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; | |
5cb350ba DG |
165 | /* spinlock to serialize modification to shares */ |
166 | spinlock_t lock; | |
29f59db3 SV |
167 | }; |
168 | ||
169 | /* Default task group's sched entity on each cpu */ | |
170 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
171 | /* Default task group's cfs_rq on each cpu */ | |
172 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
173 | ||
9b5b7751 SV |
174 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
175 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
176 | |
177 | /* Default task group. | |
3a252015 | 178 | * Every task in system belong to this group at bootup. |
29f59db3 | 179 | */ |
4cf86d77 | 180 | struct task_group init_task_group = { |
3a252015 IM |
181 | .se = init_sched_entity_p, |
182 | .cfs_rq = init_cfs_rq_p, | |
183 | }; | |
9b5b7751 | 184 | |
24e377a8 | 185 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 186 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 187 | #else |
3a252015 | 188 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
189 | #endif |
190 | ||
4cf86d77 | 191 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
192 | |
193 | /* return group to which a task belongs */ | |
4cf86d77 | 194 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 195 | { |
4cf86d77 | 196 | struct task_group *tg; |
9b5b7751 | 197 | |
24e377a8 SV |
198 | #ifdef CONFIG_FAIR_USER_SCHED |
199 | tg = p->user->tg; | |
200 | #else | |
4cf86d77 | 201 | tg = &init_task_group; |
24e377a8 | 202 | #endif |
9b5b7751 SV |
203 | |
204 | return tg; | |
29f59db3 SV |
205 | } |
206 | ||
207 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
208 | static inline void set_task_cfs_rq(struct task_struct *p) | |
209 | { | |
4cf86d77 IM |
210 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
211 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
212 | } |
213 | ||
214 | #else | |
215 | ||
216 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
217 | ||
218 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
219 | ||
6aa645ea IM |
220 | /* CFS-related fields in a runqueue */ |
221 | struct cfs_rq { | |
222 | struct load_weight load; | |
223 | unsigned long nr_running; | |
224 | ||
6aa645ea | 225 | u64 exec_clock; |
e9acbff6 | 226 | u64 min_vruntime; |
6aa645ea IM |
227 | |
228 | struct rb_root tasks_timeline; | |
229 | struct rb_node *rb_leftmost; | |
230 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
231 | /* 'curr' points to currently running entity on this cfs_rq. |
232 | * It is set to NULL otherwise (i.e when none are currently running). | |
233 | */ | |
234 | struct sched_entity *curr; | |
ddc97297 PZ |
235 | |
236 | unsigned long nr_spread_over; | |
237 | ||
62160e3f | 238 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
239 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
240 | ||
241 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
242 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
243 | * (like users, containers etc.) | |
244 | * | |
245 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
246 | * list is used during load balance. | |
247 | */ | |
248 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 249 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 250 | struct rcu_head rcu; |
6aa645ea IM |
251 | #endif |
252 | }; | |
1da177e4 | 253 | |
6aa645ea IM |
254 | /* Real-Time classes' related field in a runqueue: */ |
255 | struct rt_rq { | |
256 | struct rt_prio_array active; | |
257 | int rt_load_balance_idx; | |
258 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
259 | }; | |
260 | ||
1da177e4 LT |
261 | /* |
262 | * This is the main, per-CPU runqueue data structure. | |
263 | * | |
264 | * Locking rule: those places that want to lock multiple runqueues | |
265 | * (such as the load balancing or the thread migration code), lock | |
266 | * acquire operations must be ordered by ascending &runqueue. | |
267 | */ | |
70b97a7f | 268 | struct rq { |
6aa645ea | 269 | spinlock_t lock; /* runqueue lock */ |
1da177e4 LT |
270 | |
271 | /* | |
272 | * nr_running and cpu_load should be in the same cacheline because | |
273 | * remote CPUs use both these fields when doing load calculation. | |
274 | */ | |
275 | unsigned long nr_running; | |
6aa645ea IM |
276 | #define CPU_LOAD_IDX_MAX 5 |
277 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 278 | unsigned char idle_at_tick; |
46cb4b7c SS |
279 | #ifdef CONFIG_NO_HZ |
280 | unsigned char in_nohz_recently; | |
281 | #endif | |
495eca49 | 282 | struct load_weight load; /* capture load from *all* tasks on this cpu */ |
6aa645ea IM |
283 | unsigned long nr_load_updates; |
284 | u64 nr_switches; | |
285 | ||
286 | struct cfs_rq cfs; | |
287 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
288 | struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */ | |
1da177e4 | 289 | #endif |
6aa645ea | 290 | struct rt_rq rt; |
1da177e4 LT |
291 | |
292 | /* | |
293 | * This is part of a global counter where only the total sum | |
294 | * over all CPUs matters. A task can increase this counter on | |
295 | * one CPU and if it got migrated afterwards it may decrease | |
296 | * it on another CPU. Always updated under the runqueue lock: | |
297 | */ | |
298 | unsigned long nr_uninterruptible; | |
299 | ||
36c8b586 | 300 | struct task_struct *curr, *idle; |
c9819f45 | 301 | unsigned long next_balance; |
1da177e4 | 302 | struct mm_struct *prev_mm; |
6aa645ea | 303 | |
6aa645ea IM |
304 | u64 clock, prev_clock_raw; |
305 | s64 clock_max_delta; | |
306 | ||
307 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
308 | u64 idle_clock; |
309 | unsigned int clock_deep_idle_events; | |
529c7726 | 310 | u64 tick_timestamp; |
6aa645ea | 311 | |
1da177e4 LT |
312 | atomic_t nr_iowait; |
313 | ||
314 | #ifdef CONFIG_SMP | |
315 | struct sched_domain *sd; | |
316 | ||
317 | /* For active balancing */ | |
318 | int active_balance; | |
319 | int push_cpu; | |
0a2966b4 | 320 | int cpu; /* cpu of this runqueue */ |
1da177e4 | 321 | |
36c8b586 | 322 | struct task_struct *migration_thread; |
1da177e4 LT |
323 | struct list_head migration_queue; |
324 | #endif | |
325 | ||
326 | #ifdef CONFIG_SCHEDSTATS | |
327 | /* latency stats */ | |
328 | struct sched_info rq_sched_info; | |
329 | ||
330 | /* sys_sched_yield() stats */ | |
331 | unsigned long yld_exp_empty; | |
332 | unsigned long yld_act_empty; | |
333 | unsigned long yld_both_empty; | |
2d72376b | 334 | unsigned long yld_count; |
1da177e4 LT |
335 | |
336 | /* schedule() stats */ | |
337 | unsigned long sched_switch; | |
2d72376b | 338 | unsigned long sched_count; |
1da177e4 LT |
339 | unsigned long sched_goidle; |
340 | ||
341 | /* try_to_wake_up() stats */ | |
2d72376b | 342 | unsigned long ttwu_count; |
1da177e4 | 343 | unsigned long ttwu_local; |
b8efb561 IM |
344 | |
345 | /* BKL stats */ | |
2d72376b | 346 | unsigned long bkl_count; |
1da177e4 | 347 | #endif |
fcb99371 | 348 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
349 | }; |
350 | ||
f34e3b61 | 351 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 352 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 353 | |
dd41f596 IM |
354 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
355 | { | |
356 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
357 | } | |
358 | ||
0a2966b4 CL |
359 | static inline int cpu_of(struct rq *rq) |
360 | { | |
361 | #ifdef CONFIG_SMP | |
362 | return rq->cpu; | |
363 | #else | |
364 | return 0; | |
365 | #endif | |
366 | } | |
367 | ||
d5036e89 IM |
368 | static inline int is_migration_thread(struct task_struct *p, struct rq *rq) |
369 | { | |
370 | #ifdef CONFIG_SMP | |
371 | return p == rq->migration_thread; | |
372 | #else | |
373 | return 0; | |
374 | #endif | |
375 | } | |
376 | ||
20d315d4 | 377 | /* |
b04a0f4c IM |
378 | * Update the per-runqueue clock, as finegrained as the platform can give |
379 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 380 | */ |
b04a0f4c | 381 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
382 | { |
383 | u64 prev_raw = rq->prev_clock_raw; | |
384 | u64 now = sched_clock(); | |
385 | s64 delta = now - prev_raw; | |
386 | u64 clock = rq->clock; | |
387 | ||
b04a0f4c IM |
388 | #ifdef CONFIG_SCHED_DEBUG |
389 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
390 | #endif | |
20d315d4 IM |
391 | /* |
392 | * Protect against sched_clock() occasionally going backwards: | |
393 | */ | |
394 | if (unlikely(delta < 0)) { | |
395 | clock++; | |
396 | rq->clock_warps++; | |
397 | } else { | |
398 | /* | |
399 | * Catch too large forward jumps too: | |
400 | */ | |
529c7726 IM |
401 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
402 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
403 | clock = rq->tick_timestamp + TICK_NSEC; | |
404 | else | |
405 | clock++; | |
20d315d4 IM |
406 | rq->clock_overflows++; |
407 | } else { | |
408 | if (unlikely(delta > rq->clock_max_delta)) | |
409 | rq->clock_max_delta = delta; | |
410 | clock += delta; | |
411 | } | |
412 | } | |
413 | ||
414 | rq->prev_clock_raw = now; | |
415 | rq->clock = clock; | |
b04a0f4c | 416 | } |
20d315d4 | 417 | |
b04a0f4c IM |
418 | static void update_rq_clock(struct rq *rq) |
419 | { | |
420 | if (likely(smp_processor_id() == cpu_of(rq))) | |
421 | __update_rq_clock(rq); | |
20d315d4 IM |
422 | } |
423 | ||
674311d5 NP |
424 | /* |
425 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 426 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
427 | * |
428 | * The domain tree of any CPU may only be accessed from within | |
429 | * preempt-disabled sections. | |
430 | */ | |
48f24c4d IM |
431 | #define for_each_domain(cpu, __sd) \ |
432 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
433 | |
434 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
435 | #define this_rq() (&__get_cpu_var(runqueues)) | |
436 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
437 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
438 | ||
bf5c91ba IM |
439 | /* |
440 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
441 | */ | |
442 | #ifdef CONFIG_SCHED_DEBUG | |
443 | # define const_debug __read_mostly | |
444 | #else | |
445 | # define const_debug static const | |
446 | #endif | |
447 | ||
448 | /* | |
449 | * Debugging: various feature bits | |
450 | */ | |
451 | enum { | |
bbdba7c0 IM |
452 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
453 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 454 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 455 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 456 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 457 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
458 | }; |
459 | ||
460 | const_debug unsigned int sysctl_sched_features = | |
bf5c91ba | 461 | SCHED_FEAT_NEW_FAIR_SLEEPERS *1 | |
94dfb5e7 | 462 | SCHED_FEAT_START_DEBIT *1 | |
06877c33 | 463 | SCHED_FEAT_TREE_AVG *0 | |
ce6c1311 | 464 | SCHED_FEAT_APPROX_AVG *0 | |
95938a35 MG |
465 | SCHED_FEAT_WAKEUP_PREEMPT *1 | |
466 | SCHED_FEAT_PREEMPT_RESTRICT *1; | |
bf5c91ba IM |
467 | |
468 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
469 | ||
e436d800 IM |
470 | /* |
471 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
472 | * clock constructed from sched_clock(): | |
473 | */ | |
474 | unsigned long long cpu_clock(int cpu) | |
475 | { | |
e436d800 IM |
476 | unsigned long long now; |
477 | unsigned long flags; | |
b04a0f4c | 478 | struct rq *rq; |
e436d800 | 479 | |
2cd4d0ea | 480 | local_irq_save(flags); |
b04a0f4c IM |
481 | rq = cpu_rq(cpu); |
482 | update_rq_clock(rq); | |
483 | now = rq->clock; | |
2cd4d0ea | 484 | local_irq_restore(flags); |
e436d800 IM |
485 | |
486 | return now; | |
487 | } | |
a58f6f25 | 488 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 489 | |
1da177e4 | 490 | #ifndef prepare_arch_switch |
4866cde0 NP |
491 | # define prepare_arch_switch(next) do { } while (0) |
492 | #endif | |
493 | #ifndef finish_arch_switch | |
494 | # define finish_arch_switch(prev) do { } while (0) | |
495 | #endif | |
496 | ||
497 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 498 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
499 | { |
500 | return rq->curr == p; | |
501 | } | |
502 | ||
70b97a7f | 503 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
504 | { |
505 | } | |
506 | ||
70b97a7f | 507 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 508 | { |
da04c035 IM |
509 | #ifdef CONFIG_DEBUG_SPINLOCK |
510 | /* this is a valid case when another task releases the spinlock */ | |
511 | rq->lock.owner = current; | |
512 | #endif | |
8a25d5de IM |
513 | /* |
514 | * If we are tracking spinlock dependencies then we have to | |
515 | * fix up the runqueue lock - which gets 'carried over' from | |
516 | * prev into current: | |
517 | */ | |
518 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
519 | ||
4866cde0 NP |
520 | spin_unlock_irq(&rq->lock); |
521 | } | |
522 | ||
523 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 524 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
525 | { |
526 | #ifdef CONFIG_SMP | |
527 | return p->oncpu; | |
528 | #else | |
529 | return rq->curr == p; | |
530 | #endif | |
531 | } | |
532 | ||
70b97a7f | 533 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
534 | { |
535 | #ifdef CONFIG_SMP | |
536 | /* | |
537 | * We can optimise this out completely for !SMP, because the | |
538 | * SMP rebalancing from interrupt is the only thing that cares | |
539 | * here. | |
540 | */ | |
541 | next->oncpu = 1; | |
542 | #endif | |
543 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
544 | spin_unlock_irq(&rq->lock); | |
545 | #else | |
546 | spin_unlock(&rq->lock); | |
547 | #endif | |
548 | } | |
549 | ||
70b97a7f | 550 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
551 | { |
552 | #ifdef CONFIG_SMP | |
553 | /* | |
554 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
555 | * We must ensure this doesn't happen until the switch is completely | |
556 | * finished. | |
557 | */ | |
558 | smp_wmb(); | |
559 | prev->oncpu = 0; | |
560 | #endif | |
561 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
562 | local_irq_enable(); | |
1da177e4 | 563 | #endif |
4866cde0 NP |
564 | } |
565 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 566 | |
b29739f9 IM |
567 | /* |
568 | * __task_rq_lock - lock the runqueue a given task resides on. | |
569 | * Must be called interrupts disabled. | |
570 | */ | |
70b97a7f | 571 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
572 | __acquires(rq->lock) |
573 | { | |
3a5c359a AK |
574 | for (;;) { |
575 | struct rq *rq = task_rq(p); | |
576 | spin_lock(&rq->lock); | |
577 | if (likely(rq == task_rq(p))) | |
578 | return rq; | |
b29739f9 | 579 | spin_unlock(&rq->lock); |
b29739f9 | 580 | } |
b29739f9 IM |
581 | } |
582 | ||
1da177e4 LT |
583 | /* |
584 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
585 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
586 | * explicitly disabling preemption. | |
587 | */ | |
70b97a7f | 588 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
589 | __acquires(rq->lock) |
590 | { | |
70b97a7f | 591 | struct rq *rq; |
1da177e4 | 592 | |
3a5c359a AK |
593 | for (;;) { |
594 | local_irq_save(*flags); | |
595 | rq = task_rq(p); | |
596 | spin_lock(&rq->lock); | |
597 | if (likely(rq == task_rq(p))) | |
598 | return rq; | |
1da177e4 | 599 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 600 | } |
1da177e4 LT |
601 | } |
602 | ||
a9957449 | 603 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
604 | __releases(rq->lock) |
605 | { | |
606 | spin_unlock(&rq->lock); | |
607 | } | |
608 | ||
70b97a7f | 609 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
610 | __releases(rq->lock) |
611 | { | |
612 | spin_unlock_irqrestore(&rq->lock, *flags); | |
613 | } | |
614 | ||
1da177e4 | 615 | /* |
cc2a73b5 | 616 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 617 | */ |
a9957449 | 618 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
619 | __acquires(rq->lock) |
620 | { | |
70b97a7f | 621 | struct rq *rq; |
1da177e4 LT |
622 | |
623 | local_irq_disable(); | |
624 | rq = this_rq(); | |
625 | spin_lock(&rq->lock); | |
626 | ||
627 | return rq; | |
628 | } | |
629 | ||
1b9f19c2 | 630 | /* |
2aa44d05 | 631 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 632 | */ |
2aa44d05 | 633 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 634 | { |
2aa44d05 IM |
635 | struct rq *rq = cpu_rq(smp_processor_id()); |
636 | ||
637 | spin_lock(&rq->lock); | |
638 | __update_rq_clock(rq); | |
639 | spin_unlock(&rq->lock); | |
640 | rq->clock_deep_idle_events++; | |
641 | } | |
642 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
643 | ||
644 | /* | |
645 | * We just idled delta nanoseconds (called with irqs disabled): | |
646 | */ | |
647 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
648 | { | |
649 | struct rq *rq = cpu_rq(smp_processor_id()); | |
650 | u64 now = sched_clock(); | |
1b9f19c2 | 651 | |
2aa44d05 IM |
652 | rq->idle_clock += delta_ns; |
653 | /* | |
654 | * Override the previous timestamp and ignore all | |
655 | * sched_clock() deltas that occured while we idled, | |
656 | * and use the PM-provided delta_ns to advance the | |
657 | * rq clock: | |
658 | */ | |
659 | spin_lock(&rq->lock); | |
660 | rq->prev_clock_raw = now; | |
661 | rq->clock += delta_ns; | |
662 | spin_unlock(&rq->lock); | |
1b9f19c2 | 663 | } |
2aa44d05 | 664 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 665 | |
c24d20db IM |
666 | /* |
667 | * resched_task - mark a task 'to be rescheduled now'. | |
668 | * | |
669 | * On UP this means the setting of the need_resched flag, on SMP it | |
670 | * might also involve a cross-CPU call to trigger the scheduler on | |
671 | * the target CPU. | |
672 | */ | |
673 | #ifdef CONFIG_SMP | |
674 | ||
675 | #ifndef tsk_is_polling | |
676 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
677 | #endif | |
678 | ||
679 | static void resched_task(struct task_struct *p) | |
680 | { | |
681 | int cpu; | |
682 | ||
683 | assert_spin_locked(&task_rq(p)->lock); | |
684 | ||
685 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
686 | return; | |
687 | ||
688 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
689 | ||
690 | cpu = task_cpu(p); | |
691 | if (cpu == smp_processor_id()) | |
692 | return; | |
693 | ||
694 | /* NEED_RESCHED must be visible before we test polling */ | |
695 | smp_mb(); | |
696 | if (!tsk_is_polling(p)) | |
697 | smp_send_reschedule(cpu); | |
698 | } | |
699 | ||
700 | static void resched_cpu(int cpu) | |
701 | { | |
702 | struct rq *rq = cpu_rq(cpu); | |
703 | unsigned long flags; | |
704 | ||
705 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
706 | return; | |
707 | resched_task(cpu_curr(cpu)); | |
708 | spin_unlock_irqrestore(&rq->lock, flags); | |
709 | } | |
710 | #else | |
711 | static inline void resched_task(struct task_struct *p) | |
712 | { | |
713 | assert_spin_locked(&task_rq(p)->lock); | |
714 | set_tsk_need_resched(p); | |
715 | } | |
716 | #endif | |
717 | ||
45bf76df IM |
718 | #if BITS_PER_LONG == 32 |
719 | # define WMULT_CONST (~0UL) | |
720 | #else | |
721 | # define WMULT_CONST (1UL << 32) | |
722 | #endif | |
723 | ||
724 | #define WMULT_SHIFT 32 | |
725 | ||
194081eb IM |
726 | /* |
727 | * Shift right and round: | |
728 | */ | |
cf2ab469 | 729 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 730 | |
cb1c4fc9 | 731 | static unsigned long |
45bf76df IM |
732 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
733 | struct load_weight *lw) | |
734 | { | |
735 | u64 tmp; | |
736 | ||
737 | if (unlikely(!lw->inv_weight)) | |
194081eb | 738 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
739 | |
740 | tmp = (u64)delta_exec * weight; | |
741 | /* | |
742 | * Check whether we'd overflow the 64-bit multiplication: | |
743 | */ | |
194081eb | 744 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 745 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
746 | WMULT_SHIFT/2); |
747 | else | |
cf2ab469 | 748 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 749 | |
ecf691da | 750 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
751 | } |
752 | ||
753 | static inline unsigned long | |
754 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
755 | { | |
756 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
757 | } | |
758 | ||
1091985b | 759 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
760 | { |
761 | lw->weight += inc; | |
45bf76df IM |
762 | } |
763 | ||
1091985b | 764 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
765 | { |
766 | lw->weight -= dec; | |
45bf76df IM |
767 | } |
768 | ||
2dd73a4f PW |
769 | /* |
770 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
771 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
772 | * each task makes to its run queue's load is weighted according to its | |
773 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
774 | * scaled version of the new time slice allocation that they receive on time | |
775 | * slice expiry etc. | |
776 | */ | |
777 | ||
dd41f596 IM |
778 | #define WEIGHT_IDLEPRIO 2 |
779 | #define WMULT_IDLEPRIO (1 << 31) | |
780 | ||
781 | /* | |
782 | * Nice levels are multiplicative, with a gentle 10% change for every | |
783 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
784 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
785 | * that remained on nice 0. | |
786 | * | |
787 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
788 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
789 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
790 | * If a task goes up by ~10% and another task goes down by ~10% then | |
791 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
792 | */ |
793 | static const int prio_to_weight[40] = { | |
254753dc IM |
794 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
795 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
796 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
797 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
798 | /* 0 */ 1024, 820, 655, 526, 423, | |
799 | /* 5 */ 335, 272, 215, 172, 137, | |
800 | /* 10 */ 110, 87, 70, 56, 45, | |
801 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
802 | }; |
803 | ||
5714d2de IM |
804 | /* |
805 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
806 | * | |
807 | * In cases where the weight does not change often, we can use the | |
808 | * precalculated inverse to speed up arithmetics by turning divisions | |
809 | * into multiplications: | |
810 | */ | |
dd41f596 | 811 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
812 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
813 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
814 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
815 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
816 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
817 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
818 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
819 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 820 | }; |
2dd73a4f | 821 | |
dd41f596 IM |
822 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
823 | ||
824 | /* | |
825 | * runqueue iterator, to support SMP load-balancing between different | |
826 | * scheduling classes, without having to expose their internal data | |
827 | * structures to the load-balancing proper: | |
828 | */ | |
829 | struct rq_iterator { | |
830 | void *arg; | |
831 | struct task_struct *(*start)(void *); | |
832 | struct task_struct *(*next)(void *); | |
833 | }; | |
834 | ||
835 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
836 | unsigned long max_nr_move, unsigned long max_load_move, | |
837 | struct sched_domain *sd, enum cpu_idle_type idle, | |
838 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 839 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
840 | |
841 | #include "sched_stats.h" | |
dd41f596 | 842 | #include "sched_idletask.c" |
5522d5d5 IM |
843 | #include "sched_fair.c" |
844 | #include "sched_rt.c" | |
dd41f596 IM |
845 | #ifdef CONFIG_SCHED_DEBUG |
846 | # include "sched_debug.c" | |
847 | #endif | |
848 | ||
849 | #define sched_class_highest (&rt_sched_class) | |
850 | ||
9c217245 IM |
851 | /* |
852 | * Update delta_exec, delta_fair fields for rq. | |
853 | * | |
854 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 855 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
856 | * delta_exec advances at the same rate as wall-clock (provided |
857 | * cpu is not idle). | |
858 | * | |
859 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
860 | * runqueue over any given interval. This (smoothened) load is used | |
861 | * during load balance. | |
862 | * | |
495eca49 | 863 | * This function is called /before/ updating rq->load |
9c217245 IM |
864 | * and when switching tasks. |
865 | */ | |
29b4b623 | 866 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 867 | { |
495eca49 | 868 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
869 | } |
870 | ||
79b5dddf | 871 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 872 | { |
495eca49 | 873 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
874 | } |
875 | ||
e5fa2237 | 876 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
877 | { |
878 | rq->nr_running++; | |
29b4b623 | 879 | inc_load(rq, p); |
9c217245 IM |
880 | } |
881 | ||
db53181e | 882 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
883 | { |
884 | rq->nr_running--; | |
79b5dddf | 885 | dec_load(rq, p); |
9c217245 IM |
886 | } |
887 | ||
45bf76df IM |
888 | static void set_load_weight(struct task_struct *p) |
889 | { | |
890 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
891 | p->se.load.weight = prio_to_weight[0] * 2; |
892 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
893 | return; | |
894 | } | |
45bf76df | 895 | |
dd41f596 IM |
896 | /* |
897 | * SCHED_IDLE tasks get minimal weight: | |
898 | */ | |
899 | if (p->policy == SCHED_IDLE) { | |
900 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
901 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
902 | return; | |
903 | } | |
71f8bd46 | 904 | |
dd41f596 IM |
905 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
906 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
907 | } |
908 | ||
8159f87e | 909 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 910 | { |
dd41f596 | 911 | sched_info_queued(p); |
fd390f6a | 912 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 913 | p->se.on_rq = 1; |
71f8bd46 IM |
914 | } |
915 | ||
69be72c1 | 916 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 917 | { |
f02231e5 | 918 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 919 | p->se.on_rq = 0; |
71f8bd46 IM |
920 | } |
921 | ||
14531189 | 922 | /* |
dd41f596 | 923 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 924 | */ |
14531189 IM |
925 | static inline int __normal_prio(struct task_struct *p) |
926 | { | |
dd41f596 | 927 | return p->static_prio; |
14531189 IM |
928 | } |
929 | ||
b29739f9 IM |
930 | /* |
931 | * Calculate the expected normal priority: i.e. priority | |
932 | * without taking RT-inheritance into account. Might be | |
933 | * boosted by interactivity modifiers. Changes upon fork, | |
934 | * setprio syscalls, and whenever the interactivity | |
935 | * estimator recalculates. | |
936 | */ | |
36c8b586 | 937 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
938 | { |
939 | int prio; | |
940 | ||
e05606d3 | 941 | if (task_has_rt_policy(p)) |
b29739f9 IM |
942 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
943 | else | |
944 | prio = __normal_prio(p); | |
945 | return prio; | |
946 | } | |
947 | ||
948 | /* | |
949 | * Calculate the current priority, i.e. the priority | |
950 | * taken into account by the scheduler. This value might | |
951 | * be boosted by RT tasks, or might be boosted by | |
952 | * interactivity modifiers. Will be RT if the task got | |
953 | * RT-boosted. If not then it returns p->normal_prio. | |
954 | */ | |
36c8b586 | 955 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
956 | { |
957 | p->normal_prio = normal_prio(p); | |
958 | /* | |
959 | * If we are RT tasks or we were boosted to RT priority, | |
960 | * keep the priority unchanged. Otherwise, update priority | |
961 | * to the normal priority: | |
962 | */ | |
963 | if (!rt_prio(p->prio)) | |
964 | return p->normal_prio; | |
965 | return p->prio; | |
966 | } | |
967 | ||
1da177e4 | 968 | /* |
dd41f596 | 969 | * activate_task - move a task to the runqueue. |
1da177e4 | 970 | */ |
dd41f596 | 971 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 972 | { |
dd41f596 IM |
973 | if (p->state == TASK_UNINTERRUPTIBLE) |
974 | rq->nr_uninterruptible--; | |
1da177e4 | 975 | |
8159f87e | 976 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 977 | inc_nr_running(p, rq); |
1da177e4 LT |
978 | } |
979 | ||
1da177e4 LT |
980 | /* |
981 | * deactivate_task - remove a task from the runqueue. | |
982 | */ | |
2e1cb74a | 983 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 984 | { |
dd41f596 IM |
985 | if (p->state == TASK_UNINTERRUPTIBLE) |
986 | rq->nr_uninterruptible++; | |
987 | ||
69be72c1 | 988 | dequeue_task(rq, p, sleep); |
db53181e | 989 | dec_nr_running(p, rq); |
1da177e4 LT |
990 | } |
991 | ||
1da177e4 LT |
992 | /** |
993 | * task_curr - is this task currently executing on a CPU? | |
994 | * @p: the task in question. | |
995 | */ | |
36c8b586 | 996 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
997 | { |
998 | return cpu_curr(task_cpu(p)) == p; | |
999 | } | |
1000 | ||
2dd73a4f PW |
1001 | /* Used instead of source_load when we know the type == 0 */ |
1002 | unsigned long weighted_cpuload(const int cpu) | |
1003 | { | |
495eca49 | 1004 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1005 | } |
1006 | ||
1007 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1008 | { | |
1009 | #ifdef CONFIG_SMP | |
1010 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1011 | #endif |
29f59db3 | 1012 | set_task_cfs_rq(p); |
2dd73a4f PW |
1013 | } |
1014 | ||
1da177e4 | 1015 | #ifdef CONFIG_SMP |
c65cc870 | 1016 | |
dd41f596 | 1017 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1018 | { |
dd41f596 IM |
1019 | int old_cpu = task_cpu(p); |
1020 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1021 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1022 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1023 | u64 clock_offset; |
dd41f596 IM |
1024 | |
1025 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1026 | |
1027 | #ifdef CONFIG_SCHEDSTATS | |
1028 | if (p->se.wait_start) | |
1029 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1030 | if (p->se.sleep_start) |
1031 | p->se.sleep_start -= clock_offset; | |
1032 | if (p->se.block_start) | |
1033 | p->se.block_start -= clock_offset; | |
6cfb0d5d | 1034 | #endif |
2830cf8c SV |
1035 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1036 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1037 | |
1038 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1039 | } |
1040 | ||
70b97a7f | 1041 | struct migration_req { |
1da177e4 | 1042 | struct list_head list; |
1da177e4 | 1043 | |
36c8b586 | 1044 | struct task_struct *task; |
1da177e4 LT |
1045 | int dest_cpu; |
1046 | ||
1da177e4 | 1047 | struct completion done; |
70b97a7f | 1048 | }; |
1da177e4 LT |
1049 | |
1050 | /* | |
1051 | * The task's runqueue lock must be held. | |
1052 | * Returns true if you have to wait for migration thread. | |
1053 | */ | |
36c8b586 | 1054 | static int |
70b97a7f | 1055 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1056 | { |
70b97a7f | 1057 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1058 | |
1059 | /* | |
1060 | * If the task is not on a runqueue (and not running), then | |
1061 | * it is sufficient to simply update the task's cpu field. | |
1062 | */ | |
dd41f596 | 1063 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1064 | set_task_cpu(p, dest_cpu); |
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | init_completion(&req->done); | |
1da177e4 LT |
1069 | req->task = p; |
1070 | req->dest_cpu = dest_cpu; | |
1071 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1072 | |
1da177e4 LT |
1073 | return 1; |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * wait_task_inactive - wait for a thread to unschedule. | |
1078 | * | |
1079 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1080 | * else this function might spin for a *long* time. This function can't | |
1081 | * be called with interrupts off, or it may introduce deadlock with | |
1082 | * smp_call_function() if an IPI is sent by the same process we are | |
1083 | * waiting to become inactive. | |
1084 | */ | |
36c8b586 | 1085 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1086 | { |
1087 | unsigned long flags; | |
dd41f596 | 1088 | int running, on_rq; |
70b97a7f | 1089 | struct rq *rq; |
1da177e4 | 1090 | |
3a5c359a AK |
1091 | for (;;) { |
1092 | /* | |
1093 | * We do the initial early heuristics without holding | |
1094 | * any task-queue locks at all. We'll only try to get | |
1095 | * the runqueue lock when things look like they will | |
1096 | * work out! | |
1097 | */ | |
1098 | rq = task_rq(p); | |
fa490cfd | 1099 | |
3a5c359a AK |
1100 | /* |
1101 | * If the task is actively running on another CPU | |
1102 | * still, just relax and busy-wait without holding | |
1103 | * any locks. | |
1104 | * | |
1105 | * NOTE! Since we don't hold any locks, it's not | |
1106 | * even sure that "rq" stays as the right runqueue! | |
1107 | * But we don't care, since "task_running()" will | |
1108 | * return false if the runqueue has changed and p | |
1109 | * is actually now running somewhere else! | |
1110 | */ | |
1111 | while (task_running(rq, p)) | |
1112 | cpu_relax(); | |
fa490cfd | 1113 | |
3a5c359a AK |
1114 | /* |
1115 | * Ok, time to look more closely! We need the rq | |
1116 | * lock now, to be *sure*. If we're wrong, we'll | |
1117 | * just go back and repeat. | |
1118 | */ | |
1119 | rq = task_rq_lock(p, &flags); | |
1120 | running = task_running(rq, p); | |
1121 | on_rq = p->se.on_rq; | |
1122 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1123 | |
3a5c359a AK |
1124 | /* |
1125 | * Was it really running after all now that we | |
1126 | * checked with the proper locks actually held? | |
1127 | * | |
1128 | * Oops. Go back and try again.. | |
1129 | */ | |
1130 | if (unlikely(running)) { | |
1131 | cpu_relax(); | |
1132 | continue; | |
1133 | } | |
fa490cfd | 1134 | |
3a5c359a AK |
1135 | /* |
1136 | * It's not enough that it's not actively running, | |
1137 | * it must be off the runqueue _entirely_, and not | |
1138 | * preempted! | |
1139 | * | |
1140 | * So if it wa still runnable (but just not actively | |
1141 | * running right now), it's preempted, and we should | |
1142 | * yield - it could be a while. | |
1143 | */ | |
1144 | if (unlikely(on_rq)) { | |
1145 | schedule_timeout_uninterruptible(1); | |
1146 | continue; | |
1147 | } | |
fa490cfd | 1148 | |
3a5c359a AK |
1149 | /* |
1150 | * Ahh, all good. It wasn't running, and it wasn't | |
1151 | * runnable, which means that it will never become | |
1152 | * running in the future either. We're all done! | |
1153 | */ | |
1154 | break; | |
1155 | } | |
1da177e4 LT |
1156 | } |
1157 | ||
1158 | /*** | |
1159 | * kick_process - kick a running thread to enter/exit the kernel | |
1160 | * @p: the to-be-kicked thread | |
1161 | * | |
1162 | * Cause a process which is running on another CPU to enter | |
1163 | * kernel-mode, without any delay. (to get signals handled.) | |
1164 | * | |
1165 | * NOTE: this function doesnt have to take the runqueue lock, | |
1166 | * because all it wants to ensure is that the remote task enters | |
1167 | * the kernel. If the IPI races and the task has been migrated | |
1168 | * to another CPU then no harm is done and the purpose has been | |
1169 | * achieved as well. | |
1170 | */ | |
36c8b586 | 1171 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1172 | { |
1173 | int cpu; | |
1174 | ||
1175 | preempt_disable(); | |
1176 | cpu = task_cpu(p); | |
1177 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1178 | smp_send_reschedule(cpu); | |
1179 | preempt_enable(); | |
1180 | } | |
1181 | ||
1182 | /* | |
2dd73a4f PW |
1183 | * Return a low guess at the load of a migration-source cpu weighted |
1184 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1185 | * |
1186 | * We want to under-estimate the load of migration sources, to | |
1187 | * balance conservatively. | |
1188 | */ | |
a9957449 | 1189 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1190 | { |
70b97a7f | 1191 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1192 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1193 | |
3b0bd9bc | 1194 | if (type == 0) |
dd41f596 | 1195 | return total; |
b910472d | 1196 | |
dd41f596 | 1197 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | /* | |
2dd73a4f PW |
1201 | * Return a high guess at the load of a migration-target cpu weighted |
1202 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1203 | */ |
a9957449 | 1204 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1205 | { |
70b97a7f | 1206 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1207 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1208 | |
7897986b | 1209 | if (type == 0) |
dd41f596 | 1210 | return total; |
3b0bd9bc | 1211 | |
dd41f596 | 1212 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1213 | } |
1214 | ||
1215 | /* | |
1216 | * Return the average load per task on the cpu's run queue | |
1217 | */ | |
1218 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1219 | { | |
70b97a7f | 1220 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1221 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1222 | unsigned long n = rq->nr_running; |
1223 | ||
dd41f596 | 1224 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1225 | } |
1226 | ||
147cbb4b NP |
1227 | /* |
1228 | * find_idlest_group finds and returns the least busy CPU group within the | |
1229 | * domain. | |
1230 | */ | |
1231 | static struct sched_group * | |
1232 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1233 | { | |
1234 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1235 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1236 | int load_idx = sd->forkexec_idx; | |
1237 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1238 | ||
1239 | do { | |
1240 | unsigned long load, avg_load; | |
1241 | int local_group; | |
1242 | int i; | |
1243 | ||
da5a5522 BD |
1244 | /* Skip over this group if it has no CPUs allowed */ |
1245 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1246 | continue; |
da5a5522 | 1247 | |
147cbb4b | 1248 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1249 | |
1250 | /* Tally up the load of all CPUs in the group */ | |
1251 | avg_load = 0; | |
1252 | ||
1253 | for_each_cpu_mask(i, group->cpumask) { | |
1254 | /* Bias balancing toward cpus of our domain */ | |
1255 | if (local_group) | |
1256 | load = source_load(i, load_idx); | |
1257 | else | |
1258 | load = target_load(i, load_idx); | |
1259 | ||
1260 | avg_load += load; | |
1261 | } | |
1262 | ||
1263 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1264 | avg_load = sg_div_cpu_power(group, |
1265 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1266 | |
1267 | if (local_group) { | |
1268 | this_load = avg_load; | |
1269 | this = group; | |
1270 | } else if (avg_load < min_load) { | |
1271 | min_load = avg_load; | |
1272 | idlest = group; | |
1273 | } | |
3a5c359a | 1274 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1275 | |
1276 | if (!idlest || 100*this_load < imbalance*min_load) | |
1277 | return NULL; | |
1278 | return idlest; | |
1279 | } | |
1280 | ||
1281 | /* | |
0feaece9 | 1282 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1283 | */ |
95cdf3b7 IM |
1284 | static int |
1285 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1286 | { |
da5a5522 | 1287 | cpumask_t tmp; |
147cbb4b NP |
1288 | unsigned long load, min_load = ULONG_MAX; |
1289 | int idlest = -1; | |
1290 | int i; | |
1291 | ||
da5a5522 BD |
1292 | /* Traverse only the allowed CPUs */ |
1293 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1294 | ||
1295 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1296 | load = weighted_cpuload(i); |
147cbb4b NP |
1297 | |
1298 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1299 | min_load = load; | |
1300 | idlest = i; | |
1301 | } | |
1302 | } | |
1303 | ||
1304 | return idlest; | |
1305 | } | |
1306 | ||
476d139c NP |
1307 | /* |
1308 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1309 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1310 | * SD_BALANCE_EXEC. | |
1311 | * | |
1312 | * Balance, ie. select the least loaded group. | |
1313 | * | |
1314 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1315 | * | |
1316 | * preempt must be disabled. | |
1317 | */ | |
1318 | static int sched_balance_self(int cpu, int flag) | |
1319 | { | |
1320 | struct task_struct *t = current; | |
1321 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1322 | |
c96d145e | 1323 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1324 | /* |
1325 | * If power savings logic is enabled for a domain, stop there. | |
1326 | */ | |
5c45bf27 SS |
1327 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1328 | break; | |
476d139c NP |
1329 | if (tmp->flags & flag) |
1330 | sd = tmp; | |
c96d145e | 1331 | } |
476d139c NP |
1332 | |
1333 | while (sd) { | |
1334 | cpumask_t span; | |
1335 | struct sched_group *group; | |
1a848870 SS |
1336 | int new_cpu, weight; |
1337 | ||
1338 | if (!(sd->flags & flag)) { | |
1339 | sd = sd->child; | |
1340 | continue; | |
1341 | } | |
476d139c NP |
1342 | |
1343 | span = sd->span; | |
1344 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1345 | if (!group) { |
1346 | sd = sd->child; | |
1347 | continue; | |
1348 | } | |
476d139c | 1349 | |
da5a5522 | 1350 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1351 | if (new_cpu == -1 || new_cpu == cpu) { |
1352 | /* Now try balancing at a lower domain level of cpu */ | |
1353 | sd = sd->child; | |
1354 | continue; | |
1355 | } | |
476d139c | 1356 | |
1a848870 | 1357 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1358 | cpu = new_cpu; |
476d139c NP |
1359 | sd = NULL; |
1360 | weight = cpus_weight(span); | |
1361 | for_each_domain(cpu, tmp) { | |
1362 | if (weight <= cpus_weight(tmp->span)) | |
1363 | break; | |
1364 | if (tmp->flags & flag) | |
1365 | sd = tmp; | |
1366 | } | |
1367 | /* while loop will break here if sd == NULL */ | |
1368 | } | |
1369 | ||
1370 | return cpu; | |
1371 | } | |
1372 | ||
1373 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1374 | |
1375 | /* | |
1376 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1377 | * not idle and an idle cpu is available. The span of cpus to | |
1378 | * search starts with cpus closest then further out as needed, | |
1379 | * so we always favor a closer, idle cpu. | |
1380 | * | |
1381 | * Returns the CPU we should wake onto. | |
1382 | */ | |
1383 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1384 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1385 | { |
1386 | cpumask_t tmp; | |
1387 | struct sched_domain *sd; | |
1388 | int i; | |
1389 | ||
4953198b SS |
1390 | /* |
1391 | * If it is idle, then it is the best cpu to run this task. | |
1392 | * | |
1393 | * This cpu is also the best, if it has more than one task already. | |
1394 | * Siblings must be also busy(in most cases) as they didn't already | |
1395 | * pickup the extra load from this cpu and hence we need not check | |
1396 | * sibling runqueue info. This will avoid the checks and cache miss | |
1397 | * penalities associated with that. | |
1398 | */ | |
1399 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1400 | return cpu; |
1401 | ||
1402 | for_each_domain(cpu, sd) { | |
1403 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1404 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 LT |
1405 | for_each_cpu_mask(i, tmp) { |
1406 | if (idle_cpu(i)) | |
1407 | return i; | |
1408 | } | |
9761eea8 | 1409 | } else { |
e0f364f4 | 1410 | break; |
9761eea8 | 1411 | } |
1da177e4 LT |
1412 | } |
1413 | return cpu; | |
1414 | } | |
1415 | #else | |
36c8b586 | 1416 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1417 | { |
1418 | return cpu; | |
1419 | } | |
1420 | #endif | |
1421 | ||
1422 | /*** | |
1423 | * try_to_wake_up - wake up a thread | |
1424 | * @p: the to-be-woken-up thread | |
1425 | * @state: the mask of task states that can be woken | |
1426 | * @sync: do a synchronous wakeup? | |
1427 | * | |
1428 | * Put it on the run-queue if it's not already there. The "current" | |
1429 | * thread is always on the run-queue (except when the actual | |
1430 | * re-schedule is in progress), and as such you're allowed to do | |
1431 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1432 | * runnable without the overhead of this. | |
1433 | * | |
1434 | * returns failure only if the task is already active. | |
1435 | */ | |
36c8b586 | 1436 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 LT |
1437 | { |
1438 | int cpu, this_cpu, success = 0; | |
1439 | unsigned long flags; | |
1440 | long old_state; | |
70b97a7f | 1441 | struct rq *rq; |
1da177e4 | 1442 | #ifdef CONFIG_SMP |
7897986b | 1443 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1444 | unsigned long load, this_load; |
1da177e4 LT |
1445 | int new_cpu; |
1446 | #endif | |
1447 | ||
1448 | rq = task_rq_lock(p, &flags); | |
1449 | old_state = p->state; | |
1450 | if (!(old_state & state)) | |
1451 | goto out; | |
1452 | ||
dd41f596 | 1453 | if (p->se.on_rq) |
1da177e4 LT |
1454 | goto out_running; |
1455 | ||
1456 | cpu = task_cpu(p); | |
1457 | this_cpu = smp_processor_id(); | |
1458 | ||
1459 | #ifdef CONFIG_SMP | |
1460 | if (unlikely(task_running(rq, p))) | |
1461 | goto out_activate; | |
1462 | ||
7897986b NP |
1463 | new_cpu = cpu; |
1464 | ||
2d72376b | 1465 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1466 | if (cpu == this_cpu) { |
1467 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1468 | goto out_set_cpu; |
1469 | } | |
1470 | ||
1471 | for_each_domain(this_cpu, sd) { | |
1472 | if (cpu_isset(cpu, sd->span)) { | |
1473 | schedstat_inc(sd, ttwu_wake_remote); | |
1474 | this_sd = sd; | |
1475 | break; | |
1da177e4 LT |
1476 | } |
1477 | } | |
1da177e4 | 1478 | |
7897986b | 1479 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1480 | goto out_set_cpu; |
1481 | ||
1da177e4 | 1482 | /* |
7897986b | 1483 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1484 | */ |
7897986b NP |
1485 | if (this_sd) { |
1486 | int idx = this_sd->wake_idx; | |
1487 | unsigned int imbalance; | |
1da177e4 | 1488 | |
a3f21bce NP |
1489 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1490 | ||
7897986b NP |
1491 | load = source_load(cpu, idx); |
1492 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1493 | |
7897986b NP |
1494 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1495 | ||
a3f21bce NP |
1496 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1497 | unsigned long tl = this_load; | |
33859f7f MOS |
1498 | unsigned long tl_per_task; |
1499 | ||
1500 | tl_per_task = cpu_avg_load_per_task(this_cpu); | |
2dd73a4f | 1501 | |
1da177e4 | 1502 | /* |
a3f21bce NP |
1503 | * If sync wakeup then subtract the (maximum possible) |
1504 | * effect of the currently running task from the load | |
1505 | * of the current CPU: | |
1da177e4 | 1506 | */ |
a3f21bce | 1507 | if (sync) |
dd41f596 | 1508 | tl -= current->se.load.weight; |
a3f21bce NP |
1509 | |
1510 | if ((tl <= load && | |
2dd73a4f | 1511 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1512 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1513 | /* |
1514 | * This domain has SD_WAKE_AFFINE and | |
1515 | * p is cache cold in this domain, and | |
1516 | * there is no bad imbalance. | |
1517 | */ | |
1518 | schedstat_inc(this_sd, ttwu_move_affine); | |
1519 | goto out_set_cpu; | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * Start passive balancing when half the imbalance_pct | |
1525 | * limit is reached. | |
1526 | */ | |
1527 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1528 | if (imbalance*this_load <= 100*load) { | |
1529 | schedstat_inc(this_sd, ttwu_move_balance); | |
1530 | goto out_set_cpu; | |
1531 | } | |
1da177e4 LT |
1532 | } |
1533 | } | |
1534 | ||
1535 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1536 | out_set_cpu: | |
1537 | new_cpu = wake_idle(new_cpu, p); | |
1538 | if (new_cpu != cpu) { | |
1539 | set_task_cpu(p, new_cpu); | |
1540 | task_rq_unlock(rq, &flags); | |
1541 | /* might preempt at this point */ | |
1542 | rq = task_rq_lock(p, &flags); | |
1543 | old_state = p->state; | |
1544 | if (!(old_state & state)) | |
1545 | goto out; | |
dd41f596 | 1546 | if (p->se.on_rq) |
1da177e4 LT |
1547 | goto out_running; |
1548 | ||
1549 | this_cpu = smp_processor_id(); | |
1550 | cpu = task_cpu(p); | |
1551 | } | |
1552 | ||
1553 | out_activate: | |
1554 | #endif /* CONFIG_SMP */ | |
2daa3577 | 1555 | update_rq_clock(rq); |
dd41f596 | 1556 | activate_task(rq, p, 1); |
1da177e4 LT |
1557 | /* |
1558 | * Sync wakeups (i.e. those types of wakeups where the waker | |
1559 | * has indicated that it will leave the CPU in short order) | |
1560 | * don't trigger a preemption, if the woken up task will run on | |
1561 | * this cpu. (in this case the 'I will reschedule' promise of | |
1562 | * the waker guarantees that the freshly woken up task is going | |
1563 | * to be considered on this CPU.) | |
1564 | */ | |
dd41f596 IM |
1565 | if (!sync || cpu != this_cpu) |
1566 | check_preempt_curr(rq, p); | |
1da177e4 LT |
1567 | success = 1; |
1568 | ||
1569 | out_running: | |
1570 | p->state = TASK_RUNNING; | |
1571 | out: | |
1572 | task_rq_unlock(rq, &flags); | |
1573 | ||
1574 | return success; | |
1575 | } | |
1576 | ||
36c8b586 | 1577 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1578 | { |
1579 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1580 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1581 | } | |
1da177e4 LT |
1582 | EXPORT_SYMBOL(wake_up_process); |
1583 | ||
36c8b586 | 1584 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1585 | { |
1586 | return try_to_wake_up(p, state, 0); | |
1587 | } | |
1588 | ||
1da177e4 LT |
1589 | /* |
1590 | * Perform scheduler related setup for a newly forked process p. | |
1591 | * p is forked by current. | |
dd41f596 IM |
1592 | * |
1593 | * __sched_fork() is basic setup used by init_idle() too: | |
1594 | */ | |
1595 | static void __sched_fork(struct task_struct *p) | |
1596 | { | |
dd41f596 IM |
1597 | p->se.exec_start = 0; |
1598 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1599 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1600 | |
1601 | #ifdef CONFIG_SCHEDSTATS | |
1602 | p->se.wait_start = 0; | |
dd41f596 IM |
1603 | p->se.sum_sleep_runtime = 0; |
1604 | p->se.sleep_start = 0; | |
dd41f596 IM |
1605 | p->se.block_start = 0; |
1606 | p->se.sleep_max = 0; | |
1607 | p->se.block_max = 0; | |
1608 | p->se.exec_max = 0; | |
eba1ed4b | 1609 | p->se.slice_max = 0; |
dd41f596 | 1610 | p->se.wait_max = 0; |
6cfb0d5d | 1611 | #endif |
476d139c | 1612 | |
dd41f596 IM |
1613 | INIT_LIST_HEAD(&p->run_list); |
1614 | p->se.on_rq = 0; | |
476d139c | 1615 | |
e107be36 AK |
1616 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1617 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1618 | #endif | |
1619 | ||
1da177e4 LT |
1620 | /* |
1621 | * We mark the process as running here, but have not actually | |
1622 | * inserted it onto the runqueue yet. This guarantees that | |
1623 | * nobody will actually run it, and a signal or other external | |
1624 | * event cannot wake it up and insert it on the runqueue either. | |
1625 | */ | |
1626 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1627 | } |
1628 | ||
1629 | /* | |
1630 | * fork()/clone()-time setup: | |
1631 | */ | |
1632 | void sched_fork(struct task_struct *p, int clone_flags) | |
1633 | { | |
1634 | int cpu = get_cpu(); | |
1635 | ||
1636 | __sched_fork(p); | |
1637 | ||
1638 | #ifdef CONFIG_SMP | |
1639 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1640 | #endif | |
02e4bac2 | 1641 | set_task_cpu(p, cpu); |
b29739f9 IM |
1642 | |
1643 | /* | |
1644 | * Make sure we do not leak PI boosting priority to the child: | |
1645 | */ | |
1646 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1647 | if (!rt_prio(p->prio)) |
1648 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1649 | |
52f17b6c | 1650 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1651 | if (likely(sched_info_on())) |
52f17b6c | 1652 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1653 | #endif |
d6077cb8 | 1654 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1655 | p->oncpu = 0; |
1656 | #endif | |
1da177e4 | 1657 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1658 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1659 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1660 | #endif |
476d139c | 1661 | put_cpu(); |
1da177e4 LT |
1662 | } |
1663 | ||
1664 | /* | |
1665 | * wake_up_new_task - wake up a newly created task for the first time. | |
1666 | * | |
1667 | * This function will do some initial scheduler statistics housekeeping | |
1668 | * that must be done for every newly created context, then puts the task | |
1669 | * on the runqueue and wakes it. | |
1670 | */ | |
36c8b586 | 1671 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1672 | { |
1673 | unsigned long flags; | |
dd41f596 | 1674 | struct rq *rq; |
1da177e4 LT |
1675 | |
1676 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1677 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1678 | update_rq_clock(rq); |
1da177e4 LT |
1679 | |
1680 | p->prio = effective_prio(p); | |
1681 | ||
00bf7bfc | 1682 | if (!p->sched_class->task_new || !current->se.on_rq || !rq->cfs.curr) { |
dd41f596 | 1683 | activate_task(rq, p, 0); |
1da177e4 | 1684 | } else { |
1da177e4 | 1685 | /* |
dd41f596 IM |
1686 | * Let the scheduling class do new task startup |
1687 | * management (if any): | |
1da177e4 | 1688 | */ |
ee0827d8 | 1689 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1690 | inc_nr_running(p, rq); |
1da177e4 | 1691 | } |
dd41f596 IM |
1692 | check_preempt_curr(rq, p); |
1693 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1694 | } |
1695 | ||
e107be36 AK |
1696 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1697 | ||
1698 | /** | |
421cee29 RD |
1699 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1700 | * @notifier: notifier struct to register | |
e107be36 AK |
1701 | */ |
1702 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1703 | { | |
1704 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1705 | } | |
1706 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1707 | ||
1708 | /** | |
1709 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1710 | * @notifier: notifier struct to unregister |
e107be36 AK |
1711 | * |
1712 | * This is safe to call from within a preemption notifier. | |
1713 | */ | |
1714 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1715 | { | |
1716 | hlist_del(¬ifier->link); | |
1717 | } | |
1718 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1719 | ||
1720 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1721 | { | |
1722 | struct preempt_notifier *notifier; | |
1723 | struct hlist_node *node; | |
1724 | ||
1725 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1726 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1727 | } | |
1728 | ||
1729 | static void | |
1730 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1731 | struct task_struct *next) | |
1732 | { | |
1733 | struct preempt_notifier *notifier; | |
1734 | struct hlist_node *node; | |
1735 | ||
1736 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1737 | notifier->ops->sched_out(notifier, next); | |
1738 | } | |
1739 | ||
1740 | #else | |
1741 | ||
1742 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1743 | { | |
1744 | } | |
1745 | ||
1746 | static void | |
1747 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1748 | struct task_struct *next) | |
1749 | { | |
1750 | } | |
1751 | ||
1752 | #endif | |
1753 | ||
4866cde0 NP |
1754 | /** |
1755 | * prepare_task_switch - prepare to switch tasks | |
1756 | * @rq: the runqueue preparing to switch | |
421cee29 | 1757 | * @prev: the current task that is being switched out |
4866cde0 NP |
1758 | * @next: the task we are going to switch to. |
1759 | * | |
1760 | * This is called with the rq lock held and interrupts off. It must | |
1761 | * be paired with a subsequent finish_task_switch after the context | |
1762 | * switch. | |
1763 | * | |
1764 | * prepare_task_switch sets up locking and calls architecture specific | |
1765 | * hooks. | |
1766 | */ | |
e107be36 AK |
1767 | static inline void |
1768 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1769 | struct task_struct *next) | |
4866cde0 | 1770 | { |
e107be36 | 1771 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1772 | prepare_lock_switch(rq, next); |
1773 | prepare_arch_switch(next); | |
1774 | } | |
1775 | ||
1da177e4 LT |
1776 | /** |
1777 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1778 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1779 | * @prev: the thread we just switched away from. |
1780 | * | |
4866cde0 NP |
1781 | * finish_task_switch must be called after the context switch, paired |
1782 | * with a prepare_task_switch call before the context switch. | |
1783 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1784 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1785 | * |
1786 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1787 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1788 | * with the lock held can cause deadlocks; see schedule() for | |
1789 | * details.) | |
1790 | */ | |
a9957449 | 1791 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1792 | __releases(rq->lock) |
1793 | { | |
1da177e4 | 1794 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1795 | long prev_state; |
1da177e4 LT |
1796 | |
1797 | rq->prev_mm = NULL; | |
1798 | ||
1799 | /* | |
1800 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1801 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1802 | * schedule one last time. The schedule call will never return, and |
1803 | * the scheduled task must drop that reference. | |
c394cc9f | 1804 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1805 | * still held, otherwise prev could be scheduled on another cpu, die |
1806 | * there before we look at prev->state, and then the reference would | |
1807 | * be dropped twice. | |
1808 | * Manfred Spraul <manfred@colorfullife.com> | |
1809 | */ | |
55a101f8 | 1810 | prev_state = prev->state; |
4866cde0 NP |
1811 | finish_arch_switch(prev); |
1812 | finish_lock_switch(rq, prev); | |
e107be36 | 1813 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1814 | if (mm) |
1815 | mmdrop(mm); | |
c394cc9f | 1816 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1817 | /* |
1818 | * Remove function-return probe instances associated with this | |
1819 | * task and put them back on the free list. | |
9761eea8 | 1820 | */ |
c6fd91f0 | 1821 | kprobe_flush_task(prev); |
1da177e4 | 1822 | put_task_struct(prev); |
c6fd91f0 | 1823 | } |
1da177e4 LT |
1824 | } |
1825 | ||
1826 | /** | |
1827 | * schedule_tail - first thing a freshly forked thread must call. | |
1828 | * @prev: the thread we just switched away from. | |
1829 | */ | |
36c8b586 | 1830 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1831 | __releases(rq->lock) |
1832 | { | |
70b97a7f IM |
1833 | struct rq *rq = this_rq(); |
1834 | ||
4866cde0 NP |
1835 | finish_task_switch(rq, prev); |
1836 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1837 | /* In this case, finish_task_switch does not reenable preemption */ | |
1838 | preempt_enable(); | |
1839 | #endif | |
1da177e4 LT |
1840 | if (current->set_child_tid) |
1841 | put_user(current->pid, current->set_child_tid); | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * context_switch - switch to the new MM and the new | |
1846 | * thread's register state. | |
1847 | */ | |
dd41f596 | 1848 | static inline void |
70b97a7f | 1849 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1850 | struct task_struct *next) |
1da177e4 | 1851 | { |
dd41f596 | 1852 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1853 | |
e107be36 | 1854 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1855 | mm = next->mm; |
1856 | oldmm = prev->active_mm; | |
9226d125 ZA |
1857 | /* |
1858 | * For paravirt, this is coupled with an exit in switch_to to | |
1859 | * combine the page table reload and the switch backend into | |
1860 | * one hypercall. | |
1861 | */ | |
1862 | arch_enter_lazy_cpu_mode(); | |
1863 | ||
dd41f596 | 1864 | if (unlikely(!mm)) { |
1da177e4 LT |
1865 | next->active_mm = oldmm; |
1866 | atomic_inc(&oldmm->mm_count); | |
1867 | enter_lazy_tlb(oldmm, next); | |
1868 | } else | |
1869 | switch_mm(oldmm, mm, next); | |
1870 | ||
dd41f596 | 1871 | if (unlikely(!prev->mm)) { |
1da177e4 | 1872 | prev->active_mm = NULL; |
1da177e4 LT |
1873 | rq->prev_mm = oldmm; |
1874 | } | |
3a5f5e48 IM |
1875 | /* |
1876 | * Since the runqueue lock will be released by the next | |
1877 | * task (which is an invalid locking op but in the case | |
1878 | * of the scheduler it's an obvious special-case), so we | |
1879 | * do an early lockdep release here: | |
1880 | */ | |
1881 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1882 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1883 | #endif |
1da177e4 LT |
1884 | |
1885 | /* Here we just switch the register state and the stack. */ | |
1886 | switch_to(prev, next, prev); | |
1887 | ||
dd41f596 IM |
1888 | barrier(); |
1889 | /* | |
1890 | * this_rq must be evaluated again because prev may have moved | |
1891 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1892 | * frame will be invalid. | |
1893 | */ | |
1894 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1895 | } |
1896 | ||
1897 | /* | |
1898 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1899 | * | |
1900 | * externally visible scheduler statistics: current number of runnable | |
1901 | * threads, current number of uninterruptible-sleeping threads, total | |
1902 | * number of context switches performed since bootup. | |
1903 | */ | |
1904 | unsigned long nr_running(void) | |
1905 | { | |
1906 | unsigned long i, sum = 0; | |
1907 | ||
1908 | for_each_online_cpu(i) | |
1909 | sum += cpu_rq(i)->nr_running; | |
1910 | ||
1911 | return sum; | |
1912 | } | |
1913 | ||
1914 | unsigned long nr_uninterruptible(void) | |
1915 | { | |
1916 | unsigned long i, sum = 0; | |
1917 | ||
0a945022 | 1918 | for_each_possible_cpu(i) |
1da177e4 LT |
1919 | sum += cpu_rq(i)->nr_uninterruptible; |
1920 | ||
1921 | /* | |
1922 | * Since we read the counters lockless, it might be slightly | |
1923 | * inaccurate. Do not allow it to go below zero though: | |
1924 | */ | |
1925 | if (unlikely((long)sum < 0)) | |
1926 | sum = 0; | |
1927 | ||
1928 | return sum; | |
1929 | } | |
1930 | ||
1931 | unsigned long long nr_context_switches(void) | |
1932 | { | |
cc94abfc SR |
1933 | int i; |
1934 | unsigned long long sum = 0; | |
1da177e4 | 1935 | |
0a945022 | 1936 | for_each_possible_cpu(i) |
1da177e4 LT |
1937 | sum += cpu_rq(i)->nr_switches; |
1938 | ||
1939 | return sum; | |
1940 | } | |
1941 | ||
1942 | unsigned long nr_iowait(void) | |
1943 | { | |
1944 | unsigned long i, sum = 0; | |
1945 | ||
0a945022 | 1946 | for_each_possible_cpu(i) |
1da177e4 LT |
1947 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1948 | ||
1949 | return sum; | |
1950 | } | |
1951 | ||
db1b1fef JS |
1952 | unsigned long nr_active(void) |
1953 | { | |
1954 | unsigned long i, running = 0, uninterruptible = 0; | |
1955 | ||
1956 | for_each_online_cpu(i) { | |
1957 | running += cpu_rq(i)->nr_running; | |
1958 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1959 | } | |
1960 | ||
1961 | if (unlikely((long)uninterruptible < 0)) | |
1962 | uninterruptible = 0; | |
1963 | ||
1964 | return running + uninterruptible; | |
1965 | } | |
1966 | ||
48f24c4d | 1967 | /* |
dd41f596 IM |
1968 | * Update rq->cpu_load[] statistics. This function is usually called every |
1969 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 1970 | */ |
dd41f596 | 1971 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 1972 | { |
495eca49 | 1973 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
1974 | int i, scale; |
1975 | ||
1976 | this_rq->nr_load_updates++; | |
dd41f596 IM |
1977 | |
1978 | /* Update our load: */ | |
1979 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
1980 | unsigned long old_load, new_load; | |
1981 | ||
1982 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
1983 | ||
1984 | old_load = this_rq->cpu_load[i]; | |
1985 | new_load = this_load; | |
a25707f3 IM |
1986 | /* |
1987 | * Round up the averaging division if load is increasing. This | |
1988 | * prevents us from getting stuck on 9 if the load is 10, for | |
1989 | * example. | |
1990 | */ | |
1991 | if (new_load > old_load) | |
1992 | new_load += scale-1; | |
dd41f596 IM |
1993 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
1994 | } | |
48f24c4d IM |
1995 | } |
1996 | ||
dd41f596 IM |
1997 | #ifdef CONFIG_SMP |
1998 | ||
1da177e4 LT |
1999 | /* |
2000 | * double_rq_lock - safely lock two runqueues | |
2001 | * | |
2002 | * Note this does not disable interrupts like task_rq_lock, | |
2003 | * you need to do so manually before calling. | |
2004 | */ | |
70b97a7f | 2005 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2006 | __acquires(rq1->lock) |
2007 | __acquires(rq2->lock) | |
2008 | { | |
054b9108 | 2009 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2010 | if (rq1 == rq2) { |
2011 | spin_lock(&rq1->lock); | |
2012 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2013 | } else { | |
c96d145e | 2014 | if (rq1 < rq2) { |
1da177e4 LT |
2015 | spin_lock(&rq1->lock); |
2016 | spin_lock(&rq2->lock); | |
2017 | } else { | |
2018 | spin_lock(&rq2->lock); | |
2019 | spin_lock(&rq1->lock); | |
2020 | } | |
2021 | } | |
6e82a3be IM |
2022 | update_rq_clock(rq1); |
2023 | update_rq_clock(rq2); | |
1da177e4 LT |
2024 | } |
2025 | ||
2026 | /* | |
2027 | * double_rq_unlock - safely unlock two runqueues | |
2028 | * | |
2029 | * Note this does not restore interrupts like task_rq_unlock, | |
2030 | * you need to do so manually after calling. | |
2031 | */ | |
70b97a7f | 2032 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2033 | __releases(rq1->lock) |
2034 | __releases(rq2->lock) | |
2035 | { | |
2036 | spin_unlock(&rq1->lock); | |
2037 | if (rq1 != rq2) | |
2038 | spin_unlock(&rq2->lock); | |
2039 | else | |
2040 | __release(rq2->lock); | |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2045 | */ | |
70b97a7f | 2046 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2047 | __releases(this_rq->lock) |
2048 | __acquires(busiest->lock) | |
2049 | __acquires(this_rq->lock) | |
2050 | { | |
054b9108 KK |
2051 | if (unlikely(!irqs_disabled())) { |
2052 | /* printk() doesn't work good under rq->lock */ | |
2053 | spin_unlock(&this_rq->lock); | |
2054 | BUG_ON(1); | |
2055 | } | |
1da177e4 | 2056 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2057 | if (busiest < this_rq) { |
1da177e4 LT |
2058 | spin_unlock(&this_rq->lock); |
2059 | spin_lock(&busiest->lock); | |
2060 | spin_lock(&this_rq->lock); | |
2061 | } else | |
2062 | spin_lock(&busiest->lock); | |
2063 | } | |
2064 | } | |
2065 | ||
1da177e4 LT |
2066 | /* |
2067 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2068 | * This is accomplished by forcing the cpu_allowed mask to only | |
2069 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2070 | * the cpu_allowed mask is restored. | |
2071 | */ | |
36c8b586 | 2072 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2073 | { |
70b97a7f | 2074 | struct migration_req req; |
1da177e4 | 2075 | unsigned long flags; |
70b97a7f | 2076 | struct rq *rq; |
1da177e4 LT |
2077 | |
2078 | rq = task_rq_lock(p, &flags); | |
2079 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2080 | || unlikely(cpu_is_offline(dest_cpu))) | |
2081 | goto out; | |
2082 | ||
2083 | /* force the process onto the specified CPU */ | |
2084 | if (migrate_task(p, dest_cpu, &req)) { | |
2085 | /* Need to wait for migration thread (might exit: take ref). */ | |
2086 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2087 | |
1da177e4 LT |
2088 | get_task_struct(mt); |
2089 | task_rq_unlock(rq, &flags); | |
2090 | wake_up_process(mt); | |
2091 | put_task_struct(mt); | |
2092 | wait_for_completion(&req.done); | |
36c8b586 | 2093 | |
1da177e4 LT |
2094 | return; |
2095 | } | |
2096 | out: | |
2097 | task_rq_unlock(rq, &flags); | |
2098 | } | |
2099 | ||
2100 | /* | |
476d139c NP |
2101 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2102 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2103 | */ |
2104 | void sched_exec(void) | |
2105 | { | |
1da177e4 | 2106 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2107 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2108 | put_cpu(); |
476d139c NP |
2109 | if (new_cpu != this_cpu) |
2110 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2111 | } |
2112 | ||
2113 | /* | |
2114 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2115 | * Both runqueues must be locked. | |
2116 | */ | |
dd41f596 IM |
2117 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2118 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2119 | { |
2e1cb74a | 2120 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2121 | set_task_cpu(p, this_cpu); |
dd41f596 | 2122 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2123 | /* |
2124 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2125 | * to be always true for them. | |
2126 | */ | |
dd41f596 | 2127 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2128 | } |
2129 | ||
2130 | /* | |
2131 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2132 | */ | |
858119e1 | 2133 | static |
70b97a7f | 2134 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2135 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2136 | int *all_pinned) |
1da177e4 LT |
2137 | { |
2138 | /* | |
2139 | * We do not migrate tasks that are: | |
2140 | * 1) running (obviously), or | |
2141 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2142 | * 3) are cache-hot on their current CPU. | |
2143 | */ | |
1da177e4 LT |
2144 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
2145 | return 0; | |
81026794 NP |
2146 | *all_pinned = 0; |
2147 | ||
2148 | if (task_running(rq, p)) | |
2149 | return 0; | |
1da177e4 | 2150 | |
1da177e4 LT |
2151 | return 1; |
2152 | } | |
2153 | ||
dd41f596 | 2154 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2155 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2156 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2157 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2158 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2159 | { |
dd41f596 IM |
2160 | int pulled = 0, pinned = 0, skip_for_load; |
2161 | struct task_struct *p; | |
2162 | long rem_load_move = max_load_move; | |
1da177e4 | 2163 | |
2dd73a4f | 2164 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2165 | goto out; |
2166 | ||
81026794 NP |
2167 | pinned = 1; |
2168 | ||
1da177e4 | 2169 | /* |
dd41f596 | 2170 | * Start the load-balancing iterator: |
1da177e4 | 2171 | */ |
dd41f596 IM |
2172 | p = iterator->start(iterator->arg); |
2173 | next: | |
2174 | if (!p) | |
1da177e4 | 2175 | goto out; |
50ddd969 PW |
2176 | /* |
2177 | * To help distribute high priority tasks accross CPUs we don't | |
2178 | * skip a task if it will be the highest priority task (i.e. smallest | |
2179 | * prio value) on its new queue regardless of its load weight | |
2180 | */ | |
dd41f596 IM |
2181 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2182 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2183 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2184 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2185 | p = iterator->next(iterator->arg); |
2186 | goto next; | |
1da177e4 LT |
2187 | } |
2188 | ||
dd41f596 | 2189 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2190 | pulled++; |
dd41f596 | 2191 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2192 | |
2dd73a4f PW |
2193 | /* |
2194 | * We only want to steal up to the prescribed number of tasks | |
2195 | * and the prescribed amount of weighted load. | |
2196 | */ | |
2197 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2198 | if (p->prio < *this_best_prio) |
2199 | *this_best_prio = p->prio; | |
dd41f596 IM |
2200 | p = iterator->next(iterator->arg); |
2201 | goto next; | |
1da177e4 LT |
2202 | } |
2203 | out: | |
2204 | /* | |
2205 | * Right now, this is the only place pull_task() is called, | |
2206 | * so we can safely collect pull_task() stats here rather than | |
2207 | * inside pull_task(). | |
2208 | */ | |
2209 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2210 | |
2211 | if (all_pinned) | |
2212 | *all_pinned = pinned; | |
dd41f596 | 2213 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2214 | return pulled; |
2215 | } | |
2216 | ||
dd41f596 | 2217 | /* |
43010659 PW |
2218 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2219 | * this_rq, as part of a balancing operation within domain "sd". | |
2220 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2221 | * |
2222 | * Called with both runqueues locked. | |
2223 | */ | |
2224 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2225 | unsigned long max_load_move, |
dd41f596 IM |
2226 | struct sched_domain *sd, enum cpu_idle_type idle, |
2227 | int *all_pinned) | |
2228 | { | |
5522d5d5 | 2229 | const struct sched_class *class = sched_class_highest; |
43010659 | 2230 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2231 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2232 | |
2233 | do { | |
43010659 PW |
2234 | total_load_moved += |
2235 | class->load_balance(this_rq, this_cpu, busiest, | |
2236 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2237 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2238 | class = class->next; |
43010659 | 2239 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2240 | |
43010659 PW |
2241 | return total_load_moved > 0; |
2242 | } | |
2243 | ||
2244 | /* | |
2245 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2246 | * part of active balancing operations within "domain". | |
2247 | * Returns 1 if successful and 0 otherwise. | |
2248 | * | |
2249 | * Called with both runqueues locked. | |
2250 | */ | |
2251 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2252 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2253 | { | |
5522d5d5 | 2254 | const struct sched_class *class; |
a4ac01c3 | 2255 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2256 | |
2257 | for (class = sched_class_highest; class; class = class->next) | |
2258 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2259 | 1, ULONG_MAX, sd, idle, NULL, |
2260 | &this_best_prio)) | |
43010659 PW |
2261 | return 1; |
2262 | ||
2263 | return 0; | |
dd41f596 IM |
2264 | } |
2265 | ||
1da177e4 LT |
2266 | /* |
2267 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2268 | * domain. It calculates and returns the amount of weighted load which |
2269 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2270 | */ |
2271 | static struct sched_group * | |
2272 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2273 | unsigned long *imbalance, enum cpu_idle_type idle, |
2274 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2275 | { |
2276 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2277 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2278 | unsigned long max_pull; |
2dd73a4f PW |
2279 | unsigned long busiest_load_per_task, busiest_nr_running; |
2280 | unsigned long this_load_per_task, this_nr_running; | |
7897986b | 2281 | int load_idx; |
5c45bf27 SS |
2282 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2283 | int power_savings_balance = 1; | |
2284 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2285 | unsigned long min_nr_running = ULONG_MAX; | |
2286 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2287 | #endif | |
1da177e4 LT |
2288 | |
2289 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2290 | busiest_load_per_task = busiest_nr_running = 0; |
2291 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2292 | if (idle == CPU_NOT_IDLE) |
7897986b | 2293 | load_idx = sd->busy_idx; |
d15bcfdb | 2294 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2295 | load_idx = sd->newidle_idx; |
2296 | else | |
2297 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2298 | |
2299 | do { | |
5c45bf27 | 2300 | unsigned long load, group_capacity; |
1da177e4 LT |
2301 | int local_group; |
2302 | int i; | |
783609c6 | 2303 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2304 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2305 | |
2306 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2307 | ||
783609c6 SS |
2308 | if (local_group) |
2309 | balance_cpu = first_cpu(group->cpumask); | |
2310 | ||
1da177e4 | 2311 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2312 | sum_weighted_load = sum_nr_running = avg_load = 0; |
1da177e4 LT |
2313 | |
2314 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2315 | struct rq *rq; |
2316 | ||
2317 | if (!cpu_isset(i, *cpus)) | |
2318 | continue; | |
2319 | ||
2320 | rq = cpu_rq(i); | |
2dd73a4f | 2321 | |
9439aab8 | 2322 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2323 | *sd_idle = 0; |
2324 | ||
1da177e4 | 2325 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2326 | if (local_group) { |
2327 | if (idle_cpu(i) && !first_idle_cpu) { | |
2328 | first_idle_cpu = 1; | |
2329 | balance_cpu = i; | |
2330 | } | |
2331 | ||
a2000572 | 2332 | load = target_load(i, load_idx); |
783609c6 | 2333 | } else |
a2000572 | 2334 | load = source_load(i, load_idx); |
1da177e4 LT |
2335 | |
2336 | avg_load += load; | |
2dd73a4f | 2337 | sum_nr_running += rq->nr_running; |
dd41f596 | 2338 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2339 | } |
2340 | ||
783609c6 SS |
2341 | /* |
2342 | * First idle cpu or the first cpu(busiest) in this sched group | |
2343 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2344 | * domains. In the newly idle case, we will allow all the cpu's |
2345 | * to do the newly idle load balance. | |
783609c6 | 2346 | */ |
9439aab8 SS |
2347 | if (idle != CPU_NEWLY_IDLE && local_group && |
2348 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2349 | *balance = 0; |
2350 | goto ret; | |
2351 | } | |
2352 | ||
1da177e4 | 2353 | total_load += avg_load; |
5517d86b | 2354 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2355 | |
2356 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2357 | avg_load = sg_div_cpu_power(group, |
2358 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2359 | |
5517d86b | 2360 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2361 | |
1da177e4 LT |
2362 | if (local_group) { |
2363 | this_load = avg_load; | |
2364 | this = group; | |
2dd73a4f PW |
2365 | this_nr_running = sum_nr_running; |
2366 | this_load_per_task = sum_weighted_load; | |
2367 | } else if (avg_load > max_load && | |
5c45bf27 | 2368 | sum_nr_running > group_capacity) { |
1da177e4 LT |
2369 | max_load = avg_load; |
2370 | busiest = group; | |
2dd73a4f PW |
2371 | busiest_nr_running = sum_nr_running; |
2372 | busiest_load_per_task = sum_weighted_load; | |
1da177e4 | 2373 | } |
5c45bf27 SS |
2374 | |
2375 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2376 | /* | |
2377 | * Busy processors will not participate in power savings | |
2378 | * balance. | |
2379 | */ | |
dd41f596 IM |
2380 | if (idle == CPU_NOT_IDLE || |
2381 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2382 | goto group_next; | |
5c45bf27 SS |
2383 | |
2384 | /* | |
2385 | * If the local group is idle or completely loaded | |
2386 | * no need to do power savings balance at this domain | |
2387 | */ | |
2388 | if (local_group && (this_nr_running >= group_capacity || | |
2389 | !this_nr_running)) | |
2390 | power_savings_balance = 0; | |
2391 | ||
dd41f596 | 2392 | /* |
5c45bf27 SS |
2393 | * If a group is already running at full capacity or idle, |
2394 | * don't include that group in power savings calculations | |
dd41f596 IM |
2395 | */ |
2396 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2397 | || !sum_nr_running) |
dd41f596 | 2398 | goto group_next; |
5c45bf27 | 2399 | |
dd41f596 | 2400 | /* |
5c45bf27 | 2401 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2402 | * This is the group from where we need to pick up the load |
2403 | * for saving power | |
2404 | */ | |
2405 | if ((sum_nr_running < min_nr_running) || | |
2406 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2407 | first_cpu(group->cpumask) < |
2408 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2409 | group_min = group; |
2410 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2411 | min_load_per_task = sum_weighted_load / |
2412 | sum_nr_running; | |
dd41f596 | 2413 | } |
5c45bf27 | 2414 | |
dd41f596 | 2415 | /* |
5c45bf27 | 2416 | * Calculate the group which is almost near its |
dd41f596 IM |
2417 | * capacity but still has some space to pick up some load |
2418 | * from other group and save more power | |
2419 | */ | |
2420 | if (sum_nr_running <= group_capacity - 1) { | |
2421 | if (sum_nr_running > leader_nr_running || | |
2422 | (sum_nr_running == leader_nr_running && | |
2423 | first_cpu(group->cpumask) > | |
2424 | first_cpu(group_leader->cpumask))) { | |
2425 | group_leader = group; | |
2426 | leader_nr_running = sum_nr_running; | |
2427 | } | |
48f24c4d | 2428 | } |
5c45bf27 SS |
2429 | group_next: |
2430 | #endif | |
1da177e4 LT |
2431 | group = group->next; |
2432 | } while (group != sd->groups); | |
2433 | ||
2dd73a4f | 2434 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2435 | goto out_balanced; |
2436 | ||
2437 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2438 | ||
2439 | if (this_load >= avg_load || | |
2440 | 100*max_load <= sd->imbalance_pct*this_load) | |
2441 | goto out_balanced; | |
2442 | ||
2dd73a4f | 2443 | busiest_load_per_task /= busiest_nr_running; |
1da177e4 LT |
2444 | /* |
2445 | * We're trying to get all the cpus to the average_load, so we don't | |
2446 | * want to push ourselves above the average load, nor do we wish to | |
2447 | * reduce the max loaded cpu below the average load, as either of these | |
2448 | * actions would just result in more rebalancing later, and ping-pong | |
2449 | * tasks around. Thus we look for the minimum possible imbalance. | |
2450 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2451 | * be counted as no imbalance for these purposes -- we can't fix that | |
2452 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2453 | * appear as very large values with unsigned longs. | |
2454 | */ | |
2dd73a4f PW |
2455 | if (max_load <= busiest_load_per_task) |
2456 | goto out_balanced; | |
2457 | ||
2458 | /* | |
2459 | * In the presence of smp nice balancing, certain scenarios can have | |
2460 | * max load less than avg load(as we skip the groups at or below | |
2461 | * its cpu_power, while calculating max_load..) | |
2462 | */ | |
2463 | if (max_load < avg_load) { | |
2464 | *imbalance = 0; | |
2465 | goto small_imbalance; | |
2466 | } | |
0c117f1b SS |
2467 | |
2468 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2469 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2470 | |
1da177e4 | 2471 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2472 | *imbalance = min(max_pull * busiest->__cpu_power, |
2473 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2474 | / SCHED_LOAD_SCALE; |
2475 | ||
2dd73a4f PW |
2476 | /* |
2477 | * if *imbalance is less than the average load per runnable task | |
2478 | * there is no gaurantee that any tasks will be moved so we'll have | |
2479 | * a think about bumping its value to force at least one task to be | |
2480 | * moved | |
2481 | */ | |
7fd0d2dd | 2482 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2483 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2484 | unsigned int imbn; |
2485 | ||
2486 | small_imbalance: | |
2487 | pwr_move = pwr_now = 0; | |
2488 | imbn = 2; | |
2489 | if (this_nr_running) { | |
2490 | this_load_per_task /= this_nr_running; | |
2491 | if (busiest_load_per_task > this_load_per_task) | |
2492 | imbn = 1; | |
2493 | } else | |
2494 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2495 | |
dd41f596 IM |
2496 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2497 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2498 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2499 | return busiest; |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * OK, we don't have enough imbalance to justify moving tasks, | |
2504 | * however we may be able to increase total CPU power used by | |
2505 | * moving them. | |
2506 | */ | |
2507 | ||
5517d86b ED |
2508 | pwr_now += busiest->__cpu_power * |
2509 | min(busiest_load_per_task, max_load); | |
2510 | pwr_now += this->__cpu_power * | |
2511 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2512 | pwr_now /= SCHED_LOAD_SCALE; |
2513 | ||
2514 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2515 | tmp = sg_div_cpu_power(busiest, |
2516 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2517 | if (max_load > tmp) |
5517d86b | 2518 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2519 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2520 | |
2521 | /* Amount of load we'd add */ | |
5517d86b | 2522 | if (max_load * busiest->__cpu_power < |
33859f7f | 2523 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2524 | tmp = sg_div_cpu_power(this, |
2525 | max_load * busiest->__cpu_power); | |
1da177e4 | 2526 | else |
5517d86b ED |
2527 | tmp = sg_div_cpu_power(this, |
2528 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2529 | pwr_move += this->__cpu_power * | |
2530 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2531 | pwr_move /= SCHED_LOAD_SCALE; |
2532 | ||
2533 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2534 | if (pwr_move > pwr_now) |
2535 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2536 | } |
2537 | ||
1da177e4 LT |
2538 | return busiest; |
2539 | ||
2540 | out_balanced: | |
5c45bf27 | 2541 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2542 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2543 | goto ret; |
1da177e4 | 2544 | |
5c45bf27 SS |
2545 | if (this == group_leader && group_leader != group_min) { |
2546 | *imbalance = min_load_per_task; | |
2547 | return group_min; | |
2548 | } | |
5c45bf27 | 2549 | #endif |
783609c6 | 2550 | ret: |
1da177e4 LT |
2551 | *imbalance = 0; |
2552 | return NULL; | |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2557 | */ | |
70b97a7f | 2558 | static struct rq * |
d15bcfdb | 2559 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2560 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2561 | { |
70b97a7f | 2562 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2563 | unsigned long max_load = 0; |
1da177e4 LT |
2564 | int i; |
2565 | ||
2566 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2567 | unsigned long wl; |
0a2966b4 CL |
2568 | |
2569 | if (!cpu_isset(i, *cpus)) | |
2570 | continue; | |
2571 | ||
48f24c4d | 2572 | rq = cpu_rq(i); |
dd41f596 | 2573 | wl = weighted_cpuload(i); |
2dd73a4f | 2574 | |
dd41f596 | 2575 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2576 | continue; |
1da177e4 | 2577 | |
dd41f596 IM |
2578 | if (wl > max_load) { |
2579 | max_load = wl; | |
48f24c4d | 2580 | busiest = rq; |
1da177e4 LT |
2581 | } |
2582 | } | |
2583 | ||
2584 | return busiest; | |
2585 | } | |
2586 | ||
77391d71 NP |
2587 | /* |
2588 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2589 | * so long as it is large enough. | |
2590 | */ | |
2591 | #define MAX_PINNED_INTERVAL 512 | |
2592 | ||
1da177e4 LT |
2593 | /* |
2594 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2595 | * tasks if there is an imbalance. | |
1da177e4 | 2596 | */ |
70b97a7f | 2597 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2598 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2599 | int *balance) |
1da177e4 | 2600 | { |
43010659 | 2601 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2602 | struct sched_group *group; |
1da177e4 | 2603 | unsigned long imbalance; |
70b97a7f | 2604 | struct rq *busiest; |
0a2966b4 | 2605 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2606 | unsigned long flags; |
5969fe06 | 2607 | |
89c4710e SS |
2608 | /* |
2609 | * When power savings policy is enabled for the parent domain, idle | |
2610 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2611 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2612 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2613 | */ |
d15bcfdb | 2614 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2615 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2616 | sd_idle = 1; |
1da177e4 | 2617 | |
2d72376b | 2618 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2619 | |
0a2966b4 CL |
2620 | redo: |
2621 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2622 | &cpus, balance); |
2623 | ||
06066714 | 2624 | if (*balance == 0) |
783609c6 | 2625 | goto out_balanced; |
783609c6 | 2626 | |
1da177e4 LT |
2627 | if (!group) { |
2628 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2629 | goto out_balanced; | |
2630 | } | |
2631 | ||
0a2966b4 | 2632 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2633 | if (!busiest) { |
2634 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2635 | goto out_balanced; | |
2636 | } | |
2637 | ||
db935dbd | 2638 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2639 | |
2640 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2641 | ||
43010659 | 2642 | ld_moved = 0; |
1da177e4 LT |
2643 | if (busiest->nr_running > 1) { |
2644 | /* | |
2645 | * Attempt to move tasks. If find_busiest_group has found | |
2646 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2647 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2648 | * correctly treated as an imbalance. |
2649 | */ | |
fe2eea3f | 2650 | local_irq_save(flags); |
e17224bf | 2651 | double_rq_lock(this_rq, busiest); |
43010659 | 2652 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2653 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2654 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2655 | local_irq_restore(flags); |
81026794 | 2656 | |
46cb4b7c SS |
2657 | /* |
2658 | * some other cpu did the load balance for us. | |
2659 | */ | |
43010659 | 2660 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2661 | resched_cpu(this_cpu); |
2662 | ||
81026794 | 2663 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2664 | if (unlikely(all_pinned)) { |
2665 | cpu_clear(cpu_of(busiest), cpus); | |
2666 | if (!cpus_empty(cpus)) | |
2667 | goto redo; | |
81026794 | 2668 | goto out_balanced; |
0a2966b4 | 2669 | } |
1da177e4 | 2670 | } |
81026794 | 2671 | |
43010659 | 2672 | if (!ld_moved) { |
1da177e4 LT |
2673 | schedstat_inc(sd, lb_failed[idle]); |
2674 | sd->nr_balance_failed++; | |
2675 | ||
2676 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2677 | |
fe2eea3f | 2678 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2679 | |
2680 | /* don't kick the migration_thread, if the curr | |
2681 | * task on busiest cpu can't be moved to this_cpu | |
2682 | */ | |
2683 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2684 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2685 | all_pinned = 1; |
2686 | goto out_one_pinned; | |
2687 | } | |
2688 | ||
1da177e4 LT |
2689 | if (!busiest->active_balance) { |
2690 | busiest->active_balance = 1; | |
2691 | busiest->push_cpu = this_cpu; | |
81026794 | 2692 | active_balance = 1; |
1da177e4 | 2693 | } |
fe2eea3f | 2694 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2695 | if (active_balance) |
1da177e4 LT |
2696 | wake_up_process(busiest->migration_thread); |
2697 | ||
2698 | /* | |
2699 | * We've kicked active balancing, reset the failure | |
2700 | * counter. | |
2701 | */ | |
39507451 | 2702 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2703 | } |
81026794 | 2704 | } else |
1da177e4 LT |
2705 | sd->nr_balance_failed = 0; |
2706 | ||
81026794 | 2707 | if (likely(!active_balance)) { |
1da177e4 LT |
2708 | /* We were unbalanced, so reset the balancing interval */ |
2709 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2710 | } else { |
2711 | /* | |
2712 | * If we've begun active balancing, start to back off. This | |
2713 | * case may not be covered by the all_pinned logic if there | |
2714 | * is only 1 task on the busy runqueue (because we don't call | |
2715 | * move_tasks). | |
2716 | */ | |
2717 | if (sd->balance_interval < sd->max_interval) | |
2718 | sd->balance_interval *= 2; | |
1da177e4 LT |
2719 | } |
2720 | ||
43010659 | 2721 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2722 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2723 | return -1; |
43010659 | 2724 | return ld_moved; |
1da177e4 LT |
2725 | |
2726 | out_balanced: | |
1da177e4 LT |
2727 | schedstat_inc(sd, lb_balanced[idle]); |
2728 | ||
16cfb1c0 | 2729 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2730 | |
2731 | out_one_pinned: | |
1da177e4 | 2732 | /* tune up the balancing interval */ |
77391d71 NP |
2733 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2734 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2735 | sd->balance_interval *= 2; |
2736 | ||
48f24c4d | 2737 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2738 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2739 | return -1; |
1da177e4 LT |
2740 | return 0; |
2741 | } | |
2742 | ||
2743 | /* | |
2744 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2745 | * tasks if there is an imbalance. | |
2746 | * | |
d15bcfdb | 2747 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2748 | * this_rq is locked. |
2749 | */ | |
48f24c4d | 2750 | static int |
70b97a7f | 2751 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2752 | { |
2753 | struct sched_group *group; | |
70b97a7f | 2754 | struct rq *busiest = NULL; |
1da177e4 | 2755 | unsigned long imbalance; |
43010659 | 2756 | int ld_moved = 0; |
5969fe06 | 2757 | int sd_idle = 0; |
969bb4e4 | 2758 | int all_pinned = 0; |
0a2966b4 | 2759 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2760 | |
89c4710e SS |
2761 | /* |
2762 | * When power savings policy is enabled for the parent domain, idle | |
2763 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2764 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2765 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2766 | */ |
2767 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2768 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2769 | sd_idle = 1; |
1da177e4 | 2770 | |
2d72376b | 2771 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2772 | redo: |
d15bcfdb | 2773 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2774 | &sd_idle, &cpus, NULL); |
1da177e4 | 2775 | if (!group) { |
d15bcfdb | 2776 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2777 | goto out_balanced; |
1da177e4 LT |
2778 | } |
2779 | ||
d15bcfdb | 2780 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2781 | &cpus); |
db935dbd | 2782 | if (!busiest) { |
d15bcfdb | 2783 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2784 | goto out_balanced; |
1da177e4 LT |
2785 | } |
2786 | ||
db935dbd NP |
2787 | BUG_ON(busiest == this_rq); |
2788 | ||
d15bcfdb | 2789 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2790 | |
43010659 | 2791 | ld_moved = 0; |
d6d5cfaf NP |
2792 | if (busiest->nr_running > 1) { |
2793 | /* Attempt to move tasks */ | |
2794 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2795 | /* this_rq->clock is already updated */ |
2796 | update_rq_clock(busiest); | |
43010659 | 2797 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2798 | imbalance, sd, CPU_NEWLY_IDLE, |
2799 | &all_pinned); | |
d6d5cfaf | 2800 | spin_unlock(&busiest->lock); |
0a2966b4 | 2801 | |
969bb4e4 | 2802 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2803 | cpu_clear(cpu_of(busiest), cpus); |
2804 | if (!cpus_empty(cpus)) | |
2805 | goto redo; | |
2806 | } | |
d6d5cfaf NP |
2807 | } |
2808 | ||
43010659 | 2809 | if (!ld_moved) { |
d15bcfdb | 2810 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2811 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2812 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2813 | return -1; |
2814 | } else | |
16cfb1c0 | 2815 | sd->nr_balance_failed = 0; |
1da177e4 | 2816 | |
43010659 | 2817 | return ld_moved; |
16cfb1c0 NP |
2818 | |
2819 | out_balanced: | |
d15bcfdb | 2820 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2821 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2822 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2823 | return -1; |
16cfb1c0 | 2824 | sd->nr_balance_failed = 0; |
48f24c4d | 2825 | |
16cfb1c0 | 2826 | return 0; |
1da177e4 LT |
2827 | } |
2828 | ||
2829 | /* | |
2830 | * idle_balance is called by schedule() if this_cpu is about to become | |
2831 | * idle. Attempts to pull tasks from other CPUs. | |
2832 | */ | |
70b97a7f | 2833 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2834 | { |
2835 | struct sched_domain *sd; | |
dd41f596 IM |
2836 | int pulled_task = -1; |
2837 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2838 | |
2839 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2840 | unsigned long interval; |
2841 | ||
2842 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2843 | continue; | |
2844 | ||
2845 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2846 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2847 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2848 | this_rq, sd); |
2849 | ||
2850 | interval = msecs_to_jiffies(sd->balance_interval); | |
2851 | if (time_after(next_balance, sd->last_balance + interval)) | |
2852 | next_balance = sd->last_balance + interval; | |
2853 | if (pulled_task) | |
2854 | break; | |
1da177e4 | 2855 | } |
dd41f596 | 2856 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2857 | /* |
2858 | * We are going idle. next_balance may be set based on | |
2859 | * a busy processor. So reset next_balance. | |
2860 | */ | |
2861 | this_rq->next_balance = next_balance; | |
dd41f596 | 2862 | } |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /* | |
2866 | * active_load_balance is run by migration threads. It pushes running tasks | |
2867 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2868 | * running on each physical CPU where possible, and avoids physical / | |
2869 | * logical imbalances. | |
2870 | * | |
2871 | * Called with busiest_rq locked. | |
2872 | */ | |
70b97a7f | 2873 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2874 | { |
39507451 | 2875 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2876 | struct sched_domain *sd; |
2877 | struct rq *target_rq; | |
39507451 | 2878 | |
48f24c4d | 2879 | /* Is there any task to move? */ |
39507451 | 2880 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2881 | return; |
2882 | ||
2883 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2884 | |
2885 | /* | |
39507451 NP |
2886 | * This condition is "impossible", if it occurs |
2887 | * we need to fix it. Originally reported by | |
2888 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2889 | */ |
39507451 | 2890 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2891 | |
39507451 NP |
2892 | /* move a task from busiest_rq to target_rq */ |
2893 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2894 | update_rq_clock(busiest_rq); |
2895 | update_rq_clock(target_rq); | |
39507451 NP |
2896 | |
2897 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2898 | for_each_domain(target_cpu, sd) { |
39507451 | 2899 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2900 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2901 | break; |
c96d145e | 2902 | } |
39507451 | 2903 | |
48f24c4d | 2904 | if (likely(sd)) { |
2d72376b | 2905 | schedstat_inc(sd, alb_count); |
39507451 | 2906 | |
43010659 PW |
2907 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2908 | sd, CPU_IDLE)) | |
48f24c4d IM |
2909 | schedstat_inc(sd, alb_pushed); |
2910 | else | |
2911 | schedstat_inc(sd, alb_failed); | |
2912 | } | |
39507451 | 2913 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2914 | } |
2915 | ||
46cb4b7c SS |
2916 | #ifdef CONFIG_NO_HZ |
2917 | static struct { | |
2918 | atomic_t load_balancer; | |
2919 | cpumask_t cpu_mask; | |
2920 | } nohz ____cacheline_aligned = { | |
2921 | .load_balancer = ATOMIC_INIT(-1), | |
2922 | .cpu_mask = CPU_MASK_NONE, | |
2923 | }; | |
2924 | ||
7835b98b | 2925 | /* |
46cb4b7c SS |
2926 | * This routine will try to nominate the ilb (idle load balancing) |
2927 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
2928 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
2929 | * go into this tickless mode, then there will be no ilb owner (as there is | |
2930 | * no need for one) and all the cpus will sleep till the next wakeup event | |
2931 | * arrives... | |
2932 | * | |
2933 | * For the ilb owner, tick is not stopped. And this tick will be used | |
2934 | * for idle load balancing. ilb owner will still be part of | |
2935 | * nohz.cpu_mask.. | |
7835b98b | 2936 | * |
46cb4b7c SS |
2937 | * While stopping the tick, this cpu will become the ilb owner if there |
2938 | * is no other owner. And will be the owner till that cpu becomes busy | |
2939 | * or if all cpus in the system stop their ticks at which point | |
2940 | * there is no need for ilb owner. | |
2941 | * | |
2942 | * When the ilb owner becomes busy, it nominates another owner, during the | |
2943 | * next busy scheduler_tick() | |
2944 | */ | |
2945 | int select_nohz_load_balancer(int stop_tick) | |
2946 | { | |
2947 | int cpu = smp_processor_id(); | |
2948 | ||
2949 | if (stop_tick) { | |
2950 | cpu_set(cpu, nohz.cpu_mask); | |
2951 | cpu_rq(cpu)->in_nohz_recently = 1; | |
2952 | ||
2953 | /* | |
2954 | * If we are going offline and still the leader, give up! | |
2955 | */ | |
2956 | if (cpu_is_offline(cpu) && | |
2957 | atomic_read(&nohz.load_balancer) == cpu) { | |
2958 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2959 | BUG(); | |
2960 | return 0; | |
2961 | } | |
2962 | ||
2963 | /* time for ilb owner also to sleep */ | |
2964 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
2965 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2966 | atomic_set(&nohz.load_balancer, -1); | |
2967 | return 0; | |
2968 | } | |
2969 | ||
2970 | if (atomic_read(&nohz.load_balancer) == -1) { | |
2971 | /* make me the ilb owner */ | |
2972 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
2973 | return 1; | |
2974 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
2975 | return 1; | |
2976 | } else { | |
2977 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
2978 | return 0; | |
2979 | ||
2980 | cpu_clear(cpu, nohz.cpu_mask); | |
2981 | ||
2982 | if (atomic_read(&nohz.load_balancer) == cpu) | |
2983 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
2984 | BUG(); | |
2985 | } | |
2986 | return 0; | |
2987 | } | |
2988 | #endif | |
2989 | ||
2990 | static DEFINE_SPINLOCK(balancing); | |
2991 | ||
2992 | /* | |
7835b98b CL |
2993 | * It checks each scheduling domain to see if it is due to be balanced, |
2994 | * and initiates a balancing operation if so. | |
2995 | * | |
2996 | * Balancing parameters are set up in arch_init_sched_domains. | |
2997 | */ | |
a9957449 | 2998 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 2999 | { |
46cb4b7c SS |
3000 | int balance = 1; |
3001 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3002 | unsigned long interval; |
3003 | struct sched_domain *sd; | |
46cb4b7c | 3004 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3005 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3006 | int update_next_balance = 0; |
1da177e4 | 3007 | |
46cb4b7c | 3008 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3009 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3010 | continue; | |
3011 | ||
3012 | interval = sd->balance_interval; | |
d15bcfdb | 3013 | if (idle != CPU_IDLE) |
1da177e4 LT |
3014 | interval *= sd->busy_factor; |
3015 | ||
3016 | /* scale ms to jiffies */ | |
3017 | interval = msecs_to_jiffies(interval); | |
3018 | if (unlikely(!interval)) | |
3019 | interval = 1; | |
dd41f596 IM |
3020 | if (interval > HZ*NR_CPUS/10) |
3021 | interval = HZ*NR_CPUS/10; | |
3022 | ||
1da177e4 | 3023 | |
08c183f3 CL |
3024 | if (sd->flags & SD_SERIALIZE) { |
3025 | if (!spin_trylock(&balancing)) | |
3026 | goto out; | |
3027 | } | |
3028 | ||
c9819f45 | 3029 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3030 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3031 | /* |
3032 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3033 | * longer idle, or one of our SMT siblings is |
3034 | * not idle. | |
3035 | */ | |
d15bcfdb | 3036 | idle = CPU_NOT_IDLE; |
1da177e4 | 3037 | } |
1bd77f2d | 3038 | sd->last_balance = jiffies; |
1da177e4 | 3039 | } |
08c183f3 CL |
3040 | if (sd->flags & SD_SERIALIZE) |
3041 | spin_unlock(&balancing); | |
3042 | out: | |
f549da84 | 3043 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3044 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3045 | update_next_balance = 1; |
3046 | } | |
783609c6 SS |
3047 | |
3048 | /* | |
3049 | * Stop the load balance at this level. There is another | |
3050 | * CPU in our sched group which is doing load balancing more | |
3051 | * actively. | |
3052 | */ | |
3053 | if (!balance) | |
3054 | break; | |
1da177e4 | 3055 | } |
f549da84 SS |
3056 | |
3057 | /* | |
3058 | * next_balance will be updated only when there is a need. | |
3059 | * When the cpu is attached to null domain for ex, it will not be | |
3060 | * updated. | |
3061 | */ | |
3062 | if (likely(update_next_balance)) | |
3063 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3064 | } |
3065 | ||
3066 | /* | |
3067 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3068 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3069 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3070 | */ | |
3071 | static void run_rebalance_domains(struct softirq_action *h) | |
3072 | { | |
dd41f596 IM |
3073 | int this_cpu = smp_processor_id(); |
3074 | struct rq *this_rq = cpu_rq(this_cpu); | |
3075 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3076 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3077 | |
dd41f596 | 3078 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3079 | |
3080 | #ifdef CONFIG_NO_HZ | |
3081 | /* | |
3082 | * If this cpu is the owner for idle load balancing, then do the | |
3083 | * balancing on behalf of the other idle cpus whose ticks are | |
3084 | * stopped. | |
3085 | */ | |
dd41f596 IM |
3086 | if (this_rq->idle_at_tick && |
3087 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3088 | cpumask_t cpus = nohz.cpu_mask; |
3089 | struct rq *rq; | |
3090 | int balance_cpu; | |
3091 | ||
dd41f596 | 3092 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3093 | for_each_cpu_mask(balance_cpu, cpus) { |
3094 | /* | |
3095 | * If this cpu gets work to do, stop the load balancing | |
3096 | * work being done for other cpus. Next load | |
3097 | * balancing owner will pick it up. | |
3098 | */ | |
3099 | if (need_resched()) | |
3100 | break; | |
3101 | ||
de0cf899 | 3102 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3103 | |
3104 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3105 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3106 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3107 | } |
3108 | } | |
3109 | #endif | |
3110 | } | |
3111 | ||
3112 | /* | |
3113 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3114 | * | |
3115 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3116 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3117 | * if the whole system is idle. | |
3118 | */ | |
dd41f596 | 3119 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3120 | { |
46cb4b7c SS |
3121 | #ifdef CONFIG_NO_HZ |
3122 | /* | |
3123 | * If we were in the nohz mode recently and busy at the current | |
3124 | * scheduler tick, then check if we need to nominate new idle | |
3125 | * load balancer. | |
3126 | */ | |
3127 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3128 | rq->in_nohz_recently = 0; | |
3129 | ||
3130 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3131 | cpu_clear(cpu, nohz.cpu_mask); | |
3132 | atomic_set(&nohz.load_balancer, -1); | |
3133 | } | |
3134 | ||
3135 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3136 | /* | |
3137 | * simple selection for now: Nominate the | |
3138 | * first cpu in the nohz list to be the next | |
3139 | * ilb owner. | |
3140 | * | |
3141 | * TBD: Traverse the sched domains and nominate | |
3142 | * the nearest cpu in the nohz.cpu_mask. | |
3143 | */ | |
3144 | int ilb = first_cpu(nohz.cpu_mask); | |
3145 | ||
3146 | if (ilb != NR_CPUS) | |
3147 | resched_cpu(ilb); | |
3148 | } | |
3149 | } | |
3150 | ||
3151 | /* | |
3152 | * If this cpu is idle and doing idle load balancing for all the | |
3153 | * cpus with ticks stopped, is it time for that to stop? | |
3154 | */ | |
3155 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3156 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3157 | resched_cpu(cpu); | |
3158 | return; | |
3159 | } | |
3160 | ||
3161 | /* | |
3162 | * If this cpu is idle and the idle load balancing is done by | |
3163 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3164 | */ | |
3165 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3166 | cpu_isset(cpu, nohz.cpu_mask)) | |
3167 | return; | |
3168 | #endif | |
3169 | if (time_after_eq(jiffies, rq->next_balance)) | |
3170 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3171 | } |
dd41f596 IM |
3172 | |
3173 | #else /* CONFIG_SMP */ | |
3174 | ||
1da177e4 LT |
3175 | /* |
3176 | * on UP we do not need to balance between CPUs: | |
3177 | */ | |
70b97a7f | 3178 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3179 | { |
3180 | } | |
dd41f596 IM |
3181 | |
3182 | /* Avoid "used but not defined" warning on UP */ | |
3183 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3184 | unsigned long max_nr_move, unsigned long max_load_move, | |
3185 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3186 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3187 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3188 | { |
3189 | *load_moved = 0; | |
3190 | ||
3191 | return 0; | |
3192 | } | |
3193 | ||
1da177e4 LT |
3194 | #endif |
3195 | ||
1da177e4 LT |
3196 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3197 | ||
3198 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3199 | ||
3200 | /* | |
41b86e9c IM |
3201 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3202 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3203 | */ |
41b86e9c | 3204 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3205 | { |
1da177e4 | 3206 | unsigned long flags; |
41b86e9c IM |
3207 | u64 ns, delta_exec; |
3208 | struct rq *rq; | |
48f24c4d | 3209 | |
41b86e9c IM |
3210 | rq = task_rq_lock(p, &flags); |
3211 | ns = p->se.sum_exec_runtime; | |
3212 | if (rq->curr == p) { | |
a8e504d2 IM |
3213 | update_rq_clock(rq); |
3214 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3215 | if ((s64)delta_exec > 0) |
3216 | ns += delta_exec; | |
3217 | } | |
3218 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3219 | |
1da177e4 LT |
3220 | return ns; |
3221 | } | |
3222 | ||
1da177e4 LT |
3223 | /* |
3224 | * Account user cpu time to a process. | |
3225 | * @p: the process that the cpu time gets accounted to | |
3226 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3227 | * @cputime: the cpu time spent in user space since the last update | |
3228 | */ | |
3229 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3230 | { | |
3231 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3232 | cputime64_t tmp; | |
3233 | ||
3234 | p->utime = cputime_add(p->utime, cputime); | |
3235 | ||
3236 | /* Add user time to cpustat. */ | |
3237 | tmp = cputime_to_cputime64(cputime); | |
3238 | if (TASK_NICE(p) > 0) | |
3239 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3240 | else | |
3241 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3242 | } | |
3243 | ||
3244 | /* | |
3245 | * Account system cpu time to a process. | |
3246 | * @p: the process that the cpu time gets accounted to | |
3247 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3248 | * @cputime: the cpu time spent in kernel space since the last update | |
3249 | */ | |
3250 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3251 | cputime_t cputime) | |
3252 | { | |
3253 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3254 | struct rq *rq = this_rq(); |
1da177e4 LT |
3255 | cputime64_t tmp; |
3256 | ||
3257 | p->stime = cputime_add(p->stime, cputime); | |
3258 | ||
3259 | /* Add system time to cpustat. */ | |
3260 | tmp = cputime_to_cputime64(cputime); | |
3261 | if (hardirq_count() - hardirq_offset) | |
3262 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3263 | else if (softirq_count()) | |
3264 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3265 | else if (p != rq->idle) | |
3266 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3267 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3268 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3269 | else | |
3270 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3271 | /* Account for system time used */ | |
3272 | acct_update_integrals(p); | |
1da177e4 LT |
3273 | } |
3274 | ||
3275 | /* | |
3276 | * Account for involuntary wait time. | |
3277 | * @p: the process from which the cpu time has been stolen | |
3278 | * @steal: the cpu time spent in involuntary wait | |
3279 | */ | |
3280 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3281 | { | |
3282 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3283 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3284 | struct rq *rq = this_rq(); |
1da177e4 LT |
3285 | |
3286 | if (p == rq->idle) { | |
3287 | p->stime = cputime_add(p->stime, steal); | |
3288 | if (atomic_read(&rq->nr_iowait) > 0) | |
3289 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3290 | else | |
3291 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3292 | } else | |
3293 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3294 | } | |
3295 | ||
7835b98b CL |
3296 | /* |
3297 | * This function gets called by the timer code, with HZ frequency. | |
3298 | * We call it with interrupts disabled. | |
3299 | * | |
3300 | * It also gets called by the fork code, when changing the parent's | |
3301 | * timeslices. | |
3302 | */ | |
3303 | void scheduler_tick(void) | |
3304 | { | |
7835b98b CL |
3305 | int cpu = smp_processor_id(); |
3306 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3307 | struct task_struct *curr = rq->curr; |
529c7726 | 3308 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3309 | |
3310 | spin_lock(&rq->lock); | |
546fe3c9 | 3311 | __update_rq_clock(rq); |
529c7726 IM |
3312 | /* |
3313 | * Let rq->clock advance by at least TICK_NSEC: | |
3314 | */ | |
3315 | if (unlikely(rq->clock < next_tick)) | |
3316 | rq->clock = next_tick; | |
3317 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3318 | update_cpu_load(rq); |
dd41f596 IM |
3319 | if (curr != rq->idle) /* FIXME: needed? */ |
3320 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3321 | spin_unlock(&rq->lock); |
7835b98b | 3322 | |
e418e1c2 | 3323 | #ifdef CONFIG_SMP |
dd41f596 IM |
3324 | rq->idle_at_tick = idle_cpu(cpu); |
3325 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3326 | #endif |
1da177e4 LT |
3327 | } |
3328 | ||
1da177e4 LT |
3329 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3330 | ||
3331 | void fastcall add_preempt_count(int val) | |
3332 | { | |
3333 | /* | |
3334 | * Underflow? | |
3335 | */ | |
9a11b49a IM |
3336 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3337 | return; | |
1da177e4 LT |
3338 | preempt_count() += val; |
3339 | /* | |
3340 | * Spinlock count overflowing soon? | |
3341 | */ | |
33859f7f MOS |
3342 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3343 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3344 | } |
3345 | EXPORT_SYMBOL(add_preempt_count); | |
3346 | ||
3347 | void fastcall sub_preempt_count(int val) | |
3348 | { | |
3349 | /* | |
3350 | * Underflow? | |
3351 | */ | |
9a11b49a IM |
3352 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3353 | return; | |
1da177e4 LT |
3354 | /* |
3355 | * Is the spinlock portion underflowing? | |
3356 | */ | |
9a11b49a IM |
3357 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3358 | !(preempt_count() & PREEMPT_MASK))) | |
3359 | return; | |
3360 | ||
1da177e4 LT |
3361 | preempt_count() -= val; |
3362 | } | |
3363 | EXPORT_SYMBOL(sub_preempt_count); | |
3364 | ||
3365 | #endif | |
3366 | ||
3367 | /* | |
dd41f596 | 3368 | * Print scheduling while atomic bug: |
1da177e4 | 3369 | */ |
dd41f596 | 3370 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3371 | { |
dd41f596 IM |
3372 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3373 | prev->comm, preempt_count(), prev->pid); | |
3374 | debug_show_held_locks(prev); | |
3375 | if (irqs_disabled()) | |
3376 | print_irqtrace_events(prev); | |
3377 | dump_stack(); | |
3378 | } | |
1da177e4 | 3379 | |
dd41f596 IM |
3380 | /* |
3381 | * Various schedule()-time debugging checks and statistics: | |
3382 | */ | |
3383 | static inline void schedule_debug(struct task_struct *prev) | |
3384 | { | |
1da177e4 LT |
3385 | /* |
3386 | * Test if we are atomic. Since do_exit() needs to call into | |
3387 | * schedule() atomically, we ignore that path for now. | |
3388 | * Otherwise, whine if we are scheduling when we should not be. | |
3389 | */ | |
dd41f596 IM |
3390 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3391 | __schedule_bug(prev); | |
3392 | ||
1da177e4 LT |
3393 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3394 | ||
2d72376b | 3395 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3396 | #ifdef CONFIG_SCHEDSTATS |
3397 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3398 | schedstat_inc(this_rq(), bkl_count); |
3399 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3400 | } |
3401 | #endif | |
dd41f596 IM |
3402 | } |
3403 | ||
3404 | /* | |
3405 | * Pick up the highest-prio task: | |
3406 | */ | |
3407 | static inline struct task_struct * | |
ff95f3df | 3408 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3409 | { |
5522d5d5 | 3410 | const struct sched_class *class; |
dd41f596 | 3411 | struct task_struct *p; |
1da177e4 LT |
3412 | |
3413 | /* | |
dd41f596 IM |
3414 | * Optimization: we know that if all tasks are in |
3415 | * the fair class we can call that function directly: | |
1da177e4 | 3416 | */ |
dd41f596 | 3417 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3418 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3419 | if (likely(p)) |
3420 | return p; | |
1da177e4 LT |
3421 | } |
3422 | ||
dd41f596 IM |
3423 | class = sched_class_highest; |
3424 | for ( ; ; ) { | |
fb8d4724 | 3425 | p = class->pick_next_task(rq); |
dd41f596 IM |
3426 | if (p) |
3427 | return p; | |
3428 | /* | |
3429 | * Will never be NULL as the idle class always | |
3430 | * returns a non-NULL p: | |
3431 | */ | |
3432 | class = class->next; | |
3433 | } | |
3434 | } | |
1da177e4 | 3435 | |
dd41f596 IM |
3436 | /* |
3437 | * schedule() is the main scheduler function. | |
3438 | */ | |
3439 | asmlinkage void __sched schedule(void) | |
3440 | { | |
3441 | struct task_struct *prev, *next; | |
3442 | long *switch_count; | |
3443 | struct rq *rq; | |
dd41f596 IM |
3444 | int cpu; |
3445 | ||
3446 | need_resched: | |
3447 | preempt_disable(); | |
3448 | cpu = smp_processor_id(); | |
3449 | rq = cpu_rq(cpu); | |
3450 | rcu_qsctr_inc(cpu); | |
3451 | prev = rq->curr; | |
3452 | switch_count = &prev->nivcsw; | |
3453 | ||
3454 | release_kernel_lock(prev); | |
3455 | need_resched_nonpreemptible: | |
3456 | ||
3457 | schedule_debug(prev); | |
1da177e4 | 3458 | |
1e819950 IM |
3459 | /* |
3460 | * Do the rq-clock update outside the rq lock: | |
3461 | */ | |
3462 | local_irq_disable(); | |
c1b3da3e | 3463 | __update_rq_clock(rq); |
1e819950 IM |
3464 | spin_lock(&rq->lock); |
3465 | clear_tsk_need_resched(prev); | |
1da177e4 | 3466 | |
1da177e4 | 3467 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3468 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3469 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3470 | prev->state = TASK_RUNNING; |
dd41f596 | 3471 | } else { |
2e1cb74a | 3472 | deactivate_task(rq, prev, 1); |
1da177e4 | 3473 | } |
dd41f596 | 3474 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3475 | } |
3476 | ||
dd41f596 | 3477 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3478 | idle_balance(cpu, rq); |
1da177e4 | 3479 | |
31ee529c | 3480 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3481 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3482 | |
3483 | sched_info_switch(prev, next); | |
dd41f596 | 3484 | |
1da177e4 | 3485 | if (likely(prev != next)) { |
1da177e4 LT |
3486 | rq->nr_switches++; |
3487 | rq->curr = next; | |
3488 | ++*switch_count; | |
3489 | ||
dd41f596 | 3490 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3491 | } else |
3492 | spin_unlock_irq(&rq->lock); | |
3493 | ||
dd41f596 IM |
3494 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3495 | cpu = smp_processor_id(); | |
3496 | rq = cpu_rq(cpu); | |
1da177e4 | 3497 | goto need_resched_nonpreemptible; |
dd41f596 | 3498 | } |
1da177e4 LT |
3499 | preempt_enable_no_resched(); |
3500 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3501 | goto need_resched; | |
3502 | } | |
1da177e4 LT |
3503 | EXPORT_SYMBOL(schedule); |
3504 | ||
3505 | #ifdef CONFIG_PREEMPT | |
3506 | /* | |
2ed6e34f | 3507 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3508 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3509 | * occur there and call schedule directly. | |
3510 | */ | |
3511 | asmlinkage void __sched preempt_schedule(void) | |
3512 | { | |
3513 | struct thread_info *ti = current_thread_info(); | |
3514 | #ifdef CONFIG_PREEMPT_BKL | |
3515 | struct task_struct *task = current; | |
3516 | int saved_lock_depth; | |
3517 | #endif | |
3518 | /* | |
3519 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3520 | * we do not want to preempt the current task. Just return.. | |
3521 | */ | |
beed33a8 | 3522 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3523 | return; |
3524 | ||
3a5c359a AK |
3525 | do { |
3526 | add_preempt_count(PREEMPT_ACTIVE); | |
3527 | ||
3528 | /* | |
3529 | * We keep the big kernel semaphore locked, but we | |
3530 | * clear ->lock_depth so that schedule() doesnt | |
3531 | * auto-release the semaphore: | |
3532 | */ | |
1da177e4 | 3533 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3534 | saved_lock_depth = task->lock_depth; |
3535 | task->lock_depth = -1; | |
1da177e4 | 3536 | #endif |
3a5c359a | 3537 | schedule(); |
1da177e4 | 3538 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3539 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3540 | #endif |
3a5c359a | 3541 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3542 | |
3a5c359a AK |
3543 | /* |
3544 | * Check again in case we missed a preemption opportunity | |
3545 | * between schedule and now. | |
3546 | */ | |
3547 | barrier(); | |
3548 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3549 | } |
1da177e4 LT |
3550 | EXPORT_SYMBOL(preempt_schedule); |
3551 | ||
3552 | /* | |
2ed6e34f | 3553 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3554 | * off of irq context. |
3555 | * Note, that this is called and return with irqs disabled. This will | |
3556 | * protect us against recursive calling from irq. | |
3557 | */ | |
3558 | asmlinkage void __sched preempt_schedule_irq(void) | |
3559 | { | |
3560 | struct thread_info *ti = current_thread_info(); | |
3561 | #ifdef CONFIG_PREEMPT_BKL | |
3562 | struct task_struct *task = current; | |
3563 | int saved_lock_depth; | |
3564 | #endif | |
2ed6e34f | 3565 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3566 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3567 | ||
3a5c359a AK |
3568 | do { |
3569 | add_preempt_count(PREEMPT_ACTIVE); | |
3570 | ||
3571 | /* | |
3572 | * We keep the big kernel semaphore locked, but we | |
3573 | * clear ->lock_depth so that schedule() doesnt | |
3574 | * auto-release the semaphore: | |
3575 | */ | |
1da177e4 | 3576 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3577 | saved_lock_depth = task->lock_depth; |
3578 | task->lock_depth = -1; | |
1da177e4 | 3579 | #endif |
3a5c359a AK |
3580 | local_irq_enable(); |
3581 | schedule(); | |
3582 | local_irq_disable(); | |
1da177e4 | 3583 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3584 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3585 | #endif |
3a5c359a | 3586 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3587 | |
3a5c359a AK |
3588 | /* |
3589 | * Check again in case we missed a preemption opportunity | |
3590 | * between schedule and now. | |
3591 | */ | |
3592 | barrier(); | |
3593 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3594 | } |
3595 | ||
3596 | #endif /* CONFIG_PREEMPT */ | |
3597 | ||
95cdf3b7 IM |
3598 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3599 | void *key) | |
1da177e4 | 3600 | { |
48f24c4d | 3601 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3602 | } |
1da177e4 LT |
3603 | EXPORT_SYMBOL(default_wake_function); |
3604 | ||
3605 | /* | |
3606 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3607 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3608 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3609 | * | |
3610 | * There are circumstances in which we can try to wake a task which has already | |
3611 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3612 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3613 | */ | |
3614 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3615 | int nr_exclusive, int sync, void *key) | |
3616 | { | |
2e45874c | 3617 | wait_queue_t *curr, *next; |
1da177e4 | 3618 | |
2e45874c | 3619 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3620 | unsigned flags = curr->flags; |
3621 | ||
1da177e4 | 3622 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3623 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3624 | break; |
3625 | } | |
3626 | } | |
3627 | ||
3628 | /** | |
3629 | * __wake_up - wake up threads blocked on a waitqueue. | |
3630 | * @q: the waitqueue | |
3631 | * @mode: which threads | |
3632 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3633 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3634 | */ |
3635 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3636 | int nr_exclusive, void *key) |
1da177e4 LT |
3637 | { |
3638 | unsigned long flags; | |
3639 | ||
3640 | spin_lock_irqsave(&q->lock, flags); | |
3641 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3642 | spin_unlock_irqrestore(&q->lock, flags); | |
3643 | } | |
1da177e4 LT |
3644 | EXPORT_SYMBOL(__wake_up); |
3645 | ||
3646 | /* | |
3647 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3648 | */ | |
3649 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3650 | { | |
3651 | __wake_up_common(q, mode, 1, 0, NULL); | |
3652 | } | |
3653 | ||
3654 | /** | |
67be2dd1 | 3655 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3656 | * @q: the waitqueue |
3657 | * @mode: which threads | |
3658 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3659 | * | |
3660 | * The sync wakeup differs that the waker knows that it will schedule | |
3661 | * away soon, so while the target thread will be woken up, it will not | |
3662 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3663 | * with each other. This can prevent needless bouncing between CPUs. | |
3664 | * | |
3665 | * On UP it can prevent extra preemption. | |
3666 | */ | |
95cdf3b7 IM |
3667 | void fastcall |
3668 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3669 | { |
3670 | unsigned long flags; | |
3671 | int sync = 1; | |
3672 | ||
3673 | if (unlikely(!q)) | |
3674 | return; | |
3675 | ||
3676 | if (unlikely(!nr_exclusive)) | |
3677 | sync = 0; | |
3678 | ||
3679 | spin_lock_irqsave(&q->lock, flags); | |
3680 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3681 | spin_unlock_irqrestore(&q->lock, flags); | |
3682 | } | |
3683 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3684 | ||
3685 | void fastcall complete(struct completion *x) | |
3686 | { | |
3687 | unsigned long flags; | |
3688 | ||
3689 | spin_lock_irqsave(&x->wait.lock, flags); | |
3690 | x->done++; | |
3691 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3692 | 1, 0, NULL); | |
3693 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3694 | } | |
3695 | EXPORT_SYMBOL(complete); | |
3696 | ||
3697 | void fastcall complete_all(struct completion *x) | |
3698 | { | |
3699 | unsigned long flags; | |
3700 | ||
3701 | spin_lock_irqsave(&x->wait.lock, flags); | |
3702 | x->done += UINT_MAX/2; | |
3703 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3704 | 0, 0, NULL); | |
3705 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3706 | } | |
3707 | EXPORT_SYMBOL(complete_all); | |
3708 | ||
8cbbe86d AK |
3709 | static inline long __sched |
3710 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3711 | { |
1da177e4 LT |
3712 | if (!x->done) { |
3713 | DECLARE_WAITQUEUE(wait, current); | |
3714 | ||
3715 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3716 | __add_wait_queue_tail(&x->wait, &wait); | |
3717 | do { | |
8cbbe86d AK |
3718 | if (state == TASK_INTERRUPTIBLE && |
3719 | signal_pending(current)) { | |
3720 | __remove_wait_queue(&x->wait, &wait); | |
3721 | return -ERESTARTSYS; | |
3722 | } | |
3723 | __set_current_state(state); | |
1da177e4 LT |
3724 | spin_unlock_irq(&x->wait.lock); |
3725 | timeout = schedule_timeout(timeout); | |
3726 | spin_lock_irq(&x->wait.lock); | |
3727 | if (!timeout) { | |
3728 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3729 | return timeout; |
1da177e4 LT |
3730 | } |
3731 | } while (!x->done); | |
3732 | __remove_wait_queue(&x->wait, &wait); | |
3733 | } | |
3734 | x->done--; | |
1da177e4 LT |
3735 | return timeout; |
3736 | } | |
1da177e4 | 3737 | |
8cbbe86d AK |
3738 | static long __sched |
3739 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3740 | { |
1da177e4 LT |
3741 | might_sleep(); |
3742 | ||
3743 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3744 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3745 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3746 | return timeout; |
3747 | } | |
1da177e4 | 3748 | |
8cbbe86d AK |
3749 | void fastcall __sched wait_for_completion(struct completion *x) |
3750 | { | |
3751 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3752 | } |
8cbbe86d | 3753 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 LT |
3754 | |
3755 | unsigned long fastcall __sched | |
8cbbe86d | 3756 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3757 | { |
8cbbe86d | 3758 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3759 | } |
8cbbe86d | 3760 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3761 | |
8cbbe86d | 3762 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3763 | { |
8cbbe86d | 3764 | return wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
0fec171c | 3765 | } |
8cbbe86d | 3766 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3767 | |
8cbbe86d AK |
3768 | unsigned long fastcall __sched |
3769 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3770 | unsigned long timeout) | |
0fec171c | 3771 | { |
8cbbe86d | 3772 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3773 | } |
8cbbe86d | 3774 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3775 | |
8cbbe86d AK |
3776 | static long __sched |
3777 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3778 | { |
0fec171c IM |
3779 | unsigned long flags; |
3780 | wait_queue_t wait; | |
3781 | ||
3782 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3783 | |
8cbbe86d | 3784 | __set_current_state(state); |
1da177e4 | 3785 | |
8cbbe86d AK |
3786 | spin_lock_irqsave(&q->lock, flags); |
3787 | __add_wait_queue(q, &wait); | |
3788 | spin_unlock(&q->lock); | |
3789 | timeout = schedule_timeout(timeout); | |
3790 | spin_lock_irq(&q->lock); | |
3791 | __remove_wait_queue(q, &wait); | |
3792 | spin_unlock_irqrestore(&q->lock, flags); | |
3793 | ||
3794 | return timeout; | |
3795 | } | |
3796 | ||
3797 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3798 | { | |
3799 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3800 | } |
1da177e4 LT |
3801 | EXPORT_SYMBOL(interruptible_sleep_on); |
3802 | ||
0fec171c | 3803 | long __sched |
95cdf3b7 | 3804 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3805 | { |
8cbbe86d | 3806 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3807 | } |
1da177e4 LT |
3808 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3809 | ||
0fec171c | 3810 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3811 | { |
8cbbe86d | 3812 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3813 | } |
1da177e4 LT |
3814 | EXPORT_SYMBOL(sleep_on); |
3815 | ||
0fec171c | 3816 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3817 | { |
8cbbe86d | 3818 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3819 | } |
1da177e4 LT |
3820 | EXPORT_SYMBOL(sleep_on_timeout); |
3821 | ||
b29739f9 IM |
3822 | #ifdef CONFIG_RT_MUTEXES |
3823 | ||
3824 | /* | |
3825 | * rt_mutex_setprio - set the current priority of a task | |
3826 | * @p: task | |
3827 | * @prio: prio value (kernel-internal form) | |
3828 | * | |
3829 | * This function changes the 'effective' priority of a task. It does | |
3830 | * not touch ->normal_prio like __setscheduler(). | |
3831 | * | |
3832 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3833 | */ | |
36c8b586 | 3834 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3835 | { |
3836 | unsigned long flags; | |
83b699ed | 3837 | int oldprio, on_rq, running; |
70b97a7f | 3838 | struct rq *rq; |
b29739f9 IM |
3839 | |
3840 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3841 | ||
3842 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3843 | update_rq_clock(rq); |
b29739f9 | 3844 | |
d5f9f942 | 3845 | oldprio = p->prio; |
dd41f596 | 3846 | on_rq = p->se.on_rq; |
83b699ed SV |
3847 | running = task_running(rq, p); |
3848 | if (on_rq) { | |
69be72c1 | 3849 | dequeue_task(rq, p, 0); |
83b699ed SV |
3850 | if (running) |
3851 | p->sched_class->put_prev_task(rq, p); | |
3852 | } | |
dd41f596 IM |
3853 | |
3854 | if (rt_prio(prio)) | |
3855 | p->sched_class = &rt_sched_class; | |
3856 | else | |
3857 | p->sched_class = &fair_sched_class; | |
3858 | ||
b29739f9 IM |
3859 | p->prio = prio; |
3860 | ||
dd41f596 | 3861 | if (on_rq) { |
83b699ed SV |
3862 | if (running) |
3863 | p->sched_class->set_curr_task(rq); | |
8159f87e | 3864 | enqueue_task(rq, p, 0); |
b29739f9 IM |
3865 | /* |
3866 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3867 | * our priority decreased, or if we are not currently running on |
3868 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3869 | */ |
83b699ed | 3870 | if (running) { |
d5f9f942 AM |
3871 | if (p->prio > oldprio) |
3872 | resched_task(rq->curr); | |
dd41f596 IM |
3873 | } else { |
3874 | check_preempt_curr(rq, p); | |
3875 | } | |
b29739f9 IM |
3876 | } |
3877 | task_rq_unlock(rq, &flags); | |
3878 | } | |
3879 | ||
3880 | #endif | |
3881 | ||
36c8b586 | 3882 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 3883 | { |
dd41f596 | 3884 | int old_prio, delta, on_rq; |
1da177e4 | 3885 | unsigned long flags; |
70b97a7f | 3886 | struct rq *rq; |
1da177e4 LT |
3887 | |
3888 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
3889 | return; | |
3890 | /* | |
3891 | * We have to be careful, if called from sys_setpriority(), | |
3892 | * the task might be in the middle of scheduling on another CPU. | |
3893 | */ | |
3894 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3895 | update_rq_clock(rq); |
1da177e4 LT |
3896 | /* |
3897 | * The RT priorities are set via sched_setscheduler(), but we still | |
3898 | * allow the 'normal' nice value to be set - but as expected | |
3899 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 3900 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 3901 | */ |
e05606d3 | 3902 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
3903 | p->static_prio = NICE_TO_PRIO(nice); |
3904 | goto out_unlock; | |
3905 | } | |
dd41f596 IM |
3906 | on_rq = p->se.on_rq; |
3907 | if (on_rq) { | |
69be72c1 | 3908 | dequeue_task(rq, p, 0); |
79b5dddf | 3909 | dec_load(rq, p); |
2dd73a4f | 3910 | } |
1da177e4 | 3911 | |
1da177e4 | 3912 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 3913 | set_load_weight(p); |
b29739f9 IM |
3914 | old_prio = p->prio; |
3915 | p->prio = effective_prio(p); | |
3916 | delta = p->prio - old_prio; | |
1da177e4 | 3917 | |
dd41f596 | 3918 | if (on_rq) { |
8159f87e | 3919 | enqueue_task(rq, p, 0); |
29b4b623 | 3920 | inc_load(rq, p); |
1da177e4 | 3921 | /* |
d5f9f942 AM |
3922 | * If the task increased its priority or is running and |
3923 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 3924 | */ |
d5f9f942 | 3925 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
3926 | resched_task(rq->curr); |
3927 | } | |
3928 | out_unlock: | |
3929 | task_rq_unlock(rq, &flags); | |
3930 | } | |
1da177e4 LT |
3931 | EXPORT_SYMBOL(set_user_nice); |
3932 | ||
e43379f1 MM |
3933 | /* |
3934 | * can_nice - check if a task can reduce its nice value | |
3935 | * @p: task | |
3936 | * @nice: nice value | |
3937 | */ | |
36c8b586 | 3938 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 3939 | { |
024f4747 MM |
3940 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
3941 | int nice_rlim = 20 - nice; | |
48f24c4d | 3942 | |
e43379f1 MM |
3943 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
3944 | capable(CAP_SYS_NICE)); | |
3945 | } | |
3946 | ||
1da177e4 LT |
3947 | #ifdef __ARCH_WANT_SYS_NICE |
3948 | ||
3949 | /* | |
3950 | * sys_nice - change the priority of the current process. | |
3951 | * @increment: priority increment | |
3952 | * | |
3953 | * sys_setpriority is a more generic, but much slower function that | |
3954 | * does similar things. | |
3955 | */ | |
3956 | asmlinkage long sys_nice(int increment) | |
3957 | { | |
48f24c4d | 3958 | long nice, retval; |
1da177e4 LT |
3959 | |
3960 | /* | |
3961 | * Setpriority might change our priority at the same moment. | |
3962 | * We don't have to worry. Conceptually one call occurs first | |
3963 | * and we have a single winner. | |
3964 | */ | |
e43379f1 MM |
3965 | if (increment < -40) |
3966 | increment = -40; | |
1da177e4 LT |
3967 | if (increment > 40) |
3968 | increment = 40; | |
3969 | ||
3970 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
3971 | if (nice < -20) | |
3972 | nice = -20; | |
3973 | if (nice > 19) | |
3974 | nice = 19; | |
3975 | ||
e43379f1 MM |
3976 | if (increment < 0 && !can_nice(current, nice)) |
3977 | return -EPERM; | |
3978 | ||
1da177e4 LT |
3979 | retval = security_task_setnice(current, nice); |
3980 | if (retval) | |
3981 | return retval; | |
3982 | ||
3983 | set_user_nice(current, nice); | |
3984 | return 0; | |
3985 | } | |
3986 | ||
3987 | #endif | |
3988 | ||
3989 | /** | |
3990 | * task_prio - return the priority value of a given task. | |
3991 | * @p: the task in question. | |
3992 | * | |
3993 | * This is the priority value as seen by users in /proc. | |
3994 | * RT tasks are offset by -200. Normal tasks are centered | |
3995 | * around 0, value goes from -16 to +15. | |
3996 | */ | |
36c8b586 | 3997 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
3998 | { |
3999 | return p->prio - MAX_RT_PRIO; | |
4000 | } | |
4001 | ||
4002 | /** | |
4003 | * task_nice - return the nice value of a given task. | |
4004 | * @p: the task in question. | |
4005 | */ | |
36c8b586 | 4006 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4007 | { |
4008 | return TASK_NICE(p); | |
4009 | } | |
1da177e4 | 4010 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4011 | |
4012 | /** | |
4013 | * idle_cpu - is a given cpu idle currently? | |
4014 | * @cpu: the processor in question. | |
4015 | */ | |
4016 | int idle_cpu(int cpu) | |
4017 | { | |
4018 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4019 | } | |
4020 | ||
1da177e4 LT |
4021 | /** |
4022 | * idle_task - return the idle task for a given cpu. | |
4023 | * @cpu: the processor in question. | |
4024 | */ | |
36c8b586 | 4025 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4026 | { |
4027 | return cpu_rq(cpu)->idle; | |
4028 | } | |
4029 | ||
4030 | /** | |
4031 | * find_process_by_pid - find a process with a matching PID value. | |
4032 | * @pid: the pid in question. | |
4033 | */ | |
a9957449 | 4034 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4035 | { |
4036 | return pid ? find_task_by_pid(pid) : current; | |
4037 | } | |
4038 | ||
4039 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4040 | static void |
4041 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4042 | { |
dd41f596 | 4043 | BUG_ON(p->se.on_rq); |
48f24c4d | 4044 | |
1da177e4 | 4045 | p->policy = policy; |
dd41f596 IM |
4046 | switch (p->policy) { |
4047 | case SCHED_NORMAL: | |
4048 | case SCHED_BATCH: | |
4049 | case SCHED_IDLE: | |
4050 | p->sched_class = &fair_sched_class; | |
4051 | break; | |
4052 | case SCHED_FIFO: | |
4053 | case SCHED_RR: | |
4054 | p->sched_class = &rt_sched_class; | |
4055 | break; | |
4056 | } | |
4057 | ||
1da177e4 | 4058 | p->rt_priority = prio; |
b29739f9 IM |
4059 | p->normal_prio = normal_prio(p); |
4060 | /* we are holding p->pi_lock already */ | |
4061 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4062 | set_load_weight(p); |
1da177e4 LT |
4063 | } |
4064 | ||
4065 | /** | |
72fd4a35 | 4066 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4067 | * @p: the task in question. |
4068 | * @policy: new policy. | |
4069 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4070 | * |
72fd4a35 | 4071 | * NOTE that the task may be already dead. |
1da177e4 | 4072 | */ |
95cdf3b7 IM |
4073 | int sched_setscheduler(struct task_struct *p, int policy, |
4074 | struct sched_param *param) | |
1da177e4 | 4075 | { |
83b699ed | 4076 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4077 | unsigned long flags; |
70b97a7f | 4078 | struct rq *rq; |
1da177e4 | 4079 | |
66e5393a SR |
4080 | /* may grab non-irq protected spin_locks */ |
4081 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4082 | recheck: |
4083 | /* double check policy once rq lock held */ | |
4084 | if (policy < 0) | |
4085 | policy = oldpolicy = p->policy; | |
4086 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4087 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4088 | policy != SCHED_IDLE) | |
b0a9499c | 4089 | return -EINVAL; |
1da177e4 LT |
4090 | /* |
4091 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4092 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4093 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4094 | */ |
4095 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4096 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4097 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4098 | return -EINVAL; |
e05606d3 | 4099 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4100 | return -EINVAL; |
4101 | ||
37e4ab3f OC |
4102 | /* |
4103 | * Allow unprivileged RT tasks to decrease priority: | |
4104 | */ | |
4105 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4106 | if (rt_policy(policy)) { |
8dc3e909 | 4107 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4108 | |
4109 | if (!lock_task_sighand(p, &flags)) | |
4110 | return -ESRCH; | |
4111 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4112 | unlock_task_sighand(p, &flags); | |
4113 | ||
4114 | /* can't set/change the rt policy */ | |
4115 | if (policy != p->policy && !rlim_rtprio) | |
4116 | return -EPERM; | |
4117 | ||
4118 | /* can't increase priority */ | |
4119 | if (param->sched_priority > p->rt_priority && | |
4120 | param->sched_priority > rlim_rtprio) | |
4121 | return -EPERM; | |
4122 | } | |
dd41f596 IM |
4123 | /* |
4124 | * Like positive nice levels, dont allow tasks to | |
4125 | * move out of SCHED_IDLE either: | |
4126 | */ | |
4127 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4128 | return -EPERM; | |
5fe1d75f | 4129 | |
37e4ab3f OC |
4130 | /* can't change other user's priorities */ |
4131 | if ((current->euid != p->euid) && | |
4132 | (current->euid != p->uid)) | |
4133 | return -EPERM; | |
4134 | } | |
1da177e4 LT |
4135 | |
4136 | retval = security_task_setscheduler(p, policy, param); | |
4137 | if (retval) | |
4138 | return retval; | |
b29739f9 IM |
4139 | /* |
4140 | * make sure no PI-waiters arrive (or leave) while we are | |
4141 | * changing the priority of the task: | |
4142 | */ | |
4143 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4144 | /* |
4145 | * To be able to change p->policy safely, the apropriate | |
4146 | * runqueue lock must be held. | |
4147 | */ | |
b29739f9 | 4148 | rq = __task_rq_lock(p); |
1da177e4 LT |
4149 | /* recheck policy now with rq lock held */ |
4150 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4151 | policy = oldpolicy = -1; | |
b29739f9 IM |
4152 | __task_rq_unlock(rq); |
4153 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4154 | goto recheck; |
4155 | } | |
2daa3577 | 4156 | update_rq_clock(rq); |
dd41f596 | 4157 | on_rq = p->se.on_rq; |
83b699ed SV |
4158 | running = task_running(rq, p); |
4159 | if (on_rq) { | |
2e1cb74a | 4160 | deactivate_task(rq, p, 0); |
83b699ed SV |
4161 | if (running) |
4162 | p->sched_class->put_prev_task(rq, p); | |
4163 | } | |
f6b53205 | 4164 | |
1da177e4 | 4165 | oldprio = p->prio; |
dd41f596 | 4166 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4167 | |
dd41f596 | 4168 | if (on_rq) { |
83b699ed SV |
4169 | if (running) |
4170 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4171 | activate_task(rq, p, 0); |
1da177e4 LT |
4172 | /* |
4173 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4174 | * our priority decreased, or if we are not currently running on |
4175 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4176 | */ |
83b699ed | 4177 | if (running) { |
d5f9f942 AM |
4178 | if (p->prio > oldprio) |
4179 | resched_task(rq->curr); | |
dd41f596 IM |
4180 | } else { |
4181 | check_preempt_curr(rq, p); | |
4182 | } | |
1da177e4 | 4183 | } |
b29739f9 IM |
4184 | __task_rq_unlock(rq); |
4185 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4186 | ||
95e02ca9 TG |
4187 | rt_mutex_adjust_pi(p); |
4188 | ||
1da177e4 LT |
4189 | return 0; |
4190 | } | |
4191 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4192 | ||
95cdf3b7 IM |
4193 | static int |
4194 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4195 | { |
1da177e4 LT |
4196 | struct sched_param lparam; |
4197 | struct task_struct *p; | |
36c8b586 | 4198 | int retval; |
1da177e4 LT |
4199 | |
4200 | if (!param || pid < 0) | |
4201 | return -EINVAL; | |
4202 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4203 | return -EFAULT; | |
5fe1d75f ON |
4204 | |
4205 | rcu_read_lock(); | |
4206 | retval = -ESRCH; | |
1da177e4 | 4207 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4208 | if (p != NULL) |
4209 | retval = sched_setscheduler(p, policy, &lparam); | |
4210 | rcu_read_unlock(); | |
36c8b586 | 4211 | |
1da177e4 LT |
4212 | return retval; |
4213 | } | |
4214 | ||
4215 | /** | |
4216 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4217 | * @pid: the pid in question. | |
4218 | * @policy: new policy. | |
4219 | * @param: structure containing the new RT priority. | |
4220 | */ | |
4221 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4222 | struct sched_param __user *param) | |
4223 | { | |
c21761f1 JB |
4224 | /* negative values for policy are not valid */ |
4225 | if (policy < 0) | |
4226 | return -EINVAL; | |
4227 | ||
1da177e4 LT |
4228 | return do_sched_setscheduler(pid, policy, param); |
4229 | } | |
4230 | ||
4231 | /** | |
4232 | * sys_sched_setparam - set/change the RT priority of a thread | |
4233 | * @pid: the pid in question. | |
4234 | * @param: structure containing the new RT priority. | |
4235 | */ | |
4236 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4237 | { | |
4238 | return do_sched_setscheduler(pid, -1, param); | |
4239 | } | |
4240 | ||
4241 | /** | |
4242 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4243 | * @pid: the pid in question. | |
4244 | */ | |
4245 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4246 | { | |
36c8b586 | 4247 | struct task_struct *p; |
3a5c359a | 4248 | int retval; |
1da177e4 LT |
4249 | |
4250 | if (pid < 0) | |
3a5c359a | 4251 | return -EINVAL; |
1da177e4 LT |
4252 | |
4253 | retval = -ESRCH; | |
4254 | read_lock(&tasklist_lock); | |
4255 | p = find_process_by_pid(pid); | |
4256 | if (p) { | |
4257 | retval = security_task_getscheduler(p); | |
4258 | if (!retval) | |
4259 | retval = p->policy; | |
4260 | } | |
4261 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4262 | return retval; |
4263 | } | |
4264 | ||
4265 | /** | |
4266 | * sys_sched_getscheduler - get the RT priority of a thread | |
4267 | * @pid: the pid in question. | |
4268 | * @param: structure containing the RT priority. | |
4269 | */ | |
4270 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4271 | { | |
4272 | struct sched_param lp; | |
36c8b586 | 4273 | struct task_struct *p; |
3a5c359a | 4274 | int retval; |
1da177e4 LT |
4275 | |
4276 | if (!param || pid < 0) | |
3a5c359a | 4277 | return -EINVAL; |
1da177e4 LT |
4278 | |
4279 | read_lock(&tasklist_lock); | |
4280 | p = find_process_by_pid(pid); | |
4281 | retval = -ESRCH; | |
4282 | if (!p) | |
4283 | goto out_unlock; | |
4284 | ||
4285 | retval = security_task_getscheduler(p); | |
4286 | if (retval) | |
4287 | goto out_unlock; | |
4288 | ||
4289 | lp.sched_priority = p->rt_priority; | |
4290 | read_unlock(&tasklist_lock); | |
4291 | ||
4292 | /* | |
4293 | * This one might sleep, we cannot do it with a spinlock held ... | |
4294 | */ | |
4295 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4296 | ||
1da177e4 LT |
4297 | return retval; |
4298 | ||
4299 | out_unlock: | |
4300 | read_unlock(&tasklist_lock); | |
4301 | return retval; | |
4302 | } | |
4303 | ||
4304 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4305 | { | |
1da177e4 | 4306 | cpumask_t cpus_allowed; |
36c8b586 IM |
4307 | struct task_struct *p; |
4308 | int retval; | |
1da177e4 | 4309 | |
5be9361c | 4310 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4311 | read_lock(&tasklist_lock); |
4312 | ||
4313 | p = find_process_by_pid(pid); | |
4314 | if (!p) { | |
4315 | read_unlock(&tasklist_lock); | |
5be9361c | 4316 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4317 | return -ESRCH; |
4318 | } | |
4319 | ||
4320 | /* | |
4321 | * It is not safe to call set_cpus_allowed with the | |
4322 | * tasklist_lock held. We will bump the task_struct's | |
4323 | * usage count and then drop tasklist_lock. | |
4324 | */ | |
4325 | get_task_struct(p); | |
4326 | read_unlock(&tasklist_lock); | |
4327 | ||
4328 | retval = -EPERM; | |
4329 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4330 | !capable(CAP_SYS_NICE)) | |
4331 | goto out_unlock; | |
4332 | ||
e7834f8f DQ |
4333 | retval = security_task_setscheduler(p, 0, NULL); |
4334 | if (retval) | |
4335 | goto out_unlock; | |
4336 | ||
1da177e4 LT |
4337 | cpus_allowed = cpuset_cpus_allowed(p); |
4338 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4339 | retval = set_cpus_allowed(p, new_mask); | |
4340 | ||
4341 | out_unlock: | |
4342 | put_task_struct(p); | |
5be9361c | 4343 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4344 | return retval; |
4345 | } | |
4346 | ||
4347 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4348 | cpumask_t *new_mask) | |
4349 | { | |
4350 | if (len < sizeof(cpumask_t)) { | |
4351 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4352 | } else if (len > sizeof(cpumask_t)) { | |
4353 | len = sizeof(cpumask_t); | |
4354 | } | |
4355 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4356 | } | |
4357 | ||
4358 | /** | |
4359 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4360 | * @pid: pid of the process | |
4361 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4362 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4363 | */ | |
4364 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4365 | unsigned long __user *user_mask_ptr) | |
4366 | { | |
4367 | cpumask_t new_mask; | |
4368 | int retval; | |
4369 | ||
4370 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4371 | if (retval) | |
4372 | return retval; | |
4373 | ||
4374 | return sched_setaffinity(pid, new_mask); | |
4375 | } | |
4376 | ||
4377 | /* | |
4378 | * Represents all cpu's present in the system | |
4379 | * In systems capable of hotplug, this map could dynamically grow | |
4380 | * as new cpu's are detected in the system via any platform specific | |
4381 | * method, such as ACPI for e.g. | |
4382 | */ | |
4383 | ||
4cef0c61 | 4384 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4385 | EXPORT_SYMBOL(cpu_present_map); |
4386 | ||
4387 | #ifndef CONFIG_SMP | |
4cef0c61 | 4388 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4389 | EXPORT_SYMBOL(cpu_online_map); |
4390 | ||
4cef0c61 | 4391 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4392 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4393 | #endif |
4394 | ||
4395 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4396 | { | |
36c8b586 | 4397 | struct task_struct *p; |
1da177e4 | 4398 | int retval; |
1da177e4 | 4399 | |
5be9361c | 4400 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4401 | read_lock(&tasklist_lock); |
4402 | ||
4403 | retval = -ESRCH; | |
4404 | p = find_process_by_pid(pid); | |
4405 | if (!p) | |
4406 | goto out_unlock; | |
4407 | ||
e7834f8f DQ |
4408 | retval = security_task_getscheduler(p); |
4409 | if (retval) | |
4410 | goto out_unlock; | |
4411 | ||
2f7016d9 | 4412 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4413 | |
4414 | out_unlock: | |
4415 | read_unlock(&tasklist_lock); | |
5be9361c | 4416 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4417 | |
9531b62f | 4418 | return retval; |
1da177e4 LT |
4419 | } |
4420 | ||
4421 | /** | |
4422 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4423 | * @pid: pid of the process | |
4424 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4425 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4426 | */ | |
4427 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4428 | unsigned long __user *user_mask_ptr) | |
4429 | { | |
4430 | int ret; | |
4431 | cpumask_t mask; | |
4432 | ||
4433 | if (len < sizeof(cpumask_t)) | |
4434 | return -EINVAL; | |
4435 | ||
4436 | ret = sched_getaffinity(pid, &mask); | |
4437 | if (ret < 0) | |
4438 | return ret; | |
4439 | ||
4440 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4441 | return -EFAULT; | |
4442 | ||
4443 | return sizeof(cpumask_t); | |
4444 | } | |
4445 | ||
4446 | /** | |
4447 | * sys_sched_yield - yield the current processor to other threads. | |
4448 | * | |
dd41f596 IM |
4449 | * This function yields the current CPU to other tasks. If there are no |
4450 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4451 | */ |
4452 | asmlinkage long sys_sched_yield(void) | |
4453 | { | |
70b97a7f | 4454 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4455 | |
2d72376b | 4456 | schedstat_inc(rq, yld_count); |
4530d7ab | 4457 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4458 | |
4459 | /* | |
4460 | * Since we are going to call schedule() anyway, there's | |
4461 | * no need to preempt or enable interrupts: | |
4462 | */ | |
4463 | __release(rq->lock); | |
8a25d5de | 4464 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4465 | _raw_spin_unlock(&rq->lock); |
4466 | preempt_enable_no_resched(); | |
4467 | ||
4468 | schedule(); | |
4469 | ||
4470 | return 0; | |
4471 | } | |
4472 | ||
e7b38404 | 4473 | static void __cond_resched(void) |
1da177e4 | 4474 | { |
8e0a43d8 IM |
4475 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4476 | __might_sleep(__FILE__, __LINE__); | |
4477 | #endif | |
5bbcfd90 IM |
4478 | /* |
4479 | * The BKS might be reacquired before we have dropped | |
4480 | * PREEMPT_ACTIVE, which could trigger a second | |
4481 | * cond_resched() call. | |
4482 | */ | |
1da177e4 LT |
4483 | do { |
4484 | add_preempt_count(PREEMPT_ACTIVE); | |
4485 | schedule(); | |
4486 | sub_preempt_count(PREEMPT_ACTIVE); | |
4487 | } while (need_resched()); | |
4488 | } | |
4489 | ||
4490 | int __sched cond_resched(void) | |
4491 | { | |
9414232f IM |
4492 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4493 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4494 | __cond_resched(); |
4495 | return 1; | |
4496 | } | |
4497 | return 0; | |
4498 | } | |
1da177e4 LT |
4499 | EXPORT_SYMBOL(cond_resched); |
4500 | ||
4501 | /* | |
4502 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4503 | * call schedule, and on return reacquire the lock. | |
4504 | * | |
4505 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4506 | * operations here to prevent schedule() from being called twice (once via | |
4507 | * spin_unlock(), once by hand). | |
4508 | */ | |
95cdf3b7 | 4509 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4510 | { |
6df3cecb JK |
4511 | int ret = 0; |
4512 | ||
1da177e4 LT |
4513 | if (need_lockbreak(lock)) { |
4514 | spin_unlock(lock); | |
4515 | cpu_relax(); | |
6df3cecb | 4516 | ret = 1; |
1da177e4 LT |
4517 | spin_lock(lock); |
4518 | } | |
9414232f | 4519 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4520 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4521 | _raw_spin_unlock(lock); |
4522 | preempt_enable_no_resched(); | |
4523 | __cond_resched(); | |
6df3cecb | 4524 | ret = 1; |
1da177e4 | 4525 | spin_lock(lock); |
1da177e4 | 4526 | } |
6df3cecb | 4527 | return ret; |
1da177e4 | 4528 | } |
1da177e4 LT |
4529 | EXPORT_SYMBOL(cond_resched_lock); |
4530 | ||
4531 | int __sched cond_resched_softirq(void) | |
4532 | { | |
4533 | BUG_ON(!in_softirq()); | |
4534 | ||
9414232f | 4535 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4536 | local_bh_enable(); |
1da177e4 LT |
4537 | __cond_resched(); |
4538 | local_bh_disable(); | |
4539 | return 1; | |
4540 | } | |
4541 | return 0; | |
4542 | } | |
1da177e4 LT |
4543 | EXPORT_SYMBOL(cond_resched_softirq); |
4544 | ||
1da177e4 LT |
4545 | /** |
4546 | * yield - yield the current processor to other threads. | |
4547 | * | |
72fd4a35 | 4548 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4549 | * thread runnable and calls sys_sched_yield(). |
4550 | */ | |
4551 | void __sched yield(void) | |
4552 | { | |
4553 | set_current_state(TASK_RUNNING); | |
4554 | sys_sched_yield(); | |
4555 | } | |
1da177e4 LT |
4556 | EXPORT_SYMBOL(yield); |
4557 | ||
4558 | /* | |
4559 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4560 | * that process accounting knows that this is a task in IO wait state. | |
4561 | * | |
4562 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4563 | * has set its backing_dev_info: the queue against which it should throttle) | |
4564 | */ | |
4565 | void __sched io_schedule(void) | |
4566 | { | |
70b97a7f | 4567 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4568 | |
0ff92245 | 4569 | delayacct_blkio_start(); |
1da177e4 LT |
4570 | atomic_inc(&rq->nr_iowait); |
4571 | schedule(); | |
4572 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4573 | delayacct_blkio_end(); |
1da177e4 | 4574 | } |
1da177e4 LT |
4575 | EXPORT_SYMBOL(io_schedule); |
4576 | ||
4577 | long __sched io_schedule_timeout(long timeout) | |
4578 | { | |
70b97a7f | 4579 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4580 | long ret; |
4581 | ||
0ff92245 | 4582 | delayacct_blkio_start(); |
1da177e4 LT |
4583 | atomic_inc(&rq->nr_iowait); |
4584 | ret = schedule_timeout(timeout); | |
4585 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4586 | delayacct_blkio_end(); |
1da177e4 LT |
4587 | return ret; |
4588 | } | |
4589 | ||
4590 | /** | |
4591 | * sys_sched_get_priority_max - return maximum RT priority. | |
4592 | * @policy: scheduling class. | |
4593 | * | |
4594 | * this syscall returns the maximum rt_priority that can be used | |
4595 | * by a given scheduling class. | |
4596 | */ | |
4597 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4598 | { | |
4599 | int ret = -EINVAL; | |
4600 | ||
4601 | switch (policy) { | |
4602 | case SCHED_FIFO: | |
4603 | case SCHED_RR: | |
4604 | ret = MAX_USER_RT_PRIO-1; | |
4605 | break; | |
4606 | case SCHED_NORMAL: | |
b0a9499c | 4607 | case SCHED_BATCH: |
dd41f596 | 4608 | case SCHED_IDLE: |
1da177e4 LT |
4609 | ret = 0; |
4610 | break; | |
4611 | } | |
4612 | return ret; | |
4613 | } | |
4614 | ||
4615 | /** | |
4616 | * sys_sched_get_priority_min - return minimum RT priority. | |
4617 | * @policy: scheduling class. | |
4618 | * | |
4619 | * this syscall returns the minimum rt_priority that can be used | |
4620 | * by a given scheduling class. | |
4621 | */ | |
4622 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4623 | { | |
4624 | int ret = -EINVAL; | |
4625 | ||
4626 | switch (policy) { | |
4627 | case SCHED_FIFO: | |
4628 | case SCHED_RR: | |
4629 | ret = 1; | |
4630 | break; | |
4631 | case SCHED_NORMAL: | |
b0a9499c | 4632 | case SCHED_BATCH: |
dd41f596 | 4633 | case SCHED_IDLE: |
1da177e4 LT |
4634 | ret = 0; |
4635 | } | |
4636 | return ret; | |
4637 | } | |
4638 | ||
4639 | /** | |
4640 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4641 | * @pid: pid of the process. | |
4642 | * @interval: userspace pointer to the timeslice value. | |
4643 | * | |
4644 | * this syscall writes the default timeslice value of a given process | |
4645 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4646 | */ | |
4647 | asmlinkage | |
4648 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4649 | { | |
36c8b586 | 4650 | struct task_struct *p; |
a4ec24b4 | 4651 | unsigned int time_slice; |
3a5c359a | 4652 | int retval; |
1da177e4 | 4653 | struct timespec t; |
1da177e4 LT |
4654 | |
4655 | if (pid < 0) | |
3a5c359a | 4656 | return -EINVAL; |
1da177e4 LT |
4657 | |
4658 | retval = -ESRCH; | |
4659 | read_lock(&tasklist_lock); | |
4660 | p = find_process_by_pid(pid); | |
4661 | if (!p) | |
4662 | goto out_unlock; | |
4663 | ||
4664 | retval = security_task_getscheduler(p); | |
4665 | if (retval) | |
4666 | goto out_unlock; | |
4667 | ||
a4ec24b4 DA |
4668 | if (p->policy == SCHED_FIFO) |
4669 | time_slice = 0; | |
4670 | else if (p->policy == SCHED_RR) | |
4671 | time_slice = DEF_TIMESLICE; | |
4672 | else { | |
4673 | struct sched_entity *se = &p->se; | |
4674 | unsigned long flags; | |
4675 | struct rq *rq; | |
4676 | ||
4677 | rq = task_rq_lock(p, &flags); | |
4678 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4679 | task_rq_unlock(rq, &flags); | |
4680 | } | |
1da177e4 | 4681 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4682 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4683 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4684 | return retval; |
3a5c359a | 4685 | |
1da177e4 LT |
4686 | out_unlock: |
4687 | read_unlock(&tasklist_lock); | |
4688 | return retval; | |
4689 | } | |
4690 | ||
2ed6e34f | 4691 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4692 | |
4693 | static void show_task(struct task_struct *p) | |
1da177e4 | 4694 | { |
1da177e4 | 4695 | unsigned long free = 0; |
36c8b586 | 4696 | unsigned state; |
1da177e4 | 4697 | |
1da177e4 | 4698 | state = p->state ? __ffs(p->state) + 1 : 0; |
2ed6e34f AM |
4699 | printk("%-13.13s %c", p->comm, |
4700 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); | |
4bd77321 | 4701 | #if BITS_PER_LONG == 32 |
1da177e4 | 4702 | if (state == TASK_RUNNING) |
4bd77321 | 4703 | printk(" running "); |
1da177e4 | 4704 | else |
4bd77321 | 4705 | printk(" %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4706 | #else |
4707 | if (state == TASK_RUNNING) | |
4bd77321 | 4708 | printk(" running task "); |
1da177e4 LT |
4709 | else |
4710 | printk(" %016lx ", thread_saved_pc(p)); | |
4711 | #endif | |
4712 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4713 | { | |
10ebffde | 4714 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4715 | while (!*n) |
4716 | n++; | |
10ebffde | 4717 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4718 | } |
4719 | #endif | |
4bd77321 | 4720 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4721 | |
4722 | if (state != TASK_RUNNING) | |
4723 | show_stack(p, NULL); | |
4724 | } | |
4725 | ||
e59e2ae2 | 4726 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4727 | { |
36c8b586 | 4728 | struct task_struct *g, *p; |
1da177e4 | 4729 | |
4bd77321 IM |
4730 | #if BITS_PER_LONG == 32 |
4731 | printk(KERN_INFO | |
4732 | " task PC stack pid father\n"); | |
1da177e4 | 4733 | #else |
4bd77321 IM |
4734 | printk(KERN_INFO |
4735 | " task PC stack pid father\n"); | |
1da177e4 LT |
4736 | #endif |
4737 | read_lock(&tasklist_lock); | |
4738 | do_each_thread(g, p) { | |
4739 | /* | |
4740 | * reset the NMI-timeout, listing all files on a slow | |
4741 | * console might take alot of time: | |
4742 | */ | |
4743 | touch_nmi_watchdog(); | |
39bc89fd | 4744 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4745 | show_task(p); |
1da177e4 LT |
4746 | } while_each_thread(g, p); |
4747 | ||
04c9167f JF |
4748 | touch_all_softlockup_watchdogs(); |
4749 | ||
dd41f596 IM |
4750 | #ifdef CONFIG_SCHED_DEBUG |
4751 | sysrq_sched_debug_show(); | |
4752 | #endif | |
1da177e4 | 4753 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4754 | /* |
4755 | * Only show locks if all tasks are dumped: | |
4756 | */ | |
4757 | if (state_filter == -1) | |
4758 | debug_show_all_locks(); | |
1da177e4 LT |
4759 | } |
4760 | ||
1df21055 IM |
4761 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4762 | { | |
dd41f596 | 4763 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4764 | } |
4765 | ||
f340c0d1 IM |
4766 | /** |
4767 | * init_idle - set up an idle thread for a given CPU | |
4768 | * @idle: task in question | |
4769 | * @cpu: cpu the idle task belongs to | |
4770 | * | |
4771 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4772 | * flag, to make booting more robust. | |
4773 | */ | |
5c1e1767 | 4774 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4775 | { |
70b97a7f | 4776 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4777 | unsigned long flags; |
4778 | ||
dd41f596 IM |
4779 | __sched_fork(idle); |
4780 | idle->se.exec_start = sched_clock(); | |
4781 | ||
b29739f9 | 4782 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4783 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4784 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4785 | |
4786 | spin_lock_irqsave(&rq->lock, flags); | |
4787 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4788 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4789 | idle->oncpu = 1; | |
4790 | #endif | |
1da177e4 LT |
4791 | spin_unlock_irqrestore(&rq->lock, flags); |
4792 | ||
4793 | /* Set the preempt count _outside_ the spinlocks! */ | |
4794 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4795 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4796 | #else |
a1261f54 | 4797 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4798 | #endif |
dd41f596 IM |
4799 | /* |
4800 | * The idle tasks have their own, simple scheduling class: | |
4801 | */ | |
4802 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4803 | } |
4804 | ||
4805 | /* | |
4806 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4807 | * indicates which cpus entered this state. This is used | |
4808 | * in the rcu update to wait only for active cpus. For system | |
4809 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4810 | * always be CPU_MASK_NONE. | |
4811 | */ | |
4812 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4813 | ||
4814 | #ifdef CONFIG_SMP | |
4815 | /* | |
4816 | * This is how migration works: | |
4817 | * | |
70b97a7f | 4818 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4819 | * runqueue and wake up that CPU's migration thread. |
4820 | * 2) we down() the locked semaphore => thread blocks. | |
4821 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4822 | * thread off the CPU) | |
4823 | * 4) it gets the migration request and checks whether the migrated | |
4824 | * task is still in the wrong runqueue. | |
4825 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4826 | * it and puts it into the right queue. | |
4827 | * 6) migration thread up()s the semaphore. | |
4828 | * 7) we wake up and the migration is done. | |
4829 | */ | |
4830 | ||
4831 | /* | |
4832 | * Change a given task's CPU affinity. Migrate the thread to a | |
4833 | * proper CPU and schedule it away if the CPU it's executing on | |
4834 | * is removed from the allowed bitmask. | |
4835 | * | |
4836 | * NOTE: the caller must have a valid reference to the task, the | |
4837 | * task must not exit() & deallocate itself prematurely. The | |
4838 | * call is not atomic; no spinlocks may be held. | |
4839 | */ | |
36c8b586 | 4840 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4841 | { |
70b97a7f | 4842 | struct migration_req req; |
1da177e4 | 4843 | unsigned long flags; |
70b97a7f | 4844 | struct rq *rq; |
48f24c4d | 4845 | int ret = 0; |
1da177e4 LT |
4846 | |
4847 | rq = task_rq_lock(p, &flags); | |
4848 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4849 | ret = -EINVAL; | |
4850 | goto out; | |
4851 | } | |
4852 | ||
4853 | p->cpus_allowed = new_mask; | |
4854 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4855 | if (cpu_isset(task_cpu(p), new_mask)) | |
4856 | goto out; | |
4857 | ||
4858 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4859 | /* Need help from migration thread: drop lock and wait. */ | |
4860 | task_rq_unlock(rq, &flags); | |
4861 | wake_up_process(rq->migration_thread); | |
4862 | wait_for_completion(&req.done); | |
4863 | tlb_migrate_finish(p->mm); | |
4864 | return 0; | |
4865 | } | |
4866 | out: | |
4867 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4868 | |
1da177e4 LT |
4869 | return ret; |
4870 | } | |
1da177e4 LT |
4871 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4872 | ||
4873 | /* | |
4874 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
4875 | * this because either it can't run here any more (set_cpus_allowed() | |
4876 | * away from this CPU, or CPU going down), or because we're | |
4877 | * attempting to rebalance this task on exec (sched_exec). | |
4878 | * | |
4879 | * So we race with normal scheduler movements, but that's OK, as long | |
4880 | * as the task is no longer on this CPU. | |
efc30814 KK |
4881 | * |
4882 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 4883 | */ |
efc30814 | 4884 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 4885 | { |
70b97a7f | 4886 | struct rq *rq_dest, *rq_src; |
dd41f596 | 4887 | int ret = 0, on_rq; |
1da177e4 LT |
4888 | |
4889 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 4890 | return ret; |
1da177e4 LT |
4891 | |
4892 | rq_src = cpu_rq(src_cpu); | |
4893 | rq_dest = cpu_rq(dest_cpu); | |
4894 | ||
4895 | double_rq_lock(rq_src, rq_dest); | |
4896 | /* Already moved. */ | |
4897 | if (task_cpu(p) != src_cpu) | |
4898 | goto out; | |
4899 | /* Affinity changed (again). */ | |
4900 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
4901 | goto out; | |
4902 | ||
dd41f596 | 4903 | on_rq = p->se.on_rq; |
6e82a3be | 4904 | if (on_rq) |
2e1cb74a | 4905 | deactivate_task(rq_src, p, 0); |
6e82a3be | 4906 | |
1da177e4 | 4907 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
4908 | if (on_rq) { |
4909 | activate_task(rq_dest, p, 0); | |
4910 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 4911 | } |
efc30814 | 4912 | ret = 1; |
1da177e4 LT |
4913 | out: |
4914 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 4915 | return ret; |
1da177e4 LT |
4916 | } |
4917 | ||
4918 | /* | |
4919 | * migration_thread - this is a highprio system thread that performs | |
4920 | * thread migration by bumping thread off CPU then 'pushing' onto | |
4921 | * another runqueue. | |
4922 | */ | |
95cdf3b7 | 4923 | static int migration_thread(void *data) |
1da177e4 | 4924 | { |
1da177e4 | 4925 | int cpu = (long)data; |
70b97a7f | 4926 | struct rq *rq; |
1da177e4 LT |
4927 | |
4928 | rq = cpu_rq(cpu); | |
4929 | BUG_ON(rq->migration_thread != current); | |
4930 | ||
4931 | set_current_state(TASK_INTERRUPTIBLE); | |
4932 | while (!kthread_should_stop()) { | |
70b97a7f | 4933 | struct migration_req *req; |
1da177e4 | 4934 | struct list_head *head; |
1da177e4 | 4935 | |
1da177e4 LT |
4936 | spin_lock_irq(&rq->lock); |
4937 | ||
4938 | if (cpu_is_offline(cpu)) { | |
4939 | spin_unlock_irq(&rq->lock); | |
4940 | goto wait_to_die; | |
4941 | } | |
4942 | ||
4943 | if (rq->active_balance) { | |
4944 | active_load_balance(rq, cpu); | |
4945 | rq->active_balance = 0; | |
4946 | } | |
4947 | ||
4948 | head = &rq->migration_queue; | |
4949 | ||
4950 | if (list_empty(head)) { | |
4951 | spin_unlock_irq(&rq->lock); | |
4952 | schedule(); | |
4953 | set_current_state(TASK_INTERRUPTIBLE); | |
4954 | continue; | |
4955 | } | |
70b97a7f | 4956 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
4957 | list_del_init(head->next); |
4958 | ||
674311d5 NP |
4959 | spin_unlock(&rq->lock); |
4960 | __migrate_task(req->task, cpu, req->dest_cpu); | |
4961 | local_irq_enable(); | |
1da177e4 LT |
4962 | |
4963 | complete(&req->done); | |
4964 | } | |
4965 | __set_current_state(TASK_RUNNING); | |
4966 | return 0; | |
4967 | ||
4968 | wait_to_die: | |
4969 | /* Wait for kthread_stop */ | |
4970 | set_current_state(TASK_INTERRUPTIBLE); | |
4971 | while (!kthread_should_stop()) { | |
4972 | schedule(); | |
4973 | set_current_state(TASK_INTERRUPTIBLE); | |
4974 | } | |
4975 | __set_current_state(TASK_RUNNING); | |
4976 | return 0; | |
4977 | } | |
4978 | ||
4979 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 KK |
4980 | /* |
4981 | * Figure out where task on dead CPU should go, use force if neccessary. | |
4982 | * NOTE: interrupts should be disabled by the caller | |
4983 | */ | |
48f24c4d | 4984 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 4985 | { |
efc30814 | 4986 | unsigned long flags; |
1da177e4 | 4987 | cpumask_t mask; |
70b97a7f IM |
4988 | struct rq *rq; |
4989 | int dest_cpu; | |
1da177e4 | 4990 | |
3a5c359a AK |
4991 | do { |
4992 | /* On same node? */ | |
4993 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
4994 | cpus_and(mask, mask, p->cpus_allowed); | |
4995 | dest_cpu = any_online_cpu(mask); | |
4996 | ||
4997 | /* On any allowed CPU? */ | |
4998 | if (dest_cpu == NR_CPUS) | |
4999 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5000 | ||
5001 | /* No more Mr. Nice Guy. */ | |
5002 | if (dest_cpu == NR_CPUS) { | |
5003 | rq = task_rq_lock(p, &flags); | |
5004 | cpus_setall(p->cpus_allowed); | |
5005 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5006 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5007 | |
3a5c359a AK |
5008 | /* |
5009 | * Don't tell them about moving exiting tasks or | |
5010 | * kernel threads (both mm NULL), since they never | |
5011 | * leave kernel. | |
5012 | */ | |
5013 | if (p->mm && printk_ratelimit()) | |
5014 | printk(KERN_INFO "process %d (%s) no " | |
5015 | "longer affine to cpu%d\n", | |
5016 | p->pid, p->comm, dead_cpu); | |
5017 | } | |
5018 | } while (!__migrate_task(p, dead_cpu, dest_cpu)); | |
1da177e4 LT |
5019 | } |
5020 | ||
5021 | /* | |
5022 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5023 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5024 | * for performance reasons the counter is not stricly tracking tasks to | |
5025 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5026 | * to keep the global sum constant after CPU-down: | |
5027 | */ | |
70b97a7f | 5028 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5029 | { |
70b97a7f | 5030 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5031 | unsigned long flags; |
5032 | ||
5033 | local_irq_save(flags); | |
5034 | double_rq_lock(rq_src, rq_dest); | |
5035 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5036 | rq_src->nr_uninterruptible = 0; | |
5037 | double_rq_unlock(rq_src, rq_dest); | |
5038 | local_irq_restore(flags); | |
5039 | } | |
5040 | ||
5041 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5042 | static void migrate_live_tasks(int src_cpu) | |
5043 | { | |
48f24c4d | 5044 | struct task_struct *p, *t; |
1da177e4 LT |
5045 | |
5046 | write_lock_irq(&tasklist_lock); | |
5047 | ||
48f24c4d IM |
5048 | do_each_thread(t, p) { |
5049 | if (p == current) | |
1da177e4 LT |
5050 | continue; |
5051 | ||
48f24c4d IM |
5052 | if (task_cpu(p) == src_cpu) |
5053 | move_task_off_dead_cpu(src_cpu, p); | |
5054 | } while_each_thread(t, p); | |
1da177e4 LT |
5055 | |
5056 | write_unlock_irq(&tasklist_lock); | |
5057 | } | |
5058 | ||
a9957449 AD |
5059 | /* |
5060 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5061 | */ | |
5062 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5063 | { | |
5064 | update_rq_clock(rq); | |
5065 | ||
5066 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5067 | rq->nr_uninterruptible--; | |
5068 | ||
5069 | enqueue_task(rq, p, 0); | |
5070 | inc_nr_running(p, rq); | |
5071 | } | |
5072 | ||
dd41f596 IM |
5073 | /* |
5074 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5075 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5076 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5077 | */ |
5078 | void sched_idle_next(void) | |
5079 | { | |
48f24c4d | 5080 | int this_cpu = smp_processor_id(); |
70b97a7f | 5081 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5082 | struct task_struct *p = rq->idle; |
5083 | unsigned long flags; | |
5084 | ||
5085 | /* cpu has to be offline */ | |
48f24c4d | 5086 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5087 | |
48f24c4d IM |
5088 | /* |
5089 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5090 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5091 | */ |
5092 | spin_lock_irqsave(&rq->lock, flags); | |
5093 | ||
dd41f596 | 5094 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5095 | |
5096 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5097 | activate_idle_task(p, rq); |
1da177e4 LT |
5098 | |
5099 | spin_unlock_irqrestore(&rq->lock, flags); | |
5100 | } | |
5101 | ||
48f24c4d IM |
5102 | /* |
5103 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5104 | * offline. |
5105 | */ | |
5106 | void idle_task_exit(void) | |
5107 | { | |
5108 | struct mm_struct *mm = current->active_mm; | |
5109 | ||
5110 | BUG_ON(cpu_online(smp_processor_id())); | |
5111 | ||
5112 | if (mm != &init_mm) | |
5113 | switch_mm(mm, &init_mm, current); | |
5114 | mmdrop(mm); | |
5115 | } | |
5116 | ||
054b9108 | 5117 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5118 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5119 | { |
70b97a7f | 5120 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5121 | |
5122 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5123 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5124 | |
5125 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5126 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5127 | |
48f24c4d | 5128 | get_task_struct(p); |
1da177e4 LT |
5129 | |
5130 | /* | |
5131 | * Drop lock around migration; if someone else moves it, | |
5132 | * that's OK. No task can be added to this CPU, so iteration is | |
5133 | * fine. | |
054b9108 | 5134 | * NOTE: interrupts should be left disabled --dev@ |
1da177e4 | 5135 | */ |
054b9108 | 5136 | spin_unlock(&rq->lock); |
48f24c4d | 5137 | move_task_off_dead_cpu(dead_cpu, p); |
054b9108 | 5138 | spin_lock(&rq->lock); |
1da177e4 | 5139 | |
48f24c4d | 5140 | put_task_struct(p); |
1da177e4 LT |
5141 | } |
5142 | ||
5143 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5144 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5145 | { | |
70b97a7f | 5146 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5147 | struct task_struct *next; |
48f24c4d | 5148 | |
dd41f596 IM |
5149 | for ( ; ; ) { |
5150 | if (!rq->nr_running) | |
5151 | break; | |
a8e504d2 | 5152 | update_rq_clock(rq); |
ff95f3df | 5153 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5154 | if (!next) |
5155 | break; | |
5156 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5157 | |
1da177e4 LT |
5158 | } |
5159 | } | |
5160 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5161 | ||
e692ab53 NP |
5162 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5163 | ||
5164 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5165 | { |
5166 | .procname = "sched_domain", | |
c57baf1e | 5167 | .mode = 0555, |
e0361851 | 5168 | }, |
e692ab53 NP |
5169 | {0,}, |
5170 | }; | |
5171 | ||
5172 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5173 | { |
c57baf1e | 5174 | .ctl_name = CTL_KERN, |
e0361851 | 5175 | .procname = "kernel", |
c57baf1e | 5176 | .mode = 0555, |
e0361851 AD |
5177 | .child = sd_ctl_dir, |
5178 | }, | |
e692ab53 NP |
5179 | {0,}, |
5180 | }; | |
5181 | ||
5182 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5183 | { | |
5184 | struct ctl_table *entry = | |
5185 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); | |
5186 | ||
5187 | BUG_ON(!entry); | |
5188 | memset(entry, 0, n * sizeof(struct ctl_table)); | |
5189 | ||
5190 | return entry; | |
5191 | } | |
5192 | ||
5193 | static void | |
e0361851 | 5194 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5195 | const char *procname, void *data, int maxlen, |
5196 | mode_t mode, proc_handler *proc_handler) | |
5197 | { | |
e692ab53 NP |
5198 | entry->procname = procname; |
5199 | entry->data = data; | |
5200 | entry->maxlen = maxlen; | |
5201 | entry->mode = mode; | |
5202 | entry->proc_handler = proc_handler; | |
5203 | } | |
5204 | ||
5205 | static struct ctl_table * | |
5206 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5207 | { | |
ace8b3d6 | 5208 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5209 | |
e0361851 | 5210 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5211 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5212 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5213 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5214 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5215 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5216 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5217 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5218 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5219 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5220 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5221 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5222 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5223 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5224 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5225 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5226 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5227 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5228 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5229 | &sd->cache_nice_tries, |
5230 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5231 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 NP |
5232 | sizeof(int), 0644, proc_dointvec_minmax); |
5233 | ||
5234 | return table; | |
5235 | } | |
5236 | ||
5237 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) | |
5238 | { | |
5239 | struct ctl_table *entry, *table; | |
5240 | struct sched_domain *sd; | |
5241 | int domain_num = 0, i; | |
5242 | char buf[32]; | |
5243 | ||
5244 | for_each_domain(cpu, sd) | |
5245 | domain_num++; | |
5246 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
5247 | ||
5248 | i = 0; | |
5249 | for_each_domain(cpu, sd) { | |
5250 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5251 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5252 | entry->mode = 0555; |
e692ab53 NP |
5253 | entry->child = sd_alloc_ctl_domain_table(sd); |
5254 | entry++; | |
5255 | i++; | |
5256 | } | |
5257 | return table; | |
5258 | } | |
5259 | ||
5260 | static struct ctl_table_header *sd_sysctl_header; | |
5261 | static void init_sched_domain_sysctl(void) | |
5262 | { | |
5263 | int i, cpu_num = num_online_cpus(); | |
5264 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5265 | char buf[32]; | |
5266 | ||
5267 | sd_ctl_dir[0].child = entry; | |
5268 | ||
5269 | for (i = 0; i < cpu_num; i++, entry++) { | |
5270 | snprintf(buf, 32, "cpu%d", i); | |
e692ab53 | 5271 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5272 | entry->mode = 0555; |
e692ab53 NP |
5273 | entry->child = sd_alloc_ctl_cpu_table(i); |
5274 | } | |
5275 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5276 | } | |
5277 | #else | |
5278 | static void init_sched_domain_sysctl(void) | |
5279 | { | |
5280 | } | |
5281 | #endif | |
5282 | ||
1da177e4 LT |
5283 | /* |
5284 | * migration_call - callback that gets triggered when a CPU is added. | |
5285 | * Here we can start up the necessary migration thread for the new CPU. | |
5286 | */ | |
48f24c4d IM |
5287 | static int __cpuinit |
5288 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5289 | { |
1da177e4 | 5290 | struct task_struct *p; |
48f24c4d | 5291 | int cpu = (long)hcpu; |
1da177e4 | 5292 | unsigned long flags; |
70b97a7f | 5293 | struct rq *rq; |
1da177e4 LT |
5294 | |
5295 | switch (action) { | |
5be9361c GS |
5296 | case CPU_LOCK_ACQUIRE: |
5297 | mutex_lock(&sched_hotcpu_mutex); | |
5298 | break; | |
5299 | ||
1da177e4 | 5300 | case CPU_UP_PREPARE: |
8bb78442 | 5301 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5302 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5303 | if (IS_ERR(p)) |
5304 | return NOTIFY_BAD; | |
1da177e4 LT |
5305 | kthread_bind(p, cpu); |
5306 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5307 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5308 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5309 | task_rq_unlock(rq, &flags); |
5310 | cpu_rq(cpu)->migration_thread = p; | |
5311 | break; | |
48f24c4d | 5312 | |
1da177e4 | 5313 | case CPU_ONLINE: |
8bb78442 | 5314 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5315 | /* Strictly unneccessary, as first user will wake it. */ |
5316 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5317 | break; | |
48f24c4d | 5318 | |
1da177e4 LT |
5319 | #ifdef CONFIG_HOTPLUG_CPU |
5320 | case CPU_UP_CANCELED: | |
8bb78442 | 5321 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5322 | if (!cpu_rq(cpu)->migration_thread) |
5323 | break; | |
1da177e4 | 5324 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5325 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5326 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5327 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5328 | cpu_rq(cpu)->migration_thread = NULL; | |
5329 | break; | |
48f24c4d | 5330 | |
1da177e4 | 5331 | case CPU_DEAD: |
8bb78442 | 5332 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5333 | migrate_live_tasks(cpu); |
5334 | rq = cpu_rq(cpu); | |
5335 | kthread_stop(rq->migration_thread); | |
5336 | rq->migration_thread = NULL; | |
5337 | /* Idle task back to normal (off runqueue, low prio) */ | |
5338 | rq = task_rq_lock(rq->idle, &flags); | |
a8e504d2 | 5339 | update_rq_clock(rq); |
2e1cb74a | 5340 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5341 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5342 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5343 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5344 | migrate_dead_tasks(cpu); |
5345 | task_rq_unlock(rq, &flags); | |
5346 | migrate_nr_uninterruptible(rq); | |
5347 | BUG_ON(rq->nr_running != 0); | |
5348 | ||
5349 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5350 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5351 | * the requestors. */ |
5352 | spin_lock_irq(&rq->lock); | |
5353 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5354 | struct migration_req *req; |
5355 | ||
1da177e4 | 5356 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5357 | struct migration_req, list); |
1da177e4 LT |
5358 | list_del_init(&req->list); |
5359 | complete(&req->done); | |
5360 | } | |
5361 | spin_unlock_irq(&rq->lock); | |
5362 | break; | |
5363 | #endif | |
5be9361c GS |
5364 | case CPU_LOCK_RELEASE: |
5365 | mutex_unlock(&sched_hotcpu_mutex); | |
5366 | break; | |
1da177e4 LT |
5367 | } |
5368 | return NOTIFY_OK; | |
5369 | } | |
5370 | ||
5371 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5372 | * happens before everything else. | |
5373 | */ | |
26c2143b | 5374 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5375 | .notifier_call = migration_call, |
5376 | .priority = 10 | |
5377 | }; | |
5378 | ||
5379 | int __init migration_init(void) | |
5380 | { | |
5381 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5382 | int err; |
48f24c4d IM |
5383 | |
5384 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5385 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5386 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5387 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5388 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5389 | |
1da177e4 LT |
5390 | return 0; |
5391 | } | |
5392 | #endif | |
5393 | ||
5394 | #ifdef CONFIG_SMP | |
476f3534 CL |
5395 | |
5396 | /* Number of possible processor ids */ | |
5397 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5398 | EXPORT_SYMBOL(nr_cpu_ids); | |
5399 | ||
3e9830dc | 5400 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5401 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5402 | { | |
5403 | int level = 0; | |
5404 | ||
41c7ce9a NP |
5405 | if (!sd) { |
5406 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5407 | return; | |
5408 | } | |
5409 | ||
1da177e4 LT |
5410 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5411 | ||
5412 | do { | |
5413 | int i; | |
5414 | char str[NR_CPUS]; | |
5415 | struct sched_group *group = sd->groups; | |
5416 | cpumask_t groupmask; | |
5417 | ||
5418 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5419 | cpus_clear(groupmask); | |
5420 | ||
5421 | printk(KERN_DEBUG); | |
5422 | for (i = 0; i < level + 1; i++) | |
5423 | printk(" "); | |
5424 | printk("domain %d: ", level); | |
5425 | ||
5426 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5427 | printk("does not load-balance\n"); | |
5428 | if (sd->parent) | |
33859f7f MOS |
5429 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5430 | " has parent"); | |
1da177e4 LT |
5431 | break; |
5432 | } | |
5433 | ||
5434 | printk("span %s\n", str); | |
5435 | ||
5436 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5437 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5438 | "CPU%d\n", cpu); | |
1da177e4 | 5439 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5440 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5441 | " CPU%d\n", cpu); | |
1da177e4 LT |
5442 | |
5443 | printk(KERN_DEBUG); | |
5444 | for (i = 0; i < level + 2; i++) | |
5445 | printk(" "); | |
5446 | printk("groups:"); | |
5447 | do { | |
5448 | if (!group) { | |
5449 | printk("\n"); | |
5450 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5451 | break; | |
5452 | } | |
5453 | ||
5517d86b | 5454 | if (!group->__cpu_power) { |
1da177e4 | 5455 | printk("\n"); |
33859f7f MOS |
5456 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5457 | "set\n"); | |
26797a34 | 5458 | break; |
1da177e4 LT |
5459 | } |
5460 | ||
5461 | if (!cpus_weight(group->cpumask)) { | |
5462 | printk("\n"); | |
5463 | printk(KERN_ERR "ERROR: empty group\n"); | |
26797a34 | 5464 | break; |
1da177e4 LT |
5465 | } |
5466 | ||
5467 | if (cpus_intersects(groupmask, group->cpumask)) { | |
5468 | printk("\n"); | |
5469 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
26797a34 | 5470 | break; |
1da177e4 LT |
5471 | } |
5472 | ||
5473 | cpus_or(groupmask, groupmask, group->cpumask); | |
5474 | ||
5475 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
5476 | printk(" %s", str); | |
5477 | ||
5478 | group = group->next; | |
5479 | } while (group != sd->groups); | |
5480 | printk("\n"); | |
5481 | ||
5482 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5483 | printk(KERN_ERR "ERROR: groups don't span " |
5484 | "domain->span\n"); | |
1da177e4 LT |
5485 | |
5486 | level++; | |
5487 | sd = sd->parent; | |
33859f7f MOS |
5488 | if (!sd) |
5489 | continue; | |
1da177e4 | 5490 | |
33859f7f MOS |
5491 | if (!cpus_subset(groupmask, sd->span)) |
5492 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5493 | "of domain->span\n"); | |
1da177e4 LT |
5494 | |
5495 | } while (sd); | |
5496 | } | |
5497 | #else | |
48f24c4d | 5498 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5499 | #endif |
5500 | ||
1a20ff27 | 5501 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5502 | { |
5503 | if (cpus_weight(sd->span) == 1) | |
5504 | return 1; | |
5505 | ||
5506 | /* Following flags need at least 2 groups */ | |
5507 | if (sd->flags & (SD_LOAD_BALANCE | | |
5508 | SD_BALANCE_NEWIDLE | | |
5509 | SD_BALANCE_FORK | | |
89c4710e SS |
5510 | SD_BALANCE_EXEC | |
5511 | SD_SHARE_CPUPOWER | | |
5512 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5513 | if (sd->groups != sd->groups->next) |
5514 | return 0; | |
5515 | } | |
5516 | ||
5517 | /* Following flags don't use groups */ | |
5518 | if (sd->flags & (SD_WAKE_IDLE | | |
5519 | SD_WAKE_AFFINE | | |
5520 | SD_WAKE_BALANCE)) | |
5521 | return 0; | |
5522 | ||
5523 | return 1; | |
5524 | } | |
5525 | ||
48f24c4d IM |
5526 | static int |
5527 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5528 | { |
5529 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5530 | ||
5531 | if (sd_degenerate(parent)) | |
5532 | return 1; | |
5533 | ||
5534 | if (!cpus_equal(sd->span, parent->span)) | |
5535 | return 0; | |
5536 | ||
5537 | /* Does parent contain flags not in child? */ | |
5538 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5539 | if (cflags & SD_WAKE_AFFINE) | |
5540 | pflags &= ~SD_WAKE_BALANCE; | |
5541 | /* Flags needing groups don't count if only 1 group in parent */ | |
5542 | if (parent->groups == parent->groups->next) { | |
5543 | pflags &= ~(SD_LOAD_BALANCE | | |
5544 | SD_BALANCE_NEWIDLE | | |
5545 | SD_BALANCE_FORK | | |
89c4710e SS |
5546 | SD_BALANCE_EXEC | |
5547 | SD_SHARE_CPUPOWER | | |
5548 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5549 | } |
5550 | if (~cflags & pflags) | |
5551 | return 0; | |
5552 | ||
5553 | return 1; | |
5554 | } | |
5555 | ||
1da177e4 LT |
5556 | /* |
5557 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5558 | * hold the hotplug lock. | |
5559 | */ | |
9c1cfda2 | 5560 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5561 | { |
70b97a7f | 5562 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5563 | struct sched_domain *tmp; |
5564 | ||
5565 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5566 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5567 | struct sched_domain *parent = tmp->parent; | |
5568 | if (!parent) | |
5569 | break; | |
1a848870 | 5570 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5571 | tmp->parent = parent->parent; |
1a848870 SS |
5572 | if (parent->parent) |
5573 | parent->parent->child = tmp; | |
5574 | } | |
245af2c7 SS |
5575 | } |
5576 | ||
1a848870 | 5577 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5578 | sd = sd->parent; |
1a848870 SS |
5579 | if (sd) |
5580 | sd->child = NULL; | |
5581 | } | |
1da177e4 LT |
5582 | |
5583 | sched_domain_debug(sd, cpu); | |
5584 | ||
674311d5 | 5585 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5586 | } |
5587 | ||
5588 | /* cpus with isolated domains */ | |
67af63a6 | 5589 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5590 | |
5591 | /* Setup the mask of cpus configured for isolated domains */ | |
5592 | static int __init isolated_cpu_setup(char *str) | |
5593 | { | |
5594 | int ints[NR_CPUS], i; | |
5595 | ||
5596 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5597 | cpus_clear(cpu_isolated_map); | |
5598 | for (i = 1; i <= ints[0]; i++) | |
5599 | if (ints[i] < NR_CPUS) | |
5600 | cpu_set(ints[i], cpu_isolated_map); | |
5601 | return 1; | |
5602 | } | |
5603 | ||
8927f494 | 5604 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5605 | |
5606 | /* | |
6711cab4 SS |
5607 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5608 | * to a function which identifies what group(along with sched group) a CPU | |
5609 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5610 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5611 | * |
5612 | * init_sched_build_groups will build a circular linked list of the groups | |
5613 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5614 | * and ->cpu_power to 0. | |
5615 | */ | |
a616058b | 5616 | static void |
6711cab4 SS |
5617 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5618 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5619 | struct sched_group **sg)) | |
1da177e4 LT |
5620 | { |
5621 | struct sched_group *first = NULL, *last = NULL; | |
5622 | cpumask_t covered = CPU_MASK_NONE; | |
5623 | int i; | |
5624 | ||
5625 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5626 | struct sched_group *sg; |
5627 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5628 | int j; |
5629 | ||
5630 | if (cpu_isset(i, covered)) | |
5631 | continue; | |
5632 | ||
5633 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5634 | sg->__cpu_power = 0; |
1da177e4 LT |
5635 | |
5636 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5637 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5638 | continue; |
5639 | ||
5640 | cpu_set(j, covered); | |
5641 | cpu_set(j, sg->cpumask); | |
5642 | } | |
5643 | if (!first) | |
5644 | first = sg; | |
5645 | if (last) | |
5646 | last->next = sg; | |
5647 | last = sg; | |
5648 | } | |
5649 | last->next = first; | |
5650 | } | |
5651 | ||
9c1cfda2 | 5652 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5653 | |
9c1cfda2 | 5654 | #ifdef CONFIG_NUMA |
198e2f18 | 5655 | |
9c1cfda2 JH |
5656 | /** |
5657 | * find_next_best_node - find the next node to include in a sched_domain | |
5658 | * @node: node whose sched_domain we're building | |
5659 | * @used_nodes: nodes already in the sched_domain | |
5660 | * | |
5661 | * Find the next node to include in a given scheduling domain. Simply | |
5662 | * finds the closest node not already in the @used_nodes map. | |
5663 | * | |
5664 | * Should use nodemask_t. | |
5665 | */ | |
5666 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5667 | { | |
5668 | int i, n, val, min_val, best_node = 0; | |
5669 | ||
5670 | min_val = INT_MAX; | |
5671 | ||
5672 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5673 | /* Start at @node */ | |
5674 | n = (node + i) % MAX_NUMNODES; | |
5675 | ||
5676 | if (!nr_cpus_node(n)) | |
5677 | continue; | |
5678 | ||
5679 | /* Skip already used nodes */ | |
5680 | if (test_bit(n, used_nodes)) | |
5681 | continue; | |
5682 | ||
5683 | /* Simple min distance search */ | |
5684 | val = node_distance(node, n); | |
5685 | ||
5686 | if (val < min_val) { | |
5687 | min_val = val; | |
5688 | best_node = n; | |
5689 | } | |
5690 | } | |
5691 | ||
5692 | set_bit(best_node, used_nodes); | |
5693 | return best_node; | |
5694 | } | |
5695 | ||
5696 | /** | |
5697 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5698 | * @node: node whose cpumask we're constructing | |
5699 | * @size: number of nodes to include in this span | |
5700 | * | |
5701 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5702 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5703 | * out optimally. | |
5704 | */ | |
5705 | static cpumask_t sched_domain_node_span(int node) | |
5706 | { | |
9c1cfda2 | 5707 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5708 | cpumask_t span, nodemask; |
5709 | int i; | |
9c1cfda2 JH |
5710 | |
5711 | cpus_clear(span); | |
5712 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5713 | ||
5714 | nodemask = node_to_cpumask(node); | |
5715 | cpus_or(span, span, nodemask); | |
5716 | set_bit(node, used_nodes); | |
5717 | ||
5718 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5719 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5720 | |
9c1cfda2 JH |
5721 | nodemask = node_to_cpumask(next_node); |
5722 | cpus_or(span, span, nodemask); | |
5723 | } | |
5724 | ||
5725 | return span; | |
5726 | } | |
5727 | #endif | |
5728 | ||
5c45bf27 | 5729 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5730 | |
9c1cfda2 | 5731 | /* |
48f24c4d | 5732 | * SMT sched-domains: |
9c1cfda2 | 5733 | */ |
1da177e4 LT |
5734 | #ifdef CONFIG_SCHED_SMT |
5735 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5736 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5737 | |
6711cab4 SS |
5738 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5739 | struct sched_group **sg) | |
1da177e4 | 5740 | { |
6711cab4 SS |
5741 | if (sg) |
5742 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5743 | return cpu; |
5744 | } | |
5745 | #endif | |
5746 | ||
48f24c4d IM |
5747 | /* |
5748 | * multi-core sched-domains: | |
5749 | */ | |
1e9f28fa SS |
5750 | #ifdef CONFIG_SCHED_MC |
5751 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5752 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5753 | #endif |
5754 | ||
5755 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5756 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5757 | struct sched_group **sg) | |
1e9f28fa | 5758 | { |
6711cab4 | 5759 | int group; |
a616058b SS |
5760 | cpumask_t mask = cpu_sibling_map[cpu]; |
5761 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 SS |
5762 | group = first_cpu(mask); |
5763 | if (sg) | |
5764 | *sg = &per_cpu(sched_group_core, group); | |
5765 | return group; | |
1e9f28fa SS |
5766 | } |
5767 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5768 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5769 | struct sched_group **sg) | |
1e9f28fa | 5770 | { |
6711cab4 SS |
5771 | if (sg) |
5772 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5773 | return cpu; |
5774 | } | |
5775 | #endif | |
5776 | ||
1da177e4 | 5777 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5778 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5779 | |
6711cab4 SS |
5780 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5781 | struct sched_group **sg) | |
1da177e4 | 5782 | { |
6711cab4 | 5783 | int group; |
48f24c4d | 5784 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5785 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5786 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5787 | group = first_cpu(mask); |
1e9f28fa | 5788 | #elif defined(CONFIG_SCHED_SMT) |
a616058b SS |
5789 | cpumask_t mask = cpu_sibling_map[cpu]; |
5790 | cpus_and(mask, mask, *cpu_map); | |
6711cab4 | 5791 | group = first_cpu(mask); |
1da177e4 | 5792 | #else |
6711cab4 | 5793 | group = cpu; |
1da177e4 | 5794 | #endif |
6711cab4 SS |
5795 | if (sg) |
5796 | *sg = &per_cpu(sched_group_phys, group); | |
5797 | return group; | |
1da177e4 LT |
5798 | } |
5799 | ||
5800 | #ifdef CONFIG_NUMA | |
1da177e4 | 5801 | /* |
9c1cfda2 JH |
5802 | * The init_sched_build_groups can't handle what we want to do with node |
5803 | * groups, so roll our own. Now each node has its own list of groups which | |
5804 | * gets dynamically allocated. | |
1da177e4 | 5805 | */ |
9c1cfda2 | 5806 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5807 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5808 | |
9c1cfda2 | 5809 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5810 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5811 | |
6711cab4 SS |
5812 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5813 | struct sched_group **sg) | |
9c1cfda2 | 5814 | { |
6711cab4 SS |
5815 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5816 | int group; | |
5817 | ||
5818 | cpus_and(nodemask, nodemask, *cpu_map); | |
5819 | group = first_cpu(nodemask); | |
5820 | ||
5821 | if (sg) | |
5822 | *sg = &per_cpu(sched_group_allnodes, group); | |
5823 | return group; | |
1da177e4 | 5824 | } |
6711cab4 | 5825 | |
08069033 SS |
5826 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
5827 | { | |
5828 | struct sched_group *sg = group_head; | |
5829 | int j; | |
5830 | ||
5831 | if (!sg) | |
5832 | return; | |
3a5c359a AK |
5833 | do { |
5834 | for_each_cpu_mask(j, sg->cpumask) { | |
5835 | struct sched_domain *sd; | |
08069033 | 5836 | |
3a5c359a AK |
5837 | sd = &per_cpu(phys_domains, j); |
5838 | if (j != first_cpu(sd->groups->cpumask)) { | |
5839 | /* | |
5840 | * Only add "power" once for each | |
5841 | * physical package. | |
5842 | */ | |
5843 | continue; | |
5844 | } | |
08069033 | 5845 | |
3a5c359a AK |
5846 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
5847 | } | |
5848 | sg = sg->next; | |
5849 | } while (sg != group_head); | |
08069033 | 5850 | } |
1da177e4 LT |
5851 | #endif |
5852 | ||
a616058b | 5853 | #ifdef CONFIG_NUMA |
51888ca2 SV |
5854 | /* Free memory allocated for various sched_group structures */ |
5855 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5856 | { | |
a616058b | 5857 | int cpu, i; |
51888ca2 SV |
5858 | |
5859 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
5860 | struct sched_group **sched_group_nodes |
5861 | = sched_group_nodes_bycpu[cpu]; | |
5862 | ||
51888ca2 SV |
5863 | if (!sched_group_nodes) |
5864 | continue; | |
5865 | ||
5866 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5867 | cpumask_t nodemask = node_to_cpumask(i); | |
5868 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
5869 | ||
5870 | cpus_and(nodemask, nodemask, *cpu_map); | |
5871 | if (cpus_empty(nodemask)) | |
5872 | continue; | |
5873 | ||
5874 | if (sg == NULL) | |
5875 | continue; | |
5876 | sg = sg->next; | |
5877 | next_sg: | |
5878 | oldsg = sg; | |
5879 | sg = sg->next; | |
5880 | kfree(oldsg); | |
5881 | if (oldsg != sched_group_nodes[i]) | |
5882 | goto next_sg; | |
5883 | } | |
5884 | kfree(sched_group_nodes); | |
5885 | sched_group_nodes_bycpu[cpu] = NULL; | |
5886 | } | |
51888ca2 | 5887 | } |
a616058b SS |
5888 | #else |
5889 | static void free_sched_groups(const cpumask_t *cpu_map) | |
5890 | { | |
5891 | } | |
5892 | #endif | |
51888ca2 | 5893 | |
89c4710e SS |
5894 | /* |
5895 | * Initialize sched groups cpu_power. | |
5896 | * | |
5897 | * cpu_power indicates the capacity of sched group, which is used while | |
5898 | * distributing the load between different sched groups in a sched domain. | |
5899 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
5900 | * there are asymmetries in the topology. If there are asymmetries, group | |
5901 | * having more cpu_power will pickup more load compared to the group having | |
5902 | * less cpu_power. | |
5903 | * | |
5904 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
5905 | * the maximum number of tasks a group can handle in the presence of other idle | |
5906 | * or lightly loaded groups in the same sched domain. | |
5907 | */ | |
5908 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
5909 | { | |
5910 | struct sched_domain *child; | |
5911 | struct sched_group *group; | |
5912 | ||
5913 | WARN_ON(!sd || !sd->groups); | |
5914 | ||
5915 | if (cpu != first_cpu(sd->groups->cpumask)) | |
5916 | return; | |
5917 | ||
5918 | child = sd->child; | |
5919 | ||
5517d86b ED |
5920 | sd->groups->__cpu_power = 0; |
5921 | ||
89c4710e SS |
5922 | /* |
5923 | * For perf policy, if the groups in child domain share resources | |
5924 | * (for example cores sharing some portions of the cache hierarchy | |
5925 | * or SMT), then set this domain groups cpu_power such that each group | |
5926 | * can handle only one task, when there are other idle groups in the | |
5927 | * same sched domain. | |
5928 | */ | |
5929 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
5930 | (child->flags & | |
5931 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 5932 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
5933 | return; |
5934 | } | |
5935 | ||
89c4710e SS |
5936 | /* |
5937 | * add cpu_power of each child group to this groups cpu_power | |
5938 | */ | |
5939 | group = child->groups; | |
5940 | do { | |
5517d86b | 5941 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
5942 | group = group->next; |
5943 | } while (group != child->groups); | |
5944 | } | |
5945 | ||
1da177e4 | 5946 | /* |
1a20ff27 DG |
5947 | * Build sched domains for a given set of cpus and attach the sched domains |
5948 | * to the individual cpus | |
1da177e4 | 5949 | */ |
51888ca2 | 5950 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
5951 | { |
5952 | int i; | |
d1b55138 JH |
5953 | #ifdef CONFIG_NUMA |
5954 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 5955 | int sd_allnodes = 0; |
d1b55138 JH |
5956 | |
5957 | /* | |
5958 | * Allocate the per-node list of sched groups | |
5959 | */ | |
dd41f596 | 5960 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
d3a5aa98 | 5961 | GFP_KERNEL); |
d1b55138 JH |
5962 | if (!sched_group_nodes) { |
5963 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 5964 | return -ENOMEM; |
d1b55138 JH |
5965 | } |
5966 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
5967 | #endif | |
1da177e4 LT |
5968 | |
5969 | /* | |
1a20ff27 | 5970 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 5971 | */ |
1a20ff27 | 5972 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
5973 | struct sched_domain *sd = NULL, *p; |
5974 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
5975 | ||
1a20ff27 | 5976 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
5977 | |
5978 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
5979 | if (cpus_weight(*cpu_map) > |
5980 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
5981 | sd = &per_cpu(allnodes_domains, i); |
5982 | *sd = SD_ALLNODES_INIT; | |
5983 | sd->span = *cpu_map; | |
6711cab4 | 5984 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 5985 | p = sd; |
6711cab4 | 5986 | sd_allnodes = 1; |
9c1cfda2 JH |
5987 | } else |
5988 | p = NULL; | |
5989 | ||
1da177e4 | 5990 | sd = &per_cpu(node_domains, i); |
1da177e4 | 5991 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
5992 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
5993 | sd->parent = p; | |
1a848870 SS |
5994 | if (p) |
5995 | p->child = sd; | |
9c1cfda2 | 5996 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
5997 | #endif |
5998 | ||
5999 | p = sd; | |
6000 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6001 | *sd = SD_CPU_INIT; |
6002 | sd->span = nodemask; | |
6003 | sd->parent = p; | |
1a848870 SS |
6004 | if (p) |
6005 | p->child = sd; | |
6711cab4 | 6006 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6007 | |
1e9f28fa SS |
6008 | #ifdef CONFIG_SCHED_MC |
6009 | p = sd; | |
6010 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6011 | *sd = SD_MC_INIT; |
6012 | sd->span = cpu_coregroup_map(i); | |
6013 | cpus_and(sd->span, sd->span, *cpu_map); | |
6014 | sd->parent = p; | |
1a848870 | 6015 | p->child = sd; |
6711cab4 | 6016 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6017 | #endif |
6018 | ||
1da177e4 LT |
6019 | #ifdef CONFIG_SCHED_SMT |
6020 | p = sd; | |
6021 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 LT |
6022 | *sd = SD_SIBLING_INIT; |
6023 | sd->span = cpu_sibling_map[i]; | |
1a20ff27 | 6024 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6025 | sd->parent = p; |
1a848870 | 6026 | p->child = sd; |
6711cab4 | 6027 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6028 | #endif |
6029 | } | |
6030 | ||
6031 | #ifdef CONFIG_SCHED_SMT | |
6032 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6033 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 6034 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
1a20ff27 | 6035 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6036 | if (i != first_cpu(this_sibling_map)) |
6037 | continue; | |
6038 | ||
dd41f596 IM |
6039 | init_sched_build_groups(this_sibling_map, cpu_map, |
6040 | &cpu_to_cpu_group); | |
1da177e4 LT |
6041 | } |
6042 | #endif | |
6043 | ||
1e9f28fa SS |
6044 | #ifdef CONFIG_SCHED_MC |
6045 | /* Set up multi-core groups */ | |
6046 | for_each_cpu_mask(i, *cpu_map) { | |
6047 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6048 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6049 | if (i != first_cpu(this_core_map)) | |
6050 | continue; | |
dd41f596 IM |
6051 | init_sched_build_groups(this_core_map, cpu_map, |
6052 | &cpu_to_core_group); | |
1e9f28fa SS |
6053 | } |
6054 | #endif | |
6055 | ||
1da177e4 LT |
6056 | /* Set up physical groups */ |
6057 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6058 | cpumask_t nodemask = node_to_cpumask(i); | |
6059 | ||
1a20ff27 | 6060 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6061 | if (cpus_empty(nodemask)) |
6062 | continue; | |
6063 | ||
6711cab4 | 6064 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6065 | } |
6066 | ||
6067 | #ifdef CONFIG_NUMA | |
6068 | /* Set up node groups */ | |
6711cab4 | 6069 | if (sd_allnodes) |
dd41f596 IM |
6070 | init_sched_build_groups(*cpu_map, cpu_map, |
6071 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6072 | |
6073 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6074 | /* Set up node groups */ | |
6075 | struct sched_group *sg, *prev; | |
6076 | cpumask_t nodemask = node_to_cpumask(i); | |
6077 | cpumask_t domainspan; | |
6078 | cpumask_t covered = CPU_MASK_NONE; | |
6079 | int j; | |
6080 | ||
6081 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6082 | if (cpus_empty(nodemask)) { |
6083 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6084 | continue; |
d1b55138 | 6085 | } |
9c1cfda2 JH |
6086 | |
6087 | domainspan = sched_domain_node_span(i); | |
6088 | cpus_and(domainspan, domainspan, *cpu_map); | |
6089 | ||
15f0b676 | 6090 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6091 | if (!sg) { |
6092 | printk(KERN_WARNING "Can not alloc domain group for " | |
6093 | "node %d\n", i); | |
6094 | goto error; | |
6095 | } | |
9c1cfda2 JH |
6096 | sched_group_nodes[i] = sg; |
6097 | for_each_cpu_mask(j, nodemask) { | |
6098 | struct sched_domain *sd; | |
9761eea8 | 6099 | |
9c1cfda2 JH |
6100 | sd = &per_cpu(node_domains, j); |
6101 | sd->groups = sg; | |
9c1cfda2 | 6102 | } |
5517d86b | 6103 | sg->__cpu_power = 0; |
9c1cfda2 | 6104 | sg->cpumask = nodemask; |
51888ca2 | 6105 | sg->next = sg; |
9c1cfda2 JH |
6106 | cpus_or(covered, covered, nodemask); |
6107 | prev = sg; | |
6108 | ||
6109 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6110 | cpumask_t tmp, notcovered; | |
6111 | int n = (i + j) % MAX_NUMNODES; | |
6112 | ||
6113 | cpus_complement(notcovered, covered); | |
6114 | cpus_and(tmp, notcovered, *cpu_map); | |
6115 | cpus_and(tmp, tmp, domainspan); | |
6116 | if (cpus_empty(tmp)) | |
6117 | break; | |
6118 | ||
6119 | nodemask = node_to_cpumask(n); | |
6120 | cpus_and(tmp, tmp, nodemask); | |
6121 | if (cpus_empty(tmp)) | |
6122 | continue; | |
6123 | ||
15f0b676 SV |
6124 | sg = kmalloc_node(sizeof(struct sched_group), |
6125 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6126 | if (!sg) { |
6127 | printk(KERN_WARNING | |
6128 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6129 | goto error; |
9c1cfda2 | 6130 | } |
5517d86b | 6131 | sg->__cpu_power = 0; |
9c1cfda2 | 6132 | sg->cpumask = tmp; |
51888ca2 | 6133 | sg->next = prev->next; |
9c1cfda2 JH |
6134 | cpus_or(covered, covered, tmp); |
6135 | prev->next = sg; | |
6136 | prev = sg; | |
6137 | } | |
9c1cfda2 | 6138 | } |
1da177e4 LT |
6139 | #endif |
6140 | ||
6141 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6142 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6143 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6144 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6145 | ||
89c4710e | 6146 | init_sched_groups_power(i, sd); |
5c45bf27 | 6147 | } |
1da177e4 | 6148 | #endif |
1e9f28fa | 6149 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6150 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6151 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6152 | ||
89c4710e | 6153 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6154 | } |
6155 | #endif | |
1e9f28fa | 6156 | |
5c45bf27 | 6157 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6158 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6159 | ||
89c4710e | 6160 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6161 | } |
6162 | ||
9c1cfda2 | 6163 | #ifdef CONFIG_NUMA |
08069033 SS |
6164 | for (i = 0; i < MAX_NUMNODES; i++) |
6165 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6166 | |
6711cab4 SS |
6167 | if (sd_allnodes) { |
6168 | struct sched_group *sg; | |
f712c0c7 | 6169 | |
6711cab4 | 6170 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6171 | init_numa_sched_groups_power(sg); |
6172 | } | |
9c1cfda2 JH |
6173 | #endif |
6174 | ||
1da177e4 | 6175 | /* Attach the domains */ |
1a20ff27 | 6176 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6177 | struct sched_domain *sd; |
6178 | #ifdef CONFIG_SCHED_SMT | |
6179 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6180 | #elif defined(CONFIG_SCHED_MC) |
6181 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6182 | #else |
6183 | sd = &per_cpu(phys_domains, i); | |
6184 | #endif | |
6185 | cpu_attach_domain(sd, i); | |
6186 | } | |
51888ca2 SV |
6187 | |
6188 | return 0; | |
6189 | ||
a616058b | 6190 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6191 | error: |
6192 | free_sched_groups(cpu_map); | |
6193 | return -ENOMEM; | |
a616058b | 6194 | #endif |
1da177e4 | 6195 | } |
1a20ff27 DG |
6196 | /* |
6197 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6198 | */ | |
51888ca2 | 6199 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6200 | { |
6201 | cpumask_t cpu_default_map; | |
51888ca2 | 6202 | int err; |
1da177e4 | 6203 | |
1a20ff27 DG |
6204 | /* |
6205 | * Setup mask for cpus without special case scheduling requirements. | |
6206 | * For now this just excludes isolated cpus, but could be used to | |
6207 | * exclude other special cases in the future. | |
6208 | */ | |
6209 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6210 | ||
51888ca2 SV |
6211 | err = build_sched_domains(&cpu_default_map); |
6212 | ||
6213 | return err; | |
1a20ff27 DG |
6214 | } |
6215 | ||
6216 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6217 | { |
51888ca2 | 6218 | free_sched_groups(cpu_map); |
9c1cfda2 | 6219 | } |
1da177e4 | 6220 | |
1a20ff27 DG |
6221 | /* |
6222 | * Detach sched domains from a group of cpus specified in cpu_map | |
6223 | * These cpus will now be attached to the NULL domain | |
6224 | */ | |
858119e1 | 6225 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6226 | { |
6227 | int i; | |
6228 | ||
6229 | for_each_cpu_mask(i, *cpu_map) | |
6230 | cpu_attach_domain(NULL, i); | |
6231 | synchronize_sched(); | |
6232 | arch_destroy_sched_domains(cpu_map); | |
6233 | } | |
6234 | ||
6235 | /* | |
6236 | * Partition sched domains as specified by the cpumasks below. | |
6237 | * This attaches all cpus from the cpumasks to the NULL domain, | |
6238 | * waits for a RCU quiescent period, recalculates sched | |
6239 | * domain information and then attaches them back to the | |
6240 | * correct sched domains | |
6241 | * Call with hotplug lock held | |
6242 | */ | |
51888ca2 | 6243 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
1a20ff27 DG |
6244 | { |
6245 | cpumask_t change_map; | |
51888ca2 | 6246 | int err = 0; |
1a20ff27 DG |
6247 | |
6248 | cpus_and(*partition1, *partition1, cpu_online_map); | |
6249 | cpus_and(*partition2, *partition2, cpu_online_map); | |
6250 | cpus_or(change_map, *partition1, *partition2); | |
6251 | ||
6252 | /* Detach sched domains from all of the affected cpus */ | |
6253 | detach_destroy_domains(&change_map); | |
6254 | if (!cpus_empty(*partition1)) | |
51888ca2 SV |
6255 | err = build_sched_domains(partition1); |
6256 | if (!err && !cpus_empty(*partition2)) | |
6257 | err = build_sched_domains(partition2); | |
6258 | ||
6259 | return err; | |
1a20ff27 DG |
6260 | } |
6261 | ||
5c45bf27 | 6262 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6263 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6264 | { |
6265 | int err; | |
6266 | ||
5be9361c | 6267 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6268 | detach_destroy_domains(&cpu_online_map); |
6269 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6270 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6271 | |
6272 | return err; | |
6273 | } | |
6274 | ||
6275 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6276 | { | |
6277 | int ret; | |
6278 | ||
6279 | if (buf[0] != '0' && buf[0] != '1') | |
6280 | return -EINVAL; | |
6281 | ||
6282 | if (smt) | |
6283 | sched_smt_power_savings = (buf[0] == '1'); | |
6284 | else | |
6285 | sched_mc_power_savings = (buf[0] == '1'); | |
6286 | ||
6287 | ret = arch_reinit_sched_domains(); | |
6288 | ||
6289 | return ret ? ret : count; | |
6290 | } | |
6291 | ||
5c45bf27 SS |
6292 | #ifdef CONFIG_SCHED_MC |
6293 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6294 | { | |
6295 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6296 | } | |
48f24c4d IM |
6297 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6298 | const char *buf, size_t count) | |
5c45bf27 SS |
6299 | { |
6300 | return sched_power_savings_store(buf, count, 0); | |
6301 | } | |
6707de00 AB |
6302 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6303 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6304 | #endif |
6305 | ||
6306 | #ifdef CONFIG_SCHED_SMT | |
6307 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6308 | { | |
6309 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6310 | } | |
48f24c4d IM |
6311 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6312 | const char *buf, size_t count) | |
5c45bf27 SS |
6313 | { |
6314 | return sched_power_savings_store(buf, count, 1); | |
6315 | } | |
6707de00 AB |
6316 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6317 | sched_smt_power_savings_store); | |
6318 | #endif | |
6319 | ||
6320 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6321 | { | |
6322 | int err = 0; | |
6323 | ||
6324 | #ifdef CONFIG_SCHED_SMT | |
6325 | if (smt_capable()) | |
6326 | err = sysfs_create_file(&cls->kset.kobj, | |
6327 | &attr_sched_smt_power_savings.attr); | |
6328 | #endif | |
6329 | #ifdef CONFIG_SCHED_MC | |
6330 | if (!err && mc_capable()) | |
6331 | err = sysfs_create_file(&cls->kset.kobj, | |
6332 | &attr_sched_mc_power_savings.attr); | |
6333 | #endif | |
6334 | return err; | |
6335 | } | |
5c45bf27 SS |
6336 | #endif |
6337 | ||
1da177e4 LT |
6338 | /* |
6339 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6340 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6341 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6342 | * which will prevent rebalancing while the sched domains are recalculated. |
6343 | */ | |
6344 | static int update_sched_domains(struct notifier_block *nfb, | |
6345 | unsigned long action, void *hcpu) | |
6346 | { | |
1da177e4 LT |
6347 | switch (action) { |
6348 | case CPU_UP_PREPARE: | |
8bb78442 | 6349 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6350 | case CPU_DOWN_PREPARE: |
8bb78442 | 6351 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6352 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6353 | return NOTIFY_OK; |
6354 | ||
6355 | case CPU_UP_CANCELED: | |
8bb78442 | 6356 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6357 | case CPU_DOWN_FAILED: |
8bb78442 | 6358 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6359 | case CPU_ONLINE: |
8bb78442 | 6360 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6361 | case CPU_DEAD: |
8bb78442 | 6362 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6363 | /* |
6364 | * Fall through and re-initialise the domains. | |
6365 | */ | |
6366 | break; | |
6367 | default: | |
6368 | return NOTIFY_DONE; | |
6369 | } | |
6370 | ||
6371 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6372 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6373 | |
6374 | return NOTIFY_OK; | |
6375 | } | |
1da177e4 LT |
6376 | |
6377 | void __init sched_init_smp(void) | |
6378 | { | |
5c1e1767 NP |
6379 | cpumask_t non_isolated_cpus; |
6380 | ||
5be9361c | 6381 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6382 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6383 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6384 | if (cpus_empty(non_isolated_cpus)) |
6385 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6386 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6387 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6388 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 | 6389 | |
e692ab53 NP |
6390 | init_sched_domain_sysctl(); |
6391 | ||
5c1e1767 NP |
6392 | /* Move init over to a non-isolated CPU */ |
6393 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6394 | BUG(); | |
1da177e4 LT |
6395 | } |
6396 | #else | |
6397 | void __init sched_init_smp(void) | |
6398 | { | |
6399 | } | |
6400 | #endif /* CONFIG_SMP */ | |
6401 | ||
6402 | int in_sched_functions(unsigned long addr) | |
6403 | { | |
6404 | /* Linker adds these: start and end of __sched functions */ | |
6405 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6406 | |
1da177e4 LT |
6407 | return in_lock_functions(addr) || |
6408 | (addr >= (unsigned long)__sched_text_start | |
6409 | && addr < (unsigned long)__sched_text_end); | |
6410 | } | |
6411 | ||
a9957449 | 6412 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6413 | { |
6414 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6415 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6416 | cfs_rq->rq = rq; | |
6417 | #endif | |
67e9fb2a | 6418 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6419 | } |
6420 | ||
1da177e4 LT |
6421 | void __init sched_init(void) |
6422 | { | |
476f3534 | 6423 | int highest_cpu = 0; |
dd41f596 IM |
6424 | int i, j; |
6425 | ||
0a945022 | 6426 | for_each_possible_cpu(i) { |
dd41f596 | 6427 | struct rt_prio_array *array; |
70b97a7f | 6428 | struct rq *rq; |
1da177e4 LT |
6429 | |
6430 | rq = cpu_rq(i); | |
6431 | spin_lock_init(&rq->lock); | |
fcb99371 | 6432 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6433 | rq->nr_running = 0; |
dd41f596 IM |
6434 | rq->clock = 1; |
6435 | init_cfs_rq(&rq->cfs, rq); | |
6436 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6437 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6438 | { |
6439 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6440 | struct sched_entity *se = | |
6441 | &per_cpu(init_sched_entity, i); | |
6442 | ||
6443 | init_cfs_rq_p[i] = cfs_rq; | |
6444 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6445 | cfs_rq->tg = &init_task_group; |
3a252015 | 6446 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6447 | &rq->leaf_cfs_rq_list); |
6448 | ||
3a252015 IM |
6449 | init_sched_entity_p[i] = se; |
6450 | se->cfs_rq = &rq->cfs; | |
6451 | se->my_q = cfs_rq; | |
4cf86d77 | 6452 | se->load.weight = init_task_group_load; |
9b5b7751 | 6453 | se->load.inv_weight = |
4cf86d77 | 6454 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6455 | se->parent = NULL; |
6456 | } | |
4cf86d77 | 6457 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6458 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6459 | #endif |
1da177e4 | 6460 | |
dd41f596 IM |
6461 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6462 | rq->cpu_load[j] = 0; | |
1da177e4 | 6463 | #ifdef CONFIG_SMP |
41c7ce9a | 6464 | rq->sd = NULL; |
1da177e4 | 6465 | rq->active_balance = 0; |
dd41f596 | 6466 | rq->next_balance = jiffies; |
1da177e4 | 6467 | rq->push_cpu = 0; |
0a2966b4 | 6468 | rq->cpu = i; |
1da177e4 LT |
6469 | rq->migration_thread = NULL; |
6470 | INIT_LIST_HEAD(&rq->migration_queue); | |
6471 | #endif | |
6472 | atomic_set(&rq->nr_iowait, 0); | |
6473 | ||
dd41f596 IM |
6474 | array = &rq->rt.active; |
6475 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6476 | INIT_LIST_HEAD(array->queue + j); | |
6477 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6478 | } |
476f3534 | 6479 | highest_cpu = i; |
dd41f596 IM |
6480 | /* delimiter for bitsearch: */ |
6481 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6482 | } |
6483 | ||
2dd73a4f | 6484 | set_load_weight(&init_task); |
b50f60ce | 6485 | |
e107be36 AK |
6486 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6487 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6488 | #endif | |
6489 | ||
c9819f45 | 6490 | #ifdef CONFIG_SMP |
476f3534 | 6491 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6492 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6493 | #endif | |
6494 | ||
b50f60ce HC |
6495 | #ifdef CONFIG_RT_MUTEXES |
6496 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6497 | #endif | |
6498 | ||
1da177e4 LT |
6499 | /* |
6500 | * The boot idle thread does lazy MMU switching as well: | |
6501 | */ | |
6502 | atomic_inc(&init_mm.mm_count); | |
6503 | enter_lazy_tlb(&init_mm, current); | |
6504 | ||
6505 | /* | |
6506 | * Make us the idle thread. Technically, schedule() should not be | |
6507 | * called from this thread, however somewhere below it might be, | |
6508 | * but because we are the idle thread, we just pick up running again | |
6509 | * when this runqueue becomes "idle". | |
6510 | */ | |
6511 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6512 | /* |
6513 | * During early bootup we pretend to be a normal task: | |
6514 | */ | |
6515 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6516 | } |
6517 | ||
6518 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6519 | void __might_sleep(char *file, int line) | |
6520 | { | |
48f24c4d | 6521 | #ifdef in_atomic |
1da177e4 LT |
6522 | static unsigned long prev_jiffy; /* ratelimiting */ |
6523 | ||
6524 | if ((in_atomic() || irqs_disabled()) && | |
6525 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6526 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6527 | return; | |
6528 | prev_jiffy = jiffies; | |
91368d73 | 6529 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6530 | " context at %s:%d\n", file, line); |
6531 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6532 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6533 | debug_show_held_locks(current); |
3117df04 IM |
6534 | if (irqs_disabled()) |
6535 | print_irqtrace_events(current); | |
1da177e4 LT |
6536 | dump_stack(); |
6537 | } | |
6538 | #endif | |
6539 | } | |
6540 | EXPORT_SYMBOL(__might_sleep); | |
6541 | #endif | |
6542 | ||
6543 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6544 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6545 | { | |
6546 | int on_rq; | |
6547 | update_rq_clock(rq); | |
6548 | on_rq = p->se.on_rq; | |
6549 | if (on_rq) | |
6550 | deactivate_task(rq, p, 0); | |
6551 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6552 | if (on_rq) { | |
6553 | activate_task(rq, p, 0); | |
6554 | resched_task(rq->curr); | |
6555 | } | |
6556 | } | |
6557 | ||
1da177e4 LT |
6558 | void normalize_rt_tasks(void) |
6559 | { | |
a0f98a1c | 6560 | struct task_struct *g, *p; |
1da177e4 | 6561 | unsigned long flags; |
70b97a7f | 6562 | struct rq *rq; |
1da177e4 LT |
6563 | |
6564 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6565 | do_each_thread(g, p) { |
6cfb0d5d | 6566 | p->se.exec_start = 0; |
6cfb0d5d | 6567 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6568 | p->se.wait_start = 0; |
dd41f596 | 6569 | p->se.sleep_start = 0; |
dd41f596 | 6570 | p->se.block_start = 0; |
6cfb0d5d | 6571 | #endif |
dd41f596 IM |
6572 | task_rq(p)->clock = 0; |
6573 | ||
6574 | if (!rt_task(p)) { | |
6575 | /* | |
6576 | * Renice negative nice level userspace | |
6577 | * tasks back to 0: | |
6578 | */ | |
6579 | if (TASK_NICE(p) < 0 && p->mm) | |
6580 | set_user_nice(p, 0); | |
1da177e4 | 6581 | continue; |
dd41f596 | 6582 | } |
1da177e4 | 6583 | |
b29739f9 IM |
6584 | spin_lock_irqsave(&p->pi_lock, flags); |
6585 | rq = __task_rq_lock(p); | |
1da177e4 | 6586 | |
3a5e4dc1 AK |
6587 | if (!is_migration_thread(p, rq)) |
6588 | normalize_task(rq, p); | |
6589 | ||
b29739f9 IM |
6590 | __task_rq_unlock(rq); |
6591 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6592 | } while_each_thread(g, p); |
6593 | ||
1da177e4 LT |
6594 | read_unlock_irq(&tasklist_lock); |
6595 | } | |
6596 | ||
6597 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6598 | |
6599 | #ifdef CONFIG_IA64 | |
6600 | /* | |
6601 | * These functions are only useful for the IA64 MCA handling. | |
6602 | * | |
6603 | * They can only be called when the whole system has been | |
6604 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6605 | * activity can take place. Using them for anything else would | |
6606 | * be a serious bug, and as a result, they aren't even visible | |
6607 | * under any other configuration. | |
6608 | */ | |
6609 | ||
6610 | /** | |
6611 | * curr_task - return the current task for a given cpu. | |
6612 | * @cpu: the processor in question. | |
6613 | * | |
6614 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6615 | */ | |
36c8b586 | 6616 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6617 | { |
6618 | return cpu_curr(cpu); | |
6619 | } | |
6620 | ||
6621 | /** | |
6622 | * set_curr_task - set the current task for a given cpu. | |
6623 | * @cpu: the processor in question. | |
6624 | * @p: the task pointer to set. | |
6625 | * | |
6626 | * Description: This function must only be used when non-maskable interrupts | |
6627 | * are serviced on a separate stack. It allows the architecture to switch the | |
6628 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6629 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6630 | * and caller must save the original value of the current task (see | |
6631 | * curr_task() above) and restore that value before reenabling interrupts and | |
6632 | * re-starting the system. | |
6633 | * | |
6634 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6635 | */ | |
36c8b586 | 6636 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6637 | { |
6638 | cpu_curr(cpu) = p; | |
6639 | } | |
6640 | ||
6641 | #endif | |
29f59db3 SV |
6642 | |
6643 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6644 | ||
29f59db3 | 6645 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6646 | struct task_group *sched_create_group(void) |
29f59db3 | 6647 | { |
4cf86d77 | 6648 | struct task_group *tg; |
29f59db3 SV |
6649 | struct cfs_rq *cfs_rq; |
6650 | struct sched_entity *se; | |
9b5b7751 | 6651 | struct rq *rq; |
29f59db3 SV |
6652 | int i; |
6653 | ||
29f59db3 SV |
6654 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6655 | if (!tg) | |
6656 | return ERR_PTR(-ENOMEM); | |
6657 | ||
9b5b7751 | 6658 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6659 | if (!tg->cfs_rq) |
6660 | goto err; | |
9b5b7751 | 6661 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6662 | if (!tg->se) |
6663 | goto err; | |
6664 | ||
6665 | for_each_possible_cpu(i) { | |
9b5b7751 | 6666 | rq = cpu_rq(i); |
29f59db3 SV |
6667 | |
6668 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6669 | cpu_to_node(i)); | |
6670 | if (!cfs_rq) | |
6671 | goto err; | |
6672 | ||
6673 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6674 | cpu_to_node(i)); | |
6675 | if (!se) | |
6676 | goto err; | |
6677 | ||
6678 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6679 | memset(se, 0, sizeof(struct sched_entity)); | |
6680 | ||
6681 | tg->cfs_rq[i] = cfs_rq; | |
6682 | init_cfs_rq(cfs_rq, rq); | |
6683 | cfs_rq->tg = tg; | |
29f59db3 SV |
6684 | |
6685 | tg->se[i] = se; | |
6686 | se->cfs_rq = &rq->cfs; | |
6687 | se->my_q = cfs_rq; | |
6688 | se->load.weight = NICE_0_LOAD; | |
6689 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6690 | se->parent = NULL; | |
6691 | } | |
6692 | ||
9b5b7751 SV |
6693 | for_each_possible_cpu(i) { |
6694 | rq = cpu_rq(i); | |
6695 | cfs_rq = tg->cfs_rq[i]; | |
6696 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6697 | } | |
29f59db3 | 6698 | |
9b5b7751 | 6699 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6700 | spin_lock_init(&tg->lock); |
29f59db3 | 6701 | |
9b5b7751 | 6702 | return tg; |
29f59db3 SV |
6703 | |
6704 | err: | |
6705 | for_each_possible_cpu(i) { | |
a65914b3 | 6706 | if (tg->cfs_rq) |
29f59db3 | 6707 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6708 | if (tg->se) |
29f59db3 SV |
6709 | kfree(tg->se[i]); |
6710 | } | |
a65914b3 IM |
6711 | kfree(tg->cfs_rq); |
6712 | kfree(tg->se); | |
6713 | kfree(tg); | |
29f59db3 SV |
6714 | |
6715 | return ERR_PTR(-ENOMEM); | |
6716 | } | |
6717 | ||
9b5b7751 SV |
6718 | /* rcu callback to free various structures associated with a task group */ |
6719 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6720 | { |
9b5b7751 | 6721 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6722 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6723 | struct sched_entity *se; |
6724 | int i; | |
6725 | ||
29f59db3 SV |
6726 | /* now it should be safe to free those cfs_rqs */ |
6727 | for_each_possible_cpu(i) { | |
6728 | cfs_rq = tg->cfs_rq[i]; | |
6729 | kfree(cfs_rq); | |
6730 | ||
6731 | se = tg->se[i]; | |
6732 | kfree(se); | |
6733 | } | |
6734 | ||
6735 | kfree(tg->cfs_rq); | |
6736 | kfree(tg->se); | |
6737 | kfree(tg); | |
6738 | } | |
6739 | ||
9b5b7751 | 6740 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6741 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 6742 | { |
9b5b7751 SV |
6743 | struct cfs_rq *cfs_rq; |
6744 | int i; | |
29f59db3 | 6745 | |
9b5b7751 SV |
6746 | for_each_possible_cpu(i) { |
6747 | cfs_rq = tg->cfs_rq[i]; | |
6748 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
6749 | } | |
6750 | ||
6751 | cfs_rq = tg->cfs_rq[0]; | |
6752 | ||
6753 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6754 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
6755 | } |
6756 | ||
9b5b7751 | 6757 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6758 | * The caller of this function should have put the task in its new group |
6759 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6760 | * reflect its new group. | |
9b5b7751 SV |
6761 | */ |
6762 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
6763 | { |
6764 | int on_rq, running; | |
6765 | unsigned long flags; | |
6766 | struct rq *rq; | |
6767 | ||
6768 | rq = task_rq_lock(tsk, &flags); | |
6769 | ||
6770 | if (tsk->sched_class != &fair_sched_class) | |
6771 | goto done; | |
6772 | ||
6773 | update_rq_clock(rq); | |
6774 | ||
6775 | running = task_running(rq, tsk); | |
6776 | on_rq = tsk->se.on_rq; | |
6777 | ||
83b699ed | 6778 | if (on_rq) { |
29f59db3 | 6779 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
6780 | if (unlikely(running)) |
6781 | tsk->sched_class->put_prev_task(rq, tsk); | |
6782 | } | |
29f59db3 SV |
6783 | |
6784 | set_task_cfs_rq(tsk); | |
6785 | ||
83b699ed SV |
6786 | if (on_rq) { |
6787 | if (unlikely(running)) | |
6788 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 6789 | enqueue_task(rq, tsk, 0); |
83b699ed | 6790 | } |
29f59db3 SV |
6791 | |
6792 | done: | |
6793 | task_rq_unlock(rq, &flags); | |
6794 | } | |
6795 | ||
6796 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
6797 | { | |
6798 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
6799 | struct rq *rq = cfs_rq->rq; | |
6800 | int on_rq; | |
6801 | ||
6802 | spin_lock_irq(&rq->lock); | |
6803 | ||
6804 | on_rq = se->on_rq; | |
6805 | if (on_rq) | |
6806 | dequeue_entity(cfs_rq, se, 0); | |
6807 | ||
6808 | se->load.weight = shares; | |
6809 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
6810 | ||
6811 | if (on_rq) | |
6812 | enqueue_entity(cfs_rq, se, 0); | |
6813 | ||
6814 | spin_unlock_irq(&rq->lock); | |
6815 | } | |
6816 | ||
4cf86d77 | 6817 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
6818 | { |
6819 | int i; | |
29f59db3 | 6820 | |
5cb350ba | 6821 | spin_lock(&tg->lock); |
9b5b7751 | 6822 | if (tg->shares == shares) |
5cb350ba | 6823 | goto done; |
29f59db3 | 6824 | |
9b5b7751 | 6825 | tg->shares = shares; |
29f59db3 | 6826 | for_each_possible_cpu(i) |
9b5b7751 | 6827 | set_se_shares(tg->se[i], shares); |
29f59db3 | 6828 | |
5cb350ba DG |
6829 | done: |
6830 | spin_unlock(&tg->lock); | |
9b5b7751 | 6831 | return 0; |
29f59db3 SV |
6832 | } |
6833 | ||
5cb350ba DG |
6834 | unsigned long sched_group_shares(struct task_group *tg) |
6835 | { | |
6836 | return tg->shares; | |
6837 | } | |
6838 | ||
3a252015 | 6839 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |