]>
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 | */ | |
a4ec24b4 | 99 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
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 | * | |
a4ec24b4 | 108 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
109 | * Timeslices get refilled after they expire. |
110 | */ | |
1da177e4 | 111 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 112 | |
5517d86b ED |
113 | #ifdef CONFIG_SMP |
114 | /* | |
115 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
116 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
117 | */ | |
118 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
119 | { | |
120 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Each time a sched group cpu_power is changed, | |
125 | * we must compute its reciprocal value | |
126 | */ | |
127 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
128 | { | |
129 | sg->__cpu_power += val; | |
130 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
131 | } | |
132 | #endif | |
133 | ||
e05606d3 IM |
134 | static inline int rt_policy(int policy) |
135 | { | |
136 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
137 | return 1; | |
138 | return 0; | |
139 | } | |
140 | ||
141 | static inline int task_has_rt_policy(struct task_struct *p) | |
142 | { | |
143 | return rt_policy(p->policy); | |
144 | } | |
145 | ||
1da177e4 | 146 | /* |
6aa645ea | 147 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 148 | */ |
6aa645ea IM |
149 | struct rt_prio_array { |
150 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
151 | struct list_head queue[MAX_RT_PRIO]; | |
152 | }; | |
153 | ||
29f59db3 SV |
154 | #ifdef CONFIG_FAIR_GROUP_SCHED |
155 | ||
29f59db3 SV |
156 | struct cfs_rq; |
157 | ||
158 | /* task group related information */ | |
4cf86d77 | 159 | struct task_group { |
29f59db3 SV |
160 | /* schedulable entities of this group on each cpu */ |
161 | struct sched_entity **se; | |
162 | /* runqueue "owned" by this group on each cpu */ | |
163 | struct cfs_rq **cfs_rq; | |
164 | unsigned long shares; | |
165 | }; | |
166 | ||
167 | /* Default task group's sched entity on each cpu */ | |
168 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
169 | /* Default task group's cfs_rq on each cpu */ | |
170 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
171 | ||
9b5b7751 SV |
172 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
173 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
174 | |
175 | /* Default task group. | |
3a252015 | 176 | * Every task in system belong to this group at bootup. |
29f59db3 | 177 | */ |
4cf86d77 | 178 | struct task_group init_task_group = { |
3a252015 IM |
179 | .se = init_sched_entity_p, |
180 | .cfs_rq = init_cfs_rq_p, | |
181 | }; | |
9b5b7751 | 182 | |
24e377a8 | 183 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 184 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 185 | #else |
3a252015 | 186 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
187 | #endif |
188 | ||
4cf86d77 | 189 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
190 | |
191 | /* return group to which a task belongs */ | |
4cf86d77 | 192 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 193 | { |
4cf86d77 | 194 | struct task_group *tg; |
9b5b7751 | 195 | |
24e377a8 SV |
196 | #ifdef CONFIG_FAIR_USER_SCHED |
197 | tg = p->user->tg; | |
198 | #else | |
4cf86d77 | 199 | tg = &init_task_group; |
24e377a8 | 200 | #endif |
9b5b7751 SV |
201 | |
202 | return tg; | |
29f59db3 SV |
203 | } |
204 | ||
205 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
206 | static inline void set_task_cfs_rq(struct task_struct *p) | |
207 | { | |
4cf86d77 IM |
208 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
209 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
210 | } |
211 | ||
212 | #else | |
213 | ||
214 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
215 | ||
216 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
217 | ||
6aa645ea IM |
218 | /* CFS-related fields in a runqueue */ |
219 | struct cfs_rq { | |
220 | struct load_weight load; | |
221 | unsigned long nr_running; | |
222 | ||
6aa645ea | 223 | u64 exec_clock; |
e9acbff6 | 224 | u64 min_vruntime; |
6aa645ea IM |
225 | |
226 | struct rb_root tasks_timeline; | |
227 | struct rb_node *rb_leftmost; | |
228 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
229 | /* 'curr' points to currently running entity on this cfs_rq. |
230 | * It is set to NULL otherwise (i.e when none are currently running). | |
231 | */ | |
232 | struct sched_entity *curr; | |
ddc97297 PZ |
233 | |
234 | unsigned long nr_spread_over; | |
235 | ||
62160e3f | 236 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
237 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
238 | ||
239 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
240 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
241 | * (like users, containers etc.) | |
242 | * | |
243 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
244 | * list is used during load balance. | |
245 | */ | |
246 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 247 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 248 | struct rcu_head rcu; |
6aa645ea IM |
249 | #endif |
250 | }; | |
1da177e4 | 251 | |
6aa645ea IM |
252 | /* Real-Time classes' related field in a runqueue: */ |
253 | struct rt_rq { | |
254 | struct rt_prio_array active; | |
255 | int rt_load_balance_idx; | |
256 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
257 | }; | |
258 | ||
1da177e4 LT |
259 | /* |
260 | * This is the main, per-CPU runqueue data structure. | |
261 | * | |
262 | * Locking rule: those places that want to lock multiple runqueues | |
263 | * (such as the load balancing or the thread migration code), lock | |
264 | * acquire operations must be ordered by ascending &runqueue. | |
265 | */ | |
70b97a7f | 266 | struct rq { |
6aa645ea | 267 | spinlock_t lock; /* runqueue lock */ |
1da177e4 LT |
268 | |
269 | /* | |
270 | * nr_running and cpu_load should be in the same cacheline because | |
271 | * remote CPUs use both these fields when doing load calculation. | |
272 | */ | |
273 | unsigned long nr_running; | |
6aa645ea IM |
274 | #define CPU_LOAD_IDX_MAX 5 |
275 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 276 | unsigned char idle_at_tick; |
46cb4b7c SS |
277 | #ifdef CONFIG_NO_HZ |
278 | unsigned char in_nohz_recently; | |
279 | #endif | |
495eca49 | 280 | struct load_weight load; /* capture load from *all* tasks on this cpu */ |
6aa645ea IM |
281 | unsigned long nr_load_updates; |
282 | u64 nr_switches; | |
283 | ||
284 | struct cfs_rq cfs; | |
285 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
286 | struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */ | |
1da177e4 | 287 | #endif |
6aa645ea | 288 | struct rt_rq rt; |
1da177e4 LT |
289 | |
290 | /* | |
291 | * This is part of a global counter where only the total sum | |
292 | * over all CPUs matters. A task can increase this counter on | |
293 | * one CPU and if it got migrated afterwards it may decrease | |
294 | * it on another CPU. Always updated under the runqueue lock: | |
295 | */ | |
296 | unsigned long nr_uninterruptible; | |
297 | ||
36c8b586 | 298 | struct task_struct *curr, *idle; |
c9819f45 | 299 | unsigned long next_balance; |
1da177e4 | 300 | struct mm_struct *prev_mm; |
6aa645ea | 301 | |
6aa645ea IM |
302 | u64 clock, prev_clock_raw; |
303 | s64 clock_max_delta; | |
304 | ||
305 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
306 | u64 idle_clock; |
307 | unsigned int clock_deep_idle_events; | |
529c7726 | 308 | u64 tick_timestamp; |
6aa645ea | 309 | |
1da177e4 LT |
310 | atomic_t nr_iowait; |
311 | ||
312 | #ifdef CONFIG_SMP | |
313 | struct sched_domain *sd; | |
314 | ||
315 | /* For active balancing */ | |
316 | int active_balance; | |
317 | int push_cpu; | |
0a2966b4 | 318 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 319 | |
36c8b586 | 320 | struct task_struct *migration_thread; |
1da177e4 LT |
321 | struct list_head migration_queue; |
322 | #endif | |
323 | ||
324 | #ifdef CONFIG_SCHEDSTATS | |
325 | /* latency stats */ | |
326 | struct sched_info rq_sched_info; | |
327 | ||
328 | /* sys_sched_yield() stats */ | |
329 | unsigned long yld_exp_empty; | |
330 | unsigned long yld_act_empty; | |
331 | unsigned long yld_both_empty; | |
2d72376b | 332 | unsigned long yld_count; |
1da177e4 LT |
333 | |
334 | /* schedule() stats */ | |
335 | unsigned long sched_switch; | |
2d72376b | 336 | unsigned long sched_count; |
1da177e4 LT |
337 | unsigned long sched_goidle; |
338 | ||
339 | /* try_to_wake_up() stats */ | |
2d72376b | 340 | unsigned long ttwu_count; |
1da177e4 | 341 | unsigned long ttwu_local; |
b8efb561 IM |
342 | |
343 | /* BKL stats */ | |
2d72376b | 344 | unsigned long bkl_count; |
1da177e4 | 345 | #endif |
fcb99371 | 346 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
347 | }; |
348 | ||
f34e3b61 | 349 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 350 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 351 | |
dd41f596 IM |
352 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
353 | { | |
354 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
355 | } | |
356 | ||
0a2966b4 CL |
357 | static inline int cpu_of(struct rq *rq) |
358 | { | |
359 | #ifdef CONFIG_SMP | |
360 | return rq->cpu; | |
361 | #else | |
362 | return 0; | |
363 | #endif | |
364 | } | |
365 | ||
20d315d4 | 366 | /* |
b04a0f4c IM |
367 | * Update the per-runqueue clock, as finegrained as the platform can give |
368 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 369 | */ |
b04a0f4c | 370 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
371 | { |
372 | u64 prev_raw = rq->prev_clock_raw; | |
373 | u64 now = sched_clock(); | |
374 | s64 delta = now - prev_raw; | |
375 | u64 clock = rq->clock; | |
376 | ||
b04a0f4c IM |
377 | #ifdef CONFIG_SCHED_DEBUG |
378 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
379 | #endif | |
20d315d4 IM |
380 | /* |
381 | * Protect against sched_clock() occasionally going backwards: | |
382 | */ | |
383 | if (unlikely(delta < 0)) { | |
384 | clock++; | |
385 | rq->clock_warps++; | |
386 | } else { | |
387 | /* | |
388 | * Catch too large forward jumps too: | |
389 | */ | |
529c7726 IM |
390 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
391 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
392 | clock = rq->tick_timestamp + TICK_NSEC; | |
393 | else | |
394 | clock++; | |
20d315d4 IM |
395 | rq->clock_overflows++; |
396 | } else { | |
397 | if (unlikely(delta > rq->clock_max_delta)) | |
398 | rq->clock_max_delta = delta; | |
399 | clock += delta; | |
400 | } | |
401 | } | |
402 | ||
403 | rq->prev_clock_raw = now; | |
404 | rq->clock = clock; | |
b04a0f4c | 405 | } |
20d315d4 | 406 | |
b04a0f4c IM |
407 | static void update_rq_clock(struct rq *rq) |
408 | { | |
409 | if (likely(smp_processor_id() == cpu_of(rq))) | |
410 | __update_rq_clock(rq); | |
20d315d4 IM |
411 | } |
412 | ||
674311d5 NP |
413 | /* |
414 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 415 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
416 | * |
417 | * The domain tree of any CPU may only be accessed from within | |
418 | * preempt-disabled sections. | |
419 | */ | |
48f24c4d IM |
420 | #define for_each_domain(cpu, __sd) \ |
421 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
422 | |
423 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
424 | #define this_rq() (&__get_cpu_var(runqueues)) | |
425 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
426 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
427 | ||
bf5c91ba IM |
428 | /* |
429 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
430 | */ | |
431 | #ifdef CONFIG_SCHED_DEBUG | |
432 | # define const_debug __read_mostly | |
433 | #else | |
434 | # define const_debug static const | |
435 | #endif | |
436 | ||
437 | /* | |
438 | * Debugging: various feature bits | |
439 | */ | |
440 | enum { | |
bbdba7c0 IM |
441 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
442 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 443 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 444 | SCHED_FEAT_APPROX_AVG = 8, |
bf5c91ba IM |
445 | }; |
446 | ||
447 | const_debug unsigned int sysctl_sched_features = | |
bf5c91ba | 448 | SCHED_FEAT_NEW_FAIR_SLEEPERS *1 | |
94dfb5e7 | 449 | SCHED_FEAT_START_DEBIT *1 | |
06877c33 | 450 | SCHED_FEAT_TREE_AVG *0 | |
94dfb5e7 | 451 | SCHED_FEAT_APPROX_AVG *0; |
bf5c91ba IM |
452 | |
453 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
454 | ||
e436d800 IM |
455 | /* |
456 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
457 | * clock constructed from sched_clock(): | |
458 | */ | |
459 | unsigned long long cpu_clock(int cpu) | |
460 | { | |
e436d800 IM |
461 | unsigned long long now; |
462 | unsigned long flags; | |
b04a0f4c | 463 | struct rq *rq; |
e436d800 | 464 | |
2cd4d0ea | 465 | local_irq_save(flags); |
b04a0f4c IM |
466 | rq = cpu_rq(cpu); |
467 | update_rq_clock(rq); | |
468 | now = rq->clock; | |
2cd4d0ea | 469 | local_irq_restore(flags); |
e436d800 IM |
470 | |
471 | return now; | |
472 | } | |
473 | ||
1da177e4 | 474 | #ifndef prepare_arch_switch |
4866cde0 NP |
475 | # define prepare_arch_switch(next) do { } while (0) |
476 | #endif | |
477 | #ifndef finish_arch_switch | |
478 | # define finish_arch_switch(prev) do { } while (0) | |
479 | #endif | |
480 | ||
481 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 482 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
483 | { |
484 | return rq->curr == p; | |
485 | } | |
486 | ||
70b97a7f | 487 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
488 | { |
489 | } | |
490 | ||
70b97a7f | 491 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 492 | { |
da04c035 IM |
493 | #ifdef CONFIG_DEBUG_SPINLOCK |
494 | /* this is a valid case when another task releases the spinlock */ | |
495 | rq->lock.owner = current; | |
496 | #endif | |
8a25d5de IM |
497 | /* |
498 | * If we are tracking spinlock dependencies then we have to | |
499 | * fix up the runqueue lock - which gets 'carried over' from | |
500 | * prev into current: | |
501 | */ | |
502 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
503 | ||
4866cde0 NP |
504 | spin_unlock_irq(&rq->lock); |
505 | } | |
506 | ||
507 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 508 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
509 | { |
510 | #ifdef CONFIG_SMP | |
511 | return p->oncpu; | |
512 | #else | |
513 | return rq->curr == p; | |
514 | #endif | |
515 | } | |
516 | ||
70b97a7f | 517 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
518 | { |
519 | #ifdef CONFIG_SMP | |
520 | /* | |
521 | * We can optimise this out completely for !SMP, because the | |
522 | * SMP rebalancing from interrupt is the only thing that cares | |
523 | * here. | |
524 | */ | |
525 | next->oncpu = 1; | |
526 | #endif | |
527 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
528 | spin_unlock_irq(&rq->lock); | |
529 | #else | |
530 | spin_unlock(&rq->lock); | |
531 | #endif | |
532 | } | |
533 | ||
70b97a7f | 534 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
535 | { |
536 | #ifdef CONFIG_SMP | |
537 | /* | |
538 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
539 | * We must ensure this doesn't happen until the switch is completely | |
540 | * finished. | |
541 | */ | |
542 | smp_wmb(); | |
543 | prev->oncpu = 0; | |
544 | #endif | |
545 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
546 | local_irq_enable(); | |
1da177e4 | 547 | #endif |
4866cde0 NP |
548 | } |
549 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 550 | |
b29739f9 IM |
551 | /* |
552 | * __task_rq_lock - lock the runqueue a given task resides on. | |
553 | * Must be called interrupts disabled. | |
554 | */ | |
70b97a7f | 555 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
556 | __acquires(rq->lock) |
557 | { | |
70b97a7f | 558 | struct rq *rq; |
b29739f9 IM |
559 | |
560 | repeat_lock_task: | |
561 | rq = task_rq(p); | |
562 | spin_lock(&rq->lock); | |
563 | if (unlikely(rq != task_rq(p))) { | |
564 | spin_unlock(&rq->lock); | |
565 | goto repeat_lock_task; | |
566 | } | |
567 | return rq; | |
568 | } | |
569 | ||
1da177e4 LT |
570 | /* |
571 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
572 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
573 | * explicitly disabling preemption. | |
574 | */ | |
70b97a7f | 575 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
576 | __acquires(rq->lock) |
577 | { | |
70b97a7f | 578 | struct rq *rq; |
1da177e4 LT |
579 | |
580 | repeat_lock_task: | |
581 | local_irq_save(*flags); | |
582 | rq = task_rq(p); | |
583 | spin_lock(&rq->lock); | |
584 | if (unlikely(rq != task_rq(p))) { | |
585 | spin_unlock_irqrestore(&rq->lock, *flags); | |
586 | goto repeat_lock_task; | |
587 | } | |
588 | return rq; | |
589 | } | |
590 | ||
a9957449 | 591 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
592 | __releases(rq->lock) |
593 | { | |
594 | spin_unlock(&rq->lock); | |
595 | } | |
596 | ||
70b97a7f | 597 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
598 | __releases(rq->lock) |
599 | { | |
600 | spin_unlock_irqrestore(&rq->lock, *flags); | |
601 | } | |
602 | ||
1da177e4 | 603 | /* |
cc2a73b5 | 604 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 605 | */ |
a9957449 | 606 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
607 | __acquires(rq->lock) |
608 | { | |
70b97a7f | 609 | struct rq *rq; |
1da177e4 LT |
610 | |
611 | local_irq_disable(); | |
612 | rq = this_rq(); | |
613 | spin_lock(&rq->lock); | |
614 | ||
615 | return rq; | |
616 | } | |
617 | ||
1b9f19c2 | 618 | /* |
2aa44d05 | 619 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 620 | */ |
2aa44d05 | 621 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 622 | { |
2aa44d05 IM |
623 | struct rq *rq = cpu_rq(smp_processor_id()); |
624 | ||
625 | spin_lock(&rq->lock); | |
626 | __update_rq_clock(rq); | |
627 | spin_unlock(&rq->lock); | |
628 | rq->clock_deep_idle_events++; | |
629 | } | |
630 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
631 | ||
632 | /* | |
633 | * We just idled delta nanoseconds (called with irqs disabled): | |
634 | */ | |
635 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
636 | { | |
637 | struct rq *rq = cpu_rq(smp_processor_id()); | |
638 | u64 now = sched_clock(); | |
1b9f19c2 | 639 | |
2aa44d05 IM |
640 | rq->idle_clock += delta_ns; |
641 | /* | |
642 | * Override the previous timestamp and ignore all | |
643 | * sched_clock() deltas that occured while we idled, | |
644 | * and use the PM-provided delta_ns to advance the | |
645 | * rq clock: | |
646 | */ | |
647 | spin_lock(&rq->lock); | |
648 | rq->prev_clock_raw = now; | |
649 | rq->clock += delta_ns; | |
650 | spin_unlock(&rq->lock); | |
1b9f19c2 | 651 | } |
2aa44d05 | 652 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 653 | |
c24d20db IM |
654 | /* |
655 | * resched_task - mark a task 'to be rescheduled now'. | |
656 | * | |
657 | * On UP this means the setting of the need_resched flag, on SMP it | |
658 | * might also involve a cross-CPU call to trigger the scheduler on | |
659 | * the target CPU. | |
660 | */ | |
661 | #ifdef CONFIG_SMP | |
662 | ||
663 | #ifndef tsk_is_polling | |
664 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
665 | #endif | |
666 | ||
667 | static void resched_task(struct task_struct *p) | |
668 | { | |
669 | int cpu; | |
670 | ||
671 | assert_spin_locked(&task_rq(p)->lock); | |
672 | ||
673 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
674 | return; | |
675 | ||
676 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
677 | ||
678 | cpu = task_cpu(p); | |
679 | if (cpu == smp_processor_id()) | |
680 | return; | |
681 | ||
682 | /* NEED_RESCHED must be visible before we test polling */ | |
683 | smp_mb(); | |
684 | if (!tsk_is_polling(p)) | |
685 | smp_send_reschedule(cpu); | |
686 | } | |
687 | ||
688 | static void resched_cpu(int cpu) | |
689 | { | |
690 | struct rq *rq = cpu_rq(cpu); | |
691 | unsigned long flags; | |
692 | ||
693 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
694 | return; | |
695 | resched_task(cpu_curr(cpu)); | |
696 | spin_unlock_irqrestore(&rq->lock, flags); | |
697 | } | |
698 | #else | |
699 | static inline void resched_task(struct task_struct *p) | |
700 | { | |
701 | assert_spin_locked(&task_rq(p)->lock); | |
702 | set_tsk_need_resched(p); | |
703 | } | |
704 | #endif | |
705 | ||
45bf76df IM |
706 | #if BITS_PER_LONG == 32 |
707 | # define WMULT_CONST (~0UL) | |
708 | #else | |
709 | # define WMULT_CONST (1UL << 32) | |
710 | #endif | |
711 | ||
712 | #define WMULT_SHIFT 32 | |
713 | ||
194081eb IM |
714 | /* |
715 | * Shift right and round: | |
716 | */ | |
cf2ab469 | 717 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 718 | |
cb1c4fc9 | 719 | static unsigned long |
45bf76df IM |
720 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
721 | struct load_weight *lw) | |
722 | { | |
723 | u64 tmp; | |
724 | ||
725 | if (unlikely(!lw->inv_weight)) | |
194081eb | 726 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
727 | |
728 | tmp = (u64)delta_exec * weight; | |
729 | /* | |
730 | * Check whether we'd overflow the 64-bit multiplication: | |
731 | */ | |
194081eb | 732 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 733 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
734 | WMULT_SHIFT/2); |
735 | else | |
cf2ab469 | 736 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 737 | |
ecf691da | 738 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
739 | } |
740 | ||
741 | static inline unsigned long | |
742 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
743 | { | |
744 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
745 | } | |
746 | ||
1091985b | 747 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
748 | { |
749 | lw->weight += inc; | |
45bf76df IM |
750 | } |
751 | ||
1091985b | 752 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
753 | { |
754 | lw->weight -= dec; | |
45bf76df IM |
755 | } |
756 | ||
2dd73a4f PW |
757 | /* |
758 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
759 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
760 | * each task makes to its run queue's load is weighted according to its | |
761 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
762 | * scaled version of the new time slice allocation that they receive on time | |
763 | * slice expiry etc. | |
764 | */ | |
765 | ||
dd41f596 IM |
766 | #define WEIGHT_IDLEPRIO 2 |
767 | #define WMULT_IDLEPRIO (1 << 31) | |
768 | ||
769 | /* | |
770 | * Nice levels are multiplicative, with a gentle 10% change for every | |
771 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
772 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
773 | * that remained on nice 0. | |
774 | * | |
775 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
776 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
777 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
778 | * If a task goes up by ~10% and another task goes down by ~10% then | |
779 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
780 | */ |
781 | static const int prio_to_weight[40] = { | |
254753dc IM |
782 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
783 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
784 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
785 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
786 | /* 0 */ 1024, 820, 655, 526, 423, | |
787 | /* 5 */ 335, 272, 215, 172, 137, | |
788 | /* 10 */ 110, 87, 70, 56, 45, | |
789 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
790 | }; |
791 | ||
5714d2de IM |
792 | /* |
793 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
794 | * | |
795 | * In cases where the weight does not change often, we can use the | |
796 | * precalculated inverse to speed up arithmetics by turning divisions | |
797 | * into multiplications: | |
798 | */ | |
dd41f596 | 799 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
800 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
801 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
802 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
803 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
804 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
805 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
806 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
807 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 808 | }; |
2dd73a4f | 809 | |
dd41f596 IM |
810 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
811 | ||
812 | /* | |
813 | * runqueue iterator, to support SMP load-balancing between different | |
814 | * scheduling classes, without having to expose their internal data | |
815 | * structures to the load-balancing proper: | |
816 | */ | |
817 | struct rq_iterator { | |
818 | void *arg; | |
819 | struct task_struct *(*start)(void *); | |
820 | struct task_struct *(*next)(void *); | |
821 | }; | |
822 | ||
823 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
824 | unsigned long max_nr_move, unsigned long max_load_move, | |
825 | struct sched_domain *sd, enum cpu_idle_type idle, | |
826 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 827 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
828 | |
829 | #include "sched_stats.h" | |
dd41f596 | 830 | #include "sched_idletask.c" |
5522d5d5 IM |
831 | #include "sched_fair.c" |
832 | #include "sched_rt.c" | |
dd41f596 IM |
833 | #ifdef CONFIG_SCHED_DEBUG |
834 | # include "sched_debug.c" | |
835 | #endif | |
836 | ||
837 | #define sched_class_highest (&rt_sched_class) | |
838 | ||
9c217245 IM |
839 | /* |
840 | * Update delta_exec, delta_fair fields for rq. | |
841 | * | |
842 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 843 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
844 | * delta_exec advances at the same rate as wall-clock (provided |
845 | * cpu is not idle). | |
846 | * | |
847 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
848 | * runqueue over any given interval. This (smoothened) load is used | |
849 | * during load balance. | |
850 | * | |
495eca49 | 851 | * This function is called /before/ updating rq->load |
9c217245 IM |
852 | * and when switching tasks. |
853 | */ | |
29b4b623 | 854 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 855 | { |
495eca49 | 856 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
857 | } |
858 | ||
79b5dddf | 859 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 860 | { |
495eca49 | 861 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
862 | } |
863 | ||
e5fa2237 | 864 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
865 | { |
866 | rq->nr_running++; | |
29b4b623 | 867 | inc_load(rq, p); |
9c217245 IM |
868 | } |
869 | ||
db53181e | 870 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
871 | { |
872 | rq->nr_running--; | |
79b5dddf | 873 | dec_load(rq, p); |
9c217245 IM |
874 | } |
875 | ||
45bf76df IM |
876 | static void set_load_weight(struct task_struct *p) |
877 | { | |
878 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
879 | p->se.load.weight = prio_to_weight[0] * 2; |
880 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
881 | return; | |
882 | } | |
45bf76df | 883 | |
dd41f596 IM |
884 | /* |
885 | * SCHED_IDLE tasks get minimal weight: | |
886 | */ | |
887 | if (p->policy == SCHED_IDLE) { | |
888 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
889 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
890 | return; | |
891 | } | |
71f8bd46 | 892 | |
dd41f596 IM |
893 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
894 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
895 | } |
896 | ||
8159f87e | 897 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 898 | { |
dd41f596 | 899 | sched_info_queued(p); |
fd390f6a | 900 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 901 | p->se.on_rq = 1; |
71f8bd46 IM |
902 | } |
903 | ||
69be72c1 | 904 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 905 | { |
f02231e5 | 906 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 907 | p->se.on_rq = 0; |
71f8bd46 IM |
908 | } |
909 | ||
14531189 | 910 | /* |
dd41f596 | 911 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 912 | */ |
14531189 IM |
913 | static inline int __normal_prio(struct task_struct *p) |
914 | { | |
dd41f596 | 915 | return p->static_prio; |
14531189 IM |
916 | } |
917 | ||
b29739f9 IM |
918 | /* |
919 | * Calculate the expected normal priority: i.e. priority | |
920 | * without taking RT-inheritance into account. Might be | |
921 | * boosted by interactivity modifiers. Changes upon fork, | |
922 | * setprio syscalls, and whenever the interactivity | |
923 | * estimator recalculates. | |
924 | */ | |
36c8b586 | 925 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
926 | { |
927 | int prio; | |
928 | ||
e05606d3 | 929 | if (task_has_rt_policy(p)) |
b29739f9 IM |
930 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
931 | else | |
932 | prio = __normal_prio(p); | |
933 | return prio; | |
934 | } | |
935 | ||
936 | /* | |
937 | * Calculate the current priority, i.e. the priority | |
938 | * taken into account by the scheduler. This value might | |
939 | * be boosted by RT tasks, or might be boosted by | |
940 | * interactivity modifiers. Will be RT if the task got | |
941 | * RT-boosted. If not then it returns p->normal_prio. | |
942 | */ | |
36c8b586 | 943 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
944 | { |
945 | p->normal_prio = normal_prio(p); | |
946 | /* | |
947 | * If we are RT tasks or we were boosted to RT priority, | |
948 | * keep the priority unchanged. Otherwise, update priority | |
949 | * to the normal priority: | |
950 | */ | |
951 | if (!rt_prio(p->prio)) | |
952 | return p->normal_prio; | |
953 | return p->prio; | |
954 | } | |
955 | ||
1da177e4 | 956 | /* |
dd41f596 | 957 | * activate_task - move a task to the runqueue. |
1da177e4 | 958 | */ |
dd41f596 | 959 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 960 | { |
dd41f596 IM |
961 | if (p->state == TASK_UNINTERRUPTIBLE) |
962 | rq->nr_uninterruptible--; | |
1da177e4 | 963 | |
8159f87e | 964 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 965 | inc_nr_running(p, rq); |
1da177e4 LT |
966 | } |
967 | ||
1da177e4 LT |
968 | /* |
969 | * deactivate_task - remove a task from the runqueue. | |
970 | */ | |
2e1cb74a | 971 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 972 | { |
dd41f596 IM |
973 | if (p->state == TASK_UNINTERRUPTIBLE) |
