RS485: fix inconsistencies in the meaning of some variables
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / idle / intel_idle.c
blob5d2f8e13cf0e670e83b48b6dc243402d781b9758
1 /*
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
14 * more details.
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.
29 * Design Assumptions
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
38 * Known limitations
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 */
54 #define DEBUG
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/mwait.h>
66 #include <asm/msr.h>
68 #define INTEL_IDLE_VERSION "0.4"
69 #define PREFIX "intel_idle: "
71 static struct cpuidle_driver intel_idle_driver = {
72 .name = "intel_idle",
73 .owner = THIS_MODULE,
75 /* intel_idle.max_cstate=0 disables driver */
76 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
78 static unsigned int mwait_substates;
80 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
81 /* Reliable LAPIC Timer States, bit 1 for C1 etc. */
82 static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
84 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
85 static int intel_idle(struct cpuidle_device *dev,
86 struct cpuidle_driver *drv, int index);
88 static struct cpuidle_state *cpuidle_state_table;
91 * Hardware C-state auto-demotion may not always be optimal.
92 * Indicate which enable bits to clear here.
94 static unsigned long long auto_demotion_disable_flags;
97 * Set this flag for states where the HW flushes the TLB for us
98 * and so we don't need cross-calls to keep it consistent.
99 * If this flag is set, SW flushes the TLB, so even if the
100 * HW doesn't do the flushing, this flag is safe to use.
102 #define CPUIDLE_FLAG_TLB_FLUSHED 0x10000
105 * States are indexed by the cstate number,
106 * which is also the index into the MWAIT hint array.
107 * Thus C0 is a dummy.
109 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
110 { /* MWAIT C0 */ },
111 { /* MWAIT C1 */
112 .name = "C1-NHM",
113 .desc = "MWAIT 0x00",
114 .flags = CPUIDLE_FLAG_TIME_VALID,
115 .exit_latency = 3,
116 .target_residency = 6,
117 .enter = &intel_idle },
118 { /* MWAIT C2 */
119 .name = "C3-NHM",
120 .desc = "MWAIT 0x10",
121 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
122 .exit_latency = 20,
123 .target_residency = 80,
124 .enter = &intel_idle },
125 { /* MWAIT C3 */
126 .name = "C6-NHM",
127 .desc = "MWAIT 0x20",
128 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
129 .exit_latency = 200,
130 .target_residency = 800,
131 .enter = &intel_idle },
134 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
135 { /* MWAIT C0 */ },
136 { /* MWAIT C1 */
137 .name = "C1-SNB",
138 .desc = "MWAIT 0x00",
139 .flags = CPUIDLE_FLAG_TIME_VALID,
140 .exit_latency = 1,
141 .target_residency = 1,
142 .enter = &intel_idle },
143 { /* MWAIT C2 */
144 .name = "C3-SNB",
145 .desc = "MWAIT 0x10",
146 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
147 .exit_latency = 80,
148 .target_residency = 211,
149 .enter = &intel_idle },
150 { /* MWAIT C3 */
151 .name = "C6-SNB",
152 .desc = "MWAIT 0x20",
153 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
154 .exit_latency = 104,
155 .target_residency = 345,
156 .enter = &intel_idle },
157 { /* MWAIT C4 */
158 .name = "C7-SNB",
159 .desc = "MWAIT 0x30",
160 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
161 .exit_latency = 109,
162 .target_residency = 345,
163 .enter = &intel_idle },
166 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
167 { /* MWAIT C0 */ },
168 { /* MWAIT C1 */
169 .name = "C1-ATM",
170 .desc = "MWAIT 0x00",
171 .flags = CPUIDLE_FLAG_TIME_VALID,
172 .exit_latency = 1,
173 .target_residency = 4,
174 .enter = &intel_idle },
175 { /* MWAIT C2 */
176 .name = "C2-ATM",
177 .desc = "MWAIT 0x10",
178 .flags = CPUIDLE_FLAG_TIME_VALID,
179 .exit_latency = 20,
180 .target_residency = 80,
181 .enter = &intel_idle },
182 { /* MWAIT C3 */ },
183 { /* MWAIT C4 */
184 .name = "C4-ATM",
185 .desc = "MWAIT 0x30",
186 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
187 .exit_latency = 100,
188 .target_residency = 400,
189 .enter = &intel_idle },
190 { /* MWAIT C5 */ },
191 { /* MWAIT C6 */
192 .name = "C6-ATM",
193 .desc = "MWAIT 0x52",
194 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
195 .exit_latency = 140,
196 .target_residency = 560,
197 .enter = &intel_idle },
200 static int get_driver_data(int cstate)
202 int driver_data;
203 switch (cstate) {
205 case 1: /* MWAIT C1 */
206 driver_data = 0x00;
207 break;
208 case 2: /* MWAIT C2 */
209 driver_data = 0x10;
210 break;
211 case 3: /* MWAIT C3 */
212 driver_data = 0x20;
213 break;
214 case 4: /* MWAIT C4 */
215 driver_data = 0x30;
216 break;
217 case 5: /* MWAIT C5 */
218 driver_data = 0x40;
219 break;
220 case 6: /* MWAIT C6 */
221 driver_data = 0x52;
222 break;
223 default:
224 driver_data = 0x00;
226 return driver_data;
230 * intel_idle
231 * @dev: cpuidle_device
232 * @drv: cpuidle driver
233 * @index: index of cpuidle state
236 static int intel_idle(struct cpuidle_device *dev,
237 struct cpuidle_driver *drv, int index)
239 unsigned long ecx = 1; /* break on interrupt flag */
240 struct cpuidle_state *state = &drv->states[index];
241 struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
242 unsigned long eax = (unsigned long)cpuidle_get_statedata(state_usage);
243 unsigned int cstate;
244 ktime_t kt_before, kt_after;
245 s64 usec_delta;
246 int cpu = smp_processor_id();
248 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
250 local_irq_disable();
253 * leave_mm() to avoid costly and often unnecessary wakeups
254 * for flushing the user TLB's associated with the active mm.
256 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
257 leave_mm(cpu);
259 if (!(lapic_timer_reliable_states & (1 << (cstate))))
260 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
262 kt_before = ktime_get_real();
264 stop_critical_timings();
265 if (!need_resched()) {
267 __monitor((void *)&current_thread_info()->flags, 0, 0);
268 smp_mb();
269 if (!need_resched())
270 __mwait(eax, ecx);
273 start_critical_timings();
275 kt_after = ktime_get_real();
276 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
278 local_irq_enable();
280 if (!(lapic_timer_reliable_states & (1 << (cstate))))
281 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
283 /* Update cpuidle counters */
284 dev->last_residency = (int)usec_delta;
286 return index;
289 static void __setup_broadcast_timer(void *arg)
291 unsigned long reason = (unsigned long)arg;
292 int cpu = smp_processor_id();
294 reason = reason ?
295 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
297 clockevents_notify(reason, &cpu);
300 static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
301 unsigned long action, void *hcpu)
303 int hotcpu = (unsigned long)hcpu;
305 switch (action & 0xf) {
306 case CPU_ONLINE:
307 smp_call_function_single(hotcpu, __setup_broadcast_timer,
308 (void *)true, 1);
309 break;
311 return NOTIFY_OK;
314 static struct notifier_block setup_broadcast_notifier = {
315 .notifier_call = setup_broadcast_cpuhp_notify,
318 static void auto_demotion_disable(void *dummy)
320 unsigned long long msr_bits;
322 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
323 msr_bits &= ~auto_demotion_disable_flags;
324 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
328 * intel_idle_probe()
330 static int intel_idle_probe(void)
332 unsigned int eax, ebx, ecx;
334 if (max_cstate == 0) {
335 pr_debug(PREFIX "disabled\n");
336 return -EPERM;
339 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
340 return -ENODEV;
342 if (!boot_cpu_has(X86_FEATURE_MWAIT))
343 return -ENODEV;
345 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
346 return -ENODEV;
348 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
350 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
351 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
352 return -ENODEV;
354 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
357 if (boot_cpu_data.x86 != 6) /* family 6 */
358 return -ENODEV;
360 switch (boot_cpu_data.x86_model) {
362 case 0x1A: /* Core i7, Xeon 5500 series */
363 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
364 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
365 case 0x2E: /* Nehalem-EX Xeon */
366 case 0x2F: /* Westmere-EX Xeon */
367 case 0x25: /* Westmere */
368 case 0x2C: /* Westmere */
369 cpuidle_state_table = nehalem_cstates;
370 auto_demotion_disable_flags =
371 (NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE);
372 break;
374 case 0x1C: /* 28 - Atom Processor */
375 cpuidle_state_table = atom_cstates;
376 break;
378 case 0x26: /* 38 - Lincroft Atom Processor */
379 cpuidle_state_table = atom_cstates;
380 auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE;
381 break;
383 case 0x2A: /* SNB */
384 case 0x2D: /* SNB Xeon */
385 cpuidle_state_table = snb_cstates;
386 break;
388 default:
389 pr_debug(PREFIX "does not run on family %d model %d\n",
390 boot_cpu_data.x86, boot_cpu_data.x86_model);
391 return -ENODEV;
394 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
395 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
396 else {
397 smp_call_function(__setup_broadcast_timer, (void *)true, 1);
398 register_cpu_notifier(&setup_broadcast_notifier);
401 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
402 " model 0x%X\n", boot_cpu_data.x86_model);
404 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
405 lapic_timer_reliable_states);
406 return 0;
410 * intel_idle_cpuidle_devices_uninit()
411 * unregister, free cpuidle_devices
413 static void intel_idle_cpuidle_devices_uninit(void)
415 int i;
416 struct cpuidle_device *dev;
418 for_each_online_cpu(i) {
419 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
420 cpuidle_unregister_device(dev);
423 free_percpu(intel_idle_cpuidle_devices);
424 return;
427 * intel_idle_cpuidle_driver_init()
428 * allocate, initialize cpuidle_states
430 static int intel_idle_cpuidle_driver_init(void)
432 int cstate;
433 struct cpuidle_driver *drv = &intel_idle_driver;
435 drv->state_count = 1;
437 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
438 int num_substates;
440 if (cstate > max_cstate) {
441 printk(PREFIX "max_cstate %d reached\n",
442 max_cstate);
443 break;
446 /* does the state exist in CPUID.MWAIT? */
447 num_substates = (mwait_substates >> ((cstate) * 4))
448 & MWAIT_SUBSTATE_MASK;
449 if (num_substates == 0)
450 continue;
451 /* is the state not enabled? */
452 if (cpuidle_state_table[cstate].enter == NULL) {
453 /* does the driver not know about the state? */
454 if (*cpuidle_state_table[cstate].name == '\0')
455 pr_debug(PREFIX "unaware of model 0x%x"
456 " MWAIT %d please"
457 " contact lenb@kernel.org",
458 boot_cpu_data.x86_model, cstate);
459 continue;
462 if ((cstate > 2) &&
463 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
464 mark_tsc_unstable("TSC halts in idle"
465 " states deeper than C2");
467 drv->states[drv->state_count] = /* structure copy */
468 cpuidle_state_table[cstate];
470 drv->state_count += 1;
473 if (auto_demotion_disable_flags)
474 smp_call_function(auto_demotion_disable, NULL, 1);
476 return 0;
481 * intel_idle_cpuidle_devices_init()
482 * allocate, initialize, register cpuidle_devices
484 static int intel_idle_cpuidle_devices_init(void)
486 int i, cstate;
487 struct cpuidle_device *dev;
489 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
490 if (intel_idle_cpuidle_devices == NULL)
491 return -ENOMEM;
493 for_each_online_cpu(i) {
494 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
496 dev->state_count = 1;
498 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
499 int num_substates;
501 if (cstate > max_cstate) {
502 printk(PREFIX "max_cstate %d reached\n",
503 max_cstate);
504 break;
507 /* does the state exist in CPUID.MWAIT? */
508 num_substates = (mwait_substates >> ((cstate) * 4))
509 & MWAIT_SUBSTATE_MASK;
510 if (num_substates == 0)
511 continue;
512 /* is the state not enabled? */
513 if (cpuidle_state_table[cstate].enter == NULL) {
514 continue;
517 dev->states_usage[dev->state_count].driver_data =
518 (void *)get_driver_data(cstate);
520 dev->state_count += 1;
523 dev->cpu = i;
524 if (cpuidle_register_device(dev)) {
525 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
527 intel_idle_cpuidle_devices_uninit();
528 return -EIO;
532 return 0;
536 static int __init intel_idle_init(void)
538 int retval;
540 /* Do not load intel_idle at all for now if idle= is passed */
541 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
542 return -ENODEV;
544 retval = intel_idle_probe();
545 if (retval)
546 return retval;
548 intel_idle_cpuidle_driver_init();
549 retval = cpuidle_register_driver(&intel_idle_driver);
550 if (retval) {
551 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
552 cpuidle_get_driver()->name);
553 return retval;
556 retval = intel_idle_cpuidle_devices_init();
557 if (retval) {
558 cpuidle_unregister_driver(&intel_idle_driver);
559 return retval;
562 return 0;
565 static void __exit intel_idle_exit(void)
567 intel_idle_cpuidle_devices_uninit();
568 cpuidle_unregister_driver(&intel_idle_driver);
570 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
571 smp_call_function(__setup_broadcast_timer, (void *)false, 1);
572 unregister_cpu_notifier(&setup_broadcast_notifier);
575 return;
578 module_init(intel_idle_init);
579 module_exit(intel_idle_exit);
581 module_param(max_cstate, int, 0444);
583 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
584 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
585 MODULE_LICENSE("GPL");