]>
Commit | Line | Data |
---|---|---|
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 | #include <asm/mwait.h> | |
63 | ||
64 | #define INTEL_IDLE_VERSION "0.4" | |
65 | #define PREFIX "intel_idle: " | |
66 | ||
67 | static struct cpuidle_driver intel_idle_driver = { | |
68 | .name = "intel_idle", | |
69 | .owner = THIS_MODULE, | |
70 | }; | |
71 | /* intel_idle.max_cstate=0 disables driver */ | |
72 | static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1; | |
73 | ||
74 | static unsigned int mwait_substates; | |
75 | ||
76 | /* Reliable LAPIC Timer States, bit 1 for C1 etc. */ | |
77 | static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */ | |
78 | ||
79 | static struct cpuidle_device __percpu *intel_idle_cpuidle_devices; | |
80 | static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state); | |
81 | ||
82 | static 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 | */ | |
89 | static 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, | |
97 | .target_residency = 6, | |
98 | .enter = &intel_idle }, | |
99 | { /* MWAIT C2 */ | |
100 | .name = "NHM-C3", | |
101 | .desc = "MWAIT 0x10", | |
102 | .driver_data = (void *) 0x10, | |
103 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
104 | .exit_latency = 20, | |
105 | .target_residency = 80, | |
106 | .enter = &intel_idle }, | |
107 | { /* MWAIT C3 */ | |
108 | .name = "NHM-C6", | |
109 | .desc = "MWAIT 0x20", | |
110 | .driver_data = (void *) 0x20, | |
111 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
112 | .exit_latency = 200, | |
113 | .target_residency = 800, | |
114 | .enter = &intel_idle }, | |
115 | }; | |
116 | ||
117 | static 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 = 1, | |
126 | .enter = &intel_idle }, | |
127 | { /* MWAIT C2 */ | |
128 | .name = "SNB-C3", | |
129 | .desc = "MWAIT 0x10", | |
130 | .driver_data = (void *) 0x10, | |
131 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
132 | .exit_latency = 80, | |
133 | .target_residency = 211, | |
134 | .enter = &intel_idle }, | |
135 | { /* MWAIT C3 */ | |
136 | .name = "SNB-C6", | |
137 | .desc = "MWAIT 0x20", | |
138 | .driver_data = (void *) 0x20, | |
139 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
140 | .exit_latency = 104, | |
141 | .target_residency = 345, | |
142 | .enter = &intel_idle }, | |
143 | { /* MWAIT C4 */ | |
144 | .name = "SNB-C7", | |
145 | .desc = "MWAIT 0x30", | |
146 | .driver_data = (void *) 0x30, | |
147 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
148 | .exit_latency = 109, | |
149 | .target_residency = 345, | |
150 | .enter = &intel_idle }, | |
151 | }; | |
152 | ||
153 | static 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, | |
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, | |
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, | |
176 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
177 | .exit_latency = 100, | |
178 | .target_residency = 400, | |
179 | .enter = &intel_idle }, | |
180 | { /* MWAIT C5 */ }, | |
181 | { /* MWAIT C6 */ | |
182 | .name = "ATM-C6", | |
183 | .desc = "MWAIT 0x52", | |
184 | .driver_data = (void *) 0x52, | |
185 | .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, | |
186 | .exit_latency = 140, | |
187 | .target_residency = 560, | |
188 | .enter = &intel_idle }, | |
189 | }; | |
190 | ||
191 | /** | |
192 | * intel_idle | |
193 | * @dev: cpuidle_device | |
194 | * @state: cpuidle state | |
195 | * | |
196 | */ | |
197 | static 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 | ||
208 | local_irq_disable(); | |
209 | ||
210 | /* | |
211 | * leave_mm() to avoid costly and often unnecessary wakeups | |
212 | * for flushing the user TLB's associated with the active mm. | |
213 | */ | |
214 | if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED) | |
215 | leave_mm(cpu); | |
216 | ||
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 | |
224 | trace_power_start(POWER_CSTATE, (eax >> 4) + 1, cpu); | |
225 | #endif | |
226 | if (!need_resched()) { | |
227 | ||
228 | __monitor((void *)¤t_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 | */ | |
250 | static int intel_idle_probe(void) | |
251 | { | |
252 | unsigned int eax, ebx, ecx; | |
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 | ||
268 | cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates); | |
269 | ||
270 | if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) || | |
271 | !(ecx & CPUID5_ECX_INTERRUPT_BREAK)) | |
272 | return -ENODEV; | |
273 | ||
274 | pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates); | |
275 | ||
276 | ||
277 | if (boot_cpu_data.x86 != 6) /* family 6 */ | |
278 | return -ENODEV; | |
279 | ||
280 | switch (boot_cpu_data.x86_model) { | |
281 | ||
282 | case 0x1A: /* Core i7, Xeon 5500 series */ | |
283 | case 0x1E: /* Core i7 and i5 Processor - Lynnfield Jasper Forest */ | |
284 | case 0x1F: /* Core i7 and i5 Processor - Nehalem */ | |
285 | case 0x2E: /* Nehalem-EX Xeon */ | |
286 | case 0x2F: /* Westmere-EX Xeon */ | |
287 | case 0x25: /* Westmere */ | |
288 | case 0x2C: /* Westmere */ | |
289 | cpuidle_state_table = nehalem_cstates; | |
290 | break; | |
291 | ||
292 | case 0x1C: /* 28 - Atom Processor */ | |
293 | case 0x26: /* 38 - Lincroft Atom Processor */ | |
294 | cpuidle_state_table = atom_cstates; | |
295 | break; | |
296 | ||
297 | case 0x2A: /* SNB */ | |
298 | case 0x2D: /* SNB Xeon */ | |
299 | cpuidle_state_table = snb_cstates; | |
300 | break; | |
301 | ||
302 | default: | |
303 | pr_debug(PREFIX "does not run on family %d model %d\n", | |
304 | boot_cpu_data.x86, boot_cpu_data.x86_model); | |
305 | return -ENODEV; | |
306 | } | |
307 | ||
308 | if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */ | |
309 | lapic_timer_reliable_states = 0xFFFFFFFF; | |
310 | ||
311 | pr_debug(PREFIX "v" INTEL_IDLE_VERSION | |
312 | " model 0x%X\n", boot_cpu_data.x86_model); | |
313 | ||
314 | pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n", | |
315 | lapic_timer_reliable_states); | |
316 | return 0; | |
317 | } | |
318 | ||
319 | /* | |
320 | * intel_idle_cpuidle_devices_uninit() | |
321 | * unregister, free cpuidle_devices | |
322 | */ | |
323 | static void intel_idle_cpuidle_devices_uninit(void) | |
324 | { | |
325 | int i; | |
326 | struct cpuidle_device *dev; | |
327 | ||
328 | for_each_online_cpu(i) { | |
329 | dev = per_cpu_ptr(intel_idle_cpuidle_devices, i); | |
330 | cpuidle_unregister_device(dev); | |
331 | } | |
332 | ||
333 | free_percpu(intel_idle_cpuidle_devices); | |
334 | return; | |
335 | } | |
336 | /* | |
337 | * intel_idle_cpuidle_devices_init() | |
338 | * allocate, initialize, register cpuidle_devices | |
339 | */ | |
340 | static int intel_idle_cpuidle_devices_init(void) | |
341 | { | |
342 | int i, cstate; | |
343 | struct cpuidle_device *dev; | |
344 | ||
345 | intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device); | |
346 | if (intel_idle_cpuidle_devices == NULL) | |
347 | return -ENOMEM; | |
348 | ||
349 | for_each_online_cpu(i) { | |
350 | dev = per_cpu_ptr(intel_idle_cpuidle_devices, i); | |
351 | ||
352 | dev->state_count = 1; | |
353 | ||
354 | for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) { | |
355 | int num_substates; | |
356 | ||
357 | if (cstate > max_cstate) { | |
358 | printk(PREFIX "max_cstate %d reached\n", | |
359 | max_cstate); | |
360 | break; | |
361 | } | |
362 | ||
363 | /* does the state exist in CPUID.MWAIT? */ | |
364 | num_substates = (mwait_substates >> ((cstate) * 4)) | |
365 | & MWAIT_SUBSTATE_MASK; | |
366 | if (num_substates == 0) | |
367 | continue; | |
368 | /* is the state not enabled? */ | |
369 | if (cpuidle_state_table[cstate].enter == NULL) { | |
370 | /* does the driver not know about the state? */ | |
371 | if (*cpuidle_state_table[cstate].name == '\0') | |
372 | pr_debug(PREFIX "unaware of model 0x%x" | |
373 | " MWAIT %d please" | |
374 | " contact lenb@kernel.org", | |
375 | boot_cpu_data.x86_model, cstate); | |
376 | continue; | |
377 | } | |
378 | ||
379 | if ((cstate > 2) && | |
380 | !boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) | |
381 | mark_tsc_unstable("TSC halts in idle" | |
382 | " states deeper than C2"); | |
383 | ||
384 | dev->states[dev->state_count] = /* structure copy */ | |
385 | cpuidle_state_table[cstate]; | |
386 | ||
387 | dev->state_count += 1; | |
388 | } | |
389 | ||
390 | dev->cpu = i; | |
391 | if (cpuidle_register_device(dev)) { | |
392 | pr_debug(PREFIX "cpuidle_register_device %d failed!\n", | |
393 | i); | |
394 | intel_idle_cpuidle_devices_uninit(); | |
395 | return -EIO; | |
396 | } | |
397 | } | |
398 | ||
399 | return 0; | |
400 | } | |
401 | ||
402 | ||
403 | static int __init intel_idle_init(void) | |
404 | { | |
405 | int retval; | |
406 | ||
407 | /* Do not load intel_idle at all for now if idle= is passed */ | |
408 | if (boot_option_idle_override != IDLE_NO_OVERRIDE) | |
409 | return -ENODEV; | |
410 | ||
411 | retval = intel_idle_probe(); | |
412 | if (retval) | |
413 | return retval; | |
414 | ||
415 | retval = cpuidle_register_driver(&intel_idle_driver); | |
416 | if (retval) { | |
417 | printk(KERN_DEBUG PREFIX "intel_idle yielding to %s", | |
418 | cpuidle_get_driver()->name); | |
419 | return retval; | |
420 | } | |
421 | ||
422 | retval = intel_idle_cpuidle_devices_init(); | |
423 | if (retval) { | |
424 | cpuidle_unregister_driver(&intel_idle_driver); | |
425 | return retval; | |
426 | } | |
427 | ||
428 | return 0; | |
429 | } | |
430 | ||
431 | static void __exit intel_idle_exit(void) | |
432 | { | |
433 | intel_idle_cpuidle_devices_uninit(); | |
434 | cpuidle_unregister_driver(&intel_idle_driver); | |
435 | ||
436 | return; | |
437 | } | |
438 | ||
439 | module_init(intel_idle_init); | |
440 | module_exit(intel_idle_exit); | |
441 | ||
442 | module_param(max_cstate, int, 0444); | |
443 | ||
444 | MODULE_AUTHOR("Len Brown <len.brown@intel.com>"); | |
445 | MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION); | |
446 | MODULE_LICENSE("GPL"); |