]>
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> | |
e692ab53 | 56 | #include <linux/sysctl.h> |
1da177e4 LT |
57 | #include <linux/syscalls.h> |
58 | #include <linux/times.h> | |
8f0ab514 | 59 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 60 | #include <linux/kprobes.h> |
0ff92245 | 61 | #include <linux/delayacct.h> |
5517d86b | 62 | #include <linux/reciprocal_div.h> |
dff06c15 | 63 | #include <linux/unistd.h> |
1da177e4 | 64 | |
5517d86b | 65 | #include <asm/tlb.h> |
1da177e4 | 66 | |
b035b6de AD |
67 | /* |
68 | * Scheduler clock - returns current time in nanosec units. | |
69 | * This is default implementation. | |
70 | * Architectures and sub-architectures can override this. | |
71 | */ | |
72 | unsigned long long __attribute__((weak)) sched_clock(void) | |
73 | { | |
74 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
75 | } | |
76 | ||
1da177e4 LT |
77 | /* |
78 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
79 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
80 | * and back. | |
81 | */ | |
82 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
83 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
84 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
85 | ||
86 | /* | |
87 | * 'User priority' is the nice value converted to something we | |
88 | * can work with better when scaling various scheduler parameters, | |
89 | * it's a [ 0 ... 39 ] range. | |
90 | */ | |
91 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
92 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
93 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
94 | ||
95 | /* | |
96 | * Some helpers for converting nanosecond timing to jiffy resolution | |
97 | */ | |
98 | #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) | |
99 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) | |
100 | ||
6aa645ea IM |
101 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
102 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
103 | ||
1da177e4 LT |
104 | /* |
105 | * These are the 'tuning knobs' of the scheduler: | |
106 | * | |
107 | * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger), | |
108 | * default timeslice is 100 msecs, maximum timeslice is 800 msecs. | |
109 | * Timeslices get refilled after they expire. | |
110 | */ | |
111 | #define MIN_TIMESLICE max(5 * HZ / 1000, 1) | |
112 | #define DEF_TIMESLICE (100 * HZ / 1000) | |
2dd73a4f | 113 | |
5517d86b ED |
114 | #ifdef CONFIG_SMP |
115 | /* | |
116 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
117 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
118 | */ | |
119 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
120 | { | |
121 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
122 | } | |
123 | ||
124 | /* | |
125 | * Each time a sched group cpu_power is changed, | |
126 | * we must compute its reciprocal value | |
127 | */ | |
128 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
129 | { | |
130 | sg->__cpu_power += val; | |
131 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
132 | } | |
133 | #endif | |
134 | ||
634fa8c9 IM |
135 | #define SCALE_PRIO(x, prio) \ |
136 | max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE) | |
137 | ||
91fcdd4e | 138 | /* |
634fa8c9 | 139 | * static_prio_timeslice() scales user-nice values [ -20 ... 0 ... 19 ] |
91fcdd4e | 140 | * to time slice values: [800ms ... 100ms ... 5ms] |
91fcdd4e | 141 | */ |
634fa8c9 | 142 | static unsigned int static_prio_timeslice(int static_prio) |
2dd73a4f | 143 | { |
634fa8c9 IM |
144 | if (static_prio == NICE_TO_PRIO(19)) |
145 | return 1; | |
146 | ||
147 | if (static_prio < NICE_TO_PRIO(0)) | |
148 | return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio); | |
149 | else | |
150 | return SCALE_PRIO(DEF_TIMESLICE, static_prio); | |
2dd73a4f PW |
151 | } |
152 | ||
e05606d3 IM |
153 | static inline int rt_policy(int policy) |
154 | { | |
155 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
156 | return 1; | |
157 | return 0; | |
158 | } | |
159 | ||
160 | static inline int task_has_rt_policy(struct task_struct *p) | |
161 | { | |
162 | return rt_policy(p->policy); | |
163 | } | |
164 | ||
1da177e4 | 165 | /* |
6aa645ea | 166 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 167 | */ |
6aa645ea IM |
168 | struct rt_prio_array { |
169 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
170 | struct list_head queue[MAX_RT_PRIO]; | |
171 | }; | |
172 | ||
173 | struct load_stat { | |
174 | struct load_weight load; | |
175 | u64 load_update_start, load_update_last; | |
176 | unsigned long delta_fair, delta_exec, delta_stat; | |
177 | }; | |
178 | ||
179 | /* CFS-related fields in a runqueue */ | |
180 | struct cfs_rq { | |
181 | struct load_weight load; | |
182 | unsigned long nr_running; | |
183 | ||
184 | s64 fair_clock; | |
185 | u64 exec_clock; | |
186 | s64 wait_runtime; | |
187 | u64 sleeper_bonus; | |
188 | unsigned long wait_runtime_overruns, wait_runtime_underruns; | |
189 | ||
190 | struct rb_root tasks_timeline; | |
191 | struct rb_node *rb_leftmost; | |
192 | struct rb_node *rb_load_balance_curr; | |
193 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
194 | /* 'curr' points to currently running entity on this cfs_rq. | |
195 | * It is set to NULL otherwise (i.e when none are currently running). | |
196 | */ | |
197 | struct sched_entity *curr; | |
198 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ | |
199 | ||
200 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
201 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
202 | * (like users, containers etc.) | |
203 | * | |
204 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
205 | * list is used during load balance. | |
206 | */ | |
207 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
208 | #endif | |
209 | }; | |
1da177e4 | 210 | |
6aa645ea IM |
211 | /* Real-Time classes' related field in a runqueue: */ |
212 | struct rt_rq { | |
213 | struct rt_prio_array active; | |
214 | int rt_load_balance_idx; | |
215 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
216 | }; | |
217 | ||
1da177e4 LT |
218 | /* |
219 | * This is the main, per-CPU runqueue data structure. | |
220 | * | |
221 | * Locking rule: those places that want to lock multiple runqueues | |
222 | * (such as the load balancing or the thread migration code), lock | |
223 | * acquire operations must be ordered by ascending &runqueue. | |
224 | */ | |
70b97a7f | 225 | struct rq { |
6aa645ea | 226 | spinlock_t lock; /* runqueue lock */ |
1da177e4 LT |
227 | |
228 | /* | |
229 | * nr_running and cpu_load should be in the same cacheline because | |
230 | * remote CPUs use both these fields when doing load calculation. | |
231 | */ | |
232 | unsigned long nr_running; | |
6aa645ea IM |
233 | #define CPU_LOAD_IDX_MAX 5 |
234 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 235 | unsigned char idle_at_tick; |
46cb4b7c SS |
236 | #ifdef CONFIG_NO_HZ |
237 | unsigned char in_nohz_recently; | |
238 | #endif | |
6aa645ea IM |
239 | struct load_stat ls; /* capture load from *all* tasks on this cpu */ |
240 | unsigned long nr_load_updates; | |
241 | u64 nr_switches; | |
242 | ||
243 | struct cfs_rq cfs; | |
244 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
245 | struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */ | |
1da177e4 | 246 | #endif |
6aa645ea | 247 | struct rt_rq rt; |
1da177e4 LT |
248 | |
249 | /* | |
250 | * This is part of a global counter where only the total sum | |
251 | * over all CPUs matters. A task can increase this counter on | |
252 | * one CPU and if it got migrated afterwards it may decrease | |
253 | * it on another CPU. Always updated under the runqueue lock: | |
254 | */ | |
255 | unsigned long nr_uninterruptible; | |
256 | ||
36c8b586 | 257 | struct task_struct *curr, *idle; |
c9819f45 | 258 | unsigned long next_balance; |
1da177e4 | 259 | struct mm_struct *prev_mm; |
6aa645ea | 260 | |
6aa645ea IM |
261 | u64 clock, prev_clock_raw; |
262 | s64 clock_max_delta; | |
263 | ||
264 | unsigned int clock_warps, clock_overflows; | |
265 | unsigned int clock_unstable_events; | |
266 | ||
1da177e4 LT |
267 | atomic_t nr_iowait; |
268 | ||
269 | #ifdef CONFIG_SMP | |
270 | struct sched_domain *sd; | |
271 | ||
272 | /* For active balancing */ | |
273 | int active_balance; | |
274 | int push_cpu; | |
0a2966b4 | 275 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 276 | |
36c8b586 | 277 | struct task_struct *migration_thread; |
1da177e4 LT |
278 | struct list_head migration_queue; |
279 | #endif | |
280 | ||
281 | #ifdef CONFIG_SCHEDSTATS | |
282 | /* latency stats */ | |
283 | struct sched_info rq_sched_info; | |
284 | ||
285 | /* sys_sched_yield() stats */ | |
286 | unsigned long yld_exp_empty; | |
287 | unsigned long yld_act_empty; | |
288 | unsigned long yld_both_empty; | |
289 | unsigned long yld_cnt; | |
290 | ||
291 | /* schedule() stats */ | |
292 | unsigned long sched_switch; | |
293 | unsigned long sched_cnt; | |
294 | unsigned long sched_goidle; | |
295 | ||
296 | /* try_to_wake_up() stats */ | |
297 | unsigned long ttwu_cnt; | |
298 | unsigned long ttwu_local; | |
299 | #endif | |
fcb99371 | 300 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
301 | }; |
302 | ||
f34e3b61 | 303 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 304 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 305 | |
dd41f596 IM |
306 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
307 | { | |
308 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
309 | } | |
310 | ||
0a2966b4 CL |
311 | static inline int cpu_of(struct rq *rq) |
312 | { | |
313 | #ifdef CONFIG_SMP | |
314 | return rq->cpu; | |
315 | #else | |
316 | return 0; | |
317 | #endif | |
318 | } | |
319 | ||
20d315d4 IM |
320 | /* |
321 | * Per-runqueue clock, as finegrained as the platform can give us: | |
322 | */ | |
323 | static unsigned long long __rq_clock(struct rq *rq) | |
324 | { | |
325 | u64 prev_raw = rq->prev_clock_raw; | |
326 | u64 now = sched_clock(); | |
327 | s64 delta = now - prev_raw; | |
328 | u64 clock = rq->clock; | |
329 | ||
330 | /* | |
331 | * Protect against sched_clock() occasionally going backwards: | |
332 | */ | |
333 | if (unlikely(delta < 0)) { | |
334 | clock++; | |
335 | rq->clock_warps++; | |
336 | } else { | |
337 | /* | |
338 | * Catch too large forward jumps too: | |
339 | */ | |
340 | if (unlikely(delta > 2*TICK_NSEC)) { | |
341 | clock++; | |
342 | rq->clock_overflows++; | |
343 | } else { | |
344 | if (unlikely(delta > rq->clock_max_delta)) | |
345 | rq->clock_max_delta = delta; | |
346 | clock += delta; | |
347 | } | |
348 | } | |
349 | ||
350 | rq->prev_clock_raw = now; | |
351 | rq->clock = clock; | |
352 | ||
353 | return clock; | |
354 | } | |
355 | ||
7bfd0485 | 356 | static unsigned long long rq_clock(struct rq *rq) |
20d315d4 IM |
357 | { |
358 | int this_cpu = smp_processor_id(); | |
359 | ||
360 | if (this_cpu == cpu_of(rq)) | |
361 | return __rq_clock(rq); | |
362 | ||
363 | return rq->clock; | |
364 | } | |
365 | ||
674311d5 NP |
366 | /* |
367 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 368 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
369 | * |
370 | * The domain tree of any CPU may only be accessed from within | |
371 | * preempt-disabled sections. | |
372 | */ | |
48f24c4d IM |
373 | #define for_each_domain(cpu, __sd) \ |
374 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
375 | |
376 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
377 | #define this_rq() (&__get_cpu_var(runqueues)) | |
378 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
379 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
380 | ||
e436d800 IM |
381 | /* |
382 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
383 | * clock constructed from sched_clock(): | |
384 | */ | |
385 | unsigned long long cpu_clock(int cpu) | |
386 | { | |
e436d800 IM |
387 | unsigned long long now; |
388 | unsigned long flags; | |
389 | ||
2cd4d0ea IM |
390 | local_irq_save(flags); |
391 | now = rq_clock(cpu_rq(cpu)); | |
392 | local_irq_restore(flags); | |
e436d800 IM |
393 | |
394 | return now; | |
395 | } | |
396 | ||
138a8aeb IM |
397 | #ifdef CONFIG_FAIR_GROUP_SCHED |
398 | /* Change a task's ->cfs_rq if it moves across CPUs */ | |
399 | static inline void set_task_cfs_rq(struct task_struct *p) | |
400 | { | |
401 | p->se.cfs_rq = &task_rq(p)->cfs; | |
402 | } | |
403 | #else | |
404 | static inline void set_task_cfs_rq(struct task_struct *p) | |
405 | { | |
406 | } | |
407 | #endif | |
408 | ||
1da177e4 | 409 | #ifndef prepare_arch_switch |
4866cde0 NP |
410 | # define prepare_arch_switch(next) do { } while (0) |
411 | #endif | |
412 | #ifndef finish_arch_switch | |
413 | # define finish_arch_switch(prev) do { } while (0) | |
414 | #endif | |
415 | ||
416 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 417 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
418 | { |
419 | return rq->curr == p; | |
420 | } | |
421 | ||
70b97a7f | 422 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
423 | { |
424 | } | |
425 | ||
70b97a7f | 426 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 427 | { |
da04c035 IM |
428 | #ifdef CONFIG_DEBUG_SPINLOCK |
429 | /* this is a valid case when another task releases the spinlock */ | |
430 | rq->lock.owner = current; | |
431 | #endif | |
8a25d5de IM |
432 | /* |
433 | * If we are tracking spinlock dependencies then we have to | |
434 | * fix up the runqueue lock - which gets 'carried over' from | |
435 | * prev into current: | |
436 | */ | |
437 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
438 | ||
4866cde0 NP |
439 | spin_unlock_irq(&rq->lock); |
440 | } | |
441 | ||
442 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 443 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
444 | { |
445 | #ifdef CONFIG_SMP | |
446 | return p->oncpu; | |
447 | #else | |
448 | return rq->curr == p; | |
449 | #endif | |
450 | } | |
451 | ||
70b97a7f | 452 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
453 | { |
454 | #ifdef CONFIG_SMP | |
455 | /* | |
456 | * We can optimise this out completely for !SMP, because the | |
457 | * SMP rebalancing from interrupt is the only thing that cares | |
458 | * here. | |
459 | */ | |
460 | next->oncpu = 1; | |
461 | #endif | |
462 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
463 | spin_unlock_irq(&rq->lock); | |
464 | #else | |
465 | spin_unlock(&rq->lock); | |
466 | #endif | |
467 | } | |
468 | ||
70b97a7f | 469 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
470 | { |
471 | #ifdef CONFIG_SMP | |
472 | /* | |
473 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
474 | * We must ensure this doesn't happen until the switch is completely | |
475 | * finished. | |
476 | */ | |
477 | smp_wmb(); | |
478 | prev->oncpu = 0; | |
479 | #endif | |
480 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
481 | local_irq_enable(); | |
1da177e4 | 482 | #endif |
4866cde0 NP |
483 | } |
484 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 485 | |
b29739f9 IM |
486 | /* |
487 | * __task_rq_lock - lock the runqueue a given task resides on. | |
488 | * Must be called interrupts disabled. | |
489 | */ | |
70b97a7f | 490 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
491 | __acquires(rq->lock) |
492 | { | |
70b97a7f | 493 | struct rq *rq; |
b29739f9 IM |
494 | |
495 | repeat_lock_task: | |
496 | rq = task_rq(p); | |
497 | spin_lock(&rq->lock); | |
498 | if (unlikely(rq != task_rq(p))) { | |
499 | spin_unlock(&rq->lock); | |
500 | goto repeat_lock_task; | |
501 | } | |
502 | return rq; | |
503 | } | |
504 | ||
1da177e4 LT |
505 | /* |
506 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
507 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
508 | * explicitly disabling preemption. | |
509 | */ | |
70b97a7f | 510 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
511 | __acquires(rq->lock) |
512 | { | |
70b97a7f | 513 | struct rq *rq; |
1da177e4 LT |
514 | |
515 | repeat_lock_task: | |
516 | local_irq_save(*flags); | |
517 | rq = task_rq(p); | |
518 | spin_lock(&rq->lock); | |
519 | if (unlikely(rq != task_rq(p))) { | |
520 | spin_unlock_irqrestore(&rq->lock, *flags); | |
521 | goto repeat_lock_task; | |
522 | } | |
523 | return rq; | |
524 | } | |
525 | ||
70b97a7f | 526 | static inline void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
527 | __releases(rq->lock) |
528 | { | |
529 | spin_unlock(&rq->lock); | |
530 | } | |
531 | ||
70b97a7f | 532 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
533 | __releases(rq->lock) |
534 | { | |
535 | spin_unlock_irqrestore(&rq->lock, *flags); | |
536 | } | |
537 | ||
1da177e4 | 538 | /* |
cc2a73b5 | 539 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 540 | */ |
70b97a7f | 541 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
542 | __acquires(rq->lock) |
543 | { | |
70b97a7f | 544 | struct rq *rq; |
1da177e4 LT |
545 | |
546 | local_irq_disable(); | |
547 | rq = this_rq(); | |
548 | spin_lock(&rq->lock); | |
549 | ||
550 | return rq; | |
551 | } | |
552 | ||
1b9f19c2 IM |
553 | /* |
554 | * CPU frequency is/was unstable - start new by setting prev_clock_raw: | |
555 | */ | |
556 | void sched_clock_unstable_event(void) | |
557 | { | |
558 | unsigned long flags; | |
559 | struct rq *rq; | |
560 | ||
561 | rq = task_rq_lock(current, &flags); | |
562 | rq->prev_clock_raw = sched_clock(); | |
563 | rq->clock_unstable_events++; | |
564 | task_rq_unlock(rq, &flags); | |
565 | } | |
566 | ||
c24d20db IM |
567 | /* |
568 | * resched_task - mark a task 'to be rescheduled now'. | |
569 | * | |
570 | * On UP this means the setting of the need_resched flag, on SMP it | |
571 | * might also involve a cross-CPU call to trigger the scheduler on | |
572 | * the target CPU. | |
573 | */ | |
574 | #ifdef CONFIG_SMP | |
575 | ||
576 | #ifndef tsk_is_polling | |
577 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
578 | #endif | |
579 | ||
580 | static void resched_task(struct task_struct *p) | |
581 | { | |
582 | int cpu; | |
583 | ||
584 | assert_spin_locked(&task_rq(p)->lock); | |
585 | ||
586 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
587 | return; | |
588 | ||
589 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
590 | ||
591 | cpu = task_cpu(p); | |
592 | if (cpu == smp_processor_id()) | |
593 | return; | |
594 | ||
595 | /* NEED_RESCHED must be visible before we test polling */ | |
596 | smp_mb(); | |
597 | if (!tsk_is_polling(p)) | |
598 | smp_send_reschedule(cpu); | |
599 | } | |
600 | ||
601 | static void resched_cpu(int cpu) | |
602 | { | |
603 | struct rq *rq = cpu_rq(cpu); | |
604 | unsigned long flags; | |
605 | ||
606 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
607 | return; | |
608 | resched_task(cpu_curr(cpu)); | |
609 | spin_unlock_irqrestore(&rq->lock, flags); | |
610 | } | |
611 | #else | |
612 | static inline void resched_task(struct task_struct *p) | |
613 | { | |
614 | assert_spin_locked(&task_rq(p)->lock); | |
615 | set_tsk_need_resched(p); | |
616 | } | |
617 | #endif | |
618 | ||
45bf76df IM |
619 | static u64 div64_likely32(u64 divident, unsigned long divisor) |
620 | { | |
621 | #if BITS_PER_LONG == 32 | |
622 | if (likely(divident <= 0xffffffffULL)) | |
623 | return (u32)divident / divisor; | |
624 | do_div(divident, divisor); | |
625 | ||
626 | return divident; | |
627 | #else | |
628 | return divident / divisor; | |
629 | #endif | |
630 | } | |
631 | ||
632 | #if BITS_PER_LONG == 32 | |
633 | # define WMULT_CONST (~0UL) | |
634 | #else | |
635 | # define WMULT_CONST (1UL << 32) | |
636 | #endif | |
637 | ||
638 | #define WMULT_SHIFT 32 | |
639 | ||
cb1c4fc9 | 640 | static unsigned long |
45bf76df IM |
641 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
642 | struct load_weight *lw) | |
643 | { | |
644 | u64 tmp; | |
645 | ||
646 | if (unlikely(!lw->inv_weight)) | |
647 | lw->inv_weight = WMULT_CONST / lw->weight; | |
648 | ||
649 | tmp = (u64)delta_exec * weight; | |
650 | /* | |
651 | * Check whether we'd overflow the 64-bit multiplication: | |
652 | */ | |
653 | if (unlikely(tmp > WMULT_CONST)) { | |
654 | tmp = ((tmp >> WMULT_SHIFT/2) * lw->inv_weight) | |
655 | >> (WMULT_SHIFT/2); | |
656 | } else { | |
657 | tmp = (tmp * lw->inv_weight) >> WMULT_SHIFT; | |
658 | } | |
659 | ||
ecf691da | 660 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
661 | } |
662 | ||
663 | static inline unsigned long | |
664 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
665 | { | |
666 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
667 | } | |
668 | ||
669 | static void update_load_add(struct load_weight *lw, unsigned long inc) | |
670 | { | |
671 | lw->weight += inc; | |
672 | lw->inv_weight = 0; | |
673 | } | |
674 | ||
675 | static void update_load_sub(struct load_weight *lw, unsigned long dec) | |
676 | { | |
677 | lw->weight -= dec; | |
678 | lw->inv_weight = 0; | |
679 | } | |
680 | ||
2dd73a4f PW |
681 | /* |
682 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
683 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
684 | * each task makes to its run queue's load is weighted according to its | |
685 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
686 | * scaled version of the new time slice allocation that they receive on time | |
687 | * slice expiry etc. | |
688 | */ | |
689 | ||
dd41f596 IM |
690 | #define WEIGHT_IDLEPRIO 2 |
691 | #define WMULT_IDLEPRIO (1 << 31) | |
692 | ||
693 | /* | |
694 | * Nice levels are multiplicative, with a gentle 10% change for every | |
695 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
696 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
697 | * that remained on nice 0. | |
698 | * | |
699 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
700 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
701 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
702 | * If a task goes up by ~10% and another task goes down by ~10% then | |
703 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
704 | */ |
705 | static const int prio_to_weight[40] = { | |
706 | /* -20 */ 88818, 71054, 56843, 45475, 36380, 29104, 23283, 18626, 14901, 11921, | |
707 | /* -10 */ 9537, 7629, 6103, 4883, 3906, 3125, 2500, 2000, 1600, 1280, | |
708 | /* 0 */ NICE_0_LOAD /* 1024 */, | |
709 | /* 1 */ 819, 655, 524, 419, 336, 268, 215, 172, 137, | |
710 | /* 10 */ 110, 87, 70, 56, 45, 36, 29, 23, 18, 15, | |
711 | }; | |
712 | ||
5714d2de IM |
713 | /* |
714 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
715 | * | |
716 | * In cases where the weight does not change often, we can use the | |
717 | * precalculated inverse to speed up arithmetics by turning divisions | |
718 | * into multiplications: | |
719 | */ | |
dd41f596 | 720 | static const u32 prio_to_wmult[40] = { |
e4af30be IM |
721 | /* -20 */ 48356, 60446, 75558, 94446, 118058, |
722 | /* -15 */ 147573, 184467, 230589, 288233, 360285, | |
723 | /* -10 */ 450347, 562979, 703746, 879575, 1099582, | |
724 | /* -5 */ 1374389, 1717986, 2147483, 2684354, 3355443, | |
725 | /* 0 */ 4194304, 5244160, 6557201, 8196502, 10250518, | |
726 | /* 5 */ 12782640, 16025997, 19976592, 24970740, 31350126, | |
727 | /* 10 */ 39045157, 49367440, 61356675, 76695844, 95443717, | |
728 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 729 | }; |
2dd73a4f | 730 | |
dd41f596 IM |
731 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
732 | ||
733 | /* | |
734 | * runqueue iterator, to support SMP load-balancing between different | |
735 | * scheduling classes, without having to expose their internal data | |
736 | * structures to the load-balancing proper: | |
737 | */ | |
738 | struct rq_iterator { | |
739 | void *arg; | |
740 | struct task_struct *(*start)(void *); | |
741 | struct task_struct *(*next)(void *); | |
742 | }; | |
743 | ||
744 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
745 | unsigned long max_nr_move, unsigned long max_load_move, | |
746 | struct sched_domain *sd, enum cpu_idle_type idle, | |
747 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 748 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
749 | |
750 | #include "sched_stats.h" | |
751 | #include "sched_rt.c" | |
752 | #include "sched_fair.c" | |
753 | #include "sched_idletask.c" | |
754 | #ifdef CONFIG_SCHED_DEBUG | |
755 | # include "sched_debug.c" | |
756 | #endif | |
757 | ||
758 | #define sched_class_highest (&rt_sched_class) | |
759 | ||
9c217245 IM |
760 | static void __update_curr_load(struct rq *rq, struct load_stat *ls) |
761 | { | |
762 | if (rq->curr != rq->idle && ls->load.weight) { | |
763 | ls->delta_exec += ls->delta_stat; | |
764 | ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load); | |
765 | ls->delta_stat = 0; | |
766 | } | |
767 | } | |
768 | ||
769 | /* | |
770 | * Update delta_exec, delta_fair fields for rq. | |
771 | * | |
772 | * delta_fair clock advances at a rate inversely proportional to | |
773 | * total load (rq->ls.load.weight) on the runqueue, while | |
774 | * delta_exec advances at the same rate as wall-clock (provided | |
775 | * cpu is not idle). | |
776 | * | |
777 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
778 | * runqueue over any given interval. This (smoothened) load is used | |
779 | * during load balance. | |
780 | * | |
781 | * This function is called /before/ updating rq->ls.load | |
782 | * and when switching tasks. | |
783 | */ | |
784 | static void update_curr_load(struct rq *rq, u64 now) | |
785 | { | |
786 | struct load_stat *ls = &rq->ls; | |
787 | u64 start; | |
788 | ||
789 | start = ls->load_update_start; | |
790 | ls->load_update_start = now; | |
791 | ls->delta_stat += now - start; | |
792 | /* | |
793 | * Stagger updates to ls->delta_fair. Very frequent updates | |
794 | * can be expensive. | |
795 | */ | |
796 | if (ls->delta_stat >= sysctl_sched_stat_granularity) | |
797 | __update_curr_load(rq, ls); | |
798 | } | |
799 | ||
800 | static inline void | |
801 | inc_load(struct rq *rq, const struct task_struct *p, u64 now) | |
802 | { | |
803 | update_curr_load(rq, now); | |
804 | update_load_add(&rq->ls.load, p->se.load.weight); | |
805 | } | |
806 | ||
807 | static inline void | |
808 | dec_load(struct rq *rq, const struct task_struct *p, u64 now) | |
809 | { | |
810 | update_curr_load(rq, now); | |
811 | update_load_sub(&rq->ls.load, p->se.load.weight); | |
812 | } | |
813 | ||
814 | static void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now) | |
815 | { | |
816 | rq->nr_running++; | |
817 | inc_load(rq, p, now); | |
818 | } | |
819 | ||
820 | static void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now) | |
821 | { | |
822 | rq->nr_running--; | |
823 | dec_load(rq, p, now); | |
824 | } | |
825 | ||
45bf76df IM |
826 | static void set_load_weight(struct task_struct *p) |
827 | { | |
dd41f596 IM |
828 | task_rq(p)->cfs.wait_runtime -= p->se.wait_runtime; |
829 | p->se.wait_runtime = 0; | |
830 | ||
45bf76df | 831 | if (task_has_rt_policy(p)) { |
dd41f596 IM |
832 | p->se.load.weight = prio_to_weight[0] * 2; |
833 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
834 | return; | |
835 | } | |
45bf76df | 836 | |
dd41f596 IM |
837 | /* |
838 | * SCHED_IDLE tasks get minimal weight: | |
839 | */ | |
840 | if (p->policy == SCHED_IDLE) { | |
841 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
842 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
843 | return; | |
844 | } | |
71f8bd46 | 845 | |
dd41f596 IM |
846 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
847 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
848 | } |
849 | ||
dd41f596 IM |
850 | static void |
851 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now) | |
71f8bd46 | 852 | { |
dd41f596 IM |
853 | sched_info_queued(p); |
854 | p->sched_class->enqueue_task(rq, p, wakeup, now); | |
855 | p->se.on_rq = 1; | |
71f8bd46 IM |
856 | } |
857 | ||
dd41f596 IM |
858 | static void |
859 | dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now) | |
71f8bd46 | 860 | { |
dd41f596 IM |
861 | p->sched_class->dequeue_task(rq, p, sleep, now); |
862 | p->se.on_rq = 0; | |
71f8bd46 IM |
863 | } |
864 | ||
14531189 | 865 | /* |
dd41f596 | 866 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 867 | */ |
14531189 IM |
868 | static inline int __normal_prio(struct task_struct *p) |
869 | { | |
dd41f596 | 870 | return p->static_prio; |
14531189 IM |
871 | } |
872 | ||
b29739f9 IM |
873 | /* |
874 | * Calculate the expected normal priority: i.e. priority | |
875 | * without taking RT-inheritance into account. Might be | |
876 | * boosted by interactivity modifiers. Changes upon fork, | |
877 | * setprio syscalls, and whenever the interactivity | |
878 | * estimator recalculates. | |
879 | */ | |
36c8b586 | 880 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
881 | { |
882 | int prio; | |
883 | ||
e05606d3 | 884 | if (task_has_rt_policy(p)) |
b29739f9 IM |
885 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
886 | else | |
887 | prio = __normal_prio(p); | |
888 | return prio; | |
889 | } | |
890 | ||
891 | /* | |
892 | * Calculate the current priority, i.e. the priority | |
893 | * taken into account by the scheduler. This value might | |
894 | * be boosted by RT tasks, or might be boosted by | |
895 | * interactivity modifiers. Will be RT if the task got | |
896 | * RT-boosted. If not then it returns p->normal_prio. | |
897 | */ | |
36c8b586 | 898 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
899 | { |
900 | p->normal_prio = normal_prio(p); | |
901 | /* | |
902 | * If we are RT tasks or we were boosted to RT priority, | |
903 | * keep the priority unchanged. Otherwise, update priority | |
904 | * to the normal priority: | |
