]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
19 | */ | |
20 | ||
21 | #include <linux/mm.h> | |
22 | #include <linux/module.h> | |
23 | #include <linux/nmi.h> | |
24 | #include <linux/init.h> | |
25 | #include <asm/uaccess.h> | |
26 | #include <linux/highmem.h> | |
27 | #include <linux/smp_lock.h> | |
28 | #include <asm/mmu_context.h> | |
29 | #include <linux/interrupt.h> | |
c59ede7b | 30 | #include <linux/capability.h> |
1da177e4 LT |
31 | #include <linux/completion.h> |
32 | #include <linux/kernel_stat.h> | |
9a11b49a | 33 | #include <linux/debug_locks.h> |
1da177e4 LT |
34 | #include <linux/security.h> |
35 | #include <linux/notifier.h> | |
36 | #include <linux/profile.h> | |
7dfb7103 | 37 | #include <linux/freezer.h> |
198e2f18 | 38 | #include <linux/vmalloc.h> |
1da177e4 LT |
39 | #include <linux/blkdev.h> |
40 | #include <linux/delay.h> | |
41 | #include <linux/smp.h> | |
42 | #include <linux/threads.h> | |
43 | #include <linux/timer.h> | |
44 | #include <linux/rcupdate.h> | |
45 | #include <linux/cpu.h> | |
46 | #include <linux/cpuset.h> | |
47 | #include <linux/percpu.h> | |
48 | #include <linux/kthread.h> | |
49 | #include <linux/seq_file.h> | |
50 | #include <linux/syscalls.h> | |
51 | #include <linux/times.h> | |
8f0ab514 | 52 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 53 | #include <linux/kprobes.h> |
0ff92245 | 54 | #include <linux/delayacct.h> |
5517d86b | 55 | #include <linux/reciprocal_div.h> |
1da177e4 | 56 | |
5517d86b | 57 | #include <asm/tlb.h> |
1da177e4 LT |
58 | #include <asm/unistd.h> |
59 | ||
b035b6de AD |
60 | /* |
61 | * Scheduler clock - returns current time in nanosec units. | |
62 | * This is default implementation. | |
63 | * Architectures and sub-architectures can override this. | |
64 | */ | |
65 | unsigned long long __attribute__((weak)) sched_clock(void) | |
66 | { | |
67 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
68 | } | |
69 | ||
1da177e4 LT |
70 | /* |
71 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
72 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
73 | * and back. | |
74 | */ | |
75 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
76 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
77 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
78 | ||
79 | /* | |
80 | * 'User priority' is the nice value converted to something we | |
81 | * can work with better when scaling various scheduler parameters, | |
82 | * it's a [ 0 ... 39 ] range. | |
83 | */ | |
84 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
85 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
86 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
87 | ||
88 | /* | |
89 | * Some helpers for converting nanosecond timing to jiffy resolution | |
90 | */ | |
91 | #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) | |
92 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) | |
93 | ||
94 | /* | |
95 | * These are the 'tuning knobs' of the scheduler: | |
96 | * | |
97 | * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger), | |
98 | * default timeslice is 100 msecs, maximum timeslice is 800 msecs. | |
99 | * Timeslices get refilled after they expire. | |
100 | */ | |
101 | #define MIN_TIMESLICE max(5 * HZ / 1000, 1) | |
102 | #define DEF_TIMESLICE (100 * HZ / 1000) | |
103 | #define ON_RUNQUEUE_WEIGHT 30 | |
104 | #define CHILD_PENALTY 95 | |
105 | #define PARENT_PENALTY 100 | |
106 | #define EXIT_WEIGHT 3 | |
107 | #define PRIO_BONUS_RATIO 25 | |
108 | #define MAX_BONUS (MAX_USER_PRIO * PRIO_BONUS_RATIO / 100) | |
109 | #define INTERACTIVE_DELTA 2 | |
110 | #define MAX_SLEEP_AVG (DEF_TIMESLICE * MAX_BONUS) | |
111 | #define STARVATION_LIMIT (MAX_SLEEP_AVG) | |
112 | #define NS_MAX_SLEEP_AVG (JIFFIES_TO_NS(MAX_SLEEP_AVG)) | |
113 | ||
114 | /* | |
115 | * If a task is 'interactive' then we reinsert it in the active | |
116 | * array after it has expired its current timeslice. (it will not | |
117 | * continue to run immediately, it will still roundrobin with | |
118 | * other interactive tasks.) | |
119 | * | |
120 | * This part scales the interactivity limit depending on niceness. | |
121 | * | |
122 | * We scale it linearly, offset by the INTERACTIVE_DELTA delta. | |
123 | * Here are a few examples of different nice levels: | |
124 | * | |
125 | * TASK_INTERACTIVE(-20): [1,1,1,1,1,1,1,1,1,0,0] | |
126 | * TASK_INTERACTIVE(-10): [1,1,1,1,1,1,1,0,0,0,0] | |
127 | * TASK_INTERACTIVE( 0): [1,1,1,1,0,0,0,0,0,0,0] | |
128 | * TASK_INTERACTIVE( 10): [1,1,0,0,0,0,0,0,0,0,0] | |
129 | * TASK_INTERACTIVE( 19): [0,0,0,0,0,0,0,0,0,0,0] | |
130 | * | |
131 | * (the X axis represents the possible -5 ... 0 ... +5 dynamic | |
132 | * priority range a task can explore, a value of '1' means the | |
133 | * task is rated interactive.) | |
134 | * | |
135 | * Ie. nice +19 tasks can never get 'interactive' enough to be | |
136 | * reinserted into the active array. And only heavily CPU-hog nice -20 | |
137 | * tasks will be expired. Default nice 0 tasks are somewhere between, | |
138 | * it takes some effort for them to get interactive, but it's not | |
139 | * too hard. | |
140 | */ | |
141 | ||
142 | #define CURRENT_BONUS(p) \ | |
143 | (NS_TO_JIFFIES((p)->sleep_avg) * MAX_BONUS / \ | |
144 | MAX_SLEEP_AVG) | |
145 | ||
146 | #define GRANULARITY (10 * HZ / 1000 ? : 1) | |
147 | ||
148 | #ifdef CONFIG_SMP | |
149 | #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \ | |
150 | (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1)) * \ | |
151 | num_online_cpus()) | |
152 | #else | |
153 | #define TIMESLICE_GRANULARITY(p) (GRANULARITY * \ | |
154 | (1 << (((MAX_BONUS - CURRENT_BONUS(p)) ? : 1) - 1))) | |
155 | #endif | |
156 | ||
157 | #define SCALE(v1,v1_max,v2_max) \ | |
158 | (v1) * (v2_max) / (v1_max) | |
159 | ||
160 | #define DELTA(p) \ | |
013d3868 MA |
161 | (SCALE(TASK_NICE(p) + 20, 40, MAX_BONUS) - 20 * MAX_BONUS / 40 + \ |
162 | INTERACTIVE_DELTA) | |
1da177e4 LT |
163 | |
164 | #define TASK_INTERACTIVE(p) \ | |
165 | ((p)->prio <= (p)->static_prio - DELTA(p)) | |
166 | ||
167 | #define INTERACTIVE_SLEEP(p) \ | |
168 | (JIFFIES_TO_NS(MAX_SLEEP_AVG * \ | |
169 | (MAX_BONUS / 2 + DELTA((p)) + 1) / MAX_BONUS - 1)) | |
170 | ||
171 | #define TASK_PREEMPTS_CURR(p, rq) \ | |
d5f9f942 | 172 | ((p)->prio < (rq)->curr->prio) |
1da177e4 | 173 | |
1da177e4 | 174 | #define SCALE_PRIO(x, prio) \ |
2dd73a4f | 175 | max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE) |
1da177e4 | 176 | |
2dd73a4f | 177 | static unsigned int static_prio_timeslice(int static_prio) |
1da177e4 | 178 | { |
2dd73a4f PW |
179 | if (static_prio < NICE_TO_PRIO(0)) |
180 | return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio); | |
1da177e4 | 181 | else |
2dd73a4f | 182 | return SCALE_PRIO(DEF_TIMESLICE, static_prio); |
1da177e4 | 183 | } |
2dd73a4f | 184 | |
5517d86b ED |
185 | #ifdef CONFIG_SMP |
186 | /* | |
187 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
188 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
189 | */ | |
190 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
191 | { | |
192 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
193 | } | |
194 | ||
195 | /* | |
196 | * Each time a sched group cpu_power is changed, | |
197 | * we must compute its reciprocal value | |
198 | */ | |
199 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
200 | { | |
201 | sg->__cpu_power += val; | |
202 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
203 | } | |
204 | #endif | |
205 | ||
91fcdd4e BP |
206 | /* |
207 | * task_timeslice() scales user-nice values [ -20 ... 0 ... 19 ] | |
208 | * to time slice values: [800ms ... 100ms ... 5ms] | |
209 | * | |
210 | * The higher a thread's priority, the bigger timeslices | |
211 | * it gets during one round of execution. But even the lowest | |
212 | * priority thread gets MIN_TIMESLICE worth of execution time. | |
213 | */ | |
214 | ||
36c8b586 | 215 | static inline unsigned int task_timeslice(struct task_struct *p) |
2dd73a4f PW |
216 | { |
217 | return static_prio_timeslice(p->static_prio); | |
218 | } | |
219 | ||
1da177e4 LT |
220 | /* |
221 | * These are the runqueue data structures: | |
222 | */ | |
223 | ||
1da177e4 LT |
224 | struct prio_array { |
225 | unsigned int nr_active; | |
d444886e | 226 | DECLARE_BITMAP(bitmap, MAX_PRIO+1); /* include 1 bit for delimiter */ |
1da177e4 LT |
227 | struct list_head queue[MAX_PRIO]; |
228 | }; | |
229 | ||
230 | /* | |
231 | * This is the main, per-CPU runqueue data structure. | |
232 | * | |
233 | * Locking rule: those places that want to lock multiple runqueues | |
234 | * (such as the load balancing or the thread migration code), lock | |
235 | * acquire operations must be ordered by ascending &runqueue. | |
236 | */ | |
70b97a7f | 237 | struct rq { |
1da177e4 LT |
238 | spinlock_t lock; |
239 | ||
240 | /* | |
241 | * nr_running and cpu_load should be in the same cacheline because | |
242 | * remote CPUs use both these fields when doing load calculation. | |
243 | */ | |
244 | unsigned long nr_running; | |
2dd73a4f | 245 | unsigned long raw_weighted_load; |
1da177e4 | 246 | #ifdef CONFIG_SMP |
7897986b | 247 | unsigned long cpu_load[3]; |
bdecea3a | 248 | unsigned char idle_at_tick; |
46cb4b7c SS |
249 | #ifdef CONFIG_NO_HZ |
250 | unsigned char in_nohz_recently; | |
251 | #endif | |
1da177e4 LT |
252 | #endif |
253 | unsigned long long nr_switches; | |
254 | ||
255 | /* | |
256 | * This is part of a global counter where only the total sum | |
257 | * over all CPUs matters. A task can increase this counter on | |
258 | * one CPU and if it got migrated afterwards it may decrease | |
259 | * it on another CPU. Always updated under the runqueue lock: | |
260 | */ | |
261 | unsigned long nr_uninterruptible; | |
262 | ||
263 | unsigned long expired_timestamp; | |
b18ec803 MG |
264 | /* Cached timestamp set by update_cpu_clock() */ |
265 | unsigned long long most_recent_timestamp; | |
36c8b586 | 266 | struct task_struct *curr, *idle; |
c9819f45 | 267 | unsigned long next_balance; |
1da177e4 | 268 | struct mm_struct *prev_mm; |
70b97a7f | 269 | struct prio_array *active, *expired, arrays[2]; |
1da177e4 LT |
270 | int best_expired_prio; |
271 | atomic_t nr_iowait; | |
272 | ||
273 | #ifdef CONFIG_SMP | |
274 | struct sched_domain *sd; | |
275 | ||
276 | /* For active balancing */ | |
277 | int active_balance; | |
278 | int push_cpu; | |
0a2966b4 | 279 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 280 | |
36c8b586 | 281 | struct task_struct *migration_thread; |
1da177e4 LT |
282 | struct list_head migration_queue; |
283 | #endif | |
284 | ||
285 | #ifdef CONFIG_SCHEDSTATS | |
286 | /* latency stats */ | |
287 | struct sched_info rq_sched_info; | |
288 | ||
289 | /* sys_sched_yield() stats */ | |
290 | unsigned long yld_exp_empty; | |
291 | unsigned long yld_act_empty; | |
292 | unsigned long yld_both_empty; | |
293 | unsigned long yld_cnt; | |
294 | ||
295 | /* schedule() stats */ | |
296 | unsigned long sched_switch; | |
297 | unsigned long sched_cnt; | |
298 | unsigned long sched_goidle; | |
299 | ||
300 | /* try_to_wake_up() stats */ | |
301 | unsigned long ttwu_cnt; | |
302 | unsigned long ttwu_local; | |
303 | #endif | |
fcb99371 | 304 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
305 | }; |
306 | ||
c3396620 | 307 | static DEFINE_PER_CPU(struct rq, runqueues) ____cacheline_aligned_in_smp; |
5be9361c | 308 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 309 | |
0a2966b4 CL |
310 | static inline int cpu_of(struct rq *rq) |
311 | { | |
312 | #ifdef CONFIG_SMP | |
313 | return rq->cpu; | |
314 | #else | |
315 | return 0; | |
316 | #endif | |
317 | } | |
318 | ||
674311d5 NP |
319 | /* |
320 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 321 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
322 | * |
323 | * The domain tree of any CPU may only be accessed from within | |
324 | * preempt-disabled sections. | |
325 | */ | |
48f24c4d IM |
326 | #define for_each_domain(cpu, __sd) \ |
327 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
328 | |
329 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
330 | #define this_rq() (&__get_cpu_var(runqueues)) | |
331 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
332 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
333 | ||
1da177e4 | 334 | #ifndef prepare_arch_switch |
4866cde0 NP |
335 | # define prepare_arch_switch(next) do { } while (0) |
336 | #endif | |
337 | #ifndef finish_arch_switch | |
338 | # define finish_arch_switch(prev) do { } while (0) | |
339 | #endif | |
340 | ||
341 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 342 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
343 | { |
344 | return rq->curr == p; | |
345 | } | |
346 | ||
70b97a7f | 347 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
348 | { |
349 | } | |
350 | ||
70b97a7f | 351 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 352 | { |
da04c035 IM |
353 | #ifdef CONFIG_DEBUG_SPINLOCK |
354 | /* this is a valid case when another task releases the spinlock */ | |
355 | rq->lock.owner = current; | |
356 | #endif | |
8a25d5de IM |
357 | /* |
358 | * If we are tracking spinlock dependencies then we have to | |
359 | * fix up the runqueue lock - which gets 'carried over' from | |
360 | * prev into current: | |
361 | */ | |
362 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
363 | ||
4866cde0 NP |
364 | spin_unlock_irq(&rq->lock); |
365 | } | |
366 | ||
367 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 368 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
369 | { |
370 | #ifdef CONFIG_SMP | |
371 | return p->oncpu; | |
372 | #else | |
373 | return rq->curr == p; | |
374 | #endif | |
375 | } | |
376 | ||
70b97a7f | 377 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
378 | { |
379 | #ifdef CONFIG_SMP | |
380 | /* | |
381 | * We can optimise this out completely for !SMP, because the | |
382 | * SMP rebalancing from interrupt is the only thing that cares | |
383 | * here. | |
384 | */ | |
385 | next->oncpu = 1; | |
386 | #endif | |
387 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
388 | spin_unlock_irq(&rq->lock); | |
389 | #else | |
390 | spin_unlock(&rq->lock); | |
391 | #endif | |
392 | } | |
393 | ||
70b97a7f | 394 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
395 | { |
396 | #ifdef CONFIG_SMP | |
397 | /* | |
398 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
399 | * We must ensure this doesn't happen until the switch is completely | |
400 | * finished. | |
401 | */ | |
402 | smp_wmb(); | |
403 | prev->oncpu = 0; | |
404 | #endif | |
405 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
406 | local_irq_enable(); | |
1da177e4 | 407 | #endif |
4866cde0 NP |
408 | } |
409 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 410 | |
b29739f9 IM |
411 | /* |
412 | * __task_rq_lock - lock the runqueue a given task resides on. | |
413 | * Must be called interrupts disabled. | |
414 | */ | |
70b97a7f | 415 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
416 | __acquires(rq->lock) |
417 | { | |
70b97a7f | 418 | struct rq *rq; |
b29739f9 IM |
419 | |
420 | repeat_lock_task: | |
421 | rq = task_rq(p); | |
422 | spin_lock(&rq->lock); | |
423 | if (unlikely(rq != task_rq(p))) { | |
424 | spin_unlock(&rq->lock); | |
425 | goto repeat_lock_task; | |
426 | } | |
427 | return rq; | |
428 | } | |
429 | ||
1da177e4 LT |
430 | /* |
431 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
432 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
433 | * explicitly disabling preemption. | |
434 | */ | |
70b97a7f | 435 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
436 | __acquires(rq->lock) |
437 | { | |
70b97a7f | 438 | struct rq *rq; |
1da177e4 LT |
439 | |
440 | repeat_lock_task: | |
441 | local_irq_save(*flags); | |
442 | rq = task_rq(p); | |
443 | spin_lock(&rq->lock); | |
444 | if (unlikely(rq != task_rq(p))) { | |
445 | spin_unlock_irqrestore(&rq->lock, *flags); | |
446 | goto repeat_lock_task; | |
447 | } | |
448 | return rq; | |
449 | } | |
450 | ||
70b97a7f | 451 | static inline void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
452 | __releases(rq->lock) |
453 | { | |
454 | spin_unlock(&rq->lock); | |
455 | } | |
456 | ||
70b97a7f | 457 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
458 | __releases(rq->lock) |
459 | { | |
460 | spin_unlock_irqrestore(&rq->lock, *flags); | |
461 | } | |
462 | ||
463 | #ifdef CONFIG_SCHEDSTATS | |
464 | /* | |
465 | * bump this up when changing the output format or the meaning of an existing | |
466 | * format, so that tools can adapt (or abort) | |
467 | */ | |
06066714 | 468 | #define SCHEDSTAT_VERSION 14 |
1da177e4 LT |
469 | |
470 | static int show_schedstat(struct seq_file *seq, void *v) | |
471 | { | |
472 | int cpu; | |
473 | ||
474 | seq_printf(seq, "version %d\n", SCHEDSTAT_VERSION); | |
475 | seq_printf(seq, "timestamp %lu\n", jiffies); | |
476 | for_each_online_cpu(cpu) { | |
70b97a7f | 477 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
478 | #ifdef CONFIG_SMP |
479 | struct sched_domain *sd; | |
480 | int dcnt = 0; | |
481 | #endif | |
482 | ||
483 | /* runqueue-specific stats */ | |
484 | seq_printf(seq, | |
485 | "cpu%d %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu", | |
486 | cpu, rq->yld_both_empty, | |
487 | rq->yld_act_empty, rq->yld_exp_empty, rq->yld_cnt, | |
488 | rq->sched_switch, rq->sched_cnt, rq->sched_goidle, | |
489 | rq->ttwu_cnt, rq->ttwu_local, | |
490 | rq->rq_sched_info.cpu_time, | |
491 | rq->rq_sched_info.run_delay, rq->rq_sched_info.pcnt); | |
492 | ||
493 | seq_printf(seq, "\n"); | |
494 | ||
495 | #ifdef CONFIG_SMP | |
496 | /* domain-specific stats */ | |
674311d5 | 497 | preempt_disable(); |
1da177e4 | 498 | for_each_domain(cpu, sd) { |
d15bcfdb | 499 | enum cpu_idle_type itype; |
1da177e4 LT |
500 | char mask_str[NR_CPUS]; |
501 | ||
502 | cpumask_scnprintf(mask_str, NR_CPUS, sd->span); | |
503 | seq_printf(seq, "domain%d %s", dcnt++, mask_str); | |
d15bcfdb | 504 | for (itype = CPU_IDLE; itype < CPU_MAX_IDLE_TYPES; |
1da177e4 | 505 | itype++) { |
33859f7f MOS |
506 | seq_printf(seq, " %lu %lu %lu %lu %lu %lu %lu " |
507 | "%lu", | |
1da177e4 LT |
508 | sd->lb_cnt[itype], |
509 | sd->lb_balanced[itype], | |
510 | sd->lb_failed[itype], | |
511 | sd->lb_imbalance[itype], | |
512 | sd->lb_gained[itype], | |
513 | sd->lb_hot_gained[itype], | |
514 | sd->lb_nobusyq[itype], | |
06066714 | 515 | sd->lb_nobusyg[itype]); |
1da177e4 | 516 | } |
33859f7f MOS |
517 | seq_printf(seq, " %lu %lu %lu %lu %lu %lu %lu %lu %lu" |
518 | " %lu %lu %lu\n", | |
1da177e4 | 519 | sd->alb_cnt, sd->alb_failed, sd->alb_pushed, |
68767a0a NP |
520 | sd->sbe_cnt, sd->sbe_balanced, sd->sbe_pushed, |
521 | sd->sbf_cnt, sd->sbf_balanced, sd->sbf_pushed, | |
33859f7f MOS |
522 | sd->ttwu_wake_remote, sd->ttwu_move_affine, |
523 | sd->ttwu_move_balance); | |
1da177e4 | 524 | } |
674311d5 | 525 | preempt_enable(); |
1da177e4 LT |
526 | #endif |
527 | } | |
528 | return 0; | |
529 | } | |
530 | ||
531 | static int schedstat_open(struct inode *inode, struct file *file) | |
532 | { | |
533 | unsigned int size = PAGE_SIZE * (1 + num_online_cpus() / 32); | |
534 | char *buf = kmalloc(size, GFP_KERNEL); | |
535 | struct seq_file *m; | |
536 | int res; | |
537 | ||
538 | if (!buf) | |
539 | return -ENOMEM; | |
540 | res = single_open(file, show_schedstat, NULL); | |
541 | if (!res) { | |
542 | m = file->private_data; | |
543 | m->buf = buf; | |
544 | m->size = size; | |
545 | } else | |
546 | kfree(buf); | |
547 | return res; | |
548 | } | |
549 | ||
15ad7cdc | 550 | const struct file_operations proc_schedstat_operations = { |
1da177e4 LT |
551 | .open = schedstat_open, |
552 | .read = seq_read, | |
553 | .llseek = seq_lseek, | |
554 | .release = single_release, | |
555 | }; | |
556 | ||
52f17b6c CS |
557 | /* |
558 | * Expects runqueue lock to be held for atomicity of update | |
559 | */ | |
560 | static inline void | |
561 | rq_sched_info_arrive(struct rq *rq, unsigned long delta_jiffies) | |
562 | { | |
563 | if (rq) { | |
564 | rq->rq_sched_info.run_delay += delta_jiffies; | |
565 | rq->rq_sched_info.pcnt++; | |
566 | } | |
567 | } | |
568 | ||
569 | /* | |
570 | * Expects runqueue lock to be held for atomicity of update | |
571 | */ | |
572 | static inline void | |
573 | rq_sched_info_depart(struct rq *rq, unsigned long delta_jiffies) | |
574 | { | |
575 | if (rq) | |
576 | rq->rq_sched_info.cpu_time += delta_jiffies; | |
577 | } | |
1da177e4 LT |
578 | # define schedstat_inc(rq, field) do { (rq)->field++; } while (0) |
579 | # define schedstat_add(rq, field, amt) do { (rq)->field += (amt); } while (0) | |
580 | #else /* !CONFIG_SCHEDSTATS */ | |
52f17b6c CS |
581 | static inline void |
582 | rq_sched_info_arrive(struct rq *rq, unsigned long delta_jiffies) | |
583 | {} | |
584 | static inline void | |
585 | rq_sched_info_depart(struct rq *rq, unsigned long delta_jiffies) | |
586 | {} | |
1da177e4 LT |
587 | # define schedstat_inc(rq, field) do { } while (0) |
588 | # define schedstat_add(rq, field, amt) do { } while (0) | |
589 | #endif | |
590 | ||
591 | /* | |
cc2a73b5 | 592 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 593 | */ |
70b97a7f | 594 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
595 | __acquires(rq->lock) |
596 | { | |
70b97a7f | 597 | struct rq *rq; |
1da177e4 LT |
598 | |
599 | local_irq_disable(); | |
600 | rq = this_rq(); | |
601 | spin_lock(&rq->lock); | |
602 | ||
603 | return rq; | |
604 | } | |
605 | ||
52f17b6c | 606 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
1da177e4 LT |
607 | /* |
608 | * Called when a process is dequeued from the active array and given | |
609 | * the cpu. We should note that with the exception of interactive | |
610 | * tasks, the expired queue will become the active queue after the active | |
611 | * queue is empty, without explicitly dequeuing and requeuing tasks in the | |
612 | * expired queue. (Interactive tasks may be requeued directly to the | |
613 | * active queue, thus delaying tasks in the expired queue from running; | |
614 | * see scheduler_tick()). | |
615 | * | |
616 | * This function is only called from sched_info_arrive(), rather than | |
617 | * dequeue_task(). Even though a task may be queued and dequeued multiple | |
618 | * times as it is shuffled about, we're really interested in knowing how | |
619 | * long it was from the *first* time it was queued to the time that it | |
620 | * finally hit a cpu. | |
621 | */ | |
36c8b586 | 622 | static inline void sched_info_dequeued(struct task_struct *t) |
1da177e4 LT |
623 | { |
624 | t->sched_info.last_queued = 0; | |
625 | } | |
626 | ||
627 | /* | |
628 | * Called when a task finally hits the cpu. We can now calculate how | |
629 | * long it was waiting to run. We also note when it began so that we | |
630 | * can keep stats on how long its timeslice is. | |
631 | */ | |
36c8b586 | 632 | static void sched_info_arrive(struct task_struct *t) |
1da177e4 | 633 | { |
52f17b6c | 634 | unsigned long now = jiffies, delta_jiffies = 0; |
1da177e4 LT |
635 | |
636 | if (t->sched_info.last_queued) | |
52f17b6c | 637 | delta_jiffies = now - t->sched_info.last_queued; |
1da177e4 | 638 | sched_info_dequeued(t); |
52f17b6c | 639 | t->sched_info.run_delay += delta_jiffies; |
1da177e4 LT |
640 | t->sched_info.last_arrival = now; |
641 | t->sched_info.pcnt++; | |
642 | ||
52f17b6c | 643 | rq_sched_info_arrive(task_rq(t), delta_jiffies); |
1da177e4 LT |
644 | } |
645 | ||
646 | /* | |
647 | * Called when a process is queued into either the active or expired | |
648 | * array. The time is noted and later used to determine how long we | |
649 | * had to wait for us to reach the cpu. Since the expired queue will | |
650 | * become the active queue after active queue is empty, without dequeuing | |
651 | * and requeuing any tasks, we are interested in queuing to either. It | |
652 | * is unusual but not impossible for tasks to be dequeued and immediately | |
653 | * requeued in the same or another array: this can happen in sched_yield(), | |
654 | * set_user_nice(), and even load_balance() as it moves tasks from runqueue | |
655 | * to runqueue. | |
656 | * | |
657 | * This function is only called from enqueue_task(), but also only updates | |
658 | * the timestamp if it is already not set. It's assumed that | |
659 | * sched_info_dequeued() will clear that stamp when appropriate. | |
660 | */ | |
36c8b586 | 661 | static inline void sched_info_queued(struct task_struct *t) |
1da177e4 | 662 | { |
52f17b6c CS |
663 | if (unlikely(sched_info_on())) |
664 | if (!t->sched_info.last_queued) | |
665 | t->sched_info.last_queued = jiffies; | |
1da177e4 LT |
666 | } |
667 | ||
668 | /* | |
669 | * Called when a process ceases being the active-running process, either | |
670 | * voluntarily or involuntarily. Now we can calculate how long we ran. | |
671 | */ | |
36c8b586 | 672 | static inline void sched_info_depart(struct task_struct *t) |
1da177e4 | 673 | { |
52f17b6c | 674 | unsigned long delta_jiffies = jiffies - t->sched_info.last_arrival; |
1da177e4 | 675 | |
52f17b6c CS |
676 | t->sched_info.cpu_time += delta_jiffies; |
677 | rq_sched_info_depart(task_rq(t), delta_jiffies); | |
1da177e4 LT |
678 | } |
679 | ||
680 | /* | |
681 | * Called when tasks are switched involuntarily due, typically, to expiring | |
682 | * their time slice. (This may also be called when switching to or from | |
683 | * the idle task.) We are only called when prev != next. | |
684 | */ | |
36c8b586 | 685 | static inline void |
52f17b6c | 686 | __sched_info_switch(struct task_struct *prev, struct task_struct *next) |
1da177e4 | 687 | { |
70b97a7f | 688 | struct rq *rq = task_rq(prev); |
1da177e4 LT |
689 | |
690 | /* | |
691 | * prev now departs the cpu. It's not interesting to record | |
692 | * stats about how efficient we were at scheduling the idle | |
693 | * process, however. | |
694 | */ | |
695 | if (prev != rq->idle) | |
696 | sched_info_depart(prev); | |
697 | ||
698 | if (next != rq->idle) | |
699 | sched_info_arrive(next); | |
700 | } | |
52f17b6c CS |
701 | static inline void |
702 | sched_info_switch(struct task_struct *prev, struct task_struct *next) | |
703 | { | |
704 | if (unlikely(sched_info_on())) | |
705 | __sched_info_switch(prev, next); | |
706 | } | |
1da177e4 LT |
707 | #else |
708 | #define sched_info_queued(t) do { } while (0) | |
709 | #define sched_info_switch(t, next) do { } while (0) | |
52f17b6c | 710 | #endif /* CONFIG_SCHEDSTATS || CONFIG_TASK_DELAY_ACCT */ |
1da177e4 LT |
711 | |
712 | /* | |
713 | * Adding/removing a task to/from a priority array: | |
714 | */ | |
70b97a7f | 715 | static void dequeue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
716 | { |
717 | array->nr_active--; | |
718 | list_del(&p->run_list); | |
719 | if (list_empty(array->queue + p->prio)) | |
720 | __clear_bit(p->prio, array->bitmap); | |
721 | } | |
722 | ||
70b97a7f | 723 | static void enqueue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
724 | { |
725 | sched_info_queued(p); | |
726 | list_add_tail(&p->run_list, array->queue + p->prio); | |
727 | __set_bit(p->prio, array->bitmap); | |
728 | array->nr_active++; | |
729 | p->array = array; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Put task to the end of the run list without the overhead of dequeue | |
734 | * followed by enqueue. | |
735 | */ | |
70b97a7f | 736 | static void requeue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
737 | { |
738 | list_move_tail(&p->run_list, array->queue + p->prio); | |
739 | } | |
740 | ||
70b97a7f IM |
741 | static inline void |
742 | enqueue_task_head(struct task_struct *p, struct prio_array *array) | |
1da177e4 LT |
743 | { |
744 | list_add(&p->run_list, array->queue + p->prio); | |
745 | __set_bit(p->prio, array->bitmap); | |
746 | array->nr_active++; | |
747 | p->array = array; | |
748 | } | |
749 | ||
750 | /* | |
b29739f9 | 751 | * __normal_prio - return the priority that is based on the static |
1da177e4 LT |
752 | * priority but is modified by bonuses/penalties. |
753 | * | |
754 | * We scale the actual sleep average [0 .... MAX_SLEEP_AVG] | |
755 | * into the -5 ... 0 ... +5 bonus/penalty range. | |
756 | * | |
757 | * We use 25% of the full 0...39 priority range so that: | |
758 | * | |
759 | * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs. | |
760 | * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks. | |
761 | * | |
762 | * Both properties are important to certain workloads. | |
763 | */ | |
b29739f9 | 764 | |
36c8b586 | 765 | static inline int __normal_prio(struct task_struct *p) |
1da177e4 LT |
766 | { |
767 | int bonus, prio; | |
768 | ||
1da177e4 LT |
769 | bonus = CURRENT_BONUS(p) - MAX_BONUS / 2; |
770 | ||
771 | prio = p->static_prio - bonus; | |
772 | if (prio < MAX_RT_PRIO) | |
773 | prio = MAX_RT_PRIO; | |
774 | if (prio > MAX_PRIO-1) | |
775 | prio = MAX_PRIO-1; | |
776 | return prio; | |
777 | } | |
778 | ||
2dd73a4f PW |
779 | /* |
780 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
781 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
782 | * each task makes to its run queue's load is weighted according to its | |
783 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
784 | * scaled version of the new time slice allocation that they receive on time | |
785 | * slice expiry etc. | |
786 | */ | |
787 | ||
788 | /* | |
789 | * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE | |
790 | * If static_prio_timeslice() is ever changed to break this assumption then | |
791 | * this code will need modification | |
792 | */ | |
793 | #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE | |
794 | #define LOAD_WEIGHT(lp) \ | |
795 | (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO) | |
796 | #define PRIO_TO_LOAD_WEIGHT(prio) \ | |
797 | LOAD_WEIGHT(static_prio_timeslice(prio)) | |
798 | #define RTPRIO_TO_LOAD_WEIGHT(rp) \ | |
799 | (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + LOAD_WEIGHT(rp)) | |
800 | ||
36c8b586 | 801 | static void set_load_weight(struct task_struct *p) |
2dd73a4f | 802 | { |
b29739f9 | 803 | if (has_rt_policy(p)) { |
2dd73a4f PW |
804 | #ifdef CONFIG_SMP |
805 | if (p == task_rq(p)->migration_thread) | |
806 | /* | |
807 | * The migration thread does the actual balancing. | |
808 | * Giving its load any weight will skew balancing | |
809 | * adversely. | |
810 | */ | |
811 | p->load_weight = 0; | |
812 | else | |
813 | #endif | |
814 | p->load_weight = RTPRIO_TO_LOAD_WEIGHT(p->rt_priority); | |
815 | } else | |
816 | p->load_weight = PRIO_TO_LOAD_WEIGHT(p->static_prio); | |
817 | } | |
818 | ||
36c8b586 | 819 | static inline void |
70b97a7f | 820 | inc_raw_weighted_load(struct rq *rq, const struct task_struct *p) |
2dd73a4f PW |
821 | { |
822 | rq->raw_weighted_load += p->load_weight; | |
823 | } | |
824 | ||
36c8b586 | 825 | static inline void |
70b97a7f | 826 | dec_raw_weighted_load(struct rq *rq, const struct task_struct *p) |
2dd73a4f PW |
827 | { |
828 | rq->raw_weighted_load -= p->load_weight; | |
829 | } | |
830 | ||
70b97a7f | 831 | static inline void inc_nr_running(struct task_struct *p, struct rq *rq) |
2dd73a4f PW |
832 | { |
833 | rq->nr_running++; | |
834 | inc_raw_weighted_load(rq, p); | |
835 | } | |
836 | ||
70b97a7f | 837 | static inline void dec_nr_running(struct task_struct *p, struct rq *rq) |
2dd73a4f PW |
838 | { |
839 | rq->nr_running--; | |
840 | dec_raw_weighted_load(rq, p); | |
841 | } | |
842 | ||
b29739f9 IM |
843 | /* |
844 | * Calculate the expected normal priority: i.e. priority | |
845 | * without taking RT-inheritance into account. Might be | |
846 | * boosted by interactivity modifiers. Changes upon fork, | |
847 | * setprio syscalls, and whenever the interactivity | |
848 | * estimator recalculates. | |
849 | */ | |
36c8b586 | 850 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
851 | { |
852 | int prio; | |
853 | ||
854 | if (has_rt_policy(p)) | |
855 | prio = MAX_RT_PRIO-1 - p->rt_priority; | |
856 | else | |
857 | prio = __normal_prio(p); | |
858 | return prio; | |
859 | } | |
860 | ||
861 | /* | |
862 | * Calculate the current priority, i.e. the priority | |
863 | * taken into account by the scheduler. This value might | |
864 | * be boosted by RT tasks, or might be boosted by | |
865 | * interactivity modifiers. Will be RT if the task got | |
866 | * RT-boosted. If not then it returns p->normal_prio. | |
867 | */ | |
36c8b586 | 868 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
869 | { |
870 | p->normal_prio = normal_prio(p); | |
871 | /* | |
872 | * If we are RT tasks or we were boosted to RT priority, | |
873 | * keep the priority unchanged. Otherwise, update priority | |
874 | * to the normal priority: | |
875 | */ | |
876 | if (!rt_prio(p->prio)) | |
877 | return p->normal_prio; | |
878 | return p->prio; | |
879 | } | |
880 | ||
1da177e4 LT |
881 | /* |
882 | * __activate_task - move a task to the runqueue. | |
883 | */ | |
70b97a7f | 884 | static void __activate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 885 | { |
70b97a7f | 886 | struct prio_array *target = rq->active; |
d425b274 | 887 | |
f1adad78 | 888 | if (batch_task(p)) |
d425b274 CK |
889 | target = rq->expired; |
890 | enqueue_task(p, target); | |
2dd73a4f | 891 | inc_nr_running(p, rq); |
1da177e4 LT |
892 | } |
893 | ||
894 | /* | |
895 | * __activate_idle_task - move idle task to the _front_ of runqueue. | |
896 | */ | |
70b97a7f | 897 | static inline void __activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 LT |
898 | { |
899 | enqueue_task_head(p, rq->active); | |
2dd73a4f | 900 | inc_nr_running(p, rq); |
1da177e4 LT |
901 | } |
902 | ||
b29739f9 IM |
903 | /* |
904 | * Recalculate p->normal_prio and p->prio after having slept, | |
905 | * updating the sleep-average too: | |
906 | */ | |
36c8b586 | 907 | static int recalc_task_prio(struct task_struct *p, unsigned long long now) |
1da177e4 LT |
908 | { |
909 | /* Caller must always ensure 'now >= p->timestamp' */ | |
72d2854d | 910 | unsigned long sleep_time = now - p->timestamp; |
1da177e4 | 911 | |
d425b274 | 912 | if (batch_task(p)) |
b0a9499c | 913 | sleep_time = 0; |
1da177e4 LT |
914 | |
915 | if (likely(sleep_time > 0)) { | |
916 | /* | |
72d2854d CK |
917 | * This ceiling is set to the lowest priority that would allow |
918 | * a task to be reinserted into the active array on timeslice | |
919 | * completion. | |
1da177e4 | 920 | */ |
72d2854d | 921 | unsigned long ceiling = INTERACTIVE_SLEEP(p); |
e72ff0bb | 922 | |
72d2854d CK |
923 | if (p->mm && sleep_time > ceiling && p->sleep_avg < ceiling) { |
924 | /* | |
925 | * Prevents user tasks from achieving best priority | |
926 | * with one single large enough sleep. | |
927 | */ | |
928 | p->sleep_avg = ceiling; | |
929 | /* | |
930 | * Using INTERACTIVE_SLEEP() as a ceiling places a | |
931 | * nice(0) task 1ms sleep away from promotion, and | |
932 | * gives it 700ms to round-robin with no chance of | |
933 | * being demoted. This is more than generous, so | |
934 | * mark this sleep as non-interactive to prevent the | |
935 | * on-runqueue bonus logic from intervening should | |
936 | * this task not receive cpu immediately. | |
937 | */ | |
938 | p->sleep_type = SLEEP_NONINTERACTIVE; | |
1da177e4 | 939 | } else { |
1da177e4 LT |
940 | /* |
941 | * Tasks waking from uninterruptible sleep are | |
942 | * limited in their sleep_avg rise as they | |
943 | * are likely to be waiting on I/O | |
944 | */ | |
3dee386e | 945 | if (p->sleep_type == SLEEP_NONINTERACTIVE && p->mm) { |
72d2854d | 946 | if (p->sleep_avg >= ceiling) |
1da177e4 LT |
947 | sleep_time = 0; |
948 | else if (p->sleep_avg + sleep_time >= | |
72d2854d CK |
949 | ceiling) { |
950 | p->sleep_avg = ceiling; | |
951 | sleep_time = 0; | |
1da177e4 LT |
952 | } |
953 | } | |
954 | ||
955 | /* | |
956 | * This code gives a bonus to interactive tasks. | |
957 | * | |
958 | * The boost works by updating the 'average sleep time' | |
959 | * value here, based on ->timestamp. The more time a | |
960 | * task spends sleeping, the higher the average gets - | |
961 | * and the higher the priority boost gets as well. | |
962 | */ | |
963 | p->sleep_avg += sleep_time; | |
964 | ||
1da177e4 | 965 | } |
72d2854d CK |
966 | if (p->sleep_avg > NS_MAX_SLEEP_AVG) |
967 | p->sleep_avg = NS_MAX_SLEEP_AVG; | |
1da177e4 LT |
968 | } |
969 | ||
a3464a10 | 970 | return effective_prio(p); |
1da177e4 LT |
971 | } |
972 | ||
973 | /* | |
974 | * activate_task - move a task to the runqueue and do priority recalculation | |
975 | * | |
976 | * Update all the scheduling statistics stuff. (sleep average | |
977 | * calculation, priority modifiers, etc.) | |
978 | */ | |
70b97a7f | 979 | static void activate_task(struct task_struct *p, struct rq *rq, int local) |
1da177e4 LT |
980 | { |
981 | unsigned long long now; | |
982 | ||
62ab616d KC |
983 | if (rt_task(p)) |
984 | goto out; | |
985 | ||
1da177e4 LT |
986 | now = sched_clock(); |
987 | #ifdef CONFIG_SMP | |
988 | if (!local) { | |
989 | /* Compensate for drifting sched_clock */ | |
70b97a7f | 990 | struct rq *this_rq = this_rq(); |
b18ec803 MG |
991 | now = (now - this_rq->most_recent_timestamp) |
992 | + rq->most_recent_timestamp; | |
1da177e4 LT |
993 | } |
994 | #endif | |
995 | ||
ece8a684 IM |
996 | /* |
997 | * Sleep time is in units of nanosecs, so shift by 20 to get a | |
998 | * milliseconds-range estimation of the amount of time that the task | |
999 | * spent sleeping: | |
1000 | */ | |
1001 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
1002 | if (p->state == TASK_UNINTERRUPTIBLE) | |
1003 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(p), | |
1004 | (now - p->timestamp) >> 20); | |
1005 | } | |
1006 | ||
62ab616d | 1007 | p->prio = recalc_task_prio(p, now); |
1da177e4 LT |
1008 | |
1009 | /* | |
1010 | * This checks to make sure it's not an uninterruptible task | |
1011 | * that is now waking up. | |
1012 | */ | |
3dee386e | 1013 | if (p->sleep_type == SLEEP_NORMAL) { |
1da177e4 LT |
1014 | /* |
1015 | * Tasks which were woken up by interrupts (ie. hw events) | |
1016 | * are most likely of interactive nature. So we give them | |
1017 | * the credit of extending their sleep time to the period | |
1018 | * of time they spend on the runqueue, waiting for execution | |
1019 | * on a CPU, first time around: | |
1020 | */ | |
1021 | if (in_interrupt()) | |
3dee386e | 1022 | p->sleep_type = SLEEP_INTERRUPTED; |
1da177e4 LT |
1023 | else { |
1024 | /* | |
1025 | * Normal first-time wakeups get a credit too for | |
1026 | * on-runqueue time, but it will be weighted down: | |
1027 | */ | |
3dee386e | 1028 | p->sleep_type = SLEEP_INTERACTIVE; |
1da177e4 LT |
1029 | } |
1030 | } | |
1031 | p->timestamp = now; | |
62ab616d | 1032 | out: |
1da177e4 LT |
1033 | __activate_task(p, rq); |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * deactivate_task - remove a task from the runqueue. | |
1038 | */ | |
70b97a7f | 1039 | static void deactivate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 1040 | { |
2dd73a4f | 1041 | dec_nr_running(p, rq); |
1da177e4 LT |
1042 | dequeue_task(p, p->array); |
1043 | p->array = NULL; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * resched_task - mark a task 'to be rescheduled now'. | |
1048 | * | |
1049 | * On UP this means the setting of the need_resched flag, on SMP it | |
1050 | * might also involve a cross-CPU call to trigger the scheduler on | |
1051 | * the target CPU. | |
1052 | */ | |
1053 | #ifdef CONFIG_SMP | |
495ab9c0 AK |
1054 | |
1055 | #ifndef tsk_is_polling | |
1056 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1057 | #endif | |
1058 | ||
36c8b586 | 1059 | static void resched_task(struct task_struct *p) |
1da177e4 | 1060 | { |
64c7c8f8 | 1061 | int cpu; |
1da177e4 LT |
1062 | |
1063 | assert_spin_locked(&task_rq(p)->lock); | |
1064 | ||
64c7c8f8 NP |
1065 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
1066 | return; | |
1067 | ||
1068 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
1da177e4 | 1069 | |
64c7c8f8 NP |
1070 | cpu = task_cpu(p); |
1071 | if (cpu == smp_processor_id()) | |
1072 | return; | |
1073 | ||
495ab9c0 | 1074 | /* NEED_RESCHED must be visible before we test polling */ |
64c7c8f8 | 1075 | smp_mb(); |
495ab9c0 | 1076 | if (!tsk_is_polling(p)) |
64c7c8f8 | 1077 | smp_send_reschedule(cpu); |
1da177e4 | 1078 | } |
46cb4b7c SS |
1079 | |
1080 | static void resched_cpu(int cpu) | |
1081 | { | |
1082 | struct rq *rq = cpu_rq(cpu); | |
1083 | unsigned long flags; | |
1084 | ||
1085 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1086 | return; | |
1087 | resched_task(cpu_curr(cpu)); | |
1088 | spin_unlock_irqrestore(&rq->lock, flags); | |
1089 | } | |
1da177e4 | 1090 | #else |
36c8b586 | 1091 | static inline void resched_task(struct task_struct *p) |
1da177e4 | 1092 | { |
64c7c8f8 | 1093 | assert_spin_locked(&task_rq(p)->lock); |
1da177e4 LT |
1094 | set_tsk_need_resched(p); |
1095 | } | |
1096 | #endif | |
1097 | ||
1098 | /** | |
1099 | * task_curr - is this task currently executing on a CPU? | |
1100 | * @p: the task in question. | |
1101 | */ | |
36c8b586 | 1102 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1103 | { |
1104 | return cpu_curr(task_cpu(p)) == p; | |
1105 | } | |
1106 | ||
2dd73a4f PW |
1107 | /* Used instead of source_load when we know the type == 0 */ |
1108 | unsigned long weighted_cpuload(const int cpu) | |
1109 | { | |
1110 | return cpu_rq(cpu)->raw_weighted_load; | |
1111 | } | |
1112 | ||
1da177e4 | 1113 | #ifdef CONFIG_SMP |
c65cc870 IM |
1114 | |
1115 | void set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1116 | { | |
1117 | task_thread_info(p)->cpu = cpu; | |
1118 | } | |
1119 | ||
70b97a7f | 1120 | struct migration_req { |
1da177e4 | 1121 | struct list_head list; |
1da177e4 | 1122 | |
36c8b586 | 1123 | struct task_struct *task; |
1da177e4 LT |
1124 | int dest_cpu; |
1125 | ||
1da177e4 | 1126 | struct completion done; |
70b97a7f | 1127 | }; |
1da177e4 LT |
1128 | |
1129 | /* | |
1130 | * The task's runqueue lock must be held. | |
1131 | * Returns true if you have to wait for migration thread. | |
1132 | */ | |
36c8b586 | 1133 | static int |
70b97a7f | 1134 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1135 | { |
70b97a7f | 1136 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1137 | |
1138 | /* | |
1139 | * If the task is not on a runqueue (and not running), then | |
1140 | * it is sufficient to simply update the task's cpu field. | |
1141 | */ | |
1142 | if (!p->array && !task_running(rq, p)) { | |
1143 | set_task_cpu(p, dest_cpu); | |
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | init_completion(&req->done); | |
1da177e4 LT |
1148 | req->task = p; |
1149 | req->dest_cpu = dest_cpu; | |
1150 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1151 | |
1da177e4 LT |
1152 | return 1; |
1153 | } | |
1154 | ||
1155 | /* | |
1156 | * wait_task_inactive - wait for a thread to unschedule. | |
1157 | * | |
1158 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1159 | * else this function might spin for a *long* time. This function can't | |
1160 | * be called with interrupts off, or it may introduce deadlock with | |
1161 | * smp_call_function() if an IPI is sent by the same process we are | |
1162 | * waiting to become inactive. | |
1163 | */ | |
36c8b586 | 1164 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1165 | { |
1166 | unsigned long flags; | |
70b97a7f | 1167 | struct rq *rq; |
fa490cfd LT |
1168 | struct prio_array *array; |
1169 | int running; | |
1da177e4 LT |
1170 | |
1171 | repeat: | |
fa490cfd LT |
1172 | /* |
1173 | * We do the initial early heuristics without holding | |
1174 | * any task-queue locks at all. We'll only try to get | |
1175 | * the runqueue lock when things look like they will | |
1176 | * work out! | |
1177 | */ | |
1178 | rq = task_rq(p); | |
1179 | ||
1180 | /* | |
1181 | * If the task is actively running on another CPU | |
1182 | * still, just relax and busy-wait without holding | |
1183 | * any locks. | |
1184 | * | |
1185 | * NOTE! Since we don't hold any locks, it's not | |
1186 | * even sure that "rq" stays as the right runqueue! | |
1187 | * But we don't care, since "task_running()" will | |
1188 | * return false if the runqueue has changed and p | |
1189 | * is actually now running somewhere else! | |
1190 | */ | |
1191 | while (task_running(rq, p)) | |
1192 | cpu_relax(); | |
1193 | ||
1194 | /* | |
1195 | * Ok, time to look more closely! We need the rq | |
1196 | * lock now, to be *sure*. If we're wrong, we'll | |
1197 | * just go back and repeat. | |
1198 | */ | |
1da177e4 | 1199 | rq = task_rq_lock(p, &flags); |
fa490cfd LT |
1200 | running = task_running(rq, p); |
1201 | array = p->array; | |
1202 | task_rq_unlock(rq, &flags); | |
1203 | ||
1204 | /* | |
1205 | * Was it really running after all now that we | |
1206 | * checked with the proper locks actually held? | |
1207 | * | |
1208 | * Oops. Go back and try again.. | |
1209 | */ | |
1210 | if (unlikely(running)) { | |
1da177e4 | 1211 | cpu_relax(); |
1da177e4 LT |
1212 | goto repeat; |
1213 | } | |
fa490cfd LT |
1214 | |
1215 | /* | |
1216 | * It's not enough that it's not actively running, | |
1217 | * it must be off the runqueue _entirely_, and not | |
1218 | * preempted! | |
1219 | * | |
1220 | * So if it wa still runnable (but just not actively | |
1221 | * running right now), it's preempted, and we should | |
1222 | * yield - it could be a while. | |
1223 | */ | |
1224 | if (unlikely(array)) { | |
1225 | yield(); | |
1226 | goto repeat; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Ahh, all good. It wasn't running, and it wasn't | |
1231 | * runnable, which means that it will never become | |
1232 | * running in the future either. We're all done! | |
1233 | */ | |
1da177e4 LT |
1234 | } |
1235 | ||
1236 | /*** | |
1237 | * kick_process - kick a running thread to enter/exit the kernel | |
1238 | * @p: the to-be-kicked thread | |
1239 | * | |
1240 | * Cause a process which is running on another CPU to enter | |
1241 | * kernel-mode, without any delay. (to get signals handled.) | |
1242 | * | |
1243 | * NOTE: this function doesnt have to take the runqueue lock, | |
1244 | * because all it wants to ensure is that the remote task enters | |
1245 | * the kernel. If the IPI races and the task has been migrated | |
1246 | * to another CPU then no harm is done and the purpose has been | |
1247 | * achieved as well. | |
1248 | */ | |
36c8b586 | 1249 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1250 | { |
1251 | int cpu; | |
1252 | ||
1253 | preempt_disable(); | |
1254 | cpu = task_cpu(p); | |
1255 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1256 | smp_send_reschedule(cpu); | |
1257 | preempt_enable(); | |
1258 | } | |
1259 | ||
1260 | /* | |
2dd73a4f PW |
1261 | * Return a low guess at the load of a migration-source cpu weighted |
1262 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1263 | * |
1264 | * We want to under-estimate the load of migration sources, to | |
1265 | * balance conservatively. | |
1266 | */ | |
a2000572 | 1267 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1268 | { |
70b97a7f | 1269 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1270 | |
3b0bd9bc | 1271 | if (type == 0) |
2dd73a4f | 1272 | return rq->raw_weighted_load; |
b910472d | 1273 | |
2dd73a4f | 1274 | return min(rq->cpu_load[type-1], rq->raw_weighted_load); |
1da177e4 LT |
1275 | } |
1276 | ||
1277 | /* | |
2dd73a4f PW |
1278 | * Return a high guess at the load of a migration-target cpu weighted |
1279 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1280 | */ |
a2000572 | 1281 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1282 | { |
70b97a7f | 1283 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1284 | |
7897986b | 1285 | if (type == 0) |
2dd73a4f | 1286 | return rq->raw_weighted_load; |
3b0bd9bc | 1287 | |
2dd73a4f PW |
1288 | return max(rq->cpu_load[type-1], rq->raw_weighted_load); |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * Return the average load per task on the cpu's run queue | |
1293 | */ | |
1294 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1295 | { | |
70b97a7f | 1296 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f PW |
1297 | unsigned long n = rq->nr_running; |
1298 | ||
48f24c4d | 1299 | return n ? rq->raw_weighted_load / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1300 | } |
1301 | ||
147cbb4b NP |
1302 | /* |
1303 | * find_idlest_group finds and returns the least busy CPU group within the | |
1304 | * domain. | |
1305 | */ | |
1306 | static struct sched_group * | |
1307 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1308 | { | |
1309 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1310 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1311 | int load_idx = sd->forkexec_idx; | |
1312 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1313 | ||
1314 | do { | |
1315 | unsigned long load, avg_load; | |
1316 | int local_group; | |
1317 | int i; | |
1318 | ||
da5a5522 BD |
1319 | /* Skip over this group if it has no CPUs allowed */ |
1320 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1321 | goto nextgroup; | |
1322 | ||
147cbb4b | 1323 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1324 | |
1325 | /* Tally up the load of all CPUs in the group */ | |
1326 | avg_load = 0; | |
1327 | ||
1328 | for_each_cpu_mask(i, group->cpumask) { | |
1329 | /* Bias balancing toward cpus of our domain */ | |
1330 | if (local_group) | |
1331 | load = source_load(i, load_idx); | |
1332 | else | |
1333 | load = target_load(i, load_idx); | |
1334 | ||
1335 | avg_load += load; | |
1336 | } | |
1337 | ||
1338 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1339 | avg_load = sg_div_cpu_power(group, |
1340 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1341 | |
1342 | if (local_group) { | |
1343 | this_load = avg_load; | |
1344 | this = group; | |
1345 | } else if (avg_load < min_load) { | |
1346 | min_load = avg_load; | |
1347 | idlest = group; | |
1348 | } | |
da5a5522 | 1349 | nextgroup: |
147cbb4b NP |
1350 | group = group->next; |
1351 | } while (group != sd->groups); | |
1352 | ||
1353 | if (!idlest || 100*this_load < imbalance*min_load) | |
1354 | return NULL; | |
1355 | return idlest; | |
1356 | } | |
1357 | ||
1358 | /* | |
0feaece9 | 1359 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1360 | */ |
95cdf3b7 IM |
1361 | static int |
1362 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1363 | { |
da5a5522 | 1364 | cpumask_t tmp; |
147cbb4b NP |
1365 | unsigned long load, min_load = ULONG_MAX; |
1366 | int idlest = -1; | |
1367 | int i; | |
1368 | ||
da5a5522 BD |
1369 | /* Traverse only the allowed CPUs */ |
1370 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1371 | ||
1372 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1373 | load = weighted_cpuload(i); |
147cbb4b NP |
1374 | |
1375 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1376 | min_load = load; | |
1377 | idlest = i; | |
1378 | } | |
1379 | } | |
1380 | ||
1381 | return idlest; | |
1382 | } | |
1383 | ||
476d139c NP |
1384 | /* |
1385 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1386 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1387 | * SD_BALANCE_EXEC. | |
1388 | * | |
1389 | * Balance, ie. select the least loaded group. | |
1390 | * | |
1391 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1392 | * | |
1393 | * preempt must be disabled. | |
1394 | */ | |
1395 | static int sched_balance_self(int cpu, int flag) | |
1396 | { | |
1397 | struct task_struct *t = current; | |
1398 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1399 | |
c96d145e | 1400 | for_each_domain(cpu, tmp) { |
5c45bf27 SS |
1401 | /* |
1402 | * If power savings logic is enabled for a domain, stop there. | |
1403 | */ | |
1404 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) | |
1405 | break; | |
476d139c NP |
1406 | if (tmp->flags & flag) |
1407 | sd = tmp; | |
c96d145e | 1408 | } |
476d139c NP |
1409 | |
1410 | while (sd) { | |
1411 | cpumask_t span; | |
1412 | struct sched_group *group; | |
1a848870 SS |
1413 | int new_cpu, weight; |
1414 | ||
1415 | if (!(sd->flags & flag)) { | |
1416 | sd = sd->child; | |
1417 | continue; | |
1418 | } | |
476d139c NP |
1419 | |
1420 | span = sd->span; | |
1421 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1422 | if (!group) { |
1423 | sd = sd->child; | |
1424 | continue; | |
1425 | } | |
476d139c | 1426 | |
da5a5522 | 1427 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1428 | if (new_cpu == -1 || new_cpu == cpu) { |
1429 | /* Now try balancing at a lower domain level of cpu */ | |
1430 | sd = sd->child; | |
1431 | continue; | |
1432 | } | |
476d139c | 1433 | |
1a848870 | 1434 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1435 | cpu = new_cpu; |
476d139c NP |
1436 | sd = NULL; |
1437 | weight = cpus_weight(span); | |
1438 | for_each_domain(cpu, tmp) { | |
1439 | if (weight <= cpus_weight(tmp->span)) | |
1440 | break; | |
1441 | if (tmp->flags & flag) | |
1442 | sd = tmp; | |
1443 | } | |
1444 | /* while loop will break here if sd == NULL */ | |
1445 | } | |
1446 | ||
1447 | return cpu; | |
1448 | } | |
1449 | ||
1450 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1451 | |
1452 | /* | |
1453 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1454 | * not idle and an idle cpu is available. The span of cpus to | |
1455 | * search starts with cpus closest then further out as needed, | |
1456 | * so we always favor a closer, idle cpu. | |
1457 | * | |
1458 | * Returns the CPU we should wake onto. | |
1459 | */ | |
1460 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1461 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1462 | { |
1463 | cpumask_t tmp; | |
1464 | struct sched_domain *sd; | |
1465 | int i; | |
1466 | ||
4953198b SS |
1467 | /* |
1468 | * If it is idle, then it is the best cpu to run this task. | |
1469 | * | |
1470 | * This cpu is also the best, if it has more than one task already. | |
1471 | * Siblings must be also busy(in most cases) as they didn't already | |
1472 | * pickup the extra load from this cpu and hence we need not check | |
1473 | * sibling runqueue info. This will avoid the checks and cache miss | |
1474 | * penalities associated with that. | |
1475 | */ | |
1476 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1477 | return cpu; |
1478 | ||
1479 | for_each_domain(cpu, sd) { | |
1480 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1481 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1482 | for_each_cpu_mask(i, tmp) { |
1483 | if (idle_cpu(i)) | |
1484 | return i; | |
1485 | } | |
1486 | } | |
e0f364f4 NP |
1487 | else |
1488 | break; | |
1da177e4 LT |
1489 | } |
1490 | return cpu; | |
1491 | } | |
1492 | #else | |
36c8b586 | 1493 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1494 | { |
1495 | return cpu; | |
1496 | } | |
1497 | #endif | |
1498 | ||
1499 | /*** | |
1500 | * try_to_wake_up - wake up a thread | |
1501 | * @p: the to-be-woken-up thread | |
1502 | * @state: the mask of task states that can be woken | |
1503 | * @sync: do a synchronous wakeup? | |
1504 | * | |
1505 | * Put it on the run-queue if it's not already there. The "current" | |
1506 | * thread is always on the run-queue (except when the actual | |
1507 | * re-schedule is in progress), and as such you're allowed to do | |
1508 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1509 | * runnable without the overhead of this. | |
1510 | * | |
1511 | * returns failure only if the task is already active. | |
1512 | */ | |
36c8b586 | 1513 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1514 | { |
1515 | int cpu, this_cpu, success = 0; | |
1516 | unsigned long flags; | |
1517 | long old_state; | |
70b97a7f | 1518 | struct rq *rq; |
1da177e4 | 1519 | #ifdef CONFIG_SMP |
7897986b | 1520 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1521 | unsigned long load, this_load; |
1da177e4 LT |
1522 | int new_cpu; |
1523 | #endif | |
1524 | ||
1525 | rq = task_rq_lock(p, &flags); | |
1526 | old_state = p->state; | |
1527 | if (!(old_state & state)) | |
1528 | goto out; | |
1529 | ||
1530 | if (p->array) | |
1531 | goto out_running; | |
1532 | ||
1533 | cpu = task_cpu(p); | |
1534 | this_cpu = smp_processor_id(); | |
1535 | ||
1536 | #ifdef CONFIG_SMP | |
1537 | if (unlikely(task_running(rq, p))) | |
1538 | goto out_activate; | |
1539 | ||
7897986b NP |
1540 | new_cpu = cpu; |
1541 | ||
1da177e4 LT |
1542 | schedstat_inc(rq, ttwu_cnt); |
1543 | if (cpu == this_cpu) { | |
1544 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1545 | goto out_set_cpu; |
1546 | } | |
1547 | ||
1548 | for_each_domain(this_cpu, sd) { | |
1549 | if (cpu_isset(cpu, sd->span)) { | |
1550 | schedstat_inc(sd, ttwu_wake_remote); | |
1551 | this_sd = sd; | |
1552 | break; | |
1da177e4 LT |
1553 | } |
1554 | } | |
1da177e4 | 1555 | |
7897986b | 1556 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1557 | goto out_set_cpu; |
1558 | ||
1da177e4 | 1559 | /* |
7897986b | 1560 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1561 | */ |
7897986b NP |
1562 | if (this_sd) { |
1563 | int idx = this_sd->wake_idx; | |
1564 | unsigned int imbalance; | |
1da177e4 | 1565 | |
a3f21bce NP |
1566 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1567 | ||
7897986b NP |
1568 | load = source_load(cpu, idx); |
1569 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1570 | |
7897986b NP |
1571 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1572 | ||
a3f21bce NP |
1573 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1574 | unsigned long tl = this_load; | |
33859f7f MOS |
1575 | unsigned long tl_per_task; |
1576 | ||
1577 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1578 | |
1da177e4 | 1579 | /* |
a3f21bce NP |
1580 | * If sync wakeup then subtract the (maximum possible) |
1581 | * effect of the currently running task from the load | |
1582 | * of the current CPU: | |
1da177e4 | 1583 | */ |
a3f21bce | 1584 | if (sync) |
2dd73a4f | 1585 | tl -= current->load_weight; |
a3f21bce NP |
1586 | |
1587 | if ((tl <= load && | |
2dd73a4f PW |
1588 | tl + target_load(cpu, idx) <= tl_per_task) || |
1589 | 100*(tl + p->load_weight) <= imbalance*load) { | |
a3f21bce NP |
1590 | /* |
1591 | * This domain has SD_WAKE_AFFINE and | |
1592 | * p is cache cold in this domain, and | |
1593 | * there is no bad imbalance. | |
1594 | */ | |
1595 | schedstat_inc(this_sd, ttwu_move_affine); | |
1596 | goto out_set_cpu; | |
1597 | } | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Start passive balancing when half the imbalance_pct | |
1602 | * limit is reached. | |
1603 | */ | |
1604 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1605 | if (imbalance*this_load <= 100*load) { | |
1606 | schedstat_inc(this_sd, ttwu_move_balance); | |
1607 | goto out_set_cpu; | |
1608 | } | |
1da177e4 LT |
1609 | } |
1610 | } | |
1611 | ||
1612 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1613 | out_set_cpu: | |
1614 | new_cpu = wake_idle(new_cpu, p); | |
1615 | if (new_cpu != cpu) { | |
1616 | set_task_cpu(p, new_cpu); | |
1617 | task_rq_unlock(rq, &flags); | |
1618 | /* might preempt at this point */ | |
1619 | rq = task_rq_lock(p, &flags); | |
1620 | old_state = p->state; | |
1621 | if (!(old_state & state)) | |
1622 | goto out; | |
1623 | if (p->array) | |
1624 | goto out_running; | |
1625 | ||
1626 | this_cpu = smp_processor_id(); | |
1627 | cpu = task_cpu(p); | |
1628 | } | |
1629 | ||
1630 | out_activate: | |
1631 | #endif /* CONFIG_SMP */ | |
1632 | if (old_state == TASK_UNINTERRUPTIBLE) { | |
1633 | rq->nr_uninterruptible--; | |
1634 | /* | |
1635 | * Tasks on involuntary sleep don't earn | |
1636 | * sleep_avg beyond just interactive state. | |
1637 | */ | |
3dee386e | 1638 | p->sleep_type = SLEEP_NONINTERACTIVE; |
e7c38cb4 | 1639 | } else |
1da177e4 | 1640 | |
d79fc0fc IM |
1641 | /* |
1642 | * Tasks that have marked their sleep as noninteractive get | |
e7c38cb4 CK |
1643 | * woken up with their sleep average not weighted in an |
1644 | * interactive way. | |
d79fc0fc | 1645 | */ |
e7c38cb4 CK |
1646 | if (old_state & TASK_NONINTERACTIVE) |
1647 | p->sleep_type = SLEEP_NONINTERACTIVE; | |
1648 | ||
1649 | ||
1650 | activate_task(p, rq, cpu == this_cpu); | |
1da177e4 LT |
1651 | /* |
1652 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1653 | * has indicated that it will leave the CPU in short order) | |
1654 | * don't trigger a preemption, if the woken up task will run on | |
1655 | * this cpu. (in this case the 'I will reschedule' promise of | |
1656 | * the waker guarantees that the freshly woken up task is going | |
1657 | * to be considered on this CPU.) | |
1658 | */ | |
1da177e4 LT |
1659 | if (!sync || cpu != this_cpu) { |
1660 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1661 | resched_task(rq->curr); | |
1662 | } | |
1663 | success = 1; | |
1664 | ||
1665 | out_running: | |
1666 | p->state = TASK_RUNNING; | |
1667 | out: | |
1668 | task_rq_unlock(rq, &flags); | |
1669 | ||
1670 | return success; | |
1671 | } | |
1672 | ||
36c8b586 | 1673 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1674 | { |
1675 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1676 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1677 | } | |
1da177e4 LT |
1678 | EXPORT_SYMBOL(wake_up_process); |
1679 | ||
36c8b586 | 1680 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1681 | { |
1682 | return try_to_wake_up(p, state, 0); | |
1683 | } | |
1684 | ||
bc947631 | 1685 | static void task_running_tick(struct rq *rq, struct task_struct *p); |
1da177e4 LT |
1686 | /* |
1687 | * Perform scheduler related setup for a newly forked process p. | |
1688 | * p is forked by current. | |
1689 | */ | |
36c8b586 | 1690 | void fastcall sched_fork(struct task_struct *p, int clone_flags) |
1da177e4 | 1691 | { |
476d139c NP |
1692 | int cpu = get_cpu(); |
1693 | ||
1694 | #ifdef CONFIG_SMP | |
1695 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1696 | #endif | |
1697 | set_task_cpu(p, cpu); | |
1698 | ||
1da177e4 LT |
1699 | /* |
1700 | * We mark the process as running here, but have not actually | |
1701 | * inserted it onto the runqueue yet. This guarantees that | |
1702 | * nobody will actually run it, and a signal or other external | |
1703 | * event cannot wake it up and insert it on the runqueue either. | |
1704 | */ | |
1705 | p->state = TASK_RUNNING; | |
b29739f9 IM |
1706 | |
1707 | /* | |
1708 | * Make sure we do not leak PI boosting priority to the child: | |
1709 | */ | |
1710 | p->prio = current->normal_prio; | |
1711 | ||
1da177e4 LT |
1712 | INIT_LIST_HEAD(&p->run_list); |
1713 | p->array = NULL; | |
52f17b6c CS |
1714 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
1715 | if (unlikely(sched_info_on())) | |
1716 | memset(&p->sched_info, 0, sizeof(p->sched_info)); | |
1da177e4 | 1717 | #endif |
d6077cb8 | 1718 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1719 | p->oncpu = 0; |
1720 | #endif | |
1da177e4 | 1721 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1722 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1723 | task_thread_info(p)->preempt_count = 1; |
1da177e4 LT |
1724 | #endif |
1725 | /* | |
1726 | * Share the timeslice between parent and child, thus the | |
1727 | * total amount of pending timeslices in the system doesn't change, | |
1728 | * resulting in more scheduling fairness. | |
1729 | */ | |
1730 | local_irq_disable(); | |
1731 | p->time_slice = (current->time_slice + 1) >> 1; | |
1732 | /* | |
1733 | * The remainder of the first timeslice might be recovered by | |
1734 | * the parent if the child exits early enough. | |
1735 | */ | |
1736 | p->first_time_slice = 1; | |
1737 | current->time_slice >>= 1; | |
1738 | p->timestamp = sched_clock(); | |
1739 | if (unlikely(!current->time_slice)) { | |
1740 | /* | |
1741 | * This case is rare, it happens when the parent has only | |
1742 | * a single jiffy left from its timeslice. Taking the | |
1743 | * runqueue lock is not a problem. | |
1744 | */ | |
1745 | current->time_slice = 1; | |
bc947631 | 1746 | task_running_tick(cpu_rq(cpu), current); |
476d139c NP |
1747 | } |
1748 | local_irq_enable(); | |
1749 | put_cpu(); | |
1da177e4 LT |
1750 | } |
1751 | ||
1752 | /* | |
1753 | * wake_up_new_task - wake up a newly created task for the first time. | |
1754 | * | |
1755 | * This function will do some initial scheduler statistics housekeeping | |
1756 | * that must be done for every newly created context, then puts the task | |
1757 | * on the runqueue and wakes it. | |
1758 | */ | |
36c8b586 | 1759 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 | 1760 | { |
70b97a7f | 1761 | struct rq *rq, *this_rq; |
1da177e4 LT |
1762 | unsigned long flags; |
1763 | int this_cpu, cpu; | |
1da177e4 LT |
1764 | |
1765 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1766 | BUG_ON(p->state != TASK_RUNNING); |
1da177e4 | 1767 | this_cpu = smp_processor_id(); |
147cbb4b | 1768 | cpu = task_cpu(p); |
1da177e4 | 1769 | |
1da177e4 LT |
1770 | /* |
1771 | * We decrease the sleep average of forking parents | |
1772 | * and children as well, to keep max-interactive tasks | |
1773 | * from forking tasks that are max-interactive. The parent | |
1774 | * (current) is done further down, under its lock. | |
1775 | */ | |
1776 | p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) * | |
1777 | CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1778 | ||
1779 | p->prio = effective_prio(p); | |
1780 | ||
1781 | if (likely(cpu == this_cpu)) { | |
1782 | if (!(clone_flags & CLONE_VM)) { | |
1783 | /* | |
1784 | * The VM isn't cloned, so we're in a good position to | |
1785 | * do child-runs-first in anticipation of an exec. This | |
1786 | * usually avoids a lot of COW overhead. | |
1787 | */ | |
1788 | if (unlikely(!current->array)) | |
1789 | __activate_task(p, rq); | |
1790 | else { | |
1791 | p->prio = current->prio; | |
b29739f9 | 1792 | p->normal_prio = current->normal_prio; |
1da177e4 LT |
1793 | list_add_tail(&p->run_list, ¤t->run_list); |
1794 | p->array = current->array; | |
1795 | p->array->nr_active++; | |
2dd73a4f | 1796 | inc_nr_running(p, rq); |
1da177e4 LT |
1797 | } |
1798 | set_need_resched(); | |
1799 | } else | |
1800 | /* Run child last */ | |
1801 | __activate_task(p, rq); | |
1802 | /* | |
1803 | * We skip the following code due to cpu == this_cpu | |
1804 | * | |
1805 | * task_rq_unlock(rq, &flags); | |
1806 | * this_rq = task_rq_lock(current, &flags); | |
1807 | */ | |
1808 | this_rq = rq; | |
1809 | } else { | |
1810 | this_rq = cpu_rq(this_cpu); | |
1811 | ||
1812 | /* | |
1813 | * Not the local CPU - must adjust timestamp. This should | |
1814 | * get optimised away in the !CONFIG_SMP case. | |
1815 | */ | |
b18ec803 MG |
1816 | p->timestamp = (p->timestamp - this_rq->most_recent_timestamp) |
1817 | + rq->most_recent_timestamp; | |
1da177e4 LT |
1818 | __activate_task(p, rq); |
1819 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1820 | resched_task(rq->curr); | |
1821 | ||
1822 | /* | |
1823 | * Parent and child are on different CPUs, now get the | |
1824 | * parent runqueue to update the parent's ->sleep_avg: | |
1825 | */ | |
1826 | task_rq_unlock(rq, &flags); | |
1827 | this_rq = task_rq_lock(current, &flags); | |
1828 | } | |
1829 | current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) * | |
1830 | PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1831 | task_rq_unlock(this_rq, &flags); | |
1832 | } | |
1833 | ||
4866cde0 NP |
1834 | /** |
1835 | * prepare_task_switch - prepare to switch tasks | |
1836 | * @rq: the runqueue preparing to switch | |
1837 | * @next: the task we are going to switch to. | |
1838 | * | |
1839 | * This is called with the rq lock held and interrupts off. It must | |
1840 | * be paired with a subsequent finish_task_switch after the context | |
1841 | * switch. | |
1842 | * | |
1843 | * prepare_task_switch sets up locking and calls architecture specific | |
1844 | * hooks. | |
1845 | */ | |
70b97a7f | 1846 | static inline void prepare_task_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
1847 | { |
1848 | prepare_lock_switch(rq, next); | |
1849 | prepare_arch_switch(next); | |
1850 | } | |
1851 | ||
1da177e4 LT |
1852 | /** |
1853 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1854 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1855 | * @prev: the thread we just switched away from. |
1856 | * | |
4866cde0 NP |
1857 | * finish_task_switch must be called after the context switch, paired |
1858 | * with a prepare_task_switch call before the context switch. | |
1859 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1860 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1861 | * |
1862 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1863 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1864 | * with the lock held can cause deadlocks; see schedule() for | |
1865 | * details.) | |
1866 | */ | |
70b97a7f | 1867 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1868 | __releases(rq->lock) |
1869 | { | |
1da177e4 | 1870 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1871 | long prev_state; |
1da177e4 LT |
1872 | |
1873 | rq->prev_mm = NULL; | |
1874 | ||
1875 | /* | |
1876 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1877 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1878 | * schedule one last time. The schedule call will never return, and |
1879 | * the scheduled task must drop that reference. | |
c394cc9f | 1880 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1881 | * still held, otherwise prev could be scheduled on another cpu, die |
1882 | * there before we look at prev->state, and then the reference would | |
1883 | * be dropped twice. | |
1884 | * Manfred Spraul <manfred@colorfullife.com> | |
1885 | */ | |
55a101f8 | 1886 | prev_state = prev->state; |
4866cde0 NP |
1887 | finish_arch_switch(prev); |
1888 | finish_lock_switch(rq, prev); | |
1da177e4 LT |
1889 | if (mm) |
1890 | mmdrop(mm); | |
c394cc9f | 1891 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1892 | /* |
1893 | * Remove function-return probe instances associated with this | |
1894 | * task and put them back on the free list. | |
1895 | */ | |
1896 | kprobe_flush_task(prev); | |
1da177e4 | 1897 | put_task_struct(prev); |
c6fd91f0 | 1898 | } |
1da177e4 LT |
1899 | } |
1900 | ||
1901 | /** | |
1902 | * schedule_tail - first thing a freshly forked thread must call. | |
1903 | * @prev: the thread we just switched away from. | |
1904 | */ | |
36c8b586 | 1905 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1906 | __releases(rq->lock) |
1907 | { | |
70b97a7f IM |
1908 | struct rq *rq = this_rq(); |
1909 | ||
4866cde0 NP |
1910 | finish_task_switch(rq, prev); |
1911 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1912 | /* In this case, finish_task_switch does not reenable preemption */ | |
1913 | preempt_enable(); | |
1914 | #endif | |
1da177e4 LT |
1915 | if (current->set_child_tid) |
1916 | put_user(current->pid, current->set_child_tid); | |
1917 | } | |
1918 | ||
1919 | /* | |
1920 | * context_switch - switch to the new MM and the new | |
1921 | * thread's register state. | |
1922 | */ | |
36c8b586 | 1923 | static inline struct task_struct * |
70b97a7f | 1924 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1925 | struct task_struct *next) |
1da177e4 LT |
1926 | { |
1927 | struct mm_struct *mm = next->mm; | |
1928 | struct mm_struct *oldmm = prev->active_mm; | |
1929 | ||
9226d125 ZA |
1930 | /* |
1931 | * For paravirt, this is coupled with an exit in switch_to to | |
1932 | * combine the page table reload and the switch backend into | |
1933 | * one hypercall. | |
1934 | */ | |
1935 | arch_enter_lazy_cpu_mode(); | |
1936 | ||
beed33a8 | 1937 | if (!mm) { |
1da177e4 LT |
1938 | next->active_mm = oldmm; |
1939 | atomic_inc(&oldmm->mm_count); | |
1940 | enter_lazy_tlb(oldmm, next); | |
1941 | } else | |
1942 | switch_mm(oldmm, mm, next); | |
1943 | ||
beed33a8 | 1944 | if (!prev->mm) { |
1da177e4 LT |
1945 | prev->active_mm = NULL; |
1946 | WARN_ON(rq->prev_mm); | |
1947 | rq->prev_mm = oldmm; | |
1948 | } | |
3a5f5e48 IM |
1949 | /* |
1950 | * Since the runqueue lock will be released by the next | |
1951 | * task (which is an invalid locking op but in the case | |
1952 | * of the scheduler it's an obvious special-case), so we | |
1953 | * do an early lockdep release here: | |
1954 | */ | |
1955 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1956 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1957 | #endif |
1da177e4 LT |
1958 | |
1959 | /* Here we just switch the register state and the stack. */ | |
1960 | switch_to(prev, next, prev); | |
1961 | ||
1962 | return prev; | |
1963 | } | |
1964 | ||
1965 | /* | |
1966 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1967 | * | |
1968 | * externally visible scheduler statistics: current number of runnable | |
1969 | * threads, current number of uninterruptible-sleeping threads, total | |
1970 | * number of context switches performed since bootup. | |
1971 | */ | |
1972 | unsigned long nr_running(void) | |
1973 | { | |
1974 | unsigned long i, sum = 0; | |
1975 | ||
1976 | for_each_online_cpu(i) | |
1977 | sum += cpu_rq(i)->nr_running; | |
1978 | ||
1979 | return sum; | |
1980 | } | |
1981 | ||
1982 | unsigned long nr_uninterruptible(void) | |
1983 | { | |
1984 | unsigned long i, sum = 0; | |
1985 | ||
0a945022 | 1986 | for_each_possible_cpu(i) |
1da177e4 LT |
1987 | sum += cpu_rq(i)->nr_uninterruptible; |
1988 | ||
1989 | /* | |
1990 | * Since we read the counters lockless, it might be slightly | |
1991 | * inaccurate. Do not allow it to go below zero though: | |
1992 | */ | |
1993 | if (unlikely((long)sum < 0)) | |
1994 | sum = 0; | |
1995 | ||
1996 | return sum; | |
1997 | } | |
1998 | ||
1999 | unsigned long long nr_context_switches(void) | |
2000 | { | |
cc94abfc SR |
2001 | int i; |
2002 | unsigned long long sum = 0; | |
1da177e4 | 2003 | |
0a945022 | 2004 | for_each_possible_cpu(i) |
1da177e4 LT |
2005 | sum += cpu_rq(i)->nr_switches; |
2006 | ||
2007 | return sum; | |
2008 | } | |
2009 | ||
2010 | unsigned long nr_iowait(void) | |
2011 | { | |
2012 | unsigned long i, sum = 0; | |
2013 | ||
0a945022 | 2014 | for_each_possible_cpu(i) |
1da177e4 LT |
2015 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2016 | ||
2017 | return sum; | |
2018 | } | |
2019 | ||
db1b1fef JS |
2020 | unsigned long nr_active(void) |
2021 | { | |
2022 | unsigned long i, running = 0, uninterruptible = 0; | |
2023 | ||
2024 | for_each_online_cpu(i) { | |
2025 | running += cpu_rq(i)->nr_running; | |
2026 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2027 | } | |
2028 | ||
2029 | if (unlikely((long)uninterruptible < 0)) | |
2030 | uninterruptible = 0; | |
2031 | ||
2032 | return running + uninterruptible; | |
2033 | } | |
2034 | ||
1da177e4 LT |
2035 | #ifdef CONFIG_SMP |
2036 | ||
48f24c4d IM |
2037 | /* |
2038 | * Is this task likely cache-hot: | |
2039 | */ | |
2040 | static inline int | |
2041 | task_hot(struct task_struct *p, unsigned long long now, struct sched_domain *sd) | |
2042 | { | |
2043 | return (long long)(now - p->last_ran) < (long long)sd->cache_hot_time; | |
2044 | } | |
2045 | ||
1da177e4 LT |
2046 | /* |
2047 | * double_rq_lock - safely lock two runqueues | |
2048 | * | |
2049 | * Note this does not disable interrupts like task_rq_lock, | |
2050 | * you need to do so manually before calling. | |
2051 | */ | |
70b97a7f | 2052 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2053 | __acquires(rq1->lock) |
2054 | __acquires(rq2->lock) | |
2055 | { | |
054b9108 | 2056 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2057 | if (rq1 == rq2) { |
2058 | spin_lock(&rq1->lock); | |
2059 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2060 | } else { | |
c96d145e | 2061 | if (rq1 < rq2) { |
1da177e4 LT |
2062 | spin_lock(&rq1->lock); |
2063 | spin_lock(&rq2->lock); | |
2064 | } else { | |
2065 | spin_lock(&rq2->lock); | |
2066 | spin_lock(&rq1->lock); | |
2067 | } | |
2068 | } | |
2069 | } | |
2070 | ||
2071 | /* | |
2072 | * double_rq_unlock - safely unlock two runqueues | |
2073 | * | |
2074 | * Note this does not restore interrupts like task_rq_unlock, | |
2075 | * you need to do so manually after calling. | |
2076 | */ | |
70b97a7f | 2077 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2078 | __releases(rq1->lock) |
2079 | __releases(rq2->lock) | |
2080 | { | |
2081 | spin_unlock(&rq1->lock); | |
2082 | if (rq1 != rq2) | |
2083 | spin_unlock(&rq2->lock); | |
2084 | else | |
2085 | __release(rq2->lock); | |
2086 | } | |
2087 | ||
2088 | /* | |
2089 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2090 | */ | |
70b97a7f | 2091 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2092 | __releases(this_rq->lock) |
2093 | __acquires(busiest->lock) | |
2094 | __acquires(this_rq->lock) | |
2095 | { | |
054b9108 KK |
2096 | if (unlikely(!irqs_disabled())) { |
2097 | /* printk() doesn't work good under rq->lock */ | |
2098 | spin_unlock(&this_rq->lock); | |
2099 | BUG_ON(1); | |
2100 | } | |
1da177e4 | 2101 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2102 | if (busiest < this_rq) { |
1da177e4 LT |
2103 | spin_unlock(&this_rq->lock); |
2104 | spin_lock(&busiest->lock); | |
2105 | spin_lock(&this_rq->lock); | |
2106 | } else | |
2107 | spin_lock(&busiest->lock); | |
2108 | } | |
2109 | } | |
2110 | ||
1da177e4 LT |
2111 | /* |
2112 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2113 | * This is accomplished by forcing the cpu_allowed mask to only | |
2114 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2115 | * the cpu_allowed mask is restored. | |
2116 | */ | |
36c8b586 | 2117 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2118 | { |
70b97a7f | 2119 | struct migration_req req; |
1da177e4 | 2120 | unsigned long flags; |
70b97a7f | 2121 | struct rq *rq; |
1da177e4 LT |
2122 | |
2123 | rq = task_rq_lock(p, &flags); | |
2124 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2125 | || unlikely(cpu_is_offline(dest_cpu))) | |
2126 | goto out; | |
2127 | ||
2128 | /* force the process onto the specified CPU */ | |
2129 | if (migrate_task(p, dest_cpu, &req)) { | |
2130 | /* Need to wait for migration thread (might exit: take ref). */ | |
2131 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2132 | |
1da177e4 LT |
2133 | get_task_struct(mt); |
2134 | task_rq_unlock(rq, &flags); | |
2135 | wake_up_process(mt); | |
2136 | put_task_struct(mt); | |
2137 | wait_for_completion(&req.done); | |
36c8b586 | 2138 | |
1da177e4 LT |
2139 | return; |
2140 | } | |
2141 | out: | |
2142 | task_rq_unlock(rq, &flags); | |
2143 | } | |
2144 | ||
2145 | /* | |
476d139c NP |
2146 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2147 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2148 | */ |
2149 | void sched_exec(void) | |
2150 | { | |
1da177e4 | 2151 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2152 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2153 | put_cpu(); |
476d139c NP |
2154 | if (new_cpu != this_cpu) |
2155 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2156 | } |
2157 | ||
2158 | /* | |
2159 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2160 | * Both runqueues must be locked. | |
2161 | */ | |
70b97a7f IM |
2162 | static void pull_task(struct rq *src_rq, struct prio_array *src_array, |
2163 | struct task_struct *p, struct rq *this_rq, | |
2164 | struct prio_array *this_array, int this_cpu) | |
1da177e4 LT |
2165 | { |
2166 | dequeue_task(p, src_array); | |
2dd73a4f | 2167 | dec_nr_running(p, src_rq); |
1da177e4 | 2168 | set_task_cpu(p, this_cpu); |
2dd73a4f | 2169 | inc_nr_running(p, this_rq); |
1da177e4 | 2170 | enqueue_task(p, this_array); |
b18ec803 MG |
2171 | p->timestamp = (p->timestamp - src_rq->most_recent_timestamp) |
2172 | + this_rq->most_recent_timestamp; | |
1da177e4 LT |
2173 | /* |
2174 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2175 | * to be always true for them. | |
2176 | */ | |
2177 | if (TASK_PREEMPTS_CURR(p, this_rq)) | |
2178 | resched_task(this_rq->curr); | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2183 | */ | |
858119e1 | 2184 | static |
70b97a7f | 2185 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2186 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2187 | int *all_pinned) |
1da177e4 LT |
2188 | { |
2189 | /* | |
2190 | * We do not migrate tasks that are: | |
2191 | * 1) running (obviously), or | |
2192 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2193 | * 3) are cache-hot on their current CPU. | |
2194 | */ | |
1da177e4 LT |
2195 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2196 | return 0; | |
81026794 NP |
2197 | *all_pinned = 0; |
2198 | ||
2199 | if (task_running(rq, p)) | |
2200 | return 0; | |
1da177e4 LT |
2201 | |
2202 | /* | |
2203 | * Aggressive migration if: | |
cafb20c1 | 2204 | * 1) task is cache cold, or |
1da177e4 LT |
2205 | * 2) too many balance attempts have failed. |
2206 | */ | |
2207 | ||
b18ec803 MG |
2208 | if (sd->nr_balance_failed > sd->cache_nice_tries) { |
2209 | #ifdef CONFIG_SCHEDSTATS | |
2210 | if (task_hot(p, rq->most_recent_timestamp, sd)) | |
2211 | schedstat_inc(sd, lb_hot_gained[idle]); | |
2212 | #endif | |
1da177e4 | 2213 | return 1; |
b18ec803 | 2214 | } |
1da177e4 | 2215 | |
b18ec803 | 2216 | if (task_hot(p, rq->most_recent_timestamp, sd)) |
81026794 | 2217 | return 0; |
1da177e4 LT |
2218 | return 1; |
2219 | } | |
2220 | ||
615052dc | 2221 | #define rq_best_prio(rq) min((rq)->curr->prio, (rq)->best_expired_prio) |
48f24c4d | 2222 | |
1da177e4 | 2223 | /* |
2dd73a4f PW |
2224 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted |
2225 | * load from busiest to this_rq, as part of a balancing operation within | |
2226 | * "domain". Returns the number of tasks moved. | |
1da177e4 LT |
2227 | * |
2228 | * Called with both runqueues locked. | |
2229 | */ | |
70b97a7f | 2230 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2231 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2232 | struct sched_domain *sd, enum cpu_idle_type idle, |
2dd73a4f | 2233 | int *all_pinned) |
1da177e4 | 2234 | { |
48f24c4d IM |
2235 | int idx, pulled = 0, pinned = 0, this_best_prio, best_prio, |
2236 | best_prio_seen, skip_for_load; | |
70b97a7f | 2237 | struct prio_array *array, *dst_array; |
1da177e4 | 2238 | struct list_head *head, *curr; |
36c8b586 | 2239 | struct task_struct *tmp; |
2dd73a4f | 2240 | long rem_load_move; |
1da177e4 | 2241 | |
2dd73a4f | 2242 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2243 | goto out; |
2244 | ||
2dd73a4f | 2245 | rem_load_move = max_load_move; |
81026794 | 2246 | pinned = 1; |
615052dc | 2247 | this_best_prio = rq_best_prio(this_rq); |
48f24c4d | 2248 | best_prio = rq_best_prio(busiest); |
615052dc PW |
2249 | /* |
2250 | * Enable handling of the case where there is more than one task | |
2251 | * with the best priority. If the current running task is one | |
48f24c4d | 2252 | * of those with prio==best_prio we know it won't be moved |
615052dc PW |
2253 | * and therefore it's safe to override the skip (based on load) of |
2254 | * any task we find with that prio. | |
2255 | */ | |
48f24c4d | 2256 | best_prio_seen = best_prio == busiest->curr->prio; |
81026794 | 2257 | |
1da177e4 LT |
2258 | /* |
2259 | * We first consider expired tasks. Those will likely not be | |
2260 | * executed in the near future, and they are most likely to | |
2261 | * be cache-cold, thus switching CPUs has the least effect | |
2262 | * on them. | |
2263 | */ | |
2264 | if (busiest->expired->nr_active) { | |
2265 | array = busiest->expired; | |
2266 | dst_array = this_rq->expired; | |
2267 | } else { | |
2268 | array = busiest->active; | |
2269 | dst_array = this_rq->active; | |
2270 | } | |
2271 | ||
2272 | new_array: | |
2273 | /* Start searching at priority 0: */ | |
2274 | idx = 0; | |
2275 | skip_bitmap: | |
2276 | if (!idx) | |
2277 | idx = sched_find_first_bit(array->bitmap); | |
2278 | else | |
2279 | idx = find_next_bit(array->bitmap, MAX_PRIO, idx); | |
2280 | if (idx >= MAX_PRIO) { | |
2281 | if (array == busiest->expired && busiest->active->nr_active) { | |
2282 | array = busiest->active; | |
2283 | dst_array = this_rq->active; | |
2284 | goto new_array; | |
2285 | } | |
2286 | goto out; | |
2287 | } | |
2288 | ||
2289 | head = array->queue + idx; | |
2290 | curr = head->prev; | |
2291 | skip_queue: | |
36c8b586 | 2292 | tmp = list_entry(curr, struct task_struct, run_list); |
1da177e4 LT |
2293 | |
2294 | curr = curr->prev; | |
2295 | ||
50ddd969 PW |
2296 | /* |
2297 | * To help distribute high priority tasks accross CPUs we don't | |
2298 | * skip a task if it will be the highest priority task (i.e. smallest | |
2299 | * prio value) on its new queue regardless of its load weight | |
2300 | */ | |
615052dc PW |
2301 | skip_for_load = tmp->load_weight > rem_load_move; |
2302 | if (skip_for_load && idx < this_best_prio) | |
48f24c4d | 2303 | skip_for_load = !best_prio_seen && idx == best_prio; |
615052dc | 2304 | if (skip_for_load || |
2dd73a4f | 2305 | !can_migrate_task(tmp, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d IM |
2306 | |
2307 | best_prio_seen |= idx == best_prio; | |
1da177e4 LT |
2308 | if (curr != head) |
2309 | goto skip_queue; | |
2310 | idx++; | |
2311 | goto skip_bitmap; | |
2312 | } | |
2313 | ||
1da177e4 LT |
2314 | pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu); |
2315 | pulled++; | |
2dd73a4f | 2316 | rem_load_move -= tmp->load_weight; |
1da177e4 | 2317 | |
2dd73a4f PW |
2318 | /* |
2319 | * We only want to steal up to the prescribed number of tasks | |
2320 | * and the prescribed amount of weighted load. | |
2321 | */ | |
2322 | if (pulled < max_nr_move && rem_load_move > 0) { | |
615052dc PW |
2323 | if (idx < this_best_prio) |
2324 | this_best_prio = idx; | |
1da177e4 LT |
2325 | if (curr != head) |
2326 | goto skip_queue; | |
2327 | idx++; | |
2328 | goto skip_bitmap; | |
2329 | } | |
2330 | out: | |
2331 | /* | |
2332 | * Right now, this is the only place pull_task() is called, | |
2333 | * so we can safely collect pull_task() stats here rather than | |
2334 | * inside pull_task(). | |
2335 | */ | |
2336 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2337 | |
2338 | if (all_pinned) | |
2339 | *all_pinned = pinned; | |
1da177e4 LT |
2340 | return pulled; |
2341 | } | |
2342 | ||
2343 | /* | |
2344 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2345 | * domain. It calculates and returns the amount of weighted load which |
2346 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2347 | */ |
2348 | static struct sched_group * | |
2349 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
d15bcfdb | 2350 | unsigned long *imbalance, enum cpu_idle_type idle, int *sd_idle, |
783609c6 | 2351 | cpumask_t *cpus, int *balance) |
1da177e4 LT |
2352 | { |
2353 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2354 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2355 | unsigned long max_pull; |
2dd73a4f PW |
2356 | unsigned long busiest_load_per_task, busiest_nr_running; |
2357 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2358 | int load_idx; |
5c45bf27 SS |
2359 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2360 | int power_savings_balance = 1; | |
2361 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2362 | unsigned long min_nr_running = ULONG_MAX; | |
2363 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2364 | #endif | |
1da177e4 LT |
2365 | |
2366 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2367 | busiest_load_per_task = busiest_nr_running = 0; |
2368 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2369 | if (idle == CPU_NOT_IDLE) |
7897986b | 2370 | load_idx = sd->busy_idx; |
d15bcfdb | 2371 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2372 | load_idx = sd->newidle_idx; |
2373 | else | |
2374 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2375 | |
2376 | do { | |
5c45bf27 | 2377 | unsigned long load, group_capacity; |
1da177e4 LT |
2378 | int local_group; |
2379 | int i; | |
783609c6 | 2380 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2381 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2382 | |
2383 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2384 | ||
783609c6 SS |
2385 | if (local_group) |
2386 | balance_cpu = first_cpu(group->cpumask); | |
2387 | ||
1da177e4 | 2388 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2389 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2390 | |
2391 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2392 | struct rq *rq; |
2393 | ||
2394 | if (!cpu_isset(i, *cpus)) | |
2395 | continue; | |
2396 | ||
2397 | rq = cpu_rq(i); | |
2dd73a4f | 2398 | |
5969fe06 NP |
2399 | if (*sd_idle && !idle_cpu(i)) |
2400 | *sd_idle = 0; | |
2401 | ||
1da177e4 | 2402 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2403 | if (local_group) { |
2404 | if (idle_cpu(i) && !first_idle_cpu) { | |
2405 | first_idle_cpu = 1; | |
2406 | balance_cpu = i; | |
2407 | } | |
2408 | ||
a2000572 | 2409 | load = target_load(i, load_idx); |
783609c6 | 2410 | } else |
a2000572 | 2411 | load = source_load(i, load_idx); |
1da177e4 LT |
2412 | |
2413 | avg_load += load; | |
2dd73a4f PW |
2414 | sum_nr_running += rq->nr_running; |
2415 | sum_weighted_load += rq->raw_weighted_load; | |
1da177e4 LT |
2416 | } |
2417 | ||
783609c6 SS |
2418 | /* |
2419 | * First idle cpu or the first cpu(busiest) in this sched group | |
2420 | * is eligible for doing load balancing at this and above | |
2421 | * domains. | |
2422 | */ | |
2423 | if (local_group && balance_cpu != this_cpu && balance) { | |
2424 | *balance = 0; | |
2425 | goto ret; | |
2426 | } | |
2427 | ||
1da177e4 | 2428 | total_load += avg_load; |
5517d86b | 2429 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2430 | |
2431 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2432 | avg_load = sg_div_cpu_power(group, |
2433 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2434 | |
5517d86b | 2435 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2436 | |
1da177e4 LT |
2437 | if (local_group) { |
2438 | this_load = avg_load; | |
2439 | this = group; | |
2dd73a4f PW |
2440 | this_nr_running = sum_nr_running; |
2441 | this_load_per_task = sum_weighted_load; | |
2442 | } else if (avg_load > max_load && | |
5c45bf27 | 2443 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2444 | max_load = avg_load; |
2445 | busiest = group; | |
2dd73a4f PW |
2446 | busiest_nr_running = sum_nr_running; |
2447 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2448 | } |
5c45bf27 SS |
2449 | |
2450 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2451 | /* | |
2452 | * Busy processors will not participate in power savings | |
2453 | * balance. | |
2454 | */ | |
d15bcfdb | 2455 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 SS |
2456 | goto group_next; |
2457 | ||
2458 | /* | |
2459 | * If the local group is idle or completely loaded | |
2460 | * no need to do power savings balance at this domain | |
2461 | */ | |
2462 | if (local_group && (this_nr_running >= group_capacity || | |
2463 | !this_nr_running)) | |
2464 | power_savings_balance = 0; | |
2465 | ||
2466 | /* | |
2467 | * If a group is already running at full capacity or idle, | |
2468 | * don't include that group in power savings calculations | |
2469 | */ | |
2470 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
2471 | || !sum_nr_running) | |
2472 | goto group_next; | |
2473 | ||
2474 | /* | |
2475 | * Calculate the group which has the least non-idle load. | |
2476 | * This is the group from where we need to pick up the load | |
2477 | * for saving power | |
2478 | */ | |
2479 | if ((sum_nr_running < min_nr_running) || | |
2480 | (sum_nr_running == min_nr_running && | |
2481 | first_cpu(group->cpumask) < | |
2482 | first_cpu(group_min->cpumask))) { | |
2483 | group_min = group; | |
2484 | min_nr_running = sum_nr_running; | |
2485 | min_load_per_task = sum_weighted_load / | |
2486 | sum_nr_running; | |
2487 | } | |
2488 | ||
2489 | /* | |
2490 | * Calculate the group which is almost near its | |
2491 | * capacity but still has some space to pick up some load | |
2492 | * from other group and save more power | |
2493 | */ | |
48f24c4d | 2494 | if (sum_nr_running <= group_capacity - 1) { |
5c45bf27 SS |
2495 | if (sum_nr_running > leader_nr_running || |
2496 | (sum_nr_running == leader_nr_running && | |
2497 | first_cpu(group->cpumask) > | |
2498 | first_cpu(group_leader->cpumask))) { | |
2499 | group_leader = group; | |
2500 | leader_nr_running = sum_nr_running; | |
2501 | } | |
48f24c4d | 2502 | } |
5c45bf27 SS |
2503 | group_next: |
2504 | #endif | |
1da177e4 LT |
2505 | group = group->next; |
2506 | } while (group != sd->groups); | |
2507 | ||
2dd73a4f | 2508 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2509 | goto out_balanced; |
2510 | ||
2511 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2512 | ||
2513 | if (this_load >= avg_load || | |
2514 | 100*max_load <= sd->imbalance_pct*this_load) | |
2515 | goto out_balanced; | |
2516 | ||
2dd73a4f | 2517 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2518 | /* |
2519 | * We're trying to get all the cpus to the average_load, so we don't | |
2520 | * want to push ourselves above the average load, nor do we wish to | |
2521 | * reduce the max loaded cpu below the average load, as either of these | |
2522 | * actions would just result in more rebalancing later, and ping-pong | |
2523 | * tasks around. Thus we look for the minimum possible imbalance. | |
2524 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2525 | * be counted as no imbalance for these purposes -- we can't fix that | |
2526 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2527 | * appear as very large values with unsigned longs. | |
2528 | */ | |
2dd73a4f PW |
2529 | if (max_load <= busiest_load_per_task) |
2530 | goto out_balanced; | |
2531 | ||
2532 | /* | |
2533 | * In the presence of smp nice balancing, certain scenarios can have | |
2534 | * max load less than avg load(as we skip the groups at or below | |
2535 | * its cpu_power, while calculating max_load..) | |
2536 | */ | |
2537 | if (max_load < avg_load) { | |
2538 | *imbalance = 0; | |
2539 | goto small_imbalance; | |
2540 | } | |
0c117f1b SS |
2541 | |
2542 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2543 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2544 | |
1da177e4 | 2545 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2546 | *imbalance = min(max_pull * busiest->__cpu_power, |
2547 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2548 | / SCHED_LOAD_SCALE; |
2549 | ||
2dd73a4f PW |
2550 | /* |
2551 | * if *imbalance is less than the average load per runnable task | |
2552 | * there is no gaurantee that any tasks will be moved so we'll have | |
2553 | * a think about bumping its value to force at least one task to be | |
2554 | * moved | |
2555 | */ | |
2556 | if (*imbalance < busiest_load_per_task) { | |
48f24c4d | 2557 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2558 | unsigned int imbn; |
2559 | ||
2560 | small_imbalance: | |
2561 | pwr_move = pwr_now = 0; | |
2562 | imbn = 2; | |
2563 | if (this_nr_running) { | |
2564 | this_load_per_task /= this_nr_running; | |
2565 | if (busiest_load_per_task > this_load_per_task) | |
2566 | imbn = 1; | |
2567 | } else | |
2568 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2569 | |
2dd73a4f PW |
2570 | if (max_load - this_load >= busiest_load_per_task * imbn) { |
2571 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2572 | return busiest; |
2573 | } | |
2574 | ||
2575 | /* | |
2576 | * OK, we don't have enough imbalance to justify moving tasks, | |
2577 | * however we may be able to increase total CPU power used by | |
2578 | * moving them. | |
2579 | */ | |
2580 | ||
5517d86b ED |
2581 | pwr_now += busiest->__cpu_power * |
2582 | min(busiest_load_per_task, max_load); | |
2583 | pwr_now += this->__cpu_power * | |
2584 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2585 | pwr_now /= SCHED_LOAD_SCALE; |
2586 | ||
2587 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2588 | tmp = sg_div_cpu_power(busiest, |
2589 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2590 | if (max_load > tmp) |
5517d86b | 2591 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2592 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2593 | |
2594 | /* Amount of load we'd add */ | |
5517d86b | 2595 | if (max_load * busiest->__cpu_power < |
33859f7f | 2596 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2597 | tmp = sg_div_cpu_power(this, |
2598 | max_load * busiest->__cpu_power); | |
1da177e4 | 2599 | else |
5517d86b ED |
2600 | tmp = sg_div_cpu_power(this, |
2601 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2602 | pwr_move += this->__cpu_power * | |
2603 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2604 | pwr_move /= SCHED_LOAD_SCALE; |
2605 | ||
2606 | /* Move if we gain throughput */ | |
2607 | if (pwr_move <= pwr_now) | |
2608 | goto out_balanced; | |
2609 | ||
2dd73a4f | 2610 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2611 | } |
2612 | ||
1da177e4 LT |
2613 | return busiest; |
2614 | ||
2615 | out_balanced: | |
5c45bf27 | 2616 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2617 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2618 | goto ret; |
1da177e4 | 2619 | |
5c45bf27 SS |
2620 | if (this == group_leader && group_leader != group_min) { |
2621 | *imbalance = min_load_per_task; | |
2622 | return group_min; | |
2623 | } | |
5c45bf27 | 2624 | #endif |
783609c6 | 2625 | ret: |
1da177e4 LT |
2626 | *imbalance = 0; |
2627 | return NULL; | |
2628 | } | |
2629 | ||
2630 | /* | |
2631 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2632 | */ | |
70b97a7f | 2633 | static struct rq * |
d15bcfdb | 2634 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2635 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2636 | { |
70b97a7f | 2637 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2638 | unsigned long max_load = 0; |
1da177e4 LT |
2639 | int i; |
2640 | ||
2641 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2642 | |
2643 | if (!cpu_isset(i, *cpus)) | |
2644 | continue; | |
2645 | ||
48f24c4d | 2646 | rq = cpu_rq(i); |
2dd73a4f | 2647 | |
48f24c4d | 2648 | if (rq->nr_running == 1 && rq->raw_weighted_load > imbalance) |
2dd73a4f | 2649 | continue; |
1da177e4 | 2650 | |
48f24c4d IM |
2651 | if (rq->raw_weighted_load > max_load) { |
2652 | max_load = rq->raw_weighted_load; | |
2653 | busiest = rq; | |
1da177e4 LT |
2654 | } |
2655 | } | |
2656 | ||
2657 | return busiest; | |
2658 | } | |
2659 | ||
77391d71 NP |
2660 | /* |
2661 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2662 | * so long as it is large enough. | |
2663 | */ | |
2664 | #define MAX_PINNED_INTERVAL 512 | |
2665 | ||
48f24c4d IM |
2666 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2667 | { | |
2668 | return n > 0 ? n - 1 : 0; | |
2669 | } | |
2670 | ||
1da177e4 LT |
2671 | /* |
2672 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2673 | * tasks if there is an imbalance. | |
1da177e4 | 2674 | */ |
70b97a7f | 2675 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2676 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2677 | int *balance) |
1da177e4 | 2678 | { |
48f24c4d | 2679 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2680 | struct sched_group *group; |
1da177e4 | 2681 | unsigned long imbalance; |
70b97a7f | 2682 | struct rq *busiest; |
0a2966b4 | 2683 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2684 | unsigned long flags; |
5969fe06 | 2685 | |
89c4710e SS |
2686 | /* |
2687 | * When power savings policy is enabled for the parent domain, idle | |
2688 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2689 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2690 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2691 | */ |
d15bcfdb | 2692 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2693 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2694 | sd_idle = 1; |
1da177e4 | 2695 | |
1da177e4 LT |
2696 | schedstat_inc(sd, lb_cnt[idle]); |
2697 | ||
0a2966b4 CL |
2698 | redo: |
2699 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2700 | &cpus, balance); |
2701 | ||
06066714 | 2702 | if (*balance == 0) |
783609c6 | 2703 | goto out_balanced; |
783609c6 | 2704 | |
1da177e4 LT |
2705 | if (!group) { |
2706 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2707 | goto out_balanced; | |
2708 | } | |
2709 | ||
0a2966b4 | 2710 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2711 | if (!busiest) { |
2712 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2713 | goto out_balanced; | |
2714 | } | |
2715 | ||
db935dbd | 2716 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2717 | |
2718 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2719 | ||
2720 | nr_moved = 0; | |
2721 | if (busiest->nr_running > 1) { | |
2722 | /* | |
2723 | * Attempt to move tasks. If find_busiest_group has found | |
2724 | * an imbalance but busiest->nr_running <= 1, the group is | |
2725 | * still unbalanced. nr_moved simply stays zero, so it is | |
2726 | * correctly treated as an imbalance. | |
2727 | */ | |
fe2eea3f | 2728 | local_irq_save(flags); |
e17224bf | 2729 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2730 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2731 | minus_1_or_zero(busiest->nr_running), |
2732 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2733 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2734 | local_irq_restore(flags); |
81026794 | 2735 | |
46cb4b7c SS |
2736 | /* |
2737 | * some other cpu did the load balance for us. | |
2738 | */ | |
2739 | if (nr_moved && this_cpu != smp_processor_id()) | |
2740 | resched_cpu(this_cpu); | |
2741 | ||
81026794 | 2742 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2743 | if (unlikely(all_pinned)) { |
2744 | cpu_clear(cpu_of(busiest), cpus); | |
2745 | if (!cpus_empty(cpus)) | |
2746 | goto redo; | |
81026794 | 2747 | goto out_balanced; |
0a2966b4 | 2748 | } |
1da177e4 | 2749 | } |
81026794 | 2750 | |
1da177e4 LT |
2751 | if (!nr_moved) { |
2752 | schedstat_inc(sd, lb_failed[idle]); | |
2753 | sd->nr_balance_failed++; | |
2754 | ||
2755 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2756 | |
fe2eea3f | 2757 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2758 | |
2759 | /* don't kick the migration_thread, if the curr | |
2760 | * task on busiest cpu can't be moved to this_cpu | |
2761 | */ | |
2762 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2763 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2764 | all_pinned = 1; |
2765 | goto out_one_pinned; | |
2766 | } | |
2767 | ||
1da177e4 LT |
2768 | if (!busiest->active_balance) { |
2769 | busiest->active_balance = 1; | |
2770 | busiest->push_cpu = this_cpu; | |
81026794 | 2771 | active_balance = 1; |
1da177e4 | 2772 | } |
fe2eea3f | 2773 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2774 | if (active_balance) |
1da177e4 LT |
2775 | wake_up_process(busiest->migration_thread); |
2776 | ||
2777 | /* | |
2778 | * We've kicked active balancing, reset the failure | |
2779 | * counter. | |
2780 | */ | |
39507451 | 2781 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2782 | } |
81026794 | 2783 | } else |
1da177e4 LT |
2784 | sd->nr_balance_failed = 0; |
2785 | ||
81026794 | 2786 | if (likely(!active_balance)) { |
1da177e4 LT |
2787 | /* We were unbalanced, so reset the balancing interval */ |
2788 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2789 | } else { |
2790 | /* | |
2791 | * If we've begun active balancing, start to back off. This | |
2792 | * case may not be covered by the all_pinned logic if there | |
2793 | * is only 1 task on the busy runqueue (because we don't call | |
2794 | * move_tasks). | |
2795 | */ | |
2796 | if (sd->balance_interval < sd->max_interval) | |
2797 | sd->balance_interval *= 2; | |
1da177e4 LT |
2798 | } |
2799 | ||
5c45bf27 | 2800 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2801 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2802 | return -1; |
1da177e4 LT |
2803 | return nr_moved; |
2804 | ||
2805 | out_balanced: | |
1da177e4 LT |
2806 | schedstat_inc(sd, lb_balanced[idle]); |
2807 | ||
16cfb1c0 | 2808 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2809 | |
2810 | out_one_pinned: | |
1da177e4 | 2811 | /* tune up the balancing interval */ |
77391d71 NP |
2812 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2813 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2814 | sd->balance_interval *= 2; |
2815 | ||
48f24c4d | 2816 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2817 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2818 | return -1; |
1da177e4 LT |
2819 | return 0; |
2820 | } | |
2821 | ||
2822 | /* | |
2823 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2824 | * tasks if there is an imbalance. | |
2825 | * | |
d15bcfdb | 2826 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2827 | * this_rq is locked. |
2828 | */ | |
48f24c4d | 2829 | static int |
70b97a7f | 2830 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2831 | { |
2832 | struct sched_group *group; | |
70b97a7f | 2833 | struct rq *busiest = NULL; |
1da177e4 LT |
2834 | unsigned long imbalance; |
2835 | int nr_moved = 0; | |
5969fe06 | 2836 | int sd_idle = 0; |
0a2966b4 | 2837 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2838 | |
89c4710e SS |
2839 | /* |
2840 | * When power savings policy is enabled for the parent domain, idle | |
2841 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2842 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2843 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2844 | */ |
2845 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2846 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2847 | sd_idle = 1; |
1da177e4 | 2848 | |
d15bcfdb | 2849 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2850 | redo: |
d15bcfdb | 2851 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2852 | &sd_idle, &cpus, NULL); |
1da177e4 | 2853 | if (!group) { |
d15bcfdb | 2854 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2855 | goto out_balanced; |
1da177e4 LT |
2856 | } |
2857 | ||
d15bcfdb | 2858 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2859 | &cpus); |
db935dbd | 2860 | if (!busiest) { |
d15bcfdb | 2861 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2862 | goto out_balanced; |
1da177e4 LT |
2863 | } |
2864 | ||
db935dbd NP |
2865 | BUG_ON(busiest == this_rq); |
2866 | ||
d15bcfdb | 2867 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2868 | |
2869 | nr_moved = 0; | |
2870 | if (busiest->nr_running > 1) { | |
2871 | /* Attempt to move tasks */ | |
2872 | double_lock_balance(this_rq, busiest); | |
2873 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2874 | minus_1_or_zero(busiest->nr_running), |
d15bcfdb | 2875 | imbalance, sd, CPU_NEWLY_IDLE, NULL); |
d6d5cfaf | 2876 | spin_unlock(&busiest->lock); |
0a2966b4 CL |
2877 | |
2878 | if (!nr_moved) { | |
2879 | cpu_clear(cpu_of(busiest), cpus); | |
2880 | if (!cpus_empty(cpus)) | |
2881 | goto redo; | |
2882 | } | |
d6d5cfaf NP |
2883 | } |
2884 | ||
5969fe06 | 2885 | if (!nr_moved) { |
d15bcfdb | 2886 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2887 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2888 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2889 | return -1; |
2890 | } else | |
16cfb1c0 | 2891 | sd->nr_balance_failed = 0; |
1da177e4 | 2892 | |
1da177e4 | 2893 | return nr_moved; |
16cfb1c0 NP |
2894 | |
2895 | out_balanced: | |
d15bcfdb | 2896 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2897 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2898 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2899 | return -1; |
16cfb1c0 | 2900 | sd->nr_balance_failed = 0; |
48f24c4d | 2901 | |
16cfb1c0 | 2902 | return 0; |
1da177e4 LT |
2903 | } |
2904 | ||
2905 | /* | |
2906 | * idle_balance is called by schedule() if this_cpu is about to become | |
2907 | * idle. Attempts to pull tasks from other CPUs. | |
2908 | */ | |
70b97a7f | 2909 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2910 | { |
2911 | struct sched_domain *sd; | |
1bd77f2d CL |
2912 | int pulled_task = 0; |
2913 | unsigned long next_balance = jiffies + 60 * HZ; | |
1da177e4 LT |
2914 | |
2915 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2916 | unsigned long interval; |
2917 | ||
2918 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2919 | continue; | |
2920 | ||
2921 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2922 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2923 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2924 | this_rq, sd); |
2925 | ||
2926 | interval = msecs_to_jiffies(sd->balance_interval); | |
2927 | if (time_after(next_balance, sd->last_balance + interval)) | |
2928 | next_balance = sd->last_balance + interval; | |
2929 | if (pulled_task) | |
2930 | break; | |
1da177e4 | 2931 | } |
1bd77f2d CL |
2932 | if (!pulled_task) |
2933 | /* | |
2934 | * We are going idle. next_balance may be set based on | |
2935 | * a busy processor. So reset next_balance. | |
2936 | */ | |
2937 | this_rq->next_balance = next_balance; | |
1da177e4 LT |
2938 | } |
2939 | ||
2940 | /* | |
2941 | * active_load_balance is run by migration threads. It pushes running tasks | |
2942 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2943 | * running on each physical CPU where possible, and avoids physical / | |
2944 | * logical imbalances. | |
2945 | * | |
2946 | * Called with busiest_rq locked. | |
2947 | */ | |
70b97a7f | 2948 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2949 | { |
39507451 | 2950 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2951 | struct sched_domain *sd; |
2952 | struct rq *target_rq; | |
39507451 | 2953 | |
48f24c4d | 2954 | /* Is there any task to move? */ |
39507451 | 2955 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2956 | return; |
2957 | ||
2958 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2959 | |
2960 | /* | |
39507451 NP |
2961 | * This condition is "impossible", if it occurs |
2962 | * we need to fix it. Originally reported by | |
2963 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2964 | */ |
39507451 | 2965 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2966 | |
39507451 NP |
2967 | /* move a task from busiest_rq to target_rq */ |
2968 | double_lock_balance(busiest_rq, target_rq); | |
2969 | ||
2970 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2971 | for_each_domain(target_cpu, sd) { |
39507451 | 2972 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2973 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2974 | break; |
c96d145e | 2975 | } |
39507451 | 2976 | |
48f24c4d IM |
2977 | if (likely(sd)) { |
2978 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2979 | |
48f24c4d | 2980 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
d15bcfdb | 2981 | RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE, |
48f24c4d IM |
2982 | NULL)) |
2983 | schedstat_inc(sd, alb_pushed); | |
2984 | else | |
2985 | schedstat_inc(sd, alb_failed); | |
2986 | } | |
39507451 | 2987 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2988 | } |
2989 | ||
7835b98b | 2990 | static void update_load(struct rq *this_rq) |
1da177e4 | 2991 | { |
7835b98b | 2992 | unsigned long this_load; |
ff91691b | 2993 | unsigned int i, scale; |
1da177e4 | 2994 | |
2dd73a4f | 2995 | this_load = this_rq->raw_weighted_load; |
48f24c4d IM |
2996 | |
2997 | /* Update our load: */ | |
ff91691b | 2998 | for (i = 0, scale = 1; i < 3; i++, scale += scale) { |
48f24c4d IM |
2999 | unsigned long old_load, new_load; |
3000 | ||
ff91691b NP |
3001 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
3002 | ||
7897986b | 3003 | old_load = this_rq->cpu_load[i]; |
48f24c4d | 3004 | new_load = this_load; |
7897986b NP |
3005 | /* |
3006 | * Round up the averaging division if load is increasing. This | |
3007 | * prevents us from getting stuck on 9 if the load is 10, for | |
3008 | * example. | |
3009 | */ | |
3010 | if (new_load > old_load) | |
3011 | new_load += scale-1; | |
ff91691b | 3012 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
7897986b | 3013 | } |
7835b98b CL |
3014 | } |
3015 | ||
46cb4b7c SS |
3016 | #ifdef CONFIG_NO_HZ |
3017 | static struct { | |
3018 | atomic_t load_balancer; | |
3019 | cpumask_t cpu_mask; | |
3020 | } nohz ____cacheline_aligned = { | |
3021 | .load_balancer = ATOMIC_INIT(-1), | |
3022 | .cpu_mask = CPU_MASK_NONE, | |
3023 | }; | |
3024 | ||
7835b98b | 3025 | /* |
46cb4b7c SS |
3026 | * This routine will try to nominate the ilb (idle load balancing) |
3027 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3028 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3029 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3030 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3031 | * arrives... | |
3032 | * | |
3033 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3034 | * for idle load balancing. ilb owner will still be part of | |
3035 | * nohz.cpu_mask.. | |
7835b98b | 3036 | * |
46cb4b7c SS |
3037 | * While stopping the tick, this cpu will become the ilb owner if there |
3038 | * is no other owner. And will be the owner till that cpu becomes busy | |
3039 | * or if all cpus in the system stop their ticks at which point | |
3040 | * there is no need for ilb owner. | |
3041 | * | |
3042 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3043 | * next busy scheduler_tick() | |
3044 | */ | |
3045 | int select_nohz_load_balancer(int stop_tick) | |
3046 | { | |
3047 | int cpu = smp_processor_id(); | |
3048 | ||
3049 | if (stop_tick) { | |
3050 | cpu_set(cpu, nohz.cpu_mask); | |
3051 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3052 | ||
3053 | /* | |
3054 | * If we are going offline and still the leader, give up! | |
3055 | */ | |
3056 | if (cpu_is_offline(cpu) && | |
3057 | atomic_read(&nohz.load_balancer) == cpu) { | |
3058 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3059 | BUG(); | |
3060 | return 0; | |
3061 | } | |
3062 | ||
3063 | /* time for ilb owner also to sleep */ | |
3064 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3065 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3066 | atomic_set(&nohz.load_balancer, -1); | |
3067 | return 0; | |
3068 | } | |
3069 | ||
3070 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3071 | /* make me the ilb owner */ | |
3072 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3073 | return 1; | |
3074 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3075 | return 1; | |
3076 | } else { | |
3077 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3078 | return 0; | |
3079 | ||
3080 | cpu_clear(cpu, nohz.cpu_mask); | |
3081 | ||
3082 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3083 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3084 | BUG(); | |
3085 | } | |
3086 | return 0; | |
3087 | } | |
3088 | #endif | |
3089 | ||
3090 | static DEFINE_SPINLOCK(balancing); | |
3091 | ||
3092 | /* | |
7835b98b CL |
3093 | * It checks each scheduling domain to see if it is due to be balanced, |
3094 | * and initiates a balancing operation if so. | |
3095 | * | |
3096 | * Balancing parameters are set up in arch_init_sched_domains. | |
3097 | */ | |
d15bcfdb | 3098 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3099 | { |
46cb4b7c SS |
3100 | int balance = 1; |
3101 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3102 | unsigned long interval; |
3103 | struct sched_domain *sd; | |
46cb4b7c | 3104 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3105 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 3106 | |
46cb4b7c | 3107 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3108 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3109 | continue; | |
3110 | ||
3111 | interval = sd->balance_interval; | |
d15bcfdb | 3112 | if (idle != CPU_IDLE) |
1da177e4 LT |
3113 | interval *= sd->busy_factor; |
3114 | ||
3115 | /* scale ms to jiffies */ | |
3116 | interval = msecs_to_jiffies(interval); | |
3117 | if (unlikely(!interval)) | |
3118 | interval = 1; | |
3119 | ||
08c183f3 CL |
3120 | if (sd->flags & SD_SERIALIZE) { |
3121 | if (!spin_trylock(&balancing)) | |
3122 | goto out; | |
3123 | } | |
3124 | ||
c9819f45 | 3125 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3126 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3127 | /* |
3128 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3129 | * longer idle, or one of our SMT siblings is |
3130 | * not idle. | |
3131 | */ | |
d15bcfdb | 3132 | idle = CPU_NOT_IDLE; |
1da177e4 | 3133 | } |
1bd77f2d | 3134 | sd->last_balance = jiffies; |
1da177e4 | 3135 | } |
08c183f3 CL |
3136 | if (sd->flags & SD_SERIALIZE) |
3137 | spin_unlock(&balancing); | |
3138 | out: | |
c9819f45 CL |
3139 | if (time_after(next_balance, sd->last_balance + interval)) |
3140 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
3141 | |
3142 | /* | |
3143 | * Stop the load balance at this level. There is another | |
3144 | * CPU in our sched group which is doing load balancing more | |
3145 | * actively. | |
3146 | */ | |
3147 | if (!balance) | |
3148 | break; | |
1da177e4 | 3149 | } |
46cb4b7c SS |
3150 | rq->next_balance = next_balance; |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3155 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3156 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3157 | */ | |
3158 | static void run_rebalance_domains(struct softirq_action *h) | |
3159 | { | |
3160 | int local_cpu = smp_processor_id(); | |
3161 | struct rq *local_rq = cpu_rq(local_cpu); | |
d15bcfdb | 3162 | enum cpu_idle_type idle = local_rq->idle_at_tick ? CPU_IDLE : CPU_NOT_IDLE; |
46cb4b7c SS |
3163 | |
3164 | rebalance_domains(local_cpu, idle); | |
3165 | ||
3166 | #ifdef CONFIG_NO_HZ | |
3167 | /* | |
3168 | * If this cpu is the owner for idle load balancing, then do the | |
3169 | * balancing on behalf of the other idle cpus whose ticks are | |
3170 | * stopped. | |
3171 | */ | |
3172 | if (local_rq->idle_at_tick && | |
3173 | atomic_read(&nohz.load_balancer) == local_cpu) { | |
3174 | cpumask_t cpus = nohz.cpu_mask; | |
3175 | struct rq *rq; | |
3176 | int balance_cpu; | |
3177 | ||
3178 | cpu_clear(local_cpu, cpus); | |
3179 | for_each_cpu_mask(balance_cpu, cpus) { | |
3180 | /* | |
3181 | * If this cpu gets work to do, stop the load balancing | |
3182 | * work being done for other cpus. Next load | |
3183 | * balancing owner will pick it up. | |
3184 | */ | |
3185 | if (need_resched()) | |
3186 | break; | |
3187 | ||
d15bcfdb | 3188 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3189 | |
3190 | rq = cpu_rq(balance_cpu); | |
3191 | if (time_after(local_rq->next_balance, rq->next_balance)) | |
3192 | local_rq->next_balance = rq->next_balance; | |
3193 | } | |
3194 | } | |
3195 | #endif | |
3196 | } | |
3197 | ||
3198 | /* | |
3199 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3200 | * | |
3201 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3202 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3203 | * if the whole system is idle. | |
3204 | */ | |
3205 | static inline void trigger_load_balance(int cpu) | |
3206 | { | |
3207 | struct rq *rq = cpu_rq(cpu); | |
3208 | #ifdef CONFIG_NO_HZ | |
3209 | /* | |
3210 | * If we were in the nohz mode recently and busy at the current | |
3211 | * scheduler tick, then check if we need to nominate new idle | |
3212 | * load balancer. | |
3213 | */ | |
3214 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3215 | rq->in_nohz_recently = 0; | |
3216 | ||
3217 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3218 | cpu_clear(cpu, nohz.cpu_mask); | |
3219 | atomic_set(&nohz.load_balancer, -1); | |
3220 | } | |
3221 | ||
3222 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3223 | /* | |
3224 | * simple selection for now: Nominate the | |
3225 | * first cpu in the nohz list to be the next | |
3226 | * ilb owner. | |
3227 | * | |
3228 | * TBD: Traverse the sched domains and nominate | |
3229 | * the nearest cpu in the nohz.cpu_mask. | |
3230 | */ | |
3231 | int ilb = first_cpu(nohz.cpu_mask); | |
3232 | ||
3233 | if (ilb != NR_CPUS) | |
3234 | resched_cpu(ilb); | |
3235 | } | |
3236 | } | |
3237 | ||
3238 | /* | |
3239 | * If this cpu is idle and doing idle load balancing for all the | |
3240 | * cpus with ticks stopped, is it time for that to stop? | |
3241 | */ | |
3242 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3243 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3244 | resched_cpu(cpu); | |
3245 | return; | |
3246 | } | |
3247 | ||
3248 | /* | |
3249 | * If this cpu is idle and the idle load balancing is done by | |
3250 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3251 | */ | |
3252 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3253 | cpu_isset(cpu, nohz.cpu_mask)) | |
3254 | return; | |
3255 | #endif | |
3256 | if (time_after_eq(jiffies, rq->next_balance)) | |
3257 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 LT |
3258 | } |
3259 | #else | |
3260 | /* | |
3261 | * on UP we do not need to balance between CPUs: | |
3262 | */ | |
70b97a7f | 3263 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3264 | { |
3265 | } | |
3266 | #endif | |
3267 | ||
1da177e4 LT |
3268 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3269 | ||
3270 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3271 | ||
3272 | /* | |
3273 | * This is called on clock ticks and on context switches. | |
3274 | * Bank in p->sched_time the ns elapsed since the last tick or switch. | |
3275 | */ | |
48f24c4d | 3276 | static inline void |
70b97a7f | 3277 | update_cpu_clock(struct task_struct *p, struct rq *rq, unsigned long long now) |
1da177e4 | 3278 | { |
b18ec803 MG |
3279 | p->sched_time += now - p->last_ran; |
3280 | p->last_ran = rq->most_recent_timestamp = now; | |
1da177e4 LT |
3281 | } |
3282 | ||
3283 | /* | |
3284 | * Return current->sched_time plus any more ns on the sched_clock | |
3285 | * that have not yet been banked. | |
3286 | */ | |
36c8b586 | 3287 | unsigned long long current_sched_time(const struct task_struct *p) |
1da177e4 LT |
3288 | { |
3289 | unsigned long long ns; | |
3290 | unsigned long flags; | |
48f24c4d | 3291 | |
1da177e4 | 3292 | local_irq_save(flags); |
b18ec803 | 3293 | ns = p->sched_time + sched_clock() - p->last_ran; |
1da177e4 | 3294 | local_irq_restore(flags); |
48f24c4d | 3295 | |
1da177e4 LT |
3296 | return ns; |
3297 | } | |
3298 | ||
f1adad78 LT |
3299 | /* |
3300 | * We place interactive tasks back into the active array, if possible. | |
3301 | * | |
3302 | * To guarantee that this does not starve expired tasks we ignore the | |
3303 | * interactivity of a task if the first expired task had to wait more | |
3304 | * than a 'reasonable' amount of time. This deadline timeout is | |
3305 | * load-dependent, as the frequency of array switched decreases with | |
3306 | * increasing number of running tasks. We also ignore the interactivity | |
3307 | * if a better static_prio task has expired: | |
3308 | */ | |
70b97a7f | 3309 | static inline int expired_starving(struct rq *rq) |
48f24c4d IM |
3310 | { |
3311 | if (rq->curr->static_prio > rq->best_expired_prio) | |
3312 | return 1; | |
3313 | if (!STARVATION_LIMIT || !rq->expired_timestamp) | |
3314 | return 0; | |
3315 | if (jiffies - rq->expired_timestamp > STARVATION_LIMIT * rq->nr_running) | |
3316 | return 1; | |
3317 | return 0; | |
3318 | } | |
f1adad78 | 3319 | |
1da177e4 LT |
3320 | /* |
3321 | * Account user cpu time to a process. | |
3322 | * @p: the process that the cpu time gets accounted to | |
3323 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3324 | * @cputime: the cpu time spent in user space since the last update | |
3325 | */ | |
3326 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3327 | { | |
3328 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3329 | cputime64_t tmp; | |
3330 | ||
3331 | p->utime = cputime_add(p->utime, cputime); | |
3332 | ||
3333 | /* Add user time to cpustat. */ | |
3334 | tmp = cputime_to_cputime64(cputime); | |
3335 | if (TASK_NICE(p) > 0) | |
3336 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3337 | else | |
3338 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * Account system cpu time to a process. | |
3343 | * @p: the process that the cpu time gets accounted to | |
3344 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3345 | * @cputime: the cpu time spent in kernel space since the last update | |
3346 | */ | |
3347 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3348 | cputime_t cputime) | |
3349 | { | |
3350 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3351 | struct rq *rq = this_rq(); |
1da177e4 LT |
3352 | cputime64_t tmp; |
3353 | ||
3354 | p->stime = cputime_add(p->stime, cputime); | |
3355 | ||
3356 | /* Add system time to cpustat. */ | |
3357 | tmp = cputime_to_cputime64(cputime); | |
3358 | if (hardirq_count() - hardirq_offset) | |
3359 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3360 | else if (softirq_count()) | |
3361 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3362 | else if (p != rq->idle) | |
3363 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3364 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3365 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3366 | else | |
3367 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3368 | /* Account for system time used */ | |
3369 | acct_update_integrals(p); | |
1da177e4 LT |
3370 | } |
3371 | ||
3372 | /* | |
3373 | * Account for involuntary wait time. | |
3374 | * @p: the process from which the cpu time has been stolen | |
3375 | * @steal: the cpu time spent in involuntary wait | |
3376 | */ | |
3377 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3378 | { | |
3379 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3380 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3381 | struct rq *rq = this_rq(); |
1da177e4 LT |
3382 | |
3383 | if (p == rq->idle) { | |
3384 | p->stime = cputime_add(p->stime, steal); | |
3385 | if (atomic_read(&rq->nr_iowait) > 0) | |
3386 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3387 | else | |
3388 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3389 | } else | |
3390 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3391 | } | |
3392 | ||
7835b98b | 3393 | static void task_running_tick(struct rq *rq, struct task_struct *p) |
1da177e4 | 3394 | { |
1da177e4 | 3395 | if (p->array != rq->active) { |
7835b98b | 3396 | /* Task has expired but was not scheduled yet */ |
1da177e4 | 3397 | set_tsk_need_resched(p); |
7835b98b | 3398 | return; |
1da177e4 LT |
3399 | } |
3400 | spin_lock(&rq->lock); | |
3401 | /* | |
3402 | * The task was running during this tick - update the | |
3403 | * time slice counter. Note: we do not update a thread's | |
3404 | * priority until it either goes to sleep or uses up its | |
3405 | * timeslice. This makes it possible for interactive tasks | |
3406 | * to use up their timeslices at their highest priority levels. | |
3407 | */ | |
3408 | if (rt_task(p)) { | |
3409 | /* | |
3410 | * RR tasks need a special form of timeslice management. | |
3411 | * FIFO tasks have no timeslices. | |
3412 | */ | |
3413 | if ((p->policy == SCHED_RR) && !--p->time_slice) { | |
3414 | p->time_slice = task_timeslice(p); | |
3415 | p->first_time_slice = 0; | |
3416 | set_tsk_need_resched(p); | |
3417 | ||
3418 | /* put it at the end of the queue: */ | |
3419 | requeue_task(p, rq->active); | |
3420 | } | |
3421 | goto out_unlock; | |
3422 | } | |
3423 | if (!--p->time_slice) { | |
3424 | dequeue_task(p, rq->active); | |
3425 | set_tsk_need_resched(p); | |
3426 | p->prio = effective_prio(p); | |
3427 | p->time_slice = task_timeslice(p); | |
3428 | p->first_time_slice = 0; | |
3429 | ||
3430 | if (!rq->expired_timestamp) | |
3431 | rq->expired_timestamp = jiffies; | |
48f24c4d | 3432 | if (!TASK_INTERACTIVE(p) || expired_starving(rq)) { |
1da177e4 LT |
3433 | enqueue_task(p, rq->expired); |
3434 | if (p->static_prio < rq->best_expired_prio) | |
3435 | rq->best_expired_prio = p->static_prio; | |
3436 | } else | |
3437 | enqueue_task(p, rq->active); | |
3438 | } else { | |
3439 | /* | |
3440 | * Prevent a too long timeslice allowing a task to monopolize | |
3441 | * the CPU. We do this by splitting up the timeslice into | |
3442 | * smaller pieces. | |
3443 | * | |
3444 | * Note: this does not mean the task's timeslices expire or | |
3445 | * get lost in any way, they just might be preempted by | |
3446 | * another task of equal priority. (one with higher | |
3447 | * priority would have preempted this task already.) We | |
3448 | * requeue this task to the end of the list on this priority | |
3449 | * level, which is in essence a round-robin of tasks with | |
3450 | * equal priority. | |
3451 | * | |
3452 | * This only applies to tasks in the interactive | |
3453 | * delta range with at least TIMESLICE_GRANULARITY to requeue. | |
3454 | */ | |
3455 | if (TASK_INTERACTIVE(p) && !((task_timeslice(p) - | |
3456 | p->time_slice) % TIMESLICE_GRANULARITY(p)) && | |
3457 | (p->time_slice >= TIMESLICE_GRANULARITY(p)) && | |
3458 | (p->array == rq->active)) { | |
3459 | ||
3460 | requeue_task(p, rq->active); | |
3461 | set_tsk_need_resched(p); | |
3462 | } | |
3463 | } | |
3464 | out_unlock: | |
3465 | spin_unlock(&rq->lock); | |
7835b98b CL |
3466 | } |
3467 | ||
3468 | /* | |
3469 | * This function gets called by the timer code, with HZ frequency. | |
3470 | * We call it with interrupts disabled. | |
3471 | * | |
3472 | * It also gets called by the fork code, when changing the parent's | |
3473 | * timeslices. | |
3474 | */ | |
3475 | void scheduler_tick(void) | |
3476 | { | |
3477 | unsigned long long now = sched_clock(); | |
3478 | struct task_struct *p = current; | |
3479 | int cpu = smp_processor_id(); | |
bdecea3a | 3480 | int idle_at_tick = idle_cpu(cpu); |
7835b98b | 3481 | struct rq *rq = cpu_rq(cpu); |
7835b98b CL |
3482 | |
3483 | update_cpu_clock(p, rq, now); | |
3484 | ||
bdecea3a | 3485 | if (!idle_at_tick) |
7835b98b | 3486 | task_running_tick(rq, p); |
e418e1c2 | 3487 | #ifdef CONFIG_SMP |
7835b98b | 3488 | update_load(rq); |
bdecea3a | 3489 | rq->idle_at_tick = idle_at_tick; |
46cb4b7c | 3490 | trigger_load_balance(cpu); |
e418e1c2 | 3491 | #endif |
1da177e4 LT |
3492 | } |
3493 | ||
1da177e4 LT |
3494 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3495 | ||
3496 | void fastcall add_preempt_count(int val) | |
3497 | { | |
3498 | /* | |
3499 | * Underflow? | |
3500 | */ | |
9a11b49a IM |
3501 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3502 | return; | |
1da177e4 LT |
3503 | preempt_count() += val; |
3504 | /* | |
3505 | * Spinlock count overflowing soon? | |
3506 | */ | |
33859f7f MOS |
3507 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3508 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3509 | } |
3510 | EXPORT_SYMBOL(add_preempt_count); | |
3511 | ||
3512 | void fastcall sub_preempt_count(int val) | |
3513 | { | |
3514 | /* | |
3515 | * Underflow? | |
3516 | */ | |
9a11b49a IM |
3517 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3518 | return; | |
1da177e4 LT |
3519 | /* |
3520 | * Is the spinlock portion underflowing? | |
3521 | */ | |
9a11b49a IM |
3522 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3523 | !(preempt_count() & PREEMPT_MASK))) | |
3524 | return; | |
3525 | ||
1da177e4 LT |
3526 | preempt_count() -= val; |
3527 | } | |
3528 | EXPORT_SYMBOL(sub_preempt_count); | |
3529 | ||
3530 | #endif | |
3531 | ||
3dee386e CK |
3532 | static inline int interactive_sleep(enum sleep_type sleep_type) |
3533 | { | |
3534 | return (sleep_type == SLEEP_INTERACTIVE || | |
3535 | sleep_type == SLEEP_INTERRUPTED); | |
3536 | } | |
3537 | ||
1da177e4 LT |
3538 | /* |
3539 | * schedule() is the main scheduler function. | |
3540 | */ | |
3541 | asmlinkage void __sched schedule(void) | |
3542 | { | |
36c8b586 | 3543 | struct task_struct *prev, *next; |
70b97a7f | 3544 | struct prio_array *array; |
1da177e4 LT |
3545 | struct list_head *queue; |
3546 | unsigned long long now; | |
3547 | unsigned long run_time; | |
a3464a10 | 3548 | int cpu, idx, new_prio; |
48f24c4d | 3549 | long *switch_count; |
70b97a7f | 3550 | struct rq *rq; |
1da177e4 LT |
3551 | |
3552 | /* | |
3553 | * Test if we are atomic. Since do_exit() needs to call into | |
3554 | * schedule() atomically, we ignore that path for now. | |
3555 | * Otherwise, whine if we are scheduling when we should not be. | |
3556 | */ | |
77e4bfbc AM |
3557 | if (unlikely(in_atomic() && !current->exit_state)) { |
3558 | printk(KERN_ERR "BUG: scheduling while atomic: " | |
3559 | "%s/0x%08x/%d\n", | |
3560 | current->comm, preempt_count(), current->pid); | |
a4c410f0 | 3561 | debug_show_held_locks(current); |
3117df04 IM |
3562 | if (irqs_disabled()) |
3563 | print_irqtrace_events(current); | |
77e4bfbc | 3564 | dump_stack(); |
1da177e4 LT |
3565 | } |
3566 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); | |
3567 | ||
3568 | need_resched: | |
3569 | preempt_disable(); | |
3570 | prev = current; | |
3571 | release_kernel_lock(prev); | |
3572 | need_resched_nonpreemptible: | |
3573 | rq = this_rq(); | |
3574 | ||
3575 | /* | |
3576 | * The idle thread is not allowed to schedule! | |
3577 | * Remove this check after it has been exercised a bit. | |
3578 | */ | |
3579 | if (unlikely(prev == rq->idle) && prev->state != TASK_RUNNING) { | |
3580 | printk(KERN_ERR "bad: scheduling from the idle thread!\n"); | |
3581 | dump_stack(); | |
3582 | } | |
3583 | ||
3584 | schedstat_inc(rq, sched_cnt); | |
3585 | now = sched_clock(); | |
238628ed | 3586 | if (likely((long long)(now - prev->timestamp) < NS_MAX_SLEEP_AVG)) { |
1da177e4 | 3587 | run_time = now - prev->timestamp; |
238628ed | 3588 | if (unlikely((long long)(now - prev->timestamp) < 0)) |
1da177e4 LT |
3589 | run_time = 0; |
3590 | } else | |
3591 | run_time = NS_MAX_SLEEP_AVG; | |
3592 | ||
3593 | /* | |
3594 | * Tasks charged proportionately less run_time at high sleep_avg to | |
3595 | * delay them losing their interactive status | |
3596 | */ | |
3597 | run_time /= (CURRENT_BONUS(prev) ? : 1); | |
3598 | ||
3599 | spin_lock_irq(&rq->lock); | |
3600 | ||
1da177e4 LT |
3601 | switch_count = &prev->nivcsw; |
3602 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { | |
3603 | switch_count = &prev->nvcsw; | |
3604 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && | |
3605 | unlikely(signal_pending(prev)))) | |
3606 | prev->state = TASK_RUNNING; | |
3607 | else { | |
3608 | if (prev->state == TASK_UNINTERRUPTIBLE) | |
3609 | rq->nr_uninterruptible++; | |
3610 | deactivate_task(prev, rq); | |
3611 | } | |
3612 | } | |
3613 | ||
3614 | cpu = smp_processor_id(); | |
3615 | if (unlikely(!rq->nr_running)) { | |
1da177e4 LT |
3616 | idle_balance(cpu, rq); |
3617 | if (!rq->nr_running) { | |
3618 | next = rq->idle; | |
3619 | rq->expired_timestamp = 0; | |
1da177e4 LT |
3620 | goto switch_tasks; |
3621 | } | |
1da177e4 LT |
3622 | } |
3623 | ||
3624 | array = rq->active; | |
3625 | if (unlikely(!array->nr_active)) { | |
3626 | /* | |
3627 | * Switch the active and expired arrays. | |
3628 | */ | |
3629 | schedstat_inc(rq, sched_switch); | |
3630 | rq->active = rq->expired; | |
3631 | rq->expired = array; | |
3632 | array = rq->active; | |
3633 | rq->expired_timestamp = 0; | |
3634 | rq->best_expired_prio = MAX_PRIO; | |
3635 | } | |
3636 | ||
3637 | idx = sched_find_first_bit(array->bitmap); | |
3638 | queue = array->queue + idx; | |
36c8b586 | 3639 | next = list_entry(queue->next, struct task_struct, run_list); |
1da177e4 | 3640 | |
3dee386e | 3641 | if (!rt_task(next) && interactive_sleep(next->sleep_type)) { |
1da177e4 | 3642 | unsigned long long delta = now - next->timestamp; |
238628ed | 3643 | if (unlikely((long long)(now - next->timestamp) < 0)) |
1da177e4 LT |
3644 | delta = 0; |
3645 | ||
3dee386e | 3646 | if (next->sleep_type == SLEEP_INTERACTIVE) |
1da177e4 LT |
3647 | delta = delta * (ON_RUNQUEUE_WEIGHT * 128 / 100) / 128; |
3648 | ||
3649 | array = next->array; | |
a3464a10 CS |
3650 | new_prio = recalc_task_prio(next, next->timestamp + delta); |
3651 | ||
3652 | if (unlikely(next->prio != new_prio)) { | |
3653 | dequeue_task(next, array); | |
3654 | next->prio = new_prio; | |
3655 | enqueue_task(next, array); | |
7c4bb1f9 | 3656 | } |
1da177e4 | 3657 | } |
3dee386e | 3658 | next->sleep_type = SLEEP_NORMAL; |
1da177e4 LT |
3659 | switch_tasks: |
3660 | if (next == rq->idle) | |
3661 | schedstat_inc(rq, sched_goidle); | |
3662 | prefetch(next); | |
383f2835 | 3663 | prefetch_stack(next); |
1da177e4 LT |
3664 | clear_tsk_need_resched(prev); |
3665 | rcu_qsctr_inc(task_cpu(prev)); | |
3666 | ||
3667 | update_cpu_clock(prev, rq, now); | |
3668 | ||
3669 | prev->sleep_avg -= run_time; | |
3670 | if ((long)prev->sleep_avg <= 0) | |
3671 | prev->sleep_avg = 0; | |
3672 | prev->timestamp = prev->last_ran = now; | |
3673 | ||
3674 | sched_info_switch(prev, next); | |
3675 | if (likely(prev != next)) { | |
c1e16aa2 | 3676 | next->timestamp = next->last_ran = now; |
1da177e4 LT |
3677 | rq->nr_switches++; |
3678 | rq->curr = next; | |
3679 | ++*switch_count; | |
3680 | ||
4866cde0 | 3681 | prepare_task_switch(rq, next); |
1da177e4 LT |
3682 | prev = context_switch(rq, prev, next); |
3683 | barrier(); | |
4866cde0 NP |
3684 | /* |
3685 | * this_rq must be evaluated again because prev may have moved | |
3686 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3687 | * frame will be invalid. | |
3688 | */ | |
3689 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3690 | } else |
3691 | spin_unlock_irq(&rq->lock); | |
3692 | ||
3693 | prev = current; | |
3694 | if (unlikely(reacquire_kernel_lock(prev) < 0)) | |
3695 | goto need_resched_nonpreemptible; | |
3696 | preempt_enable_no_resched(); | |
3697 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3698 | goto need_resched; | |
3699 | } | |
1da177e4 LT |
3700 | EXPORT_SYMBOL(schedule); |
3701 | ||
3702 | #ifdef CONFIG_PREEMPT | |
3703 | /* | |
2ed6e34f | 3704 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3705 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3706 | * occur there and call schedule directly. | |
3707 | */ | |
3708 | asmlinkage void __sched preempt_schedule(void) | |
3709 | { | |
3710 | struct thread_info *ti = current_thread_info(); | |
3711 | #ifdef CONFIG_PREEMPT_BKL | |
3712 | struct task_struct *task = current; | |
3713 | int saved_lock_depth; | |
3714 | #endif | |
3715 | /* | |
3716 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3717 | * we do not want to preempt the current task. Just return.. | |
3718 | */ | |
beed33a8 | 3719 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3720 | return; |
3721 | ||
3722 | need_resched: | |
3723 | add_preempt_count(PREEMPT_ACTIVE); | |
3724 | /* | |
3725 | * We keep the big kernel semaphore locked, but we | |
3726 | * clear ->lock_depth so that schedule() doesnt | |
3727 | * auto-release the semaphore: | |
3728 | */ | |
3729 | #ifdef CONFIG_PREEMPT_BKL | |
3730 | saved_lock_depth = task->lock_depth; | |
3731 | task->lock_depth = -1; | |
3732 | #endif | |
3733 | schedule(); | |
3734 | #ifdef CONFIG_PREEMPT_BKL | |
3735 | task->lock_depth = saved_lock_depth; | |
3736 | #endif | |
3737 | sub_preempt_count(PREEMPT_ACTIVE); | |
3738 | ||
3739 | /* we could miss a preemption opportunity between schedule and now */ | |
3740 | barrier(); | |
3741 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3742 | goto need_resched; | |
3743 | } | |
1da177e4 LT |
3744 | EXPORT_SYMBOL(preempt_schedule); |
3745 | ||
3746 | /* | |
2ed6e34f | 3747 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3748 | * off of irq context. |
3749 | * Note, that this is called and return with irqs disabled. This will | |
3750 | * protect us against recursive calling from irq. | |
3751 | */ | |
3752 | asmlinkage void __sched preempt_schedule_irq(void) | |
3753 | { | |
3754 | struct thread_info *ti = current_thread_info(); | |
3755 | #ifdef CONFIG_PREEMPT_BKL | |
3756 | struct task_struct *task = current; | |
3757 | int saved_lock_depth; | |
3758 | #endif | |
2ed6e34f | 3759 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3760 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3761 | ||
3762 | need_resched: | |
3763 | add_preempt_count(PREEMPT_ACTIVE); | |
3764 | /* | |
3765 | * We keep the big kernel semaphore locked, but we | |
3766 | * clear ->lock_depth so that schedule() doesnt | |
3767 | * auto-release the semaphore: | |
3768 | */ | |
3769 | #ifdef CONFIG_PREEMPT_BKL | |
3770 | saved_lock_depth = task->lock_depth; | |
3771 | task->lock_depth = -1; | |
3772 | #endif | |
3773 | local_irq_enable(); | |
3774 | schedule(); | |
3775 | local_irq_disable(); | |
3776 | #ifdef CONFIG_PREEMPT_BKL | |
3777 | task->lock_depth = saved_lock_depth; | |
3778 | #endif | |
3779 | sub_preempt_count(PREEMPT_ACTIVE); | |
3780 | ||
3781 | /* we could miss a preemption opportunity between schedule and now */ | |
3782 | barrier(); | |
3783 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3784 | goto need_resched; | |
3785 | } | |
3786 | ||
3787 | #endif /* CONFIG_PREEMPT */ | |
3788 | ||
95cdf3b7 IM |
3789 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3790 | void *key) | |
1da177e4 | 3791 | { |
48f24c4d | 3792 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3793 | } |
1da177e4 LT |
3794 | EXPORT_SYMBOL(default_wake_function); |
3795 | ||
3796 | /* | |
3797 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3798 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3799 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3800 | * | |
3801 | * There are circumstances in which we can try to wake a task which has already | |
3802 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3803 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3804 | */ | |
3805 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3806 | int nr_exclusive, int sync, void *key) | |
3807 | { | |
3808 | struct list_head *tmp, *next; | |
3809 | ||
3810 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3811 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3812 | unsigned flags = curr->flags; | |
3813 | ||
1da177e4 | 3814 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3815 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3816 | break; |
3817 | } | |
3818 | } | |
3819 | ||
3820 | /** | |
3821 | * __wake_up - wake up threads blocked on a waitqueue. | |
3822 | * @q: the waitqueue | |
3823 | * @mode: which threads | |
3824 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3825 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3826 | */ |
3827 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3828 | int nr_exclusive, void *key) |
1da177e4 LT |
3829 | { |
3830 | unsigned long flags; | |
3831 | ||
3832 | spin_lock_irqsave(&q->lock, flags); | |
3833 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3834 | spin_unlock_irqrestore(&q->lock, flags); | |
3835 | } | |
1da177e4 LT |
3836 | EXPORT_SYMBOL(__wake_up); |
3837 | ||
3838 | /* | |
3839 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3840 | */ | |
3841 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3842 | { | |
3843 | __wake_up_common(q, mode, 1, 0, NULL); | |
3844 | } | |
3845 | ||
3846 | /** | |
67be2dd1 | 3847 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3848 | * @q: the waitqueue |
3849 | * @mode: which threads | |
3850 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3851 | * | |
3852 | * The sync wakeup differs that the waker knows that it will schedule | |
3853 | * away soon, so while the target thread will be woken up, it will not | |
3854 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3855 | * with each other. This can prevent needless bouncing between CPUs. | |
3856 | * | |
3857 | * On UP it can prevent extra preemption. | |
3858 | */ | |
95cdf3b7 IM |
3859 | void fastcall |
3860 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3861 | { |
3862 | unsigned long flags; | |
3863 | int sync = 1; | |
3864 | ||
3865 | if (unlikely(!q)) | |
3866 | return; | |
3867 | ||
3868 | if (unlikely(!nr_exclusive)) | |
3869 | sync = 0; | |
3870 | ||
3871 | spin_lock_irqsave(&q->lock, flags); | |
3872 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3873 | spin_unlock_irqrestore(&q->lock, flags); | |
3874 | } | |
3875 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3876 | ||
3877 | void fastcall complete(struct completion *x) | |
3878 | { | |
3879 | unsigned long flags; | |
3880 | ||
3881 | spin_lock_irqsave(&x->wait.lock, flags); | |
3882 | x->done++; | |
3883 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3884 | 1, 0, NULL); | |
3885 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3886 | } | |
3887 | EXPORT_SYMBOL(complete); | |
3888 | ||
3889 | void fastcall complete_all(struct completion *x) | |
3890 | { | |
3891 | unsigned long flags; | |
3892 | ||
3893 | spin_lock_irqsave(&x->wait.lock, flags); | |
3894 | x->done += UINT_MAX/2; | |
3895 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3896 | 0, 0, NULL); | |
3897 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3898 | } | |
3899 | EXPORT_SYMBOL(complete_all); | |
3900 | ||
3901 | void fastcall __sched wait_for_completion(struct completion *x) | |
3902 | { | |
3903 | might_sleep(); | |
48f24c4d | 3904 | |
1da177e4 LT |
3905 | spin_lock_irq(&x->wait.lock); |
3906 | if (!x->done) { | |
3907 | DECLARE_WAITQUEUE(wait, current); | |
3908 | ||
3909 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3910 | __add_wait_queue_tail(&x->wait, &wait); | |
3911 | do { | |
3912 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3913 | spin_unlock_irq(&x->wait.lock); | |
3914 | schedule(); | |
3915 | spin_lock_irq(&x->wait.lock); | |
3916 | } while (!x->done); | |
3917 | __remove_wait_queue(&x->wait, &wait); | |
3918 | } | |
3919 | x->done--; | |
3920 | spin_unlock_irq(&x->wait.lock); | |
3921 | } | |
3922 | EXPORT_SYMBOL(wait_for_completion); | |
3923 | ||
3924 | unsigned long fastcall __sched | |
3925 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3926 | { | |
3927 | might_sleep(); | |
3928 | ||
3929 | spin_lock_irq(&x->wait.lock); | |
3930 | if (!x->done) { | |
3931 | DECLARE_WAITQUEUE(wait, current); | |
3932 | ||
3933 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3934 | __add_wait_queue_tail(&x->wait, &wait); | |
3935 | do { | |
3936 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3937 | spin_unlock_irq(&x->wait.lock); | |
3938 | timeout = schedule_timeout(timeout); | |
3939 | spin_lock_irq(&x->wait.lock); | |
3940 | if (!timeout) { | |
3941 | __remove_wait_queue(&x->wait, &wait); | |
3942 | goto out; | |
3943 | } | |
3944 | } while (!x->done); | |
3945 | __remove_wait_queue(&x->wait, &wait); | |
3946 | } | |
3947 | x->done--; | |
3948 | out: | |
3949 | spin_unlock_irq(&x->wait.lock); | |
3950 | return timeout; | |
3951 | } | |
3952 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3953 | ||
3954 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3955 | { | |
3956 | int ret = 0; | |
3957 | ||
3958 | might_sleep(); | |
3959 | ||
3960 | spin_lock_irq(&x->wait.lock); | |
3961 | if (!x->done) { | |
3962 | DECLARE_WAITQUEUE(wait, current); | |
3963 | ||
3964 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3965 | __add_wait_queue_tail(&x->wait, &wait); | |
3966 | do { | |
3967 | if (signal_pending(current)) { | |
3968 | ret = -ERESTARTSYS; | |
3969 | __remove_wait_queue(&x->wait, &wait); | |
3970 | goto out; | |
3971 | } | |
3972 | __set_current_state(TASK_INTERRUPTIBLE); | |
3973 | spin_unlock_irq(&x->wait.lock); | |
3974 | schedule(); | |
3975 | spin_lock_irq(&x->wait.lock); | |
3976 | } while (!x->done); | |
3977 | __remove_wait_queue(&x->wait, &wait); | |
3978 | } | |
3979 | x->done--; | |
3980 | out: | |
3981 | spin_unlock_irq(&x->wait.lock); | |
3982 | ||
3983 | return ret; | |
3984 | } | |
3985 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3986 | ||
3987 | unsigned long fastcall __sched | |
3988 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3989 | unsigned long timeout) | |
3990 | { | |
3991 | might_sleep(); | |
3992 | ||
3993 | spin_lock_irq(&x->wait.lock); | |
3994 | if (!x->done) { | |
3995 | DECLARE_WAITQUEUE(wait, current); | |
3996 | ||
3997 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3998 | __add_wait_queue_tail(&x->wait, &wait); | |
3999 | do { | |
4000 | if (signal_pending(current)) { | |
4001 | timeout = -ERESTARTSYS; | |
4002 | __remove_wait_queue(&x->wait, &wait); | |
4003 | goto out; | |
4004 | } | |
4005 | __set_current_state(TASK_INTERRUPTIBLE); | |
4006 | spin_unlock_irq(&x->wait.lock); | |
4007 | timeout = schedule_timeout(timeout); | |
4008 | spin_lock_irq(&x->wait.lock); | |
4009 | if (!timeout) { | |
4010 | __remove_wait_queue(&x->wait, &wait); | |
4011 | goto out; | |
4012 | } | |
4013 | } while (!x->done); | |
4014 | __remove_wait_queue(&x->wait, &wait); | |
4015 | } | |
4016 | x->done--; | |
4017 | out: | |
4018 | spin_unlock_irq(&x->wait.lock); | |
4019 | return timeout; | |
4020 | } | |
4021 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
4022 | ||
4023 | ||
4024 | #define SLEEP_ON_VAR \ | |
4025 | unsigned long flags; \ | |
4026 | wait_queue_t wait; \ | |
4027 | init_waitqueue_entry(&wait, current); | |
4028 | ||
4029 | #define SLEEP_ON_HEAD \ | |
4030 | spin_lock_irqsave(&q->lock,flags); \ | |
4031 | __add_wait_queue(q, &wait); \ | |
4032 | spin_unlock(&q->lock); | |
4033 | ||
4034 | #define SLEEP_ON_TAIL \ | |
4035 | spin_lock_irq(&q->lock); \ | |
4036 | __remove_wait_queue(q, &wait); \ | |
4037 | spin_unlock_irqrestore(&q->lock, flags); | |
4038 | ||
4039 | void fastcall __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4040 | { | |
4041 | SLEEP_ON_VAR | |
4042 | ||
4043 | current->state = TASK_INTERRUPTIBLE; | |
4044 | ||
4045 | SLEEP_ON_HEAD | |
4046 | schedule(); | |
4047 | SLEEP_ON_TAIL | |
4048 | } | |
1da177e4 LT |
4049 | EXPORT_SYMBOL(interruptible_sleep_on); |
4050 | ||
95cdf3b7 IM |
4051 | long fastcall __sched |
4052 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
1da177e4 LT |
4053 | { |
4054 | SLEEP_ON_VAR | |
4055 | ||
4056 | current->state = TASK_INTERRUPTIBLE; | |
4057 | ||
4058 | SLEEP_ON_HEAD | |
4059 | timeout = schedule_timeout(timeout); | |
4060 | SLEEP_ON_TAIL | |
4061 | ||
4062 | return timeout; | |
4063 | } | |
1da177e4 LT |
4064 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4065 | ||
4066 | void fastcall __sched sleep_on(wait_queue_head_t *q) | |
4067 | { | |
4068 | SLEEP_ON_VAR | |
4069 | ||
4070 | current->state = TASK_UNINTERRUPTIBLE; | |
4071 | ||
4072 | SLEEP_ON_HEAD | |
4073 | schedule(); | |
4074 | SLEEP_ON_TAIL | |
4075 | } | |
1da177e4 LT |
4076 | EXPORT_SYMBOL(sleep_on); |
4077 | ||
4078 | long fastcall __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
4079 | { | |
4080 | SLEEP_ON_VAR | |
4081 | ||
4082 | current->state = TASK_UNINTERRUPTIBLE; | |
4083 | ||
4084 | SLEEP_ON_HEAD | |
4085 | timeout = schedule_timeout(timeout); | |
4086 | SLEEP_ON_TAIL | |
4087 | ||
4088 | return timeout; | |
4089 | } | |
4090 | ||
4091 | EXPORT_SYMBOL(sleep_on_timeout); | |
4092 | ||
b29739f9 IM |
4093 | #ifdef CONFIG_RT_MUTEXES |
4094 | ||
4095 | /* | |
4096 | * rt_mutex_setprio - set the current priority of a task | |
4097 | * @p: task | |
4098 | * @prio: prio value (kernel-internal form) | |
4099 | * | |
4100 | * This function changes the 'effective' priority of a task. It does | |
4101 | * not touch ->normal_prio like __setscheduler(). | |
4102 | * | |
4103 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4104 | */ | |
36c8b586 | 4105 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4106 | { |
70b97a7f | 4107 | struct prio_array *array; |
b29739f9 | 4108 | unsigned long flags; |
70b97a7f | 4109 | struct rq *rq; |
d5f9f942 | 4110 | int oldprio; |
b29739f9 IM |
4111 | |
4112 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4113 | ||
4114 | rq = task_rq_lock(p, &flags); | |
4115 | ||
d5f9f942 | 4116 | oldprio = p->prio; |
b29739f9 IM |
4117 | array = p->array; |
4118 | if (array) | |
4119 | dequeue_task(p, array); | |
4120 | p->prio = prio; | |
4121 | ||
4122 | if (array) { | |
4123 | /* | |
4124 | * If changing to an RT priority then queue it | |
4125 | * in the active array! | |
4126 | */ | |
4127 | if (rt_task(p)) | |
4128 | array = rq->active; | |
4129 | enqueue_task(p, array); | |
4130 | /* | |
4131 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4132 | * our priority decreased, or if we are not currently running on |
4133 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4134 | */ |
d5f9f942 AM |
4135 | if (task_running(rq, p)) { |
4136 | if (p->prio > oldprio) | |
4137 | resched_task(rq->curr); | |
4138 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
b29739f9 IM |
4139 | resched_task(rq->curr); |
4140 | } | |
4141 | task_rq_unlock(rq, &flags); | |
4142 | } | |
4143 | ||
4144 | #endif | |
4145 | ||
36c8b586 | 4146 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4147 | { |
70b97a7f | 4148 | struct prio_array *array; |
48f24c4d | 4149 | int old_prio, delta; |
1da177e4 | 4150 | unsigned long flags; |
70b97a7f | 4151 | struct rq *rq; |
1da177e4 LT |
4152 | |
4153 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4154 | return; | |
4155 | /* | |
4156 | * We have to be careful, if called from sys_setpriority(), | |
4157 | * the task might be in the middle of scheduling on another CPU. | |
4158 | */ | |
4159 | rq = task_rq_lock(p, &flags); | |
4160 | /* | |
4161 | * The RT priorities are set via sched_setscheduler(), but we still | |
4162 | * allow the 'normal' nice value to be set - but as expected | |
4163 | * it wont have any effect on scheduling until the task is | |
b0a9499c | 4164 | * not SCHED_NORMAL/SCHED_BATCH: |
1da177e4 | 4165 | */ |
b29739f9 | 4166 | if (has_rt_policy(p)) { |
1da177e4 LT |
4167 | p->static_prio = NICE_TO_PRIO(nice); |
4168 | goto out_unlock; | |
4169 | } | |
4170 | array = p->array; | |
2dd73a4f | 4171 | if (array) { |
1da177e4 | 4172 | dequeue_task(p, array); |
2dd73a4f PW |
4173 | dec_raw_weighted_load(rq, p); |
4174 | } | |
1da177e4 | 4175 | |
1da177e4 | 4176 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4177 | set_load_weight(p); |
b29739f9 IM |
4178 | old_prio = p->prio; |
4179 | p->prio = effective_prio(p); | |
4180 | delta = p->prio - old_prio; | |
1da177e4 LT |
4181 | |
4182 | if (array) { | |
4183 | enqueue_task(p, array); | |
2dd73a4f | 4184 | inc_raw_weighted_load(rq, p); |
1da177e4 | 4185 | /* |
d5f9f942 AM |
4186 | * If the task increased its priority or is running and |
4187 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4188 | */ |
d5f9f942 | 4189 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4190 | resched_task(rq->curr); |
4191 | } | |
4192 | out_unlock: | |
4193 | task_rq_unlock(rq, &flags); | |
4194 | } | |
1da177e4 LT |
4195 | EXPORT_SYMBOL(set_user_nice); |
4196 | ||
e43379f1 MM |
4197 | /* |
4198 | * can_nice - check if a task can reduce its nice value | |
4199 | * @p: task | |
4200 | * @nice: nice value | |
4201 | */ | |
36c8b586 | 4202 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4203 | { |
024f4747 MM |
4204 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4205 | int nice_rlim = 20 - nice; | |
48f24c4d | 4206 | |
e43379f1 MM |
4207 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4208 | capable(CAP_SYS_NICE)); | |
4209 | } | |
4210 | ||
1da177e4 LT |
4211 | #ifdef __ARCH_WANT_SYS_NICE |
4212 | ||
4213 | /* | |
4214 | * sys_nice - change the priority of the current process. | |
4215 | * @increment: priority increment | |
4216 | * | |
4217 | * sys_setpriority is a more generic, but much slower function that | |
4218 | * does similar things. | |
4219 | */ | |
4220 | asmlinkage long sys_nice(int increment) | |
4221 | { | |
48f24c4d | 4222 | long nice, retval; |
1da177e4 LT |
4223 | |
4224 | /* | |
4225 | * Setpriority might change our priority at the same moment. | |
4226 | * We don't have to worry. Conceptually one call occurs first | |
4227 | * and we have a single winner. | |
4228 | */ | |
e43379f1 MM |
4229 | if (increment < -40) |
4230 | increment = -40; | |
1da177e4 LT |
4231 | if (increment > 40) |
4232 | increment = 40; | |
4233 | ||
4234 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4235 | if (nice < -20) | |
4236 | nice = -20; | |
4237 | if (nice > 19) | |
4238 | nice = 19; | |
4239 | ||
e43379f1 MM |
4240 | if (increment < 0 && !can_nice(current, nice)) |
4241 | return -EPERM; | |
4242 | ||
1da177e4 LT |
4243 | retval = security_task_setnice(current, nice); |
4244 | if (retval) | |
4245 | return retval; | |
4246 | ||
4247 | set_user_nice(current, nice); | |
4248 | return 0; | |
4249 | } | |
4250 | ||
4251 | #endif | |
4252 | ||
4253 | /** | |
4254 | * task_prio - return the priority value of a given task. | |
4255 | * @p: the task in question. | |
4256 | * | |
4257 | * This is the priority value as seen by users in /proc. | |
4258 | * RT tasks are offset by -200. Normal tasks are centered | |
4259 | * around 0, value goes from -16 to +15. | |
4260 | */ | |
36c8b586 | 4261 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4262 | { |
4263 | return p->prio - MAX_RT_PRIO; | |
4264 | } | |
4265 | ||
4266 | /** | |
4267 | * task_nice - return the nice value of a given task. | |
4268 | * @p: the task in question. | |
4269 | */ | |
36c8b586 | 4270 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4271 | { |
4272 | return TASK_NICE(p); | |
4273 | } | |
1da177e4 | 4274 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4275 | |
4276 | /** | |
4277 | * idle_cpu - is a given cpu idle currently? | |
4278 | * @cpu: the processor in question. | |
4279 | */ | |
4280 | int idle_cpu(int cpu) | |
4281 | { | |
4282 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4283 | } | |
4284 | ||
1da177e4 LT |
4285 | /** |
4286 | * idle_task - return the idle task for a given cpu. | |
4287 | * @cpu: the processor in question. | |
4288 | */ | |
36c8b586 | 4289 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4290 | { |
4291 | return cpu_rq(cpu)->idle; | |
4292 | } | |
4293 | ||
4294 | /** | |
4295 | * find_process_by_pid - find a process with a matching PID value. | |
4296 | * @pid: the pid in question. | |
4297 | */ | |
36c8b586 | 4298 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4299 | { |
4300 | return pid ? find_task_by_pid(pid) : current; | |
4301 | } | |
4302 | ||
4303 | /* Actually do priority change: must hold rq lock. */ | |
4304 | static void __setscheduler(struct task_struct *p, int policy, int prio) | |
4305 | { | |
4306 | BUG_ON(p->array); | |
48f24c4d | 4307 | |
1da177e4 LT |
4308 | p->policy = policy; |
4309 | p->rt_priority = prio; | |
b29739f9 IM |
4310 | p->normal_prio = normal_prio(p); |
4311 | /* we are holding p->pi_lock already */ | |
4312 | p->prio = rt_mutex_getprio(p); | |
4313 | /* | |
4314 | * SCHED_BATCH tasks are treated as perpetual CPU hogs: | |
4315 | */ | |
4316 | if (policy == SCHED_BATCH) | |
4317 | p->sleep_avg = 0; | |
2dd73a4f | 4318 | set_load_weight(p); |
1da177e4 LT |
4319 | } |
4320 | ||
4321 | /** | |
72fd4a35 | 4322 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4323 | * @p: the task in question. |
4324 | * @policy: new policy. | |
4325 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4326 | * |
72fd4a35 | 4327 | * NOTE that the task may be already dead. |
1da177e4 | 4328 | */ |
95cdf3b7 IM |
4329 | int sched_setscheduler(struct task_struct *p, int policy, |
4330 | struct sched_param *param) | |
1da177e4 | 4331 | { |
48f24c4d | 4332 | int retval, oldprio, oldpolicy = -1; |
70b97a7f | 4333 | struct prio_array *array; |
1da177e4 | 4334 | unsigned long flags; |
70b97a7f | 4335 | struct rq *rq; |
1da177e4 | 4336 | |
66e5393a SR |
4337 | /* may grab non-irq protected spin_locks */ |
4338 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4339 | recheck: |
4340 | /* double check policy once rq lock held */ | |
4341 | if (policy < 0) | |
4342 | policy = oldpolicy = p->policy; | |
4343 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
b0a9499c IM |
4344 | policy != SCHED_NORMAL && policy != SCHED_BATCH) |
4345 | return -EINVAL; | |
1da177e4 LT |
4346 | /* |
4347 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
b0a9499c IM |
4348 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL and |
4349 | * SCHED_BATCH is 0. | |
1da177e4 LT |
4350 | */ |
4351 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4352 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4353 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4354 | return -EINVAL; |
57a6f51c | 4355 | if (is_rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4356 | return -EINVAL; |
4357 | ||
37e4ab3f OC |
4358 | /* |
4359 | * Allow unprivileged RT tasks to decrease priority: | |
4360 | */ | |
4361 | if (!capable(CAP_SYS_NICE)) { | |
8dc3e909 ON |
4362 | if (is_rt_policy(policy)) { |
4363 | unsigned long rlim_rtprio; | |
4364 | unsigned long flags; | |
4365 | ||
4366 | if (!lock_task_sighand(p, &flags)) | |
4367 | return -ESRCH; | |
4368 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4369 | unlock_task_sighand(p, &flags); | |
4370 | ||
4371 | /* can't set/change the rt policy */ | |
4372 | if (policy != p->policy && !rlim_rtprio) | |
4373 | return -EPERM; | |
4374 | ||
4375 | /* can't increase priority */ | |
4376 | if (param->sched_priority > p->rt_priority && | |
4377 | param->sched_priority > rlim_rtprio) | |
4378 | return -EPERM; | |
4379 | } | |
5fe1d75f | 4380 | |
37e4ab3f OC |
4381 | /* can't change other user's priorities */ |
4382 | if ((current->euid != p->euid) && | |
4383 | (current->euid != p->uid)) | |
4384 | return -EPERM; | |
4385 | } | |
1da177e4 LT |
4386 | |
4387 | retval = security_task_setscheduler(p, policy, param); | |
4388 | if (retval) | |
4389 | return retval; | |
b29739f9 IM |
4390 | /* |
4391 | * make sure no PI-waiters arrive (or leave) while we are | |
4392 | * changing the priority of the task: | |
4393 | */ | |
4394 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4395 | /* |
4396 | * To be able to change p->policy safely, the apropriate | |
4397 | * runqueue lock must be held. | |
4398 | */ | |
b29739f9 | 4399 | rq = __task_rq_lock(p); |
1da177e4 LT |
4400 | /* recheck policy now with rq lock held */ |
4401 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4402 | policy = oldpolicy = -1; | |
b29739f9 IM |
4403 | __task_rq_unlock(rq); |
4404 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4405 | goto recheck; |
4406 | } | |
4407 | array = p->array; | |
4408 | if (array) | |
4409 | deactivate_task(p, rq); | |
4410 | oldprio = p->prio; | |
4411 | __setscheduler(p, policy, param->sched_priority); | |
4412 | if (array) { | |
4413 | __activate_task(p, rq); | |
4414 | /* | |
4415 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4416 | * our priority decreased, or if we are not currently running on |
4417 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4418 | */ |
d5f9f942 AM |
4419 | if (task_running(rq, p)) { |
4420 | if (p->prio > oldprio) | |
4421 | resched_task(rq->curr); | |
4422 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
1da177e4 LT |
4423 | resched_task(rq->curr); |
4424 | } | |
b29739f9 IM |
4425 | __task_rq_unlock(rq); |
4426 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4427 | ||
95e02ca9 TG |
4428 | rt_mutex_adjust_pi(p); |
4429 | ||
1da177e4 LT |
4430 | return 0; |
4431 | } | |
4432 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4433 | ||
95cdf3b7 IM |
4434 | static int |
4435 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4436 | { |
1da177e4 LT |
4437 | struct sched_param lparam; |
4438 | struct task_struct *p; | |
36c8b586 | 4439 | int retval; |
1da177e4 LT |
4440 | |
4441 | if (!param || pid < 0) | |
4442 | return -EINVAL; | |
4443 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4444 | return -EFAULT; | |
5fe1d75f ON |
4445 | |
4446 | rcu_read_lock(); | |
4447 | retval = -ESRCH; | |
1da177e4 | 4448 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4449 | if (p != NULL) |
4450 | retval = sched_setscheduler(p, policy, &lparam); | |
4451 | rcu_read_unlock(); | |
36c8b586 | 4452 | |
1da177e4 LT |
4453 | return retval; |
4454 | } | |
4455 | ||
4456 | /** | |
4457 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4458 | * @pid: the pid in question. | |
4459 | * @policy: new policy. | |
4460 | * @param: structure containing the new RT priority. | |
4461 | */ | |
4462 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4463 | struct sched_param __user *param) | |
4464 | { | |
c21761f1 JB |
4465 | /* negative values for policy are not valid */ |
4466 | if (policy < 0) | |
4467 | return -EINVAL; | |
4468 | ||
1da177e4 LT |
4469 | return do_sched_setscheduler(pid, policy, param); |
4470 | } | |
4471 | ||
4472 | /** | |
4473 | * sys_sched_setparam - set/change the RT priority of a thread | |
4474 | * @pid: the pid in question. | |
4475 | * @param: structure containing the new RT priority. | |
4476 | */ | |
4477 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4478 | { | |
4479 | return do_sched_setscheduler(pid, -1, param); | |
4480 | } | |
4481 | ||
4482 | /** | |
4483 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4484 | * @pid: the pid in question. | |
4485 | */ | |
4486 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4487 | { | |
36c8b586 | 4488 | struct task_struct *p; |
1da177e4 | 4489 | int retval = -EINVAL; |
1da177e4 LT |
4490 | |
4491 | if (pid < 0) | |
4492 | goto out_nounlock; | |
4493 | ||
4494 | retval = -ESRCH; | |
4495 | read_lock(&tasklist_lock); | |
4496 | p = find_process_by_pid(pid); | |
4497 | if (p) { | |
4498 | retval = security_task_getscheduler(p); | |
4499 | if (!retval) | |
4500 | retval = p->policy; | |
4501 | } | |
4502 | read_unlock(&tasklist_lock); | |
4503 | ||
4504 | out_nounlock: | |
4505 | return retval; | |
4506 | } | |
4507 | ||
4508 | /** | |
4509 | * sys_sched_getscheduler - get the RT priority of a thread | |
4510 | * @pid: the pid in question. | |
4511 | * @param: structure containing the RT priority. | |
4512 | */ | |
4513 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4514 | { | |
4515 | struct sched_param lp; | |
36c8b586 | 4516 | struct task_struct *p; |
1da177e4 | 4517 | int retval = -EINVAL; |
1da177e4 LT |
4518 | |
4519 | if (!param || pid < 0) | |
4520 | goto out_nounlock; | |
4521 | ||
4522 | read_lock(&tasklist_lock); | |
4523 | p = find_process_by_pid(pid); | |
4524 | retval = -ESRCH; | |
4525 | if (!p) | |
4526 | goto out_unlock; | |
4527 | ||
4528 | retval = security_task_getscheduler(p); | |
4529 | if (retval) | |
4530 | goto out_unlock; | |
4531 | ||
4532 | lp.sched_priority = p->rt_priority; | |
4533 | read_unlock(&tasklist_lock); | |
4534 | ||
4535 | /* | |
4536 | * This one might sleep, we cannot do it with a spinlock held ... | |
4537 | */ | |
4538 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4539 | ||
4540 | out_nounlock: | |
4541 | return retval; | |
4542 | ||
4543 | out_unlock: | |
4544 | read_unlock(&tasklist_lock); | |
4545 | return retval; | |
4546 | } | |
4547 | ||
4548 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4549 | { | |
1da177e4 | 4550 | cpumask_t cpus_allowed; |
36c8b586 IM |
4551 | struct task_struct *p; |
4552 | int retval; | |
1da177e4 | 4553 | |
5be9361c | 4554 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4555 | read_lock(&tasklist_lock); |
4556 | ||
4557 | p = find_process_by_pid(pid); | |
4558 | if (!p) { | |
4559 | read_unlock(&tasklist_lock); | |
5be9361c | 4560 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4561 | return -ESRCH; |
4562 | } | |
4563 | ||
4564 | /* | |
4565 | * It is not safe to call set_cpus_allowed with the | |
4566 | * tasklist_lock held. We will bump the task_struct's | |
4567 | * usage count and then drop tasklist_lock. | |
4568 | */ | |
4569 | get_task_struct(p); | |
4570 | read_unlock(&tasklist_lock); | |
4571 | ||
4572 | retval = -EPERM; | |
4573 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4574 | !capable(CAP_SYS_NICE)) | |
4575 | goto out_unlock; | |
4576 | ||
e7834f8f DQ |
4577 | retval = security_task_setscheduler(p, 0, NULL); |
4578 | if (retval) | |
4579 | goto out_unlock; | |
4580 | ||
1da177e4 LT |
4581 | cpus_allowed = cpuset_cpus_allowed(p); |
4582 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4583 | retval = set_cpus_allowed(p, new_mask); | |
4584 | ||
4585 | out_unlock: | |
4586 | put_task_struct(p); | |
5be9361c | 4587 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4588 | return retval; |
4589 | } | |
4590 | ||
4591 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4592 | cpumask_t *new_mask) | |
4593 | { | |
4594 | if (len < sizeof(cpumask_t)) { | |
4595 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4596 | } else if (len > sizeof(cpumask_t)) { | |
4597 | len = sizeof(cpumask_t); | |
4598 | } | |
4599 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4600 | } | |
4601 | ||
4602 | /** | |
4603 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4604 | * @pid: pid of the process | |
4605 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4606 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4607 | */ | |
4608 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4609 | unsigned long __user *user_mask_ptr) | |
4610 | { | |
4611 | cpumask_t new_mask; | |
4612 | int retval; | |
4613 | ||
4614 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4615 | if (retval) | |
4616 | return retval; | |
4617 | ||
4618 | return sched_setaffinity(pid, new_mask); | |
4619 | } | |
4620 | ||
4621 | /* | |
4622 | * Represents all cpu's present in the system | |
4623 | * In systems capable of hotplug, this map could dynamically grow | |
4624 | * as new cpu's are detected in the system via any platform specific | |
4625 | * method, such as ACPI for e.g. | |
4626 | */ | |
4627 | ||
4cef0c61 | 4628 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4629 | EXPORT_SYMBOL(cpu_present_map); |
4630 | ||
4631 | #ifndef CONFIG_SMP | |
4cef0c61 | 4632 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4633 | EXPORT_SYMBOL(cpu_online_map); |
4634 | ||
4cef0c61 | 4635 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4636 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4637 | #endif |
4638 | ||
4639 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4640 | { | |
36c8b586 | 4641 | struct task_struct *p; |
1da177e4 | 4642 | int retval; |
1da177e4 | 4643 | |
5be9361c | 4644 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4645 | read_lock(&tasklist_lock); |
4646 | ||
4647 | retval = -ESRCH; | |
4648 | p = find_process_by_pid(pid); | |
4649 | if (!p) | |
4650 | goto out_unlock; | |
4651 | ||
e7834f8f DQ |
4652 | retval = security_task_getscheduler(p); |
4653 | if (retval) | |
4654 | goto out_unlock; | |
4655 | ||
2f7016d9 | 4656 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4657 | |
4658 | out_unlock: | |
4659 | read_unlock(&tasklist_lock); | |
5be9361c | 4660 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4661 | if (retval) |
4662 | return retval; | |
4663 | ||
4664 | return 0; | |
4665 | } | |
4666 | ||
4667 | /** | |
4668 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4669 | * @pid: pid of the process | |
4670 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4671 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4672 | */ | |
4673 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4674 | unsigned long __user *user_mask_ptr) | |
4675 | { | |
4676 | int ret; | |
4677 | cpumask_t mask; | |
4678 | ||
4679 | if (len < sizeof(cpumask_t)) | |
4680 | return -EINVAL; | |
4681 | ||
4682 | ret = sched_getaffinity(pid, &mask); | |
4683 | if (ret < 0) | |
4684 | return ret; | |
4685 | ||
4686 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4687 | return -EFAULT; | |
4688 | ||
4689 | return sizeof(cpumask_t); | |
4690 | } | |
4691 | ||
4692 | /** | |
4693 | * sys_sched_yield - yield the current processor to other threads. | |
4694 | * | |
72fd4a35 | 4695 | * This function yields the current CPU by moving the calling thread |
1da177e4 LT |
4696 | * to the expired array. If there are no other threads running on this |
4697 | * CPU then this function will return. | |
4698 | */ | |
4699 | asmlinkage long sys_sched_yield(void) | |
4700 | { | |
70b97a7f IM |
4701 | struct rq *rq = this_rq_lock(); |
4702 | struct prio_array *array = current->array, *target = rq->expired; | |
1da177e4 LT |
4703 | |
4704 | schedstat_inc(rq, yld_cnt); | |
4705 | /* | |
4706 | * We implement yielding by moving the task into the expired | |
4707 | * queue. | |
4708 | * | |
4709 | * (special rule: RT tasks will just roundrobin in the active | |
4710 | * array.) | |
4711 | */ | |
4712 | if (rt_task(current)) | |
4713 | target = rq->active; | |
4714 | ||
5927ad78 | 4715 | if (array->nr_active == 1) { |
1da177e4 LT |
4716 | schedstat_inc(rq, yld_act_empty); |
4717 | if (!rq->expired->nr_active) | |
4718 | schedstat_inc(rq, yld_both_empty); | |
4719 | } else if (!rq->expired->nr_active) | |
4720 | schedstat_inc(rq, yld_exp_empty); | |
4721 | ||
4722 | if (array != target) { | |
4723 | dequeue_task(current, array); | |
4724 | enqueue_task(current, target); | |
4725 | } else | |
4726 | /* | |
4727 | * requeue_task is cheaper so perform that if possible. | |
4728 | */ | |
4729 | requeue_task(current, array); | |
4730 | ||
4731 | /* | |
4732 | * Since we are going to call schedule() anyway, there's | |
4733 | * no need to preempt or enable interrupts: | |
4734 | */ | |
4735 | __release(rq->lock); | |
8a25d5de | 4736 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4737 | _raw_spin_unlock(&rq->lock); |
4738 | preempt_enable_no_resched(); | |
4739 | ||
4740 | schedule(); | |
4741 | ||
4742 | return 0; | |
4743 | } | |
4744 | ||
e7b38404 | 4745 | static void __cond_resched(void) |
1da177e4 | 4746 | { |
8e0a43d8 IM |
4747 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4748 | __might_sleep(__FILE__, __LINE__); | |
4749 | #endif | |
5bbcfd90 IM |
4750 | /* |
4751 | * The BKS might be reacquired before we have dropped | |
4752 | * PREEMPT_ACTIVE, which could trigger a second | |
4753 | * cond_resched() call. | |
4754 | */ | |
1da177e4 LT |
4755 | do { |
4756 | add_preempt_count(PREEMPT_ACTIVE); | |
4757 | schedule(); | |
4758 | sub_preempt_count(PREEMPT_ACTIVE); | |
4759 | } while (need_resched()); | |
4760 | } | |
4761 | ||
4762 | int __sched cond_resched(void) | |
4763 | { | |
9414232f IM |
4764 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4765 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4766 | __cond_resched(); |
4767 | return 1; | |
4768 | } | |
4769 | return 0; | |
4770 | } | |
1da177e4 LT |
4771 | EXPORT_SYMBOL(cond_resched); |
4772 | ||
4773 | /* | |
4774 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4775 | * call schedule, and on return reacquire the lock. | |
4776 | * | |
4777 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4778 | * operations here to prevent schedule() from being called twice (once via | |
4779 | * spin_unlock(), once by hand). | |
4780 | */ | |
95cdf3b7 | 4781 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4782 | { |
6df3cecb JK |
4783 | int ret = 0; |
4784 | ||
1da177e4 LT |
4785 | if (need_lockbreak(lock)) { |
4786 | spin_unlock(lock); | |
4787 | cpu_relax(); | |
6df3cecb | 4788 | ret = 1; |
1da177e4 LT |
4789 | spin_lock(lock); |
4790 | } | |
9414232f | 4791 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4792 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4793 | _raw_spin_unlock(lock); |
4794 | preempt_enable_no_resched(); | |
4795 | __cond_resched(); | |
6df3cecb | 4796 | ret = 1; |
1da177e4 | 4797 | spin_lock(lock); |
1da177e4 | 4798 | } |
6df3cecb | 4799 | return ret; |
1da177e4 | 4800 | } |
1da177e4 LT |
4801 | EXPORT_SYMBOL(cond_resched_lock); |
4802 | ||
4803 | int __sched cond_resched_softirq(void) | |
4804 | { | |
4805 | BUG_ON(!in_softirq()); | |
4806 | ||
9414232f | 4807 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4808 | local_bh_enable(); |
1da177e4 LT |
4809 | __cond_resched(); |
4810 | local_bh_disable(); | |
4811 | return 1; | |
4812 | } | |
4813 | return 0; | |
4814 | } | |
1da177e4 LT |
4815 | EXPORT_SYMBOL(cond_resched_softirq); |
4816 | ||
1da177e4 LT |
4817 | /** |
4818 | * yield - yield the current processor to other threads. | |
4819 | * | |
72fd4a35 | 4820 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4821 | * thread runnable and calls sys_sched_yield(). |
4822 | */ | |
4823 | void __sched yield(void) | |
4824 | { | |
4825 | set_current_state(TASK_RUNNING); | |
4826 | sys_sched_yield(); | |
4827 | } | |
1da177e4 LT |
4828 | EXPORT_SYMBOL(yield); |
4829 | ||
4830 | /* | |
4831 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4832 | * that process accounting knows that this is a task in IO wait state. | |
4833 | * | |
4834 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4835 | * has set its backing_dev_info: the queue against which it should throttle) | |
4836 | */ | |
4837 | void __sched io_schedule(void) | |
4838 | { | |
70b97a7f | 4839 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4840 | |
0ff92245 | 4841 | delayacct_blkio_start(); |
1da177e4 LT |
4842 | atomic_inc(&rq->nr_iowait); |
4843 | schedule(); | |
4844 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4845 | delayacct_blkio_end(); |
1da177e4 | 4846 | } |
1da177e4 LT |
4847 | EXPORT_SYMBOL(io_schedule); |
4848 | ||
4849 | long __sched io_schedule_timeout(long timeout) | |
4850 | { | |
70b97a7f | 4851 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4852 | long ret; |
4853 | ||
0ff92245 | 4854 | delayacct_blkio_start(); |
1da177e4 LT |
4855 | atomic_inc(&rq->nr_iowait); |
4856 | ret = schedule_timeout(timeout); | |
4857 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4858 | delayacct_blkio_end(); |
1da177e4 LT |
4859 | return ret; |
4860 | } | |
4861 | ||
4862 | /** | |
4863 | * sys_sched_get_priority_max - return maximum RT priority. | |
4864 | * @policy: scheduling class. | |
4865 | * | |
4866 | * this syscall returns the maximum rt_priority that can be used | |
4867 | * by a given scheduling class. | |
4868 | */ | |
4869 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4870 | { | |
4871 | int ret = -EINVAL; | |
4872 | ||
4873 | switch (policy) { | |
4874 | case SCHED_FIFO: | |
4875 | case SCHED_RR: | |
4876 | ret = MAX_USER_RT_PRIO-1; | |
4877 | break; | |
4878 | case SCHED_NORMAL: | |
b0a9499c | 4879 | case SCHED_BATCH: |
1da177e4 LT |
4880 | ret = 0; |
4881 | break; | |
4882 | } | |
4883 | return ret; | |
4884 | } | |
4885 | ||
4886 | /** | |
4887 | * sys_sched_get_priority_min - return minimum RT priority. | |
4888 | * @policy: scheduling class. | |
4889 | * | |
4890 | * this syscall returns the minimum rt_priority that can be used | |
4891 | * by a given scheduling class. | |
4892 | */ | |
4893 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4894 | { | |
4895 | int ret = -EINVAL; | |
4896 | ||
4897 | switch (policy) { | |
4898 | case SCHED_FIFO: | |
4899 | case SCHED_RR: | |
4900 | ret = 1; | |
4901 | break; | |
4902 | case SCHED_NORMAL: | |
b0a9499c | 4903 | case SCHED_BATCH: |
1da177e4 LT |
4904 | ret = 0; |
4905 | } | |
4906 | return ret; | |
4907 | } | |
4908 | ||
4909 | /** | |
4910 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4911 | * @pid: pid of the process. | |
4912 | * @interval: userspace pointer to the timeslice value. | |
4913 | * | |
4914 | * this syscall writes the default timeslice value of a given process | |
4915 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4916 | */ | |
4917 | asmlinkage | |
4918 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4919 | { | |
36c8b586 | 4920 | struct task_struct *p; |
1da177e4 LT |
4921 | int retval = -EINVAL; |
4922 | struct timespec t; | |
1da177e4 LT |
4923 | |
4924 | if (pid < 0) | |
4925 | goto out_nounlock; | |
4926 | ||
4927 | retval = -ESRCH; | |
4928 | read_lock(&tasklist_lock); | |
4929 | p = find_process_by_pid(pid); | |
4930 | if (!p) | |
4931 | goto out_unlock; | |
4932 | ||
4933 | retval = security_task_getscheduler(p); | |
4934 | if (retval) | |
4935 | goto out_unlock; | |
4936 | ||
b78709cf | 4937 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
1da177e4 LT |
4938 | 0 : task_timeslice(p), &t); |
4939 | read_unlock(&tasklist_lock); | |
4940 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4941 | out_nounlock: | |
4942 | return retval; | |
4943 | out_unlock: | |
4944 | read_unlock(&tasklist_lock); | |
4945 | return retval; | |
4946 | } | |
4947 | ||
2ed6e34f | 4948 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4949 | |
4950 | static void show_task(struct task_struct *p) | |
1da177e4 | 4951 | { |
1da177e4 | 4952 | unsigned long free = 0; |
36c8b586 | 4953 | unsigned state; |
1da177e4 | 4954 | |
1da177e4 | 4955 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4956 | printk("%-13.13s %c", p->comm, |
4957 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
1da177e4 LT |
4958 | #if (BITS_PER_LONG == 32) |
4959 | if (state == TASK_RUNNING) | |
4960 | printk(" running "); | |
4961 | else | |
4962 | printk(" %08lX ", thread_saved_pc(p)); | |
4963 | #else | |
4964 | if (state == TASK_RUNNING) | |
4965 | printk(" running task "); | |
4966 | else | |
4967 | printk(" %016lx ", thread_saved_pc(p)); | |
4968 | #endif | |
4969 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4970 | { | |
10ebffde | 4971 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4972 | while (!*n) |
4973 | n++; | |
10ebffde | 4974 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4975 | } |
4976 | #endif | |
35f6f753 | 4977 | printk("%5lu %5d %6d", free, p->pid, p->parent->pid); |
1da177e4 LT |
4978 | if (!p->mm) |
4979 | printk(" (L-TLB)\n"); | |
4980 | else | |
4981 | printk(" (NOTLB)\n"); | |
4982 | ||
4983 | if (state != TASK_RUNNING) | |
4984 | show_stack(p, NULL); | |
4985 | } | |
4986 | ||
e59e2ae2 | 4987 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4988 | { |
36c8b586 | 4989 | struct task_struct *g, *p; |
1da177e4 LT |
4990 | |
4991 | #if (BITS_PER_LONG == 32) | |
4992 | printk("\n" | |
301827ac CC |
4993 | " free sibling\n"); |
4994 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
4995 | #else |
4996 | printk("\n" | |
301827ac CC |
4997 | " free sibling\n"); |
4998 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
4999 | #endif |
5000 | read_lock(&tasklist_lock); | |
5001 | do_each_thread(g, p) { | |
5002 | /* | |
5003 | * reset the NMI-timeout, listing all files on a slow | |
5004 | * console might take alot of time: | |
5005 | */ | |
5006 | touch_nmi_watchdog(); | |
39bc89fd | 5007 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 5008 | show_task(p); |
1da177e4 LT |
5009 | } while_each_thread(g, p); |
5010 | ||
04c9167f JF |
5011 | touch_all_softlockup_watchdogs(); |
5012 | ||
1da177e4 | 5013 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5014 | /* |
5015 | * Only show locks if all tasks are dumped: | |
5016 | */ | |
5017 | if (state_filter == -1) | |
5018 | debug_show_all_locks(); | |
1da177e4 LT |
5019 | } |
5020 | ||
f340c0d1 IM |
5021 | /** |
5022 | * init_idle - set up an idle thread for a given CPU | |
5023 | * @idle: task in question | |
5024 | * @cpu: cpu the idle task belongs to | |
5025 | * | |
5026 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5027 | * flag, to make booting more robust. | |
5028 | */ | |
5c1e1767 | 5029 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5030 | { |
70b97a7f | 5031 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5032 | unsigned long flags; |
5033 | ||
81c29a85 | 5034 | idle->timestamp = sched_clock(); |
1da177e4 LT |
5035 | idle->sleep_avg = 0; |
5036 | idle->array = NULL; | |
b29739f9 | 5037 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 LT |
5038 | idle->state = TASK_RUNNING; |
5039 | idle->cpus_allowed = cpumask_of_cpu(cpu); | |
5040 | set_task_cpu(idle, cpu); | |
5041 | ||
5042 | spin_lock_irqsave(&rq->lock, flags); | |
5043 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5044 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5045 | idle->oncpu = 1; | |
5046 | #endif | |
1da177e4 LT |
5047 | spin_unlock_irqrestore(&rq->lock, flags); |
5048 | ||
5049 | /* Set the preempt count _outside_ the spinlocks! */ | |
5050 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 5051 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 5052 | #else |
a1261f54 | 5053 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 LT |
5054 | #endif |
5055 | } | |
5056 | ||
5057 | /* | |
5058 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5059 | * indicates which cpus entered this state. This is used | |
5060 | * in the rcu update to wait only for active cpus. For system | |
5061 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5062 | * always be CPU_MASK_NONE. | |
5063 | */ | |
5064 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5065 | ||
5066 | #ifdef CONFIG_SMP | |
5067 | /* | |
5068 | * This is how migration works: | |
5069 | * | |
70b97a7f | 5070 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5071 | * runqueue and wake up that CPU's migration thread. |
5072 | * 2) we down() the locked semaphore => thread blocks. | |
5073 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5074 | * thread off the CPU) | |
5075 | * 4) it gets the migration request and checks whether the migrated | |
5076 | * task is still in the wrong runqueue. | |
5077 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5078 | * it and puts it into the right queue. | |
5079 | * 6) migration thread up()s the semaphore. | |
5080 | * 7) we wake up and the migration is done. | |
5081 | */ | |
5082 | ||
5083 | /* | |
5084 | * Change a given task's CPU affinity. Migrate the thread to a | |
5085 | * proper CPU and schedule it away if the CPU it's executing on | |
5086 | * is removed from the allowed bitmask. | |
5087 | * | |
5088 | * NOTE: the caller must have a valid reference to the task, the | |
5089 | * task must not exit() & deallocate itself prematurely. The | |
5090 | * call is not atomic; no spinlocks may be held. | |
5091 | */ | |
36c8b586 | 5092 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5093 | { |
70b97a7f | 5094 | struct migration_req req; |
1da177e4 | 5095 | unsigned long flags; |
70b97a7f | 5096 | struct rq *rq; |
48f24c4d | 5097 | int ret = 0; |
1da177e4 LT |
5098 | |
5099 | rq = task_rq_lock(p, &flags); | |
5100 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5101 | ret = -EINVAL; | |
5102 | goto out; | |
5103 | } | |
5104 | ||
5105 | p->cpus_allowed = new_mask; | |
5106 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5107 | if (cpu_isset(task_cpu(p), new_mask)) | |
5108 | goto out; | |
5109 | ||
5110 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5111 | /* Need help from migration thread: drop lock and wait. */ | |
5112 | task_rq_unlock(rq, &flags); | |
5113 | wake_up_process(rq->migration_thread); | |
5114 | wait_for_completion(&req.done); | |
5115 | tlb_migrate_finish(p->mm); | |
5116 | return 0; | |
5117 | } | |
5118 | out: | |
5119 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5120 | |
1da177e4 LT |
5121 | return ret; |
5122 | } | |
1da177e4 LT |
5123 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5124 | ||
5125 | /* | |
5126 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5127 | * this because either it can't run here any more (set_cpus_allowed() | |
5128 | * away from this CPU, or CPU going down), or because we're | |
5129 | * attempting to rebalance this task on exec (sched_exec). | |
5130 | * | |
5131 | * So we race with normal scheduler movements, but that's OK, as long | |
5132 | * as the task is no longer on this CPU. | |
efc30814 KK |
5133 | * |
5134 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5135 | */ |
efc30814 | 5136 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5137 | { |
70b97a7f | 5138 | struct rq *rq_dest, *rq_src; |
efc30814 | 5139 | int ret = 0; |
1da177e4 LT |
5140 | |
5141 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5142 | return ret; |
1da177e4 LT |
5143 | |
5144 | rq_src = cpu_rq(src_cpu); | |
5145 | rq_dest = cpu_rq(dest_cpu); | |
5146 | ||
5147 | double_rq_lock(rq_src, rq_dest); | |
5148 | /* Already moved. */ | |
5149 | if (task_cpu(p) != src_cpu) | |
5150 | goto out; | |
5151 | /* Affinity changed (again). */ | |
5152 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5153 | goto out; | |
5154 | ||
5155 | set_task_cpu(p, dest_cpu); | |
5156 | if (p->array) { | |
5157 | /* | |
5158 | * Sync timestamp with rq_dest's before activating. | |
5159 | * The same thing could be achieved by doing this step | |
5160 | * afterwards, and pretending it was a local activate. | |
5161 | * This way is cleaner and logically correct. | |
5162 | */ | |
b18ec803 MG |
5163 | p->timestamp = p->timestamp - rq_src->most_recent_timestamp |
5164 | + rq_dest->most_recent_timestamp; | |
1da177e4 | 5165 | deactivate_task(p, rq_src); |
0a565f79 | 5166 | __activate_task(p, rq_dest); |
1da177e4 LT |
5167 | if (TASK_PREEMPTS_CURR(p, rq_dest)) |
5168 | resched_task(rq_dest->curr); | |
5169 | } | |
efc30814 | 5170 | ret = 1; |
1da177e4 LT |
5171 | out: |
5172 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5173 | return ret; |
1da177e4 LT |
5174 | } |
5175 | ||
5176 | /* | |
5177 | * migration_thread - this is a highprio system thread that performs | |
5178 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5179 | * another runqueue. | |
5180 | */ | |
95cdf3b7 | 5181 | static int migration_thread(void *data) |
1da177e4 | 5182 | { |
1da177e4 | 5183 | int cpu = (long)data; |
70b97a7f | 5184 | struct rq *rq; |
1da177e4 LT |
5185 | |
5186 | rq = cpu_rq(cpu); | |
5187 | BUG_ON(rq->migration_thread != current); | |
5188 | ||
5189 | set_current_state(TASK_INTERRUPTIBLE); | |
5190 | while (!kthread_should_stop()) { | |
70b97a7f | 5191 | struct migration_req *req; |
1da177e4 | 5192 | struct list_head *head; |
1da177e4 | 5193 | |
3e1d1d28 | 5194 | try_to_freeze(); |
1da177e4 LT |
5195 | |
5196 | spin_lock_irq(&rq->lock); | |
5197 | ||
5198 | if (cpu_is_offline(cpu)) { | |
5199 | spin_unlock_irq(&rq->lock); | |
5200 | goto wait_to_die; | |
5201 | } | |
5202 | ||
5203 | if (rq->active_balance) { | |
5204 | active_load_balance(rq, cpu); | |
5205 | rq->active_balance = 0; | |
5206 | } | |
5207 | ||
5208 | head = &rq->migration_queue; | |
5209 | ||
5210 | if (list_empty(head)) { | |
5211 | spin_unlock_irq(&rq->lock); | |
5212 | schedule(); | |
5213 | set_current_state(TASK_INTERRUPTIBLE); | |
5214 | continue; | |
5215 | } | |
70b97a7f | 5216 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5217 | list_del_init(head->next); |
5218 | ||
674311d5 NP |
5219 | spin_unlock(&rq->lock); |
5220 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5221 | local_irq_enable(); | |
1da177e4 LT |
5222 | |
5223 | complete(&req->done); | |
5224 | } | |
5225 | __set_current_state(TASK_RUNNING); | |
5226 | return 0; | |
5227 | ||
5228 | wait_to_die: | |
5229 | /* Wait for kthread_stop */ | |
5230 | set_current_state(TASK_INTERRUPTIBLE); | |
5231 | while (!kthread_should_stop()) { | |
5232 | schedule(); | |
5233 | set_current_state(TASK_INTERRUPTIBLE); | |
5234 | } | |
5235 | __set_current_state(TASK_RUNNING); | |
5236 | return 0; | |
5237 | } | |
5238 | ||
5239 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5240 | /* |
5241 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5242 | * NOTE: interrupts should be disabled by the caller | |
5243 | */ | |
48f24c4d | 5244 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5245 | { |
efc30814 | 5246 | unsigned long flags; |
1da177e4 | 5247 | cpumask_t mask; |
70b97a7f IM |
5248 | struct rq *rq; |
5249 | int dest_cpu; | |
1da177e4 | 5250 | |
efc30814 | 5251 | restart: |
1da177e4 LT |
5252 | /* On same node? */ |
5253 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5254 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5255 | dest_cpu = any_online_cpu(mask); |
5256 | ||
5257 | /* On any allowed CPU? */ | |
5258 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5259 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5260 | |
5261 | /* No more Mr. Nice Guy. */ | |
5262 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5263 | rq = task_rq_lock(p, &flags); |
5264 | cpus_setall(p->cpus_allowed); | |
5265 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5266 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5267 | |
5268 | /* | |
5269 | * Don't tell them about moving exiting tasks or | |
5270 | * kernel threads (both mm NULL), since they never | |
5271 | * leave kernel. | |
5272 | */ | |
48f24c4d | 5273 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5274 | printk(KERN_INFO "process %d (%s) no " |
5275 | "longer affine to cpu%d\n", | |
48f24c4d | 5276 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5277 | } |
48f24c4d | 5278 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5279 | goto restart; |
1da177e4 LT |
5280 | } |
5281 | ||
5282 | /* | |
5283 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5284 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5285 | * for performance reasons the counter is not stricly tracking tasks to | |
5286 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5287 | * to keep the global sum constant after CPU-down: | |
5288 | */ | |
70b97a7f | 5289 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5290 | { |
70b97a7f | 5291 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5292 | unsigned long flags; |
5293 | ||
5294 | local_irq_save(flags); | |
5295 | double_rq_lock(rq_src, rq_dest); | |
5296 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5297 | rq_src->nr_uninterruptible = 0; | |
5298 | double_rq_unlock(rq_src, rq_dest); | |
5299 | local_irq_restore(flags); | |
5300 | } | |
5301 | ||
5302 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5303 | static void migrate_live_tasks(int src_cpu) | |
5304 | { | |
48f24c4d | 5305 | struct task_struct *p, *t; |
1da177e4 LT |
5306 | |
5307 | write_lock_irq(&tasklist_lock); | |
5308 | ||
48f24c4d IM |
5309 | do_each_thread(t, p) { |
5310 | if (p == current) | |
1da177e4 LT |
5311 | continue; |
5312 | ||
48f24c4d IM |
5313 | if (task_cpu(p) == src_cpu) |
5314 | move_task_off_dead_cpu(src_cpu, p); | |
5315 | } while_each_thread(t, p); | |
1da177e4 LT |
5316 | |
5317 | write_unlock_irq(&tasklist_lock); | |
5318 | } | |
5319 | ||
5320 | /* Schedules idle task to be the next runnable task on current CPU. | |
5321 | * It does so by boosting its priority to highest possible and adding it to | |
48f24c4d | 5322 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5323 | */ |
5324 | void sched_idle_next(void) | |
5325 | { | |
48f24c4d | 5326 | int this_cpu = smp_processor_id(); |
70b97a7f | 5327 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5328 | struct task_struct *p = rq->idle; |
5329 | unsigned long flags; | |
5330 | ||
5331 | /* cpu has to be offline */ | |
48f24c4d | 5332 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5333 | |
48f24c4d IM |
5334 | /* |
5335 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5336 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5337 | */ |
5338 | spin_lock_irqsave(&rq->lock, flags); | |
5339 | ||
5340 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
48f24c4d IM |
5341 | |
5342 | /* Add idle task to the _front_ of its priority queue: */ | |
1da177e4 LT |
5343 | __activate_idle_task(p, rq); |
5344 | ||
5345 | spin_unlock_irqrestore(&rq->lock, flags); | |
5346 | } | |
5347 | ||
48f24c4d IM |
5348 | /* |
5349 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5350 | * offline. |
5351 | */ | |
5352 | void idle_task_exit(void) | |
5353 | { | |
5354 | struct mm_struct *mm = current->active_mm; | |
5355 | ||
5356 | BUG_ON(cpu_online(smp_processor_id())); | |
5357 | ||
5358 | if (mm != &init_mm) | |
5359 | switch_mm(mm, &init_mm, current); | |
5360 | mmdrop(mm); | |
5361 | } | |
5362 | ||
054b9108 | 5363 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5364 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5365 | { |
70b97a7f | 5366 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5367 | |
5368 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5369 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5370 | |
5371 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5372 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5373 | |
48f24c4d | 5374 | get_task_struct(p); |
1da177e4 LT |
5375 | |
5376 | /* | |
5377 | * Drop lock around migration; if someone else moves it, | |
5378 | * that's OK. No task can be added to this CPU, so iteration is | |
5379 | * fine. | |
054b9108 | 5380 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5381 | */ |
054b9108 | 5382 | spin_unlock(&rq->lock); |
48f24c4d | 5383 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5384 | spin_lock(&rq->lock); |
1da177e4 | 5385 | |
48f24c4d | 5386 | put_task_struct(p); |
1da177e4 LT |
5387 | } |
5388 | ||
5389 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5390 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5391 | { | |
70b97a7f | 5392 | struct rq *rq = cpu_rq(dead_cpu); |
48f24c4d | 5393 | unsigned int arr, i; |
1da177e4 LT |
5394 | |
5395 | for (arr = 0; arr < 2; arr++) { | |
5396 | for (i = 0; i < MAX_PRIO; i++) { | |
5397 | struct list_head *list = &rq->arrays[arr].queue[i]; | |
48f24c4d | 5398 | |
1da177e4 | 5399 | while (!list_empty(list)) |
36c8b586 IM |
5400 | migrate_dead(dead_cpu, list_entry(list->next, |
5401 | struct task_struct, run_list)); | |
1da177e4 LT |
5402 | } |
5403 | } | |
5404 | } | |
5405 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5406 | ||
5407 | /* | |
5408 | * migration_call - callback that gets triggered when a CPU is added. | |
5409 | * Here we can start up the necessary migration thread for the new CPU. | |
5410 | */ | |
48f24c4d IM |
5411 | static int __cpuinit |
5412 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5413 | { |
1da177e4 | 5414 | struct task_struct *p; |
48f24c4d | 5415 | int cpu = (long)hcpu; |
1da177e4 | 5416 | unsigned long flags; |
70b97a7f | 5417 | struct rq *rq; |
1da177e4 LT |
5418 | |
5419 | switch (action) { | |
5be9361c GS |
5420 | case CPU_LOCK_ACQUIRE: |
5421 | mutex_lock(&sched_hotcpu_mutex); | |
5422 | break; | |
5423 | ||
1da177e4 | 5424 | case CPU_UP_PREPARE: |
8bb78442 | 5425 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
5426 | p = kthread_create(migration_thread, hcpu, "migration/%d",cpu); |
5427 | if (IS_ERR(p)) | |
5428 | return NOTIFY_BAD; | |
5429 | p->flags |= PF_NOFREEZE; | |
5430 | kthread_bind(p, cpu); | |
5431 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5432 | rq = task_rq_lock(p, &flags); | |
5433 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
5434 | task_rq_unlock(rq, &flags); | |
5435 | cpu_rq(cpu)->migration_thread = p; | |
5436 | break; | |
48f24c4d | 5437 | |
1da177e4 | 5438 | case CPU_ONLINE: |
8bb78442 | 5439 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5440 | /* Strictly unneccessary, as first user will wake it. */ |
5441 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5442 | break; | |
48f24c4d | 5443 | |
1da177e4 LT |
5444 | #ifdef CONFIG_HOTPLUG_CPU |
5445 | case CPU_UP_CANCELED: | |
8bb78442 | 5446 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5447 | if (!cpu_rq(cpu)->migration_thread) |
5448 | break; | |
1da177e4 | 5449 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5450 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5451 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5452 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5453 | cpu_rq(cpu)->migration_thread = NULL; | |
5454 | break; | |
48f24c4d | 5455 | |
1da177e4 | 5456 | case CPU_DEAD: |
8bb78442 | 5457 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5458 | migrate_live_tasks(cpu); |
5459 | rq = cpu_rq(cpu); | |
5460 | kthread_stop(rq->migration_thread); | |
5461 | rq->migration_thread = NULL; | |
5462 | /* Idle task back to normal (off runqueue, low prio) */ | |
5463 | rq = task_rq_lock(rq->idle, &flags); | |
5464 | deactivate_task(rq->idle, rq); | |
5465 | rq->idle->static_prio = MAX_PRIO; | |
5466 | __setscheduler(rq->idle, SCHED_NORMAL, 0); | |
5467 | migrate_dead_tasks(cpu); | |
5468 | task_rq_unlock(rq, &flags); | |
5469 | migrate_nr_uninterruptible(rq); | |
5470 | BUG_ON(rq->nr_running != 0); | |
5471 | ||
5472 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5473 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5474 | * the requestors. */ |
5475 | spin_lock_irq(&rq->lock); | |
5476 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5477 | struct migration_req *req; |
5478 | ||
1da177e4 | 5479 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5480 | struct migration_req, list); |
1da177e4 LT |
5481 | list_del_init(&req->list); |
5482 | complete(&req->done); | |
5483 | } | |
5484 | spin_unlock_irq(&rq->lock); | |
5485 | break; | |
5486 | #endif | |
5be9361c GS |
5487 | case CPU_LOCK_RELEASE: |
5488 | mutex_unlock(&sched_hotcpu_mutex); | |
5489 | break; | |
1da177e4 LT |
5490 | } |
5491 | return NOTIFY_OK; | |
5492 | } | |
5493 | ||
5494 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5495 | * happens before everything else. | |
5496 | */ | |
26c2143b | 5497 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5498 | .notifier_call = migration_call, |
5499 | .priority = 10 | |
5500 | }; | |
5501 | ||
5502 | int __init migration_init(void) | |
5503 | { | |
5504 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5505 | int err; |
48f24c4d IM |
5506 | |
5507 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5508 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5509 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5510 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5511 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5512 | |
1da177e4 LT |
5513 | return 0; |
5514 | } | |
5515 | #endif | |
5516 | ||
5517 | #ifdef CONFIG_SMP | |
476f3534 CL |
5518 | |
5519 | /* Number of possible processor ids */ | |
5520 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5521 | EXPORT_SYMBOL(nr_cpu_ids); | |
5522 | ||
1a20ff27 | 5523 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5524 | #ifdef SCHED_DOMAIN_DEBUG |
5525 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5526 | { | |
5527 | int level = 0; | |
5528 | ||
41c7ce9a NP |
5529 | if (!sd) { |
5530 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5531 | return; | |
5532 | } | |
5533 | ||
1da177e4 LT |
5534 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5535 | ||
5536 | do { | |
5537 | int i; | |
5538 | char str[NR_CPUS]; | |
5539 | struct sched_group *group = sd->groups; | |
5540 | cpumask_t groupmask; | |
5541 | ||
5542 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5543 | cpus_clear(groupmask); | |
5544 | ||
5545 | printk(KERN_DEBUG); | |
5546 | for (i = 0; i < level + 1; i++) | |
5547 | printk(" "); | |
5548 | printk("domain %d: ", level); | |
5549 | ||
5550 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5551 | printk("does not load-balance\n"); | |
5552 | if (sd->parent) | |
33859f7f MOS |
5553 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5554 | " has parent"); | |
1da177e4 LT |
5555 | break; |
5556 | } | |
5557 | ||
5558 | printk("span %s\n", str); | |
5559 | ||
5560 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5561 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5562 | "CPU%d\n", cpu); | |
1da177e4 | 5563 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5564 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5565 | " CPU%d\n", cpu); | |
1da177e4 LT |
5566 | |
5567 | printk(KERN_DEBUG); | |
5568 | for (i = 0; i < level + 2; i++) | |
5569 | printk(" "); | |
5570 | printk("groups:"); | |
5571 | do { | |
5572 | if (!group) { | |
5573 | printk("\n"); | |
5574 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5575 | break; | |
5576 | } | |
5577 | ||
5517d86b | 5578 | if (!group->__cpu_power) { |
1da177e4 | 5579 | printk("\n"); |
33859f7f MOS |
5580 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5581 | "set\n"); | |
1da177e4 LT |
5582 | } |
5583 | ||
5584 | if (!cpus_weight(group->cpumask)) { | |
5585 | printk("\n"); | |
5586 | printk(KERN_ERR "ERROR: empty group\n"); | |
5587 | } | |
5588 | ||
5589 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5590 | printk("\n"); | |
5591 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5592 | } | |
5593 | ||
5594 | cpus_or(groupmask, groupmask, group->cpumask); | |
5595 | ||
5596 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5597 | printk(" %s", str); | |
5598 | ||
5599 | group = group->next; | |
5600 | } while (group != sd->groups); | |
5601 | printk("\n"); | |
5602 | ||
5603 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5604 | printk(KERN_ERR "ERROR: groups don't span " |
5605 | "domain->span\n"); | |
1da177e4 LT |
5606 | |
5607 | level++; | |
5608 | sd = sd->parent; | |
33859f7f MOS |
5609 | if (!sd) |
5610 | continue; | |
1da177e4 | 5611 | |
33859f7f MOS |
5612 | if (!cpus_subset(groupmask, sd->span)) |
5613 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5614 | "of domain->span\n"); | |
1da177e4 LT |
5615 | |
5616 | } while (sd); | |
5617 | } | |
5618 | #else | |
48f24c4d | 5619 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5620 | #endif |
5621 | ||
1a20ff27 | 5622 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5623 | { |
5624 | if (cpus_weight(sd->span) == 1) | |
5625 | return 1; | |
5626 | ||
5627 | /* Following flags need at least 2 groups */ | |
5628 | if (sd->flags & (SD_LOAD_BALANCE | | |
5629 | SD_BALANCE_NEWIDLE | | |
5630 | SD_BALANCE_FORK | | |
89c4710e SS |
5631 | SD_BALANCE_EXEC | |
5632 | SD_SHARE_CPUPOWER | | |
5633 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5634 | if (sd->groups != sd->groups->next) |
5635 | return 0; | |
5636 | } | |
5637 | ||
5638 | /* Following flags don't use groups */ | |
5639 | if (sd->flags & (SD_WAKE_IDLE | | |
5640 | SD_WAKE_AFFINE | | |
5641 | SD_WAKE_BALANCE)) | |
5642 | return 0; | |
5643 | ||
5644 | return 1; | |
5645 | } | |
5646 | ||
48f24c4d IM |
5647 | static int |
5648 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5649 | { |
5650 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5651 | ||
5652 | if (sd_degenerate(parent)) | |
5653 | return 1; | |
5654 | ||
5655 | if (!cpus_equal(sd->span, parent->span)) | |
5656 | return 0; | |
5657 | ||
5658 | /* Does parent contain flags not in child? */ | |
5659 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5660 | if (cflags & SD_WAKE_AFFINE) | |
5661 | pflags &= ~SD_WAKE_BALANCE; | |
5662 | /* Flags needing groups don't count if only 1 group in parent */ | |
5663 | if (parent->groups == parent->groups->next) { | |
5664 | pflags &= ~(SD_LOAD_BALANCE | | |
5665 | SD_BALANCE_NEWIDLE | | |
5666 | SD_BALANCE_FORK | | |
89c4710e SS |
5667 | SD_BALANCE_EXEC | |
5668 | SD_SHARE_CPUPOWER | | |
5669 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5670 | } |
5671 | if (~cflags & pflags) | |
5672 | return 0; | |
5673 | ||
5674 | return 1; | |
5675 | } | |
5676 | ||
1da177e4 LT |
5677 | /* |
5678 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5679 | * hold the hotplug lock. | |
5680 | */ | |
9c1cfda2 | 5681 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5682 | { |
70b97a7f | 5683 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5684 | struct sched_domain *tmp; |
5685 | ||
5686 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5687 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5688 | struct sched_domain *parent = tmp->parent; | |
5689 | if (!parent) | |
5690 | break; | |
1a848870 | 5691 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5692 | tmp->parent = parent->parent; |
1a848870 SS |
5693 | if (parent->parent) |
5694 | parent->parent->child = tmp; | |
5695 | } | |
245af2c7 SS |
5696 | } |
5697 | ||
1a848870 | 5698 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5699 | sd = sd->parent; |
1a848870 SS |
5700 | if (sd) |
5701 | sd->child = NULL; | |
5702 | } | |
1da177e4 LT |
5703 | |
5704 | sched_domain_debug(sd, cpu); | |
5705 | ||
674311d5 | 5706 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5707 | } |
5708 | ||
5709 | /* cpus with isolated domains */ | |
67af63a6 | 5710 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5711 | |
5712 | /* Setup the mask of cpus configured for isolated domains */ | |
5713 | static int __init isolated_cpu_setup(char *str) | |
5714 | { | |
5715 | int ints[NR_CPUS], i; | |
5716 | ||
5717 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5718 | cpus_clear(cpu_isolated_map); | |
5719 | for (i = 1; i <= ints[0]; i++) | |
5720 | if (ints[i] < NR_CPUS) | |
5721 | cpu_set(ints[i], cpu_isolated_map); | |
5722 | return 1; | |
5723 | } | |
5724 | ||
5725 | __setup ("isolcpus=", isolated_cpu_setup); | |
5726 | ||
5727 | /* | |
6711cab4 SS |
5728 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5729 | * to a function which identifies what group(along with sched group) a CPU | |
5730 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5731 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5732 | * |
5733 | * init_sched_build_groups will build a circular linked list of the groups | |
5734 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5735 | * and ->cpu_power to 0. | |
5736 | */ | |
a616058b | 5737 | static void |
6711cab4 SS |
5738 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5739 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5740 | struct sched_group **sg)) | |
1da177e4 LT |
5741 | { |
5742 | struct sched_group *first = NULL, *last = NULL; | |
5743 | cpumask_t covered = CPU_MASK_NONE; | |
5744 | int i; | |
5745 | ||
5746 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5747 | struct sched_group *sg; |
5748 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5749 | int j; |
5750 | ||
5751 | if (cpu_isset(i, covered)) | |
5752 | continue; | |
5753 | ||
5754 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5755 | sg->__cpu_power = 0; |
1da177e4 LT |
5756 | |
5757 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5758 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5759 | continue; |
5760 | ||
5761 | cpu_set(j, covered); | |
5762 | cpu_set(j, sg->cpumask); | |
5763 | } | |
5764 | if (!first) | |
5765 | first = sg; | |
5766 | if (last) | |
5767 | last->next = sg; | |
5768 | last = sg; | |
5769 | } | |
5770 | last->next = first; | |
5771 | } | |
5772 | ||
9c1cfda2 | 5773 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5774 | |
9c1cfda2 | 5775 | #ifdef CONFIG_NUMA |
198e2f18 | 5776 | |
9c1cfda2 JH |
5777 | /** |
5778 | * find_next_best_node - find the next node to include in a sched_domain | |
5779 | * @node: node whose sched_domain we're building | |
5780 | * @used_nodes: nodes already in the sched_domain | |
5781 | * | |
5782 | * Find the next node to include in a given scheduling domain. Simply | |
5783 | * finds the closest node not already in the @used_nodes map. | |
5784 | * | |
5785 | * Should use nodemask_t. | |
5786 | */ | |
5787 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5788 | { | |
5789 | int i, n, val, min_val, best_node = 0; | |
5790 | ||
5791 | min_val = INT_MAX; | |
5792 | ||
5793 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5794 | /* Start at @node */ | |
5795 | n = (node + i) % MAX_NUMNODES; | |
5796 | ||
5797 | if (!nr_cpus_node(n)) | |
5798 | continue; | |
5799 | ||
5800 | /* Skip already used nodes */ | |
5801 | if (test_bit(n, used_nodes)) | |
5802 | continue; | |
5803 | ||
5804 | /* Simple min distance search */ | |
5805 | val = node_distance(node, n); | |
5806 | ||
5807 | if (val < min_val) { | |
5808 | min_val = val; | |
5809 | best_node = n; | |
5810 | } | |
5811 | } | |
5812 | ||
5813 | set_bit(best_node, used_nodes); | |
5814 | return best_node; | |
5815 | } | |
5816 | ||
5817 | /** | |
5818 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5819 | * @node: node whose cpumask we're constructing | |
5820 | * @size: number of nodes to include in this span | |
5821 | * | |
5822 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5823 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5824 | * out optimally. | |
5825 | */ | |
5826 | static cpumask_t sched_domain_node_span(int node) | |
5827 | { | |
9c1cfda2 | 5828 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5829 | cpumask_t span, nodemask; |
5830 | int i; | |
9c1cfda2 JH |
5831 | |
5832 | cpus_clear(span); | |
5833 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5834 | ||
5835 | nodemask = node_to_cpumask(node); | |
5836 | cpus_or(span, span, nodemask); | |
5837 | set_bit(node, used_nodes); | |
5838 | ||
5839 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5840 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5841 | |
9c1cfda2 JH |
5842 | nodemask = node_to_cpumask(next_node); |
5843 | cpus_or(span, span, nodemask); | |
5844 | } | |
5845 | ||
5846 | return span; | |
5847 | } | |
5848 | #endif | |
5849 | ||
5c45bf27 | 5850 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5851 | |
9c1cfda2 | 5852 | /* |
48f24c4d | 5853 | * SMT sched-domains: |
9c1cfda2 | 5854 | */ |
1da177e4 LT |
5855 | #ifdef CONFIG_SCHED_SMT |
5856 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5857 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5858 | |
6711cab4 SS |
5859 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5860 | struct sched_group **sg) | |
1da177e4 | 5861 | { |
6711cab4 SS |
5862 | if (sg) |
5863 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5864 | return cpu; |
5865 | } | |
5866 | #endif | |
5867 | ||
48f24c4d IM |
5868 | /* |
5869 | * multi-core sched-domains: | |
5870 | */ | |
1e9f28fa SS |
5871 | #ifdef CONFIG_SCHED_MC |
5872 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5873 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5874 | #endif |
5875 | ||
5876 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5877 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5878 | struct sched_group **sg) | |
1e9f28fa | 5879 | { |
6711cab4 | 5880 | int group; |
a616058b SS |
5881 | cpumask_t mask = cpu_sibling_map[cpu]; |
5882 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5883 | group = first_cpu(mask); |
5884 | if (sg) | |
5885 | *sg = &per_cpu(sched_group_core, group); | |
5886 | return group; | |
1e9f28fa SS |
5887 | } |
5888 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5889 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5890 | struct sched_group **sg) | |
1e9f28fa | 5891 | { |
6711cab4 SS |
5892 | if (sg) |
5893 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5894 | return cpu; |
5895 | } | |
5896 | #endif | |
5897 | ||
1da177e4 | 5898 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5899 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5900 | |
6711cab4 SS |
5901 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5902 | struct sched_group **sg) | |
1da177e4 | 5903 | { |
6711cab4 | 5904 | int group; |
48f24c4d | 5905 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5906 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5907 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5908 | group = first_cpu(mask); |
1e9f28fa | 5909 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5910 | cpumask_t mask = cpu_sibling_map[cpu]; |
5911 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5912 | group = first_cpu(mask); |
1da177e4 | 5913 | #else |
6711cab4 | 5914 | group = cpu; |
1da177e4 | 5915 | #endif |
6711cab4 SS |
5916 | if (sg) |
5917 | *sg = &per_cpu(sched_group_phys, group); | |
5918 | return group; | |
1da177e4 LT |
5919 | } |
5920 | ||
5921 | #ifdef CONFIG_NUMA | |
1da177e4 | 5922 | /* |
9c1cfda2 JH |
5923 | * The init_sched_build_groups can't handle what we want to do with node |
5924 | * groups, so roll our own. Now each node has its own list of groups which | |
5925 | * gets dynamically allocated. | |
1da177e4 | 5926 | */ |
9c1cfda2 | 5927 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5928 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5929 | |
9c1cfda2 | 5930 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5931 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5932 | |
6711cab4 SS |
5933 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5934 | struct sched_group **sg) | |
9c1cfda2 | 5935 | { |
6711cab4 SS |
5936 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5937 | int group; | |
5938 | ||
5939 | cpus_and(nodemask, nodemask, *cpu_map); | |
5940 | group = first_cpu(nodemask); | |
5941 | ||
5942 | if (sg) | |
5943 | *sg = &per_cpu(sched_group_allnodes, group); | |
5944 | return group; | |
1da177e4 | 5945 | } |
6711cab4 | 5946 | |
08069033 SS |
5947 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5948 | { | |
5949 | struct sched_group *sg = group_head; | |
5950 | int j; | |
5951 | ||
5952 | if (!sg) | |
5953 | return; | |
5954 | next_sg: | |
5955 | for_each_cpu_mask(j, sg->cpumask) { | |
5956 | struct sched_domain *sd; | |
5957 | ||
5958 | sd = &per_cpu(phys_domains, j); | |
5959 | if (j != first_cpu(sd->groups->cpumask)) { | |
5960 | /* | |
5961 | * Only add "power" once for each | |
5962 | * physical package. | |
5963 | */ | |
5964 | continue; | |
5965 | } | |
5966 | ||
5517d86b | 5967 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5968 | } |
5969 | sg = sg->next; | |
5970 | if (sg != group_head) | |
5971 | goto next_sg; | |
5972 | } | |
1da177e4 LT |
5973 | #endif |
5974 | ||
a616058b | 5975 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5976 | /* Free memory allocated for various sched_group structures */ |
5977 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5978 | { | |
a616058b | 5979 | int cpu, i; |
51888ca2 SV |
5980 | |
5981 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5982 | struct sched_group **sched_group_nodes |
5983 | = sched_group_nodes_bycpu[cpu]; | |
5984 | ||
51888ca2 SV |
5985 | if (!sched_group_nodes) |
5986 | continue; | |
5987 | ||
5988 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5989 | cpumask_t nodemask = node_to_cpumask(i); | |
5990 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5991 | ||
5992 | cpus_and(nodemask, nodemask, *cpu_map); | |
5993 | if (cpus_empty(nodemask)) | |
5994 | continue; | |
5995 | ||
5996 | if (sg == NULL) | |
5997 | continue; | |
5998 | sg = sg->next; | |
5999 | next_sg: | |
6000 | oldsg = sg; | |
6001 | sg = sg->next; | |
6002 | kfree(oldsg); | |
6003 | if (oldsg != sched_group_nodes[i]) | |
6004 | goto next_sg; | |
6005 | } | |
6006 | kfree(sched_group_nodes); | |
6007 | sched_group_nodes_bycpu[cpu] = NULL; | |
6008 | } | |
51888ca2 | 6009 | } |
a616058b SS |
6010 | #else |
6011 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6012 | { | |
6013 | } | |
6014 | #endif | |
51888ca2 | 6015 | |
89c4710e SS |
6016 | /* |
6017 | * Initialize sched groups cpu_power. | |
6018 | * | |
6019 | * cpu_power indicates the capacity of sched group, which is used while | |
6020 | * distributing the load between different sched groups in a sched domain. | |
6021 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6022 | * there are asymmetries in the topology. If there are asymmetries, group | |
6023 | * having more cpu_power will pickup more load compared to the group having | |
6024 | * less cpu_power. | |
6025 | * | |
6026 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6027 | * the maximum number of tasks a group can handle in the presence of other idle | |
6028 | * or lightly loaded groups in the same sched domain. | |
6029 | */ | |
6030 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6031 | { | |
6032 | struct sched_domain *child; | |
6033 | struct sched_group *group; | |
6034 | ||
6035 | WARN_ON(!sd || !sd->groups); | |
6036 | ||
6037 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6038 | return; | |
6039 | ||
6040 | child = sd->child; | |
6041 | ||
5517d86b ED |
6042 | sd->groups->__cpu_power = 0; |
6043 | ||
89c4710e SS |
6044 | /* |
6045 | * For perf policy, if the groups in child domain share resources | |
6046 | * (for example cores sharing some portions of the cache hierarchy | |
6047 | * or SMT), then set this domain groups cpu_power such that each group | |
6048 | * can handle only one task, when there are other idle groups in the | |
6049 | * same sched domain. | |
6050 | */ | |
6051 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6052 | (child->flags & | |
6053 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6054 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6055 | return; |
6056 | } | |
6057 | ||
89c4710e SS |
6058 | /* |
6059 | * add cpu_power of each child group to this groups cpu_power | |
6060 | */ | |
6061 | group = child->groups; | |
6062 | do { | |
5517d86b | 6063 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6064 | group = group->next; |
6065 | } while (group != child->groups); | |
6066 | } | |
6067 | ||
1da177e4 | 6068 | /* |
1a20ff27 DG |
6069 | * Build sched domains for a given set of cpus and attach the sched domains |
6070 | * to the individual cpus | |
1da177e4 | 6071 | */ |
51888ca2 | 6072 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6073 | { |
6074 | int i; | |
89c4710e | 6075 | struct sched_domain *sd; |
d1b55138 JH |
6076 | #ifdef CONFIG_NUMA |
6077 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6078 | int sd_allnodes = 0; |
d1b55138 JH |
6079 | |
6080 | /* | |
6081 | * Allocate the per-node list of sched groups | |
6082 | */ | |
51888ca2 | 6083 | sched_group_nodes = kzalloc(sizeof(struct sched_group*)*MAX_NUMNODES, |
d3a5aa98 | 6084 | GFP_KERNEL); |
d1b55138 JH |
6085 | if (!sched_group_nodes) { |
6086 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6087 | return -ENOMEM; |
d1b55138 JH |
6088 | } |
6089 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6090 | #endif | |
1da177e4 LT |
6091 | |
6092 | /* | |
1a20ff27 | 6093 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6094 | */ |
1a20ff27 | 6095 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6096 | struct sched_domain *sd = NULL, *p; |
6097 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6098 | ||
1a20ff27 | 6099 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6100 | |
6101 | #ifdef CONFIG_NUMA | |
d1b55138 | 6102 | if (cpus_weight(*cpu_map) |
9c1cfda2 JH |
6103 | > SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { |
6104 | sd = &per_cpu(allnodes_domains, i); | |
6105 | *sd = SD_ALLNODES_INIT; | |
6106 | sd->span = *cpu_map; | |
6711cab4 | 6107 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6108 | p = sd; |
6711cab4 | 6109 | sd_allnodes = 1; |
9c1cfda2 JH |
6110 | } else |
6111 | p = NULL; | |
6112 | ||
1da177e4 | 6113 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6114 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6115 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6116 | sd->parent = p; | |
1a848870 SS |
6117 | if (p) |
6118 | p->child = sd; | |
9c1cfda2 | 6119 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6120 | #endif |
6121 | ||
6122 | p = sd; | |
6123 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6124 | *sd = SD_CPU_INIT; |
6125 | sd->span = nodemask; | |
6126 | sd->parent = p; | |
1a848870 SS |
6127 | if (p) |
6128 | p->child = sd; | |
6711cab4 | 6129 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6130 | |
1e9f28fa SS |
6131 | #ifdef CONFIG_SCHED_MC |
6132 | p = sd; | |
6133 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6134 | *sd = SD_MC_INIT; |
6135 | sd->span = cpu_coregroup_map(i); | |
6136 | cpus_and(sd->span, sd->span, *cpu_map); | |
6137 | sd->parent = p; | |
1a848870 | 6138 | p->child = sd; |
6711cab4 | 6139 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6140 | #endif |
6141 | ||
1da177e4 LT |
6142 | #ifdef CONFIG_SCHED_SMT |
6143 | p = sd; | |
6144 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6145 | *sd = SD_SIBLING_INIT; |
6146 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6147 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6148 | sd->parent = p; |
1a848870 | 6149 | p->child = sd; |
6711cab4 | 6150 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6151 | #endif |
6152 | } | |
6153 | ||
6154 | #ifdef CONFIG_SCHED_SMT | |
6155 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6156 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6157 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6158 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6159 | if (i != first_cpu(this_sibling_map)) |
6160 | continue; | |
6161 | ||
6711cab4 | 6162 | init_sched_build_groups(this_sibling_map, cpu_map, &cpu_to_cpu_group); |
1da177e4 LT |
6163 | } |
6164 | #endif | |
6165 | ||
1e9f28fa SS |
6166 | #ifdef CONFIG_SCHED_MC |
6167 | /* Set up multi-core groups */ | |
6168 | for_each_cpu_mask(i, *cpu_map) { | |
6169 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6170 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6171 | if (i != first_cpu(this_core_map)) | |
6172 | continue; | |
6711cab4 | 6173 | init_sched_build_groups(this_core_map, cpu_map, &cpu_to_core_group); |
1e9f28fa SS |
6174 | } |
6175 | #endif | |
6176 | ||
6177 | ||
1da177e4 LT |
6178 | /* Set up physical groups */ |
6179 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6180 | cpumask_t nodemask = node_to_cpumask(i); | |
6181 | ||
1a20ff27 | 6182 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6183 | if (cpus_empty(nodemask)) |
6184 | continue; | |
6185 | ||
6711cab4 | 6186 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6187 | } |
6188 | ||
6189 | #ifdef CONFIG_NUMA | |
6190 | /* Set up node groups */ | |
6711cab4 SS |
6191 | if (sd_allnodes) |
6192 | init_sched_build_groups(*cpu_map, cpu_map, &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6193 | |
6194 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6195 | /* Set up node groups */ | |
6196 | struct sched_group *sg, *prev; | |
6197 | cpumask_t nodemask = node_to_cpumask(i); | |
6198 | cpumask_t domainspan; | |
6199 | cpumask_t covered = CPU_MASK_NONE; | |
6200 | int j; | |
6201 | ||
6202 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6203 | if (cpus_empty(nodemask)) { |
6204 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6205 | continue; |
d1b55138 | 6206 | } |
9c1cfda2 JH |
6207 | |
6208 | domainspan = sched_domain_node_span(i); | |
6209 | cpus_and(domainspan, domainspan, *cpu_map); | |
6210 | ||
15f0b676 | 6211 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6212 | if (!sg) { |
6213 | printk(KERN_WARNING "Can not alloc domain group for " | |
6214 | "node %d\n", i); | |
6215 | goto error; | |
6216 | } | |
9c1cfda2 JH |
6217 | sched_group_nodes[i] = sg; |
6218 | for_each_cpu_mask(j, nodemask) { | |
6219 | struct sched_domain *sd; | |
6220 | sd = &per_cpu(node_domains, j); | |
6221 | sd->groups = sg; | |
9c1cfda2 | 6222 | } |
5517d86b | 6223 | sg->__cpu_power = 0; |
9c1cfda2 | 6224 | sg->cpumask = nodemask; |
51888ca2 | 6225 | sg->next = sg; |
9c1cfda2 JH |
6226 | cpus_or(covered, covered, nodemask); |
6227 | prev = sg; | |
6228 | ||
6229 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6230 | cpumask_t tmp, notcovered; | |
6231 | int n = (i + j) % MAX_NUMNODES; | |
6232 | ||
6233 | cpus_complement(notcovered, covered); | |
6234 | cpus_and(tmp, notcovered, *cpu_map); | |
6235 | cpus_and(tmp, tmp, domainspan); | |
6236 | if (cpus_empty(tmp)) | |
6237 | break; | |
6238 | ||
6239 | nodemask = node_to_cpumask(n); | |
6240 | cpus_and(tmp, tmp, nodemask); | |
6241 | if (cpus_empty(tmp)) | |
6242 | continue; | |
6243 | ||
15f0b676 SV |
6244 | sg = kmalloc_node(sizeof(struct sched_group), |
6245 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6246 | if (!sg) { |
6247 | printk(KERN_WARNING | |
6248 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6249 | goto error; |
9c1cfda2 | 6250 | } |
5517d86b | 6251 | sg->__cpu_power = 0; |
9c1cfda2 | 6252 | sg->cpumask = tmp; |
51888ca2 | 6253 | sg->next = prev->next; |
9c1cfda2 JH |
6254 | cpus_or(covered, covered, tmp); |
6255 | prev->next = sg; | |
6256 | prev = sg; | |
6257 | } | |
9c1cfda2 | 6258 | } |
1da177e4 LT |
6259 | #endif |
6260 | ||
6261 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6262 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6263 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6264 | sd = &per_cpu(cpu_domains, i); |
89c4710e | 6265 | init_sched_groups_power(i, sd); |
5c45bf27 | 6266 | } |
1da177e4 | 6267 | #endif |
1e9f28fa | 6268 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6269 | for_each_cpu_mask(i, *cpu_map) { |
1e9f28fa | 6270 | sd = &per_cpu(core_domains, i); |
89c4710e | 6271 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6272 | } |
6273 | #endif | |
1e9f28fa | 6274 | |
5c45bf27 | 6275 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6276 | sd = &per_cpu(phys_domains, i); |
89c4710e | 6277 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6278 | } |
6279 | ||
9c1cfda2 | 6280 | #ifdef CONFIG_NUMA |
08069033 SS |
6281 | for (i = 0; i < MAX_NUMNODES; i++) |
6282 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6283 | |
6711cab4 SS |
6284 | if (sd_allnodes) { |
6285 | struct sched_group *sg; | |
f712c0c7 | 6286 | |
6711cab4 | 6287 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6288 | init_numa_sched_groups_power(sg); |
6289 | } | |
9c1cfda2 JH |
6290 | #endif |
6291 | ||
1da177e4 | 6292 | /* Attach the domains */ |
1a20ff27 | 6293 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6294 | struct sched_domain *sd; |
6295 | #ifdef CONFIG_SCHED_SMT | |
6296 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6297 | #elif defined(CONFIG_SCHED_MC) |
6298 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6299 | #else |
6300 | sd = &per_cpu(phys_domains, i); | |
6301 | #endif | |
6302 | cpu_attach_domain(sd, i); | |
6303 | } | |
51888ca2 SV |
6304 | |
6305 | return 0; | |
6306 | ||
a616058b | 6307 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6308 | error: |
6309 | free_sched_groups(cpu_map); | |
6310 | return -ENOMEM; | |
a616058b | 6311 | #endif |
1da177e4 | 6312 | } |
1a20ff27 DG |
6313 | /* |
6314 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6315 | */ | |
51888ca2 | 6316 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6317 | { |
6318 | cpumask_t cpu_default_map; | |
51888ca2 | 6319 | int err; |
1da177e4 | 6320 | |
1a20ff27 DG |
6321 | /* |
6322 | * Setup mask for cpus without special case scheduling requirements. | |
6323 | * For now this just excludes isolated cpus, but could be used to | |
6324 | * exclude other special cases in the future. | |
6325 | */ | |
6326 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6327 | ||
51888ca2 SV |
6328 | err = build_sched_domains(&cpu_default_map); |
6329 | ||
6330 | return err; | |
1a20ff27 DG |
6331 | } |
6332 | ||
6333 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6334 | { |
51888ca2 | 6335 | free_sched_groups(cpu_map); |
9c1cfda2 | 6336 | } |
1da177e4 | 6337 | |
1a20ff27 DG |
6338 | /* |
6339 | * Detach sched domains from a group of cpus specified in cpu_map | |
6340 | * These cpus will now be attached to the NULL domain | |
6341 | */ | |
858119e1 | 6342 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6343 | { |
6344 | int i; | |
6345 | ||
6346 | for_each_cpu_mask(i, *cpu_map) | |
6347 | cpu_attach_domain(NULL, i); | |
6348 | synchronize_sched(); | |
6349 | arch_destroy_sched_domains(cpu_map); | |
6350 | } | |
6351 | ||
6352 | /* | |
6353 | * Partition sched domains as specified by the cpumasks below. | |
6354 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6355 | * waits for a RCU quiescent period, recalculates sched | |
6356 | * domain information and then attaches them back to the | |
6357 | * correct sched domains | |
6358 | * Call with hotplug lock held | |
6359 | */ | |
51888ca2 | 6360 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6361 | { |
6362 | cpumask_t change_map; | |
51888ca2 | 6363 | int err = 0; |
1a20ff27 DG |
6364 | |
6365 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6366 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6367 | cpus_or(change_map, *partition1, *partition2); | |
6368 | ||
6369 | /* Detach sched domains from all of the affected cpus */ | |
6370 | detach_destroy_domains(&change_map); | |
6371 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6372 | err = build_sched_domains(partition1); |
6373 | if (!err && !cpus_empty(*partition2)) | |
6374 | err = build_sched_domains(partition2); | |
6375 | ||
6376 | return err; | |
1a20ff27 DG |
6377 | } |
6378 | ||
5c45bf27 SS |
6379 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6380 | int arch_reinit_sched_domains(void) | |
6381 | { | |
6382 | int err; | |
6383 | ||
5be9361c | 6384 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6385 | detach_destroy_domains(&cpu_online_map); |
6386 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6387 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6388 | |
6389 | return err; | |
6390 | } | |
6391 | ||
6392 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6393 | { | |
6394 | int ret; | |
6395 | ||
6396 | if (buf[0] != '0' && buf[0] != '1') | |
6397 | return -EINVAL; | |
6398 | ||
6399 | if (smt) | |
6400 | sched_smt_power_savings = (buf[0] == '1'); | |
6401 | else | |
6402 | sched_mc_power_savings = (buf[0] == '1'); | |
6403 | ||
6404 | ret = arch_reinit_sched_domains(); | |
6405 | ||
6406 | return ret ? ret : count; | |
6407 | } | |
6408 | ||
6409 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6410 | { | |
6411 | int err = 0; | |
48f24c4d | 6412 | |
5c45bf27 SS |
6413 | #ifdef CONFIG_SCHED_SMT |
6414 | if (smt_capable()) | |
6415 | err = sysfs_create_file(&cls->kset.kobj, | |
6416 | &attr_sched_smt_power_savings.attr); | |
6417 | #endif | |
6418 | #ifdef CONFIG_SCHED_MC | |
6419 | if (!err && mc_capable()) | |
6420 | err = sysfs_create_file(&cls->kset.kobj, | |
6421 | &attr_sched_mc_power_savings.attr); | |
6422 | #endif | |
6423 | return err; | |
6424 | } | |
6425 | #endif | |
6426 | ||
6427 | #ifdef CONFIG_SCHED_MC | |
6428 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6429 | { | |
6430 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6431 | } | |
48f24c4d IM |
6432 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6433 | const char *buf, size_t count) | |
5c45bf27 SS |
6434 | { |
6435 | return sched_power_savings_store(buf, count, 0); | |
6436 | } | |
6437 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6438 | sched_mc_power_savings_store); | |
6439 | #endif | |
6440 | ||
6441 | #ifdef CONFIG_SCHED_SMT | |
6442 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6443 | { | |
6444 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6445 | } | |
48f24c4d IM |
6446 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6447 | const char *buf, size_t count) | |
5c45bf27 SS |
6448 | { |
6449 | return sched_power_savings_store(buf, count, 1); | |
6450 | } | |
6451 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6452 | sched_smt_power_savings_store); | |
6453 | #endif | |
6454 | ||
1da177e4 LT |
6455 | /* |
6456 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6457 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6458 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6459 | * which will prevent rebalancing while the sched domains are recalculated. |
6460 | */ | |
6461 | static int update_sched_domains(struct notifier_block *nfb, | |
6462 | unsigned long action, void *hcpu) | |
6463 | { | |
1da177e4 LT |
6464 | switch (action) { |
6465 | case CPU_UP_PREPARE: | |
8bb78442 | 6466 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6467 | case CPU_DOWN_PREPARE: |
8bb78442 | 6468 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6469 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6470 | return NOTIFY_OK; |
6471 | ||
6472 | case CPU_UP_CANCELED: | |
8bb78442 | 6473 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6474 | case CPU_DOWN_FAILED: |
8bb78442 | 6475 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6476 | case CPU_ONLINE: |
8bb78442 | 6477 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6478 | case CPU_DEAD: |
8bb78442 | 6479 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6480 | /* |
6481 | * Fall through and re-initialise the domains. | |
6482 | */ | |
6483 | break; | |
6484 | default: | |
6485 | return NOTIFY_DONE; | |
6486 | } | |
6487 | ||
6488 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6489 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6490 | |
6491 | return NOTIFY_OK; | |
6492 | } | |
1da177e4 LT |
6493 | |
6494 | void __init sched_init_smp(void) | |
6495 | { | |
5c1e1767 NP |
6496 | cpumask_t non_isolated_cpus; |
6497 | ||
5be9361c | 6498 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6499 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6500 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6501 | if (cpus_empty(non_isolated_cpus)) |
6502 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6503 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6504 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6505 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6506 | |
6507 | /* Move init over to a non-isolated CPU */ | |
6508 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6509 | BUG(); | |
1da177e4 LT |
6510 | } |
6511 | #else | |
6512 | void __init sched_init_smp(void) | |
6513 | { | |
6514 | } | |
6515 | #endif /* CONFIG_SMP */ | |
6516 | ||
6517 | int in_sched_functions(unsigned long addr) | |
6518 | { | |
6519 | /* Linker adds these: start and end of __sched functions */ | |
6520 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6521 | |
1da177e4 LT |
6522 | return in_lock_functions(addr) || |
6523 | (addr >= (unsigned long)__sched_text_start | |
6524 | && addr < (unsigned long)__sched_text_end); | |
6525 | } | |
6526 | ||
6527 | void __init sched_init(void) | |
6528 | { | |
1da177e4 | 6529 | int i, j, k; |
476f3534 | 6530 | int highest_cpu = 0; |
1da177e4 | 6531 | |
0a945022 | 6532 | for_each_possible_cpu(i) { |
70b97a7f IM |
6533 | struct prio_array *array; |
6534 | struct rq *rq; | |
1da177e4 LT |
6535 | |
6536 | rq = cpu_rq(i); | |
6537 | spin_lock_init(&rq->lock); | |
fcb99371 | 6538 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6539 | rq->nr_running = 0; |
1da177e4 LT |
6540 | rq->active = rq->arrays; |
6541 | rq->expired = rq->arrays + 1; | |
6542 | rq->best_expired_prio = MAX_PRIO; | |
6543 | ||
6544 | #ifdef CONFIG_SMP | |
41c7ce9a | 6545 | rq->sd = NULL; |
7897986b NP |
6546 | for (j = 1; j < 3; j++) |
6547 | rq->cpu_load[j] = 0; | |
1da177e4 LT |
6548 | rq->active_balance = 0; |
6549 | rq->push_cpu = 0; | |
0a2966b4 | 6550 | rq->cpu = i; |
1da177e4 LT |
6551 | rq->migration_thread = NULL; |
6552 | INIT_LIST_HEAD(&rq->migration_queue); | |
6553 | #endif | |
6554 | atomic_set(&rq->nr_iowait, 0); | |
6555 | ||
6556 | for (j = 0; j < 2; j++) { | |
6557 | array = rq->arrays + j; | |
6558 | for (k = 0; k < MAX_PRIO; k++) { | |
6559 | INIT_LIST_HEAD(array->queue + k); | |
6560 | __clear_bit(k, array->bitmap); | |
6561 | } | |
6562 | // delimiter for bitsearch | |
6563 | __set_bit(MAX_PRIO, array->bitmap); | |
6564 | } | |
476f3534 | 6565 | highest_cpu = i; |
1da177e4 LT |
6566 | } |
6567 | ||
2dd73a4f | 6568 | set_load_weight(&init_task); |
b50f60ce | 6569 | |
c9819f45 | 6570 | #ifdef CONFIG_SMP |
476f3534 | 6571 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6572 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6573 | #endif | |
6574 | ||
b50f60ce HC |
6575 | #ifdef CONFIG_RT_MUTEXES |
6576 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6577 | #endif | |
6578 | ||
1da177e4 LT |
6579 | /* |
6580 | * The boot idle thread does lazy MMU switching as well: | |
6581 | */ | |
6582 | atomic_inc(&init_mm.mm_count); | |
6583 | enter_lazy_tlb(&init_mm, current); | |
6584 | ||
6585 | /* | |
6586 | * Make us the idle thread. Technically, schedule() should not be | |
6587 | * called from this thread, however somewhere below it might be, | |
6588 | * but because we are the idle thread, we just pick up running again | |
6589 | * when this runqueue becomes "idle". | |
6590 | */ | |
6591 | init_idle(current, smp_processor_id()); | |
6592 | } | |
6593 | ||
6594 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6595 | void __might_sleep(char *file, int line) | |
6596 | { | |
48f24c4d | 6597 | #ifdef in_atomic |
1da177e4 LT |
6598 | static unsigned long prev_jiffy; /* ratelimiting */ |
6599 | ||
6600 | if ((in_atomic() || irqs_disabled()) && | |
6601 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6602 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6603 | return; | |
6604 | prev_jiffy = jiffies; | |
91368d73 | 6605 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6606 | " context at %s:%d\n", file, line); |
6607 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6608 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6609 | debug_show_held_locks(current); |
3117df04 IM |
6610 | if (irqs_disabled()) |
6611 | print_irqtrace_events(current); | |
1da177e4 LT |
6612 | dump_stack(); |
6613 | } | |
6614 | #endif | |
6615 | } | |
6616 | EXPORT_SYMBOL(__might_sleep); | |
6617 | #endif | |
6618 | ||
6619 | #ifdef CONFIG_MAGIC_SYSRQ | |
6620 | void normalize_rt_tasks(void) | |
6621 | { | |
70b97a7f | 6622 | struct prio_array *array; |
a0f98a1c | 6623 | struct task_struct *g, *p; |
1da177e4 | 6624 | unsigned long flags; |
70b97a7f | 6625 | struct rq *rq; |
1da177e4 LT |
6626 | |
6627 | read_lock_irq(&tasklist_lock); | |
a0f98a1c IM |
6628 | |
6629 | do_each_thread(g, p) { | |
1da177e4 LT |
6630 | if (!rt_task(p)) |
6631 | continue; | |
6632 | ||
b29739f9 IM |
6633 | spin_lock_irqsave(&p->pi_lock, flags); |
6634 | rq = __task_rq_lock(p); | |
1da177e4 LT |
6635 | |
6636 | array = p->array; | |
6637 | if (array) | |
6638 | deactivate_task(p, task_rq(p)); | |
6639 | __setscheduler(p, SCHED_NORMAL, 0); | |
6640 | if (array) { | |
6641 | __activate_task(p, task_rq(p)); | |
6642 | resched_task(rq->curr); | |
6643 | } | |
6644 | ||
b29739f9 IM |
6645 | __task_rq_unlock(rq); |
6646 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6647 | } while_each_thread(g, p); |
6648 | ||
1da177e4 LT |
6649 | read_unlock_irq(&tasklist_lock); |
6650 | } | |
6651 | ||
6652 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6653 | |
6654 | #ifdef CONFIG_IA64 | |
6655 | /* | |
6656 | * These functions are only useful for the IA64 MCA handling. | |
6657 | * | |
6658 | * They can only be called when the whole system has been | |
6659 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6660 | * activity can take place. Using them for anything else would | |
6661 | * be a serious bug, and as a result, they aren't even visible | |
6662 | * under any other configuration. | |
6663 | */ | |
6664 | ||
6665 | /** | |
6666 | * curr_task - return the current task for a given cpu. | |
6667 | * @cpu: the processor in question. | |
6668 | * | |
6669 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6670 | */ | |
36c8b586 | 6671 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6672 | { |
6673 | return cpu_curr(cpu); | |
6674 | } | |
6675 | ||
6676 | /** | |
6677 | * set_curr_task - set the current task for a given cpu. | |
6678 | * @cpu: the processor in question. | |
6679 | * @p: the task pointer to set. | |
6680 | * | |
6681 | * Description: This function must only be used when non-maskable interrupts | |
6682 | * are serviced on a separate stack. It allows the architecture to switch the | |
6683 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6684 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6685 | * and caller must save the original value of the current task (see | |
6686 | * curr_task() above) and restore that value before reenabling interrupts and | |
6687 | * re-starting the system. | |
6688 | * | |
6689 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6690 | */ | |
36c8b586 | 6691 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6692 | { |
6693 | cpu_curr(cpu) = p; | |
6694 | } | |
6695 | ||
6696 | #endif |