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> |
f5ff8422 | 64 | #include <linux/pagemap.h> |
1da177e4 | 65 | |
5517d86b | 66 | #include <asm/tlb.h> |
1da177e4 | 67 | |
b035b6de AD |
68 | /* |
69 | * Scheduler clock - returns current time in nanosec units. | |
70 | * This is default implementation. | |
71 | * Architectures and sub-architectures can override this. | |
72 | */ | |
73 | unsigned long long __attribute__((weak)) sched_clock(void) | |
74 | { | |
75 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
76 | } | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
80 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
81 | * and back. | |
82 | */ | |
83 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
84 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
85 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
86 | ||
87 | /* | |
88 | * 'User priority' is the nice value converted to something we | |
89 | * can work with better when scaling various scheduler parameters, | |
90 | * it's a [ 0 ... 39 ] range. | |
91 | */ | |
92 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
93 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
94 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
95 | ||
96 | /* | |
97 | * Some helpers for converting nanosecond timing to jiffy resolution | |
98 | */ | |
99 | #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) | |
100 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) | |
101 | ||
6aa645ea IM |
102 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
103 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
104 | ||
1da177e4 LT |
105 | /* |
106 | * These are the 'tuning knobs' of the scheduler: | |
107 | * | |
108 | * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger), | |
109 | * default timeslice is 100 msecs, maximum timeslice is 800 msecs. | |
110 | * Timeslices get refilled after they expire. | |
111 | */ | |
112 | #define MIN_TIMESLICE max(5 * HZ / 1000, 1) | |
113 | #define DEF_TIMESLICE (100 * HZ / 1000) | |
2dd73a4f | 114 | |
5517d86b ED |
115 | #ifdef CONFIG_SMP |
116 | /* | |
117 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
118 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
119 | */ | |
120 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
121 | { | |
122 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Each time a sched group cpu_power is changed, | |
127 | * we must compute its reciprocal value | |
128 | */ | |
129 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
130 | { | |
131 | sg->__cpu_power += val; | |
132 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
133 | } | |
134 | #endif | |
135 | ||
634fa8c9 IM |
136 | #define SCALE_PRIO(x, prio) \ |
137 | max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE) | |
138 | ||
91fcdd4e | 139 | /* |
634fa8c9 | 140 | * static_prio_timeslice() scales user-nice values [ -20 ... 0 ... 19 ] |
91fcdd4e | 141 | * to time slice values: [800ms ... 100ms ... 5ms] |
91fcdd4e | 142 | */ |
634fa8c9 | 143 | static unsigned int static_prio_timeslice(int static_prio) |
2dd73a4f | 144 | { |
634fa8c9 IM |
145 | if (static_prio == NICE_TO_PRIO(19)) |
146 | return 1; | |
147 | ||
148 | if (static_prio < NICE_TO_PRIO(0)) | |
149 | return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio); | |
150 | else | |
151 | return SCALE_PRIO(DEF_TIMESLICE, static_prio); | |
2dd73a4f PW |
152 | } |
153 | ||
e05606d3 IM |
154 | static inline int rt_policy(int policy) |
155 | { | |
156 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
157 | return 1; | |
158 | return 0; | |
159 | } | |
160 | ||
161 | static inline int task_has_rt_policy(struct task_struct *p) | |
162 | { | |
163 | return rt_policy(p->policy); | |
164 | } | |
165 | ||
1da177e4 | 166 | /* |
6aa645ea | 167 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 168 | */ |
6aa645ea IM |
169 | struct rt_prio_array { |
170 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
171 | struct list_head queue[MAX_RT_PRIO]; | |
172 | }; | |
173 | ||
174 | struct load_stat { | |
175 | struct load_weight load; | |
6aa645ea IM |
176 | }; |
177 | ||
178 | /* CFS-related fields in a runqueue */ | |
179 | struct cfs_rq { | |
180 | struct load_weight load; | |
181 | unsigned long nr_running; | |
182 | ||
183 | s64 fair_clock; | |
184 | u64 exec_clock; | |
e9acbff6 | 185 | u64 min_vruntime; |
6aa645ea IM |
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; | |
6aa645ea IM |
193 | /* 'curr' points to currently running entity on this cfs_rq. |
194 | * It is set to NULL otherwise (i.e when none are currently running). | |
195 | */ | |
196 | struct sched_entity *curr; | |
62160e3f | 197 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
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; | |
2aa44d05 IM |
265 | u64 idle_clock; |
266 | unsigned int clock_deep_idle_events; | |
529c7726 | 267 | u64 tick_timestamp; |
6aa645ea | 268 | |
1da177e4 LT |
269 | atomic_t nr_iowait; |
270 | ||
271 | #ifdef CONFIG_SMP | |
272 | struct sched_domain *sd; | |
273 | ||
274 | /* For active balancing */ | |
275 | int active_balance; | |
276 | int push_cpu; | |
0a2966b4 | 277 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 278 | |
36c8b586 | 279 | struct task_struct *migration_thread; |
1da177e4 LT |
280 | struct list_head migration_queue; |
281 | #endif | |
282 | ||
283 | #ifdef CONFIG_SCHEDSTATS | |
284 | /* latency stats */ | |
285 | struct sched_info rq_sched_info; | |
286 | ||
287 | /* sys_sched_yield() stats */ | |
288 | unsigned long yld_exp_empty; | |
289 | unsigned long yld_act_empty; | |
290 | unsigned long yld_both_empty; | |
291 | unsigned long yld_cnt; | |
292 | ||
293 | /* schedule() stats */ | |
294 | unsigned long sched_switch; | |
295 | unsigned long sched_cnt; | |
296 | unsigned long sched_goidle; | |
297 | ||
298 | /* try_to_wake_up() stats */ | |
299 | unsigned long ttwu_cnt; | |
300 | unsigned long ttwu_local; | |
301 | #endif | |
fcb99371 | 302 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
303 | }; |
304 | ||
f34e3b61 | 305 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 306 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 307 | |
dd41f596 IM |
308 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
309 | { | |
310 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
311 | } | |
312 | ||
0a2966b4 CL |
313 | static inline int cpu_of(struct rq *rq) |
314 | { | |
315 | #ifdef CONFIG_SMP | |
316 | return rq->cpu; | |
317 | #else | |
318 | return 0; | |
319 | #endif | |
320 | } | |
321 | ||
20d315d4 | 322 | /* |
b04a0f4c IM |
323 | * Update the per-runqueue clock, as finegrained as the platform can give |
324 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 325 | */ |
b04a0f4c | 326 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
327 | { |
328 | u64 prev_raw = rq->prev_clock_raw; | |
329 | u64 now = sched_clock(); | |
330 | s64 delta = now - prev_raw; | |
331 | u64 clock = rq->clock; | |
332 | ||
b04a0f4c IM |
333 | #ifdef CONFIG_SCHED_DEBUG |
334 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
335 | #endif | |
20d315d4 IM |
336 | /* |
337 | * Protect against sched_clock() occasionally going backwards: | |
338 | */ | |
339 | if (unlikely(delta < 0)) { | |
340 | clock++; | |
341 | rq->clock_warps++; | |
342 | } else { | |
343 | /* | |
344 | * Catch too large forward jumps too: | |
345 | */ | |
529c7726 IM |
346 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
347 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
348 | clock = rq->tick_timestamp + TICK_NSEC; | |
349 | else | |
350 | clock++; | |
20d315d4 IM |
351 | rq->clock_overflows++; |
352 | } else { | |
353 | if (unlikely(delta > rq->clock_max_delta)) | |
354 | rq->clock_max_delta = delta; | |
355 | clock += delta; | |
356 | } | |
357 | } | |
358 | ||
359 | rq->prev_clock_raw = now; | |
360 | rq->clock = clock; | |
b04a0f4c | 361 | } |
20d315d4 | 362 | |
b04a0f4c IM |
363 | static void update_rq_clock(struct rq *rq) |
364 | { | |
365 | if (likely(smp_processor_id() == cpu_of(rq))) | |
366 | __update_rq_clock(rq); | |
20d315d4 IM |
367 | } |
368 | ||
674311d5 NP |
369 | /* |
370 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 371 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
372 | * |
373 | * The domain tree of any CPU may only be accessed from within | |
374 | * preempt-disabled sections. | |
375 | */ | |
48f24c4d IM |
376 | #define for_each_domain(cpu, __sd) \ |
377 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
378 | |
379 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
380 | #define this_rq() (&__get_cpu_var(runqueues)) | |
381 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
382 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
383 | ||
bf5c91ba IM |
384 | /* |
385 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
386 | */ | |
387 | #ifdef CONFIG_SCHED_DEBUG | |
388 | # define const_debug __read_mostly | |
389 | #else | |
390 | # define const_debug static const | |
391 | #endif | |
392 | ||
393 | /* | |
394 | * Debugging: various feature bits | |
395 | */ | |
396 | enum { | |
397 | SCHED_FEAT_FAIR_SLEEPERS = 1, | |
398 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 2, | |
399 | SCHED_FEAT_SLEEPER_AVG = 4, | |
400 | SCHED_FEAT_SLEEPER_LOAD_AVG = 8, | |
401 | SCHED_FEAT_START_DEBIT = 16, | |
bf5c91ba IM |
402 | }; |
403 | ||
404 | const_debug unsigned int sysctl_sched_features = | |
405 | SCHED_FEAT_FAIR_SLEEPERS *0 | | |
406 | SCHED_FEAT_NEW_FAIR_SLEEPERS *1 | | |
407 | SCHED_FEAT_SLEEPER_AVG *0 | | |
408 | SCHED_FEAT_SLEEPER_LOAD_AVG *1 | | |
28a1f6fa | 409 | SCHED_FEAT_START_DEBIT *1; |
bf5c91ba IM |
410 | |
411 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
412 | ||
e436d800 IM |
413 | /* |
414 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
415 | * clock constructed from sched_clock(): | |
416 | */ | |
417 | unsigned long long cpu_clock(int cpu) | |
418 | { | |
e436d800 IM |
419 | unsigned long long now; |
420 | unsigned long flags; | |
b04a0f4c | 421 | struct rq *rq; |
e436d800 | 422 | |
2cd4d0ea | 423 | local_irq_save(flags); |
b04a0f4c IM |
424 | rq = cpu_rq(cpu); |
425 | update_rq_clock(rq); | |
426 | now = rq->clock; | |
2cd4d0ea | 427 | local_irq_restore(flags); |
e436d800 IM |
428 | |
429 | return now; | |
430 | } | |
431 | ||
138a8aeb IM |
432 | #ifdef CONFIG_FAIR_GROUP_SCHED |
433 | /* Change a task's ->cfs_rq if it moves across CPUs */ | |
434 | static inline void set_task_cfs_rq(struct task_struct *p) | |
435 | { | |
436 | p->se.cfs_rq = &task_rq(p)->cfs; | |
437 | } | |
438 | #else | |
439 | static inline void set_task_cfs_rq(struct task_struct *p) | |
440 | { | |
441 | } | |
442 | #endif | |
443 | ||
1da177e4 | 444 | #ifndef prepare_arch_switch |
4866cde0 NP |
445 | # define prepare_arch_switch(next) do { } while (0) |
446 | #endif | |
447 | #ifndef finish_arch_switch | |
448 | # define finish_arch_switch(prev) do { } while (0) | |
449 | #endif | |
450 | ||
451 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 452 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
453 | { |
454 | return rq->curr == p; | |
455 | } | |
456 | ||
70b97a7f | 457 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
458 | { |
459 | } | |
460 | ||
70b97a7f | 461 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 462 | { |
da04c035 IM |
463 | #ifdef CONFIG_DEBUG_SPINLOCK |
464 | /* this is a valid case when another task releases the spinlock */ | |
465 | rq->lock.owner = current; | |
466 | #endif | |
8a25d5de IM |
467 | /* |
468 | * If we are tracking spinlock dependencies then we have to | |
469 | * fix up the runqueue lock - which gets 'carried over' from | |
470 | * prev into current: | |
471 | */ | |
472 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
473 | ||
4866cde0 NP |
474 | spin_unlock_irq(&rq->lock); |
475 | } | |
476 | ||
477 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 478 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
479 | { |
480 | #ifdef CONFIG_SMP | |
481 | return p->oncpu; | |
482 | #else | |
483 | return rq->curr == p; | |
484 | #endif | |
485 | } | |
486 | ||
70b97a7f | 487 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
488 | { |
489 | #ifdef CONFIG_SMP | |
490 | /* | |
491 | * We can optimise this out completely for !SMP, because the | |
492 | * SMP rebalancing from interrupt is the only thing that cares | |
493 | * here. | |
494 | */ | |
495 | next->oncpu = 1; | |
496 | #endif | |
497 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
498 | spin_unlock_irq(&rq->lock); | |
499 | #else | |
500 | spin_unlock(&rq->lock); | |
501 | #endif | |
502 | } | |
503 | ||
70b97a7f | 504 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
505 | { |
506 | #ifdef CONFIG_SMP | |
507 | /* | |
508 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
509 | * We must ensure this doesn't happen until the switch is completely | |
510 | * finished. | |
511 | */ | |
512 | smp_wmb(); | |
513 | prev->oncpu = 0; | |
514 | #endif | |
515 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
516 | local_irq_enable(); | |
1da177e4 | 517 | #endif |
4866cde0 NP |
518 | } |
519 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 520 | |
b29739f9 IM |
521 | /* |
522 | * __task_rq_lock - lock the runqueue a given task resides on. | |
523 | * Must be called interrupts disabled. | |
524 | */ | |
70b97a7f | 525 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
526 | __acquires(rq->lock) |
527 | { | |
70b97a7f | 528 | struct rq *rq; |
b29739f9 IM |
529 | |
530 | repeat_lock_task: | |
531 | rq = task_rq(p); | |
532 | spin_lock(&rq->lock); | |
533 | if (unlikely(rq != task_rq(p))) { | |
534 | spin_unlock(&rq->lock); | |
535 | goto repeat_lock_task; | |
536 | } | |
537 | return rq; | |
538 | } | |
539 | ||
1da177e4 LT |
540 | /* |
541 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
542 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
543 | * explicitly disabling preemption. | |
544 | */ | |
70b97a7f | 545 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
546 | __acquires(rq->lock) |
547 | { | |
70b97a7f | 548 | struct rq *rq; |
1da177e4 LT |
549 | |
550 | repeat_lock_task: | |
551 | local_irq_save(*flags); | |
552 | rq = task_rq(p); | |
553 | spin_lock(&rq->lock); | |
554 | if (unlikely(rq != task_rq(p))) { | |
555 | spin_unlock_irqrestore(&rq->lock, *flags); | |
556 | goto repeat_lock_task; | |
557 | } | |
558 | return rq; | |
559 | } | |
560 | ||
70b97a7f | 561 | static inline void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
562 | __releases(rq->lock) |
563 | { | |
564 | spin_unlock(&rq->lock); | |
565 | } | |
566 | ||
70b97a7f | 567 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
568 | __releases(rq->lock) |
569 | { | |
570 | spin_unlock_irqrestore(&rq->lock, *flags); | |
571 | } | |
572 | ||
1da177e4 | 573 | /* |
cc2a73b5 | 574 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 575 | */ |
70b97a7f | 576 | static inline struct rq *this_rq_lock(void) |
1da177e4 LT |
577 | __acquires(rq->lock) |
578 | { | |
70b97a7f | 579 | struct rq *rq; |
1da177e4 LT |
580 | |
581 | local_irq_disable(); | |
582 | rq = this_rq(); | |
583 | spin_lock(&rq->lock); | |
584 | ||
585 | return rq; | |
586 | } | |
587 | ||
1b9f19c2 | 588 | /* |
2aa44d05 | 589 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 590 | */ |
2aa44d05 | 591 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 592 | { |
2aa44d05 IM |
593 | struct rq *rq = cpu_rq(smp_processor_id()); |
594 | ||
595 | spin_lock(&rq->lock); | |
596 | __update_rq_clock(rq); | |
597 | spin_unlock(&rq->lock); | |
598 | rq->clock_deep_idle_events++; | |
599 | } | |
600 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
601 | ||
602 | /* | |
603 | * We just idled delta nanoseconds (called with irqs disabled): | |
604 | */ | |
605 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
606 | { | |
607 | struct rq *rq = cpu_rq(smp_processor_id()); | |
608 | u64 now = sched_clock(); | |
1b9f19c2 | 609 | |
2aa44d05 IM |
610 | rq->idle_clock += delta_ns; |
611 | /* | |
612 | * Override the previous timestamp and ignore all | |
613 | * sched_clock() deltas that occured while we idled, | |
614 | * and use the PM-provided delta_ns to advance the | |
615 | * rq clock: | |
616 | */ | |
617 | spin_lock(&rq->lock); | |
618 | rq->prev_clock_raw = now; | |
619 | rq->clock += delta_ns; | |
620 | spin_unlock(&rq->lock); | |
1b9f19c2 | 621 | } |
2aa44d05 | 622 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 623 | |
c24d20db IM |
624 | /* |
625 | * resched_task - mark a task 'to be rescheduled now'. | |
626 | * | |
627 | * On UP this means the setting of the need_resched flag, on SMP it | |
628 | * might also involve a cross-CPU call to trigger the scheduler on | |
629 | * the target CPU. | |
630 | */ | |
631 | #ifdef CONFIG_SMP | |
632 | ||
633 | #ifndef tsk_is_polling | |
634 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
635 | #endif | |
636 | ||
637 | static void resched_task(struct task_struct *p) | |
638 | { | |
639 | int cpu; | |
640 | ||
641 | assert_spin_locked(&task_rq(p)->lock); | |
642 | ||
643 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
644 | return; | |
645 | ||
646 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
647 | ||
648 | cpu = task_cpu(p); | |
649 | if (cpu == smp_processor_id()) | |
650 | return; | |
651 | ||
652 | /* NEED_RESCHED must be visible before we test polling */ | |
653 | smp_mb(); | |
654 | if (!tsk_is_polling(p)) | |
655 | smp_send_reschedule(cpu); | |
656 | } | |
657 | ||
658 | static void resched_cpu(int cpu) | |
659 | { | |
660 | struct rq *rq = cpu_rq(cpu); | |
661 | unsigned long flags; | |
662 | ||
663 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
664 | return; | |
665 | resched_task(cpu_curr(cpu)); | |
666 | spin_unlock_irqrestore(&rq->lock, flags); | |
667 | } | |
668 | #else | |
669 | static inline void resched_task(struct task_struct *p) | |
670 | { | |
671 | assert_spin_locked(&task_rq(p)->lock); | |
672 | set_tsk_need_resched(p); | |
673 | } | |
674 | #endif | |
675 | ||
45bf76df IM |
676 | static u64 div64_likely32(u64 divident, unsigned long divisor) |
677 | { | |
678 | #if BITS_PER_LONG == 32 | |
679 | if (likely(divident <= 0xffffffffULL)) | |
680 | return (u32)divident / divisor; | |
681 | do_div(divident, divisor); | |
682 | ||
683 | return divident; | |
684 | #else | |
685 | return divident / divisor; | |
686 | #endif | |
687 | } | |
688 | ||
689 | #if BITS_PER_LONG == 32 | |
690 | # define WMULT_CONST (~0UL) | |
691 | #else | |
692 | # define WMULT_CONST (1UL << 32) | |
693 | #endif | |
694 | ||
695 | #define WMULT_SHIFT 32 | |
696 | ||
194081eb IM |
697 | /* |
698 | * Shift right and round: | |
699 | */ | |
cf2ab469 | 700 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 701 | |
cb1c4fc9 | 702 | static unsigned long |
45bf76df IM |
703 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
704 | struct load_weight *lw) | |
705 | { | |
706 | u64 tmp; | |
707 | ||
708 | if (unlikely(!lw->inv_weight)) | |
194081eb | 709 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
710 | |
711 | tmp = (u64)delta_exec * weight; | |
712 | /* | |
713 | * Check whether we'd overflow the 64-bit multiplication: | |
714 | */ | |
194081eb | 715 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 716 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
717 | WMULT_SHIFT/2); |
718 | else | |
cf2ab469 | 719 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 720 | |
ecf691da | 721 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
722 | } |
723 | ||
724 | static inline unsigned long | |
725 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
726 | { | |
727 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
728 | } | |
729 | ||
1091985b | 730 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
731 | { |
732 | lw->weight += inc; | |
6cb58195 IM |
733 | if (sched_feat(FAIR_SLEEPERS)) |
734 | lw->inv_weight = WMULT_CONST / lw->weight; | |
45bf76df IM |
735 | } |
736 | ||
1091985b | 737 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
738 | { |
739 | lw->weight -= dec; | |
6cb58195 | 740 | if (sched_feat(FAIR_SLEEPERS) && likely(lw->weight)) |
1091985b | 741 | lw->inv_weight = WMULT_CONST / lw->weight; |
45bf76df IM |
742 | } |
743 | ||
2dd73a4f PW |
744 | /* |
745 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
746 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
747 | * each task makes to its run queue's load is weighted according to its | |
748 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
749 | * scaled version of the new time slice allocation that they receive on time | |
750 | * slice expiry etc. | |
751 | */ | |
752 | ||
dd41f596 IM |
753 | #define WEIGHT_IDLEPRIO 2 |
754 | #define WMULT_IDLEPRIO (1 << 31) | |
755 | ||
756 | /* | |
757 | * Nice levels are multiplicative, with a gentle 10% change for every | |
758 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
759 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
760 | * that remained on nice 0. | |
761 | * | |
762 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
763 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
764 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
765 | * If a task goes up by ~10% and another task goes down by ~10% then | |
766 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
767 | */ |
768 | static const int prio_to_weight[40] = { | |
254753dc IM |
769 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
770 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
771 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
772 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
773 | /* 0 */ 1024, 820, 655, 526, 423, | |
774 | /* 5 */ 335, 272, 215, 172, 137, | |
775 | /* 10 */ 110, 87, 70, 56, 45, | |
776 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
777 | }; |
778 | ||
5714d2de IM |
779 | /* |
780 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
781 | * | |
782 | * In cases where the weight does not change often, we can use the | |
783 | * precalculated inverse to speed up arithmetics by turning divisions | |
784 | * into multiplications: | |
785 | */ | |
dd41f596 | 786 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
787 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
788 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
789 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
790 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
791 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
792 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
793 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
794 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 795 | }; |
2dd73a4f | 796 | |
dd41f596 IM |
797 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
798 | ||
799 | /* | |
800 | * runqueue iterator, to support SMP load-balancing between different | |
801 | * scheduling classes, without having to expose their internal data | |
802 | * structures to the load-balancing proper: | |
803 | */ | |
804 | struct rq_iterator { | |
805 | void *arg; | |
806 | struct task_struct *(*start)(void *); | |
807 | struct task_struct *(*next)(void *); | |
808 | }; | |
809 | ||
810 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
811 | unsigned long max_nr_move, unsigned long max_load_move, | |
812 | struct sched_domain *sd, enum cpu_idle_type idle, | |
813 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 814 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
815 | |
816 | #include "sched_stats.h" | |
817 | #include "sched_rt.c" | |
818 | #include "sched_fair.c" | |
819 | #include "sched_idletask.c" | |
820 | #ifdef CONFIG_SCHED_DEBUG | |
821 | # include "sched_debug.c" | |
822 | #endif | |
823 | ||
824 | #define sched_class_highest (&rt_sched_class) | |
825 | ||
9c217245 IM |
826 | /* |
827 | * Update delta_exec, delta_fair fields for rq. | |
828 | * | |
829 | * delta_fair clock advances at a rate inversely proportional to | |
830 | * total load (rq->ls.load.weight) on the runqueue, while | |
831 | * delta_exec advances at the same rate as wall-clock (provided | |
832 | * cpu is not idle). | |
833 | * | |
834 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
835 | * runqueue over any given interval. This (smoothened) load is used | |
836 | * during load balance. | |
837 | * | |
838 | * This function is called /before/ updating rq->ls.load | |
839 | * and when switching tasks. | |
840 | */ | |
29b4b623 | 841 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 842 | { |
9c217245 IM |
843 | update_load_add(&rq->ls.load, p->se.load.weight); |
844 | } | |
845 | ||
79b5dddf | 846 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 847 | { |
9c217245 IM |
848 | update_load_sub(&rq->ls.load, p->se.load.weight); |
849 | } | |
850 | ||
e5fa2237 | 851 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
852 | { |
853 | rq->nr_running++; | |
29b4b623 | 854 | inc_load(rq, p); |
9c217245 IM |
855 | } |
856 | ||
db53181e | 857 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
858 | { |
859 | rq->nr_running--; | |
79b5dddf | 860 | dec_load(rq, p); |
9c217245 IM |
861 | } |
862 | ||
45bf76df IM |
863 | static void set_load_weight(struct task_struct *p) |
864 | { | |
dd41f596 IM |
865 | p->se.wait_runtime = 0; |
866 | ||
45bf76df | 867 | if (task_has_rt_policy(p)) { |
dd41f596 IM |
868 | p->se.load.weight = prio_to_weight[0] * 2; |
869 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
870 | return; | |
871 | } | |
45bf76df | 872 | |
dd41f596 IM |
873 | /* |
874 | * SCHED_IDLE tasks get minimal weight: | |
875 | */ | |
876 | if (p->policy == SCHED_IDLE) { | |
877 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
878 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
879 | return; | |
880 | } | |
71f8bd46 | 881 | |
dd41f596 IM |
882 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
883 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
884 | } |
885 | ||
8159f87e | 886 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 887 | { |
dd41f596 | 888 | sched_info_queued(p); |
fd390f6a | 889 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 890 | p->se.on_rq = 1; |
71f8bd46 IM |
891 | } |
892 | ||
69be72c1 | 893 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 894 | { |
f02231e5 | 895 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 896 | p->se.on_rq = 0; |
71f8bd46 IM |
897 | } |
898 | ||
14531189 | 899 | /* |
dd41f596 | 900 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 901 | */ |
14531189 IM |
902 | static inline int __normal_prio(struct task_struct *p) |
903 | { | |
dd41f596 | 904 | return p->static_prio; |
14531189 IM |
905 | } |
906 | ||
b29739f9 IM |
907 | /* |
908 | * Calculate the expected normal priority: i.e. priority | |
909 | * without taking RT-inheritance into account. Might be | |
910 | * boosted by interactivity modifiers. Changes upon fork, | |
911 | * setprio syscalls, and whenever the interactivity | |
912 | * estimator recalculates. | |
913 | */ | |
36c8b586 | 914 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
915 | { |
916 | int prio; | |
917 | ||
e05606d3 | 918 | if (task_has_rt_policy(p)) |
b29739f9 IM |
919 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
920 | else | |
921 | prio = __normal_prio(p); | |
922 | return prio; | |
923 | } | |
924 | ||
925 | /* | |
926 | * Calculate the current priority, i.e. the priority | |
927 | * taken into account by the scheduler. This value might | |
928 | * be boosted by RT tasks, or might be boosted by | |
929 | * interactivity modifiers. Will be RT if the task got | |
930 | * RT-boosted. If not then it returns p->normal_prio. | |
931 | */ | |
36c8b586 | 932 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
933 | { |
934 | p->normal_prio = normal_prio(p); | |
935 | /* | |
936 | * If we are RT tasks or we were boosted to RT priority, | |
937 | * keep the priority unchanged. Otherwise, update priority | |
938 | * to the normal priority: | |
939 | */ | |
940 | if (!rt_prio(p->prio)) | |
941 | return p->normal_prio; | |
942 | return p->prio; | |
943 | } | |
944 | ||
1da177e4 | 945 | /* |
dd41f596 | 946 | * activate_task - move a task to the runqueue. |
1da177e4 | 947 | */ |
dd41f596 | 948 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 949 | { |
dd41f596 IM |
950 | if (p->state == TASK_UNINTERRUPTIBLE) |
951 | rq->nr_uninterruptible--; | |
1da177e4 | 952 | |
8159f87e | 953 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 954 | inc_nr_running(p, rq); |
1da177e4 LT |
955 | } |
956 | ||
957 | /* | |
dd41f596 | 958 | * activate_idle_task - move idle task to the _front_ of runqueue. |
1da177e4 | 959 | */ |
dd41f596 | 960 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 961 | { |
a8e504d2 | 962 | update_rq_clock(rq); |
1da177e4 | 963 | |
dd41f596 IM |
964 | if (p->state == TASK_UNINTERRUPTIBLE) |
965 | rq->nr_uninterruptible--; | |
ece8a684 | 966 | |
8159f87e | 967 | enqueue_task(rq, p, 0); |
e5fa2237 | 968 | inc_nr_running(p, rq); |
1da177e4 LT |
969 | } |
970 | ||
971 | /* | |
972 | * deactivate_task - remove a task from the runqueue. | |
973 | */ | |
2e1cb74a | 974 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 975 | { |
dd41f596 IM |
976 | if (p->state == TASK_UNINTERRUPTIBLE) |
977 | rq->nr_uninterruptible++; | |
978 | ||
69be72c1 | 979 | dequeue_task(rq, p, sleep); |
db53181e | 980 | dec_nr_running(p, rq); |
1da177e4 LT |
981 | } |
982 | ||
1da177e4 LT |
983 | /** |
984 | * task_curr - is this task currently executing on a CPU? | |
985 | * @p: the task in question. | |
986 | */ | |
36c8b586 | 987 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
988 | { |
989 | return cpu_curr(task_cpu(p)) == p; | |
990 | } | |
991 | ||
2dd73a4f PW |
992 | /* Used instead of source_load when we know the type == 0 */ |
993 | unsigned long weighted_cpuload(const int cpu) | |
994 | { | |
dd41f596 IM |
995 | return cpu_rq(cpu)->ls.load.weight; |
996 | } | |
997 | ||
998 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
999 | { | |
1000 | #ifdef CONFIG_SMP | |
1001 | task_thread_info(p)->cpu = cpu; | |
1002 | set_task_cfs_rq(p); | |
1003 | #endif | |
2dd73a4f PW |
1004 | } |
1005 | ||
1da177e4 | 1006 | #ifdef CONFIG_SMP |
c65cc870 | 1007 | |
dd41f596 | 1008 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1009 | { |
dd41f596 IM |
1010 | int old_cpu = task_cpu(p); |
1011 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
1012 | u64 clock_offset, fair_clock_offset; | |
1013 | ||
1014 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1015 | fair_clock_offset = old_rq->cfs.fair_clock - new_rq->cfs.fair_clock; |
1016 | ||
dd41f596 IM |
1017 | if (p->se.wait_start_fair) |
1018 | p->se.wait_start_fair -= fair_clock_offset; | |
6cfb0d5d IM |
1019 | if (p->se.sleep_start_fair) |
1020 | p->se.sleep_start_fair -= fair_clock_offset; | |
1021 | ||
1022 | #ifdef CONFIG_SCHEDSTATS | |
1023 | if (p->se.wait_start) | |
1024 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1025 | if (p->se.sleep_start) |
1026 | p->se.sleep_start -= clock_offset; | |
1027 | if (p->se.block_start) | |
1028 | p->se.block_start -= clock_offset; | |
6cfb0d5d | 1029 | #endif |
dd41f596 IM |
1030 | |
1031 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1032 | } |
1033 | ||
70b97a7f | 1034 | struct migration_req { |
1da177e4 | 1035 | struct list_head list; |
1da177e4 | 1036 | |
36c8b586 | 1037 | struct task_struct *task; |
1da177e4 LT |
1038 | int dest_cpu; |
1039 | ||
1da177e4 | 1040 | struct completion done; |
70b97a7f | 1041 | }; |
1da177e4 LT |
1042 | |
1043 | /* | |
1044 | * The task's runqueue lock must be held. | |
1045 | * Returns true if you have to wait for migration thread. | |
1046 | */ | |
36c8b586 | 1047 | static int |
70b97a7f | 1048 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1049 | { |
70b97a7f | 1050 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1051 | |
1052 | /* | |
1053 | * If the task is not on a runqueue (and not running), then | |
1054 | * it is sufficient to simply update the task's cpu field. | |
1055 | */ | |
dd41f596 | 1056 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1057 | set_task_cpu(p, dest_cpu); |
1058 | return 0; | |
1059 | } | |
1060 | ||
1061 | init_completion(&req->done); | |
1da177e4 LT |
1062 | req->task = p; |
1063 | req->dest_cpu = dest_cpu; | |
1064 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1065 | |
1da177e4 LT |
1066 | return 1; |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * wait_task_inactive - wait for a thread to unschedule. | |
1071 | * | |
1072 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1073 | * else this function might spin for a *long* time. This function can't | |
1074 | * be called with interrupts off, or it may introduce deadlock with | |
1075 | * smp_call_function() if an IPI is sent by the same process we are | |
1076 | * waiting to become inactive. | |
1077 | */ | |
36c8b586 | 1078 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1079 | { |
1080 | unsigned long flags; | |
dd41f596 | 1081 | int running, on_rq; |
70b97a7f | 1082 | struct rq *rq; |
1da177e4 LT |
1083 | |
1084 | repeat: | |
fa490cfd LT |
1085 | /* |
1086 | * We do the initial early heuristics without holding | |
1087 | * any task-queue locks at all. We'll only try to get | |
1088 | * the runqueue lock when things look like they will | |
1089 | * work out! | |
1090 | */ | |
1091 | rq = task_rq(p); | |
1092 | ||
1093 | /* | |
1094 | * If the task is actively running on another CPU | |
1095 | * still, just relax and busy-wait without holding | |
1096 | * any locks. | |
1097 | * | |
1098 | * NOTE! Since we don't hold any locks, it's not | |
1099 | * even sure that "rq" stays as the right runqueue! | |
1100 | * But we don't care, since "task_running()" will | |
1101 | * return false if the runqueue has changed and p | |
1102 | * is actually now running somewhere else! | |
1103 | */ | |
1104 | while (task_running(rq, p)) | |
1105 | cpu_relax(); | |
1106 | ||
1107 | /* | |
1108 | * Ok, time to look more closely! We need the rq | |
1109 | * lock now, to be *sure*. If we're wrong, we'll | |
1110 | * just go back and repeat. | |
1111 | */ | |
1da177e4 | 1112 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1113 | running = task_running(rq, p); |
dd41f596 | 1114 | on_rq = p->se.on_rq; |
fa490cfd LT |
1115 | task_rq_unlock(rq, &flags); |
1116 | ||
1117 | /* | |
1118 | * Was it really running after all now that we | |
1119 | * checked with the proper locks actually held? | |
1120 | * | |
1121 | * Oops. Go back and try again.. | |
1122 | */ | |
1123 | if (unlikely(running)) { | |
1da177e4 | 1124 | cpu_relax(); |
1da177e4 LT |
1125 | goto repeat; |
1126 | } | |
fa490cfd LT |
1127 | |
1128 | /* | |
1129 | * It's not enough that it's not actively running, | |
1130 | * it must be off the runqueue _entirely_, and not | |
1131 | * preempted! | |
1132 | * | |
1133 | * So if it wa still runnable (but just not actively | |
1134 | * running right now), it's preempted, and we should | |
1135 | * yield - it could be a while. | |
1136 | */ | |
dd41f596 | 1137 | if (unlikely(on_rq)) { |
fa490cfd LT |
1138 | yield(); |
1139 | goto repeat; | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * Ahh, all good. It wasn't running, and it wasn't | |
1144 | * runnable, which means that it will never become | |
1145 | * running in the future either. We're all done! | |
1146 | */ | |
1da177e4 LT |
1147 | } |
1148 | ||
1149 | /*** | |
1150 | * kick_process - kick a running thread to enter/exit the kernel | |
1151 | * @p: the to-be-kicked thread | |
1152 | * | |
1153 | * Cause a process which is running on another CPU to enter | |
1154 | * kernel-mode, without any delay. (to get signals handled.) | |
1155 | * | |
1156 | * NOTE: this function doesnt have to take the runqueue lock, | |
1157 | * because all it wants to ensure is that the remote task enters | |
1158 | * the kernel. If the IPI races and the task has been migrated | |
1159 | * to another CPU then no harm is done and the purpose has been | |
1160 | * achieved as well. | |
1161 | */ | |
36c8b586 | 1162 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1163 | { |
1164 | int cpu; | |
1165 | ||
1166 | preempt_disable(); | |
1167 | cpu = task_cpu(p); | |
1168 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1169 | smp_send_reschedule(cpu); | |
1170 | preempt_enable(); | |
1171 | } | |
1172 | ||
1173 | /* | |
2dd73a4f PW |
1174 | * Return a low guess at the load of a migration-source cpu weighted |
1175 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1176 | * |
1177 | * We want to under-estimate the load of migration sources, to | |
1178 | * balance conservatively. | |
1179 | */ | |
a2000572 | 1180 | static inline unsigned long source_load(int cpu, int type) |
1da177e4 | 1181 | { |
70b97a7f | 1182 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1183 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1184 | |
3b0bd9bc | 1185 | if (type == 0) |
dd41f596 | 1186 | return total; |
b910472d | 1187 | |
dd41f596 | 1188 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1189 | } |
1190 | ||
1191 | /* | |
2dd73a4f PW |
1192 | * Return a high guess at the load of a migration-target cpu weighted |
1193 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1194 | */ |
a2000572 | 1195 | static inline unsigned long target_load(int cpu, int type) |
1da177e4 | 1196 | { |
70b97a7f | 1197 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1198 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1199 | |
7897986b | 1200 | if (type == 0) |
dd41f596 | 1201 | return total; |
3b0bd9bc | 1202 | |
dd41f596 | 1203 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1204 | } |
1205 | ||
1206 | /* | |
1207 | * Return the average load per task on the cpu's run queue | |
1208 | */ | |
1209 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1210 | { | |
70b97a7f | 1211 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1212 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1213 | unsigned long n = rq->nr_running; |
1214 | ||
dd41f596 | 1215 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1216 | } |
1217 | ||
147cbb4b NP |
1218 | /* |
1219 | * find_idlest_group finds and returns the least busy CPU group within the | |
1220 | * domain. | |
1221 | */ | |
1222 | static struct sched_group * | |
1223 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1224 | { | |
1225 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1226 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1227 | int load_idx = sd->forkexec_idx; | |
1228 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1229 | ||
1230 | do { | |
1231 | unsigned long load, avg_load; | |
1232 | int local_group; | |
1233 | int i; | |
1234 | ||
da5a5522 BD |
1235 | /* Skip over this group if it has no CPUs allowed */ |
1236 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1237 | goto nextgroup; | |
1238 | ||
147cbb4b | 1239 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1240 | |
1241 | /* Tally up the load of all CPUs in the group */ | |
1242 | avg_load = 0; | |
1243 | ||
1244 | for_each_cpu_mask(i, group->cpumask) { | |
1245 | /* Bias balancing toward cpus of our domain */ | |
1246 | if (local_group) | |
1247 | load = source_load(i, load_idx); | |
1248 | else | |
1249 | load = target_load(i, load_idx); | |
1250 | ||
1251 | avg_load += load; | |
1252 | } | |
1253 | ||
1254 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1255 | avg_load = sg_div_cpu_power(group, |
1256 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1257 | |
1258 | if (local_group) { | |
1259 | this_load = avg_load; | |
1260 | this = group; | |
1261 | } else if (avg_load < min_load) { | |
1262 | min_load = avg_load; | |
1263 | idlest = group; | |
1264 | } | |
da5a5522 | 1265 | nextgroup: |
147cbb4b NP |
1266 | group = group->next; |
1267 | } while (group != sd->groups); | |
1268 | ||
1269 | if (!idlest || 100*this_load < imbalance*min_load) | |
1270 | return NULL; | |
1271 | return idlest; | |
1272 | } | |
1273 | ||
1274 | /* | |
0feaece9 | 1275 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1276 | */ |
95cdf3b7 IM |
1277 | static int |
1278 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1279 | { |
da5a5522 | 1280 | cpumask_t tmp; |
147cbb4b NP |
1281 | unsigned long load, min_load = ULONG_MAX; |
1282 | int idlest = -1; | |
1283 | int i; | |
1284 | ||
da5a5522 BD |
1285 | /* Traverse only the allowed CPUs */ |
1286 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1287 | ||
1288 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1289 | load = weighted_cpuload(i); |
147cbb4b NP |
1290 | |
1291 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1292 | min_load = load; | |
1293 | idlest = i; | |
1294 | } | |
1295 | } | |
1296 | ||
1297 | return idlest; | |
1298 | } | |
1299 | ||
476d139c NP |
1300 | /* |
1301 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1302 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1303 | * SD_BALANCE_EXEC. | |
1304 | * | |
1305 | * Balance, ie. select the least loaded group. | |
1306 | * | |
1307 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1308 | * | |
1309 | * preempt must be disabled. | |
1310 | */ | |
1311 | static int sched_balance_self(int cpu, int flag) | |
1312 | { | |
1313 | struct task_struct *t = current; | |
1314 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1315 | |
c96d145e | 1316 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1317 | /* |
1318 | * If power savings logic is enabled for a domain, stop there. | |
1319 | */ | |
5c45bf27 SS |
1320 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1321 | break; | |
476d139c NP |
1322 | if (tmp->flags & flag) |
1323 | sd = tmp; | |
c96d145e | 1324 | } |
476d139c NP |
1325 | |
1326 | while (sd) { | |
1327 | cpumask_t span; | |
1328 | struct sched_group *group; | |
1a848870 SS |
1329 | int new_cpu, weight; |
1330 | ||
1331 | if (!(sd->flags & flag)) { | |
1332 | sd = sd->child; | |
1333 | continue; | |
1334 | } | |
476d139c NP |
1335 | |
1336 | span = sd->span; | |
1337 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1338 | if (!group) { |
1339 | sd = sd->child; | |
1340 | continue; | |
1341 | } | |
476d139c | 1342 | |
da5a5522 | 1343 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1344 | if (new_cpu == -1 || new_cpu == cpu) { |
1345 | /* Now try balancing at a lower domain level of cpu */ | |
1346 | sd = sd->child; | |
1347 | continue; | |
1348 | } | |
476d139c | 1349 | |
1a848870 | 1350 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1351 | cpu = new_cpu; |
476d139c NP |
1352 | sd = NULL; |
1353 | weight = cpus_weight(span); | |
1354 | for_each_domain(cpu, tmp) { | |
1355 | if (weight <= cpus_weight(tmp->span)) | |
1356 | break; | |
1357 | if (tmp->flags & flag) | |
1358 | sd = tmp; | |
1359 | } | |
1360 | /* while loop will break here if sd == NULL */ | |
1361 | } | |
1362 | ||
1363 | return cpu; | |
1364 | } | |
1365 | ||
1366 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1367 | |
1368 | /* | |
1369 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1370 | * not idle and an idle cpu is available. The span of cpus to | |
1371 | * search starts with cpus closest then further out as needed, | |
1372 | * so we always favor a closer, idle cpu. | |
1373 | * | |
1374 | * Returns the CPU we should wake onto. | |
1375 | */ | |
1376 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1377 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1378 | { |
1379 | cpumask_t tmp; | |
1380 | struct sched_domain *sd; | |
1381 | int i; | |
1382 | ||
4953198b SS |
1383 | /* |
1384 | * If it is idle, then it is the best cpu to run this task. | |
1385 | * | |
1386 | * This cpu is also the best, if it has more than one task already. | |
1387 | * Siblings must be also busy(in most cases) as they didn't already | |
1388 | * pickup the extra load from this cpu and hence we need not check | |
1389 | * sibling runqueue info. This will avoid the checks and cache miss | |
1390 | * penalities associated with that. | |
1391 | */ | |
1392 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1393 | return cpu; |
1394 | ||
1395 | for_each_domain(cpu, sd) { | |
1396 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1397 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1398 | for_each_cpu_mask(i, tmp) { |
1399 | if (idle_cpu(i)) | |
1400 | return i; | |
1401 | } | |
9761eea8 | 1402 | } else { |
e0f364f4 | 1403 | break; |
9761eea8 | 1404 | } |
1da177e4 LT |
1405 | } |
1406 | return cpu; | |
1407 | } | |
1408 | #else | |
36c8b586 | 1409 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1410 | { |
1411 | return cpu; | |
1412 | } | |
1413 | #endif | |
1414 | ||
1415 | /*** | |
1416 | * try_to_wake_up - wake up a thread | |
1417 | * @p: the to-be-woken-up thread | |
1418 | * @state: the mask of task states that can be woken | |
1419 | * @sync: do a synchronous wakeup? | |
1420 | * | |
1421 | * Put it on the run-queue if it's not already there. The "current" | |
1422 | * thread is always on the run-queue (except when the actual | |
1423 | * re-schedule is in progress), and as such you're allowed to do | |
1424 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1425 | * runnable without the overhead of this. | |
1426 | * | |
1427 | * returns failure only if the task is already active. | |
1428 | */ | |
36c8b586 | 1429 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1430 | { |
1431 | int cpu, this_cpu, success = 0; | |
1432 | unsigned long flags; | |
1433 | long old_state; | |
70b97a7f | 1434 | struct rq *rq; |
1da177e4 | 1435 | #ifdef CONFIG_SMP |
7897986b | 1436 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1437 | unsigned long load, this_load; |
1da177e4 LT |
1438 | int new_cpu; |
1439 | #endif | |
1440 | ||
1441 | rq = task_rq_lock(p, &flags); | |
1442 | old_state = p->state; | |
1443 | if (!(old_state & state)) | |
1444 | goto out; | |
1445 | ||
dd41f596 | 1446 | if (p->se.on_rq) |
1da177e4 LT |
1447 | goto out_running; |
1448 | ||
1449 | cpu = task_cpu(p); | |
1450 | this_cpu = smp_processor_id(); | |
1451 | ||
1452 | #ifdef CONFIG_SMP | |
1453 | if (unlikely(task_running(rq, p))) | |
1454 | goto out_activate; | |
1455 | ||
7897986b NP |
1456 | new_cpu = cpu; |
1457 | ||
1da177e4 LT |
1458 | schedstat_inc(rq, ttwu_cnt); |
1459 | if (cpu == this_cpu) { | |
1460 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1461 | goto out_set_cpu; |
1462 | } | |
1463 | ||
1464 | for_each_domain(this_cpu, sd) { | |
1465 | if (cpu_isset(cpu, sd->span)) { | |
1466 | schedstat_inc(sd, ttwu_wake_remote); | |
1467 | this_sd = sd; | |
1468 | break; | |
1da177e4 LT |
1469 | } |
1470 | } | |
1da177e4 | 1471 | |
7897986b | 1472 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1473 | goto out_set_cpu; |
1474 | ||
1da177e4 | 1475 | /* |
7897986b | 1476 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1477 | */ |
7897986b NP |
1478 | if (this_sd) { |
1479 | int idx = this_sd->wake_idx; | |
1480 | unsigned int imbalance; | |
1da177e4 | 1481 | |
a3f21bce NP |
1482 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1483 | ||
7897986b NP |
1484 | load = source_load(cpu, idx); |
1485 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1486 | |
7897986b NP |
1487 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1488 | ||
a3f21bce NP |
1489 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1490 | unsigned long tl = this_load; | |
33859f7f MOS |
1491 | unsigned long tl_per_task; |
1492 | ||
1493 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1494 | |
1da177e4 | 1495 | /* |
a3f21bce NP |
1496 | * If sync wakeup then subtract the (maximum possible) |
1497 | * effect of the currently running task from the load | |
1498 | * of the current CPU: | |
1da177e4 | 1499 | */ |
a3f21bce | 1500 | if (sync) |
dd41f596 | 1501 | tl -= current->se.load.weight; |
a3f21bce NP |
1502 | |
1503 | if ((tl <= load && | |
2dd73a4f | 1504 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1505 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1506 | /* |
1507 | * This domain has SD_WAKE_AFFINE and | |
1508 | * p is cache cold in this domain, and | |
1509 | * there is no bad imbalance. | |
1510 | */ | |
1511 | schedstat_inc(this_sd, ttwu_move_affine); | |
1512 | goto out_set_cpu; | |
1513 | } | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * Start passive balancing when half the imbalance_pct | |
1518 | * limit is reached. | |
1519 | */ | |
1520 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1521 | if (imbalance*this_load <= 100*load) { | |
1522 | schedstat_inc(this_sd, ttwu_move_balance); | |
1523 | goto out_set_cpu; | |
1524 | } | |
1da177e4 LT |
1525 | } |
1526 | } | |
1527 | ||
1528 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1529 | out_set_cpu: | |
1530 | new_cpu = wake_idle(new_cpu, p); | |
1531 | if (new_cpu != cpu) { | |
1532 | set_task_cpu(p, new_cpu); | |
1533 | task_rq_unlock(rq, &flags); | |
1534 | /* might preempt at this point */ | |
1535 | rq = task_rq_lock(p, &flags); | |
1536 | old_state = p->state; | |
1537 | if (!(old_state & state)) | |
1538 | goto out; | |
dd41f596 | 1539 | if (p->se.on_rq) |
1da177e4 LT |
1540 | goto out_running; |
1541 | ||
1542 | this_cpu = smp_processor_id(); | |
1543 | cpu = task_cpu(p); | |
1544 | } | |
1545 | ||
1546 | out_activate: | |
1547 | #endif /* CONFIG_SMP */ | |
2daa3577 | 1548 | update_rq_clock(rq); |
dd41f596 | 1549 | activate_task(rq, p, 1); |
1da177e4 LT |
1550 | /* |
1551 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1552 | * has indicated that it will leave the CPU in short order) | |
1553 | * don't trigger a preemption, if the woken up task will run on | |
1554 | * this cpu. (in this case the 'I will reschedule' promise of | |
1555 | * the waker guarantees that the freshly woken up task is going | |
1556 | * to be considered on this CPU.) | |
1557 | */ | |
dd41f596 IM |
1558 | if (!sync || cpu != this_cpu) |
1559 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1560 | success = 1; |
1561 | ||
1562 | out_running: | |
1563 | p->state = TASK_RUNNING; | |
1564 | out: | |
1565 | task_rq_unlock(rq, &flags); | |
1566 | ||
1567 | return success; | |
1568 | } | |
1569 | ||
36c8b586 | 1570 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1571 | { |
1572 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1573 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1574 | } | |
1da177e4 LT |
1575 | EXPORT_SYMBOL(wake_up_process); |
1576 | ||
36c8b586 | 1577 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1578 | { |
1579 | return try_to_wake_up(p, state, 0); | |
1580 | } | |
1581 | ||
1da177e4 LT |
1582 | /* |
1583 | * Perform scheduler related setup for a newly forked process p. | |
1584 | * p is forked by current. | |
dd41f596 IM |
1585 | * |
1586 | * __sched_fork() is basic setup used by init_idle() too: | |
1587 | */ | |
1588 | static void __sched_fork(struct task_struct *p) | |
1589 | { | |
1590 | p->se.wait_start_fair = 0; | |
dd41f596 IM |
1591 | p->se.exec_start = 0; |
1592 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1593 | p->se.prev_sum_exec_runtime = 0; |
dd41f596 | 1594 | p->se.wait_runtime = 0; |
6cfb0d5d IM |
1595 | p->se.sleep_start_fair = 0; |
1596 | ||
1597 | #ifdef CONFIG_SCHEDSTATS | |
1598 | p->se.wait_start = 0; | |
dd41f596 IM |
1599 | p->se.sum_wait_runtime = 0; |
1600 | p->se.sum_sleep_runtime = 0; | |
1601 | p->se.sleep_start = 0; | |
dd41f596 IM |
1602 | p->se.block_start = 0; |
1603 | p->se.sleep_max = 0; | |
1604 | p->se.block_max = 0; | |
1605 | p->se.exec_max = 0; | |
eba1ed4b | 1606 | p->se.slice_max = 0; |
dd41f596 IM |
1607 | p->se.wait_max = 0; |
1608 | p->se.wait_runtime_overruns = 0; | |
1609 | p->se.wait_runtime_underruns = 0; | |
6cfb0d5d | 1610 | #endif |
476d139c | 1611 | |
dd41f596 IM |
1612 | INIT_LIST_HEAD(&p->run_list); |
1613 | p->se.on_rq = 0; | |
476d139c | 1614 | |
e107be36 AK |
1615 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1616 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1617 | #endif | |
1618 | ||
1da177e4 LT |
1619 | /* |
1620 | * We mark the process as running here, but have not actually | |
1621 | * inserted it onto the runqueue yet. This guarantees that | |
1622 | * nobody will actually run it, and a signal or other external | |
1623 | * event cannot wake it up and insert it on the runqueue either. | |
1624 | */ | |
1625 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1626 | } |
1627 | ||
1628 | /* | |
1629 | * fork()/clone()-time setup: | |
1630 | */ | |
1631 | void sched_fork(struct task_struct *p, int clone_flags) | |
1632 | { | |
1633 | int cpu = get_cpu(); | |
1634 | ||
1635 | __sched_fork(p); | |
1636 | ||
1637 | #ifdef CONFIG_SMP | |
1638 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1639 | #endif | |
1640 | __set_task_cpu(p, cpu); | |
b29739f9 IM |
1641 | |
1642 | /* | |
1643 | * Make sure we do not leak PI boosting priority to the child: | |
1644 | */ | |
1645 | p->prio = current->normal_prio; | |
1646 | ||
52f17b6c | 1647 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1648 | if (likely(sched_info_on())) |
52f17b6c | 1649 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1650 | #endif |
d6077cb8 | 1651 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1652 | p->oncpu = 0; |
1653 | #endif | |
1da177e4 | 1654 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1655 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1656 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1657 | #endif |
476d139c | 1658 | put_cpu(); |
1da177e4 LT |
1659 | } |
1660 | ||
1661 | /* | |
1662 | * wake_up_new_task - wake up a newly created task for the first time. | |
1663 | * | |
1664 | * This function will do some initial scheduler statistics housekeeping | |
1665 | * that must be done for every newly created context, then puts the task | |
1666 | * on the runqueue and wakes it. | |
1667 | */ | |
36c8b586 | 1668 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1669 | { |
1670 | unsigned long flags; | |
dd41f596 IM |
1671 | struct rq *rq; |
1672 | int this_cpu; | |
1da177e4 LT |
1673 | |
1674 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1675 | BUG_ON(p->state != TASK_RUNNING); |
dd41f596 | 1676 | this_cpu = smp_processor_id(); /* parent's CPU */ |
a8e504d2 | 1677 | update_rq_clock(rq); |
1da177e4 LT |
1678 | |
1679 | p->prio = effective_prio(p); | |
1680 | ||
9c95e731 HS |
1681 | if (rt_prio(p->prio)) |
1682 | p->sched_class = &rt_sched_class; | |
1683 | else | |
1684 | p->sched_class = &fair_sched_class; | |
1685 | ||
44142fac IM |
1686 | if (task_cpu(p) != this_cpu || !p->sched_class->task_new || |
1687 | !current->se.on_rq) { | |
dd41f596 | 1688 | activate_task(rq, p, 0); |
1da177e4 | 1689 | } else { |
1da177e4 | 1690 | /* |
dd41f596 IM |
1691 | * Let the scheduling class do new task startup |
1692 | * management (if any): | |
1da177e4 | 1693 | */ |
ee0827d8 | 1694 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1695 | inc_nr_running(p, rq); |
1da177e4 | 1696 | } |
dd41f596 IM |
1697 | check_preempt_curr(rq, p); |
1698 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1699 | } |
1700 | ||
e107be36 AK |
1701 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1702 | ||
1703 | /** | |
421cee29 RD |
1704 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1705 | * @notifier: notifier struct to register | |
e107be36 AK |
1706 | */ |
1707 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1708 | { | |
1709 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1710 | } | |
1711 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1712 | ||
1713 | /** | |
1714 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1715 | * @notifier: notifier struct to unregister |
e107be36 AK |
1716 | * |
1717 | * This is safe to call from within a preemption notifier. | |
1718 | */ | |
1719 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1720 | { | |
1721 | hlist_del(¬ifier->link); | |
1722 | } | |
1723 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1724 | ||
1725 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1726 | { | |
1727 | struct preempt_notifier *notifier; | |
1728 | struct hlist_node *node; | |
1729 | ||
1730 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1731 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1732 | } | |
1733 | ||
1734 | static void | |
1735 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1736 | struct task_struct *next) | |
1737 | { | |
1738 | struct preempt_notifier *notifier; | |
1739 | struct hlist_node *node; | |
1740 | ||
1741 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1742 | notifier->ops->sched_out(notifier, next); | |
1743 | } | |
1744 | ||
1745 | #else | |
1746 | ||
1747 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1748 | { | |
1749 | } | |
1750 | ||
1751 | static void | |
1752 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1753 | struct task_struct *next) | |
1754 | { | |
1755 | } | |
1756 | ||
1757 | #endif | |
1758 | ||
4866cde0 NP |
1759 | /** |
1760 | * prepare_task_switch - prepare to switch tasks | |
1761 | * @rq: the runqueue preparing to switch | |
421cee29 | 1762 | * @prev: the current task that is being switched out |
4866cde0 NP |
1763 | * @next: the task we are going to switch to. |
1764 | * | |
1765 | * This is called with the rq lock held and interrupts off. It must | |
1766 | * be paired with a subsequent finish_task_switch after the context | |
1767 | * switch. | |
1768 | * | |
1769 | * prepare_task_switch sets up locking and calls architecture specific | |
1770 | * hooks. | |
1771 | */ | |
e107be36 AK |
1772 | static inline void |
1773 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1774 | struct task_struct *next) | |
4866cde0 | 1775 | { |
e107be36 | 1776 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1777 | prepare_lock_switch(rq, next); |
1778 | prepare_arch_switch(next); | |
1779 | } | |
1780 | ||
1da177e4 LT |
1781 | /** |
1782 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1783 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1784 | * @prev: the thread we just switched away from. |
1785 | * | |
4866cde0 NP |
1786 | * finish_task_switch must be called after the context switch, paired |
1787 | * with a prepare_task_switch call before the context switch. | |
1788 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1789 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1790 | * |
1791 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1792 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1793 | * with the lock held can cause deadlocks; see schedule() for | |
1794 | * details.) | |
1795 | */ | |
70b97a7f | 1796 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1797 | __releases(rq->lock) |
1798 | { | |
1da177e4 | 1799 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1800 | long prev_state; |
1da177e4 LT |
1801 | |
1802 | rq->prev_mm = NULL; | |
1803 | ||
1804 | /* | |
1805 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1806 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1807 | * schedule one last time. The schedule call will never return, and |
1808 | * the scheduled task must drop that reference. | |
c394cc9f | 1809 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1810 | * still held, otherwise prev could be scheduled on another cpu, die |
1811 | * there before we look at prev->state, and then the reference would | |
1812 | * be dropped twice. | |
1813 | * Manfred Spraul <manfred@colorfullife.com> | |
1814 | */ | |
55a101f8 | 1815 | prev_state = prev->state; |
4866cde0 NP |
1816 | finish_arch_switch(prev); |
1817 | finish_lock_switch(rq, prev); | |
e107be36 | 1818 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1819 | if (mm) |
1820 | mmdrop(mm); | |
c394cc9f | 1821 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1822 | /* |
1823 | * Remove function-return probe instances associated with this | |
1824 | * task and put them back on the free list. | |
9761eea8 | 1825 | */ |
c6fd91f0 | 1826 | kprobe_flush_task(prev); |
1da177e4 | 1827 | put_task_struct(prev); |
c6fd91f0 | 1828 | } |
1da177e4 LT |
1829 | } |
1830 | ||
1831 | /** | |
1832 | * schedule_tail - first thing a freshly forked thread must call. | |
1833 | * @prev: the thread we just switched away from. | |
1834 | */ | |
36c8b586 | 1835 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1836 | __releases(rq->lock) |
1837 | { | |
70b97a7f IM |
1838 | struct rq *rq = this_rq(); |
1839 | ||
4866cde0 NP |
1840 | finish_task_switch(rq, prev); |
1841 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1842 | /* In this case, finish_task_switch does not reenable preemption */ | |
1843 | preempt_enable(); | |
1844 | #endif | |
1da177e4 LT |
1845 | if (current->set_child_tid) |
1846 | put_user(current->pid, current->set_child_tid); | |
1847 | } | |
1848 | ||
1849 | /* | |
1850 | * context_switch - switch to the new MM and the new | |
1851 | * thread's register state. | |
1852 | */ | |
dd41f596 | 1853 | static inline void |
70b97a7f | 1854 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1855 | struct task_struct *next) |
1da177e4 | 1856 | { |
dd41f596 | 1857 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1858 | |
e107be36 | 1859 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1860 | mm = next->mm; |
1861 | oldmm = prev->active_mm; | |
9226d125 ZA |
1862 | /* |
1863 | * For paravirt, this is coupled with an exit in switch_to to | |
1864 | * combine the page table reload and the switch backend into | |
1865 | * one hypercall. | |
1866 | */ | |
1867 | arch_enter_lazy_cpu_mode(); | |
1868 | ||
dd41f596 | 1869 | if (unlikely(!mm)) { |
1da177e4 LT |
1870 | next->active_mm = oldmm; |
1871 | atomic_inc(&oldmm->mm_count); | |
1872 | enter_lazy_tlb(oldmm, next); | |
1873 | } else | |
1874 | switch_mm(oldmm, mm, next); | |
1875 | ||
dd41f596 | 1876 | if (unlikely(!prev->mm)) { |
1da177e4 | 1877 | prev->active_mm = NULL; |
1da177e4 LT |
1878 | rq->prev_mm = oldmm; |
1879 | } | |
3a5f5e48 IM |
1880 | /* |
1881 | * Since the runqueue lock will be released by the next | |
1882 | * task (which is an invalid locking op but in the case | |
1883 | * of the scheduler it's an obvious special-case), so we | |
1884 | * do an early lockdep release here: | |
1885 | */ | |
1886 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1887 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1888 | #endif |
1da177e4 LT |
1889 | |
1890 | /* Here we just switch the register state and the stack. */ | |
1891 | switch_to(prev, next, prev); | |
1892 | ||
dd41f596 IM |
1893 | barrier(); |
1894 | /* | |
1895 | * this_rq must be evaluated again because prev may have moved | |
1896 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1897 | * frame will be invalid. | |
1898 | */ | |
1899 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1900 | } |
1901 | ||
1902 | /* | |
1903 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1904 | * | |
1905 | * externally visible scheduler statistics: current number of runnable | |
1906 | * threads, current number of uninterruptible-sleeping threads, total | |
1907 | * number of context switches performed since bootup. | |
1908 | */ | |
1909 | unsigned long nr_running(void) | |
1910 | { | |
1911 | unsigned long i, sum = 0; | |
1912 | ||
1913 | for_each_online_cpu(i) | |
1914 | sum += cpu_rq(i)->nr_running; | |
1915 | ||
1916 | return sum; | |
1917 | } | |
1918 | ||
1919 | unsigned long nr_uninterruptible(void) | |
1920 | { | |
1921 | unsigned long i, sum = 0; | |
1922 | ||
0a945022 | 1923 | for_each_possible_cpu(i) |
1da177e4 LT |
1924 | sum += cpu_rq(i)->nr_uninterruptible; |
1925 | ||
1926 | /* | |
1927 | * Since we read the counters lockless, it might be slightly | |
1928 | * inaccurate. Do not allow it to go below zero though: | |
1929 | */ | |
1930 | if (unlikely((long)sum < 0)) | |
1931 | sum = 0; | |
1932 | ||
1933 | return sum; | |
1934 | } | |
1935 | ||
1936 | unsigned long long nr_context_switches(void) | |
1937 | { | |
cc94abfc SR |
1938 | int i; |
1939 | unsigned long long sum = 0; | |
1da177e4 | 1940 | |
0a945022 | 1941 | for_each_possible_cpu(i) |
1da177e4 LT |
1942 | sum += cpu_rq(i)->nr_switches; |
1943 | ||
1944 | return sum; | |
1945 | } | |
1946 | ||
1947 | unsigned long nr_iowait(void) | |
1948 | { | |
1949 | unsigned long i, sum = 0; | |
1950 | ||
0a945022 | 1951 | for_each_possible_cpu(i) |
1da177e4 LT |
1952 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1953 | ||
1954 | return sum; | |
1955 | } | |
1956 | ||
db1b1fef JS |
1957 | unsigned long nr_active(void) |
1958 | { | |
1959 | unsigned long i, running = 0, uninterruptible = 0; | |
1960 | ||
1961 | for_each_online_cpu(i) { | |
1962 | running += cpu_rq(i)->nr_running; | |
1963 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1964 | } | |
1965 | ||
1966 | if (unlikely((long)uninterruptible < 0)) | |
1967 | uninterruptible = 0; | |
1968 | ||
1969 | return running + uninterruptible; | |
1970 | } | |
1971 | ||
48f24c4d | 1972 | /* |
dd41f596 IM |
1973 | * Update rq->cpu_load[] statistics. This function is usually called every |
1974 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1975 | */ |
dd41f596 | 1976 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1977 | { |
53df556e | 1978 | unsigned long this_load = this_rq->ls.load.weight; |
dd41f596 IM |
1979 | int i, scale; |
1980 | ||
1981 | this_rq->nr_load_updates++; | |
dd41f596 IM |
1982 | |
1983 | /* Update our load: */ | |
1984 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1985 | unsigned long old_load, new_load; | |
1986 | ||
1987 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1988 | ||
1989 | old_load = this_rq->cpu_load[i]; | |
1990 | new_load = this_load; | |
a25707f3 IM |
1991 | /* |
1992 | * Round up the averaging division if load is increasing. This | |
1993 | * prevents us from getting stuck on 9 if the load is 10, for | |
1994 | * example. | |
1995 | */ | |
1996 | if (new_load > old_load) | |
1997 | new_load += scale-1; | |
dd41f596 IM |
1998 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
1999 | } | |
48f24c4d IM |
2000 | } |
2001 | ||
dd41f596 IM |
2002 | #ifdef CONFIG_SMP |
2003 | ||
1da177e4 LT |
2004 | /* |
2005 | * double_rq_lock - safely lock two runqueues | |
2006 | * | |
2007 | * Note this does not disable interrupts like task_rq_lock, | |
2008 | * you need to do so manually before calling. | |
2009 | */ | |
70b97a7f | 2010 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2011 | __acquires(rq1->lock) |
2012 | __acquires(rq2->lock) | |
2013 | { | |
054b9108 | 2014 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2015 | if (rq1 == rq2) { |
2016 | spin_lock(&rq1->lock); | |
2017 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2018 | } else { | |
c96d145e | 2019 | if (rq1 < rq2) { |
1da177e4 LT |
2020 | spin_lock(&rq1->lock); |
2021 | spin_lock(&rq2->lock); | |
2022 | } else { | |
2023 | spin_lock(&rq2->lock); | |
2024 | spin_lock(&rq1->lock); | |
2025 | } | |
2026 | } | |
6e82a3be IM |
2027 | update_rq_clock(rq1); |
2028 | update_rq_clock(rq2); | |
1da177e4 LT |
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 | { |
2e1cb74a | 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 | 2155 | |
1da177e4 LT |
2156 | return 1; |
2157 | } | |
2158 | ||
dd41f596 | 2159 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2160 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2161 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2162 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2163 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2164 | { |
dd41f596 IM |
2165 | int pulled = 0, pinned = 0, skip_for_load; |
2166 | struct task_struct *p; | |
2167 | long rem_load_move = max_load_move; | |
1da177e4 | 2168 | |
2dd73a4f | 2169 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2170 | goto out; |
2171 | ||
81026794 NP |
2172 | pinned = 1; |
2173 | ||
1da177e4 | 2174 | /* |
dd41f596 | 2175 | * Start the load-balancing iterator: |
1da177e4 | 2176 | */ |
dd41f596 IM |
2177 | p = iterator->start(iterator->arg); |
2178 | next: | |
2179 | if (!p) | |
1da177e4 | 2180 | goto out; |
50ddd969 PW |
2181 | /* |
2182 | * To help distribute high priority tasks accross CPUs we don't | |
2183 | * skip a task if it will be the highest priority task (i.e. smallest | |
2184 | * prio value) on its new queue regardless of its load weight | |
2185 | */ | |
dd41f596 IM |
2186 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2187 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2188 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2189 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2190 | p = iterator->next(iterator->arg); |
2191 | goto next; | |
1da177e4 LT |
2192 | } |
2193 | ||
dd41f596 | 2194 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2195 | pulled++; |
dd41f596 | 2196 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2197 | |
2dd73a4f PW |
2198 | /* |
2199 | * We only want to steal up to the prescribed number of tasks | |
2200 | * and the prescribed amount of weighted load. | |
2201 | */ | |
2202 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2203 | if (p->prio < *this_best_prio) |
2204 | *this_best_prio = p->prio; | |
dd41f596 IM |
2205 | p = iterator->next(iterator->arg); |
2206 | goto next; | |
1da177e4 LT |
2207 | } |
2208 | out: | |
2209 | /* | |
2210 | * Right now, this is the only place pull_task() is called, | |
2211 | * so we can safely collect pull_task() stats here rather than | |
2212 | * inside pull_task(). | |
2213 | */ | |
2214 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2215 | |
2216 | if (all_pinned) | |
2217 | *all_pinned = pinned; | |
dd41f596 | 2218 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2219 | return pulled; |
2220 | } | |
2221 | ||
dd41f596 | 2222 | /* |
43010659 PW |
2223 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2224 | * this_rq, as part of a balancing operation within domain "sd". | |
2225 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2226 | * |
2227 | * Called with both runqueues locked. | |
2228 | */ | |
2229 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2230 | unsigned long max_load_move, |
dd41f596 IM |
2231 | struct sched_domain *sd, enum cpu_idle_type idle, |
2232 | int *all_pinned) | |
2233 | { | |
2234 | struct sched_class *class = sched_class_highest; | |
43010659 | 2235 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2236 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2237 | |
2238 | do { | |
43010659 PW |
2239 | total_load_moved += |
2240 | class->load_balance(this_rq, this_cpu, busiest, | |
2241 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2242 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2243 | class = class->next; |
43010659 | 2244 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2245 | |
43010659 PW |
2246 | return total_load_moved > 0; |
2247 | } | |
2248 | ||
2249 | /* | |
2250 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2251 | * part of active balancing operations within "domain". | |
2252 | * Returns 1 if successful and 0 otherwise. | |
2253 | * | |
2254 | * Called with both runqueues locked. | |
2255 | */ | |
2256 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2257 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2258 | { | |
2259 | struct sched_class *class; | |
a4ac01c3 | 2260 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2261 | |
2262 | for (class = sched_class_highest; class; class = class->next) | |
2263 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2264 | 1, ULONG_MAX, sd, idle, NULL, |
2265 | &this_best_prio)) | |
43010659 PW |
2266 | return 1; |
2267 | ||
2268 | return 0; | |
dd41f596 IM |
2269 | } |
2270 | ||
1da177e4 LT |
2271 | /* |
2272 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2273 | * domain. It calculates and returns the amount of weighted load which |
2274 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2275 | */ |
2276 | static struct sched_group * | |
2277 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2278 | unsigned long *imbalance, enum cpu_idle_type idle, |
2279 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2280 | { |
2281 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2282 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2283 | unsigned long max_pull; |
2dd73a4f PW |
2284 | unsigned long busiest_load_per_task, busiest_nr_running; |
2285 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2286 | int load_idx; |
5c45bf27 SS |
2287 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2288 | int power_savings_balance = 1; | |
2289 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2290 | unsigned long min_nr_running = ULONG_MAX; | |
2291 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2292 | #endif | |
1da177e4 LT |
2293 | |
2294 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2295 | busiest_load_per_task = busiest_nr_running = 0; |
2296 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2297 | if (idle == CPU_NOT_IDLE) |
7897986b | 2298 | load_idx = sd->busy_idx; |
d15bcfdb | 2299 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2300 | load_idx = sd->newidle_idx; |
2301 | else | |
2302 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2303 | |
2304 | do { | |
5c45bf27 | 2305 | unsigned long load, group_capacity; |
1da177e4 LT |
2306 | int local_group; |
2307 | int i; | |
783609c6 | 2308 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2309 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2310 | |
2311 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2312 | ||
783609c6 SS |
2313 | if (local_group) |
2314 | balance_cpu = first_cpu(group->cpumask); | |
2315 | ||
1da177e4 | 2316 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2317 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2318 | |
2319 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2320 | struct rq *rq; |
2321 | ||
2322 | if (!cpu_isset(i, *cpus)) | |
2323 | continue; | |
2324 | ||
2325 | rq = cpu_rq(i); | |
2dd73a4f | 2326 | |
9439aab8 | 2327 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2328 | *sd_idle = 0; |
2329 | ||
1da177e4 | 2330 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2331 | if (local_group) { |
2332 | if (idle_cpu(i) && !first_idle_cpu) { | |
2333 | first_idle_cpu = 1; | |
2334 | balance_cpu = i; | |
2335 | } | |
2336 | ||
a2000572 | 2337 | load = target_load(i, load_idx); |
783609c6 | 2338 | } else |
a2000572 | 2339 | load = source_load(i, load_idx); |
1da177e4 LT |
2340 | |
2341 | avg_load += load; | |
2dd73a4f | 2342 | sum_nr_running += rq->nr_running; |
dd41f596 | 2343 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2344 | } |
2345 | ||
783609c6 SS |
2346 | /* |
2347 | * First idle cpu or the first cpu(busiest) in this sched group | |
2348 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2349 | * domains. In the newly idle case, we will allow all the cpu's |
2350 | * to do the newly idle load balance. | |
783609c6 | 2351 | */ |
9439aab8 SS |
2352 | if (idle != CPU_NEWLY_IDLE && local_group && |
2353 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2354 | *balance = 0; |
2355 | goto ret; | |
2356 | } | |
2357 | ||
1da177e4 | 2358 | total_load += avg_load; |
5517d86b | 2359 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2360 | |
2361 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2362 | avg_load = sg_div_cpu_power(group, |
2363 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2364 | |
5517d86b | 2365 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2366 | |
1da177e4 LT |
2367 | if (local_group) { |
2368 | this_load = avg_load; | |
2369 | this = group; | |
2dd73a4f PW |
2370 | this_nr_running = sum_nr_running; |
2371 | this_load_per_task = sum_weighted_load; | |
2372 | } else if (avg_load > max_load && | |
5c45bf27 | 2373 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2374 | max_load = avg_load; |
2375 | busiest = group; | |
2dd73a4f PW |
2376 | busiest_nr_running = sum_nr_running; |
2377 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2378 | } |
5c45bf27 SS |
2379 | |
2380 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2381 | /* | |
2382 | * Busy processors will not participate in power savings | |
2383 | * balance. | |
2384 | */ | |
dd41f596 IM |
2385 | if (idle == CPU_NOT_IDLE || |
2386 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2387 | goto group_next; | |
5c45bf27 SS |
2388 | |
2389 | /* | |
2390 | * If the local group is idle or completely loaded | |
2391 | * no need to do power savings balance at this domain | |
2392 | */ | |
2393 | if (local_group && (this_nr_running >= group_capacity || | |
2394 | !this_nr_running)) | |
2395 | power_savings_balance = 0; | |
2396 | ||
dd41f596 | 2397 | /* |
5c45bf27 SS |
2398 | * If a group is already running at full capacity or idle, |
2399 | * don't include that group in power savings calculations | |
dd41f596 IM |
2400 | */ |
2401 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2402 | || !sum_nr_running) |
dd41f596 | 2403 | goto group_next; |
5c45bf27 | 2404 | |
dd41f596 | 2405 | /* |
5c45bf27 | 2406 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2407 | * This is the group from where we need to pick up the load |
2408 | * for saving power | |
2409 | */ | |
2410 | if ((sum_nr_running < min_nr_running) || | |
2411 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2412 | first_cpu(group->cpumask) < |
2413 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2414 | group_min = group; |
2415 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2416 | min_load_per_task = sum_weighted_load / |
2417 | sum_nr_running; | |
dd41f596 | 2418 | } |
5c45bf27 | 2419 | |
dd41f596 | 2420 | /* |
5c45bf27 | 2421 | * Calculate the group which is almost near its |
dd41f596 IM |
2422 | * capacity but still has some space to pick up some load |
2423 | * from other group and save more power | |
2424 | */ | |
2425 | if (sum_nr_running <= group_capacity - 1) { | |
2426 | if (sum_nr_running > leader_nr_running || | |
2427 | (sum_nr_running == leader_nr_running && | |
2428 | first_cpu(group->cpumask) > | |
2429 | first_cpu(group_leader->cpumask))) { | |
2430 | group_leader = group; | |
2431 | leader_nr_running = sum_nr_running; | |
2432 | } | |
48f24c4d | 2433 | } |
5c45bf27 SS |
2434 | group_next: |
2435 | #endif | |
1da177e4 LT |
2436 | group = group->next; |
2437 | } while (group != sd->groups); | |
2438 | ||
2dd73a4f | 2439 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2440 | goto out_balanced; |
2441 | ||
2442 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2443 | ||
2444 | if (this_load >= avg_load || | |
2445 | 100*max_load <= sd->imbalance_pct*this_load) | |
2446 | goto out_balanced; | |
2447 | ||
2dd73a4f | 2448 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2449 | /* |
2450 | * We're trying to get all the cpus to the average_load, so we don't | |
2451 | * want to push ourselves above the average load, nor do we wish to | |
2452 | * reduce the max loaded cpu below the average load, as either of these | |
2453 | * actions would just result in more rebalancing later, and ping-pong | |
2454 | * tasks around. Thus we look for the minimum possible imbalance. | |
2455 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2456 | * be counted as no imbalance for these purposes -- we can't fix that | |
2457 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2458 | * appear as very large values with unsigned longs. | |
2459 | */ | |
2dd73a4f PW |
2460 | if (max_load <= busiest_load_per_task) |
2461 | goto out_balanced; | |
2462 | ||
2463 | /* | |
2464 | * In the presence of smp nice balancing, certain scenarios can have | |
2465 | * max load less than avg load(as we skip the groups at or below | |
2466 | * its cpu_power, while calculating max_load..) | |
2467 | */ | |
2468 | if (max_load < avg_load) { | |
2469 | *imbalance = 0; | |
2470 | goto small_imbalance; | |
2471 | } | |
0c117f1b SS |
2472 | |
2473 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2474 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2475 | |
1da177e4 | 2476 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2477 | *imbalance = min(max_pull * busiest->__cpu_power, |
2478 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2479 | / SCHED_LOAD_SCALE; |
2480 | ||
2dd73a4f PW |
2481 | /* |
2482 | * if *imbalance is less than the average load per runnable task | |
2483 | * there is no gaurantee that any tasks will be moved so we'll have | |
2484 | * a think about bumping its value to force at least one task to be | |
2485 | * moved | |
2486 | */ | |
7fd0d2dd | 2487 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2488 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2489 | unsigned int imbn; |
2490 | ||
2491 | small_imbalance: | |
2492 | pwr_move = pwr_now = 0; | |
2493 | imbn = 2; | |
2494 | if (this_nr_running) { | |
2495 | this_load_per_task /= this_nr_running; | |
2496 | if (busiest_load_per_task > this_load_per_task) | |
2497 | imbn = 1; | |
2498 | } else | |
2499 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2500 | |
dd41f596 IM |
2501 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2502 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2503 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2504 | return busiest; |
2505 | } | |
2506 | ||
2507 | /* | |
2508 | * OK, we don't have enough imbalance to justify moving tasks, | |
2509 | * however we may be able to increase total CPU power used by | |
2510 | * moving them. | |
2511 | */ | |
2512 | ||
5517d86b ED |
2513 | pwr_now += busiest->__cpu_power * |
2514 | min(busiest_load_per_task, max_load); | |
2515 | pwr_now += this->__cpu_power * | |
2516 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2517 | pwr_now /= SCHED_LOAD_SCALE; |
2518 | ||
2519 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2520 | tmp = sg_div_cpu_power(busiest, |
2521 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2522 | if (max_load > tmp) |
5517d86b | 2523 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2524 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2525 | |
2526 | /* Amount of load we'd add */ | |
5517d86b | 2527 | if (max_load * busiest->__cpu_power < |
33859f7f | 2528 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2529 | tmp = sg_div_cpu_power(this, |
2530 | max_load * busiest->__cpu_power); | |
1da177e4 | 2531 | else |
5517d86b ED |
2532 | tmp = sg_div_cpu_power(this, |
2533 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2534 | pwr_move += this->__cpu_power * | |
2535 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2536 | pwr_move /= SCHED_LOAD_SCALE; |
2537 | ||
2538 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2539 | if (pwr_move > pwr_now) |
2540 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2541 | } |
2542 | ||
1da177e4 LT |
2543 | return busiest; |
2544 | ||
2545 | out_balanced: | |
5c45bf27 | 2546 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2547 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2548 | goto ret; |
1da177e4 | 2549 | |
5c45bf27 SS |
2550 | if (this == group_leader && group_leader != group_min) { |
2551 | *imbalance = min_load_per_task; | |
2552 | return group_min; | |
2553 | } | |
5c45bf27 | 2554 | #endif |
783609c6 | 2555 | ret: |
1da177e4 LT |
2556 | *imbalance = 0; |
2557 | return NULL; | |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2562 | */ | |
70b97a7f | 2563 | static struct rq * |
d15bcfdb | 2564 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2565 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2566 | { |
70b97a7f | 2567 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2568 | unsigned long max_load = 0; |
1da177e4 LT |
2569 | int i; |
2570 | ||
2571 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2572 | unsigned long wl; |
0a2966b4 CL |
2573 | |
2574 | if (!cpu_isset(i, *cpus)) | |
2575 | continue; | |
2576 | ||
48f24c4d | 2577 | rq = cpu_rq(i); |
dd41f596 | 2578 | wl = weighted_cpuload(i); |
2dd73a4f | 2579 | |
dd41f596 | 2580 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2581 | continue; |
1da177e4 | 2582 | |
dd41f596 IM |
2583 | if (wl > max_load) { |
2584 | max_load = wl; | |
48f24c4d | 2585 | busiest = rq; |
1da177e4 LT |
2586 | } |
2587 | } | |
2588 | ||
2589 | return busiest; | |
2590 | } | |
2591 | ||
77391d71 NP |
2592 | /* |
2593 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2594 | * so long as it is large enough. | |
2595 | */ | |
2596 | #define MAX_PINNED_INTERVAL 512 | |
2597 | ||
1da177e4 LT |
2598 | /* |
2599 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2600 | * tasks if there is an imbalance. | |
1da177e4 | 2601 | */ |
70b97a7f | 2602 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2603 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2604 | int *balance) |
1da177e4 | 2605 | { |
43010659 | 2606 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2607 | struct sched_group *group; |
1da177e4 | 2608 | unsigned long imbalance; |
70b97a7f | 2609 | struct rq *busiest; |
0a2966b4 | 2610 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2611 | unsigned long flags; |
5969fe06 | 2612 | |
89c4710e SS |
2613 | /* |
2614 | * When power savings policy is enabled for the parent domain, idle | |
2615 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2616 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2617 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2618 | */ |
d15bcfdb | 2619 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2620 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2621 | sd_idle = 1; |
1da177e4 | 2622 | |
1da177e4 LT |
2623 | schedstat_inc(sd, lb_cnt[idle]); |
2624 | ||
0a2966b4 CL |
2625 | redo: |
2626 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2627 | &cpus, balance); |
2628 | ||
06066714 | 2629 | if (*balance == 0) |
783609c6 | 2630 | goto out_balanced; |
783609c6 | 2631 | |
1da177e4 LT |
2632 | if (!group) { |
2633 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2634 | goto out_balanced; | |
2635 | } | |
2636 | ||
0a2966b4 | 2637 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2638 | if (!busiest) { |
2639 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2640 | goto out_balanced; | |
2641 | } | |
2642 | ||
db935dbd | 2643 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2644 | |
2645 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2646 | ||
43010659 | 2647 | ld_moved = 0; |
1da177e4 LT |
2648 | if (busiest->nr_running > 1) { |
2649 | /* | |
2650 | * Attempt to move tasks. If find_busiest_group has found | |
2651 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2652 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2653 | * correctly treated as an imbalance. |
2654 | */ | |
fe2eea3f | 2655 | local_irq_save(flags); |
e17224bf | 2656 | double_rq_lock(this_rq, busiest); |
43010659 | 2657 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2658 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2659 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2660 | local_irq_restore(flags); |
81026794 | 2661 | |
46cb4b7c SS |
2662 | /* |
2663 | * some other cpu did the load balance for us. | |
2664 | */ | |
43010659 | 2665 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2666 | resched_cpu(this_cpu); |
2667 | ||
81026794 | 2668 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2669 | if (unlikely(all_pinned)) { |
2670 | cpu_clear(cpu_of(busiest), cpus); | |
2671 | if (!cpus_empty(cpus)) | |
2672 | goto redo; | |
81026794 | 2673 | goto out_balanced; |
0a2966b4 | 2674 | } |
1da177e4 | 2675 | } |
81026794 | 2676 | |
43010659 | 2677 | if (!ld_moved) { |
1da177e4 LT |
2678 | schedstat_inc(sd, lb_failed[idle]); |
2679 | sd->nr_balance_failed++; | |
2680 | ||
2681 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2682 | |
fe2eea3f | 2683 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2684 | |
2685 | /* don't kick the migration_thread, if the curr | |
2686 | * task on busiest cpu can't be moved to this_cpu | |
2687 | */ | |
2688 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2689 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2690 | all_pinned = 1; |
2691 | goto out_one_pinned; | |
2692 | } | |
2693 | ||
1da177e4 LT |
2694 | if (!busiest->active_balance) { |
2695 | busiest->active_balance = 1; | |
2696 | busiest->push_cpu = this_cpu; | |
81026794 | 2697 | active_balance = 1; |
1da177e4 | 2698 | } |
fe2eea3f | 2699 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2700 | if (active_balance) |
1da177e4 LT |
2701 | wake_up_process(busiest->migration_thread); |
2702 | ||
2703 | /* | |
2704 | * We've kicked active balancing, reset the failure | |
2705 | * counter. | |
2706 | */ | |
39507451 | 2707 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2708 | } |
81026794 | 2709 | } else |
1da177e4 LT |
2710 | sd->nr_balance_failed = 0; |
2711 | ||
81026794 | 2712 | if (likely(!active_balance)) { |
1da177e4 LT |
2713 | /* We were unbalanced, so reset the balancing interval */ |
2714 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2715 | } else { |
2716 | /* | |
2717 | * If we've begun active balancing, start to back off. This | |
2718 | * case may not be covered by the all_pinned logic if there | |
2719 | * is only 1 task on the busy runqueue (because we don't call | |
2720 | * move_tasks). | |
2721 | */ | |
2722 | if (sd->balance_interval < sd->max_interval) | |
2723 | sd->balance_interval *= 2; | |
1da177e4 LT |
2724 | } |
2725 | ||
43010659 | 2726 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2727 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2728 | return -1; |
43010659 | 2729 | return ld_moved; |
1da177e4 LT |
2730 | |
2731 | out_balanced: | |
1da177e4 LT |
2732 | schedstat_inc(sd, lb_balanced[idle]); |
2733 | ||
16cfb1c0 | 2734 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2735 | |
2736 | out_one_pinned: | |
1da177e4 | 2737 | /* tune up the balancing interval */ |
77391d71 NP |
2738 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2739 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2740 | sd->balance_interval *= 2; |
2741 | ||
48f24c4d | 2742 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2743 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2744 | return -1; |
1da177e4 LT |
2745 | return 0; |
2746 | } | |
2747 | ||
2748 | /* | |
2749 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2750 | * tasks if there is an imbalance. | |
2751 | * | |
d15bcfdb | 2752 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2753 | * this_rq is locked. |
2754 | */ | |
48f24c4d | 2755 | static int |
70b97a7f | 2756 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2757 | { |
2758 | struct sched_group *group; | |
70b97a7f | 2759 | struct rq *busiest = NULL; |
1da177e4 | 2760 | unsigned long imbalance; |
43010659 | 2761 | int ld_moved = 0; |
5969fe06 | 2762 | int sd_idle = 0; |
969bb4e4 | 2763 | int all_pinned = 0; |
0a2966b4 | 2764 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2765 | |
89c4710e SS |
2766 | /* |
2767 | * When power savings policy is enabled for the parent domain, idle | |
2768 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2769 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2770 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2771 | */ |
2772 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2773 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2774 | sd_idle = 1; |
1da177e4 | 2775 | |
d15bcfdb | 2776 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
0a2966b4 | 2777 | redo: |
d15bcfdb | 2778 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2779 | &sd_idle, &cpus, NULL); |
1da177e4 | 2780 | if (!group) { |
d15bcfdb | 2781 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2782 | goto out_balanced; |
1da177e4 LT |
2783 | } |
2784 | ||
d15bcfdb | 2785 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2786 | &cpus); |
db935dbd | 2787 | if (!busiest) { |
d15bcfdb | 2788 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2789 | goto out_balanced; |
1da177e4 LT |
2790 | } |
2791 | ||
db935dbd NP |
2792 | BUG_ON(busiest == this_rq); |
2793 | ||
d15bcfdb | 2794 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2795 | |
43010659 | 2796 | ld_moved = 0; |
d6d5cfaf NP |
2797 | if (busiest->nr_running > 1) { |
2798 | /* Attempt to move tasks */ | |
2799 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2800 | /* this_rq->clock is already updated */ |
2801 | update_rq_clock(busiest); | |
43010659 | 2802 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2803 | imbalance, sd, CPU_NEWLY_IDLE, |
2804 | &all_pinned); | |
d6d5cfaf | 2805 | spin_unlock(&busiest->lock); |
0a2966b4 | 2806 | |
969bb4e4 | 2807 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2808 | cpu_clear(cpu_of(busiest), cpus); |
2809 | if (!cpus_empty(cpus)) | |
2810 | goto redo; | |
2811 | } | |
d6d5cfaf NP |
2812 | } |
2813 | ||
43010659 | 2814 | if (!ld_moved) { |
d15bcfdb | 2815 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2816 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2817 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2818 | return -1; |
2819 | } else | |
16cfb1c0 | 2820 | sd->nr_balance_failed = 0; |
1da177e4 | 2821 | |
43010659 | 2822 | return ld_moved; |
16cfb1c0 NP |
2823 | |
2824 | out_balanced: | |
d15bcfdb | 2825 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2826 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2827 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2828 | return -1; |
16cfb1c0 | 2829 | sd->nr_balance_failed = 0; |
48f24c4d | 2830 | |
16cfb1c0 | 2831 | return 0; |
1da177e4 LT |
2832 | } |
2833 | ||
2834 | /* | |
2835 | * idle_balance is called by schedule() if this_cpu is about to become | |
2836 | * idle. Attempts to pull tasks from other CPUs. | |
2837 | */ | |
70b97a7f | 2838 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2839 | { |
2840 | struct sched_domain *sd; | |
dd41f596 IM |
2841 | int pulled_task = -1; |
2842 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2843 | |
2844 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2845 | unsigned long interval; |
2846 | ||
2847 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2848 | continue; | |
2849 | ||
2850 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2851 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2852 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2853 | this_rq, sd); |
2854 | ||
2855 | interval = msecs_to_jiffies(sd->balance_interval); | |
2856 | if (time_after(next_balance, sd->last_balance + interval)) | |
2857 | next_balance = sd->last_balance + interval; | |
2858 | if (pulled_task) | |
2859 | break; | |
1da177e4 | 2860 | } |
dd41f596 | 2861 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2862 | /* |
2863 | * We are going idle. next_balance may be set based on | |
2864 | * a busy processor. So reset next_balance. | |
2865 | */ | |
2866 | this_rq->next_balance = next_balance; | |
dd41f596 | 2867 | } |
1da177e4 LT |
2868 | } |
2869 | ||
2870 | /* | |
2871 | * active_load_balance is run by migration threads. It pushes running tasks | |
2872 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2873 | * running on each physical CPU where possible, and avoids physical / | |
2874 | * logical imbalances. | |
2875 | * | |
2876 | * Called with busiest_rq locked. | |
2877 | */ | |
70b97a7f | 2878 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2879 | { |
39507451 | 2880 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2881 | struct sched_domain *sd; |
2882 | struct rq *target_rq; | |
39507451 | 2883 | |
48f24c4d | 2884 | /* Is there any task to move? */ |
39507451 | 2885 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2886 | return; |
2887 | ||
2888 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2889 | |
2890 | /* | |
39507451 NP |
2891 | * This condition is "impossible", if it occurs |
2892 | * we need to fix it. Originally reported by | |
2893 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2894 | */ |
39507451 | 2895 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2896 | |
39507451 NP |
2897 | /* move a task from busiest_rq to target_rq */ |
2898 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2899 | update_rq_clock(busiest_rq); |
2900 | update_rq_clock(target_rq); | |
39507451 NP |
2901 | |
2902 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2903 | for_each_domain(target_cpu, sd) { |
39507451 | 2904 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2905 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2906 | break; |
c96d145e | 2907 | } |
39507451 | 2908 | |
48f24c4d IM |
2909 | if (likely(sd)) { |
2910 | schedstat_inc(sd, alb_cnt); | |
39507451 | 2911 | |
43010659 PW |
2912 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2913 | sd, CPU_IDLE)) | |
48f24c4d IM |
2914 | schedstat_inc(sd, alb_pushed); |
2915 | else | |
2916 | schedstat_inc(sd, alb_failed); | |
2917 | } | |
39507451 | 2918 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2919 | } |
2920 | ||
46cb4b7c SS |
2921 | #ifdef CONFIG_NO_HZ |
2922 | static struct { | |
2923 | atomic_t load_balancer; | |
2924 | cpumask_t cpu_mask; | |
2925 | } nohz ____cacheline_aligned = { | |
2926 | .load_balancer = ATOMIC_INIT(-1), | |
2927 | .cpu_mask = CPU_MASK_NONE, | |
2928 | }; | |
2929 | ||
7835b98b | 2930 | /* |
46cb4b7c SS |
2931 | * This routine will try to nominate the ilb (idle load balancing) |
2932 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2933 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2934 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2935 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2936 | * arrives... | |
2937 | * | |
2938 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2939 | * for idle load balancing. ilb owner will still be part of | |
2940 | * nohz.cpu_mask.. | |
7835b98b | 2941 | * |
46cb4b7c SS |
2942 | * While stopping the tick, this cpu will become the ilb owner if there |
2943 | * is no other owner. And will be the owner till that cpu becomes busy | |
2944 | * or if all cpus in the system stop their ticks at which point | |
2945 | * there is no need for ilb owner. | |
2946 | * | |
2947 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2948 | * next busy scheduler_tick() | |
2949 | */ | |
2950 | int select_nohz_load_balancer(int stop_tick) | |
2951 | { | |
2952 | int cpu = smp_processor_id(); | |
2953 | ||
2954 | if (stop_tick) { | |
2955 | cpu_set(cpu, nohz.cpu_mask); | |
2956 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2957 | ||
2958 | /* | |
2959 | * If we are going offline and still the leader, give up! | |
2960 | */ | |
2961 | if (cpu_is_offline(cpu) && | |
2962 | atomic_read(&nohz.load_balancer) == cpu) { | |
2963 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2964 | BUG(); | |
2965 | return 0; | |
2966 | } | |
2967 | ||
2968 | /* time for ilb owner also to sleep */ | |
2969 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2970 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2971 | atomic_set(&nohz.load_balancer, -1); | |
2972 | return 0; | |
2973 | } | |
2974 | ||
2975 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2976 | /* make me the ilb owner */ | |
2977 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2978 | return 1; | |
2979 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2980 | return 1; | |
2981 | } else { | |
2982 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2983 | return 0; | |
2984 | ||
2985 | cpu_clear(cpu, nohz.cpu_mask); | |
2986 | ||
2987 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2988 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2989 | BUG(); | |
2990 | } | |
2991 | return 0; | |
2992 | } | |
2993 | #endif | |
2994 | ||
2995 | static DEFINE_SPINLOCK(balancing); | |
2996 | ||
2997 | /* | |
7835b98b CL |
2998 | * It checks each scheduling domain to see if it is due to be balanced, |
2999 | * and initiates a balancing operation if so. | |
3000 | * | |
3001 | * Balancing parameters are set up in arch_init_sched_domains. | |
3002 | */ | |
d15bcfdb | 3003 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3004 | { |
46cb4b7c SS |
3005 | int balance = 1; |
3006 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3007 | unsigned long interval; |
3008 | struct sched_domain *sd; | |
46cb4b7c | 3009 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3010 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3011 | int update_next_balance = 0; |
1da177e4 | 3012 | |
46cb4b7c | 3013 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3014 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3015 | continue; | |
3016 | ||
3017 | interval = sd->balance_interval; | |
d15bcfdb | 3018 | if (idle != CPU_IDLE) |
1da177e4 LT |
3019 | interval *= sd->busy_factor; |
3020 | ||
3021 | /* scale ms to jiffies */ | |
3022 | interval = msecs_to_jiffies(interval); | |
3023 | if (unlikely(!interval)) | |
3024 | interval = 1; | |
dd41f596 IM |
3025 | if (interval > HZ*NR_CPUS/10) |
3026 | interval = HZ*NR_CPUS/10; | |
3027 | ||
1da177e4 | 3028 | |
08c183f3 CL |
3029 | if (sd->flags & SD_SERIALIZE) { |
3030 | if (!spin_trylock(&balancing)) | |
3031 | goto out; | |
3032 | } | |
3033 | ||
c9819f45 | 3034 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3035 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3036 | /* |
3037 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3038 | * longer idle, or one of our SMT siblings is |
3039 | * not idle. | |
3040 | */ | |
d15bcfdb | 3041 | idle = CPU_NOT_IDLE; |
1da177e4 | 3042 | } |
1bd77f2d | 3043 | sd->last_balance = jiffies; |
1da177e4 | 3044 | } |
08c183f3 CL |
3045 | if (sd->flags & SD_SERIALIZE) |
3046 | spin_unlock(&balancing); | |
3047 | out: | |
f549da84 | 3048 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3049 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3050 | update_next_balance = 1; |
3051 | } | |
783609c6 SS |
3052 | |
3053 | /* | |
3054 | * Stop the load balance at this level. There is another | |
3055 | * CPU in our sched group which is doing load balancing more | |
3056 | * actively. | |
3057 | */ | |
3058 | if (!balance) | |
3059 | break; | |
1da177e4 | 3060 | } |
f549da84 SS |
3061 | |
3062 | /* | |
3063 | * next_balance will be updated only when there is a need. | |
3064 | * When the cpu is attached to null domain for ex, it will not be | |
3065 | * updated. | |
3066 | */ | |
3067 | if (likely(update_next_balance)) | |
3068 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3069 | } |
3070 | ||
3071 | /* | |
3072 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3073 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3074 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3075 | */ | |
3076 | static void run_rebalance_domains(struct softirq_action *h) | |
3077 | { | |
dd41f596 IM |
3078 | int this_cpu = smp_processor_id(); |
3079 | struct rq *this_rq = cpu_rq(this_cpu); | |
3080 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3081 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3082 | |
dd41f596 | 3083 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3084 | |
3085 | #ifdef CONFIG_NO_HZ | |
3086 | /* | |
3087 | * If this cpu is the owner for idle load balancing, then do the | |
3088 | * balancing on behalf of the other idle cpus whose ticks are | |
3089 | * stopped. | |
3090 | */ | |
dd41f596 IM |
3091 | if (this_rq->idle_at_tick && |
3092 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3093 | cpumask_t cpus = nohz.cpu_mask; |
3094 | struct rq *rq; | |
3095 | int balance_cpu; | |
3096 | ||
dd41f596 | 3097 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3098 | for_each_cpu_mask(balance_cpu, cpus) { |
3099 | /* | |
3100 | * If this cpu gets work to do, stop the load balancing | |
3101 | * work being done for other cpus. Next load | |
3102 | * balancing owner will pick it up. | |
3103 | */ | |
3104 | if (need_resched()) | |
3105 | break; | |
3106 | ||
de0cf899 | 3107 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3108 | |
3109 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3110 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3111 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3112 | } |
3113 | } | |
3114 | #endif | |
3115 | } | |
3116 | ||
3117 | /* | |
3118 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3119 | * | |
3120 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3121 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3122 | * if the whole system is idle. | |
3123 | */ | |
dd41f596 | 3124 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3125 | { |
46cb4b7c SS |
3126 | #ifdef CONFIG_NO_HZ |
3127 | /* | |
3128 | * If we were in the nohz mode recently and busy at the current | |
3129 | * scheduler tick, then check if we need to nominate new idle | |
3130 | * load balancer. | |
3131 | */ | |
3132 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3133 | rq->in_nohz_recently = 0; | |
3134 | ||
3135 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3136 | cpu_clear(cpu, nohz.cpu_mask); | |
3137 | atomic_set(&nohz.load_balancer, -1); | |
3138 | } | |
3139 | ||
3140 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3141 | /* | |
3142 | * simple selection for now: Nominate the | |
3143 | * first cpu in the nohz list to be the next | |
3144 | * ilb owner. | |
3145 | * | |
3146 | * TBD: Traverse the sched domains and nominate | |
3147 | * the nearest cpu in the nohz.cpu_mask. | |
3148 | */ | |
3149 | int ilb = first_cpu(nohz.cpu_mask); | |
3150 | ||
3151 | if (ilb != NR_CPUS) | |
3152 | resched_cpu(ilb); | |
3153 | } | |
3154 | } | |
3155 | ||
3156 | /* | |
3157 | * If this cpu is idle and doing idle load balancing for all the | |
3158 | * cpus with ticks stopped, is it time for that to stop? | |
3159 | */ | |
3160 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3161 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3162 | resched_cpu(cpu); | |
3163 | return; | |
3164 | } | |
3165 | ||
3166 | /* | |
3167 | * If this cpu is idle and the idle load balancing is done by | |
3168 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3169 | */ | |
3170 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3171 | cpu_isset(cpu, nohz.cpu_mask)) | |
3172 | return; | |
3173 | #endif | |
3174 | if (time_after_eq(jiffies, rq->next_balance)) | |
3175 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3176 | } |
dd41f596 IM |
3177 | |
3178 | #else /* CONFIG_SMP */ | |
3179 | ||
1da177e4 LT |
3180 | /* |
3181 | * on UP we do not need to balance between CPUs: | |
3182 | */ | |
70b97a7f | 3183 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3184 | { |
3185 | } | |
dd41f596 IM |
3186 | |
3187 | /* Avoid "used but not defined" warning on UP */ | |
3188 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3189 | unsigned long max_nr_move, unsigned long max_load_move, | |
3190 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3191 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3192 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3193 | { |
3194 | *load_moved = 0; | |
3195 | ||
3196 | return 0; | |
3197 | } | |
3198 | ||
1da177e4 LT |
3199 | #endif |
3200 | ||
1da177e4 LT |
3201 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3202 | ||
3203 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3204 | ||
3205 | /* | |
41b86e9c IM |
3206 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3207 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3208 | */ |
41b86e9c | 3209 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3210 | { |
1da177e4 | 3211 | unsigned long flags; |
41b86e9c IM |
3212 | u64 ns, delta_exec; |
3213 | struct rq *rq; | |
48f24c4d | 3214 | |
41b86e9c IM |
3215 | rq = task_rq_lock(p, &flags); |
3216 | ns = p->se.sum_exec_runtime; | |
3217 | if (rq->curr == p) { | |
a8e504d2 IM |
3218 | update_rq_clock(rq); |
3219 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3220 | if ((s64)delta_exec > 0) |
3221 | ns += delta_exec; | |
3222 | } | |
3223 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3224 | |
1da177e4 LT |
3225 | return ns; |
3226 | } | |
3227 | ||
1da177e4 LT |
3228 | /* |
3229 | * Account user cpu time to a process. | |
3230 | * @p: the process that the cpu time gets accounted to | |
3231 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3232 | * @cputime: the cpu time spent in user space since the last update | |
3233 | */ | |
3234 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3235 | { | |
3236 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3237 | cputime64_t tmp; | |
3238 | ||
3239 | p->utime = cputime_add(p->utime, cputime); | |
3240 | ||
3241 | /* Add user time to cpustat. */ | |
3242 | tmp = cputime_to_cputime64(cputime); | |
3243 | if (TASK_NICE(p) > 0) | |
3244 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3245 | else | |
3246 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3247 | } | |
3248 | ||
3249 | /* | |
3250 | * Account system cpu time to a process. | |
3251 | * @p: the process that the cpu time gets accounted to | |
3252 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3253 | * @cputime: the cpu time spent in kernel space since the last update | |
3254 | */ | |
3255 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3256 | cputime_t cputime) | |
3257 | { | |
3258 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3259 | struct rq *rq = this_rq(); |
1da177e4 LT |
3260 | cputime64_t tmp; |
3261 | ||
3262 | p->stime = cputime_add(p->stime, cputime); | |
3263 | ||
3264 | /* Add system time to cpustat. */ | |
3265 | tmp = cputime_to_cputime64(cputime); | |
3266 | if (hardirq_count() - hardirq_offset) | |
3267 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3268 | else if (softirq_count()) | |
3269 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3270 | else if (p != rq->idle) | |
3271 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3272 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3273 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3274 | else | |
3275 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3276 | /* Account for system time used */ | |
3277 | acct_update_integrals(p); | |
1da177e4 LT |
3278 | } |
3279 | ||
3280 | /* | |
3281 | * Account for involuntary wait time. | |
3282 | * @p: the process from which the cpu time has been stolen | |
3283 | * @steal: the cpu time spent in involuntary wait | |
3284 | */ | |
3285 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3286 | { | |
3287 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3288 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3289 | struct rq *rq = this_rq(); |
1da177e4 LT |
3290 | |
3291 | if (p == rq->idle) { | |
3292 | p->stime = cputime_add(p->stime, steal); | |
3293 | if (atomic_read(&rq->nr_iowait) > 0) | |
3294 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3295 | else | |
3296 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3297 | } else | |
3298 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3299 | } | |
3300 | ||
7835b98b CL |
3301 | /* |
3302 | * This function gets called by the timer code, with HZ frequency. | |
3303 | * We call it with interrupts disabled. | |
3304 | * | |
3305 | * It also gets called by the fork code, when changing the parent's | |
3306 | * timeslices. | |
3307 | */ | |
3308 | void scheduler_tick(void) | |
3309 | { | |
7835b98b CL |
3310 | int cpu = smp_processor_id(); |
3311 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3312 | struct task_struct *curr = rq->curr; |
529c7726 | 3313 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3314 | |
3315 | spin_lock(&rq->lock); | |
546fe3c9 | 3316 | __update_rq_clock(rq); |
529c7726 IM |
3317 | /* |
3318 | * Let rq->clock advance by at least TICK_NSEC: | |
3319 | */ | |
3320 | if (unlikely(rq->clock < next_tick)) | |
3321 | rq->clock = next_tick; | |
3322 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3323 | update_cpu_load(rq); |
dd41f596 IM |
3324 | if (curr != rq->idle) /* FIXME: needed? */ |
3325 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3326 | spin_unlock(&rq->lock); |
7835b98b | 3327 | |
e418e1c2 | 3328 | #ifdef CONFIG_SMP |
dd41f596 IM |
3329 | rq->idle_at_tick = idle_cpu(cpu); |
3330 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3331 | #endif |
1da177e4 LT |
3332 | } |
3333 | ||
1da177e4 LT |
3334 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3335 | ||
3336 | void fastcall add_preempt_count(int val) | |
3337 | { | |
3338 | /* | |
3339 | * Underflow? | |
3340 | */ | |
9a11b49a IM |
3341 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3342 | return; | |
1da177e4 LT |
3343 | preempt_count() += val; |
3344 | /* | |
3345 | * Spinlock count overflowing soon? | |
3346 | */ | |
33859f7f MOS |
3347 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3348 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3349 | } |
3350 | EXPORT_SYMBOL(add_preempt_count); | |
3351 | ||
3352 | void fastcall sub_preempt_count(int val) | |
3353 | { | |
3354 | /* | |
3355 | * Underflow? | |
3356 | */ | |
9a11b49a IM |
3357 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3358 | return; | |
1da177e4 LT |
3359 | /* |
3360 | * Is the spinlock portion underflowing? | |
3361 | */ | |
9a11b49a IM |
3362 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3363 | !(preempt_count() & PREEMPT_MASK))) | |
3364 | return; | |
3365 | ||
1da177e4 LT |
3366 | preempt_count() -= val; |
3367 | } | |
3368 | EXPORT_SYMBOL(sub_preempt_count); | |
3369 | ||
3370 | #endif | |
3371 | ||
3372 | /* | |
dd41f596 | 3373 | * Print scheduling while atomic bug: |
1da177e4 | 3374 | */ |
dd41f596 | 3375 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3376 | { |
dd41f596 IM |
3377 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3378 | prev->comm, preempt_count(), prev->pid); | |
3379 | debug_show_held_locks(prev); | |
3380 | if (irqs_disabled()) | |
3381 | print_irqtrace_events(prev); | |
3382 | dump_stack(); | |
3383 | } | |
1da177e4 | 3384 | |
dd41f596 IM |
3385 | /* |
3386 | * Various schedule()-time debugging checks and statistics: | |
3387 | */ | |
3388 | static inline void schedule_debug(struct task_struct *prev) | |
3389 | { | |
1da177e4 LT |
3390 | /* |
3391 | * Test if we are atomic. Since do_exit() needs to call into | |
3392 | * schedule() atomically, we ignore that path for now. | |
3393 | * Otherwise, whine if we are scheduling when we should not be. | |
3394 | */ | |
dd41f596 IM |
3395 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3396 | __schedule_bug(prev); | |
3397 | ||
1da177e4 LT |
3398 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3399 | ||
dd41f596 IM |
3400 | schedstat_inc(this_rq(), sched_cnt); |
3401 | } | |
3402 | ||
3403 | /* | |
3404 | * Pick up the highest-prio task: | |
3405 | */ | |
3406 | static inline struct task_struct * | |
ff95f3df | 3407 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 IM |
3408 | { |
3409 | struct sched_class *class; | |
3410 | struct task_struct *p; | |
1da177e4 LT |
3411 | |
3412 | /* | |
dd41f596 IM |
3413 | * Optimization: we know that if all tasks are in |
3414 | * the fair class we can call that function directly: | |
1da177e4 | 3415 | */ |
dd41f596 | 3416 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3417 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3418 | if (likely(p)) |
3419 | return p; | |
1da177e4 LT |
3420 | } |
3421 | ||
dd41f596 IM |
3422 | class = sched_class_highest; |
3423 | for ( ; ; ) { | |
fb8d4724 | 3424 | p = class->pick_next_task(rq); |
dd41f596 IM |
3425 | if (p) |
3426 | return p; | |
3427 | /* | |
3428 | * Will never be NULL as the idle class always | |
3429 | * returns a non-NULL p: | |
3430 | */ | |
3431 | class = class->next; | |
3432 | } | |
3433 | } | |
1da177e4 | 3434 | |
dd41f596 IM |
3435 | /* |
3436 | * schedule() is the main scheduler function. | |
3437 | */ | |
3438 | asmlinkage void __sched schedule(void) | |
3439 | { | |
3440 | struct task_struct *prev, *next; | |
3441 | long *switch_count; | |
3442 | struct rq *rq; | |
dd41f596 IM |
3443 | int cpu; |
3444 | ||
3445 | need_resched: | |
3446 | preempt_disable(); | |
3447 | cpu = smp_processor_id(); | |
3448 | rq = cpu_rq(cpu); | |
3449 | rcu_qsctr_inc(cpu); | |
3450 | prev = rq->curr; | |
3451 | switch_count = &prev->nivcsw; | |
3452 | ||
3453 | release_kernel_lock(prev); | |
3454 | need_resched_nonpreemptible: | |
3455 | ||
3456 | schedule_debug(prev); | |
1da177e4 LT |
3457 | |
3458 | spin_lock_irq(&rq->lock); | |
dd41f596 | 3459 | clear_tsk_need_resched(prev); |
c1b3da3e | 3460 | __update_rq_clock(rq); |
1da177e4 | 3461 | |
1da177e4 | 3462 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3463 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3464 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3465 | prev->state = TASK_RUNNING; |
dd41f596 | 3466 | } else { |
2e1cb74a | 3467 | deactivate_task(rq, prev, 1); |
1da177e4 | 3468 | } |
dd41f596 | 3469 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3470 | } |
3471 | ||
dd41f596 | 3472 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3473 | idle_balance(cpu, rq); |
1da177e4 | 3474 | |
31ee529c | 3475 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3476 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3477 | |
3478 | sched_info_switch(prev, next); | |
dd41f596 | 3479 | |
1da177e4 | 3480 | if (likely(prev != next)) { |
1da177e4 LT |
3481 | rq->nr_switches++; |
3482 | rq->curr = next; | |
3483 | ++*switch_count; | |
3484 | ||
dd41f596 | 3485 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3486 | } else |
3487 | spin_unlock_irq(&rq->lock); | |
3488 | ||
dd41f596 IM |
3489 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3490 | cpu = smp_processor_id(); | |
3491 | rq = cpu_rq(cpu); | |
1da177e4 | 3492 | goto need_resched_nonpreemptible; |
dd41f596 | 3493 | } |
1da177e4 LT |
3494 | preempt_enable_no_resched(); |
3495 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3496 | goto need_resched; | |
3497 | } | |
1da177e4 LT |
3498 | EXPORT_SYMBOL(schedule); |
3499 | ||
3500 | #ifdef CONFIG_PREEMPT | |
3501 | /* | |
2ed6e34f | 3502 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3503 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3504 | * occur there and call schedule directly. | |
3505 | */ | |
3506 | asmlinkage void __sched preempt_schedule(void) | |
3507 | { | |
3508 | struct thread_info *ti = current_thread_info(); | |
3509 | #ifdef CONFIG_PREEMPT_BKL | |
3510 | struct task_struct *task = current; | |
3511 | int saved_lock_depth; | |
3512 | #endif | |
3513 | /* | |
3514 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3515 | * we do not want to preempt the current task. Just return.. | |
3516 | */ | |
beed33a8 | 3517 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3518 | return; |
3519 | ||
3520 | need_resched: | |
3521 | add_preempt_count(PREEMPT_ACTIVE); | |
3522 | /* | |
3523 | * We keep the big kernel semaphore locked, but we | |
3524 | * clear ->lock_depth so that schedule() doesnt | |
3525 | * auto-release the semaphore: | |
3526 | */ | |
3527 | #ifdef CONFIG_PREEMPT_BKL | |
3528 | saved_lock_depth = task->lock_depth; | |
3529 | task->lock_depth = -1; | |
3530 | #endif | |
3531 | schedule(); | |
3532 | #ifdef CONFIG_PREEMPT_BKL | |
3533 | task->lock_depth = saved_lock_depth; | |
3534 | #endif | |
3535 | sub_preempt_count(PREEMPT_ACTIVE); | |
3536 | ||
3537 | /* we could miss a preemption opportunity between schedule and now */ | |
3538 | barrier(); | |
3539 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3540 | goto need_resched; | |
3541 | } | |
1da177e4 LT |
3542 | EXPORT_SYMBOL(preempt_schedule); |
3543 | ||
3544 | /* | |
2ed6e34f | 3545 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3546 | * off of irq context. |
3547 | * Note, that this is called and return with irqs disabled. This will | |
3548 | * protect us against recursive calling from irq. | |
3549 | */ | |
3550 | asmlinkage void __sched preempt_schedule_irq(void) | |
3551 | { | |
3552 | struct thread_info *ti = current_thread_info(); | |
3553 | #ifdef CONFIG_PREEMPT_BKL | |
3554 | struct task_struct *task = current; | |
3555 | int saved_lock_depth; | |
3556 | #endif | |
2ed6e34f | 3557 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3558 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3559 | ||
3560 | need_resched: | |
3561 | add_preempt_count(PREEMPT_ACTIVE); | |
3562 | /* | |
3563 | * We keep the big kernel semaphore locked, but we | |
3564 | * clear ->lock_depth so that schedule() doesnt | |
3565 | * auto-release the semaphore: | |
3566 | */ | |
3567 | #ifdef CONFIG_PREEMPT_BKL | |
3568 | saved_lock_depth = task->lock_depth; | |
3569 | task->lock_depth = -1; | |
3570 | #endif | |
3571 | local_irq_enable(); | |
3572 | schedule(); | |
3573 | local_irq_disable(); | |
3574 | #ifdef CONFIG_PREEMPT_BKL | |
3575 | task->lock_depth = saved_lock_depth; | |
3576 | #endif | |
3577 | sub_preempt_count(PREEMPT_ACTIVE); | |
3578 | ||
3579 | /* we could miss a preemption opportunity between schedule and now */ | |
3580 | barrier(); | |
3581 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3582 | goto need_resched; | |
3583 | } | |
3584 | ||
3585 | #endif /* CONFIG_PREEMPT */ | |
3586 | ||
95cdf3b7 IM |
3587 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3588 | void *key) | |
1da177e4 | 3589 | { |
48f24c4d | 3590 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3591 | } |
1da177e4 LT |
3592 | EXPORT_SYMBOL(default_wake_function); |
3593 | ||
3594 | /* | |
3595 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3596 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3597 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3598 | * | |
3599 | * There are circumstances in which we can try to wake a task which has already | |
3600 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3601 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3602 | */ | |
3603 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3604 | int nr_exclusive, int sync, void *key) | |
3605 | { | |
2e45874c | 3606 | wait_queue_t *curr, *next; |
1da177e4 | 3607 | |
2e45874c | 3608 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3609 | unsigned flags = curr->flags; |
3610 | ||
1da177e4 | 3611 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3612 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3613 | break; |
3614 | } | |
3615 | } | |
3616 | ||
3617 | /** | |
3618 | * __wake_up - wake up threads blocked on a waitqueue. | |
3619 | * @q: the waitqueue | |
3620 | * @mode: which threads | |
3621 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3622 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3623 | */ |
3624 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3625 | int nr_exclusive, void *key) |
1da177e4 LT |
3626 | { |
3627 | unsigned long flags; | |
3628 | ||
3629 | spin_lock_irqsave(&q->lock, flags); | |
3630 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3631 | spin_unlock_irqrestore(&q->lock, flags); | |
3632 | } | |
1da177e4 LT |
3633 | EXPORT_SYMBOL(__wake_up); |
3634 | ||
3635 | /* | |
3636 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3637 | */ | |
3638 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3639 | { | |
3640 | __wake_up_common(q, mode, 1, 0, NULL); | |
3641 | } | |
3642 | ||
3643 | /** | |
67be2dd1 | 3644 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3645 | * @q: the waitqueue |
3646 | * @mode: which threads | |
3647 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3648 | * | |
3649 | * The sync wakeup differs that the waker knows that it will schedule | |
3650 | * away soon, so while the target thread will be woken up, it will not | |
3651 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3652 | * with each other. This can prevent needless bouncing between CPUs. | |
3653 | * | |
3654 | * On UP it can prevent extra preemption. | |
3655 | */ | |
95cdf3b7 IM |
3656 | void fastcall |
3657 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3658 | { |
3659 | unsigned long flags; | |
3660 | int sync = 1; | |
3661 | ||
3662 | if (unlikely(!q)) | |
3663 | return; | |
3664 | ||
3665 | if (unlikely(!nr_exclusive)) | |
3666 | sync = 0; | |
3667 | ||
3668 | spin_lock_irqsave(&q->lock, flags); | |
3669 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3670 | spin_unlock_irqrestore(&q->lock, flags); | |
3671 | } | |
3672 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3673 | ||
3674 | void fastcall complete(struct completion *x) | |
3675 | { | |
3676 | unsigned long flags; | |
3677 | ||
3678 | spin_lock_irqsave(&x->wait.lock, flags); | |
3679 | x->done++; | |
3680 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3681 | 1, 0, NULL); | |
3682 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3683 | } | |
3684 | EXPORT_SYMBOL(complete); | |
3685 | ||
3686 | void fastcall complete_all(struct completion *x) | |
3687 | { | |
3688 | unsigned long flags; | |
3689 | ||
3690 | spin_lock_irqsave(&x->wait.lock, flags); | |
3691 | x->done += UINT_MAX/2; | |
3692 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3693 | 0, 0, NULL); | |
3694 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3695 | } | |
3696 | EXPORT_SYMBOL(complete_all); | |
3697 | ||
3698 | void fastcall __sched wait_for_completion(struct completion *x) | |
3699 | { | |
3700 | might_sleep(); | |
48f24c4d | 3701 | |
1da177e4 LT |
3702 | spin_lock_irq(&x->wait.lock); |
3703 | if (!x->done) { | |
3704 | DECLARE_WAITQUEUE(wait, current); | |
3705 | ||
3706 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3707 | __add_wait_queue_tail(&x->wait, &wait); | |
3708 | do { | |
3709 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3710 | spin_unlock_irq(&x->wait.lock); | |
3711 | schedule(); | |
3712 | spin_lock_irq(&x->wait.lock); | |
3713 | } while (!x->done); | |
3714 | __remove_wait_queue(&x->wait, &wait); | |
3715 | } | |
3716 | x->done--; | |
3717 | spin_unlock_irq(&x->wait.lock); | |
3718 | } | |
3719 | EXPORT_SYMBOL(wait_for_completion); | |
3720 | ||
3721 | unsigned long fastcall __sched | |
3722 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3723 | { | |
3724 | might_sleep(); | |
3725 | ||
3726 | spin_lock_irq(&x->wait.lock); | |
3727 | if (!x->done) { | |
3728 | DECLARE_WAITQUEUE(wait, current); | |
3729 | ||
3730 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3731 | __add_wait_queue_tail(&x->wait, &wait); | |
3732 | do { | |
3733 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3734 | spin_unlock_irq(&x->wait.lock); | |
3735 | timeout = schedule_timeout(timeout); | |
3736 | spin_lock_irq(&x->wait.lock); | |
3737 | if (!timeout) { | |
3738 | __remove_wait_queue(&x->wait, &wait); | |
3739 | goto out; | |
3740 | } | |
3741 | } while (!x->done); | |
3742 | __remove_wait_queue(&x->wait, &wait); | |
3743 | } | |
3744 | x->done--; | |
3745 | out: | |
3746 | spin_unlock_irq(&x->wait.lock); | |
3747 | return timeout; | |
3748 | } | |
3749 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3750 | ||
3751 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3752 | { | |
3753 | int ret = 0; | |
3754 | ||
3755 | might_sleep(); | |
3756 | ||
3757 | spin_lock_irq(&x->wait.lock); | |
3758 | if (!x->done) { | |
3759 | DECLARE_WAITQUEUE(wait, current); | |
3760 | ||
3761 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3762 | __add_wait_queue_tail(&x->wait, &wait); | |
3763 | do { | |
3764 | if (signal_pending(current)) { | |
3765 | ret = -ERESTARTSYS; | |
3766 | __remove_wait_queue(&x->wait, &wait); | |
3767 | goto out; | |
3768 | } | |
3769 | __set_current_state(TASK_INTERRUPTIBLE); | |
3770 | spin_unlock_irq(&x->wait.lock); | |
3771 | schedule(); | |
3772 | spin_lock_irq(&x->wait.lock); | |
3773 | } while (!x->done); | |
3774 | __remove_wait_queue(&x->wait, &wait); | |
3775 | } | |
3776 | x->done--; | |
3777 | out: | |
3778 | spin_unlock_irq(&x->wait.lock); | |
3779 | ||
3780 | return ret; | |
3781 | } | |
3782 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3783 | ||
3784 | unsigned long fastcall __sched | |
3785 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3786 | unsigned long timeout) | |
3787 | { | |
3788 | might_sleep(); | |
3789 | ||
3790 | spin_lock_irq(&x->wait.lock); | |
3791 | if (!x->done) { | |
3792 | DECLARE_WAITQUEUE(wait, current); | |
3793 | ||
3794 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3795 | __add_wait_queue_tail(&x->wait, &wait); | |
3796 | do { | |
3797 | if (signal_pending(current)) { | |
3798 | timeout = -ERESTARTSYS; | |
3799 | __remove_wait_queue(&x->wait, &wait); | |
3800 | goto out; | |
3801 | } | |
3802 | __set_current_state(TASK_INTERRUPTIBLE); | |
3803 | spin_unlock_irq(&x->wait.lock); | |
3804 | timeout = schedule_timeout(timeout); | |
3805 | spin_lock_irq(&x->wait.lock); | |
3806 | if (!timeout) { | |
3807 | __remove_wait_queue(&x->wait, &wait); | |
3808 | goto out; | |
3809 | } | |
3810 | } while (!x->done); | |
3811 | __remove_wait_queue(&x->wait, &wait); | |
3812 | } | |
3813 | x->done--; | |
3814 | out: | |
3815 | spin_unlock_irq(&x->wait.lock); | |
3816 | return timeout; | |
3817 | } | |
3818 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3819 | ||
0fec171c IM |
3820 | static inline void |
3821 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3822 | { | |
3823 | spin_lock_irqsave(&q->lock, *flags); | |
3824 | __add_wait_queue(q, wait); | |
1da177e4 | 3825 | spin_unlock(&q->lock); |
0fec171c | 3826 | } |
1da177e4 | 3827 | |
0fec171c IM |
3828 | static inline void |
3829 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3830 | { | |
3831 | spin_lock_irq(&q->lock); | |
3832 | __remove_wait_queue(q, wait); | |
3833 | spin_unlock_irqrestore(&q->lock, *flags); | |
3834 | } | |
1da177e4 | 3835 | |
0fec171c | 3836 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3837 | { |
0fec171c IM |
3838 | unsigned long flags; |
3839 | wait_queue_t wait; | |
3840 | ||
3841 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3842 | |
3843 | current->state = TASK_INTERRUPTIBLE; | |
3844 | ||
0fec171c | 3845 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3846 | schedule(); |
0fec171c | 3847 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3848 | } |
1da177e4 LT |
3849 | EXPORT_SYMBOL(interruptible_sleep_on); |
3850 | ||
0fec171c | 3851 | long __sched |
95cdf3b7 | 3852 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3853 | { |
0fec171c IM |
3854 | unsigned long flags; |
3855 | wait_queue_t wait; | |
3856 | ||
3857 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3858 | |
3859 | current->state = TASK_INTERRUPTIBLE; | |
3860 | ||
0fec171c | 3861 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3862 | timeout = schedule_timeout(timeout); |
0fec171c | 3863 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3864 | |
3865 | return timeout; | |
3866 | } | |
1da177e4 LT |
3867 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3868 | ||
0fec171c | 3869 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3870 | { |
0fec171c IM |
3871 | unsigned long flags; |
3872 | wait_queue_t wait; | |
3873 | ||
3874 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3875 | |
3876 | current->state = TASK_UNINTERRUPTIBLE; | |
3877 | ||
0fec171c | 3878 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3879 | schedule(); |
0fec171c | 3880 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3881 | } |
1da177e4 LT |
3882 | EXPORT_SYMBOL(sleep_on); |
3883 | ||
0fec171c | 3884 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3885 | { |
0fec171c IM |
3886 | unsigned long flags; |
3887 | wait_queue_t wait; | |
3888 | ||
3889 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3890 | |
3891 | current->state = TASK_UNINTERRUPTIBLE; | |
3892 | ||
0fec171c | 3893 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3894 | timeout = schedule_timeout(timeout); |
0fec171c | 3895 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3896 | |
3897 | return timeout; | |
3898 | } | |
1da177e4 LT |
3899 | EXPORT_SYMBOL(sleep_on_timeout); |
3900 | ||
b29739f9 IM |
3901 | #ifdef CONFIG_RT_MUTEXES |
3902 | ||
3903 | /* | |
3904 | * rt_mutex_setprio - set the current priority of a task | |
3905 | * @p: task | |
3906 | * @prio: prio value (kernel-internal form) | |
3907 | * | |
3908 | * This function changes the 'effective' priority of a task. It does | |
3909 | * not touch ->normal_prio like __setscheduler(). | |
3910 | * | |
3911 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3912 | */ | |
36c8b586 | 3913 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3914 | { |
3915 | unsigned long flags; | |
dd41f596 | 3916 | int oldprio, on_rq; |
70b97a7f | 3917 | struct rq *rq; |
b29739f9 IM |
3918 | |
3919 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3920 | ||
3921 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3922 | update_rq_clock(rq); |
b29739f9 | 3923 | |
d5f9f942 | 3924 | oldprio = p->prio; |
dd41f596 IM |
3925 | on_rq = p->se.on_rq; |
3926 | if (on_rq) | |
69be72c1 | 3927 | dequeue_task(rq, p, 0); |
dd41f596 IM |
3928 | |
3929 | if (rt_prio(prio)) | |
3930 | p->sched_class = &rt_sched_class; | |
3931 | else | |
3932 | p->sched_class = &fair_sched_class; | |
3933 | ||
b29739f9 IM |
3934 | p->prio = prio; |
3935 | ||
dd41f596 | 3936 | if (on_rq) { |
8159f87e | 3937 | enqueue_task(rq, p, 0); |
b29739f9 IM |
3938 | /* |
3939 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3940 | * our priority decreased, or if we are not currently running on |
3941 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3942 | */ |
d5f9f942 AM |
3943 | if (task_running(rq, p)) { |
3944 | if (p->prio > oldprio) | |
3945 | resched_task(rq->curr); | |
dd41f596 IM |
3946 | } else { |
3947 | check_preempt_curr(rq, p); | |
3948 | } | |
b29739f9 IM |
3949 | } |
3950 | task_rq_unlock(rq, &flags); | |
3951 | } | |
3952 | ||
3953 | #endif | |
3954 | ||
36c8b586 | 3955 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3956 | { |
dd41f596 | 3957 | int old_prio, delta, on_rq; |
1da177e4 | 3958 | unsigned long flags; |
70b97a7f | 3959 | struct rq *rq; |
1da177e4 LT |
3960 | |
3961 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3962 | return; | |
3963 | /* | |
3964 | * We have to be careful, if called from sys_setpriority(), | |
3965 | * the task might be in the middle of scheduling on another CPU. | |
3966 | */ | |
3967 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3968 | update_rq_clock(rq); |
1da177e4 LT |
3969 | /* |
3970 | * The RT priorities are set via sched_setscheduler(), but we still | |
3971 | * allow the 'normal' nice value to be set - but as expected | |
3972 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3973 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3974 | */ |
e05606d3 | 3975 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3976 | p->static_prio = NICE_TO_PRIO(nice); |
3977 | goto out_unlock; | |
3978 | } | |
dd41f596 IM |
3979 | on_rq = p->se.on_rq; |
3980 | if (on_rq) { | |
69be72c1 | 3981 | dequeue_task(rq, p, 0); |
79b5dddf | 3982 | dec_load(rq, p); |
2dd73a4f | 3983 | } |
1da177e4 | 3984 | |
1da177e4 | 3985 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3986 | set_load_weight(p); |
b29739f9 IM |
3987 | old_prio = p->prio; |
3988 | p->prio = effective_prio(p); | |
3989 | delta = p->prio - old_prio; | |
1da177e4 | 3990 | |
dd41f596 | 3991 | if (on_rq) { |
8159f87e | 3992 | enqueue_task(rq, p, 0); |
29b4b623 | 3993 | inc_load(rq, p); |
1da177e4 | 3994 | /* |
d5f9f942 AM |
3995 | * If the task increased its priority or is running and |
3996 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3997 | */ |
d5f9f942 | 3998 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3999 | resched_task(rq->curr); |
4000 | } | |
4001 | out_unlock: | |
4002 | task_rq_unlock(rq, &flags); | |
4003 | } | |
1da177e4 LT |
4004 | EXPORT_SYMBOL(set_user_nice); |
4005 | ||
e43379f1 MM |
4006 | /* |
4007 | * can_nice - check if a task can reduce its nice value | |
4008 | * @p: task | |
4009 | * @nice: nice value | |
4010 | */ | |
36c8b586 | 4011 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4012 | { |
024f4747 MM |
4013 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4014 | int nice_rlim = 20 - nice; | |
48f24c4d | 4015 | |
e43379f1 MM |
4016 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4017 | capable(CAP_SYS_NICE)); | |
4018 | } | |
4019 | ||
1da177e4 LT |
4020 | #ifdef __ARCH_WANT_SYS_NICE |
4021 | ||
4022 | /* | |
4023 | * sys_nice - change the priority of the current process. | |
4024 | * @increment: priority increment | |
4025 | * | |
4026 | * sys_setpriority is a more generic, but much slower function that | |
4027 | * does similar things. | |
4028 | */ | |
4029 | asmlinkage long sys_nice(int increment) | |
4030 | { | |
48f24c4d | 4031 | long nice, retval; |
1da177e4 LT |
4032 | |
4033 | /* | |
4034 | * Setpriority might change our priority at the same moment. | |
4035 | * We don't have to worry. Conceptually one call occurs first | |
4036 | * and we have a single winner. | |
4037 | */ | |
e43379f1 MM |
4038 | if (increment < -40) |
4039 | increment = -40; | |
1da177e4 LT |
4040 | if (increment > 40) |
4041 | increment = 40; | |
4042 | ||
4043 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4044 | if (nice < -20) | |
4045 | nice = -20; | |
4046 | if (nice > 19) | |
4047 | nice = 19; | |
4048 | ||
e43379f1 MM |
4049 | if (increment < 0 && !can_nice(current, nice)) |
4050 | return -EPERM; | |
4051 | ||
1da177e4 LT |
4052 | retval = security_task_setnice(current, nice); |
4053 | if (retval) | |
4054 | return retval; | |
4055 | ||
4056 | set_user_nice(current, nice); | |
4057 | return 0; | |
4058 | } | |
4059 | ||
4060 | #endif | |
4061 | ||
4062 | /** | |
4063 | * task_prio - return the priority value of a given task. | |
4064 | * @p: the task in question. | |
4065 | * | |
4066 | * This is the priority value as seen by users in /proc. | |
4067 | * RT tasks are offset by -200. Normal tasks are centered | |
4068 | * around 0, value goes from -16 to +15. | |
4069 | */ | |
36c8b586 | 4070 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4071 | { |
4072 | return p->prio - MAX_RT_PRIO; | |
4073 | } | |
4074 | ||
4075 | /** | |
4076 | * task_nice - return the nice value of a given task. | |
4077 | * @p: the task in question. | |
4078 | */ | |
36c8b586 | 4079 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4080 | { |
4081 | return TASK_NICE(p); | |
4082 | } | |
1da177e4 | 4083 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4084 | |
4085 | /** | |
4086 | * idle_cpu - is a given cpu idle currently? | |
4087 | * @cpu: the processor in question. | |
4088 | */ | |
4089 | int idle_cpu(int cpu) | |
4090 | { | |
4091 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4092 | } | |
4093 | ||
1da177e4 LT |
4094 | /** |
4095 | * idle_task - return the idle task for a given cpu. | |
4096 | * @cpu: the processor in question. | |
4097 | */ | |
36c8b586 | 4098 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4099 | { |
4100 | return cpu_rq(cpu)->idle; | |
4101 | } | |
4102 | ||
4103 | /** | |
4104 | * find_process_by_pid - find a process with a matching PID value. | |
4105 | * @pid: the pid in question. | |
4106 | */ | |
36c8b586 | 4107 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4108 | { |
4109 | return pid ? find_task_by_pid(pid) : current; | |
4110 | } | |
4111 | ||
4112 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4113 | static void |
4114 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4115 | { |
dd41f596 | 4116 | BUG_ON(p->se.on_rq); |
48f24c4d | 4117 | |
1da177e4 | 4118 | p->policy = policy; |
dd41f596 IM |
4119 | switch (p->policy) { |
4120 | case SCHED_NORMAL: | |
4121 | case SCHED_BATCH: | |
4122 | case SCHED_IDLE: | |
4123 | p->sched_class = &fair_sched_class; | |
4124 | break; | |
4125 | case SCHED_FIFO: | |
4126 | case SCHED_RR: | |
4127 | p->sched_class = &rt_sched_class; | |
4128 | break; | |
4129 | } | |
4130 | ||
1da177e4 | 4131 | p->rt_priority = prio; |
b29739f9 IM |
4132 | p->normal_prio = normal_prio(p); |
4133 | /* we are holding p->pi_lock already */ | |
4134 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4135 | set_load_weight(p); |
1da177e4 LT |
4136 | } |
4137 | ||
4138 | /** | |
72fd4a35 | 4139 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4140 | * @p: the task in question. |
4141 | * @policy: new policy. | |
4142 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4143 | * |
72fd4a35 | 4144 | * NOTE that the task may be already dead. |
1da177e4 | 4145 | */ |
95cdf3b7 IM |
4146 | int sched_setscheduler(struct task_struct *p, int policy, |
4147 | struct sched_param *param) | |
1da177e4 | 4148 | { |
dd41f596 | 4149 | int retval, oldprio, oldpolicy = -1, on_rq; |
1da177e4 | 4150 | unsigned long flags; |
70b97a7f | 4151 | struct rq *rq; |
1da177e4 | 4152 | |
66e5393a SR |
4153 | /* may grab non-irq protected spin_locks */ |
4154 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4155 | recheck: |
4156 | /* double check policy once rq lock held */ | |
4157 | if (policy < 0) | |
4158 | policy = oldpolicy = p->policy; | |
4159 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4160 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4161 | policy != SCHED_IDLE) | |
b0a9499c | 4162 | return -EINVAL; |
1da177e4 LT |
4163 | /* |
4164 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4165 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4166 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4167 | */ |
4168 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4169 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4170 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4171 | return -EINVAL; |
e05606d3 | 4172 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4173 | return -EINVAL; |
4174 | ||
37e4ab3f OC |
4175 | /* |
4176 | * Allow unprivileged RT tasks to decrease priority: | |
4177 | */ | |
4178 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4179 | if (rt_policy(policy)) { |
8dc3e909 | 4180 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4181 | |
4182 | if (!lock_task_sighand(p, &flags)) | |
4183 | return -ESRCH; | |
4184 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4185 | unlock_task_sighand(p, &flags); | |
4186 | ||
4187 | /* can't set/change the rt policy */ | |
4188 | if (policy != p->policy && !rlim_rtprio) | |
4189 | return -EPERM; | |
4190 | ||
4191 | /* can't increase priority */ | |
4192 | if (param->sched_priority > p->rt_priority && | |
4193 | param->sched_priority > rlim_rtprio) | |
4194 | return -EPERM; | |
4195 | } | |
dd41f596 IM |
4196 | /* |
4197 | * Like positive nice levels, dont allow tasks to | |
4198 | * move out of SCHED_IDLE either: | |
4199 | */ | |
4200 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4201 | return -EPERM; | |
5fe1d75f | 4202 | |
37e4ab3f OC |
4203 | /* can't change other user's priorities */ |
4204 | if ((current->euid != p->euid) && | |
4205 | (current->euid != p->uid)) | |
4206 | return -EPERM; | |
4207 | } | |
1da177e4 LT |
4208 | |
4209 | retval = security_task_setscheduler(p, policy, param); | |
4210 | if (retval) | |
4211 | return retval; | |
b29739f9 IM |
4212 | /* |
4213 | * make sure no PI-waiters arrive (or leave) while we are | |
4214 | * changing the priority of the task: | |
4215 | */ | |
4216 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4217 | /* |
4218 | * To be able to change p->policy safely, the apropriate | |
4219 | * runqueue lock must be held. | |
4220 | */ | |
b29739f9 | 4221 | rq = __task_rq_lock(p); |
1da177e4 LT |
4222 | /* recheck policy now with rq lock held */ |
4223 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4224 | policy = oldpolicy = -1; | |
b29739f9 IM |
4225 | __task_rq_unlock(rq); |
4226 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4227 | goto recheck; |
4228 | } | |
2daa3577 | 4229 | update_rq_clock(rq); |
dd41f596 | 4230 | on_rq = p->se.on_rq; |
2daa3577 | 4231 | if (on_rq) |
2e1cb74a | 4232 | deactivate_task(rq, p, 0); |
1da177e4 | 4233 | oldprio = p->prio; |
dd41f596 IM |
4234 | __setscheduler(rq, p, policy, param->sched_priority); |
4235 | if (on_rq) { | |
4236 | activate_task(rq, p, 0); | |
1da177e4 LT |
4237 | /* |
4238 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4239 | * our priority decreased, or if we are not currently running on |
4240 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4241 | */ |
d5f9f942 AM |
4242 | if (task_running(rq, p)) { |
4243 | if (p->prio > oldprio) | |
4244 | resched_task(rq->curr); | |
dd41f596 IM |
4245 | } else { |
4246 | check_preempt_curr(rq, p); | |
4247 | } | |
1da177e4 | 4248 | } |
b29739f9 IM |
4249 | __task_rq_unlock(rq); |
4250 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4251 | ||
95e02ca9 TG |
4252 | rt_mutex_adjust_pi(p); |
4253 | ||
1da177e4 LT |
4254 | return 0; |
4255 | } | |
4256 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4257 | ||
95cdf3b7 IM |
4258 | static int |
4259 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4260 | { |
1da177e4 LT |
4261 | struct sched_param lparam; |
4262 | struct task_struct *p; | |
36c8b586 | 4263 | int retval; |
1da177e4 LT |
4264 | |
4265 | if (!param || pid < 0) | |
4266 | return -EINVAL; | |
4267 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4268 | return -EFAULT; | |
5fe1d75f ON |
4269 | |
4270 | rcu_read_lock(); | |
4271 | retval = -ESRCH; | |
1da177e4 | 4272 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4273 | if (p != NULL) |
4274 | retval = sched_setscheduler(p, policy, &lparam); | |
4275 | rcu_read_unlock(); | |
36c8b586 | 4276 | |
1da177e4 LT |
4277 | return retval; |
4278 | } | |
4279 | ||
4280 | /** | |
4281 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4282 | * @pid: the pid in question. | |
4283 | * @policy: new policy. | |
4284 | * @param: structure containing the new RT priority. | |
4285 | */ | |
4286 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4287 | struct sched_param __user *param) | |
4288 | { | |
c21761f1 JB |
4289 | /* negative values for policy are not valid */ |
4290 | if (policy < 0) | |
4291 | return -EINVAL; | |
4292 | ||
1da177e4 LT |
4293 | return do_sched_setscheduler(pid, policy, param); |
4294 | } | |
4295 | ||
4296 | /** | |
4297 | * sys_sched_setparam - set/change the RT priority of a thread | |
4298 | * @pid: the pid in question. | |
4299 | * @param: structure containing the new RT priority. | |
4300 | */ | |
4301 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4302 | { | |