905 | */ | |
906 | if (!rt_prio(p->prio)) | |
907 | return p->normal_prio; | |
908 | return p->prio; | |
909 | } | |
910 | ||
1da177e4 | 911 | /* |
dd41f596 | 912 | * activate_task - move a task to the runqueue. |
1da177e4 | 913 | */ |
dd41f596 | 914 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 915 | { |
dd41f596 | 916 | u64 now = rq_clock(rq); |
d425b274 | 917 | |
dd41f596 IM |
918 | if (p->state == TASK_UNINTERRUPTIBLE) |
919 | rq->nr_uninterruptible--; | |
1da177e4 | 920 | |
dd41f596 IM |
921 | enqueue_task(rq, p, wakeup, now); |
922 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
923 | } |
924 | ||
925 | /* | |
dd41f596 | 926 | * activate_idle_task - move idle task to the _front_ of runqueue. |
1da177e4 | 927 | */ |
dd41f596 | 928 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 929 | { |
dd41f596 | 930 | u64 now = rq_clock(rq); |
1da177e4 | 931 | |
dd41f596 IM |
932 | if (p->state == TASK_UNINTERRUPTIBLE) |
933 | rq->nr_uninterruptible--; | |
ece8a684 | 934 | |
dd41f596 IM |
935 | enqueue_task(rq, p, 0, now); |
936 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
937 | } |
938 | ||
939 | /* | |
940 | * deactivate_task - remove a task from the runqueue. | |
941 | */ | |
8e717b19 IM |
942 | static void |
943 | deactivate_task(struct rq *rq, struct task_struct *p, int sleep, u64 now) | |
1da177e4 | 944 | { |
dd41f596 IM |
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; | |
6cfb0d5d IM |
984 | fair_clock_offset = old_rq->cfs.fair_clock - new_rq->cfs.fair_clock; |
985 | ||
dd41f596 IM |
986 | if (p->se.wait_start_fair) |
987 | p->se.wait_start_fair -= fair_clock_offset; | |
6cfb0d5d IM |
988 | if (p->se.sleep_start_fair) |
989 | p->se.sleep_start_fair -= fair_clock_offset; | |
990 | ||
991 | #ifdef CONFIG_SCHEDSTATS | |
992 | if (p->se.wait_start) | |
993 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
994 | if (p->se.sleep_start) |
995 | p->se.sleep_start -= clock_offset; | |
996 | if (p->se.block_start) | |
997 | p->se.block_start -= clock_offset; | |
6cfb0d5d | 998 | #endif |
dd41f596 IM |
999 | |
1000 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1001 | } |
1002 | ||
70b97a7f | 1003 | struct migration_req { |
1da177e4 | 1004 | struct list_head list; |
1da177e4 | 1005 | |
36c8b586 | 1006 | struct task_struct *task; |
1da177e4 LT |
1007 | int dest_cpu; |
1008 | ||
1da177e4 | 1009 | struct completion done; |
70b97a7f | 1010 | }; |
1da177e4 LT |
1011 | |
1012 | /* | |
1013 | * The task's runqueue lock must be held. | |
1014 | * Returns true if you have to wait for migration thread. | |
1015 | */ | |
36c8b586 | 1016 | static int |
70b97a7f | 1017 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1018 | { |
70b97a7f | 1019 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1020 | |
1021 | /* | |
1022 | * If the task is not on a runqueue (and not running), then | |
1023 | * it is sufficient to simply update the task's cpu field. | |
1024 | */ | |
dd41f596 | 1025 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1026 | set_task_cpu(p, dest_cpu); |
1027 | return 0; | |
1028 | } | |
1029 | ||
1030 | init_completion(&req->done); | |
1da177e4 LT |
1031 | req->task = p; |
1032 | req->dest_cpu = dest_cpu; | |
1033 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1034 | |
1da177e4 LT |
1035 | return 1; |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * wait_task_inactive - wait for a thread to unschedule. | |
1040 | * | |
1041 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1042 | * else this function might spin for a *long* time. This function can't | |
1043 | * be called with interrupts off, or it may introduce deadlock with | |
1044 | * smp_call_function() if an IPI is sent by the same process we are | |
1045 | * waiting to become inactive. | |
1046 | */ | |
36c8b586 | 1047 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1048 | { |
1049 | unsigned long flags; | |
dd41f596 | 1050 | int running, on_rq; |
70b97a7f | 1051 | struct rq *rq; |
1da177e4 LT |
1052 | |
1053 | repeat: | |
fa490cfd LT |
1054 | /* |
1055 | * We do the initial early heuristics without holding | |
1056 | * any task-queue locks at all. We'll only try to get | |
1057 | * the runqueue lock when things look like they will | |
1058 | * work out! | |
1059 | */ | |
1060 | rq = task_rq(p); | |
1061 | ||
1062 | /* | |
1063 | * If the task is actively running on another CPU | |
1064 | * still, just relax and busy-wait without holding | |
1065 | * any locks. | |
1066 | * | |
1067 | * NOTE! Since we don't hold any locks, it's not | |
1068 | * even sure that "rq" stays as the right runqueue! | |
1069 | * But we don't care, since "task_running()" will | |
1070 | * return false if the runqueue has changed and p | |
1071 | * is actually now running somewhere else! | |
1072 | */ | |
1073 | while (task_running(rq, p)) | |
1074 | cpu_relax(); | |
1075 | ||
1076 | /* | |
1077 | * Ok, time to look more closely! We need the rq | |
1078 | * lock now, to be *sure*. If we're wrong, we'll | |
1079 | * just go back and repeat. | |
1080 | */ | |
1da177e4 | 1081 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1082 | running = task_running(rq, p); |
dd41f596 | 1083 | on_rq = p->se.on_rq; |
fa490cfd LT |
1084 | task_rq_unlock(rq, &flags); |
1085 | ||
1086 | /* | |
1087 | * Was it really running after all now that we | |
1088 | * checked with the proper locks actually held? | |
1089 | * | |
1090 | * Oops. Go back and try again.. | |
1091 | */ | |
1092 | if (unlikely(running)) { | |
1da177e4 | 1093 | cpu_relax(); |
1da177e4 LT |
1094 | goto repeat; |
1095 | } | |
fa490cfd LT |
1096 | |
1097 | /* | |
1098 | * It's not enough that it's not actively running, | |
1099 | * it must be off the runqueue _entirely_, and not | |
1100 | * preempted! | |
1101 | * | |
1102 | * So if it wa still runnable (but just not actively | |
1103 | * running right now), it's preempted, and we should | |
1104 | * yield - it could be a while. | |
1105 | */ | |
dd41f596 | 1106 | if (unlikely(on_rq)) { |
fa490cfd LT |
1107 | yield(); |
1108 | goto repeat; | |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * Ahh, all good. It wasn't running, and it wasn't | |
1113 | * runnable, which means that it will never become | |
1114 | * running in the future either. We're all done! | |
1115 | */ | |
1da177e4 LT |
1116 | } |
1117 | ||
1118 | /*** | |
1119 | * kick_process - kick a running thread to enter/exit the kernel | |
1120 | * @p: the to-be-kicked thread | |
1121 | * | |
1122 | * Cause a process which is running on another CPU to enter | |
1123 | * kernel-mode, without any delay. (to get signals handled.) | |
1124 | * | |
1125 | * NOTE: this function doesnt have to take the runqueue lock, | |
1126 | * because all it wants to ensure is that the remote task enters | |
1127 | * the kernel. If the IPI races and the task has been migrated | |
1128 | * to another CPU then no harm is done and the purpose has been | |
1129 | * achieved as well. | |
1130 | */ | |
36c8b586 | 1131 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1132 | { |
1133 | int cpu; | |
1134 | ||
1135 | preempt_disable(); | |
1136 | cpu = task_cpu(p); | |
1137 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1138 | smp_send_reschedule(cpu); | |
1139 | preempt_enable(); | |
1140 | } | |
1141 | ||
1142 | /* | |
2dd73a4f PW |
1143 | * Return a low guess at the load of a migration-source cpu weighted |
1144 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1145 | * |
1146 | * We want to under-estimate the load of migration sources, to | |
1147 | * balance conservatively. | |
1148 | */ | |
a2000572 | 1149 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1150 | { |
70b97a7f | 1151 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1152 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1153 | |
3b0bd9bc | 1154 | if (type == 0) |
dd41f596 | 1155 | return total; |
b910472d | 1156 | |
dd41f596 | 1157 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1158 | } |
1159 | ||
1160 | /* | |
2dd73a4f PW |
1161 | * Return a high guess at the load of a migration-target cpu weighted |
1162 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1163 | */ |
a2000572 | 1164 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1165 | { |
70b97a7f | 1166 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1167 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1168 | |
7897986b | 1169 | if (type == 0) |
dd41f596 | 1170 | return total; |
3b0bd9bc | 1171 | |
dd41f596 | 1172 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1173 | } |
1174 | ||
1175 | /* | |
1176 | * Return the average load per task on the cpu's run queue | |
1177 | */ | |
1178 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1179 | { | |
70b97a7f | 1180 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1181 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1182 | unsigned long n = rq->nr_running; |
1183 | ||
dd41f596 | 1184 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1185 | } |
1186 | ||
147cbb4b NP |
1187 | /* |
1188 | * find_idlest_group finds and returns the least busy CPU group within the | |
1189 | * domain. | |
1190 | */ | |
1191 | static struct sched_group * | |
1192 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1193 | { | |
1194 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1195 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1196 | int load_idx = sd->forkexec_idx; | |
1197 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1198 | ||
1199 | do { | |
1200 | unsigned long load, avg_load; | |
1201 | int local_group; | |
1202 | int i; | |
1203 | ||
da5a5522 BD |
1204 | /* Skip over this group if it has no CPUs allowed */ |
1205 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1206 | goto nextgroup; | |
1207 | ||
147cbb4b | 1208 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1209 | |
1210 | /* Tally up the load of all CPUs in the group */ | |
1211 | avg_load = 0; | |
1212 | ||
1213 | for_each_cpu_mask(i, group->cpumask) { | |
1214 | /* Bias balancing toward cpus of our domain */ | |
1215 | if (local_group) | |
1216 | load = source_load(i, load_idx); | |
1217 | else | |
1218 | load = target_load(i, load_idx); | |
1219 | ||
1220 | avg_load += load; | |
1221 | } | |
1222 | ||
1223 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1224 | avg_load = sg_div_cpu_power(group, |
1225 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1226 | |
1227 | if (local_group) { | |
1228 | this_load = avg_load; | |
1229 | this = group; | |
1230 | } else if (avg_load < min_load) { | |
1231 | min_load = avg_load; | |
1232 | idlest = group; | |
1233 | } | |
da5a5522 | 1234 | nextgroup: |
147cbb4b NP |
1235 | group = group->next; |
1236 | } while (group != sd->groups); | |
1237 | ||
1238 | if (!idlest || 100*this_load < imbalance*min_load) | |
1239 | return NULL; | |
1240 | return idlest; | |
1241 | } | |
1242 | ||
1243 | /* | |
0feaece9 | 1244 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1245 | */ |
95cdf3b7 IM |
1246 | static int |
1247 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1248 | { |
da5a5522 | 1249 | cpumask_t tmp; |
147cbb4b NP |
1250 | unsigned long load, min_load = ULONG_MAX; |
1251 | int idlest = -1; | |
1252 | int i; | |
1253 | ||
da5a5522 BD |
1254 | /* Traverse only the allowed CPUs */ |
1255 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1256 | ||
1257 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1258 | load = weighted_cpuload(i); |
147cbb4b NP |
1259 | |
1260 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1261 | min_load = load; | |
1262 | idlest = i; | |
1263 | } | |
1264 | } | |
1265 | ||
1266 | return idlest; | |
1267 | } | |
1268 | ||
476d139c NP |
1269 | /* |
1270 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1271 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1272 | * SD_BALANCE_EXEC. | |
1273 | * | |
1274 | * Balance, ie. select the least loaded group. | |
1275 | * | |
1276 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1277 | * | |
1278 | * preempt must be disabled. | |
1279 | */ | |
1280 | static int sched_balance_self(int cpu, int flag) | |
1281 | { | |
1282 | struct task_struct *t = current; | |
1283 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1284 | |
c96d145e | 1285 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1286 | /* |
1287 | * If power savings logic is enabled for a domain, stop there. | |
1288 | */ | |
5c45bf27 SS |
1289 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1290 | break; | |
476d139c NP |
1291 | if (tmp->flags & flag) |
1292 | sd = tmp; | |
c96d145e | 1293 | } |
476d139c NP |
1294 | |
1295 | while (sd) { | |
1296 | cpumask_t span; | |
1297 | struct sched_group *group; | |
1a848870 SS |
1298 | int new_cpu, weight; |
1299 | ||
1300 | if (!(sd->flags & flag)) { | |
1301 | sd = sd->child; | |
1302 | continue; | |
1303 | } | |
476d139c NP |
1304 | |
1305 | span = sd->span; | |
1306 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1307 | if (!group) { |
1308 | sd = sd->child; | |
1309 | continue; | |
1310 | } | |
476d139c | 1311 | |
da5a5522 | 1312 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1313 | if (new_cpu == -1 || new_cpu == cpu) { |
1314 | /* Now try balancing at a lower domain level of cpu */ | |
1315 | sd = sd->child; | |
1316 | continue; | |
1317 | } | |
476d139c | 1318 | |
1a848870 | 1319 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1320 | cpu = new_cpu; |
476d139c NP |
1321 | sd = NULL; |
1322 | weight = cpus_weight(span); | |
1323 | for_each_domain(cpu, tmp) { | |
1324 | if (weight <= cpus_weight(tmp->span)) | |
1325 | break; | |
1326 | if (tmp->flags & flag) | |
1327 | sd = tmp; | |
1328 | } | |
1329 | /* while loop will break here if sd == NULL */ | |
1330 | } | |
1331 | ||
1332 | return cpu; | |
1333 | } | |
1334 | ||
1335 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1336 | |
1337 | /* | |
1338 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1339 | * not idle and an idle cpu is available. The span of cpus to | |
1340 | * search starts with cpus closest then further out as needed, | |
1341 | * so we always favor a closer, idle cpu. | |
1342 | * | |
1343 | * Returns the CPU we should wake onto. | |
1344 | */ | |
1345 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1346 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1347 | { |
1348 | cpumask_t tmp; | |
1349 | struct sched_domain *sd; | |
1350 | int i; | |
1351 | ||
4953198b SS |
1352 | /* |
1353 | * If it is idle, then it is the best cpu to run this task. | |
1354 | * | |
1355 | * This cpu is also the best, if it has more than one task already. | |
1356 | * Siblings must be also busy(in most cases) as they didn't already | |
1357 | * pickup the extra load from this cpu and hence we need not check | |
1358 | * sibling runqueue info. This will avoid the checks and cache miss | |
1359 | * penalities associated with that. | |
1360 | */ | |
1361 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1362 | return cpu; |
1363 | ||
1364 | for_each_domain(cpu, sd) { | |
1365 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1366 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1367 | for_each_cpu_mask(i, tmp) { |
1368 | if (idle_cpu(i)) | |
1369 | return i; | |
1370 | } | |
9761eea8 | 1371 | } else { |
e0f364f4 | 1372 | break; |
9761eea8 | 1373 | } |
1da177e4 LT |
1374 | } |
1375 | return cpu; | |
1376 | } | |
1377 | #else | |
36c8b586 | 1378 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1379 | { |
1380 | return cpu; | |
1381 | } | |
1382 | #endif | |
1383 | ||
1384 | /*** | |
1385 | * try_to_wake_up - wake up a thread | |
1386 | * @p: the to-be-woken-up thread | |
1387 | * @state: the mask of task states that can be woken | |
1388 | * @sync: do a synchronous wakeup? | |
1389 | * | |
1390 | * Put it on the run-queue if it's not already there. The "current" | |
1391 | * thread is always on the run-queue (except when the actual | |
1392 | * re-schedule is in progress), and as such you're allowed to do | |
1393 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1394 | * runnable without the overhead of this. | |
1395 | * | |
1396 | * returns failure only if the task is already active. | |
1397 | */ | |
36c8b586 | 1398 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1399 | { |
1400 | int cpu, this_cpu, success = 0; | |
1401 | unsigned long flags; | |
1402 | long old_state; | |
70b97a7f | 1403 | struct rq *rq; |
1da177e4 | 1404 | #ifdef CONFIG_SMP |
7897986b | 1405 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1406 | unsigned long load, this_load; |
1da177e4 LT |
1407 | int new_cpu; |
1408 | #endif | |
1409 | ||
1410 | rq = task_rq_lock(p, &flags); | |
1411 | old_state = p->state; | |
1412 | if (!(old_state & state)) | |
1413 | goto out; | |
1414 | ||
dd41f596 | 1415 | if (p->se.on_rq) |
1da177e4 LT |
1416 | goto out_running; |
1417 | ||
1418 | cpu = task_cpu(p); | |
1419 | this_cpu = smp_processor_id(); | |
1420 | ||
1421 | #ifdef CONFIG_SMP | |
1422 | if (unlikely(task_running(rq, p))) | |
1423 | goto out_activate; | |
1424 | ||
7897986b NP |
1425 | new_cpu = cpu; |
1426 | ||
1da177e4 LT |
1427 | schedstat_inc(rq, ttwu_cnt); |
1428 | if (cpu == this_cpu) { | |
1429 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1430 | goto out_set_cpu; |
1431 | } | |
1432 | ||
1433 | for_each_domain(this_cpu, sd) { | |
1434 | if (cpu_isset(cpu, sd->span)) { | |
1435 | schedstat_inc(sd, ttwu_wake_remote); | |
1436 | this_sd = sd; | |
1437 | break; | |
1da177e4 LT |
1438 | } |
1439 | } | |
1da177e4 | 1440 | |
7897986b | 1441 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1442 | goto out_set_cpu; |
1443 | ||
1da177e4 | 1444 | /* |
7897986b | 1445 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1446 | */ |
7897986b NP |
1447 | if (this_sd) { |
1448 | int idx = this_sd->wake_idx; | |
1449 | unsigned int imbalance; | |
1da177e4 | 1450 | |
a3f21bce NP |
1451 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1452 | ||
7897986b NP |
1453 | load = source_load(cpu, idx); |
1454 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1455 | |
7897986b NP |
1456 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1457 | ||
a3f21bce NP |
1458 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1459 | unsigned long tl = this_load; | |
33859f7f MOS |
1460 | unsigned long tl_per_task; |
1461 | ||
1462 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1463 | |
1da177e4 | 1464 | /* |
a3f21bce NP |
1465 | * If sync wakeup then subtract the (maximum possible) |
1466 | * effect of the currently running task from the load | |
1467 | * of the current CPU: | |
1da177e4 | 1468 | */ |
a3f21bce | 1469 | if (sync) |
dd41f596 | 1470 | tl -= current->se.load.weight; |
a3f21bce NP |
1471 | |
1472 | if ((tl <= load && | |
2dd73a4f | 1473 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1474 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1475 | /* |
1476 | * This domain has SD_WAKE_AFFINE and | |
1477 | * p is cache cold in this domain, and | |
1478 | * there is no bad imbalance. | |
1479 | */ | |
1480 | schedstat_inc(this_sd, ttwu_move_affine); | |
1481 | goto out_set_cpu; | |
1482 | } | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * Start passive balancing when half the imbalance_pct | |
1487 | * limit is reached. | |
1488 | */ | |
1489 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1490 | if (imbalance*this_load <= 100*load) { | |
1491 | schedstat_inc(this_sd, ttwu_move_balance); | |
1492 | goto out_set_cpu; | |
1493 | } | |
1da177e4 LT |
1494 | } |
1495 | } | |
1496 | ||
1497 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1498 | out_set_cpu: | |
1499 | new_cpu = wake_idle(new_cpu, p); | |
1500 | if (new_cpu != cpu) { | |
1501 | set_task_cpu(p, new_cpu); | |
1502 | task_rq_unlock(rq, &flags); | |
1503 | /* might preempt at this point */ | |
1504 | rq = task_rq_lock(p, &flags); | |
1505 | old_state = p->state; | |
1506 | if (!(old_state & state)) | |
1507 | goto out; | |
dd41f596 | 1508 | if (p->se.on_rq) |
1da177e4 LT |
1509 | goto out_running; |
1510 | ||
1511 | this_cpu = smp_processor_id(); | |
1512 | cpu = task_cpu(p); | |
1513 | } | |
1514 | ||
1515 | out_activate: | |
1516 | #endif /* CONFIG_SMP */ | |
dd41f596 | 1517 | activate_task(rq, p, 1); |
1da177e4 LT |
1518 | /* |
1519 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1520 | * has indicated that it will leave the CPU in short order) | |
1521 | * don't trigger a preemption, if the woken up task will run on | |
1522 | * this cpu. (in this case the 'I will reschedule' promise of | |
1523 | * the waker guarantees that the freshly woken up task is going | |
1524 | * to be considered on this CPU.) | |
1525 | */ | |
dd41f596 IM |
1526 | if (!sync || cpu != this_cpu) |
1527 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1528 | success = 1; |
1529 | ||
1530 | out_running: | |
1531 | p->state = TASK_RUNNING; | |
1532 | out: | |
1533 | task_rq_unlock(rq, &flags); | |
1534 | ||
1535 | return success; | |
1536 | } | |
1537 | ||
36c8b586 | 1538 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1539 | { |
1540 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1541 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1542 | } | |
1da177e4 LT |
1543 | EXPORT_SYMBOL(wake_up_process); |
1544 | ||
36c8b586 | 1545 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1546 | { |
1547 | return try_to_wake_up(p, state, 0); | |
1548 | } | |
1549 | ||
1da177e4 LT |
1550 | /* |
1551 | * Perform scheduler related setup for a newly forked process p. | |
1552 | * p is forked by current. | |
dd41f596 IM |
1553 | * |
1554 | * __sched_fork() is basic setup used by init_idle() too: | |
1555 | */ | |
1556 | static void __sched_fork(struct task_struct *p) | |
1557 | { | |
1558 | p->se.wait_start_fair = 0; | |
dd41f596 IM |
1559 | p->se.exec_start = 0; |
1560 | p->se.sum_exec_runtime = 0; | |
1561 | p->se.delta_exec = 0; | |
1562 | p->se.delta_fair_run = 0; | |
1563 | p->se.delta_fair_sleep = 0; | |
1564 | p->se.wait_runtime = 0; | |
6cfb0d5d IM |
1565 | p->se.sleep_start_fair = 0; |
1566 | ||
1567 | #ifdef CONFIG_SCHEDSTATS | |
1568 | p->se.wait_start = 0; | |
dd41f596 IM |
1569 | p->se.sum_wait_runtime = 0; |
1570 | p->se.sum_sleep_runtime = 0; | |
1571 | p->se.sleep_start = 0; | |
dd41f596 IM |
1572 | p->se.block_start = 0; |
1573 | p->se.sleep_max = 0; | |
1574 | p->se.block_max = 0; | |
1575 | p->se.exec_max = 0; | |
1576 | p->se.wait_max = 0; | |
1577 | p->se.wait_runtime_overruns = 0; | |
1578 | p->se.wait_runtime_underruns = 0; | |
6cfb0d5d | 1579 | #endif |
476d139c | 1580 | |
dd41f596 IM |
1581 | INIT_LIST_HEAD(&p->run_list); |
1582 | p->se.on_rq = 0; | |
476d139c | 1583 | |
e107be36 AK |
1584 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1585 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1586 | #endif | |
1587 | ||
1da177e4 LT |
1588 | /* |
1589 | * We mark the process as running here, but have not actually | |
1590 | * inserted it onto the runqueue yet. This guarantees that | |
1591 | * nobody will actually run it, and a signal or other external | |
1592 | * event cannot wake it up and insert it on the runqueue either. | |
1593 | */ | |
1594 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1595 | } |
1596 | ||
1597 | /* | |
1598 | * fork()/clone()-time setup: | |
1599 | */ | |
1600 | void sched_fork(struct task_struct *p, int clone_flags) | |
1601 | { | |
1602 | int cpu = get_cpu(); | |
1603 | ||
1604 | __sched_fork(p); | |
1605 | ||
1606 | #ifdef CONFIG_SMP | |
1607 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1608 | #endif | |
1609 | __set_task_cpu(p, cpu); | |
b29739f9 IM |
1610 | |
1611 | /* | |
1612 | * Make sure we do not leak PI boosting priority to the child: | |
1613 | */ | |
1614 | p->prio = current->normal_prio; | |
1615 | ||
52f17b6c | 1616 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1617 | if (likely(sched_info_on())) |
52f17b6c | 1618 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1619 | #endif |
d6077cb8 | 1620 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1621 | p->oncpu = 0; |
1622 | #endif | |
1da177e4 | 1623 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1624 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1625 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1626 | #endif |
476d139c | 1627 | put_cpu(); |
1da177e4 LT |
1628 | } |
1629 | ||
dd41f596 IM |
1630 | /* |
1631 | * After fork, child runs first. (default) If set to 0 then | |
1632 | * parent will (try to) run first. | |
1633 | */ | |
1634 | unsigned int __read_mostly sysctl_sched_child_runs_first = 1; | |
1635 | ||
1da177e4 LT |
1636 | /* |
1637 | * wake_up_new_task - wake up a newly created task for the first time. | |
1638 | * | |
1639 | * This function will do some initial scheduler statistics housekeeping | |
1640 | * that must be done for every newly created context, then puts the task | |
1641 | * on the runqueue and wakes it. | |
1642 | */ | |
36c8b586 | 1643 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1644 | { |
1645 | unsigned long flags; | |
dd41f596 IM |
1646 | struct rq *rq; |
1647 | int this_cpu; | |
cad60d93 | 1648 | u64 now; |
1da177e4 LT |
1649 | |
1650 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1651 | BUG_ON(p->state != TASK_RUNNING); |
dd41f596 | 1652 | this_cpu = smp_processor_id(); /* parent's CPU */ |
cad60d93 | 1653 | now = rq_clock(rq); |
1da177e4 LT |
1654 | |
1655 | p->prio = effective_prio(p); | |
1656 | ||
cad60d93 IM |
1657 | if (!p->sched_class->task_new || !sysctl_sched_child_runs_first || |
1658 | (clone_flags & CLONE_VM) || task_cpu(p) != this_cpu || | |
1659 | !current->se.on_rq) { | |
1660 | ||
dd41f596 | 1661 | activate_task(rq, p, 0); |
1da177e4 | 1662 | } else { |
1da177e4 | 1663 | /* |
dd41f596 IM |
1664 | * Let the scheduling class do new task startup |
1665 | * management (if any): | |
1da177e4 | 1666 | */ |
cad60d93 IM |
1667 | p->sched_class->task_new(rq, p, now); |
1668 | inc_nr_running(p, rq, now); | |
1da177e4 | 1669 | } |
dd41f596 IM |
1670 | check_preempt_curr(rq, p); |
1671 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1672 | } |
1673 | ||
e107be36 AK |
1674 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1675 | ||
1676 | /** | |
421cee29 RD |
1677 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1678 | * @notifier: notifier struct to register | |
e107be36 AK |
1679 | */ |
1680 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1681 | { | |
1682 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1683 | } | |
1684 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1685 | ||
1686 | /** | |
1687 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1688 | * @notifier: notifier struct to unregister |
e107be36 AK |
1689 | * |
1690 | * This is safe to call from within a preemption notifier. | |
1691 | */ | |
1692 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1693 | { | |
1694 | hlist_del(¬ifier->link); | |
1695 | } | |
1696 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1697 | ||
1698 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1699 | { | |
1700 | struct preempt_notifier *notifier; | |
1701 | struct hlist_node *node; | |
1702 | ||
1703 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1704 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1705 | } | |
1706 | ||
1707 | static void | |
1708 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1709 | struct task_struct *next) | |
1710 | { | |
1711 | struct preempt_notifier *notifier; | |
1712 | struct hlist_node *node; | |
1713 | ||
1714 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1715 | notifier->ops->sched_out(notifier, next); | |
1716 | } | |
1717 | ||
1718 | #else | |
1719 | ||
1720 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1721 | { | |
1722 | } | |
1723 | ||
1724 | static void | |
1725 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1726 | struct task_struct *next) | |
1727 | { | |
1728 | } | |
1729 | ||
1730 | #endif | |
1731 | ||
4866cde0 NP |
1732 | /** |
1733 | * prepare_task_switch - prepare to switch tasks | |
1734 | * @rq: the runqueue preparing to switch | |
421cee29 | 1735 | * @prev: the current task that is being switched out |
4866cde0 NP |
1736 | * @next: the task we are going to switch to. |
1737 | * | |
1738 | * This is called with the rq lock held and interrupts off. It must | |
1739 | * be paired with a subsequent finish_task_switch after the context | |
1740 | * switch. | |
1741 | * | |
1742 | * prepare_task_switch sets up locking and calls architecture specific | |
1743 | * hooks. | |
1744 | */ | |
e107be36 AK |
1745 | static inline void |
1746 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1747 | struct task_struct *next) | |
4866cde0 | 1748 | { |
e107be36 | 1749 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1750 | prepare_lock_switch(rq, next); |
1751 | prepare_arch_switch(next); | |
1752 | } | |
1753 | ||
1da177e4 LT |
1754 | /** |
1755 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1756 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1757 | * @prev: the thread we just switched away from. |
1758 | * | |
4866cde0 NP |
1759 | * finish_task_switch must be called after the context switch, paired |
1760 | * with a prepare_task_switch call before the context switch. | |
1761 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1762 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1763 | * |
1764 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1765 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1766 | * with the lock held can cause deadlocks; see schedule() for | |
1767 | * details.) | |
1768 | */ | |
70b97a7f | 1769 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1770 | __releases(rq->lock) |
