2 * intel_idle.c - native hardware idle loop for modern Intel processors
4 * Copyright (c) 2010, Intel Corporation.
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 <asm/mwait.h>
66 #define INTEL_IDLE_VERSION "0.4"
67 #define PREFIX "intel_idle: "
69 static struct cpuidle_driver intel_idle_driver = {
73 /* intel_idle.max_cstate=0 disables driver */
74 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
76 static unsigned int mwait_substates;
78 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
79 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */
80 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
82 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
83 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
85 static struct cpuidle_state *cpuidle_state_table;
88 * Set this flag for states where the HW flushes the TLB for us
89 * and so we don't need cross-calls to keep it consistent.
90 * If this flag is set, SW flushes the TLB, so even if the
91 * HW doesn't do the flushing, this flag is safe to use.
93 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
96 * States are indexed by the cstate number,
97 * which is also the index into the MWAIT hint array.
100 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
104 .desc = "MWAIT 0x00",
105 .driver_data = (void *) 0x00,
106 .flags = CPUIDLE_FLAG_TIME_VALID,
108 .target_residency = 6,
109 .enter = &intel_idle },
112 .desc = "MWAIT 0x10",
113 .driver_data = (void *) 0x10,
114 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
116 .target_residency = 80,
117 .enter = &intel_idle },
120 .desc = "MWAIT 0x20",
121 .driver_data = (void *) 0x20,
122 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
124 .target_residency = 800,
125 .enter = &intel_idle },
128 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
132 .desc = "MWAIT 0x00",
133 .driver_data = (void *) 0x00,
134 .flags = CPUIDLE_FLAG_TIME_VALID,
136 .target_residency = 1,
137 .enter = &intel_idle },
140 .desc = "MWAIT 0x10",
141 .driver_data = (void *) 0x10,
142 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
144 .target_residency = 211,
145 .enter = &intel_idle },
148 .desc = "MWAIT 0x20",
149 .driver_data = (void *) 0x20,
150 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
152 .target_residency = 345,
153 .enter = &intel_idle },
156 .desc = "MWAIT 0x30",
157 .driver_data = (void *) 0x30,
158 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
160 .target_residency = 345,
161 .enter = &intel_idle },
164 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
168 .desc = "MWAIT 0x00",
169 .driver_data = (void *) 0x00,
170 .flags = CPUIDLE_FLAG_TIME_VALID,
172 .target_residency = 4,
173 .enter = &intel_idle },
176 .desc = "MWAIT 0x10",
177 .driver_data = (void *) 0x10,
178 .flags = CPUIDLE_FLAG_TIME_VALID,
180 .target_residency = 80,
181 .enter = &intel_idle },
185 .desc = "MWAIT 0x30",
186 .driver_data = (void *) 0x30,
187 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
189 .target_residency = 400,
190 .enter = &intel_idle },
194 .desc = "MWAIT 0x52",
195 .driver_data = (void *) 0x52,
196 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
198 .target_residency = 560,
199 .enter = &intel_idle },
204 * @dev: cpuidle_device
205 * @state: cpuidle state
208 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
210 unsigned long ecx = 1; /* break on interrupt flag */
211 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
213 ktime_t kt_before, kt_after;
215 int cpu = smp_processor_id();
217 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
222 * leave_mm() to avoid costly and often unnecessary wakeups
223 * for flushing the user TLB's associated with the active mm.
225 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
228 if (!(lapic_timer_reliable_states & (1 << (cstate))))
229 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
231 kt_before = ktime_get_real();
233 stop_critical_timings();
234 if (!need_resched()) {
236 __monitor((void *)¤t_thread_info()->flags, 0, 0);
242 start_critical_timings();
244 kt_after = ktime_get_real();
245 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
249 if (!(lapic_timer_reliable_states & (1 << (cstate))))
250 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
255 static void __setup_broadcast_timer(void *arg)
257 unsigned long reason = (unsigned long)arg;
258 int cpu = smp_processor_id();
261 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
263 clockevents_notify(reason, &cpu);
266 static int __cpuinit setup_broadcast_cpuhp_notify(struct notifier_block *n,
267 unsigned long action, void *hcpu)
269 int hotcpu = (unsigned long)hcpu;
271 switch (action & 0xf) {
273 smp_call_function_single(hotcpu, __setup_broadcast_timer,
276 case CPU_DOWN_PREPARE:
277 smp_call_function_single(hotcpu, __setup_broadcast_timer,
284 static struct notifier_block __cpuinitdata setup_broadcast_notifier = {
285 .notifier_call = setup_broadcast_cpuhp_notify,
291 static int intel_idle_probe(void)
293 unsigned int eax, ebx, ecx;
295 if (max_cstate == 0) {
296 pr_debug(PREFIX "disabled\n");
300 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
303 if (!boot_cpu_has(X86_FEATURE_MWAIT))
306 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
309 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
311 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
312 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
315 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
318 if (boot_cpu_data.x86 != 6) /* family 6 */
321 switch (boot_cpu_data.x86_model) {
323 case 0x1A: /* Core i7, Xeon 5500 series */
324 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
325 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
326 case 0x2E: /* Nehalem-EX Xeon */
327 case 0x2F: /* Westmere-EX Xeon */
328 case 0x25: /* Westmere */
329 case 0x2C: /* Westmere */
330 cpuidle_state_table = nehalem_cstates;
333 case 0x1C: /* 28 - Atom Processor */
334 case 0x26: /* 38 - Lincroft Atom Processor */
335 cpuidle_state_table = atom_cstates;
339 case 0x2D: /* SNB Xeon */
340 cpuidle_state_table = snb_cstates;
344 pr_debug(PREFIX "does not run on family %d model %d\n",
345 boot_cpu_data.x86, boot_cpu_data.x86_model);
349 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
350 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
352 smp_call_function(__setup_broadcast_timer, (void *)true, 1);
353 register_cpu_notifier(&setup_broadcast_notifier);
356 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
357 " model 0x%X\n", boot_cpu_data.x86_model);
359 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
360 lapic_timer_reliable_states);
365 * intel_idle_cpuidle_devices_uninit()
366 * unregister, free cpuidle_devices
368 static void intel_idle_cpuidle_devices_uninit(void)
371 struct cpuidle_device *dev;
373 for_each_online_cpu(i) {
374 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
375 cpuidle_unregister_device(dev);
378 free_percpu(intel_idle_cpuidle_devices);
382 * intel_idle_cpuidle_devices_init()
383 * allocate, initialize, register cpuidle_devices
385 static int intel_idle_cpuidle_devices_init(void)
388 struct cpuidle_device *dev;
390 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
391 if (intel_idle_cpuidle_devices == NULL)
394 for_each_online_cpu(i) {
395 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
397 dev->state_count = 1;
399 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
402 if (cstate > max_cstate) {
403 printk(PREFIX "max_cstate %d reached\n",
408 /* does the state exist in CPUID.MWAIT? */
409 num_substates = (mwait_substates >> ((cstate) * 4))
410 & MWAIT_SUBSTATE_MASK;
411 if (num_substates == 0)
413 /* is the state not enabled? */
414 if (cpuidle_state_table[cstate].enter == NULL) {
415 /* does the driver not know about the state? */
416 if (*cpuidle_state_table[cstate].name == '\0')
417 pr_debug(PREFIX "unaware of model 0x%x"
420 boot_cpu_data.x86_model, cstate);
425 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
426 mark_tsc_unstable("TSC halts in idle"
427 " states deeper than C2");
429 dev->states[dev->state_count] = /* structure copy */
430 cpuidle_state_table[cstate];
432 dev->state_count += 1;
436 if (cpuidle_register_device(dev)) {
437 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
439 intel_idle_cpuidle_devices_uninit();
448 static int __init intel_idle_init(void)
452 /* Do not load intel_idle at all for now if idle= is passed */
453 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
456 retval = intel_idle_probe();
460 retval = cpuidle_register_driver(&intel_idle_driver);
462 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
463 cpuidle_get_driver()->name);
467 retval = intel_idle_cpuidle_devices_init();
469 cpuidle_unregister_driver(&intel_idle_driver);
476 static void __exit intel_idle_exit(void)
478 intel_idle_cpuidle_devices_uninit();
479 cpuidle_unregister_driver(&intel_idle_driver);
481 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
482 smp_call_function(__setup_broadcast_timer, (void *)false, 1);
483 unregister_cpu_notifier(&setup_broadcast_notifier);
489 module_init(intel_idle_init);
490 module_exit(intel_idle_exit);
492 module_param(max_cstate, int, 0444);
495 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
496 MODULE_LICENSE("GPL");