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 | ||
356 | static inline unsigned long long rq_clock(struct rq *rq) | |
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, | |
748 | int this_best_prio, int best_prio, int best_prio_seen, | |
749 | struct rq_iterator *iterator); | |
750 | ||
751 | #include "sched_stats.h" | |
752 | #include "sched_rt.c" | |
753 | #include "sched_fair.c" | |
754 | #include "sched_idletask.c" | |
755 | #ifdef CONFIG_SCHED_DEBUG | |
756 | # include "sched_debug.c" | |
757 | #endif | |
758 | ||
759 | #define sched_class_highest (&rt_sched_class) | |
760 | ||
9c217245 IM |
761 | static void __update_curr_load(struct rq *rq, struct load_stat *ls) |
762 | { | |
763 | if (rq->curr != rq->idle && ls->load.weight) { | |
764 | ls->delta_exec += ls->delta_stat; | |
765 | ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load); | |
766 | ls->delta_stat = 0; | |
767 | } | |
768 | } | |
769 | ||
770 | /* | |
771 | * Update delta_exec, delta_fair fields for rq. | |
772 | * | |
773 | * delta_fair clock advances at a rate inversely proportional to | |
774 | * total load (rq->ls.load.weight) on the runqueue, while | |
775 | * delta_exec advances at the same rate as wall-clock (provided | |
776 | * cpu is not idle). | |
777 | * | |
778 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
779 | * runqueue over any given interval. This (smoothened) load is used | |
780 | * during load balance. | |
781 | * | |
782 | * This function is called /before/ updating rq->ls.load | |
783 | * and when switching tasks. | |
784 | */ | |
785 | static void update_curr_load(struct rq *rq, u64 now) | |
786 | { | |
787 | struct load_stat *ls = &rq->ls; | |
788 | u64 start; | |
789 | ||
790 | start = ls->load_update_start; | |
791 | ls->load_update_start = now; | |
792 | ls->delta_stat += now - start; | |
793 | /* | |
794 | * Stagger updates to ls->delta_fair. Very frequent updates | |
795 | * can be expensive. | |
796 | */ | |
797 | if (ls->delta_stat >= sysctl_sched_stat_granularity) | |
798 | __update_curr_load(rq, ls); | |
799 | } | |
800 | ||
801 | static inline void | |
802 | inc_load(struct rq *rq, const struct task_struct *p, u64 now) | |
803 | { | |
804 | update_curr_load(rq, now); | |
805 | update_load_add(&rq->ls.load, p->se.load.weight); | |
806 | } | |
807 | ||
808 | static inline void | |
809 | dec_load(struct rq *rq, const struct task_struct *p, u64 now) | |
810 | { | |
811 | update_curr_load(rq, now); | |
812 | update_load_sub(&rq->ls.load, p->se.load.weight); | |
813 | } | |
814 | ||
815 | static void inc_nr_running(struct task_struct *p, struct rq *rq, u64 now) | |
816 | { | |
817 | rq->nr_running++; | |
818 | inc_load(rq, p, now); | |
819 | } | |
820 | ||
821 | static void dec_nr_running(struct task_struct *p, struct rq *rq, u64 now) | |
822 | { | |
823 | rq->nr_running--; | |
824 | dec_load(rq, p, now); | |
825 | } | |
826 | ||
45bf76df IM |
827 | static void set_load_weight(struct task_struct *p) |
828 | { | |
dd41f596 IM |
829 | task_rq(p)->cfs.wait_runtime -= p->se.wait_runtime; |
830 | p->se.wait_runtime = 0; | |
831 | ||
45bf76df | 832 | if (task_has_rt_policy(p)) { |
dd41f596 IM |
833 | p->se.load.weight = prio_to_weight[0] * 2; |
834 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
835 | return; | |
836 | } | |
45bf76df | 837 | |
dd41f596 IM |
838 | /* |
839 | * SCHED_IDLE tasks get minimal weight: | |
840 | */ | |
841 | if (p->policy == SCHED_IDLE) { | |
842 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
843 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
844 | return; | |
845 | } | |
71f8bd46 | 846 | |
dd41f596 IM |
847 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
848 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
849 | } |
850 | ||
dd41f596 IM |
851 | static void |
852 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, u64 now) | |
71f8bd46 | 853 | { |
dd41f596 IM |
854 | sched_info_queued(p); |
855 | p->sched_class->enqueue_task(rq, p, wakeup, now); | |
856 | p->se.on_rq = 1; | |
71f8bd46 IM |
857 | } |
858 | ||
dd41f596 IM |
859 | static void |
860 | dequeue_task(struct rq *rq, struct task_struct *p, int sleep, u64 now) | |
71f8bd46 | 861 | { |
dd41f596 IM |
862 | p->sched_class->dequeue_task(rq, p, sleep, now); |
863 | p->se.on_rq = 0; | |
71f8bd46 IM |
864 | } |
865 | ||
14531189 | 866 | /* |
dd41f596 | 867 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 868 | */ |
14531189 IM |
869 | static inline int __normal_prio(struct task_struct *p) |
870 | { | |
dd41f596 | 871 | return p->static_prio; |
14531189 IM |
872 | } |
873 | ||
b29739f9 IM |
874 | /* |
875 | * Calculate the expected normal priority: i.e. priority | |
876 | * without taking RT-inheritance into account. Might be | |
877 | * boosted by interactivity modifiers. Changes upon fork, | |
878 | * setprio syscalls, and whenever the interactivity | |
879 | * estimator recalculates. | |
880 | */ | |
36c8b586 | 881 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
882 | { |
883 | int prio; | |
884 | ||
e05606d3 | 885 | if (task_has_rt_policy(p)) |
b29739f9 IM |
886 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
887 | else | |
888 | prio = __normal_prio(p); | |
889 | return prio; | |
890 | } | |
891 | ||
892 | /* | |
893 | * Calculate the current priority, i.e. the priority | |
894 | * taken into account by the scheduler. This value might | |
895 | * be boosted by RT tasks, or might be boosted by | |
896 | * interactivity modifiers. Will be RT if the task got | |
897 | * RT-boosted. If not then it returns p->normal_prio. | |
898 | */ | |
36c8b586 | 899 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
900 | { |
901 | p->normal_prio = normal_prio(p); | |
902 | /* | |
903 | * If we are RT tasks or we were boosted to RT priority, | |
904 | * keep the priority unchanged. Otherwise, update priority | |
905 | * to the normal priority: | |
906 | */ | |
907 | if (!rt_prio(p->prio)) | |
908 | return p->normal_prio; | |
909 | return p->prio; | |
910 | } | |
911 | ||
1da177e4 | 912 | /* |
dd41f596 | 913 | * activate_task - move a task to the runqueue. |
1da177e4 | 914 | */ |
dd41f596 | 915 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 916 | { |
dd41f596 | 917 | u64 now = rq_clock(rq); |
d425b274 | 918 | |
dd41f596 IM |
919 | if (p->state == TASK_UNINTERRUPTIBLE) |
920 | rq->nr_uninterruptible--; | |
1da177e4 | 921 | |
dd41f596 IM |
922 | enqueue_task(rq, p, wakeup, now); |
923 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
924 | } |
925 | ||
926 | /* | |
dd41f596 | 927 | * activate_idle_task - move idle task to the _front_ of runqueue. |
1da177e4 | 928 | */ |
dd41f596 | 929 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 930 | { |
dd41f596 | 931 | u64 now = rq_clock(rq); |
1da177e4 | 932 | |
dd41f596 IM |
933 | if (p->state == TASK_UNINTERRUPTIBLE) |
934 | rq->nr_uninterruptible--; | |
ece8a684 | 935 | |
dd41f596 IM |
936 | enqueue_task(rq, p, 0, now); |
937 | inc_nr_running(p, rq, now); | |
1da177e4 LT |
938 | } |
939 | ||
940 | /* | |
941 | * deactivate_task - remove a task from the runqueue. | |
942 | */ | |
dd41f596 | 943 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 944 | { |
dd41f596 IM |
945 | u64 now = rq_clock(rq); |
946 | ||
947 | if (p->state == TASK_UNINTERRUPTIBLE) | |
948 | rq->nr_uninterruptible++; | |
949 | ||
950 | dequeue_task(rq, p, sleep, now); | |
951 | dec_nr_running(p, rq, now); | |
1da177e4 LT |
952 | } |
953 | ||
1da177e4 LT |
954 | /** |
955 | * task_curr - is this task currently executing on a CPU? | |
956 | * @p: the task in question. | |
957 | */ | |
36c8b586 | 958 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
959 | { |
960 | return cpu_curr(task_cpu(p)) == p; | |
961 | } | |
962 | ||
2dd73a4f PW |
963 | /* Used instead of source_load when we know the type == 0 */ |
964 | unsigned long weighted_cpuload(const int cpu) | |
965 | { | |
dd41f596 IM |
966 | return cpu_rq(cpu)->ls.load.weight; |
967 | } | |
968 | ||
969 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
970 | { | |
971 | #ifdef CONFIG_SMP | |
972 | task_thread_info(p)->cpu = cpu; | |
973 | set_task_cfs_rq(p); | |
974 | #endif | |
2dd73a4f PW |
975 | } |
976 | ||
1da177e4 | 977 | #ifdef CONFIG_SMP |
c65cc870 | 978 | |
dd41f596 | 979 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 980 | { |
dd41f596 IM |
981 | int old_cpu = task_cpu(p); |
982 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
983 | u64 clock_offset, fair_clock_offset; | |
984 | ||
985 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
986 | fair_clock_offset = old_rq->cfs.fair_clock - new_rq->cfs.fair_clock; |
987 | ||
dd41f596 IM |
988 | if (p->se.wait_start_fair) |
989 | p->se.wait_start_fair -= fair_clock_offset; | |
6cfb0d5d IM |
990 | if (p->se.sleep_start_fair) |
991 | p->se.sleep_start_fair -= fair_clock_offset; | |
992 | ||
993 | #ifdef CONFIG_SCHEDSTATS | |
994 | if (p->se.wait_start) | |
995 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
996 | if (p->se.sleep_start) |
997 | p->se.sleep_start -= clock_offset; | |
998 | if (p->se.block_start) | |
999 | p->se.block_start -= clock_offset; | |
6cfb0d5d | 1000 | #endif |
dd41f596 IM |
1001 | |
1002 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1003 | } |
1004 | ||
70b97a7f | 1005 | struct migration_req { |
1da177e4 | 1006 | struct list_head list; |
1da177e4 | 1007 | |
36c8b586 | 1008 | struct task_struct *task; |
1da177e4 LT |
1009 | int dest_cpu; |
1010 | ||
1da177e4 | 1011 | struct completion done; |
70b97a7f | 1012 | }; |
1da177e4 LT |
1013 | |
1014 | /* | |
1015 | * The task's runqueue lock must be held. | |
1016 | * Returns true if you have to wait for migration thread. | |
1017 | */ | |
36c8b586 | 1018 | static int |
70b97a7f | 1019 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1020 | { |
70b97a7f | 1021 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1022 | |
1023 | /* | |
1024 | * If the task is not on a runqueue (and not running), then | |
1025 | * it is sufficient to simply update the task's cpu field. | |
1026 | */ | |
dd41f596 | 1027 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1028 | set_task_cpu(p, dest_cpu); |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | init_completion(&req->done); | |
1da177e4 LT |
1033 | req->task = p; |
1034 | req->dest_cpu = dest_cpu; | |
1035 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1036 | |
1da177e4 LT |
1037 | return 1; |
1038 | } | |
1039 | ||
1040 | /* | |
1041 | * wait_task_inactive - wait for a thread to unschedule. | |
1042 | * | |
1043 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1044 | * else this function might spin for a *long* time. This function can't | |
1045 | * be called with interrupts off, or it may introduce deadlock with | |
1046 | * smp_call_function() if an IPI is sent by the same process we are | |
1047 | * waiting to become inactive. | |
1048 | */ | |
36c8b586 | 1049 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1050 | { |
1051 | unsigned long flags; | |
dd41f596 | 1052 | int running, on_rq; |
70b97a7f | 1053 | struct rq *rq; |
1da177e4 LT |
1054 | |
1055 | repeat: | |
fa490cfd LT |
1056 | /* |
1057 | * We do the initial early heuristics without holding | |
1058 | * any task-queue locks at all. We'll only try to get | |
1059 | * the runqueue lock when things look like they will | |
1060 | * work out! | |
1061 | */ | |
1062 | rq = task_rq(p); | |
1063 | ||
1064 | /* | |
1065 | * If the task is actively running on another CPU | |
1066 | * still, just relax and busy-wait without holding | |
1067 | * any locks. | |
1068 | * | |
1069 | * NOTE! Since we don't hold any locks, it's not | |
1070 | * even sure that "rq" stays as the right runqueue! | |
1071 | * But we don't care, since "task_running()" will | |
1072 | * return false if the runqueue has changed and p | |
1073 | * is actually now running somewhere else! | |
1074 | */ | |
1075 | while (task_running(rq, p)) | |
1076 | cpu_relax(); | |
1077 | ||
1078 | /* | |
1079 | * Ok, time to look more closely! We need the rq | |
1080 | * lock now, to be *sure*. If we're wrong, we'll | |
1081 | * just go back and repeat. | |
1082 | */ | |
1da177e4 | 1083 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1084 | running = task_running(rq, p); |
dd41f596 | 1085 | on_rq = p->se.on_rq; |
fa490cfd LT |
1086 | task_rq_unlock(rq, &flags); |
1087 | ||
1088 | /* | |
1089 | * Was it really running after all now that we | |
1090 | * checked with the proper locks actually held? | |
1091 | * | |
1092 | * Oops. Go back and try again.. | |
1093 | */ | |
1094 | if (unlikely(running)) { | |
1da177e4 | 1095 | cpu_relax(); |
1da177e4 LT |
1096 | goto repeat; |
1097 | } | |
fa490cfd LT |
1098 | |
1099 | /* | |
1100 | * It's not enough that it's not actively running, | |
1101 | * it must be off the runqueue _entirely_, and not | |
1102 | * preempted! | |
1103 | * | |
1104 | * So if it wa still runnable (but just not actively | |
1105 | * running right now), it's preempted, and we should | |
1106 | * yield - it could be a while. | |
1107 | */ | |
dd41f596 | 1108 | if (unlikely(on_rq)) { |
fa490cfd LT |
1109 | yield(); |
1110 | goto repeat; | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * Ahh, all good. It wasn't running, and it wasn't | |
1115 | * runnable, which means that it will never become | |
1116 | * running in the future either. We're all done! | |
1117 | */ | |
1da177e4 LT |
1118 | } |
1119 | ||
1120 | /*** | |
1121 | * kick_process - kick a running thread to enter/exit the kernel | |
1122 | * @p: the to-be-kicked thread | |
1123 | * | |
1124 | * Cause a process which is running on another CPU to enter | |
1125 | * kernel-mode, without any delay. (to get signals handled.) | |
1126 | * | |
1127 | * NOTE: this function doesnt have to take the runqueue lock, | |
1128 | * because all it wants to ensure is that the remote task enters | |
1129 | * the kernel. If the IPI races and the task has been migrated | |
1130 | * to another CPU then no harm is done and the purpose has been | |
1131 | * achieved as well. | |
1132 | */ | |
36c8b586 | 1133 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1134 | { |
1135 | int cpu; | |
1136 | ||
1137 | preempt_disable(); | |
1138 | cpu = task_cpu(p); | |
1139 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1140 | smp_send_reschedule(cpu); | |
1141 | preempt_enable(); | |
1142 | } | |
1143 | ||
1144 | /* | |
2dd73a4f PW |
1145 | * Return a low guess at the load of a migration-source cpu weighted |
1146 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1147 | * |
1148 | * We want to under-estimate the load of migration sources, to | |
1149 | * balance conservatively. | |
1150 | */ | |
a2000572 | 1151 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1152 | { |
70b97a7f | 1153 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1154 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1155 | |
3b0bd9bc | 1156 | if (type == 0) |
dd41f596 | 1157 | return total; |
b910472d | 1158 | |
dd41f596 | 1159 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1160 | } |
1161 | ||
1162 | /* | |
2dd73a4f PW |
1163 | * Return a high guess at the load of a migration-target cpu weighted |
1164 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1165 | */ |
a2000572 | 1166 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1167 | { |
70b97a7f | 1168 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1169 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1170 | |
7897986b | 1171 | if (type == 0) |
dd41f596 | 1172 | return total; |
3b0bd9bc | 1173 | |
dd41f596 | 1174 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1175 | } |
1176 | ||
1177 | /* | |
1178 | * Return the average load per task on the cpu's run queue | |
1179 | */ | |
1180 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1181 | { | |
70b97a7f | 1182 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1183 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1184 | unsigned long n = rq->nr_running; |
1185 | ||
dd41f596 | 1186 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1187 | } |
1188 | ||
147cbb4b NP |
1189 | /* |
1190 | * find_idlest_group finds and returns the least busy CPU group within the | |
1191 | * domain. | |
1192 | */ | |
1193 | static struct sched_group * | |
1194 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1195 | { | |
1196 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1197 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1198 | int load_idx = sd->forkexec_idx; | |
1199 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1200 | ||
1201 | do { | |
1202 | unsigned long load, avg_load; | |
1203 | int local_group; | |
1204 | int i; | |
1205 | ||
da5a5522 BD |
1206 | /* Skip over this group if it has no CPUs allowed */ |
1207 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1208 | goto nextgroup; | |
1209 | ||
147cbb4b | 1210 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1211 | |
1212 | /* Tally up the load of all CPUs in the group */ | |
1213 | avg_load = 0; | |
1214 | ||
1215 | for_each_cpu_mask(i, group->cpumask) { | |
1216 | /* Bias balancing toward cpus of our domain */ | |
1217 | if (local_group) | |
1218 | load = source_load(i, load_idx); | |
1219 | else | |
1220 | load = target_load(i, load_idx); | |
1221 | ||
1222 | avg_load += load; | |
1223 | } | |
1224 | ||
1225 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1226 | avg_load = sg_div_cpu_power(group, |
1227 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1228 | |
1229 | if (local_group) { | |
1230 | this_load = avg_load; | |
1231 | this = group; | |
1232 | } else if (avg_load < min_load) { | |
1233 | min_load = avg_load; | |
1234 | idlest = group; | |
1235 | } | |
da5a5522 | 1236 | nextgroup: |
147cbb4b NP |
1237 | group = group->next; |
1238 | } while (group != sd->groups); | |
1239 | ||
1240 | if (!idlest || 100*this_load < imbalance*min_load) | |
1241 | return NULL; | |
1242 | return idlest; | |
1243 | } | |
1244 | ||
1245 | /* | |
0feaece9 | 1246 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1247 | */ |
95cdf3b7 IM |
1248 | static int |
1249 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1250 | { |
da5a5522 | 1251 | cpumask_t tmp; |
147cbb4b NP |
1252 | unsigned long load, min_load = ULONG_MAX; |
1253 | int idlest = -1; | |
1254 | int i; | |
1255 | ||
da5a5522 BD |
1256 | /* Traverse only the allowed CPUs */ |
1257 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1258 | ||
1259 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1260 | load = weighted_cpuload(i); |
147cbb4b NP |
1261 | |
1262 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1263 | min_load = load; | |
1264 | idlest = i; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | return idlest; | |
1269 | } | |
1270 | ||
476d139c NP |
1271 | /* |
1272 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1273 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1274 | * SD_BALANCE_EXEC. | |
1275 | * | |
1276 | * Balance, ie. select the least loaded group. | |
1277 | * | |
1278 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1279 | * | |
1280 | * preempt must be disabled. | |
1281 | */ | |
1282 | static int sched_balance_self(int cpu, int flag) | |
1283 | { | |
1284 | struct task_struct *t = current; | |
1285 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1286 | |
c96d145e | 1287 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1288 | /* |
1289 | * If power savings logic is enabled for a domain, stop there. | |
1290 | */ | |
5c45bf27 SS |
1291 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1292 | break; | |
476d139c NP |
1293 | if (tmp->flags & flag) |
1294 | sd = tmp; | |
c96d145e | 1295 | } |
476d139c NP |
1296 | |
1297 | while (sd) { | |
1298 | cpumask_t span; | |
1299 | struct sched_group *group; | |
1a848870 SS |
1300 | int new_cpu, weight; |
1301 | ||
1302 | if (!(sd->flags & flag)) { | |
1303 | sd = sd->child; | |
1304 | continue; | |
1305 | } | |
476d139c NP |
1306 | |
1307 | span = sd->span; | |
1308 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1309 | if (!group) { |
1310 | sd = sd->child; | |
1311 | continue; | |
1312 | } | |
476d139c | 1313 | |
da5a5522 | 1314 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1315 | if (new_cpu == -1 || new_cpu == cpu) { |
1316 | /* Now try balancing at a lower domain level of cpu */ | |
1317 | sd = sd->child; | |
1318 | continue; | |
1319 | } | |
476d139c | 1320 | |
1a848870 | 1321 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1322 | cpu = new_cpu; |
476d139c NP |
1323 | sd = NULL; |
1324 | weight = cpus_weight(span); | |
1325 | for_each_domain(cpu, tmp) { | |
1326 | if (weight <= cpus_weight(tmp->span)) | |
1327 | break; | |
1328 | if (tmp->flags & flag) | |
1329 | sd = tmp; | |
1330 | } | |
1331 | /* while loop will break here if sd == NULL */ | |
1332 | } | |
1333 | ||
1334 | return cpu; | |
1335 | } | |
1336 | ||
1337 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1338 | |
1339 | /* | |
1340 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1341 | * not idle and an idle cpu is available. The span of cpus to | |
1342 | * search starts with cpus closest then further out as needed, | |
1343 | * so we always favor a closer, idle cpu. | |
1344 | * | |
1345 | * Returns the CPU we should wake onto. | |
1346 | */ | |
1347 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1348 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1349 | { |
1350 | cpumask_t tmp; | |
1351 | struct sched_domain *sd; | |
1352 | int i; | |
1353 | ||
4953198b SS |
1354 | /* |
1355 | * If it is idle, then it is the best cpu to run this task. | |
1356 | * | |
1357 | * This cpu is also the best, if it has more than one task already. | |
1358 | * Siblings must be also busy(in most cases) as they didn't already | |
1359 | * pickup the extra load from this cpu and hence we need not check | |
1360 | * sibling runqueue info. This will avoid the checks and cache miss | |
1361 | * penalities associated with that. | |
1362 | */ | |
1363 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1364 | return cpu; |
1365 | ||
1366 | for_each_domain(cpu, sd) { | |
1367 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1368 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1369 | for_each_cpu_mask(i, tmp) { |
1370 | if (idle_cpu(i)) | |
1371 | return i; | |
1372 | } | |
9761eea8 | 1373 | } else { |
e0f364f4 | 1374 | break; |
9761eea8 | 1375 | } |
1da177e4 LT |
1376 | } |
1377 | return cpu; | |
1378 | } | |
1379 | #else | |
36c8b586 | 1380 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1381 | { |
1382 | return cpu; | |
1383 | } | |
1384 | #endif | |
1385 | ||
1386 | /*** | |
1387 | * try_to_wake_up - wake up a thread | |
1388 | * @p: the to-be-woken-up thread | |
1389 | * @state: the mask of task states that can be woken | |
1390 | * @sync: do a synchronous wakeup? | |
1391 | * | |
1392 | * Put it on the run-queue if it's not already there. The "current" | |
1393 | * thread is always on the run-queue (except when the actual | |
1394 | * re-schedule is in progress), and as such you're allowed to do | |
1395 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1396 | * runnable without the overhead of this. | |
1397 | * | |
1398 | * returns failure only if the task is already active. | |
1399 | */ | |
36c8b586 | 1400 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1401 | { |
1402 | int cpu, this_cpu, success = 0; | |
1403 | unsigned long flags; | |
1404 | long old_state; | |
70b97a7f | 1405 | struct rq *rq; |
1da177e4 | 1406 | #ifdef CONFIG_SMP |
7897986b | 1407 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1408 | unsigned long load, this_load; |
1da177e4 LT |
1409 | int new_cpu; |
1410 | #endif | |
1411 | ||
1412 | rq = task_rq_lock(p, &flags); | |
1413 | old_state = p->state; | |
1414 | if (!(old_state & state)) | |
1415 | goto out; | |
1416 | ||
dd41f596 | 1417 | if (p->se.on_rq) |
1da177e4 LT |
1418 | goto out_running; |
1419 | ||
1420 | cpu = task_cpu(p); | |
1421 | this_cpu = smp_processor_id(); | |
1422 | ||
1423 | #ifdef CONFIG_SMP | |
1424 | if (unlikely(task_running(rq, p))) | |
1425 | goto out_activate; | |
1426 | ||
7897986b NP |
1427 | new_cpu = cpu; |
1428 | ||
1da177e4 LT |
1429 | schedstat_inc(rq, ttwu_cnt); |
1430 | if (cpu == this_cpu) { | |
1431 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1432 | goto out_set_cpu; |
1433 | } | |
1434 | ||
1435 | for_each_domain(this_cpu, sd) { | |
1436 | if (cpu_isset(cpu, sd->span)) { | |
1437 | schedstat_inc(sd, ttwu_wake_remote); | |
1438 | this_sd = sd; | |
1439 | break; | |
1da177e4 LT |
1440 | } |
1441 | } | |
1da177e4 | 1442 | |
7897986b | 1443 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1444 | goto out_set_cpu; |
1445 | ||
1da177e4 | 1446 | /* |
7897986b | 1447 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1448 | */ |
7897986b NP |
1449 | if (this_sd) { |
1450 | int idx = this_sd->wake_idx; | |
1451 | unsigned int imbalance; | |
1da177e4 | 1452 | |
a3f21bce NP |
1453 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1454 | ||
7897986b NP |
1455 | load = source_load(cpu, idx); |
1456 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1457 | |
7897986b NP |
1458 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1459 | ||
a3f21bce NP |
1460 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1461 | unsigned long tl = this_load; | |
33859f7f MOS |
1462 | unsigned long tl_per_task; |
1463 | ||
1464 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1465 | |
1da177e4 | 1466 | /* |
a3f21bce NP |
1467 | * If sync wakeup then subtract the (maximum possible) |
1468 | * effect of the currently running task from the load | |
1469 | * of the current CPU: | |
1da177e4 | 1470 | */ |
a3f21bce | 1471 | if (sync) |
dd41f596 | 1472 | tl -= current->se.load.weight; |
a3f21bce NP |
1473 | |
1474 | if ((tl <= load && | |
2dd73a4f | 1475 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1476 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1477 | /* |
1478 | * This domain has SD_WAKE_AFFINE and | |
1479 | * p is cache cold in this domain, and | |
1480 | * there is no bad imbalance. | |
1481 | */ | |
1482 | schedstat_inc(this_sd, ttwu_move_affine); | |
1483 | goto out_set_cpu; | |
1484 | } | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Start passive balancing when half the imbalance_pct | |
1489 | * limit is reached. | |
1490 | */ | |
1491 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1492 | if (imbalance*this_load <= 100*load) { | |
1493 | schedstat_inc(this_sd, ttwu_move_balance); | |
1494 | goto out_set_cpu; | |
1495 | } | |
1da177e4 LT |
1496 | } |
1497 | } | |
1498 | ||
1499 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1500 | out_set_cpu: | |
1501 | new_cpu = wake_idle(new_cpu, p); | |
1502 | if (new_cpu != cpu) { | |
1503 | set_task_cpu(p, new_cpu); | |
