Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
1da177e4 LT |
25 | */ |
26 | ||
27 | #include <linux/mm.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/nmi.h> | |
30 | #include <linux/init.h> | |
dff06c15 | 31 | #include <linux/uaccess.h> |
1da177e4 LT |
32 | #include <linux/highmem.h> |
33 | #include <linux/smp_lock.h> | |
34 | #include <asm/mmu_context.h> | |
35 | #include <linux/interrupt.h> | |
c59ede7b | 36 | #include <linux/capability.h> |
1da177e4 LT |
37 | #include <linux/completion.h> |
38 | #include <linux/kernel_stat.h> | |
9a11b49a | 39 | #include <linux/debug_locks.h> |
1da177e4 LT |
40 | #include <linux/security.h> |
41 | #include <linux/notifier.h> | |
42 | #include <linux/profile.h> | |
7dfb7103 | 43 | #include <linux/freezer.h> |
198e2f18 | 44 | #include <linux/vmalloc.h> |
1da177e4 LT |
45 | #include <linux/blkdev.h> |
46 | #include <linux/delay.h> | |
47 | #include <linux/smp.h> | |
48 | #include <linux/threads.h> | |
49 | #include <linux/timer.h> | |
50 | #include <linux/rcupdate.h> | |
51 | #include <linux/cpu.h> | |
52 | #include <linux/cpuset.h> | |
53 | #include <linux/percpu.h> | |
54 | #include <linux/kthread.h> | |
55 | #include <linux/seq_file.h> | |
e692ab53 | 56 | #include <linux/sysctl.h> |
1da177e4 LT |
57 | #include <linux/syscalls.h> |
58 | #include <linux/times.h> | |
8f0ab514 | 59 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 60 | #include <linux/kprobes.h> |
0ff92245 | 61 | #include <linux/delayacct.h> |
5517d86b | 62 | #include <linux/reciprocal_div.h> |
dff06c15 | 63 | #include <linux/unistd.h> |
f5ff8422 | 64 | #include <linux/pagemap.h> |
1da177e4 | 65 | |
5517d86b | 66 | #include <asm/tlb.h> |
1da177e4 | 67 | |
b035b6de AD |
68 | /* |
69 | * Scheduler clock - returns current time in nanosec units. | |
70 | * This is default implementation. | |
71 | * Architectures and sub-architectures can override this. | |
72 | */ | |
73 | unsigned long long __attribute__((weak)) sched_clock(void) | |
74 | { | |
75 | return (unsigned long long)jiffies * (1000000000 / HZ); | |
76 | } | |
77 | ||
1da177e4 LT |
78 | /* |
79 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
80 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
81 | * and back. | |
82 | */ | |
83 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
84 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
85 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
86 | ||
87 | /* | |
88 | * 'User priority' is the nice value converted to something we | |
89 | * can work with better when scaling various scheduler parameters, | |
90 | * it's a [ 0 ... 39 ] range. | |
91 | */ | |
92 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
93 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
94 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
95 | ||
96 | /* | |
97 | * Some helpers for converting nanosecond timing to jiffy resolution | |
98 | */ | |
a4ec24b4 | 99 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (1000000000 / HZ)) |
1da177e4 LT |
100 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
101 | ||
6aa645ea IM |
102 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
103 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
104 | ||
1da177e4 LT |
105 | /* |
106 | * These are the 'tuning knobs' of the scheduler: | |
107 | * | |
a4ec24b4 | 108 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
109 | * Timeslices get refilled after they expire. |
110 | */ | |
1da177e4 | 111 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 112 | |
5517d86b ED |
113 | #ifdef CONFIG_SMP |
114 | /* | |
115 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
116 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
117 | */ | |
118 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
119 | { | |
120 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
121 | } | |
122 | ||
123 | /* | |
124 | * Each time a sched group cpu_power is changed, | |
125 | * we must compute its reciprocal value | |
126 | */ | |
127 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
128 | { | |
129 | sg->__cpu_power += val; | |
130 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
131 | } | |
132 | #endif | |
133 | ||
e05606d3 IM |
134 | static inline int rt_policy(int policy) |
135 | { | |
136 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
137 | return 1; | |
138 | return 0; | |
139 | } | |
140 | ||
141 | static inline int task_has_rt_policy(struct task_struct *p) | |
142 | { | |
143 | return rt_policy(p->policy); | |
144 | } | |
145 | ||
1da177e4 | 146 | /* |
6aa645ea | 147 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 148 | */ |
6aa645ea IM |
149 | struct rt_prio_array { |
150 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
151 | struct list_head queue[MAX_RT_PRIO]; | |
152 | }; | |
153 | ||
29f59db3 SV |
154 | #ifdef CONFIG_FAIR_GROUP_SCHED |
155 | ||
29f59db3 SV |
156 | struct cfs_rq; |
157 | ||
158 | /* task group related information */ | |
4cf86d77 | 159 | struct task_group { |
29f59db3 SV |
160 | /* schedulable entities of this group on each cpu */ |
161 | struct sched_entity **se; | |
162 | /* runqueue "owned" by this group on each cpu */ | |
163 | struct cfs_rq **cfs_rq; | |
164 | unsigned long shares; | |
5cb350ba DG |
165 | /* spinlock to serialize modification to shares */ |
166 | spinlock_t lock; | |
29f59db3 SV |
167 | }; |
168 | ||
169 | /* Default task group's sched entity on each cpu */ | |
170 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
171 | /* Default task group's cfs_rq on each cpu */ | |
172 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
173 | ||
9b5b7751 SV |
174 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
175 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
176 | |
177 | /* Default task group. | |
3a252015 | 178 | * Every task in system belong to this group at bootup. |
29f59db3 | 179 | */ |
4cf86d77 | 180 | struct task_group init_task_group = { |
3a252015 IM |
181 | .se = init_sched_entity_p, |
182 | .cfs_rq = init_cfs_rq_p, | |
183 | }; | |
9b5b7751 | 184 | |
24e377a8 | 185 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 186 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 187 | #else |
3a252015 | 188 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
189 | #endif |
190 | ||
4cf86d77 | 191 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
192 | |
193 | /* return group to which a task belongs */ | |
4cf86d77 | 194 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 195 | { |
4cf86d77 | 196 | struct task_group *tg; |
9b5b7751 | 197 | |
24e377a8 SV |
198 | #ifdef CONFIG_FAIR_USER_SCHED |
199 | tg = p->user->tg; | |
200 | #else | |
4cf86d77 | 201 | tg = &init_task_group; |
24e377a8 | 202 | #endif |
9b5b7751 SV |
203 | |
204 | return tg; | |
29f59db3 SV |
205 | } |
206 | ||
207 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
208 | static inline void set_task_cfs_rq(struct task_struct *p) | |
209 | { | |
4cf86d77 IM |
210 | p->se.cfs_rq = task_group(p)->cfs_rq[task_cpu(p)]; |
211 | p->se.parent = task_group(p)->se[task_cpu(p)]; | |
29f59db3 SV |
212 | } |
213 | ||
214 | #else | |
215 | ||
216 | static inline void set_task_cfs_rq(struct task_struct *p) { } | |
217 | ||
218 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
219 | ||
6aa645ea IM |
220 | /* CFS-related fields in a runqueue */ |
221 | struct cfs_rq { | |
222 | struct load_weight load; | |
223 | unsigned long nr_running; | |
224 | ||
6aa645ea | 225 | u64 exec_clock; |
e9acbff6 | 226 | u64 min_vruntime; |
6aa645ea IM |
227 | |
228 | struct rb_root tasks_timeline; | |
229 | struct rb_node *rb_leftmost; | |
230 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
231 | /* 'curr' points to currently running entity on this cfs_rq. |
232 | * It is set to NULL otherwise (i.e when none are currently running). | |
233 | */ | |
234 | struct sched_entity *curr; | |
ddc97297 PZ |
235 | |
236 | unsigned long nr_spread_over; | |
237 | ||
62160e3f | 238 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
239 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
240 | ||
241 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
242 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities | |
243 | * (like users, containers etc.) | |
244 | * | |
245 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
246 | * list is used during load balance. | |
247 | */ | |
248 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ | |
4cf86d77 | 249 | struct task_group *tg; /* group that "owns" this runqueue */ |
9b5b7751 | 250 | struct rcu_head rcu; |
6aa645ea IM |
251 | #endif |
252 | }; | |
1da177e4 | 253 | |
6aa645ea IM |
254 | /* Real-Time classes' related field in a runqueue: */ |
255 | struct rt_rq { | |
256 | struct rt_prio_array active; | |
257 | int rt_load_balance_idx; | |
258 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
259 | }; | |
260 | ||
1da177e4 LT |
261 | /* |
262 | * This is the main, per-CPU runqueue data structure. | |
263 | * | |
264 | * Locking rule: those places that want to lock multiple runqueues | |
265 | * (such as the load balancing or the thread migration code), lock | |
266 | * acquire operations must be ordered by ascending &runqueue. | |
267 | */ | |
70b97a7f | 268 | struct rq { |
d8016491 IM |
269 | /* runqueue lock: */ |
270 | spinlock_t lock; | |
1da177e4 LT |
271 | |
272 | /* | |
273 | * nr_running and cpu_load should be in the same cacheline because | |
274 | * remote CPUs use both these fields when doing load calculation. | |
275 | */ | |
276 | unsigned long nr_running; | |
6aa645ea IM |
277 | #define CPU_LOAD_IDX_MAX 5 |
278 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 279 | unsigned char idle_at_tick; |
46cb4b7c SS |
280 | #ifdef CONFIG_NO_HZ |
281 | unsigned char in_nohz_recently; | |
282 | #endif | |
d8016491 IM |
283 | /* capture load from *all* tasks on this cpu: */ |
284 | struct load_weight load; | |
6aa645ea IM |
285 | unsigned long nr_load_updates; |
286 | u64 nr_switches; | |
287 | ||
288 | struct cfs_rq cfs; | |
289 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
290 | /* list of leaf cfs_rq on this cpu: */ |
291 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 292 | #endif |
6aa645ea | 293 | struct rt_rq rt; |
1da177e4 LT |
294 | |
295 | /* | |
296 | * This is part of a global counter where only the total sum | |
297 | * over all CPUs matters. A task can increase this counter on | |
298 | * one CPU and if it got migrated afterwards it may decrease | |
299 | * it on another CPU. Always updated under the runqueue lock: | |
300 | */ | |
301 | unsigned long nr_uninterruptible; | |
302 | ||
36c8b586 | 303 | struct task_struct *curr, *idle; |
c9819f45 | 304 | unsigned long next_balance; |
1da177e4 | 305 | struct mm_struct *prev_mm; |
6aa645ea | 306 | |
6aa645ea IM |
307 | u64 clock, prev_clock_raw; |
308 | s64 clock_max_delta; | |
309 | ||
310 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
311 | u64 idle_clock; |
312 | unsigned int clock_deep_idle_events; | |
529c7726 | 313 | u64 tick_timestamp; |
6aa645ea | 314 | |
1da177e4 LT |
315 | atomic_t nr_iowait; |
316 | ||
317 | #ifdef CONFIG_SMP | |
318 | struct sched_domain *sd; | |
319 | ||
320 | /* For active balancing */ | |
321 | int active_balance; | |
322 | int push_cpu; | |
d8016491 IM |
323 | /* cpu of this runqueue: */ |
324 | int cpu; | |
1da177e4 | 325 | |
36c8b586 | 326 | struct task_struct *migration_thread; |
1da177e4 LT |
327 | struct list_head migration_queue; |
328 | #endif | |
329 | ||
330 | #ifdef CONFIG_SCHEDSTATS | |
331 | /* latency stats */ | |
332 | struct sched_info rq_sched_info; | |
333 | ||
334 | /* sys_sched_yield() stats */ | |
480b9434 KC |
335 | unsigned int yld_exp_empty; |
336 | unsigned int yld_act_empty; | |
337 | unsigned int yld_both_empty; | |
338 | unsigned int yld_count; | |
1da177e4 LT |
339 | |
340 | /* schedule() stats */ | |
480b9434 KC |
341 | unsigned int sched_switch; |
342 | unsigned int sched_count; | |
343 | unsigned int sched_goidle; | |
1da177e4 LT |
344 | |
345 | /* try_to_wake_up() stats */ | |
480b9434 KC |
346 | unsigned int ttwu_count; |
347 | unsigned int ttwu_local; | |
b8efb561 IM |
348 | |
349 | /* BKL stats */ | |
480b9434 | 350 | unsigned int bkl_count; |
1da177e4 | 351 | #endif |
fcb99371 | 352 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
353 | }; |
354 | ||
f34e3b61 | 355 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 356 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 357 | |
dd41f596 IM |
358 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
359 | { | |
360 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
361 | } | |
362 | ||
0a2966b4 CL |
363 | static inline int cpu_of(struct rq *rq) |
364 | { | |
365 | #ifdef CONFIG_SMP | |
366 | return rq->cpu; | |
367 | #else | |
368 | return 0; | |
369 | #endif | |
370 | } | |
371 | ||
20d315d4 | 372 | /* |
b04a0f4c IM |
373 | * Update the per-runqueue clock, as finegrained as the platform can give |
374 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 375 | */ |
b04a0f4c | 376 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
377 | { |
378 | u64 prev_raw = rq->prev_clock_raw; | |
379 | u64 now = sched_clock(); | |
380 | s64 delta = now - prev_raw; | |
381 | u64 clock = rq->clock; | |
382 | ||
b04a0f4c IM |
383 | #ifdef CONFIG_SCHED_DEBUG |
384 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
385 | #endif | |
20d315d4 IM |
386 | /* |
387 | * Protect against sched_clock() occasionally going backwards: | |
388 | */ | |
389 | if (unlikely(delta < 0)) { | |
390 | clock++; | |
391 | rq->clock_warps++; | |
392 | } else { | |
393 | /* | |
394 | * Catch too large forward jumps too: | |
395 | */ | |
529c7726 IM |
396 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
397 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
398 | clock = rq->tick_timestamp + TICK_NSEC; | |
399 | else | |
400 | clock++; | |
20d315d4 IM |
401 | rq->clock_overflows++; |
402 | } else { | |
403 | if (unlikely(delta > rq->clock_max_delta)) | |
404 | rq->clock_max_delta = delta; | |
405 | clock += delta; | |
406 | } | |
407 | } | |
408 | ||
409 | rq->prev_clock_raw = now; | |
410 | rq->clock = clock; | |
b04a0f4c | 411 | } |
20d315d4 | 412 | |
b04a0f4c IM |
413 | static void update_rq_clock(struct rq *rq) |
414 | { | |
415 | if (likely(smp_processor_id() == cpu_of(rq))) | |
416 | __update_rq_clock(rq); | |
20d315d4 IM |
417 | } |
418 | ||
674311d5 NP |
419 | /* |
420 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 421 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
422 | * |
423 | * The domain tree of any CPU may only be accessed from within | |
424 | * preempt-disabled sections. | |
425 | */ | |
48f24c4d IM |
426 | #define for_each_domain(cpu, __sd) \ |
427 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
428 | |
429 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
430 | #define this_rq() (&__get_cpu_var(runqueues)) | |
431 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
432 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
433 | ||
bf5c91ba IM |
434 | /* |
435 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
436 | */ | |
437 | #ifdef CONFIG_SCHED_DEBUG | |
438 | # define const_debug __read_mostly | |
439 | #else | |
440 | # define const_debug static const | |
441 | #endif | |
442 | ||
443 | /* | |
444 | * Debugging: various feature bits | |
445 | */ | |
446 | enum { | |
bbdba7c0 IM |
447 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
448 | SCHED_FEAT_START_DEBIT = 2, | |
06877c33 | 449 | SCHED_FEAT_TREE_AVG = 4, |
bbdba7c0 | 450 | SCHED_FEAT_APPROX_AVG = 8, |
ce6c1311 | 451 | SCHED_FEAT_WAKEUP_PREEMPT = 16, |
95938a35 | 452 | SCHED_FEAT_PREEMPT_RESTRICT = 32, |
bf5c91ba IM |
453 | }; |
454 | ||
455 | const_debug unsigned int sysctl_sched_features = | |
8401f775 IM |
456 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
457 | SCHED_FEAT_START_DEBIT * 1 | | |
458 | SCHED_FEAT_TREE_AVG * 0 | | |
459 | SCHED_FEAT_APPROX_AVG * 0 | | |
460 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | | |
461 | SCHED_FEAT_PREEMPT_RESTRICT * 1; | |
bf5c91ba IM |
462 | |
463 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
464 | ||
e436d800 IM |
465 | /* |
466 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
467 | * clock constructed from sched_clock(): | |
468 | */ | |
469 | unsigned long long cpu_clock(int cpu) | |
470 | { | |
e436d800 IM |
471 | unsigned long long now; |
472 | unsigned long flags; | |
b04a0f4c | 473 | struct rq *rq; |
e436d800 | 474 | |
2cd4d0ea | 475 | local_irq_save(flags); |
b04a0f4c IM |
476 | rq = cpu_rq(cpu); |
477 | update_rq_clock(rq); | |
478 | now = rq->clock; | |
2cd4d0ea | 479 | local_irq_restore(flags); |
e436d800 IM |
480 | |
481 | return now; | |
482 | } | |
a58f6f25 | 483 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 484 | |
1da177e4 | 485 | #ifndef prepare_arch_switch |
4866cde0 NP |
486 | # define prepare_arch_switch(next) do { } while (0) |
487 | #endif | |
488 | #ifndef finish_arch_switch | |
489 | # define finish_arch_switch(prev) do { } while (0) | |
490 | #endif | |
491 | ||
492 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 493 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
494 | { |
495 | return rq->curr == p; | |
496 | } | |
497 | ||
70b97a7f | 498 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
499 | { |
500 | } | |
501 | ||
70b97a7f | 502 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 503 | { |
da04c035 IM |
504 | #ifdef CONFIG_DEBUG_SPINLOCK |
505 | /* this is a valid case when another task releases the spinlock */ | |
506 | rq->lock.owner = current; | |
507 | #endif | |
8a25d5de IM |
508 | /* |
509 | * If we are tracking spinlock dependencies then we have to | |
510 | * fix up the runqueue lock - which gets 'carried over' from | |
511 | * prev into current: | |
512 | */ | |
513 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
514 | ||
4866cde0 NP |
515 | spin_unlock_irq(&rq->lock); |
516 | } | |
517 | ||
518 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 519 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
520 | { |
521 | #ifdef CONFIG_SMP | |
522 | return p->oncpu; | |
523 | #else | |
524 | return rq->curr == p; | |
525 | #endif | |
526 | } | |
527 | ||
70b97a7f | 528 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
529 | { |
530 | #ifdef CONFIG_SMP | |
531 | /* | |
532 | * We can optimise this out completely for !SMP, because the | |
533 | * SMP rebalancing from interrupt is the only thing that cares | |
534 | * here. | |
535 | */ | |
536 | next->oncpu = 1; | |
537 | #endif | |
538 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
539 | spin_unlock_irq(&rq->lock); | |
540 | #else | |
541 | spin_unlock(&rq->lock); | |
542 | #endif | |
543 | } | |
544 | ||
70b97a7f | 545 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
546 | { |
547 | #ifdef CONFIG_SMP | |
548 | /* | |
549 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
550 | * We must ensure this doesn't happen until the switch is completely | |
551 | * finished. | |
552 | */ | |
553 | smp_wmb(); | |
554 | prev->oncpu = 0; | |
555 | #endif | |
556 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
557 | local_irq_enable(); | |
1da177e4 | 558 | #endif |
4866cde0 NP |
559 | } |
560 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 561 | |
b29739f9 IM |
562 | /* |
563 | * __task_rq_lock - lock the runqueue a given task resides on. | |
564 | * Must be called interrupts disabled. | |
565 | */ | |
70b97a7f | 566 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
567 | __acquires(rq->lock) |
568 | { | |
3a5c359a AK |
569 | for (;;) { |
570 | struct rq *rq = task_rq(p); | |
571 | spin_lock(&rq->lock); | |
572 | if (likely(rq == task_rq(p))) | |
573 | return rq; | |
b29739f9 | 574 | spin_unlock(&rq->lock); |
b29739f9 | 575 | } |
b29739f9 IM |
576 | } |
577 | ||
1da177e4 LT |
578 | /* |
579 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
580 | * interrupts. Note the ordering: we can safely lookup the task_rq without | |
581 | * explicitly disabling preemption. | |
582 | */ | |
70b97a7f | 583 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
584 | __acquires(rq->lock) |
585 | { | |
70b97a7f | 586 | struct rq *rq; |
1da177e4 | 587 | |
3a5c359a AK |
588 | for (;;) { |
589 | local_irq_save(*flags); | |
590 | rq = task_rq(p); | |
591 | spin_lock(&rq->lock); | |
592 | if (likely(rq == task_rq(p))) | |
593 | return rq; | |
1da177e4 | 594 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 595 | } |
1da177e4 LT |
596 | } |
597 | ||
a9957449 | 598 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
599 | __releases(rq->lock) |
600 | { | |
601 | spin_unlock(&rq->lock); | |
602 | } | |
603 | ||
70b97a7f | 604 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
605 | __releases(rq->lock) |
606 | { | |
607 | spin_unlock_irqrestore(&rq->lock, *flags); | |
608 | } | |
609 | ||
1da177e4 | 610 | /* |
cc2a73b5 | 611 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 612 | */ |
a9957449 | 613 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
614 | __acquires(rq->lock) |
615 | { | |
70b97a7f | 616 | struct rq *rq; |
1da177e4 LT |
617 | |
618 | local_irq_disable(); | |
619 | rq = this_rq(); | |
620 | spin_lock(&rq->lock); | |
621 | ||
622 | return rq; | |
623 | } | |
624 | ||
1b9f19c2 | 625 | /* |
2aa44d05 | 626 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 627 | */ |
2aa44d05 | 628 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 629 | { |
2aa44d05 IM |
630 | struct rq *rq = cpu_rq(smp_processor_id()); |
631 | ||
632 | spin_lock(&rq->lock); | |
633 | __update_rq_clock(rq); | |
634 | spin_unlock(&rq->lock); | |
635 | rq->clock_deep_idle_events++; | |
636 | } | |
637 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
638 | ||
639 | /* | |
640 | * We just idled delta nanoseconds (called with irqs disabled): | |
641 | */ | |
642 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
643 | { | |
644 | struct rq *rq = cpu_rq(smp_processor_id()); | |
645 | u64 now = sched_clock(); | |
1b9f19c2 | 646 | |
2aa44d05 IM |
647 | rq->idle_clock += delta_ns; |
648 | /* | |
649 | * Override the previous timestamp and ignore all | |
650 | * sched_clock() deltas that occured while we idled, | |
651 | * and use the PM-provided delta_ns to advance the | |
652 | * rq clock: | |
653 | */ | |
654 | spin_lock(&rq->lock); | |
655 | rq->prev_clock_raw = now; | |
656 | rq->clock += delta_ns; | |
657 | spin_unlock(&rq->lock); | |
1b9f19c2 | 658 | } |
2aa44d05 | 659 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 660 | |
c24d20db IM |
661 | /* |
662 | * resched_task - mark a task 'to be rescheduled now'. | |
663 | * | |
664 | * On UP this means the setting of the need_resched flag, on SMP it | |
665 | * might also involve a cross-CPU call to trigger the scheduler on | |
666 | * the target CPU. | |
667 | */ | |
668 | #ifdef CONFIG_SMP | |
669 | ||
670 | #ifndef tsk_is_polling | |
671 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
672 | #endif | |
673 | ||
674 | static void resched_task(struct task_struct *p) | |
675 | { | |
676 | int cpu; | |
677 | ||
678 | assert_spin_locked(&task_rq(p)->lock); | |
679 | ||
680 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
681 | return; | |
682 | ||
683 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
684 | ||
685 | cpu = task_cpu(p); | |
686 | if (cpu == smp_processor_id()) | |
687 | return; | |
688 | ||
689 | /* NEED_RESCHED must be visible before we test polling */ | |
690 | smp_mb(); | |
691 | if (!tsk_is_polling(p)) | |
692 | smp_send_reschedule(cpu); | |
693 | } | |
694 | ||
695 | static void resched_cpu(int cpu) | |
696 | { | |
697 | struct rq *rq = cpu_rq(cpu); | |
698 | unsigned long flags; | |
699 | ||
700 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
701 | return; | |
702 | resched_task(cpu_curr(cpu)); | |
703 | spin_unlock_irqrestore(&rq->lock, flags); | |
704 | } | |
705 | #else | |
706 | static inline void resched_task(struct task_struct *p) | |
707 | { | |
708 | assert_spin_locked(&task_rq(p)->lock); | |
709 | set_tsk_need_resched(p); | |
710 | } | |
711 | #endif | |
712 | ||
45bf76df IM |
713 | #if BITS_PER_LONG == 32 |
714 | # define WMULT_CONST (~0UL) | |
715 | #else | |
716 | # define WMULT_CONST (1UL << 32) | |
717 | #endif | |
718 | ||
719 | #define WMULT_SHIFT 32 | |
720 | ||
194081eb IM |
721 | /* |
722 | * Shift right and round: | |
723 | */ | |
cf2ab469 | 724 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 725 | |
cb1c4fc9 | 726 | static unsigned long |
45bf76df IM |
727 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
728 | struct load_weight *lw) | |
729 | { | |
730 | u64 tmp; | |
731 | ||
732 | if (unlikely(!lw->inv_weight)) | |
194081eb | 733 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
734 | |
735 | tmp = (u64)delta_exec * weight; | |
736 | /* | |
737 | * Check whether we'd overflow the 64-bit multiplication: | |
738 | */ | |
194081eb | 739 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 740 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
741 | WMULT_SHIFT/2); |
742 | else | |
cf2ab469 | 743 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 744 | |
ecf691da | 745 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
746 | } |
747 | ||
748 | static inline unsigned long | |
749 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
750 | { | |
751 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
752 | } | |
753 | ||
1091985b | 754 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
755 | { |
756 | lw->weight += inc; | |
45bf76df IM |
757 | } |
758 | ||
1091985b | 759 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
760 | { |
761 | lw->weight -= dec; | |
45bf76df IM |
762 | } |
763 | ||
2dd73a4f PW |
764 | /* |
765 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
766 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
767 | * each task makes to its run queue's load is weighted according to its | |
768 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a | |
769 | * scaled version of the new time slice allocation that they receive on time | |
770 | * slice expiry etc. | |
771 | */ | |
772 | ||
dd41f596 IM |
773 | #define WEIGHT_IDLEPRIO 2 |
774 | #define WMULT_IDLEPRIO (1 << 31) | |
775 | ||
776 | /* | |
777 | * Nice levels are multiplicative, with a gentle 10% change for every | |
778 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
779 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
780 | * that remained on nice 0. | |
781 | * | |
782 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
783 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
784 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
785 | * If a task goes up by ~10% and another task goes down by ~10% then | |
786 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
787 | */ |
788 | static const int prio_to_weight[40] = { | |
254753dc IM |
789 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
790 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
791 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
792 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
793 | /* 0 */ 1024, 820, 655, 526, 423, | |
794 | /* 5 */ 335, 272, 215, 172, 137, | |
795 | /* 10 */ 110, 87, 70, 56, 45, | |
796 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
797 | }; |
798 | ||
5714d2de IM |
799 | /* |
800 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
801 | * | |
802 | * In cases where the weight does not change often, we can use the | |
803 | * precalculated inverse to speed up arithmetics by turning divisions | |
804 | * into multiplications: | |
805 | */ | |
dd41f596 | 806 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
807 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
808 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
809 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
810 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
811 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
812 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
813 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
814 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 815 | }; |
2dd73a4f | 816 | |
dd41f596 IM |
817 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
818 | ||
819 | /* | |
820 | * runqueue iterator, to support SMP load-balancing between different | |
821 | * scheduling classes, without having to expose their internal data | |
822 | * structures to the load-balancing proper: | |
823 | */ | |
824 | struct rq_iterator { | |
825 | void *arg; | |
826 | struct task_struct *(*start)(void *); | |
827 | struct task_struct *(*next)(void *); | |
828 | }; | |
829 | ||
830 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
831 | unsigned long max_nr_move, unsigned long max_load_move, | |
832 | struct sched_domain *sd, enum cpu_idle_type idle, | |
833 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 834 | int *this_best_prio, struct rq_iterator *iterator); |
dd41f596 IM |
835 | |
836 | #include "sched_stats.h" | |
dd41f596 | 837 | #include "sched_idletask.c" |
5522d5d5 IM |
838 | #include "sched_fair.c" |
839 | #include "sched_rt.c" | |
dd41f596 IM |
840 | #ifdef CONFIG_SCHED_DEBUG |
841 | # include "sched_debug.c" | |
842 | #endif | |
843 | ||
844 | #define sched_class_highest (&rt_sched_class) | |
845 | ||
9c217245 IM |
846 | /* |
847 | * Update delta_exec, delta_fair fields for rq. | |
848 | * | |
849 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 850 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
851 | * delta_exec advances at the same rate as wall-clock (provided |
852 | * cpu is not idle). | |
853 | * | |
854 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
855 | * runqueue over any given interval. This (smoothened) load is used | |
856 | * during load balance. | |
857 | * | |
495eca49 | 858 | * This function is called /before/ updating rq->load |
9c217245 IM |
859 | * and when switching tasks. |
860 | */ | |
29b4b623 | 861 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 862 | { |
495eca49 | 863 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
864 | } |
865 | ||
79b5dddf | 866 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 867 | { |
495eca49 | 868 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
869 | } |
870 | ||
e5fa2237 | 871 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
872 | { |
873 | rq->nr_running++; | |
29b4b623 | 874 | inc_load(rq, p); |
9c217245 IM |
875 | } |
876 | ||
db53181e | 877 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
878 | { |
879 | rq->nr_running--; | |
79b5dddf | 880 | dec_load(rq, p); |
9c217245 IM |
881 | } |
882 | ||
45bf76df IM |
883 | static void set_load_weight(struct task_struct *p) |
884 | { | |
885 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
886 | p->se.load.weight = prio_to_weight[0] * 2; |
887 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
888 | return; | |
889 | } | |
45bf76df | 890 | |
dd41f596 IM |
891 | /* |
892 | * SCHED_IDLE tasks get minimal weight: | |
893 | */ | |
894 | if (p->policy == SCHED_IDLE) { | |
895 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
896 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
897 | return; | |
898 | } | |
71f8bd46 | 899 | |
dd41f596 IM |
900 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
901 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
902 | } |
903 | ||
8159f87e | 904 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 905 | { |
dd41f596 | 906 | sched_info_queued(p); |
fd390f6a | 907 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 908 | p->se.on_rq = 1; |
