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 | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
1da177e4 LT |
25 | */ |
26 | ||
27 | #include <linux/mm.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/nmi.h> | |
30 | #include <linux/init.h> | |
dff06c15 | 31 | #include <linux/uaccess.h> |
1da177e4 LT |
32 | #include <linux/highmem.h> |
33 | #include <linux/smp_lock.h> | |
34 | #include <asm/mmu_context.h> | |
35 | #include <linux/interrupt.h> | |
c59ede7b | 36 | #include <linux/capability.h> |
1da177e4 LT |
37 | #include <linux/completion.h> |
38 | #include <linux/kernel_stat.h> | |
9a11b49a | 39 | #include <linux/debug_locks.h> |
1da177e4 LT |
40 | #include <linux/security.h> |
41 | #include <linux/notifier.h> | |
42 | #include <linux/profile.h> | |
7dfb7103 | 43 | #include <linux/freezer.h> |
198e2f18 | 44 | #include <linux/vmalloc.h> |
1da177e4 LT |
45 | #include <linux/blkdev.h> |
46 | #include <linux/delay.h> | |
47 | #include <linux/smp.h> | |
48 | #include <linux/threads.h> | |
49 | #include <linux/timer.h> | |
50 | #include <linux/rcupdate.h> | |
51 | #include <linux/cpu.h> | |
52 | #include <linux/cpuset.h> | |
53 | #include <linux/percpu.h> | |
54 | #include <linux/kthread.h> | |
55 | #include <linux/seq_file.h> | |
56 | #include <linux/syscalls.h> | |
57 | #include <linux/times.h> | |
8f0ab514 | 58 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 59 | #include <linux/kprobes.h> |
0ff92245 | 60 | #include <linux/delayacct.h> |
5517d86b | 61 | #include <linux/reciprocal_div.h> |
dff06c15 | 62 | #include <linux/unistd.h> |
1da177e4 | 63 | |
5517d86b | 64 | #include <asm/tlb.h> |
1da177e4 | 65 | |
b035b6de AD |
66 | /* |
67 | * Scheduler clock - returns current time in nanosec units. | |
68 | * This is default implementation. | |
69 | * Architectures and sub-architectures can override this. | |
70 | */ | |
71 | unsigned long long __attribute__((weak)) sched_clock(void) | |
72 | { | |
73 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
74 | } | |
75 | ||
1da177e4 LT |
76 | /* |
77 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
78 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
79 | * and back. | |
80 | */ | |
81 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
82 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
83 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
84 | ||
85 | /* | |
86 | * 'User priority' is the nice value converted to something we | |
87 | * can work with better when scaling various scheduler parameters, | |
88 | * it's a [ 0 ... 39 ] range. | |
89 | */ | |
90 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
91 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
92 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
93 | ||
94 | /* | |
95 | * Some helpers for converting nanosecond timing to jiffy resolution | |
96 | */ | |
97 | #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) | |
98 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) | |
99 | ||
6aa645ea IM |
100 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
101 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
102 | ||
1da177e4 LT |
103 | /* |
104 | * These are the 'tuning knobs' of the scheduler: | |
105 | * | |
106 | * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger), | |
107 | * default timeslice is 100 msecs, maximum timeslice is 800 msecs. | |
108 | * Timeslices get refilled after they expire. | |
109 | */ | |
110 | #define MIN_TIMESLICE max(5 * HZ / 1000, 1) | |
111 | #define DEF_TIMESLICE (100 * HZ / 1000) | |
2dd73a4f | 112 | |
5517d86b ED |
113 | #ifdef CONFIG_SMP |
114 | /* | |
115 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
116 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
117 | */ | |
118 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
119 | { | |
120 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Each time a sched group cpu_power is changed, | |
125 | * we must compute its reciprocal value | |
126 | */ | |
127 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
128 | { | |
129 | sg->__cpu_power += val; | |
130 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
131 | } | |
132 | #endif | |
133 | ||
634fa8c9 IM |
134 | #define SCALE_PRIO(x, prio) \ |
135 | max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE) | |
136 | ||
91fcdd4e | 137 | /* |
634fa8c9 | 138 | * static_prio_timeslice() scales user-nice values [ -20 ... 0 ... 19 ] |
91fcdd4e | 139 | * to time slice values: [800ms ... 100ms ... 5ms] |
91fcdd4e | 140 | */ |
634fa8c9 | 141 | static unsigned int static_prio_timeslice(int static_prio) |
2dd73a4f | 142 | { |
634fa8c9 IM |
143 | if (static_prio == NICE_TO_PRIO(19)) |
144 | return 1; | |
145 | ||
146 | if (static_prio < NICE_TO_PRIO(0)) | |
147 | return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio); | |
148 | else | |
149 | return SCALE_PRIO(DEF_TIMESLICE, static_prio); | |
2dd73a4f PW |
150 | } |
151 | ||
e05606d3 IM |
152 | static inline int rt_policy(int policy) |
153 | { | |
154 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
155 | return 1; | |
156 | return 0; | |
157 | } | |
158 | ||
159 | static inline int task_has_rt_policy(struct task_struct *p) | |
160 | { | |
161 | return rt_policy(p->policy); | |
162 | } | |
163 | ||
1da177e4 | 164 | /* |
6aa645ea | 165 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 166 | */ |
6aa645ea IM |
167 | struct rt_prio_array { |
168 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
169 | struct list_head queue[MAX_RT_PRIO]; | |
170 | }; | |
171 | ||
172 | struct load_stat { | |
173 | struct load_weight load; | |
174 | u64 load_update_start, load_update_last; | |
175 | unsigned long delta_fair, delta_exec, delta_stat; | |
176 | }; | |
177 | ||
178 | /* CFS-related fields in a runqueue */ | |
179 | struct cfs_rq { | |
180 | struct load_weight load; | |
181 | unsigned long nr_running; | |
182 | ||
183 | s64 fair_clock; | |
184 | u64 exec_clock; | |
185 | s64 wait_runtime; | |
186 | u64 sleeper_bonus; | |
187 | unsigned long wait_runtime_overruns, wait_runtime_underruns; | |
188 | ||
189 | struct rb_root tasks_timeline; | |
190 | struct rb_node *rb_leftmost; | |
191 | struct rb_node *rb_load_balance_curr; | |
192 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
193 | /* 'curr' points to currently running entity on this cfs_rq. | |
194 | * It is set to NULL otherwise (i.e when none are currently running). | |
195 | */ | |
196 | struct sched_entity *curr; | |
197 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
198 | ||
199 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
200 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
201 | * (like users, containers etc.) | |
202 | * | |
203 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
204 | * list is used during load balance. | |
205 | */ | |
206 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
207 | #endif | |
208 | }; | |
1da177e4 | 209 | |
6aa645ea IM |
210 | /* Real-Time classes' related field in a runqueue: */ |
211 | struct rt_rq { | |
212 | struct rt_prio_array active; | |
213 | int rt_load_balance_idx; | |
214 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
215 | }; | |
216 | ||
1da177e4 LT |
217 | /* |
218 | * This is the main, per-CPU runqueue data structure. | |
219 | * | |
220 | * Locking rule: those places that want to lock multiple runqueues | |
221 | * (such as the load balancing or the thread migration code), lock | |
222 | * acquire operations must be ordered by ascending &runqueue. | |
223 | */ | |
70b97a7f | 224 | struct rq { |
6aa645ea | 225 | spinlock_t lock; /* runqueue lock */ |
1da177e4 LT |
226 | |
227 | /* | |
228 | * nr_running and cpu_load should be in the same cacheline because | |
229 | * remote CPUs use both these fields when doing load calculation. | |
230 | */ | |
231 | unsigned long nr_running; | |
6aa645ea IM |
232 | #define CPU_LOAD_IDX_MAX 5 |
233 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 234 | unsigned char idle_at_tick; |
46cb4b7c SS |
235 | #ifdef CONFIG_NO_HZ |
236 | unsigned char in_nohz_recently; | |
237 | #endif | |
6aa645ea IM |
238 | struct load_stat ls; /* capture load from *all* tasks on this cpu */ |
239 | unsigned long nr_load_updates; | |
240 | u64 nr_switches; | |
241 | ||
242 | struct cfs_rq cfs; | |
243 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
244 | struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */ | |
1da177e4 | 245 | #endif |
6aa645ea | 246 | struct rt_rq rt; |
1da177e4 LT |
247 | |
248 | /* | |
249 | * This is part of a global counter where only the total sum | |
250 | * over all CPUs matters. A task can increase this counter on | |
251 | * one CPU and if it got migrated afterwards it may decrease | |
252 | * it on another CPU. Always updated under the runqueue lock: | |
253 | */ | |
254 | unsigned long nr_uninterruptible; | |
255 | ||
36c8b586 | 256 | struct task_struct *curr, *idle; |
c9819f45 | 257 | unsigned long next_balance; |
1da177e4 | 258 | struct mm_struct *prev_mm; |
6aa645ea | 259 | |
6aa645ea IM |
260 | u64 clock, prev_clock_raw; |
261 | s64 clock_max_delta; | |
262 | ||
263 | unsigned int clock_warps, clock_overflows; | |
264 | unsigned int clock_unstable_events; | |
265 | ||
1da177e4 LT |
266 | atomic_t nr_iowait; |
267 | ||
268 | #ifdef CONFIG_SMP | |
269 | struct sched_domain *sd; | |
270 | ||
271 | /* For active balancing */ | |
272 | int active_balance; | |
273 | int push_cpu; | |
0a2966b4 | 274 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 275 | |
36c8b586 | 276 | struct task_struct *migration_thread; |
1da177e4 LT |
277 | struct list_head migration_queue; |
278 | #endif | |
279 | ||
280 | #ifdef CONFIG_SCHEDSTATS | |
281 | /* latency stats */ | |
282 | struct sched_info rq_sched_info; | |
283 | ||
284 | /* sys_sched_yield() stats */ | |
285 | unsigned long yld_exp_empty; | |
286 | unsigned long yld_act_empty; | |
287 | unsigned long yld_both_empty; | |
288 | unsigned long yld_cnt; | |
289 | ||
290 | /* schedule() stats */ | |
291 | unsigned long sched_switch; | |
292 | unsigned long sched_cnt; | |
293 | unsigned long sched_goidle; | |
294 | ||
295 | /* try_to_wake_up() stats */ | |
296 | unsigned long ttwu_cnt; | |
297 | unsigned long ttwu_local; | |
298 | #endif | |
fcb99371 | 299 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
300 | }; |
301 | ||
f34e3b61 | 302 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 303 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 304 | |
dd41f596 IM |
305 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
306 | { | |
307 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
308 | } | |
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 | ||
20d315d4 IM |
319 | /* |
320 | * Per-runqueue clock, as finegrained as the platform can give us: | |
321 | */ | |
322 | static unsigned long long __rq_clock(struct rq *rq) | |
323 | { | |
324 | u64 prev_raw = rq->prev_clock_raw; | |
325 | u64 now = sched_clock(); | |
326 | s64 delta = now - prev_raw; | |
327 | u64 clock = rq->clock; | |
328 | ||
329 | /* | |
330 | * Protect against sched_clock() occasionally going backwards: | |
331 | */ | |
332 | if (unlikely(delta < 0)) { | |
333 | clock++; | |
334 | rq->clock_warps++; | |
335 | } else { | |
336 | /* | |
337 | * Catch too large forward jumps too: | |
338 | */ | |
339 | if (unlikely(delta > 2*TICK_NSEC)) { | |
340 | clock++; | |
341 | rq->clock_overflows++; | |
342 | } else { | |
343 | if (unlikely(delta > rq->clock_max_delta)) | |
344 | rq->clock_max_delta = delta; | |
345 | clock += delta; | |
346 | } | |
347 | } | |
348 | ||
349 | rq->prev_clock_raw = now; | |
350 | rq->clock = clock; | |
351 | ||
352 | return clock; | |
353 | } | |
354 | ||
355 | static inline unsigned long long rq_clock(struct rq *rq) | |
356 | { | |
357 | int this_cpu = smp_processor_id(); | |
358 | ||
359 | if (this_cpu == cpu_of(rq)) | |
360 | return __rq_clock(rq); | |
361 | ||
362 | return rq->clock; | |
363 | } | |
364 | ||
674311d5 NP |
365 | /* |
366 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 367 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
368 | * |
369 | * The domain tree of any CPU may only be accessed from within | |
370 | * preempt-disabled sections. | |
371 | */ | |
48f24c4d IM |
372 | #define for_each_domain(cpu, __sd) \ |
373 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
374 | |
375 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
376 | #define this_rq() (&__get_cpu_var(runqueues)) | |
377 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
378 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
379 | ||
e436d800 IM |
380 | /* |
381 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
382 | * clock constructed from sched_clock(): | |
383 | */ | |
384 | unsigned long long cpu_clock(int cpu) | |
385 | { | |
386 | struct rq *rq = cpu_rq(cpu); | |
387 | unsigned long long now; | |
388 | unsigned long flags; | |
389 | ||
390 | spin_lock_irqsave(&rq->lock, flags); | |
391 | now = rq_clock(rq); | |
392 | spin_unlock_irqrestore(&rq->lock, flags); | |
393 | ||
394 | return now; | |
395 | } | |
396 | ||
138a8aeb IM |
397 | #ifdef CONFIG_FAIR_GROUP_SCHED |
398 | /* Change a task's ->cfs_rq if it moves across CPUs */ | |
399 | static inline void set_task_cfs_rq(struct task_struct *p) | |
400 | { | |
401 | p->se.cfs_rq = &task_rq(p)->cfs; | |
402 | } | |
403 | #else | |
404 | static inline void set_task_cfs_rq(struct task_struct *p) | |
405 | { | |
406 | } | |
407 | #endif | |
408 | ||
1da177e4 | 409 | #ifndef prepare_arch_switch |
4866cde0 NP |
410 | # define prepare_arch_switch(next) do { } while (0) |
411 | #endif | |
412 | #ifndef finish_arch_switch | |
413 | # define finish_arch_switch(prev) do { } while (0) | |
414 | #endif | |
415 | ||
416 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 417 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
418 | { |
419 | return rq->curr == p; | |
420 | } | |
421 | ||
70b97a7f | 422 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
423 | { |
424 | } | |
425 | ||
70b97a7f | 426 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 427 | { |
da04c035 IM |
428 | #ifdef CONFIG_DEBUG_SPINLOCK |
429 | /* this is a valid case when another task releases the spinlock */ | |
430 | rq->lock.owner = current; | |
431 | #endif | |
8a25d5de IM |
432 | /* |
433 | * If we are tracking spinlock dependencies then we have to | |
434 | * fix up the runqueue lock - which gets 'carried over' from | |
435 | * prev into current: | |
436 | */ | |
437 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
438 | ||
4866cde0 NP |
439 | spin_unlock_irq(&rq->lock); |
440 | } | |
441 | ||
442 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 443 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
444 | { |
445 | #ifdef CONFIG_SMP | |
446 | return p->oncpu; | |
447 | #else | |
448 | return rq->curr == p; | |
449 | #endif | |
450 | } | |
451 | ||
70b97a7f | 452 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
453 | { |
454 | #ifdef CONFIG_SMP | |
455 | /* | |
456 | * We can optimise this out completely for !SMP, because the | |
457 | * SMP rebalancing from interrupt is the only thing that cares | |
458 | * here. | |
459 | */ | |
460 | next->oncpu = 1; | |
461 | #endif | |
462 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
463 | spin_unlock_irq(&rq->lock); | |
464 | #else | |
465 | spin_unlock(&rq->lock); | |
466 | #endif | |
467 | } | |
468 | ||
70b97a7f | 469 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
470 | { |
471 | #ifdef CONFIG_SMP | |
472 | /* | |
473 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
474 | * We must ensure this doesn't happen until the switch is completely | |
475 | * finished. | |
476 | */ | |
477 | smp_wmb(); | |
478 | prev->oncpu = 0; | |
479 | #endif | |
480 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
481 | local_irq_enable(); | |
1da177e4 | 482 | #endif |
4866cde0 NP |
483 | } |
484 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 485 | |
b29739f9 IM |
486 | /* |
487 | * __task_rq_lock - lock the runqueue a given task resides on. | |
488 | * Must be called interrupts disabled. | |
489 | */ | |
70b97a7f | 490 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
491 | __acquires(rq->lock) |
492 | { | |
70b97a7f | 493 | struct rq *rq; |
b29739f9 IM |
494 | |
495 | repeat_lock_task: | |
496 | rq = task_rq(p); | |
497 | spin_lock(&rq->lock); | |
498 | if (unlikely(rq != task_rq(p))) { | |
499 | spin_unlock(&rq->lock); | |
500 | goto repeat_lock_task; | |
501 | } | |
502 | return rq; | |
503 | } | |
504 | ||
1da177e4 LT |
505 | /* |
506 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
507 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
508 | * explicitly disabling preemption. | |
509 | */ | |
70b97a7f | 510 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
511 | __acquires(rq->lock) |
512 | { | |
70b97a7f | 513 | struct rq *rq; |
1da177e4 LT |
514 | |
515 | repeat_lock_task: | |
516 | local_irq_save(*flags); | |
517 | rq = task_rq(p); | |
518 | spin_lock(&rq->lock); | |
519 | if (unlikely(rq != task_rq(p))) { | |
520 | spin_unlock_irqrestore(&rq->lock, *flags); | |
521 | goto repeat_lock_task; | |
522 | } | |
523 | return rq; | |
524 | } | |
525 | ||
70b97a7f | 526 | static inline void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
527 | __releases(rq->lock) |
528 | { | |
529 | spin_unlock(&rq->lock); | |
530 | } | |
531 | ||
70b97a7f | 532 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
533 | __releases(rq->lock) |
534 | { | |
535 | spin_unlock_irqrestore(&rq->lock, *flags); | |
536 | } | |
537 | ||
1da177e4 | 538 | /* |
cc2a73b5 | 539 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 540 | */ |
70b97a7f | 541 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
542 | __acquires(rq->lock) |
543 | { | |
70b97a7f | 544 | struct rq *rq; |
1da177e4 LT |
545 | |
546 | local_irq_disable(); | |
547 | rq = this_rq(); | |
548 | spin_lock(&rq->lock); | |
549 | ||
550 | return rq; | |
551 | } | |
552 | ||
1b9f19c2 IM |
553 | /* |
554 | * CPU frequency is/was unstable - start new by setting prev_clock_raw: | |
555 | */ | |
556 | void sched_clock_unstable_event(void) | |
557 | { | |
558 | unsigned long flags; | |
559 | struct rq *rq; | |
560 | ||
561 | rq = task_rq_lock(current, &flags); | |
562 | rq->prev_clock_raw = sched_clock(); | |
563 | rq->clock_unstable_events++; | |
564 | task_rq_unlock(rq, &flags); | |
565 | } | |
566 | ||
c24d20db IM |
567 | /* |
568 | * resched_task - mark a task 'to be rescheduled now'. | |
569 | * | |
570 | * On UP this means the setting of the need_resched flag, on SMP it | |
571 | * might also involve a cross-CPU call to trigger the scheduler on | |
572 | * the target CPU. | |
573 | */ | |
574 | #ifdef CONFIG_SMP | |
575 | ||
576 | #ifndef tsk_is_polling | |
577 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
578 | #endif | |
579 | ||
580 | static void resched_task(struct task_struct *p) | |
581 | { | |
582 | int cpu; | |
583 | ||
584 | assert_spin_locked(&task_rq(p)->lock); | |
585 | ||
586 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
587 | return; | |
588 | ||
589 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
590 | ||
591 | cpu = task_cpu(p); | |
592 | if (cpu == smp_processor_id()) | |
593 | return; | |
594 | ||
595 | /* NEED_RESCHED must be visible before we test polling */ | |
596 | smp_mb(); | |
597 | if (!tsk_is_polling(p)) | |
598 | smp_send_reschedule(cpu); | |
599 | } | |
600 | ||
601 | static void resched_cpu(int cpu) | |
602 | { | |
603 | struct rq *rq = cpu_rq(cpu); | |
604 | unsigned long flags; | |
605 | ||
606 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
607 | return; | |
608 | resched_task(cpu_curr(cpu)); | |
609 | spin_unlock_irqrestore(&rq->lock, flags); | |
610 | } | |
611 | #else | |
612 | static inline void resched_task(struct task_struct *p) | |
613 | { | |
614 | assert_spin_locked(&task_rq(p)->lock); | |
615 | set_tsk_need_resched(p); | |
616 | } | |
617 | #endif | |
618 | ||
45bf76df IM |
619 | static u64 div64_likely32(u64 divident, unsigned long divisor) |
620 | { | |
621 | #if BITS_PER_LONG == 32 | |
622 | if (likely(divident <= 0xffffffffULL)) | |
623 | return (u32)divident / divisor; | |
624 | do_div(divident, divisor); | |
625 | ||
626 | return divident; | |
627 | #else | |
628 | return divident / divisor; | |
629 | #endif | |
630 | } | |
631 | ||
632 | #if BITS_PER_LONG == 32 | |
633 | # define WMULT_CONST (~0UL) | |
634 | #else | |
635 | # define WMULT_CONST (1UL << 32) | |
636 | #endif | |
637 | ||
638 | #define WMULT_SHIFT 32 | |
639 | ||
640 | static inline unsigned long | |
641 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, | |
642 | struct load_weight *lw) | |
643 | { | |
644 | u64 tmp; | |
645 | ||
646 | if (unlikely(!lw->inv_weight)) | |
647 | lw->inv_weight = WMULT_CONST / lw->weight; | |
648 | ||
649 | tmp = (u64)delta_exec * weight; | |
650 | /* | |
651 | * Check whether we'd overflow the 64-bit multiplication: | |
652 | */ | |
653 | if (unlikely(tmp > WMULT_CONST)) { | |
654 | tmp = ((tmp >> WMULT_SHIFT/2) * lw->inv_weight) | |
655 | >> (WMULT_SHIFT/2); | |
656 | } else { | |
657 | tmp = (tmp * lw->inv_weight) >> WMULT_SHIFT; | |
658 | } | |
659 | ||
660 | return (unsigned long)min(tmp, (u64)sysctl_sched_runtime_limit); | |
661 | } | |
662 | ||
663 | static inline unsigned long | |
664 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
665 | { | |
666 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
667 | } | |
668 | ||
669 | static void update_load_add(struct load_weight *lw, unsigned long inc) | |
670 | { | |
671 | lw->weight += inc; | |
672 | lw->inv_weight = 0; | |
673 | } | |
674 | ||
675 | static void update_load_sub(struct load_weight *lw, unsigned long dec) | |
676 | { | |
677 | lw->weight -= dec; | |
678 | lw->inv_weight = 0; | |
679 | } | |
680 | ||
681 | static void __update_curr_load(struct rq *rq, struct load_stat *ls) | |
682 | { | |
683 | if (rq->curr != rq->idle && ls->load.weight) { | |
684 | ls->delta_exec += ls->delta_stat; | |
685 | ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load); | |
686 | ls->delta_stat = 0; | |
687 | } | |
688 | } | |
689 | ||
690 | /* | |
691 | * Update delta_exec, delta_fair fields for rq. | |
692 | * | |
693 | * delta_fair clock advances at a rate inversely proportional to | |
694 | * total load (rq->ls.load.weight) on the runqueue, while | |
695 | * delta_exec advances at the same rate as wall-clock (provided | |
696 | * cpu is not idle). | |
697 | * | |
698 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
699 | * runqueue over any given interval. This (smoothened) load is used | |
700 | * during load balance. | |
701 | * | |
702 | * This function is called /before/ updating rq->ls.load | |
703 | * and when switching tasks. | |
704 | */ | |
705 | static void update_curr_load(struct rq *rq, u64 now) | |
706 | { | |
707 | struct load_stat *ls = &rq->ls; | |
708 | u64 start; | |
709 | ||
710 | start = ls->load_update_start; | |
711 | ls->load_update_start = now; | |
712 | ls->delta_stat += now - start; | |
713 | /* | |
714 | * Stagger updates to ls->delta_fair. Very frequent updates | |
715 | * can be expensive. | |
716 | */ | |
717 | if (ls->delta_stat >= sysctl_sched_stat_granularity) | |
718 | __update_curr_load(rq, ls); | |
719 | } | |
720 | ||
2dd73a4f PW |
721 | /* |
722 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
723 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
724 | * each task makes to its run queue's load is weighted according to its | |
725 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
726 | * scaled version of the new time slice allocation that they receive on time | |
727 | * slice expiry etc. | |
728 | */ | |
729 | ||
730 | /* | |
731 | * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE | |
732 | * If static_prio_timeslice() is ever changed to break this assumption then | |
733 | * this code will need modification | |
734 | */ | |
735 | #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE | |
dd41f596 | 736 | #define load_weight(lp) \ |
2dd73a4f PW |
737 | (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO) |
738 | #define PRIO_TO_LOAD_WEIGHT(prio) \ | |
dd41f596 | 739 | load_weight(static_prio_timeslice(prio)) |
2dd73a4f | 740 | #define RTPRIO_TO_LOAD_WEIGHT(rp) \ |
dd41f596 IM |
741 | (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + load_weight(rp)) |
742 | ||
743 | #define WEIGHT_IDLEPRIO 2 | |
744 | #define WMULT_IDLEPRIO (1 << 31) | |
745 | ||
746 | /* | |
747 | * Nice levels are multiplicative, with a gentle 10% change for every | |
748 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
749 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
750 | * that remained on nice 0. | |
751 | * | |
752 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
753 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
754 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
755 | * If a task goes up by ~10% and another task goes down by ~10% then | |
756 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
757 | */ |
758 | static const int prio_to_weight[40] = { | |
759 | /* -20 */ 88818, 71054, 56843, 45475, 36380, 29104, 23283, 18626, 14901, 11921, | |
760 | /* -10 */ 9537, 7629, 6103, 4883, 3906, 3125, 2500, 2000, 1600, 1280, | |
761 | /* 0 */ NICE_0_LOAD /* 1024 */, | |
762 | /* 1 */ 819, 655, 524, 419, 336, 268, 215, 172, 137, | |
763 | /* 10 */ 110, 87, 70, 56, 45, 36, 29, 23, 18, 15, | |
764 | }; | |
765 | ||
5714d2de IM |
766 | /* |
767 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
768 | * | |
769 | * In cases where the weight does not change often, we can use the | |
770 | * precalculated inverse to speed up arithmetics by turning divisions | |
771 | * into multiplications: | |
772 | */ | |
dd41f596 | 773 | static const u32 prio_to_wmult[40] = { |
e4af30be IM |
774 | /* -20 */ 48356, 60446, 75558, 94446, 118058, |
775 | /* -15 */ 147573, 184467, 230589, 288233, 360285, | |
776 | /* -10 */ 450347, 562979, 703746, 879575, 1099582, | |
777 | /* -5 */ 1374389, 1717986, 2147483, 2684354, 3355443, | |
778 | /* 0 */ 4194304, 5244160, 6557201, 8196502, 10250518, | |
779 | /* 5 */ 12782640, 16025997, 19976592, 24970740, 31350126, | |
780 | /* 10 */ 39045157, 49367440, 61356675, 76695844, 95443717, | |
781 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 782 | }; |
2dd73a4f | 783 | |
36c8b586 | 784 | static inline void |
dd41f596 | 785 | inc_load(struct rq *rq, const struct task_struct *p, u64 now) |
2dd73a4f | 786 | { |
dd41f596 IM |
787 | update_curr_load(rq, now); |
788 | update_load_add(&rq->ls.load, p->se.load.weight); | |
2dd73a4f PW |
789 | } |
790 | ||
36c8b586 | 791 | static inline void |
dd41f596 | 792 | dec_load(struct rq *rq, const struct task_struct *p, u64 now) |
2dd73a4f | 793 | { |
dd41f596 IM |
794 | update_curr_load(rq, now); |
795 | update_load_sub(&rq->ls.load, p->se.load.weight); | |
2dd73a4f PW |
796 | } |
797 | ||
dd41f596 | 798 | static inline void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now) |
2dd73a4f PW |
799 | { |
800 | rq->nr_running++; | |
dd41f596 | 801 | inc_load(rq, p, now); |
2dd73a4f PW |
802 | } |
803 | ||
dd41f596 | 804 | static inline void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now) |
2dd73a4f PW |
805 | { |
806 | rq->nr_running--; | |
dd41f596 | 807 | dec_load(rq, p, now); |
2dd73a4f PW |
808 | } |
809 | ||
dd41f596 IM |
810 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
811 | ||
812 | /* | |
813 | * runqueue iterator, to support SMP load-balancing between different | |
814 | * scheduling classes, without having to expose their internal data | |
815 | * structures to the load-balancing proper: | |
816 | */ | |
817 | struct rq_iterator { | |
818 | void *arg; | |
819 | struct task_struct *(*start)(void *); | |
820 | struct task_struct *(*next)(void *); | |
821 | }; | |
822 | ||
823 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
824 | unsigned long max_nr_move, unsigned long max_load_move, | |
825 | struct sched_domain *sd, enum cpu_idle_type idle, | |
826 | int *all_pinned, unsigned long *load_moved, | |
827 | int this_best_prio, int best_prio, int best_prio_seen, | |
828 | struct rq_iterator *iterator); | |
829 | ||
830 | #include "sched_stats.h" | |
831 | #include "sched_rt.c" | |
832 | #include "sched_fair.c" | |
833 | #include "sched_idletask.c" | |
834 | #ifdef CONFIG_SCHED_DEBUG | |
835 | # include "sched_debug.c" | |
836 | #endif | |
837 | ||
838 | #define sched_class_highest (&rt_sched_class) | |
839 | ||
45bf76df IM |
840 | static void set_load_weight(struct task_struct *p) |
841 | { | |
dd41f596 IM |
842 | task_rq(p)->cfs.wait_runtime -= p->se.wait_runtime; |
843 | p->se.wait_runtime = 0; | |
844 | ||
45bf76df | 845 | if (task_has_rt_policy(p)) { |
dd41f596 IM |
846 | p->se.load.weight = prio_to_weight[0] * 2; |
847 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
848 | return; | |
849 | } | |
45bf76df | 850 | |
dd41f596 IM |
851 | /* |
852 | * SCHED_IDLE tasks get minimal weight: | |
853 | */ | |
854 | if (p->policy == SCHED_IDLE) { | |
855 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
856 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
857 | return; | |
858 | } | |
71f8bd46 | 859 | |
dd41f596 IM |
860 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
861 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
