2 * intel_idle.c - native hardware idle loop for modern Intel processors
4 * Copyright (c) 2010, Intel Corporation.
5 * Len Brown <len.brown@intel.com>
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.
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
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.
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.
31 * All CPUs have same idle states as boot CPU
33 * Chipset BM_STS (bus master status) bit is a NOP
34 * for preventing entry into deep C-stats
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.
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.
49 * There is currently no kernel-based automatic probing/loading mechanism
50 * if the driver is built as a module.
53 /* un-comment DEBUG to enable pr_debug() statements */
56 #include <linux/kernel.h>
57 #include <linux/cpuidle.h>
58 #include <linux/clockchips.h>
59 #include <trace/events/power.h>
60 #include <linux/sched.h>
61 #include <linux/notifier.h>
62 #include <linux/cpu.h>
63 #include <linux/module.h>
64 #include <asm/cpu_device_id.h>
65 #include <asm/mwait.h>
68 #define INTEL_IDLE_VERSION "0.4"
69 #define PREFIX "intel_idle: "
71 static struct cpuidle_driver intel_idle_driver = {
74 .en_core_tk_irqen = 1,
76 /* intel_idle.max_cstate=0 disables driver */
77 static int max_cstate = CPUIDLE_STATE_MAX - 1;
79 static unsigned int mwait_substates;
81 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
82 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */
83 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
86 struct cpuidle_state *state_table;
89 * Hardware C-state auto-demotion may not always be optimal.
90 * Indicate which enable bits to clear here.
92 unsigned long auto_demotion_disable_flags;
93 bool disable_promotion_to_c1e;
96 static const struct idle_cpu *icpu;
97 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
98 static int intel_idle(struct cpuidle_device *dev,
99 struct cpuidle_driver *drv, int index);
100 static int intel_idle_cpu_init(int cpu);
102 static struct cpuidle_state *cpuidle_state_table;
105 * Set this flag for states where the HW flushes the TLB for us
106 * and so we don't need cross-calls to keep it consistent.
107 * If this flag is set, SW flushes the TLB, so even if the
108 * HW doesn't do the flushing, this flag is safe to use.
110 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
113 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
114 * the C-state (top nibble) and sub-state (bottom nibble)
115 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
117 * We store the hint at the top of our "flags" for each state.
119 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
120 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
123 * States are indexed by the cstate number,
124 * which is also the index into the MWAIT hint array.
125 * Thus C0 is a dummy.
127 static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
130 .desc = "MWAIT 0x00",
131 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
133 .target_residency = 6,
134 .enter = &intel_idle },
137 .desc = "MWAIT 0x01",
138 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
140 .target_residency = 20,
141 .enter = &intel_idle },
144 .desc = "MWAIT 0x10",
145 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
147 .target_residency = 80,
148 .enter = &intel_idle },
151 .desc = "MWAIT 0x20",
152 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
154 .target_residency = 800,
155 .enter = &intel_idle },
160 static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
163 .desc = "MWAIT 0x00",
164 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
166 .target_residency = 2,
167 .enter = &intel_idle },
170 .desc = "MWAIT 0x01",
171 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
173 .target_residency = 20,
174 .enter = &intel_idle },
177 .desc = "MWAIT 0x10",
178 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
180 .target_residency = 211,
181 .enter = &intel_idle },
184 .desc = "MWAIT 0x20",
185 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
187 .target_residency = 345,
188 .enter = &intel_idle },
191 .desc = "MWAIT 0x30",
192 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
194 .target_residency = 345,
195 .enter = &intel_idle },
200 static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
203 .desc = "MWAIT 0x00",
204 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
206 .target_residency = 1,
207 .enter = &intel_idle },
210 .desc = "MWAIT 0x01",
211 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
213 .target_residency = 20,
214 .enter = &intel_idle },
217 .desc = "MWAIT 0x10",
218 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
220 .target_residency = 156,
221 .enter = &intel_idle },
224 .desc = "MWAIT 0x20",
225 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
227 .target_residency = 300,
228 .enter = &intel_idle },
231 .desc = "MWAIT 0x30",
232 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
234 .target_residency = 300,
235 .enter = &intel_idle },
240 static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
243 .desc = "MWAIT 0x00",
244 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
246 .target_residency = 2,
247 .enter = &intel_idle },
250 .desc = "MWAIT 0x01",
251 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
253 .target_residency = 20,
254 .enter = &intel_idle },
257 .desc = "MWAIT 0x10",
258 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
260 .target_residency = 100,
261 .enter = &intel_idle },
264 .desc = "MWAIT 0x20",
265 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
267 .target_residency = 400,
268 .enter = &intel_idle },
271 .desc = "MWAIT 0x32",
272 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
274 .target_residency = 500,
275 .enter = &intel_idle },
280 static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = {
283 .desc = "MWAIT 0x00",
284 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
286 .target_residency = 20,
287 .enter = &intel_idle },
290 .desc = "MWAIT 0x10",
291 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID,
293 .target_residency = 80,
294 .enter = &intel_idle },
297 .desc = "MWAIT 0x30",
298 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
300 .target_residency = 400,
301 .enter = &intel_idle },
304 .desc = "MWAIT 0x52",
305 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
307 .target_residency = 560,
308 .enter = &intel_idle },
315 * @dev: cpuidle_device
316 * @drv: cpuidle driver
317 * @index: index of cpuidle state
319 * Must be called under local_irq_disable().
321 static int intel_idle(struct cpuidle_device *dev,
322 struct cpuidle_driver *drv, int index)
324 unsigned long ecx = 1; /* break on interrupt flag */
325 struct cpuidle_state *state = &drv->states[index];
326 unsigned long eax = flg2MWAIT(state->flags);
328 int cpu = smp_processor_id();
330 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
333 * leave_mm() to avoid costly and often unnecessary wakeups
334 * for flushing the user TLB's associated with the active mm.
336 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
339 if (!(lapic_timer_reliable_states & (1 << (cstate))))
340 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
342 stop_critical_timings();
343 if (!need_resched()) {
345 __monitor((void *)¤t_thread_info()->flags, 0, 0);
351 start_critical_timings();
353 if (!(lapic_timer_reliable_states & (1 << (cstate))))
354 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
359 static void __setup_broadcast_timer(void *arg)
361 unsigned long reason = (unsigned long)arg;
362 int cpu = smp_processor_id();
365 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
367 clockevents_notify(reason, &cpu);
370 static int cpu_hotplug_notify(struct notifier_block *n,
371 unsigned long action, void *hcpu)
373 int hotcpu = (unsigned long)hcpu;
374 struct cpuidle_device *dev;
376 switch (action & 0xf) {
379 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
380 smp_call_function_single(hotcpu, __setup_broadcast_timer,
384 * Some systems can hotplug a cpu at runtime after
385 * the kernel has booted, we have to initialize the
386 * driver in this case
388 dev = per_cpu_ptr(intel_idle_cpuidle_devices, hotcpu);
389 if (!dev->registered)
390 intel_idle_cpu_init(hotcpu);
397 static struct notifier_block cpu_hotplug_notifier = {
398 .notifier_call = cpu_hotplug_notify,
401 static void auto_demotion_disable(void *dummy)
403 unsigned long long msr_bits;
405 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
406 msr_bits &= ~(icpu->auto_demotion_disable_flags);
407 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
409 static void c1e_promotion_disable(void *dummy)
411 unsigned long long msr_bits;
413 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
415 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
418 static const struct idle_cpu idle_cpu_nehalem = {
419 .state_table = nehalem_cstates,
420 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
421 .disable_promotion_to_c1e = true,
424 static const struct idle_cpu idle_cpu_atom = {
425 .state_table = atom_cstates,
428 static const struct idle_cpu idle_cpu_lincroft = {
429 .state_table = atom_cstates,
430 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
433 static const struct idle_cpu idle_cpu_snb = {
434 .state_table = snb_cstates,
435 .disable_promotion_to_c1e = true,
438 static const struct idle_cpu idle_cpu_ivb = {
439 .state_table = ivb_cstates,
440 .disable_promotion_to_c1e = true,
443 static const struct idle_cpu idle_cpu_hsw = {
444 .state_table = hsw_cstates,
445 .disable_promotion_to_c1e = true,
448 #define ICPU(model, cpu) \
449 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
451 static const struct x86_cpu_id intel_idle_ids[] = {
452 ICPU(0x1a, idle_cpu_nehalem),
453 ICPU(0x1e, idle_cpu_nehalem),
454 ICPU(0x1f, idle_cpu_nehalem),
455 ICPU(0x25, idle_cpu_nehalem),
456 ICPU(0x2c, idle_cpu_nehalem),
457 ICPU(0x2e, idle_cpu_nehalem),
458 ICPU(0x1c, idle_cpu_atom),
459 ICPU(0x26, idle_cpu_lincroft),
460 ICPU(0x2f, idle_cpu_nehalem),
461 ICPU(0x2a, idle_cpu_snb),
462 ICPU(0x2d, idle_cpu_snb),
463 ICPU(0x3a, idle_cpu_ivb),
464 ICPU(0x3e, idle_cpu_ivb),
465 ICPU(0x3c, idle_cpu_hsw),
466 ICPU(0x3f, idle_cpu_hsw),
467 ICPU(0x45, idle_cpu_hsw),
470 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
475 static int intel_idle_probe(void)
477 unsigned int eax, ebx, ecx;
478 const struct x86_cpu_id *id;
480 if (max_cstate == 0) {
481 pr_debug(PREFIX "disabled\n");
485 id = x86_match_cpu(intel_idle_ids);
487 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
488 boot_cpu_data.x86 == 6)
489 pr_debug(PREFIX "does not run on family %d model %d\n",
490 boot_cpu_data.x86, boot_cpu_data.x86_model);
494 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
497 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
499 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
500 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
504 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
506 icpu = (const struct idle_cpu *)id->driver_data;
507 cpuidle_state_table = icpu->state_table;
509 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
510 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
512 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
514 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
515 " model 0x%X\n", boot_cpu_data.x86_model);
517 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
518 lapic_timer_reliable_states);
523 * intel_idle_cpuidle_devices_uninit()
524 * unregister, free cpuidle_devices
526 static void intel_idle_cpuidle_devices_uninit(void)
529 struct cpuidle_device *dev;
531 for_each_online_cpu(i) {
532 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
533 cpuidle_unregister_device(dev);
536 free_percpu(intel_idle_cpuidle_devices);
540 * intel_idle_cpuidle_driver_init()
541 * allocate, initialize cpuidle_states
543 static int intel_idle_cpuidle_driver_init(void)
546 struct cpuidle_driver *drv = &intel_idle_driver;
548 drv->state_count = 1;
550 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
551 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
553 if (cpuidle_state_table[cstate].enter == NULL)
556 if (cstate + 1 > max_cstate) {
557 printk(PREFIX "max_cstate %d reached\n",
562 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
563 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
564 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
566 /* does the state exist in CPUID.MWAIT? */
567 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
568 & MWAIT_SUBSTATE_MASK;
570 /* if sub-state in table is not enumerated by CPUID */
571 if ((mwait_substate + 1) > num_substates)
574 if (((mwait_cstate + 1) > 2) &&
575 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
576 mark_tsc_unstable("TSC halts in idle"
577 " states deeper than C2");
579 drv->states[drv->state_count] = /* structure copy */
580 cpuidle_state_table[cstate];
582 drv->state_count += 1;
585 if (icpu->auto_demotion_disable_flags)
586 on_each_cpu(auto_demotion_disable, NULL, 1);
588 if (icpu->disable_promotion_to_c1e) /* each-cpu is redundant */
589 on_each_cpu(c1e_promotion_disable, NULL, 1);
596 * intel_idle_cpu_init()
597 * allocate, initialize, register cpuidle_devices
598 * @cpu: cpu/core to initialize
600 static int intel_idle_cpu_init(int cpu)
603 struct cpuidle_device *dev;
605 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
607 dev->state_count = 1;
609 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
610 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
612 if (cpuidle_state_table[cstate].enter == NULL)
615 if (cstate + 1 > max_cstate) {
616 printk(PREFIX "max_cstate %d reached\n", max_cstate);
620 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
621 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
622 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
624 /* does the state exist in CPUID.MWAIT? */
625 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
626 & MWAIT_SUBSTATE_MASK;
628 /* if sub-state in table is not enumerated by CPUID */
629 if ((mwait_substate + 1) > num_substates)
632 dev->state_count += 1;
637 if (cpuidle_register_device(dev)) {
638 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
639 intel_idle_cpuidle_devices_uninit();
643 if (icpu->auto_demotion_disable_flags)
644 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
649 static int __init intel_idle_init(void)
653 /* Do not load intel_idle at all for now if idle= is passed */
654 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
657 retval = intel_idle_probe();
661 intel_idle_cpuidle_driver_init();
662 retval = cpuidle_register_driver(&intel_idle_driver);
664 struct cpuidle_driver *drv = cpuidle_get_driver();
665 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
666 drv ? drv->name : "none");
670 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
671 if (intel_idle_cpuidle_devices == NULL)
674 for_each_online_cpu(i) {
675 retval = intel_idle_cpu_init(i);
677 cpuidle_unregister_driver(&intel_idle_driver);
681 register_cpu_notifier(&cpu_hotplug_notifier);
686 static void __exit intel_idle_exit(void)
688 intel_idle_cpuidle_devices_uninit();
689 cpuidle_unregister_driver(&intel_idle_driver);
692 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
693 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
694 unregister_cpu_notifier(&cpu_hotplug_notifier);
699 module_init(intel_idle_init);
700 module_exit(intel_idle_exit);
702 module_param(max_cstate, int, 0444);
704 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
705 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
706 MODULE_LICENSE("GPL");