]>
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 | ||
266 | struct sched_class *load_balance_class; | |
267 | ||
1da177e4 LT |
268 | atomic_t nr_iowait; |
269 | ||
270 | #ifdef CONFIG_SMP | |
271 | struct sched_domain *sd; | |
272 | ||
273 | /* For active balancing */ | |
274 | int active_balance; | |
275 | int push_cpu; | |
0a2966b4 | 276 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 277 | |
36c8b586 | 278 | struct task_struct *migration_thread; |
1da177e4 LT |
279 | struct list_head migration_queue; |
280 | #endif | |
281 | ||
282 | #ifdef CONFIG_SCHEDSTATS | |
283 | /* latency stats */ | |
284 | struct sched_info rq_sched_info; | |
285 | ||
286 | /* sys_sched_yield() stats */ | |
287 | unsigned long yld_exp_empty; | |
288 | unsigned long yld_act_empty; | |
289 | unsigned long yld_both_empty; | |
290 | unsigned long yld_cnt; | |
291 | ||
292 | /* schedule() stats */ | |
293 | unsigned long sched_switch; | |
294 | unsigned long sched_cnt; | |
295 | unsigned long sched_goidle; | |
296 | ||
297 | /* try_to_wake_up() stats */ | |
298 | unsigned long ttwu_cnt; | |
299 | unsigned long ttwu_local; | |
300 | #endif | |
fcb99371 | 301 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
302 | }; |
303 | ||
c3396620 | 304 | static DEFINE_PER_CPU(struct rq, runqueues) ____cacheline_aligned_in_smp; |
5be9361c | 305 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 306 | |
dd41f596 IM |
307 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
308 | { | |
309 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
310 | } | |
311 | ||
0a2966b4 CL |
312 | static inline int cpu_of(struct rq *rq) |
313 | { | |
314 | #ifdef CONFIG_SMP | |
315 | return rq->cpu; | |
316 | #else | |
317 | return 0; | |
318 | #endif | |
319 | } | |
320 | ||
20d315d4 IM |
321 | /* |
322 | * Per-runqueue clock, as finegrained as the platform can give us: | |
323 | */ | |
324 | static unsigned long long __rq_clock(struct rq *rq) | |
325 | { | |
326 | u64 prev_raw = rq->prev_clock_raw; | |
327 | u64 now = sched_clock(); | |
328 | s64 delta = now - prev_raw; | |
329 | u64 clock = rq->clock; | |
330 | ||
331 | /* | |
332 | * Protect against sched_clock() occasionally going backwards: | |
333 | */ | |
334 | if (unlikely(delta < 0)) { | |
335 | clock++; | |
336 | rq->clock_warps++; | |
337 | } else { | |
338 | /* | |
339 | * Catch too large forward jumps too: | |
340 | */ | |
341 | if (unlikely(delta > 2*TICK_NSEC)) { | |
342 | clock++; | |
343 | rq->clock_overflows++; | |
344 | } else { | |
345 | if (unlikely(delta > rq->clock_max_delta)) | |
346 | rq->clock_max_delta = delta; | |
347 | clock += delta; | |
348 | } | |
349 | } | |
350 | ||
351 | rq->prev_clock_raw = now; | |
352 | rq->clock = clock; | |
353 | ||
354 | return clock; | |
355 | } | |
356 | ||
357 | static inline unsigned long long rq_clock(struct rq *rq) | |
358 | { | |
359 | int this_cpu = smp_processor_id(); | |
360 | ||
361 | if (this_cpu == cpu_of(rq)) | |
362 | return __rq_clock(rq); | |
363 | ||
364 | return rq->clock; | |
365 | } | |
366 | ||
674311d5 NP |
367 | /* |
368 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 369 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
370 | * |
371 | * The domain tree of any CPU may only be accessed from within | |
372 | * preempt-disabled sections. | |
373 | */ | |
48f24c4d IM |
374 | #define for_each_domain(cpu, __sd) \ |
375 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
376 | |
377 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
378 | #define this_rq() (&__get_cpu_var(runqueues)) | |
379 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
380 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
381 | ||
138a8aeb IM |
382 | #ifdef CONFIG_FAIR_GROUP_SCHED |
383 | /* Change a task's ->cfs_rq if it moves across CPUs */ | |
384 | static inline void set_task_cfs_rq(struct task_struct *p) | |
385 | { | |
386 | p->se.cfs_rq = &task_rq(p)->cfs; | |
387 | } | |
388 | #else | |
389 | static inline void set_task_cfs_rq(struct task_struct *p) | |
390 | { | |
391 | } | |
392 | #endif | |
393 | ||
1da177e4 | 394 | #ifndef prepare_arch_switch |
4866cde0 NP |
395 | # define prepare_arch_switch(next) do { } while (0) |
396 | #endif | |
397 | #ifndef finish_arch_switch | |
398 | # define finish_arch_switch(prev) do { } while (0) | |
399 | #endif | |
400 | ||
401 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 402 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
403 | { |
404 | return rq->curr == p; | |
405 | } | |
406 | ||
70b97a7f | 407 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
408 | { |
409 | } | |
410 | ||
70b97a7f | 411 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 412 | { |
da04c035 IM |
413 | #ifdef CONFIG_DEBUG_SPINLOCK |
414 | /* this is a valid case when another task releases the spinlock */ | |
415 | rq->lock.owner = current; | |
416 | #endif | |
8a25d5de IM |
417 | /* |
418 | * If we are tracking spinlock dependencies then we have to | |
419 | * fix up the runqueue lock - which gets 'carried over' from | |
420 | * prev into current: | |
421 | */ | |
422 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
423 | ||
4866cde0 NP |
424 | spin_unlock_irq(&rq->lock); |
425 | } | |
426 | ||
427 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 428 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
429 | { |
430 | #ifdef CONFIG_SMP | |
431 | return p->oncpu; | |
432 | #else | |
433 | return rq->curr == p; | |
434 | #endif | |
435 | } | |
436 | ||
70b97a7f | 437 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
438 | { |
439 | #ifdef CONFIG_SMP | |
440 | /* | |
441 | * We can optimise this out completely for !SMP, because the | |
442 | * SMP rebalancing from interrupt is the only thing that cares | |
443 | * here. | |
444 | */ | |
445 | next->oncpu = 1; | |
446 | #endif | |
447 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
448 | spin_unlock_irq(&rq->lock); | |
449 | #else | |
450 | spin_unlock(&rq->lock); | |
451 | #endif | |
452 | } | |
453 | ||
70b97a7f | 454 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
455 | { |
456 | #ifdef CONFIG_SMP | |
457 | /* | |
458 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
459 | * We must ensure this doesn't happen until the switch is completely | |
460 | * finished. | |
461 | */ | |
462 | smp_wmb(); | |
463 | prev->oncpu = 0; | |
464 | #endif | |
465 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
466 | local_irq_enable(); | |
1da177e4 | 467 | #endif |
4866cde0 NP |
468 | } |
469 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 470 | |
b29739f9 IM |
471 | /* |
472 | * __task_rq_lock - lock the runqueue a given task resides on. | |
473 | * Must be called interrupts disabled. | |
474 | */ | |
70b97a7f | 475 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
476 | __acquires(rq->lock) |
477 | { | |
70b97a7f | 478 | struct rq *rq; |
b29739f9 IM |
479 | |
480 | repeat_lock_task: | |
481 | rq = task_rq(p); | |
482 | spin_lock(&rq->lock); | |
483 | if (unlikely(rq != task_rq(p))) { | |
484 | spin_unlock(&rq->lock); | |
485 | goto repeat_lock_task; | |
486 | } | |
487 | return rq; | |
488 | } | |
489 | ||
1da177e4 LT |
490 | /* |
491 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
492 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
493 | * explicitly disabling preemption. | |
494 | */ | |
70b97a7f | 495 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
496 | __acquires(rq->lock) |
497 | { | |
70b97a7f | 498 | struct rq *rq; |
1da177e4 LT |
499 | |
500 | repeat_lock_task: | |
501 | local_irq_save(*flags); | |
502 | rq = task_rq(p); | |
503 | spin_lock(&rq->lock); | |
504 | if (unlikely(rq != task_rq(p))) { | |
505 | spin_unlock_irqrestore(&rq->lock, *flags); | |
506 | goto repeat_lock_task; | |
507 | } | |
508 | return rq; | |
509 | } | |
510 | ||
70b97a7f | 511 | static inline void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
512 | __releases(rq->lock) |
513 | { | |
514 | spin_unlock(&rq->lock); | |
515 | } | |
516 | ||
70b97a7f | 517 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
518 | __releases(rq->lock) |
519 | { | |
520 | spin_unlock_irqrestore(&rq->lock, *flags); | |
521 | } | |
522 | ||
1da177e4 | 523 | /* |
cc2a73b5 | 524 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 525 | */ |
70b97a7f | 526 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
527 | __acquires(rq->lock) |
528 | { | |
70b97a7f | 529 | struct rq *rq; |
1da177e4 LT |
530 | |
531 | local_irq_disable(); | |
532 | rq = this_rq(); | |
533 | spin_lock(&rq->lock); | |
534 | ||
535 | return rq; | |
536 | } | |
537 | ||
1b9f19c2 IM |
538 | /* |
539 | * CPU frequency is/was unstable - start new by setting prev_clock_raw: | |
540 | */ | |
541 | void sched_clock_unstable_event(void) | |
542 | { | |
543 | unsigned long flags; | |
544 | struct rq *rq; | |
545 | ||
546 | rq = task_rq_lock(current, &flags); | |
547 | rq->prev_clock_raw = sched_clock(); | |
548 | rq->clock_unstable_events++; | |
549 | task_rq_unlock(rq, &flags); | |
550 | } | |
551 | ||
c24d20db IM |
552 | /* |
553 | * resched_task - mark a task 'to be rescheduled now'. | |
554 | * | |
555 | * On UP this means the setting of the need_resched flag, on SMP it | |
556 | * might also involve a cross-CPU call to trigger the scheduler on | |
557 | * the target CPU. | |
558 | */ | |
559 | #ifdef CONFIG_SMP | |
560 | ||
561 | #ifndef tsk_is_polling | |
562 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
563 | #endif | |
564 | ||
565 | static void resched_task(struct task_struct *p) | |
566 | { | |
567 | int cpu; | |
568 | ||
569 | assert_spin_locked(&task_rq(p)->lock); | |
570 | ||
571 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
572 | return; | |
573 | ||
574 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
575 | ||
576 | cpu = task_cpu(p); | |
577 | if (cpu == smp_processor_id()) | |
578 | return; | |
579 | ||
580 | /* NEED_RESCHED must be visible before we test polling */ | |
581 | smp_mb(); | |
582 | if (!tsk_is_polling(p)) | |
583 | smp_send_reschedule(cpu); | |
584 | } | |
585 | ||
586 | static void resched_cpu(int cpu) | |
587 | { | |
588 | struct rq *rq = cpu_rq(cpu); | |
589 | unsigned long flags; | |
590 | ||
591 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
592 | return; | |
593 | resched_task(cpu_curr(cpu)); | |
594 | spin_unlock_irqrestore(&rq->lock, flags); | |
595 | } | |
596 | #else | |
597 | static inline void resched_task(struct task_struct *p) | |
598 | { | |
599 | assert_spin_locked(&task_rq(p)->lock); | |
600 | set_tsk_need_resched(p); | |
601 | } | |
602 | #endif | |
603 | ||
45bf76df IM |
604 | static u64 div64_likely32(u64 divident, unsigned long divisor) |
605 | { | |
606 | #if BITS_PER_LONG == 32 | |
607 | if (likely(divident <= 0xffffffffULL)) | |
608 | return (u32)divident / divisor; | |
609 | do_div(divident, divisor); | |
610 | ||
611 | return divident; | |
612 | #else | |
613 | return divident / divisor; | |
614 | #endif | |
615 | } | |
616 | ||
617 | #if BITS_PER_LONG == 32 | |
618 | # define WMULT_CONST (~0UL) | |
619 | #else | |
620 | # define WMULT_CONST (1UL << 32) | |
621 | #endif | |
622 | ||
623 | #define WMULT_SHIFT 32 | |
624 | ||
625 | static inline unsigned long | |
626 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, | |
627 | struct load_weight *lw) | |
628 | { | |
629 | u64 tmp; | |
630 | ||
631 | if (unlikely(!lw->inv_weight)) | |
632 | lw->inv_weight = WMULT_CONST / lw->weight; | |
633 | ||
634 | tmp = (u64)delta_exec * weight; | |
635 | /* | |
636 | * Check whether we'd overflow the 64-bit multiplication: | |
637 | */ | |
638 | if (unlikely(tmp > WMULT_CONST)) { | |
639 | tmp = ((tmp >> WMULT_SHIFT/2) * lw->inv_weight) | |
640 | >> (WMULT_SHIFT/2); | |
641 | } else { | |
642 | tmp = (tmp * lw->inv_weight) >> WMULT_SHIFT; | |
643 | } | |
644 | ||
645 | return (unsigned long)min(tmp, (u64)sysctl_sched_runtime_limit); | |
646 | } | |
647 | ||
648 | static inline unsigned long | |
649 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
650 | { | |
651 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
652 | } | |
653 | ||
654 | static void update_load_add(struct load_weight *lw, unsigned long inc) | |
655 | { | |
656 | lw->weight += inc; | |
657 | lw->inv_weight = 0; | |
658 | } | |
659 | ||
660 | static void update_load_sub(struct load_weight *lw, unsigned long dec) | |
661 | { | |
662 | lw->weight -= dec; | |
663 | lw->inv_weight = 0; | |
664 | } | |
665 | ||
666 | static void __update_curr_load(struct rq *rq, struct load_stat *ls) | |
667 | { | |
668 | if (rq->curr != rq->idle && ls->load.weight) { | |
669 | ls->delta_exec += ls->delta_stat; | |
670 | ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load); | |
671 | ls->delta_stat = 0; | |
672 | } | |
673 | } | |
674 | ||
675 | /* | |
676 | * Update delta_exec, delta_fair fields for rq. | |
677 | * | |
678 | * delta_fair clock advances at a rate inversely proportional to | |
679 | * total load (rq->ls.load.weight) on the runqueue, while | |
680 | * delta_exec advances at the same rate as wall-clock (provided | |
681 | * cpu is not idle). | |
682 | * | |
683 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
684 | * runqueue over any given interval. This (smoothened) load is used | |
685 | * during load balance. | |
686 | * | |
687 | * This function is called /before/ updating rq->ls.load | |
688 | * and when switching tasks. | |
689 | */ | |
690 | static void update_curr_load(struct rq *rq, u64 now) | |
691 | { | |
692 | struct load_stat *ls = &rq->ls; | |
693 | u64 start; | |
694 | ||
695 | start = ls->load_update_start; | |
696 | ls->load_update_start = now; | |
697 | ls->delta_stat += now - start; | |
698 | /* | |
699 | * Stagger updates to ls->delta_fair. Very frequent updates | |
700 | * can be expensive. | |
701 | */ | |
702 | if (ls->delta_stat >= sysctl_sched_stat_granularity) | |
703 | __update_curr_load(rq, ls); | |
704 | } | |
705 | ||
2dd73a4f PW |
706 | /* |
707 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
708 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
709 | * each task makes to its run queue's load is weighted according to its | |
710 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
711 | * scaled version of the new time slice allocation that they receive on time | |
712 | * slice expiry etc. | |
713 | */ | |
714 | ||
715 | /* | |
716 | * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE | |
717 | * If static_prio_timeslice() is ever changed to break this assumption then | |
718 | * this code will need modification | |
719 | */ | |
720 | #define TIME_SLICE_NICE_ZERO DEF_TIMESLICE | |
dd41f596 | 721 | #define load_weight(lp) \ |
2dd73a4f PW |
722 | (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO) |
723 | #define PRIO_TO_LOAD_WEIGHT(prio) \ | |
dd41f596 | 724 | load_weight(static_prio_timeslice(prio)) |
2dd73a4f | 725 | #define RTPRIO_TO_LOAD_WEIGHT(rp) \ |
dd41f596 IM |
726 | (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + load_weight(rp)) |
727 | ||
728 | #define WEIGHT_IDLEPRIO 2 | |
729 | #define WMULT_IDLEPRIO (1 << 31) | |
730 | ||
731 | /* | |
732 | * Nice levels are multiplicative, with a gentle 10% change for every | |
733 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
734 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
735 | * that remained on nice 0. | |
736 | * | |
737 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
738 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
739 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
740 | * If a task goes up by ~10% and another task goes down by ~10% then | |
741 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
742 | */ |
743 | static const int prio_to_weight[40] = { | |
744 | /* -20 */ 88818, 71054, 56843, 45475, 36380, 29104, 23283, 18626, 14901, 11921, | |
745 | /* -10 */ 9537, 7629, 6103, 4883, 3906, 3125, 2500, 2000, 1600, 1280, | |
746 | /* 0 */ NICE_0_LOAD /* 1024 */, | |
747 | /* 1 */ 819, 655, 524, 419, 336, 268, 215, 172, 137, | |
748 | /* 10 */ 110, 87, 70, 56, 45, 36, 29, 23, 18, 15, | |
749 | }; | |
750 | ||
5714d2de IM |
751 | /* |
752 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
753 | * | |
754 | * In cases where the weight does not change often, we can use the | |
755 | * precalculated inverse to speed up arithmetics by turning divisions | |
756 | * into multiplications: | |
757 | */ | |
dd41f596 | 758 | static const u32 prio_to_wmult[40] = { |
e4af30be IM |
759 | /* -20 */ 48356, 60446, 75558, 94446, 118058, |
760 | /* -15 */ 147573, 184467, 230589, 288233, 360285, | |
761 | /* -10 */ 450347, 562979, 703746, 879575, 1099582, | |
762 | /* -5 */ 1374389, 1717986, 2147483, 2684354, 3355443, | |
763 | /* 0 */ 4194304, 5244160, 6557201, 8196502, 10250518, | |
764 | /* 5 */ 12782640, 16025997, 19976592, 24970740, 31350126, | |
765 | /* 10 */ 39045157, 49367440, 61356675, 76695844, 95443717, | |
766 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 767 | }; |
2dd73a4f | 768 | |
36c8b586 | 769 | static inline void |
dd41f596 | 770 | inc_load(struct rq *rq, const struct task_struct *p, u64 now) |
2dd73a4f | 771 | { |
dd41f596 IM |
772 | update_curr_load(rq, now); |
773 | update_load_add(&rq->ls.load, p->se.load.weight); | |
2dd73a4f PW |
774 | } |
775 | ||
36c8b586 | 776 | static inline void |
dd41f596 | 777 | dec_load(struct rq *rq, const struct task_struct *p, u64 now) |
2dd73a4f | 778 | { |
dd41f596 IM |
779 | update_curr_load(rq, now); |
780 | update_load_sub(&rq->ls.load, p->se.load.weight); | |
2dd73a4f PW |
781 | } |
782 | ||
dd41f596 | 783 | static inline void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now) |
2dd73a4f PW |
784 | { |
785 | rq->nr_running++; | |
dd41f596 | 786 | inc_load(rq, p, now); |
2dd73a4f PW |
787 | } |
788 | ||
dd41f596 | 789 | static inline void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now) |
2dd73a4f PW |
790 | { |
791 | rq->nr_running--; | |
dd41f596 | 792 | dec_load(rq, p, now); |
2dd73a4f PW |
793 | } |
794 | ||
dd41f596 IM |
795 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
796 | ||
797 | /* | |
798 | * runqueue iterator, to support SMP load-balancing between different | |
799 | * scheduling classes, without having to expose their internal data | |
800 | * structures to the load-balancing proper: | |
801 | */ | |
802 | struct rq_iterator { | |
803 | void *arg; | |
804 | struct task_struct *(*start)(void *); | |
805 | struct task_struct *(*next)(void *); | |
806 | }; | |
807 | ||
808 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
809 | unsigned long max_nr_move, unsigned long max_load_move, | |
810 | struct sched_domain *sd, enum cpu_idle_type idle, | |
811 | int *all_pinned, unsigned long *load_moved, | |
812 | int this_best_prio, int best_prio, int best_prio_seen, | |
813 | struct rq_iterator *iterator); | |
814 | ||
815 | #include "sched_stats.h" | |
816 | #include "sched_rt.c" | |
817 | #include "sched_fair.c" | |
818 | #include "sched_idletask.c" | |
819 | #ifdef CONFIG_SCHED_DEBUG | |
820 | # include "sched_debug.c" | |
821 | #endif | |
822 | ||
823 | #define sched_class_highest (&rt_sched_class) | |
824 | ||
45bf76df IM |
825 | static void set_load_weight(struct task_struct *p) |
826 | { | |
dd41f596 IM |
827 | task_rq(p)->cfs.wait_runtime -= p->se.wait_runtime; |
828 | p->se.wait_runtime = 0; | |
829 | ||
45bf76df | 830 | if (task_has_rt_policy(p)) { |
dd41f596 IM |
831 | p->se.load.weight = prio_to_weight[0] * 2; |
832 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
833 | return; | |
834 | } | |
45bf76df | 835 | |
dd41f596 IM |
836 | /* |
837 | * SCHED_IDLE tasks get minimal weight: | |
838 | */ | |
839 | if (p->policy == SCHED_IDLE) { | |
840 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
841 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
842 | return; | |
843 | } | |
71f8bd46 | 844 | |
dd41f596 IM |
845 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
846 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
847 | } |
848 | ||
dd41f596 IM |
849 | static void |
850 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now) | |
71f8bd46 | 851 | { |
dd41f596 IM |
852 | sched_info_queued(p); |
853 | p->sched_class->enqueue_task(rq, p, wakeup, now); | |
854 | p->se.on_rq = 1; | |
71f8bd46 IM |
855 | } |
856 | ||
dd41f596 IM |
857 | static void |
858 | dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now) | |
71f8bd46 | 859 | { |
dd41f596 IM |
860 | p->sched_class->dequeue_task(rq, p, sleep, now); |
861 | p->se.on_rq = 0; | |
71f8bd46 IM |
862 | } |
863 | ||
14531189 | 864 | /* |
dd41f596 | 865 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 866 | */ |
14531189 IM |
867 | static inline int __normal_prio(struct task_struct *p) |
868 | { | |
dd41f596 | 869 | return p->static_prio; |
14531189 IM |
870 | } |
871 | ||
b29739f9 IM |
872 | /* |
873 | * Calculate the expected normal priority: i.e. priority | |
874 | * without taking RT-inheritance into account. Might be | |
875 | * boosted by interactivity modifiers. Changes upon fork, | |
876 | * setprio syscalls, and whenever the interactivity | |
877 | * estimator recalculates. | |
878 | */ | |
36c8b586 | 879 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
880 | { |
881 | int prio; | |
882 | ||
e05606d3 | 883 | if (task_has_rt_policy(p)) |
b29739f9 IM |
884 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
885 | else | |
886 | prio = __normal_prio(p); | |
887 | return prio; | |
888 | } | |
889 | ||
890 | /* | |
891 | * Calculate the current priority, i.e. the priority | |
892 | * taken into account by the scheduler. This value might | |
893 | * be boosted by RT tasks, or might be boosted by | |
894 | * interactivity modifiers. Will be RT if the task got | |
895 | * RT-boosted. If not then it returns p->normal_prio. | |
896 | */ | |
36c8b586 | 897 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
898 | { |
899 | p->normal_prio = normal_prio(p); | |
900 | /* | |
901 | * If we are RT tasks or we were boosted to RT priority, | |
902 | * keep the priority unchanged. Otherwise, update priority | |
903 | * to the normal priority: | |
904 | */ | |
905 | if (!rt_prio(p->prio)) | |
906 | return p->normal_prio; | |
907 | return p->prio; | |
908 | } | |
909 | ||
1da177e4 | 910 | /* |
dd41f596 | 911 | * activate_task - move a task to the runqueue. |
1da177e4 | 912 | */ |
dd41f596 | 913 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 914 | { |
dd41f596 | 915 | u64 now = rq_clock(rq); |
d425b274 | 916 | |
dd41f596 IM |
917 | if (p->state == TASK_UNINTERRUPTIBLE) |
918 | rq->nr_uninterruptible--; | |
1da177e4 | 919 | |
dd41f596 IM |
920 | enqueue_task(rq, p, wakeup, now); |
921 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
922 | } |
923 | ||
924 | /* | |
dd41f596 | 925 | * activate_idle_task - move idle task to the _front_ of runqueue. |
1da177e4 | 926 | */ |
dd41f596 | 927 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 928 | { |
dd41f596 | 929 | u64 now = rq_clock(rq); |
1da177e4 | 930 | |
dd41f596 IM |
931 | if (p->state == TASK_UNINTERRUPTIBLE) |
932 | rq->nr_uninterruptible--; | |
ece8a684 | 933 | |
dd41f596 IM |
934 | enqueue_task(rq, p, 0, now); |
935 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
936 | } |
937 | ||
938 | /* | |
939 | * deactivate_task - remove a task from the runqueue. | |
940 | */ | |
dd41f596 | 941 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 942 | { |
dd41f596 IM |
943 | u64 now = rq_clock(rq); |
944 | ||
945 | if (p->state == TASK_UNINTERRUPTIBLE) | |
946 | rq->nr_uninterruptible++; | |
947 | ||
948 | dequeue_task(rq, p, sleep, now); | |
949 | dec_nr_running(p, rq, now); | |
1da177e4 LT |
950 | } |
951 | ||
1da177e4 LT |
952 | /** |
953 | * task_curr - is this task currently executing on a CPU? | |
954 | * @p: the task in question. | |
955 | */ | |
36c8b586 | 956 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
957 | { |
958 | return cpu_curr(task_cpu(p)) == p; | |
959 | } | |
960 | ||
2dd73a4f PW |
961 | /* Used instead of source_load when we know the type == 0 */ |
962 | unsigned long weighted_cpuload(const int cpu) | |
963 | { | |
dd41f596 IM |
964 | return cpu_rq(cpu)->ls.load.weight; |
965 | } | |
966 | ||
967 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
968 | { | |
969 | #ifdef CONFIG_SMP | |
970 | task_thread_info(p)->cpu = cpu; | |
971 | set_task_cfs_rq(p); | |
972 | #endif | |
2dd73a4f PW |
973 | } |
974 | ||
1da177e4 | 975 | #ifdef CONFIG_SMP |
c65cc870 | 976 | |
dd41f596 | 977 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 978 | { |
dd41f596 IM |
979 | int old_cpu = task_cpu(p); |
980 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
981 | u64 clock_offset, fair_clock_offset; | |
982 | ||
983 | clock_offset = old_rq->clock - new_rq->clock; | |
984 | fair_clock_offset = old_rq->cfs.fair_clock - | |
985 | new_rq->cfs.fair_clock; | |
986 | if (p->se.wait_start) | |
987 | p->se.wait_start -= clock_offset; | |
988 | if (p->se.wait_start_fair) | |
989 | p->se.wait_start_fair -= fair_clock_offset; | |
990 | if (p->se.sleep_start) | |
991 | p->se.sleep_start -= clock_offset; | |
992 | if (p->se.block_start) | |
993 | p->se.block_start -= clock_offset; | |
994 | if (p->se.sleep_start_fair) | |
995 | p->se.sleep_start_fair -= fair_clock_offset; | |
996 | ||
997 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
998 | } |
999 | ||
70b97a7f | 1000 | struct migration_req { |
1da177e4 | 1001 | struct list_head list; |
1da177e4 | 1002 | |
36c8b586 | 1003 | struct task_struct *task; |
1da177e4 LT |
1004 | int dest_cpu; |
1005 | ||
1da177e4 | 1006 | struct completion done; |
70b97a7f | 1007 | }; |
1da177e4 LT |
1008 | |
1009 | /* | |
1010 | * The task's runqueue lock must be held. | |
1011 | * Returns true if you have to wait for migration thread. | |
1012 | */ | |
36c8b586 | 1013 | static int |
70b97a7f | 1014 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1015 | { |
70b97a7f | 1016 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1017 | |
1018 | /* | |
1019 | * If the task is not on a runqueue (and not running), then | |
1020 | * it is sufficient to simply update the task's cpu field. | |
1021 | */ | |
dd41f596 | 1022 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1023 | set_task_cpu(p, dest_cpu); |
1024 | return 0; | |
1025 | } | |
1026 | ||
1027 | init_completion(&req->done); | |
1da177e4 LT |
1028 | req->task = p; |
1029 | req->dest_cpu = dest_cpu; | |
1030 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1031 | |
1da177e4 LT |
1032 | return 1; |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * wait_task_inactive - wait for a thread to unschedule. | |
1037 | * | |
1038 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1039 | * else this function might spin for a *long* time. This function can't | |
1040 | * be called with interrupts off, or it may introduce deadlock with | |
1041 | * smp_call_function() if an IPI is sent by the same process we are | |
1042 | * waiting to become inactive. | |
1043 | */ | |
36c8b586 | 1044 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1045 | { |
1046 | unsigned long flags; | |
dd41f596 | 1047 | int running, on_rq; |
70b97a7f | 1048 | struct rq *rq; |
1da177e4 LT |
1049 | |
1050 | repeat: | |
fa490cfd LT |
1051 | /* |
1052 | * We do the initial early heuristics without holding | |
1053 | * any task-queue locks at all. We'll only try to get | |
1054 | * the runqueue lock when things look like they will | |
1055 | * work out! | |
1056 | */ | |
1057 | rq = task_rq(p); | |
1058 | ||
1059 | /* | |
1060 | * If the task is actively running on another CPU | |
1061 | * still, just relax and busy-wait without holding | |
1062 | * any locks. | |
1063 | * | |
1064 | * NOTE! Since we don't hold any locks, it's not | |
1065 | * even sure that "rq" stays as the right runqueue! | |
1066 | * But we don't care, since "task_running()" will | |
