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 = 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 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
113 * the C-state (top nibble) and sub-state (bottom nibble)
114 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
116 * We store the hint at the top of our "flags" for each state.
118 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
119 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
122 * States are indexed by the cstate number,
123 * which is also the index into the MWAIT hint array.
124 * Thus C0 is a dummy.
126 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
130 .desc = "MWAIT 0x00",
131 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
133 .target_residency = 6,
134 .enter = &intel_idle },
137 .desc = "MWAIT 0x10",
138 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
140 .target_residency = 80,
141 .enter = &intel_idle },
144 .desc = "MWAIT 0x20",
145 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
147 .target_residency = 800,
148 .enter = &intel_idle },
151 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
155 .desc = "MWAIT 0x00",
156 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
158 .target_residency = 1,
159 .enter = &intel_idle },
162 .desc = "MWAIT 0x10",
163 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
165 .target_residency = 211,
166 .enter = &intel_idle },
169 .desc = "MWAIT 0x20",
170 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
172 .target_residency = 345,
173 .enter = &intel_idle },
176 .desc = "MWAIT 0x30",
177 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
179 .target_residency = 345,
180 .enter = &intel_idle },
183 static struct cpuidle_state ivb_cstates[MWAIT_MAX_NUM_CSTATES] = {
187 .desc = "MWAIT 0x00",
188 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
190 .target_residency = 1,
191 .enter = &intel_idle },
194 .desc = "MWAIT 0x10",
195 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
197 .target_residency = 156,
198 .enter = &intel_idle },
201 .desc = "MWAIT 0x20",
202 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
204 .target_residency = 300,
205 .enter = &intel_idle },
208 .desc = "MWAIT 0x30",
209 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
211 .target_residency = 300,
212 .enter = &intel_idle },
215 static struct cpuidle_state hsw_cstates[MWAIT_MAX_NUM_CSTATES] = {
219 .desc = "MWAIT 0x00",
220 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
222 .target_residency = 2,
223 .enter = &intel_idle },
226 .desc = "MWAIT 0x10",
227 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
229 .target_residency = 100,
230 .enter = &intel_idle },
233 .desc = "MWAIT 0x20",
234 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
236 .target_residency = 400,
237 .enter = &intel_idle },
240 .desc = "MWAIT 0x32",
241 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
243 .target_residency = 500,
244 .enter = &intel_idle },
247 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
251 .desc = "MWAIT 0x00",
252 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
254 .target_residency = 4,
255 .enter = &intel_idle },
258 .desc = "MWAIT 0x10",
259 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID,
261 .target_residency = 80,
262 .enter = &intel_idle },
266 .desc = "MWAIT 0x30",
267 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
269 .target_residency = 400,
270 .enter = &intel_idle },
274 .desc = "MWAIT 0x52",
275 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
277 .target_residency = 560,
278 .enter = &intel_idle },
283 * @dev: cpuidle_device
284 * @drv: cpuidle driver
285 * @index: index of cpuidle state
287 * Must be called under local_irq_disable().
289 static int intel_idle(struct cpuidle_device *dev,
290 struct cpuidle_driver *drv, int index)
292 unsigned long ecx = 1; /* break on interrupt flag */
293 struct cpuidle_state *state = &drv->states[index];
294 unsigned long eax = flg2MWAIT(state->flags);
296 int cpu = smp_processor_id();
298 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
301 * leave_mm() to avoid costly and often unnecessary wakeups
302 * for flushing the user TLB's associated with the active mm.
304 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
307 if (!(lapic_timer_reliable_states & (1 << (cstate))))
308 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
310 stop_critical_timings();
311 if (!need_resched()) {
313 __monitor((void *)¤t_thread_info()->flags, 0, 0);
319 start_critical_timings();
321 if (!(lapic_timer_reliable_states & (1 << (cstate))))
322 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
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 static const struct idle_cpu idle_cpu_hsw = {
401 .state_table = hsw_cstates,
404 #define ICPU(model, cpu) \
405 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
407 static const struct x86_cpu_id intel_idle_ids[] = {
408 ICPU(0x1a, idle_cpu_nehalem),
409 ICPU(0x1e, idle_cpu_nehalem),
410 ICPU(0x1f, idle_cpu_nehalem),
411 ICPU(0x25, idle_cpu_nehalem),
412 ICPU(0x2c, idle_cpu_nehalem),
413 ICPU(0x2e, idle_cpu_nehalem),
414 ICPU(0x1c, idle_cpu_atom),
415 ICPU(0x26, idle_cpu_lincroft),
416 ICPU(0x2f, idle_cpu_nehalem),
417 ICPU(0x2a, idle_cpu_snb),
418 ICPU(0x2d, idle_cpu_snb),
419 ICPU(0x3a, idle_cpu_ivb),
420 ICPU(0x3e, idle_cpu_ivb),
421 ICPU(0x3c, idle_cpu_hsw),
422 ICPU(0x3f, idle_cpu_hsw),
423 ICPU(0x45, idle_cpu_hsw),
426 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
431 static int intel_idle_probe(void)
433 unsigned int eax, ebx, ecx;
434 const struct x86_cpu_id *id;
436 if (max_cstate == 0) {
437 pr_debug(PREFIX "disabled\n");
441 id = x86_match_cpu(intel_idle_ids);
443 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
444 boot_cpu_data.x86 == 6)
445 pr_debug(PREFIX "does not run on family %d model %d\n",
446 boot_cpu_data.x86, boot_cpu_data.x86_model);
450 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
453 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
455 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
456 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
460 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
462 icpu = (const struct idle_cpu *)id->driver_data;
463 cpuidle_state_table = icpu->state_table;
465 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
466 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
468 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
470 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
471 " model 0x%X\n", boot_cpu_data.x86_model);
473 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
474 lapic_timer_reliable_states);
479 * intel_idle_cpuidle_devices_uninit()
480 * unregister, free cpuidle_devices
482 static void intel_idle_cpuidle_devices_uninit(void)
485 struct cpuidle_device *dev;
487 for_each_online_cpu(i) {
488 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
489 cpuidle_unregister_device(dev);
492 free_percpu(intel_idle_cpuidle_devices);
496 * intel_idle_cpuidle_driver_init()
497 * allocate, initialize cpuidle_states
499 static int intel_idle_cpuidle_driver_init(void)
502 struct cpuidle_driver *drv = &intel_idle_driver;
504 drv->state_count = 1;
506 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
509 if (cstate > max_cstate) {
510 printk(PREFIX "max_cstate %d reached\n",
515 /* does the state exist in CPUID.MWAIT? */
516 num_substates = (mwait_substates >> ((cstate) * 4))
517 & MWAIT_SUBSTATE_MASK;
518 if (num_substates == 0)
520 /* is the state not enabled? */
521 if (cpuidle_state_table[cstate].enter == NULL) {
522 /* does the driver not know about the state? */
523 if (*cpuidle_state_table[cstate].name == '\0')
524 pr_debug(PREFIX "unaware of model 0x%x"
526 " contact lenb@kernel.org\n",
527 boot_cpu_data.x86_model, cstate);
532 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
533 mark_tsc_unstable("TSC halts in idle"
534 " states deeper than C2");
536 drv->states[drv->state_count] = /* structure copy */
537 cpuidle_state_table[cstate];
539 drv->state_count += 1;
542 if (icpu->auto_demotion_disable_flags)
543 on_each_cpu(auto_demotion_disable, NULL, 1);
550 * intel_idle_cpu_init()
551 * allocate, initialize, register cpuidle_devices
552 * @cpu: cpu/core to initialize
554 static int intel_idle_cpu_init(int cpu)
557 struct cpuidle_device *dev;
559 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
561 dev->state_count = 1;
563 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
566 if (cstate > max_cstate) {
567 printk(PREFIX "max_cstate %d reached\n", max_cstate);
571 /* does the state exist in CPUID.MWAIT? */
572 num_substates = (mwait_substates >> ((cstate) * 4))
573 & MWAIT_SUBSTATE_MASK;
574 if (num_substates == 0)
576 /* is the state not enabled? */
577 if (cpuidle_state_table[cstate].enter == NULL)
580 dev->state_count += 1;
585 if (cpuidle_register_device(dev)) {
586 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
587 intel_idle_cpuidle_devices_uninit();
591 if (icpu->auto_demotion_disable_flags)
592 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
597 static int __init intel_idle_init(void)
601 /* Do not load intel_idle at all for now if idle= is passed */
602 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
605 retval = intel_idle_probe();
609 intel_idle_cpuidle_driver_init();
610 retval = cpuidle_register_driver(&intel_idle_driver);
612 struct cpuidle_driver *drv = cpuidle_get_driver();
613 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
614 drv ? drv->name : "none");
618 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
619 if (intel_idle_cpuidle_devices == NULL)
622 for_each_online_cpu(i) {
623 retval = intel_idle_cpu_init(i);
625 cpuidle_unregister_driver(&intel_idle_driver);
629 register_cpu_notifier(&cpu_hotplug_notifier);
634 static void __exit intel_idle_exit(void)
636 intel_idle_cpuidle_devices_uninit();
637 cpuidle_unregister_driver(&intel_idle_driver);
640 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
641 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
642 unregister_cpu_notifier(&cpu_hotplug_notifier);
647 module_init(intel_idle_init);
648 module_exit(intel_idle_exit);
650 module_param(max_cstate, int, 0444);
652 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
653 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
654 MODULE_LICENSE("GPL");