]>
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> | |
b488893a | 47 | #include <linux/pid_namespace.h> |
1da177e4 LT |
48 | #include <linux/smp.h> |
49 | #include <linux/threads.h> | |
50 | #include <linux/timer.h> | |
51 | #include <linux/rcupdate.h> | |
52 | #include <linux/cpu.h> | |
53 | #include <linux/cpuset.h> | |
54 | #include <linux/percpu.h> | |
62d0df64 | 55 | #include <linux/cpu_acct.h> |
1da177e4 LT |
56 | #include <linux/kthread.h> |
57 | #include <linux/seq_file.h> | |
e692ab53 | 58 | #include <linux/sysctl.h> |
1da177e4 LT |
59 | #include <linux/syscalls.h> |
60 | #include <linux/times.h> | |
8f0ab514 | 61 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 62 | #include <linux/kprobes.h> |
0ff92245 | 63 | #include <linux/delayacct.h> |
5517d86b | 64 | #include <linux/reciprocal_div.h> |
dff06c15 | 65 | #include <linux/unistd.h> |
f5ff8422 | 66 | #include <linux/pagemap.h> |
1da177e4 | 67 | |
5517d86b | 68 | #include <asm/tlb.h> |
1da177e4 | 69 | |
b035b6de AD |
70 | /* |
71 | * Scheduler clock - returns current time in nanosec units. | |
72 | * This is default implementation. | |
73 | * Architectures and sub-architectures can override this. | |
74 | */ | |
75 | unsigned long long __attribute__((weak)) sched_clock(void) | |
76 | { | |
77 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
78 | } | |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
99 | * Some helpers for converting nanosecond timing to jiffy resolution | |
100 | */ | |
a4ec24b4 | 101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
102 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
103 | ||
6aa645ea IM |
104 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
105 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
106 | ||
1da177e4 LT |
107 | /* |
108 | * These are the 'tuning knobs' of the scheduler: | |
109 | * | |
a4ec24b4 | 110 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
111 | * Timeslices get refilled after they expire. |
112 | */ | |
1da177e4 | 113 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 114 | |
5517d86b ED |
115 | #ifdef CONFIG_SMP |
116 | /* | |
117 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
118 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
119 | */ | |
120 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
121 | { | |
122 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Each time a sched group cpu_power is changed, | |
127 | * we must compute its reciprocal value | |
128 | */ | |
129 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
130 | { | |
131 | sg->__cpu_power += val; | |
132 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
133 | } | |
134 | #endif | |
135 | ||
e05606d3 IM |
136 | static inline int rt_policy(int policy) |
137 | { | |
138 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
139 | return 1; | |
140 | return 0; | |
141 | } | |
142 | ||
143 | static inline int task_has_rt_policy(struct task_struct *p) | |
144 | { | |
145 | return rt_policy(p->policy); | |
146 | } | |
147 | ||
1da177e4 | 148 | /* |
6aa645ea | 149 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 150 | */ |
6aa645ea IM |
151 | struct rt_prio_array { |
152 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
153 | struct list_head queue[MAX_RT_PRIO]; | |
154 | }; | |
155 | ||
29f59db3 SV |
156 | #ifdef CONFIG_FAIR_GROUP_SCHED |
157 | ||
29f59db3 SV |
158 | struct cfs_rq; |
159 | ||
160 | /* task group related information */ | |
4cf86d77 | 161 | struct task_group { |
29f59db3 SV |
162 | /* schedulable entities of this group on each cpu */ |
163 | struct sched_entity **se; | |
164 | /* runqueue "owned" by this group on each cpu */ | |
165 | struct cfs_rq **cfs_rq; | |
166 | unsigned long shares; | |
5cb350ba DG |
167 | /* spinlock to serialize modification to shares */ |
168 | spinlock_t lock; | |
29f59db3 SV |
169 | }; |
170 | ||
171 | /* Default task group's sched entity on each cpu */ | |
172 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
173 | /* Default task group's cfs_rq on each cpu */ | |
174 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
175 | ||
9b5b7751 SV |
176 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
177 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
178 | |
179 | /* Default task group. | |
3a252015 | 180 | * Every task in system belong to this group at bootup. |
29f59db3 | 181 | */ |
4cf86d77 | 182 | struct task_group init_task_group = { |
3a252015 IM |
183 | .se = init_sched_entity_p, |
184 | .cfs_rq = init_cfs_rq_p, | |
185 | }; | |
9b5b7751 | 186 | |
24e377a8 | 187 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 188 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 189 | #else |
3a252015 | 190 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
191 | #endif |
192 | ||
4cf86d77 | 193 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
194 | |
195 | /* return group to which a task belongs */ | |
4cf86d77 | 196 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 197 | { |
4cf86d77 | 198 | struct task_group *tg; |
9b5b7751 | 199 | |
24e377a8 SV |
200 | #ifdef CONFIG_FAIR_USER_SCHED |
201 | tg = p->user->tg; | |
202 | #else | |
4cf86d77 | 203 | tg = &init_task_group; |
24e377a8 | 204 | #endif |
9b5b7751 SV |
205 | |
206 | return tg; | |
29f59db3 SV |
207 | } |
208 | ||
209 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
210 | static inline void set_task_cfs_rq(struct task_struct *p) | |
211 | { | |
4cf86d77 IM |
212 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
213 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
214 | } |
215 | ||
216 | #else | |
217 | ||
218 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
219 | ||
220 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
221 | ||
6aa645ea IM |
222 | /* CFS-related fields in a runqueue */ |
223 | struct cfs_rq { | |
224 | struct load_weight load; | |
225 | unsigned long nr_running; | |
226 | ||
6aa645ea | 227 | u64 exec_clock; |
e9acbff6 | 228 | u64 min_vruntime; |
6aa645ea IM |
229 | |
230 | struct rb_root tasks_timeline; | |
231 | struct rb_node *rb_leftmost; | |
232 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
233 | /* 'curr' points to currently running entity on this cfs_rq. |
234 | * It is set to NULL otherwise (i.e when none are currently running). | |
235 | */ | |
236 | struct sched_entity *curr; | |
ddc97297 PZ |
237 | |
238 | unsigned long nr_spread_over; | |
239 | ||
62160e3f | 240 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
241 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
242 | ||
243 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
244 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
245 | * (like users, containers etc.) | |
246 | * | |
247 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
248 | * list is used during load balance. | |
249 | */ | |
250 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 251 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 252 | struct rcu_head rcu; |
6aa645ea IM |
253 | #endif |
254 | }; | |
1da177e4 | 255 | |
6aa645ea IM |
256 | /* Real-Time classes' related field in a runqueue: */ |
257 | struct rt_rq { | |
258 | struct rt_prio_array active; | |
259 | int rt_load_balance_idx; | |
260 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
261 | }; | |
262 | ||
1da177e4 LT |
263 | /* |
264 | * This is the main, per-CPU runqueue data structure. | |
265 | * | |
266 | * Locking rule: those places that want to lock multiple runqueues | |
267 | * (such as the load balancing or the thread migration code), lock | |
268 | * acquire operations must be ordered by ascending &runqueue. | |
269 | */ | |
70b97a7f | 270 | struct rq { |
d8016491 IM |
271 | /* runqueue lock: */ |
272 | spinlock_t lock; | |
1da177e4 LT |
273 | |
274 | /* | |
275 | * nr_running and cpu_load should be in the same cacheline because | |
276 | * remote CPUs use both these fields when doing load calculation. | |
277 | */ | |
278 | unsigned long nr_running; | |
6aa645ea IM |
279 | #define CPU_LOAD_IDX_MAX 5 |
280 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 281 | unsigned char idle_at_tick; |
46cb4b7c SS |
282 | #ifdef CONFIG_NO_HZ |
283 | unsigned char in_nohz_recently; | |
284 | #endif | |
d8016491 IM |
285 | /* capture load from *all* tasks on this cpu: */ |
286 | struct load_weight load; | |
6aa645ea IM |
287 | unsigned long nr_load_updates; |
288 | u64 nr_switches; | |
289 | ||
290 | struct cfs_rq cfs; | |
291 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
292 | /* list of leaf cfs_rq on this cpu: */ |
293 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 294 | #endif |
6aa645ea | 295 | struct rt_rq rt; |
1da177e4 LT |
296 | |
297 | /* | |
298 | * This is part of a global counter where only the total sum | |
299 | * over all CPUs matters. A task can increase this counter on | |
300 | * one CPU and if it got migrated afterwards it may decrease | |
301 | * it on another CPU. Always updated under the runqueue lock: | |
302 | */ | |
303 | unsigned long nr_uninterruptible; | |
304 | ||
36c8b586 | 305 | struct task_struct *curr, *idle; |
c9819f45 | 306 | unsigned long next_balance; |
1da177e4 | 307 | struct mm_struct *prev_mm; |
6aa645ea | 308 | |
6aa645ea IM |
309 | u64 clock, prev_clock_raw; |
310 | s64 clock_max_delta; | |
311 | ||
312 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
313 | u64 idle_clock; |
314 | unsigned int clock_deep_idle_events; | |
529c7726 | 315 | u64 tick_timestamp; |
6aa645ea | 316 | |
1da177e4 LT |
317 | atomic_t nr_iowait; |
318 | ||
319 | #ifdef CONFIG_SMP | |
320 | struct sched_domain *sd; | |
321 | ||
322 | /* For active balancing */ | |
323 | int active_balance; | |
324 | int push_cpu; | |
d8016491 IM |
325 | /* cpu of this runqueue: */ |
326 | int cpu; | |
1da177e4 | 327 | |
36c8b586 | 328 | struct task_struct *migration_thread; |
1da177e4 LT |
329 | struct list_head migration_queue; |
330 | #endif | |
331 | ||
332 | #ifdef CONFIG_SCHEDSTATS | |
333 | /* latency stats */ | |
334 | struct sched_info rq_sched_info; | |
335 | ||
336 | /* sys_sched_yield() stats */ | |
480b9434 KC |
337 | unsigned int yld_exp_empty; |
338 | unsigned int yld_act_empty; | |
339 | unsigned int yld_both_empty; | |
340 | unsigned int yld_count; | |
1da177e4 LT |
341 | |
342 | /* schedule() stats */ | |
480b9434 KC |
343 | unsigned int sched_switch; |
344 | unsigned int sched_count; | |
345 | unsigned int sched_goidle; | |
1da177e4 LT |
346 | |
347 | /* try_to_wake_up() stats */ | |
480b9434 KC |
348 | unsigned int ttwu_count; |
349 | unsigned int ttwu_local; | |
b8efb561 IM |
350 | |
351 | /* BKL stats */ | |
480b9434 | 352 | unsigned int bkl_count; |
1da177e4 | 353 | #endif |
fcb99371 | 354 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
355 | }; |
356 | ||
f34e3b61 | 357 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 358 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 359 | |
dd41f596 IM |
360 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
361 | { | |
362 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
363 | } | |
364 | ||
0a2966b4 CL |
365 | static inline int cpu_of(struct rq *rq) |
366 | { | |
367 | #ifdef CONFIG_SMP | |
368 | return rq->cpu; | |
369 | #else | |
370 | return 0; | |
371 | #endif | |
372 | } | |
373 | ||
20d315d4 | 374 | /* |
b04a0f4c IM |
375 | * Update the per-runqueue clock, as finegrained as the platform can give |
376 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 377 | */ |
b04a0f4c | 378 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
379 | { |
380 | u64 prev_raw = rq->prev_clock_raw; | |
381 | u64 now = sched_clock(); | |
382 | s64 delta = now - prev_raw; | |
383 | u64 clock = rq->clock; | |
384 | ||
b04a0f4c IM |
385 | #ifdef CONFIG_SCHED_DEBUG |
386 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
387 | #endif | |
20d315d4 IM |
388 | /* |
389 | * Protect against sched_clock() occasionally going backwards: | |
390 | */ | |
391 | if (unlikely(delta < 0)) { | |
392 | clock++; | |
393 | rq->clock_warps++; | |
394 | } else { | |
395 | /* | |
396 | * Catch too large forward jumps too: | |
397 | */ | |
529c7726 IM |
398 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
399 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
400 | clock = rq->tick_timestamp + TICK_NSEC; | |
401 | else | |
402 | clock++; | |
20d315d4 IM |
403 | rq->clock_overflows++; |
404 | } else { | |
405 | if (unlikely(delta > rq->clock_max_delta)) | |
406 | rq->clock_max_delta = delta; | |
407 | clock += delta; | |
408 | } | |
409 | } | |
410 | ||
411 | rq->prev_clock_raw = now; | |
412 | rq->clock = clock; | |
b04a0f4c | 413 | } |
20d315d4 | 414 | |
b04a0f4c IM |
415 | static void update_rq_clock(struct rq *rq) |
416 | { | |
417 | if (likely(smp_processor_id() == cpu_of(rq))) | |
418 | __update_rq_clock(rq); | |
20d315d4 IM |
419 | } |
420 | ||
674311d5 NP |
421 | /* |
422 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 423 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
424 | * |
425 | * The domain tree of any CPU may only be accessed from within | |
426 | * preempt-disabled sections. | |
427 | */ | |
48f24c4d IM |
428 | #define for_each_domain(cpu, __sd) \ |
429 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
430 | |
431 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
432 | #define this_rq() (&__get_cpu_var(runqueues)) | |
433 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
434 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
435 | ||
bf5c91ba IM |
436 | /* |
437 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
438 | */ | |
439 | #ifdef CONFIG_SCHED_DEBUG | |
440 | # define const_debug __read_mostly | |
441 | #else | |
442 | # define const_debug static const | |
443 | #endif | |
444 | ||
445 | /* | |
446 | * Debugging: various feature bits | |
447 | */ | |
448 | enum { | |
bbdba7c0 IM |
449 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
450 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 451 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 452 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 453 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 454 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
455 | }; |
456 | ||
457 | const_debug unsigned int sysctl_sched_features = | |
8401f775 IM |
458 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
459 | SCHED_FEAT_START_DEBIT * 1 | | |
460 | SCHED_FEAT_TREE_AVG * 0 | | |
461 | SCHED_FEAT_APPROX_AVG * 0 | | |
462 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | | |
463 | SCHED_FEAT_PREEMPT_RESTRICT * 1; | |
bf5c91ba IM |
464 | |
465 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
466 | ||
e436d800 IM |
467 | /* |
468 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
469 | * clock constructed from sched_clock(): | |
470 | */ | |
471 | unsigned long long cpu_clock(int cpu) | |
472 | { | |
e436d800 IM |
473 | unsigned long long now; |
474 | unsigned long flags; | |
b04a0f4c | 475 | struct rq *rq; |
e436d800 | 476 | |
2cd4d0ea | 477 | local_irq_save(flags); |
b04a0f4c IM |
478 | rq = cpu_rq(cpu); |
479 | update_rq_clock(rq); | |
480 | now = rq->clock; | |
2cd4d0ea | 481 | local_irq_restore(flags); |
e436d800 IM |
482 | |
483 | return now; | |
484 | } | |
a58f6f25 | 485 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 486 | |
1da177e4 | 487 | #ifndef prepare_arch_switch |
4866cde0 NP |
488 | # define prepare_arch_switch(next) do { } while (0) |
489 | #endif | |
490 | #ifndef finish_arch_switch | |
491 | # define finish_arch_switch(prev) do { } while (0) | |
492 | #endif | |
493 | ||
494 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 495 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
496 | { |
497 | return rq->curr == p; | |
498 | } | |
499 | ||
70b97a7f | 500 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
501 | { |
502 | } | |
503 | ||
70b97a7f | 504 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 505 | { |
da04c035 IM |
506 | #ifdef CONFIG_DEBUG_SPINLOCK |
507 | /* this is a valid case when another task releases the spinlock */ | |
508 | rq->lock.owner = current; | |
509 | #endif | |
8a25d5de IM |
510 | /* |
511 | * If we are tracking spinlock dependencies then we have to | |
512 | * fix up the runqueue lock - which gets 'carried over' from | |
513 | * prev into current: | |
514 | */ | |
515 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
516 | ||
4866cde0 NP |
517 | spin_unlock_irq(&rq->lock); |
518 | } | |
519 | ||
520 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 521 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
522 | { |
523 | #ifdef CONFIG_SMP | |
524 | return p->oncpu; | |
525 | #else | |
526 | return rq->curr == p; | |
527 | #endif | |
528 | } | |
529 | ||
70b97a7f | 530 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
531 | { |
532 | #ifdef CONFIG_SMP | |
533 | /* | |
534 | * We can optimise this out completely for !SMP, because the | |
535 | * SMP rebalancing from interrupt is the only thing that cares | |
536 | * here. | |
537 | */ | |
538 | next->oncpu = 1; | |
539 | #endif | |
540 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
541 | spin_unlock_irq(&rq->lock); | |
542 | #else | |
543 | spin_unlock(&rq->lock); | |
544 | #endif | |
545 | } | |
546 | ||
70b97a7f | 547 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
548 | { |
549 | #ifdef CONFIG_SMP | |
550 | /* | |
551 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
552 | * We must ensure this doesn't happen until the switch is completely | |
553 | * finished. | |
554 | */ | |
555 | smp_wmb(); | |
556 | prev->oncpu = 0; | |
557 | #endif | |
558 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
559 | local_irq_enable(); | |
1da177e4 | 560 | #endif |
4866cde0 NP |
561 | } |
562 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 563 | |
b29739f9 IM |
564 | /* |
565 | * __task_rq_lock - lock the runqueue a given task resides on. | |
566 | * Must be called interrupts disabled. | |
567 | */ | |
70b97a7f | 568 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
569 | __acquires(rq->lock) |
570 | { | |
3a5c359a AK |
571 | for (;;) { |
572 | struct rq *rq = task_rq(p); | |
573 | spin_lock(&rq->lock); | |
574 | if (likely(rq == task_rq(p))) | |
575 | return rq; | |
b29739f9 | 576 | spin_unlock(&rq->lock); |
b29739f9 | 577 | } |
b29739f9 IM |
578 | } |
579 | ||
1da177e4 LT |
580 | /* |
581 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
582 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
583 | * explicitly disabling preemption. | |
584 | */ | |
70b97a7f | 585 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
586 | __acquires(rq->lock) |
587 | { | |
70b97a7f | 588 | struct rq *rq; |
1da177e4 | 589 | |
3a5c359a AK |
590 | for (;;) { |
591 | local_irq_save(*flags); | |
592 | rq = task_rq(p); | |
593 | spin_lock(&rq->lock); | |
594 | if (likely(rq == task_rq(p))) | |
595 | return rq; | |
1da177e4 | 596 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 597 | } |
1da177e4 LT |
598 | } |
599 | ||
a9957449 | 600 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
601 | __releases(rq->lock) |
602 | { | |
603 | spin_unlock(&rq->lock); | |
604 | } | |
605 | ||
70b97a7f | 606 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
607 | __releases(rq->lock) |
608 | { | |
609 | spin_unlock_irqrestore(&rq->lock, *flags); | |
610 | } | |
611 | ||
1da177e4 | 612 | /* |
cc2a73b5 | 613 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 614 | */ |
a9957449 | 615 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
616 | __acquires(rq->lock) |
617 | { | |
70b97a7f | 618 | struct rq *rq; |
1da177e4 LT |
619 | |
620 | local_irq_disable(); | |
621 | rq = this_rq(); | |
622 | spin_lock(&rq->lock); | |
623 | ||
624 | return rq; | |
625 | } | |
626 | ||
1b9f19c2 | 627 | /* |
2aa44d05 | 628 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 629 | */ |
2aa44d05 | 630 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 631 | { |
2aa44d05 IM |
632 | struct rq *rq = cpu_rq(smp_processor_id()); |
633 | ||
634 | spin_lock(&rq->lock); | |
635 | __update_rq_clock(rq); | |
636 | spin_unlock(&rq->lock); | |
637 | rq->clock_deep_idle_events++; | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
640 | ||
641 | /* | |
642 | * We just idled delta nanoseconds (called with irqs disabled): | |
643 | */ | |
644 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
645 | { | |
646 | struct rq *rq = cpu_rq(smp_processor_id()); | |
647 | u64 now = sched_clock(); | |
1b9f19c2 | 648 | |
2aa44d05 IM |
649 | rq->idle_clock += delta_ns; |
650 | /* | |
651 | * Override the previous timestamp and ignore all | |
652 | * sched_clock() deltas that occured while we idled, | |
653 | * and use the PM-provided delta_ns to advance the | |
654 | * rq clock: | |
655 | */ | |
656 | spin_lock(&rq->lock); | |
657 | rq->prev_clock_raw = now; | |
658 | rq->clock += delta_ns; | |
659 | spin_unlock(&rq->lock); | |
1b9f19c2 | 660 | } |
2aa44d05 | 661 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 662 | |
c24d20db IM |
663 | /* |
664 | * resched_task - mark a task 'to be rescheduled now'. | |
665 | * | |
666 | * On UP this means the setting of the need_resched flag, on SMP it | |
667 | * might also involve a cross-CPU call to trigger the scheduler on | |
668 | * the target CPU. | |
669 | */ | |
670 | #ifdef CONFIG_SMP | |
671 | ||
672 | #ifndef tsk_is_polling | |
673 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
674 | #endif | |
675 | ||
676 | static void resched_task(struct task_struct *p) | |
677 | { | |
678 | int cpu; | |
679 | ||
680 | assert_spin_locked(&task_rq(p)->lock); | |
681 | ||
682 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
683 | return; | |
684 | ||
685 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
686 | ||
687 | cpu = task_cpu(p); | |
688 | if (cpu == smp_processor_id()) | |
689 | return; | |
690 | ||
691 | /* NEED_RESCHED must be visible before we test polling */ | |
692 | smp_mb(); | |
693 | if (!tsk_is_polling(p)) | |
694 | smp_send_reschedule(cpu); | |
695 | } | |
696 | ||
697 | static void resched_cpu(int cpu) | |
698 | { | |
699 | struct rq *rq = cpu_rq(cpu); | |
700 | unsigned long flags; | |
701 | ||
702 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
703 | return; | |
704 | resched_task(cpu_curr(cpu)); | |
705 | spin_unlock_irqrestore(&rq->lock, flags); | |
706 | } | |
707 | #else | |
708 | static inline void resched_task(struct task_struct *p) | |
709 | { | |
710 | assert_spin_locked(&task_rq(p)->lock); | |
711 | set_tsk_need_resched(p); | |
712 | } | |
713 | #endif | |
714 | ||
45bf76df IM |
715 | #if BITS_PER_LONG == 32 |
716 | # define WMULT_CONST (~0UL) | |
717 | #else | |
718 | # define WMULT_CONST (1UL << 32) | |
719 | #endif | |
720 | ||
721 | #define WMULT_SHIFT 32 | |
722 | ||
194081eb IM |
723 | /* |
724 | * Shift right and round: | |
725 | */ | |
cf2ab469 | 726 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 727 | |
cb1c4fc9 | 728 | static unsigned long |
45bf76df IM |
729 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
730 | struct load_weight *lw) | |
731 | { | |
732 | u64 tmp; | |
733 | ||
734 | if (unlikely(!lw->inv_weight)) | |
194081eb | 735 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
736 | |
737 | tmp = (u64)delta_exec * weight; | |
738 | /* | |
739 | * Check whether we'd overflow the 64-bit multiplication: | |
740 | */ | |
194081eb | 741 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 742 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
743 | WMULT_SHIFT/2); |
744 | else | |
cf2ab469 | 745 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 746 | |
ecf691da | 747 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
748 | } |
749 | ||
750 | static inline unsigned long | |
751 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
752 | { | |
753 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
754 | } | |
755 | ||
1091985b | 756 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
757 | { |
758 | lw->weight += inc; | |
45bf76df IM |
759 | } |
760 | ||
1091985b | 761 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
762 | { |
763 | lw->weight -= dec; | |
45bf76df IM |
764 | } |
765 | ||
2dd73a4f PW |
766 | /* |
767 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
768 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
769 | * each task makes to its run queue's load is weighted according to its | |
770 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
771 | * scaled version of the new time slice allocation that they receive on time | |
772 | * slice expiry etc. | |
773 | */ | |
774 | ||
dd41f596 IM |
775 | #define WEIGHT_IDLEPRIO 2 |
776 | #define WMULT_IDLEPRIO (1 << 31) | |
777 | ||
778 | /* | |
779 | * Nice levels are multiplicative, with a gentle 10% change for every | |
780 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
781 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
782 | * that remained on nice 0. | |
783 | * | |
784 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
785 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
786 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
787 | * If a task goes up by ~10% and another task goes down by ~10% then | |
788 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
789 | */ |
790 | static const int prio_to_weight[40] = { | |
254753dc IM |
791 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
792 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
793 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
794 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
795 | /* 0 */ 1024, 820, 655, 526, 423, | |
796 | /* 5 */ 335, 272, 215, 172, 137, | |
797 | /* 10 */ 110, 87, 70, 56, 45, | |
798 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
799 | }; |
800 | ||
5714d2de IM |
801 | /* |
802 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
803 | * | |
804 | * In cases where the weight does not change often, we can use the | |
805 | * precalculated inverse to speed up arithmetics by turning divisions | |
806 | * into multiplications: | |
807 | */ | |
dd41f596 | 808 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
809 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
810 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
811 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
812 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
813 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
814 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
815 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
816 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 817 | }; |
2dd73a4f | 818 | |
dd41f596 IM |
819 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
820 | ||
821 | /* | |
822 | * runqueue iterator, to support SMP load-balancing between different | |
823 | * scheduling classes, without having to expose their internal data | |
824 | * structures to the load-balancing proper: | |
825 | */ | |
826 | struct rq_iterator { | |
827 | void *arg; | |
828 | struct task_struct *(*start)(void *); | |
829 | struct task_struct *(*next)(void *); | |
830 | }; | |
831 | ||
832 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
833 | unsigned long max_nr_move, unsigned long max_load_move, | |
834 | struct sched_domain *sd, enum cpu_idle_type idle, | |
835 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 836 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
837 | |
838 | #include "sched_stats.h" | |
dd41f596 | 839 | #include "sched_idletask.c" |
5522d5d5 IM |
840 | #include "sched_fair.c" |
841 | #include "sched_rt.c" | |
dd41f596 IM |
842 | #ifdef CONFIG_SCHED_DEBUG |
843 | # include "sched_debug.c" | |
844 | #endif | |
845 | ||
846 | #define sched_class_highest (&rt_sched_class) | |
847 | ||
9c217245 IM |
848 | /* |
849 | * Update delta_exec, delta_fair fields for rq. | |
850 | * | |
851 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 852 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
853 | * delta_exec advances at the same rate as wall-clock (provided |
854 | * cpu is not idle). | |
855 | * | |
856 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
857 | * runqueue over any given interval. This (smoothened) load is used | |
858 | * during load balance. | |
859 | * | |
495eca49 | 860 | * This function is called /before/ updating rq->load |
9c217245 IM |
861 | * and when switching tasks. |
862 | */ | |
29b4b623 | 863 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 864 | { |
495eca49 | 865 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
866 | } |
867 | ||
79b5dddf | 868 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 869 | { |
495eca49 | 870 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
871 | } |
872 | ||
e5fa2237 | 873 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
874 | { |
875 | rq->nr_running++; | |
29b4b623 | 876 | inc_load(rq, p); |
9c217245 IM |
877 | } |
878 | ||
db53181e | 879 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
880 | { |
881 | rq->nr_running--; | |
79b5dddf | 882 | dec_load(rq, p); |
9c217245 IM |
883 | } |
884 | ||
45bf76df IM |
885 | static void set_load_weight(struct task_struct *p) |
886 | { | |
887 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
888 | p->se.load.weight = prio_to_weight[0] * 2; |
889 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
890 | return; | |
891 | } | |
45bf76df | 892 | |
dd41f596 IM |
893 | /* |
894 | * SCHED_IDLE tasks get minimal weight: | |
895 | */ | |
896 | if (p->policy == SCHED_IDLE) { | |
897 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
898 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
899 | return; | |
900 | } | |
71f8bd46 | 901 | |
dd41f596 IM |
902 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
903 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
904 | } |
905 | ||
8159f87e | 906 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 907 | { |
dd41f596 | 908 | sched_info_queued(p); |
fd390f6a | 909 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 910 | p->se.on_rq = 1; |
71f8bd46 IM |
911 | } |
912 | ||
69be72c1 | 913 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 914 | { |
f02231e5 | 915 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 916 | p->se.on_rq = 0; |
71f8bd46 IM |
917 | } |
918 | ||
14531189 | 919 | /* |
dd41f596 | 920 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 921 | */ |
14531189 IM |
922 | static inline int __normal_prio(struct task_struct *p) |
923 | { | |
dd41f596 | 924 | return p->static_prio; |
14531189 IM |
925 | } |
926 | ||
b29739f9 IM |
927 | /* |
928 | * Calculate the expected normal priority: i.e. priority | |
929 | * without taking RT-inheritance into account. Might be | |
930 | * boosted by interactivity modifiers. Changes upon fork, | |
931 | * setprio syscalls, and whenever the interactivity | |
932 | * estimator recalculates. | |
933 | */ | |
36c8b586 | 934 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
935 | { |
936 | int prio; | |
937 | ||
e05606d3 | 938 | if (task_has_rt_policy(p)) |
b29739f9 IM |
939 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
940 | else | |
941 | prio = __normal_prio(p); | |
942 | return prio; | |
943 | } | |
944 | ||
945 | /* | |
946 | * Calculate the current priority, i.e. the priority | |
947 | * taken into account by the scheduler. This value might | |
948 | * be boosted by RT tasks, or might be boosted by | |
949 | * interactivity modifiers. Will be RT if the task got | |
950 | * RT-boosted. If not then it returns p->normal_prio. | |
951 | */ | |
36c8b586 | 952 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
953 | { |
954 | p->normal_prio = normal_prio(p); | |
955 | /* | |
956 | * If we are RT tasks or we were boosted to RT priority, | |
957 | * keep the priority unchanged. Otherwise, update priority | |
958 | * to the normal priority: | |
959 | */ | |
960 | if (!rt_prio(p->prio)) | |
961 | return p->normal_prio; | |
962 | return p->prio; | |
963 | } | |
964 | ||
1da177e4 | 965 | /* |
dd41f596 | 966 | * activate_task - move a task to the runqueue. |
1da177e4 | 967 | */ |
dd41f596 | 968 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 969 | { |
dd41f596 IM |
970 | if (p->state == TASK_UNINTERRUPTIBLE) |
971 | rq->nr_uninterruptible--; | |
1da177e4 | 972 | |
8159f87e | 973 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 974 | inc_nr_running(p, rq); |
1da177e4 LT |
975 | } |
976 | ||
1da177e4 LT |
977 | /* |
978 | * deactivate_task - remove a task from the runqueue. | |
979 | */ | |
2e1cb74a | 980 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 981 | { |
dd41f596 IM |
982 | if (p->state == TASK_UNINTERRUPTIBLE) |
983 | rq->nr_uninterruptible++; | |
984 | ||
69be72c1 | 985 | dequeue_task(rq, p, sleep); |
db53181e | 986 | dec_nr_running(p, rq); |
1da177e4 LT |
987 | } |
988 | ||
1da177e4 LT |
989 | /** |
990 | * task_curr - is this task currently executing on a CPU? | |
991 | * @p: the task in question. | |
992 | */ | |
36c8b586 | 993 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
994 | { |
995 | return cpu_curr(task_cpu(p)) == p; | |
996 | } | |
997 | ||
2dd73a4f PW |
998 | /* Used instead of source_load when we know the type == 0 */ |
999 | unsigned long weighted_cpuload(const int cpu) | |
1000 | { | |
495eca49 | 1001 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1002 | } |
1003 | ||
1004 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1005 | { | |
1006 | #ifdef CONFIG_SMP | |
1007 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1008 | #endif |
29f59db3 | 1009 | set_task_cfs_rq(p); |
2dd73a4f PW |
1010 | } |
1011 | ||
1da177e4 | 1012 | #ifdef CONFIG_SMP |
c65cc870 | 1013 | |
cc367732 IM |
1014 | /* |
1015 | * Is this task likely cache-hot: | |
1016 | */ | |
1017 | static inline int | |
1018 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1019 | { | |
1020 | s64 delta; | |
1021 | ||
1022 | if (p->sched_class != &fair_sched_class) | |
1023 | return 0; | |
1024 | ||
6bc1665b IM |
1025 | if (sysctl_sched_migration_cost == -1) |
1026 | return 1; | |
1027 | if (sysctl_sched_migration_cost == 0) | |
1028 | return 0; | |
1029 | ||
cc367732 IM |
1030 | delta = now - p->se.exec_start; |
1031 | ||
1032 | return delta < (s64)sysctl_sched_migration_cost; | |
1033 | } | |
1034 | ||
1035 | ||
dd41f596 | 1036 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1037 | { |
dd41f596 IM |
1038 | int old_cpu = task_cpu(p); |
1039 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1040 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1041 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1042 | u64 clock_offset; |
dd41f596 IM |
1043 | |
1044 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1045 | |
1046 | #ifdef CONFIG_SCHEDSTATS | |
1047 | if (p->se.wait_start) | |
1048 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1049 | if (p->se.sleep_start) |
1050 | p->se.sleep_start -= clock_offset; | |
1051 | if (p->se.block_start) | |
1052 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1053 | if (old_cpu != new_cpu) { |
1054 | schedstat_inc(p, se.nr_migrations); | |
1055 | if (task_hot(p, old_rq->clock, NULL)) | |
1056 | schedstat_inc(p, se.nr_forced2_migrations); | |
1057 | } | |
6cfb0d5d | 1058 | #endif |
2830cf8c SV |
1059 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1060 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1061 | |
1062 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1063 | } |
1064 | ||
70b97a7f | 1065 | struct migration_req { |
1da177e4 | 1066 | struct list_head list; |
1da177e4 | 1067 | |
36c8b586 | 1068 | struct task_struct *task; |
1da177e4 LT |
1069 | int dest_cpu; |
1070 | ||
1da177e4 | 1071 | struct completion done; |
70b97a7f | 1072 | }; |
1da177e4 LT |
1073 | |
1074 | /* | |
1075 | * The task's runqueue lock must be held. | |
1076 | * Returns true if you have to wait for migration thread. | |
1077 | */ | |
36c8b586 | 1078 | static int |
70b97a7f | 1079 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1080 | { |
70b97a7f | 1081 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1082 | |
1083 | /* | |
1084 | * If the task is not on a runqueue (and not running), then | |
1085 | * it is sufficient to simply update the task's cpu field. | |
1086 | */ | |
dd41f596 | 1087 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1088 | set_task_cpu(p, dest_cpu); |
1089 | return 0; | |
1090 | } | |
1091 | ||
1092 | init_completion(&req->done); | |
1da177e4 LT |
1093 | req->task = p; |
1094 | req->dest_cpu = dest_cpu; | |
1095 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1096 | |
1da177e4 LT |
1097 | return 1; |
1098 | } | |
1099 | ||
1100 | /* | |
1101 | * wait_task_inactive - wait for a thread to unschedule. | |
1102 | * | |
1103 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1104 | * else this function might spin for a *long* time. This function can't | |
1105 | * be called with interrupts off, or it may introduce deadlock with | |
1106 | * smp_call_function() if an IPI is sent by the same process we are | |
1107 | * waiting to become inactive. | |
1108 | */ | |
36c8b586 | 1109 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1110 | { |
1111 | unsigned long flags; | |
dd41f596 | 1112 | int running, on_rq; |
70b97a7f | 1113 | struct rq *rq; |
1da177e4 | 1114 | |
3a5c359a AK |
1115 | for (;;) { |
1116 | /* | |
1117 | * We do the initial early heuristics without holding | |
1118 | * any task-queue locks at all. We'll only try to get | |
1119 | * the runqueue lock when things look like they will | |
1120 | * work out! | |
1121 | */ | |
1122 | rq = task_rq(p); | |
fa490cfd | 1123 | |
3a5c359a AK |
1124 | /* |
1125 | * If the task is actively running on another CPU | |
1126 | * still, just relax and busy-wait without holding | |
1127 | * any locks. | |
1128 | * | |
1129 | * NOTE! Since we don't hold any locks, it's not | |
1130 | * even sure that "rq" stays as the right runqueue! | |
1131 | * But we don't care, since "task_running()" will | |
1132 | * return false if the runqueue has changed and p | |
1133 | * is actually now running somewhere else! | |
1134 | */ | |
1135 | while (task_running(rq, p)) | |
1136 | cpu_relax(); | |
fa490cfd | 1137 | |
3a5c359a AK |
1138 | /* |
1139 | * Ok, time to look more closely! We need the rq | |
1140 | * lock now, to be *sure*. If we're wrong, we'll | |
1141 | * just go back and repeat. | |
1142 | */ | |
1143 | rq = task_rq_lock(p, &flags); | |
1144 | running = task_running(rq, p); | |
1145 | on_rq = p->se.on_rq; | |
1146 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1147 | |
3a5c359a AK |
1148 | /* |
1149 | * Was it really running after all now that we | |
1150 | * checked with the proper locks actually held? | |
1151 | * | |
1152 | * Oops. Go back and try again.. | |
1153 | */ | |
1154 | if (unlikely(running)) { | |
1155 | cpu_relax(); | |
1156 | continue; | |
1157 | } | |
fa490cfd | 1158 | |
3a5c359a AK |
1159 | /* |
1160 | * It's not enough that it's not actively running, | |
1161 | * it must be off the runqueue _entirely_, and not | |
1162 | * preempted! | |
1163 | * | |
1164 | * So if it wa still runnable (but just not actively | |
1165 | * running right now), it's preempted, and we should | |
1166 | * yield - it could be a while. | |
1167 | */ | |
1168 | if (unlikely(on_rq)) { | |
1169 | schedule_timeout_uninterruptible(1); | |
1170 | continue; | |
1171 | } | |
fa490cfd | 1172 | |
3a5c359a AK |
1173 | /* |
1174 | * Ahh, all good. It wasn't running, and it wasn't | |
1175 | * runnable, which means that it will never become | |
1176 | * running in the future either. We're all done! | |
1177 | */ | |
1178 | break; | |
1179 | } | |
1da177e4 LT |
1180 | } |
1181 | ||
1182 | /*** | |
1183 | * kick_process - kick a running thread to enter/exit the kernel | |
1184 | * @p: the to-be-kicked thread | |
1185 | * | |
1186 | * Cause a process which is running on another CPU to enter | |
1187 | * kernel-mode, without any delay. (to get signals handled.) | |
1188 | * | |
1189 | * NOTE: this function doesnt have to take the runqueue lock, | |
1190 | * because all it wants to ensure is that the remote task enters | |
1191 | * the kernel. If the IPI races and the task has been migrated | |
1192 | * to another CPU then no harm is done and the purpose has been | |
1193 | * achieved as well. | |
1194 | */ | |
36c8b586 | 1195 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1196 | { |
1197 | int cpu; | |
1198 | ||
1199 | preempt_disable(); | |
1200 | cpu = task_cpu(p); | |
1201 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1202 | smp_send_reschedule(cpu); | |
1203 | preempt_enable(); | |
1204 | } | |
1205 | ||
1206 | /* | |
2dd73a4f PW |
1207 | * Return a low guess at the load of a migration-source cpu weighted |
1208 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1209 | * |
1210 | * We want to under-estimate the load of migration sources, to | |
1211 | * balance conservatively. | |
1212 | */ | |
a9957449 | 1213 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1214 | { |
70b97a7f | 1215 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1216 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1217 | |
3b0bd9bc | 1218 | if (type == 0) |
dd41f596 | 1219 | return total; |
b910472d | 1220 | |
dd41f596 | 1221 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1222 | } |
1223 | ||
1224 | /* | |
2dd73a4f PW |
1225 | * Return a high guess at the load of a migration-target cpu weighted |
1226 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1227 | */ |
a9957449 | 1228 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1229 | { |
70b97a7f | 1230 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1231 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1232 | |
7897986b | 1233 | if (type == 0) |
dd41f596 | 1234 | return total; |
3b0bd9bc | 1235 | |
dd41f596 | 1236 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1237 | } |
1238 | ||
1239 | /* | |
1240 | * Return the average load per task on the cpu's run queue | |
1241 | */ | |
1242 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1243 | { | |
70b97a7f | 1244 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1245 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1246 | unsigned long n = rq->nr_running; |
1247 | ||
dd41f596 | 1248 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1249 | } |
1250 | ||
147cbb4b NP |
1251 | /* |
1252 | * find_idlest_group finds and returns the least busy CPU group within the | |
1253 | * domain. | |
1254 | */ | |
1255 | static struct sched_group * | |
1256 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1257 | { | |
1258 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1259 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1260 | int load_idx = sd->forkexec_idx; | |
1261 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1262 | ||
1263 | do { | |
1264 | unsigned long load, avg_load; | |
1265 | int local_group; | |
1266 | int i; | |
1267 | ||
da5a5522 BD |
1268 | /* Skip over this group if it has no CPUs allowed */ |
1269 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1270 | continue; |
da5a5522 | 1271 | |
147cbb4b | 1272 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1273 | |
1274 | /* Tally up the load of all CPUs in the group */ | |
1275 | avg_load = 0; | |
1276 | ||
1277 | for_each_cpu_mask(i, group->cpumask) { | |
1278 | /* Bias balancing toward cpus of our domain */ | |
1279 | if (local_group) | |
1280 | load = source_load(i, load_idx); | |
1281 | else | |
1282 | load = target_load(i, load_idx); | |
1283 | ||
1284 | avg_load += load; | |
1285 | } | |
1286 | ||
1287 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1288 | avg_load = sg_div_cpu_power(group, |
1289 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1290 | |
1291 | if (local_group) { | |
1292 | this_load = avg_load; | |
1293 | this = group; | |
1294 | } else if (avg_load < min_load) { | |
1295 | min_load = avg_load; | |
1296 | idlest = group; | |
1297 | } | |
3a5c359a | 1298 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1299 | |
1300 | if (!idlest || 100*this_load < imbalance*min_load) | |
1301 | return NULL; | |
1302 | return idlest; | |
1303 | } | |
1304 | ||
1305 | /* | |
0feaece9 | 1306 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1307 | */ |
95cdf3b7 IM |
1308 | static int |
1309 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1310 | { |
da5a5522 | 1311 | cpumask_t tmp; |
147cbb4b NP |
1312 | unsigned long load, min_load = ULONG_MAX; |
1313 | int idlest = -1; | |
1314 | int i; | |
1315 | ||
da5a5522 BD |
1316 | /* Traverse only the allowed CPUs */ |
1317 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1318 | ||
1319 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1320 | load = weighted_cpuload(i); |
147cbb4b NP |
1321 | |
1322 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1323 | min_load = load; | |
1324 | idlest = i; | |
1325 | } | |
1326 | } | |
1327 | ||
1328 | return idlest; | |
1329 | } | |
1330 | ||
476d139c NP |
1331 | /* |
1332 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1333 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1334 | * SD_BALANCE_EXEC. | |
1335 | * | |
1336 | * Balance, ie. select the least loaded group. | |
1337 | * | |
1338 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1339 | * | |
1340 | * preempt must be disabled. | |
1341 | */ | |
1342 | static int sched_balance_self(int cpu, int flag) | |
1343 | { | |
1344 | struct task_struct *t = current; | |
1345 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1346 | |
c96d145e | 1347 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1348 | /* |
1349 | * If power savings logic is enabled for a domain, stop there. | |
1350 | */ | |
5c45bf27 SS |
1351 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1352 | break; | |
476d139c NP |
1353 | if (tmp->flags & flag) |
1354 | sd = tmp; | |
c96d145e | 1355 | } |
476d139c NP |
1356 | |
1357 | while (sd) { | |
1358 | cpumask_t span; | |
1359 | struct sched_group *group; | |
1a848870 SS |
1360 | int new_cpu, weight; |
1361 | ||
1362 | if (!(sd->flags & flag)) { | |
1363 | sd = sd->child; | |
1364 | continue; | |
1365 | } | |
476d139c NP |
1366 | |
1367 | span = sd->span; | |
1368 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1369 | if (!group) { |
1370 | sd = sd->child; | |
1371 | continue; | |
1372 | } | |
476d139c | 1373 | |
da5a5522 | 1374 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1375 | if (new_cpu == -1 || new_cpu == cpu) { |
1376 | /* Now try balancing at a lower domain level of cpu */ | |
1377 | sd = sd->child; | |
1378 | continue; | |
1379 | } | |
476d139c | 1380 | |
1a848870 | 1381 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1382 | cpu = new_cpu; |
476d139c NP |
1383 | sd = NULL; |
1384 | weight = cpus_weight(span); | |
1385 | for_each_domain(cpu, tmp) { | |
1386 | if (weight <= cpus_weight(tmp->span)) | |
1387 | break; | |
1388 | if (tmp->flags & flag) | |
1389 | sd = tmp; | |
1390 | } | |
1391 | /* while loop will break here if sd == NULL */ | |
1392 | } | |
1393 | ||
1394 | return cpu; | |
1395 | } | |
1396 | ||
1397 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1398 | |
1399 | /* | |
1400 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1401 | * not idle and an idle cpu is available. The span of cpus to | |
1402 | * search starts with cpus closest then further out as needed, | |
1403 | * so we always favor a closer, idle cpu. | |
1404 | * | |
1405 | * Returns the CPU we should wake onto. | |
1406 | */ | |
1407 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1408 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1409 | { |
1410 | cpumask_t tmp; | |
1411 | struct sched_domain *sd; | |
1412 | int i; | |
1413 | ||
4953198b SS |
1414 | /* |
1415 | * If it is idle, then it is the best cpu to run this task. | |
1416 | * | |
1417 | * This cpu is also the best, if it has more than one task already. | |
1418 | * Siblings must be also busy(in most cases) as they didn't already | |
1419 | * pickup the extra load from this cpu and hence we need not check | |
1420 | * sibling runqueue info. This will avoid the checks and cache miss | |
1421 | * penalities associated with that. | |
1422 | */ | |
1423 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1424 | return cpu; |
1425 | ||
1426 | for_each_domain(cpu, sd) { | |
1427 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1428 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1429 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1430 | if (idle_cpu(i)) { |
1431 | if (i != task_cpu(p)) { | |
1432 | schedstat_inc(p, | |
1433 | se.nr_wakeups_idle); | |
1434 | } | |
1da177e4 | 1435 | return i; |
cc367732 | 1436 | } |
1da177e4 | 1437 | } |
9761eea8 | 1438 | } else { |
e0f364f4 | 1439 | break; |
9761eea8 | 1440 | } |
1da177e4 LT |
1441 | } |
1442 | return cpu; | |
1443 | } | |
1444 | #else | |
36c8b586 | 1445 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1446 | { |
1447 | return cpu; | |
1448 | } | |
1449 | #endif | |
1450 | ||
1451 | /*** | |
1452 | * try_to_wake_up - wake up a thread | |
1453 | * @p: the to-be-woken-up thread | |
1454 | * @state: the mask of task states that can be woken | |
1455 | * @sync: do a synchronous wakeup? | |
1456 | * | |
1457 | * Put it on the run-queue if it's not already there. The "current" | |
1458 | * thread is always on the run-queue (except when the actual | |
1459 | * re-schedule is in progress), and as such you're allowed to do | |
1460 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1461 | * runnable without the overhead of this. | |
1462 | * | |
1463 | * returns failure only if the task is already active. | |
1464 | */ | |
36c8b586 | 1465 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1466 | { |
cc367732 | 1467 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1468 | unsigned long flags; |
1469 | long old_state; | |
70b97a7f | 1470 | struct rq *rq; |
1da177e4 | 1471 | #ifdef CONFIG_SMP |
7897986b | 1472 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1473 | unsigned long load, this_load; |
1da177e4 LT |
1474 | int new_cpu; |
1475 | #endif | |
1476 | ||
1477 | rq = task_rq_lock(p, &flags); | |
1478 | old_state = p->state; | |
1479 | if (!(old_state & state)) | |
1480 | goto out; | |
1481 | ||
dd41f596 | 1482 | if (p->se.on_rq) |
1da177e4 LT |
1483 | goto out_running; |
1484 | ||
1485 | cpu = task_cpu(p); | |
cc367732 | 1486 | orig_cpu = cpu; |
1da177e4 LT |
1487 | this_cpu = smp_processor_id(); |
1488 | ||
1489 | #ifdef CONFIG_SMP | |
1490 | if (unlikely(task_running(rq, p))) | |
1491 | goto out_activate; | |
1492 | ||
7897986b NP |
1493 | new_cpu = cpu; |
1494 | ||
2d72376b | 1495 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1496 | if (cpu == this_cpu) { |
1497 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1498 | goto out_set_cpu; |
1499 | } | |
1500 | ||
1501 | for_each_domain(this_cpu, sd) { | |
1502 | if (cpu_isset(cpu, sd->span)) { | |
1503 | schedstat_inc(sd, ttwu_wake_remote); | |
1504 | this_sd = sd; | |
1505 | break; | |
1da177e4 LT |
1506 | } |
1507 | } | |
1da177e4 | 1508 | |
7897986b | 1509 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1510 | goto out_set_cpu; |
1511 | ||
1da177e4 | 1512 | /* |
7897986b | 1513 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1514 | */ |
7897986b NP |
1515 | if (this_sd) { |
1516 | int idx = this_sd->wake_idx; | |
1517 | unsigned int imbalance; | |
1da177e4 | 1518 | |
a3f21bce NP |
1519 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1520 | ||
7897986b NP |
1521 | load = source_load(cpu, idx); |
1522 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1523 | |
7897986b NP |
1524 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1525 | ||
a3f21bce NP |
1526 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1527 | unsigned long tl = this_load; | |
33859f7f MOS |
1528 | unsigned long tl_per_task; |
1529 | ||
71e20f18 IM |
1530 | /* |
1531 | * Attract cache-cold tasks on sync wakeups: | |
1532 | */ | |
1533 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1534 | goto out_set_cpu; | |
1535 | ||
cc367732 | 1536 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1537 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1538 | |
1da177e4 | 1539 | /* |
a3f21bce NP |
1540 | * If sync wakeup then subtract the (maximum possible) |
1541 | * effect of the currently running task from the load | |
1542 | * of the current CPU: | |
1da177e4 | 1543 | */ |
a3f21bce | 1544 | if (sync) |
dd41f596 | 1545 | tl -= current->se.load.weight; |
a3f21bce NP |
1546 | |
1547 | if ((tl <= load && | |
2dd73a4f | 1548 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1549 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1550 | /* |
1551 | * This domain has SD_WAKE_AFFINE and | |
1552 | * p is cache cold in this domain, and | |
1553 | * there is no bad imbalance. | |
1554 | */ | |
1555 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1556 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1557 | goto out_set_cpu; |
1558 | } | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * Start passive balancing when half the imbalance_pct | |
1563 | * limit is reached. | |
1564 | */ | |
1565 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1566 | if (imbalance*this_load <= 100*load) { | |
1567 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1568 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1569 | goto out_set_cpu; |
1570 | } | |
1da177e4 LT |
1571 | } |
1572 | } | |
1573 | ||
1574 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1575 | out_set_cpu: | |
1576 | new_cpu = wake_idle(new_cpu, p); | |
1577 | if (new_cpu != cpu) { | |
1578 | set_task_cpu(p, new_cpu); | |
1579 | task_rq_unlock(rq, &flags); | |
1580 | /* might preempt at this point */ | |
1581 | rq = task_rq_lock(p, &flags); | |
1582 | old_state = p->state; | |
1583 | if (!(old_state & state)) | |
1584 | goto out; | |
dd41f596 | 1585 | if (p->se.on_rq) |
1da177e4 LT |
1586 | goto out_running; |
1587 | ||
1588 | this_cpu = smp_processor_id(); | |
1589 | cpu = task_cpu(p); | |
1590 | } | |
1591 | ||
1592 | out_activate: | |
1593 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1594 | schedstat_inc(p, se.nr_wakeups); |
1595 | if (sync) | |
1596 | schedstat_inc(p, se.nr_wakeups_sync); | |
1597 | if (orig_cpu != cpu) | |
1598 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1599 | if (cpu == this_cpu) | |
1600 | schedstat_inc(p, se.nr_wakeups_local); | |
1601 | else | |
1602 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1603 | update_rq_clock(rq); |
dd41f596 | 1604 | activate_task(rq, p, 1); |
9c63d9c0 | 1605 | check_preempt_curr(rq, p); |
1da177e4 LT |
1606 | success = 1; |
1607 | ||
1608 | out_running: | |
1609 | p->state = TASK_RUNNING; | |
1610 | out: | |
1611 | task_rq_unlock(rq, &flags); | |
1612 | ||
1613 | return success; | |
1614 | } | |
1615 | ||
36c8b586 | 1616 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1617 | { |
1618 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1619 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1620 | } | |
1da177e4 LT |
1621 | EXPORT_SYMBOL(wake_up_process); |
1622 | ||
36c8b586 | 1623 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1624 | { |
1625 | return try_to_wake_up(p, state, 0); | |
1626 | } | |
1627 | ||
1da177e4 LT |
1628 | /* |
1629 | * Perform scheduler related setup for a newly forked process p. | |
1630 | * p is forked by current. | |
dd41f596 IM |
1631 | * |
1632 | * __sched_fork() is basic setup used by init_idle() too: | |
1633 | */ | |
1634 | static void __sched_fork(struct task_struct *p) | |
1635 | { | |
dd41f596 IM |
1636 | p->se.exec_start = 0; |
1637 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1638 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1639 | |
1640 | #ifdef CONFIG_SCHEDSTATS | |
1641 | p->se.wait_start = 0; | |
dd41f596 IM |
1642 | p->se.sum_sleep_runtime = 0; |
1643 | p->se.sleep_start = 0; | |
dd41f596 IM |
1644 | p->se.block_start = 0; |
1645 | p->se.sleep_max = 0; | |
1646 | p->se.block_max = 0; | |
1647 | p->se.exec_max = 0; | |
eba1ed4b | 1648 | p->se.slice_max = 0; |
dd41f596 | 1649 | p->se.wait_max = 0; |
6cfb0d5d | 1650 | #endif |
476d139c | 1651 | |
dd41f596 IM |
1652 | INIT_LIST_HEAD(&p->run_list); |
1653 | p->se.on_rq = 0; | |
476d139c | 1654 | |
e107be36 AK |
1655 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1656 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1657 | #endif | |
1658 | ||
1da177e4 LT |
1659 | /* |
1660 | * We mark the process as running here, but have not actually | |
1661 | * inserted it onto the runqueue yet. This guarantees that | |
1662 | * nobody will actually run it, and a signal or other external | |
1663 | * event cannot wake it up and insert it on the runqueue either. | |
1664 | */ | |
1665 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1666 | } |
1667 | ||
1668 | /* | |
1669 | * fork()/clone()-time setup: | |
1670 | */ | |
1671 | void sched_fork(struct task_struct *p, int clone_flags) | |
1672 | { | |
1673 | int cpu = get_cpu(); | |
1674 | ||
1675 | __sched_fork(p); | |
1676 | ||
1677 | #ifdef CONFIG_SMP | |
1678 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1679 | #endif | |
02e4bac2 | 1680 | set_task_cpu(p, cpu); |
b29739f9 IM |
1681 | |
1682 | /* | |
1683 | * Make sure we do not leak PI boosting priority to the child: | |
1684 | */ | |
1685 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1686 | if (!rt_prio(p->prio)) |
1687 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1688 | |
52f17b6c | 1689 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1690 | if (likely(sched_info_on())) |
52f17b6c | 1691 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1692 | #endif |
d6077cb8 | 1693 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1694 | p->oncpu = 0; |
1695 | #endif | |
1da177e4 | 1696 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1697 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1698 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1699 | #endif |
476d139c | 1700 | put_cpu(); |
1da177e4 LT |
1701 | } |
1702 | ||
1703 | /* | |
1704 | * wake_up_new_task - wake up a newly created task for the first time. | |
1705 | * | |
1706 | * This function will do some initial scheduler statistics housekeeping | |
1707 | * that must be done for every newly created context, then puts the task | |
1708 | * on the runqueue and wakes it. | |
1709 | */ | |
36c8b586 | 1710 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1711 | { |
1712 | unsigned long flags; | |
dd41f596 | 1713 | struct rq *rq; |
1da177e4 LT |
1714 | |
1715 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1716 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1717 | update_rq_clock(rq); |
1da177e4 LT |
1718 | |
1719 | p->prio = effective_prio(p); | |
1720 | ||
b9dca1e0 | 1721 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1722 | activate_task(rq, p, 0); |
1da177e4 | 1723 | } else { |
1da177e4 | 1724 | /* |
dd41f596 IM |
1725 | * Let the scheduling class do new task startup |
1726 | * management (if any): | |
1da177e4 | 1727 | */ |
ee0827d8 | 1728 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1729 | inc_nr_running(p, rq); |
1da177e4 | 1730 | } |
dd41f596 IM |
1731 | check_preempt_curr(rq, p); |
1732 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1733 | } |
1734 | ||
e107be36 AK |
1735 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1736 | ||
1737 | /** | |
421cee29 RD |
1738 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1739 | * @notifier: notifier struct to register | |
e107be36 AK |
1740 | */ |
1741 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1742 | { | |
1743 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1744 | } | |
1745 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1746 | ||
1747 | /** | |
1748 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1749 | * @notifier: notifier struct to unregister |
e107be36 AK |
1750 | * |
1751 | * This is safe to call from within a preemption notifier. | |
1752 | */ | |
1753 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1754 | { | |
1755 | hlist_del(¬ifier->link); | |
1756 | } | |
1757 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1758 | ||
1759 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1760 | { | |
1761 | struct preempt_notifier *notifier; | |
1762 | struct hlist_node *node; | |
1763 | ||
1764 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1765 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1766 | } | |
1767 | ||
1768 | static void | |
1769 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1770 | struct task_struct *next) | |
1771 | { | |
1772 | struct preempt_notifier *notifier; | |
1773 | struct hlist_node *node; | |
1774 | ||
1775 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1776 | notifier->ops->sched_out(notifier, next); | |
1777 | } | |
1778 | ||
1779 | #else | |
1780 | ||
1781 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1782 | { | |
1783 | } | |
1784 | ||
1785 | static void | |
1786 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1787 | struct task_struct *next) | |
1788 | { | |
1789 | } | |
1790 | ||
1791 | #endif | |
1792 | ||
4866cde0 NP |
1793 | /** |
1794 | * prepare_task_switch - prepare to switch tasks | |
1795 | * @rq: the runqueue preparing to switch | |
421cee29 | 1796 | * @prev: the current task that is being switched out |
4866cde0 NP |
1797 | * @next: the task we are going to switch to. |
1798 | * | |
1799 | * This is called with the rq lock held and interrupts off. It must | |
1800 | * be paired with a subsequent finish_task_switch after the context | |
1801 | * switch. | |
1802 | * | |
1803 | * prepare_task_switch sets up locking and calls architecture specific | |
1804 | * hooks. | |
1805 | */ | |
e107be36 AK |
1806 | static inline void |
1807 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1808 | struct task_struct *next) | |
4866cde0 | 1809 | { |
e107be36 | 1810 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1811 | prepare_lock_switch(rq, next); |
1812 | prepare_arch_switch(next); | |
1813 | } | |
1814 | ||
1da177e4 LT |
1815 | /** |
1816 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1817 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1818 | * @prev: the thread we just switched away from. |
1819 | * | |
4866cde0 NP |
1820 | * finish_task_switch must be called after the context switch, paired |
1821 | * with a prepare_task_switch call before the context switch. | |
1822 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1823 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1824 | * |
1825 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1826 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1827 | * with the lock held can cause deadlocks; see schedule() for | |
1828 | * details.) | |
1829 | */ | |
a9957449 | 1830 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1831 | __releases(rq->lock) |
1832 | { | |
1da177e4 | 1833 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1834 | long prev_state; |
1da177e4 LT |
1835 | |
1836 | rq->prev_mm = NULL; | |
1837 | ||
1838 | /* | |
1839 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1840 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1841 | * schedule one last time. The schedule call will never return, and |
1842 | * the scheduled task must drop that reference. | |
c394cc9f | 1843 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1844 | * still held, otherwise prev could be scheduled on another cpu, die |