1067 | * return false if the runqueue has changed and p | |
1068 | * is actually now running somewhere else! | |
1069 | */ | |
1070 | while (task_running(rq, p)) | |
1071 | cpu_relax(); | |
1072 | ||
1073 | /* | |
1074 | * Ok, time to look more closely! We need the rq | |
1075 | * lock now, to be *sure*. If we're wrong, we'll | |
1076 | * just go back and repeat. | |
1077 | */ | |
1da177e4 | 1078 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1079 | running = task_running(rq, p); |
dd41f596 | 1080 | on_rq = p->se.on_rq; |
fa490cfd LT |
1081 | task_rq_unlock(rq, &flags); |
1082 | ||
1083 | /* | |
1084 | * Was it really running after all now that we | |
1085 | * checked with the proper locks actually held? | |
1086 | * | |
1087 | * Oops. Go back and try again.. | |
1088 | */ | |
1089 | if (unlikely(running)) { | |
1da177e4 | 1090 | cpu_relax(); |
1da177e4 LT |
1091 | goto repeat; |
1092 | } | |
fa490cfd LT |
1093 | |
1094 | /* | |
1095 | * It's not enough that it's not actively running, | |
1096 | * it must be off the runqueue _entirely_, and not | |
1097 | * preempted! | |
1098 | * | |
1099 | * So if it wa still runnable (but just not actively | |
1100 | * running right now), it's preempted, and we should | |
1101 | * yield - it could be a while. | |
1102 | */ | |
dd41f596 | 1103 | if (unlikely(on_rq)) { |
fa490cfd LT |
1104 | yield(); |
1105 | goto repeat; | |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * Ahh, all good. It wasn't running, and it wasn't | |
1110 | * runnable, which means that it will never become | |
1111 | * running in the future either. We're all done! | |
1112 | */ | |
1da177e4 LT |
1113 | } |
1114 | ||
1115 | /*** | |
1116 | * kick_process - kick a running thread to enter/exit the kernel | |
1117 | * @p: the to-be-kicked thread | |
1118 | * | |
1119 | * Cause a process which is running on another CPU to enter | |
1120 | * kernel-mode, without any delay. (to get signals handled.) | |
1121 | * | |
1122 | * NOTE: this function doesnt have to take the runqueue lock, | |
1123 | * because all it wants to ensure is that the remote task enters | |
1124 | * the kernel. If the IPI races and the task has been migrated | |
1125 | * to another CPU then no harm is done and the purpose has been | |
1126 | * achieved as well. | |
1127 | */ | |
36c8b586 | 1128 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1129 | { |
1130 | int cpu; | |
1131 | ||
1132 | preempt_disable(); | |
1133 | cpu = task_cpu(p); | |
1134 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1135 | smp_send_reschedule(cpu); | |
1136 | preempt_enable(); | |
1137 | } | |
1138 | ||
1139 | /* | |
2dd73a4f PW |
1140 | * Return a low guess at the load of a migration-source cpu weighted |
1141 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1142 | * |
1143 | * We want to under-estimate the load of migration sources, to | |
1144 | * balance conservatively. | |
1145 | */ | |
a2000572 | 1146 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1147 | { |
70b97a7f | 1148 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1149 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1150 | |
3b0bd9bc | 1151 | if (type == 0) |
dd41f596 | 1152 | return total; |
b910472d | 1153 | |
dd41f596 | 1154 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1155 | } |
1156 | ||
1157 | /* | |
2dd73a4f PW |
1158 | * Return a high guess at the load of a migration-target cpu weighted |
1159 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1160 | */ |
a2000572 | 1161 | static inline unsigned long target_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 | |
7897986b | 1166 | if (type == 0) |
dd41f596 | 1167 | return total; |
3b0bd9bc | 1168 | |
dd41f596 | 1169 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1170 | } |
1171 | ||
1172 | /* | |
1173 | * Return the average load per task on the cpu's run queue | |
1174 | */ | |
1175 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1176 | { | |
70b97a7f | 1177 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1178 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1179 | unsigned long n = rq->nr_running; |
1180 | ||
dd41f596 | 1181 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1182 | } |
1183 | ||
147cbb4b NP |
1184 | /* |
1185 | * find_idlest_group finds and returns the least busy CPU group within the | |
1186 | * domain. | |
1187 | */ | |
1188 | static struct sched_group * | |
1189 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1190 | { | |
1191 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1192 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1193 | int load_idx = sd->forkexec_idx; | |
1194 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1195 | ||
1196 | do { | |
1197 | unsigned long load, avg_load; | |
1198 | int local_group; | |
1199 | int i; | |
1200 | ||
da5a5522 BD |
1201 | /* Skip over this group if it has no CPUs allowed */ |
1202 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1203 | goto nextgroup; | |
1204 | ||
147cbb4b | 1205 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1206 | |
1207 | /* Tally up the load of all CPUs in the group */ | |
1208 | avg_load = 0; | |
1209 | ||
1210 | for_each_cpu_mask(i, group->cpumask) { | |
1211 | /* Bias balancing toward cpus of our domain */ | |
1212 | if (local_group) | |
1213 | load = source_load(i, load_idx); | |
1214 | else | |
1215 | load = target_load(i, load_idx); | |
1216 | ||
1217 | avg_load += load; | |
1218 | } | |
1219 | ||
1220 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1221 | avg_load = sg_div_cpu_power(group, |
1222 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1223 | |
1224 | if (local_group) { | |
1225 | this_load = avg_load; | |
1226 | this = group; | |
1227 | } else if (avg_load < min_load) { | |
1228 | min_load = avg_load; | |
1229 | idlest = group; | |
1230 | } | |
da5a5522 | 1231 | nextgroup: |
147cbb4b NP |
1232 | group = group->next; |
1233 | } while (group != sd->groups); | |
1234 | ||
1235 | if (!idlest || 100*this_load < imbalance*min_load) | |
1236 | return NULL; | |
1237 | return idlest; | |
1238 | } | |
1239 | ||
1240 | /* | |
0feaece9 | 1241 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1242 | */ |
95cdf3b7 IM |
1243 | static int |
1244 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1245 | { |
da5a5522 | 1246 | cpumask_t tmp; |
147cbb4b NP |
1247 | unsigned long load, min_load = ULONG_MAX; |
1248 | int idlest = -1; | |
1249 | int i; | |
1250 | ||
da5a5522 BD |
1251 | /* Traverse only the allowed CPUs */ |
1252 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1253 | ||
1254 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1255 | load = weighted_cpuload(i); |
147cbb4b NP |
1256 | |
1257 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1258 | min_load = load; | |
1259 | idlest = i; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | return idlest; | |
1264 | } | |
1265 | ||
476d139c NP |
1266 | /* |
1267 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1268 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1269 | * SD_BALANCE_EXEC. | |
1270 | * | |
1271 | * Balance, ie. select the least loaded group. | |
1272 | * | |
1273 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1274 | * | |
1275 | * preempt must be disabled. | |
1276 | */ | |
1277 | static int sched_balance_self(int cpu, int flag) | |
1278 | { | |
1279 | struct task_struct *t = current; | |
1280 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1281 | |
c96d145e | 1282 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1283 | /* |
1284 | * If power savings logic is enabled for a domain, stop there. | |
1285 | */ | |
5c45bf27 SS |
1286 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1287 | break; | |
476d139c NP |
1288 | if (tmp->flags & flag) |
1289 | sd = tmp; | |
c96d145e | 1290 | } |
476d139c NP |
1291 | |
1292 | while (sd) { | |
1293 | cpumask_t span; | |
1294 | struct sched_group *group; | |
1a848870 SS |
1295 | int new_cpu, weight; |
1296 | ||
1297 | if (!(sd->flags & flag)) { | |
1298 | sd = sd->child; | |
1299 | continue; | |
1300 | } | |
476d139c NP |
1301 | |
1302 | span = sd->span; | |
1303 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1304 | if (!group) { |
1305 | sd = sd->child; | |
1306 | continue; | |
1307 | } | |
476d139c | 1308 | |
da5a5522 | 1309 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1310 | if (new_cpu == -1 || new_cpu == cpu) { |
1311 | /* Now try balancing at a lower domain level of cpu */ | |
1312 | sd = sd->child; | |
1313 | continue; | |
1314 | } | |
476d139c | 1315 | |
1a848870 | 1316 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1317 | cpu = new_cpu; |
476d139c NP |
1318 | sd = NULL; |
1319 | weight = cpus_weight(span); | |
1320 | for_each_domain(cpu, tmp) { | |
1321 | if (weight <= cpus_weight(tmp->span)) | |
1322 | break; | |
1323 | if (tmp->flags & flag) | |
1324 | sd = tmp; | |
1325 | } | |
1326 | /* while loop will break here if sd == NULL */ | |
1327 | } | |
1328 | ||
1329 | return cpu; | |
1330 | } | |
1331 | ||
1332 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1333 | |
1334 | /* | |
1335 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1336 | * not idle and an idle cpu is available. The span of cpus to | |
1337 | * search starts with cpus closest then further out as needed, | |
1338 | * so we always favor a closer, idle cpu. | |
1339 | * | |
1340 | * Returns the CPU we should wake onto. | |
1341 | */ | |
1342 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1343 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1344 | { |
1345 | cpumask_t tmp; | |
1346 | struct sched_domain *sd; | |
1347 | int i; | |
1348 | ||
4953198b SS |
1349 | /* |
1350 | * If it is idle, then it is the best cpu to run this task. | |
1351 | * | |
1352 | * This cpu is also the best, if it has more than one task already. | |
1353 | * Siblings must be also busy(in most cases) as they didn't already | |
1354 | * pickup the extra load from this cpu and hence we need not check | |
1355 | * sibling runqueue info. This will avoid the checks and cache miss | |
1356 | * penalities associated with that. | |
1357 | */ | |
1358 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1359 | return cpu; |
1360 | ||
1361 | for_each_domain(cpu, sd) { | |
1362 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1363 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1364 | for_each_cpu_mask(i, tmp) { |
1365 | if (idle_cpu(i)) | |
1366 | return i; | |
1367 | } | |
9761eea8 | 1368 | } else { |
e0f364f4 | 1369 | break; |
9761eea8 | 1370 | } |
1da177e4 LT |
1371 | } |
1372 | return cpu; | |
1373 | } | |
1374 | #else | |
36c8b586 | 1375 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1376 | { |
1377 | return cpu; | |
1378 | } | |
1379 | #endif | |
1380 | ||
1381 | /*** | |
1382 | * try_to_wake_up - wake up a thread | |
1383 | * @p: the to-be-woken-up thread | |
1384 | * @state: the mask of task states that can be woken | |
1385 | * @sync: do a synchronous wakeup? | |
1386 | * | |
1387 | * Put it on the run-queue if it's not already there. The "current" | |
1388 | * thread is always on the run-queue (except when the actual | |
1389 | * re-schedule is in progress), and as such you're allowed to do | |
1390 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1391 | * runnable without the overhead of this. | |
1392 | * | |
1393 | * returns failure only if the task is already active. | |
1394 | */ | |
36c8b586 | 1395 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1396 | { |
1397 | int cpu, this_cpu, success = 0; | |
1398 | unsigned long flags; | |
1399 | long old_state; | |
70b97a7f | 1400 | struct rq *rq; |
1da177e4 | 1401 | #ifdef CONFIG_SMP |
7897986b | 1402 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1403 | unsigned long load, this_load; |
1da177e4 LT |
1404 | int new_cpu; |
1405 | #endif | |
1406 | ||
1407 | rq = task_rq_lock(p, &flags); | |
1408 | old_state = p->state; | |
1409 | if (!(old_state & state)) | |
1410 | goto out; | |
1411 | ||
dd41f596 | 1412 | if (p->se.on_rq) |
1da177e4 LT |
1413 | goto out_running; |
1414 | ||
1415 | cpu = task_cpu(p); | |
1416 | this_cpu = smp_processor_id(); | |
1417 | ||
1418 | #ifdef CONFIG_SMP | |
1419 | if (unlikely(task_running(rq, p))) | |
1420 | goto out_activate; | |
1421 | ||
7897986b NP |
1422 | new_cpu = cpu; |
1423 | ||
1da177e4 LT |
1424 | schedstat_inc(rq, ttwu_cnt); |
1425 | if (cpu == this_cpu) { | |
1426 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1427 | goto out_set_cpu; |
1428 | } | |
1429 | ||
1430 | for_each_domain(this_cpu, sd) { | |
1431 | if (cpu_isset(cpu, sd->span)) { | |
1432 | schedstat_inc(sd, ttwu_wake_remote); | |
1433 | this_sd = sd; | |
1434 | break; | |
1da177e4 LT |
1435 | } |
1436 | } | |
1da177e4 | 1437 | |
7897986b | 1438 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1439 | goto out_set_cpu; |
1440 | ||
1da177e4 | 1441 | /* |
7897986b | 1442 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1443 | */ |
7897986b NP |
1444 | if (this_sd) { |
1445 | int idx = this_sd->wake_idx; | |
1446 | unsigned int imbalance; | |
1da177e4 | 1447 | |
a3f21bce NP |
1448 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1449 | ||
7897986b NP |
1450 | load = source_load(cpu, idx); |
1451 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1452 | |
7897986b NP |
1453 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1454 | ||
a3f21bce NP |
1455 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1456 | unsigned long tl = this_load; | |
33859f7f MOS |
1457 | unsigned long tl_per_task; |
1458 | ||
1459 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1460 | |
1da177e4 | 1461 | /* |
a3f21bce NP |
1462 | * If sync wakeup then subtract the (maximum possible) |
1463 | * effect of the currently running task from the load | |
1464 | * of the current CPU: | |
1da177e4 | 1465 | */ |
a3f21bce | 1466 | if (sync) |
dd41f596 | 1467 | tl -= current->se.load.weight; |
a3f21bce NP |
1468 | |
1469 | if ((tl <= load && | |
2dd73a4f | 1470 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1471 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1472 | /* |
1473 | * This domain has SD_WAKE_AFFINE and | |
1474 | * p is cache cold in this domain, and | |
1475 | * there is no bad imbalance. | |
1476 | */ | |
1477 | schedstat_inc(this_sd, ttwu_move_affine); | |
1478 | goto out_set_cpu; | |
1479 | } | |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * Start passive balancing when half the imbalance_pct | |
1484 | * limit is reached. | |
1485 | */ | |
1486 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1487 | if (imbalance*this_load <= 100*load) { | |
1488 | schedstat_inc(this_sd, ttwu_move_balance); | |
1489 | goto out_set_cpu; | |
1490 | } | |
1da177e4 LT |
1491 | } |
1492 | } | |
1493 | ||
1494 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1495 | out_set_cpu: | |
1496 | new_cpu = wake_idle(new_cpu, p); | |
1497 | if (new_cpu != cpu) { | |
1498 | set_task_cpu(p, new_cpu); | |
1499 | task_rq_unlock(rq, &flags); | |
1500 | /* might preempt at this point */ | |
1501 | rq = task_rq_lock(p, &flags); | |
1502 | old_state = p->state; | |
1503 | if (!(old_state & state)) | |
1504 | goto out; | |
dd41f596 | 1505 | if (p->se.on_rq) |
1da177e4 LT |
1506 | goto out_running; |
1507 | ||
1508 | this_cpu = smp_processor_id(); | |
1509 | cpu = task_cpu(p); | |
1510 | } | |
1511 | ||
1512 | out_activate: | |
1513 | #endif /* CONFIG_SMP */ | |
dd41f596 | 1514 | activate_task(rq, p, 1); |
1da177e4 LT |
1515 | /* |
1516 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1517 | * has indicated that it will leave the CPU in short order) | |
1518 | * don't trigger a preemption, if the woken up task will run on | |
1519 | * this cpu. (in this case the 'I will reschedule' promise of | |
1520 | * the waker guarantees that the freshly woken up task is going | |
1521 | * to be considered on this CPU.) | |
1522 | */ | |
dd41f596 IM |
1523 | if (!sync || cpu != this_cpu) |
1524 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1525 | success = 1; |
1526 | ||
1527 | out_running: | |
1528 | p->state = TASK_RUNNING; | |
1529 | out: | |
1530 | task_rq_unlock(rq, &flags); | |
1531 | ||
1532 | return success; | |
1533 | } | |
1534 | ||
36c8b586 | 1535 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1536 | { |
1537 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1538 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1539 | } | |
1da177e4 LT |
1540 | EXPORT_SYMBOL(wake_up_process); |
1541 | ||
36c8b586 | 1542 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1543 | { |
1544 | return try_to_wake_up(p, state, 0); | |
1545 | } | |
1546 | ||
1da177e4 LT |
1547 | /* |
1548 | * Perform scheduler related setup for a newly forked process p. | |
1549 | * p is forked by current. | |
dd41f596 IM |
1550 | * |
1551 | * __sched_fork() is basic setup used by init_idle() too: | |
1552 | */ | |
1553 | static void __sched_fork(struct task_struct *p) | |
1554 | { | |
1555 | p->se.wait_start_fair = 0; | |
1556 | p->se.wait_start = 0; | |
1557 | p->se.exec_start = 0; | |
1558 | p->se.sum_exec_runtime = 0; | |
1559 | p->se.delta_exec = 0; | |
1560 | p->se.delta_fair_run = 0; | |
1561 | p->se.delta_fair_sleep = 0; | |
1562 | p->se.wait_runtime = 0; | |
1563 | p->se.sum_wait_runtime = 0; | |
1564 | p->se.sum_sleep_runtime = 0; | |
1565 | p->se.sleep_start = 0; | |
1566 | p->se.sleep_start_fair = 0; | |
1567 | p->se.block_start = 0; | |
1568 | p->se.sleep_max = 0; | |
1569 | p->se.block_max = 0; | |
1570 | p->se.exec_max = 0; | |
1571 | p->se.wait_max = 0; | |
1572 | p->se.wait_runtime_overruns = 0; | |
1573 | p->se.wait_runtime_underruns = 0; | |
476d139c | 1574 | |
dd41f596 IM |
1575 | INIT_LIST_HEAD(&p->run_list); |
1576 | p->se.on_rq = 0; | |
476d139c | 1577 | |
1da177e4 LT |
1578 | /* |
1579 | * We mark the process as running here, but have not actually | |
1580 | * inserted it onto the runqueue yet. This guarantees that | |
1581 | * nobody will actually run it, and a signal or other external | |
1582 | * event cannot wake it up and insert it on the runqueue either. | |
1583 | */ | |
1584 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1585 | } |
1586 | ||
1587 | /* | |
1588 | * fork()/clone()-time setup: | |
1589 | */ | |
1590 | void sched_fork(struct task_struct *p, int clone_flags) | |
1591 | { | |
1592 | int cpu = get_cpu(); | |
1593 | ||
1594 | __sched_fork(p); | |
1595 | ||
1596 | #ifdef CONFIG_SMP | |
1597 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1598 | #endif | |
1599 | __set_task_cpu(p, cpu); | |
b29739f9 IM |
1600 | |
1601 | /* | |
1602 | * Make sure we do not leak PI boosting priority to the child: | |
1603 | */ | |
1604 | p->prio = current->normal_prio; | |
1605 | ||
52f17b6c | 1606 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1607 | if (likely(sched_info_on())) |
52f17b6c | 1608 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1609 | #endif |
d6077cb8 | 1610 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1611 | p->oncpu = 0; |
1612 | #endif | |
1da177e4 | 1613 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1614 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1615 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1616 | #endif |
476d139c | 1617 | put_cpu(); |
1da177e4 LT |
1618 | } |
1619 | ||
dd41f596 IM |
1620 | /* |
1621 | * After fork, child runs first. (default) If set to 0 then | |
1622 | * parent will (try to) run first. | |
1623 | */ | |
1624 | unsigned int __read_mostly sysctl_sched_child_runs_first = 1; | |
1625 | ||
1da177e4 LT |
1626 | /* |
1627 | * wake_up_new_task - wake up a newly created task for the first time. | |
1628 | * | |
1629 | * This function will do some initial scheduler statistics housekeeping | |
1630 | * that must be done for every newly created context, then puts the task | |
1631 | * on the runqueue and wakes it. | |
1632 | */ | |
36c8b586 | 1633 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1634 | { |
1635 | unsigned long flags; | |
dd41f596 IM |
1636 | struct rq *rq; |
1637 | int this_cpu; | |
1da177e4 LT |
1638 | |
1639 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1640 | BUG_ON(p->state != TASK_RUNNING); |
dd41f596 | 1641 | this_cpu = smp_processor_id(); /* parent's CPU */ |
1da177e4 LT |
1642 | |
1643 | p->prio = effective_prio(p); | |
1644 | ||
dd41f596 IM |
1645 | if (!sysctl_sched_child_runs_first || (clone_flags & CLONE_VM) || |
1646 | task_cpu(p) != this_cpu || !current->se.on_rq) { | |
1647 | activate_task(rq, p, 0); | |
1da177e4 | 1648 | } else { |
1da177e4 | 1649 | /* |
dd41f596 IM |
1650 | * Let the scheduling class do new task startup |
1651 | * management (if any): | |
1da177e4 | 1652 | */ |
dd41f596 | 1653 | p->sched_class->task_new(rq, p); |
1da177e4 | 1654 | } |
dd41f596 IM |
1655 | check_preempt_curr(rq, p); |
1656 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1657 | } |
1658 | ||
4866cde0 NP |
1659 | /** |
1660 | * prepare_task_switch - prepare to switch tasks | |
1661 | * @rq: the runqueue preparing to switch | |
1662 | * @next: the task we are going to switch to. | |
1663 | * | |
1664 | * This is called with the rq lock held and interrupts off. It must | |
1665 | * be paired with a subsequent finish_task_switch after the context | |
1666 | * switch. | |
1667 | * | |
1668 | * prepare_task_switch sets up locking and calls architecture specific | |
1669 | * hooks. | |
1670 | */ | |
70b97a7f | 1671 | static inline void prepare_task_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
1672 | { |
1673 | prepare_lock_switch(rq, next); | |
1674 | prepare_arch_switch(next); | |
1675 | } | |
1676 | ||
1da177e4 LT |
1677 | /** |
1678 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1679 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1680 | * @prev: the thread we just switched away from. |
1681 | * | |
4866cde0 NP |
1682 | * finish_task_switch must be called after the context switch, paired |
1683 | * with a prepare_task_switch call before the context switch. | |
1684 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1685 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1686 | * |
1687 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1688 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1689 | * with the lock held can cause deadlocks; see schedule() for | |
1690 | * details.) | |
1691 | */ | |
70b97a7f | 1692 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1693 | __releases(rq->lock) |
1694 | { | |
1da177e4 | 1695 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1696 | long prev_state; |
1da177e4 LT |
1697 | |
1698 | rq->prev_mm = NULL; | |
1699 | ||
1700 | /* | |
1701 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1702 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1703 | * schedule one last time. The schedule call will never return, and |
1704 | * the scheduled task must drop that reference. | |
c394cc9f | 1705 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1706 | * still held, otherwise prev could be scheduled on another cpu, die |
1707 | * there before we look at prev->state, and then the reference would | |
1708 | * be dropped twice. | |
1709 | * Manfred Spraul <manfred@colorfullife.com> | |
1710 | */ | |
55a101f8 | 1711 | prev_state = prev->state; |
4866cde0 NP |
1712 | finish_arch_switch(prev); |
1713 | finish_lock_switch(rq, prev); | |
1da177e4 LT |
1714 | if (mm) |
1715 | mmdrop(mm); | |
c394cc9f | 1716 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1717 | /* |
1718 | * Remove function-return probe instances associated with this | |
1719 | * task and put them back on the free list. | |
9761eea8 | 1720 | */ |
c6fd91f0 | 1721 | kprobe_flush_task(prev); |
1da177e4 | 1722 | put_task_struct(prev); |
c6fd91f0 | 1723 | } |
1da177e4 LT |
1724 | } |
1725 | ||
1726 | /** | |
1727 | * schedule_tail - first thing a freshly forked thread must call. | |
1728 | * @prev: the thread we just switched away from. | |
1729 | */ | |
36c8b586 | 1730 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1731 | __releases(rq->lock) |
1732 | { | |
70b97a7f IM |
1733 | struct rq *rq = this_rq(); |
1734 | ||
4866cde0 NP |
1735 | finish_task_switch(rq, prev); |
1736 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1737 | /* In this case, finish_task_switch does not reenable preemption */ | |
1738 | preempt_enable(); | |
1739 | #endif | |
1da177e4 LT |
1740 | if (current->set_child_tid) |
1741 | put_user(current->pid, current->set_child_tid); | |
1742 | } | |
1743 | ||
1744 | /* | |
1745 | * context_switch - switch to the new MM and the new | |
1746 | * thread's register state. | |
1747 | */ | |
dd41f596 | 1748 | static inline void |
70b97a7f | 1749 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1750 | struct task_struct *next) |
1da177e4 | 1751 | { |
dd41f596 | 1752 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1753 | |
dd41f596 IM |
1754 | prepare_task_switch(rq, next); |
1755 | mm = next->mm; | |
1756 | oldmm = prev->active_mm; | |
9226d125 ZA |
1757 | /* |
1758 | * For paravirt, this is coupled with an exit in switch_to to | |
1759 | * combine the page table reload and the switch backend into | |
1760 | * one hypercall. | |
1761 | */ | |
1762 | arch_enter_lazy_cpu_mode(); | |
1763 | ||
dd41f596 | 1764 | if (unlikely(!mm)) { |
1da177e4 LT |
1765 | next->active_mm = oldmm; |
1766 | atomic_inc(&oldmm->mm_count); | |
1767 | enter_lazy_tlb(oldmm, next); | |
1768 | } else | |
1769 | switch_mm(oldmm, mm, next); | |
1770 | ||
dd41f596 | 1771 | if (unlikely(!prev->mm)) { |
1da177e4 | 1772 | prev->active_mm = NULL; |
1da177e4 LT |
1773 | rq->prev_mm = oldmm; |
1774 | } | |
3a5f5e48 IM |
1775 | /* |
1776 | * Since the runqueue lock will be released by the next | |
1777 | * task (which is an invalid locking op but in the case | |
1778 | * of the scheduler it's an obvious special-case), so we | |
1779 | * do an early lockdep release here: | |
1780 | */ | |
1781 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1782 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1783 | #endif |
1da177e4 LT |
1784 | |
1785 | /* Here we just switch the register state and the stack. */ | |
1786 | switch_to(prev, next, prev); | |
1787 | ||
dd41f596 IM |
1788 | barrier(); |
1789 | /* | |
1790 | * this_rq must be evaluated again because prev may have moved | |
1791 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1792 | * frame will be invalid. | |
1793 | */ | |
1794 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1799 | * | |
1800 | * externally visible scheduler statistics: current number of runnable | |
1801 | * threads, current number of uninterruptible-sleeping threads, total | |
1802 | * number of context switches performed since bootup. | |
1803 | */ | |
1804 | unsigned long nr_running(void) | |
1805 | { | |
1806 | unsigned long i, sum = 0; | |
1807 | ||
1808 | for_each_online_cpu(i) | |
1809 | sum += cpu_rq(i)->nr_running; | |
1810 | ||
1811 | return sum; | |
1812 | } | |
1813 | ||
1814 | unsigned long nr_uninterruptible(void) | |
1815 | { | |
1816 | unsigned long i, sum = 0; | |
1817 | ||
0a945022 | 1818 | for_each_possible_cpu(i) |
1da177e4 LT |
1819 | sum += cpu_rq(i)->nr_uninterruptible; |
1820 | ||
1821 | /* | |
1822 | * Since we read the counters lockless, it might be slightly | |
1823 | * inaccurate. Do not allow it to go below zero though: | |
1824 | */ | |
1825 | if (unlikely((long)sum < 0)) | |
1826 | sum = 0; | |
1827 | ||
1828 | return sum; | |
1829 | } | |
1830 | ||
1831 | unsigned long long nr_context_switches(void) | |
1832 | { | |
cc94abfc SR |
1833 | int i; |
1834 | unsigned long long sum = 0; | |
1da177e4 | 1835 | |
0a945022 | 1836 | for_each_possible_cpu(i) |
1da177e4 LT |
1837 | sum += cpu_rq(i)->nr_switches; |
1838 | ||
1839 | return sum; | |
1840 | } | |
1841 | ||
1842 | unsigned long nr_iowait(void) | |
1843 | { | |
1844 | unsigned long i, sum = 0; | |
1845 | ||
0a945022 | 1846 | for_each_possible_cpu(i) |
1da177e4 LT |
1847 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1848 | ||
1849 | return sum; | |
1850 | } | |
1851 | ||
db1b1fef JS |
1852 | unsigned long nr_active(void) |
1853 | { | |
1854 | unsigned long i, running = 0, uninterruptible = 0; | |
1855 | ||
1856 | for_each_online_cpu(i) { | |
1857 | running += cpu_rq(i)->nr_running; | |
1858 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1859 | } | |
1860 | ||
1861 | if (unlikely((long)uninterruptible < 0)) | |
1862 | uninterruptible = 0; | |
1863 | ||
1864 | return running + uninterruptible; | |
1865 | } | |
1866 | ||
48f24c4d | 1867 | /* |
dd41f596 IM |
1868 | * Update rq->cpu_load[] statistics. This function is usually called every |
1869 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1870 | */ |
dd41f596 | 1871 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1872 | { |
dd41f596 IM |
1873 | u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64; |
1874 | unsigned long total_load = this_rq->ls.load.weight; | |
1875 | unsigned long this_load = total_load; | |
1876 | struct load_stat *ls = &this_rq->ls; | |
1877 | u64 now = __rq_clock(this_rq); | |
1878 | int i, scale; | |
1879 | ||
1880 | this_rq->nr_load_updates++; | |
1881 | if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD))) | |
1882 | goto do_avg; | |
1883 | ||
1884 | /* Update delta_fair/delta_exec fields first */ | |
1885 | update_curr_load(this_rq, now); | |
1886 | ||
1887 | fair_delta64 = ls->delta_fair + 1; | |
1888 | ls->delta_fair = 0; | |
1889 | ||
1890 | exec_delta64 = ls->delta_exec + 1; | |
1891 | ls->delta_exec = 0; | |
1892 | ||
1893 | sample_interval64 = now - ls->load_update_last; | |
1894 | ls->load_update_last = now; | |
1895 | ||
1896 | if ((s64)sample_interval64 < (s64)TICK_NSEC) | |
1897 | sample_interval64 = TICK_NSEC; | |
1898 | ||
1899 | if (exec_delta64 > sample_interval64) | |
1900 | exec_delta64 = sample_interval64; | |
1901 | ||
1902 | idle_delta64 = sample_interval64 - exec_delta64; | |
1903 | ||
1904 | tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64); | |
1905 | tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64); | |
1906 | ||
1907 | this_load = (unsigned long)tmp64; | |
1908 | ||
1909 | do_avg: | |
1910 | ||
1911 | /* Update our load: */ | |
1912 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1913 | unsigned long old_load, new_load; | |
1914 | ||
1915 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1916 | ||
1917 | old_load = this_rq->cpu_load[i]; | |
1918 | new_load = this_load; | |
1919 | ||
1920 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; | |
1921 | } | |
48f24c4d IM |
1922 | } |
1923 | ||
dd41f596 IM |
1924 | #ifdef CONFIG_SMP |
1925 | ||
1da177e4 LT |
1926 | /* |
1927 | * double_rq_lock - safely lock two runqueues | |
1928 | * | |
1929 | * Note this does not disable interrupts like task_rq_lock, | |
1930 | * you need to do so manually before calling. | |
1931 | */ | |
70b97a7f | 1932 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
1933 | __acquires(rq1->lock) |
1934 | __acquires(rq2->lock) | |
1935 | { | |
054b9108 | 1936 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
1937 | if (rq1 == rq2) { |
1938 | spin_lock(&rq1->lock); | |
1939 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1940 | } else { | |
c96d145e | 1941 | if (rq1 < rq2) { |
1da177e4 LT |
1942 | spin_lock(&rq1->lock); |
1943 | spin_lock(&rq2->lock); | |
1944 | } else { | |
1945 | spin_lock(&rq2->lock); | |
1946 | spin_lock(&rq1->lock); | |
1947 | } | |
1948 | } | |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * double_rq_unlock - safely unlock two runqueues | |
1953 | * | |
1954 | * Note this does not restore interrupts like task_rq_unlock, | |
1955 | * you need to do so manually after calling. | |
1956 | */ | |
70b97a7f | 1957 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
1958 | __releases(rq1->lock) |
1959 | __releases(rq2->lock) | |
1960 | { | |
1961 | spin_unlock(&rq1->lock); | |
1962 | if (rq1 != rq2) | |
1963 | spin_unlock(&rq2->lock); | |
1964 | else | |
1965 | __release(rq2->lock); | |
1966 | } | |
1967 | ||
1968 | /* | |
1969 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1970 | */ | |
70b97a7f | 1971 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
1972 | __releases(this_rq->lock) |
1973 | __acquires(busiest->lock) | |
1974 | __acquires(this_rq->lock) | |
1975 | { | |
054b9108 KK |
1976 | if (unlikely(!irqs_disabled())) { |
1977 | /* printk() doesn't work good under rq->lock */ | |
1978 | spin_unlock(&this_rq->lock); | |
1979 | BUG_ON(1); | |
1980 | } | |
1da177e4 | 1981 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 1982 | if (busiest < this_rq) { |
1da177e4 LT |
1983 | spin_unlock(&this_rq->lock); |
1984 | spin_lock(&busiest->lock); | |
1985 | spin_lock(&this_rq->lock); | |
1986 | } else | |
1987 | spin_lock(&busiest->lock); | |
1988 | } | |
1989 | } | |
1990 | ||
1da177e4 LT |
1991 | /* |
1992 | * If dest_cpu is allowed for this process, migrate the task to it. | |
1993 | * This is accomplished by forcing the cpu_allowed mask to only | |
1994 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
1995 | * the cpu_allowed mask is restored. | |
1996 | */ | |
36c8b586 | 1997 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 1998 | { |
70b97a7f | 1999 | struct migration_req req; |
1da177e4 | 2000 | unsigned long flags; |
70b97a7f | 2001 | struct rq *rq; |
1da177e4 LT |
2002 | |
2003 | rq = task_rq_lock(p, &flags); | |
2004 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2005 | || unlikely(cpu_is_offline(dest_cpu))) | |
2006 | goto out; | |
2007 | ||
2008 | /* force the process onto the specified CPU */ | |
2009 | if (migrate_task(p, dest_cpu, &req)) { | |
2010 | /* Need to wait for migration thread (might exit: take ref). */ | |
2011 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2012 | |
1da177e4 LT |
2013 | get_task_struct(mt); |
2014 | task_rq_unlock(rq, &flags); | |
2015 | wake_up_process(mt); | |
2016 | put_task_struct(mt); | |
2017 | wait_for_completion(&req.done); | |
36c8b586 | 2018 | |
1da177e4 LT |
2019 | return; |
2020 | } | |
2021 | out: | |
2022 | task_rq_unlock(rq, &flags); | |
2023 | } | |
2024 | ||
2025 | /* | |
476d139c NP |
2026 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2027 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2028 | */ |
2029 | void sched_exec(void) | |
2030 | { | |
1da177e4 | 2031 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2032 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2033 | put_cpu(); |
476d139c NP |
2034 | if (new_cpu != this_cpu) |
2035 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2036 | } |
2037 | ||
2038 | /* | |
2039 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2040 | * Both runqueues must be locked. | |
2041 | */ | |
dd41f596 IM |
2042 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2043 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2044 | { |
dd41f596 | 2045 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2046 | set_task_cpu(p, this_cpu); |
dd41f596 | 2047 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2048 | /* |
2049 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2050 | * to be always true for them. | |
2051 | */ | |
dd41f596 | 2052 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2053 | } |
2054 | ||
2055 | /* | |
2056 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2057 | */ | |
858119e1 | 2058 | static |
70b97a7f | 2059 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2060 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2061 | int *all_pinned) |
1da177e4 LT |
2062 | { |
2063 | /* | |
2064 | * We do not migrate tasks that are: | |
2065 | * 1) running (obviously), or | |
2066 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2067 | * 3) are cache-hot on their current CPU. | |
2068 | */ | |
1da177e4 LT |
2069 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2070 | return 0; | |
81026794 NP |
2071 | *all_pinned = 0; |
2072 | ||
2073 | if (task_running(rq, p)) | |
2074 | return 0; | |
1da177e4 LT |
2075 | |
2076 | /* | |
dd41f596 | 2077 | * Aggressive migration if too many balance attempts have failed: |
1da177e4 | 2078 | */ |
dd41f596 | 2079 | if (sd->nr_balance_failed > sd->cache_nice_tries) |
1da177e4 LT |
2080 | return 1; |
2081 | ||
1da177e4 LT |
2082 | return 1; |
2083 | } | |
2084 | ||
dd41f596 | 2085 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2086 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2087 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 IM |
2088 | int *all_pinned, unsigned long *load_moved, |
2089 | int this_best_prio, int best_prio, int best_prio_seen, | |
2090 | struct rq_iterator *iterator) | |
1da177e4 | 2091 | { |
dd41f596 IM |
2092 | int pulled = 0, pinned = 0, skip_for_load; |
2093 | struct task_struct *p; | |
2094 | long rem_load_move = max_load_move; | |
1da177e4 | 2095 | |
2dd73a4f | 2096 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2097 | goto out; |
2098 | ||
81026794 NP |
2099 | pinned = 1; |
2100 | ||
1da177e4 | 2101 | /* |
dd41f596 | 2102 | * Start the load-balancing iterator: |
1da177e4 | 2103 | */ |
dd41f596 IM |
2104 | p = iterator->start(iterator->arg); |
2105 | next: | |
2106 | if (!p) | |
1da177e4 | 2107 | goto out; |
50ddd969 PW |
2108 | /* |
2109 | * To help distribute high priority tasks accross CPUs we don't | |
2110 | * skip a task if it will be the highest priority task (i.e. smallest | |
2111 | * prio value) on its new queue regardless of its load weight | |
2112 | */ | |
dd41f596 IM |
2113 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2114 | SCHED_LOAD_SCALE_FUZZ; | |
2115 | if (skip_for_load && p->prio < this_best_prio) | |
2116 | skip_for_load = !best_prio_seen && p->prio == best_prio; | |
615052dc | 2117 | if (skip_for_load || |
dd41f596 | 2118 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d | 2119 | |
dd41f596 IM |
2120 | best_prio_seen |= p->prio == best_prio; |
2121 | p = iterator->next(iterator->arg); | |
2122 | goto next; | |
1da177e4 LT |
2123 | } |
2124 | ||
dd41f596 | 2125 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2126 | pulled++; |
dd41f596 | 2127 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2128 | |
2dd73a4f PW |
2129 | /* |
2130 | * We only want to steal up to the prescribed number of tasks | |
2131 | * and the prescribed amount of weighted load. | |
2132 | */ | |
2133 | if (pulled < max_nr_move && rem_load_move > 0) { | |
dd41f596 IM |
2134 | if (p->prio < this_best_prio) |
2135 | this_best_prio = p->prio; | |
2136 | p = iterator->next(iterator->arg); | |
2137 | goto next; | |
1da177e4 LT |
2138 | } |
2139 | out: | |
2140 | /* | |
2141 | * Right now, this is the only place pull_task() is called, | |
2142 | * so we can safely collect pull_task() stats here rather than | |
2143 | * inside pull_task(). | |
2144 | */ | |
2145 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2146 | |
2147 | if (all_pinned) | |
2148 | *all_pinned = pinned; | |
dd41f596 | 2149 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2150 | return pulled; |
2151 | } | |
2152 | ||
dd41f596 IM |
2153 | /* |
2154 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted | |
2155 | * load from busiest to this_rq, as part of a balancing operation within | |
2156 | * "domain". Returns the number of tasks moved. | |
2157 | * | |
2158 | * Called with both runqueues locked. | |
2159 | */ | |
2160 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2161 | unsigned long max_nr_move, unsigned long max_load_move, | |
2162 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2163 | int *all_pinned) | |
2164 | { | |
2165 | struct sched_class *class = sched_class_highest; | |
2166 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
2167 | long rem_load_move = max_load_move; | |
2168 | ||
2169 | do { | |
2170 | nr_moved = class->load_balance(this_rq, this_cpu, busiest, | |
2171 | max_nr_move, (unsigned long)rem_load_move, | |
2172 | sd, idle, all_pinned, &load_moved); | |
2173 | total_nr_moved += nr_moved; | |
2174 | max_nr_move -= nr_moved; | |
2175 | rem_load_move -= load_moved; | |
2176 | class = class->next; | |
2177 | } while (class && max_nr_move && rem_load_move > 0); | |
2178 | ||
2179 | return total_nr_moved; | |
2180 | } | |
2181 | ||
1da177e4 LT |
2182 | /* |
2183 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2184 | * domain. It calculates and returns the amount of weighted load which |
2185 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2186 | */ |
2187 | static struct sched_group * | |
2188 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2189 | unsigned long *imbalance, enum cpu_idle_type idle, |
2190 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2191 | { |
2192 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2193 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2194 | unsigned long max_pull; |
2dd73a4f PW |
2195 | unsigned long busiest_load_per_task, busiest_nr_running; |
2196 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2197 | int load_idx; |
5c45bf27 SS |
2198 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2199 | int power_savings_balance = 1; | |
2200 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2201 | unsigned long min_nr_running = ULONG_MAX; | |
2202 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2203 | #endif | |
1da177e4 LT |
2204 | |
2205 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2206 | busiest_load_per_task = busiest_nr_running = 0; |
2207 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2208 | if (idle == CPU_NOT_IDLE) |
7897986b | 2209 | load_idx = sd->busy_idx; |
d15bcfdb | 2210 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2211 | load_idx = sd->newidle_idx; |
2212 | else | |
2213 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2214 | |
2215 | do { | |
5c45bf27 | 2216 | unsigned long load, group_capacity; |
1da177e4 LT |
2217 | int local_group; |
2218 | int i; | |
783609c6 | 2219 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2220 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2221 | |
2222 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2223 | ||
783609c6 SS |
2224 | if (local_group) |
2225 | balance_cpu = first_cpu(group->cpumask); | |
2226 | ||
1da177e4 | 2227 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2228 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2229 | |
2230 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2231 | struct rq *rq; |
2232 | ||
2233 | if (!cpu_isset(i, *cpus)) | |
2234 | continue; | |
2235 | ||
2236 | rq = cpu_rq(i); | |
2dd73a4f | 2237 | |
5969fe06 NP |
2238 | if (*sd_idle && !idle_cpu(i)) |
2239 | *sd_idle = 0; | |
2240 | ||
1da177e4 | 2241 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2242 | if (local_group) { |
2243 | if (idle_cpu(i) && !first_idle_cpu) { | |
2244 | first_idle_cpu = 1; | |
2245 | balance_cpu = i; | |
2246 | } | |
2247 | ||
a2000572 | 2248 | load = target_load(i, load_idx); |
783609c6 | 2249 | } else |
a2000572 | 2250 | load = source_load(i, load_idx); |
1da177e4 LT |
2251 | |
2252 | avg_load += load; | |
2dd73a4f | 2253 | sum_nr_running += rq->nr_running; |
dd41f596 | 2254 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2255 | } |
2256 | ||
783609c6 SS |
2257 | /* |
2258 | * First idle cpu or the first cpu(busiest) in this sched group | |
2259 | * is eligible for doing load balancing at this and above | |
2260 | * domains. | |
2261 | */ | |
2262 | if (local_group && balance_cpu != this_cpu && balance) { | |
2263 | *balance = 0; | |
2264 | goto ret; | |
2265 | } | |
2266 | ||
1da177e4 | 2267 | total_load += avg_load; |
5517d86b | 2268 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2269 | |
2270 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2271 | avg_load = sg_div_cpu_power(group, |
2272 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2273 | |
5517d86b | 2274 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2275 | |
1da177e4 LT |
2276 | if (local_group) { |
2277 | this_load = avg_load; | |
2278 | this = group; | |
2dd73a4f PW |
2279 | this_nr_running = sum_nr_running; |
2280 | this_load_per_task = sum_weighted_load; | |
2281 | } else if (avg_load > max_load && | |
5c45bf27 | 2282 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2283 | max_load = avg_load; |
2284 | busiest = group; | |
2dd73a4f PW |
2285 | busiest_nr_running = sum_nr_running; |
2286 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2287 | } |
5c45bf27 SS |
2288 | |
2289 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2290 | /* | |
2291 | * Busy processors will not participate in power savings | |
2292 | * balance. | |
2293 | */ | |
dd41f596 IM |
2294 | if (idle == CPU_NOT_IDLE || |
2295 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2296 | goto group_next; | |
5c45bf27 SS |
2297 | |
2298 | /* | |
2299 | * If the local group is idle or completely loaded | |
2300 | * no need to do power savings balance at this domain | |
2301 | */ | |
2302 | if (local_group && (this_nr_running >= group_capacity || | |
2303 | !this_nr_running)) | |
2304 | power_savings_balance = 0; | |
2305 | ||
dd41f596 | 2306 | /* |
5c45bf27 SS |
2307 | * If a group is already running at full capacity or idle, |
2308 | * don't include that group in power savings calculations | |
dd41f596 IM |
2309 | */ |
2310 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2311 | || !sum_nr_running) |
dd41f596 | 2312 | goto group_next; |
5c45bf27 | 2313 | |
dd41f596 | 2314 | /* |
5c45bf27 | 2315 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2316 | * This is the group from where we need to pick up the load |
2317 | * for saving power | |
2318 | */ | |
2319 | if ((sum_nr_running < min_nr_running) || | |
2320 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2321 | first_cpu(group->cpumask) < |
2322 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2323 | group_min = group; |
2324 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2325 | min_load_per_task = sum_weighted_load / |
2326 | sum_nr_running; | |
dd41f596 | 2327 | } |
5c45bf27 | 2328 | |
dd41f596 | 2329 | /* |
5c45bf27 | 2330 | * Calculate the group which is almost near its |
dd41f596 IM |
2331 | * capacity but still has some space to pick up some load |
2332 | * from other group and save more power | |
2333 | */ | |
2334 | if (sum_nr_running <= group_capacity - 1) { | |
2335 | if (sum_nr_running > leader_nr_running || | |
2336 | (sum_nr_running == leader_nr_running && | |
2337 | first_cpu(group->cpumask) > | |
2338 | first_cpu(group_leader->cpumask))) { | |
2339 | group_leader = group; | |
2340 | leader_nr_running = sum_nr_running; | |
2341 | } | |
48f24c4d | 2342 | } |
5c45bf27 SS |
2343 | group_next: |
2344 | #endif | |
1da177e4 LT |
2345 | group = group->next; |
2346 | } while (group != sd->groups); | |
2347 | ||
2dd73a4f | 2348 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2349 | goto out_balanced; |
2350 | ||
2351 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2352 | ||
2353 | if (this_load >= avg_load || | |
2354 | 100*max_load <= sd->imbalance_pct*this_load) | |
2355 | goto out_balanced; | |
2356 | ||
2dd73a4f | 2357 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2358 | /* |
2359 | * We're trying to get all the cpus to the average_load, so we don't | |
2360 | * want to push ourselves above the average load, nor do we wish to | |
2361 | * reduce the max loaded cpu below the average load, as either of these | |
2362 | * actions would just result in more rebalancing later, and ping-pong | |
2363 | * tasks around. Thus we look for the minimum possible imbalance. | |
2364 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2365 | * be counted as no imbalance for these purposes -- we can't fix that | |
2366 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2367 | * appear as very large values with unsigned longs. | |
2368 | */ | |
2dd73a4f PW |
2369 | if (max_load <= busiest_load_per_task) |
2370 | goto out_balanced; | |
2371 | ||
2372 | /* | |
2373 | * In the presence of smp nice balancing, certain scenarios can have | |
2374 | * max load less than avg load(as we skip the groups at or below | |
2375 | * its cpu_power, while calculating max_load..) | |
2376 | */ | |
2377 | if (max_load < avg_load) { | |
2378 | *imbalance = 0; | |
2379 | goto small_imbalance; | |
2380 | } | |
0c117f1b SS |
2381 | |
2382 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2383 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2384 | |
1da177e4 | 2385 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2386 | *imbalance = min(max_pull * busiest->__cpu_power, |
2387 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2388 | / SCHED_LOAD_SCALE; |
2389 | ||
2dd73a4f PW |
2390 | /* |
2391 | * if *imbalance is less than the average load per runnable task | |
2392 | * there is no gaurantee that any tasks will be moved so we'll have | |
2393 | * a think about bumping its value to force at least one task to be | |
2394 | * moved | |
2395 | */ | |
dd41f596 | 2396 | if (*imbalance + SCHED_LOAD_SCALE_FUZZ < busiest_load_per_task/2) { |
48f24c4d | 2397 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2398 | unsigned int imbn; |
2399 | ||
2400 | small_imbalance: | |
2401 | pwr_move = pwr_now = 0; | |
2402 | imbn = 2; | |
2403 | if (this_nr_running) { | |
2404 | this_load_per_task /= this_nr_running; | |
2405 | if (busiest_load_per_task > this_load_per_task) | |
2406 | imbn = 1; | |
2407 | } else | |
2408 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2409 | |
dd41f596 IM |
2410 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2411 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2412 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2413 | return busiest; |
2414 | } | |
2415 | ||
2416 | /* | |
2417 | * OK, we don't have enough imbalance to justify moving tasks, | |
2418 | * however we may be able to increase total CPU power used by | |
2419 | * moving them. | |
2420 | */ | |
2421 | ||
5517d86b ED |
2422 | pwr_now += busiest->__cpu_power * |
2423 | min(busiest_load_per_task, max_load); | |
2424 | pwr_now += this->__cpu_power * | |
2425 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2426 | pwr_now /= SCHED_LOAD_SCALE; |
2427 | ||
2428 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2429 | tmp = sg_div_cpu_power(busiest, |
2430 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2431 | if (max_load > tmp) |
5517d86b | 2432 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2433 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2434 | |
2435 | /* Amount of load we'd add */ | |
5517d86b | 2436 | if (max_load * busiest->__cpu_power < |
33859f7f | 2437 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2438 | tmp = sg_div_cpu_power(this, |
2439 | max_load * busiest->__cpu_power); | |
1da177e4 | 2440 | else |
5517d86b ED |
2441 | tmp = sg_div_cpu_power(this, |
2442 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2443 | pwr_move += this->__cpu_power * | |
2444 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2445 | pwr_move /= SCHED_LOAD_SCALE; |
2446 | ||
2447 | /* Move if we gain throughput */ | |
2448 | if (pwr_move <= pwr_now) | |
2449 | goto out_balanced; | |
2450 | ||
2dd73a4f | 2451 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2452 | } |
2453 | ||
1da177e4 LT |
2454 | return busiest; |
2455 | ||
2456 | out_balanced: | |
5c45bf27 | 2457 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2458 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2459 | goto ret; |
1da177e4 | 2460 | |
5c45bf27 SS |
2461 | if (this == group_leader && group_leader != group_min) { |
2462 | *imbalance = min_load_per_task; | |
2463 | return group_min; | |
2464 | } | |
5c45bf27 | 2465 | #endif |
783609c6 | 2466 | ret: |
1da177e4 LT |
2467 | *imbalance = 0; |
2468 | return NULL; | |
2469 | } | |
2470 | ||
2471 | /* | |
2472 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2473 | */ | |
70b97a7f | 2474 | static struct rq * |
d15bcfdb | 2475 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2476 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2477 | { |
70b97a7f | 2478 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2479 | unsigned long max_load = 0; |
1da177e4 LT |
2480 | int i; |
2481 | ||
2482 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2483 | unsigned long wl; |
0a2966b4 CL |
2484 | |
2485 | if (!cpu_isset(i, *cpus)) | |
2486 | continue; | |
2487 | ||
48f24c4d | 2488 | rq = cpu_rq(i); |
dd41f596 | 2489 | wl = weighted_cpuload(i); |
2dd73a4f | 2490 | |
dd41f596 | 2491 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2492 | continue; |
1da177e4 | 2493 | |
dd41f596 IM |
2494 | if (wl > max_load) { |
2495 | max_load = wl; | |
48f24c4d | 2496 | busiest = rq; |
1da177e4 LT |
2497 | } |
2498 | } | |
2499 | ||
2500 | return busiest; | |
2501 | } | |
2502 | ||
77391d71 NP |
2503 | /* |
2504 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2505 | * so long as it is large enough. | |
2506 | */ | |
2507 | #define MAX_PINNED_INTERVAL 512 | |
2508 | ||
48f24c4d IM |
2509 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2510 | { | |
2511 | return n > 0 ? n - 1 : 0; | |
2512 | } | |
2513 | ||
1da177e4 LT |
2514 | /* |
2515 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2516 | * tasks if there is an imbalance. | |
1da177e4 | 2517 | */ |
70b97a7f | 2518 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2519 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2520 | int *balance) |
1da177e4 | 2521 | { |
48f24c4d | 2522 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2523 | struct sched_group *group; |
1da177e4 | 2524 | unsigned long imbalance; |
70b97a7f | 2525 | struct rq *busiest; |
0a2966b4 | 2526 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2527 | unsigned long flags; |
5969fe06 | 2528 | |
89c4710e SS |
2529 | /* |
2530 | * When power savings policy is enabled for the parent domain, idle | |
2531 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2532 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2533 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2534 | */ |
d15bcfdb | 2535 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2536 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2537 | sd_idle = 1; |
1da177e4 | 2538 | |
1da177e4 LT |
2539 | schedstat_inc(sd, lb_cnt[idle]); |
2540 | ||
0a2966b4 CL |
2541 | redo: |
2542 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2543 | &cpus, balance); |
2544 | ||
06066714 | 2545 | if (*balance == 0) |
783609c6 | 2546 | goto out_balanced; |
783609c6 | 2547 | |
1da177e4 LT |
2548 | if (!group) { |
2549 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2550 | goto out_balanced; | |
2551 | } | |
2552 | ||
0a2966b4 | 2553 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2554 | if (!busiest) { |
2555 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2556 | goto out_balanced; | |
2557 | } | |
2558 | ||
db935dbd | 2559 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2560 | |
2561 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2562 | ||
2563 | nr_moved = 0; | |
2564 | if (busiest->nr_running > 1) { | |
2565 | /* | |
2566 | * Attempt to move tasks. If find_busiest_group has found | |
2567 | * an imbalance but busiest->nr_running <= 1, the group is | |
2568 | * still unbalanced. nr_moved simply stays zero, so it is | |
2569 | * correctly treated as an imbalance. | |
2570 | */ | |
fe2eea3f | 2571 | local_irq_save(flags); |
e17224bf | 2572 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2573 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2574 | minus_1_or_zero(busiest->nr_running), |
2575 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2576 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2577 | local_irq_restore(flags); |
81026794 | 2578 | |
46cb4b7c SS |
2579 | /* |
2580 | * some other cpu did the load balance for us. | |
