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 <asm/mwait.h>
64 #define INTEL_IDLE_VERSION "0.4"
65 #define PREFIX "intel_idle: "
67 static struct cpuidle_driver intel_idle_driver = {
71 /* intel_idle.max_cstate=0 disables driver */
72 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
74 static unsigned int mwait_substates;
76 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */
77 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
79 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
80 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
82 static struct cpuidle_state *cpuidle_state_table;
85 * States are indexed by the cstate number,
86 * which is also the index into the MWAIT hint array.
89 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
94 .driver_data = (void *) 0x00,
95 .flags = CPUIDLE_FLAG_TIME_VALID,
97 .target_residency = 6,
98 .enter = &intel_idle },
101 .desc = "MWAIT 0x10",
102 .driver_data = (void *) 0x10,
103 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
105 .target_residency = 80,
106 .enter = &intel_idle },
109 .desc = "MWAIT 0x20",
110 .driver_data = (void *) 0x20,
111 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
113 .target_residency = 800,
114 .enter = &intel_idle },
117 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
121 .desc = "MWAIT 0x00",
122 .driver_data = (void *) 0x00,
123 .flags = CPUIDLE_FLAG_TIME_VALID,
125 .target_residency = 4,
126 .enter = &intel_idle },
129 .desc = "MWAIT 0x10",
130 .driver_data = (void *) 0x10,
131 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
133 .target_residency = 160,
134 .enter = &intel_idle },
137 .desc = "MWAIT 0x20",
138 .driver_data = (void *) 0x20,
139 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
141 .target_residency = 208,
142 .enter = &intel_idle },
145 .desc = "MWAIT 0x30",
146 .driver_data = (void *) 0x30,
147 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
149 .target_residency = 300,
150 .enter = &intel_idle },
153 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
157 .desc = "MWAIT 0x00",
158 .driver_data = (void *) 0x00,
159 .flags = CPUIDLE_FLAG_TIME_VALID,
161 .target_residency = 4,
162 .enter = &intel_idle },
165 .desc = "MWAIT 0x10",
166 .driver_data = (void *) 0x10,
167 .flags = CPUIDLE_FLAG_TIME_VALID,
169 .target_residency = 80,
170 .enter = &intel_idle },
174 .desc = "MWAIT 0x30",
175 .driver_data = (void *) 0x30,
176 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
178 .target_residency = 400,
179 .enter = &intel_idle },
183 .desc = "MWAIT 0x52",
184 .driver_data = (void *) 0x52,
185 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
187 .target_residency = 560,
188 .enter = &intel_idle },
193 * @dev: cpuidle_device
194 * @state: cpuidle state
197 static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
199 unsigned long ecx = 1; /* break on interrupt flag */
200 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
202 ktime_t kt_before, kt_after;
204 int cpu = smp_processor_id();
206 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
211 * leave_mm() to avoid costly and often unnecessary wakeups
212 * for flushing the user TLB's associated with the active mm.
214 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
217 if (!(lapic_timer_reliable_states & (1 << (cstate))))
218 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
220 kt_before = ktime_get_real();
222 stop_critical_timings();
223 trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu);
224 trace_cpu_idle((eax >> 4) + 1, cpu);
225 if (!need_resched()) {
227 __monitor((void *)¤t_thread_info()->flags, 0, 0);
233 start_critical_timings();
235 kt_after = ktime_get_real();
236 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
240 if (!(lapic_timer_reliable_states & (1 << (cstate))))
241 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
249 static int intel_idle_probe(void)
251 unsigned int eax, ebx, ecx;
253 if (max_cstate == 0) {
254 pr_debug(PREFIX "disabled\n");
258 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
261 if (!boot_cpu_has(X86_FEATURE_MWAIT))
264 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
267 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
269 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
270 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
273 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
276 if (boot_cpu_data.x86 != 6) /* family 6 */
279 switch (boot_cpu_data.x86_model) {
281 case 0x1A: /* Core i7, Xeon 5500 series */
282 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
283 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
284 case 0x2E: /* Nehalem-EX Xeon */
285 case 0x2F: /* Westmere-EX Xeon */
286 case 0x25: /* Westmere */
287 case 0x2C: /* Westmere */
288 cpuidle_state_table = nehalem_cstates;
291 case 0x1C: /* 28 - Atom Processor */
292 case 0x26: /* 38 - Lincroft Atom Processor */
293 cpuidle_state_table = atom_cstates;
297 case 0x2D: /* SNB Xeon */
298 cpuidle_state_table = snb_cstates;
302 pr_debug(PREFIX "does not run on family %d model %d\n",
303 boot_cpu_data.x86, boot_cpu_data.x86_model);
307 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
308 lapic_timer_reliable_states = 0xFFFFFFFF;
310 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
311 " model 0x%X\n", boot_cpu_data.x86_model);
313 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
314 lapic_timer_reliable_states);
319 * intel_idle_cpuidle_devices_uninit()
320 * unregister, free cpuidle_devices
322 static void intel_idle_cpuidle_devices_uninit(void)
325 struct cpuidle_device *dev;
327 for_each_online_cpu(i) {
328 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
329 cpuidle_unregister_device(dev);
332 free_percpu(intel_idle_cpuidle_devices);
336 * intel_idle_cpuidle_devices_init()
337 * allocate, initialize, register cpuidle_devices
339 static int intel_idle_cpuidle_devices_init(void)
342 struct cpuidle_device *dev;
344 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
345 if (intel_idle_cpuidle_devices == NULL)
348 for_each_online_cpu(i) {
349 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
351 dev->state_count = 1;
353 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
356 if (cstate > max_cstate) {
357 printk(PREFIX "max_cstate %d reached\n",
362 /* does the state exist in CPUID.MWAIT? */
363 num_substates = (mwait_substates >> ((cstate) * 4))
364 & MWAIT_SUBSTATE_MASK;
365 if (num_substates == 0)
367 /* is the state not enabled? */
368 if (cpuidle_state_table[cstate].enter == NULL) {
369 /* does the driver not know about the state? */
370 if (*cpuidle_state_table[cstate].name == '\0')
371 pr_debug(PREFIX "unaware of model 0x%x"
373 " contact lenb@kernel.org",
374 boot_cpu_data.x86_model, cstate);
379 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
380 mark_tsc_unstable("TSC halts in idle"
381 " states deeper than C2");
383 dev->states[dev->state_count] = /* structure copy */
384 cpuidle_state_table[cstate];
386 dev->state_count += 1;
390 if (cpuidle_register_device(dev)) {
391 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
393 intel_idle_cpuidle_devices_uninit();
402 static int __init intel_idle_init(void)
406 retval = intel_idle_probe();
410 retval = cpuidle_register_driver(&intel_idle_driver);
412 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
413 cpuidle_get_driver()->name);
417 retval = intel_idle_cpuidle_devices_init();
419 cpuidle_unregister_driver(&intel_idle_driver);
426 static void __exit intel_idle_exit(void)
428 intel_idle_cpuidle_devices_uninit();
429 cpuidle_unregister_driver(&intel_idle_driver);
434 module_init(intel_idle_init);
435 module_exit(intel_idle_exit);
437 module_param(max_cstate, int, 0444);
439 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
440 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
441 MODULE_LICENSE("GPL");