1771 | { | |
1da177e4 | 1772 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1773 | long prev_state; |
1da177e4 LT |
1774 | |
1775 | rq->prev_mm = NULL; | |
1776 | ||
1777 | /* | |
1778 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1779 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1780 | * schedule one last time. The schedule call will never return, and |
1781 | * the scheduled task must drop that reference. | |
c394cc9f | 1782 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1783 | * still held, otherwise prev could be scheduled on another cpu, die |
1784 | * there before we look at prev->state, and then the reference would | |
1785 | * be dropped twice. | |
1786 | * Manfred Spraul <manfred@colorfullife.com> | |
1787 | */ | |
55a101f8 | 1788 | prev_state = prev->state; |
4866cde0 NP |
1789 | finish_arch_switch(prev); |
1790 | finish_lock_switch(rq, prev); | |
e107be36 | 1791 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1792 | if (mm) |
1793 | mmdrop(mm); | |
c394cc9f | 1794 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1795 | /* |
1796 | * Remove function-return probe instances associated with this | |
1797 | * task and put them back on the free list. | |
9761eea8 | 1798 | */ |
c6fd91f0 | 1799 | kprobe_flush_task(prev); |
1da177e4 | 1800 | put_task_struct(prev); |
c6fd91f0 | 1801 | } |
1da177e4 LT |
1802 | } |
1803 | ||
1804 | /** | |
1805 | * schedule_tail - first thing a freshly forked thread must call. | |
1806 | * @prev: the thread we just switched away from. | |
1807 | */ | |
36c8b586 | 1808 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1809 | __releases(rq->lock) |
1810 | { | |
70b97a7f IM |
1811 | struct rq *rq = this_rq(); |
1812 | ||
4866cde0 NP |
1813 | finish_task_switch(rq, prev); |
1814 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1815 | /* In this case, finish_task_switch does not reenable preemption */ | |
1816 | preempt_enable(); | |
1817 | #endif | |
1da177e4 LT |
1818 | if (current->set_child_tid) |
1819 | put_user(current->pid, current->set_child_tid); | |
1820 | } | |
1821 | ||
1822 | /* | |
1823 | * context_switch - switch to the new MM and the new | |
1824 | * thread's register state. | |
1825 | */ | |
dd41f596 | 1826 | static inline void |
70b97a7f | 1827 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1828 | struct task_struct *next) |
1da177e4 | 1829 | { |
dd41f596 | 1830 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1831 | |
e107be36 | 1832 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1833 | mm = next->mm; |
1834 | oldmm = prev->active_mm; | |
9226d125 ZA |
1835 | /* |
1836 | * For paravirt, this is coupled with an exit in switch_to to | |
1837 | * combine the page table reload and the switch backend into | |
1838 | * one hypercall. | |
1839 | */ | |
1840 | arch_enter_lazy_cpu_mode(); | |
1841 | ||
dd41f596 | 1842 | if (unlikely(!mm)) { |
1da177e4 LT |
1843 | next->active_mm = oldmm; |
1844 | atomic_inc(&oldmm->mm_count); | |
1845 | enter_lazy_tlb(oldmm, next); | |
1846 | } else | |
1847 | switch_mm(oldmm, mm, next); | |
1848 | ||
dd41f596 | 1849 | if (unlikely(!prev->mm)) { |
1da177e4 | 1850 | prev->active_mm = NULL; |
1da177e4 LT |
1851 | rq->prev_mm = oldmm; |
1852 | } | |
3a5f5e48 IM |
1853 | /* |
1854 | * Since the runqueue lock will be released by the next | |
1855 | * task (which is an invalid locking op but in the case | |
1856 | * of the scheduler it's an obvious special-case), so we | |
1857 | * do an early lockdep release here: | |
1858 | */ | |
1859 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1860 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1861 | #endif |
1da177e4 LT |
1862 | |
1863 | /* Here we just switch the register state and the stack. */ | |
1864 | switch_to(prev, next, prev); | |
1865 | ||
dd41f596 IM |
1866 | barrier(); |
1867 | /* | |
1868 | * this_rq must be evaluated again because prev may have moved | |
1869 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1870 | * frame will be invalid. | |
1871 | */ | |
1872 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1873 | } |
1874 | ||
1875 | /* | |
1876 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1877 | * | |
1878 | * externally visible scheduler statistics: current number of runnable | |
1879 | * threads, current number of uninterruptible-sleeping threads, total | |
1880 | * number of context switches performed since bootup. | |
1881 | */ | |
1882 | unsigned long nr_running(void) | |
1883 | { | |
1884 | unsigned long i, sum = 0; | |
1885 | ||
1886 | for_each_online_cpu(i) | |
1887 | sum += cpu_rq(i)->nr_running; | |
1888 | ||
1889 | return sum; | |
1890 | } | |
1891 | ||
1892 | unsigned long nr_uninterruptible(void) | |
1893 | { | |
1894 | unsigned long i, sum = 0; | |
1895 | ||
0a945022 | 1896 | for_each_possible_cpu(i) |
1da177e4 LT |
1897 | sum += cpu_rq(i)->nr_uninterruptible; |
1898 | ||
1899 | /* | |
1900 | * Since we read the counters lockless, it might be slightly | |
1901 | * inaccurate. Do not allow it to go below zero though: | |
1902 | */ | |
1903 | if (unlikely((long)sum < 0)) | |
1904 | sum = 0; | |
1905 | ||
1906 | return sum; | |
1907 | } | |
1908 | ||
1909 | unsigned long long nr_context_switches(void) | |
1910 | { | |
cc94abfc SR |
1911 | int i; |
1912 | unsigned long long sum = 0; | |
1da177e4 | 1913 | |
0a945022 | 1914 | for_each_possible_cpu(i) |
1da177e4 LT |
1915 | sum += cpu_rq(i)->nr_switches; |
1916 | ||
1917 | return sum; | |
1918 | } | |
1919 | ||
1920 | unsigned long nr_iowait(void) | |
1921 | { | |
1922 | unsigned long i, sum = 0; | |
1923 | ||
0a945022 | 1924 | for_each_possible_cpu(i) |
1da177e4 LT |
1925 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1926 | ||
1927 | return sum; | |
1928 | } | |
1929 | ||
db1b1fef JS |
1930 | unsigned long nr_active(void) |
1931 | { | |
1932 | unsigned long i, running = 0, uninterruptible = 0; | |
1933 | ||
1934 | for_each_online_cpu(i) { | |
1935 | running += cpu_rq(i)->nr_running; | |
1936 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1937 | } | |
1938 | ||
1939 | if (unlikely((long)uninterruptible < 0)) | |
1940 | uninterruptible = 0; | |
1941 | ||
1942 | return running + uninterruptible; | |
1943 | } | |
1944 | ||
48f24c4d | 1945 | /* |
dd41f596 IM |
1946 | * Update rq->cpu_load[] statistics. This function is usually called every |
1947 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1948 | */ |
dd41f596 | 1949 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1950 | { |
dd41f596 IM |
1951 | u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64; |
1952 | unsigned long total_load = this_rq->ls.load.weight; | |
1953 | unsigned long this_load = total_load; | |
1954 | struct load_stat *ls = &this_rq->ls; | |
1955 | u64 now = __rq_clock(this_rq); | |
1956 | int i, scale; | |
1957 | ||
1958 | this_rq->nr_load_updates++; | |
1959 | if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD))) | |
1960 | goto do_avg; | |
1961 | ||
1962 | /* Update delta_fair/delta_exec fields first */ | |
1963 | update_curr_load(this_rq, now); | |
1964 | ||
1965 | fair_delta64 = ls->delta_fair + 1; | |
1966 | ls->delta_fair = 0; | |
1967 | ||
1968 | exec_delta64 = ls->delta_exec + 1; | |
1969 | ls->delta_exec = 0; | |
1970 | ||
1971 | sample_interval64 = now - ls->load_update_last; | |
1972 | ls->load_update_last = now; | |
1973 | ||
1974 | if ((s64)sample_interval64 < (s64)TICK_NSEC) | |
1975 | sample_interval64 = TICK_NSEC; | |
1976 | ||
1977 | if (exec_delta64 > sample_interval64) | |
1978 | exec_delta64 = sample_interval64; | |
1979 | ||
1980 | idle_delta64 = sample_interval64 - exec_delta64; | |
1981 | ||
1982 | tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64); | |
1983 | tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64); | |
1984 | ||
1985 | this_load = (unsigned long)tmp64; | |
1986 | ||
1987 | do_avg: | |
1988 | ||
1989 | /* Update our load: */ | |
1990 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1991 | unsigned long old_load, new_load; | |
1992 | ||
1993 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1994 | ||
1995 | old_load = this_rq->cpu_load[i]; | |
1996 | new_load = this_load; | |
1997 | ||
1998 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; | |
1999 | } | |
48f24c4d IM |
2000 | } |
2001 | ||
dd41f596 IM |
2002 | #ifdef CONFIG_SMP |
2003 | ||
1da177e4 LT |
2004 | /* |
2005 | * double_rq_lock - safely lock two runqueues | |
2006 | * | |
2007 | * Note this does not disable interrupts like task_rq_lock, | |
2008 | * you need to do so manually before calling. | |
2009 | */ | |
70b97a7f | 2010 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2011 | __acquires(rq1->lock) |
2012 | __acquires(rq2->lock) | |
2013 | { | |
054b9108 | 2014 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2015 | if (rq1 == rq2) { |
2016 | spin_lock(&rq1->lock); | |
2017 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2018 | } else { | |
c96d145e | 2019 | if (rq1 < rq2) { |
1da177e4 LT |
2020 | spin_lock(&rq1->lock); |
2021 | spin_lock(&rq2->lock); | |
2022 | } else { | |
2023 | spin_lock(&rq2->lock); | |
2024 | spin_lock(&rq1->lock); | |
2025 | } | |
2026 | } | |
2027 | } | |
2028 | ||
2029 | /* | |
2030 | * double_rq_unlock - safely unlock two runqueues | |
2031 | * | |
2032 | * Note this does not restore interrupts like task_rq_unlock, | |
2033 | * you need to do so manually after calling. | |
2034 | */ | |
70b97a7f | 2035 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2036 | __releases(rq1->lock) |
2037 | __releases(rq2->lock) | |
2038 | { | |
2039 | spin_unlock(&rq1->lock); | |
2040 | if (rq1 != rq2) | |
2041 | spin_unlock(&rq2->lock); | |
2042 | else | |
2043 | __release(rq2->lock); | |
2044 | } | |
2045 | ||
2046 | /* | |
2047 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2048 | */ | |
70b97a7f | 2049 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2050 | __releases(this_rq->lock) |
2051 | __acquires(busiest->lock) | |
2052 | __acquires(this_rq->lock) | |
2053 | { | |
054b9108 KK |
2054 | if (unlikely(!irqs_disabled())) { |
2055 | /* printk() doesn't work good under rq->lock */ | |
2056 | spin_unlock(&this_rq->lock); | |
2057 | BUG_ON(1); | |
2058 | } | |
1da177e4 | 2059 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2060 | if (busiest < this_rq) { |
1da177e4 LT |
2061 | spin_unlock(&this_rq->lock); |
2062 | spin_lock(&busiest->lock); | |
2063 | spin_lock(&this_rq->lock); | |
2064 | } else | |
2065 | spin_lock(&busiest->lock); | |
2066 | } | |
2067 | } | |
2068 | ||
1da177e4 LT |
2069 | /* |
2070 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2071 | * This is accomplished by forcing the cpu_allowed mask to only | |
2072 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2073 | * the cpu_allowed mask is restored. | |
2074 | */ | |
36c8b586 | 2075 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2076 | { |
70b97a7f | 2077 | struct migration_req req; |
1da177e4 | 2078 | unsigned long flags; |
70b97a7f | 2079 | struct rq *rq; |
1da177e4 LT |
2080 | |
2081 | rq = task_rq_lock(p, &flags); | |
2082 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2083 | || unlikely(cpu_is_offline(dest_cpu))) | |
2084 | goto out; | |
2085 | ||
2086 | /* force the process onto the specified CPU */ | |
2087 | if (migrate_task(p, dest_cpu, &req)) { | |
2088 | /* Need to wait for migration thread (might exit: take ref). */ | |
2089 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2090 | |
1da177e4 LT |
2091 | get_task_struct(mt); |
2092 | task_rq_unlock(rq, &flags); | |
2093 | wake_up_process(mt); | |
2094 | put_task_struct(mt); | |
2095 | wait_for_completion(&req.done); | |
36c8b586 | 2096 | |
1da177e4 LT |
2097 | return; |
2098 | } | |
2099 | out: | |
2100 | task_rq_unlock(rq, &flags); | |
2101 | } | |
2102 | ||
2103 | /* | |
476d139c NP |
2104 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2105 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2106 | */ |
2107 | void sched_exec(void) | |
2108 | { | |
1da177e4 | 2109 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2110 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2111 | put_cpu(); |
476d139c NP |
2112 | if (new_cpu != this_cpu) |
2113 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2114 | } |
2115 | ||
2116 | /* | |
2117 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2118 | * Both runqueues must be locked. | |
2119 | */ | |
dd41f596 IM |
2120 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2121 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2122 | { |
8e717b19 | 2123 | deactivate_task(src_rq, p, 0, rq_clock(src_rq)); |
1da177e4 | 2124 | set_task_cpu(p, this_cpu); |
dd41f596 | 2125 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2126 | /* |
2127 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2128 | * to be always true for them. | |
2129 | */ | |
dd41f596 | 2130 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2131 | } |
2132 | ||
2133 | /* | |
2134 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2135 | */ | |
858119e1 | 2136 | static |
70b97a7f | 2137 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2138 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2139 | int *all_pinned) |
1da177e4 LT |
2140 | { |
2141 | /* | |
2142 | * We do not migrate tasks that are: | |
2143 | * 1) running (obviously), or | |
2144 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2145 | * 3) are cache-hot on their current CPU. | |
2146 | */ | |
1da177e4 LT |
2147 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2148 | return 0; | |
81026794 NP |
2149 | *all_pinned = 0; |
2150 | ||
2151 | if (task_running(rq, p)) | |
2152 | return 0; | |
1da177e4 LT |
2153 | |
2154 | /* | |
dd41f596 | 2155 | * Aggressive migration if too many balance attempts have failed: |
1da177e4 | 2156 | */ |
dd41f596 | 2157 | if (sd->nr_balance_failed > sd->cache_nice_tries) |
1da177e4 LT |
2158 | return 1; |
2159 | ||
1da177e4 LT |
2160 | return 1; |
2161 | } | |
2162 | ||
dd41f596 | 2163 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2164 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2165 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2166 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2167 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2168 | { |
dd41f596 IM |
2169 | int pulled = 0, pinned = 0, skip_for_load; |
2170 | struct task_struct *p; | |
2171 | long rem_load_move = max_load_move; | |
1da177e4 | 2172 | |
2dd73a4f | 2173 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2174 | goto out; |
2175 | ||
81026794 NP |
2176 | pinned = 1; |
2177 | ||
1da177e4 | 2178 | /* |
dd41f596 | 2179 | * Start the load-balancing iterator: |
1da177e4 | 2180 | */ |
dd41f596 IM |
2181 | p = iterator->start(iterator->arg); |
2182 | next: | |
2183 | if (!p) | |
1da177e4 | 2184 | goto out; |
50ddd969 PW |
2185 | /* |
2186 | * To help distribute high priority tasks accross CPUs we don't | |
2187 | * skip a task if it will be the highest priority task (i.e. smallest | |
2188 | * prio value) on its new queue regardless of its load weight | |
2189 | */ | |
dd41f596 IM |
2190 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2191 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2192 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2193 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2194 | p = iterator->next(iterator->arg); |
2195 | goto next; | |
1da177e4 LT |
2196 | } |
2197 | ||
dd41f596 | 2198 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2199 | pulled++; |
dd41f596 | 2200 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2201 | |
2dd73a4f PW |
2202 | /* |
2203 | * We only want to steal up to the prescribed number of tasks | |
2204 | * and the prescribed amount of weighted load. | |
2205 | */ | |
2206 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2207 | if (p->prio < *this_best_prio) |
2208 | *this_best_prio = p->prio; | |
dd41f596 IM |
2209 | p = iterator->next(iterator->arg); |
2210 | goto next; | |
1da177e4 LT |
2211 | } |
2212 | out: | |
2213 | /* | |
2214 | * Right now, this is the only place pull_task() is called, | |
2215 | * so we can safely collect pull_task() stats here rather than | |
2216 | * inside pull_task(). | |
2217 | */ | |
2218 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2219 | |
2220 | if (all_pinned) | |
2221 | *all_pinned = pinned; | |
dd41f596 | 2222 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2223 | return pulled; |
2224 | } | |
2225 | ||
dd41f596 | 2226 | /* |
43010659 PW |
2227 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2228 | * this_rq, as part of a balancing operation within domain "sd". | |
2229 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2230 | * |
2231 | * Called with both runqueues locked. | |
2232 | */ | |
2233 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2234 | unsigned long max_load_move, |
dd41f596 IM |
2235 | struct sched_domain *sd, enum cpu_idle_type idle, |
2236 | int *all_pinned) | |
2237 | { | |
2238 | struct sched_class *class = sched_class_highest; | |
43010659 | 2239 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2240 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2241 | |
2242 | do { | |
43010659 PW |
2243 | total_load_moved += |
2244 | class->load_balance(this_rq, this_cpu, busiest, | |
2245 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2246 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2247 | class = class->next; |
43010659 | 2248 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2249 | |
43010659 PW |
2250 | return total_load_moved > 0; |
2251 | } | |
2252 | ||
2253 | /* | |
2254 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2255 | * part of active balancing operations within "domain". | |
2256 | * Returns 1 if successful and 0 otherwise. | |
2257 | * | |
2258 | * Called with both runqueues locked. | |
2259 | */ | |
2260 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2261 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2262 | { | |
2263 | struct sched_class *class; | |
a4ac01c3 | 2264 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2265 | |
2266 | for (class = sched_class_highest; class; class = class->next) | |
2267 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2268 | 1, ULONG_MAX, sd, idle, NULL, |
2269 | &this_best_prio)) | |
43010659 PW |
2270 | return 1; |
2271 | ||
2272 | return 0; | |
dd41f596 IM |
2273 | } |
2274 | ||
1da177e4 LT |
2275 | /* |
2276 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2277 | * domain. It calculates and returns the amount of weighted load which |
2278 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2279 | */ |
2280 | static struct sched_group * | |
2281 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2282 | unsigned long *imbalance, enum cpu_idle_type idle, |
2283 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2284 | { |
2285 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2286 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2287 | unsigned long max_pull; |
2dd73a4f PW |
2288 | unsigned long busiest_load_per_task, busiest_nr_running; |
2289 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2290 | int load_idx; |
5c45bf27 SS |
2291 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2292 | int power_savings_balance = 1; | |
2293 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2294 | unsigned long min_nr_running = ULONG_MAX; | |
2295 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2296 | #endif | |
1da177e4 LT |
2297 | |
2298 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2299 | busiest_load_per_task = busiest_nr_running = 0; |
2300 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2301 | if (idle == CPU_NOT_IDLE) |
7897986b | 2302 | load_idx = sd->busy_idx; |
d15bcfdb | 2303 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2304 | load_idx = sd->newidle_idx; |
2305 | else | |
2306 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2307 | |
2308 | do { | |
5c45bf27 | 2309 | unsigned long load, group_capacity; |
1da177e4 LT |
2310 | int local_group; |
2311 | int i; | |
783609c6 | 2312 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2313 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2314 | |
2315 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2316 | ||
783609c6 SS |
2317 | if (local_group) |
2318 | balance_cpu = first_cpu(group->cpumask); | |
2319 | ||
1da177e4 | 2320 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2321 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2322 | |
2323 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2324 | struct rq *rq; |
2325 | ||
2326 | if (!cpu_isset(i, *cpus)) | |
2327 | continue; | |
2328 | ||
2329 | rq = cpu_rq(i); | |
2dd73a4f | 2330 | |
9439aab8 | 2331 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2332 | *sd_idle = 0; |
2333 | ||
1da177e4 | 2334 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2335 | if (local_group) { |
2336 | if (idle_cpu(i) && !first_idle_cpu) { | |
2337 | first_idle_cpu = 1; | |
2338 | balance_cpu = i; | |
2339 | } | |
2340 | ||
a2000572 | 2341 | load = target_load(i, load_idx); |
783609c6 | 2342 | } else |
a2000572 | 2343 | load = source_load(i, load_idx); |
1da177e4 LT |
2344 | |
2345 | avg_load += load; | |
2dd73a4f | 2346 | sum_nr_running += rq->nr_running; |
dd41f596 | 2347 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2348 | } |
2349 | ||
783609c6 SS |
2350 | /* |
2351 | * First idle cpu or the first cpu(busiest) in this sched group | |
2352 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2353 | * domains. In the newly idle case, we will allow all the cpu's |
2354 | * to do the newly idle load balance. | |
783609c6 | 2355 | */ |
9439aab8 SS |
2356 | if (idle != CPU_NEWLY_IDLE && local_group && |
2357 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2358 | *balance = 0; |
2359 | goto ret; | |
2360 | } | |
2361 | ||
1da177e4 | 2362 | total_load += avg_load; |
5517d86b | 2363 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2364 | |
2365 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2366 | avg_load = sg_div_cpu_power(group, |
2367 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2368 | |
5517d86b | 2369 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2370 | |
1da177e4 LT |
2371 | if (local_group) { |
2372 | this_load = avg_load; | |
2373 | this = group; | |
2dd73a4f PW |
2374 | this_nr_running = sum_nr_running; |
2375 | this_load_per_task = sum_weighted_load; | |
2376 | } else if (avg_load > max_load && | |
5c45bf27 | 2377 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2378 | max_load = avg_load; |
2379 | busiest = group; | |
2dd73a4f PW |
2380 | busiest_nr_running = sum_nr_running; |
2381 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2382 | } |
5c45bf27 SS |
2383 | |
2384 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2385 | /* | |
2386 | * Busy processors will not participate in power savings | |
2387 | * balance. | |
2388 | */ | |
dd41f596 IM |
2389 | if (idle == CPU_NOT_IDLE || |
2390 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2391 | goto group_next; | |
5c45bf27 SS |
2392 | |
2393 | /* | |
2394 | * If the local group is idle or completely loaded | |
2395 | * no need to do power savings balance at this domain | |
2396 | */ | |
2397 | if (local_group && (this_nr_running >= group_capacity || | |
2398 | !this_nr_running)) | |
2399 | power_savings_balance = 0; | |
2400 | ||
dd41f596 | 2401 | /* |
5c45bf27 SS |
2402 | * If a group is already running at full capacity or idle, |
2403 | * don't include that group in power savings calculations | |
dd41f596 IM |
2404 | */ |
2405 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2406 | || !sum_nr_running) |
dd41f596 | 2407 | goto group_next; |
5c45bf27 | 2408 | |
dd41f596 | 2409 | /* |
5c45bf27 | 2410 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2411 | * This is the group from where we need to pick up the load |
2412 | * for saving power | |
2413 | */ | |
2414 | if ((sum_nr_running < min_nr_running) || | |
2415 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2416 | first_cpu(group->cpumask) < |
2417 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2418 | group_min = group; |
2419 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2420 | min_load_per_task = sum_weighted_load / |
2421 | sum_nr_running; | |
dd41f596 | 2422 | } |
5c45bf27 | 2423 | |
dd41f596 | 2424 | /* |
5c45bf27 | 2425 | * Calculate the group which is almost near its |
dd41f596 IM |
2426 | * capacity but still has some space to pick up some load |
2427 | * from other group and save more power | |
2428 | */ | |
2429 | if (sum_nr_running <= group_capacity - 1) { | |
2430 | if (sum_nr_running > leader_nr_running || | |
2431 | (sum_nr_running == leader_nr_running && | |
2432 | first_cpu(group->cpumask) > | |
2433 | first_cpu(group_leader->cpumask))) { | |
2434 | group_leader = group; | |
2435 | leader_nr_running = sum_nr_running; | |
2436 | } | |
48f24c4d | 2437 | } |
5c45bf27 SS |
2438 | group_next: |
2439 | #endif | |
1da177e4 LT |
2440 | group = group->next; |
2441 | } while (group != sd->groups); | |
2442 | ||
2dd73a4f | 2443 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2444 | goto out_balanced; |
2445 | ||
2446 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2447 | ||
2448 | if (this_load >= avg_load || | |
2449 | 100*max_load <= sd->imbalance_pct*this_load) | |
2450 | goto out_balanced; | |
2451 | ||
2dd73a4f | 2452 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2453 | /* |
2454 | * We're trying to get all the cpus to the average_load, so we don't | |
2455 | * want to push ourselves above the average load, nor do we wish to | |
2456 | * reduce the max loaded cpu below the average load, as either of these | |
2457 | * actions would just result in more rebalancing later, and ping-pong | |
2458 | * tasks around. Thus we look for the minimum possible imbalance. | |
2459 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2460 | * be counted as no imbalance for these purposes -- we can't fix that | |
2461 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2462 | * appear as very large values with unsigned longs. | |
2463 | */ | |
2dd73a4f PW |
2464 | if (max_load <= busiest_load_per_task) |
2465 | goto out_balanced; | |
2466 | ||
2467 | /* | |
2468 | * In the presence of smp nice balancing, certain scenarios can have | |
2469 | * max load less than avg load(as we skip the groups at or below | |
2470 | * its cpu_power, while calculating max_load..) | |
2471 | */ | |
2472 | if (max_load < avg_load) { | |
2473 | *imbalance = 0; | |
2474 | goto small_imbalance; | |
2475 | } | |
0c117f1b SS |
2476 | |
2477 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2478 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2479 | |
1da177e4 | 2480 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2481 | *imbalance = min(max_pull * busiest->__cpu_power, |
2482 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2483 | / SCHED_LOAD_SCALE; |
2484 | ||
2dd73a4f PW |
2485 | /* |
2486 | * if *imbalance is less than the average load per runnable task | |
2487 | * there is no gaurantee that any tasks will be moved so we'll have | |
2488 | * a think about bumping its value to force at least one task to be | |
2489 | * moved | |
2490 | */ | |
dd41f596 | 2491 | if (*imbalance + SCHED_LOAD_SCALE_FUZZ < busiest_load_per_task/2) { |
48f24c4d | 2492 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2493 | unsigned int imbn; |
2494 | ||
2495 | small_imbalance: | |
2496 | pwr_move = pwr_now = 0; | |
2497 | imbn = 2; | |
2498 | if (this_nr_running) { | |
2499 | this_load_per_task /= this_nr_running; | |
2500 | if (busiest_load_per_task > this_load_per_task) | |
2501 | imbn = 1; | |
2502 | } else | |
2503 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2504 | |
dd41f596 IM |
2505 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2506 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2507 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2508 | return busiest; |
2509 | } | |
2510 | ||
2511 | /* | |
2512 | * OK, we don't have enough imbalance to justify moving tasks, | |
2513 | * however we may be able to increase total CPU power used by | |
2514 | * moving them. | |
2515 | */ | |
2516 | ||
5517d86b ED |
2517 | pwr_now += busiest->__cpu_power * |
2518 | min(busiest_load_per_task, max_load); | |
2519 | pwr_now += this->__cpu_power * | |
2520 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2521 | pwr_now /= SCHED_LOAD_SCALE; |
2522 | ||
2523 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2524 | tmp = sg_div_cpu_power(busiest, |
2525 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2526 | if (max_load > tmp) |
5517d86b | 2527 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2528 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2529 | |
2530 | /* Amount of load we'd add */ | |
5517d86b | 2531 | if (max_load * busiest->__cpu_power < |
33859f7f | 2532 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2533 | tmp = sg_div_cpu_power(this, |
2534 | max_load * busiest->__cpu_power); | |
1da177e4 | 2535 | else |
5517d86b ED |
2536 | tmp = sg_div_cpu_power(this, |
2537 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2538 | pwr_move += this->__cpu_power * | |
2539 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2540 | pwr_move /= SCHED_LOAD_SCALE; |
2541 | ||
2542 | /* Move if we gain throughput */ | |
2543 | if (pwr_move <= pwr_now) | |
2544 | goto out_balanced; | |
2545 | ||
2dd73a4f | 2546 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2547 | } |
2548 | ||
1da177e4 LT |
2549 | return busiest; |