2581 | */ | |
2582 | if (nr_moved && this_cpu != smp_processor_id()) | |
2583 | resched_cpu(this_cpu); | |
2584 | ||
81026794 | 2585 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2586 | if (unlikely(all_pinned)) { |
2587 | cpu_clear(cpu_of(busiest), cpus); | |
2588 | if (!cpus_empty(cpus)) | |
2589 | goto redo; | |
81026794 | 2590 | goto out_balanced; |
0a2966b4 | 2591 | } |
1da177e4 | 2592 | } |
81026794 | 2593 | |
1da177e4 LT |
2594 | if (!nr_moved) { |
2595 | schedstat_inc(sd, lb_failed[idle]); | |
2596 | sd->nr_balance_failed++; | |
2597 | ||
2598 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2599 | |
fe2eea3f | 2600 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2601 | |
2602 | /* don't kick the migration_thread, if the curr | |
2603 | * task on busiest cpu can't be moved to this_cpu | |
2604 | */ | |
2605 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2606 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2607 | all_pinned = 1; |
2608 | goto out_one_pinned; | |
2609 | } | |
2610 | ||
1da177e4 LT |
2611 | if (!busiest->active_balance) { |
2612 | busiest->active_balance = 1; | |
2613 | busiest->push_cpu = this_cpu; | |
81026794 | 2614 | active_balance = 1; |
1da177e4 | 2615 | } |
fe2eea3f | 2616 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2617 | if (active_balance) |
1da177e4 LT |
2618 | wake_up_process(busiest->migration_thread); |
2619 | ||
2620 | /* | |
2621 | * We've kicked active balancing, reset the failure | |
2622 | * counter. | |
2623 | */ | |
39507451 | 2624 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2625 | } |
81026794 | 2626 | } else |
1da177e4 LT |
2627 | sd->nr_balance_failed = 0; |
2628 | ||
81026794 | 2629 | if (likely(!active_balance)) { |
1da177e4 LT |
2630 | /* We were unbalanced, so reset the balancing interval */ |
2631 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2632 | } else { |
2633 | /* | |
2634 | * If we've begun active balancing, start to back off. This | |
2635 | * case may not be covered by the all_pinned logic if there | |
2636 | * is only 1 task on the busy runqueue (because we don't call | |
2637 | * move_tasks). | |
2638 | */ | |
2639 | if (sd->balance_interval < sd->max_interval) | |
2640 | sd->balance_interval *= 2; | |
1da177e4 LT |
2641 | } |
2642 | ||
5c45bf27 | 2643 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2644 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2645 | return -1; |
1da177e4 LT |
2646 | return nr_moved; |
2647 | ||
2648 | out_balanced: | |
1da177e4 LT |
2649 | schedstat_inc(sd, lb_balanced[idle]); |
2650 | ||
16cfb1c0 | 2651 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2652 | |
2653 | out_one_pinned: | |
1da177e4 | 2654 | /* tune up the balancing interval */ |
77391d71 NP |
2655 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2656 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2657 | sd->balance_interval *= 2; |
2658 | ||
48f24c4d | 2659 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2660 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2661 | return -1; |
1da177e4 LT |
2662 | return 0; |
2663 | } | |
2664 | ||
2665 | /* | |
2666 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2667 | * tasks if there is an imbalance. | |
2668 | * | |
d15bcfdb | 2669 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2670 | * this_rq is locked. |
2671 | */ | |
48f24c4d | 2672 | static int |
70b97a7f | 2673 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2674 | { |
2675 | struct sched_group *group; | |
70b97a7f | 2676 | struct rq *busiest = NULL; |
1da177e4 LT |
2677 | unsigned long imbalance; |
2678 | int nr_moved = 0; | |
5969fe06 | 2679 | int sd_idle = 0; |
0a2966b4 | 2680 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2681 | |
89c4710e SS |
2682 | /* |
2683 | * When power savings policy is enabled for the parent domain, idle | |
2684 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2685 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2686 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2687 | */ |
2688 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2689 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2690 | sd_idle = 1; |
1da177e4 | 2691 | |
d15bcfdb | 2692 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2693 | redo: |
d15bcfdb | 2694 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2695 | &sd_idle, &cpus, NULL); |
1da177e4 | 2696 | if (!group) { |
d15bcfdb | 2697 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2698 | goto out_balanced; |
1da177e4 LT |
2699 | } |
2700 | ||
d15bcfdb | 2701 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2702 | &cpus); |
db935dbd | 2703 | if (!busiest) { |
d15bcfdb | 2704 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2705 | goto out_balanced; |
1da177e4 LT |
2706 | } |
2707 | ||
db935dbd NP |
2708 | BUG_ON(busiest == this_rq); |
2709 | ||
d15bcfdb | 2710 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2711 | |
2712 | nr_moved = 0; | |
2713 | if (busiest->nr_running > 1) { | |
2714 | /* Attempt to move tasks */ | |
2715 | double_lock_balance(this_rq, busiest); | |
2716 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2717 | minus_1_or_zero(busiest->nr_running), |
d15bcfdb | 2718 | imbalance, sd, CPU_NEWLY_IDLE, NULL); |
d6d5cfaf | 2719 | spin_unlock(&busiest->lock); |
0a2966b4 CL |
2720 | |
2721 | if (!nr_moved) { | |
2722 | cpu_clear(cpu_of(busiest), cpus); | |
2723 | if (!cpus_empty(cpus)) | |
2724 | goto redo; | |
2725 | } | |
d6d5cfaf NP |
2726 | } |
2727 | ||
5969fe06 | 2728 | if (!nr_moved) { |
d15bcfdb | 2729 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2730 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2731 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2732 | return -1; |
2733 | } else | |
16cfb1c0 | 2734 | sd->nr_balance_failed = 0; |
1da177e4 | 2735 | |
1da177e4 | 2736 | return nr_moved; |
16cfb1c0 NP |
2737 | |
2738 | out_balanced: | |
d15bcfdb | 2739 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2740 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2741 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2742 | return -1; |
16cfb1c0 | 2743 | sd->nr_balance_failed = 0; |
48f24c4d | 2744 | |
16cfb1c0 | 2745 | return 0; |
1da177e4 LT |
2746 | } |
2747 | ||
2748 | /* | |
2749 | * idle_balance is called by schedule() if this_cpu is about to become | |
2750 | * idle. Attempts to pull tasks from other CPUs. | |
2751 | */ | |
70b97a7f | 2752 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2753 | { |
2754 | struct sched_domain *sd; | |
dd41f596 IM |
2755 | int pulled_task = -1; |
2756 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2757 | |
2758 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2759 | unsigned long interval; |
2760 | ||
2761 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2762 | continue; | |
2763 | ||
2764 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2765 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2766 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2767 | this_rq, sd); |
2768 | ||
2769 | interval = msecs_to_jiffies(sd->balance_interval); | |
2770 | if (time_after(next_balance, sd->last_balance + interval)) | |
2771 | next_balance = sd->last_balance + interval; | |
2772 | if (pulled_task) | |
2773 | break; | |
1da177e4 | 2774 | } |
dd41f596 | 2775 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2776 | /* |
2777 | * We are going idle. next_balance may be set based on | |
2778 | * a busy processor. So reset next_balance. | |
2779 | */ | |
2780 | this_rq->next_balance = next_balance; | |
dd41f596 | 2781 | } |
1da177e4 LT |
2782 | } |
2783 | ||
2784 | /* | |
2785 | * active_load_balance is run by migration threads. It pushes running tasks | |
2786 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2787 | * running on each physical CPU where possible, and avoids physical / | |
2788 | * logical imbalances. | |
2789 | * | |
2790 | * Called with busiest_rq locked. | |
2791 | */ | |
70b97a7f | 2792 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2793 | { |
39507451 | 2794 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2795 | struct sched_domain *sd; |
2796 | struct rq *target_rq; | |
39507451 | 2797 | |
48f24c4d | 2798 | /* Is there any task to move? */ |
39507451 | 2799 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2800 | return; |
2801 | ||
2802 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2803 | |
2804 | /* | |
39507451 NP |
2805 | * This condition is "impossible", if it occurs |
2806 | * we need to fix it. Originally reported by | |
2807 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2808 | */ |
39507451 | 2809 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2810 | |
39507451 NP |
2811 | /* move a task from busiest_rq to target_rq */ |
2812 | double_lock_balance(busiest_rq, target_rq); | |
2813 | ||
2814 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2815 | for_each_domain(target_cpu, sd) { |
39507451 | 2816 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2817 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2818 | break; |
c96d145e | 2819 | } |
39507451 | 2820 | |
48f24c4d IM |
2821 | if (likely(sd)) { |
2822 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2823 | |
48f24c4d | 2824 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
d15bcfdb | 2825 | RTPRIO_TO_LOAD_WEIGHT(100), sd, CPU_IDLE, |
48f24c4d IM |
2826 | NULL)) |
2827 | schedstat_inc(sd, alb_pushed); | |
2828 | else | |
2829 | schedstat_inc(sd, alb_failed); | |
2830 | } | |
39507451 | 2831 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2832 | } |
2833 | ||
46cb4b7c SS |
2834 | #ifdef CONFIG_NO_HZ |
2835 | static struct { | |
2836 | atomic_t load_balancer; | |
2837 | cpumask_t cpu_mask; | |
2838 | } nohz ____cacheline_aligned = { | |
2839 | .load_balancer = ATOMIC_INIT(-1), | |
2840 | .cpu_mask = CPU_MASK_NONE, | |
2841 | }; | |
2842 | ||
7835b98b | 2843 | /* |
46cb4b7c SS |
2844 | * This routine will try to nominate the ilb (idle load balancing) |
2845 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2846 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2847 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2848 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2849 | * arrives... | |
2850 | * | |
2851 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2852 | * for idle load balancing. ilb owner will still be part of | |
2853 | * nohz.cpu_mask.. | |
7835b98b | 2854 | * |
46cb4b7c SS |
2855 | * While stopping the tick, this cpu will become the ilb owner if there |
2856 | * is no other owner. And will be the owner till that cpu becomes busy | |
2857 | * or if all cpus in the system stop their ticks at which point | |
2858 | * there is no need for ilb owner. | |
2859 | * | |
2860 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2861 | * next busy scheduler_tick() | |
2862 | */ | |
2863 | int select_nohz_load_balancer(int stop_tick) | |
2864 | { | |
2865 | int cpu = smp_processor_id(); | |
2866 | ||
2867 | if (stop_tick) { | |
2868 | cpu_set(cpu, nohz.cpu_mask); | |
2869 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2870 | ||
2871 | /* | |
2872 | * If we are going offline and still the leader, give up! | |
2873 | */ | |
2874 | if (cpu_is_offline(cpu) && | |
2875 | atomic_read(&nohz.load_balancer) == cpu) { | |
2876 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2877 | BUG(); | |
2878 | return 0; | |
2879 | } | |
2880 | ||
2881 | /* time for ilb owner also to sleep */ | |
2882 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2883 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2884 | atomic_set(&nohz.load_balancer, -1); | |
2885 | return 0; | |
2886 | } | |
2887 | ||
2888 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2889 | /* make me the ilb owner */ | |
2890 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2891 | return 1; | |
2892 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2893 | return 1; | |
2894 | } else { | |
2895 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2896 | return 0; | |
2897 | ||
2898 | cpu_clear(cpu, nohz.cpu_mask); | |
2899 | ||
2900 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2901 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2902 | BUG(); | |
2903 | } | |
2904 | return 0; | |
2905 | } | |
2906 | #endif | |
2907 | ||
2908 | static DEFINE_SPINLOCK(balancing); | |
2909 | ||
2910 | /* | |
7835b98b CL |
2911 | * It checks each scheduling domain to see if it is due to be balanced, |
2912 | * and initiates a balancing operation if so. | |
2913 | * | |
2914 | * Balancing parameters are set up in arch_init_sched_domains. | |
2915 | */ | |
d15bcfdb | 2916 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 2917 | { |
46cb4b7c SS |
2918 | int balance = 1; |
2919 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
2920 | unsigned long interval; |
2921 | struct sched_domain *sd; | |
46cb4b7c | 2922 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 2923 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 2924 | |
46cb4b7c | 2925 | for_each_domain(cpu, sd) { |
1da177e4 LT |
2926 | if (!(sd->flags & SD_LOAD_BALANCE)) |
2927 | continue; | |
2928 | ||
2929 | interval = sd->balance_interval; | |
d15bcfdb | 2930 | if (idle != CPU_IDLE) |
1da177e4 LT |
2931 | interval *= sd->busy_factor; |
2932 | ||
2933 | /* scale ms to jiffies */ | |
2934 | interval = msecs_to_jiffies(interval); | |
2935 | if (unlikely(!interval)) | |
2936 | interval = 1; | |
dd41f596 IM |
2937 | if (interval > HZ*NR_CPUS/10) |
2938 | interval = HZ*NR_CPUS/10; | |
2939 | ||
1da177e4 | 2940 | |
08c183f3 CL |
2941 | if (sd->flags & SD_SERIALIZE) { |
2942 | if (!spin_trylock(&balancing)) | |
2943 | goto out; | |
2944 | } | |
2945 | ||
c9819f45 | 2946 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 2947 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
2948 | /* |
2949 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
2950 | * longer idle, or one of our SMT siblings is |
2951 | * not idle. | |
2952 | */ | |
d15bcfdb | 2953 | idle = CPU_NOT_IDLE; |
1da177e4 | 2954 | } |
1bd77f2d | 2955 | sd->last_balance = jiffies; |
1da177e4 | 2956 | } |
08c183f3 CL |
2957 | if (sd->flags & SD_SERIALIZE) |
2958 | spin_unlock(&balancing); | |
2959 | out: | |
c9819f45 CL |
2960 | if (time_after(next_balance, sd->last_balance + interval)) |
2961 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
2962 | |
2963 | /* | |
2964 | * Stop the load balance at this level. There is another | |
2965 | * CPU in our sched group which is doing load balancing more | |
2966 | * actively. | |
2967 | */ | |
2968 | if (!balance) | |
2969 | break; | |
1da177e4 | 2970 | } |
46cb4b7c SS |
2971 | rq->next_balance = next_balance; |
2972 | } | |
2973 | ||
2974 | /* | |
2975 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
2976 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
2977 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
2978 | */ | |
2979 | static void run_rebalance_domains(struct softirq_action *h) | |
2980 | { | |
dd41f596 IM |
2981 | int this_cpu = smp_processor_id(); |
2982 | struct rq *this_rq = cpu_rq(this_cpu); | |
2983 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
2984 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 2985 | |
dd41f596 | 2986 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
2987 | |
2988 | #ifdef CONFIG_NO_HZ | |
2989 | /* | |
2990 | * If this cpu is the owner for idle load balancing, then do the | |
2991 | * balancing on behalf of the other idle cpus whose ticks are | |
2992 | * stopped. | |
2993 | */ | |
dd41f596 IM |
2994 | if (this_rq->idle_at_tick && |
2995 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
2996 | cpumask_t cpus = nohz.cpu_mask; |
2997 | struct rq *rq; | |
2998 | int balance_cpu; | |
2999 | ||
dd41f596 | 3000 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3001 | for_each_cpu_mask(balance_cpu, cpus) { |
3002 | /* | |
3003 | * If this cpu gets work to do, stop the load balancing | |
3004 | * work being done for other cpus. Next load | |
3005 | * balancing owner will pick it up. | |
3006 | */ | |
3007 | if (need_resched()) | |
3008 | break; | |
3009 | ||
dd41f596 | 3010 | rebalance_domains(balance_cpu, SCHED_IDLE); |
46cb4b7c SS |
3011 | |
3012 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3013 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3014 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3015 | } |
3016 | } | |
3017 | #endif | |
3018 | } | |
3019 | ||
3020 | /* | |
3021 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3022 | * | |
3023 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3024 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3025 | * if the whole system is idle. | |
3026 | */ | |
dd41f596 | 3027 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3028 | { |
46cb4b7c SS |
3029 | #ifdef CONFIG_NO_HZ |
3030 | /* | |
3031 | * If we were in the nohz mode recently and busy at the current | |
3032 | * scheduler tick, then check if we need to nominate new idle | |
3033 | * load balancer. | |
3034 | */ | |
3035 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3036 | rq->in_nohz_recently = 0; | |
3037 | ||
3038 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3039 | cpu_clear(cpu, nohz.cpu_mask); | |
3040 | atomic_set(&nohz.load_balancer, -1); | |
3041 | } | |
3042 | ||
3043 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3044 | /* | |
3045 | * simple selection for now: Nominate the | |
3046 | * first cpu in the nohz list to be the next | |
3047 | * ilb owner. | |
3048 | * | |
3049 | * TBD: Traverse the sched domains and nominate | |
3050 | * the nearest cpu in the nohz.cpu_mask. | |
3051 | */ | |
3052 | int ilb = first_cpu(nohz.cpu_mask); | |
3053 | ||
3054 | if (ilb != NR_CPUS) | |
3055 | resched_cpu(ilb); | |
3056 | } | |
3057 | } | |
3058 | ||
3059 | /* | |
3060 | * If this cpu is idle and doing idle load balancing for all the | |
3061 | * cpus with ticks stopped, is it time for that to stop? | |
3062 | */ | |
3063 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3064 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3065 | resched_cpu(cpu); | |
3066 | return; | |
3067 | } | |
3068 | ||
3069 | /* | |
3070 | * If this cpu is idle and the idle load balancing is done by | |
3071 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3072 | */ | |
3073 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3074 | cpu_isset(cpu, nohz.cpu_mask)) | |
3075 | return; | |
3076 | #endif | |
3077 | if (time_after_eq(jiffies, rq->next_balance)) | |
3078 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3079 | } |
dd41f596 IM |
3080 | |
3081 | #else /* CONFIG_SMP */ | |
3082 | ||
1da177e4 LT |
3083 | /* |
3084 | * on UP we do not need to balance between CPUs: | |
3085 | */ | |
70b97a7f | 3086 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3087 | { |
3088 | } | |
dd41f596 IM |
3089 | |
3090 | /* Avoid "used but not defined" warning on UP */ | |
3091 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3092 | unsigned long max_nr_move, unsigned long max_load_move, | |
3093 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3094 | int *all_pinned, unsigned long *load_moved, | |
3095 | int this_best_prio, int best_prio, int best_prio_seen, | |
3096 | struct rq_iterator *iterator) | |
3097 | { | |
3098 | *load_moved = 0; | |
3099 | ||
3100 | return 0; | |
3101 | } | |
3102 | ||
1da177e4 LT |
3103 | #endif |
3104 | ||
1da177e4 LT |
3105 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3106 | ||
3107 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3108 | ||
3109 | /* | |
41b86e9c IM |
3110 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3111 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3112 | */ |
41b86e9c | 3113 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3114 | { |
1da177e4 | 3115 | unsigned long flags; |
41b86e9c IM |
3116 | u64 ns, delta_exec; |
3117 | struct rq *rq; | |
48f24c4d | 3118 | |
41b86e9c IM |
3119 | rq = task_rq_lock(p, &flags); |
3120 | ns = p->se.sum_exec_runtime; | |
3121 | if (rq->curr == p) { | |
3122 | delta_exec = rq_clock(rq) - p->se.exec_start; | |
3123 | if ((s64)delta_exec > 0) | |
3124 | ns += delta_exec; | |
3125 | } | |
3126 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3127 | |
1da177e4 LT |
3128 | return ns; |
3129 | } | |
3130 | ||
1da177e4 LT |
3131 | /* |
3132 | * Account user cpu time to a process. | |
3133 | * @p: the process that the cpu time gets accounted to | |
3134 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3135 | * @cputime: the cpu time spent in user space since the last update | |
3136 | */ | |
3137 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3138 | { | |
3139 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3140 | cputime64_t tmp; | |
3141 | ||
3142 | p->utime = cputime_add(p->utime, cputime); | |
3143 | ||
3144 | /* Add user time to cpustat. */ | |
3145 | tmp = cputime_to_cputime64(cputime); | |
3146 | if (TASK_NICE(p) > 0) | |
3147 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3148 | else | |
3149 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * Account system cpu time to a process. | |
3154 | * @p: the process that the cpu time gets accounted to | |
3155 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3156 | * @cputime: the cpu time spent in kernel space since the last update | |
3157 | */ | |
3158 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3159 | cputime_t cputime) | |
3160 | { | |
3161 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3162 | struct rq *rq = this_rq(); |
1da177e4 LT |
3163 | cputime64_t tmp; |
3164 | ||
3165 | p->stime = cputime_add(p->stime, cputime); | |
3166 | ||
3167 | /* Add system time to cpustat. */ | |
3168 | tmp = cputime_to_cputime64(cputime); | |
3169 | if (hardirq_count() - hardirq_offset) | |
3170 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3171 | else if (softirq_count()) | |
3172 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3173 | else if (p != rq->idle) | |
3174 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3175 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3176 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3177 | else | |
3178 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3179 | /* Account for system time used */ | |
3180 | acct_update_integrals(p); | |
1da177e4 LT |
3181 | } |
3182 | ||
3183 | /* | |
3184 | * Account for involuntary wait time. | |
3185 | * @p: the process from which the cpu time has been stolen | |
3186 | * @steal: the cpu time spent in involuntary wait | |
3187 | */ | |
3188 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3189 | { | |
3190 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3191 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3192 | struct rq *rq = this_rq(); |
1da177e4 LT |
3193 | |
3194 | if (p == rq->idle) { | |
3195 | p->stime = cputime_add(p->stime, steal); | |
3196 | if (atomic_read(&rq->nr_iowait) > 0) | |
3197 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3198 | else | |
3199 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3200 | } else | |
3201 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3202 | } | |
3203 | ||
7835b98b CL |
3204 | /* |
3205 | * This function gets called by the timer code, with HZ frequency. | |
3206 | * We call it with interrupts disabled. | |
3207 | * | |
3208 | * It also gets called by the fork code, when changing the parent's | |
3209 | * timeslices. | |
3210 | */ | |
3211 | void scheduler_tick(void) | |
3212 | { | |
7835b98b CL |
3213 | int cpu = smp_processor_id(); |
3214 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 IM |
3215 | struct task_struct *curr = rq->curr; |
3216 | ||
3217 | spin_lock(&rq->lock); | |
3218 | if (curr != rq->idle) /* FIXME: needed? */ | |
3219 | curr->sched_class->task_tick(rq, curr); | |
3220 | update_cpu_load(rq); | |
3221 | spin_unlock(&rq->lock); | |
7835b98b | 3222 | |
e418e1c2 | 3223 | #ifdef CONFIG_SMP |
dd41f596 IM |
3224 | rq->idle_at_tick = idle_cpu(cpu); |
3225 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3226 | #endif |
1da177e4 LT |
3227 | } |
3228 | ||
1da177e4 LT |
3229 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3230 | ||
3231 | void fastcall add_preempt_count(int val) | |
3232 | { | |
3233 | /* | |
3234 | * Underflow? | |
3235 | */ | |
9a11b49a IM |
3236 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3237 | return; | |
1da177e4 LT |
3238 | preempt_count() += val; |
3239 | /* | |
3240 | * Spinlock count overflowing soon? | |
3241 | */ | |
33859f7f MOS |
3242 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3243 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3244 | } |
3245 | EXPORT_SYMBOL(add_preempt_count); | |
3246 | ||
3247 | void fastcall sub_preempt_count(int val) | |
3248 | { | |
3249 | /* | |
3250 | * Underflow? | |
3251 | */ | |
9a11b49a IM |
3252 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3253 | return; | |
1da177e4 LT |
3254 | /* |
3255 | * Is the spinlock portion underflowing? | |
3256 | */ | |
9a11b49a IM |
3257 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3258 | !(preempt_count() & PREEMPT_MASK))) | |
3259 | return; | |
3260 | ||
1da177e4 LT |
3261 | preempt_count() -= val; |
3262 | } | |
3263 | EXPORT_SYMBOL(sub_preempt_count); | |
3264 | ||
3265 | #endif | |
3266 | ||
3267 | /* | |
dd41f596 | 3268 | * Print scheduling while atomic bug: |
1da177e4 | 3269 | */ |
dd41f596 | 3270 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3271 | { |
dd41f596 IM |
3272 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3273 | prev->comm, preempt_count(), prev->pid); | |
3274 | debug_show_held_locks(prev); | |
3275 | if (irqs_disabled()) | |
3276 | print_irqtrace_events(prev); | |
3277 | dump_stack(); | |
3278 | } | |
1da177e4 | 3279 | |
dd41f596 IM |
3280 | /* |
3281 | * Various schedule()-time debugging checks and statistics: | |
3282 | */ | |
3283 | static inline void schedule_debug(struct task_struct *prev) | |
3284 | { | |
1da177e4 LT |
3285 | /* |
3286 | * Test if we are atomic. Since do_exit() needs to call into | |
3287 | * schedule() atomically, we ignore that path for now. | |
3288 | * Otherwise, whine if we are scheduling when we should not be. | |
3289 | */ | |
dd41f596 IM |
3290 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3291 | __schedule_bug(prev); | |
3292 | ||
1da177e4 LT |
3293 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3294 | ||
dd41f596 IM |
3295 | schedstat_inc(this_rq(), sched_cnt); |
3296 | } | |
3297 | ||
3298 | /* | |
3299 | * Pick up the highest-prio task: | |
3300 | */ | |
3301 | static inline struct task_struct * | |
3302 | pick_next_task(struct rq *rq, struct task_struct *prev, u64 now) | |
3303 | { | |
3304 | struct sched_class *class; | |
3305 | struct task_struct *p; | |
1da177e4 LT |
3306 | |
3307 | /* | |
dd41f596 IM |
3308 | * Optimization: we know that if all tasks are in |
3309 | * the fair class we can call that function directly: | |
1da177e4 | 3310 | */ |
dd41f596 IM |
3311 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
3312 | p = fair_sched_class.pick_next_task(rq, now); | |
3313 | if (likely(p)) | |
3314 | return p; | |
1da177e4 LT |
3315 | } |
3316 | ||
dd41f596 IM |
3317 | class = sched_class_highest; |
3318 | for ( ; ; ) { | |
3319 | p = class->pick_next_task(rq, now); | |
3320 | if (p) | |
3321 | return p; | |
3322 | /* | |
3323 | * Will never be NULL as the idle class always | |
3324 | * returns a non-NULL p: | |
3325 | */ | |
3326 | class = class->next; | |
3327 | } | |
3328 | } | |
1da177e4 | 3329 | |
dd41f596 IM |
3330 | /* |
3331 | * schedule() is the main scheduler function. | |
3332 | */ | |
3333 | asmlinkage void __sched schedule(void) | |
3334 | { | |
3335 | struct task_struct *prev, *next; | |
3336 | long *switch_count; | |
3337 | struct rq *rq; | |
3338 | u64 now; | |
3339 | int cpu; | |
3340 | ||
3341 | need_resched: | |
3342 | preempt_disable(); | |
3343 | cpu = smp_processor_id(); | |
3344 | rq = cpu_rq(cpu); | |
3345 | rcu_qsctr_inc(cpu); | |
3346 | prev = rq->curr; | |
3347 | switch_count = &prev->nivcsw; | |
3348 | ||
3349 | release_kernel_lock(prev); | |
3350 | need_resched_nonpreemptible: | |
3351 | ||
3352 | schedule_debug(prev); | |
1da177e4 LT |
3353 | |
3354 | spin_lock_irq(&rq->lock); | |
dd41f596 | 3355 | clear_tsk_need_resched(prev); |
1da177e4 | 3356 | |
1da177e4 | 3357 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3358 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3359 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3360 | prev->state = TASK_RUNNING; |
dd41f596 IM |
3361 | } else { |
3362 | deactivate_task(rq, prev, 1); | |
1da177e4 | 3363 | } |
dd41f596 | 3364 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3365 | } |
3366 | ||
dd41f596 | 3367 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3368 | idle_balance(cpu, rq); |
1da177e4 | 3369 | |
dd41f596 IM |
3370 | now = __rq_clock(rq); |
3371 | prev->sched_class->put_prev_task(rq, prev, now); | |
3372 | next = pick_next_task(rq, prev, now); | |
1da177e4 LT |
3373 | |
3374 | sched_info_switch(prev, next); | |
dd41f596 | 3375 | |
1da177e4 | 3376 | if (likely(prev != next)) { |
1da177e4 LT |
3377 | rq->nr_switches++; |
3378 | rq->curr = next; | |
3379 | ++*switch_count; | |
3380 | ||
dd41f596 | 3381 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3382 | } else |
3383 | spin_unlock_irq(&rq->lock); | |
3384 | ||
dd41f596 IM |
3385 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3386 | cpu = smp_processor_id(); | |
3387 | rq = cpu_rq(cpu); | |
1da177e4 | 3388 | goto need_resched_nonpreemptible; |
dd41f596 | 3389 | } |
1da177e4 LT |
3390 | preempt_enable_no_resched(); |
3391 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3392 | goto need_resched; | |
3393 | } | |
1da177e4 LT |
3394 | EXPORT_SYMBOL(schedule); |
3395 | ||
3396 | #ifdef CONFIG_PREEMPT | |
3397 | /* | |
2ed6e34f | 3398 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3399 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3400 | * occur there and call schedule directly. | |
3401 | */ | |
3402 | asmlinkage void __sched preempt_schedule(void) | |
3403 | { | |
3404 | struct thread_info *ti = current_thread_info(); | |
3405 | #ifdef CONFIG_PREEMPT_BKL | |
3406 | struct task_struct *task = current; | |
3407 | int saved_lock_depth; | |
3408 | #endif | |
3409 | /* | |
3410 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3411 | * we do not want to preempt the current task. Just return.. | |
3412 | */ | |
beed33a8 | 3413 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3414 | return; |
3415 | ||
3416 | need_resched: | |
3417 | add_preempt_count(PREEMPT_ACTIVE); | |
3418 | /* | |
3419 | * We keep the big kernel semaphore locked, but we | |
3420 | * clear ->lock_depth so that schedule() doesnt | |
3421 | * auto-release the semaphore: | |
3422 | */ | |
3423 | #ifdef CONFIG_PREEMPT_BKL | |
3424 | saved_lock_depth = task->lock_depth; | |
3425 | task->lock_depth = -1; | |
3426 | #endif | |
3427 | schedule(); | |
3428 | #ifdef CONFIG_PREEMPT_BKL | |
3429 | task->lock_depth = saved_lock_depth; | |
3430 | #endif | |
3431 | sub_preempt_count(PREEMPT_ACTIVE); | |
3432 | ||
3433 | /* we could miss a preemption opportunity between schedule and now */ | |
3434 | barrier(); | |
3435 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3436 | goto need_resched; | |
3437 | } | |
1da177e4 LT |
3438 | EXPORT_SYMBOL(preempt_schedule); |
3439 | ||
3440 | /* | |
2ed6e34f | 3441 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3442 | * off of irq context. |
3443 | * Note, that this is called and return with irqs disabled. This will | |
3444 | * protect us against recursive calling from irq. | |
3445 | */ | |
3446 | asmlinkage void __sched preempt_schedule_irq(void) | |
3447 | { | |
3448 | struct thread_info *ti = current_thread_info(); | |
3449 | #ifdef CONFIG_PREEMPT_BKL | |
3450 | struct task_struct *task = current; | |
3451 | int saved_lock_depth; | |
3452 | #endif | |
2ed6e34f | 3453 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3454 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3455 | ||
3456 | need_resched: | |
3457 | add_preempt_count(PREEMPT_ACTIVE); | |
3458 | /* | |
3459 | * We keep the big kernel semaphore locked, but we | |
3460 | * clear ->lock_depth so that schedule() doesnt | |
3461 | * auto-release the semaphore: | |
3462 | */ | |
3463 | #ifdef CONFIG_PREEMPT_BKL | |
3464 | saved_lock_depth = task->lock_depth; | |
3465 | task->lock_depth = -1; | |
3466 | #endif | |
3467 | local_irq_enable(); | |
3468 | schedule(); | |
3469 | local_irq_disable(); | |
3470 | #ifdef CONFIG_PREEMPT_BKL | |
3471 | task->lock_depth = saved_lock_depth; | |
3472 | #endif | |
3473 | sub_preempt_count(PREEMPT_ACTIVE); | |
3474 | ||
3475 | /* we could miss a preemption opportunity between schedule and now */ | |
3476 | barrier(); | |
3477 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3478 | goto need_resched; | |
3479 | } | |
3480 | ||
3481 | #endif /* CONFIG_PREEMPT */ | |
3482 | ||
95cdf3b7 IM |
3483 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3484 | void *key) | |
1da177e4 | 3485 | { |
48f24c4d | 3486 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3487 | } |
1da177e4 LT |
3488 | EXPORT_SYMBOL(default_wake_function); |
3489 | ||
3490 | /* | |
3491 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3492 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3493 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3494 | * | |
3495 | * There are circumstances in which we can try to wake a task which has already | |
3496 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3497 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3498 | */ | |
3499 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3500 | int nr_exclusive, int sync, void *key) | |
3501 | { | |
3502 | struct list_head *tmp, *next; | |
3503 | ||
3504 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3505 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3506 | unsigned flags = curr->flags; | |
3507 | ||
1da177e4 | 3508 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3509 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3510 | break; |
3511 | } | |
3512 | } | |
3513 | ||
3514 | /** | |
3515 | * __wake_up - wake up threads blocked on a waitqueue. | |
3516 | * @q: the waitqueue | |
3517 | * @mode: which threads | |
3518 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3519 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3520 | */ |
3521 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3522 | int nr_exclusive, void *key) |
1da177e4 LT |
3523 | { |
3524 | unsigned long flags; | |
3525 | ||
3526 | spin_lock_irqsave(&q->lock, flags); | |
3527 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3528 | spin_unlock_irqrestore(&q->lock, flags); | |
3529 | } | |
1da177e4 LT |
3530 | EXPORT_SYMBOL(__wake_up); |
3531 | ||
3532 | /* | |
3533 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3534 | */ | |
3535 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3536 | { | |
3537 | __wake_up_common(q, mode, 1, 0, NULL); | |
3538 | } | |
3539 | ||
3540 | /** | |
67be2dd1 | 3541 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3542 | * @q: the waitqueue |
3543 | * @mode: which threads | |
3544 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3545 | * | |
3546 | * The sync wakeup differs that the waker knows that it will schedule | |
3547 | * away soon, so while the target thread will be woken up, it will not | |
3548 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3549 | * with each other. This can prevent needless bouncing between CPUs. | |
3550 | * | |
3551 | * On UP it can prevent extra preemption. | |
3552 | */ | |
95cdf3b7 IM |
3553 | void fastcall |
3554 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3555 | { |
3556 | unsigned long flags; | |
3557 | int sync = 1; | |
3558 | ||
3559 | if (unlikely(!q)) | |
3560 | return; | |
3561 | ||
3562 | if (unlikely(!nr_exclusive)) | |
3563 | sync = 0; | |
3564 | ||
3565 | spin_lock_irqsave(&q->lock, flags); | |
3566 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3567 | spin_unlock_irqrestore(&q->lock, flags); | |
3568 | } | |
3569 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3570 | ||
3571 | void fastcall complete(struct completion *x) | |
3572 | { | |
3573 | unsigned long flags; | |
3574 | ||
3575 | spin_lock_irqsave(&x->wait.lock, flags); | |
3576 | x->done++; | |
3577 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3578 | 1, 0, NULL); | |
3579 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3580 | } | |
3581 | EXPORT_SYMBOL(complete); | |
3582 | ||
3583 | void fastcall complete_all(struct completion *x) | |
3584 | { | |
3585 | unsigned long flags; | |
3586 | ||
3587 | spin_lock_irqsave(&x->wait.lock, flags); | |
3588 | x->done += UINT_MAX/2; | |
3589 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3590 | 0, 0, NULL); | |
3591 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3592 | } | |
3593 | EXPORT_SYMBOL(complete_all); | |
3594 | ||
3595 | void fastcall __sched wait_for_completion(struct completion *x) | |
3596 | { | |
3597 | might_sleep(); | |
48f24c4d | 3598 | |
1da177e4 LT |
3599 | spin_lock_irq(&x->wait.lock); |
3600 | if (!x->done) { | |
3601 | DECLARE_WAITQUEUE(wait, current); | |
3602 | ||
3603 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3604 | __add_wait_queue_tail(&x->wait, &wait); | |
3605 | do { | |
3606 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3607 | spin_unlock_irq(&x->wait.lock); | |
3608 | schedule(); | |
3609 | spin_lock_irq(&x->wait.lock); | |
3610 | } while (!x->done); | |
3611 | __remove_wait_queue(&x->wait, &wait); | |
3612 | } | |
3613 | x->done--; | |
3614 | spin_unlock_irq(&x->wait.lock); | |
3615 | } | |
3616 | EXPORT_SYMBOL(wait_for_completion); | |
3617 | ||
3618 | unsigned long fastcall __sched | |
3619 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3620 | { | |
3621 | might_sleep(); | |
3622 | ||
3623 | spin_lock_irq(&x->wait.lock); | |
3624 | if (!x->done) { | |
3625 | DECLARE_WAITQUEUE(wait, current); | |
3626 | ||
3627 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3628 | __add_wait_queue_tail(&x->wait, &wait); | |
3629 | do { | |
3630 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3631 | spin_unlock_irq(&x->wait.lock); | |
3632 | timeout = schedule_timeout(timeout); | |
3633 | spin_lock_irq(&x->wait.lock); | |
3634 | if (!timeout) { | |
3635 | __remove_wait_queue(&x->wait, &wait); | |
3636 | goto out; | |
3637 | } | |
3638 | } while (!x->done); | |
3639 | __remove_wait_queue(&x->wait, &wait); | |
3640 | } | |
3641 | x->done--; | |
3642 | out: | |
3643 | spin_unlock_irq(&x->wait.lock); | |
3644 | return timeout; | |
3645 | } | |
3646 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3647 | ||
3648 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3649 | { | |
3650 | int ret = 0; | |
3651 | ||
3652 | might_sleep(); | |
3653 | ||
3654 | spin_lock_irq(&x->wait.lock); | |
3655 | if (!x->done) { | |
3656 | DECLARE_WAITQUEUE(wait, current); | |
3657 | ||
3658 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3659 | __add_wait_queue_tail(&x->wait, &wait); | |
3660 | do { | |
3661 | if (signal_pending(current)) { | |
3662 | ret = -ERESTARTSYS; | |
3663 | __remove_wait_queue(&x->wait, &wait); | |
3664 | goto out; | |
3665 | } | |
3666 | __set_current_state(TASK_INTERRUPTIBLE); | |
3667 | spin_unlock_irq(&x->wait.lock); | |
3668 | schedule(); | |
3669 | spin_lock_irq(&x->wait.lock); | |
3670 | } while (!x->done); | |
3671 | __remove_wait_queue(&x->wait, &wait); | |
3672 | } | |
3673 | x->done--; | |
3674 | out: | |
3675 | spin_unlock_irq(&x->wait.lock); | |
3676 | ||
3677 | return ret; | |
3678 | } | |
3679 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3680 | ||
3681 | unsigned long fastcall __sched | |
3682 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3683 | unsigned long timeout) | |
3684 | { | |
3685 | might_sleep(); | |
3686 | ||
3687 | spin_lock_irq(&x->wait.lock); | |
3688 | if (!x->done) { | |
3689 | DECLARE_WAITQUEUE(wait, current); | |
3690 | ||
3691 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3692 | __add_wait_queue_tail(&x->wait, &wait); | |
3693 | do { | |
3694 | if (signal_pending(current)) { | |
3695 | timeout = -ERESTARTSYS; | |
3696 | __remove_wait_queue(&x->wait, &wait); | |
3697 | goto out; | |
3698 | } | |
3699 | __set_current_state(TASK_INTERRUPTIBLE); | |
3700 | spin_unlock_irq(&x->wait.lock); | |
3701 | timeout = schedule_timeout(timeout); | |
3702 | spin_lock_irq(&x->wait.lock); | |
3703 | if (!timeout) { | |
3704 | __remove_wait_queue(&x->wait, &wait); | |
3705 | goto out; | |
3706 | } | |
3707 | } while (!x->done); | |
3708 | __remove_wait_queue(&x->wait, &wait); | |
3709 | } | |
3710 | x->done--; | |
3711 | out: | |
3712 | spin_unlock_irq(&x->wait.lock); | |
3713 | return timeout; | |
3714 | } | |
3715 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3716 | ||
0fec171c IM |
3717 | static inline void |
3718 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3719 | { | |
3720 | spin_lock_irqsave(&q->lock, *flags); | |
3721 | __add_wait_queue(q, wait); | |
1da177e4 | 3722 | spin_unlock(&q->lock); |
0fec171c | 3723 | } |
1da177e4 | 3724 | |
0fec171c IM |
3725 | static inline void |
3726 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3727 | { | |
3728 | spin_lock_irq(&q->lock); | |
3729 | __remove_wait_queue(q, wait); | |
3730 | spin_unlock_irqrestore(&q->lock, *flags); | |
3731 | } | |
1da177e4 | 3732 | |
0fec171c | 3733 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3734 | { |
0fec171c IM |
3735 | unsigned long flags; |
3736 | wait_queue_t wait; | |
3737 | ||
3738 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3739 | |
3740 | current->state = TASK_INTERRUPTIBLE; | |
3741 | ||
0fec171c | 3742 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3743 | schedule(); |
0fec171c | 3744 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3745 | } |
1da177e4 LT |
3746 | EXPORT_SYMBOL(interruptible_sleep_on); |
3747 | ||
0fec171c | 3748 | long __sched |
95cdf3b7 | 3749 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3750 | { |
0fec171c IM |
3751 | unsigned long flags; |
3752 | wait_queue_t wait; | |
3753 | ||
3754 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3755 | |
3756 | current->state = TASK_INTERRUPTIBLE; | |
3757 | ||
0fec171c | 3758 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3759 | timeout = schedule_timeout(timeout); |
0fec171c | 3760 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3761 | |
3762 | return timeout; | |
3763 | } | |
1da177e4 LT |
3764 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3765 | ||
0fec171c | 3766 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3767 | { |
0fec171c IM |
3768 | unsigned long flags; |
3769 | wait_queue_t wait; | |
3770 | ||
3771 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3772 | |
3773 | current->state = TASK_UNINTERRUPTIBLE; | |
3774 | ||
0fec171c | 3775 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3776 | schedule(); |
0fec171c | 3777 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3778 | } |
1da177e4 LT |
3779 | EXPORT_SYMBOL(sleep_on); |
3780 | ||
0fec171c | 3781 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3782 | { |
0fec171c IM |
3783 | unsigned long flags; |
3784 | wait_queue_t wait; | |
3785 | ||
3786 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3787 | |
3788 | current->state = TASK_UNINTERRUPTIBLE; | |
3789 | ||
0fec171c | 3790 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3791 | timeout = schedule_timeout(timeout); |
0fec171c | 3792 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3793 | |
3794 | return timeout; | |
3795 | } | |
1da177e4 LT |
3796 | EXPORT_SYMBOL(sleep_on_timeout); |
3797 | ||
b29739f9 IM |
3798 | #ifdef CONFIG_RT_MUTEXES |
3799 | ||
3800 | /* | |
3801 | * rt_mutex_setprio - set the current priority of a task | |
3802 | * @p: task | |
3803 | * @prio: prio value (kernel-internal form) | |
3804 | * | |
3805 | * This function changes the 'effective' priority of a task. It does | |
3806 | * not touch ->normal_prio like __setscheduler(). | |
3807 | * | |
3808 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3809 | */ | |
36c8b586 | 3810 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3811 | { |
3812 | unsigned long flags; | |
dd41f596 | 3813 | int oldprio, on_rq; |
70b97a7f | 3814 | struct rq *rq; |
dd41f596 | 3815 | u64 now; |
b29739f9 IM |
3816 | |
3817 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3818 | ||
3819 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3820 | now = rq_clock(rq); |
b29739f9 | 3821 | |
d5f9f942 | 3822 | oldprio = p->prio; |
dd41f596 IM |
3823 | on_rq = p->se.on_rq; |
3824 | if (on_rq) | |
3825 | dequeue_task(rq, p, 0, now); | |
3826 | ||
3827 | if (rt_prio(prio)) | |
3828 | p->sched_class = &rt_sched_class; | |
3829 | else | |
3830 | p->sched_class = &fair_sched_class; | |
3831 | ||
b29739f9 IM |
3832 | p->prio = prio; |
3833 | ||
dd41f596 IM |
3834 | if (on_rq) { |
3835 | enqueue_task(rq, p, 0, now); | |
b29739f9 IM |
3836 | /* |
3837 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3838 | * our priority decreased, or if we are not currently running on |
3839 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3840 | */ |
d5f9f942 AM |
3841 | if (task_running(rq, p)) { |
3842 | if (p->prio > oldprio) | |
3843 | resched_task(rq->curr); | |
dd41f596 IM |
3844 | } else { |
3845 | check_preempt_curr(rq, p); | |
3846 | } | |
b29739f9 IM |
3847 | } |
3848 | task_rq_unlock(rq, &flags); | |
3849 | } | |
3850 | ||
3851 | #endif | |
3852 | ||
36c8b586 | 3853 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3854 | { |
dd41f596 | 3855 | int old_prio, delta, on_rq; |
1da177e4 | 3856 | unsigned long flags; |
70b97a7f | 3857 | struct rq *rq; |
dd41f596 | 3858 | u64 now; |
1da177e4 LT |
3859 | |
3860 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3861 | return; | |
3862 | /* | |
3863 | * We have to be careful, if called from sys_setpriority(), | |
3864 | * the task might be in the middle of scheduling on another CPU. | |
3865 | */ | |
3866 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3867 | now = rq_clock(rq); |
1da177e4 LT |
3868 | /* |
3869 | * The RT priorities are set via sched_setscheduler(), but we still | |
3870 | * allow the 'normal' nice value to be set - but as expected | |
3871 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3872 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3873 | */ |
e05606d3 | 3874 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3875 | p->static_prio = NICE_TO_PRIO(nice); |
3876 | goto out_unlock; | |
3877 | } | |
dd41f596 IM |
3878 | on_rq = p->se.on_rq; |
3879 | if (on_rq) { | |
3880 | dequeue_task(rq, p, 0, now); | |
3881 | dec_load(rq, p, now); | |
2dd73a4f | 3882 | } |
1da177e4 | 3883 | |
1da177e4 | 3884 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3885 | set_load_weight(p); |
b29739f9 IM |
3886 | old_prio = p->prio; |
3887 | p->prio = effective_prio(p); | |
3888 | delta = p->prio - old_prio; | |
1da177e4 | 3889 | |
dd41f596 IM |
3890 | if (on_rq) { |
3891 | enqueue_task(rq, p, 0, now); | |
3892 | inc_load(rq, p, now); | |
1da177e4 | 3893 | /* |
d5f9f942 AM |
3894 | * If the task increased its priority or is running and |
3895 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3896 | */ |
d5f9f942 | 3897 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3898 | resched_task(rq->curr); |
3899 | } | |
3900 | out_unlock: | |
3901 | task_rq_unlock(rq, &flags); | |
3902 | } | |
1da177e4 LT |
3903 | EXPORT_SYMBOL(set_user_nice); |
3904 | ||
e43379f1 MM |
3905 | /* |
3906 | * can_nice - check if a task can reduce its nice value | |
3907 | * @p: task | |
3908 | * @nice: nice value | |
3909 | */ | |
36c8b586 | 3910 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3911 | { |
024f4747 MM |
3912 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3913 | int nice_rlim = 20 - nice; | |
48f24c4d | 3914 | |
e43379f1 MM |
3915 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
3916 | capable(CAP_SYS_NICE)); | |
3917 | } | |
3918 | ||
1da177e4 LT |
3919 | #ifdef __ARCH_WANT_SYS_NICE |
3920 | ||
3921 | /* | |
3922 | * sys_nice - change the priority of the current process. | |
3923 | * @increment: priority increment | |
3924 | * | |
3925 | * sys_setpriority is a more generic, but much slower function that | |
3926 | * does similar things. | |
3927 | */ | |
3928 | asmlinkage long sys_nice(int increment) | |
3929 | { | |
48f24c4d | 3930 | long nice, retval; |
1da177e4 LT |
3931 | |
3932 | /* | |
3933 | * Setpriority might change our priority at the same moment. | |
3934 | * We don't have to worry. Conceptually one call occurs first | |
3935 | * and we have a single winner. | |
3936 | */ | |
e43379f1 MM |
3937 | if (increment < -40) |
3938 | increment = -40; | |
1da177e4 LT |
3939 | if (increment > 40) |
3940 | increment = 40; | |
3941 | ||
3942 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
3943 | if (nice < -20) | |
3944 | nice = -20; | |
3945 | if (nice > 19) | |
3946 | nice = 19; | |
3947 | ||
e43379f1 MM |
3948 | if (increment < 0 && !can_nice(current, nice)) |
3949 | return -EPERM; | |
3950 | ||
1da177e4 LT |
3951 | retval = security_task_setnice(current, nice); |
3952 | if (retval) | |
3953 | return retval; | |
3954 | ||
3955 | set_user_nice(current, nice); | |
3956 | return 0; | |
3957 | } | |
3958 | ||
3959 | #endif | |
3960 | ||
3961 | /** | |
3962 | * task_prio - return the priority value of a given task. | |
3963 | * @p: the task in question. | |
3964 | * | |
3965 | * This is the priority value as seen by users in /proc. | |
3966 | * RT tasks are offset by -200. Normal tasks are centered | |
3967 | * around 0, value goes from -16 to +15. | |
3968 | */ | |
36c8b586 | 3969 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3970 | { |
3971 | return p->prio - MAX_RT_PRIO; | |
3972 | } | |
3973 | ||
3974 | /** | |
3975 | * task_nice - return the nice value of a given task. | |
3976 | * @p: the task in question. | |
3977 | */ | |
36c8b586 | 3978 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
3979 | { |
3980 | return TASK_NICE(p); | |
3981 | } | |
1da177e4 | 3982 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
3983 | |
3984 | /** | |
3985 | * idle_cpu - is a given cpu idle currently? | |
3986 | * @cpu: the processor in question. | |
3987 | */ | |
3988 | int idle_cpu(int cpu) | |
3989 | { | |
3990 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
3991 | } | |
3992 | ||
1da177e4 LT |
3993 | /** |
3994 | * idle_task - return the idle task for a given cpu. | |
3995 | * @cpu: the processor in question. | |
3996 | */ | |
36c8b586 | 3997 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
3998 | { |
3999 | return cpu_rq(cpu)->idle; | |
4000 | } | |
4001 | ||
4002 | /** | |
4003 | * find_process_by_pid - find a process with a matching PID value. | |
4004 | * @pid: the pid in question. | |
4005 | */ | |
36c8b586 | 4006 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4007 | { |
4008 | return pid ? find_task_by_pid(pid) : current; | |
4009 | } | |
4010 | ||
4011 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4012 | static void |
4013 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4014 | { |
dd41f596 | 4015 | BUG_ON(p->se.on_rq); |
48f24c4d | 4016 | |
1da177e4 | 4017 | p->policy = policy; |
dd41f596 IM |
4018 | switch (p->policy) { |
4019 | case SCHED_NORMAL: | |
4020 | case SCHED_BATCH: | |
4021 | case SCHED_IDLE: | |
4022 | p->sched_class = &fair_sched_class; | |
4023 | break; | |
4024 | case SCHED_FIFO: | |
4025 | case SCHED_RR: | |
4026 | p->sched_class = &rt_sched_class; | |
4027 | break; | |
4028 | } | |
4029 | ||
1da177e4 | 4030 | p->rt_priority = prio; |
b29739f9 IM |
4031 | p->normal_prio = normal_prio(p); |
4032 | /* we are holding p->pi_lock already */ | |
4033 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4034 | set_load_weight(p); |
1da177e4 LT |
4035 | } |
4036 | ||
4037 | /** | |
72fd4a35 | 4038 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4039 | * @p: the task in question. |
4040 | * @policy: new policy. | |
4041 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4042 | * |
72fd4a35 | 4043 | * NOTE that the task may be already dead. |
1da177e4 | 4044 | */ |
95cdf3b7 IM |
4045 | int sched_setscheduler(struct task_struct *p, int policy, |
4046 | struct sched_param *param) | |
1da177e4 | 4047 | { |
dd41f596 | 4048 | int retval, oldprio, oldpolicy = -1, on_rq; |
1da177e4 | 4049 | unsigned long flags; |
70b97a7f | 4050 | struct rq *rq; |
1da177e4 | 4051 | |
66e5393a SR |
4052 | /* may grab non-irq protected spin_locks */ |
4053 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4054 | recheck: |
4055 | /* double check policy once rq lock held */ | |
4056 | if (policy < 0) | |
4057 | policy = oldpolicy = p->policy; | |
4058 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4059 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4060 | policy != SCHED_IDLE) | |
b0a9499c | 4061 | return -EINVAL; |
1da177e4 LT |
4062 | /* |
4063 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4064 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4065 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4066 | */ |
4067 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4068 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4069 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4070 | return -EINVAL; |
e05606d3 | 4071 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4072 | return -EINVAL; |
4073 | ||
37e4ab3f OC |
4074 | /* |
4075 | * Allow unprivileged RT tasks to decrease priority: | |
4076 | */ | |
4077 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4078 | if (rt_policy(policy)) { |
8dc3e909 | 4079 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4080 | |
4081 | if (!lock_task_sighand(p, &flags)) | |
4082 | return -ESRCH; | |
4083 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4084 | unlock_task_sighand(p, &flags); | |
4085 | ||
4086 | /* can't set/change the rt policy */ | |
4087 | if (policy != p->policy && !rlim_rtprio) | |
4088 | return -EPERM; | |
4089 | ||
4090 | /* can't increase priority */ | |
4091 | if (param->sched_priority > p->rt_priority && | |
4092 | param->sched_priority > rlim_rtprio) | |
4093 | return -EPERM; | |
4094 | } | |
dd41f596 IM |
4095 | /* |
4096 | * Like positive nice levels, dont allow tasks to | |
4097 | * move out of SCHED_IDLE either: | |
4098 | */ | |
4099 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4100 | return -EPERM; | |
5fe1d75f | 4101 | |
37e4ab3f OC |
4102 | /* can't change other user's priorities */ |
4103 | if ((current->euid != p->euid) && | |
4104 | (current->euid != p->uid)) | |
4105 | return -EPERM; | |
4106 | } | |
1da177e4 LT |
4107 | |
4108 | retval = security_task_setscheduler(p, policy, param); | |
4109 | if (retval) | |
4110 | return retval; | |
b29739f9 IM |
4111 | /* |
4112 | * make sure no PI-waiters arrive (or leave) while we are | |
4113 | * changing the priority of the task: | |
4114 | */ | |
4115 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4116 | /* |
4117 | * To be able to change p->policy safely, the apropriate | |
4118 | * runqueue lock must be held. | |
4119 | */ | |
b29739f9 | 4120 | rq = __task_rq_lock(p); |
1da177e4 LT |
4121 | /* recheck policy now with rq lock held */ |
4122 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4123 | policy = oldpolicy = -1; | |
b29739f9 IM |
4124 | __task_rq_unlock(rq); |
4125 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4126 | goto recheck; |
4127 | } | |
dd41f596 IM |
4128 | on_rq = p->se.on_rq; |
4129 | if (on_rq) | |
4130 | deactivate_task(rq, p, 0); | |
1da177e4 | 4131 | oldprio = p->prio; |
dd41f596 IM |
4132 | __setscheduler(rq, p, policy, param->sched_priority); |
4133 | if (on_rq) { | |
4134 | activate_task(rq, p, 0); | |
1da177e4 LT |
4135 | /* |
4136 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4137 | * our priority decreased, or if we are not currently running on |
4138 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4139 | */ |
d5f9f942 AM |
4140 | if (task_running(rq, p)) { |
4141 | if (p->prio > oldprio) | |
4142 | resched_task(rq->curr); | |
dd41f596 IM |
4143 | } else { |
4144 | check_preempt_curr(rq, p); | |
4145 | } | |
1da177e4 | 4146 | } |
b29739f9 IM |
4147 | __task_rq_unlock(rq); |
4148 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4149 | ||
95e02ca9 TG |
4150 | rt_mutex_adjust_pi(p); |
4151 | ||
1da177e4 LT |
4152 | return 0; |
4153 | } | |
4154 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4155 | ||
95cdf3b7 IM |
4156 | static int |
4157 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4158 | { |
1da177e4 LT |
4159 | struct sched_param lparam; |
4160 | struct task_struct *p; | |
36c8b586 | 4161 | int retval; |
1da177e4 LT |
4162 | |
4163 | if (!param || pid < 0) | |
4164 | return -EINVAL; | |
4165 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4166 | return -EFAULT; | |
5fe1d75f ON |
4167 | |
4168 | rcu_read_lock(); | |
4169 | retval = -ESRCH; | |
1da177e4 | 4170 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4171 | if (p != NULL) |
4172 | retval = sched_setscheduler(p, policy, &lparam); | |
4173 | rcu_read_unlock(); | |
36c8b586 | 4174 | |
1da177e4 LT |
4175 | return retval; |
4176 | } | |
4177 | ||
4178 | /** | |
4179 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4180 | * @pid: the pid in question. | |
4181 | * @policy: new policy. | |
4182 | * @param: structure containing the new RT priority. | |
4183 | */ | |
4184 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4185 | struct sched_param __user *param) | |
4186 | { | |
c21761f1 JB |
4187 | /* negative values for policy are not valid */ |
4188 | if (policy < 0) | |
4189 | return -EINVAL; | |
4190 | ||
1da177e4 LT |
4191 | return do_sched_setscheduler(pid, policy, param); |
4192 | } | |
4193 | ||
4194 | /** | |
4195 | * sys_sched_setparam - set/change the RT priority of a thread | |
4196 | * @pid: the pid in question. | |
4197 | * @param: structure containing the new RT priority. | |
4198 | */ | |
4199 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4200 | { | |
4201 | return do_sched_setscheduler(pid, -1, param); | |
4202 | } | |
4203 | ||
4204 | /** | |
4205 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4206 | * @pid: the pid in question. | |
4207 | */ | |
4208 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4209 | { | |
36c8b586 | 4210 | struct task_struct *p; |
1da177e4 | 4211 | int retval = -EINVAL; |
1da177e4 LT |
4212 | |
4213 | if (pid < 0) | |
4214 | goto out_nounlock; | |
4215 | ||
4216 | retval = -ESRCH; | |
4217 | read_lock(&tasklist_lock); | |
4218 | p = find_process_by_pid(pid); | |
4219 | if (p) { | |
4220 | retval = security_task_getscheduler(p); | |
4221 | if (!retval) | |
4222 | retval = p->policy; | |
4223 | } | |
4224 | read_unlock(&tasklist_lock); | |
4225 | ||
4226 | out_nounlock: | |
4227 | return retval; | |
4228 | } | |
4229 | ||
4230 | /** | |
4231 | * sys_sched_getscheduler - get the RT priority of a thread | |
4232 | * @pid: the pid in question. | |
4233 | * @param: structure containing the RT priority. | |
4234 | */ | |
4235 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4236 | { | |
4237 | struct sched_param lp; | |
36c8b586 | 4238 | struct task_struct *p; |
1da177e4 | 4239 | int retval = -EINVAL; |
1da177e4 LT |
4240 | |
4241 | if (!param || pid < 0) | |
4242 | goto out_nounlock; | |
4243 | ||
4244 | read_lock(&tasklist_lock); | |
4245 | p = find_process_by_pid(pid); | |
4246 | retval = -ESRCH; | |
4247 | if (!p) | |
4248 | goto out_unlock; | |
4249 | ||
4250 | retval = security_task_getscheduler(p); | |
4251 | if (retval) | |
4252 | goto out_unlock; | |
4253 | ||
4254 | lp.sched_priority = p->rt_priority; | |
4255 | read_unlock(&tasklist_lock); | |
4256 | ||
4257 | /* | |
4258 | * This one might sleep, we cannot do it with a spinlock held ... | |
4259 | */ | |
4260 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4261 | ||
4262 | out_nounlock: | |
4263 | return retval; | |
4264 | ||
4265 | out_unlock: | |
4266 | read_unlock(&tasklist_lock); | |
4267 | return retval; | |
4268 | } | |
4269 | ||
4270 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4271 | { | |
1da177e4 | 4272 | cpumask_t cpus_allowed; |
36c8b586 IM |
4273 | struct task_struct *p; |
4274 | int retval; | |
1da177e4 | 4275 | |
5be9361c | 4276 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4277 | read_lock(&tasklist_lock); |
4278 | ||
4279 | p = find_process_by_pid(pid); | |
4280 | if (!p) { | |
4281 | read_unlock(&tasklist_lock); | |
5be9361c | 4282 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4283 | return -ESRCH; |
4284 | } | |
4285 | ||
4286 | /* | |
4287 | * It is not safe to call set_cpus_allowed with the | |
4288 | * tasklist_lock held. We will bump the task_struct's | |
4289 | * usage count and then drop tasklist_lock. | |
4290 | */ | |
4291 | get_task_struct(p); | |
4292 | read_unlock(&tasklist_lock); | |
4293 | ||
4294 | retval = -EPERM; | |
4295 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4296 | !capable(CAP_SYS_NICE)) | |
4297 | goto out_unlock; | |
4298 | ||
e7834f8f DQ |
4299 | retval = security_task_setscheduler(p, 0, NULL); |
4300 | if (retval) | |
4301 | goto out_unlock; | |
4302 | ||
1da177e4 LT |
4303 | cpus_allowed = cpuset_cpus_allowed(p); |
4304 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4305 | retval = set_cpus_allowed(p, new_mask); | |
4306 | ||
4307 | out_unlock: | |
4308 | put_task_struct(p); | |
5be9361c | 4309 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4310 | return retval; |
4311 | } | |
4312 | ||
4313 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4314 | cpumask_t *new_mask) | |
4315 | { | |
4316 | if (len < sizeof(cpumask_t)) { | |
4317 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4318 | } else if (len > sizeof(cpumask_t)) { | |
4319 | len = sizeof(cpumask_t); | |
4320 | } | |
4321 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4322 | } | |
4323 | ||
4324 | /** | |
4325 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4326 | * @pid: pid of the process | |
4327 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4328 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4329 | */ | |
4330 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4331 | unsigned long __user *user_mask_ptr) | |
4332 | { | |
4333 | cpumask_t new_mask; | |
4334 | int retval; | |
4335 | ||
4336 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4337 | if (retval) | |
4338 | return retval; | |
4339 | ||
4340 | return sched_setaffinity(pid, new_mask); | |
4341 | } | |
4342 | ||
4343 | /* | |
4344 | * Represents all cpu's present in the system | |
4345 | * In systems capable of hotplug, this map could dynamically grow | |
4346 | * as new cpu's are detected in the system via any platform specific | |
4347 | * method, such as ACPI for e.g. | |
4348 | */ | |
4349 | ||
4cef0c61 | 4350 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4351 | EXPORT_SYMBOL(cpu_present_map); |
4352 | ||
4353 | #ifndef CONFIG_SMP | |
4cef0c61 | 4354 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4355 | EXPORT_SYMBOL(cpu_online_map); |
4356 | ||
4cef0c61 | 4357 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4358 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4359 | #endif |
4360 | ||
4361 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4362 | { | |
36c8b586 | 4363 | struct task_struct *p; |
1da177e4 | 4364 | int retval; |
1da177e4 | 4365 | |
5be9361c | 4366 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4367 | read_lock(&tasklist_lock); |
4368 | ||
4369 | retval = -ESRCH; | |
4370 | p = find_process_by_pid(pid); | |
4371 | if (!p) | |
4372 | goto out_unlock; | |
4373 | ||
e7834f8f DQ |
4374 | retval = security_task_getscheduler(p); |
4375 | if (retval) | |
4376 | goto out_unlock; | |
4377 | ||
2f7016d9 | 4378 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4379 | |
4380 | out_unlock: | |
4381 | read_unlock(&tasklist_lock); | |
5be9361c | 4382 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4383 | if (retval) |
4384 | return retval; | |
4385 | ||
4386 | return 0; | |
4387 | } | |
4388 | ||
4389 | /** | |
4390 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4391 | * @pid: pid of the process | |
4392 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4393 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4394 | */ | |
4395 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4396 | unsigned long __user *user_mask_ptr) | |
4397 | { | |
4398 | int ret; | |
4399 | cpumask_t mask; | |
4400 | ||
4401 | if (len < sizeof(cpumask_t)) | |
4402 | return -EINVAL; | |
4403 | ||
4404 | ret = sched_getaffinity(pid, &mask); | |
4405 | if (ret < 0) | |
4406 | return ret; | |
4407 | ||
4408 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4409 | return -EFAULT; | |
4410 | ||
4411 | return sizeof(cpumask_t); | |
4412 | } | |
4413 | ||
4414 | /** | |
4415 | * sys_sched_yield - yield the current processor to other threads. | |
4416 | * | |
dd41f596 IM |
4417 | * This function yields the current CPU to other tasks. If there are no |
4418 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4419 | */ |
4420 | asmlinkage long sys_sched_yield(void) | |
4421 | { | |
70b97a7f | 4422 | struct rq *rq = this_rq_lock(); |
1da177e4 LT |
4423 | |
4424 | schedstat_inc(rq, yld_cnt); | |
dd41f596 | 4425 | if (unlikely(rq->nr_running == 1)) |
1da177e4 | 4426 | schedstat_inc(rq, yld_act_empty); |
dd41f596 IM |
4427 | else |
4428 | current->sched_class->yield_task(rq, current); | |
1da177e4 LT |
4429 | |
4430 | /* | |
4431 | * Since we are going to call schedule() anyway, there's | |
4432 | * no need to preempt or enable interrupts: | |
4433 | */ | |
4434 | __release(rq->lock); | |
8a25d5de | 4435 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4436 | _raw_spin_unlock(&rq->lock); |
4437 | preempt_enable_no_resched(); | |
4438 | ||
4439 | schedule(); | |
4440 | ||
4441 | return 0; | |
4442 | } | |
4443 | ||
e7b38404 | 4444 | static void __cond_resched(void) |
1da177e4 | 4445 | { |
8e0a43d8 IM |
4446 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4447 | __might_sleep(__FILE__, __LINE__); | |
4448 | #endif | |
5bbcfd90 IM |
4449 | /* |
4450 | * The BKS might be reacquired before we have dropped | |
4451 | * PREEMPT_ACTIVE, which could trigger a second | |
4452 | * cond_resched() call. | |
4453 | */ | |
1da177e4 LT |
4454 | do { |
4455 | add_preempt_count(PREEMPT_ACTIVE); | |
4456 | schedule(); | |
4457 | sub_preempt_count(PREEMPT_ACTIVE); | |
4458 | } while (need_resched()); | |
4459 | } | |
4460 | ||
4461 | int __sched cond_resched(void) | |
4462 | { | |
9414232f IM |
4463 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4464 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4465 | __cond_resched(); |
4466 | return 1; | |
4467 | } | |
4468 | return 0; | |
4469 | } | |
1da177e4 LT |
4470 | EXPORT_SYMBOL(cond_resched); |
4471 | ||
4472 | /* | |
4473 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4474 | * call schedule, and on return reacquire the lock. | |
4475 | * | |
4476 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4477 | * operations here to prevent schedule() from being called twice (once via | |
4478 | * spin_unlock(), once by hand). | |
4479 | */ | |
95cdf3b7 | 4480 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4481 | { |
6df3cecb JK |
4482 | int ret = 0; |
4483 | ||
1da177e4 LT |
4484 | if (need_lockbreak(lock)) { |
4485 | spin_unlock(lock); | |
4486 | cpu_relax(); | |
6df3cecb | 4487 | ret = 1; |
1da177e4 LT |
4488 | spin_lock(lock); |
4489 | } | |
9414232f | 4490 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4491 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4492 | _raw_spin_unlock(lock); |
4493 | preempt_enable_no_resched(); | |
4494 | __cond_resched(); | |
6df3cecb | 4495 | ret = 1; |
1da177e4 | 4496 | spin_lock(lock); |
1da177e4 | 4497 | } |
6df3cecb | 4498 | return ret; |
1da177e4 | 4499 | } |
1da177e4 LT |
4500 | EXPORT_SYMBOL(cond_resched_lock); |
4501 | ||
4502 | int __sched cond_resched_softirq(void) | |
4503 | { | |
4504 | BUG_ON(!in_softirq()); | |
4505 | ||
9414232f | 4506 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4507 | local_bh_enable(); |
1da177e4 LT |
4508 | __cond_resched(); |
4509 | local_bh_disable(); | |
4510 | return 1; | |
4511 | } | |
4512 | return 0; | |
4513 | } | |
1da177e4 LT |
4514 | EXPORT_SYMBOL(cond_resched_softirq); |
4515 | ||
1da177e4 LT |
4516 | /** |
4517 | * yield - yield the current processor to other threads. | |
4518 | * | |
72fd4a35 | 4519 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4520 | * thread runnable and calls sys_sched_yield(). |
4521 | */ | |
4522 | void __sched yield(void) | |
4523 | { | |
4524 | set_current_state(TASK_RUNNING); | |
4525 | sys_sched_yield(); | |
4526 | } | |
1da177e4 LT |
4527 | EXPORT_SYMBOL(yield); |
4528 | ||
4529 | /* | |
4530 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4531 | * that process accounting knows that this is a task in IO wait state. | |
4532 | * | |
4533 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4534 | * has set its backing_dev_info: the queue against which it should throttle) | |
4535 | */ | |
4536 | void __sched io_schedule(void) | |
4537 | { | |
70b97a7f | 4538 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4539 | |
0ff92245 | 4540 | delayacct_blkio_start(); |
1da177e4 LT |
4541 | atomic_inc(&rq->nr_iowait); |
4542 | schedule(); | |
4543 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4544 | delayacct_blkio_end(); |
1da177e4 | 4545 | } |
1da177e4 LT |
4546 | EXPORT_SYMBOL(io_schedule); |
4547 | ||
4548 | long __sched io_schedule_timeout(long timeout) | |
4549 | { | |
70b97a7f | 4550 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4551 | long ret; |
4552 | ||
0ff92245 | 4553 | delayacct_blkio_start(); |
1da177e4 LT |
4554 | atomic_inc(&rq->nr_iowait); |
4555 | ret = schedule_timeout(timeout); | |
4556 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4557 | delayacct_blkio_end(); |
1da177e4 LT |
4558 | return ret; |
4559 | } | |
4560 | ||
4561 | /** | |
4562 | * sys_sched_get_priority_max - return maximum RT priority. | |
4563 | * @policy: scheduling class. | |
4564 | * | |
4565 | * this syscall returns the maximum rt_priority that can be used | |
4566 | * by a given scheduling class. | |
4567 | */ | |
4568 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4569 | { | |
4570 | int ret = -EINVAL; | |
4571 | ||
4572 | switch (policy) { | |
4573 | case SCHED_FIFO: | |
4574 | case SCHED_RR: | |
4575 | ret = MAX_USER_RT_PRIO-1; | |
4576 | break; | |
4577 | case SCHED_NORMAL: | |
b0a9499c | 4578 | case SCHED_BATCH: |
dd41f596 | 4579 | case SCHED_IDLE: |
1da177e4 LT |
4580 | ret = 0; |
4581 | break; | |
4582 | } | |
4583 | return ret; | |
4584 | } | |
4585 | ||
4586 | /** | |
4587 | * sys_sched_get_priority_min - return minimum RT priority. | |
4588 | * @policy: scheduling class. | |
4589 | * | |
4590 | * this syscall returns the minimum rt_priority that can be used | |
4591 | * by a given scheduling class. | |
4592 | */ | |
4593 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4594 | { | |
4595 | int ret = -EINVAL; | |
4596 | ||
4597 | switch (policy) { | |
4598 | case SCHED_FIFO: | |
4599 | case SCHED_RR: | |
4600 | ret = 1; | |
4601 | break; | |
4602 | case SCHED_NORMAL: | |
b0a9499c | 4603 | case SCHED_BATCH: |
dd41f596 | 4604 | case SCHED_IDLE: |
1da177e4 LT |
4605 | ret = 0; |
4606 | } | |
4607 | return ret; | |
4608 | } | |
4609 | ||
4610 | /** | |
4611 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4612 | * @pid: pid of the process. | |
4613 | * @interval: userspace pointer to the timeslice value. | |
4614 | * | |
4615 | * this syscall writes the default timeslice value of a given process | |
4616 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4617 | */ | |
4618 | asmlinkage | |
4619 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4620 | { | |
36c8b586 | 4621 | struct task_struct *p; |
1da177e4 LT |
4622 | int retval = -EINVAL; |
4623 | struct timespec t; | |
1da177e4 LT |
4624 | |
4625 | if (pid < 0) | |
4626 | goto out_nounlock; | |
4627 | ||
4628 | retval = -ESRCH; | |
4629 | read_lock(&tasklist_lock); | |
4630 | p = find_process_by_pid(pid); | |
4631 | if (!p) | |
4632 | goto out_unlock; | |
4633 | ||
4634 | retval = security_task_getscheduler(p); | |
4635 | if (retval) | |
4636 | goto out_unlock; | |
4637 | ||
b78709cf | 4638 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
dd41f596 | 4639 | 0 : static_prio_timeslice(p->static_prio), &t); |
1da177e4 LT |
4640 | read_unlock(&tasklist_lock); |
4641 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4642 | out_nounlock: | |
4643 | return retval; | |
4644 | out_unlock: | |
4645 | read_unlock(&tasklist_lock); | |
4646 | return retval; | |
4647 | } | |
4648 | ||
2ed6e34f | 4649 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4650 | |
4651 | static void show_task(struct task_struct *p) | |
1da177e4 | 4652 | { |
1da177e4 | 4653 | unsigned long free = 0; |
36c8b586 | 4654 | unsigned state; |
1da177e4 | 4655 | |
1da177e4 | 4656 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4657 | printk("%-13.13s %c", p->comm, |
4658 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4659 | #if BITS_PER_LONG == 32 |
1da177e4 | 4660 | if (state == TASK_RUNNING) |
4bd77321 | 4661 | printk(" running "); |
1da177e4 | 4662 | else |
4bd77321 | 4663 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4664 | #else |
4665 | if (state == TASK_RUNNING) | |
4bd77321 | 4666 | printk(" running task "); |
1da177e4 LT |
4667 | else |
4668 | printk(" %016lx ", thread_saved_pc(p)); | |
4669 | #endif | |
4670 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4671 | { | |
10ebffde | 4672 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4673 | while (!*n) |
4674 | n++; | |
10ebffde | 4675 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4676 | } |
4677 | #endif | |
4bd77321 | 4678 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4679 | |
4680 | if (state != TASK_RUNNING) | |
4681 | show_stack(p, NULL); | |
4682 | } | |
4683 | ||
e59e2ae2 | 4684 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4685 | { |
36c8b586 | 4686 | struct task_struct *g, *p; |
1da177e4 | 4687 | |
4bd77321 IM |
4688 | #if BITS_PER_LONG == 32 |
4689 | printk(KERN_INFO | |
4690 | " task PC stack pid father\n"); | |
1da177e4 | 4691 | #else |
4bd77321 IM |
4692 | printk(KERN_INFO |
4693 | " task PC stack pid father\n"); | |
1da177e4 LT |
4694 | #endif |
4695 | read_lock(&tasklist_lock); | |
4696 | do_each_thread(g, p) { | |
4697 | /* | |
4698 | * reset the NMI-timeout, listing all files on a slow | |
4699 | * console might take alot of time: | |
4700 | */ | |
4701 | touch_nmi_watchdog(); | |
39bc89fd | 4702 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4703 | show_task(p); |
1da177e4 LT |
4704 | } while_each_thread(g, p); |
4705 | ||
04c9167f JF |
4706 | touch_all_softlockup_watchdogs(); |
4707 | ||
dd41f596 IM |
4708 | #ifdef CONFIG_SCHED_DEBUG |
4709 | sysrq_sched_debug_show(); | |
4710 | #endif | |
1da177e4 | 4711 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4712 | /* |
4713 | * Only show locks if all tasks are dumped: | |
4714 | */ | |
4715 | if (state_filter == -1) | |
4716 | debug_show_all_locks(); | |
1da177e4 LT |
4717 | } |
4718 | ||
1df21055 IM |
4719 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4720 | { | |
dd41f596 | 4721 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4722 | } |
4723 | ||
f340c0d1 IM |
4724 | /** |
4725 | * init_idle - set up an idle thread for a given CPU | |
4726 | * @idle: task in question | |
4727 | * @cpu: cpu the idle task belongs to | |
4728 | * | |
4729 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4730 | * flag, to make booting more robust. | |
4731 | */ | |
5c1e1767 | 4732 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4733 | { |
70b97a7f | 4734 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4735 | unsigned long flags; |
4736 | ||
dd41f596 IM |
4737 | __sched_fork(idle); |
4738 | idle->se.exec_start = sched_clock(); | |
4739 | ||
b29739f9 | 4740 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4741 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4742 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4743 | |
4744 | spin_lock_irqsave(&rq->lock, flags); | |
4745 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4746 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4747 | idle->oncpu = 1; | |
4748 | #endif | |
1da177e4 LT |
4749 | spin_unlock_irqrestore(&rq->lock, flags); |
4750 | ||
4751 | /* Set the preempt count _outside_ the spinlocks! */ | |
4752 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4753 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4754 | #else |
a1261f54 | 4755 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4756 | #endif |
dd41f596 IM |
4757 | /* |
4758 | * The idle tasks have their own, simple scheduling class: | |
4759 | */ | |
4760 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4761 | } |
4762 | ||
4763 | /* | |
4764 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4765 | * indicates which cpus entered this state. This is used | |
4766 | * in the rcu update to wait only for active cpus. For system | |
4767 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4768 | * always be CPU_MASK_NONE. | |
4769 | */ | |
4770 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4771 | ||
dd41f596 IM |
4772 | /* |
4773 | * Increase the granularity value when there are more CPUs, | |
4774 | * because with more CPUs the 'effective latency' as visible | |
4775 | * to users decreases. But the relationship is not linear, | |
4776 | * so pick a second-best guess by going with the log2 of the | |
4777 | * number of CPUs. | |
4778 | * | |
4779 | * This idea comes from the SD scheduler of Con Kolivas: | |
4780 | */ | |
4781 | static inline void sched_init_granularity(void) | |
4782 | { | |
4783 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
a5968df8 | 4784 | const unsigned long gran_limit = 100000000; |
dd41f596 IM |
4785 | |
4786 | sysctl_sched_granularity *= factor; | |
4787 | if (sysctl_sched_granularity > gran_limit) | |
4788 | sysctl_sched_granularity = gran_limit; | |
4789 | ||
4790 | sysctl_sched_runtime_limit = sysctl_sched_granularity * 4; | |
4791 | sysctl_sched_wakeup_granularity = sysctl_sched_granularity / 2; | |
4792 | } | |
4793 | ||
1da177e4 LT |
4794 | #ifdef CONFIG_SMP |
4795 | /* | |
4796 | * This is how migration works: | |
4797 | * | |
70b97a7f | 4798 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4799 | * runqueue and wake up that CPU's migration thread. |
4800 | * 2) we down() the locked semaphore => thread blocks. | |
4801 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4802 | * thread off the CPU) | |
4803 | * 4) it gets the migration request and checks whether the migrated | |
4804 | * task is still in the wrong runqueue. | |
4805 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4806 | * it and puts it into the right queue. | |
4807 | * 6) migration thread up()s the semaphore. | |
4808 | * 7) we wake up and the migration is done. | |
4809 | */ | |
4810 | ||
4811 | /* | |
4812 | * Change a given task's CPU affinity. Migrate the thread to a | |
4813 | * proper CPU and schedule it away if the CPU it's executing on | |
4814 | * is removed from the allowed bitmask. | |
4815 | * | |
4816 | * NOTE: the caller must have a valid reference to the task, the | |
4817 | * task must not exit() & deallocate itself prematurely. The | |
4818 | * call is not atomic; no spinlocks may be held. | |
4819 | */ | |
36c8b586 | 4820 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4821 | { |
70b97a7f | 4822 | struct migration_req req; |
1da177e4 | 4823 | unsigned long flags; |
70b97a7f | 4824 | struct rq *rq; |
48f24c4d | 4825 | int ret = 0; |
1da177e4 LT |
4826 | |
4827 | rq = task_rq_lock(p, &flags); | |
4828 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4829 | ret = -EINVAL; | |
4830 | goto out; | |
4831 | } | |
4832 | ||
4833 | p->cpus_allowed = new_mask; | |
4834 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4835 | if (cpu_isset(task_cpu(p), new_mask)) | |
4836 | goto out; | |
4837 | ||
4838 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4839 | /* Need help from migration thread: drop lock and wait. */ | |
4840 | task_rq_unlock(rq, &flags); | |
4841 | wake_up_process(rq->migration_thread); | |
4842 | wait_for_completion(&req.done); | |
4843 | tlb_migrate_finish(p->mm); | |
4844 | return 0; | |
4845 | } | |
4846 | out: | |
4847 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4848 | |
1da177e4 LT |
4849 | return ret; |
4850 | } | |
1da177e4 LT |
4851 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4852 | ||
4853 | /* | |
4854 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4855 | * this because either it can't run here any more (set_cpus_allowed() | |
4856 | * away from this CPU, or CPU going down), or because we're | |
4857 | * attempting to rebalance this task on exec (sched_exec). | |
4858 | * | |
4859 | * So we race with normal scheduler movements, but that's OK, as long | |
4860 | * as the task is no longer on this CPU. | |
efc30814 KK |
4861 | * |
4862 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4863 | */ |
efc30814 | 4864 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4865 | { |
70b97a7f | 4866 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4867 | int ret = 0, on_rq; |
1da177e4 LT |
4868 | |
4869 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4870 | return ret; |
1da177e4 LT |
4871 | |
4872 | rq_src = cpu_rq(src_cpu); | |
4873 | rq_dest = cpu_rq(dest_cpu); | |
4874 | ||
4875 | double_rq_lock(rq_src, rq_dest); | |
4876 | /* Already moved. */ | |
4877 | if (task_cpu(p) != src_cpu) | |
4878 | goto out; | |
4879 | /* Affinity changed (again). */ | |
4880 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4881 | goto out; | |
4882 | ||
dd41f596 IM |
4883 | on_rq = p->se.on_rq; |
4884 | if (on_rq) | |
4885 | deactivate_task(rq_src, p, 0); | |
1da177e4 | 4886 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4887 | if (on_rq) { |
4888 | activate_task(rq_dest, p, 0); | |
4889 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4890 | } |
efc30814 | 4891 | ret = 1; |
1da177e4 LT |
4892 | out: |
4893 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4894 | return ret; |
1da177e4 LT |
4895 | } |
4896 | ||
4897 | /* | |
4898 | * migration_thread - this is a highprio system thread that performs | |
4899 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4900 | * another runqueue. | |
4901 | */ | |
95cdf3b7 | 4902 | static int migration_thread(void *data) |
1da177e4 | 4903 | { |
1da177e4 | 4904 | int cpu = (long)data; |
70b97a7f | 4905 | struct rq *rq; |
1da177e4 LT |
4906 | |
4907 | rq = cpu_rq(cpu); | |
4908 | BUG_ON(rq->migration_thread != current); | |
4909 | ||
4910 | set_current_state(TASK_INTERRUPTIBLE); | |
4911 | while (!kthread_should_stop()) { | |
70b97a7f | 4912 | struct migration_req *req; |
1da177e4 | 4913 | struct list_head *head; |
1da177e4 | 4914 | |
1da177e4 LT |
4915 | spin_lock_irq(&rq->lock); |
4916 | ||
4917 | if (cpu_is_offline(cpu)) { | |
4918 | spin_unlock_irq(&rq->lock); | |
4919 | goto wait_to_die; | |
4920 | } | |
4921 | ||
4922 | if (rq->active_balance) { | |
4923 | active_load_balance(rq, cpu); | |
4924 | rq->active_balance = 0; | |
4925 | } | |
4926 | ||
4927 | head = &rq->migration_queue; | |
4928 | ||
4929 | if (list_empty(head)) { | |
4930 | spin_unlock_irq(&rq->lock); | |
4931 | schedule(); | |
4932 | set_current_state(TASK_INTERRUPTIBLE); | |
4933 | continue; | |
4934 | } | |
70b97a7f | 4935 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
4936 | list_del_init(head->next); |
4937 | ||
674311d5 NP |
4938 | spin_unlock(&rq->lock); |
4939 | __migrate_task(req->task, cpu, req->dest_cpu); | |
4940 | local_irq_enable(); | |
1da177e4 LT |
4941 | |
4942 | complete(&req->done); | |
4943 | } | |
4944 | __set_current_state(TASK_RUNNING); | |
4945 | return 0; | |
4946 | ||
4947 | wait_to_die: | |
4948 | /* Wait for kthread_stop */ | |
4949 | set_current_state(TASK_INTERRUPTIBLE); | |
4950 | while (!kthread_should_stop()) { | |
4951 | schedule(); | |
4952 | set_current_state(TASK_INTERRUPTIBLE); | |
4953 | } | |
4954 | __set_current_state(TASK_RUNNING); | |
4955 | return 0; | |
4956 | } | |
4957 | ||
4958 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
4959 | /* |
4960 | * Figure out where task on dead CPU should go, use force if neccessary. | |
4961 | * NOTE: interrupts should be disabled by the caller | |
4962 | */ | |
48f24c4d | 4963 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 4964 | { |
efc30814 | 4965 | unsigned long flags; |
1da177e4 | 4966 | cpumask_t mask; |
70b97a7f IM |
4967 | struct rq *rq; |
4968 | int dest_cpu; | |
1da177e4 | 4969 | |
efc30814 | 4970 | restart: |
1da177e4 LT |
4971 | /* On same node? */ |
4972 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 4973 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
4974 | dest_cpu = any_online_cpu(mask); |
4975 | ||
4976 | /* On any allowed CPU? */ | |
4977 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 4978 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
4979 | |
4980 | /* No more Mr. Nice Guy. */ | |
4981 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
4982 | rq = task_rq_lock(p, &flags); |
4983 | cpus_setall(p->cpus_allowed); | |
4984 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 4985 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4986 | |
4987 | /* | |
4988 | * Don't tell them about moving exiting tasks or | |
4989 | * kernel threads (both mm NULL), since they never | |
4990 | * leave kernel. | |
4991 | */ | |
48f24c4d | 4992 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
4993 | printk(KERN_INFO "process %d (%s) no " |
4994 | "longer affine to cpu%d\n", | |
48f24c4d | 4995 | p->pid, p->comm, dead_cpu); |
1da177e4 | 4996 | } |
48f24c4d | 4997 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 4998 | goto restart; |
1da177e4 LT |
4999 | } |
5000 | ||
5001 | /* | |
5002 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5003 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5004 | * for performance reasons the counter is not stricly tracking tasks to | |
5005 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5006 | * to keep the global sum constant after CPU-down: | |
5007 | */ | |
70b97a7f | 5008 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5009 | { |
70b97a7f | 5010 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5011 | unsigned long flags; |
5012 | ||
5013 | local_irq_save(flags); | |
5014 | double_rq_lock(rq_src, rq_dest); | |
5015 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5016 | rq_src->nr_uninterruptible = 0; | |
5017 | double_rq_unlock(rq_src, rq_dest); | |
5018 | local_irq_restore(flags); | |
5019 | } | |
5020 | ||
5021 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5022 | static void migrate_live_tasks(int src_cpu) | |
5023 | { | |
48f24c4d | 5024 | struct task_struct *p, *t; |
1da177e4 LT |
5025 | |
5026 | write_lock_irq(&tasklist_lock); | |
5027 | ||
48f24c4d IM |
5028 | do_each_thread(t, p) { |
5029 | if (p == current) | |
1da177e4 LT |
5030 | continue; |
5031 | ||
48f24c4d IM |
5032 | if (task_cpu(p) == src_cpu) |
5033 | move_task_off_dead_cpu(src_cpu, p); | |
5034 | } while_each_thread(t, p); | |
1da177e4 LT |
5035 | |
5036 | write_unlock_irq(&tasklist_lock); | |
5037 | } | |
5038 | ||
dd41f596 IM |
5039 | /* |
5040 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5041 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5042 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5043 | */ |
5044 | void sched_idle_next(void) | |
5045 | { | |
48f24c4d | 5046 | int this_cpu = smp_processor_id(); |
70b97a7f | 5047 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5048 | struct task_struct *p = rq->idle; |
5049 | unsigned long flags; | |
5050 | ||
5051 | /* cpu has to be offline */ | |
48f24c4d | 5052 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5053 | |
48f24c4d IM |
5054 | /* |
5055 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5056 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5057 | */ |
5058 | spin_lock_irqsave(&rq->lock, flags); | |
5059 | ||
dd41f596 | 5060 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5061 | |
5062 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5063 | activate_idle_task(p, rq); |
1da177e4 LT |
5064 | |
5065 | spin_unlock_irqrestore(&rq->lock, flags); | |
5066 | } | |
5067 | ||
48f24c4d IM |
5068 | /* |
5069 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5070 | * offline. |
5071 | */ | |
5072 | void idle_task_exit(void) | |
5073 | { | |
5074 | struct mm_struct *mm = current->active_mm; | |
5075 | ||
5076 | BUG_ON(cpu_online(smp_processor_id())); | |
5077 | ||
5078 | if (mm != &init_mm) | |
5079 | switch_mm(mm, &init_mm, current); | |
5080 | mmdrop(mm); | |
5081 | } | |
5082 | ||
054b9108 | 5083 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5084 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5085 | { |
70b97a7f | 5086 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5087 | |
5088 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5089 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5090 | |
5091 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5092 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5093 | |
48f24c4d | 5094 | get_task_struct(p); |
1da177e4 LT |
5095 | |
5096 | /* | |
5097 | * Drop lock around migration; if someone else moves it, | |
5098 | * that's OK. No task can be added to this CPU, so iteration is | |
5099 | * fine. | |
054b9108 | 5100 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5101 | */ |
054b9108 | 5102 | spin_unlock(&rq->lock); |
48f24c4d | 5103 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5104 | spin_lock(&rq->lock); |
1da177e4 | 5105 | |
48f24c4d | 5106 | put_task_struct(p); |
1da177e4 LT |
5107 | } |
5108 | ||
5109 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5110 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5111 | { | |
70b97a7f | 5112 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5113 | struct task_struct *next; |
48f24c4d | 5114 | |
dd41f596 IM |
5115 | for ( ; ; ) { |
5116 | if (!rq->nr_running) | |
5117 | break; | |
5118 | next = pick_next_task(rq, rq->curr, rq_clock(rq)); | |
5119 | if (!next) | |
5120 | break; | |
5121 | migrate_dead(dead_cpu, next); | |
1da177e4 LT |
5122 | } |
5123 | } | |
5124 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5125 | ||
5126 | /* | |
5127 | * migration_call - callback that gets triggered when a CPU is added. | |
5128 | * Here we can start up the necessary migration thread for the new CPU. | |
5129 | */ | |
48f24c4d IM |
5130 | static int __cpuinit |
5131 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5132 | { |
1da177e4 | 5133 | struct task_struct *p; |
48f24c4d | 5134 | int cpu = (long)hcpu; |
1da177e4 | 5135 | unsigned long flags; |
70b97a7f | 5136 | struct rq *rq; |
1da177e4 LT |
5137 | |
5138 | switch (action) { | |
5be9361c GS |
5139 | case CPU_LOCK_ACQUIRE: |
5140 | mutex_lock(&sched_hotcpu_mutex); | |
5141 | break; | |
5142 | ||
1da177e4 | 5143 | case CPU_UP_PREPARE: |
8bb78442 | 5144 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5145 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5146 | if (IS_ERR(p)) |
5147 | return NOTIFY_BAD; | |
1da177e4 LT |
5148 | kthread_bind(p, cpu); |
5149 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5150 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5151 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5152 | task_rq_unlock(rq, &flags); |
5153 | cpu_rq(cpu)->migration_thread = p; | |
5154 | break; | |
48f24c4d | 5155 | |
1da177e4 | 5156 | case CPU_ONLINE: |
8bb78442 | 5157 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5158 | /* Strictly unneccessary, as first user will wake it. */ |
5159 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5160 | break; | |
48f24c4d | 5161 | |
1da177e4 LT |
5162 | #ifdef CONFIG_HOTPLUG_CPU |
5163 | case CPU_UP_CANCELED: | |
8bb78442 | 5164 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5165 | if (!cpu_rq(cpu)->migration_thread) |
5166 | break; | |
1da177e4 | 5167 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5168 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5169 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5170 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5171 | cpu_rq(cpu)->migration_thread = NULL; | |
5172 | break; | |
48f24c4d | 5173 | |
1da177e4 | 5174 | case CPU_DEAD: |
8bb78442 | 5175 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5176 | migrate_live_tasks(cpu); |
5177 | rq = cpu_rq(cpu); | |
5178 | kthread_stop(rq->migration_thread); | |
5179 | rq->migration_thread = NULL; | |
5180 | /* Idle task back to normal (off runqueue, low prio) */ | |
5181 | rq = task_rq_lock(rq->idle, &flags); | |
dd41f596 | 5182 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5183 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5184 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5185 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5186 | migrate_dead_tasks(cpu); |
5187 | task_rq_unlock(rq, &flags); | |
5188 | migrate_nr_uninterruptible(rq); | |
5189 | BUG_ON(rq->nr_running != 0); | |
5190 | ||
5191 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5192 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5193 | * the requestors. */ |
5194 | spin_lock_irq(&rq->lock); | |
5195 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5196 | struct migration_req *req; |
5197 | ||
1da177e4 | 5198 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5199 | struct migration_req, list); |
1da177e4 LT |
5200 | list_del_init(&req->list); |
5201 | complete(&req->done); | |
5202 | } | |
5203 | spin_unlock_irq(&rq->lock); | |
5204 | break; | |
5205 | #endif | |
5be9361c GS |
5206 | case CPU_LOCK_RELEASE: |
5207 | mutex_unlock(&sched_hotcpu_mutex); | |
5208 | break; | |
1da177e4 LT |
5209 | } |
5210 | return NOTIFY_OK; | |
5211 | } | |
5212 | ||
5213 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5214 | * happens before everything else. | |
5215 | */ | |
26c2143b | 5216 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5217 | .notifier_call = migration_call, |
5218 | .priority = 10 | |
5219 | }; | |
5220 | ||
5221 | int __init migration_init(void) | |
5222 | { | |
5223 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5224 | int err; |
48f24c4d IM |
5225 | |
5226 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5227 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5228 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5229 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5230 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5231 | |
1da177e4 LT |
5232 | return 0; |
5233 | } | |
5234 | #endif | |
5235 | ||
5236 | #ifdef CONFIG_SMP | |
476f3534 CL |
5237 | |
5238 | /* Number of possible processor ids */ | |
5239 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5240 | EXPORT_SYMBOL(nr_cpu_ids); | |
5241 | ||
1a20ff27 | 5242 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5243 | #ifdef SCHED_DOMAIN_DEBUG |
5244 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5245 | { | |
5246 | int level = 0; | |
5247 | ||
41c7ce9a NP |
5248 | if (!sd) { |
5249 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5250 | return; | |
5251 | } | |
5252 | ||
1da177e4 LT |
5253 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5254 | ||
5255 | do { | |
5256 | int i; | |
5257 | char str[NR_CPUS]; | |
5258 | struct sched_group *group = sd->groups; | |
5259 | cpumask_t groupmask; | |
5260 | ||
5261 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5262 | cpus_clear(groupmask); | |
5263 | ||
5264 | printk(KERN_DEBUG); | |
5265 | for (i = 0; i < level + 1; i++) | |
5266 | printk(" "); | |
5267 | printk("domain %d: ", level); | |
5268 | ||
5269 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5270 | printk("does not load-balance\n"); | |
5271 | if (sd->parent) | |
33859f7f MOS |
5272 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5273 | " has parent"); | |
1da177e4 LT |
5274 | break; |
5275 | } | |
5276 | ||
5277 | printk("span %s\n", str); | |
5278 | ||
5279 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5280 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5281 | "CPU%d\n", cpu); | |
1da177e4 | 5282 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5283 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5284 | " CPU%d\n", cpu); | |
1da177e4 LT |
5285 | |
5286 | printk(KERN_DEBUG); | |
5287 | for (i = 0; i < level + 2; i++) | |
5288 | printk(" "); | |
5289 | printk("groups:"); | |
5290 | do { | |
5291 | if (!group) { | |
5292 | printk("\n"); | |
5293 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5294 | break; | |
5295 | } | |
5296 | ||
5517d86b | 5297 | if (!group->__cpu_power) { |
1da177e4 | 5298 | printk("\n"); |
33859f7f MOS |
5299 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5300 | "set\n"); | |
1da177e4 LT |
5301 | } |
5302 | ||
5303 | if (!cpus_weight(group->cpumask)) { | |
5304 | printk("\n"); | |
5305 | printk(KERN_ERR "ERROR: empty group\n"); | |
5306 | } | |
5307 | ||
5308 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5309 | printk("\n"); | |
5310 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5311 | } | |
5312 | ||
5313 | cpus_or(groupmask, groupmask, group->cpumask); | |
5314 | ||
5315 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5316 | printk(" %s", str); | |
5317 | ||
5318 | group = group->next; | |
5319 | } while (group != sd->groups); | |
5320 | printk("\n"); | |
5321 | ||
5322 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5323 | printk(KERN_ERR "ERROR: groups don't span " |
5324 | "domain->span\n"); | |
1da177e4 LT |
5325 | |
5326 | level++; | |
5327 | sd = sd->parent; | |
33859f7f MOS |
5328 | if (!sd) |
5329 | continue; | |
1da177e4 | 5330 | |
33859f7f MOS |
5331 | if (!cpus_subset(groupmask, sd->span)) |
5332 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5333 | "of domain->span\n"); | |
1da177e4 LT |
5334 | |
5335 | } while (sd); | |
5336 | } | |
5337 | #else | |
48f24c4d | 5338 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5339 | #endif |
5340 | ||
1a20ff27 | 5341 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5342 | { |
5343 | if (cpus_weight(sd->span) == 1) | |
5344 | return 1; | |
5345 | ||
5346 | /* Following flags need at least 2 groups */ | |
5347 | if (sd->flags & (SD_LOAD_BALANCE | | |
5348 | SD_BALANCE_NEWIDLE | | |
5349 | SD_BALANCE_FORK | | |
89c4710e SS |
5350 | SD_BALANCE_EXEC | |
5351 | SD_SHARE_CPUPOWER | | |
5352 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5353 | if (sd->groups != sd->groups->next) |
5354 | return 0; | |
5355 | } | |
5356 | ||
5357 | /* Following flags don't use groups */ | |
5358 | if (sd->flags & (SD_WAKE_IDLE | | |
5359 | SD_WAKE_AFFINE | | |
5360 | SD_WAKE_BALANCE)) | |
5361 | return 0; | |
5362 | ||
5363 | return 1; | |
5364 | } | |
5365 | ||
48f24c4d IM |
5366 | static int |
5367 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5368 | { |
5369 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5370 | ||
5371 | if (sd_degenerate(parent)) | |
5372 | return 1; | |
5373 | ||
5374 | if (!cpus_equal(sd->span, parent->span)) | |
5375 | return 0; | |
5376 | ||
5377 | /* Does parent contain flags not in child? */ | |
5378 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5379 | if (cflags & SD_WAKE_AFFINE) | |
5380 | pflags &= ~SD_WAKE_BALANCE; | |
5381 | /* Flags needing groups don't count if only 1 group in parent */ | |
5382 | if (parent->groups == parent->groups->next) { | |
5383 | pflags &= ~(SD_LOAD_BALANCE | | |
5384 | SD_BALANCE_NEWIDLE | | |
5385 | SD_BALANCE_FORK | | |
89c4710e SS |
5386 | SD_BALANCE_EXEC | |
5387 | SD_SHARE_CPUPOWER | | |
5388 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5389 | } |
5390 | if (~cflags & pflags) | |
5391 | return 0; | |
5392 | ||
5393 | return 1; | |
5394 | } | |
5395 | ||
1da177e4 LT |
5396 | /* |
5397 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5398 | * hold the hotplug lock. | |
5399 | */ | |
9c1cfda2 | 5400 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5401 | { |
70b97a7f | 5402 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5403 | struct sched_domain *tmp; |
5404 | ||
5405 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5406 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5407 | struct sched_domain *parent = tmp->parent; | |
5408 | if (!parent) | |
5409 | break; | |
1a848870 | 5410 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5411 | tmp->parent = parent->parent; |
1a848870 SS |
5412 | if (parent->parent) |
5413 | parent->parent->child = tmp; | |
5414 | } | |
245af2c7 SS |
5415 | } |
5416 | ||
1a848870 | 5417 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5418 | sd = sd->parent; |
1a848870 SS |
5419 | if (sd) |
5420 | sd->child = NULL; | |
5421 | } | |
1da177e4 LT |
5422 | |
5423 | sched_domain_debug(sd, cpu); | |
5424 | ||
674311d5 | 5425 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5426 | } |
5427 | ||
5428 | /* cpus with isolated domains */ | |
67af63a6 | 5429 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5430 | |
5431 | /* Setup the mask of cpus configured for isolated domains */ | |
5432 | static int __init isolated_cpu_setup(char *str) | |
5433 | { | |
5434 | int ints[NR_CPUS], i; | |
5435 | ||
5436 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5437 | cpus_clear(cpu_isolated_map); | |
5438 | for (i = 1; i <= ints[0]; i++) | |
5439 | if (ints[i] < NR_CPUS) | |
5440 | cpu_set(ints[i], cpu_isolated_map); | |
5441 | return 1; | |
5442 | } | |
5443 | ||
5444 | __setup ("isolcpus=", isolated_cpu_setup); | |
5445 | ||
5446 | /* | |
6711cab4 SS |
5447 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5448 | * to a function which identifies what group(along with sched group) a CPU | |
5449 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5450 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5451 | * |
5452 | * init_sched_build_groups will build a circular linked list of the groups | |
5453 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5454 | * and ->cpu_power to 0. | |
5455 | */ | |
a616058b | 5456 | static void |
6711cab4 SS |
5457 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5458 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5459 | struct sched_group **sg)) | |
1da177e4 LT |
5460 | { |
5461 | struct sched_group *first = NULL, *last = NULL; | |
5462 | cpumask_t covered = CPU_MASK_NONE; | |
5463 | int i; | |
5464 | ||
5465 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5466 | struct sched_group *sg; |
5467 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5468 | int j; |
5469 | ||
5470 | if (cpu_isset(i, covered)) | |
5471 | continue; | |
5472 | ||
5473 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5474 | sg->__cpu_power = 0; |
1da177e4 LT |
5475 | |
5476 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5477 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5478 | continue; |
5479 | ||
5480 | cpu_set(j, covered); | |
5481 | cpu_set(j, sg->cpumask); | |
5482 | } | |
5483 | if (!first) | |
5484 | first = sg; | |
5485 | if (last) | |
5486 | last->next = sg; | |
5487 | last = sg; | |
5488 | } | |
5489 | last->next = first; | |
5490 | } | |
5491 | ||
9c1cfda2 | 5492 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5493 | |
9c1cfda2 | 5494 | #ifdef CONFIG_NUMA |
198e2f18 | 5495 | |
9c1cfda2 JH |
5496 | /** |
5497 | * find_next_best_node - find the next node to include in a sched_domain | |
5498 | * @node: node whose sched_domain we're building | |
5499 | * @used_nodes: nodes already in the sched_domain | |
5500 | * | |
5501 | * Find the next node to include in a given scheduling domain. Simply | |
5502 | * finds the closest node not already in the @used_nodes map. | |
5503 | * | |
5504 | * Should use nodemask_t. | |
5505 | */ | |
5506 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5507 | { | |
5508 | int i, n, val, min_val, best_node = 0; | |
5509 | ||
5510 | min_val = INT_MAX; | |
5511 | ||
5512 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5513 | /* Start at @node */ | |
5514 | n = (node + i) % MAX_NUMNODES; | |
5515 | ||
5516 | if (!nr_cpus_node(n)) | |
5517 | continue; | |
5518 | ||
5519 | /* Skip already used nodes */ | |
5520 | if (test_bit(n, used_nodes)) | |
5521 | continue; | |
5522 | ||
5523 | /* Simple min distance search */ | |
5524 | val = node_distance(node, n); | |
5525 | ||
5526 | if (val < min_val) { | |
5527 | min_val = val; | |
5528 | best_node = n; | |
5529 | } | |
5530 | } | |
5531 | ||
5532 | set_bit(best_node, used_nodes); | |
5533 | return best_node; | |
5534 | } | |
5535 | ||
5536 | /** | |
5537 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5538 | * @node: node whose cpumask we're constructing | |
5539 | * @size: number of nodes to include in this span | |
5540 | * | |
5541 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5542 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5543 | * out optimally. | |
5544 | */ | |
5545 | static cpumask_t sched_domain_node_span(int node) | |
5546 | { | |
9c1cfda2 | 5547 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5548 | cpumask_t span, nodemask; |
5549 | int i; | |
9c1cfda2 JH |
5550 | |
5551 | cpus_clear(span); | |
5552 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5553 | ||
5554 | nodemask = node_to_cpumask(node); | |
5555 | cpus_or(span, span, nodemask); | |
5556 | set_bit(node, used_nodes); | |
5557 | ||
5558 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5559 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5560 | |
9c1cfda2 JH |
5561 | nodemask = node_to_cpumask(next_node); |
5562 | cpus_or(span, span, nodemask); | |
5563 | } | |
5564 | ||
5565 | return span; | |
5566 | } | |
5567 | #endif | |
5568 | ||
5c45bf27 | 5569 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5570 | |
9c1cfda2 | 5571 | /* |
48f24c4d | 5572 | * SMT sched-domains: |
9c1cfda2 | 5573 | */ |
1da177e4 LT |
5574 | #ifdef CONFIG_SCHED_SMT |
5575 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5576 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5577 | |
6711cab4 SS |
5578 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5579 | struct sched_group **sg) | |
1da177e4 | 5580 | { |
6711cab4 SS |
5581 | if (sg) |
5582 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5583 | return cpu; |
5584 | } | |
5585 | #endif | |
5586 | ||
48f24c4d IM |
5587 | /* |
5588 | * multi-core sched-domains: | |
5589 | */ | |
1e9f28fa SS |
5590 | #ifdef CONFIG_SCHED_MC |
5591 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5592 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5593 | #endif |
5594 | ||
5595 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5596 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5597 | struct sched_group **sg) | |
1e9f28fa | 5598 | { |
6711cab4 | 5599 | int group; |
a616058b SS |
5600 | cpumask_t mask = cpu_sibling_map[cpu]; |
5601 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5602 | group = first_cpu(mask); |
5603 | if (sg) | |
5604 | *sg = &per_cpu(sched_group_core, group); | |
5605 | return group; | |
1e9f28fa SS |
5606 | } |
5607 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5608 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5609 | struct sched_group **sg) | |
1e9f28fa | 5610 | { |
6711cab4 SS |
5611 | if (sg) |
5612 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5613 | return cpu; |
5614 | } | |
5615 | #endif | |
5616 | ||
1da177e4 | 5617 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5618 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5619 | |
6711cab4 SS |
5620 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5621 | struct sched_group **sg) | |
1da177e4 | 5622 | { |
6711cab4 | 5623 | int group; |
48f24c4d | 5624 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5625 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5626 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5627 | group = first_cpu(mask); |
1e9f28fa | 5628 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5629 | cpumask_t mask = cpu_sibling_map[cpu]; |
5630 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5631 | group = first_cpu(mask); |
1da177e4 | 5632 | #else |
6711cab4 | 5633 | group = cpu; |
1da177e4 | 5634 | #endif |
6711cab4 SS |
5635 | if (sg) |
5636 | *sg = &per_cpu(sched_group_phys, group); | |
5637 | return group; | |
1da177e4 LT |
5638 | } |
5639 | ||
5640 | #ifdef CONFIG_NUMA | |
1da177e4 | 5641 | /* |
9c1cfda2 JH |
5642 | * The init_sched_build_groups can't handle what we want to do with node |
5643 | * groups, so roll our own. Now each node has its own list of groups which | |
5644 | * gets dynamically allocated. | |
1da177e4 | 5645 | */ |
9c1cfda2 | 5646 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5647 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5648 | |
9c1cfda2 | 5649 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5650 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5651 | |
6711cab4 SS |
5652 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5653 | struct sched_group **sg) | |
9c1cfda2 | 5654 | { |
6711cab4 SS |
5655 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5656 | int group; | |
5657 | ||
5658 | cpus_and(nodemask, nodemask, *cpu_map); | |
5659 | group = first_cpu(nodemask); | |
5660 | ||
5661 | if (sg) | |
5662 | *sg = &per_cpu(sched_group_allnodes, group); | |
5663 | return group; | |
1da177e4 | 5664 | } |
6711cab4 | 5665 | |
08069033 SS |
5666 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5667 | { | |
5668 | struct sched_group *sg = group_head; | |
5669 | int j; | |
5670 | ||
5671 | if (!sg) | |
5672 | return; | |
5673 | next_sg: | |
5674 | for_each_cpu_mask(j, sg->cpumask) { | |
5675 | struct sched_domain *sd; | |
5676 | ||
5677 | sd = &per_cpu(phys_domains, j); | |
5678 | if (j != first_cpu(sd->groups->cpumask)) { | |
5679 | /* | |
5680 | * Only add "power" once for each | |
5681 | * physical package. | |
5682 | */ | |
5683 | continue; | |
5684 | } | |
5685 | ||
5517d86b | 5686 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5687 | } |
5688 | sg = sg->next; | |
5689 | if (sg != group_head) | |
5690 | goto next_sg; | |
5691 | } | |
1da177e4 LT |
5692 | #endif |
5693 | ||
a616058b | 5694 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5695 | /* Free memory allocated for various sched_group structures */ |
5696 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5697 | { | |
a616058b | 5698 | int cpu, i; |
51888ca2 SV |
5699 | |
5700 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5701 | struct sched_group **sched_group_nodes |
5702 | = sched_group_nodes_bycpu[cpu]; | |
5703 | ||
51888ca2 SV |
5704 | if (!sched_group_nodes) |
5705 | continue; | |
5706 | ||
5707 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5708 | cpumask_t nodemask = node_to_cpumask(i); | |
5709 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5710 | ||
5711 | cpus_and(nodemask, nodemask, *cpu_map); | |
5712 | if (cpus_empty(nodemask)) | |
5713 | continue; | |
5714 | ||
5715 | if (sg == NULL) | |
5716 | continue; | |
5717 | sg = sg->next; | |
5718 | next_sg: | |
5719 | oldsg = sg; | |
5720 | sg = sg->next; | |
5721 | kfree(oldsg); | |
5722 | if (oldsg != sched_group_nodes[i]) | |
5723 | goto next_sg; | |
5724 | } | |
5725 | kfree(sched_group_nodes); | |
5726 | sched_group_nodes_bycpu[cpu] = NULL; | |
5727 | } | |
51888ca2 | 5728 | } |
a616058b SS |
5729 | #else |
5730 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5731 | { | |
5732 | } | |
5733 | #endif | |
51888ca2 | 5734 | |
89c4710e SS |
5735 | /* |
5736 | * Initialize sched groups cpu_power. | |
5737 | * | |
5738 | * cpu_power indicates the capacity of sched group, which is used while | |
5739 | * distributing the load between different sched groups in a sched domain. | |
5740 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5741 | * there are asymmetries in the topology. If there are asymmetries, group | |
5742 | * having more cpu_power will pickup more load compared to the group having | |
5743 | * less cpu_power. | |
5744 | * | |
5745 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5746 | * the maximum number of tasks a group can handle in the presence of other idle | |
5747 | * or lightly loaded groups in the same sched domain. | |
5748 | */ | |
5749 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5750 | { | |
5751 | struct sched_domain *child; | |
5752 | struct sched_group *group; | |
5753 | ||
5754 | WARN_ON(!sd || !sd->groups); | |
5755 | ||
5756 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5757 | return; | |
5758 | ||
5759 | child = sd->child; | |
5760 | ||
5517d86b ED |
5761 | sd->groups->__cpu_power = 0; |
5762 | ||
89c4710e SS |
5763 | /* |
5764 | * For perf policy, if the groups in child domain share resources | |
5765 | * (for example cores sharing some portions of the cache hierarchy | |
5766 | * or SMT), then set this domain groups cpu_power such that each group | |
5767 | * can handle only one task, when there are other idle groups in the | |
5768 | * same sched domain. | |
5769 | */ | |
5770 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5771 | (child->flags & | |
5772 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5773 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5774 | return; |
5775 | } | |
5776 | ||
89c4710e SS |
5777 | /* |
5778 | * add cpu_power of each child group to this groups cpu_power | |
5779 | */ | |
5780 | group = child->groups; | |
5781 | do { | |
5517d86b | 5782 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5783 | group = group->next; |
5784 | } while (group != child->groups); | |
5785 | } | |
5786 | ||
1da177e4 | 5787 | /* |
1a20ff27 DG |
5788 | * Build sched domains for a given set of cpus and attach the sched domains |
5789 | * to the individual cpus | |
1da177e4 | 5790 | */ |
51888ca2 | 5791 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5792 | { |
5793 | int i; | |
d1b55138 JH |
5794 | #ifdef CONFIG_NUMA |
5795 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 5796 | int sd_allnodes = 0; |
d1b55138 JH |
5797 | |
5798 | /* | |
5799 | * Allocate the per-node list of sched groups | |
5800 | */ | |
dd41f596 | 5801 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 5802 | GFP_KERNEL); |
d1b55138 JH |
5803 | if (!sched_group_nodes) { |
5804 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 5805 | return -ENOMEM; |
d1b55138 JH |
5806 | } |
5807 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
5808 | #endif | |
1da177e4 LT |
5809 | |
5810 | /* | |
1a20ff27 | 5811 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 5812 | */ |
1a20ff27 | 5813 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
5814 | struct sched_domain *sd = NULL, *p; |
5815 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
5816 | ||
1a20ff27 | 5817 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5818 | |
5819 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
5820 | if (cpus_weight(*cpu_map) > |
5821 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
5822 | sd = &per_cpu(allnodes_domains, i); |
5823 | *sd = SD_ALLNODES_INIT; | |
5824 | sd->span = *cpu_map; | |
6711cab4 | 5825 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 5826 | p = sd; |
6711cab4 | 5827 | sd_allnodes = 1; |
9c1cfda2 JH |
5828 | } else |
5829 | p = NULL; | |
5830 | ||
1da177e4 | 5831 | sd = &per_cpu(node_domains, i); |
1da177e4 | 5832 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
5833 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
5834 | sd->parent = p; | |
1a848870 SS |
5835 | if (p) |
5836 | p->child = sd; | |
9c1cfda2 | 5837 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
5838 | #endif |
5839 | ||
5840 | p = sd; | |
5841 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
5842 | *sd = SD_CPU_INIT; |
5843 | sd->span = nodemask; | |
5844 | sd->parent = p; | |
1a848870 SS |
5845 | if (p) |
5846 | p->child = sd; | |
6711cab4 | 5847 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 5848 | |
1e9f28fa SS |
5849 | #ifdef CONFIG_SCHED_MC |
5850 | p = sd; | |
5851 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
5852 | *sd = SD_MC_INIT; |
5853 | sd->span = cpu_coregroup_map(i); | |
5854 | cpus_and(sd->span, sd->span, *cpu_map); | |
5855 | sd->parent = p; | |
1a848870 | 5856 | p->child = sd; |
6711cab4 | 5857 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
5858 | #endif |
5859 | ||
1da177e4 LT |
5860 | #ifdef CONFIG_SCHED_SMT |
5861 | p = sd; | |
5862 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
5863 | *sd = SD_SIBLING_INIT; |
5864 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 5865 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 5866 | sd->parent = p; |
1a848870 | 5867 | p->child = sd; |
6711cab4 | 5868 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
5869 | #endif |
5870 | } | |
5871 | ||
5872 | #ifdef CONFIG_SCHED_SMT | |
5873 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 5874 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 5875 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 5876 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
5877 | if (i != first_cpu(this_sibling_map)) |
5878 | continue; | |
5879 | ||
dd41f596 IM |
5880 | init_sched_build_groups(this_sibling_map, cpu_map, |
5881 | &cpu_to_cpu_group); | |
1da177e4 LT |
5882 | } |
5883 | #endif | |
5884 | ||
1e9f28fa SS |
5885 | #ifdef CONFIG_SCHED_MC |
5886 | /* Set up multi-core groups */ | |
5887 | for_each_cpu_mask(i, *cpu_map) { | |
5888 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
5889 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
5890 | if (i != first_cpu(this_core_map)) | |
5891 | continue; | |
dd41f596 IM |
5892 | init_sched_build_groups(this_core_map, cpu_map, |
5893 | &cpu_to_core_group); | |
1e9f28fa SS |
5894 | } |
5895 | #endif | |
5896 | ||
1da177e4 LT |
5897 | /* Set up physical groups */ |
5898 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5899 | cpumask_t nodemask = node_to_cpumask(i); | |
5900 | ||
1a20ff27 | 5901 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5902 | if (cpus_empty(nodemask)) |
5903 | continue; | |
5904 | ||
6711cab4 | 5905 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
5906 | } |
5907 | ||
5908 | #ifdef CONFIG_NUMA | |
5909 | /* Set up node groups */ | |
6711cab4 | 5910 | if (sd_allnodes) |
dd41f596 IM |
5911 | init_sched_build_groups(*cpu_map, cpu_map, |
5912 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
5913 | |
5914 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5915 | /* Set up node groups */ | |
5916 | struct sched_group *sg, *prev; | |
5917 | cpumask_t nodemask = node_to_cpumask(i); | |
5918 | cpumask_t domainspan; | |
5919 | cpumask_t covered = CPU_MASK_NONE; | |
5920 | int j; | |
5921 | ||
5922 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
5923 | if (cpus_empty(nodemask)) { |
5924 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 5925 | continue; |
d1b55138 | 5926 | } |
9c1cfda2 JH |
5927 | |
5928 | domainspan = sched_domain_node_span(i); | |
5929 | cpus_and(domainspan, domainspan, *cpu_map); | |
5930 | ||
15f0b676 | 5931 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
5932 | if (!sg) { |
5933 | printk(KERN_WARNING "Can not alloc domain group for " | |
5934 | "node %d\n", i); | |
5935 | goto error; | |
5936 | } | |
9c1cfda2 JH |
5937 | sched_group_nodes[i] = sg; |
5938 | for_each_cpu_mask(j, nodemask) { | |
5939 | struct sched_domain *sd; | |
9761eea8 | 5940 | |
9c1cfda2 JH |
5941 | sd = &per_cpu(node_domains, j); |
5942 | sd->groups = sg; | |
9c1cfda2 | 5943 | } |
5517d86b | 5944 | sg->__cpu_power = 0; |
9c1cfda2 | 5945 | sg->cpumask = nodemask; |
51888ca2 | 5946 | sg->next = sg; |
9c1cfda2 JH |
5947 | cpus_or(covered, covered, nodemask); |
5948 | prev = sg; | |
5949 | ||
5950 | for (j = 0; j < MAX_NUMNODES; j++) { | |
5951 | cpumask_t tmp, notcovered; | |
5952 | int n = (i + j) % MAX_NUMNODES; | |
5953 | ||
5954 | cpus_complement(notcovered, covered); | |
5955 | cpus_and(tmp, notcovered, *cpu_map); | |
5956 | cpus_and(tmp, tmp, domainspan); | |
5957 | if (cpus_empty(tmp)) | |
5958 | break; | |
5959 | ||
5960 | nodemask = node_to_cpumask(n); | |
5961 | cpus_and(tmp, tmp, nodemask); | |
5962 | if (cpus_empty(tmp)) | |
5963 | continue; | |
5964 | ||
15f0b676 SV |
5965 | sg = kmalloc_node(sizeof(struct sched_group), |
5966 | GFP_KERNEL, i); | |
9c1cfda2 JH |
5967 | if (!sg) { |
5968 | printk(KERN_WARNING | |
5969 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 5970 | goto error; |
9c1cfda2 | 5971 | } |
5517d86b | 5972 | sg->__cpu_power = 0; |
9c1cfda2 | 5973 | sg->cpumask = tmp; |
51888ca2 | 5974 | sg->next = prev->next; |
9c1cfda2 JH |
5975 | cpus_or(covered, covered, tmp); |
5976 | prev->next = sg; | |
5977 | prev = sg; | |
5978 | } | |
9c1cfda2 | 5979 | } |
1da177e4 LT |
5980 | #endif |
5981 | ||
5982 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 5983 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 5984 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
5985 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
5986 | ||
89c4710e | 5987 | init_sched_groups_power(i, sd); |
5c45bf27 | 5988 | } |
1da177e4 | 5989 | #endif |
1e9f28fa | 5990 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 5991 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
5992 | struct sched_domain *sd = &per_cpu(core_domains, i); |
5993 | ||
89c4710e | 5994 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
5995 | } |
5996 | #endif | |
1e9f28fa | 5997 | |
5c45bf27 | 5998 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
5999 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6000 | ||
89c4710e | 6001 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6002 | } |
6003 | ||
9c1cfda2 | 6004 | #ifdef CONFIG_NUMA |
08069033 SS |
6005 | for (i = 0; i < MAX_NUMNODES; i++) |
6006 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6007 | |
6711cab4 SS |
6008 | if (sd_allnodes) { |
6009 | struct sched_group *sg; | |
f712c0c7 | 6010 | |
6711cab4 | 6011 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6012 | init_numa_sched_groups_power(sg); |
6013 | } | |
9c1cfda2 JH |
6014 | #endif |
6015 | ||
1da177e4 | 6016 | /* Attach the domains */ |
1a20ff27 | 6017 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6018 | struct sched_domain *sd; |
6019 | #ifdef CONFIG_SCHED_SMT | |
6020 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6021 | #elif defined(CONFIG_SCHED_MC) |
6022 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6023 | #else |
6024 | sd = &per_cpu(phys_domains, i); | |
6025 | #endif | |
6026 | cpu_attach_domain(sd, i); | |
6027 | } | |
51888ca2 SV |
6028 | |
6029 | return 0; | |
6030 | ||
a616058b | 6031 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6032 | error: |
6033 | free_sched_groups(cpu_map); | |
6034 | return -ENOMEM; | |
a616058b | 6035 | #endif |
1da177e4 | 6036 | } |
1a20ff27 DG |
6037 | /* |
6038 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6039 | */ | |
51888ca2 | 6040 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6041 | { |
6042 | cpumask_t cpu_default_map; | |
51888ca2 | 6043 | int err; |
1da177e4 | 6044 | |
1a20ff27 DG |
6045 | /* |
6046 | * Setup mask for cpus without special case scheduling requirements. | |
6047 | * For now this just excludes isolated cpus, but could be used to | |
6048 | * exclude other special cases in the future. | |
6049 | */ | |
6050 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6051 | ||
51888ca2 SV |
6052 | err = build_sched_domains(&cpu_default_map); |
6053 | ||
6054 | return err; | |
1a20ff27 DG |
6055 | } |
6056 | ||
6057 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6058 | { |
51888ca2 | 6059 | free_sched_groups(cpu_map); |
9c1cfda2 | 6060 | } |
1da177e4 | 6061 | |
1a20ff27 DG |
6062 | /* |
6063 | * Detach sched domains from a group of cpus specified in cpu_map | |
6064 | * These cpus will now be attached to the NULL domain | |
6065 | */ | |
858119e1 | 6066 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6067 | { |
6068 | int i; | |
6069 | ||
6070 | for_each_cpu_mask(i, *cpu_map) | |
6071 | cpu_attach_domain(NULL, i); | |
6072 | synchronize_sched(); | |
6073 | arch_destroy_sched_domains(cpu_map); | |
6074 | } | |
6075 | ||
6076 | /* | |
6077 | * Partition sched domains as specified by the cpumasks below. | |
6078 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6079 | * waits for a RCU quiescent period, recalculates sched | |
6080 | * domain information and then attaches them back to the | |
6081 | * correct sched domains | |
6082 | * Call with hotplug lock held | |
6083 | */ | |
51888ca2 | 6084 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6085 | { |
6086 | cpumask_t change_map; | |
51888ca2 | 6087 | int err = 0; |
1a20ff27 DG |
6088 | |
6089 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6090 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6091 | cpus_or(change_map, *partition1, *partition2); | |
6092 | ||
6093 | /* Detach sched domains from all of the affected cpus */ | |
6094 | detach_destroy_domains(&change_map); | |
6095 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6096 | err = build_sched_domains(partition1); |
6097 | if (!err && !cpus_empty(*partition2)) | |
6098 | err = build_sched_domains(partition2); | |
6099 | ||
6100 | return err; | |
1a20ff27 DG |
6101 | } |
6102 | ||
5c45bf27 SS |
6103 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6104 | int arch_reinit_sched_domains(void) | |
6105 | { | |
6106 | int err; | |
6107 | ||
5be9361c | 6108 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6109 | detach_destroy_domains(&cpu_online_map); |
6110 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6111 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6112 | |
6113 | return err; | |
6114 | } | |
6115 | ||
6116 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6117 | { | |
6118 | int ret; | |
6119 | ||
6120 | if (buf[0] != '0' && buf[0] != '1') | |
6121 | return -EINVAL; | |
6122 | ||
6123 | if (smt) | |
6124 | sched_smt_power_savings = (buf[0] == '1'); | |
6125 | else | |
6126 | sched_mc_power_savings = (buf[0] == '1'); | |
6127 | ||
6128 | ret = arch_reinit_sched_domains(); | |
6129 | ||
6130 | return ret ? ret : count; | |
6131 | } | |
6132 | ||
6133 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6134 | { | |
6135 | int err = 0; | |
48f24c4d | 6136 | |
5c45bf27 SS |
6137 | #ifdef CONFIG_SCHED_SMT |
6138 | if (smt_capable()) | |
6139 | err = sysfs_create_file(&cls->kset.kobj, | |
6140 | &attr_sched_smt_power_savings.attr); | |
6141 | #endif | |
6142 | #ifdef CONFIG_SCHED_MC | |
6143 | if (!err && mc_capable()) | |
6144 | err = sysfs_create_file(&cls->kset.kobj, | |
6145 | &attr_sched_mc_power_savings.attr); | |
6146 | #endif | |
6147 | return err; | |
6148 | } | |
6149 | #endif | |
6150 | ||
6151 | #ifdef CONFIG_SCHED_MC | |
6152 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6153 | { | |
6154 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6155 | } | |
48f24c4d IM |
6156 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6157 | const char *buf, size_t count) | |
5c45bf27 SS |
6158 | { |
6159 | return sched_power_savings_store(buf, count, 0); | |
6160 | } | |
6161 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6162 | sched_mc_power_savings_store); | |
6163 | #endif | |
6164 | ||
6165 | #ifdef CONFIG_SCHED_SMT | |
6166 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6167 | { | |
6168 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6169 | } | |
48f24c4d IM |
6170 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6171 | const char *buf, size_t count) | |
5c45bf27 SS |
6172 | { |
6173 | return sched_power_savings_store(buf, count, 1); | |
6174 | } | |
6175 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6176 | sched_smt_power_savings_store); | |
6177 | #endif | |
6178 | ||
1da177e4 LT |
6179 | /* |
6180 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6181 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6182 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6183 | * which will prevent rebalancing while the sched domains are recalculated. |
6184 | */ | |
6185 | static int update_sched_domains(struct notifier_block *nfb, | |
6186 | unsigned long action, void *hcpu) | |
6187 | { | |
1da177e4 LT |
6188 | switch (action) { |
6189 | case CPU_UP_PREPARE: | |
8bb78442 | 6190 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6191 | case CPU_DOWN_PREPARE: |
8bb78442 | 6192 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6193 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6194 | return NOTIFY_OK; |
6195 | ||
6196 | case CPU_UP_CANCELED: | |
8bb78442 | 6197 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6198 | case CPU_DOWN_FAILED: |
8bb78442 | 6199 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6200 | case CPU_ONLINE: |
8bb78442 | 6201 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6202 | case CPU_DEAD: |
8bb78442 | 6203 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6204 | /* |
6205 | * Fall through and re-initialise the domains. | |
6206 | */ | |
6207 | break; | |
6208 | default: | |
6209 | return NOTIFY_DONE; | |
6210 | } | |
6211 | ||
6212 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6213 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6214 | |
6215 | return NOTIFY_OK; | |
6216 | } | |
1da177e4 LT |
6217 | |
6218 | void __init sched_init_smp(void) | |
6219 | { | |
5c1e1767 NP |
6220 | cpumask_t non_isolated_cpus; |
6221 | ||
5be9361c | 6222 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6223 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6224 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6225 | if (cpus_empty(non_isolated_cpus)) |
6226 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6227 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6228 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6229 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6230 | |
6231 | /* Move init over to a non-isolated CPU */ | |
6232 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6233 | BUG(); | |
dd41f596 | 6234 | sched_init_granularity(); |
1da177e4 LT |
6235 | } |
6236 | #else | |
6237 | void __init sched_init_smp(void) | |
6238 | { | |
dd41f596 | 6239 | sched_init_granularity(); |
1da177e4 LT |
6240 | } |
6241 | #endif /* CONFIG_SMP */ | |
6242 | ||
6243 | int in_sched_functions(unsigned long addr) | |
6244 | { | |
6245 | /* Linker adds these: start and end of __sched functions */ | |
6246 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6247 | |
1da177e4 LT |
6248 | return in_lock_functions(addr) || |
6249 | (addr >= (unsigned long)__sched_text_start | |
6250 | && addr < (unsigned long)__sched_text_end); | |
6251 | } | |
6252 | ||
dd41f596 IM |
6253 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
6254 | { | |
6255 | cfs_rq->tasks_timeline = RB_ROOT; | |
6256 | cfs_rq->fair_clock = 1; | |
6257 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6258 | cfs_rq->rq = rq; | |
6259 | #endif | |
6260 | } | |
6261 | ||
1da177e4 LT |
6262 | void __init sched_init(void) |
6263 | { | |
dd41f596 | 6264 | u64 now = sched_clock(); |
476f3534 | 6265 | int highest_cpu = 0; |
dd41f596 IM |
6266 | int i, j; |
6267 | ||
6268 | /* | |
6269 | * Link up the scheduling class hierarchy: | |
6270 | */ | |
6271 | rt_sched_class.next = &fair_sched_class; | |
6272 | fair_sched_class.next = &idle_sched_class; | |
6273 | idle_sched_class.next = NULL; | |
1da177e4 | 6274 | |
0a945022 | 6275 | for_each_possible_cpu(i) { |
dd41f596 | 6276 | struct rt_prio_array *array; |
70b97a7f | 6277 | struct rq *rq; |
1da177e4 LT |
6278 | |
6279 | rq = cpu_rq(i); | |
6280 | spin_lock_init(&rq->lock); | |
fcb99371 | 6281 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6282 | rq->nr_running = 0; |
dd41f596 IM |
6283 | rq->clock = 1; |
6284 | init_cfs_rq(&rq->cfs, rq); | |
6285 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6286 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
6287 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6288 | #endif | |
6289 | rq->ls.load_update_last = now; | |
6290 | rq->ls.load_update_start = now; | |
1da177e4 | 6291 | |
dd41f596 IM |
6292 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6293 | rq->cpu_load[j] = 0; | |
1da177e4 | 6294 | #ifdef CONFIG_SMP |
41c7ce9a | 6295 | rq->sd = NULL; |
1da177e4 | 6296 | rq->active_balance = 0; |
dd41f596 | 6297 | rq->next_balance = jiffies; |
1da177e4 | 6298 | rq->push_cpu = 0; |
0a2966b4 | 6299 | rq->cpu = i; |
1da177e4 LT |
6300 | rq->migration_thread = NULL; |
6301 | INIT_LIST_HEAD(&rq->migration_queue); | |
6302 | #endif | |
6303 | atomic_set(&rq->nr_iowait, 0); | |
6304 | ||
dd41f596 IM |
6305 | array = &rq->rt.active; |
6306 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6307 | INIT_LIST_HEAD(array->queue + j); | |
6308 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6309 | } |
476f3534 | 6310 | highest_cpu = i; |
dd41f596 IM |
6311 | /* delimiter for bitsearch: */ |
6312 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6313 | } |
6314 | ||
2dd73a4f | 6315 | set_load_weight(&init_task); |
b50f60ce | 6316 | |
c9819f45 | 6317 | #ifdef CONFIG_SMP |
476f3534 | 6318 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6319 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6320 | #endif | |
6321 | ||
b50f60ce HC |
6322 | #ifdef CONFIG_RT_MUTEXES |
6323 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6324 | #endif | |
6325 | ||
1da177e4 LT |
6326 | /* |
6327 | * The boot idle thread does lazy MMU switching as well: | |
6328 | */ | |
6329 | atomic_inc(&init_mm.mm_count); | |
6330 | enter_lazy_tlb(&init_mm, current); | |
6331 | ||
6332 | /* | |
6333 | * Make us the idle thread. Technically, schedule() should not be | |
6334 | * called from this thread, however somewhere below it might be, | |
6335 | * but because we are the idle thread, we just pick up running again | |
6336 | * when this runqueue becomes "idle". | |
6337 | */ | |
6338 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6339 | /* |
6340 | * During early bootup we pretend to be a normal task: | |
6341 | */ | |
6342 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6343 | } |
6344 | ||
6345 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6346 | void __might_sleep(char *file, int line) | |
6347 | { | |
48f24c4d | 6348 | #ifdef in_atomic |
1da177e4 LT |
6349 | static unsigned long prev_jiffy; /* ratelimiting */ |
6350 | ||
6351 | if ((in_atomic() || irqs_disabled()) && | |
6352 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6353 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6354 | return; | |
6355 | prev_jiffy = jiffies; | |
91368d73 | 6356 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6357 | " context at %s:%d\n", file, line); |
6358 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6359 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6360 | debug_show_held_locks(current); |
3117df04 IM |
6361 | if (irqs_disabled()) |
6362 | print_irqtrace_events(current); | |
1da177e4 LT |
6363 | dump_stack(); |
6364 | } | |
6365 | #endif | |
6366 | } | |
6367 | EXPORT_SYMBOL(__might_sleep); | |
6368 | #endif | |
6369 | ||
6370 | #ifdef CONFIG_MAGIC_SYSRQ | |
6371 | void normalize_rt_tasks(void) | |
6372 | { | |
a0f98a1c | 6373 | struct task_struct *g, *p; |
1da177e4 | 6374 | unsigned long flags; |
70b97a7f | 6375 | struct rq *rq; |
dd41f596 | 6376 | int on_rq; |
1da177e4 LT |
6377 | |
6378 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6379 | do_each_thread(g, p) { |
dd41f596 IM |
6380 | p->se.fair_key = 0; |
6381 | p->se.wait_runtime = 0; | |
6382 | p->se.wait_start_fair = 0; | |
6383 | p->se.wait_start = 0; | |
6384 | p->se.exec_start = 0; | |
6385 | p->se.sleep_start = 0; | |
6386 | p->se.sleep_start_fair = 0; | |
6387 | p->se.block_start = 0; | |
6388 | task_rq(p)->cfs.fair_clock = 0; | |
6389 | task_rq(p)->clock = 0; | |
6390 | ||
6391 | if (!rt_task(p)) { | |
6392 | /* | |
6393 | * Renice negative nice level userspace | |
6394 | * tasks back to 0: | |
6395 | */ | |
6396 | if (TASK_NICE(p) < 0 && p->mm) | |
6397 | set_user_nice(p, 0); | |
1da177e4 | 6398 | continue; |
dd41f596 | 6399 | } |
1da177e4 | 6400 | |
b29739f9 IM |
6401 | spin_lock_irqsave(&p->pi_lock, flags); |
6402 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6403 | #ifdef CONFIG_SMP |
6404 | /* | |
6405 | * Do not touch the migration thread: | |
6406 | */ | |
6407 | if (p == rq->migration_thread) | |
6408 | goto out_unlock; | |
6409 | #endif | |
1da177e4 | 6410 | |
dd41f596 IM |
6411 | on_rq = p->se.on_rq; |
6412 | if (on_rq) | |
6413 | deactivate_task(task_rq(p), p, 0); | |
6414 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6415 | if (on_rq) { | |
6416 | activate_task(task_rq(p), p, 0); | |
1da177e4 LT |
6417 | resched_task(rq->curr); |
6418 | } | |
dd41f596 IM |
6419 | #ifdef CONFIG_SMP |
6420 | out_unlock: | |
6421 | #endif | |
b29739f9 IM |
6422 | __task_rq_unlock(rq); |
6423 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6424 | } while_each_thread(g, p); |
6425 | ||
1da177e4 LT |
6426 | read_unlock_irq(&tasklist_lock); |
6427 | } | |
6428 | ||
6429 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6430 | |
6431 | #ifdef CONFIG_IA64 | |
6432 | /* | |
6433 | * These functions are only useful for the IA64 MCA handling. | |
6434 | * | |
6435 | * They can only be called when the whole system has been | |
6436 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6437 | * activity can take place. Using them for anything else would | |
6438 | * be a serious bug, and as a result, they aren't even visible | |
6439 | * under any other configuration. | |
6440 | */ | |
6441 | ||
6442 | /** | |
6443 | * curr_task - return the current task for a given cpu. | |
6444 | * @cpu: the processor in question. | |
6445 | * | |
6446 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6447 | */ | |
36c8b586 | 6448 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6449 | { |
6450 | return cpu_curr(cpu); | |
6451 | } | |
6452 | ||
6453 | /** | |
6454 | * set_curr_task - set the current task for a given cpu. | |
6455 | * @cpu: the processor in question. | |
6456 | * @p: the task pointer to set. | |
6457 | * | |
6458 | * Description: This function must only be used when non-maskable interrupts | |
6459 | * are serviced on a separate stack. It allows the architecture to switch the | |
6460 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6461 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6462 | * and caller must save the original value of the current task (see | |
6463 | * curr_task() above) and restore that value before reenabling interrupts and | |
6464 | * re-starting the system. | |
6465 | * | |
6466 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6467 | */ | |
36c8b586 | 6468 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6469 | { |
6470 | cpu_curr(cpu) = p; | |
6471 | } | |
6472 | ||
6473 | #endif |