974 | rq->nr_uninterruptible++; | |
975 | ||
69be72c1 | 976 | dequeue_task(rq, p, sleep); |
db53181e | 977 | dec_nr_running(p, rq); |
1da177e4 LT |
978 | } |
979 | ||
1da177e4 LT |
980 | /** |
981 | * task_curr - is this task currently executing on a CPU? | |
982 | * @p: the task in question. | |
983 | */ | |
36c8b586 | 984 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
985 | { |
986 | return cpu_curr(task_cpu(p)) == p; | |
987 | } | |
988 | ||
2dd73a4f PW |
989 | /* Used instead of source_load when we know the type == 0 */ |
990 | unsigned long weighted_cpuload(const int cpu) | |
991 | { | |
495eca49 | 992 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
993 | } |
994 | ||
995 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
996 | { | |
997 | #ifdef CONFIG_SMP | |
998 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 999 | #endif |
29f59db3 | 1000 | set_task_cfs_rq(p); |
2dd73a4f PW |
1001 | } |
1002 | ||
1da177e4 | 1003 | #ifdef CONFIG_SMP |
c65cc870 | 1004 | |
dd41f596 | 1005 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1006 | { |
dd41f596 IM |
1007 | int old_cpu = task_cpu(p); |
1008 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1009 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1010 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1011 | u64 clock_offset; |
dd41f596 IM |
1012 | |
1013 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1014 | |
1015 | #ifdef CONFIG_SCHEDSTATS | |
1016 | if (p->se.wait_start) | |
1017 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1018 | if (p->se.sleep_start) |
1019 | p->se.sleep_start -= clock_offset; | |
1020 | if (p->se.block_start) | |
1021 | p->se.block_start -= clock_offset; | |
6cfb0d5d | 1022 | #endif |
2830cf8c SV |
1023 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1024 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1025 | |
1026 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1027 | } |
1028 | ||
70b97a7f | 1029 | struct migration_req { |
1da177e4 | 1030 | struct list_head list; |
1da177e4 | 1031 | |
36c8b586 | 1032 | struct task_struct *task; |
1da177e4 LT |
1033 | int dest_cpu; |
1034 | ||
1da177e4 | 1035 | struct completion done; |
70b97a7f | 1036 | }; |
1da177e4 LT |
1037 | |
1038 | /* | |
1039 | * The task's runqueue lock must be held. | |
1040 | * Returns true if you have to wait for migration thread. | |
1041 | */ | |
36c8b586 | 1042 | static int |
70b97a7f | 1043 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1044 | { |
70b97a7f | 1045 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1046 | |
1047 | /* | |
1048 | * If the task is not on a runqueue (and not running), then | |
1049 | * it is sufficient to simply update the task's cpu field. | |
1050 | */ | |
dd41f596 | 1051 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1052 | set_task_cpu(p, dest_cpu); |
1053 | return 0; | |
1054 | } | |
1055 | ||
1056 | init_completion(&req->done); | |
1da177e4 LT |
1057 | req->task = p; |
1058 | req->dest_cpu = dest_cpu; | |
1059 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1060 | |
1da177e4 LT |
1061 | return 1; |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * wait_task_inactive - wait for a thread to unschedule. | |
1066 | * | |
1067 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1068 | * else this function might spin for a *long* time. This function can't | |
1069 | * be called with interrupts off, or it may introduce deadlock with | |
1070 | * smp_call_function() if an IPI is sent by the same process we are | |
1071 | * waiting to become inactive. | |
1072 | */ | |
36c8b586 | 1073 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1074 | { |
1075 | unsigned long flags; | |
dd41f596 | 1076 | int running, on_rq; |
70b97a7f | 1077 | struct rq *rq; |
1da177e4 LT |
1078 | |
1079 | repeat: | |
fa490cfd LT |
1080 | /* |
1081 | * We do the initial early heuristics without holding | |
1082 | * any task-queue locks at all. We'll only try to get | |
1083 | * the runqueue lock when things look like they will | |
1084 | * work out! | |
1085 | */ | |
1086 | rq = task_rq(p); | |
1087 | ||
1088 | /* | |
1089 | * If the task is actively running on another CPU | |
1090 | * still, just relax and busy-wait without holding | |
1091 | * any locks. | |
1092 | * | |
1093 | * NOTE! Since we don't hold any locks, it's not | |
1094 | * even sure that "rq" stays as the right runqueue! | |
1095 | * But we don't care, since "task_running()" will | |
1096 | * return false if the runqueue has changed and p | |
1097 | * is actually now running somewhere else! | |
1098 | */ | |
1099 | while (task_running(rq, p)) | |
1100 | cpu_relax(); | |
1101 | ||
1102 | /* | |
1103 | * Ok, time to look more closely! We need the rq | |
1104 | * lock now, to be *sure*. If we're wrong, we'll | |
1105 | * just go back and repeat. | |
1106 | */ | |
1da177e4 | 1107 | rq = task_rq_lock(p, &flags); |
fa490cfd | 1108 | running = task_running(rq, p); |
dd41f596 | 1109 | on_rq = p->se.on_rq; |
fa490cfd LT |
1110 | task_rq_unlock(rq, &flags); |
1111 | ||
1112 | /* | |
1113 | * Was it really running after all now that we | |
1114 | * checked with the proper locks actually held? | |
1115 | * | |
1116 | * Oops. Go back and try again.. | |
1117 | */ | |
1118 | if (unlikely(running)) { | |
1da177e4 | 1119 | cpu_relax(); |
1da177e4 LT |
1120 | goto repeat; |
1121 | } | |
fa490cfd LT |
1122 | |
1123 | /* | |
1124 | * It's not enough that it's not actively running, | |
1125 | * it must be off the runqueue _entirely_, and not | |
1126 | * preempted! | |
1127 | * | |
1128 | * So if it wa still runnable (but just not actively | |
1129 | * running right now), it's preempted, and we should | |
1130 | * yield - it could be a while. | |
1131 | */ | |
dd41f596 | 1132 | if (unlikely(on_rq)) { |
fa490cfd LT |
1133 | yield(); |
1134 | goto repeat; | |
1135 | } | |
1136 | ||
1137 | /* | |
1138 | * Ahh, all good. It wasn't running, and it wasn't | |
1139 | * runnable, which means that it will never become | |
1140 | * running in the future either. We're all done! | |
1141 | */ | |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | /*** | |
1145 | * kick_process - kick a running thread to enter/exit the kernel | |
1146 | * @p: the to-be-kicked thread | |
1147 | * | |
1148 | * Cause a process which is running on another CPU to enter | |
1149 | * kernel-mode, without any delay. (to get signals handled.) | |
1150 | * | |
1151 | * NOTE: this function doesnt have to take the runqueue lock, | |
1152 | * because all it wants to ensure is that the remote task enters | |
1153 | * the kernel. If the IPI races and the task has been migrated | |
1154 | * to another CPU then no harm is done and the purpose has been | |
1155 | * achieved as well. | |
1156 | */ | |
36c8b586 | 1157 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1158 | { |
1159 | int cpu; | |
1160 | ||
1161 | preempt_disable(); | |
1162 | cpu = task_cpu(p); | |
1163 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1164 | smp_send_reschedule(cpu); | |
1165 | preempt_enable(); | |
1166 | } | |
1167 | ||
1168 | /* | |
2dd73a4f PW |
1169 | * Return a low guess at the load of a migration-source cpu weighted |
1170 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1171 | * |
1172 | * We want to under-estimate the load of migration sources, to | |
1173 | * balance conservatively. | |
1174 | */ | |
a9957449 | 1175 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1176 | { |
70b97a7f | 1177 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1178 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1179 | |
3b0bd9bc | 1180 | if (type == 0) |
dd41f596 | 1181 | return total; |
b910472d | 1182 | |
dd41f596 | 1183 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1184 | } |
1185 | ||
1186 | /* | |
2dd73a4f PW |
1187 | * Return a high guess at the load of a migration-target cpu weighted |
1188 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1189 | */ |
a9957449 | 1190 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1191 | { |
70b97a7f | 1192 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1193 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1194 | |
7897986b | 1195 | if (type == 0) |
dd41f596 | 1196 | return total; |
3b0bd9bc | 1197 | |
dd41f596 | 1198 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1199 | } |
1200 | ||
1201 | /* | |
1202 | * Return the average load per task on the cpu's run queue | |
1203 | */ | |
1204 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1205 | { | |
70b97a7f | 1206 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1207 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1208 | unsigned long n = rq->nr_running; |
1209 | ||
dd41f596 | 1210 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1211 | } |
1212 | ||
147cbb4b NP |
1213 | /* |
1214 | * find_idlest_group finds and returns the least busy CPU group within the | |
1215 | * domain. | |
1216 | */ | |
1217 | static struct sched_group * | |
1218 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1219 | { | |
1220 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1221 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1222 | int load_idx = sd->forkexec_idx; | |
1223 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1224 | ||
1225 | do { | |
1226 | unsigned long load, avg_load; | |
1227 | int local_group; | |
1228 | int i; | |
1229 | ||
da5a5522 BD |
1230 | /* Skip over this group if it has no CPUs allowed */ |
1231 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
1232 | goto nextgroup; | |
1233 | ||
147cbb4b | 1234 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1235 | |
1236 | /* Tally up the load of all CPUs in the group */ | |
1237 | avg_load = 0; | |
1238 | ||
1239 | for_each_cpu_mask(i, group->cpumask) { | |
1240 | /* Bias balancing toward cpus of our domain */ | |
1241 | if (local_group) | |
1242 | load = source_load(i, load_idx); | |
1243 | else | |
1244 | load = target_load(i, load_idx); | |
1245 | ||
1246 | avg_load += load; | |
1247 | } | |
1248 | ||
1249 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1250 | avg_load = sg_div_cpu_power(group, |
1251 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1252 | |
1253 | if (local_group) { | |
1254 | this_load = avg_load; | |
1255 | this = group; | |
1256 | } else if (avg_load < min_load) { | |
1257 | min_load = avg_load; | |
1258 | idlest = group; | |
1259 | } | |
da5a5522 | 1260 | nextgroup: |
147cbb4b NP |
1261 | group = group->next; |
1262 | } while (group != sd->groups); | |
1263 | ||
1264 | if (!idlest || 100*this_load < imbalance*min_load) | |
1265 | return NULL; | |
1266 | return idlest; | |
1267 | } | |
1268 | ||
1269 | /* | |
0feaece9 | 1270 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1271 | */ |
95cdf3b7 IM |
1272 | static int |
1273 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1274 | { |
da5a5522 | 1275 | cpumask_t tmp; |
147cbb4b NP |
1276 | unsigned long load, min_load = ULONG_MAX; |
1277 | int idlest = -1; | |
1278 | int i; | |
1279 | ||
da5a5522 BD |
1280 | /* Traverse only the allowed CPUs */ |
1281 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1282 | ||
1283 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1284 | load = weighted_cpuload(i); |
147cbb4b NP |
1285 | |
1286 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1287 | min_load = load; | |
1288 | idlest = i; | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | return idlest; | |
1293 | } | |
1294 | ||
476d139c NP |
1295 | /* |
1296 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1297 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1298 | * SD_BALANCE_EXEC. | |
1299 | * | |
1300 | * Balance, ie. select the least loaded group. | |
1301 | * | |
1302 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1303 | * | |
1304 | * preempt must be disabled. | |
1305 | */ | |
1306 | static int sched_balance_self(int cpu, int flag) | |
1307 | { | |
1308 | struct task_struct *t = current; | |
1309 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1310 | |
c96d145e | 1311 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1312 | /* |
1313 | * If power savings logic is enabled for a domain, stop there. | |
1314 | */ | |
5c45bf27 SS |
1315 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1316 | break; | |
476d139c NP |
1317 | if (tmp->flags & flag) |
1318 | sd = tmp; | |
c96d145e | 1319 | } |
476d139c NP |
1320 | |
1321 | while (sd) { | |
1322 | cpumask_t span; | |
1323 | struct sched_group *group; | |
1a848870 SS |
1324 | int new_cpu, weight; |
1325 | ||
1326 | if (!(sd->flags & flag)) { | |
1327 | sd = sd->child; | |
1328 | continue; | |
1329 | } | |
476d139c NP |
1330 | |
1331 | span = sd->span; | |
1332 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1333 | if (!group) { |
1334 | sd = sd->child; | |
1335 | continue; | |
1336 | } | |
476d139c | 1337 | |
da5a5522 | 1338 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1339 | if (new_cpu == -1 || new_cpu == cpu) { |
1340 | /* Now try balancing at a lower domain level of cpu */ | |
1341 | sd = sd->child; | |
1342 | continue; | |
1343 | } | |
476d139c | 1344 | |
1a848870 | 1345 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1346 | cpu = new_cpu; |
476d139c NP |
1347 | sd = NULL; |
1348 | weight = cpus_weight(span); | |
1349 | for_each_domain(cpu, tmp) { | |
1350 | if (weight <= cpus_weight(tmp->span)) | |
1351 | break; | |
1352 | if (tmp->flags & flag) | |
1353 | sd = tmp; | |
1354 | } | |
1355 | /* while loop will break here if sd == NULL */ | |
1356 | } | |
1357 | ||
1358 | return cpu; | |
1359 | } | |
1360 | ||
1361 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1362 | |
1363 | /* | |
1364 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1365 | * not idle and an idle cpu is available. The span of cpus to | |
1366 | * search starts with cpus closest then further out as needed, | |
1367 | * so we always favor a closer, idle cpu. | |
1368 | * | |
1369 | * Returns the CPU we should wake onto. | |
1370 | */ | |
1371 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1372 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1373 | { |
1374 | cpumask_t tmp; | |
1375 | struct sched_domain *sd; | |
1376 | int i; | |
1377 | ||
4953198b SS |
1378 | /* |
1379 | * If it is idle, then it is the best cpu to run this task. | |
1380 | * | |
1381 | * This cpu is also the best, if it has more than one task already. | |
1382 | * Siblings must be also busy(in most cases) as they didn't already | |
1383 | * pickup the extra load from this cpu and hence we need not check | |
1384 | * sibling runqueue info. This will avoid the checks and cache miss | |
1385 | * penalities associated with that. | |
1386 | */ | |
1387 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1388 | return cpu; |
1389 | ||
1390 | for_each_domain(cpu, sd) { | |
1391 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1392 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1393 | for_each_cpu_mask(i, tmp) { |
1394 | if (idle_cpu(i)) | |
1395 | return i; | |
1396 | } | |
9761eea8 | 1397 | } else { |
e0f364f4 | 1398 | break; |
9761eea8 | 1399 | } |
1da177e4 LT |
1400 | } |
1401 | return cpu; | |
1402 | } | |
1403 | #else | |
36c8b586 | 1404 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1405 | { |
1406 | return cpu; | |
1407 | } | |
1408 | #endif | |
1409 | ||
1410 | /*** | |
1411 | * try_to_wake_up - wake up a thread | |
1412 | * @p: the to-be-woken-up thread | |
1413 | * @state: the mask of task states that can be woken | |
1414 | * @sync: do a synchronous wakeup? | |
1415 | * | |
1416 | * Put it on the run-queue if it's not already there. The "current" | |
1417 | * thread is always on the run-queue (except when the actual | |
1418 | * re-schedule is in progress), and as such you're allowed to do | |
1419 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1420 | * runnable without the overhead of this. | |
1421 | * | |
1422 | * returns failure only if the task is already active. | |
1423 | */ | |
36c8b586 | 1424 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1425 | { |
1426 | int cpu, this_cpu, success = 0; | |
1427 | unsigned long flags; | |
1428 | long old_state; | |
70b97a7f | 1429 | struct rq *rq; |
1da177e4 | 1430 | #ifdef CONFIG_SMP |
7897986b | 1431 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1432 | unsigned long load, this_load; |
1da177e4 LT |
1433 | int new_cpu; |
1434 | #endif | |
1435 | ||
1436 | rq = task_rq_lock(p, &flags); | |
1437 | old_state = p->state; | |
1438 | if (!(old_state & state)) | |
1439 | goto out; | |
1440 | ||
dd41f596 | 1441 | if (p->se.on_rq) |
1da177e4 LT |
1442 | goto out_running; |
1443 | ||
1444 | cpu = task_cpu(p); | |
1445 | this_cpu = smp_processor_id(); | |
1446 | ||
1447 | #ifdef CONFIG_SMP | |
1448 | if (unlikely(task_running(rq, p))) | |
1449 | goto out_activate; | |
1450 | ||
7897986b NP |
1451 | new_cpu = cpu; |
1452 | ||
2d72376b | 1453 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1454 | if (cpu == this_cpu) { |
1455 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1456 | goto out_set_cpu; |
1457 | } | |
1458 | ||
1459 | for_each_domain(this_cpu, sd) { | |
1460 | if (cpu_isset(cpu, sd->span)) { | |
1461 | schedstat_inc(sd, ttwu_wake_remote); | |
1462 | this_sd = sd; | |
1463 | break; | |
1da177e4 LT |
1464 | } |
1465 | } | |
1da177e4 | 1466 | |
7897986b | 1467 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1468 | goto out_set_cpu; |
1469 | ||
1da177e4 | 1470 | /* |
7897986b | 1471 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1472 | */ |
7897986b NP |
1473 | if (this_sd) { |
1474 | int idx = this_sd->wake_idx; | |
1475 | unsigned int imbalance; | |
1da177e4 | 1476 | |
a3f21bce NP |
1477 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1478 | ||
7897986b NP |
1479 | load = source_load(cpu, idx); |
1480 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1481 | |
7897986b NP |
1482 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1483 | ||
a3f21bce NP |
1484 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1485 | unsigned long tl = this_load; | |
33859f7f MOS |
1486 | unsigned long tl_per_task; |
1487 | ||
1488 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1489 | |
1da177e4 | 1490 | /* |
a3f21bce NP |
1491 | * If sync wakeup then subtract the (maximum possible) |
1492 | * effect of the currently running task from the load | |
1493 | * of the current CPU: | |
1da177e4 | 1494 | */ |
a3f21bce | 1495 | if (sync) |
dd41f596 | 1496 | tl -= current->se.load.weight; |
a3f21bce NP |
1497 | |
1498 | if ((tl <= load && | |
2dd73a4f | 1499 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1500 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1501 | /* |
1502 | * This domain has SD_WAKE_AFFINE and | |
1503 | * p is cache cold in this domain, and | |
1504 | * there is no bad imbalance. | |
1505 | */ | |
1506 | schedstat_inc(this_sd, ttwu_move_affine); | |
1507 | goto out_set_cpu; | |
1508 | } | |
1509 | } | |
1510 | ||
1511 | /* | |
1512 | * Start passive balancing when half the imbalance_pct | |
1513 | * limit is reached. | |
1514 | */ | |
1515 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1516 | if (imbalance*this_load <= 100*load) { | |
1517 | schedstat_inc(this_sd, ttwu_move_balance); | |
1518 | goto out_set_cpu; | |
1519 | } | |
1da177e4 LT |
1520 | } |
1521 | } | |
1522 | ||
1523 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1524 | out_set_cpu: | |
1525 | new_cpu = wake_idle(new_cpu, p); | |
1526 | if (new_cpu != cpu) { | |
1527 | set_task_cpu(p, new_cpu); | |
1528 | task_rq_unlock(rq, &flags); | |
1529 | /* might preempt at this point */ | |
1530 | rq = task_rq_lock(p, &flags); | |
1531 | old_state = p->state; | |
1532 | if (!(old_state & state)) | |
1533 | goto out; | |
dd41f596 | 1534 | if (p->se.on_rq) |
1da177e4 LT |
1535 | goto out_running; |
1536 | ||
1537 | this_cpu = smp_processor_id(); | |
1538 | cpu = task_cpu(p); | |
1539 | } | |
1540 | ||
1541 | out_activate: | |
1542 | #endif /* CONFIG_SMP */ | |
2daa3577 | 1543 | update_rq_clock(rq); |
dd41f596 | 1544 | activate_task(rq, p, 1); |
1da177e4 LT |
1545 | /* |
1546 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1547 | * has indicated that it will leave the CPU in short order) | |
1548 | * don't trigger a preemption, if the woken up task will run on | |
1549 | * this cpu. (in this case the 'I will reschedule' promise of | |
1550 | * the waker guarantees that the freshly woken up task is going | |
1551 | * to be considered on this CPU.) | |
1552 | */ | |
dd41f596 IM |
1553 | if (!sync || cpu != this_cpu) |
1554 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1555 | success = 1; |
1556 | ||
1557 | out_running: | |
1558 | p->state = TASK_RUNNING; | |
1559 | out: | |
1560 | task_rq_unlock(rq, &flags); | |
1561 | ||
1562 | return success; | |
1563 | } | |
1564 | ||
36c8b586 | 1565 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1566 | { |
1567 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1568 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1569 | } | |
1da177e4 LT |
1570 | EXPORT_SYMBOL(wake_up_process); |
1571 | ||
36c8b586 | 1572 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1573 | { |
1574 | return try_to_wake_up(p, state, 0); | |
1575 | } | |
1576 | ||
1da177e4 LT |
1577 | /* |
1578 | * Perform scheduler related setup for a newly forked process p. | |
1579 | * p is forked by current. | |
dd41f596 IM |
1580 | * |
1581 | * __sched_fork() is basic setup used by init_idle() too: | |
1582 | */ | |
1583 | static void __sched_fork(struct task_struct *p) | |
1584 | { | |
dd41f596 IM |
1585 | p->se.exec_start = 0; |
1586 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1587 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1588 | |
1589 | #ifdef CONFIG_SCHEDSTATS | |
1590 | p->se.wait_start = 0; | |
dd41f596 IM |
1591 | p->se.sum_sleep_runtime = 0; |
1592 | p->se.sleep_start = 0; | |
dd41f596 IM |
1593 | p->se.block_start = 0; |
1594 | p->se.sleep_max = 0; | |
1595 | p->se.block_max = 0; | |
1596 | p->se.exec_max = 0; | |
eba1ed4b | 1597 | p->se.slice_max = 0; |
dd41f596 | 1598 | p->se.wait_max = 0; |
6cfb0d5d | 1599 | #endif |
476d139c | 1600 | |
dd41f596 IM |
1601 | INIT_LIST_HEAD(&p->run_list); |
1602 | p->se.on_rq = 0; | |
476d139c | 1603 | |
e107be36 AK |
1604 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1605 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1606 | #endif | |
1607 | ||
1da177e4 LT |
1608 | /* |
1609 | * We mark the process as running here, but have not actually | |
1610 | * inserted it onto the runqueue yet. This guarantees that | |
1611 | * nobody will actually run it, and a signal or other external | |
1612 | * event cannot wake it up and insert it on the runqueue either. | |
1613 | */ | |
1614 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1615 | } |
1616 | ||
1617 | /* | |
1618 | * fork()/clone()-time setup: | |
1619 | */ | |
1620 | void sched_fork(struct task_struct *p, int clone_flags) | |
1621 | { | |
1622 | int cpu = get_cpu(); | |
1623 | ||
1624 | __sched_fork(p); | |
1625 | ||
1626 | #ifdef CONFIG_SMP | |
1627 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1628 | #endif | |
02e4bac2 | 1629 | set_task_cpu(p, cpu); |
b29739f9 IM |
1630 | |
1631 | /* | |
1632 | * Make sure we do not leak PI boosting priority to the child: | |
1633 | */ | |
1634 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1635 | if (!rt_prio(p->prio)) |
1636 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1637 | |
52f17b6c | 1638 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1639 | if (likely(sched_info_on())) |
52f17b6c | 1640 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1641 | #endif |
d6077cb8 | 1642 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1643 | p->oncpu = 0; |
1644 | #endif | |
1da177e4 | 1645 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1646 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1647 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1648 | #endif |
476d139c | 1649 | put_cpu(); |
1da177e4 LT |
1650 | } |
1651 | ||
1652 | /* | |
1653 | * wake_up_new_task - wake up a newly created task for the first time. | |
1654 | * | |
1655 | * This function will do some initial scheduler statistics housekeeping | |
1656 | * that must be done for every newly created context, then puts the task | |
1657 | * on the runqueue and wakes it. | |
1658 | */ | |
36c8b586 | 1659 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1660 | { |
1661 | unsigned long flags; | |
dd41f596 | 1662 | struct rq *rq; |
1da177e4 LT |
1663 | |
1664 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1665 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1666 | update_rq_clock(rq); |
1da177e4 LT |
1667 | |
1668 | p->prio = effective_prio(p); | |
1669 | ||
00bf7bfc | 1670 | if (!p->sched_class->task_new || !current->se.on_rq || !rq->cfs.curr) { |
dd41f596 | 1671 | activate_task(rq, p, 0); |
1da177e4 | 1672 | } else { |
1da177e4 | 1673 | /* |
dd41f596 IM |
1674 | * Let the scheduling class do new task startup |
1675 | * management (if any): | |
1da177e4 | 1676 | */ |
ee0827d8 | 1677 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1678 | inc_nr_running(p, rq); |
1da177e4 | 1679 | } |
dd41f596 IM |
1680 | check_preempt_curr(rq, p); |
1681 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1682 | } |
1683 | ||
e107be36 AK |
1684 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1685 | ||
1686 | /** | |
421cee29 RD |
1687 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1688 | * @notifier: notifier struct to register | |
e107be36 AK |
1689 | */ |
1690 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1691 | { | |
1692 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1693 | } | |
1694 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1695 | ||
1696 | /** | |
1697 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1698 | * @notifier: notifier struct to unregister |
e107be36 AK |
1699 | * |
1700 | * This is safe to call from within a preemption notifier. | |
1701 | */ | |
1702 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1703 | { | |
1704 | hlist_del(¬ifier->link); | |
1705 | } | |
1706 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1707 | ||
1708 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1709 | { | |
1710 | struct preempt_notifier *notifier; | |
1711 | struct hlist_node *node; | |
1712 | ||
1713 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1714 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1715 | } | |
1716 | ||
1717 | static void | |
1718 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1719 | struct task_struct *next) | |
1720 | { | |
1721 | struct preempt_notifier *notifier; | |
1722 | struct hlist_node *node; | |
1723 | ||
1724 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1725 | notifier->ops->sched_out(notifier, next); | |
1726 | } | |
1727 | ||
1728 | #else | |
1729 | ||
1730 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1731 | { | |
1732 | } | |
1733 | ||
1734 | static void | |
1735 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1736 | struct task_struct *next) | |
1737 | { | |
1738 | } | |
1739 | ||
1740 | #endif | |
1741 | ||
4866cde0 NP |
1742 | /** |
1743 | * prepare_task_switch - prepare to switch tasks | |
1744 | * @rq: the runqueue preparing to switch | |
421cee29 | 1745 | * @prev: the current task that is being switched out |
4866cde0 NP |
1746 | * @next: the task we are going to switch to. |
1747 | * | |
1748 | * This is called with the rq lock held and interrupts off. It must | |
1749 | * be paired with a subsequent finish_task_switch after the context | |
1750 | * switch. | |
1751 | * | |
1752 | * prepare_task_switch sets up locking and calls architecture specific | |
1753 | * hooks. | |
1754 | */ | |
e107be36 AK |
1755 | static inline void |
1756 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1757 | struct task_struct *next) | |
4866cde0 | 1758 | { |
e107be36 | 1759 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1760 | prepare_lock_switch(rq, next); |
1761 | prepare_arch_switch(next); | |
1762 | } | |
1763 | ||
1da177e4 LT |
1764 | /** |
1765 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1766 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1767 | * @prev: the thread we just switched away from. |
1768 | * | |
4866cde0 NP |
1769 | * finish_task_switch must be called after the context switch, paired |
1770 | * with a prepare_task_switch call before the context switch. | |
1771 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1772 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1773 | * |
1774 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1775 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1776 | * with the lock held can cause deadlocks; see schedule() for | |
1777 | * details.) | |
1778 | */ | |
a9957449 | 1779 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1780 | __releases(rq->lock) |
1781 | { | |
1da177e4 | 1782 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1783 | long prev_state; |
1da177e4 LT |
1784 | |
1785 | rq->prev_mm = NULL; | |
1786 | ||
1787 | /* | |
1788 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1789 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1790 | * schedule one last time. The schedule call will never return, and |
1791 | * the scheduled task must drop that reference. | |
c394cc9f | 1792 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1793 | * still held, otherwise prev could be scheduled on another cpu, die |
1794 | * there before we look at prev->state, and then the reference would | |
1795 | * be dropped twice. | |
1796 | * Manfred Spraul <manfred@colorfullife.com> | |
1797 | */ | |
55a101f8 | 1798 | prev_state = prev->state; |
4866cde0 NP |
1799 | finish_arch_switch(prev); |
1800 | finish_lock_switch(rq, prev); | |
e107be36 | 1801 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1802 | if (mm) |
1803 | mmdrop(mm); | |
c394cc9f | 1804 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1805 | /* |
1806 | * Remove function-return probe instances associated with this | |
1807 | * task and put them back on the free list. | |
9761eea8 | 1808 | */ |
c6fd91f0 | 1809 | kprobe_flush_task(prev); |
1da177e4 | 1810 | put_task_struct(prev); |
c6fd91f0 | 1811 | } |
1da177e4 LT |
1812 | } |
1813 | ||
1814 | /** | |
1815 | * schedule_tail - first thing a freshly forked thread must call. | |
1816 | * @prev: the thread we just switched away from. | |
1817 | */ | |
