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 | ||
1da177e4 | 463 | /* |
cc2a73b5 | 464 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 465 | */ |
70b97a7f | 466 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
467 | __acquires(rq->lock) |
468 | { | |
70b97a7f | 469 | struct rq *rq; |
1da177e4 LT |
470 | |
471 | local_irq_disable(); | |
472 | rq = this_rq(); | |
473 | spin_lock(&rq->lock); | |
474 | ||
475 | return rq; | |
476 | } | |
477 | ||
425e0968 | 478 | #include "sched_stats.h" |
1da177e4 LT |
479 | |
480 | /* | |
481 | * Adding/removing a task to/from a priority array: | |
482 | */ | |
70b97a7f | 483 | static void dequeue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
484 | { |
485 | array->nr_active--; | |
486 | list_del(&p->run_list); | |
487 | if (list_empty(array->queue + p->prio)) | |
488 | __clear_bit(p->prio, array->bitmap); | |
489 | } | |
490 | ||
70b97a7f | 491 | static void enqueue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
492 | { |
493 | sched_info_queued(p); | |
494 | list_add_tail(&p->run_list, array->queue + p->prio); | |
495 | __set_bit(p->prio, array->bitmap); | |
496 | array->nr_active++; | |
497 | p->array = array; | |
498 | } | |
499 | ||
500 | /* | |
501 | * Put task to the end of the run list without the overhead of dequeue | |
502 | * followed by enqueue. | |
503 | */ | |
70b97a7f | 504 | static void requeue_task(struct task_struct *p, struct prio_array *array) |
1da177e4 LT |
505 | { |
506 | list_move_tail(&p->run_list, array->queue + p->prio); | |
507 | } | |
508 | ||
70b97a7f IM |
509 | static inline void |
510 | enqueue_task_head(struct task_struct *p, struct prio_array *array) | |
1da177e4 LT |
511 | { |
512 | list_add(&p->run_list, array->queue + p->prio); | |
513 | __set_bit(p->prio, array->bitmap); | |
514 | array->nr_active++; | |
515 | p->array = array; | |
516 | } | |
517 | ||
518 | /* | |
b29739f9 | 519 | * __normal_prio - return the priority that is based on the static |
1da177e4 LT |
520 | * priority but is modified by bonuses/penalties. |
521 | * | |
522 | * We scale the actual sleep average [0 .... MAX_SLEEP_AVG] | |
523 | * into the -5 ... 0 ... +5 bonus/penalty range. | |
524 | * | |
525 | * We use 25% of the full 0...39 priority range so that: | |
526 | * | |
527 | * 1) nice +19 interactive tasks do not preempt nice 0 CPU hogs. | |
528 | * 2) nice -20 CPU hogs do not get preempted by nice 0 tasks. | |
529 | * | |
530 | * Both properties are important to certain workloads. | |
531 | */ | |
b29739f9 | 532 | |
36c8b586 | 533 | static inline int __normal_prio(struct task_struct *p) |
1da177e4 LT |
534 | { |
535 | int bonus, prio; | |
536 | ||
1da177e4 LT |
537 | bonus = CURRENT_BONUS(p) - MAX_BONUS / 2; |
538 | ||
539 | prio = p->static_prio - bonus; | |
540 | if (prio < MAX_RT_PRIO) | |
541 | prio = MAX_RT_PRIO; | |
542 | if (prio > MAX_PRIO-1) | |
543 | prio = MAX_PRIO-1; | |
544 | return prio; | |
545 | } | |
546 | ||
2dd73a4f PW |
547 | /* |
548 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
549 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
550 | * each task makes to its run queue's load is weighted according to its | |
551 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
552 | * scaled version of the new time slice allocation that they receive on time | |
553 | * slice expiry etc. | |
554 | */ | |
555 | ||
556 | /* | |
557 | * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE | |
558 | * If static_prio_timeslice() is ever changed to break this assumption then | |
559 | * this code will need modification | |
560 | */ | |
561 | #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE | |
562 | #define LOAD_WEIGHT(lp) \ | |
563 | (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO) | |
564 | #define PRIO_TO_LOAD_WEIGHT(prio) \ | |
565 | LOAD_WEIGHT(static_prio_timeslice(prio)) | |
566 | #define RTPRIO_TO_LOAD_WEIGHT(rp) \ | |
567 | (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + LOAD_WEIGHT(rp)) | |
568 | ||
36c8b586 | 569 | static void set_load_weight(struct task_struct *p) |
2dd73a4f | 570 | { |
b29739f9 | 571 | if (has_rt_policy(p)) { |
2dd73a4f PW |
572 | #ifdef CONFIG_SMP |
573 | if (p == task_rq(p)->migration_thread) | |
574 | /* | |
575 | * The migration thread does the actual balancing. | |
576 | * Giving its load any weight will skew balancing | |
577 | * adversely. | |
578 | */ | |
579 | p->load_weight = 0; | |
580 | else | |
581 | #endif | |
582 | p->load_weight = RTPRIO_TO_LOAD_WEIGHT(p->rt_priority); | |
583 | } else | |
584 | p->load_weight = PRIO_TO_LOAD_WEIGHT(p->static_prio); | |
585 | } | |
586 | ||
36c8b586 | 587 | static inline void |
70b97a7f | 588 | inc_raw_weighted_load(struct rq *rq, const struct task_struct *p) |
2dd73a4f PW |
589 | { |
590 | rq->raw_weighted_load += p->load_weight; | |
591 | } | |
592 | ||
36c8b586 | 593 | static inline void |
70b97a7f | 594 | dec_raw_weighted_load(struct rq *rq, const struct task_struct *p) |
2dd73a4f PW |
595 | { |
596 | rq->raw_weighted_load -= p->load_weight; | |
597 | } | |
598 | ||
70b97a7f | 599 | static inline void inc_nr_running(struct task_struct *p, struct rq *rq) |
2dd73a4f PW |
600 | { |
601 | rq->nr_running++; | |
602 | inc_raw_weighted_load(rq, p); | |
603 | } | |
604 | ||
70b97a7f | 605 | static inline void dec_nr_running(struct task_struct *p, struct rq *rq) |
2dd73a4f PW |
606 | { |
607 | rq->nr_running--; | |
608 | dec_raw_weighted_load(rq, p); | |
609 | } | |
610 | ||
b29739f9 IM |
611 | /* |
612 | * Calculate the expected normal priority: i.e. priority | |
613 | * without taking RT-inheritance into account. Might be | |
614 | * boosted by interactivity modifiers. Changes upon fork, | |
615 | * setprio syscalls, and whenever the interactivity | |
616 | * estimator recalculates. | |
617 | */ | |
36c8b586 | 618 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
619 | { |
620 | int prio; | |
621 | ||
622 | if (has_rt_policy(p)) | |
623 | prio = MAX_RT_PRIO-1 - p->rt_priority; | |
624 | else | |
625 | prio = __normal_prio(p); | |
626 | return prio; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Calculate the current priority, i.e. the priority | |
631 | * taken into account by the scheduler. This value might | |
632 | * be boosted by RT tasks, or might be boosted by | |
633 | * interactivity modifiers. Will be RT if the task got | |
634 | * RT-boosted. If not then it returns p->normal_prio. | |
635 | */ | |
36c8b586 | 636 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
637 | { |
638 | p->normal_prio = normal_prio(p); | |
639 | /* | |
640 | * If we are RT tasks or we were boosted to RT priority, | |
641 | * keep the priority unchanged. Otherwise, update priority | |
642 | * to the normal priority: | |
643 | */ | |
644 | if (!rt_prio(p->prio)) | |
645 | return p->normal_prio; | |
646 | return p->prio; | |
647 | } | |
648 | ||
1da177e4 LT |
649 | /* |
650 | * __activate_task - move a task to the runqueue. | |
651 | */ | |
70b97a7f | 652 | static void __activate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 653 | { |
70b97a7f | 654 | struct prio_array *target = rq->active; |
d425b274 | 655 | |
f1adad78 | 656 | if (batch_task(p)) |
d425b274 CK |
657 | target = rq->expired; |
658 | enqueue_task(p, target); | |
2dd73a4f | 659 | inc_nr_running(p, rq); |
1da177e4 LT |
660 | } |
661 | ||
662 | /* | |
663 | * __activate_idle_task - move idle task to the _front_ of runqueue. | |
664 | */ | |
70b97a7f | 665 | static inline void __activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 LT |
666 | { |
667 | enqueue_task_head(p, rq->active); | |
2dd73a4f | 668 | inc_nr_running(p, rq); |
1da177e4 LT |
669 | } |
670 | ||
b29739f9 IM |
671 | /* |
672 | * Recalculate p->normal_prio and p->prio after having slept, | |
673 | * updating the sleep-average too: | |
674 | */ | |
36c8b586 | 675 | static int recalc_task_prio(struct task_struct *p, unsigned long long now) |
1da177e4 LT |
676 | { |
677 | /* Caller must always ensure 'now >= p->timestamp' */ | |
72d2854d | 678 | unsigned long sleep_time = now - p->timestamp; |
1da177e4 | 679 | |
d425b274 | 680 | if (batch_task(p)) |
b0a9499c | 681 | sleep_time = 0; |
1da177e4 LT |
682 | |
683 | if (likely(sleep_time > 0)) { | |
684 | /* | |
72d2854d CK |
685 | * This ceiling is set to the lowest priority that would allow |
686 | * a task to be reinserted into the active array on timeslice | |
687 | * completion. | |
1da177e4 | 688 | */ |
72d2854d | 689 | unsigned long ceiling = INTERACTIVE_SLEEP(p); |
e72ff0bb | 690 | |
72d2854d CK |
691 | if (p->mm && sleep_time > ceiling && p->sleep_avg < ceiling) { |
692 | /* | |
693 | * Prevents user tasks from achieving best priority | |
694 | * with one single large enough sleep. | |
695 | */ | |
696 | p->sleep_avg = ceiling; | |
697 | /* | |
698 | * Using INTERACTIVE_SLEEP() as a ceiling places a | |
699 | * nice(0) task 1ms sleep away from promotion, and | |
700 | * gives it 700ms to round-robin with no chance of | |
701 | * being demoted. This is more than generous, so | |
702 | * mark this sleep as non-interactive to prevent the | |
703 | * on-runqueue bonus logic from intervening should | |
704 | * this task not receive cpu immediately. | |
705 | */ | |
706 | p->sleep_type = SLEEP_NONINTERACTIVE; | |
1da177e4 | 707 | } else { |
1da177e4 LT |
708 | /* |
709 | * Tasks waking from uninterruptible sleep are | |
710 | * limited in their sleep_avg rise as they | |
711 | * are likely to be waiting on I/O | |
712 | */ | |
3dee386e | 713 | if (p->sleep_type == SLEEP_NONINTERACTIVE && p->mm) { |
72d2854d | 714 | if (p->sleep_avg >= ceiling) |
1da177e4 LT |
715 | sleep_time = 0; |
716 | else if (p->sleep_avg + sleep_time >= | |
72d2854d CK |
717 | ceiling) { |
718 | p->sleep_avg = ceiling; | |
719 | sleep_time = 0; | |
1da177e4 LT |
720 | } |
721 | } | |
722 | ||
723 | /* | |
724 | * This code gives a bonus to interactive tasks. | |
725 | * | |
726 | * The boost works by updating the 'average sleep time' | |
727 | * value here, based on ->timestamp. The more time a | |
728 | * task spends sleeping, the higher the average gets - | |
729 | * and the higher the priority boost gets as well. | |
730 | */ | |
731 | p->sleep_avg += sleep_time; | |
732 | ||
1da177e4 | 733 | } |
72d2854d CK |
734 | if (p->sleep_avg > NS_MAX_SLEEP_AVG) |
735 | p->sleep_avg = NS_MAX_SLEEP_AVG; | |
1da177e4 LT |
736 | } |
737 | ||
a3464a10 | 738 | return effective_prio(p); |
1da177e4 LT |
739 | } |
740 | ||
741 | /* | |
742 | * activate_task - move a task to the runqueue and do priority recalculation | |
743 | * | |
744 | * Update all the scheduling statistics stuff. (sleep average | |
745 | * calculation, priority modifiers, etc.) | |
746 | */ | |
70b97a7f | 747 | static void activate_task(struct task_struct *p, struct rq *rq, int local) |
1da177e4 LT |
748 | { |
749 | unsigned long long now; | |
750 | ||
62ab616d CK |
751 | if (rt_task(p)) |
752 | goto out; | |
753 | ||
1da177e4 LT |
754 | now = sched_clock(); |
755 | #ifdef CONFIG_SMP | |
756 | if (!local) { | |
757 | /* Compensate for drifting sched_clock */ | |
70b97a7f | 758 | struct rq *this_rq = this_rq(); |
b18ec803 MG |
759 | now = (now - this_rq->most_recent_timestamp) |
760 | + rq->most_recent_timestamp; | |
1da177e4 LT |
761 | } |
762 | #endif | |
763 | ||
ece8a684 IM |
764 | /* |
765 | * Sleep time is in units of nanosecs, so shift by 20 to get a | |
766 | * milliseconds-range estimation of the amount of time that the task | |
767 | * spent sleeping: | |
768 | */ | |
769 | if (unlikely(prof_on == SLEEP_PROFILING)) { | |
770 | if (p->state == TASK_UNINTERRUPTIBLE) | |
771 | profile_hits(SLEEP_PROFILING, (void *)get_wchan(p), | |
772 | (now - p->timestamp) >> 20); | |
773 | } | |
774 | ||
62ab616d | 775 | p->prio = recalc_task_prio(p, now); |
1da177e4 LT |
776 | |
777 | /* | |
778 | * This checks to make sure it's not an uninterruptible task | |
779 | * that is now waking up. | |
780 | */ | |
3dee386e | 781 | if (p->sleep_type == SLEEP_NORMAL) { |
1da177e4 LT |
782 | /* |
783 | * Tasks which were woken up by interrupts (ie. hw events) | |
784 | * are most likely of interactive nature. So we give them | |
785 | * the credit of extending their sleep time to the period | |
786 | * of time they spend on the runqueue, waiting for execution | |
787 | * on a CPU, first time around: | |
788 | */ | |
789 | if (in_interrupt()) | |
3dee386e | 790 | p->sleep_type = SLEEP_INTERRUPTED; |
1da177e4 LT |
791 | else { |
792 | /* | |
793 | * Normal first-time wakeups get a credit too for | |
794 | * on-runqueue time, but it will be weighted down: | |
795 | */ | |
3dee386e | 796 | p->sleep_type = SLEEP_INTERACTIVE; |
1da177e4 LT |
797 | } |
798 | } | |
799 | p->timestamp = now; | |
62ab616d | 800 | out: |
1da177e4 LT |
801 | __activate_task(p, rq); |
802 | } | |
803 | ||
804 | /* | |
805 | * deactivate_task - remove a task from the runqueue. | |
806 | */ | |
70b97a7f | 807 | static void deactivate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 808 | { |
2dd73a4f | 809 | dec_nr_running(p, rq); |
1da177e4 LT |
810 | dequeue_task(p, p->array); |
811 | p->array = NULL; | |
812 | } | |
813 | ||
814 | /* | |
815 | * resched_task - mark a task 'to be rescheduled now'. | |
816 | * | |
817 | * On UP this means the setting of the need_resched flag, on SMP it | |
818 | * might also involve a cross-CPU call to trigger the scheduler on | |
819 | * the target CPU. | |
820 | */ | |
821 | #ifdef CONFIG_SMP | |
495ab9c0 AK |
822 | |
823 | #ifndef tsk_is_polling | |
824 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
825 | #endif | |
826 | ||
36c8b586 | 827 | static void resched_task(struct task_struct *p) |
1da177e4 | 828 | { |
64c7c8f8 | 829 | int cpu; |
1da177e4 LT |
830 | |
831 | assert_spin_locked(&task_rq(p)->lock); | |
832 | ||
64c7c8f8 NP |
833 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
834 | return; | |
835 | ||
836 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
1da177e4 | 837 | |
64c7c8f8 NP |
838 | cpu = task_cpu(p); |
839 | if (cpu == smp_processor_id()) | |
840 | return; | |
841 | ||
495ab9c0 | 842 | /* NEED_RESCHED must be visible before we test polling */ |
64c7c8f8 | 843 | smp_mb(); |
495ab9c0 | 844 | if (!tsk_is_polling(p)) |
64c7c8f8 | 845 | smp_send_reschedule(cpu); |
1da177e4 | 846 | } |
46cb4b7c SS |
847 | |
848 | static void resched_cpu(int cpu) | |
849 | { | |
850 | struct rq *rq = cpu_rq(cpu); | |
851 | unsigned long flags; | |
852 | ||
853 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
854 | return; | |
855 | resched_task(cpu_curr(cpu)); | |
856 | spin_unlock_irqrestore(&rq->lock, flags); | |
857 | } | |
1da177e4 | 858 | #else |
36c8b586 | 859 | static inline void resched_task(struct task_struct *p) |
1da177e4 | 860 | { |
64c7c8f8 | 861 | assert_spin_locked(&task_rq(p)->lock); |
1da177e4 LT |
862 | set_tsk_need_resched(p); |
863 | } | |
864 | #endif | |
865 | ||
866 | /** | |
867 | * task_curr - is this task currently executing on a CPU? | |
868 | * @p: the task in question. | |
869 | */ | |
36c8b586 | 870 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
871 | { |
872 | return cpu_curr(task_cpu(p)) == p; | |
873 | } | |
874 | ||
2dd73a4f PW |
875 | /* Used instead of source_load when we know the type == 0 */ |
876 | unsigned long weighted_cpuload(const int cpu) | |
877 | { | |
878 | return cpu_rq(cpu)->raw_weighted_load; | |
879 | } | |
880 | ||
1da177e4 | 881 | #ifdef CONFIG_SMP |
c65cc870 IM |
882 | |
883 | void set_task_cpu(struct task_struct *p, unsigned int cpu) | |
884 | { | |
885 | task_thread_info(p)->cpu = cpu; | |
886 | } | |
887 | ||
70b97a7f | 888 | struct migration_req { |
1da177e4 | 889 | struct list_head list; |
1da177e4 | 890 | |
36c8b586 | 891 | struct task_struct *task; |
1da177e4 LT |
892 | int dest_cpu; |
893 | ||
1da177e4 | 894 | struct completion done; |
70b97a7f | 895 | }; |
1da177e4 LT |
896 | |
897 | /* | |
898 | * The task's runqueue lock must be held. | |
899 | * Returns true if you have to wait for migration thread. | |
900 | */ | |
36c8b586 | 901 | static int |
70b97a7f | 902 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 903 | { |
70b97a7f | 904 | struct rq *rq = task_rq(p); |
1da177e4 LT |
905 | |
906 | /* | |
907 | * If the task is not on a runqueue (and not running), then | |
908 | * it is sufficient to simply update the task's cpu field. | |
909 | */ | |
910 | if (!p->array && !task_running(rq, p)) { | |
911 | set_task_cpu(p, dest_cpu); | |
912 | return 0; | |
913 | } | |
914 | ||
915 | init_completion(&req->done); | |
1da177e4 LT |
916 | req->task = p; |
917 | req->dest_cpu = dest_cpu; | |
918 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 919 | |
1da177e4 LT |
920 | return 1; |
921 | } | |
922 | ||
923 | /* | |
924 | * wait_task_inactive - wait for a thread to unschedule. | |
925 | * | |
926 | * The caller must ensure that the task *will* unschedule sometime soon, | |
927 | * else this function might spin for a *long* time. This function can't | |
928 | * be called with interrupts off, or it may introduce deadlock with | |
929 | * smp_call_function() if an IPI is sent by the same process we are | |
930 | * waiting to become inactive. | |
931 | */ | |
36c8b586 | 932 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
933 | { |
934 | unsigned long flags; | |
70b97a7f | 935 | struct rq *rq; |
fa490cfd LT |
936 | struct prio_array *array; |
937 | int running; | |
1da177e4 LT |
938 | |
939 | repeat: | |
fa490cfd LT |
940 | /* |
941 | * We do the initial early heuristics without holding | |
942 | * any task-queue locks at all. We'll only try to get | |
943 | * the runqueue lock when things look like they will | |
944 | * work out! | |
945 | */ | |
946 | rq = task_rq(p); | |
947 | ||
948 | /* | |
949 | * If the task is actively running on another CPU | |
950 | * still, just relax and busy-wait without holding | |
951 | * any locks. | |
952 | * | |
953 | * NOTE! Since we don't hold any locks, it's not | |
954 | * even sure that "rq" stays as the right runqueue! | |
955 | * But we don't care, since "task_running()" will | |
956 | * return false if the runqueue has changed and p | |
957 | * is actually now running somewhere else! | |
958 | */ | |
959 | while (task_running(rq, p)) | |
960 | cpu_relax(); | |
961 | ||
962 | /* | |
963 | * Ok, time to look more closely! We need the rq | |
964 | * lock now, to be *sure*. If we're wrong, we'll | |
965 | * just go back and repeat. | |
966 | */ | |
1da177e4 | 967 | rq = task_rq_lock(p, &flags); |
fa490cfd LT |
968 | running = task_running(rq, p); |
969 | array = p->array; | |
970 | task_rq_unlock(rq, &flags); | |
971 | ||
972 | /* | |
973 | * Was it really running after all now that we | |
974 | * checked with the proper locks actually held? | |
975 | * | |
976 | * Oops. Go back and try again.. | |
977 | */ | |
978 | if (unlikely(running)) { | |
1da177e4 | 979 | cpu_relax(); |
1da177e4 LT |
980 | goto repeat; |
981 | } | |
fa490cfd LT |
982 | |
983 | /* | |
984 | * It's not enough that it's not actively running, | |
985 | * it must be off the runqueue _entirely_, and not | |
986 | * preempted! | |
987 | * | |
988 | * So if it wa still runnable (but just not actively | |
989 | * running right now), it's preempted, and we should | |
990 | * yield - it could be a while. | |
991 | */ | |
992 | if (unlikely(array)) { | |
993 | yield(); | |
994 | goto repeat; | |
995 | } | |
996 | ||
997 | /* | |
998 | * Ahh, all good. It wasn't running, and it wasn't | |
999 | * runnable, which means that it will never become | |
1000 | * running in the future either. We're all done! | |
1001 | */ | |
1da177e4 LT |
1002 | } |
1003 | ||
1004 | /*** | |
1005 | * kick_process - kick a running thread to enter/exit the kernel | |
1006 | * @p: the to-be-kicked thread | |
1007 | * | |
1008 | * Cause a process which is running on another CPU to enter | |
1009 | * kernel-mode, without any delay. (to get signals handled.) | |
1010 | * | |
1011 | * NOTE: this function doesnt have to take the runqueue lock, | |
1012 | * because all it wants to ensure is that the remote task enters | |
1013 | * the kernel. If the IPI races and the task has been migrated | |
1014 | * to another CPU then no harm is done and the purpose has been | |
1015 | * achieved as well. | |
1016 | */ | |
36c8b586 | 1017 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1018 | { |
1019 | int cpu; | |
1020 | ||
1021 | preempt_disable(); | |
1022 | cpu = task_cpu(p); | |
1023 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1024 | smp_send_reschedule(cpu); | |
1025 | preempt_enable(); | |
1026 | } | |
1027 | ||
1028 | /* | |
2dd73a4f PW |
1029 | * Return a low guess at the load of a migration-source cpu weighted |
1030 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1031 | * |
1032 | * We want to under-estimate the load of migration sources, to | |
1033 | * balance conservatively. | |
1034 | */ | |
a2000572 | 1035 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1036 | { |
70b97a7f | 1037 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1038 | |
3b0bd9bc | 1039 | if (type == 0) |
2dd73a4f | 1040 | return rq->raw_weighted_load; |
b910472d | 1041 | |
2dd73a4f | 1042 | return min(rq->cpu_load[type-1], rq->raw_weighted_load); |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | /* | |
2dd73a4f PW |
1046 | * Return a high guess at the load of a migration-target cpu weighted |
1047 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1048 | */ |
a2000572 | 1049 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1050 | { |
70b97a7f | 1051 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f | 1052 | |
7897986b | 1053 | if (type == 0) |
2dd73a4f | 1054 | return rq->raw_weighted_load; |
3b0bd9bc | 1055 | |
2dd73a4f PW |
1056 | return max(rq->cpu_load[type-1], rq->raw_weighted_load); |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Return the average load per task on the cpu's run queue | |
1061 | */ | |
1062 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1063 | { | |
70b97a7f | 1064 | struct rq *rq = cpu_rq(cpu); |
2dd73a4f PW |
1065 | unsigned long n = rq->nr_running; |
1066 | ||
48f24c4d | 1067 | return n ? rq->raw_weighted_load / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1068 | } |
1069 | ||
147cbb4b NP |
1070 | /* |
1071 | * find_idlest_group finds and returns the least busy CPU group within the | |
1072 | * domain. | |
1073 | */ | |
1074 | static struct sched_group * | |
1075 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1076 | { | |
1077 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1078 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1079 | int load_idx = sd->forkexec_idx; | |
1080 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1081 | ||
1082 | do { | |
1083 | unsigned long load, avg_load; | |
1084 | int local_group; | |
1085 | int i; | |
1086 | ||
da5a5522 BD |
1087 | /* Skip over this group if it has no CPUs allowed */ |
1088 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1089 | goto nextgroup; | |
1090 | ||
147cbb4b | 1091 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1092 | |
1093 | /* Tally up the load of all CPUs in the group */ | |
1094 | avg_load = 0; | |
1095 | ||
1096 | for_each_cpu_mask(i, group->cpumask) { | |
1097 | /* Bias balancing toward cpus of our domain */ | |
1098 | if (local_group) | |
1099 | load = source_load(i, load_idx); | |
1100 | else | |
1101 | load = target_load(i, load_idx); | |
1102 | ||
1103 | avg_load += load; | |
1104 | } | |
1105 | ||
1106 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1107 | avg_load = sg_div_cpu_power(group, |
1108 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1109 | |
1110 | if (local_group) { | |
1111 | this_load = avg_load; | |
1112 | this = group; | |
1113 | } else if (avg_load < min_load) { | |
1114 | min_load = avg_load; | |
1115 | idlest = group; | |
1116 | } | |
da5a5522 | 1117 | nextgroup: |
147cbb4b NP |
1118 | group = group->next; |
1119 | } while (group != sd->groups); | |
1120 | ||
1121 | if (!idlest || 100*this_load < imbalance*min_load) | |
1122 | return NULL; | |
1123 | return idlest; | |
1124 | } | |
1125 | ||
1126 | /* | |
0feaece9 | 1127 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1128 | */ |
95cdf3b7 IM |
1129 | static int |
1130 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1131 | { |
da5a5522 | 1132 | cpumask_t tmp; |
147cbb4b NP |
1133 | unsigned long load, min_load = ULONG_MAX; |
1134 | int idlest = -1; | |
1135 | int i; | |
1136 | ||
da5a5522 BD |
1137 | /* Traverse only the allowed CPUs */ |
1138 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1139 | ||
1140 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1141 | load = weighted_cpuload(i); |
147cbb4b NP |
1142 | |
1143 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1144 | min_load = load; | |
1145 | idlest = i; | |
1146 | } | |
1147 | } | |
1148 | ||
1149 | return idlest; | |
1150 | } | |
1151 | ||
476d139c NP |
1152 | /* |
1153 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1154 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1155 | * SD_BALANCE_EXEC. | |
1156 | * | |
1157 | * Balance, ie. select the least loaded group. | |
1158 | * | |
1159 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1160 | * | |
1161 | * preempt must be disabled. | |
1162 | */ | |
1163 | static int sched_balance_self(int cpu, int flag) | |
1164 | { | |
1165 | struct task_struct *t = current; | |
1166 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1167 | |
c96d145e | 1168 | for_each_domain(cpu, tmp) { |
5c45bf27 SS |
1169 | /* |
1170 | * If power savings logic is enabled for a domain, stop there. | |
1171 | */ | |
1172 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) | |
1173 | break; | |
476d139c NP |
1174 | if (tmp->flags & flag) |
1175 | sd = tmp; | |
c96d145e | 1176 | } |
476d139c NP |
1177 | |
1178 | while (sd) { | |
1179 | cpumask_t span; | |
1180 | struct sched_group *group; | |
1a848870 SS |
1181 | int new_cpu, weight; |
1182 | ||
1183 | if (!(sd->flags & flag)) { | |
1184 | sd = sd->child; | |
1185 | continue; | |
1186 | } | |
476d139c NP |
1187 | |
1188 | span = sd->span; | |
1189 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1190 | if (!group) { |
1191 | sd = sd->child; | |
1192 | continue; | |
1193 | } | |
476d139c | 1194 | |
da5a5522 | 1195 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1196 | if (new_cpu == -1 || new_cpu == cpu) { |
1197 | /* Now try balancing at a lower domain level of cpu */ | |
1198 | sd = sd->child; | |
1199 | continue; | |
1200 | } | |
476d139c | 1201 | |
1a848870 | 1202 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1203 | cpu = new_cpu; |
476d139c NP |
1204 | sd = NULL; |
1205 | weight = cpus_weight(span); | |
1206 | for_each_domain(cpu, tmp) { | |
1207 | if (weight <= cpus_weight(tmp->span)) | |
1208 | break; | |
1209 | if (tmp->flags & flag) | |
1210 | sd = tmp; | |
1211 | } | |
1212 | /* while loop will break here if sd == NULL */ | |
1213 | } | |
1214 | ||
1215 | return cpu; | |
1216 | } | |
1217 | ||
1218 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1219 | |
1220 | /* | |
1221 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1222 | * not idle and an idle cpu is available. The span of cpus to | |
1223 | * search starts with cpus closest then further out as needed, | |
1224 | * so we always favor a closer, idle cpu. | |
1225 | * | |
1226 | * Returns the CPU we should wake onto. | |
1227 | */ | |
1228 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1229 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1230 | { |
1231 | cpumask_t tmp; | |
1232 | struct sched_domain *sd; | |
1233 | int i; | |
1234 | ||
4953198b SS |
1235 | /* |
1236 | * If it is idle, then it is the best cpu to run this task. | |
1237 | * | |
1238 | * This cpu is also the best, if it has more than one task already. | |
1239 | * Siblings must be also busy(in most cases) as they didn't already | |
1240 | * pickup the extra load from this cpu and hence we need not check | |
1241 | * sibling runqueue info. This will avoid the checks and cache miss | |
1242 | * penalities associated with that. | |
1243 | */ | |
1244 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1245 | return cpu; |
1246 | ||
1247 | for_each_domain(cpu, sd) { | |
1248 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1249 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1250 | for_each_cpu_mask(i, tmp) { |
1251 | if (idle_cpu(i)) | |
1252 | return i; | |
1253 | } | |
1254 | } | |
e0f364f4 NP |
1255 | else |
1256 | break; | |
1da177e4 LT |
1257 | } |
1258 | return cpu; | |
1259 | } | |
1260 | #else | |
36c8b586 | 1261 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1262 | { |
1263 | return cpu; | |
1264 | } | |
1265 | #endif | |
1266 | ||
1267 | /*** | |
1268 | * try_to_wake_up - wake up a thread | |
1269 | * @p: the to-be-woken-up thread | |
1270 | * @state: the mask of task states that can be woken | |
1271 | * @sync: do a synchronous wakeup? | |
1272 | * | |
1273 | * Put it on the run-queue if it's not already there. The "current" | |
1274 | * thread is always on the run-queue (except when the actual | |
1275 | * re-schedule is in progress), and as such you're allowed to do | |
1276 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1277 | * runnable without the overhead of this. | |
1278 | * | |
1279 | * returns failure only if the task is already active. | |
1280 | */ | |
36c8b586 | 1281 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1282 | { |
1283 | int cpu, this_cpu, success = 0; | |
1284 | unsigned long flags; | |
1285 | long old_state; | |
70b97a7f | 1286 | struct rq *rq; |
1da177e4 | 1287 | #ifdef CONFIG_SMP |
7897986b | 1288 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1289 | unsigned long load, this_load; |
1da177e4 LT |
1290 | int new_cpu; |
1291 | #endif | |
1292 | ||
1293 | rq = task_rq_lock(p, &flags); | |
1294 | old_state = p->state; | |
1295 | if (!(old_state & state)) | |
1296 | goto out; | |
1297 | ||
1298 | if (p->array) | |
1299 | goto out_running; | |
1300 | ||
1301 | cpu = task_cpu(p); | |
1302 | this_cpu = smp_processor_id(); | |
1303 | ||
1304 | #ifdef CONFIG_SMP | |
1305 | if (unlikely(task_running(rq, p))) | |
1306 | goto out_activate; | |
1307 | ||
7897986b NP |
1308 | new_cpu = cpu; |
1309 | ||
1da177e4 LT |
1310 | schedstat_inc(rq, ttwu_cnt); |
1311 | if (cpu == this_cpu) { | |
1312 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1313 | goto out_set_cpu; |
1314 | } | |
1315 | ||
1316 | for_each_domain(this_cpu, sd) { | |
1317 | if (cpu_isset(cpu, sd->span)) { | |
1318 | schedstat_inc(sd, ttwu_wake_remote); | |
1319 | this_sd = sd; | |
1320 | break; | |
1da177e4 LT |
1321 | } |
1322 | } | |
1da177e4 | 1323 | |
7897986b | 1324 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1325 | goto out_set_cpu; |
1326 | ||
1da177e4 | 1327 | /* |
7897986b | 1328 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1329 | */ |
7897986b NP |
1330 | if (this_sd) { |
1331 | int idx = this_sd->wake_idx; | |
1332 | unsigned int imbalance; | |
1da177e4 | 1333 | |
a3f21bce NP |
1334 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1335 | ||
7897986b NP |
1336 | load = source_load(cpu, idx); |
1337 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1338 | |
7897986b NP |
1339 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1340 | ||
a3f21bce NP |
1341 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1342 | unsigned long tl = this_load; | |
33859f7f MOS |
1343 | unsigned long tl_per_task; |
1344 | ||
1345 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1346 | |
1da177e4 | 1347 | /* |
a3f21bce NP |
1348 | * If sync wakeup then subtract the (maximum possible) |
1349 | * effect of the currently running task from the load | |
1350 | * of the current CPU: | |
1da177e4 | 1351 | */ |
a3f21bce | 1352 | if (sync) |
2dd73a4f | 1353 | tl -= current->load_weight; |
a3f21bce NP |
1354 | |
1355 | if ((tl <= load && | |
2dd73a4f PW |
1356 | tl + target_load(cpu, idx) <= tl_per_task) || |
1357 | 100*(tl + p->load_weight) <= imbalance*load) { | |
a3f21bce NP |
1358 | /* |
1359 | * This domain has SD_WAKE_AFFINE and | |
1360 | * p is cache cold in this domain, and | |
1361 | * there is no bad imbalance. | |
1362 | */ | |
1363 | schedstat_inc(this_sd, ttwu_move_affine); | |
1364 | goto out_set_cpu; | |
1365 | } | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * Start passive balancing when half the imbalance_pct | |
1370 | * limit is reached. | |
1371 | */ | |
1372 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1373 | if (imbalance*this_load <= 100*load) { | |
1374 | schedstat_inc(this_sd, ttwu_move_balance); | |
1375 | goto out_set_cpu; | |
1376 | } | |
1da177e4 LT |
1377 | } |
1378 | } | |
1379 | ||
1380 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1381 | out_set_cpu: | |
1382 | new_cpu = wake_idle(new_cpu, p); | |
1383 | if (new_cpu != cpu) { | |
1384 | set_task_cpu(p, new_cpu); | |
1385 | task_rq_unlock(rq, &flags); | |
1386 | /* might preempt at this point */ | |
1387 | rq = task_rq_lock(p, &flags); | |
1388 | old_state = p->state; | |
1389 | if (!(old_state & state)) | |
1390 | goto out; | |
1391 | if (p->array) | |
1392 | goto out_running; | |
1393 | ||
1394 | this_cpu = smp_processor_id(); | |
1395 | cpu = task_cpu(p); | |
1396 | } | |
1397 | ||
1398 | out_activate: | |
1399 | #endif /* CONFIG_SMP */ | |
1400 | if (old_state == TASK_UNINTERRUPTIBLE) { | |
1401 | rq->nr_uninterruptible--; | |
1402 | /* | |
1403 | * Tasks on involuntary sleep don't earn | |
1404 | * sleep_avg beyond just interactive state. | |
1405 | */ | |
3dee386e | 1406 | p->sleep_type = SLEEP_NONINTERACTIVE; |
e7c38cb4 | 1407 | } else |
1da177e4 | 1408 | |
d79fc0fc IM |
1409 | /* |
1410 | * Tasks that have marked their sleep as noninteractive get | |
e7c38cb4 CK |
1411 | * woken up with their sleep average not weighted in an |
1412 | * interactive way. | |
d79fc0fc | 1413 | */ |
e7c38cb4 CK |
1414 | if (old_state & TASK_NONINTERACTIVE) |
1415 | p->sleep_type = SLEEP_NONINTERACTIVE; | |
1416 | ||
1417 | ||
1418 | activate_task(p, rq, cpu == this_cpu); | |
1da177e4 LT |
1419 | /* |
1420 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1421 | * has indicated that it will leave the CPU in short order) | |
1422 | * don't trigger a preemption, if the woken up task will run on | |
1423 | * this cpu. (in this case the 'I will reschedule' promise of | |
1424 | * the waker guarantees that the freshly woken up task is going | |
1425 | * to be considered on this CPU.) | |
1426 | */ | |
1da177e4 LT |
1427 | if (!sync || cpu != this_cpu) { |
1428 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1429 | resched_task(rq->curr); | |
1430 | } | |
1431 | success = 1; | |
1432 | ||
1433 | out_running: | |
1434 | p->state = TASK_RUNNING; | |
1435 | out: | |
1436 | task_rq_unlock(rq, &flags); | |
1437 | ||
1438 | return success; | |
1439 | } | |
1440 | ||
36c8b586 | 1441 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1442 | { |
1443 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1444 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1445 | } | |
1da177e4 LT |
1446 | EXPORT_SYMBOL(wake_up_process); |
1447 | ||
36c8b586 | 1448 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1449 | { |
1450 | return try_to_wake_up(p, state, 0); | |
1451 | } | |
1452 | ||
bc947631 | 1453 | static void task_running_tick(struct rq *rq, struct task_struct *p); |
1da177e4 LT |
1454 | /* |
1455 | * Perform scheduler related setup for a newly forked process p. | |
1456 | * p is forked by current. | |
1457 | */ | |
36c8b586 | 1458 | void fastcall sched_fork(struct task_struct *p, int clone_flags) |
1da177e4 | 1459 | { |
476d139c NP |
1460 | int cpu = get_cpu(); |
1461 | ||
1462 | #ifdef CONFIG_SMP | |
1463 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1464 | #endif | |
1465 | set_task_cpu(p, cpu); | |
1466 | ||
1da177e4 LT |
1467 | /* |
1468 | * We mark the process as running here, but have not actually | |
1469 | * inserted it onto the runqueue yet. This guarantees that | |
1470 | * nobody will actually run it, and a signal or other external | |
1471 | * event cannot wake it up and insert it on the runqueue either. | |
1472 | */ | |
1473 | p->state = TASK_RUNNING; | |
b29739f9 IM |
1474 | |
1475 | /* | |
1476 | * Make sure we do not leak PI boosting priority to the child: | |
1477 | */ | |
1478 | p->prio = current->normal_prio; | |
1479 | ||
1da177e4 LT |
1480 | INIT_LIST_HEAD(&p->run_list); |
1481 | p->array = NULL; | |
52f17b6c CS |
1482 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
1483 | if (unlikely(sched_info_on())) | |
1484 | memset(&p->sched_info, 0, sizeof(p->sched_info)); | |
1da177e4 | 1485 | #endif |
d6077cb8 | 1486 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1487 | p->oncpu = 0; |
1488 | #endif | |
1da177e4 | 1489 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1490 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1491 | task_thread_info(p)->preempt_count = 1; |
1da177e4 LT |
1492 | #endif |
1493 | /* | |
1494 | * Share the timeslice between parent and child, thus the | |
1495 | * total amount of pending timeslices in the system doesn't change, | |
1496 | * resulting in more scheduling fairness. | |
1497 | */ | |
1498 | local_irq_disable(); | |
1499 | p->time_slice = (current->time_slice + 1) >> 1; | |
1500 | /* | |
1501 | * The remainder of the first timeslice might be recovered by | |
1502 | * the parent if the child exits early enough. | |
1503 | */ | |
1504 | p->first_time_slice = 1; | |
1505 | current->time_slice >>= 1; | |
1506 | p->timestamp = sched_clock(); | |
1507 | if (unlikely(!current->time_slice)) { | |
1508 | /* | |
1509 | * This case is rare, it happens when the parent has only | |
1510 | * a single jiffy left from its timeslice. Taking the | |
1511 | * runqueue lock is not a problem. | |
1512 | */ | |
1513 | current->time_slice = 1; | |
bc947631 | 1514 | task_running_tick(cpu_rq(cpu), current); |
476d139c NP |
1515 | } |
1516 | local_irq_enable(); | |
1517 | put_cpu(); | |
1da177e4 LT |
1518 | } |
1519 | ||
1520 | /* | |
1521 | * wake_up_new_task - wake up a newly created task for the first time. | |
1522 | * | |
1523 | * This function will do some initial scheduler statistics housekeeping | |
1524 | * that must be done for every newly created context, then puts the task | |
1525 | * on the runqueue and wakes it. | |
1526 | */ | |
36c8b586 | 1527 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 | 1528 | { |
70b97a7f | 1529 | struct rq *rq, *this_rq; |
1da177e4 LT |
1530 | unsigned long flags; |
1531 | int this_cpu, cpu; | |
1da177e4 LT |
1532 | |
1533 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1534 | BUG_ON(p->state != TASK_RUNNING); |
1da177e4 | 1535 | this_cpu = smp_processor_id(); |
147cbb4b | 1536 | cpu = task_cpu(p); |
1da177e4 | 1537 | |
1da177e4 LT |
1538 | /* |
1539 | * We decrease the sleep average of forking parents | |
1540 | * and children as well, to keep max-interactive tasks | |
1541 | * from forking tasks that are max-interactive. The parent | |
1542 | * (current) is done further down, under its lock. | |
1543 | */ | |
1544 | p->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(p) * | |
1545 | CHILD_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1546 | ||
1547 | p->prio = effective_prio(p); | |
1548 | ||
1549 | if (likely(cpu == this_cpu)) { | |
1550 | if (!(clone_flags & CLONE_VM)) { | |
1551 | /* | |
1552 | * The VM isn't cloned, so we're in a good position to | |
1553 | * do child-runs-first in anticipation of an exec. This | |
1554 | * usually avoids a lot of COW overhead. | |
1555 | */ | |
1556 | if (unlikely(!current->array)) | |
1557 | __activate_task(p, rq); | |
1558 | else { | |
1559 | p->prio = current->prio; | |
b29739f9 | 1560 | p->normal_prio = current->normal_prio; |
1da177e4 LT |
1561 | list_add_tail(&p->run_list, ¤t->run_list); |
1562 | p->array = current->array; | |
1563 | p->array->nr_active++; | |
2dd73a4f | 1564 | inc_nr_running(p, rq); |
1da177e4 LT |
1565 | } |
1566 | set_need_resched(); | |
1567 | } else | |
1568 | /* Run child last */ | |
1569 | __activate_task(p, rq); | |
1570 | /* | |
1571 | * We skip the following code due to cpu == this_cpu | |
1572 | * | |
1573 | * task_rq_unlock(rq, &flags); | |
1574 | * this_rq = task_rq_lock(current, &flags); | |
1575 | */ | |
1576 | this_rq = rq; | |
1577 | } else { | |
1578 | this_rq = cpu_rq(this_cpu); | |
1579 | ||
1580 | /* | |
1581 | * Not the local CPU - must adjust timestamp. This should | |
1582 | * get optimised away in the !CONFIG_SMP case. | |
1583 | */ | |
b18ec803 MG |
1584 | p->timestamp = (p->timestamp - this_rq->most_recent_timestamp) |
1585 | + rq->most_recent_timestamp; | |
1da177e4 LT |
1586 | __activate_task(p, rq); |
1587 | if (TASK_PREEMPTS_CURR(p, rq)) | |
1588 | resched_task(rq->curr); | |
1589 | ||
1590 | /* | |
1591 | * Parent and child are on different CPUs, now get the | |
1592 | * parent runqueue to update the parent's ->sleep_avg: | |
1593 | */ | |
1594 | task_rq_unlock(rq, &flags); | |
1595 | this_rq = task_rq_lock(current, &flags); | |
1596 | } | |
1597 | current->sleep_avg = JIFFIES_TO_NS(CURRENT_BONUS(current) * | |
1598 | PARENT_PENALTY / 100 * MAX_SLEEP_AVG / MAX_BONUS); | |
1599 | task_rq_unlock(this_rq, &flags); | |
1600 | } | |
1601 | ||
4866cde0 NP |
1602 | /** |
1603 | * prepare_task_switch - prepare to switch tasks | |
1604 | * @rq: the runqueue preparing to switch | |
1605 | * @next: the task we are going to switch to. | |
1606 | * | |
1607 | * This is called with the rq lock held and interrupts off. It must | |
1608 | * be paired with a subsequent finish_task_switch after the context | |
1609 | * switch. | |
1610 | * | |
1611 | * prepare_task_switch sets up locking and calls architecture specific | |
1612 | * hooks. | |
1613 | */ | |
70b97a7f | 1614 | static inline void prepare_task_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
1615 | { |
1616 | prepare_lock_switch(rq, next); | |
1617 | prepare_arch_switch(next); | |
1618 | } | |
1619 | ||
1da177e4 LT |
1620 | /** |
1621 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1622 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1623 | * @prev: the thread we just switched away from. |
1624 | * | |
4866cde0 NP |
1625 | * finish_task_switch must be called after the context switch, paired |
1626 | * with a prepare_task_switch call before the context switch. | |
1627 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1628 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1629 | * |
1630 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1631 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1632 | * with the lock held can cause deadlocks; see schedule() for | |
1633 | * details.) | |
1634 | */ | |
70b97a7f | 1635 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1636 | __releases(rq->lock) |
1637 | { | |
1da177e4 | 1638 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1639 | long prev_state; |
1da177e4 LT |
1640 | |
1641 | rq->prev_mm = NULL; | |
1642 | ||
1643 | /* | |
1644 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1645 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1646 | * schedule one last time. The schedule call will never return, and |
1647 | * the scheduled task must drop that reference. | |
c394cc9f | 1648 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1649 | * still held, otherwise prev could be scheduled on another cpu, die |
1650 | * there before we look at prev->state, and then the reference would | |
1651 | * be dropped twice. | |
1652 | * Manfred Spraul <manfred@colorfullife.com> | |
1653 | */ | |
55a101f8 | 1654 | prev_state = prev->state; |
4866cde0 NP |
1655 | finish_arch_switch(prev); |
1656 | finish_lock_switch(rq, prev); | |
1da177e4 LT |
1657 | if (mm) |
1658 | mmdrop(mm); | |
c394cc9f | 1659 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1660 | /* |
1661 | * Remove function-return probe instances associated with this | |
1662 | * task and put them back on the free list. | |
1663 | */ | |
1664 | kprobe_flush_task(prev); | |
1da177e4 | 1665 | put_task_struct(prev); |
c6fd91f0 | 1666 | } |
1da177e4 LT |
1667 | } |
1668 | ||
1669 | /** | |
1670 | * schedule_tail - first thing a freshly forked thread must call. | |
1671 | * @prev: the thread we just switched away from. | |
1672 | */ | |
36c8b586 | 1673 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1674 | __releases(rq->lock) |
1675 | { | |
70b97a7f IM |
1676 | struct rq *rq = this_rq(); |
1677 | ||
4866cde0 NP |
1678 | finish_task_switch(rq, prev); |
1679 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1680 | /* In this case, finish_task_switch does not reenable preemption */ | |
1681 | preempt_enable(); | |
1682 | #endif | |
1da177e4 LT |
1683 | if (current->set_child_tid) |
1684 | put_user(current->pid, current->set_child_tid); | |
1685 | } | |
1686 | ||
1687 | /* | |
1688 | * context_switch - switch to the new MM and the new | |
1689 | * thread's register state. | |
1690 | */ | |
36c8b586 | 1691 | static inline struct task_struct * |
70b97a7f | 1692 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1693 | struct task_struct *next) |
1da177e4 LT |
1694 | { |
1695 | struct mm_struct *mm = next->mm; | |
1696 | struct mm_struct *oldmm = prev->active_mm; | |
1697 | ||
9226d125 ZA |
1698 | /* |
1699 | * For paravirt, this is coupled with an exit in switch_to to | |
1700 | * combine the page table reload and the switch backend into | |
1701 | * one hypercall. | |
1702 | */ | |
1703 | arch_enter_lazy_cpu_mode(); | |
1704 | ||
beed33a8 | 1705 | if (!mm) { |
1da177e4 LT |
1706 | next->active_mm = oldmm; |
1707 | atomic_inc(&oldmm->mm_count); | |
1708 | enter_lazy_tlb(oldmm, next); | |
1709 | } else | |
1710 | switch_mm(oldmm, mm, next); | |
1711 | ||
beed33a8 | 1712 | if (!prev->mm) { |
1da177e4 LT |
1713 | prev->active_mm = NULL; |
1714 | WARN_ON(rq->prev_mm); | |
1715 | rq->prev_mm = oldmm; | |
1716 | } | |
3a5f5e48 IM |
1717 | /* |
1718 | * Since the runqueue lock will be released by the next | |
1719 | * task (which is an invalid locking op but in the case | |
1720 | * of the scheduler it's an obvious special-case), so we | |
1721 | * do an early lockdep release here: | |
1722 | */ | |
1723 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1724 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1725 | #endif |
1da177e4 LT |
1726 | |
1727 | /* Here we just switch the register state and the stack. */ | |
1728 | switch_to(prev, next, prev); | |
1729 | ||
1730 | return prev; | |
1731 | } | |
1732 | ||
1733 | /* | |
1734 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1735 | * | |
1736 | * externally visible scheduler statistics: current number of runnable | |
1737 | * threads, current number of uninterruptible-sleeping threads, total | |
1738 | * number of context switches performed since bootup. | |
1739 | */ | |
1740 | unsigned long nr_running(void) | |
1741 | { | |
1742 | unsigned long i, sum = 0; | |
1743 | ||
1744 | for_each_online_cpu(i) | |
1745 | sum += cpu_rq(i)->nr_running; | |
1746 | ||
1747 | return sum; | |
1748 | } | |
1749 | ||
1750 | unsigned long nr_uninterruptible(void) | |
1751 | { | |
1752 | unsigned long i, sum = 0; | |
1753 | ||
0a945022 | 1754 | for_each_possible_cpu(i) |
1da177e4 LT |
1755 | sum += cpu_rq(i)->nr_uninterruptible; |
1756 | ||
1757 | /* | |
1758 | * Since we read the counters lockless, it might be slightly | |
1759 | * inaccurate. Do not allow it to go below zero though: | |
1760 | */ | |
1761 | if (unlikely((long)sum < 0)) | |
1762 | sum = 0; | |
1763 | ||
1764 | return sum; | |
1765 | } | |
1766 | ||
1767 | unsigned long long nr_context_switches(void) | |
1768 | { | |
cc94abfc SR |
1769 | int i; |
1770 | unsigned long long sum = 0; | |
1da177e4 | 1771 | |
0a945022 | 1772 | for_each_possible_cpu(i) |
1da177e4 LT |
1773 | sum += cpu_rq(i)->nr_switches; |
1774 | ||
1775 | return sum; | |
1776 | } | |
1777 | ||
1778 | unsigned long nr_iowait(void) | |
1779 | { | |
1780 | unsigned long i, sum = 0; | |
1781 | ||
0a945022 | 1782 | for_each_possible_cpu(i) |
1da177e4 LT |
1783 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1784 | ||
1785 | return sum; | |
1786 | } | |
1787 | ||
db1b1fef JS |
1788 | unsigned long nr_active(void) |
1789 | { | |
1790 | unsigned long i, running = 0, uninterruptible = 0; | |
1791 | ||
1792 | for_each_online_cpu(i) { | |
1793 | running += cpu_rq(i)->nr_running; | |
1794 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1795 | } | |
1796 | ||
1797 | if (unlikely((long)uninterruptible < 0)) | |
1798 | uninterruptible = 0; | |
1799 | ||
1800 | return running + uninterruptible; | |
1801 | } | |
1802 | ||
1da177e4 LT |
1803 | #ifdef CONFIG_SMP |
1804 | ||
48f24c4d IM |
1805 | /* |
1806 | * Is this task likely cache-hot: | |
1807 | */ | |
1808 | static inline int | |
1809 | task_hot(struct task_struct *p, unsigned long long now, struct sched_domain *sd) | |
1810 | { | |
1811 | return (long long)(now - p->last_ran) < (long long)sd->cache_hot_time; | |
1812 | } | |
1813 | ||
1da177e4 LT |
1814 | /* |
1815 | * double_rq_lock - safely lock two runqueues | |
1816 | * | |
1817 | * Note this does not disable interrupts like task_rq_lock, | |
1818 | * you need to do so manually before calling. | |
1819 | */ | |
70b97a7f | 1820 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
1821 | __acquires(rq1->lock) |
1822 | __acquires(rq2->lock) | |
1823 | { | |
054b9108 | 1824 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
1825 | if (rq1 == rq2) { |
1826 | spin_lock(&rq1->lock); | |
1827 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1828 | } else { | |
c96d145e | 1829 | if (rq1 < rq2) { |
1da177e4 LT |
1830 | spin_lock(&rq1->lock); |
1831 | spin_lock(&rq2->lock); | |
1832 | } else { | |
1833 | spin_lock(&rq2->lock); | |
1834 | spin_lock(&rq1->lock); | |
1835 | } | |
1836 | } | |
1837 | } | |
1838 | ||
1839 | /* | |
1840 | * double_rq_unlock - safely unlock two runqueues | |
1841 | * | |
1842 | * Note this does not restore interrupts like task_rq_unlock, | |
1843 | * you need to do so manually after calling. | |
1844 | */ | |
70b97a7f | 1845 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
1846 | __releases(rq1->lock) |
1847 | __releases(rq2->lock) | |
1848 | { | |
1849 | spin_unlock(&rq1->lock); | |
1850 | if (rq1 != rq2) | |
1851 | spin_unlock(&rq2->lock); | |
1852 | else | |
1853 | __release(rq2->lock); | |
1854 | } | |
1855 | ||
1856 | /* | |
1857 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1858 | */ | |
70b97a7f | 1859 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
1860 | __releases(this_rq->lock) |
1861 | __acquires(busiest->lock) | |
1862 | __acquires(this_rq->lock) | |
1863 | { | |
054b9108 KK |
1864 | if (unlikely(!irqs_disabled())) { |
1865 | /* printk() doesn't work good under rq->lock */ | |
1866 | spin_unlock(&this_rq->lock); | |
1867 | BUG_ON(1); | |
1868 | } | |
1da177e4 | 1869 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 1870 | if (busiest < this_rq) { |
1da177e4 LT |
1871 | spin_unlock(&this_rq->lock); |
1872 | spin_lock(&busiest->lock); | |
1873 | spin_lock(&this_rq->lock); | |
1874 | } else | |
1875 | spin_lock(&busiest->lock); | |
1876 | } | |
1877 | } | |
1878 | ||
1da177e4 LT |
1879 | /* |
1880 | * If dest_cpu is allowed for this process, migrate the task to it. | |
1881 | * This is accomplished by forcing the cpu_allowed mask to only | |
1882 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
1883 | * the cpu_allowed mask is restored. | |
1884 | */ | |
36c8b586 | 1885 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 1886 | { |
70b97a7f | 1887 | struct migration_req req; |
1da177e4 | 1888 | unsigned long flags; |
70b97a7f | 1889 | struct rq *rq; |
1da177e4 LT |
1890 | |
1891 | rq = task_rq_lock(p, &flags); | |
1892 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
1893 | || unlikely(cpu_is_offline(dest_cpu))) | |
1894 | goto out; | |
1895 | ||
1896 | /* force the process onto the specified CPU */ | |
1897 | if (migrate_task(p, dest_cpu, &req)) { | |
1898 | /* Need to wait for migration thread (might exit: take ref). */ | |
1899 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 1900 | |
1da177e4 LT |
1901 | get_task_struct(mt); |
1902 | task_rq_unlock(rq, &flags); | |
1903 | wake_up_process(mt); | |
1904 | put_task_struct(mt); | |
1905 | wait_for_completion(&req.done); | |
36c8b586 | 1906 | |
1da177e4 LT |
1907 | return; |
1908 | } | |
1909 | out: | |
1910 | task_rq_unlock(rq, &flags); | |
1911 | } | |
1912 | ||
1913 | /* | |
476d139c NP |
1914 | * sched_exec - execve() is a valuable balancing opportunity, because at |
1915 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
1916 | */ |
1917 | void sched_exec(void) | |
1918 | { | |
1da177e4 | 1919 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 1920 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 1921 | put_cpu(); |
476d139c NP |
1922 | if (new_cpu != this_cpu) |
1923 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
1924 | } |
1925 | ||
1926 | /* | |
1927 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
1928 | * Both runqueues must be locked. | |
1929 | */ | |
70b97a7f IM |
1930 | static void pull_task(struct rq *src_rq, struct prio_array *src_array, |
1931 | struct task_struct *p, struct rq *this_rq, | |
1932 | struct prio_array *this_array, int this_cpu) | |
1da177e4 LT |
1933 | { |
1934 | dequeue_task(p, src_array); | |
2dd73a4f | 1935 | dec_nr_running(p, src_rq); |
1da177e4 | 1936 | set_task_cpu(p, this_cpu); |
2dd73a4f | 1937 | inc_nr_running(p, this_rq); |
1da177e4 | 1938 | enqueue_task(p, this_array); |
b18ec803 MG |
1939 | p->timestamp = (p->timestamp - src_rq->most_recent_timestamp) |
1940 | + this_rq->most_recent_timestamp; | |
1da177e4 LT |
1941 | /* |
1942 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
1943 | * to be always true for them. | |
1944 | */ | |
1945 | if (TASK_PREEMPTS_CURR(p, this_rq)) | |
1946 | resched_task(this_rq->curr); | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
1951 | */ | |
858119e1 | 1952 | static |
70b97a7f | 1953 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 1954 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 1955 | int *all_pinned) |
1da177e4 LT |
1956 | { |
1957 | /* | |
1958 | * We do not migrate tasks that are: | |
1959 | * 1) running (obviously), or | |
1960 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
1961 | * 3) are cache-hot on their current CPU. | |
1962 | */ | |
1da177e4 LT |
1963 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
1964 | return 0; | |
81026794 NP |
1965 | *all_pinned = 0; |
1966 | ||
1967 | if (task_running(rq, p)) | |
1968 | return 0; | |
1da177e4 LT |
1969 | |
1970 | /* | |
1971 | * Aggressive migration if: | |
cafb20c1 | 1972 | * 1) task is cache cold, or |
1da177e4 LT |
1973 | * 2) too many balance attempts have failed. |
1974 | */ | |
1975 | ||
b18ec803 MG |
1976 | if (sd->nr_balance_failed > sd->cache_nice_tries) { |
1977 | #ifdef CONFIG_SCHEDSTATS | |
1978 | if (task_hot(p, rq->most_recent_timestamp, sd)) | |
1979 | schedstat_inc(sd, lb_hot_gained[idle]); | |
1980 | #endif | |
1da177e4 | 1981 | return 1; |
b18ec803 | 1982 | } |
1da177e4 | 1983 | |
b18ec803 | 1984 | if (task_hot(p, rq->most_recent_timestamp, sd)) |
81026794 | 1985 | return 0; |
1da177e4 LT |
1986 | return 1; |
1987 | } | |
1988 | ||
615052dc | 1989 | #define rq_best_prio(rq) min((rq)->curr->prio, (rq)->best_expired_prio) |
48f24c4d | 1990 | |
1da177e4 | 1991 | /* |
2dd73a4f PW |
1992 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted |
1993 | * load from busiest to this_rq, as part of a balancing operation within | |
1994 | * "domain". Returns the number of tasks moved. | |
1da177e4 LT |
1995 | * |
1996 | * Called with both runqueues locked. | |
1997 | */ | |
70b97a7f | 1998 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 1999 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2000 | struct sched_domain *sd, enum cpu_idle_type idle, |
2dd73a4f | 2001 | int *all_pinned) |
1da177e4 | 2002 | { |
48f24c4d IM |
2003 | int idx, pulled = 0, pinned = 0, this_best_prio, best_prio, |
2004 | best_prio_seen, skip_for_load; | |
70b97a7f | 2005 | struct prio_array *array, *dst_array; |
1da177e4 | 2006 | struct list_head *head, *curr; |
36c8b586 | 2007 | struct task_struct *tmp; |
2dd73a4f | 2008 | long rem_load_move; |
1da177e4 | 2009 | |
2dd73a4f | 2010 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2011 | goto out; |
2012 | ||
2dd73a4f | 2013 | rem_load_move = max_load_move; |
81026794 | 2014 | pinned = 1; |
615052dc | 2015 | this_best_prio = rq_best_prio(this_rq); |
48f24c4d | 2016 | best_prio = rq_best_prio(busiest); |
615052dc PW |
2017 | /* |
2018 | * Enable handling of the case where there is more than one task | |
2019 | * with the best priority. If the current running task is one | |
48f24c4d | 2020 | * of those with prio==best_prio we know it won't be moved |
615052dc PW |
2021 | * and therefore it's safe to override the skip (based on load) of |
2022 | * any task we find with that prio. | |
2023 | */ | |
48f24c4d | 2024 | best_prio_seen = best_prio == busiest->curr->prio; |
81026794 | 2025 | |
1da177e4 LT |
2026 | /* |
2027 | * We first consider expired tasks. Those will likely not be | |
2028 | * executed in the near future, and they are most likely to | |
2029 | * be cache-cold, thus switching CPUs has the least effect | |
2030 | * on them. | |
2031 | */ | |
2032 | if (busiest->expired->nr_active) { | |
2033 | array = busiest->expired; | |
2034 | dst_array = this_rq->expired; | |
2035 | } else { | |
2036 | array = busiest->active; | |
2037 | dst_array = this_rq->active; | |
2038 | } | |
2039 | ||
2040 | new_array: | |
2041 | /* Start searching at priority 0: */ | |
2042 | idx = 0; | |
2043 | skip_bitmap: | |
2044 | if (!idx) | |
2045 | idx = sched_find_first_bit(array->bitmap); | |
2046 | else | |
2047 | idx = find_next_bit(array->bitmap, MAX_PRIO, idx); | |
2048 | if (idx >= MAX_PRIO) { | |
2049 | if (array == busiest->expired && busiest->active->nr_active) { | |
2050 | array = busiest->active; | |
2051 | dst_array = this_rq->active; | |
2052 | goto new_array; | |
2053 | } | |
2054 | goto out; | |
2055 | } | |
2056 | ||
2057 | head = array->queue + idx; | |
2058 | curr = head->prev; | |
2059 | skip_queue: | |
36c8b586 | 2060 | tmp = list_entry(curr, struct task_struct, run_list); |
1da177e4 LT |
2061 | |
2062 | curr = curr->prev; | |
2063 | ||
50ddd969 PW |
2064 | /* |
2065 | * To help distribute high priority tasks accross CPUs we don't | |
2066 | * skip a task if it will be the highest priority task (i.e. smallest | |
2067 | * prio value) on its new queue regardless of its load weight | |
2068 | */ | |
615052dc PW |
2069 | skip_for_load = tmp->load_weight > rem_load_move; |
2070 | if (skip_for_load && idx < this_best_prio) | |
48f24c4d | 2071 | skip_for_load = !best_prio_seen && idx == best_prio; |
615052dc | 2072 | if (skip_for_load || |
2dd73a4f | 2073 | !can_migrate_task(tmp, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d IM |
2074 | |
2075 | best_prio_seen |= idx == best_prio; | |
1da177e4 LT |
2076 | if (curr != head) |
2077 | goto skip_queue; | |
2078 | idx++; | |
2079 | goto skip_bitmap; | |
2080 | } | |
2081 | ||
1da177e4 LT |
2082 | pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu); |
2083 | pulled++; | |
2dd73a4f | 2084 | rem_load_move -= tmp->load_weight; |
1da177e4 | 2085 | |
2dd73a4f PW |
2086 | /* |
2087 | * We only want to steal up to the prescribed number of tasks | |
2088 | * and the prescribed amount of weighted load. | |
2089 | */ | |
2090 | if (pulled < max_nr_move && rem_load_move > 0) { | |
615052dc PW |
2091 | if (idx < this_best_prio) |
2092 | this_best_prio = idx; | |
1da177e4 LT |
2093 | if (curr != head) |
2094 | goto skip_queue; | |
2095 | idx++; | |
2096 | goto skip_bitmap; | |
2097 | } | |
2098 | out: | |
2099 | /* | |
2100 | * Right now, this is the only place pull_task() is called, | |
2101 | * so we can safely collect pull_task() stats here rather than | |
2102 | * inside pull_task(). | |
2103 | */ | |
2104 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2105 | |
2106 | if (all_pinned) | |
2107 | *all_pinned = pinned; | |
1da177e4 LT |
2108 | return pulled; |
2109 | } | |
2110 | ||
2111 | /* | |
2112 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2113 | * domain. It calculates and returns the amount of weighted load which |
2114 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2115 | */ |
2116 | static struct sched_group * | |
2117 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
d15bcfdb | 2118 | unsigned long *imbalance, enum cpu_idle_type idle, int *sd_idle, |
783609c6 | 2119 | cpumask_t *cpus, int *balance) |
1da177e4 LT |
2120 | { |
2121 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2122 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2123 | unsigned long max_pull; |
2dd73a4f PW |
2124 | unsigned long busiest_load_per_task, busiest_nr_running; |
2125 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2126 | int load_idx; |
5c45bf27 SS |
2127 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2128 | int power_savings_balance = 1; | |
2129 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2130 | unsigned long min_nr_running = ULONG_MAX; | |
2131 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2132 | #endif | |
1da177e4 LT |
2133 | |
2134 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2135 | busiest_load_per_task = busiest_nr_running = 0; |
2136 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2137 | if (idle == CPU_NOT_IDLE) |
7897986b | 2138 | load_idx = sd->busy_idx; |
d15bcfdb | 2139 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2140 | load_idx = sd->newidle_idx; |
2141 | else | |
2142 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2143 | |
2144 | do { | |
5c45bf27 | 2145 | unsigned long load, group_capacity; |
1da177e4 LT |
2146 | int local_group; |
2147 | int i; | |
783609c6 | 2148 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2149 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2150 | |
2151 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2152 | ||
783609c6 SS |
2153 | if (local_group) |
2154 | balance_cpu = first_cpu(group->cpumask); | |
2155 | ||
1da177e4 | 2156 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2157 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2158 | |
2159 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2160 | struct rq *rq; |
2161 | ||
2162 | if (!cpu_isset(i, *cpus)) | |
2163 | continue; | |
2164 | ||
2165 | rq = cpu_rq(i); | |
2dd73a4f | 2166 | |
5969fe06 NP |
2167 | if (*sd_idle && !idle_cpu(i)) |
2168 | *sd_idle = 0; | |
2169 | ||
1da177e4 | 2170 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2171 | if (local_group) { |
2172 | if (idle_cpu(i) && !first_idle_cpu) { | |
2173 | first_idle_cpu = 1; | |
2174 | balance_cpu = i; | |
2175 | } | |
2176 | ||
a2000572 | 2177 | load = target_load(i, load_idx); |
783609c6 | 2178 | } else |
a2000572 | 2179 | load = source_load(i, load_idx); |
1da177e4 LT |
2180 | |
2181 | avg_load += load; | |
2dd73a4f PW |
2182 | sum_nr_running += rq->nr_running; |
2183 | sum_weighted_load += rq->raw_weighted_load; | |
1da177e4 LT |
2184 | } |
2185 | ||
783609c6 SS |
2186 | /* |
2187 | * First idle cpu or the first cpu(busiest) in this sched group | |
2188 | * is eligible for doing load balancing at this and above | |
2189 | * domains. | |
2190 | */ | |
2191 | if (local_group && balance_cpu != this_cpu && balance) { | |
2192 | *balance = 0; | |
2193 | goto ret; | |
2194 | } | |
2195 | ||
1da177e4 | 2196 | total_load += avg_load; |
5517d86b | 2197 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2198 | |
2199 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2200 | avg_load = sg_div_cpu_power(group, |
2201 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2202 | |
5517d86b | 2203 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2204 | |
1da177e4 LT |
2205 | if (local_group) { |
2206 | this_load = avg_load; | |
2207 | this = group; | |
2dd73a4f PW |
2208 | this_nr_running = sum_nr_running; |
2209 | this_load_per_task = sum_weighted_load; | |
2210 | } else if (avg_load > max_load && | |
5c45bf27 | 2211 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2212 | max_load = avg_load; |
2213 | busiest = group; | |
2dd73a4f PW |
2214 | busiest_nr_running = sum_nr_running; |
2215 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2216 | } |
5c45bf27 SS |
2217 | |
2218 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2219 | /* | |
2220 | * Busy processors will not participate in power savings | |
2221 | * balance. | |
2222 | */ | |
d15bcfdb | 2223 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 SS |
2224 | goto group_next; |
2225 | ||
2226 | /* | |
2227 | * If the local group is idle or completely loaded | |
2228 | * no need to do power savings balance at this domain | |
2229 | */ | |
2230 | if (local_group && (this_nr_running >= group_capacity || | |
2231 | !this_nr_running)) | |
2232 | power_savings_balance = 0; | |
2233 | ||
2234 | /* | |
2235 | * If a group is already running at full capacity or idle, | |
2236 | * don't include that group in power savings calculations | |
2237 | */ | |
2238 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
2239 | || !sum_nr_running) | |
2240 | goto group_next; | |
2241 | ||
2242 | /* | |
2243 | * Calculate the group which has the least non-idle load. | |
2244 | * This is the group from where we need to pick up the load | |
2245 | * for saving power | |
2246 | */ | |
2247 | if ((sum_nr_running < min_nr_running) || | |
2248 | (sum_nr_running == min_nr_running && | |
2249 | first_cpu(group->cpumask) < | |
2250 | first_cpu(group_min->cpumask))) { | |
2251 | group_min = group; | |
2252 | min_nr_running = sum_nr_running; | |
2253 | min_load_per_task = sum_weighted_load / | |
2254 | sum_nr_running; | |
2255 | } | |
2256 | ||
2257 | /* | |
2258 | * Calculate the group which is almost near its | |
2259 | * capacity but still has some space to pick up some load | |
2260 | * from other group and save more power | |
2261 | */ | |
48f24c4d | 2262 | if (sum_nr_running <= group_capacity - 1) { |
5c45bf27 SS |
2263 | if (sum_nr_running > leader_nr_running || |
2264 | (sum_nr_running == leader_nr_running && | |
2265 | first_cpu(group->cpumask) > | |
2266 | first_cpu(group_leader->cpumask))) { | |
2267 | group_leader = group; | |
2268 | leader_nr_running = sum_nr_running; | |
2269 | } | |
48f24c4d | 2270 | } |
5c45bf27 SS |
2271 | group_next: |
2272 | #endif | |
1da177e4 LT |
2273 | group = group->next; |
2274 | } while (group != sd->groups); | |
2275 | ||
2dd73a4f | 2276 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2277 | goto out_balanced; |
2278 | ||
2279 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2280 | ||
2281 | if (this_load >= avg_load || | |
2282 | 100*max_load <= sd->imbalance_pct*this_load) | |
2283 | goto out_balanced; | |
2284 | ||
2dd73a4f | 2285 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2286 | /* |
2287 | * We're trying to get all the cpus to the average_load, so we don't | |
2288 | * want to push ourselves above the average load, nor do we wish to | |
2289 | * reduce the max loaded cpu below the average load, as either of these | |
2290 | * actions would just result in more rebalancing later, and ping-pong | |
2291 | * tasks around. Thus we look for the minimum possible imbalance. | |
2292 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2293 | * be counted as no imbalance for these purposes -- we can't fix that | |
2294 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2295 | * appear as very large values with unsigned longs. | |
2296 | */ | |
2dd73a4f PW |
2297 | if (max_load <= busiest_load_per_task) |
2298 | goto out_balanced; | |
2299 | ||
2300 | /* | |
2301 | * In the presence of smp nice balancing, certain scenarios can have | |
2302 | * max load less than avg load(as we skip the groups at or below | |
2303 | * its cpu_power, while calculating max_load..) | |
2304 | */ | |
2305 | if (max_load < avg_load) { | |
2306 | *imbalance = 0; | |
2307 | goto small_imbalance; | |
2308 | } | |
0c117f1b SS |
2309 | |
2310 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2311 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2312 | |
1da177e4 | 2313 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2314 | *imbalance = min(max_pull * busiest->__cpu_power, |
2315 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2316 | / SCHED_LOAD_SCALE; |
2317 | ||
2dd73a4f PW |
2318 | /* |
2319 | * if *imbalance is less than the average load per runnable task | |
2320 | * there is no gaurantee that any tasks will be moved so we'll have | |
2321 | * a think about bumping its value to force at least one task to be | |
2322 | * moved | |
2323 | */ | |
2324 | if (*imbalance < busiest_load_per_task) { | |
48f24c4d | 2325 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2326 | unsigned int imbn; |
2327 | ||
2328 | small_imbalance: | |
2329 | pwr_move = pwr_now = 0; | |
2330 | imbn = 2; | |
2331 | if (this_nr_running) { | |
2332 | this_load_per_task /= this_nr_running; | |
2333 | if (busiest_load_per_task > this_load_per_task) | |
2334 | imbn = 1; | |
2335 | } else | |
2336 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2337 | |
2dd73a4f PW |
2338 | if (max_load - this_load >= busiest_load_per_task * imbn) { |
2339 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2340 | return busiest; |
2341 | } | |
2342 | ||
2343 | /* | |
2344 | * OK, we don't have enough imbalance to justify moving tasks, | |
2345 | * however we may be able to increase total CPU power used by | |
2346 | * moving them. | |
2347 | */ | |
2348 | ||
5517d86b ED |
2349 | pwr_now += busiest->__cpu_power * |
2350 | min(busiest_load_per_task, max_load); | |
2351 | pwr_now += this->__cpu_power * | |
2352 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2353 | pwr_now /= SCHED_LOAD_SCALE; |
2354 | ||
2355 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2356 | tmp = sg_div_cpu_power(busiest, |
2357 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2358 | if (max_load > tmp) |
5517d86b | 2359 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2360 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2361 | |
2362 | /* Amount of load we'd add */ | |
5517d86b | 2363 | if (max_load * busiest->__cpu_power < |
33859f7f | 2364 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2365 | tmp = sg_div_cpu_power(this, |
2366 | max_load * busiest->__cpu_power); | |
1da177e4 | 2367 | else |
5517d86b ED |
2368 | tmp = sg_div_cpu_power(this, |
2369 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2370 | pwr_move += this->__cpu_power * | |
2371 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2372 | pwr_move /= SCHED_LOAD_SCALE; |
2373 | ||
2374 | /* Move if we gain throughput */ | |
2375 | if (pwr_move <= pwr_now) | |
2376 | goto out_balanced; | |
2377 | ||
2dd73a4f | 2378 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2379 | } |
2380 | ||
1da177e4 LT |
2381 | return busiest; |
2382 | ||
2383 | out_balanced: | |
5c45bf27 | 2384 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2385 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2386 | goto ret; |
1da177e4 | 2387 | |
5c45bf27 SS |
2388 | if (this == group_leader && group_leader != group_min) { |
2389 | *imbalance = min_load_per_task; | |
2390 | return group_min; | |
2391 | } | |
5c45bf27 | 2392 | #endif |
783609c6 | 2393 | ret: |
1da177e4 LT |
2394 | *imbalance = 0; |
2395 | return NULL; | |
2396 | } | |
2397 | ||
2398 | /* | |
2399 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2400 | */ | |
70b97a7f | 2401 | static struct rq * |
d15bcfdb | 2402 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2403 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2404 | { |
70b97a7f | 2405 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2406 | unsigned long max_load = 0; |
1da177e4 LT |
2407 | int i; |
2408 | ||
2409 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2410 | |
2411 | if (!cpu_isset(i, *cpus)) | |
2412 | continue; | |
2413 | ||
48f24c4d | 2414 | rq = cpu_rq(i); |
2dd73a4f | 2415 | |
48f24c4d | 2416 | if (rq->nr_running == 1 && rq->raw_weighted_load > imbalance) |
2dd73a4f | 2417 | continue; |
1da177e4 | 2418 | |
48f24c4d IM |
2419 | if (rq->raw_weighted_load > max_load) { |
2420 | max_load = rq->raw_weighted_load; | |
2421 | busiest = rq; | |
1da177e4 LT |
2422 | } |
2423 | } | |
2424 | ||
2425 | return busiest; | |
2426 | } | |
2427 | ||
77391d71 NP |
2428 | /* |
2429 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2430 | * so long as it is large enough. | |
2431 | */ | |
2432 | #define MAX_PINNED_INTERVAL 512 | |
2433 | ||
48f24c4d IM |
2434 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2435 | { | |
2436 | return n > 0 ? n - 1 : 0; | |
2437 | } | |
2438 | ||
1da177e4 LT |
2439 | /* |
2440 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2441 | * tasks if there is an imbalance. | |
1da177e4 | 2442 | */ |
70b97a7f | 2443 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2444 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2445 | int *balance) |
1da177e4 | 2446 | { |
48f24c4d | 2447 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2448 | struct sched_group *group; |
1da177e4 | 2449 | unsigned long imbalance; |
70b97a7f | 2450 | struct rq *busiest; |
0a2966b4 | 2451 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2452 | unsigned long flags; |
5969fe06 | 2453 | |
89c4710e SS |
2454 | /* |
2455 | * When power savings policy is enabled for the parent domain, idle | |
2456 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2457 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2458 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2459 | */ |
d15bcfdb | 2460 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2461 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2462 | sd_idle = 1; |
1da177e4 | 2463 | |
1da177e4 LT |
2464 | schedstat_inc(sd, lb_cnt[idle]); |
2465 | ||
0a2966b4 CL |
2466 | redo: |
2467 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2468 | &cpus, balance); |
2469 | ||
06066714 | 2470 | if (*balance == 0) |
783609c6 | 2471 | goto out_balanced; |
783609c6 | 2472 | |
1da177e4 LT |
2473 | if (!group) { |
2474 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2475 | goto out_balanced; | |
2476 | } | |
2477 | ||
0a2966b4 | 2478 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2479 | if (!busiest) { |
2480 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2481 | goto out_balanced; | |
2482 | } | |
2483 | ||
db935dbd | 2484 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2485 | |
2486 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2487 | ||
2488 | nr_moved = 0; | |
2489 | if (busiest->nr_running > 1) { | |
2490 | /* | |
2491 | * Attempt to move tasks. If find_busiest_group has found | |
2492 | * an imbalance but busiest->nr_running <= 1, the group is | |
2493 | * still unbalanced. nr_moved simply stays zero, so it is | |
2494 | * correctly treated as an imbalance. | |
2495 | */ | |
fe2eea3f | 2496 | local_irq_save(flags); |
e17224bf | 2497 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2498 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2499 | minus_1_or_zero(busiest->nr_running), |
2500 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2501 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2502 | local_irq_restore(flags); |
81026794 | 2503 | |
46cb4b7c SS |
2504 | /* |
2505 | * some other cpu did the load balance for us. | |
2506 | */ | |
2507 | if (nr_moved && this_cpu != smp_processor_id()) | |
2508 | resched_cpu(this_cpu); | |
2509 | ||
81026794 | 2510 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2511 | if (unlikely(all_pinned)) { |
2512 | cpu_clear(cpu_of(busiest), cpus); | |
2513 | if (!cpus_empty(cpus)) | |
2514 | goto redo; | |
81026794 | 2515 | goto out_balanced; |
0a2966b4 | 2516 | } |
1da177e4 | 2517 | } |
81026794 | 2518 | |
1da177e4 LT |
2519 | if (!nr_moved) { |
2520 | schedstat_inc(sd, lb_failed[idle]); | |
2521 | sd->nr_balance_failed++; | |
2522 | ||
2523 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2524 | |
fe2eea3f | 2525 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2526 | |
2527 | /* don't kick the migration_thread, if the curr | |
2528 | * task on busiest cpu can't be moved to this_cpu | |
2529 | */ | |
2530 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2531 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2532 | all_pinned = 1; |
2533 | goto out_one_pinned; | |
2534 | } | |
2535 | ||
1da177e4 LT |
2536 | if (!busiest->active_balance) { |
2537 | busiest->active_balance = 1; | |
2538 | busiest->push_cpu = this_cpu; | |
81026794 | 2539 | active_balance = 1; |
1da177e4 | 2540 | } |
fe2eea3f | 2541 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2542 | if (active_balance) |
1da177e4 LT |
2543 | wake_up_process(busiest->migration_thread); |
2544 | ||
2545 | /* | |
2546 | * We've kicked active balancing, reset the failure | |
2547 | * counter. | |
2548 | */ | |
39507451 | 2549 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2550 | } |
81026794 | 2551 | } else |
1da177e4 LT |
2552 | sd->nr_balance_failed = 0; |
2553 | ||
81026794 | 2554 | if (likely(!active_balance)) { |
1da177e4 LT |
2555 | /* We were unbalanced, so reset the balancing interval */ |
2556 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2557 | } else { |
2558 | /* | |
2559 | * If we've begun active balancing, start to back off. This | |
2560 | * case may not be covered by the all_pinned logic if there | |
2561 | * is only 1 task on the busy runqueue (because we don't call | |
2562 | * move_tasks). | |
2563 | */ | |
2564 | if (sd->balance_interval < sd->max_interval) | |
2565 | sd->balance_interval *= 2; | |
1da177e4 LT |
2566 | } |
2567 | ||
5c45bf27 | 2568 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2569 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2570 | return -1; |
1da177e4 LT |
2571 | return nr_moved; |
2572 | ||
2573 | out_balanced: | |
1da177e4 LT |
2574 | schedstat_inc(sd, lb_balanced[idle]); |
2575 | ||
16cfb1c0 | 2576 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2577 | |
2578 | out_one_pinned: | |
1da177e4 | 2579 | /* tune up the balancing interval */ |
77391d71 NP |
2580 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2581 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2582 | sd->balance_interval *= 2; |
2583 | ||
48f24c4d | 2584 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2585 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2586 | return -1; |
1da177e4 LT |
2587 | return 0; |
2588 | } | |
2589 | ||
2590 | /* | |
2591 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2592 | * tasks if there is an imbalance. | |
2593 | * | |
d15bcfdb | 2594 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2595 | * this_rq is locked. |
2596 | */ | |
48f24c4d | 2597 | static int |
70b97a7f | 2598 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2599 | { |
2600 | struct sched_group *group; | |
70b97a7f | 2601 | struct rq *busiest = NULL; |
1da177e4 LT |
2602 | unsigned long imbalance; |
2603 | int nr_moved = 0; | |
5969fe06 | 2604 | int sd_idle = 0; |
0a2966b4 | 2605 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2606 | |
89c4710e SS |
2607 | /* |
2608 | * When power savings policy is enabled for the parent domain, idle | |
2609 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2610 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2611 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2612 | */ |
2613 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2614 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2615 | sd_idle = 1; |
1da177e4 | 2616 | |
d15bcfdb | 2617 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2618 | redo: |
d15bcfdb | 2619 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2620 | &sd_idle, &cpus, NULL); |
1da177e4 | 2621 | if (!group) { |
d15bcfdb | 2622 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2623 | goto out_balanced; |
1da177e4 LT |
2624 | } |
2625 | ||
d15bcfdb | 2626 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2627 | &cpus); |
db935dbd | 2628 | if (!busiest) { |
d15bcfdb | 2629 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2630 | goto out_balanced; |
1da177e4 LT |
2631 | } |
2632 | ||
db935dbd NP |
2633 | BUG_ON(busiest == this_rq); |
2634 | ||
d15bcfdb | 2635 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2636 | |
2637 | nr_moved = 0; | |
2638 | if (busiest->nr_running > 1) { | |
2639 | /* Attempt to move tasks */ | |
2640 | double_lock_balance(this_rq, busiest); | |
2641 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2642 | minus_1_or_zero(busiest->nr_running), |
d15bcfdb | 2643 | imbalance, sd, CPU_NEWLY_IDLE, NULL); |
d6d5cfaf | 2644 | spin_unlock(&busiest->lock); |
0a2966b4 CL |
2645 | |
2646 | if (!nr_moved) { | |
2647 | cpu_clear(cpu_of(busiest), cpus); | |
2648 | if (!cpus_empty(cpus)) | |
2649 | goto redo; | |
2650 | } | |
d6d5cfaf NP |
2651 | } |
2652 | ||
5969fe06 | 2653 | if (!nr_moved) { |
d15bcfdb | 2654 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2655 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2656 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2657 | return -1; |
2658 | } else | |
16cfb1c0 | 2659 | sd->nr_balance_failed = 0; |
1da177e4 | 2660 | |
1da177e4 | 2661 | return nr_moved; |
16cfb1c0 NP |
2662 | |
2663 | out_balanced: | |
d15bcfdb | 2664 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2665 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2666 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2667 | return -1; |
16cfb1c0 | 2668 | sd->nr_balance_failed = 0; |
48f24c4d | 2669 | |
16cfb1c0 | 2670 | return 0; |
1da177e4 LT |
2671 | } |
2672 | ||
2673 | /* | |
2674 | * idle_balance is called by schedule() if this_cpu is about to become | |
2675 | * idle. Attempts to pull tasks from other CPUs. | |
2676 | */ | |
70b97a7f | 2677 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2678 | { |
2679 | struct sched_domain *sd; | |
1bd77f2d CL |
2680 | int pulled_task = 0; |
2681 | unsigned long next_balance = jiffies + 60 * HZ; | |
1da177e4 LT |
2682 | |
2683 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2684 | unsigned long interval; |
2685 | ||
2686 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2687 | continue; | |
2688 | ||
2689 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2690 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2691 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2692 | this_rq, sd); |
2693 | ||
2694 | interval = msecs_to_jiffies(sd->balance_interval); | |
2695 | if (time_after(next_balance, sd->last_balance + interval)) | |
2696 | next_balance = sd->last_balance + interval; | |
2697 | if (pulled_task) | |
2698 | break; | |
1da177e4 | 2699 | } |
1bd77f2d CL |
2700 | if (!pulled_task) |
2701 | /* | |
2702 | * We are going idle. next_balance may be set based on | |
2703 | * a busy processor. So reset next_balance. | |
2704 | */ | |
2705 | this_rq->next_balance = next_balance; | |
1da177e4 LT |
2706 | } |
2707 | ||
2708 | /* | |
2709 | * active_load_balance is run by migration threads. It pushes running tasks | |
2710 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2711 | * running on each physical CPU where possible, and avoids physical / | |
2712 | * logical imbalances. | |
2713 | * | |
2714 | * Called with busiest_rq locked. | |
2715 | */ | |
70b97a7f | 2716 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2717 | { |
39507451 | 2718 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2719 | struct sched_domain *sd; |
2720 | struct rq *target_rq; | |
39507451 | 2721 | |
48f24c4d | 2722 | /* Is there any task to move? */ |
39507451 | 2723 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2724 | return; |
2725 | ||
2726 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2727 | |
2728 | /* | |
39507451 NP |
2729 | * This condition is "impossible", if it occurs |
2730 | * we need to fix it. Originally reported by | |
2731 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2732 | */ |
39507451 | 2733 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2734 | |
39507451 NP |
2735 | /* move a task from busiest_rq to target_rq */ |
2736 | double_lock_balance(busiest_rq, target_rq); | |
2737 | ||
2738 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2739 | for_each_domain(target_cpu, sd) { |
39507451 | 2740 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2741 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2742 | break; |
c96d145e | 2743 | } |
39507451 | 2744 | |
48f24c4d IM |
2745 | if (likely(sd)) { |
2746 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2747 | |
48f24c4d | 2748 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
d15bcfdb | 2749 | RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE, |
48f24c4d IM |
2750 | NULL)) |
2751 | schedstat_inc(sd, alb_pushed); | |
2752 | else | |
2753 | schedstat_inc(sd, alb_failed); | |
2754 | } | |
39507451 | 2755 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2756 | } |
2757 | ||
7835b98b | 2758 | static void update_load(struct rq *this_rq) |
1da177e4 | 2759 | { |
7835b98b | 2760 | unsigned long this_load; |
ff91691b | 2761 | unsigned int i, scale; |
1da177e4 | 2762 | |
2dd73a4f | 2763 | this_load = this_rq->raw_weighted_load; |
48f24c4d IM |
2764 | |
2765 | /* Update our load: */ | |
ff91691b | 2766 | for (i = 0, scale = 1; i < 3; i++, scale += scale) { |
48f24c4d IM |
2767 | unsigned long old_load, new_load; |
2768 | ||
ff91691b NP |
2769 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
2770 | ||
7897986b | 2771 | old_load = this_rq->cpu_load[i]; |
48f24c4d | 2772 | new_load = this_load; |
7897986b NP |
2773 | /* |
2774 | * Round up the averaging division if load is increasing. This | |
2775 | * prevents us from getting stuck on 9 if the load is 10, for | |
2776 | * example. | |
2777 | */ | |
2778 | if (new_load > old_load) | |
2779 | new_load += scale-1; | |
ff91691b | 2780 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
7897986b | 2781 | } |
7835b98b CL |
2782 | } |
2783 | ||
46cb4b7c SS |
2784 | #ifdef CONFIG_NO_HZ |
2785 | static struct { | |
2786 | atomic_t load_balancer; | |
2787 | cpumask_t cpu_mask; | |
2788 | } nohz ____cacheline_aligned = { | |
2789 | .load_balancer = ATOMIC_INIT(-1), | |
2790 | .cpu_mask = CPU_MASK_NONE, | |
2791 | }; | |
2792 | ||
7835b98b | 2793 | /* |
46cb4b7c SS |
2794 | * This routine will try to nominate the ilb (idle load balancing) |
2795 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2796 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2797 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2798 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2799 | * arrives... | |
2800 | * | |
2801 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2802 | * for idle load balancing. ilb owner will still be part of | |
2803 | * nohz.cpu_mask.. | |
7835b98b | 2804 | * |
46cb4b7c SS |
2805 | * While stopping the tick, this cpu will become the ilb owner if there |
2806 | * is no other owner. And will be the owner till that cpu becomes busy | |
2807 | * or if all cpus in the system stop their ticks at which point | |
2808 | * there is no need for ilb owner. | |
2809 | * | |
2810 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2811 | * next busy scheduler_tick() | |
2812 | */ | |
2813 | int select_nohz_load_balancer(int stop_tick) | |
2814 | { | |
2815 | int cpu = smp_processor_id(); | |
2816 | ||
2817 | if (stop_tick) { | |
2818 | cpu_set(cpu, nohz.cpu_mask); | |
2819 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2820 | ||
2821 | /* | |
2822 | * If we are going offline and still the leader, give up! | |
2823 | */ | |
2824 | if (cpu_is_offline(cpu) && | |
2825 | atomic_read(&nohz.load_balancer) == cpu) { | |
2826 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2827 | BUG(); | |
2828 | return 0; | |
2829 | } | |
2830 | ||
2831 | /* time for ilb owner also to sleep */ | |
2832 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2833 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2834 | atomic_set(&nohz.load_balancer, -1); | |
2835 | return 0; | |
2836 | } | |
2837 | ||
2838 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2839 | /* make me the ilb owner */ | |
2840 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2841 | return 1; | |
2842 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2843 | return 1; | |
2844 | } else { | |
2845 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2846 | return 0; | |
2847 | ||
2848 | cpu_clear(cpu, nohz.cpu_mask); | |
2849 | ||
2850 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2851 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2852 | BUG(); | |
2853 | } | |
2854 | return 0; | |
2855 | } | |
2856 | #endif | |
2857 | ||
2858 | static DEFINE_SPINLOCK(balancing); | |
2859 | ||
2860 | /* | |
7835b98b CL |
2861 | * It checks each scheduling domain to see if it is due to be balanced, |
2862 | * and initiates a balancing operation if so. | |
2863 | * | |
2864 | * Balancing parameters are set up in arch_init_sched_domains. | |
2865 | */ | |
d15bcfdb | 2866 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 2867 | { |
46cb4b7c SS |
2868 | int balance = 1; |
2869 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
2870 | unsigned long interval; |
2871 | struct sched_domain *sd; | |
46cb4b7c | 2872 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 2873 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 2874 | |
46cb4b7c | 2875 | for_each_domain(cpu, sd) { |
1da177e4 LT |
2876 | if (!(sd->flags & SD_LOAD_BALANCE)) |
2877 | continue; | |
2878 | ||
2879 | interval = sd->balance_interval; | |
d15bcfdb | 2880 | if (idle != CPU_IDLE) |
1da177e4 LT |
2881 | interval *= sd->busy_factor; |
2882 | ||
2883 | /* scale ms to jiffies */ | |
2884 | interval = msecs_to_jiffies(interval); | |
2885 | if (unlikely(!interval)) | |
2886 | interval = 1; | |
2887 | ||
08c183f3 CL |
2888 | if (sd->flags & SD_SERIALIZE) { |
2889 | if (!spin_trylock(&balancing)) | |
2890 | goto out; | |
2891 | } | |
2892 | ||
c9819f45 | 2893 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 2894 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
2895 | /* |
2896 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
2897 | * longer idle, or one of our SMT siblings is |
2898 | * not idle. | |
2899 | */ | |
d15bcfdb | 2900 | idle = CPU_NOT_IDLE; |
1da177e4 | 2901 | } |
1bd77f2d | 2902 | sd->last_balance = jiffies; |
1da177e4 | 2903 | } |
08c183f3 CL |
2904 | if (sd->flags & SD_SERIALIZE) |
2905 | spin_unlock(&balancing); | |
2906 | out: | |
c9819f45 CL |
2907 | if (time_after(next_balance, sd->last_balance + interval)) |
2908 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
2909 | |
2910 | /* | |
2911 | * Stop the load balance at this level. There is another | |
2912 | * CPU in our sched group which is doing load balancing more | |
2913 | * actively. | |
2914 | */ | |
2915 | if (!balance) | |
2916 | break; | |
1da177e4 | 2917 | } |
46cb4b7c SS |
2918 | rq->next_balance = next_balance; |
2919 | } | |
2920 | ||
2921 | /* | |
2922 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
2923 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
2924 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
2925 | */ | |
2926 | static void run_rebalance_domains(struct softirq_action *h) | |
2927 | { | |
2928 | int local_cpu = smp_processor_id(); | |
2929 | struct rq *local_rq = cpu_rq(local_cpu); | |
d15bcfdb | 2930 | enum cpu_idle_type idle = local_rq->idle_at_tick ? CPU_IDLE : CPU_NOT_IDLE; |
46cb4b7c SS |
2931 | |
2932 | rebalance_domains(local_cpu, idle); | |
2933 | ||
2934 | #ifdef CONFIG_NO_HZ | |
2935 | /* | |
2936 | * If this cpu is the owner for idle load balancing, then do the | |
2937 | * balancing on behalf of the other idle cpus whose ticks are | |
2938 | * stopped. | |
2939 | */ | |
2940 | if (local_rq->idle_at_tick && | |
2941 | atomic_read(&nohz.load_balancer) == local_cpu) { | |
2942 | cpumask_t cpus = nohz.cpu_mask; | |
2943 | struct rq *rq; | |
2944 | int balance_cpu; | |
2945 | ||
2946 | cpu_clear(local_cpu, cpus); | |
2947 | for_each_cpu_mask(balance_cpu, cpus) { | |
2948 | /* | |
2949 | * If this cpu gets work to do, stop the load balancing | |
2950 | * work being done for other cpus. Next load | |
2951 | * balancing owner will pick it up. | |
2952 | */ | |
2953 | if (need_resched()) | |
2954 | break; | |
2955 | ||
d15bcfdb | 2956 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
2957 | |
2958 | rq = cpu_rq(balance_cpu); | |
2959 | if (time_after(local_rq->next_balance, rq->next_balance)) | |
2960 | local_rq->next_balance = rq->next_balance; | |
2961 | } | |
2962 | } | |
2963 | #endif | |
2964 | } | |
2965 | ||
2966 | /* | |
2967 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
2968 | * | |
2969 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
2970 | * idle load balancing owner or decide to stop the periodic load balancing, | |
2971 | * if the whole system is idle. | |
2972 | */ | |
2973 | static inline void trigger_load_balance(int cpu) | |
2974 | { | |
2975 | struct rq *rq = cpu_rq(cpu); | |
2976 | #ifdef CONFIG_NO_HZ | |
2977 | /* | |
2978 | * If we were in the nohz mode recently and busy at the current | |
2979 | * scheduler tick, then check if we need to nominate new idle | |
2980 | * load balancer. | |
2981 | */ | |
2982 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
2983 | rq->in_nohz_recently = 0; | |
2984 | ||
2985 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
2986 | cpu_clear(cpu, nohz.cpu_mask); | |
2987 | atomic_set(&nohz.load_balancer, -1); | |
2988 | } | |
2989 | ||
2990 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2991 | /* | |
2992 | * simple selection for now: Nominate the | |
2993 | * first cpu in the nohz list to be the next | |
2994 | * ilb owner. | |
2995 | * | |
2996 | * TBD: Traverse the sched domains and nominate | |
2997 | * the nearest cpu in the nohz.cpu_mask. | |
2998 | */ | |
2999 | int ilb = first_cpu(nohz.cpu_mask); | |
3000 | ||
3001 | if (ilb != NR_CPUS) | |
3002 | resched_cpu(ilb); | |
3003 | } | |
3004 | } | |
3005 | ||
3006 | /* | |
3007 | * If this cpu is idle and doing idle load balancing for all the | |
3008 | * cpus with ticks stopped, is it time for that to stop? | |
3009 | */ | |
3010 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3011 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3012 | resched_cpu(cpu); | |
3013 | return; | |
3014 | } | |
3015 | ||
3016 | /* | |
3017 | * If this cpu is idle and the idle load balancing is done by | |
3018 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3019 | */ | |
3020 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3021 | cpu_isset(cpu, nohz.cpu_mask)) | |
3022 | return; | |
3023 | #endif | |
3024 | if (time_after_eq(jiffies, rq->next_balance)) | |
3025 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 LT |
3026 | } |
3027 | #else | |
3028 | /* | |
3029 | * on UP we do not need to balance between CPUs: | |
3030 | */ | |
70b97a7f | 3031 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3032 | { |
3033 | } | |
3034 | #endif | |
3035 | ||
1da177e4 LT |
3036 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3037 | ||
3038 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3039 | ||
3040 | /* | |
3041 | * This is called on clock ticks and on context switches. | |
3042 | * Bank in p->sched_time the ns elapsed since the last tick or switch. | |
3043 | */ | |
48f24c4d | 3044 | static inline void |
70b97a7f | 3045 | update_cpu_clock(struct task_struct *p, struct rq *rq, unsigned long long now) |
1da177e4 | 3046 | { |
b18ec803 MG |
3047 | p->sched_time += now - p->last_ran; |
3048 | p->last_ran = rq->most_recent_timestamp = now; | |
1da177e4 LT |
3049 | } |
3050 | ||
3051 | /* | |
3052 | * Return current->sched_time plus any more ns on the sched_clock | |
3053 | * that have not yet been banked. | |
3054 | */ | |
36c8b586 | 3055 | unsigned long long current_sched_time(const struct task_struct *p) |
1da177e4 LT |
3056 | { |
3057 | unsigned long long ns; | |
3058 | unsigned long flags; | |
48f24c4d | 3059 | |
1da177e4 | 3060 | local_irq_save(flags); |
b18ec803 | 3061 | ns = p->sched_time + sched_clock() - p->last_ran; |
1da177e4 | 3062 | local_irq_restore(flags); |
48f24c4d | 3063 | |
1da177e4 LT |
3064 | return ns; |
3065 | } | |
3066 | ||
f1adad78 LT |
3067 | /* |
3068 | * We place interactive tasks back into the active array, if possible. | |
3069 | * | |
3070 | * To guarantee that this does not starve expired tasks we ignore the | |
3071 | * interactivity of a task if the first expired task had to wait more | |
3072 | * than a 'reasonable' amount of time. This deadline timeout is | |
3073 | * load-dependent, as the frequency of array switched decreases with | |
3074 | * increasing number of running tasks. We also ignore the interactivity | |
3075 | * if a better static_prio task has expired: | |
3076 | */ | |
70b97a7f | 3077 | static inline int expired_starving(struct rq *rq) |
48f24c4d IM |
3078 | { |
3079 | if (rq->curr->static_prio > rq->best_expired_prio) | |
3080 | return 1; | |
3081 | if (!STARVATION_LIMIT || !rq->expired_timestamp) | |
3082 | return 0; | |
3083 | if (jiffies - rq->expired_timestamp > STARVATION_LIMIT * rq->nr_running) | |
3084 | return 1; | |
3085 | return 0; | |
3086 | } | |
f1adad78 | 3087 | |
1da177e4 LT |
3088 | /* |
3089 | * Account user cpu time to a process. | |
3090 | * @p: the process that the cpu time gets accounted to | |
3091 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3092 | * @cputime: the cpu time spent in user space since the last update | |
3093 | */ | |
3094 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3095 | { | |
3096 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3097 | cputime64_t tmp; | |
3098 | ||
3099 | p->utime = cputime_add(p->utime, cputime); | |
3100 | ||
3101 | /* Add user time to cpustat. */ | |
3102 | tmp = cputime_to_cputime64(cputime); | |
3103 | if (TASK_NICE(p) > 0) | |
3104 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3105 | else | |
3106 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * Account system cpu time to a process. | |
3111 | * @p: the process that the cpu time gets accounted to | |
3112 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3113 | * @cputime: the cpu time spent in kernel space since the last update | |
3114 | */ | |
3115 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3116 | cputime_t cputime) | |
3117 | { | |
3118 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3119 | struct rq *rq = this_rq(); |
1da177e4 LT |
3120 | cputime64_t tmp; |
3121 | ||
3122 | p->stime = cputime_add(p->stime, cputime); | |
3123 | ||
3124 | /* Add system time to cpustat. */ | |
3125 | tmp = cputime_to_cputime64(cputime); | |
3126 | if (hardirq_count() - hardirq_offset) | |
3127 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3128 | else if (softirq_count()) | |
3129 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3130 | else if (p != rq->idle) | |
3131 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3132 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3133 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3134 | else | |
3135 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3136 | /* Account for system time used */ | |
3137 | acct_update_integrals(p); | |
1da177e4 LT |
3138 | } |
3139 | ||
3140 | /* | |
3141 | * Account for involuntary wait time. | |
3142 | * @p: the process from which the cpu time has been stolen | |
3143 | * @steal: the cpu time spent in involuntary wait | |
3144 | */ | |
3145 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3146 | { | |
3147 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3148 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3149 | struct rq *rq = this_rq(); |
1da177e4 LT |
3150 | |
3151 | if (p == rq->idle) { | |
3152 | p->stime = cputime_add(p->stime, steal); | |
3153 | if (atomic_read(&rq->nr_iowait) > 0) | |
3154 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3155 | else | |
3156 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3157 | } else | |
3158 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3159 | } | |
3160 | ||
7835b98b | 3161 | static void task_running_tick(struct rq *rq, struct task_struct *p) |
1da177e4 | 3162 | { |
1da177e4 | 3163 | if (p->array != rq->active) { |
7835b98b | 3164 | /* Task has expired but was not scheduled yet */ |
1da177e4 | 3165 | set_tsk_need_resched(p); |
7835b98b | 3166 | return; |
1da177e4 LT |
3167 | } |
3168 | spin_lock(&rq->lock); | |
3169 | /* | |
3170 | * The task was running during this tick - update the | |
3171 | * time slice counter. Note: we do not update a thread's | |
3172 | * priority until it either goes to sleep or uses up its | |
3173 | * timeslice. This makes it possible for interactive tasks | |
3174 | * to use up their timeslices at their highest priority levels. | |
3175 | */ | |
3176 | if (rt_task(p)) { | |
3177 | /* | |
3178 | * RR tasks need a special form of timeslice management. | |
3179 | * FIFO tasks have no timeslices. | |
3180 | */ | |
3181 | if ((p->policy == SCHED_RR) && !--p->time_slice) { | |
3182 | p->time_slice = task_timeslice(p); | |
3183 | p->first_time_slice = 0; | |
3184 | set_tsk_need_resched(p); | |
3185 | ||
3186 | /* put it at the end of the queue: */ | |
3187 | requeue_task(p, rq->active); | |
3188 | } | |
3189 | goto out_unlock; | |
3190 | } | |
3191 | if (!--p->time_slice) { | |
3192 | dequeue_task(p, rq->active); | |
3193 | set_tsk_need_resched(p); | |
3194 | p->prio = effective_prio(p); | |
3195 | p->time_slice = task_timeslice(p); | |
3196 | p->first_time_slice = 0; | |
3197 | ||
3198 | if (!rq->expired_timestamp) | |
3199 | rq->expired_timestamp = jiffies; | |
48f24c4d | 3200 | if (!TASK_INTERACTIVE(p) || expired_starving(rq)) { |
1da177e4 LT |
3201 | enqueue_task(p, rq->expired); |
3202 | if (p->static_prio < rq->best_expired_prio) | |
3203 | rq->best_expired_prio = p->static_prio; | |
3204 | } else | |
3205 | enqueue_task(p, rq->active); | |
3206 | } else { | |
3207 | /* | |
3208 | * Prevent a too long timeslice allowing a task to monopolize | |
3209 | * the CPU. We do this by splitting up the timeslice into | |
3210 | * smaller pieces. | |
3211 | * | |
3212 | * Note: this does not mean the task's timeslices expire or | |
3213 | * get lost in any way, they just might be preempted by | |
3214 | * another task of equal priority. (one with higher | |
3215 | * priority would have preempted this task already.) We | |
3216 | * requeue this task to the end of the list on this priority | |
3217 | * level, which is in essence a round-robin of tasks with | |
3218 | * equal priority. | |
3219 | * | |
3220 | * This only applies to tasks in the interactive | |
3221 | * delta range with at least TIMESLICE_GRANULARITY to requeue. | |
3222 | */ | |
3223 | if (TASK_INTERACTIVE(p) && !((task_timeslice(p) - | |
3224 | p->time_slice) % TIMESLICE_GRANULARITY(p)) && | |
3225 | (p->time_slice >= TIMESLICE_GRANULARITY(p)) && | |
3226 | (p->array == rq->active)) { | |
3227 | ||
3228 | requeue_task(p, rq->active); | |
3229 | set_tsk_need_resched(p); | |
3230 | } | |
3231 | } | |
3232 | out_unlock: | |
3233 | spin_unlock(&rq->lock); | |
7835b98b CL |
3234 | } |
3235 | ||
3236 | /* | |
3237 | * This function gets called by the timer code, with HZ frequency. | |
3238 | * We call it with interrupts disabled. | |
3239 | * | |
3240 | * It also gets called by the fork code, when changing the parent's | |
3241 | * timeslices. | |
3242 | */ | |
3243 | void scheduler_tick(void) | |
3244 | { | |
3245 | unsigned long long now = sched_clock(); | |
3246 | struct task_struct *p = current; | |
3247 | int cpu = smp_processor_id(); | |
bdecea3a | 3248 | int idle_at_tick = idle_cpu(cpu); |
7835b98b | 3249 | struct rq *rq = cpu_rq(cpu); |
7835b98b CL |
3250 | |
3251 | update_cpu_clock(p, rq, now); | |
3252 | ||
bdecea3a | 3253 | if (!idle_at_tick) |
7835b98b | 3254 | task_running_tick(rq, p); |
e418e1c2 | 3255 | #ifdef CONFIG_SMP |
7835b98b | 3256 | update_load(rq); |
bdecea3a | 3257 | rq->idle_at_tick = idle_at_tick; |
46cb4b7c | 3258 | trigger_load_balance(cpu); |
e418e1c2 | 3259 | #endif |
1da177e4 LT |
3260 | } |
3261 | ||
1da177e4 LT |
3262 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3263 | ||
3264 | void fastcall add_preempt_count(int val) | |
3265 | { | |
3266 | /* | |
3267 | * Underflow? | |
3268 | */ | |
9a11b49a IM |
3269 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3270 | return; | |
1da177e4 LT |
3271 | preempt_count() += val; |
3272 | /* | |
3273 | * Spinlock count overflowing soon? | |
3274 | */ | |
33859f7f MOS |
3275 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3276 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3277 | } |
3278 | EXPORT_SYMBOL(add_preempt_count); | |
3279 | ||
3280 | void fastcall sub_preempt_count(int val) | |
3281 | { | |
3282 | /* | |
3283 | * Underflow? | |
3284 | */ | |
9a11b49a IM |
3285 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3286 | return; | |
1da177e4 LT |
3287 | /* |
3288 | * Is the spinlock portion underflowing? | |
3289 | */ | |
9a11b49a IM |
3290 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3291 | !(preempt_count() & PREEMPT_MASK))) | |
3292 | return; | |
3293 | ||
1da177e4 LT |
3294 | preempt_count() -= val; |
3295 | } | |
3296 | EXPORT_SYMBOL(sub_preempt_count); | |
3297 | ||
3298 | #endif | |
3299 | ||
3dee386e CK |
3300 | static inline int interactive_sleep(enum sleep_type sleep_type) |
3301 | { | |
3302 | return (sleep_type == SLEEP_INTERACTIVE || | |
3303 | sleep_type == SLEEP_INTERRUPTED); | |
3304 | } | |
3305 | ||
1da177e4 LT |
3306 | /* |
3307 | * schedule() is the main scheduler function. | |
3308 | */ | |
3309 | asmlinkage void __sched schedule(void) | |
3310 | { | |
36c8b586 | 3311 | struct task_struct *prev, *next; |
70b97a7f | 3312 | struct prio_array *array; |
1da177e4 LT |
3313 | struct list_head *queue; |
3314 | unsigned long long now; | |
3315 | unsigned long run_time; | |
a3464a10 | 3316 | int cpu, idx, new_prio; |
48f24c4d | 3317 | long *switch_count; |
70b97a7f | 3318 | struct rq *rq; |
1da177e4 LT |
3319 | |
3320 | /* | |
3321 | * Test if we are atomic. Since do_exit() needs to call into | |
3322 | * schedule() atomically, we ignore that path for now. | |
3323 | * Otherwise, whine if we are scheduling when we should not be. | |
3324 | */ | |
77e4bfbc AM |
3325 | if (unlikely(in_atomic() && !current->exit_state)) { |
3326 | printk(KERN_ERR "BUG: scheduling while atomic: " | |
3327 | "%s/0x%08x/%d\n", | |
3328 | current->comm, preempt_count(), current->pid); | |
a4c410f0 | 3329 | debug_show_held_locks(current); |
3117df04 IM |
3330 | if (irqs_disabled()) |
3331 | print_irqtrace_events(current); | |
77e4bfbc | 3332 | dump_stack(); |
1da177e4 LT |
3333 | } |
3334 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); | |
3335 | ||
3336 | need_resched: | |
3337 | preempt_disable(); | |
3338 | prev = current; | |
3339 | release_kernel_lock(prev); | |
3340 | need_resched_nonpreemptible: | |
3341 | rq = this_rq(); | |
3342 | ||
3343 | /* | |
3344 | * The idle thread is not allowed to schedule! | |
3345 | * Remove this check after it has been exercised a bit. | |
3346 | */ | |
3347 | if (unlikely(prev == rq->idle) && prev->state != TASK_RUNNING) { | |
3348 | printk(KERN_ERR "bad: scheduling from the idle thread!\n"); | |
3349 | dump_stack(); | |
3350 | } | |
3351 | ||
3352 | schedstat_inc(rq, sched_cnt); | |
3353 | now = sched_clock(); | |
238628ed | 3354 | if (likely((long long)(now - prev->timestamp) < NS_MAX_SLEEP_AVG)) { |
1da177e4 | 3355 | run_time = now - prev->timestamp; |
238628ed | 3356 | if (unlikely((long long)(now - prev->timestamp) < 0)) |
1da177e4 LT |
3357 | run_time = 0; |
3358 | } else | |
3359 | run_time = NS_MAX_SLEEP_AVG; | |
3360 | ||
3361 | /* | |
3362 | * Tasks charged proportionately less run_time at high sleep_avg to | |
3363 | * delay them losing their interactive status | |
3364 | */ | |
3365 | run_time /= (CURRENT_BONUS(prev) ? : 1); | |
3366 | ||
3367 | spin_lock_irq(&rq->lock); | |
3368 | ||
1da177e4 LT |
3369 | switch_count = &prev->nivcsw; |
3370 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { | |
3371 | switch_count = &prev->nvcsw; | |
3372 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && | |
3373 | unlikely(signal_pending(prev)))) | |
3374 | prev->state = TASK_RUNNING; | |
3375 | else { | |
3376 | if (prev->state == TASK_UNINTERRUPTIBLE) | |
3377 | rq->nr_uninterruptible++; | |
3378 | deactivate_task(prev, rq); | |
3379 | } | |
3380 | } | |
3381 | ||
3382 | cpu = smp_processor_id(); | |
3383 | if (unlikely(!rq->nr_running)) { | |
1da177e4 LT |
3384 | idle_balance(cpu, rq); |
3385 | if (!rq->nr_running) { | |
3386 | next = rq->idle; | |
3387 | rq->expired_timestamp = 0; | |
1da177e4 LT |
3388 | goto switch_tasks; |
3389 | } | |
1da177e4 LT |
3390 | } |
3391 | ||
3392 | array = rq->active; | |
3393 | if (unlikely(!array->nr_active)) { | |
3394 | /* | |
3395 | * Switch the active and expired arrays. | |
3396 | */ | |
3397 | schedstat_inc(rq, sched_switch); | |
3398 | rq->active = rq->expired; | |
3399 | rq->expired = array; | |
3400 | array = rq->active; | |
3401 | rq->expired_timestamp = 0; | |
3402 | rq->best_expired_prio = MAX_PRIO; | |
3403 | } | |
3404 | ||
3405 | idx = sched_find_first_bit(array->bitmap); | |
3406 | queue = array->queue + idx; | |
36c8b586 | 3407 | next = list_entry(queue->next, struct task_struct, run_list); |
1da177e4 | 3408 | |
3dee386e | 3409 | if (!rt_task(next) && interactive_sleep(next->sleep_type)) { |
1da177e4 | 3410 | unsigned long long delta = now - next->timestamp; |
238628ed | 3411 | if (unlikely((long long)(now - next->timestamp) < 0)) |
1da177e4 LT |
3412 | delta = 0; |
3413 | ||
3dee386e | 3414 | if (next->sleep_type == SLEEP_INTERACTIVE) |
1da177e4 LT |
3415 | delta = delta * (ON_RUNQUEUE_WEIGHT * 128 / 100) / 128; |
3416 | ||
3417 | array = next->array; | |
a3464a10 CS |
3418 | new_prio = recalc_task_prio(next, next->timestamp + delta); |
3419 | ||
3420 | if (unlikely(next->prio != new_prio)) { | |
3421 | dequeue_task(next, array); | |
3422 | next->prio = new_prio; | |
3423 | enqueue_task(next, array); | |
7c4bb1f9 | 3424 | } |
1da177e4 | 3425 | } |
3dee386e | 3426 | next->sleep_type = SLEEP_NORMAL; |
1da177e4 LT |
3427 | switch_tasks: |
3428 | if (next == rq->idle) | |
3429 | schedstat_inc(rq, sched_goidle); | |
3430 | prefetch(next); | |
383f2835 | 3431 | prefetch_stack(next); |
1da177e4 LT |
3432 | clear_tsk_need_resched(prev); |
3433 | rcu_qsctr_inc(task_cpu(prev)); | |
3434 | ||
3435 | update_cpu_clock(prev, rq, now); | |
3436 | ||
3437 | prev->sleep_avg -= run_time; | |
3438 | if ((long)prev->sleep_avg <= 0) | |
3439 | prev->sleep_avg = 0; | |
3440 | prev->timestamp = prev->last_ran = now; | |
3441 | ||
3442 | sched_info_switch(prev, next); | |
3443 | if (likely(prev != next)) { | |
c1e16aa2 | 3444 | next->timestamp = next->last_ran = now; |
1da177e4 LT |
3445 | rq->nr_switches++; |
3446 | rq->curr = next; | |
3447 | ++*switch_count; | |
3448 | ||
4866cde0 | 3449 | prepare_task_switch(rq, next); |
1da177e4 LT |
3450 | prev = context_switch(rq, prev, next); |
3451 | barrier(); | |
4866cde0 NP |
3452 | /* |
3453 | * this_rq must be evaluated again because prev may have moved | |
3454 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3455 | * frame will be invalid. | |
3456 | */ | |
3457 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3458 | } else |
3459 | spin_unlock_irq(&rq->lock); | |
3460 | ||
3461 | prev = current; | |
3462 | if (unlikely(reacquire_kernel_lock(prev) < 0)) | |
3463 | goto need_resched_nonpreemptible; | |
3464 | preempt_enable_no_resched(); | |
3465 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3466 | goto need_resched; | |
3467 | } | |
1da177e4 LT |
3468 | EXPORT_SYMBOL(schedule); |
3469 | ||
3470 | #ifdef CONFIG_PREEMPT | |
3471 | /* | |
2ed6e34f | 3472 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3473 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3474 | * occur there and call schedule directly. | |
3475 | */ | |
3476 | asmlinkage void __sched preempt_schedule(void) | |
3477 | { | |
3478 | struct thread_info *ti = current_thread_info(); | |
3479 | #ifdef CONFIG_PREEMPT_BKL | |
3480 | struct task_struct *task = current; | |
3481 | int saved_lock_depth; | |
3482 | #endif | |
3483 | /* | |
3484 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3485 | * we do not want to preempt the current task. Just return.. | |
3486 | */ | |
beed33a8 | 3487 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3488 | return; |
3489 | ||
3490 | need_resched: | |
3491 | add_preempt_count(PREEMPT_ACTIVE); | |
3492 | /* | |
3493 | * We keep the big kernel semaphore locked, but we | |
3494 | * clear ->lock_depth so that schedule() doesnt | |
3495 | * auto-release the semaphore: | |
3496 | */ | |
3497 | #ifdef CONFIG_PREEMPT_BKL | |
3498 | saved_lock_depth = task->lock_depth; | |
3499 | task->lock_depth = -1; | |
3500 | #endif | |
3501 | schedule(); | |
3502 | #ifdef CONFIG_PREEMPT_BKL | |
3503 | task->lock_depth = saved_lock_depth; | |
3504 | #endif | |
3505 | sub_preempt_count(PREEMPT_ACTIVE); | |
3506 | ||
3507 | /* we could miss a preemption opportunity between schedule and now */ | |
3508 | barrier(); | |
3509 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3510 | goto need_resched; | |
3511 | } | |
1da177e4 LT |
3512 | EXPORT_SYMBOL(preempt_schedule); |
3513 | ||
3514 | /* | |
2ed6e34f | 3515 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3516 | * off of irq context. |
3517 | * Note, that this is called and return with irqs disabled. This will | |
3518 | * protect us against recursive calling from irq. | |
3519 | */ | |
3520 | asmlinkage void __sched preempt_schedule_irq(void) | |
3521 | { | |
3522 | struct thread_info *ti = current_thread_info(); | |
3523 | #ifdef CONFIG_PREEMPT_BKL | |
3524 | struct task_struct *task = current; | |
3525 | int saved_lock_depth; | |
3526 | #endif | |
2ed6e34f | 3527 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3528 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3529 | ||
3530 | need_resched: | |
3531 | add_preempt_count(PREEMPT_ACTIVE); | |
3532 | /* | |
3533 | * We keep the big kernel semaphore locked, but we | |
3534 | * clear ->lock_depth so that schedule() doesnt | |
3535 | * auto-release the semaphore: | |
3536 | */ | |
3537 | #ifdef CONFIG_PREEMPT_BKL | |
3538 | saved_lock_depth = task->lock_depth; | |
3539 | task->lock_depth = -1; | |
3540 | #endif | |
3541 | local_irq_enable(); | |
3542 | schedule(); | |
3543 | local_irq_disable(); | |
3544 | #ifdef CONFIG_PREEMPT_BKL | |
3545 | task->lock_depth = saved_lock_depth; | |
3546 | #endif | |
3547 | sub_preempt_count(PREEMPT_ACTIVE); | |
3548 | ||
3549 | /* we could miss a preemption opportunity between schedule and now */ | |
3550 | barrier(); | |
3551 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3552 | goto need_resched; | |
3553 | } | |
3554 | ||
3555 | #endif /* CONFIG_PREEMPT */ | |
3556 | ||
95cdf3b7 IM |
3557 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3558 | void *key) | |
1da177e4 | 3559 | { |
48f24c4d | 3560 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3561 | } |
1da177e4 LT |
3562 | EXPORT_SYMBOL(default_wake_function); |
3563 | ||
3564 | /* | |
3565 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3566 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3567 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3568 | * | |
3569 | * There are circumstances in which we can try to wake a task which has already | |
3570 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3571 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3572 | */ | |
3573 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3574 | int nr_exclusive, int sync, void *key) | |
3575 | { | |
3576 | struct list_head *tmp, *next; | |
3577 | ||
3578 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3579 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3580 | unsigned flags = curr->flags; | |
3581 | ||
1da177e4 | 3582 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3583 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3584 | break; |
3585 | } | |
3586 | } | |
3587 | ||
3588 | /** | |
3589 | * __wake_up - wake up threads blocked on a waitqueue. | |
3590 | * @q: the waitqueue | |
3591 | * @mode: which threads | |
3592 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3593 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3594 | */ |
3595 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3596 | int nr_exclusive, void *key) |
1da177e4 LT |
3597 | { |
3598 | unsigned long flags; | |
3599 | ||
3600 | spin_lock_irqsave(&q->lock, flags); | |
3601 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3602 | spin_unlock_irqrestore(&q->lock, flags); | |
3603 | } | |
1da177e4 LT |
3604 | EXPORT_SYMBOL(__wake_up); |
3605 | ||
3606 | /* | |
3607 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3608 | */ | |
3609 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3610 | { | |
3611 | __wake_up_common(q, mode, 1, 0, NULL); | |
3612 | } | |
3613 | ||
3614 | /** | |
67be2dd1 | 3615 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3616 | * @q: the waitqueue |
3617 | * @mode: which threads | |
3618 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3619 | * | |
3620 | * The sync wakeup differs that the waker knows that it will schedule | |
3621 | * away soon, so while the target thread will be woken up, it will not | |
3622 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3623 | * with each other. This can prevent needless bouncing between CPUs. | |
3624 | * | |
3625 | * On UP it can prevent extra preemption. | |
3626 | */ | |
95cdf3b7 IM |
3627 | void fastcall |
3628 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3629 | { |
3630 | unsigned long flags; | |
3631 | int sync = 1; | |
3632 | ||
3633 | if (unlikely(!q)) | |
3634 | return; | |
3635 | ||
3636 | if (unlikely(!nr_exclusive)) | |
3637 | sync = 0; | |
3638 | ||
3639 | spin_lock_irqsave(&q->lock, flags); | |
3640 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3641 | spin_unlock_irqrestore(&q->lock, flags); | |
3642 | } | |
3643 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3644 | ||
3645 | void fastcall complete(struct completion *x) | |
3646 | { | |
3647 | unsigned long flags; | |
3648 | ||
3649 | spin_lock_irqsave(&x->wait.lock, flags); | |
3650 | x->done++; | |
3651 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3652 | 1, 0, NULL); | |
3653 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3654 | } | |
3655 | EXPORT_SYMBOL(complete); | |
3656 | ||
3657 | void fastcall complete_all(struct completion *x) | |
3658 | { | |
3659 | unsigned long flags; | |
3660 | ||
3661 | spin_lock_irqsave(&x->wait.lock, flags); | |
3662 | x->done += UINT_MAX/2; | |
3663 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3664 | 0, 0, NULL); | |
3665 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3666 | } | |
3667 | EXPORT_SYMBOL(complete_all); | |
3668 | ||
3669 | void fastcall __sched wait_for_completion(struct completion *x) | |
3670 | { | |
3671 | might_sleep(); | |
48f24c4d | 3672 | |
1da177e4 LT |
3673 | spin_lock_irq(&x->wait.lock); |
3674 | if (!x->done) { | |
3675 | DECLARE_WAITQUEUE(wait, current); | |
3676 | ||
3677 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3678 | __add_wait_queue_tail(&x->wait, &wait); | |
3679 | do { | |
3680 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3681 | spin_unlock_irq(&x->wait.lock); | |
3682 | schedule(); | |
3683 | spin_lock_irq(&x->wait.lock); | |
3684 | } while (!x->done); | |
3685 | __remove_wait_queue(&x->wait, &wait); | |
3686 | } | |
3687 | x->done--; | |
3688 | spin_unlock_irq(&x->wait.lock); | |
3689 | } | |
3690 | EXPORT_SYMBOL(wait_for_completion); | |
3691 | ||
3692 | unsigned long fastcall __sched | |
3693 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3694 | { | |
3695 | might_sleep(); | |
3696 | ||
3697 | spin_lock_irq(&x->wait.lock); | |
3698 | if (!x->done) { | |
3699 | DECLARE_WAITQUEUE(wait, current); | |
3700 | ||
3701 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3702 | __add_wait_queue_tail(&x->wait, &wait); | |
3703 | do { | |
3704 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3705 | spin_unlock_irq(&x->wait.lock); | |
3706 | timeout = schedule_timeout(timeout); | |
3707 | spin_lock_irq(&x->wait.lock); | |
3708 | if (!timeout) { | |
3709 | __remove_wait_queue(&x->wait, &wait); | |
3710 | goto out; | |
3711 | } | |
3712 | } while (!x->done); | |
3713 | __remove_wait_queue(&x->wait, &wait); | |
3714 | } | |
3715 | x->done--; | |
3716 | out: | |
3717 | spin_unlock_irq(&x->wait.lock); | |
3718 | return timeout; | |
3719 | } | |
3720 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3721 | ||
3722 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3723 | { | |
3724 | int ret = 0; | |
3725 | ||
3726 | might_sleep(); | |
3727 | ||
3728 | spin_lock_irq(&x->wait.lock); | |
3729 | if (!x->done) { | |
3730 | DECLARE_WAITQUEUE(wait, current); | |
3731 | ||
3732 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3733 | __add_wait_queue_tail(&x->wait, &wait); | |
3734 | do { | |
3735 | if (signal_pending(current)) { | |
3736 | ret = -ERESTARTSYS; | |
3737 | __remove_wait_queue(&x->wait, &wait); | |
3738 | goto out; | |
3739 | } | |
3740 | __set_current_state(TASK_INTERRUPTIBLE); | |
3741 | spin_unlock_irq(&x->wait.lock); | |
3742 | schedule(); | |
3743 | spin_lock_irq(&x->wait.lock); | |
3744 | } while (!x->done); | |
3745 | __remove_wait_queue(&x->wait, &wait); | |
3746 | } | |
3747 | x->done--; | |
3748 | out: | |
3749 | spin_unlock_irq(&x->wait.lock); | |
3750 | ||
3751 | return ret; | |
3752 | } | |
3753 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3754 | ||
3755 | unsigned long fastcall __sched | |
3756 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3757 | unsigned long timeout) | |
3758 | { | |
3759 | might_sleep(); | |
3760 | ||
3761 | spin_lock_irq(&x->wait.lock); | |
3762 | if (!x->done) { | |
3763 | DECLARE_WAITQUEUE(wait, current); | |
3764 | ||
3765 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3766 | __add_wait_queue_tail(&x->wait, &wait); | |
3767 | do { | |
3768 | if (signal_pending(current)) { | |
3769 | timeout = -ERESTARTSYS; | |
3770 | __remove_wait_queue(&x->wait, &wait); | |
3771 | goto out; | |
3772 | } | |
3773 | __set_current_state(TASK_INTERRUPTIBLE); | |
3774 | spin_unlock_irq(&x->wait.lock); | |
3775 | timeout = schedule_timeout(timeout); | |
3776 | spin_lock_irq(&x->wait.lock); | |
3777 | if (!timeout) { | |
3778 | __remove_wait_queue(&x->wait, &wait); | |
3779 | goto out; | |
3780 | } | |
3781 | } while (!x->done); | |
3782 | __remove_wait_queue(&x->wait, &wait); | |
3783 | } | |
3784 | x->done--; | |
3785 | out: | |
3786 | spin_unlock_irq(&x->wait.lock); | |
3787 | return timeout; | |
3788 | } | |
3789 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3790 | ||
3791 | ||
3792 | #define SLEEP_ON_VAR \ | |
3793 | unsigned long flags; \ | |
3794 | wait_queue_t wait; \ | |
3795 | init_waitqueue_entry(&wait, current); | |
3796 | ||
3797 | #define SLEEP_ON_HEAD \ | |
3798 | spin_lock_irqsave(&q->lock,flags); \ | |
3799 | __add_wait_queue(q, &wait); \ | |
3800 | spin_unlock(&q->lock); | |
3801 | ||
3802 | #define SLEEP_ON_TAIL \ | |
3803 | spin_lock_irq(&q->lock); \ | |
3804 | __remove_wait_queue(q, &wait); \ | |
3805 | spin_unlock_irqrestore(&q->lock, flags); | |
3806 | ||
3807 | void fastcall __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3808 | { | |
3809 | SLEEP_ON_VAR | |
3810 | ||
3811 | current->state = TASK_INTERRUPTIBLE; | |
3812 | ||
3813 | SLEEP_ON_HEAD | |
3814 | schedule(); | |
3815 | SLEEP_ON_TAIL | |
3816 | } | |
1da177e4 LT |
3817 | EXPORT_SYMBOL(interruptible_sleep_on); |
3818 | ||
95cdf3b7 IM |
3819 | long fastcall __sched |
3820 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
1da177e4 LT |
3821 | { |
3822 | SLEEP_ON_VAR | |
3823 | ||
3824 | current->state = TASK_INTERRUPTIBLE; | |
3825 | ||
3826 | SLEEP_ON_HEAD | |
3827 | timeout = schedule_timeout(timeout); | |
3828 | SLEEP_ON_TAIL | |
3829 | ||
3830 | return timeout; | |
3831 | } | |
1da177e4 LT |
3832 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3833 | ||
3834 | void fastcall __sched sleep_on(wait_queue_head_t *q) | |
3835 | { | |
3836 | SLEEP_ON_VAR | |
3837 | ||
3838 | current->state = TASK_UNINTERRUPTIBLE; | |
3839 | ||
3840 | SLEEP_ON_HEAD | |
3841 | schedule(); | |
3842 | SLEEP_ON_TAIL | |
3843 | } | |
1da177e4 LT |
3844 | EXPORT_SYMBOL(sleep_on); |
3845 | ||
3846 | long fastcall __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) | |
3847 | { | |
3848 | SLEEP_ON_VAR | |
3849 | ||
3850 | current->state = TASK_UNINTERRUPTIBLE; | |
3851 | ||
3852 | SLEEP_ON_HEAD | |
3853 | timeout = schedule_timeout(timeout); | |
3854 | SLEEP_ON_TAIL | |
3855 | ||
3856 | return timeout; | |
3857 | } | |
3858 | ||
3859 | EXPORT_SYMBOL(sleep_on_timeout); | |
3860 | ||
b29739f9 IM |
3861 | #ifdef CONFIG_RT_MUTEXES |
3862 | ||
3863 | /* | |
3864 | * rt_mutex_setprio - set the current priority of a task | |
3865 | * @p: task | |
3866 | * @prio: prio value (kernel-internal form) | |
3867 | * | |
3868 | * This function changes the 'effective' priority of a task. It does | |
3869 | * not touch ->normal_prio like __setscheduler(). | |
3870 | * | |
3871 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3872 | */ | |
36c8b586 | 3873 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 3874 | { |
70b97a7f | 3875 | struct prio_array *array; |
b29739f9 | 3876 | unsigned long flags; |
70b97a7f | 3877 | struct rq *rq; |
d5f9f942 | 3878 | int oldprio; |
b29739f9 IM |
3879 | |
3880 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3881 | ||
3882 | rq = task_rq_lock(p, &flags); | |
3883 | ||
d5f9f942 | 3884 | oldprio = p->prio; |
b29739f9 IM |
3885 | array = p->array; |
3886 | if (array) | |
3887 | dequeue_task(p, array); | |
3888 | p->prio = prio; | |
3889 | ||
3890 | if (array) { | |
3891 | /* | |
3892 | * If changing to an RT priority then queue it | |
3893 | * in the active array! | |
3894 | */ | |
3895 | if (rt_task(p)) | |
3896 | array = rq->active; | |
3897 | enqueue_task(p, array); | |
3898 | /* | |
3899 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3900 | * our priority decreased, or if we are not currently running on |
3901 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3902 | */ |
d5f9f942 AM |
3903 | if (task_running(rq, p)) { |
3904 | if (p->prio > oldprio) | |
3905 | resched_task(rq->curr); | |
3906 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
b29739f9 IM |
3907 | resched_task(rq->curr); |
3908 | } | |
3909 | task_rq_unlock(rq, &flags); | |
3910 | } | |
3911 | ||
3912 | #endif | |
3913 | ||
36c8b586 | 3914 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3915 | { |
70b97a7f | 3916 | struct prio_array *array; |
48f24c4d | 3917 | int old_prio, delta; |
1da177e4 | 3918 | unsigned long flags; |
70b97a7f | 3919 | struct rq *rq; |
1da177e4 LT |
3920 | |
3921 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3922 | return; | |
3923 | /* | |
3924 | * We have to be careful, if called from sys_setpriority(), | |
3925 | * the task might be in the middle of scheduling on another CPU. | |
3926 | */ | |
3927 | rq = task_rq_lock(p, &flags); | |
3928 | /* | |
3929 | * The RT priorities are set via sched_setscheduler(), but we still | |
3930 | * allow the 'normal' nice value to be set - but as expected | |
3931 | * it wont have any effect on scheduling until the task is | |
b0a9499c | 3932 | * not SCHED_NORMAL/SCHED_BATCH: |
1da177e4 | 3933 | */ |
b29739f9 | 3934 | if (has_rt_policy(p)) { |
1da177e4 LT |
3935 | p->static_prio = NICE_TO_PRIO(nice); |
3936 | goto out_unlock; | |
3937 | } | |
3938 | array = p->array; | |
2dd73a4f | 3939 | if (array) { |
1da177e4 | 3940 | dequeue_task(p, array); |
2dd73a4f PW |
3941 | dec_raw_weighted_load(rq, p); |
3942 | } | |
1da177e4 | 3943 | |
1da177e4 | 3944 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3945 | set_load_weight(p); |
b29739f9 IM |
3946 | old_prio = p->prio; |
3947 | p->prio = effective_prio(p); | |
3948 | delta = p->prio - old_prio; | |
1da177e4 LT |
3949 | |
3950 | if (array) { | |
3951 | enqueue_task(p, array); | |
2dd73a4f | 3952 | inc_raw_weighted_load(rq, p); |
1da177e4 | 3953 | /* |
d5f9f942 AM |
3954 | * If the task increased its priority or is running and |
3955 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3956 | */ |
d5f9f942 | 3957 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3958 | resched_task(rq->curr); |
3959 | } | |
3960 | out_unlock: | |
3961 | task_rq_unlock(rq, &flags); | |
3962 | } | |
1da177e4 LT |
3963 | EXPORT_SYMBOL(set_user_nice); |
3964 | ||
e43379f1 MM |
3965 | /* |
3966 | * can_nice - check if a task can reduce its nice value | |
3967 | * @p: task | |
3968 | * @nice: nice value | |
3969 | */ | |
36c8b586 | 3970 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3971 | { |
024f4747 MM |
3972 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3973 | int nice_rlim = 20 - nice; | |
48f24c4d | 3974 | |
e43379f1 MM |
3975 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
3976 | capable(CAP_SYS_NICE)); | |
3977 | } | |
3978 | ||
1da177e4 LT |
3979 | #ifdef __ARCH_WANT_SYS_NICE |
3980 | ||
3981 | /* | |
3982 | * sys_nice - change the priority of the current process. | |
3983 | * @increment: priority increment | |
3984 | * | |
3985 | * sys_setpriority is a more generic, but much slower function that | |
3986 | * does similar things. | |
3987 | */ | |
3988 | asmlinkage long sys_nice(int increment) | |
3989 | { | |
48f24c4d | 3990 | long nice, retval; |
1da177e4 LT |
3991 | |
3992 | /* | |
3993 | * Setpriority might change our priority at the same moment. | |
3994 | * We don't have to worry. Conceptually one call occurs first | |
3995 | * and we have a single winner. | |
3996 | */ | |
e43379f1 MM |
3997 | if (increment < -40) |
3998 | increment = -40; | |
1da177e4 LT |
3999 | if (increment > 40) |
4000 | increment = 40; | |
4001 | ||
4002 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4003 | if (nice < -20) | |
4004 | nice = -20; | |
4005 | if (nice > 19) | |
4006 | nice = 19; | |
4007 | ||
e43379f1 MM |
4008 | if (increment < 0 && !can_nice(current, nice)) |
4009 | return -EPERM; | |
4010 | ||
1da177e4 LT |
4011 | retval = security_task_setnice(current, nice); |
4012 | if (retval) | |
4013 | return retval; | |
4014 | ||
4015 | set_user_nice(current, nice); | |
4016 | return 0; | |
4017 | } | |
4018 | ||
4019 | #endif | |
4020 | ||
4021 | /** | |
4022 | * task_prio - return the priority value of a given task. | |
4023 | * @p: the task in question. | |
4024 | * | |
4025 | * This is the priority value as seen by users in /proc. | |
4026 | * RT tasks are offset by -200. Normal tasks are centered | |
4027 | * around 0, value goes from -16 to +15. | |
4028 | */ | |
36c8b586 | 4029 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4030 | { |
4031 | return p->prio - MAX_RT_PRIO; | |
4032 | } | |
4033 | ||
4034 | /** | |
4035 | * task_nice - return the nice value of a given task. | |
4036 | * @p: the task in question. | |
4037 | */ | |
36c8b586 | 4038 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4039 | { |
4040 | return TASK_NICE(p); | |
4041 | } | |
1da177e4 | 4042 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4043 | |
4044 | /** | |
4045 | * idle_cpu - is a given cpu idle currently? | |
4046 | * @cpu: the processor in question. | |
4047 | */ | |
4048 | int idle_cpu(int cpu) | |
4049 | { | |
4050 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4051 | } | |
4052 | ||
1da177e4 LT |
4053 | /** |
4054 | * idle_task - return the idle task for a given cpu. | |
4055 | * @cpu: the processor in question. | |
4056 | */ | |
36c8b586 | 4057 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4058 | { |
4059 | return cpu_rq(cpu)->idle; | |
4060 | } | |
4061 | ||
4062 | /** | |
4063 | * find_process_by_pid - find a process with a matching PID value. | |
4064 | * @pid: the pid in question. | |
4065 | */ | |
36c8b586 | 4066 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4067 | { |
4068 | return pid ? find_task_by_pid(pid) : current; | |
4069 | } | |
4070 | ||
4071 | /* Actually do priority change: must hold rq lock. */ | |
4072 | static void __setscheduler(struct task_struct *p, int policy, int prio) | |
4073 | { | |
4074 | BUG_ON(p->array); | |
48f24c4d | 4075 | |
1da177e4 LT |
4076 | p->policy = policy; |
4077 | p->rt_priority = prio; | |
b29739f9 IM |
4078 | p->normal_prio = normal_prio(p); |
4079 | /* we are holding p->pi_lock already */ | |
4080 | p->prio = rt_mutex_getprio(p); | |
4081 | /* | |
4082 | * SCHED_BATCH tasks are treated as perpetual CPU hogs: | |
4083 | */ | |
4084 | if (policy == SCHED_BATCH) | |
4085 | p->sleep_avg = 0; | |
2dd73a4f | 4086 | set_load_weight(p); |
1da177e4 LT |
4087 | } |
4088 | ||
4089 | /** | |
72fd4a35 | 4090 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4091 | * @p: the task in question. |
4092 | * @policy: new policy. | |
4093 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4094 | * |
72fd4a35 | 4095 | * NOTE that the task may be already dead. |
1da177e4 | 4096 | */ |
95cdf3b7 IM |
4097 | int sched_setscheduler(struct task_struct *p, int policy, |
4098 | struct sched_param *param) | |
1da177e4 | 4099 | { |
48f24c4d | 4100 | int retval, oldprio, oldpolicy = -1; |
70b97a7f | 4101 | struct prio_array *array; |
1da177e4 | 4102 | unsigned long flags; |
70b97a7f | 4103 | struct rq *rq; |
1da177e4 | 4104 | |
66e5393a SR |
4105 | /* may grab non-irq protected spin_locks */ |
4106 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4107 | recheck: |
4108 | /* double check policy once rq lock held */ | |
4109 | if (policy < 0) | |
4110 | policy = oldpolicy = p->policy; | |
4111 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
b0a9499c IM |
4112 | policy != SCHED_NORMAL && policy != SCHED_BATCH) |
4113 | return -EINVAL; | |
1da177e4 LT |
4114 | /* |
4115 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
b0a9499c IM |
4116 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL and |
4117 | * SCHED_BATCH is 0. | |
1da177e4 LT |
4118 | */ |
4119 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4120 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4121 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4122 | return -EINVAL; |
57a6f51c | 4123 | if (is_rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4124 | return -EINVAL; |
4125 | ||
37e4ab3f OC |
4126 | /* |
4127 | * Allow unprivileged RT tasks to decrease priority: | |
4128 | */ | |
4129 | if (!capable(CAP_SYS_NICE)) { | |
8dc3e909 ON |
4130 | if (is_rt_policy(policy)) { |
4131 | unsigned long rlim_rtprio; | |
4132 | unsigned long flags; | |
4133 | ||
4134 | if (!lock_task_sighand(p, &flags)) | |
4135 | return -ESRCH; | |
4136 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4137 | unlock_task_sighand(p, &flags); | |
4138 | ||
4139 | /* can't set/change the rt policy */ | |
4140 | if (policy != p->policy && !rlim_rtprio) | |
4141 | return -EPERM; | |
4142 | ||
4143 | /* can't increase priority */ | |
4144 | if (param->sched_priority > p->rt_priority && | |
4145 | param->sched_priority > rlim_rtprio) | |
4146 | return -EPERM; | |
4147 | } | |
5fe1d75f | 4148 | |
37e4ab3f OC |
4149 | /* can't change other user's priorities */ |
4150 | if ((current->euid != p->euid) && | |
4151 | (current->euid != p->uid)) | |
4152 | return -EPERM; | |
4153 | } | |
1da177e4 LT |
4154 | |
4155 | retval = security_task_setscheduler(p, policy, param); | |
4156 | if (retval) | |
4157 | return retval; | |
b29739f9 IM |
4158 | /* |
4159 | * make sure no PI-waiters arrive (or leave) while we are | |
4160 | * changing the priority of the task: | |
4161 | */ | |
4162 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4163 | /* |
4164 | * To be able to change p->policy safely, the apropriate | |
4165 | * runqueue lock must be held. | |
4166 | */ | |
b29739f9 | 4167 | rq = __task_rq_lock(p); |
1da177e4 LT |
4168 | /* recheck policy now with rq lock held */ |
4169 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4170 | policy = oldpolicy = -1; | |
b29739f9 IM |
4171 | __task_rq_unlock(rq); |
4172 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4173 | goto recheck; |
4174 | } | |
4175 | array = p->array; | |
4176 | if (array) | |
4177 | deactivate_task(p, rq); | |
4178 | oldprio = p->prio; | |
4179 | __setscheduler(p, policy, param->sched_priority); | |
4180 | if (array) { | |
4181 | __activate_task(p, rq); | |
4182 | /* | |
4183 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4184 | * our priority decreased, or if we are not currently running on |
4185 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4186 | */ |
d5f9f942 AM |
4187 | if (task_running(rq, p)) { |
4188 | if (p->prio > oldprio) | |
4189 | resched_task(rq->curr); | |
4190 | } else if (TASK_PREEMPTS_CURR(p, rq)) | |
1da177e4 LT |
4191 | resched_task(rq->curr); |
4192 | } | |
b29739f9 IM |
4193 | __task_rq_unlock(rq); |
4194 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4195 | ||
95e02ca9 TG |
4196 | rt_mutex_adjust_pi(p); |
4197 | ||
1da177e4 LT |
4198 | return 0; |
4199 | } | |
4200 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4201 | ||
95cdf3b7 IM |
4202 | static int |
4203 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4204 | { |
1da177e4 LT |
4205 | struct sched_param lparam; |
4206 | struct task_struct *p; | |
36c8b586 | 4207 | int retval; |
1da177e4 LT |
4208 | |
4209 | if (!param || pid < 0) | |
4210 | return -EINVAL; | |
4211 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4212 | return -EFAULT; | |
5fe1d75f ON |
4213 | |
4214 | rcu_read_lock(); | |
4215 | retval = -ESRCH; | |
1da177e4 | 4216 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4217 | if (p != NULL) |
4218 | retval = sched_setscheduler(p, policy, &lparam); | |
4219 | rcu_read_unlock(); | |
36c8b586 | 4220 | |
1da177e4 LT |
4221 | return retval; |
4222 | } | |
4223 | ||
4224 | /** | |
4225 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4226 | * @pid: the pid in question. | |
4227 | * @policy: new policy. | |
4228 | * @param: structure containing the new RT priority. | |
4229 | */ | |
4230 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4231 | struct sched_param __user *param) | |
4232 | { | |
c21761f1 JB |
4233 | /* negative values for policy are not valid */ |
4234 | if (policy < 0) | |
4235 | return -EINVAL; | |
4236 | ||
1da177e4 LT |
4237 | return do_sched_setscheduler(pid, policy, param); |
4238 | } | |
4239 | ||
4240 | /** | |
4241 | * sys_sched_setparam - set/change the RT priority of a thread | |
4242 | * @pid: the pid in question. | |
4243 | * @param: structure containing the new RT priority. | |
4244 | */ | |
4245 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4246 | { | |
4247 | return do_sched_setscheduler(pid, -1, param); | |
4248 | } | |
4249 | ||
4250 | /** | |
4251 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4252 | * @pid: the pid in question. | |
4253 | */ | |
4254 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4255 | { | |
36c8b586 | 4256 | struct task_struct *p; |
1da177e4 | 4257 | int retval = -EINVAL; |
1da177e4 LT |
4258 | |
4259 | if (pid < 0) | |
4260 | goto out_nounlock; | |
4261 | ||
4262 | retval = -ESRCH; | |
4263 | read_lock(&tasklist_lock); | |
4264 | p = find_process_by_pid(pid); | |
4265 | if (p) { | |
4266 | retval = security_task_getscheduler(p); | |
4267 | if (!retval) | |
4268 | retval = p->policy; | |
4269 | } | |
4270 | read_unlock(&tasklist_lock); | |
4271 | ||
4272 | out_nounlock: | |
4273 | return retval; | |
4274 | } | |
4275 | ||
4276 | /** | |
4277 | * sys_sched_getscheduler - get the RT priority of a thread | |
4278 | * @pid: the pid in question. | |
4279 | * @param: structure containing the RT priority. | |
4280 | */ | |
4281 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4282 | { | |
4283 | struct sched_param lp; | |
36c8b586 | 4284 | struct task_struct *p; |
1da177e4 | 4285 | int retval = -EINVAL; |
1da177e4 LT |
4286 | |
4287 | if (!param || pid < 0) | |
4288 | goto out_nounlock; | |
4289 | ||
4290 | read_lock(&tasklist_lock); | |
4291 | p = find_process_by_pid(pid); | |
4292 | retval = -ESRCH; | |
4293 | if (!p) | |
4294 | goto out_unlock; | |
4295 | ||
4296 | retval = security_task_getscheduler(p); | |
4297 | if (retval) | |
4298 | goto out_unlock; | |
4299 | ||
4300 | lp.sched_priority = p->rt_priority; | |
4301 | read_unlock(&tasklist_lock); | |
4302 | ||
4303 | /* | |
4304 | * This one might sleep, we cannot do it with a spinlock held ... | |
4305 | */ | |
4306 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4307 | ||
4308 | out_nounlock: | |
4309 | return retval; | |
4310 | ||
4311 | out_unlock: | |
4312 | read_unlock(&tasklist_lock); | |
4313 | return retval; | |
4314 | } | |
4315 | ||
4316 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4317 | { | |
1da177e4 | 4318 | cpumask_t cpus_allowed; |
36c8b586 IM |
4319 | struct task_struct *p; |
4320 | int retval; | |
1da177e4 | 4321 | |
5be9361c | 4322 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4323 | read_lock(&tasklist_lock); |
4324 | ||
4325 | p = find_process_by_pid(pid); | |
4326 | if (!p) { | |
4327 | read_unlock(&tasklist_lock); | |
5be9361c | 4328 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4329 | return -ESRCH; |
4330 | } | |
4331 | ||
4332 | /* | |
4333 | * It is not safe to call set_cpus_allowed with the | |
4334 | * tasklist_lock held. We will bump the task_struct's | |
4335 | * usage count and then drop tasklist_lock. | |
4336 | */ | |
4337 | get_task_struct(p); | |
4338 | read_unlock(&tasklist_lock); | |
4339 | ||
4340 | retval = -EPERM; | |
4341 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4342 | !capable(CAP_SYS_NICE)) | |
4343 | goto out_unlock; | |
4344 | ||
e7834f8f DQ |
4345 | retval = security_task_setscheduler(p, 0, NULL); |
4346 | if (retval) | |
4347 | goto out_unlock; | |
4348 | ||
1da177e4 LT |
4349 | cpus_allowed = cpuset_cpus_allowed(p); |
4350 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4351 | retval = set_cpus_allowed(p, new_mask); | |
4352 | ||
4353 | out_unlock: | |
4354 | put_task_struct(p); | |
5be9361c | 4355 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4356 | return retval; |
4357 | } | |
4358 | ||
4359 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4360 | cpumask_t *new_mask) | |
4361 | { | |
4362 | if (len < sizeof(cpumask_t)) { | |
4363 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4364 | } else if (len > sizeof(cpumask_t)) { | |
4365 | len = sizeof(cpumask_t); | |
4366 | } | |
4367 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4368 | } | |
4369 | ||
4370 | /** | |
4371 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4372 | * @pid: pid of the process | |
4373 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4374 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4375 | */ | |
4376 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4377 | unsigned long __user *user_mask_ptr) | |
4378 | { | |
4379 | cpumask_t new_mask; | |
4380 | int retval; | |
4381 | ||
4382 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4383 | if (retval) | |
4384 | return retval; | |
4385 | ||
4386 | return sched_setaffinity(pid, new_mask); | |
4387 | } | |
4388 | ||
4389 | /* | |
4390 | * Represents all cpu's present in the system | |
4391 | * In systems capable of hotplug, this map could dynamically grow | |
4392 | * as new cpu's are detected in the system via any platform specific | |
4393 | * method, such as ACPI for e.g. | |
4394 | */ | |
4395 | ||
4cef0c61 | 4396 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4397 | EXPORT_SYMBOL(cpu_present_map); |
4398 | ||
4399 | #ifndef CONFIG_SMP | |
4cef0c61 | 4400 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4401 | EXPORT_SYMBOL(cpu_online_map); |
4402 | ||
4cef0c61 | 4403 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4404 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4405 | #endif |
4406 | ||
4407 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4408 | { | |
36c8b586 | 4409 | struct task_struct *p; |
1da177e4 | 4410 | int retval; |
1da177e4 | 4411 | |
5be9361c | 4412 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4413 | read_lock(&tasklist_lock); |
4414 | ||
4415 | retval = -ESRCH; | |
4416 | p = find_process_by_pid(pid); | |
4417 | if (!p) | |
4418 | goto out_unlock; | |
4419 | ||
e7834f8f DQ |
4420 | retval = security_task_getscheduler(p); |
4421 | if (retval) | |
4422 | goto out_unlock; | |
4423 | ||
2f7016d9 | 4424 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4425 | |
4426 | out_unlock: | |
4427 | read_unlock(&tasklist_lock); | |
5be9361c | 4428 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4429 | if (retval) |
4430 | return retval; | |
4431 | ||
4432 | return 0; | |
4433 | } | |
4434 | ||
4435 | /** | |
4436 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4437 | * @pid: pid of the process | |
4438 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4439 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4440 | */ | |
4441 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4442 | unsigned long __user *user_mask_ptr) | |
4443 | { | |
4444 | int ret; | |
4445 | cpumask_t mask; | |
4446 | ||
4447 | if (len < sizeof(cpumask_t)) | |
4448 | return -EINVAL; | |
4449 | ||
4450 | ret = sched_getaffinity(pid, &mask); | |
4451 | if (ret < 0) | |
4452 | return ret; | |
4453 | ||
4454 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4455 | return -EFAULT; | |
4456 | ||
4457 | return sizeof(cpumask_t); | |
4458 | } | |
4459 | ||
4460 | /** | |
4461 | * sys_sched_yield - yield the current processor to other threads. | |
4462 | * | |
72fd4a35 | 4463 | * This function yields the current CPU by moving the calling thread |
1da177e4 LT |
4464 | * to the expired array. If there are no other threads running on this |
4465 | * CPU then this function will return. | |
4466 | */ | |
4467 | asmlinkage long sys_sched_yield(void) | |
4468 | { | |
70b97a7f IM |
4469 | struct rq *rq = this_rq_lock(); |
4470 | struct prio_array *array = current->array, *target = rq->expired; | |
1da177e4 LT |
4471 | |
4472 | schedstat_inc(rq, yld_cnt); | |
4473 | /* | |
4474 | * We implement yielding by moving the task into the expired | |
4475 | * queue. | |
4476 | * | |
4477 | * (special rule: RT tasks will just roundrobin in the active | |
4478 | * array.) | |
4479 | */ | |
4480 | if (rt_task(current)) | |
4481 | target = rq->active; | |
4482 | ||
5927ad78 | 4483 | if (array->nr_active == 1) { |
1da177e4 LT |
4484 | schedstat_inc(rq, yld_act_empty); |
4485 | if (!rq->expired->nr_active) | |
4486 | schedstat_inc(rq, yld_both_empty); | |
4487 | } else if (!rq->expired->nr_active) | |
4488 | schedstat_inc(rq, yld_exp_empty); | |
4489 | ||
4490 | if (array != target) { | |
4491 | dequeue_task(current, array); | |
4492 | enqueue_task(current, target); | |
4493 | } else | |
4494 | /* | |
4495 | * requeue_task is cheaper so perform that if possible. | |
4496 | */ | |
4497 | requeue_task(current, array); | |
4498 | ||
4499 | /* | |
4500 | * Since we are going to call schedule() anyway, there's | |
4501 | * no need to preempt or enable interrupts: | |
4502 | */ | |
4503 | __release(rq->lock); | |
8a25d5de | 4504 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4505 | _raw_spin_unlock(&rq->lock); |
4506 | preempt_enable_no_resched(); | |
4507 | ||
4508 | schedule(); | |
4509 | ||
4510 | return 0; | |
4511 | } | |
4512 | ||
e7b38404 | 4513 | static void __cond_resched(void) |
1da177e4 | 4514 | { |
8e0a43d8 IM |
4515 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4516 | __might_sleep(__FILE__, __LINE__); | |
4517 | #endif | |
5bbcfd90 IM |
4518 | /* |
4519 | * The BKS might be reacquired before we have dropped | |
4520 | * PREEMPT_ACTIVE, which could trigger a second | |
4521 | * cond_resched() call. | |
4522 | */ | |
1da177e4 LT |
4523 | do { |
4524 | add_preempt_count(PREEMPT_ACTIVE); | |
4525 | schedule(); | |
4526 | sub_preempt_count(PREEMPT_ACTIVE); | |
4527 | } while (need_resched()); | |
4528 | } | |
4529 | ||
4530 | int __sched cond_resched(void) | |
4531 | { | |
9414232f IM |
4532 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4533 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4534 | __cond_resched(); |
4535 | return 1; | |
4536 | } | |
4537 | return 0; | |
4538 | } | |
1da177e4 LT |
4539 | EXPORT_SYMBOL(cond_resched); |
4540 | ||
4541 | /* | |
4542 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4543 | * call schedule, and on return reacquire the lock. | |
4544 | * | |
4545 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4546 | * operations here to prevent schedule() from being called twice (once via | |
4547 | * spin_unlock(), once by hand). | |
4548 | */ | |
95cdf3b7 | 4549 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4550 | { |
6df3cecb JK |
4551 | int ret = 0; |
4552 | ||
1da177e4 LT |
4553 | if (need_lockbreak(lock)) { |
4554 | spin_unlock(lock); | |
4555 | cpu_relax(); | |
6df3cecb | 4556 | ret = 1; |
1da177e4 LT |
4557 | spin_lock(lock); |
4558 | } | |
9414232f | 4559 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4560 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4561 | _raw_spin_unlock(lock); |
4562 | preempt_enable_no_resched(); | |
4563 | __cond_resched(); | |
6df3cecb | 4564 | ret = 1; |
1da177e4 | 4565 | spin_lock(lock); |
1da177e4 | 4566 | } |
6df3cecb | 4567 | return ret; |
1da177e4 | 4568 | } |
1da177e4 LT |
4569 | EXPORT_SYMBOL(cond_resched_lock); |
4570 | ||
4571 | int __sched cond_resched_softirq(void) | |
4572 | { | |
4573 | BUG_ON(!in_softirq()); | |
4574 | ||
9414232f | 4575 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4576 | local_bh_enable(); |
1da177e4 LT |
4577 | __cond_resched(); |
4578 | local_bh_disable(); | |
4579 | return 1; | |
4580 | } | |
4581 | return 0; | |
4582 | } | |
1da177e4 LT |
4583 | EXPORT_SYMBOL(cond_resched_softirq); |
4584 | ||
1da177e4 LT |
4585 | /** |
4586 | * yield - yield the current processor to other threads. | |
4587 | * | |
72fd4a35 | 4588 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4589 | * thread runnable and calls sys_sched_yield(). |
4590 | */ | |
4591 | void __sched yield(void) | |
4592 | { | |
4593 | set_current_state(TASK_RUNNING); | |
4594 | sys_sched_yield(); | |
4595 | } | |
1da177e4 LT |
4596 | EXPORT_SYMBOL(yield); |
4597 | ||
4598 | /* | |
4599 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4600 | * that process accounting knows that this is a task in IO wait state. | |
4601 | * | |
4602 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4603 | * has set its backing_dev_info: the queue against which it should throttle) | |
4604 | */ | |
4605 | void __sched io_schedule(void) | |
4606 | { | |
70b97a7f | 4607 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4608 | |
0ff92245 | 4609 | delayacct_blkio_start(); |
1da177e4 LT |
4610 | atomic_inc(&rq->nr_iowait); |
4611 | schedule(); | |
4612 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4613 | delayacct_blkio_end(); |
1da177e4 | 4614 | } |
1da177e4 LT |
4615 | EXPORT_SYMBOL(io_schedule); |
4616 | ||
4617 | long __sched io_schedule_timeout(long timeout) | |
4618 | { | |
70b97a7f | 4619 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4620 | long ret; |
4621 | ||
0ff92245 | 4622 | delayacct_blkio_start(); |
1da177e4 LT |
4623 | atomic_inc(&rq->nr_iowait); |
4624 | ret = schedule_timeout(timeout); | |
4625 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4626 | delayacct_blkio_end(); |
1da177e4 LT |
4627 | return ret; |
4628 | } | |
4629 | ||
4630 | /** | |
4631 | * sys_sched_get_priority_max - return maximum RT priority. | |
4632 | * @policy: scheduling class. | |
4633 | * | |
4634 | * this syscall returns the maximum rt_priority that can be used | |
4635 | * by a given scheduling class. | |
4636 | */ | |
4637 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4638 | { | |
4639 | int ret = -EINVAL; | |
4640 | ||
4641 | switch (policy) { | |
4642 | case SCHED_FIFO: | |
4643 | case SCHED_RR: | |
4644 | ret = MAX_USER_RT_PRIO-1; | |
4645 | break; | |
4646 | case SCHED_NORMAL: | |
b0a9499c | 4647 | case SCHED_BATCH: |
1da177e4 LT |
4648 | ret = 0; |
4649 | break; | |
4650 | } | |
4651 | return ret; | |
4652 | } | |
4653 | ||
4654 | /** | |
4655 | * sys_sched_get_priority_min - return minimum RT priority. | |
4656 | * @policy: scheduling class. | |
4657 | * | |
4658 | * this syscall returns the minimum rt_priority that can be used | |
4659 | * by a given scheduling class. | |
4660 | */ | |
4661 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4662 | { | |
4663 | int ret = -EINVAL; | |
4664 | ||
4665 | switch (policy) { | |
4666 | case SCHED_FIFO: | |
4667 | case SCHED_RR: | |
4668 | ret = 1; | |
4669 | break; | |
4670 | case SCHED_NORMAL: | |
b0a9499c | 4671 | case SCHED_BATCH: |
1da177e4 LT |
4672 | ret = 0; |
4673 | } | |
4674 | return ret; | |
4675 | } | |
4676 | ||
4677 | /** | |
4678 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4679 | * @pid: pid of the process. | |
4680 | * @interval: userspace pointer to the timeslice value. | |
4681 | * | |
4682 | * this syscall writes the default timeslice value of a given process | |
4683 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4684 | */ | |
4685 | asmlinkage | |
4686 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4687 | { | |
36c8b586 | 4688 | struct task_struct *p; |
1da177e4 LT |
4689 | int retval = -EINVAL; |
4690 | struct timespec t; | |
1da177e4 LT |
4691 | |
4692 | if (pid < 0) | |
4693 | goto out_nounlock; | |
4694 | ||
4695 | retval = -ESRCH; | |
4696 | read_lock(&tasklist_lock); | |
4697 | p = find_process_by_pid(pid); | |
4698 | if (!p) | |
4699 | goto out_unlock; | |
4700 | ||
4701 | retval = security_task_getscheduler(p); | |
4702 | if (retval) | |
4703 | goto out_unlock; | |
4704 | ||
b78709cf | 4705 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
1da177e4 LT |
4706 | 0 : task_timeslice(p), &t); |
4707 | read_unlock(&tasklist_lock); | |
4708 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4709 | out_nounlock: | |
4710 | return retval; | |
4711 | out_unlock: | |
4712 | read_unlock(&tasklist_lock); | |
4713 | return retval; | |
4714 | } | |
4715 | ||
2ed6e34f | 4716 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4717 | |
4718 | static void show_task(struct task_struct *p) | |
1da177e4 | 4719 | { |
1da177e4 | 4720 | unsigned long free = 0; |
36c8b586 | 4721 | unsigned state; |
1da177e4 | 4722 | |
1da177e4 | 4723 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4724 | printk("%-13.13s %c", p->comm, |
4725 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
1da177e4 LT |
4726 | #if (BITS_PER_LONG == 32) |
4727 | if (state == TASK_RUNNING) | |
4728 | printk(" running "); | |
4729 | else | |
4730 | printk(" %08lX ", thread_saved_pc(p)); | |
4731 | #else | |
4732 | if (state == TASK_RUNNING) | |
4733 | printk(" running task "); | |
4734 | else | |
4735 | printk(" %016lx ", thread_saved_pc(p)); | |
4736 | #endif | |
4737 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4738 | { | |
10ebffde | 4739 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4740 | while (!*n) |
4741 | n++; | |
10ebffde | 4742 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4743 | } |
4744 | #endif | |
35f6f753 | 4745 | printk("%5lu %5d %6d", free, p->pid, p->parent->pid); |
1da177e4 LT |
4746 | if (!p->mm) |
4747 | printk(" (L-TLB)\n"); | |
4748 | else | |
4749 | printk(" (NOTLB)\n"); | |
4750 | ||
4751 | if (state != TASK_RUNNING) | |
4752 | show_stack(p, NULL); | |
4753 | } | |
4754 | ||
e59e2ae2 | 4755 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4756 | { |
36c8b586 | 4757 | struct task_struct *g, *p; |
1da177e4 LT |
4758 | |
4759 | #if (BITS_PER_LONG == 32) | |
4760 | printk("\n" | |
301827ac CC |
4761 | " free sibling\n"); |
4762 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
4763 | #else |
4764 | printk("\n" | |
301827ac CC |
4765 | " free sibling\n"); |
4766 | printk(" task PC stack pid father child younger older\n"); | |
1da177e4 LT |
4767 | #endif |
4768 | read_lock(&tasklist_lock); | |
4769 | do_each_thread(g, p) { | |
4770 | /* | |
4771 | * reset the NMI-timeout, listing all files on a slow | |
4772 | * console might take alot of time: | |
4773 | */ | |
4774 | touch_nmi_watchdog(); | |
39bc89fd | 4775 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4776 | show_task(p); |
1da177e4 LT |
4777 | } while_each_thread(g, p); |
4778 | ||
04c9167f JF |
4779 | touch_all_softlockup_watchdogs(); |
4780 | ||
1da177e4 | 4781 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4782 | /* |
4783 | * Only show locks if all tasks are dumped: | |
4784 | */ | |
4785 | if (state_filter == -1) | |
4786 | debug_show_all_locks(); | |
1da177e4 LT |
4787 | } |
4788 | ||
1df21055 IM |
4789 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4790 | { | |
4791 | /* nothing yet */ | |
4792 | } | |
4793 | ||
f340c0d1 IM |
4794 | /** |
4795 | * init_idle - set up an idle thread for a given CPU | |
4796 | * @idle: task in question | |
4797 | * @cpu: cpu the idle task belongs to | |
4798 | * | |
4799 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4800 | * flag, to make booting more robust. | |
4801 | */ | |
5c1e1767 | 4802 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4803 | { |
70b97a7f | 4804 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4805 | unsigned long flags; |
4806 | ||
81c29a85 | 4807 | idle->timestamp = sched_clock(); |
1da177e4 LT |
4808 | idle->sleep_avg = 0; |
4809 | idle->array = NULL; | |
b29739f9 | 4810 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 LT |
4811 | idle->state = TASK_RUNNING; |
4812 | idle->cpus_allowed = cpumask_of_cpu(cpu); | |
4813 | set_task_cpu(idle, cpu); | |
4814 | ||
4815 | spin_lock_irqsave(&rq->lock, flags); | |
4816 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4817 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4818 | idle->oncpu = 1; | |
4819 | #endif | |
1da177e4 LT |
4820 | spin_unlock_irqrestore(&rq->lock, flags); |
4821 | ||
4822 | /* Set the preempt count _outside_ the spinlocks! */ | |
4823 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4824 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4825 | #else |
a1261f54 | 4826 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 LT |
4827 | #endif |
4828 | } | |
4829 | ||
4830 | /* | |
4831 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4832 | * indicates which cpus entered this state. This is used | |
4833 | * in the rcu update to wait only for active cpus. For system | |
4834 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4835 | * always be CPU_MASK_NONE. | |
4836 | */ | |
4837 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4838 | ||
4839 | #ifdef CONFIG_SMP | |
4840 | /* | |
4841 | * This is how migration works: | |
4842 | * | |
70b97a7f | 4843 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4844 | * runqueue and wake up that CPU's migration thread. |
4845 | * 2) we down() the locked semaphore => thread blocks. | |
4846 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4847 | * thread off the CPU) | |
4848 | * 4) it gets the migration request and checks whether the migrated | |
4849 | * task is still in the wrong runqueue. | |
4850 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4851 | * it and puts it into the right queue. | |
4852 | * 6) migration thread up()s the semaphore. | |
4853 | * 7) we wake up and the migration is done. | |
4854 | */ | |
4855 | ||
4856 | /* | |
4857 | * Change a given task's CPU affinity. Migrate the thread to a | |
4858 | * proper CPU and schedule it away if the CPU it's executing on | |
4859 | * is removed from the allowed bitmask. | |
4860 | * | |
4861 | * NOTE: the caller must have a valid reference to the task, the | |
4862 | * task must not exit() & deallocate itself prematurely. The | |
4863 | * call is not atomic; no spinlocks may be held. | |
4864 | */ | |
36c8b586 | 4865 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4866 | { |
70b97a7f | 4867 | struct migration_req req; |
1da177e4 | 4868 | unsigned long flags; |
70b97a7f | 4869 | struct rq *rq; |
48f24c4d | 4870 | int ret = 0; |
1da177e4 LT |
4871 | |
4872 | rq = task_rq_lock(p, &flags); | |
4873 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4874 | ret = -EINVAL; | |
4875 | goto out; | |
4876 | } | |
4877 | ||
4878 | p->cpus_allowed = new_mask; | |
4879 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4880 | if (cpu_isset(task_cpu(p), new_mask)) | |
4881 | goto out; | |
4882 | ||
4883 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4884 | /* Need help from migration thread: drop lock and wait. */ | |
4885 | task_rq_unlock(rq, &flags); | |
4886 | wake_up_process(rq->migration_thread); | |
4887 | wait_for_completion(&req.done); | |
4888 | tlb_migrate_finish(p->mm); | |
4889 | return 0; | |
4890 | } | |
4891 | out: | |
4892 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4893 | |
1da177e4 LT |
4894 | return ret; |
4895 | } | |
1da177e4 LT |
4896 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4897 | ||
4898 | /* | |
4899 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4900 | * this because either it can't run here any more (set_cpus_allowed() | |
4901 | * away from this CPU, or CPU going down), or because we're | |
4902 | * attempting to rebalance this task on exec (sched_exec). | |
4903 | * | |
4904 | * So we race with normal scheduler movements, but that's OK, as long | |
4905 | * as the task is no longer on this CPU. | |
efc30814 KK |
4906 | * |
4907 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4908 | */ |
efc30814 | 4909 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4910 | { |
70b97a7f | 4911 | struct rq *rq_dest, *rq_src; |
efc30814 | 4912 | int ret = 0; |
1da177e4 LT |
4913 | |
4914 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4915 | return ret; |
1da177e4 LT |
4916 | |
4917 | rq_src = cpu_rq(src_cpu); | |
4918 | rq_dest = cpu_rq(dest_cpu); | |
4919 | ||
4920 | double_rq_lock(rq_src, rq_dest); | |
4921 | /* Already moved. */ | |
4922 | if (task_cpu(p) != src_cpu) | |
4923 | goto out; | |
4924 | /* Affinity changed (again). */ | |
4925 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4926 | goto out; | |
4927 | ||
4928 | set_task_cpu(p, dest_cpu); | |
4929 | if (p->array) { | |
4930 | /* | |
4931 | * Sync timestamp with rq_dest's before activating. | |
4932 | * The same thing could be achieved by doing this step | |
4933 | * afterwards, and pretending it was a local activate. | |
4934 | * This way is cleaner and logically correct. | |
4935 | */ | |
b18ec803 MG |
4936 | p->timestamp = p->timestamp - rq_src->most_recent_timestamp |
4937 | + rq_dest->most_recent_timestamp; | |
1da177e4 | 4938 | deactivate_task(p, rq_src); |
0a565f79 | 4939 | __activate_task(p, rq_dest); |
1da177e4 LT |
4940 | if (TASK_PREEMPTS_CURR(p, rq_dest)) |
4941 | resched_task(rq_dest->curr); | |
4942 | } | |
efc30814 | 4943 | ret = 1; |
1da177e4 LT |
4944 | out: |
4945 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4946 | return ret; |
1da177e4 LT |
4947 | } |
4948 | ||
4949 | /* | |
4950 | * migration_thread - this is a highprio system thread that performs | |
4951 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4952 | * another runqueue. | |
4953 | */ | |
95cdf3b7 | 4954 | static int migration_thread(void *data) |
1da177e4 | 4955 | { |
1da177e4 | 4956 | int cpu = (long)data; |
70b97a7f | 4957 | struct rq *rq; |
1da177e4 LT |
4958 | |
4959 | rq = cpu_rq(cpu); | |
4960 | BUG_ON(rq->migration_thread != current); | |
4961 | ||
4962 | set_current_state(TASK_INTERRUPTIBLE); | |
4963 | while (!kthread_should_stop()) { | |
70b97a7f | 4964 | struct migration_req *req; |
1da177e4 | 4965 | struct list_head *head; |
1da177e4 | 4966 | |
3e1d1d28 | 4967 | try_to_freeze(); |
1da177e4 LT |
4968 | |
4969 | spin_lock_irq(&rq->lock); | |
4970 | ||
4971 | if (cpu_is_offline(cpu)) { | |
4972 | spin_unlock_irq(&rq->lock); | |
4973 | goto wait_to_die; | |
4974 | } | |
4975 | ||
4976 | if (rq->active_balance) { | |
4977 | active_load_balance(rq, cpu); | |
4978 | rq->active_balance = 0; | |
4979 | } | |
4980 | ||
4981 | head = &rq->migration_queue; | |
4982 | ||
4983 | if (list_empty(head)) { | |
4984 | spin_unlock_irq(&rq->lock); | |
4985 | schedule(); | |
4986 | set_current_state(TASK_INTERRUPTIBLE); | |
4987 | continue; | |
4988 | } | |
70b97a7f | 4989 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
4990 | list_del_init(head->next); |
4991 | ||
674311d5 NP |
4992 | spin_unlock(&rq->lock); |
4993 | __migrate_task(req->task, cpu, req->dest_cpu); | |
4994 | local_irq_enable(); | |
1da177e4 LT |
4995 | |
4996 | complete(&req->done); | |
4997 | } | |
4998 | __set_current_state(TASK_RUNNING); | |
4999 | return 0; | |
5000 | ||
5001 | wait_to_die: | |
5002 | /* Wait for kthread_stop */ | |
5003 | set_current_state(TASK_INTERRUPTIBLE); | |
5004 | while (!kthread_should_stop()) { | |
5005 | schedule(); | |
5006 | set_current_state(TASK_INTERRUPTIBLE); | |
5007 | } | |
5008 | __set_current_state(TASK_RUNNING); | |
5009 | return 0; | |
5010 | } | |
5011 | ||
5012 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5013 | /* |
5014 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5015 | * NOTE: interrupts should be disabled by the caller | |
5016 | */ | |
48f24c4d | 5017 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5018 | { |
efc30814 | 5019 | unsigned long flags; |
1da177e4 | 5020 | cpumask_t mask; |
70b97a7f IM |
5021 | struct rq *rq; |
5022 | int dest_cpu; | |
1da177e4 | 5023 | |
efc30814 | 5024 | restart: |
1da177e4 LT |
5025 | /* On same node? */ |
5026 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5027 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5028 | dest_cpu = any_online_cpu(mask); |
5029 | ||
5030 | /* On any allowed CPU? */ | |
5031 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5032 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5033 | |
5034 | /* No more Mr. Nice Guy. */ | |
5035 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5036 | rq = task_rq_lock(p, &flags); |
5037 | cpus_setall(p->cpus_allowed); | |
5038 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5039 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5040 | |
5041 | /* | |
5042 | * Don't tell them about moving exiting tasks or | |
5043 | * kernel threads (both mm NULL), since they never | |
5044 | * leave kernel. | |
5045 | */ | |
48f24c4d | 5046 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5047 | printk(KERN_INFO "process %d (%s) no " |
5048 | "longer affine to cpu%d\n", | |
48f24c4d | 5049 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5050 | } |
48f24c4d | 5051 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5052 | goto restart; |
1da177e4 LT |
5053 | } |
5054 | ||
5055 | /* | |
5056 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5057 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5058 | * for performance reasons the counter is not stricly tracking tasks to | |
5059 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5060 | * to keep the global sum constant after CPU-down: | |
5061 | */ | |
70b97a7f | 5062 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5063 | { |
70b97a7f | 5064 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5065 | unsigned long flags; |
5066 | ||
5067 | local_irq_save(flags); | |
5068 | double_rq_lock(rq_src, rq_dest); | |
5069 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5070 | rq_src->nr_uninterruptible = 0; | |
5071 | double_rq_unlock(rq_src, rq_dest); | |
5072 | local_irq_restore(flags); | |
5073 | } | |
5074 | ||
5075 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5076 | static void migrate_live_tasks(int src_cpu) | |
5077 | { | |
48f24c4d | 5078 | struct task_struct *p, *t; |
1da177e4 LT |
5079 | |
5080 | write_lock_irq(&tasklist_lock); | |
5081 | ||
48f24c4d IM |
5082 | do_each_thread(t, p) { |
5083 | if (p == current) | |
1da177e4 LT |
5084 | continue; |
5085 | ||
48f24c4d IM |
5086 | if (task_cpu(p) == src_cpu) |
5087 | move_task_off_dead_cpu(src_cpu, p); | |
5088 | } while_each_thread(t, p); | |
1da177e4 LT |
5089 | |
5090 | write_unlock_irq(&tasklist_lock); | |
5091 | } | |
5092 | ||
5093 | /* Schedules idle task to be the next runnable task on current CPU. | |
5094 | * It does so by boosting its priority to highest possible and adding it to | |
48f24c4d | 5095 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5096 | */ |
5097 | void sched_idle_next(void) | |
5098 | { | |
48f24c4d | 5099 | int this_cpu = smp_processor_id(); |
70b97a7f | 5100 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5101 | struct task_struct *p = rq->idle; |
5102 | unsigned long flags; | |
5103 | ||
5104 | /* cpu has to be offline */ | |
48f24c4d | 5105 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5106 | |
48f24c4d IM |
5107 | /* |
5108 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5109 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5110 | */ |
5111 | spin_lock_irqsave(&rq->lock, flags); | |
5112 | ||
5113 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
48f24c4d IM |
5114 | |
5115 | /* Add idle task to the _front_ of its priority queue: */ | |
1da177e4 LT |
5116 | __activate_idle_task(p, rq); |
5117 | ||
5118 | spin_unlock_irqrestore(&rq->lock, flags); | |
5119 | } | |
5120 | ||
48f24c4d IM |
5121 | /* |
5122 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5123 | * offline. |
5124 | */ | |
5125 | void idle_task_exit(void) | |
5126 | { | |
5127 | struct mm_struct *mm = current->active_mm; | |
5128 | ||
5129 | BUG_ON(cpu_online(smp_processor_id())); | |
5130 | ||
5131 | if (mm != &init_mm) | |
5132 | switch_mm(mm, &init_mm, current); | |
5133 | mmdrop(mm); | |
5134 | } | |
5135 | ||
054b9108 | 5136 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5137 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5138 | { |
70b97a7f | 5139 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5140 | |
5141 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5142 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5143 | |
5144 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5145 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5146 | |
48f24c4d | 5147 | get_task_struct(p); |
1da177e4 LT |
5148 | |
5149 | /* | |
5150 | * Drop lock around migration; if someone else moves it, | |
5151 | * that's OK. No task can be added to this CPU, so iteration is | |
5152 | * fine. | |
054b9108 | 5153 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5154 | */ |
054b9108 | 5155 | spin_unlock(&rq->lock); |
48f24c4d | 5156 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5157 | spin_lock(&rq->lock); |
1da177e4 | 5158 | |
48f24c4d | 5159 | put_task_struct(p); |
1da177e4 LT |
5160 | } |
5161 | ||
5162 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5163 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5164 | { | |
70b97a7f | 5165 | struct rq *rq = cpu_rq(dead_cpu); |
48f24c4d | 5166 | unsigned int arr, i; |
1da177e4 LT |
5167 | |
5168 | for (arr = 0; arr < 2; arr++) { | |
5169 | for (i = 0; i < MAX_PRIO; i++) { | |
5170 | struct list_head *list = &rq->arrays[arr].queue[i]; | |
48f24c4d | 5171 | |
1da177e4 | 5172 | while (!list_empty(list)) |
36c8b586 IM |
5173 | migrate_dead(dead_cpu, list_entry(list->next, |
5174 | struct task_struct, run_list)); | |
1da177e4 LT |
5175 | } |
5176 | } | |
5177 | } | |
5178 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5179 | ||
5180 | /* | |
5181 | * migration_call - callback that gets triggered when a CPU is added. | |
5182 | * Here we can start up the necessary migration thread for the new CPU. | |
5183 | */ | |
48f24c4d IM |
5184 | static int __cpuinit |
5185 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5186 | { |
1da177e4 | 5187 | struct task_struct *p; |
48f24c4d | 5188 | int cpu = (long)hcpu; |
1da177e4 | 5189 | unsigned long flags; |
70b97a7f | 5190 | struct rq *rq; |
1da177e4 LT |
5191 | |
5192 | switch (action) { | |
5be9361c GS |
5193 | case CPU_LOCK_ACQUIRE: |
5194 | mutex_lock(&sched_hotcpu_mutex); | |
5195 | break; | |
5196 | ||
1da177e4 | 5197 | case CPU_UP_PREPARE: |
8bb78442 | 5198 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 LT |
5199 | p = kthread_create(migration_thread, hcpu, "migration/%d",cpu); |
5200 | if (IS_ERR(p)) | |
5201 | return NOTIFY_BAD; | |
5202 | p->flags |= PF_NOFREEZE; | |
5203 | kthread_bind(p, cpu); | |
5204 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5205 | rq = task_rq_lock(p, &flags); | |
5206 | __setscheduler(p, SCHED_FIFO, MAX_RT_PRIO-1); | |
5207 | task_rq_unlock(rq, &flags); | |
5208 | cpu_rq(cpu)->migration_thread = p; | |
5209 | break; | |
48f24c4d | 5210 | |
1da177e4 | 5211 | case CPU_ONLINE: |
8bb78442 | 5212 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5213 | /* Strictly unneccessary, as first user will wake it. */ |
5214 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5215 | break; | |
48f24c4d | 5216 | |
1da177e4 LT |
5217 | #ifdef CONFIG_HOTPLUG_CPU |
5218 | case CPU_UP_CANCELED: | |
8bb78442 | 5219 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5220 | if (!cpu_rq(cpu)->migration_thread) |
5221 | break; | |
1da177e4 | 5222 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5223 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5224 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5225 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5226 | cpu_rq(cpu)->migration_thread = NULL; | |
5227 | break; | |
48f24c4d | 5228 | |
1da177e4 | 5229 | case CPU_DEAD: |
8bb78442 | 5230 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5231 | migrate_live_tasks(cpu); |
5232 | rq = cpu_rq(cpu); | |
5233 | kthread_stop(rq->migration_thread); | |
5234 | rq->migration_thread = NULL; | |
5235 | /* Idle task back to normal (off runqueue, low prio) */ | |
5236 | rq = task_rq_lock(rq->idle, &flags); | |
5237 | deactivate_task(rq->idle, rq); | |
5238 | rq->idle->static_prio = MAX_PRIO; | |
5239 | __setscheduler(rq->idle, SCHED_NORMAL, 0); | |
5240 | migrate_dead_tasks(cpu); | |
5241 | task_rq_unlock(rq, &flags); | |
5242 | migrate_nr_uninterruptible(rq); | |
5243 | BUG_ON(rq->nr_running != 0); | |
5244 | ||
5245 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5246 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5247 | * the requestors. */ |
5248 | spin_lock_irq(&rq->lock); | |
5249 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5250 | struct migration_req *req; |
5251 | ||
1da177e4 | 5252 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5253 | struct migration_req, list); |
1da177e4 LT |
5254 | list_del_init(&req->list); |
5255 | complete(&req->done); | |
5256 | } | |
5257 | spin_unlock_irq(&rq->lock); | |
5258 | break; | |
5259 | #endif | |
5be9361c GS |
5260 | case CPU_LOCK_RELEASE: |
5261 | mutex_unlock(&sched_hotcpu_mutex); | |
5262 | break; | |
1da177e4 LT |
5263 | } |
5264 | return NOTIFY_OK; | |
5265 | } | |
5266 | ||
5267 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5268 | * happens before everything else. | |
5269 | */ | |
26c2143b | 5270 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5271 | .notifier_call = migration_call, |
5272 | .priority = 10 | |
5273 | }; | |
5274 | ||
5275 | int __init migration_init(void) | |
5276 | { | |
5277 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5278 | int err; |
48f24c4d IM |
5279 | |
5280 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5281 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5282 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5283 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5284 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5285 | |
1da177e4 LT |
5286 | return 0; |
5287 | } | |
5288 | #endif | |
5289 | ||
5290 | #ifdef CONFIG_SMP | |
476f3534 CL |
5291 | |
5292 | /* Number of possible processor ids */ | |
5293 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5294 | EXPORT_SYMBOL(nr_cpu_ids); | |
5295 | ||
1a20ff27 | 5296 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5297 | #ifdef SCHED_DOMAIN_DEBUG |
5298 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5299 | { | |
5300 | int level = 0; | |
5301 | ||
41c7ce9a NP |
5302 | if (!sd) { |
5303 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5304 | return; | |
5305 | } | |
5306 | ||
1da177e4 LT |
5307 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5308 | ||
5309 | do { | |
5310 | int i; | |
5311 | char str[NR_CPUS]; | |
5312 | struct sched_group *group = sd->groups; | |
5313 | cpumask_t groupmask; | |
5314 | ||
5315 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5316 | cpus_clear(groupmask); | |
5317 | ||
5318 | printk(KERN_DEBUG); | |
5319 | for (i = 0; i < level + 1; i++) | |
5320 | printk(" "); | |
5321 | printk("domain %d: ", level); | |
5322 | ||
5323 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5324 | printk("does not load-balance\n"); | |
5325 | if (sd->parent) | |
33859f7f MOS |
5326 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5327 | " has parent"); | |
1da177e4 LT |
5328 | break; |
5329 | } | |
5330 | ||
5331 | printk("span %s\n", str); | |
5332 | ||
5333 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5334 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5335 | "CPU%d\n", cpu); | |
1da177e4 | 5336 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5337 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5338 | " CPU%d\n", cpu); | |
1da177e4 LT |
5339 | |
5340 | printk(KERN_DEBUG); | |
5341 | for (i = 0; i < level + 2; i++) | |
5342 | printk(" "); | |
5343 | printk("groups:"); | |
5344 | do { | |
5345 | if (!group) { | |
5346 | printk("\n"); | |
5347 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5348 | break; | |
5349 | } | |
5350 | ||
5517d86b | 5351 | if (!group->__cpu_power) { |
1da177e4 | 5352 | printk("\n"); |
33859f7f MOS |
5353 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5354 | "set\n"); | |
1da177e4 LT |
5355 | } |
5356 | ||
5357 | if (!cpus_weight(group->cpumask)) { | |
5358 | printk("\n"); | |
5359 | printk(KERN_ERR "ERROR: empty group\n"); | |
5360 | } | |
5361 | ||
5362 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5363 | printk("\n"); | |
5364 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5365 | } | |
5366 | ||
5367 | cpus_or(groupmask, groupmask, group->cpumask); | |
5368 | ||
5369 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5370 | printk(" %s", str); | |
5371 | ||
5372 | group = group->next; | |
5373 | } while (group != sd->groups); | |
5374 | printk("\n"); | |
5375 | ||
5376 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5377 | printk(KERN_ERR "ERROR: groups don't span " |
5378 | "domain->span\n"); | |
1da177e4 LT |
5379 | |
5380 | level++; | |
5381 | sd = sd->parent; | |
33859f7f MOS |
5382 | if (!sd) |
5383 | continue; | |
1da177e4 | 5384 | |
33859f7f MOS |
5385 | if (!cpus_subset(groupmask, sd->span)) |
5386 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5387 | "of domain->span\n"); | |
1da177e4 LT |
5388 | |
5389 | } while (sd); | |
5390 | } | |
5391 | #else | |
48f24c4d | 5392 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5393 | #endif |
5394 | ||
1a20ff27 | 5395 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5396 | { |
5397 | if (cpus_weight(sd->span) == 1) | |
5398 | return 1; | |
5399 | ||
5400 | /* Following flags need at least 2 groups */ | |
5401 | if (sd->flags & (SD_LOAD_BALANCE | | |
5402 | SD_BALANCE_NEWIDLE | | |
5403 | SD_BALANCE_FORK | | |
89c4710e SS |
5404 | SD_BALANCE_EXEC | |
5405 | SD_SHARE_CPUPOWER | | |
5406 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5407 | if (sd->groups != sd->groups->next) |
5408 | return 0; | |
5409 | } | |
5410 | ||
5411 | /* Following flags don't use groups */ | |
5412 | if (sd->flags & (SD_WAKE_IDLE | | |
5413 | SD_WAKE_AFFINE | | |
5414 | SD_WAKE_BALANCE)) | |
5415 | return 0; | |
5416 | ||
5417 | return 1; | |
5418 | } | |
5419 | ||
48f24c4d IM |
5420 | static int |
5421 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5422 | { |
5423 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5424 | ||
5425 | if (sd_degenerate(parent)) | |
5426 | return 1; | |
5427 | ||
5428 | if (!cpus_equal(sd->span, parent->span)) | |
5429 | return 0; | |
5430 | ||
5431 | /* Does parent contain flags not in child? */ | |
5432 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5433 | if (cflags & SD_WAKE_AFFINE) | |
5434 | pflags &= ~SD_WAKE_BALANCE; | |
5435 | /* Flags needing groups don't count if only 1 group in parent */ | |
5436 | if (parent->groups == parent->groups->next) { | |
5437 | pflags &= ~(SD_LOAD_BALANCE | | |
5438 | SD_BALANCE_NEWIDLE | | |
5439 | SD_BALANCE_FORK | | |
89c4710e SS |
5440 | SD_BALANCE_EXEC | |
5441 | SD_SHARE_CPUPOWER | | |
5442 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5443 | } |
5444 | if (~cflags & pflags) | |
5445 | return 0; | |
5446 | ||
5447 | return 1; | |
5448 | } | |
5449 | ||
1da177e4 LT |
5450 | /* |
5451 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5452 | * hold the hotplug lock. | |
5453 | */ | |
9c1cfda2 | 5454 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5455 | { |
70b97a7f | 5456 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5457 | struct sched_domain *tmp; |
5458 | ||
5459 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5460 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5461 | struct sched_domain *parent = tmp->parent; | |
5462 | if (!parent) | |
5463 | break; | |
1a848870 | 5464 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5465 | tmp->parent = parent->parent; |
1a848870 SS |
5466 | if (parent->parent) |
5467 | parent->parent->child = tmp; | |
5468 | } | |
245af2c7 SS |
5469 | } |
5470 | ||
1a848870 | 5471 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5472 | sd = sd->parent; |
1a848870 SS |
5473 | if (sd) |
5474 | sd->child = NULL; | |
5475 | } | |
1da177e4 LT |
5476 | |
5477 | sched_domain_debug(sd, cpu); | |
5478 | ||
674311d5 | 5479 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5480 | } |
5481 | ||
5482 | /* cpus with isolated domains */ | |
67af63a6 | 5483 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5484 | |
5485 | /* Setup the mask of cpus configured for isolated domains */ | |
5486 | static int __init isolated_cpu_setup(char *str) | |
5487 | { | |
5488 | int ints[NR_CPUS], i; | |
5489 | ||
5490 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5491 | cpus_clear(cpu_isolated_map); | |
5492 | for (i = 1; i <= ints[0]; i++) | |
5493 | if (ints[i] < NR_CPUS) | |
5494 | cpu_set(ints[i], cpu_isolated_map); | |
5495 | return 1; | |
5496 | } | |
5497 | ||
5498 | __setup ("isolcpus=", isolated_cpu_setup); | |
5499 | ||
5500 | /* | |
6711cab4 SS |
5501 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5502 | * to a function which identifies what group(along with sched group) a CPU | |
5503 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5504 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5505 | * |
5506 | * init_sched_build_groups will build a circular linked list of the groups | |
5507 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5508 | * and ->cpu_power to 0. | |
5509 | */ | |
a616058b | 5510 | static void |
6711cab4 SS |
5511 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5512 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5513 | struct sched_group **sg)) | |
1da177e4 LT |
5514 | { |
5515 | struct sched_group *first = NULL, *last = NULL; | |
5516 | cpumask_t covered = CPU_MASK_NONE; | |
5517 | int i; | |
5518 | ||
5519 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5520 | struct sched_group *sg; |
5521 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5522 | int j; |
5523 | ||
5524 | if (cpu_isset(i, covered)) | |
5525 | continue; | |
5526 | ||
5527 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5528 | sg->__cpu_power = 0; |
1da177e4 LT |
5529 | |
5530 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5531 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5532 | continue; |
5533 | ||
5534 | cpu_set(j, covered); | |
5535 | cpu_set(j, sg->cpumask); | |
5536 | } | |
5537 | if (!first) | |
5538 | first = sg; | |
5539 | if (last) | |
5540 | last->next = sg; | |
5541 | last = sg; | |
5542 | } | |
5543 | last->next = first; | |
5544 | } | |
5545 | ||
9c1cfda2 | 5546 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5547 | |
9c1cfda2 | 5548 | #ifdef CONFIG_NUMA |
198e2f18 | 5549 | |
9c1cfda2 JH |
5550 | /** |
5551 | * find_next_best_node - find the next node to include in a sched_domain | |
5552 | * @node: node whose sched_domain we're building | |
5553 | * @used_nodes: nodes already in the sched_domain | |
5554 | * | |
5555 | * Find the next node to include in a given scheduling domain. Simply | |
5556 | * finds the closest node not already in the @used_nodes map. | |
5557 | * | |
5558 | * Should use nodemask_t. | |
5559 | */ | |
5560 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5561 | { | |
5562 | int i, n, val, min_val, best_node = 0; | |
5563 | ||
5564 | min_val = INT_MAX; | |
5565 | ||
5566 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5567 | /* Start at @node */ | |
5568 | n = (node + i) % MAX_NUMNODES; | |
5569 | ||
5570 | if (!nr_cpus_node(n)) | |
5571 | continue; | |
5572 | ||
5573 | /* Skip already used nodes */ | |
5574 | if (test_bit(n, used_nodes)) | |
5575 | continue; | |
5576 | ||
5577 | /* Simple min distance search */ | |
5578 | val = node_distance(node, n); | |
5579 | ||
5580 | if (val < min_val) { | |
5581 | min_val = val; | |
5582 | best_node = n; | |
5583 | } | |
5584 | } | |
5585 | ||
5586 | set_bit(best_node, used_nodes); | |
5587 | return best_node; | |
5588 | } | |
5589 | ||
5590 | /** | |
5591 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5592 | * @node: node whose cpumask we're constructing | |
5593 | * @size: number of nodes to include in this span | |
5594 | * | |
5595 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5596 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5597 | * out optimally. | |
5598 | */ | |
5599 | static cpumask_t sched_domain_node_span(int node) | |
5600 | { | |
9c1cfda2 | 5601 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5602 | cpumask_t span, nodemask; |
5603 | int i; | |
9c1cfda2 JH |
5604 | |
5605 | cpus_clear(span); | |
5606 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5607 | ||
5608 | nodemask = node_to_cpumask(node); | |
5609 | cpus_or(span, span, nodemask); | |
5610 | set_bit(node, used_nodes); | |
5611 | ||
5612 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5613 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5614 | |
9c1cfda2 JH |
5615 | nodemask = node_to_cpumask(next_node); |
5616 | cpus_or(span, span, nodemask); | |
5617 | } | |
5618 | ||
5619 | return span; | |
5620 | } | |
5621 | #endif | |
5622 | ||
5c45bf27 | 5623 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5624 | |
9c1cfda2 | 5625 | /* |
48f24c4d | 5626 | * SMT sched-domains: |
9c1cfda2 | 5627 | */ |
1da177e4 LT |
5628 | #ifdef CONFIG_SCHED_SMT |
5629 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5630 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5631 | |
6711cab4 SS |
5632 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5633 | struct sched_group **sg) | |
1da177e4 | 5634 | { |
6711cab4 SS |
5635 | if (sg) |
5636 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5637 | return cpu; |
5638 | } | |
5639 | #endif | |
5640 | ||
48f24c4d IM |
5641 | /* |
5642 | * multi-core sched-domains: | |
5643 | */ | |
1e9f28fa SS |
5644 | #ifdef CONFIG_SCHED_MC |
5645 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5646 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5647 | #endif |
5648 | ||
5649 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5650 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5651 | struct sched_group **sg) | |
1e9f28fa | 5652 | { |
6711cab4 | 5653 | int group; |
a616058b SS |
5654 | cpumask_t mask = cpu_sibling_map[cpu]; |
5655 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5656 | group = first_cpu(mask); |
5657 | if (sg) | |
5658 | *sg = &per_cpu(sched_group_core, group); | |
5659 | return group; | |
1e9f28fa SS |
5660 | } |
5661 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5662 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5663 | struct sched_group **sg) | |
1e9f28fa | 5664 | { |
6711cab4 SS |
5665 | if (sg) |
5666 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5667 | return cpu; |
5668 | } | |
5669 | #endif | |
5670 | ||
1da177e4 | 5671 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5672 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5673 | |
6711cab4 SS |
5674 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5675 | struct sched_group **sg) | |
1da177e4 | 5676 | { |
6711cab4 | 5677 | int group; |
48f24c4d | 5678 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5679 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5680 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5681 | group = first_cpu(mask); |
1e9f28fa | 5682 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5683 | cpumask_t mask = cpu_sibling_map[cpu]; |
5684 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5685 | group = first_cpu(mask); |
1da177e4 | 5686 | #else |
6711cab4 | 5687 | group = cpu; |
1da177e4 | 5688 | #endif |
6711cab4 SS |
5689 | if (sg) |
5690 | *sg = &per_cpu(sched_group_phys, group); | |
5691 | return group; | |
1da177e4 LT |
5692 | } |
5693 | ||
5694 | #ifdef CONFIG_NUMA | |
1da177e4 | 5695 | /* |
9c1cfda2 JH |
5696 | * The init_sched_build_groups can't handle what we want to do with node |
5697 | * groups, so roll our own. Now each node has its own list of groups which | |
5698 | * gets dynamically allocated. | |
1da177e4 | 5699 | */ |
9c1cfda2 | 5700 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5701 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5702 | |
9c1cfda2 | 5703 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5704 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5705 | |
6711cab4 SS |
5706 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5707 | struct sched_group **sg) | |
9c1cfda2 | 5708 | { |
6711cab4 SS |
5709 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5710 | int group; | |
5711 | ||
5712 | cpus_and(nodemask, nodemask, *cpu_map); | |
5713 | group = first_cpu(nodemask); | |
5714 | ||
5715 | if (sg) | |
5716 | *sg = &per_cpu(sched_group_allnodes, group); | |
5717 | return group; | |
1da177e4 | 5718 | } |
6711cab4 | 5719 | |
08069033 SS |
5720 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5721 | { | |
5722 | struct sched_group *sg = group_head; | |
5723 | int j; | |
5724 | ||
5725 | if (!sg) | |
5726 | return; | |
5727 | next_sg: | |
5728 | for_each_cpu_mask(j, sg->cpumask) { | |
5729 | struct sched_domain *sd; | |
5730 | ||
5731 | sd = &per_cpu(phys_domains, j); | |
5732 | if (j != first_cpu(sd->groups->cpumask)) { | |
5733 | /* | |
5734 | * Only add "power" once for each | |
5735 | * physical package. | |
5736 | */ | |
5737 | continue; | |
5738 | } | |
5739 | ||
5517d86b | 5740 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5741 | } |
5742 | sg = sg->next; | |
5743 | if (sg != group_head) | |
5744 | goto next_sg; | |
5745 | } | |
1da177e4 LT |
5746 | #endif |
5747 | ||
a616058b | 5748 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5749 | /* Free memory allocated for various sched_group structures */ |
5750 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5751 | { | |
a616058b | 5752 | int cpu, i; |
51888ca2 SV |
5753 | |
5754 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5755 | struct sched_group **sched_group_nodes |
5756 | = sched_group_nodes_bycpu[cpu]; | |
5757 | ||
51888ca2 SV |
5758 | if (!sched_group_nodes) |
5759 | continue; | |
5760 | ||
5761 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5762 | cpumask_t nodemask = node_to_cpumask(i); | |
5763 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5764 | ||
5765 | cpus_and(nodemask, nodemask, *cpu_map); | |
5766 | if (cpus_empty(nodemask)) | |
5767 | continue; | |
5768 | ||
5769 | if (sg == NULL) | |
5770 | continue; | |
5771 | sg = sg->next; | |
5772 | next_sg: | |
5773 | oldsg = sg; | |
5774 | sg = sg->next; | |
5775 | kfree(oldsg); | |
5776 | if (oldsg != sched_group_nodes[i]) | |
5777 | goto next_sg; | |
5778 | } | |
5779 | kfree(sched_group_nodes); | |
5780 | sched_group_nodes_bycpu[cpu] = NULL; | |
5781 | } | |
51888ca2 | 5782 | } |
a616058b SS |
5783 | #else |
5784 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5785 | { | |
5786 | } | |
5787 | #endif | |
51888ca2 | 5788 | |
89c4710e SS |
5789 | /* |
5790 | * Initialize sched groups cpu_power. | |
5791 | * | |
5792 | * cpu_power indicates the capacity of sched group, which is used while | |
5793 | * distributing the load between different sched groups in a sched domain. | |
5794 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5795 | * there are asymmetries in the topology. If there are asymmetries, group | |
5796 | * having more cpu_power will pickup more load compared to the group having | |
5797 | * less cpu_power. | |
5798 | * | |
5799 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5800 | * the maximum number of tasks a group can handle in the presence of other idle | |
5801 | * or lightly loaded groups in the same sched domain. | |
5802 | */ | |
5803 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5804 | { | |
5805 | struct sched_domain *child; | |
5806 | struct sched_group *group; | |
5807 | ||
5808 | WARN_ON(!sd || !sd->groups); | |
5809 | ||
5810 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5811 | return; | |
5812 | ||
5813 | child = sd->child; | |
5814 | ||
5517d86b ED |
5815 | sd->groups->__cpu_power = 0; |
5816 | ||
89c4710e SS |
5817 | /* |
5818 | * For perf policy, if the groups in child domain share resources | |
5819 | * (for example cores sharing some portions of the cache hierarchy | |
5820 | * or SMT), then set this domain groups cpu_power such that each group | |
5821 | * can handle only one task, when there are other idle groups in the | |
5822 | * same sched domain. | |
5823 | */ | |
5824 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5825 | (child->flags & | |
5826 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5827 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5828 | return; |
5829 | } | |
5830 | ||
89c4710e SS |
5831 | /* |
5832 | * add cpu_power of each child group to this groups cpu_power | |
5833 | */ | |
5834 | group = child->groups; | |
5835 | do { | |
5517d86b | 5836 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5837 | group = group->next; |
5838 | } while (group != child->groups); | |
5839 | } | |
5840 | ||
1da177e4 | 5841 | /* |
1a20ff27 DG |
5842 | * Build sched domains for a given set of cpus and attach the sched domains |
5843 | * to the individual cpus | |
1da177e4 | 5844 | */ |
51888ca2 | 5845 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5846 | { |
5847 | int i; | |
89c4710e | 5848 | struct sched_domain *sd; |
d1b55138 JH |
5849 | #ifdef CONFIG_NUMA |
5850 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 5851 | int sd_allnodes = 0; |
d1b55138 JH |
5852 | |
5853 | /* | |
5854 | * Allocate the per-node list of sched groups | |
5855 | */ | |
51888ca2 | 5856 | sched_group_nodes = kzalloc(sizeof(struct sched_group*)*MAX_NUMNODES, |
d3a5aa98 | 5857 | GFP_KERNEL); |
d1b55138 JH |
5858 | if (!sched_group_nodes) { |
5859 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 5860 | return -ENOMEM; |
d1b55138 JH |
5861 | } |
5862 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
5863 | #endif | |
1da177e4 LT |
5864 | |
5865 | /* | |
1a20ff27 | 5866 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 5867 | */ |
1a20ff27 | 5868 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
5869 | struct sched_domain *sd = NULL, *p; |
5870 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
5871 | ||
1a20ff27 | 5872 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5873 | |
5874 | #ifdef CONFIG_NUMA | |
d1b55138 | 5875 | if (cpus_weight(*cpu_map) |
9c1cfda2 JH |
5876 | > SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { |
5877 | sd = &per_cpu(allnodes_domains, i); | |
5878 | *sd = SD_ALLNODES_INIT; | |
5879 | sd->span = *cpu_map; | |
6711cab4 | 5880 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 5881 | p = sd; |
6711cab4 | 5882 | sd_allnodes = 1; |
9c1cfda2 JH |
5883 | } else |
5884 | p = NULL; | |
5885 | ||
1da177e4 | 5886 | sd = &per_cpu(node_domains, i); |
1da177e4 | 5887 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
5888 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
5889 | sd->parent = p; | |
1a848870 SS |
5890 | if (p) |
5891 | p->child = sd; | |
9c1cfda2 | 5892 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
5893 | #endif |
5894 | ||
5895 | p = sd; | |
5896 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
5897 | *sd = SD_CPU_INIT; |
5898 | sd->span = nodemask; | |
5899 | sd->parent = p; | |
1a848870 SS |
5900 | if (p) |
5901 | p->child = sd; | |
6711cab4 | 5902 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 5903 | |
1e9f28fa SS |
5904 | #ifdef CONFIG_SCHED_MC |
5905 | p = sd; | |
5906 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
5907 | *sd = SD_MC_INIT; |
5908 | sd->span = cpu_coregroup_map(i); | |
5909 | cpus_and(sd->span, sd->span, *cpu_map); | |
5910 | sd->parent = p; | |
1a848870 | 5911 | p->child = sd; |
6711cab4 | 5912 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
5913 | #endif |
5914 | ||
1da177e4 LT |
5915 | #ifdef CONFIG_SCHED_SMT |
5916 | p = sd; | |
5917 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
5918 | *sd = SD_SIBLING_INIT; |
5919 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 5920 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 5921 | sd->parent = p; |
1a848870 | 5922 | p->child = sd; |
6711cab4 | 5923 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
5924 | #endif |
5925 | } | |
5926 | ||
5927 | #ifdef CONFIG_SCHED_SMT | |
5928 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 5929 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 5930 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 5931 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
5932 | if (i != first_cpu(this_sibling_map)) |
5933 | continue; | |
5934 | ||
6711cab4 | 5935 | init_sched_build_groups(this_sibling_map, cpu_map, &cpu_to_cpu_group); |
1da177e4 LT |
5936 | } |
5937 | #endif | |
5938 | ||
1e9f28fa SS |
5939 | #ifdef CONFIG_SCHED_MC |
5940 | /* Set up multi-core groups */ | |
5941 | for_each_cpu_mask(i, *cpu_map) { | |
5942 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
5943 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
5944 | if (i != first_cpu(this_core_map)) | |
5945 | continue; | |
6711cab4 | 5946 | init_sched_build_groups(this_core_map, cpu_map, &cpu_to_core_group); |
1e9f28fa SS |
5947 | } |
5948 | #endif | |
5949 | ||
5950 | ||
1da177e4 LT |
5951 | /* Set up physical groups */ |
5952 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5953 | cpumask_t nodemask = node_to_cpumask(i); | |
5954 | ||
1a20ff27 | 5955 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5956 | if (cpus_empty(nodemask)) |
5957 | continue; | |
5958 | ||
6711cab4 | 5959 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
5960 | } |
5961 | ||
5962 | #ifdef CONFIG_NUMA | |
5963 | /* Set up node groups */ | |
6711cab4 SS |
5964 | if (sd_allnodes) |
5965 | init_sched_build_groups(*cpu_map, cpu_map, &cpu_to_allnodes_group); | |
9c1cfda2 JH |
5966 | |
5967 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5968 | /* Set up node groups */ | |
5969 | struct sched_group *sg, *prev; | |
5970 | cpumask_t nodemask = node_to_cpumask(i); | |
5971 | cpumask_t domainspan; | |
5972 | cpumask_t covered = CPU_MASK_NONE; | |
5973 | int j; | |
5974 | ||
5975 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
5976 | if (cpus_empty(nodemask)) { |
5977 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 5978 | continue; |
d1b55138 | 5979 | } |
9c1cfda2 JH |
5980 | |
5981 | domainspan = sched_domain_node_span(i); | |
5982 | cpus_and(domainspan, domainspan, *cpu_map); | |
5983 | ||
15f0b676 | 5984 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
5985 | if (!sg) { |
5986 | printk(KERN_WARNING "Can not alloc domain group for " | |
5987 | "node %d\n", i); | |
5988 | goto error; | |
5989 | } | |
9c1cfda2 JH |
5990 | sched_group_nodes[i] = sg; |
5991 | for_each_cpu_mask(j, nodemask) { | |
5992 | struct sched_domain *sd; | |
5993 | sd = &per_cpu(node_domains, j); | |
5994 | sd->groups = sg; | |
9c1cfda2 | 5995 | } |
5517d86b | 5996 | sg->__cpu_power = 0; |
9c1cfda2 | 5997 | sg->cpumask = nodemask; |
51888ca2 | 5998 | sg->next = sg; |
9c1cfda2 JH |
5999 | cpus_or(covered, covered, nodemask); |
6000 | prev = sg; | |
6001 | ||
6002 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6003 | cpumask_t tmp, notcovered; | |
6004 | int n = (i + j) % MAX_NUMNODES; | |
6005 | ||
6006 | cpus_complement(notcovered, covered); | |
6007 | cpus_and(tmp, notcovered, *cpu_map); | |
6008 | cpus_and(tmp, tmp, domainspan); | |
6009 | if (cpus_empty(tmp)) | |
6010 | break; | |
6011 | ||
6012 | nodemask = node_to_cpumask(n); | |
6013 | cpus_and(tmp, tmp, nodemask); | |
6014 | if (cpus_empty(tmp)) | |
6015 | continue; | |
6016 | ||
15f0b676 SV |
6017 | sg = kmalloc_node(sizeof(struct sched_group), |
6018 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6019 | if (!sg) { |
6020 | printk(KERN_WARNING | |
6021 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6022 | goto error; |
9c1cfda2 | 6023 | } |
5517d86b | 6024 | sg->__cpu_power = 0; |
9c1cfda2 | 6025 | sg->cpumask = tmp; |
51888ca2 | 6026 | sg->next = prev->next; |
9c1cfda2 JH |
6027 | cpus_or(covered, covered, tmp); |
6028 | prev->next = sg; | |
6029 | prev = sg; | |
6030 | } | |
9c1cfda2 | 6031 | } |
1da177e4 LT |
6032 | #endif |
6033 | ||
6034 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6035 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6036 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6037 | sd = &per_cpu(cpu_domains, i); |
89c4710e | 6038 | init_sched_groups_power(i, sd); |
5c45bf27 | 6039 | } |
1da177e4 | 6040 | #endif |
1e9f28fa | 6041 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6042 | for_each_cpu_mask(i, *cpu_map) { |
1e9f28fa | 6043 | sd = &per_cpu(core_domains, i); |
89c4710e | 6044 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6045 | } |
6046 | #endif | |
1e9f28fa | 6047 | |
5c45bf27 | 6048 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6049 | sd = &per_cpu(phys_domains, i); |
89c4710e | 6050 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6051 | } |
6052 | ||
9c1cfda2 | 6053 | #ifdef CONFIG_NUMA |
08069033 SS |
6054 | for (i = 0; i < MAX_NUMNODES; i++) |
6055 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6056 | |
6711cab4 SS |
6057 | if (sd_allnodes) { |
6058 | struct sched_group *sg; | |
f712c0c7 | 6059 | |
6711cab4 | 6060 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6061 | init_numa_sched_groups_power(sg); |
6062 | } | |
9c1cfda2 JH |
6063 | #endif |
6064 | ||
1da177e4 | 6065 | /* Attach the domains */ |
1a20ff27 | 6066 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6067 | struct sched_domain *sd; |
6068 | #ifdef CONFIG_SCHED_SMT | |
6069 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6070 | #elif defined(CONFIG_SCHED_MC) |
6071 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6072 | #else |
6073 | sd = &per_cpu(phys_domains, i); | |
6074 | #endif | |
6075 | cpu_attach_domain(sd, i); | |
6076 | } | |
51888ca2 SV |
6077 | |
6078 | return 0; | |
6079 | ||
a616058b | 6080 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6081 | error: |
6082 | free_sched_groups(cpu_map); | |
6083 | return -ENOMEM; | |
a616058b | 6084 | #endif |
1da177e4 | 6085 | } |
1a20ff27 DG |
6086 | /* |
6087 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6088 | */ | |
51888ca2 | 6089 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6090 | { |
6091 | cpumask_t cpu_default_map; | |
51888ca2 | 6092 | int err; |
1da177e4 | 6093 | |
1a20ff27 DG |
6094 | /* |
6095 | * Setup mask for cpus without special case scheduling requirements. | |
6096 | * For now this just excludes isolated cpus, but could be used to | |
6097 | * exclude other special cases in the future. | |
6098 | */ | |
6099 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6100 | ||
51888ca2 SV |
6101 | err = build_sched_domains(&cpu_default_map); |
6102 | ||
6103 | return err; | |
1a20ff27 DG |
6104 | } |
6105 | ||
6106 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6107 | { |
51888ca2 | 6108 | free_sched_groups(cpu_map); |
9c1cfda2 | 6109 | } |
1da177e4 | 6110 | |
1a20ff27 DG |
6111 | /* |
6112 | * Detach sched domains from a group of cpus specified in cpu_map | |
6113 | * These cpus will now be attached to the NULL domain | |
6114 | */ | |
858119e1 | 6115 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6116 | { |
6117 | int i; | |
6118 | ||
6119 | for_each_cpu_mask(i, *cpu_map) | |
6120 | cpu_attach_domain(NULL, i); | |
6121 | synchronize_sched(); | |
6122 | arch_destroy_sched_domains(cpu_map); | |
6123 | } | |
6124 | ||
6125 | /* | |
6126 | * Partition sched domains as specified by the cpumasks below. | |
6127 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6128 | * waits for a RCU quiescent period, recalculates sched | |
6129 | * domain information and then attaches them back to the | |
6130 | * correct sched domains | |
6131 | * Call with hotplug lock held | |
6132 | */ | |
51888ca2 | 6133 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6134 | { |
6135 | cpumask_t change_map; | |
51888ca2 | 6136 | int err = 0; |
1a20ff27 DG |
6137 | |
6138 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6139 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6140 | cpus_or(change_map, *partition1, *partition2); | |
6141 | ||
6142 | /* Detach sched domains from all of the affected cpus */ | |
6143 | detach_destroy_domains(&change_map); | |
6144 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6145 | err = build_sched_domains(partition1); |
6146 | if (!err && !cpus_empty(*partition2)) | |
6147 | err = build_sched_domains(partition2); | |
6148 | ||
6149 | return err; | |
1a20ff27 DG |
6150 | } |
6151 | ||
5c45bf27 SS |
6152 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6153 | int arch_reinit_sched_domains(void) | |
6154 | { | |
6155 | int err; | |
6156 | ||
5be9361c | 6157 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6158 | detach_destroy_domains(&cpu_online_map); |
6159 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6160 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6161 | |
6162 | return err; | |
6163 | } | |
6164 | ||
6165 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6166 | { | |
6167 | int ret; | |
6168 | ||
6169 | if (buf[0] != '0' && buf[0] != '1') | |
6170 | return -EINVAL; | |
6171 | ||
6172 | if (smt) | |
6173 | sched_smt_power_savings = (buf[0] == '1'); | |
6174 | else | |
6175 | sched_mc_power_savings = (buf[0] == '1'); | |
6176 | ||
6177 | ret = arch_reinit_sched_domains(); | |
6178 | ||
6179 | return ret ? ret : count; | |
6180 | } | |
6181 | ||
6182 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6183 | { | |
6184 | int err = 0; | |
48f24c4d | 6185 | |
5c45bf27 SS |
6186 | #ifdef CONFIG_SCHED_SMT |
6187 | if (smt_capable()) | |
6188 | err = sysfs_create_file(&cls->kset.kobj, | |
6189 | &attr_sched_smt_power_savings.attr); | |
6190 | #endif | |
6191 | #ifdef CONFIG_SCHED_MC | |
6192 | if (!err && mc_capable()) | |
6193 | err = sysfs_create_file(&cls->kset.kobj, | |
6194 | &attr_sched_mc_power_savings.attr); | |
6195 | #endif | |
6196 | return err; | |
6197 | } | |
6198 | #endif | |
6199 | ||
6200 | #ifdef CONFIG_SCHED_MC | |
6201 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6202 | { | |
6203 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6204 | } | |
48f24c4d IM |
6205 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6206 | const char *buf, size_t count) | |
5c45bf27 SS |
6207 | { |
6208 | return sched_power_savings_store(buf, count, 0); | |
6209 | } | |
6210 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6211 | sched_mc_power_savings_store); | |
6212 | #endif | |
6213 | ||
6214 | #ifdef CONFIG_SCHED_SMT | |
6215 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6216 | { | |
6217 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6218 | } | |
48f24c4d IM |
6219 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6220 | const char *buf, size_t count) | |
5c45bf27 SS |
6221 | { |
6222 | return sched_power_savings_store(buf, count, 1); | |
6223 | } | |
6224 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6225 | sched_smt_power_savings_store); | |
6226 | #endif | |
6227 | ||
1da177e4 LT |
6228 | /* |
6229 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6230 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6231 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6232 | * which will prevent rebalancing while the sched domains are recalculated. |
6233 | */ | |
6234 | static int update_sched_domains(struct notifier_block *nfb, | |
6235 | unsigned long action, void *hcpu) | |
6236 | { | |
1da177e4 LT |
6237 | switch (action) { |
6238 | case CPU_UP_PREPARE: | |
8bb78442 | 6239 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6240 | case CPU_DOWN_PREPARE: |
8bb78442 | 6241 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6242 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6243 | return NOTIFY_OK; |
6244 | ||
6245 | case CPU_UP_CANCELED: | |
8bb78442 | 6246 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6247 | case CPU_DOWN_FAILED: |
8bb78442 | 6248 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6249 | case CPU_ONLINE: |
8bb78442 | 6250 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6251 | case CPU_DEAD: |
8bb78442 | 6252 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6253 | /* |
6254 | * Fall through and re-initialise the domains. | |
6255 | */ | |
6256 | break; | |
6257 | default: | |
6258 | return NOTIFY_DONE; | |
6259 | } | |
6260 | ||
6261 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6262 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6263 | |
6264 | return NOTIFY_OK; | |
6265 | } | |
1da177e4 LT |
6266 | |
6267 | void __init sched_init_smp(void) | |
6268 | { | |
5c1e1767 NP |
6269 | cpumask_t non_isolated_cpus; |
6270 | ||
5be9361c | 6271 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6272 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6273 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6274 | if (cpus_empty(non_isolated_cpus)) |
6275 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6276 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6277 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6278 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6279 | |
6280 | /* Move init over to a non-isolated CPU */ | |
6281 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6282 | BUG(); | |
1da177e4 LT |
6283 | } |
6284 | #else | |
6285 | void __init sched_init_smp(void) | |
6286 | { | |
6287 | } | |
6288 | #endif /* CONFIG_SMP */ | |
6289 | ||
6290 | int in_sched_functions(unsigned long addr) | |
6291 | { | |
6292 | /* Linker adds these: start and end of __sched functions */ | |
6293 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6294 | |
1da177e4 LT |
6295 | return in_lock_functions(addr) || |
6296 | (addr >= (unsigned long)__sched_text_start | |
6297 | && addr < (unsigned long)__sched_text_end); | |
6298 | } | |
6299 | ||
6300 | void __init sched_init(void) | |
6301 | { | |
1da177e4 | 6302 | int i, j, k; |
476f3534 | 6303 | int highest_cpu = 0; |
1da177e4 | 6304 | |
0a945022 | 6305 | for_each_possible_cpu(i) { |
70b97a7f IM |
6306 | struct prio_array *array; |
6307 | struct rq *rq; | |
1da177e4 LT |
6308 | |
6309 | rq = cpu_rq(i); | |
6310 | spin_lock_init(&rq->lock); | |
fcb99371 | 6311 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6312 | rq->nr_running = 0; |
1da177e4 LT |
6313 | rq->active = rq->arrays; |
6314 | rq->expired = rq->arrays + 1; | |
6315 | rq->best_expired_prio = MAX_PRIO; | |
6316 | ||
6317 | #ifdef CONFIG_SMP | |
41c7ce9a | 6318 | rq->sd = NULL; |
7897986b NP |
6319 | for (j = 1; j < 3; j++) |
6320 | rq->cpu_load[j] = 0; | |
1da177e4 LT |
6321 | rq->active_balance = 0; |
6322 | rq->push_cpu = 0; | |
0a2966b4 | 6323 | rq->cpu = i; |
1da177e4 LT |
6324 | rq->migration_thread = NULL; |
6325 | INIT_LIST_HEAD(&rq->migration_queue); | |
6326 | #endif | |
6327 | atomic_set(&rq->nr_iowait, 0); | |
6328 | ||
6329 | for (j = 0; j < 2; j++) { | |
6330 | array = rq->arrays + j; | |
6331 | for (k = 0; k < MAX_PRIO; k++) { | |
6332 | INIT_LIST_HEAD(array->queue + k); | |
6333 | __clear_bit(k, array->bitmap); | |
6334 | } | |
6335 | // delimiter for bitsearch | |
6336 | __set_bit(MAX_PRIO, array->bitmap); | |
6337 | } | |
476f3534 | 6338 | highest_cpu = i; |
1da177e4 LT |
6339 | } |
6340 | ||
2dd73a4f | 6341 | set_load_weight(&init_task); |
b50f60ce | 6342 | |
c9819f45 | 6343 | #ifdef CONFIG_SMP |
476f3534 | 6344 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6345 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6346 | #endif | |
6347 | ||
b50f60ce HC |
6348 | #ifdef CONFIG_RT_MUTEXES |
6349 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6350 | #endif | |
6351 | ||
1da177e4 LT |
6352 | /* |
6353 | * The boot idle thread does lazy MMU switching as well: | |
6354 | */ | |
6355 | atomic_inc(&init_mm.mm_count); | |
6356 | enter_lazy_tlb(&init_mm, current); | |
6357 | ||
6358 | /* | |
6359 | * Make us the idle thread. Technically, schedule() should not be | |
6360 | * called from this thread, however somewhere below it might be, | |
6361 | * but because we are the idle thread, we just pick up running again | |
6362 | * when this runqueue becomes "idle". | |
6363 | */ | |
6364 | init_idle(current, smp_processor_id()); | |
6365 | } | |
6366 | ||
6367 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6368 | void __might_sleep(char *file, int line) | |
6369 | { | |
48f24c4d | 6370 | #ifdef in_atomic |
1da177e4 LT |
6371 | static unsigned long prev_jiffy; /* ratelimiting */ |
6372 | ||
6373 | if ((in_atomic() || irqs_disabled()) && | |
6374 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6375 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6376 | return; | |
6377 | prev_jiffy = jiffies; | |
91368d73 | 6378 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6379 | " context at %s:%d\n", file, line); |
6380 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6381 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6382 | debug_show_held_locks(current); |
3117df04 IM |
6383 | if (irqs_disabled()) |
6384 | print_irqtrace_events(current); | |
1da177e4 LT |
6385 | dump_stack(); |
6386 | } | |
6387 | #endif | |
6388 | } | |
6389 | EXPORT_SYMBOL(__might_sleep); | |
6390 | #endif | |
6391 | ||
6392 | #ifdef CONFIG_MAGIC_SYSRQ | |
6393 | void normalize_rt_tasks(void) | |
6394 | { | |
70b97a7f | 6395 | struct prio_array *array; |
a0f98a1c | 6396 | struct task_struct *g, *p; |
1da177e4 | 6397 | unsigned long flags; |
70b97a7f | 6398 | struct rq *rq; |
1da177e4 LT |
6399 | |
6400 | read_lock_irq(&tasklist_lock); | |
a0f98a1c IM |
6401 | |
6402 | do_each_thread(g, p) { | |
1da177e4 LT |
6403 | if (!rt_task(p)) |
6404 | continue; | |
6405 | ||
b29739f9 IM |
6406 | spin_lock_irqsave(&p->pi_lock, flags); |
6407 | rq = __task_rq_lock(p); | |
1da177e4 LT |
6408 | |
6409 | array = p->array; | |
6410 | if (array) | |
6411 | deactivate_task(p, task_rq(p)); | |
6412 | __setscheduler(p, SCHED_NORMAL, 0); | |
6413 | if (array) { | |
6414 | __activate_task(p, task_rq(p)); | |
6415 | resched_task(rq->curr); | |
6416 | } | |
6417 | ||
b29739f9 IM |
6418 | __task_rq_unlock(rq); |
6419 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6420 | } while_each_thread(g, p); |
6421 | ||
1da177e4 LT |
6422 | read_unlock_irq(&tasklist_lock); |
6423 | } | |
6424 | ||
6425 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6426 | |
6427 | #ifdef CONFIG_IA64 | |
6428 | /* | |
6429 | * These functions are only useful for the IA64 MCA handling. | |
6430 | * | |
6431 | * They can only be called when the whole system has been | |
6432 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6433 | * activity can take place. Using them for anything else would | |
6434 | * be a serious bug, and as a result, they aren't even visible | |
6435 | * under any other configuration. | |
6436 | */ | |
6437 | ||
6438 | /** | |
6439 | * curr_task - return the current task for a given cpu. | |
6440 | * @cpu: the processor in question. | |
6441 | * | |
6442 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6443 | */ | |
36c8b586 | 6444 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6445 | { |
6446 | return cpu_curr(cpu); | |
6447 | } | |
6448 | ||
6449 | /** | |
6450 | * set_curr_task - set the current task for a given cpu. | |
6451 | * @cpu: the processor in question. | |
6452 | * @p: the task pointer to set. | |
6453 | * | |
6454 | * Description: This function must only be used when non-maskable interrupts | |
6455 | * are serviced on a separate stack. It allows the architecture to switch the | |
6456 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6457 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6458 | * and caller must save the original value of the current task (see | |
6459 | * curr_task() above) and restore that value before reenabling interrupts and | |
6460 | * re-starting the system. | |
6461 | * | |
6462 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6463 | */ | |
36c8b586 | 6464 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6465 | { |
6466 | cpu_curr(cpu) = p; | |
6467 | } | |
6468 | ||
6469 | #endif |