2550 | ||
2551 | out_balanced: | |
5c45bf27 | 2552 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2553 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2554 | goto ret; |
1da177e4 | 2555 | |
5c45bf27 SS |
2556 | if (this == group_leader && group_leader != group_min) { |
2557 | *imbalance = min_load_per_task; | |
2558 | return group_min; | |
2559 | } | |
5c45bf27 | 2560 | #endif |
783609c6 | 2561 | ret: |
1da177e4 LT |
2562 | *imbalance = 0; |
2563 | return NULL; | |
2564 | } | |
2565 | ||
2566 | /* | |
2567 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2568 | */ | |
70b97a7f | 2569 | static struct rq * |
d15bcfdb | 2570 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2571 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2572 | { |
70b97a7f | 2573 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2574 | unsigned long max_load = 0; |
1da177e4 LT |
2575 | int i; |
2576 | ||
2577 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2578 | unsigned long wl; |
0a2966b4 CL |
2579 | |
2580 | if (!cpu_isset(i, *cpus)) | |
2581 | continue; | |
2582 | ||
48f24c4d | 2583 | rq = cpu_rq(i); |
dd41f596 | 2584 | wl = weighted_cpuload(i); |
2dd73a4f | 2585 | |
dd41f596 | 2586 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2587 | continue; |
1da177e4 | 2588 | |
dd41f596 IM |
2589 | if (wl > max_load) { |
2590 | max_load = wl; | |
48f24c4d | 2591 | busiest = rq; |
1da177e4 LT |
2592 | } |
2593 | } | |
2594 | ||
2595 | return busiest; | |
2596 | } | |
2597 | ||
77391d71 NP |
2598 | /* |
2599 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2600 | * so long as it is large enough. | |
2601 | */ | |
2602 | #define MAX_PINNED_INTERVAL 512 | |
2603 | ||
1da177e4 LT |
2604 | /* |
2605 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2606 | * tasks if there is an imbalance. | |
1da177e4 | 2607 | */ |
70b97a7f | 2608 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2609 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2610 | int *balance) |
1da177e4 | 2611 | { |
43010659 | 2612 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2613 | struct sched_group *group; |
1da177e4 | 2614 | unsigned long imbalance; |
70b97a7f | 2615 | struct rq *busiest; |
0a2966b4 | 2616 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2617 | unsigned long flags; |
5969fe06 | 2618 | |
89c4710e SS |
2619 | /* |
2620 | * When power savings policy is enabled for the parent domain, idle | |
2621 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2622 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2623 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2624 | */ |
d15bcfdb | 2625 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2626 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2627 | sd_idle = 1; |
1da177e4 | 2628 | |
1da177e4 LT |
2629 | schedstat_inc(sd, lb_cnt[idle]); |
2630 | ||
0a2966b4 CL |
2631 | redo: |
2632 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2633 | &cpus, balance); |
2634 | ||
06066714 | 2635 | if (*balance == 0) |
783609c6 | 2636 | goto out_balanced; |
783609c6 | 2637 | |
1da177e4 LT |
2638 | if (!group) { |
2639 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2640 | goto out_balanced; | |
2641 | } | |
2642 | ||
0a2966b4 | 2643 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2644 | if (!busiest) { |
2645 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2646 | goto out_balanced; | |
2647 | } | |
2648 | ||
db935dbd | 2649 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2650 | |
2651 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2652 | ||
43010659 | 2653 | ld_moved = 0; |
1da177e4 LT |
2654 | if (busiest->nr_running > 1) { |
2655 | /* | |
2656 | * Attempt to move tasks. If find_busiest_group has found | |
2657 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2658 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2659 | * correctly treated as an imbalance. |
2660 | */ | |
fe2eea3f | 2661 | local_irq_save(flags); |
e17224bf | 2662 | double_rq_lock(this_rq, busiest); |
43010659 | 2663 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2664 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2665 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2666 | local_irq_restore(flags); |
81026794 | 2667 | |
46cb4b7c SS |
2668 | /* |
2669 | * some other cpu did the load balance for us. | |
2670 | */ | |
43010659 | 2671 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2672 | resched_cpu(this_cpu); |
2673 | ||
81026794 | 2674 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2675 | if (unlikely(all_pinned)) { |
2676 | cpu_clear(cpu_of(busiest), cpus); | |
2677 | if (!cpus_empty(cpus)) | |
2678 | goto redo; | |
81026794 | 2679 | goto out_balanced; |
0a2966b4 | 2680 | } |
1da177e4 | 2681 | } |
81026794 | 2682 | |
43010659 | 2683 | if (!ld_moved) { |
1da177e4 LT |
2684 | schedstat_inc(sd, lb_failed[idle]); |
2685 | sd->nr_balance_failed++; | |
2686 | ||
2687 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2688 | |
fe2eea3f | 2689 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2690 | |
2691 | /* don't kick the migration_thread, if the curr | |
2692 | * task on busiest cpu can't be moved to this_cpu | |
2693 | */ | |
2694 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2695 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2696 | all_pinned = 1; |
2697 | goto out_one_pinned; | |
2698 | } | |
2699 | ||
1da177e4 LT |
2700 | if (!busiest->active_balance) { |
2701 | busiest->active_balance = 1; | |
2702 | busiest->push_cpu = this_cpu; | |
81026794 | 2703 | active_balance = 1; |
1da177e4 | 2704 | } |
fe2eea3f | 2705 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2706 | if (active_balance) |
1da177e4 LT |
2707 | wake_up_process(busiest->migration_thread); |
2708 | ||
2709 | /* | |
2710 | * We've kicked active balancing, reset the failure | |
2711 | * counter. | |
2712 | */ | |
39507451 | 2713 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2714 | } |
81026794 | 2715 | } else |
1da177e4 LT |
2716 | sd->nr_balance_failed = 0; |
2717 | ||
81026794 | 2718 | if (likely(!active_balance)) { |
1da177e4 LT |
2719 | /* We were unbalanced, so reset the balancing interval */ |
2720 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2721 | } else { |
2722 | /* | |
2723 | * If we've begun active balancing, start to back off. This | |
2724 | * case may not be covered by the all_pinned logic if there | |
2725 | * is only 1 task on the busy runqueue (because we don't call | |
2726 | * move_tasks). | |
2727 | */ | |
2728 | if (sd->balance_interval < sd->max_interval) | |
2729 | sd->balance_interval *= 2; | |
1da177e4 LT |
2730 | } |
2731 | ||
43010659 | 2732 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2733 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2734 | return -1; |
43010659 | 2735 | return ld_moved; |
1da177e4 LT |
2736 | |
2737 | out_balanced: | |
1da177e4 LT |
2738 | schedstat_inc(sd, lb_balanced[idle]); |
2739 | ||
16cfb1c0 | 2740 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2741 | |
2742 | out_one_pinned: | |
1da177e4 | 2743 | /* tune up the balancing interval */ |
77391d71 NP |
2744 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2745 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2746 | sd->balance_interval *= 2; |
2747 | ||
48f24c4d | 2748 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2749 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2750 | return -1; |
1da177e4 LT |
2751 | return 0; |
2752 | } | |
2753 | ||
2754 | /* | |
2755 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2756 | * tasks if there is an imbalance. | |
2757 | * | |
d15bcfdb | 2758 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2759 | * this_rq is locked. |
2760 | */ | |
48f24c4d | 2761 | static int |
70b97a7f | 2762 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2763 | { |
2764 | struct sched_group *group; | |
70b97a7f | 2765 | struct rq *busiest = NULL; |
1da177e4 | 2766 | unsigned long imbalance; |
43010659 | 2767 | int ld_moved = 0; |
5969fe06 | 2768 | int sd_idle = 0; |
969bb4e4 | 2769 | int all_pinned = 0; |
0a2966b4 | 2770 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2771 | |
89c4710e SS |
2772 | /* |
2773 | * When power savings policy is enabled for the parent domain, idle | |
2774 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2775 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2776 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2777 | */ |
2778 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2779 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2780 | sd_idle = 1; |
1da177e4 | 2781 | |
d15bcfdb | 2782 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2783 | redo: |
d15bcfdb | 2784 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2785 | &sd_idle, &cpus, NULL); |
1da177e4 | 2786 | if (!group) { |
d15bcfdb | 2787 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2788 | goto out_balanced; |
1da177e4 LT |
2789 | } |
2790 | ||
d15bcfdb | 2791 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2792 | &cpus); |
db935dbd | 2793 | if (!busiest) { |
d15bcfdb | 2794 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2795 | goto out_balanced; |
1da177e4 LT |
2796 | } |
2797 | ||
db935dbd NP |
2798 | BUG_ON(busiest == this_rq); |
2799 | ||
d15bcfdb | 2800 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2801 | |
43010659 | 2802 | ld_moved = 0; |
d6d5cfaf NP |
2803 | if (busiest->nr_running > 1) { |
2804 | /* Attempt to move tasks */ | |
2805 | double_lock_balance(this_rq, busiest); | |
43010659 | 2806 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2807 | imbalance, sd, CPU_NEWLY_IDLE, |
2808 | &all_pinned); | |
d6d5cfaf | 2809 | spin_unlock(&busiest->lock); |
0a2966b4 | 2810 | |
969bb4e4 | 2811 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2812 | cpu_clear(cpu_of(busiest), cpus); |
2813 | if (!cpus_empty(cpus)) | |
2814 | goto redo; | |
2815 | } | |
d6d5cfaf NP |
2816 | } |
2817 | ||
43010659 | 2818 | if (!ld_moved) { |
d15bcfdb | 2819 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2820 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2821 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2822 | return -1; |
2823 | } else | |
16cfb1c0 | 2824 | sd->nr_balance_failed = 0; |
1da177e4 | 2825 | |
43010659 | 2826 | return ld_moved; |
16cfb1c0 NP |
2827 | |
2828 | out_balanced: | |
d15bcfdb | 2829 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2830 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2831 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2832 | return -1; |
16cfb1c0 | 2833 | sd->nr_balance_failed = 0; |
48f24c4d | 2834 | |
16cfb1c0 | 2835 | return 0; |
1da177e4 LT |
2836 | } |
2837 | ||
2838 | /* | |
2839 | * idle_balance is called by schedule() if this_cpu is about to become | |
2840 | * idle. Attempts to pull tasks from other CPUs. | |
2841 | */ | |
70b97a7f | 2842 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2843 | { |
2844 | struct sched_domain *sd; | |
dd41f596 IM |
2845 | int pulled_task = -1; |
2846 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2847 | |
2848 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2849 | unsigned long interval; |
2850 | ||
2851 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2852 | continue; | |
2853 | ||
2854 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2855 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2856 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2857 | this_rq, sd); |
2858 | ||
2859 | interval = msecs_to_jiffies(sd->balance_interval); | |
2860 | if (time_after(next_balance, sd->last_balance + interval)) | |
2861 | next_balance = sd->last_balance + interval; | |
2862 | if (pulled_task) | |
2863 | break; | |
1da177e4 | 2864 | } |
dd41f596 | 2865 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2866 | /* |
2867 | * We are going idle. next_balance may be set based on | |
2868 | * a busy processor. So reset next_balance. | |
2869 | */ | |
2870 | this_rq->next_balance = next_balance; | |
dd41f596 | 2871 | } |
1da177e4 LT |
2872 | } |
2873 | ||
2874 | /* | |
2875 | * active_load_balance is run by migration threads. It pushes running tasks | |
2876 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2877 | * running on each physical CPU where possible, and avoids physical / | |
2878 | * logical imbalances. | |
2879 | * | |
2880 | * Called with busiest_rq locked. | |
2881 | */ | |
70b97a7f | 2882 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2883 | { |
39507451 | 2884 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2885 | struct sched_domain *sd; |
2886 | struct rq *target_rq; | |
39507451 | 2887 | |
48f24c4d | 2888 | /* Is there any task to move? */ |
39507451 | 2889 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2890 | return; |
2891 | ||
2892 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2893 | |
2894 | /* | |
39507451 NP |
2895 | * This condition is "impossible", if it occurs |
2896 | * we need to fix it. Originally reported by | |
2897 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2898 | */ |
39507451 | 2899 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2900 | |
39507451 NP |
2901 | /* move a task from busiest_rq to target_rq */ |
2902 | double_lock_balance(busiest_rq, target_rq); | |
2903 | ||
2904 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2905 | for_each_domain(target_cpu, sd) { |
39507451 | 2906 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2907 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2908 | break; |
c96d145e | 2909 | } |
39507451 | 2910 | |
48f24c4d IM |
2911 | if (likely(sd)) { |
2912 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2913 | |
43010659 PW |
2914 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2915 | sd, CPU_IDLE)) | |
48f24c4d IM |
2916 | schedstat_inc(sd, alb_pushed); |
2917 | else | |
2918 | schedstat_inc(sd, alb_failed); | |
2919 | } | |
39507451 | 2920 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2921 | } |
2922 | ||
46cb4b7c SS |
2923 | #ifdef CONFIG_NO_HZ |
2924 | static struct { | |
2925 | atomic_t load_balancer; | |
2926 | cpumask_t cpu_mask; | |
2927 | } nohz ____cacheline_aligned = { | |
2928 | .load_balancer = ATOMIC_INIT(-1), | |
2929 | .cpu_mask = CPU_MASK_NONE, | |
2930 | }; | |
2931 | ||
7835b98b | 2932 | /* |
46cb4b7c SS |
2933 | * This routine will try to nominate the ilb (idle load balancing) |
2934 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2935 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2936 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2937 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2938 | * arrives... | |
2939 | * | |
2940 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2941 | * for idle load balancing. ilb owner will still be part of | |
2942 | * nohz.cpu_mask.. | |
7835b98b | 2943 | * |
46cb4b7c SS |
2944 | * While stopping the tick, this cpu will become the ilb owner if there |
2945 | * is no other owner. And will be the owner till that cpu becomes busy | |
2946 | * or if all cpus in the system stop their ticks at which point | |
2947 | * there is no need for ilb owner. | |
2948 | * | |
2949 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2950 | * next busy scheduler_tick() | |
2951 | */ | |
2952 | int select_nohz_load_balancer(int stop_tick) | |
2953 | { | |
2954 | int cpu = smp_processor_id(); | |
2955 | ||
2956 | if (stop_tick) { | |
2957 | cpu_set(cpu, nohz.cpu_mask); | |
2958 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2959 | ||
2960 | /* | |
2961 | * If we are going offline and still the leader, give up! | |
2962 | */ | |
2963 | if (cpu_is_offline(cpu) && | |
2964 | atomic_read(&nohz.load_balancer) == cpu) { | |
2965 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2966 | BUG(); | |
2967 | return 0; | |
2968 | } | |
2969 | ||
2970 | /* time for ilb owner also to sleep */ | |
2971 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2972 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2973 | atomic_set(&nohz.load_balancer, -1); | |
2974 | return 0; | |
2975 | } | |
2976 | ||
2977 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2978 | /* make me the ilb owner */ | |
2979 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2980 | return 1; | |
2981 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2982 | return 1; | |
2983 | } else { | |
2984 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2985 | return 0; | |
2986 | ||
2987 | cpu_clear(cpu, nohz.cpu_mask); | |
2988 | ||
2989 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2990 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2991 | BUG(); | |
2992 | } | |
2993 | return 0; | |
2994 | } | |
2995 | #endif | |
2996 | ||
2997 | static DEFINE_SPINLOCK(balancing); | |
2998 | ||
2999 | /* | |
7835b98b CL |
3000 | * It checks each scheduling domain to see if it is due to be balanced, |
3001 | * and initiates a balancing operation if so. | |
3002 | * | |
3003 | * Balancing parameters are set up in arch_init_sched_domains. | |
3004 | */ | |
d15bcfdb | 3005 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3006 | { |
46cb4b7c SS |
3007 | int balance = 1; |
3008 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3009 | unsigned long interval; |
3010 | struct sched_domain *sd; | |
46cb4b7c | 3011 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3012 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 3013 | |
46cb4b7c | 3014 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3015 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3016 | continue; | |
3017 | ||
3018 | interval = sd->balance_interval; | |
d15bcfdb | 3019 | if (idle != CPU_IDLE) |
1da177e4 LT |
3020 | interval *= sd->busy_factor; |
3021 | ||
3022 | /* scale ms to jiffies */ | |
3023 | interval = msecs_to_jiffies(interval); | |
3024 | if (unlikely(!interval)) | |
3025 | interval = 1; | |
dd41f596 IM |
3026 | if (interval > HZ*NR_CPUS/10) |
3027 | interval = HZ*NR_CPUS/10; | |
3028 | ||
1da177e4 | 3029 | |
08c183f3 CL |
3030 | if (sd->flags & SD_SERIALIZE) { |
3031 | if (!spin_trylock(&balancing)) | |
3032 | goto out; | |
3033 | } | |
3034 | ||
c9819f45 | 3035 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3036 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3037 | /* |
3038 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3039 | * longer idle, or one of our SMT siblings is |
3040 | * not idle. | |
3041 | */ | |
d15bcfdb | 3042 | idle = CPU_NOT_IDLE; |
1da177e4 | 3043 | } |
1bd77f2d | 3044 | sd->last_balance = jiffies; |
1da177e4 | 3045 | } |
08c183f3 CL |
3046 | if (sd->flags & SD_SERIALIZE) |
3047 | spin_unlock(&balancing); | |
3048 | out: | |
c9819f45 CL |
3049 | if (time_after(next_balance, sd->last_balance + interval)) |
3050 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
3051 | |
3052 | /* | |
3053 | * Stop the load balance at this level. There is another | |
3054 | * CPU in our sched group which is doing load balancing more | |
3055 | * actively. | |
3056 | */ | |
3057 | if (!balance) | |
3058 | break; | |
1da177e4 | 3059 | } |
46cb4b7c SS |
3060 | rq->next_balance = next_balance; |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3065 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3066 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3067 | */ | |
3068 | static void run_rebalance_domains(struct softirq_action *h) | |
3069 | { | |
dd41f596 IM |
3070 | int this_cpu = smp_processor_id(); |
3071 | struct rq *this_rq = cpu_rq(this_cpu); | |
3072 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3073 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3074 | |
dd41f596 | 3075 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3076 | |
3077 | #ifdef CONFIG_NO_HZ | |
3078 | /* | |
3079 | * If this cpu is the owner for idle load balancing, then do the | |
3080 | * balancing on behalf of the other idle cpus whose ticks are | |
3081 | * stopped. | |
3082 | */ | |
dd41f596 IM |
3083 | if (this_rq->idle_at_tick && |
3084 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3085 | cpumask_t cpus = nohz.cpu_mask; |
3086 | struct rq *rq; | |
3087 | int balance_cpu; | |
3088 | ||
dd41f596 | 3089 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3090 | for_each_cpu_mask(balance_cpu, cpus) { |
3091 | /* | |
3092 | * If this cpu gets work to do, stop the load balancing | |
3093 | * work being done for other cpus. Next load | |
3094 | * balancing owner will pick it up. | |
3095 | */ | |
3096 | if (need_resched()) | |
3097 | break; | |
3098 | ||
dd41f596 | 3099 | rebalance_domains(balance_cpu, SCHED_IDLE); |
46cb4b7c SS |
3100 | |
3101 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3102 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3103 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3104 | } |
3105 | } | |
3106 | #endif | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3111 | * | |
3112 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3113 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3114 | * if the whole system is idle. | |
3115 | */ | |
dd41f596 | 3116 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3117 | { |
46cb4b7c SS |
3118 | #ifdef CONFIG_NO_HZ |
3119 | /* | |
3120 | * If we were in the nohz mode recently and busy at the current | |
3121 | * scheduler tick, then check if we need to nominate new idle | |
3122 | * load balancer. | |
3123 | */ | |
3124 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3125 | rq->in_nohz_recently = 0; | |
3126 | ||
3127 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3128 | cpu_clear(cpu, nohz.cpu_mask); | |
3129 | atomic_set(&nohz.load_balancer, -1); | |
3130 | } | |
3131 | ||
3132 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3133 | /* | |
3134 | * simple selection for now: Nominate the | |
3135 | * first cpu in the nohz list to be the next | |
3136 | * ilb owner. | |
3137 | * | |
3138 | * TBD: Traverse the sched domains and nominate | |
3139 | * the nearest cpu in the nohz.cpu_mask. | |
3140 | */ | |
3141 | int ilb = first_cpu(nohz.cpu_mask); | |
3142 | ||
3143 | if (ilb != NR_CPUS) | |
3144 | resched_cpu(ilb); | |
3145 | } | |
3146 | } | |
3147 | ||
3148 | /* | |
3149 | * If this cpu is idle and doing idle load balancing for all the | |
3150 | * cpus with ticks stopped, is it time for that to stop? | |
3151 | */ | |
3152 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3153 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3154 | resched_cpu(cpu); | |
3155 | return; | |
3156 | } | |
3157 | ||
3158 | /* | |
3159 | * If this cpu is idle and the idle load balancing is done by | |
3160 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3161 | */ | |
3162 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3163 | cpu_isset(cpu, nohz.cpu_mask)) | |
3164 | return; | |
3165 | #endif | |
3166 | if (time_after_eq(jiffies, rq->next_balance)) | |
3167 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3168 | } |
dd41f596 IM |
3169 | |
3170 | #else /* CONFIG_SMP */ | |
3171 | ||
1da177e4 LT |
3172 | /* |
3173 | * on UP we do not need to balance between CPUs: | |
3174 | */ | |
70b97a7f | 3175 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3176 | { |
3177 | } | |
dd41f596 IM |
3178 | |
3179 | /* Avoid "used but not defined" warning on UP */ | |
3180 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3181 | unsigned long max_nr_move, unsigned long max_load_move, | |
3182 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3183 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3184 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3185 | { |
3186 | *load_moved = 0; | |
3187 | ||
3188 | return 0; | |
3189 | } | |
3190 | ||
1da177e4 LT |
3191 | #endif |
3192 | ||
1da177e4 LT |
3193 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3194 | ||
3195 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3196 | ||
3197 | /* | |
41b86e9c IM |
3198 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3199 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3200 | */ |
41b86e9c | 3201 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3202 | { |
1da177e4 | 3203 | unsigned long flags; |
41b86e9c IM |
3204 | u64 ns, delta_exec; |
3205 | struct rq *rq; | |
48f24c4d | 3206 | |
41b86e9c IM |
3207 | rq = task_rq_lock(p, &flags); |
3208 | ns = p->se.sum_exec_runtime; | |
3209 | if (rq->curr == p) { | |
3210 | delta_exec = rq_clock(rq) - p->se.exec_start; | |
3211 | if ((s64)delta_exec > 0) | |
3212 | ns += delta_exec; | |
3213 | } | |
3214 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3215 | |
1da177e4 LT |
3216 | return ns; |
3217 | } | |
3218 | ||
1da177e4 LT |
3219 | /* |
3220 | * Account user cpu time to a process. | |
3221 | * @p: the process that the cpu time gets accounted to | |
3222 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3223 | * @cputime: the cpu time spent in user space since the last update | |
3224 | */ | |
3225 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3226 | { | |
3227 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3228 | cputime64_t tmp; | |
3229 | ||
3230 | p->utime = cputime_add(p->utime, cputime); | |
3231 | ||
3232 | /* Add user time to cpustat. */ | |
3233 | tmp = cputime_to_cputime64(cputime); | |
3234 | if (TASK_NICE(p) > 0) | |
3235 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3236 | else | |
3237 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3238 | } | |
3239 | ||
3240 | /* | |
3241 | * Account system cpu time to a process. | |
3242 | * @p: the process that the cpu time gets accounted to | |
3243 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3244 | * @cputime: the cpu time spent in kernel space since the last update | |
3245 | */ | |
3246 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3247 | cputime_t cputime) | |
3248 | { | |
3249 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3250 | struct rq *rq = this_rq(); |
1da177e4 LT |
3251 | cputime64_t tmp; |
3252 | ||
3253 | p->stime = cputime_add(p->stime, cputime); | |
3254 | ||
3255 | /* Add system time to cpustat. */ | |
3256 | tmp = cputime_to_cputime64(cputime); | |
3257 | if (hardirq_count() - hardirq_offset) | |
3258 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3259 | else if (softirq_count()) | |
3260 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3261 | else if (p != rq->idle) | |
3262 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3263 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3264 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3265 | else | |
3266 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3267 | /* Account for system time used */ | |
3268 | acct_update_integrals(p); | |
1da177e4 LT |
3269 | } |
3270 | ||
3271 | /* | |
3272 | * Account for involuntary wait time. | |
3273 | * @p: the process from which the cpu time has been stolen | |
3274 | * @steal: the cpu time spent in involuntary wait | |
3275 | */ | |
3276 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3277 | { | |
3278 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3279 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3280 | struct rq *rq = this_rq(); |
1da177e4 LT |
3281 | |
3282 | if (p == rq->idle) { | |
3283 | p->stime = cputime_add(p->stime, steal); | |
3284 | if (atomic_read(&rq->nr_iowait) > 0) | |
3285 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3286 | else | |
3287 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3288 | } else | |
3289 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3290 | } | |
3291 | ||
7835b98b CL |
3292 | /* |
3293 | * This function gets called by the timer code, with HZ frequency. | |
3294 | * We call it with interrupts disabled. | |
3295 | * | |
3296 | * It also gets called by the fork code, when changing the parent's | |
3297 | * timeslices. | |
3298 | */ | |
3299 | void scheduler_tick(void) | |
3300 | { | |
7835b98b CL |
3301 | int cpu = smp_processor_id(); |
3302 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 IM |
3303 | struct task_struct *curr = rq->curr; |
3304 | ||
3305 | spin_lock(&rq->lock); | |
f1a438d8 | 3306 | update_cpu_load(rq); |
dd41f596 IM |
3307 | if (curr != rq->idle) /* FIXME: needed? */ |
3308 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3309 | spin_unlock(&rq->lock); |
7835b98b | 3310 | |
e418e1c2 | 3311 | #ifdef CONFIG_SMP |
dd41f596 IM |
3312 | rq->idle_at_tick = idle_cpu(cpu); |
3313 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3314 | #endif |
1da177e4 LT |
3315 | } |
3316 | ||
1da177e4 LT |
3317 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3318 | ||
3319 | void fastcall add_preempt_count(int val) | |
3320 | { | |
3321 | /* | |
3322 | * Underflow? | |
3323 | */ | |
9a11b49a IM |
3324 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3325 | return; | |
1da177e4 LT |
3326 | preempt_count() += val; |
3327 | /* | |
3328 | * Spinlock count overflowing soon? | |
3329 | */ | |
33859f7f MOS |
3330 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3331 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3332 | } |
3333 | EXPORT_SYMBOL(add_preempt_count); | |
3334 | ||
3335 | void fastcall sub_preempt_count(int val) | |
3336 | { | |
3337 | /* | |
3338 | * Underflow? | |
3339 | */ | |
9a11b49a IM |
3340 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3341 | return; | |
1da177e4 LT |
3342 | /* |
3343 | * Is the spinlock portion underflowing? | |
3344 | */ | |
9a11b49a IM |
3345 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3346 | !(preempt_count() & PREEMPT_MASK))) | |
3347 | return; | |
3348 | ||
1da177e4 LT |
3349 | preempt_count() -= val; |
3350 | } | |
3351 | EXPORT_SYMBOL(sub_preempt_count); | |
3352 | ||
3353 | #endif | |
3354 | ||
3355 | /* | |
dd41f596 | 3356 | * Print scheduling while atomic bug: |
1da177e4 | 3357 | */ |
dd41f596 | 3358 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3359 | { |
dd41f596 IM |
3360 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3361 | prev->comm, preempt_count(), prev->pid); | |
3362 | debug_show_held_locks(prev); | |
3363 | if (irqs_disabled()) | |
3364 | print_irqtrace_events(prev); | |
3365 | dump_stack(); | |
3366 | } | |
1da177e4 | 3367 | |
dd41f596 IM |
3368 | /* |
3369 | * Various schedule()-time debugging checks and statistics: | |
3370 | */ | |
3371 | static inline void schedule_debug(struct task_struct *prev) | |
3372 | { | |
1da177e4 LT |
3373 | /* |
3374 | * Test if we are atomic. Since do_exit() needs to call into | |
3375 | * schedule() atomically, we ignore that path for now. | |
3376 | * Otherwise, whine if we are scheduling when we should not be. | |
3377 | */ | |
dd41f596 IM |
3378 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3379 | __schedule_bug(prev); | |
3380 | ||
1da177e4 LT |
3381 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3382 | ||
dd41f596 IM |
3383 | schedstat_inc(this_rq(), sched_cnt); |
3384 | } | |
3385 | ||
3386 | /* | |
3387 | * Pick up the highest-prio task: | |
3388 | */ | |
3389 | static inline struct task_struct * | |
3390 | pick_next_task(struct rq *rq, struct task_struct *prev, u64 now) | |
3391 | { | |
3392 | struct sched_class *class; | |
3393 | struct task_struct *p; | |
1da177e4 LT |
3394 | |
3395 | /* | |
dd41f596 IM |
3396 | * Optimization: we know that if all tasks are in |
3397 | * the fair class we can call that function directly: | |
1da177e4 | 3398 | */ |
dd41f596 IM |
3399 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
3400 | p = fair_sched_class.pick_next_task(rq, now); | |
3401 | if (likely(p)) | |
3402 | return p; | |
1da177e4 LT |
3403 | } |
3404 | ||
dd41f596 IM |
3405 | class = sched_class_highest; |
3406 | for ( ; ; ) { | |
3407 | p = class->pick_next_task(rq, now); | |
3408 | if (p) | |
3409 | return p; | |
3410 | /* | |
3411 | * Will never be NULL as the idle class always | |
3412 | * returns a non-NULL p: | |
3413 | */ | |
3414 | class = class->next; | |
3415 | } | |
3416 | } | |
1da177e4 | 3417 | |
dd41f596 IM |
3418 | /* |
3419 | * schedule() is the main scheduler function. | |
3420 | */ | |
3421 | asmlinkage void __sched schedule(void) | |
3422 | { | |
3423 | struct task_struct *prev, *next; | |
3424 | long *switch_count; | |
3425 | struct rq *rq; | |
3426 | u64 now; | |
3427 | int cpu; | |
3428 | ||
3429 | need_resched: | |
3430 | preempt_disable(); | |
3431 | cpu = smp_processor_id(); | |
3432 | rq = cpu_rq(cpu); | |
3433 | rcu_qsctr_inc(cpu); | |
3434 | prev = rq->curr; | |
3435 | switch_count = &prev->nivcsw; | |
3436 | ||
3437 | release_kernel_lock(prev); | |
3438 | need_resched_nonpreemptible: | |
3439 | ||
3440 | schedule_debug(prev); | |
1da177e4 LT |
3441 | |
3442 | spin_lock_irq(&rq->lock); | |
dd41f596 | 3443 | clear_tsk_need_resched(prev); |
8e717b19 | 3444 | now = __rq_clock(rq); |
1da177e4 | 3445 | |
1da177e4 | 3446 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3447 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3448 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3449 | prev->state = TASK_RUNNING; |
dd41f596 | 3450 | } else { |
8e717b19 | 3451 | deactivate_task(rq, prev, 1, now); |
1da177e4 | 3452 | } |
dd41f596 | 3453 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3454 | } |
3455 | ||
dd41f596 | 3456 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3457 | idle_balance(cpu, rq); |
1da177e4 | 3458 | |
dd41f596 IM |
3459 | prev->sched_class->put_prev_task(rq, prev, now); |
3460 | next = pick_next_task(rq, prev, now); | |
1da177e4 LT |
3461 | |
3462 | sched_info_switch(prev, next); | |
dd41f596 | 3463 | |
1da177e4 | 3464 | if (likely(prev != next)) { |
1da177e4 LT |
3465 | rq->nr_switches++; |
3466 | rq->curr = next; | |
3467 | ++*switch_count; | |
3468 | ||
dd41f596 | 3469 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3470 | } else |
3471 | spin_unlock_irq(&rq->lock); | |
3472 | ||
dd41f596 IM |
3473 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3474 | cpu = smp_processor_id(); | |
3475 | rq = cpu_rq(cpu); | |
1da177e4 | 3476 | goto need_resched_nonpreemptible; |
dd41f596 | 3477 | } |
1da177e4 LT |
3478 | preempt_enable_no_resched(); |
3479 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3480 | goto need_resched; | |
3481 | } | |
1da177e4 LT |
3482 | EXPORT_SYMBOL(schedule); |
3483 | ||
3484 | #ifdef CONFIG_PREEMPT | |
3485 | /* | |
2ed6e34f | 3486 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3487 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3488 | * occur there and call schedule directly. | |
3489 | */ | |
3490 | asmlinkage void __sched preempt_schedule(void) | |
3491 | { | |
3492 | struct thread_info *ti = current_thread_info(); | |
3493 | #ifdef CONFIG_PREEMPT_BKL | |
3494 | struct task_struct *task = current; | |
3495 | int saved_lock_depth; | |
3496 | #endif | |
3497 | /* | |
3498 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3499 | * we do not want to preempt the current task. Just return.. | |
3500 | */ | |
beed33a8 | 3501 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3502 | return; |
3503 | ||
3504 | need_resched: | |
3505 | add_preempt_count(PREEMPT_ACTIVE); | |
3506 | /* | |
3507 | * We keep the big kernel semaphore locked, but we | |
3508 | * clear ->lock_depth so that schedule() doesnt | |
3509 | * auto-release the semaphore: | |
3510 | */ | |
3511 | #ifdef CONFIG_PREEMPT_BKL | |
3512 | saved_lock_depth = task->lock_depth; | |
3513 | task->lock_depth = -1; | |
3514 | #endif | |
3515 | schedule(); | |
3516 | #ifdef CONFIG_PREEMPT_BKL | |
3517 | task->lock_depth = saved_lock_depth; | |
3518 | #endif | |
3519 | sub_preempt_count(PREEMPT_ACTIVE); | |
3520 | ||
3521 | /* we could miss a preemption opportunity between schedule and now */ | |
3522 | barrier(); | |
3523 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3524 | goto need_resched; | |
3525 | } | |
1da177e4 LT |
3526 | EXPORT_SYMBOL(preempt_schedule); |
3527 | ||
3528 | /* | |
2ed6e34f | 3529 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3530 | * off of irq context. |
3531 | * Note, that this is called and return with irqs disabled. This will | |
3532 | * protect us against recursive calling from irq. | |
3533 | */ | |
3534 | asmlinkage void __sched preempt_schedule_irq(void) | |
3535 | { | |
3536 | struct thread_info *ti = current_thread_info(); | |
3537 | #ifdef CONFIG_PREEMPT_BKL | |
3538 | struct task_struct *task = current; | |
3539 | int saved_lock_depth; | |
3540 | #endif | |
2ed6e34f | 3541 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3542 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3543 | ||
3544 | need_resched: | |
3545 | add_preempt_count(PREEMPT_ACTIVE); | |
3546 | /* | |
3547 | * We keep the big kernel semaphore locked, but we | |
3548 | * clear ->lock_depth so that schedule() doesnt | |
3549 | * auto-release the semaphore: | |
3550 | */ | |
3551 | #ifdef CONFIG_PREEMPT_BKL | |
3552 | saved_lock_depth = task->lock_depth; | |
3553 | task->lock_depth = -1; | |
3554 | #endif | |
3555 | local_irq_enable(); | |
3556 | schedule(); | |
3557 | local_irq_disable(); | |
3558 | #ifdef CONFIG_PREEMPT_BKL | |
3559 | task->lock_depth = saved_lock_depth; | |
3560 | #endif | |
3561 | sub_preempt_count(PREEMPT_ACTIVE); | |
3562 | ||
3563 | /* we could miss a preemption opportunity between schedule and now */ | |
3564 | barrier(); | |
3565 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3566 | goto need_resched; | |
3567 | } | |
3568 | ||
3569 | #endif /* CONFIG_PREEMPT */ | |
3570 | ||
95cdf3b7 IM |
3571 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3572 | void *key) | |
1da177e4 | 3573 | { |
48f24c4d | 3574 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3575 | } |
1da177e4 LT |
3576 | EXPORT_SYMBOL(default_wake_function); |
3577 | ||
3578 | /* | |
3579 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3580 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3581 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3582 | * | |
3583 | * There are circumstances in which we can try to wake a task which has already | |
3584 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3585 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3586 | */ | |
3587 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3588 | int nr_exclusive, int sync, void *key) | |
3589 | { | |
3590 | struct list_head *tmp, *next; | |
3591 | ||
3592 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3593 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3594 | unsigned flags = curr->flags; | |
3595 | ||
1da177e4 | 3596 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3597 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3598 | break; |
3599 | } | |
3600 | } | |
3601 | ||
3602 | /** | |
3603 | * __wake_up - wake up threads blocked on a waitqueue. | |
3604 | * @q: the waitqueue | |
3605 | * @mode: which threads | |
3606 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3607 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3608 | */ |
3609 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3610 | int nr_exclusive, void *key) |
1da177e4 LT |
3611 | { |
3612 | unsigned long flags; | |
3613 | ||
3614 | spin_lock_irqsave(&q->lock, flags); | |
3615 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3616 | spin_unlock_irqrestore(&q->lock, flags); | |
3617 | } | |
1da177e4 LT |
3618 | EXPORT_SYMBOL(__wake_up); |
3619 | ||
3620 | /* | |
3621 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3622 | */ | |
3623 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3624 | { | |
3625 | __wake_up_common(q, mode, 1, 0, NULL); | |
3626 | } | |
3627 | ||
3628 | /** | |
67be2dd1 | 3629 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3630 | * @q: the waitqueue |
3631 | * @mode: which threads | |
3632 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3633 | * | |
3634 | * The sync wakeup differs that the waker knows that it will schedule | |
3635 | * away soon, so while the target thread will be woken up, it will not | |
3636 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3637 | * with each other. This can prevent needless bouncing between CPUs. | |
3638 | * | |
3639 | * On UP it can prevent extra preemption. | |
3640 | */ | |
95cdf3b7 IM |
3641 | void fastcall |
3642 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3643 | { |
3644 | unsigned long flags; | |
3645 | int sync = 1; | |
3646 | ||
3647 | if (unlikely(!q)) | |
3648 | return; | |
3649 | ||
3650 | if (unlikely(!nr_exclusive)) | |
3651 | sync = 0; | |
3652 | ||
3653 | spin_lock_irqsave(&q->lock, flags); | |
3654 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3655 | spin_unlock_irqrestore(&q->lock, flags); | |
3656 | } | |
3657 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3658 | ||
3659 | void fastcall complete(struct completion *x) | |
3660 | { | |
3661 | unsigned long flags; | |
3662 | ||
3663 | spin_lock_irqsave(&x->wait.lock, flags); | |
3664 | x->done++; | |
3665 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3666 | 1, 0, NULL); | |
3667 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3668 | } | |
3669 | EXPORT_SYMBOL(complete); | |
3670 | ||
3671 | void fastcall complete_all(struct completion *x) | |
3672 | { | |
3673 | unsigned long flags; | |
3674 | ||
3675 | spin_lock_irqsave(&x->wait.lock, flags); | |
3676 | x->done += UINT_MAX/2; | |
3677 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3678 | 0, 0, NULL); | |
3679 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3680 | } | |
3681 | EXPORT_SYMBOL(complete_all); | |
3682 | ||
3683 | void fastcall __sched wait_for_completion(struct completion *x) | |
3684 | { | |
3685 | might_sleep(); | |
48f24c4d | 3686 | |
1da177e4 LT |
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 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3695 | spin_unlock_irq(&x->wait.lock); | |
3696 | schedule(); | |
3697 | spin_lock_irq(&x->wait.lock); | |
3698 | } while (!x->done); | |
3699 | __remove_wait_queue(&x->wait, &wait); | |
3700 | } | |
3701 | x->done--; | |
3702 | spin_unlock_irq(&x->wait.lock); | |
3703 | } | |
3704 | EXPORT_SYMBOL(wait_for_completion); | |
3705 | ||
3706 | unsigned long fastcall __sched | |
3707 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3708 | { | |
3709 | might_sleep(); | |
3710 | ||
3711 | spin_lock_irq(&x->wait.lock); | |
3712 | if (!x->done) { | |
3713 | DECLARE_WAITQUEUE(wait, current); | |
3714 | ||
3715 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3716 | __add_wait_queue_tail(&x->wait, &wait); | |
3717 | do { | |
3718 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3719 | spin_unlock_irq(&x->wait.lock); | |
3720 | timeout = schedule_timeout(timeout); | |
3721 | spin_lock_irq(&x->wait.lock); | |
3722 | if (!timeout) { | |
3723 | __remove_wait_queue(&x->wait, &wait); | |
3724 | goto out; | |
3725 | } | |
3726 | } while (!x->done); | |
3727 | __remove_wait_queue(&x->wait, &wait); | |
3728 | } | |
3729 | x->done--; | |
3730 | out: | |
3731 | spin_unlock_irq(&x->wait.lock); | |
3732 | return timeout; | |
3733 | } | |
3734 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3735 | ||
3736 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3737 | { | |
3738 | int ret = 0; | |
3739 | ||
3740 | might_sleep(); | |
3741 | ||
3742 | spin_lock_irq(&x->wait.lock); | |
3743 | if (!x->done) { | |
3744 | DECLARE_WAITQUEUE(wait, current); | |
3745 | ||
3746 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3747 | __add_wait_queue_tail(&x->wait, &wait); | |
3748 | do { | |
3749 | if (signal_pending(current)) { | |
3750 | ret = -ERESTARTSYS; | |
3751 | __remove_wait_queue(&x->wait, &wait); | |
3752 | goto out; | |
3753 | } | |
3754 | __set_current_state(TASK_INTERRUPTIBLE); | |
3755 | spin_unlock_irq(&x->wait.lock); | |
3756 | schedule(); | |
3757 | spin_lock_irq(&x->wait.lock); | |
3758 | } while (!x->done); | |
3759 | __remove_wait_queue(&x->wait, &wait); | |
3760 | } | |
3761 | x->done--; | |
3762 | out: | |
3763 | spin_unlock_irq(&x->wait.lock); | |
3764 | ||
3765 | return ret; | |
3766 | } | |
3767 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3768 | ||
3769 | unsigned long fastcall __sched | |
3770 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3771 | unsigned long timeout) | |
3772 | { | |
3773 | might_sleep(); | |
3774 | ||
3775 | spin_lock_irq(&x->wait.lock); | |
3776 | if (!x->done) { | |
3777 | DECLARE_WAITQUEUE(wait, current); | |
3778 | ||
3779 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3780 | __add_wait_queue_tail(&x->wait, &wait); | |
3781 | do { | |
3782 | if (signal_pending(current)) { | |
3783 | timeout = -ERESTARTSYS; | |
3784 | __remove_wait_queue(&x->wait, &wait); | |
3785 | goto out; | |
3786 | } | |
3787 | __set_current_state(TASK_INTERRUPTIBLE); | |
3788 | spin_unlock_irq(&x->wait.lock); | |
3789 | timeout = schedule_timeout(timeout); | |
3790 | spin_lock_irq(&x->wait.lock); | |
3791 | if (!timeout) { | |
3792 | __remove_wait_queue(&x->wait, &wait); | |
3793 | goto out; | |
3794 | } | |
3795 | } while (!x->done); | |
3796 | __remove_wait_queue(&x->wait, &wait); | |
3797 | } | |
3798 | x->done--; | |
3799 | out: | |
3800 | spin_unlock_irq(&x->wait.lock); | |
3801 | return timeout; | |
3802 | } | |
3803 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3804 | ||
0fec171c IM |
3805 | static inline void |
3806 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3807 | { | |
3808 | spin_lock_irqsave(&q->lock, *flags); | |
3809 | __add_wait_queue(q, wait); | |
1da177e4 | 3810 | spin_unlock(&q->lock); |
0fec171c | 3811 | } |
1da177e4 | 3812 | |
0fec171c IM |
3813 | static inline void |
3814 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3815 | { | |
3816 | spin_lock_irq(&q->lock); | |
3817 | __remove_wait_queue(q, wait); | |
3818 | spin_unlock_irqrestore(&q->lock, *flags); | |
3819 | } | |
1da177e4 | 3820 | |
0fec171c | 3821 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3822 | { |
0fec171c IM |
3823 | unsigned long flags; |
3824 | wait_queue_t wait; | |
3825 | ||
3826 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3827 | |
3828 | current->state = TASK_INTERRUPTIBLE; | |
3829 | ||
0fec171c | 3830 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3831 | schedule(); |
0fec171c | 3832 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3833 | } |
1da177e4 LT |
3834 | EXPORT_SYMBOL(interruptible_sleep_on); |
3835 | ||
0fec171c | 3836 | long __sched |
95cdf3b7 | 3837 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3838 | { |
0fec171c IM |
3839 | unsigned long flags; |
3840 | wait_queue_t wait; | |
3841 | ||
3842 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3843 | |
3844 | current->state = TASK_INTERRUPTIBLE; | |
3845 | ||
0fec171c | 3846 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3847 | timeout = schedule_timeout(timeout); |
0fec171c | 3848 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3849 | |
3850 | return timeout; | |
3851 | } | |
1da177e4 LT |
3852 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3853 | ||
0fec171c | 3854 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3855 | { |
0fec171c IM |
3856 | unsigned long flags; |
3857 | wait_queue_t wait; | |
3858 | ||
3859 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3860 | |
3861 | current->state = TASK_UNINTERRUPTIBLE; | |
3862 | ||
0fec171c | 3863 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3864 | schedule(); |
0fec171c | 3865 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3866 | } |
1da177e4 LT |
3867 | EXPORT_SYMBOL(sleep_on); |
3868 | ||
0fec171c | 3869 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3870 | { |
0fec171c IM |
3871 | unsigned long flags; |
3872 | wait_queue_t wait; | |
3873 | ||
3874 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3875 | |
3876 | current->state = TASK_UNINTERRUPTIBLE; | |
3877 | ||
0fec171c | 3878 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3879 | timeout = schedule_timeout(timeout); |
0fec171c | 3880 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3881 | |
3882 | return timeout; | |
3883 | } | |
1da177e4 LT |
3884 | EXPORT_SYMBOL(sleep_on_timeout); |
3885 | ||
b29739f9 IM |
3886 | #ifdef CONFIG_RT_MUTEXES |
3887 | ||
3888 | /* | |
3889 | * rt_mutex_setprio - set the current priority of a task | |
3890 | * @p: task | |
3891 | * @prio: prio value (kernel-internal form) | |
3892 | * | |
3893 | * This function changes the 'effective' priority of a task. It does | |
3894 | * not touch ->normal_prio like __setscheduler(). | |
3895 | * | |
3896 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3897 | */ | |
36c8b586 | 3898 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3899 | { |
3900 | unsigned long flags; | |
dd41f596 | 3901 | int oldprio, on_rq; |
70b97a7f | 3902 | struct rq *rq; |
dd41f596 | 3903 | u64 now; |
b29739f9 IM |
3904 | |
3905 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3906 | ||
3907 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3908 | now = rq_clock(rq); |
b29739f9 | 3909 | |
d5f9f942 | 3910 | oldprio = p->prio; |
dd41f596 IM |
3911 | on_rq = p->se.on_rq; |
3912 | if (on_rq) | |
3913 | dequeue_task(rq, p, 0, now); | |
3914 | ||
3915 | if (rt_prio(prio)) | |
3916 | p->sched_class = &rt_sched_class; | |
3917 | else | |
3918 | p->sched_class = &fair_sched_class; | |
3919 | ||
b29739f9 IM |
3920 | p->prio = prio; |
3921 | ||
dd41f596 IM |
3922 | if (on_rq) { |
3923 | enqueue_task(rq, p, 0, now); | |
b29739f9 IM |
3924 | /* |
3925 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3926 | * our priority decreased, or if we are not currently running on |
3927 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3928 | */ |
d5f9f942 AM |
3929 | if (task_running(rq, p)) { |
3930 | if (p->prio > oldprio) | |
3931 | resched_task(rq->curr); | |
dd41f596 IM |
3932 | } else { |
3933 | check_preempt_curr(rq, p); | |
3934 | } | |
b29739f9 IM |
3935 | } |
3936 | task_rq_unlock(rq, &flags); | |
3937 | } | |
3938 | ||
3939 | #endif | |
3940 | ||
36c8b586 | 3941 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3942 | { |
dd41f596 | 3943 | int old_prio, delta, on_rq; |
1da177e4 | 3944 | unsigned long flags; |
70b97a7f | 3945 | struct rq *rq; |
dd41f596 | 3946 | u64 now; |
1da177e4 LT |
3947 | |
3948 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3949 | return; | |
3950 | /* | |
3951 | * We have to be careful, if called from sys_setpriority(), | |
3952 | * the task might be in the middle of scheduling on another CPU. | |
3953 | */ | |
3954 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3955 | now = rq_clock(rq); |
1da177e4 LT |
3956 | /* |
3957 | * The RT priorities are set via sched_setscheduler(), but we still | |
3958 | * allow the 'normal' nice value to be set - but as expected | |
3959 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3960 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3961 | */ |
e05606d3 | 3962 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3963 | p->static_prio = NICE_TO_PRIO(nice); |
3964 | goto out_unlock; | |
3965 | } | |
dd41f596 IM |
3966 | on_rq = p->se.on_rq; |
3967 | if (on_rq) { | |
3968 | dequeue_task(rq, p, 0, now); | |
3969 | dec_load(rq, p, now); | |
2dd73a4f | 3970 | } |
1da177e4 | 3971 | |
1da177e4 | 3972 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3973 | set_load_weight(p); |
b29739f9 IM |
3974 | old_prio = p->prio; |
3975 | p->prio = effective_prio(p); | |
3976 | delta = p->prio - old_prio; | |
1da177e4 | 3977 | |
dd41f596 IM |
3978 | if (on_rq) { |
3979 | enqueue_task(rq, p, 0, now); | |
3980 | inc_load(rq, p, now); | |
1da177e4 | 3981 | /* |
d5f9f942 AM |
3982 | * If the task increased its priority or is running and |
3983 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3984 | */ |
d5f9f942 | 3985 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3986 | resched_task(rq->curr); |
3987 | } | |
3988 | out_unlock: | |
3989 | task_rq_unlock(rq, &flags); | |
3990 | } | |
1da177e4 LT |
3991 | EXPORT_SYMBOL(set_user_nice); |
3992 | ||
e43379f1 MM |
3993 | /* |
3994 | * can_nice - check if a task can reduce its nice value | |
3995 | * @p: task | |
3996 | * @nice: nice value | |
3997 | */ | |
36c8b586 | 3998 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3999 | { |
024f4747 MM |
4000 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4001 | int nice_rlim = 20 - nice; | |
48f24c4d | 4002 | |
e43379f1 MM |
4003 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4004 | capable(CAP_SYS_NICE)); | |
4005 | } | |
4006 | ||
1da177e4 LT |
4007 | #ifdef __ARCH_WANT_SYS_NICE |
4008 | ||
4009 | /* | |
4010 | * sys_nice - change the priority of the current process. | |
4011 | * @increment: priority increment | |
4012 | * | |
4013 | * sys_setpriority is a more generic, but much slower function that | |
4014 | * does similar things. | |
4015 | */ | |
4016 | asmlinkage long sys_nice(int increment) | |
4017 | { | |
48f24c4d | 4018 | long nice, retval; |
1da177e4 LT |
4019 | |
4020 | /* | |
4021 | * Setpriority might change our priority at the same moment. | |
4022 | * We don't have to worry. Conceptually one call occurs first | |
4023 | * and we have a single winner. | |
4024 | */ | |
e43379f1 MM |
4025 | if (increment < -40) |
4026 | increment = -40; | |
1da177e4 LT |
4027 | if (increment > 40) |
4028 | increment = 40; | |
4029 | ||
4030 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4031 | if (nice < -20) | |
4032 | nice = -20; | |
4033 | if (nice > 19) | |
4034 | nice = 19; | |
4035 | ||
e43379f1 MM |
4036 | if (increment < 0 && !can_nice(current, nice)) |
4037 | return -EPERM; | |
4038 | ||
1da177e4 LT |
4039 | retval = security_task_setnice(current, nice); |
4040 | if (retval) | |
4041 | return retval; | |
4042 | ||
4043 | set_user_nice(current, nice); | |
4044 | return 0; | |
4045 | } | |
4046 | ||
4047 | #endif | |
4048 | ||
4049 | /** | |
4050 | * task_prio - return the priority value of a given task. | |
4051 | * @p: the task in question. | |
4052 | * | |
4053 | * This is the priority value as seen by users in /proc. | |
4054 | * RT tasks are offset by -200. Normal tasks are centered | |
4055 | * around 0, value goes from -16 to +15. | |
4056 | */ | |
36c8b586 | 4057 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4058 | { |
4059 | return p->prio - MAX_RT_PRIO; | |
4060 | } | |
4061 | ||
4062 | /** | |
4063 | * task_nice - return the nice value of a given task. | |
4064 | * @p: the task in question. | |
4065 | */ | |
36c8b586 | 4066 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4067 | { |
4068 | return TASK_NICE(p); | |
4069 | } | |
1da177e4 | 4070 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4071 | |
4072 | /** | |
4073 | * idle_cpu - is a given cpu idle currently? | |
4074 | * @cpu: the processor in question. | |
4075 | */ | |
4076 | int idle_cpu(int cpu) | |
4077 | { | |
4078 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4079 | } | |
4080 | ||
1da177e4 LT |
4081 | /** |
4082 | * idle_task - return the idle task for a given cpu. | |
4083 | * @cpu: the processor in question. | |
4084 | */ | |
36c8b586 | 4085 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4086 | { |
4087 | return cpu_rq(cpu)->idle; | |
4088 | } | |
4089 | ||
4090 | /** | |
4091 | * find_process_by_pid - find a process with a matching PID value. | |
4092 | * @pid: the pid in question. | |
4093 | */ | |
36c8b586 | 4094 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4095 | { |
4096 | return pid ? find_task_by_pid(pid) : current; | |
4097 | } | |
4098 | ||
4099 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4100 | static void |
4101 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4102 | { |
dd41f596 | 4103 | BUG_ON(p->se.on_rq); |
48f24c4d | 4104 | |
1da177e4 | 4105 | p->policy = policy; |
dd41f596 IM |
4106 | switch (p->policy) { |
4107 | case SCHED_NORMAL: | |
4108 | case SCHED_BATCH: | |
4109 | case SCHED_IDLE: | |
4110 | p->sched_class = &fair_sched_class; | |
4111 | break; | |
4112 | case SCHED_FIFO: | |
4113 | case SCHED_RR: | |
4114 | p->sched_class = &rt_sched_class; | |
4115 | break; | |
4116 | } | |
4117 | ||
1da177e4 | 4118 | p->rt_priority = prio; |
b29739f9 IM |
4119 | p->normal_prio = normal_prio(p); |
4120 | /* we are holding p->pi_lock already */ | |
4121 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4122 | set_load_weight(p); |
1da177e4 LT |
4123 | } |
4124 | ||
4125 | /** | |
72fd4a35 | 4126 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4127 | * @p: the task in question. |
4128 | * @policy: new policy. | |
4129 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4130 | * |
72fd4a35 | 4131 | * NOTE that the task may be already dead. |
1da177e4 | 4132 | */ |
95cdf3b7 IM |
4133 | int sched_setscheduler(struct task_struct *p, int policy, |
4134 | struct sched_param *param) | |
1da177e4 | 4135 | { |
dd41f596 | 4136 | int retval, oldprio, oldpolicy = -1, on_rq; |
1da177e4 | 4137 | unsigned long flags; |
70b97a7f | 4138 | struct rq *rq; |
1da177e4 | 4139 | |
66e5393a SR |
4140 | /* may grab non-irq protected spin_locks */ |
4141 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4142 | recheck: |
4143 | /* double check policy once rq lock held */ | |
4144 | if (policy < 0) | |
4145 | policy = oldpolicy = p->policy; | |
4146 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4147 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4148 | policy != SCHED_IDLE) | |
b0a9499c | 4149 | return -EINVAL; |
1da177e4 LT |
4150 | /* |
4151 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4152 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4153 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4154 | */ |
4155 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4156 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4157 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4158 | return -EINVAL; |
e05606d3 | 4159 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4160 | return -EINVAL; |
4161 | ||
37e4ab3f OC |
4162 | /* |
4163 | * Allow unprivileged RT tasks to decrease priority: | |
4164 | */ | |
4165 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4166 | if (rt_policy(policy)) { |
8dc3e909 | 4167 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4168 | |
4169 | if (!lock_task_sighand(p, &flags)) | |
4170 | return -ESRCH; | |
4171 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4172 | unlock_task_sighand(p, &flags); | |
4173 | ||
4174 | /* can't set/change the rt policy */ | |
4175 | if (policy != p->policy && !rlim_rtprio) | |
4176 | return -EPERM; | |
4177 | ||
4178 | /* can't increase priority */ | |
4179 | if (param->sched_priority > p->rt_priority && | |
4180 | param->sched_priority > rlim_rtprio) | |
4181 | return -EPERM; | |
4182 | } | |
dd41f596 IM |
4183 | /* |
4184 | * Like positive nice levels, dont allow tasks to | |
4185 | * move out of SCHED_IDLE either: | |
4186 | */ | |
4187 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4188 | return -EPERM; | |
5fe1d75f | 4189 | |
37e4ab3f OC |
4190 | /* can't change other user's priorities */ |
4191 | if ((current->euid != p->euid) && | |
4192 | (current->euid != p->uid)) | |
4193 | return -EPERM; | |
4194 | } | |
1da177e4 LT |
4195 | |
4196 | retval = security_task_setscheduler(p, policy, param); | |
4197 | if (retval) | |
4198 | return retval; | |
b29739f9 IM |
4199 | /* |
4200 | * make sure no PI-waiters arrive (or leave) while we are | |
4201 | * changing the priority of the task: | |
4202 | */ | |
4203 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4204 | /* |
4205 | * To be able to change p->policy safely, the apropriate | |
4206 | * runqueue lock must be held. | |
4207 | */ | |
b29739f9 | 4208 | rq = __task_rq_lock(p); |
1da177e4 LT |
4209 | /* recheck policy now with rq lock held */ |
4210 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4211 | policy = oldpolicy = -1; | |
b29739f9 IM |
4212 | __task_rq_unlock(rq); |
4213 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4214 | goto recheck; |
4215 | } | |
dd41f596 IM |
4216 | on_rq = p->se.on_rq; |
4217 | if (on_rq) | |
8e717b19 | 4218 | deactivate_task(rq, p, 0, rq_clock(rq)); |
1da177e4 | 4219 | oldprio = p->prio; |
dd41f596 IM |
4220 | __setscheduler(rq, p, policy, param->sched_priority); |
4221 | if (on_rq) { | |
4222 | activate_task(rq, p, 0); | |
1da177e4 LT |
4223 | /* |
4224 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4225 | * our priority decreased, or if we are not currently running on |
4226 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4227 | */ |
d5f9f942 AM |
4228 | if (task_running(rq, p)) { |
4229 | if (p->prio > oldprio) | |
4230 | resched_task(rq->curr); | |
dd41f596 IM |
4231 | } else { |
4232 | check_preempt_curr(rq, p); | |
4233 | } | |
1da177e4 | 4234 | } |
b29739f9 IM |
4235 | __task_rq_unlock(rq); |
4236 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4237 | ||
95e02ca9 TG |
4238 | rt_mutex_adjust_pi(p); |
4239 | ||
1da177e4 LT |
4240 | return 0; |
4241 | } | |
4242 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4243 | ||
95cdf3b7 IM |
4244 | static int |
4245 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4246 | { |
1da177e4 LT |
4247 | struct sched_param lparam; |
4248 | struct task_struct *p; | |
36c8b586 | 4249 | int retval; |
1da177e4 LT |
4250 | |
4251 | if (!param || pid < 0) | |
4252 | return -EINVAL; | |
4253 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4254 | return -EFAULT; | |
5fe1d75f ON |
4255 | |
4256 | rcu_read_lock(); | |
4257 | retval = -ESRCH; | |
1da177e4 | 4258 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4259 | if (p != NULL) |
4260 | retval = sched_setscheduler(p, policy, &lparam); | |
4261 | rcu_read_unlock(); | |
36c8b586 | 4262 | |
1da177e4 LT |
4263 | return retval; |
4264 | } | |
4265 | ||
4266 | /** | |
4267 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4268 | * @pid: the pid in question. | |
4269 | * @policy: new policy. | |
4270 | * @param: structure containing the new RT priority. | |
4271 | */ | |
4272 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4273 | struct sched_param __user *param) | |
4274 | { | |
c21761f1 JB |
4275 | /* negative values for policy are not valid */ |
4276 | if (policy < 0) | |
4277 | return -EINVAL; | |
4278 | ||
1da177e4 LT |
4279 | return do_sched_setscheduler(pid, policy, param); |
4280 | } | |
4281 | ||
4282 | /** | |
4283 | * sys_sched_setparam - set/change the RT priority of a thread | |
4284 | * @pid: the pid in question. | |
4285 | * @param: structure containing the new RT priority. | |
4286 | */ | |
4287 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4288 | { | |
4289 | return do_sched_setscheduler(pid, -1, param); | |
4290 | } | |
4291 | ||
4292 | /** | |
4293 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4294 | * @pid: the pid in question. | |
4295 | */ | |
4296 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4297 | { | |
36c8b586 | 4298 | struct task_struct *p; |
1da177e4 | 4299 | int retval = -EINVAL; |
1da177e4 LT |
4300 | |
4301 | if (pid < 0) | |
4302 | goto out_nounlock; | |
4303 | ||
4304 | retval = -ESRCH; | |
4305 | read_lock(&tasklist_lock); | |
4306 | p = find_process_by_pid(pid); | |
4307 | if (p) { | |
4308 | retval = security_task_getscheduler(p); | |
4309 | if (!retval) | |
4310 | retval = p->policy; | |
4311 | } | |
4312 | read_unlock(&tasklist_lock); | |
4313 | ||
4314 | out_nounlock: | |
4315 | return retval; | |
4316 | } | |
4317 | ||
4318 | /** | |
4319 | * sys_sched_getscheduler - get the RT priority of a thread | |
4320 | * @pid: the pid in question. | |
4321 | * @param: structure containing the RT priority. | |
4322 | */ | |
4323 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4324 | { | |
4325 | struct sched_param lp; | |
36c8b586 | 4326 | struct task_struct *p; |
1da177e4 | 4327 | int retval = -EINVAL; |
1da177e4 LT |
4328 | |
4329 | if (!param || pid < 0) | |
4330 | goto out_nounlock; | |
4331 | ||
4332 | read_lock(&tasklist_lock); | |
4333 | p = find_process_by_pid(pid); | |
4334 | retval = -ESRCH; | |
4335 | if (!p) | |
4336 | goto out_unlock; | |
4337 | ||
4338 | retval = security_task_getscheduler(p); | |
4339 | if (retval) | |
4340 | goto out_unlock; | |
4341 | ||
4342 | lp.sched_priority = p->rt_priority; | |
4343 | read_unlock(&tasklist_lock); | |
4344 | ||
4345 | /* | |
4346 | * This one might sleep, we cannot do it with a spinlock held ... | |
4347 | */ | |
4348 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4349 | ||
4350 | out_nounlock: | |
4351 | return retval; | |
4352 | ||
4353 | out_unlock: | |
4354 | read_unlock(&tasklist_lock); | |
4355 | return retval; | |
4356 | } | |
4357 | ||
4358 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4359 | { | |
1da177e4 | 4360 | cpumask_t cpus_allowed; |
36c8b586 IM |
4361 | struct task_struct *p; |
4362 | int retval; | |
1da177e4 | 4363 | |
5be9361c | 4364 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4365 | read_lock(&tasklist_lock); |
4366 | ||
4367 | p = find_process_by_pid(pid); | |
4368 | if (!p) { | |
4369 | read_unlock(&tasklist_lock); | |
5be9361c | 4370 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4371 | return -ESRCH; |
4372 | } | |
4373 | ||
4374 | /* | |
4375 | * It is not safe to call set_cpus_allowed with the | |
4376 | * tasklist_lock held. We will bump the task_struct's | |
4377 | * usage count and then drop tasklist_lock. | |
4378 | */ | |
4379 | get_task_struct(p); | |
4380 | read_unlock(&tasklist_lock); | |
4381 | ||
4382 | retval = -EPERM; | |
4383 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4384 | !capable(CAP_SYS_NICE)) | |
4385 | goto out_unlock; | |
4386 | ||
e7834f8f DQ |
4387 | retval = security_task_setscheduler(p, 0, NULL); |
4388 | if (retval) | |
4389 | goto out_unlock; | |
4390 | ||
1da177e4 LT |
4391 | cpus_allowed = cpuset_cpus_allowed(p); |
4392 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4393 | retval = set_cpus_allowed(p, new_mask); | |
4394 | ||
4395 | out_unlock: | |
4396 | put_task_struct(p); | |
5be9361c | 4397 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4398 | return retval; |
4399 | } | |
4400 | ||
4401 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4402 | cpumask_t *new_mask) | |
4403 | { | |
4404 | if (len < sizeof(cpumask_t)) { | |
4405 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4406 | } else if (len > sizeof(cpumask_t)) { | |
4407 | len = sizeof(cpumask_t); | |
4408 | } | |
4409 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4410 | } | |
4411 | ||
4412 | /** | |
4413 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4414 | * @pid: pid of the process | |
4415 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4416 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4417 | */ | |
4418 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4419 | unsigned long __user *user_mask_ptr) | |
4420 | { | |
4421 | cpumask_t new_mask; | |
4422 | int retval; | |
4423 | ||
4424 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4425 | if (retval) | |
4426 | return retval; | |
4427 | ||
4428 | return sched_setaffinity(pid, new_mask); | |
4429 | } | |
4430 | ||
4431 | /* | |
4432 | * Represents all cpu's present in the system | |
4433 | * In systems capable of hotplug, this map could dynamically grow | |
4434 | * as new cpu's are detected in the system via any platform specific | |
4435 | * method, such as ACPI for e.g. | |
4436 | */ | |
4437 | ||
4cef0c61 | 4438 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4439 | EXPORT_SYMBOL(cpu_present_map); |
4440 | ||
4441 | #ifndef CONFIG_SMP | |
4cef0c61 | 4442 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4443 | EXPORT_SYMBOL(cpu_online_map); |
4444 | ||
4cef0c61 | 4445 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4446 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4447 | #endif |
4448 | ||
4449 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4450 | { | |
36c8b586 | 4451 | struct task_struct *p; |
1da177e4 | 4452 | int retval; |
1da177e4 | 4453 | |
5be9361c | 4454 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4455 | read_lock(&tasklist_lock); |
4456 | ||
4457 | retval = -ESRCH; | |
4458 | p = find_process_by_pid(pid); | |
4459 | if (!p) | |
4460 | goto out_unlock; | |
4461 | ||
e7834f8f DQ |
4462 | retval = security_task_getscheduler(p); |
4463 | if (retval) | |
4464 | goto out_unlock; | |
4465 | ||
2f7016d9 | 4466 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4467 | |
4468 | out_unlock: | |
4469 | read_unlock(&tasklist_lock); | |
5be9361c | 4470 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4471 | |
9531b62f | 4472 | return retval; |
1da177e4 LT |
4473 | } |
4474 | ||
4475 | /** | |
4476 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4477 | * @pid: pid of the process | |
4478 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4479 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4480 | */ | |
4481 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4482 | unsigned long __user *user_mask_ptr) | |
4483 | { | |
4484 | int ret; | |
4485 | cpumask_t mask; | |
4486 | ||
4487 | if (len < sizeof(cpumask_t)) | |
4488 | return -EINVAL; | |
4489 | ||
4490 | ret = sched_getaffinity(pid, &mask); | |
4491 | if (ret < 0) | |
4492 | return ret; | |
4493 | ||
4494 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4495 | return -EFAULT; | |
4496 | ||
4497 | return sizeof(cpumask_t); | |
4498 | } | |
4499 | ||
4500 | /** | |
4501 | * sys_sched_yield - yield the current processor to other threads. | |
4502 | * | |
dd41f596 IM |
4503 | * This function yields the current CPU to other tasks. If there are no |
4504 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4505 | */ |
4506 | asmlinkage long sys_sched_yield(void) | |
4507 | { | |
70b97a7f | 4508 | struct rq *rq = this_rq_lock(); |
1da177e4 LT |
4509 | |
4510 | schedstat_inc(rq, yld_cnt); | |
dd41f596 | 4511 | if (unlikely(rq->nr_running == 1)) |
1da177e4 | 4512 | schedstat_inc(rq, yld_act_empty); |
dd41f596 IM |
4513 | else |
4514 | current->sched_class->yield_task(rq, current); | |
1da177e4 LT |
4515 | |
4516 | /* | |
4517 | * Since we are going to call schedule() anyway, there's | |
4518 | * no need to preempt or enable interrupts: | |
4519 | */ | |
4520 | __release(rq->lock); | |
8a25d5de | 4521 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4522 | _raw_spin_unlock(&rq->lock); |
4523 | preempt_enable_no_resched(); | |
4524 | ||
4525 | schedule(); | |
4526 | ||
4527 | return 0; | |
4528 | } | |
4529 | ||
e7b38404 | 4530 | static void __cond_resched(void) |
1da177e4 | 4531 | { |
8e0a43d8 IM |
4532 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4533 | __might_sleep(__FILE__, __LINE__); | |
4534 | #endif | |
5bbcfd90 IM |
4535 | /* |
4536 | * The BKS might be reacquired before we have dropped | |
4537 | * PREEMPT_ACTIVE, which could trigger a second | |
4538 | * cond_resched() call. | |
4539 | */ | |
1da177e4 LT |
4540 | do { |
4541 | add_preempt_count(PREEMPT_ACTIVE); | |
4542 | schedule(); | |
4543 | sub_preempt_count(PREEMPT_ACTIVE); | |
4544 | } while (need_resched()); | |
4545 | } | |
4546 | ||
4547 | int __sched cond_resched(void) | |
4548 | { | |
9414232f IM |
4549 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4550 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4551 | __cond_resched(); |
4552 | return 1; | |
4553 | } | |
4554 | return 0; | |
4555 | } | |
1da177e4 LT |
4556 | EXPORT_SYMBOL(cond_resched); |
4557 | ||
4558 | /* | |
4559 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4560 | * call schedule, and on return reacquire the lock. | |
4561 | * | |
4562 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4563 | * operations here to prevent schedule() from being called twice (once via | |
4564 | * spin_unlock(), once by hand). | |
4565 | */ | |
95cdf3b7 | 4566 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4567 | { |
6df3cecb JK |
4568 | int ret = 0; |
4569 | ||
1da177e4 LT |
4570 | if (need_lockbreak(lock)) { |
4571 | spin_unlock(lock); | |
4572 | cpu_relax(); | |
6df3cecb | 4573 | ret = 1; |
1da177e4 LT |
4574 | spin_lock(lock); |
4575 | } | |
9414232f | 4576 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4577 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4578 | _raw_spin_unlock(lock); |
4579 | preempt_enable_no_resched(); | |
4580 | __cond_resched(); | |
6df3cecb | 4581 | ret = 1; |
1da177e4 | 4582 | spin_lock(lock); |
1da177e4 | 4583 | } |
6df3cecb | 4584 | return ret; |
1da177e4 | 4585 | } |
1da177e4 LT |
4586 | EXPORT_SYMBOL(cond_resched_lock); |
4587 | ||
4588 | int __sched cond_resched_softirq(void) | |
4589 | { | |
4590 | BUG_ON(!in_softirq()); | |
4591 | ||
9414232f | 4592 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4593 | local_bh_enable(); |
1da177e4 LT |
4594 | __cond_resched(); |
4595 | local_bh_disable(); | |
4596 | return 1; | |
4597 | } | |
4598 | return 0; | |
4599 | } | |
1da177e4 LT |
4600 | EXPORT_SYMBOL(cond_resched_softirq); |
4601 | ||
1da177e4 LT |
4602 | /** |
4603 | * yield - yield the current processor to other threads. | |
4604 | * | |
72fd4a35 | 4605 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4606 | * thread runnable and calls sys_sched_yield(). |
4607 | */ | |
4608 | void __sched yield(void) | |
4609 | { | |
4610 | set_current_state(TASK_RUNNING); | |
4611 | sys_sched_yield(); | |
4612 | } | |
1da177e4 LT |
4613 | EXPORT_SYMBOL(yield); |
4614 | ||
4615 | /* | |
4616 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4617 | * that process accounting knows that this is a task in IO wait state. | |
4618 | * | |
4619 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4620 | * has set its backing_dev_info: the queue against which it should throttle) | |
4621 | */ | |
4622 | void __sched io_schedule(void) | |
4623 | { | |
70b97a7f | 4624 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4625 | |
0ff92245 | 4626 | delayacct_blkio_start(); |
1da177e4 LT |
4627 | atomic_inc(&rq->nr_iowait); |
4628 | schedule(); | |
4629 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4630 | delayacct_blkio_end(); |
1da177e4 | 4631 | } |
1da177e4 LT |
4632 | EXPORT_SYMBOL(io_schedule); |
4633 | ||
4634 | long __sched io_schedule_timeout(long timeout) | |
4635 | { | |
70b97a7f | 4636 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4637 | long ret; |
4638 | ||
0ff92245 | 4639 | delayacct_blkio_start(); |
1da177e4 LT |
4640 | atomic_inc(&rq->nr_iowait); |
4641 | ret = schedule_timeout(timeout); | |
4642 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4643 | delayacct_blkio_end(); |
1da177e4 LT |
4644 | return ret; |
4645 | } | |
4646 | ||
4647 | /** | |
4648 | * sys_sched_get_priority_max - return maximum RT priority. | |
4649 | * @policy: scheduling class. | |
4650 | * | |
4651 | * this syscall returns the maximum rt_priority that can be used | |
4652 | * by a given scheduling class. | |
4653 | */ | |
4654 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4655 | { | |
4656 | int ret = -EINVAL; | |
4657 | ||
4658 | switch (policy) { | |
4659 | case SCHED_FIFO: | |
4660 | case SCHED_RR: | |
4661 | ret = MAX_USER_RT_PRIO-1; | |
4662 | break; | |
4663 | case SCHED_NORMAL: | |
b0a9499c | 4664 | case SCHED_BATCH: |
dd41f596 | 4665 | case SCHED_IDLE: |
1da177e4 LT |
4666 | ret = 0; |
4667 | break; | |
4668 | } | |
4669 | return ret; | |
4670 | } | |
4671 | ||
4672 | /** | |
4673 | * sys_sched_get_priority_min - return minimum RT priority. | |
4674 | * @policy: scheduling class. | |
4675 | * | |
4676 | * this syscall returns the minimum rt_priority that can be used | |
4677 | * by a given scheduling class. | |
4678 | */ | |
4679 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4680 | { | |
4681 | int ret = -EINVAL; | |
4682 | ||
4683 | switch (policy) { | |
4684 | case SCHED_FIFO: | |
4685 | case SCHED_RR: | |
4686 | ret = 1; | |
4687 | break; | |
4688 | case SCHED_NORMAL: | |
b0a9499c | 4689 | case SCHED_BATCH: |
dd41f596 | 4690 | case SCHED_IDLE: |
1da177e4 LT |
4691 | ret = 0; |
4692 | } | |
4693 | return ret; | |
4694 | } | |
4695 | ||
4696 | /** | |
4697 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4698 | * @pid: pid of the process. | |
4699 | * @interval: userspace pointer to the timeslice value. | |
4700 | * | |
4701 | * this syscall writes the default timeslice value of a given process | |
4702 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4703 | */ | |
4704 | asmlinkage | |
4705 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4706 | { | |
36c8b586 | 4707 | struct task_struct *p; |
1da177e4 LT |
4708 | int retval = -EINVAL; |
4709 | struct timespec t; | |
1da177e4 LT |
4710 | |
4711 | if (pid < 0) | |
4712 | goto out_nounlock; | |
4713 | ||
4714 | retval = -ESRCH; | |
4715 | read_lock(&tasklist_lock); | |
4716 | p = find_process_by_pid(pid); | |
4717 | if (!p) | |
4718 | goto out_unlock; | |
4719 | ||
4720 | retval = security_task_getscheduler(p); | |
4721 | if (retval) | |
4722 | goto out_unlock; | |
4723 | ||
b78709cf | 4724 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
dd41f596 | 4725 | 0 : static_prio_timeslice(p->static_prio), &t); |
1da177e4 LT |
4726 | read_unlock(&tasklist_lock); |
4727 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4728 | out_nounlock: | |
4729 | return retval; | |
4730 | out_unlock: | |
4731 | read_unlock(&tasklist_lock); | |
4732 | return retval; | |
4733 | } | |
4734 | ||
2ed6e34f | 4735 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4736 | |
4737 | static void show_task(struct task_struct *p) | |
1da177e4 | 4738 | { |
1da177e4 | 4739 | unsigned long free = 0; |
36c8b586 | 4740 | unsigned state; |
1da177e4 | 4741 | |
1da177e4 | 4742 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4743 | printk("%-13.13s %c", p->comm, |
4744 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4745 | #if BITS_PER_LONG == 32 |
1da177e4 | 4746 | if (state == TASK_RUNNING) |
4bd77321 | 4747 | printk(" running "); |
1da177e4 | 4748 | else |
4bd77321 | 4749 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4750 | #else |
4751 | if (state == TASK_RUNNING) | |
4bd77321 | 4752 | printk(" running task "); |
1da177e4 LT |
4753 | else |
4754 | printk(" %016lx ", thread_saved_pc(p)); | |
4755 | #endif | |
4756 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4757 | { | |
10ebffde | 4758 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4759 | while (!*n) |
4760 | n++; | |
10ebffde | 4761 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4762 | } |
4763 | #endif | |
4bd77321 | 4764 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4765 | |
4766 | if (state != TASK_RUNNING) | |
4767 | show_stack(p, NULL); | |
4768 | } | |
4769 | ||
e59e2ae2 | 4770 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4771 | { |
36c8b586 | 4772 | struct task_struct *g, *p; |
1da177e4 | 4773 | |
4bd77321 IM |
4774 | #if BITS_PER_LONG == 32 |
4775 | printk(KERN_INFO | |
4776 | " task PC stack pid father\n"); | |
1da177e4 | 4777 | #else |
4bd77321 IM |
4778 | printk(KERN_INFO |
4779 | " task PC stack pid father\n"); | |
1da177e4 LT |
4780 | #endif |
4781 | read_lock(&tasklist_lock); | |
4782 | do_each_thread(g, p) { | |
4783 | /* | |
4784 | * reset the NMI-timeout, listing all files on a slow | |
4785 | * console might take alot of time: | |
4786 | */ | |
4787 | touch_nmi_watchdog(); | |
39bc89fd | 4788 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4789 | show_task(p); |
1da177e4 LT |
4790 | } while_each_thread(g, p); |
4791 | ||
04c9167f JF |
4792 | touch_all_softlockup_watchdogs(); |
4793 | ||
dd41f596 IM |
4794 | #ifdef CONFIG_SCHED_DEBUG |
4795 | sysrq_sched_debug_show(); | |
4796 | #endif | |
1da177e4 | 4797 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4798 | /* |
4799 | * Only show locks if all tasks are dumped: | |
4800 | */ | |
4801 | if (state_filter == -1) | |
4802 | debug_show_all_locks(); | |
1da177e4 LT |
4803 | } |
4804 | ||
1df21055 IM |
4805 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4806 | { | |
dd41f596 | 4807 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4808 | } |
4809 | ||
f340c0d1 IM |
4810 | /** |
4811 | * init_idle - set up an idle thread for a given CPU | |
4812 | * @idle: task in question | |
4813 | * @cpu: cpu the idle task belongs to | |
4814 | * | |
4815 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4816 | * flag, to make booting more robust. | |
4817 | */ | |
5c1e1767 | 4818 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4819 | { |
70b97a7f | 4820 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4821 | unsigned long flags; |
4822 | ||
dd41f596 IM |
4823 | __sched_fork(idle); |
4824 | idle->se.exec_start = sched_clock(); | |
4825 | ||
b29739f9 | 4826 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4827 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4828 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4829 | |
4830 | spin_lock_irqsave(&rq->lock, flags); | |
4831 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4832 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4833 | idle->oncpu = 1; | |
4834 | #endif | |
1da177e4 LT |
4835 | spin_unlock_irqrestore(&rq->lock, flags); |
4836 | ||
4837 | /* Set the preempt count _outside_ the spinlocks! */ | |
4838 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4839 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4840 | #else |
a1261f54 | 4841 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4842 | #endif |
dd41f596 IM |
4843 | /* |
4844 | * The idle tasks have their own, simple scheduling class: | |
4845 | */ | |
4846 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4847 | } |
4848 | ||
4849 | /* | |
4850 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4851 | * indicates which cpus entered this state. This is used | |
4852 | * in the rcu update to wait only for active cpus. For system | |
4853 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4854 | * always be CPU_MASK_NONE. | |
4855 | */ | |
4856 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4857 | ||
dd41f596 IM |
4858 | /* |
4859 | * Increase the granularity value when there are more CPUs, | |
4860 | * because with more CPUs the 'effective latency' as visible | |
4861 | * to users decreases. But the relationship is not linear, | |
4862 | * so pick a second-best guess by going with the log2 of the | |
4863 | * number of CPUs. | |
4864 | * | |
4865 | * This idea comes from the SD scheduler of Con Kolivas: | |
4866 | */ | |
4867 | static inline void sched_init_granularity(void) | |
4868 | { | |
4869 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
a5968df8 | 4870 | const unsigned long gran_limit = 100000000; |
dd41f596 IM |
4871 | |
4872 | sysctl_sched_granularity *= factor; | |
4873 | if (sysctl_sched_granularity > gran_limit) | |
4874 | sysctl_sched_granularity = gran_limit; | |
4875 | ||
4876 | sysctl_sched_runtime_limit = sysctl_sched_granularity * 4; | |
4877 | sysctl_sched_wakeup_granularity = sysctl_sched_granularity / 2; | |
4878 | } | |
4879 | ||
1da177e4 LT |
4880 | #ifdef CONFIG_SMP |
4881 | /* | |
4882 | * This is how migration works: | |
4883 | * | |
70b97a7f | 4884 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4885 | * runqueue and wake up that CPU's migration thread. |
4886 | * 2) we down() the locked semaphore => thread blocks. | |
4887 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4888 | * thread off the CPU) | |
4889 | * 4) it gets the migration request and checks whether the migrated | |
4890 | * task is still in the wrong runqueue. | |
4891 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4892 | * it and puts it into the right queue. | |
4893 | * 6) migration thread up()s the semaphore. | |
4894 | * 7) we wake up and the migration is done. | |
4895 | */ | |
4896 | ||
4897 | /* | |
4898 | * Change a given task's CPU affinity. Migrate the thread to a | |
4899 | * proper CPU and schedule it away if the CPU it's executing on | |
4900 | * is removed from the allowed bitmask. | |
4901 | * | |
4902 | * NOTE: the caller must have a valid reference to the task, the | |
4903 | * task must not exit() & deallocate itself prematurely. The | |
4904 | * call is not atomic; no spinlocks may be held. | |
4905 | */ | |
36c8b586 | 4906 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4907 | { |
70b97a7f | 4908 | struct migration_req req; |
1da177e4 | 4909 | unsigned long flags; |
70b97a7f | 4910 | struct rq *rq; |
48f24c4d | 4911 | int ret = 0; |
1da177e4 LT |
4912 | |
4913 | rq = task_rq_lock(p, &flags); | |
4914 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4915 | ret = -EINVAL; | |
4916 | goto out; | |
4917 | } | |
4918 | ||
4919 | p->cpus_allowed = new_mask; | |
4920 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4921 | if (cpu_isset(task_cpu(p), new_mask)) | |
4922 | goto out; | |
4923 | ||
4924 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4925 | /* Need help from migration thread: drop lock and wait. */ | |
4926 | task_rq_unlock(rq, &flags); | |
4927 | wake_up_process(rq->migration_thread); | |
4928 | wait_for_completion(&req.done); | |
4929 | tlb_migrate_finish(p->mm); | |
4930 | return 0; | |
4931 | } | |
4932 | out: | |
4933 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4934 | |
1da177e4 LT |
4935 | return ret; |
4936 | } | |
1da177e4 LT |
4937 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4938 | ||
4939 | /* | |
4940 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4941 | * this because either it can't run here any more (set_cpus_allowed() | |
4942 | * away from this CPU, or CPU going down), or because we're | |
4943 | * attempting to rebalance this task on exec (sched_exec). | |
4944 | * | |
4945 | * So we race with normal scheduler movements, but that's OK, as long | |
4946 | * as the task is no longer on this CPU. | |
efc30814 KK |
4947 | * |
4948 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4949 | */ |
efc30814 | 4950 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4951 | { |
70b97a7f | 4952 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4953 | int ret = 0, on_rq; |
1da177e4 LT |
4954 | |
4955 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4956 | return ret; |
1da177e4 LT |
4957 | |
4958 | rq_src = cpu_rq(src_cpu); | |
4959 | rq_dest = cpu_rq(dest_cpu); | |
4960 | ||
4961 | double_rq_lock(rq_src, rq_dest); | |
4962 | /* Already moved. */ | |
4963 | if (task_cpu(p) != src_cpu) | |
4964 | goto out; | |
4965 | /* Affinity changed (again). */ | |
4966 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4967 | goto out; | |
4968 | ||
dd41f596 IM |
4969 | on_rq = p->se.on_rq; |
4970 | if (on_rq) | |
8e717b19 | 4971 | deactivate_task(rq_src, p, 0, rq_clock(rq_src)); |
1da177e4 | 4972 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4973 | if (on_rq) { |
4974 | activate_task(rq_dest, p, 0); | |
4975 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4976 | } |
efc30814 | 4977 | ret = 1; |
1da177e4 LT |
4978 | out: |
4979 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4980 | return ret; |
1da177e4 LT |
4981 | } |
4982 | ||
4983 | /* | |
4984 | * migration_thread - this is a highprio system thread that performs | |
4985 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4986 | * another runqueue. | |
4987 | */ | |
95cdf3b7 | 4988 | static int migration_thread(void *data) |
1da177e4 | 4989 | { |
1da177e4 | 4990 | int cpu = (long)data; |
70b97a7f | 4991 | struct rq *rq; |
1da177e4 LT |
4992 | |
4993 | rq = cpu_rq(cpu); | |
4994 | BUG_ON(rq->migration_thread != current); | |
4995 | ||
4996 | set_current_state(TASK_INTERRUPTIBLE); | |
4997 | while (!kthread_should_stop()) { | |
70b97a7f | 4998 | struct migration_req *req; |
1da177e4 | 4999 | struct list_head *head; |
1da177e4 | 5000 | |
1da177e4 LT |
5001 | spin_lock_irq(&rq->lock); |
5002 | ||
5003 | if (cpu_is_offline(cpu)) { | |
5004 | spin_unlock_irq(&rq->lock); | |
5005 | goto wait_to_die; | |
5006 | } | |
5007 | ||
5008 | if (rq->active_balance) { | |
5009 | active_load_balance(rq, cpu); | |
5010 | rq->active_balance = 0; | |
5011 | } | |
5012 | ||
5013 | head = &rq->migration_queue; | |
5014 | ||
5015 | if (list_empty(head)) { | |
5016 | spin_unlock_irq(&rq->lock); | |
5017 | schedule(); | |
5018 | set_current_state(TASK_INTERRUPTIBLE); | |
5019 | continue; | |
5020 | } | |
70b97a7f | 5021 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5022 | list_del_init(head->next); |
5023 | ||
674311d5 NP |
5024 | spin_unlock(&rq->lock); |
5025 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5026 | local_irq_enable(); | |
1da177e4 LT |
5027 | |
5028 | complete(&req->done); | |
5029 | } | |
5030 | __set_current_state(TASK_RUNNING); | |
5031 | return 0; | |
5032 | ||
5033 | wait_to_die: | |
5034 | /* Wait for kthread_stop */ | |
5035 | set_current_state(TASK_INTERRUPTIBLE); | |
5036 | while (!kthread_should_stop()) { | |
5037 | schedule(); | |
5038 | set_current_state(TASK_INTERRUPTIBLE); | |
5039 | } | |
5040 | __set_current_state(TASK_RUNNING); | |
5041 | return 0; | |
5042 | } | |
5043 | ||
5044 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5045 | /* |
5046 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5047 | * NOTE: interrupts should be disabled by the caller | |
5048 | */ | |
48f24c4d | 5049 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5050 | { |
efc30814 | 5051 | unsigned long flags; |
1da177e4 | 5052 | cpumask_t mask; |
70b97a7f IM |
5053 | struct rq *rq; |
5054 | int dest_cpu; | |
1da177e4 | 5055 | |
efc30814 | 5056 | restart: |
1da177e4 LT |
5057 | /* On same node? */ |
5058 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5059 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5060 | dest_cpu = any_online_cpu(mask); |
5061 | ||
5062 | /* On any allowed CPU? */ | |
5063 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5064 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5065 | |
5066 | /* No more Mr. Nice Guy. */ | |
5067 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5068 | rq = task_rq_lock(p, &flags); |
5069 | cpus_setall(p->cpus_allowed); | |
5070 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5071 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5072 | |
5073 | /* | |
5074 | * Don't tell them about moving exiting tasks or | |
5075 | * kernel threads (both mm NULL), since they never | |
5076 | * leave kernel. | |
5077 | */ | |
48f24c4d | 5078 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5079 | printk(KERN_INFO "process %d (%s) no " |
5080 | "longer affine to cpu%d\n", | |
48f24c4d | 5081 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5082 | } |
48f24c4d | 5083 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5084 | goto restart; |
1da177e4 LT |
5085 | } |
5086 | ||
5087 | /* | |
5088 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5089 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5090 | * for performance reasons the counter is not stricly tracking tasks to | |
5091 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5092 | * to keep the global sum constant after CPU-down: | |
5093 | */ | |
70b97a7f | 5094 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5095 | { |
70b97a7f | 5096 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5097 | unsigned long flags; |
5098 | ||
5099 | local_irq_save(flags); | |
5100 | double_rq_lock(rq_src, rq_dest); | |
5101 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5102 | rq_src->nr_uninterruptible = 0; | |
5103 | double_rq_unlock(rq_src, rq_dest); | |
5104 | local_irq_restore(flags); | |
5105 | } | |
5106 | ||
5107 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5108 | static void migrate_live_tasks(int src_cpu) | |
5109 | { | |
48f24c4d | 5110 | struct task_struct *p, *t; |
1da177e4 LT |
5111 | |
5112 | write_lock_irq(&tasklist_lock); | |
5113 | ||
48f24c4d IM |
5114 | do_each_thread(t, p) { |
5115 | if (p == current) | |
1da177e4 LT |
5116 | continue; |
5117 | ||
48f24c4d IM |
5118 | if (task_cpu(p) == src_cpu) |
5119 | move_task_off_dead_cpu(src_cpu, p); | |
5120 | } while_each_thread(t, p); | |
1da177e4 LT |
5121 | |
5122 | write_unlock_irq(&tasklist_lock); | |
5123 | } | |
5124 | ||
dd41f596 IM |
5125 | /* |
5126 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5127 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5128 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5129 | */ |
5130 | void sched_idle_next(void) | |
5131 | { | |
48f24c4d | 5132 | int this_cpu = smp_processor_id(); |
70b97a7f | 5133 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5134 | struct task_struct *p = rq->idle; |
5135 | unsigned long flags; | |
5136 | ||
5137 | /* cpu has to be offline */ | |
48f24c4d | 5138 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5139 | |
48f24c4d IM |
5140 | /* |
5141 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5142 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5143 | */ |
5144 | spin_lock_irqsave(&rq->lock, flags); | |
5145 | ||
dd41f596 | 5146 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5147 | |
5148 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5149 | activate_idle_task(p, rq); |
1da177e4 LT |
5150 | |
5151 | spin_unlock_irqrestore(&rq->lock, flags); | |
5152 | } | |
5153 | ||
48f24c4d IM |
5154 | /* |
5155 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5156 | * offline. |
5157 | */ | |
5158 | void idle_task_exit(void) | |
5159 | { | |
5160 | struct mm_struct *mm = current->active_mm; | |
5161 | ||
5162 | BUG_ON(cpu_online(smp_processor_id())); | |
5163 | ||
5164 | if (mm != &init_mm) | |
5165 | switch_mm(mm, &init_mm, current); | |
5166 | mmdrop(mm); | |
5167 | } | |
5168 | ||
054b9108 | 5169 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5170 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5171 | { |
70b97a7f | 5172 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5173 | |
5174 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5175 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5176 | |
5177 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5178 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5179 | |
48f24c4d | 5180 | get_task_struct(p); |
1da177e4 LT |
5181 | |
5182 | /* | |
5183 | * Drop lock around migration; if someone else moves it, | |
5184 | * that's OK. No task can be added to this CPU, so iteration is | |
5185 | * fine. | |
054b9108 | 5186 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5187 | */ |
054b9108 | 5188 | spin_unlock(&rq->lock); |
48f24c4d | 5189 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5190 | spin_lock(&rq->lock); |
1da177e4 | 5191 | |
48f24c4d | 5192 | put_task_struct(p); |
1da177e4 LT |
5193 | } |
5194 | ||
5195 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5196 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5197 | { | |
70b97a7f | 5198 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5199 | struct task_struct *next; |
48f24c4d | 5200 | |
dd41f596 IM |
5201 | for ( ; ; ) { |
5202 | if (!rq->nr_running) | |
5203 | break; | |
5204 | next = pick_next_task(rq, rq->curr, rq_clock(rq)); | |
5205 | if (!next) | |
5206 | break; | |
5207 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5208 | |
1da177e4 LT |
5209 | } |
5210 | } | |
5211 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5212 | ||
e692ab53 NP |
5213 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5214 | ||
5215 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5216 | { |
5217 | .procname = "sched_domain", | |
5218 | .mode = 0755, | |
5219 | }, | |
e692ab53 NP |
5220 | {0,}, |
5221 | }; | |
5222 | ||
5223 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5224 | { |
5225 | .procname = "kernel", | |
5226 | .mode = 0755, | |
5227 | .child = sd_ctl_dir, | |
5228 | }, | |
e692ab53 NP |
5229 | {0,}, |
5230 | }; | |
5231 | ||
5232 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5233 | { | |
5234 | struct ctl_table *entry = | |
5235 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); | |
5236 | ||
5237 | BUG_ON(!entry); | |
5238 | memset(entry, 0, n * sizeof(struct ctl_table)); | |
5239 | ||
5240 | return entry; | |
5241 | } | |
5242 | ||
5243 | static void | |
e0361851 | 5244 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5245 | const char *procname, void *data, int maxlen, |
5246 | mode_t mode, proc_handler *proc_handler) | |
5247 | { | |
e692ab53 NP |
5248 | entry->procname = procname; |
5249 | entry->data = data; | |
5250 | entry->maxlen = maxlen; | |
5251 | entry->mode = mode; | |
5252 | entry->proc_handler = proc_handler; | |
5253 | } | |
5254 | ||
5255 | static struct ctl_table * | |
5256 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5257 | { | |
5258 | struct ctl_table *table = sd_alloc_ctl_entry(14); | |
5259 | ||
e0361851 | 5260 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5261 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5262 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5263 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5264 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5265 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5266 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5267 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5268 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5269 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5270 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5271 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5272 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5273 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5274 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5275 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5276 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5277 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5278 | set_table_entry(&table[10], "cache_nice_tries", |
e692ab53 NP |
5279 | &sd->cache_nice_tries, |
5280 | sizeof(int), 0644, proc_dointvec_minmax); | |
e0361851 | 5281 | set_table_entry(&table[12], "flags", &sd->flags, |
e692ab53 NP |
5282 | sizeof(int), 0644, proc_dointvec_minmax); |
5283 | ||
5284 | return table; | |
5285 | } | |
5286 | ||
5287 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | |
5288 | { | |
5289 | struct ctl_table *entry, *table; | |
5290 | struct sched_domain *sd; | |
5291 | int domain_num = 0, i; | |
5292 | char buf[32]; | |
5293 | ||
5294 | for_each_domain(cpu, sd) | |
5295 | domain_num++; | |
5296 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
5297 | ||
5298 | i = 0; | |
5299 | for_each_domain(cpu, sd) { | |
5300 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 NP |
5301 | entry->procname = kstrdup(buf, GFP_KERNEL); |
5302 | entry->mode = 0755; | |
5303 | entry->child = sd_alloc_ctl_domain_table(sd); | |
5304 | entry++; | |
5305 | i++; | |
5306 | } | |
5307 | return table; | |
5308 | } | |
5309 | ||
5310 | static struct ctl_table_header *sd_sysctl_header; | |
5311 | static void init_sched_domain_sysctl(void) | |
5312 | { | |
5313 | int i, cpu_num = num_online_cpus(); | |
5314 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5315 | char buf[32]; | |
5316 | ||
5317 | sd_ctl_dir[0].child = entry; | |
5318 | ||
5319 | for (i = 0; i < cpu_num; i++, entry++) { | |
5320 | snprintf(buf, 32, "cpu%d", i); | |
e692ab53 NP |
5321 | entry->procname = kstrdup(buf, GFP_KERNEL); |
5322 | entry->mode = 0755; | |
5323 | entry->child = sd_alloc_ctl_cpu_table(i); | |
5324 | } | |
5325 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5326 | } | |
5327 | #else | |
5328 | static void init_sched_domain_sysctl(void) | |
5329 | { | |
5330 | } | |
5331 | #endif | |
5332 | ||
1da177e4 LT |
5333 | /* |
5334 | * migration_call - callback that gets triggered when a CPU is added. | |
5335 | * Here we can start up the necessary migration thread for the new CPU. | |
5336 | */ | |
48f24c4d IM |
5337 | static int __cpuinit |
5338 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5339 | { |
1da177e4 | 5340 | struct task_struct *p; |
48f24c4d | 5341 | int cpu = (long)hcpu; |
1da177e4 | 5342 | unsigned long flags; |
70b97a7f | 5343 | struct rq *rq; |
1da177e4 LT |
5344 | |
5345 | switch (action) { | |
5be9361c GS |
5346 | case CPU_LOCK_ACQUIRE: |
5347 | mutex_lock(&sched_hotcpu_mutex); | |
5348 | break; | |
5349 | ||
1da177e4 | 5350 | case CPU_UP_PREPARE: |
8bb78442 | 5351 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5352 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5353 | if (IS_ERR(p)) |
5354 | return NOTIFY_BAD; | |
1da177e4 LT |
5355 | kthread_bind(p, cpu); |
5356 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5357 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5358 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5359 | task_rq_unlock(rq, &flags); |
5360 | cpu_rq(cpu)->migration_thread = p; | |
5361 | break; | |
48f24c4d | 5362 | |
1da177e4 | 5363 | case CPU_ONLINE: |
8bb78442 | 5364 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5365 | /* Strictly unneccessary, as first user will wake it. */ |
5366 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5367 | break; | |
48f24c4d | 5368 | |
1da177e4 LT |
5369 | #ifdef CONFIG_HOTPLUG_CPU |
5370 | case CPU_UP_CANCELED: | |
8bb78442 | 5371 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5372 | if (!cpu_rq(cpu)->migration_thread) |
5373 | break; | |
1da177e4 | 5374 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5375 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5376 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5377 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5378 | cpu_rq(cpu)->migration_thread = NULL; | |
5379 | break; | |
48f24c4d | 5380 | |
1da177e4 | 5381 | case CPU_DEAD: |
8bb78442 | 5382 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5383 | migrate_live_tasks(cpu); |
5384 | rq = cpu_rq(cpu); | |
5385 | kthread_stop(rq->migration_thread); | |
5386 | rq->migration_thread = NULL; | |
5387 | /* Idle task back to normal (off runqueue, low prio) */ | |
5388 | rq = task_rq_lock(rq->idle, &flags); | |
8e717b19 | 5389 | deactivate_task(rq, rq->idle, 0, rq_clock(rq)); |
1da177e4 | 5390 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5391 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5392 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5393 | migrate_dead_tasks(cpu); |
5394 | task_rq_unlock(rq, &flags); | |
5395 | migrate_nr_uninterruptible(rq); | |
5396 | BUG_ON(rq->nr_running != 0); | |
5397 | ||
5398 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5399 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5400 | * the requestors. */ |
5401 | spin_lock_irq(&rq->lock); | |
5402 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5403 | struct migration_req *req; |
5404 | ||
1da177e4 | 5405 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5406 | struct migration_req, list); |
1da177e4 LT |
5407 | list_del_init(&req->list); |
5408 | complete(&req->done); | |
5409 | } | |
5410 | spin_unlock_irq(&rq->lock); | |
5411 | break; | |
5412 | #endif | |
5be9361c GS |
5413 | case CPU_LOCK_RELEASE: |
5414 | mutex_unlock(&sched_hotcpu_mutex); | |
5415 | break; | |
1da177e4 LT |
5416 | } |
5417 | return NOTIFY_OK; | |
5418 | } | |
5419 | ||
5420 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5421 | * happens before everything else. | |
5422 | */ | |
26c2143b | 5423 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5424 | .notifier_call = migration_call, |
5425 | .priority = 10 | |
5426 | }; | |
5427 | ||
5428 | int __init migration_init(void) | |
5429 | { | |
5430 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5431 | int err; |
48f24c4d IM |
5432 | |
5433 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5434 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5435 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5436 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5437 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5438 | |
1da177e4 LT |
5439 | return 0; |
5440 | } | |
5441 | #endif | |
5442 | ||
5443 | #ifdef CONFIG_SMP | |
476f3534 CL |
5444 | |
5445 | /* Number of possible processor ids */ | |
5446 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5447 | EXPORT_SYMBOL(nr_cpu_ids); | |
5448 | ||
1a20ff27 | 5449 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5450 | #ifdef SCHED_DOMAIN_DEBUG |
5451 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5452 | { | |
5453 | int level = 0; | |
5454 | ||
41c7ce9a NP |
5455 | if (!sd) { |
5456 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5457 | return; | |
5458 | } | |
5459 | ||
1da177e4 LT |
5460 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5461 | ||
5462 | do { | |
5463 | int i; | |
5464 | char str[NR_CPUS]; | |
5465 | struct sched_group *group = sd->groups; | |
5466 | cpumask_t groupmask; | |
5467 | ||
5468 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5469 | cpus_clear(groupmask); | |
5470 | ||
5471 | printk(KERN_DEBUG); | |
5472 | for (i = 0; i < level + 1; i++) | |
5473 | printk(" "); | |
5474 | printk("domain %d: ", level); | |
5475 | ||
5476 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5477 | printk("does not load-balance\n"); | |
5478 | if (sd->parent) | |
33859f7f MOS |
5479 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5480 | " has parent"); | |
1da177e4 LT |
5481 | break; |
5482 | } | |
5483 | ||
5484 | printk("span %s\n", str); | |
5485 | ||
5486 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5487 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5488 | "CPU%d\n", cpu); | |
1da177e4 | 5489 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5490 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5491 | " CPU%d\n", cpu); | |
1da177e4 LT |
5492 | |
5493 | printk(KERN_DEBUG); | |
5494 | for (i = 0; i < level + 2; i++) | |
5495 | printk(" "); | |
5496 | printk("groups:"); | |
5497 | do { | |
5498 | if (!group) { | |
5499 | printk("\n"); | |
5500 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5501 | break; | |
5502 | } | |
5503 | ||
5517d86b | 5504 | if (!group->__cpu_power) { |
1da177e4 | 5505 | printk("\n"); |
33859f7f MOS |
5506 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5507 | "set\n"); | |
1da177e4 LT |
5508 | } |
5509 | ||
5510 | if (!cpus_weight(group->cpumask)) { | |
5511 | printk("\n"); | |
5512 | printk(KERN_ERR "ERROR: empty group\n"); | |
5513 | } | |
5514 | ||
5515 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5516 | printk("\n"); | |
5517 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5518 | } | |
5519 | ||
5520 | cpus_or(groupmask, groupmask, group->cpumask); | |
5521 | ||
5522 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5523 | printk(" %s", str); | |
5524 | ||
5525 | group = group->next; | |
5526 | } while (group != sd->groups); | |
5527 | printk("\n"); | |
5528 | ||
5529 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5530 | printk(KERN_ERR "ERROR: groups don't span " |
5531 | "domain->span\n"); | |
1da177e4 LT |
5532 | |
5533 | level++; | |
5534 | sd = sd->parent; | |
33859f7f MOS |
5535 | if (!sd) |
5536 | continue; | |
1da177e4 | 5537 | |
33859f7f MOS |
5538 | if (!cpus_subset(groupmask, sd->span)) |
5539 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5540 | "of domain->span\n"); | |
1da177e4 LT |
5541 | |
5542 | } while (sd); | |
5543 | } | |
5544 | #else | |
48f24c4d | 5545 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5546 | #endif |
5547 | ||
1a20ff27 | 5548 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5549 | { |
5550 | if (cpus_weight(sd->span) == 1) | |
5551 | return 1; | |
5552 | ||
5553 | /* Following flags need at least 2 groups */ | |
5554 | if (sd->flags & (SD_LOAD_BALANCE | | |
5555 | SD_BALANCE_NEWIDLE | | |
5556 | SD_BALANCE_FORK | | |
89c4710e SS |
5557 | SD_BALANCE_EXEC | |
5558 | SD_SHARE_CPUPOWER | | |
5559 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5560 | if (sd->groups != sd->groups->next) |
5561 | return 0; | |
5562 | } | |
5563 | ||
5564 | /* Following flags don't use groups */ | |
5565 | if (sd->flags & (SD_WAKE_IDLE | | |
5566 | SD_WAKE_AFFINE | | |
5567 | SD_WAKE_BALANCE)) | |
5568 | return 0; | |
5569 | ||
5570 | return 1; | |
5571 | } | |
5572 | ||
48f24c4d IM |
5573 | static int |
5574 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5575 | { |
5576 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5577 | ||
5578 | if (sd_degenerate(parent)) | |
5579 | return 1; | |
5580 | ||
5581 | if (!cpus_equal(sd->span, parent->span)) | |
5582 | return 0; | |
5583 | ||
5584 | /* Does parent contain flags not in child? */ | |
5585 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5586 | if (cflags & SD_WAKE_AFFINE) | |
5587 | pflags &= ~SD_WAKE_BALANCE; | |
5588 | /* Flags needing groups don't count if only 1 group in parent */ | |
5589 | if (parent->groups == parent->groups->next) { | |
5590 | pflags &= ~(SD_LOAD_BALANCE | | |
5591 | SD_BALANCE_NEWIDLE | | |
5592 | SD_BALANCE_FORK | | |
89c4710e SS |
5593 | SD_BALANCE_EXEC | |
5594 | SD_SHARE_CPUPOWER | | |
5595 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5596 | } |
5597 | if (~cflags & pflags) | |
5598 | return 0; | |
5599 | ||
5600 | return 1; | |
5601 | } | |
5602 | ||
1da177e4 LT |
5603 | /* |
5604 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5605 | * hold the hotplug lock. | |
5606 | */ | |
9c1cfda2 | 5607 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5608 | { |
70b97a7f | 5609 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5610 | struct sched_domain *tmp; |
5611 | ||
5612 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5613 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5614 | struct sched_domain *parent = tmp->parent; | |
5615 | if (!parent) | |
5616 | break; | |
1a848870 | 5617 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5618 | tmp->parent = parent->parent; |
1a848870 SS |
5619 | if (parent->parent) |
5620 | parent->parent->child = tmp; | |
5621 | } | |
245af2c7 SS |
5622 | } |
5623 | ||
1a848870 | 5624 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5625 | sd = sd->parent; |
1a848870 SS |
5626 | if (sd) |
5627 | sd->child = NULL; | |
5628 | } | |
1da177e4 LT |
5629 | |
5630 | sched_domain_debug(sd, cpu); | |
5631 | ||
674311d5 | 5632 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5633 | } |
5634 | ||
5635 | /* cpus with isolated domains */ | |
67af63a6 | 5636 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5637 | |
5638 | /* Setup the mask of cpus configured for isolated domains */ | |
5639 | static int __init isolated_cpu_setup(char *str) | |
5640 | { | |
5641 | int ints[NR_CPUS], i; | |
5642 | ||
5643 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5644 | cpus_clear(cpu_isolated_map); | |
5645 | for (i = 1; i <= ints[0]; i++) | |
5646 | if (ints[i] < NR_CPUS) | |
5647 | cpu_set(ints[i], cpu_isolated_map); | |
5648 | return 1; | |
5649 | } | |
5650 | ||
5651 | __setup ("isolcpus=", isolated_cpu_setup); | |
5652 | ||
5653 | /* | |
6711cab4 SS |
5654 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5655 | * to a function which identifies what group(along with sched group) a CPU | |
5656 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5657 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5658 | * |
5659 | * init_sched_build_groups will build a circular linked list of the groups | |
5660 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5661 | * and ->cpu_power to 0. | |
5662 | */ | |
a616058b | 5663 | static void |
6711cab4 SS |
5664 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5665 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5666 | struct sched_group **sg)) | |
1da177e4 LT |
5667 | { |
5668 | struct sched_group *first = NULL, *last = NULL; | |
5669 | cpumask_t covered = CPU_MASK_NONE; | |
5670 | int i; | |
5671 | ||
5672 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5673 | struct sched_group *sg; |
5674 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5675 | int j; |
5676 | ||
5677 | if (cpu_isset(i, covered)) | |
5678 | continue; | |
5679 | ||
5680 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5681 | sg->__cpu_power = 0; |
1da177e4 LT |
5682 | |
5683 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5684 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5685 | continue; |
5686 | ||
5687 | cpu_set(j, covered); | |
5688 | cpu_set(j, sg->cpumask); | |
5689 | } | |
5690 | if (!first) | |
5691 | first = sg; | |
5692 | if (last) | |
5693 | last->next = sg; | |
5694 | last = sg; | |
5695 | } | |
5696 | last->next = first; | |
5697 | } | |
5698 | ||
9c1cfda2 | 5699 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5700 | |
9c1cfda2 | 5701 | #ifdef CONFIG_NUMA |
198e2f18 | 5702 | |
9c1cfda2 JH |
5703 | /** |
5704 | * find_next_best_node - find the next node to include in a sched_domain | |
5705 | * @node: node whose sched_domain we're building | |
5706 | * @used_nodes: nodes already in the sched_domain | |
5707 | * | |
5708 | * Find the next node to include in a given scheduling domain. Simply | |
5709 | * finds the closest node not already in the @used_nodes map. | |
5710 | * | |
5711 | * Should use nodemask_t. | |
5712 | */ | |
5713 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5714 | { | |
5715 | int i, n, val, min_val, best_node = 0; | |
5716 | ||
5717 | min_val = INT_MAX; | |
5718 | ||
5719 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5720 | /* Start at @node */ | |
5721 | n = (node + i) % MAX_NUMNODES; | |
5722 | ||
5723 | if (!nr_cpus_node(n)) | |
5724 | continue; | |
5725 | ||
5726 | /* Skip already used nodes */ | |
5727 | if (test_bit(n, used_nodes)) | |
5728 | continue; | |
5729 | ||
5730 | /* Simple min distance search */ | |
5731 | val = node_distance(node, n); | |
5732 | ||
5733 | if (val < min_val) { | |
5734 | min_val = val; | |
5735 | best_node = n; | |
5736 | } | |
5737 | } | |
5738 | ||
5739 | set_bit(best_node, used_nodes); | |
5740 | return best_node; | |
5741 | } | |
5742 | ||
5743 | /** | |
5744 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5745 | * @node: node whose cpumask we're constructing | |
5746 | * @size: number of nodes to include in this span | |
5747 | * | |
5748 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5749 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5750 | * out optimally. | |
5751 | */ | |
5752 | static cpumask_t sched_domain_node_span(int node) | |
5753 | { | |
9c1cfda2 | 5754 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5755 | cpumask_t span, nodemask; |
5756 | int i; | |
9c1cfda2 JH |
5757 | |
5758 | cpus_clear(span); | |
5759 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5760 | ||
5761 | nodemask = node_to_cpumask(node); | |
5762 | cpus_or(span, span, nodemask); | |
5763 | set_bit(node, used_nodes); | |
5764 | ||
5765 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5766 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5767 | |
9c1cfda2 JH |
5768 | nodemask = node_to_cpumask(next_node); |
5769 | cpus_or(span, span, nodemask); | |
5770 | } | |
5771 | ||
5772 | return span; | |
5773 | } | |
5774 | #endif | |
5775 | ||
5c45bf27 | 5776 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5777 | |
9c1cfda2 | 5778 | /* |
48f24c4d | 5779 | * SMT sched-domains: |
9c1cfda2 | 5780 | */ |
1da177e4 LT |
5781 | #ifdef CONFIG_SCHED_SMT |
5782 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5783 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5784 | |
6711cab4 SS |
5785 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5786 | struct sched_group **sg) | |
1da177e4 | 5787 | { |
6711cab4 SS |
5788 | if (sg) |
5789 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5790 | return cpu; |
5791 | } | |
5792 | #endif | |
5793 | ||
48f24c4d IM |
5794 | /* |
5795 | * multi-core sched-domains: | |
5796 | */ | |
1e9f28fa SS |
5797 | #ifdef CONFIG_SCHED_MC |
5798 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5799 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5800 | #endif |
5801 | ||
5802 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5803 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5804 | struct sched_group **sg) | |
1e9f28fa | 5805 | { |
6711cab4 | 5806 | int group; |
a616058b SS |
5807 | cpumask_t mask = cpu_sibling_map[cpu]; |
5808 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5809 | group = first_cpu(mask); |
5810 | if (sg) | |
5811 | *sg = &per_cpu(sched_group_core, group); | |
5812 | return group; | |
1e9f28fa SS |
5813 | } |
5814 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5815 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5816 | struct sched_group **sg) | |
1e9f28fa | 5817 | { |
6711cab4 SS |
5818 | if (sg) |
5819 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5820 | return cpu; |
5821 | } | |
5822 | #endif | |
5823 | ||
1da177e4 | 5824 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5825 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5826 | |
6711cab4 SS |
5827 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5828 | struct sched_group **sg) | |
1da177e4 | 5829 | { |
6711cab4 | 5830 | int group; |
48f24c4d | 5831 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5832 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5833 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5834 | group = first_cpu(mask); |
1e9f28fa | 5835 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5836 | cpumask_t mask = cpu_sibling_map[cpu]; |
5837 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5838 | group = first_cpu(mask); |
1da177e4 | 5839 | #else |
6711cab4 | 5840 | group = cpu; |
1da177e4 | 5841 | #endif |
6711cab4 SS |
5842 | if (sg) |
5843 | *sg = &per_cpu(sched_group_phys, group); | |
5844 | return group; | |
1da177e4 LT |
5845 | } |
5846 | ||
5847 | #ifdef CONFIG_NUMA | |
1da177e4 | 5848 | /* |
9c1cfda2 JH |
5849 | * The init_sched_build_groups can't handle what we want to do with node |
5850 | * groups, so roll our own. Now each node has its own list of groups which | |
5851 | * gets dynamically allocated. | |
1da177e4 | 5852 | */ |
9c1cfda2 | 5853 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5854 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5855 | |
9c1cfda2 | 5856 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5857 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5858 | |
6711cab4 SS |
5859 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5860 | struct sched_group **sg) | |
9c1cfda2 | 5861 | { |
6711cab4 SS |
5862 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5863 | int group; | |
5864 | ||
5865 | cpus_and(nodemask, nodemask, *cpu_map); | |
5866 | group = first_cpu(nodemask); | |
5867 | ||
5868 | if (sg) | |
5869 | *sg = &per_cpu(sched_group_allnodes, group); | |
5870 | return group; | |
1da177e4 | 5871 | } |
6711cab4 | 5872 | |
08069033 SS |
5873 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5874 | { | |
5875 | struct sched_group *sg = group_head; | |
5876 | int j; | |
5877 | ||
5878 | if (!sg) | |
5879 | return; | |
5880 | next_sg: | |
5881 | for_each_cpu_mask(j, sg->cpumask) { | |
5882 | struct sched_domain *sd; | |
5883 | ||
5884 | sd = &per_cpu(phys_domains, j); | |
5885 | if (j != first_cpu(sd->groups->cpumask)) { | |
5886 | /* | |
5887 | * Only add "power" once for each | |
5888 | * physical package. | |
5889 | */ | |
5890 | continue; | |
5891 | } | |
5892 | ||
5517d86b | 5893 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5894 | } |
5895 | sg = sg->next; | |
5896 | if (sg != group_head) | |
5897 | goto next_sg; | |
5898 | } | |
1da177e4 LT |
5899 | #endif |
5900 | ||
a616058b | 5901 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5902 | /* Free memory allocated for various sched_group structures */ |
5903 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5904 | { | |
a616058b | 5905 | int cpu, i; |
51888ca2 SV |
5906 | |
5907 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5908 | struct sched_group **sched_group_nodes |
5909 | = sched_group_nodes_bycpu[cpu]; | |
5910 | ||
51888ca2 SV |
5911 | if (!sched_group_nodes) |
5912 | continue; | |
5913 | ||
5914 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5915 | cpumask_t nodemask = node_to_cpumask(i); | |
5916 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5917 | ||
5918 | cpus_and(nodemask, nodemask, *cpu_map); | |
5919 | if (cpus_empty(nodemask)) | |
5920 | continue; | |
5921 | ||
5922 | if (sg == NULL) | |
5923 | continue; | |
5924 | sg = sg->next; | |
5925 | next_sg: | |
5926 | oldsg = sg; | |
5927 | sg = sg->next; | |
5928 | kfree(oldsg); | |
5929 | if (oldsg != sched_group_nodes[i]) | |
5930 | goto next_sg; | |
5931 | } | |
5932 | kfree(sched_group_nodes); | |
5933 | sched_group_nodes_bycpu[cpu] = NULL; | |
5934 | } | |
51888ca2 | 5935 | } |
a616058b SS |
5936 | #else |
5937 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5938 | { | |
5939 | } | |
5940 | #endif | |
51888ca2 | 5941 | |
89c4710e SS |
5942 | /* |
5943 | * Initialize sched groups cpu_power. | |
5944 | * | |
5945 | * cpu_power indicates the capacity of sched group, which is used while | |
5946 | * distributing the load between different sched groups in a sched domain. | |
5947 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5948 | * there are asymmetries in the topology. If there are asymmetries, group | |
5949 | * having more cpu_power will pickup more load compared to the group having | |
5950 | * less cpu_power. | |
5951 | * | |
5952 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5953 | * the maximum number of tasks a group can handle in the presence of other idle | |
5954 | * or lightly loaded groups in the same sched domain. | |
5955 | */ | |
5956 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5957 | { | |
5958 | struct sched_domain *child; | |
5959 | struct sched_group *group; | |
5960 | ||
5961 | WARN_ON(!sd || !sd->groups); | |
5962 | ||
5963 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5964 | return; | |
5965 | ||
5966 | child = sd->child; | |
5967 | ||
5517d86b ED |
5968 | sd->groups->__cpu_power = 0; |
5969 | ||
89c4710e SS |
5970 | /* |
5971 | * For perf policy, if the groups in child domain share resources | |
5972 | * (for example cores sharing some portions of the cache hierarchy | |
5973 | * or SMT), then set this domain groups cpu_power such that each group | |
5974 | * can handle only one task, when there are other idle groups in the | |
5975 | * same sched domain. | |
5976 | */ | |
5977 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5978 | (child->flags & | |
5979 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5980 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5981 | return; |
5982 | } | |
5983 | ||
89c4710e SS |
5984 | /* |
5985 | * add cpu_power of each child group to this groups cpu_power | |
5986 | */ | |
5987 | group = child->groups; | |
5988 | do { | |
5517d86b | 5989 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5990 | group = group->next; |
5991 | } while (group != child->groups); | |
5992 | } | |
5993 | ||
1da177e4 | 5994 | /* |
1a20ff27 DG |
5995 | * Build sched domains for a given set of cpus and attach the sched domains |
5996 | * to the individual cpus | |
1da177e4 | 5997 | */ |
51888ca2 | 5998 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5999 | { |
6000 | int i; | |
d1b55138 JH |
6001 | #ifdef CONFIG_NUMA |
6002 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6003 | int sd_allnodes = 0; |
d1b55138 JH |
6004 | |
6005 | /* | |
6006 | * Allocate the per-node list of sched groups | |
6007 | */ | |
dd41f596 | 6008 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 6009 | GFP_KERNEL); |
d1b55138 JH |
6010 | if (!sched_group_nodes) { |
6011 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6012 | return -ENOMEM; |
d1b55138 JH |
6013 | } |
6014 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6015 | #endif | |
1da177e4 LT |
6016 | |
6017 | /* | |
1a20ff27 | 6018 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6019 | */ |
1a20ff27 | 6020 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6021 | struct sched_domain *sd = NULL, *p; |
6022 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6023 | ||
1a20ff27 | 6024 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6025 | |
6026 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6027 | if (cpus_weight(*cpu_map) > |
6028 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6029 | sd = &per_cpu(allnodes_domains, i); |
6030 | *sd = SD_ALLNODES_INIT; | |
6031 | sd->span = *cpu_map; | |
6711cab4 | 6032 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6033 | p = sd; |
6711cab4 | 6034 | sd_allnodes = 1; |
9c1cfda2 JH |
6035 | } else |
6036 | p = NULL; | |
6037 | ||
1da177e4 | 6038 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6039 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6040 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6041 | sd->parent = p; | |
1a848870 SS |
6042 | if (p) |
6043 | p->child = sd; | |
9c1cfda2 | 6044 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6045 | #endif |
6046 | ||
6047 | p = sd; | |
6048 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6049 | *sd = SD_CPU_INIT; |
6050 | sd->span = nodemask; | |
6051 | sd->parent = p; | |
1a848870 SS |
6052 | if (p) |
6053 | p->child = sd; | |
6711cab4 | 6054 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6055 | |
1e9f28fa SS |
6056 | #ifdef CONFIG_SCHED_MC |
6057 | p = sd; | |
6058 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6059 | *sd = SD_MC_INIT; |
6060 | sd->span = cpu_coregroup_map(i); | |
6061 | cpus_and(sd->span, sd->span, *cpu_map); | |
6062 | sd->parent = p; | |
1a848870 | 6063 | p->child = sd; |
6711cab4 | 6064 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6065 | #endif |
6066 | ||
1da177e4 LT |
6067 | #ifdef CONFIG_SCHED_SMT |
6068 | p = sd; | |
6069 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6070 | *sd = SD_SIBLING_INIT; |
6071 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6072 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6073 | sd->parent = p; |
1a848870 | 6074 | p->child = sd; |
6711cab4 | 6075 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6076 | #endif |
6077 | } | |
6078 | ||
6079 | #ifdef CONFIG_SCHED_SMT | |
6080 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6081 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6082 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6083 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6084 | if (i != first_cpu(this_sibling_map)) |
6085 | continue; | |
6086 | ||
dd41f596 IM |
6087 | init_sched_build_groups(this_sibling_map, cpu_map, |
6088 | &cpu_to_cpu_group); | |
1da177e4 LT |
6089 | } |
6090 | #endif | |
6091 | ||
1e9f28fa SS |
6092 | #ifdef CONFIG_SCHED_MC |
6093 | /* Set up multi-core groups */ | |
6094 | for_each_cpu_mask(i, *cpu_map) { | |
6095 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6096 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6097 | if (i != first_cpu(this_core_map)) | |
6098 | continue; | |
dd41f596 IM |
6099 | init_sched_build_groups(this_core_map, cpu_map, |
6100 | &cpu_to_core_group); | |
1e9f28fa SS |
6101 | } |
6102 | #endif | |
6103 | ||
1da177e4 LT |
6104 | /* Set up physical groups */ |
6105 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6106 | cpumask_t nodemask = node_to_cpumask(i); | |
6107 | ||
1a20ff27 | 6108 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6109 | if (cpus_empty(nodemask)) |
6110 | continue; | |
6111 | ||
6711cab4 | 6112 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6113 | } |
6114 | ||
6115 | #ifdef CONFIG_NUMA | |
6116 | /* Set up node groups */ | |
6711cab4 | 6117 | if (sd_allnodes) |
dd41f596 IM |
6118 | init_sched_build_groups(*cpu_map, cpu_map, |
6119 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6120 | |
6121 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6122 | /* Set up node groups */ | |
6123 | struct sched_group *sg, *prev; | |
6124 | cpumask_t nodemask = node_to_cpumask(i); | |
6125 | cpumask_t domainspan; | |
6126 | cpumask_t covered = CPU_MASK_NONE; | |
6127 | int j; | |
6128 | ||
6129 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6130 | if (cpus_empty(nodemask)) { |
6131 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6132 | continue; |
d1b55138 | 6133 | } |
9c1cfda2 JH |
6134 | |
6135 | domainspan = sched_domain_node_span(i); | |
6136 | cpus_and(domainspan, domainspan, *cpu_map); | |
6137 | ||
15f0b676 | 6138 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6139 | if (!sg) { |
6140 | printk(KERN_WARNING "Can not alloc domain group for " | |
6141 | "node %d\n", i); | |
6142 | goto error; | |
6143 | } | |
9c1cfda2 JH |
6144 | sched_group_nodes[i] = sg; |
6145 | for_each_cpu_mask(j, nodemask) { | |
6146 | struct sched_domain *sd; | |
9761eea8 | 6147 | |
9c1cfda2 JH |
6148 | sd = &per_cpu(node_domains, j); |
6149 | sd->groups = sg; | |
9c1cfda2 | 6150 | } |
5517d86b | 6151 | sg->__cpu_power = 0; |
9c1cfda2 | 6152 | sg->cpumask = nodemask; |
51888ca2 | 6153 | sg->next = sg; |
9c1cfda2 JH |
6154 | cpus_or(covered, covered, nodemask); |
6155 | prev = sg; | |
6156 | ||
6157 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6158 | cpumask_t tmp, notcovered; | |
6159 | int n = (i + j) % MAX_NUMNODES; | |
6160 | ||
6161 | cpus_complement(notcovered, covered); | |
6162 | cpus_and(tmp, notcovered, *cpu_map); | |
6163 | cpus_and(tmp, tmp, domainspan); | |
6164 | if (cpus_empty(tmp)) | |
6165 | break; | |
6166 | ||
6167 | nodemask = node_to_cpumask(n); | |
6168 | cpus_and(tmp, tmp, nodemask); | |
6169 | if (cpus_empty(tmp)) | |
6170 | continue; | |
6171 | ||
15f0b676 SV |
6172 | sg = kmalloc_node(sizeof(struct sched_group), |
6173 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6174 | if (!sg) { |
6175 | printk(KERN_WARNING | |
6176 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6177 | goto error; |
9c1cfda2 | 6178 | } |
5517d86b | 6179 | sg->__cpu_power = 0; |
9c1cfda2 | 6180 | sg->cpumask = tmp; |
51888ca2 | 6181 | sg->next = prev->next; |
9c1cfda2 JH |
6182 | cpus_or(covered, covered, tmp); |
6183 | prev->next = sg; | |
6184 | prev = sg; | |
6185 | } | |
9c1cfda2 | 6186 | } |
1da177e4 LT |
6187 | #endif |
6188 | ||
6189 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6190 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6191 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6192 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6193 | ||
89c4710e | 6194 | init_sched_groups_power(i, sd); |
5c45bf27 | 6195 | } |
1da177e4 | 6196 | #endif |
1e9f28fa | 6197 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6198 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6199 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6200 | ||
89c4710e | 6201 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6202 | } |
6203 | #endif | |
1e9f28fa | 6204 | |
5c45bf27 | 6205 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6206 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6207 | ||
89c4710e | 6208 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6209 | } |
6210 | ||
9c1cfda2 | 6211 | #ifdef CONFIG_NUMA |
08069033 SS |
6212 | for (i = 0; i < MAX_NUMNODES; i++) |
6213 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6214 | |
6711cab4 SS |
6215 | if (sd_allnodes) { |
6216 | struct sched_group *sg; | |
f712c0c7 | 6217 | |
6711cab4 | 6218 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6219 | init_numa_sched_groups_power(sg); |
6220 | } | |
9c1cfda2 JH |
6221 | #endif |
6222 | ||
1da177e4 | 6223 | /* Attach the domains */ |
1a20ff27 | 6224 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6225 | struct sched_domain *sd; |
6226 | #ifdef CONFIG_SCHED_SMT | |
6227 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6228 | #elif defined(CONFIG_SCHED_MC) |
6229 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6230 | #else |
6231 | sd = &per_cpu(phys_domains, i); | |
6232 | #endif | |
6233 | cpu_attach_domain(sd, i); | |
6234 | } | |
51888ca2 SV |
6235 | |
6236 | return 0; | |
6237 | ||
a616058b | 6238 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6239 | error: |
6240 | free_sched_groups(cpu_map); | |
6241 | return -ENOMEM; | |
a616058b | 6242 | #endif |
1da177e4 | 6243 | } |
1a20ff27 DG |
6244 | /* |
6245 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6246 | */ | |
51888ca2 | 6247 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6248 | { |
6249 | cpumask_t cpu_default_map; | |
51888ca2 | 6250 | int err; |
1da177e4 | 6251 | |
1a20ff27 DG |
6252 | /* |
6253 | * Setup mask for cpus without special case scheduling requirements. | |
6254 | * For now this just excludes isolated cpus, but could be used to | |
6255 | * exclude other special cases in the future. | |
6256 | */ | |
6257 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6258 | ||
51888ca2 SV |
6259 | err = build_sched_domains(&cpu_default_map); |
6260 | ||
6261 | return err; | |
1a20ff27 DG |
6262 | } |
6263 | ||
6264 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6265 | { |
51888ca2 | 6266 | free_sched_groups(cpu_map); |
9c1cfda2 | 6267 | } |
1da177e4 | 6268 | |
1a20ff27 DG |
6269 | /* |
6270 | * Detach sched domains from a group of cpus specified in cpu_map | |
6271 | * These cpus will now be attached to the NULL domain | |
6272 | */ | |
858119e1 | 6273 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6274 | { |
6275 | int i; | |
6276 | ||
6277 | for_each_cpu_mask(i, *cpu_map) | |
6278 | cpu_attach_domain(NULL, i); | |
6279 | synchronize_sched(); | |
6280 | arch_destroy_sched_domains(cpu_map); | |
6281 | } | |
6282 | ||
6283 | /* | |
6284 | * Partition sched domains as specified by the cpumasks below. | |
6285 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6286 | * waits for a RCU quiescent period, recalculates sched | |
6287 | * domain information and then attaches them back to the | |
6288 | * correct sched domains | |
6289 | * Call with hotplug lock held | |
6290 | */ | |
51888ca2 | 6291 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6292 | { |
6293 | cpumask_t change_map; | |
51888ca2 | 6294 | int err = 0; |
1a20ff27 DG |
6295 | |
6296 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6297 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6298 | cpus_or(change_map, *partition1, *partition2); | |
6299 | ||
6300 | /* Detach sched domains from all of the affected cpus */ | |
6301 | detach_destroy_domains(&change_map); | |
6302 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6303 | err = build_sched_domains(partition1); |
6304 | if (!err && !cpus_empty(*partition2)) | |
6305 | err = build_sched_domains(partition2); | |
6306 | ||
6307 | return err; | |
1a20ff27 DG |
6308 | } |
6309 | ||
5c45bf27 SS |
6310 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6311 | int arch_reinit_sched_domains(void) | |
6312 | { | |
6313 | int err; | |
6314 | ||
5be9361c | 6315 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6316 | detach_destroy_domains(&cpu_online_map); |
6317 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6318 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6319 | |
6320 | return err; | |
6321 | } | |
6322 | ||
6323 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6324 | { | |
6325 | int ret; | |
6326 | ||
6327 | if (buf[0] != '0' && buf[0] != '1') | |
6328 | return -EINVAL; | |
6329 | ||
6330 | if (smt) | |
6331 | sched_smt_power_savings = (buf[0] == '1'); | |
6332 | else | |
6333 | sched_mc_power_savings = (buf[0] == '1'); | |
6334 | ||
6335 | ret = arch_reinit_sched_domains(); | |
6336 | ||
6337 | return ret ? ret : count; | |
6338 | } | |
6339 | ||
6340 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6341 | { | |
6342 | int err = 0; | |
48f24c4d | 6343 | |
5c45bf27 SS |
6344 | #ifdef CONFIG_SCHED_SMT |
6345 | if (smt_capable()) | |
6346 | err = sysfs_create_file(&cls->kset.kobj, | |
6347 | &attr_sched_smt_power_savings.attr); | |
6348 | #endif | |
6349 | #ifdef CONFIG_SCHED_MC | |
6350 | if (!err && mc_capable()) | |
6351 | err = sysfs_create_file(&cls->kset.kobj, | |
6352 | &attr_sched_mc_power_savings.attr); | |
6353 | #endif | |
6354 | return err; | |
6355 | } | |
6356 | #endif | |
6357 | ||
6358 | #ifdef CONFIG_SCHED_MC | |
6359 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6360 | { | |
6361 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6362 | } | |
48f24c4d IM |
6363 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6364 | const char *buf, size_t count) | |
5c45bf27 SS |
6365 | { |
6366 | return sched_power_savings_store(buf, count, 0); | |
6367 | } | |
6368 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6369 | sched_mc_power_savings_store); | |
6370 | #endif | |
6371 | ||
6372 | #ifdef CONFIG_SCHED_SMT | |
6373 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6374 | { | |
6375 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6376 | } | |
48f24c4d IM |
6377 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6378 | const char *buf, size_t count) | |
5c45bf27 SS |
6379 | { |
6380 | return sched_power_savings_store(buf, count, 1); | |
6381 | } | |
6382 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6383 | sched_smt_power_savings_store); | |
6384 | #endif | |
6385 | ||
1da177e4 LT |
6386 | /* |
6387 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6388 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6389 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6390 | * which will prevent rebalancing while the sched domains are recalculated. |
6391 | */ | |
6392 | static int update_sched_domains(struct notifier_block *nfb, | |
6393 | unsigned long action, void *hcpu) | |
6394 | { | |
1da177e4 LT |
6395 | switch (action) { |
6396 | case CPU_UP_PREPARE: | |
8bb78442 | 6397 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6398 | case CPU_DOWN_PREPARE: |
8bb78442 | 6399 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6400 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6401 | return NOTIFY_OK; |
6402 | ||
6403 | case CPU_UP_CANCELED: | |
8bb78442 | 6404 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6405 | case CPU_DOWN_FAILED: |
8bb78442 | 6406 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6407 | case CPU_ONLINE: |
8bb78442 | 6408 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6409 | case CPU_DEAD: |
8bb78442 | 6410 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6411 | /* |
6412 | * Fall through and re-initialise the domains. | |
6413 | */ | |
6414 | break; | |
6415 | default: | |
6416 | return NOTIFY_DONE; | |
6417 | } | |
6418 | ||
6419 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6420 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6421 | |
6422 | return NOTIFY_OK; | |
6423 | } | |
1da177e4 LT |
6424 | |
6425 | void __init sched_init_smp(void) | |
6426 | { | |
5c1e1767 NP |
6427 | cpumask_t non_isolated_cpus; |
6428 | ||
5be9361c | 6429 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6430 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6431 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6432 | if (cpus_empty(non_isolated_cpus)) |
6433 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6434 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6435 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6436 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 | 6437 | |
e692ab53 NP |
6438 | init_sched_domain_sysctl(); |
6439 | ||
5c1e1767 NP |
6440 | /* Move init over to a non-isolated CPU */ |
6441 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6442 | BUG(); | |
dd41f596 | 6443 | sched_init_granularity(); |
1da177e4 LT |
6444 | } |
6445 | #else | |
6446 | void __init sched_init_smp(void) | |
6447 | { | |
dd41f596 | 6448 | sched_init_granularity(); |
1da177e4 LT |
6449 | } |
6450 | #endif /* CONFIG_SMP */ | |
6451 | ||
6452 | int in_sched_functions(unsigned long addr) | |
6453 | { | |
6454 | /* Linker adds these: start and end of __sched functions */ | |
6455 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6456 | |
1da177e4 LT |
6457 | return in_lock_functions(addr) || |
6458 | (addr >= (unsigned long)__sched_text_start | |
6459 | && addr < (unsigned long)__sched_text_end); | |
6460 | } | |
6461 | ||
dd41f596 IM |
6462 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
6463 | { | |
6464 | cfs_rq->tasks_timeline = RB_ROOT; | |
6465 | cfs_rq->fair_clock = 1; | |
6466 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6467 | cfs_rq->rq = rq; | |
6468 | #endif | |
6469 | } | |
6470 | ||
1da177e4 LT |
6471 | void __init sched_init(void) |
6472 | { | |
dd41f596 | 6473 | u64 now = sched_clock(); |
476f3534 | 6474 | int highest_cpu = 0; |
dd41f596 IM |
6475 | int i, j; |
6476 | ||
6477 | /* | |
6478 | * Link up the scheduling class hierarchy: | |
6479 | */ | |
6480 | rt_sched_class.next = &fair_sched_class; | |
6481 | fair_sched_class.next = &idle_sched_class; | |
6482 | idle_sched_class.next = NULL; | |
1da177e4 | 6483 | |
0a945022 | 6484 | for_each_possible_cpu(i) { |
dd41f596 | 6485 | struct rt_prio_array *array; |
70b97a7f | 6486 | struct rq *rq; |
1da177e4 LT |
6487 | |
6488 | rq = cpu_rq(i); | |
6489 | spin_lock_init(&rq->lock); | |
fcb99371 | 6490 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6491 | rq->nr_running = 0; |
dd41f596 IM |
6492 | rq->clock = 1; |
6493 | init_cfs_rq(&rq->cfs, rq); | |
6494 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6495 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
6496 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6497 | #endif | |
6498 | rq->ls.load_update_last = now; | |
6499 | rq->ls.load_update_start = now; | |
1da177e4 | 6500 | |
dd41f596 IM |
6501 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6502 | rq->cpu_load[j] = 0; | |
1da177e4 | 6503 | #ifdef CONFIG_SMP |
41c7ce9a | 6504 | rq->sd = NULL; |
1da177e4 | 6505 | rq->active_balance = 0; |
dd41f596 | 6506 | rq->next_balance = jiffies; |
1da177e4 | 6507 | rq->push_cpu = 0; |
0a2966b4 | 6508 | rq->cpu = i; |
1da177e4 LT |
6509 | rq->migration_thread = NULL; |
6510 | INIT_LIST_HEAD(&rq->migration_queue); | |
6511 | #endif | |
6512 | atomic_set(&rq->nr_iowait, 0); | |
6513 | ||
dd41f596 IM |
6514 | array = &rq->rt.active; |
6515 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6516 | INIT_LIST_HEAD(array->queue + j); | |
6517 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6518 | } |
476f3534 | 6519 | highest_cpu = i; |
dd41f596 IM |
6520 | /* delimiter for bitsearch: */ |
6521 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6522 | } |
6523 | ||
2dd73a4f | 6524 | set_load_weight(&init_task); |
b50f60ce | 6525 | |
e107be36 AK |
6526 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6527 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6528 | #endif | |
6529 | ||
c9819f45 | 6530 | #ifdef CONFIG_SMP |
476f3534 | 6531 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6532 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6533 | #endif | |
6534 | ||
b50f60ce HC |
6535 | #ifdef CONFIG_RT_MUTEXES |
6536 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6537 | #endif | |
6538 | ||
1da177e4 LT |
6539 | /* |
6540 | * The boot idle thread does lazy MMU switching as well: | |
6541 | */ | |
6542 | atomic_inc(&init_mm.mm_count); | |
6543 | enter_lazy_tlb(&init_mm, current); | |
6544 | ||
6545 | /* | |
6546 | * Make us the idle thread. Technically, schedule() should not be | |
6547 | * called from this thread, however somewhere below it might be, | |
6548 | * but because we are the idle thread, we just pick up running again | |
6549 | * when this runqueue becomes "idle". | |
6550 | */ | |
6551 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6552 | /* |
6553 | * During early bootup we pretend to be a normal task: | |
6554 | */ | |
6555 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6556 | } |
6557 | ||
6558 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6559 | void __might_sleep(char *file, int line) | |
6560 | { | |
48f24c4d | 6561 | #ifdef in_atomic |
1da177e4 LT |
6562 | static unsigned long prev_jiffy; /* ratelimiting */ |
6563 | ||
6564 | if ((in_atomic() || irqs_disabled()) && | |
6565 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6566 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6567 | return; | |
6568 | prev_jiffy = jiffies; | |
91368d73 | 6569 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6570 | " context at %s:%d\n", file, line); |
6571 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6572 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6573 | debug_show_held_locks(current); |
3117df04 IM |
6574 | if (irqs_disabled()) |
6575 | print_irqtrace_events(current); | |
1da177e4 LT |
6576 | dump_stack(); |
6577 | } | |
6578 | #endif | |
6579 | } | |
6580 | EXPORT_SYMBOL(__might_sleep); | |
6581 | #endif | |
6582 | ||
6583 | #ifdef CONFIG_MAGIC_SYSRQ | |
6584 | void normalize_rt_tasks(void) | |
6585 | { | |
a0f98a1c | 6586 | struct task_struct *g, *p; |
1da177e4 | 6587 | unsigned long flags; |
70b97a7f | 6588 | struct rq *rq; |
dd41f596 | 6589 | int on_rq; |
1da177e4 LT |
6590 | |
6591 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6592 | do_each_thread(g, p) { |
dd41f596 IM |
6593 | p->se.fair_key = 0; |
6594 | p->se.wait_runtime = 0; | |
6cfb0d5d | 6595 | p->se.exec_start = 0; |
dd41f596 | 6596 | p->se.wait_start_fair = 0; |
6cfb0d5d IM |
6597 | p->se.sleep_start_fair = 0; |
6598 | #ifdef CONFIG_SCHEDSTATS | |
dd41f596 | 6599 | p->se.wait_start = 0; |
dd41f596 | 6600 | p->se.sleep_start = 0; |
dd41f596 | 6601 | p->se.block_start = 0; |
6cfb0d5d | 6602 | #endif |
dd41f596 IM |
6603 | task_rq(p)->cfs.fair_clock = 0; |
6604 | task_rq(p)->clock = 0; | |
6605 | ||
6606 | if (!rt_task(p)) { | |
6607 | /* | |
6608 | * Renice negative nice level userspace | |
6609 | * tasks back to 0: | |
6610 | */ | |
6611 | if (TASK_NICE(p) < 0 && p->mm) | |
6612 | set_user_nice(p, 0); | |
1da177e4 | 6613 | continue; |
dd41f596 | 6614 | } |
1da177e4 | 6615 | |
b29739f9 IM |
6616 | spin_lock_irqsave(&p->pi_lock, flags); |
6617 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6618 | #ifdef CONFIG_SMP |
6619 | /* | |
6620 | * Do not touch the migration thread: | |
6621 | */ | |
6622 | if (p == rq->migration_thread) | |
6623 | goto out_unlock; | |
6624 | #endif | |
1da177e4 | 6625 | |
dd41f596 IM |
6626 | on_rq = p->se.on_rq; |
6627 | if (on_rq) | |
8e717b19 | 6628 | deactivate_task(task_rq(p), p, 0, rq_clock(task_rq(p))); |
dd41f596 IM |
6629 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
6630 | if (on_rq) { | |
6631 | activate_task(task_rq(p), p, 0); | |
1da177e4 LT |
6632 | resched_task(rq->curr); |
6633 | } | |
dd41f596 IM |
6634 | #ifdef CONFIG_SMP |
6635 | out_unlock: | |
6636 | #endif | |
b29739f9 IM |
6637 | __task_rq_unlock(rq); |
6638 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6639 | } while_each_thread(g, p); |
6640 | ||
1da177e4 LT |
6641 | read_unlock_irq(&tasklist_lock); |
6642 | } | |
6643 | ||
6644 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6645 | |
6646 | #ifdef CONFIG_IA64 | |
6647 | /* | |
6648 | * These functions are only useful for the IA64 MCA handling. | |
6649 | * | |
6650 | * They can only be called when the whole system has been | |
6651 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6652 | * activity can take place. Using them for anything else would | |
6653 | * be a serious bug, and as a result, they aren't even visible | |
6654 | * under any other configuration. | |
6655 | */ | |
6656 | ||
6657 | /** | |
6658 | * curr_task - return the current task for a given cpu. | |
6659 | * @cpu: the processor in question. | |
6660 | * | |
6661 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6662 | */ | |
36c8b586 | 6663 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6664 | { |
6665 | return cpu_curr(cpu); | |
6666 | } | |
6667 | ||
6668 | /** | |
6669 | * set_curr_task - set the current task for a given cpu. | |
6670 | * @cpu: the processor in question. | |
6671 | * @p: the task pointer to set. | |
6672 | * | |
6673 | * Description: This function must only be used when non-maskable interrupts | |
6674 | * are serviced on a separate stack. It allows the architecture to switch the | |
6675 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6676 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6677 | * and caller must save the original value of the current task (see | |
6678 | * curr_task() above) and restore that value before reenabling interrupts and | |
6679 | * re-starting the system. | |
6680 | * | |
6681 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6682 | */ | |
36c8b586 | 6683 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6684 | { |
6685 | cpu_curr(cpu) = p; | |
6686 | } | |
6687 | ||
6688 | #endif |