1504 | task_rq_unlock(rq, &flags); | |
1505 | /* might preempt at this point */ | |
1506 | rq = task_rq_lock(p, &flags); | |
1507 | old_state = p->state; | |
1508 | if (!(old_state & state)) | |
1509 | goto out; | |
dd41f596 | 1510 | if (p->se.on_rq) |
1da177e4 LT |
1511 | goto out_running; |
1512 | ||
1513 | this_cpu = smp_processor_id(); | |
1514 | cpu = task_cpu(p); | |
1515 | } | |
1516 | ||
1517 | out_activate: | |
1518 | #endif /* CONFIG_SMP */ | |
dd41f596 | 1519 | activate_task(rq, p, 1); |
1da177e4 LT |
1520 | /* |
1521 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1522 | * has indicated that it will leave the CPU in short order) | |
1523 | * don't trigger a preemption, if the woken up task will run on | |
1524 | * this cpu. (in this case the 'I will reschedule' promise of | |
1525 | * the waker guarantees that the freshly woken up task is going | |
1526 | * to be considered on this CPU.) | |
1527 | */ | |
dd41f596 IM |
1528 | if (!sync || cpu != this_cpu) |
1529 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1530 | success = 1; |
1531 | ||
1532 | out_running: | |
1533 | p->state = TASK_RUNNING; | |
1534 | out: | |
1535 | task_rq_unlock(rq, &flags); | |
1536 | ||
1537 | return success; | |
1538 | } | |
1539 | ||
36c8b586 | 1540 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1541 | { |
1542 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1543 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1544 | } | |
1da177e4 LT |
1545 | EXPORT_SYMBOL(wake_up_process); |
1546 | ||
36c8b586 | 1547 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1548 | { |
1549 | return try_to_wake_up(p, state, 0); | |
1550 | } | |
1551 | ||
1da177e4 LT |
1552 | /* |
1553 | * Perform scheduler related setup for a newly forked process p. | |
1554 | * p is forked by current. | |
dd41f596 IM |
1555 | * |
1556 | * __sched_fork() is basic setup used by init_idle() too: | |
1557 | */ | |
1558 | static void __sched_fork(struct task_struct *p) | |
1559 | { | |
1560 | p->se.wait_start_fair = 0; | |
dd41f596 IM |
1561 | p->se.exec_start = 0; |
1562 | p->se.sum_exec_runtime = 0; | |
1563 | p->se.delta_exec = 0; | |
1564 | p->se.delta_fair_run = 0; | |
1565 | p->se.delta_fair_sleep = 0; | |
1566 | p->se.wait_runtime = 0; | |
6cfb0d5d IM |
1567 | p->se.sleep_start_fair = 0; |
1568 | ||
1569 | #ifdef CONFIG_SCHEDSTATS | |
1570 | p->se.wait_start = 0; | |
dd41f596 IM |
1571 | p->se.sum_wait_runtime = 0; |
1572 | p->se.sum_sleep_runtime = 0; | |
1573 | p->se.sleep_start = 0; | |
dd41f596 IM |
1574 | p->se.block_start = 0; |
1575 | p->se.sleep_max = 0; | |
1576 | p->se.block_max = 0; | |
1577 | p->se.exec_max = 0; | |
1578 | p->se.wait_max = 0; | |
1579 | p->se.wait_runtime_overruns = 0; | |
1580 | p->se.wait_runtime_underruns = 0; | |
6cfb0d5d | 1581 | #endif |
476d139c | 1582 | |
dd41f596 IM |
1583 | INIT_LIST_HEAD(&p->run_list); |
1584 | p->se.on_rq = 0; | |
476d139c | 1585 | |
e107be36 AK |
1586 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1587 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1588 | #endif | |
1589 | ||
1da177e4 LT |
1590 | /* |
1591 | * We mark the process as running here, but have not actually | |
1592 | * inserted it onto the runqueue yet. This guarantees that | |
1593 | * nobody will actually run it, and a signal or other external | |
1594 | * event cannot wake it up and insert it on the runqueue either. | |
1595 | */ | |
1596 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1597 | } |
1598 | ||
1599 | /* | |
1600 | * fork()/clone()-time setup: | |
1601 | */ | |
1602 | void sched_fork(struct task_struct *p, int clone_flags) | |
1603 | { | |
1604 | int cpu = get_cpu(); | |
1605 | ||
1606 | __sched_fork(p); | |
1607 | ||
1608 | #ifdef CONFIG_SMP | |
1609 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1610 | #endif | |
1611 | __set_task_cpu(p, cpu); | |
b29739f9 IM |
1612 | |
1613 | /* | |
1614 | * Make sure we do not leak PI boosting priority to the child: | |
1615 | */ | |
1616 | p->prio = current->normal_prio; | |
1617 | ||
52f17b6c | 1618 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1619 | if (likely(sched_info_on())) |
52f17b6c | 1620 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1621 | #endif |
d6077cb8 | 1622 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1623 | p->oncpu = 0; |
1624 | #endif | |
1da177e4 | 1625 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1626 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1627 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1628 | #endif |
476d139c | 1629 | put_cpu(); |
1da177e4 LT |
1630 | } |
1631 | ||
dd41f596 IM |
1632 | /* |
1633 | * After fork, child runs first. (default) If set to 0 then | |
1634 | * parent will (try to) run first. | |
1635 | */ | |
1636 | unsigned int __read_mostly sysctl_sched_child_runs_first = 1; | |
1637 | ||
1da177e4 LT |
1638 | /* |
1639 | * wake_up_new_task - wake up a newly created task for the first time. | |
1640 | * | |
1641 | * This function will do some initial scheduler statistics housekeeping | |
1642 | * that must be done for every newly created context, then puts the task | |
1643 | * on the runqueue and wakes it. | |
1644 | */ | |
36c8b586 | 1645 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1646 | { |
1647 | unsigned long flags; | |
dd41f596 IM |
1648 | struct rq *rq; |
1649 | int this_cpu; | |
cad60d93 | 1650 | u64 now; |
1da177e4 LT |
1651 | |
1652 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1653 | BUG_ON(p->state != TASK_RUNNING); |
dd41f596 | 1654 | this_cpu = smp_processor_id(); /* parent's CPU */ |
cad60d93 | 1655 | now = rq_clock(rq); |
1da177e4 LT |
1656 | |
1657 | p->prio = effective_prio(p); | |
1658 | ||
cad60d93 IM |
1659 | if (!p->sched_class->task_new || !sysctl_sched_child_runs_first || |
1660 | (clone_flags & CLONE_VM) || task_cpu(p) != this_cpu || | |
1661 | !current->se.on_rq) { | |
1662 | ||
dd41f596 | 1663 | activate_task(rq, p, 0); |
1da177e4 | 1664 | } else { |
1da177e4 | 1665 | /* |
dd41f596 IM |
1666 | * Let the scheduling class do new task startup |
1667 | * management (if any): | |
1da177e4 | 1668 | */ |
cad60d93 IM |
1669 | p->sched_class->task_new(rq, p, now); |
1670 | inc_nr_running(p, rq, now); | |
1da177e4 | 1671 | } |
dd41f596 IM |
1672 | check_preempt_curr(rq, p); |
1673 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1674 | } |
1675 | ||
e107be36 AK |
1676 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1677 | ||
1678 | /** | |
421cee29 RD |
1679 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1680 | * @notifier: notifier struct to register | |
e107be36 AK |
1681 | */ |
1682 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1683 | { | |
1684 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1685 | } | |
1686 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1687 | ||
1688 | /** | |
1689 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1690 | * @notifier: notifier struct to unregister |
e107be36 AK |
1691 | * |
1692 | * This is safe to call from within a preemption notifier. | |
1693 | */ | |
1694 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1695 | { | |
1696 | hlist_del(¬ifier->link); | |
1697 | } | |
1698 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1699 | ||
1700 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1701 | { | |
1702 | struct preempt_notifier *notifier; | |
1703 | struct hlist_node *node; | |
1704 | ||
1705 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1706 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1707 | } | |
1708 | ||
1709 | static void | |
1710 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1711 | struct task_struct *next) | |
1712 | { | |
1713 | struct preempt_notifier *notifier; | |
1714 | struct hlist_node *node; | |
1715 | ||
1716 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1717 | notifier->ops->sched_out(notifier, next); | |
1718 | } | |
1719 | ||
1720 | #else | |
1721 | ||
1722 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1723 | { | |
1724 | } | |
1725 | ||
1726 | static void | |
1727 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1728 | struct task_struct *next) | |
1729 | { | |
1730 | } | |
1731 | ||
1732 | #endif | |
1733 | ||
4866cde0 NP |
1734 | /** |
1735 | * prepare_task_switch - prepare to switch tasks | |
1736 | * @rq: the runqueue preparing to switch | |
421cee29 | 1737 | * @prev: the current task that is being switched out |
4866cde0 NP |
1738 | * @next: the task we are going to switch to. |
1739 | * | |
1740 | * This is called with the rq lock held and interrupts off. It must | |
1741 | * be paired with a subsequent finish_task_switch after the context | |
1742 | * switch. | |
1743 | * | |
1744 | * prepare_task_switch sets up locking and calls architecture specific | |
1745 | * hooks. | |
1746 | */ | |
e107be36 AK |
1747 | static inline void |
1748 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1749 | struct task_struct *next) | |
4866cde0 | 1750 | { |
e107be36 | 1751 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1752 | prepare_lock_switch(rq, next); |
1753 | prepare_arch_switch(next); | |
1754 | } | |
1755 | ||
1da177e4 LT |
1756 | /** |
1757 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1758 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1759 | * @prev: the thread we just switched away from. |
1760 | * | |
4866cde0 NP |
1761 | * finish_task_switch must be called after the context switch, paired |
1762 | * with a prepare_task_switch call before the context switch. | |
1763 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1764 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1765 | * |
1766 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1767 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1768 | * with the lock held can cause deadlocks; see schedule() for | |
1769 | * details.) | |
1770 | */ | |
70b97a7f | 1771 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1772 | __releases(rq->lock) |
1773 | { | |
1da177e4 | 1774 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1775 | long prev_state; |
1da177e4 LT |
1776 | |
1777 | rq->prev_mm = NULL; | |
1778 | ||
1779 | /* | |
1780 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1781 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1782 | * schedule one last time. The schedule call will never return, and |
1783 | * the scheduled task must drop that reference. | |
c394cc9f | 1784 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1785 | * still held, otherwise prev could be scheduled on another cpu, die |
1786 | * there before we look at prev->state, and then the reference would | |
1787 | * be dropped twice. | |
1788 | * Manfred Spraul <manfred@colorfullife.com> | |
1789 | */ | |
55a101f8 | 1790 | prev_state = prev->state; |
4866cde0 NP |
1791 | finish_arch_switch(prev); |
1792 | finish_lock_switch(rq, prev); | |
e107be36 | 1793 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1794 | if (mm) |
1795 | mmdrop(mm); | |
c394cc9f | 1796 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1797 | /* |
1798 | * Remove function-return probe instances associated with this | |
1799 | * task and put them back on the free list. | |
9761eea8 | 1800 | */ |
c6fd91f0 | 1801 | kprobe_flush_task(prev); |
1da177e4 | 1802 | put_task_struct(prev); |
c6fd91f0 | 1803 | } |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | /** | |
1807 | * schedule_tail - first thing a freshly forked thread must call. | |
1808 | * @prev: the thread we just switched away from. | |
1809 | */ | |
36c8b586 | 1810 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1811 | __releases(rq->lock) |
1812 | { | |
70b97a7f IM |
1813 | struct rq *rq = this_rq(); |
1814 | ||
4866cde0 NP |
1815 | finish_task_switch(rq, prev); |
1816 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1817 | /* In this case, finish_task_switch does not reenable preemption */ | |
1818 | preempt_enable(); | |
1819 | #endif | |
1da177e4 LT |
1820 | if (current->set_child_tid) |
1821 | put_user(current->pid, current->set_child_tid); | |
1822 | } | |
1823 | ||
1824 | /* | |
1825 | * context_switch - switch to the new MM and the new | |
1826 | * thread's register state. | |
1827 | */ | |
dd41f596 | 1828 | static inline void |
70b97a7f | 1829 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1830 | struct task_struct *next) |
1da177e4 | 1831 | { |
dd41f596 | 1832 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1833 | |
e107be36 | 1834 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1835 | mm = next->mm; |
1836 | oldmm = prev->active_mm; | |
9226d125 ZA |
1837 | /* |
1838 | * For paravirt, this is coupled with an exit in switch_to to | |
1839 | * combine the page table reload and the switch backend into | |
1840 | * one hypercall. | |
1841 | */ | |
1842 | arch_enter_lazy_cpu_mode(); | |
1843 | ||
dd41f596 | 1844 | if (unlikely(!mm)) { |
1da177e4 LT |
1845 | next->active_mm = oldmm; |
1846 | atomic_inc(&oldmm->mm_count); | |
1847 | enter_lazy_tlb(oldmm, next); | |
1848 | } else | |
1849 | switch_mm(oldmm, mm, next); | |
1850 | ||
dd41f596 | 1851 | if (unlikely(!prev->mm)) { |
1da177e4 | 1852 | prev->active_mm = NULL; |
1da177e4 LT |
1853 | rq->prev_mm = oldmm; |
1854 | } | |
3a5f5e48 IM |
1855 | /* |
1856 | * Since the runqueue lock will be released by the next | |
1857 | * task (which is an invalid locking op but in the case | |
1858 | * of the scheduler it's an obvious special-case), so we | |
1859 | * do an early lockdep release here: | |
1860 | */ | |
1861 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1862 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1863 | #endif |
1da177e4 LT |
1864 | |
1865 | /* Here we just switch the register state and the stack. */ | |
1866 | switch_to(prev, next, prev); | |
1867 | ||
dd41f596 IM |
1868 | barrier(); |
1869 | /* | |
1870 | * this_rq must be evaluated again because prev may have moved | |
1871 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1872 | * frame will be invalid. | |
1873 | */ | |
1874 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1875 | } |
1876 | ||
1877 | /* | |
1878 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1879 | * | |
1880 | * externally visible scheduler statistics: current number of runnable | |
1881 | * threads, current number of uninterruptible-sleeping threads, total | |
1882 | * number of context switches performed since bootup. | |
1883 | */ | |
1884 | unsigned long nr_running(void) | |
1885 | { | |
1886 | unsigned long i, sum = 0; | |
1887 | ||
1888 | for_each_online_cpu(i) | |
1889 | sum += cpu_rq(i)->nr_running; | |
1890 | ||
1891 | return sum; | |
1892 | } | |
1893 | ||
1894 | unsigned long nr_uninterruptible(void) | |
1895 | { | |
1896 | unsigned long i, sum = 0; | |
1897 | ||
0a945022 | 1898 | for_each_possible_cpu(i) |
1da177e4 LT |
1899 | sum += cpu_rq(i)->nr_uninterruptible; |
1900 | ||
1901 | /* | |
1902 | * Since we read the counters lockless, it might be slightly | |
1903 | * inaccurate. Do not allow it to go below zero though: | |
1904 | */ | |
1905 | if (unlikely((long)sum < 0)) | |
1906 | sum = 0; | |
1907 | ||
1908 | return sum; | |
1909 | } | |
1910 | ||
1911 | unsigned long long nr_context_switches(void) | |
1912 | { | |
cc94abfc SR |
1913 | int i; |
1914 | unsigned long long sum = 0; | |
1da177e4 | 1915 | |
0a945022 | 1916 | for_each_possible_cpu(i) |
1da177e4 LT |
1917 | sum += cpu_rq(i)->nr_switches; |
1918 | ||
1919 | return sum; | |
1920 | } | |
1921 | ||
1922 | unsigned long nr_iowait(void) | |
1923 | { | |
1924 | unsigned long i, sum = 0; | |
1925 | ||
0a945022 | 1926 | for_each_possible_cpu(i) |
1da177e4 LT |
1927 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1928 | ||
1929 | return sum; | |
1930 | } | |
1931 | ||
db1b1fef JS |
1932 | unsigned long nr_active(void) |
1933 | { | |
1934 | unsigned long i, running = 0, uninterruptible = 0; | |
1935 | ||
1936 | for_each_online_cpu(i) { | |
1937 | running += cpu_rq(i)->nr_running; | |
1938 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1939 | } | |
1940 | ||
1941 | if (unlikely((long)uninterruptible < 0)) | |
1942 | uninterruptible = 0; | |
1943 | ||
1944 | return running + uninterruptible; | |
1945 | } | |
1946 | ||
48f24c4d | 1947 | /* |
dd41f596 IM |
1948 | * Update rq->cpu_load[] statistics. This function is usually called every |
1949 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1950 | */ |
dd41f596 | 1951 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1952 | { |
dd41f596 IM |
1953 | u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64; |
1954 | unsigned long total_load = this_rq->ls.load.weight; | |
1955 | unsigned long this_load = total_load; | |
1956 | struct load_stat *ls = &this_rq->ls; | |
1957 | u64 now = __rq_clock(this_rq); | |
1958 | int i, scale; | |
1959 | ||
1960 | this_rq->nr_load_updates++; | |
1961 | if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD))) | |
1962 | goto do_avg; | |
1963 | ||
1964 | /* Update delta_fair/delta_exec fields first */ | |
1965 | update_curr_load(this_rq, now); | |
1966 | ||
1967 | fair_delta64 = ls->delta_fair + 1; | |
1968 | ls->delta_fair = 0; | |
1969 | ||
1970 | exec_delta64 = ls->delta_exec + 1; | |
1971 | ls->delta_exec = 0; | |
1972 | ||
1973 | sample_interval64 = now - ls->load_update_last; | |
1974 | ls->load_update_last = now; | |
1975 | ||
1976 | if ((s64)sample_interval64 < (s64)TICK_NSEC) | |
1977 | sample_interval64 = TICK_NSEC; | |
1978 | ||
1979 | if (exec_delta64 > sample_interval64) | |
1980 | exec_delta64 = sample_interval64; | |
1981 | ||
1982 | idle_delta64 = sample_interval64 - exec_delta64; | |
1983 | ||
1984 | tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64); | |
1985 | tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64); | |
1986 | ||
1987 | this_load = (unsigned long)tmp64; | |
1988 | ||
1989 | do_avg: | |
1990 | ||
1991 | /* Update our load: */ | |
1992 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1993 | unsigned long old_load, new_load; | |
1994 | ||
1995 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1996 | ||
1997 | old_load = this_rq->cpu_load[i]; | |
1998 | new_load = this_load; | |
1999 | ||
2000 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; | |
2001 | } | |
48f24c4d IM |
2002 | } |
2003 | ||
dd41f596 IM |
2004 | #ifdef CONFIG_SMP |
2005 | ||
1da177e4 LT |
2006 | /* |
2007 | * double_rq_lock - safely lock two runqueues | |
2008 | * | |
2009 | * Note this does not disable interrupts like task_rq_lock, | |
2010 | * you need to do so manually before calling. | |
2011 | */ | |
70b97a7f | 2012 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2013 | __acquires(rq1->lock) |
2014 | __acquires(rq2->lock) | |
2015 | { | |
054b9108 | 2016 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2017 | if (rq1 == rq2) { |
2018 | spin_lock(&rq1->lock); | |
2019 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2020 | } else { | |
c96d145e | 2021 | if (rq1 < rq2) { |
1da177e4 LT |
2022 | spin_lock(&rq1->lock); |
2023 | spin_lock(&rq2->lock); | |
2024 | } else { | |
2025 | spin_lock(&rq2->lock); | |
2026 | spin_lock(&rq1->lock); | |
2027 | } | |
2028 | } | |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * double_rq_unlock - safely unlock two runqueues | |
2033 | * | |
2034 | * Note this does not restore interrupts like task_rq_unlock, | |
2035 | * you need to do so manually after calling. | |
2036 | */ | |
70b97a7f | 2037 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2038 | __releases(rq1->lock) |
2039 | __releases(rq2->lock) | |
2040 | { | |
2041 | spin_unlock(&rq1->lock); | |
2042 | if (rq1 != rq2) | |
2043 | spin_unlock(&rq2->lock); | |
2044 | else | |
2045 | __release(rq2->lock); | |
2046 | } | |
2047 | ||
2048 | /* | |
2049 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2050 | */ | |
70b97a7f | 2051 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2052 | __releases(this_rq->lock) |
2053 | __acquires(busiest->lock) | |
2054 | __acquires(this_rq->lock) | |
2055 | { | |
054b9108 KK |
2056 | if (unlikely(!irqs_disabled())) { |
2057 | /* printk() doesn't work good under rq->lock */ | |
2058 | spin_unlock(&this_rq->lock); | |
2059 | BUG_ON(1); | |
2060 | } | |
1da177e4 | 2061 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2062 | if (busiest < this_rq) { |
1da177e4 LT |
2063 | spin_unlock(&this_rq->lock); |
2064 | spin_lock(&busiest->lock); | |
2065 | spin_lock(&this_rq->lock); | |
2066 | } else | |
2067 | spin_lock(&busiest->lock); | |
2068 | } | |
2069 | } | |
2070 | ||
1da177e4 LT |
2071 | /* |
2072 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2073 | * This is accomplished by forcing the cpu_allowed mask to only | |
2074 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2075 | * the cpu_allowed mask is restored. | |
2076 | */ | |
36c8b586 | 2077 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2078 | { |
70b97a7f | 2079 | struct migration_req req; |
1da177e4 | 2080 | unsigned long flags; |
70b97a7f | 2081 | struct rq *rq; |
1da177e4 LT |
2082 | |
2083 | rq = task_rq_lock(p, &flags); | |
2084 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2085 | || unlikely(cpu_is_offline(dest_cpu))) | |
2086 | goto out; | |
2087 | ||
2088 | /* force the process onto the specified CPU */ | |
2089 | if (migrate_task(p, dest_cpu, &req)) { | |
2090 | /* Need to wait for migration thread (might exit: take ref). */ | |
2091 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2092 | |
1da177e4 LT |
2093 | get_task_struct(mt); |
2094 | task_rq_unlock(rq, &flags); | |
2095 | wake_up_process(mt); | |
2096 | put_task_struct(mt); | |
2097 | wait_for_completion(&req.done); | |
36c8b586 | 2098 | |
1da177e4 LT |
2099 | return; |
2100 | } | |
2101 | out: | |
2102 | task_rq_unlock(rq, &flags); | |
2103 | } | |
2104 | ||
2105 | /* | |
476d139c NP |
2106 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2107 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2108 | */ |
2109 | void sched_exec(void) | |
2110 | { | |
1da177e4 | 2111 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2112 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2113 | put_cpu(); |
476d139c NP |
2114 | if (new_cpu != this_cpu) |
2115 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2120 | * Both runqueues must be locked. | |
2121 | */ | |
dd41f596 IM |
2122 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2123 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2124 | { |
dd41f596 | 2125 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2126 | set_task_cpu(p, this_cpu); |
dd41f596 | 2127 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2128 | /* |
2129 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2130 | * to be always true for them. | |
2131 | */ | |
dd41f596 | 2132 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2133 | } |
2134 | ||
2135 | /* | |
2136 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2137 | */ | |
858119e1 | 2138 | static |
70b97a7f | 2139 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2140 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2141 | int *all_pinned) |
1da177e4 LT |
2142 | { |
2143 | /* | |
2144 | * We do not migrate tasks that are: | |
2145 | * 1) running (obviously), or | |
2146 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2147 | * 3) are cache-hot on their current CPU. | |
2148 | */ | |
1da177e4 LT |
2149 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2150 | return 0; | |
81026794 NP |
2151 | *all_pinned = 0; |
2152 | ||
2153 | if (task_running(rq, p)) | |
2154 | return 0; | |
1da177e4 LT |
2155 | |
2156 | /* | |
dd41f596 | 2157 | * Aggressive migration if too many balance attempts have failed: |
1da177e4 | 2158 | */ |
dd41f596 | 2159 | if (sd->nr_balance_failed > sd->cache_nice_tries) |
1da177e4 LT |
2160 | return 1; |
2161 | ||
1da177e4 LT |
2162 | return 1; |
2163 | } | |
2164 | ||
dd41f596 | 2165 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2166 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2167 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 IM |
2168 | int *all_pinned, unsigned long *load_moved, |
2169 | int this_best_prio, int best_prio, int best_prio_seen, | |
2170 | struct rq_iterator *iterator) | |
1da177e4 | 2171 | { |
dd41f596 IM |
2172 | int pulled = 0, pinned = 0, skip_for_load; |
2173 | struct task_struct *p; | |
2174 | long rem_load_move = max_load_move; | |
1da177e4 | 2175 | |
2dd73a4f | 2176 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2177 | goto out; |
2178 | ||
81026794 NP |
2179 | pinned = 1; |
2180 | ||
1da177e4 | 2181 | /* |
dd41f596 | 2182 | * Start the load-balancing iterator: |
1da177e4 | 2183 | */ |
dd41f596 IM |
2184 | p = iterator->start(iterator->arg); |
2185 | next: | |
2186 | if (!p) | |
1da177e4 | 2187 | goto out; |
50ddd969 PW |
2188 | /* |
2189 | * To help distribute high priority tasks accross CPUs we don't | |
2190 | * skip a task if it will be the highest priority task (i.e. smallest | |
2191 | * prio value) on its new queue regardless of its load weight | |
2192 | */ | |
dd41f596 IM |
2193 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2194 | SCHED_LOAD_SCALE_FUZZ; | |
2195 | if (skip_for_load && p->prio < this_best_prio) | |
2196 | skip_for_load = !best_prio_seen && p->prio == best_prio; | |
615052dc | 2197 | if (skip_for_load || |
dd41f596 | 2198 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
48f24c4d | 2199 | |
dd41f596 IM |
2200 | best_prio_seen |= p->prio == best_prio; |
2201 | p = iterator->next(iterator->arg); | |
2202 | goto next; | |
1da177e4 LT |
2203 | } |
2204 | ||
dd41f596 | 2205 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2206 | pulled++; |
dd41f596 | 2207 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2208 | |
2dd73a4f PW |
2209 | /* |
2210 | * We only want to steal up to the prescribed number of tasks | |
2211 | * and the prescribed amount of weighted load. | |
2212 | */ | |
2213 | if (pulled < max_nr_move && rem_load_move > 0) { | |
dd41f596 IM |
2214 | if (p->prio < this_best_prio) |
2215 | this_best_prio = p->prio; | |
2216 | p = iterator->next(iterator->arg); | |
2217 | goto next; | |
1da177e4 LT |
2218 | } |
2219 | out: | |
2220 | /* | |
2221 | * Right now, this is the only place pull_task() is called, | |
2222 | * so we can safely collect pull_task() stats here rather than | |
2223 | * inside pull_task(). | |
2224 | */ | |
2225 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2226 | |
2227 | if (all_pinned) | |
2228 | *all_pinned = pinned; | |
dd41f596 | 2229 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2230 | return pulled; |
2231 | } | |
2232 | ||
dd41f596 IM |
2233 | /* |
2234 | * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted | |
2235 | * load from busiest to this_rq, as part of a balancing operation within | |
2236 | * "domain". Returns the number of tasks moved. | |
2237 | * | |
2238 | * Called with both runqueues locked. | |
2239 | */ | |
2240 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2241 | unsigned long max_nr_move, unsigned long max_load_move, | |
2242 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2243 | int *all_pinned) | |
2244 | { | |
2245 | struct sched_class *class = sched_class_highest; | |
2246 | unsigned long load_moved, total_nr_moved = 0, nr_moved; | |
2247 | long rem_load_move = max_load_move; | |
2248 | ||
2249 | do { | |
2250 | nr_moved = class->load_balance(this_rq, this_cpu, busiest, | |
2251 | max_nr_move, (unsigned long)rem_load_move, | |
2252 | sd, idle, all_pinned, &load_moved); | |
2253 | total_nr_moved += nr_moved; | |
2254 | max_nr_move -= nr_moved; | |
2255 | rem_load_move -= load_moved; | |
2256 | class = class->next; | |
2257 | } while (class && max_nr_move && rem_load_move > 0); | |
2258 | ||
2259 | return total_nr_moved; | |
2260 | } | |
2261 | ||
1da177e4 LT |
2262 | /* |
2263 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2264 | * domain. It calculates and returns the amount of weighted load which |
2265 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2266 | */ |
2267 | static struct sched_group * | |
2268 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2269 | unsigned long *imbalance, enum cpu_idle_type idle, |
2270 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2271 | { |
2272 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2273 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2274 | unsigned long max_pull; |
2dd73a4f PW |
2275 | unsigned long busiest_load_per_task, busiest_nr_running; |
2276 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2277 | int load_idx; |
5c45bf27 SS |
2278 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2279 | int power_savings_balance = 1; | |
2280 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2281 | unsigned long min_nr_running = ULONG_MAX; | |
2282 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2283 | #endif | |
1da177e4 LT |
2284 | |
2285 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2286 | busiest_load_per_task = busiest_nr_running = 0; |
2287 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2288 | if (idle == CPU_NOT_IDLE) |
7897986b | 2289 | load_idx = sd->busy_idx; |
d15bcfdb | 2290 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2291 | load_idx = sd->newidle_idx; |
2292 | else | |
2293 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2294 | |
2295 | do { | |
5c45bf27 | 2296 | unsigned long load, group_capacity; |
1da177e4 LT |
2297 | int local_group; |
2298 | int i; | |
783609c6 | 2299 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2300 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2301 | |
2302 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2303 | ||
783609c6 SS |
2304 | if (local_group) |
2305 | balance_cpu = first_cpu(group->cpumask); | |
2306 | ||
1da177e4 | 2307 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2308 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2309 | |
2310 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2311 | struct rq *rq; |
2312 | ||
2313 | if (!cpu_isset(i, *cpus)) | |
2314 | continue; | |
2315 | ||
2316 | rq = cpu_rq(i); | |
2dd73a4f | 2317 | |
9439aab8 | 2318 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2319 | *sd_idle = 0; |
2320 | ||
1da177e4 | 2321 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2322 | if (local_group) { |
2323 | if (idle_cpu(i) && !first_idle_cpu) { | |
2324 | first_idle_cpu = 1; | |
2325 | balance_cpu = i; | |
2326 | } | |
2327 | ||
a2000572 | 2328 | load = target_load(i, load_idx); |
783609c6 | 2329 | } else |
a2000572 | 2330 | load = source_load(i, load_idx); |
1da177e4 LT |
2331 | |
2332 | avg_load += load; | |
2dd73a4f | 2333 | sum_nr_running += rq->nr_running; |
dd41f596 | 2334 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2335 | } |
2336 | ||
783609c6 SS |
2337 | /* |
2338 | * First idle cpu or the first cpu(busiest) in this sched group | |
2339 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2340 | * domains. In the newly idle case, we will allow all the cpu's |
2341 | * to do the newly idle load balance. | |
783609c6 | 2342 | */ |
9439aab8 SS |
2343 | if (idle != CPU_NEWLY_IDLE && local_group && |
2344 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2345 | *balance = 0; |
2346 | goto ret; | |
2347 | } | |
2348 | ||
1da177e4 | 2349 | total_load += avg_load; |
5517d86b | 2350 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2351 | |
2352 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2353 | avg_load = sg_div_cpu_power(group, |
2354 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2355 | |
5517d86b | 2356 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2357 | |
1da177e4 LT |
2358 | if (local_group) { |
2359 | this_load = avg_load; | |
2360 | this = group; | |
2dd73a4f PW |
2361 | this_nr_running = sum_nr_running; |
2362 | this_load_per_task = sum_weighted_load; | |
2363 | } else if (avg_load > max_load && | |
5c45bf27 | 2364 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2365 | max_load = avg_load; |
2366 | busiest = group; | |
2dd73a4f PW |
2367 | busiest_nr_running = sum_nr_running; |
2368 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2369 | } |
5c45bf27 SS |
2370 | |
2371 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2372 | /* | |
2373 | * Busy processors will not participate in power savings | |
2374 | * balance. | |
2375 | */ | |
dd41f596 IM |
2376 | if (idle == CPU_NOT_IDLE || |
2377 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2378 | goto group_next; | |
5c45bf27 SS |
2379 | |
2380 | /* | |
2381 | * If the local group is idle or completely loaded | |
2382 | * no need to do power savings balance at this domain | |
2383 | */ | |
2384 | if (local_group && (this_nr_running >= group_capacity || | |
2385 | !this_nr_running)) | |
2386 | power_savings_balance = 0; | |
2387 | ||
dd41f596 | 2388 | /* |
5c45bf27 SS |
2389 | * If a group is already running at full capacity or idle, |
2390 | * don't include that group in power savings calculations | |
dd41f596 IM |
2391 | */ |
2392 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2393 | || !sum_nr_running) |
dd41f596 | 2394 | goto group_next; |
5c45bf27 | 2395 | |
dd41f596 | 2396 | /* |
5c45bf27 | 2397 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2398 | * This is the group from where we need to pick up the load |
2399 | * for saving power | |
2400 | */ | |
2401 | if ((sum_nr_running < min_nr_running) || | |
2402 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2403 | first_cpu(group->cpumask) < |
2404 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2405 | group_min = group; |
2406 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2407 | min_load_per_task = sum_weighted_load / |
2408 | sum_nr_running; | |
dd41f596 | 2409 | } |
5c45bf27 | 2410 | |
dd41f596 | 2411 | /* |
5c45bf27 | 2412 | * Calculate the group which is almost near its |
dd41f596 IM |
2413 | * capacity but still has some space to pick up some load |
2414 | * from other group and save more power | |
2415 | */ | |
2416 | if (sum_nr_running <= group_capacity - 1) { | |
2417 | if (sum_nr_running > leader_nr_running || | |
2418 | (sum_nr_running == leader_nr_running && | |
2419 | first_cpu(group->cpumask) > | |
2420 | first_cpu(group_leader->cpumask))) { | |
2421 | group_leader = group; | |
2422 | leader_nr_running = sum_nr_running; | |
2423 | } | |
48f24c4d | 2424 | } |
5c45bf27 SS |
2425 | group_next: |
2426 | #endif | |
1da177e4 LT |
2427 | group = group->next; |
2428 | } while (group != sd->groups); | |
2429 | ||
2dd73a4f | 2430 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2431 | goto out_balanced; |
2432 | ||
2433 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2434 | ||
2435 | if (this_load >= avg_load || | |
2436 | 100*max_load <= sd->imbalance_pct*this_load) | |
2437 | goto out_balanced; | |
2438 | ||
2dd73a4f | 2439 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2440 | /* |
2441 | * We're trying to get all the cpus to the average_load, so we don't | |
2442 | * want to push ourselves above the average load, nor do we wish to | |
2443 | * reduce the max loaded cpu below the average load, as either of these | |
2444 | * actions would just result in more rebalancing later, and ping-pong | |
2445 | * tasks around. Thus we look for the minimum possible imbalance. | |
2446 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2447 | * be counted as no imbalance for these purposes -- we can't fix that | |
2448 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2449 | * appear as very large values with unsigned longs. | |
2450 | */ | |
2dd73a4f PW |
2451 | if (max_load <= busiest_load_per_task) |
2452 | goto out_balanced; | |
2453 | ||
2454 | /* | |
2455 | * In the presence of smp nice balancing, certain scenarios can have | |
2456 | * max load less than avg load(as we skip the groups at or below | |
2457 | * its cpu_power, while calculating max_load..) | |
2458 | */ | |
2459 | if (max_load < avg_load) { | |
2460 | *imbalance = 0; | |
2461 | goto small_imbalance; | |
2462 | } | |
0c117f1b SS |
2463 | |
2464 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2465 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2466 | |
1da177e4 | 2467 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2468 | *imbalance = min(max_pull * busiest->__cpu_power, |
2469 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2470 | / SCHED_LOAD_SCALE; |
2471 | ||
2dd73a4f PW |
2472 | /* |
2473 | * if *imbalance is less than the average load per runnable task | |
2474 | * there is no gaurantee that any tasks will be moved so we'll have | |
2475 | * a think about bumping its value to force at least one task to be | |
2476 | * moved | |
2477 | */ | |
dd41f596 | 2478 | if (*imbalance + SCHED_LOAD_SCALE_FUZZ < busiest_load_per_task/2) { |
48f24c4d | 2479 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2480 | unsigned int imbn; |
2481 | ||
2482 | small_imbalance: | |
2483 | pwr_move = pwr_now = 0; | |
2484 | imbn = 2; | |
2485 | if (this_nr_running) { | |
2486 | this_load_per_task /= this_nr_running; | |
2487 | if (busiest_load_per_task > this_load_per_task) | |
2488 | imbn = 1; | |
2489 | } else | |
2490 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2491 | |
dd41f596 IM |
2492 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2493 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2494 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2495 | return busiest; |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * OK, we don't have enough imbalance to justify moving tasks, | |
2500 | * however we may be able to increase total CPU power used by | |
2501 | * moving them. | |
2502 | */ | |
2503 | ||
5517d86b ED |
2504 | pwr_now += busiest->__cpu_power * |
2505 | min(busiest_load_per_task, max_load); | |
2506 | pwr_now += this->__cpu_power * | |
2507 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2508 | pwr_now /= SCHED_LOAD_SCALE; |
2509 | ||
2510 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2511 | tmp = sg_div_cpu_power(busiest, |
2512 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2513 | if (max_load > tmp) |
5517d86b | 2514 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2515 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2516 | |
2517 | /* Amount of load we'd add */ | |
5517d86b | 2518 | if (max_load * busiest->__cpu_power < |
33859f7f | 2519 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2520 | tmp = sg_div_cpu_power(this, |
2521 | max_load * busiest->__cpu_power); | |
1da177e4 | 2522 | else |
5517d86b ED |
2523 | tmp = sg_div_cpu_power(this, |
2524 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2525 | pwr_move += this->__cpu_power * | |
2526 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2527 | pwr_move /= SCHED_LOAD_SCALE; |
2528 | ||
2529 | /* Move if we gain throughput */ | |
2530 | if (pwr_move <= pwr_now) | |
2531 | goto out_balanced; | |
2532 | ||
2dd73a4f | 2533 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2534 | } |
2535 | ||
1da177e4 LT |
2536 | return busiest; |
2537 | ||
2538 | out_balanced: | |
5c45bf27 | 2539 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2540 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2541 | goto ret; |
1da177e4 | 2542 | |
5c45bf27 SS |
2543 | if (this == group_leader && group_leader != group_min) { |
2544 | *imbalance = min_load_per_task; | |
2545 | return group_min; | |
2546 | } | |
5c45bf27 | 2547 | #endif |
783609c6 | 2548 | ret: |
1da177e4 LT |
2549 | *imbalance = 0; |
2550 | return NULL; | |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2555 | */ | |
70b97a7f | 2556 | static struct rq * |
d15bcfdb | 2557 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2558 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2559 | { |
70b97a7f | 2560 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2561 | unsigned long max_load = 0; |
1da177e4 LT |
2562 | int i; |
2563 | ||
2564 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2565 | unsigned long wl; |
0a2966b4 CL |
2566 | |
2567 | if (!cpu_isset(i, *cpus)) | |
2568 | continue; | |
2569 | ||
48f24c4d | 2570 | rq = cpu_rq(i); |
dd41f596 | 2571 | wl = weighted_cpuload(i); |
2dd73a4f | 2572 | |
dd41f596 | 2573 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2574 | continue; |
1da177e4 | 2575 | |
dd41f596 IM |
2576 | if (wl > max_load) { |
2577 | max_load = wl; | |
48f24c4d | 2578 | busiest = rq; |
1da177e4 LT |
2579 | } |
2580 | } | |
2581 | ||
2582 | return busiest; | |
2583 | } | |
2584 | ||
77391d71 NP |
2585 | /* |
2586 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2587 | * so long as it is large enough. | |
2588 | */ | |
2589 | #define MAX_PINNED_INTERVAL 512 | |
2590 | ||
48f24c4d IM |
2591 | static inline unsigned long minus_1_or_zero(unsigned long n) |
2592 | { | |
2593 | return n > 0 ? n - 1 : 0; | |
2594 | } | |
2595 | ||
1da177e4 LT |
2596 | /* |
2597 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2598 | * tasks if there is an imbalance. | |
1da177e4 | 2599 | */ |
70b97a7f | 2600 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2601 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2602 | int *balance) |
1da177e4 | 2603 | { |
48f24c4d | 2604 | int nr_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2605 | struct sched_group *group; |
1da177e4 | 2606 | unsigned long imbalance; |
70b97a7f | 2607 | struct rq *busiest; |
0a2966b4 | 2608 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2609 | unsigned long flags; |
5969fe06 | 2610 | |
89c4710e SS |
2611 | /* |
2612 | * When power savings policy is enabled for the parent domain, idle | |
2613 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2614 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2615 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2616 | */ |
d15bcfdb | 2617 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2618 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2619 | sd_idle = 1; |
1da177e4 | 2620 | |
1da177e4 LT |
2621 | schedstat_inc(sd, lb_cnt[idle]); |
2622 | ||
0a2966b4 CL |
2623 | redo: |
2624 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2625 | &cpus, balance); |
2626 | ||
06066714 | 2627 | if (*balance == 0) |
783609c6 | 2628 | goto out_balanced; |
783609c6 | 2629 | |
1da177e4 LT |
2630 | if (!group) { |
2631 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2632 | goto out_balanced; | |
2633 | } | |
2634 | ||
0a2966b4 | 2635 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2636 | if (!busiest) { |
2637 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2638 | goto out_balanced; | |
2639 | } | |
2640 | ||
db935dbd | 2641 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2642 | |
2643 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2644 | ||
2645 | nr_moved = 0; | |
2646 | if (busiest->nr_running > 1) { | |
2647 | /* | |
2648 | * Attempt to move tasks. If find_busiest_group has found | |
2649 | * an imbalance but busiest->nr_running <= 1, the group is | |
2650 | * still unbalanced. nr_moved simply stays zero, so it is | |
2651 | * correctly treated as an imbalance. | |
2652 | */ | |
fe2eea3f | 2653 | local_irq_save(flags); |
e17224bf | 2654 | double_rq_lock(this_rq, busiest); |
1da177e4 | 2655 | nr_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d IM |
2656 | minus_1_or_zero(busiest->nr_running), |
2657 | imbalance, sd, idle, &all_pinned); | |
e17224bf | 2658 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2659 | local_irq_restore(flags); |
81026794 | 2660 | |
46cb4b7c SS |
2661 | /* |
2662 | * some other cpu did the load balance for us. | |
2663 | */ | |
2664 | if (nr_moved && this_cpu != smp_processor_id()) | |
2665 | resched_cpu(this_cpu); | |
2666 | ||
81026794 | 2667 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2668 | if (unlikely(all_pinned)) { |
2669 | cpu_clear(cpu_of(busiest), cpus); | |
2670 | if (!cpus_empty(cpus)) | |
2671 | goto redo; | |
81026794 | 2672 | goto out_balanced; |
0a2966b4 | 2673 | } |
1da177e4 | 2674 | } |
81026794 | 2675 | |
1da177e4 LT |
2676 | if (!nr_moved) { |
2677 | schedstat_inc(sd, lb_failed[idle]); | |
2678 | sd->nr_balance_failed++; | |
2679 | ||
2680 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2681 | |
fe2eea3f | 2682 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2683 | |
2684 | /* don't kick the migration_thread, if the curr | |
2685 | * task on busiest cpu can't be moved to this_cpu | |
2686 | */ | |
2687 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2688 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2689 | all_pinned = 1; |
2690 | goto out_one_pinned; | |
2691 | } | |
2692 | ||
1da177e4 LT |
2693 | if (!busiest->active_balance) { |
2694 | busiest->active_balance = 1; | |
2695 | busiest->push_cpu = this_cpu; | |
81026794 | 2696 | active_balance = 1; |
1da177e4 | 2697 | } |
fe2eea3f | 2698 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2699 | if (active_balance) |
1da177e4 LT |
2700 | wake_up_process(busiest->migration_thread); |
2701 | ||
2702 | /* | |
2703 | * We've kicked active balancing, reset the failure | |
2704 | * counter. | |
2705 | */ | |
39507451 | 2706 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2707 | } |
81026794 | 2708 | } else |
1da177e4 LT |
2709 | sd->nr_balance_failed = 0; |
2710 | ||
81026794 | 2711 | if (likely(!active_balance)) { |
1da177e4 LT |
2712 | /* We were unbalanced, so reset the balancing interval */ |
2713 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2714 | } else { |
2715 | /* | |
2716 | * If we've begun active balancing, start to back off. This | |
2717 | * case may not be covered by the all_pinned logic if there | |
2718 | * is only 1 task on the busy runqueue (because we don't call | |
2719 | * move_tasks). | |
2720 | */ | |
2721 | if (sd->balance_interval < sd->max_interval) | |
2722 | sd->balance_interval *= 2; | |
1da177e4 LT |
2723 | } |
2724 | ||
5c45bf27 | 2725 | if (!nr_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2726 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2727 | return -1; |
1da177e4 LT |
2728 | return nr_moved; |
2729 | ||
2730 | out_balanced: | |
1da177e4 LT |
2731 | schedstat_inc(sd, lb_balanced[idle]); |
2732 | ||
16cfb1c0 | 2733 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2734 | |
2735 | out_one_pinned: | |
1da177e4 | 2736 | /* tune up the balancing interval */ |
77391d71 NP |
2737 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2738 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2739 | sd->balance_interval *= 2; |
2740 | ||
48f24c4d | 2741 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2742 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2743 | return -1; |
1da177e4 LT |
2744 | return 0; |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2749 | * tasks if there is an imbalance. | |
2750 | * | |
d15bcfdb | 2751 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2752 | * this_rq is locked. |
2753 | */ | |
48f24c4d | 2754 | static int |
70b97a7f | 2755 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2756 | { |
2757 | struct sched_group *group; | |
70b97a7f | 2758 | struct rq *busiest = NULL; |
1da177e4 LT |
2759 | unsigned long imbalance; |
2760 | int nr_moved = 0; | |
5969fe06 | 2761 | int sd_idle = 0; |
969bb4e4 | 2762 | int all_pinned = 0; |
0a2966b4 | 2763 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2764 | |
89c4710e SS |
2765 | /* |
2766 | * When power savings policy is enabled for the parent domain, idle | |
2767 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2768 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2769 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2770 | */ |
2771 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2772 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2773 | sd_idle = 1; |
1da177e4 | 2774 | |
d15bcfdb | 2775 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2776 | redo: |
d15bcfdb | 2777 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2778 | &sd_idle, &cpus, NULL); |
1da177e4 | 2779 | if (!group) { |
d15bcfdb | 2780 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2781 | goto out_balanced; |
1da177e4 LT |
2782 | } |
2783 | ||
d15bcfdb | 2784 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2785 | &cpus); |
db935dbd | 2786 | if (!busiest) { |
d15bcfdb | 2787 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2788 | goto out_balanced; |
1da177e4 LT |
2789 | } |
2790 | ||
db935dbd NP |
2791 | BUG_ON(busiest == this_rq); |
2792 | ||
d15bcfdb | 2793 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf NP |
2794 | |
2795 | nr_moved = 0; | |
2796 | if (busiest->nr_running > 1) { | |
2797 | /* Attempt to move tasks */ | |
2798 | double_lock_balance(this_rq, busiest); | |
2799 | nr_moved = move_tasks(this_rq, this_cpu, busiest, | |
2dd73a4f | 2800 | minus_1_or_zero(busiest->nr_running), |
969bb4e4 SS |
2801 | imbalance, sd, CPU_NEWLY_IDLE, |
2802 | &all_pinned); | |
d6d5cfaf | 2803 | spin_unlock(&busiest->lock); |
0a2966b4 | 2804 | |
969bb4e4 | 2805 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2806 | cpu_clear(cpu_of(busiest), cpus); |
2807 | if (!cpus_empty(cpus)) | |
2808 | goto redo; | |
2809 | } | |
d6d5cfaf NP |
2810 | } |
2811 | ||
5969fe06 | 2812 | if (!nr_moved) { |
d15bcfdb | 2813 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2814 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2815 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2816 | return -1; |
2817 | } else | |
16cfb1c0 | 2818 | sd->nr_balance_failed = 0; |
1da177e4 | 2819 | |
1da177e4 | 2820 | return nr_moved; |
16cfb1c0 NP |
2821 | |
2822 | out_balanced: | |
d15bcfdb | 2823 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2824 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2825 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2826 | return -1; |
16cfb1c0 | 2827 | sd->nr_balance_failed = 0; |
48f24c4d | 2828 | |
16cfb1c0 | 2829 | return 0; |
1da177e4 LT |
2830 | } |
2831 | ||
2832 | /* | |
2833 | * idle_balance is called by schedule() if this_cpu is about to become | |
2834 | * idle. Attempts to pull tasks from other CPUs. | |
2835 | */ | |
70b97a7f | 2836 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2837 | { |
2838 | struct sched_domain *sd; | |
dd41f596 IM |
2839 | int pulled_task = -1; |
2840 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2841 | |
2842 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2843 | unsigned long interval; |
2844 | ||
2845 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2846 | continue; | |
2847 | ||
2848 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2849 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2850 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2851 | this_rq, sd); |
2852 | ||
2853 | interval = msecs_to_jiffies(sd->balance_interval); | |
2854 | if (time_after(next_balance, sd->last_balance + interval)) | |
2855 | next_balance = sd->last_balance + interval; | |
2856 | if (pulled_task) | |
2857 | break; | |
1da177e4 | 2858 | } |
dd41f596 | 2859 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2860 | /* |
2861 | * We are going idle. next_balance may be set based on | |
2862 | * a busy processor. So reset next_balance. | |
2863 | */ | |
2864 | this_rq->next_balance = next_balance; | |
dd41f596 | 2865 | } |
1da177e4 LT |
2866 | } |
2867 | ||
2868 | /* | |
2869 | * active_load_balance is run by migration threads. It pushes running tasks | |
2870 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2871 | * running on each physical CPU where possible, and avoids physical / | |
2872 | * logical imbalances. | |
2873 | * | |
2874 | * Called with busiest_rq locked. | |
2875 | */ | |
70b97a7f | 2876 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2877 | { |
39507451 | 2878 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2879 | struct sched_domain *sd; |
2880 | struct rq *target_rq; | |
39507451 | 2881 | |
48f24c4d | 2882 | /* Is there any task to move? */ |
39507451 | 2883 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2884 | return; |
2885 | ||
2886 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2887 | |
2888 | /* | |
39507451 NP |
2889 | * This condition is "impossible", if it occurs |
2890 | * we need to fix it. Originally reported by | |
2891 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2892 | */ |
39507451 | 2893 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2894 | |
39507451 NP |
2895 | /* move a task from busiest_rq to target_rq */ |
2896 | double_lock_balance(busiest_rq, target_rq); | |
2897 | ||
2898 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2899 | for_each_domain(target_cpu, sd) { |
39507451 | 2900 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2901 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2902 | break; |
c96d145e | 2903 | } |
39507451 | 2904 | |
48f24c4d IM |
2905 | if (likely(sd)) { |
2906 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2907 | |
48f24c4d | 2908 | if (move_tasks(target_rq, target_cpu, busiest_rq, 1, |
5a4f3ea7 | 2909 | ULONG_MAX, sd, CPU_IDLE, NULL)) |
48f24c4d IM |
2910 | schedstat_inc(sd, alb_pushed); |
2911 | else | |
2912 | schedstat_inc(sd, alb_failed); | |
2913 | } | |
39507451 | 2914 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2915 | } |
2916 | ||
46cb4b7c SS |
2917 | #ifdef CONFIG_NO_HZ |
2918 | static struct { | |
2919 | atomic_t load_balancer; | |
2920 | cpumask_t cpu_mask; | |
2921 | } nohz ____cacheline_aligned = { | |
2922 | .load_balancer = ATOMIC_INIT(-1), | |
2923 | .cpu_mask = CPU_MASK_NONE, | |
2924 | }; | |
2925 | ||
7835b98b | 2926 | /* |
46cb4b7c SS |
2927 | * This routine will try to nominate the ilb (idle load balancing) |
2928 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2929 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2930 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2931 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2932 | * arrives... | |
2933 | * | |
2934 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2935 | * for idle load balancing. ilb owner will still be part of | |
2936 | * nohz.cpu_mask.. | |
7835b98b | 2937 | * |
46cb4b7c SS |
2938 | * While stopping the tick, this cpu will become the ilb owner if there |
2939 | * is no other owner. And will be the owner till that cpu becomes busy | |
2940 | * or if all cpus in the system stop their ticks at which point | |
2941 | * there is no need for ilb owner. | |
2942 | * | |
2943 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2944 | * next busy scheduler_tick() | |
2945 | */ | |
2946 | int select_nohz_load_balancer(int stop_tick) | |
2947 | { | |
2948 | int cpu = smp_processor_id(); | |
2949 | ||
2950 | if (stop_tick) { | |
2951 | cpu_set(cpu, nohz.cpu_mask); | |
2952 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2953 | ||
2954 | /* | |
2955 | * If we are going offline and still the leader, give up! | |
2956 | */ | |
2957 | if (cpu_is_offline(cpu) && | |
2958 | atomic_read(&nohz.load_balancer) == cpu) { | |
2959 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2960 | BUG(); | |
2961 | return 0; | |
2962 | } | |
2963 | ||
2964 | /* time for ilb owner also to sleep */ | |
2965 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2966 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2967 | atomic_set(&nohz.load_balancer, -1); | |
2968 | return 0; | |
2969 | } | |
2970 | ||
2971 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2972 | /* make me the ilb owner */ | |
2973 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2974 | return 1; | |
2975 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2976 | return 1; | |
2977 | } else { | |
2978 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2979 | return 0; | |
2980 | ||
2981 | cpu_clear(cpu, nohz.cpu_mask); | |
2982 | ||
2983 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2984 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2985 | BUG(); | |
2986 | } | |
2987 | return 0; | |
2988 | } | |
2989 | #endif | |
2990 | ||
2991 | static DEFINE_SPINLOCK(balancing); | |
2992 | ||
2993 | /* | |
7835b98b CL |
2994 | * It checks each scheduling domain to see if it is due to be balanced, |
2995 | * and initiates a balancing operation if so. | |
2996 | * | |
2997 | * Balancing parameters are set up in arch_init_sched_domains. | |
2998 | */ | |
d15bcfdb | 2999 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3000 | { |
46cb4b7c SS |
3001 | int balance = 1; |
3002 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3003 | unsigned long interval; |
3004 | struct sched_domain *sd; | |
46cb4b7c | 3005 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3006 | unsigned long next_balance = jiffies + 60*HZ; |
1da177e4 | 3007 | |
46cb4b7c | 3008 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3009 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3010 | continue; | |
3011 | ||
3012 | interval = sd->balance_interval; | |
d15bcfdb | 3013 | if (idle != CPU_IDLE) |
1da177e4 LT |
3014 | interval *= sd->busy_factor; |
3015 | ||
3016 | /* scale ms to jiffies */ | |
3017 | interval = msecs_to_jiffies(interval); | |
3018 | if (unlikely(!interval)) | |
3019 | interval = 1; | |
dd41f596 IM |
3020 | if (interval > HZ*NR_CPUS/10) |
3021 | interval = HZ*NR_CPUS/10; | |
3022 | ||
1da177e4 | 3023 | |
08c183f3 CL |
3024 | if (sd->flags & SD_SERIALIZE) { |
3025 | if (!spin_trylock(&balancing)) | |
3026 | goto out; | |
3027 | } | |
3028 | ||
c9819f45 | 3029 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3030 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3031 | /* |
3032 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3033 | * longer idle, or one of our SMT siblings is |
3034 | * not idle. | |
3035 | */ | |
d15bcfdb | 3036 | idle = CPU_NOT_IDLE; |
1da177e4 | 3037 | } |
1bd77f2d | 3038 | sd->last_balance = jiffies; |
1da177e4 | 3039 | } |
08c183f3 CL |
3040 | if (sd->flags & SD_SERIALIZE) |
3041 | spin_unlock(&balancing); | |
3042 | out: | |
c9819f45 CL |
3043 | if (time_after(next_balance, sd->last_balance + interval)) |
3044 | next_balance = sd->last_balance + interval; | |
783609c6 SS |
3045 | |
3046 | /* | |
3047 | * Stop the load balance at this level. There is another | |
3048 | * CPU in our sched group which is doing load balancing more | |
3049 | * actively. | |
3050 | */ | |
3051 | if (!balance) | |
3052 | break; | |
1da177e4 | 3053 | } |
46cb4b7c SS |
3054 | rq->next_balance = next_balance; |
3055 | } | |
3056 | ||
3057 | /* | |
3058 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3059 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3060 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3061 | */ | |
3062 | static void run_rebalance_domains(struct softirq_action *h) | |
3063 | { | |
dd41f596 IM |
3064 | int this_cpu = smp_processor_id(); |
3065 | struct rq *this_rq = cpu_rq(this_cpu); | |
3066 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3067 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3068 | |
dd41f596 | 3069 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3070 | |
3071 | #ifdef CONFIG_NO_HZ | |
3072 | /* | |
3073 | * If this cpu is the owner for idle load balancing, then do the | |
3074 | * balancing on behalf of the other idle cpus whose ticks are | |
3075 | * stopped. | |
3076 | */ | |
dd41f596 IM |
3077 | if (this_rq->idle_at_tick && |
3078 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3079 | cpumask_t cpus = nohz.cpu_mask; |
3080 | struct rq *rq; | |
3081 | int balance_cpu; | |
3082 | ||
dd41f596 | 3083 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3084 | for_each_cpu_mask(balance_cpu, cpus) { |
3085 | /* | |
3086 | * If this cpu gets work to do, stop the load balancing | |
3087 | * work being done for other cpus. Next load | |
3088 | * balancing owner will pick it up. | |
3089 | */ | |
3090 | if (need_resched()) | |
3091 | break; | |
3092 | ||
dd41f596 | 3093 | rebalance_domains(balance_cpu, SCHED_IDLE); |
46cb4b7c SS |
3094 | |
3095 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3096 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3097 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3098 | } |
3099 | } | |
3100 | #endif | |
3101 | } | |
3102 | ||
3103 | /* | |
3104 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3105 | * | |
3106 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3107 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3108 | * if the whole system is idle. | |
3109 | */ | |
dd41f596 | 3110 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3111 | { |
46cb4b7c SS |
3112 | #ifdef CONFIG_NO_HZ |
3113 | /* | |
3114 | * If we were in the nohz mode recently and busy at the current | |
3115 | * scheduler tick, then check if we need to nominate new idle | |
3116 | * load balancer. | |
3117 | */ | |
3118 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3119 | rq->in_nohz_recently = 0; | |
3120 | ||
3121 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3122 | cpu_clear(cpu, nohz.cpu_mask); | |
3123 | atomic_set(&nohz.load_balancer, -1); | |
3124 | } | |
3125 | ||
3126 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3127 | /* | |
3128 | * simple selection for now: Nominate the | |
3129 | * first cpu in the nohz list to be the next | |
3130 | * ilb owner. | |
3131 | * | |
3132 | * TBD: Traverse the sched domains and nominate | |
3133 | * the nearest cpu in the nohz.cpu_mask. | |
3134 | */ | |
3135 | int ilb = first_cpu(nohz.cpu_mask); | |
3136 | ||
3137 | if (ilb != NR_CPUS) | |
3138 | resched_cpu(ilb); | |
3139 | } | |
3140 | } | |
3141 | ||
3142 | /* | |
3143 | * If this cpu is idle and doing idle load balancing for all the | |
3144 | * cpus with ticks stopped, is it time for that to stop? | |
3145 | */ | |
3146 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3147 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3148 | resched_cpu(cpu); | |
3149 | return; | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * If this cpu is idle and the idle load balancing is done by | |
3154 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3155 | */ | |
3156 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3157 | cpu_isset(cpu, nohz.cpu_mask)) | |
3158 | return; | |
3159 | #endif | |
3160 | if (time_after_eq(jiffies, rq->next_balance)) | |
3161 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3162 | } |
dd41f596 IM |
3163 | |
3164 | #else /* CONFIG_SMP */ | |
3165 | ||
1da177e4 LT |
3166 | /* |
3167 | * on UP we do not need to balance between CPUs: | |
3168 | */ | |
70b97a7f | 3169 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3170 | { |
3171 | } | |
dd41f596 IM |
3172 | |
3173 | /* Avoid "used but not defined" warning on UP */ | |
3174 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3175 | unsigned long max_nr_move, unsigned long max_load_move, | |
3176 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3177 | int *all_pinned, unsigned long *load_moved, | |
3178 | int this_best_prio, int best_prio, int best_prio_seen, | |
3179 | struct rq_iterator *iterator) | |
3180 | { | |
3181 | *load_moved = 0; | |
3182 | ||
3183 | return 0; | |
3184 | } | |
3185 | ||
1da177e4 LT |
3186 | #endif |
3187 | ||
1da177e4 LT |
3188 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3189 | ||
3190 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3191 | ||
3192 | /* | |
41b86e9c IM |
3193 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3194 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3195 | */ |
41b86e9c | 3196 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3197 | { |
1da177e4 | 3198 | unsigned long flags; |
41b86e9c IM |
3199 | u64 ns, delta_exec; |
3200 | struct rq *rq; | |
48f24c4d | 3201 | |
41b86e9c IM |
3202 | rq = task_rq_lock(p, &flags); |
3203 | ns = p->se.sum_exec_runtime; | |
3204 | if (rq->curr == p) { | |
3205 | delta_exec = rq_clock(rq) - p->se.exec_start; | |
3206 | if ((s64)delta_exec > 0) | |
3207 | ns += delta_exec; | |
3208 | } | |
3209 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3210 | |
1da177e4 LT |
3211 | return ns; |
3212 | } | |
3213 | ||
1da177e4 LT |
3214 | /* |
3215 | * Account user cpu time to a process. | |
3216 | * @p: the process that the cpu time gets accounted to | |
3217 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3218 | * @cputime: the cpu time spent in user space since the last update | |
3219 | */ | |
3220 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3221 | { | |
3222 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3223 | cputime64_t tmp; | |
3224 | ||
3225 | p->utime = cputime_add(p->utime, cputime); | |
3226 | ||
3227 | /* Add user time to cpustat. */ | |
3228 | tmp = cputime_to_cputime64(cputime); | |
3229 | if (TASK_NICE(p) > 0) | |
3230 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3231 | else | |
3232 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3233 | } | |
3234 | ||
3235 | /* | |
3236 | * Account system cpu time to a process. | |
3237 | * @p: the process that the cpu time gets accounted to | |
3238 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3239 | * @cputime: the cpu time spent in kernel space since the last update | |
3240 | */ | |
3241 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3242 | cputime_t cputime) | |
3243 | { | |
3244 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3245 | struct rq *rq = this_rq(); |
1da177e4 LT |
3246 | cputime64_t tmp; |
3247 | ||
3248 | p->stime = cputime_add(p->stime, cputime); | |
3249 | ||
3250 | /* Add system time to cpustat. */ | |
3251 | tmp = cputime_to_cputime64(cputime); | |
3252 | if (hardirq_count() - hardirq_offset) | |
3253 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3254 | else if (softirq_count()) | |
3255 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3256 | else if (p != rq->idle) | |
3257 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3258 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3259 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3260 | else | |
3261 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3262 | /* Account for system time used */ | |
3263 | acct_update_integrals(p); | |
1da177e4 LT |
3264 | } |
3265 | ||
3266 | /* | |
3267 | * Account for involuntary wait time. | |
3268 | * @p: the process from which the cpu time has been stolen | |
3269 | * @steal: the cpu time spent in involuntary wait | |
3270 | */ | |
3271 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3272 | { | |
3273 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3274 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3275 | struct rq *rq = this_rq(); |
1da177e4 LT |
3276 | |
3277 | if (p == rq->idle) { | |
3278 | p->stime = cputime_add(p->stime, steal); | |
3279 | if (atomic_read(&rq->nr_iowait) > 0) | |
3280 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3281 | else | |
3282 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3283 | } else | |
3284 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3285 | } | |
3286 | ||
7835b98b CL |
3287 | /* |
3288 | * This function gets called by the timer code, with HZ frequency. | |
3289 | * We call it with interrupts disabled. | |
3290 | * | |
3291 | * It also gets called by the fork code, when changing the parent's | |
3292 | * timeslices. | |
3293 | */ | |
3294 | void scheduler_tick(void) | |
3295 | { | |
7835b98b CL |
3296 | int cpu = smp_processor_id(); |
3297 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 IM |
3298 | struct task_struct *curr = rq->curr; |
3299 | ||
3300 | spin_lock(&rq->lock); | |
3301 | if (curr != rq->idle) /* FIXME: needed? */ | |
3302 | curr->sched_class->task_tick(rq, curr); | |
3303 | update_cpu_load(rq); | |
3304 | spin_unlock(&rq->lock); | |
7835b98b | 3305 | |
e418e1c2 | 3306 | #ifdef CONFIG_SMP |
dd41f596 IM |
3307 | rq->idle_at_tick = idle_cpu(cpu); |
3308 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3309 | #endif |
1da177e4 LT |
3310 | } |
3311 | ||
1da177e4 LT |
3312 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3313 | ||
3314 | void fastcall add_preempt_count(int val) | |
3315 | { | |
3316 | /* | |
3317 | * Underflow? | |
3318 | */ | |
9a11b49a IM |
3319 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3320 | return; | |
1da177e4 LT |
3321 | preempt_count() += val; |
3322 | /* | |
3323 | * Spinlock count overflowing soon? | |
3324 | */ | |
33859f7f MOS |
3325 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3326 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3327 | } |
3328 | EXPORT_SYMBOL(add_preempt_count); | |
3329 | ||
3330 | void fastcall sub_preempt_count(int val) | |
3331 | { | |
3332 | /* | |
3333 | * Underflow? | |
3334 | */ | |
9a11b49a IM |
3335 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3336 | return; | |
1da177e4 LT |
3337 | /* |
3338 | * Is the spinlock portion underflowing? | |
3339 | */ | |
9a11b49a IM |
3340 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3341 | !(preempt_count() & PREEMPT_MASK))) | |
3342 | return; | |
3343 | ||
1da177e4 LT |
3344 | preempt_count() -= val; |
3345 | } | |
3346 | EXPORT_SYMBOL(sub_preempt_count); | |
3347 | ||
3348 | #endif | |
3349 | ||
3350 | /* | |
dd41f596 | 3351 | * Print scheduling while atomic bug: |
1da177e4 | 3352 | */ |
dd41f596 | 3353 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3354 | { |
dd41f596 IM |
3355 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3356 | prev->comm, preempt_count(), prev->pid); | |
3357 | debug_show_held_locks(prev); | |
3358 | if (irqs_disabled()) | |
3359 | print_irqtrace_events(prev); | |
3360 | dump_stack(); | |
3361 | } | |
1da177e4 | 3362 | |
dd41f596 IM |
3363 | /* |
3364 | * Various schedule()-time debugging checks and statistics: | |
3365 | */ | |
3366 | static inline void schedule_debug(struct task_struct *prev) | |
3367 | { | |
1da177e4 LT |
3368 | /* |
3369 | * Test if we are atomic. Since do_exit() needs to call into | |
3370 | * schedule() atomically, we ignore that path for now. | |
3371 | * Otherwise, whine if we are scheduling when we should not be. | |
3372 | */ | |
dd41f596 IM |
3373 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3374 | __schedule_bug(prev); | |
3375 | ||
1da177e4 LT |
3376 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3377 | ||
dd41f596 IM |
3378 | schedstat_inc(this_rq(), sched_cnt); |
3379 | } | |
3380 | ||
3381 | /* | |
3382 | * Pick up the highest-prio task: | |
3383 | */ | |
3384 | static inline struct task_struct * | |
3385 | pick_next_task(struct rq *rq, struct task_struct *prev, u64 now) | |
3386 | { | |
3387 | struct sched_class *class; | |
3388 | struct task_struct *p; | |
1da177e4 LT |
3389 | |
3390 | /* | |
dd41f596 IM |
3391 | * Optimization: we know that if all tasks are in |
3392 | * the fair class we can call that function directly: | |
1da177e4 | 3393 | */ |
dd41f596 IM |
3394 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
3395 | p = fair_sched_class.pick_next_task(rq, now); | |
3396 | if (likely(p)) | |
3397 | return p; | |
1da177e4 LT |
3398 | } |
3399 | ||
dd41f596 IM |
3400 | class = sched_class_highest; |
3401 | for ( ; ; ) { | |
3402 | p = class->pick_next_task(rq, now); | |
3403 | if (p) | |
3404 | return p; | |
3405 | /* | |
3406 | * Will never be NULL as the idle class always | |
3407 | * returns a non-NULL p: | |
3408 | */ | |
3409 | class = class->next; | |
3410 | } | |
3411 | } | |
1da177e4 | 3412 | |
dd41f596 IM |
3413 | /* |
3414 | * schedule() is the main scheduler function. | |
3415 | */ | |
3416 | asmlinkage void __sched schedule(void) | |
3417 | { | |
3418 | struct task_struct *prev, *next; | |
3419 | long *switch_count; | |
3420 | struct rq *rq; | |
3421 | u64 now; | |
3422 | int cpu; | |
3423 | ||
3424 | need_resched: | |
3425 | preempt_disable(); | |
3426 | cpu = smp_processor_id(); | |
3427 | rq = cpu_rq(cpu); | |
3428 | rcu_qsctr_inc(cpu); | |
3429 | prev = rq->curr; | |
3430 | switch_count = &prev->nivcsw; | |
3431 | ||
3432 | release_kernel_lock(prev); | |
3433 | need_resched_nonpreemptible: | |
3434 | ||
3435 | schedule_debug(prev); | |
1da177e4 LT |
3436 | |
3437 | spin_lock_irq(&rq->lock); | |
dd41f596 | 3438 | clear_tsk_need_resched(prev); |
1da177e4 | 3439 | |
1da177e4 | 3440 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3441 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3442 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3443 | prev->state = TASK_RUNNING; |
dd41f596 IM |
3444 | } else { |
3445 | deactivate_task(rq, prev, 1); | |
1da177e4 | 3446 | } |
dd41f596 | 3447 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3448 | } |
3449 | ||
dd41f596 | 3450 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3451 | idle_balance(cpu, rq); |
1da177e4 | 3452 | |
dd41f596 IM |
3453 | now = __rq_clock(rq); |
3454 | prev->sched_class->put_prev_task(rq, prev, now); | |
3455 | next = pick_next_task(rq, prev, now); | |
1da177e4 LT |
3456 | |
3457 | sched_info_switch(prev, next); | |
dd41f596 | 3458 | |
1da177e4 | 3459 | if (likely(prev != next)) { |
1da177e4 LT |
3460 | rq->nr_switches++; |
3461 | rq->curr = next; | |
3462 | ++*switch_count; | |
3463 | ||
dd41f596 | 3464 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3465 | } else |
3466 | spin_unlock_irq(&rq->lock); | |
3467 | ||
dd41f596 IM |
3468 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3469 | cpu = smp_processor_id(); | |
3470 | rq = cpu_rq(cpu); | |
1da177e4 | 3471 | goto need_resched_nonpreemptible; |
dd41f596 | 3472 | } |
1da177e4 LT |
3473 | preempt_enable_no_resched(); |
3474 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3475 | goto need_resched; | |
3476 | } | |
1da177e4 LT |
3477 | EXPORT_SYMBOL(schedule); |
3478 | ||
3479 | #ifdef CONFIG_PREEMPT | |
3480 | /* | |
2ed6e34f | 3481 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3482 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3483 | * occur there and call schedule directly. | |
3484 | */ | |
3485 | asmlinkage void __sched preempt_schedule(void) | |
3486 | { | |
3487 | struct thread_info *ti = current_thread_info(); | |
3488 | #ifdef CONFIG_PREEMPT_BKL | |
3489 | struct task_struct *task = current; | |
3490 | int saved_lock_depth; | |
3491 | #endif | |
3492 | /* | |
3493 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3494 | * we do not want to preempt the current task. Just return.. | |
3495 | */ | |
beed33a8 | 3496 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3497 | return; |
3498 | ||
3499 | need_resched: | |
3500 | add_preempt_count(PREEMPT_ACTIVE); | |
3501 | /* | |
3502 | * We keep the big kernel semaphore locked, but we | |
3503 | * clear ->lock_depth so that schedule() doesnt | |
3504 | * auto-release the semaphore: | |
3505 | */ | |
3506 | #ifdef CONFIG_PREEMPT_BKL | |
3507 | saved_lock_depth = task->lock_depth; | |
3508 | task->lock_depth = -1; | |
3509 | #endif | |
3510 | schedule(); | |
3511 | #ifdef CONFIG_PREEMPT_BKL | |
3512 | task->lock_depth = saved_lock_depth; | |
3513 | #endif | |
3514 | sub_preempt_count(PREEMPT_ACTIVE); | |
3515 | ||
3516 | /* we could miss a preemption opportunity between schedule and now */ | |
3517 | barrier(); | |
3518 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3519 | goto need_resched; | |
3520 | } | |
1da177e4 LT |
3521 | EXPORT_SYMBOL(preempt_schedule); |
3522 | ||
3523 | /* | |
2ed6e34f | 3524 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3525 | * off of irq context. |
3526 | * Note, that this is called and return with irqs disabled. This will | |
3527 | * protect us against recursive calling from irq. | |
3528 | */ | |
3529 | asmlinkage void __sched preempt_schedule_irq(void) | |
3530 | { | |
3531 | struct thread_info *ti = current_thread_info(); | |
3532 | #ifdef CONFIG_PREEMPT_BKL | |
3533 | struct task_struct *task = current; | |
3534 | int saved_lock_depth; | |
3535 | #endif | |
2ed6e34f | 3536 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3537 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3538 | ||
3539 | need_resched: | |
3540 | add_preempt_count(PREEMPT_ACTIVE); | |
3541 | /* | |
3542 | * We keep the big kernel semaphore locked, but we | |
3543 | * clear ->lock_depth so that schedule() doesnt | |
3544 | * auto-release the semaphore: | |
3545 | */ | |
3546 | #ifdef CONFIG_PREEMPT_BKL | |
3547 | saved_lock_depth = task->lock_depth; | |
3548 | task->lock_depth = -1; | |
3549 | #endif | |
3550 | local_irq_enable(); | |
3551 | schedule(); | |
3552 | local_irq_disable(); | |
3553 | #ifdef CONFIG_PREEMPT_BKL | |
3554 | task->lock_depth = saved_lock_depth; | |
3555 | #endif | |
3556 | sub_preempt_count(PREEMPT_ACTIVE); | |
3557 | ||
3558 | /* we could miss a preemption opportunity between schedule and now */ | |
3559 | barrier(); | |
3560 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3561 | goto need_resched; | |
3562 | } | |
3563 | ||
3564 | #endif /* CONFIG_PREEMPT */ | |
3565 | ||
95cdf3b7 IM |
3566 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3567 | void *key) | |
1da177e4 | 3568 | { |
48f24c4d | 3569 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3570 | } |
1da177e4 LT |
3571 | EXPORT_SYMBOL(default_wake_function); |
3572 | ||
3573 | /* | |
3574 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3575 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3576 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3577 | * | |
3578 | * There are circumstances in which we can try to wake a task which has already | |
3579 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3580 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3581 | */ | |
3582 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3583 | int nr_exclusive, int sync, void *key) | |
3584 | { | |
3585 | struct list_head *tmp, *next; | |
3586 | ||
3587 | list_for_each_safe(tmp, next, &q->task_list) { | |
48f24c4d IM |
3588 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
3589 | unsigned flags = curr->flags; | |
3590 | ||
1da177e4 | 3591 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3592 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3593 | break; |
3594 | } | |
3595 | } | |
3596 | ||
3597 | /** | |
3598 | * __wake_up - wake up threads blocked on a waitqueue. | |
3599 | * @q: the waitqueue | |
3600 | * @mode: which threads | |
3601 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3602 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3603 | */ |
3604 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3605 | int nr_exclusive, void *key) |
1da177e4 LT |
3606 | { |
3607 | unsigned long flags; | |
3608 | ||
3609 | spin_lock_irqsave(&q->lock, flags); | |
3610 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3611 | spin_unlock_irqrestore(&q->lock, flags); | |
3612 | } | |
1da177e4 LT |
3613 | EXPORT_SYMBOL(__wake_up); |
3614 | ||
3615 | /* | |
3616 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3617 | */ | |
3618 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3619 | { | |
3620 | __wake_up_common(q, mode, 1, 0, NULL); | |
3621 | } | |
3622 | ||
3623 | /** | |
67be2dd1 | 3624 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3625 | * @q: the waitqueue |
3626 | * @mode: which threads | |
3627 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3628 | * | |
3629 | * The sync wakeup differs that the waker knows that it will schedule | |
3630 | * away soon, so while the target thread will be woken up, it will not | |
3631 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3632 | * with each other. This can prevent needless bouncing between CPUs. | |
3633 | * | |
3634 | * On UP it can prevent extra preemption. | |
3635 | */ | |
95cdf3b7 IM |
3636 | void fastcall |
3637 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3638 | { |
3639 | unsigned long flags; | |
3640 | int sync = 1; | |
3641 | ||
3642 | if (unlikely(!q)) | |
3643 | return; | |
3644 | ||
3645 | if (unlikely(!nr_exclusive)) | |
3646 | sync = 0; | |
3647 | ||
3648 | spin_lock_irqsave(&q->lock, flags); | |
3649 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3650 | spin_unlock_irqrestore(&q->lock, flags); | |
3651 | } | |
3652 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3653 | ||
3654 | void fastcall complete(struct completion *x) | |
3655 | { | |
3656 | unsigned long flags; | |
3657 | ||
3658 | spin_lock_irqsave(&x->wait.lock, flags); | |
3659 | x->done++; | |
3660 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3661 | 1, 0, NULL); | |
3662 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3663 | } | |
3664 | EXPORT_SYMBOL(complete); | |
3665 | ||
3666 | void fastcall complete_all(struct completion *x) | |
3667 | { | |
3668 | unsigned long flags; | |
3669 | ||
3670 | spin_lock_irqsave(&x->wait.lock, flags); | |
3671 | x->done += UINT_MAX/2; | |
3672 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3673 | 0, 0, NULL); | |
3674 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3675 | } | |
3676 | EXPORT_SYMBOL(complete_all); | |
3677 | ||
3678 | void fastcall __sched wait_for_completion(struct completion *x) | |
3679 | { | |
3680 | might_sleep(); | |
48f24c4d | 3681 | |
1da177e4 LT |
3682 | spin_lock_irq(&x->wait.lock); |
3683 | if (!x->done) { | |
3684 | DECLARE_WAITQUEUE(wait, current); | |
3685 | ||
3686 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3687 | __add_wait_queue_tail(&x->wait, &wait); | |
3688 | do { | |
3689 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3690 | spin_unlock_irq(&x->wait.lock); | |
3691 | schedule(); | |
3692 | spin_lock_irq(&x->wait.lock); | |
3693 | } while (!x->done); | |
3694 | __remove_wait_queue(&x->wait, &wait); | |
3695 | } | |
3696 | x->done--; | |
3697 | spin_unlock_irq(&x->wait.lock); | |
3698 | } | |
3699 | EXPORT_SYMBOL(wait_for_completion); | |
3700 | ||
3701 | unsigned long fastcall __sched | |
3702 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3703 | { | |
3704 | might_sleep(); | |
3705 | ||
3706 | spin_lock_irq(&x->wait.lock); | |
3707 | if (!x->done) { | |
3708 | DECLARE_WAITQUEUE(wait, current); | |
3709 | ||
3710 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3711 | __add_wait_queue_tail(&x->wait, &wait); | |
3712 | do { | |
3713 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3714 | spin_unlock_irq(&x->wait.lock); | |
3715 | timeout = schedule_timeout(timeout); | |
3716 | spin_lock_irq(&x->wait.lock); | |
3717 | if (!timeout) { | |
3718 | __remove_wait_queue(&x->wait, &wait); | |
3719 | goto out; | |
3720 | } | |
3721 | } while (!x->done); | |
3722 | __remove_wait_queue(&x->wait, &wait); | |
3723 | } | |
3724 | x->done--; | |
3725 | out: | |
3726 | spin_unlock_irq(&x->wait.lock); | |
3727 | return timeout; | |
3728 | } | |
3729 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3730 | ||
3731 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3732 | { | |
3733 | int ret = 0; | |
3734 | ||
3735 | might_sleep(); | |
3736 | ||
3737 | spin_lock_irq(&x->wait.lock); | |
3738 | if (!x->done) { | |
3739 | DECLARE_WAITQUEUE(wait, current); | |
3740 | ||
3741 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3742 | __add_wait_queue_tail(&x->wait, &wait); | |
3743 | do { | |
3744 | if (signal_pending(current)) { | |
3745 | ret = -ERESTARTSYS; | |
3746 | __remove_wait_queue(&x->wait, &wait); | |
3747 | goto out; | |
3748 | } | |
3749 | __set_current_state(TASK_INTERRUPTIBLE); | |
3750 | spin_unlock_irq(&x->wait.lock); | |
3751 | schedule(); | |
3752 | spin_lock_irq(&x->wait.lock); | |
3753 | } while (!x->done); | |
3754 | __remove_wait_queue(&x->wait, &wait); | |
3755 | } | |
3756 | x->done--; | |
3757 | out: | |
3758 | spin_unlock_irq(&x->wait.lock); | |
3759 | ||
3760 | return ret; | |
3761 | } | |
3762 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3763 | ||
3764 | unsigned long fastcall __sched | |
3765 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3766 | unsigned long timeout) | |
3767 | { | |
3768 | might_sleep(); | |
3769 | ||
3770 | spin_lock_irq(&x->wait.lock); | |
3771 | if (!x->done) { | |
3772 | DECLARE_WAITQUEUE(wait, current); | |
3773 | ||
3774 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3775 | __add_wait_queue_tail(&x->wait, &wait); | |
3776 | do { | |
3777 | if (signal_pending(current)) { | |
3778 | timeout = -ERESTARTSYS; | |
3779 | __remove_wait_queue(&x->wait, &wait); | |
3780 | goto out; | |
3781 | } | |
3782 | __set_current_state(TASK_INTERRUPTIBLE); | |
3783 | spin_unlock_irq(&x->wait.lock); | |
3784 | timeout = schedule_timeout(timeout); | |
3785 | spin_lock_irq(&x->wait.lock); | |
3786 | if (!timeout) { | |
3787 | __remove_wait_queue(&x->wait, &wait); | |
3788 | goto out; | |
3789 | } | |
3790 | } while (!x->done); | |
3791 | __remove_wait_queue(&x->wait, &wait); | |
3792 | } | |
3793 | x->done--; | |
3794 | out: | |
3795 | spin_unlock_irq(&x->wait.lock); | |
3796 | return timeout; | |
3797 | } | |
3798 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3799 | ||
0fec171c IM |
3800 | static inline void |
3801 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3802 | { | |
3803 | spin_lock_irqsave(&q->lock, *flags); | |
3804 | __add_wait_queue(q, wait); | |
1da177e4 | 3805 | spin_unlock(&q->lock); |
0fec171c | 3806 | } |
1da177e4 | 3807 | |
0fec171c IM |
3808 | static inline void |
3809 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3810 | { | |
3811 | spin_lock_irq(&q->lock); | |
3812 | __remove_wait_queue(q, wait); | |
3813 | spin_unlock_irqrestore(&q->lock, *flags); | |
3814 | } | |
1da177e4 | 3815 | |
0fec171c | 3816 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3817 | { |
0fec171c IM |
3818 | unsigned long flags; |
3819 | wait_queue_t wait; | |
3820 | ||
3821 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3822 | |
3823 | current->state = TASK_INTERRUPTIBLE; | |
3824 | ||
0fec171c | 3825 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3826 | schedule(); |
0fec171c | 3827 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3828 | } |
1da177e4 LT |
3829 | EXPORT_SYMBOL(interruptible_sleep_on); |
3830 | ||
0fec171c | 3831 | long __sched |
95cdf3b7 | 3832 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3833 | { |
0fec171c IM |
3834 | unsigned long flags; |
3835 | wait_queue_t wait; | |
3836 | ||
3837 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3838 | |
3839 | current->state = TASK_INTERRUPTIBLE; | |
3840 | ||
0fec171c | 3841 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3842 | timeout = schedule_timeout(timeout); |
0fec171c | 3843 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3844 | |
3845 | return timeout; | |
3846 | } | |
1da177e4 LT |
3847 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3848 | ||
0fec171c | 3849 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3850 | { |
0fec171c IM |
3851 | unsigned long flags; |
3852 | wait_queue_t wait; | |
3853 | ||
3854 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3855 | |
3856 | current->state = TASK_UNINTERRUPTIBLE; | |
3857 | ||
0fec171c | 3858 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3859 | schedule(); |
0fec171c | 3860 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3861 | } |
1da177e4 LT |
3862 | EXPORT_SYMBOL(sleep_on); |
3863 | ||
0fec171c | 3864 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3865 | { |
0fec171c IM |
3866 | unsigned long flags; |
3867 | wait_queue_t wait; | |
3868 | ||
3869 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3870 | |
3871 | current->state = TASK_UNINTERRUPTIBLE; | |
3872 | ||
0fec171c | 3873 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3874 | timeout = schedule_timeout(timeout); |
0fec171c | 3875 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3876 | |
3877 | return timeout; | |
3878 | } | |
1da177e4 LT |
3879 | EXPORT_SYMBOL(sleep_on_timeout); |
3880 | ||
b29739f9 IM |
3881 | #ifdef CONFIG_RT_MUTEXES |
3882 | ||
3883 | /* | |
3884 | * rt_mutex_setprio - set the current priority of a task | |
3885 | * @p: task | |
3886 | * @prio: prio value (kernel-internal form) | |
3887 | * | |
3888 | * This function changes the 'effective' priority of a task. It does | |
3889 | * not touch ->normal_prio like __setscheduler(). | |
3890 | * | |
3891 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3892 | */ | |
36c8b586 | 3893 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3894 | { |
3895 | unsigned long flags; | |
dd41f596 | 3896 | int oldprio, on_rq; |
70b97a7f | 3897 | struct rq *rq; |
dd41f596 | 3898 | u64 now; |
b29739f9 IM |
3899 | |
3900 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3901 | ||
3902 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3903 | now = rq_clock(rq); |
b29739f9 | 3904 | |
d5f9f942 | 3905 | oldprio = p->prio; |
dd41f596 IM |
3906 | on_rq = p->se.on_rq; |
3907 | if (on_rq) | |
3908 | dequeue_task(rq, p, 0, now); | |
3909 | ||
3910 | if (rt_prio(prio)) | |
3911 | p->sched_class = &rt_sched_class; | |
3912 | else | |
3913 | p->sched_class = &fair_sched_class; | |
3914 | ||
b29739f9 IM |
3915 | p->prio = prio; |
3916 | ||
dd41f596 IM |
3917 | if (on_rq) { |
3918 | enqueue_task(rq, p, 0, now); | |
b29739f9 IM |
3919 | /* |
3920 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3921 | * our priority decreased, or if we are not currently running on |
3922 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3923 | */ |
d5f9f942 AM |
3924 | if (task_running(rq, p)) { |
3925 | if (p->prio > oldprio) | |
3926 | resched_task(rq->curr); | |
dd41f596 IM |
3927 | } else { |
3928 | check_preempt_curr(rq, p); | |
3929 | } | |
b29739f9 IM |
3930 | } |
3931 | task_rq_unlock(rq, &flags); | |
3932 | } | |
3933 | ||
3934 | #endif | |
3935 | ||
36c8b586 | 3936 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3937 | { |
dd41f596 | 3938 | int old_prio, delta, on_rq; |
1da177e4 | 3939 | unsigned long flags; |
70b97a7f | 3940 | struct rq *rq; |
dd41f596 | 3941 | u64 now; |
1da177e4 LT |
3942 | |
3943 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3944 | return; | |
3945 | /* | |
3946 | * We have to be careful, if called from sys_setpriority(), | |
3947 | * the task might be in the middle of scheduling on another CPU. | |
3948 | */ | |
3949 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 3950 | now = rq_clock(rq); |
1da177e4 LT |
3951 | /* |
3952 | * The RT priorities are set via sched_setscheduler(), but we still | |
3953 | * allow the 'normal' nice value to be set - but as expected | |
3954 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3955 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3956 | */ |
e05606d3 | 3957 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3958 | p->static_prio = NICE_TO_PRIO(nice); |
3959 | goto out_unlock; | |
3960 | } | |
dd41f596 IM |
3961 | on_rq = p->se.on_rq; |
3962 | if (on_rq) { | |
3963 | dequeue_task(rq, p, 0, now); | |
3964 | dec_load(rq, p, now); | |
2dd73a4f | 3965 | } |
1da177e4 | 3966 | |
1da177e4 | 3967 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3968 | set_load_weight(p); |
b29739f9 IM |
3969 | old_prio = p->prio; |
3970 | p->prio = effective_prio(p); | |
3971 | delta = p->prio - old_prio; | |
1da177e4 | 3972 | |
dd41f596 IM |
3973 | if (on_rq) { |
3974 | enqueue_task(rq, p, 0, now); | |
3975 | inc_load(rq, p, now); | |
1da177e4 | 3976 | /* |
d5f9f942 AM |
3977 | * If the task increased its priority or is running and |
3978 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3979 | */ |
d5f9f942 | 3980 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3981 | resched_task(rq->curr); |
3982 | } | |
3983 | out_unlock: | |
3984 | task_rq_unlock(rq, &flags); | |
3985 | } | |
1da177e4 LT |
3986 | EXPORT_SYMBOL(set_user_nice); |
3987 | ||
e43379f1 MM |
3988 | /* |
3989 | * can_nice - check if a task can reduce its nice value | |
3990 | * @p: task | |
3991 | * @nice: nice value | |
3992 | */ | |
36c8b586 | 3993 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3994 | { |
024f4747 MM |
3995 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3996 | int nice_rlim = 20 - nice; | |
48f24c4d | 3997 | |
e43379f1 MM |
3998 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
3999 | capable(CAP_SYS_NICE)); | |
4000 | } | |
4001 | ||
1da177e4 LT |
4002 | #ifdef __ARCH_WANT_SYS_NICE |
4003 | ||
4004 | /* | |
4005 | * sys_nice - change the priority of the current process. | |
4006 | * @increment: priority increment | |
4007 | * | |
4008 | * sys_setpriority is a more generic, but much slower function that | |
4009 | * does similar things. | |
4010 | */ | |
4011 | asmlinkage long sys_nice(int increment) | |
4012 | { | |
48f24c4d | 4013 | long nice, retval; |
1da177e4 LT |
4014 | |
4015 | /* | |
4016 | * Setpriority might change our priority at the same moment. | |
4017 | * We don't have to worry. Conceptually one call occurs first | |
4018 | * and we have a single winner. | |
4019 | */ | |
e43379f1 MM |
4020 | if (increment < -40) |
4021 | increment = -40; | |
1da177e4 LT |
4022 | if (increment > 40) |
4023 | increment = 40; | |
4024 | ||
4025 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4026 | if (nice < -20) | |
4027 | nice = -20; | |
4028 | if (nice > 19) | |
4029 | nice = 19; | |
4030 | ||
e43379f1 MM |
4031 | if (increment < 0 && !can_nice(current, nice)) |
4032 | return -EPERM; | |
4033 | ||
1da177e4 LT |
4034 | retval = security_task_setnice(current, nice); |
4035 | if (retval) | |
4036 | return retval; | |
4037 | ||
4038 | set_user_nice(current, nice); | |
4039 | return 0; | |
4040 | } | |
4041 | ||
4042 | #endif | |
4043 | ||
4044 | /** | |
4045 | * task_prio - return the priority value of a given task. | |
4046 | * @p: the task in question. | |
4047 | * | |
4048 | * This is the priority value as seen by users in /proc. | |
4049 | * RT tasks are offset by -200. Normal tasks are centered | |
4050 | * around 0, value goes from -16 to +15. | |
4051 | */ | |
36c8b586 | 4052 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4053 | { |
4054 | return p->prio - MAX_RT_PRIO; | |
4055 | } | |
4056 | ||
4057 | /** | |
4058 | * task_nice - return the nice value of a given task. | |
4059 | * @p: the task in question. | |
4060 | */ | |
36c8b586 | 4061 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4062 | { |
4063 | return TASK_NICE(p); | |
4064 | } | |
1da177e4 | 4065 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4066 | |
4067 | /** | |
4068 | * idle_cpu - is a given cpu idle currently? | |
4069 | * @cpu: the processor in question. | |
4070 | */ | |
4071 | int idle_cpu(int cpu) | |
4072 | { | |
4073 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4074 | } | |
4075 | ||
1da177e4 LT |
4076 | /** |
4077 | * idle_task - return the idle task for a given cpu. | |
4078 | * @cpu: the processor in question. | |
4079 | */ | |
36c8b586 | 4080 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4081 | { |
4082 | return cpu_rq(cpu)->idle; | |
4083 | } | |
4084 | ||
4085 | /** | |
4086 | * find_process_by_pid - find a process with a matching PID value. | |
4087 | * @pid: the pid in question. | |
4088 | */ | |
36c8b586 | 4089 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4090 | { |
4091 | return pid ? find_task_by_pid(pid) : current; | |
4092 | } | |
4093 | ||
4094 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4095 | static void |
4096 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4097 | { |
dd41f596 | 4098 | BUG_ON(p->se.on_rq); |
48f24c4d | 4099 | |
1da177e4 | 4100 | p->policy = policy; |
dd41f596 IM |
4101 | switch (p->policy) { |
4102 | case SCHED_NORMAL: | |
4103 | case SCHED_BATCH: | |
4104 | case SCHED_IDLE: | |
4105 | p->sched_class = &fair_sched_class; | |
4106 | break; | |
4107 | case SCHED_FIFO: | |
4108 | case SCHED_RR: | |
4109 | p->sched_class = &rt_sched_class; | |
4110 | break; | |
4111 | } | |
4112 | ||
1da177e4 | 4113 | p->rt_priority = prio; |
b29739f9 IM |
4114 | p->normal_prio = normal_prio(p); |
4115 | /* we are holding p->pi_lock already */ | |
4116 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4117 | set_load_weight(p); |
1da177e4 LT |
4118 | } |
4119 | ||
4120 | /** | |
72fd4a35 | 4121 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4122 | * @p: the task in question. |
4123 | * @policy: new policy. | |
4124 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4125 | * |
72fd4a35 | 4126 | * NOTE that the task may be already dead. |
1da177e4 | 4127 | */ |
95cdf3b7 IM |
4128 | int sched_setscheduler(struct task_struct *p, int policy, |
4129 | struct sched_param *param) | |
1da177e4 | 4130 | { |
dd41f596 | 4131 | int retval, oldprio, oldpolicy = -1, on_rq; |
1da177e4 | 4132 | unsigned long flags; |
70b97a7f | 4133 | struct rq *rq; |
1da177e4 | 4134 | |
66e5393a SR |
4135 | /* may grab non-irq protected spin_locks */ |
4136 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4137 | recheck: |
4138 | /* double check policy once rq lock held */ | |
4139 | if (policy < 0) | |
4140 | policy = oldpolicy = p->policy; | |
4141 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4142 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4143 | policy != SCHED_IDLE) | |
b0a9499c | 4144 | return -EINVAL; |
1da177e4 LT |
4145 | /* |
4146 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4147 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4148 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4149 | */ |
4150 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4151 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4152 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4153 | return -EINVAL; |
e05606d3 | 4154 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4155 | return -EINVAL; |
4156 | ||
37e4ab3f OC |
4157 | /* |
4158 | * Allow unprivileged RT tasks to decrease priority: | |
4159 | */ | |
4160 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4161 | if (rt_policy(policy)) { |
8dc3e909 | 4162 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4163 | |
4164 | if (!lock_task_sighand(p, &flags)) | |
4165 | return -ESRCH; | |
4166 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4167 | unlock_task_sighand(p, &flags); | |
4168 | ||
4169 | /* can't set/change the rt policy */ | |
4170 | if (policy != p->policy && !rlim_rtprio) | |
4171 | return -EPERM; | |
4172 | ||
4173 | /* can't increase priority */ | |
4174 | if (param->sched_priority > p->rt_priority && | |
4175 | param->sched_priority > rlim_rtprio) | |
4176 | return -EPERM; | |
4177 | } | |
dd41f596 IM |
4178 | /* |
4179 | * Like positive nice levels, dont allow tasks to | |
4180 | * move out of SCHED_IDLE either: | |
4181 | */ | |
4182 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4183 | return -EPERM; | |
5fe1d75f | 4184 | |
37e4ab3f OC |
4185 | /* can't change other user's priorities */ |
4186 | if ((current->euid != p->euid) && | |
4187 | (current->euid != p->uid)) | |
4188 | return -EPERM; | |
4189 | } | |
1da177e4 LT |
4190 | |
4191 | retval = security_task_setscheduler(p, policy, param); | |
4192 | if (retval) | |
4193 | return retval; | |
b29739f9 IM |
4194 | /* |
4195 | * make sure no PI-waiters arrive (or leave) while we are | |
4196 | * changing the priority of the task: | |
4197 | */ | |
4198 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4199 | /* |
4200 | * To be able to change p->policy safely, the apropriate | |
4201 | * runqueue lock must be held. | |
4202 | */ | |
b29739f9 | 4203 | rq = __task_rq_lock(p); |
1da177e4 LT |
4204 | /* recheck policy now with rq lock held */ |
4205 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4206 | policy = oldpolicy = -1; | |
b29739f9 IM |
4207 | __task_rq_unlock(rq); |
4208 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4209 | goto recheck; |
4210 | } | |
dd41f596 IM |
4211 | on_rq = p->se.on_rq; |
4212 | if (on_rq) | |
4213 | deactivate_task(rq, p, 0); | |
1da177e4 | 4214 | oldprio = p->prio; |
dd41f596 IM |
4215 | __setscheduler(rq, p, policy, param->sched_priority); |
4216 | if (on_rq) { | |
4217 | activate_task(rq, p, 0); | |
1da177e4 LT |
4218 | /* |
4219 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4220 | * our priority decreased, or if we are not currently running on |
4221 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4222 | */ |
d5f9f942 AM |
4223 | if (task_running(rq, p)) { |
4224 | if (p->prio > oldprio) | |
4225 | resched_task(rq->curr); | |
dd41f596 IM |
4226 | } else { |
4227 | check_preempt_curr(rq, p); | |
4228 | } | |
1da177e4 | 4229 | } |
b29739f9 IM |
4230 | __task_rq_unlock(rq); |
4231 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4232 | ||
95e02ca9 TG |
4233 | rt_mutex_adjust_pi(p); |
4234 | ||
1da177e4 LT |
4235 | return 0; |
4236 | } | |
4237 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4238 | ||
95cdf3b7 IM |
4239 | static int |
4240 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4241 | { |
1da177e4 LT |
4242 | struct sched_param lparam; |
4243 | struct task_struct *p; | |
36c8b586 | 4244 | int retval; |
1da177e4 LT |
4245 | |
4246 | if (!param || pid < 0) | |
4247 | return -EINVAL; | |
4248 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4249 | return -EFAULT; | |
5fe1d75f ON |
4250 | |
4251 | rcu_read_lock(); | |
4252 | retval = -ESRCH; | |
1da177e4 | 4253 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4254 | if (p != NULL) |
4255 | retval = sched_setscheduler(p, policy, &lparam); | |
4256 | rcu_read_unlock(); | |
36c8b586 | 4257 | |
1da177e4 LT |
4258 | return retval; |
4259 | } | |
4260 | ||
4261 | /** | |
4262 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4263 | * @pid: the pid in question. | |
4264 | * @policy: new policy. | |
4265 | * @param: structure containing the new RT priority. | |
4266 | */ | |
4267 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4268 | struct sched_param __user *param) | |
4269 | { | |
c21761f1 JB |
4270 | /* negative values for policy are not valid */ |
4271 | if (policy < 0) | |
4272 | return -EINVAL; | |
4273 | ||
1da177e4 LT |
4274 | return do_sched_setscheduler(pid, policy, param); |
4275 | } | |
4276 | ||
4277 | /** | |
4278 | * sys_sched_setparam - set/change the RT priority of a thread | |
4279 | * @pid: the pid in question. | |
4280 | * @param: structure containing the new RT priority. | |
4281 | */ | |
4282 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4283 | { | |
4284 | return do_sched_setscheduler(pid, -1, param); | |
4285 | } | |
4286 | ||
4287 | /** | |
4288 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4289 | * @pid: the pid in question. | |
4290 | */ | |
4291 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4292 | { | |
36c8b586 | 4293 | struct task_struct *p; |
1da177e4 | 4294 | int retval = -EINVAL; |
1da177e4 LT |
4295 | |
4296 | if (pid < 0) | |
4297 | goto out_nounlock; | |
4298 | ||
4299 | retval = -ESRCH; | |
4300 | read_lock(&tasklist_lock); | |
4301 | p = find_process_by_pid(pid); | |
4302 | if (p) { | |
4303 | retval = security_task_getscheduler(p); | |
4304 | if (!retval) | |
4305 | retval = p->policy; | |
4306 | } | |
4307 | read_unlock(&tasklist_lock); | |
4308 | ||
4309 | out_nounlock: | |
4310 | return retval; | |
4311 | } | |
4312 | ||
4313 | /** | |
4314 | * sys_sched_getscheduler - get the RT priority of a thread | |
4315 | * @pid: the pid in question. | |
4316 | * @param: structure containing the RT priority. | |
4317 | */ | |
4318 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4319 | { | |
4320 | struct sched_param lp; | |
36c8b586 | 4321 | struct task_struct *p; |
1da177e4 | 4322 | int retval = -EINVAL; |
1da177e4 LT |
4323 | |
4324 | if (!param || pid < 0) | |
4325 | goto out_nounlock; | |
4326 | ||
4327 | read_lock(&tasklist_lock); | |
4328 | p = find_process_by_pid(pid); | |
4329 | retval = -ESRCH; | |
4330 | if (!p) | |
4331 | goto out_unlock; | |
4332 | ||
4333 | retval = security_task_getscheduler(p); | |
4334 | if (retval) | |
4335 | goto out_unlock; | |
4336 | ||
4337 | lp.sched_priority = p->rt_priority; | |
4338 | read_unlock(&tasklist_lock); | |
4339 | ||
4340 | /* | |
4341 | * This one might sleep, we cannot do it with a spinlock held ... | |
4342 | */ | |
4343 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4344 | ||
4345 | out_nounlock: | |
4346 | return retval; | |
4347 | ||
4348 | out_unlock: | |
4349 | read_unlock(&tasklist_lock); | |
4350 | return retval; | |
4351 | } | |
4352 | ||
4353 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4354 | { | |
1da177e4 | 4355 | cpumask_t cpus_allowed; |
36c8b586 IM |
4356 | struct task_struct *p; |
4357 | int retval; | |
1da177e4 | 4358 | |
5be9361c | 4359 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4360 | read_lock(&tasklist_lock); |
4361 | ||
4362 | p = find_process_by_pid(pid); | |
4363 | if (!p) { | |
4364 | read_unlock(&tasklist_lock); | |
5be9361c | 4365 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4366 | return -ESRCH; |
4367 | } | |
4368 | ||
4369 | /* | |
4370 | * It is not safe to call set_cpus_allowed with the | |
4371 | * tasklist_lock held. We will bump the task_struct's | |
4372 | * usage count and then drop tasklist_lock. | |
4373 | */ | |
4374 | get_task_struct(p); | |
4375 | read_unlock(&tasklist_lock); | |
4376 | ||
4377 | retval = -EPERM; | |
4378 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4379 | !capable(CAP_SYS_NICE)) | |
4380 | goto out_unlock; | |
4381 | ||
e7834f8f DQ |
4382 | retval = security_task_setscheduler(p, 0, NULL); |
4383 | if (retval) | |
4384 | goto out_unlock; | |
4385 | ||
1da177e4 LT |
4386 | cpus_allowed = cpuset_cpus_allowed(p); |
4387 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4388 | retval = set_cpus_allowed(p, new_mask); | |
4389 | ||
4390 | out_unlock: | |
4391 | put_task_struct(p); | |
5be9361c | 4392 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4393 | return retval; |
4394 | } | |
4395 | ||
4396 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4397 | cpumask_t *new_mask) | |
4398 | { | |
4399 | if (len < sizeof(cpumask_t)) { | |
4400 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4401 | } else if (len > sizeof(cpumask_t)) { | |
4402 | len = sizeof(cpumask_t); | |
4403 | } | |
4404 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4405 | } | |
4406 | ||
4407 | /** | |
4408 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4409 | * @pid: pid of the process | |
4410 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4411 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4412 | */ | |
4413 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4414 | unsigned long __user *user_mask_ptr) | |
4415 | { | |
4416 | cpumask_t new_mask; | |
4417 | int retval; | |
4418 | ||
4419 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4420 | if (retval) | |
4421 | return retval; | |
4422 | ||
4423 | return sched_setaffinity(pid, new_mask); | |
4424 | } | |
4425 | ||
4426 | /* | |
4427 | * Represents all cpu's present in the system | |
4428 | * In systems capable of hotplug, this map could dynamically grow | |
4429 | * as new cpu's are detected in the system via any platform specific | |
4430 | * method, such as ACPI for e.g. | |
4431 | */ | |
4432 | ||
4cef0c61 | 4433 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4434 | EXPORT_SYMBOL(cpu_present_map); |
4435 | ||
4436 | #ifndef CONFIG_SMP | |
4cef0c61 | 4437 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4438 | EXPORT_SYMBOL(cpu_online_map); |
4439 | ||
4cef0c61 | 4440 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4441 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4442 | #endif |
4443 | ||
4444 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4445 | { | |
36c8b586 | 4446 | struct task_struct *p; |
1da177e4 | 4447 | int retval; |
1da177e4 | 4448 | |