1845 | * there before we look at prev->state, and then the reference would | |
1846 | * be dropped twice. | |
1847 | * Manfred Spraul <manfred@colorfullife.com> | |
1848 | */ | |
55a101f8 | 1849 | prev_state = prev->state; |
4866cde0 NP |
1850 | finish_arch_switch(prev); |
1851 | finish_lock_switch(rq, prev); | |
e107be36 | 1852 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1853 | if (mm) |
1854 | mmdrop(mm); | |
c394cc9f | 1855 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1856 | /* |
1857 | * Remove function-return probe instances associated with this | |
1858 | * task and put them back on the free list. | |
9761eea8 | 1859 | */ |
c6fd91f0 | 1860 | kprobe_flush_task(prev); |
1da177e4 | 1861 | put_task_struct(prev); |
c6fd91f0 | 1862 | } |
1da177e4 LT |
1863 | } |
1864 | ||
1865 | /** | |
1866 | * schedule_tail - first thing a freshly forked thread must call. | |
1867 | * @prev: the thread we just switched away from. | |
1868 | */ | |
36c8b586 | 1869 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1870 | __releases(rq->lock) |
1871 | { | |
70b97a7f IM |
1872 | struct rq *rq = this_rq(); |
1873 | ||
4866cde0 NP |
1874 | finish_task_switch(rq, prev); |
1875 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1876 | /* In this case, finish_task_switch does not reenable preemption */ | |
1877 | preempt_enable(); | |
1878 | #endif | |
1da177e4 | 1879 | if (current->set_child_tid) |
b488893a | 1880 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1881 | } |
1882 | ||
1883 | /* | |
1884 | * context_switch - switch to the new MM and the new | |
1885 | * thread's register state. | |
1886 | */ | |
dd41f596 | 1887 | static inline void |
70b97a7f | 1888 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1889 | struct task_struct *next) |
1da177e4 | 1890 | { |
dd41f596 | 1891 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1892 | |
e107be36 | 1893 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1894 | mm = next->mm; |
1895 | oldmm = prev->active_mm; | |
9226d125 ZA |
1896 | /* |
1897 | * For paravirt, this is coupled with an exit in switch_to to | |
1898 | * combine the page table reload and the switch backend into | |
1899 | * one hypercall. | |
1900 | */ | |
1901 | arch_enter_lazy_cpu_mode(); | |
1902 | ||
dd41f596 | 1903 | if (unlikely(!mm)) { |
1da177e4 LT |
1904 | next->active_mm = oldmm; |
1905 | atomic_inc(&oldmm->mm_count); | |
1906 | enter_lazy_tlb(oldmm, next); | |
1907 | } else | |
1908 | switch_mm(oldmm, mm, next); | |
1909 | ||
dd41f596 | 1910 | if (unlikely(!prev->mm)) { |
1da177e4 | 1911 | prev->active_mm = NULL; |
1da177e4 LT |
1912 | rq->prev_mm = oldmm; |
1913 | } | |
3a5f5e48 IM |
1914 | /* |
1915 | * Since the runqueue lock will be released by the next | |
1916 | * task (which is an invalid locking op but in the case | |
1917 | * of the scheduler it's an obvious special-case), so we | |
1918 | * do an early lockdep release here: | |
1919 | */ | |
1920 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1921 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1922 | #endif |
1da177e4 LT |
1923 | |
1924 | /* Here we just switch the register state and the stack. */ | |
1925 | switch_to(prev, next, prev); | |
1926 | ||
dd41f596 IM |
1927 | barrier(); |
1928 | /* | |
1929 | * this_rq must be evaluated again because prev may have moved | |
1930 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1931 | * frame will be invalid. | |
1932 | */ | |
1933 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1934 | } |
1935 | ||
1936 | /* | |
1937 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1938 | * | |
1939 | * externally visible scheduler statistics: current number of runnable | |
1940 | * threads, current number of uninterruptible-sleeping threads, total | |
1941 | * number of context switches performed since bootup. | |
1942 | */ | |
1943 | unsigned long nr_running(void) | |
1944 | { | |
1945 | unsigned long i, sum = 0; | |
1946 | ||
1947 | for_each_online_cpu(i) | |
1948 | sum += cpu_rq(i)->nr_running; | |
1949 | ||
1950 | return sum; | |
1951 | } | |
1952 | ||
1953 | unsigned long nr_uninterruptible(void) | |
1954 | { | |
1955 | unsigned long i, sum = 0; | |
1956 | ||
0a945022 | 1957 | for_each_possible_cpu(i) |
1da177e4 LT |
1958 | sum += cpu_rq(i)->nr_uninterruptible; |
1959 | ||
1960 | /* | |
1961 | * Since we read the counters lockless, it might be slightly | |
1962 | * inaccurate. Do not allow it to go below zero though: | |
1963 | */ | |
1964 | if (unlikely((long)sum < 0)) | |
1965 | sum = 0; | |
1966 | ||
1967 | return sum; | |
1968 | } | |
1969 | ||
1970 | unsigned long long nr_context_switches(void) | |
1971 | { | |
cc94abfc SR |
1972 | int i; |
1973 | unsigned long long sum = 0; | |
1da177e4 | 1974 | |
0a945022 | 1975 | for_each_possible_cpu(i) |
1da177e4 LT |
1976 | sum += cpu_rq(i)->nr_switches; |
1977 | ||
1978 | return sum; | |
1979 | } | |
1980 | ||
1981 | unsigned long nr_iowait(void) | |
1982 | { | |
1983 | unsigned long i, sum = 0; | |
1984 | ||
0a945022 | 1985 | for_each_possible_cpu(i) |
1da177e4 LT |
1986 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1987 | ||
1988 | return sum; | |
1989 | } | |
1990 | ||
db1b1fef JS |
1991 | unsigned long nr_active(void) |
1992 | { | |
1993 | unsigned long i, running = 0, uninterruptible = 0; | |
1994 | ||
1995 | for_each_online_cpu(i) { | |
1996 | running += cpu_rq(i)->nr_running; | |
1997 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1998 | } | |
1999 | ||
2000 | if (unlikely((long)uninterruptible < 0)) | |
2001 | uninterruptible = 0; | |
2002 | ||
2003 | return running + uninterruptible; | |
2004 | } | |
2005 | ||
48f24c4d | 2006 | /* |
dd41f596 IM |
2007 | * Update rq->cpu_load[] statistics. This function is usually called every |
2008 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2009 | */ |
dd41f596 | 2010 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2011 | { |
495eca49 | 2012 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2013 | int i, scale; |
2014 | ||
2015 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2016 | |
2017 | /* Update our load: */ | |
2018 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2019 | unsigned long old_load, new_load; | |
2020 | ||
2021 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2022 | ||
2023 | old_load = this_rq->cpu_load[i]; | |
2024 | new_load = this_load; | |
a25707f3 IM |
2025 | /* |
2026 | * Round up the averaging division if load is increasing. This | |
2027 | * prevents us from getting stuck on 9 if the load is 10, for | |
2028 | * example. | |
2029 | */ | |
2030 | if (new_load > old_load) | |
2031 | new_load += scale-1; | |
dd41f596 IM |
2032 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2033 | } | |
48f24c4d IM |
2034 | } |
2035 | ||
dd41f596 IM |
2036 | #ifdef CONFIG_SMP |
2037 | ||
1da177e4 LT |
2038 | /* |
2039 | * double_rq_lock - safely lock two runqueues | |
2040 | * | |
2041 | * Note this does not disable interrupts like task_rq_lock, | |
2042 | * you need to do so manually before calling. | |
2043 | */ | |
70b97a7f | 2044 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2045 | __acquires(rq1->lock) |
2046 | __acquires(rq2->lock) | |
2047 | { | |
054b9108 | 2048 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2049 | if (rq1 == rq2) { |
2050 | spin_lock(&rq1->lock); | |
2051 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2052 | } else { | |
c96d145e | 2053 | if (rq1 < rq2) { |
1da177e4 LT |
2054 | spin_lock(&rq1->lock); |
2055 | spin_lock(&rq2->lock); | |
2056 | } else { | |
2057 | spin_lock(&rq2->lock); | |
2058 | spin_lock(&rq1->lock); | |
2059 | } | |
2060 | } | |
6e82a3be IM |
2061 | update_rq_clock(rq1); |
2062 | update_rq_clock(rq2); | |
1da177e4 LT |
2063 | } |
2064 | ||
2065 | /* | |
2066 | * double_rq_unlock - safely unlock two runqueues | |
2067 | * | |
2068 | * Note this does not restore interrupts like task_rq_unlock, | |
2069 | * you need to do so manually after calling. | |
2070 | */ | |
70b97a7f | 2071 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2072 | __releases(rq1->lock) |
2073 | __releases(rq2->lock) | |
2074 | { | |
2075 | spin_unlock(&rq1->lock); | |
2076 | if (rq1 != rq2) | |
2077 | spin_unlock(&rq2->lock); | |
2078 | else | |
2079 | __release(rq2->lock); | |
2080 | } | |
2081 | ||
2082 | /* | |
2083 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2084 | */ | |
70b97a7f | 2085 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2086 | __releases(this_rq->lock) |
2087 | __acquires(busiest->lock) | |
2088 | __acquires(this_rq->lock) | |
2089 | { | |
054b9108 KK |
2090 | if (unlikely(!irqs_disabled())) { |
2091 | /* printk() doesn't work good under rq->lock */ | |
2092 | spin_unlock(&this_rq->lock); | |
2093 | BUG_ON(1); | |
2094 | } | |
1da177e4 | 2095 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2096 | if (busiest < this_rq) { |
1da177e4 LT |
2097 | spin_unlock(&this_rq->lock); |
2098 | spin_lock(&busiest->lock); | |
2099 | spin_lock(&this_rq->lock); | |
2100 | } else | |
2101 | spin_lock(&busiest->lock); | |
2102 | } | |
2103 | } | |
2104 | ||
1da177e4 LT |
2105 | /* |
2106 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2107 | * This is accomplished by forcing the cpu_allowed mask to only | |
2108 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2109 | * the cpu_allowed mask is restored. | |
2110 | */ | |
36c8b586 | 2111 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2112 | { |
70b97a7f | 2113 | struct migration_req req; |
1da177e4 | 2114 | unsigned long flags; |
70b97a7f | 2115 | struct rq *rq; |
1da177e4 LT |
2116 | |
2117 | rq = task_rq_lock(p, &flags); | |
2118 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2119 | || unlikely(cpu_is_offline(dest_cpu))) | |
2120 | goto out; | |
2121 | ||
2122 | /* force the process onto the specified CPU */ | |
2123 | if (migrate_task(p, dest_cpu, &req)) { | |
2124 | /* Need to wait for migration thread (might exit: take ref). */ | |
2125 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2126 | |
1da177e4 LT |
2127 | get_task_struct(mt); |
2128 | task_rq_unlock(rq, &flags); | |
2129 | wake_up_process(mt); | |
2130 | put_task_struct(mt); | |
2131 | wait_for_completion(&req.done); | |
36c8b586 | 2132 | |
1da177e4 LT |
2133 | return; |
2134 | } | |
2135 | out: | |
2136 | task_rq_unlock(rq, &flags); | |
2137 | } | |
2138 | ||
2139 | /* | |
476d139c NP |
2140 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2141 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2142 | */ |
2143 | void sched_exec(void) | |
2144 | { | |
1da177e4 | 2145 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2146 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2147 | put_cpu(); |
476d139c NP |
2148 | if (new_cpu != this_cpu) |
2149 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2150 | } |
2151 | ||
2152 | /* | |
2153 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2154 | * Both runqueues must be locked. | |
2155 | */ | |
dd41f596 IM |
2156 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2157 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2158 | { |
2e1cb74a | 2159 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2160 | set_task_cpu(p, this_cpu); |
dd41f596 | 2161 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2162 | /* |
2163 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2164 | * to be always true for them. | |
2165 | */ | |
dd41f596 | 2166 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2167 | } |
2168 | ||
2169 | /* | |
2170 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2171 | */ | |
858119e1 | 2172 | static |
70b97a7f | 2173 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2174 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2175 | int *all_pinned) |
1da177e4 LT |
2176 | { |
2177 | /* | |
2178 | * We do not migrate tasks that are: | |
2179 | * 1) running (obviously), or | |
2180 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2181 | * 3) are cache-hot on their current CPU. | |
2182 | */ | |
cc367732 IM |
2183 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2184 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2185 | return 0; |
cc367732 | 2186 | } |
81026794 NP |
2187 | *all_pinned = 0; |
2188 | ||
cc367732 IM |
2189 | if (task_running(rq, p)) { |
2190 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2191 | return 0; |
cc367732 | 2192 | } |
1da177e4 | 2193 | |
da84d961 IM |
2194 | /* |
2195 | * Aggressive migration if: | |
2196 | * 1) task is cache cold, or | |
2197 | * 2) too many balance attempts have failed. | |
2198 | */ | |
2199 | ||
6bc1665b IM |
2200 | if (!task_hot(p, rq->clock, sd) || |
2201 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2202 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2203 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2204 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2205 | schedstat_inc(p, se.nr_forced_migrations); |
2206 | } | |
da84d961 IM |
2207 | #endif |
2208 | return 1; | |
2209 | } | |
2210 | ||
cc367732 IM |
2211 | if (task_hot(p, rq->clock, sd)) { |
2212 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2213 | return 0; |
cc367732 | 2214 | } |
1da177e4 LT |
2215 | return 1; |
2216 | } | |
2217 | ||
dd41f596 | 2218 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2219 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2220 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2221 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2222 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2223 | { |
dd41f596 IM |
2224 | int pulled = 0, pinned = 0, skip_for_load; |
2225 | struct task_struct *p; | |
2226 | long rem_load_move = max_load_move; | |
1da177e4 | 2227 | |
2dd73a4f | 2228 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2229 | goto out; |
2230 | ||
81026794 NP |
2231 | pinned = 1; |
2232 | ||
1da177e4 | 2233 | /* |
dd41f596 | 2234 | * Start the load-balancing iterator: |
1da177e4 | 2235 | */ |
dd41f596 IM |
2236 | p = iterator->start(iterator->arg); |
2237 | next: | |
2238 | if (!p) | |
1da177e4 | 2239 | goto out; |
50ddd969 PW |
2240 | /* |
2241 | * To help distribute high priority tasks accross CPUs we don't | |
2242 | * skip a task if it will be the highest priority task (i.e. smallest | |
2243 | * prio value) on its new queue regardless of its load weight | |
2244 | */ | |
dd41f596 IM |
2245 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2246 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2247 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2248 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2249 | p = iterator->next(iterator->arg); |
2250 | goto next; | |
1da177e4 LT |
2251 | } |
2252 | ||
dd41f596 | 2253 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2254 | pulled++; |
dd41f596 | 2255 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2256 | |
2dd73a4f PW |
2257 | /* |
2258 | * We only want to steal up to the prescribed number of tasks | |
2259 | * and the prescribed amount of weighted load. | |
2260 | */ | |
2261 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2262 | if (p->prio < *this_best_prio) |
2263 | *this_best_prio = p->prio; | |
dd41f596 IM |
2264 | p = iterator->next(iterator->arg); |
2265 | goto next; | |
1da177e4 LT |
2266 | } |
2267 | out: | |
2268 | /* | |
2269 | * Right now, this is the only place pull_task() is called, | |
2270 | * so we can safely collect pull_task() stats here rather than | |
2271 | * inside pull_task(). | |
2272 | */ | |
2273 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2274 | |
2275 | if (all_pinned) | |
2276 | *all_pinned = pinned; | |
dd41f596 | 2277 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2278 | return pulled; |
2279 | } | |
2280 | ||
dd41f596 | 2281 | /* |
43010659 PW |
2282 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2283 | * this_rq, as part of a balancing operation within domain "sd". | |
2284 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2285 | * |
2286 | * Called with both runqueues locked. | |
2287 | */ | |
2288 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2289 | unsigned long max_load_move, |
dd41f596 IM |
2290 | struct sched_domain *sd, enum cpu_idle_type idle, |
2291 | int *all_pinned) | |
2292 | { | |
5522d5d5 | 2293 | const struct sched_class *class = sched_class_highest; |
43010659 | 2294 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2295 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2296 | |
2297 | do { | |
43010659 PW |
2298 | total_load_moved += |
2299 | class->load_balance(this_rq, this_cpu, busiest, | |
2300 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2301 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2302 | class = class->next; |
43010659 | 2303 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2304 | |
43010659 PW |
2305 | return total_load_moved > 0; |
2306 | } | |
2307 | ||
2308 | /* | |
2309 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2310 | * part of active balancing operations within "domain". | |
2311 | * Returns 1 if successful and 0 otherwise. | |
2312 | * | |
2313 | * Called with both runqueues locked. | |
2314 | */ | |
2315 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2316 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2317 | { | |
5522d5d5 | 2318 | const struct sched_class *class; |
a4ac01c3 | 2319 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2320 | |
2321 | for (class = sched_class_highest; class; class = class->next) | |
2322 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2323 | 1, ULONG_MAX, sd, idle, NULL, |
2324 | &this_best_prio)) | |
43010659 PW |
2325 | return 1; |
2326 | ||
2327 | return 0; | |
dd41f596 IM |
2328 | } |
2329 | ||
1da177e4 LT |
2330 | /* |
2331 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2332 | * domain. It calculates and returns the amount of weighted load which |
2333 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2334 | */ |
2335 | static struct sched_group * | |
2336 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2337 | unsigned long *imbalance, enum cpu_idle_type idle, |
2338 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2339 | { |
2340 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2341 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2342 | unsigned long max_pull; |
2dd73a4f PW |
2343 | unsigned long busiest_load_per_task, busiest_nr_running; |
2344 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2345 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2346 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2347 | int power_savings_balance = 1; | |
2348 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2349 | unsigned long min_nr_running = ULONG_MAX; | |
2350 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2351 | #endif | |
1da177e4 LT |
2352 | |
2353 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2354 | busiest_load_per_task = busiest_nr_running = 0; |
2355 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2356 | if (idle == CPU_NOT_IDLE) |
7897986b | 2357 | load_idx = sd->busy_idx; |
d15bcfdb | 2358 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2359 | load_idx = sd->newidle_idx; |
2360 | else | |
2361 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2362 | |
2363 | do { | |
908a7c1b | 2364 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2365 | int local_group; |
2366 | int i; | |
908a7c1b | 2367 | int __group_imb = 0; |
783609c6 | 2368 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2369 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2370 | |
2371 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2372 | ||
783609c6 SS |
2373 | if (local_group) |
2374 | balance_cpu = first_cpu(group->cpumask); | |
2375 | ||
1da177e4 | 2376 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2377 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2378 | max_cpu_load = 0; |
2379 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2380 | |
2381 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2382 | struct rq *rq; |
2383 | ||
2384 | if (!cpu_isset(i, *cpus)) | |
2385 | continue; | |
2386 | ||
2387 | rq = cpu_rq(i); | |
2dd73a4f | 2388 | |
9439aab8 | 2389 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2390 | *sd_idle = 0; |
2391 | ||
1da177e4 | 2392 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2393 | if (local_group) { |
2394 | if (idle_cpu(i) && !first_idle_cpu) { | |
2395 | first_idle_cpu = 1; | |
2396 | balance_cpu = i; | |
2397 | } | |
2398 | ||
a2000572 | 2399 | load = target_load(i, load_idx); |
908a7c1b | 2400 | } else { |
a2000572 | 2401 | load = source_load(i, load_idx); |
908a7c1b KC |
2402 | if (load > max_cpu_load) |
2403 | max_cpu_load = load; | |
2404 | if (min_cpu_load > load) | |
2405 | min_cpu_load = load; | |
2406 | } | |
1da177e4 LT |
2407 | |
2408 | avg_load += load; | |
2dd73a4f | 2409 | sum_nr_running += rq->nr_running; |
dd41f596 | 2410 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2411 | } |
2412 | ||
783609c6 SS |
2413 | /* |
2414 | * First idle cpu or the first cpu(busiest) in this sched group | |
2415 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2416 | * domains. In the newly idle case, we will allow all the cpu's |
2417 | * to do the newly idle load balance. | |
783609c6 | 2418 | */ |
9439aab8 SS |
2419 | if (idle != CPU_NEWLY_IDLE && local_group && |
2420 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2421 | *balance = 0; |
2422 | goto ret; | |
2423 | } | |
2424 | ||
1da177e4 | 2425 | total_load += avg_load; |
5517d86b | 2426 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2427 | |
2428 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2429 | avg_load = sg_div_cpu_power(group, |
2430 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2431 | |
908a7c1b KC |
2432 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2433 | __group_imb = 1; | |
2434 | ||
5517d86b | 2435 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2436 | |
1da177e4 LT |
2437 | if (local_group) { |
2438 | this_load = avg_load; | |
2439 | this = group; | |
2dd73a4f PW |
2440 | this_nr_running = sum_nr_running; |
2441 | this_load_per_task = sum_weighted_load; | |
2442 | } else if (avg_load > max_load && | |
908a7c1b | 2443 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2444 | max_load = avg_load; |
2445 | busiest = group; | |
2dd73a4f PW |
2446 | busiest_nr_running = sum_nr_running; |
2447 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2448 | group_imb = __group_imb; |
1da177e4 | 2449 | } |
5c45bf27 SS |
2450 | |
2451 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2452 | /* | |
2453 | * Busy processors will not participate in power savings | |
2454 | * balance. | |
2455 | */ | |
dd41f596 IM |
2456 | if (idle == CPU_NOT_IDLE || |
2457 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2458 | goto group_next; | |
5c45bf27 SS |
2459 | |
2460 | /* | |
2461 | * If the local group is idle or completely loaded | |
2462 | * no need to do power savings balance at this domain | |
2463 | */ | |
2464 | if (local_group && (this_nr_running >= group_capacity || | |
2465 | !this_nr_running)) | |
2466 | power_savings_balance = 0; | |
2467 | ||
dd41f596 | 2468 | /* |
5c45bf27 SS |
2469 | * If a group is already running at full capacity or idle, |
2470 | * don't include that group in power savings calculations | |
dd41f596 IM |
2471 | */ |
2472 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2473 | || !sum_nr_running) |
dd41f596 | 2474 | goto group_next; |
5c45bf27 | 2475 | |
dd41f596 | 2476 | /* |
5c45bf27 | 2477 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2478 | * This is the group from where we need to pick up the load |
2479 | * for saving power | |
2480 | */ | |
2481 | if ((sum_nr_running < min_nr_running) || | |
2482 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2483 | first_cpu(group->cpumask) < |
2484 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2485 | group_min = group; |
2486 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2487 | min_load_per_task = sum_weighted_load / |
2488 | sum_nr_running; | |
dd41f596 | 2489 | } |
5c45bf27 | 2490 | |
dd41f596 | 2491 | /* |
5c45bf27 | 2492 | * Calculate the group which is almost near its |
dd41f596 IM |
2493 | * capacity but still has some space to pick up some load |
2494 | * from other group and save more power | |
2495 | */ | |
2496 | if (sum_nr_running <= group_capacity - 1) { | |
2497 | if (sum_nr_running > leader_nr_running || | |
2498 | (sum_nr_running == leader_nr_running && | |
2499 | first_cpu(group->cpumask) > | |
2500 | first_cpu(group_leader->cpumask))) { | |
2501 | group_leader = group; | |
2502 | leader_nr_running = sum_nr_running; | |
2503 | } | |
48f24c4d | 2504 | } |
5c45bf27 SS |
2505 | group_next: |
2506 | #endif | |
1da177e4 LT |
2507 | group = group->next; |
2508 | } while (group != sd->groups); | |
2509 | ||
2dd73a4f | 2510 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2511 | goto out_balanced; |
2512 | ||
2513 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2514 | ||
2515 | if (this_load >= avg_load || | |
2516 | 100*max_load <= sd->imbalance_pct*this_load) | |
2517 | goto out_balanced; | |
2518 | ||
2dd73a4f | 2519 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2520 | if (group_imb) |
2521 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2522 | ||
1da177e4 LT |
2523 | /* |
2524 | * We're trying to get all the cpus to the average_load, so we don't | |
2525 | * want to push ourselves above the average load, nor do we wish to | |
2526 | * reduce the max loaded cpu below the average load, as either of these | |
2527 | * actions would just result in more rebalancing later, and ping-pong | |
2528 | * tasks around. Thus we look for the minimum possible imbalance. | |
2529 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2530 | * be counted as no imbalance for these purposes -- we can't fix that | |
2531 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2532 | * appear as very large values with unsigned longs. | |
2533 | */ | |
2dd73a4f PW |
2534 | if (max_load <= busiest_load_per_task) |
2535 | goto out_balanced; | |
2536 | ||
2537 | /* | |
2538 | * In the presence of smp nice balancing, certain scenarios can have | |
2539 | * max load less than avg load(as we skip the groups at or below | |
2540 | * its cpu_power, while calculating max_load..) | |
2541 | */ | |
2542 | if (max_load < avg_load) { | |
2543 | *imbalance = 0; | |
2544 | goto small_imbalance; | |
2545 | } | |
0c117f1b SS |
2546 | |
2547 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2548 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2549 | |
1da177e4 | 2550 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2551 | *imbalance = min(max_pull * busiest->__cpu_power, |
2552 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2553 | / SCHED_LOAD_SCALE; |
2554 | ||
2dd73a4f PW |
2555 | /* |
2556 | * if *imbalance is less than the average load per runnable task | |
2557 | * there is no gaurantee that any tasks will be moved so we'll have | |
2558 | * a think about bumping its value to force at least one task to be | |
2559 | * moved | |
2560 | */ | |
7fd0d2dd | 2561 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2562 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2563 | unsigned int imbn; |
2564 | ||
2565 | small_imbalance: | |
2566 | pwr_move = pwr_now = 0; | |
2567 | imbn = 2; | |
2568 | if (this_nr_running) { | |
2569 | this_load_per_task /= this_nr_running; | |
2570 | if (busiest_load_per_task > this_load_per_task) | |
2571 | imbn = 1; | |
2572 | } else | |
2573 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2574 | |
dd41f596 IM |
2575 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2576 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2577 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2578 | return busiest; |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * OK, we don't have enough imbalance to justify moving tasks, | |
2583 | * however we may be able to increase total CPU power used by | |
2584 | * moving them. | |
2585 | */ | |
2586 | ||
5517d86b ED |
2587 | pwr_now += busiest->__cpu_power * |
2588 | min(busiest_load_per_task, max_load); | |
2589 | pwr_now += this->__cpu_power * | |
2590 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2591 | pwr_now /= SCHED_LOAD_SCALE; |
2592 | ||
2593 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2594 | tmp = sg_div_cpu_power(busiest, |
2595 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2596 | if (max_load > tmp) |
5517d86b | 2597 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2598 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2599 | |
2600 | /* Amount of load we'd add */ | |
5517d86b | 2601 | if (max_load * busiest->__cpu_power < |
33859f7f | 2602 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2603 | tmp = sg_div_cpu_power(this, |
2604 | max_load * busiest->__cpu_power); | |
1da177e4 | 2605 | else |
5517d86b ED |
2606 | tmp = sg_div_cpu_power(this, |
2607 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2608 | pwr_move += this->__cpu_power * | |
2609 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2610 | pwr_move /= SCHED_LOAD_SCALE; |
2611 | ||
2612 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2613 | if (pwr_move > pwr_now) |
2614 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2615 | } |
2616 | ||
1da177e4 LT |
2617 | return busiest; |
2618 | ||
2619 | out_balanced: | |
5c45bf27 | 2620 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2621 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2622 | goto ret; |
1da177e4 | 2623 | |
5c45bf27 SS |
2624 | if (this == group_leader && group_leader != group_min) { |
2625 | *imbalance = min_load_per_task; | |
2626 | return group_min; | |
2627 | } | |
5c45bf27 | 2628 | #endif |
783609c6 | 2629 | ret: |
1da177e4 LT |
2630 | *imbalance = 0; |
2631 | return NULL; | |
2632 | } | |
2633 | ||
2634 | /* | |
2635 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2636 | */ | |
70b97a7f | 2637 | static struct rq * |
d15bcfdb | 2638 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2639 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2640 | { |
70b97a7f | 2641 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2642 | unsigned long max_load = 0; |
1da177e4 LT |
2643 | int i; |
2644 | ||
2645 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2646 | unsigned long wl; |
0a2966b4 CL |
2647 | |
2648 | if (!cpu_isset(i, *cpus)) | |
2649 | continue; | |
2650 | ||
48f24c4d | 2651 | rq = cpu_rq(i); |
dd41f596 | 2652 | wl = weighted_cpuload(i); |
2dd73a4f | 2653 | |
dd41f596 | 2654 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2655 | continue; |
1da177e4 | 2656 | |
dd41f596 IM |
2657 | if (wl > max_load) { |
2658 | max_load = wl; | |
48f24c4d | 2659 | busiest = rq; |
1da177e4 LT |
2660 | } |
2661 | } | |
2662 | ||
2663 | return busiest; | |
2664 | } | |
2665 | ||
77391d71 NP |
2666 | /* |
2667 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2668 | * so long as it is large enough. | |
2669 | */ | |
2670 | #define MAX_PINNED_INTERVAL 512 | |
2671 | ||
1da177e4 LT |
2672 | /* |
2673 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2674 | * tasks if there is an imbalance. | |
1da177e4 | 2675 | */ |
70b97a7f | 2676 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2677 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2678 | int *balance) |
1da177e4 | 2679 | { |
43010659 | 2680 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2681 | struct sched_group *group; |
1da177e4 | 2682 | unsigned long imbalance; |
70b97a7f | 2683 | struct rq *busiest; |
0a2966b4 | 2684 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2685 | unsigned long flags; |