36c8b586 | 1818 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1819 | __releases(rq->lock) |
1820 | { | |
70b97a7f IM |
1821 | struct rq *rq = this_rq(); |
1822 | ||
4866cde0 NP |
1823 | finish_task_switch(rq, prev); |
1824 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1825 | /* In this case, finish_task_switch does not reenable preemption */ | |
1826 | preempt_enable(); | |
1827 | #endif | |
1da177e4 LT |
1828 | if (current->set_child_tid) |
1829 | put_user(current->pid, current->set_child_tid); | |
1830 | } | |
1831 | ||
1832 | /* | |
1833 | * context_switch - switch to the new MM and the new | |
1834 | * thread's register state. | |
1835 | */ | |
dd41f596 | 1836 | static inline void |
70b97a7f | 1837 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1838 | struct task_struct *next) |
1da177e4 | 1839 | { |
dd41f596 | 1840 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1841 | |
e107be36 | 1842 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1843 | mm = next->mm; |
1844 | oldmm = prev->active_mm; | |
9226d125 ZA |
1845 | /* |
1846 | * For paravirt, this is coupled with an exit in switch_to to | |
1847 | * combine the page table reload and the switch backend into | |
1848 | * one hypercall. | |
1849 | */ | |
1850 | arch_enter_lazy_cpu_mode(); | |
1851 | ||
dd41f596 | 1852 | if (unlikely(!mm)) { |
1da177e4 LT |
1853 | next->active_mm = oldmm; |
1854 | atomic_inc(&oldmm->mm_count); | |
1855 | enter_lazy_tlb(oldmm, next); | |
1856 | } else | |
1857 | switch_mm(oldmm, mm, next); | |
1858 | ||
dd41f596 | 1859 | if (unlikely(!prev->mm)) { |
1da177e4 | 1860 | prev->active_mm = NULL; |
1da177e4 LT |
1861 | rq->prev_mm = oldmm; |
1862 | } | |
3a5f5e48 IM |
1863 | /* |
1864 | * Since the runqueue lock will be released by the next | |
1865 | * task (which is an invalid locking op but in the case | |
1866 | * of the scheduler it's an obvious special-case), so we | |
1867 | * do an early lockdep release here: | |
1868 | */ | |
1869 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1870 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1871 | #endif |
1da177e4 LT |
1872 | |
1873 | /* Here we just switch the register state and the stack. */ | |
1874 | switch_to(prev, next, prev); | |
1875 | ||
dd41f596 IM |
1876 | barrier(); |
1877 | /* | |
1878 | * this_rq must be evaluated again because prev may have moved | |
1879 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1880 | * frame will be invalid. | |
1881 | */ | |
1882 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1883 | } |
1884 | ||
1885 | /* | |
1886 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1887 | * | |
1888 | * externally visible scheduler statistics: current number of runnable | |
1889 | * threads, current number of uninterruptible-sleeping threads, total | |
1890 | * number of context switches performed since bootup. | |
1891 | */ | |
1892 | unsigned long nr_running(void) | |
1893 | { | |
1894 | unsigned long i, sum = 0; | |
1895 | ||
1896 | for_each_online_cpu(i) | |
1897 | sum += cpu_rq(i)->nr_running; | |
1898 | ||
1899 | return sum; | |
1900 | } | |
1901 | ||
1902 | unsigned long nr_uninterruptible(void) | |
1903 | { | |
1904 | unsigned long i, sum = 0; | |
1905 | ||
0a945022 | 1906 | for_each_possible_cpu(i) |
1da177e4 LT |
1907 | sum += cpu_rq(i)->nr_uninterruptible; |
1908 | ||
1909 | /* | |
1910 | * Since we read the counters lockless, it might be slightly | |
1911 | * inaccurate. Do not allow it to go below zero though: | |
1912 | */ | |
1913 | if (unlikely((long)sum < 0)) | |
1914 | sum = 0; | |
1915 | ||
1916 | return sum; | |
1917 | } | |
1918 | ||
1919 | unsigned long long nr_context_switches(void) | |
1920 | { | |
cc94abfc SR |
1921 | int i; |
1922 | unsigned long long sum = 0; | |
1da177e4 | 1923 | |
0a945022 | 1924 | for_each_possible_cpu(i) |
1da177e4 LT |
1925 | sum += cpu_rq(i)->nr_switches; |
1926 | ||
1927 | return sum; | |
1928 | } | |
1929 | ||
1930 | unsigned long nr_iowait(void) | |
1931 | { | |
1932 | unsigned long i, sum = 0; | |
1933 | ||
0a945022 | 1934 | for_each_possible_cpu(i) |
1da177e4 LT |
1935 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1936 | ||
1937 | return sum; | |
1938 | } | |
1939 | ||
db1b1fef JS |
1940 | unsigned long nr_active(void) |
1941 | { | |
1942 | unsigned long i, running = 0, uninterruptible = 0; | |
1943 | ||
1944 | for_each_online_cpu(i) { | |
1945 | running += cpu_rq(i)->nr_running; | |
1946 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1947 | } | |
1948 | ||
1949 | if (unlikely((long)uninterruptible < 0)) | |
1950 | uninterruptible = 0; | |
1951 | ||
1952 | return running + uninterruptible; | |
1953 | } | |
1954 | ||
48f24c4d | 1955 | /* |
dd41f596 IM |
1956 | * Update rq->cpu_load[] statistics. This function is usually called every |
1957 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1958 | */ |
dd41f596 | 1959 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1960 | { |
495eca49 | 1961 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
1962 | int i, scale; |
1963 | ||
1964 | this_rq->nr_load_updates++; | |
dd41f596 IM |
1965 | |
1966 | /* Update our load: */ | |
1967 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1968 | unsigned long old_load, new_load; | |
1969 | ||
1970 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1971 | ||
1972 | old_load = this_rq->cpu_load[i]; | |
1973 | new_load = this_load; | |
a25707f3 IM |
1974 | /* |
1975 | * Round up the averaging division if load is increasing. This | |
1976 | * prevents us from getting stuck on 9 if the load is 10, for | |
1977 | * example. | |
1978 | */ | |
1979 | if (new_load > old_load) | |
1980 | new_load += scale-1; | |
dd41f596 IM |
1981 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
1982 | } | |
48f24c4d IM |
1983 | } |
1984 | ||
dd41f596 IM |
1985 | #ifdef CONFIG_SMP |
1986 | ||
1da177e4 LT |
1987 | /* |
1988 | * double_rq_lock - safely lock two runqueues | |
1989 | * | |
1990 | * Note this does not disable interrupts like task_rq_lock, | |
1991 | * you need to do so manually before calling. | |
1992 | */ | |
70b97a7f | 1993 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
1994 | __acquires(rq1->lock) |
1995 | __acquires(rq2->lock) | |
1996 | { | |
054b9108 | 1997 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
1998 | if (rq1 == rq2) { |
1999 | spin_lock(&rq1->lock); | |
2000 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2001 | } else { | |
c96d145e | 2002 | if (rq1 < rq2) { |
1da177e4 LT |
2003 | spin_lock(&rq1->lock); |
2004 | spin_lock(&rq2->lock); | |
2005 | } else { | |
2006 | spin_lock(&rq2->lock); | |
2007 | spin_lock(&rq1->lock); | |
2008 | } | |
2009 | } | |
6e82a3be IM |
2010 | update_rq_clock(rq1); |
2011 | update_rq_clock(rq2); | |
1da177e4 LT |
2012 | } |
2013 | ||
2014 | /* | |
2015 | * double_rq_unlock - safely unlock two runqueues | |
2016 | * | |
2017 | * Note this does not restore interrupts like task_rq_unlock, | |
2018 | * you need to do so manually after calling. | |
2019 | */ | |
70b97a7f | 2020 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2021 | __releases(rq1->lock) |
2022 | __releases(rq2->lock) | |
2023 | { | |
2024 | spin_unlock(&rq1->lock); | |
2025 | if (rq1 != rq2) | |
2026 | spin_unlock(&rq2->lock); | |
2027 | else | |
2028 | __release(rq2->lock); | |
2029 | } | |
2030 | ||
2031 | /* | |
2032 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2033 | */ | |
70b97a7f | 2034 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2035 | __releases(this_rq->lock) |
2036 | __acquires(busiest->lock) | |
2037 | __acquires(this_rq->lock) | |
2038 | { | |
054b9108 KK |
2039 | if (unlikely(!irqs_disabled())) { |
2040 | /* printk() doesn't work good under rq->lock */ | |
2041 | spin_unlock(&this_rq->lock); | |
2042 | BUG_ON(1); | |
2043 | } | |
1da177e4 | 2044 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2045 | if (busiest < this_rq) { |
1da177e4 LT |
2046 | spin_unlock(&this_rq->lock); |
2047 | spin_lock(&busiest->lock); | |
2048 | spin_lock(&this_rq->lock); | |
2049 | } else | |
2050 | spin_lock(&busiest->lock); | |
2051 | } | |
2052 | } | |
2053 | ||
1da177e4 LT |
2054 | /* |
2055 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2056 | * This is accomplished by forcing the cpu_allowed mask to only | |
2057 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2058 | * the cpu_allowed mask is restored. | |
2059 | */ | |
36c8b586 | 2060 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2061 | { |
70b97a7f | 2062 | struct migration_req req; |
1da177e4 | 2063 | unsigned long flags; |
70b97a7f | 2064 | struct rq *rq; |
1da177e4 LT |
2065 | |
2066 | rq = task_rq_lock(p, &flags); | |
2067 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2068 | || unlikely(cpu_is_offline(dest_cpu))) | |
2069 | goto out; | |
2070 | ||
2071 | /* force the process onto the specified CPU */ | |
2072 | if (migrate_task(p, dest_cpu, &req)) { | |
2073 | /* Need to wait for migration thread (might exit: take ref). */ | |
2074 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2075 | |
1da177e4 LT |
2076 | get_task_struct(mt); |
2077 | task_rq_unlock(rq, &flags); | |
2078 | wake_up_process(mt); | |
2079 | put_task_struct(mt); | |
2080 | wait_for_completion(&req.done); | |
36c8b586 | 2081 | |
1da177e4 LT |
2082 | return; |
2083 | } | |
2084 | out: | |
2085 | task_rq_unlock(rq, &flags); | |
2086 | } | |
2087 | ||
2088 | /* | |
476d139c NP |
2089 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2090 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2091 | */ |
2092 | void sched_exec(void) | |
2093 | { | |
1da177e4 | 2094 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2095 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2096 | put_cpu(); |
476d139c NP |
2097 | if (new_cpu != this_cpu) |
2098 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2099 | } |
2100 | ||
2101 | /* | |
2102 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2103 | * Both runqueues must be locked. | |
2104 | */ | |
dd41f596 IM |
2105 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2106 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2107 | { |
2e1cb74a | 2108 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2109 | set_task_cpu(p, this_cpu); |
dd41f596 | 2110 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2111 | /* |
2112 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2113 | * to be always true for them. | |
2114 | */ | |
dd41f596 | 2115 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2120 | */ | |
858119e1 | 2121 | static |
70b97a7f | 2122 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2123 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2124 | int *all_pinned) |
1da177e4 LT |
2125 | { |
2126 | /* | |
2127 | * We do not migrate tasks that are: | |
2128 | * 1) running (obviously), or | |
2129 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2130 | * 3) are cache-hot on their current CPU. | |
2131 | */ | |
1da177e4 LT |
2132 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2133 | return 0; | |
81026794 NP |
2134 | *all_pinned = 0; |
2135 | ||
2136 | if (task_running(rq, p)) | |
2137 | return 0; | |
1da177e4 | 2138 | |
1da177e4 LT |
2139 | return 1; |
2140 | } | |
2141 | ||
dd41f596 | 2142 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2143 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2144 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2145 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2146 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2147 | { |
dd41f596 IM |
2148 | int pulled = 0, pinned = 0, skip_for_load; |
2149 | struct task_struct *p; | |
2150 | long rem_load_move = max_load_move; | |
1da177e4 | 2151 | |
2dd73a4f | 2152 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2153 | goto out; |
2154 | ||
81026794 NP |
2155 | pinned = 1; |
2156 | ||
1da177e4 | 2157 | /* |
dd41f596 | 2158 | * Start the load-balancing iterator: |
1da177e4 | 2159 | */ |
dd41f596 IM |
2160 | p = iterator->start(iterator->arg); |
2161 | next: | |
2162 | if (!p) | |
1da177e4 | 2163 | goto out; |
50ddd969 PW |
2164 | /* |
2165 | * To help distribute high priority tasks accross CPUs we don't | |
2166 | * skip a task if it will be the highest priority task (i.e. smallest | |
2167 | * prio value) on its new queue regardless of its load weight | |
2168 | */ | |
dd41f596 IM |
2169 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2170 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2171 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2172 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2173 | p = iterator->next(iterator->arg); |
2174 | goto next; | |
1da177e4 LT |
2175 | } |
2176 | ||
dd41f596 | 2177 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2178 | pulled++; |
dd41f596 | 2179 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2180 | |
2dd73a4f PW |
2181 | /* |
2182 | * We only want to steal up to the prescribed number of tasks | |
2183 | * and the prescribed amount of weighted load. | |
2184 | */ | |
2185 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2186 | if (p->prio < *this_best_prio) |
2187 | *this_best_prio = p->prio; | |
dd41f596 IM |
2188 | p = iterator->next(iterator->arg); |
2189 | goto next; | |
1da177e4 LT |
2190 | } |
2191 | out: | |
2192 | /* | |
2193 | * Right now, this is the only place pull_task() is called, | |
2194 | * so we can safely collect pull_task() stats here rather than | |
2195 | * inside pull_task(). | |
2196 | */ | |
2197 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2198 | |
2199 | if (all_pinned) | |
2200 | *all_pinned = pinned; | |
dd41f596 | 2201 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2202 | return pulled; |
2203 | } | |
2204 | ||
dd41f596 | 2205 | /* |
43010659 PW |
2206 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2207 | * this_rq, as part of a balancing operation within domain "sd". | |
2208 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2209 | * |
2210 | * Called with both runqueues locked. | |
2211 | */ | |
2212 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2213 | unsigned long max_load_move, |
dd41f596 IM |
2214 | struct sched_domain *sd, enum cpu_idle_type idle, |
2215 | int *all_pinned) | |
2216 | { | |
5522d5d5 | 2217 | const struct sched_class *class = sched_class_highest; |
43010659 | 2218 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2219 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2220 | |
2221 | do { | |
43010659 PW |
2222 | total_load_moved += |
2223 | class->load_balance(this_rq, this_cpu, busiest, | |
2224 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2225 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2226 | class = class->next; |
43010659 | 2227 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2228 | |
43010659 PW |
2229 | return total_load_moved > 0; |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2234 | * part of active balancing operations within "domain". | |
2235 | * Returns 1 if successful and 0 otherwise. | |
2236 | * | |
2237 | * Called with both runqueues locked. | |
2238 | */ | |
2239 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2240 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2241 | { | |
5522d5d5 | 2242 | const struct sched_class *class; |
a4ac01c3 | 2243 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2244 | |
2245 | for (class = sched_class_highest; class; class = class->next) | |
2246 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2247 | 1, ULONG_MAX, sd, idle, NULL, |
2248 | &this_best_prio)) | |
43010659 PW |
2249 | return 1; |
2250 | ||
2251 | return 0; | |
dd41f596 IM |
2252 | } |
2253 | ||
1da177e4 LT |
2254 | /* |
2255 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2256 | * domain. It calculates and returns the amount of weighted load which |
2257 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2258 | */ |
2259 | static struct sched_group * | |
2260 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2261 | unsigned long *imbalance, enum cpu_idle_type idle, |
2262 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2263 | { |
2264 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2265 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2266 | unsigned long max_pull; |
2dd73a4f PW |
2267 | unsigned long busiest_load_per_task, busiest_nr_running; |
2268 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2269 | int load_idx; |
5c45bf27 SS |
2270 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2271 | int power_savings_balance = 1; | |
2272 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2273 | unsigned long min_nr_running = ULONG_MAX; | |
2274 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2275 | #endif | |
1da177e4 LT |
2276 | |
2277 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2278 | busiest_load_per_task = busiest_nr_running = 0; |
2279 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2280 | if (idle == CPU_NOT_IDLE) |
7897986b | 2281 | load_idx = sd->busy_idx; |
d15bcfdb | 2282 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2283 | load_idx = sd->newidle_idx; |
2284 | else | |
2285 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2286 | |
2287 | do { | |
5c45bf27 | 2288 | unsigned long load, group_capacity; |
1da177e4 LT |
2289 | int local_group; |
2290 | int i; | |
783609c6 | 2291 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2292 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2293 | |
2294 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2295 | ||
783609c6 SS |
2296 | if (local_group) |
2297 | balance_cpu = first_cpu(group->cpumask); | |
2298 | ||
1da177e4 | 2299 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2300 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2301 | |
2302 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2303 | struct rq *rq; |
2304 | ||
2305 | if (!cpu_isset(i, *cpus)) | |
2306 | continue; | |
2307 | ||
2308 | rq = cpu_rq(i); | |
2dd73a4f | 2309 | |
9439aab8 | 2310 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2311 | *sd_idle = 0; |
2312 | ||
1da177e4 | 2313 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2314 | if (local_group) { |
2315 | if (idle_cpu(i) && !first_idle_cpu) { | |
2316 | first_idle_cpu = 1; | |
2317 | balance_cpu = i; | |
2318 | } | |
2319 | ||
a2000572 | 2320 | load = target_load(i, load_idx); |
783609c6 | 2321 | } else |
a2000572 | 2322 | load = source_load(i, load_idx); |
1da177e4 LT |
2323 | |
2324 | avg_load += load; | |
2dd73a4f | 2325 | sum_nr_running += rq->nr_running; |
dd41f596 | 2326 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2327 | } |
2328 | ||
783609c6 SS |
2329 | /* |
2330 | * First idle cpu or the first cpu(busiest) in this sched group | |
2331 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2332 | * domains. In the newly idle case, we will allow all the cpu's |
2333 | * to do the newly idle load balance. | |
783609c6 | 2334 | */ |
9439aab8 SS |
2335 | if (idle != CPU_NEWLY_IDLE && local_group && |
2336 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2337 | *balance = 0; |
2338 | goto ret; | |
2339 | } | |
2340 | ||
1da177e4 | 2341 | total_load += avg_load; |
5517d86b | 2342 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2343 | |
2344 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2345 | avg_load = sg_div_cpu_power(group, |
2346 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2347 | |
5517d86b | 2348 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2349 | |
1da177e4 LT |
2350 | if (local_group) { |
2351 | this_load = avg_load; | |
2352 | this = group; | |
2dd73a4f PW |
2353 | this_nr_running = sum_nr_running; |
2354 | this_load_per_task = sum_weighted_load; | |
2355 | } else if (avg_load > max_load && | |
5c45bf27 | 2356 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2357 | max_load = avg_load; |
2358 | busiest = group; | |
2dd73a4f PW |
2359 | busiest_nr_running = sum_nr_running; |
2360 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2361 | } |
5c45bf27 SS |
2362 | |
2363 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2364 | /* | |
2365 | * Busy processors will not participate in power savings | |
2366 | * balance. | |
2367 | */ | |
dd41f596 IM |
2368 | if (idle == CPU_NOT_IDLE || |
2369 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2370 | goto group_next; | |
5c45bf27 SS |
2371 | |
2372 | /* | |
2373 | * If the local group is idle or completely loaded | |
2374 | * no need to do power savings balance at this domain | |
2375 | */ | |
2376 | if (local_group && (this_nr_running >= group_capacity || | |
2377 | !this_nr_running)) | |
2378 | power_savings_balance = 0; | |
2379 | ||
dd41f596 | 2380 | /* |
5c45bf27 SS |
2381 | * If a group is already running at full capacity or idle, |
2382 | * don't include that group in power savings calculations | |
dd41f596 IM |
2383 | */ |
2384 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2385 | || !sum_nr_running) |
dd41f596 | 2386 | goto group_next; |
5c45bf27 | 2387 | |
dd41f596 | 2388 | /* |
5c45bf27 | 2389 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2390 | * This is the group from where we need to pick up the load |
2391 | * for saving power | |
2392 | */ | |
2393 | if ((sum_nr_running < min_nr_running) || | |
2394 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2395 | first_cpu(group->cpumask) < |
2396 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2397 | group_min = group; |
2398 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2399 | min_load_per_task = sum_weighted_load / |
2400 | sum_nr_running; | |
dd41f596 | 2401 | } |
5c45bf27 | 2402 | |
dd41f596 | 2403 | /* |
5c45bf27 | 2404 | * Calculate the group which is almost near its |
dd41f596 IM |
2405 | * capacity but still has some space to pick up some load |
2406 | * from other group and save more power | |
2407 | */ | |
2408 | if (sum_nr_running <= group_capacity - 1) { | |
2409 | if (sum_nr_running > leader_nr_running || | |
2410 | (sum_nr_running == leader_nr_running && | |
2411 | first_cpu(group->cpumask) > | |
2412 | first_cpu(group_leader->cpumask))) { | |
2413 | group_leader = group; | |
2414 | leader_nr_running = sum_nr_running; | |
2415 | } | |
48f24c4d | 2416 | } |
5c45bf27 SS |
2417 | group_next: |
2418 | #endif | |
1da177e4 LT |
2419 | group = group->next; |
2420 | } while (group != sd->groups); | |
2421 | ||
2dd73a4f | 2422 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2423 | goto out_balanced; |
2424 | ||
2425 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2426 | ||
2427 | if (this_load >= avg_load || | |
2428 | 100*max_load <= sd->imbalance_pct*this_load) | |
2429 | goto out_balanced; | |
2430 | ||
2dd73a4f | 2431 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2432 | /* |
2433 | * We're trying to get all the cpus to the average_load, so we don't | |
2434 | * want to push ourselves above the average load, nor do we wish to | |
2435 | * reduce the max loaded cpu below the average load, as either of these | |
2436 | * actions would just result in more rebalancing later, and ping-pong | |
2437 | * tasks around. Thus we look for the minimum possible imbalance. | |
2438 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2439 | * be counted as no imbalance for these purposes -- we can't fix that | |
2440 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2441 | * appear as very large values with unsigned longs. | |
2442 | */ | |
2dd73a4f PW |
2443 | if (max_load <= busiest_load_per_task) |
2444 | goto out_balanced; | |
2445 | ||
2446 | /* | |
2447 | * In the presence of smp nice balancing, certain scenarios can have | |
2448 | * max load less than avg load(as we skip the groups at or below | |
2449 | * its cpu_power, while calculating max_load..) | |
2450 | */ | |
2451 | if (max_load < avg_load) { | |
2452 | *imbalance = 0; | |
2453 | goto small_imbalance; | |
2454 | } | |
0c117f1b SS |
2455 | |
2456 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2457 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2458 | |
1da177e4 | 2459 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2460 | *imbalance = min(max_pull * busiest->__cpu_power, |
2461 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2462 | / SCHED_LOAD_SCALE; |
2463 | ||
2dd73a4f PW |
2464 | /* |
2465 | * if *imbalance is less than the average load per runnable task | |
2466 | * there is no gaurantee that any tasks will be moved so we'll have | |
2467 | * a think about bumping its value to force at least one task to be | |
2468 | * moved | |
2469 | */ | |
7fd0d2dd | 2470 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2471 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2472 | unsigned int imbn; |
2473 | ||
2474 | small_imbalance: | |
2475 | pwr_move = pwr_now = 0; | |
2476 | imbn = 2; | |
2477 | if (this_nr_running) { | |
2478 | this_load_per_task /= this_nr_running; | |
2479 | if (busiest_load_per_task > this_load_per_task) | |
2480 | imbn = 1; | |
2481 | } else | |
2482 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2483 | |
dd41f596 IM |
2484 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2485 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2486 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2487 | return busiest; |
2488 | } | |
2489 | ||
2490 | /* | |
2491 | * OK, we don't have enough imbalance to justify moving tasks, | |
2492 | * however we may be able to increase total CPU power used by | |
2493 | * moving them. | |
2494 | */ | |
2495 | ||
5517d86b ED |
2496 | pwr_now += busiest->__cpu_power * |
2497 | min(busiest_load_per_task, max_load); | |
2498 | pwr_now += this->__cpu_power * | |
2499 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2500 | pwr_now /= SCHED_LOAD_SCALE; |
2501 | ||
2502 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2503 | tmp = sg_div_cpu_power(busiest, |
2504 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2505 | if (max_load > tmp) |
5517d86b | 2506 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2507 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2508 | |
2509 | /* Amount of load we'd add */ | |
5517d86b | 2510 | if (max_load * busiest->__cpu_power < |
33859f7f | 2511 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2512 | tmp = sg_div_cpu_power(this, |
2513 | max_load * busiest->__cpu_power); | |
1da177e4 | 2514 | else |
5517d86b ED |
2515 | tmp = sg_div_cpu_power(this, |
2516 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2517 | pwr_move += this->__cpu_power * | |
2518 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2519 | pwr_move /= SCHED_LOAD_SCALE; |
2520 | ||
2521 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2522 | if (pwr_move > pwr_now) |
2523 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2524 | } |
2525 | ||
1da177e4 LT |
2526 | return busiest; |
2527 | ||
2528 | out_balanced: | |
5c45bf27 | 2529 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2530 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2531 | goto ret; |
1da177e4 | 2532 | |
5c45bf27 SS |
2533 | if (this == group_leader && group_leader != group_min) { |
2534 | *imbalance = min_load_per_task; | |
2535 | return group_min; | |
2536 | } | |
5c45bf27 | 2537 | #endif |
783609c6 | 2538 | ret: |
1da177e4 LT |
2539 | *imbalance = 0; |
2540 | return NULL; | |
2541 | } | |
2542 | ||
2543 | /* | |
2544 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2545 | */ | |
70b97a7f | 2546 | static struct rq * |
d15bcfdb | 2547 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2548 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2549 | { |
70b97a7f | 2550 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2551 | unsigned long max_load = 0; |
1da177e4 LT |
2552 | int i; |
2553 | ||
2554 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2555 | unsigned long wl; |
0a2966b4 CL |
2556 | |
2557 | if (!cpu_isset(i, *cpus)) | |
2558 | continue; | |
2559 | ||
48f24c4d | 2560 | rq = cpu_rq(i); |
dd41f596 | 2561 | wl = weighted_cpuload(i); |
2dd73a4f | 2562 | |
dd41f596 | 2563 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2564 | continue; |
1da177e4 | 2565 | |
dd41f596 IM |
2566 | if (wl > max_load) { |
2567 | max_load = wl; | |
48f24c4d | 2568 | busiest = rq; |
1da177e4 LT |
2569 | } |
2570 | } | |
2571 | ||
2572 | return busiest; | |
2573 | } | |
2574 | ||
77391d71 NP |
2575 | /* |
2576 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2577 | * so long as it is large enough. | |
2578 | */ | |
2579 | #define MAX_PINNED_INTERVAL 512 | |
2580 | ||
1da177e4 LT |
2581 | /* |
2582 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2583 | * tasks if there is an imbalance. | |
1da177e4 | 2584 | */ |
70b97a7f | 2585 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2586 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2587 | int *balance) |
1da177e4 | 2588 | { |
43010659 | 2589 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2590 | struct sched_group *group; |
1da177e4 | 2591 | unsigned long imbalance; |
70b97a7f | 2592 | struct rq *busiest; |
0a2966b4 | 2593 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2594 | unsigned long flags; |
5969fe06 | 2595 | |
89c4710e SS |
2596 | /* |
2597 | * When power savings policy is enabled for the parent domain, idle | |
2598 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2599 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2600 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2601 | */ |
d15bcfdb | 2602 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2603 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2604 | sd_idle = 1; |
1da177e4 | 2605 | |
2d72376b | 2606 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2607 | |
0a2966b4 CL |
2608 | redo: |
2609 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2610 | &cpus, balance); |
2611 | ||
06066714 | 2612 | if (*balance == 0) |
783609c6 | 2613 | goto out_balanced; |
783609c6 | 2614 | |
1da177e4 LT |
2615 | if (!group) { |
2616 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2617 | goto out_balanced; | |
2618 | } | |
2619 | ||
0a2966b4 | 2620 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2621 | if (!busiest) { |
2622 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2623 | goto out_balanced; | |
2624 | } | |
2625 | ||
db935dbd | 2626 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2627 | |
2628 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2629 | ||
43010659 | 2630 | ld_moved = 0; |
1da177e4 LT |
2631 | if (busiest->nr_running > 1) { |
2632 | /* | |
2633 | * Attempt to move tasks. If find_busiest_group has found | |
2634 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2635 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2636 | * correctly treated as an imbalance. |
2637 | */ | |
fe2eea3f | 2638 | local_irq_save(flags); |
e17224bf | 2639 | double_rq_lock(this_rq, busiest); |
43010659 | 2640 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2641 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2642 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2643 | local_irq_restore(flags); |
81026794 | 2644 | |
46cb4b7c SS |
2645 | /* |
2646 | * some other cpu did the load balance for us. | |
2647 | */ | |
43010659 | 2648 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2649 | resched_cpu(this_cpu); |
2650 | ||
81026794 | 2651 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2652 | if (unlikely(all_pinned)) { |
2653 | cpu_clear(cpu_of(busiest), cpus); | |
2654 | if (!cpus_empty(cpus)) | |
2655 | goto redo; | |
81026794 | 2656 | goto out_balanced; |
0a2966b4 | 2657 | } |
1da177e4 | 2658 | } |
81026794 | 2659 | |
43010659 | 2660 | if (!ld_moved) { |
1da177e4 LT |
2661 | schedstat_inc(sd, lb_failed[idle]); |
2662 | sd->nr_balance_failed++; | |
2663 | ||
2664 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2665 | |
fe2eea3f | 2666 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2667 | |
2668 | /* don't kick the migration_thread, if the curr | |
2669 | * task on busiest cpu can't be moved to this_cpu | |
2670 | */ | |
2671 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2672 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2673 | all_pinned = 1; |
2674 | goto out_one_pinned; | |
2675 | } | |
2676 | ||
1da177e4 LT |
2677 | if (!busiest->active_balance) { |
2678 | busiest->active_balance = 1; | |
2679 | busiest->push_cpu = this_cpu; | |
81026794 | 2680 | active_balance = 1; |
1da177e4 | 2681 | } |
fe2eea3f | 2682 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2683 | if (active_balance) |
1da177e4 LT |
2684 | wake_up_process(busiest->migration_thread); |
2685 | ||
2686 | /* | |
2687 | * We've kicked active balancing, reset the failure | |
2688 | * counter. | |
2689 | */ | |
39507451 | 2690 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2691 | } |
81026794 | 2692 | } else |
1da177e4 LT |
2693 | sd->nr_balance_failed = 0; |
2694 | ||
81026794 | 2695 | if (likely(!active_balance)) { |
1da177e4 LT |
2696 | /* We were unbalanced, so reset the balancing interval */ |
2697 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2698 | } else { |
2699 | /* | |
2700 | * If we've begun active balancing, start to back off. This | |
2701 | * case may not be covered by the all_pinned logic if there | |
2702 | * is only 1 task on the busy runqueue (because we don't call | |
2703 | * move_tasks). | |
2704 | */ | |
2705 | if (sd->balance_interval < sd->max_interval) | |
2706 | sd->balance_interval *= 2; | |
1da177e4 LT |
2707 | } |
2708 | ||
43010659 | 2709 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2710 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2711 | return -1; |
43010659 | 2712 | return ld_moved; |
1da177e4 LT |
2713 | |
2714 | out_balanced: | |
1da177e4 LT |
2715 | schedstat_inc(sd, lb_balanced[idle]); |
2716 | ||
16cfb1c0 | 2717 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2718 | |
2719 | out_one_pinned: | |
1da177e4 | 2720 | /* tune up the balancing interval */ |
77391d71 NP |
2721 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2722 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2723 | sd->balance_interval *= 2; |
2724 | ||
48f24c4d | 2725 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2726 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2727 | return -1; |
1da177e4 LT |
2728 | return 0; |
2729 | } | |
2730 | ||
2731 | /* | |
2732 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2733 | * tasks if there is an imbalance. | |
2734 | * | |
d15bcfdb | 2735 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2736 | * this_rq is locked. |
2737 | */ | |
48f24c4d | 2738 | static int |
70b97a7f | 2739 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2740 | { |
2741 | struct sched_group *group; | |
70b97a7f | 2742 | struct rq *busiest = NULL; |
1da177e4 | 2743 | unsigned long imbalance; |
43010659 | 2744 | int ld_moved = 0; |
5969fe06 | 2745 | int sd_idle = 0; |
969bb4e4 | 2746 | int all_pinned = 0; |
0a2966b4 | 2747 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2748 | |
89c4710e SS |
2749 | /* |
2750 | * When power savings policy is enabled for the parent domain, idle | |
2751 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2752 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2753 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2754 | */ |
2755 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2756 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2757 | sd_idle = 1; |
1da177e4 | 2758 | |
2d72376b | 2759 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2760 | redo: |
d15bcfdb | 2761 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2762 | &sd_idle, &cpus, NULL); |
1da177e4 | 2763 | if (!group) { |
d15bcfdb | 2764 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2765 | goto out_balanced; |
1da177e4 LT |
2766 | } |
2767 | ||
d15bcfdb | 2768 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2769 | &cpus); |
db935dbd | 2770 | if (!busiest) { |
d15bcfdb | 2771 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2772 | goto out_balanced; |
1da177e4 LT |
2773 | } |
2774 | ||
db935dbd NP |
2775 | BUG_ON(busiest == this_rq); |
2776 | ||
d15bcfdb | 2777 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2778 | |
43010659 | 2779 | ld_moved = 0; |
d6d5cfaf NP |
2780 | if (busiest->nr_running > 1) { |
2781 | /* Attempt to move tasks */ | |
2782 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2783 | /* this_rq->clock is already updated */ |
2784 | update_rq_clock(busiest); | |
43010659 | 2785 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2786 | imbalance, sd, CPU_NEWLY_IDLE, |
2787 | &all_pinned); | |
d6d5cfaf | 2788 | spin_unlock(&busiest->lock); |
0a2966b4 | 2789 | |
969bb4e4 | 2790 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2791 | cpu_clear(cpu_of(busiest), cpus); |
2792 | if (!cpus_empty(cpus)) | |
2793 | goto redo; | |
2794 | } | |
d6d5cfaf NP |
2795 | } |
2796 | ||
43010659 | 2797 | if (!ld_moved) { |
d15bcfdb | 2798 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2799 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2800 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2801 | return -1; |
2802 | } else | |
16cfb1c0 | 2803 | sd->nr_balance_failed = 0; |
1da177e4 | 2804 | |
43010659 | 2805 | return ld_moved; |
16cfb1c0 NP |
2806 | |
2807 | out_balanced: | |
d15bcfdb | 2808 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2809 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2810 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2811 | return -1; |
16cfb1c0 | 2812 | sd->nr_balance_failed = 0; |
48f24c4d | 2813 | |
16cfb1c0 | 2814 | return 0; |
1da177e4 LT |
2815 | } |
2816 | ||
2817 | /* | |
2818 | * idle_balance is called by schedule() if this_cpu is about to become | |
2819 | * idle. Attempts to pull tasks from other CPUs. | |
2820 | */ | |
70b97a7f | 2821 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2822 | { |
2823 | struct sched_domain *sd; | |
dd41f596 IM |
2824 | int pulled_task = -1; |
2825 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2826 | |
2827 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2828 | unsigned long interval; |
2829 | ||
2830 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2831 | continue; | |
2832 | ||
2833 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2834 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2835 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2836 | this_rq, sd); |
2837 | ||
2838 | interval = msecs_to_jiffies(sd->balance_interval); | |
2839 | if (time_after(next_balance, sd->last_balance + interval)) | |
2840 | next_balance = sd->last_balance + interval; | |
2841 | if (pulled_task) | |
2842 | break; | |
1da177e4 | 2843 | } |
dd41f596 | 2844 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2845 | /* |
2846 | * We are going idle. next_balance may be set based on | |
2847 | * a busy processor. So reset next_balance. | |
2848 | */ | |
2849 | this_rq->next_balance = next_balance; | |
dd41f596 | 2850 | } |
1da177e4 LT |
2851 | } |
2852 | ||
2853 | /* | |
2854 | * active_load_balance is run by migration threads. It pushes running tasks | |
2855 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2856 | * running on each physical CPU where possible, and avoids physical / | |
2857 | * logical imbalances. | |
2858 | * | |
2859 | * Called with busiest_rq locked. | |
2860 | */ | |
70b97a7f | 2861 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2862 | { |
39507451 | 2863 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2864 | struct sched_domain *sd; |
2865 | struct rq *target_rq; | |
39507451 | 2866 | |
48f24c4d | 2867 | /* Is there any task to move? */ |
39507451 | 2868 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2869 | return; |
2870 | ||
2871 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2872 | |
2873 | /* | |
39507451 NP |
2874 | * This condition is "impossible", if it occurs |
2875 | * we need to fix it. Originally reported by | |
2876 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2877 | */ |
39507451 | 2878 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2879 | |
39507451 NP |
2880 | /* move a task from busiest_rq to target_rq */ |
2881 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2882 | update_rq_clock(busiest_rq); |
2883 | update_rq_clock(target_rq); | |
39507451 NP |
2884 | |
2885 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2886 | for_each_domain(target_cpu, sd) { |
39507451 | 2887 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2888 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2889 | break; |
c96d145e | 2890 | } |
39507451 | 2891 | |
48f24c4d | 2892 | if (likely(sd)) { |
2d72376b | 2893 | schedstat_inc(sd, alb_count); |
39507451 | 2894 | |
43010659 PW |
2895 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2896 | sd, CPU_IDLE)) | |
48f24c4d IM |
2897 | schedstat_inc(sd, alb_pushed); |
2898 | else | |
2899 | schedstat_inc(sd, alb_failed); | |
2900 | } | |
39507451 | 2901 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2902 | } |
2903 | ||
46cb4b7c SS |
2904 | #ifdef CONFIG_NO_HZ |
2905 | static struct { | |
2906 | atomic_t load_balancer; | |
2907 | cpumask_t cpu_mask; | |
2908 | } nohz ____cacheline_aligned = { | |
2909 | .load_balancer = ATOMIC_INIT(-1), | |
2910 | .cpu_mask = CPU_MASK_NONE, | |
2911 | }; | |
2912 | ||
7835b98b | 2913 | /* |
46cb4b7c SS |
2914 | * This routine will try to nominate the ilb (idle load balancing) |
2915 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2916 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2917 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2918 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2919 | * arrives... | |
2920 | * | |
2921 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2922 | * for idle load balancing. ilb owner will still be part of | |
2923 | * nohz.cpu_mask.. | |
7835b98b | 2924 | * |
46cb4b7c SS |
2925 | * While stopping the tick, this cpu will become the ilb owner if there |
2926 | * is no other owner. And will be the owner till that cpu becomes busy | |
2927 | * or if all cpus in the system stop their ticks at which point | |
2928 | * there is no need for ilb owner. | |
2929 | * | |
2930 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2931 | * next busy scheduler_tick() | |
2932 | */ | |
2933 | int select_nohz_load_balancer(int stop_tick) | |
2934 | { | |
2935 | int cpu = smp_processor_id(); | |
2936 | ||
2937 | if (stop_tick) { | |
2938 | cpu_set(cpu, nohz.cpu_mask); | |
2939 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2940 | ||
2941 | /* | |
2942 | * If we are going offline and still the leader, give up! | |
2943 | */ | |
2944 | if (cpu_is_offline(cpu) && | |
2945 | atomic_read(&nohz.load_balancer) == cpu) { | |
2946 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2947 | BUG(); | |
2948 | return 0; | |
2949 | } | |
2950 | ||
2951 | /* time for ilb owner also to sleep */ | |
2952 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2953 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2954 | atomic_set(&nohz.load_balancer, -1); | |
2955 | return 0; | |
2956 | } | |
2957 | ||
2958 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2959 | /* make me the ilb owner */ | |
2960 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2961 | return 1; | |
2962 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2963 | return 1; | |
2964 | } else { | |
2965 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2966 | return 0; | |
2967 | ||
2968 | cpu_clear(cpu, nohz.cpu_mask); | |
2969 | ||
2970 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2971 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2972 | BUG(); | |
2973 | } | |
2974 | return 0; | |
2975 | } | |
2976 | #endif | |
2977 | ||
2978 | static DEFINE_SPINLOCK(balancing); | |
2979 | ||
2980 | /* | |
7835b98b CL |
2981 | * It checks each scheduling domain to see if it is due to be balanced, |
2982 | * and initiates a balancing operation if so. | |
2983 | * | |
2984 | * Balancing parameters are set up in arch_init_sched_domains. | |
2985 | */ | |
a9957449 | 2986 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 2987 | { |
46cb4b7c SS |
2988 | int balance = 1; |
2989 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
2990 | unsigned long interval; |
2991 | struct sched_domain *sd; | |
46cb4b7c | 2992 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 2993 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 2994 | int update_next_balance = 0; |
1da177e4 | 2995 | |
46cb4b7c | 2996 | for_each_domain(cpu, sd) { |
1da177e4 LT |
2997 | if (!(sd->flags & SD_LOAD_BALANCE)) |
2998 | continue; | |
2999 | ||
3000 | interval = sd->balance_interval; | |
d15bcfdb | 3001 | if (idle != CPU_IDLE) |
1da177e4 LT |
3002 | interval *= sd->busy_factor; |
3003 | ||
3004 | /* scale ms to jiffies */ | |
3005 | interval = msecs_to_jiffies(interval); | |
3006 | if (unlikely(!interval)) | |
3007 | interval = 1; | |
dd41f596 IM |
3008 | if (interval > HZ*NR_CPUS/10) |
3009 | interval = HZ*NR_CPUS/10; | |
3010 | ||
1da177e4 | 3011 | |
08c183f3 CL |
3012 | if (sd->flags & SD_SERIALIZE) { |
3013 | if (!spin_trylock(&balancing)) | |
3014 | goto out; | |
3015 | } | |
3016 | ||
c9819f45 | 3017 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3018 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3019 | /* |
3020 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3021 | * longer idle, or one of our SMT siblings is |
3022 | * not idle. | |
3023 | */ | |
d15bcfdb | 3024 | idle = CPU_NOT_IDLE; |
1da177e4 | 3025 | } |
1bd77f2d | 3026 | sd->last_balance = jiffies; |
1da177e4 | 3027 | } |
08c183f3 CL |
3028 | if (sd->flags & SD_SERIALIZE) |
3029 | spin_unlock(&balancing); | |
3030 | out: | |
f549da84 | 3031 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3032 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3033 | update_next_balance = 1; |
3034 | } | |
783609c6 SS |
3035 | |
3036 | /* | |
3037 | * Stop the load balance at this level. There is another | |
3038 | * CPU in our sched group which is doing load balancing more | |
3039 | * actively. | |
3040 | */ | |
3041 | if (!balance) | |
3042 | break; | |
1da177e4 | 3043 | } |
f549da84 SS |
3044 | |
3045 | /* | |
3046 | * next_balance will be updated only when there is a need. | |
3047 | * When the cpu is attached to null domain for ex, it will not be | |
3048 | * updated. | |
3049 | */ | |
3050 | if (likely(update_next_balance)) | |
3051 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3052 | } |
3053 | ||
3054 | /* | |
3055 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3056 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3057 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3058 | */ | |
3059 | static void run_rebalance_domains(struct softirq_action *h) | |
3060 | { | |
dd41f596 IM |
3061 | int this_cpu = smp_processor_id(); |
3062 | struct rq *this_rq = cpu_rq(this_cpu); | |
3063 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3064 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3065 | |
dd41f596 | 3066 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3067 | |
3068 | #ifdef CONFIG_NO_HZ | |
3069 | /* | |
3070 | * If this cpu is the owner for idle load balancing, then do the | |
3071 | * balancing on behalf of the other idle cpus whose ticks are | |
3072 | * stopped. | |
3073 | */ | |
dd41f596 IM |
3074 | if (this_rq->idle_at_tick && |
3075 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3076 | cpumask_t cpus = nohz.cpu_mask; |
3077 | struct rq *rq; | |
3078 | int balance_cpu; | |
3079 | ||
dd41f596 | 3080 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3081 | for_each_cpu_mask(balance_cpu, cpus) { |
3082 | /* | |
3083 | * If this cpu gets work to do, stop the load balancing | |
3084 | * work being done for other cpus. Next load | |
3085 | * balancing owner will pick it up. | |
3086 | */ | |
3087 | if (need_resched()) | |
3088 | break; | |
3089 | ||
de0cf899 | 3090 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3091 | |
3092 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3093 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3094 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3095 | } |
3096 | } | |
3097 | #endif | |
3098 | } | |
3099 | ||
3100 | /* | |
3101 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3102 | * | |
3103 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3104 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3105 | * if the whole system is idle. | |
3106 | */ | |
dd41f596 | 3107 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3108 | { |
46cb4b7c SS |
3109 | #ifdef CONFIG_NO_HZ |
3110 | /* | |
3111 | * If we were in the nohz mode recently and busy at the current | |
3112 | * scheduler tick, then check if we need to nominate new idle | |
3113 | * load balancer. | |
3114 | */ | |
3115 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3116 | rq->in_nohz_recently = 0; | |
3117 | ||
3118 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3119 | cpu_clear(cpu, nohz.cpu_mask); | |
3120 | atomic_set(&nohz.load_balancer, -1); | |
3121 | } | |
3122 | ||
3123 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3124 | /* | |
3125 | * simple selection for now: Nominate the | |
3126 | * first cpu in the nohz list to be the next | |
3127 | * ilb owner. | |
3128 | * | |
3129 | * TBD: Traverse the sched domains and nominate | |
3130 | * the nearest cpu in the nohz.cpu_mask. | |
3131 | */ | |
3132 | int ilb = first_cpu(nohz.cpu_mask); | |
3133 | ||
3134 | if (ilb != NR_CPUS) | |
3135 | resched_cpu(ilb); | |
3136 | } | |
3137 | } | |
3138 | ||
3139 | /* | |
3140 | * If this cpu is idle and doing idle load balancing for all the | |
3141 | * cpus with ticks stopped, is it time for that to stop? | |
3142 | */ | |
3143 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3144 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3145 | resched_cpu(cpu); | |
3146 | return; | |
3147 | } | |
3148 | ||
3149 | /* | |
3150 | * If this cpu is idle and the idle load balancing is done by | |
3151 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3152 | */ | |
3153 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3154 | cpu_isset(cpu, nohz.cpu_mask)) | |
3155 | return; | |
3156 | #endif | |
3157 | if (time_after_eq(jiffies, rq->next_balance)) | |
3158 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3159 | } |
dd41f596 IM |
3160 | |
3161 | #else /* CONFIG_SMP */ | |
3162 | ||
1da177e4 LT |
3163 | /* |
3164 | * on UP we do not need to balance between CPUs: | |
3165 | */ | |
70b97a7f | 3166 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3167 | { |
3168 | } | |
dd41f596 IM |
3169 | |
3170 | /* Avoid "used but not defined" warning on UP */ | |
3171 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3172 | unsigned long max_nr_move, unsigned long max_load_move, | |
3173 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3174 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3175 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3176 | { |
3177 | *load_moved = 0; | |
3178 | ||
3179 | return 0; | |
3180 | } | |
3181 | ||
1da177e4 LT |
3182 | #endif |
3183 | ||
1da177e4 LT |
3184 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3185 | ||
3186 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3187 | ||
3188 | /* | |
41b86e9c IM |
3189 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3190 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3191 | */ |
41b86e9c | 3192 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3193 | { |
1da177e4 | 3194 | unsigned long flags; |
41b86e9c IM |
3195 | u64 ns, delta_exec; |
3196 | struct rq *rq; | |
48f24c4d | 3197 | |
41b86e9c IM |
3198 | rq = task_rq_lock(p, &flags); |
3199 | ns = p->se.sum_exec_runtime; | |
3200 | if (rq->curr == p) { | |
a8e504d2 IM |
3201 | update_rq_clock(rq); |
3202 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3203 | if ((s64)delta_exec > 0) |
3204 | ns += delta_exec; | |
3205 | } | |
3206 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3207 | |
1da177e4 LT |
3208 | return ns; |
3209 | } | |
3210 | ||
1da177e4 LT |
3211 | /* |
3212 | * Account user cpu time to a process. | |
3213 | * @p: the process that the cpu time gets accounted to | |
3214 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3215 | * @cputime: the cpu time spent in user space since the last update | |
3216 | */ | |
3217 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3218 | { | |
3219 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3220 | cputime64_t tmp; | |
3221 | ||
3222 | p->utime = cputime_add(p->utime, cputime); | |
3223 | ||
3224 | /* Add user time to cpustat. */ | |
3225 | tmp = cputime_to_cputime64(cputime); | |
3226 | if (TASK_NICE(p) > 0) | |
3227 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3228 | else | |
3229 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3230 | } | |
3231 | ||
3232 | /* | |
3233 | * Account system cpu time to a process. | |
3234 | * @p: the process that the cpu time gets accounted to | |
3235 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3236 | * @cputime: the cpu time spent in kernel space since the last update | |
3237 | */ | |
3238 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3239 | cputime_t cputime) | |
3240 | { | |
3241 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3242 | struct rq *rq = this_rq(); |
1da177e4 LT |
3243 | cputime64_t tmp; |
3244 | ||
3245 | p->stime = cputime_add(p->stime, cputime); | |
3246 | ||
3247 | /* Add system time to cpustat. */ | |
3248 | tmp = cputime_to_cputime64(cputime); | |
3249 | if (hardirq_count() - hardirq_offset) | |
3250 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3251 | else if (softirq_count()) | |
3252 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3253 | else if (p != rq->idle) | |
3254 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3255 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3256 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3257 | else | |
3258 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3259 | /* Account for system time used */ | |
3260 | acct_update_integrals(p); | |
1da177e4 LT |
3261 | } |
3262 | ||
3263 | /* | |
3264 | * Account for involuntary wait time. | |
3265 | * @p: the process from which the cpu time has been stolen | |
3266 | * @steal: the cpu time spent in involuntary wait | |
3267 | */ | |
3268 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3269 | { | |
3270 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3271 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3272 | struct rq *rq = this_rq(); |
1da177e4 LT |
3273 | |
3274 | if (p == rq->idle) { | |
3275 | p->stime = cputime_add(p->stime, steal); | |
3276 | if (atomic_read(&rq->nr_iowait) > 0) | |
3277 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3278 | else | |
3279 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3280 | } else | |
3281 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3282 | } | |
3283 | ||
7835b98b CL |
3284 | /* |
3285 | * This function gets called by the timer code, with HZ frequency. | |
3286 | * We call it with interrupts disabled. | |
3287 | * | |
3288 | * It also gets called by the fork code, when changing the parent's | |
3289 | * timeslices. | |
3290 | */ | |
3291 | void scheduler_tick(void) | |
3292 | { | |
7835b98b CL |
3293 | int cpu = smp_processor_id(); |
3294 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3295 | struct task_struct *curr = rq->curr; |
529c7726 | 3296 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3297 | |
3298 | spin_lock(&rq->lock); | |
546fe3c9 | 3299 | __update_rq_clock(rq); |
529c7726 IM |
3300 | /* |
3301 | * Let rq->clock advance by at least TICK_NSEC: | |
3302 | */ | |
3303 | if (unlikely(rq->clock < next_tick)) | |
3304 | rq->clock = next_tick; | |
3305 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3306 | update_cpu_load(rq); |
dd41f596 IM |
3307 | if (curr != rq->idle) /* FIXME: needed? */ |
3308 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3309 | spin_unlock(&rq->lock); |
7835b98b | 3310 | |
e418e1c2 | 3311 | #ifdef CONFIG_SMP |
dd41f596 IM |
3312 | rq->idle_at_tick = idle_cpu(cpu); |
3313 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3314 | #endif |
1da177e4 LT |
3315 | } |
3316 | ||
1da177e4 LT |
3317 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3318 | ||
3319 | void fastcall add_preempt_count(int val) | |
3320 | { | |
3321 | /* | |
3322 | * Underflow? | |
3323 | */ | |
9a11b49a IM |
3324 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3325 | return; | |
1da177e4 LT |
3326 | preempt_count() += val; |
3327 | /* | |
3328 | * Spinlock count overflowing soon? | |
3329 | */ | |
33859f7f MOS |
3330 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3331 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3332 | } |
3333 | EXPORT_SYMBOL(add_preempt_count); | |
3334 | ||
3335 | void fastcall sub_preempt_count(int val) | |
3336 | { | |
3337 | /* | |
3338 | * Underflow? | |
3339 | */ | |
9a11b49a IM |
3340 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3341 | return; | |
1da177e4 LT |
3342 | /* |
3343 | * Is the spinlock portion underflowing? | |
3344 | */ | |
9a11b49a IM |
3345 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3346 | !(preempt_count() & PREEMPT_MASK))) | |
3347 | return; | |
3348 | ||
1da177e4 LT |
3349 | preempt_count() -= val; |
3350 | } | |
3351 | EXPORT_SYMBOL(sub_preempt_count); | |
3352 | ||
3353 | #endif | |
3354 | ||
3355 | /* | |
dd41f596 | 3356 | * Print scheduling while atomic bug: |
1da177e4 | 3357 | */ |
dd41f596 | 3358 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3359 | { |
dd41f596 IM |
3360 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3361 | prev->comm, preempt_count(), prev->pid); | |
3362 | debug_show_held_locks(prev); | |
3363 | if (irqs_disabled()) | |
3364 | print_irqtrace_events(prev); | |
3365 | dump_stack(); | |
3366 | } | |
1da177e4 | 3367 | |
dd41f596 IM |
3368 | /* |
3369 | * Various schedule()-time debugging checks and statistics: | |
3370 | */ | |
3371 | static inline void schedule_debug(struct task_struct *prev) | |
3372 | { | |
1da177e4 LT |
3373 | /* |
3374 | * Test if we are atomic. Since do_exit() needs to call into | |
3375 | * schedule() atomically, we ignore that path for now. | |
3376 | * Otherwise, whine if we are scheduling when we should not be. | |
3377 | */ | |
dd41f596 IM |
3378 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3379 | __schedule_bug(prev); | |
3380 | ||
1da177e4 LT |
3381 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3382 | ||
2d72376b | 3383 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3384 | #ifdef CONFIG_SCHEDSTATS |
3385 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3386 | schedstat_inc(this_rq(), bkl_count); |
3387 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3388 | } |
3389 | #endif | |
dd41f596 IM |
3390 | } |
3391 | ||
3392 | /* | |
3393 | * Pick up the highest-prio task: | |
3394 | */ | |
3395 | static inline struct task_struct * | |
ff95f3df | 3396 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3397 | { |
5522d5d5 | 3398 | const struct sched_class *class; |
dd41f596 | 3399 | struct task_struct *p; |
1da177e4 LT |
3400 | |
3401 | /* | |
dd41f596 IM |
3402 | * Optimization: we know that if all tasks are in |
3403 | * the fair class we can call that function directly: | |
1da177e4 | 3404 | */ |
dd41f596 | 3405 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3406 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3407 | if (likely(p)) |
3408 | return p; | |
1da177e4 LT |
3409 | } |
3410 | ||
dd41f596 IM |
3411 | class = sched_class_highest; |
3412 | for ( ; ; ) { | |
fb8d4724 | 3413 | p = class->pick_next_task(rq); |
dd41f596 IM |
3414 | if (p) |
3415 | return p; | |
3416 | /* | |
3417 | * Will never be NULL as the idle class always | |
3418 | * returns a non-NULL p: | |
3419 | */ | |
3420 | class = class->next; | |
3421 | } | |
3422 | } | |
1da177e4 | 3423 | |
dd41f596 IM |
3424 | /* |
3425 | * schedule() is the main scheduler function. | |
3426 | */ | |
3427 | asmlinkage void __sched schedule(void) | |
3428 | { | |
3429 | struct task_struct *prev, *next; | |
3430 | long *switch_count; | |
3431 | struct rq *rq; | |
dd41f596 IM |
3432 | int cpu; |
3433 | ||
3434 | need_resched: | |
3435 | preempt_disable(); | |
3436 | cpu = smp_processor_id(); | |
3437 | rq = cpu_rq(cpu); | |
3438 | rcu_qsctr_inc(cpu); | |
3439 | prev = rq->curr; | |
3440 | switch_count = &prev->nivcsw; | |
3441 | ||
3442 | release_kernel_lock(prev); | |
3443 | need_resched_nonpreemptible: | |
3444 | ||
3445 | schedule_debug(prev); | |
1da177e4 | 3446 | |
1e819950 IM |
3447 | /* |
3448 | * Do the rq-clock update outside the rq lock: | |
3449 | */ | |
3450 | local_irq_disable(); | |
c1b3da3e | 3451 | __update_rq_clock(rq); |
1e819950 IM |
3452 | spin_lock(&rq->lock); |
3453 | clear_tsk_need_resched(prev); | |
1da177e4 | 3454 | |
1da177e4 | 3455 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3456 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3457 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3458 | prev->state = TASK_RUNNING; |