71f8bd46 IM |
909 | } |
910 | ||
69be72c1 | 911 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 912 | { |
f02231e5 | 913 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 914 | p->se.on_rq = 0; |
71f8bd46 IM |
915 | } |
916 | ||
14531189 | 917 | /* |
dd41f596 | 918 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 919 | */ |
14531189 IM |
920 | static inline int __normal_prio(struct task_struct *p) |
921 | { | |
dd41f596 | 922 | return p->static_prio; |
14531189 IM |
923 | } |
924 | ||
b29739f9 IM |
925 | /* |
926 | * Calculate the expected normal priority: i.e. priority | |
927 | * without taking RT-inheritance into account. Might be | |
928 | * boosted by interactivity modifiers. Changes upon fork, | |
929 | * setprio syscalls, and whenever the interactivity | |
930 | * estimator recalculates. | |
931 | */ | |
36c8b586 | 932 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
933 | { |
934 | int prio; | |
935 | ||
e05606d3 | 936 | if (task_has_rt_policy(p)) |
b29739f9 IM |
937 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
938 | else | |
939 | prio = __normal_prio(p); | |
940 | return prio; | |
941 | } | |
942 | ||
943 | /* | |
944 | * Calculate the current priority, i.e. the priority | |
945 | * taken into account by the scheduler. This value might | |
946 | * be boosted by RT tasks, or might be boosted by | |
947 | * interactivity modifiers. Will be RT if the task got | |
948 | * RT-boosted. If not then it returns p->normal_prio. | |
949 | */ | |
36c8b586 | 950 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
951 | { |
952 | p->normal_prio = normal_prio(p); | |
953 | /* | |
954 | * If we are RT tasks or we were boosted to RT priority, | |
955 | * keep the priority unchanged. Otherwise, update priority | |
956 | * to the normal priority: | |
957 | */ | |
958 | if (!rt_prio(p->prio)) | |
959 | return p->normal_prio; | |
960 | return p->prio; | |
961 | } | |
962 | ||
1da177e4 | 963 | /* |
dd41f596 | 964 | * activate_task - move a task to the runqueue. |
1da177e4 | 965 | */ |
dd41f596 | 966 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 967 | { |
dd41f596 IM |
968 | if (p->state == TASK_UNINTERRUPTIBLE) |
969 | rq->nr_uninterruptible--; | |
1da177e4 | 970 | |
8159f87e | 971 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 972 | inc_nr_running(p, rq); |
1da177e4 LT |
973 | } |
974 | ||
1da177e4 LT |
975 | /* |
976 | * deactivate_task - remove a task from the runqueue. | |
977 | */ | |
2e1cb74a | 978 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 979 | { |
dd41f596 IM |
980 | if (p->state == TASK_UNINTERRUPTIBLE) |
981 | rq->nr_uninterruptible++; | |
982 | ||
69be72c1 | 983 | dequeue_task(rq, p, sleep); |
db53181e | 984 | dec_nr_running(p, rq); |
1da177e4 LT |
985 | } |
986 | ||
1da177e4 LT |
987 | /** |
988 | * task_curr - is this task currently executing on a CPU? | |
989 | * @p: the task in question. | |
990 | */ | |
36c8b586 | 991 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
992 | { |
993 | return cpu_curr(task_cpu(p)) == p; | |
994 | } | |
995 | ||
2dd73a4f PW |
996 | /* Used instead of source_load when we know the type == 0 */ |
997 | unsigned long weighted_cpuload(const int cpu) | |
998 | { | |
495eca49 | 999 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1000 | } |
1001 | ||
1002 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1003 | { | |
1004 | #ifdef CONFIG_SMP | |
1005 | task_thread_info(p)->cpu = cpu; | |
dd41f596 | 1006 | #endif |
29f59db3 | 1007 | set_task_cfs_rq(p); |
2dd73a4f PW |
1008 | } |
1009 | ||
1da177e4 | 1010 | #ifdef CONFIG_SMP |
c65cc870 | 1011 | |
cc367732 IM |
1012 | /* |
1013 | * Is this task likely cache-hot: | |
1014 | */ | |
1015 | static inline int | |
1016 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1017 | { | |
1018 | s64 delta; | |
1019 | ||
1020 | if (p->sched_class != &fair_sched_class) | |
1021 | return 0; | |
1022 | ||
6bc1665b IM |
1023 | if (sysctl_sched_migration_cost == -1) |
1024 | return 1; | |
1025 | if (sysctl_sched_migration_cost == 0) | |
1026 | return 0; | |
1027 | ||
cc367732 IM |
1028 | delta = now - p->se.exec_start; |
1029 | ||
1030 | return delta < (s64)sysctl_sched_migration_cost; | |
1031 | } | |
1032 | ||
1033 | ||
dd41f596 | 1034 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1035 | { |
dd41f596 IM |
1036 | int old_cpu = task_cpu(p); |
1037 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1038 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1039 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1040 | u64 clock_offset; |
dd41f596 IM |
1041 | |
1042 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1043 | |
1044 | #ifdef CONFIG_SCHEDSTATS | |
1045 | if (p->se.wait_start) | |
1046 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1047 | if (p->se.sleep_start) |
1048 | p->se.sleep_start -= clock_offset; | |
1049 | if (p->se.block_start) | |
1050 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1051 | if (old_cpu != new_cpu) { |
1052 | schedstat_inc(p, se.nr_migrations); | |
1053 | if (task_hot(p, old_rq->clock, NULL)) | |
1054 | schedstat_inc(p, se.nr_forced2_migrations); | |
1055 | } | |
6cfb0d5d | 1056 | #endif |
2830cf8c SV |
1057 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1058 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1059 | |
1060 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1061 | } |
1062 | ||
70b97a7f | 1063 | struct migration_req { |
1da177e4 | 1064 | struct list_head list; |
1da177e4 | 1065 | |
36c8b586 | 1066 | struct task_struct *task; |
1da177e4 LT |
1067 | int dest_cpu; |
1068 | ||
1da177e4 | 1069 | struct completion done; |
70b97a7f | 1070 | }; |
1da177e4 LT |
1071 | |
1072 | /* | |
1073 | * The task's runqueue lock must be held. | |
1074 | * Returns true if you have to wait for migration thread. | |
1075 | */ | |
36c8b586 | 1076 | static int |
70b97a7f | 1077 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1078 | { |
70b97a7f | 1079 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1080 | |
1081 | /* | |
1082 | * If the task is not on a runqueue (and not running), then | |
1083 | * it is sufficient to simply update the task's cpu field. | |
1084 | */ | |
dd41f596 | 1085 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1086 | set_task_cpu(p, dest_cpu); |
1087 | return 0; | |
1088 | } | |
1089 | ||
1090 | init_completion(&req->done); | |
1da177e4 LT |
1091 | req->task = p; |
1092 | req->dest_cpu = dest_cpu; | |
1093 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1094 | |
1da177e4 LT |
1095 | return 1; |
1096 | } | |
1097 | ||
1098 | /* | |
1099 | * wait_task_inactive - wait for a thread to unschedule. | |
1100 | * | |
1101 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1102 | * else this function might spin for a *long* time. This function can't | |
1103 | * be called with interrupts off, or it may introduce deadlock with | |
1104 | * smp_call_function() if an IPI is sent by the same process we are | |
1105 | * waiting to become inactive. | |
1106 | */ | |
36c8b586 | 1107 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1108 | { |
1109 | unsigned long flags; | |
dd41f596 | 1110 | int running, on_rq; |
70b97a7f | 1111 | struct rq *rq; |
1da177e4 | 1112 | |
3a5c359a AK |
1113 | for (;;) { |
1114 | /* | |
1115 | * We do the initial early heuristics without holding | |
1116 | * any task-queue locks at all. We'll only try to get | |
1117 | * the runqueue lock when things look like they will | |
1118 | * work out! | |
1119 | */ | |
1120 | rq = task_rq(p); | |
fa490cfd | 1121 | |
3a5c359a AK |
1122 | /* |
1123 | * If the task is actively running on another CPU | |
1124 | * still, just relax and busy-wait without holding | |
1125 | * any locks. | |
1126 | * | |
1127 | * NOTE! Since we don't hold any locks, it's not | |
1128 | * even sure that "rq" stays as the right runqueue! | |
1129 | * But we don't care, since "task_running()" will | |
1130 | * return false if the runqueue has changed and p | |
1131 | * is actually now running somewhere else! | |
1132 | */ | |
1133 | while (task_running(rq, p)) | |
1134 | cpu_relax(); | |
fa490cfd | 1135 | |
3a5c359a AK |
1136 | /* |
1137 | * Ok, time to look more closely! We need the rq | |
1138 | * lock now, to be *sure*. If we're wrong, we'll | |
1139 | * just go back and repeat. | |
1140 | */ | |
1141 | rq = task_rq_lock(p, &flags); | |
1142 | running = task_running(rq, p); | |
1143 | on_rq = p->se.on_rq; | |
1144 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1145 | |
3a5c359a AK |
1146 | /* |
1147 | * Was it really running after all now that we | |
1148 | * checked with the proper locks actually held? | |
1149 | * | |
1150 | * Oops. Go back and try again.. | |
1151 | */ | |
1152 | if (unlikely(running)) { | |
1153 | cpu_relax(); | |
1154 | continue; | |
1155 | } | |
fa490cfd | 1156 | |
3a5c359a AK |
1157 | /* |
1158 | * It's not enough that it's not actively running, | |
1159 | * it must be off the runqueue _entirely_, and not | |
1160 | * preempted! | |
1161 | * | |
1162 | * So if it wa still runnable (but just not actively | |
1163 | * running right now), it's preempted, and we should | |
1164 | * yield - it could be a while. | |
1165 | */ | |
1166 | if (unlikely(on_rq)) { | |
1167 | schedule_timeout_uninterruptible(1); | |
1168 | continue; | |
1169 | } | |
fa490cfd | 1170 | |
3a5c359a AK |
1171 | /* |
1172 | * Ahh, all good. It wasn't running, and it wasn't | |
1173 | * runnable, which means that it will never become | |
1174 | * running in the future either. We're all done! | |
1175 | */ | |
1176 | break; | |
1177 | } | |
1da177e4 LT |
1178 | } |
1179 | ||
1180 | /*** | |
1181 | * kick_process - kick a running thread to enter/exit the kernel | |
1182 | * @p: the to-be-kicked thread | |
1183 | * | |
1184 | * Cause a process which is running on another CPU to enter | |
1185 | * kernel-mode, without any delay. (to get signals handled.) | |
1186 | * | |
1187 | * NOTE: this function doesnt have to take the runqueue lock, | |
1188 | * because all it wants to ensure is that the remote task enters | |
1189 | * the kernel. If the IPI races and the task has been migrated | |
1190 | * to another CPU then no harm is done and the purpose has been | |
1191 | * achieved as well. | |
1192 | */ | |
36c8b586 | 1193 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1194 | { |
1195 | int cpu; | |
1196 | ||
1197 | preempt_disable(); | |
1198 | cpu = task_cpu(p); | |
1199 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1200 | smp_send_reschedule(cpu); | |
1201 | preempt_enable(); | |
1202 | } | |
1203 | ||
1204 | /* | |
2dd73a4f PW |
1205 | * Return a low guess at the load of a migration-source cpu weighted |
1206 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1207 | * |
1208 | * We want to under-estimate the load of migration sources, to | |
1209 | * balance conservatively. | |
1210 | */ | |
a9957449 | 1211 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1212 | { |
70b97a7f | 1213 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1214 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1215 | |
3b0bd9bc | 1216 | if (type == 0) |
dd41f596 | 1217 | return total; |
b910472d | 1218 | |
dd41f596 | 1219 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1220 | } |
1221 | ||
1222 | /* | |
2dd73a4f PW |
1223 | * Return a high guess at the load of a migration-target cpu weighted |
1224 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1225 | */ |
a9957449 | 1226 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1227 | { |
70b97a7f | 1228 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1229 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1230 | |
7897986b | 1231 | if (type == 0) |
dd41f596 | 1232 | return total; |
3b0bd9bc | 1233 | |
dd41f596 | 1234 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1235 | } |
1236 | ||
1237 | /* | |
1238 | * Return the average load per task on the cpu's run queue | |
1239 | */ | |
1240 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1241 | { | |
70b97a7f | 1242 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1243 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1244 | unsigned long n = rq->nr_running; |
1245 | ||
dd41f596 | 1246 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1247 | } |
1248 | ||
147cbb4b NP |
1249 | /* |
1250 | * find_idlest_group finds and returns the least busy CPU group within the | |
1251 | * domain. | |
1252 | */ | |
1253 | static struct sched_group * | |
1254 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1255 | { | |
1256 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1257 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1258 | int load_idx = sd->forkexec_idx; | |
1259 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1260 | ||
1261 | do { | |
1262 | unsigned long load, avg_load; | |
1263 | int local_group; | |
1264 | int i; | |
1265 | ||
da5a5522 BD |
1266 | /* Skip over this group if it has no CPUs allowed */ |
1267 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1268 | continue; |
da5a5522 | 1269 | |
147cbb4b | 1270 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1271 | |
1272 | /* Tally up the load of all CPUs in the group */ | |
1273 | avg_load = 0; | |
1274 | ||
1275 | for_each_cpu_mask(i, group->cpumask) { | |
1276 | /* Bias balancing toward cpus of our domain */ | |
1277 | if (local_group) | |
1278 | load = source_load(i, load_idx); | |
1279 | else | |
1280 | load = target_load(i, load_idx); | |
1281 | ||
1282 | avg_load += load; | |
1283 | } | |
1284 | ||
1285 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1286 | avg_load = sg_div_cpu_power(group, |
1287 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1288 | |
1289 | if (local_group) { | |
1290 | this_load = avg_load; | |
1291 | this = group; | |
1292 | } else if (avg_load < min_load) { | |
1293 | min_load = avg_load; | |
1294 | idlest = group; | |
1295 | } | |
3a5c359a | 1296 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1297 | |
1298 | if (!idlest || 100*this_load < imbalance*min_load) | |
1299 | return NULL; | |
1300 | return idlest; | |
1301 | } | |
1302 | ||
1303 | /* | |
0feaece9 | 1304 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1305 | */ |
95cdf3b7 IM |
1306 | static int |
1307 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1308 | { |
da5a5522 | 1309 | cpumask_t tmp; |
147cbb4b NP |
1310 | unsigned long load, min_load = ULONG_MAX; |
1311 | int idlest = -1; | |
1312 | int i; | |
1313 | ||
da5a5522 BD |
1314 | /* Traverse only the allowed CPUs */ |
1315 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1316 | ||
1317 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1318 | load = weighted_cpuload(i); |
147cbb4b NP |
1319 | |
1320 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1321 | min_load = load; | |
1322 | idlest = i; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | return idlest; | |
1327 | } | |
1328 | ||
476d139c NP |
1329 | /* |
1330 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1331 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1332 | * SD_BALANCE_EXEC. | |
1333 | * | |
1334 | * Balance, ie. select the least loaded group. | |
1335 | * | |
1336 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1337 | * | |
1338 | * preempt must be disabled. | |
1339 | */ | |
1340 | static int sched_balance_self(int cpu, int flag) | |
1341 | { | |
1342 | struct task_struct *t = current; | |
1343 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1344 | |
c96d145e | 1345 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1346 | /* |
1347 | * If power savings logic is enabled for a domain, stop there. | |
1348 | */ | |
5c45bf27 SS |
1349 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1350 | break; | |
476d139c NP |
1351 | if (tmp->flags & flag) |
1352 | sd = tmp; | |
c96d145e | 1353 | } |
476d139c NP |
1354 | |
1355 | while (sd) { | |
1356 | cpumask_t span; | |
1357 | struct sched_group *group; | |
1a848870 SS |
1358 | int new_cpu, weight; |
1359 | ||
1360 | if (!(sd->flags & flag)) { | |
1361 | sd = sd->child; | |
1362 | continue; | |
1363 | } | |
476d139c NP |
1364 | |
1365 | span = sd->span; | |
1366 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1367 | if (!group) { |
1368 | sd = sd->child; | |
1369 | continue; | |
1370 | } | |
476d139c | 1371 | |
da5a5522 | 1372 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1373 | if (new_cpu == -1 || new_cpu == cpu) { |
1374 | /* Now try balancing at a lower domain level of cpu */ | |
1375 | sd = sd->child; | |
1376 | continue; | |
1377 | } | |
476d139c | 1378 | |
1a848870 | 1379 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1380 | cpu = new_cpu; |
476d139c NP |
1381 | sd = NULL; |
1382 | weight = cpus_weight(span); | |
1383 | for_each_domain(cpu, tmp) { | |
1384 | if (weight <= cpus_weight(tmp->span)) | |
1385 | break; | |
1386 | if (tmp->flags & flag) | |
1387 | sd = tmp; | |
1388 | } | |
1389 | /* while loop will break here if sd == NULL */ | |
1390 | } | |
1391 | ||
1392 | return cpu; | |
1393 | } | |
1394 | ||
1395 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1396 | |
1397 | /* | |
1398 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1399 | * not idle and an idle cpu is available. The span of cpus to | |
1400 | * search starts with cpus closest then further out as needed, | |
1401 | * so we always favor a closer, idle cpu. | |
1402 | * | |
1403 | * Returns the CPU we should wake onto. | |
1404 | */ | |
1405 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1406 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1407 | { |
1408 | cpumask_t tmp; | |
1409 | struct sched_domain *sd; | |
1410 | int i; | |
1411 | ||
4953198b SS |
1412 | /* |
1413 | * If it is idle, then it is the best cpu to run this task. | |
1414 | * | |
1415 | * This cpu is also the best, if it has more than one task already. | |
1416 | * Siblings must be also busy(in most cases) as they didn't already | |
1417 | * pickup the extra load from this cpu and hence we need not check | |
1418 | * sibling runqueue info. This will avoid the checks and cache miss | |
1419 | * penalities associated with that. | |
1420 | */ | |
1421 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1422 | return cpu; |
1423 | ||
1424 | for_each_domain(cpu, sd) { | |
1425 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1426 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1427 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1428 | if (idle_cpu(i)) { |
1429 | if (i != task_cpu(p)) { | |
1430 | schedstat_inc(p, | |
1431 | se.nr_wakeups_idle); | |
1432 | } | |
1da177e4 | 1433 | return i; |
cc367732 | 1434 | } |
1da177e4 | 1435 | } |
9761eea8 | 1436 | } else { |
e0f364f4 | 1437 | break; |
9761eea8 | 1438 | } |
1da177e4 LT |
1439 | } |
1440 | return cpu; | |
1441 | } | |
1442 | #else | |
36c8b586 | 1443 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1444 | { |
1445 | return cpu; | |
1446 | } | |
1447 | #endif | |
1448 | ||
1449 | /*** | |
1450 | * try_to_wake_up - wake up a thread | |
1451 | * @p: the to-be-woken-up thread | |
1452 | * @state: the mask of task states that can be woken | |
1453 | * @sync: do a synchronous wakeup? | |
1454 | * | |
1455 | * Put it on the run-queue if it's not already there. The "current" | |
1456 | * thread is always on the run-queue (except when the actual | |
1457 | * re-schedule is in progress), and as such you're allowed to do | |
1458 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1459 | * runnable without the overhead of this. | |
1460 | * | |
1461 | * returns failure only if the task is already active. | |
1462 | */ | |
36c8b586 | 1463 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1464 | { |
cc367732 | 1465 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1466 | unsigned long flags; |
1467 | long old_state; | |
70b97a7f | 1468 | struct rq *rq; |
1da177e4 | 1469 | #ifdef CONFIG_SMP |
7897986b | 1470 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1471 | unsigned long load, this_load; |
1da177e4 LT |
1472 | int new_cpu; |
1473 | #endif | |
1474 | ||
1475 | rq = task_rq_lock(p, &flags); | |
1476 | old_state = p->state; | |
1477 | if (!(old_state & state)) | |
1478 | goto out; | |
1479 | ||
dd41f596 | 1480 | if (p->se.on_rq) |
1da177e4 LT |
1481 | goto out_running; |
1482 | ||
1483 | cpu = task_cpu(p); | |
cc367732 | 1484 | orig_cpu = cpu; |
1da177e4 LT |
1485 | this_cpu = smp_processor_id(); |
1486 | ||
1487 | #ifdef CONFIG_SMP | |
1488 | if (unlikely(task_running(rq, p))) | |
1489 | goto out_activate; | |
1490 | ||
7897986b NP |
1491 | new_cpu = cpu; |
1492 | ||
2d72376b | 1493 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1494 | if (cpu == this_cpu) { |
1495 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1496 | goto out_set_cpu; |
1497 | } | |
1498 | ||
1499 | for_each_domain(this_cpu, sd) { | |
1500 | if (cpu_isset(cpu, sd->span)) { | |
1501 | schedstat_inc(sd, ttwu_wake_remote); | |
1502 | this_sd = sd; | |
1503 | break; | |
1da177e4 LT |
1504 | } |
1505 | } | |
1da177e4 | 1506 | |
7897986b | 1507 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1508 | goto out_set_cpu; |
1509 | ||
1da177e4 | 1510 | /* |
7897986b | 1511 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1512 | */ |
7897986b NP |
1513 | if (this_sd) { |
1514 | int idx = this_sd->wake_idx; | |
1515 | unsigned int imbalance; | |
1da177e4 | 1516 | |
a3f21bce NP |
1517 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1518 | ||
7897986b NP |
1519 | load = source_load(cpu, idx); |
1520 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1521 | |
7897986b NP |
1522 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1523 | ||
a3f21bce NP |
1524 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1525 | unsigned long tl = this_load; | |
33859f7f MOS |
1526 | unsigned long tl_per_task; |
1527 | ||
71e20f18 IM |
1528 | /* |
1529 | * Attract cache-cold tasks on sync wakeups: | |
1530 | */ | |
1531 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1532 | goto out_set_cpu; | |
1533 | ||
cc367732 | 1534 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1535 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1536 | |
1da177e4 | 1537 | /* |
a3f21bce NP |
1538 | * If sync wakeup then subtract the (maximum possible) |
1539 | * effect of the currently running task from the load | |
1540 | * of the current CPU: | |
1da177e4 | 1541 | */ |
a3f21bce | 1542 | if (sync) |
dd41f596 | 1543 | tl -= current->se.load.weight; |
a3f21bce NP |
1544 | |
1545 | if ((tl <= load && | |
2dd73a4f | 1546 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1547 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1548 | /* |
1549 | * This domain has SD_WAKE_AFFINE and | |
1550 | * p is cache cold in this domain, and | |
1551 | * there is no bad imbalance. | |
1552 | */ | |
1553 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1554 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1555 | goto out_set_cpu; |
1556 | } | |
1557 | } | |
1558 | ||
1559 | /* | |
1560 | * Start passive balancing when half the imbalance_pct | |
1561 | * limit is reached. | |
1562 | */ | |
1563 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1564 | if (imbalance*this_load <= 100*load) { | |
1565 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1566 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1567 | goto out_set_cpu; |
1568 | } | |
1da177e4 LT |
1569 | } |
1570 | } | |
1571 | ||
1572 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1573 | out_set_cpu: | |
1574 | new_cpu = wake_idle(new_cpu, p); | |
1575 | if (new_cpu != cpu) { | |
1576 | set_task_cpu(p, new_cpu); | |
1577 | task_rq_unlock(rq, &flags); | |
1578 | /* might preempt at this point */ | |
1579 | rq = task_rq_lock(p, &flags); | |
1580 | old_state = p->state; | |
1581 | if (!(old_state & state)) | |
1582 | goto out; | |
dd41f596 | 1583 | if (p->se.on_rq) |
1da177e4 LT |
1584 | goto out_running; |
1585 | ||
1586 | this_cpu = smp_processor_id(); | |
1587 | cpu = task_cpu(p); | |
1588 | } | |
1589 | ||
1590 | out_activate: | |
1591 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1592 | schedstat_inc(p, se.nr_wakeups); |
1593 | if (sync) | |
1594 | schedstat_inc(p, se.nr_wakeups_sync); | |
1595 | if (orig_cpu != cpu) | |
1596 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1597 | if (cpu == this_cpu) | |
1598 | schedstat_inc(p, se.nr_wakeups_local); | |
1599 | else | |
1600 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1601 | update_rq_clock(rq); |
dd41f596 | 1602 | activate_task(rq, p, 1); |
9c63d9c0 | 1603 | check_preempt_curr(rq, p); |
1da177e4 LT |
1604 | success = 1; |
1605 | ||
1606 | out_running: | |
1607 | p->state = TASK_RUNNING; | |
1608 | out: | |
1609 | task_rq_unlock(rq, &flags); | |
1610 | ||
1611 | return success; | |
1612 | } | |
1613 | ||
36c8b586 | 1614 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1615 | { |
1616 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1617 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1618 | } | |
1da177e4 LT |
1619 | EXPORT_SYMBOL(wake_up_process); |
1620 | ||
36c8b586 | 1621 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1622 | { |
1623 | return try_to_wake_up(p, state, 0); | |
1624 | } | |
1625 | ||
1da177e4 LT |
1626 | /* |
1627 | * Perform scheduler related setup for a newly forked process p. | |
1628 | * p is forked by current. | |
dd41f596 IM |
1629 | * |
1630 | * __sched_fork() is basic setup used by init_idle() too: | |
1631 | */ | |
1632 | static void __sched_fork(struct task_struct *p) | |
1633 | { | |
dd41f596 IM |
1634 | p->se.exec_start = 0; |
1635 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1636 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1637 | |
1638 | #ifdef CONFIG_SCHEDSTATS | |
1639 | p->se.wait_start = 0; | |
dd41f596 IM |
1640 | p->se.sum_sleep_runtime = 0; |
1641 | p->se.sleep_start = 0; | |
dd41f596 IM |
1642 | p->se.block_start = 0; |
1643 | p->se.sleep_max = 0; | |
1644 | p->se.block_max = 0; | |
1645 | p->se.exec_max = 0; | |
eba1ed4b | 1646 | p->se.slice_max = 0; |
dd41f596 | 1647 | p->se.wait_max = 0; |
6cfb0d5d | 1648 | #endif |
476d139c | 1649 | |
dd41f596 IM |
1650 | INIT_LIST_HEAD(&p->run_list); |
1651 | p->se.on_rq = 0; | |
476d139c | 1652 | |
e107be36 AK |
1653 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1654 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1655 | #endif | |
1656 | ||
1da177e4 LT |
1657 | /* |
1658 | * We mark the process as running here, but have not actually | |
1659 | * inserted it onto the runqueue yet. This guarantees that | |
1660 | * nobody will actually run it, and a signal or other external | |
1661 | * event cannot wake it up and insert it on the runqueue either. | |
1662 | */ | |
1663 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1664 | } |
1665 | ||
1666 | /* | |
1667 | * fork()/clone()-time setup: | |
1668 | */ | |
1669 | void sched_fork(struct task_struct *p, int clone_flags) | |
1670 | { | |
1671 | int cpu = get_cpu(); | |
1672 | ||
1673 | __sched_fork(p); | |
1674 | ||
1675 | #ifdef CONFIG_SMP | |
1676 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1677 | #endif | |
02e4bac2 | 1678 | set_task_cpu(p, cpu); |
b29739f9 IM |
1679 | |
1680 | /* | |
1681 | * Make sure we do not leak PI boosting priority to the child: | |
1682 | */ | |
1683 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1684 | if (!rt_prio(p->prio)) |
1685 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1686 | |
52f17b6c | 1687 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1688 | if (likely(sched_info_on())) |
52f17b6c | 1689 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1690 | #endif |
d6077cb8 | 1691 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1692 | p->oncpu = 0; |
1693 | #endif | |
1da177e4 | 1694 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1695 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1696 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1697 | #endif |
476d139c | 1698 | put_cpu(); |
1da177e4 LT |
1699 | } |
1700 | ||
1701 | /* | |
1702 | * wake_up_new_task - wake up a newly created task for the first time. | |
1703 | * | |
1704 | * This function will do some initial scheduler statistics housekeeping | |
1705 | * that must be done for every newly created context, then puts the task | |
1706 | * on the runqueue and wakes it. | |
1707 | */ | |
36c8b586 | 1708 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1709 | { |
1710 | unsigned long flags; | |
dd41f596 | 1711 | struct rq *rq; |
1da177e4 LT |
1712 | |
1713 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1714 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1715 | update_rq_clock(rq); |
1da177e4 LT |
1716 | |
1717 | p->prio = effective_prio(p); | |
1718 | ||
b9dca1e0 | 1719 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1720 | activate_task(rq, p, 0); |
1da177e4 | 1721 | } else { |
1da177e4 | 1722 | /* |
dd41f596 IM |
1723 | * Let the scheduling class do new task startup |
1724 | * management (if any): | |
1da177e4 | 1725 | */ |
ee0827d8 | 1726 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1727 | inc_nr_running(p, rq); |
1da177e4 | 1728 | } |
dd41f596 IM |
1729 | check_preempt_curr(rq, p); |
1730 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1731 | } |
1732 | ||
e107be36 AK |
1733 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1734 | ||
1735 | /** | |
421cee29 RD |
1736 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1737 | * @notifier: notifier struct to register | |
e107be36 AK |
1738 | */ |
1739 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1740 | { | |
1741 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1742 | } | |
1743 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1744 | ||
1745 | /** | |
1746 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1747 | * @notifier: notifier struct to unregister |
e107be36 AK |
1748 | * |
1749 | * This is safe to call from within a preemption notifier. | |
1750 | */ | |
1751 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1752 | { | |
1753 | hlist_del(¬ifier->link); | |
1754 | } | |
1755 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1756 | ||
1757 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1758 | { | |
1759 | struct preempt_notifier *notifier; | |
1760 | struct hlist_node *node; | |
1761 | ||
1762 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1763 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1764 | } | |
1765 | ||
1766 | static void | |
1767 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1768 | struct task_struct *next) | |
1769 | { | |
1770 | struct preempt_notifier *notifier; | |
1771 | struct hlist_node *node; | |
1772 | ||
1773 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1774 | notifier->ops->sched_out(notifier, next); | |
1775 | } | |
1776 | ||
1777 | #else | |
1778 | ||
1779 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1780 | { | |
1781 | } | |
1782 | ||
1783 | static void | |
1784 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1785 | struct task_struct *next) | |
1786 | { | |
1787 | } | |
1788 | ||
1789 | #endif | |
1790 | ||
4866cde0 NP |
1791 | /** |
1792 | * prepare_task_switch - prepare to switch tasks | |
1793 | * @rq: the runqueue preparing to switch | |
421cee29 | 1794 | * @prev: the current task that is being switched out |
4866cde0 NP |
1795 | * @next: the task we are going to switch to. |
1796 | * | |
1797 | * This is called with the rq lock held and interrupts off. It must | |
1798 | * be paired with a subsequent finish_task_switch after the context | |
1799 | * switch. | |
1800 | * | |
1801 | * prepare_task_switch sets up locking and calls architecture specific | |
1802 | * hooks. | |
1803 | */ | |
e107be36 AK |
1804 | static inline void |
1805 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1806 | struct task_struct *next) | |
4866cde0 | 1807 | { |
e107be36 | 1808 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1809 | prepare_lock_switch(rq, next); |
1810 | prepare_arch_switch(next); | |
1811 | } | |
1812 | ||
1da177e4 LT |
1813 | /** |
1814 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1815 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1816 | * @prev: the thread we just switched away from. |
1817 | * | |
4866cde0 NP |
1818 | * finish_task_switch must be called after the context switch, paired |
1819 | * with a prepare_task_switch call before the context switch. | |
1820 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1821 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1822 | * |
1823 | * Note that we may have delayed dropping an mm in context_switch(). If | |
1824 | * so, we finish that here outside of the runqueue lock. (Doing it | |
1825 | * with the lock held can cause deadlocks; see schedule() for | |
1826 | * details.) | |
1827 | */ | |
a9957449 | 1828 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1829 | __releases(rq->lock) |
1830 | { | |
1da177e4 | 1831 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1832 | long prev_state; |
1da177e4 LT |
1833 | |
1834 | rq->prev_mm = NULL; | |
1835 | ||
1836 | /* | |
1837 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1838 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1839 | * schedule one last time. The schedule call will never return, and |
1840 | * the scheduled task must drop that reference. | |
c394cc9f | 1841 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1842 | * still held, otherwise prev could be scheduled on another cpu, die |
1843 | * there before we look at prev->state, and then the reference would | |
1844 | * be dropped twice. | |
1845 | * Manfred Spraul <manfred@colorfullife.com> | |
1846 | */ | |
55a101f8 | 1847 | prev_state = prev->state; |
4866cde0 NP |
1848 | finish_arch_switch(prev); |
1849 | finish_lock_switch(rq, prev); | |
e107be36 | 1850 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1851 | if (mm) |
1852 | mmdrop(mm); | |
c394cc9f | 1853 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1854 | /* |
1855 | * Remove function-return probe instances associated with this | |
1856 | * task and put them back on the free list. | |
9761eea8 | 1857 | */ |
c6fd91f0 | 1858 | kprobe_flush_task(prev); |
1da177e4 | 1859 | put_task_struct(prev); |
c6fd91f0 | 1860 | } |
1da177e4 LT |
1861 | } |
1862 | ||
1863 | /** | |
1864 | * schedule_tail - first thing a freshly forked thread must call. | |
1865 | * @prev: the thread we just switched away from. | |
1866 | */ | |
36c8b586 | 1867 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1868 | __releases(rq->lock) |
1869 | { | |
70b97a7f IM |
1870 | struct rq *rq = this_rq(); |
1871 | ||
4866cde0 NP |
1872 | finish_task_switch(rq, prev); |
1873 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1874 | /* In this case, finish_task_switch does not reenable preemption */ | |
1875 | preempt_enable(); | |
1876 | #endif | |
1da177e4 LT |
1877 | if (current->set_child_tid) |
1878 | put_user(current->pid, current->set_child_tid); | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * context_switch - switch to the new MM and the new | |
1883 | * thread's register state. | |
1884 | */ | |
dd41f596 | 1885 | static inline void |
70b97a7f | 1886 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1887 | struct task_struct *next) |
1da177e4 | 1888 | { |
dd41f596 | 1889 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1890 | |
e107be36 | 1891 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1892 | mm = next->mm; |
1893 | oldmm = prev->active_mm; | |
9226d125 ZA |
1894 | /* |
1895 | * For paravirt, this is coupled with an exit in switch_to to | |
1896 | * combine the page table reload and the switch backend into | |
1897 | * one hypercall. | |
1898 | */ | |
1899 | arch_enter_lazy_cpu_mode(); | |
1900 | ||
dd41f596 | 1901 | if (unlikely(!mm)) { |
1da177e4 LT |
1902 | next->active_mm = oldmm; |
1903 | atomic_inc(&oldmm->mm_count); | |
1904 | enter_lazy_tlb(oldmm, next); | |
1905 | } else | |
1906 | switch_mm(oldmm, mm, next); | |
1907 | ||
dd41f596 | 1908 | if (unlikely(!prev->mm)) { |
1da177e4 | 1909 | prev->active_mm = NULL; |
1da177e4 LT |
1910 | rq->prev_mm = oldmm; |
1911 | } | |
3a5f5e48 IM |
1912 | /* |
1913 | * Since the runqueue lock will be released by the next | |
1914 | * task (which is an invalid locking op but in the case | |
1915 | * of the scheduler it's an obvious special-case), so we | |
1916 | * do an early lockdep release here: | |
1917 | */ | |
1918 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1919 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1920 | #endif |
1da177e4 LT |
1921 | |
1922 | /* Here we just switch the register state and the stack. */ | |
1923 | switch_to(prev, next, prev); | |
1924 | ||
dd41f596 IM |
1925 | barrier(); |
1926 | /* | |
1927 | * this_rq must be evaluated again because prev may have moved | |
1928 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1929 | * frame will be invalid. | |
1930 | */ | |
1931 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1932 | } |
1933 | ||
1934 | /* | |
1935 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1936 | * | |
1937 | * externally visible scheduler statistics: current number of runnable | |
1938 | * threads, current number of uninterruptible-sleeping threads, total | |
1939 | * number of context switches performed since bootup. | |
1940 | */ | |
1941 | unsigned long nr_running(void) | |
1942 | { | |
1943 | unsigned long i, sum = 0; | |
1944 | ||
1945 | for_each_online_cpu(i) | |
1946 | sum += cpu_rq(i)->nr_running; | |
1947 | ||
1948 | return sum; | |
1949 | } | |
1950 | ||
1951 | unsigned long nr_uninterruptible(void) | |
1952 | { | |
1953 | unsigned long i, sum = 0; | |
1954 | ||
0a945022 | 1955 | for_each_possible_cpu(i) |
1da177e4 LT |
1956 | sum += cpu_rq(i)->nr_uninterruptible; |
1957 | ||
1958 | /* | |
1959 | * Since we read the counters lockless, it might be slightly | |
1960 | * inaccurate. Do not allow it to go below zero though: | |
1961 | */ | |
1962 | if (unlikely((long)sum < 0)) | |
1963 | sum = 0; | |
1964 | ||
1965 | return sum; | |
1966 | } | |
1967 | ||
1968 | unsigned long long nr_context_switches(void) | |
1969 | { | |
cc94abfc SR |
1970 | int i; |
1971 | unsigned long long sum = 0; | |
1da177e4 | 1972 | |
0a945022 | 1973 | for_each_possible_cpu(i) |
1da177e4 LT |
1974 | sum += cpu_rq(i)->nr_switches; |
1975 | ||
1976 | return sum; | |
1977 | } | |
1978 | ||
1979 | unsigned long nr_iowait(void) | |
1980 | { | |
1981 | unsigned long i, sum = 0; | |
1982 | ||
0a945022 | 1983 | for_each_possible_cpu(i) |
1da177e4 LT |
1984 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
1985 | ||
1986 | return sum; | |
1987 | } | |
1988 | ||
db1b1fef JS |
1989 | unsigned long nr_active(void) |
1990 | { | |
1991 | unsigned long i, running = 0, uninterruptible = 0; | |
1992 | ||
1993 | for_each_online_cpu(i) { | |
1994 | running += cpu_rq(i)->nr_running; | |
1995 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
1996 | } | |
1997 | ||
1998 | if (unlikely((long)uninterruptible < 0)) | |
1999 | uninterruptible = 0; | |
2000 | ||
2001 | return running + uninterruptible; | |
2002 | } | |
2003 | ||
48f24c4d | 2004 | /* |
dd41f596 IM |
2005 | * Update rq->cpu_load[] statistics. This function is usually called every |
2006 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2007 | */ |
dd41f596 | 2008 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2009 | { |
495eca49 | 2010 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2011 | int i, scale; |
2012 | ||
2013 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2014 | |
2015 | /* Update our load: */ | |
2016 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2017 | unsigned long old_load, new_load; | |
2018 | ||
2019 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2020 | ||
2021 | old_load = this_rq->cpu_load[i]; | |
2022 | new_load = this_load; | |
a25707f3 IM |
2023 | /* |
2024 | * Round up the averaging division if load is increasing. This | |
2025 | * prevents us from getting stuck on 9 if the load is 10, for | |
2026 | * example. | |
2027 | */ | |
2028 | if (new_load > old_load) | |
2029 | new_load += scale-1; | |
dd41f596 IM |
2030 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2031 | } | |
48f24c4d IM |
2032 | } |
2033 | ||
dd41f596 IM |
2034 | #ifdef CONFIG_SMP |
2035 | ||
1da177e4 LT |
2036 | /* |
2037 | * double_rq_lock - safely lock two runqueues | |
2038 | * | |
2039 | * Note this does not disable interrupts like task_rq_lock, | |
2040 | * you need to do so manually before calling. | |
2041 | */ | |
70b97a7f | 2042 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2043 | __acquires(rq1->lock) |
2044 | __acquires(rq2->lock) | |
2045 | { | |
054b9108 | 2046 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2047 | if (rq1 == rq2) { |
2048 | spin_lock(&rq1->lock); | |
2049 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2050 | } else { | |
c96d145e | 2051 | if (rq1 < rq2) { |
1da177e4 LT |
2052 | spin_lock(&rq1->lock); |
2053 | spin_lock(&rq2->lock); | |
2054 | } else { | |
2055 | spin_lock(&rq2->lock); | |
2056 | spin_lock(&rq1->lock); | |
2057 | } | |
2058 | } | |
6e82a3be IM |
2059 | update_rq_clock(rq1); |
2060 | update_rq_clock(rq2); | |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | /* | |
2064 | * double_rq_unlock - safely unlock two runqueues | |
2065 | * | |
2066 | * Note this does not restore interrupts like task_rq_unlock, | |
2067 | * you need to do so manually after calling. | |
2068 | */ | |
70b97a7f | 2069 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2070 | __releases(rq1->lock) |
2071 | __releases(rq2->lock) | |
2072 | { | |
2073 | spin_unlock(&rq1->lock); | |
2074 | if (rq1 != rq2) | |
2075 | spin_unlock(&rq2->lock); | |
2076 | else | |
2077 | __release(rq2->lock); | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2082 | */ | |
70b97a7f | 2083 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2084 | __releases(this_rq->lock) |
2085 | __acquires(busiest->lock) | |
2086 | __acquires(this_rq->lock) | |
2087 | { | |
054b9108 KK |
2088 | if (unlikely(!irqs_disabled())) { |
2089 | /* printk() doesn't work good under rq->lock */ | |
2090 | spin_unlock(&this_rq->lock); | |
2091 | BUG_ON(1); | |
2092 | } | |
1da177e4 | 2093 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2094 | if (busiest < this_rq) { |
1da177e4 LT |
2095 | spin_unlock(&this_rq->lock); |
2096 | spin_lock(&busiest->lock); | |
2097 | spin_lock(&this_rq->lock); | |
2098 | } else | |
2099 | spin_lock(&busiest->lock); | |
2100 | } | |
2101 | } | |
2102 | ||
1da177e4 LT |
2103 | /* |
2104 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2105 | * This is accomplished by forcing the cpu_allowed mask to only | |
2106 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then | |
2107 | * the cpu_allowed mask is restored. | |
2108 | */ | |
36c8b586 | 2109 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2110 | { |
70b97a7f | 2111 | struct migration_req req; |
1da177e4 | 2112 | unsigned long flags; |
70b97a7f | 2113 | struct rq *rq; |
1da177e4 LT |
2114 | |
2115 | rq = task_rq_lock(p, &flags); | |
2116 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2117 | || unlikely(cpu_is_offline(dest_cpu))) | |
2118 | goto out; | |
2119 | ||
2120 | /* force the process onto the specified CPU */ | |
2121 | if (migrate_task(p, dest_cpu, &req)) { | |
2122 | /* Need to wait for migration thread (might exit: take ref). */ | |
2123 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2124 | |
1da177e4 LT |
2125 | get_task_struct(mt); |
2126 | task_rq_unlock(rq, &flags); | |
2127 | wake_up_process(mt); | |
2128 | put_task_struct(mt); | |
2129 | wait_for_completion(&req.done); | |
36c8b586 | 2130 | |
1da177e4 LT |
2131 | return; |
2132 | } | |
2133 | out: | |
2134 | task_rq_unlock(rq, &flags); | |
2135 | } | |
2136 | ||
2137 | /* | |
476d139c NP |
2138 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2139 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2140 | */ |
2141 | void sched_exec(void) | |
2142 | { | |
1da177e4 | 2143 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2144 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2145 | put_cpu(); |
476d139c NP |
2146 | if (new_cpu != this_cpu) |
2147 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2148 | } |
2149 | ||
2150 | /* | |
2151 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2152 | * Both runqueues must be locked. | |
2153 | */ | |
dd41f596 IM |
2154 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2155 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2156 | { |
2e1cb74a | 2157 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2158 | set_task_cpu(p, this_cpu); |
dd41f596 | 2159 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2160 | /* |
2161 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2162 | * to be always true for them. | |
2163 | */ | |
dd41f596 | 2164 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2165 | } |
2166 | ||
2167 | /* | |
2168 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2169 | */ | |
858119e1 | 2170 | static |
70b97a7f | 2171 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2172 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2173 | int *all_pinned) |
1da177e4 LT |
2174 | { |
2175 | /* | |
2176 | * We do not migrate tasks that are: | |
2177 | * 1) running (obviously), or | |
2178 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2179 | * 3) are cache-hot on their current CPU. | |
2180 | */ | |
cc367732 IM |
2181 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2182 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2183 | return 0; |
cc367732 | 2184 | } |
81026794 NP |
2185 | *all_pinned = 0; |
2186 | ||
cc367732 IM |
2187 | if (task_running(rq, p)) { |
2188 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2189 | return 0; |
cc367732 | 2190 | } |
1da177e4 | 2191 | |
da84d961 IM |
2192 | /* |
2193 | * Aggressive migration if: | |
2194 | * 1) task is cache cold, or | |
2195 | * 2) too many balance attempts have failed. | |
2196 | */ | |
2197 | ||
6bc1665b IM |
2198 | if (!task_hot(p, rq->clock, sd) || |
2199 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2200 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2201 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2202 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2203 | schedstat_inc(p, se.nr_forced_migrations); |
2204 | } | |
da84d961 IM |
2205 | #endif |
2206 | return 1; | |
2207 | } | |
2208 | ||
cc367732 IM |
2209 | if (task_hot(p, rq->clock, sd)) { |
2210 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2211 | return 0; |
cc367732 | 2212 | } |
1da177e4 LT |
2213 | return 1; |
2214 | } | |
2215 | ||
dd41f596 | 2216 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
2dd73a4f | 2217 | unsigned long max_nr_move, unsigned long max_load_move, |
d15bcfdb | 2218 | struct sched_domain *sd, enum cpu_idle_type idle, |
dd41f596 | 2219 | int *all_pinned, unsigned long *load_moved, |
a4ac01c3 | 2220 | int *this_best_prio, struct rq_iterator *iterator) |
1da177e4 | 2221 | { |
dd41f596 IM |
2222 | int pulled = 0, pinned = 0, skip_for_load; |
2223 | struct task_struct *p; | |
2224 | long rem_load_move = max_load_move; | |
1da177e4 | 2225 | |
2dd73a4f | 2226 | if (max_nr_move == 0 || max_load_move == 0) |
1da177e4 LT |
2227 | goto out; |
2228 | ||
81026794 NP |
2229 | pinned = 1; |
2230 | ||
1da177e4 | 2231 | /* |
dd41f596 | 2232 | * Start the load-balancing iterator: |
1da177e4 | 2233 | */ |
dd41f596 IM |
2234 | p = iterator->start(iterator->arg); |
2235 | next: | |
2236 | if (!p) | |
1da177e4 | 2237 | goto out; |
50ddd969 PW |
2238 | /* |
2239 | * To help distribute high priority tasks accross CPUs we don't | |
2240 | * skip a task if it will be the highest priority task (i.e. smallest | |
2241 | * prio value) on its new queue regardless of its load weight | |
2242 | */ | |
dd41f596 IM |
2243 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2244 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2245 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2246 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2247 | p = iterator->next(iterator->arg); |
2248 | goto next; | |
1da177e4 LT |
2249 | } |
2250 | ||
dd41f596 | 2251 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2252 | pulled++; |
dd41f596 | 2253 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2254 | |
2dd73a4f PW |
2255 | /* |
2256 | * We only want to steal up to the prescribed number of tasks | |
2257 | * and the prescribed amount of weighted load. | |
2258 | */ | |
2259 | if (pulled < max_nr_move && rem_load_move > 0) { | |
a4ac01c3 PW |
2260 | if (p->prio < *this_best_prio) |
2261 | *this_best_prio = p->prio; | |
dd41f596 IM |
2262 | p = iterator->next(iterator->arg); |
2263 | goto next; | |
1da177e4 LT |
2264 | } |
2265 | out: | |
2266 | /* | |
2267 | * Right now, this is the only place pull_task() is called, | |
2268 | * so we can safely collect pull_task() stats here rather than | |
2269 | * inside pull_task(). | |
2270 | */ | |
2271 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2272 | |
2273 | if (all_pinned) | |
2274 | *all_pinned = pinned; | |
dd41f596 | 2275 | *load_moved = max_load_move - rem_load_move; |
1da177e4 LT |
2276 | return pulled; |
2277 | } | |
2278 | ||
dd41f596 | 2279 | /* |
43010659 PW |
2280 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2281 | * this_rq, as part of a balancing operation within domain "sd". | |
2282 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2283 | * |
2284 | * Called with both runqueues locked. | |
2285 | */ | |
2286 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2287 | unsigned long max_load_move, |
dd41f596 IM |
2288 | struct sched_domain *sd, enum cpu_idle_type idle, |
2289 | int *all_pinned) | |
2290 | { | |
5522d5d5 | 2291 | const struct sched_class *class = sched_class_highest; |
43010659 | 2292 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2293 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2294 | |
2295 | do { | |
43010659 PW |
2296 | total_load_moved += |
2297 | class->load_balance(this_rq, this_cpu, busiest, | |
2298 | ULONG_MAX, max_load_move - total_load_moved, | |
a4ac01c3 | 2299 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2300 | class = class->next; |
43010659 | 2301 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2302 | |
43010659 PW |
2303 | return total_load_moved > 0; |
2304 | } | |
2305 | ||
2306 | /* | |
2307 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2308 | * part of active balancing operations within "domain". | |
2309 | * Returns 1 if successful and 0 otherwise. | |
2310 | * | |
2311 | * Called with both runqueues locked. | |
2312 | */ | |
2313 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2314 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2315 | { | |
5522d5d5 | 2316 | const struct sched_class *class; |
a4ac01c3 | 2317 | int this_best_prio = MAX_PRIO; |
43010659 PW |
2318 | |
2319 | for (class = sched_class_highest; class; class = class->next) | |
2320 | if (class->load_balance(this_rq, this_cpu, busiest, | |
a4ac01c3 PW |
2321 | 1, ULONG_MAX, sd, idle, NULL, |
2322 | &this_best_prio)) | |
43010659 PW |
2323 | return 1; |
2324 | ||
2325 | return 0; | |
dd41f596 IM |
2326 | } |
2327 | ||
1da177e4 LT |
2328 | /* |
2329 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2330 | * domain. It calculates and returns the amount of weighted load which |
2331 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2332 | */ |
2333 | static struct sched_group * | |
2334 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2335 | unsigned long *imbalance, enum cpu_idle_type idle, |
2336 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2337 | { |
2338 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2339 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2340 | unsigned long max_pull; |
2dd73a4f PW |
2341 | unsigned long busiest_load_per_task, busiest_nr_running; |
2342 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2343 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2344 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2345 | int power_savings_balance = 1; | |
2346 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2347 | unsigned long min_nr_running = ULONG_MAX; | |
2348 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2349 | #endif | |
1da177e4 LT |
2350 | |
2351 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2352 | busiest_load_per_task = busiest_nr_running = 0; |
2353 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2354 | if (idle == CPU_NOT_IDLE) |
7897986b | 2355 | load_idx = sd->busy_idx; |
d15bcfdb | 2356 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2357 | load_idx = sd->newidle_idx; |
2358 | else | |
2359 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2360 | |
2361 | do { | |
908a7c1b | 2362 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2363 | int local_group; |
2364 | int i; | |
908a7c1b | 2365 | int __group_imb = 0; |
783609c6 | 2366 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2367 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2368 | |
2369 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2370 | ||
783609c6 SS |
2371 | if (local_group) |
2372 | balance_cpu = first_cpu(group->cpumask); | |
2373 | ||
1da177e4 | 2374 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2375 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2376 | max_cpu_load = 0; |
2377 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2378 | |
2379 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2380 | struct rq *rq; |
2381 | ||
2382 | if (!cpu_isset(i, *cpus)) | |
2383 | continue; | |
2384 | ||
2385 | rq = cpu_rq(i); | |
2dd73a4f | 2386 | |
9439aab8 | 2387 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2388 | *sd_idle = 0; |
2389 | ||
1da177e4 | 2390 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2391 | if (local_group) { |
2392 | if (idle_cpu(i) && !first_idle_cpu) { | |
2393 | first_idle_cpu = 1; | |
2394 | balance_cpu = i; | |
2395 | } | |
2396 | ||
a2000572 | 2397 | load = target_load(i, load_idx); |
908a7c1b | 2398 | } else { |
a2000572 | 2399 | load = source_load(i, load_idx); |
908a7c1b KC |
2400 | if (load > max_cpu_load) |
2401 | max_cpu_load = load; | |
2402 | if (min_cpu_load > load) | |
2403 | min_cpu_load = load; | |
2404 | } | |
1da177e4 LT |
2405 | |
2406 | avg_load += load; | |
2dd73a4f | 2407 | sum_nr_running += rq->nr_running; |
dd41f596 | 2408 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2409 | } |
2410 | ||
783609c6 SS |
2411 | /* |
2412 | * First idle cpu or the first cpu(busiest) in this sched group | |
2413 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2414 | * domains. In the newly idle case, we will allow all the cpu's |
2415 | * to do the newly idle load balance. | |
783609c6 | 2416 | */ |
9439aab8 SS |
2417 | if (idle != CPU_NEWLY_IDLE && local_group && |
2418 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2419 | *balance = 0; |
2420 | goto ret; | |
2421 | } | |
2422 | ||
1da177e4 | 2423 | total_load += avg_load; |
5517d86b | 2424 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2425 | |
2426 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2427 | avg_load = sg_div_cpu_power(group, |
2428 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2429 | |
908a7c1b KC |
2430 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2431 | __group_imb = 1; | |
2432 | ||
5517d86b | 2433 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2434 | |
1da177e4 LT |
2435 | if (local_group) { |
2436 | this_load = avg_load; | |
2437 | this = group; | |
2dd73a4f PW |
2438 | this_nr_running = sum_nr_running; |
2439 | this_load_per_task = sum_weighted_load; | |
2440 | } else if (avg_load > max_load && | |
908a7c1b | 2441 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2442 | max_load = avg_load; |
2443 | busiest = group; | |
2dd73a4f PW |
2444 | busiest_nr_running = sum_nr_running; |
2445 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2446 | group_imb = __group_imb; |
1da177e4 | 2447 | } |
5c45bf27 SS |
2448 | |
2449 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2450 | /* | |
2451 | * Busy processors will not participate in power savings | |
2452 | * balance. | |
2453 | */ | |
dd41f596 IM |
2454 | if (idle == CPU_NOT_IDLE || |
2455 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2456 | goto group_next; | |
5c45bf27 SS |
2457 | |
2458 | /* | |
2459 | * If the local group is idle or completely loaded | |
2460 | * no need to do power savings balance at this domain | |
2461 | */ | |
2462 | if (local_group && (this_nr_running >= group_capacity || | |
2463 | !this_nr_running)) | |
2464 | power_savings_balance = 0; | |
2465 | ||
dd41f596 | 2466 | /* |
5c45bf27 SS |
2467 | * If a group is already running at full capacity or idle, |
2468 | * don't include that group in power savings calculations | |
dd41f596 IM |
2469 | */ |
2470 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2471 | || !sum_nr_running) |
dd41f596 | 2472 | goto group_next; |
5c45bf27 | 2473 | |
dd41f596 | 2474 | /* |
5c45bf27 | 2475 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2476 | * This is the group from where we need to pick up the load |
2477 | * for saving power | |
2478 | */ | |
2479 | if ((sum_nr_running < min_nr_running) || | |
2480 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2481 | first_cpu(group->cpumask) < |
2482 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2483 | group_min = group; |
2484 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2485 | min_load_per_task = sum_weighted_load / |
2486 | sum_nr_running; | |
dd41f596 | 2487 | } |
5c45bf27 | 2488 | |
dd41f596 | 2489 | /* |
5c45bf27 | 2490 | * Calculate the group which is almost near its |
dd41f596 IM |
2491 | * capacity but still has some space to pick up some load |
2492 | * from other group and save more power | |
2493 | */ | |
2494 | if (sum_nr_running <= group_capacity - 1) { | |
2495 | if (sum_nr_running > leader_nr_running || | |
2496 | (sum_nr_running == leader_nr_running && | |
2497 | first_cpu(group->cpumask) > | |
2498 | first_cpu(group_leader->cpumask))) { | |
2499 | group_leader = group; | |
2500 | leader_nr_running = sum_nr_running; | |
2501 | } | |
48f24c4d | 2502 | } |
5c45bf27 SS |
2503 | group_next: |
2504 | #endif | |
1da177e4 LT |
2505 | group = group->next; |
2506 | } while (group != sd->groups); | |
2507 | ||
2dd73a4f | 2508 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2509 | goto out_balanced; |
2510 | ||
2511 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2512 | ||
2513 | if (this_load >= avg_load || | |
2514 | 100*max_load <= sd->imbalance_pct*this_load) | |
2515 | goto out_balanced; | |
2516 | ||
2dd73a4f | 2517 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2518 | if (group_imb) |
2519 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2520 | ||
1da177e4 LT |
2521 | /* |
2522 | * We're trying to get all the cpus to the average_load, so we don't | |
2523 | * want to push ourselves above the average load, nor do we wish to | |
2524 | * reduce the max loaded cpu below the average load, as either of these | |
2525 | * actions would just result in more rebalancing later, and ping-pong | |
2526 | * tasks around. Thus we look for the minimum possible imbalance. | |
2527 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2528 | * be counted as no imbalance for these purposes -- we can't fix that | |
2529 | * by pulling tasks to us. Be careful of negative numbers as they'll | |
2530 | * appear as very large values with unsigned longs. | |
2531 | */ | |
2dd73a4f PW |
2532 | if (max_load <= busiest_load_per_task) |
2533 | goto out_balanced; | |
2534 | ||
2535 | /* | |
2536 | * In the presence of smp nice balancing, certain scenarios can have | |
2537 | * max load less than avg load(as we skip the groups at or below | |
2538 | * its cpu_power, while calculating max_load..) | |
2539 | */ | |
2540 | if (max_load < avg_load) { | |
2541 | *imbalance = 0; | |
2542 | goto small_imbalance; | |
2543 | } | |
0c117f1b SS |
2544 | |
2545 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2546 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2547 | |
1da177e4 | 2548 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2549 | *imbalance = min(max_pull * busiest->__cpu_power, |
2550 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2551 | / SCHED_LOAD_SCALE; |
2552 | ||
2dd73a4f PW |
2553 | /* |
2554 | * if *imbalance is less than the average load per runnable task | |
2555 | * there is no gaurantee that any tasks will be moved so we'll have | |
2556 | * a think about bumping its value to force at least one task to be | |
2557 | * moved | |
2558 | */ | |
7fd0d2dd | 2559 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2560 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2561 | unsigned int imbn; |
2562 | ||
2563 | small_imbalance: | |
2564 | pwr_move = pwr_now = 0; | |
2565 | imbn = 2; | |
2566 | if (this_nr_running) { | |
2567 | this_load_per_task /= this_nr_running; | |
2568 | if (busiest_load_per_task > this_load_per_task) | |
2569 | imbn = 1; | |
2570 | } else | |
2571 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2572 | |
dd41f596 IM |
2573 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2574 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2575 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2576 | return busiest; |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * OK, we don't have enough imbalance to justify moving tasks, | |
2581 | * however we may be able to increase total CPU power used by | |
2582 | * moving them. | |
2583 | */ | |
2584 | ||
5517d86b ED |
2585 | pwr_now += busiest->__cpu_power * |
2586 | min(busiest_load_per_task, max_load); | |
2587 | pwr_now += this->__cpu_power * | |
2588 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2589 | pwr_now /= SCHED_LOAD_SCALE; |
2590 | ||
2591 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2592 | tmp = sg_div_cpu_power(busiest, |
2593 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2594 | if (max_load > tmp) |
5517d86b | 2595 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2596 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2597 | |
2598 | /* Amount of load we'd add */ | |
5517d86b | 2599 | if (max_load * busiest->__cpu_power < |
33859f7f | 2600 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2601 | tmp = sg_div_cpu_power(this, |
2602 | max_load * busiest->__cpu_power); | |
1da177e4 | 2603 | else |
5517d86b ED |
2604 | tmp = sg_div_cpu_power(this, |
2605 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2606 | pwr_move += this->__cpu_power * | |
2607 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2608 | pwr_move /= SCHED_LOAD_SCALE; |
2609 | ||
2610 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2611 | if (pwr_move > pwr_now) |
2612 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2613 | } |
2614 | ||
1da177e4 LT |
2615 | return busiest; |
2616 | ||
2617 | out_balanced: | |
5c45bf27 | 2618 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2619 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2620 | goto ret; |
