]>
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> | |
62d0df64 | 54 | #include <linux/cpu_acct.h> |
1da177e4 LT |
55 | #include <linux/kthread.h> |
56 | #include <linux/seq_file.h> | |
e692ab53 | 57 | #include <linux/sysctl.h> |
1da177e4 LT |
58 | #include <linux/syscalls.h> |
59 | #include <linux/times.h> | |
8f0ab514 | 60 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 61 | #include <linux/kprobes.h> |
0ff92245 | 62 | #include <linux/delayacct.h> |
5517d86b | 63 | #include <linux/reciprocal_div.h> |
dff06c15 | 64 | #include <linux/unistd.h> |
f5ff8422 | 65 | #include <linux/pagemap.h> |
1da177e4 | 66 | |
5517d86b | 67 | #include <asm/tlb.h> |
1da177e4 | 68 | |
b035b6de AD |
69 | /* |
70 | * Scheduler clock - returns current time in nanosec units. | |
71 | * This is default implementation. | |
72 | * Architectures and sub-architectures can override this. | |
73 | */ | |
74 | unsigned long long __attribute__((weak)) sched_clock(void) | |
75 | { | |
76 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
77 | } | |
78 | ||
1da177e4 LT |
79 | /* |
80 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
81 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
82 | * and back. | |
83 | */ | |
84 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
85 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
86 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
87 | ||
88 | /* | |
89 | * 'User priority' is the nice value converted to something we | |
90 | * can work with better when scaling various scheduler parameters, | |
91 | * it's a [ 0 ... 39 ] range. | |
92 | */ | |
93 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
94 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
95 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
96 | ||
97 | /* | |
98 | * Some helpers for converting nanosecond timing to jiffy resolution | |
99 | */ | |
a4ec24b4 | 100 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
101 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
102 | ||
6aa645ea IM |
103 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
104 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
105 | ||
1da177e4 LT |
106 | /* |
107 | * These are the 'tuning knobs' of the scheduler: | |
108 | * | |
a4ec24b4 | 109 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
110 | * Timeslices get refilled after they expire. |
111 | */ | |
1da177e4 | 112 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 113 | |
5517d86b ED |
114 | #ifdef CONFIG_SMP |
115 | /* | |
116 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
117 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
118 | */ | |
119 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
120 | { | |
121 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
122 | } | |
123 | ||
124 | /* | |
125 | * Each time a sched group cpu_power is changed, | |
126 | * we must compute its reciprocal value | |
127 | */ | |
128 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
129 | { | |
130 | sg->__cpu_power += val; | |
131 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
132 | } | |
133 | #endif | |
134 | ||
e05606d3 IM |
135 | static inline int rt_policy(int policy) |
136 | { | |
137 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
138 | return 1; | |
139 | return 0; | |
140 | } | |
141 | ||
142 | static inline int task_has_rt_policy(struct task_struct *p) | |
143 | { | |
144 | return rt_policy(p->policy); | |
145 | } | |
146 | ||
1da177e4 | 147 | /* |
6aa645ea | 148 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 149 | */ |
6aa645ea IM |
150 | struct rt_prio_array { |
151 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
152 | struct list_head queue[MAX_RT_PRIO]; | |
153 | }; | |
154 | ||
29f59db3 SV |
155 | #ifdef CONFIG_FAIR_GROUP_SCHED |
156 | ||
29f59db3 SV |
157 | struct cfs_rq; |
158 | ||
159 | /* task group related information */ | |
4cf86d77 | 160 | struct task_group { |
29f59db3 SV |
161 | /* schedulable entities of this group on each cpu */ |
162 | struct sched_entity **se; | |
163 | /* runqueue "owned" by this group on each cpu */ | |
164 | struct cfs_rq **cfs_rq; | |
165 | unsigned long shares; | |
5cb350ba DG |
166 | /* spinlock to serialize modification to shares */ |
167 | spinlock_t lock; | |
29f59db3 SV |
168 | }; |
169 | ||
170 | /* Default task group's sched entity on each cpu */ | |
171 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
172 | /* Default task group's cfs_rq on each cpu */ | |
173 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
174 | ||
9b5b7751 SV |
175 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
176 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
177 | |
178 | /* Default task group. | |
3a252015 | 179 | * Every task in system belong to this group at bootup. |
29f59db3 | 180 | */ |
4cf86d77 | 181 | struct task_group init_task_group = { |
3a252015 IM |
182 | .se = init_sched_entity_p, |
183 | .cfs_rq = init_cfs_rq_p, | |
184 | }; | |
9b5b7751 | 185 | |
24e377a8 | 186 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 187 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 188 | #else |
3a252015 | 189 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
190 | #endif |
191 | ||
4cf86d77 | 192 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
193 | |
194 | /* return group to which a task belongs */ | |
4cf86d77 | 195 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 196 | { |
4cf86d77 | 197 | struct task_group *tg; |
9b5b7751 | 198 | |
24e377a8 SV |
199 | #ifdef CONFIG_FAIR_USER_SCHED |
200 | tg = p->user->tg; | |
201 | #else | |
4cf86d77 | 202 | tg = &init_task_group; |
24e377a8 | 203 | #endif |
9b5b7751 SV |
204 | |
205 | return tg; | |
29f59db3 SV |
206 | } |
207 | ||
208 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
209 | static inline void set_task_cfs_rq(struct task_struct *p) | |
210 | { | |
4cf86d77 IM |
211 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
212 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
213 | } |
214 | ||
215 | #else | |
216 | ||
217 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
218 | ||
219 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
220 | ||
6aa645ea IM |
221 | /* CFS-related fields in a runqueue */ |
222 | struct cfs_rq { | |
223 | struct load_weight load; | |
224 | unsigned long nr_running; | |
225 | ||
6aa645ea | 226 | u64 exec_clock; |
e9acbff6 | 227 | u64 min_vruntime; |
6aa645ea IM |
228 | |
229 | struct rb_root tasks_timeline; | |
230 | struct rb_node *rb_leftmost; | |
231 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
232 | /* 'curr' points to currently running entity on this cfs_rq. |
233 | * It is set to NULL otherwise (i.e when none are currently running). | |
234 | */ | |
235 | struct sched_entity *curr; | |
ddc97297 PZ |
236 | |
237 | unsigned long nr_spread_over; | |
238 | ||
62160e3f | 239 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
240 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
241 | ||
242 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
243 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
244 | * (like users, containers etc.) | |
245 | * | |
246 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
247 | * list is used during load balance. | |
248 | */ | |
249 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 250 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 251 | struct rcu_head rcu; |
6aa645ea IM |
252 | #endif |
253 | }; | |
1da177e4 | 254 | |
6aa645ea IM |
255 | /* Real-Time classes' related field in a runqueue: */ |
256 | struct rt_rq { | |
257 | struct rt_prio_array active; | |
258 | int rt_load_balance_idx; | |
259 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
260 | }; | |
261 | ||
1da177e4 LT |
262 | /* |
263 | * This is the main, per-CPU runqueue data structure. | |
264 | * | |
265 | * Locking rule: those places that want to lock multiple runqueues | |
266 | * (such as the load balancing or the thread migration code), lock | |
267 | * acquire operations must be ordered by ascending &runqueue. | |
268 | */ | |
70b97a7f | 269 | struct rq { |
d8016491 IM |
270 | /* runqueue lock: */ |
271 | spinlock_t lock; | |
1da177e4 LT |
272 | |
273 | /* | |
274 | * nr_running and cpu_load should be in the same cacheline because | |
275 | * remote CPUs use both these fields when doing load calculation. | |
276 | */ | |
277 | unsigned long nr_running; | |
6aa645ea IM |
278 | #define CPU_LOAD_IDX_MAX 5 |
279 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 280 | unsigned char idle_at_tick; |
46cb4b7c SS |
281 | #ifdef CONFIG_NO_HZ |
282 | unsigned char in_nohz_recently; | |
283 | #endif | |
d8016491 IM |
284 | /* capture load from *all* tasks on this cpu: */ |
285 | struct load_weight load; | |
6aa645ea IM |
286 | unsigned long nr_load_updates; |
287 | u64 nr_switches; | |
288 | ||
289 | struct cfs_rq cfs; | |
290 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
291 | /* list of leaf cfs_rq on this cpu: */ |
292 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 293 | #endif |
6aa645ea | 294 | struct rt_rq rt; |
1da177e4 LT |
295 | |
296 | /* | |
297 | * This is part of a global counter where only the total sum | |
298 | * over all CPUs matters. A task can increase this counter on | |
299 | * one CPU and if it got migrated afterwards it may decrease | |
300 | * it on another CPU. Always updated under the runqueue lock: | |
301 | */ | |
302 | unsigned long nr_uninterruptible; | |
303 | ||
36c8b586 | 304 | struct task_struct *curr, *idle; |
c9819f45 | 305 | unsigned long next_balance; |
1da177e4 | 306 | struct mm_struct *prev_mm; |
6aa645ea | 307 | |
6aa645ea IM |
308 | u64 clock, prev_clock_raw; |
309 | s64 clock_max_delta; | |
310 | ||
311 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
312 | u64 idle_clock; |
313 | unsigned int clock_deep_idle_events; | |
529c7726 | 314 | u64 tick_timestamp; |
6aa645ea | 315 | |
1da177e4 LT |
316 | atomic_t nr_iowait; |
317 | ||
318 | #ifdef CONFIG_SMP | |
319 | struct sched_domain *sd; | |
320 | ||
321 | /* For active balancing */ | |
322 | int active_balance; | |
323 | int push_cpu; | |
d8016491 IM |
324 | /* cpu of this runqueue: */ |
325 | int cpu; | |
1da177e4 | 326 | |
36c8b586 | 327 | struct task_struct *migration_thread; |
1da177e4 LT |
328 | struct list_head migration_queue; |
329 | #endif | |
330 | ||
331 | #ifdef CONFIG_SCHEDSTATS | |
332 | /* latency stats */ | |
333 | struct sched_info rq_sched_info; | |
334 | ||
335 | /* sys_sched_yield() stats */ | |
480b9434 KC |
336 | unsigned int yld_exp_empty; |
337 | unsigned int yld_act_empty; | |
338 | unsigned int yld_both_empty; | |
339 | unsigned int yld_count; | |
1da177e4 LT |
340 | |
341 | /* schedule() stats */ | |
480b9434 KC |
342 | unsigned int sched_switch; |
343 | unsigned int sched_count; | |
344 | unsigned int sched_goidle; | |
1da177e4 LT |
345 | |
346 | /* try_to_wake_up() stats */ | |
480b9434 KC |
347 | unsigned int ttwu_count; |
348 | unsigned int ttwu_local; | |
b8efb561 IM |
349 | |
350 | /* BKL stats */ | |
480b9434 | 351 | unsigned int bkl_count; |
1da177e4 | 352 | #endif |
fcb99371 | 353 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
354 | }; |
355 | ||
f34e3b61 | 356 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 357 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 358 | |
dd41f596 IM |
359 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
360 | { | |
361 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
362 | } | |
363 | ||
0a2966b4 CL |
364 | static inline int cpu_of(struct rq *rq) |
365 | { | |
366 | #ifdef CONFIG_SMP | |
367 | return rq->cpu; | |
368 | #else | |
369 | return 0; | |
370 | #endif | |
371 | } | |
372 | ||
20d315d4 | 373 | /* |
b04a0f4c IM |
374 | * Update the per-runqueue clock, as finegrained as the platform can give |
375 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 376 | */ |
b04a0f4c | 377 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
378 | { |
379 | u64 prev_raw = rq->prev_clock_raw; | |
380 | u64 now = sched_clock(); | |
381 | s64 delta = now - prev_raw; | |
382 | u64 clock = rq->clock; | |
383 | ||
b04a0f4c IM |
384 | #ifdef CONFIG_SCHED_DEBUG |
385 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
386 | #endif | |
20d315d4 IM |
387 | /* |
388 | * Protect against sched_clock() occasionally going backwards: | |
389 | */ | |
390 | if (unlikely(delta < 0)) { | |
391 | clock++; | |
392 | rq->clock_warps++; | |
393 | } else { | |
394 | /* | |
395 | * Catch too large forward jumps too: | |
396 | */ | |
529c7726 IM |
397 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
398 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
399 | clock = rq->tick_timestamp + TICK_NSEC; | |
400 | else | |
401 | clock++; | |
20d315d4 IM |
402 | rq->clock_overflows++; |
403 | } else { | |
404 | if (unlikely(delta > rq->clock_max_delta)) | |
405 | rq->clock_max_delta = delta; | |
406 | clock += delta; | |
407 | } | |
408 | } | |
409 | ||
410 | rq->prev_clock_raw = now; | |
411 | rq->clock = clock; | |
b04a0f4c | 412 | } |
20d315d4 | 413 | |
b04a0f4c IM |
414 | static void update_rq_clock(struct rq *rq) |
415 | { | |
416 | if (likely(smp_processor_id() == cpu_of(rq))) | |
417 | __update_rq_clock(rq); | |
20d315d4 IM |
418 | } |
419 | ||
674311d5 NP |
420 | /* |
421 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 422 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
423 | * |
424 | * The domain tree of any CPU may only be accessed from within | |
425 | * preempt-disabled sections. | |
426 | */ | |
48f24c4d IM |
427 | #define for_each_domain(cpu, __sd) \ |
428 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
429 | |
430 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
431 | #define this_rq() (&__get_cpu_var(runqueues)) | |
432 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
433 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
434 | ||
bf5c91ba IM |
435 | /* |
436 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
437 | */ | |
438 | #ifdef CONFIG_SCHED_DEBUG | |
439 | # define const_debug __read_mostly | |
440 | #else | |
441 | # define const_debug static const | |
442 | #endif | |
443 | ||
444 | /* | |
445 | * Debugging: various feature bits | |
446 | */ | |
447 | enum { | |
bbdba7c0 IM |
448 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
449 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 450 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 451 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 452 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 453 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
454 | }; |
455 | ||
456 | const_debug unsigned int sysctl_sched_features = | |
8401f775 IM |
457 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
458 | SCHED_FEAT_START_DEBIT * 1 | | |
459 | SCHED_FEAT_TREE_AVG * 0 | | |
460 | SCHED_FEAT_APPROX_AVG * 0 | | |
461 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | | |
462 | SCHED_FEAT_PREEMPT_RESTRICT * 1; | |
bf5c91ba IM |
463 | |
464 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
465 | ||
e436d800 IM |
466 | /* |
467 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
468 | * clock constructed from sched_clock(): | |
469 | */ | |
470 | unsigned long long cpu_clock(int cpu) | |
471 | { | |
e436d800 IM |
472 | unsigned long long now; |
473 | unsigned long flags; | |
b04a0f4c | 474 | struct rq *rq; |
e436d800 | 475 | |
2cd4d0ea | 476 | local_irq_save(flags); |
b04a0f4c IM |
477 | rq = cpu_rq(cpu); |
478 | update_rq_clock(rq); | |
479 | now = rq->clock; | |
2cd4d0ea | 480 | local_irq_restore(flags); |
e436d800 IM |
481 | |
482 | return now; | |
483 | } | |
a58f6f25 | 484 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 485 | |
1da177e4 | 486 | #ifndef prepare_arch_switch |
4866cde0 NP |
487 | # define prepare_arch_switch(next) do { } while (0) |
488 | #endif | |
489 | #ifndef finish_arch_switch | |
490 | # define finish_arch_switch(prev) do { } while (0) | |
491 | #endif | |
492 | ||
493 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 494 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
495 | { |
496 | return rq->curr == p; | |
497 | } | |
498 | ||
70b97a7f | 499 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
500 | { |
501 | } | |
502 | ||
70b97a7f | 503 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 504 | { |
da04c035 IM |
505 | #ifdef CONFIG_DEBUG_SPINLOCK |
506 | /* this is a valid case when another task releases the spinlock */ | |
507 | rq->lock.owner = current; | |
508 | #endif | |
8a25d5de IM |
509 | /* |
510 | * If we are tracking spinlock dependencies then we have to | |
511 | * fix up the runqueue lock - which gets 'carried over' from | |
512 | * prev into current: | |
513 | */ | |
514 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
515 | ||
4866cde0 NP |
516 | spin_unlock_irq(&rq->lock); |
517 | } | |
518 | ||
519 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 520 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
521 | { |
522 | #ifdef CONFIG_SMP | |
523 | return p->oncpu; | |
524 | #else | |
525 | return rq->curr == p; | |
526 | #endif | |
527 | } | |
528 | ||
70b97a7f | 529 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
530 | { |
531 | #ifdef CONFIG_SMP | |
532 | /* | |
533 | * We can optimise this out completely for !SMP, because the | |
534 | * SMP rebalancing from interrupt is the only thing that cares | |
535 | * here. | |
536 | */ | |
537 | next->oncpu = 1; | |
538 | #endif | |
539 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
540 | spin_unlock_irq(&rq->lock); | |
541 | #else | |
542 | spin_unlock(&rq->lock); | |
543 | #endif | |
544 | } | |
545 | ||
70b97a7f | 546 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
547 | { |
548 | #ifdef CONFIG_SMP | |
549 | /* | |
550 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
551 | * We must ensure this doesn't happen until the switch is completely | |
552 | * finished. | |
553 | */ | |
554 | smp_wmb(); | |
555 | prev->oncpu = 0; | |
556 | #endif | |
557 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
558 | local_irq_enable(); | |
1da177e4 | 559 | #endif |
4866cde0 NP |
560 | } |
561 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 562 | |
b29739f9 IM |
563 | /* |
564 | * __task_rq_lock - lock the runqueue a given task resides on. | |
565 | * Must be called interrupts disabled. | |
566 | */ | |
70b97a7f | 567 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
568 | __acquires(rq->lock) |
569 | { | |
3a5c359a AK |
570 | for (;;) { |
571 | struct rq *rq = task_rq(p); | |
572 | spin_lock(&rq->lock); | |
573 | if (likely(rq == task_rq(p))) | |
574 | return rq; | |
b29739f9 | 575 | spin_unlock(&rq->lock); |
b29739f9 | 576 | } |
b29739f9 IM |
577 | } |
578 | ||
1da177e4 LT |
579 | /* |
580 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
581 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
582 | * explicitly disabling preemption. | |
583 | */ | |
70b97a7f | 584 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
585 | __acquires(rq->lock) |
586 | { | |
70b97a7f | 587 | struct rq *rq; |
1da177e4 | 588 | |
3a5c359a AK |
589 | for (;;) { |
590 | local_irq_save(*flags); | |
591 | rq = task_rq(p); | |
592 | spin_lock(&rq->lock); | |
593 | if (likely(rq == task_rq(p))) | |
594 | return rq; | |
1da177e4 | 595 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 596 | } |
1da177e4 LT |
597 | } |
598 | ||
a9957449 | 599 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
600 | __releases(rq->lock) |
601 | { | |
602 | spin_unlock(&rq->lock); | |
603 | } | |
604 | ||
70b97a7f | 605 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
606 | __releases(rq->lock) |
607 | { | |
608 | spin_unlock_irqrestore(&rq->lock, *flags); | |
609 | } | |
610 | ||
1da177e4 | 611 | /* |
cc2a73b5 | 612 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 613 | */ |
a9957449 | 614 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
615 | __acquires(rq->lock) |
616 | { | |
70b97a7f | 617 | struct rq *rq; |
1da177e4 LT |
618 | |
619 | local_irq_disable(); | |
620 | rq = this_rq(); | |
621 | spin_lock(&rq->lock); | |
622 | ||
623 | return rq; | |
624 | } | |
625 | ||
1b9f19c2 | 626 | /* |
2aa44d05 | 627 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 628 | */ |
2aa44d05 | 629 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 630 | { |
2aa44d05 IM |
631 | struct rq *rq = cpu_rq(smp_processor_id()); |
632 | ||
633 | spin_lock(&rq->lock); | |
634 | __update_rq_clock(rq); | |
635 | spin_unlock(&rq->lock); | |
636 | rq->clock_deep_idle_events++; | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
639 | ||
640 | /* | |
641 | * We just idled delta nanoseconds (called with irqs disabled): | |
642 | */ | |
643 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
644 | { | |
645 | struct rq *rq = cpu_rq(smp_processor_id()); | |
646 | u64 now = sched_clock(); | |
1b9f19c2 | 647 | |
2aa44d05 IM |
648 | rq->idle_clock += delta_ns; |
649 | /* | |
650 | * Override the previous timestamp and ignore all | |
651 | * sched_clock() deltas that occured while we idled, | |
652 | * and use the PM-provided delta_ns to advance the | |
653 | * rq clock: | |
654 | */ | |
655 | spin_lock(&rq->lock); | |
656 | rq->prev_clock_raw = now; | |
657 | rq->clock += delta_ns; | |
658 | spin_unlock(&rq->lock); | |
1b9f19c2 | 659 | } |
2aa44d05 | 660 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 661 | |
c24d20db IM |
662 | /* |
663 | * resched_task - mark a task 'to be rescheduled now'. | |
664 | * | |
665 | * On UP this means the setting of the need_resched flag, on SMP it | |
666 | * might also involve a cross-CPU call to trigger the scheduler on | |
667 | * the target CPU. | |
668 | */ | |
669 | #ifdef CONFIG_SMP | |
670 | ||
671 | #ifndef tsk_is_polling | |
672 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
673 | #endif | |
674 | ||
675 | static void resched_task(struct task_struct *p) | |
676 | { | |
677 | int cpu; | |
678 | ||
679 | assert_spin_locked(&task_rq(p)->lock); | |
680 | ||
681 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
682 | return; | |
683 | ||
684 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
685 | ||
686 | cpu = task_cpu(p); | |
687 | if (cpu == smp_processor_id()) | |
688 | return; | |
689 | ||
690 | /* NEED_RESCHED must be visible before we test polling */ | |
691 | smp_mb(); | |
692 | if (!tsk_is_polling(p)) | |
693 | smp_send_reschedule(cpu); | |
694 | } | |
695 | ||
696 | static void resched_cpu(int cpu) | |
697 | { | |
698 | struct rq *rq = cpu_rq(cpu); | |
699 | unsigned long flags; | |
700 | ||
701 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
702 | return; | |
703 | resched_task(cpu_curr(cpu)); | |
704 | spin_unlock_irqrestore(&rq->lock, flags); | |
705 | } | |
706 | #else | |
707 | static inline void resched_task(struct task_struct *p) | |
708 | { | |
709 | assert_spin_locked(&task_rq(p)->lock); | |
710 | set_tsk_need_resched(p); | |
711 | } | |
712 | #endif | |
713 | ||
45bf76df IM |
714 | #if BITS_PER_LONG == 32 |
715 | # define WMULT_CONST (~0UL) | |
716 | #else | |
717 | # define WMULT_CONST (1UL << 32) | |
718 | #endif | |
719 | ||
720 | #define WMULT_SHIFT 32 | |
721 | ||
194081eb IM |
722 | /* |
723 | * Shift right and round: | |
724 | */ | |
cf2ab469 | 725 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 726 | |
cb1c4fc9 | 727 | static unsigned long |
45bf76df IM |
728 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
729 | struct load_weight *lw) | |
730 | { | |
731 | u64 tmp; | |
732 | ||
733 | if (unlikely(!lw->inv_weight)) | |
194081eb | 734 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
735 | |
736 | tmp = (u64)delta_exec * weight; | |
737 | /* | |
738 | * Check whether we'd overflow the 64-bit multiplication: | |
739 | */ | |
194081eb | 740 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 741 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
742 | WMULT_SHIFT/2); |
743 | else | |
cf2ab469 | 744 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 745 | |
ecf691da | 746 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
747 | } |
748 | ||
749 | static inline unsigned long | |
750 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
751 | { | |
752 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
753 | } | |
754 | ||
1091985b | 755 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
756 | { |
757 | lw->weight += inc; | |
45bf76df IM |
758 | } |
759 | ||
1091985b | 760 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
761 | { |
762 | lw->weight -= dec; | |
45bf76df IM |
763 | } |
764 | ||
2dd73a4f PW |
765 | /* |
766 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
767 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
768 | * each task makes to its run queue's load is weighted according to its | |
769 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
770 | * scaled version of the new time slice allocation that they receive on time | |
771 | * slice expiry etc. | |
772 | */ | |
773 | ||
dd41f596 IM |
774 | #define WEIGHT_IDLEPRIO 2 |
775 | #define WMULT_IDLEPRIO (1 << 31) | |
776 | ||
777 | /* | |
778 | * Nice levels are multiplicative, with a gentle 10% change for every | |
779 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
780 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
781 | * that remained on nice 0. | |
782 | * | |
783 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
784 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
785 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
786 | * If a task goes up by ~10% and another task goes down by ~10% then | |
787 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
788 | */ |
789 | static const int prio_to_weight[40] = { | |
254753dc IM |
790 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
791 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
792 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
793 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
794 | /* 0 */ 1024, 820, 655, 526, 423, | |
795 | /* 5 */ 335, 272, 215, 172, 137, | |
796 | /* 10 */ 110, 87, 70, 56, 45, | |
797 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
798 | }; |
799 | ||
5714d2de IM |
800 | /* |
801 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
802 | * | |
803 | * In cases where the weight does not change often, we can use the | |
804 | * precalculated inverse to speed up arithmetics by turning divisions | |
805 | * into multiplications: | |
806 | */ | |
dd41f596 | 807 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
808 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
809 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
810 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
811 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
812 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
813 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
814 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
815 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 816 | }; |
2dd73a4f | 817 | |
dd41f596 IM |
818 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
819 | ||
820 | /* | |
821 | * runqueue iterator, to support SMP load-balancing between different | |
822 | * scheduling classes, without having to expose their internal data | |
823 | * structures to the load-balancing proper: | |
824 | */ | |
825 | struct rq_iterator { | |
826 | void *arg; | |
827 | struct task_struct *(*start)(void *); | |
828 | struct task_struct *(*next)(void *); | |
829 | }; | |
830 | ||
831 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
832 | unsigned long max_nr_move, unsigned long max_load_move, | |
833 | struct sched_domain *sd, enum cpu_idle_type idle, | |
834 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 835 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
836 | |
837 | #include "sched_stats.h" | |
dd41f596 | 838 | #include "sched_idletask.c" |
5522d5d5 IM |
839 | #include "sched_fair.c" |
840 | #include "sched_rt.c" | |
dd41f596 IM |
841 | #ifdef CONFIG_SCHED_DEBUG |
842 | # include "sched_debug.c" | |
843 | #endif | |
844 | ||
845 | #define sched_class_highest (&rt_sched_class) | |
846 | ||
9c217245 IM |
847 | /* |
848 | * Update delta_exec, delta_fair fields for rq. | |
849 | * | |
850 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 851 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
852 | * delta_exec advances at the same rate as wall-clock (provided |
853 | * cpu is not idle). | |
854 | * | |
855 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
856 | * runqueue over any given interval. This (smoothened) load is used | |
857 | * during load balance. | |
858 | * | |
495eca49 | 859 | * This function is called /before/ updating rq->load |
9c217245 IM |
860 | * and when switching tasks. |
861 | */ | |
29b4b623 | 862 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 863 | { |
495eca49 | 864 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
865 | } |
866 | ||
79b5dddf | 867 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 868 | { |
495eca49 | 869 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
870 | } |
871 | ||
e5fa2237 | 872 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
873 | { |
874 | rq->nr_running++; | |
29b4b623 | 875 | inc_load(rq, p); |
9c217245 IM |
876 | } |
877 | ||
db53181e | 878 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
879 | { |
880 | rq->nr_running--; | |
79b5dddf | 881 | dec_load(rq, p); |
9c217245 IM |
882 | } |
883 | ||
45bf76df IM |
884 | static void set_load_weight(struct task_struct *p) |
885 | { | |
886 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
887 | p->se.load.weight = prio_to_weight[0] * 2; |
888 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
889 | return; | |
890 | } | |
45bf76df | 891 | |
dd41f596 IM |
892 | /* |
893 | * SCHED_IDLE tasks get minimal weight: | |
894 | */ | |
895 | if (p->policy == SCHED_IDLE) { | |
896 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
897 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
898 | return; | |
899 | } | |
71f8bd46 | 900 | |
dd41f596 IM |
901 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
902 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
903 | } |
904 | ||
8159f87e | 905 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 906 | { |
dd41f596 | 907 | sched_info_queued(p); |
fd390f6a | 908 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 909 | p->se.on_rq = 1; |
71f8bd46 IM |
910 | } |
911 | ||
69be72c1 | 912 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 913 | { |
f02231e5 | 914 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 915 | p->se.on_rq = 0; |
71f8bd46 IM |
916 | } |
917 | ||
14531189 | 918 | /* |
dd41f596 | 919 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 920 | */ |
14531189 IM |
921 | static inline int __normal_prio(struct task_struct *p) |
922 | { | |
dd41f596 | 923 | return p->static_prio; |
14531189 IM |
924 | } |
925 | ||
b29739f9 IM |
926 | /* |
927 | * Calculate the expected normal priority: i.e. priority | |
928 | * without taking RT-inheritance into account. Might be | |
929 | * boosted by interactivity modifiers. Changes upon fork, | |
930 | * setprio syscalls, and whenever the interactivity | |
931 | * estimator recalculates. | |
932 | */ | |
36c8b586 | 933 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
934 | { |
935 | int prio; | |
936 | ||
e05606d3 | 937 | if (task_has_rt_policy(p)) |
b29739f9 IM |
938 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
939 | else | |
940 | prio = __normal_prio(p); | |
941 | return prio; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Calculate the current priority, i.e. the priority | |
946 | * taken into account by the scheduler. This value might | |
947 | * be boosted by RT tasks, or might be boosted by | |
948 | * interactivity modifiers. Will be RT if the task got | |
949 | * RT-boosted. If not then it returns p->normal_prio. | |
950 | */ | |
36c8b586 | 951 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
952 | { |
953 | p->normal_prio = normal_prio(p); | |
954 | /* | |
955 | * If we are RT tasks or we were boosted to RT priority, | |
956 | * keep the priority unchanged. Otherwise, update priority | |
957 | * to the normal priority: | |
958 | */ | |
959 | if (!rt_prio(p->prio)) | |
960 | return p->normal_prio; | |
961 | return p->prio; | |
962 | } | |
963 | ||
1da177e4 | 964 | /* |
dd41f596 | 965 | * activate_task - move a task to the runqueue. |
1da177e4 | 966 | */ |
dd41f596 | 967 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 968 | { |
dd41f596 IM |
969 | if (p->state == TASK_UNINTERRUPTIBLE) |
970 | rq->nr_uninterruptible--; | |
1da177e4 | 971 | |
8159f87e | 972 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 973 | inc_nr_running(p, rq); |
1da177e4 LT |
974 | } |
975 | ||
1da177e4 LT |
976 | /* |
977 | * deactivate_task - remove a task from the runqueue. | |
978 | */ | |
2e1cb74a | 979 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 980 | { |
dd41f596 IM |
981 | if (p->state == TASK_UNINTERRUPTIBLE) |
982 | rq->nr_uninterruptible++; | |
983 | ||
69be72c1 | 984 | dequeue_task(rq, p, sleep); |
db53181e | 985 | dec_nr_running(p, rq); |
1da177e4 LT |
986 | } |
987 | ||
1da177e4 LT |
988 | /** |
989 | * task_curr - is this task currently executing on a CPU? | |
990 | * @p: the task in question. | |
991 | */ | |
36c8b586 | 992 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
993 | { |
994 | return cpu_curr(task_cpu(p)) == p; | |
995 | } | |
996 | ||
2dd73a4f PW |
997 | /* Used instead of source_load when we know the type == 0 */ |
998 | unsigned long weighted_cpuload(const int cpu) | |
999 | { | |
495eca49 | 1000 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1001 | } |
1002 | ||
1003 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1004 | { | |
1005 | #ifdef CONFIG_SMP | |
1006 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1007 | #endif |
29f59db3 | 1008 | set_task_cfs_rq(p); |
2dd73a4f PW |
1009 | } |
1010 | ||
1da177e4 | 1011 | #ifdef CONFIG_SMP |
c65cc870 | 1012 | |
cc367732 IM |
1013 | /* |
1014 | * Is this task likely cache-hot: | |
1015 | */ | |
1016 | static inline int | |
1017 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1018 | { | |
1019 | s64 delta; | |
1020 | ||
1021 | if (p->sched_class != &fair_sched_class) | |
1022 | return 0; | |
1023 | ||
6bc1665b IM |
1024 | if (sysctl_sched_migration_cost == -1) |
1025 | return 1; | |
1026 | if (sysctl_sched_migration_cost == 0) | |
1027 | return 0; | |
1028 | ||
cc367732 IM |
1029 | delta = now - p->se.exec_start; |
1030 | ||
1031 | return delta < (s64)sysctl_sched_migration_cost; | |
1032 | } | |
1033 | ||
1034 | ||
dd41f596 | 1035 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1036 | { |
dd41f596 IM |
1037 | int old_cpu = task_cpu(p); |
1038 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1039 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1040 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1041 | u64 clock_offset; |
dd41f596 IM |
1042 | |
1043 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1044 | |
1045 | #ifdef CONFIG_SCHEDSTATS | |
1046 | if (p->se.wait_start) | |
1047 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1048 | if (p->se.sleep_start) |
1049 | p->se.sleep_start -= clock_offset; | |
1050 | if (p->se.block_start) | |
1051 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1052 | if (old_cpu != new_cpu) { |
1053 | schedstat_inc(p, se.nr_migrations); | |
1054 | if (task_hot(p, old_rq->clock, NULL)) | |
1055 | schedstat_inc(p, se.nr_forced2_migrations); | |
1056 | } | |
6cfb0d5d | 1057 | #endif |
2830cf8c SV |
1058 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1059 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1060 | |
1061 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1062 | } |
1063 | ||
70b97a7f | 1064 | struct migration_req { |
1da177e4 | 1065 | struct list_head list; |
1da177e4 | 1066 | |
36c8b586 | 1067 | struct task_struct *task; |
1da177e4 LT |
1068 | int dest_cpu; |
1069 | ||
1da177e4 | 1070 | struct completion done; |
70b97a7f | 1071 | }; |
1da177e4 LT |
1072 | |
1073 | /* | |
1074 | * The task's runqueue lock must be held. | |
1075 | * Returns true if you have to wait for migration thread. | |
1076 | */ | |
36c8b586 | 1077 | static int |
70b97a7f | 1078 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1079 | { |
70b97a7f | 1080 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1081 | |
1082 | /* | |
1083 | * If the task is not on a runqueue (and not running), then | |
1084 | * it is sufficient to simply update the task's cpu field. | |
1085 | */ | |
dd41f596 | 1086 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1087 | set_task_cpu(p, dest_cpu); |
1088 | return 0; | |
1089 | } | |
1090 | ||
1091 | init_completion(&req->done); | |
1da177e4 LT |
1092 | req->task = p; |
1093 | req->dest_cpu = dest_cpu; | |
1094 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1095 | |
1da177e4 LT |
1096 | return 1; |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * wait_task_inactive - wait for a thread to unschedule. | |
1101 | * | |
1102 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1103 | * else this function might spin for a *long* time. This function can't | |
1104 | * be called with interrupts off, or it may introduce deadlock with | |
1105 | * smp_call_function() if an IPI is sent by the same process we are | |
1106 | * waiting to become inactive. | |
1107 | */ | |
36c8b586 | 1108 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1109 | { |
1110 | unsigned long flags; | |
dd41f596 | 1111 | int running, on_rq; |
70b97a7f | 1112 | struct rq *rq; |
1da177e4 | 1113 | |
3a5c359a AK |
1114 | for (;;) { |
1115 | /* | |
1116 | * We do the initial early heuristics without holding | |
1117 | * any task-queue locks at all. We'll only try to get | |
1118 | * the runqueue lock when things look like they will | |
1119 | * work out! | |
1120 | */ | |
1121 | rq = task_rq(p); | |
fa490cfd | 1122 | |
3a5c359a AK |
1123 | /* |
1124 | * If the task is actively running on another CPU | |
1125 | * still, just relax and busy-wait without holding | |
1126 | * any locks. | |
1127 | * | |
1128 | * NOTE! Since we don't hold any locks, it's not | |
1129 | * even sure that "rq" stays as the right runqueue! | |
1130 | * But we don't care, since "task_running()" will | |
1131 | * return false if the runqueue has changed and p | |
1132 | * is actually now running somewhere else! | |
1133 | */ | |
1134 | while (task_running(rq, p)) | |
1135 | cpu_relax(); | |
fa490cfd | 1136 | |
3a5c359a AK |
1137 | /* |
1138 | * Ok, time to look more closely! We need the rq | |
1139 | * lock now, to be *sure*. If we're wrong, we'll | |
1140 | * just go back and repeat. | |
1141 | */ | |
1142 | rq = task_rq_lock(p, &flags); | |
1143 | running = task_running(rq, p); | |
1144 | on_rq = p->se.on_rq; | |
1145 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1146 | |
3a5c359a AK |
1147 | /* |
1148 | * Was it really running after all now that we | |
1149 | * checked with the proper locks actually held? | |
1150 | * | |
1151 | * Oops. Go back and try again.. | |
1152 | */ | |
1153 | if (unlikely(running)) { | |
1154 | cpu_relax(); | |
1155 | continue; | |
1156 | } | |
fa490cfd | 1157 | |
3a5c359a AK |
1158 | /* |
1159 | * It's not enough that it's not actively running, | |
1160 | * it must be off the runqueue _entirely_, and not | |
1161 | * preempted! | |
1162 | * | |
1163 | * So if it wa still runnable (but just not actively | |
1164 | * running right now), it's preempted, and we should | |
1165 | * yield - it could be a while. | |
1166 | */ | |
1167 | if (unlikely(on_rq)) { | |
1168 | schedule_timeout_uninterruptible(1); | |
1169 | continue; | |
1170 | } | |
fa490cfd | 1171 | |
3a5c359a AK |
1172 | /* |
1173 | * Ahh, all good. It wasn't running, and it wasn't | |
1174 | * runnable, which means that it will never become | |
1175 | * running in the future either. We're all done! | |
1176 | */ | |
1177 | break; | |
1178 | } | |
1da177e4 LT |
1179 | } |
1180 | ||
1181 | /*** | |
1182 | * kick_process - kick a running thread to enter/exit the kernel | |
1183 | * @p: the to-be-kicked thread | |
1184 | * | |
1185 | * Cause a process which is running on another CPU to enter | |
1186 | * kernel-mode, without any delay. (to get signals handled.) | |
1187 | * | |
1188 | * NOTE: this function doesnt have to take the runqueue lock, | |
1189 | * because all it wants to ensure is that the remote task enters | |
1190 | * the kernel. If the IPI races and the task has been migrated | |
1191 | * to another CPU then no harm is done and the purpose has been | |
1192 | * achieved as well. | |
1193 | */ | |
36c8b586 | 1194 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1195 | { |
1196 | int cpu; | |
1197 | ||
1198 | preempt_disable(); | |
1199 | cpu = task_cpu(p); | |
1200 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1201 | smp_send_reschedule(cpu); | |
1202 | preempt_enable(); | |
1203 | } | |
1204 | ||
1205 | /* | |
2dd73a4f PW |
1206 | * Return a low guess at the load of a migration-source cpu weighted |
1207 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1208 | * |
1209 | * We want to under-estimate the load of migration sources, to | |
1210 | * balance conservatively. | |
1211 | */ | |
a9957449 | 1212 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1213 | { |
70b97a7f | 1214 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1215 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1216 | |
3b0bd9bc | 1217 | if (type == 0) |
dd41f596 | 1218 | return total; |
b910472d | 1219 | |
dd41f596 | 1220 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
2dd73a4f PW |
1224 | * Return a high guess at the load of a migration-target cpu weighted |
1225 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1226 | */ |
a9957449 | 1227 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1228 | { |
70b97a7f | 1229 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1230 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1231 | |
7897986b | 1232 | if (type == 0) |
dd41f596 | 1233 | return total; |
3b0bd9bc | 1234 | |
dd41f596 | 1235 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1236 | } |
1237 | ||
1238 | /* | |
1239 | * Return the average load per task on the cpu's run queue | |
1240 | */ | |
1241 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1242 | { | |
70b97a7f | 1243 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1244 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1245 | unsigned long n = rq->nr_running; |
1246 | ||
dd41f596 | 1247 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1248 | } |
1249 | ||
147cbb4b NP |
1250 | /* |
1251 | * find_idlest_group finds and returns the least busy CPU group within the | |
1252 | * domain. | |
1253 | */ | |
1254 | static struct sched_group * | |
1255 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1256 | { | |
1257 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1258 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1259 | int load_idx = sd->forkexec_idx; | |
1260 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1261 | ||
1262 | do { | |
1263 | unsigned long load, avg_load; | |
1264 | int local_group; | |
1265 | int i; | |
1266 | ||
da5a5522 BD |
1267 | /* Skip over this group if it has no CPUs allowed */ |
1268 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1269 | continue; |
da5a5522 | 1270 | |
147cbb4b | 1271 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1272 | |
1273 | /* Tally up the load of all CPUs in the group */ | |
1274 | avg_load = 0; | |
1275 | ||
1276 | for_each_cpu_mask(i, group->cpumask) { | |
1277 | /* Bias balancing toward cpus of our domain */ | |
1278 | if (local_group) | |
1279 | load = source_load(i, load_idx); | |
1280 | else | |
1281 | load = target_load(i, load_idx); | |
1282 | ||
1283 | avg_load += load; | |
1284 | } | |
1285 | ||
1286 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1287 | avg_load = sg_div_cpu_power(group, |
1288 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1289 | |
1290 | if (local_group) { | |
1291 | this_load = avg_load; | |
1292 | this = group; | |
1293 | } else if (avg_load < min_load) { | |
1294 | min_load = avg_load; | |
1295 | idlest = group; | |
1296 | } | |
3a5c359a | 1297 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1298 | |
1299 | if (!idlest || 100*this_load < imbalance*min_load) | |
1300 | return NULL; | |
1301 | return idlest; | |
1302 | } | |
1303 | ||
1304 | /* | |
0feaece9 | 1305 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1306 | */ |
95cdf3b7 IM |
1307 | static int |
1308 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1309 | { |
da5a5522 | 1310 | cpumask_t tmp; |
147cbb4b NP |
1311 | unsigned long load, min_load = ULONG_MAX; |
1312 | int idlest = -1; | |
1313 | int i; | |
1314 | ||
da5a5522 BD |
1315 | /* Traverse only the allowed CPUs */ |
1316 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1317 | ||
1318 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1319 | load = weighted_cpuload(i); |
147cbb4b NP |
1320 | |
1321 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1322 | min_load = load; | |
1323 | idlest = i; | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | return idlest; | |
1328 | } | |
1329 | ||
476d139c NP |
1330 | /* |
1331 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1332 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1333 | * SD_BALANCE_EXEC. | |
1334 | * | |
1335 | * Balance, ie. select the least loaded group. | |
1336 | * | |
1337 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1338 | * | |
1339 | * preempt must be disabled. | |
1340 | */ | |
1341 | static int sched_balance_self(int cpu, int flag) | |
1342 | { | |
1343 | struct task_struct *t = current; | |
1344 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1345 | |
c96d145e | 1346 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1347 | /* |
1348 | * If power savings logic is enabled for a domain, stop there. | |
1349 | */ | |
5c45bf27 SS |
1350 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1351 | break; | |
476d139c NP |
1352 | if (tmp->flags & flag) |
1353 | sd = tmp; | |
c96d145e | 1354 | } |
476d139c NP |
1355 | |
1356 | while (sd) { | |
1357 | cpumask_t span; | |
1358 | struct sched_group *group; | |
1a848870 SS |
1359 | int new_cpu, weight; |
1360 | ||
1361 | if (!(sd->flags & flag)) { | |
1362 | sd = sd->child; | |
1363 | continue; | |
1364 | } | |
476d139c NP |
1365 | |
1366 | span = sd->span; | |
1367 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1368 | if (!group) { |
1369 | sd = sd->child; | |
1370 | continue; | |
1371 | } | |
476d139c | 1372 | |
da5a5522 | 1373 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1374 | if (new_cpu == -1 || new_cpu == cpu) { |
1375 | /* Now try balancing at a lower domain level of cpu */ | |
1376 | sd = sd->child; | |
1377 | continue; | |
1378 | } | |
476d139c | 1379 | |
1a848870 | 1380 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1381 | cpu = new_cpu; |
476d139c NP |
1382 | sd = NULL; |
1383 | weight = cpus_weight(span); | |
1384 | for_each_domain(cpu, tmp) { | |
1385 | if (weight <= cpus_weight(tmp->span)) | |
1386 | break; | |
1387 | if (tmp->flags & flag) | |
1388 | sd = tmp; | |
1389 | } | |
1390 | /* while loop will break here if sd == NULL */ | |
1391 | } | |
1392 | ||
1393 | return cpu; | |
1394 | } | |
1395 | ||
1396 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1397 | |
1398 | /* | |
1399 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1400 | * not idle and an idle cpu is available. The span of cpus to | |
1401 | * search starts with cpus closest then further out as needed, | |
1402 | * so we always favor a closer, idle cpu. | |
1403 | * | |
1404 | * Returns the CPU we should wake onto. | |
1405 | */ | |
1406 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1407 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1408 | { |
1409 | cpumask_t tmp; | |
1410 | struct sched_domain *sd; | |
1411 | int i; | |
1412 | ||
4953198b SS |
1413 | /* |
1414 | * If it is idle, then it is the best cpu to run this task. | |
1415 | * | |
1416 | * This cpu is also the best, if it has more than one task already. | |
1417 | * Siblings must be also busy(in most cases) as they didn't already | |
1418 | * pickup the extra load from this cpu and hence we need not check | |
1419 | * sibling runqueue info. This will avoid the checks and cache miss | |
1420 | * penalities associated with that. | |
1421 | */ | |
1422 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1423 | return cpu; |
1424 | ||
1425 | for_each_domain(cpu, sd) { | |
1426 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1427 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1428 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1429 | if (idle_cpu(i)) { |
1430 | if (i != task_cpu(p)) { | |
1431 | schedstat_inc(p, | |
1432 | se.nr_wakeups_idle); | |
1433 | } | |
1da177e4 | 1434 | return i; |
cc367732 | 1435 | } |
1da177e4 | 1436 | } |
9761eea8 | 1437 | } else { |
e0f364f4 | 1438 | break; |
9761eea8 | 1439 | } |
1da177e4 LT |
1440 | } |
1441 | return cpu; | |
1442 | } | |
1443 | #else | |
36c8b586 | 1444 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1445 | { |
1446 | return cpu; | |
1447 | } | |
1448 | #endif | |
1449 | ||
1450 | /*** | |
1451 | * try_to_wake_up - wake up a thread | |
1452 | * @p: the to-be-woken-up thread | |
1453 | * @state: the mask of task states that can be woken | |
1454 | * @sync: do a synchronous wakeup? | |
1455 | * | |
1456 | * Put it on the run-queue if it's not already there. The "current" | |
1457 | * thread is always on the run-queue (except when the actual | |
1458 | * re-schedule is in progress), and as such you're allowed to do | |
1459 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1460 | * runnable without the overhead of this. | |
1461 | * | |
1462 | * returns failure only if the task is already active. | |
1463 | */ | |
36c8b586 | 1464 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1465 | { |
cc367732 | 1466 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1467 | unsigned long flags; |
1468 | long old_state; | |
70b97a7f | 1469 | struct rq *rq; |
1da177e4 | 1470 | #ifdef CONFIG_SMP |
7897986b | 1471 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1472 | unsigned long load, this_load; |
1da177e4 LT |
1473 | int new_cpu; |
1474 | #endif | |
1475 | ||
1476 | rq = task_rq_lock(p, &flags); | |
1477 | old_state = p->state; | |
1478 | if (!(old_state & state)) | |
1479 | goto out; | |
1480 | ||
dd41f596 | 1481 | if (p->se.on_rq) |
1da177e4 LT |
1482 | goto out_running; |
1483 | ||
1484 | cpu = task_cpu(p); | |
cc367732 | 1485 | orig_cpu = cpu; |
1da177e4 LT |
1486 | this_cpu = smp_processor_id(); |
1487 | ||
1488 | #ifdef CONFIG_SMP | |
1489 | if (unlikely(task_running(rq, p))) | |
1490 | goto out_activate; | |
1491 | ||
7897986b NP |
1492 | new_cpu = cpu; |
1493 | ||
2d72376b | 1494 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1495 | if (cpu == this_cpu) { |
1496 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1497 | goto out_set_cpu; |
1498 | } | |
1499 | ||
1500 | for_each_domain(this_cpu, sd) { | |
1501 | if (cpu_isset(cpu, sd->span)) { | |
1502 | schedstat_inc(sd, ttwu_wake_remote); | |
1503 | this_sd = sd; | |
1504 | break; | |
1da177e4 LT |
1505 | } |
1506 | } | |
1da177e4 | 1507 | |
7897986b | 1508 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1509 | goto out_set_cpu; |
1510 | ||
1da177e4 | 1511 | /* |
7897986b | 1512 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1513 | */ |
7897986b NP |
1514 | if (this_sd) { |
1515 | int idx = this_sd->wake_idx; | |
1516 | unsigned int imbalance; | |
1da177e4 | 1517 | |
a3f21bce NP |
1518 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1519 | ||
7897986b NP |
1520 | load = source_load(cpu, idx); |
1521 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1522 | |
7897986b NP |
1523 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1524 | ||
a3f21bce NP |
1525 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1526 | unsigned long tl = this_load; | |
33859f7f MOS |
1527 | unsigned long tl_per_task; |
1528 | ||
71e20f18 IM |
1529 | /* |
1530 | * Attract cache-cold tasks on sync wakeups: | |
1531 | */ | |
1532 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1533 | goto out_set_cpu; | |
1534 | ||
cc367732 | 1535 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1536 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1537 | |
1da177e4 | 1538 | /* |
a3f21bce NP |
1539 | * If sync wakeup then subtract the (maximum possible) |
1540 | * effect of the currently running task from the load | |
1541 | * of the current CPU: | |
1da177e4 | 1542 | */ |
a3f21bce | 1543 | if (sync) |
dd41f596 | 1544 | tl -= current->se.load.weight; |
a3f21bce NP |
1545 | |
1546 | if ((tl <= load && | |
2dd73a4f | 1547 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1548 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1549 | /* |
1550 | * This domain has SD_WAKE_AFFINE and | |
1551 | * p is cache cold in this domain, and | |
1552 | * there is no bad imbalance. | |
1553 | */ | |
1554 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1555 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1556 | goto out_set_cpu; |
1557 | } | |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * Start passive balancing when half the imbalance_pct | |
1562 | * limit is reached. | |
1563 | */ | |
1564 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1565 | if (imbalance*this_load <= 100*load) { | |
1566 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1567 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1568 | goto out_set_cpu; |
1569 | } | |
1da177e4 LT |
1570 | } |
1571 | } | |
1572 | ||
1573 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1574 | out_set_cpu: | |
1575 | new_cpu = wake_idle(new_cpu, p); | |
1576 | if (new_cpu != cpu) { | |
1577 | set_task_cpu(p, new_cpu); | |
1578 | task_rq_unlock(rq, &flags); | |
1579 | /* might preempt at this point */ | |
1580 | rq = task_rq_lock(p, &flags); | |
1581 | old_state = p->state; | |
1582 | if (!(old_state & state)) | |
1583 | goto out; | |
dd41f596 | 1584 | if (p->se.on_rq) |
1da177e4 LT |
1585 | goto out_running; |
1586 | ||
1587 | this_cpu = smp_processor_id(); | |
1588 | cpu = task_cpu(p); | |
1589 | } | |
1590 | ||
1591 | out_activate: | |
1592 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1593 | schedstat_inc(p, se.nr_wakeups); |
1594 | if (sync) | |
1595 | schedstat_inc(p, se.nr_wakeups_sync); | |
1596 | if (orig_cpu != cpu) | |
1597 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1598 | if (cpu == this_cpu) | |
1599 | schedstat_inc(p, se.nr_wakeups_local); | |
1600 | else | |
1601 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1602 | update_rq_clock(rq); |
dd41f596 | 1603 | activate_task(rq, p, 1); |
9c63d9c0 | 1604 | check_preempt_curr(rq, p); |
1da177e4 LT |
1605 | success = 1; |
1606 | ||
1607 | out_running: | |
1608 | p->state = TASK_RUNNING; | |
1609 | out: | |
1610 | task_rq_unlock(rq, &flags); | |
1611 | ||
1612 | return success; | |
1613 | } | |
1614 | ||
36c8b586 | 1615 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1616 | { |
1617 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1618 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1619 | } | |
1da177e4 LT |
1620 | EXPORT_SYMBOL(wake_up_process); |
1621 | ||
36c8b586 | 1622 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1623 | { |
1624 | return try_to_wake_up(p, state, 0); | |
1625 | } | |
1626 | ||
1da177e4 LT |
1627 | /* |
1628 | * Perform scheduler related setup for a newly forked process p. | |
1629 | * p is forked by current. | |
dd41f596 IM |
1630 | * |
1631 | * __sched_fork() is basic setup used by init_idle() too: | |
1632 | */ | |
1633 | static void __sched_fork(struct task_struct *p) | |
1634 | { | |
dd41f596 IM |
1635 | p->se.exec_start = 0; |
1636 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1637 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1638 | |
1639 | #ifdef CONFIG_SCHEDSTATS | |
1640 | p->se.wait_start = 0; | |
dd41f596 IM |
1641 | p->se.sum_sleep_runtime = 0; |
1642 | p->se.sleep_start = 0; | |
dd41f596 IM |
1643 | p->se.block_start = 0; |
1644 | p->se.sleep_max = 0; | |
1645 | p->se.block_max = 0; | |
1646 | p->se.exec_max = 0; | |
eba1ed4b | 1647 | p->se.slice_max = 0; |
dd41f596 | 1648 | p->se.wait_max = 0; |
6cfb0d5d | 1649 | #endif |
476d139c | 1650 | |
dd41f596 IM |
1651 | INIT_LIST_HEAD(&p->run_list); |
1652 | p->se.on_rq = 0; | |
476d139c | 1653 | |
e107be36 AK |
1654 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1655 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1656 | #endif | |
1657 | ||
1da177e4 LT |
1658 | /* |
1659 | * We mark the process as running here, but have not actually | |
1660 | * inserted it onto the runqueue yet. This guarantees that | |
1661 | * nobody will actually run it, and a signal or other external | |
1662 | * event cannot wake it up and insert it on the runqueue either. | |
1663 | */ | |
1664 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1665 | } |
1666 | ||
1667 | /* | |
1668 | * fork()/clone()-time setup: | |
1669 | */ | |
1670 | void sched_fork(struct task_struct *p, int clone_flags) | |
1671 | { | |
1672 | int cpu = get_cpu(); | |
1673 | ||
1674 | __sched_fork(p); | |
1675 | ||
1676 | #ifdef CONFIG_SMP | |
1677 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1678 | #endif | |
02e4bac2 | 1679 | set_task_cpu(p, cpu); |
b29739f9 IM |
1680 | |
1681 | /* | |
1682 | * Make sure we do not leak PI boosting priority to the child: | |
1683 | */ | |
1684 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1685 | if (!rt_prio(p->prio)) |
1686 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1687 | |
52f17b6c | 1688 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1689 | if (likely(sched_info_on())) |
52f17b6c | 1690 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1691 | #endif |
d6077cb8 | 1692 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1693 | p->oncpu = 0; |
1694 | #endif | |
1da177e4 | 1695 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1696 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1697 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1698 | #endif |
476d139c | 1699 | put_cpu(); |
1da177e4 LT |
1700 | } |
1701 | ||
1702 | /* | |
1703 | * wake_up_new_task - wake up a newly created task for the first time. | |
1704 | * | |
1705 | * This function will do some initial scheduler statistics housekeeping | |
1706 | * that must be done for every newly created context, then puts the task | |
1707 | * on the runqueue and wakes it. | |
1708 | */ | |
36c8b586 | 1709 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1710 | { |
1711 | unsigned long flags; | |
dd41f596 | 1712 | struct rq *rq; |
1da177e4 LT |
1713 | |
1714 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1715 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1716 | update_rq_clock(rq); |
1da177e4 LT |
1717 | |
1718 | p->prio = effective_prio(p); | |
1719 | ||
b9dca1e0 | 1720 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1721 | activate_task(rq, p, 0); |
1da177e4 | 1722 | } else { |
1da177e4 | 1723 | /* |
dd41f596 IM |
1724 | * Let the scheduling class do new task startup |
1725 | * management (if any): | |
1da177e4 | 1726 | */ |
ee0827d8 | 1727 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1728 | inc_nr_running(p, rq); |
1da177e4 | 1729 | } |
dd41f596 IM |
1730 | check_preempt_curr(rq, p); |
1731 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1732 | } |
1733 | ||
e107be36 AK |
1734 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1735 | ||
1736 | /** | |
421cee29 RD |
1737 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1738 | * @notifier: notifier struct to register | |
e107be36 AK |
1739 | */ |
1740 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1741 | { | |
1742 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1743 | } | |
1744 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1745 | ||
1746 | /** | |
1747 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1748 | * @notifier: notifier struct to unregister |
e107be36 AK |
1749 | * |
1750 | * This is safe to call from within a preemption notifier. | |
1751 | */ | |
1752 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1753 | { | |
1754 | hlist_del(¬ifier->link); | |
1755 | } | |
1756 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1757 | ||
1758 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1759 | { | |
1760 | struct preempt_notifier *notifier; | |
1761 | struct hlist_node *node; | |
1762 | ||
1763 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1764 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1765 | } | |
1766 | ||
1767 | static void | |
1768 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1769 | struct task_struct *next) | |
1770 | { | |
1771 | struct preempt_notifier *notifier; | |
1772 | struct hlist_node *node; | |
1773 | ||
1774 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1775 | notifier->ops->sched_out(notifier, next); | |
1776 | } | |
1777 | ||
1778 | #else | |
1779 | ||
1780 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1781 | { | |
1782 | } | |
1783 | ||
1784 | static void | |
1785 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1786 | struct task_struct *next) | |
1787 | { | |
1788 | } | |
1789 | ||
1790 | #endif | |
1791 | ||
4866cde0 NP |
1792 | /** |
1793 | * prepare_task_switch - prepare to switch tasks | |
1794 | * @rq: the runqueue preparing to switch | |
421cee29 | 1795 | * @prev: the current task that is being switched out |
4866cde0 NP |
1796 | * @next: the task we are going to switch to. |
1797 | * | |
1798 | * This is called with the rq lock held and interrupts off. It must | |
1799 | * be paired with a subsequent finish_task_switch after the context | |
1800 | * switch. | |
1801 | * | |
1802 | * prepare_task_switch sets up locking and calls architecture specific | |
1803 | * hooks. | |
1804 | */ | |
e107be36 AK |
1805 | static inline void |
1806 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1807 | struct task_struct *next) | |
4866cde0 | 1808 | { |
e107be36 | 1809 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1810 | prepare_lock_switch(rq, next); |
1811 | prepare_arch_switch(next); | |
1812 | } | |
1813 | ||
1da177e4 LT |
1814 | /** |
1815 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1816 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1817 | * @prev: the thread we just switched away from. |
1818 | * | |
4866cde0 NP |
1819 | * finish_task_switch must be called after the context switch, paired |
1820 | * with a prepare_task_switch call before the context switch. | |
1821 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1822 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1823 | * |
1824 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1825 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1826 | * with the lock held can cause deadlocks; see schedule() for | |
1827 | * details.) | |
1828 | */ | |
a9957449 | 1829 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1830 | __releases(rq->lock) |
1831 | { | |
1da177e4 | 1832 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1833 | long prev_state; |
1da177e4 LT |
1834 | |
1835 | rq->prev_mm = NULL; | |
1836 | ||
1837 | /* | |
1838 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1839 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1840 | * schedule one last time. The schedule call will never return, and |
1841 | * the scheduled task must drop that reference. | |
c394cc9f | 1842 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1843 | * still held, otherwise prev could be scheduled on another cpu, die |
1844 | * there before we look at prev->state, and then the reference would | |
1845 | * be dropped twice. | |
1846 | * Manfred Spraul <manfred@colorfullife.com> | |
1847 | */ | |
55a101f8 | 1848 | prev_state = prev->state; |
4866cde0 NP |
1849 | finish_arch_switch(prev); |
1850 | finish_lock_switch(rq, prev); | |
e107be36 | 1851 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1852 | if (mm) |
1853 | mmdrop(mm); | |
c394cc9f | 1854 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1855 | /* |
1856 | * Remove function-return probe instances associated with this | |
1857 | * task and put them back on the free list. | |
9761eea8 | 1858 | */ |
c6fd91f0 | 1859 | kprobe_flush_task(prev); |
1da177e4 | 1860 | put_task_struct(prev); |
c6fd91f0 | 1861 | } |
1da177e4 LT |
1862 | } |
1863 | ||
1864 | /** | |
1865 | * schedule_tail - first thing a freshly forked thread must call. | |
1866 | * @prev: the thread we just switched away from. | |
1867 | */ | |
36c8b586 | 1868 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1869 | __releases(rq->lock) |
1870 | { | |
70b97a7f IM |
1871 | struct rq *rq = this_rq(); |
1872 | ||
4866cde0 NP |
1873 | finish_task_switch(rq, prev); |
1874 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1875 | /* In this case, finish_task_switch does not reenable preemption */ | |
1876 | preempt_enable(); | |
1877 | #endif | |
1da177e4 LT |
1878 | if (current->set_child_tid) |
1879 | put_user(current->pid, current->set_child_tid); | |
1880 | } | |
1881 | ||
1882 | /* | |
1883 | * context_switch - switch to the new MM and the new | |
1884 | * thread's register state. | |
1885 | */ | |
dd41f596 | 1886 | static inline void |
70b97a7f | 1887 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1888 | struct task_struct *next) |
1da177e4 | 1889 | { |
dd41f596 | 1890 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1891 | |
e107be36 | 1892 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1893 | mm = next->mm; |
1894 | oldmm = prev->active_mm; | |
9226d125 ZA |
1895 | /* |
1896 | * For paravirt, this is coupled with an exit in switch_to to | |
1897 | * combine the page table reload and the switch backend into | |
1898 | * one hypercall. | |
1899 | */ | |
1900 | arch_enter_lazy_cpu_mode(); | |
1901 | ||
dd41f596 | 1902 | if (unlikely(!mm)) { |
1da177e4 LT |
1903 | next->active_mm = oldmm; |
1904 | atomic_inc(&oldmm->mm_count); | |
1905 | enter_lazy_tlb(oldmm, next); | |
1906 | } else | |
1907 | switch_mm(oldmm, mm, next); | |
1908 | ||
dd41f596 | 1909 | if (unlikely(!prev->mm)) { |
1da177e4 | 1910 | prev->active_mm = NULL; |
1da177e4 LT |
1911 | rq->prev_mm = oldmm; |
1912 | } | |
3a5f5e48 IM |
1913 | /* |
1914 | * Since the runqueue lock will be released by the next | |
1915 | * task (which is an invalid locking op but in the case | |
1916 | * of the scheduler it's an obvious special-case), so we | |
1917 | * do an early lockdep release here: | |
1918 | */ | |
1919 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1920 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1921 | #endif |
1da177e4 LT |
1922 | |
1923 | /* Here we just switch the register state and the stack. */ | |
1924 | switch_to(prev, next, prev); | |
1925 | ||
dd41f596 IM |
1926 | barrier(); |
1927 | /* | |
1928 | * this_rq must be evaluated again because prev may have moved | |
1929 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1930 | * frame will be invalid. | |
1931 | */ | |
1932 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1933 | } |
1934 | ||
1935 | /* | |
1936 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1937 | * | |
1938 | * externally visible scheduler statistics: current number of runnable | |
1939 | * threads, current number of uninterruptible-sleeping threads, total | |
1940 | * number of context switches performed since bootup. | |
1941 | */ | |
1942 | unsigned long nr_running(void) | |
1943 | { | |
1944 | unsigned long i, sum = 0; | |
1945 | ||
1946 | for_each_online_cpu(i) | |
1947 | sum += cpu_rq(i)->nr_running; | |
1948 | ||
1949 | return sum; | |
1950 | } | |
1951 | ||
1952 | unsigned long nr_uninterruptible(void) | |
1953 | { | |
1954 | unsigned long i, sum = 0; | |
1955 | ||
0a945022 | 1956 | for_each_possible_cpu(i) |
1da177e4 LT |
1957 | sum += cpu_rq(i)->nr_uninterruptible; |
1958 | ||
1959 | /* | |
1960 | * Since we read the counters lockless, it might be slightly | |
1961 | * inaccurate. Do not allow it to go below zero though: | |
1962 | */ | |
1963 | if (unlikely((long)sum < 0)) | |
1964 | sum = 0; | |
1965 | ||
1966 | return sum; | |
1967 | } | |
1968 | ||
1969 | unsigned long long nr_context_switches(void) | |
1970 | { | |
cc94abfc SR |
1971 | int i; |
1972 | unsigned long long sum = 0; | |
1da177e4 | 1973 | |
0a945022 | 1974 | for_each_possible_cpu(i) |
1da177e4 LT |
1975 | sum += cpu_rq(i)->nr_switches; |
1976 | ||
1977 | return sum; | |
1978 | } | |
1979 | ||
1980 | unsigned long nr_iowait(void) | |
1981 | { | |
1982 | unsigned long i, sum = 0; | |
1983 | ||
0a945022 | 1984 | for_each_possible_cpu(i) |
1da177e4 LT |
1985 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1986 | ||
1987 | return sum; | |
1988 | } | |
1989 | ||
db1b1fef JS |
1990 | unsigned long nr_active(void) |
1991 | { | |
1992 | unsigned long i, running = 0, uninterruptible = 0; | |
1993 | ||
1994 | for_each_online_cpu(i) { | |
1995 | running += cpu_rq(i)->nr_running; | |
1996 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1997 | } | |
1998 | ||
1999 | if (unlikely((long)uninterruptible < 0)) | |
2000 | uninterruptible = 0; | |
2001 | ||
2002 | return running + uninterruptible; | |
2003 | } | |
2004 | ||
48f24c4d | 2005 | /* |
dd41f596 IM |
2006 | * Update rq->cpu_load[] statistics. This function is usually called every |
2007 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2008 | */ |
dd41f596 | 2009 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2010 | { |
495eca49 | 2011 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2012 | int i, scale; |
2013 | ||
2014 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2015 | |
2016 | /* Update our load: */ | |
2017 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2018 | unsigned long old_load, new_load; | |
2019 | ||
2020 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2021 | ||
2022 | old_load = this_rq->cpu_load[i]; | |
2023 | new_load = this_load; | |
a25707f3 IM |
2024 | /* |
2025 | * Round up the averaging division if load is increasing. This | |
2026 | * prevents us from getting stuck on 9 if the load is 10, for | |
2027 | * example. | |
2028 | */ | |
2029 | if (new_load > old_load) | |
2030 | new_load += scale-1; | |
dd41f596 IM |
2031 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2032 | } | |
48f24c4d IM |
2033 | } |
2034 | ||
dd41f596 IM |
2035 | #ifdef CONFIG_SMP |
2036 | ||
1da177e4 LT |
2037 | /* |
2038 | * double_rq_lock - safely lock two runqueues | |
2039 | * | |
2040 | * Note this does not disable interrupts like task_rq_lock, | |
2041 | * you need to do so manually before calling. | |
2042 | */ | |
70b97a7f | 2043 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2044 | __acquires(rq1->lock) |
2045 | __acquires(rq2->lock) | |
2046 | { | |
054b9108 | 2047 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2048 | if (rq1 == rq2) { |
2049 | spin_lock(&rq1->lock); | |
2050 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2051 | } else { | |
c96d145e | 2052 | if (rq1 < rq2) { |
1da177e4 LT |
2053 | spin_lock(&rq1->lock); |
2054 | spin_lock(&rq2->lock); | |
2055 | } else { | |
2056 | spin_lock(&rq2->lock); | |
2057 | spin_lock(&rq1->lock); | |
2058 | } | |
2059 | } | |
6e82a3be IM |
2060 | update_rq_clock(rq1); |
2061 | update_rq_clock(rq2); | |
1da177e4 LT |
2062 | } |
2063 | ||
2064 | /* | |
2065 | * double_rq_unlock - safely unlock two runqueues | |
2066 | * | |
2067 | * Note this does not restore interrupts like task_rq_unlock, | |
2068 | * you need to do so manually after calling. | |
2069 | */ | |
70b97a7f | 2070 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2071 | __releases(rq1->lock) |
2072 | __releases(rq2->lock) | |
2073 | { | |
2074 | spin_unlock(&rq1->lock); | |
2075 | if (rq1 != rq2) | |
2076 | spin_unlock(&rq2->lock); | |
2077 | else | |
2078 | __release(rq2->lock); | |
2079 | } | |
2080 | ||
2081 | /* | |
2082 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2083 | */ | |
70b97a7f | 2084 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2085 | __releases(this_rq->lock) |
2086 | __acquires(busiest->lock) | |
2087 | __acquires(this_rq->lock) | |
2088 | { | |
054b9108 KK |
2089 | if (unlikely(!irqs_disabled())) { |
2090 | /* printk() doesn't work good under rq->lock */ | |
2091 | spin_unlock(&this_rq->lock); | |
2092 | BUG_ON(1); | |
2093 | } | |
1da177e4 | 2094 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2095 | if (busiest < this_rq) { |
1da177e4 LT |
2096 | spin_unlock(&this_rq->lock); |
2097 | spin_lock(&busiest->lock); | |
2098 | spin_lock(&this_rq->lock); | |
2099 | } else | |
2100 | spin_lock(&busiest->lock); | |
2101 | } | |
2102 | } | |
2103 | ||
1da177e4 LT |
2104 | /* |
2105 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2106 | * This is accomplished by forcing the cpu_allowed mask to only | |
2107 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2108 | * the cpu_allowed mask is restored. | |
2109 | */ | |
36c8b586 | 2110 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2111 | { |
70b97a7f | 2112 | struct migration_req req; |
1da177e4 | 2113 | unsigned long flags; |
70b97a7f | 2114 | struct rq *rq; |
1da177e4 LT |
2115 | |
2116 | rq = task_rq_lock(p, &flags); | |
2117 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2118 | || unlikely(cpu_is_offline(dest_cpu))) | |
2119 | goto out; | |
2120 | ||
2121 | /* force the process onto the specified CPU */ | |
2122 | if (migrate_task(p, dest_cpu, &req)) { | |
2123 | /* Need to wait for migration thread (might exit: take ref). */ | |
2124 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2125 | |
1da177e4 LT |
2126 | get_task_struct(mt); |
2127 | task_rq_unlock(rq, &flags); | |
2128 | wake_up_process(mt); | |
2129 | put_task_struct(mt); | |
2130 | wait_for_completion(&req.done); | |
36c8b586 | 2131 | |
1da177e4 LT |
2132 | return; |
2133 | } | |
2134 | out: | |
2135 | task_rq_unlock(rq, &flags); | |
2136 | } | |
2137 | ||
2138 | /* | |
476d139c NP |
2139 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2140 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2141 | */ |
2142 | void sched_exec(void) | |
2143 | { | |
1da177e4 | 2144 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2145 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2146 | put_cpu(); |
476d139c NP |
2147 | if (new_cpu != this_cpu) |
2148 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2149 | } |
2150 | ||
2151 | /* | |
2152 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2153 | * Both runqueues must be locked. | |
2154 | */ | |
dd41f596 IM |
2155 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2156 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2157 | { |
2e1cb74a | 2158 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2159 | set_task_cpu(p, this_cpu); |
dd41f596 | 2160 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2161 | /* |
2162 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2163 | * to be always true for them. | |
2164 | */ | |
dd41f596 | 2165 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2166 | } |
2167 | ||
2168 | /* | |
2169 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2170 | */ | |
858119e1 | 2171 | static |
70b97a7f | 2172 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2173 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2174 | int *all_pinned) |
1da177e4 LT |
2175 | { |
2176 | /* | |
2177 | * We do not migrate tasks that are: | |
2178 | * 1) running (obviously), or | |
2179 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2180 | * 3) are cache-hot on their current CPU. | |
2181 | */ | |
cc367732 IM |
2182 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2183 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2184 | return 0; |
cc367732 | 2185 | } |
81026794 NP |
2186 | *all_pinned = 0; |
2187 | ||
cc367732 IM |
2188 | if (task_running(rq, p)) { |
2189 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2190 | return 0; |
cc367732 | 2191 | } |
1da177e4 | 2192 | |
da84d961 IM |
2193 | /* |
2194 | * Aggressive migration if: | |
2195 | * 1) task is cache cold, or | |
2196 | * 2) too many balance attempts have failed. | |
2197 | */ | |
2198 | ||
6bc1665b IM |
2199 | if (!task_hot(p, rq->clock, sd) || |
2200 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2201 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2202 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2203 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2204 | schedstat_inc(p, se.nr_forced_migrations); |
2205 | } | |
da84d961 IM |
2206 | #endif |
2207 | return 1; | |
2208 | } | |
2209 | ||
cc367732 IM |
2210 | if (task_hot(p, rq->clock, sd)) { |
2211 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2212 | return 0; |
cc367732 | 2213 | } |
1da177e4 LT |
2214 | return 1; |
2215 | } | |
2216 | ||
dd41f596 | 2217 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2218 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2219 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2220 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2221 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2222 | { |
dd41f596 IM |
2223 | int pulled = 0, pinned = 0, skip_for_load; |
2224 | struct task_struct *p; | |
2225 | long rem_load_move = max_load_move; | |
1da177e4 | 2226 | |
2dd73a4f | 2227 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2228 | goto out; |
2229 | ||
81026794 NP |
2230 | pinned = 1; |
2231 | ||
1da177e4 | 2232 | /* |
dd41f596 | 2233 | * Start the load-balancing iterator: |
1da177e4 | 2234 | */ |
dd41f596 IM |
2235 | p = iterator->start(iterator->arg); |
2236 | next: | |
2237 | if (!p) | |
1da177e4 | 2238 | goto out; |
50ddd969 PW |
2239 | /* |
2240 | * To help distribute high priority tasks accross CPUs we don't | |
2241 | * skip a task if it will be the highest priority task (i.e. smallest | |
2242 | * prio value) on its new queue regardless of its load weight | |
2243 | */ | |
dd41f596 IM |
2244 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2245 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2246 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2247 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2248 | p = iterator->next(iterator->arg); |
2249 | goto next; | |
1da177e4 LT |
2250 | } |
2251 | ||
dd41f596 | 2252 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2253 | pulled++; |
dd41f596 | 2254 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2255 | |
2dd73a4f PW |
2256 | /* |
2257 | * We only want to steal up to the prescribed number of tasks | |
2258 | * and the prescribed amount of weighted load. | |
2259 | */ | |
2260 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2261 | if (p->prio < *this_best_prio) |
2262 | *this_best_prio = p->prio; | |
dd41f596 IM |
2263 | p = iterator->next(iterator->arg); |
2264 | goto next; | |
1da177e4 LT |
2265 | } |
2266 | out: | |
2267 | /* | |
2268 | * Right now, this is the only place pull_task() is called, | |
2269 | * so we can safely collect pull_task() stats here rather than | |
2270 | * inside pull_task(). | |
2271 | */ | |
2272 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2273 | |
2274 | if (all_pinned) | |
2275 | *all_pinned = pinned; | |
dd41f596 | 2276 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2277 | return pulled; |
2278 | } | |
2279 | ||
dd41f596 | 2280 | /* |
43010659 PW |
2281 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2282 | * this_rq, as part of a balancing operation within domain "sd". | |
2283 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2284 | * |
2285 | * Called with both runqueues locked. | |
2286 | */ | |
2287 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2288 | unsigned long max_load_move, |
dd41f596 IM |
2289 | struct sched_domain *sd, enum cpu_idle_type idle, |
2290 | int *all_pinned) | |
2291 | { | |
5522d5d5 | 2292 | const struct sched_class *class = sched_class_highest; |
43010659 | 2293 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2294 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2295 | |
2296 | do { | |
43010659 PW |
2297 | total_load_moved += |
2298 | class->load_balance(this_rq, this_cpu, busiest, | |
2299 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2300 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2301 | class = class->next; |
43010659 | 2302 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2303 | |
43010659 PW |
2304 | return total_load_moved > 0; |
2305 | } | |
2306 | ||
2307 | /* | |
2308 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2309 | * part of active balancing operations within "domain". | |
2310 | * Returns 1 if successful and 0 otherwise. | |
2311 | * | |
2312 | * Called with both runqueues locked. | |
2313 | */ | |
2314 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2315 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2316 | { | |
5522d5d5 | 2317 | const struct sched_class *class; |
a4ac01c3 | 2318 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2319 | |
2320 | for (class = sched_class_highest; class; class = class->next) | |
2321 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2322 | 1, ULONG_MAX, sd, idle, NULL, |
2323 | &this_best_prio)) | |
43010659 PW |
2324 | return 1; |
2325 | ||
2326 | return 0; | |
dd41f596 IM |
2327 | } |
2328 | ||
1da177e4 LT |
2329 | /* |
2330 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2331 | * domain. It calculates and returns the amount of weighted load which |
2332 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2333 | */ |
2334 | static struct sched_group * | |
2335 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2336 | unsigned long *imbalance, enum cpu_idle_type idle, |
2337 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2338 | { |
2339 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2340 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2341 | unsigned long max_pull; |
2dd73a4f PW |
2342 | unsigned long busiest_load_per_task, busiest_nr_running; |
2343 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2344 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2345 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2346 | int power_savings_balance = 1; | |
2347 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2348 | unsigned long min_nr_running = ULONG_MAX; | |
2349 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2350 | #endif | |
1da177e4 LT |
2351 | |
2352 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2353 | busiest_load_per_task = busiest_nr_running = 0; |
2354 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2355 | if (idle == CPU_NOT_IDLE) |
7897986b | 2356 | load_idx = sd->busy_idx; |
d15bcfdb | 2357 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2358 | load_idx = sd->newidle_idx; |
2359 | else | |
2360 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2361 | |
2362 | do { | |
908a7c1b | 2363 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2364 | int local_group; |
2365 | int i; | |
908a7c1b | 2366 | int __group_imb = 0; |
783609c6 | 2367 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2368 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2369 | |
2370 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2371 | ||
783609c6 SS |
2372 | if (local_group) |
2373 | balance_cpu = first_cpu(group->cpumask); | |
2374 | ||
1da177e4 | 2375 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2376 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2377 | max_cpu_load = 0; |
2378 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2379 | |
2380 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2381 | struct rq *rq; |
2382 | ||
2383 | if (!cpu_isset(i, *cpus)) | |
2384 | continue; | |
2385 | ||
2386 | rq = cpu_rq(i); | |
2dd73a4f | 2387 | |
9439aab8 | 2388 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2389 | *sd_idle = 0; |
2390 | ||
1da177e4 | 2391 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2392 | if (local_group) { |
2393 | if (idle_cpu(i) && !first_idle_cpu) { | |
2394 | first_idle_cpu = 1; | |
2395 | balance_cpu = i; | |
2396 | } | |
2397 | ||
a2000572 | 2398 | load = target_load(i, load_idx); |
908a7c1b | 2399 | } else { |
a2000572 | 2400 | load = source_load(i, load_idx); |
908a7c1b KC |
2401 | if (load > max_cpu_load) |
2402 | max_cpu_load = load; | |
2403 | if (min_cpu_load > load) | |
2404 | min_cpu_load = load; | |
2405 | } | |
1da177e4 LT |
2406 | |
2407 | avg_load += load; | |
2dd73a4f | 2408 | sum_nr_running += rq->nr_running; |
dd41f596 | 2409 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2410 | } |
2411 | ||
783609c6 SS |
2412 | /* |
2413 | * First idle cpu or the first cpu(busiest) in this sched group | |
2414 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2415 | * domains. In the newly idle case, we will allow all the cpu's |
2416 | * to do the newly idle load balance. | |
783609c6 | 2417 | */ |
9439aab8 SS |
2418 | if (idle != CPU_NEWLY_IDLE && local_group && |
2419 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2420 | *balance = 0; |
2421 | goto ret; | |
2422 | } | |
2423 | ||
1da177e4 | 2424 | total_load += avg_load; |
5517d86b | 2425 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2426 | |
2427 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2428 | avg_load = sg_div_cpu_power(group, |
2429 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2430 | |
908a7c1b KC |
2431 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2432 | __group_imb = 1; | |
2433 | ||
5517d86b | 2434 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2435 | |
1da177e4 LT |
2436 | if (local_group) { |
2437 | this_load = avg_load; | |
2438 | this = group; | |
2dd73a4f PW |
2439 | this_nr_running = sum_nr_running; |
2440 | this_load_per_task = sum_weighted_load; | |
2441 | } else if (avg_load > max_load && | |
908a7c1b | 2442 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2443 | max_load = avg_load; |
2444 | busiest = group; | |
2dd73a4f PW |
2445 | busiest_nr_running = sum_nr_running; |
2446 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2447 | group_imb = __group_imb; |
1da177e4 | 2448 | } |
5c45bf27 SS |
2449 | |
2450 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2451 | /* | |
2452 | * Busy processors will not participate in power savings | |
2453 | * balance. | |
2454 | */ | |
dd41f596 IM |
2455 | if (idle == CPU_NOT_IDLE || |
2456 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2457 | goto group_next; | |
5c45bf27 SS |
2458 | |
2459 | /* | |
2460 | * If the local group is idle or completely loaded | |
2461 | * no need to do power savings balance at this domain | |
2462 | */ | |
2463 | if (local_group && (this_nr_running >= group_capacity || | |
2464 | !this_nr_running)) | |
2465 | power_savings_balance = 0; | |
2466 | ||
dd41f596 | 2467 | /* |
5c45bf27 SS |
2468 | * If a group is already running at full capacity or idle, |
2469 | * don't include that group in power savings calculations | |
dd41f596 IM |
2470 | */ |
2471 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2472 | || !sum_nr_running) |
dd41f596 | 2473 | goto group_next; |
5c45bf27 | 2474 | |
dd41f596 | 2475 | /* |
5c45bf27 | 2476 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2477 | * This is the group from where we need to pick up the load |
2478 | * for saving power | |
2479 | */ | |
2480 | if ((sum_nr_running < min_nr_running) || | |
2481 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2482 | first_cpu(group->cpumask) < |
2483 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2484 | group_min = group; |
2485 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2486 | min_load_per_task = sum_weighted_load / |
2487 | sum_nr_running; | |
dd41f596 | 2488 | } |
5c45bf27 | 2489 | |
dd41f596 | 2490 | /* |
5c45bf27 | 2491 | * Calculate the group which is almost near its |
dd41f596 IM |
2492 | * capacity but still has some space to pick up some load |
2493 | * from other group and save more power | |
2494 | */ | |
2495 | if (sum_nr_running <= group_capacity - 1) { | |
2496 | if (sum_nr_running > leader_nr_running || | |
2497 | (sum_nr_running == leader_nr_running && | |
2498 | first_cpu(group->cpumask) > | |
2499 | first_cpu(group_leader->cpumask))) { | |
2500 | group_leader = group; | |
2501 | leader_nr_running = sum_nr_running; | |
2502 | } | |
48f24c4d | 2503 | } |
5c45bf27 SS |
2504 | group_next: |
2505 | #endif | |
1da177e4 LT |
2506 | group = group->next; |
2507 | } while (group != sd->groups); | |
2508 | ||
2dd73a4f | 2509 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2510 | goto out_balanced; |
2511 | ||
2512 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2513 | ||
2514 | if (this_load >= avg_load || | |
2515 | 100*max_load <= sd->imbalance_pct*this_load) | |
2516 | goto out_balanced; | |
2517 | ||
2dd73a4f | 2518 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2519 | if (group_imb) |
2520 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2521 | ||
1da177e4 LT |
2522 | /* |
2523 | * We're trying to get all the cpus to the average_load, so we don't | |
2524 | * want to push ourselves above the average load, nor do we wish to | |
2525 | * reduce the max loaded cpu below the average load, as either of these | |
2526 | * actions would just result in more rebalancing later, and ping-pong | |
2527 | * tasks around. Thus we look for the minimum possible imbalance. | |
2528 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2529 | * be counted as no imbalance for these purposes -- we can't fix that | |
2530 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2531 | * appear as very large values with unsigned longs. | |
2532 | */ | |
2dd73a4f PW |
2533 | if (max_load <= busiest_load_per_task) |
2534 | goto out_balanced; | |
2535 | ||
2536 | /* | |
2537 | * In the presence of smp nice balancing, certain scenarios can have | |
2538 | * max load less than avg load(as we skip the groups at or below | |
2539 | * its cpu_power, while calculating max_load..) | |
2540 | */ | |
2541 | if (max_load < avg_load) { | |
2542 | *imbalance = 0; | |
2543 | goto small_imbalance; | |
2544 | } | |
0c117f1b SS |
2545 | |
2546 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2547 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2548 | |
1da177e4 | 2549 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2550 | *imbalance = min(max_pull * busiest->__cpu_power, |
2551 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2552 | / SCHED_LOAD_SCALE; |
2553 | ||
2dd73a4f PW |
2554 | /* |
2555 | * if *imbalance is less than the average load per runnable task | |
2556 | * there is no gaurantee that any tasks will be moved so we'll have | |
2557 | * a think about bumping its value to force at least one task to be | |
2558 | * moved | |
2559 | */ | |
7fd0d2dd | 2560 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2561 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2562 | unsigned int imbn; |
2563 | ||
2564 | small_imbalance: | |
2565 | pwr_move = pwr_now = 0; | |
2566 | imbn = 2; | |
2567 | if (this_nr_running) { | |
2568 | this_load_per_task /= this_nr_running; | |
2569 | if (busiest_load_per_task > this_load_per_task) | |
2570 | imbn = 1; | |
2571 | } else | |
2572 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2573 | |
dd41f596 IM |
2574 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2575 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2576 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2577 | return busiest; |
2578 | } | |
2579 | ||
2580 | /* | |
2581 | * OK, we don't have enough imbalance to justify moving tasks, | |
2582 | * however we may be able to increase total CPU power used by | |
2583 | * moving them. | |
2584 | */ | |
2585 | ||
5517d86b ED |
2586 | pwr_now += busiest->__cpu_power * |
2587 | min(busiest_load_per_task, max_load); | |
2588 | pwr_now += this->__cpu_power * | |
2589 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2590 | pwr_now /= SCHED_LOAD_SCALE; |
2591 | ||
2592 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2593 | tmp = sg_div_cpu_power(busiest, |
2594 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2595 | if (max_load > tmp) |
5517d86b | 2596 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2597 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2598 | |
2599 | /* Amount of load we'd add */ | |
5517d86b | 2600 | if (max_load * busiest->__cpu_power < |
33859f7f | 2601 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2602 | tmp = sg_div_cpu_power(this, |
2603 | max_load * busiest->__cpu_power); | |
1da177e4 | 2604 | else |
5517d86b ED |
2605 | tmp = sg_div_cpu_power(this, |
2606 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2607 | pwr_move += this->__cpu_power * | |
2608 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2609 | pwr_move /= SCHED_LOAD_SCALE; |
2610 | ||
2611 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2612 | if (pwr_move > pwr_now) |
2613 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2614 | } |
2615 | ||
1da177e4 LT |
2616 | return busiest; |
2617 | ||
2618 | out_balanced: | |
5c45bf27 | 2619 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2620 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2621 | goto ret; |
1da177e4 | 2622 | |
5c45bf27 SS |
2623 | if (this == group_leader && group_leader != group_min) { |
2624 | *imbalance = min_load_per_task; | |
2625 | return group_min; | |
2626 | } | |
5c45bf27 | 2627 | #endif |
783609c6 | 2628 | ret: |
1da177e4 LT |
2629 | *imbalance = 0; |
2630 | return NULL; | |
2631 | } | |
2632 | ||
2633 | /* | |
2634 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2635 | */ | |
70b97a7f | 2636 | static struct rq * |
d15bcfdb | 2637 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2638 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2639 | { |
70b97a7f | 2640 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2641 | unsigned long max_load = 0; |
1da177e4 LT |
2642 | int i; |
2643 | ||
2644 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2645 | unsigned long wl; |
0a2966b4 CL |
2646 | |
2647 | if (!cpu_isset(i, *cpus)) | |
2648 | continue; | |
2649 | ||
48f24c4d | 2650 | rq = cpu_rq(i); |
dd41f596 | 2651 | wl = weighted_cpuload(i); |
2dd73a4f | 2652 | |
dd41f596 | 2653 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2654 | continue; |
1da177e4 | 2655 | |
dd41f596 IM |
2656 | if (wl > max_load) { |
2657 | max_load = wl; | |
48f24c4d | 2658 | busiest = rq; |
1da177e4 LT |
2659 | } |
2660 | } | |
2661 | ||
2662 | return busiest; | |
2663 | } | |
2664 | ||
77391d71 NP |
2665 | /* |
2666 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2667 | * so long as it is large enough. | |
2668 | */ | |
2669 | #define MAX_PINNED_INTERVAL 512 | |
2670 | ||
1da177e4 LT |
2671 | /* |
2672 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2673 | * tasks if there is an imbalance. | |
1da177e4 | 2674 | */ |
70b97a7f | 2675 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2676 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2677 | int *balance) |
1da177e4 | 2678 | { |
43010659 | 2679 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2680 | struct sched_group *group; |
1da177e4 | 2681 | unsigned long imbalance; |
70b97a7f | 2682 | struct rq *busiest; |
0a2966b4 | 2683 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2684 | unsigned long flags; |
5969fe06 | 2685 | |
89c4710e SS |
2686 | /* |
2687 | * When power savings policy is enabled for the parent domain, idle | |
2688 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2689 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2690 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2691 | */ |
d15bcfdb | 2692 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2693 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2694 | sd_idle = 1; |
1da177e4 | 2695 | |
2d72376b | 2696 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2697 | |
0a2966b4 CL |
2698 | redo: |
2699 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2700 | &cpus, balance); |
2701 | ||
06066714 | 2702 | if (*balance == 0) |
783609c6 | 2703 | goto out_balanced; |
783609c6 | 2704 | |
1da177e4 LT |
2705 | if (!group) { |
2706 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2707 | goto out_balanced; | |
2708 | } | |
2709 | ||
0a2966b4 | 2710 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2711 | if (!busiest) { |
2712 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2713 | goto out_balanced; | |
2714 | } | |
2715 | ||
db935dbd | 2716 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2717 | |
2718 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2719 | ||
43010659 | 2720 | ld_moved = 0; |
1da177e4 LT |
2721 | if (busiest->nr_running > 1) { |
2722 | /* | |
2723 | * Attempt to move tasks. If find_busiest_group has found | |
2724 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2725 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2726 | * correctly treated as an imbalance. |
2727 | */ | |
fe2eea3f | 2728 | local_irq_save(flags); |
e17224bf | 2729 | double_rq_lock(this_rq, busiest); |
43010659 | 2730 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2731 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2732 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2733 | local_irq_restore(flags); |
81026794 | 2734 | |
46cb4b7c SS |
2735 | /* |
2736 | * some other cpu did the load balance for us. | |
2737 | */ | |
43010659 | 2738 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2739 | resched_cpu(this_cpu); |
2740 | ||
81026794 | 2741 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2742 | if (unlikely(all_pinned)) { |
2743 | cpu_clear(cpu_of(busiest), cpus); | |
2744 | if (!cpus_empty(cpus)) | |
2745 | goto redo; | |
81026794 | 2746 | goto out_balanced; |
0a2966b4 | 2747 | } |
1da177e4 | 2748 | } |
81026794 | 2749 | |
43010659 | 2750 | if (!ld_moved) { |
1da177e4 LT |
2751 | schedstat_inc(sd, lb_failed[idle]); |
2752 | sd->nr_balance_failed++; | |
2753 | ||
2754 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2755 | |
fe2eea3f | 2756 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2757 | |
2758 | /* don't kick the migration_thread, if the curr | |
2759 | * task on busiest cpu can't be moved to this_cpu | |
2760 | */ | |
2761 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2762 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2763 | all_pinned = 1; |
2764 | goto out_one_pinned; | |
2765 | } | |
2766 | ||
1da177e4 LT |
2767 | if (!busiest->active_balance) { |
2768 | busiest->active_balance = 1; | |
2769 | busiest->push_cpu = this_cpu; | |
81026794 | 2770 | active_balance = 1; |
1da177e4 | 2771 | } |
fe2eea3f | 2772 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2773 | if (active_balance) |
1da177e4 LT |
2774 | wake_up_process(busiest->migration_thread); |
2775 | ||
2776 | /* | |
2777 | * We've kicked active balancing, reset the failure | |
2778 | * counter. | |
2779 | */ | |
39507451 | 2780 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2781 | } |
81026794 | 2782 | } else |
1da177e4 LT |
2783 | sd->nr_balance_failed = 0; |
2784 | ||
81026794 | 2785 | if (likely(!active_balance)) { |
1da177e4 LT |
2786 | /* We were unbalanced, so reset the balancing interval */ |
2787 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2788 | } else { |
2789 | /* | |
2790 | * If we've begun active balancing, start to back off. This | |
2791 | * case may not be covered by the all_pinned logic if there | |
2792 | * is only 1 task on the busy runqueue (because we don't call | |
2793 | * move_tasks). | |
2794 | */ | |
2795 | if (sd->balance_interval < sd->max_interval) | |
2796 | sd->balance_interval *= 2; | |
1da177e4 LT |
2797 | } |
2798 | ||
43010659 | 2799 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2800 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2801 | return -1; |
43010659 | 2802 | return ld_moved; |
1da177e4 LT |
2803 | |
2804 | out_balanced: | |
1da177e4 LT |
2805 | schedstat_inc(sd, lb_balanced[idle]); |
2806 | ||
16cfb1c0 | 2807 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2808 | |
2809 | out_one_pinned: | |
1da177e4 | 2810 | /* tune up the balancing interval */ |
77391d71 NP |
2811 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2812 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2813 | sd->balance_interval *= 2; |
2814 | ||
48f24c4d | 2815 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2816 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2817 | return -1; |
1da177e4 LT |
2818 | return 0; |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2823 | * tasks if there is an imbalance. | |
2824 | * | |
d15bcfdb | 2825 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2826 | * this_rq is locked. |
2827 | */ | |
48f24c4d | 2828 | static int |
70b97a7f | 2829 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2830 | { |
2831 | struct sched_group *group; | |
70b97a7f | 2832 | struct rq *busiest = NULL; |
1da177e4 | 2833 | unsigned long imbalance; |
43010659 | 2834 | int ld_moved = 0; |
5969fe06 | 2835 | int sd_idle = 0; |
969bb4e4 | 2836 | int all_pinned = 0; |
0a2966b4 | 2837 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2838 | |
89c4710e SS |
2839 | /* |
2840 | * When power savings policy is enabled for the parent domain, idle | |
2841 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2842 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2843 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2844 | */ |
2845 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2846 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2847 | sd_idle = 1; |
1da177e4 | 2848 | |
2d72376b | 2849 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2850 | redo: |
d15bcfdb | 2851 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2852 | &sd_idle, &cpus, NULL); |
1da177e4 | 2853 | if (!group) { |
d15bcfdb | 2854 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2855 | goto out_balanced; |
1da177e4 LT |
2856 | } |
2857 | ||
d15bcfdb | 2858 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2859 | &cpus); |
db935dbd | 2860 | if (!busiest) { |
d15bcfdb | 2861 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2862 | goto out_balanced; |
1da177e4 LT |
2863 | } |
2864 | ||
db935dbd NP |
2865 | BUG_ON(busiest == this_rq); |
2866 | ||
d15bcfdb | 2867 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2868 | |
43010659 | 2869 | ld_moved = 0; |
d6d5cfaf NP |
2870 | if (busiest->nr_running > 1) { |
2871 | /* Attempt to move tasks */ | |
2872 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2873 | /* this_rq->clock is already updated */ |
2874 | update_rq_clock(busiest); | |
43010659 | 2875 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2876 | imbalance, sd, CPU_NEWLY_IDLE, |
2877 | &all_pinned); | |
d6d5cfaf | 2878 | spin_unlock(&busiest->lock); |
0a2966b4 | 2879 | |
969bb4e4 | 2880 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2881 | cpu_clear(cpu_of(busiest), cpus); |
2882 | if (!cpus_empty(cpus)) | |
2883 | goto redo; | |
2884 | } | |
d6d5cfaf NP |
2885 | } |
2886 | ||
43010659 | 2887 | if (!ld_moved) { |
d15bcfdb | 2888 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2889 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2890 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2891 | return -1; |
2892 | } else | |
16cfb1c0 | 2893 | sd->nr_balance_failed = 0; |
1da177e4 | 2894 | |
43010659 | 2895 | return ld_moved; |
16cfb1c0 NP |
2896 | |
2897 | out_balanced: | |
d15bcfdb | 2898 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2899 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2900 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2901 | return -1; |
16cfb1c0 | 2902 | sd->nr_balance_failed = 0; |
48f24c4d | 2903 | |
16cfb1c0 | 2904 | return 0; |
1da177e4 LT |
2905 | } |
2906 | ||
2907 | /* | |
2908 | * idle_balance is called by schedule() if this_cpu is about to become | |
2909 | * idle. Attempts to pull tasks from other CPUs. | |
2910 | */ | |
70b97a7f | 2911 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2912 | { |
2913 | struct sched_domain *sd; | |
dd41f596 IM |
2914 | int pulled_task = -1; |
2915 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2916 | |
2917 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2918 | unsigned long interval; |
2919 | ||
2920 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2921 | continue; | |
2922 | ||
2923 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2924 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2925 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2926 | this_rq, sd); |
2927 | ||
2928 | interval = msecs_to_jiffies(sd->balance_interval); | |
2929 | if (time_after(next_balance, sd->last_balance + interval)) | |
2930 | next_balance = sd->last_balance + interval; | |
2931 | if (pulled_task) | |
2932 | break; | |
1da177e4 | 2933 | } |
dd41f596 | 2934 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2935 | /* |
2936 | * We are going idle. next_balance may be set based on | |
2937 | * a busy processor. So reset next_balance. | |
2938 | */ | |
2939 | this_rq->next_balance = next_balance; | |
dd41f596 | 2940 | } |
1da177e4 LT |
2941 | } |
2942 | ||
2943 | /* | |
2944 | * active_load_balance is run by migration threads. It pushes running tasks | |
2945 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2946 | * running on each physical CPU where possible, and avoids physical / | |
2947 | * logical imbalances. | |
2948 | * | |
2949 | * Called with busiest_rq locked. | |
2950 | */ | |
70b97a7f | 2951 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2952 | { |
39507451 | 2953 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2954 | struct sched_domain *sd; |
2955 | struct rq *target_rq; | |
39507451 | 2956 | |
48f24c4d | 2957 | /* Is there any task to move? */ |
39507451 | 2958 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2959 | return; |
2960 | ||
2961 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2962 | |
2963 | /* | |
39507451 NP |
2964 | * This condition is "impossible", if it occurs |
2965 | * we need to fix it. Originally reported by | |
2966 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2967 | */ |
39507451 | 2968 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2969 | |
39507451 NP |
2970 | /* move a task from busiest_rq to target_rq */ |
2971 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2972 | update_rq_clock(busiest_rq); |
2973 | update_rq_clock(target_rq); | |
39507451 NP |
2974 | |
2975 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2976 | for_each_domain(target_cpu, sd) { |
39507451 | 2977 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2978 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2979 | break; |
c96d145e | 2980 | } |
39507451 | 2981 | |
48f24c4d | 2982 | if (likely(sd)) { |
2d72376b | 2983 | schedstat_inc(sd, alb_count); |
39507451 | 2984 | |
43010659 PW |
2985 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2986 | sd, CPU_IDLE)) | |
48f24c4d IM |
2987 | schedstat_inc(sd, alb_pushed); |
2988 | else | |
2989 | schedstat_inc(sd, alb_failed); | |
2990 | } | |
39507451 | 2991 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2992 | } |
2993 | ||
46cb4b7c SS |
2994 | #ifdef CONFIG_NO_HZ |
2995 | static struct { | |
2996 | atomic_t load_balancer; | |
2997 | cpumask_t cpu_mask; | |
2998 | } nohz ____cacheline_aligned = { | |
2999 | .load_balancer = ATOMIC_INIT(-1), | |
3000 | .cpu_mask = CPU_MASK_NONE, | |
3001 | }; | |
3002 | ||
7835b98b | 3003 | /* |
46cb4b7c SS |
3004 | * This routine will try to nominate the ilb (idle load balancing) |
3005 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3006 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3007 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3008 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3009 | * arrives... | |
3010 | * | |
3011 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3012 | * for idle load balancing. ilb owner will still be part of | |
3013 | * nohz.cpu_mask.. | |
7835b98b | 3014 | * |
46cb4b7c SS |
3015 | * While stopping the tick, this cpu will become the ilb owner if there |
3016 | * is no other owner. And will be the owner till that cpu becomes busy | |
3017 | * or if all cpus in the system stop their ticks at which point | |
3018 | * there is no need for ilb owner. | |
3019 | * | |
3020 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3021 | * next busy scheduler_tick() | |
3022 | */ | |
3023 | int select_nohz_load_balancer(int stop_tick) | |
3024 | { | |
3025 | int cpu = smp_processor_id(); | |
3026 | ||
3027 | if (stop_tick) { | |
3028 | cpu_set(cpu, nohz.cpu_mask); | |
3029 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3030 | ||
3031 | /* | |
3032 | * If we are going offline and still the leader, give up! | |
3033 | */ | |
3034 | if (cpu_is_offline(cpu) && | |
3035 | atomic_read(&nohz.load_balancer) == cpu) { | |
3036 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3037 | BUG(); | |
3038 | return 0; | |
3039 | } | |
3040 | ||
3041 | /* time for ilb owner also to sleep */ | |
3042 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3043 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3044 | atomic_set(&nohz.load_balancer, -1); | |
3045 | return 0; | |
3046 | } | |
3047 | ||
3048 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3049 | /* make me the ilb owner */ | |
3050 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3051 | return 1; | |
3052 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3053 | return 1; | |
3054 | } else { | |
3055 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3056 | return 0; | |
3057 | ||
3058 | cpu_clear(cpu, nohz.cpu_mask); | |
3059 | ||
3060 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3061 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3062 | BUG(); | |
3063 | } | |
3064 | return 0; | |
3065 | } | |
3066 | #endif | |
3067 | ||
3068 | static DEFINE_SPINLOCK(balancing); | |
3069 | ||
3070 | /* | |
7835b98b CL |
3071 | * It checks each scheduling domain to see if it is due to be balanced, |
3072 | * and initiates a balancing operation if so. | |
3073 | * | |
3074 | * Balancing parameters are set up in arch_init_sched_domains. | |
3075 | */ | |
a9957449 | 3076 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3077 | { |
46cb4b7c SS |
3078 | int balance = 1; |
3079 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3080 | unsigned long interval; |
3081 | struct sched_domain *sd; | |
46cb4b7c | 3082 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3083 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3084 | int update_next_balance = 0; |
1da177e4 | 3085 | |
46cb4b7c | 3086 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3087 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3088 | continue; | |
3089 | ||
3090 | interval = sd->balance_interval; | |
d15bcfdb | 3091 | if (idle != CPU_IDLE) |
1da177e4 LT |
3092 | interval *= sd->busy_factor; |
3093 | ||
3094 | /* scale ms to jiffies */ | |
3095 | interval = msecs_to_jiffies(interval); | |
3096 | if (unlikely(!interval)) | |
3097 | interval = 1; | |
dd41f596 IM |
3098 | if (interval > HZ*NR_CPUS/10) |
3099 | interval = HZ*NR_CPUS/10; | |
3100 | ||
1da177e4 | 3101 | |
08c183f3 CL |
3102 | if (sd->flags & SD_SERIALIZE) { |
3103 | if (!spin_trylock(&balancing)) | |
3104 | goto out; | |
3105 | } | |
3106 | ||
c9819f45 | 3107 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3108 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3109 | /* |
3110 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3111 | * longer idle, or one of our SMT siblings is |
3112 | * not idle. | |
3113 | */ | |
d15bcfdb | 3114 | idle = CPU_NOT_IDLE; |
1da177e4 | 3115 | } |
1bd77f2d | 3116 | sd->last_balance = jiffies; |
1da177e4 | 3117 | } |
08c183f3 CL |
3118 | if (sd->flags & SD_SERIALIZE) |
3119 | spin_unlock(&balancing); | |
3120 | out: | |
f549da84 | 3121 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3122 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3123 | update_next_balance = 1; |
3124 | } | |
783609c6 SS |
3125 | |
3126 | /* | |
3127 | * Stop the load balance at this level. There is another | |
3128 | * CPU in our sched group which is doing load balancing more | |
3129 | * actively. | |
3130 | */ | |
3131 | if (!balance) | |
3132 | break; | |
1da177e4 | 3133 | } |
f549da84 SS |
3134 | |
3135 | /* | |
3136 | * next_balance will be updated only when there is a need. | |
3137 | * When the cpu is attached to null domain for ex, it will not be | |
3138 | * updated. | |
3139 | */ | |
3140 | if (likely(update_next_balance)) | |
3141 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3142 | } |
3143 | ||
3144 | /* | |
3145 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3146 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3147 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3148 | */ | |
3149 | static void run_rebalance_domains(struct softirq_action *h) | |
3150 | { | |
dd41f596 IM |
3151 | int this_cpu = smp_processor_id(); |
3152 | struct rq *this_rq = cpu_rq(this_cpu); | |
3153 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3154 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3155 | |
dd41f596 | 3156 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3157 | |
3158 | #ifdef CONFIG_NO_HZ | |
3159 | /* | |
3160 | * If this cpu is the owner for idle load balancing, then do the | |
3161 | * balancing on behalf of the other idle cpus whose ticks are | |
3162 | * stopped. | |
3163 | */ | |
dd41f596 IM |
3164 | if (this_rq->idle_at_tick && |
3165 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3166 | cpumask_t cpus = nohz.cpu_mask; |
3167 | struct rq *rq; | |
3168 | int balance_cpu; | |
3169 | ||
dd41f596 | 3170 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3171 | for_each_cpu_mask(balance_cpu, cpus) { |
3172 | /* | |
3173 | * If this cpu gets work to do, stop the load balancing | |
3174 | * work being done for other cpus. Next load | |
3175 | * balancing owner will pick it up. | |
3176 | */ | |
3177 | if (need_resched()) | |
3178 | break; | |
3179 | ||
de0cf899 | 3180 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3181 | |
3182 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3183 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3184 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3185 | } |
3186 | } | |
3187 | #endif | |
3188 | } | |
3189 | ||
3190 | /* | |
3191 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3192 | * | |
3193 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3194 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3195 | * if the whole system is idle. | |
3196 | */ | |
dd41f596 | 3197 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3198 | { |
46cb4b7c SS |
3199 | #ifdef CONFIG_NO_HZ |
3200 | /* | |
3201 | * If we were in the nohz mode recently and busy at the current | |
3202 | * scheduler tick, then check if we need to nominate new idle | |
3203 | * load balancer. | |
3204 | */ | |
3205 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3206 | rq->in_nohz_recently = 0; | |
3207 | ||
3208 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3209 | cpu_clear(cpu, nohz.cpu_mask); | |
3210 | atomic_set(&nohz.load_balancer, -1); | |
3211 | } | |
3212 | ||
3213 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3214 | /* | |
3215 | * simple selection for now: Nominate the | |
3216 | * first cpu in the nohz list to be the next | |
3217 | * ilb owner. | |
3218 | * | |
3219 | * TBD: Traverse the sched domains and nominate | |
3220 | * the nearest cpu in the nohz.cpu_mask. | |
3221 | */ | |
3222 | int ilb = first_cpu(nohz.cpu_mask); | |
3223 | ||
3224 | if (ilb != NR_CPUS) | |
3225 | resched_cpu(ilb); | |
3226 | } | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * If this cpu is idle and doing idle load balancing for all the | |
3231 | * cpus with ticks stopped, is it time for that to stop? | |
3232 | */ | |
3233 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3234 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3235 | resched_cpu(cpu); | |
3236 | return; | |
3237 | } | |
3238 | ||
3239 | /* | |
3240 | * If this cpu is idle and the idle load balancing is done by | |
3241 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3242 | */ | |
3243 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3244 | cpu_isset(cpu, nohz.cpu_mask)) | |
3245 | return; | |
3246 | #endif | |
3247 | if (time_after_eq(jiffies, rq->next_balance)) | |
3248 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3249 | } |
dd41f596 IM |
3250 | |
3251 | #else /* CONFIG_SMP */ | |
3252 | ||
1da177e4 LT |
3253 | /* |
3254 | * on UP we do not need to balance between CPUs: | |
3255 | */ | |
70b97a7f | 3256 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3257 | { |
3258 | } | |
dd41f596 IM |
3259 | |
3260 | /* Avoid "used but not defined" warning on UP */ | |
3261 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3262 | unsigned long max_nr_move, unsigned long max_load_move, | |
3263 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3264 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3265 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3266 | { |
3267 | *load_moved = 0; | |
3268 | ||
3269 | return 0; | |
3270 | } | |
3271 | ||
1da177e4 LT |
3272 | #endif |
3273 | ||
1da177e4 LT |
3274 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3275 | ||
3276 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3277 | ||
3278 | /* | |
41b86e9c IM |
3279 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3280 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3281 | */ |
41b86e9c | 3282 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3283 | { |
1da177e4 | 3284 | unsigned long flags; |
41b86e9c IM |
3285 | u64 ns, delta_exec; |
3286 | struct rq *rq; | |
48f24c4d | 3287 | |
41b86e9c IM |
3288 | rq = task_rq_lock(p, &flags); |
3289 | ns = p->se.sum_exec_runtime; | |
3290 | if (rq->curr == p) { | |
a8e504d2 IM |
3291 | update_rq_clock(rq); |
3292 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3293 | if ((s64)delta_exec > 0) |
3294 | ns += delta_exec; | |
3295 | } | |
3296 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3297 | |
1da177e4 LT |
3298 | return ns; |
3299 | } | |
3300 | ||
1da177e4 LT |
3301 | /* |
3302 | * Account user cpu time to a process. | |
3303 | * @p: the process that the cpu time gets accounted to | |
3304 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3305 | * @cputime: the cpu time spent in user space since the last update | |
3306 | */ | |
3307 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3308 | { | |
3309 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3310 | cputime64_t tmp; | |
62d0df64 | 3311 | struct rq *rq = this_rq(); |
1da177e4 LT |
3312 | |
3313 | p->utime = cputime_add(p->utime, cputime); | |
3314 | ||
62d0df64 PM |
3315 | if (p != rq->idle) |
3316 | cpuacct_charge(p, cputime); | |
3317 | ||
1da177e4 LT |
3318 | /* Add user time to cpustat. */ |
3319 | tmp = cputime_to_cputime64(cputime); | |
3320 | if (TASK_NICE(p) > 0) | |
3321 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3322 | else | |
3323 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3324 | } | |
3325 | ||
94886b84 LV |
3326 | /* |
3327 | * Account guest cpu time to a process. | |
3328 | * @p: the process that the cpu time gets accounted to | |
3329 | * @cputime: the cpu time spent in virtual machine since the last update | |
3330 | */ | |
3331 | void account_guest_time(struct task_struct *p, cputime_t cputime) | |
3332 | { | |
3333 | cputime64_t tmp; | |
3334 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3335 | ||
3336 | tmp = cputime_to_cputime64(cputime); | |
3337 | ||
3338 | p->utime = cputime_add(p->utime, cputime); | |
3339 | p->gtime = cputime_add(p->gtime, cputime); | |
3340 | ||
3341 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3342 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3343 | } | |
3344 | ||
c66f08be MN |
3345 | /* |
3346 | * Account scaled user cpu time to a process. | |
3347 | * @p: the process that the cpu time gets accounted to | |
3348 | * @cputime: the cpu time spent in user space since the last update | |
3349 | */ | |
3350 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3351 | { | |
3352 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3353 | } | |
3354 | ||
1da177e4 LT |
3355 | /* |
3356 | * Account system cpu time to a process. | |
3357 | * @p: the process that the cpu time gets accounted to | |
3358 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3359 | * @cputime: the cpu time spent in kernel space since the last update | |
3360 | */ | |
3361 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3362 | cputime_t cputime) | |
3363 | { | |
3364 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3365 | struct rq *rq = this_rq(); |
1da177e4 LT |
3366 | cputime64_t tmp; |
3367 | ||
94886b84 LV |
3368 | if (p->flags & PF_VCPU) { |
3369 | account_guest_time(p, cputime); | |
3370 | p->flags &= ~PF_VCPU; | |
3371 | return; | |
3372 | } | |
3373 | ||
1da177e4 LT |
3374 | p->stime = cputime_add(p->stime, cputime); |
3375 | ||
3376 | /* Add system time to cpustat. */ | |
3377 | tmp = cputime_to_cputime64(cputime); | |
3378 | if (hardirq_count() - hardirq_offset) | |
3379 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3380 | else if (softirq_count()) | |
3381 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
62d0df64 | 3382 | else if (p != rq->idle) { |
1da177e4 | 3383 | cpustat->system = cputime64_add(cpustat->system, tmp); |
62d0df64 PM |
3384 | cpuacct_charge(p, cputime); |
3385 | } else if (atomic_read(&rq->nr_iowait) > 0) | |
1da177e4 LT |
3386 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3387 | else | |
3388 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3389 | /* Account for system time used */ | |
3390 | acct_update_integrals(p); | |
1da177e4 LT |
3391 | } |
3392 | ||
c66f08be MN |
3393 | /* |
3394 | * Account scaled system cpu time to a process. | |
3395 | * @p: the process that the cpu time gets accounted to | |
3396 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3397 | * @cputime: the cpu time spent in kernel space since the last update | |
3398 | */ | |
3399 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3400 | { | |
3401 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3402 | } | |
3403 | ||
1da177e4 LT |
3404 | /* |
3405 | * Account for involuntary wait time. | |
3406 | * @p: the process from which the cpu time has been stolen | |
3407 | * @steal: the cpu time spent in involuntary wait | |
3408 | */ | |
3409 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3410 | { | |
3411 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3412 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3413 | struct rq *rq = this_rq(); |
1da177e4 LT |
3414 | |
3415 | if (p == rq->idle) { | |
3416 | p->stime = cputime_add(p->stime, steal); | |
3417 | if (atomic_read(&rq->nr_iowait) > 0) | |
3418 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3419 | else | |
3420 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
62d0df64 | 3421 | } else { |
1da177e4 | 3422 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
62d0df64 PM |
3423 | cpuacct_charge(p, -tmp); |
3424 | } | |
1da177e4 LT |
3425 | } |
3426 | ||
7835b98b CL |
3427 | /* |
3428 | * This function gets called by the timer code, with HZ frequency. | |
3429 | * We call it with interrupts disabled. | |
3430 | * | |
3431 | * It also gets called by the fork code, when changing the parent's | |
3432 | * timeslices. | |
3433 | */ | |
3434 | void scheduler_tick(void) | |
3435 | { | |
7835b98b CL |
3436 | int cpu = smp_processor_id(); |
3437 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3438 | struct task_struct *curr = rq->curr; |
529c7726 | 3439 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3440 | |
3441 | spin_lock(&rq->lock); | |
546fe3c9 | 3442 | __update_rq_clock(rq); |
529c7726 IM |
3443 | /* |
3444 | * Let rq->clock advance by at least TICK_NSEC: | |
3445 | */ | |
3446 | if (unlikely(rq->clock < next_tick)) | |
3447 | rq->clock = next_tick; | |
3448 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3449 | update_cpu_load(rq); |
dd41f596 IM |
3450 | if (curr != rq->idle) /* FIXME: needed? */ |
3451 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3452 | spin_unlock(&rq->lock); |
7835b98b | 3453 | |
e418e1c2 | 3454 | #ifdef CONFIG_SMP |
dd41f596 IM |
3455 | rq->idle_at_tick = idle_cpu(cpu); |
3456 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3457 | #endif |
1da177e4 LT |
3458 | } |
3459 | ||
1da177e4 LT |
3460 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3461 | ||
3462 | void fastcall add_preempt_count(int val) | |
3463 | { | |
3464 | /* | |
3465 | * Underflow? | |
3466 | */ | |
9a11b49a IM |
3467 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3468 | return; | |
1da177e4 LT |
3469 | preempt_count() += val; |
3470 | /* | |
3471 | * Spinlock count overflowing soon? | |
3472 | */ | |
33859f7f MOS |
3473 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3474 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3475 | } |
3476 | EXPORT_SYMBOL(add_preempt_count); | |
3477 | ||
3478 | void fastcall sub_preempt_count(int val) | |
3479 | { | |
3480 | /* | |
3481 | * Underflow? | |
3482 | */ | |
9a11b49a IM |
3483 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3484 | return; | |
1da177e4 LT |
3485 | /* |
3486 | * Is the spinlock portion underflowing? | |
3487 | */ | |
9a11b49a IM |
3488 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3489 | !(preempt_count() & PREEMPT_MASK))) | |
3490 | return; | |
3491 | ||
1da177e4 LT |
3492 | preempt_count() -= val; |
3493 | } | |
3494 | EXPORT_SYMBOL(sub_preempt_count); | |
3495 | ||
3496 | #endif | |
3497 | ||
3498 | /* | |
dd41f596 | 3499 | * Print scheduling while atomic bug: |
1da177e4 | 3500 | */ |
dd41f596 | 3501 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3502 | { |
dd41f596 IM |
3503 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3504 | prev->comm, preempt_count(), prev->pid); | |
3505 | debug_show_held_locks(prev); | |
3506 | if (irqs_disabled()) | |
3507 | print_irqtrace_events(prev); | |
3508 | dump_stack(); | |
3509 | } | |
1da177e4 | 3510 | |
dd41f596 IM |
3511 | /* |
3512 | * Various schedule()-time debugging checks and statistics: | |
3513 | */ | |
3514 | static inline void schedule_debug(struct task_struct *prev) | |
3515 | { | |
1da177e4 LT |
3516 | /* |
3517 | * Test if we are atomic. Since do_exit() needs to call into | |
3518 | * schedule() atomically, we ignore that path for now. | |
3519 | * Otherwise, whine if we are scheduling when we should not be. | |
3520 | */ | |
dd41f596 IM |
3521 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3522 | __schedule_bug(prev); | |
3523 | ||
1da177e4 LT |
3524 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3525 | ||
2d72376b | 3526 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3527 | #ifdef CONFIG_SCHEDSTATS |
3528 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3529 | schedstat_inc(this_rq(), bkl_count); |
3530 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3531 | } |
3532 | #endif | |
dd41f596 IM |
3533 | } |
3534 | ||
3535 | /* | |
3536 | * Pick up the highest-prio task: | |
3537 | */ | |
3538 | static inline struct task_struct * | |
ff95f3df | 3539 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3540 | { |
5522d5d5 | 3541 | const struct sched_class *class; |
dd41f596 | 3542 | struct task_struct *p; |
1da177e4 LT |
3543 | |
3544 | /* | |
dd41f596 IM |
3545 | * Optimization: we know that if all tasks are in |
3546 | * the fair class we can call that function directly: | |
1da177e4 | 3547 | */ |
dd41f596 | 3548 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3549 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3550 | if (likely(p)) |
3551 | return p; | |
1da177e4 LT |
3552 | } |
3553 | ||
dd41f596 IM |
3554 | class = sched_class_highest; |
3555 | for ( ; ; ) { | |
fb8d4724 | 3556 | p = class->pick_next_task(rq); |
dd41f596 IM |
3557 | if (p) |
3558 | return p; | |
3559 | /* | |
3560 | * Will never be NULL as the idle class always | |
3561 | * returns a non-NULL p: | |
3562 | */ | |
3563 | class = class->next; | |
3564 | } | |
3565 | } | |
1da177e4 | 3566 | |
dd41f596 IM |
3567 | /* |
3568 | * schedule() is the main scheduler function. | |
3569 | */ | |
3570 | asmlinkage void __sched schedule(void) | |
3571 | { | |
3572 | struct task_struct *prev, *next; | |
3573 | long *switch_count; | |
3574 | struct rq *rq; | |
dd41f596 IM |
3575 | int cpu; |
3576 | ||
3577 | need_resched: | |
3578 | preempt_disable(); | |
3579 | cpu = smp_processor_id(); | |
3580 | rq = cpu_rq(cpu); | |
3581 | rcu_qsctr_inc(cpu); | |
3582 | prev = rq->curr; | |
3583 | switch_count = &prev->nivcsw; | |
3584 | ||
3585 | release_kernel_lock(prev); | |
3586 | need_resched_nonpreemptible: | |
3587 | ||
3588 | schedule_debug(prev); | |
1da177e4 | 3589 | |
1e819950 IM |
3590 | /* |
3591 | * Do the rq-clock update outside the rq lock: | |
3592 | */ | |
3593 | local_irq_disable(); | |
c1b3da3e | 3594 | __update_rq_clock(rq); |
1e819950 IM |
3595 | spin_lock(&rq->lock); |
3596 | clear_tsk_need_resched(prev); | |
1da177e4 | 3597 | |
1da177e4 | 3598 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3599 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3600 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3601 | prev->state = TASK_RUNNING; |
dd41f596 | 3602 | } else { |
2e1cb74a | 3603 | deactivate_task(rq, prev, 1); |
1da177e4 | 3604 | } |
dd41f596 | 3605 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3606 | } |
3607 | ||
dd41f596 | 3608 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3609 | idle_balance(cpu, rq); |
1da177e4 | 3610 | |
31ee529c | 3611 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3612 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3613 | |
3614 | sched_info_switch(prev, next); | |
dd41f596 | 3615 | |
1da177e4 | 3616 | if (likely(prev != next)) { |
1da177e4 LT |
3617 | rq->nr_switches++; |
3618 | rq->curr = next; | |
3619 | ++*switch_count; | |
3620 | ||
dd41f596 | 3621 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3622 | } else |
3623 | spin_unlock_irq(&rq->lock); | |
3624 | ||
dd41f596 IM |
3625 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3626 | cpu = smp_processor_id(); | |
3627 | rq = cpu_rq(cpu); | |
1da177e4 | 3628 | goto need_resched_nonpreemptible; |
dd41f596 | 3629 | } |
1da177e4 LT |
3630 | preempt_enable_no_resched(); |
3631 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3632 | goto need_resched; | |
3633 | } | |
1da177e4 LT |
3634 | EXPORT_SYMBOL(schedule); |
3635 | ||
3636 | #ifdef CONFIG_PREEMPT | |
3637 | /* | |
2ed6e34f | 3638 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3639 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3640 | * occur there and call schedule directly. | |
3641 | */ | |
3642 | asmlinkage void __sched preempt_schedule(void) | |
3643 | { | |
3644 | struct thread_info *ti = current_thread_info(); | |
3645 | #ifdef CONFIG_PREEMPT_BKL | |
3646 | struct task_struct *task = current; | |
3647 | int saved_lock_depth; | |
3648 | #endif | |
3649 | /* | |
3650 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3651 | * we do not want to preempt the current task. Just return.. | |
3652 | */ | |
beed33a8 | 3653 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3654 | return; |
3655 | ||
3a5c359a AK |
3656 | do { |
3657 | add_preempt_count(PREEMPT_ACTIVE); | |
3658 | ||
3659 | /* | |
3660 | * We keep the big kernel semaphore locked, but we | |
3661 | * clear ->lock_depth so that schedule() doesnt | |
3662 | * auto-release the semaphore: | |
3663 | */ | |
1da177e4 | 3664 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3665 | saved_lock_depth = task->lock_depth; |
3666 | task->lock_depth = -1; | |
1da177e4 | 3667 | #endif |
3a5c359a | 3668 | schedule(); |
1da177e4 | 3669 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3670 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3671 | #endif |
3a5c359a | 3672 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3673 | |
3a5c359a AK |
3674 | /* |
3675 | * Check again in case we missed a preemption opportunity | |
3676 | * between schedule and now. | |
3677 | */ | |
3678 | barrier(); | |
3679 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3680 | } |
1da177e4 LT |
3681 | EXPORT_SYMBOL(preempt_schedule); |
3682 | ||
3683 | /* | |
2ed6e34f | 3684 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3685 | * off of irq context. |
3686 | * Note, that this is called and return with irqs disabled. This will | |
3687 | * protect us against recursive calling from irq. | |
3688 | */ | |
3689 | asmlinkage void __sched preempt_schedule_irq(void) | |
3690 | { | |
3691 | struct thread_info *ti = current_thread_info(); | |
3692 | #ifdef CONFIG_PREEMPT_BKL | |
3693 | struct task_struct *task = current; | |
3694 | int saved_lock_depth; | |
3695 | #endif | |
2ed6e34f | 3696 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3697 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3698 | ||
3a5c359a AK |
3699 | do { |
3700 | add_preempt_count(PREEMPT_ACTIVE); | |
3701 | ||
3702 | /* | |
3703 | * We keep the big kernel semaphore locked, but we | |
3704 | * clear ->lock_depth so that schedule() doesnt | |
3705 | * auto-release the semaphore: | |
3706 | */ | |
1da177e4 | 3707 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3708 | saved_lock_depth = task->lock_depth; |
3709 | task->lock_depth = -1; | |
1da177e4 | 3710 | #endif |
3a5c359a AK |
3711 | local_irq_enable(); |
3712 | schedule(); | |
3713 | local_irq_disable(); | |
1da177e4 | 3714 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3715 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3716 | #endif |
3a5c359a | 3717 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3718 | |
3a5c359a AK |
3719 | /* |
3720 | * Check again in case we missed a preemption opportunity | |
3721 | * between schedule and now. | |
3722 | */ | |
3723 | barrier(); | |
3724 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3725 | } |
3726 | ||
3727 | #endif /* CONFIG_PREEMPT */ | |
3728 | ||
95cdf3b7 IM |
3729 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3730 | void *key) | |
1da177e4 | 3731 | { |
48f24c4d | 3732 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3733 | } |
1da177e4 LT |
3734 | EXPORT_SYMBOL(default_wake_function); |
3735 | ||
3736 | /* | |
3737 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3738 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3739 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3740 | * | |
3741 | * There are circumstances in which we can try to wake a task which has already | |
3742 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3743 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3744 | */ | |
3745 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3746 | int nr_exclusive, int sync, void *key) | |
3747 | { | |
2e45874c | 3748 | wait_queue_t *curr, *next; |
1da177e4 | 3749 | |
2e45874c | 3750 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3751 | unsigned flags = curr->flags; |
3752 | ||
1da177e4 | 3753 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3754 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3755 | break; |
3756 | } | |
3757 | } | |
3758 | ||
3759 | /** | |
3760 | * __wake_up - wake up threads blocked on a waitqueue. | |
3761 | * @q: the waitqueue | |
3762 | * @mode: which threads | |
3763 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3764 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3765 | */ |
3766 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3767 | int nr_exclusive, void *key) |
1da177e4 LT |
3768 | { |
3769 | unsigned long flags; | |
3770 | ||
3771 | spin_lock_irqsave(&q->lock, flags); | |
3772 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3773 | spin_unlock_irqrestore(&q->lock, flags); | |
3774 | } | |
1da177e4 LT |
3775 | EXPORT_SYMBOL(__wake_up); |
3776 | ||
3777 | /* | |
3778 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3779 | */ | |
3780 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3781 | { | |
3782 | __wake_up_common(q, mode, 1, 0, NULL); | |
3783 | } | |
3784 | ||
3785 | /** | |
67be2dd1 | 3786 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3787 | * @q: the waitqueue |
3788 | * @mode: which threads | |
3789 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3790 | * | |
3791 | * The sync wakeup differs that the waker knows that it will schedule | |
3792 | * away soon, so while the target thread will be woken up, it will not | |
3793 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3794 | * with each other. This can prevent needless bouncing between CPUs. | |
3795 | * | |
3796 | * On UP it can prevent extra preemption. | |
3797 | */ | |
95cdf3b7 IM |
3798 | void fastcall |
3799 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3800 | { |
3801 | unsigned long flags; | |
3802 | int sync = 1; | |
3803 | ||
3804 | if (unlikely(!q)) | |
3805 | return; | |
3806 | ||
3807 | if (unlikely(!nr_exclusive)) | |
3808 | sync = 0; | |
3809 | ||
3810 | spin_lock_irqsave(&q->lock, flags); | |
3811 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3812 | spin_unlock_irqrestore(&q->lock, flags); | |
3813 | } | |
3814 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3815 | ||
3816 | void fastcall complete(struct completion *x) | |
3817 | { | |
3818 | unsigned long flags; | |
3819 | ||
3820 | spin_lock_irqsave(&x->wait.lock, flags); | |
3821 | x->done++; | |
3822 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3823 | 1, 0, NULL); | |
3824 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3825 | } | |
3826 | EXPORT_SYMBOL(complete); | |
3827 | ||
3828 | void fastcall complete_all(struct completion *x) | |
3829 | { | |
3830 | unsigned long flags; | |
3831 | ||
3832 | spin_lock_irqsave(&x->wait.lock, flags); | |
3833 | x->done += UINT_MAX/2; | |
3834 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3835 | 0, 0, NULL); | |
3836 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3837 | } | |
3838 | EXPORT_SYMBOL(complete_all); | |
3839 | ||
8cbbe86d AK |
3840 | static inline long __sched |
3841 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3842 | { |
1da177e4 LT |
3843 | if (!x->done) { |
3844 | DECLARE_WAITQUEUE(wait, current); | |
3845 | ||
3846 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3847 | __add_wait_queue_tail(&x->wait, &wait); | |
3848 | do { | |
8cbbe86d AK |
3849 | if (state == TASK_INTERRUPTIBLE && |
3850 | signal_pending(current)) { | |
3851 | __remove_wait_queue(&x->wait, &wait); | |
3852 | return -ERESTARTSYS; | |
3853 | } | |
3854 | __set_current_state(state); | |
1da177e4 LT |
3855 | spin_unlock_irq(&x->wait.lock); |
3856 | timeout = schedule_timeout(timeout); | |
3857 | spin_lock_irq(&x->wait.lock); | |
3858 | if (!timeout) { | |
3859 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3860 | return timeout; |
1da177e4 LT |
3861 | } |
3862 | } while (!x->done); | |
3863 | __remove_wait_queue(&x->wait, &wait); | |
3864 | } | |
3865 | x->done--; | |
1da177e4 LT |
3866 | return timeout; |
3867 | } | |
1da177e4 | 3868 | |
8cbbe86d AK |
3869 | static long __sched |
3870 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3871 | { |
1da177e4 LT |
3872 | might_sleep(); |
3873 | ||
3874 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3875 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3876 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3877 | return timeout; |
3878 | } | |
1da177e4 | 3879 | |
8cbbe86d AK |
3880 | void fastcall __sched wait_for_completion(struct completion *x) |
3881 | { | |
3882 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3883 | } |
8cbbe86d | 3884 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 LT |
3885 | |
3886 | unsigned long fastcall __sched | |
8cbbe86d | 3887 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3888 | { |
8cbbe86d | 3889 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3890 | } |
8cbbe86d | 3891 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3892 | |
8cbbe86d | 3893 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3894 | { |
51e97990 AK |
3895 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3896 | if (t == -ERESTARTSYS) | |
3897 | return t; | |
3898 | return 0; | |
0fec171c | 3899 | } |
8cbbe86d | 3900 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3901 | |
8cbbe86d AK |
3902 | unsigned long fastcall __sched |
3903 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3904 | unsigned long timeout) | |
0fec171c | 3905 | { |
8cbbe86d | 3906 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3907 | } |
8cbbe86d | 3908 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3909 | |
8cbbe86d AK |
3910 | static long __sched |
3911 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3912 | { |
0fec171c IM |
3913 | unsigned long flags; |
3914 | wait_queue_t wait; | |
3915 | ||
3916 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3917 | |
8cbbe86d | 3918 | __set_current_state(state); |
1da177e4 | 3919 | |
8cbbe86d AK |
3920 | spin_lock_irqsave(&q->lock, flags); |
3921 | __add_wait_queue(q, &wait); | |
3922 | spin_unlock(&q->lock); | |
3923 | timeout = schedule_timeout(timeout); | |
3924 | spin_lock_irq(&q->lock); | |
3925 | __remove_wait_queue(q, &wait); | |
3926 | spin_unlock_irqrestore(&q->lock, flags); | |
3927 | ||
3928 | return timeout; | |
3929 | } | |
3930 | ||
3931 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3932 | { | |
3933 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3934 | } |
1da177e4 LT |
3935 | EXPORT_SYMBOL(interruptible_sleep_on); |
3936 | ||
0fec171c | 3937 | long __sched |
95cdf3b7 | 3938 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3939 | { |
8cbbe86d | 3940 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3941 | } |
1da177e4 LT |
3942 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3943 | ||
0fec171c | 3944 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3945 | { |
8cbbe86d | 3946 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3947 | } |
1da177e4 LT |
3948 | EXPORT_SYMBOL(sleep_on); |
3949 | ||
0fec171c | 3950 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3951 | { |
8cbbe86d | 3952 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3953 | } |
1da177e4 LT |
3954 | EXPORT_SYMBOL(sleep_on_timeout); |
3955 | ||
b29739f9 IM |
3956 | #ifdef CONFIG_RT_MUTEXES |
3957 | ||
3958 | /* | |
3959 | * rt_mutex_setprio - set the current priority of a task | |
3960 | * @p: task | |
3961 | * @prio: prio value (kernel-internal form) | |
3962 | * | |
3963 | * This function changes the 'effective' priority of a task. It does | |
3964 | * not touch ->normal_prio like __setscheduler(). | |
3965 | * | |
3966 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3967 | */ | |
36c8b586 | 3968 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3969 | { |
3970 | unsigned long flags; | |
83b699ed | 3971 | int oldprio, on_rq, running; |
70b97a7f | 3972 | struct rq *rq; |
b29739f9 IM |
3973 | |
3974 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3975 | ||
3976 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3977 | update_rq_clock(rq); |
b29739f9 | 3978 | |
d5f9f942 | 3979 | oldprio = p->prio; |
dd41f596 | 3980 | on_rq = p->se.on_rq; |
83b699ed SV |
3981 | running = task_running(rq, p); |
3982 | if (on_rq) { | |
69be72c1 | 3983 | dequeue_task(rq, p, 0); |
83b699ed SV |
3984 | if (running) |
3985 | p->sched_class->put_prev_task(rq, p); | |
3986 | } | |
dd41f596 IM |
3987 | |
3988 | if (rt_prio(prio)) | |
3989 | p->sched_class = &rt_sched_class; | |
3990 | else | |
3991 | p->sched_class = &fair_sched_class; | |
3992 | ||
b29739f9 IM |
3993 | p->prio = prio; |
3994 | ||
dd41f596 | 3995 | if (on_rq) { |
83b699ed SV |
3996 | if (running) |
3997 | p->sched_class->set_curr_task(rq); | |
8159f87e | 3998 | enqueue_task(rq, p, 0); |
b29739f9 IM |
3999 | /* |
4000 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4001 | * our priority decreased, or if we are not currently running on |
4002 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4003 | */ |
83b699ed | 4004 | if (running) { |
d5f9f942 AM |
4005 | if (p->prio > oldprio) |
4006 | resched_task(rq->curr); | |
dd41f596 IM |
4007 | } else { |
4008 | check_preempt_curr(rq, p); | |
4009 | } | |
b29739f9 IM |
4010 | } |
4011 | task_rq_unlock(rq, &flags); | |
4012 | } | |
4013 | ||
4014 | #endif | |
4015 | ||
36c8b586 | 4016 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4017 | { |
dd41f596 | 4018 | int old_prio, delta, on_rq; |
1da177e4 | 4019 | unsigned long flags; |
70b97a7f | 4020 | struct rq *rq; |
1da177e4 LT |
4021 | |
4022 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4023 | return; | |
4024 | /* | |
4025 | * We have to be careful, if called from sys_setpriority(), | |
4026 | * the task might be in the middle of scheduling on another CPU. | |
4027 | */ | |
4028 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4029 | update_rq_clock(rq); |
1da177e4 LT |
4030 | /* |
4031 | * The RT priorities are set via sched_setscheduler(), but we still | |
4032 | * allow the 'normal' nice value to be set - but as expected | |
4033 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4034 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4035 | */ |
e05606d3 | 4036 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4037 | p->static_prio = NICE_TO_PRIO(nice); |
4038 | goto out_unlock; | |
4039 | } | |
dd41f596 IM |
4040 | on_rq = p->se.on_rq; |
4041 | if (on_rq) { | |
69be72c1 | 4042 | dequeue_task(rq, p, 0); |
79b5dddf | 4043 | dec_load(rq, p); |
2dd73a4f | 4044 | } |
1da177e4 | 4045 | |
1da177e4 | 4046 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4047 | set_load_weight(p); |
b29739f9 IM |
4048 | old_prio = p->prio; |
4049 | p->prio = effective_prio(p); | |
4050 | delta = p->prio - old_prio; | |
1da177e4 | 4051 | |
dd41f596 | 4052 | if (on_rq) { |
8159f87e | 4053 | enqueue_task(rq, p, 0); |
29b4b623 | 4054 | inc_load(rq, p); |
1da177e4 | 4055 | /* |
d5f9f942 AM |
4056 | * If the task increased its priority or is running and |
4057 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4058 | */ |
d5f9f942 | 4059 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4060 | resched_task(rq->curr); |
4061 | } | |
4062 | out_unlock: | |
4063 | task_rq_unlock(rq, &flags); | |
4064 | } | |
1da177e4 LT |
4065 | EXPORT_SYMBOL(set_user_nice); |
4066 | ||
e43379f1 MM |
4067 | /* |
4068 | * can_nice - check if a task can reduce its nice value | |
4069 | * @p: task | |
4070 | * @nice: nice value | |
4071 | */ | |
36c8b586 | 4072 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4073 | { |
024f4747 MM |
4074 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4075 | int nice_rlim = 20 - nice; | |
48f24c4d | 4076 | |
e43379f1 MM |
4077 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4078 | capable(CAP_SYS_NICE)); | |
4079 | } | |
4080 | ||
1da177e4 LT |
4081 | #ifdef __ARCH_WANT_SYS_NICE |
4082 | ||
4083 | /* | |
4084 | * sys_nice - change the priority of the current process. | |
4085 | * @increment: priority increment | |
4086 | * | |
4087 | * sys_setpriority is a more generic, but much slower function that | |
4088 | * does similar things. | |
4089 | */ | |
4090 | asmlinkage long sys_nice(int increment) | |
4091 | { | |
48f24c4d | 4092 | long nice, retval; |
1da177e4 LT |
4093 | |
4094 | /* | |
4095 | * Setpriority might change our priority at the same moment. | |
4096 | * We don't have to worry. Conceptually one call occurs first | |
4097 | * and we have a single winner. | |
4098 | */ | |
e43379f1 MM |
4099 | if (increment < -40) |
4100 | increment = -40; | |
1da177e4 LT |
4101 | if (increment > 40) |
4102 | increment = 40; | |
4103 | ||
4104 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4105 | if (nice < -20) | |
4106 | nice = -20; | |
4107 | if (nice > 19) | |
4108 | nice = 19; | |
4109 | ||
e43379f1 MM |
4110 | if (increment < 0 && !can_nice(current, nice)) |
4111 | return -EPERM; | |
4112 | ||
1da177e4 LT |
4113 | retval = security_task_setnice(current, nice); |
4114 | if (retval) | |
4115 | return retval; | |
4116 | ||
4117 | set_user_nice(current, nice); | |
4118 | return 0; | |
4119 | } | |
4120 | ||
4121 | #endif | |
4122 | ||
4123 | /** | |
4124 | * task_prio - return the priority value of a given task. | |
4125 | * @p: the task in question. | |
4126 | * | |
4127 | * This is the priority value as seen by users in /proc. | |
4128 | * RT tasks are offset by -200. Normal tasks are centered | |
4129 | * around 0, value goes from -16 to +15. | |
4130 | */ | |
36c8b586 | 4131 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4132 | { |
4133 | return p->prio - MAX_RT_PRIO; | |
4134 | } | |
4135 | ||
4136 | /** | |
4137 | * task_nice - return the nice value of a given task. | |
4138 | * @p: the task in question. | |
4139 | */ | |
36c8b586 | 4140 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4141 | { |
4142 | return TASK_NICE(p); | |
4143 | } | |
1da177e4 | 4144 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4145 | |
4146 | /** | |
4147 | * idle_cpu - is a given cpu idle currently? | |
4148 | * @cpu: the processor in question. | |
4149 | */ | |
4150 | int idle_cpu(int cpu) | |
4151 | { | |
4152 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4153 | } | |
4154 | ||
1da177e4 LT |
4155 | /** |
4156 | * idle_task - return the idle task for a given cpu. | |
4157 | * @cpu: the processor in question. | |
4158 | */ | |
36c8b586 | 4159 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4160 | { |
4161 | return cpu_rq(cpu)->idle; | |
4162 | } | |
4163 | ||
4164 | /** | |
4165 | * find_process_by_pid - find a process with a matching PID value. | |
4166 | * @pid: the pid in question. | |
4167 | */ | |
a9957449 | 4168 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4169 | { |
4170 | return pid ? find_task_by_pid(pid) : current; | |
4171 | } | |
4172 | ||
4173 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4174 | static void |
4175 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4176 | { |
dd41f596 | 4177 | BUG_ON(p->se.on_rq); |
48f24c4d | 4178 | |
1da177e4 | 4179 | p->policy = policy; |
dd41f596 IM |
4180 | switch (p->policy) { |
4181 | case SCHED_NORMAL: | |
4182 | case SCHED_BATCH: | |
4183 | case SCHED_IDLE: | |
4184 | p->sched_class = &fair_sched_class; | |
4185 | break; | |
4186 | case SCHED_FIFO: | |
4187 | case SCHED_RR: | |
4188 | p->sched_class = &rt_sched_class; | |
4189 | break; | |
4190 | } | |
4191 | ||
1da177e4 | 4192 | p->rt_priority = prio; |
b29739f9 IM |
4193 | p->normal_prio = normal_prio(p); |
4194 | /* we are holding p->pi_lock already */ | |
4195 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4196 | set_load_weight(p); |
1da177e4 LT |
4197 | } |
4198 | ||
4199 | /** | |
72fd4a35 | 4200 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4201 | * @p: the task in question. |
4202 | * @policy: new policy. | |
4203 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4204 | * |
72fd4a35 | 4205 | * NOTE that the task may be already dead. |
1da177e4 | 4206 | */ |
95cdf3b7 IM |
4207 | int sched_setscheduler(struct task_struct *p, int policy, |
4208 | struct sched_param *param) | |
1da177e4 | 4209 | { |
83b699ed | 4210 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4211 | unsigned long flags; |
70b97a7f | 4212 | struct rq *rq; |
1da177e4 | 4213 | |
66e5393a SR |
4214 | /* may grab non-irq protected spin_locks */ |
4215 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4216 | recheck: |
4217 | /* double check policy once rq lock held */ | |
4218 | if (policy < 0) | |
4219 | policy = oldpolicy = p->policy; | |
4220 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4221 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4222 | policy != SCHED_IDLE) | |
b0a9499c | 4223 | return -EINVAL; |
1da177e4 LT |
4224 | /* |
4225 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4226 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4227 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4228 | */ |
4229 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4230 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4231 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4232 | return -EINVAL; |
e05606d3 | 4233 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4234 | return -EINVAL; |
4235 | ||
37e4ab3f OC |
4236 | /* |
4237 | * Allow unprivileged RT tasks to decrease priority: | |
4238 | */ | |
4239 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4240 | if (rt_policy(policy)) { |
8dc3e909 | 4241 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4242 | |
4243 | if (!lock_task_sighand(p, &flags)) | |
4244 | return -ESRCH; | |
4245 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4246 | unlock_task_sighand(p, &flags); | |
4247 | ||
4248 | /* can't set/change the rt policy */ | |
4249 | if (policy != p->policy && !rlim_rtprio) | |
4250 | return -EPERM; | |
4251 | ||
4252 | /* can't increase priority */ | |
4253 | if (param->sched_priority > p->rt_priority && | |
4254 | param->sched_priority > rlim_rtprio) | |
4255 | return -EPERM; | |
4256 | } | |
dd41f596 IM |
4257 | /* |
4258 | * Like positive nice levels, dont allow tasks to | |
4259 | * move out of SCHED_IDLE either: | |
4260 | */ | |
4261 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4262 | return -EPERM; | |
5fe1d75f | 4263 | |
37e4ab3f OC |
4264 | /* can't change other user's priorities */ |
4265 | if ((current->euid != p->euid) && | |
4266 | (current->euid != p->uid)) | |
4267 | return -EPERM; | |
4268 | } | |
1da177e4 LT |
4269 | |
4270 | retval = security_task_setscheduler(p, policy, param); | |
4271 | if (retval) | |
4272 | return retval; | |
b29739f9 IM |
4273 | /* |
4274 | * make sure no PI-waiters arrive (or leave) while we are | |
4275 | * changing the priority of the task: | |
4276 | */ | |
4277 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4278 | /* |
4279 | * To be able to change p->policy safely, the apropriate | |
4280 | * runqueue lock must be held. | |
4281 | */ | |
b29739f9 | 4282 | rq = __task_rq_lock(p); |
1da177e4 LT |
4283 | /* recheck policy now with rq lock held */ |
4284 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4285 | policy = oldpolicy = -1; | |
b29739f9 IM |
4286 | __task_rq_unlock(rq); |
4287 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4288 | goto recheck; |
4289 | } | |
2daa3577 | 4290 | update_rq_clock(rq); |
dd41f596 | 4291 | on_rq = p->se.on_rq; |
83b699ed SV |
4292 | running = task_running(rq, p); |
4293 | if (on_rq) { | |
2e1cb74a | 4294 | deactivate_task(rq, p, 0); |
83b699ed SV |
4295 | if (running) |
4296 | p->sched_class->put_prev_task(rq, p); | |
4297 | } | |
f6b53205 | 4298 | |
1da177e4 | 4299 | oldprio = p->prio; |
dd41f596 | 4300 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4301 | |
dd41f596 | 4302 | if (on_rq) { |
83b699ed SV |
4303 | if (running) |
4304 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4305 | activate_task(rq, p, 0); |
1da177e4 LT |
4306 | /* |
4307 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4308 | * our priority decreased, or if we are not currently running on |
4309 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4310 | */ |
83b699ed | 4311 | if (running) { |
d5f9f942 AM |
4312 | if (p->prio > oldprio) |
4313 | resched_task(rq->curr); | |
dd41f596 IM |
4314 | } else { |
4315 | check_preempt_curr(rq, p); | |
4316 | } | |
1da177e4 | 4317 | } |
b29739f9 IM |
4318 | __task_rq_unlock(rq); |
4319 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4320 | ||
95e02ca9 TG |
4321 | rt_mutex_adjust_pi(p); |
4322 | ||
1da177e4 LT |
4323 | return 0; |
4324 | } | |
4325 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4326 | ||
95cdf3b7 IM |
4327 | static int |
4328 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4329 | { |
1da177e4 LT |
4330 | struct sched_param lparam; |
4331 | struct task_struct *p; | |
36c8b586 | 4332 | int retval; |
1da177e4 LT |
4333 | |
4334 | if (!param || pid < 0) | |
4335 | return -EINVAL; | |
4336 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4337 | return -EFAULT; | |
5fe1d75f ON |
4338 | |
4339 | rcu_read_lock(); | |
4340 | retval = -ESRCH; | |
1da177e4 | 4341 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4342 | if (p != NULL) |
4343 | retval = sched_setscheduler(p, policy, &lparam); | |
4344 | rcu_read_unlock(); | |
36c8b586 | 4345 | |
1da177e4 LT |
4346 | return retval; |
4347 | } | |
4348 | ||
4349 | /** | |
4350 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4351 | * @pid: the pid in question. | |
4352 | * @policy: new policy. | |
4353 | * @param: structure containing the new RT priority. | |
4354 | */ | |
4355 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4356 | struct sched_param __user *param) | |
4357 | { | |
c21761f1 JB |
4358 | /* negative values for policy are not valid */ |
4359 | if (policy < 0) | |
4360 | return -EINVAL; | |
4361 | ||
1da177e4 LT |
4362 | return do_sched_setscheduler(pid, policy, param); |
4363 | } | |
4364 | ||
4365 | /** | |
4366 | * sys_sched_setparam - set/change the RT priority of a thread | |
4367 | * @pid: the pid in question. | |
4368 | * @param: structure containing the new RT priority. | |
4369 | */ | |
4370 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4371 | { | |
4372 | return do_sched_setscheduler(pid, -1, param); | |
4373 | } | |
4374 | ||
4375 | /** | |
4376 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4377 | * @pid: the pid in question. | |
4378 | */ | |
4379 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4380 | { | |
36c8b586 | 4381 | struct task_struct *p; |
3a5c359a | 4382 | int retval; |
1da177e4 LT |
4383 | |
4384 | if (pid < 0) | |
3a5c359a | 4385 | return -EINVAL; |
1da177e4 LT |
4386 | |
4387 | retval = -ESRCH; | |
4388 | read_lock(&tasklist_lock); | |
4389 | p = find_process_by_pid(pid); | |
4390 | if (p) { | |
4391 | retval = security_task_getscheduler(p); | |
4392 | if (!retval) | |
4393 | retval = p->policy; | |
4394 | } | |
4395 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4396 | return retval; |
4397 | } | |
4398 | ||
4399 | /** | |
4400 | * sys_sched_getscheduler - get the RT priority of a thread | |
4401 | * @pid: the pid in question. | |
4402 | * @param: structure containing the RT priority. | |
4403 | */ | |
4404 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4405 | { | |
4406 | struct sched_param lp; | |
36c8b586 | 4407 | struct task_struct *p; |
3a5c359a | 4408 | int retval; |
1da177e4 LT |
4409 | |
4410 | if (!param || pid < 0) | |
3a5c359a | 4411 | return -EINVAL; |
1da177e4 LT |
4412 | |
4413 | read_lock(&tasklist_lock); | |
4414 | p = find_process_by_pid(pid); | |
4415 | retval = -ESRCH; | |
4416 | if (!p) | |
4417 | goto out_unlock; | |
4418 | ||
4419 | retval = security_task_getscheduler(p); | |
4420 | if (retval) | |
4421 | goto out_unlock; | |
4422 | ||
4423 | lp.sched_priority = p->rt_priority; | |
4424 | read_unlock(&tasklist_lock); | |
4425 | ||
4426 | /* | |
4427 | * This one might sleep, we cannot do it with a spinlock held ... | |
4428 | */ | |
4429 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4430 | ||
1da177e4 LT |
4431 | return retval; |
4432 | ||
4433 | out_unlock: | |
4434 | read_unlock(&tasklist_lock); | |
4435 | return retval; | |
4436 | } | |
4437 | ||
4438 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4439 | { | |
1da177e4 | 4440 | cpumask_t cpus_allowed; |
36c8b586 IM |
4441 | struct task_struct *p; |
4442 | int retval; | |
1da177e4 | 4443 | |
5be9361c | 4444 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4445 | read_lock(&tasklist_lock); |
4446 | ||
4447 | p = find_process_by_pid(pid); | |
4448 | if (!p) { | |
4449 | read_unlock(&tasklist_lock); | |
5be9361c | 4450 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4451 | return -ESRCH; |
4452 | } | |
4453 | ||
4454 | /* | |
4455 | * It is not safe to call set_cpus_allowed with the | |
4456 | * tasklist_lock held. We will bump the task_struct's | |
4457 | * usage count and then drop tasklist_lock. | |
4458 | */ | |
4459 | get_task_struct(p); | |
4460 | read_unlock(&tasklist_lock); | |
4461 | ||
4462 | retval = -EPERM; | |
4463 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4464 | !capable(CAP_SYS_NICE)) | |
4465 | goto out_unlock; | |
4466 | ||
e7834f8f DQ |
4467 | retval = security_task_setscheduler(p, 0, NULL); |
4468 | if (retval) | |
4469 | goto out_unlock; | |
4470 | ||
1da177e4 LT |
4471 | cpus_allowed = cpuset_cpus_allowed(p); |
4472 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4473 | retval = set_cpus_allowed(p, new_mask); | |
4474 | ||
4475 | out_unlock: | |
4476 | put_task_struct(p); | |
5be9361c | 4477 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4478 | return retval; |
4479 | } | |
4480 | ||
4481 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4482 | cpumask_t *new_mask) | |
4483 | { | |
4484 | if (len < sizeof(cpumask_t)) { | |
4485 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4486 | } else if (len > sizeof(cpumask_t)) { | |
4487 | len = sizeof(cpumask_t); | |
4488 | } | |
4489 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4490 | } | |
4491 | ||
4492 | /** | |
4493 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4494 | * @pid: pid of the process | |
4495 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4496 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4497 | */ | |
4498 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4499 | unsigned long __user *user_mask_ptr) | |
4500 | { | |
4501 | cpumask_t new_mask; | |
4502 | int retval; | |
4503 | ||
4504 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4505 | if (retval) | |
4506 | return retval; | |
4507 | ||
4508 | return sched_setaffinity(pid, new_mask); | |
4509 | } | |
4510 | ||
4511 | /* | |
4512 | * Represents all cpu's present in the system | |
4513 | * In systems capable of hotplug, this map could dynamically grow | |
4514 | * as new cpu's are detected in the system via any platform specific | |
4515 | * method, such as ACPI for e.g. | |
4516 | */ | |
4517 | ||
4cef0c61 | 4518 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4519 | EXPORT_SYMBOL(cpu_present_map); |
4520 | ||
4521 | #ifndef CONFIG_SMP | |
4cef0c61 | 4522 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4523 | EXPORT_SYMBOL(cpu_online_map); |
4524 | ||
4cef0c61 | 4525 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4526 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4527 | #endif |
4528 | ||
4529 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4530 | { | |
36c8b586 | 4531 | struct task_struct *p; |
1da177e4 | 4532 | int retval; |
1da177e4 | 4533 | |
5be9361c | 4534 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4535 | read_lock(&tasklist_lock); |
4536 | ||
4537 | retval = -ESRCH; | |
4538 | p = find_process_by_pid(pid); | |
4539 | if (!p) | |
4540 | goto out_unlock; | |
4541 | ||
e7834f8f DQ |
4542 | retval = security_task_getscheduler(p); |
4543 | if (retval) | |
4544 | goto out_unlock; | |
4545 | ||
2f7016d9 | 4546 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4547 | |
4548 | out_unlock: | |
4549 | read_unlock(&tasklist_lock); | |
5be9361c | 4550 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4551 | |
9531b62f | 4552 | return retval; |
1da177e4 LT |
4553 | } |
4554 | ||
4555 | /** | |
4556 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4557 | * @pid: pid of the process | |
4558 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4559 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4560 | */ | |
4561 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4562 | unsigned long __user *user_mask_ptr) | |
4563 | { | |
4564 | int ret; | |
4565 | cpumask_t mask; | |
4566 | ||
4567 | if (len < sizeof(cpumask_t)) | |
4568 | return -EINVAL; | |
4569 | ||
4570 | ret = sched_getaffinity(pid, &mask); | |
4571 | if (ret < 0) | |
4572 | return ret; | |
4573 | ||
4574 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4575 | return -EFAULT; | |
4576 | ||
4577 | return sizeof(cpumask_t); | |
4578 | } | |
4579 | ||
4580 | /** | |
4581 | * sys_sched_yield - yield the current processor to other threads. | |
4582 | * | |
dd41f596 IM |
4583 | * This function yields the current CPU to other tasks. If there are no |
4584 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4585 | */ |
4586 | asmlinkage long sys_sched_yield(void) | |
4587 | { | |
70b97a7f | 4588 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4589 | |
2d72376b | 4590 | schedstat_inc(rq, yld_count); |
4530d7ab | 4591 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4592 | |
4593 | /* | |
4594 | * Since we are going to call schedule() anyway, there's | |
4595 | * no need to preempt or enable interrupts: | |
4596 | */ | |
4597 | __release(rq->lock); | |
8a25d5de | 4598 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4599 | _raw_spin_unlock(&rq->lock); |
4600 | preempt_enable_no_resched(); | |
4601 | ||
4602 | schedule(); | |
4603 | ||
4604 | return 0; | |
4605 | } | |
4606 | ||
e7b38404 | 4607 | static void __cond_resched(void) |
1da177e4 | 4608 | { |
8e0a43d8 IM |
4609 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4610 | __might_sleep(__FILE__, __LINE__); | |
4611 | #endif | |
5bbcfd90 IM |
4612 | /* |
4613 | * The BKS might be reacquired before we have dropped | |
4614 | * PREEMPT_ACTIVE, which could trigger a second | |
4615 | * cond_resched() call. | |
4616 | */ | |
1da177e4 LT |
4617 | do { |
4618 | add_preempt_count(PREEMPT_ACTIVE); | |
4619 | schedule(); | |
4620 | sub_preempt_count(PREEMPT_ACTIVE); | |
4621 | } while (need_resched()); | |
4622 | } | |
4623 | ||
4624 | int __sched cond_resched(void) | |
4625 | { | |
9414232f IM |
4626 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4627 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4628 | __cond_resched(); |
4629 | return 1; | |
4630 | } | |
4631 | return 0; | |
4632 | } | |
1da177e4 LT |
4633 | EXPORT_SYMBOL(cond_resched); |
4634 | ||
4635 | /* | |
4636 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4637 | * call schedule, and on return reacquire the lock. | |
4638 | * | |
4639 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4640 | * operations here to prevent schedule() from being called twice (once via | |
4641 | * spin_unlock(), once by hand). | |
4642 | */ | |
95cdf3b7 | 4643 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4644 | { |
6df3cecb JK |
4645 | int ret = 0; |
4646 | ||
1da177e4 LT |
4647 | if (need_lockbreak(lock)) { |
4648 | spin_unlock(lock); | |
4649 | cpu_relax(); | |
6df3cecb | 4650 | ret = 1; |
1da177e4 LT |
4651 | spin_lock(lock); |
4652 | } | |
9414232f | 4653 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4654 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4655 | _raw_spin_unlock(lock); |
4656 | preempt_enable_no_resched(); | |
4657 | __cond_resched(); | |
6df3cecb | 4658 | ret = 1; |
1da177e4 | 4659 | spin_lock(lock); |
1da177e4 | 4660 | } |
6df3cecb | 4661 | return ret; |
1da177e4 | 4662 | } |
1da177e4 LT |
4663 | EXPORT_SYMBOL(cond_resched_lock); |
4664 | ||
4665 | int __sched cond_resched_softirq(void) | |
4666 | { | |
4667 | BUG_ON(!in_softirq()); | |
4668 | ||
9414232f | 4669 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4670 | local_bh_enable(); |
1da177e4 LT |
4671 | __cond_resched(); |
4672 | local_bh_disable(); | |
4673 | return 1; | |
4674 | } | |
4675 | return 0; | |
4676 | } | |
1da177e4 LT |
4677 | EXPORT_SYMBOL(cond_resched_softirq); |
4678 | ||
1da177e4 LT |
4679 | /** |
4680 | * yield - yield the current processor to other threads. | |
4681 | * | |
72fd4a35 | 4682 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4683 | * thread runnable and calls sys_sched_yield(). |
4684 | */ | |
4685 | void __sched yield(void) | |
4686 | { | |
4687 | set_current_state(TASK_RUNNING); | |
4688 | sys_sched_yield(); | |
4689 | } | |
1da177e4 LT |
4690 | EXPORT_SYMBOL(yield); |
4691 | ||
4692 | /* | |
4693 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4694 | * that process accounting knows that this is a task in IO wait state. | |
4695 | * | |
4696 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4697 | * has set its backing_dev_info: the queue against which it should throttle) | |
4698 | */ | |
4699 | void __sched io_schedule(void) | |
4700 | { | |
70b97a7f | 4701 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4702 | |
0ff92245 | 4703 | delayacct_blkio_start(); |
1da177e4 LT |
4704 | atomic_inc(&rq->nr_iowait); |
4705 | schedule(); | |
4706 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4707 | delayacct_blkio_end(); |
1da177e4 | 4708 | } |
1da177e4 LT |
4709 | EXPORT_SYMBOL(io_schedule); |
4710 | ||
4711 | long __sched io_schedule_timeout(long timeout) | |
4712 | { | |
70b97a7f | 4713 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4714 | long ret; |
4715 | ||
0ff92245 | 4716 | delayacct_blkio_start(); |
1da177e4 LT |
4717 | atomic_inc(&rq->nr_iowait); |
4718 | ret = schedule_timeout(timeout); | |
4719 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4720 | delayacct_blkio_end(); |
1da177e4 LT |
4721 | return ret; |
4722 | } | |
4723 | ||
4724 | /** | |
4725 | * sys_sched_get_priority_max - return maximum RT priority. | |
4726 | * @policy: scheduling class. | |
4727 | * | |
4728 | * this syscall returns the maximum rt_priority that can be used | |
4729 | * by a given scheduling class. | |
4730 | */ | |
4731 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4732 | { | |
4733 | int ret = -EINVAL; | |
4734 | ||
4735 | switch (policy) { | |
4736 | case SCHED_FIFO: | |
4737 | case SCHED_RR: | |
4738 | ret = MAX_USER_RT_PRIO-1; | |
4739 | break; | |
4740 | case SCHED_NORMAL: | |
b0a9499c | 4741 | case SCHED_BATCH: |
dd41f596 | 4742 | case SCHED_IDLE: |
1da177e4 LT |
4743 | ret = 0; |
4744 | break; | |
4745 | } | |
4746 | return ret; | |
4747 | } | |
4748 | ||
4749 | /** | |
4750 | * sys_sched_get_priority_min - return minimum RT priority. | |
4751 | * @policy: scheduling class. | |
4752 | * | |
4753 | * this syscall returns the minimum rt_priority that can be used | |
4754 | * by a given scheduling class. | |
4755 | */ | |
4756 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4757 | { | |
4758 | int ret = -EINVAL; | |
4759 | ||
4760 | switch (policy) { | |
4761 | case SCHED_FIFO: | |
4762 | case SCHED_RR: | |
4763 | ret = 1; | |
4764 | break; | |
4765 | case SCHED_NORMAL: | |
b0a9499c | 4766 | case SCHED_BATCH: |
dd41f596 | 4767 | case SCHED_IDLE: |
1da177e4 LT |
4768 | ret = 0; |
4769 | } | |
4770 | return ret; | |
4771 | } | |
4772 | ||
4773 | /** | |
4774 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4775 | * @pid: pid of the process. | |
4776 | * @interval: userspace pointer to the timeslice value. | |
4777 | * | |
4778 | * this syscall writes the default timeslice value of a given process | |
4779 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4780 | */ | |
4781 | asmlinkage | |
4782 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4783 | { | |
36c8b586 | 4784 | struct task_struct *p; |
a4ec24b4 | 4785 | unsigned int time_slice; |
3a5c359a | 4786 | int retval; |
1da177e4 | 4787 | struct timespec t; |
1da177e4 LT |
4788 | |
4789 | if (pid < 0) | |
3a5c359a | 4790 | return -EINVAL; |
1da177e4 LT |
4791 | |
4792 | retval = -ESRCH; | |
4793 | read_lock(&tasklist_lock); | |
4794 | p = find_process_by_pid(pid); | |
4795 | if (!p) | |
4796 | goto out_unlock; | |
4797 | ||
4798 | retval = security_task_getscheduler(p); | |
4799 | if (retval) | |
4800 | goto out_unlock; | |
4801 | ||
a4ec24b4 DA |
4802 | if (p->policy == SCHED_FIFO) |
4803 | time_slice = 0; | |
4804 | else if (p->policy == SCHED_RR) | |
4805 | time_slice = DEF_TIMESLICE; | |
4806 | else { | |
4807 | struct sched_entity *se = &p->se; | |
4808 | unsigned long flags; | |
4809 | struct rq *rq; | |
4810 | ||
4811 | rq = task_rq_lock(p, &flags); | |
4812 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4813 | task_rq_unlock(rq, &flags); | |
4814 | } | |
1da177e4 | 4815 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4816 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4817 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4818 | return retval; |
3a5c359a | 4819 | |
1da177e4 LT |
4820 | out_unlock: |
4821 | read_unlock(&tasklist_lock); | |
4822 | return retval; | |
4823 | } | |
4824 | ||
2ed6e34f | 4825 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4826 | |
4827 | static void show_task(struct task_struct *p) | |
1da177e4 | 4828 | { |
1da177e4 | 4829 | unsigned long free = 0; |
36c8b586 | 4830 | unsigned state; |
1da177e4 | 4831 | |
1da177e4 | 4832 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4833 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4834 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4835 | #if BITS_PER_LONG == 32 |
1da177e4 | 4836 | if (state == TASK_RUNNING) |
cc4ea795 | 4837 | printk(KERN_CONT " running "); |
1da177e4 | 4838 | else |
cc4ea795 | 4839 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4840 | #else |
4841 | if (state == TASK_RUNNING) | |
cc4ea795 | 4842 | printk(KERN_CONT " running task "); |
1da177e4 | 4843 | else |
cc4ea795 | 4844 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4845 | #endif |
4846 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4847 | { | |
10ebffde | 4848 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4849 | while (!*n) |
4850 | n++; | |
10ebffde | 4851 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4852 | } |
4853 | #endif | |
cc4ea795 | 4854 | printk(KERN_CONT "%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4855 | |
4856 | if (state != TASK_RUNNING) | |
4857 | show_stack(p, NULL); | |
4858 | } | |
4859 | ||
e59e2ae2 | 4860 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4861 | { |
36c8b586 | 4862 | struct task_struct *g, *p; |
1da177e4 | 4863 | |
4bd77321 IM |
4864 | #if BITS_PER_LONG == 32 |
4865 | printk(KERN_INFO | |
4866 | " task PC stack pid father\n"); | |
1da177e4 | 4867 | #else |
4bd77321 IM |
4868 | printk(KERN_INFO |
4869 | " task PC stack pid father\n"); | |
1da177e4 LT |
4870 | #endif |
4871 | read_lock(&tasklist_lock); | |
4872 | do_each_thread(g, p) { | |
4873 | /* | |
4874 | * reset the NMI-timeout, listing all files on a slow | |
4875 | * console might take alot of time: | |
4876 | */ | |
4877 | touch_nmi_watchdog(); | |
39bc89fd | 4878 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4879 | show_task(p); |
1da177e4 LT |
4880 | } while_each_thread(g, p); |
4881 | ||
04c9167f JF |
4882 | touch_all_softlockup_watchdogs(); |
4883 | ||
dd41f596 IM |
4884 | #ifdef CONFIG_SCHED_DEBUG |
4885 | sysrq_sched_debug_show(); | |
4886 | #endif | |
1da177e4 | 4887 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4888 | /* |
4889 | * Only show locks if all tasks are dumped: | |
4890 | */ | |
4891 | if (state_filter == -1) | |
4892 | debug_show_all_locks(); | |
1da177e4 LT |
4893 | } |
4894 | ||
1df21055 IM |
4895 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4896 | { | |
dd41f596 | 4897 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4898 | } |
4899 | ||
f340c0d1 IM |
4900 | /** |
4901 | * init_idle - set up an idle thread for a given CPU | |
4902 | * @idle: task in question | |
4903 | * @cpu: cpu the idle task belongs to | |
4904 | * | |
4905 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4906 | * flag, to make booting more robust. | |
4907 | */ | |
5c1e1767 | 4908 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4909 | { |
70b97a7f | 4910 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4911 | unsigned long flags; |
4912 | ||
dd41f596 IM |
4913 | __sched_fork(idle); |
4914 | idle->se.exec_start = sched_clock(); | |
4915 | ||
b29739f9 | 4916 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4917 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4918 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4919 | |
4920 | spin_lock_irqsave(&rq->lock, flags); | |
4921 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4922 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4923 | idle->oncpu = 1; | |
4924 | #endif | |
1da177e4 LT |
4925 | spin_unlock_irqrestore(&rq->lock, flags); |
4926 | ||
4927 | /* Set the preempt count _outside_ the spinlocks! */ | |
4928 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4929 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4930 | #else |
a1261f54 | 4931 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4932 | #endif |
dd41f596 IM |
4933 | /* |
4934 | * The idle tasks have their own, simple scheduling class: | |
4935 | */ | |
4936 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4937 | } |
4938 | ||
4939 | /* | |
4940 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4941 | * indicates which cpus entered this state. This is used | |
4942 | * in the rcu update to wait only for active cpus. For system | |
4943 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4944 | * always be CPU_MASK_NONE. | |
4945 | */ | |
4946 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4947 | ||
4948 | #ifdef CONFIG_SMP | |
4949 | /* | |
4950 | * This is how migration works: | |
4951 | * | |
70b97a7f | 4952 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4953 | * runqueue and wake up that CPU's migration thread. |
4954 | * 2) we down() the locked semaphore => thread blocks. | |
4955 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4956 | * thread off the CPU) | |
4957 | * 4) it gets the migration request and checks whether the migrated | |
4958 | * task is still in the wrong runqueue. | |
4959 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4960 | * it and puts it into the right queue. | |
4961 | * 6) migration thread up()s the semaphore. | |
4962 | * 7) we wake up and the migration is done. | |
4963 | */ | |
4964 | ||
4965 | /* | |
4966 | * Change a given task's CPU affinity. Migrate the thread to a | |
4967 | * proper CPU and schedule it away if the CPU it's executing on | |
4968 | * is removed from the allowed bitmask. | |
4969 | * | |
4970 | * NOTE: the caller must have a valid reference to the task, the | |
4971 | * task must not exit() & deallocate itself prematurely. The | |
4972 | * call is not atomic; no spinlocks may be held. | |
4973 | */ | |
36c8b586 | 4974 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4975 | { |
70b97a7f | 4976 | struct migration_req req; |
1da177e4 | 4977 | unsigned long flags; |
70b97a7f | 4978 | struct rq *rq; |
48f24c4d | 4979 | int ret = 0; |
1da177e4 LT |
4980 | |
4981 | rq = task_rq_lock(p, &flags); | |
4982 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4983 | ret = -EINVAL; | |
4984 | goto out; | |
4985 | } | |
4986 | ||
4987 | p->cpus_allowed = new_mask; | |
4988 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4989 | if (cpu_isset(task_cpu(p), new_mask)) | |
4990 | goto out; | |
4991 | ||
4992 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4993 | /* Need help from migration thread: drop lock and wait. */ | |
4994 | task_rq_unlock(rq, &flags); | |
4995 | wake_up_process(rq->migration_thread); | |
4996 | wait_for_completion(&req.done); | |
4997 | tlb_migrate_finish(p->mm); | |
4998 | return 0; | |
4999 | } | |
5000 | out: | |
5001 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5002 | |
1da177e4 LT |
5003 | return ret; |
5004 | } | |
1da177e4 LT |
5005 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5006 | ||
5007 | /* | |
5008 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5009 | * this because either it can't run here any more (set_cpus_allowed() | |
5010 | * away from this CPU, or CPU going down), or because we're | |
5011 | * attempting to rebalance this task on exec (sched_exec). | |
5012 | * | |
5013 | * So we race with normal scheduler movements, but that's OK, as long | |
5014 | * as the task is no longer on this CPU. | |
efc30814 KK |
5015 | * |
5016 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5017 | */ |
efc30814 | 5018 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5019 | { |
70b97a7f | 5020 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5021 | int ret = 0, on_rq; |
1da177e4 LT |
5022 | |
5023 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5024 | return ret; |
1da177e4 LT |
5025 | |
5026 | rq_src = cpu_rq(src_cpu); | |
5027 | rq_dest = cpu_rq(dest_cpu); | |
5028 | ||
5029 | double_rq_lock(rq_src, rq_dest); | |
5030 | /* Already moved. */ | |
5031 | if (task_cpu(p) != src_cpu) | |
5032 | goto out; | |
5033 | /* Affinity changed (again). */ | |
5034 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5035 | goto out; | |
5036 | ||
dd41f596 | 5037 | on_rq = p->se.on_rq; |
6e82a3be | 5038 | if (on_rq) |
2e1cb74a | 5039 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5040 | |
1da177e4 | 5041 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5042 | if (on_rq) { |
5043 | activate_task(rq_dest, p, 0); | |
5044 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5045 | } |
efc30814 | 5046 | ret = 1; |
1da177e4 LT |
5047 | out: |
5048 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5049 | return ret; |
1da177e4 LT |
5050 | } |
5051 | ||
5052 | /* | |
5053 | * migration_thread - this is a highprio system thread that performs | |
5054 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5055 | * another runqueue. | |
5056 | */ | |
95cdf3b7 | 5057 | static int migration_thread(void *data) |
1da177e4 | 5058 | { |
1da177e4 | 5059 | int cpu = (long)data; |
70b97a7f | 5060 | struct rq *rq; |
1da177e4 LT |
5061 | |
5062 | rq = cpu_rq(cpu); | |
5063 | BUG_ON(rq->migration_thread != current); | |
5064 | ||
5065 | set_current_state(TASK_INTERRUPTIBLE); | |
5066 | while (!kthread_should_stop()) { | |
70b97a7f | 5067 | struct migration_req *req; |
1da177e4 | 5068 | struct list_head *head; |
1da177e4 | 5069 | |
1da177e4 LT |
5070 | spin_lock_irq(&rq->lock); |
5071 | ||
5072 | if (cpu_is_offline(cpu)) { | |
5073 | spin_unlock_irq(&rq->lock); | |
5074 | goto wait_to_die; | |
5075 | } | |
5076 | ||
5077 | if (rq->active_balance) { | |
5078 | active_load_balance(rq, cpu); | |
5079 | rq->active_balance = 0; | |
5080 | } | |
5081 | ||
5082 | head = &rq->migration_queue; | |
5083 | ||
5084 | if (list_empty(head)) { | |
5085 | spin_unlock_irq(&rq->lock); | |
5086 | schedule(); | |
5087 | set_current_state(TASK_INTERRUPTIBLE); | |
5088 | continue; | |
5089 | } | |
70b97a7f | 5090 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5091 | list_del_init(head->next); |
5092 | ||
674311d5 NP |
5093 | spin_unlock(&rq->lock); |
5094 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5095 | local_irq_enable(); | |
1da177e4 LT |
5096 | |
5097 | complete(&req->done); | |
5098 | } | |
5099 | __set_current_state(TASK_RUNNING); | |
5100 | return 0; | |
5101 | ||
5102 | wait_to_die: | |
5103 | /* Wait for kthread_stop */ | |
5104 | set_current_state(TASK_INTERRUPTIBLE); | |
5105 | while (!kthread_should_stop()) { | |
5106 | schedule(); | |
5107 | set_current_state(TASK_INTERRUPTIBLE); | |
5108 | } | |
5109 | __set_current_state(TASK_RUNNING); | |
5110 | return 0; | |
5111 | } | |
5112 | ||
5113 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5114 | |
5115 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5116 | { | |
5117 | int ret; | |
5118 | ||
5119 | local_irq_disable(); | |
5120 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5121 | local_irq_enable(); | |
5122 | return ret; | |
5123 | } | |
5124 | ||
054b9108 KK |
5125 | /* |
5126 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5127 | * NOTE: interrupts should be disabled by the caller | |
5128 | */ | |
48f24c4d | 5129 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5130 | { |
efc30814 | 5131 | unsigned long flags; |
1da177e4 | 5132 | cpumask_t mask; |
70b97a7f IM |
5133 | struct rq *rq; |
5134 | int dest_cpu; | |
1da177e4 | 5135 | |
3a5c359a AK |
5136 | do { |
5137 | /* On same node? */ | |
5138 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5139 | cpus_and(mask, mask, p->cpus_allowed); | |
5140 | dest_cpu = any_online_cpu(mask); | |
5141 | ||
5142 | /* On any allowed CPU? */ | |
5143 | if (dest_cpu == NR_CPUS) | |
5144 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5145 | ||
5146 | /* No more Mr. Nice Guy. */ | |
5147 | if (dest_cpu == NR_CPUS) { | |
5148 | rq = task_rq_lock(p, &flags); | |
5149 | cpus_setall(p->cpus_allowed); | |
5150 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5151 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5152 | |
3a5c359a AK |
5153 | /* |
5154 | * Don't tell them about moving exiting tasks or | |
5155 | * kernel threads (both mm NULL), since they never | |
5156 | * leave kernel. | |
5157 | */ | |
5158 | if (p->mm && printk_ratelimit()) | |
5159 | printk(KERN_INFO "process %d (%s) no " | |
5160 | "longer affine to cpu%d\n", | |
5161 | p->pid, p->comm, dead_cpu); | |
5162 | } | |
f7b4cddc | 5163 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5164 | } |
5165 | ||
5166 | /* | |
5167 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5168 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5169 | * for performance reasons the counter is not stricly tracking tasks to | |
5170 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5171 | * to keep the global sum constant after CPU-down: | |
5172 | */ | |
70b97a7f | 5173 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5174 | { |
70b97a7f | 5175 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5176 | unsigned long flags; |
5177 | ||
5178 | local_irq_save(flags); | |
5179 | double_rq_lock(rq_src, rq_dest); | |
5180 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5181 | rq_src->nr_uninterruptible = 0; | |
5182 | double_rq_unlock(rq_src, rq_dest); | |
5183 | local_irq_restore(flags); | |
5184 | } | |
5185 | ||
5186 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5187 | static void migrate_live_tasks(int src_cpu) | |
5188 | { | |
48f24c4d | 5189 | struct task_struct *p, *t; |
1da177e4 | 5190 | |
f7b4cddc | 5191 | read_lock(&tasklist_lock); |
1da177e4 | 5192 | |
48f24c4d IM |
5193 | do_each_thread(t, p) { |
5194 | if (p == current) | |
1da177e4 LT |
5195 | continue; |
5196 | ||
48f24c4d IM |
5197 | if (task_cpu(p) == src_cpu) |
5198 | move_task_off_dead_cpu(src_cpu, p); | |
5199 | } while_each_thread(t, p); | |
1da177e4 | 5200 | |
f7b4cddc | 5201 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5202 | } |
5203 | ||
a9957449 AD |
5204 | /* |
5205 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5206 | */ | |
5207 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5208 | { | |
5209 | update_rq_clock(rq); | |
5210 | ||
5211 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5212 | rq->nr_uninterruptible--; | |
5213 | ||
5214 | enqueue_task(rq, p, 0); | |
5215 | inc_nr_running(p, rq); | |
5216 | } | |
5217 | ||
dd41f596 IM |
5218 | /* |
5219 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5220 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5221 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5222 | */ |
5223 | void sched_idle_next(void) | |
5224 | { | |
48f24c4d | 5225 | int this_cpu = smp_processor_id(); |
70b97a7f | 5226 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5227 | struct task_struct *p = rq->idle; |
5228 | unsigned long flags; | |
5229 | ||
5230 | /* cpu has to be offline */ | |
48f24c4d | 5231 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5232 | |
48f24c4d IM |
5233 | /* |
5234 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5235 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5236 | */ |
5237 | spin_lock_irqsave(&rq->lock, flags); | |
5238 | ||
dd41f596 | 5239 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5240 | |
5241 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5242 | activate_idle_task(p, rq); |
1da177e4 LT |
5243 | |
5244 | spin_unlock_irqrestore(&rq->lock, flags); | |
5245 | } | |
5246 | ||
48f24c4d IM |
5247 | /* |
5248 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5249 | * offline. |
5250 | */ | |
5251 | void idle_task_exit(void) | |
5252 | { | |
5253 | struct mm_struct *mm = current->active_mm; | |
5254 | ||
5255 | BUG_ON(cpu_online(smp_processor_id())); | |
5256 | ||
5257 | if (mm != &init_mm) | |
5258 | switch_mm(mm, &init_mm, current); | |
5259 | mmdrop(mm); | |
5260 | } | |
5261 | ||
054b9108 | 5262 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5263 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5264 | { |
70b97a7f | 5265 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5266 | |
5267 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5268 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5269 | |
5270 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5271 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5272 | |
48f24c4d | 5273 | get_task_struct(p); |
1da177e4 LT |
5274 | |
5275 | /* | |
5276 | * Drop lock around migration; if someone else moves it, | |
5277 | * that's OK. No task can be added to this CPU, so iteration is | |
5278 | * fine. | |
5279 | */ | |
f7b4cddc | 5280 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5281 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5282 | spin_lock_irq(&rq->lock); |
1da177e4 | 5283 | |
48f24c4d | 5284 | put_task_struct(p); |
1da177e4 LT |
5285 | } |
5286 | ||
5287 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5288 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5289 | { | |
70b97a7f | 5290 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5291 | struct task_struct *next; |
48f24c4d | 5292 | |
dd41f596 IM |
5293 | for ( ; ; ) { |
5294 | if (!rq->nr_running) | |
5295 | break; | |
a8e504d2 | 5296 | update_rq_clock(rq); |
ff95f3df | 5297 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5298 | if (!next) |
5299 | break; | |
5300 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5301 | |
1da177e4 LT |
5302 | } |
5303 | } | |
5304 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5305 | ||
e692ab53 NP |
5306 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5307 | ||
5308 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5309 | { |
5310 | .procname = "sched_domain", | |
c57baf1e | 5311 | .mode = 0555, |
e0361851 | 5312 | }, |
e692ab53 NP |
5313 | {0,}, |
5314 | }; | |
5315 | ||
5316 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5317 | { |
c57baf1e | 5318 | .ctl_name = CTL_KERN, |
e0361851 | 5319 | .procname = "kernel", |
c57baf1e | 5320 | .mode = 0555, |
e0361851 AD |
5321 | .child = sd_ctl_dir, |
5322 | }, | |
e692ab53 NP |
5323 | {0,}, |
5324 | }; | |
5325 | ||
5326 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5327 | { | |
5328 | struct ctl_table *entry = | |
5cf9f062 | 5329 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5330 | |
e692ab53 NP |
5331 | return entry; |
5332 | } | |
5333 | ||
6382bc90 MM |
5334 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5335 | { | |
cd790076 | 5336 | struct ctl_table *entry; |
6382bc90 | 5337 | |
cd790076 MM |
5338 | /* |
5339 | * In the intermediate directories, both the child directory and | |
5340 | * procname are dynamically allocated and could fail but the mode | |
5341 | * will always be set. In the lowest directory the names are | |
5342 | * static strings and all have proc handlers. | |
5343 | */ | |
5344 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5345 | if (entry->child) |
5346 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5347 | if (entry->proc_handler == NULL) |
5348 | kfree(entry->procname); | |
5349 | } | |
6382bc90 MM |
5350 | |
5351 | kfree(*tablep); | |
5352 | *tablep = NULL; | |
5353 | } | |
5354 | ||
e692ab53 | 5355 | static void |
e0361851 | 5356 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5357 | const char *procname, void *data, int maxlen, |
5358 | mode_t mode, proc_handler *proc_handler) | |
5359 | { | |
e692ab53 NP |
5360 | entry->procname = procname; |
5361 | entry->data = data; | |
5362 | entry->maxlen = maxlen; | |
5363 | entry->mode = mode; | |
5364 | entry->proc_handler = proc_handler; | |
5365 | } | |
5366 | ||
5367 | static struct ctl_table * | |
5368 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5369 | { | |
ace8b3d6 | 5370 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5371 | |
ad1cdc1d MM |
5372 | if (table == NULL) |
5373 | return NULL; | |
5374 | ||
e0361851 | 5375 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5376 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5377 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5378 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5379 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5380 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5381 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5382 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5383 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5384 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5385 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5386 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5387 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5388 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5389 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5390 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5391 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5392 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5393 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5394 | &sd->cache_nice_tries, |
5395 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5396 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5397 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5398 | /* &table[11] is terminator */ |
e692ab53 NP |
5399 | |
5400 | return table; | |
5401 | } | |
5402 | ||
8401f775 | 5403 | static ctl_table * sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5404 | { |
5405 | struct ctl_table *entry, *table; | |
5406 | struct sched_domain *sd; | |
5407 | int domain_num = 0, i; | |
5408 | char buf[32]; | |
5409 | ||
5410 | for_each_domain(cpu, sd) | |
5411 | domain_num++; | |
5412 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5413 | if (table == NULL) |
5414 | return NULL; | |
e692ab53 NP |
5415 | |
5416 | i = 0; | |
5417 | for_each_domain(cpu, sd) { | |
5418 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5419 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5420 | entry->mode = 0555; |
e692ab53 NP |
5421 | entry->child = sd_alloc_ctl_domain_table(sd); |
5422 | entry++; | |
5423 | i++; | |
5424 | } | |
5425 | return table; | |
5426 | } | |
5427 | ||
5428 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5429 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5430 | { |
5431 | int i, cpu_num = num_online_cpus(); | |
5432 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5433 | char buf[32]; | |
5434 | ||
ad1cdc1d MM |
5435 | if (entry == NULL) |
5436 | return; | |
5437 | ||
e692ab53 NP |
5438 | sd_ctl_dir[0].child = entry; |
5439 | ||
97b6ea7b | 5440 | for_each_online_cpu(i) { |
e692ab53 | 5441 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5442 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5443 | entry->mode = 0555; |
e692ab53 | 5444 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5445 | entry++; |
e692ab53 NP |
5446 | } |
5447 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5448 | } | |
6382bc90 MM |
5449 | |
5450 | static void unregister_sched_domain_sysctl(void) | |
5451 | { | |
5452 | unregister_sysctl_table(sd_sysctl_header); | |
5453 | sd_sysctl_header = NULL; | |
5454 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
5455 | } | |
e692ab53 | 5456 | #else |
6382bc90 MM |
5457 | static void register_sched_domain_sysctl(void) |
5458 | { | |
5459 | } | |
5460 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5461 | { |
5462 | } | |
5463 | #endif | |
5464 | ||
1da177e4 LT |
5465 | /* |
5466 | * migration_call - callback that gets triggered when a CPU is added. | |
5467 | * Here we can start up the necessary migration thread for the new CPU. | |
5468 | */ | |
48f24c4d IM |
5469 | static int __cpuinit |
5470 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5471 | { |
1da177e4 | 5472 | struct task_struct *p; |
48f24c4d | 5473 | int cpu = (long)hcpu; |
1da177e4 | 5474 | unsigned long flags; |
70b97a7f | 5475 | struct rq *rq; |
1da177e4 LT |
5476 | |
5477 | switch (action) { | |
5be9361c GS |
5478 | case CPU_LOCK_ACQUIRE: |
5479 | mutex_lock(&sched_hotcpu_mutex); | |
5480 | break; | |
5481 | ||
1da177e4 | 5482 | case CPU_UP_PREPARE: |
8bb78442 | 5483 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5484 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5485 | if (IS_ERR(p)) |
5486 | return NOTIFY_BAD; | |
1da177e4 LT |
5487 | kthread_bind(p, cpu); |
5488 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5489 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5490 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5491 | task_rq_unlock(rq, &flags); |
5492 | cpu_rq(cpu)->migration_thread = p; | |
5493 | break; | |
48f24c4d | 5494 | |
1da177e4 | 5495 | case CPU_ONLINE: |
8bb78442 | 5496 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5497 | /* Strictly unneccessary, as first user will wake it. */ |
5498 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5499 | break; | |
48f24c4d | 5500 | |
1da177e4 LT |
5501 | #ifdef CONFIG_HOTPLUG_CPU |
5502 | case CPU_UP_CANCELED: | |
8bb78442 | 5503 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5504 | if (!cpu_rq(cpu)->migration_thread) |
5505 | break; | |
1da177e4 | 5506 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5507 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5508 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5509 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5510 | cpu_rq(cpu)->migration_thread = NULL; | |
5511 | break; | |
48f24c4d | 5512 | |
1da177e4 | 5513 | case CPU_DEAD: |
8bb78442 | 5514 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5515 | migrate_live_tasks(cpu); |
5516 | rq = cpu_rq(cpu); | |
5517 | kthread_stop(rq->migration_thread); | |
5518 | rq->migration_thread = NULL; | |
5519 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5520 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5521 | update_rq_clock(rq); |
2e1cb74a | 5522 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5523 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5524 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5525 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5526 | migrate_dead_tasks(cpu); |
d2da272a | 5527 | spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5528 | migrate_nr_uninterruptible(rq); |
5529 | BUG_ON(rq->nr_running != 0); | |
5530 | ||
5531 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5532 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5533 | * the requestors. */ |
5534 | spin_lock_irq(&rq->lock); | |
5535 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5536 | struct migration_req *req; |
5537 | ||
1da177e4 | 5538 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5539 | struct migration_req, list); |
1da177e4 LT |
5540 | list_del_init(&req->list); |
5541 | complete(&req->done); | |
5542 | } | |
5543 | spin_unlock_irq(&rq->lock); | |
5544 | break; | |
5545 | #endif | |
5be9361c GS |
5546 | case CPU_LOCK_RELEASE: |
5547 | mutex_unlock(&sched_hotcpu_mutex); | |
5548 | break; | |
1da177e4 LT |
5549 | } |
5550 | return NOTIFY_OK; | |
5551 | } | |
5552 | ||
5553 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5554 | * happens before everything else. | |
5555 | */ | |
26c2143b | 5556 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5557 | .notifier_call = migration_call, |
5558 | .priority = 10 | |
5559 | }; | |
5560 | ||
5561 | int __init migration_init(void) | |
5562 | { | |
5563 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5564 | int err; |
48f24c4d IM |
5565 | |
5566 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5567 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5568 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5569 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5570 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5571 | |
1da177e4 LT |
5572 | return 0; |
5573 | } | |
5574 | #endif | |
5575 | ||
5576 | #ifdef CONFIG_SMP | |
476f3534 CL |
5577 | |
5578 | /* Number of possible processor ids */ | |
5579 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5580 | EXPORT_SYMBOL(nr_cpu_ids); | |
5581 | ||
3e9830dc | 5582 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5583 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5584 | { | |
5585 | int level = 0; | |
5586 | ||
41c7ce9a NP |
5587 | if (!sd) { |
5588 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5589 | return; | |
5590 | } | |
5591 | ||
1da177e4 LT |
5592 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5593 | ||
5594 | do { | |
5595 | int i; | |
5596 | char str[NR_CPUS]; | |
5597 | struct sched_group *group = sd->groups; | |
5598 | cpumask_t groupmask; | |
5599 | ||
5600 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5601 | cpus_clear(groupmask); | |
5602 | ||
5603 | printk(KERN_DEBUG); | |
5604 | for (i = 0; i < level + 1; i++) | |
5605 | printk(" "); | |
5606 | printk("domain %d: ", level); | |
5607 | ||
5608 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5609 | printk("does not load-balance\n"); | |
5610 | if (sd->parent) | |
33859f7f MOS |
5611 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5612 | " has parent"); | |
1da177e4 LT |
5613 | break; |
5614 | } | |
5615 | ||
5616 | printk("span %s\n", str); | |
5617 | ||
5618 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5619 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5620 | "CPU%d\n", cpu); | |
1da177e4 | 5621 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5622 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5623 | " CPU%d\n", cpu); | |
1da177e4 LT |
5624 | |
5625 | printk(KERN_DEBUG); | |
5626 | for (i = 0; i < level + 2; i++) | |
5627 | printk(" "); | |
5628 | printk("groups:"); | |
5629 | do { | |
5630 | if (!group) { | |
5631 | printk("\n"); | |
5632 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5633 | break; | |
5634 | } | |
5635 | ||
5517d86b | 5636 | if (!group->__cpu_power) { |
cc4ea795 | 5637 | printk(KERN_CONT "\n"); |
33859f7f MOS |
5638 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5639 | "set\n"); | |
26797a34 | 5640 | break; |
1da177e4 LT |
5641 | } |
5642 | ||
5643 | if (!cpus_weight(group->cpumask)) { | |
cc4ea795 | 5644 | printk(KERN_CONT "\n"); |
1da177e4 | 5645 | printk(KERN_ERR "ERROR: empty group\n"); |
26797a34 | 5646 | break; |
1da177e4 LT |
5647 | } |
5648 | ||
5649 | if (cpus_intersects(groupmask, group->cpumask)) { | |
cc4ea795 | 5650 | printk(KERN_CONT "\n"); |
1da177e4 | 5651 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
26797a34 | 5652 | break; |
1da177e4 LT |
5653 | } |
5654 | ||
5655 | cpus_or(groupmask, groupmask, group->cpumask); | |
5656 | ||
5657 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
cc4ea795 | 5658 | printk(KERN_CONT " %s", str); |
1da177e4 LT |
5659 | |
5660 | group = group->next; | |
5661 | } while (group != sd->groups); | |
cc4ea795 | 5662 | printk(KERN_CONT "\n"); |
1da177e4 LT |
5663 | |
5664 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5665 | printk(KERN_ERR "ERROR: groups don't span " |
5666 | "domain->span\n"); | |
1da177e4 LT |
5667 | |
5668 | level++; | |
5669 | sd = sd->parent; | |
33859f7f MOS |
5670 | if (!sd) |
5671 | continue; | |
1da177e4 | 5672 | |
33859f7f MOS |
5673 | if (!cpus_subset(groupmask, sd->span)) |
5674 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5675 | "of domain->span\n"); | |
1da177e4 LT |
5676 | |
5677 | } while (sd); | |
5678 | } | |
5679 | #else | |
48f24c4d | 5680 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5681 | #endif |
5682 | ||
1a20ff27 | 5683 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5684 | { |
5685 | if (cpus_weight(sd->span) == 1) | |
5686 | return 1; | |
5687 | ||
5688 | /* Following flags need at least 2 groups */ | |
5689 | if (sd->flags & (SD_LOAD_BALANCE | | |
5690 | SD_BALANCE_NEWIDLE | | |
5691 | SD_BALANCE_FORK | | |
89c4710e SS |
5692 | SD_BALANCE_EXEC | |
5693 | SD_SHARE_CPUPOWER | | |
5694 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5695 | if (sd->groups != sd->groups->next) |
5696 | return 0; | |
5697 | } | |
5698 | ||
5699 | /* Following flags don't use groups */ | |
5700 | if (sd->flags & (SD_WAKE_IDLE | | |
5701 | SD_WAKE_AFFINE | | |
5702 | SD_WAKE_BALANCE)) | |
5703 | return 0; | |
5704 | ||
5705 | return 1; | |
5706 | } | |
5707 | ||
48f24c4d IM |
5708 | static int |
5709 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5710 | { |
5711 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5712 | ||
5713 | if (sd_degenerate(parent)) | |
5714 | return 1; | |
5715 | ||
5716 | if (!cpus_equal(sd->span, parent->span)) | |
5717 | return 0; | |
5718 | ||
5719 | /* Does parent contain flags not in child? */ | |
5720 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5721 | if (cflags & SD_WAKE_AFFINE) | |
5722 | pflags &= ~SD_WAKE_BALANCE; | |
5723 | /* Flags needing groups don't count if only 1 group in parent */ | |
5724 | if (parent->groups == parent->groups->next) { | |
5725 | pflags &= ~(SD_LOAD_BALANCE | | |
5726 | SD_BALANCE_NEWIDLE | | |
5727 | SD_BALANCE_FORK | | |
89c4710e SS |
5728 | SD_BALANCE_EXEC | |
5729 | SD_SHARE_CPUPOWER | | |
5730 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5731 | } |
5732 | if (~cflags & pflags) | |
5733 | return 0; | |
5734 | ||
5735 | return 1; | |
5736 | } | |
5737 | ||
1da177e4 LT |
5738 | /* |
5739 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5740 | * hold the hotplug lock. | |
5741 | */ | |
9c1cfda2 | 5742 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5743 | { |
70b97a7f | 5744 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5745 | struct sched_domain *tmp; |
5746 | ||
5747 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5748 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5749 | struct sched_domain *parent = tmp->parent; | |
5750 | if (!parent) | |
5751 | break; | |
1a848870 | 5752 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5753 | tmp->parent = parent->parent; |
1a848870 SS |
5754 | if (parent->parent) |
5755 | parent->parent->child = tmp; | |
5756 | } | |
245af2c7 SS |
5757 | } |
5758 | ||
1a848870 | 5759 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5760 | sd = sd->parent; |
1a848870 SS |
5761 | if (sd) |
5762 | sd->child = NULL; | |
5763 | } | |
1da177e4 LT |
5764 | |
5765 | sched_domain_debug(sd, cpu); | |
5766 | ||
674311d5 | 5767 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5768 | } |
5769 | ||
5770 | /* cpus with isolated domains */ | |
67af63a6 | 5771 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5772 | |
5773 | /* Setup the mask of cpus configured for isolated domains */ | |
5774 | static int __init isolated_cpu_setup(char *str) | |
5775 | { | |
5776 | int ints[NR_CPUS], i; | |
5777 | ||
5778 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5779 | cpus_clear(cpu_isolated_map); | |
5780 | for (i = 1; i <= ints[0]; i++) | |
5781 | if (ints[i] < NR_CPUS) | |
5782 | cpu_set(ints[i], cpu_isolated_map); | |
5783 | return 1; | |
5784 | } | |
5785 | ||
8927f494 | 5786 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5787 | |
5788 | /* | |
6711cab4 SS |
5789 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5790 | * to a function which identifies what group(along with sched group) a CPU | |
5791 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5792 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5793 | * |
5794 | * init_sched_build_groups will build a circular linked list of the groups | |
5795 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5796 | * and ->cpu_power to 0. | |
5797 | */ | |
a616058b | 5798 | static void |
6711cab4 SS |
5799 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5800 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5801 | struct sched_group **sg)) | |
1da177e4 LT |
5802 | { |
5803 | struct sched_group *first = NULL, *last = NULL; | |
5804 | cpumask_t covered = CPU_MASK_NONE; | |
5805 | int i; | |
5806 | ||
5807 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5808 | struct sched_group *sg; |
5809 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5810 | int j; |
5811 | ||
5812 | if (cpu_isset(i, covered)) | |
5813 | continue; | |
5814 | ||
5815 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5816 | sg->__cpu_power = 0; |
1da177e4 LT |
5817 | |
5818 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5819 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5820 | continue; |
5821 | ||
5822 | cpu_set(j, covered); | |
5823 | cpu_set(j, sg->cpumask); | |
5824 | } | |
5825 | if (!first) | |
5826 | first = sg; | |
5827 | if (last) | |
5828 | last->next = sg; | |
5829 | last = sg; | |
5830 | } | |
5831 | last->next = first; | |
5832 | } | |
5833 | ||
9c1cfda2 | 5834 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5835 | |
9c1cfda2 | 5836 | #ifdef CONFIG_NUMA |
198e2f18 | 5837 | |
9c1cfda2 JH |
5838 | /** |
5839 | * find_next_best_node - find the next node to include in a sched_domain | |
5840 | * @node: node whose sched_domain we're building | |
5841 | * @used_nodes: nodes already in the sched_domain | |
5842 | * | |
5843 | * Find the next node to include in a given scheduling domain. Simply | |
5844 | * finds the closest node not already in the @used_nodes map. | |
5845 | * | |
5846 | * Should use nodemask_t. | |
5847 | */ | |
5848 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5849 | { | |
5850 | int i, n, val, min_val, best_node = 0; | |
5851 | ||
5852 | min_val = INT_MAX; | |
5853 | ||
5854 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5855 | /* Start at @node */ | |
5856 | n = (node + i) % MAX_NUMNODES; | |
5857 | ||
5858 | if (!nr_cpus_node(n)) | |
5859 | continue; | |
5860 | ||
5861 | /* Skip already used nodes */ | |
5862 | if (test_bit(n, used_nodes)) | |
5863 | continue; | |
5864 | ||
5865 | /* Simple min distance search */ | |
5866 | val = node_distance(node, n); | |
5867 | ||
5868 | if (val < min_val) { | |
5869 | min_val = val; | |
5870 | best_node = n; | |
5871 | } | |
5872 | } | |
5873 | ||
5874 | set_bit(best_node, used_nodes); | |
5875 | return best_node; | |
5876 | } | |
5877 | ||
5878 | /** | |
5879 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5880 | * @node: node whose cpumask we're constructing | |
5881 | * @size: number of nodes to include in this span | |
5882 | * | |
5883 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5884 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5885 | * out optimally. | |
5886 | */ | |
5887 | static cpumask_t sched_domain_node_span(int node) | |
5888 | { | |
9c1cfda2 | 5889 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5890 | cpumask_t span, nodemask; |
5891 | int i; | |
9c1cfda2 JH |
5892 | |
5893 | cpus_clear(span); | |
5894 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5895 | ||
5896 | nodemask = node_to_cpumask(node); | |
5897 | cpus_or(span, span, nodemask); | |
5898 | set_bit(node, used_nodes); | |
5899 | ||
5900 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5901 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5902 | |
9c1cfda2 JH |
5903 | nodemask = node_to_cpumask(next_node); |
5904 | cpus_or(span, span, nodemask); | |
5905 | } | |
5906 | ||
5907 | return span; | |
5908 | } | |
5909 | #endif | |
5910 | ||
5c45bf27 | 5911 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5912 | |
9c1cfda2 | 5913 | /* |
48f24c4d | 5914 | * SMT sched-domains: |
9c1cfda2 | 5915 | */ |
1da177e4 LT |
5916 | #ifdef CONFIG_SCHED_SMT |
5917 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5918 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5919 | |
6711cab4 SS |
5920 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5921 | struct sched_group **sg) | |
1da177e4 | 5922 | { |
6711cab4 SS |
5923 | if (sg) |
5924 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5925 | return cpu; |
5926 | } | |
5927 | #endif | |
5928 | ||
48f24c4d IM |
5929 | /* |
5930 | * multi-core sched-domains: | |
5931 | */ | |
1e9f28fa SS |
5932 | #ifdef CONFIG_SCHED_MC |
5933 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5934 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5935 | #endif |
5936 | ||
5937 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5938 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5939 | struct sched_group **sg) | |
1e9f28fa | 5940 | { |
6711cab4 | 5941 | int group; |
d5a7430d | 5942 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5943 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
5944 | group = first_cpu(mask); |
5945 | if (sg) | |
5946 | *sg = &per_cpu(sched_group_core, group); | |
5947 | return group; | |
1e9f28fa SS |
5948 | } |
5949 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5950 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5951 | struct sched_group **sg) | |
1e9f28fa | 5952 | { |
6711cab4 SS |
5953 | if (sg) |
5954 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5955 | return cpu; |
5956 | } | |
5957 | #endif | |
5958 | ||
1da177e4 | 5959 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5960 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5961 | |
6711cab4 SS |
5962 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5963 | struct sched_group **sg) | |
1da177e4 | 5964 | { |
6711cab4 | 5965 | int group; |
48f24c4d | 5966 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5967 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5968 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5969 | group = first_cpu(mask); |
1e9f28fa | 5970 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 5971 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5972 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5973 | group = first_cpu(mask); |
1da177e4 | 5974 | #else |
6711cab4 | 5975 | group = cpu; |
1da177e4 | 5976 | #endif |
6711cab4 SS |
5977 | if (sg) |
5978 | *sg = &per_cpu(sched_group_phys, group); | |
5979 | return group; | |
1da177e4 LT |
5980 | } |
5981 | ||
5982 | #ifdef CONFIG_NUMA | |
1da177e4 | 5983 | /* |
9c1cfda2 JH |
5984 | * The init_sched_build_groups can't handle what we want to do with node |
5985 | * groups, so roll our own. Now each node has its own list of groups which | |
5986 | * gets dynamically allocated. | |
1da177e4 | 5987 | */ |
9c1cfda2 | 5988 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5989 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5990 | |
9c1cfda2 | 5991 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5992 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5993 | |
6711cab4 SS |
5994 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5995 | struct sched_group **sg) | |
9c1cfda2 | 5996 | { |
6711cab4 SS |
5997 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5998 | int group; | |
5999 | ||
6000 | cpus_and(nodemask, nodemask, *cpu_map); | |
6001 | group = first_cpu(nodemask); | |
6002 | ||
6003 | if (sg) | |
6004 | *sg = &per_cpu(sched_group_allnodes, group); | |
6005 | return group; | |
1da177e4 | 6006 | } |
6711cab4 | 6007 | |
08069033 SS |
6008 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6009 | { | |
6010 | struct sched_group *sg = group_head; | |
6011 | int j; | |
6012 | ||
6013 | if (!sg) | |
6014 | return; | |
3a5c359a AK |
6015 | do { |
6016 | for_each_cpu_mask(j, sg->cpumask) { | |
6017 | struct sched_domain *sd; | |
08069033 | 6018 | |
3a5c359a AK |
6019 | sd = &per_cpu(phys_domains, j); |
6020 | if (j != first_cpu(sd->groups->cpumask)) { | |
6021 | /* | |
6022 | * Only add "power" once for each | |
6023 | * physical package. | |
6024 | */ | |
6025 | continue; | |
6026 | } | |
08069033 | 6027 | |
3a5c359a AK |
6028 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6029 | } | |
6030 | sg = sg->next; | |
6031 | } while (sg != group_head); | |
08069033 | 6032 | } |
1da177e4 LT |
6033 | #endif |
6034 | ||
a616058b | 6035 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6036 | /* Free memory allocated for various sched_group structures */ |
6037 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6038 | { | |
a616058b | 6039 | int cpu, i; |
51888ca2 SV |
6040 | |
6041 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6042 | struct sched_group **sched_group_nodes |
6043 | = sched_group_nodes_bycpu[cpu]; | |
6044 | ||
51888ca2 SV |
6045 | if (!sched_group_nodes) |
6046 | continue; | |
6047 | ||
6048 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6049 | cpumask_t nodemask = node_to_cpumask(i); | |
6050 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6051 | ||
6052 | cpus_and(nodemask, nodemask, *cpu_map); | |
6053 | if (cpus_empty(nodemask)) | |
6054 | continue; | |
6055 | ||
6056 | if (sg == NULL) | |
6057 | continue; | |
6058 | sg = sg->next; | |
6059 | next_sg: | |
6060 | oldsg = sg; | |
6061 | sg = sg->next; | |
6062 | kfree(oldsg); | |
6063 | if (oldsg != sched_group_nodes[i]) | |
6064 | goto next_sg; | |
6065 | } | |
6066 | kfree(sched_group_nodes); | |
6067 | sched_group_nodes_bycpu[cpu] = NULL; | |
6068 | } | |
51888ca2 | 6069 | } |
a616058b SS |
6070 | #else |
6071 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6072 | { | |
6073 | } | |
6074 | #endif | |
51888ca2 | 6075 | |
89c4710e SS |
6076 | /* |
6077 | * Initialize sched groups cpu_power. | |
6078 | * | |
6079 | * cpu_power indicates the capacity of sched group, which is used while | |
6080 | * distributing the load between different sched groups in a sched domain. | |
6081 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6082 | * there are asymmetries in the topology. If there are asymmetries, group | |
6083 | * having more cpu_power will pickup more load compared to the group having | |
6084 | * less cpu_power. | |
6085 | * | |
6086 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6087 | * the maximum number of tasks a group can handle in the presence of other idle | |
6088 | * or lightly loaded groups in the same sched domain. | |
6089 | */ | |
6090 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6091 | { | |
6092 | struct sched_domain *child; | |
6093 | struct sched_group *group; | |
6094 | ||
6095 | WARN_ON(!sd || !sd->groups); | |
6096 | ||
6097 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6098 | return; | |
6099 | ||
6100 | child = sd->child; | |
6101 | ||
5517d86b ED |
6102 | sd->groups->__cpu_power = 0; |
6103 | ||
89c4710e SS |
6104 | /* |
6105 | * For perf policy, if the groups in child domain share resources | |
6106 | * (for example cores sharing some portions of the cache hierarchy | |
6107 | * or SMT), then set this domain groups cpu_power such that each group | |
6108 | * can handle only one task, when there are other idle groups in the | |
6109 | * same sched domain. | |
6110 | */ | |
6111 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6112 | (child->flags & | |
6113 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6114 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6115 | return; |
6116 | } | |
6117 | ||
89c4710e SS |
6118 | /* |
6119 | * add cpu_power of each child group to this groups cpu_power | |
6120 | */ | |
6121 | group = child->groups; | |
6122 | do { | |
5517d86b | 6123 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6124 | group = group->next; |
6125 | } while (group != child->groups); | |
6126 | } | |
6127 | ||
1da177e4 | 6128 | /* |
1a20ff27 DG |
6129 | * Build sched domains for a given set of cpus and attach the sched domains |
6130 | * to the individual cpus | |
1da177e4 | 6131 | */ |
51888ca2 | 6132 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6133 | { |
6134 | int i; | |
d1b55138 JH |
6135 | #ifdef CONFIG_NUMA |
6136 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6137 | int sd_allnodes = 0; |
d1b55138 JH |
6138 | |
6139 | /* | |
6140 | * Allocate the per-node list of sched groups | |
6141 | */ | |
5cf9f062 | 6142 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
d3a5aa98 | 6143 | GFP_KERNEL); |
d1b55138 JH |
6144 | if (!sched_group_nodes) { |
6145 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6146 | return -ENOMEM; |
d1b55138 JH |
6147 | } |
6148 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6149 | #endif | |
1da177e4 LT |
6150 | |
6151 | /* | |
1a20ff27 | 6152 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6153 | */ |
1a20ff27 | 6154 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6155 | struct sched_domain *sd = NULL, *p; |
6156 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6157 | ||
1a20ff27 | 6158 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6159 | |
6160 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6161 | if (cpus_weight(*cpu_map) > |
6162 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6163 | sd = &per_cpu(allnodes_domains, i); |
6164 | *sd = SD_ALLNODES_INIT; | |
6165 | sd->span = *cpu_map; | |
6711cab4 | 6166 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6167 | p = sd; |
6711cab4 | 6168 | sd_allnodes = 1; |
9c1cfda2 JH |
6169 | } else |
6170 | p = NULL; | |
6171 | ||
1da177e4 | 6172 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6173 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6174 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6175 | sd->parent = p; | |
1a848870 SS |
6176 | if (p) |
6177 | p->child = sd; | |
9c1cfda2 | 6178 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6179 | #endif |
6180 | ||
6181 | p = sd; | |
6182 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6183 | *sd = SD_CPU_INIT; |
6184 | sd->span = nodemask; | |
6185 | sd->parent = p; | |
1a848870 SS |
6186 | if (p) |
6187 | p->child = sd; | |
6711cab4 | 6188 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6189 | |
1e9f28fa SS |
6190 | #ifdef CONFIG_SCHED_MC |
6191 | p = sd; | |
6192 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6193 | *sd = SD_MC_INIT; |
6194 | sd->span = cpu_coregroup_map(i); | |
6195 | cpus_and(sd->span, sd->span, *cpu_map); | |
6196 | sd->parent = p; | |
1a848870 | 6197 | p->child = sd; |
6711cab4 | 6198 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6199 | #endif |
6200 | ||
1da177e4 LT |
6201 | #ifdef CONFIG_SCHED_SMT |
6202 | p = sd; | |
6203 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6204 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6205 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6206 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6207 | sd->parent = p; |
1a848870 | 6208 | p->child = sd; |
6711cab4 | 6209 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6210 | #endif |
6211 | } | |
6212 | ||
6213 | #ifdef CONFIG_SCHED_SMT | |
6214 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6215 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6216 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6217 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6218 | if (i != first_cpu(this_sibling_map)) |
6219 | continue; | |
6220 | ||
dd41f596 IM |
6221 | init_sched_build_groups(this_sibling_map, cpu_map, |
6222 | &cpu_to_cpu_group); | |
1da177e4 LT |
6223 | } |
6224 | #endif | |
6225 | ||
1e9f28fa SS |
6226 | #ifdef CONFIG_SCHED_MC |
6227 | /* Set up multi-core groups */ | |
6228 | for_each_cpu_mask(i, *cpu_map) { | |
6229 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6230 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6231 | if (i != first_cpu(this_core_map)) | |
6232 | continue; | |
dd41f596 IM |
6233 | init_sched_build_groups(this_core_map, cpu_map, |
6234 | &cpu_to_core_group); | |
1e9f28fa SS |
6235 | } |
6236 | #endif | |
6237 | ||
1da177e4 LT |
6238 | /* Set up physical groups */ |
6239 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6240 | cpumask_t nodemask = node_to_cpumask(i); | |
6241 | ||
1a20ff27 | 6242 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6243 | if (cpus_empty(nodemask)) |
6244 | continue; | |
6245 | ||
6711cab4 | 6246 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6247 | } |
6248 | ||
6249 | #ifdef CONFIG_NUMA | |
6250 | /* Set up node groups */ | |
6711cab4 | 6251 | if (sd_allnodes) |
dd41f596 IM |
6252 | init_sched_build_groups(*cpu_map, cpu_map, |
6253 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6254 | |
6255 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6256 | /* Set up node groups */ | |
6257 | struct sched_group *sg, *prev; | |
6258 | cpumask_t nodemask = node_to_cpumask(i); | |
6259 | cpumask_t domainspan; | |
6260 | cpumask_t covered = CPU_MASK_NONE; | |
6261 | int j; | |
6262 | ||
6263 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6264 | if (cpus_empty(nodemask)) { |
6265 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6266 | continue; |
d1b55138 | 6267 | } |
9c1cfda2 JH |
6268 | |
6269 | domainspan = sched_domain_node_span(i); | |
6270 | cpus_and(domainspan, domainspan, *cpu_map); | |
6271 | ||
15f0b676 | 6272 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6273 | if (!sg) { |
6274 | printk(KERN_WARNING "Can not alloc domain group for " | |
6275 | "node %d\n", i); | |
6276 | goto error; | |
6277 | } | |
9c1cfda2 JH |
6278 | sched_group_nodes[i] = sg; |
6279 | for_each_cpu_mask(j, nodemask) { | |
6280 | struct sched_domain *sd; | |
9761eea8 | 6281 | |
9c1cfda2 JH |
6282 | sd = &per_cpu(node_domains, j); |
6283 | sd->groups = sg; | |
9c1cfda2 | 6284 | } |
5517d86b | 6285 | sg->__cpu_power = 0; |
9c1cfda2 | 6286 | sg->cpumask = nodemask; |
51888ca2 | 6287 | sg->next = sg; |
9c1cfda2 JH |
6288 | cpus_or(covered, covered, nodemask); |
6289 | prev = sg; | |
6290 | ||
6291 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6292 | cpumask_t tmp, notcovered; | |
6293 | int n = (i + j) % MAX_NUMNODES; | |
6294 | ||
6295 | cpus_complement(notcovered, covered); | |
6296 | cpus_and(tmp, notcovered, *cpu_map); | |
6297 | cpus_and(tmp, tmp, domainspan); | |
6298 | if (cpus_empty(tmp)) | |
6299 | break; | |
6300 | ||
6301 | nodemask = node_to_cpumask(n); | |
6302 | cpus_and(tmp, tmp, nodemask); | |
6303 | if (cpus_empty(tmp)) | |
6304 | continue; | |
6305 | ||
15f0b676 SV |
6306 | sg = kmalloc_node(sizeof(struct sched_group), |
6307 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6308 | if (!sg) { |
6309 | printk(KERN_WARNING | |
6310 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6311 | goto error; |
9c1cfda2 | 6312 | } |
5517d86b | 6313 | sg->__cpu_power = 0; |
9c1cfda2 | 6314 | sg->cpumask = tmp; |
51888ca2 | 6315 | sg->next = prev->next; |
9c1cfda2 JH |
6316 | cpus_or(covered, covered, tmp); |
6317 | prev->next = sg; | |
6318 | prev = sg; | |
6319 | } | |
9c1cfda2 | 6320 | } |
1da177e4 LT |
6321 | #endif |
6322 | ||
6323 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6324 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6325 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6326 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6327 | ||
89c4710e | 6328 | init_sched_groups_power(i, sd); |
5c45bf27 | 6329 | } |
1da177e4 | 6330 | #endif |
1e9f28fa | 6331 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6332 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6333 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6334 | ||
89c4710e | 6335 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6336 | } |
6337 | #endif | |
1e9f28fa | 6338 | |
5c45bf27 | 6339 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6340 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6341 | ||
89c4710e | 6342 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6343 | } |
6344 | ||
9c1cfda2 | 6345 | #ifdef CONFIG_NUMA |
08069033 SS |
6346 | for (i = 0; i < MAX_NUMNODES; i++) |
6347 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6348 | |
6711cab4 SS |
6349 | if (sd_allnodes) { |
6350 | struct sched_group *sg; | |
f712c0c7 | 6351 | |
6711cab4 | 6352 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6353 | init_numa_sched_groups_power(sg); |
6354 | } | |
9c1cfda2 JH |
6355 | #endif |
6356 | ||
1da177e4 | 6357 | /* Attach the domains */ |
1a20ff27 | 6358 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6359 | struct sched_domain *sd; |
6360 | #ifdef CONFIG_SCHED_SMT | |
6361 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6362 | #elif defined(CONFIG_SCHED_MC) |
6363 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6364 | #else |
6365 | sd = &per_cpu(phys_domains, i); | |
6366 | #endif | |
6367 | cpu_attach_domain(sd, i); | |
6368 | } | |
51888ca2 SV |
6369 | |
6370 | return 0; | |
6371 | ||
a616058b | 6372 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6373 | error: |
6374 | free_sched_groups(cpu_map); | |
6375 | return -ENOMEM; | |
a616058b | 6376 | #endif |
1da177e4 | 6377 | } |
1a20ff27 DG |
6378 | /* |
6379 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6380 | */ | |
51888ca2 | 6381 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6382 | { |
6383 | cpumask_t cpu_default_map; | |
51888ca2 | 6384 | int err; |
1da177e4 | 6385 | |
1a20ff27 DG |
6386 | /* |
6387 | * Setup mask for cpus without special case scheduling requirements. | |
6388 | * For now this just excludes isolated cpus, but could be used to | |
6389 | * exclude other special cases in the future. | |
6390 | */ | |
6391 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6392 | ||
51888ca2 SV |
6393 | err = build_sched_domains(&cpu_default_map); |
6394 | ||
6382bc90 MM |
6395 | register_sched_domain_sysctl(); |
6396 | ||
51888ca2 | 6397 | return err; |
1a20ff27 DG |
6398 | } |
6399 | ||
6400 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6401 | { |
51888ca2 | 6402 | free_sched_groups(cpu_map); |
9c1cfda2 | 6403 | } |
1da177e4 | 6404 | |
1a20ff27 DG |
6405 | /* |
6406 | * Detach sched domains from a group of cpus specified in cpu_map | |
6407 | * These cpus will now be attached to the NULL domain | |
6408 | */ | |
858119e1 | 6409 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6410 | { |
6411 | int i; | |
6412 | ||
6382bc90 MM |
6413 | unregister_sched_domain_sysctl(); |
6414 | ||
1a20ff27 DG |
6415 | for_each_cpu_mask(i, *cpu_map) |
6416 | cpu_attach_domain(NULL, i); | |
6417 | synchronize_sched(); | |
6418 | arch_destroy_sched_domains(cpu_map); | |
6419 | } | |
6420 | ||
5c45bf27 | 6421 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6422 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6423 | { |
6424 | int err; | |
6425 | ||
5be9361c | 6426 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6427 | detach_destroy_domains(&cpu_online_map); |
6428 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6429 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6430 | |
6431 | return err; | |
6432 | } | |
6433 | ||
6434 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6435 | { | |
6436 | int ret; | |
6437 | ||
6438 | if (buf[0] != '0' && buf[0] != '1') | |
6439 | return -EINVAL; | |
6440 | ||
6441 | if (smt) | |
6442 | sched_smt_power_savings = (buf[0] == '1'); | |
6443 | else | |
6444 | sched_mc_power_savings = (buf[0] == '1'); | |
6445 | ||
6446 | ret = arch_reinit_sched_domains(); | |
6447 | ||
6448 | return ret ? ret : count; | |
6449 | } | |
6450 | ||
5c45bf27 SS |
6451 | #ifdef CONFIG_SCHED_MC |
6452 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6453 | { | |
6454 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6455 | } | |
48f24c4d IM |
6456 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6457 | const char *buf, size_t count) | |
5c45bf27 SS |
6458 | { |
6459 | return sched_power_savings_store(buf, count, 0); | |
6460 | } | |
6707de00 AB |
6461 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6462 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6463 | #endif |
6464 | ||
6465 | #ifdef CONFIG_SCHED_SMT | |
6466 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6467 | { | |
6468 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6469 | } | |
48f24c4d IM |
6470 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6471 | const char *buf, size_t count) | |
5c45bf27 SS |
6472 | { |
6473 | return sched_power_savings_store(buf, count, 1); | |
6474 | } | |
6707de00 AB |
6475 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6476 | sched_smt_power_savings_store); | |
6477 | #endif | |
6478 | ||
6479 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6480 | { | |
6481 | int err = 0; | |
6482 | ||
6483 | #ifdef CONFIG_SCHED_SMT | |
6484 | if (smt_capable()) | |
6485 | err = sysfs_create_file(&cls->kset.kobj, | |
6486 | &attr_sched_smt_power_savings.attr); | |
6487 | #endif | |
6488 | #ifdef CONFIG_SCHED_MC | |
6489 | if (!err && mc_capable()) | |
6490 | err = sysfs_create_file(&cls->kset.kobj, | |
6491 | &attr_sched_mc_power_savings.attr); | |
6492 | #endif | |
6493 | return err; | |
6494 | } | |
5c45bf27 SS |
6495 | #endif |
6496 | ||
1da177e4 LT |
6497 | /* |
6498 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6499 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6500 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6501 | * which will prevent rebalancing while the sched domains are recalculated. |
6502 | */ | |
6503 | static int update_sched_domains(struct notifier_block *nfb, | |
6504 | unsigned long action, void *hcpu) | |
6505 | { | |
1da177e4 LT |
6506 | switch (action) { |
6507 | case CPU_UP_PREPARE: | |
8bb78442 | 6508 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6509 | case CPU_DOWN_PREPARE: |
8bb78442 | 6510 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6511 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6512 | return NOTIFY_OK; |
6513 | ||
6514 | case CPU_UP_CANCELED: | |
8bb78442 | 6515 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6516 | case CPU_DOWN_FAILED: |
8bb78442 | 6517 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6518 | case CPU_ONLINE: |
8bb78442 | 6519 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6520 | case CPU_DEAD: |
8bb78442 | 6521 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6522 | /* |
6523 | * Fall through and re-initialise the domains. | |
6524 | */ | |
6525 | break; | |
6526 | default: | |
6527 | return NOTIFY_DONE; | |
6528 | } | |
6529 | ||
6530 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6531 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6532 | |
6533 | return NOTIFY_OK; | |
6534 | } | |
1da177e4 LT |
6535 | |
6536 | void __init sched_init_smp(void) | |
6537 | { | |
5c1e1767 NP |
6538 | cpumask_t non_isolated_cpus; |
6539 | ||
5be9361c | 6540 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6541 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6542 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6543 | if (cpus_empty(non_isolated_cpus)) |
6544 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6545 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6546 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6547 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6548 | |
6549 | /* Move init over to a non-isolated CPU */ | |
6550 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6551 | BUG(); | |
1da177e4 LT |
6552 | } |
6553 | #else | |
6554 | void __init sched_init_smp(void) | |
6555 | { | |
6556 | } | |
6557 | #endif /* CONFIG_SMP */ | |
6558 | ||
6559 | int in_sched_functions(unsigned long addr) | |
6560 | { | |
6561 | /* Linker adds these: start and end of __sched functions */ | |
6562 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6563 | |
1da177e4 LT |
6564 | return in_lock_functions(addr) || |
6565 | (addr >= (unsigned long)__sched_text_start | |
6566 | && addr < (unsigned long)__sched_text_end); | |
6567 | } | |
6568 | ||
a9957449 | 6569 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6570 | { |
6571 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6572 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6573 | cfs_rq->rq = rq; | |
6574 | #endif | |
67e9fb2a | 6575 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6576 | } |
6577 | ||
1da177e4 LT |
6578 | void __init sched_init(void) |
6579 | { | |
476f3534 | 6580 | int highest_cpu = 0; |
dd41f596 IM |
6581 | int i, j; |
6582 | ||
0a945022 | 6583 | for_each_possible_cpu(i) { |
dd41f596 | 6584 | struct rt_prio_array *array; |
70b97a7f | 6585 | struct rq *rq; |
1da177e4 LT |
6586 | |
6587 | rq = cpu_rq(i); | |
6588 | spin_lock_init(&rq->lock); | |
fcb99371 | 6589 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6590 | rq->nr_running = 0; |
dd41f596 IM |
6591 | rq->clock = 1; |
6592 | init_cfs_rq(&rq->cfs, rq); | |
6593 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6594 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6595 | { |
6596 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6597 | struct sched_entity *se = | |
6598 | &per_cpu(init_sched_entity, i); | |
6599 | ||
6600 | init_cfs_rq_p[i] = cfs_rq; | |
6601 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6602 | cfs_rq->tg = &init_task_group; |
3a252015 | 6603 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6604 | &rq->leaf_cfs_rq_list); |
6605 | ||
3a252015 IM |
6606 | init_sched_entity_p[i] = se; |
6607 | se->cfs_rq = &rq->cfs; | |
6608 | se->my_q = cfs_rq; | |
4cf86d77 | 6609 | se->load.weight = init_task_group_load; |
9b5b7751 | 6610 | se->load.inv_weight = |
4cf86d77 | 6611 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6612 | se->parent = NULL; |
6613 | } | |
4cf86d77 | 6614 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6615 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6616 | #endif |
1da177e4 | 6617 | |
dd41f596 IM |
6618 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6619 | rq->cpu_load[j] = 0; | |
1da177e4 | 6620 | #ifdef CONFIG_SMP |
41c7ce9a | 6621 | rq->sd = NULL; |
1da177e4 | 6622 | rq->active_balance = 0; |
dd41f596 | 6623 | rq->next_balance = jiffies; |
1da177e4 | 6624 | rq->push_cpu = 0; |
0a2966b4 | 6625 | rq->cpu = i; |
1da177e4 LT |
6626 | rq->migration_thread = NULL; |
6627 | INIT_LIST_HEAD(&rq->migration_queue); | |
6628 | #endif | |
6629 | atomic_set(&rq->nr_iowait, 0); | |
6630 | ||
dd41f596 IM |
6631 | array = &rq->rt.active; |
6632 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6633 | INIT_LIST_HEAD(array->queue + j); | |
6634 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6635 | } |
476f3534 | 6636 | highest_cpu = i; |
dd41f596 IM |
6637 | /* delimiter for bitsearch: */ |
6638 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6639 | } |
6640 | ||
2dd73a4f | 6641 | set_load_weight(&init_task); |
b50f60ce | 6642 | |
e107be36 AK |
6643 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6644 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6645 | #endif | |
6646 | ||
c9819f45 | 6647 | #ifdef CONFIG_SMP |
476f3534 | 6648 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6649 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6650 | #endif | |
6651 | ||
b50f60ce HC |
6652 | #ifdef CONFIG_RT_MUTEXES |
6653 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6654 | #endif | |
6655 | ||
1da177e4 LT |
6656 | /* |
6657 | * The boot idle thread does lazy MMU switching as well: | |
6658 | */ | |
6659 | atomic_inc(&init_mm.mm_count); | |
6660 | enter_lazy_tlb(&init_mm, current); | |
6661 | ||
6662 | /* | |
6663 | * Make us the idle thread. Technically, schedule() should not be | |
6664 | * called from this thread, however somewhere below it might be, | |
6665 | * but because we are the idle thread, we just pick up running again | |
6666 | * when this runqueue becomes "idle". | |
6667 | */ | |
6668 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6669 | /* |
6670 | * During early bootup we pretend to be a normal task: | |
6671 | */ | |
6672 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6673 | } |
6674 | ||
6675 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6676 | void __might_sleep(char *file, int line) | |
6677 | { | |
48f24c4d | 6678 | #ifdef in_atomic |
1da177e4 LT |
6679 | static unsigned long prev_jiffy; /* ratelimiting */ |
6680 | ||
6681 | if ((in_atomic() || irqs_disabled()) && | |
6682 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6683 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6684 | return; | |
6685 | prev_jiffy = jiffies; | |
91368d73 | 6686 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6687 | " context at %s:%d\n", file, line); |
6688 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6689 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6690 | debug_show_held_locks(current); |
3117df04 IM |
6691 | if (irqs_disabled()) |
6692 | print_irqtrace_events(current); | |
1da177e4 LT |
6693 | dump_stack(); |
6694 | } | |
6695 | #endif | |
6696 | } | |
6697 | EXPORT_SYMBOL(__might_sleep); | |
6698 | #endif | |
6699 | ||
6700 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6701 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6702 | { | |
6703 | int on_rq; | |
6704 | update_rq_clock(rq); | |
6705 | on_rq = p->se.on_rq; | |
6706 | if (on_rq) | |
6707 | deactivate_task(rq, p, 0); | |
6708 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6709 | if (on_rq) { | |
6710 | activate_task(rq, p, 0); | |
6711 | resched_task(rq->curr); | |
6712 | } | |
6713 | } | |
6714 | ||
1da177e4 LT |
6715 | void normalize_rt_tasks(void) |
6716 | { | |
a0f98a1c | 6717 | struct task_struct *g, *p; |
1da177e4 | 6718 | unsigned long flags; |
70b97a7f | 6719 | struct rq *rq; |
1da177e4 LT |
6720 | |
6721 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6722 | do_each_thread(g, p) { |
178be793 IM |
6723 | /* |
6724 | * Only normalize user tasks: | |
6725 | */ | |
6726 | if (!p->mm) | |
6727 | continue; | |
6728 | ||
6cfb0d5d | 6729 | p->se.exec_start = 0; |
6cfb0d5d | 6730 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6731 | p->se.wait_start = 0; |
dd41f596 | 6732 | p->se.sleep_start = 0; |
dd41f596 | 6733 | p->se.block_start = 0; |
6cfb0d5d | 6734 | #endif |
dd41f596 IM |
6735 | task_rq(p)->clock = 0; |
6736 | ||
6737 | if (!rt_task(p)) { | |
6738 | /* | |
6739 | * Renice negative nice level userspace | |
6740 | * tasks back to 0: | |
6741 | */ | |
6742 | if (TASK_NICE(p) < 0 && p->mm) | |
6743 | set_user_nice(p, 0); | |
1da177e4 | 6744 | continue; |
dd41f596 | 6745 | } |
1da177e4 | 6746 | |
b29739f9 IM |
6747 | spin_lock_irqsave(&p->pi_lock, flags); |
6748 | rq = __task_rq_lock(p); | |
1da177e4 | 6749 | |
178be793 | 6750 | normalize_task(rq, p); |
3a5e4dc1 | 6751 | |
b29739f9 IM |
6752 | __task_rq_unlock(rq); |
6753 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6754 | } while_each_thread(g, p); |
6755 | ||
1da177e4 LT |
6756 | read_unlock_irq(&tasklist_lock); |
6757 | } | |
6758 | ||
6759 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6760 | |
6761 | #ifdef CONFIG_IA64 | |
6762 | /* | |
6763 | * These functions are only useful for the IA64 MCA handling. | |
6764 | * | |
6765 | * They can only be called when the whole system has been | |
6766 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6767 | * activity can take place. Using them for anything else would | |
6768 | * be a serious bug, and as a result, they aren't even visible | |
6769 | * under any other configuration. | |
6770 | */ | |
6771 | ||
6772 | /** | |
6773 | * curr_task - return the current task for a given cpu. | |
6774 | * @cpu: the processor in question. | |
6775 | * | |
6776 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6777 | */ | |
36c8b586 | 6778 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6779 | { |
6780 | return cpu_curr(cpu); | |
6781 | } | |
6782 | ||
6783 | /** | |
6784 | * set_curr_task - set the current task for a given cpu. | |
6785 | * @cpu: the processor in question. | |
6786 | * @p: the task pointer to set. | |
6787 | * | |
6788 | * Description: This function must only be used when non-maskable interrupts | |
6789 | * are serviced on a separate stack. It allows the architecture to switch the | |
6790 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6791 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6792 | * and caller must save the original value of the current task (see | |
6793 | * curr_task() above) and restore that value before reenabling interrupts and | |
6794 | * re-starting the system. | |
6795 | * | |
6796 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6797 | */ | |
36c8b586 | 6798 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6799 | { |
6800 | cpu_curr(cpu) = p; | |
6801 | } | |
6802 | ||
6803 | #endif | |
29f59db3 SV |
6804 | |
6805 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6806 | ||
29f59db3 | 6807 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6808 | struct task_group *sched_create_group(void) |
29f59db3 | 6809 | { |
4cf86d77 | 6810 | struct task_group *tg; |
29f59db3 SV |
6811 | struct cfs_rq *cfs_rq; |
6812 | struct sched_entity *se; | |
9b5b7751 | 6813 | struct rq *rq; |
29f59db3 SV |
6814 | int i; |
6815 | ||
29f59db3 SV |
6816 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6817 | if (!tg) | |
6818 | return ERR_PTR(-ENOMEM); | |
6819 | ||
9b5b7751 | 6820 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6821 | if (!tg->cfs_rq) |
6822 | goto err; | |
9b5b7751 | 6823 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6824 | if (!tg->se) |
6825 | goto err; | |
6826 | ||
6827 | for_each_possible_cpu(i) { | |
9b5b7751 | 6828 | rq = cpu_rq(i); |
29f59db3 SV |
6829 | |
6830 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6831 | cpu_to_node(i)); | |
6832 | if (!cfs_rq) | |
6833 | goto err; | |
6834 | ||
6835 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6836 | cpu_to_node(i)); | |
6837 | if (!se) | |
6838 | goto err; | |
6839 | ||
6840 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6841 | memset(se, 0, sizeof(struct sched_entity)); | |
6842 | ||
6843 | tg->cfs_rq[i] = cfs_rq; | |
6844 | init_cfs_rq(cfs_rq, rq); | |
6845 | cfs_rq->tg = tg; | |
29f59db3 SV |
6846 | |
6847 | tg->se[i] = se; | |
6848 | se->cfs_rq = &rq->cfs; | |
6849 | se->my_q = cfs_rq; | |
6850 | se->load.weight = NICE_0_LOAD; | |
6851 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6852 | se->parent = NULL; | |
6853 | } | |
6854 | ||
9b5b7751 SV |
6855 | for_each_possible_cpu(i) { |
6856 | rq = cpu_rq(i); | |
6857 | cfs_rq = tg->cfs_rq[i]; | |
6858 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6859 | } | |
29f59db3 | 6860 | |
9b5b7751 | 6861 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6862 | spin_lock_init(&tg->lock); |
29f59db3 | 6863 | |
9b5b7751 | 6864 | return tg; |
29f59db3 SV |
6865 | |
6866 | err: | |
6867 | for_each_possible_cpu(i) { | |
a65914b3 | 6868 | if (tg->cfs_rq) |
29f59db3 | 6869 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6870 | if (tg->se) |
29f59db3 SV |
6871 | kfree(tg->se[i]); |
6872 | } | |
a65914b3 IM |
6873 | kfree(tg->cfs_rq); |
6874 | kfree(tg->se); | |
6875 | kfree(tg); | |
29f59db3 SV |
6876 | |
6877 | return ERR_PTR(-ENOMEM); | |
6878 | } | |
6879 | ||
9b5b7751 SV |
6880 | /* rcu callback to free various structures associated with a task group */ |
6881 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6882 | { |
9b5b7751 | 6883 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6884 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6885 | struct sched_entity *se; |
6886 | int i; | |
6887 | ||
29f59db3 SV |
6888 | /* now it should be safe to free those cfs_rqs */ |
6889 | for_each_possible_cpu(i) { | |
6890 | cfs_rq = tg->cfs_rq[i]; | |
6891 | kfree(cfs_rq); | |
6892 | ||
6893 | se = tg->se[i]; | |
6894 | kfree(se); | |
6895 | } | |
6896 | ||
6897 | kfree(tg->cfs_rq); | |
6898 | kfree(tg->se); | |
6899 | kfree(tg); | |
6900 | } | |
6901 | ||
9b5b7751 | 6902 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6903 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 6904 | { |
9b5b7751 SV |
6905 | struct cfs_rq *cfs_rq; |
6906 | int i; | |
29f59db3 | 6907 | |
9b5b7751 SV |
6908 | for_each_possible_cpu(i) { |
6909 | cfs_rq = tg->cfs_rq[i]; | |
6910 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
6911 | } | |
6912 | ||
6913 | cfs_rq = tg->cfs_rq[0]; | |
6914 | ||
6915 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6916 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
6917 | } |
6918 | ||
9b5b7751 | 6919 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6920 | * The caller of this function should have put the task in its new group |
6921 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6922 | * reflect its new group. | |
9b5b7751 SV |
6923 | */ |
6924 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
6925 | { |
6926 | int on_rq, running; | |
6927 | unsigned long flags; | |
6928 | struct rq *rq; | |
6929 | ||
6930 | rq = task_rq_lock(tsk, &flags); | |
6931 | ||
6932 | if (tsk->sched_class != &fair_sched_class) | |
6933 | goto done; | |
6934 | ||
6935 | update_rq_clock(rq); | |
6936 | ||
6937 | running = task_running(rq, tsk); | |
6938 | on_rq = tsk->se.on_rq; | |
6939 | ||
83b699ed | 6940 | if (on_rq) { |
29f59db3 | 6941 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
6942 | if (unlikely(running)) |
6943 | tsk->sched_class->put_prev_task(rq, tsk); | |
6944 | } | |
29f59db3 SV |
6945 | |
6946 | set_task_cfs_rq(tsk); | |
6947 | ||
83b699ed SV |
6948 | if (on_rq) { |
6949 | if (unlikely(running)) | |
6950 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 6951 | enqueue_task(rq, tsk, 0); |
83b699ed | 6952 | } |
29f59db3 SV |
6953 | |
6954 | done: | |
6955 | task_rq_unlock(rq, &flags); | |
6956 | } | |
6957 | ||
6958 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
6959 | { | |
6960 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
6961 | struct rq *rq = cfs_rq->rq; | |
6962 | int on_rq; | |
6963 | ||
6964 | spin_lock_irq(&rq->lock); | |
6965 | ||
6966 | on_rq = se->on_rq; | |
6967 | if (on_rq) | |
6968 | dequeue_entity(cfs_rq, se, 0); | |
6969 | ||
6970 | se->load.weight = shares; | |
6971 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
6972 | ||
6973 | if (on_rq) | |
6974 | enqueue_entity(cfs_rq, se, 0); | |
6975 | ||
6976 | spin_unlock_irq(&rq->lock); | |
6977 | } | |
6978 | ||
4cf86d77 | 6979 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
6980 | { |
6981 | int i; | |
29f59db3 | 6982 | |
5cb350ba | 6983 | spin_lock(&tg->lock); |
9b5b7751 | 6984 | if (tg->shares == shares) |
5cb350ba | 6985 | goto done; |
29f59db3 | 6986 | |
9b5b7751 | 6987 | tg->shares = shares; |
29f59db3 | 6988 | for_each_possible_cpu(i) |
9b5b7751 | 6989 | set_se_shares(tg->se[i], shares); |
29f59db3 | 6990 | |
5cb350ba DG |
6991 | done: |
6992 | spin_unlock(&tg->lock); | |
9b5b7751 | 6993 | return 0; |
29f59db3 SV |
6994 | } |
6995 | ||
5cb350ba DG |
6996 | unsigned long sched_group_shares(struct task_group *tg) |
6997 | { | |
6998 | return tg->shares; | |
6999 | } | |
7000 | ||
3a252015 | 7001 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |