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/notifier.h>
51#include <linux/cpu.h>
52#include <linux/moduleparam.h>
53#include <asm/cpu_device_id.h>
54#include <asm/intel-family.h>
55#include <asm/mwait.h>
56#include <asm/msr.h>
57
58#define INTEL_IDLE_VERSION "0.5.1"
59
60static struct cpuidle_driver intel_idle_driver = {
61 .name = "intel_idle",
62 .owner = THIS_MODULE,
63};
64/* intel_idle.max_cstate=0 disables driver */
65static int max_cstate = CPUIDLE_STATE_MAX - 1;
66static unsigned int disabled_states_mask;
67
68static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
69
70static unsigned long auto_demotion_disable_flags;
71static bool disable_promotion_to_c1e;
72
73struct idle_cpu {
74 struct cpuidle_state *state_table;
75
76 /*
77 * Hardware C-state auto-demotion may not always be optimal.
78 * Indicate which enable bits to clear here.
79 */
80 unsigned long auto_demotion_disable_flags;
81 bool byt_auto_demotion_disable_flag;
82 bool disable_promotion_to_c1e;
83 bool use_acpi;
84};
85
86static const struct idle_cpu *icpu __initdata;
87static struct cpuidle_state *cpuidle_state_table __initdata;
88
89static unsigned int mwait_substates __initdata;
90
91/*
92 * Enable this state by default even if the ACPI _CST does not list it.
93 */
94#define CPUIDLE_FLAG_ALWAYS_ENABLE BIT(15)
95
96/*
97 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
98 * the C-state (top nibble) and sub-state (bottom nibble)
99 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
100 *
101 * We store the hint at the top of our "flags" for each state.
102 */
103#define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
104#define MWAIT2flg(eax) ((eax & 0xFF) << 24)
105
106/**
107 * intel_idle - Ask the processor to enter the given idle state.
108 * @dev: cpuidle device of the target CPU.
109 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
110 * @index: Target idle state index.
111 *
112 * Use the MWAIT instruction to notify the processor that the CPU represented by
113 * @dev is idle and it can try to enter the idle state corresponding to @index.
114 *
115 * If the local APIC timer is not known to be reliable in the target idle state,
116 * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
117 *
118 * Optionally call leave_mm() for the target CPU upfront to avoid wakeups due to
119 * flushing user TLBs.
120 *
121 * Must be called under local_irq_disable().
122 */
123static __cpuidle int intel_idle(struct cpuidle_device *dev,
124 struct cpuidle_driver *drv, int index)
125{
126 struct cpuidle_state *state = &drv->states[index];
127 unsigned long eax = flg2MWAIT(state->flags);
128 unsigned long ecx = 1; /* break on interrupt flag */
129
130 mwait_idle_with_hints(eax, ecx);
131
132 return index;
133}
134
135/**
136 * intel_idle_s2idle - Ask the processor to enter the given idle state.
137 * @dev: cpuidle device of the target CPU.
138 * @drv: cpuidle driver (assumed to point to intel_idle_driver).
139 * @index: Target idle state index.
140 *
141 * Use the MWAIT instruction to notify the processor that the CPU represented by
142 * @dev is idle and it can try to enter the idle state corresponding to @index.
143 *
144 * Invoked as a suspend-to-idle callback routine with frozen user space, frozen
145 * scheduler tick and suspended scheduler clock on the target CPU.
146 */
147static __cpuidle int intel_idle_s2idle(struct cpuidle_device *dev,
148 struct cpuidle_driver *drv, int index)
149{
150 unsigned long eax = flg2MWAIT(drv->states[index].flags);
151 unsigned long ecx = 1; /* break on interrupt flag */
152
153 mwait_idle_with_hints(eax, ecx);
154
155 return 0;
156}
157
158/*
159 * States are indexed by the cstate number,
160 * which is also the index into the MWAIT hint array.
161 * Thus C0 is a dummy.
162 */
163static struct cpuidle_state nehalem_cstates[] __initdata = {
164 {
165 .name = "C1",
166 .desc = "MWAIT 0x00",
167 .flags = MWAIT2flg(0x00),
168 .exit_latency = 3,
169 .target_residency = 6,
170 .enter = &intel_idle,
171 .enter_s2idle = intel_idle_s2idle, },
172 {
173 .name = "C1E",
174 .desc = "MWAIT 0x01",
175 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
176 .exit_latency = 10,
177 .target_residency = 20,
178 .enter = &intel_idle,
179 .enter_s2idle = intel_idle_s2idle, },
180 {
181 .name = "C3",
182 .desc = "MWAIT 0x10",
183 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
184 .exit_latency = 20,
185 .target_residency = 80,
186 .enter = &intel_idle,
187 .enter_s2idle = intel_idle_s2idle, },
188 {
189 .name = "C6",
190 .desc = "MWAIT 0x20",
191 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
192 .exit_latency = 200,
193 .target_residency = 800,
194 .enter = &intel_idle,
195 .enter_s2idle = intel_idle_s2idle, },
196 {
197 .enter = NULL }
198};
199
200static struct cpuidle_state snb_cstates[] __initdata = {
201 {
202 .name = "C1",
203 .desc = "MWAIT 0x00",
204 .flags = MWAIT2flg(0x00),
205 .exit_latency = 2,
206 .target_residency = 2,
207 .enter = &intel_idle,
208 .enter_s2idle = intel_idle_s2idle, },
209 {
210 .name = "C1E",
211 .desc = "MWAIT 0x01",
212 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
213 .exit_latency = 10,
214 .target_residency = 20,
215 .enter = &intel_idle,
216 .enter_s2idle = intel_idle_s2idle, },
217 {
218 .name = "C3",
219 .desc = "MWAIT 0x10",
220 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
221 .exit_latency = 80,
222 .target_residency = 211,
223 .enter = &intel_idle,
224 .enter_s2idle = intel_idle_s2idle, },
225 {
226 .name = "C6",
227 .desc = "MWAIT 0x20",
228 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
229 .exit_latency = 104,
230 .target_residency = 345,
231 .enter = &intel_idle,
232 .enter_s2idle = intel_idle_s2idle, },
233 {
234 .name = "C7",
235 .desc = "MWAIT 0x30",
236 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
237 .exit_latency = 109,
238 .target_residency = 345,
239 .enter = &intel_idle,
240 .enter_s2idle = intel_idle_s2idle, },
241 {
242 .enter = NULL }
243};
244
245static struct cpuidle_state byt_cstates[] __initdata = {
246 {
247 .name = "C1",
248 .desc = "MWAIT 0x00",
249 .flags = MWAIT2flg(0x00),
250 .exit_latency = 1,
251 .target_residency = 1,
252 .enter = &intel_idle,
253 .enter_s2idle = intel_idle_s2idle, },
254 {
255 .name = "C6N",
256 .desc = "MWAIT 0x58",
257 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
258 .exit_latency = 300,
259 .target_residency = 275,
260 .enter = &intel_idle,
261 .enter_s2idle = intel_idle_s2idle, },
262 {
263 .name = "C6S",
264 .desc = "MWAIT 0x52",
265 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
266 .exit_latency = 500,
267 .target_residency = 560,
268 .enter = &intel_idle,
269 .enter_s2idle = intel_idle_s2idle, },
270 {
271 .name = "C7",
272 .desc = "MWAIT 0x60",
273 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
274 .exit_latency = 1200,
275 .target_residency = 4000,
276 .enter = &intel_idle,
277 .enter_s2idle = intel_idle_s2idle, },
278 {
279 .name = "C7S",
280 .desc = "MWAIT 0x64",
281 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
282 .exit_latency = 10000,
283 .target_residency = 20000,
284 .enter = &intel_idle,
285 .enter_s2idle = intel_idle_s2idle, },
286 {
287 .enter = NULL }
288};
289
290static struct cpuidle_state cht_cstates[] __initdata = {
291 {
292 .name = "C1",
293 .desc = "MWAIT 0x00",
294 .flags = MWAIT2flg(0x00),
295 .exit_latency = 1,
296 .target_residency = 1,
297 .enter = &intel_idle,
298 .enter_s2idle = intel_idle_s2idle, },
299 {
300 .name = "C6N",
301 .desc = "MWAIT 0x58",
302 .flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
303 .exit_latency = 80,
304 .target_residency = 275,
305 .enter = &intel_idle,
306 .enter_s2idle = intel_idle_s2idle, },
307 {
308 .name = "C6S",
309 .desc = "MWAIT 0x52",
310 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
311 .exit_latency = 200,
312 .target_residency = 560,
313 .enter = &intel_idle,
314 .enter_s2idle = intel_idle_s2idle, },
315 {
316 .name = "C7",
317 .desc = "MWAIT 0x60",
318 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
319 .exit_latency = 1200,
320 .target_residency = 4000,
321 .enter = &intel_idle,
322 .enter_s2idle = intel_idle_s2idle, },
323 {
324 .name = "C7S",
325 .desc = "MWAIT 0x64",
326 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
327 .exit_latency = 10000,
328 .target_residency = 20000,
329 .enter = &intel_idle,
330 .enter_s2idle = intel_idle_s2idle, },
331 {
332 .enter = NULL }
333};
334
335static struct cpuidle_state ivb_cstates[] __initdata = {
336 {
337 .name = "C1",
338 .desc = "MWAIT 0x00",
339 .flags = MWAIT2flg(0x00),
340 .exit_latency = 1,
341 .target_residency = 1,
342 .enter = &intel_idle,
343 .enter_s2idle = intel_idle_s2idle, },
344 {
345 .name = "C1E",
346 .desc = "MWAIT 0x01",
347 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
348 .exit_latency = 10,
349 .target_residency = 20,
350 .enter = &intel_idle,
351 .enter_s2idle = intel_idle_s2idle, },
352 {
353 .name = "C3",
354 .desc = "MWAIT 0x10",
355 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
356 .exit_latency = 59,
357 .target_residency = 156,
358 .enter = &intel_idle,
359 .enter_s2idle = intel_idle_s2idle, },
360 {
361 .name = "C6",
362 .desc = "MWAIT 0x20",
363 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
364 .exit_latency = 80,
365 .target_residency = 300,
366 .enter = &intel_idle,
367 .enter_s2idle = intel_idle_s2idle, },
368 {
369 .name = "C7",
370 .desc = "MWAIT 0x30",
371 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
372 .exit_latency = 87,
373 .target_residency = 300,
374 .enter = &intel_idle,
375 .enter_s2idle = intel_idle_s2idle, },
376 {
377 .enter = NULL }
378};
379
380static struct cpuidle_state ivt_cstates[] __initdata = {
381 {
382 .name = "C1",
383 .desc = "MWAIT 0x00",
384 .flags = MWAIT2flg(0x00),
385 .exit_latency = 1,
386 .target_residency = 1,
387 .enter = &intel_idle,
388 .enter_s2idle = intel_idle_s2idle, },
389 {
390 .name = "C1E",
391 .desc = "MWAIT 0x01",
392 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
393 .exit_latency = 10,
394 .target_residency = 80,
395 .enter = &intel_idle,
396 .enter_s2idle = intel_idle_s2idle, },
397 {
398 .name = "C3",
399 .desc = "MWAIT 0x10",
400 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
401 .exit_latency = 59,
402 .target_residency = 156,
403 .enter = &intel_idle,
404 .enter_s2idle = intel_idle_s2idle, },
405 {
406 .name = "C6",
407 .desc = "MWAIT 0x20",
408 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
409 .exit_latency = 82,
410 .target_residency = 300,
411 .enter = &intel_idle,
412 .enter_s2idle = intel_idle_s2idle, },
413 {
414 .enter = NULL }
415};
416
417static struct cpuidle_state ivt_cstates_4s[] __initdata = {
418 {
419 .name = "C1",
420 .desc = "MWAIT 0x00",
421 .flags = MWAIT2flg(0x00),
422 .exit_latency = 1,
423 .target_residency = 1,
424 .enter = &intel_idle,
425 .enter_s2idle = intel_idle_s2idle, },
426 {
427 .name = "C1E",
428 .desc = "MWAIT 0x01",
429 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
430 .exit_latency = 10,
431 .target_residency = 250,
432 .enter = &intel_idle,
433 .enter_s2idle = intel_idle_s2idle, },
434 {
435 .name = "C3",
436 .desc = "MWAIT 0x10",
437 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
438 .exit_latency = 59,
439 .target_residency = 300,
440 .enter = &intel_idle,
441 .enter_s2idle = intel_idle_s2idle, },
442 {
443 .name = "C6",
444 .desc = "MWAIT 0x20",
445 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
446 .exit_latency = 84,
447 .target_residency = 400,
448 .enter = &intel_idle,
449 .enter_s2idle = intel_idle_s2idle, },
450 {
451 .enter = NULL }
452};
453
454static struct cpuidle_state ivt_cstates_8s[] __initdata = {
455 {
456 .name = "C1",
457 .desc = "MWAIT 0x00",
458 .flags = MWAIT2flg(0x00),
459 .exit_latency = 1,
460 .target_residency = 1,
461 .enter = &intel_idle,
462 .enter_s2idle = intel_idle_s2idle, },
463 {
464 .name = "C1E",
465 .desc = "MWAIT 0x01",
466 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
467 .exit_latency = 10,
468 .target_residency = 500,
469 .enter = &intel_idle,
470 .enter_s2idle = intel_idle_s2idle, },
471 {
472 .name = "C3",
473 .desc = "MWAIT 0x10",
474 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
475 .exit_latency = 59,
476 .target_residency = 600,
477 .enter = &intel_idle,
478 .enter_s2idle = intel_idle_s2idle, },
479 {
480 .name = "C6",
481 .desc = "MWAIT 0x20",
482 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
483 .exit_latency = 88,
484 .target_residency = 700,
485 .enter = &intel_idle,
486 .enter_s2idle = intel_idle_s2idle, },
487 {
488 .enter = NULL }
489};
490
491static struct cpuidle_state hsw_cstates[] __initdata = {
492 {
493 .name = "C1",
494 .desc = "MWAIT 0x00",
495 .flags = MWAIT2flg(0x00),
496 .exit_latency = 2,
497 .target_residency = 2,
498 .enter = &intel_idle,
499 .enter_s2idle = intel_idle_s2idle, },
500 {
501 .name = "C1E",
502 .desc = "MWAIT 0x01",
503 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
504 .exit_latency = 10,
505 .target_residency = 20,
506 .enter = &intel_idle,
507 .enter_s2idle = intel_idle_s2idle, },
508 {
509 .name = "C3",
510 .desc = "MWAIT 0x10",
511 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
512 .exit_latency = 33,
513 .target_residency = 100,
514 .enter = &intel_idle,
515 .enter_s2idle = intel_idle_s2idle, },
516 {
517 .name = "C6",
518 .desc = "MWAIT 0x20",
519 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
520 .exit_latency = 133,
521 .target_residency = 400,
522 .enter = &intel_idle,
523 .enter_s2idle = intel_idle_s2idle, },
524 {
525 .name = "C7s",
526 .desc = "MWAIT 0x32",
527 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
528 .exit_latency = 166,
529 .target_residency = 500,
530 .enter = &intel_idle,
531 .enter_s2idle = intel_idle_s2idle, },
532 {
533 .name = "C8",
534 .desc = "MWAIT 0x40",
535 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
536 .exit_latency = 300,
537 .target_residency = 900,
538 .enter = &intel_idle,
539 .enter_s2idle = intel_idle_s2idle, },
540 {
541 .name = "C9",
542 .desc = "MWAIT 0x50",
543 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
544 .exit_latency = 600,
545 .target_residency = 1800,
546 .enter = &intel_idle,
547 .enter_s2idle = intel_idle_s2idle, },
548 {
549 .name = "C10",
550 .desc = "MWAIT 0x60",
551 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
552 .exit_latency = 2600,
553 .target_residency = 7700,
554 .enter = &intel_idle,
555 .enter_s2idle = intel_idle_s2idle, },
556 {
557 .enter = NULL }
558};
559static struct cpuidle_state bdw_cstates[] __initdata = {
560 {
561 .name = "C1",
562 .desc = "MWAIT 0x00",
563 .flags = MWAIT2flg(0x00),
564 .exit_latency = 2,
565 .target_residency = 2,
566 .enter = &intel_idle,
567 .enter_s2idle = intel_idle_s2idle, },
568 {
569 .name = "C1E",
570 .desc = "MWAIT 0x01",
571 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
572 .exit_latency = 10,
573 .target_residency = 20,
574 .enter = &intel_idle,
575 .enter_s2idle = intel_idle_s2idle, },
576 {
577 .name = "C3",
578 .desc = "MWAIT 0x10",
579 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
580 .exit_latency = 40,
581 .target_residency = 100,
582 .enter = &intel_idle,
583 .enter_s2idle = intel_idle_s2idle, },
584 {
585 .name = "C6",
586 .desc = "MWAIT 0x20",
587 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
588 .exit_latency = 133,
589 .target_residency = 400,
590 .enter = &intel_idle,
591 .enter_s2idle = intel_idle_s2idle, },
592 {
593 .name = "C7s",
594 .desc = "MWAIT 0x32",
595 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
596 .exit_latency = 166,
597 .target_residency = 500,
598 .enter = &intel_idle,
599 .enter_s2idle = intel_idle_s2idle, },
600 {
601 .name = "C8",
602 .desc = "MWAIT 0x40",
603 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
604 .exit_latency = 300,
605 .target_residency = 900,
606 .enter = &intel_idle,
607 .enter_s2idle = intel_idle_s2idle, },
608 {
609 .name = "C9",
610 .desc = "MWAIT 0x50",
611 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
612 .exit_latency = 600,
613 .target_residency = 1800,
614 .enter = &intel_idle,
615 .enter_s2idle = intel_idle_s2idle, },
616 {
617 .name = "C10",
618 .desc = "MWAIT 0x60",
619 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
620 .exit_latency = 2600,
621 .target_residency = 7700,
622 .enter = &intel_idle,
623 .enter_s2idle = intel_idle_s2idle, },
624 {
625 .enter = NULL }
626};
627
628static struct cpuidle_state skl_cstates[] __initdata = {
629 {
630 .name = "C1",
631 .desc = "MWAIT 0x00",
632 .flags = MWAIT2flg(0x00),
633 .exit_latency = 2,
634 .target_residency = 2,
635 .enter = &intel_idle,
636 .enter_s2idle = intel_idle_s2idle, },
637 {
638 .name = "C1E",
639 .desc = "MWAIT 0x01",
640 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
641 .exit_latency = 10,
642 .target_residency = 20,
643 .enter = &intel_idle,
644 .enter_s2idle = intel_idle_s2idle, },
645 {
646 .name = "C3",
647 .desc = "MWAIT 0x10",
648 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
649 .exit_latency = 70,
650 .target_residency = 100,
651 .enter = &intel_idle,
652 .enter_s2idle = intel_idle_s2idle, },
653 {
654 .name = "C6",
655 .desc = "MWAIT 0x20",
656 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
657 .exit_latency = 85,
658 .target_residency = 200,
659 .enter = &intel_idle,
660 .enter_s2idle = intel_idle_s2idle, },
661 {
662 .name = "C7s",
663 .desc = "MWAIT 0x33",
664 .flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED,
665 .exit_latency = 124,
666 .target_residency = 800,
667 .enter = &intel_idle,
668 .enter_s2idle = intel_idle_s2idle, },
669 {
670 .name = "C8",
671 .desc = "MWAIT 0x40",
672 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
673 .exit_latency = 200,
674 .target_residency = 800,
675 .enter = &intel_idle,
676 .enter_s2idle = intel_idle_s2idle, },
677 {
678 .name = "C9",
679 .desc = "MWAIT 0x50",
680 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
681 .exit_latency = 480,
682 .target_residency = 5000,
683 .enter = &intel_idle,
684 .enter_s2idle = intel_idle_s2idle, },
685 {
686 .name = "C10",
687 .desc = "MWAIT 0x60",
688 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
689 .exit_latency = 890,
690 .target_residency = 5000,
691 .enter = &intel_idle,
692 .enter_s2idle = intel_idle_s2idle, },
693 {
694 .enter = NULL }
695};
696
697static struct cpuidle_state skx_cstates[] __initdata = {
698 {
699 .name = "C1",
700 .desc = "MWAIT 0x00",
701 .flags = MWAIT2flg(0x00),
702 .exit_latency = 2,
703 .target_residency = 2,
704 .enter = &intel_idle,
705 .enter_s2idle = intel_idle_s2idle, },
706 {
707 .name = "C1E",
708 .desc = "MWAIT 0x01",
709 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
710 .exit_latency = 10,
711 .target_residency = 20,
712 .enter = &intel_idle,
713 .enter_s2idle = intel_idle_s2idle, },
714 {
715 .name = "C6",
716 .desc = "MWAIT 0x20",
717 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
718 .exit_latency = 133,
719 .target_residency = 600,
720 .enter = &intel_idle,
721 .enter_s2idle = intel_idle_s2idle, },
722 {
723 .enter = NULL }
724};
725
726static struct cpuidle_state icx_cstates[] __initdata = {
727 {
728 .name = "C1",
729 .desc = "MWAIT 0x00",
730 .flags = MWAIT2flg(0x00),
731 .exit_latency = 1,
732 .target_residency = 1,
733 .enter = &intel_idle,
734 .enter_s2idle = intel_idle_s2idle, },
735 {
736 .name = "C1E",
737 .desc = "MWAIT 0x01",
738 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
739 .exit_latency = 4,
740 .target_residency = 4,
741 .enter = &intel_idle,
742 .enter_s2idle = intel_idle_s2idle, },
743 {
744 .name = "C6",
745 .desc = "MWAIT 0x20",
746 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
747 .exit_latency = 170,
748 .target_residency = 600,
749 .enter = &intel_idle,
750 .enter_s2idle = intel_idle_s2idle, },
751 {
752 .enter = NULL }
753};
754
755static struct cpuidle_state atom_cstates[] __initdata = {
756 {
757 .name = "C1E",
758 .desc = "MWAIT 0x00",
759 .flags = MWAIT2flg(0x00),
760 .exit_latency = 10,
761 .target_residency = 20,
762 .enter = &intel_idle,
763 .enter_s2idle = intel_idle_s2idle, },
764 {
765 .name = "C2",
766 .desc = "MWAIT 0x10",
767 .flags = MWAIT2flg(0x10),
768 .exit_latency = 20,
769 .target_residency = 80,
770 .enter = &intel_idle,
771 .enter_s2idle = intel_idle_s2idle, },
772 {
773 .name = "C4",
774 .desc = "MWAIT 0x30",
775 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
776 .exit_latency = 100,
777 .target_residency = 400,
778 .enter = &intel_idle,
779 .enter_s2idle = intel_idle_s2idle, },
780 {
781 .name = "C6",
782 .desc = "MWAIT 0x52",
783 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
784 .exit_latency = 140,
785 .target_residency = 560,
786 .enter = &intel_idle,
787 .enter_s2idle = intel_idle_s2idle, },
788 {
789 .enter = NULL }
790};
791static struct cpuidle_state tangier_cstates[] __initdata = {
792 {
793 .name = "C1",
794 .desc = "MWAIT 0x00",
795 .flags = MWAIT2flg(0x00),
796 .exit_latency = 1,
797 .target_residency = 4,
798 .enter = &intel_idle,
799 .enter_s2idle = intel_idle_s2idle, },
800 {
801 .name = "C4",
802 .desc = "MWAIT 0x30",
803 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
804 .exit_latency = 100,
805 .target_residency = 400,
806 .enter = &intel_idle,
807 .enter_s2idle = intel_idle_s2idle, },
808 {
809 .name = "C6",
810 .desc = "MWAIT 0x52",
811 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
812 .exit_latency = 140,
813 .target_residency = 560,
814 .enter = &intel_idle,
815 .enter_s2idle = intel_idle_s2idle, },
816 {
817 .name = "C7",
818 .desc = "MWAIT 0x60",
819 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
820 .exit_latency = 1200,
821 .target_residency = 4000,
822 .enter = &intel_idle,
823 .enter_s2idle = intel_idle_s2idle, },
824 {
825 .name = "C9",
826 .desc = "MWAIT 0x64",
827 .flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
828 .exit_latency = 10000,
829 .target_residency = 20000,
830 .enter = &intel_idle,
831 .enter_s2idle = intel_idle_s2idle, },
832 {
833 .enter = NULL }
834};
835static struct cpuidle_state avn_cstates[] __initdata = {
836 {
837 .name = "C1",
838 .desc = "MWAIT 0x00",
839 .flags = MWAIT2flg(0x00),
840 .exit_latency = 2,
841 .target_residency = 2,
842 .enter = &intel_idle,
843 .enter_s2idle = intel_idle_s2idle, },
844 {
845 .name = "C6",
846 .desc = "MWAIT 0x51",
847 .flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
848 .exit_latency = 15,
849 .target_residency = 45,
850 .enter = &intel_idle,
851 .enter_s2idle = intel_idle_s2idle, },
852 {
853 .enter = NULL }
854};
855static struct cpuidle_state knl_cstates[] __initdata = {
856 {
857 .name = "C1",
858 .desc = "MWAIT 0x00",
859 .flags = MWAIT2flg(0x00),
860 .exit_latency = 1,
861 .target_residency = 2,
862 .enter = &intel_idle,
863 .enter_s2idle = intel_idle_s2idle },
864 {
865 .name = "C6",
866 .desc = "MWAIT 0x10",
867 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
868 .exit_latency = 120,
869 .target_residency = 500,
870 .enter = &intel_idle,
871 .enter_s2idle = intel_idle_s2idle },
872 {
873 .enter = NULL }
874};
875
876static struct cpuidle_state bxt_cstates[] __initdata = {
877 {
878 .name = "C1",
879 .desc = "MWAIT 0x00",
880 .flags = MWAIT2flg(0x00),
881 .exit_latency = 2,
882 .target_residency = 2,
883 .enter = &intel_idle,
884 .enter_s2idle = intel_idle_s2idle, },
885 {
886 .name = "C1E",
887 .desc = "MWAIT 0x01",
888 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
889 .exit_latency = 10,
890 .target_residency = 20,
891 .enter = &intel_idle,
892 .enter_s2idle = intel_idle_s2idle, },
893 {
894 .name = "C6",
895 .desc = "MWAIT 0x20",
896 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
897 .exit_latency = 133,
898 .target_residency = 133,
899 .enter = &intel_idle,
900 .enter_s2idle = intel_idle_s2idle, },
901 {
902 .name = "C7s",
903 .desc = "MWAIT 0x31",
904 .flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
905 .exit_latency = 155,
906 .target_residency = 155,
907 .enter = &intel_idle,
908 .enter_s2idle = intel_idle_s2idle, },
909 {
910 .name = "C8",
911 .desc = "MWAIT 0x40",
912 .flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
913 .exit_latency = 1000,
914 .target_residency = 1000,
915 .enter = &intel_idle,
916 .enter_s2idle = intel_idle_s2idle, },
917 {
918 .name = "C9",
919 .desc = "MWAIT 0x50",
920 .flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
921 .exit_latency = 2000,
922 .target_residency = 2000,
923 .enter = &intel_idle,
924 .enter_s2idle = intel_idle_s2idle, },
925 {
926 .name = "C10",
927 .desc = "MWAIT 0x60",
928 .flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
929 .exit_latency = 10000,
930 .target_residency = 10000,
931 .enter = &intel_idle,
932 .enter_s2idle = intel_idle_s2idle, },
933 {
934 .enter = NULL }
935};
936
937static struct cpuidle_state dnv_cstates[] __initdata = {
938 {
939 .name = "C1",
940 .desc = "MWAIT 0x00",
941 .flags = MWAIT2flg(0x00),
942 .exit_latency = 2,
943 .target_residency = 2,
944 .enter = &intel_idle,
945 .enter_s2idle = intel_idle_s2idle, },
946 {
947 .name = "C1E",
948 .desc = "MWAIT 0x01",
949 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
950 .exit_latency = 10,
951 .target_residency = 20,
952 .enter = &intel_idle,
953 .enter_s2idle = intel_idle_s2idle, },
954 {
955 .name = "C6",
956 .desc = "MWAIT 0x20",
957 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
958 .exit_latency = 50,
959 .target_residency = 500,
960 .enter = &intel_idle,
961 .enter_s2idle = intel_idle_s2idle, },
962 {
963 .enter = NULL }
964};
965
966/*
967 * Note, depending on HW and FW revision, SnowRidge SoC may or may not support
968 * C6, and this is indicated in the CPUID mwait leaf.
969 */
970static struct cpuidle_state snr_cstates[] __initdata = {
971 {
972 .name = "C1",
973 .desc = "MWAIT 0x00",
974 .flags = MWAIT2flg(0x00),
975 .exit_latency = 2,
976 .target_residency = 2,
977 .enter = &intel_idle,
978 .enter_s2idle = intel_idle_s2idle, },
979 {
980 .name = "C1E",
981 .desc = "MWAIT 0x01",
982 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_ALWAYS_ENABLE,
983 .exit_latency = 15,
984 .target_residency = 25,
985 .enter = &intel_idle,
986 .enter_s2idle = intel_idle_s2idle, },
987 {
988 .name = "C6",
989 .desc = "MWAIT 0x20",
990 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
991 .exit_latency = 130,
992 .target_residency = 500,
993 .enter = &intel_idle,
994 .enter_s2idle = intel_idle_s2idle, },
995 {
996 .enter = NULL }
997};
998
999static const struct idle_cpu idle_cpu_nehalem __initconst = {
1000 .state_table = nehalem_cstates,
1001 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1002 .disable_promotion_to_c1e = true,
1003};
1004
1005static const struct idle_cpu idle_cpu_nhx __initconst = {
1006 .state_table = nehalem_cstates,
1007 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
1008 .disable_promotion_to_c1e = true,
1009 .use_acpi = true,
1010};
1011
1012static const struct idle_cpu idle_cpu_atom __initconst = {
1013 .state_table = atom_cstates,
1014};
1015
1016static const struct idle_cpu idle_cpu_tangier __initconst = {
1017 .state_table = tangier_cstates,
1018};
1019
1020static const struct idle_cpu idle_cpu_lincroft __initconst = {
1021 .state_table = atom_cstates,
1022 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1023};
1024
1025static const struct idle_cpu idle_cpu_snb __initconst = {
1026 .state_table = snb_cstates,
1027 .disable_promotion_to_c1e = true,
1028};
1029
1030static const struct idle_cpu idle_cpu_snx __initconst = {
1031 .state_table = snb_cstates,
1032 .disable_promotion_to_c1e = true,
1033 .use_acpi = true,
1034};
1035
1036static const struct idle_cpu idle_cpu_byt __initconst = {
1037 .state_table = byt_cstates,
1038 .disable_promotion_to_c1e = true,
1039 .byt_auto_demotion_disable_flag = true,
1040};
1041
1042static const struct idle_cpu idle_cpu_cht __initconst = {
1043 .state_table = cht_cstates,
1044 .disable_promotion_to_c1e = true,
1045 .byt_auto_demotion_disable_flag = true,
1046};
1047
1048static const struct idle_cpu idle_cpu_ivb __initconst = {
1049 .state_table = ivb_cstates,
1050 .disable_promotion_to_c1e = true,
1051};
1052
1053static const struct idle_cpu idle_cpu_ivt __initconst = {
1054 .state_table = ivt_cstates,
1055 .disable_promotion_to_c1e = true,
1056 .use_acpi = true,
1057};
1058
1059static const struct idle_cpu idle_cpu_hsw __initconst = {
1060 .state_table = hsw_cstates,
1061 .disable_promotion_to_c1e = true,
1062};
1063
1064static const struct idle_cpu idle_cpu_hsx __initconst = {
1065 .state_table = hsw_cstates,
1066 .disable_promotion_to_c1e = true,
1067 .use_acpi = true,
1068};
1069
1070static const struct idle_cpu idle_cpu_bdw __initconst = {
1071 .state_table = bdw_cstates,
1072 .disable_promotion_to_c1e = true,
1073};
1074
1075static const struct idle_cpu idle_cpu_bdx __initconst = {
1076 .state_table = bdw_cstates,
1077 .disable_promotion_to_c1e = true,
1078 .use_acpi = true,
1079};
1080
1081static const struct idle_cpu idle_cpu_skl __initconst = {
1082 .state_table = skl_cstates,
1083 .disable_promotion_to_c1e = true,
1084};
1085
1086static const struct idle_cpu idle_cpu_skx __initconst = {
1087 .state_table = skx_cstates,
1088 .disable_promotion_to_c1e = true,
1089 .use_acpi = true,
1090};
1091
1092static const struct idle_cpu idle_cpu_icx __initconst = {
1093 .state_table = icx_cstates,
1094 .disable_promotion_to_c1e = true,
1095 .use_acpi = true,
1096};
1097
1098static const struct idle_cpu idle_cpu_avn __initconst = {
1099 .state_table = avn_cstates,
1100 .disable_promotion_to_c1e = true,
1101 .use_acpi = true,
1102};
1103
1104static const struct idle_cpu idle_cpu_knl __initconst = {
1105 .state_table = knl_cstates,
1106 .use_acpi = true,
1107};
1108
1109static const struct idle_cpu idle_cpu_bxt __initconst = {
1110 .state_table = bxt_cstates,
1111 .disable_promotion_to_c1e = true,
1112};
1113
1114static const struct idle_cpu idle_cpu_dnv __initconst = {
1115 .state_table = dnv_cstates,
1116 .disable_promotion_to_c1e = true,
1117 .use_acpi = true,
1118};
1119
1120static const struct idle_cpu idle_cpu_snr __initconst = {
1121 .state_table = snr_cstates,
1122 .disable_promotion_to_c1e = true,
1123 .use_acpi = true,
1124};
1125
1126static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1127 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EP, &idle_cpu_nhx),
1128 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM, &idle_cpu_nehalem),
1129 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_G, &idle_cpu_nehalem),
1130 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE, &idle_cpu_nehalem),
1131 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EP, &idle_cpu_nhx),
1132 X86_MATCH_INTEL_FAM6_MODEL(NEHALEM_EX, &idle_cpu_nhx),
1133 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL, &idle_cpu_atom),
1134 X86_MATCH_INTEL_FAM6_MODEL(ATOM_BONNELL_MID, &idle_cpu_lincroft),
1135 X86_MATCH_INTEL_FAM6_MODEL(WESTMERE_EX, &idle_cpu_nhx),
1136 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE, &idle_cpu_snb),
1137 X86_MATCH_INTEL_FAM6_MODEL(SANDYBRIDGE_X, &idle_cpu_snx),
1138 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SALTWELL, &idle_cpu_atom),
1139 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT, &idle_cpu_byt),
1140 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_MID, &idle_cpu_tangier),
1141 X86_MATCH_INTEL_FAM6_MODEL(ATOM_AIRMONT, &idle_cpu_cht),
1142 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE, &idle_cpu_ivb),
1143 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &idle_cpu_ivt),
1144 X86_MATCH_INTEL_FAM6_MODEL(HASWELL, &idle_cpu_hsw),
1145 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &idle_cpu_hsx),
1146 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_L, &idle_cpu_hsw),
1147 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_G, &idle_cpu_hsw),
1148 X86_MATCH_INTEL_FAM6_MODEL(ATOM_SILVERMONT_D, &idle_cpu_avn),
1149 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL, &idle_cpu_bdw),
1150 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_G, &idle_cpu_bdw),
1151 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &idle_cpu_bdx),
1152 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &idle_cpu_bdx),
1153 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_L, &idle_cpu_skl),
1154 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE, &idle_cpu_skl),
1155 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE_L, &idle_cpu_skl),
1156 X86_MATCH_INTEL_FAM6_MODEL(KABYLAKE, &idle_cpu_skl),
1157 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &idle_cpu_skx),
1158 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &idle_cpu_icx),
1159 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &idle_cpu_icx),
1160 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &idle_cpu_knl),
1161 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &idle_cpu_knl),
1162 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT, &idle_cpu_bxt),
1163 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_PLUS, &idle_cpu_bxt),
1164 X86_MATCH_INTEL_FAM6_MODEL(ATOM_GOLDMONT_D, &idle_cpu_dnv),
1165 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, &idle_cpu_snr),
1166 {}
1167};
1168
1169static const struct x86_cpu_id intel_mwait_ids[] __initconst = {
1170 X86_MATCH_VENDOR_FAM_FEATURE(INTEL, 6, X86_FEATURE_MWAIT, NULL),
1171 {}
1172};
1173
1174static bool __init intel_idle_max_cstate_reached(int cstate)
1175{
1176 if (cstate + 1 > max_cstate) {
1177 pr_info("max_cstate %d reached\n", max_cstate);
1178 return true;
1179 }
1180 return false;
1181}
1182
1183static bool __init intel_idle_state_needs_timer_stop(struct cpuidle_state *state)
1184{
1185 unsigned long eax = flg2MWAIT(state->flags);
1186
1187 if (boot_cpu_has(X86_FEATURE_ARAT))
1188 return false;
1189
1190 /*
1191 * Switch over to one-shot tick broadcast if the target C-state
1192 * is deeper than C1.
1193 */
1194 return !!((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK);
1195}
1196
1197#ifdef CONFIG_ACPI_PROCESSOR_CSTATE
1198#include <acpi/processor.h>
1199
1200static bool no_acpi __read_mostly;
1201module_param(no_acpi, bool, 0444);
1202MODULE_PARM_DESC(no_acpi, "Do not use ACPI _CST for building the idle states list");
1203
1204static bool force_use_acpi __read_mostly; /* No effect if no_acpi is set. */
1205module_param_named(use_acpi, force_use_acpi, bool, 0444);
1206MODULE_PARM_DESC(use_acpi, "Use ACPI _CST for building the idle states list");
1207
1208static struct acpi_processor_power acpi_state_table __initdata;
1209
1210/**
1211 * intel_idle_cst_usable - Check if the _CST information can be used.
1212 *
1213 * Check if all of the C-states listed by _CST in the max_cstate range are
1214 * ACPI_CSTATE_FFH, which means that they should be entered via MWAIT.
1215 */
1216static bool __init intel_idle_cst_usable(void)
1217{
1218 int cstate, limit;
1219
1220 limit = min_t(int, min_t(int, CPUIDLE_STATE_MAX, max_cstate + 1),
1221 acpi_state_table.count);
1222
1223 for (cstate = 1; cstate < limit; cstate++) {
1224 struct acpi_processor_cx *cx = &acpi_state_table.states[cstate];
1225
1226 if (cx->entry_method != ACPI_CSTATE_FFH)
1227 return false;
1228 }
1229
1230 return true;
1231}
1232
1233static bool __init intel_idle_acpi_cst_extract(void)
1234{
1235 unsigned int cpu;
1236
1237 if (no_acpi) {
1238 pr_debug("Not allowed to use ACPI _CST\n");
1239 return false;
1240 }
1241
1242 for_each_possible_cpu(cpu) {
1243 struct acpi_processor *pr = per_cpu(processors, cpu);
1244
1245 if (!pr)
1246 continue;
1247
1248 if (acpi_processor_evaluate_cst(pr->handle, cpu, &acpi_state_table))
1249 continue;
1250
1251 acpi_state_table.count++;
1252
1253 if (!intel_idle_cst_usable())
1254 continue;
1255
1256 if (!acpi_processor_claim_cst_control())
1257 break;
1258
1259 return true;
1260 }
1261
1262 acpi_state_table.count = 0;
1263 pr_debug("ACPI _CST not found or not usable\n");
1264 return false;
1265}
1266
1267static void __init intel_idle_init_cstates_acpi(struct cpuidle_driver *drv)
1268{
1269 int cstate, limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1270
1271 /*
1272 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1273 * the interesting states are ACPI_CSTATE_FFH.
1274 */
1275 for (cstate = 1; cstate < limit; cstate++) {
1276 struct acpi_processor_cx *cx;
1277 struct cpuidle_state *state;
1278
1279 if (intel_idle_max_cstate_reached(cstate - 1))
1280 break;
1281
1282 cx = &acpi_state_table.states[cstate];
1283
1284 state = &drv->states[drv->state_count++];
1285
1286 snprintf(state->name, CPUIDLE_NAME_LEN, "C%d_ACPI", cstate);
1287 strlcpy(state->desc, cx->desc, CPUIDLE_DESC_LEN);
1288 state->exit_latency = cx->latency;
1289 /*
1290 * For C1-type C-states use the same number for both the exit
1291 * latency and target residency, because that is the case for
1292 * C1 in the majority of the static C-states tables above.
1293 * For the other types of C-states, however, set the target
1294 * residency to 3 times the exit latency which should lead to
1295 * a reasonable balance between energy-efficiency and
1296 * performance in the majority of interesting cases.
1297 */
1298 state->target_residency = cx->latency;
1299 if (cx->type > ACPI_STATE_C1)
1300 state->target_residency *= 3;
1301
1302 state->flags = MWAIT2flg(cx->address);
1303 if (cx->type > ACPI_STATE_C2)
1304 state->flags |= CPUIDLE_FLAG_TLB_FLUSHED;
1305
1306 if (disabled_states_mask & BIT(cstate))
1307 state->flags |= CPUIDLE_FLAG_OFF;
1308
1309 if (intel_idle_state_needs_timer_stop(state))
1310 state->flags |= CPUIDLE_FLAG_TIMER_STOP;
1311
1312 state->enter = intel_idle;
1313 state->enter_s2idle = intel_idle_s2idle;
1314 }
1315}
1316
1317static bool __init intel_idle_off_by_default(u32 mwait_hint)
1318{
1319 int cstate, limit;
1320
1321 /*
1322 * If there are no _CST C-states, do not disable any C-states by
1323 * default.
1324 */
1325 if (!acpi_state_table.count)
1326 return false;
1327
1328 limit = min_t(int, CPUIDLE_STATE_MAX, acpi_state_table.count);
1329 /*
1330 * If limit > 0, intel_idle_cst_usable() has returned 'true', so all of
1331 * the interesting states are ACPI_CSTATE_FFH.
1332 */
1333 for (cstate = 1; cstate < limit; cstate++) {
1334 if (acpi_state_table.states[cstate].address == mwait_hint)
1335 return false;
1336 }
1337 return true;
1338}
1339#else /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1340#define force_use_acpi (false)
1341
1342static inline bool intel_idle_acpi_cst_extract(void) { return false; }
1343static inline void intel_idle_init_cstates_acpi(struct cpuidle_driver *drv) { }
1344static inline bool intel_idle_off_by_default(u32 mwait_hint) { return false; }
1345#endif /* !CONFIG_ACPI_PROCESSOR_CSTATE */
1346
1347/**
1348 * ivt_idle_state_table_update - Tune the idle states table for Ivy Town.
1349 *
1350 * Tune IVT multi-socket targets.
1351 * Assumption: num_sockets == (max_package_num + 1).
1352 */
1353static void __init ivt_idle_state_table_update(void)
1354{
1355 /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1356 int cpu, package_num, num_sockets = 1;
1357
1358 for_each_online_cpu(cpu) {
1359 package_num = topology_physical_package_id(cpu);
1360 if (package_num + 1 > num_sockets) {
1361 num_sockets = package_num + 1;
1362
1363 if (num_sockets > 4) {
1364 cpuidle_state_table = ivt_cstates_8s;
1365 return;
1366 }
1367 }
1368 }
1369
1370 if (num_sockets > 2)
1371 cpuidle_state_table = ivt_cstates_4s;
1372
1373 /* else, 1 and 2 socket systems use default ivt_cstates */
1374}
1375
1376/**
1377 * irtl_2_usec - IRTL to microseconds conversion.
1378 * @irtl: IRTL MSR value.
1379 *
1380 * Translate the IRTL (Interrupt Response Time Limit) MSR value to microseconds.
1381 */
1382static unsigned long long __init irtl_2_usec(unsigned long long irtl)
1383{
1384 static const unsigned int irtl_ns_units[] __initconst = {
1385 1, 32, 1024, 32768, 1048576, 33554432, 0, 0
1386 };
1387 unsigned long long ns;
1388
1389 if (!irtl)
1390 return 0;
1391
1392 ns = irtl_ns_units[(irtl >> 10) & 0x7];
1393
1394 return div_u64((irtl & 0x3FF) * ns, NSEC_PER_USEC);
1395}
1396
1397/**
1398 * bxt_idle_state_table_update - Fix up the Broxton idle states table.
1399 *
1400 * On BXT, trust the IRTL (Interrupt Response Time Limit) MSR to show the
1401 * definitive maximum latency and use the same value for target_residency.
1402 */
1403static void __init bxt_idle_state_table_update(void)
1404{
1405 unsigned long long msr;
1406 unsigned int usec;
1407
1408 rdmsrl(MSR_PKGC6_IRTL, msr);
1409 usec = irtl_2_usec(msr);
1410 if (usec) {
1411 bxt_cstates[2].exit_latency = usec;
1412 bxt_cstates[2].target_residency = usec;
1413 }
1414
1415 rdmsrl(MSR_PKGC7_IRTL, msr);
1416 usec = irtl_2_usec(msr);
1417 if (usec) {
1418 bxt_cstates[3].exit_latency = usec;
1419 bxt_cstates[3].target_residency = usec;
1420 }
1421
1422 rdmsrl(MSR_PKGC8_IRTL, msr);
1423 usec = irtl_2_usec(msr);
1424 if (usec) {
1425 bxt_cstates[4].exit_latency = usec;
1426 bxt_cstates[4].target_residency = usec;
1427 }
1428
1429 rdmsrl(MSR_PKGC9_IRTL, msr);
1430 usec = irtl_2_usec(msr);
1431 if (usec) {
1432 bxt_cstates[5].exit_latency = usec;
1433 bxt_cstates[5].target_residency = usec;
1434 }
1435
1436 rdmsrl(MSR_PKGC10_IRTL, msr);
1437 usec = irtl_2_usec(msr);
1438 if (usec) {
1439 bxt_cstates[6].exit_latency = usec;
1440 bxt_cstates[6].target_residency = usec;
1441 }
1442
1443}
1444
1445/**
1446 * sklh_idle_state_table_update - Fix up the Sky Lake idle states table.
1447 *
1448 * On SKL-H (model 0x5e) skip C8 and C9 if C10 is enabled and SGX disabled.
1449 */
1450static void __init sklh_idle_state_table_update(void)
1451{
1452 unsigned long long msr;
1453 unsigned int eax, ebx, ecx, edx;
1454
1455
1456 /* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1457 if (max_cstate <= 7)
1458 return;
1459
1460 /* if PC10 not present in CPUID.MWAIT.EDX */
1461 if ((mwait_substates & (0xF << 28)) == 0)
1462 return;
1463
1464 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1465
1466 /* PC10 is not enabled in PKG C-state limit */
1467 if ((msr & 0xF) != 8)
1468 return;
1469
1470 ecx = 0;
1471 cpuid(7, &eax, &ebx, &ecx, &edx);
1472
1473 /* if SGX is present */
1474 if (ebx & (1 << 2)) {
1475
1476 rdmsrl(MSR_IA32_FEAT_CTL, msr);
1477
1478 /* if SGX is enabled */
1479 if (msr & (1 << 18))
1480 return;
1481 }
1482
1483 skl_cstates[5].flags |= CPUIDLE_FLAG_UNUSABLE; /* C8-SKL */
1484 skl_cstates[6].flags |= CPUIDLE_FLAG_UNUSABLE; /* C9-SKL */
1485}
1486
1487/**
1488 * skx_idle_state_table_update - Adjust the Sky Lake/Cascade Lake
1489 * idle states table.
1490 */
1491static void __init skx_idle_state_table_update(void)
1492{
1493 unsigned long long msr;
1494
1495 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1496
1497 /*
1498 * 000b: C0/C1 (no package C-state support)
1499 * 001b: C2
1500 * 010b: C6 (non-retention)
1501 * 011b: C6 (retention)
1502 * 111b: No Package C state limits.
1503 */
1504 if ((msr & 0x7) < 2) {
1505 /*
1506 * Uses the CC6 + PC0 latency and 3 times of
1507 * latency for target_residency if the PC6
1508 * is disabled in BIOS. This is consistent
1509 * with how intel_idle driver uses _CST
1510 * to set the target_residency.
1511 */
1512 skx_cstates[2].exit_latency = 92;
1513 skx_cstates[2].target_residency = 276;
1514 }
1515}
1516
1517static bool __init intel_idle_verify_cstate(unsigned int mwait_hint)
1518{
1519 unsigned int mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint) + 1;
1520 unsigned int num_substates = (mwait_substates >> mwait_cstate * 4) &
1521 MWAIT_SUBSTATE_MASK;
1522
1523 /* Ignore the C-state if there are NO sub-states in CPUID for it. */
1524 if (num_substates == 0)
1525 return false;
1526
1527 if (mwait_cstate > 2 && !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1528 mark_tsc_unstable("TSC halts in idle states deeper than C2");
1529
1530 return true;
1531}
1532
1533static void __init intel_idle_init_cstates_icpu(struct cpuidle_driver *drv)
1534{
1535 int cstate;
1536
1537 switch (boot_cpu_data.x86_model) {
1538 case INTEL_FAM6_IVYBRIDGE_X:
1539 ivt_idle_state_table_update();
1540 break;
1541 case INTEL_FAM6_ATOM_GOLDMONT:
1542 case INTEL_FAM6_ATOM_GOLDMONT_PLUS:
1543 bxt_idle_state_table_update();
1544 break;
1545 case INTEL_FAM6_SKYLAKE:
1546 sklh_idle_state_table_update();
1547 break;
1548 case INTEL_FAM6_SKYLAKE_X:
1549 skx_idle_state_table_update();
1550 break;
1551 }
1552
1553 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1554 unsigned int mwait_hint;
1555
1556 if (intel_idle_max_cstate_reached(cstate))
1557 break;
1558
1559 if (!cpuidle_state_table[cstate].enter &&
1560 !cpuidle_state_table[cstate].enter_s2idle)
1561 break;
1562
1563 /* If marked as unusable, skip this state. */
1564 if (cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_UNUSABLE) {
1565 pr_debug("state %s is disabled\n",
1566 cpuidle_state_table[cstate].name);
1567 continue;
1568 }
1569
1570 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1571 if (!intel_idle_verify_cstate(mwait_hint))
1572 continue;
1573
1574 /* Structure copy. */
1575 drv->states[drv->state_count] = cpuidle_state_table[cstate];
1576
1577 if ((disabled_states_mask & BIT(drv->state_count)) ||
1578 ((icpu->use_acpi || force_use_acpi) &&
1579 intel_idle_off_by_default(mwait_hint) &&
1580 !(cpuidle_state_table[cstate].flags & CPUIDLE_FLAG_ALWAYS_ENABLE)))
1581 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_OFF;
1582
1583 if (intel_idle_state_needs_timer_stop(&drv->states[drv->state_count]))
1584 drv->states[drv->state_count].flags |= CPUIDLE_FLAG_TIMER_STOP;
1585
1586 drv->state_count++;
1587 }
1588
1589 if (icpu->byt_auto_demotion_disable_flag) {
1590 wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
1591 wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
1592 }
1593}
1594
1595/**
1596 * intel_idle_cpuidle_driver_init - Create the list of available idle states.
1597 * @drv: cpuidle driver structure to initialize.
1598 */
1599static void __init intel_idle_cpuidle_driver_init(struct cpuidle_driver *drv)
1600{
1601 cpuidle_poll_state_init(drv);
1602
1603 if (disabled_states_mask & BIT(0))
1604 drv->states[0].flags |= CPUIDLE_FLAG_OFF;
1605
1606 drv->state_count = 1;
1607
1608 if (icpu)
1609 intel_idle_init_cstates_icpu(drv);
1610 else
1611 intel_idle_init_cstates_acpi(drv);
1612}
1613
1614static void auto_demotion_disable(void)
1615{
1616 unsigned long long msr_bits;
1617
1618 rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1619 msr_bits &= ~auto_demotion_disable_flags;
1620 wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
1621}
1622
1623static void c1e_promotion_disable(void)
1624{
1625 unsigned long long msr_bits;
1626
1627 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
1628 msr_bits &= ~0x2;
1629 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
1630}
1631
1632/**
1633 * intel_idle_cpu_init - Register the target CPU with the cpuidle core.
1634 * @cpu: CPU to initialize.
1635 *
1636 * Register a cpuidle device object for @cpu and update its MSRs in accordance
1637 * with the processor model flags.
1638 */
1639static int intel_idle_cpu_init(unsigned int cpu)
1640{
1641 struct cpuidle_device *dev;
1642
1643 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1644 dev->cpu = cpu;
1645
1646 if (cpuidle_register_device(dev)) {
1647 pr_debug("cpuidle_register_device %d failed!\n", cpu);
1648 return -EIO;
1649 }
1650
1651 if (auto_demotion_disable_flags)
1652 auto_demotion_disable();
1653
1654 if (disable_promotion_to_c1e)
1655 c1e_promotion_disable();
1656
1657 return 0;
1658}
1659
1660static int intel_idle_cpu_online(unsigned int cpu)
1661{
1662 struct cpuidle_device *dev;
1663
1664 if (!boot_cpu_has(X86_FEATURE_ARAT))
1665 tick_broadcast_enable();
1666
1667 /*
1668 * Some systems can hotplug a cpu at runtime after
1669 * the kernel has booted, we have to initialize the
1670 * driver in this case
1671 */
1672 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1673 if (!dev->registered)
1674 return intel_idle_cpu_init(cpu);
1675
1676 return 0;
1677}
1678
1679/**
1680 * intel_idle_cpuidle_devices_uninit - Unregister all cpuidle devices.
1681 */
1682static void __init intel_idle_cpuidle_devices_uninit(void)
1683{
1684 int i;
1685
1686 for_each_online_cpu(i)
1687 cpuidle_unregister_device(per_cpu_ptr(intel_idle_cpuidle_devices, i));
1688}
1689
1690static int __init intel_idle_init(void)
1691{
1692 const struct x86_cpu_id *id;
1693 unsigned int eax, ebx, ecx;
1694 int retval;
1695
1696 /* Do not load intel_idle at all for now if idle= is passed */
1697 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
1698 return -ENODEV;
1699
1700 if (max_cstate == 0) {
1701 pr_debug("disabled\n");
1702 return -EPERM;
1703 }
1704
1705 id = x86_match_cpu(intel_idle_ids);
1706 if (id) {
1707 if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
1708 pr_debug("Please enable MWAIT in BIOS SETUP\n");
1709 return -ENODEV;
1710 }
1711 } else {
1712 id = x86_match_cpu(intel_mwait_ids);
1713 if (!id)
1714 return -ENODEV;
1715 }
1716
1717 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
1718 return -ENODEV;
1719
1720 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
1721
1722 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
1723 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
1724 !mwait_substates)
1725 return -ENODEV;
1726
1727 pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
1728
1729 icpu = (const struct idle_cpu *)id->driver_data;
1730 if (icpu) {
1731 cpuidle_state_table = icpu->state_table;
1732 auto_demotion_disable_flags = icpu->auto_demotion_disable_flags;
1733 disable_promotion_to_c1e = icpu->disable_promotion_to_c1e;
1734 if (icpu->use_acpi || force_use_acpi)
1735 intel_idle_acpi_cst_extract();
1736 } else if (!intel_idle_acpi_cst_extract()) {
1737 return -ENODEV;
1738 }
1739
1740 pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
1741 boot_cpu_data.x86_model);
1742
1743 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
1744 if (!intel_idle_cpuidle_devices)
1745 return -ENOMEM;
1746
1747 intel_idle_cpuidle_driver_init(&intel_idle_driver);
1748
1749 retval = cpuidle_register_driver(&intel_idle_driver);
1750 if (retval) {
1751 struct cpuidle_driver *drv = cpuidle_get_driver();
1752 printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
1753 drv ? drv->name : "none");
1754 goto init_driver_fail;
1755 }
1756
1757 retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
1758 intel_idle_cpu_online, NULL);
1759 if (retval < 0)
1760 goto hp_setup_fail;
1761
1762 pr_debug("Local APIC timer is reliable in %s\n",
1763 boot_cpu_has(X86_FEATURE_ARAT) ? "all C-states" : "C1");
1764
1765 return 0;
1766
1767hp_setup_fail:
1768 intel_idle_cpuidle_devices_uninit();
1769 cpuidle_unregister_driver(&intel_idle_driver);
1770init_driver_fail:
1771 free_percpu(intel_idle_cpuidle_devices);
1772 return retval;
1773
1774}
1775device_initcall(intel_idle_init);
1776
1777/*
1778 * We are not really modular, but we used to support that. Meaning we also
1779 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
1780 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
1781 * is the easiest way (currently) to continue doing that.
1782 */
1783module_param(max_cstate, int, 0444);
1784/*
1785 * The positions of the bits that are set in this number are the indices of the
1786 * idle states to be disabled by default (as reflected by the names of the
1787 * corresponding idle state directories in sysfs, "state0", "state1" ...
1788 * "state<i>" ..., where <i> is the index of the given state).
1789 */
1790module_param_named(states_off, disabled_states_mask, uint, 0444);
1791MODULE_PARM_DESC(states_off, "Mask of disabled idle states");