5969fe06 | 2686 | |
89c4710e SS |
2687 | /* |
2688 | * When power savings policy is enabled for the parent domain, idle | |
2689 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2690 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2691 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2692 | */ |
d15bcfdb | 2693 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2694 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2695 | sd_idle = 1; |
1da177e4 | 2696 | |
2d72376b | 2697 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2698 | |
0a2966b4 CL |
2699 | redo: |
2700 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2701 | &cpus, balance); |
2702 | ||
06066714 | 2703 | if (*balance == 0) |
783609c6 | 2704 | goto out_balanced; |
783609c6 | 2705 | |
1da177e4 LT |
2706 | if (!group) { |
2707 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2708 | goto out_balanced; | |
2709 | } | |
2710 | ||
0a2966b4 | 2711 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2712 | if (!busiest) { |
2713 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2714 | goto out_balanced; | |
2715 | } | |
2716 | ||
db935dbd | 2717 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2718 | |
2719 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2720 | ||
43010659 | 2721 | ld_moved = 0; |
1da177e4 LT |
2722 | if (busiest->nr_running > 1) { |
2723 | /* | |
2724 | * Attempt to move tasks. If find_busiest_group has found | |
2725 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2726 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2727 | * correctly treated as an imbalance. |
2728 | */ | |
fe2eea3f | 2729 | local_irq_save(flags); |
e17224bf | 2730 | double_rq_lock(this_rq, busiest); |
43010659 | 2731 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2732 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2733 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2734 | local_irq_restore(flags); |
81026794 | 2735 | |
46cb4b7c SS |
2736 | /* |
2737 | * some other cpu did the load balance for us. | |
2738 | */ | |
43010659 | 2739 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2740 | resched_cpu(this_cpu); |
2741 | ||
81026794 | 2742 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2743 | if (unlikely(all_pinned)) { |
2744 | cpu_clear(cpu_of(busiest), cpus); | |
2745 | if (!cpus_empty(cpus)) | |
2746 | goto redo; | |
81026794 | 2747 | goto out_balanced; |
0a2966b4 | 2748 | } |
1da177e4 | 2749 | } |
81026794 | 2750 | |
43010659 | 2751 | if (!ld_moved) { |
1da177e4 LT |
2752 | schedstat_inc(sd, lb_failed[idle]); |
2753 | sd->nr_balance_failed++; | |
2754 | ||
2755 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2756 | |
fe2eea3f | 2757 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2758 | |
2759 | /* don't kick the migration_thread, if the curr | |
2760 | * task on busiest cpu can't be moved to this_cpu | |
2761 | */ | |
2762 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2763 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2764 | all_pinned = 1; |
2765 | goto out_one_pinned; | |
2766 | } | |
2767 | ||
1da177e4 LT |
2768 | if (!busiest->active_balance) { |
2769 | busiest->active_balance = 1; | |
2770 | busiest->push_cpu = this_cpu; | |
81026794 | 2771 | active_balance = 1; |
1da177e4 | 2772 | } |
fe2eea3f | 2773 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2774 | if (active_balance) |
1da177e4 LT |
2775 | wake_up_process(busiest->migration_thread); |
2776 | ||
2777 | /* | |
2778 | * We've kicked active balancing, reset the failure | |
2779 | * counter. | |
2780 | */ | |
39507451 | 2781 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2782 | } |
81026794 | 2783 | } else |
1da177e4 LT |
2784 | sd->nr_balance_failed = 0; |
2785 | ||
81026794 | 2786 | if (likely(!active_balance)) { |
1da177e4 LT |
2787 | /* We were unbalanced, so reset the balancing interval */ |
2788 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2789 | } else { |
2790 | /* | |
2791 | * If we've begun active balancing, start to back off. This | |
2792 | * case may not be covered by the all_pinned logic if there | |
2793 | * is only 1 task on the busy runqueue (because we don't call | |
2794 | * move_tasks). | |
2795 | */ | |
2796 | if (sd->balance_interval < sd->max_interval) | |
2797 | sd->balance_interval *= 2; | |
1da177e4 LT |
2798 | } |
2799 | ||
43010659 | 2800 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2801 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2802 | return -1; |
43010659 | 2803 | return ld_moved; |
1da177e4 LT |
2804 | |
2805 | out_balanced: | |
1da177e4 LT |
2806 | schedstat_inc(sd, lb_balanced[idle]); |
2807 | ||
16cfb1c0 | 2808 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2809 | |
2810 | out_one_pinned: | |
1da177e4 | 2811 | /* tune up the balancing interval */ |
77391d71 NP |
2812 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2813 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2814 | sd->balance_interval *= 2; |
2815 | ||
48f24c4d | 2816 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2817 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2818 | return -1; |
1da177e4 LT |
2819 | return 0; |
2820 | } | |
2821 | ||
2822 | /* | |
2823 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2824 | * tasks if there is an imbalance. | |
2825 | * | |
d15bcfdb | 2826 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2827 | * this_rq is locked. |
2828 | */ | |
48f24c4d | 2829 | static int |
70b97a7f | 2830 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2831 | { |
2832 | struct sched_group *group; | |
70b97a7f | 2833 | struct rq *busiest = NULL; |
1da177e4 | 2834 | unsigned long imbalance; |
43010659 | 2835 | int ld_moved = 0; |
5969fe06 | 2836 | int sd_idle = 0; |
969bb4e4 | 2837 | int all_pinned = 0; |
0a2966b4 | 2838 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2839 | |
89c4710e SS |
2840 | /* |
2841 | * When power savings policy is enabled for the parent domain, idle | |
2842 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2843 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2844 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2845 | */ |
2846 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2847 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2848 | sd_idle = 1; |
1da177e4 | 2849 | |
2d72376b | 2850 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2851 | redo: |
d15bcfdb | 2852 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2853 | &sd_idle, &cpus, NULL); |
1da177e4 | 2854 | if (!group) { |
d15bcfdb | 2855 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2856 | goto out_balanced; |
1da177e4 LT |
2857 | } |
2858 | ||
d15bcfdb | 2859 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2860 | &cpus); |
db935dbd | 2861 | if (!busiest) { |
d15bcfdb | 2862 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2863 | goto out_balanced; |
1da177e4 LT |
2864 | } |
2865 | ||
db935dbd NP |
2866 | BUG_ON(busiest == this_rq); |
2867 | ||
d15bcfdb | 2868 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2869 | |
43010659 | 2870 | ld_moved = 0; |
d6d5cfaf NP |
2871 | if (busiest->nr_running > 1) { |
2872 | /* Attempt to move tasks */ | |
2873 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2874 | /* this_rq->clock is already updated */ |
2875 | update_rq_clock(busiest); | |
43010659 | 2876 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2877 | imbalance, sd, CPU_NEWLY_IDLE, |
2878 | &all_pinned); | |
d6d5cfaf | 2879 | spin_unlock(&busiest->lock); |
0a2966b4 | 2880 | |
969bb4e4 | 2881 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2882 | cpu_clear(cpu_of(busiest), cpus); |
2883 | if (!cpus_empty(cpus)) | |
2884 | goto redo; | |
2885 | } | |
d6d5cfaf NP |
2886 | } |
2887 | ||
43010659 | 2888 | if (!ld_moved) { |
d15bcfdb | 2889 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2890 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2891 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2892 | return -1; |
2893 | } else | |
16cfb1c0 | 2894 | sd->nr_balance_failed = 0; |
1da177e4 | 2895 | |
43010659 | 2896 | return ld_moved; |
16cfb1c0 NP |
2897 | |
2898 | out_balanced: | |
d15bcfdb | 2899 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2900 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2901 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2902 | return -1; |
16cfb1c0 | 2903 | sd->nr_balance_failed = 0; |
48f24c4d | 2904 | |
16cfb1c0 | 2905 | return 0; |
1da177e4 LT |
2906 | } |
2907 | ||
2908 | /* | |
2909 | * idle_balance is called by schedule() if this_cpu is about to become | |
2910 | * idle. Attempts to pull tasks from other CPUs. | |
2911 | */ | |
70b97a7f | 2912 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2913 | { |
2914 | struct sched_domain *sd; | |
dd41f596 IM |
2915 | int pulled_task = -1; |
2916 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2917 | |
2918 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2919 | unsigned long interval; |
2920 | ||
2921 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2922 | continue; | |
2923 | ||
2924 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2925 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2926 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2927 | this_rq, sd); |
2928 | ||
2929 | interval = msecs_to_jiffies(sd->balance_interval); | |
2930 | if (time_after(next_balance, sd->last_balance + interval)) | |
2931 | next_balance = sd->last_balance + interval; | |
2932 | if (pulled_task) | |
2933 | break; | |
1da177e4 | 2934 | } |
dd41f596 | 2935 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2936 | /* |
2937 | * We are going idle. next_balance may be set based on | |
2938 | * a busy processor. So reset next_balance. | |
2939 | */ | |
2940 | this_rq->next_balance = next_balance; | |
dd41f596 | 2941 | } |
1da177e4 LT |
2942 | } |
2943 | ||
2944 | /* | |
2945 | * active_load_balance is run by migration threads. It pushes running tasks | |
2946 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2947 | * running on each physical CPU where possible, and avoids physical / | |
2948 | * logical imbalances. | |
2949 | * | |
2950 | * Called with busiest_rq locked. | |
2951 | */ | |
70b97a7f | 2952 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2953 | { |
39507451 | 2954 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2955 | struct sched_domain *sd; |
2956 | struct rq *target_rq; | |
39507451 | 2957 | |
48f24c4d | 2958 | /* Is there any task to move? */ |
39507451 | 2959 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2960 | return; |
2961 | ||
2962 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2963 | |
2964 | /* | |
39507451 NP |
2965 | * This condition is "impossible", if it occurs |
2966 | * we need to fix it. Originally reported by | |
2967 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2968 | */ |
39507451 | 2969 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2970 | |
39507451 NP |
2971 | /* move a task from busiest_rq to target_rq */ |
2972 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2973 | update_rq_clock(busiest_rq); |
2974 | update_rq_clock(target_rq); | |
39507451 NP |
2975 | |
2976 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2977 | for_each_domain(target_cpu, sd) { |
39507451 | 2978 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2979 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2980 | break; |
c96d145e | 2981 | } |
39507451 | 2982 | |
48f24c4d | 2983 | if (likely(sd)) { |
2d72376b | 2984 | schedstat_inc(sd, alb_count); |
39507451 | 2985 | |
43010659 PW |
2986 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2987 | sd, CPU_IDLE)) | |
48f24c4d IM |
2988 | schedstat_inc(sd, alb_pushed); |
2989 | else | |
2990 | schedstat_inc(sd, alb_failed); | |
2991 | } | |
39507451 | 2992 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2993 | } |
2994 | ||
46cb4b7c SS |
2995 | #ifdef CONFIG_NO_HZ |
2996 | static struct { | |
2997 | atomic_t load_balancer; | |
2998 | cpumask_t cpu_mask; | |
2999 | } nohz ____cacheline_aligned = { | |
3000 | .load_balancer = ATOMIC_INIT(-1), | |
3001 | .cpu_mask = CPU_MASK_NONE, | |
3002 | }; | |
3003 | ||
7835b98b | 3004 | /* |
46cb4b7c SS |
3005 | * This routine will try to nominate the ilb (idle load balancing) |
3006 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3007 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3008 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3009 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3010 | * arrives... | |
3011 | * | |
3012 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3013 | * for idle load balancing. ilb owner will still be part of | |
3014 | * nohz.cpu_mask.. | |
7835b98b | 3015 | * |
46cb4b7c SS |
3016 | * While stopping the tick, this cpu will become the ilb owner if there |
3017 | * is no other owner. And will be the owner till that cpu becomes busy | |
3018 | * or if all cpus in the system stop their ticks at which point | |
3019 | * there is no need for ilb owner. | |
3020 | * | |
3021 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3022 | * next busy scheduler_tick() | |
3023 | */ | |
3024 | int select_nohz_load_balancer(int stop_tick) | |
3025 | { | |
3026 | int cpu = smp_processor_id(); | |
3027 | ||
3028 | if (stop_tick) { | |
3029 | cpu_set(cpu, nohz.cpu_mask); | |
3030 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3031 | ||
3032 | /* | |
3033 | * If we are going offline and still the leader, give up! | |
3034 | */ | |
3035 | if (cpu_is_offline(cpu) && | |
3036 | atomic_read(&nohz.load_balancer) == cpu) { | |
3037 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3038 | BUG(); | |
3039 | return 0; | |
3040 | } | |
3041 | ||
3042 | /* time for ilb owner also to sleep */ | |
3043 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3044 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3045 | atomic_set(&nohz.load_balancer, -1); | |
3046 | return 0; | |
3047 | } | |
3048 | ||
3049 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3050 | /* make me the ilb owner */ | |
3051 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3052 | return 1; | |
3053 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3054 | return 1; | |
3055 | } else { | |
3056 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3057 | return 0; | |
3058 | ||
3059 | cpu_clear(cpu, nohz.cpu_mask); | |
3060 | ||
3061 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3062 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3063 | BUG(); | |
3064 | } | |
3065 | return 0; | |
3066 | } | |
3067 | #endif | |
3068 | ||
3069 | static DEFINE_SPINLOCK(balancing); | |
3070 | ||
3071 | /* | |
7835b98b CL |
3072 | * It checks each scheduling domain to see if it is due to be balanced, |
3073 | * and initiates a balancing operation if so. | |
3074 | * | |
3075 | * Balancing parameters are set up in arch_init_sched_domains. | |
3076 | */ | |
a9957449 | 3077 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3078 | { |
46cb4b7c SS |
3079 | int balance = 1; |
3080 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3081 | unsigned long interval; |
3082 | struct sched_domain *sd; | |
46cb4b7c | 3083 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3084 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3085 | int update_next_balance = 0; |
1da177e4 | 3086 | |
46cb4b7c | 3087 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3088 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3089 | continue; | |
3090 | ||
3091 | interval = sd->balance_interval; | |
d15bcfdb | 3092 | if (idle != CPU_IDLE) |
1da177e4 LT |
3093 | interval *= sd->busy_factor; |
3094 | ||
3095 | /* scale ms to jiffies */ | |
3096 | interval = msecs_to_jiffies(interval); | |
3097 | if (unlikely(!interval)) | |
3098 | interval = 1; | |
dd41f596 IM |
3099 | if (interval > HZ*NR_CPUS/10) |
3100 | interval = HZ*NR_CPUS/10; | |
3101 | ||
1da177e4 | 3102 | |
08c183f3 CL |
3103 | if (sd->flags & SD_SERIALIZE) { |
3104 | if (!spin_trylock(&balancing)) | |
3105 | goto out; | |
3106 | } | |
3107 | ||
c9819f45 | 3108 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3109 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3110 | /* |
3111 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3112 | * longer idle, or one of our SMT siblings is |
3113 | * not idle. | |
3114 | */ | |
d15bcfdb | 3115 | idle = CPU_NOT_IDLE; |
1da177e4 | 3116 | } |
1bd77f2d | 3117 | sd->last_balance = jiffies; |
1da177e4 | 3118 | } |
08c183f3 CL |
3119 | if (sd->flags & SD_SERIALIZE) |
3120 | spin_unlock(&balancing); | |
3121 | out: | |
f549da84 | 3122 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3123 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3124 | update_next_balance = 1; |
3125 | } | |
783609c6 SS |
3126 | |
3127 | /* | |
3128 | * Stop the load balance at this level. There is another | |
3129 | * CPU in our sched group which is doing load balancing more | |
3130 | * actively. | |
3131 | */ | |
3132 | if (!balance) | |
3133 | break; | |
1da177e4 | 3134 | } |
f549da84 SS |
3135 | |
3136 | /* | |
3137 | * next_balance will be updated only when there is a need. | |
3138 | * When the cpu is attached to null domain for ex, it will not be | |
3139 | * updated. | |
3140 | */ | |
3141 | if (likely(update_next_balance)) | |
3142 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3143 | } |
3144 | ||
3145 | /* | |
3146 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3147 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3148 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3149 | */ | |
3150 | static void run_rebalance_domains(struct softirq_action *h) | |
3151 | { | |
dd41f596 IM |
3152 | int this_cpu = smp_processor_id(); |
3153 | struct rq *this_rq = cpu_rq(this_cpu); | |
3154 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3155 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3156 | |
dd41f596 | 3157 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3158 | |
3159 | #ifdef CONFIG_NO_HZ | |
3160 | /* | |
3161 | * If this cpu is the owner for idle load balancing, then do the | |
3162 | * balancing on behalf of the other idle cpus whose ticks are | |
3163 | * stopped. | |
3164 | */ | |
dd41f596 IM |
3165 | if (this_rq->idle_at_tick && |
3166 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3167 | cpumask_t cpus = nohz.cpu_mask; |
3168 | struct rq *rq; | |
3169 | int balance_cpu; | |
3170 | ||
dd41f596 | 3171 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3172 | for_each_cpu_mask(balance_cpu, cpus) { |
3173 | /* | |
3174 | * If this cpu gets work to do, stop the load balancing | |
3175 | * work being done for other cpus. Next load | |
3176 | * balancing owner will pick it up. | |
3177 | */ | |
3178 | if (need_resched()) | |
3179 | break; | |
3180 | ||
de0cf899 | 3181 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3182 | |
3183 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3184 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3185 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3186 | } |
3187 | } | |
3188 | #endif | |
3189 | } | |
3190 | ||
3191 | /* | |
3192 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3193 | * | |
3194 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3195 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3196 | * if the whole system is idle. | |
3197 | */ | |
dd41f596 | 3198 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3199 | { |
46cb4b7c SS |
3200 | #ifdef CONFIG_NO_HZ |
3201 | /* | |
3202 | * If we were in the nohz mode recently and busy at the current | |
3203 | * scheduler tick, then check if we need to nominate new idle | |
3204 | * load balancer. | |
3205 | */ | |
3206 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3207 | rq->in_nohz_recently = 0; | |
3208 | ||
3209 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3210 | cpu_clear(cpu, nohz.cpu_mask); | |
3211 | atomic_set(&nohz.load_balancer, -1); | |
3212 | } | |
3213 | ||
3214 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3215 | /* | |
3216 | * simple selection for now: Nominate the | |
3217 | * first cpu in the nohz list to be the next | |
3218 | * ilb owner. | |
3219 | * | |
3220 | * TBD: Traverse the sched domains and nominate | |
3221 | * the nearest cpu in the nohz.cpu_mask. | |
3222 | */ | |
3223 | int ilb = first_cpu(nohz.cpu_mask); | |
3224 | ||
3225 | if (ilb != NR_CPUS) | |
3226 | resched_cpu(ilb); | |
3227 | } | |
3228 | } | |
3229 | ||
3230 | /* | |
3231 | * If this cpu is idle and doing idle load balancing for all the | |
3232 | * cpus with ticks stopped, is it time for that to stop? | |
3233 | */ | |
3234 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3235 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3236 | resched_cpu(cpu); | |
3237 | return; | |
3238 | } | |
3239 | ||
3240 | /* | |
3241 | * If this cpu is idle and the idle load balancing is done by | |
3242 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3243 | */ | |
3244 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3245 | cpu_isset(cpu, nohz.cpu_mask)) | |
3246 | return; | |
3247 | #endif | |
3248 | if (time_after_eq(jiffies, rq->next_balance)) | |
3249 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3250 | } |
dd41f596 IM |
3251 | |
3252 | #else /* CONFIG_SMP */ | |
3253 | ||
1da177e4 LT |
3254 | /* |
3255 | * on UP we do not need to balance between CPUs: | |
3256 | */ | |
70b97a7f | 3257 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3258 | { |
3259 | } | |
dd41f596 IM |
3260 | |
3261 | /* Avoid "used but not defined" warning on UP */ | |
3262 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3263 | unsigned long max_nr_move, unsigned long max_load_move, | |
3264 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3265 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3266 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3267 | { |
3268 | *load_moved = 0; | |
3269 | ||
3270 | return 0; | |
3271 | } | |
3272 | ||
1da177e4 LT |
3273 | #endif |
3274 | ||
1da177e4 LT |
3275 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3276 | ||
3277 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3278 | ||
3279 | /* | |
41b86e9c IM |
3280 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3281 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3282 | */ |
41b86e9c | 3283 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3284 | { |
1da177e4 | 3285 | unsigned long flags; |
41b86e9c IM |
3286 | u64 ns, delta_exec; |
3287 | struct rq *rq; | |
48f24c4d | 3288 | |
41b86e9c IM |
3289 | rq = task_rq_lock(p, &flags); |
3290 | ns = p->se.sum_exec_runtime; | |
3291 | if (rq->curr == p) { | |
a8e504d2 IM |
3292 | update_rq_clock(rq); |
3293 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3294 | if ((s64)delta_exec > 0) |
3295 | ns += delta_exec; | |
3296 | } | |
3297 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3298 | |
1da177e4 LT |
3299 | return ns; |
3300 | } | |
3301 | ||
1da177e4 LT |
3302 | /* |
3303 | * Account user cpu time to a process. | |
3304 | * @p: the process that the cpu time gets accounted to | |
3305 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3306 | * @cputime: the cpu time spent in user space since the last update | |
3307 | */ | |
3308 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3309 | { | |
3310 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3311 | cputime64_t tmp; | |
62d0df64 | 3312 | struct rq *rq = this_rq(); |
1da177e4 LT |
3313 | |
3314 | p->utime = cputime_add(p->utime, cputime); | |
3315 | ||
62d0df64 PM |
3316 | if (p != rq->idle) |
3317 | cpuacct_charge(p, cputime); | |
3318 | ||
1da177e4 LT |
3319 | /* Add user time to cpustat. */ |
3320 | tmp = cputime_to_cputime64(cputime); | |
3321 | if (TASK_NICE(p) > 0) | |
3322 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3323 | else | |
3324 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3325 | } | |
3326 | ||
94886b84 LV |
3327 | /* |
3328 | * Account guest cpu time to a process. | |
3329 | * @p: the process that the cpu time gets accounted to | |
3330 | * @cputime: the cpu time spent in virtual machine since the last update | |
3331 | */ | |
3332 | void account_guest_time(struct task_struct *p, cputime_t cputime) | |
3333 | { | |
3334 | cputime64_t tmp; | |
3335 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3336 | ||
3337 | tmp = cputime_to_cputime64(cputime); | |
3338 | ||
3339 | p->utime = cputime_add(p->utime, cputime); | |
3340 | p->gtime = cputime_add(p->gtime, cputime); | |
3341 | ||
3342 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3343 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3344 | } | |
3345 | ||
c66f08be MN |
3346 | /* |
3347 | * Account scaled user cpu time to a process. | |
3348 | * @p: the process that the cpu time gets accounted to | |
3349 | * @cputime: the cpu time spent in user space since the last update | |
3350 | */ | |
3351 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3352 | { | |
3353 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3354 | } | |
3355 | ||
1da177e4 LT |
3356 | /* |
3357 | * Account system cpu time to a process. | |
3358 | * @p: the process that the cpu time gets accounted to | |
3359 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3360 | * @cputime: the cpu time spent in kernel space since the last update | |
3361 | */ | |
3362 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3363 | cputime_t cputime) | |
3364 | { | |
3365 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3366 | struct rq *rq = this_rq(); |
1da177e4 LT |
3367 | cputime64_t tmp; |
3368 | ||
94886b84 LV |
3369 | if (p->flags & PF_VCPU) { |
3370 | account_guest_time(p, cputime); | |
3371 | p->flags &= ~PF_VCPU; | |
3372 | return; | |
3373 | } | |
3374 | ||
1da177e4 LT |
3375 | p->stime = cputime_add(p->stime, cputime); |
3376 | ||
3377 | /* Add system time to cpustat. */ | |
3378 | tmp = cputime_to_cputime64(cputime); | |
3379 | if (hardirq_count() - hardirq_offset) | |
3380 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3381 | else if (softirq_count()) | |
3382 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
62d0df64 | 3383 | else if (p != rq->idle) { |
1da177e4 | 3384 | cpustat->system = cputime64_add(cpustat->system, tmp); |
62d0df64 PM |
3385 | cpuacct_charge(p, cputime); |
3386 | } else if (atomic_read(&rq->nr_iowait) > 0) | |
1da177e4 LT |
3387 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3388 | else | |
3389 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3390 | /* Account for system time used */ | |
3391 | acct_update_integrals(p); | |
1da177e4 LT |
3392 | } |
3393 | ||
c66f08be MN |
3394 | /* |
3395 | * Account scaled system cpu time to a process. | |
3396 | * @p: the process that the cpu time gets accounted to | |
3397 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3398 | * @cputime: the cpu time spent in kernel space since the last update | |
3399 | */ | |
3400 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3401 | { | |
3402 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3403 | } | |
3404 | ||
1da177e4 LT |
3405 | /* |
3406 | * Account for involuntary wait time. | |
3407 | * @p: the process from which the cpu time has been stolen | |
3408 | * @steal: the cpu time spent in involuntary wait | |
3409 | */ | |
3410 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3411 | { | |
3412 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3413 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3414 | struct rq *rq = this_rq(); |
1da177e4 LT |
3415 | |
3416 | if (p == rq->idle) { | |
3417 | p->stime = cputime_add(p->stime, steal); | |
3418 | if (atomic_read(&rq->nr_iowait) > 0) | |
3419 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3420 | else | |
3421 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
62d0df64 | 3422 | } else { |
1da177e4 | 3423 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
62d0df64 PM |
3424 | cpuacct_charge(p, -tmp); |
3425 | } | |
1da177e4 LT |
3426 | } |
3427 | ||
7835b98b CL |
3428 | /* |
3429 | * This function gets called by the timer code, with HZ frequency. | |
3430 | * We call it with interrupts disabled. | |
3431 | * | |
3432 | * It also gets called by the fork code, when changing the parent's | |
3433 | * timeslices. | |
3434 | */ | |
3435 | void scheduler_tick(void) | |
3436 | { | |
7835b98b CL |
3437 | int cpu = smp_processor_id(); |
3438 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3439 | struct task_struct *curr = rq->curr; |
529c7726 | 3440 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3441 | |
3442 | spin_lock(&rq->lock); | |
546fe3c9 | 3443 | __update_rq_clock(rq); |
529c7726 IM |
3444 | /* |
3445 | * Let rq->clock advance by at least TICK_NSEC: | |
3446 | */ | |
3447 | if (unlikely(rq->clock < next_tick)) | |
3448 | rq->clock = next_tick; | |
3449 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3450 | update_cpu_load(rq); |
dd41f596 IM |
3451 | if (curr != rq->idle) /* FIXME: needed? */ |
3452 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3453 | spin_unlock(&rq->lock); |
7835b98b | 3454 | |
e418e1c2 | 3455 | #ifdef CONFIG_SMP |
dd41f596 IM |
3456 | rq->idle_at_tick = idle_cpu(cpu); |
3457 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3458 | #endif |
1da177e4 LT |
3459 | } |
3460 | ||
1da177e4 LT |
3461 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3462 | ||
3463 | void fastcall add_preempt_count(int val) | |
3464 | { | |
3465 | /* | |
3466 | * Underflow? | |
3467 | */ | |
9a11b49a IM |
3468 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3469 | return; | |
1da177e4 LT |
3470 | preempt_count() += val; |
3471 | /* | |
3472 | * Spinlock count overflowing soon? | |
3473 | */ | |
33859f7f MOS |
3474 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3475 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3476 | } |
3477 | EXPORT_SYMBOL(add_preempt_count); | |
3478 | ||
3479 | void fastcall sub_preempt_count(int val) | |
3480 | { | |
3481 | /* | |
3482 | * Underflow? | |
3483 | */ | |
9a11b49a IM |
3484 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3485 | return; | |
1da177e4 LT |
3486 | /* |
3487 | * Is the spinlock portion underflowing? | |
3488 | */ | |
9a11b49a IM |
3489 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3490 | !(preempt_count() & PREEMPT_MASK))) | |
3491 | return; | |
3492 | ||
1da177e4 LT |
3493 | preempt_count() -= val; |
3494 | } | |
3495 | EXPORT_SYMBOL(sub_preempt_count); | |
3496 | ||
3497 | #endif | |
3498 | ||
3499 | /* | |
dd41f596 | 3500 | * Print scheduling while atomic bug: |
1da177e4 | 3501 | */ |
dd41f596 | 3502 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3503 | { |
dd41f596 IM |
3504 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3505 | prev->comm, preempt_count(), prev->pid); | |
3506 | debug_show_held_locks(prev); | |
3507 | if (irqs_disabled()) | |
3508 | print_irqtrace_events(prev); | |
3509 | dump_stack(); | |
3510 | } | |
1da177e4 | 3511 | |
dd41f596 IM |
3512 | /* |
3513 | * Various schedule()-time debugging checks and statistics: | |
3514 | */ | |
3515 | static inline void schedule_debug(struct task_struct *prev) | |
3516 | { | |
1da177e4 LT |
3517 | /* |
3518 | * Test if we are atomic. Since do_exit() needs to call into | |
3519 | * schedule() atomically, we ignore that path for now. | |
3520 | * Otherwise, whine if we are scheduling when we should not be. | |
3521 | */ | |
dd41f596 IM |
3522 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3523 | __schedule_bug(prev); | |
3524 | ||
1da177e4 LT |
3525 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3526 | ||
2d72376b | 3527 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3528 | #ifdef CONFIG_SCHEDSTATS |