1da177e4 | 2621 | |
5c45bf27 SS |
2622 | if (this == group_leader && group_leader != group_min) { |
2623 | *imbalance = min_load_per_task; | |
2624 | return group_min; | |
2625 | } | |
5c45bf27 | 2626 | #endif |
783609c6 | 2627 | ret: |
1da177e4 LT |
2628 | *imbalance = 0; |
2629 | return NULL; | |
2630 | } | |
2631 | ||
2632 | /* | |
2633 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2634 | */ | |
70b97a7f | 2635 | static struct rq * |
d15bcfdb | 2636 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2637 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2638 | { |
70b97a7f | 2639 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2640 | unsigned long max_load = 0; |
1da177e4 LT |
2641 | int i; |
2642 | ||
2643 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2644 | unsigned long wl; |
0a2966b4 CL |
2645 | |
2646 | if (!cpu_isset(i, *cpus)) | |
2647 | continue; | |
2648 | ||
48f24c4d | 2649 | rq = cpu_rq(i); |
dd41f596 | 2650 | wl = weighted_cpuload(i); |
2dd73a4f | 2651 | |
dd41f596 | 2652 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2653 | continue; |
1da177e4 | 2654 | |
dd41f596 IM |
2655 | if (wl > max_load) { |
2656 | max_load = wl; | |
48f24c4d | 2657 | busiest = rq; |
1da177e4 LT |
2658 | } |
2659 | } | |
2660 | ||
2661 | return busiest; | |
2662 | } | |
2663 | ||
77391d71 NP |
2664 | /* |
2665 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2666 | * so long as it is large enough. | |
2667 | */ | |
2668 | #define MAX_PINNED_INTERVAL 512 | |
2669 | ||
1da177e4 LT |
2670 | /* |
2671 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2672 | * tasks if there is an imbalance. | |
1da177e4 | 2673 | */ |
70b97a7f | 2674 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2675 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2676 | int *balance) |
1da177e4 | 2677 | { |
43010659 | 2678 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2679 | struct sched_group *group; |
1da177e4 | 2680 | unsigned long imbalance; |
70b97a7f | 2681 | struct rq *busiest; |
0a2966b4 | 2682 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2683 | unsigned long flags; |
5969fe06 | 2684 | |
89c4710e SS |
2685 | /* |
2686 | * When power savings policy is enabled for the parent domain, idle | |
2687 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2688 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2689 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2690 | */ |
d15bcfdb | 2691 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2692 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2693 | sd_idle = 1; |
1da177e4 | 2694 | |
2d72376b | 2695 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2696 | |
0a2966b4 CL |
2697 | redo: |
2698 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2699 | &cpus, balance); |
2700 | ||
06066714 | 2701 | if (*balance == 0) |
783609c6 | 2702 | goto out_balanced; |
783609c6 | 2703 | |
1da177e4 LT |
2704 | if (!group) { |
2705 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2706 | goto out_balanced; | |
2707 | } | |
2708 | ||
0a2966b4 | 2709 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2710 | if (!busiest) { |
2711 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2712 | goto out_balanced; | |
2713 | } | |
2714 | ||
db935dbd | 2715 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2716 | |
2717 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2718 | ||
43010659 | 2719 | ld_moved = 0; |
1da177e4 LT |
2720 | if (busiest->nr_running > 1) { |
2721 | /* | |
2722 | * Attempt to move tasks. If find_busiest_group has found | |
2723 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2724 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2725 | * correctly treated as an imbalance. |
2726 | */ | |
fe2eea3f | 2727 | local_irq_save(flags); |
e17224bf | 2728 | double_rq_lock(this_rq, busiest); |
43010659 | 2729 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2730 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2731 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2732 | local_irq_restore(flags); |
81026794 | 2733 | |
46cb4b7c SS |
2734 | /* |
2735 | * some other cpu did the load balance for us. | |
2736 | */ | |
43010659 | 2737 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2738 | resched_cpu(this_cpu); |
2739 | ||
81026794 | 2740 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2741 | if (unlikely(all_pinned)) { |
2742 | cpu_clear(cpu_of(busiest), cpus); | |
2743 | if (!cpus_empty(cpus)) | |
2744 | goto redo; | |
81026794 | 2745 | goto out_balanced; |
0a2966b4 | 2746 | } |
1da177e4 | 2747 | } |
81026794 | 2748 | |
43010659 | 2749 | if (!ld_moved) { |
1da177e4 LT |
2750 | schedstat_inc(sd, lb_failed[idle]); |
2751 | sd->nr_balance_failed++; | |
2752 | ||
2753 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2754 | |
fe2eea3f | 2755 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2756 | |
2757 | /* don't kick the migration_thread, if the curr | |
2758 | * task on busiest cpu can't be moved to this_cpu | |
2759 | */ | |
2760 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2761 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2762 | all_pinned = 1; |
2763 | goto out_one_pinned; | |
2764 | } | |
2765 | ||
1da177e4 LT |
2766 | if (!busiest->active_balance) { |
2767 | busiest->active_balance = 1; | |
2768 | busiest->push_cpu = this_cpu; | |
81026794 | 2769 | active_balance = 1; |
1da177e4 | 2770 | } |
fe2eea3f | 2771 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2772 | if (active_balance) |
1da177e4 LT |
2773 | wake_up_process(busiest->migration_thread); |
2774 | ||
2775 | /* | |
2776 | * We've kicked active balancing, reset the failure | |
2777 | * counter. | |
2778 | */ | |
39507451 | 2779 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2780 | } |
81026794 | 2781 | } else |
1da177e4 LT |
2782 | sd->nr_balance_failed = 0; |
2783 | ||
81026794 | 2784 | if (likely(!active_balance)) { |
1da177e4 LT |
2785 | /* We were unbalanced, so reset the balancing interval */ |
2786 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2787 | } else { |
2788 | /* | |
2789 | * If we've begun active balancing, start to back off. This | |
2790 | * case may not be covered by the all_pinned logic if there | |
2791 | * is only 1 task on the busy runqueue (because we don't call | |
2792 | * move_tasks). | |
2793 | */ | |
2794 | if (sd->balance_interval < sd->max_interval) | |
2795 | sd->balance_interval *= 2; | |
1da177e4 LT |
2796 | } |
2797 | ||
43010659 | 2798 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2799 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2800 | return -1; |
43010659 | 2801 | return ld_moved; |
1da177e4 LT |
2802 | |
2803 | out_balanced: | |
1da177e4 LT |
2804 | schedstat_inc(sd, lb_balanced[idle]); |
2805 | ||
16cfb1c0 | 2806 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2807 | |
2808 | out_one_pinned: | |
1da177e4 | 2809 | /* tune up the balancing interval */ |
77391d71 NP |
2810 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2811 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2812 | sd->balance_interval *= 2; |
2813 | ||
48f24c4d | 2814 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2815 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2816 | return -1; |
1da177e4 LT |
2817 | return 0; |
2818 | } | |
2819 | ||
2820 | /* | |
2821 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2822 | * tasks if there is an imbalance. | |
2823 | * | |
d15bcfdb | 2824 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2825 | * this_rq is locked. |
2826 | */ | |
48f24c4d | 2827 | static int |
70b97a7f | 2828 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2829 | { |
2830 | struct sched_group *group; | |
70b97a7f | 2831 | struct rq *busiest = NULL; |
1da177e4 | 2832 | unsigned long imbalance; |
43010659 | 2833 | int ld_moved = 0; |
5969fe06 | 2834 | int sd_idle = 0; |
969bb4e4 | 2835 | int all_pinned = 0; |
0a2966b4 | 2836 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2837 | |
89c4710e SS |
2838 | /* |
2839 | * When power savings policy is enabled for the parent domain, idle | |
2840 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2841 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2842 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2843 | */ |
2844 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2845 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2846 | sd_idle = 1; |
1da177e4 | 2847 | |
2d72376b | 2848 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2849 | redo: |
d15bcfdb | 2850 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2851 | &sd_idle, &cpus, NULL); |
1da177e4 | 2852 | if (!group) { |
d15bcfdb | 2853 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2854 | goto out_balanced; |
1da177e4 LT |
2855 | } |
2856 | ||
d15bcfdb | 2857 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2858 | &cpus); |
db935dbd | 2859 | if (!busiest) { |
d15bcfdb | 2860 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2861 | goto out_balanced; |
1da177e4 LT |
2862 | } |
2863 | ||
db935dbd NP |
2864 | BUG_ON(busiest == this_rq); |
2865 | ||
d15bcfdb | 2866 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2867 | |
43010659 | 2868 | ld_moved = 0; |
d6d5cfaf NP |
2869 | if (busiest->nr_running > 1) { |
2870 | /* Attempt to move tasks */ | |
2871 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2872 | /* this_rq->clock is already updated */ |
2873 | update_rq_clock(busiest); | |
43010659 | 2874 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2875 | imbalance, sd, CPU_NEWLY_IDLE, |
2876 | &all_pinned); | |
d6d5cfaf | 2877 | spin_unlock(&busiest->lock); |
0a2966b4 | 2878 | |
969bb4e4 | 2879 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2880 | cpu_clear(cpu_of(busiest), cpus); |
2881 | if (!cpus_empty(cpus)) | |
2882 | goto redo; | |
2883 | } | |
d6d5cfaf NP |
2884 | } |
2885 | ||
43010659 | 2886 | if (!ld_moved) { |
d15bcfdb | 2887 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2888 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2889 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2890 | return -1; |
2891 | } else | |
16cfb1c0 | 2892 | sd->nr_balance_failed = 0; |
1da177e4 | 2893 | |
43010659 | 2894 | return ld_moved; |
16cfb1c0 NP |
2895 | |
2896 | out_balanced: | |
d15bcfdb | 2897 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2898 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2899 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2900 | return -1; |
16cfb1c0 | 2901 | sd->nr_balance_failed = 0; |
48f24c4d | 2902 | |
16cfb1c0 | 2903 | return 0; |
1da177e4 LT |
2904 | } |
2905 | ||
2906 | /* | |
2907 | * idle_balance is called by schedule() if this_cpu is about to become | |
2908 | * idle. Attempts to pull tasks from other CPUs. | |
2909 | */ | |
70b97a7f | 2910 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2911 | { |
2912 | struct sched_domain *sd; | |
dd41f596 IM |
2913 | int pulled_task = -1; |
2914 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2915 | |
2916 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2917 | unsigned long interval; |
2918 | ||
2919 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2920 | continue; | |
2921 | ||
2922 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2923 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2924 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2925 | this_rq, sd); |
2926 | ||
2927 | interval = msecs_to_jiffies(sd->balance_interval); | |
2928 | if (time_after(next_balance, sd->last_balance + interval)) | |
2929 | next_balance = sd->last_balance + interval; | |
2930 | if (pulled_task) | |
2931 | break; | |
1da177e4 | 2932 | } |
dd41f596 | 2933 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2934 | /* |
2935 | * We are going idle. next_balance may be set based on | |
2936 | * a busy processor. So reset next_balance. | |
2937 | */ | |
2938 | this_rq->next_balance = next_balance; | |
dd41f596 | 2939 | } |
1da177e4 LT |
2940 | } |
2941 | ||
2942 | /* | |
2943 | * active_load_balance is run by migration threads. It pushes running tasks | |
2944 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2945 | * running on each physical CPU where possible, and avoids physical / | |
2946 | * logical imbalances. | |
2947 | * | |
2948 | * Called with busiest_rq locked. | |
2949 | */ | |
70b97a7f | 2950 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 2951 | { |
39507451 | 2952 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
2953 | struct sched_domain *sd; |
2954 | struct rq *target_rq; | |
39507451 | 2955 | |
48f24c4d | 2956 | /* Is there any task to move? */ |
39507451 | 2957 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
2958 | return; |
2959 | ||
2960 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
2961 | |
2962 | /* | |
39507451 NP |
2963 | * This condition is "impossible", if it occurs |
2964 | * we need to fix it. Originally reported by | |
2965 | * Bjorn Helgaas on a 128-cpu setup. | |
1da177e4 | 2966 | */ |
39507451 | 2967 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 2968 | |
39507451 NP |
2969 | /* move a task from busiest_rq to target_rq */ |
2970 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
2971 | update_rq_clock(busiest_rq); |
2972 | update_rq_clock(target_rq); | |
39507451 NP |
2973 | |
2974 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 2975 | for_each_domain(target_cpu, sd) { |
39507451 | 2976 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 2977 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 2978 | break; |
c96d145e | 2979 | } |
39507451 | 2980 | |
48f24c4d | 2981 | if (likely(sd)) { |
2d72376b | 2982 | schedstat_inc(sd, alb_count); |
39507451 | 2983 | |
43010659 PW |
2984 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
2985 | sd, CPU_IDLE)) | |
48f24c4d IM |
2986 | schedstat_inc(sd, alb_pushed); |
2987 | else | |
2988 | schedstat_inc(sd, alb_failed); | |
2989 | } | |
39507451 | 2990 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
2991 | } |
2992 | ||
46cb4b7c SS |
2993 | #ifdef CONFIG_NO_HZ |
2994 | static struct { | |
2995 | atomic_t load_balancer; | |
2996 | cpumask_t cpu_mask; | |
2997 | } nohz ____cacheline_aligned = { | |
2998 | .load_balancer = ATOMIC_INIT(-1), | |
2999 | .cpu_mask = CPU_MASK_NONE, | |
3000 | }; | |
3001 | ||
7835b98b | 3002 | /* |
46cb4b7c SS |
3003 | * This routine will try to nominate the ilb (idle load balancing) |
3004 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3005 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3006 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3007 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3008 | * arrives... | |
3009 | * | |
3010 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3011 | * for idle load balancing. ilb owner will still be part of | |
3012 | * nohz.cpu_mask.. | |
7835b98b | 3013 | * |
46cb4b7c SS |
3014 | * While stopping the tick, this cpu will become the ilb owner if there |
3015 | * is no other owner. And will be the owner till that cpu becomes busy | |
3016 | * or if all cpus in the system stop their ticks at which point | |
3017 | * there is no need for ilb owner. | |
3018 | * | |
3019 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3020 | * next busy scheduler_tick() | |
3021 | */ | |
3022 | int select_nohz_load_balancer(int stop_tick) | |
3023 | { | |
3024 | int cpu = smp_processor_id(); | |
3025 | ||
3026 | if (stop_tick) { | |
3027 | cpu_set(cpu, nohz.cpu_mask); | |
3028 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3029 | ||
3030 | /* | |
3031 | * If we are going offline and still the leader, give up! | |
3032 | */ | |
3033 | if (cpu_is_offline(cpu) && | |
3034 | atomic_read(&nohz.load_balancer) == cpu) { | |
3035 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3036 | BUG(); | |
3037 | return 0; | |
3038 | } | |
3039 | ||
3040 | /* time for ilb owner also to sleep */ | |
3041 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3042 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3043 | atomic_set(&nohz.load_balancer, -1); | |
3044 | return 0; | |
3045 | } | |
3046 | ||
3047 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3048 | /* make me the ilb owner */ | |
3049 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3050 | return 1; | |
3051 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3052 | return 1; | |
3053 | } else { | |
3054 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3055 | return 0; | |
3056 | ||
3057 | cpu_clear(cpu, nohz.cpu_mask); | |
3058 | ||
3059 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3060 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3061 | BUG(); | |
3062 | } | |
3063 | return 0; | |
3064 | } | |
3065 | #endif | |
3066 | ||
3067 | static DEFINE_SPINLOCK(balancing); | |
3068 | ||
3069 | /* | |
7835b98b CL |
3070 | * It checks each scheduling domain to see if it is due to be balanced, |
3071 | * and initiates a balancing operation if so. | |
3072 | * | |
3073 | * Balancing parameters are set up in arch_init_sched_domains. | |
3074 | */ | |
a9957449 | 3075 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3076 | { |
46cb4b7c SS |
3077 | int balance = 1; |
3078 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3079 | unsigned long interval; |
3080 | struct sched_domain *sd; | |
46cb4b7c | 3081 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3082 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3083 | int update_next_balance = 0; |
1da177e4 | 3084 | |
46cb4b7c | 3085 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3086 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3087 | continue; | |
3088 | ||
3089 | interval = sd->balance_interval; | |
d15bcfdb | 3090 | if (idle != CPU_IDLE) |
1da177e4 LT |
3091 | interval *= sd->busy_factor; |
3092 | ||
3093 | /* scale ms to jiffies */ | |
3094 | interval = msecs_to_jiffies(interval); | |
3095 | if (unlikely(!interval)) | |
3096 | interval = 1; | |
dd41f596 IM |
3097 | if (interval > HZ*NR_CPUS/10) |
3098 | interval = HZ*NR_CPUS/10; | |
3099 | ||
1da177e4 | 3100 | |
08c183f3 CL |
3101 | if (sd->flags & SD_SERIALIZE) { |
3102 | if (!spin_trylock(&balancing)) | |
3103 | goto out; | |
3104 | } | |
3105 | ||
c9819f45 | 3106 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3107 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3108 | /* |
3109 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3110 | * longer idle, or one of our SMT siblings is |
3111 | * not idle. | |
3112 | */ | |
d15bcfdb | 3113 | idle = CPU_NOT_IDLE; |
1da177e4 | 3114 | } |
1bd77f2d | 3115 | sd->last_balance = jiffies; |
1da177e4 | 3116 | } |
08c183f3 CL |
3117 | if (sd->flags & SD_SERIALIZE) |
3118 | spin_unlock(&balancing); | |
3119 | out: | |
f549da84 | 3120 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3121 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3122 | update_next_balance = 1; |
3123 | } | |
783609c6 SS |
3124 | |
3125 | /* | |
3126 | * Stop the load balance at this level. There is another | |
3127 | * CPU in our sched group which is doing load balancing more | |
3128 | * actively. | |
3129 | */ | |
3130 | if (!balance) | |
3131 | break; | |
1da177e4 | 3132 | } |
f549da84 SS |
3133 | |
3134 | /* | |
3135 | * next_balance will be updated only when there is a need. | |
3136 | * When the cpu is attached to null domain for ex, it will not be | |
3137 | * updated. | |
3138 | */ | |
3139 | if (likely(update_next_balance)) | |
3140 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3141 | } |
3142 | ||
3143 | /* | |
3144 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3145 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3146 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3147 | */ | |
3148 | static void run_rebalance_domains(struct softirq_action *h) | |
3149 | { | |
dd41f596 IM |
3150 | int this_cpu = smp_processor_id(); |
3151 | struct rq *this_rq = cpu_rq(this_cpu); | |
3152 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3153 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3154 | |
dd41f596 | 3155 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3156 | |
3157 | #ifdef CONFIG_NO_HZ | |
3158 | /* | |
3159 | * If this cpu is the owner for idle load balancing, then do the | |
3160 | * balancing on behalf of the other idle cpus whose ticks are | |
3161 | * stopped. | |
3162 | */ | |
dd41f596 IM |
3163 | if (this_rq->idle_at_tick && |
3164 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3165 | cpumask_t cpus = nohz.cpu_mask; |
3166 | struct rq *rq; | |
3167 | int balance_cpu; | |
3168 | ||
dd41f596 | 3169 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3170 | for_each_cpu_mask(balance_cpu, cpus) { |
3171 | /* | |
3172 | * If this cpu gets work to do, stop the load balancing | |
3173 | * work being done for other cpus. Next load | |
3174 | * balancing owner will pick it up. | |
3175 | */ | |
3176 | if (need_resched()) | |
3177 | break; | |
3178 | ||
de0cf899 | 3179 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3180 | |
3181 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3182 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3183 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3184 | } |
3185 | } | |
3186 | #endif | |
3187 | } | |
3188 | ||
3189 | /* | |
3190 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3191 | * | |
3192 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3193 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3194 | * if the whole system is idle. | |
3195 | */ | |
dd41f596 | 3196 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3197 | { |
46cb4b7c SS |
3198 | #ifdef CONFIG_NO_HZ |
3199 | /* | |
3200 | * If we were in the nohz mode recently and busy at the current | |
3201 | * scheduler tick, then check if we need to nominate new idle | |
3202 | * load balancer. | |
3203 | */ | |
3204 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3205 | rq->in_nohz_recently = 0; | |
3206 | ||
3207 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3208 | cpu_clear(cpu, nohz.cpu_mask); | |
3209 | atomic_set(&nohz.load_balancer, -1); | |
3210 | } | |
3211 | ||
3212 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3213 | /* | |
3214 | * simple selection for now: Nominate the | |
3215 | * first cpu in the nohz list to be the next | |
3216 | * ilb owner. | |
3217 | * | |
3218 | * TBD: Traverse the sched domains and nominate | |
3219 | * the nearest cpu in the nohz.cpu_mask. | |
3220 | */ | |
3221 | int ilb = first_cpu(nohz.cpu_mask); | |
3222 | ||
3223 | if (ilb != NR_CPUS) | |
3224 | resched_cpu(ilb); | |
3225 | } | |
3226 | } | |
3227 | ||
3228 | /* | |
3229 | * If this cpu is idle and doing idle load balancing for all the | |
3230 | * cpus with ticks stopped, is it time for that to stop? | |
3231 | */ | |
3232 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3233 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3234 | resched_cpu(cpu); | |
3235 | return; | |
3236 | } | |
3237 | ||
3238 | /* | |
3239 | * If this cpu is idle and the idle load balancing is done by | |
3240 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3241 | */ | |
3242 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3243 | cpu_isset(cpu, nohz.cpu_mask)) | |
3244 | return; | |
3245 | #endif | |
3246 | if (time_after_eq(jiffies, rq->next_balance)) | |
3247 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3248 | } |
dd41f596 IM |
3249 | |
3250 | #else /* CONFIG_SMP */ | |
3251 | ||
1da177e4 LT |
3252 | /* |
3253 | * on UP we do not need to balance between CPUs: | |
3254 | */ | |
70b97a7f | 3255 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3256 | { |
3257 | } | |
dd41f596 IM |
3258 | |
3259 | /* Avoid "used but not defined" warning on UP */ | |
3260 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3261 | unsigned long max_nr_move, unsigned long max_load_move, | |
3262 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3263 | int *all_pinned, unsigned long *load_moved, | |
a4ac01c3 | 3264 | int *this_best_prio, struct rq_iterator *iterator) |
dd41f596 IM |
3265 | { |
3266 | *load_moved = 0; | |
3267 | ||
3268 | return 0; | |
3269 | } | |
3270 | ||
1da177e4 LT |
3271 | #endif |
3272 | ||
1da177e4 LT |
3273 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3274 | ||
3275 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3276 | ||
3277 | /* | |
41b86e9c IM |
3278 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3279 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3280 | */ |
41b86e9c | 3281 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3282 | { |
1da177e4 | 3283 | unsigned long flags; |
41b86e9c IM |
3284 | u64 ns, delta_exec; |
3285 | struct rq *rq; | |
48f24c4d | 3286 | |
41b86e9c IM |
3287 | rq = task_rq_lock(p, &flags); |
3288 | ns = p->se.sum_exec_runtime; | |
3289 | if (rq->curr == p) { | |
a8e504d2 IM |
3290 | update_rq_clock(rq); |
3291 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3292 | if ((s64)delta_exec > 0) |
3293 | ns += delta_exec; | |
3294 | } | |
3295 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3296 | |
1da177e4 LT |
3297 | return ns; |
3298 | } | |
3299 | ||
1da177e4 LT |
3300 | /* |
3301 | * Account user cpu time to a process. | |
3302 | * @p: the process that the cpu time gets accounted to | |
3303 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3304 | * @cputime: the cpu time spent in user space since the last update | |
3305 | */ | |
3306 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3307 | { | |
3308 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3309 | cputime64_t tmp; | |
3310 | ||
3311 | p->utime = cputime_add(p->utime, cputime); | |
3312 | ||
3313 | /* Add user time to cpustat. */ | |
3314 | tmp = cputime_to_cputime64(cputime); | |
3315 | if (TASK_NICE(p) > 0) | |
3316 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3317 | else | |
3318 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3319 | } | |
3320 | ||
94886b84 LV |
3321 | /* |
3322 | * Account guest cpu time to a process. | |
3323 | * @p: the process that the cpu time gets accounted to | |
3324 | * @cputime: the cpu time spent in virtual machine since the last update | |
3325 | */ | |
3326 | void account_guest_time(struct task_struct *p, cputime_t cputime) | |
3327 | { | |
3328 | cputime64_t tmp; | |
3329 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3330 | ||
3331 | tmp = cputime_to_cputime64(cputime); | |
3332 | ||
3333 | p->utime = cputime_add(p->utime, cputime); | |
3334 | p->gtime = cputime_add(p->gtime, cputime); | |
3335 | ||
3336 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3337 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3338 | } | |
3339 | ||
c66f08be MN |
3340 | /* |
3341 | * Account scaled user cpu time to a process. | |
3342 | * @p: the process that the cpu time gets accounted to | |
3343 | * @cputime: the cpu time spent in user space since the last update | |
3344 | */ | |
3345 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3346 | { | |
3347 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3348 | } | |
3349 | ||
1da177e4 LT |
3350 | /* |
3351 | * Account system cpu time to a process. | |
3352 | * @p: the process that the cpu time gets accounted to | |
3353 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3354 | * @cputime: the cpu time spent in kernel space since the last update | |
3355 | */ | |
3356 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3357 | cputime_t cputime) | |
3358 | { | |
3359 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3360 | struct rq *rq = this_rq(); |
1da177e4 LT |
3361 | cputime64_t tmp; |
3362 | ||
94886b84 LV |
3363 | if (p->flags & PF_VCPU) { |
3364 | account_guest_time(p, cputime); | |
3365 | p->flags &= ~PF_VCPU; | |
3366 | return; | |
3367 | } | |
3368 | ||
1da177e4 LT |
3369 | p->stime = cputime_add(p->stime, cputime); |
3370 | ||
3371 | /* Add system time to cpustat. */ | |
3372 | tmp = cputime_to_cputime64(cputime); | |
3373 | if (hardirq_count() - hardirq_offset) | |
3374 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3375 | else if (softirq_count()) | |
3376 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
3377 | else if (p != rq->idle) | |
3378 | cpustat->system = cputime64_add(cpustat->system, tmp); | |
3379 | else if (atomic_read(&rq->nr_iowait) > 0) | |
3380 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3381 | else | |
3382 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3383 | /* Account for system time used */ | |
3384 | acct_update_integrals(p); | |
1da177e4 LT |
3385 | } |
3386 | ||
c66f08be MN |
3387 | /* |
3388 | * Account scaled system cpu time to a process. | |
3389 | * @p: the process that the cpu time gets accounted to | |
3390 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3391 | * @cputime: the cpu time spent in kernel space since the last update | |
3392 | */ | |
3393 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3394 | { | |
3395 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3396 | } | |
3397 | ||
1da177e4 LT |
3398 | /* |
3399 | * Account for involuntary wait time. | |
3400 | * @p: the process from which the cpu time has been stolen | |
3401 | * @steal: the cpu time spent in involuntary wait | |
3402 | */ | |
3403 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3404 | { | |
3405 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3406 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3407 | struct rq *rq = this_rq(); |
1da177e4 LT |
3408 | |
3409 | if (p == rq->idle) { | |
3410 | p->stime = cputime_add(p->stime, steal); | |
3411 | if (atomic_read(&rq->nr_iowait) > 0) | |
3412 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3413 | else | |
3414 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3415 | } else | |
3416 | cpustat->steal = cputime64_add(cpustat->steal, tmp); | |
3417 | } | |
3418 | ||
7835b98b CL |
3419 | /* |
3420 | * This function gets called by the timer code, with HZ frequency. | |
3421 | * We call it with interrupts disabled. | |
3422 | * | |
3423 | * It also gets called by the fork code, when changing the parent's | |
3424 | * timeslices. | |
3425 | */ | |
3426 | void scheduler_tick(void) | |
3427 | { | |
7835b98b CL |
3428 | int cpu = smp_processor_id(); |
3429 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3430 | struct task_struct *curr = rq->curr; |
529c7726 | 3431 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3432 | |
3433 | spin_lock(&rq->lock); | |
546fe3c9 | 3434 | __update_rq_clock(rq); |
529c7726 IM |
3435 | /* |
3436 | * Let rq->clock advance by at least TICK_NSEC: | |
3437 | */ | |
3438 | if (unlikely(rq->clock < next_tick)) | |
3439 | rq->clock = next_tick; | |
3440 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3441 | update_cpu_load(rq); |
dd41f596 IM |
3442 | if (curr != rq->idle) /* FIXME: needed? */ |
3443 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3444 | spin_unlock(&rq->lock); |
7835b98b | 3445 | |
e418e1c2 | 3446 | #ifdef CONFIG_SMP |
dd41f596 IM |
3447 | rq->idle_at_tick = idle_cpu(cpu); |
3448 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3449 | #endif |
1da177e4 LT |
3450 | } |
3451 | ||
1da177e4 LT |
3452 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3453 | ||
3454 | void fastcall add_preempt_count(int val) | |
3455 | { | |
3456 | /* | |
3457 | * Underflow? | |
3458 | */ | |
9a11b49a IM |
3459 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3460 | return; | |
1da177e4 LT |
3461 | preempt_count() += val; |
3462 | /* | |
3463 | * Spinlock count overflowing soon? | |
3464 | */ | |
33859f7f MOS |
3465 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3466 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3467 | } |
3468 | EXPORT_SYMBOL(add_preempt_count); | |
3469 | ||
3470 | void fastcall sub_preempt_count(int val) | |
3471 | { | |
3472 | /* | |
3473 | * Underflow? | |
3474 | */ | |
9a11b49a IM |
3475 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3476 | return; | |
1da177e4 LT |
3477 | /* |
3478 | * Is the spinlock portion underflowing? | |
3479 | */ | |
9a11b49a IM |
3480 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3481 | !(preempt_count() & PREEMPT_MASK))) | |
3482 | return; | |
3483 | ||
1da177e4 LT |
3484 | preempt_count() -= val; |
3485 | } | |
3486 | EXPORT_SYMBOL(sub_preempt_count); | |
3487 | ||
3488 | #endif | |
3489 | ||
3490 | /* | |
dd41f596 | 3491 | * Print scheduling while atomic bug: |
1da177e4 | 3492 | */ |
dd41f596 | 3493 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3494 | { |
dd41f596 IM |
3495 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
3496 | prev->comm, preempt_count(), prev->pid); | |
3497 | debug_show_held_locks(prev); | |
3498 | if (irqs_disabled()) | |
3499 | print_irqtrace_events(prev); | |
3500 | dump_stack(); | |
3501 | } | |
1da177e4 | 3502 | |
dd41f596 IM |
3503 | /* |
3504 | * Various schedule()-time debugging checks and statistics: | |
3505 | */ | |
3506 | static inline void schedule_debug(struct task_struct *prev) | |
3507 | { | |
1da177e4 LT |
3508 | /* |
3509 | * Test if we are atomic. Since do_exit() needs to call into | |
3510 | * schedule() atomically, we ignore that path for now. | |
3511 | * Otherwise, whine if we are scheduling when we should not be. | |
3512 | */ | |
dd41f596 IM |
3513 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3514 | __schedule_bug(prev); | |
3515 | ||
1da177e4 LT |
3516 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3517 | ||
2d72376b | 3518 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3519 | #ifdef CONFIG_SCHEDSTATS |
3520 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3521 | schedstat_inc(this_rq(), bkl_count); |
3522 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3523 | } |
3524 | #endif | |
dd41f596 IM |
3525 | } |
3526 | ||
3527 | /* | |
3528 | * Pick up the highest-prio task: | |
3529 | */ | |
3530 | static inline struct task_struct * | |
ff95f3df | 3531 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3532 | { |
5522d5d5 | 3533 | const struct sched_class *class; |
dd41f596 | 3534 | struct task_struct *p; |
1da177e4 LT |
3535 | |
3536 | /* | |
dd41f596 IM |
3537 | * Optimization: we know that if all tasks are in |
3538 | * the fair class we can call that function directly: | |
1da177e4 | 3539 | */ |
dd41f596 | 3540 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3541 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3542 | if (likely(p)) |
3543 | return p; | |
1da177e4 LT |
3544 | } |
3545 | ||
dd41f596 IM |
3546 | class = sched_class_highest; |
3547 | for ( ; ; ) { | |
fb8d4724 | 3548 | p = class->pick_next_task(rq); |
dd41f596 IM |
3549 | if (p) |
3550 | return p; | |
3551 | /* | |
3552 | * Will never be NULL as the idle class always | |
3553 | * returns a non-NULL p: | |
3554 | */ | |
3555 | class = class->next; | |
3556 | } | |
3557 | } | |
1da177e4 | 3558 | |
dd41f596 IM |
3559 | /* |
3560 | * schedule() is the main scheduler function. | |
3561 | */ | |
3562 | asmlinkage void __sched schedule(void) | |
3563 | { | |
3564 | struct task_struct *prev, *next; | |
3565 | long *switch_count; | |
3566 | struct rq *rq; | |
dd41f596 IM |
3567 | int cpu; |
3568 | ||
3569 | need_resched: | |
3570 | preempt_disable(); | |
3571 | cpu = smp_processor_id(); | |
3572 | rq = cpu_rq(cpu); | |
3573 | rcu_qsctr_inc(cpu); | |
3574 | prev = rq->curr; | |
3575 | switch_count = &prev->nivcsw; | |
3576 | ||
3577 | release_kernel_lock(prev); | |
3578 | need_resched_nonpreemptible: | |
3579 | ||
3580 | schedule_debug(prev); | |
1da177e4 | 3581 | |
1e819950 IM |
3582 | /* |
3583 | * Do the rq-clock update outside the rq lock: | |
3584 | */ | |
3585 | local_irq_disable(); | |
c1b3da3e | 3586 | __update_rq_clock(rq); |
1e819950 IM |
3587 | spin_lock(&rq->lock); |
3588 | clear_tsk_need_resched(prev); | |
1da177e4 | 3589 | |
1da177e4 | 3590 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3591 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3592 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3593 | prev->state = TASK_RUNNING; |
dd41f596 | 3594 | } else { |
2e1cb74a | 3595 | deactivate_task(rq, prev, 1); |
1da177e4 | 3596 | } |
dd41f596 | 3597 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3598 | } |
3599 | ||
dd41f596 | 3600 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3601 | idle_balance(cpu, rq); |
1da177e4 | 3602 | |
31ee529c | 3603 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3604 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3605 | |
3606 | sched_info_switch(prev, next); | |
dd41f596 | 3607 | |
1da177e4 | 3608 | if (likely(prev != next)) { |
1da177e4 LT |
3609 | rq->nr_switches++; |
3610 | rq->curr = next; | |
3611 | ++*switch_count; | |
3612 | ||
dd41f596 | 3613 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3614 | } else |
3615 | spin_unlock_irq(&rq->lock); | |
3616 | ||
dd41f596 IM |
3617 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3618 | cpu = smp_processor_id(); | |
3619 | rq = cpu_rq(cpu); | |
1da177e4 | 3620 | goto need_resched_nonpreemptible; |
dd41f596 | 3621 | } |
1da177e4 LT |
3622 | preempt_enable_no_resched(); |
3623 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3624 | goto need_resched; | |
3625 | } | |
1da177e4 LT |
3626 | EXPORT_SYMBOL(schedule); |
3627 | ||
3628 | #ifdef CONFIG_PREEMPT | |
3629 | /* | |
2ed6e34f | 3630 | * this is the entry point to schedule() from in-kernel preemption |
1da177e4 LT |
3631 | * off of preempt_enable. Kernel preemptions off return from interrupt |
3632 | * occur there and call schedule directly. | |
3633 | */ | |
3634 | asmlinkage void __sched preempt_schedule(void) | |
3635 | { | |
3636 | struct thread_info *ti = current_thread_info(); | |
3637 | #ifdef CONFIG_PREEMPT_BKL | |
3638 | struct task_struct *task = current; | |
3639 | int saved_lock_depth; | |
3640 | #endif | |
3641 | /* | |
3642 | * If there is a non-zero preempt_count or interrupts are disabled, | |
3643 | * we do not want to preempt the current task. Just return.. | |
3644 | */ | |
beed33a8 | 3645 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3646 | return; |
3647 | ||
3a5c359a AK |
3648 | do { |
3649 | add_preempt_count(PREEMPT_ACTIVE); | |
3650 | ||
3651 | /* | |
3652 | * We keep the big kernel semaphore locked, but we | |
3653 | * clear ->lock_depth so that schedule() doesnt | |
3654 | * auto-release the semaphore: | |
3655 | */ | |
1da177e4 | 3656 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3657 | saved_lock_depth = task->lock_depth; |
3658 | task->lock_depth = -1; | |
1da177e4 | 3659 | #endif |
3a5c359a | 3660 | schedule(); |
1da177e4 | 3661 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3662 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3663 | #endif |
3a5c359a | 3664 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3665 | |
3a5c359a AK |
3666 | /* |
3667 | * Check again in case we missed a preemption opportunity | |
3668 | * between schedule and now. | |
3669 | */ | |
3670 | barrier(); | |
3671 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3672 | } |
1da177e4 LT |
3673 | EXPORT_SYMBOL(preempt_schedule); |
3674 | ||
3675 | /* | |
2ed6e34f | 3676 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3677 | * off of irq context. |
3678 | * Note, that this is called and return with irqs disabled. This will | |
3679 | * protect us against recursive calling from irq. | |
3680 | */ | |
3681 | asmlinkage void __sched preempt_schedule_irq(void) | |
3682 | { | |
3683 | struct thread_info *ti = current_thread_info(); | |
3684 | #ifdef CONFIG_PREEMPT_BKL | |
3685 | struct task_struct *task = current; | |
3686 | int saved_lock_depth; | |
3687 | #endif | |
2ed6e34f | 3688 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3689 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3690 | ||
3a5c359a AK |
3691 | do { |
3692 | add_preempt_count(PREEMPT_ACTIVE); | |
3693 | ||
3694 | /* | |
3695 | * We keep the big kernel semaphore locked, but we | |
3696 | * clear ->lock_depth so that schedule() doesnt | |
3697 | * auto-release the semaphore: | |
3698 | */ | |
1da177e4 | 3699 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3700 | saved_lock_depth = task->lock_depth; |
3701 | task->lock_depth = -1; | |
1da177e4 | 3702 | #endif |
3a5c359a AK |
3703 | local_irq_enable(); |
3704 | schedule(); | |
3705 | local_irq_disable(); | |
1da177e4 | 3706 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3707 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3708 | #endif |
3a5c359a | 3709 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3710 | |
3a5c359a AK |
3711 | /* |
3712 | * Check again in case we missed a preemption opportunity | |
3713 | * between schedule and now. | |
3714 | */ | |
3715 | barrier(); | |
3716 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3717 | } |
3718 | ||
3719 | #endif /* CONFIG_PREEMPT */ | |
3720 | ||
95cdf3b7 IM |
3721 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3722 | void *key) | |
1da177e4 | 3723 | { |
48f24c4d | 3724 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3725 | } |
1da177e4 LT |
3726 | EXPORT_SYMBOL(default_wake_function); |
3727 | ||
3728 | /* | |
3729 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
3730 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
3731 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
3732 | * | |
3733 | * There are circumstances in which we can try to wake a task which has already | |
3734 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
3735 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
3736 | */ | |
3737 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3738 | int nr_exclusive, int sync, void *key) | |
3739 | { | |
2e45874c | 3740 | wait_queue_t *curr, *next; |
1da177e4 | 3741 | |
2e45874c | 3742 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3743 | unsigned flags = curr->flags; |
3744 | ||
1da177e4 | 3745 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3746 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3747 | break; |
3748 | } | |
3749 | } | |
3750 | ||
3751 | /** | |
3752 | * __wake_up - wake up threads blocked on a waitqueue. | |
3753 | * @q: the waitqueue | |
3754 | * @mode: which threads | |
3755 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3756 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3757 | */ |
3758 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3759 | int nr_exclusive, void *key) |
1da177e4 LT |
3760 | { |
3761 | unsigned long flags; | |
3762 | ||
3763 | spin_lock_irqsave(&q->lock, flags); | |
3764 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3765 | spin_unlock_irqrestore(&q->lock, flags); | |
3766 | } | |
1da177e4 LT |
3767 | EXPORT_SYMBOL(__wake_up); |
3768 | ||
3769 | /* | |
3770 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3771 | */ | |
3772 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3773 | { | |
3774 | __wake_up_common(q, mode, 1, 0, NULL); | |
3775 | } | |
3776 | ||
3777 | /** | |
67be2dd1 | 3778 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3779 | * @q: the waitqueue |
3780 | * @mode: which threads | |
3781 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3782 | * | |
3783 | * The sync wakeup differs that the waker knows that it will schedule | |
3784 | * away soon, so while the target thread will be woken up, it will not | |
3785 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3786 | * with each other. This can prevent needless bouncing between CPUs. | |
3787 | * | |
3788 | * On UP it can prevent extra preemption. | |
3789 | */ | |
95cdf3b7 IM |
3790 | void fastcall |
3791 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3792 | { |
3793 | unsigned long flags; | |
3794 | int sync = 1; | |
3795 | ||
3796 | if (unlikely(!q)) | |
3797 | return; | |
3798 | ||
3799 | if (unlikely(!nr_exclusive)) | |
3800 | sync = 0; | |
3801 | ||
3802 | spin_lock_irqsave(&q->lock, flags); | |
3803 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3804 | spin_unlock_irqrestore(&q->lock, flags); | |
3805 | } | |
3806 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3807 | ||
3808 | void fastcall complete(struct completion *x) | |
3809 | { | |
3810 | unsigned long flags; | |
3811 | ||
3812 | spin_lock_irqsave(&x->wait.lock, flags); | |
3813 | x->done++; | |
3814 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3815 | 1, 0, NULL); | |
3816 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3817 | } | |
3818 | EXPORT_SYMBOL(complete); | |
3819 | ||
3820 | void fastcall complete_all(struct completion *x) | |
3821 | { | |
3822 | unsigned long flags; | |
3823 | ||
3824 | spin_lock_irqsave(&x->wait.lock, flags); | |
3825 | x->done += UINT_MAX/2; | |
3826 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3827 | 0, 0, NULL); | |
3828 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3829 | } | |
3830 | EXPORT_SYMBOL(complete_all); | |
3831 | ||
8cbbe86d AK |
3832 | static inline long __sched |
3833 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3834 | { |
1da177e4 LT |
3835 | if (!x->done) { |
3836 | DECLARE_WAITQUEUE(wait, current); | |
3837 | ||
3838 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3839 | __add_wait_queue_tail(&x->wait, &wait); | |
3840 | do { | |
8cbbe86d AK |
3841 | if (state == TASK_INTERRUPTIBLE && |
3842 | signal_pending(current)) { | |
3843 | __remove_wait_queue(&x->wait, &wait); | |
3844 | return -ERESTARTSYS; | |
3845 | } | |
3846 | __set_current_state(state); | |
1da177e4 LT |
3847 | spin_unlock_irq(&x->wait.lock); |
3848 | timeout = schedule_timeout(timeout); | |
3849 | spin_lock_irq(&x->wait.lock); | |
3850 | if (!timeout) { | |
3851 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3852 | return timeout; |
1da177e4 LT |
3853 | } |
3854 | } while (!x->done); | |
3855 | __remove_wait_queue(&x->wait, &wait); | |
3856 | } | |
3857 | x->done--; | |
1da177e4 LT |
3858 | return timeout; |
3859 | } | |
1da177e4 | 3860 | |
8cbbe86d AK |
3861 | static long __sched |
3862 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3863 | { |
1da177e4 LT |
3864 | might_sleep(); |
3865 | ||
3866 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3867 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3868 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3869 | return timeout; |
3870 | } | |
1da177e4 | 3871 | |
8cbbe86d AK |
3872 | void fastcall __sched wait_for_completion(struct completion *x) |
3873 | { | |
3874 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3875 | } |
8cbbe86d | 3876 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 LT |
3877 | |
3878 | unsigned long fastcall __sched | |
8cbbe86d | 3879 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3880 | { |
8cbbe86d | 3881 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3882 | } |
8cbbe86d | 3883 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3884 | |
8cbbe86d | 3885 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3886 | { |
51e97990 AK |
3887 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3888 | if (t == -ERESTARTSYS) | |
3889 | return t; | |
3890 | return 0; | |
0fec171c | 3891 | } |
8cbbe86d | 3892 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3893 | |
8cbbe86d AK |
3894 | unsigned long fastcall __sched |
3895 | wait_for_completion_interruptible_timeout(struct completion *x, | |
3896 | unsigned long timeout) | |
0fec171c | 3897 | { |
8cbbe86d | 3898 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3899 | } |
8cbbe86d | 3900 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3901 | |
8cbbe86d AK |
3902 | static long __sched |
3903 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3904 | { |
0fec171c IM |
3905 | unsigned long flags; |
3906 | wait_queue_t wait; | |
3907 | ||
3908 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3909 | |
8cbbe86d | 3910 | __set_current_state(state); |
1da177e4 | 3911 | |
8cbbe86d AK |
3912 | spin_lock_irqsave(&q->lock, flags); |
3913 | __add_wait_queue(q, &wait); | |
3914 | spin_unlock(&q->lock); | |
3915 | timeout = schedule_timeout(timeout); | |
3916 | spin_lock_irq(&q->lock); | |
3917 | __remove_wait_queue(q, &wait); | |
3918 | spin_unlock_irqrestore(&q->lock, flags); | |
3919 | ||
3920 | return timeout; | |
3921 | } | |
3922 | ||
3923 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3924 | { | |
3925 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3926 | } |
1da177e4 LT |
3927 | EXPORT_SYMBOL(interruptible_sleep_on); |
3928 | ||
0fec171c | 3929 | long __sched |
95cdf3b7 | 3930 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3931 | { |
8cbbe86d | 3932 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3933 | } |
1da177e4 LT |
3934 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3935 | ||
0fec171c | 3936 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3937 | { |
8cbbe86d | 3938 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3939 | } |
1da177e4 LT |
3940 | EXPORT_SYMBOL(sleep_on); |
3941 | ||
0fec171c | 3942 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3943 | { |
8cbbe86d | 3944 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3945 | } |
1da177e4 LT |
3946 | EXPORT_SYMBOL(sleep_on_timeout); |
3947 | ||
b29739f9 IM |
3948 | #ifdef CONFIG_RT_MUTEXES |
3949 | ||
3950 | /* | |
3951 | * rt_mutex_setprio - set the current priority of a task | |
3952 | * @p: task | |
3953 | * @prio: prio value (kernel-internal form) | |
3954 | * | |
3955 | * This function changes the 'effective' priority of a task. It does | |
3956 | * not touch ->normal_prio like __setscheduler(). | |
3957 | * | |
3958 | * Used by the rt_mutex code to implement priority inheritance logic. | |
3959 | */ | |
36c8b586 | 3960 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
3961 | { |
3962 | unsigned long flags; | |
83b699ed | 3963 | int oldprio, on_rq, running; |
70b97a7f | 3964 | struct rq *rq; |
b29739f9 IM |
3965 | |
3966 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
3967 | ||
3968 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 3969 | update_rq_clock(rq); |
b29739f9 | 3970 | |
d5f9f942 | 3971 | oldprio = p->prio; |
dd41f596 | 3972 | on_rq = p->se.on_rq; |
83b699ed SV |
3973 | running = task_running(rq, p); |
3974 | if (on_rq) { | |
69be72c1 | 3975 | dequeue_task(rq, p, 0); |
83b699ed SV |
3976 | if (running) |
3977 | p->sched_class->put_prev_task(rq, p); | |
3978 | } | |
dd41f596 IM |
3979 | |
3980 | if (rt_prio(prio)) | |
3981 | p->sched_class = &rt_sched_class; | |
3982 | else | |
3983 | p->sched_class = &fair_sched_class; | |
3984 | ||
b29739f9 IM |
3985 | p->prio = prio; |
3986 | ||
dd41f596 | 3987 | if (on_rq) { |
83b699ed SV |
3988 | if (running) |
3989 | p->sched_class->set_curr_task(rq); | |
8159f87e | 3990 | enqueue_task(rq, p, 0); |
b29739f9 IM |
3991 | /* |
3992 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
3993 | * our priority decreased, or if we are not currently running on |
3994 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 3995 | */ |
83b699ed | 3996 | if (running) { |
d5f9f942 AM |
3997 | if (p->prio > oldprio) |
3998 | resched_task(rq->curr); | |
dd41f596 IM |
3999 | } else { |
4000 | check_preempt_curr(rq, p); | |
4001 | } | |
b29739f9 IM |
4002 | } |
4003 | task_rq_unlock(rq, &flags); | |
4004 | } | |
4005 | ||
4006 | #endif | |
4007 | ||
36c8b586 | 4008 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4009 | { |
dd41f596 | 4010 | int old_prio, delta, on_rq; |
1da177e4 | 4011 | unsigned long flags; |
70b97a7f | 4012 | struct rq *rq; |
1da177e4 LT |
4013 | |
4014 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4015 | return; | |
4016 | /* | |
4017 | * We have to be careful, if called from sys_setpriority(), | |
4018 | * the task might be in the middle of scheduling on another CPU. | |
4019 | */ | |
4020 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4021 | update_rq_clock(rq); |
1da177e4 LT |
4022 | /* |
4023 | * The RT priorities are set via sched_setscheduler(), but we still | |
4024 | * allow the 'normal' nice value to be set - but as expected | |
4025 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4026 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4027 | */ |
e05606d3 | 4028 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4029 | p->static_prio = NICE_TO_PRIO(nice); |
4030 | goto out_unlock; | |
4031 | } | |
dd41f596 IM |
4032 | on_rq = p->se.on_rq; |
4033 | if (on_rq) { | |
69be72c1 | 4034 | dequeue_task(rq, p, 0); |
79b5dddf | 4035 | dec_load(rq, p); |
2dd73a4f | 4036 | } |
1da177e4 | 4037 | |
1da177e4 | 4038 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4039 | set_load_weight(p); |
b29739f9 IM |
4040 | old_prio = p->prio; |
4041 | p->prio = effective_prio(p); | |
4042 | delta = p->prio - old_prio; | |
1da177e4 | 4043 | |
dd41f596 | 4044 | if (on_rq) { |
8159f87e | 4045 | enqueue_task(rq, p, 0); |
29b4b623 | 4046 | inc_load(rq, p); |
1da177e4 | 4047 | /* |
d5f9f942 AM |
4048 | * If the task increased its priority or is running and |
4049 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4050 | */ |
d5f9f942 | 4051 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4052 | resched_task(rq->curr); |
4053 | } | |
4054 | out_unlock: | |
4055 | task_rq_unlock(rq, &flags); | |
4056 | } | |
1da177e4 LT |
4057 | EXPORT_SYMBOL(set_user_nice); |
4058 | ||
e43379f1 MM |
4059 | /* |
4060 | * can_nice - check if a task can reduce its nice value | |
4061 | * @p: task | |
4062 | * @nice: nice value | |
4063 | */ | |
36c8b586 | 4064 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4065 | { |
024f4747 MM |
4066 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4067 | int nice_rlim = 20 - nice; | |
48f24c4d | 4068 | |
e43379f1 MM |
4069 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4070 | capable(CAP_SYS_NICE)); | |
4071 | } | |
4072 | ||
1da177e4 LT |
4073 | #ifdef __ARCH_WANT_SYS_NICE |
4074 | ||
4075 | /* | |
4076 | * sys_nice - change the priority of the current process. | |
4077 | * @increment: priority increment | |
4078 | * | |
4079 | * sys_setpriority is a more generic, but much slower function that | |
4080 | * does similar things. | |
4081 | */ | |
4082 | asmlinkage long sys_nice(int increment) | |
4083 | { | |
48f24c4d | 4084 | long nice, retval; |
1da177e4 LT |
4085 | |
4086 | /* | |
4087 | * Setpriority might change our priority at the same moment. | |
4088 | * We don't have to worry. Conceptually one call occurs first | |
4089 | * and we have a single winner. | |
4090 | */ | |
e43379f1 MM |
4091 | if (increment < -40) |
4092 | increment = -40; | |
1da177e4 LT |
4093 | if (increment > 40) |
4094 | increment = 40; | |
4095 | ||
4096 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4097 | if (nice < -20) | |
4098 | nice = -20; | |
4099 | if (nice > 19) | |
4100 | nice = 19; | |
4101 | ||
e43379f1 MM |
4102 | if (increment < 0 && !can_nice(current, nice)) |
4103 | return -EPERM; | |
4104 | ||
1da177e4 LT |
4105 | retval = security_task_setnice(current, nice); |
4106 | if (retval) | |
4107 | return retval; | |
4108 | ||
4109 | set_user_nice(current, nice); | |
4110 | return 0; | |
4111 | } | |
4112 | ||
4113 | #endif | |
4114 | ||
4115 | /** | |
4116 | * task_prio - return the priority value of a given task. | |
4117 | * @p: the task in question. | |
4118 | * | |
4119 | * This is the priority value as seen by users in /proc. | |
4120 | * RT tasks are offset by -200. Normal tasks are centered | |
4121 | * around 0, value goes from -16 to +15. | |
4122 | */ | |
36c8b586 | 4123 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4124 | { |
4125 | return p->prio - MAX_RT_PRIO; | |
4126 | } | |
4127 | ||
4128 | /** | |
4129 | * task_nice - return the nice value of a given task. | |
4130 | * @p: the task in question. | |
4131 | */ | |
36c8b586 | 4132 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4133 | { |
4134 | return TASK_NICE(p); | |
4135 | } | |
1da177e4 | 4136 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4137 | |
4138 | /** | |
4139 | * idle_cpu - is a given cpu idle currently? | |
4140 | * @cpu: the processor in question. | |
4141 | */ | |
4142 | int idle_cpu(int cpu) | |
4143 | { | |
4144 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4145 | } | |
4146 | ||
1da177e4 LT |
4147 | /** |
4148 | * idle_task - return the idle task for a given cpu. | |
4149 | * @cpu: the processor in question. | |
4150 | */ | |
36c8b586 | 4151 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4152 | { |
4153 | return cpu_rq(cpu)->idle; | |
4154 | } | |
4155 | ||
4156 | /** | |
4157 | * find_process_by_pid - find a process with a matching PID value. | |
4158 | * @pid: the pid in question. | |
4159 | */ | |
a9957449 | 4160 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 LT |
4161 | { |
4162 | return pid ? find_task_by_pid(pid) : current; | |
4163 | } | |
4164 | ||
4165 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4166 | static void |
4167 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4168 | { |
dd41f596 | 4169 | BUG_ON(p->se.on_rq); |
48f24c4d | 4170 | |
1da177e4 | 4171 | p->policy = policy; |
dd41f596 IM |
4172 | switch (p->policy) { |
4173 | case SCHED_NORMAL: | |
4174 | case SCHED_BATCH: | |
4175 | case SCHED_IDLE: | |
4176 | p->sched_class = &fair_sched_class; | |
4177 | break; | |
4178 | case SCHED_FIFO: | |
4179 | case SCHED_RR: | |
4180 | p->sched_class = &rt_sched_class; | |
4181 | break; | |
4182 | } | |
4183 | ||
1da177e4 | 4184 | p->rt_priority = prio; |
b29739f9 IM |
4185 | p->normal_prio = normal_prio(p); |
4186 | /* we are holding p->pi_lock already */ | |
4187 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4188 | set_load_weight(p); |
1da177e4 LT |
4189 | } |
4190 | ||
4191 | /** | |
72fd4a35 | 4192 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4193 | * @p: the task in question. |
4194 | * @policy: new policy. | |
4195 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4196 | * |
72fd4a35 | 4197 | * NOTE that the task may be already dead. |
1da177e4 | 4198 | */ |
95cdf3b7 IM |
4199 | int sched_setscheduler(struct task_struct *p, int policy, |
4200 | struct sched_param *param) | |
1da177e4 | 4201 | { |
83b699ed | 4202 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4203 | unsigned long flags; |
70b97a7f | 4204 | struct rq *rq; |
1da177e4 | 4205 | |
66e5393a SR |
4206 | /* may grab non-irq protected spin_locks */ |
4207 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4208 | recheck: |
4209 | /* double check policy once rq lock held */ | |
4210 | if (policy < 0) | |
4211 | policy = oldpolicy = p->policy; | |
4212 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4213 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4214 | policy != SCHED_IDLE) | |
b0a9499c | 4215 | return -EINVAL; |
1da177e4 LT |
4216 | /* |
4217 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4218 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4219 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4220 | */ |
4221 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4222 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4223 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4224 | return -EINVAL; |
e05606d3 | 4225 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4226 | return -EINVAL; |
4227 | ||
37e4ab3f OC |
4228 | /* |
4229 | * Allow unprivileged RT tasks to decrease priority: | |
4230 | */ | |
4231 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4232 | if (rt_policy(policy)) { |
8dc3e909 | 4233 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4234 | |
4235 | if (!lock_task_sighand(p, &flags)) | |
4236 | return -ESRCH; | |
4237 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4238 | unlock_task_sighand(p, &flags); | |
4239 | ||
4240 | /* can't set/change the rt policy */ | |
4241 | if (policy != p->policy && !rlim_rtprio) | |
4242 | return -EPERM; | |
4243 | ||
4244 | /* can't increase priority */ | |
4245 | if (param->sched_priority > p->rt_priority && | |
4246 | param->sched_priority > rlim_rtprio) | |
4247 | return -EPERM; | |
4248 | } | |
dd41f596 IM |
4249 | /* |
4250 | * Like positive nice levels, dont allow tasks to | |
4251 | * move out of SCHED_IDLE either: | |
4252 | */ | |
4253 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4254 | return -EPERM; | |
5fe1d75f | 4255 | |
37e4ab3f OC |
4256 | /* can't change other user's priorities */ |
4257 | if ((current->euid != p->euid) && | |
4258 | (current->euid != p->uid)) | |
4259 | return -EPERM; | |
4260 | } | |
1da177e4 LT |
4261 | |
4262 | retval = security_task_setscheduler(p, policy, param); | |
4263 | if (retval) | |
4264 | return retval; | |
b29739f9 IM |
4265 | /* |
4266 | * make sure no PI-waiters arrive (or leave) while we are | |
4267 | * changing the priority of the task: | |
4268 | */ | |
4269 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4270 | /* |
4271 | * To be able to change p->policy safely, the apropriate | |
4272 | * runqueue lock must be held. | |
4273 | */ | |
b29739f9 | 4274 | rq = __task_rq_lock(p); |
1da177e4 LT |
4275 | /* recheck policy now with rq lock held */ |
4276 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4277 | policy = oldpolicy = -1; | |
b29739f9 IM |
4278 | __task_rq_unlock(rq); |
4279 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4280 | goto recheck; |
4281 | } | |
2daa3577 | 4282 | update_rq_clock(rq); |
dd41f596 | 4283 | on_rq = p->se.on_rq; |
83b699ed SV |
4284 | running = task_running(rq, p); |
4285 | if (on_rq) { | |
2e1cb74a | 4286 | deactivate_task(rq, p, 0); |
83b699ed SV |
4287 | if (running) |
4288 | p->sched_class->put_prev_task(rq, p); | |
4289 | } | |
f6b53205 | 4290 | |
1da177e4 | 4291 | oldprio = p->prio; |
dd41f596 | 4292 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4293 | |
dd41f596 | 4294 | if (on_rq) { |
83b699ed SV |
4295 | if (running) |
4296 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4297 | activate_task(rq, p, 0); |
1da177e4 LT |
4298 | /* |
4299 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4300 | * our priority decreased, or if we are not currently running on |
4301 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4302 | */ |
83b699ed | 4303 | if (running) { |
d5f9f942 AM |
4304 | if (p->prio > oldprio) |
4305 | resched_task(rq->curr); | |
dd41f596 IM |
4306 | } else { |
4307 | check_preempt_curr(rq, p); | |
4308 | } | |
1da177e4 | 4309 | } |
b29739f9 IM |
4310 | __task_rq_unlock(rq); |
4311 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4312 | ||
95e02ca9 TG |
4313 | rt_mutex_adjust_pi(p); |
4314 | ||
1da177e4 LT |
4315 | return 0; |
4316 | } | |
4317 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4318 | ||
95cdf3b7 IM |
4319 | static int |
4320 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4321 | { |
1da177e4 LT |
4322 | struct sched_param lparam; |
4323 | struct task_struct *p; | |
36c8b586 | 4324 | int retval; |
1da177e4 LT |
4325 | |
4326 | if (!param || pid < 0) | |
4327 | return -EINVAL; | |
4328 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4329 | return -EFAULT; | |
5fe1d75f ON |
4330 | |
4331 | rcu_read_lock(); | |
4332 | retval = -ESRCH; | |
1da177e4 | 4333 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4334 | if (p != NULL) |
4335 | retval = sched_setscheduler(p, policy, &lparam); | |
4336 | rcu_read_unlock(); | |
36c8b586 | 4337 | |
1da177e4 LT |
4338 | return retval; |
4339 | } | |
4340 | ||
4341 | /** | |
4342 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4343 | * @pid: the pid in question. | |
4344 | * @policy: new policy. | |
4345 | * @param: structure containing the new RT priority. | |
4346 | */ | |
4347 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, | |
4348 | struct sched_param __user *param) | |
4349 | { | |
c21761f1 JB |
4350 | /* negative values for policy are not valid */ |
4351 | if (policy < 0) | |
4352 | return -EINVAL; | |
4353 | ||
1da177e4 LT |
4354 | return do_sched_setscheduler(pid, policy, param); |
4355 | } | |
4356 | ||
4357 | /** | |
4358 | * sys_sched_setparam - set/change the RT priority of a thread | |
4359 | * @pid: the pid in question. | |
4360 | * @param: structure containing the new RT priority. | |
4361 | */ | |
4362 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4363 | { | |
4364 | return do_sched_setscheduler(pid, -1, param); | |
4365 | } | |
4366 | ||
4367 | /** | |
4368 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4369 | * @pid: the pid in question. | |
4370 | */ | |
4371 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4372 | { | |
36c8b586 | 4373 | struct task_struct *p; |
3a5c359a | 4374 | int retval; |
1da177e4 LT |
4375 | |
4376 | if (pid < 0) | |
3a5c359a | 4377 | return -EINVAL; |
1da177e4 LT |
4378 | |
4379 | retval = -ESRCH; | |
4380 | read_lock(&tasklist_lock); | |
4381 | p = find_process_by_pid(pid); | |
4382 | if (p) { | |
4383 | retval = security_task_getscheduler(p); | |
4384 | if (!retval) | |
4385 | retval = p->policy; | |
4386 | } | |
4387 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4388 | return retval; |
4389 | } | |
4390 | ||
4391 | /** | |
4392 | * sys_sched_getscheduler - get the RT priority of a thread | |
4393 | * @pid: the pid in question. | |
4394 | * @param: structure containing the RT priority. | |
4395 | */ | |
4396 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4397 | { | |
4398 | struct sched_param lp; | |
36c8b586 | 4399 | struct task_struct *p; |
3a5c359a | 4400 | int retval; |
1da177e4 LT |
4401 | |
4402 | if (!param || pid < 0) | |
3a5c359a | 4403 | return -EINVAL; |
1da177e4 LT |
4404 | |
4405 | read_lock(&tasklist_lock); | |
4406 | p = find_process_by_pid(pid); | |
4407 | retval = -ESRCH; | |
4408 | if (!p) | |
4409 | goto out_unlock; | |
4410 | ||
4411 | retval = security_task_getscheduler(p); | |
4412 | if (retval) | |
4413 | goto out_unlock; | |
4414 | ||
4415 | lp.sched_priority = p->rt_priority; | |
4416 | read_unlock(&tasklist_lock); | |
4417 | ||
4418 | /* | |
4419 | * This one might sleep, we cannot do it with a spinlock held ... | |
4420 | */ | |
4421 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4422 | ||
1da177e4 LT |
4423 | return retval; |
4424 | ||
4425 | out_unlock: | |
4426 | read_unlock(&tasklist_lock); | |
4427 | return retval; | |
4428 | } | |
4429 | ||
4430 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4431 | { | |
1da177e4 | 4432 | cpumask_t cpus_allowed; |
36c8b586 IM |
4433 | struct task_struct *p; |
4434 | int retval; | |
1da177e4 | 4435 | |
5be9361c | 4436 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4437 | read_lock(&tasklist_lock); |
4438 | ||
4439 | p = find_process_by_pid(pid); | |
4440 | if (!p) { | |
4441 | read_unlock(&tasklist_lock); | |
5be9361c | 4442 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4443 | return -ESRCH; |
4444 | } | |
4445 | ||
4446 | /* | |
4447 | * It is not safe to call set_cpus_allowed with the | |
4448 | * tasklist_lock held. We will bump the task_struct's | |
4449 | * usage count and then drop tasklist_lock. | |
4450 | */ | |
4451 | get_task_struct(p); | |
4452 | read_unlock(&tasklist_lock); | |
4453 | ||
4454 | retval = -EPERM; | |
4455 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4456 | !capable(CAP_SYS_NICE)) | |
4457 | goto out_unlock; | |
4458 | ||
e7834f8f DQ |
4459 | retval = security_task_setscheduler(p, 0, NULL); |
4460 | if (retval) | |
4461 | goto out_unlock; | |
4462 | ||
1da177e4 LT |
4463 | cpus_allowed = cpuset_cpus_allowed(p); |
4464 | cpus_and(new_mask, new_mask, cpus_allowed); | |
4465 | retval = set_cpus_allowed(p, new_mask); | |
4466 | ||
4467 | out_unlock: | |
4468 | put_task_struct(p); | |
5be9361c | 4469 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4470 | return retval; |
4471 | } | |
4472 | ||
4473 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4474 | cpumask_t *new_mask) | |
4475 | { | |
4476 | if (len < sizeof(cpumask_t)) { | |
4477 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4478 | } else if (len > sizeof(cpumask_t)) { | |
4479 | len = sizeof(cpumask_t); | |
4480 | } | |
4481 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4482 | } | |
4483 | ||
4484 | /** | |
4485 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4486 | * @pid: pid of the process | |
4487 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4488 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4489 | */ | |
4490 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4491 | unsigned long __user *user_mask_ptr) | |
4492 | { | |
4493 | cpumask_t new_mask; | |
4494 | int retval; | |
4495 | ||
4496 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4497 | if (retval) | |
4498 | return retval; | |
4499 | ||
4500 | return sched_setaffinity(pid, new_mask); | |
4501 | } | |
4502 | ||
4503 | /* | |
4504 | * Represents all cpu's present in the system | |
4505 | * In systems capable of hotplug, this map could dynamically grow | |
4506 | * as new cpu's are detected in the system via any platform specific | |
4507 | * method, such as ACPI for e.g. | |
4508 | */ | |
4509 | ||
4cef0c61 | 4510 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4511 | EXPORT_SYMBOL(cpu_present_map); |
4512 | ||
4513 | #ifndef CONFIG_SMP | |
4cef0c61 | 4514 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4515 | EXPORT_SYMBOL(cpu_online_map); |
4516 | ||
4cef0c61 | 4517 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4518 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4519 | #endif |
4520 | ||
4521 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4522 | { | |
36c8b586 | 4523 | struct task_struct *p; |
1da177e4 | 4524 | int retval; |
1da177e4 | 4525 | |
5be9361c | 4526 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4527 | read_lock(&tasklist_lock); |
4528 | ||
4529 | retval = -ESRCH; | |
4530 | p = find_process_by_pid(pid); | |
4531 | if (!p) | |
4532 | goto out_unlock; | |
4533 | ||
e7834f8f DQ |
4534 | retval = security_task_getscheduler(p); |
4535 | if (retval) | |
4536 | goto out_unlock; | |
4537 | ||
2f7016d9 | 4538 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4539 | |
4540 | out_unlock: | |
4541 | read_unlock(&tasklist_lock); | |
5be9361c | 4542 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4543 | |
9531b62f | 4544 | return retval; |
1da177e4 LT |
4545 | } |
4546 | ||
4547 | /** | |
4548 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4549 | * @pid: pid of the process | |
4550 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4551 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4552 | */ | |
4553 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4554 | unsigned long __user *user_mask_ptr) | |
4555 | { | |
4556 | int ret; | |
4557 | cpumask_t mask; | |
4558 | ||
4559 | if (len < sizeof(cpumask_t)) | |
4560 | return -EINVAL; | |
4561 | ||
4562 | ret = sched_getaffinity(pid, &mask); | |
4563 | if (ret < 0) | |
4564 | return ret; | |
4565 | ||
4566 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4567 | return -EFAULT; | |
4568 | ||
4569 | return sizeof(cpumask_t); | |
4570 | } | |
4571 | ||
4572 | /** | |
4573 | * sys_sched_yield - yield the current processor to other threads. | |
4574 | * | |
dd41f596 IM |
4575 | * This function yields the current CPU to other tasks. If there are no |
4576 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4577 | */ |
4578 | asmlinkage long sys_sched_yield(void) | |
4579 | { | |
70b97a7f | 4580 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4581 | |
2d72376b | 4582 | schedstat_inc(rq, yld_count); |
4530d7ab | 4583 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4584 | |
4585 | /* | |
4586 | * Since we are going to call schedule() anyway, there's | |
4587 | * no need to preempt or enable interrupts: | |
4588 | */ | |
4589 | __release(rq->lock); | |
8a25d5de | 4590 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4591 | _raw_spin_unlock(&rq->lock); |
4592 | preempt_enable_no_resched(); | |
4593 | ||
4594 | schedule(); | |
4595 | ||
4596 | return 0; | |
4597 | } | |
4598 | ||
e7b38404 | 4599 | static void __cond_resched(void) |
1da177e4 | 4600 | { |
8e0a43d8 IM |
4601 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4602 | __might_sleep(__FILE__, __LINE__); | |
4603 | #endif | |
5bbcfd90 IM |
4604 | /* |
4605 | * The BKS might be reacquired before we have dropped | |
4606 | * PREEMPT_ACTIVE, which could trigger a second | |
4607 | * cond_resched() call. | |
4608 | */ | |
1da177e4 LT |
4609 | do { |
4610 | add_preempt_count(PREEMPT_ACTIVE); | |
4611 | schedule(); | |
4612 | sub_preempt_count(PREEMPT_ACTIVE); | |
4613 | } while (need_resched()); | |
4614 | } | |
4615 | ||
4616 | int __sched cond_resched(void) | |
4617 | { | |
9414232f IM |
4618 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4619 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4620 | __cond_resched(); |
4621 | return 1; | |
4622 | } | |
4623 | return 0; | |
4624 | } | |
1da177e4 LT |
4625 | EXPORT_SYMBOL(cond_resched); |
4626 | ||
4627 | /* | |
4628 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4629 | * call schedule, and on return reacquire the lock. | |
4630 | * | |
4631 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level | |
4632 | * operations here to prevent schedule() from being called twice (once via | |
4633 | * spin_unlock(), once by hand). | |
4634 | */ | |
95cdf3b7 | 4635 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4636 | { |
6df3cecb JK |
4637 | int ret = 0; |
4638 | ||
1da177e4 LT |
4639 | if (need_lockbreak(lock)) { |
4640 | spin_unlock(lock); | |
4641 | cpu_relax(); | |
6df3cecb | 4642 | ret = 1; |
1da177e4 LT |
4643 | spin_lock(lock); |
4644 | } | |
9414232f | 4645 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4646 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4647 | _raw_spin_unlock(lock); |
4648 | preempt_enable_no_resched(); | |
4649 | __cond_resched(); | |
6df3cecb | 4650 | ret = 1; |
1da177e4 | 4651 | spin_lock(lock); |
1da177e4 | 4652 | } |
6df3cecb | 4653 | return ret; |
1da177e4 | 4654 | } |
1da177e4 LT |
4655 | EXPORT_SYMBOL(cond_resched_lock); |
4656 | ||
4657 | int __sched cond_resched_softirq(void) | |
4658 | { | |
4659 | BUG_ON(!in_softirq()); | |
4660 | ||
9414232f | 4661 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4662 | local_bh_enable(); |
1da177e4 LT |
4663 | __cond_resched(); |
4664 | local_bh_disable(); | |
4665 | return 1; | |
4666 | } | |
4667 | return 0; | |
4668 | } | |
1da177e4 LT |
4669 | EXPORT_SYMBOL(cond_resched_softirq); |
4670 | ||
1da177e4 LT |
4671 | /** |
4672 | * yield - yield the current processor to other threads. | |
4673 | * | |
72fd4a35 | 4674 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4675 | * thread runnable and calls sys_sched_yield(). |
4676 | */ | |
4677 | void __sched yield(void) | |
4678 | { | |
4679 | set_current_state(TASK_RUNNING); | |
4680 | sys_sched_yield(); | |
4681 | } | |
1da177e4 LT |
4682 | EXPORT_SYMBOL(yield); |
4683 | ||
4684 | /* | |
4685 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so | |
4686 | * that process accounting knows that this is a task in IO wait state. | |
4687 | * | |
4688 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4689 | * has set its backing_dev_info: the queue against which it should throttle) | |
4690 | */ | |
4691 | void __sched io_schedule(void) | |
4692 | { | |
70b97a7f | 4693 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4694 | |
0ff92245 | 4695 | delayacct_blkio_start(); |
1da177e4 LT |
4696 | atomic_inc(&rq->nr_iowait); |
4697 | schedule(); | |
4698 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4699 | delayacct_blkio_end(); |
1da177e4 | 4700 | } |
1da177e4 LT |
4701 | EXPORT_SYMBOL(io_schedule); |
4702 | ||
4703 | long __sched io_schedule_timeout(long timeout) | |
4704 | { | |
70b97a7f | 4705 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4706 | long ret; |
4707 | ||
0ff92245 | 4708 | delayacct_blkio_start(); |
1da177e4 LT |
4709 | atomic_inc(&rq->nr_iowait); |
4710 | ret = schedule_timeout(timeout); | |
4711 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4712 | delayacct_blkio_end(); |
1da177e4 LT |
4713 | return ret; |
4714 | } | |
4715 | ||
4716 | /** | |
4717 | * sys_sched_get_priority_max - return maximum RT priority. | |
4718 | * @policy: scheduling class. | |
4719 | * | |
4720 | * this syscall returns the maximum rt_priority that can be used | |
4721 | * by a given scheduling class. | |
4722 | */ | |
4723 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4724 | { | |
4725 | int ret = -EINVAL; | |
4726 | ||
4727 | switch (policy) { | |
4728 | case SCHED_FIFO: | |
4729 | case SCHED_RR: | |
4730 | ret = MAX_USER_RT_PRIO-1; | |
4731 | break; | |
4732 | case SCHED_NORMAL: | |
b0a9499c | 4733 | case SCHED_BATCH: |
dd41f596 | 4734 | case SCHED_IDLE: |
1da177e4 LT |
4735 | ret = 0; |
4736 | break; | |
4737 | } | |
4738 | return ret; | |
4739 | } | |
4740 | ||
4741 | /** | |
4742 | * sys_sched_get_priority_min - return minimum RT priority. | |
4743 | * @policy: scheduling class. | |
4744 | * | |
4745 | * this syscall returns the minimum rt_priority that can be used | |
4746 | * by a given scheduling class. | |
4747 | */ | |
4748 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4749 | { | |
4750 | int ret = -EINVAL; | |
4751 | ||
4752 | switch (policy) { | |
4753 | case SCHED_FIFO: | |
4754 | case SCHED_RR: | |
4755 | ret = 1; | |
4756 | break; | |
4757 | case SCHED_NORMAL: | |
b0a9499c | 4758 | case SCHED_BATCH: |
dd41f596 | 4759 | case SCHED_IDLE: |
1da177e4 LT |
4760 | ret = 0; |
4761 | } | |
4762 | return ret; | |
4763 | } | |
4764 | ||
4765 | /** | |
4766 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4767 | * @pid: pid of the process. | |
4768 | * @interval: userspace pointer to the timeslice value. | |
4769 | * | |
4770 | * this syscall writes the default timeslice value of a given process | |
4771 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4772 | */ | |
4773 | asmlinkage | |
4774 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4775 | { | |
36c8b586 | 4776 | struct task_struct *p; |
a4ec24b4 | 4777 | unsigned int time_slice; |
3a5c359a | 4778 | int retval; |
1da177e4 | 4779 | struct timespec t; |
1da177e4 LT |
4780 | |
4781 | if (pid < 0) | |
3a5c359a | 4782 | return -EINVAL; |
1da177e4 LT |
4783 | |
4784 | retval = -ESRCH; | |
4785 | read_lock(&tasklist_lock); | |
4786 | p = find_process_by_pid(pid); | |
4787 | if (!p) | |
4788 | goto out_unlock; | |
4789 | ||
4790 | retval = security_task_getscheduler(p); | |
4791 | if (retval) | |
4792 | goto out_unlock; | |
4793 | ||
a4ec24b4 DA |
4794 | if (p->policy == SCHED_FIFO) |
4795 | time_slice = 0; | |
4796 | else if (p->policy == SCHED_RR) | |
4797 | time_slice = DEF_TIMESLICE; | |
4798 | else { | |
4799 | struct sched_entity *se = &p->se; | |
4800 | unsigned long flags; | |
4801 | struct rq *rq; | |
4802 | ||
4803 | rq = task_rq_lock(p, &flags); | |
4804 | time_slice = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se)); | |
4805 | task_rq_unlock(rq, &flags); | |
4806 | } | |
1da177e4 | 4807 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4808 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4809 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4810 | return retval; |
3a5c359a | 4811 | |
1da177e4 LT |
4812 | out_unlock: |
4813 | read_unlock(&tasklist_lock); | |
4814 | return retval; | |
4815 | } | |
4816 | ||
2ed6e34f | 4817 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4818 | |
4819 | static void show_task(struct task_struct *p) | |
1da177e4 | 4820 | { |
1da177e4 | 4821 | unsigned long free = 0; |
36c8b586 | 4822 | unsigned state; |
1da177e4 | 4823 | |
1da177e4 | 4824 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4825 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4826 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4827 | #if BITS_PER_LONG == 32 |
1da177e4 | 4828 | if (state == TASK_RUNNING) |
cc4ea795 | 4829 | printk(KERN_CONT " running "); |
1da177e4 | 4830 | else |
cc4ea795 | 4831 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4832 | #else |
4833 | if (state == TASK_RUNNING) | |
cc4ea795 | 4834 | printk(KERN_CONT " running task "); |
1da177e4 | 4835 | else |
cc4ea795 | 4836 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4837 | #endif |
4838 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4839 | { | |
10ebffde | 4840 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4841 | while (!*n) |
4842 | n++; | |
10ebffde | 4843 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4844 | } |
4845 | #endif | |
cc4ea795 | 4846 | printk(KERN_CONT "%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
1da177e4 LT |
4847 | |
4848 | if (state != TASK_RUNNING) | |
4849 | show_stack(p, NULL); | |
4850 | } | |
4851 | ||
e59e2ae2 | 4852 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4853 | { |
36c8b586 | 4854 | struct task_struct *g, *p; |
1da177e4 | 4855 | |
4bd77321 IM |
4856 | #if BITS_PER_LONG == 32 |
4857 | printk(KERN_INFO | |
4858 | " task PC stack pid father\n"); | |
1da177e4 | 4859 | #else |
4bd77321 IM |
4860 | printk(KERN_INFO |
4861 | " task PC stack pid father\n"); | |
1da177e4 LT |
4862 | #endif |
4863 | read_lock(&tasklist_lock); | |
4864 | do_each_thread(g, p) { | |
4865 | /* | |
4866 | * reset the NMI-timeout, listing all files on a slow | |
4867 | * console might take alot of time: | |
4868 | */ | |
4869 | touch_nmi_watchdog(); | |
39bc89fd | 4870 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4871 | show_task(p); |
1da177e4 LT |
4872 | } while_each_thread(g, p); |
4873 | ||
04c9167f JF |
4874 | touch_all_softlockup_watchdogs(); |
4875 | ||
dd41f596 IM |
4876 | #ifdef CONFIG_SCHED_DEBUG |
4877 | sysrq_sched_debug_show(); | |
4878 | #endif | |
1da177e4 | 4879 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4880 | /* |
4881 | * Only show locks if all tasks are dumped: | |
4882 | */ | |
4883 | if (state_filter == -1) | |
4884 | debug_show_all_locks(); | |
1da177e4 LT |
4885 | } |
4886 | ||
1df21055 IM |
4887 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4888 | { | |
dd41f596 | 4889 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4890 | } |
4891 | ||
f340c0d1 IM |
4892 | /** |
4893 | * init_idle - set up an idle thread for a given CPU | |
4894 | * @idle: task in question | |
4895 | * @cpu: cpu the idle task belongs to | |
4896 | * | |
4897 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4898 | * flag, to make booting more robust. | |
4899 | */ | |
5c1e1767 | 4900 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4901 | { |
70b97a7f | 4902 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4903 | unsigned long flags; |
4904 | ||
dd41f596 IM |
4905 | __sched_fork(idle); |
4906 | idle->se.exec_start = sched_clock(); | |
4907 | ||
b29739f9 | 4908 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4909 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4910 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4911 | |
4912 | spin_lock_irqsave(&rq->lock, flags); | |
4913 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4914 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4915 | idle->oncpu = 1; | |
4916 | #endif | |
1da177e4 LT |
4917 | spin_unlock_irqrestore(&rq->lock, flags); |
4918 | ||
4919 | /* Set the preempt count _outside_ the spinlocks! */ | |
4920 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4921 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4922 | #else |
a1261f54 | 4923 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4924 | #endif |
dd41f596 IM |
4925 | /* |
4926 | * The idle tasks have their own, simple scheduling class: | |
4927 | */ | |
4928 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4929 | } |
4930 | ||
4931 | /* | |
4932 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4933 | * indicates which cpus entered this state. This is used | |
4934 | * in the rcu update to wait only for active cpus. For system | |
4935 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4936 | * always be CPU_MASK_NONE. | |
4937 | */ | |
4938 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4939 | ||
4940 | #ifdef CONFIG_SMP | |
4941 | /* | |
4942 | * This is how migration works: | |
4943 | * | |
70b97a7f | 4944 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
4945 | * runqueue and wake up that CPU's migration thread. |
4946 | * 2) we down() the locked semaphore => thread blocks. | |
4947 | * 3) migration thread wakes up (implicitly it forces the migrated | |
4948 | * thread off the CPU) | |
4949 | * 4) it gets the migration request and checks whether the migrated | |
4950 | * task is still in the wrong runqueue. | |
4951 | * 5) if it's in the wrong runqueue then the migration thread removes | |
4952 | * it and puts it into the right queue. | |
4953 | * 6) migration thread up()s the semaphore. | |
4954 | * 7) we wake up and the migration is done. | |
4955 | */ | |
4956 | ||
4957 | /* | |
4958 | * Change a given task's CPU affinity. Migrate the thread to a | |
4959 | * proper CPU and schedule it away if the CPU it's executing on | |
4960 | * is removed from the allowed bitmask. | |
4961 | * | |
4962 | * NOTE: the caller must have a valid reference to the task, the | |
4963 | * task must not exit() & deallocate itself prematurely. The | |
4964 | * call is not atomic; no spinlocks may be held. | |
4965 | */ | |
36c8b586 | 4966 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 4967 | { |
70b97a7f | 4968 | struct migration_req req; |
1da177e4 | 4969 | unsigned long flags; |
70b97a7f | 4970 | struct rq *rq; |
48f24c4d | 4971 | int ret = 0; |
1da177e4 LT |
4972 | |
4973 | rq = task_rq_lock(p, &flags); | |
4974 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
4975 | ret = -EINVAL; | |
4976 | goto out; | |
4977 | } | |
4978 | ||
4979 | p->cpus_allowed = new_mask; | |
4980 | /* Can the task run on the task's current CPU? If so, we're done */ | |
4981 | if (cpu_isset(task_cpu(p), new_mask)) | |
4982 | goto out; | |
4983 | ||
4984 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
4985 | /* Need help from migration thread: drop lock and wait. */ | |
4986 | task_rq_unlock(rq, &flags); | |
4987 | wake_up_process(rq->migration_thread); | |
4988 | wait_for_completion(&req.done); | |
4989 | tlb_migrate_finish(p->mm); | |
4990 | return 0; | |
4991 | } | |
4992 | out: | |
4993 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4994 | |
1da177e4 LT |
4995 | return ret; |
4996 | } | |
1da177e4 LT |
4997 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
4998 | ||
4999 | /* | |
5000 | * Move (not current) task off this cpu, onto dest cpu. We're doing | |
5001 | * this because either it can't run here any more (set_cpus_allowed() | |
5002 | * away from this CPU, or CPU going down), or because we're | |
5003 | * attempting to rebalance this task on exec (sched_exec). | |
5004 | * | |
5005 | * So we race with normal scheduler movements, but that's OK, as long | |
5006 | * as the task is no longer on this CPU. | |
efc30814 KK |
5007 | * |
5008 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5009 | */ |
efc30814 | 5010 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5011 | { |
70b97a7f | 5012 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5013 | int ret = 0, on_rq; |
1da177e4 LT |
5014 | |
5015 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5016 | return ret; |
1da177e4 LT |
5017 | |
5018 | rq_src = cpu_rq(src_cpu); | |
5019 | rq_dest = cpu_rq(dest_cpu); | |
5020 | ||
5021 | double_rq_lock(rq_src, rq_dest); | |
5022 | /* Already moved. */ | |
5023 | if (task_cpu(p) != src_cpu) | |
5024 | goto out; | |
5025 | /* Affinity changed (again). */ | |
5026 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5027 | goto out; | |
5028 | ||
dd41f596 | 5029 | on_rq = p->se.on_rq; |
6e82a3be | 5030 | if (on_rq) |
2e1cb74a | 5031 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5032 | |
1da177e4 | 5033 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5034 | if (on_rq) { |
5035 | activate_task(rq_dest, p, 0); | |
5036 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5037 | } |
efc30814 | 5038 | ret = 1; |
1da177e4 LT |
5039 | out: |
5040 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5041 | return ret; |
1da177e4 LT |
5042 | } |
5043 | ||
5044 | /* | |
5045 | * migration_thread - this is a highprio system thread that performs | |
5046 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5047 | * another runqueue. | |
5048 | */ | |
95cdf3b7 | 5049 | static int migration_thread(void *data) |
1da177e4 | 5050 | { |
1da177e4 | 5051 | int cpu = (long)data; |
70b97a7f | 5052 | struct rq *rq; |
1da177e4 LT |
5053 | |
5054 | rq = cpu_rq(cpu); | |
5055 | BUG_ON(rq->migration_thread != current); | |
5056 | ||
5057 | set_current_state(TASK_INTERRUPTIBLE); | |
5058 | while (!kthread_should_stop()) { | |
70b97a7f | 5059 | struct migration_req *req; |
1da177e4 | 5060 | struct list_head *head; |
1da177e4 | 5061 | |
1da177e4 LT |
5062 | spin_lock_irq(&rq->lock); |
5063 | ||
5064 | if (cpu_is_offline(cpu)) { | |
5065 | spin_unlock_irq(&rq->lock); | |
5066 | goto wait_to_die; | |
5067 | } | |
5068 | ||
5069 | if (rq->active_balance) { | |
5070 | active_load_balance(rq, cpu); | |
5071 | rq->active_balance = 0; | |
5072 | } | |
5073 | ||
5074 | head = &rq->migration_queue; | |
5075 | ||
5076 | if (list_empty(head)) { | |
5077 | spin_unlock_irq(&rq->lock); | |
5078 | schedule(); | |
5079 | set_current_state(TASK_INTERRUPTIBLE); | |
5080 | continue; | |
5081 | } | |
70b97a7f | 5082 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5083 | list_del_init(head->next); |
5084 | ||
674311d5 NP |
5085 | spin_unlock(&rq->lock); |
5086 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5087 | local_irq_enable(); | |
1da177e4 LT |
5088 | |
5089 | complete(&req->done); | |
5090 | } | |
5091 | __set_current_state(TASK_RUNNING); | |
5092 | return 0; | |
5093 | ||
5094 | wait_to_die: | |
5095 | /* Wait for kthread_stop */ | |
5096 | set_current_state(TASK_INTERRUPTIBLE); | |
5097 | while (!kthread_should_stop()) { | |
5098 | schedule(); | |
5099 | set_current_state(TASK_INTERRUPTIBLE); | |
5100 | } | |
5101 | __set_current_state(TASK_RUNNING); | |
5102 | return 0; | |
5103 | } | |
5104 | ||
5105 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5106 | |
5107 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5108 | { | |
5109 | int ret; | |
5110 | ||
5111 | local_irq_disable(); | |
5112 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5113 | local_irq_enable(); | |
5114 | return ret; | |
5115 | } | |
5116 | ||
054b9108 KK |
5117 | /* |
5118 | * Figure out where task on dead CPU should go, use force if neccessary. | |
5119 | * NOTE: interrupts should be disabled by the caller | |
5120 | */ | |
48f24c4d | 5121 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5122 | { |
efc30814 | 5123 | unsigned long flags; |
1da177e4 | 5124 | cpumask_t mask; |
70b97a7f IM |
5125 | struct rq *rq; |
5126 | int dest_cpu; | |
1da177e4 | 5127 | |
3a5c359a AK |
5128 | do { |
5129 | /* On same node? */ | |
5130 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5131 | cpus_and(mask, mask, p->cpus_allowed); | |
5132 | dest_cpu = any_online_cpu(mask); | |
5133 | ||
5134 | /* On any allowed CPU? */ | |
5135 | if (dest_cpu == NR_CPUS) | |
5136 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5137 | ||
5138 | /* No more Mr. Nice Guy. */ | |
5139 | if (dest_cpu == NR_CPUS) { | |
5140 | rq = task_rq_lock(p, &flags); | |
5141 | cpus_setall(p->cpus_allowed); | |
5142 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5143 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5144 | |
3a5c359a AK |
5145 | /* |
5146 | * Don't tell them about moving exiting tasks or | |
5147 | * kernel threads (both mm NULL), since they never | |
5148 | * leave kernel. | |
5149 | */ | |
5150 | if (p->mm && printk_ratelimit()) | |
5151 | printk(KERN_INFO "process %d (%s) no " | |
5152 | "longer affine to cpu%d\n", | |
5153 | p->pid, p->comm, dead_cpu); | |
5154 | } | |
f7b4cddc | 5155 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5156 | } |
5157 | ||
5158 | /* | |
5159 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5160 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5161 | * for performance reasons the counter is not stricly tracking tasks to | |
5162 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5163 | * to keep the global sum constant after CPU-down: | |
5164 | */ | |
70b97a7f | 5165 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5166 | { |
70b97a7f | 5167 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5168 | unsigned long flags; |
5169 | ||
5170 | local_irq_save(flags); | |
5171 | double_rq_lock(rq_src, rq_dest); | |
5172 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5173 | rq_src->nr_uninterruptible = 0; | |
5174 | double_rq_unlock(rq_src, rq_dest); | |
5175 | local_irq_restore(flags); | |
5176 | } | |
5177 | ||
5178 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5179 | static void migrate_live_tasks(int src_cpu) | |
5180 | { | |
48f24c4d | 5181 | struct task_struct *p, *t; |
1da177e4 | 5182 | |
f7b4cddc | 5183 | read_lock(&tasklist_lock); |
1da177e4 | 5184 | |
48f24c4d IM |
5185 | do_each_thread(t, p) { |
5186 | if (p == current) | |
1da177e4 LT |
5187 | continue; |
5188 | ||
48f24c4d IM |
5189 | if (task_cpu(p) == src_cpu) |
5190 | move_task_off_dead_cpu(src_cpu, p); | |
5191 | } while_each_thread(t, p); | |
1da177e4 | 5192 | |
f7b4cddc | 5193 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5194 | } |
5195 | ||
a9957449 AD |
5196 | /* |
5197 | * activate_idle_task - move idle task to the _front_ of runqueue. | |
5198 | */ | |
5199 | static void activate_idle_task(struct task_struct *p, struct rq *rq) | |
5200 | { | |
5201 | update_rq_clock(rq); | |
5202 | ||
5203 | if (p->state == TASK_UNINTERRUPTIBLE) | |
5204 | rq->nr_uninterruptible--; | |
5205 | ||
5206 | enqueue_task(rq, p, 0); | |
5207 | inc_nr_running(p, rq); | |
5208 | } | |
5209 | ||
dd41f596 IM |
5210 | /* |
5211 | * Schedules idle task to be the next runnable task on current CPU. | |
1da177e4 | 5212 | * It does so by boosting its priority to highest possible and adding it to |
48f24c4d | 5213 | * the _front_ of the runqueue. Used by CPU offline code. |
1da177e4 LT |
5214 | */ |
5215 | void sched_idle_next(void) | |
5216 | { | |
48f24c4d | 5217 | int this_cpu = smp_processor_id(); |
70b97a7f | 5218 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5219 | struct task_struct *p = rq->idle; |
5220 | unsigned long flags; | |
5221 | ||
5222 | /* cpu has to be offline */ | |
48f24c4d | 5223 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5224 | |
48f24c4d IM |
5225 | /* |
5226 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5227 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5228 | */ |
5229 | spin_lock_irqsave(&rq->lock, flags); | |
5230 | ||
dd41f596 | 5231 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d IM |
5232 | |
5233 | /* Add idle task to the _front_ of its priority queue: */ | |
dd41f596 | 5234 | activate_idle_task(p, rq); |
1da177e4 LT |
5235 | |
5236 | spin_unlock_irqrestore(&rq->lock, flags); | |
5237 | } | |
5238 | ||
48f24c4d IM |
5239 | /* |
5240 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5241 | * offline. |
5242 | */ | |
5243 | void idle_task_exit(void) | |
5244 | { | |
5245 | struct mm_struct *mm = current->active_mm; | |
5246 | ||
5247 | BUG_ON(cpu_online(smp_processor_id())); | |
5248 | ||
5249 | if (mm != &init_mm) | |
5250 | switch_mm(mm, &init_mm, current); | |
5251 | mmdrop(mm); | |
5252 | } | |
5253 | ||
054b9108 | 5254 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5255 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5256 | { |
70b97a7f | 5257 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5258 | |
5259 | /* Must be exiting, otherwise would be on tasklist. */ | |
48f24c4d | 5260 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
1da177e4 LT |
5261 | |
5262 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5263 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5264 | |
48f24c4d | 5265 | get_task_struct(p); |
1da177e4 LT |
5266 | |
5267 | /* | |
5268 | * Drop lock around migration; if someone else moves it, | |
5269 | * that's OK. No task can be added to this CPU, so iteration is | |
5270 | * fine. | |
5271 | */ | |
f7b4cddc | 5272 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5273 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5274 | spin_lock_irq(&rq->lock); |
1da177e4 | 5275 | |
48f24c4d | 5276 | put_task_struct(p); |
1da177e4 LT |
5277 | } |
5278 | ||
5279 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5280 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5281 | { | |
70b97a7f | 5282 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5283 | struct task_struct *next; |
48f24c4d | 5284 | |
dd41f596 IM |
5285 | for ( ; ; ) { |
5286 | if (!rq->nr_running) | |
5287 | break; | |
a8e504d2 | 5288 | update_rq_clock(rq); |
ff95f3df | 5289 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5290 | if (!next) |
5291 | break; | |
5292 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5293 | |
1da177e4 LT |
5294 | } |
5295 | } | |
5296 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5297 | ||
e692ab53 NP |
5298 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5299 | ||
5300 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5301 | { |
5302 | .procname = "sched_domain", | |
c57baf1e | 5303 | .mode = 0555, |
e0361851 | 5304 | }, |
e692ab53 NP |
5305 | {0,}, |
5306 | }; | |
5307 | ||
5308 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5309 | { |
c57baf1e | 5310 | .ctl_name = CTL_KERN, |
e0361851 | 5311 | .procname = "kernel", |
c57baf1e | 5312 | .mode = 0555, |
e0361851 AD |
5313 | .child = sd_ctl_dir, |
5314 | }, | |
e692ab53 NP |
5315 | {0,}, |
5316 | }; | |
5317 | ||
5318 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5319 | { | |
5320 | struct ctl_table *entry = | |
5cf9f062 | 5321 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5322 | |
e692ab53 NP |
5323 | return entry; |
5324 | } | |
5325 | ||
6382bc90 MM |
5326 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5327 | { | |
cd790076 | 5328 | struct ctl_table *entry; |
6382bc90 | 5329 | |
cd790076 MM |
5330 | /* |
5331 | * In the intermediate directories, both the child directory and | |
5332 | * procname are dynamically allocated and could fail but the mode | |
5333 | * will always be set. In the lowest directory the names are | |
5334 | * static strings and all have proc handlers. | |
5335 | */ | |
5336 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5337 | if (entry->child) |
5338 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5339 | if (entry->proc_handler == NULL) |
5340 | kfree(entry->procname); | |
5341 | } | |
6382bc90 MM |
5342 | |
5343 | kfree(*tablep); | |
5344 | *tablep = NULL; | |
5345 | } | |
5346 | ||
e692ab53 | 5347 | static void |
e0361851 | 5348 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5349 | const char *procname, void *data, int maxlen, |
5350 | mode_t mode, proc_handler *proc_handler) | |
5351 | { | |
e692ab53 NP |
5352 | entry->procname = procname; |
5353 | entry->data = data; | |
5354 | entry->maxlen = maxlen; | |
5355 | entry->mode = mode; | |
5356 | entry->proc_handler = proc_handler; | |
5357 | } | |
5358 | ||
5359 | static struct ctl_table * | |
5360 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5361 | { | |
ace8b3d6 | 5362 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5363 | |
ad1cdc1d MM |
5364 | if (table == NULL) |
5365 | return NULL; | |
5366 | ||
e0361851 | 5367 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5368 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5369 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5370 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5371 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5372 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5373 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5374 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5375 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5376 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5377 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5378 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5379 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5380 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5381 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5382 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5383 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5384 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5385 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5386 | &sd->cache_nice_tries, |
5387 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5388 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5389 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5390 | /* &table[11] is terminator */ |
e692ab53 NP |
5391 | |
5392 | return table; | |
5393 | } | |
5394 | ||
8401f775 | 5395 | static ctl_table * sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5396 | { |
5397 | struct ctl_table *entry, *table; | |
5398 | struct sched_domain *sd; | |
5399 | int domain_num = 0, i; | |
5400 | char buf[32]; | |
5401 | ||
5402 | for_each_domain(cpu, sd) | |
5403 | domain_num++; | |
5404 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5405 | if (table == NULL) |
5406 | return NULL; | |
e692ab53 NP |
5407 | |
5408 | i = 0; | |
5409 | for_each_domain(cpu, sd) { | |
5410 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5411 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5412 | entry->mode = 0555; |
e692ab53 NP |
5413 | entry->child = sd_alloc_ctl_domain_table(sd); |
5414 | entry++; | |
5415 | i++; | |
5416 | } | |
5417 | return table; | |
5418 | } | |
5419 | ||
5420 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5421 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5422 | { |
5423 | int i, cpu_num = num_online_cpus(); | |
5424 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5425 | char buf[32]; | |
5426 | ||
ad1cdc1d MM |
5427 | if (entry == NULL) |
5428 | return; | |
5429 | ||
e692ab53 NP |
5430 | sd_ctl_dir[0].child = entry; |
5431 | ||
97b6ea7b | 5432 | for_each_online_cpu(i) { |
e692ab53 | 5433 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5434 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5435 | entry->mode = 0555; |
e692ab53 | 5436 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5437 | entry++; |
e692ab53 NP |
5438 | } |
5439 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); | |
5440 | } | |
6382bc90 MM |
5441 | |
5442 | static void unregister_sched_domain_sysctl(void) | |
5443 | { | |
5444 | unregister_sysctl_table(sd_sysctl_header); | |
5445 | sd_sysctl_header = NULL; | |
5446 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
5447 | } | |
e692ab53 | 5448 | #else |
6382bc90 MM |
5449 | static void register_sched_domain_sysctl(void) |
5450 | { | |
5451 | } | |
5452 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5453 | { |
5454 | } | |
5455 | #endif | |
5456 | ||
1da177e4 LT |
5457 | /* |
5458 | * migration_call - callback that gets triggered when a CPU is added. | |
5459 | * Here we can start up the necessary migration thread for the new CPU. | |
5460 | */ | |
48f24c4d IM |
5461 | static int __cpuinit |
5462 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5463 | { |
1da177e4 | 5464 | struct task_struct *p; |
48f24c4d | 5465 | int cpu = (long)hcpu; |
1da177e4 | 5466 | unsigned long flags; |
70b97a7f | 5467 | struct rq *rq; |
1da177e4 LT |
5468 | |
5469 | switch (action) { | |
5be9361c GS |
5470 | case CPU_LOCK_ACQUIRE: |
5471 | mutex_lock(&sched_hotcpu_mutex); | |
5472 | break; | |
5473 | ||
1da177e4 | 5474 | case CPU_UP_PREPARE: |
8bb78442 | 5475 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5476 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5477 | if (IS_ERR(p)) |
5478 | return NOTIFY_BAD; | |
1da177e4 LT |
5479 | kthread_bind(p, cpu); |
5480 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5481 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5482 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5483 | task_rq_unlock(rq, &flags); |
5484 | cpu_rq(cpu)->migration_thread = p; | |
5485 | break; | |
48f24c4d | 5486 | |
1da177e4 | 5487 | case CPU_ONLINE: |
8bb78442 | 5488 | case CPU_ONLINE_FROZEN: |
1da177e4 LT |
5489 | /* Strictly unneccessary, as first user will wake it. */ |
5490 | wake_up_process(cpu_rq(cpu)->migration_thread); | |
5491 | break; | |
48f24c4d | 5492 | |
1da177e4 LT |
5493 | #ifdef CONFIG_HOTPLUG_CPU |
5494 | case CPU_UP_CANCELED: | |
8bb78442 | 5495 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5496 | if (!cpu_rq(cpu)->migration_thread) |
5497 | break; | |
1da177e4 | 5498 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5499 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5500 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5501 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5502 | cpu_rq(cpu)->migration_thread = NULL; | |
5503 | break; | |
48f24c4d | 5504 | |
1da177e4 | 5505 | case CPU_DEAD: |
8bb78442 | 5506 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5507 | migrate_live_tasks(cpu); |
5508 | rq = cpu_rq(cpu); | |
5509 | kthread_stop(rq->migration_thread); | |
5510 | rq->migration_thread = NULL; | |
5511 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5512 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5513 | update_rq_clock(rq); |
2e1cb74a | 5514 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5515 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5516 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5517 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5518 | migrate_dead_tasks(cpu); |
d2da272a | 5519 | spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5520 | migrate_nr_uninterruptible(rq); |
5521 | BUG_ON(rq->nr_running != 0); | |
5522 | ||
5523 | /* No need to migrate the tasks: it was best-effort if | |
5be9361c | 5524 | * they didn't take sched_hotcpu_mutex. Just wake up |
1da177e4 LT |
5525 | * the requestors. */ |
5526 | spin_lock_irq(&rq->lock); | |
5527 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5528 | struct migration_req *req; |
5529 | ||
1da177e4 | 5530 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5531 | struct migration_req, list); |
1da177e4 LT |
5532 | list_del_init(&req->list); |
5533 | complete(&req->done); | |
5534 | } | |
5535 | spin_unlock_irq(&rq->lock); | |
5536 | break; | |
5537 | #endif | |
5be9361c GS |
5538 | case CPU_LOCK_RELEASE: |
5539 | mutex_unlock(&sched_hotcpu_mutex); | |
5540 | break; | |
1da177e4 LT |
5541 | } |
5542 | return NOTIFY_OK; | |
5543 | } | |
5544 | ||
5545 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5546 | * happens before everything else. | |
5547 | */ | |
26c2143b | 5548 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5549 | .notifier_call = migration_call, |
5550 | .priority = 10 | |
5551 | }; | |
5552 | ||
5553 | int __init migration_init(void) | |
5554 | { | |
5555 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5556 | int err; |
48f24c4d IM |
5557 | |
5558 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5559 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5560 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5561 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5562 | register_cpu_notifier(&migration_notifier); | |
48f24c4d | 5563 | |
1da177e4 LT |
5564 | return 0; |
5565 | } | |
5566 | #endif | |
5567 | ||
5568 | #ifdef CONFIG_SMP | |
476f3534 CL |
5569 | |
5570 | /* Number of possible processor ids */ | |
5571 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5572 | EXPORT_SYMBOL(nr_cpu_ids); | |
5573 | ||
3e9830dc | 5574 | #ifdef CONFIG_SCHED_DEBUG |
1da177e4 LT |
5575 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5576 | { | |
5577 | int level = 0; | |
5578 | ||
41c7ce9a NP |
5579 | if (!sd) { |
5580 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5581 | return; | |
5582 | } | |
5583 | ||
1da177e4 LT |
5584 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5585 | ||
5586 | do { | |
5587 | int i; | |
5588 | char str[NR_CPUS]; | |
5589 | struct sched_group *group = sd->groups; | |
5590 | cpumask_t groupmask; | |
5591 | ||
5592 | cpumask_scnprintf(str, NR_CPUS, sd->span); | |
5593 | cpus_clear(groupmask); | |
5594 | ||
5595 | printk(KERN_DEBUG); | |
5596 | for (i = 0; i < level + 1; i++) | |
5597 | printk(" "); | |
5598 | printk("domain %d: ", level); | |
5599 | ||
5600 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5601 | printk("does not load-balance\n"); | |
5602 | if (sd->parent) | |
33859f7f MOS |
5603 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5604 | " has parent"); | |
1da177e4 LT |
5605 | break; |
5606 | } | |
5607 | ||
5608 | printk("span %s\n", str); | |
5609 | ||
5610 | if (!cpu_isset(cpu, sd->span)) | |
33859f7f MOS |
5611 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5612 | "CPU%d\n", cpu); | |
1da177e4 | 5613 | if (!cpu_isset(cpu, group->cpumask)) |
33859f7f MOS |
5614 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5615 | " CPU%d\n", cpu); | |
1da177e4 LT |
5616 | |
5617 | printk(KERN_DEBUG); | |
5618 | for (i = 0; i < level + 2; i++) | |
5619 | printk(" "); | |
5620 | printk("groups:"); | |
5621 | do { | |
5622 | if (!group) { | |
5623 | printk("\n"); | |
5624 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
5625 | break; | |
5626 | } | |
5627 | ||
5517d86b | 5628 | if (!group->__cpu_power) { |
cc4ea795 | 5629 | printk(KERN_CONT "\n"); |
33859f7f MOS |
5630 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
5631 | "set\n"); | |
26797a34 | 5632 | break; |
1da177e4 LT |
5633 | } |
5634 | ||
5635 | if (!cpus_weight(group->cpumask)) { | |
cc4ea795 | 5636 | printk(KERN_CONT "\n"); |
1da177e4 | 5637 | printk(KERN_ERR "ERROR: empty group\n"); |
26797a34 | 5638 | break; |
1da177e4 LT |
5639 | } |
5640 | ||
5641 | if (cpus_intersects(groupmask, group->cpumask)) { | |
cc4ea795 | 5642 | printk(KERN_CONT "\n"); |
1da177e4 | 5643 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
26797a34 | 5644 | break; |
1da177e4 LT |
5645 | } |
5646 | ||
5647 | cpus_or(groupmask, groupmask, group->cpumask); | |
5648 | ||
5649 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); | |
cc4ea795 | 5650 | printk(KERN_CONT " %s", str); |
1da177e4 LT |
5651 | |
5652 | group = group->next; | |
5653 | } while (group != sd->groups); | |
cc4ea795 | 5654 | printk(KERN_CONT "\n"); |
1da177e4 LT |
5655 | |
5656 | if (!cpus_equal(sd->span, groupmask)) | |
33859f7f MOS |
5657 | printk(KERN_ERR "ERROR: groups don't span " |
5658 | "domain->span\n"); | |
1da177e4 LT |
5659 | |
5660 | level++; | |
5661 | sd = sd->parent; | |
33859f7f MOS |
5662 | if (!sd) |
5663 | continue; | |
1da177e4 | 5664 | |
33859f7f MOS |
5665 | if (!cpus_subset(groupmask, sd->span)) |
5666 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5667 | "of domain->span\n"); | |
1da177e4 LT |
5668 | |
5669 | } while (sd); | |
5670 | } | |
5671 | #else | |
48f24c4d | 5672 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5673 | #endif |
5674 | ||
1a20ff27 | 5675 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5676 | { |
5677 | if (cpus_weight(sd->span) == 1) | |
5678 | return 1; | |
5679 | ||
5680 | /* Following flags need at least 2 groups */ | |
5681 | if (sd->flags & (SD_LOAD_BALANCE | | |
5682 | SD_BALANCE_NEWIDLE | | |
5683 | SD_BALANCE_FORK | | |
89c4710e SS |
5684 | SD_BALANCE_EXEC | |
5685 | SD_SHARE_CPUPOWER | | |
5686 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5687 | if (sd->groups != sd->groups->next) |
5688 | return 0; | |
5689 | } | |
5690 | ||
5691 | /* Following flags don't use groups */ | |
5692 | if (sd->flags & (SD_WAKE_IDLE | | |
5693 | SD_WAKE_AFFINE | | |
5694 | SD_WAKE_BALANCE)) | |
5695 | return 0; | |
5696 | ||
5697 | return 1; | |
5698 | } | |
5699 | ||
48f24c4d IM |
5700 | static int |
5701 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5702 | { |
5703 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5704 | ||
5705 | if (sd_degenerate(parent)) | |
5706 | return 1; | |
5707 | ||
5708 | if (!cpus_equal(sd->span, parent->span)) | |
5709 | return 0; | |
5710 | ||
5711 | /* Does parent contain flags not in child? */ | |
5712 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5713 | if (cflags & SD_WAKE_AFFINE) | |
5714 | pflags &= ~SD_WAKE_BALANCE; | |
5715 | /* Flags needing groups don't count if only 1 group in parent */ | |
5716 | if (parent->groups == parent->groups->next) { | |
5717 | pflags &= ~(SD_LOAD_BALANCE | | |
5718 | SD_BALANCE_NEWIDLE | | |
5719 | SD_BALANCE_FORK | | |
89c4710e SS |
5720 | SD_BALANCE_EXEC | |
5721 | SD_SHARE_CPUPOWER | | |
5722 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5723 | } |
5724 | if (~cflags & pflags) | |
5725 | return 0; | |
5726 | ||
5727 | return 1; | |
5728 | } | |
5729 | ||
1da177e4 LT |
5730 | /* |
5731 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5732 | * hold the hotplug lock. | |
5733 | */ | |
9c1cfda2 | 5734 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5735 | { |
70b97a7f | 5736 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5737 | struct sched_domain *tmp; |
5738 | ||
5739 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5740 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5741 | struct sched_domain *parent = tmp->parent; | |
5742 | if (!parent) | |
5743 | break; | |
1a848870 | 5744 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5745 | tmp->parent = parent->parent; |
1a848870 SS |
5746 | if (parent->parent) |
5747 | parent->parent->child = tmp; | |
5748 | } | |
245af2c7 SS |
5749 | } |
5750 | ||
1a848870 | 5751 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5752 | sd = sd->parent; |
1a848870 SS |
5753 | if (sd) |
5754 | sd->child = NULL; | |
5755 | } | |
1da177e4 LT |
5756 | |
5757 | sched_domain_debug(sd, cpu); | |
5758 | ||
674311d5 | 5759 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5760 | } |
5761 | ||
5762 | /* cpus with isolated domains */ | |
67af63a6 | 5763 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5764 | |
5765 | /* Setup the mask of cpus configured for isolated domains */ | |
5766 | static int __init isolated_cpu_setup(char *str) | |
5767 | { | |
5768 | int ints[NR_CPUS], i; | |
5769 | ||
5770 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5771 | cpus_clear(cpu_isolated_map); | |
5772 | for (i = 1; i <= ints[0]; i++) | |
5773 | if (ints[i] < NR_CPUS) | |
5774 | cpu_set(ints[i], cpu_isolated_map); | |
5775 | return 1; | |
5776 | } | |
5777 | ||
8927f494 | 5778 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5779 | |
5780 | /* | |
6711cab4 SS |
5781 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5782 | * to a function which identifies what group(along with sched group) a CPU | |
5783 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5784 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5785 | * |
5786 | * init_sched_build_groups will build a circular linked list of the groups | |
5787 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5788 | * and ->cpu_power to 0. | |
5789 | */ | |
a616058b | 5790 | static void |
6711cab4 SS |
5791 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5792 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5793 | struct sched_group **sg)) | |
1da177e4 LT |
5794 | { |
5795 | struct sched_group *first = NULL, *last = NULL; | |
5796 | cpumask_t covered = CPU_MASK_NONE; | |
5797 | int i; | |
5798 | ||
5799 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5800 | struct sched_group *sg; |
5801 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5802 | int j; |
5803 | ||
5804 | if (cpu_isset(i, covered)) | |
5805 | continue; | |
5806 | ||
5807 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5808 | sg->__cpu_power = 0; |
1da177e4 LT |
5809 | |
5810 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5811 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5812 | continue; |
5813 | ||
5814 | cpu_set(j, covered); | |
5815 | cpu_set(j, sg->cpumask); | |
5816 | } | |
5817 | if (!first) | |
5818 | first = sg; | |
5819 | if (last) | |
5820 | last->next = sg; | |
5821 | last = sg; | |
5822 | } | |
5823 | last->next = first; | |
5824 | } | |
5825 | ||
9c1cfda2 | 5826 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5827 | |
9c1cfda2 | 5828 | #ifdef CONFIG_NUMA |
198e2f18 | 5829 | |
9c1cfda2 JH |
5830 | /** |
5831 | * find_next_best_node - find the next node to include in a sched_domain | |
5832 | * @node: node whose sched_domain we're building | |
5833 | * @used_nodes: nodes already in the sched_domain | |
5834 | * | |
5835 | * Find the next node to include in a given scheduling domain. Simply | |
5836 | * finds the closest node not already in the @used_nodes map. | |
5837 | * | |
5838 | * Should use nodemask_t. | |
5839 | */ | |
5840 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5841 | { | |
5842 | int i, n, val, min_val, best_node = 0; | |
5843 | ||
5844 | min_val = INT_MAX; | |
5845 | ||
5846 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5847 | /* Start at @node */ | |
5848 | n = (node + i) % MAX_NUMNODES; | |
5849 | ||
5850 | if (!nr_cpus_node(n)) | |
5851 | continue; | |
5852 | ||
5853 | /* Skip already used nodes */ | |
5854 | if (test_bit(n, used_nodes)) | |
5855 | continue; | |
5856 | ||
5857 | /* Simple min distance search */ | |
5858 | val = node_distance(node, n); | |
5859 | ||
5860 | if (val < min_val) { | |
5861 | min_val = val; | |
5862 | best_node = n; | |
5863 | } | |
5864 | } | |
5865 | ||
5866 | set_bit(best_node, used_nodes); | |
5867 | return best_node; | |
5868 | } | |
5869 | ||
5870 | /** | |
5871 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5872 | * @node: node whose cpumask we're constructing | |
5873 | * @size: number of nodes to include in this span | |
5874 | * | |
5875 | * Given a node, construct a good cpumask for its sched_domain to span. It | |
5876 | * should be one that prevents unnecessary balancing, but also spreads tasks | |
5877 | * out optimally. | |
5878 | */ | |
5879 | static cpumask_t sched_domain_node_span(int node) | |
5880 | { | |
9c1cfda2 | 5881 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5882 | cpumask_t span, nodemask; |
5883 | int i; | |
9c1cfda2 JH |
5884 | |
5885 | cpus_clear(span); | |
5886 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5887 | ||
5888 | nodemask = node_to_cpumask(node); | |
5889 | cpus_or(span, span, nodemask); | |
5890 | set_bit(node, used_nodes); | |
5891 | ||
5892 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5893 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5894 | |
9c1cfda2 JH |
5895 | nodemask = node_to_cpumask(next_node); |
5896 | cpus_or(span, span, nodemask); | |
5897 | } | |
5898 | ||
5899 | return span; | |
5900 | } | |
5901 | #endif | |
5902 | ||
5c45bf27 | 5903 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5904 | |
9c1cfda2 | 5905 | /* |
48f24c4d | 5906 | * SMT sched-domains: |
9c1cfda2 | 5907 | */ |
1da177e4 LT |
5908 | #ifdef CONFIG_SCHED_SMT |
5909 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5910 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 5911 | |
6711cab4 SS |
5912 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
5913 | struct sched_group **sg) | |
1da177e4 | 5914 | { |
6711cab4 SS |
5915 | if (sg) |
5916 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
5917 | return cpu; |
5918 | } | |
5919 | #endif | |
5920 | ||
48f24c4d IM |
5921 | /* |
5922 | * multi-core sched-domains: | |
5923 | */ | |
1e9f28fa SS |
5924 | #ifdef CONFIG_SCHED_MC |
5925 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 5926 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
5927 | #endif |
5928 | ||
5929 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
6711cab4 SS |
5930 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5931 | struct sched_group **sg) | |
1e9f28fa | 5932 | { |
6711cab4 | 5933 | int group; |
d5a7430d | 5934 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5935 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
5936 | group = first_cpu(mask); |
5937 | if (sg) | |
5938 | *sg = &per_cpu(sched_group_core, group); | |
5939 | return group; | |
1e9f28fa SS |
5940 | } |
5941 | #elif defined(CONFIG_SCHED_MC) | |
6711cab4 SS |
5942 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
5943 | struct sched_group **sg) | |
1e9f28fa | 5944 | { |
6711cab4 SS |
5945 | if (sg) |
5946 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
5947 | return cpu; |
5948 | } | |
5949 | #endif | |
5950 | ||
1da177e4 | 5951 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 5952 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 5953 | |
6711cab4 SS |
5954 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
5955 | struct sched_group **sg) | |
1da177e4 | 5956 | { |
6711cab4 | 5957 | int group; |
48f24c4d | 5958 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 5959 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 5960 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5961 | group = first_cpu(mask); |
1e9f28fa | 5962 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 5963 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 5964 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 5965 | group = first_cpu(mask); |
1da177e4 | 5966 | #else |
6711cab4 | 5967 | group = cpu; |
1da177e4 | 5968 | #endif |
6711cab4 SS |
5969 | if (sg) |
5970 | *sg = &per_cpu(sched_group_phys, group); | |
5971 | return group; | |
1da177e4 LT |
5972 | } |
5973 | ||
5974 | #ifdef CONFIG_NUMA | |
1da177e4 | 5975 | /* |
9c1cfda2 JH |
5976 | * The init_sched_build_groups can't handle what we want to do with node |
5977 | * groups, so roll our own. Now each node has its own list of groups which | |
5978 | * gets dynamically allocated. | |
1da177e4 | 5979 | */ |
9c1cfda2 | 5980 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 5981 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 5982 | |
9c1cfda2 | 5983 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 5984 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 5985 | |
6711cab4 SS |
5986 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
5987 | struct sched_group **sg) | |
9c1cfda2 | 5988 | { |
6711cab4 SS |
5989 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
5990 | int group; | |
5991 | ||
5992 | cpus_and(nodemask, nodemask, *cpu_map); | |
5993 | group = first_cpu(nodemask); | |
5994 | ||
5995 | if (sg) | |
5996 | *sg = &per_cpu(sched_group_allnodes, group); | |
5997 | return group; | |
1da177e4 | 5998 | } |
6711cab4 | 5999 | |
08069033 SS |
6000 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6001 | { | |
6002 | struct sched_group *sg = group_head; | |
6003 | int j; | |
6004 | ||
6005 | if (!sg) | |
6006 | return; | |
3a5c359a AK |
6007 | do { |
6008 | for_each_cpu_mask(j, sg->cpumask) { | |
6009 | struct sched_domain *sd; | |
08069033 | 6010 | |
3a5c359a AK |
6011 | sd = &per_cpu(phys_domains, j); |
6012 | if (j != first_cpu(sd->groups->cpumask)) { | |
6013 | /* | |
6014 | * Only add "power" once for each | |
6015 | * physical package. | |
6016 | */ | |
6017 | continue; | |
6018 | } | |
08069033 | 6019 | |
3a5c359a AK |
6020 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6021 | } | |
6022 | sg = sg->next; | |
6023 | } while (sg != group_head); | |
08069033 | 6024 | } |
1da177e4 LT |
6025 | #endif |
6026 | ||
a616058b | 6027 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6028 | /* Free memory allocated for various sched_group structures */ |
6029 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6030 | { | |
a616058b | 6031 | int cpu, i; |
51888ca2 SV |
6032 | |
6033 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6034 | struct sched_group **sched_group_nodes |
6035 | = sched_group_nodes_bycpu[cpu]; | |
6036 | ||
51888ca2 SV |
6037 | if (!sched_group_nodes) |
6038 | continue; | |
6039 | ||
6040 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6041 | cpumask_t nodemask = node_to_cpumask(i); | |
6042 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6043 | ||
6044 | cpus_and(nodemask, nodemask, *cpu_map); | |
6045 | if (cpus_empty(nodemask)) | |
6046 | continue; | |
6047 | ||
6048 | if (sg == NULL) | |
6049 | continue; | |
6050 | sg = sg->next; | |
6051 | next_sg: | |
6052 | oldsg = sg; | |
6053 | sg = sg->next; | |
6054 | kfree(oldsg); | |
6055 | if (oldsg != sched_group_nodes[i]) | |
6056 | goto next_sg; | |
6057 | } | |
6058 | kfree(sched_group_nodes); | |
6059 | sched_group_nodes_bycpu[cpu] = NULL; | |
6060 | } | |
51888ca2 | 6061 | } |
a616058b SS |
6062 | #else |
6063 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6064 | { | |
6065 | } | |
6066 | #endif | |
51888ca2 | 6067 | |
89c4710e SS |
6068 | /* |
6069 | * Initialize sched groups cpu_power. | |
6070 | * | |
6071 | * cpu_power indicates the capacity of sched group, which is used while | |
6072 | * distributing the load between different sched groups in a sched domain. | |
6073 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6074 | * there are asymmetries in the topology. If there are asymmetries, group | |
6075 | * having more cpu_power will pickup more load compared to the group having | |
6076 | * less cpu_power. | |
6077 | * | |
6078 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6079 | * the maximum number of tasks a group can handle in the presence of other idle | |
6080 | * or lightly loaded groups in the same sched domain. | |
6081 | */ | |
6082 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6083 | { | |
6084 | struct sched_domain *child; | |
6085 | struct sched_group *group; | |
6086 | ||
6087 | WARN_ON(!sd || !sd->groups); | |
6088 | ||
6089 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6090 | return; | |
6091 | ||
6092 | child = sd->child; | |
6093 | ||
5517d86b ED |
6094 | sd->groups->__cpu_power = 0; |
6095 | ||
89c4710e SS |
6096 | /* |
6097 | * For perf policy, if the groups in child domain share resources | |
6098 | * (for example cores sharing some portions of the cache hierarchy | |
6099 | * or SMT), then set this domain groups cpu_power such that each group | |
6100 | * can handle only one task, when there are other idle groups in the | |
6101 | * same sched domain. | |
6102 | */ | |
6103 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6104 | (child->flags & | |
6105 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6106 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6107 | return; |
6108 | } | |
6109 | ||
89c4710e SS |
6110 | /* |
6111 | * add cpu_power of each child group to this groups cpu_power | |
6112 | */ | |
6113 | group = child->groups; | |
6114 | do { | |
5517d86b | 6115 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6116 | group = group->next; |
6117 | } while (group != child->groups); | |
6118 | } | |
6119 | ||
1da177e4 | 6120 | /* |
1a20ff27 DG |
6121 | * Build sched domains for a given set of cpus and attach the sched domains |
6122 | * to the individual cpus | |
1da177e4 | 6123 | */ |
51888ca2 | 6124 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6125 | { |
6126 | int i; | |
d1b55138 JH |
6127 | #ifdef CONFIG_NUMA |
6128 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6129 | int sd_allnodes = 0; |
d1b55138 JH |
6130 | |
6131 | /* | |
6132 | * Allocate the per-node list of sched groups | |
6133 | */ | |
5cf9f062 | 6134 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
d3a5aa98 | 6135 | GFP_KERNEL); |
d1b55138 JH |
6136 | if (!sched_group_nodes) { |
6137 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6138 | return -ENOMEM; |
d1b55138 JH |
6139 | } |
6140 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6141 | #endif | |
1da177e4 LT |
6142 | |
6143 | /* | |
1a20ff27 | 6144 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6145 | */ |
1a20ff27 | 6146 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6147 | struct sched_domain *sd = NULL, *p; |
6148 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6149 | ||
1a20ff27 | 6150 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6151 | |
6152 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6153 | if (cpus_weight(*cpu_map) > |
6154 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6155 | sd = &per_cpu(allnodes_domains, i); |
6156 | *sd = SD_ALLNODES_INIT; | |
6157 | sd->span = *cpu_map; | |
6711cab4 | 6158 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6159 | p = sd; |
6711cab4 | 6160 | sd_allnodes = 1; |
9c1cfda2 JH |
6161 | } else |
6162 | p = NULL; | |
6163 | ||
1da177e4 | 6164 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6165 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6166 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6167 | sd->parent = p; | |
1a848870 SS |
6168 | if (p) |
6169 | p->child = sd; | |
9c1cfda2 | 6170 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6171 | #endif |
6172 | ||
6173 | p = sd; | |
6174 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6175 | *sd = SD_CPU_INIT; |
6176 | sd->span = nodemask; | |
6177 | sd->parent = p; | |
1a848870 SS |
6178 | if (p) |
6179 | p->child = sd; | |
6711cab4 | 6180 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6181 | |
1e9f28fa SS |
6182 | #ifdef CONFIG_SCHED_MC |
6183 | p = sd; | |
6184 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6185 | *sd = SD_MC_INIT; |
6186 | sd->span = cpu_coregroup_map(i); | |
6187 | cpus_and(sd->span, sd->span, *cpu_map); | |
6188 | sd->parent = p; | |
1a848870 | 6189 | p->child = sd; |
6711cab4 | 6190 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6191 | #endif |
6192 | ||
1da177e4 LT |
6193 | #ifdef CONFIG_SCHED_SMT |
6194 | p = sd; | |
6195 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6196 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6197 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6198 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6199 | sd->parent = p; |
1a848870 | 6200 | p->child = sd; |
6711cab4 | 6201 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6202 | #endif |
6203 | } | |
6204 | ||
6205 | #ifdef CONFIG_SCHED_SMT | |
6206 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6207 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6208 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6209 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6210 | if (i != first_cpu(this_sibling_map)) |
6211 | continue; | |
6212 | ||
dd41f596 IM |
6213 | init_sched_build_groups(this_sibling_map, cpu_map, |
6214 | &cpu_to_cpu_group); | |
1da177e4 LT |
6215 | } |
6216 | #endif | |
6217 | ||
1e9f28fa SS |
6218 | #ifdef CONFIG_SCHED_MC |
6219 | /* Set up multi-core groups */ | |
6220 | for_each_cpu_mask(i, *cpu_map) { | |
6221 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6222 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6223 | if (i != first_cpu(this_core_map)) | |
6224 | continue; | |
dd41f596 IM |
6225 | init_sched_build_groups(this_core_map, cpu_map, |
6226 | &cpu_to_core_group); | |
1e9f28fa SS |
6227 | } |
6228 | #endif | |
6229 | ||
1da177e4 LT |
6230 | /* Set up physical groups */ |
6231 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6232 | cpumask_t nodemask = node_to_cpumask(i); | |
6233 | ||
1a20ff27 | 6234 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6235 | if (cpus_empty(nodemask)) |
6236 | continue; | |
6237 | ||
6711cab4 | 6238 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6239 | } |
6240 | ||
6241 | #ifdef CONFIG_NUMA | |
6242 | /* Set up node groups */ | |
6711cab4 | 6243 | if (sd_allnodes) |
dd41f596 IM |
6244 | init_sched_build_groups(*cpu_map, cpu_map, |
6245 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6246 | |
6247 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6248 | /* Set up node groups */ | |
6249 | struct sched_group *sg, *prev; | |
6250 | cpumask_t nodemask = node_to_cpumask(i); | |
6251 | cpumask_t domainspan; | |
6252 | cpumask_t covered = CPU_MASK_NONE; | |
6253 | int j; | |
6254 | ||
6255 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6256 | if (cpus_empty(nodemask)) { |
6257 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6258 | continue; |
d1b55138 | 6259 | } |
9c1cfda2 JH |
6260 | |
6261 | domainspan = sched_domain_node_span(i); | |
6262 | cpus_and(domainspan, domainspan, *cpu_map); | |
6263 | ||
15f0b676 | 6264 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6265 | if (!sg) { |
6266 | printk(KERN_WARNING "Can not alloc domain group for " | |
6267 | "node %d\n", i); | |
6268 | goto error; | |
6269 | } | |
9c1cfda2 JH |
6270 | sched_group_nodes[i] = sg; |
6271 | for_each_cpu_mask(j, nodemask) { | |
6272 | struct sched_domain *sd; | |
9761eea8 | 6273 | |
9c1cfda2 JH |
6274 | sd = &per_cpu(node_domains, j); |
6275 | sd->groups = sg; | |
9c1cfda2 | 6276 | } |
5517d86b | 6277 | sg->__cpu_power = 0; |
9c1cfda2 | 6278 | sg->cpumask = nodemask; |
51888ca2 | 6279 | sg->next = sg; |
9c1cfda2 JH |
6280 | cpus_or(covered, covered, nodemask); |
6281 | prev = sg; | |
6282 | ||
6283 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6284 | cpumask_t tmp, notcovered; | |
6285 | int n = (i + j) % MAX_NUMNODES; | |
6286 | ||
6287 | cpus_complement(notcovered, covered); | |
6288 | cpus_and(tmp, notcovered, *cpu_map); | |
6289 | cpus_and(tmp, tmp, domainspan); | |
6290 | if (cpus_empty(tmp)) | |
6291 | break; | |
6292 | ||
6293 | nodemask = node_to_cpumask(n); | |
6294 | cpus_and(tmp, tmp, nodemask); | |
6295 | if (cpus_empty(tmp)) | |
6296 | continue; | |
6297 | ||
15f0b676 SV |
6298 | sg = kmalloc_node(sizeof(struct sched_group), |
6299 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6300 | if (!sg) { |
6301 | printk(KERN_WARNING | |
6302 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6303 | goto error; |
9c1cfda2 | 6304 | } |
5517d86b | 6305 | sg->__cpu_power = 0; |
9c1cfda2 | 6306 | sg->cpumask = tmp; |
51888ca2 | 6307 | sg->next = prev->next; |
9c1cfda2 JH |
6308 | cpus_or(covered, covered, tmp); |
6309 | prev->next = sg; | |
6310 | prev = sg; | |
6311 | } | |
9c1cfda2 | 6312 | } |
1da177e4 LT |
6313 | #endif |
6314 | ||
6315 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6316 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6317 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6318 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6319 | ||
89c4710e | 6320 | init_sched_groups_power(i, sd); |
5c45bf27 | 6321 | } |
1da177e4 | 6322 | #endif |
1e9f28fa | 6323 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6324 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6325 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6326 | ||
89c4710e | 6327 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6328 | } |
6329 | #endif | |
1e9f28fa | 6330 | |
5c45bf27 | 6331 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6332 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6333 | ||
89c4710e | 6334 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6335 | } |
6336 | ||
9c1cfda2 | 6337 | #ifdef CONFIG_NUMA |
08069033 SS |
6338 | for (i = 0; i < MAX_NUMNODES; i++) |
6339 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6340 | |
6711cab4 SS |
6341 | if (sd_allnodes) { |
6342 | struct sched_group *sg; | |
f712c0c7 | 6343 | |
6711cab4 | 6344 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6345 | init_numa_sched_groups_power(sg); |
6346 | } | |
9c1cfda2 JH |
6347 | #endif |
6348 | ||
1da177e4 | 6349 | /* Attach the domains */ |
1a20ff27 | 6350 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6351 | struct sched_domain *sd; |
6352 | #ifdef CONFIG_SCHED_SMT | |
6353 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6354 | #elif defined(CONFIG_SCHED_MC) |
6355 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6356 | #else |
6357 | sd = &per_cpu(phys_domains, i); | |
6358 | #endif | |
6359 | cpu_attach_domain(sd, i); | |
6360 | } | |
51888ca2 SV |
6361 | |
6362 | return 0; | |
6363 | ||
a616058b | 6364 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6365 | error: |
6366 | free_sched_groups(cpu_map); | |
6367 | return -ENOMEM; | |
a616058b | 6368 | #endif |
1da177e4 | 6369 | } |
1a20ff27 DG |
6370 | /* |
6371 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. | |
6372 | */ | |
51888ca2 | 6373 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6374 | { |
6375 | cpumask_t cpu_default_map; | |
51888ca2 | 6376 | int err; |
1da177e4 | 6377 | |
1a20ff27 DG |
6378 | /* |
6379 | * Setup mask for cpus without special case scheduling requirements. | |
6380 | * For now this just excludes isolated cpus, but could be used to | |
6381 | * exclude other special cases in the future. | |
6382 | */ | |
6383 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); | |
6384 | ||
51888ca2 SV |
6385 | err = build_sched_domains(&cpu_default_map); |
6386 | ||
6382bc90 MM |
6387 | register_sched_domain_sysctl(); |
6388 | ||
51888ca2 | 6389 | return err; |
1a20ff27 DG |
6390 | } |
6391 | ||
6392 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6393 | { |
51888ca2 | 6394 | free_sched_groups(cpu_map); |
9c1cfda2 | 6395 | } |
1da177e4 | 6396 | |
1a20ff27 DG |
6397 | /* |
6398 | * Detach sched domains from a group of cpus specified in cpu_map | |
6399 | * These cpus will now be attached to the NULL domain | |
6400 | */ | |
858119e1 | 6401 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6402 | { |
6403 | int i; | |
6404 | ||
6382bc90 MM |
6405 | unregister_sched_domain_sysctl(); |
6406 | ||
1a20ff27 DG |
6407 | for_each_cpu_mask(i, *cpu_map) |
6408 | cpu_attach_domain(NULL, i); | |
6409 | synchronize_sched(); | |
6410 | arch_destroy_sched_domains(cpu_map); | |
6411 | } | |
6412 | ||
5c45bf27 | 6413 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6414 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6415 | { |
6416 | int err; | |
6417 | ||
5be9361c | 6418 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6419 | detach_destroy_domains(&cpu_online_map); |
6420 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6421 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6422 | |
6423 | return err; | |
6424 | } | |
6425 | ||
6426 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6427 | { | |
6428 | int ret; | |
6429 | ||
6430 | if (buf[0] != '0' && buf[0] != '1') | |
6431 | return -EINVAL; | |
6432 | ||
6433 | if (smt) | |
6434 | sched_smt_power_savings = (buf[0] == '1'); | |
6435 | else | |
6436 | sched_mc_power_savings = (buf[0] == '1'); | |
6437 | ||
6438 | ret = arch_reinit_sched_domains(); | |
6439 | ||
6440 | return ret ? ret : count; | |
6441 | } | |
6442 | ||
5c45bf27 SS |
6443 | #ifdef CONFIG_SCHED_MC |
6444 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6445 | { | |
6446 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6447 | } | |
48f24c4d IM |
6448 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6449 | const char *buf, size_t count) | |
5c45bf27 SS |
6450 | { |
6451 | return sched_power_savings_store(buf, count, 0); | |
6452 | } | |
6707de00 AB |
6453 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6454 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6455 | #endif |
6456 | ||
6457 | #ifdef CONFIG_SCHED_SMT | |
6458 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6459 | { | |
6460 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6461 | } | |
48f24c4d IM |
6462 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6463 | const char *buf, size_t count) | |
5c45bf27 SS |
6464 | { |
6465 | return sched_power_savings_store(buf, count, 1); | |
6466 | } | |
6707de00 AB |
6467 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6468 | sched_smt_power_savings_store); | |
6469 | #endif | |
6470 | ||
6471 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6472 | { | |
6473 | int err = 0; | |
6474 | ||
6475 | #ifdef CONFIG_SCHED_SMT | |
6476 | if (smt_capable()) | |
6477 | err = sysfs_create_file(&cls->kset.kobj, | |
6478 | &attr_sched_smt_power_savings.attr); | |
6479 | #endif | |
6480 | #ifdef CONFIG_SCHED_MC | |
6481 | if (!err && mc_capable()) | |
6482 | err = sysfs_create_file(&cls->kset.kobj, | |
6483 | &attr_sched_mc_power_savings.attr); | |
6484 | #endif | |
6485 | return err; | |
6486 | } | |
5c45bf27 SS |
6487 | #endif |
6488 | ||
1da177e4 LT |
6489 | /* |
6490 | * Force a reinitialization of the sched domains hierarchy. The domains | |
6491 | * and groups cannot be updated in place without racing with the balancing | |
41c7ce9a | 6492 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6493 | * which will prevent rebalancing while the sched domains are recalculated. |
6494 | */ | |
6495 | static int update_sched_domains(struct notifier_block *nfb, | |
6496 | unsigned long action, void *hcpu) | |
6497 | { | |
1da177e4 LT |
6498 | switch (action) { |
6499 | case CPU_UP_PREPARE: | |
8bb78442 | 6500 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6501 | case CPU_DOWN_PREPARE: |
8bb78442 | 6502 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6503 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6504 | return NOTIFY_OK; |
6505 | ||
6506 | case CPU_UP_CANCELED: | |
8bb78442 | 6507 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6508 | case CPU_DOWN_FAILED: |
8bb78442 | 6509 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6510 | case CPU_ONLINE: |
8bb78442 | 6511 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6512 | case CPU_DEAD: |
8bb78442 | 6513 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6514 | /* |
6515 | * Fall through and re-initialise the domains. | |
6516 | */ | |
6517 | break; | |
6518 | default: | |
6519 | return NOTIFY_DONE; | |
6520 | } | |
6521 | ||
6522 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6523 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6524 | |
6525 | return NOTIFY_OK; | |
6526 | } | |
1da177e4 LT |
6527 | |
6528 | void __init sched_init_smp(void) | |
6529 | { | |
5c1e1767 NP |
6530 | cpumask_t non_isolated_cpus; |
6531 | ||
5be9361c | 6532 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6533 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6534 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6535 | if (cpus_empty(non_isolated_cpus)) |
6536 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6537 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6538 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6539 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6540 | |
6541 | /* Move init over to a non-isolated CPU */ | |
6542 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6543 | BUG(); | |
1da177e4 LT |
6544 | } |
6545 | #else | |
6546 | void __init sched_init_smp(void) | |
6547 | { | |
6548 | } | |
6549 | #endif /* CONFIG_SMP */ | |
6550 | ||
6551 | int in_sched_functions(unsigned long addr) | |
6552 | { | |
6553 | /* Linker adds these: start and end of __sched functions */ | |
6554 | extern char __sched_text_start[], __sched_text_end[]; | |
48f24c4d | 6555 | |
1da177e4 LT |
6556 | return in_lock_functions(addr) || |
6557 | (addr >= (unsigned long)__sched_text_start | |
6558 | && addr < (unsigned long)__sched_text_end); | |
6559 | } | |
6560 | ||
a9957449 | 6561 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6562 | { |
6563 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6564 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6565 | cfs_rq->rq = rq; | |
6566 | #endif | |
67e9fb2a | 6567 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6568 | } |
6569 | ||
1da177e4 LT |
6570 | void __init sched_init(void) |
6571 | { | |
476f3534 | 6572 | int highest_cpu = 0; |
dd41f596 IM |
6573 | int i, j; |
6574 | ||
0a945022 | 6575 | for_each_possible_cpu(i) { |
dd41f596 | 6576 | struct rt_prio_array *array; |
70b97a7f | 6577 | struct rq *rq; |
1da177e4 LT |
6578 | |
6579 | rq = cpu_rq(i); | |
6580 | spin_lock_init(&rq->lock); | |
fcb99371 | 6581 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6582 | rq->nr_running = 0; |
dd41f596 IM |
6583 | rq->clock = 1; |
6584 | init_cfs_rq(&rq->cfs, rq); | |
6585 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6586 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6587 | { |
6588 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6589 | struct sched_entity *se = | |
6590 | &per_cpu(init_sched_entity, i); | |
6591 | ||
6592 | init_cfs_rq_p[i] = cfs_rq; | |
6593 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6594 | cfs_rq->tg = &init_task_group; |
3a252015 | 6595 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6596 | &rq->leaf_cfs_rq_list); |
6597 | ||
3a252015 IM |
6598 | init_sched_entity_p[i] = se; |
6599 | se->cfs_rq = &rq->cfs; | |
6600 | se->my_q = cfs_rq; | |
4cf86d77 | 6601 | se->load.weight = init_task_group_load; |
9b5b7751 | 6602 | se->load.inv_weight = |
4cf86d77 | 6603 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6604 | se->parent = NULL; |
6605 | } | |
4cf86d77 | 6606 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6607 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6608 | #endif |
1da177e4 | 6609 | |
dd41f596 IM |
6610 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6611 | rq->cpu_load[j] = 0; | |
1da177e4 | 6612 | #ifdef CONFIG_SMP |
41c7ce9a | 6613 | rq->sd = NULL; |
1da177e4 | 6614 | rq->active_balance = 0; |
dd41f596 | 6615 | rq->next_balance = jiffies; |
1da177e4 | 6616 | rq->push_cpu = 0; |
0a2966b4 | 6617 | rq->cpu = i; |
1da177e4 LT |
6618 | rq->migration_thread = NULL; |
6619 | INIT_LIST_HEAD(&rq->migration_queue); | |
6620 | #endif | |
6621 | atomic_set(&rq->nr_iowait, 0); | |
6622 | ||
dd41f596 IM |
6623 | array = &rq->rt.active; |
6624 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6625 | INIT_LIST_HEAD(array->queue + j); | |
6626 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6627 | } |
476f3534 | 6628 | highest_cpu = i; |
dd41f596 IM |
6629 | /* delimiter for bitsearch: */ |
6630 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6631 | } |
6632 | ||
2dd73a4f | 6633 | set_load_weight(&init_task); |
b50f60ce | 6634 | |
e107be36 AK |
6635 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6636 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6637 | #endif | |
6638 | ||
c9819f45 | 6639 | #ifdef CONFIG_SMP |
476f3534 | 6640 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6641 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6642 | #endif | |
6643 | ||
b50f60ce HC |
6644 | #ifdef CONFIG_RT_MUTEXES |
6645 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6646 | #endif | |
6647 | ||
1da177e4 LT |
6648 | /* |
6649 | * The boot idle thread does lazy MMU switching as well: | |
6650 | */ | |
6651 | atomic_inc(&init_mm.mm_count); | |
6652 | enter_lazy_tlb(&init_mm, current); | |
6653 | ||
6654 | /* | |
6655 | * Make us the idle thread. Technically, schedule() should not be | |
6656 | * called from this thread, however somewhere below it might be, | |
6657 | * but because we are the idle thread, we just pick up running again | |
6658 | * when this runqueue becomes "idle". | |
6659 | */ | |
6660 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6661 | /* |
6662 | * During early bootup we pretend to be a normal task: | |
6663 | */ | |
6664 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6665 | } |
6666 | ||
6667 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6668 | void __might_sleep(char *file, int line) | |
6669 | { | |
48f24c4d | 6670 | #ifdef in_atomic |
1da177e4 LT |
6671 | static unsigned long prev_jiffy; /* ratelimiting */ |
6672 | ||
6673 | if ((in_atomic() || irqs_disabled()) && | |
6674 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6675 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6676 | return; | |
6677 | prev_jiffy = jiffies; | |
91368d73 | 6678 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6679 | " context at %s:%d\n", file, line); |
6680 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6681 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6682 | debug_show_held_locks(current); |
3117df04 IM |
6683 | if (irqs_disabled()) |
6684 | print_irqtrace_events(current); | |
1da177e4 LT |
6685 | dump_stack(); |
6686 | } | |
6687 | #endif | |
6688 | } | |
6689 | EXPORT_SYMBOL(__might_sleep); | |
6690 | #endif | |
6691 | ||
6692 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6693 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6694 | { | |
6695 | int on_rq; | |
6696 | update_rq_clock(rq); | |
6697 | on_rq = p->se.on_rq; | |
6698 | if (on_rq) | |
6699 | deactivate_task(rq, p, 0); | |
6700 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6701 | if (on_rq) { | |
6702 | activate_task(rq, p, 0); | |
6703 | resched_task(rq->curr); | |
6704 | } | |
6705 | } | |
6706 | ||
1da177e4 LT |
6707 | void normalize_rt_tasks(void) |
6708 | { | |
a0f98a1c | 6709 | struct task_struct *g, *p; |
1da177e4 | 6710 | unsigned long flags; |
70b97a7f | 6711 | struct rq *rq; |
1da177e4 LT |
6712 | |
6713 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6714 | do_each_thread(g, p) { |
178be793 IM |
6715 | /* |
6716 | * Only normalize user tasks: | |
6717 | */ | |
6718 | if (!p->mm) | |
6719 | continue; | |
6720 | ||
6cfb0d5d | 6721 | p->se.exec_start = 0; |
6cfb0d5d | 6722 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6723 | p->se.wait_start = 0; |
dd41f596 | 6724 | p->se.sleep_start = 0; |
dd41f596 | 6725 | p->se.block_start = 0; |
6cfb0d5d | 6726 | #endif |
dd41f596 IM |
6727 | task_rq(p)->clock = 0; |
6728 | ||
6729 | if (!rt_task(p)) { | |
6730 | /* | |
6731 | * Renice negative nice level userspace | |
6732 | * tasks back to 0: | |
6733 | */ | |
6734 | if (TASK_NICE(p) < 0 && p->mm) | |
6735 | set_user_nice(p, 0); | |
1da177e4 | 6736 | continue; |
dd41f596 | 6737 | } |
1da177e4 | 6738 | |
b29739f9 IM |
6739 | spin_lock_irqsave(&p->pi_lock, flags); |
6740 | rq = __task_rq_lock(p); | |
1da177e4 | 6741 | |
178be793 | 6742 | normalize_task(rq, p); |
3a5e4dc1 | 6743 | |
b29739f9 IM |
6744 | __task_rq_unlock(rq); |
6745 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6746 | } while_each_thread(g, p); |
6747 | ||
1da177e4 LT |
6748 | read_unlock_irq(&tasklist_lock); |
6749 | } | |
6750 | ||
6751 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6752 | |
6753 | #ifdef CONFIG_IA64 | |
6754 | /* | |
6755 | * These functions are only useful for the IA64 MCA handling. | |
6756 | * | |
6757 | * They can only be called when the whole system has been | |
6758 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6759 | * activity can take place. Using them for anything else would | |
6760 | * be a serious bug, and as a result, they aren't even visible | |
6761 | * under any other configuration. | |
6762 | */ | |
6763 | ||
6764 | /** | |
6765 | * curr_task - return the current task for a given cpu. | |
6766 | * @cpu: the processor in question. | |
6767 | * | |
6768 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6769 | */ | |
36c8b586 | 6770 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6771 | { |
6772 | return cpu_curr(cpu); | |
6773 | } | |
6774 | ||
6775 | /** | |
6776 | * set_curr_task - set the current task for a given cpu. | |
6777 | * @cpu: the processor in question. | |
6778 | * @p: the task pointer to set. | |
6779 | * | |
6780 | * Description: This function must only be used when non-maskable interrupts | |
6781 | * are serviced on a separate stack. It allows the architecture to switch the | |
6782 | * notion of the current task on a cpu in a non-blocking manner. This function | |
6783 | * must be called with all CPU's synchronized, and interrupts disabled, the | |
6784 | * and caller must save the original value of the current task (see | |
6785 | * curr_task() above) and restore that value before reenabling interrupts and | |
6786 | * re-starting the system. | |
6787 | * | |
6788 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6789 | */ | |
36c8b586 | 6790 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6791 | { |
6792 | cpu_curr(cpu) = p; | |
6793 | } | |
6794 | ||
6795 | #endif | |
29f59db3 SV |
6796 | |
6797 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6798 | ||
29f59db3 | 6799 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6800 | struct task_group *sched_create_group(void) |
29f59db3 | 6801 | { |
4cf86d77 | 6802 | struct task_group *tg; |
29f59db3 SV |
6803 | struct cfs_rq *cfs_rq; |
6804 | struct sched_entity *se; | |
9b5b7751 | 6805 | struct rq *rq; |
29f59db3 SV |
6806 | int i; |
6807 | ||
29f59db3 SV |
6808 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6809 | if (!tg) | |
6810 | return ERR_PTR(-ENOMEM); | |
6811 | ||
9b5b7751 | 6812 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6813 | if (!tg->cfs_rq) |
6814 | goto err; | |
9b5b7751 | 6815 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6816 | if (!tg->se) |
6817 | goto err; | |
6818 | ||
6819 | for_each_possible_cpu(i) { | |
9b5b7751 | 6820 | rq = cpu_rq(i); |
29f59db3 SV |
6821 | |
6822 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6823 | cpu_to_node(i)); | |
6824 | if (!cfs_rq) | |
6825 | goto err; | |
6826 | ||
6827 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6828 | cpu_to_node(i)); | |
6829 | if (!se) | |
6830 | goto err; | |
6831 | ||
6832 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6833 | memset(se, 0, sizeof(struct sched_entity)); | |
6834 | ||
6835 | tg->cfs_rq[i] = cfs_rq; | |
6836 | init_cfs_rq(cfs_rq, rq); | |
6837 | cfs_rq->tg = tg; | |
29f59db3 SV |
6838 | |
6839 | tg->se[i] = se; | |
6840 | se->cfs_rq = &rq->cfs; | |
6841 | se->my_q = cfs_rq; | |
6842 | se->load.weight = NICE_0_LOAD; | |
6843 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
6844 | se->parent = NULL; | |
6845 | } | |
6846 | ||
9b5b7751 SV |
6847 | for_each_possible_cpu(i) { |
6848 | rq = cpu_rq(i); | |
6849 | cfs_rq = tg->cfs_rq[i]; | |
6850 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
6851 | } | |
29f59db3 | 6852 | |
9b5b7751 | 6853 | tg->shares = NICE_0_LOAD; |
5cb350ba | 6854 | spin_lock_init(&tg->lock); |
29f59db3 | 6855 | |
9b5b7751 | 6856 | return tg; |
29f59db3 SV |
6857 | |
6858 | err: | |
6859 | for_each_possible_cpu(i) { | |
a65914b3 | 6860 | if (tg->cfs_rq) |
29f59db3 | 6861 | kfree(tg->cfs_rq[i]); |
a65914b3 | 6862 | if (tg->se) |
29f59db3 SV |
6863 | kfree(tg->se[i]); |
6864 | } | |
a65914b3 IM |
6865 | kfree(tg->cfs_rq); |
6866 | kfree(tg->se); | |
6867 | kfree(tg); | |
29f59db3 SV |
6868 | |
6869 | return ERR_PTR(-ENOMEM); | |
6870 | } | |
6871 | ||
9b5b7751 SV |
6872 | /* rcu callback to free various structures associated with a task group */ |
6873 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 6874 | { |
9b5b7751 | 6875 | struct cfs_rq *cfs_rq = container_of(rhp, struct cfs_rq, rcu); |
4cf86d77 | 6876 | struct task_group *tg = cfs_rq->tg; |
29f59db3 SV |
6877 | struct sched_entity *se; |
6878 | int i; | |
6879 | ||
29f59db3 SV |
6880 | /* now it should be safe to free those cfs_rqs */ |
6881 | for_each_possible_cpu(i) { | |
6882 | cfs_rq = tg->cfs_rq[i]; | |
6883 | kfree(cfs_rq); | |
6884 | ||
6885 | se = tg->se[i]; | |
6886 | kfree(se); | |
6887 | } | |
6888 | ||
6889 | kfree(tg->cfs_rq); | |
6890 | kfree(tg->se); | |
6891 | kfree(tg); | |
6892 | } | |
6893 | ||
9b5b7751 | 6894 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 6895 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 6896 | { |
9b5b7751 SV |
6897 | struct cfs_rq *cfs_rq; |
6898 | int i; | |
29f59db3 | 6899 | |
9b5b7751 SV |
6900 | for_each_possible_cpu(i) { |
6901 | cfs_rq = tg->cfs_rq[i]; | |
6902 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
6903 | } | |
6904 | ||
6905 | cfs_rq = tg->cfs_rq[0]; | |
6906 | ||
6907 | /* wait for possible concurrent references to cfs_rqs complete */ | |
6908 | call_rcu(&cfs_rq->rcu, free_sched_group); | |
29f59db3 SV |
6909 | } |
6910 | ||
9b5b7751 | 6911 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
6912 | * The caller of this function should have put the task in its new group |
6913 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
6914 | * reflect its new group. | |
9b5b7751 SV |
6915 | */ |
6916 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
6917 | { |
6918 | int on_rq, running; | |
6919 | unsigned long flags; | |
6920 | struct rq *rq; | |
6921 | ||
6922 | rq = task_rq_lock(tsk, &flags); | |
6923 | ||
6924 | if (tsk->sched_class != &fair_sched_class) | |
6925 | goto done; | |
6926 | ||
6927 | update_rq_clock(rq); | |
6928 | ||
6929 | running = task_running(rq, tsk); | |
6930 | on_rq = tsk->se.on_rq; | |
6931 | ||
83b699ed | 6932 | if (on_rq) { |
29f59db3 | 6933 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
6934 | if (unlikely(running)) |
6935 | tsk->sched_class->put_prev_task(rq, tsk); | |
6936 | } | |
29f59db3 SV |
6937 | |
6938 | set_task_cfs_rq(tsk); | |
6939 | ||
83b699ed SV |
6940 | if (on_rq) { |
6941 | if (unlikely(running)) | |
6942 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 6943 | enqueue_task(rq, tsk, 0); |
83b699ed | 6944 | } |
29f59db3 SV |
6945 | |
6946 | done: | |
6947 | task_rq_unlock(rq, &flags); | |
6948 | } | |
6949 | ||
6950 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
6951 | { | |
6952 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
6953 | struct rq *rq = cfs_rq->rq; | |
6954 | int on_rq; | |
6955 | ||
6956 | spin_lock_irq(&rq->lock); | |
6957 | ||
6958 | on_rq = se->on_rq; | |
6959 | if (on_rq) | |
6960 | dequeue_entity(cfs_rq, se, 0); | |
6961 | ||
6962 | se->load.weight = shares; | |
6963 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
6964 | ||
6965 | if (on_rq) | |
6966 | enqueue_entity(cfs_rq, se, 0); | |
6967 | ||
6968 | spin_unlock_irq(&rq->lock); | |
6969 | } | |
6970 | ||
4cf86d77 | 6971 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
6972 | { |
6973 | int i; | |
29f59db3 | 6974 | |
5cb350ba | 6975 | spin_lock(&tg->lock); |
9b5b7751 | 6976 | if (tg->shares == shares) |
5cb350ba | 6977 | goto done; |
29f59db3 | 6978 | |
9b5b7751 | 6979 | tg->shares = shares; |
29f59db3 | 6980 | for_each_possible_cpu(i) |
9b5b7751 | 6981 | set_se_shares(tg->se[i], shares); |
29f59db3 | 6982 | |
5cb350ba DG |
6983 | done: |
6984 | spin_unlock(&tg->lock); | |
9b5b7751 | 6985 | return 0; |
29f59db3 SV |
6986 | } |
6987 | ||
5cb350ba DG |
6988 | unsigned long sched_group_shares(struct task_group *tg) |
6989 | { | |
6990 | return tg->shares; | |
6991 | } | |
6992 | ||
3a252015 | 6993 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |