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);
99 static int intel_idle_cpu_init(int cpu);
101 static struct cpuidle_state *cpuidle_state_table;
104 * Set this flag for states where the HW flushes the TLB for us
105 * and so we don't need cross-calls to keep it consistent.
106 * If this flag is set, SW flushes the TLB, so even if the
107 * HW doesn't do the flushing, this flag is safe to use.
109 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
112 * States are indexed by the cstate number,
113 * which is also the index into the MWAIT hint array.
114 * Thus C0 is a dummy.
116 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
120 .desc = "MWAIT 0x00",
121 .flags = CPUIDLE_FLAG_TIME_VALID,
123 .target_residency = 6,
124 .enter = &intel_idle },
127 .desc = "MWAIT 0x10",
128 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
130 .target_residency = 80,
131 .enter = &intel_idle },
134 .desc = "MWAIT 0x20",
135 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
137 .target_residency = 800,
138 .enter = &intel_idle },
141 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
145 .desc = "MWAIT 0x00",
146 .flags = CPUIDLE_FLAG_TIME_VALID,
148 .target_residency = 1,
149 .enter = &intel_idle },
152 .desc = "MWAIT 0x10",
153 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
155 .target_residency = 211,
156 .enter = &intel_idle },
159 .desc = "MWAIT 0x20",
160 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
162 .target_residency = 345,
163 .enter = &intel_idle },
166 .desc = "MWAIT 0x30",
167 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
169 .target_residency = 345,
170 .enter = &intel_idle },
173 static struct cpuidle_state ivb_cstates[MWAIT_MAX_NUM_CSTATES] = {
177 .desc = "MWAIT 0x00",
178 .flags = CPUIDLE_FLAG_TIME_VALID,
180 .target_residency = 1,
181 .enter = &intel_idle },
184 .desc = "MWAIT 0x10",
185 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
187 .target_residency = 156,
188 .enter = &intel_idle },
191 .desc = "MWAIT 0x20",
192 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
194 .target_residency = 300,
195 .enter = &intel_idle },
198 .desc = "MWAIT 0x30",
199 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
201 .target_residency = 300,
202 .enter = &intel_idle },
205 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
209 .desc = "MWAIT 0x00",
210 .flags = CPUIDLE_FLAG_TIME_VALID,
212 .target_residency = 4,
213 .enter = &intel_idle },
216 .desc = "MWAIT 0x10",
217 .flags = CPUIDLE_FLAG_TIME_VALID,
219 .target_residency = 80,
220 .enter = &intel_idle },
224 .desc = "MWAIT 0x30",
225 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
227 .target_residency = 400,
228 .enter = &intel_idle },
232 .desc = "MWAIT 0x52",
233 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
235 .target_residency = 560,
236 .enter = &intel_idle },
239 static long get_driver_data(int cstate)
244 case 1: /* MWAIT C1 */
247 case 2: /* MWAIT C2 */
250 case 3: /* MWAIT C3 */
253 case 4: /* MWAIT C4 */
256 case 5: /* MWAIT C5 */
259 case 6: /* MWAIT C6 */
270 * @dev: cpuidle_device
271 * @drv: cpuidle driver
272 * @index: index of cpuidle state
274 * Must be called under local_irq_disable().
276 static int intel_idle(struct cpuidle_device *dev,
277 struct cpuidle_driver *drv, int index)
279 unsigned long ecx = 1; /* break on interrupt flag */
280 struct cpuidle_state *state = &drv->states[index];
281 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
282 unsigned long eax = (unsigned long)cpuidle_get_statedata(state_usage);
284 ktime_t kt_before, kt_after;
286 int cpu = smp_processor_id();
288 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
291 * leave_mm() to avoid costly and often unnecessary wakeups
292 * for flushing the user TLB's associated with the active mm.
294 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
297 if (!(lapic_timer_reliable_states & (1 << (cstate))))
298 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
300 kt_before = ktime_get_real();
302 stop_critical_timings();
303 if (!need_resched()) {
305 __monitor((void *)¤t_thread_info()->flags, 0, 0);
311 start_critical_timings();
313 kt_after = ktime_get_real();
314 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
318 if (!(lapic_timer_reliable_states & (1 << (cstate))))
319 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
321 /* Update cpuidle counters */
322 dev->last_residency = (int)usec_delta;
327 static void __setup_broadcast_timer(void *arg)
329 unsigned long reason = (unsigned long)arg;
330 int cpu = smp_processor_id();
333 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
335 clockevents_notify(reason, &cpu);
338 static int cpu_hotplug_notify(struct notifier_block *n,
339 unsigned long action, void *hcpu)
341 int hotcpu = (unsigned long)hcpu;
342 struct cpuidle_device *dev;
344 switch (action & 0xf) {
347 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
348 smp_call_function_single(hotcpu, __setup_broadcast_timer,
352 * Some systems can hotplug a cpu at runtime after
353 * the kernel has booted, we have to initialize the
354 * driver in this case
356 dev = per_cpu_ptr(intel_idle_cpuidle_devices, hotcpu);
357 if (!dev->registered)
358 intel_idle_cpu_init(hotcpu);
365 static struct notifier_block cpu_hotplug_notifier = {
366 .notifier_call = cpu_hotplug_notify,
369 static void auto_demotion_disable(void *dummy)
371 unsigned long long msr_bits;
373 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
374 msr_bits &= ~(icpu->auto_demotion_disable_flags);
375 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
378 static const struct idle_cpu idle_cpu_nehalem = {
379 .state_table = nehalem_cstates,
380 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
383 static const struct idle_cpu idle_cpu_atom = {
384 .state_table = atom_cstates,
387 static const struct idle_cpu idle_cpu_lincroft = {
388 .state_table = atom_cstates,
389 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
392 static const struct idle_cpu idle_cpu_snb = {
393 .state_table = snb_cstates,
396 static const struct idle_cpu idle_cpu_ivb = {
397 .state_table = ivb_cstates,
400 #define ICPU(model, cpu) \
401 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
403 static const struct x86_cpu_id intel_idle_ids[] = {
404 ICPU(0x1a, idle_cpu_nehalem),
405 ICPU(0x1e, idle_cpu_nehalem),
406 ICPU(0x1f, idle_cpu_nehalem),
407 ICPU(0x25, idle_cpu_nehalem),
408 ICPU(0x2c, idle_cpu_nehalem),
409 ICPU(0x2e, idle_cpu_nehalem),
410 ICPU(0x1c, idle_cpu_atom),
411 ICPU(0x26, idle_cpu_lincroft),
412 ICPU(0x2f, idle_cpu_nehalem),
413 ICPU(0x2a, idle_cpu_snb),
414 ICPU(0x2d, idle_cpu_snb),
415 ICPU(0x3a, idle_cpu_ivb),
416 ICPU(0x3e, idle_cpu_ivb),
419 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
424 static int intel_idle_probe(void)
426 unsigned int eax, ebx, ecx;
427 const struct x86_cpu_id *id;
429 if (max_cstate == 0) {
430 pr_debug(PREFIX "disabled\n");
434 id = x86_match_cpu(intel_idle_ids);
436 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
437 boot_cpu_data.x86 == 6)
438 pr_debug(PREFIX "does not run on family %d model %d\n",
439 boot_cpu_data.x86, boot_cpu_data.x86_model);
443 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
446 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
448 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
449 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
453 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
455 icpu = (const struct idle_cpu *)id->driver_data;
456 cpuidle_state_table = icpu->state_table;
458 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
459 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
461 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
463 register_cpu_notifier(&cpu_hotplug_notifier);
465 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
466 " model 0x%X\n", boot_cpu_data.x86_model);
468 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
469 lapic_timer_reliable_states);
474 * intel_idle_cpuidle_devices_uninit()
475 * unregister, free cpuidle_devices
477 static void intel_idle_cpuidle_devices_uninit(void)
480 struct cpuidle_device *dev;
482 for_each_online_cpu(i) {
483 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
484 cpuidle_unregister_device(dev);
487 free_percpu(intel_idle_cpuidle_devices);
491 * intel_idle_cpuidle_driver_init()
492 * allocate, initialize cpuidle_states
494 static int intel_idle_cpuidle_driver_init(void)
497 struct cpuidle_driver *drv = &intel_idle_driver;
499 drv->state_count = 1;
501 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
504 if (cstate > max_cstate) {
505 printk(PREFIX "max_cstate %d reached\n",
510 /* does the state exist in CPUID.MWAIT? */
511 num_substates = (mwait_substates >> ((cstate) * 4))
512 & MWAIT_SUBSTATE_MASK;
513 if (num_substates == 0)
515 /* is the state not enabled? */
516 if (cpuidle_state_table[cstate].enter == NULL) {
517 /* does the driver not know about the state? */
518 if (*cpuidle_state_table[cstate].name == '\0')
519 pr_debug(PREFIX "unaware of model 0x%x"
521 " contact lenb@kernel.org",
522 boot_cpu_data.x86_model, cstate);
527 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
528 mark_tsc_unstable("TSC halts in idle"
529 " states deeper than C2");
531 drv->states[drv->state_count] = /* structure copy */
532 cpuidle_state_table[cstate];
534 drv->state_count += 1;
537 if (icpu->auto_demotion_disable_flags)
538 on_each_cpu(auto_demotion_disable, NULL, 1);
545 * intel_idle_cpu_init()
546 * allocate, initialize, register cpuidle_devices
547 * @cpu: cpu/core to initialize
549 static int intel_idle_cpu_init(int cpu)
552 struct cpuidle_device *dev;
554 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
556 dev->state_count = 1;
558 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
561 if (cstate > max_cstate) {
562 printk(PREFIX "max_cstate %d reached\n", max_cstate);
566 /* does the state exist in CPUID.MWAIT? */
567 num_substates = (mwait_substates >> ((cstate) * 4))
568 & MWAIT_SUBSTATE_MASK;
569 if (num_substates == 0)
571 /* is the state not enabled? */
572 if (cpuidle_state_table[cstate].enter == NULL)
575 dev->states_usage[dev->state_count].driver_data =
576 (void *)get_driver_data(cstate);
578 dev->state_count += 1;
583 if (cpuidle_register_device(dev)) {
584 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
585 intel_idle_cpuidle_devices_uninit();
589 if (icpu->auto_demotion_disable_flags)
590 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
595 static int __init intel_idle_init(void)
599 /* Do not load intel_idle at all for now if idle= is passed */
600 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
603 retval = intel_idle_probe();
607 intel_idle_cpuidle_driver_init();
608 retval = cpuidle_register_driver(&intel_idle_driver);
610 struct cpuidle_driver *drv = cpuidle_get_driver();
611 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
612 drv ? drv->name : "none");
616 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
617 if (intel_idle_cpuidle_devices == NULL)
620 for_each_online_cpu(i) {
621 retval = intel_idle_cpu_init(i);
623 cpuidle_unregister_driver(&intel_idle_driver);
631 static void __exit intel_idle_exit(void)
633 intel_idle_cpuidle_devices_uninit();
634 cpuidle_unregister_driver(&intel_idle_driver);
637 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
638 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
639 unregister_cpu_notifier(&cpu_hotplug_notifier);
644 module_init(intel_idle_init);
645 module_exit(intel_idle_exit);
647 module_param(max_cstate, int, 0444);
649 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
650 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
651 MODULE_LICENSE("GPL");