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 CK |
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 |
70b97a7f | 1114 | struct migration_req { |
1da177e4 | 1115 | struct list_head list; |
1da177e4 | 1116 | |
36c8b586 | 1117 | struct task_struct *task; |
1da177e4 LT |
1118 | int dest_cpu; |
1119 | ||
1da177e4 | 1120 | struct completion done; |
70b97a7f | 1121 | }; |
1da177e4 LT |
1122 | |
1123 | /* | |
1124 | * The task's runqueue lock must be held. | |
1125 | * Returns true if you have to wait for migration thread. | |
1126 | */ | |
36c8b586 | 1127 | static int |
70b97a7f | 1128 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1129 | { |
70b97a7f | 1130 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1131 | |
1132 | /* | |
1133 | * If the task is not on a runqueue (and not running), then | |
1134 | * it is sufficient to simply update the task's cpu field. | |
1135 | */ | |
1136 | if (!p->array && !task_running(rq, p)) { | |
1137 | set_task_cpu(p, dest_cpu); | |
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | init_completion(&req->done); | |
1da177e4 LT |
1142 | req->task = p; |
1143 | req->dest_cpu = dest_cpu; | |
1144 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1145 | |
1da177e4 LT |
1146 | return 1; |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * wait_task_inactive - wait for a thread to unschedule. | |
1151 | * | |
1152 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1153 | * else this function might spin for a *long* time. This function can't | |
1154 | * be called with interrupts off, or it may introduce deadlock with | |
1155 | * smp_call_function() if an IPI is sent by the same process we are | |
1156 | * waiting to become inactive. | |
1157 | */ | |
36c8b586 | 1158 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1159 | { |
1160 | unsigned long flags; | |
70b97a7f | 1161 | struct rq *rq; |
fa490cfd LT |
1162 | struct prio_array *array; |
1163 | int running; | |
1da177e4 LT |
1164 | |
1165 | repeat: | |
fa490cfd LT |
1166 | /* |
1167 | * We do the initial early heuristics without holding | |
1168 | * any task-queue locks at all. We'll only try to get | |
1169 | * the runqueue lock when things look like they will | |
1170 | * work out! | |
1171 | */ | |
1172 | rq = task_rq(p); | |
1173 | ||
1174 | /* | |
1175 | * If the task is actively running on another CPU | |
1176 | * still, just relax and busy-wait without holding | |
1177 | * any locks. | |
1178 | * | |
1179 | * NOTE! Since we don't hold any locks, it's not | |
1180 | * even sure that "rq" stays as the right runqueue! | |
1181 | * But we don't care, since "task_running()" will | |
1182 | * return false if the runqueue has changed and p | |
1183 | * is actually now running somewhere else! | |
1184 | */ | |
1185 | while (task_running(rq, p)) | |
1186 | cpu_relax(); | |
1187 | ||
1188 | /* | |
1189 | * Ok, time to look more closely! We need the rq | |
1190 | * lock now, to be *sure*. If we're wrong, we'll | |
1191 | * just go back and repeat. | |
1192 | */ | |
1da177e4 | 1193 | rq = task_rq_lock(p, &flags); |
fa490cfd LT |
1194 | running = task_running(rq, p); |
1195 | array = p->array; | |
1196 | task_rq_unlock(rq, &flags); | |
1197 | ||
1198 | /* | |
1199 | * Was it really running after all now that we | |
1200 | * checked with the proper locks actually held? | |
1201 | * | |
1202 | * Oops. Go back and try again.. | |
1203 | */ | |
1204 | if (unlikely(running)) { | |
1da177e4 | 1205 | cpu_relax(); |
1da177e4 LT |
1206 | goto repeat; |
1207 | } | |
fa490cfd LT |
1208 | |
1209 | /* | |
1210 | * It's not enough that it's not actively running, | |
1211 | * it must be off the runqueue _entirely_, and not | |
1212 | * preempted! | |
1213 | * | |
1214 | * So if it wa still runnable (but just not actively | |
1215 | * running right now), it's preempted, and we should | |
1216 | * yield - it could be a while. | |
1217 | */ | |
1218 | if (unlikely(array)) { | |
1219 | yield(); | |
1220 | goto repeat; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * Ahh, all good. It wasn't running, and it wasn't | |
1225 | * runnable, which means that it will never become | |
1226 | * running in the future either. We're all done! | |
1227 | */ | |
1da177e4 LT |
1228 | } |
1229 | ||
1230 | /*** | |
1231 | * kick_process - kick a running thread to enter/exit the kernel | |
1232 | * @p: the to-be-kicked thread | |
1233 | * | |
1234 | * Cause a process which is running on another CPU to enter | |
1235 | * kernel-mode, without any delay. (to get signals handled.) | |
1236 | * | |
1237 | * NOTE: this function doesnt have to take the runqueue lock, | |
1238 | * because all it wants to ensure is that the remote task enters | |
1239 | * the kernel. If the IPI races and the task has been migrated | |
1240 | * to another CPU then no harm is done and the purpose has been | |
1241 | * achieved as well. | |
1242 | */ | |
36c8b586 | 1243 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1244 | { |
1245 | int cpu; | |
1246 | ||
1247 | preempt_disable(); | |
1248 | cpu = task_cpu(p); | |
1249 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1250 | smp_send_reschedule(cpu); | |
1251 | preempt_enable(); | |
1252 | } | |
1253 | ||
1254 | /* | |
2dd73a4f PW |
1255 | * Return a low guess at the load of a migration-source cpu weighted |
1256 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1257 | * |
1258 | * We want to under-estimate the load of migration sources, to | |
1259 | * balance conservatively. | |
1260 | */ | |
a2000572 | 1261 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1262 | { |
70b97a7f | 1263 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1264 | |
3b0bd9bc | 1265 | if (type == 0) |
2dd73a4f | 1266 | return rq->raw_weighted_load; |
b910472d | 1267 | |
2dd73a4f | 1268 | return min(rq->cpu_load[type-1], rq->raw_weighted_load); |
1da177e4 LT |
1269 | } |
1270 | ||
1271 | /* | |
2dd73a4f PW |
1272 | * Return a high guess at the load of a migration-target cpu weighted |
1273 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1274 | */ |
a2000572 | 1275 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1276 | { |
70b97a7f | 1277 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1278 | |
7897986b | 1279 | if (type == 0) |
2dd73a4f | 1280 | return rq->raw_weighted_load; |
3b0bd9bc | 1281 | |
2dd73a4f PW |
1282 | return max(rq->cpu_load[type-1], rq->raw_weighted_load); |
1283 | } | |
1284 | ||
1285 | /* | |
1286 | * Return the average load per task on the cpu's run queue | |
1287 | */ | |
1288 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1289 | { | |
70b97a7f | 1290 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f PW |
1291 | unsigned long n = rq->nr_running; |
1292 | ||
48f24c4d | 1293 | return n ? rq->raw_weighted_load / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1294 | } |
1295 | ||
147cbb4b NP |
1296 | /* |
1297 | * find_idlest_group finds and returns the least busy CPU group within the | |
1298 | * domain. | |
1299 | */ | |
1300 | static struct sched_group * | |
1301 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1302 | { | |
1303 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1304 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1305 | int load_idx = sd->forkexec_idx; | |
1306 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1307 | ||
1308 | do { | |
1309 | unsigned long load, avg_load; | |
1310 | int local_group; | |
1311 | int i; | |
1312 | ||
da5a5522 BD |
1313 | /* Skip over this group if it has no CPUs allowed */ |
1314 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1315 | goto nextgroup; | |
1316 | ||
147cbb4b | 1317 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1318 | |
1319 | /* Tally up the load of all CPUs in the group */ | |
1320 | avg_load = 0; | |
1321 | ||
1322 | for_each_cpu_mask(i, group->cpumask) { | |
1323 | /* Bias balancing toward cpus of our domain */ | |
1324 | if (local_group) | |
1325 | load = source_load(i, load_idx); | |
1326 | else | |
1327 | load = target_load(i, load_idx); | |
1328 | ||
1329 | avg_load += load; | |
1330 | } | |
1331 | ||
1332 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1333 | avg_load = sg_div_cpu_power(group, |
1334 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1335 | |
1336 | if (local_group) { | |
1337 | this_load = avg_load; | |
1338 | this = group; | |
1339 | } else if (avg_load < min_load) { | |
1340 | min_load = avg_load; | |
1341 | idlest = group; | |
1342 | } | |
da5a5522 | 1343 | nextgroup: |
147cbb4b NP |
1344 | group = group->next; |
1345 | } while (group != sd->groups); | |
1346 | ||
1347 | if (!idlest || 100*this_load < imbalance*min_load) | |
1348 | return NULL; | |
1349 | return idlest; | |
1350 | } | |
1351 | ||
1352 | /* | |
0feaece9 | 1353 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1354 | */ |
95cdf3b7 IM |
1355 | static int |
1356 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1357 | { |
da5a5522 | 1358 | cpumask_t tmp; |
147cbb4b NP |
1359 | unsigned long load, min_load = ULONG_MAX; |
1360 | int idlest = -1; | |
1361 | int i; | |
1362 | ||
da5a5522 BD |
1363 | /* Traverse only the allowed CPUs */ |
1364 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1365 | ||
1366 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1367 | load = weighted_cpuload(i); |
147cbb4b NP |
1368 | |
1369 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1370 | min_load = load; | |
1371 | idlest = i; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | return idlest; | |
1376 | } | |
1377 | ||
476d139c NP |
1378 | /* |
1379 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1380 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1381 | * SD_BALANCE_EXEC. | |
1382 | * | |
1383 | * Balance, ie. select the least loaded group. | |
1384 | * | |
1385 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1386 | * | |
1387 | * preempt must be disabled. | |
1388 | */ | |
1389 | static int sched_balance_self(int cpu, int flag) | |
1390 | { | |
1391 | struct task_struct *t = current; | |
1392 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1393 | |
c96d145e | 1394 | for_each_domain(cpu, tmp) { |
5c45bf27 SS |
1395 | /* |
1396 | * If power savings logic is enabled for a domain, stop there. | |
1397 | */ | |
1398 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) | |
1399 | break; | |
476d139c NP |
1400 | if (tmp->flags & flag) |
1401 | sd = tmp; | |
c96d145e | 1402 | } |
476d139c NP |
1403 | |
1404 | while (sd) { | |
1405 | cpumask_t span; | |
1406 | struct sched_group *group; | |
1a848870 SS |
1407 | int new_cpu, weight; |
1408 | ||
1409 | if (!(sd->flags & flag)) { | |
1410 | sd = sd->child; | |
1411 | continue; | |
1412 | } | |
476d139c NP |
1413 | |
1414 | span = sd->span; | |
1415 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1416 | if (!group) { |
1417 | sd = sd->child; | |
1418 | continue; | |
1419 | } | |
476d139c | 1420 | |
da5a5522 | 1421 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1422 | if (new_cpu == -1 || new_cpu == cpu) { |
1423 | /* Now try balancing at a lower domain level of cpu */ | |
1424 | sd = sd->child; | |
1425 | continue; | |
1426 | } | |
476d139c | 1427 | |
1a848870 | 1428 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1429 | cpu = new_cpu; |
476d139c NP |
1430 | sd = NULL; |
1431 | weight = cpus_weight(span); | |
1432 | for_each_domain(cpu, tmp) { | |
1433 | if (weight <= cpus_weight(tmp->span)) | |
1434 | break; | |
1435 | if (tmp->flags & flag) | |
1436 | sd = tmp; | |
1437 | } | |
1438 | /* while loop will break here if sd == NULL */ | |
1439 | } | |
1440 | ||
1441 | return cpu; | |
1442 | } | |
1443 | ||
1444 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1445 | |
1446 | /* | |
1447 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1448 | * not idle and an idle cpu is available. The span of cpus to | |
1449 | * search starts with cpus closest then further out as needed, | |
1450 | * so we always favor a closer, idle cpu. | |
1451 | * | |
1452 | * Returns the CPU we should wake onto. | |
1453 | */ | |
1454 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1455 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1456 | { |
1457 | cpumask_t tmp; | |
1458 | struct sched_domain *sd; | |
1459 | int i; | |
1460 | ||
4953198b SS |
1461 | /* |
1462 | * If it is idle, then it is the best cpu to run this task. | |
1463 | * | |
1464 | * This cpu is also the best, if it has more than one task already. | |
1465 | * Siblings must be also busy(in most cases) as they didn't already | |
1466 | * pickup the extra load from this cpu and hence we need not check | |
1467 | * sibling runqueue info. This will avoid the checks and cache miss | |
1468 | * penalities associated with that. | |
1469 | */ | |
1470 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1471 | return cpu; |
1472 | ||
1473 | for_each_domain(cpu, sd) { | |
1474 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1475 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1476 | for_each_cpu_mask(i, tmp) { |
1477 | if (idle_cpu(i)) | |
1478 | return i; | |
1479 | } | |
1480 | } | |
e0f364f4 NP |
1481 | else |
1482 | break; | |
1da177e4 LT |
1483 | } |
1484 | return cpu; | |
1485 | } | |
1486 | #else | |
36c8b586 | 1487 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1488 | { |
1489 | return cpu; | |
1490 | } | |
1491 | #endif | |
1492 | ||
1493 | /*** | |
1494 | * try_to_wake_up - wake up a thread | |
1495 | * @p: the to-be-woken-up thread | |
1496 | * @state: the mask of task states that can be woken | |
1497 | * @sync: do a synchronous wakeup? | |
1498 | * | |
1499 | * Put it on the run-queue if it's not already there. The "current" | |
1500 | * thread is always on the run-queue (except when the actual | |
1501 | * re-schedule is in progress), and as such you're allowed to do | |
1502 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1503 | * runnable without the overhead of this. | |
1504 | * | |
1505 | * returns failure only if the task is already active. | |
1506 | */ | |
36c8b586 | 1507 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1508 | { |
1509 | int cpu, this_cpu, success = 0; | |
1510 | unsigned long flags; | |
1511 | long old_state; | |
70b97a7f | 1512 | struct rq *rq; |
1da177e4 | 1513 | #ifdef CONFIG_SMP |
7897986b | 1514 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1515 | unsigned long load, this_load; |
1da177e4 LT |
1516 | int new_cpu; |
1517 | #endif | |
1518 | ||
1519 | rq = task_rq_lock(p, &flags); | |
1520 | old_state = p->state; | |
1521 | if (!(old_state & state)) | |
1522 | goto out; | |
1523 | ||
1524 | if (p->array) | |
1525 | goto out_running; | |
1526 | ||
1527 | cpu = task_cpu(p); | |
1528 | this_cpu = smp_processor_id(); | |
1529 | ||
1530 | #ifdef CONFIG_SMP | |
1531 | if (unlikely(task_running(rq, p))) | |
1532 | goto out_activate; | |
1533 | ||
7897986b NP |
1534 | new_cpu = cpu; |
1535 | ||
1da177e4 LT |
1536 | schedstat_inc(rq, ttwu_cnt); |
1537 | if (cpu == this_cpu) { | |
1538 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1539 | goto out_set_cpu; |
1540 | } | |
1541 | ||
1542 | for_each_domain(this_cpu, sd) { | |
1543 | if (cpu_isset(cpu, sd->span)) { | |
1544 | schedstat_inc(sd, ttwu_wake_remote); | |
1545 | this_sd = sd; | |
1546 | break; | |
1da177e4 LT |
1547 | } |
1548 | } | |
1da177e4 | 1549 | |
7897986b | 1550 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1551 | goto out_set_cpu; |
1552 | ||
1da177e4 | 1553 | /* |
7897986b | 1554 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1555 | */ |
7897986b NP |
1556 | if (this_sd) { |
1557 | int idx = this_sd->wake_idx; | |
1558 | unsigned int imbalance; | |
1da177e4 | 1559 | |
a3f21bce NP |
1560 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1561 | ||
7897986b NP |
1562 | load = source_load(cpu, idx); |
1563 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1564 | |
7897986b NP |
1565 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1566 | ||
a3f21bce NP |
1567 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1568 | unsigned long tl = this_load; | |
33859f7f MOS |
1569 | unsigned long tl_per_task; |
1570 | ||
1571 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1572 | |
1da177e4 | 1573 | /* |
a3f21bce NP |
1574 | * If sync wakeup then subtract the (maximum possible) |
1575 | * effect of the currently running task from the load | |
1576 | * of the current CPU: | |
1da177e4 | 1577 | */ |
a3f21bce | 1578 | if (sync) |
2dd73a4f | 1579 | tl -= current->load_weight; |
a3f21bce NP |
1580 | |
1581 | if ((tl <= load && | |
2dd73a4f PW |
1582 | tl + target_load(cpu, idx) <= tl_per_task) || |
1583 | 100*(tl + p->load_weight) <= imbalance*load) { | |
a3f21bce NP |
1584 | /* |
1585 | * This domain has SD_WAKE_AFFINE and | |
1586 | * p is cache cold in this domain, and | |
1587 | * there is no bad imbalance. | |
1588 | */ | |
1589 | schedstat_inc(this_sd, ttwu_move_affine); | |
1590 | goto out_set_cpu; | |
1591 | } | |
1592 | } | |
1593 | ||
1594 | /* | |
1595 | * Start passive balancing when half the imbalance_pct | |
1596 | * limit is reached. | |
1597 | */ | |
1598 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1599 | if (imbalance*this_load <= 100*load) { | |
1600 | schedstat_inc(this_sd, ttwu_move_balance); | |
1601 | goto out_set_cpu; | |
1602 | } | |
1da177e4 LT |
1603 | } |
1604 | } | |
1605 | ||
1606 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1607 | out_set_cpu: | |
1608 | new_cpu = wake_idle(new_cpu, p); | |
1609 | if (new_cpu != cpu) { | |
1610 | set_task_cpu(p, new_cpu); | |
1611 | task_rq_unlock(rq, &flags); | |
1612 | /* might preempt at this point */ | |
1613 | rq = task_rq_lock(p, &flags); | |
1614 | old_state = p->state; | |
1615 | if (!(old_state & state)) | |
1616 | goto out; | |
1617 | if (p->array) | |
1618 | goto out_running; | |
1619 | ||
1620 | this_cpu = smp_processor_id(); | |
1621 | cpu = task_cpu(p); | |
1622 | } | |
1623 | ||
1624 | out_activate: | |
1625 | #endif /* CONFIG_SMP */ | |
1626 | if (old_state == TASK_UNINTERRUPTIBLE) { | |
1627 | rq->nr_uninterruptible--; | |
1628 | /* | |
1629 | * Tasks on involuntary sleep don't earn | |
1630 | * sleep_avg beyond just interactive state. | |
1631 | */ | |
3dee386e | 1632 | p->sleep_type = SLEEP_NONINTERACTIVE; |
e7c38cb4 | 1633 | } else |
1da177e4 | 1634 | |
d79fc0fc IM |
1635 | /* |
1636 | * Tasks that have marked their sleep as noninteractive get | |
e7c38cb4 CK |
1637 | * woken up with their sleep average not weighted in an |
1638 | * interactive way. | |
d79fc0fc | 1639 | */ |
e7c38cb4 CK |
1640 | if (old_state & TASK_NONINTERACTIVE) |
1641 | p->sleep_type = SLEEP_NONINTERACTIVE; | |
1642 | ||
1643 | ||
1644 | activate_task(p, rq, cpu == this_cpu); | |
1da177e4 LT |
1645 | /* |
1646 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1647 | * has indicated that it will leave the CPU in short order) | |
1648 | * don't trigger a preemption, if the woken up task will run on | |
1649 | * this cpu. (in this case the 'I will reschedule' promise of | |
1650 | * the waker guarantees that the freshly woken up task is going | |
1651 | * to be considered on this CPU.) | |
1652 | */ | |
1da177e4 LT |
1653 | if (!sync || cpu != this_cpu) { |
1654 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1655 | resched_task(rq->curr); | |
1656 | } | |
1657 | success = 1; | |
1658 | ||
1659 | out_running: | |
1660 | p->state = TASK_RUNNING; | |
1661 | out: | |
1662 | task_rq_unlock(rq, &flags); | |
1663 | ||
1664 | return success; | |
1665 | } | |
1666 | ||
36c8b586 | 1667 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1668 | { |
1669 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1670 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1671 | } | |
1da177e4 LT |
1672 | EXPORT_SYMBOL(wake_up_process); |
1673 | ||
36c8b586 | 1674 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1675 | { |
1676 | return try_to_wake_up(p, state, 0); | |
1677 | } | |
1678 | ||
bc947631 | 1679 | static void task_running_tick(struct rq *rq, struct task_struct *p); |
1da177e4 LT |
1680 | /* |
1681 | * Perform scheduler related setup for a newly forked process p. | |
1682 | * p is forked by current. | |
1683 | */ | |
36c8b586 | 1684 | void fastcall sched_fork(struct task_struct *p, int clone_flags) |
1da177e4 | 1685 | { |
476d139c NP |
1686 | int cpu = get_cpu(); |
1687 | ||
1688 | #ifdef CONFIG_SMP | |
1689 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1690 | #endif | |
1691 | set_task_cpu(p, cpu); | |
1692 | ||
1da177e4 LT |
1693 | /* |
1694 | * We mark the process as running here, but have not actually | |
1695 | * inserted it onto the runqueue yet. This guarantees that | |
1696 | * nobody will actually run it, and a signal or other external | |
1697 | * event cannot wake it up and insert it on the runqueue either. | |
1698 | */ | |
1699 | p->state = TASK_RUNNING; | |
b29739f9 IM |
1700 | |
1701 | /* | |
1702 | * Make sure we do not leak PI boosting priority to the child: | |
1703 | */ | |
1704 | p->prio = current->normal_prio; | |
1705 | ||
1da177e4 LT |
1706 | INIT_LIST_HEAD(&p->run_list); |
1707 | p->array = NULL; | |
52f17b6c CS |
1708 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
1709 | if (unlikely(sched_info_on())) | |
1710 | memset(&p->sched_info, 0, sizeof(p->sched_info)); | |
1da177e4 | 1711 | #endif |
d6077cb8 | 1712 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1713 | p->oncpu = 0; |
1714 | #endif | |
1da177e4 | 1715 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1716 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1717 | task_thread_info(p)->preempt_count = 1; |
1da177e4 LT |
1718 | #endif |
1719 | /* | |
1720 | * Share the timeslice between parent and child, thus the | |
1721 | * total amount of pending timeslices in the system doesn't change, | |
1722 | * resulting in more scheduling fairness. | |
1723 | */ | |
1724 | local_irq_disable(); | |
1725 | p->time_slice = (current->time_slice + 1) >> 1; | |
1726 | /* | |
1727 | * The remainder of the first timeslice might be recovered by | |
1728 | * the parent if the child exits early enough. | |
1729 | */ | |
1730 | p->first_time_slice = 1; | |
1731 | current->time_slice >>= 1; | |
1732 | p->timestamp = sched_clock(); | |
1733 | if (unlikely(!current->time_slice)) { | |
1734 | /* | |
1735 | * This case is rare, it happens when the parent has only | |
1736 | * a single jiffy left from its timeslice. Taking the | |
1737 | * runqueue lock is not a problem. | |
1738 | */ | |
1739 | current->time_slice = 1; | |
bc947631 | 1740 | task_running_tick(cpu_rq(cpu), current); |
476d139c NP |
1741 | } |
1742 | local_irq_enable(); | |
1743 | put_cpu(); | |
1da177e4 LT |
1744 | } |
1745 | ||
1746 | /* | |
1747 | * wake_up_new_task - wake up a newly created task for the first time. | |
1748 | * | |
1749 | * This function will do some initial scheduler statistics housekeeping | |
1750 | * that must be done for every newly created context, then puts the task | |
1751 | * on the runqueue and wakes it. | |
1752 | */ | |
36c8b586 | 1753 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 | 1754 | { |
70b97a7f | 1755 | struct rq *rq, *this_rq; |
1da177e4 LT |
1756 | unsigned long flags; |
1757 | int this_cpu, cpu; | |
1da177e4 LT |
1758 | |
1759 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1760 | BUG_ON(p->state != TASK_RUNNING); |
1da177e4 | 1761 | this_cpu = smp_processor_id(); |
147cbb4b | 1762 | cpu = task_cpu(p); |
1da177e4 | 1763 | |
1da177e4 LT |
1764 | /* |
1765 | * We decrease the sleep average of forking parents | |
1766 | * and children as well, to keep max-interactive tasks | |
1767 | * from forking tasks that are max-interactive. The parent | |
1768 | * (current) is done further down, under its lock. | |
1769 | */ | |
1770 | p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) * | |
1771 | CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1772 | ||
1773 | p->prio = effective_prio(p); | |
1774 | ||
1775 | if (likely(cpu == this_cpu)) { | |
1776 | if (!(clone_flags & CLONE_VM)) { | |
1777 | /* | |
1778 | * The VM isn't cloned, so we're in a good position to | |
1779 | * do child-runs-first in anticipation of an exec. This | |
1780 | * usually avoids a lot of COW overhead. | |
1781 | */ | |
1782 | if (unlikely(!current->array)) | |
1783 | __activate_task(p, rq); | |
1784 | else { | |
1785 | p->prio = current->prio; | |
b29739f9 | 1786 | p->normal_prio = current->normal_prio; |
1da177e4 LT |
1787 | list_add_tail(&p->run_list, ¤t->run_list); |
1788 | p->array = current->array; | |
1789 | p->array->nr_active++; | |
2dd73a4f | 1790 | inc_nr_running(p, rq); |
1da177e4 LT |
1791 | } |
1792 | set_need_resched(); | |
1793 | } else | |
1794 | /* Run child last */ | |
1795 | __activate_task(p, rq); | |
1796 | /* | |
1797 | * We skip the following code due to cpu == this_cpu | |
1798 | * | |
1799 | * task_rq_unlock(rq, &flags); | |
1800 | * this_rq = task_rq_lock(current, &flags); | |
1801 | */ | |
1802 | this_rq = rq; | |
1803 | } else { | |
1804 | this_rq = cpu_rq(this_cpu); | |
1805 | ||
1806 | /* | |
1807 | * Not the local CPU - must adjust timestamp. This should | |
1808 | * get optimised away in the !CONFIG_SMP case. | |
1809 | */ | |
b18ec803 MG |
1810 | p->timestamp = (p->timestamp - this_rq->most_recent_timestamp) |
1811 | + rq->most_recent_timestamp; | |
1da177e4 LT |
1812 | __activate_task(p, rq); |
1813 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1814 | resched_task(rq->curr); | |
1815 | ||
1816 | /* | |
1817 | * Parent and child are on different CPUs, now get the | |
1818 | * parent runqueue to update the parent's ->sleep_avg: | |
1819 | */ | |
1820 | task_rq_unlock(rq, &flags); | |
1821 | this_rq = task_rq_lock(current, &flags); | |
1822 | } | |
1823 | current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) * | |
1824 | PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1825 | task_rq_unlock(this_rq, &flags); | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * Potentially available exiting-child timeslices are | |
1830 | * retrieved here - this way the parent does not get | |
1831 | * penalized for creating too many threads. | |
1832 | * | |
1833 | * (this cannot be used to 'generate' timeslices | |
1834 | * artificially, because any timeslice recovered here | |
1835 | * was given away by the parent in the first place.) | |
1836 | */ | |
36c8b586 | 1837 | void fastcall sched_exit(struct task_struct *p) |
1da177e4 LT |
1838 | { |
1839 | unsigned long flags; | |
70b97a7f | 1840 | struct rq *rq; |
1da177e4 LT |
1841 | |
1842 | /* | |
1843 | * If the child was a (relative-) CPU hog then decrease | |
1844 | * the sleep_avg of the parent as well. | |
1845 | */ | |
1846 | rq = task_rq_lock(p->parent, &flags); | |
889dfafe | 1847 | if (p->first_time_slice && task_cpu(p) == task_cpu(p->parent)) { |
1da177e4 LT |
1848 | p->parent->time_slice += p->time_slice; |
1849 | if (unlikely(p->parent->time_slice > task_timeslice(p))) | |
1850 | p->parent->time_slice = task_timeslice(p); | |
1851 | } | |
1852 | if (p->sleep_avg < p->parent->sleep_avg) | |
1853 | p->parent->sleep_avg = p->parent->sleep_avg / | |
1854 | (EXIT_WEIGHT + 1) * EXIT_WEIGHT + p->sleep_avg / | |
1855 | (EXIT_WEIGHT + 1); | |
1856 | task_rq_unlock(rq, &flags); | |
1857 | } | |
1858 | ||
4866cde0 NP |
1859 | /** |
1860 | * prepare_task_switch - prepare to switch tasks | |
1861 | * @rq: the runqueue preparing to switch | |
1862 | * @next: the task we are going to switch to. | |
1863 | * | |
1864 | * This is called with the rq lock held and interrupts off. It must | |
1865 | * be paired with a subsequent finish_task_switch after the context | |
1866 | * switch. | |
1867 | * | |
1868 | * prepare_task_switch sets up locking and calls architecture specific | |
1869 | * hooks. | |
1870 | */ | |
70b97a7f | 1871 | static inline void prepare_task_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
1872 | { |
1873 | prepare_lock_switch(rq, next); | |
1874 | prepare_arch_switch(next); | |
1875 | } | |
1876 | ||
1da177e4 LT |
1877 | /** |
1878 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1879 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1880 | * @prev: the thread we just switched away from. |
1881 | * | |
4866cde0 NP |
1882 | * finish_task_switch must be called after the context switch, paired |
1883 | * with a prepare_task_switch call before the context switch. | |
1884 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1885 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1886 | * |
1887 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1888 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1889 | * with the lock held can cause deadlocks; see schedule() for | |
1890 | * details.) | |
1891 | */ | |
70b97a7f | 1892 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1893 | __releases(rq->lock) |
1894 | { | |
1da177e4 | 1895 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1896 | long prev_state; |
1da177e4 LT |
1897 | |
1898 | rq->prev_mm = NULL; | |
1899 | ||
1900 | /* | |
1901 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1902 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1903 | * schedule one last time. The schedule call will never return, and |
1904 | * the scheduled task must drop that reference. | |
c394cc9f | 1905 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1906 | * still held, otherwise prev could be scheduled on another cpu, die |
1907 | * there before we look at prev->state, and then the reference would | |
1908 | * be dropped twice. | |
1909 | * Manfred Spraul <manfred@colorfullife.com> | |
1910 | */ | |
55a101f8 | 1911 | prev_state = prev->state; |
4866cde0 NP |
1912 | finish_arch_switch(prev); |
1913 | finish_lock_switch(rq, prev); | |
1da177e4 LT |
1914 | if (mm) |
1915 | mmdrop(mm); | |
c394cc9f | 1916 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1917 | /* |
1918 | * Remove function-return probe instances associated with this | |
1919 | * task and put them back on the free list. | |
1920 | */ | |
1921 | kprobe_flush_task(prev); | |
1da177e4 | 1922 | put_task_struct(prev); |
c6fd91f0 | 1923 | } |
1da177e4 LT |
1924 | } |
1925 | ||
1926 | /** | |
1927 | * schedule_tail - first thing a freshly forked thread must call. | |
1928 | * @prev: the thread we just switched away from. | |
1929 | */ | |
36c8b586 | 1930 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1931 | __releases(rq->lock) |
1932 | { | |
70b97a7f IM |
1933 | struct rq *rq = this_rq(); |
1934 | ||
4866cde0 NP |
1935 | finish_task_switch(rq, prev); |
1936 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1937 | /* In this case, finish_task_switch does not reenable preemption */ | |
1938 | preempt_enable(); | |
1939 | #endif | |
1da177e4 LT |
1940 | if (current->set_child_tid) |
1941 | put_user(current->pid, current->set_child_tid); | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * context_switch - switch to the new MM and the new | |
1946 | * thread's register state. | |
1947 | */ | |
36c8b586 | 1948 | static inline struct task_struct * |
70b97a7f | 1949 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1950 | struct task_struct *next) |
1da177e4 LT |
1951 | { |
1952 | struct mm_struct *mm = next->mm; | |
1953 | struct mm_struct *oldmm = prev->active_mm; | |
1954 | ||
9226d125 ZA |
1955 | /* |
1956 | * For paravirt, this is coupled with an exit in switch_to to | |
1957 | * combine the page table reload and the switch backend into | |
1958 | * one hypercall. | |
1959 | */ | |
1960 | arch_enter_lazy_cpu_mode(); | |
1961 | ||
beed33a8 | 1962 | if (!mm) { |
1da177e4 LT |
1963 | next->active_mm = oldmm; |
1964 | atomic_inc(&oldmm->mm_count); | |
1965 | enter_lazy_tlb(oldmm, next); | |
1966 | } else | |
1967 | switch_mm(oldmm, mm, next); | |
1968 | ||
beed33a8 | 1969 | if (!prev->mm) { |
1da177e4 LT |
1970 | prev->active_mm = NULL; |
1971 | WARN_ON(rq->prev_mm); | |
1972 | rq->prev_mm = oldmm; | |
1973 | } | |
3a5f5e48 IM |
1974 | /* |
1975 | * Since the runqueue lock will be released by the next | |
1976 | * task (which is an invalid locking op but in the case | |
1977 | * of the scheduler it's an obvious special-case), so we | |
1978 | * do an early lockdep release here: | |
1979 | */ | |
1980 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1981 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1982 | #endif |
1da177e4 LT |
1983 | |
1984 | /* Here we just switch the register state and the stack. */ | |
1985 | switch_to(prev, next, prev); | |
1986 | ||
1987 | return prev; | |
1988 | } | |
1989 | ||
1990 | /* | |
1991 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1992 | * | |
1993 | * externally visible scheduler statistics: current number of runnable | |
1994 | * threads, current number of uninterruptible-sleeping threads, total | |
1995 | * number of context switches performed since bootup. | |
1996 | */ | |
1997 | unsigned long nr_running(void) | |
1998 | { | |
1999 | unsigned long i, sum = 0; | |
2000 | ||
2001 | for_each_online_cpu(i) | |
2002 | sum += cpu_rq(i)->nr_running; | |
2003 | ||
2004 | return sum; | |
2005 | } | |
2006 | ||
2007 | unsigned long nr_uninterruptible(void) | |
2008 | { | |
2009 | unsigned long i, sum = 0; | |
2010 | ||
0a945022 | 2011 | for_each_possible_cpu(i) |
1da177e4 LT |
2012 | sum += cpu_rq(i)->nr_uninterruptible; |
2013 | ||
2014 | /* | |
2015 | * Since we read the counters lockless, it might be slightly | |
2016 | * inaccurate. Do not allow it to go below zero though: | |
2017 | */ | |
2018 | if (unlikely((long)sum < 0)) | |
2019 | sum = 0; | |
2020 | ||
2021 | return sum; | |
2022 | } | |
2023 | ||
2024 | unsigned long long nr_context_switches(void) | |
2025 | { | |
cc94abfc SR |
2026 | int i; |
2027 | unsigned long long sum = 0; | |
1da177e4 | 2028 | |
0a945022 | 2029 | for_each_possible_cpu(i) |
1da177e4 LT |
2030 | sum += cpu_rq(i)->nr_switches; |
2031 | ||
2032 | return sum; | |
2033 | } | |
2034 | ||
2035 | unsigned long nr_iowait(void) | |
2036 | { | |
2037 | unsigned long i, sum = 0; | |
2038 | ||
0a945022 | 2039 | for_each_possible_cpu(i) |
1da177e4 LT |
2040 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2041 | ||
2042 | return sum; | |
2043 | } | |
2044 | ||
db1b1fef JS |
2045 | unsigned long nr_active(void) |
2046 | { | |
2047 | unsigned long i, running = 0, uninterruptible = 0; | |
2048 | ||
2049 | for_each_online_cpu(i) { | |
2050 | running += cpu_rq(i)->nr_running; | |
2051 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2052 | } | |
2053 | ||
2054 | if (unlikely((long)uninterruptible < 0)) | |
2055 | uninterruptible = 0; | |
2056 | ||
2057 | return running + uninterruptible; | |
2058 | } | |
2059 | ||
1da177e4 LT |
2060 | #ifdef CONFIG_SMP |
2061 | ||
48f24c4d IM |
2062 | /* |
2063 | * Is this task likely cache-hot: | |
2064 | */ | |
2065 | static inline int | |
2066 | task_hot(struct task_struct *p, unsigned long long now, struct sched_domain *sd) | |
2067 | { | |
2068 | return (long long)(now - p->last_ran) < (long long)sd->cache_hot_time; | |
2069 | } | |
2070 | ||
1da177e4 LT |
2071 | /* |
2072 | * double_rq_lock - safely lock two runqueues | |
2073 | * | |
2074 | * Note this does not disable interrupts like task_rq_lock, | |
2075 | * you need to do so manually before calling. | |
2076 | */ | |
70b97a7f | 2077 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2078 | __acquires(rq1->lock) |
2079 | __acquires(rq2->lock) | |
2080 | { | |
054b9108 | 2081 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2082 | if (rq1 == rq2) { |
2083 | spin_lock(&rq1->lock); | |
2084 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2085 | } else { | |
c96d145e | 2086 | if (rq1 < rq2) { |
1da177e4 LT |
2087 | spin_lock(&rq1->lock); |
2088 | spin_lock(&rq2->lock); | |
2089 | } else { | |
2090 | spin_lock(&rq2->lock); | |
2091 | spin_lock(&rq1->lock); | |
2092 | } | |
2093 | } | |
2094 | } | |
2095 | ||
2096 | /* | |
2097 | * double_rq_unlock - safely unlock two runqueues | |
2098 | * | |
2099 | * Note this does not restore interrupts like task_rq_unlock, | |
2100 | * you need to do so manually after calling. | |
2101 | */ | |
70b97a7f | 2102 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2103 | __releases(rq1->lock) |
2104 | __releases(rq2->lock) | |
2105 | { | |
2106 | spin_unlock(&rq1->lock); | |
2107 | if (rq1 != rq2) | |
2108 | spin_unlock(&rq2->lock); | |
2109 | else | |
2110 | __release(rq2->lock); | |
2111 | } | |
2112 | ||
2113 | /* | |
2114 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2115 | */ | |
70b97a7f | 2116 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2117 | __releases(this_rq->lock) |
2118 | __acquires(busiest->lock) | |
2119 | __acquires(this_rq->lock) | |
2120 | { | |
054b9108 KK |
2121 | if (unlikely(!irqs_disabled())) { |
2122 | /* printk() doesn't work good under rq->lock */ | |
2123 | spin_unlock(&this_rq->lock); | |
2124 | BUG_ON(1); | |
2125 | } | |
1da177e4 | 2126 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2127 | if (busiest < this_rq) { |
1da177e4 LT |
2128 | spin_unlock(&this_rq->lock); |
2129 | spin_lock(&busiest->lock); | |
2130 | spin_lock(&this_rq->lock); | |
2131 | } else | |
2132 | spin_lock(&busiest->lock); | |
2133 | } | |
2134 | } | |
2135 | ||
1da177e4 LT |
2136 | /* |
2137 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2138 | * This is accomplished by forcing the cpu_allowed mask to only | |
2139 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2140 | * the cpu_allowed mask is restored. | |
2141 | */ | |
36c8b586 | 2142 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2143 | { |
70b97a7f | 2144 | struct migration_req req; |
1da177e4 | 2145 | unsigned long flags; |
70b97a7f | 2146 | struct rq *rq; |
1da177e4 LT |
2147 | |
2148 | rq = task_rq_lock(p, &flags); | |
2149 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2150 | || unlikely(cpu_is_offline(dest_cpu))) | |
2151 | goto out; | |
2152 | ||
2153 | /* force the process onto the specified CPU */ | |
2154 | if (migrate_task(p, dest_cpu, &req)) { | |
2155 | /* Need to wait for migration thread (might exit: take ref). */ | |
2156 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2157 | |
1da177e4 LT |
2158 | get_task_struct(mt); |
2159 | task_rq_unlock(rq, &flags); | |
2160 | wake_up_process(mt); | |
2161 | put_task_struct(mt); | |
2162 | wait_for_completion(&req.done); | |
36c8b586 | 2163 | |
1da177e4 LT |
2164 | return; |
2165 | } | |
2166 | out: | |
2167 | task_rq_unlock(rq, &flags); | |
2168 | } | |
2169 | ||
2170 | /* | |
476d139c NP |
2171 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2172 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2173 | */ |
2174 | void sched_exec(void) | |
2175 | { | |
1da177e4 | 2176 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2177 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2178 | put_cpu(); |
476d139c NP |
2179 | if (new_cpu != this_cpu) |
2180 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2181 | } |
2182 | ||
2183 | /* | |
2184 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2185 | * Both runqueues must be locked. | |
2186 | */ | |
70b97a7f IM |
2187 | static void pull_task(struct rq *src_rq, struct prio_array *src_array, |
2188 | struct task_struct *p, struct rq *this_rq, | |
2189 | struct prio_array *this_array, int this_cpu) | |
1da177e4 LT |
2190 | { |
2191 | dequeue_task(p, src_array); | |
2dd73a4f | 2192 | dec_nr_running(p, src_rq); |
1da177e4 | 2193 | set_task_cpu(p, this_cpu); |
2dd73a4f | 2194 | inc_nr_running(p, this_rq); |
1da177e4 | 2195 | enqueue_task(p, this_array); |
b18ec803 MG |
2196 | p->timestamp = (p->timestamp - src_rq->most_recent_timestamp) |
2197 | + this_rq->most_recent_timestamp; | |
1da177e4 LT |
2198 | /* |
2199 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2200 | * to be always true for them. | |
2201 | */ | |
2202 | if (TASK_PREEMPTS_CURR(p, this_rq)) | |
2203 | resched_task(this_rq->curr); | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2208 | */ | |
858119e1 | 2209 | static |
70b97a7f | 2210 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2211 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2212 | int *all_pinned) |
1da177e4 LT |
2213 | { |
2214 | /* | |
2215 | * We do not migrate tasks that are: | |
2216 | * 1) running (obviously), or | |
2217 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2218 | * 3) are cache-hot on their current CPU. | |
2219 | */ | |
1da177e4 LT |
2220 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2221 | return 0; | |
81026794 NP |
2222 | *all_pinned = 0; |
2223 | ||
2224 | if (task_running(rq, p)) | |
2225 | return 0; | |
1da177e4 LT |
2226 | |
2227 | /* | |
2228 | * Aggressive migration if: | |
cafb20c1 | 2229 | * 1) task is cache cold, or |
1da177e4 LT |
2230 | * 2) too many balance attempts have failed. |
2231 | */ | |
2232 | ||
b18ec803 MG |
2233 | if (sd->nr_balance_failed > sd->cache_nice_tries) { |
2234 | #ifdef CONFIG_SCHEDSTATS | |
2235 | if (task_hot(p, rq->most_recent_timestamp, sd)) | |
2236 | schedstat_inc(sd, lb_hot_gained[idle]); | |
2237 | #endif | |
1da177e4 | 2238 | return 1; |
b18ec803 | 2239 | } |
1da177e4 | 2240 | |
b18ec803 | 2241 | if (task_hot(p, rq->most_recent_timestamp, sd)) |
81026794 | 2242 | return 0; |
1da177e4 LT |
2243 | return 1; |
2244 | } | |
2245 | ||
615052dc | 2246 | #define rq_best_prio(rq) min((rq)->curr->prio, (rq)->best_expired_prio) |
48f24c4d | 2247 | |
1da177e4 | 2248 | /* |
2dd73a4f PW |
2249 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted |
2250 | * load from busiest to this_rq, as part of a balancing operation within | |
2251 | * "domain". Returns the number of tasks moved. | |
1da177e4 LT |
2252 | * |
2253 | * Called with both runqueues locked. | |
2254 | */ | |
70b97a7f | 2255 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2256 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2257 | struct sched_domain *sd, enum cpu_idle_type idle, |
2dd73a4f | 2258 | int *all_pinned) |
1da177e4 | 2259 | { |
48f24c4d IM |
2260 | int idx, pulled = 0, pinned = 0, this_best_prio, best_prio, |
2261 | best_prio_seen, skip_for_load; | |
70b97a7f | 2262 | struct prio_array *array, *dst_array; |
1da177e4 | 2263 | struct list_head *head, *curr; |
36c8b586 | 2264 | struct task_struct *tmp; |
2dd73a4f | 2265 | long rem_load_move; |
1da177e4 | 2266 | |
2dd73a4f | 2267 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2268 | goto out; |
2269 | ||
2dd73a4f | 2270 | rem_load_move = max_load_move; |
81026794 | 2271 | pinned = 1; |
615052dc | 2272 | this_best_prio = rq_best_prio(this_rq); |
48f24c4d | 2273 | best_prio = rq_best_prio(busiest); |
615052dc PW |
2274 | /* |
2275 | * Enable handling of the case where there is more than one task | |
2276 | * with the best priority. If the current running task is one | |
48f24c4d | 2277 | * of those with prio==best_prio we know it won't be moved |
615052dc PW |
2278 | * and therefore it's safe to override the skip (based on load) of |
2279 | * any task we find with that prio. | |
2280 | */ | |
48f24c4d | 2281 | best_prio_seen = best_prio == busiest->curr->prio; |
81026794 | 2282 | |
1da177e4 LT |
2283 | /* |
2284 | * We first consider expired tasks. Those will likely not be | |
2285 | * executed in the near future, and they are most likely to | |
2286 | * be cache-cold, thus switching CPUs has the least effect | |
2287 | * on them. | |
2288 | */ | |
2289 | if (busiest->expired->nr_active) { | |
2290 | array = busiest->expired; | |
2291 | dst_array = this_rq->expired; | |
2292 | } else { | |
2293 | array = busiest->active; | |
2294 | dst_array = this_rq->active; | |
2295 | } | |
2296 | ||
2297 | new_array: | |
2298 | /* Start searching at priority 0: */ | |
2299 | idx = 0; | |
2300 | skip_bitmap: | |
2301 | if (!idx) | |
2302 | idx = sched_find_first_bit(array->bitmap); | |
2303 | else | |
2304 | idx = find_next_bit(array->bitmap, MAX_PRIO, idx); | |
2305 | if (idx >= MAX_PRIO) { | |
2306 | if (array == busiest->expired && busiest->active->nr_active) { | |
2307 | array = busiest->active; | |
2308 | dst_array = this_rq->active; | |
2309 | goto new_array; | |
2310 | } | |
2311 | goto out; | |
2312 | } | |
2313 | ||
2314 | head = array->queue + idx; | |
2315 | curr = head->prev; | |
2316 | skip_queue: | |
36c8b586 | 2317 | tmp = list_entry(curr, struct task_struct, run_list); |
1da177e4 LT |
2318 | |
2319 | curr = curr->prev; | |
2320 | ||
50ddd969 PW |
2321 | /* |
2322 | * To help distribute high priority tasks accross CPUs we don't | |
2323 | * skip a task if it will be the highest priority task (i.e. smallest | |
2324 | * prio value) on its new queue regardless of its load weight | |
2325 | */ | |
615052dc PW |
2326 | skip_for_load = tmp->load_weight > rem_load_move; |
2327 | if (skip_for_load && idx < this_best_prio) | |
48f24c4d | 2328 | skip_for_load = !best_prio_seen && idx == best_prio; |
615052dc | 2329 | if (skip_for_load || |
2dd73a4f | 2330 | !can_migrate_task(tmp, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d IM |
2331 | |
2332 | best_prio_seen |= idx == best_prio; | |
1da177e4 LT |
2333 | if (curr != head) |
2334 | goto skip_queue; | |
2335 | idx++; | |
2336 | goto skip_bitmap; | |
2337 | } | |
2338 | ||
1da177e4 LT |
2339 | pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu); |
2340 | pulled++; | |
2dd73a4f | 2341 | rem_load_move -= tmp->load_weight; |
1da177e4 | 2342 | |
2dd73a4f PW |
2343 | /* |
2344 | * We only want to steal up to the prescribed number of tasks | |
2345 | * and the prescribed amount of weighted load. | |
2346 | */ | |
2347 | if (pulled < max_nr_move && rem_load_move > 0) { | |
615052dc PW |
2348 | if (idx < this_best_prio) |
2349 | this_best_prio = idx; | |
1da177e4 LT |
2350 | if (curr != head) |
2351 | goto skip_queue; | |
2352 | idx++; | |
2353 | goto skip_bitmap; | |
2354 | } | |
2355 | out: | |
2356 | /* | |
2357 | * Right now, this is the only place pull_task() is called, | |
2358 | * so we can safely collect pull_task() stats here rather than | |
2359 | * inside pull_task(). | |
2360 | */ | |
2361 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2362 | |
2363 | if (all_pinned) | |
2364 | *all_pinned = pinned; | |
1da177e4 LT |
2365 | return pulled; |
2366 | } | |
2367 | ||
2368 | /* | |
2369 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2370 | * domain. It calculates and returns the amount of weighted load which |
2371 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2372 | */ |
2373 | static struct sched_group * | |
2374 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
d15bcfdb | 2375 | unsigned long *imbalance, enum cpu_idle_type idle, int *sd_idle, |
783609c6 | 2376 | cpumask_t *cpus, int *balance) |
1da177e4 LT |
2377 | { |
2378 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2379 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2380 | unsigned long max_pull; |
2dd73a4f PW |
2381 | unsigned long busiest_load_per_task, busiest_nr_running; |
2382 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2383 | int load_idx; |
5c45bf27 SS |
2384 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2385 | int power_savings_balance = 1; | |
2386 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2387 | unsigned long min_nr_running = ULONG_MAX; | |
2388 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2389 | #endif | |
1da177e4 LT |
2390 | |
2391 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2392 | busiest_load_per_task = busiest_nr_running = 0; |
2393 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2394 | if (idle == CPU_NOT_IDLE) |
7897986b | 2395 | load_idx = sd->busy_idx; |
d15bcfdb | 2396 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2397 | load_idx = sd->newidle_idx; |
2398 | else | |
2399 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2400 | |
2401 | do { | |
5c45bf27 | 2402 | unsigned long load, group_capacity; |
1da177e4 LT |
2403 | int local_group; |
2404 | int i; | |
783609c6 | 2405 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2406 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2407 | |
2408 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2409 | ||
783609c6 SS |
2410 | if (local_group) |
2411 | balance_cpu = first_cpu(group->cpumask); | |
2412 | ||
1da177e4 | 2413 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2414 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2415 | |
2416 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2417 | struct rq *rq; |
2418 | ||
2419 | if (!cpu_isset(i, *cpus)) | |
2420 | continue; | |
2421 | ||
2422 | rq = cpu_rq(i); | |
2dd73a4f | 2423 | |
5969fe06 NP |
2424 | if (*sd_idle && !idle_cpu(i)) |
2425 | *sd_idle = 0; | |
2426 | ||
1da177e4 | 2427 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2428 | if (local_group) { |
2429 | if (idle_cpu(i) && !first_idle_cpu) { | |
2430 | first_idle_cpu = 1; | |
2431 | balance_cpu = i; | |
2432 | } | |
2433 | ||
a2000572 | 2434 | load = target_load(i, load_idx); |
783609c6 | 2435 | } else |
a2000572 | 2436 | load = source_load(i, load_idx); |
1da177e4 LT |
2437 | |
2438 | avg_load += load; | |
2dd73a4f PW |
2439 | sum_nr_running += rq->nr_running; |
2440 | sum_weighted_load += rq->raw_weighted_load; | |
1da177e4 LT |
2441 | } |
2442 | ||
783609c6 SS |
2443 | /* |
2444 | * First idle cpu or the first cpu(busiest) in this sched group | |
2445 | * is eligible for doing load balancing at this and above | |
2446 | * domains. | |
2447 | */ | |
2448 | if (local_group && balance_cpu != this_cpu && balance) { | |
2449 | *balance = 0; | |
2450 | goto ret; | |
2451 | } | |
2452 | ||
1da177e4 | 2453 | total_load += avg_load; |
5517d86b | 2454 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2455 | |
2456 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2457 | avg_load = sg_div_cpu_power(group, |
2458 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2459 | |
5517d86b | 2460 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2461 | |
1da177e4 LT |
2462 | if (local_group) { |
2463 | this_load = avg_load; | |
2464 | this = group; | |
2dd73a4f PW |
2465 | this_nr_running = sum_nr_running; |
2466 | this_load_per_task = sum_weighted_load; | |
2467 | } else if (avg_load > max_load && | |
5c45bf27 | 2468 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2469 | max_load = avg_load; |
2470 | busiest = group; | |
2dd73a4f PW |
2471 | busiest_nr_running = sum_nr_running; |
2472 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2473 | } |
5c45bf27 SS |
2474 | |
2475 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2476 | /* | |
2477 | * Busy processors will not participate in power savings | |
2478 | * balance. | |
2479 | */ | |
d15bcfdb | 2480 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 SS |
2481 | goto group_next; |
2482 | ||
2483 | /* | |
2484 | * If the local group is idle or completely loaded | |
2485 | * no need to do power savings balance at this domain | |
2486 | */ | |
2487 | if (local_group && (this_nr_running >= group_capacity || | |
2488 | !this_nr_running)) | |
2489 | power_savings_balance = 0; | |
2490 | ||
2491 | /* | |
2492 | * If a group is already running at full capacity or idle, | |
2493 | * don't include that group in power savings calculations | |
2494 | */ | |
2495 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
2496 | || !sum_nr_running) | |
2497 | goto group_next; | |
2498 | ||
2499 | /* | |
2500 | * Calculate the group which has the least non-idle load. | |
2501 | * This is the group from where we need to pick up the load | |
2502 | * for saving power | |
2503 | */ | |
2504 | if ((sum_nr_running < min_nr_running) || | |
2505 | (sum_nr_running == min_nr_running && | |
2506 | first_cpu(group->cpumask) < | |
2507 | first_cpu(group_min->cpumask))) { | |
2508 | group_min = group; | |
2509 | min_nr_running = sum_nr_running; | |
2510 | min_load_per_task = sum_weighted_load / | |
2511 | sum_nr_running; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * Calculate the group which is almost near its | |
2516 | * capacity but still has some space to pick up some load | |
2517 | * from other group and save more power | |
2518 | */ | |
48f24c4d | 2519 | if (sum_nr_running <= group_capacity - 1) { |
5c45bf27 SS |
2520 | if (sum_nr_running > leader_nr_running || |
2521 | (sum_nr_running == leader_nr_running && | |
2522 | first_cpu(group->cpumask) > | |
2523 | first_cpu(group_leader->cpumask))) { | |
2524 | group_leader = group; | |
2525 | leader_nr_running = sum_nr_running; | |
2526 | } | |
48f24c4d | 2527 | } |
5c45bf27 SS |
2528 | group_next: |
2529 | #endif | |
1da177e4 LT |
2530 | group = group->next; |
2531 | } while (group != sd->groups); | |
2532 | ||
2dd73a4f | 2533 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2534 | goto out_balanced; |
2535 | ||
2536 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2537 | ||
2538 | if (this_load >= avg_load || | |
2539 | 100*max_load <= sd->imbalance_pct*this_load) | |
2540 | goto out_balanced; | |
2541 | ||
2dd73a4f | 2542 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2543 | /* |
2544 | * We're trying to get all the cpus to the average_load, so we don't | |
2545 | * want to push ourselves above the average load, nor do we wish to | |
2546 | * reduce the max loaded cpu below the average load, as either of these | |
2547 | * actions would just result in more rebalancing later, and ping-pong | |
2548 | * tasks around. Thus we look for the minimum possible imbalance. | |
2549 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2550 | * be counted as no imbalance for these purposes -- we can't fix that | |
2551 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2552 | * appear as very large values with unsigned longs. | |
2553 | */ | |
2dd73a4f PW |
2554 | if (max_load <= busiest_load_per_task) |
2555 | goto out_balanced; | |
2556 | ||
2557 | /* | |
2558 | * In the presence of smp nice balancing, certain scenarios can have | |
2559 | * max load less than avg load(as we skip the groups at or below | |
2560 | * its cpu_power, while calculating max_load..) | |
2561 | */ | |
2562 | if (max_load < avg_load) { | |
2563 | *imbalance = 0; | |
2564 | goto small_imbalance; | |
2565 | } | |
0c117f1b SS |
2566 | |
2567 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2568 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2569 | |
1da177e4 | 2570 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2571 | *imbalance = min(max_pull * busiest->__cpu_power, |
2572 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2573 | / SCHED_LOAD_SCALE; |
2574 | ||
2dd73a4f PW |
2575 | /* |
2576 | * if *imbalance is less than the average load per runnable task | |
2577 | * there is no gaurantee that any tasks will be moved so we'll have | |
2578 | * a think about bumping its value to force at least one task to be | |
2579 | * moved | |
2580 | */ | |
2581 | if (*imbalance < busiest_load_per_task) { | |
48f24c4d | 2582 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2583 | unsigned int imbn; |
2584 | ||
2585 | small_imbalance: | |
2586 | pwr_move = pwr_now = 0; | |
2587 | imbn = 2; | |
2588 | if (this_nr_running) { | |
2589 | this_load_per_task /= this_nr_running; | |
2590 | if (busiest_load_per_task > this_load_per_task) | |
2591 | imbn = 1; | |
2592 | } else | |
2593 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2594 | |
2dd73a4f PW |
2595 | if (max_load - this_load >= busiest_load_per_task * imbn) { |
2596 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2597 | return busiest; |
2598 | } | |
2599 | ||
2600 | /* | |
2601 | * OK, we don't have enough imbalance to justify moving tasks, | |
2602 | * however we may be able to increase total CPU power used by | |
2603 | * moving them. | |
2604 | */ | |
2605 | ||
5517d86b ED |
2606 | pwr_now += busiest->__cpu_power * |
2607 | min(busiest_load_per_task, max_load); | |
2608 | pwr_now += this->__cpu_power * | |
2609 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2610 | pwr_now /= SCHED_LOAD_SCALE; |
2611 | ||
2612 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2613 | tmp = sg_div_cpu_power(busiest, |
2614 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2615 | if (max_load > tmp) |
5517d86b | 2616 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2617 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2618 | |
2619 | /* Amount of load we'd add */ | |
5517d86b | 2620 | if (max_load * busiest->__cpu_power < |
33859f7f | 2621 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2622 | tmp = sg_div_cpu_power(this, |
2623 | max_load * busiest->__cpu_power); | |
1da177e4 | 2624 | else |
5517d86b ED |
2625 | tmp = sg_div_cpu_power(this, |
2626 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2627 | pwr_move += this->__cpu_power * | |
2628 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2629 | pwr_move /= SCHED_LOAD_SCALE; |
2630 | ||
2631 | /* Move if we gain throughput */ | |
2632 | if (pwr_move <= pwr_now) | |
2633 | goto out_balanced; | |
2634 | ||
2dd73a4f | 2635 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2636 | } |
2637 | ||
1da177e4 LT |
2638 | return busiest; |
2639 | ||
2640 | out_balanced: | |
5c45bf27 | 2641 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2642 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2643 | goto ret; |
1da177e4 | 2644 | |
5c45bf27 SS |
2645 | if (this == group_leader && group_leader != group_min) { |
2646 | *imbalance = min_load_per_task; | |
2647 | return group_min; | |
2648 | } | |
5c45bf27 | 2649 | #endif |
783609c6 | 2650 | ret: |
1da177e4 LT |
2651 | *imbalance = 0; |
2652 | return NULL; | |
2653 | } | |
2654 | ||
2655 | /* | |
2656 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2657 | */ | |
70b97a7f | 2658 | static struct rq * |
d15bcfdb | 2659 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2660 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2661 | { |
70b97a7f | 2662 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2663 | unsigned long max_load = 0; |
1da177e4 LT |
2664 | int i; |
2665 | ||
2666 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2667 | |
2668 | if (!cpu_isset(i, *cpus)) | |
2669 | continue; | |
2670 | ||
48f24c4d | 2671 | rq = cpu_rq(i); |
2dd73a4f | 2672 | |
48f24c4d | 2673 | if (rq->nr_running == 1 && rq->raw_weighted_load > imbalance) |
2dd73a4f | 2674 | continue; |
1da177e4 | 2675 | |
48f24c4d IM |
2676 | if (rq->raw_weighted_load > max_load) { |
2677 | max_load = rq->raw_weighted_load; | |
2678 | busiest = rq; | |
1da177e4 LT |
2679 | } |
2680 | } | |
2681 | ||
2682 | return busiest; | |
2683 | } | |
2684 | ||
77391d71 NP |
2685 | /* |
2686 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2687 | * so long as it is large enough. | |
2688 | */ | |
2689 | #define MAX_PINNED_INTERVAL 512 | |
2690 | ||
48f24c4d IM |
2691 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2692 | { | |
2693 | return n > 0 ? n - 1 : 0; | |
2694 | } | |
2695 | ||
1da177e4 LT |
2696 | /* |
2697 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2698 | * tasks if there is an imbalance. | |
1da177e4 | 2699 | */ |
70b97a7f | 2700 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2701 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2702 | int *balance) |
1da177e4 | 2703 | { |
48f24c4d | 2704 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2705 | struct sched_group *group; |
1da177e4 | 2706 | unsigned long imbalance; |
70b97a7f | 2707 | struct rq *busiest; |
0a2966b4 | 2708 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2709 | unsigned long flags; |
5969fe06 | 2710 | |
89c4710e SS |
2711 | /* |
2712 | * When power savings policy is enabled for the parent domain, idle | |
2713 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2714 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2715 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2716 | */ |
d15bcfdb | 2717 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2718 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2719 | sd_idle = 1; |
1da177e4 | 2720 | |
1da177e4 LT |
2721 | schedstat_inc(sd, lb_cnt[idle]); |
2722 | ||
0a2966b4 CL |
2723 | redo: |
2724 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2725 | &cpus, balance); |
2726 | ||
06066714 | 2727 | if (*balance == 0) |
783609c6 | 2728 | goto out_balanced; |
783609c6 | 2729 | |
1da177e4 LT |
2730 | if (!group) { |
2731 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2732 | goto out_balanced; | |
2733 | } | |
2734 | ||
0a2966b4 | 2735 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2736 | if (!busiest) { |
2737 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2738 | goto out_balanced; | |
2739 | } | |
2740 | ||
db935dbd | 2741 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2742 | |
2743 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2744 | ||
2745 | nr_moved = 0; | |
2746 | if (busiest->nr_running > 1) { | |
2747 | /* | |
2748 | * Attempt to move tasks. If find_busiest_group has found | |
2749 | * an imbalance but busiest->nr_running <= 1, the group is | |
2750 | * still unbalanced. nr_moved simply stays zero, so it is | |
2751 | * correctly treated as an imbalance. | |
2752 | */ | |
fe2eea3f | 2753 | local_irq_save(flags); |
e17224bf | 2754 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2755 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2756 | minus_1_or_zero(busiest->nr_running), |
2757 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2758 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2759 | local_irq_restore(flags); |
81026794 | 2760 | |
46cb4b7c SS |
2761 | /* |
2762 | * some other cpu did the load balance for us. | |
2763 | */ | |
2764 | if (nr_moved && this_cpu != smp_processor_id()) | |
2765 | resched_cpu(this_cpu); | |
2766 | ||
81026794 | 2767 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2768 | if (unlikely(all_pinned)) { |
2769 | cpu_clear(cpu_of(busiest), cpus); | |
2770 | if (!cpus_empty(cpus)) | |
2771 | goto redo; | |
81026794 | 2772 | goto out_balanced; |
0a2966b4 | 2773 | } |
1da177e4 | 2774 | } |
81026794 | 2775 | |
1da177e4 LT |
2776 | if (!nr_moved) { |
2777 | schedstat_inc(sd, lb_failed[idle]); | |
2778 | sd->nr_balance_failed++; | |
2779 | ||
2780 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2781 | |
fe2eea3f | 2782 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2783 | |
2784 | /* don't kick the migration_thread, if the curr | |
2785 | * task on busiest cpu can't be moved to this_cpu | |
2786 | */ | |
2787 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2788 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2789 | all_pinned = 1; |
2790 | goto out_one_pinned; | |
2791 | } | |
2792 | ||
1da177e4 LT |
2793 | if (!busiest->active_balance) { |
2794 | busiest->active_balance = 1; | |
2795 | busiest->push_cpu = this_cpu; | |
81026794 | 2796 | active_balance = 1; |
1da177e4 | 2797 | } |
fe2eea3f | 2798 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2799 | if (active_balance) |
1da177e4 LT |
2800 | wake_up_process(busiest->migration_thread); |
2801 | ||
2802 | /* | |
2803 | * We've kicked active balancing, reset the failure | |
2804 | * counter. | |
2805 | */ | |
39507451 | 2806 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2807 | } |
81026794 | 2808 | } else |
1da177e4 LT |
2809 | sd->nr_balance_failed = 0; |
2810 | ||
81026794 | 2811 | if (likely(!active_balance)) { |
1da177e4 LT |
2812 | /* We were unbalanced, so reset the balancing interval */ |
2813 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2814 | } else { |
2815 | /* | |
2816 | * If we've begun active balancing, start to back off. This | |
2817 | * case may not be covered by the all_pinned logic if there | |
2818 | * is only 1 task on the busy runqueue (because we don't call | |
2819 | * move_tasks). | |
2820 | */ | |
2821 | if (sd->balance_interval < sd->max_interval) | |
2822 | sd->balance_interval *= 2; | |
1da177e4 LT |
2823 | } |
2824 | ||
5c45bf27 | 2825 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2826 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2827 | return -1; |
1da177e4 LT |
2828 | return nr_moved; |
2829 | ||
2830 | out_balanced: | |
1da177e4 LT |
2831 | schedstat_inc(sd, lb_balanced[idle]); |
2832 | ||
16cfb1c0 | 2833 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2834 | |
2835 | out_one_pinned: | |
1da177e4 | 2836 | /* tune up the balancing interval */ |
77391d71 NP |
2837 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2838 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2839 | sd->balance_interval *= 2; |
2840 | ||
48f24c4d | 2841 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2842 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2843 | return -1; |
1da177e4 LT |
2844 | return 0; |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2849 | * tasks if there is an imbalance. | |
2850 | * | |
d15bcfdb | 2851 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2852 | * this_rq is locked. |
2853 | */ | |
48f24c4d | 2854 | static int |
70b97a7f | 2855 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2856 | { |
2857 | struct sched_group *group; | |
70b97a7f | 2858 | struct rq *busiest = NULL; |
1da177e4 LT |
2859 | unsigned long imbalance; |
2860 | int nr_moved = 0; | |
5969fe06 | 2861 | int sd_idle = 0; |
0a2966b4 | 2862 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2863 | |
89c4710e SS |
2864 | /* |
2865 | * When power savings policy is enabled for the parent domain, idle | |
2866 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2867 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2868 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2869 | */ |
2870 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2871 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2872 | sd_idle = 1; |
1da177e4 | 2873 | |
d15bcfdb | 2874 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2875 | redo: |
d15bcfdb | 2876 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2877 | &sd_idle, &cpus, NULL); |
1da177e4 | 2878 | if (!group) { |
d15bcfdb | 2879 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2880 | goto out_balanced; |
1da177e4 LT |
2881 | } |
2882 | ||
d15bcfdb | 2883 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2884 | &cpus); |
db935dbd | 2885 | if (!busiest) { |
d15bcfdb | 2886 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2887 | goto out_balanced; |
1da177e4 LT |
2888 | } |
2889 | ||
db935dbd NP |
2890 | BUG_ON(busiest == this_rq); |
2891 | ||
d15bcfdb | 2892 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2893 | |
2894 | nr_moved = 0; | |
2895 | if (busiest->nr_running > 1) { | |
2896 | /* Attempt to move tasks */ | |
2897 | double_lock_balance(this_rq, busiest); | |
2898 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2899 | minus_1_or_zero(busiest->nr_running), |
d15bcfdb | 2900 | imbalance, sd, CPU_NEWLY_IDLE, NULL); |
d6d5cfaf | 2901 | spin_unlock(&busiest->lock); |
0a2966b4 CL |
2902 | |
2903 | if (!nr_moved) { | |
2904 | cpu_clear(cpu_of(busiest), cpus); | |
2905 | if (!cpus_empty(cpus)) | |
2906 | goto redo; | |
2907 | } | |
d6d5cfaf NP |
2908 | } |
2909 | ||
5969fe06 | 2910 | if (!nr_moved) { |
d15bcfdb | 2911 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2912 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2913 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2914 | return -1; |
2915 | } else | |
16cfb1c0 | 2916 | sd->nr_balance_failed = 0; |
1da177e4 | 2917 | |
1da177e4 | 2918 | return nr_moved; |
16cfb1c0 NP |
2919 | |
2920 | out_balanced: | |
d15bcfdb | 2921 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2922 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2923 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2924 | return -1; |
16cfb1c0 | 2925 | sd->nr_balance_failed = 0; |
48f24c4d | 2926 | |
16cfb1c0 | 2927 | return 0; |
1da177e4 LT |
2928 | } |
2929 | ||
2930 | /* | |
2931 | * idle_balance is called by schedule() if this_cpu is about to become | |
2932 | * idle. Attempts to pull tasks from other CPUs. | |
2933 | */ | |
70b97a7f | 2934 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2935 | { |
2936 | struct sched_domain *sd; | |
1bd77f2d CL |
2937 | int pulled_task = 0; |
2938 | unsigned long next_balance = jiffies + 60 * HZ; | |
1da177e4 LT |
2939 | |
2940 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2941 | unsigned long interval; |
2942 | ||
2943 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2944 | continue; | |
2945 | ||
2946 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2947 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2948 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2949 | this_rq, sd); |
2950 | ||
2951 | interval = msecs_to_jiffies(sd->balance_interval); | |
2952 | if (time_after(next_balance, sd->last_balance + interval)) | |
2953 | next_balance = sd->last_balance + interval; | |
2954 | if (pulled_task) | |
2955 | break; | |
1da177e4 | 2956 | } |
1bd77f2d CL |
2957 | if (!pulled_task) |
2958 | /* | |
2959 | * We are going idle. next_balance may be set based on | |
2960 | * a busy processor. So reset next_balance. | |
2961 | */ | |
2962 | this_rq->next_balance = next_balance; | |
1da177e4 LT |
2963 | } |
2964 | ||
2965 | /* | |
2966 | * active_load_balance is run by migration threads. It pushes running tasks | |
2967 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2968 | * running on each physical CPU where possible, and avoids physical / | |
2969 | * logical imbalances. | |
2970 | * | |
2971 | * Called with busiest_rq locked. | |
2972 | */ | |
70b97a7f | 2973 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2974 | { |
39507451 | 2975 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2976 | struct sched_domain *sd; |
2977 | struct rq *target_rq; | |
39507451 | 2978 | |
48f24c4d | 2979 | /* Is there any task to move? */ |
39507451 | 2980 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2981 | return; |
2982 | ||
2983 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2984 | |
2985 | /* | |
39507451 NP |
2986 | * This condition is "impossible", if it occurs |
2987 | * we need to fix it. Originally reported by | |
2988 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2989 | */ |
39507451 | 2990 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2991 | |
39507451 NP |
2992 | /* move a task from busiest_rq to target_rq */ |
2993 | double_lock_balance(busiest_rq, target_rq); | |
2994 | ||
2995 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2996 | for_each_domain(target_cpu, sd) { |
39507451 | 2997 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2998 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2999 | break; |
c96d145e | 3000 | } |
39507451 | 3001 | |
48f24c4d IM |
3002 | if (likely(sd)) { |
3003 | schedstat_inc(sd, alb_cnt); | |
39507451 | 3004 | |
48f24c4d | 3005 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
d15bcfdb | 3006 | RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE, |
48f24c4d IM |
3007 | NULL)) |
3008 | schedstat_inc(sd, alb_pushed); | |
3009 | else | |
3010 | schedstat_inc(sd, alb_failed); | |
3011 | } | |
39507451 | 3012 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3013 | } |
3014 | ||
7835b98b | 3015 | static void update_load(struct rq *this_rq) |
1da177e4 | 3016 | { |
7835b98b | 3017 | unsigned long this_load; |
ff91691b | 3018 | unsigned int i, scale; |
1da177e4 | 3019 | |
2dd73a4f | 3020 | this_load = this_rq->raw_weighted_load; |
48f24c4d IM |
3021 | |
3022 | /* Update our load: */ | |
ff91691b | 3023 | for (i = 0, scale = 1; i < 3; i++, scale += scale) { |
48f24c4d IM |
3024 | unsigned long old_load, new_load; |
3025 | ||
ff91691b NP |
3026 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
3027 | ||
7897986b | 3028 | old_load = this_rq->cpu_load[i]; |
48f24c4d | 3029 | new_load = this_load; |
7897986b NP |
3030 | /* |
3031 | * Round up the averaging division if load is increasing. This | |
3032 | * prevents us from getting stuck on 9 if the load is 10, for | |
3033 | * example. | |
3034 | */ | |
3035 | if (new_load > old_load) | |
3036 | new_load += scale-1; | |
ff91691b | 3037 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
7897986b | 3038 | } |
7835b98b CL |
3039 | } |
3040 | ||
46cb4b7c SS |
3041 | #ifdef CONFIG_NO_HZ |
3042 | static struct { | |
3043 | atomic_t load_balancer; | |
3044 | cpumask_t cpu_mask; | |
3045 | } nohz ____cacheline_aligned = { | |
3046 | .load_balancer = ATOMIC_INIT(-1), | |
3047 | .cpu_mask = CPU_MASK_NONE, | |
3048 | }; | |
3049 | ||
7835b98b | 3050 | /* |
46cb4b7c SS |
3051 | * This routine will try to nominate the ilb (idle load balancing) |
3052 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3053 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3054 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3055 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3056 | * arrives... | |
3057 | * | |
3058 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3059 | * for idle load balancing. ilb owner will still be part of | |
3060 | * nohz.cpu_mask.. | |
7835b98b | 3061 | * |
46cb4b7c SS |
3062 | * While stopping the tick, this cpu will become the ilb owner if there |
3063 | * is no other owner. And will be the owner till that cpu becomes busy | |
3064 | * or if all cpus in the system stop their ticks at which point | |
3065 | * there is no need for ilb owner. | |
3066 | * | |
3067 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3068 | * next busy scheduler_tick() | |
3069 | */ | |
3070 | int select_nohz_load_balancer(int stop_tick) | |
3071 | { | |
3072 | int cpu = smp_processor_id(); | |
3073 | ||
3074 | if (stop_tick) { | |
3075 | cpu_set(cpu, nohz.cpu_mask); | |
3076 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3077 | ||
3078 | /* | |
3079 | * If we are going offline and still the leader, give up! | |
3080 | */ | |
3081 | if (cpu_is_offline(cpu) && | |
3082 | atomic_read(&nohz.load_balancer) == cpu) { | |
3083 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3084 | BUG(); | |
3085 | return 0; | |
3086 | } | |
3087 | ||
3088 | /* time for ilb owner also to sleep */ | |
3089 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3090 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3091 | atomic_set(&nohz.load_balancer, -1); | |
3092 | return 0; | |
3093 | } | |
3094 | ||
3095 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3096 | /* make me the ilb owner */ | |
3097 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3098 | return 1; | |
3099 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3100 | return 1; | |
3101 | } else { | |
3102 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3103 | return 0; | |
3104 | ||
3105 | cpu_clear(cpu, nohz.cpu_mask); | |
3106 | ||
3107 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3108 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3109 | BUG(); | |
3110 | } | |
3111 | return 0; | |
3112 | } | |
3113 | #endif | |
3114 | ||
3115 | static DEFINE_SPINLOCK(balancing); | |
3116 | ||
3117 | /* | |
7835b98b CL |
3118 | * It checks each scheduling domain to see if it is due to be balanced, |
3119 | * and initiates a balancing operation if so. | |
3120 | * | |
3121 | * Balancing parameters are set up in arch_init_sched_domains. | |
3122 | */ | |
d15bcfdb | 3123 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3124 | { |
46cb4b7c SS |
3125 | int balance = 1; |
3126 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3127 | unsigned long interval; |
3128 | struct sched_domain *sd; | |
46cb4b7c | 3129 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3130 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 3131 | |
46cb4b7c | 3132 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3133 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3134 | continue; | |
3135 | ||
3136 | interval = sd->balance_interval; | |
d15bcfdb | 3137 | if (idle != CPU_IDLE) |
1da177e4 LT |
3138 | interval *= sd->busy_factor; |
3139 | ||
3140 | /* scale ms to jiffies */ | |
3141 | interval = msecs_to_jiffies(interval); | |
3142 | if (unlikely(!interval)) | |
3143 | interval = 1; | |
3144 | ||
08c183f3 CL |
3145 | if (sd->flags & SD_SERIALIZE) { |
3146 | if (!spin_trylock(&balancing)) | |
3147 | goto out; | |
3148 | } | |
3149 | ||
c9819f45 | 3150 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3151 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3152 | /* |
3153 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3154 | * longer idle, or one of our SMT siblings is |
3155 | * not idle. | |
3156 | */ | |
d15bcfdb | 3157 | idle = CPU_NOT_IDLE; |
1da177e4 | 3158 | } |
1bd77f2d | 3159 | sd->last_balance = jiffies; |
1da177e4 | 3160 | } |
08c183f3 CL |
3161 | if (sd->flags & SD_SERIALIZE) |
3162 | spin_unlock(&balancing); | |
3163 | out: | |
c9819f45 CL |
3164 | if (time_after(next_balance, sd->last_balance + interval)) |
3165 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
3166 | |
3167 | /* | |
3168 | * Stop the load balance at this level. There is another | |
3169 | * CPU in our sched group which is doing load balancing more | |
3170 | * actively. | |
3171 | */ | |
3172 | if (!balance) | |
3173 | break; | |
1da177e4 | 3174 | } |
46cb4b7c SS |
3175 | rq->next_balance = next_balance; |
3176 | } | |
3177 | ||
3178 | /* | |
3179 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3180 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3181 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3182 | */ | |
3183 | static void run_rebalance_domains(struct softirq_action *h) | |
3184 | { | |
3185 | int local_cpu = smp_processor_id(); | |
3186 | struct rq *local_rq = cpu_rq(local_cpu); | |
d15bcfdb | 3187 | enum cpu_idle_type idle = local_rq->idle_at_tick ? CPU_IDLE : CPU_NOT_IDLE; |
46cb4b7c SS |
3188 | |
3189 | rebalance_domains(local_cpu, idle); | |
3190 | ||
3191 | #ifdef CONFIG_NO_HZ | |
3192 | /* | |
3193 | * If this cpu is the owner for idle load balancing, then do the | |
3194 | * balancing on behalf of the other idle cpus whose ticks are | |
3195 | * stopped. | |
3196 | */ | |
3197 | if (local_rq->idle_at_tick && | |
3198 | atomic_read(&nohz.load_balancer) == local_cpu) { | |
3199 | cpumask_t cpus = nohz.cpu_mask; | |
3200 | struct rq *rq; | |
3201 | int balance_cpu; | |
3202 | ||
3203 | cpu_clear(local_cpu, cpus); | |
3204 | for_each_cpu_mask(balance_cpu, cpus) { | |
3205 | /* | |
3206 | * If this cpu gets work to do, stop the load balancing | |
3207 | * work being done for other cpus. Next load | |
3208 | * balancing owner will pick it up. | |
3209 | */ | |
3210 | if (need_resched()) | |
3211 | break; | |
3212 | ||
d15bcfdb | 3213 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3214 | |
3215 | rq = cpu_rq(balance_cpu); | |
3216 | if (time_after(local_rq->next_balance, rq->next_balance)) | |
3217 | local_rq->next_balance = rq->next_balance; | |
3218 | } | |
3219 | } | |
3220 | #endif | |
3221 | } | |
3222 | ||
3223 | /* | |
3224 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3225 | * | |
3226 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3227 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3228 | * if the whole system is idle. | |
3229 | */ | |
3230 | static inline void trigger_load_balance(int cpu) | |
3231 | { | |
3232 | struct rq *rq = cpu_rq(cpu); | |
3233 | #ifdef CONFIG_NO_HZ | |
3234 | /* | |
3235 | * If we were in the nohz mode recently and busy at the current | |
3236 | * scheduler tick, then check if we need to nominate new idle | |
3237 | * load balancer. | |
3238 | */ | |
3239 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3240 | rq->in_nohz_recently = 0; | |
3241 | ||
3242 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3243 | cpu_clear(cpu, nohz.cpu_mask); | |
3244 | atomic_set(&nohz.load_balancer, -1); | |
3245 | } | |
3246 | ||
3247 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3248 | /* | |
3249 | * simple selection for now: Nominate the | |
3250 | * first cpu in the nohz list to be the next | |
3251 | * ilb owner. | |
3252 | * | |
3253 | * TBD: Traverse the sched domains and nominate | |
3254 | * the nearest cpu in the nohz.cpu_mask. | |
3255 | */ | |
3256 | int ilb = first_cpu(nohz.cpu_mask); | |
3257 | ||
3258 | if (ilb != NR_CPUS) | |
3259 | resched_cpu(ilb); | |
3260 | } | |
3261 | } | |
3262 | ||
3263 | /* | |
3264 | * If this cpu is idle and doing idle load balancing for all the | |
3265 | * cpus with ticks stopped, is it time for that to stop? | |
3266 | */ | |
3267 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3268 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3269 | resched_cpu(cpu); | |
3270 | return; | |
3271 | } | |
3272 | ||
3273 | /* | |
3274 | * If this cpu is idle and the idle load balancing is done by | |
3275 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3276 | */ | |
3277 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3278 | cpu_isset(cpu, nohz.cpu_mask)) | |
3279 | return; | |
3280 | #endif | |
3281 | if (time_after_eq(jiffies, rq->next_balance)) | |
3282 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 LT |
3283 | } |
3284 | #else | |
3285 | /* | |
3286 | * on UP we do not need to balance between CPUs: | |
3287 | */ | |
70b97a7f | 3288 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3289 | { |
3290 | } | |
3291 | #endif | |
3292 | ||
1da177e4 LT |
3293 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3294 | ||
3295 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3296 | ||
3297 | /* | |
3298 | * This is called on clock ticks and on context switches. | |
3299 | * Bank in p->sched_time the ns elapsed since the last tick or switch. | |
3300 | */ | |
48f24c4d | 3301 | static inline void |
70b97a7f | 3302 | update_cpu_clock(struct task_struct *p, struct rq *rq, unsigned long long now) |
1da177e4 | 3303 | { |
b18ec803 MG |
3304 | p->sched_time += now - p->last_ran; |
3305 | p->last_ran = rq->most_recent_timestamp = now; | |
1da177e4 LT |
3306 | } |
3307 | ||
3308 | /* | |
3309 | * Return current->sched_time plus any more ns on the sched_clock | |
3310 | * that have not yet been banked. | |
3311 | */ | |
36c8b586 | 3312 | unsigned long long current_sched_time(const struct task_struct *p) |
1da177e4 LT |
3313 | { |
3314 | unsigned long long ns; | |
3315 | unsigned long flags; | |
48f24c4d | 3316 | |
1da177e4 | 3317 | local_irq_save(flags); |
b18ec803 | 3318 | ns = p->sched_time + sched_clock() - p->last_ran; |
1da177e4 | 3319 | local_irq_restore(flags); |
48f24c4d | 3320 | |
1da177e4 LT |
3321 | return ns; |
3322 | } | |
3323 | ||
f1adad78 LT |
3324 | /* |
3325 | * We place interactive tasks back into the active array, if possible. | |
3326 | * | |
3327 | * To guarantee that this does not starve expired tasks we ignore the | |
3328 | * interactivity of a task if the first expired task had to wait more | |
3329 | * than a 'reasonable' amount of time. This deadline timeout is | |
3330 | * load-dependent, as the frequency of array switched decreases with | |
3331 | * increasing number of running tasks. We also ignore the interactivity | |
3332 | * if a better static_prio task has expired: | |
3333 | */ | |
70b97a7f | 3334 | static inline int expired_starving(struct rq *rq) |
48f24c4d IM |
3335 | { |
3336 | if (rq->curr->static_prio > rq->best_expired_prio) | |
3337 | return 1; | |
3338 | if (!STARVATION_LIMIT || !rq->expired_timestamp) | |
3339 | return 0; | |
3340 | if (jiffies - rq->expired_timestamp > STARVATION_LIMIT * rq->nr_running) | |
3341 | return 1; | |
3342 | return 0; | |
3343 | } | |
f1adad78 | 3344 | |
1da177e4 LT |
3345 | /* |
3346 | * Account user cpu time to a process. | |
3347 | * @p: the process that the cpu time gets accounted to | |
3348 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3349 | * @cputime: the cpu time spent in user space since the last update | |
3350 | */ | |
3351 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3352 | { | |
3353 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3354 | cputime64_t tmp; | |
3355 | ||
3356 | p->utime = cputime_add(p->utime, cputime); | |
3357 | ||
3358 | /* Add user time to cpustat. */ | |
3359 | tmp = cputime_to_cputime64(cputime); | |
3360 | if (TASK_NICE(p) > 0) | |
3361 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3362 | else | |
3363 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3364 | } | |
3365 | ||
3366 | /* | |
3367 | * Account system cpu time to a process. | |
3368 | * @p: the process that the cpu time gets accounted to | |
3369 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3370 | * @cputime: the cpu time spent in kernel space since the last update | |
3371 | */ | |
3372 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3373 | cputime_t cputime) | |
3374 | { | |
3375 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3376 | struct rq *rq = this_rq(); |
1da177e4 LT |
3377 | cputime64_t tmp; |
3378 | ||
3379 | p->stime = cputime_add(p->stime, cputime); | |
3380 | ||
3381 | /* Add system time to cpustat. */ | |
3382 | tmp = cputime_to_cputime64(cputime); | |
3383 | if (hardirq_count() - hardirq_offset) | |
3384 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3385 | else if (softirq_count()) | |
3386 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3387 | else if (p != rq->idle) | |
3388 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3389 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3390 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3391 | else | |
3392 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3393 | /* Account for system time used */ | |
3394 | acct_update_integrals(p); | |
1da177e4 LT |
3395 | } |
3396 | ||
3397 | /* | |
3398 | * Account for involuntary wait time. | |
3399 | * @p: the process from which the cpu time has been stolen | |
3400 | * @steal: the cpu time spent in involuntary wait | |
3401 | */ | |
3402 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3403 | { | |
3404 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3405 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3406 | struct rq *rq = this_rq(); |
1da177e4 LT |
3407 | |
3408 | if (p == rq->idle) { | |
3409 | p->stime = cputime_add(p->stime, steal); | |
3410 | if (atomic_read(&rq->nr_iowait) > 0) | |
3411 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3412 | else | |
3413 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3414 | } else | |
3415 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3416 | } | |
3417 | ||
7835b98b | 3418 | static void task_running_tick(struct rq *rq, struct task_struct *p) |
1da177e4 | 3419 | { |
1da177e4 | 3420 | if (p->array != rq->active) { |
7835b98b | 3421 | /* Task has expired but was not scheduled yet */ |
1da177e4 | 3422 | set_tsk_need_resched(p); |
7835b98b | 3423 | return; |
1da177e4 LT |
3424 | } |
3425 | spin_lock(&rq->lock); | |
3426 | /* | |
3427 | * The task was running during this tick - update the | |
3428 | * time slice counter. Note: we do not update a thread's | |
3429 | * priority until it either goes to sleep or uses up its | |
3430 | * timeslice. This makes it possible for interactive tasks | |
3431 | * to use up their timeslices at their highest priority levels. | |
3432 | */ | |
3433 | if (rt_task(p)) { | |
3434 | /* | |
3435 | * RR tasks need a special form of timeslice management. | |
3436 | * FIFO tasks have no timeslices. | |
3437 | */ | |
3438 | if ((p->policy == SCHED_RR) && !--p->time_slice) { | |
3439 | p->time_slice = task_timeslice(p); | |
3440 | p->first_time_slice = 0; | |
3441 | set_tsk_need_resched(p); | |
3442 | ||
3443 | /* put it at the end of the queue: */ | |
3444 | requeue_task(p, rq->active); | |
3445 | } | |
3446 | goto out_unlock; | |
3447 | } | |
3448 | if (!--p->time_slice) { | |
3449 | dequeue_task(p, rq->active); | |
3450 | set_tsk_need_resched(p); | |
3451 | p->prio = effective_prio(p); | |
3452 | p->time_slice = task_timeslice(p); | |
3453 | p->first_time_slice = 0; | |
3454 | ||
3455 | if (!rq->expired_timestamp) | |
3456 | rq->expired_timestamp = jiffies; | |
48f24c4d | 3457 | if (!TASK_INTERACTIVE(p) || expired_starving(rq)) { |
1da177e4 LT |
3458 | enqueue_task(p, rq->expired); |
3459 | if (p->static_prio < rq->best_expired_prio) | |
3460 | rq->best_expired_prio = p->static_prio; | |
3461 | } else | |
3462 | enqueue_task(p, rq->active); | |
3463 | } else { | |
3464 | /* | |
3465 | * Prevent a too long timeslice allowing a task to monopolize | |
3466 | * the CPU. We do this by splitting up the timeslice into | |
3467 | * smaller pieces. | |
3468 | * | |
3469 | * Note: this does not mean the task's timeslices expire or | |
3470 | * get lost in any way, they just might be preempted by | |
3471 | * another task of equal priority. (one with higher | |
3472 | * priority would have preempted this task already.) We | |
3473 | * requeue this task to the end of the list on this priority | |
3474 | * level, which is in essence a round-robin of tasks with | |
3475 | * equal priority. | |
3476 | * | |
3477 | * This only applies to tasks in the interactive | |
3478 | * delta range with at least TIMESLICE_GRANULARITY to requeue. | |
3479 | */ | |
3480 | if (TASK_INTERACTIVE(p) && !((task_timeslice(p) - | |
3481 | p->time_slice) % TIMESLICE_GRANULARITY(p)) && | |
3482 | (p->time_slice >= TIMESLICE_GRANULARITY(p)) && | |
3483 | (p->array == rq->active)) { | |
3484 | ||
3485 | requeue_task(p, rq->active); | |
3486 | set_tsk_need_resched(p); | |
3487 | } | |
3488 | } | |
3489 | out_unlock: | |
3490 | spin_unlock(&rq->lock); | |
7835b98b CL |
3491 | } |
3492 | ||
3493 | /* | |
3494 | * This function gets called by the timer code, with HZ frequency. | |
3495 | * We call it with interrupts disabled. | |
3496 | * | |
3497 | * It also gets called by the fork code, when changing the parent's | |
3498 | * timeslices. | |
3499 | */ | |
3500 | void scheduler_tick(void) | |
3501 | { | |
3502 | unsigned long long now = sched_clock(); | |
3503 | struct task_struct *p = current; | |
3504 | int cpu = smp_processor_id(); | |
bdecea3a | 3505 | int idle_at_tick = idle_cpu(cpu); |
7835b98b | 3506 | struct rq *rq = cpu_rq(cpu); |
7835b98b CL |
3507 | |
3508 | update_cpu_clock(p, rq, now); | |
3509 | ||
bdecea3a | 3510 | if (!idle_at_tick) |
7835b98b | 3511 | task_running_tick(rq, p); |
e418e1c2 | 3512 | #ifdef CONFIG_SMP |
7835b98b | 3513 | update_load(rq); |
bdecea3a | 3514 | rq->idle_at_tick = idle_at_tick; |
46cb4b7c | 3515 | trigger_load_balance(cpu); |
e418e1c2 | 3516 | #endif |
1da177e4 LT |
3517 | } |
3518 | ||
1da177e4 LT |
3519 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3520 | ||
3521 | void fastcall add_preempt_count(int val) | |
3522 | { | |
3523 | /* | |
3524 | * Underflow? | |
3525 | */ | |
9a11b49a IM |
3526 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3527 | return; | |
1da177e4 LT |
3528 | preempt_count() += val; |
3529 | /* | |
3530 | * Spinlock count overflowing soon? | |
3531 | */ | |
33859f7f MOS |
3532 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3533 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3534 | } |
3535 | EXPORT_SYMBOL(add_preempt_count); | |
3536 | ||
3537 | void fastcall sub_preempt_count(int val) | |
3538 | { | |
3539 | /* | |
3540 | * Underflow? | |
3541 | */ | |
9a11b49a IM |
3542 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3543 | return; | |
1da177e4 LT |
3544 | /* |
3545 | * Is the spinlock portion underflowing? | |
3546 | */ | |
9a11b49a IM |
3547 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3548 | !(preempt_count() & PREEMPT_MASK))) | |
3549 | return; | |
3550 | ||
1da177e4 LT |
3551 | preempt_count() -= val; |
3552 | } | |
3553 | EXPORT_SYMBOL(sub_preempt_count); | |
3554 | ||
3555 | #endif | |
3556 | ||
3dee386e CK |
3557 | static inline int interactive_sleep(enum sleep_type sleep_type) |
3558 | { | |
3559 | return (sleep_type == SLEEP_INTERACTIVE || | |
3560 | sleep_type == SLEEP_INTERRUPTED); | |
3561 | } | |
3562 | ||
1da177e4 LT |
3563 | /* |
3564 | * schedule() is the main scheduler function. | |
3565 | */ | |
3566 | asmlinkage void __sched schedule(void) | |
3567 | { | |
36c8b586 | 3568 | struct task_struct *prev, *next; |
70b97a7f | 3569 | struct prio_array *array; |
1da177e4 LT |
3570 | struct list_head *queue; |
3571 | unsigned long long now; | |
3572 | unsigned long run_time; | |
a3464a10 | 3573 | int cpu, idx, new_prio; |
48f24c4d | 3574 | long *switch_count; |
70b97a7f | 3575 | struct rq *rq; |
1da177e4 LT |
3576 | |
3577 | /* | |
3578 | * Test if we are atomic. Since do_exit() needs to call into | |
3579 | * schedule() atomically, we ignore that path for now. | |
3580 | * Otherwise, whine if we are scheduling when we should not be. | |
3581 | */ | |
77e4bfbc AM |
3582 | if (unlikely(in_atomic() && !current->exit_state)) { |
3583 | printk(KERN_ERR "BUG: scheduling while atomic: " | |
3584 | "%s/0x%08x/%d\n", | |
3585 | current->comm, preempt_count(), current->pid); | |
a4c410f0 | 3586 | debug_show_held_locks(current); |
3117df04 IM |
3587 | if (irqs_disabled()) |
3588 | print_irqtrace_events(current); | |
77e4bfbc | 3589 | dump_stack(); |
1da177e4 LT |
3590 | } |
3591 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); | |
3592 | ||
3593 | need_resched: | |
3594 | preempt_disable(); | |
3595 | prev = current; | |
3596 | release_kernel_lock(prev); | |
3597 | need_resched_nonpreemptible: | |
3598 | rq = this_rq(); | |
3599 | ||
3600 | /* | |
3601 | * The idle thread is not allowed to schedule! | |
3602 | * Remove this check after it has been exercised a bit. | |
3603 | */ | |
3604 | if (unlikely(prev == rq->idle) && prev->state != TASK_RUNNING) { | |
3605 | printk(KERN_ERR "bad: scheduling from the idle thread!\n"); | |
3606 | dump_stack(); | |
3607 | } | |
3608 | ||
3609 | schedstat_inc(rq, sched_cnt); | |
3610 | now = sched_clock(); | |
238628ed | 3611 | if (likely((long long)(now - prev->timestamp) < NS_MAX_SLEEP_AVG)) { |
1da177e4 | 3612 | run_time = now - prev->timestamp; |
238628ed | 3613 | if (unlikely((long long)(now - prev->timestamp) < 0)) |
1da177e4 LT |
3614 | run_time = 0; |
3615 | } else | |
3616 | run_time = NS_MAX_SLEEP_AVG; | |
3617 | ||
3618 | /* | |
3619 | * Tasks charged proportionately less run_time at high sleep_avg to | |
3620 | * delay them losing their interactive status | |
3621 | */ | |
3622 | run_time /= (CURRENT_BONUS(prev) ? : 1); | |
3623 | ||
3624 | spin_lock_irq(&rq->lock); | |
3625 | ||
1da177e4 LT |
3626 | switch_count = &prev->nivcsw; |
3627 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { | |
3628 | switch_count = &prev->nvcsw; | |
3629 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && | |
3630 | unlikely(signal_pending(prev)))) | |
3631 | prev->state = TASK_RUNNING; | |
3632 | else { | |
3633 | if (prev->state == TASK_UNINTERRUPTIBLE) | |
3634 | rq->nr_uninterruptible++; | |
3635 | deactivate_task(prev, rq); | |
3636 | } | |
3637 | } | |
3638 | ||
3639 | cpu = smp_processor_id(); | |
3640 | if (unlikely(!rq->nr_running)) { | |
1da177e4 LT |
3641 | idle_balance(cpu, rq); |
3642 | if (!rq->nr_running) { | |
3643 | next = rq->idle; | |
3644 | rq->expired_timestamp = 0; | |
1da177e4 LT |
3645 | goto switch_tasks; |
3646 | } | |
1da177e4 LT |
3647 | } |
3648 | ||
3649 | array = rq->active; | |
3650 | if (unlikely(!array->nr_active)) { | |
3651 | /* | |
3652 | * Switch the active and expired arrays. | |
3653 | */ | |
3654 | schedstat_inc(rq, sched_switch); | |
3655 | rq->active = rq->expired; | |
3656 | rq->expired = array; | |
3657 | array = rq->active; | |
3658 | rq->expired_timestamp = 0; | |
3659 | rq->best_expired_prio = MAX_PRIO; | |
3660 | } | |
3661 | ||
3662 | idx = sched_find_first_bit(array->bitmap); | |
3663 | queue = array->queue + idx; | |
36c8b586 | 3664 | next = list_entry(queue->next, struct task_struct, run_list); |
1da177e4 | 3665 | |
3dee386e | 3666 | if (!rt_task(next) && interactive_sleep(next->sleep_type)) { |
1da177e4 | 3667 | unsigned long long delta = now - next->timestamp; |
238628ed | 3668 | if (unlikely((long long)(now - next->timestamp) < 0)) |
1da177e4 LT |
3669 | delta = 0; |
3670 | ||
3dee386e | 3671 | if (next->sleep_type == SLEEP_INTERACTIVE) |
1da177e4 LT |
3672 | delta = delta * (ON_RUNQUEUE_WEIGHT * 128 / 100) / 128; |
3673 | ||
3674 | array = next->array; | |
a3464a10 CS |
3675 | new_prio = recalc_task_prio(next, next->timestamp + delta); |
3676 | ||
3677 | if (unlikely(next->prio != new_prio)) { | |
3678 | dequeue_task(next, array); | |
3679 | next->prio = new_prio; | |
3680 | enqueue_task(next, array); | |
7c4bb1f9 | 3681 | } |
1da177e4 | 3682 | } |
3dee386e | 3683 | next->sleep_type = SLEEP_NORMAL; |
1da177e4 LT |
3684 | switch_tasks: |
3685 | if (next == rq->idle) | |
3686 | schedstat_inc(rq, sched_goidle); | |
3687 | prefetch(next); | |
383f2835 | 3688 | prefetch_stack(next); |
1da177e4 LT |
3689 | clear_tsk_need_resched(prev); |
3690 | rcu_qsctr_inc(task_cpu(prev)); | |
3691 | ||
3692 | update_cpu_clock(prev, rq, now); | |
3693 | ||
3694 | prev->sleep_avg -= run_time; | |
3695 | if ((long)prev->sleep_avg <= 0) | |
3696 | prev->sleep_avg = 0; | |
3697 | prev->timestamp = prev->last_ran = now; | |
3698 | ||
3699 | sched_info_switch(prev, next); | |
3700 | if (likely(prev != next)) { | |
c1e16aa2 | 3701 | next->timestamp = next->last_ran = now; |
1da177e4 LT |
3702 | rq->nr_switches++; |
3703 | rq->curr = next; | |
3704 | ++*switch_count; | |
3705 | ||
4866cde0 | 3706 | prepare_task_switch(rq, next); |
1da177e4 LT |
3707 | prev = context_switch(rq, prev, next); |
3708 | barrier(); | |
4866cde0 NP |
3709 | /* |
3710 | * this_rq must be evaluated again because prev may have moved | |
3711 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3712 | * frame will be invalid. | |
3713 | */ | |
3714 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3715 | } else |
3716 | spin_unlock_irq(&rq->lock); | |
3717 | ||
3718 | prev = current; | |
3719 | if (unlikely(reacquire_kernel_lock(prev) < 0)) | |
3720 | goto need_resched_nonpreemptible; | |
3721 | preempt_enable_no_resched(); | |
3722 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3723 | goto need_resched; | |
3724 | } | |
1da177e4 LT |
3725 | EXPORT_SYMBOL(schedule); |
3726 | ||
3727 | #ifdef CONFIG_PREEMPT | |
3728 | /* | |
2ed6e34f | 3729 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3730 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3731 | * occur there and call schedule directly. | |
3732 | */ | |
3733 | asmlinkage void __sched preempt_schedule(void) | |
3734 | { | |
3735 | struct thread_info *ti = current_thread_info(); | |
3736 | #ifdef CONFIG_PREEMPT_BKL | |
3737 | struct task_struct *task = current; | |
3738 | int saved_lock_depth; | |
3739 | #endif | |
3740 | /* | |
3741 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3742 | * we do not want to preempt the current task. Just return.. | |
3743 | */ | |
beed33a8 | 3744 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3745 | return; |
3746 | ||
3747 | need_resched: | |
3748 | add_preempt_count(PREEMPT_ACTIVE); | |
3749 | /* | |
3750 | * We keep the big kernel semaphore locked, but we | |
3751 | * clear ->lock_depth so that schedule() doesnt | |
3752 | * auto-release the semaphore: | |
3753 | */ | |
3754 | #ifdef CONFIG_PREEMPT_BKL | |
3755 | saved_lock_depth = task->lock_depth; | |
3756 | task->lock_depth = -1; | |
3757 | #endif | |
3758 | schedule(); | |
3759 | #ifdef CONFIG_PREEMPT_BKL | |
3760 | task->lock_depth = saved_lock_depth; | |
3761 | #endif | |
3762 | sub_preempt_count(PREEMPT_ACTIVE); | |
3763 | ||
3764 | /* we could miss a preemption opportunity between schedule and now */ | |
3765 | barrier(); | |
3766 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3767 | goto need_resched; | |
3768 | } | |
1da177e4 LT |
3769 | EXPORT_SYMBOL(preempt_schedule); |
3770 | ||
3771 | /* | |
2ed6e34f | 3772 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3773 | * off of irq context. |
3774 | * Note, that this is called and return with irqs disabled. This will | |
3775 | * protect us against recursive calling from irq. | |
3776 | */ | |
3777 | asmlinkage void __sched preempt_schedule_irq(void) | |
3778 | { | |
3779 | struct thread_info *ti = current_thread_info(); | |
3780 | #ifdef CONFIG_PREEMPT_BKL | |
3781 | struct task_struct *task = current; | |
3782 | int saved_lock_depth; | |
3783 | #endif | |
2ed6e34f | 3784 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3785 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3786 | ||
3787 | need_resched: | |
3788 | add_preempt_count(PREEMPT_ACTIVE); | |
3789 | /* | |
3790 | * We keep the big kernel semaphore locked, but we | |
3791 | * clear ->lock_depth so that schedule() doesnt | |
3792 | * auto-release the semaphore: | |
3793 | */ | |
3794 | #ifdef CONFIG_PREEMPT_BKL | |
3795 | saved_lock_depth = task->lock_depth; | |
3796 | task->lock_depth = -1; | |
3797 | #endif | |
3798 | local_irq_enable(); | |
3799 | schedule(); | |
3800 | local_irq_disable(); | |
3801 | #ifdef CONFIG_PREEMPT_BKL | |
3802 | task->lock_depth = saved_lock_depth; | |
3803 | #endif | |
3804 | sub_preempt_count(PREEMPT_ACTIVE); | |
3805 | ||
3806 | /* we could miss a preemption opportunity between schedule and now */ | |
3807 | barrier(); | |
3808 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3809 | goto need_resched; | |
3810 | } | |
3811 | ||
3812 | #endif /* CONFIG_PREEMPT */ | |
3813 | ||
95cdf3b7 IM |
3814 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3815 | void *key) | |
1da177e4 | 3816 | { |
48f24c4d | 3817 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3818 | } |
1da177e4 LT |
3819 | EXPORT_SYMBOL(default_wake_function); |
3820 | ||
3821 | /* | |
3822 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3823 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3824 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3825 | * | |
3826 | * There are circumstances in which we can try to wake a task which has already | |
3827 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3828 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3829 | */ | |
3830 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3831 | int nr_exclusive, int sync, void *key) | |
3832 | { | |
3833 | struct list_head *tmp, *next; | |
3834 | ||
3835 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3836 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3837 | unsigned flags = curr->flags; | |
3838 | ||
1da177e4 | 3839 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3840 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3841 | break; |
3842 | } | |
3843 | } | |
3844 | ||
3845 | /** | |
3846 | * __wake_up - wake up threads blocked on a waitqueue. | |
3847 | * @q: the waitqueue | |
3848 | * @mode: which threads | |
3849 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3850 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3851 | */ |
3852 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3853 | int nr_exclusive, void *key) |
1da177e4 LT |
3854 | { |
3855 | unsigned long flags; | |
3856 | ||
3857 | spin_lock_irqsave(&q->lock, flags); | |
3858 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3859 | spin_unlock_irqrestore(&q->lock, flags); | |
3860 | } | |
1da177e4 LT |
3861 | EXPORT_SYMBOL(__wake_up); |
3862 | ||
3863 | /* | |
3864 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3865 | */ | |
3866 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3867 | { | |
3868 | __wake_up_common(q, mode, 1, 0, NULL); | |
3869 | } | |
3870 | ||
3871 | /** | |
67be2dd1 | 3872 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3873 | * @q: the waitqueue |
3874 | * @mode: which threads | |
3875 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3876 | * | |
3877 | * The sync wakeup differs that the waker knows that it will schedule | |
3878 | * away soon, so while the target thread will be woken up, it will not | |
3879 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3880 | * with each other. This can prevent needless bouncing between CPUs. | |
3881 | * | |
3882 | * On UP it can prevent extra preemption. | |
3883 | */ | |
95cdf3b7 IM |
3884 | void fastcall |
3885 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3886 | { |
3887 | unsigned long flags; | |
3888 | int sync = 1; | |
3889 | ||
3890 | if (unlikely(!q)) | |
3891 | return; | |
3892 | ||
3893 | if (unlikely(!nr_exclusive)) | |
3894 | sync = 0; | |
3895 | ||
3896 | spin_lock_irqsave(&q->lock, flags); | |
3897 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3898 | spin_unlock_irqrestore(&q->lock, flags); | |
3899 | } | |
3900 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3901 | ||
3902 | void fastcall complete(struct completion *x) | |
3903 | { | |
3904 | unsigned long flags; | |
3905 | ||
3906 | spin_lock_irqsave(&x->wait.lock, flags); | |
3907 | x->done++; | |
3908 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3909 | 1, 0, NULL); | |
3910 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3911 | } | |
3912 | EXPORT_SYMBOL(complete); | |
3913 | ||
3914 | void fastcall complete_all(struct completion *x) | |
3915 | { | |
3916 | unsigned long flags; | |
3917 | ||
3918 | spin_lock_irqsave(&x->wait.lock, flags); | |
3919 | x->done += UINT_MAX/2; | |
3920 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3921 | 0, 0, NULL); | |
3922 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3923 | } | |
3924 | EXPORT_SYMBOL(complete_all); | |
3925 | ||
3926 | void fastcall __sched wait_for_completion(struct completion *x) | |
3927 | { | |
3928 | might_sleep(); | |
48f24c4d | 3929 | |
1da177e4 LT |
3930 | spin_lock_irq(&x->wait.lock); |
3931 | if (!x->done) { | |
3932 | DECLARE_WAITQUEUE(wait, current); | |
3933 | ||
3934 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3935 | __add_wait_queue_tail(&x->wait, &wait); | |
3936 | do { | |
3937 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3938 | spin_unlock_irq(&x->wait.lock); | |
3939 | schedule(); | |
3940 | spin_lock_irq(&x->wait.lock); | |
3941 | } while (!x->done); | |
3942 | __remove_wait_queue(&x->wait, &wait); | |
3943 | } | |
3944 | x->done--; | |
3945 | spin_unlock_irq(&x->wait.lock); | |
3946 | } | |
3947 | EXPORT_SYMBOL(wait_for_completion); | |
3948 | ||
3949 | unsigned long fastcall __sched | |
3950 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3951 | { | |
3952 | might_sleep(); | |
3953 | ||
3954 | spin_lock_irq(&x->wait.lock); | |
3955 | if (!x->done) { | |
3956 | DECLARE_WAITQUEUE(wait, current); | |
3957 | ||
3958 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3959 | __add_wait_queue_tail(&x->wait, &wait); | |
3960 | do { | |
3961 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3962 | spin_unlock_irq(&x->wait.lock); | |
3963 | timeout = schedule_timeout(timeout); | |
3964 | spin_lock_irq(&x->wait.lock); | |
3965 | if (!timeout) { | |
3966 | __remove_wait_queue(&x->wait, &wait); | |
3967 | goto out; | |
3968 | } | |
3969 | } while (!x->done); | |
3970 | __remove_wait_queue(&x->wait, &wait); | |
3971 | } | |
3972 | x->done--; | |
3973 | out: | |
3974 | spin_unlock_irq(&x->wait.lock); | |
3975 | return timeout; | |
3976 | } | |
3977 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3978 | ||
3979 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3980 | { | |
3981 | int ret = 0; | |
3982 | ||
3983 | might_sleep(); | |
3984 | ||
3985 | spin_lock_irq(&x->wait.lock); | |
3986 | if (!x->done) { | |
3987 | DECLARE_WAITQUEUE(wait, current); | |
3988 | ||
3989 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3990 | __add_wait_queue_tail(&x->wait, &wait); | |
3991 | do { | |
3992 | if (signal_pending(current)) { | |
3993 | ret = -ERESTARTSYS; | |
3994 | __remove_wait_queue(&x->wait, &wait); | |
3995 | goto out; | |
3996 | } | |
3997 | __set_current_state(TASK_INTERRUPTIBLE); | |
3998 | spin_unlock_irq(&x->wait.lock); | |
3999 | schedule(); | |
4000 | spin_lock_irq(&x->wait.lock); | |
4001 | } while (!x->done); | |
4002 | __remove_wait_queue(&x->wait, &wait); | |
4003 | } | |
4004 | x->done--; | |
4005 | out: | |
4006 | spin_unlock_irq(&x->wait.lock); | |
4007 | ||
4008 | return ret; | |
4009 | } | |
4010 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
4011 | ||
4012 | unsigned long fastcall __sched | |
4013 | wait_for_completion_interruptible_timeout(struct completion *x, | |
4014 | unsigned long timeout) | |
4015 | { | |
4016 | might_sleep(); | |
4017 | ||
4018 | spin_lock_irq(&x->wait.lock); | |
4019 | if (!x->done) { | |
4020 | DECLARE_WAITQUEUE(wait, current); | |
4021 | ||
4022 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4023 | __add_wait_queue_tail(&x->wait, &wait); | |
4024 | do { | |
4025 | if (signal_pending(current)) { | |
4026 | timeout = -ERESTARTSYS; | |
4027 | __remove_wait_queue(&x->wait, &wait); | |
4028 | goto out; | |
4029 | } | |
4030 | __set_current_state(TASK_INTERRUPTIBLE); | |
4031 | spin_unlock_irq(&x->wait.lock); | |
4032 | timeout = schedule_timeout(timeout); | |
4033 | spin_lock_irq(&x->wait.lock); | |
4034 | if (!timeout) { | |
4035 | __remove_wait_queue(&x->wait, &wait); | |
4036 | goto out; | |
4037 | } | |
4038 | } while (!x->done); | |
4039 | __remove_wait_queue(&x->wait, &wait); | |
4040 | } | |
4041 | x->done--; | |
4042 | out: | |
4043 | spin_unlock_irq(&x->wait.lock); | |
4044 | return timeout; | |
4045 | } | |
4046 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
4047 | ||
4048 | ||
4049 | #define SLEEP_ON_VAR \ | |
4050 | unsigned long flags; \ | |
4051 | wait_queue_t wait; \ | |
4052 | init_waitqueue_entry(&wait, current); | |
4053 | ||
4054 | #define SLEEP_ON_HEAD \ | |
4055 | spin_lock_irqsave(&q->lock,flags); \ | |
4056 | __add_wait_queue(q, &wait); \ | |
4057 | spin_unlock(&q->lock); | |
4058 | ||
4059 | #define SLEEP_ON_TAIL \ | |
4060 | spin_lock_irq(&q->lock); \ | |
4061 | __remove_wait_queue(q, &wait); \ | |
4062 | spin_unlock_irqrestore(&q->lock, flags); | |
4063 | ||
4064 | void fastcall __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4065 | { | |
4066 | SLEEP_ON_VAR | |
4067 | ||
4068 | current->state = TASK_INTERRUPTIBLE; | |
4069 | ||
4070 | SLEEP_ON_HEAD | |
4071 | schedule(); | |
4072 | SLEEP_ON_TAIL | |
4073 | } | |
1da177e4 LT |
4074 | EXPORT_SYMBOL(interruptible_sleep_on); |
4075 | ||
95cdf3b7 IM |
4076 | long fastcall __sched |
4077 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
1da177e4 LT |
4078 | { |
4079 | SLEEP_ON_VAR | |
4080 | ||
4081 | current->state = TASK_INTERRUPTIBLE; | |
4082 | ||
4083 | SLEEP_ON_HEAD | |
4084 | timeout = schedule_timeout(timeout); | |
4085 | SLEEP_ON_TAIL | |
4086 | ||
4087 | return timeout; | |
4088 | } | |
1da177e4 LT |
4089 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4090 | ||
4091 | void fastcall __sched sleep_on(wait_queue_head_t *q) | |
4092 | { | |
4093 | SLEEP_ON_VAR | |
4094 | ||
4095 | current->state = TASK_UNINTERRUPTIBLE; | |
4096 | ||
4097 | SLEEP_ON_HEAD | |
4098 | schedule(); | |
4099 | SLEEP_ON_TAIL | |
4100 | } | |
1da177e4 LT |
4101 | EXPORT_SYMBOL(sleep_on); |
4102 | ||
4103 | long fastcall __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
4104 | { | |
4105 | SLEEP_ON_VAR | |
4106 | ||
4107 | current->state = TASK_UNINTERRUPTIBLE; | |
4108 | ||
4109 | SLEEP_ON_HEAD | |
4110 | timeout = schedule_timeout(timeout); | |
4111 | SLEEP_ON_TAIL | |
4112 | ||
4113 | return timeout; | |
4114 | } | |
4115 | ||
4116 | EXPORT_SYMBOL(sleep_on_timeout); | |
4117 | ||
b29739f9 IM |
4118 | #ifdef CONFIG_RT_MUTEXES |
4119 | ||
4120 | /* | |
4121 | * rt_mutex_setprio - set the current priority of a task | |
4122 | * @p: task | |
4123 | * @prio: prio value (kernel-internal form) | |
4124 | * | |
4125 | * This function changes the 'effective' priority of a task. It does | |
4126 | * not touch ->normal_prio like __setscheduler(). | |
4127 | * | |
4128 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4129 | */ | |
36c8b586 | 4130 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4131 | { |
70b97a7f | 4132 | struct prio_array *array; |
b29739f9 | 4133 | unsigned long flags; |
70b97a7f | 4134 | struct rq *rq; |
d5f9f942 | 4135 | int oldprio; |
b29739f9 IM |
4136 | |
4137 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4138 | ||
4139 | rq = task_rq_lock(p, &flags); | |
4140 | ||
d5f9f942 | 4141 | oldprio = p->prio; |
b29739f9 IM |
4142 | array = p->array; |
4143 | if (array) | |
4144 | dequeue_task(p, array); | |
4145 | p->prio = prio; | |
4146 | ||
4147 | if (array) { | |
4148 | /* | |
4149 | * If changing to an RT priority then queue it | |
4150 | * in the active array! | |
4151 | */ | |
4152 | if (rt_task(p)) | |
4153 | array = rq->active; | |
4154 | enqueue_task(p, array); | |
4155 | /* | |
4156 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4157 | * our priority decreased, or if we are not currently running on |
4158 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4159 | */ |
d5f9f942 AM |
4160 | if (task_running(rq, p)) { |
4161 | if (p->prio > oldprio) | |
4162 | resched_task(rq->curr); | |
4163 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
b29739f9 IM |
4164 | resched_task(rq->curr); |
4165 | } | |
4166 | task_rq_unlock(rq, &flags); | |
4167 | } | |
4168 | ||
4169 | #endif | |
4170 | ||
36c8b586 | 4171 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4172 | { |
70b97a7f | 4173 | struct prio_array *array; |
48f24c4d | 4174 | int old_prio, delta; |
1da177e4 | 4175 | unsigned long flags; |
70b97a7f | 4176 | struct rq *rq; |
1da177e4 LT |
4177 | |
4178 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4179 | return; | |
4180 | /* | |
4181 | * We have to be careful, if called from sys_setpriority(), | |
4182 | * the task might be in the middle of scheduling on another CPU. | |
4183 | */ | |
4184 | rq = task_rq_lock(p, &flags); | |
4185 | /* | |
4186 | * The RT priorities are set via sched_setscheduler(), but we still | |
4187 | * allow the 'normal' nice value to be set - but as expected | |
4188 | * it wont have any effect on scheduling until the task is | |
b0a9499c | 4189 | * not SCHED_NORMAL/SCHED_BATCH: |
1da177e4 | 4190 | */ |
b29739f9 | 4191 | if (has_rt_policy(p)) { |
1da177e4 LT |
4192 | p->static_prio = NICE_TO_PRIO(nice); |
4193 | goto out_unlock; | |
4194 | } | |
4195 | array = p->array; | |
2dd73a4f | 4196 | if (array) { |
1da177e4 | 4197 | dequeue_task(p, array); |
2dd73a4f PW |
4198 | dec_raw_weighted_load(rq, p); |
4199 | } | |
1da177e4 | 4200 | |
1da177e4 | 4201 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4202 | set_load_weight(p); |
b29739f9 IM |
4203 | old_prio = p->prio; |
4204 | p->prio = effective_prio(p); | |
4205 | delta = p->prio - old_prio; | |
1da177e4 LT |
4206 | |
4207 | if (array) { | |
4208 | enqueue_task(p, array); | |
2dd73a4f | 4209 | inc_raw_weighted_load(rq, p); |
1da177e4 | 4210 | /* |
d5f9f942 AM |
4211 | * If the task increased its priority or is running and |
4212 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4213 | */ |
d5f9f942 | 4214 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4215 | resched_task(rq->curr); |
4216 | } | |
4217 | out_unlock: | |
4218 | task_rq_unlock(rq, &flags); | |
4219 | } | |
1da177e4 LT |
4220 | EXPORT_SYMBOL(set_user_nice); |
4221 | ||
e43379f1 MM |
4222 | /* |
4223 | * can_nice - check if a task can reduce its nice value | |
4224 | * @p: task | |
4225 | * @nice: nice value | |
4226 | */ | |
36c8b586 | 4227 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4228 | { |
024f4747 MM |
4229 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4230 | int nice_rlim = 20 - nice; | |
48f24c4d | 4231 | |
e43379f1 MM |
4232 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4233 | capable(CAP_SYS_NICE)); | |
4234 | } | |
4235 | ||
1da177e4 LT |
4236 | #ifdef __ARCH_WANT_SYS_NICE |
4237 | ||
4238 | /* | |
4239 | * sys_nice - change the priority of the current process. | |
4240 | * @increment: priority increment | |
4241 | * | |
4242 | * sys_setpriority is a more generic, but much slower function that | |
4243 | * does similar things. | |
4244 | */ | |
4245 | asmlinkage long sys_nice(int increment) | |
4246 | { | |
48f24c4d | 4247 | long nice, retval; |
1da177e4 LT |
4248 | |
4249 | /* | |
4250 | * Setpriority might change our priority at the same moment. | |
4251 | * We don't have to worry. Conceptually one call occurs first | |
4252 | * and we have a single winner. | |
4253 | */ | |
e43379f1 MM |
4254 | if (increment < -40) |
4255 | increment = -40; | |
1da177e4 LT |
4256 | if (increment > 40) |
4257 | increment = 40; | |
4258 | ||
4259 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4260 | if (nice < -20) | |
4261 | nice = -20; | |
4262 | if (nice > 19) | |
4263 | nice = 19; | |
4264 | ||
e43379f1 MM |
4265 | if (increment < 0 && !can_nice(current, nice)) |
4266 | return -EPERM; | |
4267 | ||
1da177e4 LT |
4268 | retval = security_task_setnice(current, nice); |
4269 | if (retval) | |
4270 | return retval; | |
4271 | ||
4272 | set_user_nice(current, nice); | |
4273 | return 0; | |
4274 | } | |
4275 | ||
4276 | #endif | |
4277 | ||
4278 | /** | |
4279 | * task_prio - return the priority value of a given task. | |
4280 | * @p: the task in question. | |
4281 | * | |
4282 | * This is the priority value as seen by users in /proc. | |
4283 | * RT tasks are offset by -200. Normal tasks are centered | |
4284 | * around 0, value goes from -16 to +15. | |
4285 | */ | |
36c8b586 | 4286 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4287 | { |
4288 | return p->prio - MAX_RT_PRIO; | |
4289 | } | |
4290 | ||
4291 | /** | |
4292 | * task_nice - return the nice value of a given task. | |
4293 | * @p: the task in question. | |
4294 | */ | |
36c8b586 | 4295 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4296 | { |
4297 | return TASK_NICE(p); | |
4298 | } | |
1da177e4 | 4299 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4300 | |
4301 | /** | |
4302 | * idle_cpu - is a given cpu idle currently? | |
4303 | * @cpu: the processor in question. | |
4304 | */ | |
4305 | int idle_cpu(int cpu) | |
4306 | { | |
4307 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4308 | } | |
4309 | ||
1da177e4 LT |
4310 | /** |
4311 | * idle_task - return the idle task for a given cpu. | |
4312 | * @cpu: the processor in question. | |
4313 | */ | |
36c8b586 | 4314 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4315 | { |
4316 | return cpu_rq(cpu)->idle; | |
4317 | } | |
4318 | ||
4319 | /** | |
4320 | * find_process_by_pid - find a process with a matching PID value. | |
4321 | * @pid: the pid in question. | |
4322 | */ | |
36c8b586 | 4323 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4324 | { |
4325 | return pid ? find_task_by_pid(pid) : current; | |
4326 | } | |
4327 | ||
4328 | /* Actually do priority change: must hold rq lock. */ | |
4329 | static void __setscheduler(struct task_struct *p, int policy, int prio) | |
4330 | { | |
4331 | BUG_ON(p->array); | |
48f24c4d | 4332 | |
1da177e4 LT |
4333 | p->policy = policy; |
4334 | p->rt_priority = prio; | |
b29739f9 IM |
4335 | p->normal_prio = normal_prio(p); |
4336 | /* we are holding p->pi_lock already */ | |
4337 | p->prio = rt_mutex_getprio(p); | |
4338 | /* | |
4339 | * SCHED_BATCH tasks are treated as perpetual CPU hogs: | |
4340 | */ | |
4341 | if (policy == SCHED_BATCH) | |
4342 | p->sleep_avg = 0; | |
2dd73a4f | 4343 | set_load_weight(p); |
1da177e4 LT |
4344 | } |
4345 | ||
4346 | /** | |
72fd4a35 | 4347 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4348 | * @p: the task in question. |
4349 | * @policy: new policy. | |
4350 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4351 | * |
72fd4a35 | 4352 | * NOTE that the task may be already dead. |
1da177e4 | 4353 | */ |
95cdf3b7 IM |
4354 | int sched_setscheduler(struct task_struct *p, int policy, |
4355 | struct sched_param *param) | |
1da177e4 | 4356 | { |
48f24c4d | 4357 | int retval, oldprio, oldpolicy = -1; |
70b97a7f | 4358 | struct prio_array *array; |
1da177e4 | 4359 | unsigned long flags; |
70b97a7f | 4360 | struct rq *rq; |
1da177e4 | 4361 | |
66e5393a SR |
4362 | /* may grab non-irq protected spin_locks */ |
4363 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4364 | recheck: |
4365 | /* double check policy once rq lock held */ | |
4366 | if (policy < 0) | |
4367 | policy = oldpolicy = p->policy; | |
4368 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
b0a9499c IM |
4369 | policy != SCHED_NORMAL && policy != SCHED_BATCH) |
4370 | return -EINVAL; | |
1da177e4 LT |
4371 | /* |
4372 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
b0a9499c IM |
4373 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL and |
4374 | * SCHED_BATCH is 0. | |
1da177e4 LT |
4375 | */ |
4376 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4377 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4378 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4379 | return -EINVAL; |
57a6f51c | 4380 | if (is_rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4381 | return -EINVAL; |
4382 | ||
37e4ab3f OC |
4383 | /* |
4384 | * Allow unprivileged RT tasks to decrease priority: | |
4385 | */ | |
4386 | if (!capable(CAP_SYS_NICE)) { | |
8dc3e909 ON |
4387 | if (is_rt_policy(policy)) { |
4388 | unsigned long rlim_rtprio; | |
4389 | unsigned long flags; | |
4390 | ||
4391 | if (!lock_task_sighand(p, &flags)) | |
4392 | return -ESRCH; | |
4393 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4394 | unlock_task_sighand(p, &flags); | |
4395 | ||
4396 | /* can't set/change the rt policy */ | |
4397 | if (policy != p->policy && !rlim_rtprio) | |
4398 | return -EPERM; | |
4399 | ||
4400 | /* can't increase priority */ | |
4401 | if (param->sched_priority > p->rt_priority && | |
4402 | param->sched_priority > rlim_rtprio) | |
4403 | return -EPERM; | |
4404 | } | |
5fe1d75f | 4405 | |
37e4ab3f OC |
4406 | /* can't change other user's priorities */ |
4407 | if ((current->euid != p->euid) && | |
4408 | (current->euid != p->uid)) | |
4409 | return -EPERM; | |
4410 | } | |
1da177e4 LT |
4411 | |
4412 | retval = security_task_setscheduler(p, policy, param); | |
4413 | if (retval) | |
4414 | return retval; | |
b29739f9 IM |
4415 | /* |
4416 | * make sure no PI-waiters arrive (or leave) while we are | |
4417 | * changing the priority of the task: | |
4418 | */ | |
4419 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4420 | /* |
4421 | * To be able to change p->policy safely, the apropriate | |
4422 | * runqueue lock must be held. | |
4423 | */ | |
b29739f9 | 4424 | rq = __task_rq_lock(p); |
1da177e4 LT |
4425 | /* recheck policy now with rq lock held */ |
4426 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4427 | policy = oldpolicy = -1; | |
b29739f9 IM |
4428 | __task_rq_unlock(rq); |
4429 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4430 | goto recheck; |
4431 | } | |
4432 | array = p->array; | |
4433 | if (array) | |
4434 | deactivate_task(p, rq); | |
4435 | oldprio = p->prio; | |
4436 | __setscheduler(p, policy, param->sched_priority); | |
4437 | if (array) { | |
4438 | __activate_task(p, rq); | |
4439 | /* | |
4440 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4441 | * our priority decreased, or if we are not currently running on |
4442 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4443 | */ |
d5f9f942 AM |
4444 | if (task_running(rq, p)) { |
4445 | if (p->prio > oldprio) | |
4446 | resched_task(rq->curr); | |
4447 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
1da177e4 LT |
4448 | resched_task(rq->curr); |
4449 | } | |
b29739f9 IM |
4450 | __task_rq_unlock(rq); |
4451 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4452 | ||
95e02ca9 TG |
4453 | rt_mutex_adjust_pi(p); |
4454 | ||
1da177e4 LT |
4455 | return 0; |
4456 | } | |
4457 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4458 | ||
95cdf3b7 IM |
4459 | static int |
4460 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4461 | { |
1da177e4 LT |
4462 | struct sched_param lparam; |
4463 | struct task_struct *p; | |
36c8b586 | 4464 | int retval; |
1da177e4 LT |
4465 | |
4466 | if (!param || pid < 0) | |
4467 | return -EINVAL; | |
4468 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4469 | return -EFAULT; | |
5fe1d75f ON |
4470 | |
4471 | rcu_read_lock(); | |
4472 | retval = -ESRCH; | |
1da177e4 | 4473 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4474 | if (p != NULL) |
4475 | retval = sched_setscheduler(p, policy, &lparam); | |
4476 | rcu_read_unlock(); | |
36c8b586 | 4477 | |
1da177e4 LT |
4478 | return retval; |
4479 | } | |
4480 | ||
4481 | /** | |
4482 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4483 | * @pid: the pid in question. | |
4484 | * @policy: new policy. | |
4485 | * @param: structure containing the new RT priority. | |
4486 | */ | |
4487 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4488 | struct sched_param __user *param) | |
4489 | { | |
c21761f1 JB |
4490 | /* negative values for policy are not valid */ |
4491 | if (policy < 0) | |
4492 | return -EINVAL; | |
4493 | ||
1da177e4 LT |
4494 | return do_sched_setscheduler(pid, policy, param); |
4495 | } | |
4496 | ||
4497 | /** | |
4498 | * sys_sched_setparam - set/change the RT priority of a thread | |
4499 | * @pid: the pid in question. | |
4500 | * @param: structure containing the new RT priority. | |
4501 | */ | |
4502 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4503 | { | |
4504 | return do_sched_setscheduler(pid, -1, param); | |
4505 | } | |
4506 | ||
4507 | /** | |
4508 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4509 | * @pid: the pid in question. | |
4510 | */ | |
4511 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4512 | { | |
36c8b586 | 4513 | struct task_struct *p; |
1da177e4 | 4514 | int retval = -EINVAL; |
1da177e4 LT |
4515 | |
4516 | if (pid < 0) | |
4517 | goto out_nounlock; | |
4518 | ||
4519 | retval = -ESRCH; | |
4520 | read_lock(&tasklist_lock); | |
4521 | p = find_process_by_pid(pid); | |
4522 | if (p) { | |
4523 | retval = security_task_getscheduler(p); | |
4524 | if (!retval) | |
4525 | retval = p->policy; | |
4526 | } | |
4527 | read_unlock(&tasklist_lock); | |
4528 | ||
4529 | out_nounlock: | |
4530 | return retval; | |
4531 | } | |
4532 | ||
4533 | /** | |
4534 | * sys_sched_getscheduler - get the RT priority of a thread | |
4535 | * @pid: the pid in question. | |
4536 | * @param: structure containing the RT priority. | |
4537 | */ | |
4538 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4539 | { | |
4540 | struct sched_param lp; | |
36c8b586 | 4541 | struct task_struct *p; |
1da177e4 | 4542 | int retval = -EINVAL; |
1da177e4 LT |
4543 | |
4544 | if (!param || pid < 0) | |
4545 | goto out_nounlock; | |
4546 | ||
4547 | read_lock(&tasklist_lock); | |
4548 | p = find_process_by_pid(pid); | |
4549 | retval = -ESRCH; | |
4550 | if (!p) | |
4551 | goto out_unlock; | |
4552 | ||
4553 | retval = security_task_getscheduler(p); | |
4554 | if (retval) | |
4555 | goto out_unlock; | |
4556 | ||
4557 | lp.sched_priority = p->rt_priority; | |
4558 | read_unlock(&tasklist_lock); | |
4559 | ||
4560 | /* | |
4561 | * This one might sleep, we cannot do it with a spinlock held ... | |
4562 | */ | |
4563 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4564 | ||
4565 | out_nounlock: | |
4566 | return retval; | |
4567 | ||
4568 | out_unlock: | |
4569 | read_unlock(&tasklist_lock); | |
4570 | return retval; | |
4571 | } | |
4572 | ||
4573 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4574 | { | |
1da177e4 | 4575 | cpumask_t cpus_allowed; |
36c8b586 IM |
4576 | struct task_struct *p; |
4577 | int retval; | |
1da177e4 | 4578 | |
5be9361c | 4579 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4580 | read_lock(&tasklist_lock); |
4581 | ||
4582 | p = find_process_by_pid(pid); | |
4583 | if (!p) { | |
4584 | read_unlock(&tasklist_lock); | |
5be9361c | 4585 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4586 | return -ESRCH; |
4587 | } | |
4588 | ||
4589 | /* | |
4590 | * It is not safe to call set_cpus_allowed with the | |
4591 | * tasklist_lock held. We will bump the task_struct's | |
4592 | * usage count and then drop tasklist_lock. | |
4593 | */ | |
4594 | get_task_struct(p); | |
4595 | read_unlock(&tasklist_lock); | |
4596 | ||
4597 | retval = -EPERM; | |
4598 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4599 | !capable(CAP_SYS_NICE)) | |
4600 | goto out_unlock; | |
4601 | ||
e7834f8f DQ |
4602 | retval = security_task_setscheduler(p, 0, NULL); |
4603 | if (retval) | |
4604 | goto out_unlock; | |
4605 | ||
1da177e4 LT |
4606 | cpus_allowed = cpuset_cpus_allowed(p); |
4607 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4608 | retval = set_cpus_allowed(p, new_mask); | |
4609 | ||
4610 | out_unlock: | |
4611 | put_task_struct(p); | |
5be9361c | 4612 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4613 | return retval; |
4614 | } | |
4615 | ||
4616 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4617 | cpumask_t *new_mask) | |
4618 | { | |
4619 | if (len < sizeof(cpumask_t)) { | |
4620 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4621 | } else if (len > sizeof(cpumask_t)) { | |
4622 | len = sizeof(cpumask_t); | |
4623 | } | |
4624 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4625 | } | |
4626 | ||
4627 | /** | |
4628 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4629 | * @pid: pid of the process | |
4630 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4631 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4632 | */ | |
4633 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4634 | unsigned long __user *user_mask_ptr) | |
4635 | { | |
4636 | cpumask_t new_mask; | |
4637 | int retval; | |
4638 | ||
4639 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4640 | if (retval) | |
4641 | return retval; | |
4642 | ||
4643 | return sched_setaffinity(pid, new_mask); | |
4644 | } | |
4645 | ||
4646 | /* | |
4647 | * Represents all cpu's present in the system | |
4648 | * In systems capable of hotplug, this map could dynamically grow | |
4649 | * as new cpu's are detected in the system via any platform specific | |
4650 | * method, such as ACPI for e.g. | |
4651 | */ | |
4652 | ||
4cef0c61 | 4653 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4654 | EXPORT_SYMBOL(cpu_present_map); |
4655 | ||
4656 | #ifndef CONFIG_SMP | |
4cef0c61 | 4657 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4658 | EXPORT_SYMBOL(cpu_online_map); |
4659 | ||
4cef0c61 | 4660 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4661 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4662 | #endif |
4663 | ||
4664 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4665 | { | |
36c8b586 | 4666 | struct task_struct *p; |
1da177e4 | 4667 | int retval; |
1da177e4 | 4668 | |
5be9361c | 4669 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4670 | read_lock(&tasklist_lock); |
4671 | ||
4672 | retval = -ESRCH; | |
4673 | p = find_process_by_pid(pid); | |
4674 | if (!p) | |
4675 | goto out_unlock; | |
4676 | ||
e7834f8f DQ |
4677 | retval = security_task_getscheduler(p); |
4678 | if (retval) | |
4679 | goto out_unlock; | |
4680 | ||
2f7016d9 | 4681 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4682 | |
4683 | out_unlock: | |
4684 | read_unlock(&tasklist_lock); | |
5be9361c | 4685 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4686 | if (retval) |
4687 | return retval; | |
4688 | ||
4689 | return 0; | |
4690 | } | |
4691 | ||
4692 | /** | |
4693 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4694 | * @pid: pid of the process | |
4695 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4696 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4697 | */ | |
4698 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4699 | unsigned long __user *user_mask_ptr) | |
4700 | { | |
4701 | int ret; | |
4702 | cpumask_t mask; | |
4703 | ||
4704 | if (len < sizeof(cpumask_t)) | |
4705 | return -EINVAL; | |
4706 | ||
4707 | ret = sched_getaffinity(pid, &mask); | |
4708 | if (ret < 0) | |
4709 | return ret; | |
4710 | ||
4711 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4712 | return -EFAULT; | |
4713 | ||
4714 | return sizeof(cpumask_t); | |
4715 | } | |
4716 | ||
4717 | /** | |
4718 | * sys_sched_yield - yield the current processor to other threads. | |
4719 | * | |
72fd4a35 | 4720 | * This function yields the current CPU by moving the calling thread |
1da177e4 LT |
4721 | * to the expired array. If there are no other threads running on this |
4722 | * CPU then this function will return. | |
4723 | */ | |
4724 | asmlinkage long sys_sched_yield(void) | |
4725 | { | |
70b97a7f IM |
4726 | struct rq *rq = this_rq_lock(); |
4727 | struct prio_array *array = current->array, *target = rq->expired; | |
1da177e4 LT |
4728 | |
4729 | schedstat_inc(rq, yld_cnt); | |
4730 | /* | |
4731 | * We implement yielding by moving the task into the expired | |
4732 | * queue. | |
4733 | * | |
4734 | * (special rule: RT tasks will just roundrobin in the active | |
4735 | * array.) | |
4736 | */ | |
4737 | if (rt_task(current)) | |
4738 | target = rq->active; | |
4739 | ||
5927ad78 | 4740 | if (array->nr_active == 1) { |
1da177e4 LT |
4741 | schedstat_inc(rq, yld_act_empty); |
4742 | if (!rq->expired->nr_active) | |
4743 | schedstat_inc(rq, yld_both_empty); | |
4744 | } else if (!rq->expired->nr_active) | |
4745 | schedstat_inc(rq, yld_exp_empty); | |
4746 | ||
4747 | if (array != target) { | |
4748 | dequeue_task(current, array); | |
4749 | enqueue_task(current, target); | |
4750 | } else | |
4751 | /* | |
4752 | * requeue_task is cheaper so perform that if possible. | |
4753 | */ | |
4754 | requeue_task(current, array); | |
4755 | ||
4756 | /* | |
4757 | * Since we are going to call schedule() anyway, there's | |
4758 | * no need to preempt or enable interrupts: | |
4759 | */ | |
4760 | __release(rq->lock); | |
8a25d5de | 4761 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4762 | _raw_spin_unlock(&rq->lock); |
4763 | preempt_enable_no_resched(); | |
4764 | ||
4765 | schedule(); | |
4766 | ||
4767 | return 0; | |
4768 | } | |
4769 | ||
e7b38404 | 4770 | static void __cond_resched(void) |
1da177e4 | 4771 | { |
8e0a43d8 IM |
4772 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4773 | __might_sleep(__FILE__, __LINE__); | |
4774 | #endif | |
5bbcfd90 IM |
4775 | /* |
4776 | * The BKS might be reacquired before we have dropped | |
4777 | * PREEMPT_ACTIVE, which could trigger a second | |
4778 | * cond_resched() call. | |
4779 | */ | |
1da177e4 LT |
4780 | do { |
4781 | add_preempt_count(PREEMPT_ACTIVE); | |
4782 | schedule(); | |
4783 | sub_preempt_count(PREEMPT_ACTIVE); | |
4784 | } while (need_resched()); | |
4785 | } | |
4786 | ||
4787 | int __sched cond_resched(void) | |
4788 | { | |
9414232f IM |
4789 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4790 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4791 | __cond_resched(); |
4792 | return 1; | |
4793 | } | |
4794 | return 0; | |
4795 | } | |
1da177e4 LT |
4796 | EXPORT_SYMBOL(cond_resched); |
4797 | ||
4798 | /* | |
4799 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4800 | * call schedule, and on return reacquire the lock. | |
4801 | * | |
4802 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4803 | * operations here to prevent schedule() from being called twice (once via | |
4804 | * spin_unlock(), once by hand). | |
4805 | */ | |
95cdf3b7 | 4806 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4807 | { |
6df3cecb JK |
4808 | int ret = 0; |
4809 | ||
1da177e4 LT |
4810 | if (need_lockbreak(lock)) { |
4811 | spin_unlock(lock); | |
4812 | cpu_relax(); | |
6df3cecb | 4813 | ret = 1; |
1da177e4 LT |
4814 | spin_lock(lock); |
4815 | } | |
9414232f | 4816 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4817 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4818 | _raw_spin_unlock(lock); |
4819 | preempt_enable_no_resched(); | |
4820 | __cond_resched(); | |
6df3cecb | 4821 | ret = 1; |
1da177e4 | 4822 | spin_lock(lock); |
1da177e4 | 4823 | } |
6df3cecb | 4824 | return ret; |
1da177e4 | 4825 | } |
1da177e4 LT |
4826 | EXPORT_SYMBOL(cond_resched_lock); |
4827 | ||
4828 | int __sched cond_resched_softirq(void) | |
4829 | { | |
4830 | BUG_ON(!in_softirq()); | |
4831 | ||
9414232f | 4832 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4833 | local_bh_enable(); |
1da177e4 LT |
4834 | __cond_resched(); |
4835 | local_bh_disable(); | |
4836 | return 1; | |
4837 | } | |
4838 | return 0; | |
4839 | } | |
1da177e4 LT |
4840 | EXPORT_SYMBOL(cond_resched_softirq); |
4841 | ||
1da177e4 LT |
4842 | /** |
4843 | * yield - yield the current processor to other threads. | |
4844 | * | |
72fd4a35 | 4845 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4846 | * thread runnable and calls sys_sched_yield(). |
4847 | */ | |
4848 | void __sched yield(void) | |
4849 | { | |
4850 | set_current_state(TASK_RUNNING); | |
4851 | sys_sched_yield(); | |
4852 | } | |
1da177e4 LT |
4853 | EXPORT_SYMBOL(yield); |
4854 | ||
4855 | /* | |
4856 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4857 | * that process accounting knows that this is a task in IO wait state. | |
4858 | * | |
4859 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4860 | * has set its backing_dev_info: the queue against which it should throttle) | |
4861 | */ | |
4862 | void __sched io_schedule(void) | |
4863 | { | |
70b97a7f | 4864 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4865 | |
0ff92245 | 4866 | delayacct_blkio_start(); |
1da177e4 LT |
4867 | atomic_inc(&rq->nr_iowait); |
4868 | schedule(); | |
4869 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4870 | delayacct_blkio_end(); |
1da177e4 | 4871 | } |
1da177e4 LT |
4872 | EXPORT_SYMBOL(io_schedule); |
4873 | ||
4874 | long __sched io_schedule_timeout(long timeout) | |
4875 | { | |
70b97a7f | 4876 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4877 | long ret; |
4878 | ||
0ff92245 | 4879 | delayacct_blkio_start(); |
1da177e4 LT |
4880 | atomic_inc(&rq->nr_iowait); |
4881 | ret = schedule_timeout(timeout); | |
4882 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4883 | delayacct_blkio_end(); |
1da177e4 LT |
4884 | return ret; |
4885 | } | |
4886 | ||
4887 | /** | |
4888 | * sys_sched_get_priority_max - return maximum RT priority. | |
4889 | * @policy: scheduling class. | |
4890 | * | |
4891 | * this syscall returns the maximum rt_priority that can be used | |
4892 | * by a given scheduling class. | |
4893 | */ | |
4894 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4895 | { | |
4896 | int ret = -EINVAL; | |
4897 | ||
4898 | switch (policy) { | |
4899 | case SCHED_FIFO: | |
4900 | case SCHED_RR: | |
4901 | ret = MAX_USER_RT_PRIO-1; | |
4902 | break; | |
4903 | case SCHED_NORMAL: | |
b0a9499c | 4904 | case SCHED_BATCH: |
1da177e4 LT |
4905 | ret = 0; |
4906 | break; | |
4907 | } | |
4908 | return ret; | |
4909 | } | |
4910 | ||
4911 | /** | |
4912 | * sys_sched_get_priority_min - return minimum RT priority. | |
4913 | * @policy: scheduling class. | |
4914 | * | |
4915 | * this syscall returns the minimum rt_priority that can be used | |
4916 | * by a given scheduling class. | |
4917 | */ | |
4918 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4919 | { | |
4920 | int ret = -EINVAL; | |
4921 | ||
4922 | switch (policy) { | |
4923 | case SCHED_FIFO: | |
4924 | case SCHED_RR: | |
4925 | ret = 1; | |
4926 | break; | |
4927 | case SCHED_NORMAL: | |
b0a9499c | 4928 | case SCHED_BATCH: |
1da177e4 LT |
4929 | ret = 0; |
4930 | } | |
4931 | return ret; | |
4932 | } | |
4933 | ||
4934 | /** | |
4935 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4936 | * @pid: pid of the process. | |
4937 | * @interval: userspace pointer to the timeslice value. | |
4938 | * | |
4939 | * this syscall writes the default timeslice value of a given process | |
4940 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4941 | */ | |
4942 | asmlinkage | |
4943 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4944 | { | |
36c8b586 | 4945 | struct task_struct *p; |
1da177e4 LT |
4946 | int retval = -EINVAL; |
4947 | struct timespec t; | |
1da177e4 LT |
4948 | |
4949 | if (pid < 0) | |
4950 | goto out_nounlock; | |
4951 | ||
4952 | retval = -ESRCH; | |
4953 | read_lock(&tasklist_lock); | |
4954 | p = find_process_by_pid(pid); | |
4955 | if (!p) | |
4956 | goto out_unlock; | |
4957 | ||
4958 | retval = security_task_getscheduler(p); | |
4959 | if (retval) | |
4960 | goto out_unlock; | |
4961 | ||
b78709cf | 4962 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
1da177e4 LT |
4963 | 0 : task_timeslice(p), &t); |
4964 | read_unlock(&tasklist_lock); | |
4965 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4966 | out_nounlock: | |
4967 | return retval; | |
4968 | out_unlock: | |
4969 | read_unlock(&tasklist_lock); | |
4970 | return retval; | |
4971 | } | |
4972 | ||
2ed6e34f | 4973 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4974 | |
4975 | static void show_task(struct task_struct *p) | |
1da177e4 | 4976 | { |
1da177e4 | 4977 | unsigned long free = 0; |
36c8b586 | 4978 | unsigned state; |
1da177e4 | 4979 | |
1da177e4 | 4980 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4981 | printk("%-13.13s %c", p->comm, |
4982 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
1da177e4 LT |
4983 | #if (BITS_PER_LONG == 32) |
4984 | if (state == TASK_RUNNING) | |
4985 | printk(" running "); | |
4986 | else | |
4987 | printk(" %08lX ", thread_saved_pc(p)); | |
4988 | #else | |
4989 | if (state == TASK_RUNNING) | |
4990 | printk(" running task "); | |
4991 | else | |
4992 | printk(" %016lx ", thread_saved_pc(p)); | |
4993 | #endif | |
4994 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4995 | { | |
10ebffde | 4996 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4997 | while (!*n) |
4998 | n++; | |
10ebffde | 4999 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5000 | } |
5001 | #endif | |
35f6f753 | 5002 | printk("%5lu %5d %6d", free, p->pid, p->parent->pid); |
1da177e4 LT |
5003 | if (!p->mm) |
5004 | printk(" (L-TLB)\n"); | |
5005 | else | |
5006 | printk(" (NOTLB)\n"); | |
5007 | ||
5008 | if (state != TASK_RUNNING) | |
5009 | show_stack(p, NULL); | |
5010 | } | |
5011 | ||
e59e2ae2 | 5012 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5013 | { |
36c8b586 | 5014 | struct task_struct *g, *p; |
1da177e4 LT |
5015 | |
5016 | #if (BITS_PER_LONG == 32) | |
5017 | printk("\n" | |
301827ac CC |
5018 | " free sibling\n"); |
5019 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
5020 | #else |
5021 | printk("\n" | |
301827ac CC |
5022 | " free sibling\n"); |
5023 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
5024 | #endif |
5025 | read_lock(&tasklist_lock); | |
5026 | do_each_thread(g, p) { | |
5027 | /* | |
5028 | * reset the NMI-timeout, listing all files on a slow | |
5029 | * console might take alot of time: | |
5030 | */ | |
5031 | touch_nmi_watchdog(); | |
39bc89fd | 5032 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 5033 | show_task(p); |
1da177e4 LT |
5034 | } while_each_thread(g, p); |
5035 | ||
04c9167f JF |
5036 | touch_all_softlockup_watchdogs(); |
5037 | ||
1da177e4 | 5038 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5039 | /* |
5040 | * Only show locks if all tasks are dumped: | |
5041 | */ | |
5042 | if (state_filter == -1) | |
5043 | debug_show_all_locks(); | |
1da177e4 LT |
5044 | } |
5045 | ||
f340c0d1 IM |
5046 | /** |
5047 | * init_idle - set up an idle thread for a given CPU | |
5048 | * @idle: task in question | |
5049 | * @cpu: cpu the idle task belongs to | |
5050 | * | |
5051 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5052 | * flag, to make booting more robust. | |
5053 | */ | |
5c1e1767 | 5054 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5055 | { |
70b97a7f | 5056 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5057 | unsigned long flags; |
5058 | ||
81c29a85 | 5059 | idle->timestamp = sched_clock(); |
1da177e4 LT |
5060 | idle->sleep_avg = 0; |
5061 | idle->array = NULL; | |
b29739f9 | 5062 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 LT |
5063 | idle->state = TASK_RUNNING; |
5064 | idle->cpus_allowed = cpumask_of_cpu(cpu); | |
5065 | set_task_cpu(idle, cpu); | |
5066 | ||
5067 | spin_lock_irqsave(&rq->lock, flags); | |
5068 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5069 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5070 | idle->oncpu = 1; | |
5071 | #endif | |
1da177e4 LT |
5072 | spin_unlock_irqrestore(&rq->lock, flags); |
5073 | ||
5074 | /* Set the preempt count _outside_ the spinlocks! */ | |
5075 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 5076 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 5077 | #else |
a1261f54 | 5078 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 LT |
5079 | #endif |
5080 | } | |
5081 | ||
5082 | /* | |
5083 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5084 | * indicates which cpus entered this state. This is used | |
5085 | * in the rcu update to wait only for active cpus. For system | |
5086 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5087 | * always be CPU_MASK_NONE. | |
5088 | */ | |
5089 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5090 | ||
5091 | #ifdef CONFIG_SMP | |
5092 | /* | |
5093 | * This is how migration works: | |
5094 | * | |
70b97a7f | 5095 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5096 | * runqueue and wake up that CPU's migration thread. |
5097 | * 2) we down() the locked semaphore => thread blocks. | |
5098 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5099 | * thread off the CPU) | |
5100 | * 4) it gets the migration request and checks whether the migrated | |
5101 | * task is still in the wrong runqueue. | |
5102 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5103 | * it and puts it into the right queue. | |
5104 | * 6) migration thread up()s the semaphore. | |
5105 | * 7) we wake up and the migration is done. | |
5106 | */ | |
5107 | ||
5108 | /* | |
5109 | * Change a given task's CPU affinity. Migrate the thread to a | |
5110 | * proper CPU and schedule it away if the CPU it's executing on | |
5111 | * is removed from the allowed bitmask. | |
5112 | * | |
5113 | * NOTE: the caller must have a valid reference to the task, the | |
5114 | * task must not exit() & deallocate itself prematurely. The | |
5115 | * call is not atomic; no spinlocks may be held. | |
5116 | */ | |
36c8b586 | 5117 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5118 | { |
70b97a7f | 5119 | struct migration_req req; |
1da177e4 | 5120 | unsigned long flags; |
70b97a7f | 5121 | struct rq *rq; |
48f24c4d | 5122 | int ret = 0; |
1da177e4 LT |
5123 | |
5124 | rq = task_rq_lock(p, &flags); | |
5125 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5126 | ret = -EINVAL; | |
5127 | goto out; | |
5128 | } | |
5129 | ||
5130 | p->cpus_allowed = new_mask; | |
5131 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5132 | if (cpu_isset(task_cpu(p), new_mask)) | |
5133 | goto out; | |
5134 | ||
5135 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5136 | /* Need help from migration thread: drop lock and wait. */ | |
5137 | task_rq_unlock(rq, &flags); | |
5138 | wake_up_process(rq->migration_thread); | |
5139 | wait_for_completion(&req.done); | |
5140 | tlb_migrate_finish(p->mm); | |
5141 | return 0; | |
5142 | } | |
5143 | out: | |
5144 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5145 | |
1da177e4 LT |
5146 | return ret; |
5147 | } | |
1da177e4 LT |
5148 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5149 | ||
5150 | /* | |
5151 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5152 | * this because either it can't run here any more (set_cpus_allowed() | |
5153 | * away from this CPU, or CPU going down), or because we're | |
5154 | * attempting to rebalance this task on exec (sched_exec). | |
5155 | * | |
5156 | * So we race with normal scheduler movements, but that's OK, as long | |
5157 | * as the task is no longer on this CPU. | |
efc30814 KK |
5158 | * |
5159 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5160 | */ |
efc30814 | 5161 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5162 | { |
70b97a7f | 5163 | struct rq *rq_dest, *rq_src; |
efc30814 | 5164 | int ret = 0; |
1da177e4 LT |
5165 | |
5166 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5167 | return ret; |
1da177e4 LT |
5168 | |
5169 | rq_src = cpu_rq(src_cpu); | |
5170 | rq_dest = cpu_rq(dest_cpu); | |
5171 | ||
5172 | double_rq_lock(rq_src, rq_dest); | |
5173 | /* Already moved. */ | |
5174 | if (task_cpu(p) != src_cpu) | |
5175 | goto out; | |
5176 | /* Affinity changed (again). */ | |
5177 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5178 | goto out; | |
5179 | ||
5180 | set_task_cpu(p, dest_cpu); | |
5181 | if (p->array) { | |
5182 | /* | |
5183 | * Sync timestamp with rq_dest's before activating. | |
5184 | * The same thing could be achieved by doing this step | |
5185 | * afterwards, and pretending it was a local activate. | |
5186 | * This way is cleaner and logically correct. | |
5187 | */ | |
b18ec803 MG |
5188 | p->timestamp = p->timestamp - rq_src->most_recent_timestamp |
5189 | + rq_dest->most_recent_timestamp; | |
1da177e4 | 5190 | deactivate_task(p, rq_src); |
0a565f79 | 5191 | __activate_task(p, rq_dest); |
1da177e4 LT |
5192 | if (TASK_PREEMPTS_CURR(p, rq_dest)) |
5193 | resched_task(rq_dest->curr); | |
5194 | } | |
efc30814 | 5195 | ret = 1; |
1da177e4 LT |
5196 | out: |
5197 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5198 | return ret; |
1da177e4 LT |
5199 | } |
5200 | ||
5201 | /* | |
5202 | * migration_thread - this is a highprio system thread that performs | |
5203 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5204 | * another runqueue. | |
5205 | */ | |
95cdf3b7 | 5206 | static int migration_thread(void *data) |
1da177e4 | 5207 | { |
1da177e4 | 5208 | int cpu = (long)data; |
70b97a7f | 5209 | struct rq *rq; |
1da177e4 LT |
5210 | |
5211 | rq = cpu_rq(cpu); | |
5212 | BUG_ON(rq->migration_thread != current); | |
5213 | ||
5214 | set_current_state(TASK_INTERRUPTIBLE); | |
5215 | while (!kthread_should_stop()) { | |
70b97a7f | 5216 | struct migration_req *req; |
1da177e4 | 5217 | struct list_head *head; |
1da177e4 | 5218 | |
3e1d1d28 | 5219 | try_to_freeze(); |
1da177e4 LT |
5220 | |
5221 | spin_lock_irq(&rq->lock); | |
5222 | ||
5223 | if (cpu_is_offline(cpu)) { | |
5224 | spin_unlock_irq(&rq->lock); | |
5225 | goto wait_to_die; | |
5226 | } | |
5227 | ||
5228 | if (rq->active_balance) { | |
5229 | active_load_balance(rq, cpu); | |
5230 | rq->active_balance = 0; | |
5231 | } | |
5232 | ||
5233 | head = &rq->migration_queue; | |
5234 | ||
5235 | if (list_empty(head)) { | |
5236 | spin_unlock_irq(&rq->lock); | |
5237 | schedule(); | |
5238 | set_current_state(TASK_INTERRUPTIBLE); | |
5239 | continue; | |
5240 | } | |
70b97a7f | 5241 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5242 | list_del_init(head->next); |
5243 | ||
674311d5 NP |
5244 | spin_unlock(&rq->lock); |
5245 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5246 | local_irq_enable(); | |
1da177e4 LT |
5247 | |
5248 | complete(&req->done); | |
5249 | } | |
5250 | __set_current_state(TASK_RUNNING); | |
5251 | return 0; | |
5252 | ||
5253 | wait_to_die: | |
5254 | /* Wait for kthread_stop */ | |
5255 | set_current_state(TASK_INTERRUPTIBLE); | |
5256 | while (!kthread_should_stop()) { | |
5257 | schedule(); | |
5258 | set_current_state(TASK_INTERRUPTIBLE); | |
5259 | } | |
5260 | __set_current_state(TASK_RUNNING); | |
5261 | return 0; | |
5262 | } | |
5263 | ||
5264 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5265 | /* |
5266 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5267 | * NOTE: interrupts should be disabled by the caller | |
5268 | */ | |
48f24c4d | 5269 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5270 | { |
efc30814 | 5271 | unsigned long flags; |
1da177e4 | 5272 | cpumask_t mask; |
70b97a7f IM |
5273 | struct rq *rq; |
5274 | int dest_cpu; | |
1da177e4 | 5275 | |
efc30814 | 5276 | restart: |
1da177e4 LT |
5277 | /* On same node? */ |
5278 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5279 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5280 | dest_cpu = any_online_cpu(mask); |
5281 | ||
5282 | /* On any allowed CPU? */ | |
5283 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5284 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5285 | |
5286 | /* No more Mr. Nice Guy. */ | |
5287 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5288 | rq = task_rq_lock(p, &flags); |
5289 | cpus_setall(p->cpus_allowed); | |
5290 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5291 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5292 | |
5293 | /* | |
5294 | * Don't tell them about moving exiting tasks or | |
5295 | * kernel threads (both mm NULL), since they never | |
5296 | * leave kernel. | |
5297 | */ | |
48f24c4d | 5298 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5299 | printk(KERN_INFO "process %d (%s) no " |
5300 | "longer affine to cpu%d\n", | |
48f24c4d | 5301 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5302 | } |
48f24c4d | 5303 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5304 | goto restart; |
1da177e4 LT |
5305 | } |
5306 | ||
5307 | /* | |
5308 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5309 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5310 | * for performance reasons the counter is not stricly tracking tasks to | |
5311 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5312 | * to keep the global sum constant after CPU-down: | |
5313 | */ | |
70b97a7f | 5314 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5315 | { |
70b97a7f | 5316 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5317 | unsigned long flags; |
5318 | ||
5319 | local_irq_save(flags); | |
5320 | double_rq_lock(rq_src, rq_dest); | |
5321 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5322 | rq_src->nr_uninterruptible = 0; | |
5323 | double_rq_unlock(rq_src, rq_dest); | |
5324 | local_irq_restore(flags); | |
5325 | } | |
5326 | ||
5327 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5328 | static void migrate_live_tasks(int src_cpu) | |
5329 | { | |
48f24c4d | 5330 | struct task_struct *p, *t; |
1da177e4 LT |
5331 | |
5332 | write_lock_irq(&tasklist_lock); | |
5333 | ||
48f24c4d IM |
5334 | do_each_thread(t, p) { |
5335 | if (p == current) | |
1da177e4 LT |
5336 | continue; |
5337 | ||
48f24c4d IM |
5338 | if (task_cpu(p) == src_cpu) |
5339 | move_task_off_dead_cpu(src_cpu, p); | |
5340 | } while_each_thread(t, p); | |
1da177e4 LT |
5341 | |
5342 | write_unlock_irq(&tasklist_lock); | |
5343 | } | |
5344 | ||
5345 | /* Schedules idle task to be the next runnable task on current CPU. | |
5346 | * It does so by boosting its priority to highest possible and adding it to | |
48f24c4d | 5347 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5348 | */ |
5349 | void sched_idle_next(void) | |
5350 | { | |
48f24c4d | 5351 | int this_cpu = smp_processor_id(); |
70b97a7f | 5352 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5353 | struct task_struct *p = rq->idle; |
5354 | unsigned long flags; | |
5355 | ||
5356 | /* cpu has to be offline */ | |
48f24c4d | 5357 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5358 | |
48f24c4d IM |
5359 | /* |
5360 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5361 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5362 | */ |
5363 | spin_lock_irqsave(&rq->lock, flags); | |
5364 | ||
5365 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
48f24c4d IM |
5366 | |
5367 | /* Add idle task to the _front_ of its priority queue: */ | |
1da177e4 LT |
5368 | __activate_idle_task(p, rq); |
5369 | ||
5370 | spin_unlock_irqrestore(&rq->lock, flags); | |
5371 | } | |
5372 | ||
48f24c4d IM |
5373 | /* |
5374 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5375 | * offline. |
5376 | */ | |
5377 | void idle_task_exit(void) | |
5378 | { | |
5379 | struct mm_struct *mm = current->active_mm; | |
5380 | ||
5381 | BUG_ON(cpu_online(smp_processor_id())); | |
5382 | ||
5383 | if (mm != &init_mm) | |
5384 | switch_mm(mm, &init_mm, current); | |
5385 | mmdrop(mm); | |
5386 | } | |
5387 | ||
054b9108 | 5388 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5389 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5390 | { |
70b97a7f | 5391 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5392 | |
5393 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5394 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5395 | |
5396 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5397 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5398 | |
48f24c4d | 5399 | get_task_struct(p); |
1da177e4 LT |
5400 | |
5401 | /* | |
5402 | * Drop lock around migration; if someone else moves it, | |
5403 | * that's OK. No task can be added to this CPU, so iteration is | |
5404 | * fine. | |
054b9108 | 5405 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5406 | */ |
054b9108 | 5407 | spin_unlock(&rq->lock); |
48f24c4d | 5408 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5409 | spin_lock(&rq->lock); |
1da177e4 | 5410 | |
48f24c4d | 5411 | put_task_struct(p); |
1da177e4 LT |
5412 | } |
5413 | ||
5414 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5415 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5416 | { | |
70b97a7f | 5417 | struct rq *rq = cpu_rq(dead_cpu); |
48f24c4d | 5418 | unsigned int arr, i; |
1da177e4 LT |
5419 | |
5420 | for (arr = 0; arr < 2; arr++) { | |
5421 | for (i = 0; i < MAX_PRIO; i++) { | |
5422 | struct list_head *list = &rq->arrays[arr].queue[i]; | |
48f24c4d | 5423 | |
1da177e4 | 5424 | while (!list_empty(list)) |
36c8b586 IM |
5425 | migrate_dead(dead_cpu, list_entry(list->next, |
5426 | struct task_struct, run_list)); | |
1da177e4 LT |
5427 | } |
5428 | } | |
5429 | } | |
5430 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5431 | ||
5432 | /* | |
5433 | * migration_call - callback that gets triggered when a CPU is added. | |
5434 | * Here we can start up the necessary migration thread for the new CPU. | |
5435 | */ | |
48f24c4d IM |
5436 | static int __cpuinit |
5437 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5438 | { |
1da177e4 | 5439 | struct task_struct *p; |
48f24c4d | 5440 | int cpu = (long)hcpu; |
1da177e4 | 5441 | unsigned long flags; |
70b97a7f | 5442 | struct rq *rq; |
1da177e4 LT |
5443 | |
5444 | switch (action) { | |
5be9361c GS |
5445 | case CPU_LOCK_ACQUIRE: |
5446 | mutex_lock(&sched_hotcpu_mutex); | |
5447 | break; | |
5448 | ||
1da177e4 | 5449 | case CPU_UP_PREPARE: |
8bb78442 | 5450 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
5451 | p = kthread_create(migration_thread, hcpu, "migration/%d",cpu); |
5452 | if (IS_ERR(p)) | |
5453 | return NOTIFY_BAD; | |
5454 | p->flags |= PF_NOFREEZE; | |
5455 | kthread_bind(p, cpu); | |
5456 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5457 | rq = task_rq_lock(p, &flags); | |
5458 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
5459 | task_rq_unlock(rq, &flags); | |
5460 | cpu_rq(cpu)->migration_thread = p; | |
5461 | break; | |
48f24c4d | 5462 | |
1da177e4 | 5463 | case CPU_ONLINE: |
8bb78442 | 5464 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5465 | /* Strictly unneccessary, as first user will wake it. */ |
5466 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5467 | break; | |
48f24c4d | 5468 | |
1da177e4 LT |
5469 | #ifdef CONFIG_HOTPLUG_CPU |
5470 | case CPU_UP_CANCELED: | |
8bb78442 | 5471 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5472 | if (!cpu_rq(cpu)->migration_thread) |
5473 | break; | |
1da177e4 | 5474 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5475 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5476 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5477 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5478 | cpu_rq(cpu)->migration_thread = NULL; | |
5479 | break; | |
48f24c4d | 5480 | |
1da177e4 | 5481 | case CPU_DEAD: |
8bb78442 | 5482 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5483 | migrate_live_tasks(cpu); |
5484 | rq = cpu_rq(cpu); | |
5485 | kthread_stop(rq->migration_thread); | |
5486 | rq->migration_thread = NULL; | |
5487 | /* Idle task back to normal (off runqueue, low prio) */ | |
5488 | rq = task_rq_lock(rq->idle, &flags); | |
5489 | deactivate_task(rq->idle, rq); | |
5490 | rq->idle->static_prio = MAX_PRIO; | |
5491 | __setscheduler(rq->idle, SCHED_NORMAL, 0); | |
5492 | migrate_dead_tasks(cpu); | |
5493 | task_rq_unlock(rq, &flags); | |
5494 | migrate_nr_uninterruptible(rq); | |
5495 | BUG_ON(rq->nr_running != 0); | |
5496 | ||
5497 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5498 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5499 | * the requestors. */ |
5500 | spin_lock_irq(&rq->lock); | |
5501 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5502 | struct migration_req *req; |
5503 | ||
1da177e4 | 5504 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5505 | struct migration_req, list); |
1da177e4 LT |
5506 | list_del_init(&req->list); |
5507 | complete(&req->done); | |
5508 | } | |
5509 | spin_unlock_irq(&rq->lock); | |
5510 | break; | |
5511 | #endif | |
5be9361c GS |
5512 | case CPU_LOCK_RELEASE: |
5513 | mutex_unlock(&sched_hotcpu_mutex); | |
5514 | break; | |
1da177e4 LT |
5515 | } |
5516 | return NOTIFY_OK; | |
5517 | } | |
5518 | ||
5519 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5520 | * happens before everything else. | |
5521 | */ | |
26c2143b | 5522 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5523 | .notifier_call = migration_call, |
5524 | .priority = 10 | |
5525 | }; | |
5526 | ||
5527 | int __init migration_init(void) | |
5528 | { | |
5529 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5530 | int err; |
48f24c4d IM |
5531 | |
5532 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5533 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5534 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5535 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5536 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5537 | |
1da177e4 LT |
5538 | return 0; |
5539 | } | |
5540 | #endif | |
5541 | ||
5542 | #ifdef CONFIG_SMP | |
476f3534 CL |
5543 | |
5544 | /* Number of possible processor ids */ | |
5545 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5546 | EXPORT_SYMBOL(nr_cpu_ids); | |
5547 | ||
1a20ff27 | 5548 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5549 | #ifdef SCHED_DOMAIN_DEBUG |
5550 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5551 | { | |
5552 | int level = 0; | |
5553 | ||
41c7ce9a NP |
5554 | if (!sd) { |
5555 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5556 | return; | |
5557 | } | |
5558 | ||
1da177e4 LT |
5559 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5560 | ||
5561 | do { | |
5562 | int i; | |
5563 | char str[NR_CPUS]; | |
5564 | struct sched_group *group = sd->groups; | |
5565 | cpumask_t groupmask; | |
5566 | ||
5567 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5568 | cpus_clear(groupmask); | |
5569 | ||
5570 | printk(KERN_DEBUG); | |
5571 | for (i = 0; i < level + 1; i++) | |
5572 | printk(" "); | |
5573 | printk("domain %d: ", level); | |
5574 | ||
5575 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5576 | printk("does not load-balance\n"); | |
5577 | if (sd->parent) | |
33859f7f MOS |
5578 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5579 | " has parent"); | |
1da177e4 LT |
5580 | break; |
5581 | } | |
5582 | ||
5583 | printk("span %s\n", str); | |
5584 | ||
5585 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5586 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5587 | "CPU%d\n", cpu); | |
1da177e4 | 5588 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5589 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5590 | " CPU%d\n", cpu); | |
1da177e4 LT |
5591 | |
5592 | printk(KERN_DEBUG); | |
5593 | for (i = 0; i < level + 2; i++) | |
5594 | printk(" "); | |
5595 | printk("groups:"); | |
5596 | do { | |
5597 | if (!group) { | |
5598 | printk("\n"); | |
5599 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5600 | break; | |
5601 | } | |
5602 | ||
5517d86b | 5603 | if (!group->__cpu_power) { |
1da177e4 | 5604 | printk("\n"); |
33859f7f MOS |
5605 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5606 | "set\n"); | |
1da177e4 LT |
5607 | } |
5608 | ||
5609 | if (!cpus_weight(group->cpumask)) { | |
5610 | printk("\n"); | |
5611 | printk(KERN_ERR "ERROR: empty group\n"); | |
5612 | } | |
5613 | ||
5614 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5615 | printk("\n"); | |
5616 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5617 | } | |
5618 | ||
5619 | cpus_or(groupmask, groupmask, group->cpumask); | |
5620 | ||
5621 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5622 | printk(" %s", str); | |
5623 | ||
5624 | group = group->next; | |
5625 | } while (group != sd->groups); | |
5626 | printk("\n"); | |
5627 | ||
5628 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5629 | printk(KERN_ERR "ERROR: groups don't span " |
5630 | "domain->span\n"); | |
1da177e4 LT |
5631 | |
5632 | level++; | |
5633 | sd = sd->parent; | |
33859f7f MOS |
5634 | if (!sd) |
5635 | continue; | |
1da177e4 | 5636 | |
33859f7f MOS |
5637 | if (!cpus_subset(groupmask, sd->span)) |
5638 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5639 | "of domain->span\n"); | |
1da177e4 LT |
5640 | |
5641 | } while (sd); | |
5642 | } | |
5643 | #else | |
48f24c4d | 5644 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5645 | #endif |
5646 | ||
1a20ff27 | 5647 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5648 | { |
5649 | if (cpus_weight(sd->span) == 1) | |
5650 | return 1; | |
5651 | ||
5652 | /* Following flags need at least 2 groups */ | |
5653 | if (sd->flags & (SD_LOAD_BALANCE | | |
5654 | SD_BALANCE_NEWIDLE | | |
5655 | SD_BALANCE_FORK | | |
89c4710e SS |
5656 | SD_BALANCE_EXEC | |
5657 | SD_SHARE_CPUPOWER | | |
5658 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5659 | if (sd->groups != sd->groups->next) |
5660 | return 0; | |
5661 | } | |
5662 | ||
5663 | /* Following flags don't use groups */ | |
5664 | if (sd->flags & (SD_WAKE_IDLE | | |
5665 | SD_WAKE_AFFINE | | |
5666 | SD_WAKE_BALANCE)) | |
5667 | return 0; | |
5668 | ||
5669 | return 1; | |
5670 | } | |
5671 | ||
48f24c4d IM |
5672 | static int |
5673 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5674 | { |
5675 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5676 | ||
5677 | if (sd_degenerate(parent)) | |
5678 | return 1; | |
5679 | ||
5680 | if (!cpus_equal(sd->span, parent->span)) | |
5681 | return 0; | |
5682 | ||
5683 | /* Does parent contain flags not in child? */ | |
5684 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5685 | if (cflags & SD_WAKE_AFFINE) | |
5686 | pflags &= ~SD_WAKE_BALANCE; | |
5687 | /* Flags needing groups don't count if only 1 group in parent */ | |
5688 | if (parent->groups == parent->groups->next) { | |
5689 | pflags &= ~(SD_LOAD_BALANCE | | |
5690 | SD_BALANCE_NEWIDLE | | |
5691 | SD_BALANCE_FORK | | |
89c4710e SS |
5692 | SD_BALANCE_EXEC | |
5693 | SD_SHARE_CPUPOWER | | |
5694 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5695 | } |
5696 | if (~cflags & pflags) | |
5697 | return 0; | |
5698 | ||
5699 | return 1; | |
5700 | } | |
5701 | ||
1da177e4 LT |
5702 | /* |
5703 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5704 | * hold the hotplug lock. | |
5705 | */ | |
9c1cfda2 | 5706 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5707 | { |
70b97a7f | 5708 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5709 | struct sched_domain *tmp; |
5710 | ||
5711 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5712 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5713 | struct sched_domain *parent = tmp->parent; | |
5714 | if (!parent) | |
5715 | break; | |
1a848870 | 5716 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5717 | tmp->parent = parent->parent; |
1a848870 SS |
5718 | if (parent->parent) |
5719 | parent->parent->child = tmp; | |
5720 | } | |
245af2c7 SS |
5721 | } |
5722 | ||
1a848870 | 5723 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5724 | sd = sd->parent; |
1a848870 SS |
5725 | if (sd) |
5726 | sd->child = NULL; | |
5727 | } | |
1da177e4 LT |
5728 | |
5729 | sched_domain_debug(sd, cpu); | |
5730 | ||
674311d5 | 5731 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5732 | } |
5733 | ||
5734 | /* cpus with isolated domains */ | |
67af63a6 | 5735 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5736 | |
5737 | /* Setup the mask of cpus configured for isolated domains */ | |
5738 | static int __init isolated_cpu_setup(char *str) | |
5739 | { | |
5740 | int ints[NR_CPUS], i; | |
5741 | ||
5742 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5743 | cpus_clear(cpu_isolated_map); | |
5744 | for (i = 1; i <= ints[0]; i++) | |
5745 | if (ints[i] < NR_CPUS) | |
5746 | cpu_set(ints[i], cpu_isolated_map); | |
5747 | return 1; | |
5748 | } | |
5749 | ||
5750 | __setup ("isolcpus=", isolated_cpu_setup); | |
5751 | ||
5752 | /* | |
6711cab4 SS |
5753 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5754 | * to a function which identifies what group(along with sched group) a CPU | |
5755 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5756 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5757 | * |
5758 | * init_sched_build_groups will build a circular linked list of the groups | |
5759 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5760 | * and ->cpu_power to 0. | |
5761 | */ | |
a616058b | 5762 | static void |
6711cab4 SS |
5763 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5764 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5765 | struct sched_group **sg)) | |
1da177e4 LT |
5766 | { |
5767 | struct sched_group *first = NULL, *last = NULL; | |
5768 | cpumask_t covered = CPU_MASK_NONE; | |
5769 | int i; | |
5770 | ||
5771 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5772 | struct sched_group *sg; |
5773 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5774 | int j; |
5775 | ||
5776 | if (cpu_isset(i, covered)) | |
5777 | continue; | |
5778 | ||
5779 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5780 | sg->__cpu_power = 0; |
1da177e4 LT |
5781 | |
5782 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5783 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5784 | continue; |
5785 | ||
5786 | cpu_set(j, covered); | |
5787 | cpu_set(j, sg->cpumask); | |
5788 | } | |
5789 | if (!first) | |
5790 | first = sg; | |
5791 | if (last) | |
5792 | last->next = sg; | |
5793 | last = sg; | |
5794 | } | |
5795 | last->next = first; | |
5796 | } | |
5797 | ||
9c1cfda2 | 5798 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5799 | |
9c1cfda2 | 5800 | #ifdef CONFIG_NUMA |
198e2f18 | 5801 | |
9c1cfda2 JH |
5802 | /** |
5803 | * find_next_best_node - find the next node to include in a sched_domain | |
5804 | * @node: node whose sched_domain we're building | |
5805 | * @used_nodes: nodes already in the sched_domain | |
5806 | * | |
5807 | * Find the next node to include in a given scheduling domain. Simply | |
5808 | * finds the closest node not already in the @used_nodes map. | |
5809 | * | |
5810 | * Should use nodemask_t. | |
5811 | */ | |
5812 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5813 | { | |
5814 | int i, n, val, min_val, best_node = 0; | |
5815 | ||
5816 | min_val = INT_MAX; | |
5817 | ||
5818 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5819 | /* Start at @node */ | |
5820 | n = (node + i) % MAX_NUMNODES; | |
5821 | ||
5822 | if (!nr_cpus_node(n)) | |
5823 | continue; | |
5824 | ||
5825 | /* Skip already used nodes */ | |
5826 | if (test_bit(n, used_nodes)) | |
5827 | continue; | |
5828 | ||
5829 | /* Simple min distance search */ | |
5830 | val = node_distance(node, n); | |
5831 | ||
5832 | if (val < min_val) { | |
5833 | min_val = val; | |
5834 | best_node = n; | |
5835 | } | |
5836 | } | |
5837 | ||
5838 | set_bit(best_node, used_nodes); | |
5839 | return best_node; | |
5840 | } | |
5841 | ||
5842 | /** | |
5843 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5844 | * @node: node whose cpumask we're constructing | |
5845 | * @size: number of nodes to include in this span | |
5846 | * | |
5847 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5848 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5849 | * out optimally. | |
5850 | */ | |
5851 | static cpumask_t sched_domain_node_span(int node) | |
5852 | { | |
9c1cfda2 | 5853 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5854 | cpumask_t span, nodemask; |
5855 | int i; | |
9c1cfda2 JH |
5856 | |
5857 | cpus_clear(span); | |
5858 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5859 | ||
5860 | nodemask = node_to_cpumask(node); | |
5861 | cpus_or(span, span, nodemask); | |
5862 | set_bit(node, used_nodes); | |
5863 | ||
5864 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5865 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5866 | |
9c1cfda2 JH |
5867 | nodemask = node_to_cpumask(next_node); |
5868 | cpus_or(span, span, nodemask); | |
5869 | } | |
5870 | ||
5871 | return span; | |
5872 | } | |
5873 | #endif | |
5874 | ||
5c45bf27 | 5875 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5876 | |
9c1cfda2 | 5877 | /* |
48f24c4d | 5878 | * SMT sched-domains: |
9c1cfda2 | 5879 | */ |
1da177e4 LT |
5880 | #ifdef CONFIG_SCHED_SMT |
5881 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5882 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5883 | |
6711cab4 SS |
5884 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5885 | struct sched_group **sg) | |
1da177e4 | 5886 | { |
6711cab4 SS |
5887 | if (sg) |
5888 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5889 | return cpu; |
5890 | } | |
5891 | #endif | |
5892 | ||
48f24c4d IM |
5893 | /* |
5894 | * multi-core sched-domains: | |
5895 | */ | |
1e9f28fa SS |
5896 | #ifdef CONFIG_SCHED_MC |
5897 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5898 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5899 | #endif |
5900 | ||
5901 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5902 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5903 | struct sched_group **sg) | |
1e9f28fa | 5904 | { |
6711cab4 | 5905 | int group; |
a616058b SS |
5906 | cpumask_t mask = cpu_sibling_map[cpu]; |
5907 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5908 | group = first_cpu(mask); |
5909 | if (sg) | |
5910 | *sg = &per_cpu(sched_group_core, group); | |
5911 | return group; | |
1e9f28fa SS |
5912 | } |
5913 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5914 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5915 | struct sched_group **sg) | |
1e9f28fa | 5916 | { |
6711cab4 SS |
5917 | if (sg) |
5918 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5919 | return cpu; |
5920 | } | |
5921 | #endif | |
5922 | ||
1da177e4 | 5923 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5924 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5925 | |
6711cab4 SS |
5926 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5927 | struct sched_group **sg) | |
1da177e4 | 5928 | { |
6711cab4 | 5929 | int group; |
48f24c4d | 5930 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5931 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5932 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5933 | group = first_cpu(mask); |
1e9f28fa | 5934 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5935 | cpumask_t mask = cpu_sibling_map[cpu]; |
5936 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5937 | group = first_cpu(mask); |
1da177e4 | 5938 | #else |
6711cab4 | 5939 | group = cpu; |
1da177e4 | 5940 | #endif |
6711cab4 SS |
5941 | if (sg) |
5942 | *sg = &per_cpu(sched_group_phys, group); | |
5943 | return group; | |
1da177e4 LT |
5944 | } |
5945 | ||
5946 | #ifdef CONFIG_NUMA | |
1da177e4 | 5947 | /* |
9c1cfda2 JH |
5948 | * The init_sched_build_groups can't handle what we want to do with node |
5949 | * groups, so roll our own. Now each node has its own list of groups which | |
5950 | * gets dynamically allocated. | |
1da177e4 | 5951 | */ |
9c1cfda2 | 5952 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5953 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5954 | |
9c1cfda2 | 5955 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5956 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5957 | |
6711cab4 SS |
5958 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5959 | struct sched_group **sg) | |
9c1cfda2 | 5960 | { |
6711cab4 SS |
5961 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5962 | int group; | |
5963 | ||
5964 | cpus_and(nodemask, nodemask, *cpu_map); | |
5965 | group = first_cpu(nodemask); | |
5966 | ||
5967 | if (sg) | |
5968 | *sg = &per_cpu(sched_group_allnodes, group); | |
5969 | return group; | |
1da177e4 | 5970 | } |
6711cab4 | 5971 | |
08069033 SS |
5972 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5973 | { | |
5974 | struct sched_group *sg = group_head; | |
5975 | int j; | |
5976 | ||
5977 | if (!sg) | |
5978 | return; | |
5979 | next_sg: | |
5980 | for_each_cpu_mask(j, sg->cpumask) { | |
5981 | struct sched_domain *sd; | |
5982 | ||
5983 | sd = &per_cpu(phys_domains, j); | |
5984 | if (j != first_cpu(sd->groups->cpumask)) { | |
5985 | /* | |
5986 | * Only add "power" once for each | |
5987 | * physical package. | |
5988 | */ | |
5989 | continue; | |
5990 | } | |
5991 | ||
5517d86b | 5992 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5993 | } |
5994 | sg = sg->next; | |
5995 | if (sg != group_head) | |
5996 | goto next_sg; | |
5997 | } | |
1da177e4 LT |
5998 | #endif |
5999 | ||
a616058b | 6000 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6001 | /* Free memory allocated for various sched_group structures */ |
6002 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6003 | { | |
a616058b | 6004 | int cpu, i; |
51888ca2 SV |
6005 | |
6006 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6007 | struct sched_group **sched_group_nodes |
6008 | = sched_group_nodes_bycpu[cpu]; | |
6009 | ||
51888ca2 SV |
6010 | if (!sched_group_nodes) |
6011 | continue; | |
6012 | ||
6013 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6014 | cpumask_t nodemask = node_to_cpumask(i); | |
6015 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6016 | ||
6017 | cpus_and(nodemask, nodemask, *cpu_map); | |
6018 | if (cpus_empty(nodemask)) | |
6019 | continue; | |
6020 | ||
6021 | if (sg == NULL) | |
6022 | continue; | |
6023 | sg = sg->next; | |
6024 | next_sg: | |
6025 | oldsg = sg; | |
6026 | sg = sg->next; | |
6027 | kfree(oldsg); | |
6028 | if (oldsg != sched_group_nodes[i]) | |
6029 | goto next_sg; | |
6030 | } | |
6031 | kfree(sched_group_nodes); | |
6032 | sched_group_nodes_bycpu[cpu] = NULL; | |
6033 | } | |
51888ca2 | 6034 | } |
a616058b SS |
6035 | #else |
6036 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6037 | { | |
6038 | } | |
6039 | #endif | |
51888ca2 | 6040 | |
89c4710e SS |
6041 | /* |
6042 | * Initialize sched groups cpu_power. | |
6043 | * | |
6044 | * cpu_power indicates the capacity of sched group, which is used while | |
6045 | * distributing the load between different sched groups in a sched domain. | |
6046 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6047 | * there are asymmetries in the topology. If there are asymmetries, group | |
6048 | * having more cpu_power will pickup more load compared to the group having | |
6049 | * less cpu_power. | |
6050 | * | |
6051 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6052 | * the maximum number of tasks a group can handle in the presence of other idle | |
6053 | * or lightly loaded groups in the same sched domain. | |
6054 | */ | |
6055 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6056 | { | |
6057 | struct sched_domain *child; | |
6058 | struct sched_group *group; | |
6059 | ||
6060 | WARN_ON(!sd || !sd->groups); | |
6061 | ||
6062 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6063 | return; | |
6064 | ||
6065 | child = sd->child; | |
6066 | ||
5517d86b ED |
6067 | sd->groups->__cpu_power = 0; |
6068 | ||
89c4710e SS |
6069 | /* |
6070 | * For perf policy, if the groups in child domain share resources | |
6071 | * (for example cores sharing some portions of the cache hierarchy | |
6072 | * or SMT), then set this domain groups cpu_power such that each group | |
6073 | * can handle only one task, when there are other idle groups in the | |
6074 | * same sched domain. | |
6075 | */ | |
6076 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6077 | (child->flags & | |
6078 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6079 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6080 | return; |
6081 | } | |
6082 | ||
89c4710e SS |
6083 | /* |
6084 | * add cpu_power of each child group to this groups cpu_power | |
6085 | */ | |
6086 | group = child->groups; | |
6087 | do { | |
5517d86b | 6088 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6089 | group = group->next; |
6090 | } while (group != child->groups); | |
6091 | } | |
6092 | ||
1da177e4 | 6093 | /* |
1a20ff27 DG |
6094 | * Build sched domains for a given set of cpus and attach the sched domains |
6095 | * to the individual cpus | |
1da177e4 | 6096 | */ |
51888ca2 | 6097 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6098 | { |
6099 | int i; | |
89c4710e | 6100 | struct sched_domain *sd; |
d1b55138 JH |
6101 | #ifdef CONFIG_NUMA |
6102 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6103 | int sd_allnodes = 0; |
d1b55138 JH |
6104 | |
6105 | /* | |
6106 | * Allocate the per-node list of sched groups | |
6107 | */ | |
51888ca2 | 6108 | sched_group_nodes = kzalloc(sizeof(struct sched_group*)*MAX_NUMNODES, |
d3a5aa98 | 6109 | GFP_KERNEL); |
d1b55138 JH |
6110 | if (!sched_group_nodes) { |
6111 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6112 | return -ENOMEM; |
d1b55138 JH |
6113 | } |
6114 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6115 | #endif | |
1da177e4 LT |
6116 | |
6117 | /* | |
1a20ff27 | 6118 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6119 | */ |
1a20ff27 | 6120 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6121 | struct sched_domain *sd = NULL, *p; |
6122 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6123 | ||
1a20ff27 | 6124 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6125 | |
6126 | #ifdef CONFIG_NUMA | |
d1b55138 | 6127 | if (cpus_weight(*cpu_map) |
9c1cfda2 JH |
6128 | > SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { |
6129 | sd = &per_cpu(allnodes_domains, i); | |
6130 | *sd = SD_ALLNODES_INIT; | |
6131 | sd->span = *cpu_map; | |
6711cab4 | 6132 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6133 | p = sd; |
6711cab4 | 6134 | sd_allnodes = 1; |
9c1cfda2 JH |
6135 | } else |
6136 | p = NULL; | |
6137 | ||
1da177e4 | 6138 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6139 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6140 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6141 | sd->parent = p; | |
1a848870 SS |
6142 | if (p) |
6143 | p->child = sd; | |
9c1cfda2 | 6144 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6145 | #endif |
6146 | ||
6147 | p = sd; | |
6148 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6149 | *sd = SD_CPU_INIT; |
6150 | sd->span = nodemask; | |
6151 | sd->parent = p; | |
1a848870 SS |
6152 | if (p) |
6153 | p->child = sd; | |
6711cab4 | 6154 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6155 | |
1e9f28fa SS |
6156 | #ifdef CONFIG_SCHED_MC |
6157 | p = sd; | |
6158 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6159 | *sd = SD_MC_INIT; |
6160 | sd->span = cpu_coregroup_map(i); | |
6161 | cpus_and(sd->span, sd->span, *cpu_map); | |
6162 | sd->parent = p; | |
1a848870 | 6163 | p->child = sd; |
6711cab4 | 6164 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6165 | #endif |
6166 | ||
1da177e4 LT |
6167 | #ifdef CONFIG_SCHED_SMT |
6168 | p = sd; | |
6169 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6170 | *sd = SD_SIBLING_INIT; |
6171 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6172 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6173 | sd->parent = p; |
1a848870 | 6174 | p->child = sd; |
6711cab4 | 6175 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6176 | #endif |
6177 | } | |
6178 | ||
6179 | #ifdef CONFIG_SCHED_SMT | |
6180 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6181 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6182 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6183 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6184 | if (i != first_cpu(this_sibling_map)) |
6185 | continue; | |
6186 | ||
6711cab4 | 6187 | init_sched_build_groups(this_sibling_map, cpu_map, &cpu_to_cpu_group); |
1da177e4 LT |
6188 | } |
6189 | #endif | |
6190 | ||
1e9f28fa SS |
6191 | #ifdef CONFIG_SCHED_MC |
6192 | /* Set up multi-core groups */ | |
6193 | for_each_cpu_mask(i, *cpu_map) { | |
6194 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6195 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6196 | if (i != first_cpu(this_core_map)) | |
6197 | continue; | |
6711cab4 | 6198 | init_sched_build_groups(this_core_map, cpu_map, &cpu_to_core_group); |
1e9f28fa SS |
6199 | } |
6200 | #endif | |
6201 | ||
6202 | ||
1da177e4 LT |
6203 | /* Set up physical groups */ |
6204 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6205 | cpumask_t nodemask = node_to_cpumask(i); | |
6206 | ||
1a20ff27 | 6207 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6208 | if (cpus_empty(nodemask)) |
6209 | continue; | |
6210 | ||
6711cab4 | 6211 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6212 | } |
6213 | ||
6214 | #ifdef CONFIG_NUMA | |
6215 | /* Set up node groups */ | |
6711cab4 SS |
6216 | if (sd_allnodes) |
6217 | init_sched_build_groups(*cpu_map, cpu_map, &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6218 | |
6219 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6220 | /* Set up node groups */ | |
6221 | struct sched_group *sg, *prev; | |
6222 | cpumask_t nodemask = node_to_cpumask(i); | |
6223 | cpumask_t domainspan; | |
6224 | cpumask_t covered = CPU_MASK_NONE; | |
6225 | int j; | |
6226 | ||
6227 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6228 | if (cpus_empty(nodemask)) { |
6229 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6230 | continue; |
d1b55138 | 6231 | } |
9c1cfda2 JH |
6232 | |
6233 | domainspan = sched_domain_node_span(i); | |
6234 | cpus_and(domainspan, domainspan, *cpu_map); | |
6235 | ||
15f0b676 | 6236 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6237 | if (!sg) { |
6238 | printk(KERN_WARNING "Can not alloc domain group for " | |
6239 | "node %d\n", i); | |
6240 | goto error; | |
6241 | } | |
9c1cfda2 JH |
6242 | sched_group_nodes[i] = sg; |
6243 | for_each_cpu_mask(j, nodemask) { | |
6244 | struct sched_domain *sd; | |
6245 | sd = &per_cpu(node_domains, j); | |
6246 | sd->groups = sg; | |
9c1cfda2 | 6247 | } |
5517d86b | 6248 | sg->__cpu_power = 0; |
9c1cfda2 | 6249 | sg->cpumask = nodemask; |
51888ca2 | 6250 | sg->next = sg; |
9c1cfda2 JH |
6251 | cpus_or(covered, covered, nodemask); |
6252 | prev = sg; | |
6253 | ||
6254 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6255 | cpumask_t tmp, notcovered; | |
6256 | int n = (i + j) % MAX_NUMNODES; | |
6257 | ||
6258 | cpus_complement(notcovered, covered); | |
6259 | cpus_and(tmp, notcovered, *cpu_map); | |
6260 | cpus_and(tmp, tmp, domainspan); | |
6261 | if (cpus_empty(tmp)) | |
6262 | break; | |
6263 | ||
6264 | nodemask = node_to_cpumask(n); | |
6265 | cpus_and(tmp, tmp, nodemask); | |
6266 | if (cpus_empty(tmp)) | |
6267 | continue; | |
6268 | ||
15f0b676 SV |
6269 | sg = kmalloc_node(sizeof(struct sched_group), |
6270 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6271 | if (!sg) { |
6272 | printk(KERN_WARNING | |
6273 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6274 | goto error; |
9c1cfda2 | 6275 | } |
5517d86b | 6276 | sg->__cpu_power = 0; |
9c1cfda2 | 6277 | sg->cpumask = tmp; |
51888ca2 | 6278 | sg->next = prev->next; |
9c1cfda2 JH |
6279 | cpus_or(covered, covered, tmp); |
6280 | prev->next = sg; | |
6281 | prev = sg; | |
6282 | } | |
9c1cfda2 | 6283 | } |
1da177e4 LT |
6284 | #endif |
6285 | ||
6286 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6287 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6288 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6289 | sd = &per_cpu(cpu_domains, i); |
89c4710e | 6290 | init_sched_groups_power(i, sd); |
5c45bf27 | 6291 | } |
1da177e4 | 6292 | #endif |
1e9f28fa | 6293 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6294 | for_each_cpu_mask(i, *cpu_map) { |
1e9f28fa | 6295 | sd = &per_cpu(core_domains, i); |
89c4710e | 6296 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6297 | } |
6298 | #endif | |
1e9f28fa | 6299 | |
5c45bf27 | 6300 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6301 | sd = &per_cpu(phys_domains, i); |
89c4710e | 6302 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6303 | } |
6304 | ||
9c1cfda2 | 6305 | #ifdef CONFIG_NUMA |
08069033 SS |
6306 | for (i = 0; i < MAX_NUMNODES; i++) |
6307 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6308 | |
6711cab4 SS |
6309 | if (sd_allnodes) { |
6310 | struct sched_group *sg; | |
f712c0c7 | 6311 | |
6711cab4 | 6312 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6313 | init_numa_sched_groups_power(sg); |
6314 | } | |
9c1cfda2 JH |
6315 | #endif |
6316 | ||
1da177e4 | 6317 | /* Attach the domains */ |
1a20ff27 | 6318 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6319 | struct sched_domain *sd; |
6320 | #ifdef CONFIG_SCHED_SMT | |
6321 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6322 | #elif defined(CONFIG_SCHED_MC) |
6323 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6324 | #else |
6325 | sd = &per_cpu(phys_domains, i); | |
6326 | #endif | |
6327 | cpu_attach_domain(sd, i); | |
6328 | } | |
51888ca2 SV |
6329 | |
6330 | return 0; | |
6331 | ||
a616058b | 6332 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6333 | error: |
6334 | free_sched_groups(cpu_map); | |
6335 | return -ENOMEM; | |
a616058b | 6336 | #endif |
1da177e4 | 6337 | } |
1a20ff27 DG |
6338 | /* |
6339 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6340 | */ | |
51888ca2 | 6341 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6342 | { |
6343 | cpumask_t cpu_default_map; | |
51888ca2 | 6344 | int err; |
1da177e4 | 6345 | |
1a20ff27 DG |
6346 | /* |
6347 | * Setup mask for cpus without special case scheduling requirements. | |
6348 | * For now this just excludes isolated cpus, but could be used to | |
6349 | * exclude other special cases in the future. | |
6350 | */ | |
6351 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6352 | ||
51888ca2 SV |
6353 | err = build_sched_domains(&cpu_default_map); |
6354 | ||
6355 | return err; | |
1a20ff27 DG |
6356 | } |
6357 | ||
6358 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6359 | { |
51888ca2 | 6360 | free_sched_groups(cpu_map); |
9c1cfda2 | 6361 | } |
1da177e4 | 6362 | |
1a20ff27 DG |
6363 | /* |
6364 | * Detach sched domains from a group of cpus specified in cpu_map | |
6365 | * These cpus will now be attached to the NULL domain | |
6366 | */ | |
858119e1 | 6367 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6368 | { |
6369 | int i; | |
6370 | ||
6371 | for_each_cpu_mask(i, *cpu_map) | |
6372 | cpu_attach_domain(NULL, i); | |
6373 | synchronize_sched(); | |
6374 | arch_destroy_sched_domains(cpu_map); | |
6375 | } | |
6376 | ||
6377 | /* | |
6378 | * Partition sched domains as specified by the cpumasks below. | |
6379 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6380 | * waits for a RCU quiescent period, recalculates sched | |
6381 | * domain information and then attaches them back to the | |
6382 | * correct sched domains | |
6383 | * Call with hotplug lock held | |
6384 | */ | |
51888ca2 | 6385 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6386 | { |
6387 | cpumask_t change_map; | |
51888ca2 | 6388 | int err = 0; |
1a20ff27 DG |
6389 | |
6390 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6391 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6392 | cpus_or(change_map, *partition1, *partition2); | |
6393 | ||
6394 | /* Detach sched domains from all of the affected cpus */ | |
6395 | detach_destroy_domains(&change_map); | |
6396 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6397 | err = build_sched_domains(partition1); |
6398 | if (!err && !cpus_empty(*partition2)) | |
6399 | err = build_sched_domains(partition2); | |
6400 | ||
6401 | return err; | |
1a20ff27 DG |
6402 | } |
6403 | ||
5c45bf27 SS |
6404 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6405 | int arch_reinit_sched_domains(void) | |
6406 | { | |
6407 | int err; | |
6408 | ||
5be9361c | 6409 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6410 | detach_destroy_domains(&cpu_online_map); |
6411 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6412 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6413 | |
6414 | return err; | |
6415 | } | |
6416 | ||
6417 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6418 | { | |
6419 | int ret; | |
6420 | ||
6421 | if (buf[0] != '0' && buf[0] != '1') | |
6422 | return -EINVAL; | |
6423 | ||
6424 | if (smt) | |
6425 | sched_smt_power_savings = (buf[0] == '1'); | |
6426 | else | |
6427 | sched_mc_power_savings = (buf[0] == '1'); | |
6428 | ||
6429 | ret = arch_reinit_sched_domains(); | |
6430 | ||
6431 | return ret ? ret : count; | |
6432 | } | |
6433 | ||
6434 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6435 | { | |
6436 | int err = 0; | |
48f24c4d | 6437 | |
5c45bf27 SS |
6438 | #ifdef CONFIG_SCHED_SMT |
6439 | if (smt_capable()) | |
6440 | err = sysfs_create_file(&cls->kset.kobj, | |
6441 | &attr_sched_smt_power_savings.attr); | |
6442 | #endif | |
6443 | #ifdef CONFIG_SCHED_MC | |
6444 | if (!err && mc_capable()) | |
6445 | err = sysfs_create_file(&cls->kset.kobj, | |
6446 | &attr_sched_mc_power_savings.attr); | |
6447 | #endif | |
6448 | return err; | |
6449 | } | |
6450 | #endif | |
6451 | ||
6452 | #ifdef CONFIG_SCHED_MC | |
6453 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6454 | { | |
6455 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6456 | } | |
48f24c4d IM |
6457 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6458 | const char *buf, size_t count) | |
5c45bf27 SS |
6459 | { |
6460 | return sched_power_savings_store(buf, count, 0); | |
6461 | } | |
6462 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6463 | sched_mc_power_savings_store); | |
6464 | #endif | |
6465 | ||
6466 | #ifdef CONFIG_SCHED_SMT | |
6467 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6468 | { | |
6469 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6470 | } | |
48f24c4d IM |
6471 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6472 | const char *buf, size_t count) | |
5c45bf27 SS |
6473 | { |
6474 | return sched_power_savings_store(buf, count, 1); | |
6475 | } | |
6476 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6477 | sched_smt_power_savings_store); | |
6478 | #endif | |
6479 | ||
1da177e4 LT |
6480 | /* |
6481 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6482 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6483 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6484 | * which will prevent rebalancing while the sched domains are recalculated. |
6485 | */ | |
6486 | static int update_sched_domains(struct notifier_block *nfb, | |
6487 | unsigned long action, void *hcpu) | |
6488 | { | |
1da177e4 LT |
6489 | switch (action) { |
6490 | case CPU_UP_PREPARE: | |
8bb78442 | 6491 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6492 | case CPU_DOWN_PREPARE: |
8bb78442 | 6493 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6494 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6495 | return NOTIFY_OK; |
6496 | ||
6497 | case CPU_UP_CANCELED: | |
8bb78442 | 6498 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6499 | case CPU_DOWN_FAILED: |
8bb78442 | 6500 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6501 | case CPU_ONLINE: |
8bb78442 | 6502 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6503 | case CPU_DEAD: |
8bb78442 | 6504 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6505 | /* |
6506 | * Fall through and re-initialise the domains. | |
6507 | */ | |
6508 | break; | |
6509 | default: | |
6510 | return NOTIFY_DONE; | |
6511 | } | |
6512 | ||
6513 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6514 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6515 | |
6516 | return NOTIFY_OK; | |
6517 | } | |
1da177e4 LT |
6518 | |
6519 | void __init sched_init_smp(void) | |
6520 | { | |
5c1e1767 NP |
6521 | cpumask_t non_isolated_cpus; |
6522 | ||
5be9361c | 6523 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6524 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6525 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6526 | if (cpus_empty(non_isolated_cpus)) |
6527 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6528 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6529 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6530 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6531 | |
6532 | /* Move init over to a non-isolated CPU */ | |
6533 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6534 | BUG(); | |
1da177e4 LT |
6535 | } |
6536 | #else | |
6537 | void __init sched_init_smp(void) | |
6538 | { | |
6539 | } | |
6540 | #endif /* CONFIG_SMP */ | |
6541 | ||
6542 | int in_sched_functions(unsigned long addr) | |
6543 | { | |
6544 | /* Linker adds these: start and end of __sched functions */ | |
6545 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6546 | |
1da177e4 LT |
6547 | return in_lock_functions(addr) || |
6548 | (addr >= (unsigned long)__sched_text_start | |
6549 | && addr < (unsigned long)__sched_text_end); | |
6550 | } | |
6551 | ||
6552 | void __init sched_init(void) | |
6553 | { | |
1da177e4 | 6554 | int i, j, k; |
476f3534 | 6555 | int highest_cpu = 0; |
1da177e4 | 6556 | |
0a945022 | 6557 | for_each_possible_cpu(i) { |
70b97a7f IM |
6558 | struct prio_array *array; |
6559 | struct rq *rq; | |
1da177e4 LT |
6560 | |
6561 | rq = cpu_rq(i); | |
6562 | spin_lock_init(&rq->lock); | |
fcb99371 | 6563 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6564 | rq->nr_running = 0; |
1da177e4 LT |
6565 | rq->active = rq->arrays; |
6566 | rq->expired = rq->arrays + 1; | |
6567 | rq->best_expired_prio = MAX_PRIO; | |
6568 | ||
6569 | #ifdef CONFIG_SMP | |
41c7ce9a | 6570 | rq->sd = NULL; |
7897986b NP |
6571 | for (j = 1; j < 3; j++) |
6572 | rq->cpu_load[j] = 0; | |
1da177e4 LT |
6573 | rq->active_balance = 0; |
6574 | rq->push_cpu = 0; | |
0a2966b4 | 6575 | rq->cpu = i; |
1da177e4 LT |
6576 | rq->migration_thread = NULL; |
6577 | INIT_LIST_HEAD(&rq->migration_queue); | |
6578 | #endif | |
6579 | atomic_set(&rq->nr_iowait, 0); | |
6580 | ||
6581 | for (j = 0; j < 2; j++) { | |
6582 | array = rq->arrays + j; | |
6583 | for (k = 0; k < MAX_PRIO; k++) { | |
6584 | INIT_LIST_HEAD(array->queue + k); | |
6585 | __clear_bit(k, array->bitmap); | |
6586 | } | |
6587 | // delimiter for bitsearch | |
6588 | __set_bit(MAX_PRIO, array->bitmap); | |
6589 | } | |
476f3534 | 6590 | highest_cpu = i; |
1da177e4 LT |
6591 | } |
6592 | ||
2dd73a4f | 6593 | set_load_weight(&init_task); |
b50f60ce | 6594 | |
c9819f45 | 6595 | #ifdef CONFIG_SMP |
476f3534 | 6596 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6597 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6598 | #endif | |
6599 | ||
b50f60ce HC |
6600 | #ifdef CONFIG_RT_MUTEXES |
6601 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6602 | #endif | |
6603 | ||
1da177e4 LT |
6604 | /* |
6605 | * The boot idle thread does lazy MMU switching as well: | |
6606 | */ | |
6607 | atomic_inc(&init_mm.mm_count); | |
6608 | enter_lazy_tlb(&init_mm, current); | |
6609 | ||
6610 | /* | |
6611 | * Make us the idle thread. Technically, schedule() should not be | |
6612 | * called from this thread, however somewhere below it might be, | |
6613 | * but because we are the idle thread, we just pick up running again | |
6614 | * when this runqueue becomes "idle". | |
6615 | */ | |
6616 | init_idle(current, smp_processor_id()); | |
6617 | } | |
6618 | ||
6619 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6620 | void __might_sleep(char *file, int line) | |
6621 | { | |
48f24c4d | 6622 | #ifdef in_atomic |
1da177e4 LT |
6623 | static unsigned long prev_jiffy; /* ratelimiting */ |
6624 | ||
6625 | if ((in_atomic() || irqs_disabled()) && | |
6626 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6627 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6628 | return; | |
6629 | prev_jiffy = jiffies; | |
91368d73 | 6630 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6631 | " context at %s:%d\n", file, line); |
6632 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6633 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6634 | debug_show_held_locks(current); |
3117df04 IM |
6635 | if (irqs_disabled()) |
6636 | print_irqtrace_events(current); | |
1da177e4 LT |
6637 | dump_stack(); |
6638 | } | |
6639 | #endif | |
6640 | } | |
6641 | EXPORT_SYMBOL(__might_sleep); | |
6642 | #endif | |
6643 | ||
6644 | #ifdef CONFIG_MAGIC_SYSRQ | |
6645 | void normalize_rt_tasks(void) | |
6646 | { | |
70b97a7f | 6647 | struct prio_array *array; |
a0f98a1c | 6648 | struct task_struct *g, *p; |
1da177e4 | 6649 | unsigned long flags; |
70b97a7f | 6650 | struct rq *rq; |
1da177e4 LT |
6651 | |
6652 | read_lock_irq(&tasklist_lock); | |
a0f98a1c IM |
6653 | |
6654 | do_each_thread(g, p) { | |
1da177e4 LT |
6655 | if (!rt_task(p)) |
6656 | continue; | |
6657 | ||
b29739f9 IM |
6658 | spin_lock_irqsave(&p->pi_lock, flags); |
6659 | rq = __task_rq_lock(p); | |
1da177e4 LT |
6660 | |
6661 | array = p->array; | |
6662 | if (array) | |
6663 | deactivate_task(p, task_rq(p)); | |
6664 | __setscheduler(p, SCHED_NORMAL, 0); | |
6665 | if (array) { | |
6666 | __activate_task(p, task_rq(p)); | |
6667 | resched_task(rq->curr); | |
6668 | } | |
6669 | ||
b29739f9 IM |
6670 | __task_rq_unlock(rq); |
6671 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6672 | } while_each_thread(g, p); |
6673 | ||
1da177e4 LT |
6674 | read_unlock_irq(&tasklist_lock); |
6675 | } | |
6676 | ||
6677 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6678 | |
6679 | #ifdef CONFIG_IA64 | |
6680 | /* | |
6681 | * These functions are only useful for the IA64 MCA handling. | |
6682 | * | |
6683 | * They can only be called when the whole system has been | |
6684 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6685 | * activity can take place. Using them for anything else would | |
6686 | * be a serious bug, and as a result, they aren't even visible | |
6687 | * under any other configuration. | |
6688 | */ | |
6689 | ||
6690 | /** | |
6691 | * curr_task - return the current task for a given cpu. | |
6692 | * @cpu: the processor in question. | |
6693 | * | |
6694 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6695 | */ | |
36c8b586 | 6696 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6697 | { |
6698 | return cpu_curr(cpu); | |
6699 | } | |
6700 | ||
6701 | /** | |
6702 | * set_curr_task - set the current task for a given cpu. | |
6703 | * @cpu: the processor in question. | |
6704 | * @p: the task pointer to set. | |
6705 | * | |
6706 | * Description: This function must only be used when non-maskable interrupts | |
6707 | * are serviced on a separate stack. It allows the architecture to switch the | |
6708 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6709 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6710 | * and caller must save the original value of the current task (see | |
6711 | * curr_task() above) and restore that value before reenabling interrupts and | |
6712 | * re-starting the system. | |
6713 | * | |
6714 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6715 | */ | |
36c8b586 | 6716 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6717 | { |
6718 | cpu_curr(cpu) = p; | |
6719 | } | |
6720 | ||
6721 | #endif |