862 | } |
863 | ||
dd41f596 IM |
864 | static void |
865 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now) | |
71f8bd46 | 866 | { |
dd41f596 IM |
867 | sched_info_queued(p); |
868 | p->sched_class->enqueue_task(rq, p, wakeup, now); | |
869 | p->se.on_rq = 1; | |
71f8bd46 IM |
870 | } |
871 | ||
dd41f596 IM |
872 | static void |
873 | dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now) | |
71f8bd46 | 874 | { |
dd41f596 IM |
875 | p->sched_class->dequeue_task(rq, p, sleep, now); |
876 | p->se.on_rq = 0; | |
71f8bd46 IM |
877 | } |
878 | ||
14531189 | 879 | /* |
dd41f596 | 880 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 881 | */ |
14531189 IM |
882 | static inline int __normal_prio(struct task_struct *p) |
883 | { | |
dd41f596 | 884 | return p->static_prio; |
14531189 IM |
885 | } |
886 | ||
b29739f9 IM |
887 | /* |
888 | * Calculate the expected normal priority: i.e. priority | |
889 | * without taking RT-inheritance into account. Might be | |
890 | * boosted by interactivity modifiers. Changes upon fork, | |
891 | * setprio syscalls, and whenever the interactivity | |
892 | * estimator recalculates. | |
893 | */ | |
36c8b586 | 894 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
895 | { |
896 | int prio; | |
897 | ||
e05606d3 | 898 | if (task_has_rt_policy(p)) |
b29739f9 IM |
899 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
900 | else | |
901 | prio = __normal_prio(p); | |
902 | return prio; | |
903 | } | |
904 | ||
905 | /* | |
906 | * Calculate the current priority, i.e. the priority | |
907 | * taken into account by the scheduler. This value might | |
908 | * be boosted by RT tasks, or might be boosted by | |
909 | * interactivity modifiers. Will be RT if the task got | |
910 | * RT-boosted. If not then it returns p->normal_prio. | |
911 | */ | |
36c8b586 | 912 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
913 | { |
914 | p->normal_prio = normal_prio(p); | |
915 | /* | |
916 | * If we are RT tasks or we were boosted to RT priority, | |
917 | * keep the priority unchanged. Otherwise, update priority | |
918 | * to the normal priority: | |
919 | */ | |
920 | if (!rt_prio(p->prio)) | |
921 | return p->normal_prio; | |
922 | return p->prio; | |
923 | } | |
924 | ||
1da177e4 | 925 | /* |
dd41f596 | 926 | * activate_task - move a task to the runqueue. |
1da177e4 | 927 | */ |
dd41f596 | 928 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 929 | { |
dd41f596 | 930 | u64 now = rq_clock(rq); |
d425b274 | 931 | |
dd41f596 IM |
932 | if (p->state == TASK_UNINTERRUPTIBLE) |
933 | rq->nr_uninterruptible--; | |
1da177e4 | 934 | |
dd41f596 IM |
935 | enqueue_task(rq, p, wakeup, now); |
936 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
937 | } |
938 | ||
939 | /* | |
dd41f596 | 940 | * activate_idle_task - move idle task to the _front_ of runqueue. |
1da177e4 | 941 | */ |
dd41f596 | 942 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 943 | { |
dd41f596 | 944 | u64 now = rq_clock(rq); |
1da177e4 | 945 | |
dd41f596 IM |
946 | if (p->state == TASK_UNINTERRUPTIBLE) |
947 | rq->nr_uninterruptible--; | |
ece8a684 | 948 | |
dd41f596 IM |
949 | enqueue_task(rq, p, 0, now); |
950 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
951 | } |
952 | ||
953 | /* | |
954 | * deactivate_task - remove a task from the runqueue. | |
955 | */ | |
dd41f596 | 956 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 957 | { |
dd41f596 IM |
958 | u64 now = rq_clock(rq); |
959 | ||
960 | if (p->state == TASK_UNINTERRUPTIBLE) | |
961 | rq->nr_uninterruptible++; | |
962 | ||
963 | dequeue_task(rq, p, sleep, now); | |
964 | dec_nr_running(p, rq, now); | |
1da177e4 LT |
965 | } |
966 | ||
1da177e4 LT |
967 | /** |
968 | * task_curr - is this task currently executing on a CPU? | |
969 | * @p: the task in question. | |
970 | */ | |
36c8b586 | 971 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
972 | { |
973 | return cpu_curr(task_cpu(p)) == p; | |
974 | } | |
975 | ||
2dd73a4f PW |
976 | /* Used instead of source_load when we know the type == 0 */ |
977 | unsigned long weighted_cpuload(const int cpu) | |
978 | { | |
dd41f596 IM |
979 | return cpu_rq(cpu)->ls.load.weight; |
980 | } | |
981 | ||
982 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
983 | { | |
984 | #ifdef CONFIG_SMP | |
985 | task_thread_info(p)->cpu = cpu; | |
986 | set_task_cfs_rq(p); | |
987 | #endif | |
2dd73a4f PW |
988 | } |
989 | ||
1da177e4 | 990 | #ifdef CONFIG_SMP |
c65cc870 | 991 | |
dd41f596 | 992 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 993 | { |
dd41f596 IM |
994 | int old_cpu = task_cpu(p); |
995 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
996 | u64 clock_offset, fair_clock_offset; | |
997 | ||
998 | clock_offset = old_rq->clock - new_rq->clock; | |
999 | fair_clock_offset = old_rq->cfs.fair_clock - | |
1000 | new_rq->cfs.fair_clock; | |
1001 | if (p->se.wait_start) | |
1002 | p->se.wait_start -= clock_offset; | |
1003 | if (p->se.wait_start_fair) | |
1004 | p->se.wait_start_fair -= fair_clock_offset; | |
1005 | if (p->se.sleep_start) | |
1006 | p->se.sleep_start -= clock_offset; | |
1007 | if (p->se.block_start) | |
1008 | p->se.block_start -= clock_offset; | |
1009 | if (p->se.sleep_start_fair) | |
1010 | p->se.sleep_start_fair -= fair_clock_offset; | |
1011 | ||
1012 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1013 | } |
1014 | ||
70b97a7f | 1015 | struct migration_req { |
1da177e4 | 1016 | struct list_head list; |
1da177e4 | 1017 | |
36c8b586 | 1018 | struct task_struct *task; |
1da177e4 LT |
1019 | int dest_cpu; |
1020 | ||
1da177e4 | 1021 | struct completion done; |
70b97a7f | 1022 | }; |
1da177e4 LT |
1023 | |
1024 | /* | |
1025 | * The task's runqueue lock must be held. | |
1026 | * Returns true if you have to wait for migration thread. | |
1027 | */ | |
36c8b586 | 1028 | static int |
70b97a7f | 1029 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1030 | { |
70b97a7f | 1031 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1032 | |
1033 | /* | |
1034 | * If the task is not on a runqueue (and not running), then | |
1035 | * it is sufficient to simply update the task's cpu field. | |
1036 | */ | |
dd41f596 | 1037 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1038 | set_task_cpu(p, dest_cpu); |
1039 | return 0; | |
1040 | } | |
1041 | ||
1042 | init_completion(&req->done); | |
1da177e4 LT |
1043 | req->task = p; |
1044 | req->dest_cpu = dest_cpu; | |
1045 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1046 | |
1da177e4 LT |
1047 | return 1; |
1048 | } | |
1049 | ||
1050 | /* | |
1051 | * wait_task_inactive - wait for a thread to unschedule. | |
1052 | * | |
1053 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1054 | * else this function might spin for a *long* time. This function can't | |
1055 | * be called with interrupts off, or it may introduce deadlock with | |
1056 | * smp_call_function() if an IPI is sent by the same process we are | |
1057 | * waiting to become inactive. | |
1058 | */ | |
36c8b586 | 1059 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1060 | { |
1061 | unsigned long flags; | |
dd41f596 | 1062 | int running, on_rq; |
70b97a7f | 1063 | struct rq *rq; |
1da177e4 LT |
1064 | |
1065 | repeat: | |
fa490cfd LT |
1066 | /* |
1067 | * We do the initial early heuristics without holding | |
1068 | * any task-queue locks at all. We'll only try to get | |
1069 | * the runqueue lock when things look like they will | |
1070 | * work out! | |
1071 | */ | |
1072 | rq = task_rq(p); | |
1073 | ||
1074 | /* | |
1075 | * If the task is actively running on another CPU | |
1076 | * still, just relax and busy-wait without holding | |
1077 | * any locks. | |
1078 | * | |
1079 | * NOTE! Since we don't hold any locks, it's not | |
1080 | * even sure that "rq" stays as the right runqueue! | |
1081 | * But we don't care, since "task_running()" will | |
1082 | * return false if the runqueue has changed and p | |
1083 | * is actually now running somewhere else! | |
1084 | */ | |
1085 | while (task_running(rq, p)) | |
1086 | cpu_relax(); | |
1087 | ||
1088 | /* | |
1089 | * Ok, time to look more closely! We need the rq | |
1090 | * lock now, to be *sure*. If we're wrong, we'll | |
1091 | * just go back and repeat. | |
1092 | */ | |
1da177e4 | 1093 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1094 | running = task_running(rq, p); |
dd41f596 | 1095 | on_rq = p->se.on_rq; |
fa490cfd LT |
1096 | task_rq_unlock(rq, &flags); |
1097 | ||
1098 | /* | |
1099 | * Was it really running after all now that we | |
1100 | * checked with the proper locks actually held? | |
1101 | * | |
1102 | * Oops. Go back and try again.. | |
1103 | */ | |
1104 | if (unlikely(running)) { | |
1da177e4 | 1105 | cpu_relax(); |
1da177e4 LT |
1106 | goto repeat; |
1107 | } | |
fa490cfd LT |
1108 | |
1109 | /* | |
1110 | * It's not enough that it's not actively running, | |
1111 | * it must be off the runqueue _entirely_, and not | |
1112 | * preempted! | |
1113 | * | |
1114 | * So if it wa still runnable (but just not actively | |
1115 | * running right now), it's preempted, and we should | |
1116 | * yield - it could be a while. | |
1117 | */ | |
dd41f596 | 1118 | if (unlikely(on_rq)) { |
fa490cfd LT |
1119 | yield(); |
1120 | goto repeat; | |
1121 | } | |
1122 | ||
1123 | /* | |
1124 | * Ahh, all good. It wasn't running, and it wasn't | |
1125 | * runnable, which means that it will never become | |
1126 | * running in the future either. We're all done! | |
1127 | */ | |
1da177e4 LT |
1128 | } |
1129 | ||
1130 | /*** | |
1131 | * kick_process - kick a running thread to enter/exit the kernel | |
1132 | * @p: the to-be-kicked thread | |
1133 | * | |
1134 | * Cause a process which is running on another CPU to enter | |
1135 | * kernel-mode, without any delay. (to get signals handled.) | |
1136 | * | |
1137 | * NOTE: this function doesnt have to take the runqueue lock, | |
1138 | * because all it wants to ensure is that the remote task enters | |
1139 | * the kernel. If the IPI races and the task has been migrated | |
1140 | * to another CPU then no harm is done and the purpose has been | |
1141 | * achieved as well. | |
1142 | */ | |
36c8b586 | 1143 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1144 | { |
1145 | int cpu; | |
1146 | ||
1147 | preempt_disable(); | |
1148 | cpu = task_cpu(p); | |
1149 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1150 | smp_send_reschedule(cpu); | |
1151 | preempt_enable(); | |
1152 | } | |
1153 | ||
1154 | /* | |
2dd73a4f PW |
1155 | * Return a low guess at the load of a migration-source cpu weighted |
1156 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1157 | * |
1158 | * We want to under-estimate the load of migration sources, to | |
1159 | * balance conservatively. | |
1160 | */ | |
a2000572 | 1161 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1162 | { |
70b97a7f | 1163 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1164 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1165 | |
3b0bd9bc | 1166 | if (type == 0) |
dd41f596 | 1167 | return total; |
b910472d | 1168 | |
dd41f596 | 1169 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1170 | } |
1171 | ||
1172 | /* | |
2dd73a4f PW |
1173 | * Return a high guess at the load of a migration-target cpu weighted |
1174 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1175 | */ |
a2000572 | 1176 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1177 | { |
70b97a7f | 1178 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1179 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1180 | |
7897986b | 1181 | if (type == 0) |
dd41f596 | 1182 | return total; |
3b0bd9bc | 1183 | |
dd41f596 | 1184 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1185 | } |
1186 | ||
1187 | /* | |
1188 | * Return the average load per task on the cpu's run queue | |
1189 | */ | |
1190 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1191 | { | |
70b97a7f | 1192 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1193 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1194 | unsigned long n = rq->nr_running; |
1195 | ||
dd41f596 | 1196 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1197 | } |
1198 | ||
147cbb4b NP |
1199 | /* |
1200 | * find_idlest_group finds and returns the least busy CPU group within the | |
1201 | * domain. | |
1202 | */ | |
1203 | static struct sched_group * | |
1204 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1205 | { | |
1206 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1207 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1208 | int load_idx = sd->forkexec_idx; | |
1209 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1210 | ||
1211 | do { | |
1212 | unsigned long load, avg_load; | |
1213 | int local_group; | |
1214 | int i; | |
1215 | ||
da5a5522 BD |
1216 | /* Skip over this group if it has no CPUs allowed */ |
1217 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1218 | goto nextgroup; | |
1219 | ||
147cbb4b | 1220 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1221 | |
1222 | /* Tally up the load of all CPUs in the group */ | |
1223 | avg_load = 0; | |
1224 | ||
1225 | for_each_cpu_mask(i, group->cpumask) { | |
1226 | /* Bias balancing toward cpus of our domain */ | |
1227 | if (local_group) | |
1228 | load = source_load(i, load_idx); | |
1229 | else | |
1230 | load = target_load(i, load_idx); | |
1231 | ||
1232 | avg_load += load; | |
1233 | } | |
1234 | ||
1235 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1236 | avg_load = sg_div_cpu_power(group, |
1237 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1238 | |
1239 | if (local_group) { | |
1240 | this_load = avg_load; | |
1241 | this = group; | |
1242 | } else if (avg_load < min_load) { | |
1243 | min_load = avg_load; | |
1244 | idlest = group; | |
1245 | } | |
da5a5522 | 1246 | nextgroup: |
147cbb4b NP |
1247 | group = group->next; |
1248 | } while (group != sd->groups); | |
1249 | ||
1250 | if (!idlest || 100*this_load < imbalance*min_load) | |
1251 | return NULL; | |
1252 | return idlest; | |
1253 | } | |
1254 | ||
1255 | /* | |
0feaece9 | 1256 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1257 | */ |
95cdf3b7 IM |
1258 | static int |
1259 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1260 | { |
da5a5522 | 1261 | cpumask_t tmp; |
147cbb4b NP |
1262 | unsigned long load, min_load = ULONG_MAX; |
1263 | int idlest = -1; | |
1264 | int i; | |
1265 | ||
da5a5522 BD |
1266 | /* Traverse only the allowed CPUs */ |
1267 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1268 | ||
1269 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1270 | load = weighted_cpuload(i); |
147cbb4b NP |
1271 | |
1272 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1273 | min_load = load; | |
1274 | idlest = i; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | return idlest; | |
1279 | } | |
1280 | ||
476d139c NP |
1281 | /* |
1282 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1283 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1284 | * SD_BALANCE_EXEC. | |
1285 | * | |
1286 | * Balance, ie. select the least loaded group. | |
1287 | * | |
1288 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1289 | * | |
1290 | * preempt must be disabled. | |
1291 | */ | |
1292 | static int sched_balance_self(int cpu, int flag) | |
1293 | { | |
1294 | struct task_struct *t = current; | |
1295 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1296 | |
c96d145e | 1297 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1298 | /* |
1299 | * If power savings logic is enabled for a domain, stop there. | |
1300 | */ | |
5c45bf27 SS |
1301 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1302 | break; | |
476d139c NP |
1303 | if (tmp->flags & flag) |
1304 | sd = tmp; | |
c96d145e | 1305 | } |
476d139c NP |
1306 | |
1307 | while (sd) { | |
1308 | cpumask_t span; | |
1309 | struct sched_group *group; | |
1a848870 SS |
1310 | int new_cpu, weight; |
1311 | ||
1312 | if (!(sd->flags & flag)) { | |
1313 | sd = sd->child; | |
1314 | continue; | |
1315 | } | |
476d139c NP |
1316 | |
1317 | span = sd->span; | |
1318 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1319 | if (!group) { |
1320 | sd = sd->child; | |
1321 | continue; | |
1322 | } | |
476d139c | 1323 | |
da5a5522 | 1324 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1325 | if (new_cpu == -1 || new_cpu == cpu) { |
1326 | /* Now try balancing at a lower domain level of cpu */ | |
1327 | sd = sd->child; | |
1328 | continue; | |
1329 | } | |
476d139c | 1330 | |
1a848870 | 1331 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1332 | cpu = new_cpu; |
476d139c NP |
1333 | sd = NULL; |
1334 | weight = cpus_weight(span); | |
1335 | for_each_domain(cpu, tmp) { | |
1336 | if (weight <= cpus_weight(tmp->span)) | |
1337 | break; | |
1338 | if (tmp->flags & flag) | |
1339 | sd = tmp; | |
1340 | } | |
1341 | /* while loop will break here if sd == NULL */ | |
1342 | } | |
1343 | ||
1344 | return cpu; | |
1345 | } | |
1346 | ||
1347 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1348 | |
1349 | /* | |
1350 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1351 | * not idle and an idle cpu is available. The span of cpus to | |
1352 | * search starts with cpus closest then further out as needed, | |
1353 | * so we always favor a closer, idle cpu. | |
1354 | * | |
1355 | * Returns the CPU we should wake onto. | |
1356 | */ | |
1357 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1358 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1359 | { |
1360 | cpumask_t tmp; | |
1361 | struct sched_domain *sd; | |
1362 | int i; | |
1363 | ||
4953198b SS |
1364 | /* |
1365 | * If it is idle, then it is the best cpu to run this task. | |
1366 | * | |
1367 | * This cpu is also the best, if it has more than one task already. | |
1368 | * Siblings must be also busy(in most cases) as they didn't already | |
1369 | * pickup the extra load from this cpu and hence we need not check | |
1370 | * sibling runqueue info. This will avoid the checks and cache miss | |
1371 | * penalities associated with that. | |
1372 | */ | |
1373 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1374 | return cpu; |
1375 | ||
1376 | for_each_domain(cpu, sd) { | |
1377 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1378 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1379 | for_each_cpu_mask(i, tmp) { |
1380 | if (idle_cpu(i)) | |
1381 | return i; | |
1382 | } | |
9761eea8 | 1383 | } else { |
e0f364f4 | 1384 | break; |
9761eea8 | 1385 | } |
1da177e4 LT |
1386 | } |
1387 | return cpu; | |
1388 | } | |
1389 | #else | |
36c8b586 | 1390 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1391 | { |
1392 | return cpu; | |
1393 | } | |
1394 | #endif | |
1395 | ||
1396 | /*** | |
1397 | * try_to_wake_up - wake up a thread | |
1398 | * @p: the to-be-woken-up thread | |
1399 | * @state: the mask of task states that can be woken | |
1400 | * @sync: do a synchronous wakeup? | |
1401 | * | |
1402 | * Put it on the run-queue if it's not already there. The "current" | |
1403 | * thread is always on the run-queue (except when the actual | |
1404 | * re-schedule is in progress), and as such you're allowed to do | |
1405 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1406 | * runnable without the overhead of this. | |
1407 | * | |
1408 | * returns failure only if the task is already active. | |
1409 | */ | |
36c8b586 | 1410 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1411 | { |
1412 | int cpu, this_cpu, success = 0; | |
1413 | unsigned long flags; | |
1414 | long old_state; | |
70b97a7f | 1415 | struct rq *rq; |
1da177e4 | 1416 | #ifdef CONFIG_SMP |
7897986b | 1417 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1418 | unsigned long load, this_load; |
1da177e4 LT |
1419 | int new_cpu; |
1420 | #endif | |
1421 | ||
1422 | rq = task_rq_lock(p, &flags); | |
1423 | old_state = p->state; | |
1424 | if (!(old_state & state)) | |
1425 | goto out; | |
1426 | ||
dd41f596 | 1427 | if (p->se.on_rq) |
1da177e4 LT |
1428 | goto out_running; |
1429 | ||
1430 | cpu = task_cpu(p); | |
1431 | this_cpu = smp_processor_id(); | |
1432 | ||
1433 | #ifdef CONFIG_SMP | |
1434 | if (unlikely(task_running(rq, p))) | |
1435 | goto out_activate; | |
1436 | ||
7897986b NP |
1437 | new_cpu = cpu; |
1438 | ||
1da177e4 LT |
1439 | schedstat_inc(rq, ttwu_cnt); |
1440 | if (cpu == this_cpu) { | |
1441 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1442 | goto out_set_cpu; |
1443 | } | |
1444 | ||
1445 | for_each_domain(this_cpu, sd) { | |
1446 | if (cpu_isset(cpu, sd->span)) { | |
1447 | schedstat_inc(sd, ttwu_wake_remote); | |
1448 | this_sd = sd; | |
1449 | break; | |
1da177e4 LT |
1450 | } |
1451 | } | |
1da177e4 | 1452 | |
7897986b | 1453 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1454 | goto out_set_cpu; |
1455 | ||
1da177e4 | 1456 | /* |
7897986b | 1457 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1458 | */ |
7897986b NP |
1459 | if (this_sd) { |
1460 | int idx = this_sd->wake_idx; | |
1461 | unsigned int imbalance; | |
1da177e4 | 1462 | |
a3f21bce NP |
1463 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1464 | ||
7897986b NP |
1465 | load = source_load(cpu, idx); |
1466 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1467 | |
7897986b NP |
1468 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1469 | ||
a3f21bce NP |
1470 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1471 | unsigned long tl = this_load; | |
33859f7f MOS |
1472 | unsigned long tl_per_task; |
1473 | ||
1474 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1475 | |
1da177e4 | 1476 | /* |
a3f21bce NP |
1477 | * If sync wakeup then subtract the (maximum possible) |
1478 | * effect of the currently running task from the load | |
1479 | * of the current CPU: | |
1da177e4 | 1480 | */ |
a3f21bce | 1481 | if (sync) |
dd41f596 | 1482 | tl -= current->se.load.weight; |
a3f21bce NP |
1483 | |
1484 | if ((tl <= load && | |
2dd73a4f | 1485 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1486 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1487 | /* |
1488 | * This domain has SD_WAKE_AFFINE and | |
1489 | * p is cache cold in this domain, and | |
1490 | * there is no bad imbalance. | |
1491 | */ | |
1492 | schedstat_inc(this_sd, ttwu_move_affine); | |
1493 | goto out_set_cpu; | |
1494 | } | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * Start passive balancing when half the imbalance_pct | |
1499 | * limit is reached. | |
1500 | */ | |
1501 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1502 | if (imbalance*this_load <= 100*load) { | |
1503 | schedstat_inc(this_sd, ttwu_move_balance); | |
1504 | goto out_set_cpu; | |
1505 | } | |
1da177e4 LT |
1506 | } |
1507 | } | |
1508 | ||
1509 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1510 | out_set_cpu: | |
1511 | new_cpu = wake_idle(new_cpu, p); | |
1512 | if (new_cpu != cpu) { | |
1513 | set_task_cpu(p, new_cpu); | |
1514 | task_rq_unlock(rq, &flags); | |
1515 | /* might preempt at this point */ | |
1516 | rq = task_rq_lock(p, &flags); | |
1517 | old_state = p->state; | |
1518 | if (!(old_state & state)) | |
1519 | goto out; | |
dd41f596 | 1520 | if (p->se.on_rq) |
1da177e4 LT |
1521 | goto out_running; |
1522 | ||
1523 | this_cpu = smp_processor_id(); | |
1524 | cpu = task_cpu(p); | |
1525 | } | |
1526 | ||
1527 | out_activate: | |
1528 | #endif /* CONFIG_SMP */ | |
dd41f596 | 1529 | activate_task(rq, p, 1); |
1da177e4 LT |
1530 | /* |
1531 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1532 | * has indicated that it will leave the CPU in short order) | |
1533 | * don't trigger a preemption, if the woken up task will run on | |
1534 | * this cpu. (in this case the 'I will reschedule' promise of | |
1535 | * the waker guarantees that the freshly woken up task is going | |
1536 | * to be considered on this CPU.) | |
1537 | */ | |
dd41f596 IM |
1538 | if (!sync || cpu != this_cpu) |
1539 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1540 | success = 1; |
1541 | ||
1542 | out_running: | |
1543 | p->state = TASK_RUNNING; | |
1544 | out: | |
1545 | task_rq_unlock(rq, &flags); | |
1546 | ||
1547 | return success; | |
1548 | } | |
1549 | ||
36c8b586 | 1550 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1551 | { |
1552 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1553 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1554 | } | |
1da177e4 LT |
1555 | EXPORT_SYMBOL(wake_up_process); |
1556 | ||
36c8b586 | 1557 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1558 | { |
1559 | return try_to_wake_up(p, state, 0); | |
1560 | } | |
1561 | ||
1da177e4 LT |
1562 | /* |
1563 | * Perform scheduler related setup for a newly forked process p. | |
1564 | * p is forked by current. | |
dd41f596 IM |
1565 | * |
1566 | * __sched_fork() is basic setup used by init_idle() too: | |
1567 | */ | |
1568 | static void __sched_fork(struct task_struct *p) | |
1569 | { | |
1570 | p->se.wait_start_fair = 0; | |
1571 | p->se.wait_start = 0; | |
1572 | p->se.exec_start = 0; | |
1573 | p->se.sum_exec_runtime = 0; | |
1574 | p->se.delta_exec = 0; | |
1575 | p->se.delta_fair_run = 0; | |
1576 | p->se.delta_fair_sleep = 0; | |
1577 | p->se.wait_runtime = 0; | |
1578 | p->se.sum_wait_runtime = 0; | |
1579 | p->se.sum_sleep_runtime = 0; | |
1580 | p->se.sleep_start = 0; | |
1581 | p->se.sleep_start_fair = 0; | |
1582 | p->se.block_start = 0; | |
1583 | p->se.sleep_max = 0; | |
1584 | p->se.block_max = 0; | |
1585 | p->se.exec_max = 0; | |
1586 | p->se.wait_max = 0; | |
1587 | p->se.wait_runtime_overruns = 0; | |
1588 | p->se.wait_runtime_underruns = 0; | |
476d139c | 1589 | |
dd41f596 IM |
1590 | INIT_LIST_HEAD(&p->run_list); |
1591 | p->se.on_rq = 0; | |
476d139c | 1592 | |
e107be36 AK |
1593 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1594 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1595 | #endif | |
1596 | ||
1da177e4 LT |
1597 | /* |
1598 | * We mark the process as running here, but have not actually | |
1599 | * inserted it onto the runqueue yet. This guarantees that | |
1600 | * nobody will actually run it, and a signal or other external | |
1601 | * event cannot wake it up and insert it on the runqueue either. | |
1602 | */ | |
1603 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1604 | } |
1605 | ||
1606 | /* | |
1607 | * fork()/clone()-time setup: | |
1608 | */ | |
1609 | void sched_fork(struct task_struct *p, int clone_flags) | |
1610 | { | |
1611 | int cpu = get_cpu(); | |
1612 | ||
1613 | __sched_fork(p); | |
1614 | ||
1615 | #ifdef CONFIG_SMP | |
1616 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1617 | #endif | |
1618 | __set_task_cpu(p, cpu); | |
b29739f9 IM |
1619 | |
1620 | /* | |
1621 | * Make sure we do not leak PI boosting priority to the child: | |
1622 | */ | |
1623 | p->prio = current->normal_prio; | |
1624 | ||
52f17b6c | 1625 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1626 | if (likely(sched_info_on())) |
52f17b6c | 1627 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1628 | #endif |
d6077cb8 | 1629 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1630 | p->oncpu = 0; |
1631 | #endif | |
1da177e4 | 1632 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1633 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1634 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1635 | #endif |
476d139c | 1636 | put_cpu(); |
1da177e4 LT |
1637 | } |
1638 | ||
dd41f596 IM |
1639 | /* |
1640 | * After fork, child runs first. (default) If set to 0 then | |
1641 | * parent will (try to) run first. | |
1642 | */ | |
1643 | unsigned int __read_mostly sysctl_sched_child_runs_first = 1; | |
1644 | ||
1da177e4 LT |
1645 | /* |
1646 | * wake_up_new_task - wake up a newly created task for the first time. | |
1647 | * | |
1648 | * This function will do some initial scheduler statistics housekeeping | |
1649 | * that must be done for every newly created context, then puts the task | |
1650 | * on the runqueue and wakes it. | |
1651 | */ | |
36c8b586 | 1652 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1653 | { |
1654 | unsigned long flags; | |
dd41f596 IM |
1655 | struct rq *rq; |
1656 | int this_cpu; | |
1da177e4 LT |
1657 | |
1658 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1659 | BUG_ON(p->state != TASK_RUNNING); |
dd41f596 | 1660 | this_cpu = smp_processor_id(); /* parent's CPU */ |
1da177e4 LT |
1661 | |
1662 | p->prio = effective_prio(p); | |
1663 | ||
dd41f596 IM |
1664 | if (!sysctl_sched_child_runs_first || (clone_flags & CLONE_VM) || |
1665 | task_cpu(p) != this_cpu || !current->se.on_rq) { | |
1666 | activate_task(rq, p, 0); | |
1da177e4 | 1667 | } else { |
1da177e4 | 1668 | /* |
dd41f596 IM |
1669 | * Let the scheduling class do new task startup |
1670 | * management (if any): | |
1da177e4 | 1671 | */ |
dd41f596 | 1672 | p->sched_class->task_new(rq, p); |
1da177e4 | 1673 | } |
dd41f596 IM |
1674 | check_preempt_curr(rq, p); |
1675 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1676 | } |
1677 | ||
e107be36 AK |
1678 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1679 | ||
1680 | /** | |
1681 | * preempt_notifier_register - tell me when current is being being preempted | |
1682 | * and rescheduled | |
1683 | */ | |
1684 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1685 | { | |
1686 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1687 | } | |
1688 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1689 | ||
1690 | /** | |
1691 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
1692 | * | |
1693 | * This is safe to call from within a preemption notifier. | |
1694 | */ | |
1695 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1696 | { | |
1697 | hlist_del(¬ifier->link); | |
1698 | } | |
1699 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1700 | ||
1701 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1702 | { | |
1703 | struct preempt_notifier *notifier; | |
1704 | struct hlist_node *node; | |
1705 | ||
1706 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1707 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1708 | } | |
1709 | ||
1710 | static void | |
1711 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1712 | struct task_struct *next) | |
1713 | { | |
1714 | struct preempt_notifier *notifier; | |
1715 | struct hlist_node *node; | |
1716 | ||
1717 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1718 | notifier->ops->sched_out(notifier, next); | |
1719 | } | |
1720 | ||
1721 | #else | |
1722 | ||
1723 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1724 | { | |
1725 | } | |
1726 | ||
1727 | static void | |
1728 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1729 | struct task_struct *next) | |
1730 | { | |
1731 | } | |
1732 | ||
1733 | #endif | |
1734 | ||
4866cde0 NP |
1735 | /** |
1736 | * prepare_task_switch - prepare to switch tasks | |
1737 | * @rq: the runqueue preparing to switch | |
1738 | * @next: the task we are going to switch to. | |
1739 | * | |
1740 | * This is called with the rq lock held and interrupts off. It must | |
1741 | * be paired with a subsequent finish_task_switch after the context | |
1742 | * switch. | |
1743 | * | |
1744 | * prepare_task_switch sets up locking and calls architecture specific | |
1745 | * hooks. | |
1746 | */ | |
e107be36 AK |
1747 | static inline void |
1748 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1749 | struct task_struct *next) | |
4866cde0 | 1750 | { |
e107be36 | 1751 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1752 | prepare_lock_switch(rq, next); |
1753 | prepare_arch_switch(next); | |
1754 | } | |
1755 | ||
1da177e4 LT |
1756 | /** |
1757 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1758 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1759 | * @prev: the thread we just switched away from. |
1760 | * | |
4866cde0 NP |
1761 | * finish_task_switch must be called after the context switch, paired |
1762 | * with a prepare_task_switch call before the context switch. | |
1763 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1764 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1765 | * |
1766 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1767 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1768 | * with the lock held can cause deadlocks; see schedule() for | |
1769 | * details.) | |
1770 | */ | |
70b97a7f | 1771 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1772 | __releases(rq->lock) |
1773 | { | |
1da177e4 | 1774 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1775 | long prev_state; |
1da177e4 LT |
1776 | |
1777 | rq->prev_mm = NULL; | |
1778 | ||
1779 | /* | |
1780 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1781 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1782 | * schedule one last time. The schedule call will never return, and |
1783 | * the scheduled task must drop that reference. | |
c394cc9f | 1784 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1785 | * still held, otherwise prev could be scheduled on another cpu, die |
1786 | * there before we look at prev->state, and then the reference would | |
1787 | * be dropped twice. | |
1788 | * Manfred Spraul <manfred@colorfullife.com> | |
1789 | */ | |
55a101f8 | 1790 | prev_state = prev->state; |
4866cde0 NP |
1791 | finish_arch_switch(prev); |
1792 | finish_lock_switch(rq, prev); | |
e107be36 | 1793 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1794 | if (mm) |
1795 | mmdrop(mm); | |
c394cc9f | 1796 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1797 | /* |
1798 | * Remove function-return probe instances associated with this | |
1799 | * task and put them back on the free list. | |
9761eea8 | 1800 | */ |
c6fd91f0 | 1801 | kprobe_flush_task(prev); |
1da177e4 | 1802 | put_task_struct(prev); |
c6fd91f0 | 1803 | } |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | /** | |
1807 | * schedule_tail - first thing a freshly forked thread must call. | |
1808 | * @prev: the thread we just switched away from. | |
1809 | */ | |
36c8b586 | 1810 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1811 | __releases(rq->lock) |
1812 | { | |
70b97a7f IM |
1813 | struct rq *rq = this_rq(); |
1814 | ||
4866cde0 NP |
1815 | finish_task_switch(rq, prev); |
1816 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1817 | /* In this case, finish_task_switch does not reenable preemption */ | |
1818 | preempt_enable(); | |
1819 | #endif | |
1da177e4 LT |
1820 | if (current->set_child_tid) |
1821 | put_user(current->pid, current->set_child_tid); | |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * context_switch - switch to the new MM and the new | |
1826 | * thread's register state. | |
1827 | */ | |
dd41f596 | 1828 | static inline void |
70b97a7f | 1829 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1830 | struct task_struct *next) |
1da177e4 | 1831 | { |
dd41f596 | 1832 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1833 | |
e107be36 | 1834 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1835 | mm = next->mm; |
1836 | oldmm = prev->active_mm; | |
9226d125 ZA |
1837 | /* |
1838 | * For paravirt, this is coupled with an exit in switch_to to | |
1839 | * combine the page table reload and the switch backend into | |
1840 | * one hypercall. | |
1841 | */ | |
1842 | arch_enter_lazy_cpu_mode(); | |
1843 | ||
dd41f596 | 1844 | if (unlikely(!mm)) { |
1da177e4 LT |
1845 | next->active_mm = oldmm; |
1846 | atomic_inc(&oldmm->mm_count); | |
1847 | enter_lazy_tlb(oldmm, next); | |
1848 | } else | |
1849 | switch_mm(oldmm, mm, next); | |
1850 | ||
dd41f596 | 1851 | if (unlikely(!prev->mm)) { |
1da177e4 | 1852 | prev->active_mm = NULL; |
1da177e4 LT |
1853 | rq->prev_mm = oldmm; |
1854 | } | |
3a5f5e48 IM |
1855 | /* |
1856 | * Since the runqueue lock will be released by the next | |
1857 | * task (which is an invalid locking op but in the case | |
1858 | * of the scheduler it's an obvious special-case), so we | |
1859 | * do an early lockdep release here: | |
1860 | */ | |
1861 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1862 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1863 | #endif |
1da177e4 LT |
1864 | |
1865 | /* Here we just switch the register state and the stack. */ | |
1866 | switch_to(prev, next, prev); | |
1867 | ||
dd41f596 IM |
1868 | barrier(); |
1869 | /* | |
1870 | * this_rq must be evaluated again because prev may have moved | |
1871 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1872 | * frame will be invalid. | |
1873 | */ | |
1874 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1875 | } |
1876 | ||
1877 | /* | |
1878 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1879 | * | |
1880 | * externally visible scheduler statistics: current number of runnable | |
1881 | * threads, current number of uninterruptible-sleeping threads, total | |
1882 | * number of context switches performed since bootup. | |
1883 | */ | |
1884 | unsigned long nr_running(void) | |
1885 | { | |
1886 | unsigned long i, sum = 0; | |
1887 | ||
1888 | for_each_online_cpu(i) | |
1889 | sum += cpu_rq(i)->nr_running; | |
1890 | ||
1891 | return sum; | |
1892 | } | |
1893 | ||
1894 | unsigned long nr_uninterruptible(void) | |
1895 | { | |
1896 | unsigned long i, sum = 0; | |
1897 | ||
0a945022 | 1898 | for_each_possible_cpu(i) |
1da177e4 LT |
1899 | sum += cpu_rq(i)->nr_uninterruptible; |
1900 | ||
1901 | /* | |
1902 | * Since we read the counters lockless, it might be slightly | |
1903 | * inaccurate. Do not allow it to go below zero though: | |
1904 | */ | |
1905 | if (unlikely((long)sum < 0)) | |
1906 | sum = 0; | |
1907 | ||
1908 | return sum; | |
1909 | } | |
1910 | ||
1911 | unsigned long long nr_context_switches(void) | |
1912 | { | |
cc94abfc SR |
1913 | int i; |
1914 | unsigned long long sum = 0; | |
1da177e4 | 1915 | |
0a945022 | 1916 | for_each_possible_cpu(i) |
1da177e4 LT |
1917 | sum += cpu_rq(i)->nr_switches; |
1918 | ||
1919 | return sum; | |
1920 | } | |
1921 | ||
1922 | unsigned long nr_iowait(void) | |
1923 | { | |
1924 | unsigned long i, sum = 0; | |
1925 | ||
0a945022 | 1926 | for_each_possible_cpu(i) |
1da177e4 LT |
1927 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1928 | ||
1929 | return sum; | |
1930 | } | |
1931 | ||
db1b1fef JS |
1932 | unsigned long nr_active(void) |
1933 | { | |
1934 | unsigned long i, running = 0, uninterruptible = 0; | |
1935 | ||
1936 | for_each_online_cpu(i) { | |
1937 | running += cpu_rq(i)->nr_running; | |
1938 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1939 | } | |
1940 | ||
1941 | if (unlikely((long)uninterruptible < 0)) | |
1942 | uninterruptible = 0; | |
1943 | ||
1944 | return running + uninterruptible; | |
1945 | } | |
1946 | ||
48f24c4d | 1947 | /* |
dd41f596 IM |
1948 | * Update rq->cpu_load[] statistics. This function is usually called every |
1949 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1950 | */ |
dd41f596 | 1951 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1952 | { |
dd41f596 IM |
1953 | u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64; |
1954 | unsigned long total_load = this_rq->ls.load.weight; | |
1955 | unsigned long this_load = total_load; | |
1956 | struct load_stat *ls = &this_rq->ls; | |
1957 | u64 now = __rq_clock(this_rq); | |
1958 | int i, scale; | |
1959 | ||
1960 | this_rq->nr_load_updates++; | |
1961 | if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD))) | |
1962 | goto do_avg; | |
1963 | ||
1964 | /* Update delta_fair/delta_exec fields first */ | |
1965 | update_curr_load(this_rq, now); | |
1966 | ||
1967 | fair_delta64 = ls->delta_fair + 1; | |
1968 | ls->delta_fair = 0; | |
1969 | ||
1970 | exec_delta64 = ls->delta_exec + 1; | |
1971 | ls->delta_exec = 0; | |
1972 | ||
1973 | sample_interval64 = now - ls->load_update_last; | |
1974 | ls->load_update_last = now; | |
1975 | ||
1976 | if ((s64)sample_interval64 < (s64)TICK_NSEC) | |
1977 | sample_interval64 = TICK_NSEC; | |
1978 | ||
1979 | if (exec_delta64 > sample_interval64) | |
1980 | exec_delta64 = sample_interval64; | |
1981 | ||
1982 | idle_delta64 = sample_interval64 - exec_delta64; | |
1983 | ||
1984 | tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64); | |
1985 | tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64); | |
1986 | ||
1987 | this_load = (unsigned long)tmp64; | |
1988 | ||
1989 | do_avg: | |
1990 | ||
1991 | /* Update our load: */ | |
1992 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1993 | unsigned long old_load, new_load; | |
1994 | ||
1995 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1996 | ||
1997 | old_load = this_rq->cpu_load[i]; | |
1998 | new_load = this_load; | |
1999 | ||
2000 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; | |
2001 | } | |
48f24c4d IM |
2002 | } |
2003 | ||
dd41f596 IM |
2004 | #ifdef CONFIG_SMP |
2005 | ||
1da177e4 LT |
2006 | /* |
2007 | * double_rq_lock - safely lock two runqueues | |
2008 | * | |
2009 | * Note this does not disable interrupts like task_rq_lock, | |
2010 | * you need to do so manually before calling. | |
2011 | */ | |
70b97a7f | 2012 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2013 | __acquires(rq1->lock) |
2014 | __acquires(rq2->lock) | |
2015 | { | |
054b9108 | 2016 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2017 | if (rq1 == rq2) { |
2018 | spin_lock(&rq1->lock); | |
2019 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2020 | } else { | |
c96d145e | 2021 | if (rq1 < rq2) { |
1da177e4 LT |
2022 | spin_lock(&rq1->lock); |
2023 | spin_lock(&rq2->lock); | |
2024 | } else { | |
2025 | spin_lock(&rq2->lock); | |
2026 | spin_lock(&rq1->lock); | |
2027 | } | |
2028 | } | |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * double_rq_unlock - safely unlock two runqueues | |
2033 | * | |
2034 | * Note this does not restore interrupts like task_rq_unlock, | |
2035 | * you need to do so manually after calling. | |
2036 | */ | |
70b97a7f | 2037 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2038 | __releases(rq1->lock) |
2039 | __releases(rq2->lock) | |
2040 | { | |
2041 | spin_unlock(&rq1->lock); | |
2042 | if (rq1 != rq2) | |
2043 | spin_unlock(&rq2->lock); | |
2044 | else | |
2045 | __release(rq2->lock); | |
2046 | } | |
2047 | ||
2048 | /* | |
2049 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2050 | */ | |
70b97a7f | 2051 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2052 | __releases(this_rq->lock) |
2053 | __acquires(busiest->lock) | |
2054 | __acquires(this_rq->lock) | |
2055 | { | |
054b9108 KK |
2056 | if (unlikely(!irqs_disabled())) { |
2057 | /* printk() doesn't work good under rq->lock */ | |
2058 | spin_unlock(&this_rq->lock); | |
2059 | BUG_ON(1); | |
2060 | } | |
1da177e4 | 2061 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2062 | if (busiest < this_rq) { |
1da177e4 LT |
2063 | spin_unlock(&this_rq->lock); |
2064 | spin_lock(&busiest->lock); | |
2065 | spin_lock(&this_rq->lock); | |
2066 | } else | |
2067 | spin_lock(&busiest->lock); | |
2068 | } | |
2069 | } | |
2070 | ||
1da177e4 LT |
2071 | /* |
2072 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2073 | * This is accomplished by forcing the cpu_allowed mask to only | |
2074 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2075 | * the cpu_allowed mask is restored. | |
2076 | */ | |
36c8b586 | 2077 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2078 | { |
70b97a7f | 2079 | struct migration_req req; |
1da177e4 | 2080 | unsigned long flags; |
70b97a7f | 2081 | struct rq *rq; |
1da177e4 LT |
2082 | |
2083 | rq = task_rq_lock(p, &flags); | |
2084 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2085 | || unlikely(cpu_is_offline(dest_cpu))) | |
2086 | goto out; | |
2087 | ||
2088 | /* force the process onto the specified CPU */ | |
2089 | if (migrate_task(p, dest_cpu, &req)) { | |
2090 | /* Need to wait for migration thread (might exit: take ref). */ | |
2091 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2092 | |
1da177e4 LT |
2093 | get_task_struct(mt); |
2094 | task_rq_unlock(rq, &flags); | |
2095 | wake_up_process(mt); | |
2096 | put_task_struct(mt); | |
2097 | wait_for_completion(&req.done); | |
36c8b586 | 2098 | |
1da177e4 LT |
2099 | return; |
2100 | } | |
2101 | out: | |
2102 | task_rq_unlock(rq, &flags); | |
2103 | } | |
2104 | ||
2105 | /* | |
476d139c NP |
2106 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2107 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2108 | */ |
2109 | void sched_exec(void) | |
2110 | { | |
1da177e4 | 2111 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2112 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2113 | put_cpu(); |
476d139c NP |
2114 | if (new_cpu != this_cpu) |
2115 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2120 | * Both runqueues must be locked. | |
2121 | */ | |
dd41f596 IM |
2122 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2123 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2124 | { |
dd41f596 | 2125 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2126 | set_task_cpu(p, this_cpu); |
dd41f596 | 2127 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2128 | /* |
2129 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2130 | * to be always true for them. | |
2131 | */ | |
dd41f596 | 2132 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2133 | } |
2134 | ||
2135 | /* | |
2136 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2137 | */ | |
858119e1 | 2138 | static |
70b97a7f | 2139 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2140 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2141 | int *all_pinned) |
1da177e4 LT |
2142 | { |
2143 | /* | |
2144 | * We do not migrate tasks that are: | |
2145 | * 1) running (obviously), or | |
2146 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2147 | * 3) are cache-hot on their current CPU. | |
2148 | */ | |
1da177e4 LT |
2149 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2150 | return 0; | |
81026794 NP |
2151 | *all_pinned = 0; |
2152 | ||
2153 | if (task_running(rq, p)) | |
2154 | return 0; | |
1da177e4 LT |
2155 | |
2156 | /* | |
dd41f596 | 2157 | * Aggressive migration if too many balance attempts have failed: |
1da177e4 | 2158 | */ |
dd41f596 | 2159 | if (sd->nr_balance_failed > sd->cache_nice_tries) |
1da177e4 LT |
2160 | return 1; |
2161 | ||
1da177e4 LT |
2162 | return 1; |
2163 | } | |
2164 | ||
dd41f596 | 2165 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2166 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2167 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 IM |
2168 | int *all_pinned, unsigned long *load_moved, |
2169 | int this_best_prio, int best_prio, int best_prio_seen, | |
2170 | struct rq_iterator *iterator) | |
1da177e4 | 2171 | { |
dd41f596 IM |
2172 | int pulled = 0, pinned = 0, skip_for_load; |
2173 | struct task_struct *p; | |
2174 | long rem_load_move = max_load_move; | |
1da177e4 | 2175 | |
2dd73a4f | 2176 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2177 | goto out; |
2178 | ||
81026794 NP |
2179 | pinned = 1; |
2180 | ||
1da177e4 | 2181 | /* |
dd41f596 | 2182 | * Start the load-balancing iterator: |
1da177e4 | 2183 | */ |
dd41f596 IM |
2184 | p = iterator->start(iterator->arg); |
2185 | next: | |
2186 | if (!p) | |
1da177e4 | 2187 | goto out; |
50ddd969 PW |
2188 | /* |
2189 | * To help distribute high priority tasks accross CPUs we don't | |
2190 | * skip a task if it will be the highest priority task (i.e. smallest | |
2191 | * prio value) on its new queue regardless of its load weight | |
2192 | */ | |
dd41f596 IM |
2193 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2194 | SCHED_LOAD_SCALE_FUZZ; | |
2195 | if (skip_for_load && p->prio < this_best_prio) | |
2196 | skip_for_load = !best_prio_seen && p->prio == best_prio; | |
615052dc | 2197 | if (skip_for_load || |
dd41f596 | 2198 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d | 2199 | |
dd41f596 IM |
2200 | best_prio_seen |= p->prio == best_prio; |
2201 | p = iterator->next(iterator->arg); | |
2202 | goto next; | |
1da177e4 LT |
2203 | } |
2204 | ||
dd41f596 | 2205 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2206 | pulled++; |
dd41f596 | 2207 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2208 | |
2dd73a4f PW |
2209 | /* |
2210 | * We only want to steal up to the prescribed number of tasks | |
2211 | * and the prescribed amount of weighted load. | |
2212 | */ | |
2213 | if (pulled < max_nr_move && rem_load_move > 0) { | |
dd41f596 IM |
2214 | if (p->prio < this_best_prio) |
2215 | this_best_prio = p->prio; | |
2216 | p = iterator->next(iterator->arg); | |
2217 | goto next; | |
1da177e4 LT |
2218 | } |
2219 | out: | |
2220 | /* | |
2221 | * Right now, this is the only place pull_task() is called, | |
2222 | * so we can safely collect pull_task() stats here rather than | |
2223 | * inside pull_task(). | |
2224 | */ | |
2225 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2226 | |
2227 | if (all_pinned) | |
2228 | *all_pinned = pinned; | |
dd41f596 | 2229 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2230 | return pulled; |
2231 | } | |
2232 | ||
dd41f596 IM |
2233 | /* |
2234 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted | |
2235 | * load from busiest to this_rq, as part of a balancing operation within | |
2236 | * "domain". Returns the number of tasks moved. | |
2237 | * | |
2238 | * Called with both runqueues locked. | |
2239 | */ | |
2240 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2241 | unsigned long max_nr_move, unsigned long max_load_move, | |
2242 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2243 | int *all_pinned) | |
2244 | { | |
2245 | struct sched_class *class = sched_class_highest; | |
2246 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
2247 | long rem_load_move = max_load_move; | |
2248 | ||
2249 | do { | |
2250 | nr_moved = class->load_balance(this_rq, this_cpu, busiest, | |
2251 | max_nr_move, (unsigned long)rem_load_move, | |
2252 | sd, idle, all_pinned, &load_moved); | |
2253 | total_nr_moved += nr_moved; | |
2254 | max_nr_move -= nr_moved; | |
2255 | rem_load_move -= load_moved; | |
2256 | class = class->next; | |
2257 | } while (class && max_nr_move && rem_load_move > 0); | |
2258 | ||
2259 | return total_nr_moved; | |
2260 | } | |
2261 | ||
1da177e4 LT |
2262 | /* |
2263 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2264 | * domain. It calculates and returns the amount of weighted load which |
2265 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2266 | */ |
2267 | static struct sched_group * | |
2268 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2269 | unsigned long *imbalance, enum cpu_idle_type idle, |
2270 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2271 | { |
2272 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2273 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2274 | unsigned long max_pull; |
2dd73a4f PW |
2275 | unsigned long busiest_load_per_task, busiest_nr_running; |
2276 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2277 | int load_idx; |
5c45bf27 SS |
2278 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2279 | int power_savings_balance = 1; | |
2280 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2281 | unsigned long min_nr_running = ULONG_MAX; | |
2282 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2283 | #endif | |
1da177e4 LT |
2284 | |
2285 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2286 | busiest_load_per_task = busiest_nr_running = 0; |
2287 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2288 | if (idle == CPU_NOT_IDLE) |
7897986b | 2289 | load_idx = sd->busy_idx; |
d15bcfdb | 2290 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2291 | load_idx = sd->newidle_idx; |
2292 | else | |
2293 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2294 | |
2295 | do { | |
5c45bf27 | 2296 | unsigned long load, group_capacity; |
1da177e4 LT |
2297 | int local_group; |
2298 | int i; | |
783609c6 | 2299 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2300 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2301 | |
2302 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2303 | ||
783609c6 SS |
2304 | if (local_group) |
2305 | balance_cpu = first_cpu(group->cpumask); | |
2306 | ||
1da177e4 | 2307 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2308 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2309 | |
2310 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2311 | struct rq *rq; |
2312 | ||
2313 | if (!cpu_isset(i, *cpus)) | |
2314 | continue; | |
2315 | ||
2316 | rq = cpu_rq(i); | |
2dd73a4f | 2317 | |
9439aab8 | 2318 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2319 | *sd_idle = 0; |
2320 | ||
1da177e4 | 2321 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2322 | if (local_group) { |
2323 | if (idle_cpu(i) && !first_idle_cpu) { | |
2324 | first_idle_cpu = 1; | |
2325 | balance_cpu = i; | |
2326 | } | |
2327 | ||
a2000572 | 2328 | load = target_load(i, load_idx); |
783609c6 | 2329 | } else |
a2000572 | 2330 | load = source_load(i, load_idx); |
1da177e4 LT |
2331 | |
2332 | avg_load += load; | |
2dd73a4f | 2333 | sum_nr_running += rq->nr_running; |
dd41f596 | 2334 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2335 | } |
2336 | ||
783609c6 SS |
2337 | /* |
2338 | * First idle cpu or the first cpu(busiest) in this sched group | |
2339 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2340 | * domains. In the newly idle case, we will allow all the cpu's |
2341 | * to do the newly idle load balance. | |
783609c6 | 2342 | */ |
9439aab8 SS |
2343 | if (idle != CPU_NEWLY_IDLE && local_group && |
2344 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2345 | *balance = 0; |
2346 | goto ret; | |
2347 | } | |
2348 | ||
1da177e4 | 2349 | total_load += avg_load; |
5517d86b | 2350 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2351 | |
2352 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2353 | avg_load = sg_div_cpu_power(group, |
2354 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2355 | |
5517d86b | 2356 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2357 | |
1da177e4 LT |
2358 | if (local_group) { |
2359 | this_load = avg_load; | |
2360 | this = group; | |
2dd73a4f PW |
2361 | this_nr_running = sum_nr_running; |
2362 | this_load_per_task = sum_weighted_load; | |
2363 | } else if (avg_load > max_load && | |
5c45bf27 | 2364 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2365 | max_load = avg_load; |
2366 | busiest = group; | |
2dd73a4f PW |
2367 | busiest_nr_running = sum_nr_running; |
2368 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2369 | } |
5c45bf27 SS |
2370 | |
2371 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2372 | /* | |
2373 | * Busy processors will not participate in power savings | |
2374 | * balance. | |
2375 | */ | |
dd41f596 IM |
2376 | if (idle == CPU_NOT_IDLE || |
2377 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2378 | goto group_next; | |
5c45bf27 SS |
2379 | |
2380 | /* | |
2381 | * If the local group is idle or completely loaded | |
2382 | * no need to do power savings balance at this domain | |
2383 | */ | |
2384 | if (local_group && (this_nr_running >= group_capacity || | |
2385 | !this_nr_running)) | |
2386 | power_savings_balance = 0; | |
2387 | ||
dd41f596 | 2388 | /* |
5c45bf27 SS |
2389 | * If a group is already running at full capacity or idle, |
2390 | * don't include that group in power savings calculations | |
dd41f596 IM |
2391 | */ |
2392 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2393 | || !sum_nr_running) |
dd41f596 | 2394 | goto group_next; |
5c45bf27 | 2395 | |
dd41f596 | 2396 | /* |
5c45bf27 | 2397 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2398 | * This is the group from where we need to pick up the load |
2399 | * for saving power | |
2400 | */ | |
2401 | if ((sum_nr_running < min_nr_running) || | |
2402 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2403 | first_cpu(group->cpumask) < |
2404 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2405 | group_min = group; |
2406 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2407 | min_load_per_task = sum_weighted_load / |
2408 | sum_nr_running; | |
dd41f596 | 2409 | } |
5c45bf27 | 2410 | |
dd41f596 | 2411 | /* |
5c45bf27 | 2412 | * Calculate the group which is almost near its |
dd41f596 IM |
2413 | * capacity but still has some space to pick up some load |
2414 | * from other group and save more power | |
2415 | */ | |
2416 | if (sum_nr_running <= group_capacity - 1) { | |
2417 | if (sum_nr_running > leader_nr_running || | |
2418 | (sum_nr_running == leader_nr_running && | |
2419 | first_cpu(group->cpumask) > | |
2420 | first_cpu(group_leader->cpumask))) { | |
2421 | group_leader = group; | |
2422 | leader_nr_running = sum_nr_running; | |
2423 | } | |
48f24c4d | 2424 | } |
5c45bf27 SS |
2425 | group_next: |
2426 | #endif | |
1da177e4 LT |
2427 | group = group->next; |
2428 | } while (group != sd->groups); | |
2429 | ||
2dd73a4f | 2430 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2431 | goto out_balanced; |
2432 | ||
2433 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2434 | ||
2435 | if (this_load >= avg_load || | |
2436 | 100*max_load <= sd->imbalance_pct*this_load) | |
2437 | goto out_balanced; | |
2438 | ||
2dd73a4f | 2439 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2440 | /* |
2441 | * We're trying to get all the cpus to the average_load, so we don't | |
2442 | * want to push ourselves above the average load, nor do we wish to | |
2443 | * reduce the max loaded cpu below the average load, as either of these | |
2444 | * actions would just result in more rebalancing later, and ping-pong | |
2445 | * tasks around. Thus we look for the minimum possible imbalance. | |
2446 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2447 | * be counted as no imbalance for these purposes -- we can't fix that | |
2448 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2449 | * appear as very large values with unsigned longs. | |
2450 | */ | |
2dd73a4f PW |
2451 | if (max_load <= busiest_load_per_task) |
2452 | goto out_balanced; | |
2453 | ||
2454 | /* | |
2455 | * In the presence of smp nice balancing, certain scenarios can have | |
2456 | * max load less than avg load(as we skip the groups at or below | |
2457 | * its cpu_power, while calculating max_load..) | |
2458 | */ | |
2459 | if (max_load < avg_load) { | |
2460 | *imbalance = 0; | |
2461 | goto small_imbalance; | |
2462 | } | |
0c117f1b SS |
2463 | |
2464 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2465 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2466 | |
1da177e4 | 2467 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2468 | *imbalance = min(max_pull * busiest->__cpu_power, |
2469 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2470 | / SCHED_LOAD_SCALE; |
2471 | ||
2dd73a4f PW |
2472 | /* |
2473 | * if *imbalance is less than the average load per runnable task | |
2474 | * there is no gaurantee that any tasks will be moved so we'll have | |
2475 | * a think about bumping its value to force at least one task to be | |
2476 | * moved | |
2477 | */ | |
dd41f596 | 2478 | if (*imbalance + SCHED_LOAD_SCALE_FUZZ < busiest_load_per_task/2) { |
48f24c4d | 2479 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2480 | unsigned int imbn; |
2481 | ||
2482 | small_imbalance: | |
2483 | pwr_move = pwr_now = 0; | |
2484 | imbn = 2; | |
2485 | if (this_nr_running) { | |
2486 | this_load_per_task /= this_nr_running; | |
2487 | if (busiest_load_per_task > this_load_per_task) | |
2488 | imbn = 1; | |
2489 | } else | |
2490 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2491 | |
dd41f596 IM |
2492 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2493 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2494 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2495 | return busiest; |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * OK, we don't have enough imbalance to justify moving tasks, | |
2500 | * however we may be able to increase total CPU power used by | |
2501 | * moving them. | |
2502 | */ | |
2503 | ||
5517d86b ED |
2504 | pwr_now += busiest->__cpu_power * |
2505 | min(busiest_load_per_task, max_load); | |
2506 | pwr_now += this->__cpu_power * | |
2507 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2508 | pwr_now /= SCHED_LOAD_SCALE; |
2509 | ||
2510 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2511 | tmp = sg_div_cpu_power(busiest, |
2512 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2513 | if (max_load > tmp) |
5517d86b | 2514 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2515 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2516 | |
2517 | /* Amount of load we'd add */ | |
5517d86b | 2518 | if (max_load * busiest->__cpu_power < |
33859f7f | 2519 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2520 | tmp = sg_div_cpu_power(this, |
2521 | max_load * busiest->__cpu_power); | |
1da177e4 | 2522 | else |
5517d86b ED |
2523 | tmp = sg_div_cpu_power(this, |
2524 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2525 | pwr_move += this->__cpu_power * | |
2526 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2527 | pwr_move /= SCHED_LOAD_SCALE; |
2528 | ||
2529 | /* Move if we gain throughput */ | |
2530 | if (pwr_move <= pwr_now) | |
2531 | goto out_balanced; | |
2532 | ||
2dd73a4f | 2533 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2534 | } |
2535 | ||
1da177e4 LT |
2536 | return busiest; |
2537 | ||
2538 | out_balanced: | |
5c45bf27 | 2539 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2540 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2541 | goto ret; |
1da177e4 | 2542 | |
5c45bf27 SS |
2543 | if (this == group_leader && group_leader != group_min) { |
2544 | *imbalance = min_load_per_task; | |
2545 | return group_min; | |
2546 | } | |
5c45bf27 | 2547 | #endif |
783609c6 | 2548 | ret: |
1da177e4 LT |
2549 | *imbalance = 0; |
2550 | return NULL; | |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2555 | */ | |
70b97a7f | 2556 | static struct rq * |
d15bcfdb | 2557 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2558 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2559 | { |
70b97a7f | 2560 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2561 | unsigned long max_load = 0; |
1da177e4 LT |
2562 | int i; |
2563 | ||
2564 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2565 | unsigned long wl; |
0a2966b4 CL |
2566 | |
2567 | if (!cpu_isset(i, *cpus)) | |
2568 | continue; | |
2569 | ||
48f24c4d | 2570 | rq = cpu_rq(i); |
dd41f596 | 2571 | wl = weighted_cpuload(i); |
2dd73a4f | 2572 | |
dd41f596 | 2573 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2574 | continue; |
1da177e4 | 2575 | |
dd41f596 IM |
2576 | if (wl > max_load) { |
2577 | max_load = wl; | |
48f24c4d | 2578 | busiest = rq; |
1da177e4 LT |
2579 | } |
2580 | } | |
2581 | ||
2582 | return busiest; | |
2583 | } | |
2584 | ||
77391d71 NP |
2585 | /* |
2586 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2587 | * so long as it is large enough. | |
2588 | */ | |
2589 | #define MAX_PINNED_INTERVAL 512 | |
2590 | ||
48f24c4d IM |
2591 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2592 | { | |
2593 | return n > 0 ? n - 1 : 0; | |
2594 | } | |
2595 | ||
1da177e4 LT |
2596 | /* |
2597 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2598 | * tasks if there is an imbalance. | |
1da177e4 | 2599 | */ |
70b97a7f | 2600 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2601 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2602 | int *balance) |
1da177e4 | 2603 | { |
48f24c4d | 2604 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2605 | struct sched_group *group; |
1da177e4 | 2606 | unsigned long imbalance; |
70b97a7f | 2607 | struct rq *busiest; |
0a2966b4 | 2608 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2609 | unsigned long flags; |
5969fe06 | 2610 | |
89c4710e SS |
2611 | /* |
2612 | * When power savings policy is enabled for the parent domain, idle | |
2613 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2614 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2615 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2616 | */ |
d15bcfdb | 2617 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2618 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2619 | sd_idle = 1; |
1da177e4 | 2620 | |
1da177e4 LT |
2621 | schedstat_inc(sd, lb_cnt[idle]); |
2622 | ||
0a2966b4 CL |
2623 | redo: |
2624 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2625 | &cpus, balance); |
2626 | ||
06066714 | 2627 | if (*balance == 0) |
783609c6 | 2628 | goto out_balanced; |
783609c6 | 2629 | |
1da177e4 LT |
2630 | if (!group) { |
2631 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2632 | goto out_balanced; | |
2633 | } | |
2634 | ||
0a2966b4 | 2635 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2636 | if (!busiest) { |
2637 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2638 | goto out_balanced; | |
2639 | } | |
2640 | ||
db935dbd | 2641 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2642 | |
2643 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2644 | ||
2645 | nr_moved = 0; | |
2646 | if (busiest->nr_running > 1) { | |
2647 | /* | |
2648 | * Attempt to move tasks. If find_busiest_group has found | |
2649 | * an imbalance but busiest->nr_running <= 1, the group is | |
2650 | * still unbalanced. nr_moved simply stays zero, so it is | |
2651 | * correctly treated as an imbalance. | |
2652 | */ | |
fe2eea3f | 2653 | local_irq_save(flags); |
e17224bf | 2654 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2655 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2656 | minus_1_or_zero(busiest->nr_running), |
2657 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2658 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2659 | local_irq_restore(flags); |
81026794 | 2660 | |
46cb4b7c SS |
2661 | /* |
2662 | * some other cpu did the load balance for us. | |
2663 | */ | |
2664 | if (nr_moved && this_cpu != smp_processor_id()) | |
2665 | resched_cpu(this_cpu); | |
2666 | ||
81026794 | 2667 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2668 | if (unlikely(all_pinned)) { |
2669 | cpu_clear(cpu_of(busiest), cpus); | |
2670 | if (!cpus_empty(cpus)) | |
2671 | goto redo; | |
81026794 | 2672 | goto out_balanced; |
0a2966b4 | 2673 | } |
1da177e4 | 2674 | } |
81026794 | 2675 | |
1da177e4 LT |
2676 | if (!nr_moved) { |
2677 | schedstat_inc(sd, lb_failed[idle]); | |
2678 | sd->nr_balance_failed++; | |
2679 | ||
2680 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2681 | |
fe2eea3f | 2682 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2683 | |
2684 | /* don't kick the migration_thread, if the curr | |
2685 | * task on busiest cpu can't be moved to this_cpu | |
2686 | */ | |
2687 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2688 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2689 | all_pinned = 1; |
2690 | goto out_one_pinned; | |
2691 | } | |
2692 | ||
1da177e4 LT |
2693 | if (!busiest->active_balance) { |
2694 | busiest->active_balance = 1; | |
2695 | busiest->push_cpu = this_cpu; | |
81026794 | 2696 | active_balance = 1; |
1da177e4 | 2697 | } |
fe2eea3f | 2698 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2699 | if (active_balance) |
1da177e4 LT |
2700 | wake_up_process(busiest->migration_thread); |
2701 | ||
2702 | /* | |
2703 | * We've kicked active balancing, reset the failure | |
2704 | * counter. | |
2705 | */ | |
39507451 | 2706 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2707 | } |
81026794 | 2708 | } else |
1da177e4 LT |
2709 | sd->nr_balance_failed = 0; |
2710 | ||
81026794 | 2711 | if (likely(!active_balance)) { |
1da177e4 LT |
2712 | /* We were unbalanced, so reset the balancing interval */ |
2713 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2714 | } else { |
2715 | /* | |
2716 | * If we've begun active balancing, start to back off. This | |
2717 | * case may not be covered by the all_pinned logic if there | |
2718 | * is only 1 task on the busy runqueue (because we don't call | |
2719 | * move_tasks). | |
2720 | */ | |
2721 | if (sd->balance_interval < sd->max_interval) | |
2722 | sd->balance_interval *= 2; | |
1da177e4 LT |
2723 | } |
2724 | ||
5c45bf27 | 2725 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2726 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2727 | return -1; |
1da177e4 LT |
2728 | return nr_moved; |
2729 | ||
2730 | out_balanced: | |
1da177e4 LT |
2731 | schedstat_inc(sd, lb_balanced[idle]); |
2732 | ||
16cfb1c0 | 2733 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2734 | |
2735 | out_one_pinned: | |
1da177e4 | 2736 | /* tune up the balancing interval */ |
77391d71 NP |
2737 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2738 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2739 | sd->balance_interval *= 2; |
2740 | ||
48f24c4d | 2741 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2742 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2743 | return -1; |
1da177e4 LT |
2744 | return 0; |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2749 | * tasks if there is an imbalance. | |
2750 | * | |
d15bcfdb | 2751 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2752 | * this_rq is locked. |
2753 | */ | |
48f24c4d | 2754 | static int |
70b97a7f | 2755 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2756 | { |
2757 | struct sched_group *group; | |
70b97a7f | 2758 | struct rq *busiest = NULL; |
1da177e4 LT |
2759 | unsigned long imbalance; |
2760 | int nr_moved = 0; | |
5969fe06 | 2761 | int sd_idle = 0; |
969bb4e4 | 2762 | int all_pinned = 0; |
0a2966b4 | 2763 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2764 | |
89c4710e SS |
2765 | /* |
2766 | * When power savings policy is enabled for the parent domain, idle | |
2767 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2768 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2769 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2770 | */ |
2771 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2772 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2773 | sd_idle = 1; |
1da177e4 | 2774 | |
d15bcfdb | 2775 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2776 | redo: |
d15bcfdb | 2777 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2778 | &sd_idle, &cpus, NULL); |
1da177e4 | 2779 | if (!group) { |
d15bcfdb | 2780 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2781 | goto out_balanced; |
1da177e4 LT |
2782 | } |
2783 | ||
d15bcfdb | 2784 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2785 | &cpus); |
db935dbd | 2786 | if (!busiest) { |
d15bcfdb | 2787 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2788 | goto out_balanced; |
1da177e4 LT |
2789 | } |
2790 | ||
db935dbd NP |
2791 | BUG_ON(busiest == this_rq); |
2792 | ||
d15bcfdb | 2793 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2794 | |
2795 | nr_moved = 0; | |
2796 | if (busiest->nr_running > 1) { | |
2797 | /* Attempt to move tasks */ | |
2798 | double_lock_balance(this_rq, busiest); | |
2799 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2800 | minus_1_or_zero(busiest->nr_running), |
969bb4e4 SS |
2801 | imbalance, sd, CPU_NEWLY_IDLE, |
2802 | &all_pinned); | |
d6d5cfaf | 2803 | spin_unlock(&busiest->lock); |
0a2966b4 | 2804 | |
969bb4e4 | 2805 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2806 | cpu_clear(cpu_of(busiest), cpus); |
2807 | if (!cpus_empty(cpus)) | |
2808 | goto redo; | |
2809 | } | |
d6d5cfaf NP |
2810 | } |
2811 | ||
5969fe06 | 2812 | if (!nr_moved) { |
d15bcfdb | 2813 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2814 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2815 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2816 | return -1; |
2817 | } else | |
16cfb1c0 | 2818 | sd->nr_balance_failed = 0; |
1da177e4 | 2819 | |
1da177e4 | 2820 | return nr_moved; |
16cfb1c0 NP |
2821 | |
2822 | out_balanced: | |
d15bcfdb | 2823 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2824 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2825 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2826 | return -1; |
16cfb1c0 | 2827 | sd->nr_balance_failed = 0; |
48f24c4d | 2828 | |
16cfb1c0 | 2829 | return 0; |
1da177e4 LT |
2830 | } |
2831 | ||
2832 | /* | |
2833 | * idle_balance is called by schedule() if this_cpu is about to become | |
2834 | * idle. Attempts to pull tasks from other CPUs. | |
2835 | */ | |
70b97a7f | 2836 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2837 | { |
2838 | struct sched_domain *sd; | |
dd41f596 IM |
2839 | int pulled_task = -1; |
2840 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2841 | |
2842 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2843 | unsigned long interval; |
2844 | ||
2845 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2846 | continue; | |
2847 | ||
2848 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2849 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2850 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2851 | this_rq, sd); |
2852 | ||
2853 | interval = msecs_to_jiffies(sd->balance_interval); | |
2854 | if (time_after(next_balance, sd->last_balance + interval)) | |
2855 | next_balance = sd->last_balance + interval; | |
2856 | if (pulled_task) | |
2857 | break; | |
1da177e4 | 2858 | } |
dd41f596 | 2859 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2860 | /* |
2861 | * We are going idle. next_balance may be set based on | |
2862 | * a busy processor. So reset next_balance. | |
2863 | */ | |
2864 | this_rq->next_balance = next_balance; | |
dd41f596 | 2865 | } |
1da177e4 LT |
2866 | } |
2867 | ||
2868 | /* | |
2869 | * active_load_balance is run by migration threads. It pushes running tasks | |
2870 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2871 | * running on each physical CPU where possible, and avoids physical / | |
2872 | * logical imbalances. | |
2873 | * | |
2874 | * Called with busiest_rq locked. | |
2875 | */ | |
70b97a7f | 2876 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2877 | { |
39507451 | 2878 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2879 | struct sched_domain *sd; |
2880 | struct rq *target_rq; | |
39507451 | 2881 | |
48f24c4d | 2882 | /* Is there any task to move? */ |
39507451 | 2883 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2884 | return; |
2885 | ||
2886 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2887 | |
2888 | /* | |
39507451 NP |
2889 | * This condition is "impossible", if it occurs |
2890 | * we need to fix it. Originally reported by | |
2891 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2892 | */ |
39507451 | 2893 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2894 | |
39507451 NP |
2895 | /* move a task from busiest_rq to target_rq */ |
2896 | double_lock_balance(busiest_rq, target_rq); | |
2897 | ||
2898 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2899 | for_each_domain(target_cpu, sd) { |
39507451 | 2900 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2901 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2902 | break; |
c96d145e | 2903 | } |
39507451 | 2904 | |
48f24c4d IM |
2905 | if (likely(sd)) { |
2906 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2907 | |
48f24c4d | 2908 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
d15bcfdb | 2909 | RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE, |
48f24c4d IM |
2910 | NULL)) |
2911 | schedstat_inc(sd, alb_pushed); | |
2912 | else | |
2913 | schedstat_inc(sd, alb_failed); | |
2914 | } | |
39507451 | 2915 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2916 | } |
2917 | ||
46cb4b7c SS |
2918 | #ifdef CONFIG_NO_HZ |
2919 | static struct { | |
2920 | atomic_t load_balancer; | |
2921 | cpumask_t cpu_mask; | |
2922 | } nohz ____cacheline_aligned = { | |
2923 | .load_balancer = ATOMIC_INIT(-1), | |
2924 | .cpu_mask = CPU_MASK_NONE, | |
2925 | }; | |
2926 | ||
7835b98b | 2927 | /* |
46cb4b7c SS |
2928 | * This routine will try to nominate the ilb (idle load balancing) |
2929 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2930 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2931 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2932 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2933 | * arrives... | |
2934 | * | |
2935 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2936 | * for idle load balancing. ilb owner will still be part of | |
2937 | * nohz.cpu_mask.. | |
7835b98b | 2938 | * |
46cb4b7c SS |
2939 | * While stopping the tick, this cpu will become the ilb owner if there |
2940 | * is no other owner. And will be the owner till that cpu becomes busy | |
2941 | * or if all cpus in the system stop their ticks at which point | |
2942 | * there is no need for ilb owner. | |
2943 | * | |
2944 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2945 | * next busy scheduler_tick() | |
2946 | */ | |
2947 | int select_nohz_load_balancer(int stop_tick) | |
2948 | { | |
2949 | int cpu = smp_processor_id(); | |
2950 | ||
2951 | if (stop_tick) { | |
2952 | cpu_set(cpu, nohz.cpu_mask); | |
2953 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2954 | ||
2955 | /* | |
2956 | * If we are going offline and still the leader, give up! | |
2957 | */ | |
2958 | if (cpu_is_offline(cpu) && | |
2959 | atomic_read(&nohz.load_balancer) == cpu) { | |
2960 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2961 | BUG(); | |
2962 | return 0; | |
2963 | } | |
2964 | ||
2965 | /* time for ilb owner also to sleep */ | |
2966 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2967 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2968 | atomic_set(&nohz.load_balancer, -1); | |
2969 | return 0; | |
2970 | } | |
2971 | ||
2972 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2973 | /* make me the ilb owner */ | |
2974 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2975 | return 1; | |
2976 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2977 | return 1; | |
2978 | } else { | |
2979 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2980 | return 0; | |
2981 | ||
2982 | cpu_clear(cpu, nohz.cpu_mask); | |
2983 | ||
2984 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2985 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2986 | BUG(); | |
2987 | } | |
2988 | return 0; | |
2989 | } | |
2990 | #endif | |
2991 | ||
2992 | static DEFINE_SPINLOCK(balancing); | |
2993 | ||
2994 | /* | |
7835b98b CL |
2995 | * It checks each scheduling domain to see if it is due to be balanced, |
2996 | * and initiates a balancing operation if so. | |
2997 | * | |
2998 | * Balancing parameters are set up in arch_init_sched_domains. | |
2999 | */ | |
d15bcfdb | 3000 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3001 | { |
46cb4b7c SS |
3002 | int balance = 1; |
3003 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3004 | unsigned long interval; |
3005 | struct sched_domain *sd; | |
46cb4b7c | 3006 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3007 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 3008 | |
46cb4b7c | 3009 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3010 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3011 | continue; | |
3012 | ||
3013 | interval = sd->balance_interval; | |
d15bcfdb | 3014 | if (idle != CPU_IDLE) |
1da177e4 LT |
3015 | interval *= sd->busy_factor; |
3016 | ||
3017 | /* scale ms to jiffies */ | |
3018 | interval = msecs_to_jiffies(interval); | |
3019 | if (unlikely(!interval)) | |
3020 | interval = 1; | |
dd41f596 IM |
3021 | if (interval > HZ*NR_CPUS/10) |
3022 | interval = HZ*NR_CPUS/10; | |
3023 | ||
1da177e4 | 3024 | |
08c183f3 CL |
3025 | if (sd->flags & SD_SERIALIZE) { |
3026 | if (!spin_trylock(&balancing)) | |
3027 | goto out; | |
3028 | } | |
3029 | ||
c9819f45 | 3030 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3031 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3032 | /* |
3033 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3034 | * longer idle, or one of our SMT siblings is |
3035 | * not idle. | |
3036 | */ | |
d15bcfdb | 3037 | idle = CPU_NOT_IDLE; |
1da177e4 | 3038 | } |
1bd77f2d | 3039 | sd->last_balance = jiffies; |
1da177e4 | 3040 | } |
08c183f3 CL |
3041 | if (sd->flags & SD_SERIALIZE) |
3042 | spin_unlock(&balancing); | |
3043 | out: | |
c9819f45 CL |
3044 | if (time_after(next_balance, sd->last_balance + interval)) |
3045 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
3046 | |
3047 | /* | |
3048 | * Stop the load balance at this level. There is another | |
3049 | * CPU in our sched group which is doing load balancing more | |
3050 | * actively. | |
3051 | */ | |
3052 | if (!balance) | |
3053 | break; | |
1da177e4 | 3054 | } |
46cb4b7c SS |
3055 | rq->next_balance = next_balance; |
3056 | } | |
3057 | ||
3058 | /* | |
3059 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3060 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3061 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3062 | */ | |
3063 | static void run_rebalance_domains(struct softirq_action *h) | |
3064 | { | |
dd41f596 IM |
3065 | int this_cpu = smp_processor_id(); |
3066 | struct rq *this_rq = cpu_rq(this_cpu); | |
3067 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3068 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3069 | |
dd41f596 | 3070 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3071 | |
3072 | #ifdef CONFIG_NO_HZ | |
3073 | /* | |
3074 | * If this cpu is the owner for idle load balancing, then do the | |
3075 | * balancing on behalf of the other idle cpus whose ticks are | |
3076 | * stopped. | |
3077 | */ | |
dd41f596 IM |
3078 | if (this_rq->idle_at_tick && |
3079 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3080 | cpumask_t cpus = nohz.cpu_mask; |
3081 | struct rq *rq; | |
3082 | int balance_cpu; | |
3083 | ||
dd41f596 | 3084 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3085 | for_each_cpu_mask(balance_cpu, cpus) { |
3086 | /* | |
3087 | * If this cpu gets work to do, stop the load balancing | |
3088 | * work being done for other cpus. Next load | |
3089 | * balancing owner will pick it up. | |
3090 | */ | |
3091 | if (need_resched()) | |
3092 | break; | |
3093 | ||
dd41f596 | 3094 | rebalance_domains(balance_cpu, SCHED_IDLE); |
46cb4b7c SS |
3095 | |
3096 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3097 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3098 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3099 | } |
3100 | } | |
3101 | #endif | |
3102 | } | |
3103 | ||
3104 | /* | |
3105 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3106 | * | |
3107 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3108 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3109 | * if the whole system is idle. | |
3110 | */ | |
dd41f596 | 3111 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3112 | { |
46cb4b7c SS |
3113 | #ifdef CONFIG_NO_HZ |
3114 | /* | |
3115 | * If we were in the nohz mode recently and busy at the current | |
3116 | * scheduler tick, then check if we need to nominate new idle | |
3117 | * load balancer. | |
3118 | */ | |
3119 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3120 | rq->in_nohz_recently = 0; | |
3121 | ||
3122 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3123 | cpu_clear(cpu, nohz.cpu_mask); | |
3124 | atomic_set(&nohz.load_balancer, -1); | |
3125 | } | |
3126 | ||
3127 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3128 | /* | |
3129 | * simple selection for now: Nominate the | |
3130 | * first cpu in the nohz list to be the next | |
3131 | * ilb owner. | |
3132 | * | |
3133 | * TBD: Traverse the sched domains and nominate | |
3134 | * the nearest cpu in the nohz.cpu_mask. | |
3135 | */ | |
3136 | int ilb = first_cpu(nohz.cpu_mask); | |
3137 | ||
3138 | if (ilb != NR_CPUS) | |
3139 | resched_cpu(ilb); | |
3140 | } | |
3141 | } | |
3142 | ||
3143 | /* | |
3144 | * If this cpu is idle and doing idle load balancing for all the | |
3145 | * cpus with ticks stopped, is it time for that to stop? | |
3146 | */ | |
3147 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3148 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3149 | resched_cpu(cpu); | |
3150 | return; | |
3151 | } | |
3152 | ||
3153 | /* | |
3154 | * If this cpu is idle and the idle load balancing is done by | |
3155 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3156 | */ | |
3157 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3158 | cpu_isset(cpu, nohz.cpu_mask)) | |
3159 | return; | |
3160 | #endif | |
3161 | if (time_after_eq(jiffies, rq->next_balance)) | |
3162 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3163 | } |
dd41f596 IM |
3164 | |
3165 | #else /* CONFIG_SMP */ | |
3166 | ||
1da177e4 LT |
3167 | /* |
3168 | * on UP we do not need to balance between CPUs: | |
3169 | */ | |
70b97a7f | 3170 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3171 | { |
3172 | } | |
dd41f596 IM |
3173 | |
3174 | /* Avoid "used but not defined" warning on UP */ | |
3175 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3176 | unsigned long max_nr_move, unsigned long max_load_move, | |
3177 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3178 | int *all_pinned, unsigned long *load_moved, | |
3179 | int this_best_prio, int best_prio, int best_prio_seen, | |
3180 | struct rq_iterator *iterator) | |
3181 | { | |
3182 | *load_moved = 0; | |
3183 | ||
3184 | return 0; | |
3185 | } | |
3186 | ||
1da177e4 LT |
3187 | #endif |
3188 | ||
1da177e4 LT |
3189 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3190 | ||
3191 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3192 | ||
3193 | /* | |
41b86e9c IM |
3194 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3195 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3196 | */ |
41b86e9c | 3197 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3198 | { |
1da177e4 | 3199 | unsigned long flags; |
41b86e9c IM |
3200 | u64 ns, delta_exec; |
3201 | struct rq *rq; | |
48f24c4d | 3202 | |
41b86e9c IM |
3203 | rq = task_rq_lock(p, &flags); |
3204 | ns = p->se.sum_exec_runtime; | |
3205 | if (rq->curr == p) { | |
3206 | delta_exec = rq_clock(rq) - p->se.exec_start; | |
3207 | if ((s64)delta_exec > 0) | |
3208 | ns += delta_exec; | |
3209 | } | |
3210 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3211 | |
1da177e4 LT |
3212 | return ns; |
3213 | } | |
3214 | ||
1da177e4 LT |
3215 | /* |
3216 | * Account user cpu time to a process. | |
3217 | * @p: the process that the cpu time gets accounted to | |
3218 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3219 | * @cputime: the cpu time spent in user space since the last update | |
3220 | */ | |
3221 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3222 | { | |
3223 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3224 | cputime64_t tmp; | |
3225 | ||
3226 | p->utime = cputime_add(p->utime, cputime); | |
3227 | ||
3228 | /* Add user time to cpustat. */ | |
3229 | tmp = cputime_to_cputime64(cputime); | |
3230 | if (TASK_NICE(p) > 0) | |
3231 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3232 | else | |
3233 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3234 | } | |
3235 | ||
3236 | /* | |
3237 | * Account system cpu time to a process. | |
3238 | * @p: the process that the cpu time gets accounted to | |
3239 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3240 | * @cputime: the cpu time spent in kernel space since the last update | |
3241 | */ | |
3242 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3243 | cputime_t cputime) | |
3244 | { | |
3245 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3246 | struct rq *rq = this_rq(); |
1da177e4 LT |
3247 | cputime64_t tmp; |
3248 | ||
3249 | p->stime = cputime_add(p->stime, cputime); | |
3250 | ||
3251 | /* Add system time to cpustat. */ | |
3252 | tmp = cputime_to_cputime64(cputime); | |
3253 | if (hardirq_count() - hardirq_offset) | |
3254 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3255 | else if (softirq_count()) | |
3256 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3257 | else if (p != rq->idle) | |
3258 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3259 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3260 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3261 | else | |
3262 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3263 | /* Account for system time used */ | |
3264 | acct_update_integrals(p); | |
1da177e4 LT |
3265 | } |
3266 | ||
3267 | /* | |
3268 | * Account for involuntary wait time. | |
3269 | * @p: the process from which the cpu time has been stolen | |
3270 | * @steal: the cpu time spent in involuntary wait | |
3271 | */ | |
3272 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3273 | { | |
3274 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3275 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3276 | struct rq *rq = this_rq(); |
1da177e4 LT |
3277 | |
3278 | if (p == rq->idle) { | |
3279 | p->stime = cputime_add(p->stime, steal); | |
3280 | if (atomic_read(&rq->nr_iowait) > 0) | |
3281 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3282 | else | |
3283 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3284 | } else | |
3285 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3286 | } | |
3287 | ||
7835b98b CL |
3288 | /* |
3289 | * This function gets called by the timer code, with HZ frequency. | |
3290 | * We call it with interrupts disabled. | |
3291 | * | |
3292 | * It also gets called by the fork code, when changing the parent's | |
3293 | * timeslices. | |
3294 | */ | |
3295 | void scheduler_tick(void) | |
3296 | { | |
7835b98b CL |
3297 | int cpu = smp_processor_id(); |
3298 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 IM |
3299 | struct task_struct *curr = rq->curr; |
3300 | ||
3301 | spin_lock(&rq->lock); | |
3302 | if (curr != rq->idle) /* FIXME: needed? */ | |
3303 | curr->sched_class->task_tick(rq, curr); | |
3304 | update_cpu_load(rq); | |
3305 | spin_unlock(&rq->lock); | |
7835b98b | 3306 | |
e418e1c2 | 3307 | #ifdef CONFIG_SMP |
dd41f596 IM |
3308 | rq->idle_at_tick = idle_cpu(cpu); |
3309 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3310 | #endif |
1da177e4 LT |
3311 | } |
3312 | ||
1da177e4 LT |
3313 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3314 | ||
3315 | void fastcall add_preempt_count(int val) | |
3316 | { | |
3317 | /* | |
3318 | * Underflow? | |
3319 | */ | |
9a11b49a IM |
3320 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3321 | return; | |
1da177e4 LT |
3322 | preempt_count() += val; |
3323 | /* | |
3324 | * Spinlock count overflowing soon? | |
3325 | */ | |
33859f7f MOS |
3326 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3327 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3328 | } |
3329 | EXPORT_SYMBOL(add_preempt_count); | |
3330 | ||
3331 | void fastcall sub_preempt_count(int val) | |
3332 | { | |
3333 | /* | |
3334 | * Underflow? | |
3335 | */ | |
9a11b49a IM |
3336 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3337 | return; | |
1da177e4 LT |
3338 | /* |
3339 | * Is the spinlock portion underflowing? | |
3340 | */ | |
9a11b49a IM |
3341 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3342 | !(preempt_count() & PREEMPT_MASK))) | |
3343 | return; | |
3344 | ||
1da177e4 LT |
3345 | preempt_count() -= val; |
3346 | } | |
3347 | EXPORT_SYMBOL(sub_preempt_count); | |
3348 | ||
3349 | #endif | |
3350 | ||
3351 | /* | |
dd41f596 | 3352 | * Print scheduling while atomic bug: |
1da177e4 | 3353 | */ |
dd41f596 | 3354 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3355 | { |
dd41f596 IM |
3356 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3357 | prev->comm, preempt_count(), prev->pid); | |
3358 | debug_show_held_locks(prev); | |
3359 | if (irqs_disabled()) | |
3360 | print_irqtrace_events(prev); | |
3361 | dump_stack(); | |
3362 | } | |
1da177e4 | 3363 | |
dd41f596 IM |
3364 | /* |
3365 | * Various schedule()-time debugging checks and statistics: | |
3366 | */ | |
3367 | static inline void schedule_debug(struct task_struct *prev) | |
3368 | { | |
1da177e4 LT |
3369 | /* |
3370 | * Test if we are atomic. Since do_exit() needs to call into | |
3371 | * schedule() atomically, we ignore that path for now. | |
3372 | * Otherwise, whine if we are scheduling when we should not be. | |
3373 | */ | |
dd41f596 IM |
3374 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3375 | __schedule_bug(prev); | |
3376 | ||
1da177e4 LT |
3377 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3378 | ||
dd41f596 IM |
3379 | schedstat_inc(this_rq(), sched_cnt); |
3380 | } | |
3381 | ||
3382 | /* | |
3383 | * Pick up the highest-prio task: | |
3384 | */ | |
3385 | static inline struct task_struct * | |
3386 | pick_next_task(struct rq *rq, struct task_struct *prev, u64 now) | |
3387 | { | |
3388 | struct sched_class *class; | |
3389 | struct task_struct *p; | |
1da177e4 LT |
3390 | |
3391 | /* | |
dd41f596 IM |
3392 | * Optimization: we know that if all tasks are in |
3393 | * the fair class we can call that function directly: | |
1da177e4 | 3394 | */ |
dd41f596 IM |
3395 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
3396 | p = fair_sched_class.pick_next_task(rq, now); | |
3397 | if (likely(p)) | |
3398 | return p; | |
1da177e4 LT |
3399 | } |
3400 | ||
dd41f596 IM |
3401 | class = sched_class_highest; |
3402 | for ( ; ; ) { | |
3403 | p = class->pick_next_task(rq, now); | |
3404 | if (p) | |
3405 | return p; | |
3406 | /* | |
3407 | * Will never be NULL as the idle class always | |
3408 | * returns a non-NULL p: | |
3409 | */ | |
3410 | class = class->next; | |
3411 | } | |
3412 | } | |
1da177e4 | 3413 | |
dd41f596 IM |
3414 | /* |
3415 | * schedule() is the main scheduler function. | |
3416 | */ | |
3417 | asmlinkage void __sched schedule(void) | |
3418 | { | |
3419 | struct task_struct *prev, *next; | |
3420 | long *switch_count; | |
3421 | struct rq *rq; | |
3422 | u64 now; | |
3423 | int cpu; | |
3424 | ||
3425 | need_resched: | |
3426 | preempt_disable(); | |
3427 | cpu = smp_processor_id(); | |
3428 | rq = cpu_rq(cpu); | |
3429 | rcu_qsctr_inc(cpu); | |
3430 | prev = rq->curr; | |
3431 | switch_count = &prev->nivcsw; | |
3432 | ||
3433 | release_kernel_lock(prev); | |
3434 | need_resched_nonpreemptible: | |
3435 | ||
3436 | schedule_debug(prev); | |
1da177e4 LT |
3437 | |
3438 | spin_lock_irq(&rq->lock); | |
dd41f596 | 3439 | clear_tsk_need_resched(prev); |
1da177e4 | 3440 | |
1da177e4 | 3441 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3442 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3443 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3444 | prev->state = TASK_RUNNING; |
dd41f596 IM |
3445 | } else { |
3446 | deactivate_task(rq, prev, 1); | |
1da177e4 | 3447 | } |
dd41f596 | 3448 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3449 | } |
3450 | ||
dd41f596 | 3451 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3452 | idle_balance(cpu, rq); |
1da177e4 | 3453 | |
dd41f596 IM |
3454 | now = __rq_clock(rq); |
3455 | prev->sched_class->put_prev_task(rq, prev, now); | |
3456 | next = pick_next_task(rq, prev, now); | |
1da177e4 LT |
3457 | |
3458 | sched_info_switch(prev, next); | |
dd41f596 | 3459 | |
1da177e4 | 3460 | if (likely(prev != next)) { |
1da177e4 LT |
3461 | rq->nr_switches++; |
3462 | rq->curr = next; | |
3463 | ++*switch_count; | |
3464 | ||
dd41f596 | 3465 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3466 | } else |
3467 | spin_unlock_irq(&rq->lock); | |
3468 | ||
dd41f596 IM |
3469 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3470 | cpu = smp_processor_id(); | |
3471 | rq = cpu_rq(cpu); | |
1da177e4 | 3472 | goto need_resched_nonpreemptible; |
dd41f596 | 3473 | } |
1da177e4 LT |
3474 | preempt_enable_no_resched(); |
3475 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3476 | goto need_resched; | |
3477 | } | |
1da177e4 LT |
3478 | EXPORT_SYMBOL(schedule); |
3479 | ||
3480 | #ifdef CONFIG_PREEMPT | |
3481 | /* | |
2ed6e34f | 3482 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3483 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3484 | * occur there and call schedule directly. | |
3485 | */ | |
3486 | asmlinkage void __sched preempt_schedule(void) | |
3487 | { | |
3488 | struct thread_info *ti = current_thread_info(); | |
3489 | #ifdef CONFIG_PREEMPT_BKL | |
3490 | struct task_struct *task = current; | |
3491 | int saved_lock_depth; | |
3492 | #endif | |
3493 | /* | |
3494 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3495 | * we do not want to preempt the current task. Just return.. | |
3496 | */ | |
beed33a8 | 3497 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3498 | return; |
3499 | ||
3500 | need_resched: | |
3501 | add_preempt_count(PREEMPT_ACTIVE); | |
3502 | /* | |
3503 | * We keep the big kernel semaphore locked, but we | |
3504 | * clear ->lock_depth so that schedule() doesnt | |
3505 | * auto-release the semaphore: | |
3506 | */ | |
3507 | #ifdef CONFIG_PREEMPT_BKL | |
3508 | saved_lock_depth = task->lock_depth; | |
3509 | task->lock_depth = -1; | |
3510 | #endif | |
3511 | schedule(); | |
3512 | #ifdef CONFIG_PREEMPT_BKL | |
3513 | task->lock_depth = saved_lock_depth; | |
3514 | #endif | |
3515 | sub_preempt_count(PREEMPT_ACTIVE); | |
3516 | ||
3517 | /* we could miss a preemption opportunity between schedule and now */ | |
3518 | barrier(); | |
3519 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3520 | goto need_resched; | |
3521 | } | |
1da177e4 LT |
3522 | EXPORT_SYMBOL(preempt_schedule); |
3523 | ||
3524 | /* | |
2ed6e34f | 3525 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3526 | * off of irq context. |
3527 | * Note, that this is called and return with irqs disabled. This will | |
3528 | * protect us against recursive calling from irq. | |
3529 | */ | |
3530 | asmlinkage void __sched preempt_schedule_irq(void) | |
3531 | { | |
3532 | struct thread_info *ti = current_thread_info(); | |
3533 | #ifdef CONFIG_PREEMPT_BKL | |
3534 | struct task_struct *task = current; | |
3535 | int saved_lock_depth; | |
3536 | #endif | |
2ed6e34f | 3537 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3538 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3539 | ||
3540 | need_resched: | |
3541 | add_preempt_count(PREEMPT_ACTIVE); | |
3542 | /* | |
3543 | * We keep the big kernel semaphore locked, but we | |
3544 | * clear ->lock_depth so that schedule() doesnt | |
3545 | * auto-release the semaphore: | |
3546 | */ | |
3547 | #ifdef CONFIG_PREEMPT_BKL | |
3548 | saved_lock_depth = task->lock_depth; | |
3549 | task->lock_depth = -1; | |
3550 | #endif | |
3551 | local_irq_enable(); | |
3552 | schedule(); | |
3553 | local_irq_disable(); | |
3554 | #ifdef CONFIG_PREEMPT_BKL | |
3555 | task->lock_depth = saved_lock_depth; | |
3556 | #endif | |
3557 | sub_preempt_count(PREEMPT_ACTIVE); | |
3558 | ||
3559 | /* we could miss a preemption opportunity between schedule and now */ | |
3560 | barrier(); | |
3561 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3562 | goto need_resched; | |
3563 | } | |
3564 | ||
3565 | #endif /* CONFIG_PREEMPT */ | |
3566 | ||
95cdf3b7 IM |
3567 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3568 | void *key) | |
1da177e4 | 3569 | { |
48f24c4d | 3570 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3571 | } |
1da177e4 LT |
3572 | EXPORT_SYMBOL(default_wake_function); |
3573 | ||
3574 | /* | |
3575 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3576 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3577 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3578 | * | |
3579 | * There are circumstances in which we can try to wake a task which has already | |
3580 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3581 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3582 | */ | |
3583 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3584 | int nr_exclusive, int sync, void *key) | |
3585 | { | |
3586 | struct list_head *tmp, *next; | |
3587 | ||
3588 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3589 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3590 | unsigned flags = curr->flags; | |
3591 | ||
1da177e4 | 3592 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3593 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3594 | break; |
3595 | } | |
3596 | } | |
3597 | ||
3598 | /** | |
3599 | * __wake_up - wake up threads blocked on a waitqueue. | |
3600 | * @q: the waitqueue | |
3601 | * @mode: which threads | |
3602 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3603 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3604 | */ |
3605 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3606 | int nr_exclusive, void *key) |
1da177e4 LT |
3607 | { |
3608 | unsigned long flags; | |
3609 | ||
3610 | spin_lock_irqsave(&q->lock, flags); | |
3611 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3612 | spin_unlock_irqrestore(&q->lock, flags); | |
3613 | } | |
1da177e4 LT |
3614 | EXPORT_SYMBOL(__wake_up); |
3615 | ||
3616 | /* | |
3617 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3618 | */ | |
3619 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3620 | { | |
3621 | __wake_up_common(q, mode, 1, 0, NULL); | |
3622 | } | |
3623 | ||
3624 | /** | |
67be2dd1 | 3625 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3626 | * @q: the waitqueue |
3627 | * @mode: which threads | |
3628 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3629 | * | |
3630 | * The sync wakeup differs that the waker knows that it will schedule | |
3631 | * away soon, so while the target thread will be woken up, it will not | |
3632 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3633 | * with each other. This can prevent needless bouncing between CPUs. | |
3634 | * | |
3635 | * On UP it can prevent extra preemption. | |
3636 | */ | |
95cdf3b7 IM |
3637 | void fastcall |
3638 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3639 | { |
3640 | unsigned long flags; | |
3641 | int sync = 1; | |
3642 | ||
3643 | if (unlikely(!q)) | |
3644 | return; | |
3645 | ||
3646 | if (unlikely(!nr_exclusive)) | |
3647 | sync = 0; | |
3648 | ||
3649 | spin_lock_irqsave(&q->lock, flags); | |
3650 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3651 | spin_unlock_irqrestore(&q->lock, flags); | |
3652 | } | |
3653 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3654 | ||
3655 | void fastcall complete(struct completion *x) | |
3656 | { | |
3657 | unsigned long flags; | |
3658 | ||
3659 | spin_lock_irqsave(&x->wait.lock, flags); | |
3660 | x->done++; | |
3661 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3662 | 1, 0, NULL); | |
3663 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3664 | } | |
3665 | EXPORT_SYMBOL(complete); | |
3666 | ||
3667 | void fastcall complete_all(struct completion *x) | |
3668 | { | |
3669 | unsigned long flags; | |
3670 | ||
3671 | spin_lock_irqsave(&x->wait.lock, flags); | |
3672 | x->done += UINT_MAX/2; | |
3673 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3674 | 0, 0, NULL); | |
3675 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3676 | } | |
3677 | EXPORT_SYMBOL(complete_all); | |
3678 | ||
3679 | void fastcall __sched wait_for_completion(struct completion *x) | |
3680 | { | |
3681 | might_sleep(); | |
48f24c4d | 3682 | |
1da177e4 LT |
3683 | spin_lock_irq(&x->wait.lock); |
3684 | if (!x->done) { | |
3685 | DECLARE_WAITQUEUE(wait, current); | |
3686 | ||
3687 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3688 | __add_wait_queue_tail(&x->wait, &wait); | |
3689 | do { | |
3690 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3691 | spin_unlock_irq(&x->wait.lock); | |
3692 | schedule(); | |
3693 | spin_lock_irq(&x->wait.lock); | |
3694 | } while (!x->done); | |
3695 | __remove_wait_queue(&x->wait, &wait); | |
3696 | } | |
3697 | x->done--; | |
3698 | spin_unlock_irq(&x->wait.lock); | |
3699 | } | |
3700 | EXPORT_SYMBOL(wait_for_completion); | |
3701 | ||
3702 | unsigned long fastcall __sched | |
3703 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3704 | { | |
3705 | might_sleep(); | |
3706 | ||
3707 | spin_lock_irq(&x->wait.lock); | |
3708 | if (!x->done) { | |
3709 | DECLARE_WAITQUEUE(wait, current); | |
3710 | ||
3711 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3712 | __add_wait_queue_tail(&x->wait, &wait); | |
3713 | do { | |
3714 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3715 | spin_unlock_irq(&x->wait.lock); | |
3716 | timeout = schedule_timeout(timeout); | |
3717 | spin_lock_irq(&x->wait.lock); | |
3718 | if (!timeout) { | |
3719 | __remove_wait_queue(&x->wait, &wait); | |
3720 | goto out; | |
3721 | } | |
3722 | } while (!x->done); | |
3723 | __remove_wait_queue(&x->wait, &wait); | |
3724 | } | |
3725 | x->done--; | |
3726 | out: | |
3727 | spin_unlock_irq(&x->wait.lock); | |
3728 | return timeout; | |
3729 | } | |
3730 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3731 | ||
3732 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3733 | { | |
3734 | int ret = 0; | |
3735 | ||
3736 | might_sleep(); | |
3737 | ||
3738 | spin_lock_irq(&x->wait.lock); | |
3739 | if (!x->done) { | |
3740 | DECLARE_WAITQUEUE(wait, current); | |
3741 | ||
3742 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3743 | __add_wait_queue_tail(&x->wait, &wait); | |
3744 | do { | |
3745 | if (signal_pending(current)) { | |
3746 | ret = -ERESTARTSYS; | |
3747 | __remove_wait_queue(&x->wait, &wait); | |
3748 | goto out; | |
3749 | } | |
3750 | __set_current_state(TASK_INTERRUPTIBLE); | |
3751 | spin_unlock_irq(&x->wait.lock); | |
3752 | schedule(); | |
3753 | spin_lock_irq(&x->wait.lock); | |
3754 | } while (!x->done); | |
3755 | __remove_wait_queue(&x->wait, &wait); | |
3756 | } | |
3757 | x->done--; | |
3758 | out: | |
3759 | spin_unlock_irq(&x->wait.lock); | |
3760 | ||
3761 | return ret; | |
3762 | } | |
3763 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3764 | ||
3765 | unsigned long fastcall __sched | |
3766 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3767 | unsigned long timeout) | |
3768 | { | |
3769 | might_sleep(); | |
3770 | ||
3771 | spin_lock_irq(&x->wait.lock); | |
3772 | if (!x->done) { | |
3773 | DECLARE_WAITQUEUE(wait, current); | |
3774 | ||
3775 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3776 | __add_wait_queue_tail(&x->wait, &wait); | |
3777 | do { | |
3778 | if (signal_pending(current)) { | |
3779 | timeout = -ERESTARTSYS; | |
3780 | __remove_wait_queue(&x->wait, &wait); | |
3781 | goto out; | |
3782 | } | |
3783 | __set_current_state(TASK_INTERRUPTIBLE); | |
3784 | spin_unlock_irq(&x->wait.lock); | |
3785 | timeout = schedule_timeout(timeout); | |
3786 | spin_lock_irq(&x->wait.lock); | |
3787 | if (!timeout) { | |
3788 | __remove_wait_queue(&x->wait, &wait); | |
3789 | goto out; | |
3790 | } | |
3791 | } while (!x->done); | |
3792 | __remove_wait_queue(&x->wait, &wait); | |
3793 | } | |
3794 | x->done--; | |
3795 | out: | |
3796 | spin_unlock_irq(&x->wait.lock); | |
3797 | return timeout; | |
3798 | } | |
3799 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3800 | ||
0fec171c IM |
3801 | static inline void |
3802 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3803 | { | |
3804 | spin_lock_irqsave(&q->lock, *flags); | |
3805 | __add_wait_queue(q, wait); | |
1da177e4 | 3806 | spin_unlock(&q->lock); |
0fec171c | 3807 | } |
1da177e4 | 3808 | |
0fec171c IM |
3809 | static inline void |
3810 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3811 | { | |
3812 | spin_lock_irq(&q->lock); | |
3813 | __remove_wait_queue(q, wait); | |
3814 | spin_unlock_irqrestore(&q->lock, *flags); | |
3815 | } | |
1da177e4 | 3816 | |
0fec171c | 3817 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3818 | { |
0fec171c IM |
3819 | unsigned long flags; |
3820 | wait_queue_t wait; | |
3821 | ||
3822 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3823 | |
3824 | current->state = TASK_INTERRUPTIBLE; | |
3825 | ||
0fec171c | 3826 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3827 | schedule(); |
0fec171c | 3828 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3829 | } |
1da177e4 LT |
3830 | EXPORT_SYMBOL(interruptible_sleep_on); |
3831 | ||
0fec171c | 3832 | long __sched |
95cdf3b7 | 3833 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3834 | { |
0fec171c IM |
3835 | unsigned long flags; |
3836 | wait_queue_t wait; | |
3837 | ||
3838 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3839 | |
3840 | current->state = TASK_INTERRUPTIBLE; | |
3841 | ||
0fec171c | 3842 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3843 | timeout = schedule_timeout(timeout); |
0fec171c | 3844 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3845 | |
3846 | return timeout; | |
3847 | } | |
1da177e4 LT |
3848 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3849 | ||
0fec171c | 3850 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3851 | { |
0fec171c IM |
3852 | unsigned long flags; |
3853 | wait_queue_t wait; | |
3854 | ||
3855 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3856 | |
3857 | current->state = TASK_UNINTERRUPTIBLE; | |
3858 | ||
0fec171c | 3859 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3860 | schedule(); |
0fec171c | 3861 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3862 | } |
1da177e4 LT |
3863 | EXPORT_SYMBOL(sleep_on); |
3864 | ||
0fec171c | 3865 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3866 | { |
0fec171c IM |
3867 | unsigned long flags; |
3868 | wait_queue_t wait; | |
3869 | ||
3870 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3871 | |
3872 | current->state = TASK_UNINTERRUPTIBLE; | |
3873 | ||
0fec171c | 3874 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3875 | timeout = schedule_timeout(timeout); |
0fec171c | 3876 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3877 | |
3878 | return timeout; | |
3879 | } | |
1da177e4 LT |
3880 | EXPORT_SYMBOL(sleep_on_timeout); |
3881 | ||
b29739f9 IM |
3882 | #ifdef CONFIG_RT_MUTEXES |
3883 | ||
3884 | /* | |
3885 | * rt_mutex_setprio - set the current priority of a task | |
3886 | * @p: task | |
3887 | * @prio: prio value (kernel-internal form) | |
3888 | * | |
3889 | * This function changes the 'effective' priority of a task. It does | |
3890 | * not touch ->normal_prio like __setscheduler(). | |
3891 | * | |
3892 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3893 | */ | |
36c8b586 | 3894 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3895 | { |
3896 | unsigned long flags; | |
dd41f596 | 3897 | int oldprio, on_rq; |
70b97a7f | 3898 | struct rq *rq; |
dd41f596 | 3899 | u64 now; |
b29739f9 IM |
3900 | |
3901 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3902 | ||
3903 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3904 | now = rq_clock(rq); |
b29739f9 | 3905 | |
d5f9f942 | 3906 | oldprio = p->prio; |
dd41f596 IM |
3907 | on_rq = p->se.on_rq; |
3908 | if (on_rq) | |
3909 | dequeue_task(rq, p, 0, now); | |
3910 | ||
3911 | if (rt_prio(prio)) | |
3912 | p->sched_class = &rt_sched_class; | |
3913 | else | |
3914 | p->sched_class = &fair_sched_class; | |
3915 | ||
b29739f9 IM |
3916 | p->prio = prio; |
3917 | ||
dd41f596 IM |
3918 | if (on_rq) { |
3919 | enqueue_task(rq, p, 0, now); | |
b29739f9 IM |
3920 | /* |
3921 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3922 | * our priority decreased, or if we are not currently running on |
3923 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3924 | */ |
d5f9f942 AM |
3925 | if (task_running(rq, p)) { |
3926 | if (p->prio > oldprio) | |
3927 | resched_task(rq->curr); | |
dd41f596 IM |
3928 | } else { |
3929 | check_preempt_curr(rq, p); | |
3930 | } | |
b29739f9 IM |
3931 | } |
3932 | task_rq_unlock(rq, &flags); | |
3933 | } | |
3934 | ||
3935 | #endif | |
3936 | ||
36c8b586 | 3937 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3938 | { |
dd41f596 | 3939 | int old_prio, delta, on_rq; |
1da177e4 | 3940 | unsigned long flags; |
70b97a7f | 3941 | struct rq *rq; |
dd41f596 | 3942 | u64 now; |
1da177e4 LT |
3943 | |
3944 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3945 | return; | |
3946 | /* | |
3947 | * We have to be careful, if called from sys_setpriority(), | |
3948 | * the task might be in the middle of scheduling on another CPU. | |
3949 | */ | |
3950 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3951 | now = rq_clock(rq); |
1da177e4 LT |
3952 | /* |
3953 | * The RT priorities are set via sched_setscheduler(), but we still | |
3954 | * allow the 'normal' nice value to be set - but as expected | |
3955 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3956 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3957 | */ |
e05606d3 | 3958 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3959 | p->static_prio = NICE_TO_PRIO(nice); |
3960 | goto out_unlock; | |
3961 | } | |
dd41f596 IM |
3962 | on_rq = p->se.on_rq; |
3963 | if (on_rq) { | |
3964 | dequeue_task(rq, p, 0, now); | |
3965 | dec_load(rq, p, now); | |
2dd73a4f | 3966 | } |
1da177e4 | 3967 | |
1da177e4 | 3968 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3969 | set_load_weight(p); |
b29739f9 IM |
3970 | old_prio = p->prio; |
3971 | p->prio = effective_prio(p); | |
3972 | delta = p->prio - old_prio; | |
1da177e4 | 3973 | |
dd41f596 IM |
3974 | if (on_rq) { |
3975 | enqueue_task(rq, p, 0, now); | |
3976 | inc_load(rq, p, now); | |
1da177e4 | 3977 | /* |
d5f9f942 AM |
3978 | * If the task increased its priority or is running and |
3979 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3980 | */ |
d5f9f942 | 3981 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3982 | resched_task(rq->curr); |
3983 | } | |
3984 | out_unlock: | |
3985 | task_rq_unlock(rq, &flags); | |
3986 | } | |
1da177e4 LT |
3987 | EXPORT_SYMBOL(set_user_nice); |
3988 | ||
e43379f1 MM |
3989 | /* |
3990 | * can_nice - check if a task can reduce its nice value | |
3991 | * @p: task | |
3992 | * @nice: nice value | |
3993 | */ | |
36c8b586 | 3994 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3995 | { |
024f4747 MM |
3996 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3997 | int nice_rlim = 20 - nice; | |
48f24c4d | 3998 | |
e43379f1 MM |
3999 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4000 | capable(CAP_SYS_NICE)); | |
4001 | } | |
4002 | ||
1da177e4 LT |
4003 | #ifdef __ARCH_WANT_SYS_NICE |
4004 | ||
4005 | /* | |
4006 | * sys_nice - change the priority of the current process. | |
4007 | * @increment: priority increment | |
4008 | * | |
4009 | * sys_setpriority is a more generic, but much slower function that | |
4010 | * does similar things. | |
4011 | */ | |
4012 | asmlinkage long sys_nice(int increment) | |
4013 | { | |
48f24c4d | 4014 | long nice, retval; |
1da177e4 LT |
4015 | |
4016 | /* | |
4017 | * Setpriority might change our priority at the same moment. | |
4018 | * We don't have to worry. Conceptually one call occurs first | |
4019 | * and we have a single winner. | |
4020 | */ | |
e43379f1 MM |
4021 | if (increment < -40) |
4022 | increment = -40; | |
1da177e4 LT |
4023 | if (increment > 40) |
4024 | increment = 40; | |
4025 | ||
4026 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4027 | if (nice < -20) | |
4028 | nice = -20; | |
4029 | if (nice > 19) | |
4030 | nice = 19; | |
4031 | ||
e43379f1 MM |
4032 | if (increment < 0 && !can_nice(current, nice)) |
4033 | return -EPERM; | |
4034 | ||
1da177e4 LT |
4035 | retval = security_task_setnice(current, nice); |
4036 | if (retval) | |
4037 | return retval; | |
4038 | ||
4039 | set_user_nice(current, nice); | |
4040 | return 0; | |
4041 | } | |
4042 | ||
4043 | #endif | |
4044 | ||
4045 | /** | |
4046 | * task_prio - return the priority value of a given task. | |
4047 | * @p: the task in question. | |
4048 | * | |
4049 | * This is the priority value as seen by users in /proc. | |
4050 | * RT tasks are offset by -200. Normal tasks are centered | |
4051 | * around 0, value goes from -16 to +15. | |
4052 | */ | |
36c8b586 | 4053 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4054 | { |
4055 | return p->prio - MAX_RT_PRIO; | |
4056 | } | |
4057 | ||
4058 | /** | |
4059 | * task_nice - return the nice value of a given task. | |
4060 | * @p: the task in question. | |
4061 | */ | |
36c8b586 | 4062 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4063 | { |
4064 | return TASK_NICE(p); | |
4065 | } | |
1da177e4 | 4066 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4067 | |
4068 | /** | |
4069 | * idle_cpu - is a given cpu idle currently? | |
4070 | * @cpu: the processor in question. | |
4071 | */ | |
4072 | int idle_cpu(int cpu) | |
4073 | { | |
4074 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4075 | } | |
4076 | ||
1da177e4 LT |
4077 | /** |
4078 | * idle_task - return the idle task for a given cpu. | |
4079 | * @cpu: the processor in question. | |
4080 | */ | |
36c8b586 | 4081 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4082 | { |
4083 | return cpu_rq(cpu)->idle; | |
4084 | } | |
4085 | ||
4086 | /** | |
4087 | * find_process_by_pid - find a process with a matching PID value. | |
4088 | * @pid: the pid in question. | |
4089 | */ | |
36c8b586 | 4090 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4091 | { |
4092 | return pid ? find_task_by_pid(pid) : current; | |
4093 | } | |
4094 | ||
4095 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4096 | static void |
4097 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4098 | { |
dd41f596 | 4099 | BUG_ON(p->se.on_rq); |
48f24c4d | 4100 | |
1da177e4 | 4101 | p->policy = policy; |
dd41f596 IM |
4102 | switch (p->policy) { |
4103 | case SCHED_NORMAL: | |
4104 | case SCHED_BATCH: | |
4105 | case SCHED_IDLE: | |
4106 | p->sched_class = &fair_sched_class; | |
4107 | break; | |
4108 | case SCHED_FIFO: | |
4109 | case SCHED_RR: | |
4110 | p->sched_class = &rt_sched_class; | |
4111 | break; | |
4112 | } | |
4113 | ||
1da177e4 | 4114 | p->rt_priority = prio; |
b29739f9 IM |
4115 | p->normal_prio = normal_prio(p); |
4116 | /* we are holding p->pi_lock already */ | |
4117 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4118 | set_load_weight(p); |
1da177e4 LT |
4119 | } |
4120 | ||
4121 | /** | |
72fd4a35 | 4122 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4123 | * @p: the task in question. |
4124 | * @policy: new policy. | |
4125 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4126 | * |
72fd4a35 | 4127 | * NOTE that the task may be already dead. |
1da177e4 | 4128 | */ |
95cdf3b7 IM |
4129 | int sched_setscheduler(struct task_struct *p, int policy, |
4130 | struct sched_param *param) | |
1da177e4 | 4131 | { |
dd41f596 | 4132 | int retval, oldprio, oldpolicy = -1, on_rq; |
1da177e4 | 4133 | unsigned long flags; |
70b97a7f | 4134 | struct rq *rq; |
1da177e4 | 4135 | |
66e5393a SR |
4136 | /* may grab non-irq protected spin_locks */ |
4137 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4138 | recheck: |
4139 | /* double check policy once rq lock held */ | |
4140 | if (policy < 0) | |
4141 | policy = oldpolicy = p->policy; | |
4142 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4143 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4144 | policy != SCHED_IDLE) | |
b0a9499c | 4145 | return -EINVAL; |
1da177e4 LT |
4146 | /* |
4147 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4148 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4149 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4150 | */ |
4151 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4152 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4153 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4154 | return -EINVAL; |
e05606d3 | 4155 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4156 | return -EINVAL; |
4157 | ||
37e4ab3f OC |
4158 | /* |
4159 | * Allow unprivileged RT tasks to decrease priority: | |
4160 | */ | |
4161 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4162 | if (rt_policy(policy)) { |
8dc3e909 | 4163 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4164 | |
4165 | if (!lock_task_sighand(p, &flags)) | |
4166 | return -ESRCH; | |
4167 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4168 | unlock_task_sighand(p, &flags); | |
4169 | ||
4170 | /* can't set/change the rt policy */ | |
4171 | if (policy != p->policy && !rlim_rtprio) | |
4172 | return -EPERM; | |
4173 | ||
4174 | /* can't increase priority */ | |
4175 | if (param->sched_priority > p->rt_priority && | |
4176 | param->sched_priority > rlim_rtprio) | |
4177 | return -EPERM; | |
4178 | } | |
dd41f596 IM |
4179 | /* |
4180 | * Like positive nice levels, dont allow tasks to | |
4181 | * move out of SCHED_IDLE either: | |
4182 | */ | |
4183 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4184 | return -EPERM; | |
5fe1d75f | 4185 | |
37e4ab3f OC |
4186 | /* can't change other user's priorities */ |
4187 | if ((current->euid != p->euid) && | |
4188 | (current->euid != p->uid)) | |
4189 | return -EPERM; | |
4190 | } | |
1da177e4 LT |
4191 | |
4192 | retval = security_task_setscheduler(p, policy, param); | |
4193 | if (retval) | |
4194 | return retval; | |
b29739f9 IM |
4195 | /* |
4196 | * make sure no PI-waiters arrive (or leave) while we are | |
4197 | * changing the priority of the task: | |
4198 | */ | |
4199 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4200 | /* |
4201 | * To be able to change p->policy safely, the apropriate | |
4202 | * runqueue lock must be held. | |
4203 | */ | |
b29739f9 | 4204 | rq = __task_rq_lock(p); |
1da177e4 LT |
4205 | /* recheck policy now with rq lock held */ |
4206 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4207 | policy = oldpolicy = -1; | |
b29739f9 IM |
4208 | __task_rq_unlock(rq); |
4209 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4210 | goto recheck; |
4211 | } | |
dd41f596 IM |
4212 | on_rq = p->se.on_rq; |
4213 | if (on_rq) | |
4214 | deactivate_task(rq, p, 0); | |
1da177e4 | 4215 | oldprio = p->prio; |
dd41f596 IM |
4216 | __setscheduler(rq, p, policy, param->sched_priority); |
4217 | if (on_rq) { | |
4218 | activate_task(rq, p, 0); | |
1da177e4 LT |
4219 | /* |
4220 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4221 | * our priority decreased, or if we are not currently running on |
4222 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4223 | */ |
d5f9f942 AM |
4224 | if (task_running(rq, p)) { |
4225 | if (p->prio > oldprio) | |
4226 | resched_task(rq->curr); | |
dd41f596 IM |
4227 | } else { |
4228 | check_preempt_curr(rq, p); | |
4229 | } | |
1da177e4 | 4230 | } |
b29739f9 IM |
4231 | __task_rq_unlock(rq); |
4232 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4233 | ||
95e02ca9 TG |
4234 | rt_mutex_adjust_pi(p); |
4235 | ||
1da177e4 LT |
4236 | return 0; |
4237 | } | |
4238 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4239 | ||
95cdf3b7 IM |
4240 | static int |
4241 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4242 | { |
1da177e4 LT |
4243 | struct sched_param lparam; |
4244 | struct task_struct *p; | |
36c8b586 | 4245 | int retval; |
1da177e4 LT |
4246 | |
4247 | if (!param || pid < 0) | |
4248 | return -EINVAL; | |
4249 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4250 | return -EFAULT; | |
5fe1d75f ON |
4251 | |
4252 | rcu_read_lock(); | |
4253 | retval = -ESRCH; | |
1da177e4 | 4254 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4255 | if (p != NULL) |
4256 | retval = sched_setscheduler(p, policy, &lparam); | |
4257 | rcu_read_unlock(); | |
36c8b586 | 4258 | |
1da177e4 LT |
4259 | return retval; |
4260 | } | |
4261 | ||
4262 | /** | |
4263 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4264 | * @pid: the pid in question. | |
4265 | * @policy: new policy. | |
4266 | * @param: structure containing the new RT priority. | |
4267 | */ | |
4268 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4269 | struct sched_param __user *param) | |
4270 | { | |
c21761f1 JB |
4271 | /* negative values for policy are not valid */ |
4272 | if (policy < 0) | |
4273 | return -EINVAL; | |
4274 | ||
1da177e4 LT |
4275 | return do_sched_setscheduler(pid, policy, param); |
4276 | } | |
4277 | ||
4278 | /** | |
4279 | * sys_sched_setparam - set/change the RT priority of a thread | |
4280 | * @pid: the pid in question. | |
4281 | * @param: structure containing the new RT priority. | |
4282 | */ | |
4283 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4284 | { | |
4285 | return do_sched_setscheduler(pid, -1, param); | |
4286 | } | |
4287 | ||
4288 | /** | |
4289 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4290 | * @pid: the pid in question. | |
4291 | */ | |
4292 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4293 | { | |
36c8b586 | 4294 | struct task_struct *p; |
1da177e4 | 4295 | int retval = -EINVAL; |
1da177e4 LT |
4296 | |
4297 | if (pid < 0) | |
4298 | goto out_nounlock; | |
4299 | ||
4300 | retval = -ESRCH; | |
4301 | read_lock(&tasklist_lock); | |
4302 | p = find_process_by_pid(pid); | |
4303 | if (p) { | |
4304 | retval = security_task_getscheduler(p); | |
4305 | if (!retval) | |
4306 | retval = p->policy; | |
4307 | } | |
4308 | read_unlock(&tasklist_lock); | |
4309 | ||
4310 | out_nounlock: | |
4311 | return retval; | |
4312 | } | |
4313 | ||
4314 | /** | |
4315 | * sys_sched_getscheduler - get the RT priority of a thread | |
4316 | * @pid: the pid in question. | |
4317 | * @param: structure containing the RT priority. | |
4318 | */ | |
4319 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4320 | { | |
4321 | struct sched_param lp; | |
36c8b586 | 4322 | struct task_struct *p; |
1da177e4 | 4323 | int retval = -EINVAL; |
1da177e4 LT |
4324 | |
4325 | if (!param || pid < 0) | |
4326 | goto out_nounlock; | |
4327 | ||
4328 | read_lock(&tasklist_lock); | |
4329 | p = find_process_by_pid(pid); | |
4330 | retval = -ESRCH; | |
4331 | if (!p) | |
4332 | goto out_unlock; | |
4333 | ||
4334 | retval = security_task_getscheduler(p); | |
4335 | if (retval) | |
4336 | goto out_unlock; | |
4337 | ||
4338 | lp.sched_priority = p->rt_priority; | |
4339 | read_unlock(&tasklist_lock); | |
4340 | ||
4341 | /* | |
4342 | * This one might sleep, we cannot do it with a spinlock held ... | |
4343 | */ | |
4344 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4345 | ||
4346 | out_nounlock: | |
4347 | return retval; | |
4348 | ||
4349 | out_unlock: | |
4350 | read_unlock(&tasklist_lock); | |
4351 | return retval; | |
4352 | } | |
4353 | ||
4354 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4355 | { | |
1da177e4 | 4356 | cpumask_t cpus_allowed; |
36c8b586 IM |
4357 | struct task_struct *p; |
4358 | int retval; | |
1da177e4 | 4359 | |
5be9361c | 4360 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4361 | read_lock(&tasklist_lock); |
4362 | ||
4363 | p = find_process_by_pid(pid); | |
4364 | if (!p) { | |
4365 | read_unlock(&tasklist_lock); | |
5be9361c | 4366 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4367 | return -ESRCH; |
4368 | } | |
4369 | ||
4370 | /* | |
4371 | * It is not safe to call set_cpus_allowed with the | |
4372 | * tasklist_lock held. We will bump the task_struct's | |
4373 | * usage count and then drop tasklist_lock. | |
4374 | */ | |
4375 | get_task_struct(p); | |
4376 | read_unlock(&tasklist_lock); | |
4377 | ||
4378 | retval = -EPERM; | |
4379 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4380 | !capable(CAP_SYS_NICE)) | |
4381 | goto out_unlock; | |
4382 | ||
e7834f8f DQ |
4383 | retval = security_task_setscheduler(p, 0, NULL); |
4384 | if (retval) | |
4385 | goto out_unlock; | |
4386 | ||
1da177e4 LT |
4387 | cpus_allowed = cpuset_cpus_allowed(p); |
4388 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4389 | retval = set_cpus_allowed(p, new_mask); | |
4390 | ||
4391 | out_unlock: | |
4392 | put_task_struct(p); | |
5be9361c | 4393 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4394 | return retval; |
4395 | } | |
4396 | ||
4397 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4398 | cpumask_t *new_mask) | |
4399 | { | |
4400 | if (len < sizeof(cpumask_t)) { | |
4401 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4402 | } else if (len > sizeof(cpumask_t)) { | |
4403 | len = sizeof(cpumask_t); | |
4404 | } | |
4405 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4406 | } | |
4407 | ||
4408 | /** | |
4409 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4410 | * @pid: pid of the process | |
4411 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4412 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4413 | */ | |
4414 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4415 | unsigned long __user *user_mask_ptr) | |
4416 | { | |
4417 | cpumask_t new_mask; | |
4418 | int retval; | |
4419 | ||
4420 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4421 | if (retval) | |
4422 | return retval; | |
4423 | ||
4424 | return sched_setaffinity(pid, new_mask); | |
4425 | } | |
4426 | ||
4427 | /* | |
4428 | * Represents all cpu's present in the system | |
4429 | * In systems capable of hotplug, this map could dynamically grow | |
4430 | * as new cpu's are detected in the system via any platform specific | |
4431 | * method, such as ACPI for e.g. | |
4432 | */ | |
4433 | ||
4cef0c61 | 4434 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4435 | EXPORT_SYMBOL(cpu_present_map); |
4436 | ||
4437 | #ifndef CONFIG_SMP | |
4cef0c61 | 4438 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4439 | EXPORT_SYMBOL(cpu_online_map); |
4440 | ||
4cef0c61 | 4441 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4442 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4443 | #endif |
4444 | ||
4445 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4446 | { | |
36c8b586 | 4447 | struct task_struct *p; |
1da177e4 | 4448 | int retval; |
1da177e4 | 4449 | |
5be9361c | 4450 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4451 | read_lock(&tasklist_lock); |
4452 | ||
4453 | retval = -ESRCH; | |
4454 | p = find_process_by_pid(pid); | |
4455 | if (!p) | |
4456 | goto out_unlock; | |
4457 | ||
e7834f8f DQ |
4458 | retval = security_task_getscheduler(p); |
4459 | if (retval) | |
4460 | goto out_unlock; | |
4461 | ||
2f7016d9 | 4462 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4463 | |
4464 | out_unlock: | |
4465 | read_unlock(&tasklist_lock); | |
5be9361c | 4466 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4467 | if (retval) |
4468 | return retval; | |
4469 | ||
4470 | return 0; | |
4471 | } | |
4472 | ||
4473 | /** | |
4474 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4475 | * @pid: pid of the process | |
4476 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4477 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4478 | */ | |
4479 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4480 | unsigned long __user *user_mask_ptr) | |
4481 | { | |
4482 | int ret; | |
4483 | cpumask_t mask; | |
4484 | ||
4485 | if (len < sizeof(cpumask_t)) | |
4486 | return -EINVAL; | |
4487 | ||
4488 | ret = sched_getaffinity(pid, &mask); | |
4489 | if (ret < 0) | |
4490 | return ret; | |
4491 | ||
4492 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4493 | return -EFAULT; | |
4494 | ||
4495 | return sizeof(cpumask_t); | |
4496 | } | |
4497 | ||
4498 | /** | |
4499 | * sys_sched_yield - yield the current processor to other threads. | |
4500 | * | |
dd41f596 IM |
4501 | * This function yields the current CPU to other tasks. If there are no |
4502 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4503 | */ |
4504 | asmlinkage long sys_sched_yield(void) | |
4505 | { | |
70b97a7f | 4506 | struct rq *rq = this_rq_lock(); |
1da177e4 LT |
4507 | |
4508 | schedstat_inc(rq, yld_cnt); | |
dd41f596 | 4509 | if (unlikely(rq->nr_running == 1)) |
1da177e4 | 4510 | schedstat_inc(rq, yld_act_empty); |
dd41f596 IM |
4511 | else |
4512 | current->sched_class->yield_task(rq, current); | |
1da177e4 LT |
4513 | |
4514 | /* | |
4515 | * Since we are going to call schedule() anyway, there's | |
4516 | * no need to preempt or enable interrupts: | |
4517 | */ | |
4518 | __release(rq->lock); | |
8a25d5de | 4519 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4520 | _raw_spin_unlock(&rq->lock); |
4521 | preempt_enable_no_resched(); | |
4522 | ||
4523 | schedule(); | |
4524 | ||
4525 | return 0; | |
4526 | } | |
4527 | ||
e7b38404 | 4528 | static void __cond_resched(void) |
1da177e4 | 4529 | { |
8e0a43d8 IM |
4530 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4531 | __might_sleep(__FILE__, __LINE__); | |
4532 | #endif | |
5bbcfd90 IM |
4533 | /* |
4534 | * The BKS might be reacquired before we have dropped | |
4535 | * PREEMPT_ACTIVE, which could trigger a second | |
4536 | * cond_resched() call. | |
4537 | */ | |
1da177e4 LT |
4538 | do { |
4539 | add_preempt_count(PREEMPT_ACTIVE); | |
4540 | schedule(); | |
4541 | sub_preempt_count(PREEMPT_ACTIVE); | |
4542 | } while (need_resched()); | |
4543 | } | |
4544 | ||
4545 | int __sched cond_resched(void) | |
4546 | { | |
9414232f IM |
4547 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4548 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4549 | __cond_resched(); |
4550 | return 1; | |
4551 | } | |
4552 | return 0; | |
4553 | } | |
1da177e4 LT |
4554 | EXPORT_SYMBOL(cond_resched); |
4555 | ||
4556 | /* | |
4557 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4558 | * call schedule, and on return reacquire the lock. | |
4559 | * | |
4560 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4561 | * operations here to prevent schedule() from being called twice (once via | |
4562 | * spin_unlock(), once by hand). | |
4563 | */ | |
95cdf3b7 | 4564 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4565 | { |
6df3cecb JK |
4566 | int ret = 0; |
4567 | ||
1da177e4 LT |
4568 | if (need_lockbreak(lock)) { |
4569 | spin_unlock(lock); | |
4570 | cpu_relax(); | |
6df3cecb | 4571 | ret = 1; |
1da177e4 LT |
4572 | spin_lock(lock); |
4573 | } | |
9414232f | 4574 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4575 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4576 | _raw_spin_unlock(lock); |
4577 | preempt_enable_no_resched(); | |
4578 | __cond_resched(); | |
6df3cecb | 4579 | ret = 1; |
1da177e4 | 4580 | spin_lock(lock); |
1da177e4 | 4581 | } |
6df3cecb | 4582 | return ret; |
1da177e4 | 4583 | } |
1da177e4 LT |
4584 | EXPORT_SYMBOL(cond_resched_lock); |
4585 | ||
4586 | int __sched cond_resched_softirq(void) | |
4587 | { | |
4588 | BUG_ON(!in_softirq()); | |
4589 | ||
9414232f | 4590 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4591 | local_bh_enable(); |
1da177e4 LT |
4592 | __cond_resched(); |
4593 | local_bh_disable(); | |
4594 | return 1; | |
4595 | } | |
4596 | return 0; | |
4597 | } | |
1da177e4 LT |
4598 | EXPORT_SYMBOL(cond_resched_softirq); |
4599 | ||
1da177e4 LT |
4600 | /** |
4601 | * yield - yield the current processor to other threads. | |
4602 | * | |
72fd4a35 | 4603 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4604 | * thread runnable and calls sys_sched_yield(). |
4605 | */ | |
4606 | void __sched yield(void) | |
4607 | { | |
4608 | set_current_state(TASK_RUNNING); | |
4609 | sys_sched_yield(); | |
4610 | } | |
1da177e4 LT |
4611 | EXPORT_SYMBOL(yield); |
4612 | ||
4613 | /* | |
4614 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4615 | * that process accounting knows that this is a task in IO wait state. | |
4616 | * | |
4617 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4618 | * has set its backing_dev_info: the queue against which it should throttle) | |
4619 | */ | |
4620 | void __sched io_schedule(void) | |
4621 | { | |
70b97a7f | 4622 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4623 | |
0ff92245 | 4624 | delayacct_blkio_start(); |
1da177e4 LT |
4625 | atomic_inc(&rq->nr_iowait); |
4626 | schedule(); | |
4627 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4628 | delayacct_blkio_end(); |
1da177e4 | 4629 | } |
1da177e4 LT |
4630 | EXPORT_SYMBOL(io_schedule); |
4631 | ||
4632 | long __sched io_schedule_timeout(long timeout) | |
4633 | { | |
70b97a7f | 4634 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4635 | long ret; |
4636 | ||
0ff92245 | 4637 | delayacct_blkio_start(); |
1da177e4 LT |
4638 | atomic_inc(&rq->nr_iowait); |
4639 | ret = schedule_timeout(timeout); | |
4640 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4641 | delayacct_blkio_end(); |
1da177e4 LT |
4642 | return ret; |
4643 | } | |
4644 | ||
4645 | /** | |
4646 | * sys_sched_get_priority_max - return maximum RT priority. | |
4647 | * @policy: scheduling class. | |
4648 | * | |
4649 | * this syscall returns the maximum rt_priority that can be used | |
4650 | * by a given scheduling class. | |
4651 | */ | |
4652 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4653 | { | |
4654 | int ret = -EINVAL; | |
4655 | ||
4656 | switch (policy) { | |
4657 | case SCHED_FIFO: | |
4658 | case SCHED_RR: | |
4659 | ret = MAX_USER_RT_PRIO-1; | |
4660 | break; | |
4661 | case SCHED_NORMAL: | |
b0a9499c | 4662 | case SCHED_BATCH: |
dd41f596 | 4663 | case SCHED_IDLE: |
1da177e4 LT |
4664 | ret = 0; |
4665 | break; | |
4666 | } | |
4667 | return ret; | |
4668 | } | |
4669 | ||
4670 | /** | |
4671 | * sys_sched_get_priority_min - return minimum RT priority. | |
4672 | * @policy: scheduling class. | |
4673 | * | |
4674 | * this syscall returns the minimum rt_priority that can be used | |
4675 | * by a given scheduling class. | |
4676 | */ | |
4677 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4678 | { | |
4679 | int ret = -EINVAL; | |
4680 | ||
4681 | switch (policy) { | |
4682 | case SCHED_FIFO: | |
4683 | case SCHED_RR: | |
4684 | ret = 1; | |
4685 | break; | |
4686 | case SCHED_NORMAL: | |
b0a9499c | 4687 | case SCHED_BATCH: |
dd41f596 | 4688 | case SCHED_IDLE: |
1da177e4 LT |
4689 | ret = 0; |
4690 | } | |
4691 | return ret; | |
4692 | } | |
4693 | ||
4694 | /** | |
4695 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4696 | * @pid: pid of the process. | |
4697 | * @interval: userspace pointer to the timeslice value. | |
4698 | * | |
4699 | * this syscall writes the default timeslice value of a given process | |
4700 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4701 | */ | |
4702 | asmlinkage | |
4703 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4704 | { | |
36c8b586 | 4705 | struct task_struct *p; |
1da177e4 LT |
4706 | int retval = -EINVAL; |
4707 | struct timespec t; | |
1da177e4 LT |
4708 | |
4709 | if (pid < 0) | |
4710 | goto out_nounlock; | |
4711 | ||
4712 | retval = -ESRCH; | |
4713 | read_lock(&tasklist_lock); | |
4714 | p = find_process_by_pid(pid); | |
4715 | if (!p) | |
4716 | goto out_unlock; | |
4717 | ||
4718 | retval = security_task_getscheduler(p); | |
4719 | if (retval) | |
4720 | goto out_unlock; | |
4721 | ||
b78709cf | 4722 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
dd41f596 | 4723 | 0 : static_prio_timeslice(p->static_prio), &t); |
1da177e4 LT |
4724 | read_unlock(&tasklist_lock); |
4725 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4726 | out_nounlock: | |
4727 | return retval; | |
4728 | out_unlock: | |
4729 | read_unlock(&tasklist_lock); | |
4730 | return retval; | |
4731 | } | |
4732 | ||
2ed6e34f | 4733 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4734 | |
4735 | static void show_task(struct task_struct *p) | |
1da177e4 | 4736 | { |
1da177e4 | 4737 | unsigned long free = 0; |
36c8b586 | 4738 | unsigned state; |
1da177e4 | 4739 | |
1da177e4 | 4740 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4741 | printk("%-13.13s %c", p->comm, |
4742 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4743 | #if BITS_PER_LONG == 32 |
1da177e4 | 4744 | if (state == TASK_RUNNING) |
4bd77321 | 4745 | printk(" running "); |
1da177e4 | 4746 | else |
4bd77321 | 4747 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4748 | #else |
4749 | if (state == TASK_RUNNING) | |
4bd77321 | 4750 | printk(" running task "); |
1da177e4 LT |
4751 | else |
4752 | printk(" %016lx ", thread_saved_pc(p)); | |
4753 | #endif | |
4754 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4755 | { | |
10ebffde | 4756 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4757 | while (!*n) |
4758 | n++; | |
10ebffde | 4759 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4760 | } |
4761 | #endif | |
4bd77321 | 4762 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4763 | |
4764 | if (state != TASK_RUNNING) | |
4765 | show_stack(p, NULL); | |
4766 | } | |
4767 | ||
e59e2ae2 | 4768 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4769 | { |
36c8b586 | 4770 | struct task_struct *g, *p; |
1da177e4 | 4771 | |
4bd77321 IM |
4772 | #if BITS_PER_LONG == 32 |
4773 | printk(KERN_INFO | |
4774 | " task PC stack pid father\n"); | |
1da177e4 | 4775 | #else |
4bd77321 IM |
4776 | printk(KERN_INFO |
4777 | " task PC stack pid father\n"); | |
1da177e4 LT |
4778 | #endif |
4779 | read_lock(&tasklist_lock); | |
4780 | do_each_thread(g, p) { | |
4781 | /* | |
4782 | * reset the NMI-timeout, listing all files on a slow | |
4783 | * console might take alot of time: | |
4784 | */ | |
4785 | touch_nmi_watchdog(); | |
39bc89fd | 4786 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4787 | show_task(p); |
1da177e4 LT |
4788 | } while_each_thread(g, p); |
4789 | ||
04c9167f JF |
4790 | touch_all_softlockup_watchdogs(); |
4791 | ||
dd41f596 IM |
4792 | #ifdef CONFIG_SCHED_DEBUG |
4793 | sysrq_sched_debug_show(); | |
4794 | #endif | |
1da177e4 | 4795 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4796 | /* |
4797 | * Only show locks if all tasks are dumped: | |
4798 | */ | |
4799 | if (state_filter == -1) | |
4800 | debug_show_all_locks(); | |
1da177e4 LT |
4801 | } |
4802 | ||
1df21055 IM |
4803 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4804 | { | |
dd41f596 | 4805 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4806 | } |
4807 | ||
f340c0d1 IM |
4808 | /** |
4809 | * init_idle - set up an idle thread for a given CPU | |
4810 | * @idle: task in question | |
4811 | * @cpu: cpu the idle task belongs to | |
4812 | * | |
4813 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4814 | * flag, to make booting more robust. | |
4815 | */ | |
5c1e1767 | 4816 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4817 | { |
70b97a7f | 4818 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4819 | unsigned long flags; |
4820 | ||
dd41f596 IM |
4821 | __sched_fork(idle); |
4822 | idle->se.exec_start = sched_clock(); | |
4823 | ||
b29739f9 | 4824 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4825 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4826 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4827 | |
4828 | spin_lock_irqsave(&rq->lock, flags); | |
4829 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4830 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4831 | idle->oncpu = 1; | |
4832 | #endif | |
1da177e4 LT |
4833 | spin_unlock_irqrestore(&rq->lock, flags); |
4834 | ||
4835 | /* Set the preempt count _outside_ the spinlocks! */ | |
4836 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4837 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4838 | #else |
a1261f54 | 4839 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4840 | #endif |
dd41f596 IM |
4841 | /* |
4842 | * The idle tasks have their own, simple scheduling class: | |
4843 | */ | |
4844 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4845 | } |
4846 | ||
4847 | /* | |
4848 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4849 | * indicates which cpus entered this state. This is used | |
4850 | * in the rcu update to wait only for active cpus. For system | |
4851 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4852 | * always be CPU_MASK_NONE. | |
4853 | */ | |
4854 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4855 | ||
dd41f596 IM |
4856 | /* |
4857 | * Increase the granularity value when there are more CPUs, | |
4858 | * because with more CPUs the 'effective latency' as visible | |
4859 | * to users decreases. But the relationship is not linear, | |
4860 | * so pick a second-best guess by going with the log2 of the | |
4861 | * number of CPUs. | |
4862 | * | |
4863 | * This idea comes from the SD scheduler of Con Kolivas: | |
4864 | */ | |
4865 | static inline void sched_init_granularity(void) | |
4866 | { | |
4867 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
a5968df8 | 4868 | const unsigned long gran_limit = 100000000; |
dd41f596 IM |
4869 | |
4870 | sysctl_sched_granularity *= factor; | |
4871 | if (sysctl_sched_granularity > gran_limit) | |
4872 | sysctl_sched_granularity = gran_limit; | |
4873 | ||
4874 | sysctl_sched_runtime_limit = sysctl_sched_granularity * 4; | |
4875 | sysctl_sched_wakeup_granularity = sysctl_sched_granularity / 2; | |
4876 | } | |
4877 | ||
1da177e4 LT |
4878 | #ifdef CONFIG_SMP |
4879 | /* | |
4880 | * This is how migration works: | |
4881 | * | |
70b97a7f | 4882 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4883 | * runqueue and wake up that CPU's migration thread. |
4884 | * 2) we down() the locked semaphore => thread blocks. | |
4885 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4886 | * thread off the CPU) | |
4887 | * 4) it gets the migration request and checks whether the migrated | |
4888 | * task is still in the wrong runqueue. | |
4889 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4890 | * it and puts it into the right queue. | |
4891 | * 6) migration thread up()s the semaphore. | |
4892 | * 7) we wake up and the migration is done. | |
4893 | */ | |
4894 | ||
4895 | /* | |
4896 | * Change a given task's CPU affinity. Migrate the thread to a | |
4897 | * proper CPU and schedule it away if the CPU it's executing on | |
4898 | * is removed from the allowed bitmask. | |
4899 | * | |
4900 | * NOTE: the caller must have a valid reference to the task, the | |
4901 | * task must not exit() & deallocate itself prematurely. The | |
4902 | * call is not atomic; no spinlocks may be held. | |
4903 | */ | |
36c8b586 | 4904 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4905 | { |
70b97a7f | 4906 | struct migration_req req; |
1da177e4 | 4907 | unsigned long flags; |
70b97a7f | 4908 | struct rq *rq; |
48f24c4d | 4909 | int ret = 0; |
1da177e4 LT |
4910 | |
4911 | rq = task_rq_lock(p, &flags); | |
4912 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4913 | ret = -EINVAL; | |
4914 | goto out; | |
4915 | } | |
4916 | ||
4917 | p->cpus_allowed = new_mask; | |
4918 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4919 | if (cpu_isset(task_cpu(p), new_mask)) | |
4920 | goto out; | |
4921 | ||
4922 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4923 | /* Need help from migration thread: drop lock and wait. */ | |
4924 | task_rq_unlock(rq, &flags); | |
4925 | wake_up_process(rq->migration_thread); | |
4926 | wait_for_completion(&req.done); | |
4927 | tlb_migrate_finish(p->mm); | |
4928 | return 0; | |
4929 | } | |
4930 | out: | |
4931 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4932 | |
1da177e4 LT |
4933 | return ret; |
4934 | } | |
1da177e4 LT |
4935 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4936 | ||
4937 | /* | |
4938 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4939 | * this because either it can't run here any more (set_cpus_allowed() | |
4940 | * away from this CPU, or CPU going down), or because we're | |
4941 | * attempting to rebalance this task on exec (sched_exec). | |
4942 | * | |
4943 | * So we race with normal scheduler movements, but that's OK, as long | |
4944 | * as the task is no longer on this CPU. | |
efc30814 KK |
4945 | * |
4946 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4947 | */ |
efc30814 | 4948 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4949 | { |
70b97a7f | 4950 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4951 | int ret = 0, on_rq; |
1da177e4 LT |
4952 | |
4953 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4954 | return ret; |
1da177e4 LT |
4955 | |
4956 | rq_src = cpu_rq(src_cpu); | |
4957 | rq_dest = cpu_rq(dest_cpu); | |
4958 | ||
4959 | double_rq_lock(rq_src, rq_dest); | |
4960 | /* Already moved. */ | |
4961 | if (task_cpu(p) != src_cpu) | |
4962 | goto out; | |
4963 | /* Affinity changed (again). */ | |
4964 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4965 | goto out; | |
4966 | ||
dd41f596 IM |
4967 | on_rq = p->se.on_rq; |
4968 | if (on_rq) | |
4969 | deactivate_task(rq_src, p, 0); | |
1da177e4 | 4970 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4971 | if (on_rq) { |
4972 | activate_task(rq_dest, p, 0); | |
4973 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4974 | } |
efc30814 | 4975 | ret = 1; |
1da177e4 LT |
4976 | out: |
4977 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4978 | return ret; |
1da177e4 LT |
4979 | } |
4980 | ||
4981 | /* | |
4982 | * migration_thread - this is a highprio system thread that performs | |
4983 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4984 | * another runqueue. | |
4985 | */ | |
95cdf3b7 | 4986 | static int migration_thread(void *data) |
1da177e4 | 4987 | { |
1da177e4 | 4988 | int cpu = (long)data; |
70b97a7f | 4989 | struct rq *rq; |
1da177e4 LT |
4990 | |
4991 | rq = cpu_rq(cpu); | |
4992 | BUG_ON(rq->migration_thread != current); | |
4993 | ||
4994 | set_current_state(TASK_INTERRUPTIBLE); | |
4995 | while (!kthread_should_stop()) { | |
70b97a7f | 4996 | struct migration_req *req; |
1da177e4 | 4997 | struct list_head *head; |
1da177e4 | 4998 | |
1da177e4 LT |
4999 | spin_lock_irq(&rq->lock); |
5000 | ||
5001 | if (cpu_is_offline(cpu)) { | |
5002 | spin_unlock_irq(&rq->lock); | |
5003 | goto wait_to_die; | |
5004 | } | |
5005 | ||
5006 | if (rq->active_balance) { | |
5007 | active_load_balance(rq, cpu); | |
5008 | rq->active_balance = 0; | |
5009 | } | |
5010 | ||
5011 | head = &rq->migration_queue; | |
5012 | ||
5013 | if (list_empty(head)) { | |
5014 | spin_unlock_irq(&rq->lock); | |
5015 | schedule(); | |
5016 | set_current_state(TASK_INTERRUPTIBLE); | |
5017 | continue; | |
5018 | } | |
70b97a7f | 5019 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5020 | list_del_init(head->next); |
5021 | ||
674311d5 NP |
5022 | spin_unlock(&rq->lock); |
5023 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5024 | local_irq_enable(); | |
1da177e4 LT |
5025 | |
5026 | complete(&req->done); | |
5027 | } | |
5028 | __set_current_state(TASK_RUNNING); | |
5029 | return 0; | |
5030 | ||
5031 | wait_to_die: | |
5032 | /* Wait for kthread_stop */ | |
5033 | set_current_state(TASK_INTERRUPTIBLE); | |
5034 | while (!kthread_should_stop()) { | |
5035 | schedule(); | |
5036 | set_current_state(TASK_INTERRUPTIBLE); | |
5037 | } | |
5038 | __set_current_state(TASK_RUNNING); | |
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5043 | /* |
5044 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5045 | * NOTE: interrupts should be disabled by the caller | |
5046 | */ | |
48f24c4d | 5047 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5048 | { |
efc30814 | 5049 | unsigned long flags; |
1da177e4 | 5050 | cpumask_t mask; |
70b97a7f IM |
5051 | struct rq *rq; |
5052 | int dest_cpu; | |
1da177e4 | 5053 | |
efc30814 | 5054 | restart: |
1da177e4 LT |
5055 | /* On same node? */ |
5056 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5057 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5058 | dest_cpu = any_online_cpu(mask); |
5059 | ||
5060 | /* On any allowed CPU? */ | |
5061 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5062 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5063 | |
5064 | /* No more Mr. Nice Guy. */ | |
5065 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5066 | rq = task_rq_lock(p, &flags); |
5067 | cpus_setall(p->cpus_allowed); | |
5068 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5069 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5070 | |
5071 | /* | |
5072 | * Don't tell them about moving exiting tasks or | |
5073 | * kernel threads (both mm NULL), since they never | |
5074 | * leave kernel. | |
5075 | */ | |
48f24c4d | 5076 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5077 | printk(KERN_INFO "process %d (%s) no " |
5078 | "longer affine to cpu%d\n", | |
48f24c4d | 5079 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5080 | } |
48f24c4d | 5081 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5082 | goto restart; |
1da177e4 LT |
5083 | } |
5084 | ||
5085 | /* | |
5086 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5087 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5088 | * for performance reasons the counter is not stricly tracking tasks to | |
5089 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5090 | * to keep the global sum constant after CPU-down: | |
5091 | */ | |
70b97a7f | 5092 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5093 | { |
70b97a7f | 5094 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5095 | unsigned long flags; |
5096 | ||
5097 | local_irq_save(flags); | |
5098 | double_rq_lock(rq_src, rq_dest); | |
5099 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5100 | rq_src->nr_uninterruptible = 0; | |
5101 | double_rq_unlock(rq_src, rq_dest); | |
5102 | local_irq_restore(flags); | |
5103 | } | |
5104 | ||
5105 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5106 | static void migrate_live_tasks(int src_cpu) | |
5107 | { | |
48f24c4d | 5108 | struct task_struct *p, *t; |
1da177e4 LT |
5109 | |
5110 | write_lock_irq(&tasklist_lock); | |
5111 | ||
48f24c4d IM |
5112 | do_each_thread(t, p) { |
5113 | if (p == current) | |
1da177e4 LT |
5114 | continue; |
5115 | ||
48f24c4d IM |
5116 | if (task_cpu(p) == src_cpu) |
5117 | move_task_off_dead_cpu(src_cpu, p); | |
5118 | } while_each_thread(t, p); | |
1da177e4 LT |
5119 | |
5120 | write_unlock_irq(&tasklist_lock); | |
5121 | } | |
5122 | ||
dd41f596 IM |
5123 | /* |
5124 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5125 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5126 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5127 | */ |
5128 | void sched_idle_next(void) | |
5129 | { | |
48f24c4d | 5130 | int this_cpu = smp_processor_id(); |
70b97a7f | 5131 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5132 | struct task_struct *p = rq->idle; |
5133 | unsigned long flags; | |
5134 | ||
5135 | /* cpu has to be offline */ | |
48f24c4d | 5136 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5137 | |
48f24c4d IM |
5138 | /* |
5139 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5140 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5141 | */ |
5142 | spin_lock_irqsave(&rq->lock, flags); | |
5143 | ||
dd41f596 | 5144 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5145 | |
5146 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5147 | activate_idle_task(p, rq); |
1da177e4 LT |
5148 | |
5149 | spin_unlock_irqrestore(&rq->lock, flags); | |
5150 | } | |
5151 | ||
48f24c4d IM |
5152 | /* |
5153 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5154 | * offline. |
5155 | */ | |
5156 | void idle_task_exit(void) | |
5157 | { | |
5158 | struct mm_struct *mm = current->active_mm; | |
5159 | ||
5160 | BUG_ON(cpu_online(smp_processor_id())); | |
5161 | ||
5162 | if (mm != &init_mm) | |
5163 | switch_mm(mm, &init_mm, current); | |
5164 | mmdrop(mm); | |
5165 | } | |
5166 | ||
054b9108 | 5167 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5168 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5169 | { |
70b97a7f | 5170 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5171 | |
5172 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5173 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5174 | |
5175 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5176 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5177 | |
48f24c4d | 5178 | get_task_struct(p); |
1da177e4 LT |
5179 | |
5180 | /* | |
5181 | * Drop lock around migration; if someone else moves it, | |
5182 | * that's OK. No task can be added to this CPU, so iteration is | |
5183 | * fine. | |
054b9108 | 5184 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5185 | */ |
054b9108 | 5186 | spin_unlock(&rq->lock); |
48f24c4d | 5187 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5188 | spin_lock(&rq->lock); |
1da177e4 | 5189 | |
48f24c4d | 5190 | put_task_struct(p); |
1da177e4 LT |
5191 | } |
5192 | ||
5193 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5194 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5195 | { | |
70b97a7f | 5196 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5197 | struct task_struct *next; |
48f24c4d | 5198 | |
dd41f596 IM |
5199 | for ( ; ; ) { |
5200 | if (!rq->nr_running) | |
5201 | break; | |
5202 | next = pick_next_task(rq, rq->curr, rq_clock(rq)); | |
5203 | if (!next) | |
5204 | break; | |
5205 | migrate_dead(dead_cpu, next); | |
1da177e4 LT |
5206 | } |
5207 | } | |
5208 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5209 | ||
5210 | /* | |
5211 | * migration_call - callback that gets triggered when a CPU is added. | |
5212 | * Here we can start up the necessary migration thread for the new CPU. | |
5213 | */ | |
48f24c4d IM |
5214 | static int __cpuinit |
5215 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5216 | { |
1da177e4 | 5217 | struct task_struct *p; |
48f24c4d | 5218 | int cpu = (long)hcpu; |
1da177e4 | 5219 | unsigned long flags; |
70b97a7f | 5220 | struct rq *rq; |
1da177e4 LT |
5221 | |
5222 | switch (action) { | |
5be9361c GS |
5223 | case CPU_LOCK_ACQUIRE: |
5224 | mutex_lock(&sched_hotcpu_mutex); | |
5225 | break; | |
5226 | ||
1da177e4 | 5227 | case CPU_UP_PREPARE: |
8bb78442 | 5228 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5229 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5230 | if (IS_ERR(p)) |
5231 | return NOTIFY_BAD; | |
1da177e4 LT |
5232 | kthread_bind(p, cpu); |
5233 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5234 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5235 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5236 | task_rq_unlock(rq, &flags); |
5237 | cpu_rq(cpu)->migration_thread = p; | |
5238 | break; | |
48f24c4d | 5239 | |
1da177e4 | 5240 | case CPU_ONLINE: |
8bb78442 | 5241 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5242 | /* Strictly unneccessary, as first user will wake it. */ |
5243 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5244 | break; | |
48f24c4d | 5245 | |
1da177e4 LT |
5246 | #ifdef CONFIG_HOTPLUG_CPU |
5247 | case CPU_UP_CANCELED: | |
8bb78442 | 5248 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5249 | if (!cpu_rq(cpu)->migration_thread) |
5250 | break; | |
1da177e4 | 5251 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5252 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5253 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5254 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5255 | cpu_rq(cpu)->migration_thread = NULL; | |
5256 | break; | |
48f24c4d | 5257 | |
1da177e4 | 5258 | case CPU_DEAD: |
8bb78442 | 5259 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5260 | migrate_live_tasks(cpu); |
5261 | rq = cpu_rq(cpu); | |
5262 | kthread_stop(rq->migration_thread); | |
5263 | rq->migration_thread = NULL; | |
5264 | /* Idle task back to normal (off runqueue, low prio) */ | |
5265 | rq = task_rq_lock(rq->idle, &flags); | |
dd41f596 | 5266 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5267 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5268 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5269 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5270 | migrate_dead_tasks(cpu); |
5271 | task_rq_unlock(rq, &flags); | |
5272 | migrate_nr_uninterruptible(rq); | |
5273 | BUG_ON(rq->nr_running != 0); | |
5274 | ||
5275 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5276 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5277 | * the requestors. */ |
5278 | spin_lock_irq(&rq->lock); | |
5279 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5280 | struct migration_req *req; |
5281 | ||
1da177e4 | 5282 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5283 | struct migration_req, list); |
1da177e4 LT |
5284 | list_del_init(&req->list); |
5285 | complete(&req->done); | |
5286 | } | |
5287 | spin_unlock_irq(&rq->lock); | |
5288 | break; | |
5289 | #endif | |
5be9361c GS |
5290 | case CPU_LOCK_RELEASE: |
5291 | mutex_unlock(&sched_hotcpu_mutex); | |
5292 | break; | |
1da177e4 LT |
5293 | } |
5294 | return NOTIFY_OK; | |
5295 | } | |
5296 | ||
5297 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5298 | * happens before everything else. | |
5299 | */ | |
26c2143b | 5300 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5301 | .notifier_call = migration_call, |
5302 | .priority = 10 | |
5303 | }; | |
5304 | ||
5305 | int __init migration_init(void) | |
5306 | { | |
5307 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5308 | int err; |
48f24c4d IM |
5309 | |
5310 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5311 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5312 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5313 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5314 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5315 | |
1da177e4 LT |
5316 | return 0; |
5317 | } | |
5318 | #endif | |
5319 | ||
5320 | #ifdef CONFIG_SMP | |
476f3534 CL |
5321 | |
5322 | /* Number of possible processor ids */ | |
5323 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5324 | EXPORT_SYMBOL(nr_cpu_ids); | |
5325 | ||
1a20ff27 | 5326 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5327 | #ifdef SCHED_DOMAIN_DEBUG |
5328 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5329 | { | |
5330 | int level = 0; | |
5331 | ||
41c7ce9a NP |
5332 | if (!sd) { |
5333 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5334 | return; | |
5335 | } | |
5336 | ||
1da177e4 LT |
5337 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5338 | ||
5339 | do { | |
5340 | int i; | |
5341 | char str[NR_CPUS]; | |
5342 | struct sched_group *group = sd->groups; | |
5343 | cpumask_t groupmask; | |
5344 | ||
5345 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5346 | cpus_clear(groupmask); | |
5347 | ||
5348 | printk(KERN_DEBUG); | |
5349 | for (i = 0; i < level + 1; i++) | |
5350 | printk(" "); | |
5351 | printk("domain %d: ", level); | |
5352 | ||
5353 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5354 | printk("does not load-balance\n"); | |
5355 | if (sd->parent) | |
33859f7f MOS |
5356 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5357 | " has parent"); | |
1da177e4 LT |
5358 | break; |
5359 | } | |
5360 | ||
5361 | printk("span %s\n", str); | |
5362 | ||
5363 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5364 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5365 | "CPU%d\n", cpu); | |
1da177e4 | 5366 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5367 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5368 | " CPU%d\n", cpu); | |
1da177e4 LT |
5369 | |
5370 | printk(KERN_DEBUG); | |
5371 | for (i = 0; i < level + 2; i++) | |
5372 | printk(" "); | |
5373 | printk("groups:"); | |
5374 | do { | |
5375 | if (!group) { | |
5376 | printk("\n"); | |
5377 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5378 | break; | |
5379 | } | |
5380 | ||
5517d86b | 5381 | if (!group->__cpu_power) { |
1da177e4 | 5382 | printk("\n"); |
33859f7f MOS |
5383 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5384 | "set\n"); | |
1da177e4 LT |
5385 | } |
5386 | ||
5387 | if (!cpus_weight(group->cpumask)) { | |
5388 | printk("\n"); | |
5389 | printk(KERN_ERR "ERROR: empty group\n"); | |
5390 | } | |
5391 | ||
5392 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5393 | printk("\n"); | |
5394 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5395 | } | |
5396 | ||
5397 | cpus_or(groupmask, groupmask, group->cpumask); | |
5398 | ||
5399 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5400 | printk(" %s", str); | |
5401 | ||
5402 | group = group->next; | |
5403 | } while (group != sd->groups); | |
5404 | printk("\n"); | |
5405 | ||
5406 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5407 | printk(KERN_ERR "ERROR: groups don't span " |
5408 | "domain->span\n"); | |
1da177e4 LT |
5409 | |
5410 | level++; | |
5411 | sd = sd->parent; | |
33859f7f MOS |
5412 | if (!sd) |
5413 | continue; | |
1da177e4 | 5414 | |
33859f7f MOS |
5415 | if (!cpus_subset(groupmask, sd->span)) |
5416 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5417 | "of domain->span\n"); | |
1da177e4 LT |
5418 | |
5419 | } while (sd); | |
5420 | } | |
5421 | #else | |
48f24c4d | 5422 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5423 | #endif |
5424 | ||
1a20ff27 | 5425 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5426 | { |
5427 | if (cpus_weight(sd->span) == 1) | |
5428 | return 1; | |
5429 | ||
5430 | /* Following flags need at least 2 groups */ | |
5431 | if (sd->flags & (SD_LOAD_BALANCE | | |
5432 | SD_BALANCE_NEWIDLE | | |
5433 | SD_BALANCE_FORK | | |
89c4710e SS |
5434 | SD_BALANCE_EXEC | |
5435 | SD_SHARE_CPUPOWER | | |
5436 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5437 | if (sd->groups != sd->groups->next) |
5438 | return 0; | |
5439 | } | |
5440 | ||
5441 | /* Following flags don't use groups */ | |
5442 | if (sd->flags & (SD_WAKE_IDLE | | |
5443 | SD_WAKE_AFFINE | | |
5444 | SD_WAKE_BALANCE)) | |
5445 | return 0; | |
5446 | ||
5447 | return 1; | |
5448 | } | |
5449 | ||
48f24c4d IM |
5450 | static int |
5451 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5452 | { |
5453 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5454 | ||
5455 | if (sd_degenerate(parent)) | |
5456 | return 1; | |
5457 | ||
5458 | if (!cpus_equal(sd->span, parent->span)) | |
5459 | return 0; | |
5460 | ||
5461 | /* Does parent contain flags not in child? */ | |
5462 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5463 | if (cflags & SD_WAKE_AFFINE) | |
5464 | pflags &= ~SD_WAKE_BALANCE; | |
5465 | /* Flags needing groups don't count if only 1 group in parent */ | |
5466 | if (parent->groups == parent->groups->next) { | |
5467 | pflags &= ~(SD_LOAD_BALANCE | | |
5468 | SD_BALANCE_NEWIDLE | | |
5469 | SD_BALANCE_FORK | | |
89c4710e SS |
5470 | SD_BALANCE_EXEC | |
5471 | SD_SHARE_CPUPOWER | | |
5472 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5473 | } |
5474 | if (~cflags & pflags) | |
5475 | return 0; | |
5476 | ||
5477 | return 1; | |
5478 | } | |
5479 | ||
1da177e4 LT |
5480 | /* |
5481 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5482 | * hold the hotplug lock. | |
5483 | */ | |
9c1cfda2 | 5484 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5485 | { |
70b97a7f | 5486 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5487 | struct sched_domain *tmp; |
5488 | ||
5489 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5490 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5491 | struct sched_domain *parent = tmp->parent; | |
5492 | if (!parent) | |
5493 | break; | |
1a848870 | 5494 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5495 | tmp->parent = parent->parent; |
1a848870 SS |
5496 | if (parent->parent) |
5497 | parent->parent->child = tmp; | |
5498 | } | |
245af2c7 SS |
5499 | } |
5500 | ||
1a848870 | 5501 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5502 | sd = sd->parent; |
1a848870 SS |
5503 | if (sd) |
5504 | sd->child = NULL; | |
5505 | } | |
1da177e4 LT |
5506 | |
5507 | sched_domain_debug(sd, cpu); | |
5508 | ||
674311d5 | 5509 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5510 | } |
5511 | ||
5512 | /* cpus with isolated domains */ | |
67af63a6 | 5513 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5514 | |
5515 | /* Setup the mask of cpus configured for isolated domains */ | |
5516 | static int __init isolated_cpu_setup(char *str) | |
5517 | { | |
5518 | int ints[NR_CPUS], i; | |
5519 | ||
5520 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5521 | cpus_clear(cpu_isolated_map); | |
5522 | for (i = 1; i <= ints[0]; i++) | |
5523 | if (ints[i] < NR_CPUS) | |
5524 | cpu_set(ints[i], cpu_isolated_map); | |
5525 | return 1; | |
5526 | } | |
5527 | ||
5528 | __setup ("isolcpus=", isolated_cpu_setup); | |
5529 | ||
5530 | /* | |
6711cab4 SS |
5531 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5532 | * to a function which identifies what group(along with sched group) a CPU | |
5533 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5534 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5535 | * |
5536 | * init_sched_build_groups will build a circular linked list of the groups | |
5537 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5538 | * and ->cpu_power to 0. | |
5539 | */ | |
a616058b | 5540 | static void |
6711cab4 SS |
5541 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5542 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5543 | struct sched_group **sg)) | |
1da177e4 LT |
5544 | { |
5545 | struct sched_group *first = NULL, *last = NULL; | |
5546 | cpumask_t covered = CPU_MASK_NONE; | |
5547 | int i; | |
5548 | ||
5549 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5550 | struct sched_group *sg; |
5551 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5552 | int j; |
5553 | ||
5554 | if (cpu_isset(i, covered)) | |
5555 | continue; | |
5556 | ||
5557 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5558 | sg->__cpu_power = 0; |
1da177e4 LT |
5559 | |
5560 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5561 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5562 | continue; |
5563 | ||
5564 | cpu_set(j, covered); | |
5565 | cpu_set(j, sg->cpumask); | |
5566 | } | |
5567 | if (!first) | |
5568 | first = sg; | |
5569 | if (last) | |
5570 | last->next = sg; | |
5571 | last = sg; | |
5572 | } | |
5573 | last->next = first; | |
5574 | } | |
5575 | ||
9c1cfda2 | 5576 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5577 | |
9c1cfda2 | 5578 | #ifdef CONFIG_NUMA |
198e2f18 | 5579 | |
9c1cfda2 JH |
5580 | /** |
5581 | * find_next_best_node - find the next node to include in a sched_domain | |
5582 | * @node: node whose sched_domain we're building | |
5583 | * @used_nodes: nodes already in the sched_domain | |
5584 | * | |
5585 | * Find the next node to include in a given scheduling domain. Simply | |
5586 | * finds the closest node not already in the @used_nodes map. | |
5587 | * | |
5588 | * Should use nodemask_t. | |
5589 | */ | |
5590 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5591 | { | |
5592 | int i, n, val, min_val, best_node = 0; | |
5593 | ||
5594 | min_val = INT_MAX; | |
5595 | ||
5596 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5597 | /* Start at @node */ | |
5598 | n = (node + i) % MAX_NUMNODES; | |
5599 | ||
5600 | if (!nr_cpus_node(n)) | |
5601 | continue; | |
5602 | ||
5603 | /* Skip already used nodes */ | |
5604 | if (test_bit(n, used_nodes)) | |
5605 | continue; | |
5606 | ||
5607 | /* Simple min distance search */ | |
5608 | val = node_distance(node, n); | |
5609 | ||
5610 | if (val < min_val) { | |
5611 | min_val = val; | |
5612 | best_node = n; | |
5613 | } | |
5614 | } | |
5615 | ||
5616 | set_bit(best_node, used_nodes); | |
5617 | return best_node; | |
5618 | } | |
5619 | ||
5620 | /** | |
5621 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5622 | * @node: node whose cpumask we're constructing | |
5623 | * @size: number of nodes to include in this span | |
5624 | * | |
5625 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5626 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5627 | * out optimally. | |
5628 | */ | |
5629 | static cpumask_t sched_domain_node_span(int node) | |
5630 | { | |
9c1cfda2 | 5631 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5632 | cpumask_t span, nodemask; |
5633 | int i; | |
9c1cfda2 JH |
5634 | |
5635 | cpus_clear(span); | |
5636 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5637 | ||
5638 | nodemask = node_to_cpumask(node); | |
5639 | cpus_or(span, span, nodemask); | |
5640 | set_bit(node, used_nodes); | |
5641 | ||
5642 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5643 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5644 | |
9c1cfda2 JH |
5645 | nodemask = node_to_cpumask(next_node); |
5646 | cpus_or(span, span, nodemask); | |
5647 | } | |
5648 | ||
5649 | return span; | |
5650 | } | |
5651 | #endif | |
5652 | ||
5c45bf27 | 5653 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5654 | |
9c1cfda2 | 5655 | /* |
48f24c4d | 5656 | * SMT sched-domains: |
9c1cfda2 | 5657 | */ |
1da177e4 LT |
5658 | #ifdef CONFIG_SCHED_SMT |
5659 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5660 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5661 | |
6711cab4 SS |
5662 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5663 | struct sched_group **sg) | |
1da177e4 | 5664 | { |
6711cab4 SS |
5665 | if (sg) |
5666 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5667 | return cpu; |
5668 | } | |
5669 | #endif | |
5670 | ||
48f24c4d IM |
5671 | /* |
5672 | * multi-core sched-domains: | |
5673 | */ | |
1e9f28fa SS |
5674 | #ifdef CONFIG_SCHED_MC |
5675 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5676 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5677 | #endif |
5678 | ||
5679 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5680 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5681 | struct sched_group **sg) | |
1e9f28fa | 5682 | { |
6711cab4 | 5683 | int group; |
a616058b SS |
5684 | cpumask_t mask = cpu_sibling_map[cpu]; |
5685 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5686 | group = first_cpu(mask); |
5687 | if (sg) | |
5688 | *sg = &per_cpu(sched_group_core, group); | |
5689 | return group; | |
1e9f28fa SS |
5690 | } |
5691 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5692 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5693 | struct sched_group **sg) | |
1e9f28fa | 5694 | { |
6711cab4 SS |
5695 | if (sg) |
5696 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5697 | return cpu; |
5698 | } | |
5699 | #endif | |
5700 | ||
1da177e4 | 5701 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5702 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5703 | |
6711cab4 SS |
5704 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5705 | struct sched_group **sg) | |
1da177e4 | 5706 | { |
6711cab4 | 5707 | int group; |
48f24c4d | 5708 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5709 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5710 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5711 | group = first_cpu(mask); |
1e9f28fa | 5712 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5713 | cpumask_t mask = cpu_sibling_map[cpu]; |
5714 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5715 | group = first_cpu(mask); |
1da177e4 | 5716 | #else |
6711cab4 | 5717 | group = cpu; |
1da177e4 | 5718 | #endif |
6711cab4 SS |
5719 | if (sg) |
5720 | *sg = &per_cpu(sched_group_phys, group); | |
5721 | return group; | |
1da177e4 LT |
5722 | } |
5723 | ||
5724 | #ifdef CONFIG_NUMA | |
1da177e4 | 5725 | /* |
9c1cfda2 JH |
5726 | * The init_sched_build_groups can't handle what we want to do with node |
5727 | * groups, so roll our own. Now each node has its own list of groups which | |
5728 | * gets dynamically allocated. | |
1da177e4 | 5729 | */ |
9c1cfda2 | 5730 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5731 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5732 | |
9c1cfda2 | 5733 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5734 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5735 | |
6711cab4 SS |
5736 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5737 | struct sched_group **sg) | |
9c1cfda2 | 5738 | { |
6711cab4 SS |
5739 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5740 | int group; | |
5741 | ||
5742 | cpus_and(nodemask, nodemask, *cpu_map); | |
5743 | group = first_cpu(nodemask); | |
5744 | ||
5745 | if (sg) | |
5746 | *sg = &per_cpu(sched_group_allnodes, group); | |
5747 | return group; | |
1da177e4 | 5748 | } |
6711cab4 | 5749 | |
08069033 SS |
5750 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5751 | { | |
5752 | struct sched_group *sg = group_head; | |
5753 | int j; | |
5754 | ||
5755 | if (!sg) | |
5756 | return; | |
5757 | next_sg: | |
5758 | for_each_cpu_mask(j, sg->cpumask) { | |
5759 | struct sched_domain *sd; | |
5760 | ||
5761 | sd = &per_cpu(phys_domains, j); | |
5762 | if (j != first_cpu(sd->groups->cpumask)) { | |
5763 | /* | |
5764 | * Only add "power" once for each | |
5765 | * physical package. | |
5766 | */ | |
5767 | continue; | |
5768 | } | |
5769 | ||
5517d86b | 5770 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5771 | } |
5772 | sg = sg->next; | |
5773 | if (sg != group_head) | |
5774 | goto next_sg; | |
5775 | } | |
1da177e4 LT |
5776 | #endif |
5777 | ||
a616058b | 5778 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5779 | /* Free memory allocated for various sched_group structures */ |
5780 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5781 | { | |
a616058b | 5782 | int cpu, i; |
51888ca2 SV |
5783 | |
5784 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5785 | struct sched_group **sched_group_nodes |
5786 | = sched_group_nodes_bycpu[cpu]; | |
5787 | ||
51888ca2 SV |
5788 | if (!sched_group_nodes) |
5789 | continue; | |
5790 | ||
5791 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5792 | cpumask_t nodemask = node_to_cpumask(i); | |
5793 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5794 | ||
5795 | cpus_and(nodemask, nodemask, *cpu_map); | |
5796 | if (cpus_empty(nodemask)) | |
5797 | continue; | |
5798 | ||
5799 | if (sg == NULL) | |
5800 | continue; | |
5801 | sg = sg->next; | |
5802 | next_sg: | |
5803 | oldsg = sg; | |
5804 | sg = sg->next; | |
5805 | kfree(oldsg); | |
5806 | if (oldsg != sched_group_nodes[i]) | |
5807 | goto next_sg; | |
5808 | } | |
5809 | kfree(sched_group_nodes); | |
5810 | sched_group_nodes_bycpu[cpu] = NULL; | |
5811 | } | |
51888ca2 | 5812 | } |
a616058b SS |
5813 | #else |
5814 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5815 | { | |
5816 | } | |
5817 | #endif | |
51888ca2 | 5818 | |
89c4710e SS |
5819 | /* |
5820 | * Initialize sched groups cpu_power. | |
5821 | * | |
5822 | * cpu_power indicates the capacity of sched group, which is used while | |
5823 | * distributing the load between different sched groups in a sched domain. | |
5824 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5825 | * there are asymmetries in the topology. If there are asymmetries, group | |
5826 | * having more cpu_power will pickup more load compared to the group having | |
5827 | * less cpu_power. | |
5828 | * | |
5829 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5830 | * the maximum number of tasks a group can handle in the presence of other idle | |
5831 | * or lightly loaded groups in the same sched domain. | |
5832 | */ | |
5833 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5834 | { | |
5835 | struct sched_domain *child; | |
5836 | struct sched_group *group; | |
5837 | ||
5838 | WARN_ON(!sd || !sd->groups); | |
5839 | ||
5840 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5841 | return; | |
5842 | ||
5843 | child = sd->child; | |
5844 | ||
5517d86b ED |
5845 | sd->groups->__cpu_power = 0; |
5846 | ||
89c4710e SS |
5847 | /* |
5848 | * For perf policy, if the groups in child domain share resources | |
5849 | * (for example cores sharing some portions of the cache hierarchy | |
5850 | * or SMT), then set this domain groups cpu_power such that each group | |
5851 | * can handle only one task, when there are other idle groups in the | |
5852 | * same sched domain. | |
5853 | */ | |
5854 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5855 | (child->flags & | |
5856 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5857 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5858 | return; |
5859 | } | |
5860 | ||
89c4710e SS |
5861 | /* |
5862 | * add cpu_power of each child group to this groups cpu_power | |
5863 | */ | |
5864 | group = child->groups; | |
5865 | do { | |
5517d86b | 5866 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5867 | group = group->next; |
5868 | } while (group != child->groups); | |
5869 | } | |
5870 | ||
1da177e4 | 5871 | /* |
1a20ff27 DG |
5872 | * Build sched domains for a given set of cpus and attach the sched domains |
5873 | * to the individual cpus | |
1da177e4 | 5874 | */ |
51888ca2 | 5875 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5876 | { |
5877 | int i; | |
d1b55138 JH |
5878 | #ifdef CONFIG_NUMA |
5879 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 5880 | int sd_allnodes = 0; |
d1b55138 JH |
5881 | |
5882 | /* | |
5883 | * Allocate the per-node list of sched groups | |
5884 | */ | |
dd41f596 | 5885 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 5886 | GFP_KERNEL); |
d1b55138 JH |
5887 | if (!sched_group_nodes) { |
5888 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 5889 | return -ENOMEM; |
d1b55138 JH |
5890 | } |
5891 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
5892 | #endif | |
1da177e4 LT |
5893 | |
5894 | /* | |
1a20ff27 | 5895 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 5896 | */ |
1a20ff27 | 5897 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
5898 | struct sched_domain *sd = NULL, *p; |
5899 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
5900 | ||
1a20ff27 | 5901 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5902 | |
5903 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
5904 | if (cpus_weight(*cpu_map) > |
5905 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
5906 | sd = &per_cpu(allnodes_domains, i); |
5907 | *sd = SD_ALLNODES_INIT; | |
5908 | sd->span = *cpu_map; | |
6711cab4 | 5909 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 5910 | p = sd; |
6711cab4 | 5911 | sd_allnodes = 1; |
9c1cfda2 JH |
5912 | } else |
5913 | p = NULL; | |
5914 | ||
1da177e4 | 5915 | sd = &per_cpu(node_domains, i); |
1da177e4 | 5916 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
5917 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
5918 | sd->parent = p; | |
1a848870 SS |
5919 | if (p) |
5920 | p->child = sd; | |
9c1cfda2 | 5921 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
5922 | #endif |
5923 | ||
5924 | p = sd; | |
5925 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
5926 | *sd = SD_CPU_INIT; |
5927 | sd->span = nodemask; | |
5928 | sd->parent = p; | |
1a848870 SS |
5929 | if (p) |
5930 | p->child = sd; | |
6711cab4 | 5931 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 5932 | |
1e9f28fa SS |
5933 | #ifdef CONFIG_SCHED_MC |
5934 | p = sd; | |
5935 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
5936 | *sd = SD_MC_INIT; |
5937 | sd->span = cpu_coregroup_map(i); | |
5938 | cpus_and(sd->span, sd->span, *cpu_map); | |
5939 | sd->parent = p; | |
1a848870 | 5940 | p->child = sd; |
6711cab4 | 5941 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
5942 | #endif |
5943 | ||
1da177e4 LT |
5944 | #ifdef CONFIG_SCHED_SMT |
5945 | p = sd; | |
5946 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
5947 | *sd = SD_SIBLING_INIT; |
5948 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 5949 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 5950 | sd->parent = p; |
1a848870 | 5951 | p->child = sd; |
6711cab4 | 5952 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
5953 | #endif |
5954 | } | |
5955 | ||
5956 | #ifdef CONFIG_SCHED_SMT | |
5957 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 5958 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 5959 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 5960 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
5961 | if (i != first_cpu(this_sibling_map)) |
5962 | continue; | |
5963 | ||
dd41f596 IM |
5964 | init_sched_build_groups(this_sibling_map, cpu_map, |
5965 | &cpu_to_cpu_group); | |
1da177e4 LT |
5966 | } |
5967 | #endif | |
5968 | ||
1e9f28fa SS |
5969 | #ifdef CONFIG_SCHED_MC |
5970 | /* Set up multi-core groups */ | |
5971 | for_each_cpu_mask(i, *cpu_map) { | |
5972 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
5973 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
5974 | if (i != first_cpu(this_core_map)) | |
5975 | continue; | |
dd41f596 IM |
5976 | init_sched_build_groups(this_core_map, cpu_map, |
5977 | &cpu_to_core_group); | |
1e9f28fa SS |
5978 | } |
5979 | #endif | |
5980 | ||
1da177e4 LT |
5981 | /* Set up physical groups */ |
5982 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5983 | cpumask_t nodemask = node_to_cpumask(i); | |
5984 | ||
1a20ff27 | 5985 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5986 | if (cpus_empty(nodemask)) |
5987 | continue; | |
5988 | ||
6711cab4 | 5989 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
5990 | } |
5991 | ||
5992 | #ifdef CONFIG_NUMA | |
5993 | /* Set up node groups */ | |
6711cab4 | 5994 | if (sd_allnodes) |
dd41f596 IM |
5995 | init_sched_build_groups(*cpu_map, cpu_map, |
5996 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
5997 | |
5998 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5999 | /* Set up node groups */ | |
6000 | struct sched_group *sg, *prev; | |
6001 | cpumask_t nodemask = node_to_cpumask(i); | |
6002 | cpumask_t domainspan; | |
6003 | cpumask_t covered = CPU_MASK_NONE; | |
6004 | int j; | |
6005 | ||
6006 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6007 | if (cpus_empty(nodemask)) { |
6008 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6009 | continue; |
d1b55138 | 6010 | } |
9c1cfda2 JH |
6011 | |
6012 | domainspan = sched_domain_node_span(i); | |
6013 | cpus_and(domainspan, domainspan, *cpu_map); | |
6014 | ||
15f0b676 | 6015 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6016 | if (!sg) { |
6017 | printk(KERN_WARNING "Can not alloc domain group for " | |
6018 | "node %d\n", i); | |
6019 | goto error; | |
6020 | } | |
9c1cfda2 JH |
6021 | sched_group_nodes[i] = sg; |
6022 | for_each_cpu_mask(j, nodemask) { | |
6023 | struct sched_domain *sd; | |
9761eea8 | 6024 | |
9c1cfda2 JH |
6025 | sd = &per_cpu(node_domains, j); |
6026 | sd->groups = sg; | |
9c1cfda2 | 6027 | } |
5517d86b | 6028 | sg->__cpu_power = 0; |
9c1cfda2 | 6029 | sg->cpumask = nodemask; |
51888ca2 | 6030 | sg->next = sg; |
9c1cfda2 JH |
6031 | cpus_or(covered, covered, nodemask); |
6032 | prev = sg; | |
6033 | ||
6034 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6035 | cpumask_t tmp, notcovered; | |
6036 | int n = (i + j) % MAX_NUMNODES; | |
6037 | ||
6038 | cpus_complement(notcovered, covered); | |
6039 | cpus_and(tmp, notcovered, *cpu_map); | |
6040 | cpus_and(tmp, tmp, domainspan); | |
6041 | if (cpus_empty(tmp)) | |
6042 | break; | |
6043 | ||
6044 | nodemask = node_to_cpumask(n); | |
6045 | cpus_and(tmp, tmp, nodemask); | |
6046 | if (cpus_empty(tmp)) | |
6047 | continue; | |
6048 | ||
15f0b676 SV |
6049 | sg = kmalloc_node(sizeof(struct sched_group), |
6050 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6051 | if (!sg) { |
6052 | printk(KERN_WARNING | |
6053 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6054 | goto error; |
9c1cfda2 | 6055 | } |
5517d86b | 6056 | sg->__cpu_power = 0; |
9c1cfda2 | 6057 | sg->cpumask = tmp; |
51888ca2 | 6058 | sg->next = prev->next; |
9c1cfda2 JH |
6059 | cpus_or(covered, covered, tmp); |
6060 | prev->next = sg; | |
6061 | prev = sg; | |
6062 | } | |
9c1cfda2 | 6063 | } |
1da177e4 LT |
6064 | #endif |
6065 | ||
6066 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6067 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6068 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6069 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6070 | ||
89c4710e | 6071 | init_sched_groups_power(i, sd); |
5c45bf27 | 6072 | } |
1da177e4 | 6073 | #endif |
1e9f28fa | 6074 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6075 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6076 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6077 | ||
89c4710e | 6078 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6079 | } |
6080 | #endif | |
1e9f28fa | 6081 | |
5c45bf27 | 6082 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6083 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6084 | ||
89c4710e | 6085 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6086 | } |
6087 | ||
9c1cfda2 | 6088 | #ifdef CONFIG_NUMA |
08069033 SS |
6089 | for (i = 0; i < MAX_NUMNODES; i++) |
6090 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6091 | |
6711cab4 SS |
6092 | if (sd_allnodes) { |
6093 | struct sched_group *sg; | |
f712c0c7 | 6094 | |
6711cab4 | 6095 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6096 | init_numa_sched_groups_power(sg); |
6097 | } | |
9c1cfda2 JH |
6098 | #endif |
6099 | ||
1da177e4 | 6100 | /* Attach the domains */ |
1a20ff27 | 6101 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6102 | struct sched_domain *sd; |
6103 | #ifdef CONFIG_SCHED_SMT | |
6104 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6105 | #elif defined(CONFIG_SCHED_MC) |
6106 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6107 | #else |
6108 | sd = &per_cpu(phys_domains, i); | |
6109 | #endif | |
6110 | cpu_attach_domain(sd, i); | |
6111 | } | |
51888ca2 SV |
6112 | |
6113 | return 0; | |
6114 | ||
a616058b | 6115 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6116 | error: |
6117 | free_sched_groups(cpu_map); | |
6118 | return -ENOMEM; | |
a616058b | 6119 | #endif |
1da177e4 | 6120 | } |
1a20ff27 DG |
6121 | /* |
6122 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6123 | */ | |
51888ca2 | 6124 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6125 | { |
6126 | cpumask_t cpu_default_map; | |
51888ca2 | 6127 | int err; |
1da177e4 | 6128 | |
1a20ff27 DG |
6129 | /* |
6130 | * Setup mask for cpus without special case scheduling requirements. | |
6131 | * For now this just excludes isolated cpus, but could be used to | |
6132 | * exclude other special cases in the future. | |
6133 | */ | |
6134 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6135 | ||
51888ca2 SV |
6136 | err = build_sched_domains(&cpu_default_map); |
6137 | ||
6138 | return err; | |
1a20ff27 DG |
6139 | } |
6140 | ||
6141 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6142 | { |
51888ca2 | 6143 | free_sched_groups(cpu_map); |
9c1cfda2 | 6144 | } |
1da177e4 | 6145 | |
1a20ff27 DG |
6146 | /* |
6147 | * Detach sched domains from a group of cpus specified in cpu_map | |
6148 | * These cpus will now be attached to the NULL domain | |
6149 | */ | |
858119e1 | 6150 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6151 | { |
6152 | int i; | |
6153 | ||
6154 | for_each_cpu_mask(i, *cpu_map) | |
6155 | cpu_attach_domain(NULL, i); | |
6156 | synchronize_sched(); | |
6157 | arch_destroy_sched_domains(cpu_map); | |
6158 | } | |
6159 | ||
6160 | /* | |
6161 | * Partition sched domains as specified by the cpumasks below. | |
6162 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6163 | * waits for a RCU quiescent period, recalculates sched | |
6164 | * domain information and then attaches them back to the | |
6165 | * correct sched domains | |
6166 | * Call with hotplug lock held | |
6167 | */ | |
51888ca2 | 6168 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6169 | { |
6170 | cpumask_t change_map; | |
51888ca2 | 6171 | int err = 0; |
1a20ff27 DG |
6172 | |
6173 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6174 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6175 | cpus_or(change_map, *partition1, *partition2); | |
6176 | ||
6177 | /* Detach sched domains from all of the affected cpus */ | |
6178 | detach_destroy_domains(&change_map); | |
6179 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6180 | err = build_sched_domains(partition1); |
6181 | if (!err && !cpus_empty(*partition2)) | |
6182 | err = build_sched_domains(partition2); | |
6183 | ||
6184 | return err; | |
1a20ff27 DG |
6185 | } |
6186 | ||
5c45bf27 SS |
6187 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6188 | int arch_reinit_sched_domains(void) | |
6189 | { | |
6190 | int err; | |
6191 | ||
5be9361c | 6192 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6193 | detach_destroy_domains(&cpu_online_map); |
6194 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6195 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6196 | |
6197 | return err; | |
6198 | } | |
6199 | ||
6200 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6201 | { | |
6202 | int ret; | |
6203 | ||
6204 | if (buf[0] != '0' && buf[0] != '1') | |
6205 | return -EINVAL; | |
6206 | ||
6207 | if (smt) | |
6208 | sched_smt_power_savings = (buf[0] == '1'); | |
6209 | else | |
6210 | sched_mc_power_savings = (buf[0] == '1'); | |
6211 | ||
6212 | ret = arch_reinit_sched_domains(); | |
6213 | ||
6214 | return ret ? ret : count; | |
6215 | } | |
6216 | ||
6217 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6218 | { | |
6219 | int err = 0; | |
48f24c4d | 6220 | |
5c45bf27 SS |
6221 | #ifdef CONFIG_SCHED_SMT |
6222 | if (smt_capable()) | |
6223 | err = sysfs_create_file(&cls->kset.kobj, | |
6224 | &attr_sched_smt_power_savings.attr); | |
6225 | #endif | |
6226 | #ifdef CONFIG_SCHED_MC | |
6227 | if (!err && mc_capable()) | |
6228 | err = sysfs_create_file(&cls->kset.kobj, | |
6229 | &attr_sched_mc_power_savings.attr); | |
6230 | #endif | |
6231 | return err; | |
6232 | } | |
6233 | #endif | |
6234 | ||
6235 | #ifdef CONFIG_SCHED_MC | |
6236 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6237 | { | |
6238 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6239 | } | |
48f24c4d IM |
6240 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6241 | const char *buf, size_t count) | |
5c45bf27 SS |
6242 | { |
6243 | return sched_power_savings_store(buf, count, 0); | |
6244 | } | |
6245 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6246 | sched_mc_power_savings_store); | |
6247 | #endif | |
6248 | ||
6249 | #ifdef CONFIG_SCHED_SMT | |
6250 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6251 | { | |
6252 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6253 | } | |
48f24c4d IM |
6254 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6255 | const char *buf, size_t count) | |
5c45bf27 SS |
6256 | { |
6257 | return sched_power_savings_store(buf, count, 1); | |
6258 | } | |
6259 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6260 | sched_smt_power_savings_store); | |
6261 | #endif | |
6262 | ||
1da177e4 LT |
6263 | /* |
6264 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6265 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6266 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6267 | * which will prevent rebalancing while the sched domains are recalculated. |
6268 | */ | |
6269 | static int update_sched_domains(struct notifier_block *nfb, | |
6270 | unsigned long action, void *hcpu) | |
6271 | { | |
1da177e4 LT |
6272 | switch (action) { |
6273 | case CPU_UP_PREPARE: | |
8bb78442 | 6274 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6275 | case CPU_DOWN_PREPARE: |
8bb78442 | 6276 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6277 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6278 | return NOTIFY_OK; |
6279 | ||
6280 | case CPU_UP_CANCELED: | |
8bb78442 | 6281 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6282 | case CPU_DOWN_FAILED: |
8bb78442 | 6283 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6284 | case CPU_ONLINE: |
8bb78442 | 6285 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6286 | case CPU_DEAD: |
8bb78442 | 6287 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6288 | /* |
6289 | * Fall through and re-initialise the domains. | |
6290 | */ | |
6291 | break; | |
6292 | default: | |
6293 | return NOTIFY_DONE; | |
6294 | } | |
6295 | ||
6296 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6297 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6298 | |
6299 | return NOTIFY_OK; | |
6300 | } | |
1da177e4 LT |
6301 | |
6302 | void __init sched_init_smp(void) | |
6303 | { | |
5c1e1767 NP |
6304 | cpumask_t non_isolated_cpus; |
6305 | ||
5be9361c | 6306 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6307 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6308 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6309 | if (cpus_empty(non_isolated_cpus)) |
6310 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6311 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6312 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6313 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6314 | |
6315 | /* Move init over to a non-isolated CPU */ | |
6316 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6317 | BUG(); | |
dd41f596 | 6318 | sched_init_granularity(); |
1da177e4 LT |
6319 | } |
6320 | #else | |
6321 | void __init sched_init_smp(void) | |
6322 | { | |
dd41f596 | 6323 | sched_init_granularity(); |
1da177e4 LT |
6324 | } |
6325 | #endif /* CONFIG_SMP */ | |
6326 | ||
6327 | int in_sched_functions(unsigned long addr) | |
6328 | { | |
6329 | /* Linker adds these: start and end of __sched functions */ | |
6330 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6331 | |
1da177e4 LT |
6332 | return in_lock_functions(addr) || |
6333 | (addr >= (unsigned long)__sched_text_start | |
6334 | && addr < (unsigned long)__sched_text_end); | |
6335 | } | |
6336 | ||
dd41f596 IM |
6337 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
6338 | { | |
6339 | cfs_rq->tasks_timeline = RB_ROOT; | |
6340 | cfs_rq->fair_clock = 1; | |
6341 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6342 | cfs_rq->rq = rq; | |
6343 | #endif | |
6344 | } | |
6345 | ||
1da177e4 LT |
6346 | void __init sched_init(void) |
6347 | { | |
dd41f596 | 6348 | u64 now = sched_clock(); |
476f3534 | 6349 | int highest_cpu = 0; |
dd41f596 IM |
6350 | int i, j; |
6351 | ||
6352 | /* | |
6353 | * Link up the scheduling class hierarchy: | |
6354 | */ | |
6355 | rt_sched_class.next = &fair_sched_class; | |
6356 | fair_sched_class.next = &idle_sched_class; | |
6357 | idle_sched_class.next = NULL; | |
1da177e4 | 6358 | |
0a945022 | 6359 | for_each_possible_cpu(i) { |
dd41f596 | 6360 | struct rt_prio_array *array; |
70b97a7f | 6361 | struct rq *rq; |
1da177e4 LT |
6362 | |
6363 | rq = cpu_rq(i); | |
6364 | spin_lock_init(&rq->lock); | |
fcb99371 | 6365 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6366 | rq->nr_running = 0; |
dd41f596 IM |
6367 | rq->clock = 1; |
6368 | init_cfs_rq(&rq->cfs, rq); | |
6369 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6370 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
6371 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6372 | #endif | |
6373 | rq->ls.load_update_last = now; | |
6374 | rq->ls.load_update_start = now; | |
1da177e4 | 6375 | |
dd41f596 IM |
6376 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6377 | rq->cpu_load[j] = 0; | |
1da177e4 | 6378 | #ifdef CONFIG_SMP |
41c7ce9a | 6379 | rq->sd = NULL; |
1da177e4 | 6380 | rq->active_balance = 0; |
dd41f596 | 6381 | rq->next_balance = jiffies; |
1da177e4 | 6382 | rq->push_cpu = 0; |
0a2966b4 | 6383 | rq->cpu = i; |
1da177e4 LT |
6384 | rq->migration_thread = NULL; |
6385 | INIT_LIST_HEAD(&rq->migration_queue); | |
6386 | #endif | |
6387 | atomic_set(&rq->nr_iowait, 0); | |
6388 | ||
dd41f596 IM |
6389 | array = &rq->rt.active; |
6390 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6391 | INIT_LIST_HEAD(array->queue + j); | |
6392 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6393 | } |
476f3534 | 6394 | highest_cpu = i; |
dd41f596 IM |
6395 | /* delimiter for bitsearch: */ |
6396 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6397 | } |
6398 | ||
2dd73a4f | 6399 | set_load_weight(&init_task); |
b50f60ce | 6400 | |
e107be36 AK |
6401 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6402 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6403 | #endif | |
6404 | ||
c9819f45 | 6405 | #ifdef CONFIG_SMP |
476f3534 | 6406 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6407 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6408 | #endif | |
6409 | ||
b50f60ce HC |
6410 | #ifdef CONFIG_RT_MUTEXES |
6411 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6412 | #endif | |
6413 | ||
1da177e4 LT |
6414 | /* |
6415 | * The boot idle thread does lazy MMU switching as well: | |
6416 | */ | |
6417 | atomic_inc(&init_mm.mm_count); | |
6418 | enter_lazy_tlb(&init_mm, current); | |
6419 | ||
6420 | /* | |
6421 | * Make us the idle thread. Technically, schedule() should not be | |
6422 | * called from this thread, however somewhere below it might be, | |
6423 | * but because we are the idle thread, we just pick up running again | |
6424 | * when this runqueue becomes "idle". | |
6425 | */ | |
6426 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6427 | /* |
6428 | * During early bootup we pretend to be a normal task: | |
6429 | */ | |
6430 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6431 | } |
6432 | ||
6433 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6434 | void __might_sleep(char *file, int line) | |
6435 | { | |
48f24c4d | 6436 | #ifdef in_atomic |
1da177e4 LT |
6437 | static unsigned long prev_jiffy; /* ratelimiting */ |
6438 | ||
6439 | if ((in_atomic() || irqs_disabled()) && | |
6440 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6441 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6442 | return; | |
6443 | prev_jiffy = jiffies; | |
91368d73 | 6444 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6445 | " context at %s:%d\n", file, line); |
6446 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6447 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6448 | debug_show_held_locks(current); |
3117df04 IM |
6449 | if (irqs_disabled()) |
6450 | print_irqtrace_events(current); | |
1da177e4 LT |
6451 | dump_stack(); |
6452 | } | |
6453 | #endif | |
6454 | } | |
6455 | EXPORT_SYMBOL(__might_sleep); | |
6456 | #endif | |
6457 | ||
6458 | #ifdef CONFIG_MAGIC_SYSRQ | |
6459 | void normalize_rt_tasks(void) | |
6460 | { | |
a0f98a1c | 6461 | struct task_struct *g, *p; |
1da177e4 | 6462 | unsigned long flags; |
70b97a7f | 6463 | struct rq *rq; |
dd41f596 | 6464 | int on_rq; |
1da177e4 LT |
6465 | |
6466 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6467 | do_each_thread(g, p) { |
dd41f596 IM |
6468 | p->se.fair_key = 0; |
6469 | p->se.wait_runtime = 0; | |
6470 | p->se.wait_start_fair = 0; | |
6471 | p->se.wait_start = 0; | |
6472 | p->se.exec_start = 0; | |
6473 | p->se.sleep_start = 0; | |
6474 | p->se.sleep_start_fair = 0; | |
6475 | p->se.block_start = 0; | |
6476 | task_rq(p)->cfs.fair_clock = 0; | |
6477 | task_rq(p)->clock = 0; | |
6478 | ||
6479 | if (!rt_task(p)) { | |
6480 | /* | |
6481 | * Renice negative nice level userspace | |
6482 | * tasks back to 0: | |
6483 | */ | |
6484 | if (TASK_NICE(p) < 0 && p->mm) | |
6485 | set_user_nice(p, 0); | |
1da177e4 | 6486 | continue; |
dd41f596 | 6487 | } |
1da177e4 | 6488 | |
b29739f9 IM |
6489 | spin_lock_irqsave(&p->pi_lock, flags); |
6490 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6491 | #ifdef CONFIG_SMP |
6492 | /* | |
6493 | * Do not touch the migration thread: | |
6494 | */ | |
6495 | if (p == rq->migration_thread) | |
6496 | goto out_unlock; | |
6497 | #endif | |
1da177e4 | 6498 | |
dd41f596 IM |
6499 | on_rq = p->se.on_rq; |
6500 | if (on_rq) | |
6501 | deactivate_task(task_rq(p), p, 0); | |
6502 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6503 | if (on_rq) { | |
6504 | activate_task(task_rq(p), p, 0); | |
1da177e4 LT |
6505 | resched_task(rq->curr); |
6506 | } | |
dd41f596 IM |
6507 | #ifdef CONFIG_SMP |
6508 | out_unlock: | |
6509 | #endif | |
b29739f9 IM |
6510 | __task_rq_unlock(rq); |
6511 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6512 | } while_each_thread(g, p); |
6513 | ||
1da177e4 LT |
6514 | read_unlock_irq(&tasklist_lock); |
6515 | } | |
6516 | ||
6517 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6518 | |
6519 | #ifdef CONFIG_IA64 | |
6520 | /* | |
6521 | * These functions are only useful for the IA64 MCA handling. | |
6522 | * | |
6523 | * They can only be called when the whole system has been | |
6524 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6525 | * activity can take place. Using them for anything else would | |
6526 | * be a serious bug, and as a result, they aren't even visible | |
6527 | * under any other configuration. | |
6528 | */ | |
6529 | ||
6530 | /** | |
6531 | * curr_task - return the current task for a given cpu. | |
6532 | * @cpu: the processor in question. | |
6533 | * | |
6534 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6535 | */ | |
36c8b586 | 6536 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6537 | { |
6538 | return cpu_curr(cpu); | |
6539 | } | |
6540 | ||
6541 | /** | |
6542 | * set_curr_task - set the current task for a given cpu. | |
6543 | * @cpu: the processor in question. | |
6544 | * @p: the task pointer to set. | |
6545 | * | |
6546 | * Description: This function must only be used when non-maskable interrupts | |
6547 | * are serviced on a separate stack. It allows the architecture to switch the | |
6548 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6549 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6550 | * and caller must save the original value of the current task (see | |
6551 | * curr_task() above) and restore that value before reenabling interrupts and | |
6552 | * re-starting the system. | |
6553 | * | |
6554 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6555 | */ | |
36c8b586 | 6556 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6557 | { |
6558 | cpu_curr(cpu) = p; | |
6559 | } | |
6560 | ||
6561 | #endif |