dd41f596 | 3459 | } else { |
2e1cb74a | 3460 | deactivate_task(rq, prev, 1); |
1da177e4 | 3461 | } |
dd41f596 | 3462 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3463 | } |
3464 | ||
dd41f596 | 3465 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3466 | idle_balance(cpu, rq); |
1da177e4 | 3467 | |
31ee529c | 3468 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3469 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3470 | |
3471 | sched_info_switch(prev, next); | |
dd41f596 | 3472 | |
1da177e4 | 3473 | if (likely(prev != next)) { |
1da177e4 LT |
3474 | rq->nr_switches++; |
3475 | rq->curr = next; | |
3476 | ++*switch_count; | |
3477 | ||
dd41f596 | 3478 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3479 | } else |
3480 | spin_unlock_irq(&rq->lock); | |
3481 | ||
dd41f596 IM |
3482 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3483 | cpu = smp_processor_id(); | |
3484 | rq = cpu_rq(cpu); | |
1da177e4 | 3485 | goto need_resched_nonpreemptible; |
dd41f596 | 3486 | } |
1da177e4 LT |
3487 | preempt_enable_no_resched(); |
3488 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3489 | goto need_resched; | |
3490 | } | |
1da177e4 LT |
3491 | EXPORT_SYMBOL(schedule); |
3492 | ||
3493 | #ifdef CONFIG_PREEMPT | |
3494 | /* | |
2ed6e34f | 3495 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3496 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3497 | * occur there and call schedule directly. | |
3498 | */ | |
3499 | asmlinkage void __sched preempt_schedule(void) | |
3500 | { | |
3501 | struct thread_info *ti = current_thread_info(); | |
3502 | #ifdef CONFIG_PREEMPT_BKL | |
3503 | struct task_struct *task = current; | |
3504 | int saved_lock_depth; | |
3505 | #endif | |
3506 | /* | |
3507 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3508 | * we do not want to preempt the current task. Just return.. | |
3509 | */ | |
beed33a8 | 3510 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3511 | return; |
3512 | ||
3513 | need_resched: | |
3514 | add_preempt_count(PREEMPT_ACTIVE); | |
3515 | /* | |
3516 | * We keep the big kernel semaphore locked, but we | |
3517 | * clear ->lock_depth so that schedule() doesnt | |
3518 | * auto-release the semaphore: | |
3519 | */ | |
3520 | #ifdef CONFIG_PREEMPT_BKL | |
3521 | saved_lock_depth = task->lock_depth; | |
3522 | task->lock_depth = -1; | |
3523 | #endif | |
3524 | schedule(); | |
3525 | #ifdef CONFIG_PREEMPT_BKL | |
3526 | task->lock_depth = saved_lock_depth; | |
3527 | #endif | |
3528 | sub_preempt_count(PREEMPT_ACTIVE); | |
3529 | ||
3530 | /* we could miss a preemption opportunity between schedule and now */ | |
3531 | barrier(); | |
3532 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3533 | goto need_resched; | |
3534 | } | |
1da177e4 LT |
3535 | EXPORT_SYMBOL(preempt_schedule); |
3536 | ||
3537 | /* | |
2ed6e34f | 3538 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3539 | * off of irq context. |
3540 | * Note, that this is called and return with irqs disabled. This will | |
3541 | * protect us against recursive calling from irq. | |
3542 | */ | |
3543 | asmlinkage void __sched preempt_schedule_irq(void) | |
3544 | { | |
3545 | struct thread_info *ti = current_thread_info(); | |
3546 | #ifdef CONFIG_PREEMPT_BKL | |
3547 | struct task_struct *task = current; | |
3548 | int saved_lock_depth; | |
3549 | #endif | |
2ed6e34f | 3550 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3551 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3552 | ||
3553 | need_resched: | |
3554 | add_preempt_count(PREEMPT_ACTIVE); | |
3555 | /* | |
3556 | * We keep the big kernel semaphore locked, but we | |
3557 | * clear ->lock_depth so that schedule() doesnt | |
3558 | * auto-release the semaphore: | |
3559 | */ | |
3560 | #ifdef CONFIG_PREEMPT_BKL | |
3561 | saved_lock_depth = task->lock_depth; | |
3562 | task->lock_depth = -1; | |
3563 | #endif | |
3564 | local_irq_enable(); | |
3565 | schedule(); | |
3566 | local_irq_disable(); | |
3567 | #ifdef CONFIG_PREEMPT_BKL | |
3568 | task->lock_depth = saved_lock_depth; | |
3569 | #endif | |
3570 | sub_preempt_count(PREEMPT_ACTIVE); | |
3571 | ||
3572 | /* we could miss a preemption opportunity between schedule and now */ | |
3573 | barrier(); | |
3574 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3575 | goto need_resched; | |
3576 | } | |
3577 | ||
3578 | #endif /* CONFIG_PREEMPT */ | |
3579 | ||
95cdf3b7 IM |
3580 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3581 | void *key) | |
1da177e4 | 3582 | { |
48f24c4d | 3583 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3584 | } |
1da177e4 LT |
3585 | EXPORT_SYMBOL(default_wake_function); |
3586 | ||
3587 | /* | |
3588 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3589 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3590 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3591 | * | |
3592 | * There are circumstances in which we can try to wake a task which has already | |
3593 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3594 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3595 | */ | |
3596 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3597 | int nr_exclusive, int sync, void *key) | |
3598 | { | |
2e45874c | 3599 | wait_queue_t *curr, *next; |
1da177e4 | 3600 | |
2e45874c | 3601 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3602 | unsigned flags = curr->flags; |
3603 | ||
1da177e4 | 3604 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3605 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3606 | break; |
3607 | } | |
3608 | } | |
3609 | ||
3610 | /** | |
3611 | * __wake_up - wake up threads blocked on a waitqueue. | |
3612 | * @q: the waitqueue | |
3613 | * @mode: which threads | |
3614 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3615 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3616 | */ |
3617 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3618 | int nr_exclusive, void *key) |
1da177e4 LT |
3619 | { |
3620 | unsigned long flags; | |
3621 | ||
3622 | spin_lock_irqsave(&q->lock, flags); | |
3623 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3624 | spin_unlock_irqrestore(&q->lock, flags); | |
3625 | } | |
1da177e4 LT |
3626 | EXPORT_SYMBOL(__wake_up); |
3627 | ||
3628 | /* | |
3629 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3630 | */ | |
3631 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3632 | { | |
3633 | __wake_up_common(q, mode, 1, 0, NULL); | |
3634 | } | |
3635 | ||
3636 | /** | |
67be2dd1 | 3637 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3638 | * @q: the waitqueue |
3639 | * @mode: which threads | |
3640 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3641 | * | |
3642 | * The sync wakeup differs that the waker knows that it will schedule | |
3643 | * away soon, so while the target thread will be woken up, it will not | |
3644 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3645 | * with each other. This can prevent needless bouncing between CPUs. | |
3646 | * | |
3647 | * On UP it can prevent extra preemption. | |
3648 | */ | |
95cdf3b7 IM |
3649 | void fastcall |
3650 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3651 | { |
3652 | unsigned long flags; | |
3653 | int sync = 1; | |
3654 | ||
3655 | if (unlikely(!q)) | |
3656 | return; | |
3657 | ||
3658 | if (unlikely(!nr_exclusive)) | |
3659 | sync = 0; | |
3660 | ||
3661 | spin_lock_irqsave(&q->lock, flags); | |
3662 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3663 | spin_unlock_irqrestore(&q->lock, flags); | |
3664 | } | |
3665 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3666 | ||
3667 | void fastcall complete(struct completion *x) | |
3668 | { | |
3669 | unsigned long flags; | |
3670 | ||
3671 | spin_lock_irqsave(&x->wait.lock, flags); | |
3672 | x->done++; | |
3673 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3674 | 1, 0, NULL); | |
3675 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3676 | } | |
3677 | EXPORT_SYMBOL(complete); | |
3678 | ||
3679 | void fastcall complete_all(struct completion *x) | |
3680 | { | |
3681 | unsigned long flags; | |
3682 | ||
3683 | spin_lock_irqsave(&x->wait.lock, flags); | |
3684 | x->done += UINT_MAX/2; | |
3685 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3686 | 0, 0, NULL); | |
3687 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3688 | } | |
3689 | EXPORT_SYMBOL(complete_all); | |
3690 | ||
3691 | void fastcall __sched wait_for_completion(struct completion *x) | |
3692 | { | |
3693 | might_sleep(); | |
48f24c4d | 3694 | |
1da177e4 LT |
3695 | spin_lock_irq(&x->wait.lock); |
3696 | if (!x->done) { | |
3697 | DECLARE_WAITQUEUE(wait, current); | |
3698 | ||
3699 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3700 | __add_wait_queue_tail(&x->wait, &wait); | |
3701 | do { | |
3702 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3703 | spin_unlock_irq(&x->wait.lock); | |
3704 | schedule(); | |
3705 | spin_lock_irq(&x->wait.lock); | |
3706 | } while (!x->done); | |
3707 | __remove_wait_queue(&x->wait, &wait); | |
3708 | } | |
3709 | x->done--; | |
3710 | spin_unlock_irq(&x->wait.lock); | |
3711 | } | |
3712 | EXPORT_SYMBOL(wait_for_completion); | |
3713 | ||
3714 | unsigned long fastcall __sched | |
3715 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) | |
3716 | { | |
3717 | might_sleep(); | |
3718 | ||
3719 | spin_lock_irq(&x->wait.lock); | |
3720 | if (!x->done) { | |
3721 | DECLARE_WAITQUEUE(wait, current); | |
3722 | ||
3723 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3724 | __add_wait_queue_tail(&x->wait, &wait); | |
3725 | do { | |
3726 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
3727 | spin_unlock_irq(&x->wait.lock); | |
3728 | timeout = schedule_timeout(timeout); | |
3729 | spin_lock_irq(&x->wait.lock); | |
3730 | if (!timeout) { | |
3731 | __remove_wait_queue(&x->wait, &wait); | |
3732 | goto out; | |
3733 | } | |
3734 | } while (!x->done); | |
3735 | __remove_wait_queue(&x->wait, &wait); | |
3736 | } | |
3737 | x->done--; | |
3738 | out: | |
3739 | spin_unlock_irq(&x->wait.lock); | |
3740 | return timeout; | |
3741 | } | |
3742 | EXPORT_SYMBOL(wait_for_completion_timeout); | |
3743 | ||
3744 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) | |
3745 | { | |
3746 | int ret = 0; | |
3747 | ||
3748 | might_sleep(); | |
3749 | ||
3750 | spin_lock_irq(&x->wait.lock); | |
3751 | if (!x->done) { | |
3752 | DECLARE_WAITQUEUE(wait, current); | |
3753 | ||
3754 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3755 | __add_wait_queue_tail(&x->wait, &wait); | |
3756 | do { | |
3757 | if (signal_pending(current)) { | |
3758 | ret = -ERESTARTSYS; | |
3759 | __remove_wait_queue(&x->wait, &wait); | |
3760 | goto out; | |
3761 | } | |
3762 | __set_current_state(TASK_INTERRUPTIBLE); | |
3763 | spin_unlock_irq(&x->wait.lock); | |
3764 | schedule(); | |
3765 | spin_lock_irq(&x->wait.lock); | |
3766 | } while (!x->done); | |
3767 | __remove_wait_queue(&x->wait, &wait); | |
3768 | } | |
3769 | x->done--; | |
3770 | out: | |
3771 | spin_unlock_irq(&x->wait.lock); | |
3772 | ||
3773 | return ret; | |
3774 | } | |
3775 | EXPORT_SYMBOL(wait_for_completion_interruptible); | |
3776 | ||
3777 | unsigned long fastcall __sched | |
3778 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3779 | unsigned long timeout) | |
3780 | { | |
3781 | might_sleep(); | |
3782 | ||
3783 | spin_lock_irq(&x->wait.lock); | |
3784 | if (!x->done) { | |
3785 | DECLARE_WAITQUEUE(wait, current); | |
3786 | ||
3787 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3788 | __add_wait_queue_tail(&x->wait, &wait); | |
3789 | do { | |
3790 | if (signal_pending(current)) { | |
3791 | timeout = -ERESTARTSYS; | |
3792 | __remove_wait_queue(&x->wait, &wait); | |
3793 | goto out; | |
3794 | } | |
3795 | __set_current_state(TASK_INTERRUPTIBLE); | |
3796 | spin_unlock_irq(&x->wait.lock); | |
3797 | timeout = schedule_timeout(timeout); | |
3798 | spin_lock_irq(&x->wait.lock); | |
3799 | if (!timeout) { | |
3800 | __remove_wait_queue(&x->wait, &wait); | |
3801 | goto out; | |
3802 | } | |
3803 | } while (!x->done); | |
3804 | __remove_wait_queue(&x->wait, &wait); | |
3805 | } | |
3806 | x->done--; | |
3807 | out: | |
3808 | spin_unlock_irq(&x->wait.lock); | |
3809 | return timeout; | |
3810 | } | |
3811 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); | |
3812 | ||
0fec171c IM |
3813 | static inline void |
3814 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3815 | { | |
3816 | spin_lock_irqsave(&q->lock, *flags); | |
3817 | __add_wait_queue(q, wait); | |
1da177e4 | 3818 | spin_unlock(&q->lock); |
0fec171c | 3819 | } |
1da177e4 | 3820 | |
0fec171c IM |
3821 | static inline void |
3822 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) | |
3823 | { | |
3824 | spin_lock_irq(&q->lock); | |
3825 | __remove_wait_queue(q, wait); | |
3826 | spin_unlock_irqrestore(&q->lock, *flags); | |
3827 | } | |
1da177e4 | 3828 | |
0fec171c | 3829 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
1da177e4 | 3830 | { |
0fec171c IM |
3831 | unsigned long flags; |
3832 | wait_queue_t wait; | |
3833 | ||
3834 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3835 | |
3836 | current->state = TASK_INTERRUPTIBLE; | |
3837 | ||
0fec171c | 3838 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3839 | schedule(); |
0fec171c | 3840 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3841 | } |
1da177e4 LT |
3842 | EXPORT_SYMBOL(interruptible_sleep_on); |
3843 | ||
0fec171c | 3844 | long __sched |
95cdf3b7 | 3845 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3846 | { |
0fec171c IM |
3847 | unsigned long flags; |
3848 | wait_queue_t wait; | |
3849 | ||
3850 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3851 | |
3852 | current->state = TASK_INTERRUPTIBLE; | |
3853 | ||
0fec171c | 3854 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3855 | timeout = schedule_timeout(timeout); |
0fec171c | 3856 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3857 | |
3858 | return timeout; | |
3859 | } | |
1da177e4 LT |
3860 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3861 | ||
0fec171c | 3862 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3863 | { |
0fec171c IM |
3864 | unsigned long flags; |
3865 | wait_queue_t wait; | |
3866 | ||
3867 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3868 | |
3869 | current->state = TASK_UNINTERRUPTIBLE; | |
3870 | ||
0fec171c | 3871 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3872 | schedule(); |
0fec171c | 3873 | sleep_on_tail(q, &wait, &flags); |
1da177e4 | 3874 | } |
1da177e4 LT |
3875 | EXPORT_SYMBOL(sleep_on); |
3876 | ||
0fec171c | 3877 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3878 | { |
0fec171c IM |
3879 | unsigned long flags; |
3880 | wait_queue_t wait; | |
3881 | ||
3882 | init_waitqueue_entry(&wait, current); | |
1da177e4 LT |
3883 | |
3884 | current->state = TASK_UNINTERRUPTIBLE; | |
3885 | ||
0fec171c | 3886 | sleep_on_head(q, &wait, &flags); |
1da177e4 | 3887 | timeout = schedule_timeout(timeout); |
0fec171c | 3888 | sleep_on_tail(q, &wait, &flags); |
1da177e4 LT |
3889 | |
3890 | return timeout; | |
3891 | } | |
1da177e4 LT |
3892 | EXPORT_SYMBOL(sleep_on_timeout); |
3893 | ||
b29739f9 IM |
3894 | #ifdef CONFIG_RT_MUTEXES |
3895 | ||
3896 | /* | |
3897 | * rt_mutex_setprio - set the current priority of a task | |
3898 | * @p: task | |
3899 | * @prio: prio value (kernel-internal form) | |
3900 | * | |
3901 | * This function changes the 'effective' priority of a task. It does | |
3902 | * not touch ->normal_prio like __setscheduler(). | |
3903 | * | |
3904 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3905 | */ | |
36c8b586 | 3906 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3907 | { |
3908 | unsigned long flags; | |
83b699ed | 3909 | int oldprio, on_rq, running; |
70b97a7f | 3910 | struct rq *rq; |
b29739f9 IM |
3911 | |
3912 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3913 | ||
3914 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3915 | update_rq_clock(rq); |
b29739f9 | 3916 | |
d5f9f942 | 3917 | oldprio = p->prio; |
dd41f596 | 3918 | on_rq = p->se.on_rq; |
83b699ed SV |
3919 | running = task_running(rq, p); |
3920 | if (on_rq) { | |
69be72c1 | 3921 | dequeue_task(rq, p, 0); |
83b699ed SV |
3922 | if (running) |
3923 | p->sched_class->put_prev_task(rq, p); | |
3924 | } | |
dd41f596 IM |
3925 | |
3926 | if (rt_prio(prio)) | |
3927 | p->sched_class = &rt_sched_class; | |
3928 | else | |
3929 | p->sched_class = &fair_sched_class; | |
3930 | ||
b29739f9 IM |
3931 | p->prio = prio; |
3932 | ||
dd41f596 | 3933 | if (on_rq) { |
83b699ed SV |
3934 | if (running) |
3935 | p->sched_class->set_curr_task(rq); | |
8159f87e | 3936 | enqueue_task(rq, p, 0); |
b29739f9 IM |
3937 | /* |
3938 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3939 | * our priority decreased, or if we are not currently running on |
3940 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3941 | */ |
83b699ed | 3942 | if (running) { |
d5f9f942 AM |
3943 | if (p->prio > oldprio) |
3944 | resched_task(rq->curr); | |
dd41f596 IM |
3945 | } else { |
3946 | check_preempt_curr(rq, p); | |
3947 | } | |
b29739f9 IM |
3948 | } |
3949 | task_rq_unlock(rq, &flags); | |
3950 | } | |
3951 | ||
3952 | #endif | |
3953 | ||
36c8b586 | 3954 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3955 | { |
dd41f596 | 3956 | int old_prio, delta, on_rq; |
1da177e4 | 3957 | unsigned long flags; |
70b97a7f | 3958 | struct rq *rq; |
1da177e4 LT |
3959 | |
3960 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3961 | return; | |
3962 | /* | |
3963 | * We have to be careful, if called from sys_setpriority(), | |
3964 | * the task might be in the middle of scheduling on another CPU. | |
3965 | */ | |
3966 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3967 | update_rq_clock(rq); |
1da177e4 LT |
3968 | /* |
3969 | * The RT priorities are set via sched_setscheduler(), but we still | |
3970 | * allow the 'normal' nice value to be set - but as expected | |
3971 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3972 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3973 | */ |
e05606d3 | 3974 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3975 | p->static_prio = NICE_TO_PRIO(nice); |
3976 | goto out_unlock; | |
3977 | } | |
dd41f596 IM |
3978 | on_rq = p->se.on_rq; |
3979 | if (on_rq) { | |
69be72c1 | 3980 | dequeue_task(rq, p, 0); |
79b5dddf | 3981 | dec_load(rq, p); |
2dd73a4f | 3982 | } |
1da177e4 | 3983 | |
1da177e4 | 3984 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3985 | set_load_weight(p); |
b29739f9 IM |
3986 | old_prio = p->prio; |
3987 | p->prio = effective_prio(p); | |
3988 | delta = p->prio - old_prio; | |
1da177e4 | 3989 | |
dd41f596 | 3990 | if (on_rq) { |
8159f87e | 3991 | enqueue_task(rq, p, 0); |
29b4b623 | 3992 | inc_load(rq, p); |
1da177e4 | 3993 | /* |
d5f9f942 AM |
3994 | * If the task increased its priority or is running and |
3995 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3996 | */ |
d5f9f942 | 3997 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3998 | resched_task(rq->curr); |
3999 | } | |
4000 | out_unlock: | |
4001 | task_rq_unlock(rq, &flags); | |
4002 | } | |
1da177e4 LT |
4003 | EXPORT_SYMBOL(set_user_nice); |
4004 | ||
e43379f1 MM |
4005 | /* |
4006 | * can_nice - check if a task can reduce its nice value | |
4007 | * @p: task | |
4008 | * @nice: nice value | |
4009 | */ | |
36c8b586 | 4010 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4011 | { |
024f4747 MM |
4012 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4013 | int nice_rlim = 20 - nice; | |
48f24c4d | 4014 | |
e43379f1 MM |
4015 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4016 | capable(CAP_SYS_NICE)); | |
4017 | } | |
4018 | ||
1da177e4 LT |
4019 | #ifdef __ARCH_WANT_SYS_NICE |
4020 | ||
4021 | /* | |
4022 | * sys_nice - change the priority of the current process. | |
4023 | * @increment: priority increment | |
4024 | * | |
4025 | * sys_setpriority is a more generic, but much slower function that | |
4026 | * does similar things. | |
4027 | */ | |
4028 | asmlinkage long sys_nice(int increment) | |
4029 | { | |
48f24c4d | 4030 | long nice, retval; |
1da177e4 LT |
4031 | |
4032 | /* | |
4033 | * Setpriority might change our priority at the same moment. | |
4034 | * We don't have to worry. Conceptually one call occurs first | |
4035 | * and we have a single winner. | |
4036 | */ | |
e43379f1 MM |
4037 | if (increment < -40) |
4038 | increment = -40; | |
1da177e4 LT |
4039 | if (increment > 40) |
4040 | increment = 40; | |
4041 | ||
4042 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4043 | if (nice < -20) | |
4044 | nice = -20; | |
4045 | if (nice > 19) | |
4046 | nice = 19; | |
4047 | ||
e43379f1 MM |
4048 | if (increment < 0 && !can_nice(current, nice)) |
4049 | return -EPERM; | |
4050 | ||
1da177e4 LT |
4051 | retval = security_task_setnice(current, nice); |
4052 | if (retval) | |
4053 | return retval; | |
4054 | ||
4055 | set_user_nice(current, nice); | |
4056 | return 0; | |
4057 | } | |
4058 | ||
4059 | #endif | |
4060 | ||
4061 | /** | |
4062 | * task_prio - return the priority value of a given task. | |
4063 | * @p: the task in question. | |
4064 | * | |
4065 | * This is the priority value as seen by users in /proc. | |
4066 | * RT tasks are offset by -200. Normal tasks are centered | |
4067 | * around 0, value goes from -16 to +15. | |
4068 | */ | |
36c8b586 | 4069 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4070 | { |
4071 | return p->prio - MAX_RT_PRIO; | |
4072 | } | |
4073 | ||
4074 | /** | |
4075 | * task_nice - return the nice value of a given task. | |
4076 | * @p: the task in question. | |
4077 | */ | |
36c8b586 | 4078 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4079 | { |
4080 | return TASK_NICE(p); | |
4081 | } | |
1da177e4 | 4082 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4083 | |
4084 | /** | |
4085 | * idle_cpu - is a given cpu idle currently? | |
4086 | * @cpu: the processor in question. | |
4087 | */ | |
4088 | int idle_cpu(int cpu) | |
4089 | { | |
4090 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4091 | } | |
4092 | ||
1da177e4 LT |
4093 | /** |
4094 | * idle_task - return the idle task for a given cpu. | |
4095 | * @cpu: the processor in question. | |
4096 | */ | |
36c8b586 | 4097 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4098 | { |
4099 | return cpu_rq(cpu)->idle; | |
4100 | } | |
4101 | ||
4102 | /** | |
4103 | * find_process_by_pid - find a process with a matching PID value. | |
4104 | * @pid: the pid in question. | |
4105 | */ | |
a9957449 | 4106 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4107 | { |
4108 | return pid ? find_task_by_pid(pid) : current; | |
4109 | } | |
4110 | ||
4111 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4112 | static void |
4113 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4114 | { |
dd41f596 | 4115 | BUG_ON(p->se.on_rq); |
48f24c4d | 4116 | |
1da177e4 | 4117 | p->policy = policy; |
dd41f596 IM |
4118 | switch (p->policy) { |
4119 | case SCHED_NORMAL: | |
4120 | case SCHED_BATCH: | |
4121 | case SCHED_IDLE: | |
4122 | p->sched_class = &fair_sched_class; | |
4123 | break; | |
4124 | case SCHED_FIFO: | |
4125 | case SCHED_RR: | |
4126 | p->sched_class = &rt_sched_class; | |
4127 | break; | |
4128 | } | |
4129 | ||
1da177e4 | 4130 | p->rt_priority = prio; |
b29739f9 IM |
4131 | p->normal_prio = normal_prio(p); |
4132 | /* we are holding p->pi_lock already */ | |
4133 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4134 | set_load_weight(p); |
1da177e4 LT |
4135 | } |
4136 | ||
4137 | /** | |
72fd4a35 | 4138 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4139 | * @p: the task in question. |
4140 | * @policy: new policy. | |
4141 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4142 | * |
72fd4a35 | 4143 | * NOTE that the task may be already dead. |
1da177e4 | 4144 | */ |
95cdf3b7 IM |
4145 | int sched_setscheduler(struct task_struct *p, int policy, |
4146 | struct sched_param *param) | |
1da177e4 | 4147 | { |
83b699ed | 4148 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4149 | unsigned long flags; |
70b97a7f | 4150 | struct rq *rq; |
1da177e4 | 4151 | |
66e5393a SR |
4152 | /* may grab non-irq protected spin_locks */ |
4153 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4154 | recheck: |
4155 | /* double check policy once rq lock held */ | |
4156 | if (policy < 0) | |
4157 | policy = oldpolicy = p->policy; | |
4158 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4159 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4160 | policy != SCHED_IDLE) | |
b0a9499c | 4161 | return -EINVAL; |
1da177e4 LT |
4162 | /* |
4163 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4164 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4165 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4166 | */ |
4167 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4168 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4169 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4170 | return -EINVAL; |
e05606d3 | 4171 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4172 | return -EINVAL; |
4173 | ||
37e4ab3f OC |
4174 | /* |
4175 | * Allow unprivileged RT tasks to decrease priority: | |
4176 | */ | |
4177 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4178 | if (rt_policy(policy)) { |
8dc3e909 | 4179 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4180 | |
4181 | if (!lock_task_sighand(p, &flags)) | |
4182 | return -ESRCH; | |
4183 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4184 | unlock_task_sighand(p, &flags); | |
4185 | ||
4186 | /* can't set/change the rt policy */ | |
4187 | if (policy != p->policy && !rlim_rtprio) | |
4188 | return -EPERM; | |
4189 | ||
4190 | /* can't increase priority */ | |
4191 | if (param->sched_priority > p->rt_priority && | |
4192 | param->sched_priority > rlim_rtprio) | |
4193 | return -EPERM; | |
4194 | } | |
dd41f596 IM |
4195 | /* |
4196 | * Like positive nice levels, dont allow tasks to | |
4197 | * move out of SCHED_IDLE either: | |
4198 | */ | |
4199 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4200 | return -EPERM; | |
5fe1d75f | 4201 | |
37e4ab3f OC |
4202 | /* can't change other user's priorities */ |
4203 | if ((current->euid != p->euid) && | |
4204 | (current->euid != p->uid)) | |
4205 | return -EPERM; | |
4206 | } | |
1da177e4 LT |
4207 | |
4208 | retval = security_task_setscheduler(p, policy, param); | |
4209 | if (retval) | |
4210 | return retval; | |
b29739f9 IM |
4211 | /* |
4212 | * make sure no PI-waiters arrive (or leave) while we are | |
4213 | * changing the priority of the task: | |
4214 | */ | |
4215 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4216 | /* |
4217 | * To be able to change p->policy safely, the apropriate | |
4218 | * runqueue lock must be held. | |
4219 | */ | |
b29739f9 | 4220 | rq = __task_rq_lock(p); |
1da177e4 LT |
4221 | /* recheck policy now with rq lock held */ |
4222 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4223 | policy = oldpolicy = -1; | |
b29739f9 IM |
4224 | __task_rq_unlock(rq); |
4225 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4226 | goto recheck; |
4227 | } | |
2daa3577 | 4228 | update_rq_clock(rq); |
dd41f596 | 4229 | on_rq = p->se.on_rq; |
83b699ed SV |
4230 | running = task_running(rq, p); |
4231 | if (on_rq) { | |
2e1cb74a | 4232 | deactivate_task(rq, p, 0); |
83b699ed SV |
4233 | if (running) |
4234 | p->sched_class->put_prev_task(rq, p); | |
4235 | } | |
f6b53205 | 4236 | |
1da177e4 | 4237 | oldprio = p->prio; |
dd41f596 | 4238 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4239 | |
dd41f596 | 4240 | if (on_rq) { |
83b699ed SV |
4241 | if (running) |
4242 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4243 | activate_task(rq, p, 0); |
1da177e4 LT |
4244 | /* |
4245 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4246 | * our priority decreased, or if we are not currently running on |
4247 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4248 | */ |
83b699ed | 4249 | if (running) { |
d5f9f942 AM |
4250 | if (p->prio > oldprio) |
4251 | resched_task(rq->curr); | |
dd41f596 IM |
4252 | } else { |
4253 | check_preempt_curr(rq, p); | |
4254 | } | |
1da177e4 | 4255 | } |
b29739f9 IM |
4256 | __task_rq_unlock(rq); |
4257 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4258 | ||
95e02ca9 TG |
4259 | rt_mutex_adjust_pi(p); |
4260 | ||
1da177e4 LT |
4261 | return 0; |
4262 | } | |
4263 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4264 | ||
95cdf3b7 IM |
4265 | static int |
4266 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4267 | { |
1da177e4 LT |
4268 | struct sched_param lparam; |
4269 | struct task_struct *p; | |
36c8b586 | 4270 | int retval; |
1da177e4 LT |
4271 | |
4272 | if (!param || pid < 0) | |
4273 | return -EINVAL; | |
4274 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4275 | return -EFAULT; | |
5fe1d75f ON |
4276 | |
4277 | rcu_read_lock(); | |
4278 | retval = -ESRCH; | |
1da177e4 | 4279 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4280 | if (p != NULL) |
4281 | retval = sched_setscheduler(p, policy, &lparam); | |
4282 | rcu_read_unlock(); | |
36c8b586 | 4283 | |
1da177e4 LT |
4284 | return retval; |
4285 | } | |
4286 | ||
4287 | /** | |
4288 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4289 | * @pid: the pid in question. | |
4290 | * @policy: new policy. | |
4291 | * @param: structure containing the new RT priority. | |
4292 | */ | |
4293 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4294 | struct sched_param __user *param) | |
4295 | { | |
c21761f1 JB |
4296 | /* negative values for policy are not valid */ |
4297 | if (policy < 0) | |
4298 | return -EINVAL; | |
4299 | ||
1da177e4 LT |
4300 | return do_sched_setscheduler(pid, policy, param); |
4301 | } | |
4302 | ||
4303 | /** | |
4304 | * sys_sched_setparam - set/change the RT priority of a thread | |
4305 | * @pid: the pid in question. | |
4306 | * @param: structure containing the new RT priority. | |
4307 | */ | |
4308 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4309 | { | |
4310 | return do_sched_setscheduler(pid, -1, param); | |
4311 | } | |
4312 | ||
4313 | /** | |
4314 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4315 | * @pid: the pid in question. | |
4316 | */ | |
4317 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4318 | { | |
36c8b586 | 4319 | struct task_struct *p; |
1da177e4 | 4320 | int retval = -EINVAL; |
1da177e4 LT |
4321 | |
4322 | if (pid < 0) | |
4323 | goto out_nounlock; | |
4324 | ||
4325 | retval = -ESRCH; | |
4326 | read_lock(&tasklist_lock); | |
4327 | p = find_process_by_pid(pid); | |
4328 | if (p) { | |
4329 | retval = security_task_getscheduler(p); | |
4330 | if (!retval) | |
4331 | retval = p->policy; | |
4332 | } | |
4333 | read_unlock(&tasklist_lock); | |
4334 | ||
4335 | out_nounlock: | |
4336 | return retval; | |
4337 | } | |
4338 | ||
4339 | /** | |
4340 | * sys_sched_getscheduler - get the RT priority of a thread | |
4341 | * @pid: the pid in question. | |
4342 | * @param: structure containing the RT priority. | |
4343 | */ | |
4344 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4345 | { | |
4346 | struct sched_param lp; | |
36c8b586 | 4347 | struct task_struct *p; |
1da177e4 | 4348 | int retval = -EINVAL; |
1da177e4 LT |
4349 | |
4350 | if (!param || pid < 0) | |
4351 | goto out_nounlock; | |
4352 | ||
4353 | read_lock(&tasklist_lock); | |
4354 | p = find_process_by_pid(pid); | |
4355 | retval = -ESRCH; | |
4356 | if (!p) | |
4357 | goto out_unlock; | |
4358 | ||
4359 | retval = security_task_getscheduler(p); | |
4360 | if (retval) | |
4361 | goto out_unlock; | |
4362 | ||
4363 | lp.sched_priority = p->rt_priority; | |
4364 | read_unlock(&tasklist_lock); | |
4365 | ||
4366 | /* | |
4367 | * This one might sleep, we cannot do it with a spinlock held ... | |
4368 | */ | |
4369 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4370 | ||
4371 | out_nounlock: | |
4372 | return retval; | |
4373 | ||
4374 | out_unlock: | |
4375 | read_unlock(&tasklist_lock); | |
4376 | return retval; | |
4377 | } | |
4378 | ||
4379 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4380 | { | |
1da177e4 | 4381 | cpumask_t cpus_allowed; |
36c8b586 IM |
4382 | struct task_struct *p; |
4383 | int retval; | |
1da177e4 | 4384 | |
5be9361c | 4385 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4386 | read_lock(&tasklist_lock); |
4387 | ||
4388 | p = find_process_by_pid(pid); | |
4389 | if (!p) { | |
4390 | read_unlock(&tasklist_lock); | |
5be9361c | 4391 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4392 | return -ESRCH; |
4393 | } | |
4394 | ||
4395 | /* | |
4396 | * It is not safe to call set_cpus_allowed with the | |
4397 | * tasklist_lock held. We will bump the task_struct's | |
4398 | * usage count and then drop tasklist_lock. | |
4399 | */ | |
4400 | get_task_struct(p); | |
4401 | read_unlock(&tasklist_lock); | |
4402 | ||
4403 | retval = -EPERM; | |
4404 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4405 | !capable(CAP_SYS_NICE)) | |
4406 | goto out_unlock; | |
4407 | ||
e7834f8f DQ |
4408 | retval = security_task_setscheduler(p, 0, NULL); |
4409 | if (retval) | |
4410 | goto out_unlock; | |
4411 | ||
1da177e4 LT |
4412 | cpus_allowed = cpuset_cpus_allowed(p); |
4413 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4414 | retval = set_cpus_allowed(p, new_mask); | |
4415 | ||
4416 | out_unlock: | |
4417 | put_task_struct(p); | |
5be9361c | 4418 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4419 | return retval; |
4420 | } | |
4421 | ||
4422 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4423 | cpumask_t *new_mask) | |
4424 | { | |
4425 | if (len < sizeof(cpumask_t)) { | |
4426 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4427 | } else if (len > sizeof(cpumask_t)) { | |
4428 | len = sizeof(cpumask_t); | |
4429 | } | |
4430 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4431 | } | |
4432 | ||
4433 | /** | |
4434 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4435 | * @pid: pid of the process | |
4436 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4437 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4438 | */ | |
4439 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4440 | unsigned long __user *user_mask_ptr) | |
4441 | { | |
4442 | cpumask_t new_mask; | |
4443 | int retval; | |
4444 | ||
4445 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4446 | if (retval) | |
4447 | return retval; | |
4448 | ||
4449 | return sched_setaffinity(pid, new_mask); | |
4450 | } | |
4451 | ||
4452 | /* | |
4453 | * Represents all cpu's present in the system | |
4454 | * In systems capable of hotplug, this map could dynamically grow | |
4455 | * as new cpu's are detected in the system via any platform specific | |
4456 | * method, such as ACPI for e.g. | |
4457 | */ | |
4458 | ||
4cef0c61 | 4459 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4460 | EXPORT_SYMBOL(cpu_present_map); |
4461 | ||
4462 | #ifndef CONFIG_SMP | |
4cef0c61 | 4463 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4464 | EXPORT_SYMBOL(cpu_online_map); |
4465 | ||
4cef0c61 | 4466 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4467 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4468 | #endif |
4469 | ||
4470 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4471 | { | |
36c8b586 | 4472 | struct task_struct *p; |
1da177e4 | 4473 | int retval; |
1da177e4 | 4474 | |
5be9361c | 4475 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4476 | read_lock(&tasklist_lock); |
4477 | ||
4478 | retval = -ESRCH; | |
4479 | p = find_process_by_pid(pid); | |
4480 | if (!p) | |
4481 | goto out_unlock; | |
4482 | ||
e7834f8f DQ |
4483 | retval = security_task_getscheduler(p); |
4484 | if (retval) | |
4485 | goto out_unlock; | |
4486 | ||
2f7016d9 | 4487 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4488 | |
4489 | out_unlock: | |
4490 | read_unlock(&tasklist_lock); | |
5be9361c | 4491 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4492 | |
9531b62f | 4493 | return retval; |
1da177e4 LT |
4494 | } |
4495 | ||
4496 | /** | |
4497 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4498 | * @pid: pid of the process | |
4499 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4500 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4501 | */ | |
4502 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4503 | unsigned long __user *user_mask_ptr) | |
4504 | { | |
4505 | int ret; | |
4506 | cpumask_t mask; | |
4507 | ||
4508 | if (len < sizeof(cpumask_t)) | |
4509 | return -EINVAL; | |
4510 | ||
4511 | ret = sched_getaffinity(pid, &mask); | |
4512 | if (ret < 0) | |
4513 | return ret; | |
4514 | ||
4515 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4516 | return -EFAULT; | |
4517 | ||
4518 | return sizeof(cpumask_t); | |
4519 | } | |
4520 | ||
4521 | /** | |
4522 | * sys_sched_yield - yield the current processor to other threads. | |
4523 | * | |
dd41f596 IM |
4524 | * This function yields the current CPU to other tasks. If there are no |
4525 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4526 | */ |
4527 | asmlinkage long sys_sched_yield(void) | |
4528 | { | |
70b97a7f | 4529 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4530 | |
2d72376b | 4531 | schedstat_inc(rq, yld_count); |
4530d7ab | 4532 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4533 | |
4534 | /* | |
4535 | * Since we are going to call schedule() anyway, there's | |
4536 | * no need to preempt or enable interrupts: | |
4537 | */ | |
4538 | __release(rq->lock); | |
8a25d5de | 4539 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4540 | _raw_spin_unlock(&rq->lock); |
4541 | preempt_enable_no_resched(); | |
4542 | ||
4543 | schedule(); | |
4544 | ||
4545 | return 0; | |
4546 | } | |
4547 | ||
e7b38404 | 4548 | static void __cond_resched(void) |
1da177e4 | 4549 | { |
8e0a43d8 IM |
4550 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4551 | __might_sleep(__FILE__, __LINE__); | |
4552 | #endif | |
5bbcfd90 IM |
4553 | /* |
4554 | * The BKS might be reacquired before we have dropped | |
4555 | * PREEMPT_ACTIVE, which could trigger a second | |
4556 | * cond_resched() call. | |
4557 | */ | |
1da177e4 LT |
4558 | do { |
4559 | add_preempt_count(PREEMPT_ACTIVE); | |
4560 | schedule(); | |
4561 | sub_preempt_count(PREEMPT_ACTIVE); | |
4562 | } while (need_resched()); | |
4563 | } | |
4564 | ||
4565 | int __sched cond_resched(void) | |
4566 | { | |
9414232f IM |
4567 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4568 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4569 | __cond_resched(); |
4570 | return 1; | |
4571 | } | |
4572 | return 0; | |
4573 | } | |
1da177e4 LT |
4574 | EXPORT_SYMBOL(cond_resched); |
4575 | ||
4576 | /* | |
4577 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4578 | * call schedule, and on return reacquire the lock. | |
4579 | * | |
4580 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4581 | * operations here to prevent schedule() from being called twice (once via | |
4582 | * spin_unlock(), once by hand). | |
4583 | */ | |
95cdf3b7 | 4584 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4585 | { |
6df3cecb JK |
4586 | int ret = 0; |
4587 | ||
1da177e4 LT |
4588 | if (need_lockbreak(lock)) { |
4589 | spin_unlock(lock); | |
4590 | cpu_relax(); | |
6df3cecb | 4591 | ret = 1; |
1da177e4 LT |
4592 | spin_lock(lock); |
4593 | } | |
9414232f | 4594 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4595 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4596 | _raw_spin_unlock(lock); |
4597 | preempt_enable_no_resched(); | |
4598 | __cond_resched(); | |
6df3cecb | 4599 | ret = 1; |
1da177e4 | 4600 | spin_lock(lock); |
1da177e4 | 4601 | } |
6df3cecb | 4602 | return ret; |
1da177e4 | 4603 | } |
1da177e4 LT |
4604 | EXPORT_SYMBOL(cond_resched_lock); |
4605 | ||
4606 | int __sched cond_resched_softirq(void) | |
4607 | { | |
4608 | BUG_ON(!in_softirq()); | |
4609 | ||
9414232f | 4610 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4611 | local_bh_enable(); |
1da177e4 LT |
4612 | __cond_resched(); |
4613 | local_bh_disable(); | |
4614 | return 1; | |
4615 | } | |
4616 | return 0; | |
4617 | } | |
1da177e4 LT |
4618 | EXPORT_SYMBOL(cond_resched_softirq); |
4619 | ||
1da177e4 LT |
4620 | /** |
4621 | * yield - yield the current processor to other threads. | |
4622 | * | |
72fd4a35 | 4623 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4624 | * thread runnable and calls sys_sched_yield(). |
4625 | */ | |
4626 | void __sched yield(void) | |
4627 | { | |
4628 | set_current_state(TASK_RUNNING); | |
4629 | sys_sched_yield(); | |
4630 | } | |
1da177e4 LT |
4631 | EXPORT_SYMBOL(yield); |
4632 | ||
4633 | /* | |
4634 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4635 | * that process accounting knows that this is a task in IO wait state. | |
4636 | * | |
4637 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4638 | * has set its backing_dev_info: the queue against which it should throttle) | |
4639 | */ | |
4640 | void __sched io_schedule(void) | |
4641 | { | |
70b97a7f | 4642 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4643 | |
0ff92245 | 4644 | delayacct_blkio_start(); |
1da177e4 LT |
4645 | atomic_inc(&rq->nr_iowait); |
4646 | schedule(); | |
4647 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4648 | delayacct_blkio_end(); |
1da177e4 | 4649 | } |
1da177e4 LT |
4650 | EXPORT_SYMBOL(io_schedule); |
4651 | ||
4652 | long __sched io_schedule_timeout(long timeout) | |
4653 | { | |
70b97a7f | 4654 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4655 | long ret; |
4656 | ||
0ff92245 | 4657 | delayacct_blkio_start(); |
1da177e4 LT |
4658 | atomic_inc(&rq->nr_iowait); |
4659 | ret = schedule_timeout(timeout); | |
4660 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4661 | delayacct_blkio_end(); |
1da177e4 LT |
4662 | return ret; |
4663 | } | |
4664 | ||
4665 | /** | |
4666 | * sys_sched_get_priority_max - return maximum RT priority. | |
4667 | * @policy: scheduling class. | |
4668 | * | |
4669 | * this syscall returns the maximum rt_priority that can be used | |
4670 | * by a given scheduling class. | |
4671 | */ | |
4672 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4673 | { | |
4674 | int ret = -EINVAL; | |
4675 | ||
4676 | switch (policy) { | |
4677 | case SCHED_FIFO: | |
4678 | case SCHED_RR: | |
4679 | ret = MAX_USER_RT_PRIO-1; | |
4680 | break; | |
4681 | case SCHED_NORMAL: | |
b0a9499c | 4682 | case SCHED_BATCH: |
dd41f596 | 4683 | case SCHED_IDLE: |
1da177e4 LT |
4684 | ret = 0; |
4685 | break; | |
4686 | } | |
4687 | return ret; | |
4688 | } | |
4689 | ||
4690 | /** | |
4691 | * sys_sched_get_priority_min - return minimum RT priority. | |
4692 | * @policy: scheduling class. | |
4693 | * | |
4694 | * this syscall returns the minimum rt_priority that can be used | |
4695 | * by a given scheduling class. | |
4696 | */ | |
4697 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4698 | { | |
4699 | int ret = -EINVAL; | |
4700 | ||
4701 | switch (policy) { | |
4702 | case SCHED_FIFO: | |
4703 | case SCHED_RR: | |
4704 | ret = 1; | |
4705 | break; | |
4706 | case SCHED_NORMAL: | |
b0a9499c | 4707 | case SCHED_BATCH: |
dd41f596 | 4708 | case SCHED_IDLE: |
1da177e4 LT |
4709 | ret = 0; |
4710 | } | |
4711 | return ret; | |
4712 | } | |
4713 | ||
4714 | /** | |
4715 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4716 | * @pid: pid of the process. | |
4717 | * @interval: userspace pointer to the timeslice value. | |
4718 | * | |
4719 | * this syscall writes the default timeslice value of a given process | |
4720 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4721 | */ | |
4722 | asmlinkage | |
4723 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4724 | { | |
36c8b586 | 4725 | struct task_struct *p; |
a4ec24b4 | 4726 | unsigned int time_slice; |
1da177e4 LT |
4727 | int retval = -EINVAL; |
4728 | struct timespec t; | |
1da177e4 LT |
4729 | |
4730 | if (pid < 0) | |
4731 | goto out_nounlock; | |
4732 | ||
4733 | retval = -ESRCH; | |
4734 | read_lock(&tasklist_lock); | |
4735 | p = find_process_by_pid(pid); | |
4736 | if (!p) | |
4737 | goto out_unlock; | |
4738 | ||
4739 | retval = security_task_getscheduler(p); | |
4740 | if (retval) | |
4741 | goto out_unlock; | |
4742 | ||
a4ec24b4 DA |
4743 | if (p->policy == SCHED_FIFO) |
4744 | time_slice = 0; | |
4745 | else if (p->policy == SCHED_RR) | |
4746 | time_slice = DEF_TIMESLICE; | |
4747 | else { | |
4748 | struct sched_entity *se = &p->se; | |
4749 | unsigned long flags; | |
4750 | struct rq *rq; | |
4751 | ||
4752 | rq = task_rq_lock(p, &flags); | |
4753 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4754 | task_rq_unlock(rq, &flags); | |
4755 | } | |
1da177e4 | 4756 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4757 | jiffies_to_timespec(time_slice, &t); |
1da177e4 LT |
4758 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
4759 | out_nounlock: | |
4760 | return retval; | |
4761 | out_unlock: | |
4762 | read_unlock(&tasklist_lock); | |
4763 | return retval; | |
4764 | } | |
4765 | ||
2ed6e34f | 4766 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4767 | |
4768 | static void show_task(struct task_struct *p) | |
1da177e4 | 4769 | { |
1da177e4 | 4770 | unsigned long free = 0; |
36c8b586 | 4771 | unsigned state; |
1da177e4 | 4772 | |
1da177e4 | 4773 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4774 | printk("%-13.13s %c", p->comm, |
4775 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4776 | #if BITS_PER_LONG == 32 |
1da177e4 | 4777 | if (state == TASK_RUNNING) |
4bd77321 | 4778 | printk(" running "); |
1da177e4 | 4779 | else |
4bd77321 | 4780 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4781 | #else |
4782 | if (state == TASK_RUNNING) | |
4bd77321 | 4783 | printk(" running task "); |
1da177e4 LT |
4784 | else |
4785 | printk(" %016lx ", thread_saved_pc(p)); | |
4786 | #endif | |
4787 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4788 | { | |
10ebffde | 4789 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4790 | while (!*n) |
4791 | n++; | |
10ebffde | 4792 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4793 | } |
4794 | #endif | |
4bd77321 | 4795 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4796 | |
4797 | if (state != TASK_RUNNING) | |
4798 | show_stack(p, NULL); | |
4799 | } | |
4800 | ||
e59e2ae2 | 4801 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4802 | { |
36c8b586 | 4803 | struct task_struct *g, *p; |
1da177e4 | 4804 | |
4bd77321 IM |
4805 | #if BITS_PER_LONG == 32 |
4806 | printk(KERN_INFO | |
4807 | " task PC stack pid father\n"); | |
1da177e4 | 4808 | #else |
4bd77321 IM |
4809 | printk(KERN_INFO |
4810 | " task PC stack pid father\n"); | |
1da177e4 LT |
4811 | #endif |
4812 | read_lock(&tasklist_lock); | |
4813 | do_each_thread(g, p) { | |
4814 | /* | |
4815 | * reset the NMI-timeout, listing all files on a slow | |
4816 | * console might take alot of time: | |
4817 | */ | |
4818 | touch_nmi_watchdog(); | |
39bc89fd | 4819 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4820 | show_task(p); |
1da177e4 LT |
4821 | } while_each_thread(g, p); |
4822 | ||
04c9167f JF |
4823 | touch_all_softlockup_watchdogs(); |
4824 | ||
dd41f596 IM |
4825 | #ifdef CONFIG_SCHED_DEBUG |
4826 | sysrq_sched_debug_show(); | |
4827 | #endif | |
1da177e4 | 4828 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4829 | /* |
4830 | * Only show locks if all tasks are dumped: | |
4831 | */ | |
4832 | if (state_filter == -1) | |
4833 | debug_show_all_locks(); | |
1da177e4 LT |
4834 | } |
4835 | ||
1df21055 IM |
4836 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4837 | { | |
dd41f596 | 4838 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4839 | } |
4840 | ||
f340c0d1 IM |
4841 | /** |
4842 | * init_idle - set up an idle thread for a given CPU | |
4843 | * @idle: task in question | |
4844 | * @cpu: cpu the idle task belongs to | |
4845 | * | |
4846 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4847 | * flag, to make booting more robust. | |
4848 | */ | |
5c1e1767 | 4849 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4850 | { |
70b97a7f | 4851 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4852 | unsigned long flags; |
4853 | ||
dd41f596 IM |
4854 | __sched_fork(idle); |
4855 | idle->se.exec_start = sched_clock(); | |
4856 | ||
b29739f9 | 4857 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4858 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4859 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4860 | |
4861 | spin_lock_irqsave(&rq->lock, flags); | |
4862 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4863 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4864 | idle->oncpu = 1; | |
4865 | #endif | |
1da177e4 LT |
4866 | spin_unlock_irqrestore(&rq->lock, flags); |
4867 | ||
4868 | /* Set the preempt count _outside_ the spinlocks! */ | |
4869 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4870 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4871 | #else |
a1261f54 | 4872 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4873 | #endif |
dd41f596 IM |
4874 | /* |
4875 | * The idle tasks have their own, simple scheduling class: | |
4876 | */ | |
4877 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4878 | } |
4879 | ||
4880 | /* | |
4881 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4882 | * indicates which cpus entered this state. This is used | |
4883 | * in the rcu update to wait only for active cpus. For system | |
4884 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4885 | * always be CPU_MASK_NONE. | |
4886 | */ | |
4887 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4888 | ||
4889 | #ifdef CONFIG_SMP | |
4890 | /* | |
4891 | * This is how migration works: | |
4892 | * | |
70b97a7f | 4893 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4894 | * runqueue and wake up that CPU's migration thread. |
4895 | * 2) we down() the locked semaphore => thread blocks. | |
4896 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4897 | * thread off the CPU) | |
4898 | * 4) it gets the migration request and checks whether the migrated | |
4899 | * task is still in the wrong runqueue. | |
4900 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4901 | * it and puts it into the right queue. | |
4902 | * 6) migration thread up()s the semaphore. | |
4903 | * 7) we wake up and the migration is done. | |
4904 | */ | |
4905 | ||
4906 | /* | |
4907 | * Change a given task's CPU affinity. Migrate the thread to a | |
4908 | * proper CPU and schedule it away if the CPU it's executing on | |
4909 | * is removed from the allowed bitmask. | |
4910 | * | |
4911 | * NOTE: the caller must have a valid reference to the task, the | |
4912 | * task must not exit() & deallocate itself prematurely. The | |
4913 | * call is not atomic; no spinlocks may be held. | |
4914 | */ | |
36c8b586 | 4915 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4916 | { |
70b97a7f | 4917 | struct migration_req req; |
1da177e4 | 4918 | unsigned long flags; |
70b97a7f | 4919 | struct rq *rq; |
48f24c4d | 4920 | int ret = 0; |
1da177e4 LT |
4921 | |
4922 | rq = task_rq_lock(p, &flags); | |
4923 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4924 | ret = -EINVAL; | |
4925 | goto out; | |
4926 | } | |
4927 | ||
4928 | p->cpus_allowed = new_mask; | |
4929 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4930 | if (cpu_isset(task_cpu(p), new_mask)) | |
4931 | goto out; | |
4932 | ||
4933 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4934 | /* Need help from migration thread: drop lock and wait. */ | |
4935 | task_rq_unlock(rq, &flags); | |
4936 | wake_up_process(rq->migration_thread); | |
4937 | wait_for_completion(&req.done); | |
4938 | tlb_migrate_finish(p->mm); | |
4939 | return 0; | |
4940 | } | |
4941 | out: | |
4942 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4943 | |
1da177e4 LT |
4944 | return ret; |
4945 | } | |
1da177e4 LT |
4946 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4947 | ||
4948 | /* | |
4949 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4950 | * this because either it can't run here any more (set_cpus_allowed() | |
4951 | * away from this CPU, or CPU going down), or because we're | |
4952 | * attempting to rebalance this task on exec (sched_exec). | |
4953 | * | |
4954 | * So we race with normal scheduler movements, but that's OK, as long | |
4955 | * as the task is no longer on this CPU. | |
efc30814 KK |
4956 | * |
4957 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4958 | */ |
efc30814 | 4959 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4960 | { |
70b97a7f | 4961 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4962 | int ret = 0, on_rq; |
1da177e4 LT |
4963 | |
4964 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4965 | return ret; |
1da177e4 LT |
4966 | |
4967 | rq_src = cpu_rq(src_cpu); | |
4968 | rq_dest = cpu_rq(dest_cpu); | |
4969 | ||
4970 | double_rq_lock(rq_src, rq_dest); | |
4971 | /* Already moved. */ | |
4972 | if (task_cpu(p) != src_cpu) | |
4973 | goto out; | |
4974 | /* Affinity changed (again). */ | |
4975 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4976 | goto out; | |
4977 | ||
dd41f596 | 4978 | on_rq = p->se.on_rq; |
6e82a3be | 4979 | if (on_rq) |
2e1cb74a | 4980 | deactivate_task(rq_src, p, 0); |
6e82a3be | 4981 | |
1da177e4 | 4982 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4983 | if (on_rq) { |
4984 | activate_task(rq_dest, p, 0); | |
4985 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4986 | } |
efc30814 | 4987 | ret = 1; |
1da177e4 LT |
4988 | out: |
4989 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4990 | return ret; |
1da177e4 LT |
4991 | } |
4992 | ||
4993 | /* | |
4994 | * migration_thread - this is a highprio system thread that performs | |
4995 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4996 | * another runqueue. | |
4997 | */ | |
95cdf3b7 | 4998 | static int migration_thread(void *data) |
1da177e4 | 4999 | { |
1da177e4 | 5000 | int cpu = (long)data; |
70b97a7f | 5001 | struct rq *rq; |
1da177e4 LT |
5002 | |
5003 | rq = cpu_rq(cpu); | |
5004 | BUG_ON(rq->migration_thread != current); | |
5005 | ||
5006 | set_current_state(TASK_INTERRUPTIBLE); | |
5007 | while (!kthread_should_stop()) { | |
70b97a7f | 5008 | struct migration_req *req; |
1da177e4 | 5009 | struct list_head *head; |
1da177e4 | 5010 | |
1da177e4 LT |
5011 | spin_lock_irq(&rq->lock); |
5012 | ||
5013 | if (cpu_is_offline(cpu)) { | |
5014 | spin_unlock_irq(&rq->lock); | |
5015 | goto wait_to_die; | |
5016 | } | |
5017 | ||
5018 | if (rq->active_balance) { | |
5019 | active_load_balance(rq, cpu); | |
5020 | rq->active_balance = 0; | |
5021 | } | |
5022 | ||
5023 | head = &rq->migration_queue; | |
5024 | ||
5025 | if (list_empty(head)) { | |
5026 | spin_unlock_irq(&rq->lock); | |
5027 | schedule(); | |
5028 | set_current_state(TASK_INTERRUPTIBLE); | |
5029 | continue; | |
5030 | } | |
70b97a7f | 5031 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5032 | list_del_init(head->next); |
5033 | ||
674311d5 NP |
5034 | spin_unlock(&rq->lock); |
5035 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5036 | local_irq_enable(); | |
1da177e4 LT |
5037 | |
5038 | complete(&req->done); | |
5039 | } | |
5040 | __set_current_state(TASK_RUNNING); | |
5041 | return 0; | |
5042 | ||
5043 | wait_to_die: | |
5044 | /* Wait for kthread_stop */ | |
5045 | set_current_state(TASK_INTERRUPTIBLE); | |
5046 | while (!kthread_should_stop()) { | |
5047 | schedule(); | |
5048 | set_current_state(TASK_INTERRUPTIBLE); | |
5049 | } | |
5050 | __set_current_state(TASK_RUNNING); | |
5051 | return 0; | |
5052 | } | |
5053 | ||
5054 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
5055 | /* |
5056 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5057 | * NOTE: interrupts should be disabled by the caller | |
5058 | */ | |
48f24c4d | 5059 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5060 | { |
efc30814 | 5061 | unsigned long flags; |
1da177e4 | 5062 | cpumask_t mask; |
70b97a7f IM |
5063 | struct rq *rq; |
5064 | int dest_cpu; | |
1da177e4 | 5065 | |
efc30814 | 5066 | restart: |
1da177e4 LT |
5067 | /* On same node? */ |
5068 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
48f24c4d | 5069 | cpus_and(mask, mask, p->cpus_allowed); |
1da177e4 LT |
5070 | dest_cpu = any_online_cpu(mask); |
5071 | ||
5072 | /* On any allowed CPU? */ | |
5073 | if (dest_cpu == NR_CPUS) | |
48f24c4d | 5074 | dest_cpu = any_online_cpu(p->cpus_allowed); |
1da177e4 LT |
5075 | |
5076 | /* No more Mr. Nice Guy. */ | |
5077 | if (dest_cpu == NR_CPUS) { | |
48f24c4d IM |
5078 | rq = task_rq_lock(p, &flags); |
5079 | cpus_setall(p->cpus_allowed); | |
5080 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
efc30814 | 5081 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5082 | |
5083 | /* | |
5084 | * Don't tell them about moving exiting tasks or | |
5085 | * kernel threads (both mm NULL), since they never | |
5086 | * leave kernel. | |
5087 | */ | |
48f24c4d | 5088 | if (p->mm && printk_ratelimit()) |
1da177e4 LT |
5089 | printk(KERN_INFO "process %d (%s) no " |
5090 | "longer affine to cpu%d\n", | |
48f24c4d | 5091 | p->pid, p->comm, dead_cpu); |
1da177e4 | 5092 | } |
48f24c4d | 5093 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
efc30814 | 5094 | goto restart; |
1da177e4 LT |
5095 | } |
5096 | ||
5097 | /* | |
5098 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5099 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5100 | * for performance reasons the counter is not stricly tracking tasks to | |
5101 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5102 | * to keep the global sum constant after CPU-down: | |
5103 | */ | |
70b97a7f | 5104 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5105 | { |
70b97a7f | 5106 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5107 | unsigned long flags; |
5108 | ||
5109 | local_irq_save(flags); | |
5110 | double_rq_lock(rq_src, rq_dest); | |
5111 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5112 | rq_src->nr_uninterruptible = 0; | |
5113 | double_rq_unlock(rq_src, rq_dest); | |
5114 | local_irq_restore(flags); | |
5115 | } | |
5116 | ||
5117 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5118 | static void migrate_live_tasks(int src_cpu) | |
5119 | { | |
48f24c4d | 5120 | struct task_struct *p, *t; |
1da177e4 LT |
5121 | |
5122 | write_lock_irq(&tasklist_lock); | |
5123 | ||
48f24c4d IM |
5124 | do_each_thread(t, p) { |
5125 | if (p == current) | |
1da177e4 LT |
5126 | continue; |
5127 | ||
48f24c4d IM |
5128 | if (task_cpu(p) == src_cpu) |
5129 | move_task_off_dead_cpu(src_cpu, p); | |
5130 | } while_each_thread(t, p); | |
1da177e4 LT |
5131 | |
5132 | write_unlock_irq(&tasklist_lock); | |
5133 | } | |
5134 | ||
a9957449 AD |
5135 | /* |
5136 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5137 | */ | |
5138 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5139 | { | |
5140 | update_rq_clock(rq); | |
5141 | ||
5142 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5143 | rq->nr_uninterruptible--; | |
5144 | ||
5145 | enqueue_task(rq, p, 0); | |
5146 | inc_nr_running(p, rq); | |
5147 | } | |
5148 | ||
dd41f596 IM |
5149 | /* |
5150 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5151 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5152 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5153 | */ |
5154 | void sched_idle_next(void) | |
5155 | { | |
48f24c4d | 5156 | int this_cpu = smp_processor_id(); |
70b97a7f | 5157 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5158 | struct task_struct *p = rq->idle; |
5159 | unsigned long flags; | |
5160 | ||
5161 | /* cpu has to be offline */ | |
48f24c4d | 5162 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5163 | |
48f24c4d IM |
5164 | /* |
5165 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5166 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5167 | */ |
5168 | spin_lock_irqsave(&rq->lock, flags); | |
5169 | ||
dd41f596 | 5170 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5171 | |
5172 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5173 | activate_idle_task(p, rq); |
1da177e4 LT |
5174 | |
5175 | spin_unlock_irqrestore(&rq->lock, flags); | |
5176 | } | |
5177 | ||
48f24c4d IM |
5178 | /* |
5179 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5180 | * offline. |
5181 | */ | |
5182 | void idle_task_exit(void) | |
5183 | { | |
5184 | struct mm_struct *mm = current->active_mm; | |
5185 | ||
5186 | BUG_ON(cpu_online(smp_processor_id())); | |
5187 | ||
5188 | if (mm != &init_mm) | |
5189 | switch_mm(mm, &init_mm, current); | |
5190 | mmdrop(mm); | |
5191 | } | |
5192 | ||
054b9108 | 5193 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5194 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5195 | { |
70b97a7f | 5196 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5197 | |
5198 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5199 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5200 | |
5201 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5202 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5203 | |
48f24c4d | 5204 | get_task_struct(p); |
1da177e4 LT |
5205 | |
5206 | /* | |
5207 | * Drop lock around migration; if someone else moves it, | |
5208 | * that's OK. No task can be added to this CPU, so iteration is | |
5209 | * fine. | |
054b9108 | 5210 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5211 | */ |
054b9108 | 5212 | spin_unlock(&rq->lock); |
48f24c4d | 5213 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5214 | spin_lock(&rq->lock); |
1da177e4 | 5215 | |
48f24c4d | 5216 | put_task_struct(p); |
1da177e4 LT |
5217 | } |
5218 | ||
5219 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5220 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5221 | { | |
70b97a7f | 5222 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5223 | struct task_struct *next; |
48f24c4d | 5224 | |
dd41f596 IM |
5225 | for ( ; ; ) { |
5226 | if (!rq->nr_running) | |
5227 | break; | |
a8e504d2 | 5228 | update_rq_clock(rq); |
ff95f3df | 5229 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5230 | if (!next) |
5231 | break; | |
5232 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5233 | |
1da177e4 LT |
5234 | } |
5235 | } | |
5236 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5237 | ||
e692ab53 NP |
5238 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5239 | ||
5240 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5241 | { |
5242 | .procname = "sched_domain", | |
c57baf1e | 5243 | .mode = 0555, |
e0361851 | 5244 | }, |
e692ab53 NP |
5245 | {0,}, |
5246 | }; | |
5247 | ||
5248 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5249 | { |
c57baf1e | 5250 | .ctl_name = CTL_KERN, |
e0361851 | 5251 | .procname = "kernel", |
c57baf1e | 5252 | .mode = 0555, |
e0361851 AD |
5253 | .child = sd_ctl_dir, |
5254 | }, | |
e692ab53 NP |
5255 | {0,}, |
5256 | }; | |
5257 | ||
5258 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5259 | { | |
5260 | struct ctl_table *entry = | |
5261 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); | |
5262 | ||
5263 | BUG_ON(!entry); | |
5264 | memset(entry, 0, n * sizeof(struct ctl_table)); | |
5265 | ||
5266 | return entry; | |
5267 | } | |
5268 | ||
5269 | static void | |
e0361851 | 5270 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5271 | const char *procname, void *data, int maxlen, |
5272 | mode_t mode, proc_handler *proc_handler) | |
5273 | { | |
e692ab53 NP |
5274 | entry->procname = procname; |
5275 | entry->data = data; | |
5276 | entry->maxlen = maxlen; | |
5277 | entry->mode = mode; | |
5278 | entry->proc_handler = proc_handler; | |
5279 | } | |
5280 | ||
5281 | static struct ctl_table * | |
5282 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5283 | { | |
5284 | struct ctl_table *table = sd_alloc_ctl_entry(14); | |
5285 | ||
e0361851 | 5286 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5287 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5288 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5289 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5290 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5291 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5292 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5293 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5294 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5295 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5296 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5297 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5298 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5299 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5300 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5301 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5302 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5303 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5304 | set_table_entry(&table[10], "cache_nice_tries", |
e692ab53 NP |
5305 | &sd->cache_nice_tries, |
5306 | sizeof(int), 0644, proc_dointvec_minmax); | |
e0361851 | 5307 | set_table_entry(&table[12], "flags", &sd->flags, |
e692ab53 NP |
5308 | sizeof(int), 0644, proc_dointvec_minmax); |
5309 | ||
5310 | return table; | |
5311 | } | |
5312 | ||
5313 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | |
5314 | { | |
5315 | struct ctl_table *entry, *table; | |
5316 | struct sched_domain *sd; | |
5317 | int domain_num = 0, i; | |
5318 | char buf[32]; | |
5319 | ||
5320 | for_each_domain(cpu, sd) | |
5321 | domain_num++; | |
5322 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
5323 | ||
5324 | i = 0; | |
5325 | for_each_domain(cpu, sd) { | |
5326 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5327 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5328 | entry->mode = 0555; |
e692ab53 NP |
5329 | entry->child = sd_alloc_ctl_domain_table(sd); |
5330 | entry++; | |
5331 | i++; | |
5332 | } | |
5333 | return table; | |
5334 | } | |
5335 | ||
5336 | static struct ctl_table_header *sd_sysctl_header; | |
5337 | static void init_sched_domain_sysctl(void) | |
5338 | { | |
5339 | int i, cpu_num = num_online_cpus(); | |
5340 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5341 | char buf[32]; | |
5342 | ||
5343 | sd_ctl_dir[0].child = entry; | |
5344 | ||
5345 | for (i = 0; i < cpu_num; i++, entry++) { | |
5346 | snprintf(buf, 32, "cpu%d", i); | |
e692ab53 | 5347 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5348 | entry->mode = 0555; |
e692ab53 NP |
5349 | entry->child = sd_alloc_ctl_cpu_table(i); |
5350 | } | |
5351 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5352 | } | |
5353 | #else | |
5354 | static void init_sched_domain_sysctl(void) | |
5355 | { | |
5356 | } | |
5357 | #endif | |
5358 | ||
1da177e4 LT |
5359 | /* |
5360 | * migration_call - callback that gets triggered when a CPU is added. | |
5361 | * Here we can start up the necessary migration thread for the new CPU. | |
5362 | */ | |
48f24c4d IM |
5363 | static int __cpuinit |
5364 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5365 | { |
1da177e4 | 5366 | struct task_struct *p; |
48f24c4d | 5367 | int cpu = (long)hcpu; |
1da177e4 | 5368 | unsigned long flags; |
70b97a7f | 5369 | struct rq *rq; |
1da177e4 LT |
5370 | |
5371 | switch (action) { | |
5be9361c GS |
5372 | case CPU_LOCK_ACQUIRE: |
5373 | mutex_lock(&sched_hotcpu_mutex); | |
5374 | break; | |
5375 | ||
1da177e4 | 5376 | case CPU_UP_PREPARE: |
8bb78442 | 5377 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5378 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5379 | if (IS_ERR(p)) |
5380 | return NOTIFY_BAD; | |
1da177e4 LT |
5381 | kthread_bind(p, cpu); |
5382 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5383 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5384 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5385 | task_rq_unlock(rq, &flags); |
5386 | cpu_rq(cpu)->migration_thread = p; | |
5387 | break; | |
48f24c4d | 5388 | |
1da177e4 | 5389 | case CPU_ONLINE: |
8bb78442 | 5390 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5391 | /* Strictly unneccessary, as first user will wake it. */ |
5392 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5393 | break; | |
48f24c4d | 5394 | |
1da177e4 LT |
5395 | #ifdef CONFIG_HOTPLUG_CPU |
5396 | case CPU_UP_CANCELED: | |
8bb78442 | 5397 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5398 | if (!cpu_rq(cpu)->migration_thread) |
5399 | break; | |
1da177e4 | 5400 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5401 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5402 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5403 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5404 | cpu_rq(cpu)->migration_thread = NULL; | |
5405 | break; | |
48f24c4d | 5406 | |
1da177e4 | 5407 | case CPU_DEAD: |
8bb78442 | 5408 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5409 | migrate_live_tasks(cpu); |
5410 | rq = cpu_rq(cpu); | |
5411 | kthread_stop(rq->migration_thread); | |
5412 | rq->migration_thread = NULL; | |
5413 | /* Idle task back to normal (off runqueue, low prio) */ | |
5414 | rq = task_rq_lock(rq->idle, &flags); | |
a8e504d2 | 5415 | update_rq_clock(rq); |
2e1cb74a | 5416 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5417 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5418 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5419 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5420 | migrate_dead_tasks(cpu); |
5421 | task_rq_unlock(rq, &flags); | |
5422 | migrate_nr_uninterruptible(rq); | |
5423 | BUG_ON(rq->nr_running != 0); | |
5424 | ||
5425 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5426 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5427 | * the requestors. */ |
5428 | spin_lock_irq(&rq->lock); | |
5429 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5430 | struct migration_req *req; |
5431 | ||
1da177e4 | 5432 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5433 | struct migration_req, list); |
1da177e4 LT |
5434 | list_del_init(&req->list); |
5435 | complete(&req->done); | |
5436 | } | |
5437 | spin_unlock_irq(&rq->lock); | |
5438 | break; | |
5439 | #endif | |
5be9361c GS |
5440 | case CPU_LOCK_RELEASE: |
5441 | mutex_unlock(&sched_hotcpu_mutex); | |
5442 | break; | |
1da177e4 LT |
5443 | } |
5444 | return NOTIFY_OK; | |
5445 | } | |
5446 | ||
5447 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5448 | * happens before everything else. | |
5449 | */ | |
26c2143b | 5450 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5451 | .notifier_call = migration_call, |
5452 | .priority = 10 | |
5453 | }; | |
5454 | ||
5455 | int __init migration_init(void) | |
5456 | { | |
5457 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5458 | int err; |
48f24c4d IM |
5459 | |
5460 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5461 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5462 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5463 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5464 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5465 | |
1da177e4 LT |
5466 | return 0; |
5467 | } | |
5468 | #endif | |
5469 | ||
5470 | #ifdef CONFIG_SMP | |
476f3534 CL |
5471 | |
5472 | /* Number of possible processor ids */ | |
5473 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5474 | EXPORT_SYMBOL(nr_cpu_ids); | |
5475 | ||
3e9830dc | 5476 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5477 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5478 | { | |
5479 | int level = 0; | |
5480 | ||
41c7ce9a NP |
5481 | if (!sd) { |
5482 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5483 | return; | |
5484 | } | |
5485 | ||
1da177e4 LT |
5486 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5487 | ||
5488 | do { | |
5489 | int i; | |
5490 | char str[NR_CPUS]; | |
5491 | struct sched_group *group = sd->groups; | |
5492 | cpumask_t groupmask; | |
5493 | ||
5494 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5495 | cpus_clear(groupmask); | |
5496 | ||
5497 | printk(KERN_DEBUG); | |
5498 | for (i = 0; i < level + 1; i++) | |
5499 | printk(" "); | |
5500 | printk("domain %d: ", level); | |
5501 | ||
5502 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5503 | printk("does not load-balance\n"); | |
5504 | if (sd->parent) | |
33859f7f MOS |
5505 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5506 | " has parent"); | |
1da177e4 LT |
5507 | break; |
5508 | } | |
5509 | ||
5510 | printk("span %s\n", str); | |
5511 | ||
5512 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5513 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5514 | "CPU%d\n", cpu); | |
1da177e4 | 5515 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5516 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5517 | " CPU%d\n", cpu); | |
1da177e4 LT |
5518 | |
5519 | printk(KERN_DEBUG); | |
5520 | for (i = 0; i < level + 2; i++) | |
5521 | printk(" "); | |
5522 | printk("groups:"); | |
5523 | do { | |
5524 | if (!group) { | |
5525 | printk("\n"); | |
5526 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5527 | break; | |
5528 | } | |
5529 | ||
5517d86b | 5530 | if (!group->__cpu_power) { |
1da177e4 | 5531 | printk("\n"); |
33859f7f MOS |
5532 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5533 | "set\n"); | |
26797a34 | 5534 | break; |
1da177e4 LT |
5535 | } |
5536 | ||
5537 | if (!cpus_weight(group->cpumask)) { | |
5538 | printk("\n"); | |
5539 | printk(KERN_ERR "ERROR: empty group\n"); | |
26797a34 | 5540 | break; |
1da177e4 LT |
5541 | } |
5542 | ||
5543 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5544 | printk("\n"); | |
5545 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
26797a34 | 5546 | break; |
1da177e4 LT |
5547 | } |
5548 | ||
5549 | cpus_or(groupmask, groupmask, group->cpumask); | |
5550 | ||
5551 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5552 | printk(" %s", str); | |
5553 | ||
5554 | group = group->next; | |
5555 | } while (group != sd->groups); | |
5556 | printk("\n"); | |
5557 | ||
5558 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5559 | printk(KERN_ERR "ERROR: groups don't span " |
5560 | "domain->span\n"); | |
1da177e4 LT |
5561 | |
5562 | level++; | |
5563 | sd = sd->parent; | |
33859f7f MOS |
5564 | if (!sd) |
5565 | continue; | |
1da177e4 | 5566 | |
33859f7f MOS |
5567 | if (!cpus_subset(groupmask, sd->span)) |
5568 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5569 | "of domain->span\n"); | |
1da177e4 LT |
5570 | |
5571 | } while (sd); | |
5572 | } | |
5573 | #else | |
48f24c4d | 5574 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5575 | #endif |
5576 | ||
1a20ff27 | 5577 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5578 | { |
5579 | if (cpus_weight(sd->span) == 1) | |
5580 | return 1; | |
5581 | ||
5582 | /* Following flags need at least 2 groups */ | |
5583 | if (sd->flags & (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 | if (sd->groups != sd->groups->next) |
5590 | return 0; | |
5591 | } | |
5592 | ||
5593 | /* Following flags don't use groups */ | |
5594 | if (sd->flags & (SD_WAKE_IDLE | | |
5595 | SD_WAKE_AFFINE | | |
5596 | SD_WAKE_BALANCE)) | |
5597 | return 0; | |
5598 | ||
5599 | return 1; | |
5600 | } | |
5601 | ||
48f24c4d IM |
5602 | static int |
5603 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5604 | { |
5605 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5606 | ||
5607 | if (sd_degenerate(parent)) | |
5608 | return 1; | |
5609 | ||
5610 | if (!cpus_equal(sd->span, parent->span)) | |
5611 | return 0; | |
5612 | ||
5613 | /* Does parent contain flags not in child? */ | |
5614 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5615 | if (cflags & SD_WAKE_AFFINE) | |
5616 | pflags &= ~SD_WAKE_BALANCE; | |
5617 | /* Flags needing groups don't count if only 1 group in parent */ | |
5618 | if (parent->groups == parent->groups->next) { | |
5619 | pflags &= ~(SD_LOAD_BALANCE | | |
5620 | SD_BALANCE_NEWIDLE | | |
5621 | SD_BALANCE_FORK | | |
89c4710e SS |
5622 | SD_BALANCE_EXEC | |
5623 | SD_SHARE_CPUPOWER | | |
5624 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5625 | } |
5626 | if (~cflags & pflags) | |
5627 | return 0; | |
5628 | ||
5629 | return 1; | |
5630 | } | |
5631 | ||
1da177e4 LT |
5632 | /* |
5633 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5634 | * hold the hotplug lock. | |
5635 | */ | |
9c1cfda2 | 5636 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5637 | { |
70b97a7f | 5638 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5639 | struct sched_domain *tmp; |
5640 | ||
5641 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5642 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5643 | struct sched_domain *parent = tmp->parent; | |
5644 | if (!parent) | |
5645 | break; | |
1a848870 | 5646 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5647 | tmp->parent = parent->parent; |
1a848870 SS |
5648 | if (parent->parent) |
5649 | parent->parent->child = tmp; | |
5650 | } | |
245af2c7 SS |
5651 | } |
5652 | ||
1a848870 | 5653 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5654 | sd = sd->parent; |
1a848870 SS |
5655 | if (sd) |
5656 | sd->child = NULL; | |
5657 | } | |
1da177e4 LT |
5658 | |
5659 | sched_domain_debug(sd, cpu); | |
5660 | ||
674311d5 | 5661 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5662 | } |
5663 | ||
5664 | /* cpus with isolated domains */ | |
67af63a6 | 5665 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5666 | |
5667 | /* Setup the mask of cpus configured for isolated domains */ | |
5668 | static int __init isolated_cpu_setup(char *str) | |
5669 | { | |
5670 | int ints[NR_CPUS], i; | |
5671 | ||
5672 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5673 | cpus_clear(cpu_isolated_map); | |
5674 | for (i = 1; i <= ints[0]; i++) | |
5675 | if (ints[i] < NR_CPUS) | |
5676 | cpu_set(ints[i], cpu_isolated_map); | |
5677 | return 1; | |
5678 | } | |
5679 | ||
8927f494 | 5680 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5681 | |
5682 | /* | |
6711cab4 SS |
5683 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5684 | * to a function which identifies what group(along with sched group) a CPU | |
5685 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5686 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5687 | * |
5688 | * init_sched_build_groups will build a circular linked list of the groups | |
5689 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5690 | * and ->cpu_power to 0. | |
5691 | */ | |
a616058b | 5692 | static void |
6711cab4 SS |
5693 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5694 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5695 | struct sched_group **sg)) | |
1da177e4 LT |
5696 | { |
5697 | struct sched_group *first = NULL, *last = NULL; | |
5698 | cpumask_t covered = CPU_MASK_NONE; | |
5699 | int i; | |
5700 | ||
5701 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5702 | struct sched_group *sg; |
5703 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5704 | int j; |
5705 | ||
5706 | if (cpu_isset(i, covered)) | |
5707 | continue; | |
5708 | ||
5709 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5710 | sg->__cpu_power = 0; |
1da177e4 LT |
5711 | |
5712 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5713 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5714 | continue; |
5715 | ||
5716 | cpu_set(j, covered); | |
5717 | cpu_set(j, sg->cpumask); | |
5718 | } | |
5719 | if (!first) | |
5720 | first = sg; | |
5721 | if (last) | |
5722 | last->next = sg; | |
5723 | last = sg; | |
5724 | } | |
5725 | last->next = first; | |
5726 | } | |
5727 | ||
9c1cfda2 | 5728 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5729 | |
9c1cfda2 | 5730 | #ifdef CONFIG_NUMA |
198e2f18 | 5731 | |
9c1cfda2 JH |
5732 | /** |
5733 | * find_next_best_node - find the next node to include in a sched_domain | |
5734 | * @node: node whose sched_domain we're building | |
5735 | * @used_nodes: nodes already in the sched_domain | |
5736 | * | |
5737 | * Find the next node to include in a given scheduling domain. Simply | |
5738 | * finds the closest node not already in the @used_nodes map. | |
5739 | * | |
5740 | * Should use nodemask_t. | |
5741 | */ | |
5742 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5743 | { | |
5744 | int i, n, val, min_val, best_node = 0; | |
5745 | ||
5746 | min_val = INT_MAX; | |
5747 | ||
5748 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5749 | /* Start at @node */ | |
5750 | n = (node + i) % MAX_NUMNODES; | |
5751 | ||
5752 | if (!nr_cpus_node(n)) | |
5753 | continue; | |
5754 | ||
5755 | /* Skip already used nodes */ | |
5756 | if (test_bit(n, used_nodes)) | |
5757 | continue; | |
5758 | ||
5759 | /* Simple min distance search */ | |
5760 | val = node_distance(node, n); | |
5761 | ||
5762 | if (val < min_val) { | |
5763 | min_val = val; | |
5764 | best_node = n; | |
5765 | } | |
5766 | } | |
5767 | ||
5768 | set_bit(best_node, used_nodes); | |
5769 | return best_node; | |
5770 | } | |
5771 | ||
5772 | /** | |
5773 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5774 | * @node: node whose cpumask we're constructing | |
5775 | * @size: number of nodes to include in this span | |
5776 | * | |
5777 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5778 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5779 | * out optimally. | |
5780 | */ | |
5781 | static cpumask_t sched_domain_node_span(int node) | |
5782 | { | |
9c1cfda2 | 5783 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5784 | cpumask_t span, nodemask; |
5785 | int i; | |
9c1cfda2 JH |
5786 | |
5787 | cpus_clear(span); | |
5788 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5789 | ||
5790 | nodemask = node_to_cpumask(node); | |
5791 | cpus_or(span, span, nodemask); | |
5792 | set_bit(node, used_nodes); | |
5793 | ||
5794 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5795 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5796 | |
9c1cfda2 JH |
5797 | nodemask = node_to_cpumask(next_node); |
5798 | cpus_or(span, span, nodemask); | |
5799 | } | |
5800 | ||
5801 | return span; | |
5802 | } | |
5803 | #endif | |
5804 | ||
5c45bf27 | 5805 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5806 | |
9c1cfda2 | 5807 | /* |
48f24c4d | 5808 | * SMT sched-domains: |
9c1cfda2 | 5809 | */ |
1da177e4 LT |
5810 | #ifdef CONFIG_SCHED_SMT |
5811 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5812 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5813 | |
6711cab4 SS |
5814 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5815 | struct sched_group **sg) | |
1da177e4 | 5816 | { |
6711cab4 SS |
5817 | if (sg) |
5818 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5819 | return cpu; |
5820 | } | |
5821 | #endif | |
5822 | ||
48f24c4d IM |
5823 | /* |
5824 | * multi-core sched-domains: | |
5825 | */ | |
1e9f28fa SS |
5826 | #ifdef CONFIG_SCHED_MC |
5827 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5828 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5829 | #endif |
5830 | ||
5831 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5832 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5833 | struct sched_group **sg) | |
1e9f28fa | 5834 | { |
6711cab4 | 5835 | int group; |
a616058b SS |
5836 | cpumask_t mask = cpu_sibling_map[cpu]; |
5837 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5838 | group = first_cpu(mask); |
5839 | if (sg) | |
5840 | *sg = &per_cpu(sched_group_core, group); | |
5841 | return group; | |
1e9f28fa SS |
5842 | } |
5843 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5844 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5845 | struct sched_group **sg) | |
1e9f28fa | 5846 | { |
6711cab4 SS |
5847 | if (sg) |
5848 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5849 | return cpu; |
5850 | } | |
5851 | #endif | |
5852 | ||
1da177e4 | 5853 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5854 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5855 | |
6711cab4 SS |
5856 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5857 | struct sched_group **sg) | |
1da177e4 | 5858 | { |
6711cab4 | 5859 | int group; |
48f24c4d | 5860 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5861 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5862 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5863 | group = first_cpu(mask); |
1e9f28fa | 5864 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5865 | cpumask_t mask = cpu_sibling_map[cpu]; |
5866 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5867 | group = first_cpu(mask); |
1da177e4 | 5868 | #else |
6711cab4 | 5869 | group = cpu; |
1da177e4 | 5870 | #endif |
6711cab4 SS |
5871 | if (sg) |
5872 | *sg = &per_cpu(sched_group_phys, group); | |
5873 | return group; | |
1da177e4 LT |
5874 | } |
5875 | ||
5876 | #ifdef CONFIG_NUMA | |
1da177e4 | 5877 | /* |
9c1cfda2 JH |
5878 | * The init_sched_build_groups can't handle what we want to do with node |
5879 | * groups, so roll our own. Now each node has its own list of groups which | |
5880 | * gets dynamically allocated. | |
1da177e4 | 5881 | */ |
9c1cfda2 | 5882 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5883 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5884 | |
9c1cfda2 | 5885 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5886 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5887 | |
6711cab4 SS |
5888 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5889 | struct sched_group **sg) | |
9c1cfda2 | 5890 | { |
6711cab4 SS |
5891 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5892 | int group; | |
5893 | ||
5894 | cpus_and(nodemask, nodemask, *cpu_map); | |
5895 | group = first_cpu(nodemask); | |
5896 | ||
5897 | if (sg) | |
5898 | *sg = &per_cpu(sched_group_allnodes, group); | |
5899 | return group; | |
1da177e4 | 5900 | } |
6711cab4 | 5901 | |
08069033 SS |
5902 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5903 | { | |
5904 | struct sched_group *sg = group_head; | |
5905 | int j; | |
5906 | ||
5907 | if (!sg) | |
5908 | return; | |
5909 | next_sg: | |
5910 | for_each_cpu_mask(j, sg->cpumask) { | |
5911 | struct sched_domain *sd; | |
5912 | ||
5913 | sd = &per_cpu(phys_domains, j); | |
5914 | if (j != first_cpu(sd->groups->cpumask)) { | |
5915 | /* | |
5916 | * Only add "power" once for each | |
5917 | * physical package. | |
5918 | */ | |
5919 | continue; | |
5920 | } | |
5921 | ||
5517d86b | 5922 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
08069033 SS |
5923 | } |
5924 | sg = sg->next; | |
5925 | if (sg != group_head) | |
5926 | goto next_sg; | |
5927 | } | |
1da177e4 LT |
5928 | #endif |
5929 | ||
a616058b | 5930 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5931 | /* Free memory allocated for various sched_group structures */ |
5932 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5933 | { | |
a616058b | 5934 | int cpu, i; |
51888ca2 SV |
5935 | |
5936 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5937 | struct sched_group **sched_group_nodes |
5938 | = sched_group_nodes_bycpu[cpu]; | |
5939 | ||
51888ca2 SV |
5940 | if (!sched_group_nodes) |
5941 | continue; | |
5942 | ||
5943 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5944 | cpumask_t nodemask = node_to_cpumask(i); | |
5945 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5946 | ||
5947 | cpus_and(nodemask, nodemask, *cpu_map); | |
5948 | if (cpus_empty(nodemask)) | |
5949 | continue; | |
5950 | ||
5951 | if (sg == NULL) | |
5952 | continue; | |
5953 | sg = sg->next; | |
5954 | next_sg: | |
5955 | oldsg = sg; | |
5956 | sg = sg->next; | |
5957 | kfree(oldsg); | |
5958 | if (oldsg != sched_group_nodes[i]) | |
5959 | goto next_sg; | |
5960 | } | |
5961 | kfree(sched_group_nodes); | |
5962 | sched_group_nodes_bycpu[cpu] = NULL; | |
5963 | } | |
51888ca2 | 5964 | } |
a616058b SS |
5965 | #else |
5966 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5967 | { | |
5968 | } | |
5969 | #endif | |
51888ca2 | 5970 | |
89c4710e SS |
5971 | /* |
5972 | * Initialize sched groups cpu_power. | |
5973 | * | |
5974 | * cpu_power indicates the capacity of sched group, which is used while | |
5975 | * distributing the load between different sched groups in a sched domain. | |
5976 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5977 | * there are asymmetries in the topology. If there are asymmetries, group | |
5978 | * having more cpu_power will pickup more load compared to the group having | |
5979 | * less cpu_power. | |
5980 | * | |
5981 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5982 | * the maximum number of tasks a group can handle in the presence of other idle | |
5983 | * or lightly loaded groups in the same sched domain. | |
5984 | */ | |
5985 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5986 | { | |
5987 | struct sched_domain *child; | |
5988 | struct sched_group *group; | |
5989 | ||
5990 | WARN_ON(!sd || !sd->groups); | |
5991 | ||
5992 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5993 | return; | |
5994 | ||
5995 | child = sd->child; | |
5996 | ||
5517d86b ED |
5997 | sd->groups->__cpu_power = 0; |
5998 | ||
89c4710e SS |
5999 | /* |
6000 | * For perf policy, if the groups in child domain share resources | |
6001 | * (for example cores sharing some portions of the cache hierarchy | |
6002 | * or SMT), then set this domain groups cpu_power such that each group | |
6003 | * can handle only one task, when there are other idle groups in the | |
6004 | * same sched domain. | |
6005 | */ | |
6006 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6007 | (child->flags & | |
6008 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6009 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6010 | return; |
6011 | } | |
6012 | ||
89c4710e SS |
6013 | /* |
6014 | * add cpu_power of each child group to this groups cpu_power | |
6015 | */ | |
6016 | group = child->groups; | |
6017 | do { | |
5517d86b | 6018 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6019 | group = group->next; |
6020 | } while (group != child->groups); | |
6021 | } | |
6022 | ||
1da177e4 | 6023 | /* |
1a20ff27 DG |
6024 | * Build sched domains for a given set of cpus and attach the sched domains |
6025 | * to the individual cpus | |
1da177e4 | 6026 | */ |
51888ca2 | 6027 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6028 | { |
6029 | int i; | |
d1b55138 JH |
6030 | #ifdef CONFIG_NUMA |
6031 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6032 | int sd_allnodes = 0; |
d1b55138 JH |
6033 | |
6034 | /* | |
6035 | * Allocate the per-node list of sched groups | |
6036 | */ | |
dd41f596 | 6037 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 6038 | GFP_KERNEL); |
d1b55138 JH |
6039 | if (!sched_group_nodes) { |
6040 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6041 | return -ENOMEM; |
d1b55138 JH |
6042 | } |
6043 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6044 | #endif | |
1da177e4 LT |
6045 | |
6046 | /* | |
1a20ff27 | 6047 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6048 | */ |
1a20ff27 | 6049 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6050 | struct sched_domain *sd = NULL, *p; |
6051 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6052 | ||
1a20ff27 | 6053 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6054 | |
6055 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6056 | if (cpus_weight(*cpu_map) > |
6057 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6058 | sd = &per_cpu(allnodes_domains, i); |
6059 | *sd = SD_ALLNODES_INIT; | |
6060 | sd->span = *cpu_map; | |
6711cab4 | 6061 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6062 | p = sd; |
6711cab4 | 6063 | sd_allnodes = 1; |
9c1cfda2 JH |
6064 | } else |
6065 | p = NULL; | |
6066 | ||
1da177e4 | 6067 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6068 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6069 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6070 | sd->parent = p; | |
1a848870 SS |
6071 | if (p) |
6072 | p->child = sd; | |
9c1cfda2 | 6073 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6074 | #endif |
6075 | ||
6076 | p = sd; | |
6077 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6078 | *sd = SD_CPU_INIT; |
6079 | sd->span = nodemask; | |
6080 | sd->parent = p; | |
1a848870 SS |
6081 | if (p) |
6082 | p->child = sd; | |
6711cab4 | 6083 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6084 | |
1e9f28fa SS |
6085 | #ifdef CONFIG_SCHED_MC |
6086 | p = sd; | |
6087 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6088 | *sd = SD_MC_INIT; |
6089 | sd->span = cpu_coregroup_map(i); | |
6090 | cpus_and(sd->span, sd->span, *cpu_map); | |
6091 | sd->parent = p; | |
1a848870 | 6092 | p->child = sd; |
6711cab4 | 6093 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6094 | #endif |
6095 | ||
1da177e4 LT |
6096 | #ifdef CONFIG_SCHED_SMT |
6097 | p = sd; | |
6098 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6099 | *sd = SD_SIBLING_INIT; |
6100 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6101 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6102 | sd->parent = p; |
1a848870 | 6103 | p->child = sd; |
6711cab4 | 6104 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6105 | #endif |
6106 | } | |
6107 | ||
6108 | #ifdef CONFIG_SCHED_SMT | |
6109 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6110 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6111 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6112 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6113 | if (i != first_cpu(this_sibling_map)) |
6114 | continue; | |
6115 | ||
dd41f596 IM |
6116 | init_sched_build_groups(this_sibling_map, cpu_map, |
6117 | &cpu_to_cpu_group); | |
1da177e4 LT |
6118 | } |
6119 | #endif | |
6120 | ||
1e9f28fa SS |
6121 | #ifdef CONFIG_SCHED_MC |
6122 | /* Set up multi-core groups */ | |
6123 | for_each_cpu_mask(i, *cpu_map) { | |
6124 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6125 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6126 | if (i != first_cpu(this_core_map)) | |
6127 | continue; | |
dd41f596 IM |
6128 | init_sched_build_groups(this_core_map, cpu_map, |
6129 | &cpu_to_core_group); | |
1e9f28fa SS |
6130 | } |
6131 | #endif | |
6132 | ||
1da177e4 LT |
6133 | /* Set up physical groups */ |
6134 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6135 | cpumask_t nodemask = node_to_cpumask(i); | |
6136 | ||
1a20ff27 | 6137 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6138 | if (cpus_empty(nodemask)) |
6139 | continue; | |
6140 | ||
6711cab4 | 6141 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6142 | } |
6143 | ||
6144 | #ifdef CONFIG_NUMA | |
6145 | /* Set up node groups */ | |
6711cab4 | 6146 | if (sd_allnodes) |
dd41f596 IM |
6147 | init_sched_build_groups(*cpu_map, cpu_map, |
6148 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6149 | |
6150 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6151 | /* Set up node groups */ | |
6152 | struct sched_group *sg, *prev; | |
6153 | cpumask_t nodemask = node_to_cpumask(i); | |
6154 | cpumask_t domainspan; | |
6155 | cpumask_t covered = CPU_MASK_NONE; | |
6156 | int j; | |
6157 | ||
6158 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6159 | if (cpus_empty(nodemask)) { |
6160 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6161 | continue; |
d1b55138 | 6162 | } |
9c1cfda2 JH |
6163 | |
6164 | domainspan = sched_domain_node_span(i); | |
6165 | cpus_and(domainspan, domainspan, *cpu_map); | |
6166 | ||
15f0b676 | 6167 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6168 | if (!sg) { |
6169 | printk(KERN_WARNING "Can not alloc domain group for " | |
6170 | "node %d\n", i); | |
6171 | goto error; | |
6172 | } | |
9c1cfda2 JH |
6173 | sched_group_nodes[i] = sg; |
6174 | for_each_cpu_mask(j, nodemask) { | |
6175 | struct sched_domain *sd; | |
9761eea8 | 6176 | |
9c1cfda2 JH |
6177 | sd = &per_cpu(node_domains, j); |
6178 | sd->groups = sg; | |
9c1cfda2 | 6179 | } |
5517d86b | 6180 | sg->__cpu_power = 0; |
9c1cfda2 | 6181 | sg->cpumask = nodemask; |
51888ca2 | 6182 | sg->next = sg; |
9c1cfda2 JH |
6183 | cpus_or(covered, covered, nodemask); |
6184 | prev = sg; | |
6185 | ||
6186 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6187 | cpumask_t tmp, notcovered; | |
6188 | int n = (i + j) % MAX_NUMNODES; | |
6189 | ||
6190 | cpus_complement(notcovered, covered); | |
6191 | cpus_and(tmp, notcovered, *cpu_map); | |
6192 | cpus_and(tmp, tmp, domainspan); | |
6193 | if (cpus_empty(tmp)) | |
6194 | break; | |
6195 | ||
6196 | nodemask = node_to_cpumask(n); | |
6197 | cpus_and(tmp, tmp, nodemask); | |
6198 | if (cpus_empty(tmp)) | |
6199 | continue; | |
6200 | ||
15f0b676 SV |
6201 | sg = kmalloc_node(sizeof(struct sched_group), |
6202 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6203 | if (!sg) { |
6204 | printk(KERN_WARNING | |
6205 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6206 | goto error; |
9c1cfda2 | 6207 | } |
5517d86b | 6208 | sg->__cpu_power = 0; |
9c1cfda2 | 6209 | sg->cpumask = tmp; |
51888ca2 | 6210 | sg->next = prev->next; |
9c1cfda2 JH |
6211 | cpus_or(covered, covered, tmp); |
6212 | prev->next = sg; | |
6213 | prev = sg; | |
6214 | } | |
9c1cfda2 | 6215 | } |
1da177e4 LT |
6216 | #endif |
6217 | ||
6218 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6219 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6220 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6221 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6222 | ||
89c4710e | 6223 | init_sched_groups_power(i, sd); |
5c45bf27 | 6224 | } |
1da177e4 | 6225 | #endif |
1e9f28fa | 6226 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6227 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6228 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6229 | ||
89c4710e | 6230 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6231 | } |
6232 | #endif | |
1e9f28fa | 6233 | |
5c45bf27 | 6234 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6235 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6236 | ||
89c4710e | 6237 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6238 | } |
6239 | ||
9c1cfda2 | 6240 | #ifdef CONFIG_NUMA |
08069033 SS |
6241 | for (i = 0; i < MAX_NUMNODES; i++) |
6242 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6243 | |
6711cab4 SS |
6244 | if (sd_allnodes) { |
6245 | struct sched_group *sg; | |
f712c0c7 | 6246 | |
6711cab4 | 6247 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6248 | init_numa_sched_groups_power(sg); |
6249 | } | |
9c1cfda2 JH |
6250 | #endif |
6251 | ||
1da177e4 | 6252 | /* Attach the domains */ |
1a20ff27 | 6253 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6254 | struct sched_domain *sd; |
6255 | #ifdef CONFIG_SCHED_SMT | |
6256 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6257 | #elif defined(CONFIG_SCHED_MC) |
6258 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6259 | #else |
6260 | sd = &per_cpu(phys_domains, i); | |
6261 | #endif | |
6262 | cpu_attach_domain(sd, i); | |
6263 | } | |
51888ca2 SV |
6264 | |
6265 | return 0; | |
6266 | ||
a616058b | 6267 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6268 | error: |
6269 | free_sched_groups(cpu_map); | |
6270 | return -ENOMEM; | |
a616058b | 6271 | #endif |
1da177e4 | 6272 | } |
1a20ff27 DG |
6273 | /* |
6274 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6275 | */ | |
51888ca2 | 6276 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6277 | { |
6278 | cpumask_t cpu_default_map; | |
51888ca2 | 6279 | int err; |
1da177e4 | 6280 | |
1a20ff27 DG |
6281 | /* |
6282 | * Setup mask for cpus without special case scheduling requirements. | |
6283 | * For now this just excludes isolated cpus, but could be used to | |
6284 | * exclude other special cases in the future. | |
6285 | */ | |
6286 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6287 | ||
51888ca2 SV |
6288 | err = build_sched_domains(&cpu_default_map); |
6289 | ||
6290 | return err; | |
1a20ff27 DG |
6291 | } |
6292 | ||
6293 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6294 | { |
51888ca2 | 6295 | free_sched_groups(cpu_map); |
9c1cfda2 | 6296 | } |
1da177e4 | 6297 | |
1a20ff27 DG |
6298 | /* |
6299 | * Detach sched domains from a group of cpus specified in cpu_map | |
6300 | * These cpus will now be attached to the NULL domain | |
6301 | */ | |
858119e1 | 6302 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6303 | { |
6304 | int i; | |
6305 | ||
6306 | for_each_cpu_mask(i, *cpu_map) | |
6307 | cpu_attach_domain(NULL, i); | |
6308 | synchronize_sched(); | |
6309 | arch_destroy_sched_domains(cpu_map); | |
6310 | } | |
6311 | ||
6312 | /* | |
6313 | * Partition sched domains as specified by the cpumasks below. | |
6314 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6315 | * waits for a RCU quiescent period, recalculates sched | |
6316 | * domain information and then attaches them back to the | |
6317 | * correct sched domains | |
6318 | * Call with hotplug lock held | |
6319 | */ | |
51888ca2 | 6320 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6321 | { |
6322 | cpumask_t change_map; | |
51888ca2 | 6323 | int err = 0; |
1a20ff27 DG |
6324 | |
6325 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6326 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6327 | cpus_or(change_map, *partition1, *partition2); | |
6328 | ||
6329 | /* Detach sched domains from all of the affected cpus */ | |
6330 | detach_destroy_domains(&change_map); | |
6331 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6332 | err = build_sched_domains(partition1); |
6333 | if (!err && !cpus_empty(*partition2)) | |
6334 | err = build_sched_domains(partition2); | |
6335 | ||
6336 | return err; | |
1a20ff27 DG |
6337 | } |
6338 | ||
5c45bf27 | 6339 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6340 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6341 | { |
6342 | int err; | |
6343 | ||
5be9361c | 6344 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6345 | detach_destroy_domains(&cpu_online_map); |
6346 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6347 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6348 | |
6349 | return err; | |
6350 | } | |
6351 | ||
6352 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6353 | { | |
6354 | int ret; | |
6355 | ||
6356 | if (buf[0] != '0' && buf[0] != '1') | |
6357 | return -EINVAL; | |
6358 | ||
6359 | if (smt) | |
6360 | sched_smt_power_savings = (buf[0] == '1'); | |
6361 | else | |
6362 | sched_mc_power_savings = (buf[0] == '1'); | |
6363 | ||
6364 | ret = arch_reinit_sched_domains(); | |
6365 | ||
6366 | return ret ? ret : count; | |
6367 | } | |
6368 | ||
5c45bf27 SS |
6369 | #ifdef CONFIG_SCHED_MC |
6370 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6371 | { | |
6372 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6373 | } | |
48f24c4d IM |
6374 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6375 | const char *buf, size_t count) | |
5c45bf27 SS |
6376 | { |
6377 | return sched_power_savings_store(buf, count, 0); | |
6378 | } | |
6707de00 AB |
6379 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6380 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6381 | #endif |
6382 | ||
6383 | #ifdef CONFIG_SCHED_SMT | |
6384 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6385 | { | |
6386 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6387 | } | |
48f24c4d IM |
6388 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6389 | const char *buf, size_t count) | |
5c45bf27 SS |
6390 | { |
6391 | return sched_power_savings_store(buf, count, 1); | |
6392 | } | |
6707de00 AB |
6393 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6394 | sched_smt_power_savings_store); | |
6395 | #endif | |
6396 | ||
6397 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6398 | { | |
6399 | int err = 0; | |
6400 | ||
6401 | #ifdef CONFIG_SCHED_SMT | |
6402 | if (smt_capable()) | |
6403 | err = sysfs_create_file(&cls->kset.kobj, | |
6404 | &attr_sched_smt_power_savings.attr); | |
6405 | #endif | |
6406 | #ifdef CONFIG_SCHED_MC | |
6407 | if (!err && mc_capable()) | |
6408 | err = sysfs_create_file(&cls->kset.kobj, | |
6409 | &attr_sched_mc_power_savings.attr); | |
6410 | #endif | |
6411 | return err; | |
6412 | } | |
5c45bf27 SS |
6413 | #endif |
6414 | ||
1da177e4 LT |
6415 | /* |
6416 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6417 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6418 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6419 | * which will prevent rebalancing while the sched domains are recalculated. |
6420 | */ | |
6421 | static int update_sched_domains(struct notifier_block *nfb, | |
6422 | unsigned long action, void *hcpu) | |
6423 | { | |
1da177e4 LT |
6424 | switch (action) { |
6425 | case CPU_UP_PREPARE: | |
8bb78442 | 6426 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6427 | case CPU_DOWN_PREPARE: |
8bb78442 | 6428 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6429 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6430 | return NOTIFY_OK; |
6431 | ||
6432 | case CPU_UP_CANCELED: | |
8bb78442 | 6433 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6434 | case CPU_DOWN_FAILED: |
8bb78442 | 6435 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6436 | case CPU_ONLINE: |
8bb78442 | 6437 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6438 | case CPU_DEAD: |
8bb78442 | 6439 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6440 | /* |
6441 | * Fall through and re-initialise the domains. | |
6442 | */ | |
6443 | break; | |
6444 | default: | |
6445 | return NOTIFY_DONE; | |
6446 | } | |
6447 | ||
6448 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6449 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6450 | |
6451 | return NOTIFY_OK; | |
6452 | } | |
1da177e4 LT |
6453 | |
6454 | void __init sched_init_smp(void) | |
6455 | { | |
5c1e1767 NP |
6456 | cpumask_t non_isolated_cpus; |
6457 | ||
5be9361c | 6458 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6459 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6460 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6461 | if (cpus_empty(non_isolated_cpus)) |
6462 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6463 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6464 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6465 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 | 6466 | |
e692ab53 NP |
6467 | init_sched_domain_sysctl(); |
6468 | ||
5c1e1767 NP |
6469 | /* Move init over to a non-isolated CPU */ |
6470 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6471 | BUG(); | |
1da177e4 LT |
6472 | } |
6473 | #else | |
6474 | void __init sched_init_smp(void) | |
6475 | { | |
6476 | } | |
6477 | #endif /* CONFIG_SMP */ | |
6478 | ||
6479 | int in_sched_functions(unsigned long addr) | |
6480 | { | |
6481 | /* Linker adds these: start and end of __sched functions */ | |
6482 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6483 | |
1da177e4 LT |
6484 | return in_lock_functions(addr) || |
6485 | (addr >= (unsigned long)__sched_text_start | |
6486 | && addr < (unsigned long)__sched_text_end); | |
6487 | } | |
6488 | ||
a9957449 | 6489 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6490 | { |
6491 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6492 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6493 | cfs_rq->rq = rq; | |
6494 | #endif | |
67e9fb2a | 6495 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6496 | } |
6497 | ||
1da177e4 LT |
6498 | void __init sched_init(void) |
6499 | { | |
476f3534 | 6500 | int highest_cpu = 0; |
dd41f596 IM |
6501 | int i, j; |
6502 | ||
0a945022 | 6503 | for_each_possible_cpu(i) { |
dd41f596 | 6504 | struct rt_prio_array *array; |
70b97a7f | 6505 | struct rq *rq; |
1da177e4 LT |
6506 | |
6507 | rq = cpu_rq(i); | |
6508 | spin_lock_init(&rq->lock); | |
fcb99371 | 6509 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6510 | rq->nr_running = 0; |
dd41f596 IM |
6511 | rq->clock = 1; |
6512 | init_cfs_rq(&rq->cfs, rq); | |
6513 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6514 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6515 | { |
6516 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6517 | struct sched_entity *se = | |
6518 | &per_cpu(init_sched_entity, i); | |
6519 | ||
6520 | init_cfs_rq_p[i] = cfs_rq; | |
6521 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6522 | cfs_rq->tg = &init_task_group; |
3a252015 | 6523 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6524 | &rq->leaf_cfs_rq_list); |
6525 | ||
3a252015 IM |
6526 | init_sched_entity_p[i] = se; |
6527 | se->cfs_rq = &rq->cfs; | |
6528 | se->my_q = cfs_rq; | |
4cf86d77 | 6529 | se->load.weight = init_task_group_load; |
9b5b7751 | 6530 | se->load.inv_weight = |
4cf86d77 | 6531 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6532 | se->parent = NULL; |
6533 | } | |
4cf86d77 | 6534 | init_task_group.shares = init_task_group_load; |
dd41f596 | 6535 | #endif |
1da177e4 | 6536 | |
dd41f596 IM |
6537 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6538 | rq->cpu_load[j] = 0; | |
1da177e4 | 6539 | #ifdef CONFIG_SMP |
41c7ce9a | 6540 | rq->sd = NULL; |
1da177e4 | 6541 | rq->active_balance = 0; |
dd41f596 | 6542 | rq->next_balance = jiffies; |
1da177e4 | 6543 | rq->push_cpu = 0; |
0a2966b4 | 6544 | rq->cpu = i; |
1da177e4 LT |
6545 | rq->migration_thread = NULL; |
6546 | INIT_LIST_HEAD(&rq->migration_queue); | |
6547 | #endif | |
6548 | atomic_set(&rq->nr_iowait, 0); | |
6549 | ||
dd41f596 IM |
6550 | array = &rq->rt.active; |
6551 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6552 | INIT_LIST_HEAD(array->queue + j); | |
6553 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6554 | } |
476f3534 | 6555 | highest_cpu = i; |
dd41f596 IM |
6556 | /* delimiter for bitsearch: */ |
6557 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6558 | } |
6559 | ||
2dd73a4f | 6560 | set_load_weight(&init_task); |
b50f60ce | 6561 | |
e107be36 AK |
6562 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6563 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6564 | #endif | |
6565 | ||
c9819f45 | 6566 | #ifdef CONFIG_SMP |
476f3534 | 6567 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6568 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6569 | #endif | |
6570 | ||
b50f60ce HC |
6571 | #ifdef CONFIG_RT_MUTEXES |
6572 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6573 | #endif | |
6574 | ||
1da177e4 LT |
6575 | /* |
6576 | * The boot idle thread does lazy MMU switching as well: | |
6577 | */ | |
6578 | atomic_inc(&init_mm.mm_count); | |
6579 | enter_lazy_tlb(&init_mm, current); | |
6580 | ||
6581 | /* | |
6582 | * Make us the idle thread. Technically, schedule() should not be | |
6583 | * called from this thread, however somewhere below it might be, | |
6584 | * but because we are the idle thread, we just pick up running again | |
6585 | * when this runqueue becomes "idle". | |
6586 | */ | |
6587 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6588 | /* |
6589 | * During early bootup we pretend to be a normal task: | |
6590 | */ | |
6591 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6592 | } |
6593 | ||
6594 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6595 | void __might_sleep(char *file, int line) | |
6596 | { | |
48f24c4d | 6597 | #ifdef in_atomic |
1da177e4 LT |
6598 | static unsigned long prev_jiffy; /* ratelimiting */ |
6599 | ||
6600 | if ((in_atomic() || irqs_disabled()) && | |
6601 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6602 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6603 | return; | |
6604 | prev_jiffy = jiffies; | |
91368d73 | 6605 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6606 | " context at %s:%d\n", file, line); |
6607 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6608 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6609 | debug_show_held_locks(current); |
3117df04 IM |
6610 | if (irqs_disabled()) |
6611 | print_irqtrace_events(current); | |
1da177e4 LT |
6612 | dump_stack(); |
6613 | } | |
6614 | #endif | |
6615 | } | |
6616 | EXPORT_SYMBOL(__might_sleep); | |
6617 | #endif | |
6618 | ||
6619 | #ifdef CONFIG_MAGIC_SYSRQ | |
6620 | void normalize_rt_tasks(void) | |
6621 | { | |
a0f98a1c | 6622 | struct task_struct *g, *p; |
1da177e4 | 6623 | unsigned long flags; |
70b97a7f | 6624 | struct rq *rq; |
dd41f596 | 6625 | int on_rq; |
1da177e4 LT |
6626 | |
6627 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6628 | do_each_thread(g, p) { |
6cfb0d5d | 6629 | p->se.exec_start = 0; |
6cfb0d5d | 6630 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6631 | p->se.wait_start = 0; |
dd41f596 | 6632 | p->se.sleep_start = 0; |
dd41f596 | 6633 | p->se.block_start = 0; |
6cfb0d5d | 6634 | #endif |
dd41f596 IM |
6635 | task_rq(p)->clock = 0; |
6636 | ||
6637 | if (!rt_task(p)) { | |
6638 | /* | |
6639 | * Renice negative nice level userspace | |
6640 | * tasks back to 0: | |
6641 | */ | |
6642 | if (TASK_NICE(p) < 0 && p->mm) | |
6643 | set_user_nice(p, 0); | |
1da177e4 | 6644 | continue; |
dd41f596 | 6645 | } |
1da177e4 | 6646 | |
b29739f9 IM |
6647 | spin_lock_irqsave(&p->pi_lock, flags); |
6648 | rq = __task_rq_lock(p); | |
dd41f596 IM |
6649 | #ifdef CONFIG_SMP |
6650 | /* | |
6651 | * Do not touch the migration thread: | |
6652 | */ | |
6653 | if (p == rq->migration_thread) | |
6654 | goto out_unlock; | |
6655 | #endif | |
1da177e4 | 6656 | |
2daa3577 | 6657 | update_rq_clock(rq); |
dd41f596 | 6658 | on_rq = p->se.on_rq; |
2daa3577 IM |
6659 | if (on_rq) |
6660 | deactivate_task(rq, p, 0); | |
dd41f596 IM |
6661 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
6662 | if (on_rq) { | |
2daa3577 | 6663 | activate_task(rq, p, 0); |
1da177e4 LT |
6664 | resched_task(rq->curr); |
6665 | } | |
dd41f596 IM |
6666 | #ifdef CONFIG_SMP |
6667 | out_unlock: | |
6668 | #endif | |
b29739f9 IM |
6669 | __task_rq_unlock(rq); |
6670 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6671 | } while_each_thread(g, p); |
6672 | ||
1da177e4 LT |
6673 | read_unlock_irq(&tasklist_lock); |
6674 | } | |
6675 | ||
6676 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6677 | |
6678 | #ifdef CONFIG_IA64 | |
6679 | /* | |
6680 | * These functions are only useful for the IA64 MCA handling. | |
6681 | * | |
6682 | * They can only be called when the whole system has been | |
6683 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6684 | * activity can take place. Using them for anything else would | |
6685 | * be a serious bug, and as a result, they aren't even visible | |
6686 | * under any other configuration. | |
6687 | */ | |
6688 | ||
6689 | /** | |
6690 | * curr_task - return the current task for a given cpu. | |
6691 | * @cpu: the processor in question. | |
6692 | * | |
6693 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6694 | */ | |
36c8b586 | 6695 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6696 | { |
6697 | return cpu_curr(cpu); | |
6698 | } | |
6699 | ||
6700 | /** | |
6701 | * set_curr_task - set the current task for a given cpu. | |
6702 | * @cpu: the processor in question. | |
6703 | * @p: the task pointer to set. | |
6704 | * | |
6705 | * Description: This function must only be used when non-maskable interrupts | |
6706 | * are serviced on a separate stack. It allows the architecture to switch the | |
6707 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6708 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6709 | * and caller must save the original value of the current task (see | |
6710 | * curr_task() above) and restore that value before reenabling interrupts and | |
6711 | * re-starting the system. | |
6712 | * | |
6713 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6714 | */ | |
36c8b586 | 6715 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6716 | { |
6717 | cpu_curr(cpu) = p; | |
6718 | } | |
6719 | ||
6720 | #endif | |
29f59db3 SV |
6721 | |
6722 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6723 | ||
29f59db3 | 6724 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6725 | struct task_group *sched_create_group(void) |
29f59db3 | 6726 | { |
4cf86d77 | 6727 | struct task_group *tg; |
29f59db3 SV |
6728 | struct cfs_rq *cfs_rq; |
6729 | struct sched_entity *se; | |
9b5b7751 | 6730 | struct rq *rq; |
29f59db3 SV |
6731 | int i; |
6732 | ||
29f59db3 SV |
6733 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6734 | if (!tg) | |
6735 | return ERR_PTR(-ENOMEM); | |
6736 | ||
9b5b7751 | 6737 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6738 | if (!tg->cfs_rq) |
6739 | goto err; | |
9b5b7751 | 6740 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6741 | if (!tg->se) |
6742 | goto err; | |
6743 | ||
6744 | for_each_possible_cpu(i) { | |
9b5b7751 | 6745 | rq = cpu_rq(i); |
29f59db3 SV |
6746 | |
6747 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6748 | cpu_to_node(i)); | |
6749 | if (!cfs_rq) | |
6750 | goto err; | |
6751 | ||
6752 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6753 | cpu_to_node(i)); | |
6754 | if (!se) | |
6755 | goto err; | |
6756 | ||
6757 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6758 | memset(se, 0, sizeof(struct sched_entity)); | |
6759 | ||
6760 | tg->cfs_rq[i] = cfs_rq; | |
6761 | init_cfs_rq(cfs_rq, rq); | |
6762 | cfs_rq->tg = tg; | |
29f59db3 SV |
6763 | |
6764 | tg->se[i] = se; | |
6765 | se->cfs_rq = &rq->cfs; | |
6766 | se->my_q = cfs_rq; | |
6767 | se->load.weight = NICE_0_LOAD; | |
6768 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6769 | se->parent = NULL; | |
6770 | } | |
6771 | ||
9b5b7751 SV |
6772 | for_each_possible_cpu(i) { |
6773 | rq = cpu_rq(i); | |
6774 | cfs_rq = tg->cfs_rq[i]; | |
6775 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6776 | } | |
29f59db3 | 6777 | |
9b5b7751 | 6778 | tg->shares = NICE_0_LOAD; |
29f59db3 | 6779 | |
9b5b7751 | 6780 | return tg; |
29f59db3 SV |
6781 | |
6782 | err: | |
6783 | for_each_possible_cpu(i) { | |
a65914b3 | 6784 | if (tg->cfs_rq) |
29f59db3 | 6785 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6786 | if (tg->se) |
29f59db3 SV |
6787 | kfree(tg->se[i]); |
6788 | } | |
a65914b3 IM |
6789 | kfree(tg->cfs_rq); |
6790 | kfree(tg->se); | |
6791 | kfree(tg); | |
29f59db3 SV |
6792 | |
6793 | return ERR_PTR(-ENOMEM); | |
6794 | } | |
6795 | ||
9b5b7751 SV |
6796 | /* rcu callback to free various structures associated with a task group */ |
6797 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6798 | { |
9b5b7751 | 6799 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6800 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6801 | struct sched_entity *se; |
6802 | int i; | |
6803 | ||
29f59db3 SV |
6804 | /* now it should be safe to free those cfs_rqs */ |
6805 | for_each_possible_cpu(i) { | |
6806 | cfs_rq = tg->cfs_rq[i]; | |
6807 | kfree(cfs_rq); | |
6808 | ||
6809 | se = tg->se[i]; | |
6810 | kfree(se); | |
6811 | } | |
6812 | ||
6813 | kfree(tg->cfs_rq); | |
6814 | kfree(tg->se); | |
6815 | kfree(tg); | |
6816 | } | |
6817 | ||
9b5b7751 | 6818 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6819 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 6820 | { |
9b5b7751 SV |
6821 | struct cfs_rq *cfs_rq; |
6822 | int i; | |
29f59db3 | 6823 | |
9b5b7751 SV |
6824 | for_each_possible_cpu(i) { |
6825 | cfs_rq = tg->cfs_rq[i]; | |
6826 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
6827 | } | |
6828 | ||
6829 | cfs_rq = tg->cfs_rq[0]; | |
6830 | ||
6831 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6832 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
6833 | } |
6834 | ||
9b5b7751 | 6835 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6836 | * The caller of this function should have put the task in its new group |
6837 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6838 | * reflect its new group. | |
9b5b7751 SV |
6839 | */ |
6840 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
6841 | { |
6842 | int on_rq, running; | |
6843 | unsigned long flags; | |
6844 | struct rq *rq; | |
6845 | ||
6846 | rq = task_rq_lock(tsk, &flags); | |
6847 | ||
6848 | if (tsk->sched_class != &fair_sched_class) | |
6849 | goto done; | |
6850 | ||
6851 | update_rq_clock(rq); | |
6852 | ||
6853 | running = task_running(rq, tsk); | |
6854 | on_rq = tsk->se.on_rq; | |
6855 | ||
83b699ed | 6856 | if (on_rq) { |
29f59db3 | 6857 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
6858 | if (unlikely(running)) |
6859 | tsk->sched_class->put_prev_task(rq, tsk); | |
6860 | } | |
29f59db3 SV |
6861 | |
6862 | set_task_cfs_rq(tsk); | |
6863 | ||
83b699ed SV |
6864 | if (on_rq) { |
6865 | if (unlikely(running)) | |
6866 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 6867 | enqueue_task(rq, tsk, 0); |
83b699ed | 6868 | } |
29f59db3 SV |
6869 | |
6870 | done: | |
6871 | task_rq_unlock(rq, &flags); | |
6872 | } | |
6873 | ||
6874 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
6875 | { | |
6876 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
6877 | struct rq *rq = cfs_rq->rq; | |
6878 | int on_rq; | |
6879 | ||
6880 | spin_lock_irq(&rq->lock); | |
6881 | ||
6882 | on_rq = se->on_rq; | |
6883 | if (on_rq) | |
6884 | dequeue_entity(cfs_rq, se, 0); | |
6885 | ||
6886 | se->load.weight = shares; | |
6887 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
6888 | ||
6889 | if (on_rq) | |
6890 | enqueue_entity(cfs_rq, se, 0); | |
6891 | ||
6892 | spin_unlock_irq(&rq->lock); | |
6893 | } | |
6894 | ||
4cf86d77 | 6895 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
6896 | { |
6897 | int i; | |
29f59db3 | 6898 | |
9b5b7751 SV |
6899 | if (tg->shares == shares) |
6900 | return 0; | |
29f59db3 | 6901 | |
9b5b7751 | 6902 | /* return -EINVAL if the new value is not sane */ |
29f59db3 | 6903 | |
9b5b7751 | 6904 | tg->shares = shares; |
29f59db3 | 6905 | for_each_possible_cpu(i) |
9b5b7751 | 6906 | set_se_shares(tg->se[i], shares); |
29f59db3 | 6907 | |
9b5b7751 | 6908 | return 0; |
29f59db3 SV |
6909 | } |
6910 | ||
3a252015 | 6911 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |