]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/idle/intel_idle.c
intel_idle: add support for Westmere-EX
[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>
62
63#define INTEL_IDLE_VERSION "0.4"
64#define PREFIX "intel_idle: "
65
66#define MWAIT_SUBSTATE_MASK (0xf)
67#define MWAIT_CSTATE_MASK (0xf)
68#define MWAIT_SUBSTATE_SIZE (4)
69#define MWAIT_MAX_NUM_CSTATES 8
70#define CPUID_MWAIT_LEAF (5)
71#define CPUID5_ECX_EXTENSIONS_SUPPORTED (0x1)
72#define CPUID5_ECX_INTERRUPT_BREAK (0x2)
73
74static struct cpuidle_driver intel_idle_driver = {
75 .name = "intel_idle",
76 .owner = THIS_MODULE,
77};
78/* intel_idle.max_cstate=0 disables driver */
79static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
26717172 80
c4236282 81static unsigned int mwait_substates;
26717172
LB
82
83/* Reliable LAPIC Timer States, bit 1 for C1 etc. */
84static unsigned int lapic_timer_reliable_states;
85
86static struct cpuidle_device *intel_idle_cpuidle_devices;
87static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
88
89static struct cpuidle_state *cpuidle_state_table;
90
91/*
92 * States are indexed by the cstate number,
93 * which is also the index into the MWAIT hint array.
94 * Thus C0 is a dummy.
95 */
96static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
97 { /* MWAIT C0 */ },
98 { /* MWAIT C1 */
99 .name = "NHM-C1",
100 .desc = "MWAIT 0x00",
101 .driver_data = (void *) 0x00,
102 .flags = CPUIDLE_FLAG_TIME_VALID,
103 .exit_latency = 3,
104 .power_usage = 1000,
105 .target_residency = 6,
106 .enter = &intel_idle },
107 { /* MWAIT C2 */
108 .name = "NHM-C3",
109 .desc = "MWAIT 0x10",
110 .driver_data = (void *) 0x10,
111 .flags = CPUIDLE_FLAG_TIME_VALID,
112 .exit_latency = 20,
113 .power_usage = 500,
114 .target_residency = 80,
115 .enter = &intel_idle },
116 { /* MWAIT C3 */
117 .name = "NHM-C6",
118 .desc = "MWAIT 0x20",
119 .driver_data = (void *) 0x20,
120 .flags = CPUIDLE_FLAG_TIME_VALID,
121 .exit_latency = 200,
122 .power_usage = 350,
123 .target_residency = 800,
124 .enter = &intel_idle },
125};
126
127static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
128 { /* MWAIT C0 */ },
129 { /* MWAIT C1 */
130 .name = "ATM-C1",
131 .desc = "MWAIT 0x00",
132 .driver_data = (void *) 0x00,
133 .flags = CPUIDLE_FLAG_TIME_VALID,
134 .exit_latency = 1,
135 .power_usage = 1000,
136 .target_residency = 4,
137 .enter = &intel_idle },
138 { /* MWAIT C2 */
139 .name = "ATM-C2",
140 .desc = "MWAIT 0x10",
141 .driver_data = (void *) 0x10,
142 .flags = CPUIDLE_FLAG_TIME_VALID,
143 .exit_latency = 20,
144 .power_usage = 500,
145 .target_residency = 80,
146 .enter = &intel_idle },
147 { /* MWAIT C3 */ },
148 { /* MWAIT C4 */
149 .name = "ATM-C4",
150 .desc = "MWAIT 0x30",
151 .driver_data = (void *) 0x30,
152 .flags = CPUIDLE_FLAG_TIME_VALID,
153 .exit_latency = 100,
154 .power_usage = 250,
155 .target_residency = 400,
156 .enter = &intel_idle },
157 { /* MWAIT C5 */ },
158 { /* MWAIT C6 */
159 .name = "ATM-C6",
160 .desc = "MWAIT 0x40",
161 .driver_data = (void *) 0x40,
162 .flags = CPUIDLE_FLAG_TIME_VALID,
163 .exit_latency = 200,
164 .power_usage = 150,
165 .target_residency = 800,
166 .enter = NULL }, /* disabled */
167};
168
26717172
LB
169/**
170 * intel_idle
171 * @dev: cpuidle_device
172 * @state: cpuidle state
173 *
174 */
175static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
176{
177 unsigned long ecx = 1; /* break on interrupt flag */
178 unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
179 unsigned int cstate;
180 ktime_t kt_before, kt_after;
181 s64 usec_delta;
182 int cpu = smp_processor_id();
183
184 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
185
26717172
LB
186 local_irq_disable();
187
188 if (!(lapic_timer_reliable_states & (1 << (cstate))))
189 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
190
191 kt_before = ktime_get_real();
192
193 stop_critical_timings();
194#ifndef MODULE
195 trace_power_start(POWER_CSTATE, (eax >> 4) + 1);
196#endif
197 if (!need_resched()) {
198
199 __monitor((void *)&current_thread_info()->flags, 0, 0);
200 smp_mb();
201 if (!need_resched())
202 __mwait(eax, ecx);
203 }
204
205 start_critical_timings();
206
207 kt_after = ktime_get_real();
208 usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
209
210 local_irq_enable();
211
212 if (!(lapic_timer_reliable_states & (1 << (cstate))))
213 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
214
215 return usec_delta;
216}
217
218/*
219 * intel_idle_probe()
220 */
221static int intel_idle_probe(void)
222{
c4236282 223 unsigned int eax, ebx, ecx;
26717172
LB
224
225 if (max_cstate == 0) {
226 pr_debug(PREFIX "disabled\n");
227 return -EPERM;
228 }
229
230 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
231 return -ENODEV;
232
233 if (!boot_cpu_has(X86_FEATURE_MWAIT))
234 return -ENODEV;
235
236 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
237 return -ENODEV;
238
c4236282 239 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
26717172
LB
240
241 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
242 !(ecx & CPUID5_ECX_INTERRUPT_BREAK))
243 return -ENODEV;
26717172 244
c4236282 245 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
26717172
LB
246
247 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
248 lapic_timer_reliable_states = 0xFFFFFFFF;
249
250 if (boot_cpu_data.x86 != 6) /* family 6 */
251 return -ENODEV;
252
253 switch (boot_cpu_data.x86_model) {
254
255 case 0x1A: /* Core i7, Xeon 5500 series */
256 case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */
257 case 0x1F: /* Core i7 and i5 Processor - Nehalem */
258 case 0x2E: /* Nehalem-EX Xeon */
ec67a2ba 259 case 0x2F: /* Westmere-EX Xeon */
26717172
LB
260 lapic_timer_reliable_states = (1 << 1); /* C1 */
261
262 case 0x25: /* Westmere */
263 case 0x2C: /* Westmere */
264 cpuidle_state_table = nehalem_cstates;
26717172
LB
265 break;
266
267 case 0x1C: /* 28 - Atom Processor */
268 lapic_timer_reliable_states = (1 << 2) | (1 << 1); /* C2, C1 */
269 cpuidle_state_table = atom_cstates;
26717172
LB
270 break;
271#ifdef FUTURE_USE
272 case 0x17: /* 23 - Core 2 Duo */
273 lapic_timer_reliable_states = (1 << 2) | (1 << 1); /* C2, C1 */
274#endif
275
276 default:
277 pr_debug(PREFIX "does not run on family %d model %d\n",
278 boot_cpu_data.x86, boot_cpu_data.x86_model);
279 return -ENODEV;
280 }
281
282 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
283 " model 0x%X\n", boot_cpu_data.x86_model);
284
285 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
286 lapic_timer_reliable_states);
287 return 0;
288}
289
290/*
291 * intel_idle_cpuidle_devices_uninit()
292 * unregister, free cpuidle_devices
293 */
294static void intel_idle_cpuidle_devices_uninit(void)
295{
296 int i;
297 struct cpuidle_device *dev;
298
299 for_each_online_cpu(i) {
300 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
301 cpuidle_unregister_device(dev);
302 }
303
304 free_percpu(intel_idle_cpuidle_devices);
305 return;
306}
307/*
308 * intel_idle_cpuidle_devices_init()
309 * allocate, initialize, register cpuidle_devices
310 */
311static int intel_idle_cpuidle_devices_init(void)
312{
313 int i, cstate;
314 struct cpuidle_device *dev;
315
316 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
317 if (intel_idle_cpuidle_devices == NULL)
318 return -ENOMEM;
319
320 for_each_online_cpu(i) {
321 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
322
323 dev->state_count = 1;
324
325 for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
326 int num_substates;
327
328 if (cstate > max_cstate) {
329 printk(PREFIX "max_cstate %d reached\n",
330 max_cstate);
331 break;
332 }
333
334 /* does the state exist in CPUID.MWAIT? */
c4236282 335 num_substates = (mwait_substates >> ((cstate) * 4))
26717172
LB
336 & MWAIT_SUBSTATE_MASK;
337 if (num_substates == 0)
338 continue;
339 /* is the state not enabled? */
340 if (cpuidle_state_table[cstate].enter == NULL) {
341 /* does the driver not know about the state? */
342 if (*cpuidle_state_table[cstate].name == '\0')
343 pr_debug(PREFIX "unaware of model 0x%x"
344 " MWAIT %d please"
345 " contact lenb@kernel.org",
346 boot_cpu_data.x86_model, cstate);
347 continue;
348 }
349
350 if ((cstate > 2) &&
351 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
352 mark_tsc_unstable("TSC halts in idle"
353 " states deeper than C2");
354
355 dev->states[dev->state_count] = /* structure copy */
356 cpuidle_state_table[cstate];
357
358 dev->state_count += 1;
359 }
360
361 dev->cpu = i;
362 if (cpuidle_register_device(dev)) {
363 pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
364 i);
365 intel_idle_cpuidle_devices_uninit();
366 return -EIO;
367 }
368 }
369
370 return 0;
371}
372
373
374static int __init intel_idle_init(void)
375{
376 int retval;
377
378 retval = intel_idle_probe();
379 if (retval)
380 return retval;
381
382 retval = cpuidle_register_driver(&intel_idle_driver);
383 if (retval) {
384 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
385 cpuidle_get_driver()->name);
386 return retval;
387 }
388
389 retval = intel_idle_cpuidle_devices_init();
390 if (retval) {
391 cpuidle_unregister_driver(&intel_idle_driver);
392 return retval;
393 }
394
395 return 0;
396}
397
398static void __exit intel_idle_exit(void)
399{
400 intel_idle_cpuidle_devices_uninit();
401 cpuidle_unregister_driver(&intel_idle_driver);
402
403 return;
404}
405
406module_init(intel_idle_init);
407module_exit(intel_idle_exit);
408
26717172 409module_param(max_cstate, int, 0444);
26717172
LB
410
411MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
412MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
413MODULE_LICENSE("GPL");