3529 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3530 | schedstat_inc(this_rq(), bkl_count); |
3531 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3532 | } |
3533 | #endif | |
dd41f596 IM |
3534 | } |
3535 | ||
3536 | /* | |
3537 | * Pick up the highest-prio task: | |
3538 | */ | |
3539 | static inline struct task_struct * | |
ff95f3df | 3540 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3541 | { |
5522d5d5 | 3542 | const struct sched_class *class; |
dd41f596 | 3543 | struct task_struct *p; |
1da177e4 LT |
3544 | |
3545 | /* | |
dd41f596 IM |
3546 | * Optimization: we know that if all tasks are in |
3547 | * the fair class we can call that function directly: | |
1da177e4 | 3548 | */ |
dd41f596 | 3549 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3550 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3551 | if (likely(p)) |
3552 | return p; | |
1da177e4 LT |
3553 | } |
3554 | ||
dd41f596 IM |
3555 | class = sched_class_highest; |
3556 | for ( ; ; ) { | |
fb8d4724 | 3557 | p = class->pick_next_task(rq); |
dd41f596 IM |
3558 | if (p) |
3559 | return p; | |
3560 | /* | |
3561 | * Will never be NULL as the idle class always | |
3562 | * returns a non-NULL p: | |
3563 | */ | |
3564 | class = class->next; | |
3565 | } | |
3566 | } | |
1da177e4 | 3567 | |
dd41f596 IM |
3568 | /* |
3569 | * schedule() is the main scheduler function. | |
3570 | */ | |
3571 | asmlinkage void __sched schedule(void) | |
3572 | { | |
3573 | struct task_struct *prev, *next; | |
3574 | long *switch_count; | |
3575 | struct rq *rq; | |
dd41f596 IM |
3576 | int cpu; |
3577 | ||
3578 | need_resched: | |
3579 | preempt_disable(); | |
3580 | cpu = smp_processor_id(); | |
3581 | rq = cpu_rq(cpu); | |
3582 | rcu_qsctr_inc(cpu); | |
3583 | prev = rq->curr; | |
3584 | switch_count = &prev->nivcsw; | |
3585 | ||
3586 | release_kernel_lock(prev); | |
3587 | need_resched_nonpreemptible: | |
3588 | ||
3589 | schedule_debug(prev); | |
1da177e4 | 3590 | |
1e819950 IM |
3591 | /* |
3592 | * Do the rq-clock update outside the rq lock: | |
3593 | */ | |
3594 | local_irq_disable(); | |
c1b3da3e | 3595 | __update_rq_clock(rq); |
1e819950 IM |
3596 | spin_lock(&rq->lock); |
3597 | clear_tsk_need_resched(prev); | |
1da177e4 | 3598 | |
1da177e4 | 3599 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3600 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3601 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3602 | prev->state = TASK_RUNNING; |
dd41f596 | 3603 | } else { |
2e1cb74a | 3604 | deactivate_task(rq, prev, 1); |
1da177e4 | 3605 | } |
dd41f596 | 3606 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3607 | } |
3608 | ||
dd41f596 | 3609 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3610 | idle_balance(cpu, rq); |
1da177e4 | 3611 | |
31ee529c | 3612 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3613 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3614 | |
3615 | sched_info_switch(prev, next); | |
dd41f596 | 3616 | |
1da177e4 | 3617 | if (likely(prev != next)) { |
1da177e4 LT |
3618 | rq->nr_switches++; |
3619 | rq->curr = next; | |
3620 | ++*switch_count; | |
3621 | ||
dd41f596 | 3622 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3623 | } else |
3624 | spin_unlock_irq(&rq->lock); | |
3625 | ||
dd41f596 IM |
3626 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3627 | cpu = smp_processor_id(); | |
3628 | rq = cpu_rq(cpu); | |
1da177e4 | 3629 | goto need_resched_nonpreemptible; |
dd41f596 | 3630 | } |
1da177e4 LT |
3631 | preempt_enable_no_resched(); |
3632 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3633 | goto need_resched; | |
3634 | } | |
1da177e4 LT |
3635 | EXPORT_SYMBOL(schedule); |
3636 | ||
3637 | #ifdef CONFIG_PREEMPT | |
3638 | /* | |
2ed6e34f | 3639 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3640 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3641 | * occur there and call schedule directly. | |
3642 | */ | |
3643 | asmlinkage void __sched preempt_schedule(void) | |
3644 | { | |
3645 | struct thread_info *ti = current_thread_info(); | |
3646 | #ifdef CONFIG_PREEMPT_BKL | |
3647 | struct task_struct *task = current; | |
3648 | int saved_lock_depth; | |
3649 | #endif | |
3650 | /* | |
3651 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3652 | * we do not want to preempt the current task. Just return.. | |
3653 | */ | |
beed33a8 | 3654 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3655 | return; |
3656 | ||
3a5c359a AK |
3657 | do { |
3658 | add_preempt_count(PREEMPT_ACTIVE); | |
3659 | ||
3660 | /* | |
3661 | * We keep the big kernel semaphore locked, but we | |
3662 | * clear ->lock_depth so that schedule() doesnt | |
3663 | * auto-release the semaphore: | |
3664 | */ | |
1da177e4 | 3665 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3666 | saved_lock_depth = task->lock_depth; |
3667 | task->lock_depth = -1; | |
1da177e4 | 3668 | #endif |
3a5c359a | 3669 | schedule(); |
1da177e4 | 3670 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3671 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3672 | #endif |
3a5c359a | 3673 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3674 | |
3a5c359a AK |
3675 | /* |
3676 | * Check again in case we missed a preemption opportunity | |
3677 | * between schedule and now. | |
3678 | */ | |
3679 | barrier(); | |
3680 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3681 | } |
1da177e4 LT |
3682 | EXPORT_SYMBOL(preempt_schedule); |
3683 | ||
3684 | /* | |
2ed6e34f | 3685 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3686 | * off of irq context. |
3687 | * Note, that this is called and return with irqs disabled. This will | |
3688 | * protect us against recursive calling from irq. | |
3689 | */ | |
3690 | asmlinkage void __sched preempt_schedule_irq(void) | |
3691 | { | |
3692 | struct thread_info *ti = current_thread_info(); | |
3693 | #ifdef CONFIG_PREEMPT_BKL | |
3694 | struct task_struct *task = current; | |
3695 | int saved_lock_depth; | |
3696 | #endif | |
2ed6e34f | 3697 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3698 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3699 | ||
3a5c359a AK |
3700 | do { |
3701 | add_preempt_count(PREEMPT_ACTIVE); | |
3702 | ||
3703 | /* | |
3704 | * We keep the big kernel semaphore locked, but we | |
3705 | * clear ->lock_depth so that schedule() doesnt | |
3706 | * auto-release the semaphore: | |
3707 | */ | |
1da177e4 | 3708 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3709 | saved_lock_depth = task->lock_depth; |
3710 | task->lock_depth = -1; | |
1da177e4 | 3711 | #endif |
3a5c359a AK |
3712 | local_irq_enable(); |
3713 | schedule(); | |
3714 | local_irq_disable(); | |
1da177e4 | 3715 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3716 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3717 | #endif |
3a5c359a | 3718 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3719 | |
3a5c359a AK |
3720 | /* |
3721 | * Check again in case we missed a preemption opportunity | |
3722 | * between schedule and now. | |
3723 | */ | |
3724 | barrier(); | |
3725 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3726 | } |
3727 | ||
3728 | #endif /* CONFIG_PREEMPT */ | |
3729 | ||
95cdf3b7 IM |
3730 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3731 | void *key) | |
1da177e4 | 3732 | { |
48f24c4d | 3733 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3734 | } |
1da177e4 LT |
3735 | EXPORT_SYMBOL(default_wake_function); |
3736 | ||
3737 | /* | |
3738 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3739 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3740 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3741 | * | |
3742 | * There are circumstances in which we can try to wake a task which has already | |
3743 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3744 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3745 | */ | |
3746 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3747 | int nr_exclusive, int sync, void *key) | |
3748 | { | |
2e45874c | 3749 | wait_queue_t *curr, *next; |
1da177e4 | 3750 | |
2e45874c | 3751 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3752 | unsigned flags = curr->flags; |
3753 | ||
1da177e4 | 3754 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3755 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3756 | break; |
3757 | } | |
3758 | } | |
3759 | ||
3760 | /** | |
3761 | * __wake_up - wake up threads blocked on a waitqueue. | |
3762 | * @q: the waitqueue | |
3763 | * @mode: which threads | |
3764 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3765 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3766 | */ |
3767 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3768 | int nr_exclusive, void *key) |
1da177e4 LT |
3769 | { |
3770 | unsigned long flags; | |
3771 | ||
3772 | spin_lock_irqsave(&q->lock, flags); | |
3773 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3774 | spin_unlock_irqrestore(&q->lock, flags); | |
3775 | } | |
1da177e4 LT |
3776 | EXPORT_SYMBOL(__wake_up); |
3777 | ||
3778 | /* | |
3779 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3780 | */ | |
3781 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3782 | { | |
3783 | __wake_up_common(q, mode, 1, 0, NULL); | |
3784 | } | |
3785 | ||
3786 | /** | |
67be2dd1 | 3787 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3788 | * @q: the waitqueue |
3789 | * @mode: which threads | |
3790 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3791 | * | |
3792 | * The sync wakeup differs that the waker knows that it will schedule | |
3793 | * away soon, so while the target thread will be woken up, it will not | |
3794 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3795 | * with each other. This can prevent needless bouncing between CPUs. | |
3796 | * | |
3797 | * On UP it can prevent extra preemption. | |
3798 | */ | |
95cdf3b7 IM |
3799 | void fastcall |
3800 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3801 | { |
3802 | unsigned long flags; | |
3803 | int sync = 1; | |
3804 | ||
3805 | if (unlikely(!q)) | |
3806 | return; | |
3807 | ||
3808 | if (unlikely(!nr_exclusive)) | |
3809 | sync = 0; | |
3810 | ||
3811 | spin_lock_irqsave(&q->lock, flags); | |
3812 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3813 | spin_unlock_irqrestore(&q->lock, flags); | |
3814 | } | |
3815 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3816 | ||
3817 | void fastcall complete(struct completion *x) | |
3818 | { | |
3819 | unsigned long flags; | |
3820 | ||
3821 | spin_lock_irqsave(&x->wait.lock, flags); | |
3822 | x->done++; | |
3823 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3824 | 1, 0, NULL); | |
3825 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3826 | } | |
3827 | EXPORT_SYMBOL(complete); | |
3828 | ||
3829 | void fastcall complete_all(struct completion *x) | |
3830 | { | |
3831 | unsigned long flags; | |
3832 | ||
3833 | spin_lock_irqsave(&x->wait.lock, flags); | |
3834 | x->done += UINT_MAX/2; | |
3835 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3836 | 0, 0, NULL); | |
3837 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3838 | } | |
3839 | EXPORT_SYMBOL(complete_all); | |
3840 | ||
8cbbe86d AK |
3841 | static inline long __sched |
3842 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3843 | { |
1da177e4 LT |
3844 | if (!x->done) { |
3845 | DECLARE_WAITQUEUE(wait, current); | |
3846 | ||
3847 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3848 | __add_wait_queue_tail(&x->wait, &wait); | |
3849 | do { | |
8cbbe86d AK |
3850 | if (state == TASK_INTERRUPTIBLE && |
3851 | signal_pending(current)) { | |
3852 | __remove_wait_queue(&x->wait, &wait); | |
3853 | return -ERESTARTSYS; | |
3854 | } | |
3855 | __set_current_state(state); | |
1da177e4 LT |
3856 | spin_unlock_irq(&x->wait.lock); |
3857 | timeout = schedule_timeout(timeout); | |
3858 | spin_lock_irq(&x->wait.lock); | |
3859 | if (!timeout) { | |
3860 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3861 | return timeout; |
1da177e4 LT |
3862 | } |
3863 | } while (!x->done); | |
3864 | __remove_wait_queue(&x->wait, &wait); | |
3865 | } | |
3866 | x->done--; | |
1da177e4 LT |
3867 | return timeout; |
3868 | } | |
1da177e4 | 3869 | |
8cbbe86d AK |
3870 | static long __sched |
3871 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3872 | { |
1da177e4 LT |
3873 | might_sleep(); |
3874 | ||
3875 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3876 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3877 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3878 | return timeout; |
3879 | } | |
1da177e4 | 3880 | |
8cbbe86d AK |
3881 | void fastcall __sched wait_for_completion(struct completion *x) |
3882 | { | |
3883 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3884 | } |
8cbbe86d | 3885 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 LT |
3886 | |
3887 | unsigned long fastcall __sched | |
8cbbe86d | 3888 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3889 | { |
8cbbe86d | 3890 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3891 | } |
8cbbe86d | 3892 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3893 | |
8cbbe86d | 3894 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3895 | { |
51e97990 AK |
3896 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3897 | if (t == -ERESTARTSYS) | |
3898 | return t; | |
3899 | return 0; | |
0fec171c | 3900 | } |
8cbbe86d | 3901 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3902 | |
8cbbe86d AK |
3903 | unsigned long fastcall __sched |
3904 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3905 | unsigned long timeout) | |
0fec171c | 3906 | { |
8cbbe86d | 3907 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3908 | } |
8cbbe86d | 3909 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3910 | |
8cbbe86d AK |
3911 | static long __sched |
3912 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3913 | { |
0fec171c IM |
3914 | unsigned long flags; |
3915 | wait_queue_t wait; | |
3916 | ||
3917 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3918 | |
8cbbe86d | 3919 | __set_current_state(state); |
1da177e4 | 3920 | |
8cbbe86d AK |
3921 | spin_lock_irqsave(&q->lock, flags); |
3922 | __add_wait_queue(q, &wait); | |
3923 | spin_unlock(&q->lock); | |
3924 | timeout = schedule_timeout(timeout); | |
3925 | spin_lock_irq(&q->lock); | |
3926 | __remove_wait_queue(q, &wait); | |
3927 | spin_unlock_irqrestore(&q->lock, flags); | |
3928 | ||
3929 | return timeout; | |
3930 | } | |
3931 | ||
3932 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3933 | { | |
3934 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3935 | } |
1da177e4 LT |
3936 | EXPORT_SYMBOL(interruptible_sleep_on); |
3937 | ||
0fec171c | 3938 | long __sched |
95cdf3b7 | 3939 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3940 | { |
8cbbe86d | 3941 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3942 | } |
1da177e4 LT |
3943 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3944 | ||
0fec171c | 3945 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3946 | { |
8cbbe86d | 3947 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3948 | } |
1da177e4 LT |
3949 | EXPORT_SYMBOL(sleep_on); |
3950 | ||
0fec171c | 3951 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3952 | { |
8cbbe86d | 3953 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3954 | } |
1da177e4 LT |
3955 | EXPORT_SYMBOL(sleep_on_timeout); |
3956 | ||
b29739f9 IM |
3957 | #ifdef CONFIG_RT_MUTEXES |
3958 | ||
3959 | /* | |
3960 | * rt_mutex_setprio - set the current priority of a task | |
3961 | * @p: task | |
3962 | * @prio: prio value (kernel-internal form) | |
3963 | * | |
3964 | * This function changes the 'effective' priority of a task. It does | |
3965 | * not touch ->normal_prio like __setscheduler(). | |
3966 | * | |
3967 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3968 | */ | |
36c8b586 | 3969 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3970 | { |
3971 | unsigned long flags; | |
83b699ed | 3972 | int oldprio, on_rq, running; |
70b97a7f | 3973 | struct rq *rq; |
b29739f9 IM |
3974 | |
3975 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3976 | ||
3977 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3978 | update_rq_clock(rq); |
b29739f9 | 3979 | |
d5f9f942 | 3980 | oldprio = p->prio; |
dd41f596 | 3981 | on_rq = p->se.on_rq; |
83b699ed SV |
3982 | running = task_running(rq, p); |
3983 | if (on_rq) { | |
69be72c1 | 3984 | dequeue_task(rq, p, 0); |
83b699ed SV |
3985 | if (running) |
3986 | p->sched_class->put_prev_task(rq, p); | |
3987 | } | |
dd41f596 IM |
3988 | |
3989 | if (rt_prio(prio)) | |
3990 | p->sched_class = &rt_sched_class; | |
3991 | else | |
3992 | p->sched_class = &fair_sched_class; | |
3993 | ||
b29739f9 IM |
3994 | p->prio = prio; |
3995 | ||
dd41f596 | 3996 | if (on_rq) { |
83b699ed SV |
3997 | if (running) |
3998 | p->sched_class->set_curr_task(rq); | |
8159f87e | 3999 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4000 | /* |
4001 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4002 | * our priority decreased, or if we are not currently running on |
4003 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4004 | */ |
83b699ed | 4005 | if (running) { |
d5f9f942 AM |
4006 | if (p->prio > oldprio) |
4007 | resched_task(rq->curr); | |
dd41f596 IM |
4008 | } else { |
4009 | check_preempt_curr(rq, p); | |
4010 | } | |
b29739f9 IM |
4011 | } |
4012 | task_rq_unlock(rq, &flags); | |
4013 | } | |
4014 | ||
4015 | #endif | |
4016 | ||
36c8b586 | 4017 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4018 | { |
dd41f596 | 4019 | int old_prio, delta, on_rq; |
1da177e4 | 4020 | unsigned long flags; |
70b97a7f | 4021 | struct rq *rq; |
1da177e4 LT |
4022 | |
4023 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4024 | return; | |
4025 | /* | |
4026 | * We have to be careful, if called from sys_setpriority(), | |
4027 | * the task might be in the middle of scheduling on another CPU. | |
4028 | */ | |
4029 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4030 | update_rq_clock(rq); |
1da177e4 LT |
4031 | /* |
4032 | * The RT priorities are set via sched_setscheduler(), but we still | |
4033 | * allow the 'normal' nice value to be set - but as expected | |
4034 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4035 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4036 | */ |
e05606d3 | 4037 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4038 | p->static_prio = NICE_TO_PRIO(nice); |
4039 | goto out_unlock; | |
4040 | } | |
dd41f596 IM |
4041 | on_rq = p->se.on_rq; |
4042 | if (on_rq) { | |
69be72c1 | 4043 | dequeue_task(rq, p, 0); |
79b5dddf | 4044 | dec_load(rq, p); |
2dd73a4f | 4045 | } |
1da177e4 | 4046 | |
1da177e4 | 4047 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4048 | set_load_weight(p); |
b29739f9 IM |
4049 | old_prio = p->prio; |
4050 | p->prio = effective_prio(p); | |
4051 | delta = p->prio - old_prio; | |
1da177e4 | 4052 | |
dd41f596 | 4053 | if (on_rq) { |
8159f87e | 4054 | enqueue_task(rq, p, 0); |
29b4b623 | 4055 | inc_load(rq, p); |
1da177e4 | 4056 | /* |
d5f9f942 AM |
4057 | * If the task increased its priority or is running and |
4058 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4059 | */ |
d5f9f942 | 4060 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4061 | resched_task(rq->curr); |
4062 | } | |
4063 | out_unlock: | |
4064 | task_rq_unlock(rq, &flags); | |
4065 | } | |
1da177e4 LT |
4066 | EXPORT_SYMBOL(set_user_nice); |
4067 | ||
e43379f1 MM |
4068 | /* |
4069 | * can_nice - check if a task can reduce its nice value | |
4070 | * @p: task | |
4071 | * @nice: nice value | |
4072 | */ | |
36c8b586 | 4073 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4074 | { |
024f4747 MM |
4075 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4076 | int nice_rlim = 20 - nice; | |
48f24c4d | 4077 | |
e43379f1 MM |
4078 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4079 | capable(CAP_SYS_NICE)); | |
4080 | } | |
4081 | ||
1da177e4 LT |
4082 | #ifdef __ARCH_WANT_SYS_NICE |
4083 | ||
4084 | /* | |
4085 | * sys_nice - change the priority of the current process. | |
4086 | * @increment: priority increment | |
4087 | * | |
4088 | * sys_setpriority is a more generic, but much slower function that | |
4089 | * does similar things. | |
4090 | */ | |
4091 | asmlinkage long sys_nice(int increment) | |
4092 | { | |
48f24c4d | 4093 | long nice, retval; |
1da177e4 LT |
4094 | |
4095 | /* | |
4096 | * Setpriority might change our priority at the same moment. | |
4097 | * We don't have to worry. Conceptually one call occurs first | |
4098 | * and we have a single winner. | |
4099 | */ | |
e43379f1 MM |
4100 | if (increment < -40) |
4101 | increment = -40; | |
1da177e4 LT |
4102 | if (increment > 40) |
4103 | increment = 40; | |
4104 | ||
4105 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4106 | if (nice < -20) | |
4107 | nice = -20; | |
4108 | if (nice > 19) | |
4109 | nice = 19; | |
4110 | ||
e43379f1 MM |
4111 | if (increment < 0 && !can_nice(current, nice)) |
4112 | return -EPERM; | |
4113 | ||
1da177e4 LT |
4114 | retval = security_task_setnice(current, nice); |
4115 | if (retval) | |
4116 | return retval; | |
4117 | ||
4118 | set_user_nice(current, nice); | |
4119 | return 0; | |
4120 | } | |
4121 | ||
4122 | #endif | |
4123 | ||
4124 | /** | |
4125 | * task_prio - return the priority value of a given task. | |
4126 | * @p: the task in question. | |
4127 | * | |
4128 | * This is the priority value as seen by users in /proc. | |
4129 | * RT tasks are offset by -200. Normal tasks are centered | |
4130 | * around 0, value goes from -16 to +15. | |
4131 | */ | |
36c8b586 | 4132 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4133 | { |
4134 | return p->prio - MAX_RT_PRIO; | |
4135 | } | |
4136 | ||
4137 | /** | |
4138 | * task_nice - return the nice value of a given task. | |
4139 | * @p: the task in question. | |
4140 | */ | |
36c8b586 | 4141 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4142 | { |
4143 | return TASK_NICE(p); | |
4144 | } | |
1da177e4 | 4145 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4146 | |
4147 | /** | |
4148 | * idle_cpu - is a given cpu idle currently? | |
4149 | * @cpu: the processor in question. | |
4150 | */ | |
4151 | int idle_cpu(int cpu) | |
4152 | { | |
4153 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4154 | } | |
4155 | ||
1da177e4 LT |
4156 | /** |
4157 | * idle_task - return the idle task for a given cpu. | |
4158 | * @cpu: the processor in question. | |
4159 | */ | |
36c8b586 | 4160 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4161 | { |
4162 | return cpu_rq(cpu)->idle; | |
4163 | } | |
4164 | ||
4165 | /** | |
4166 | * find_process_by_pid - find a process with a matching PID value. | |
4167 | * @pid: the pid in question. | |
4168 | */ | |
a9957449 | 4169 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4170 | { |
228ebcbe | 4171 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4172 | } |
4173 | ||
4174 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4175 | static void |
4176 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4177 | { |
dd41f596 | 4178 | BUG_ON(p->se.on_rq); |
48f24c4d | 4179 | |
1da177e4 | 4180 | p->policy = policy; |
dd41f596 IM |
4181 | switch (p->policy) { |
4182 | case SCHED_NORMAL: | |
4183 | case SCHED_BATCH: | |
4184 | case SCHED_IDLE: | |
4185 | p->sched_class = &fair_sched_class; | |
4186 | break; | |
4187 | case SCHED_FIFO: | |
4188 | case SCHED_RR: | |
4189 | p->sched_class = &rt_sched_class; | |
4190 | break; | |
4191 | } | |
4192 | ||
1da177e4 | 4193 | p->rt_priority = prio; |
b29739f9 IM |
4194 | p->normal_prio = normal_prio(p); |
4195 | /* we are holding p->pi_lock already */ | |
4196 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4197 | set_load_weight(p); |
1da177e4 LT |
4198 | } |
4199 | ||
4200 | /** | |
72fd4a35 | 4201 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4202 | * @p: the task in question. |
4203 | * @policy: new policy. | |
4204 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4205 | * |
72fd4a35 | 4206 | * NOTE that the task may be already dead. |
1da177e4 | 4207 | */ |
95cdf3b7 IM |
4208 | int sched_setscheduler(struct task_struct *p, int policy, |
4209 | struct sched_param *param) | |
1da177e4 | 4210 | { |
83b699ed | 4211 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4212 | unsigned long flags; |
70b97a7f | 4213 | struct rq *rq; |
1da177e4 | 4214 | |
66e5393a SR |
4215 | /* may grab non-irq protected spin_locks */ |
4216 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4217 | recheck: |
4218 | /* double check policy once rq lock held */ | |
4219 | if (policy < 0) | |
4220 | policy = oldpolicy = p->policy; | |
4221 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4222 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4223 | policy != SCHED_IDLE) | |
b0a9499c | 4224 | return -EINVAL; |
1da177e4 LT |
4225 | /* |
4226 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4227 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4228 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4229 | */ |
4230 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4231 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4232 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4233 | return -EINVAL; |
e05606d3 | 4234 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4235 | return -EINVAL; |
4236 | ||
37e4ab3f OC |
4237 | /* |
4238 | * Allow unprivileged RT tasks to decrease priority: | |
4239 | */ | |
4240 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4241 | if (rt_policy(policy)) { |
8dc3e909 | 4242 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4243 | |
4244 | if (!lock_task_sighand(p, &flags)) | |
4245 | return -ESRCH; | |
4246 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4247 | unlock_task_sighand(p, &flags); | |
4248 | ||
4249 | /* can't set/change the rt policy */ | |
4250 | if (policy != p->policy && !rlim_rtprio) | |
4251 | return -EPERM; | |
4252 | ||
4253 | /* can't increase priority */ | |
4254 | if (param->sched_priority > p->rt_priority && | |
4255 | param->sched_priority > rlim_rtprio) | |
4256 | return -EPERM; | |
4257 | } | |
dd41f596 IM |
4258 | /* |
4259 | * Like positive nice levels, dont allow tasks to | |
4260 | * move out of SCHED_IDLE either: | |
4261 | */ | |
4262 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4263 | return -EPERM; | |
5fe1d75f | 4264 | |
37e4ab3f OC |
4265 | /* can't change other user's priorities */ |
4266 | if ((current->euid != p->euid) && | |
4267 | (current->euid != p->uid)) | |
4268 | return -EPERM; | |
4269 | } | |
1da177e4 LT |
4270 | |
4271 | retval = security_task_setscheduler(p, policy, param); | |
4272 | if (retval) | |
4273 | return retval; | |
b29739f9 IM |
4274 | /* |
4275 | * make sure no PI-waiters arrive (or leave) while we are | |
4276 | * changing the priority of the task: | |
4277 | */ | |
4278 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4279 | /* |
4280 | * To be able to change p->policy safely, the apropriate | |
4281 | * runqueue lock must be held. | |
4282 | */ | |
b29739f9 | 4283 | rq = __task_rq_lock(p); |
1da177e4 LT |
4284 | /* recheck policy now with rq lock held */ |
4285 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4286 | policy = oldpolicy = -1; | |
b29739f9 IM |
4287 | __task_rq_unlock(rq); |
4288 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4289 | goto recheck; |
4290 | } | |
2daa3577 | 4291 | update_rq_clock(rq); |
dd41f596 | 4292 | on_rq = p->se.on_rq; |
83b699ed SV |
4293 | running = task_running(rq, p); |
4294 | if (on_rq) { | |
2e1cb74a | 4295 | deactivate_task(rq, p, 0); |
83b699ed SV |
4296 | if (running) |
4297 | p->sched_class->put_prev_task(rq, p); | |
4298 | } | |
f6b53205 | 4299 | |
1da177e4 | 4300 | oldprio = p->prio; |
dd41f596 | 4301 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4302 | |
dd41f596 | 4303 | if (on_rq) { |
83b699ed SV |
4304 | if (running) |
4305 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4306 | activate_task(rq, p, 0); |
1da177e4 LT |
4307 | /* |
4308 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4309 | * our priority decreased, or if we are not currently running on |
4310 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4311 | */ |
83b699ed | 4312 | if (running) { |
d5f9f942 AM |
4313 | if (p->prio > oldprio) |
4314 | resched_task(rq->curr); | |
dd41f596 IM |
4315 | } else { |
4316 | check_preempt_curr(rq, p); | |
4317 | } | |
1da177e4 | 4318 | } |
b29739f9 IM |
4319 | __task_rq_unlock(rq); |
4320 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4321 | ||
95e02ca9 TG |
4322 | rt_mutex_adjust_pi(p); |
4323 | ||
1da177e4 LT |
4324 | return 0; |
4325 | } | |
4326 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4327 | ||
95cdf3b7 IM |
4328 | static int |
4329 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4330 | { |
1da177e4 LT |
4331 | struct sched_param lparam; |
4332 | struct task_struct *p; | |
36c8b586 | 4333 | int retval; |
1da177e4 LT |
4334 | |
4335 | if (!param || pid < 0) | |
4336 | return -EINVAL; | |
4337 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4338 | return -EFAULT; | |
5fe1d75f ON |
4339 | |
4340 | rcu_read_lock(); | |
4341 | retval = -ESRCH; | |
1da177e4 | 4342 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4343 | if (p != NULL) |
4344 | retval = sched_setscheduler(p, policy, &lparam); | |
4345 | rcu_read_unlock(); | |
36c8b586 | 4346 | |
1da177e4 LT |
4347 | return retval; |
4348 | } | |
4349 | ||
4350 | /** | |
4351 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4352 | * @pid: the pid in question. | |
4353 | * @policy: new policy. | |
4354 | * @param: structure containing the new RT priority. | |
4355 | */ | |
4356 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4357 | struct sched_param __user *param) | |
4358 | { | |
c21761f1 JB |
4359 | /* negative values for policy are not valid */ |
4360 | if (policy < 0) | |
4361 | return -EINVAL; | |
4362 | ||
1da177e4 LT |
4363 | return do_sched_setscheduler(pid, policy, param); |
4364 | } | |
4365 | ||
4366 | /** | |
4367 | * sys_sched_setparam - set/change the RT priority of a thread | |
4368 | * @pid: the pid in question. | |
4369 | * @param: structure containing the new RT priority. | |
4370 | */ | |
4371 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4372 | { | |
4373 | return do_sched_setscheduler(pid, -1, param); | |
4374 | } | |
4375 | ||
4376 | /** | |
4377 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4378 | * @pid: the pid in question. | |
4379 | */ | |
4380 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4381 | { | |
36c8b586 | 4382 | struct task_struct *p; |
3a5c359a | 4383 | int retval; |
1da177e4 LT |
4384 | |
4385 | if (pid < 0) | |
3a5c359a | 4386 | return -EINVAL; |
1da177e4 LT |
4387 | |
4388 | retval = -ESRCH; | |
4389 | read_lock(&tasklist_lock); | |
4390 | p = find_process_by_pid(pid); | |
4391 | if (p) { | |
4392 | retval = security_task_getscheduler(p); | |
4393 | if (!retval) | |
4394 | retval = p->policy; | |
4395 | } | |
4396 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4397 | return retval; |
4398 | } | |
4399 | ||
4400 | /** | |
4401 | * sys_sched_getscheduler - get the RT priority of a thread | |
4402 | * @pid: the pid in question. | |
4403 | * @param: structure containing the RT priority. | |
4404 | */ | |
4405 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4406 | { | |
4407 | struct sched_param lp; | |
36c8b586 | 4408 | struct task_struct *p; |
3a5c359a | 4409 | int retval; |
1da177e4 LT |
4410 | |
4411 | if (!param || pid < 0) | |
3a5c359a | 4412 | return -EINVAL; |
1da177e4 LT |
4413 | |
4414 | read_lock(&tasklist_lock); | |
4415 | p = find_process_by_pid(pid); | |
4416 | retval = -ESRCH; | |
4417 | if (!p) | |
4418 | goto out_unlock; | |
4419 | ||
4420 | retval = security_task_getscheduler(p); | |
4421 | if (retval) | |
4422 | goto out_unlock; | |
4423 | ||
4424 | lp.sched_priority = p->rt_priority; | |
4425 | read_unlock(&tasklist_lock); | |
4426 | ||
4427 | /* | |
4428 | * This one might sleep, we cannot do it with a spinlock held ... | |
4429 | */ | |
4430 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4431 | ||
1da177e4 LT |
4432 | return retval; |
4433 | ||
4434 | out_unlock: | |
4435 | read_unlock(&tasklist_lock); | |
4436 | return retval; | |
4437 | } | |
4438 | ||
4439 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4440 | { | |
1da177e4 | 4441 | cpumask_t cpus_allowed; |
36c8b586 IM |
4442 | struct task_struct *p; |
4443 | int retval; | |
1da177e4 | 4444 | |
5be9361c | 4445 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4446 | read_lock(&tasklist_lock); |
4447 | ||
4448 | p = find_process_by_pid(pid); | |
4449 | if (!p) { | |
4450 | read_unlock(&tasklist_lock); | |
5be9361c | 4451 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4452 | return -ESRCH; |
4453 | } | |
4454 | ||
4455 | /* | |
4456 | * It is not safe to call set_cpus_allowed with the | |
4457 | * tasklist_lock held. We will bump the task_struct's | |
4458 | * usage count and then drop tasklist_lock. | |
4459 | */ | |
4460 | get_task_struct(p); | |
4461 | read_unlock(&tasklist_lock); | |
4462 | ||
4463 | retval = -EPERM; | |
4464 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4465 | !capable(CAP_SYS_NICE)) | |
4466 | goto out_unlock; | |
4467 | ||
e7834f8f DQ |
4468 | retval = security_task_setscheduler(p, 0, NULL); |
4469 | if (retval) | |
4470 | goto out_unlock; | |
4471 | ||
1da177e4 LT |
4472 | cpus_allowed = cpuset_cpus_allowed(p); |
4473 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4474 | again: |
1da177e4 LT |
4475 | retval = set_cpus_allowed(p, new_mask); |
4476 | ||
8707d8b8 PM |
4477 | if (!retval) { |
4478 | cpus_allowed = cpuset_cpus_allowed(p); | |
4479 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4480 | /* | |
4481 | * We must have raced with a concurrent cpuset | |
4482 | * update. Just reset the cpus_allowed to the | |
4483 | * cpuset's cpus_allowed | |
4484 | */ | |
4485 | new_mask = cpus_allowed; | |
4486 | goto again; | |
4487 | } | |
4488 | } | |
1da177e4 LT |
4489 | out_unlock: |
4490 | put_task_struct(p); | |
5be9361c | 4491 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4492 | return retval; |
4493 | } | |
4494 | ||
4495 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4496 | cpumask_t *new_mask) | |
4497 | { | |
4498 | if (len < sizeof(cpumask_t)) { | |
4499 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4500 | } else if (len > sizeof(cpumask_t)) { | |
4501 | len = sizeof(cpumask_t); | |
4502 | } | |
4503 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4504 | } | |
4505 | ||
4506 | /** | |
4507 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4508 | * @pid: pid of the process | |
4509 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4510 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4511 | */ | |
4512 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4513 | unsigned long __user *user_mask_ptr) | |
4514 | { | |
4515 | cpumask_t new_mask; | |
4516 | int retval; | |
4517 | ||
4518 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4519 | if (retval) | |
4520 | return retval; | |
4521 | ||
4522 | return sched_setaffinity(pid, new_mask); | |
4523 | } | |
4524 | ||
4525 | /* | |
4526 | * Represents all cpu's present in the system | |
4527 | * In systems capable of hotplug, this map could dynamically grow | |
4528 | * as new cpu's are detected in the system via any platform specific | |
4529 | * method, such as ACPI for e.g. | |
4530 | */ | |
4531 | ||
4cef0c61 | 4532 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4533 | EXPORT_SYMBOL(cpu_present_map); |
4534 | ||
4535 | #ifndef CONFIG_SMP | |
4cef0c61 | 4536 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4537 | EXPORT_SYMBOL(cpu_online_map); |
4538 | ||
4cef0c61 | 4539 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4540 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4541 | #endif |
4542 | ||
4543 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4544 | { | |
36c8b586 | 4545 | struct task_struct *p; |
1da177e4 | 4546 | int retval; |
1da177e4 | 4547 | |
5be9361c | 4548 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4549 | read_lock(&tasklist_lock); |
4550 | ||
4551 | retval = -ESRCH; | |
4552 | p = find_process_by_pid(pid); | |
4553 | if (!p) | |
4554 | goto out_unlock; | |
4555 | ||
e7834f8f DQ |
4556 | retval = security_task_getscheduler(p); |
4557 | if (retval) | |
4558 | goto out_unlock; | |
4559 | ||
2f7016d9 | 4560 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4561 | |
4562 | out_unlock: | |
4563 | read_unlock(&tasklist_lock); | |
5be9361c | 4564 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4565 | |
9531b62f | 4566 | return retval; |
1da177e4 LT |
4567 | } |
4568 | ||
4569 | /** | |
4570 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4571 | * @pid: pid of the process | |
4572 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4573 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4574 | */ | |
4575 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4576 | unsigned long __user *user_mask_ptr) | |
4577 | { | |
4578 | int ret; | |
4579 | cpumask_t mask; | |
4580 | ||
4581 | if (len < sizeof(cpumask_t)) | |
4582 | return -EINVAL; | |
4583 | ||
4584 | ret = sched_getaffinity(pid, &mask); | |
4585 | if (ret < 0) | |
4586 | return ret; | |
4587 | ||
4588 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4589 | return -EFAULT; | |
4590 | ||
4591 | return sizeof(cpumask_t); | |
4592 | } | |
4593 | ||
4594 | /** | |
4595 | * sys_sched_yield - yield the current processor to other threads. | |
4596 | * | |
dd41f596 IM |
4597 | * This function yields the current CPU to other tasks. If there are no |
4598 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4599 | */ |
4600 | asmlinkage long sys_sched_yield(void) | |
4601 | { | |
70b97a7f | 4602 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4603 | |
2d72376b | 4604 | schedstat_inc(rq, yld_count); |
4530d7ab | 4605 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4606 | |
4607 | /* | |
4608 | * Since we are going to call schedule() anyway, there's | |
4609 | * no need to preempt or enable interrupts: | |
4610 | */ | |
4611 | __release(rq->lock); | |
8a25d5de | 4612 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4613 | _raw_spin_unlock(&rq->lock); |
4614 | preempt_enable_no_resched(); | |
4615 | ||
4616 | schedule(); | |
4617 | ||
4618 | return 0; | |
4619 | } | |
4620 | ||
e7b38404 | 4621 | static void __cond_resched(void) |
1da177e4 | 4622 | { |
8e0a43d8 IM |
4623 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4624 | __might_sleep(__FILE__, __LINE__); | |
4625 | #endif | |
5bbcfd90 IM |
4626 | /* |
4627 | * The BKS might be reacquired before we have dropped | |
4628 | * PREEMPT_ACTIVE, which could trigger a second | |
4629 | * cond_resched() call. | |
4630 | */ | |
1da177e4 LT |
4631 | do { |
4632 | add_preempt_count(PREEMPT_ACTIVE); | |
4633 | schedule(); | |
4634 | sub_preempt_count(PREEMPT_ACTIVE); | |
4635 | } while (need_resched()); | |
4636 | } | |
4637 | ||
4638 | int __sched cond_resched(void) | |
4639 | { | |
9414232f IM |
4640 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4641 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4642 | __cond_resched(); |
4643 | return 1; | |
4644 | } | |
4645 | return 0; | |
4646 | } | |
1da177e4 LT |
4647 | EXPORT_SYMBOL(cond_resched); |
4648 | ||
4649 | /* | |
4650 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4651 | * call schedule, and on return reacquire the lock. | |
4652 | * | |
4653 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4654 | * operations here to prevent schedule() from being called twice (once via | |
4655 | * spin_unlock(), once by hand). | |
4656 | */ | |
95cdf3b7 | 4657 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4658 | { |
6df3cecb JK |
4659 | int ret = 0; |
4660 | ||
1da177e4 LT |
4661 | if (need_lockbreak(lock)) { |
4662 | spin_unlock(lock); | |
4663 | cpu_relax(); | |
6df3cecb | 4664 | ret = 1; |
1da177e4 LT |
4665 | spin_lock(lock); |
4666 | } | |
9414232f | 4667 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4668 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4669 | _raw_spin_unlock(lock); |
4670 | preempt_enable_no_resched(); | |
4671 | __cond_resched(); | |
6df3cecb | 4672 | ret = 1; |
1da177e4 | 4673 | spin_lock(lock); |
1da177e4 | 4674 | } |
6df3cecb | 4675 | return ret; |
1da177e4 | 4676 | } |
1da177e4 LT |
4677 | EXPORT_SYMBOL(cond_resched_lock); |
4678 | ||
4679 | int __sched cond_resched_softirq(void) | |
4680 | { | |
4681 | BUG_ON(!in_softirq()); | |
4682 | ||
9414232f | 4683 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4684 | local_bh_enable(); |
1da177e4 LT |
4685 | __cond_resched(); |
4686 | local_bh_disable(); | |
4687 | return 1; | |
4688 | } | |
4689 | return 0; | |
4690 | } | |
1da177e4 LT |
4691 | EXPORT_SYMBOL(cond_resched_softirq); |
4692 | ||
1da177e4 LT |
4693 | /** |
4694 | * yield - yield the current processor to other threads. | |
4695 | * | |
72fd4a35 | 4696 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4697 | * thread runnable and calls sys_sched_yield(). |
4698 | */ | |
4699 | void __sched yield(void) | |
4700 | { | |
4701 | set_current_state(TASK_RUNNING); | |
4702 | sys_sched_yield(); | |
4703 | } | |
1da177e4 LT |
4704 | EXPORT_SYMBOL(yield); |
4705 | ||
4706 | /* | |
4707 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4708 | * that process accounting knows that this is a task in IO wait state. | |
4709 | * | |
4710 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4711 | * has set its backing_dev_info: the queue against which it should throttle) | |
4712 | */ | |
4713 | void __sched io_schedule(void) | |
4714 | { | |
70b97a7f | 4715 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4716 | |
0ff92245 | 4717 | delayacct_blkio_start(); |
1da177e4 LT |
4718 | atomic_inc(&rq->nr_iowait); |
4719 | schedule(); | |
4720 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4721 | delayacct_blkio_end(); |
1da177e4 | 4722 | } |
1da177e4 LT |
4723 | EXPORT_SYMBOL(io_schedule); |
4724 | ||
4725 | long __sched io_schedule_timeout(long timeout) | |
4726 | { | |
70b97a7f | 4727 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4728 | long ret; |
4729 | ||
0ff92245 | 4730 | delayacct_blkio_start(); |
1da177e4 LT |
4731 | atomic_inc(&rq->nr_iowait); |
4732 | ret = schedule_timeout(timeout); | |
4733 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4734 | delayacct_blkio_end(); |
1da177e4 LT |
4735 | return ret; |
4736 | } | |
4737 | ||
4738 | /** | |
4739 | * sys_sched_get_priority_max - return maximum RT priority. | |
4740 | * @policy: scheduling class. | |
4741 | * | |
4742 | * this syscall returns the maximum rt_priority that can be used | |
4743 | * by a given scheduling class. | |
4744 | */ | |
4745 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4746 | { | |
4747 | int ret = -EINVAL; | |
4748 | ||
4749 | switch (policy) { | |
4750 | case SCHED_FIFO: | |
4751 | case SCHED_RR: | |
4752 | ret = MAX_USER_RT_PRIO-1; | |
4753 | break; | |
4754 | case SCHED_NORMAL: | |
b0a9499c | 4755 | case SCHED_BATCH: |
dd41f596 | 4756 | case SCHED_IDLE: |
1da177e4 LT |
4757 | ret = 0; |
4758 | break; | |
4759 | } | |
4760 | return ret; | |
4761 | } | |
4762 | ||
4763 | /** | |
4764 | * sys_sched_get_priority_min - return minimum RT priority. | |
4765 | * @policy: scheduling class. | |
4766 | * | |
4767 | * this syscall returns the minimum rt_priority that can be used | |
4768 | * by a given scheduling class. | |
4769 | */ | |
4770 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4771 | { | |
4772 | int ret = -EINVAL; | |
4773 | ||
4774 | switch (policy) { | |
4775 | case SCHED_FIFO: | |
4776 | case SCHED_RR: | |
4777 | ret = 1; | |
4778 | break; | |
4779 | case SCHED_NORMAL: | |
b0a9499c | 4780 | case SCHED_BATCH: |
dd41f596 | 4781 | case SCHED_IDLE: |
1da177e4 LT |
4782 | ret = 0; |
4783 | } | |
4784 | return ret; | |
4785 | } | |
4786 | ||
4787 | /** | |
4788 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4789 | * @pid: pid of the process. | |
4790 | * @interval: userspace pointer to the timeslice value. | |
4791 | * | |
4792 | * this syscall writes the default timeslice value of a given process | |
4793 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4794 | */ | |
4795 | asmlinkage | |
4796 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4797 | { | |
36c8b586 | 4798 | struct task_struct *p; |
a4ec24b4 | 4799 | unsigned int time_slice; |
3a5c359a | 4800 | int retval; |
1da177e4 | 4801 | struct timespec t; |
1da177e4 LT |
4802 | |
4803 | if (pid < 0) | |
3a5c359a | 4804 | return -EINVAL; |
1da177e4 LT |
4805 | |
4806 | retval = -ESRCH; | |
4807 | read_lock(&tasklist_lock); | |
4808 | p = find_process_by_pid(pid); | |
4809 | if (!p) | |
4810 | goto out_unlock; | |
4811 | ||
4812 | retval = security_task_getscheduler(p); | |
4813 | if (retval) | |
4814 | goto out_unlock; | |
4815 | ||
a4ec24b4 DA |
4816 | if (p->policy == SCHED_FIFO) |
4817 | time_slice = 0; | |
4818 | else if (p->policy == SCHED_RR) | |
4819 | time_slice = DEF_TIMESLICE; | |
4820 | else { | |
4821 | struct sched_entity *se = &p->se; | |
4822 | unsigned long flags; | |
4823 | struct rq *rq; | |
4824 | ||
4825 | rq = task_rq_lock(p, &flags); | |
4826 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4827 | task_rq_unlock(rq, &flags); | |
4828 | } | |
1da177e4 | 4829 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4830 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4831 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4832 | return retval; |
3a5c359a | 4833 | |
1da177e4 LT |
4834 | out_unlock: |
4835 | read_unlock(&tasklist_lock); | |
4836 | return retval; | |
4837 | } | |
4838 | ||
2ed6e34f | 4839 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4840 | |
4841 | static void show_task(struct task_struct *p) | |
1da177e4 | 4842 | { |
1da177e4 | 4843 | unsigned long free = 0; |
36c8b586 | 4844 | unsigned state; |
1da177e4 | 4845 | |
1da177e4 | 4846 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4847 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4848 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4849 | #if BITS_PER_LONG == 32 |
1da177e4 | 4850 | if (state == TASK_RUNNING) |
cc4ea795 | 4851 | printk(KERN_CONT " running "); |
1da177e4 | 4852 | else |
cc4ea795 | 4853 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4854 | #else |
4855 | if (state == TASK_RUNNING) | |
cc4ea795 | 4856 | printk(KERN_CONT " running task "); |
1da177e4 | 4857 | else |
cc4ea795 | 4858 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4859 | #endif |
4860 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4861 | { | |
10ebffde | 4862 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4863 | while (!*n) |
4864 | n++; | |
10ebffde | 4865 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4866 | } |
4867 | #endif | |
cc4ea795 | 4868 | printk(KERN_CONT "%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4869 | |
4870 | if (state != TASK_RUNNING) | |
4871 | show_stack(p, NULL); | |
4872 | } | |
4873 | ||
e59e2ae2 | 4874 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4875 | { |
36c8b586 | 4876 | struct task_struct *g, *p; |
1da177e4 | 4877 | |
4bd77321 IM |
4878 | #if BITS_PER_LONG == 32 |
4879 | printk(KERN_INFO | |
4880 | " task PC stack pid father\n"); | |
1da177e4 | 4881 | #else |
4bd77321 IM |
4882 | printk(KERN_INFO |
4883 | " task PC stack pid father\n"); | |
1da177e4 LT |
4884 | #endif |
4885 | read_lock(&tasklist_lock); | |
4886 | do_each_thread(g, p) { | |
4887 | /* | |
4888 | * reset the NMI-timeout, listing all files on a slow | |
4889 | * console might take alot of time: | |
4890 | */ | |
4891 | touch_nmi_watchdog(); | |
39bc89fd | 4892 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4893 | show_task(p); |
1da177e4 LT |
4894 | } while_each_thread(g, p); |
4895 | ||
04c9167f JF |
4896 | touch_all_softlockup_watchdogs(); |
4897 | ||
dd41f596 IM |
4898 | #ifdef CONFIG_SCHED_DEBUG |
4899 | sysrq_sched_debug_show(); | |
4900 | #endif | |
1da177e4 | 4901 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4902 | /* |
4903 | * Only show locks if all tasks are dumped: | |
4904 | */ | |
4905 | if (state_filter == -1) | |
4906 | debug_show_all_locks(); | |
1da177e4 LT |
4907 | } |
4908 | ||
1df21055 IM |
4909 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4910 | { | |
dd41f596 | 4911 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4912 | } |
4913 | ||
f340c0d1 IM |
4914 | /** |
4915 | * init_idle - set up an idle thread for a given CPU | |
4916 | * @idle: task in question | |
4917 | * @cpu: cpu the idle task belongs to | |
4918 | * | |
4919 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4920 | * flag, to make booting more robust. | |
4921 | */ | |
5c1e1767 | 4922 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4923 | { |
70b97a7f | 4924 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4925 | unsigned long flags; |
4926 | ||
dd41f596 IM |
4927 | __sched_fork(idle); |
4928 | idle->se.exec_start = sched_clock(); | |
4929 | ||
b29739f9 | 4930 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4931 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4932 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4933 | |
4934 | spin_lock_irqsave(&rq->lock, flags); | |
4935 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4936 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4937 | idle->oncpu = 1; | |
4938 | #endif | |
1da177e4 LT |
4939 | spin_unlock_irqrestore(&rq->lock, flags); |
4940 | ||
4941 | /* Set the preempt count _outside_ the spinlocks! */ | |
4942 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4943 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4944 | #else |
a1261f54 | 4945 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4946 | #endif |
dd41f596 IM |
4947 | /* |
4948 | * The idle tasks have their own, simple scheduling class: | |
4949 | */ | |
4950 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4951 | } |
4952 | ||
4953 | /* | |
4954 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4955 | * indicates which cpus entered this state. This is used | |
4956 | * in the rcu update to wait only for active cpus. For system | |
4957 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4958 | * always be CPU_MASK_NONE. | |
4959 | */ | |
4960 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4961 | ||
4962 | #ifdef CONFIG_SMP | |
4963 | /* | |
4964 | * This is how migration works: | |
4965 | * | |
70b97a7f | 4966 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4967 | * runqueue and wake up that CPU's migration thread. |
4968 | * 2) we down() the locked semaphore => thread blocks. | |
4969 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4970 | * thread off the CPU) | |
4971 | * 4) it gets the migration request and checks whether the migrated | |
4972 | * task is still in the wrong runqueue. | |
4973 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4974 | * it and puts it into the right queue. | |
4975 | * 6) migration thread up()s the semaphore. | |
4976 | * 7) we wake up and the migration is done. | |
4977 | */ | |
4978 | ||
4979 | /* | |
4980 | * Change a given task's CPU affinity. Migrate the thread to a | |
4981 | * proper CPU and schedule it away if the CPU it's executing on | |
4982 | * is removed from the allowed bitmask. | |
4983 | * | |
4984 | * NOTE: the caller must have a valid reference to the task, the | |
4985 | * task must not exit() & deallocate itself prematurely. The | |
4986 | * call is not atomic; no spinlocks may be held. | |
4987 | */ | |
36c8b586 | 4988 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4989 | { |
70b97a7f | 4990 | struct migration_req req; |
1da177e4 | 4991 | unsigned long flags; |
70b97a7f | 4992 | struct rq *rq; |
48f24c4d | 4993 | int ret = 0; |
1da177e4 LT |
4994 | |
4995 | rq = task_rq_lock(p, &flags); | |
4996 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4997 | ret = -EINVAL; | |
4998 | goto out; | |
4999 | } | |
5000 | ||
5001 | p->cpus_allowed = new_mask; | |
5002 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5003 | if (cpu_isset(task_cpu(p), new_mask)) | |
5004 | goto out; | |
5005 | ||
5006 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5007 | /* Need help from migration thread: drop lock and wait. */ | |
5008 | task_rq_unlock(rq, &flags); | |
5009 | wake_up_process(rq->migration_thread); | |
5010 | wait_for_completion(&req.done); | |
5011 | tlb_migrate_finish(p->mm); | |
5012 | return 0; | |
5013 | } | |
5014 | out: | |
5015 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5016 | |
1da177e4 LT |
5017 | return ret; |
5018 | } | |
1da177e4 LT |
5019 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5020 | ||
5021 | /* | |
5022 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5023 | * this because either it can't run here any more (set_cpus_allowed() | |
5024 | * away from this CPU, or CPU going down), or because we're | |
5025 | * attempting to rebalance this task on exec (sched_exec). | |
5026 | * | |
5027 | * So we race with normal scheduler movements, but that's OK, as long | |
5028 | * as the task is no longer on this CPU. | |
efc30814 KK |
5029 | * |
5030 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5031 | */ |
efc30814 | 5032 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5033 | { |
70b97a7f | 5034 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5035 | int ret = 0, on_rq; |
1da177e4 LT |
5036 | |
5037 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5038 | return ret; |
1da177e4 LT |
5039 | |
5040 | rq_src = cpu_rq(src_cpu); | |
5041 | rq_dest = cpu_rq(dest_cpu); | |
5042 | ||
5043 | double_rq_lock(rq_src, rq_dest); | |
5044 | /* Already moved. */ | |
5045 | if (task_cpu(p) != src_cpu) | |
5046 | goto out; | |
5047 | /* Affinity changed (again). */ | |
5048 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5049 | goto out; | |
5050 | ||
dd41f596 | 5051 | on_rq = p->se.on_rq; |
6e82a3be | 5052 | if (on_rq) |
2e1cb74a | 5053 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5054 | |
1da177e4 | 5055 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5056 | if (on_rq) { |
5057 | activate_task(rq_dest, p, 0); | |
5058 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5059 | } |
efc30814 | 5060 | ret = 1; |
1da177e4 LT |
5061 | out: |
5062 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5063 | return ret; |
1da177e4 LT |
5064 | } |
5065 | ||
5066 | /* | |
5067 | * migration_thread - this is a highprio system thread that performs | |
5068 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5069 | * another runqueue. | |
5070 | */ | |
95cdf3b7 | 5071 | static int migration_thread(void *data) |
1da177e4 | 5072 | { |
1da177e4 | 5073 | int cpu = (long)data; |
70b97a7f | 5074 | struct rq *rq; |
1da177e4 LT |
5075 | |
5076 | rq = cpu_rq(cpu); | |
5077 | BUG_ON(rq->migration_thread != current); | |
5078 | ||
5079 | set_current_state(TASK_INTERRUPTIBLE); | |
5080 | while (!kthread_should_stop()) { | |
70b97a7f | 5081 | struct migration_req *req; |
1da177e4 | 5082 | struct list_head *head; |
1da177e4 | 5083 | |
1da177e4 LT |
5084 | spin_lock_irq(&rq->lock); |
5085 | ||
5086 | if (cpu_is_offline(cpu)) { | |
5087 | spin_unlock_irq(&rq->lock); | |
5088 | goto wait_to_die; | |
5089 | } | |
5090 | ||
5091 | if (rq->active_balance) { | |
5092 | active_load_balance(rq, cpu); | |
5093 | rq->active_balance = 0; | |
5094 | } | |
5095 | ||
5096 | head = &rq->migration_queue; | |
5097 | ||
5098 | if (list_empty(head)) { | |
5099 | spin_unlock_irq(&rq->lock); | |
5100 | schedule(); | |
5101 | set_current_state(TASK_INTERRUPTIBLE); | |
5102 | continue; | |
5103 | } | |
70b97a7f | 5104 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5105 | list_del_init(head->next); |
5106 | ||
674311d5 NP |
5107 | spin_unlock(&rq->lock); |
5108 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5109 | local_irq_enable(); | |
1da177e4 LT |
5110 | |
5111 | complete(&req->done); | |
5112 | } | |
5113 | __set_current_state(TASK_RUNNING); | |
5114 | return 0; | |
5115 | ||
5116 | wait_to_die: | |
5117 | /* Wait for kthread_stop */ | |
5118 | set_current_state(TASK_INTERRUPTIBLE); | |
5119 | while (!kthread_should_stop()) { | |
5120 | schedule(); | |
5121 | set_current_state(TASK_INTERRUPTIBLE); | |
5122 | } | |
5123 | __set_current_state(TASK_RUNNING); | |
5124 | return 0; | |
5125 | } | |
5126 | ||
5127 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5128 | |
5129 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5130 | { | |
5131 | int ret; | |
5132 | ||
5133 | local_irq_disable(); | |
5134 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5135 | local_irq_enable(); | |
5136 | return ret; | |
5137 | } | |
5138 | ||
054b9108 KK |
5139 | /* |
5140 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5141 | * NOTE: interrupts should be disabled by the caller | |
5142 | */ | |
48f24c4d | 5143 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5144 | { |
efc30814 | 5145 | unsigned long flags; |
1da177e4 | 5146 | cpumask_t mask; |
70b97a7f IM |
5147 | struct rq *rq; |
5148 | int dest_cpu; | |
1da177e4 | 5149 | |
3a5c359a AK |
5150 | do { |
5151 | /* On same node? */ | |
5152 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5153 | cpus_and(mask, mask, p->cpus_allowed); | |
5154 | dest_cpu = any_online_cpu(mask); | |
5155 | ||
5156 | /* On any allowed CPU? */ | |
5157 | if (dest_cpu == NR_CPUS) | |
5158 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5159 | ||
5160 | /* No more Mr. Nice Guy. */ | |
5161 | if (dest_cpu == NR_CPUS) { | |
5162 | rq = task_rq_lock(p, &flags); | |
5163 | cpus_setall(p->cpus_allowed); | |
5164 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5165 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5166 | |
3a5c359a AK |
5167 | /* |
5168 | * Don't tell them about moving exiting tasks or | |
5169 | * kernel threads (both mm NULL), since they never | |
5170 | * leave kernel. | |
5171 | */ | |
5172 | if (p->mm && printk_ratelimit()) | |
5173 | printk(KERN_INFO "process %d (%s) no " | |
5174 | "longer affine to cpu%d\n", | |
5175 | p->pid, p->comm, dead_cpu); | |
5176 | } | |
f7b4cddc | 5177 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5178 | } |
5179 | ||
5180 | /* | |
5181 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5182 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5183 | * for performance reasons the counter is not stricly tracking tasks to | |
5184 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5185 | * to keep the global sum constant after CPU-down: | |
5186 | */ | |
70b97a7f | 5187 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5188 | { |
70b97a7f | 5189 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5190 | unsigned long flags; |
5191 | ||
5192 | local_irq_save(flags); | |
5193 | double_rq_lock(rq_src, rq_dest); | |
5194 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5195 | rq_src->nr_uninterruptible = 0; | |
5196 | double_rq_unlock(rq_src, rq_dest); | |
5197 | local_irq_restore(flags); | |
5198 | } | |
5199 | ||
5200 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5201 | static void migrate_live_tasks(int src_cpu) | |
5202 | { | |
48f24c4d | 5203 | struct task_struct *p, *t; |
1da177e4 | 5204 | |
f7b4cddc | 5205 | read_lock(&tasklist_lock); |
1da177e4 | 5206 | |
48f24c4d IM |
5207 | do_each_thread(t, p) { |
5208 | if (p == current) | |
1da177e4 LT |
5209 | continue; |
5210 | ||
48f24c4d IM |
5211 | if (task_cpu(p) == src_cpu) |
5212 | move_task_off_dead_cpu(src_cpu, p); | |
5213 | } while_each_thread(t, p); | |
1da177e4 | 5214 | |
f7b4cddc | 5215 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5216 | } |
5217 | ||
a9957449 AD |
5218 | /* |
5219 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5220 | */ | |
5221 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5222 | { | |
5223 | update_rq_clock(rq); | |
5224 | ||
5225 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5226 | rq->nr_uninterruptible--; | |
5227 | ||
5228 | enqueue_task(rq, p, 0); | |
5229 | inc_nr_running(p, rq); | |
5230 | } | |
5231 | ||
dd41f596 IM |
5232 | /* |
5233 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5234 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5235 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5236 | */ |
5237 | void sched_idle_next(void) | |
5238 | { | |
48f24c4d | 5239 | int this_cpu = smp_processor_id(); |
70b97a7f | 5240 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5241 | struct task_struct *p = rq->idle; |
5242 | unsigned long flags; | |
5243 | ||
5244 | /* cpu has to be offline */ | |
48f24c4d | 5245 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5246 | |
48f24c4d IM |
5247 | /* |
5248 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5249 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5250 | */ |
5251 | spin_lock_irqsave(&rq->lock, flags); | |
5252 | ||
dd41f596 | 5253 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5254 | |
5255 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5256 | activate_idle_task(p, rq); |
1da177e4 LT |
5257 | |
5258 | spin_unlock_irqrestore(&rq->lock, flags); | |
5259 | } | |
5260 | ||
48f24c4d IM |
5261 | /* |
5262 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5263 | * offline. |
5264 | */ | |
5265 | void idle_task_exit(void) | |
5266 | { | |
5267 | struct mm_struct *mm = current->active_mm; | |
5268 | ||
5269 | BUG_ON(cpu_online(smp_processor_id())); | |
5270 | ||
5271 | if (mm != &init_mm) | |
5272 | switch_mm(mm, &init_mm, current); | |
5273 | mmdrop(mm); | |
5274 | } | |
5275 | ||
054b9108 | 5276 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5277 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5278 | { |
70b97a7f | 5279 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5280 | |
5281 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5282 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5283 | |
5284 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5285 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5286 | |
48f24c4d | 5287 | get_task_struct(p); |
1da177e4 LT |
5288 | |
5289 | /* | |
5290 | * Drop lock around migration; if someone else moves it, | |
5291 | * that's OK. No task can be added to this CPU, so iteration is | |
5292 | * fine. | |
5293 | */ | |
f7b4cddc | 5294 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5295 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5296 | spin_lock_irq(&rq->lock); |
1da177e4 | 5297 | |
48f24c4d | 5298 | put_task_struct(p); |
1da177e4 LT |
5299 | } |
5300 | ||
5301 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5302 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5303 | { | |
70b97a7f | 5304 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5305 | struct task_struct *next; |
48f24c4d | 5306 | |
dd41f596 IM |
5307 | for ( ; ; ) { |
5308 | if (!rq->nr_running) | |
5309 | break; | |
a8e504d2 | 5310 | update_rq_clock(rq); |
ff95f3df | 5311 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5312 | if (!next) |
5313 | break; | |
5314 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5315 | |
1da177e4 LT |
5316 | } |
5317 | } | |
5318 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5319 | ||
