 170,
824 .target_residency = 600,
825 .enter = &intel_idle,
826 .enter_s2idle = intel_idle_s2idle, },
827 {
828 .enter = NULL }
829};
830
831/*
832 * On AlderLake C1 has to be disabled if C1E is enabled, and vice versa.
833 * C1E is enabled only if "C1E promotion" bit is set in MSR_IA32_POWER_CTL.
834 * But in this case there is effectively no C1, because C1 requests are
835 * promoted to C1E. If the "C1E promotion" bit is cleared, then both C1
836 * and C1E requests end up with C1, so there is effectively no C1E.
837 *
838 * By default we enable C1E and disable C1 by marking it with
839 * 'CPUIDLE_FLAG_UNUSABLE'.
840 */
841static struct cpuidle_state adl_cstates[] __initdata = {
842 {
843 .name = "C1",
844 .desc = "MWAIT 0x00",
845 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
846 .exit_latency = 1,
847 .target_residency = 1,
848 .enter = &intel_idle,
849 .enter_s2idle = intel_idle_s2idle, },
850 {
851 .name = "C1E",
852 .desc = "MWAIT 0x01",
853 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
854 .exit_latency = 2,
855 .target_residency = 4,
856 .enter = &intel_idle,
857 .enter_s2idle = intel_idle_s2idle, },
858 {
859 .name = "C6",
860 .desc = "MWAIT 0x20",
861 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
862 .exit_latency = 220,
863 .target_residency = 600,
864 .enter = &intel_idle,
865 .enter_s2idle = intel_idle_s2idle, },
866 {
867 .name = "C8",
868 .desc = "MWAIT 0x40",
869 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
870 .exit_latency = 280,
871 .target_residency = 800,
872 .enter = &intel_idle,
873 .enter_s2idle = intel_idle_s2idle, },
874 {
875 .name = "C10",
876 .desc = "MWAIT 0x60",
877 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
878 .exit_latency = 680,
879 .target_residency = 2000,
880 .enter = &intel_idle,
881 .enter_s2idle = intel_idle_s2idle, },
882 {
883 .enter = NULL }
884};
885
886static struct cpuidle_state adl_l_cstates[] __initdata = {
887 {
888 .name = "C1",
889 .desc = "MWAIT 0x00",
890 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
891 .exit_latency = 1,
892 .target_residency = 1,
893 .enter = &intel_idle,
894 .enter_s2idle = intel_idle_s2idle, },
895 {
896 .name = "C1E",
897 .desc = "MWAIT 0x01",
898 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
899 .exit_latency = 2,
900 .target_residency = 4,
901 .enter = &intel_idle,
902 .enter_s2idle = intel_idle_s2idle, },
903 {
904 .name = "C6",
905 .desc = "MWAIT 0x20",
906 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
907 .exit_latency = 170,
908 .target_residency = 500,
909 .enter = &intel_idle,
910 .enter_s2idle = intel_idle_s2idle, },
911 {
912 .name = "C8",
913 .desc = "MWAIT 0x40",
914 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
915 .exit_latency = 200,
916 .target_residency = 600,
917 .enter = &intel_idle,
918 .enter_s2idle = intel_idle_s2idle, },
919 {
920 .name = "C10",
921 .desc = "MWAIT 0x60",
922 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
923 .exit_latency = 230,
924 .target_residency = 700,
925 .enter = &intel_idle,
926 .enter_s2idle = intel_idle_s2idle, },
927 {
928 .enter = NULL }
929};
930
931static struct cpuidle_state adl_n_cstates[] __initdata = {
932 {
933 .name = "C1",
934 .desc = "MWAIT 0x00",
935 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_UNUSABLE,
936 .exit_latency = 1,
937 .target_residency = 1,
938 .enter = &intel_idle,
939 .enter_s2idle = intel_idle_s2idle, },
940 {
941 .name = "C1E",
942 .desc = "MWAIT 0x01",
943 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
944 .exit_latency = 2,
945 .target_residency = 4,
946 .enter = &intel_idle,
947 .enter_s2idle = intel_idle_s2idle, },
948 {
949 .name = "C6",
950 .desc = "MWAIT 0x20",
951 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
952 .exit_latency = 195,
953 .target_residency = 585,
954 .enter = &intel_idle,
955 .enter_s2idle = intel_idle_s2idle, },
956 {
957 .name = "C8",
958 .desc = "MWAIT 0x40",
959 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
960 .exit_latency = 260,
961 .target_residency = 1040,
962 .enter = &intel_idle,
963 .enter_s2idle = intel_idle_s2idle, },
964 {
965 .name = "C10",
966 .desc = "MWAIT 0x60",
967 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
968 .exit_latency = 660,
969 .target_residency = 1980,
970 .enter = &intel_idle,
971 .enter_s2idle = intel_idle_s2idle, },
972 {
973 .enter = NULL }
974};
975
976static struct cpuidle_state spr_cstates[] __initdata = {
977 {
978 .name = "C1",
979 .desc = "MWAIT 0x00",
980 .flags = MWAIT2flg(0x00),
981 .exit_latency = 1,
982 .target_residency = 1,
983 .enter = &intel_idle,
984 .enter_s2idle = intel_idle_s2idle, },
985 {
986 .name = "C1E",
987 .desc = "MWAIT 0x01",
988 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
989 .exit_latency = 2,
990 .target_residency = 4,
991 .enter = &intel_idle,
992 .enter_s2idle = intel_idle_s2idle, },
993 {
994 .name = "C6",
995 .desc = "MWAIT 0x20",
996 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED |
997 CPUIDLE_FLAG_INIT_XSTATE,
998 .exit_latency = 290,
999 .target_residency = 800,
1000 .enter = &intel_idle,
1001 .enter_s2idle = intel_idle_s2idle, },
1002 {
1003 .enter = NULL }
1004};
1005
1006static struct cpuidle_state atom_cstates[] __initdata = {
1007 {
1008 .name = "C1E",
1009 .desc = "MWAIT 0x00",
1010 .flags = MWAIT2flg(0x00),
1011 .exit_latency = 10,
1012 .target_residency = 20,
1013 .enter = &intel_idle,
1014 .enter_s2idle = intel_idle_s2idle, },
1015 {
1016 .name = "C2",
1017 .desc = "MWAIT 0x10",
1018 .flags = MWAIT2flg(0x10),
1019 .exit_latency = 20,
1020 .target_residency = 80,
1021 .enter = &intel_idle,
1022 .enter_s2idle = intel_idle_s2idle, },
1023 {
1024 .name = "C4",
1025 .desc = "MWAIT 0x30",
1026 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1027 .exit_latency = 100,
1028 .target_residency = 400,
1029 .enter = &intel_idle,
1030 .enter_s2idle = intel_idle_s2idle, },
1031 {
1032 .name = "C6",
1033 .desc = "MWAIT 0x52",
1034 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1035 .exit_latency = 140,
1036 .target_residency = 560,
1037 .enter = &intel_idle,
1038 .enter_s2idle = intel_idle_s2idle, },
1039 {
1040 .enter = NULL }
1041};
1042static struct cpuidle_state tangier_cstates[] __initdata = {
1043 {
1044 .name = "C1",
1045 .desc = "MWAIT 0x00",
1046 .flags = MWAIT2flg(0x00),
1047 .exit_latency = 1,
1048 .target_residency = 4,
1049 .enter = &intel_idle,
1050 .enter_s2idle = intel_idle_s2idle, },
1051 {
1052 .name = "C4",
1053 .desc = "MWAIT 0x30",
1054 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
1055 .exit_latency = 100,
1056 .target_residency = 400,
1057 .enter = &intel_idle,
1058 .enter_s2idle = intel_idle_s2idle, },
1059 {
1060 .name = "C6",
1061 .desc = "MWAIT 0x52",
1062 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
1063 .exit_latency = 140,
1064 .target_residency = 560,
1065 .enter = &intel_idle,
1066 .enter_s2idle = intel_idle_s2idle, },
1067 {
1068 .name = "C7",
1069 .desc = "MWAIT 0x60",
1070 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1071 .exit_latency = 1200,
1072 .target_residency = 4000,
1073 .enter = &intel_idle,
1074 .enter_s2idle = intel_idle_s2idle, },
1075 {
1076 .name = "C9",
1077 .desc = "MWAIT 0x64",
1078 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
1079 .exit_latency = 10000,
1080 .target_residency = 20000,
1081 .enter = &intel_idle,
1082 .enter_s2idle = intel_idle_s2idle, },
1083 {
1084 .enter = NULL }
1085};
1086static struct cpuidle_state avn_cstates[] __initdata = {
1087 {
1088 .name = "C1",
1089 .desc = "MWAIT 0x00",
1090 .flags = MWAIT2flg(0x00),
1091 .exit_latency = 2,
1092 .target_residency = 2,
1093 .enter = &intel_idle,
1094 .enter_s2idle = intel_idle_s2idle, },
1095 {
1096 .name = "C6",
1097 .desc = "MWAIT 0x51",
1098 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
1099 .exit_latency = 15,
1100 .target_residency = 45,
1101 .enter = &intel_idle,
1102 .enter_s2idle = intel_idle_s2idle, },
1103 {
1104 .enter = NULL }
1105};
1106static struct cpuidle_state knl_cstates[] __initdata = {
1107 {
1108 .name = "C1",
1109 .desc = "MWAIT 0x00",
1110 .flags = MWAIT2flg(0x00),
1111 .exit_latency = 1,
1112 .target_residency = 2,
1113 .enter = &intel_idle,
1114 .enter_s2idle = intel_idle_s2idle },
1115 {
1116 .name = "C6",
1117 .desc = "MWAIT 0x10",
1118 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
1119 .exit_latency = 120,
1120 .target_residency = 500,
1121 .enter = &intel_idle,
1122 .enter_s2idle = intel_idle_s2idle },
1123 {
1124 .enter = NULL }
1125};
1126
1127static struct cpuidle_state bxt_cstates[] __initdata = {
1128 {
1129 .name = "C1",
1130 .desc = "MWAIT 0x00",
1131 .flags = MWAIT2flg(0x00),
1132 .exit_latency = 2,
1133 .target_residency = 2,
1134 .enter = &intel_idle,
1135 .enter_s2idle = intel_idle_s2idle, },
1136 {
1137 .name = "C1E",
1138 .desc = "MWAIT 0x01",
1139 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1140 .exit_latency = 10,
1141 .target_residency = 20,
1142 .enter = &intel_idle,
1143 .enter_s2idle = intel_idle_s2idle, },
1144 {
1145 .name = "C6",
1146 .desc = "MWAIT 0x20",
1147 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1148 .exit_latency = 133,
1149 .target_residency = 133,
1150 .enter = &intel_idle,
1151 .enter_s2idle = intel_idle_s2idle, },
1152 {
1153 .name = "C7s",
1154 .desc = "MWAIT 0x31",
1155 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
1156 .exit_latency = 155,
1157 .target_residency = 155,
1158 .enter = &intel_idle,
1159 .enter_s2idle = intel_idle_s2idle, },
1160 {
1161 .name = "C8",
1162 .desc = "MWAIT 0x40",
1163 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
1164 .exit_latency = 1000,
1165 .target_residency = 1000,
1166 .enter = &intel_idle,
1167 .enter_s2idle = intel_idle_s2idle, },
1168 {
1169 .name = "C9",
1170 .desc = "MWAIT 0x50",
1171 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
1172 .exit_latency = 2000,
1173 .target_residency = 2000,
1174 .enter = &intel_idle,
1175 .enter_s2idle = intel_idle_s2idle, },
1176 {
1177 .name = "C10",
1178 .desc = "MWAIT 0x60",
1179 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
1180 .exit_latency = 10000,
1181 .target_residency = 10000,
1182 .enter = &intel_idle,
1183 .enter_s2idle = intel_idle_s2idle, },
1184 {
1185 .enter = NULL }
1186};
1187
1188static struct cpuidle_state dnv_cstates[] __initdata = {
1189 {
1190 .name = "C1",
1191 .desc = "MWAIT 0x00",
1192 .flags = MWAIT2flg(0x00),
1193 .exit_latency = 2,
1194 .target_residency = 2,
1195 .enter = &intel_idle,
1196 .enter_s2idle = intel_idle_s2idle, },
1197 {
1198 .name = "C1E",
1199 .desc = "MWAIT 0x01",
1200 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1201 .exit_latency = 10,
1202 .target_residency = 20,
1203 .enter = &intel_idle,
1204 .enter_s2idle = intel_idle_s2idle, },
1205 {
1206 .name = "C6",
1207 .desc = "MWAIT 0x20",
1208 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1209 .exit_latency = 50,
1210 .target_residency = 500,
1211 .enter = &intel_idle,
1212 .enter_s2idle = intel_idle_s2idle, },
1213 {
1214 .enter = NULL }
1215};
1216
1217/*
1218 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
1219 * C6, and this is indicated in the CPUID mwait leaf.
1220 */
1221static struct cpuidle_state snr_cstates[] __initdata = {
1222 {
1223 .name = "C1",
1224 .desc = "MWAIT 0x00",
1225 .flags = MWAIT2flg(0x00),
1226 .exit_latency = 2,
1227 .target_residency = 2,
1228 .enter = &intel_idle,
1229 .enter_s2idle = intel_idle_s2idle, },
1230 {
1231 .name = "C1E",
1232 .desc = "MWAIT 0x01",
1233 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
1234 .exit_latency = 15,
1235 .target_residency = 25,
1236 .enter = &intel_idle,
1237 .enter_s2idle = intel_idle_s2idle, },
1238 {
1239 .name = "C6",
1240 .desc = "MWAIT 0x20",
1241 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
1242 .exit_latency = 130,
1243 .target_residency = 500,
1244 .enter = &intel_idle,
1245 .enter_s2idle = intel_idle_s2idle, },
1246 {
1247 .enter = NULL }
1248};
1249
1250static const struct idle_cpu idle_cpu_nehalem __initconst = {
1251 .state_table = nehalem_cstates,
1252 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1253 .disable_promotion_to_c1e = true,
1254};
1255
1256static const struct idle_cpu idle_cpu_nhx __initconst = {
1257 .state_table = nehalem_cstates,
1258 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1259 .disable_promotion_to_c1e = true,
1260 .use_acpi = true,
1261};
1262
1263static const struct idle_cpu idle_cpu_atom __initconst = {
1264 .state_table = atom_cstates,
1265};
1266
1267static const struct idle_cpu idle_cpu_tangier __initconst = {
1268 .state_table = tangier_cstates,
1269};
1270
1271static const struct idle_cpu idle_cpu_lincroft __initconst = {
1272 .state_table = atom_cstates,
1273 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1274};
1275
1276static const struct idle_cpu idle_cpu_snb __initconst = {
1277 .state_table = snb_cstates,
1278 .disable_promotion_to_c1e = true,
1279};
1280
1281static const struct idle_cpu idle_cpu_snx __initconst = {
1282 .state_table = snb_cstates,
1283 .disable_promotion_to_c1e = true,
1284 .use_acpi = true,
1285};
1286
1287static const struct idle_cpu idle_cpu_byt __initconst = {
1288 .state_table = byt_cstates,
1289 .disable_promotion_to_c1e = true,
1290 .byt_auto_demotion_disable_flag = true,
1291};
1292
1293static const struct idle_cpu idle_cpu_cht __initconst = {
1294 .state_table = cht_cstates,
1295 .disable_promotion_to_c1e = true,
1296 .byt_auto_demotion_disable_flag = true,
1297};
1298
1299static const struct idle_cpu idle_cpu_ivb __initconst = {
1300 .state_table = ivb_cstates,
1301 .disable_promotion_to_c1e = true,
1302};
1303
1304static const struct idle_cpu idle_cpu_ivt __initconst = {
1305 .state_table = ivt_cstates,
1306 .disable_promotion_to_c1e = true,
1307 .use_acpi = true,
1308};
1309
1310static const struct idle_cpu idle_cpu_hsw __initconst = {
1311 .state_table = hsw_cstates,
1312 .disable_promotion_to_c1e = true,
1313};
1314
1315static const struct idle_cpu idle_cpu_hsx __initconst = {
1316 .state_table = hsw_cstates,
1317 .disable_promotion_to_c1e = true,
1318 .use_acpi = true,
1319};
1320
1321static const struct idle_cpu idle_cpu_bdw __initconst = {
1322 .state_table = bdw_cstates,
1323 .disable_promotion_to_c1e = true,
1324};
1325
1326static const struct idle_cpu idle_cpu_bdx __initconst = {
1327 .state_table = bdw_cstates,
1328 .disable_promotion_to_c1e = true,
1329 .use_acpi = true,
1330};
1331
1332static const struct idle_cpu idle_cpu_skl __initconst = {
1333 .state_table = skl_cstates,
1334 .disable_promotion_to_c1e = true,
1335};
1336
1337static const struct idle_cpu idle_cpu_skx __initconst = {
1338 .state_table = skx_cstates,
1339 .disable_promotion_to_c1e = true,
1340 .use_acpi = true,
1341};
1342
1343static const struct idle_cpu idle_cpu_icx __initconst = {
1344 .state_table = icx_cstates,
1345 .disable_promotion_to_c1e = true,
1346 .use_acpi = true,
1347};
1348
1349static const struct idle_cpu idle_cpu_adl __initconst = {
1350 .state_table = adl_cstates,
1351};
1352
1353static const struct idle_cpu idle_cpu_adl_l __initconst = {
1354 .state_table = adl_l_cstates,
1355};
1356
1357static const struct idle_cpu idle_cpu_adl_n __initconst = {
1358 .state_table = adl_n_cstates,
1359};
1360
1361static const struct idle_cpu idle_cpu_spr __initconst = {
1362 .state_table = spr_cstates,
1363 .disable_promotion_to_c1e = true,
1364 .use_acpi = true,
1365};
1366
1367static const struct idle_cpu idle_cpu_avn __initconst = {
1368 .state_table = avn_cstates,
1369 .disable_promotion_to_c1e = true,
1370 .use_acpi = true,
1371};
1372
1373static const struct idle_cpu idle_cpu_knl __initconst = {
1374 .state_table = knl_cstates,
1375 .use_acpi = true,
1376};
1377
1378static const struct idle_cpu idle_cpu_bxt __initconst = {
1379 .state_table = bxt_cstates,
1380 .disable_promotion_to_c1e = true,
1381};
1382
1383static const struct idle_cpu idle_cpu_dnv __initconst = {
1384 .state_table = dnv_cstates,
1385 .disable_promotion_to_c1e = true,
1386 .use_acpi = true,
1387};
1388
1389static const struct idle_cpu idle_cpu_snr __initconst = {
1390 .state_table = snr_cstates,
1391 .disable_promotion_to_c1e = true,
1392 .use_acpi = true,
1393};
1394
1395static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1396 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EP, &idle_cpu_nhx),
1397 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &idle_cpu_nehalem),
1398 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_G, &idle_cpu_nehalem),
1399 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE, &idle_cpu_nehalem),
1400 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EP, &idle_cpu_nhx),
1401 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EX, &idle_cpu_nhx),
1402 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL, &idle_cpu_atom),
1403 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL_MID, &idle_cpu_lincroft),
1404 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EX, &idle_cpu_nhx),
1405 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE, &idle_cpu_snb),
1406 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &idle_cpu_snx),
1407 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SALTWELL, &idle_cpu_atom),
1408 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT, &idle_cpu_byt),
1409 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1410 X86_MATCH_INTEL_FAM6_MODEL(ATOM_AIRMONT, &idle_cpu_cht),
1411 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE, &idle_cpu_ivb),
1412 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &idle_cpu_ivt),
1413 X86_MATCH_INTEL_FAM6_MODEL(HASWELL, &idle_cpu_hsw),
1414 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &idle_cpu_hsx),
1415 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_L, &idle_cpu_hsw),
1416 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_G, &idle_cpu_hsw),
1417 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_D, &idle_cpu_avn),
1418 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL, &idle_cpu_bdw),
1419 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_G, &idle_cpu_bdw),
1420 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &idle_cpu_bdx),
1421 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &idle_cpu_bdx),
1422 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_L, &idle_cpu_skl),
1423 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE, &idle_cpu_skl),
1424 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE_L, &idle_cpu_skl),
1425 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE, &idle_cpu_skl),
1426 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &idle_cpu_skx),
1427 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &idle_cpu_icx),
1428 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &idle_cpu_icx),
1429 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE, &idle_cpu_adl),
1430 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_L, &idle_cpu_adl_l),
1431 X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE_N, &idle_cpu_adl_n),
1432 X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &idle_cpu_spr),
1433 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &idle_cpu_knl),
1434 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &idle_cpu_knl),
1435 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT, &idle_cpu_bxt),
1436 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1437 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_D, &idle_cpu_dnv),
1438 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, &idle_cpu_snr),
1439 {}
1440};
1441
1442static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1443 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1444 {}
1445};
1446
1447static bool __init intel_idle_max_cstate_reached(int cstate)
1448{
1449 if (cstate + 1 > max_cstate) {
1450 pr_info("max_cstate %d reached\n", max_cstate);
1451 return true;
1452 }
1453 return false;
1454}
1455
1456static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1457{
1458 unsigned long eax = flg2MWAIT(state->flags);
1459
1460 if (boot_cpu_has(X86_FEATURE_ARAT))
1461 return false;
1462
1463 /*
1464 * Switch over to one-shot tick broadcast if the target C-state
1465 * is deeper than C1.
1466 */
1467 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1468}
1469
1470#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1471#include <acpi/processor.h>
1472
1473static bool no_acpi __read_mostly;
1474module_param(no_acpi, bool, 0444);
1475MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1476
1477static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1478module_param_named(use_acpi, force_use_acpi, bool, 0444);
1479MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1480
1481static struct acpi_processor_power acpi_state_table __initdata;
1482
1483/**
1484 * intel_idle_cst_usable - Check if the _CST information can be used.
1485 *
1486 * Check if all of the C-states listed by _CST in the max_cstate range are
1487 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1488 */
1489static bool __init intel_idle_cst_usable(void)
1490{
1491 int cstate, limit;
1492
1493 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1494 acpi_state_table.count);
1495
1496 for (cstate = 1; cstate < limit; cstate++) {
1497 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1498
1499 if (cx->entry_method != ACPI_CSTATE_FFH)
1500 return false;
1501 }
1502
1503 return true;
1504}
1505
1506static bool __init intel_idle_acpi_cst_extract(void)
1507{
1508 unsigned int cpu;
1509
1510 if (no_acpi) {
1511 pr_debug("Not allowed to use ACPI _CST\n");
1512 return false;
1513 }
1514
1515 for_each_possible_cpu(cpu) {
1516 struct acpi_processor *pr = per_cpu(processors, cpu);
1517
1518 if (!pr)
1519 continue;
1520
1521 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1522 continue;
1523
1524 acpi_state_table.count++;
1525
1526 if (!intel_idle_cst_usable())
1527 continue;
1528
1529 if (!acpi_processor_claim_cst_control())
1530 break;
1531
1532 return true;
1533 }
1534
1535 acpi_state_table.count = 0;
1536 pr_debug("ACPI _CST not found or not usable\n");
1537 return false;
1538}
1539
1540static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1541{
1542 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1543
1544 /*
1545 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1546 * the interesting states are ACPI_CSTATE_FFH.
1547 */
1548 for (cstate = 1; cstate < limit; cstate++) {
1549 struct acpi_processor_cx *cx;
1550 struct cpuidle_state *state;
1551
1552 if (intel_idle_max_cstate_reached(cstate - 1))
1553 break;
1554
1555 cx = &acpi_state_table.states[cstate];
1556
1557 state = &drv->states[drv->state_count++];
1558
1559 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1560 strscpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1561 state->exit_latency = cx->latency;
1562 /*
1563 * For C1-type C-states use the same number for both the exit
1564 * latency and target residency, because that is the case for
1565 * C1 in the majority of the static C-states tables above.
1566 * For the other types of C-states, however, set the target
1567 * residency to 3 times the exit latency which should lead to
1568 * a reasonable balance between energy-efficiency and
1569 * performance in the majority of interesting cases.
1570 */
1571 state->target_residency = cx->latency;
1572 if (cx->type > ACPI_STATE_C1)
1573 state->target_residency *= 3;
1574
1575 state->flags = MWAIT2flg(cx->address);
1576 if (cx->type > ACPI_STATE_C2)
1577 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1578
1579 if (disabled_states_mask & BIT(cstate))
1580 state->flags |= CPUIDLE_FLAG_OFF;
1581
1582 if (intel_idle_state_needs_timer_stop(state))
1583 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1584
1585 state->enter = intel_idle;
1586 state->enter_s2idle = intel_idle_s2idle;
1587 }
1588}
1589
1590static bool __init intel_idle_off_by_default(u32 mwait_hint)
1591{
1592 int cstate, limit;
1593
1594 /*
1595 * If there are no _CST C-states, do not disable any C-states by
1596 * default.
1597 */
1598 if (!acpi_state_table.count)
1599 return false;
1600
1601 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1602 /*
1603 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1604 * the interesting states are ACPI_CSTATE_FFH.
1605 */
1606 for (cstate = 1; cstate < limit; cstate++) {
1607 if (acpi_state_table.states[cstate].address == mwait_hint)
1608 return false;
1609 }
1610 return true;
1611}
1612#else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1613#define force_use_acpi (false)
1614
1615static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1616static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1617static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
1618#endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1619
1620/**
1621 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1622 *
1623 * Tune IVT multi-socket targets.
1624 * Assumption: num_sockets == (max_package_num + 1).
1625 */
1626static void __init ivt_idle_state_table_update(void)
1627{
1628 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1629 int cpu, package_num, num_sockets = 1;
1630
1631 for_each_online_cpu(cpu) {
1632 package_num = topology_physical_package_id(cpu);
1633 if (package_num + 1 > num_sockets) {
1634 num_sockets = package_num + 1;
1635
1636 if (num_sockets > 4) {
1637 cpuidle_state_table = ivt_cstates_8s;
1638 return;
1639 }
1640 }
1641 }
1642
1643 if (num_sockets > 2)
1644 cpuidle_state_table = ivt_cstates_4s;
1645
1646 /* else, 1 and 2 socket systems use default ivt_cstates */
1647}
1648
1649/**
1650 * irtl_2_usec - IRTL to microseconds conversion.
1651 * @irtl: IRTL MSR value.
1652 *
1653 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1654 */
1655static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1656{
1657 static const unsigned int irtl_ns_units[] __initconst = {
1658 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1659 };
1660 unsigned long long ns;
1661
1662 if (!irtl)
1663 return 0;
1664
1665 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1666
1667 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1668}
1669
1670/**
1671 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1672 *
1673 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1674 * definitive maximum latency and use the same value for target_residency.
1675 */
1676static void __init bxt_idle_state_table_update(void)
1677{
1678 unsigned long long msr;
1679 unsigned int usec;
1680
1681 rdmsrl(MSR_PKGC6_IRTL, msr);
1682 usec = irtl_2_usec(msr);
1683 if (usec) {
1684 bxt_cstates[2].exit_latency = usec;
1685 bxt_cstates[2].target_residency = usec;
1686 }
1687
1688 rdmsrl(MSR_PKGC7_IRTL, msr);
1689 usec = irtl_2_usec(msr);
1690 if (usec) {
1691 bxt_cstates[3].exit_latency = usec;
1692 bxt_cstates[3].target_residency = usec;
1693 }
1694
1695 rdmsrl(MSR_PKGC8_IRTL, msr);
1696 usec = irtl_2_usec(msr);
1697 if (usec) {
1698 bxt_cstates[4].exit_latency = usec;
1699 bxt_cstates[4].target_residency = usec;
1700 }
1701
1702 rdmsrl(MSR_PKGC9_IRTL, msr);
1703 usec = irtl_2_usec(msr);
1704 if (usec) {
1705 bxt_cstates[5].exit_latency = usec;
1706 bxt_cstates[5].target_residency = usec;
1707 }
1708
1709 rdmsrl(MSR_PKGC10_IRTL, msr);
1710 usec = irtl_2_usec(msr);
1711 if (usec) {
1712 bxt_cstates[6].exit_latency = usec;
1713 bxt_cstates[6].target_residency = usec;
1714 }
1715
1716}
1717
1718/**
1719 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1720 *
1721 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1722 */
1723static void __init sklh_idle_state_table_update(void)
1724{
1725 unsigned long long msr;
1726 unsigned int eax, ebx, ecx, edx;
1727
1728
1729 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1730 if (max_cstate <= 7)
1731 return;
1732
1733 /* if PC10 not present in CPUID.MWAIT.EDX */
1734 if ((mwait_substates & (0xF << 28)) == 0)
1735 return;
1736
1737 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1738
1739 /* PC10 is not enabled in PKG C-state limit */
1740 if ((msr & 0xF) != 8)
1741 return;
1742
1743 ecx = 0;
1744 cpuid(7, &eax, &ebx, &ecx, &edx);
1745
1746 /* if SGX is present */
1747 if (ebx & (1 << 2)) {
1748
1749 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1750
1751 /* if SGX is enabled */
1752 if (msr & (1 << 18))
1753 return;
1754 }
1755
1756 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1757 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1758}
1759
1760/**
1761 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1762 * idle states table.
1763 */
1764static void __init skx_idle_state_table_update(void)
1765{
1766 unsigned long long msr;
1767
1768 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1769
1770 /*
1771 * 000b: C0/C1 (no package C-state support)
1772 * 001b: C2
1773 * 010b: C6 (non-retention)
1774 * 011b: C6 (retention)
1775 * 111b: No Package C state limits.
1776 */
1777 if ((msr & 0x7) < 2) {
1778 /*
1779 * Uses the CC6 + PC0 latency and 3 times of
1780 * latency for target_residency if the PC6
1781 * is disabled in BIOS. This is consistent
1782 * with how intel_idle driver uses _CST
1783 * to set the target_residency.
1784 */
1785 skx_cstates[2].exit_latency = 92;
1786 skx_cstates[2].target_residency = 276;
1787 }
1788}
1789
1790/**
1791 * adl_idle_state_table_update - Adjust AlderLake idle states table.
1792 */
1793static void __init adl_idle_state_table_update(void)
1794{
1795 /* Check if user prefers C1 over C1E. */
1796 if (preferred_states_mask & BIT(1) && !(preferred_states_mask & BIT(2))) {
1797 cpuidle_state_table[0].flags &= ~CPUIDLE_FLAG_UNUSABLE;
1798 cpuidle_state_table[1].flags |= CPUIDLE_FLAG_UNUSABLE;
1799
1800 /* Disable C1E by clearing the "C1E promotion" bit. */
1801 c1e_promotion = C1E_PROMOTION_DISABLE;
1802 return;
1803 }
1804
1805 /* Make sure C1E is enabled by default */
1806 c1e_promotion = C1E_PROMOTION_ENABLE;
1807}
1808
1809/**
1810 * spr_idle_state_table_update - Adjust Sapphire Rapids idle states table.
1811 */
1812static void __init spr_idle_state_table_update(void)
1813{
1814 unsigned long long msr;
1815
1816 /*
1817 * By default, the C6 state assumes the worst-case scenario of package
1818 * C6. However, if PC6 is disabled, we update the numbers to match
1819 * core C6.
1820 */
1821 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1822
1823 /* Limit value 2 and above allow for PC6. */
1824 if ((msr & 0x7) < 2) {
1825 spr_cstates[2].exit_latency = 190;
1826 spr_cstates[2].target_residency = 600;
1827 }
1828}
1829
1830static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
1831{
1832 unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
1833 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
1834 MWAIT_SUBSTATE_MASK;
1835
1836 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
1837 if (num_substates == 0)
1838 return false;
1839
1840 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1841 mark_tsc_unstable("TSC halts in idle states deeper than C2");
1842
1843 return true;
1844}
1845
1846static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
1847{
1848 int cstate;
1849
1850 switch (boot_cpu_data.x86_model) {
1851 case INTEL_FAM6_IVYBRIDGE_X:
1852 ivt_idle_state_table_update();
1853 break;
1854 case INTEL_FAM6_ATOM_GOLDMONT:
1855 case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
1856 bxt_idle_state_table_update();
1857 break;
1858 case INTEL_FAM6_SKYLAKE:
1859 sklh_idle_state_table_update();
1860 break;
1861 case INTEL_FAM6_SKYLAKE_X:
1862 skx_idle_state_table_update();
1863 break;
1864 case INTEL_FAM6_SAPPHIRERAPIDS_X:
1865 spr_idle_state_table_update();
1866 break;
1867 case INTEL_FAM6_ALDERLAKE:
1868 case INTEL_FAM6_ALDERLAKE_L:
1869 case INTEL_FAM6_ALDERLAKE_N:
1870 adl_idle_state_table_update();
1871 break;
1872 }
1873
1874 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1875 unsigned int mwait_hint;
1876
1877 if (intel_idle_max_cstate_reached(cstate))
1878 break;
1879
1880 if (!cpuidle_state_table[cstate].enter &&
1881 !cpuidle_state_table[cstate].enter_s2idle)
1882 break;
1883
1884 /* If marked as unusable, skip this state. */
1885 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
1886 pr_debug("state %s is disabled\n",
1887 cpuidle_state_table[cstate].name);
1888 continue;
1889 }
1890
1891 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1892 if (!intel_idle_verify_cstate(mwait_hint))
1893 continue;
1894
1895 /* Structure copy. */
1896 drv->states[drv->state_count] = cpuidle_state_table[cstate];
1897
1898 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IRQ_ENABLE)
1899 drv->states[drv->state_count].enter = intel_idle_irq;
1900
1901 if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) &&
1902 cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IBRS) {
1903 WARN_ON_ONCE(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_IRQ_ENABLE);
1904 drv->states[drv->state_count].enter = intel_idle_ibrs;
1905 }
1906
1907 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_INIT_XSTATE)
1908 drv->states[drv->state_count].enter = intel_idle_xstate;
1909
1910 if ((disabled_states_mask & BIT(drv->state_count)) ||
1911 ((icpu->use_acpi || force_use_acpi) &&
1912 intel_idle_off_by_default(mwait_hint) &&
1913 !(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
1914 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;
1915
1916 if (intel_idle_state_needs_timer_stop(&drv->states[drv->state_count]))
1917 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_TIMER_STOP;
1918
1919 drv->state_count++;
1920 }
1921
1922 if (icpu->byt_auto_demotion_disable_flag) {
1923 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
1924 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
1925 }
1926}
1927
1928/**
1929 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1930 * @drv: cpuidle driver structure to initialize.
1931 */
1932static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
1933{
1934 cpuidle_poll_state_init(drv);
1935
1936 if (disabled_states_mask & BIT(0))
1937 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
1938
1939 drv->state_count = 1;
1940
1941 if (icpu)
1942 intel_idle_init_cstates_icpu(drv);
1943 else
1944 intel_idle_init_cstates_acpi(drv);
1945}
1946
1947static void auto_demotion_disable(void)
1948{
1949 unsigned long long msr_bits;
1950
1951 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1952 msr_bits &= ~auto_demotion_disable_flags;
1953 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1954}
1955
1956static void c1e_promotion_enable(void)
1957{
1958 unsigned long long msr_bits;
1959
1960 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1961 msr_bits |= 0x2;
1962 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1963}
1964
1965static void c1e_promotion_disable(void)
1966{
1967 unsigned long long msr_bits;
1968
1969 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1970 msr_bits &= ~0x2;
1971 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1972}
1973
1974/**
1975 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
1976 * @cpu: CPU to initialize.
1977 *
1978 * Register a cpuidle device object for @cpu and update its MSRs in accordance
1979 * with the processor model flags.
1980 */
1981static int intel_idle_cpu_init(unsigned int cpu)
1982{
1983 struct cpuidle_device *dev;
1984
1985 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1986 dev->cpu = cpu;
1987
1988 if (cpuidle_register_device(dev)) {
1989 pr_debug("cpuidle_register_device %d failed!\n", cpu);
1990 return -EIO;
1991 }
1992
1993 if (auto_demotion_disable_flags)
1994 auto_demotion_disable();
1995
1996 if (c1e_promotion == C1E_PROMOTION_ENABLE)
1997 c1e_promotion_enable();
1998 else if (c1e_promotion == C1E_PROMOTION_DISABLE)
1999 c1e_promotion_disable();
2000
2001 return 0;
2002}
2003
2004static int intel_idle_cpu_online(unsigned int cpu)
2005{
2006 struct cpuidle_device *dev;
2007
2008 if (!boot_cpu_has(X86_FEATURE_ARAT))
2009 tick_broadcast_enable();
2010
2011 /*
2012 * Some systems can hotplug a cpu at runtime after
2013 * the kernel has booted, we have to initialize the
2014 * driver in this case
2015 */
2016 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
2017 if (!dev->registered)
2018 return intel_idle_cpu_init(cpu);
2019
2020 return 0;
2021}
2022
2023/**
2024 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
2025 */
2026static void __init intel_idle_cpuidle_devices_uninit(void)
2027{
2028 int i;
2029
2030 for_each_online_cpu(i)
2031 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
2032}
2033
2034static int __init intel_idle_init(void)
2035{
2036 const struct x86_cpu_id *id;
2037 unsigned int eax, ebx, ecx;
2038 int retval;
2039
2040 /* Do not load intel_idle at all for now if idle= is passed */
2041 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
2042 return -ENODEV;
2043
2044 if (max_cstate == 0) {
2045 pr_debug("disabled\n");
2046 return -EPERM;
2047 }
2048
2049 id = x86_match_cpu(intel_idle_ids);
2050 if (id) {
2051 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
2052 pr_debug("Please enable MWAIT in BIOS SETUP\n");
2053 return -ENODEV;
2054 }
2055 } else {
2056 id = x86_match_cpu(intel_mwait_ids);
2057 if (!id)
2058 return -ENODEV;
2059 }
2060
2061 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
2062 return -ENODEV;
2063
2064 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
2065
2066 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
2067 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
2068 !mwait_substates)
2069 return -ENODEV;
2070
2071 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
2072
2073 icpu = (const struct idle_cpu *)id->driver_data;
2074 if (icpu) {
2075 cpuidle_state_table = icpu->state_table;
2076 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
2077 if (icpu->disable_promotion_to_c1e)
2078 c1e_promotion = C1E_PROMOTION_DISABLE;
2079 if (icpu->use_acpi || force_use_acpi)
2080 intel_idle_acpi_cst_extract();
2081 } else if (!intel_idle_acpi_cst_extract()) {
2082 return -ENODEV;
2083 }
2084
2085 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
2086 boot_cpu_data.x86_model);
2087
2088 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
2089 if (!intel_idle_cpuidle_devices)
2090 return -ENOMEM;
2091
2092 intel_idle_cpuidle_driver_init(&intel_idle_driver);
2093
2094 retval = cpuidle_register_driver(&intel_idle_driver);
2095 if (retval) {
2096 struct cpuidle_driver *drv = cpuidle_get_driver();
2097 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
2098 drv ? drv->name : "none");
2099 goto init_driver_fail;
2100 }
2101
2102 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
2103 intel_idle_cpu_online, NULL);
2104 if (retval < 0)
2105 goto hp_setup_fail;
2106
2107 pr_debug("Local APIC timer is reliable in %s\n",
2108 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
2109
2110 return 0;
2111
2112hp_setup_fail:
2113 intel_idle_cpuidle_devices_uninit();
2114 cpuidle_unregister_driver(&intel_idle_driver);
2115init_driver_fail:
2116 free_percpu(intel_idle_cpuidle_devices);
2117 return retval;
2118
2119}
2120device_initcall(intel_idle_init);
2121
2122/*
2123 * We are not really modular, but we used to support that. Meaning we also
2124 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
2125 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
2126 * is the easiest way (currently) to continue doing that.
2127 */
2128module_param(max_cstate, int, 0444);
2129/*
2130 * The positions of the bits that are set in this number are the indices of the
2131 * idle states to be disabled by default (as reflected by the names of the
2132 * corresponding idle state directories in sysfs, "state0", "state1" ...
2133 * "state<i>" ..., where <i> is the index of the given state).
2134 */
2135module_param_named(states_off, disabled_states_mask, uint, 0444);
2136MODULE_PARM_DESC(states_off, "Mask of disabled idle states");
2137/*
2138 * Some platforms come with mutually exclusive C-states, so that if one is
2139 * enabled, the other C-states must not be used. Example: C1 and C1E on
2140 * Sapphire Rapids platform. This parameter allows for selecting the
2141 * preferred C-states among the groups of mutually exclusive C-states - the
2142 * selected C-states will be registered, the other C-states from the mutually
2143 * exclusive group won't be registered. If the platform has no mutually
2144 * exclusive C-states, this parameter has no effect.
2145 */
2146module_param_named(preferred_cstates, preferred_states_mask, uint, 0444);
2147MODULE_PARM_DESC(preferred_cstates, "Mask of preferred idle states");