5be9361c | 4449 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4450 | read_lock(&tasklist_lock); |
4451 | ||
4452 | retval = -ESRCH; | |
4453 | p = find_process_by_pid(pid); | |
4454 | if (!p) | |
4455 | goto out_unlock; | |
4456 | ||
e7834f8f DQ |
4457 | retval = security_task_getscheduler(p); |
4458 | if (retval) | |
4459 | goto out_unlock; | |
4460 | ||
2f7016d9 | 4461 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4462 | |
4463 | out_unlock: | |
4464 | read_unlock(&tasklist_lock); | |
5be9361c | 4465 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4466 | if (retval) |
4467 | return retval; | |
4468 | ||
4469 | return 0; | |
4470 | } | |
4471 | ||
4472 | /** | |
4473 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4474 | * @pid: pid of the process | |
4475 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4476 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4477 | */ | |
4478 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4479 | unsigned long __user *user_mask_ptr) | |
4480 | { | |
4481 | int ret; | |
4482 | cpumask_t mask; | |
4483 | ||
4484 | if (len < sizeof(cpumask_t)) | |
4485 | return -EINVAL; | |
4486 | ||
4487 | ret = sched_getaffinity(pid, &mask); | |
4488 | if (ret < 0) | |
4489 | return ret; | |
4490 | ||
4491 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4492 | return -EFAULT; | |
4493 | ||
4494 | return sizeof(cpumask_t); | |
4495 | } | |
4496 | ||
4497 | /** | |
4498 | * sys_sched_yield - yield the current processor to other threads. | |
4499 | * | |
dd41f596 IM |
4500 | * This function yields the current CPU to other tasks. If there are no |
4501 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4502 | */ |
4503 | asmlinkage long sys_sched_yield(void) | |
4504 | { | |
70b97a7f | 4505 | struct rq *rq = this_rq_lock(); |
1da177e4 LT |
4506 | |
4507 | schedstat_inc(rq, yld_cnt); | |
dd41f596 | 4508 | if (unlikely(rq->nr_running == 1)) |
1da177e4 | 4509 | schedstat_inc(rq, yld_act_empty); |
dd41f596 IM |
4510 | else |
4511 | current->sched_class->yield_task(rq, current); | |
1da177e4 LT |
4512 | |
4513 | /* | |
4514 | * Since we are going to call schedule() anyway, there's | |
4515 | * no need to preempt or enable interrupts: | |
4516 | */ | |
4517 | __release(rq->lock); | |
8a25d5de | 4518 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4519 | _raw_spin_unlock(&rq->lock); |
4520 | preempt_enable_no_resched(); | |
4521 | ||
4522 | schedule(); | |
4523 | ||
4524 | return 0; | |
4525 | } | |
4526 | ||
e7b38404 | 4527 | static void __cond_resched(void) |
1da177e4 | 4528 | { |
8e0a43d8 IM |
4529 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4530 | __might_sleep(__FILE__, __LINE__); | |
4531 | #endif | |
5bbcfd90 IM |
4532 | /* |
4533 | * The BKS might be reacquired before we have dropped | |
4534 | * PREEMPT_ACTIVE, which could trigger a second | |
4535 | * cond_resched() call. | |
4536 | */ | |
1da177e4 LT |
4537 | do { |
4538 | add_preempt_count(PREEMPT_ACTIVE); | |
4539 | schedule(); | |
4540 | sub_preempt_count(PREEMPT_ACTIVE); | |
4541 | } while (need_resched()); | |
4542 | } | |
4543 | ||
4544 | int __sched cond_resched(void) | |
4545 | { | |
9414232f IM |
4546 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4547 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4548 | __cond_resched(); |
4549 | return 1; | |
4550 | } | |
4551 | return 0; | |
4552 | } | |
1da177e4 LT |
4553 | EXPORT_SYMBOL(cond_resched); |
4554 | ||
4555 | /* | |
4556 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4557 | * call schedule, and on return reacquire the lock. | |
4558 | * | |
4559 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4560 | * operations here to prevent schedule() from being called twice (once via | |
4561 | * spin_unlock(), once by hand). | |
4562 | */ | |
95cdf3b7 | 4563 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4564 | { |
6df3cecb JK |
4565 | int ret = 0; |
4566 | ||
1da177e4 LT |
4567 | if (need_lockbreak(lock)) { |
4568 | spin_unlock(lock); | |
4569 | cpu_relax(); | |
6df3cecb | 4570 | ret = 1; |
1da177e4 LT |
4571 | spin_lock(lock); |
4572 | } | |
9414232f | 4573 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4574 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4575 | _raw_spin_unlock(lock); |
4576 | preempt_enable_no_resched(); | |
4577 | __cond_resched(); | |
6df3cecb | 4578 | ret = 1; |
1da177e4 | 4579 | spin_lock(lock); |
1da177e4 | 4580 | } |
6df3cecb | 4581 | return ret; |
1da177e4 | 4582 | } |
1da177e4 LT |
4583 | EXPORT_SYMBOL(cond_resched_lock); |
4584 | ||
4585 | int __sched cond_resched_softirq(void) | |
4586 | { | |
4587 | BUG_ON(!in_softirq()); | |
4588 | ||
9414232f | 4589 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4590 | local_bh_enable(); |
1da177e4 LT |
4591 | __cond_resched(); |
4592 | local_bh_disable(); | |
4593 | return 1; | |
4594 | } | |
4595 | return 0; | |
4596 | } | |
1da177e4 LT |
4597 | EXPORT_SYMBOL(cond_resched_softirq); |
4598 | ||
1da177e4 LT |
4599 | /** |
4600 | * yield - yield the current processor to other threads. | |
4601 | * | |
72fd4a35 | 4602 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4603 | * thread runnable and calls sys_sched_yield(). |
4604 | */ | |
4605 | void __sched yield(void) | |
4606 | { | |
4607 | set_current_state(TASK_RUNNING); | |
4608 | sys_sched_yield(); | |
4609 | } | |
1da177e4 LT |
4610 | EXPORT_SYMBOL(yield); |
4611 | ||
4612 | /* | |
4613 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4614 | * that process accounting knows that this is a task in IO wait state. | |
4615 | * | |
4616 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4617 | * has set its backing_dev_info: the queue against which it should throttle) | |
4618 | */ | |
4619 | void __sched io_schedule(void) | |
4620 | { | |
70b97a7f | 4621 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4622 | |
0ff92245 | 4623 | delayacct_blkio_start(); |
1da177e4 LT |
4624 | atomic_inc(&rq->nr_iowait); |
4625 | schedule(); | |
4626 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4627 | delayacct_blkio_end(); |
1da177e4 | 4628 | } |
1da177e4 LT |
4629 | EXPORT_SYMBOL(io_schedule); |
4630 | ||
4631 | long __sched io_schedule_timeout(long timeout) | |
4632 | { | |
70b97a7f | 4633 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4634 | long ret; |
4635 | ||
0ff92245 | 4636 | delayacct_blkio_start(); |
1da177e4 LT |
4637 | atomic_inc(&rq->nr_iowait); |
4638 | ret = schedule_timeout(timeout); | |
4639 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4640 | delayacct_blkio_end(); |
1da177e4 LT |
4641 | return ret; |
4642 | } | |
4643 | ||
4644 | /** | |
4645 | * sys_sched_get_priority_max - return maximum RT priority. | |
4646 | * @policy: scheduling class. | |
4647 | * | |
4648 | * this syscall returns the maximum rt_priority that can be used | |
4649 | * by a given scheduling class. | |
4650 | */ | |
4651 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4652 | { | |
4653 | int ret = -EINVAL; | |
4654 | ||
4655 | switch (policy) { | |
4656 | case SCHED_FIFO: | |
4657 | case SCHED_RR: | |
4658 | ret = MAX_USER_RT_PRIO-1; | |
4659 | break; | |
4660 | case SCHED_NORMAL: | |
b0a9499c | 4661 | case SCHED_BATCH: |
dd41f596 | 4662 | case SCHED_IDLE: |
1da177e4 LT |
4663 | ret = 0; |
4664 | break; | |
4665 | } | |
4666 | return ret; | |
4667 | } | |
4668 | ||
4669 | /** | |
4670 | * sys_sched_get_priority_min - return minimum RT priority. | |
4671 | * @policy: scheduling class. | |
4672 | * | |
4673 | * this syscall returns the minimum rt_priority that can be used | |
4674 | * by a given scheduling class. | |
4675 | */ | |
4676 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4677 | { | |
4678 | int ret = -EINVAL; | |
4679 | ||
4680 | switch (policy) { | |
4681 | case SCHED_FIFO: | |
4682 | case SCHED_RR: | |
4683 | ret = 1; | |
4684 | break; | |
4685 | case SCHED_NORMAL: | |
b0a9499c | 4686 | case SCHED_BATCH: |
dd41f596 | 4687 | case SCHED_IDLE: |
1da177e4 LT |
4688 | ret = 0; |
4689 | } | |
4690 | return ret; | |
4691 | } | |
4692 | ||
4693 | /** | |
4694 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4695 | * @pid: pid of the process. | |
4696 | * @interval: userspace pointer to the timeslice value. | |
4697 | * | |
4698 | * this syscall writes the default timeslice value of a given process | |
4699 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4700 | */ | |
4701 | asmlinkage | |
4702 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4703 | { | |
36c8b586 | 4704 | struct task_struct *p; |
1da177e4 LT |
4705 | int retval = -EINVAL; |
4706 | struct timespec t; | |
1da177e4 LT |
4707 | |
4708 | if (pid < 0) | |
4709 | goto out_nounlock; | |
4710 | ||
4711 | retval = -ESRCH; | |
4712 | read_lock(&tasklist_lock); | |
4713 | p = find_process_by_pid(pid); | |
4714 | if (!p) | |
4715 | goto out_unlock; | |
4716 | ||
4717 | retval = security_task_getscheduler(p); | |
4718 | if (retval) | |
4719 | goto out_unlock; | |
4720 | ||
b78709cf | 4721 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
dd41f596 | 4722 | 0 : static_prio_timeslice(p->static_prio), &t); |
1da177e4 LT |
4723 | read_unlock(&tasklist_lock); |
4724 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4725 | out_nounlock: | |
4726 | return retval; | |
4727 | out_unlock: | |
4728 | read_unlock(&tasklist_lock); | |
4729 | return retval; | |
4730 | } | |
4731 | ||
2ed6e34f | 4732 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4733 | |
4734 | static void show_task(struct task_struct *p) | |
1da177e4 | 4735 | { |
1da177e4 | 4736 | unsigned long free = 0; |
36c8b586 | 4737 | unsigned state; |
1da177e4 | 4738 | |
1da177e4 | 4739 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4740 | printk("%-13.13s %c", p->comm, |
4741 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4742 | #if BITS_PER_LONG == 32 |
1da177e4 | 4743 | if (state == TASK_RUNNING) |
4bd77321 | 4744 | printk(" running "); |
1da177e4 | 4745 | else |
4bd77321 | 4746 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4747 | #else |
4748 | if (state == TASK_RUNNING) | |
4bd77321 | 4749 | printk(" running task "); |
1da177e4 LT |
4750 | else |
4751 | printk(" %016lx ", thread_saved_pc(p)); | |
4752 | #endif | |
4753 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4754 | { | |
10ebffde | 4755 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4756 | while (!*n) |
4757 | n++; | |
10ebffde | 4758 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4759 | } |
4760 | #endif | |
4bd77321 | 4761 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4762 | |
4763 | if (state != TASK_RUNNING) | |
4764 | show_stack(p, NULL); | |
4765 | } | |
4766 | ||
e59e2ae2 | 4767 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4768 | { |
36c8b586 | 4769 | struct task_struct *g, *p; |
1da177e4 | 4770 | |
4bd77321 IM |
4771 | #if BITS_PER_LONG == 32 |
4772 | printk(KERN_INFO | |
4773 | " task PC stack pid father\n"); | |
1da177e4 | 4774 | #else |
4bd77321 IM |
4775 | printk(KERN_INFO |
4776 | " task PC stack pid father\n"); | |
1da177e4 LT |
4777 | #endif |
4778 | read_lock(&tasklist_lock); | |
4779 | do_each_thread(g, p) { | |
4780 | /* | |
4781 | * reset the NMI-timeout, listing all files on a slow | |
4782 | * console might take alot of time: | |
4783 | */ | |
4784 | touch_nmi_watchdog(); | |
39bc89fd | 4785 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4786 | show_task(p); |
1da177e4 LT |
4787 | } while_each_thread(g, p); |
4788 | ||
04c9167f JF |
4789 | touch_all_softlockup_watchdogs(); |
4790 | ||
dd41f596 IM |
4791 | #ifdef CONFIG_SCHED_DEBUG |
4792 | sysrq_sched_debug_show(); | |
4793 | #endif | |
1da177e4 | 4794 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4795 | /* |
4796 | * Only show locks if all tasks are dumped: | |
4797 | */ | |
4798 | if (state_filter == -1) | |
4799 | debug_show_all_locks(); | |
1da177e4 LT |
4800 | } |
4801 | ||
1df21055 IM |
4802 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4803 | { | |
dd41f596 | 4804 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4805 | } |
4806 | ||
f340c0d1 IM |
4807 | /** |
4808 | * init_idle - set up an idle thread for a given CPU | |
4809 | * @idle: task in question | |
4810 | * @cpu: cpu the idle task belongs to | |
4811 | * | |
4812 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4813 | * flag, to make booting more robust. | |
4814 | */ | |
5c1e1767 | 4815 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4816 | { |
70b97a7f | 4817 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4818 | unsigned long flags; |
4819 | ||
dd41f596 IM |
4820 | __sched_fork(idle); |
4821 | idle->se.exec_start = sched_clock(); | |
4822 | ||
b29739f9 | 4823 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4824 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4825 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4826 | |
4827 | spin_lock_irqsave(&rq->lock, flags); | |
4828 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4829 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4830 | idle->oncpu = 1; | |
4831 | #endif | |
1da177e4 LT |
4832 | spin_unlock_irqrestore(&rq->lock, flags); |
4833 | ||
4834 | /* Set the preempt count _outside_ the spinlocks! */ | |
4835 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4836 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4837 | #else |
a1261f54 | 4838 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4839 | #endif |
dd41f596 IM |
4840 | /* |
4841 | * The idle tasks have their own, simple scheduling class: | |
4842 | */ | |
4843 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4844 | } |
4845 | ||
4846 | /* | |
4847 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4848 | * indicates which cpus entered this state. This is used | |
4849 | * in the rcu update to wait only for active cpus. For system | |
4850 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4851 | * always be CPU_MASK_NONE. | |
4852 | */ | |
4853 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4854 | ||
dd41f596 IM |
4855 | /* |
4856 | * Increase the granularity value when there are more CPUs, | |
4857 | * because with more CPUs the 'effective latency' as visible | |
4858 | * to users decreases. But the relationship is not linear, | |
4859 | * so pick a second-best guess by going with the log2 of the | |
4860 | * number of CPUs. | |
4861 | * | |
4862 | * This idea comes from the SD scheduler of Con Kolivas: | |
4863 | */ | |
4864 | static inline void sched_init_granularity(void) | |
4865 | { | |
4866 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
a5968df8 | 4867 | const unsigned long gran_limit = 100000000; |
dd41f596 IM |
4868 | |
4869 | sysctl_sched_granularity *= factor; | |
4870 | if (sysctl_sched_granularity > gran_limit) | |
4871 | sysctl_sched_granularity = gran_limit; | |
4872 | ||
4873 | sysctl_sched_runtime_limit = sysctl_sched_granularity * 4; | |
4874 | sysctl_sched_wakeup_granularity = sysctl_sched_granularity / 2; | |
4875 | } | |
4876 | ||
1da177e4 LT |
4877 | #ifdef CONFIG_SMP |
4878 | /* | |
4879 | * This is how migration works: | |
4880 | * | |
70b97a7f | 4881 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4882 | * runqueue and wake up that CPU's migration thread. |
4883 | * 2) we down() the locked semaphore => thread blocks. | |
4884 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4885 | * thread off the CPU) | |
4886 | * 4) it gets the migration request and checks whether the migrated | |
4887 | * task is still in the wrong runqueue. | |
4888 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4889 | * it and puts it into the right queue. | |
4890 | * 6) migration thread up()s the semaphore. | |
4891 | * 7) we wake up and the migration is done. | |
4892 | */ | |
4893 | ||
4894 | /* | |
4895 | * Change a given task's CPU affinity. Migrate the thread to a | |
4896 | * proper CPU and schedule it away if the CPU it's executing on | |
4897 | * is removed from the allowed bitmask. | |
4898 | * | |
4899 | * NOTE: the caller must have a valid reference to the task, the | |
4900 | * task must not exit() & deallocate itself prematurely. The | |
4901 | * call is not atomic; no spinlocks may be held. | |
4902 | */ | |
36c8b586 | 4903 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4904 | { |
70b97a7f | 4905 | struct migration_req req; |
1da177e4 | 4906 | unsigned long flags; |
70b97a7f | 4907 | struct rq *rq; |
48f24c4d | 4908 | int ret = 0; |
1da177e4 LT |
4909 | |
4910 | rq = task_rq_lock(p, &flags); | |
4911 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4912 | ret = -EINVAL; | |
4913 | goto out; | |
4914 | } | |
4915 | ||
4916 | p->cpus_allowed = new_mask; | |
4917 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4918 | if (cpu_isset(task_cpu(p), new_mask)) | |
4919 | goto out; | |
4920 | ||
4921 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4922 | /* Need help from migration thread: drop lock and wait. */ | |
4923 | task_rq_unlock(rq, &flags); | |
4924 | wake_up_process(rq->migration_thread); | |
4925 | wait_for_completion(&req.done); | |
4926 | tlb_migrate_finish(p->mm); | |
4927 | return 0; | |
4928 | } | |
4929 | out: | |
4930 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4931 | |
1da177e4 LT |
4932 | return ret; |
4933 | } | |
1da177e4 LT |
4934 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4935 | ||
4936 | /* | |
4937 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4938 | * this because either it can't run here any more (set_cpus_allowed() | |
4939 | * away from this CPU, or CPU going down), or because we're | |
4940 | * attempting to rebalance this task on exec (sched_exec). | |
4941 | * | |
4942 | * So we race with normal scheduler movements, but that's OK, as long | |
4943 | * as the task is no longer on this CPU. | |
efc30814 KK |
4944 | * |
4945 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4946 | */ |
efc30814 | 4947 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4948 | { |
70b97a7f | 4949 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4950 | int ret = 0, on_rq; |
1da177e4 LT |
4951 | |
4952 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4953 | return ret; |
1da177e4 LT |
4954 | |
4955 | rq_src = cpu_rq(src_cpu); | |
4956 | rq_dest = cpu_rq(dest_cpu); | |
4957 | ||
4958 | double_rq_lock(rq_src, rq_dest); | |
4959 | /* Already moved. */ | |
4960 | if (task_cpu(p) != src_cpu) | |
4961 | goto out; | |
4962 | /* Affinity changed (again). */ | |
4963 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4964 | goto out; | |
4965 | ||
dd41f596 IM |
4966 | on_rq = p->se.on_rq; |
4967 | if (on_rq) | |
4968 | deactivate_task(rq_src, p, 0); | |
1da177e4 | 4969 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4970 | if (on_rq) { |
4971 | activate_task(rq_dest, p, 0); | |
4972 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4973 | } |
efc30814 | 4974 | ret = 1; |
1da177e4 LT |
4975 | out: |
4976 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4977 | return ret; |
1da177e4 LT |
4978 | } |
4979 | ||
4980 | /* | |
4981 | * migration_thread - this is a highprio system thread that performs | |
4982 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4983 | * another runqueue. | |
4984 | */ | |
95cdf3b7 | 4985 | static int migration_thread(void *data) |
1da177e4 | 4986 | { |
1da177e4 | 4987 | int cpu = (long)data; |
70b97a7f | 4988 | struct rq *rq; |
1da177e4 LT |
4989 | |
4990 | rq = cpu_rq(cpu); | |
4991 | BUG_ON(rq->migration_thread != current); | |
4992 | ||
4993 | set_current_state(TASK_INTERRUPTIBLE); | |
4994 | while (!kthread_should_stop()) { | |
70b97a7f | 4995 | struct migration_req *req; |
1da177e4 | 4996 | struct list_head *head; |
1da177e4 | 4997 | |
1da177e4 LT |
4998 | spin_lock_irq(&rq->lock); |
4999 | ||
5000 | if (cpu_is_offline(cpu)) { | |
5001 | spin_unlock_irq(&rq->lock); | |
5002 | goto wait_to_die; | |
5003 | } | |
5004 | ||
5005 | if (rq->active_balance) { | |
5006 | active_load_balance(rq, cpu); | |
5007 | rq->active_balance = 0; | |
5008 | } | |
5009 | ||
5010 | head = &rq->migration_queue; | |
5011 | ||
5012 | if (list_empty(head)) { | |
5013 | spin_unlock_irq(&rq->lock); | |
5014 | schedule(); | |
5015 | set_current_state(TASK_INTERRUPTIBLE); | |
5016 | continue; | |
5017 | } | |
70b97a7f | 5018 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5019 | list_del_init(head->next); |
5020 | ||
674311d5 NP |
5021 | spin_unlock(&rq->lock); |
5022 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5023 | local_irq_enable(); | |
1da177e4 LT |
5024 | |
5025 | complete(&req->done); | |
5026 | } | |
5027 | __set_current_state(TASK_RUNNING); | |
5028 | return 0; | |
5029 | ||
5030 | wait_to_die: | |
5031 | /* Wait for kthread_stop */ | |
5032 | set_current_state(TASK_INTERRUPTIBLE); | |
5033 | while (!kthread_should_stop()) { | |
5034 | schedule(); | |
5035 | set_current_state(TASK_INTERRUPTIBLE); | |
5036 | } | |
5037 | __set_current_state(TASK_RUNNING); | |
5038 | return 0; | |
5039 | } | |
5040 | ||
5041 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5042 | /* |
5043 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5044 | * NOTE: interrupts should be disabled by the caller | |
5045 | */ | |
48f24c4d | 5046 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5047 | { |
efc30814 | 5048 | unsigned long flags; |
1da177e4 | 5049 | cpumask_t mask; |
70b97a7f IM |
5050 | struct rq *rq; |
5051 | int dest_cpu; | |
1da177e4 | 5052 | |
efc30814 | 5053 | restart: |
1da177e4 LT |
5054 | /* On same node? */ |
5055 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5056 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5057 | dest_cpu = any_online_cpu(mask); |
5058 | ||
5059 | /* On any allowed CPU? */ | |
5060 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5061 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5062 | |
5063 | /* No more Mr. Nice Guy. */ | |
5064 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5065 | rq = task_rq_lock(p, &flags); |
5066 | cpus_setall(p->cpus_allowed); | |
5067 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5068 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5069 | |
5070 | /* | |
5071 | * Don't tell them about moving exiting tasks or | |
5072 | * kernel threads (both mm NULL), since they never | |
5073 | * leave kernel. | |
5074 | */ | |
48f24c4d | 5075 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5076 | printk(KERN_INFO "process %d (%s) no " |
5077 | "longer affine to cpu%d\n", | |
48f24c4d | 5078 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5079 | } |
48f24c4d | 5080 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5081 | goto restart; |
1da177e4 LT |
5082 | } |
5083 | ||
5084 | /* | |
5085 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5086 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5087 | * for performance reasons the counter is not stricly tracking tasks to | |
5088 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5089 | * to keep the global sum constant after CPU-down: | |
5090 | */ | |
70b97a7f | 5091 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5092 | { |
70b97a7f | 5093 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5094 | unsigned long flags; |
5095 | ||
5096 | local_irq_save(flags); | |
5097 | double_rq_lock(rq_src, rq_dest); | |
5098 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5099 | rq_src->nr_uninterruptible = 0; | |
5100 | double_rq_unlock(rq_src, rq_dest); | |
5101 | local_irq_restore(flags); | |
5102 | } | |
5103 | ||
5104 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5105 | static void migrate_live_tasks(int src_cpu) | |
5106 | { | |
48f24c4d | 5107 | struct task_struct *p, *t; |
1da177e4 LT |
5108 | |
5109 | write_lock_irq(&tasklist_lock); | |
5110 | ||
48f24c4d IM |
5111 | do_each_thread(t, p) { |
5112 | if (p == current) | |
1da177e4 LT |
5113 | continue; |
5114 | ||
48f24c4d IM |
5115 | if (task_cpu(p) == src_cpu) |
5116 | move_task_off_dead_cpu(src_cpu, p); | |
5117 | } while_each_thread(t, p); | |
1da177e4 LT |
5118 | |
5119 | write_unlock_irq(&tasklist_lock); | |
5120 | } | |
5121 | ||
dd41f596 IM |
5122 | /* |
5123 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5124 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5125 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5126 | */ |
5127 | void sched_idle_next(void) | |
5128 | { | |
48f24c4d | 5129 | int this_cpu = smp_processor_id(); |
70b97a7f | 5130 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5131 | struct task_struct *p = rq->idle; |
5132 | unsigned long flags; | |
5133 | ||
5134 | /* cpu has to be offline */ | |
48f24c4d | 5135 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5136 | |
48f24c4d IM |
5137 | /* |
5138 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5139 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5140 | */ |
5141 | spin_lock_irqsave(&rq->lock, flags); | |
5142 | ||
dd41f596 | 5143 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5144 | |
5145 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5146 | activate_idle_task(p, rq); |
1da177e4 LT |
5147 | |
5148 | spin_unlock_irqrestore(&rq->lock, flags); | |
5149 | } | |
5150 | ||
48f24c4d IM |
5151 | /* |
5152 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5153 | * offline. |
5154 | */ | |
5155 | void idle_task_exit(void) | |
5156 | { | |
5157 | struct mm_struct *mm = current->active_mm; | |
5158 | ||
5159 | BUG_ON(cpu_online(smp_processor_id())); | |
5160 | ||
5161 | if (mm != &init_mm) | |
5162 | switch_mm(mm, &init_mm, current); | |
5163 | mmdrop(mm); | |
5164 | } | |
5165 | ||
054b9108 | 5166 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5167 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5168 | { |
70b97a7f | 5169 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5170 | |
5171 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5172 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5173 | |
5174 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5175 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5176 | |
48f24c4d | 5177 | get_task_struct(p); |
1da177e4 LT |
5178 | |
5179 | /* | |
5180 | * Drop lock around migration; if someone else moves it, | |
5181 | * that's OK. No task can be added to this CPU, so iteration is | |
5182 | * fine. | |
054b9108 | 5183 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5184 | */ |
054b9108 | 5185 | spin_unlock(&rq->lock); |
48f24c4d | 5186 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5187 | spin_lock(&rq->lock); |
1da177e4 | 5188 | |
48f24c4d | 5189 | put_task_struct(p); |
1da177e4 LT |
5190 | } |
5191 | ||
5192 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5193 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5194 | { | |
70b97a7f | 5195 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5196 | struct task_struct *next; |
48f24c4d | 5197 | |
dd41f596 IM |
5198 | for ( ; ; ) { |
5199 | if (!rq->nr_running) | |
5200 | break; | |
5201 | next = pick_next_task(rq, rq->curr, rq_clock(rq)); | |
5202 | if (!next) | |
5203 | break; | |
5204 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5205 | |
1da177e4 LT |
5206 | } |
5207 | } | |
5208 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5209 | ||
e692ab53 NP |
5210 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5211 | ||
5212 | static struct ctl_table sd_ctl_dir[] = { | |
5213 | {CTL_UNNUMBERED, "sched_domain", NULL, 0, 0755, NULL, }, | |
5214 | {0,}, | |
5215 | }; | |
5216 | ||
5217 | static struct ctl_table sd_ctl_root[] = { | |
5218 | {CTL_UNNUMBERED, "kernel", NULL, 0, 0755, sd_ctl_dir, }, | |
5219 | {0,}, | |
5220 | }; | |
5221 | ||
5222 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5223 | { | |
5224 | struct ctl_table *entry = | |
5225 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); | |
5226 | ||
5227 | BUG_ON(!entry); | |
5228 | memset(entry, 0, n * sizeof(struct ctl_table)); | |
5229 | ||
5230 | return entry; | |
5231 | } | |
5232 | ||
5233 | static void | |
5234 | set_table_entry(struct ctl_table *entry, int ctl_name, | |
5235 | const char *procname, void *data, int maxlen, | |
5236 | mode_t mode, proc_handler *proc_handler) | |
5237 | { | |
5238 | entry->ctl_name = ctl_name; | |
5239 | entry->procname = procname; | |
5240 | entry->data = data; | |
5241 | entry->maxlen = maxlen; | |
5242 | entry->mode = mode; | |
5243 | entry->proc_handler = proc_handler; | |
5244 | } | |
5245 | ||
5246 | static struct ctl_table * | |
5247 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5248 | { | |
5249 | struct ctl_table *table = sd_alloc_ctl_entry(14); | |
5250 | ||
5251 | set_table_entry(&table[0], 1, "min_interval", &sd->min_interval, | |
5252 | sizeof(long), 0644, proc_doulongvec_minmax); | |
5253 | set_table_entry(&table[1], 2, "max_interval", &sd->max_interval, | |
5254 | sizeof(long), 0644, proc_doulongvec_minmax); | |
5255 | set_table_entry(&table[2], 3, "busy_idx", &sd->busy_idx, | |
5256 | sizeof(int), 0644, proc_dointvec_minmax); | |
5257 | set_table_entry(&table[3], 4, "idle_idx", &sd->idle_idx, | |
5258 | sizeof(int), 0644, proc_dointvec_minmax); | |
5259 | set_table_entry(&table[4], 5, "newidle_idx", &sd->newidle_idx, | |
5260 | sizeof(int), 0644, proc_dointvec_minmax); | |
5261 | set_table_entry(&table[5], 6, "wake_idx", &sd->wake_idx, | |
5262 | sizeof(int), 0644, proc_dointvec_minmax); | |
5263 | set_table_entry(&table[6], 7, "forkexec_idx", &sd->forkexec_idx, | |
5264 | sizeof(int), 0644, proc_dointvec_minmax); | |
5265 | set_table_entry(&table[7], 8, "busy_factor", &sd->busy_factor, | |
5266 | sizeof(int), 0644, proc_dointvec_minmax); | |
5267 | set_table_entry(&table[8], 9, "imbalance_pct", &sd->imbalance_pct, | |
5268 | sizeof(int), 0644, proc_dointvec_minmax); | |
e692ab53 NP |
5269 | set_table_entry(&table[10], 11, "cache_nice_tries", |
5270 | &sd->cache_nice_tries, | |
5271 | sizeof(int), 0644, proc_dointvec_minmax); | |
5272 | set_table_entry(&table[12], 13, "flags", &sd->flags, | |
5273 | sizeof(int), 0644, proc_dointvec_minmax); | |
5274 | ||
5275 | return table; | |
5276 | } | |
5277 | ||
5278 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | |
5279 | { | |
5280 | struct ctl_table *entry, *table; | |
5281 | struct sched_domain *sd; | |
5282 | int domain_num = 0, i; | |
5283 | char buf[32]; | |
5284 | ||
5285 | for_each_domain(cpu, sd) | |
5286 | domain_num++; | |
5287 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
5288 | ||
5289 | i = 0; | |
5290 | for_each_domain(cpu, sd) { | |
5291 | snprintf(buf, 32, "domain%d", i); | |
5292 | entry->ctl_name = i + 1; | |
5293 | entry->procname = kstrdup(buf, GFP_KERNEL); | |
5294 | entry->mode = 0755; | |
5295 | entry->child = sd_alloc_ctl_domain_table(sd); | |
5296 | entry++; | |
5297 | i++; | |
5298 | } | |
5299 | return table; | |
5300 | } | |
5301 | ||
5302 | static struct ctl_table_header *sd_sysctl_header; | |
5303 | static void init_sched_domain_sysctl(void) | |
5304 | { | |
5305 | int i, cpu_num = num_online_cpus(); | |
5306 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5307 | char buf[32]; | |
5308 | ||
5309 | sd_ctl_dir[0].child = entry; | |
5310 | ||
5311 | for (i = 0; i < cpu_num; i++, entry++) { | |
5312 | snprintf(buf, 32, "cpu%d", i); | |
5313 | entry->ctl_name = i + 1; | |
5314 | entry->procname = kstrdup(buf, GFP_KERNEL); | |
5315 | entry->mode = 0755; | |
5316 | entry->child = sd_alloc_ctl_cpu_table(i); | |
5317 | } | |
5318 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5319 | } | |
5320 | #else | |
5321 | static void init_sched_domain_sysctl(void) | |
5322 | { | |
5323 | } | |
5324 | #endif | |
5325 | ||
1da177e4 LT |
5326 | /* |
5327 | * migration_call - callback that gets triggered when a CPU is added. | |
5328 | * Here we can start up the necessary migration thread for the new CPU. | |
5329 | */ | |
48f24c4d IM |
5330 | static int __cpuinit |
5331 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5332 | { |
1da177e4 | 5333 | struct task_struct *p; |
48f24c4d | 5334 | int cpu = (long)hcpu; |
1da177e4 | 5335 | unsigned long flags; |
70b97a7f | 5336 | struct rq *rq; |
1da177e4 LT |
5337 | |
5338 | switch (action) { | |
5be9361c GS |
5339 | case CPU_LOCK_ACQUIRE: |
5340 | mutex_lock(&sched_hotcpu_mutex); | |
5341 | break; | |
5342 | ||
1da177e4 | 5343 | case CPU_UP_PREPARE: |
8bb78442 | 5344 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5345 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5346 | if (IS_ERR(p)) |
5347 | return NOTIFY_BAD; | |
1da177e4 LT |
5348 | kthread_bind(p, cpu); |
5349 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5350 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5351 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5352 | task_rq_unlock(rq, &flags); |
5353 | cpu_rq(cpu)->migration_thread = p; | |
5354 | break; | |
48f24c4d | 5355 | |
1da177e4 | 5356 | case CPU_ONLINE: |
8bb78442 | 5357 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5358 | /* Strictly unneccessary, as first user will wake it. */ |
5359 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5360 | break; | |
48f24c4d | 5361 | |
1da177e4 LT |
5362 | #ifdef CONFIG_HOTPLUG_CPU |
5363 | case CPU_UP_CANCELED: | |
8bb78442 | 5364 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5365 | if (!cpu_rq(cpu)->migration_thread) |
5366 | break; | |
1da177e4 | 5367 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5368 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5369 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5370 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5371 | cpu_rq(cpu)->migration_thread = NULL; | |
5372 | break; | |
48f24c4d | 5373 | |
1da177e4 | 5374 | case CPU_DEAD: |
8bb78442 | 5375 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5376 | migrate_live_tasks(cpu); |
5377 | rq = cpu_rq(cpu); | |
5378 | kthread_stop(rq->migration_thread); | |
5379 | rq->migration_thread = NULL; | |
5380 | /* Idle task back to normal (off runqueue, low prio) */ | |
5381 | rq = task_rq_lock(rq->idle, &flags); | |
dd41f596 | 5382 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5383 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5384 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5385 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5386 | migrate_dead_tasks(cpu); |
5387 | task_rq_unlock(rq, &flags); | |
5388 | migrate_nr_uninterruptible(rq); | |
5389 | BUG_ON(rq->nr_running != 0); | |
5390 | ||
5391 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5392 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5393 | * the requestors. */ |
5394 | spin_lock_irq(&rq->lock); | |
5395 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5396 | struct migration_req *req; |
5397 | ||
1da177e4 | 5398 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5399 | struct migration_req, list); |
1da177e4 LT |
5400 | list_del_init(&req->list); |
5401 | complete(&req->done); | |
5402 | } | |
5403 | spin_unlock_irq(&rq->lock); | |
5404 | break; | |
5405 | #endif | |
5be9361c GS |
5406 | case CPU_LOCK_RELEASE: |
5407 | mutex_unlock(&sched_hotcpu_mutex); | |
5408 | break; | |
1da177e4 LT |
5409 | } |
5410 | return NOTIFY_OK; | |
5411 | } | |
5412 | ||
5413 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5414 | * happens before everything else. | |
5415 | */ | |
26c2143b | 5416 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5417 | .notifier_call = migration_call, |
5418 | .priority = 10 | |
5419 | }; | |
5420 | ||
5421 | int __init migration_init(void) | |
5422 | { | |
5423 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5424 | int err; |
48f24c4d IM |
5425 | |
5426 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5427 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5428 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5429 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5430 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5431 | |
1da177e4 LT |
5432 | return 0; |
5433 | } | |
5434 | #endif | |
5435 | ||
5436 | #ifdef CONFIG_SMP | |
476f3534 CL |
5437 | |
5438 | /* Number of possible processor ids */ | |
5439 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5440 | EXPORT_SYMBOL(nr_cpu_ids); | |
5441 | ||
1a20ff27 | 5442 | #undef SCHED_DOMAIN_DEBUG |
1da177e4 LT |
5443 | #ifdef SCHED_DOMAIN_DEBUG |
5444 | static void sched_domain_debug(struct sched_domain *sd, int cpu) | |
5445 | { | |
5446 | int level = 0; | |
5447 | ||
41c7ce9a NP |
5448 | if (!sd) { |
5449 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5450 | return; | |
5451 | } | |
5452 | ||
1da177e4 LT |
5453 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5454 | ||
5455 | do { | |
5456 | int i; | |
5457 | char str[NR_CPUS]; | |
5458 | struct sched_group *group = sd->groups; | |
5459 | cpumask_t groupmask; | |
5460 | ||
5461 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5462 | cpus_clear(groupmask); | |
5463 | ||
5464 | printk(KERN_DEBUG); | |
5465 | for (i = 0; i < level + 1; i++) | |
5466 | printk(" "); | |
5467 | printk("domain %d: ", level); | |
5468 | ||
5469 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5470 | printk("does not load-balance\n"); | |
5471 | if (sd->parent) | |
33859f7f MOS |
5472 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5473 | " has parent"); | |
1da177e4 LT |
5474 | break; |
5475 | } | |
5476 | ||
5477 | printk("span %s\n", str); | |
5478 | ||
5479 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5480 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5481 | "CPU%d\n", cpu); | |
1da177e4 | 5482 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5483 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5484 | " CPU%d\n", cpu); | |
1da177e4 LT |
5485 | |
5486 | printk(KERN_DEBUG); | |
5487 | for (i = 0; i < level + 2; i++) | |
5488 | printk(" "); | |
5489 | printk("groups:"); | |
5490 | do { | |
5491 | if (!group) { | |
5492 | printk("\n"); | |
5493 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5494 | break; | |
5495 | } | |
5496 | ||
5517d86b | 5497 | if (!group->__cpu_power) { |
1da177e4 | 5498 | printk("\n"); |
33859f7f MOS |
5499 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5500 | "set\n"); | |
1da177e4 LT |
5501 | } |
5502 | ||
5503 | if (!cpus_weight(group->cpumask)) { | |
5504 | printk("\n"); | |
5505 | printk(KERN_ERR "ERROR: empty group\n"); | |
5506 | } | |
5507 | ||
5508 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5509 | printk("\n"); | |
5510 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5511 | } | |
5512 | ||
5513 | cpus_or(groupmask, groupmask, group->cpumask); | |
5514 | ||
5515 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5516 | printk(" %s", str); | |
5517 | ||
5518 | group = group->next; | |
5519 | } while (group != sd->groups); | |
5520 | printk("\n"); | |
5521 | ||
5522 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5523 | printk(KERN_ERR "ERROR: groups don't span " |
5524 | "domain->span\n"); | |
1da177e4 LT |
5525 | |
5526 | level++; | |
5527 | sd = sd->parent; | |
33859f7f MOS |
5528 | if (!sd) |
5529 | continue; | |
1da177e4 | 5530 | |
33859f7f MOS |
5531 | if (!cpus_subset(groupmask, sd->span)) |
5532 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5533 | "of domain->span\n"); | |
1da177e4 LT |
5534 | |
5535 | } while (sd); | |
5536 | } | |
5537 | #else | |
48f24c4d | 5538 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5539 | #endif |
5540 | ||
1a20ff27 | 5541 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5542 | { |
5543 | if (cpus_weight(sd->span) == 1) | |
5544 | return 1; | |
5545 | ||
5546 | /* Following flags need at least 2 groups */ | |
5547 | if (sd->flags & (SD_LOAD_BALANCE | | |
5548 | SD_BALANCE_NEWIDLE | | |
5549 | SD_BALANCE_FORK | | |
89c4710e SS |
5550 | SD_BALANCE_EXEC | |
5551 | SD_SHARE_CPUPOWER | | |
5552 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5553 | if (sd->groups != sd->groups->next) |
5554 | return 0; | |
5555 | } | |
5556 | ||
5557 | /* Following flags don't use groups */ | |
5558 | if (sd->flags & (SD_WAKE_IDLE | | |
5559 | SD_WAKE_AFFINE | | |
5560 | SD_WAKE_BALANCE)) | |
5561 | return 0; | |
5562 | ||
5563 | return 1; | |
5564 | } | |
5565 | ||
48f24c4d IM |
5566 | static int |
5567 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5568 | { |
5569 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5570 | ||
5571 | if (sd_degenerate(parent)) | |
5572 | return 1; | |
5573 | ||
5574 | if (!cpus_equal(sd->span, parent->span)) | |
5575 | return 0; | |
5576 | ||
5577 | /* Does parent contain flags not in child? */ | |
5578 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5579 | if (cflags & SD_WAKE_AFFINE) | |
5580 | pflags &= ~SD_WAKE_BALANCE; | |
5581 | /* Flags needing groups don't count if only 1 group in parent */ | |
5582 | if (parent->groups == parent->groups->next) { | |
5583 | pflags &= ~(SD_LOAD_BALANCE | | |
5584 | SD_BALANCE_NEWIDLE | | |
5585 | SD_BALANCE_FORK | | |
89c4710e SS |
5586 | SD_BALANCE_EXEC | |
5587 | SD_SHARE_CPUPOWER | | |
5588 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5589 | } |
5590 | if (~cflags & pflags) | |
5591 | return 0; | |
5592 | ||
5593 | return 1; | |
5594 | } | |
5595 | ||
1da177e4 LT |
5596 | /* |
5597 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5598 | * hold the hotplug lock. | |
5599 | */ | |
9c1cfda2 | 5600 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5601 | { |
70b97a7f | 5602 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5603 | struct sched_domain *tmp; |
5604 | ||
5605 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5606 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5607 | struct sched_domain *parent = tmp->parent; | |
5608 | if (!parent) | |
5609 | break; | |
1a848870 | 5610 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5611 | tmp->parent = parent->parent; |
1a848870 SS |
5612 | if (parent->parent) |
5613 | parent->parent->child = tmp; | |
5614 | } | |
245af2c7 SS |
5615 | } |
5616 | ||
1a848870 | 5617 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5618 | sd = sd->parent; |
1a848870 SS |
5619 | if (sd) |
5620 | sd->child = NULL; | |
5621 | } | |
1da177e4 LT |
5622 | |
5623 | sched_domain_debug(sd, cpu); | |
5624 | ||
674311d5 | 5625 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5626 | } |
5627 | ||
5628 | /* cpus with isolated domains */ | |
67af63a6 | 5629 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5630 | |
5631 | /* Setup the mask of cpus configured for isolated domains */ | |
5632 | static int __init isolated_cpu_setup(char *str) | |
5633 | { | |
5634 | int ints[NR_CPUS], i; | |
5635 | ||
5636 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5637 | cpus_clear(cpu_isolated_map); | |
5638 | for (i = 1; i <= ints[0]; i++) | |
5639 | if (ints[i] < NR_CPUS) | |
5640 | cpu_set(ints[i], cpu_isolated_map); | |
5641 | return 1; | |
5642 | } | |
5643 | ||
5644 | __setup ("isolcpus=", isolated_cpu_setup); | |
5645 | ||
5646 | /* | |
6711cab4 SS |
5647 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5648 | * to a function which identifies what group(along with sched group) a CPU | |
5649 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5650 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5651 | * |
5652 | * init_sched_build_groups will build a circular linked list of the groups | |
5653 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5654 | * and ->cpu_power to 0. | |
5655 | */ | |
a616058b | 5656 | static void |
6711cab4 SS |
5657 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5658 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5659 | struct sched_group **sg)) | |
1da177e4 LT |
5660 | { |
5661 | struct sched_group *first = NULL, *last = NULL; | |
5662 | cpumask_t covered = CPU_MASK_NONE; | |
5663 | int i; | |
5664 | ||
5665 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5666 | struct sched_group *sg; |
5667 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5668 | int j; |
5669 | ||
5670 | if (cpu_isset(i, covered)) | |
5671 | continue; | |
5672 | ||
5673 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5674 | sg->__cpu_power = 0; |
1da177e4 LT |
5675 | |
5676 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5677 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5678 | continue; |
5679 | ||
5680 | cpu_set(j, covered); | |
5681 | cpu_set(j, sg->cpumask); | |
5682 | } | |
5683 | if (!first) | |
5684 | first = sg; | |
5685 | if (last) | |
5686 | last->next = sg; | |
5687 | last = sg; | |
5688 | } | |
5689 | last->next = first; | |
5690 | } | |
5691 | ||
9c1cfda2 | 5692 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5693 | |
9c1cfda2 | 5694 | #ifdef CONFIG_NUMA |
198e2f18 | 5695 | |
9c1cfda2 JH |
5696 | /** |
5697 | * find_next_best_node - find the next node to include in a sched_domain | |
5698 | * @node: node whose sched_domain we're building | |
5699 | * @used_nodes: nodes already in the sched_domain | |
5700 | * | |
5701 | * Find the next node to include in a given scheduling domain. Simply | |
5702 | * finds the closest node not already in the @used_nodes map. | |
5703 | * | |
5704 | * Should use nodemask_t. | |
5705 | */ | |
5706 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5707 | { | |
5708 | int i, n, val, min_val, best_node = 0; | |
5709 | ||
5710 | min_val = INT_MAX; | |
5711 | ||
5712 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5713 | /* Start at @node */ | |
5714 | n = (node + i) % MAX_NUMNODES; | |
5715 | ||
5716 | if (!nr_cpus_node(n)) | |
5717 | continue; | |
5718 | ||
5719 | /* Skip already used nodes */ | |
5720 | if (test_bit(n, used_nodes)) | |
5721 | continue; | |
5722 | ||
5723 | /* Simple min distance search */ | |
5724 | val = node_distance(node, n); | |
5725 | ||
5726 | if (val < min_val) { | |
5727 | min_val = val; | |
5728 | best_node = n; | |
5729 | } | |
5730 | } | |
5731 | ||
5732 | set_bit(best_node, used_nodes); | |
5733 | return best_node; | |
5734 | } | |
5735 | ||
5736 | /** | |
5737 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5738 | * @node: node whose cpumask we're constructing | |
5739 | * @size: number of nodes to include in this span | |
5740 | * | |
5741 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5742 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5743 | * out optimally. | |
5744 | */ | |
5745 | static cpumask_t sched_domain_node_span(int node) | |
5746 | { | |
9c1cfda2 | 5747 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5748 | cpumask_t span, nodemask; |
5749 | int i; | |
9c1cfda2 JH |
5750 | |
5751 | cpus_clear(span); | |
5752 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5753 | ||
5754 | nodemask = node_to_cpumask(node); | |
5755 | cpus_or(span, span, nodemask); | |
5756 | set_bit(node, used_nodes); | |
5757 | ||
5758 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5759 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5760 | |
9c1cfda2 JH |
5761 | nodemask = node_to_cpumask(next_node); |
5762 | cpus_or(span, span, nodemask); | |
5763 | } | |
5764 | ||
5765 | return span; | |
5766 | } | |
5767 | #endif | |
5768 | ||
5c45bf27 | 5769 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5770 | |
9c1cfda2 | 5771 | /* |
48f24c4d | 5772 | * SMT sched-domains: |
9c1cfda2 | 5773 | */ |
1da177e4 LT |
5774 | #ifdef CONFIG_SCHED_SMT |
5775 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5776 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5777 | |
6711cab4 SS |
5778 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5779 | struct sched_group **sg) | |
1da177e4 | 5780 | { |
6711cab4 SS |
5781 | if (sg) |
5782 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5783 | return cpu; |
5784 | } | |
5785 | #endif | |
5786 | ||
48f24c4d IM |
5787 | /* |
5788 | * multi-core sched-domains: | |
5789 | */ | |
1e9f28fa SS |
5790 | #ifdef CONFIG_SCHED_MC |
5791 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5792 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5793 | #endif |
5794 | ||
5795 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5796 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5797 | struct sched_group **sg) | |
1e9f28fa | 5798 | { |
6711cab4 | 5799 | int group; |
a616058b SS |
5800 | cpumask_t mask = cpu_sibling_map[cpu]; |
5801 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5802 | group = first_cpu(mask); |
5803 | if (sg) | |
5804 | *sg = &per_cpu(sched_group_core, group); | |
5805 | return group; | |
1e9f28fa SS |
5806 | } |
5807 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5808 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5809 | struct sched_group **sg) | |
1e9f28fa | 5810 | { |
6711cab4 SS |
5811 | if (sg) |
5812 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5813 | return cpu; |
5814 | } | |
5815 | #endif | |
5816 | ||
1da177e4 | 5817 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5818 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5819 | |
6711cab4 SS |
5820 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5821 | struct sched_group **sg) | |
1da177e4 | 5822 | { |
6711cab4 | 5823 | int group; |
48f24c4d | 5824 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5825 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5826 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5827 | group = first_cpu(mask); |
1e9f28fa | 5828 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5829 | cpumask_t mask = cpu_sibling_map[cpu]; |
5830 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5831 | group = first_cpu(mask); |
1da177e4 | 5832 | #else |
6711cab4 | 5833 | group = cpu; |
1da177e4 | 5834 | #endif |
6711cab4 SS |
5835 | if (sg) |
5836 | *sg = &per_cpu(sched_group_phys, group); | |
5837 | return group; | |
1da177e4 LT |
5838 | } |
5839 | ||
5840 | #ifdef CONFIG_NUMA | |
1da177e4 | 5841 | /* |
9c1cfda2 JH |
5842 | * The init_sched_build_groups can't handle what we want to do with node |
5843 | * groups, so roll our own. Now each node has its own list of groups which | |
5844 | * gets dynamically allocated. | |
1da177e4 | 5845 | */ |
9c1cfda2 | 5846 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5847 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5848 | |
9c1cfda2 | 5849 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5850 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5851 | |
6711cab4 SS |
5852 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5853 | struct sched_group **sg) | |
9c1cfda2 | 5854 | { |
6711cab4 SS |
5855 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5856 | int group; | |
5857 | ||
5858 | cpus_and(nodemask, nodemask, *cpu_map); | |
5859 | group = first_cpu(nodemask); | |
5860 | ||
5861 | if (sg) | |
5862 | *sg = &per_cpu(sched_group_allnodes, group); | |
5863 | return group; | |
1da177e4 | 5864 | } |
6711cab4 | 5865 | |
08069033 SS |
5866 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5867 | { | |
5868 | struct sched_group *sg = group_head; | |
5869 | int j; | |
5870 | ||
5871 | if (!sg) | |
5872 | return; | |
5873 | next_sg: | |
5874 | for_each_cpu_mask(j, sg->cpumask) { | |
5875 | struct sched_domain *sd; | |
5876 | ||
5877 | sd = &per_cpu(phys_domains, j); | |
5878 | if (j != first_cpu(sd->groups->cpumask)) { | |
5879 | /* | |
5880 | * Only add "power" once for each | |
5881 | * physical package. | |
5882 | */ | |
5883 | continue; | |
5884 | } | |
5885 | ||
5517d86b | 5886 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5887 | } |
5888 | sg = sg->next; | |
5889 | if (sg != group_head) | |
5890 | goto next_sg; | |
5891 | } | |
1da177e4 LT |
5892 | #endif |
5893 | ||
a616058b | 5894 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5895 | /* Free memory allocated for various sched_group structures */ |
5896 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5897 | { | |
a616058b | 5898 | int cpu, i; |
51888ca2 SV |
5899 | |
5900 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5901 | struct sched_group **sched_group_nodes |
5902 | = sched_group_nodes_bycpu[cpu]; | |
5903 | ||
51888ca2 SV |
5904 | if (!sched_group_nodes) |
5905 | continue; | |
5906 | ||
5907 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5908 | cpumask_t nodemask = node_to_cpumask(i); | |
5909 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5910 | ||
5911 | cpus_and(nodemask, nodemask, *cpu_map); | |
5912 | if (cpus_empty(nodemask)) | |
5913 | continue; | |
5914 | ||
5915 | if (sg == NULL) | |
5916 | continue; | |
5917 | sg = sg->next; | |
5918 | next_sg: | |
5919 | oldsg = sg; | |
5920 | sg = sg->next; | |
5921 | kfree(oldsg); | |
5922 | if (oldsg != sched_group_nodes[i]) | |
5923 | goto next_sg; | |
5924 | } | |
5925 | kfree(sched_group_nodes); | |
5926 | sched_group_nodes_bycpu[cpu] = NULL; | |
5927 | } | |
51888ca2 | 5928 | } |
a616058b SS |
5929 | #else |
5930 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5931 | { | |
5932 | } | |
5933 | #endif | |
51888ca2 | 5934 | |
89c4710e SS |
5935 | /* |
5936 | * Initialize sched groups cpu_power. | |
5937 | * | |
5938 | * cpu_power indicates the capacity of sched group, which is used while | |
5939 | * distributing the load between different sched groups in a sched domain. | |
5940 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5941 | * there are asymmetries in the topology. If there are asymmetries, group | |
5942 | * having more cpu_power will pickup more load compared to the group having | |
5943 | * less cpu_power. | |
5944 | * | |
5945 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5946 | * the maximum number of tasks a group can handle in the presence of other idle | |
5947 | * or lightly loaded groups in the same sched domain. | |
5948 | */ | |
5949 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5950 | { | |
5951 | struct sched_domain *child; | |
5952 | struct sched_group *group; | |
5953 | ||
5954 | WARN_ON(!sd || !sd->groups); | |
5955 | ||
5956 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5957 | return; | |
5958 | ||
5959 | child = sd->child; | |
5960 | ||
5517d86b ED |
5961 | sd->groups->__cpu_power = 0; |
5962 | ||
89c4710e SS |
5963 | /* |
5964 | * For perf policy, if the groups in child domain share resources | |
5965 | * (for example cores sharing some portions of the cache hierarchy | |
5966 | * or SMT), then set this domain groups cpu_power such that each group | |
5967 | * can handle only one task, when there are other idle groups in the | |
5968 | * same sched domain. | |
5969 | */ | |
5970 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5971 | (child->flags & | |
5972 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5973 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5974 | return; |
5975 | } | |
5976 | ||
89c4710e SS |
5977 | /* |
5978 | * add cpu_power of each child group to this groups cpu_power | |
5979 | */ | |
5980 | group = child->groups; | |
5981 | do { | |
5517d86b | 5982 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5983 | group = group->next; |
5984 | } while (group != child->groups); | |
5985 | } | |
5986 | ||
1da177e4 | 5987 | /* |
1a20ff27 DG |
5988 | * Build sched domains for a given set of cpus and attach the sched domains |
5989 | * to the individual cpus | |
1da177e4 | 5990 | */ |
51888ca2 | 5991 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5992 | { |
5993 | int i; | |
d1b55138 JH |
5994 | #ifdef CONFIG_NUMA |
5995 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 5996 | int sd_allnodes = 0; |
d1b55138 JH |
5997 | |
5998 | /* | |
5999 | * Allocate the per-node list of sched groups | |
6000 | */ | |
dd41f596 | 6001 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 6002 | GFP_KERNEL); |
d1b55138 JH |
6003 | if (!sched_group_nodes) { |
6004 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6005 | return -ENOMEM; |
d1b55138 JH |
6006 | } |
6007 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6008 | #endif | |
1da177e4 LT |
6009 | |
6010 | /* | |
1a20ff27 | 6011 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6012 | */ |
1a20ff27 | 6013 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6014 | struct sched_domain *sd = NULL, *p; |
6015 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6016 | ||
1a20ff27 | 6017 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6018 | |
6019 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6020 | if (cpus_weight(*cpu_map) > |
6021 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6022 | sd = &per_cpu(allnodes_domains, i); |
6023 | *sd = SD_ALLNODES_INIT; | |
6024 | sd->span = *cpu_map; | |
6711cab4 | 6025 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6026 | p = sd; |
6711cab4 | 6027 | sd_allnodes = 1; |
9c1cfda2 JH |
6028 | } else |
6029 | p = NULL; | |
6030 | ||
1da177e4 | 6031 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6032 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6033 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6034 | sd->parent = p; | |
1a848870 SS |
6035 | if (p) |
6036 | p->child = sd; | |
9c1cfda2 | 6037 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6038 | #endif |
6039 | ||
6040 | p = sd; | |
6041 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6042 | *sd = SD_CPU_INIT; |
6043 | sd->span = nodemask; | |
6044 | sd->parent = p; | |
1a848870 SS |
6045 | if (p) |
6046 | p->child = sd; | |
6711cab4 | 6047 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6048 | |
1e9f28fa SS |
6049 | #ifdef CONFIG_SCHED_MC |
6050 | p = sd; | |
6051 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6052 | *sd = SD_MC_INIT; |
6053 | sd->span = cpu_coregroup_map(i); | |
6054 | cpus_and(sd->span, sd->span, *cpu_map); | |
6055 | sd->parent = p; | |
1a848870 | 6056 | p->child = sd; |
6711cab4 | 6057 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6058 | #endif |
6059 | ||
1da177e4 LT |
6060 | #ifdef CONFIG_SCHED_SMT |
6061 | p = sd; | |
6062 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6063 | *sd = SD_SIBLING_INIT; |
6064 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6065 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6066 | sd->parent = p; |
1a848870 | 6067 | p->child = sd; |
6711cab4 | 6068 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6069 | #endif |
6070 | } | |
6071 | ||
6072 | #ifdef CONFIG_SCHED_SMT | |
6073 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6074 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6075 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6076 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6077 | if (i != first_cpu(this_sibling_map)) |
6078 | continue; | |
6079 | ||
dd41f596 IM |
6080 | init_sched_build_groups(this_sibling_map, cpu_map, |
6081 | &cpu_to_cpu_group); | |
1da177e4 LT |
6082 | } |
6083 | #endif | |
6084 | ||
1e9f28fa SS |
6085 | #ifdef CONFIG_SCHED_MC |
6086 | /* Set up multi-core groups */ | |
6087 | for_each_cpu_mask(i, *cpu_map) { | |
6088 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6089 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6090 | if (i != first_cpu(this_core_map)) | |
6091 | continue; | |
dd41f596 IM |
6092 | init_sched_build_groups(this_core_map, cpu_map, |
6093 | &cpu_to_core_group); | |
1e9f28fa SS |
6094 | } |
6095 | #endif | |
6096 | ||
1da177e4 LT |
6097 | /* Set up physical groups */ |
6098 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6099 | cpumask_t nodemask = node_to_cpumask(i); | |
6100 | ||
1a20ff27 | 6101 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6102 | if (cpus_empty(nodemask)) |
6103 | continue; | |
6104 | ||
6711cab4 | 6105 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6106 | } |
6107 | ||
6108 | #ifdef CONFIG_NUMA | |
6109 | /* Set up node groups */ | |
6711cab4 | 6110 | if (sd_allnodes) |
dd41f596 IM |
6111 | init_sched_build_groups(*cpu_map, cpu_map, |
6112 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6113 | |
6114 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6115 | /* Set up node groups */ | |
6116 | struct sched_group *sg, *prev; | |
6117 | cpumask_t nodemask = node_to_cpumask(i); | |
6118 | cpumask_t domainspan; | |
6119 | cpumask_t covered = CPU_MASK_NONE; | |
6120 | int j; | |
6121 | ||
6122 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6123 | if (cpus_empty(nodemask)) { |
6124 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6125 | continue; |
d1b55138 | 6126 | } |
9c1cfda2 JH |
6127 | |
6128 | domainspan = sched_domain_node_span(i); | |
6129 | cpus_and(domainspan, domainspan, *cpu_map); | |
6130 | ||
15f0b676 | 6131 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6132 | if (!sg) { |
6133 | printk(KERN_WARNING "Can not alloc domain group for " | |
6134 | "node %d\n", i); | |
6135 | goto error; | |
6136 | } | |
9c1cfda2 JH |
6137 | sched_group_nodes[i] = sg; |
6138 | for_each_cpu_mask(j, nodemask) { | |
6139 | struct sched_domain *sd; | |
9761eea8 | 6140 | |
9c1cfda2 JH |
6141 | sd = &per_cpu(node_domains, j); |
6142 | sd->groups = sg; | |
9c1cfda2 | 6143 | } |
5517d86b | 6144 | sg->__cpu_power = 0; |
9c1cfda2 | 6145 | sg->cpumask = nodemask; |
51888ca2 | 6146 | sg->next = sg; |
9c1cfda2 JH |
6147 | cpus_or(covered, covered, nodemask); |
6148 | prev = sg; | |
6149 | ||
6150 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6151 | cpumask_t tmp, notcovered; | |
6152 | int n = (i + j) % MAX_NUMNODES; | |
6153 | ||
6154 | cpus_complement(notcovered, covered); | |
6155 | cpus_and(tmp, notcovered, *cpu_map); | |
6156 | cpus_and(tmp, tmp, domainspan); | |
6157 | if (cpus_empty(tmp)) | |
6158 | break; | |
6159 | ||
6160 | nodemask = node_to_cpumask(n); | |
6161 | cpus_and(tmp, tmp, nodemask); | |
6162 | if (cpus_empty(tmp)) | |
6163 | continue; | |
6164 | ||
15f0b676 SV |
6165 | sg = kmalloc_node(sizeof(struct sched_group), |
6166 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6167 | if (!sg) { |
6168 | printk(KERN_WARNING | |
6169 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6170 | goto error; |
9c1cfda2 | 6171 | } |
5517d86b | 6172 | sg->__cpu_power = 0; |
9c1cfda2 | 6173 | sg->cpumask = tmp; |
51888ca2 | 6174 | sg->next = prev->next; |
9c1cfda2 JH |
6175 | cpus_or(covered, covered, tmp); |
6176 | prev->next = sg; | |
6177 | prev = sg; | |
6178 | } | |
9c1cfda2 | 6179 | } |
1da177e4 LT |
6180 | #endif |
6181 | ||
6182 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6183 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6184 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6185 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6186 | ||
89c4710e | 6187 | init_sched_groups_power(i, sd); |
5c45bf27 | 6188 | } |
1da177e4 | 6189 | #endif |
1e9f28fa | 6190 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6191 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6192 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6193 | ||
89c4710e | 6194 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6195 | } |
6196 | #endif | |
1e9f28fa | 6197 | |
5c45bf27 | 6198 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6199 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6200 | ||
89c4710e | 6201 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6202 | } |
6203 | ||
9c1cfda2 | 6204 | #ifdef CONFIG_NUMA |
08069033 SS |
6205 | for (i = 0; i < MAX_NUMNODES; i++) |
6206 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6207 | |
6711cab4 SS |
6208 | if (sd_allnodes) { |
6209 | struct sched_group *sg; | |
f712c0c7 | 6210 | |
6711cab4 | 6211 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6212 | init_numa_sched_groups_power(sg); |
6213 | } | |
9c1cfda2 JH |
6214 | #endif |
6215 | ||
1da177e4 | 6216 | /* Attach the domains */ |
1a20ff27 | 6217 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6218 | struct sched_domain *sd; |
6219 | #ifdef CONFIG_SCHED_SMT | |
6220 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6221 | #elif defined(CONFIG_SCHED_MC) |
6222 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6223 | #else |
6224 | sd = &per_cpu(phys_domains, i); | |
6225 | #endif | |
6226 | cpu_attach_domain(sd, i); | |
6227 | } | |
51888ca2 SV |
6228 | |
6229 | return 0; | |
6230 | ||
a616058b | 6231 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6232 | error: |
6233 | free_sched_groups(cpu_map); | |
6234 | return -ENOMEM; | |
a616058b | 6235 | #endif |
1da177e4 | 6236 | } |
1a20ff27 DG |
6237 | /* |
6238 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6239 | */ | |
51888ca2 | 6240 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6241 | { |
6242 | cpumask_t cpu_default_map; | |
51888ca2 | 6243 | int err; |
1da177e4 | 6244 | |
1a20ff27 DG |
6245 | /* |
6246 | * Setup mask for cpus without special case scheduling requirements. | |
6247 | * For now this just excludes isolated cpus, but could be used to | |
6248 | * exclude other special cases in the future. | |
6249 | */ | |
6250 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6251 | ||
51888ca2 SV |
6252 | err = build_sched_domains(&cpu_default_map); |
6253 | ||
6254 | return err; | |
1a20ff27 DG |
6255 | } |
6256 | ||
6257 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6258 | { |
51888ca2 | 6259 | free_sched_groups(cpu_map); |
9c1cfda2 | 6260 | } |
1da177e4 | 6261 | |
1a20ff27 DG |
6262 | /* |
6263 | * Detach sched domains from a group of cpus specified in cpu_map | |
6264 | * These cpus will now be attached to the NULL domain | |
6265 | */ | |
858119e1 | 6266 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6267 | { |
6268 | int i; | |
6269 | ||
6270 | for_each_cpu_mask(i, *cpu_map) | |
6271 | cpu_attach_domain(NULL, i); | |
6272 | synchronize_sched(); | |
6273 | arch_destroy_sched_domains(cpu_map); | |
6274 | } | |
6275 | ||
6276 | /* | |
6277 | * Partition sched domains as specified by the cpumasks below. | |
6278 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6279 | * waits for a RCU quiescent period, recalculates sched | |
6280 | * domain information and then attaches them back to the | |
6281 | * correct sched domains | |
6282 | * Call with hotplug lock held | |
6283 | */ | |
51888ca2 | 6284 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6285 | { |
6286 | cpumask_t change_map; | |
51888ca2 | 6287 | int err = 0; |
1a20ff27 DG |
6288 | |
6289 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6290 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6291 | cpus_or(change_map, *partition1, *partition2); | |
6292 | ||
6293 | /* Detach sched domains from all of the affected cpus */ | |
6294 | detach_destroy_domains(&change_map); | |
6295 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6296 | err = build_sched_domains(partition1); |
6297 | if (!err && !cpus_empty(*partition2)) | |
6298 | err = build_sched_domains(partition2); | |
6299 | ||
6300 | return err; | |
1a20ff27 DG |
6301 | } |
6302 | ||
5c45bf27 SS |
6303 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6304 | int arch_reinit_sched_domains(void) | |
6305 | { | |
6306 | int err; | |
6307 | ||
5be9361c | 6308 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6309 | detach_destroy_domains(&cpu_online_map); |
6310 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6311 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6312 | |
6313 | return err; | |
6314 | } | |
6315 | ||
6316 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6317 | { | |
6318 | int ret; | |
6319 | ||
6320 | if (buf[0] != '0' && buf[0] != '1') | |
6321 | return -EINVAL; | |
6322 | ||
6323 | if (smt) | |
6324 | sched_smt_power_savings = (buf[0] == '1'); | |
6325 | else | |
6326 | sched_mc_power_savings = (buf[0] == '1'); | |
6327 | ||
6328 | ret = arch_reinit_sched_domains(); | |
6329 | ||
6330 | return ret ? ret : count; | |
6331 | } | |
6332 | ||
6333 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6334 | { | |
6335 | int err = 0; | |
48f24c4d | 6336 | |
5c45bf27 SS |
6337 | #ifdef CONFIG_SCHED_SMT |
6338 | if (smt_capable()) | |
6339 | err = sysfs_create_file(&cls->kset.kobj, | |
6340 | &attr_sched_smt_power_savings.attr); | |
6341 | #endif | |
6342 | #ifdef CONFIG_SCHED_MC | |
6343 | if (!err && mc_capable()) | |
6344 | err = sysfs_create_file(&cls->kset.kobj, | |
6345 | &attr_sched_mc_power_savings.attr); | |
6346 | #endif | |
6347 | return err; | |
6348 | } | |
6349 | #endif | |
6350 | ||
6351 | #ifdef CONFIG_SCHED_MC | |
6352 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6353 | { | |
6354 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6355 | } | |
48f24c4d IM |
6356 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6357 | const char *buf, size_t count) | |
5c45bf27 SS |
6358 | { |
6359 | return sched_power_savings_store(buf, count, 0); | |
6360 | } | |
6361 | SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, | |
6362 | sched_mc_power_savings_store); | |
6363 | #endif | |
6364 | ||
6365 | #ifdef CONFIG_SCHED_SMT | |
6366 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6367 | { | |
6368 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6369 | } | |
48f24c4d IM |
6370 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6371 | const char *buf, size_t count) | |
5c45bf27 SS |
6372 | { |
6373 | return sched_power_savings_store(buf, count, 1); | |
6374 | } | |
6375 | SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, | |
6376 | sched_smt_power_savings_store); | |
6377 | #endif | |
6378 | ||
1da177e4 LT |
6379 | /* |
6380 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6381 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6382 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6383 | * which will prevent rebalancing while the sched domains are recalculated. |
6384 | */ | |
6385 | static int update_sched_domains(struct notifier_block *nfb, | |
6386 | unsigned long action, void *hcpu) | |
6387 | { | |
1da177e4 LT |
6388 | switch (action) { |
6389 | case CPU_UP_PREPARE: | |
8bb78442 | 6390 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6391 | case CPU_DOWN_PREPARE: |
8bb78442 | 6392 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6393 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6394 | return NOTIFY_OK; |
6395 | ||
6396 | case CPU_UP_CANCELED: | |
8bb78442 | 6397 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6398 | case CPU_DOWN_FAILED: |
8bb78442 | 6399 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6400 | case CPU_ONLINE: |
8bb78442 | 6401 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6402 | case CPU_DEAD: |
8bb78442 | 6403 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6404 | /* |
6405 | * Fall through and re-initialise the domains. | |
6406 | */ | |
6407 | break; | |
6408 | default: | |
6409 | return NOTIFY_DONE; | |
6410 | } | |
6411 | ||
6412 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6413 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6414 | |
6415 | return NOTIFY_OK; | |
6416 | } | |
1da177e4 LT |
6417 | |
6418 | void __init sched_init_smp(void) | |
6419 | { | |
5c1e1767 NP |
6420 | cpumask_t non_isolated_cpus; |
6421 | ||
5be9361c | 6422 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6423 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6424 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6425 | if (cpus_empty(non_isolated_cpus)) |
6426 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6427 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6428 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6429 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 | 6430 | |
e692ab53 NP |
6431 | init_sched_domain_sysctl(); |
6432 | ||
5c1e1767 NP |
6433 | /* Move init over to a non-isolated CPU */ |
6434 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6435 | BUG(); | |
dd41f596 | 6436 | sched_init_granularity(); |
1da177e4 LT |
6437 | } |
6438 | #else | |
6439 | void __init sched_init_smp(void) | |
6440 | { | |
dd41f596 | 6441 | sched_init_granularity(); |
1da177e4 LT |
6442 | } |
6443 | #endif /* CONFIG_SMP */ | |
6444 | ||
6445 | int in_sched_functions(unsigned long addr) | |
6446 | { | |
6447 | /* Linker adds these: start and end of __sched functions */ | |
6448 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6449 | |
1da177e4 LT |
6450 | return in_lock_functions(addr) || |
6451 | (addr >= (unsigned long)__sched_text_start | |
6452 | && addr < (unsigned long)__sched_text_end); | |
6453 | } | |
6454 | ||
dd41f596 IM |
6455 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
6456 | { | |
6457 | cfs_rq->tasks_timeline = RB_ROOT; | |
6458 | cfs_rq->fair_clock = 1; | |
6459 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6460 | cfs_rq->rq = rq; | |
6461 | #endif | |
6462 | } | |
6463 | ||
1da177e4 LT |
6464 | void __init sched_init(void) |
6465 | { | |
dd41f596 | 6466 | u64 now = sched_clock(); |
476f3534 | 6467 | int highest_cpu = 0; |
dd41f596 IM |
6468 | int i, j; |
6469 | ||
6470 | /* | |
6471 | * Link up the scheduling class hierarchy: | |
6472 | */ | |
6473 | rt_sched_class.next = &fair_sched_class; | |
6474 | fair_sched_class.next = &idle_sched_class; | |
6475 | idle_sched_class.next = NULL; | |
1da177e4 | 6476 | |
0a945022 | 6477 | for_each_possible_cpu(i) { |
dd41f596 | 6478 | struct rt_prio_array *array; |
70b97a7f | 6479 | struct rq *rq; |
1da177e4 LT |
6480 | |
6481 | rq = cpu_rq(i); | |
6482 | spin_lock_init(&rq->lock); | |
fcb99371 | 6483 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6484 | rq->nr_running = 0; |
dd41f596 IM |
6485 | rq->clock = 1; |
6486 | init_cfs_rq(&rq->cfs, rq); | |
6487 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6488 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
6489 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6490 | #endif | |
6491 | rq->ls.load_update_last = now; | |
6492 | rq->ls.load_update_start = now; | |
1da177e4 | 6493 | |
dd41f596 IM |
6494 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6495 | rq->cpu_load[j] = 0; | |
1da177e4 | 6496 | #ifdef CONFIG_SMP |
41c7ce9a | 6497 | rq->sd = NULL; |
1da177e4 | 6498 | rq->active_balance = 0; |
dd41f596 | 6499 | rq->next_balance = jiffies; |
1da177e4 | 6500 | rq->push_cpu = 0; |
0a2966b4 | 6501 | rq->cpu = i; |
1da177e4 LT |
6502 | rq->migration_thread = NULL; |
6503 | INIT_LIST_HEAD(&rq->migration_queue); | |
6504 | #endif | |
6505 | atomic_set(&rq->nr_iowait, 0); | |
6506 | ||
dd41f596 IM |
6507 | array = &rq->rt.active; |
6508 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6509 | INIT_LIST_HEAD(array->queue + j); | |
6510 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6511 | } |
476f3534 | 6512 | highest_cpu = i; |
dd41f596 IM |
6513 | /* delimiter for bitsearch: */ |
6514 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6515 | } |
6516 | ||
2dd73a4f | 6517 | set_load_weight(&init_task); |
b50f60ce | 6518 | |
e107be36 AK |
6519 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6520 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6521 | #endif | |
6522 | ||
c9819f45 | 6523 | #ifdef CONFIG_SMP |
476f3534 | 6524 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6525 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6526 | #endif | |
6527 | ||
b50f60ce HC |
6528 | #ifdef CONFIG_RT_MUTEXES |
6529 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6530 | #endif | |
6531 | ||
1da177e4 LT |
6532 | /* |
6533 | * The boot idle thread does lazy MMU switching as well: | |
6534 | */ | |
6535 | atomic_inc(&init_mm.mm_count); | |
6536 | enter_lazy_tlb(&init_mm, current); | |
6537 | ||
6538 | /* | |
6539 | * Make us the idle thread. Technically, schedule() should not be | |
6540 | * called from this thread, however somewhere below it might be, | |
6541 | * but because we are the idle thread, we just pick up running again | |
6542 | * when this runqueue becomes "idle". | |
6543 | */ | |
6544 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6545 | /* |
6546 | * During early bootup we pretend to be a normal task: | |
6547 | */ | |
6548 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6549 | } |
6550 | ||
6551 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6552 | void __might_sleep(char *file, int line) | |
6553 | { | |
48f24c4d | 6554 | #ifdef in_atomic |
1da177e4 LT |
6555 | static unsigned long prev_jiffy; /* ratelimiting */ |
6556 | ||
6557 | if ((in_atomic() || irqs_disabled()) && | |
6558 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6559 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6560 | return; | |
6561 | prev_jiffy = jiffies; | |
91368d73 | 6562 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6563 | " context at %s:%d\n", file, line); |
6564 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6565 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6566 | debug_show_held_locks(current); |
3117df04 IM |
6567 | if (irqs_disabled()) |
6568 | print_irqtrace_events(current); | |
1da177e4 LT |
6569 | dump_stack(); |
6570 | } | |
6571 | #endif | |
6572 | } | |
6573 | EXPORT_SYMBOL(__might_sleep); | |
6574 | #endif | |
6575 | ||
6576 | #ifdef CONFIG_MAGIC_SYSRQ | |
6577 | void normalize_rt_tasks(void) | |
6578 | { | |
a0f98a1c | 6579 | struct task_struct *g, *p; |
1da177e4 | 6580 | unsigned long flags; |
70b97a7f | 6581 | struct rq *rq; |
dd41f596 | 6582 | int on_rq; |
1da177e4 LT |
6583 | |
6584 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6585 | do_each_thread(g, p) { |
dd41f596 IM |
6586 | p->se.fair_key = 0; |
6587 | p->se.wait_runtime = 0; | |
6cfb0d5d | 6588 | p->se.exec_start = 0; |
dd41f596 | 6589 | p->se.wait_start_fair = 0; |
6cfb0d5d IM |
6590 | p->se.sleep_start_fair = 0; |
6591 | #ifdef CONFIG_SCHEDSTATS | |
dd41f596 | 6592 | p->se.wait_start = 0; |
dd41f596 | 6593 | p->se.sleep_start = 0; |
dd41f596 | 6594 | p->se.block_start = 0; |
6cfb0d5d | 6595 | #endif |
dd41f596 IM |
6596 | task_rq(p)->cfs.fair_clock = 0; |
6597 | task_rq(p)->clock = 0; | |
6598 | ||
6599 | if (!rt_task(p)) { | |
6600 | /* | |
6601 | * Renice negative nice level userspace | |
6602 | * tasks back to 0: | |
6603 | */ | |
6604 | if (TASK_NICE(p) < 0 && p->mm) | |
6605 | set_user_nice(p, 0); | |
1da177e4 | 6606 | continue; |
dd41f596 | 6607 | } |
1da177e4 | 6608 | |
b29739f9 IM |
6609 | spin_lock_irqsave(&p->pi_lock, flags); |
6610 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6611 | #ifdef CONFIG_SMP |
6612 | /* | |
6613 | * Do not touch the migration thread: | |
6614 | */ | |
6615 | if (p == rq->migration_thread) | |
6616 | goto out_unlock; | |
6617 | #endif | |
1da177e4 | 6618 | |
dd41f596 IM |
6619 | on_rq = p->se.on_rq; |
6620 | if (on_rq) | |
6621 | deactivate_task(task_rq(p), p, 0); | |
6622 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6623 | if (on_rq) { | |
6624 | activate_task(task_rq(p), p, 0); | |
1da177e4 LT |
6625 | resched_task(rq->curr); |
6626 | } | |
dd41f596 IM |
6627 | #ifdef CONFIG_SMP |
6628 | out_unlock: | |
6629 | #endif | |
b29739f9 IM |
6630 | __task_rq_unlock(rq); |
6631 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6632 | } while_each_thread(g, p); |
6633 | ||
1da177e4 LT |
6634 | read_unlock_irq(&tasklist_lock); |
6635 | } | |
6636 | ||
6637 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6638 | |
6639 | #ifdef CONFIG_IA64 | |
6640 | /* | |
6641 | * These functions are only useful for the IA64 MCA handling. | |
6642 | * | |
6643 | * They can only be called when the whole system has been | |
6644 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6645 | * activity can take place. Using them for anything else would | |
6646 | * be a serious bug, and as a result, they aren't even visible | |
6647 | * under any other configuration. | |
6648 | */ | |
6649 | ||
6650 | /** | |
6651 | * curr_task - return the current task for a given cpu. | |
6652 | * @cpu: the processor in question. | |
6653 | * | |
6654 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6655 | */ | |
36c8b586 | 6656 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6657 | { |
6658 | return cpu_curr(cpu); | |
6659 | } | |
6660 | ||
6661 | /** | |
6662 | * set_curr_task - set the current task for a given cpu. | |
6663 | * @cpu: the processor in question. | |
6664 | * @p: the task pointer to set. | |
6665 | * | |
6666 | * Description: This function must only be used when non-maskable interrupts | |
6667 | * are serviced on a separate stack. It allows the architecture to switch the | |
6668 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6669 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6670 | * and caller must save the original value of the current task (see | |
6671 | * curr_task() above) and restore that value before reenabling interrupts and | |
6672 | * re-starting the system. | |
6673 | * | |
6674 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6675 | */ | |
36c8b586 | 6676 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6677 | { |
6678 | cpu_curr(cpu) = p; | |
6679 | } | |
6680 | ||
6681 | #endif |