e692ab53 NP |
5320 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5321 | ||
5322 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5323 | { |
5324 | .procname = "sched_domain", | |
c57baf1e | 5325 | .mode = 0555, |
e0361851 | 5326 | }, |
e692ab53 NP |
5327 | {0,}, |
5328 | }; | |
5329 | ||
5330 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5331 | { |
c57baf1e | 5332 | .ctl_name = CTL_KERN, |
e0361851 | 5333 | .procname = "kernel", |
c57baf1e | 5334 | .mode = 0555, |
e0361851 AD |
5335 | .child = sd_ctl_dir, |
5336 | }, | |
e692ab53 NP |
5337 | {0,}, |
5338 | }; | |
5339 | ||
5340 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5341 | { | |
5342 | struct ctl_table *entry = | |
5cf9f062 | 5343 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5344 | |
e692ab53 NP |
5345 | return entry; |
5346 | } | |
5347 | ||
6382bc90 MM |
5348 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5349 | { | |
cd790076 | 5350 | struct ctl_table *entry; |
6382bc90 | 5351 | |
cd790076 MM |
5352 | /* |
5353 | * In the intermediate directories, both the child directory and | |
5354 | * procname are dynamically allocated and could fail but the mode | |
5355 | * will always be set. In the lowest directory the names are | |
5356 | * static strings and all have proc handlers. | |
5357 | */ | |
5358 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5359 | if (entry->child) |
5360 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5361 | if (entry->proc_handler == NULL) |
5362 | kfree(entry->procname); | |
5363 | } | |
6382bc90 MM |
5364 | |
5365 | kfree(*tablep); | |
5366 | *tablep = NULL; | |
5367 | } | |
5368 | ||
e692ab53 | 5369 | static void |
e0361851 | 5370 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5371 | const char *procname, void *data, int maxlen, |
5372 | mode_t mode, proc_handler *proc_handler) | |
5373 | { | |
e692ab53 NP |
5374 | entry->procname = procname; |
5375 | entry->data = data; | |
5376 | entry->maxlen = maxlen; | |
5377 | entry->mode = mode; | |
5378 | entry->proc_handler = proc_handler; | |
5379 | } | |
5380 | ||
5381 | static struct ctl_table * | |
5382 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5383 | { | |
ace8b3d6 | 5384 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5385 | |
ad1cdc1d MM |
5386 | if (table == NULL) |
5387 | return NULL; | |
5388 | ||
e0361851 | 5389 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5390 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5391 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5392 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5393 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5394 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5395 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5396 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5397 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5398 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5399 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5400 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5401 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5402 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5403 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5404 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5405 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5406 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5407 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5408 | &sd->cache_nice_tries, |
5409 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5410 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5411 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5412 | /* &table[11] is terminator */ |
e692ab53 NP |
5413 | |
5414 | return table; | |
5415 | } | |
5416 | ||
8401f775 | 5417 | static ctl_table * sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5418 | { |
5419 | struct ctl_table *entry, *table; | |
5420 | struct sched_domain *sd; | |
5421 | int domain_num = 0, i; | |
5422 | char buf[32]; | |
5423 | ||
5424 | for_each_domain(cpu, sd) | |
5425 | domain_num++; | |
5426 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5427 | if (table == NULL) |
5428 | return NULL; | |
e692ab53 NP |
5429 | |
5430 | i = 0; | |
5431 | for_each_domain(cpu, sd) { | |
5432 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5433 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5434 | entry->mode = 0555; |
e692ab53 NP |
5435 | entry->child = sd_alloc_ctl_domain_table(sd); |
5436 | entry++; | |
5437 | i++; | |
5438 | } | |
5439 | return table; | |
5440 | } | |
5441 | ||
5442 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5443 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5444 | { |
5445 | int i, cpu_num = num_online_cpus(); | |
5446 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5447 | char buf[32]; | |
5448 | ||
ad1cdc1d MM |
5449 | if (entry == NULL) |
5450 | return; | |
5451 | ||
e692ab53 NP |
5452 | sd_ctl_dir[0].child = entry; |
5453 | ||
97b6ea7b | 5454 | for_each_online_cpu(i) { |
e692ab53 | 5455 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5456 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5457 | entry->mode = 0555; |
e692ab53 | 5458 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5459 | entry++; |
e692ab53 NP |
5460 | } |
5461 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5462 | } | |
6382bc90 MM |
5463 | |
5464 | static void unregister_sched_domain_sysctl(void) | |
5465 | { | |
5466 | unregister_sysctl_table(sd_sysctl_header); | |
5467 | sd_sysctl_header = NULL; | |
5468 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
5469 | } | |
e692ab53 | 5470 | #else |
6382bc90 MM |
5471 | static void register_sched_domain_sysctl(void) |
5472 | { | |
5473 | } | |
5474 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5475 | { |
5476 | } | |
5477 | #endif | |
5478 | ||
1da177e4 LT |
5479 | /* |
5480 | * migration_call - callback that gets triggered when a CPU is added. | |
5481 | * Here we can start up the necessary migration thread for the new CPU. | |
5482 | */ | |
48f24c4d IM |
5483 | static int __cpuinit |
5484 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5485 | { |
1da177e4 | 5486 | struct task_struct *p; |
48f24c4d | 5487 | int cpu = (long)hcpu; |
1da177e4 | 5488 | unsigned long flags; |
70b97a7f | 5489 | struct rq *rq; |
1da177e4 LT |
5490 | |
5491 | switch (action) { | |
5be9361c GS |
5492 | case CPU_LOCK_ACQUIRE: |
5493 | mutex_lock(&sched_hotcpu_mutex); | |
5494 | break; | |
5495 | ||
1da177e4 | 5496 | case CPU_UP_PREPARE: |
8bb78442 | 5497 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5498 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5499 | if (IS_ERR(p)) |
5500 | return NOTIFY_BAD; | |
1da177e4 LT |
5501 | kthread_bind(p, cpu); |
5502 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5503 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5504 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5505 | task_rq_unlock(rq, &flags); |
5506 | cpu_rq(cpu)->migration_thread = p; | |
5507 | break; | |
48f24c4d | 5508 | |
1da177e4 | 5509 | case CPU_ONLINE: |
8bb78442 | 5510 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5511 | /* Strictly unneccessary, as first user will wake it. */ |
5512 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5513 | break; | |
48f24c4d | 5514 | |
1da177e4 LT |
5515 | #ifdef CONFIG_HOTPLUG_CPU |
5516 | case CPU_UP_CANCELED: | |
8bb78442 | 5517 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5518 | if (!cpu_rq(cpu)->migration_thread) |
5519 | break; | |
1da177e4 | 5520 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5521 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5522 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5523 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5524 | cpu_rq(cpu)->migration_thread = NULL; | |
5525 | break; | |
48f24c4d | 5526 | |
1da177e4 | 5527 | case CPU_DEAD: |
8bb78442 | 5528 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5529 | migrate_live_tasks(cpu); |
5530 | rq = cpu_rq(cpu); | |
5531 | kthread_stop(rq->migration_thread); | |
5532 | rq->migration_thread = NULL; | |
5533 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5534 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5535 | update_rq_clock(rq); |
2e1cb74a | 5536 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5537 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5538 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5539 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5540 | migrate_dead_tasks(cpu); |
d2da272a | 5541 | spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5542 | migrate_nr_uninterruptible(rq); |
5543 | BUG_ON(rq->nr_running != 0); | |
5544 | ||
5545 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5546 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5547 | * the requestors. */ |
5548 | spin_lock_irq(&rq->lock); | |
5549 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5550 | struct migration_req *req; |
5551 | ||
1da177e4 | 5552 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5553 | struct migration_req, list); |
1da177e4 LT |
5554 | list_del_init(&req->list); |
5555 | complete(&req->done); | |
5556 | } | |
5557 | spin_unlock_irq(&rq->lock); | |
5558 | break; | |
5559 | #endif | |
5be9361c GS |
5560 | case CPU_LOCK_RELEASE: |
5561 | mutex_unlock(&sched_hotcpu_mutex); | |
5562 | break; | |
1da177e4 LT |
5563 | } |
5564 | return NOTIFY_OK; | |
5565 | } | |
5566 | ||
5567 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5568 | * happens before everything else. | |
5569 | */ | |
26c2143b | 5570 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5571 | .notifier_call = migration_call, |
5572 | .priority = 10 | |
5573 | }; | |
5574 | ||
5575 | int __init migration_init(void) | |
5576 | { | |
5577 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5578 | int err; |
48f24c4d IM |
5579 | |
5580 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5581 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5582 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5583 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5584 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5585 | |
1da177e4 LT |
5586 | return 0; |
5587 | } | |
5588 | #endif | |
5589 | ||
5590 | #ifdef CONFIG_SMP | |
476f3534 CL |
5591 | |
5592 | /* Number of possible processor ids */ | |
5593 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5594 | EXPORT_SYMBOL(nr_cpu_ids); | |
5595 | ||
3e9830dc | 5596 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5597 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5598 | { | |
5599 | int level = 0; | |
5600 | ||
41c7ce9a NP |
5601 | if (!sd) { |
5602 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5603 | return; | |
5604 | } | |
5605 | ||
1da177e4 LT |
5606 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5607 | ||
5608 | do { | |
5609 | int i; | |
5610 | char str[NR_CPUS]; | |
5611 | struct sched_group *group = sd->groups; | |
5612 | cpumask_t groupmask; | |
5613 | ||
5614 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5615 | cpus_clear(groupmask); | |
5616 | ||
5617 | printk(KERN_DEBUG); | |
5618 | for (i = 0; i < level + 1; i++) | |
5619 | printk(" "); | |
5620 | printk("domain %d: ", level); | |
5621 | ||
5622 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5623 | printk("does not load-balance\n"); | |
5624 | if (sd->parent) | |
33859f7f MOS |
5625 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5626 | " has parent"); | |
1da177e4 LT |
5627 | break; |
5628 | } | |
5629 | ||
5630 | printk("span %s\n", str); | |
5631 | ||
5632 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5633 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5634 | "CPU%d\n", cpu); | |
1da177e4 | 5635 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5636 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5637 | " CPU%d\n", cpu); | |
1da177e4 LT |
5638 | |
5639 | printk(KERN_DEBUG); | |
5640 | for (i = 0; i < level + 2; i++) | |
5641 | printk(" "); | |
5642 | printk("groups:"); | |
5643 | do { | |
5644 | if (!group) { | |
5645 | printk("\n"); | |
5646 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5647 | break; | |
5648 | } | |
5649 | ||
5517d86b | 5650 | if (!group->__cpu_power) { |
cc4ea795 | 5651 | printk(KERN_CONT "\n"); |
33859f7f MOS |
5652 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5653 | "set\n"); | |
26797a34 | 5654 | break; |
1da177e4 LT |
5655 | } |
5656 | ||
5657 | if (!cpus_weight(group->cpumask)) { | |
cc4ea795 | 5658 | printk(KERN_CONT "\n"); |
1da177e4 | 5659 | printk(KERN_ERR "ERROR: empty group\n"); |
26797a34 | 5660 | break; |
1da177e4 LT |
5661 | } |
5662 | ||
5663 | if (cpus_intersects(groupmask, group->cpumask)) { | |
cc4ea795 | 5664 | printk(KERN_CONT "\n"); |
1da177e4 | 5665 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
26797a34 | 5666 | break; |
1da177e4 LT |
5667 | } |
5668 | ||
5669 | cpus_or(groupmask, groupmask, group->cpumask); | |
5670 | ||
5671 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
cc4ea795 | 5672 | printk(KERN_CONT " %s", str); |
1da177e4 LT |
5673 | |
5674 | group = group->next; | |
5675 | } while (group != sd->groups); | |
cc4ea795 | 5676 | printk(KERN_CONT "\n"); |
1da177e4 LT |
5677 | |
5678 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5679 | printk(KERN_ERR "ERROR: groups don't span " |
5680 | "domain->span\n"); | |
1da177e4 LT |
5681 | |
5682 | level++; | |
5683 | sd = sd->parent; | |
33859f7f MOS |
5684 | if (!sd) |
5685 | continue; | |
1da177e4 | 5686 | |
33859f7f MOS |
5687 | if (!cpus_subset(groupmask, sd->span)) |
5688 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5689 | "of domain->span\n"); | |
1da177e4 LT |
5690 | |
5691 | } while (sd); | |
5692 | } | |
5693 | #else | |
48f24c4d | 5694 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5695 | #endif |
5696 | ||
1a20ff27 | 5697 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5698 | { |
5699 | if (cpus_weight(sd->span) == 1) | |
5700 | return 1; | |
5701 | ||
5702 | /* Following flags need at least 2 groups */ | |
5703 | if (sd->flags & (SD_LOAD_BALANCE | | |
5704 | SD_BALANCE_NEWIDLE | | |
5705 | SD_BALANCE_FORK | | |
89c4710e SS |
5706 | SD_BALANCE_EXEC | |
5707 | SD_SHARE_CPUPOWER | | |
5708 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5709 | if (sd->groups != sd->groups->next) |
5710 | return 0; | |
5711 | } | |
5712 | ||
5713 | /* Following flags don't use groups */ | |
5714 | if (sd->flags & (SD_WAKE_IDLE | | |
5715 | SD_WAKE_AFFINE | | |
5716 | SD_WAKE_BALANCE)) | |
5717 | return 0; | |
5718 | ||
5719 | return 1; | |
5720 | } | |
5721 | ||
48f24c4d IM |
5722 | static int |
5723 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5724 | { |
5725 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5726 | ||
5727 | if (sd_degenerate(parent)) | |
5728 | return 1; | |
5729 | ||
5730 | if (!cpus_equal(sd->span, parent->span)) | |
5731 | return 0; | |
5732 | ||
5733 | /* Does parent contain flags not in child? */ | |
5734 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5735 | if (cflags & SD_WAKE_AFFINE) | |
5736 | pflags &= ~SD_WAKE_BALANCE; | |
5737 | /* Flags needing groups don't count if only 1 group in parent */ | |
5738 | if (parent->groups == parent->groups->next) { | |
5739 | pflags &= ~(SD_LOAD_BALANCE | | |
5740 | SD_BALANCE_NEWIDLE | | |
5741 | SD_BALANCE_FORK | | |
89c4710e SS |
5742 | SD_BALANCE_EXEC | |
5743 | SD_SHARE_CPUPOWER | | |
5744 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5745 | } |
5746 | if (~cflags & pflags) | |
5747 | return 0; | |
5748 | ||
5749 | return 1; | |
5750 | } | |
5751 | ||
1da177e4 LT |
5752 | /* |
5753 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5754 | * hold the hotplug lock. | |
5755 | */ | |
9c1cfda2 | 5756 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5757 | { |
70b97a7f | 5758 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5759 | struct sched_domain *tmp; |
5760 | ||
5761 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5762 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5763 | struct sched_domain *parent = tmp->parent; | |
5764 | if (!parent) | |
5765 | break; | |
1a848870 | 5766 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5767 | tmp->parent = parent->parent; |
1a848870 SS |
5768 | if (parent->parent) |
5769 | parent->parent->child = tmp; | |
5770 | } | |
245af2c7 SS |
5771 | } |
5772 | ||
1a848870 | 5773 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5774 | sd = sd->parent; |
1a848870 SS |
5775 | if (sd) |
5776 | sd->child = NULL; | |
5777 | } | |
1da177e4 LT |
5778 | |
5779 | sched_domain_debug(sd, cpu); | |
5780 | ||
674311d5 | 5781 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5782 | } |
5783 | ||
5784 | /* cpus with isolated domains */ | |
67af63a6 | 5785 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5786 | |
5787 | /* Setup the mask of cpus configured for isolated domains */ | |
5788 | static int __init isolated_cpu_setup(char *str) | |
5789 | { | |
5790 | int ints[NR_CPUS], i; | |
5791 | ||
5792 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5793 | cpus_clear(cpu_isolated_map); | |
5794 | for (i = 1; i <= ints[0]; i++) | |
5795 | if (ints[i] < NR_CPUS) | |
5796 | cpu_set(ints[i], cpu_isolated_map); | |
5797 | return 1; | |
5798 | } | |
5799 | ||
8927f494 | 5800 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5801 | |
5802 | /* | |
6711cab4 SS |
5803 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5804 | * to a function which identifies what group(along with sched group) a CPU | |
5805 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5806 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5807 | * |
5808 | * init_sched_build_groups will build a circular linked list of the groups | |
5809 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5810 | * and ->cpu_power to 0. | |
5811 | */ | |
a616058b | 5812 | static void |
6711cab4 SS |
5813 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5814 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5815 | struct sched_group **sg)) | |
1da177e4 LT |
5816 | { |
5817 | struct sched_group *first = NULL, *last = NULL; | |
5818 | cpumask_t covered = CPU_MASK_NONE; | |
5819 | int i; | |
5820 | ||
5821 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5822 | struct sched_group *sg; |
5823 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5824 | int j; |
5825 | ||
5826 | if (cpu_isset(i, covered)) | |
5827 | continue; | |
5828 | ||
5829 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5830 | sg->__cpu_power = 0; |
1da177e4 LT |
5831 | |
5832 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5833 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5834 | continue; |
5835 | ||
5836 | cpu_set(j, covered); | |
5837 | cpu_set(j, sg->cpumask); | |
5838 | } | |
5839 | if (!first) | |
5840 | first = sg; | |
5841 | if (last) | |
5842 | last->next = sg; | |
5843 | last = sg; | |
5844 | } | |
5845 | last->next = first; | |
5846 | } | |
5847 | ||
9c1cfda2 | 5848 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5849 | |
9c1cfda2 | 5850 | #ifdef CONFIG_NUMA |
198e2f18 | 5851 | |
9c1cfda2 JH |
5852 | /** |
5853 | * find_next_best_node - find the next node to include in a sched_domain | |
5854 | * @node: node whose sched_domain we're building | |
5855 | * @used_nodes: nodes already in the sched_domain | |
5856 | * | |
5857 | * Find the next node to include in a given scheduling domain. Simply | |
5858 | * finds the closest node not already in the @used_nodes map. | |
5859 | * | |
5860 | * Should use nodemask_t. | |
5861 | */ | |
5862 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5863 | { | |
5864 | int i, n, val, min_val, best_node = 0; | |
5865 | ||
5866 | min_val = INT_MAX; | |
5867 | ||
5868 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5869 | /* Start at @node */ | |
5870 | n = (node + i) % MAX_NUMNODES; | |
5871 | ||
5872 | if (!nr_cpus_node(n)) | |
5873 | continue; | |
5874 | ||
5875 | /* Skip already used nodes */ | |
5876 | if (test_bit(n, used_nodes)) | |
5877 | continue; | |
5878 | ||
5879 | /* Simple min distance search */ | |
5880 | val = node_distance(node, n); | |
5881 | ||
5882 | if (val < min_val) { | |
5883 | min_val = val; | |
5884 | best_node = n; | |
5885 | } | |
5886 | } | |
5887 | ||
5888 | set_bit(best_node, used_nodes); | |
5889 | return best_node; | |
5890 | } | |
5891 | ||
5892 | /** | |
5893 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5894 | * @node: node whose cpumask we're constructing | |
5895 | * @size: number of nodes to include in this span | |
5896 | * | |
5897 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5898 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5899 | * out optimally. | |
5900 | */ | |
5901 | static cpumask_t sched_domain_node_span(int node) | |
5902 | { | |
9c1cfda2 | 5903 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5904 | cpumask_t span, nodemask; |
5905 | int i; | |
9c1cfda2 JH |
5906 | |
5907 | cpus_clear(span); | |
5908 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5909 | ||
5910 | nodemask = node_to_cpumask(node); | |
5911 | cpus_or(span, span, nodemask); | |
5912 | set_bit(node, used_nodes); | |
5913 | ||
5914 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5915 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5916 | |
9c1cfda2 JH |
5917 | nodemask = node_to_cpumask(next_node); |
5918 | cpus_or(span, span, nodemask); | |
5919 | } | |
5920 | ||
5921 | return span; | |
5922 | } | |
5923 | #endif | |
5924 | ||
5c45bf27 | 5925 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5926 | |
9c1cfda2 | 5927 | /* |
48f24c4d | 5928 | * SMT sched-domains: |
9c1cfda2 | 5929 | */ |
1da177e4 LT |
5930 | #ifdef CONFIG_SCHED_SMT |
5931 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5932 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5933 | |
6711cab4 SS |
5934 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5935 | struct sched_group **sg) | |
1da177e4 | 5936 | { |
6711cab4 SS |
5937 | if (sg) |
5938 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5939 | return cpu; |
5940 | } | |
5941 | #endif | |
5942 | ||
48f24c4d IM |
5943 | /* |
5944 | * multi-core sched-domains: | |
5945 | */ | |
1e9f28fa SS |
5946 | #ifdef CONFIG_SCHED_MC |
5947 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5948 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5949 | #endif |
5950 | ||
5951 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5952 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5953 | struct sched_group **sg) | |
1e9f28fa | 5954 | { |
6711cab4 | 5955 | int group; |
d5a7430d | 5956 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5957 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
5958 | group = first_cpu(mask); |
5959 | if (sg) | |
5960 | *sg = &per_cpu(sched_group_core, group); | |
5961 | return group; | |
1e9f28fa SS |
5962 | } |
5963 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5964 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5965 | struct sched_group **sg) | |
1e9f28fa | 5966 | { |
6711cab4 SS |
5967 | if (sg) |
5968 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5969 | return cpu; |
5970 | } | |
5971 | #endif | |
5972 | ||
1da177e4 | 5973 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5974 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5975 | |
6711cab4 SS |
5976 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5977 | struct sched_group **sg) | |
1da177e4 | 5978 | { |
6711cab4 | 5979 | int group; |
48f24c4d | 5980 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5981 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5982 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5983 | group = first_cpu(mask); |
1e9f28fa | 5984 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 5985 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5986 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5987 | group = first_cpu(mask); |
1da177e4 | 5988 | #else |
6711cab4 | 5989 | group = cpu; |
1da177e4 | 5990 | #endif |
6711cab4 SS |
5991 | if (sg) |
5992 | *sg = &per_cpu(sched_group_phys, group); | |
5993 | return group; | |
1da177e4 LT |
5994 | } |
5995 | ||
5996 | #ifdef CONFIG_NUMA | |
1da177e4 | 5997 | /* |
9c1cfda2 JH |
5998 | * The init_sched_build_groups can't handle what we want to do with node |
5999 | * groups, so roll our own. Now each node has its own list of groups which | |
6000 | * gets dynamically allocated. | |
1da177e4 | 6001 | */ |
9c1cfda2 | 6002 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6003 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6004 | |
9c1cfda2 | 6005 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6006 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6007 | |
6711cab4 SS |
6008 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6009 | struct sched_group **sg) | |
9c1cfda2 | 6010 | { |
6711cab4 SS |
6011 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6012 | int group; | |
6013 | ||
6014 | cpus_and(nodemask, nodemask, *cpu_map); | |
6015 | group = first_cpu(nodemask); | |
6016 | ||
6017 | if (sg) | |
6018 | *sg = &per_cpu(sched_group_allnodes, group); | |
6019 | return group; | |
1da177e4 | 6020 | } |
6711cab4 | 6021 | |
08069033 SS |
6022 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6023 | { | |
6024 | struct sched_group *sg = group_head; | |
6025 | int j; | |
6026 | ||
6027 | if (!sg) | |
6028 | return; | |
3a5c359a AK |
6029 | do { |
6030 | for_each_cpu_mask(j, sg->cpumask) { | |
6031 | struct sched_domain *sd; | |
08069033 | 6032 | |
3a5c359a AK |
6033 | sd = &per_cpu(phys_domains, j); |
6034 | if (j != first_cpu(sd->groups->cpumask)) { | |
6035 | /* | |
6036 | * Only add "power" once for each | |
6037 | * physical package. | |
6038 | */ | |
6039 | continue; | |
6040 | } | |
08069033 | 6041 | |
3a5c359a AK |
6042 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6043 | } | |
6044 | sg = sg->next; | |
6045 | } while (sg != group_head); | |
08069033 | 6046 | } |
1da177e4 LT |
6047 | #endif |
6048 | ||
a616058b | 6049 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6050 | /* Free memory allocated for various sched_group structures */ |
6051 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6052 | { | |
a616058b | 6053 | int cpu, i; |
51888ca2 SV |
6054 | |
6055 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6056 | struct sched_group **sched_group_nodes |
6057 | = sched_group_nodes_bycpu[cpu]; | |
6058 | ||
51888ca2 SV |
6059 | if (!sched_group_nodes) |
6060 | continue; | |
6061 | ||
6062 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6063 | cpumask_t nodemask = node_to_cpumask(i); | |
6064 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6065 | ||
6066 | cpus_and(nodemask, nodemask, *cpu_map); | |
6067 | if (cpus_empty(nodemask)) | |
6068 | continue; | |
6069 | ||
6070 | if (sg == NULL) | |
6071 | continue; | |
6072 | sg = sg->next; | |
6073 | next_sg: | |
6074 | oldsg = sg; | |
6075 | sg = sg->next; | |
6076 | kfree(oldsg); | |
6077 | if (oldsg != sched_group_nodes[i]) | |
6078 | goto next_sg; | |
6079 | } | |
6080 | kfree(sched_group_nodes); | |
6081 | sched_group_nodes_bycpu[cpu] = NULL; | |
6082 | } | |
51888ca2 | 6083 | } |
a616058b SS |
6084 | #else |
6085 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6086 | { | |
6087 | } | |
6088 | #endif | |
51888ca2 | 6089 | |
89c4710e SS |
6090 | /* |
6091 | * Initialize sched groups cpu_power. | |
6092 | * | |
6093 | * cpu_power indicates the capacity of sched group, which is used while | |
6094 | * distributing the load between different sched groups in a sched domain. | |
6095 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6096 | * there are asymmetries in the topology. If there are asymmetries, group | |
6097 | * having more cpu_power will pickup more load compared to the group having | |
6098 | * less cpu_power. | |
6099 | * | |
6100 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6101 | * the maximum number of tasks a group can handle in the presence of other idle | |
6102 | * or lightly loaded groups in the same sched domain. | |
6103 | */ | |
6104 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6105 | { | |
6106 | struct sched_domain *child; | |
6107 | struct sched_group *group; | |
6108 | ||
6109 | WARN_ON(!sd || !sd->groups); | |
6110 | ||
6111 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6112 | return; | |
6113 | ||
6114 | child = sd->child; | |
6115 | ||
5517d86b ED |
6116 | sd->groups->__cpu_power = 0; |
6117 | ||
89c4710e SS |
6118 | /* |
6119 | * For perf policy, if the groups in child domain share resources | |
6120 | * (for example cores sharing some portions of the cache hierarchy | |
6121 | * or SMT), then set this domain groups cpu_power such that each group | |
6122 | * can handle only one task, when there are other idle groups in the | |
6123 | * same sched domain. | |
6124 | */ | |
6125 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6126 | (child->flags & | |
6127 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6128 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6129 | return; |
6130 | } | |
6131 | ||
89c4710e SS |
6132 | /* |
6133 | * add cpu_power of each child group to this groups cpu_power | |
6134 | */ | |
6135 | group = child->groups; | |
6136 | do { | |
5517d86b | 6137 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6138 | group = group->next; |
6139 | } while (group != child->groups); | |
6140 | } | |
6141 | ||
1da177e4 | 6142 | /* |
1a20ff27 DG |
6143 | * Build sched domains for a given set of cpus and attach the sched domains |
6144 | * to the individual cpus | |
1da177e4 | 6145 | */ |
51888ca2 | 6146 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6147 | { |
6148 | int i; | |
d1b55138 JH |
6149 | #ifdef CONFIG_NUMA |
6150 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6151 | int sd_allnodes = 0; |
d1b55138 JH |
6152 | |
6153 | /* | |
6154 | * Allocate the per-node list of sched groups | |
6155 | */ | |
5cf9f062 | 6156 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
d3a5aa98 | 6157 | GFP_KERNEL); |
d1b55138 JH |
6158 | if (!sched_group_nodes) { |
6159 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6160 | return -ENOMEM; |
d1b55138 JH |
6161 | } |
6162 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6163 | #endif | |
1da177e4 LT |
6164 | |
6165 | /* | |
1a20ff27 | 6166 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6167 | */ |
1a20ff27 | 6168 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6169 | struct sched_domain *sd = NULL, *p; |
6170 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6171 | ||
1a20ff27 | 6172 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6173 | |
6174 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6175 | if (cpus_weight(*cpu_map) > |
6176 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6177 | sd = &per_cpu(allnodes_domains, i); |
6178 | *sd = SD_ALLNODES_INIT; | |
6179 | sd->span = *cpu_map; | |
6711cab4 | 6180 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6181 | p = sd; |
6711cab4 | 6182 | sd_allnodes = 1; |
9c1cfda2 JH |
6183 | } else |
6184 | p = NULL; | |
6185 | ||
1da177e4 | 6186 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6187 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6188 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6189 | sd->parent = p; | |
1a848870 SS |
6190 | if (p) |
6191 | p->child = sd; | |
9c1cfda2 | 6192 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6193 | #endif |
6194 | ||
6195 | p = sd; | |
6196 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6197 | *sd = SD_CPU_INIT; |
6198 | sd->span = nodemask; | |
6199 | sd->parent = p; | |
1a848870 SS |
