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 <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 <linux/module.h>
65 #include <asm/cpu_device_id.h>
66 #include <asm/mwait.h>
69 #define INTEL_IDLE_VERSION "0.4"
70 #define PREFIX "intel_idle: "
72 static struct cpuidle_driver intel_idle_driver = {
76 /* intel_idle.max_cstate=0 disables driver */
77 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 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;
95 static const struct idle_cpu *icpu;
96 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
97 static int intel_idle(struct cpuidle_device *dev,
98 struct cpuidle_driver *drv, int index);
100 static struct cpuidle_state *cpuidle_state_table;
103 * Set this flag for states where the HW flushes the TLB for us
104 * and so we don't need cross-calls to keep it consistent.
105 * If this flag is set, SW flushes the TLB, so even if the
106 * HW doesn't do the flushing, this flag is safe to use.
108 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
111 * States are indexed by the cstate number,
112 * which is also the index into the MWAIT hint array.
113 * Thus C0 is a dummy.
115 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
119 .desc = "MWAIT 0x00",
120 .flags = CPUIDLE_FLAG_TIME_VALID,
122 .target_residency = 6,
123 .enter = &intel_idle },
126 .desc = "MWAIT 0x10",
127 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
129 .target_residency = 80,
130 .enter = &intel_idle },
133 .desc = "MWAIT 0x20",
134 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
136 .target_residency = 800,
137 .enter = &intel_idle },
140 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
144 .desc = "MWAIT 0x00",
145 .flags = CPUIDLE_FLAG_TIME_VALID,
147 .target_residency = 1,
148 .enter = &intel_idle },
151 .desc = "MWAIT 0x10",
152 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
154 .target_residency = 211,
155 .enter = &intel_idle },
158 .desc = "MWAIT 0x20",
159 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
161 .target_residency = 345,
162 .enter = &intel_idle },
165 .desc = "MWAIT 0x30",
166 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
168 .target_residency = 345,
169 .enter = &intel_idle },
172 static struct cpuidle_state ivb_cstates[MWAIT_MAX_NUM_CSTATES] = {
176 .desc = "MWAIT 0x00",
177 .flags = CPUIDLE_FLAG_TIME_VALID,
179 .target_residency = 1,
180 .enter = &intel_idle },
183 .desc = "MWAIT 0x10",
184 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
186 .target_residency = 156,
187 .enter = &intel_idle },
190 .desc = "MWAIT 0x20",
191 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
193 .target_residency = 300,
194 .enter = &intel_idle },
197 .desc = "MWAIT 0x30",
198 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
200 .target_residency = 300,
201 .enter = &intel_idle },
204 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
208 .desc = "MWAIT 0x00",
209 .flags = CPUIDLE_FLAG_TIME_VALID,
211 .target_residency = 4,
212 .enter = &intel_idle },
215 .desc = "MWAIT 0x10",
216 .flags = CPUIDLE_FLAG_TIME_VALID,
218 .target_residency = 80,
219 .enter = &intel_idle },
223 .desc = "MWAIT 0x30",
224 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
226 .target_residency = 400,
227 .enter = &intel_idle },
231 .desc = "MWAIT 0x52",
232 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
234 .target_residency = 560,
235 .enter = &intel_idle },
238 static long get_driver_data(int cstate)
243 case 1: /* MWAIT C1 */
246 case 2: /* MWAIT C2 */
249 case 3: /* MWAIT C3 */
252 case 4: /* MWAIT C4 */
255 case 5: /* MWAIT C5 */
258 case 6: /* MWAIT C6 */
269 * @dev: cpuidle_device
270 * @drv: cpuidle driver
271 * @index: index of cpuidle state
273 * Must be called under local_irq_disable().
275 static int intel_idle(struct cpuidle_device *dev,
276 struct cpuidle_driver *drv, int index)
278 unsigned long ecx = 1; /* break on interrupt flag */
279 struct cpuidle_state *state = &drv->states[index];
280 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
281 unsigned long eax = (unsigned long)cpuidle_get_statedata(state_usage);
283 ktime_t kt_before, kt_after;
285 int cpu = smp_processor_id();
287 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
290 * leave_mm() to avoid costly and often unnecessary wakeups
291 * for flushing the user TLB's associated with the active mm.
293 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
296 if (!(lapic_timer_reliable_states & (1 << (cstate))))
297 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
299 kt_before = ktime_get_real();
301 stop_critical_timings();
302 if (!need_resched()) {
304 __monitor((void *)¤t_thread_info()->flags, 0, 0);
310 start_critical_timings();
312 kt_after = ktime_get_real();
313 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
317 if (!(lapic_timer_reliable_states & (1 << (cstate))))
318 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
320 /* Update cpuidle counters */
321 dev->last_residency = (int)usec_delta;
326 static void __setup_broadcast_timer(void *arg)
328 unsigned long reason = (unsigned long)arg;
329 int cpu = smp_processor_id();
332 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
334 clockevents_notify(reason, &cpu);
337 static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
338 unsigned long action, void *hcpu)
340 int hotcpu = (unsigned long)hcpu;
342 switch (action & 0xf) {
344 smp_call_function_single(hotcpu, __setup_broadcast_timer,
351 static struct notifier_block setup_broadcast_notifier = {
352 .notifier_call = setup_broadcast_cpuhp_notify,
355 static void auto_demotion_disable(void *dummy)
357 unsigned long long msr_bits;
359 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
360 msr_bits &= ~(icpu->auto_demotion_disable_flags);
361 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
364 static const struct idle_cpu idle_cpu_nehalem = {
365 .state_table = nehalem_cstates,
366 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
369 static const struct idle_cpu idle_cpu_atom = {
370 .state_table = atom_cstates,
373 static const struct idle_cpu idle_cpu_lincroft = {
374 .state_table = atom_cstates,
375 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
378 static const struct idle_cpu idle_cpu_snb = {
379 .state_table = snb_cstates,
382 static const struct idle_cpu idle_cpu_ivb = {
383 .state_table = ivb_cstates,
386 #define ICPU(model, cpu) \
387 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
389 static const struct x86_cpu_id intel_idle_ids[] = {
390 ICPU(0x1a, idle_cpu_nehalem),
391 ICPU(0x1e, idle_cpu_nehalem),
392 ICPU(0x1f, idle_cpu_nehalem),
393 ICPU(0x25, idle_cpu_nehalem),
394 ICPU(0x2c, idle_cpu_nehalem),
395 ICPU(0x2e, idle_cpu_nehalem),
396 ICPU(0x1c, idle_cpu_atom),
397 ICPU(0x26, idle_cpu_lincroft),
398 ICPU(0x2f, idle_cpu_nehalem),
399 ICPU(0x2a, idle_cpu_snb),
400 ICPU(0x2d, idle_cpu_snb),
401 ICPU(0x3a, idle_cpu_ivb),
404 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
409 static int intel_idle_probe(void)
411 unsigned int eax, ebx, ecx;
412 const struct x86_cpu_id *id;
414 if (max_cstate == 0) {
415 pr_debug(PREFIX "disabled\n");
419 id = x86_match_cpu(intel_idle_ids);
421 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
422 boot_cpu_data.x86 == 6)
423 pr_debug(PREFIX "does not run on family %d model %d\n",
424 boot_cpu_data.x86, boot_cpu_data.x86_model);
428 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
431 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
433 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
434 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
438 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
440 icpu = (const struct idle_cpu *)id->driver_data;
441 cpuidle_state_table = icpu->state_table;
443 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
444 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
446 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
447 register_cpu_notifier(&setup_broadcast_notifier);
450 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
451 " model 0x%X\n", boot_cpu_data.x86_model);
453 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
454 lapic_timer_reliable_states);
459 * intel_idle_cpuidle_devices_uninit()
460 * unregister, free cpuidle_devices
462 static void intel_idle_cpuidle_devices_uninit(void)
465 struct cpuidle_device *dev;
467 for_each_online_cpu(i) {
468 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
469 cpuidle_unregister_device(dev);
472 free_percpu(intel_idle_cpuidle_devices);
476 * intel_idle_cpuidle_driver_init()
477 * allocate, initialize cpuidle_states
479 static int intel_idle_cpuidle_driver_init(void)
482 struct cpuidle_driver *drv = &intel_idle_driver;
484 drv->state_count = 1;
486 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
489 if (cstate > max_cstate) {
490 printk(PREFIX "max_cstate %d reached\n",
495 /* does the state exist in CPUID.MWAIT? */
496 num_substates = (mwait_substates >> ((cstate) * 4))
497 & MWAIT_SUBSTATE_MASK;
498 if (num_substates == 0)
500 /* is the state not enabled? */
501 if (cpuidle_state_table[cstate].enter == NULL) {
502 /* does the driver not know about the state? */
503 if (*cpuidle_state_table[cstate].name == '\0')
504 pr_debug(PREFIX "unaware of model 0x%x"
506 " contact lenb@kernel.org",
507 boot_cpu_data.x86_model, cstate);
512 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
513 mark_tsc_unstable("TSC halts in idle"
514 " states deeper than C2");
516 drv->states[drv->state_count] = /* structure copy */
517 cpuidle_state_table[cstate];
519 drv->state_count += 1;
522 if (icpu->auto_demotion_disable_flags)
523 on_each_cpu(auto_demotion_disable, NULL, 1);
530 * intel_idle_cpu_init()
531 * allocate, initialize, register cpuidle_devices
532 * @cpu: cpu/core to initialize
534 int intel_idle_cpu_init(int cpu)
537 struct cpuidle_device *dev;
539 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
541 dev->state_count = 1;
543 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
546 if (cstate > max_cstate) {
547 printk(PREFIX "max_cstate %d reached\n", max_cstate);
551 /* does the state exist in CPUID.MWAIT? */
552 num_substates = (mwait_substates >> ((cstate) * 4))
553 & MWAIT_SUBSTATE_MASK;
554 if (num_substates == 0)
556 /* is the state not enabled? */
557 if (cpuidle_state_table[cstate].enter == NULL)
560 dev->states_usage[dev->state_count].driver_data =
561 (void *)get_driver_data(cstate);
563 dev->state_count += 1;
568 if (cpuidle_register_device(dev)) {
569 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
570 intel_idle_cpuidle_devices_uninit();
574 if (icpu->auto_demotion_disable_flags)
575 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
579 EXPORT_SYMBOL_GPL(intel_idle_cpu_init);
581 static int __init intel_idle_init(void)
585 /* Do not load intel_idle at all for now if idle= is passed */
586 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
589 retval = intel_idle_probe();
593 intel_idle_cpuidle_driver_init();
594 retval = cpuidle_register_driver(&intel_idle_driver);
596 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
597 cpuidle_get_driver()->name);
601 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
602 if (intel_idle_cpuidle_devices == NULL)
605 for_each_online_cpu(i) {
606 retval = intel_idle_cpu_init(i);
608 cpuidle_unregister_driver(&intel_idle_driver);
616 static void __exit intel_idle_exit(void)
618 intel_idle_cpuidle_devices_uninit();
619 cpuidle_unregister_driver(&intel_idle_driver);
621 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
622 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
623 unregister_cpu_notifier(&setup_broadcast_notifier);
629 module_init(intel_idle_init);
630 module_exit(intel_idle_exit);
632 module_param(max_cstate, int, 0444);
634 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
635 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
636 MODULE_LICENSE("GPL");