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 <asm/mwait.h>
67 #define INTEL_IDLE_VERSION "0.4"
68 #define PREFIX "intel_idle: "
70 static struct cpuidle_driver intel_idle_driver = {
74 /* intel_idle.max_cstate=0 disables driver */
75 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
77 static unsigned int mwait_substates;
79 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
80 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */
81 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
83 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
84 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
86 static struct cpuidle_state *cpuidle_state_table;
89 * Hardware C-state auto-demotion may not always be optimal.
90 * Indicate which enable bits to clear here.
92 static unsigned long long auto_demotion_disable_flags;
95 * Set this flag for states where the HW flushes the TLB for us
96 * and so we don't need cross-calls to keep it consistent.
97 * If this flag is set, SW flushes the TLB, so even if the
98 * HW doesn't do the flushing, this flag is safe to use.
100 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
103 * States are indexed by the cstate number,
104 * which is also the index into the MWAIT hint array.
105 * Thus C0 is a dummy.
107 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
111 .desc = "MWAIT 0x00",
112 .driver_data = (void *) 0x00,
113 .flags = CPUIDLE_FLAG_TIME_VALID,
115 .target_residency = 6,
116 .enter = &intel_idle },
119 .desc = "MWAIT 0x10",
120 .driver_data = (void *) 0x10,
121 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
123 .target_residency = 80,
124 .enter = &intel_idle },
127 .desc = "MWAIT 0x20",
128 .driver_data = (void *) 0x20,
129 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
131 .target_residency = 800,
132 .enter = &intel_idle },
135 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
139 .desc = "MWAIT 0x00",
140 .driver_data = (void *) 0x00,
141 .flags = CPUIDLE_FLAG_TIME_VALID,
143 .target_residency = 1,
144 .enter = &intel_idle },
147 .desc = "MWAIT 0x10",
148 .driver_data = (void *) 0x10,
149 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
151 .target_residency = 211,
152 .enter = &intel_idle },
155 .desc = "MWAIT 0x20",
156 .driver_data = (void *) 0x20,
157 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
159 .target_residency = 345,
160 .enter = &intel_idle },
163 .desc = "MWAIT 0x30",
164 .driver_data = (void *) 0x30,
165 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
167 .target_residency = 345,
168 .enter = &intel_idle },
171 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
175 .desc = "MWAIT 0x00",
176 .driver_data = (void *) 0x00,
177 .flags = CPUIDLE_FLAG_TIME_VALID,
179 .target_residency = 4,
180 .enter = &intel_idle },
183 .desc = "MWAIT 0x10",
184 .driver_data = (void *) 0x10,
185 .flags = CPUIDLE_FLAG_TIME_VALID,
187 .target_residency = 80,
188 .enter = &intel_idle },
192 .desc = "MWAIT 0x30",
193 .driver_data = (void *) 0x30,
194 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
196 .target_residency = 400,
197 .enter = &intel_idle },
201 .desc = "MWAIT 0x52",
202 .driver_data = (void *) 0x52,
203 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
205 .target_residency = 560,
206 .enter = &intel_idle },
211 * @dev: cpuidle_device
212 * @state: cpuidle state
215 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
217 unsigned long ecx = 1; /* break on interrupt flag */
218 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
220 ktime_t kt_before, kt_after;
222 int cpu = smp_processor_id();
224 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
229 * leave_mm() to avoid costly and often unnecessary wakeups
230 * for flushing the user TLB's associated with the active mm.
232 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
235 if (!(lapic_timer_reliable_states & (1 << (cstate))))
236 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
238 kt_before = ktime_get_real();
240 stop_critical_timings();
241 if (!need_resched()) {
243 __monitor((void *)¤t_thread_info()->flags, 0, 0);
249 start_critical_timings();
251 kt_after = ktime_get_real();
252 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
256 if (!(lapic_timer_reliable_states & (1 << (cstate))))
257 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
262 static void __setup_broadcast_timer(void *arg)
264 unsigned long reason = (unsigned long)arg;
265 int cpu = smp_processor_id();
268 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
270 clockevents_notify(reason, &cpu);
273 static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
274 unsigned long action, void *hcpu)
276 int hotcpu = (unsigned long)hcpu;
278 switch (action & 0xf) {
280 smp_call_function_single(hotcpu, __setup_broadcast_timer,
287 static struct notifier_block setup_broadcast_notifier = {
288 .notifier_call = setup_broadcast_cpuhp_notify,
291 static void auto_demotion_disable(void *dummy)
293 unsigned long long msr_bits;
295 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
296 msr_bits &= ~auto_demotion_disable_flags;
297 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
303 static int intel_idle_probe(void)
305 unsigned int eax, ebx, ecx;
307 if (max_cstate == 0) {
308 pr_debug(PREFIX "disabled\n");
312 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
315 if (!boot_cpu_has(X86_FEATURE_MWAIT))
318 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
321 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
323 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
324 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
327 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
330 if (boot_cpu_data.x86 != 6) /* family 6 */
333 switch (boot_cpu_data.x86_model) {
335 case 0x1A: /* Core i7, Xeon 5500 series */
336 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
337 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
338 case 0x2E: /* Nehalem-EX Xeon */
339 case 0x2F: /* Westmere-EX Xeon */
340 case 0x25: /* Westmere */
341 case 0x2C: /* Westmere */
342 cpuidle_state_table = nehalem_cstates;
343 auto_demotion_disable_flags =
344 (NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE);
347 case 0x1C: /* 28 - Atom Processor */
348 case 0x26: /* 38 - Lincroft Atom Processor */
349 cpuidle_state_table = atom_cstates;
353 case 0x2D: /* SNB Xeon */
354 cpuidle_state_table = snb_cstates;
358 pr_debug(PREFIX "does not run on family %d model %d\n",
359 boot_cpu_data.x86, boot_cpu_data.x86_model);
363 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
364 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
366 smp_call_function(__setup_broadcast_timer, (void *)true, 1);
367 register_cpu_notifier(&setup_broadcast_notifier);
370 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
371 " model 0x%X\n", boot_cpu_data.x86_model);
373 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
374 lapic_timer_reliable_states);
379 * intel_idle_cpuidle_devices_uninit()
380 * unregister, free cpuidle_devices
382 static void intel_idle_cpuidle_devices_uninit(void)
385 struct cpuidle_device *dev;
387 for_each_online_cpu(i) {
388 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
389 cpuidle_unregister_device(dev);
392 free_percpu(intel_idle_cpuidle_devices);
396 * intel_idle_cpuidle_devices_init()
397 * allocate, initialize, register cpuidle_devices
399 static int intel_idle_cpuidle_devices_init(void)
402 struct cpuidle_device *dev;
404 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
405 if (intel_idle_cpuidle_devices == NULL)
408 for_each_online_cpu(i) {
409 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
411 dev->state_count = 1;
413 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
416 if (cstate > max_cstate) {
417 printk(PREFIX "max_cstate %d reached\n",
422 /* does the state exist in CPUID.MWAIT? */
423 num_substates = (mwait_substates >> ((cstate) * 4))
424 & MWAIT_SUBSTATE_MASK;
425 if (num_substates == 0)
427 /* is the state not enabled? */
428 if (cpuidle_state_table[cstate].enter == NULL) {
429 /* does the driver not know about the state? */
430 if (*cpuidle_state_table[cstate].name == '\0')
431 pr_debug(PREFIX "unaware of model 0x%x"
433 " contact lenb@kernel.org",
434 boot_cpu_data.x86_model, cstate);
439 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
440 mark_tsc_unstable("TSC halts in idle"
441 " states deeper than C2");
443 dev->states[dev->state_count] = /* structure copy */
444 cpuidle_state_table[cstate];
446 dev->state_count += 1;
450 if (cpuidle_register_device(dev)) {
451 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
453 intel_idle_cpuidle_devices_uninit();
457 if (auto_demotion_disable_flags)
458 smp_call_function(auto_demotion_disable, NULL, 1);
464 static int __init intel_idle_init(void)
468 /* Do not load intel_idle at all for now if idle= is passed */
469 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
472 retval = intel_idle_probe();
476 retval = cpuidle_register_driver(&intel_idle_driver);
478 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
479 cpuidle_get_driver()->name);
483 retval = intel_idle_cpuidle_devices_init();
485 cpuidle_unregister_driver(&intel_idle_driver);
492 static void __exit intel_idle_exit(void)
494 intel_idle_cpuidle_devices_uninit();
495 cpuidle_unregister_driver(&intel_idle_driver);
497 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
498 smp_call_function(__setup_broadcast_timer, (void *)false, 1);
499 unregister_cpu_notifier(&setup_broadcast_notifier);
505 module_init(intel_idle_init);
506 module_exit(intel_idle_exit);
508 module_param(max_cstate, int, 0444);
510 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
511 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
512 MODULE_LICENSE("GPL");