4303 | return do_sched_setscheduler(pid, -1, param); | |
4304 | } | |
4305 | ||
4306 | /** | |
4307 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4308 | * @pid: the pid in question. | |
4309 | */ | |
4310 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4311 | { | |
36c8b586 | 4312 | struct task_struct *p; |
1da177e4 | 4313 | int retval = -EINVAL; |
1da177e4 LT |
4314 | |
4315 | if (pid < 0) | |
4316 | goto out_nounlock; | |
4317 | ||
4318 | retval = -ESRCH; | |
4319 | read_lock(&tasklist_lock); | |
4320 | p = find_process_by_pid(pid); | |
4321 | if (p) { | |
4322 | retval = security_task_getscheduler(p); | |
4323 | if (!retval) | |
4324 | retval = p->policy; | |
4325 | } | |
4326 | read_unlock(&tasklist_lock); | |
4327 | ||
4328 | out_nounlock: | |
4329 | return retval; | |
4330 | } | |
4331 | ||
4332 | /** | |
4333 | * sys_sched_getscheduler - get the RT priority of a thread | |
4334 | * @pid: the pid in question. | |
4335 | * @param: structure containing the RT priority. | |
4336 | */ | |
4337 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4338 | { | |
4339 | struct sched_param lp; | |
36c8b586 | 4340 | struct task_struct *p; |
1da177e4 | 4341 | int retval = -EINVAL; |
1da177e4 LT |
4342 | |
4343 | if (!param || pid < 0) | |
4344 | goto out_nounlock; | |
4345 | ||
4346 | read_lock(&tasklist_lock); | |
4347 | p = find_process_by_pid(pid); | |
4348 | retval = -ESRCH; | |
4349 | if (!p) | |
4350 | goto out_unlock; | |
4351 | ||
4352 | retval = security_task_getscheduler(p); | |
4353 | if (retval) | |
4354 | goto out_unlock; | |
4355 | ||
4356 | lp.sched_priority = p->rt_priority; | |
4357 | read_unlock(&tasklist_lock); | |
4358 | ||
4359 | /* | |
4360 | * This one might sleep, we cannot do it with a spinlock held ... | |
4361 | */ | |
4362 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4363 | ||
4364 | out_nounlock: | |
4365 | return retval; | |
4366 | ||
4367 | out_unlock: | |
4368 | read_unlock(&tasklist_lock); | |
4369 | return retval; | |
4370 | } | |
4371 | ||
4372 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4373 | { | |
1da177e4 | 4374 | cpumask_t cpus_allowed; |
36c8b586 IM |
4375 | struct task_struct *p; |
4376 | int retval; | |
1da177e4 | 4377 | |
5be9361c | 4378 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4379 | read_lock(&tasklist_lock); |
4380 | ||
4381 | p = find_process_by_pid(pid); | |
4382 | if (!p) { | |
4383 | read_unlock(&tasklist_lock); | |
5be9361c | 4384 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4385 | return -ESRCH; |
4386 | } | |
4387 | ||
4388 | /* | |
4389 | * It is not safe to call set_cpus_allowed with the | |
4390 | * tasklist_lock held. We will bump the task_struct's | |
4391 | * usage count and then drop tasklist_lock. | |
4392 | */ | |
4393 | get_task_struct(p); | |
4394 | read_unlock(&tasklist_lock); | |
4395 | ||
4396 | retval = -EPERM; | |
4397 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4398 | !capable(CAP_SYS_NICE)) | |
4399 | goto out_unlock; | |
4400 | ||
e7834f8f DQ |
4401 | retval = security_task_setscheduler(p, 0, NULL); |
4402 | if (retval) | |
4403 | goto out_unlock; | |
4404 | ||
1da177e4 LT |
4405 | cpus_allowed = cpuset_cpus_allowed(p); |
4406 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4407 | retval = set_cpus_allowed(p, new_mask); | |
4408 | ||
4409 | out_unlock: | |
4410 | put_task_struct(p); | |
5be9361c | 4411 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4412 | return retval; |
4413 | } | |
4414 | ||
4415 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4416 | cpumask_t *new_mask) | |
4417 | { | |
4418 | if (len < sizeof(cpumask_t)) { | |
4419 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4420 | } else if (len > sizeof(cpumask_t)) { | |
4421 | len = sizeof(cpumask_t); | |
4422 | } | |
4423 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4424 | } | |
4425 | ||
4426 | /** | |
4427 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4428 | * @pid: pid of the process | |
4429 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4430 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4431 | */ | |
4432 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4433 | unsigned long __user *user_mask_ptr) | |
4434 | { | |
4435 | cpumask_t new_mask; | |
4436 | int retval; | |
4437 | ||
4438 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4439 | if (retval) | |
4440 | return retval; | |
4441 | ||
4442 | return sched_setaffinity(pid, new_mask); | |
4443 | } | |
4444 | ||
4445 | /* | |
4446 | * Represents all cpu's present in the system | |
4447 | * In systems capable of hotplug, this map could dynamically grow | |
4448 | * as new cpu's are detected in the system via any platform specific | |
4449 | * method, such as ACPI for e.g. | |
4450 | */ | |
4451 | ||
4cef0c61 | 4452 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4453 | EXPORT_SYMBOL(cpu_present_map); |
4454 | ||
4455 | #ifndef CONFIG_SMP | |
4cef0c61 | 4456 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4457 | EXPORT_SYMBOL(cpu_online_map); |
4458 | ||
4cef0c61 | 4459 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4460 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4461 | #endif |
4462 | ||
4463 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4464 | { | |
36c8b586 | 4465 | struct task_struct *p; |
1da177e4 | 4466 | int retval; |
1da177e4 | 4467 | |
5be9361c | 4468 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4469 | read_lock(&tasklist_lock); |
4470 | ||
4471 | retval = -ESRCH; | |
4472 | p = find_process_by_pid(pid); | |
4473 | if (!p) | |
4474 | goto out_unlock; | |
4475 | ||
e7834f8f DQ |
4476 | retval = security_task_getscheduler(p); |
4477 | if (retval) | |
4478 | goto out_unlock; | |
4479 | ||
2f7016d9 | 4480 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4481 | |
4482 | out_unlock: | |
4483 | read_unlock(&tasklist_lock); | |
5be9361c | 4484 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4485 | |
9531b62f | 4486 | return retval; |
1da177e4 LT |
4487 | } |
4488 | ||
4489 | /** | |
4490 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4491 | * @pid: pid of the process | |
4492 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4493 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4494 | */ | |
4495 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4496 | unsigned long __user *user_mask_ptr) | |
4497 | { | |
4498 | int ret; | |
4499 | cpumask_t mask; | |
4500 | ||
4501 | if (len < sizeof(cpumask_t)) | |
4502 | return -EINVAL; | |
4503 | ||
4504 | ret = sched_getaffinity(pid, &mask); | |
4505 | if (ret < 0) | |
4506 | return ret; | |
4507 | ||
4508 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4509 | return -EFAULT; | |
4510 | ||
4511 | return sizeof(cpumask_t); | |
4512 | } | |
4513 | ||
4514 | /** | |
4515 | * sys_sched_yield - yield the current processor to other threads. | |
4516 | * | |
dd41f596 IM |
4517 | * This function yields the current CPU to other tasks. If there are no |
4518 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4519 | */ |
4520 | asmlinkage long sys_sched_yield(void) | |
4521 | { | |
70b97a7f | 4522 | struct rq *rq = this_rq_lock(); |
1da177e4 LT |
4523 | |
4524 | schedstat_inc(rq, yld_cnt); | |
1799e35d | 4525 | current->sched_class->yield_task(rq, current); |
1da177e4 LT |
4526 | |
4527 | /* | |
4528 | * Since we are going to call schedule() anyway, there's | |
4529 | * no need to preempt or enable interrupts: | |
4530 | */ | |
4531 | __release(rq->lock); | |
8a25d5de | 4532 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4533 | _raw_spin_unlock(&rq->lock); |
4534 | preempt_enable_no_resched(); | |
4535 | ||
4536 | schedule(); | |
4537 | ||
4538 | return 0; | |
4539 | } | |
4540 | ||
e7b38404 | 4541 | static void __cond_resched(void) |
1da177e4 | 4542 | { |
8e0a43d8 IM |
4543 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4544 | __might_sleep(__FILE__, __LINE__); | |
4545 | #endif | |
5bbcfd90 IM |
4546 | /* |
4547 | * The BKS might be reacquired before we have dropped | |
4548 | * PREEMPT_ACTIVE, which could trigger a second | |
4549 | * cond_resched() call. | |
4550 | */ | |
1da177e4 LT |
4551 | do { |
4552 | add_preempt_count(PREEMPT_ACTIVE); | |
4553 | schedule(); | |
4554 | sub_preempt_count(PREEMPT_ACTIVE); | |
4555 | } while (need_resched()); | |
4556 | } | |
4557 | ||
4558 | int __sched cond_resched(void) | |
4559 | { | |
9414232f IM |
4560 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4561 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4562 | __cond_resched(); |
4563 | return 1; | |
4564 | } | |
4565 | return 0; | |
4566 | } | |
1da177e4 LT |
4567 | EXPORT_SYMBOL(cond_resched); |
4568 | ||
4569 | /* | |
4570 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4571 | * call schedule, and on return reacquire the lock. | |
4572 | * | |
4573 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4574 | * operations here to prevent schedule() from being called twice (once via | |
4575 | * spin_unlock(), once by hand). | |
4576 | */ | |
95cdf3b7 | 4577 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4578 | { |
6df3cecb JK |
4579 | int ret = 0; |
4580 | ||
1da177e4 LT |
4581 | if (need_lockbreak(lock)) { |
4582 | spin_unlock(lock); | |
4583 | cpu_relax(); | |
6df3cecb | 4584 | ret = 1; |
1da177e4 LT |
4585 | spin_lock(lock); |
4586 | } | |
9414232f | 4587 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4588 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4589 | _raw_spin_unlock(lock); |
4590 | preempt_enable_no_resched(); | |
4591 | __cond_resched(); | |
6df3cecb | 4592 | ret = 1; |
1da177e4 | 4593 | spin_lock(lock); |
1da177e4 | 4594 | } |
6df3cecb | 4595 | return ret; |
1da177e4 | 4596 | } |
1da177e4 LT |
4597 | EXPORT_SYMBOL(cond_resched_lock); |
4598 | ||
4599 | int __sched cond_resched_softirq(void) | |
4600 | { | |
4601 | BUG_ON(!in_softirq()); | |
4602 | ||
9414232f | 4603 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4604 | local_bh_enable(); |
1da177e4 LT |
4605 | __cond_resched(); |
4606 | local_bh_disable(); | |
4607 | return 1; | |
4608 | } | |
4609 | return 0; | |
4610 | } | |
1da177e4 LT |
4611 | EXPORT_SYMBOL(cond_resched_softirq); |
4612 | ||
1da177e4 LT |
4613 | /** |
4614 | * yield - yield the current processor to other threads. | |
4615 | * | |
72fd4a35 | 4616 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4617 | * thread runnable and calls sys_sched_yield(). |
4618 | */ | |
4619 | void __sched yield(void) | |
4620 | { | |
4621 | set_current_state(TASK_RUNNING); | |
4622 | sys_sched_yield(); | |
4623 | } | |
1da177e4 LT |
4624 | EXPORT_SYMBOL(yield); |
4625 | ||
4626 | /* | |
4627 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4628 | * that process accounting knows that this is a task in IO wait state. | |
4629 | * | |
4630 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4631 | * has set its backing_dev_info: the queue against which it should throttle) | |
4632 | */ | |
4633 | void __sched io_schedule(void) | |
4634 | { | |
70b97a7f | 4635 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4636 | |
0ff92245 | 4637 | delayacct_blkio_start(); |
1da177e4 LT |
4638 | atomic_inc(&rq->nr_iowait); |
4639 | schedule(); | |
4640 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4641 | delayacct_blkio_end(); |
1da177e4 | 4642 | } |
1da177e4 LT |
4643 | EXPORT_SYMBOL(io_schedule); |
4644 | ||
4645 | long __sched io_schedule_timeout(long timeout) | |
4646 | { | |
70b97a7f | 4647 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4648 | long ret; |
4649 | ||
0ff92245 | 4650 | delayacct_blkio_start(); |
1da177e4 LT |
4651 | atomic_inc(&rq->nr_iowait); |
4652 | ret = schedule_timeout(timeout); | |
4653 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4654 | delayacct_blkio_end(); |
1da177e4 LT |
4655 | return ret; |
4656 | } | |
4657 | ||
4658 | /** | |
4659 | * sys_sched_get_priority_max - return maximum RT priority. | |
4660 | * @policy: scheduling class. | |
4661 | * | |
4662 | * this syscall returns the maximum rt_priority that can be used | |
4663 | * by a given scheduling class. | |
4664 | */ | |
4665 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4666 | { | |
4667 | int ret = -EINVAL; | |
4668 | ||
4669 | switch (policy) { | |
4670 | case SCHED_FIFO: | |
4671 | case SCHED_RR: | |
4672 | ret = MAX_USER_RT_PRIO-1; | |
4673 | break; | |
4674 | case SCHED_NORMAL: | |
b0a9499c | 4675 | case SCHED_BATCH: |
dd41f596 | 4676 | case SCHED_IDLE: |
1da177e4 LT |
4677 | ret = 0; |
4678 | break; | |
4679 | } | |
4680 | return ret; | |
4681 | } | |
4682 | ||
4683 | /** | |
4684 | * sys_sched_get_priority_min - return minimum RT priority. | |
4685 | * @policy: scheduling class. | |
4686 | * | |
4687 | * this syscall returns the minimum rt_priority that can be used | |
4688 | * by a given scheduling class. | |
4689 | */ | |
4690 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4691 | { | |
4692 | int ret = -EINVAL; | |
4693 | ||
4694 | switch (policy) { | |
4695 | case SCHED_FIFO: | |
4696 | case SCHED_RR: | |
4697 | ret = 1; | |
4698 | break; | |
4699 | case SCHED_NORMAL: | |
b0a9499c | 4700 | case SCHED_BATCH: |
dd41f596 | 4701 | case SCHED_IDLE: |
1da177e4 LT |
4702 | ret = 0; |
4703 | } | |
4704 | return ret; | |
4705 | } | |
4706 | ||
4707 | /** | |
4708 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4709 | * @pid: pid of the process. | |
4710 | * @interval: userspace pointer to the timeslice value. | |
4711 | * | |
4712 | * this syscall writes the default timeslice value of a given process | |
4713 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4714 | */ | |
4715 | asmlinkage | |
4716 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4717 | { | |
36c8b586 | 4718 | struct task_struct *p; |
1da177e4 LT |
4719 | int retval = -EINVAL; |
4720 | struct timespec t; | |
1da177e4 LT |
4721 | |
4722 | if (pid < 0) | |
4723 | goto out_nounlock; | |
4724 | ||
4725 | retval = -ESRCH; | |
4726 | read_lock(&tasklist_lock); | |
4727 | p = find_process_by_pid(pid); | |
4728 | if (!p) | |
4729 | goto out_unlock; | |
4730 | ||
4731 | retval = security_task_getscheduler(p); | |
4732 | if (retval) | |
4733 | goto out_unlock; | |
4734 | ||
b78709cf | 4735 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
dd41f596 | 4736 | 0 : static_prio_timeslice(p->static_prio), &t); |
1da177e4 LT |
4737 | read_unlock(&tasklist_lock); |
4738 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; | |
4739 | out_nounlock: | |
4740 | return retval; | |
4741 | out_unlock: | |
4742 | read_unlock(&tasklist_lock); | |
4743 | return retval; | |
4744 | } | |
4745 | ||
2ed6e34f | 4746 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4747 | |
4748 | static void show_task(struct task_struct *p) | |
1da177e4 | 4749 | { |
1da177e4 | 4750 | unsigned long free = 0; |
36c8b586 | 4751 | unsigned state; |
1da177e4 | 4752 | |
1da177e4 | 4753 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4754 | printk("%-13.13s %c", p->comm, |
4755 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4756 | #if BITS_PER_LONG == 32 |
1da177e4 | 4757 | if (state == TASK_RUNNING) |
4bd77321 | 4758 | printk(" running "); |
1da177e4 | 4759 | else |
4bd77321 | 4760 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4761 | #else |
4762 | if (state == TASK_RUNNING) | |
4bd77321 | 4763 | printk(" running task "); |
1da177e4 LT |
4764 | else |
4765 | printk(" %016lx ", thread_saved_pc(p)); | |
4766 | #endif | |
4767 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4768 | { | |
10ebffde | 4769 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4770 | while (!*n) |
4771 | n++; | |
10ebffde | 4772 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4773 | } |
4774 | #endif | |
4bd77321 | 4775 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4776 | |
4777 | if (state != TASK_RUNNING) | |
4778 | show_stack(p, NULL); | |
4779 | } | |
4780 | ||
e59e2ae2 | 4781 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4782 | { |
36c8b586 | 4783 | struct task_struct *g, *p; |
1da177e4 | 4784 | |
4bd77321 IM |
4785 | #if BITS_PER_LONG == 32 |
4786 | printk(KERN_INFO | |
4787 | " task PC stack pid father\n"); | |
1da177e4 | 4788 | #else |
4bd77321 IM |
4789 | printk(KERN_INFO |
4790 | " task PC stack pid father\n"); | |
1da177e4 LT |
4791 | #endif |
4792 | read_lock(&tasklist_lock); | |
4793 | do_each_thread(g, p) { | |
4794 | /* | |
4795 | * reset the NMI-timeout, listing all files on a slow | |
4796 | * console might take alot of time: | |
4797 | */ | |
4798 | touch_nmi_watchdog(); | |
39bc89fd | 4799 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4800 | show_task(p); |
1da177e4 LT |
4801 | } while_each_thread(g, p); |
4802 | ||
04c9167f JF |
4803 | touch_all_softlockup_watchdogs(); |
4804 | ||
dd41f596 IM |
4805 | #ifdef CONFIG_SCHED_DEBUG |
4806 | sysrq_sched_debug_show(); | |
4807 | #endif | |
1da177e4 | 4808 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4809 | /* |
4810 | * Only show locks if all tasks are dumped: | |
4811 | */ | |
4812 | if (state_filter == -1) | |
4813 | debug_show_all_locks(); | |
1da177e4 LT |
4814 | } |
4815 | ||
1df21055 IM |
4816 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4817 | { | |
dd41f596 | 4818 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4819 | } |
4820 | ||
f340c0d1 IM |
4821 | /** |
4822 | * init_idle - set up an idle thread for a given CPU | |
4823 | * @idle: task in question | |
4824 | * @cpu: cpu the idle task belongs to | |
4825 | * | |
4826 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4827 | * flag, to make booting more robust. | |
4828 | */ | |
5c1e1767 | 4829 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4830 | { |
70b97a7f | 4831 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4832 | unsigned long flags; |
4833 | ||
dd41f596 IM |
4834 | __sched_fork(idle); |
4835 | idle->se.exec_start = sched_clock(); | |
4836 | ||
b29739f9 | 4837 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4838 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4839 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4840 | |
4841 | spin_lock_irqsave(&rq->lock, flags); | |
4842 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4843 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4844 | idle->oncpu = 1; | |
4845 | #endif | |
1da177e4 LT |
4846 | spin_unlock_irqrestore(&rq->lock, flags); |
4847 | ||
4848 | /* Set the preempt count _outside_ the spinlocks! */ | |
4849 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4850 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4851 | #else |
a1261f54 | 4852 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4853 | #endif |
dd41f596 IM |
4854 | /* |
4855 | * The idle tasks have their own, simple scheduling class: | |
4856 | */ | |
4857 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4858 | } |
4859 | ||
4860 | /* | |
4861 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4862 | * indicates which cpus entered this state. This is used | |
4863 | * in the rcu update to wait only for active cpus. For system | |
4864 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4865 | * always be CPU_MASK_NONE. | |
4866 | */ | |
4867 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4868 | ||
4869 | #ifdef CONFIG_SMP | |
4870 | /* | |
4871 | * This is how migration works: | |
4872 | * | |
70b97a7f | 4873 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4874 | * runqueue and wake up that CPU's migration thread. |
4875 | * 2) we down() the locked semaphore => thread blocks. | |
4876 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4877 | * thread off the CPU) | |
4878 | * 4) it gets the migration request and checks whether the migrated | |
4879 | * task is still in the wrong runqueue. | |
4880 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4881 | * it and puts it into the right queue. | |
4882 | * 6) migration thread up()s the semaphore. | |
4883 | * 7) we wake up and the migration is done. | |
4884 | */ | |
4885 | ||
4886 | /* | |
4887 | * Change a given task's CPU affinity. Migrate the thread to a | |
4888 | * proper CPU and schedule it away if the CPU it's executing on | |
4889 | * is removed from the allowed bitmask. | |
4890 | * | |
4891 | * NOTE: the caller must have a valid reference to the task, the | |
4892 | * task must not exit() & deallocate itself prematurely. The | |
4893 | * call is not atomic; no spinlocks may be held. | |
4894 | */ | |
36c8b586 | 4895 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4896 | { |
70b97a7f | 4897 | struct migration_req req; |
1da177e4 | 4898 | unsigned long flags; |
70b97a7f | 4899 | struct rq *rq; |
48f24c4d | 4900 | int ret = 0; |
1da177e4 LT |
4901 | |
4902 | rq = task_rq_lock(p, &flags); | |
4903 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4904 | ret = -EINVAL; | |
4905 | goto out; | |
4906 | } | |
4907 | ||
4908 | p->cpus_allowed = new_mask; | |
4909 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4910 | if (cpu_isset(task_cpu(p), new_mask)) | |
4911 | goto out; | |
4912 | ||
4913 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4914 | /* Need help from migration thread: drop lock and wait. */ | |
4915 | task_rq_unlock(rq, &flags); | |
4916 | wake_up_process(rq->migration_thread); | |
4917 | wait_for_completion(&req.done); | |
4918 | tlb_migrate_finish(p->mm); | |
4919 | return 0; | |
4920 | } | |
4921 | out: | |
4922 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4923 | |
1da177e4 LT |
4924 | return ret; |
4925 | } | |
1da177e4 LT |
4926 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4927 | ||
4928 | /* | |
4929 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4930 | * this because either it can't run here any more (set_cpus_allowed() | |
4931 | * away from this CPU, or CPU going down), or because we're | |
4932 | * attempting to rebalance this task on exec (sched_exec). | |
4933 | * | |
4934 | * So we race with normal scheduler movements, but that's OK, as long | |
4935 | * as the task is no longer on this CPU. | |
efc30814 KK |
4936 | * |
4937 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4938 | */ |
efc30814 | 4939 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4940 | { |
70b97a7f | 4941 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4942 | int ret = 0, on_rq; |
1da177e4 LT |
4943 | |
4944 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4945 | return ret; |
1da177e4 LT |
4946 | |
4947 | rq_src = cpu_rq(src_cpu); | |
4948 | rq_dest = cpu_rq(dest_cpu); | |
4949 | ||
4950 | double_rq_lock(rq_src, rq_dest); | |
4951 | /* Already moved. */ | |
4952 | if (task_cpu(p) != src_cpu) | |
4953 | goto out; | |
4954 | /* Affinity changed (again). */ | |
4955 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4956 | goto out; | |
4957 | ||
dd41f596 | 4958 | on_rq = p->se.on_rq; |
6e82a3be | 4959 | if (on_rq) |
2e1cb74a | 4960 | deactivate_task(rq_src, p, 0); |
6e82a3be | 4961 | |
1da177e4 | 4962 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4963 | if (on_rq) { |
4964 | activate_task(rq_dest, p, 0); | |
4965 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4966 | } |
efc30814 | 4967 | ret = 1; |
1da177e4 LT |
4968 | out: |
4969 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4970 | return ret; |
1da177e4 LT |
4971 | } |
4972 | ||
4973 | /* | |
4974 | * migration_thread - this is a highprio system thread that performs | |
4975 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4976 | * another runqueue. | |
4977 | */ | |
95cdf3b7 | 4978 | static int migration_thread(void *data) |
1da177e4 | 4979 | { |
1da177e4 | 4980 | int cpu = (long)data; |
70b97a7f | 4981 | struct rq *rq; |
1da177e4 LT |
4982 | |
4983 | rq = cpu_rq(cpu); | |
4984 | BUG_ON(rq->migration_thread != current); | |
4985 | ||
4986 | set_current_state(TASK_INTERRUPTIBLE); | |
4987 | while (!kthread_should_stop()) { | |
70b97a7f | 4988 | struct migration_req *req; |
1da177e4 | 4989 | struct list_head *head; |
1da177e4 | 4990 | |
1da177e4 LT |
4991 | spin_lock_irq(&rq->lock); |
4992 | ||
4993 | if (cpu_is_offline(cpu)) { | |
4994 | spin_unlock_irq(&rq->lock); | |
4995 | goto wait_to_die; | |
4996 | } | |
4997 | ||
4998 | if (rq->active_balance) { | |
4999 | active_load_balance(rq, cpu); | |
5000 | rq->active_balance = 0; | |
5001 | } | |
5002 | ||
5003 | head = &rq->migration_queue; | |
5004 | ||
5005 | if (list_empty(head)) { | |
5006 | spin_unlock_irq(&rq->lock); | |
5007 | schedule(); | |
5008 | set_current_state(TASK_INTERRUPTIBLE); | |
5009 | continue; | |
5010 | } | |
70b97a7f | 5011 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5012 | list_del_init(head->next); |
5013 | ||
674311d5 NP |
5014 | spin_unlock(&rq->lock); |
5015 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5016 | local_irq_enable(); | |
1da177e4 LT |
5017 | |
5018 | complete(&req->done); | |
5019 | } | |
5020 | __set_current_state(TASK_RUNNING); | |
5021 | return 0; | |
5022 | ||
5023 | wait_to_die: | |
5024 | /* Wait for kthread_stop */ | |
5025 | set_current_state(TASK_INTERRUPTIBLE); | |
5026 | while (!kthread_should_stop()) { | |
5027 | schedule(); | |
5028 | set_current_state(TASK_INTERRUPTIBLE); | |
5029 | } | |
5030 | __set_current_state(TASK_RUNNING); | |
5031 | return 0; | |
5032 | } | |
5033 | ||
5034 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5035 | /* |
5036 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5037 | * NOTE: interrupts should be disabled by the caller | |
5038 | */ | |
48f24c4d | 5039 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5040 | { |
efc30814 | 5041 | unsigned long flags; |
1da177e4 | 5042 | cpumask_t mask; |
70b97a7f IM |
5043 | struct rq *rq; |
5044 | int dest_cpu; | |
1da177e4 | 5045 | |
efc30814 | 5046 | restart: |
1da177e4 LT |
5047 | /* On same node? */ |
5048 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5049 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5050 | dest_cpu = any_online_cpu(mask); |
5051 | ||
5052 | /* On any allowed CPU? */ | |
5053 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5054 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5055 | |
5056 | /* No more Mr. Nice Guy. */ | |
5057 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5058 | rq = task_rq_lock(p, &flags); |
5059 | cpus_setall(p->cpus_allowed); | |
5060 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5061 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5062 | |
5063 | /* | |
5064 | * Don't tell them about moving exiting tasks or | |
5065 | * kernel threads (both mm NULL), since they never | |
5066 | * leave kernel. | |
5067 | */ | |
48f24c4d | 5068 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5069 | printk(KERN_INFO "process %d (%s) no " |
5070 | "longer affine to cpu%d\n", | |
48f24c4d | 5071 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5072 | } |
48f24c4d | 5073 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5074 | goto restart; |
1da177e4 LT |
5075 | } |
5076 | ||
5077 | /* | |
5078 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5079 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5080 | * for performance reasons the counter is not stricly tracking tasks to | |
5081 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5082 | * to keep the global sum constant after CPU-down: | |
5083 | */ | |
70b97a7f | 5084 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5085 | { |
70b97a7f | 5086 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5087 | unsigned long flags; |
5088 | ||
5089 | local_irq_save(flags); | |
5090 | double_rq_lock(rq_src, rq_dest); | |
5091 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5092 | rq_src->nr_uninterruptible = 0; | |
5093 | double_rq_unlock(rq_src, rq_dest); | |
5094 | local_irq_restore(flags); | |
5095 | } | |
5096 | ||
5097 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5098 | static void migrate_live_tasks(int src_cpu) | |
5099 | { | |
48f24c4d | 5100 | struct task_struct *p, *t; |
1da177e4 LT |
5101 | |
5102 | write_lock_irq(&tasklist_lock); | |
5103 | ||
48f24c4d IM |
5104 | do_each_thread(t, p) { |
5105 | if (p == current) | |
1da177e4 LT |
5106 | continue; |
5107 | ||
48f24c4d IM |
5108 | if (task_cpu(p) == src_cpu) |
5109 | move_task_off_dead_cpu(src_cpu, p); | |
5110 | } while_each_thread(t, p); | |
1da177e4 LT |
5111 | |
5112 | write_unlock_irq(&tasklist_lock); | |
5113 | } | |
5114 | ||
dd41f596 IM |
5115 | /* |
5116 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5117 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5118 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5119 | */ |
5120 | void sched_idle_next(void) | |
5121 | { | |
48f24c4d | 5122 | int this_cpu = smp_processor_id(); |
70b97a7f | 5123 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5124 | struct task_struct *p = rq->idle; |
5125 | unsigned long flags; | |
5126 | ||
5127 | /* cpu has to be offline */ | |
48f24c4d | 5128 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5129 | |
48f24c4d IM |
5130 | /* |
5131 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5132 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5133 | */ |
5134 | spin_lock_irqsave(&rq->lock, flags); | |
5135 | ||
dd41f596 | 5136 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5137 | |
5138 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5139 | activate_idle_task(p, rq); |
1da177e4 LT |
5140 | |
5141 | spin_unlock_irqrestore(&rq->lock, flags); | |
5142 | } | |
5143 | ||
48f24c4d IM |
5144 | /* |
5145 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5146 | * offline. |
5147 | */ | |
5148 | void idle_task_exit(void) | |
5149 | { | |
5150 | struct mm_struct *mm = current->active_mm; | |
5151 | ||
5152 | BUG_ON(cpu_online(smp_processor_id())); | |
5153 | ||
5154 | if (mm != &init_mm) | |
5155 | switch_mm(mm, &init_mm, current); | |
5156 | mmdrop(mm); | |
5157 | } | |
5158 | ||
054b9108 | 5159 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5160 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5161 | { |
70b97a7f | 5162 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5163 | |
5164 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5165 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5166 | |
5167 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5168 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5169 | |
48f24c4d | 5170 | get_task_struct(p); |
1da177e4 LT |
5171 | |
5172 | /* | |
5173 | * Drop lock around migration; if someone else moves it, | |
5174 | * that's OK. No task can be added to this CPU, so iteration is | |
5175 | * fine. | |
054b9108 | 5176 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5177 | */ |
054b9108 | 5178 | spin_unlock(&rq->lock); |
48f24c4d | 5179 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5180 | spin_lock(&rq->lock); |
1da177e4 | 5181 | |
48f24c4d | 5182 | put_task_struct(p); |
1da177e4 LT |
5183 | } |
5184 | ||
5185 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5186 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5187 | { | |
70b97a7f | 5188 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5189 | struct task_struct *next; |
48f24c4d | 5190 | |
dd41f596 IM |
5191 | for ( ; ; ) { |
5192 | if (!rq->nr_running) | |
5193 | break; | |
a8e504d2 | 5194 | update_rq_clock(rq); |
ff95f3df | 5195 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5196 | if (!next) |
5197 | break; | |
5198 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5199 | |
1da177e4 LT |
5200 | } |
5201 | } | |
5202 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5203 | ||
e692ab53 NP |
5204 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5205 | ||
5206 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5207 | { |
5208 | .procname = "sched_domain", | |
c57baf1e | 5209 | .mode = 0555, |
e0361851 | 5210 | }, |
e692ab53 NP |
5211 | {0,}, |
5212 | }; | |
5213 | ||
5214 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5215 | { |
c57baf1e | 5216 | .ctl_name = CTL_KERN, |
e0361851 | 5217 | .procname = "kernel", |
c57baf1e | 5218 | .mode = 0555, |
e0361851 AD |
5219 | .child = sd_ctl_dir, |
5220 | }, | |
e692ab53 NP |
5221 | {0,}, |
5222 | }; | |
5223 | ||
5224 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5225 | { | |
5226 | struct ctl_table *entry = | |
5227 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); | |
5228 | ||
5229 | BUG_ON(!entry); | |
5230 | memset(entry, 0, n * sizeof(struct ctl_table)); | |
5231 | ||
5232 | return entry; | |
5233 | } | |
5234 | ||
5235 | static void | |
e0361851 | 5236 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5237 | const char *procname, void *data, int maxlen, |
5238 | mode_t mode, proc_handler *proc_handler) | |
5239 | { | |
e692ab53 NP |
5240 | entry->procname = procname; |
5241 | entry->data = data; | |
5242 | entry->maxlen = maxlen; | |
5243 | entry->mode = mode; | |
5244 | entry->proc_handler = proc_handler; | |
5245 | } | |
5246 | ||
5247 | static struct ctl_table * | |
5248 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5249 | { | |
5250 | struct ctl_table *table = sd_alloc_ctl_entry(14); | |
5251 | ||
e0361851 | 5252 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5253 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5254 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5255 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5256 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5257 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5258 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5259 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5260 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5261 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5262 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5263 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5264 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5265 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5266 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5267 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5268 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5269 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5270 | set_table_entry(&table[10], "cache_nice_tries", |
e692ab53 NP |
5271 | &sd->cache_nice_tries, |
5272 | sizeof(int), 0644, proc_dointvec_minmax); | |
e0361851 | 5273 | set_table_entry(&table[12], "flags", &sd->flags, |
e692ab53 NP |
5274 | sizeof(int), 0644, proc_dointvec_minmax); |
5275 | ||
5276 | return table; | |
5277 | } | |
5278 | ||
5279 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | |
5280 | { | |
5281 | struct ctl_table *entry, *table; | |
5282 | struct sched_domain *sd; | |
5283 | int domain_num = 0, i; | |
5284 | char buf[32]; | |
5285 | ||
5286 | for_each_domain(cpu, sd) | |
5287 | domain_num++; | |
5288 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
5289 | ||
5290 | i = 0; | |
5291 | for_each_domain(cpu, sd) { | |
5292 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5293 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5294 | entry->mode = 0555; |
e692ab53 NP |
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); | |
e692ab53 | 5313 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5314 | entry->mode = 0555; |
e692ab53 NP |
5315 | entry->child = sd_alloc_ctl_cpu_table(i); |
5316 | } | |
5317 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5318 | } | |
5319 | #else | |
5320 | static void init_sched_domain_sysctl(void) | |
5321 | { | |
5322 | } | |
5323 | #endif | |
5324 | ||
1da177e4 LT |
5325 | /* |
5326 | * migration_call - callback that gets triggered when a CPU is added. | |
5327 | * Here we can start up the necessary migration thread for the new CPU. | |
5328 | */ | |
48f24c4d IM |
5329 | static int __cpuinit |
5330 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5331 | { |
1da177e4 | 5332 | struct task_struct *p; |
48f24c4d | 5333 | int cpu = (long)hcpu; |
1da177e4 | 5334 | unsigned long flags; |
70b97a7f | 5335 | struct rq *rq; |
1da177e4 LT |
5336 | |
5337 | switch (action) { | |
5be9361c GS |
5338 | case CPU_LOCK_ACQUIRE: |
5339 | mutex_lock(&sched_hotcpu_mutex); | |
5340 | break; | |
5341 | ||
1da177e4 | 5342 | case CPU_UP_PREPARE: |
8bb78442 | 5343 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5344 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5345 | if (IS_ERR(p)) |
5346 | return NOTIFY_BAD; | |
1da177e4 LT |
5347 | kthread_bind(p, cpu); |
5348 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5349 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5350 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5351 | task_rq_unlock(rq, &flags); |
5352 | cpu_rq(cpu)->migration_thread = p; | |
5353 | break; | |
48f24c4d | 5354 | |
1da177e4 | 5355 | case CPU_ONLINE: |
8bb78442 | 5356 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5357 | /* Strictly unneccessary, as first user will wake it. */ |
5358 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5359 | break; | |
48f24c4d | 5360 | |
1da177e4 LT |
5361 | #ifdef CONFIG_HOTPLUG_CPU |
5362 | case CPU_UP_CANCELED: | |
8bb78442 | 5363 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5364 | if (!cpu_rq(cpu)->migration_thread) |
5365 | break; | |
1da177e4 | 5366 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5367 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5368 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5369 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5370 | cpu_rq(cpu)->migration_thread = NULL; | |
5371 | break; | |
48f24c4d | 5372 | |
1da177e4 | 5373 | case CPU_DEAD: |
8bb78442 | 5374 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5375 | migrate_live_tasks(cpu); |
5376 | rq = cpu_rq(cpu); | |
5377 | kthread_stop(rq->migration_thread); | |
5378 | rq->migration_thread = NULL; | |
5379 | /* Idle task back to normal (off runqueue, low prio) */ | |
5380 | rq = task_rq_lock(rq->idle, &flags); | |
a8e504d2 | 5381 | update_rq_clock(rq); |
2e1cb74a | 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 | 6303 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6304 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
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 | ||
5c45bf27 SS |
6333 | #ifdef CONFIG_SCHED_MC |
6334 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6335 | { | |
6336 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6337 | } | |
48f24c4d IM |
6338 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6339 | const char *buf, size_t count) | |
5c45bf27 SS |
6340 | { |
6341 | return sched_power_savings_store(buf, count, 0); | |
6342 | } | |
6707de00 AB |
6343 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6344 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6345 | #endif |
6346 | ||
6347 | #ifdef CONFIG_SCHED_SMT | |
6348 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6349 | { | |
6350 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6351 | } | |
48f24c4d IM |
6352 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6353 | const char *buf, size_t count) | |
5c45bf27 SS |
6354 | { |
6355 | return sched_power_savings_store(buf, count, 1); | |
6356 | } | |
6707de00 AB |
6357 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6358 | sched_smt_power_savings_store); | |
6359 | #endif | |
6360 | ||
6361 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6362 | { | |
6363 | int err = 0; | |
6364 | ||
6365 | #ifdef CONFIG_SCHED_SMT | |
6366 | if (smt_capable()) | |
6367 | err = sysfs_create_file(&cls->kset.kobj, | |
6368 | &attr_sched_smt_power_savings.attr); | |
6369 | #endif | |
6370 | #ifdef CONFIG_SCHED_MC | |
6371 | if (!err && mc_capable()) | |
6372 | err = sysfs_create_file(&cls->kset.kobj, | |
6373 | &attr_sched_mc_power_savings.attr); | |
6374 | #endif | |
6375 | return err; | |
6376 | } | |
5c45bf27 SS |
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(); | |
1da177e4 LT |
6436 | } |
6437 | #else | |
6438 | void __init sched_init_smp(void) | |
6439 | { | |
6440 | } | |
6441 | #endif /* CONFIG_SMP */ | |
6442 | ||
6443 | int in_sched_functions(unsigned long addr) | |
6444 | { | |
6445 | /* Linker adds these: start and end of __sched functions */ | |
6446 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6447 | |
1da177e4 LT |
6448 | return in_lock_functions(addr) || |
6449 | (addr >= (unsigned long)__sched_text_start | |
6450 | && addr < (unsigned long)__sched_text_end); | |
6451 | } | |
6452 | ||
dd41f596 IM |
6453 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
6454 | { | |
6455 | cfs_rq->tasks_timeline = RB_ROOT; | |
6456 | cfs_rq->fair_clock = 1; | |
6457 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6458 | cfs_rq->rq = rq; | |
6459 | #endif | |
6460 | } | |
6461 | ||
1da177e4 LT |
6462 | void __init sched_init(void) |
6463 | { | |
476f3534 | 6464 | int highest_cpu = 0; |
dd41f596 IM |
6465 | int i, j; |
6466 | ||
6467 | /* | |
6468 | * Link up the scheduling class hierarchy: | |
6469 | */ | |
6470 | rt_sched_class.next = &fair_sched_class; | |
6471 | fair_sched_class.next = &idle_sched_class; | |
6472 | idle_sched_class.next = NULL; | |
1da177e4 | 6473 | |
0a945022 | 6474 | for_each_possible_cpu(i) { |
dd41f596 | 6475 | struct rt_prio_array *array; |
70b97a7f | 6476 | struct rq *rq; |
1da177e4 LT |
6477 | |
6478 | rq = cpu_rq(i); | |
6479 | spin_lock_init(&rq->lock); | |
fcb99371 | 6480 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6481 | rq->nr_running = 0; |
dd41f596 IM |
6482 | rq->clock = 1; |
6483 | init_cfs_rq(&rq->cfs, rq); | |
6484 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6485 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
6486 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6487 | #endif | |
1da177e4 | 6488 | |
dd41f596 IM |
6489 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6490 | rq->cpu_load[j] = 0; | |
1da177e4 | 6491 | #ifdef CONFIG_SMP |
41c7ce9a | 6492 | rq->sd = NULL; |
1da177e4 | 6493 | rq->active_balance = 0; |
dd41f596 | 6494 | rq->next_balance = jiffies; |
1da177e4 | 6495 | rq->push_cpu = 0; |
0a2966b4 | 6496 | rq->cpu = i; |
1da177e4 LT |
6497 | rq->migration_thread = NULL; |
6498 | INIT_LIST_HEAD(&rq->migration_queue); | |
6499 | #endif | |
6500 | atomic_set(&rq->nr_iowait, 0); | |
6501 | ||
dd41f596 IM |
6502 | array = &rq->rt.active; |
6503 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6504 | INIT_LIST_HEAD(array->queue + j); | |
6505 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6506 | } |
476f3534 | 6507 | highest_cpu = i; |
dd41f596 IM |
6508 | /* delimiter for bitsearch: */ |
6509 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6510 | } |
6511 | ||
2dd73a4f | 6512 | set_load_weight(&init_task); |
b50f60ce | 6513 | |
e107be36 AK |
6514 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6515 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6516 | #endif | |
6517 | ||
c9819f45 | 6518 | #ifdef CONFIG_SMP |
476f3534 | 6519 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6520 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6521 | #endif | |
6522 | ||
b50f60ce HC |
6523 | #ifdef CONFIG_RT_MUTEXES |
6524 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6525 | #endif | |
6526 | ||
1da177e4 LT |
6527 | /* |
6528 | * The boot idle thread does lazy MMU switching as well: | |
6529 | */ | |
6530 | atomic_inc(&init_mm.mm_count); | |
6531 | enter_lazy_tlb(&init_mm, current); | |
6532 | ||
6533 | /* | |
6534 | * Make us the idle thread. Technically, schedule() should not be | |
6535 | * called from this thread, however somewhere below it might be, | |
6536 | * but because we are the idle thread, we just pick up running again | |
6537 | * when this runqueue becomes "idle". | |
6538 | */ | |
6539 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6540 | /* |
6541 | * During early bootup we pretend to be a normal task: | |
6542 | */ | |
6543 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6544 | } |
6545 | ||
6546 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6547 | void __might_sleep(char *file, int line) | |
6548 | { | |
48f24c4d | 6549 | #ifdef in_atomic |
1da177e4 LT |
6550 | static unsigned long prev_jiffy; /* ratelimiting */ |
6551 | ||
6552 | if ((in_atomic() || irqs_disabled()) && | |
6553 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6554 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6555 | return; | |
6556 | prev_jiffy = jiffies; | |
91368d73 | 6557 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6558 | " context at %s:%d\n", file, line); |
6559 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6560 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6561 | debug_show_held_locks(current); |
3117df04 IM |
6562 | if (irqs_disabled()) |
6563 | print_irqtrace_events(current); | |
1da177e4 LT |
6564 | dump_stack(); |
6565 | } | |
6566 | #endif | |
6567 | } | |
6568 | EXPORT_SYMBOL(__might_sleep); | |
6569 | #endif | |
6570 | ||
6571 | #ifdef CONFIG_MAGIC_SYSRQ | |
6572 | void normalize_rt_tasks(void) | |
6573 | { | |
a0f98a1c | 6574 | struct task_struct *g, *p; |
1da177e4 | 6575 | unsigned long flags; |
70b97a7f | 6576 | struct rq *rq; |
dd41f596 | 6577 | int on_rq; |
1da177e4 LT |
6578 | |
6579 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6580 | do_each_thread(g, p) { |
dd41f596 IM |
6581 | p->se.fair_key = 0; |
6582 | p->se.wait_runtime = 0; | |
6cfb0d5d | 6583 | p->se.exec_start = 0; |
dd41f596 | 6584 | p->se.wait_start_fair = 0; |
6cfb0d5d IM |
6585 | p->se.sleep_start_fair = 0; |
6586 | #ifdef CONFIG_SCHEDSTATS | |
dd41f596 | 6587 | p->se.wait_start = 0; |
dd41f596 | 6588 | p->se.sleep_start = 0; |
dd41f596 | 6589 | p->se.block_start = 0; |
6cfb0d5d | 6590 | #endif |
dd41f596 IM |
6591 | task_rq(p)->cfs.fair_clock = 0; |
6592 | task_rq(p)->clock = 0; | |
6593 | ||
6594 | if (!rt_task(p)) { | |
6595 | /* | |
6596 | * Renice negative nice level userspace | |
6597 | * tasks back to 0: | |
6598 | */ | |
6599 | if (TASK_NICE(p) < 0 && p->mm) | |
6600 | set_user_nice(p, 0); | |
1da177e4 | 6601 | continue; |
dd41f596 | 6602 | } |
1da177e4 | 6603 | |
b29739f9 IM |
6604 | spin_lock_irqsave(&p->pi_lock, flags); |
6605 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6606 | #ifdef CONFIG_SMP |
6607 | /* | |
6608 | * Do not touch the migration thread: | |
6609 | */ | |
6610 | if (p == rq->migration_thread) | |
6611 | goto out_unlock; | |
6612 | #endif | |
1da177e4 | 6613 | |
2daa3577 | 6614 | update_rq_clock(rq); |
dd41f596 | 6615 | on_rq = p->se.on_rq; |
2daa3577 IM |
6616 | if (on_rq) |
6617 | deactivate_task(rq, p, 0); | |
dd41f596 IM |
6618 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
6619 | if (on_rq) { | |
2daa3577 | 6620 | activate_task(rq, p, 0); |
1da177e4 LT |
6621 | resched_task(rq->curr); |
6622 | } | |
dd41f596 IM |
6623 | #ifdef CONFIG_SMP |
6624 | out_unlock: | |
6625 | #endif | |
b29739f9 IM |
6626 | __task_rq_unlock(rq); |
6627 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6628 | } while_each_thread(g, p); |
6629 | ||
1da177e4 LT |
6630 | read_unlock_irq(&tasklist_lock); |
6631 | } | |
6632 | ||
6633 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6634 | |
6635 | #ifdef CONFIG_IA64 | |
6636 | /* | |
6637 | * These functions are only useful for the IA64 MCA handling. | |
6638 | * | |
6639 | * They can only be called when the whole system has been | |
6640 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6641 | * activity can take place. Using them for anything else would | |
6642 | * be a serious bug, and as a result, they aren't even visible | |
6643 | * under any other configuration. | |
6644 | */ | |
6645 | ||
6646 | /** | |
6647 | * curr_task - return the current task for a given cpu. | |
6648 | * @cpu: the processor in question. | |
6649 | * | |
6650 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6651 | */ | |
36c8b586 | 6652 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6653 | { |
6654 | return cpu_curr(cpu); | |
6655 | } | |
6656 | ||
6657 | /** | |
6658 | * set_curr_task - set the current task for a given cpu. | |
6659 | * @cpu: the processor in question. | |
6660 | * @p: the task pointer to set. | |
6661 | * | |
6662 | * Description: This function must only be used when non-maskable interrupts | |
6663 | * are serviced on a separate stack. It allows the architecture to switch the | |
6664 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6665 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6666 | * and caller must save the original value of the current task (see | |
6667 | * curr_task() above) and restore that value before reenabling interrupts and | |
6668 | * re-starting the system. | |
6669 | * | |
6670 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6671 | */ | |
36c8b586 | 6672 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6673 | { |
6674 | cpu_curr(cpu) = p; | |
6675 | } | |
6676 | ||
6677 | #endif |