6200 | if (p) |
6201 | p->child = sd; | |
6711cab4 | 6202 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6203 | |
1e9f28fa SS |
6204 | #ifdef CONFIG_SCHED_MC |
6205 | p = sd; | |
6206 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6207 | *sd = SD_MC_INIT; |
6208 | sd->span = cpu_coregroup_map(i); | |
6209 | cpus_and(sd->span, sd->span, *cpu_map); | |
6210 | sd->parent = p; | |
1a848870 | 6211 | p->child = sd; |
6711cab4 | 6212 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6213 | #endif |
6214 | ||
1da177e4 LT |
6215 | #ifdef CONFIG_SCHED_SMT |
6216 | p = sd; | |
6217 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6218 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6219 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6220 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6221 | sd->parent = p; |
1a848870 | 6222 | p->child = sd; |
6711cab4 | 6223 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6224 | #endif |
6225 | } | |
6226 | ||
6227 | #ifdef CONFIG_SCHED_SMT | |
6228 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6229 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6230 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6231 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6232 | if (i != first_cpu(this_sibling_map)) |
6233 | continue; | |
6234 | ||
dd41f596 IM |
6235 | init_sched_build_groups(this_sibling_map, cpu_map, |
6236 | &cpu_to_cpu_group); | |
1da177e4 LT |
6237 | } |
6238 | #endif | |
6239 | ||
1e9f28fa SS |
6240 | #ifdef CONFIG_SCHED_MC |
6241 | /* Set up multi-core groups */ | |
6242 | for_each_cpu_mask(i, *cpu_map) { | |
6243 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6244 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6245 | if (i != first_cpu(this_core_map)) | |
6246 | continue; | |
dd41f596 IM |
6247 | init_sched_build_groups(this_core_map, cpu_map, |
6248 | &cpu_to_core_group); | |
1e9f28fa SS |
6249 | } |
6250 | #endif | |
6251 | ||
1da177e4 LT |
6252 | /* Set up physical groups */ |
6253 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6254 | cpumask_t nodemask = node_to_cpumask(i); | |
6255 | ||
1a20ff27 | 6256 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6257 | if (cpus_empty(nodemask)) |
6258 | continue; | |
6259 | ||
6711cab4 | 6260 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6261 | } |
6262 | ||
6263 | #ifdef CONFIG_NUMA | |
6264 | /* Set up node groups */ | |
6711cab4 | 6265 | if (sd_allnodes) |
dd41f596 IM |
6266 | init_sched_build_groups(*cpu_map, cpu_map, |
6267 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6268 | |
6269 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6270 | /* Set up node groups */ | |
6271 | struct sched_group *sg, *prev; | |
6272 | cpumask_t nodemask = node_to_cpumask(i); | |
6273 | cpumask_t domainspan; | |
6274 | cpumask_t covered = CPU_MASK_NONE; | |
6275 | int j; | |
6276 | ||
6277 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6278 | if (cpus_empty(nodemask)) { |
6279 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6280 | continue; |
d1b55138 | 6281 | } |
9c1cfda2 JH |
6282 | |
6283 | domainspan = sched_domain_node_span(i); | |
6284 | cpus_and(domainspan, domainspan, *cpu_map); | |
6285 | ||
15f0b676 | 6286 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6287 | if (!sg) { |
6288 | printk(KERN_WARNING "Can not alloc domain group for " | |
6289 | "node %d\n", i); | |
6290 | goto error; | |
6291 | } | |
9c1cfda2 JH |
6292 | sched_group_nodes[i] = sg; |
6293 | for_each_cpu_mask(j, nodemask) { | |
6294 | struct sched_domain *sd; | |
9761eea8 | 6295 | |
9c1cfda2 JH |
6296 | sd = &per_cpu(node_domains, j); |
6297 | sd->groups = sg; | |
9c1cfda2 | 6298 | } |
5517d86b | 6299 | sg->__cpu_power = 0; |
9c1cfda2 | 6300 | sg->cpumask = nodemask; |
51888ca2 | 6301 | sg->next = sg; |
9c1cfda2 JH |
6302 | cpus_or(covered, covered, nodemask); |
6303 | prev = sg; | |
6304 | ||
6305 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6306 | cpumask_t tmp, notcovered; | |
6307 | int n = (i + j) % MAX_NUMNODES; | |
6308 | ||
6309 | cpus_complement(notcovered, covered); | |
6310 | cpus_and(tmp, notcovered, *cpu_map); | |
6311 | cpus_and(tmp, tmp, domainspan); | |
6312 | if (cpus_empty(tmp)) | |
6313 | break; | |
6314 | ||
6315 | nodemask = node_to_cpumask(n); | |
6316 | cpus_and(tmp, tmp, nodemask); | |
6317 | if (cpus_empty(tmp)) | |
6318 | continue; | |
6319 | ||
15f0b676 SV |
6320 | sg = kmalloc_node(sizeof(struct sched_group), |
6321 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6322 | if (!sg) { |
6323 | printk(KERN_WARNING | |
6324 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6325 | goto error; |
9c1cfda2 | 6326 | } |
5517d86b | 6327 | sg->__cpu_power = 0; |
9c1cfda2 | 6328 | sg->cpumask = tmp; |
51888ca2 | 6329 | sg->next = prev->next; |
9c1cfda2 JH |
6330 | cpus_or(covered, covered, tmp); |
6331 | prev->next = sg; | |
6332 | prev = sg; | |
6333 | } | |
9c1cfda2 | 6334 | } |
1da177e4 LT |
6335 | #endif |
6336 | ||
6337 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6338 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6339 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6340 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6341 | ||
89c4710e | 6342 | init_sched_groups_power(i, sd); |
5c45bf27 | 6343 | } |
1da177e4 | 6344 | #endif |
1e9f28fa | 6345 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6346 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6347 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6348 | ||
89c4710e | 6349 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6350 | } |
6351 | #endif | |
1e9f28fa | 6352 | |
5c45bf27 | 6353 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6354 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6355 | ||
89c4710e | 6356 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6357 | } |
6358 | ||
9c1cfda2 | 6359 | #ifdef CONFIG_NUMA |
08069033 SS |
6360 | for (i = 0; i < MAX_NUMNODES; i++) |
6361 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6362 | |
6711cab4 SS |
6363 | if (sd_allnodes) { |
6364 | struct sched_group *sg; | |
f712c0c7 | 6365 | |
6711cab4 | 6366 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6367 | init_numa_sched_groups_power(sg); |
6368 | } | |
9c1cfda2 JH |
6369 | #endif |
6370 | ||
1da177e4 | 6371 | /* Attach the domains */ |
1a20ff27 | 6372 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6373 | struct sched_domain *sd; |
6374 | #ifdef CONFIG_SCHED_SMT | |
6375 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6376 | #elif defined(CONFIG_SCHED_MC) |
6377 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6378 | #else |
6379 | sd = &per_cpu(phys_domains, i); | |
6380 | #endif | |
6381 | cpu_attach_domain(sd, i); | |
6382 | } | |
51888ca2 SV |
6383 | |
6384 | return 0; | |
6385 | ||
a616058b | 6386 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6387 | error: |
6388 | free_sched_groups(cpu_map); | |
6389 | return -ENOMEM; | |
a616058b | 6390 | #endif |
1da177e4 | 6391 | } |
029190c5 PJ |
6392 | |
6393 | static cpumask_t *doms_cur; /* current sched domains */ | |
6394 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6395 | ||
6396 | /* | |
6397 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6398 | * cpumask_t) fails, then fallback to a single sched domain, | |
6399 | * as determined by the single cpumask_t fallback_doms. | |
6400 | */ | |
6401 | static cpumask_t fallback_doms; | |
6402 | ||
1a20ff27 DG |
6403 | /* |
6404 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
029190c5 PJ |
6405 | * For now this just excludes isolated cpus, but could be used to |
6406 | * exclude other special cases in the future. | |
1a20ff27 | 6407 | */ |
51888ca2 | 6408 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6409 | { |
029190c5 PJ |
6410 | ndoms_cur = 1; |
6411 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6412 | if (!doms_cur) | |
6413 | doms_cur = &fallback_doms; | |
6414 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
6382bc90 | 6415 | register_sched_domain_sysctl(); |
029190c5 | 6416 | return build_sched_domains(doms_cur); |
1a20ff27 DG |
6417 | } |
6418 | ||
6419 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6420 | { |
51888ca2 | 6421 | free_sched_groups(cpu_map); |
9c1cfda2 | 6422 | } |
1da177e4 | 6423 | |
1a20ff27 DG |
6424 | /* |
6425 | * Detach sched domains from a group of cpus specified in cpu_map | |
6426 | * These cpus will now be attached to the NULL domain | |
6427 | */ | |
858119e1 | 6428 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6429 | { |
6430 | int i; | |
6431 | ||
6382bc90 MM |
6432 | unregister_sched_domain_sysctl(); |
6433 | ||
1a20ff27 DG |
6434 | for_each_cpu_mask(i, *cpu_map) |
6435 | cpu_attach_domain(NULL, i); | |
6436 | synchronize_sched(); | |
6437 | arch_destroy_sched_domains(cpu_map); | |
6438 | } | |
6439 | ||
029190c5 PJ |
6440 | /* |
6441 | * Partition sched domains as specified by the 'ndoms_new' | |
6442 | * cpumasks in the array doms_new[] of cpumasks. This compares | |
6443 | * doms_new[] to the current sched domain partitioning, doms_cur[]. | |
6444 | * It destroys each deleted domain and builds each new domain. | |
6445 | * | |
6446 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
6447 | * The masks don't intersect (don't overlap.) We should setup one | |
6448 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6449 | * not be load balanced. If the same cpumask appears both in the | |
6450 | * current 'doms_cur' domains and in the new 'doms_new', we can leave | |
6451 | * it as it is. | |
6452 | * | |
6453 | * The passed in 'doms_new' should be kmalloc'd. This routine takes | |
6454 | * ownership of it and will kfree it when done with it. If the caller | |
6455 | * failed the kmalloc call, then it can pass in doms_new == NULL, | |
6456 | * and partition_sched_domains() will fallback to the single partition | |
6457 | * 'fallback_doms'. | |
6458 | * | |
6459 | * Call with hotplug lock held | |
6460 | */ | |
6461 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6462 | { | |
6463 | int i, j; | |
6464 | ||
6465 | if (doms_new == NULL) { | |
6466 | ndoms_new = 1; | |
6467 | doms_new = &fallback_doms; | |
6468 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6469 | } | |
6470 | ||
6471 | /* Destroy deleted domains */ | |
6472 | for (i = 0; i < ndoms_cur; i++) { | |
6473 | for (j = 0; j < ndoms_new; j++) { | |
6474 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6475 | goto match1; | |
6476 | } | |
6477 | /* no match - a current sched domain not in new doms_new[] */ | |
6478 | detach_destroy_domains(doms_cur + i); | |
6479 | match1: | |
6480 | ; | |
6481 | } | |
6482 | ||
6483 | /* Build new domains */ | |
6484 | for (i = 0; i < ndoms_new; i++) { | |
6485 | for (j = 0; j < ndoms_cur; j++) { | |
6486 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6487 | goto match2; | |
6488 | } | |
6489 | /* no match - add a new doms_new */ | |
6490 | build_sched_domains(doms_new + i); | |
6491 | match2: | |
6492 | ; | |
6493 | } | |
6494 | ||
6495 | /* Remember the new sched domains */ | |
6496 | if (doms_cur != &fallback_doms) | |
6497 | kfree(doms_cur); | |
6498 | doms_cur = doms_new; | |
6499 | ndoms_cur = ndoms_new; | |
6500 | } | |
6501 | ||
5c45bf27 | 6502 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6503 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6504 | { |
6505 | int err; | |
6506 | ||
5be9361c | 6507 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6508 | detach_destroy_domains(&cpu_online_map); |
6509 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6510 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6511 | |
6512 | return err; | |
6513 | } | |
6514 | ||
6515 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6516 | { | |
6517 | int ret; | |
6518 | ||
6519 | if (buf[0] != '0' && buf[0] != '1') | |
6520 | return -EINVAL; | |
6521 | ||
6522 | if (smt) | |
6523 | sched_smt_power_savings = (buf[0] == '1'); | |
6524 | else | |
6525 | sched_mc_power_savings = (buf[0] == '1'); | |
6526 | ||
6527 | ret = arch_reinit_sched_domains(); | |
6528 | ||
6529 | return ret ? ret : count; | |
6530 | } | |
6531 | ||
5c45bf27 SS |
6532 | #ifdef CONFIG_SCHED_MC |
6533 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6534 | { | |
6535 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6536 | } | |
48f24c4d IM |
6537 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6538 | const char *buf, size_t count) | |
5c45bf27 SS |
6539 | { |
6540 | return sched_power_savings_store(buf, count, 0); | |
6541 | } | |
6707de00 AB |
6542 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6543 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6544 | #endif |
6545 | ||
6546 | #ifdef CONFIG_SCHED_SMT | |
6547 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6548 | { | |
6549 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6550 | } | |
48f24c4d IM |
6551 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6552 | const char *buf, size_t count) | |
5c45bf27 SS |
6553 | { |
6554 | return sched_power_savings_store(buf, count, 1); | |
6555 | } | |
6707de00 AB |
6556 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6557 | sched_smt_power_savings_store); | |
6558 | #endif | |
6559 | ||
6560 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6561 | { | |
6562 | int err = 0; | |
6563 | ||
6564 | #ifdef CONFIG_SCHED_SMT | |
6565 | if (smt_capable()) | |
6566 | err = sysfs_create_file(&cls->kset.kobj, | |
6567 | &attr_sched_smt_power_savings.attr); | |
6568 | #endif | |
6569 | #ifdef CONFIG_SCHED_MC | |
6570 | if (!err && mc_capable()) | |
6571 | err = sysfs_create_file(&cls->kset.kobj, | |
6572 | &attr_sched_mc_power_savings.attr); | |
6573 | #endif | |
6574 | return err; | |
6575 | } | |
5c45bf27 SS |
6576 | #endif |
6577 | ||
1da177e4 LT |
6578 | /* |
6579 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6580 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6581 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6582 | * which will prevent rebalancing while the sched domains are recalculated. |
6583 | */ | |
6584 | static int update_sched_domains(struct notifier_block *nfb, | |
6585 | unsigned long action, void *hcpu) | |
6586 | { | |
1da177e4 LT |
6587 | switch (action) { |
6588 | case CPU_UP_PREPARE: | |
8bb78442 | 6589 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6590 | case CPU_DOWN_PREPARE: |
8bb78442 | 6591 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6592 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6593 | return NOTIFY_OK; |
6594 | ||
6595 | case CPU_UP_CANCELED: | |
8bb78442 | 6596 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6597 | case CPU_DOWN_FAILED: |
8bb78442 | 6598 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6599 | case CPU_ONLINE: |
8bb78442 | 6600 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6601 | case CPU_DEAD: |
8bb78442 | 6602 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6603 | /* |
6604 | * Fall through and re-initialise the domains. | |
6605 | */ | |
6606 | break; | |
6607 | default: | |
6608 | return NOTIFY_DONE; | |
6609 | } | |
6610 | ||
6611 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6612 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6613 | |
6614 | return NOTIFY_OK; | |
6615 | } | |
1da177e4 LT |
6616 | |
6617 | void __init sched_init_smp(void) | |
6618 | { | |
5c1e1767 NP |
6619 | cpumask_t non_isolated_cpus; |
6620 | ||
5be9361c | 6621 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6622 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6623 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6624 | if (cpus_empty(non_isolated_cpus)) |
6625 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6626 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6627 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6628 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6629 | |
6630 | /* Move init over to a non-isolated CPU */ | |
6631 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6632 | BUG(); | |
1da177e4 LT |
6633 | } |
6634 | #else | |
6635 | void __init sched_init_smp(void) | |
6636 | { | |
6637 | } | |
6638 | #endif /* CONFIG_SMP */ | |
6639 | ||
6640 | int in_sched_functions(unsigned long addr) | |
6641 | { | |
6642 | /* Linker adds these: start and end of __sched functions */ | |
6643 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6644 | |
1da177e4 LT |
6645 | return in_lock_functions(addr) || |
6646 | (addr >= (unsigned long)__sched_text_start | |
6647 | && addr < (unsigned long)__sched_text_end); | |
6648 | } | |
6649 | ||
a9957449 | 6650 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6651 | { |
6652 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6653 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6654 | cfs_rq->rq = rq; | |
6655 | #endif | |
67e9fb2a | 6656 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6657 | } |
6658 | ||
1da177e4 LT |
6659 | void __init sched_init(void) |
6660 | { | |
476f3534 | 6661 | int highest_cpu = 0; |
dd41f596 IM |
6662 | int i, j; |
6663 | ||
0a945022 | 6664 | for_each_possible_cpu(i) { |
dd41f596 | 6665 | struct rt_prio_array *array; |
70b97a7f | 6666 | struct rq *rq; |
1da177e4 LT |
6667 | |
6668 | rq = cpu_rq(i); | |
6669 | spin_lock_init(&rq->lock); | |
fcb99371 | 6670 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6671 | rq->nr_running = 0; |
dd41f596 IM |
6672 | rq->clock = 1; |
6673 | init_cfs_rq(&rq->cfs, rq); | |
6674 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6675 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6676 | { |
6677 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6678 | struct sched_entity *se = | |
6679 | &per_cpu(init_sched_entity, i); | |
6680 | ||
6681 | init_cfs_rq_p[i] = cfs_rq; | |
6682 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6683 | cfs_rq->tg = &init_task_group; |
3a252015 | 6684 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6685 | &rq->leaf_cfs_rq_list); |
6686 | ||
3a252015 IM |
6687 | init_sched_entity_p[i] = se; |
6688 | se->cfs_rq = &rq->cfs; | |
6689 | se->my_q = cfs_rq; | |
4cf86d77 | 6690 | se->load.weight = init_task_group_load; |
9b5b7751 | 6691 | se->load.inv_weight = |
4cf86d77 | 6692 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6693 | se->parent = NULL; |
6694 | } | |
4cf86d77 | 6695 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6696 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6697 | #endif |
1da177e4 | 6698 | |
dd41f596 IM |
6699 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6700 | rq->cpu_load[j] = 0; | |
1da177e4 | 6701 | #ifdef CONFIG_SMP |
41c7ce9a | 6702 | rq->sd = NULL; |
1da177e4 | 6703 | rq->active_balance = 0; |
dd41f596 | 6704 | rq->next_balance = jiffies; |
1da177e4 | 6705 | rq->push_cpu = 0; |
0a2966b4 | 6706 | rq->cpu = i; |
1da177e4 LT |
6707 | rq->migration_thread = NULL; |
6708 | INIT_LIST_HEAD(&rq->migration_queue); | |
6709 | #endif | |
6710 | atomic_set(&rq->nr_iowait, 0); | |
6711 | ||
dd41f596 IM |
6712 | array = &rq->rt.active; |
6713 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6714 | INIT_LIST_HEAD(array->queue + j); | |
6715 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6716 | } |
476f3534 | 6717 | highest_cpu = i; |
dd41f596 IM |
6718 | /* delimiter for bitsearch: */ |
6719 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6720 | } |
6721 | ||
2dd73a4f | 6722 | set_load_weight(&init_task); |
b50f60ce | 6723 | |
e107be36 AK |
6724 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6725 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6726 | #endif | |
6727 | ||
c9819f45 | 6728 | #ifdef CONFIG_SMP |
476f3534 | 6729 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6730 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6731 | #endif | |
6732 | ||
b50f60ce HC |
6733 | #ifdef CONFIG_RT_MUTEXES |
6734 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6735 | #endif | |
6736 | ||
1da177e4 LT |
6737 | /* |
6738 | * The boot idle thread does lazy MMU switching as well: | |
6739 | */ | |
6740 | atomic_inc(&init_mm.mm_count); | |
6741 | enter_lazy_tlb(&init_mm, current); | |
6742 | ||
6743 | /* | |
6744 | * Make us the idle thread. Technically, schedule() should not be | |
6745 | * called from this thread, however somewhere below it might be, | |
6746 | * but because we are the idle thread, we just pick up running again | |
6747 | * when this runqueue becomes "idle". | |
6748 | */ | |
6749 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6750 | /* |
6751 | * During early bootup we pretend to be a normal task: | |
6752 | */ | |
6753 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6754 | } |
6755 | ||
6756 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6757 | void __might_sleep(char *file, int line) | |
6758 | { | |
48f24c4d | 6759 | #ifdef in_atomic |
1da177e4 LT |
6760 | static unsigned long prev_jiffy; /* ratelimiting */ |
6761 | ||
6762 | if ((in_atomic() || irqs_disabled()) && | |
6763 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6764 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6765 | return; | |
6766 | prev_jiffy = jiffies; | |
91368d73 | 6767 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6768 | " context at %s:%d\n", file, line); |
6769 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6770 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6771 | debug_show_held_locks(current); |
3117df04 IM |
6772 | if (irqs_disabled()) |
6773 | print_irqtrace_events(current); | |
1da177e4 LT |
6774 | dump_stack(); |
6775 | } | |
6776 | #endif | |
6777 | } | |
6778 | EXPORT_SYMBOL(__might_sleep); | |
6779 | #endif | |
6780 | ||
6781 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6782 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6783 | { | |
6784 | int on_rq; | |
6785 | update_rq_clock(rq); | |
6786 | on_rq = p->se.on_rq; | |
6787 | if (on_rq) | |
6788 | deactivate_task(rq, p, 0); | |
6789 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6790 | if (on_rq) { | |
6791 | activate_task(rq, p, 0); | |
6792 | resched_task(rq->curr); | |
6793 | } | |
6794 | } | |
6795 | ||
1da177e4 LT |
6796 | void normalize_rt_tasks(void) |
6797 | { | |
a0f98a1c | 6798 | struct task_struct *g, *p; |
1da177e4 | 6799 | unsigned long flags; |
70b97a7f | 6800 | struct rq *rq; |
1da177e4 LT |
6801 | |
6802 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6803 | do_each_thread(g, p) { |
178be793 IM |
6804 | /* |
6805 | * Only normalize user tasks: | |
6806 | */ | |
6807 | if (!p->mm) | |
6808 | continue; | |
6809 | ||
6cfb0d5d | 6810 | p->se.exec_start = 0; |
6cfb0d5d | 6811 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6812 | p->se.wait_start = 0; |
dd41f596 | 6813 | p->se.sleep_start = 0; |
dd41f596 | 6814 | p->se.block_start = 0; |
6cfb0d5d | 6815 | #endif |
dd41f596 IM |
6816 | task_rq(p)->clock = 0; |
6817 | ||
6818 | if (!rt_task(p)) { | |
6819 | /* | |
6820 | * Renice negative nice level userspace | |
6821 | * tasks back to 0: | |
6822 | */ | |
6823 | if (TASK_NICE(p) < 0 && p->mm) | |
6824 | set_user_nice(p, 0); | |
1da177e4 | 6825 | continue; |
dd41f596 | 6826 | } |
1da177e4 | 6827 | |
b29739f9 IM |
6828 | spin_lock_irqsave(&p->pi_lock, flags); |
6829 | rq = __task_rq_lock(p); | |
1da177e4 | 6830 | |
178be793 | 6831 | normalize_task(rq, p); |
3a5e4dc1 | 6832 | |
b29739f9 IM |
6833 | __task_rq_unlock(rq); |
6834 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6835 | } while_each_thread(g, p); |
6836 | ||
1da177e4 LT |
6837 | read_unlock_irq(&tasklist_lock); |
6838 | } | |
6839 | ||
6840 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6841 | |
6842 | #ifdef CONFIG_IA64 | |
6843 | /* | |
6844 | * These functions are only useful for the IA64 MCA handling. | |
6845 | * | |
6846 | * They can only be called when the whole system has been | |
6847 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6848 | * activity can take place. Using them for anything else would | |
6849 | * be a serious bug, and as a result, they aren't even visible | |
6850 | * under any other configuration. | |
6851 | */ | |
6852 | ||
6853 | /** | |
6854 | * curr_task - return the current task for a given cpu. | |
6855 | * @cpu: the processor in question. | |
6856 | * | |
6857 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6858 | */ | |
36c8b586 | 6859 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6860 | { |
6861 | return cpu_curr(cpu); | |
6862 | } | |
6863 | ||
6864 | /** | |
6865 | * set_curr_task - set the current task for a given cpu. | |
6866 | * @cpu: the processor in question. | |
6867 | * @p: the task pointer to set. | |
6868 | * | |
6869 | * Description: This function must only be used when non-maskable interrupts | |
6870 | * are serviced on a separate stack. It allows the architecture to switch the | |
6871 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6872 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6873 | * and caller must save the original value of the current task (see | |
6874 | * curr_task() above) and restore that value before reenabling interrupts and | |
6875 | * re-starting the system. | |
6876 | * | |
6877 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6878 | */ | |
36c8b586 | 6879 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6880 | { |
6881 | cpu_curr(cpu) = p; | |
6882 | } | |
6883 | ||
6884 | #endif | |
29f59db3 SV |
6885 | |
6886 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6887 | ||
29f59db3 | 6888 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6889 | struct task_group *sched_create_group(void) |
29f59db3 | 6890 | { |
4cf86d77 | 6891 | struct task_group *tg; |
29f59db3 SV |
6892 | struct cfs_rq *cfs_rq; |
6893 | struct sched_entity *se; | |
9b5b7751 | 6894 | struct rq *rq; |
29f59db3 SV |
6895 | int i; |
6896 | ||
29f59db3 SV |
6897 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6898 | if (!tg) | |
6899 | return ERR_PTR(-ENOMEM); | |
6900 | ||
9b5b7751 | 6901 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6902 | if (!tg->cfs_rq) |
6903 | goto err; | |
9b5b7751 | 6904 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6905 | if (!tg->se) |
6906 | goto err; | |
6907 | ||
6908 | for_each_possible_cpu(i) { | |
9b5b7751 | 6909 | rq = cpu_rq(i); |
29f59db3 SV |
6910 | |
6911 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6912 | cpu_to_node(i)); | |
6913 | if (!cfs_rq) | |
6914 | goto err; | |
6915 | ||
6916 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6917 | cpu_to_node(i)); | |
6918 | if (!se) | |
6919 | goto err; | |
6920 | ||
6921 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6922 | memset(se, 0, sizeof(struct sched_entity)); | |
6923 | ||
6924 | tg->cfs_rq[i] = cfs_rq; | |
6925 | init_cfs_rq(cfs_rq, rq); | |
6926 | cfs_rq->tg = tg; | |
29f59db3 SV |
6927 | |
6928 | tg->se[i] = se; | |
6929 | se->cfs_rq = &rq->cfs; | |
6930 | se->my_q = cfs_rq; | |
6931 | se->load.weight = NICE_0_LOAD; | |
6932 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6933 | se->parent = NULL; | |
6934 | } | |
6935 | ||
9b5b7751 SV |
6936 | for_each_possible_cpu(i) { |
6937 | rq = cpu_rq(i); | |
6938 | cfs_rq = tg->cfs_rq[i]; | |
6939 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6940 | } | |
29f59db3 | 6941 | |
9b5b7751 | 6942 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6943 | spin_lock_init(&tg->lock); |
29f59db3 | 6944 | |
9b5b7751 | 6945 | return tg; |
29f59db3 SV |
6946 | |
6947 | err: | |
6948 | for_each_possible_cpu(i) { | |
a65914b3 | 6949 | if (tg->cfs_rq) |
29f59db3 | 6950 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6951 | if (tg->se) |
29f59db3 SV |
6952 | kfree(tg->se[i]); |
6953 | } | |
a65914b3 IM |
6954 | kfree(tg->cfs_rq); |
6955 | kfree(tg->se); | |
6956 | kfree(tg); | |
29f59db3 SV |
6957 | |
6958 | return ERR_PTR(-ENOMEM); | |
6959 | } | |
6960 | ||
9b5b7751 SV |
6961 | /* rcu callback to free various structures associated with a task group */ |
6962 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6963 | { |
9b5b7751 | 6964 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6965 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6966 | struct sched_entity *se; |
6967 | int i; | |
6968 | ||
29f59db3 SV |
6969 | /* now it should be safe to free those cfs_rqs */ |
6970 | for_each_possible_cpu(i) { | |
6971 | cfs_rq = tg->cfs_rq[i]; | |
6972 | kfree(cfs_rq); | |
6973 | ||
6974 | se = tg->se[i]; | |
6975 | kfree(se); | |
6976 | } | |
6977 | ||
6978 | kfree(tg->cfs_rq); | |
6979 | kfree(tg->se); | |
6980 | kfree(tg); | |
6981 | } | |
6982 | ||
9b5b7751 | 6983 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6984 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 6985 | { |
9b5b7751 SV |
6986 | struct cfs_rq *cfs_rq; |
6987 | int i; | |
29f59db3 | 6988 | |
9b5b7751 SV |
6989 | for_each_possible_cpu(i) { |
6990 | cfs_rq = tg->cfs_rq[i]; | |
6991 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
6992 | } | |
6993 | ||
6994 | cfs_rq = tg->cfs_rq[0]; | |
6995 | ||
6996 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6997 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
6998 | } |
6999 | ||
9b5b7751 | 7000 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7001 | * The caller of this function should have put the task in its new group |
7002 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7003 | * reflect its new group. | |
9b5b7751 SV |
7004 | */ |
7005 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7006 | { |
7007 | int on_rq, running; | |
7008 | unsigned long flags; | |
7009 | struct rq *rq; | |
7010 | ||
7011 | rq = task_rq_lock(tsk, &flags); | |
7012 | ||
7013 | if (tsk->sched_class != &fair_sched_class) | |
7014 | goto done; | |
7015 | ||
7016 | update_rq_clock(rq); | |
7017 | ||
7018 | running = task_running(rq, tsk); | |
7019 | on_rq = tsk->se.on_rq; | |
7020 | ||
83b699ed | 7021 | if (on_rq) { |
29f59db3 | 7022 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7023 | if (unlikely(running)) |
7024 | tsk->sched_class->put_prev_task(rq, tsk); | |
7025 | } | |
29f59db3 SV |
7026 | |
7027 | set_task_cfs_rq(tsk); | |
7028 | ||
83b699ed SV |
7029 | if (on_rq) { |
7030 | if (unlikely(running)) | |
7031 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7032 | enqueue_task(rq, tsk, 0); |
83b699ed | 7033 | } |
29f59db3 SV |
7034 | |
7035 | done: | |
7036 | task_rq_unlock(rq, &flags); | |
7037 | } | |
7038 | ||
7039 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
7040 | { | |
7041 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7042 | struct rq *rq = cfs_rq->rq; | |
7043 | int on_rq; | |
7044 | ||
7045 | spin_lock_irq(&rq->lock); | |
7046 | ||
7047 | on_rq = se->on_rq; | |
7048 | if (on_rq) | |
7049 | dequeue_entity(cfs_rq, se, 0); | |
7050 | ||
7051 | se->load.weight = shares; | |
7052 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7053 | ||
7054 | if (on_rq) | |
7055 | enqueue_entity(cfs_rq, se, 0); | |
7056 | ||
7057 | spin_unlock_irq(&rq->lock); | |
7058 | } | |
7059 | ||
4cf86d77 | 7060 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7061 | { |
7062 | int i; | |
29f59db3 | 7063 | |
5cb350ba | 7064 | spin_lock(&tg->lock); |
9b5b7751 | 7065 | if (tg->shares == shares) |
5cb350ba | 7066 | goto done; |
29f59db3 | 7067 | |
9b5b7751 | 7068 | tg->shares = shares; |
29f59db3 | 7069 | for_each_possible_cpu(i) |
9b5b7751 | 7070 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7071 | |
5cb350ba DG |
7072 | done: |
7073 | spin_unlock(&tg->lock); | |
9b5b7751 | 7074 | return 0; |
29f59db3 SV |
7075 | } |
7076 | ||
5cb350ba DG |
7077 | unsigned long sched_group_shares(struct task_group *tg) |
7078 | { | |
7079 | return tg->shares; | |
7080 | } | |
7081 | ||
3a252015 | 7082 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |