]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/idle/intel_idle.c
bnx2x: Disable local BHes to prevent a dead-lock situation
[net-next-2.6.git] / drivers / idle / intel_idle.c
CommitLineData
26717172
LB
1/*
2 * intel_idle.c - native hardware idle loop for modern Intel processors
3 *
4 * Copyright (c) 2010, Intel Corporation.
5 * Len Brown <len.brown@intel.com>
6 *
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.
10 *
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.
15 *
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.
19 */
20
21/*
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.
26 */
27
28/*
29 * Design Assumptions
30 *
31 * All CPUs have same idle states as boot CPU
32 *
33 * Chipset BM_STS (bus master status) bit is a NOP
34 * for preventing entry into deep C-stats
35 */
36
37/*
38 * Known limitations
39 *
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.
44 *
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.
48 *
49 * There is currently no kernel-based automatic probing/loading mechanism
50 * if the driver is built as a module.
51 */
52
53/* un-comment DEBUG to enable pr_debug() statements */
54#define DEBUG
55
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>
bc83cccc 62#include <asm/mwait.h>
26717172
LB
63
64#define INTEL_IDLE_VERSION "0.4"
65#define PREFIX "intel_idle: "
66
26717172
LB
67static struct cpuidle_driver intel_idle_driver = {
68 .name = "intel_idle",
69 .owner = THIS_MODULE,
70};
71/* intel_idle.max_cstate=0 disables driver */
72static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
26717172 73
c4236282 74static unsigned int mwait_substates;
26717172
LB
75
76/* Reliable LAPIC Timer States, bit 1 for C1 etc. */
d13780d4 77static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
26717172 78
3265eba0 79static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
26717172
LB
80static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
81
82static struct cpuidle_state *cpuidle_state_table;
83
84/*
85 * States are indexed by the cstate number,
86 * which is also the index into the MWAIT hint array.
87 * Thus C0 is a dummy.
88 */
89static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
90 { /* MWAIT C0 */ },
91 { /* MWAIT C1 */
92 .name = "NHM-C1",
93 .desc = "MWAIT 0x00",
94 .driver_data = (void *) 0x00,
95 .flags = CPUIDLE_FLAG_TIME_VALID,
96 .exit_latency = 3,
26717172
LB
97 .target_residency = 6,
98 .enter = &intel_idle },
99 { /* MWAIT C2 */
100 .name = "NHM-C3",
101 .desc = "MWAIT 0x10",
102 .driver_data = (void *) 0x10,
6110a1f4 103 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
26717172 104 .exit_latency = 20,
26717172
LB
105 .target_residency = 80,
106 .enter = &intel_idle },
107 { /* MWAIT C3 */
108 .name = "NHM-C6",
109 .desc = "MWAIT 0x20",
110 .driver_data = (void *) 0x20,
6110a1f4 111 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
26717172 112 .exit_latency = 200,
26717172
LB
113 .target_residency = 800,
114 .enter = &intel_idle },
115};
116
d13780d4
LB
117static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
118 { /* MWAIT C0 */ },
119 { /* MWAIT C1 */
120 .name = "SNB-C1",
121 .desc = "MWAIT 0x00",
122 .driver_data = (void *) 0x00,
123 .flags = CPUIDLE_FLAG_TIME_VALID,
124 .exit_latency = 1,
125 .target_residency = 4,
126 .enter = &intel_idle },
127 { /* MWAIT C2 */
128 .name = "SNB-C3",
129 .desc = "MWAIT 0x10",
130 .driver_data = (void *) 0x10,
00527cc6 131 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
d13780d4
LB
132 .exit_latency = 80,
133 .target_residency = 160,
134 .enter = &intel_idle },
135 { /* MWAIT C3 */
136 .name = "SNB-C6",
137 .desc = "MWAIT 0x20",
138 .driver_data = (void *) 0x20,
00527cc6 139 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
d13780d4
LB
140 .exit_latency = 104,
141 .target_residency = 208,
142 .enter = &intel_idle },
143 { /* MWAIT C4 */
144 .name = "SNB-C7",
145 .desc = "MWAIT 0x30",
146 .driver_data = (void *) 0x30,
00527cc6 147 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
d13780d4
LB
148 .exit_latency = 109,
149 .target_residency = 300,
150 .enter = &intel_idle },
151};
152
26717172
LB
153static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
154 { /* MWAIT C0 */ },
155 { /* MWAIT C1 */
156 .name = "ATM-C1",
157 .desc = "MWAIT 0x00",
158 .driver_data = (void *) 0x00,
159 .flags = CPUIDLE_FLAG_TIME_VALID,
160 .exit_latency = 1,
26717172
LB
161 .target_residency = 4,
162 .enter = &intel_idle },
163 { /* MWAIT C2 */
164 .name = "ATM-C2",
165 .desc = "MWAIT 0x10",
166 .driver_data = (void *) 0x10,
167 .flags = CPUIDLE_FLAG_TIME_VALID,
168 .exit_latency = 20,
26717172
LB
169 .target_residency = 80,
170 .enter = &intel_idle },
171 { /* MWAIT C3 */ },
172 { /* MWAIT C4 */
173 .name = "ATM-C4",
174 .desc = "MWAIT 0x30",
175 .driver_data = (void *) 0x30,
6110a1f4 176 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
26717172 177 .exit_latency = 100,
26717172
LB
178 .target_residency = 400,
179 .enter = &intel_idle },
180 { /* MWAIT C5 */ },
181 { /* MWAIT C6 */
182 .name = "ATM-C6",
7fcca7d9
LB
183 .desc = "MWAIT 0x52",
184 .driver_data = (void *) 0x52,
6110a1f4 185 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
7fcca7d9 186 .exit_latency = 140,
7fcca7d9
LB
187 .target_residency = 560,
188 .enter = &intel_idle },
26717172
LB
189};
190
26717172
LB
191/**
192 * intel_idle
193 * @dev: cpuidle_device
194 * @state: cpuidle state
195 *
196 */
197static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
198{
199 unsigned long ecx = 1; /* break on interrupt flag */
200 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
201 unsigned int cstate;
202 ktime_t kt_before, kt_after;
203 s64 usec_delta;
204 int cpu = smp_processor_id();
205
206 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
207
26717172
LB
208 local_irq_disable();
209
6110a1f4 210 /*
c8381cc3
LB
211 * leave_mm() to avoid costly and often unnecessary wakeups
212 * for flushing the user TLB's associated with the active mm.
6110a1f4 213 */
c8381cc3 214 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
6110a1f4
SS
215 leave_mm(cpu);
216
26717172
LB
217 if (!(lapic_timer_reliable_states & (1 << (cstate))))
218 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
219
220 kt_before = ktime_get_real();
221
222 stop_critical_timings();
223#ifndef MODULE
8d91530c 224 trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu);
26717172
LB
225#endif
226 if (!need_resched()) {
227
228 __monitor((void *)&current_thread_info()->flags, 0, 0);
229 smp_mb();
230 if (!need_resched())
231 __mwait(eax, ecx);
232 }
233
234 start_critical_timings();
235
236 kt_after = ktime_get_real();
237 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
238
239 local_irq_enable();
240
241 if (!(lapic_timer_reliable_states & (1 << (cstate))))
242 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
243
244 return usec_delta;
245}
246
247/*
248 * intel_idle_probe()
249 */
250static int intel_idle_probe(void)
251{
c4236282 252 unsigned int eax, ebx, ecx;
26717172
LB
253
254 if (max_cstate == 0) {
255 pr_debug(PREFIX "disabled\n");
256 return -EPERM;
257 }
258
259 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
260 return -ENODEV;
261
262 if (!boot_cpu_has(X86_FEATURE_MWAIT))
263 return -ENODEV;
264
265 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
266 return -ENODEV;
267
c4236282 268 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
26717172
LB
269
270 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
271 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
272 return -ENODEV;
26717172 273
c4236282 274 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
26717172
LB
275
276 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
277 lapic_timer_reliable_states = 0xFFFFFFFF;
278
279 if (boot_cpu_data.x86 != 6) /* family 6 */
280 return -ENODEV;
281
282 switch (boot_cpu_data.x86_model) {
283
284 case 0x1A: /* Core i7, Xeon 5500 series */
285 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
286 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
287 case 0x2E: /* Nehalem-EX Xeon */
ec67a2ba 288 case 0x2F: /* Westmere-EX Xeon */
26717172
LB
289 lapic_timer_reliable_states = (1 << 1); /* C1 */
290
291 case 0x25: /* Westmere */
292 case 0x2C: /* Westmere */
293 cpuidle_state_table = nehalem_cstates;
26717172
LB
294 break;
295
296 case 0x1C: /* 28 - Atom Processor */
4725fd3c 297 case 0x26: /* 38 - Lincroft Atom Processor */
c25d2995 298 lapic_timer_reliable_states = (1 << 1); /* C1 */
26717172 299 cpuidle_state_table = atom_cstates;
26717172 300 break;
d13780d4
LB
301
302 case 0x2A: /* SNB */
303 case 0x2D: /* SNB Xeon */
304 cpuidle_state_table = snb_cstates;
d13780d4 305 break;
26717172
LB
306#ifdef FUTURE_USE
307 case 0x17: /* 23 - Core 2 Duo */
308 lapic_timer_reliable_states = (1 << 2) | (1 << 1); /* C2, C1 */
309#endif
310
311 default:
312 pr_debug(PREFIX "does not run on family %d model %d\n",
313 boot_cpu_data.x86, boot_cpu_data.x86_model);
314 return -ENODEV;
315 }
316
317 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
318 " model 0x%X\n", boot_cpu_data.x86_model);
319
320 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
321 lapic_timer_reliable_states);
322 return 0;
323}
324
325/*
326 * intel_idle_cpuidle_devices_uninit()
327 * unregister, free cpuidle_devices
328 */
329static void intel_idle_cpuidle_devices_uninit(void)
330{
331 int i;
332 struct cpuidle_device *dev;
333
334 for_each_online_cpu(i) {
335 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
336 cpuidle_unregister_device(dev);
337 }
338
339 free_percpu(intel_idle_cpuidle_devices);
340 return;
341}
342/*
343 * intel_idle_cpuidle_devices_init()
344 * allocate, initialize, register cpuidle_devices
345 */
346static int intel_idle_cpuidle_devices_init(void)
347{
348 int i, cstate;
349 struct cpuidle_device *dev;
350
351 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
352 if (intel_idle_cpuidle_devices == NULL)
353 return -ENOMEM;
354
355 for_each_online_cpu(i) {
356 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
357
358 dev->state_count = 1;
359
360 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
361 int num_substates;
362
363 if (cstate > max_cstate) {
364 printk(PREFIX "max_cstate %d reached\n",
365 max_cstate);
366 break;
367 }
368
369 /* does the state exist in CPUID.MWAIT? */
c4236282 370 num_substates = (mwait_substates >> ((cstate) * 4))
26717172
LB
371 & MWAIT_SUBSTATE_MASK;
372 if (num_substates == 0)
373 continue;
374 /* is the state not enabled? */
375 if (cpuidle_state_table[cstate].enter == NULL) {
376 /* does the driver not know about the state? */
377 if (*cpuidle_state_table[cstate].name == '\0')
378 pr_debug(PREFIX "unaware of model 0x%x"
379 " MWAIT %d please"
380 " contact lenb@kernel.org",
381 boot_cpu_data.x86_model, cstate);
382 continue;
383 }
384
385 if ((cstate > 2) &&
386 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
387 mark_tsc_unstable("TSC halts in idle"
388 " states deeper than C2");
389
390 dev->states[dev->state_count] = /* structure copy */
391 cpuidle_state_table[cstate];
392
393 dev->state_count += 1;
394 }
395
396 dev->cpu = i;
397 if (cpuidle_register_device(dev)) {
398 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
399 i);
400 intel_idle_cpuidle_devices_uninit();
401 return -EIO;
402 }
403 }
404
405 return 0;
406}
407
408
409static int __init intel_idle_init(void)
410{
411 int retval;
412
413 retval = intel_idle_probe();
414 if (retval)
415 return retval;
416
417 retval = cpuidle_register_driver(&intel_idle_driver);
418 if (retval) {
419 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
420 cpuidle_get_driver()->name);
421 return retval;
422 }
423
424 retval = intel_idle_cpuidle_devices_init();
425 if (retval) {
426 cpuidle_unregister_driver(&intel_idle_driver);
427 return retval;
428 }
429
430 return 0;
431}
432
433static void __exit intel_idle_exit(void)
434{
435 intel_idle_cpuidle_devices_uninit();
436 cpuidle_unregister_driver(&intel_idle_driver);
437
438 return;
439}
440
441module_init(intel_idle_init);
442module_exit(intel_idle_exit);
443
26717172 444module_param(max_cstate, int, 0444);
26717172
LB
445
446MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
447MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
448MODULE_LICENSE("GPL");