Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
1da177e4 LT |
25 | */ |
26 | ||
27 | #include <linux/mm.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/nmi.h> | |
30 | #include <linux/init.h> | |
dff06c15 | 31 | #include <linux/uaccess.h> |
1da177e4 LT |
32 | #include <linux/highmem.h> |
33 | #include <linux/smp_lock.h> | |
34 | #include <asm/mmu_context.h> | |
35 | #include <linux/interrupt.h> | |
c59ede7b | 36 | #include <linux/capability.h> |
1da177e4 LT |
37 | #include <linux/completion.h> |
38 | #include <linux/kernel_stat.h> | |
9a11b49a | 39 | #include <linux/debug_locks.h> |
1da177e4 LT |
40 | #include <linux/security.h> |
41 | #include <linux/notifier.h> | |
42 | #include <linux/profile.h> | |
7dfb7103 | 43 | #include <linux/freezer.h> |
198e2f18 | 44 | #include <linux/vmalloc.h> |
1da177e4 LT |
45 | #include <linux/blkdev.h> |
46 | #include <linux/delay.h> | |
b488893a | 47 | #include <linux/pid_namespace.h> |
1da177e4 LT |
48 | #include <linux/smp.h> |
49 | #include <linux/threads.h> | |
50 | #include <linux/timer.h> | |
51 | #include <linux/rcupdate.h> | |
52 | #include <linux/cpu.h> | |
53 | #include <linux/cpuset.h> | |
54 | #include <linux/percpu.h> | |
55 | #include <linux/kthread.h> | |
56 | #include <linux/seq_file.h> | |
e692ab53 | 57 | #include <linux/sysctl.h> |
1da177e4 LT |
58 | #include <linux/syscalls.h> |
59 | #include <linux/times.h> | |
8f0ab514 | 60 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 61 | #include <linux/kprobes.h> |
0ff92245 | 62 | #include <linux/delayacct.h> |
5517d86b | 63 | #include <linux/reciprocal_div.h> |
dff06c15 | 64 | #include <linux/unistd.h> |
f5ff8422 | 65 | #include <linux/pagemap.h> |
1da177e4 | 66 | |
5517d86b | 67 | #include <asm/tlb.h> |
838225b4 | 68 | #include <asm/irq_regs.h> |
1da177e4 | 69 | |
b035b6de AD |
70 | /* |
71 | * Scheduler clock - returns current time in nanosec units. | |
72 | * This is default implementation. | |
73 | * Architectures and sub-architectures can override this. | |
74 | */ | |
75 | unsigned long long __attribute__((weak)) sched_clock(void) | |
76 | { | |
d6322faf | 77 | return (unsigned long long)jiffies * (NSEC_PER_SEC / HZ); |
b035b6de AD |
78 | } |
79 | ||
1da177e4 LT |
80 | /* |
81 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
82 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
83 | * and back. | |
84 | */ | |
85 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
86 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
87 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
88 | ||
89 | /* | |
90 | * 'User priority' is the nice value converted to something we | |
91 | * can work with better when scaling various scheduler parameters, | |
92 | * it's a [ 0 ... 39 ] range. | |
93 | */ | |
94 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
95 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
96 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
97 | ||
98 | /* | |
99 | * Some helpers for converting nanosecond timing to jiffy resolution | |
100 | */ | |
d6322faf ED |
101 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
102 | #define JIFFIES_TO_NS(TIME) ((TIME) * (NSEC_PER_SEC / HZ)) | |
1da177e4 | 103 | |
6aa645ea IM |
104 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
105 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
106 | ||
1da177e4 LT |
107 | /* |
108 | * These are the 'tuning knobs' of the scheduler: | |
109 | * | |
a4ec24b4 | 110 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
111 | * Timeslices get refilled after they expire. |
112 | */ | |
1da177e4 | 113 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 114 | |
5517d86b ED |
115 | #ifdef CONFIG_SMP |
116 | /* | |
117 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
118 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
119 | */ | |
120 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
121 | { | |
122 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Each time a sched group cpu_power is changed, | |
127 | * we must compute its reciprocal value | |
128 | */ | |
129 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
130 | { | |
131 | sg->__cpu_power += val; | |
132 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
133 | } | |
134 | #endif | |
135 | ||
e05606d3 IM |
136 | static inline int rt_policy(int policy) |
137 | { | |
138 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) | |
139 | return 1; | |
140 | return 0; | |
141 | } | |
142 | ||
143 | static inline int task_has_rt_policy(struct task_struct *p) | |
144 | { | |
145 | return rt_policy(p->policy); | |
146 | } | |
147 | ||
1da177e4 | 148 | /* |
6aa645ea | 149 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 150 | */ |
6aa645ea IM |
151 | struct rt_prio_array { |
152 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
153 | struct list_head queue[MAX_RT_PRIO]; | |
154 | }; | |
155 | ||
29f59db3 SV |
156 | #ifdef CONFIG_FAIR_GROUP_SCHED |
157 | ||
68318b8e SV |
158 | #include <linux/cgroup.h> |
159 | ||
29f59db3 SV |
160 | struct cfs_rq; |
161 | ||
162 | /* task group related information */ | |
4cf86d77 | 163 | struct task_group { |
68318b8e SV |
164 | #ifdef CONFIG_FAIR_CGROUP_SCHED |
165 | struct cgroup_subsys_state css; | |
166 | #endif | |
29f59db3 SV |
167 | /* schedulable entities of this group on each cpu */ |
168 | struct sched_entity **se; | |
169 | /* runqueue "owned" by this group on each cpu */ | |
170 | struct cfs_rq **cfs_rq; | |
171 | unsigned long shares; | |
5cb350ba DG |
172 | /* spinlock to serialize modification to shares */ |
173 | spinlock_t lock; | |
ae8393e5 | 174 | struct rcu_head rcu; |
29f59db3 SV |
175 | }; |
176 | ||
177 | /* Default task group's sched entity on each cpu */ | |
178 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
179 | /* Default task group's cfs_rq on each cpu */ | |
180 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
181 | ||
9b5b7751 SV |
182 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
183 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 SV |
184 | |
185 | /* Default task group. | |
3a252015 | 186 | * Every task in system belong to this group at bootup. |
29f59db3 | 187 | */ |
4cf86d77 | 188 | struct task_group init_task_group = { |
3a252015 IM |
189 | .se = init_sched_entity_p, |
190 | .cfs_rq = init_cfs_rq_p, | |
191 | }; | |
9b5b7751 | 192 | |
24e377a8 | 193 | #ifdef CONFIG_FAIR_USER_SCHED |
3a252015 | 194 | # define INIT_TASK_GRP_LOAD 2*NICE_0_LOAD |
24e377a8 | 195 | #else |
3a252015 | 196 | # define INIT_TASK_GRP_LOAD NICE_0_LOAD |
24e377a8 SV |
197 | #endif |
198 | ||
4cf86d77 | 199 | static int init_task_group_load = INIT_TASK_GRP_LOAD; |
29f59db3 SV |
200 | |
201 | /* return group to which a task belongs */ | |
4cf86d77 | 202 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 203 | { |
4cf86d77 | 204 | struct task_group *tg; |
9b5b7751 | 205 | |
24e377a8 SV |
206 | #ifdef CONFIG_FAIR_USER_SCHED |
207 | tg = p->user->tg; | |
68318b8e SV |
208 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
209 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
210 | struct task_group, css); | |
24e377a8 | 211 | #else |
41a2d6cf | 212 | tg = &init_task_group; |
24e377a8 | 213 | #endif |
9b5b7751 | 214 | return tg; |
29f59db3 SV |
215 | } |
216 | ||
217 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
ce96b5ac | 218 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 219 | { |
ce96b5ac DA |
220 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
221 | p->se.parent = task_group(p)->se[cpu]; | |
29f59db3 SV |
222 | } |
223 | ||
224 | #else | |
225 | ||
ce96b5ac | 226 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { } |
29f59db3 SV |
227 | |
228 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
229 | ||
6aa645ea IM |
230 | /* CFS-related fields in a runqueue */ |
231 | struct cfs_rq { | |
232 | struct load_weight load; | |
233 | unsigned long nr_running; | |
234 | ||
6aa645ea | 235 | u64 exec_clock; |
e9acbff6 | 236 | u64 min_vruntime; |
6aa645ea IM |
237 | |
238 | struct rb_root tasks_timeline; | |
239 | struct rb_node *rb_leftmost; | |
240 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
241 | /* 'curr' points to currently running entity on this cfs_rq. |
242 | * It is set to NULL otherwise (i.e when none are currently running). | |
243 | */ | |
244 | struct sched_entity *curr; | |
ddc97297 PZ |
245 | |
246 | unsigned long nr_spread_over; | |
247 | ||
62160e3f | 248 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
249 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
250 | ||
41a2d6cf IM |
251 | /* |
252 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
253 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
254 | * (like users, containers etc.) | |
255 | * | |
256 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
257 | * list is used during load balance. | |
258 | */ | |
41a2d6cf IM |
259 | struct list_head leaf_cfs_rq_list; |
260 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
261 | #endif |
262 | }; | |
1da177e4 | 263 | |
6aa645ea IM |
264 | /* Real-Time classes' related field in a runqueue: */ |
265 | struct rt_rq { | |
266 | struct rt_prio_array active; | |
267 | int rt_load_balance_idx; | |
268 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
269 | }; | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * This is the main, per-CPU runqueue data structure. | |
273 | * | |
274 | * Locking rule: those places that want to lock multiple runqueues | |
275 | * (such as the load balancing or the thread migration code), lock | |
276 | * acquire operations must be ordered by ascending &runqueue. | |
277 | */ | |
70b97a7f | 278 | struct rq { |
d8016491 IM |
279 | /* runqueue lock: */ |
280 | spinlock_t lock; | |
1da177e4 LT |
281 | |
282 | /* | |
283 | * nr_running and cpu_load should be in the same cacheline because | |
284 | * remote CPUs use both these fields when doing load calculation. | |
285 | */ | |
286 | unsigned long nr_running; | |
6aa645ea IM |
287 | #define CPU_LOAD_IDX_MAX 5 |
288 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 289 | unsigned char idle_at_tick; |
46cb4b7c SS |
290 | #ifdef CONFIG_NO_HZ |
291 | unsigned char in_nohz_recently; | |
292 | #endif | |
d8016491 IM |
293 | /* capture load from *all* tasks on this cpu: */ |
294 | struct load_weight load; | |
6aa645ea IM |
295 | unsigned long nr_load_updates; |
296 | u64 nr_switches; | |
297 | ||
298 | struct cfs_rq cfs; | |
299 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
300 | /* list of leaf cfs_rq on this cpu: */ |
301 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 302 | #endif |
41a2d6cf | 303 | struct rt_rq rt; |
1da177e4 LT |
304 | |
305 | /* | |
306 | * This is part of a global counter where only the total sum | |
307 | * over all CPUs matters. A task can increase this counter on | |
308 | * one CPU and if it got migrated afterwards it may decrease | |
309 | * it on another CPU. Always updated under the runqueue lock: | |
310 | */ | |
311 | unsigned long nr_uninterruptible; | |
312 | ||
36c8b586 | 313 | struct task_struct *curr, *idle; |
c9819f45 | 314 | unsigned long next_balance; |
1da177e4 | 315 | struct mm_struct *prev_mm; |
6aa645ea | 316 | |
6aa645ea IM |
317 | u64 clock, prev_clock_raw; |
318 | s64 clock_max_delta; | |
319 | ||
320 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
321 | u64 idle_clock; |
322 | unsigned int clock_deep_idle_events; | |
529c7726 | 323 | u64 tick_timestamp; |
6aa645ea | 324 | |
1da177e4 LT |
325 | atomic_t nr_iowait; |
326 | ||
327 | #ifdef CONFIG_SMP | |
328 | struct sched_domain *sd; | |
329 | ||
330 | /* For active balancing */ | |
331 | int active_balance; | |
332 | int push_cpu; | |
d8016491 IM |
333 | /* cpu of this runqueue: */ |
334 | int cpu; | |
1da177e4 | 335 | |
36c8b586 | 336 | struct task_struct *migration_thread; |
1da177e4 LT |
337 | struct list_head migration_queue; |
338 | #endif | |
339 | ||
340 | #ifdef CONFIG_SCHEDSTATS | |
341 | /* latency stats */ | |
342 | struct sched_info rq_sched_info; | |
343 | ||
344 | /* sys_sched_yield() stats */ | |
480b9434 KC |
345 | unsigned int yld_exp_empty; |
346 | unsigned int yld_act_empty; | |
347 | unsigned int yld_both_empty; | |
348 | unsigned int yld_count; | |
1da177e4 LT |
349 | |
350 | /* schedule() stats */ | |
480b9434 KC |
351 | unsigned int sched_switch; |
352 | unsigned int sched_count; | |
353 | unsigned int sched_goidle; | |
1da177e4 LT |
354 | |
355 | /* try_to_wake_up() stats */ | |
480b9434 KC |
356 | unsigned int ttwu_count; |
357 | unsigned int ttwu_local; | |
b8efb561 IM |
358 | |
359 | /* BKL stats */ | |
480b9434 | 360 | unsigned int bkl_count; |
1da177e4 | 361 | #endif |
fcb99371 | 362 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
363 | }; |
364 | ||
f34e3b61 | 365 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
5be9361c | 366 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
1da177e4 | 367 | |
dd41f596 IM |
368 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
369 | { | |
370 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
371 | } | |
372 | ||
0a2966b4 CL |
373 | static inline int cpu_of(struct rq *rq) |
374 | { | |
375 | #ifdef CONFIG_SMP | |
376 | return rq->cpu; | |
377 | #else | |
378 | return 0; | |
379 | #endif | |
380 | } | |
381 | ||
20d315d4 | 382 | /* |
b04a0f4c IM |
383 | * Update the per-runqueue clock, as finegrained as the platform can give |
384 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 385 | */ |
b04a0f4c | 386 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
387 | { |
388 | u64 prev_raw = rq->prev_clock_raw; | |
389 | u64 now = sched_clock(); | |
390 | s64 delta = now - prev_raw; | |
391 | u64 clock = rq->clock; | |
392 | ||
b04a0f4c IM |
393 | #ifdef CONFIG_SCHED_DEBUG |
394 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
395 | #endif | |
20d315d4 IM |
396 | /* |
397 | * Protect against sched_clock() occasionally going backwards: | |
398 | */ | |
399 | if (unlikely(delta < 0)) { | |
400 | clock++; | |
401 | rq->clock_warps++; | |
402 | } else { | |
403 | /* | |
404 | * Catch too large forward jumps too: | |
405 | */ | |
529c7726 IM |
406 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
407 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
408 | clock = rq->tick_timestamp + TICK_NSEC; | |
409 | else | |
410 | clock++; | |
20d315d4 IM |
411 | rq->clock_overflows++; |
412 | } else { | |
413 | if (unlikely(delta > rq->clock_max_delta)) | |
414 | rq->clock_max_delta = delta; | |
415 | clock += delta; | |
416 | } | |
417 | } | |
418 | ||
419 | rq->prev_clock_raw = now; | |
420 | rq->clock = clock; | |
b04a0f4c | 421 | } |
20d315d4 | 422 | |
b04a0f4c IM |
423 | static void update_rq_clock(struct rq *rq) |
424 | { | |
425 | if (likely(smp_processor_id() == cpu_of(rq))) | |
426 | __update_rq_clock(rq); | |
20d315d4 IM |
427 | } |
428 | ||
674311d5 NP |
429 | /* |
430 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 431 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
432 | * |
433 | * The domain tree of any CPU may only be accessed from within | |
434 | * preempt-disabled sections. | |
435 | */ | |
48f24c4d IM |
436 | #define for_each_domain(cpu, __sd) \ |
437 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
438 | |
439 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
440 | #define this_rq() (&__get_cpu_var(runqueues)) | |
441 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
442 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
443 | ||
bf5c91ba IM |
444 | /* |
445 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
446 | */ | |
447 | #ifdef CONFIG_SCHED_DEBUG | |
448 | # define const_debug __read_mostly | |
449 | #else | |
450 | # define const_debug static const | |
451 | #endif | |
452 | ||
453 | /* | |
454 | * Debugging: various feature bits | |
455 | */ | |
456 | enum { | |
bbdba7c0 | 457 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
458 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
459 | SCHED_FEAT_START_DEBIT = 4, | |
41a2d6cf IM |
460 | SCHED_FEAT_TREE_AVG = 8, |
461 | SCHED_FEAT_APPROX_AVG = 16, | |
bf5c91ba IM |
462 | }; |
463 | ||
464 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 465 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 466 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
467 | SCHED_FEAT_START_DEBIT * 1 | |
468 | SCHED_FEAT_TREE_AVG * 0 | | |
9612633a | 469 | SCHED_FEAT_APPROX_AVG * 0; |
bf5c91ba IM |
470 | |
471 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
472 | ||
b82d9fdd PZ |
473 | /* |
474 | * Number of tasks to iterate in a single balance run. | |
475 | * Limited because this is done with IRQs disabled. | |
476 | */ | |
477 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
478 | ||
e436d800 IM |
479 | /* |
480 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
481 | * clock constructed from sched_clock(): | |
482 | */ | |
483 | unsigned long long cpu_clock(int cpu) | |
484 | { | |
e436d800 IM |
485 | unsigned long long now; |
486 | unsigned long flags; | |
b04a0f4c | 487 | struct rq *rq; |
e436d800 | 488 | |
2cd4d0ea | 489 | local_irq_save(flags); |
b04a0f4c IM |
490 | rq = cpu_rq(cpu); |
491 | update_rq_clock(rq); | |
492 | now = rq->clock; | |
2cd4d0ea | 493 | local_irq_restore(flags); |
e436d800 IM |
494 | |
495 | return now; | |
496 | } | |
a58f6f25 | 497 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 498 | |
1da177e4 | 499 | #ifndef prepare_arch_switch |
4866cde0 NP |
500 | # define prepare_arch_switch(next) do { } while (0) |
501 | #endif | |
502 | #ifndef finish_arch_switch | |
503 | # define finish_arch_switch(prev) do { } while (0) | |
504 | #endif | |
505 | ||
506 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
70b97a7f | 507 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
508 | { |
509 | return rq->curr == p; | |
510 | } | |
511 | ||
70b97a7f | 512 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
513 | { |
514 | } | |
515 | ||
70b97a7f | 516 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 517 | { |
da04c035 IM |
518 | #ifdef CONFIG_DEBUG_SPINLOCK |
519 | /* this is a valid case when another task releases the spinlock */ | |
520 | rq->lock.owner = current; | |
521 | #endif | |
8a25d5de IM |
522 | /* |
523 | * If we are tracking spinlock dependencies then we have to | |
524 | * fix up the runqueue lock - which gets 'carried over' from | |
525 | * prev into current: | |
526 | */ | |
527 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
528 | ||
4866cde0 NP |
529 | spin_unlock_irq(&rq->lock); |
530 | } | |
531 | ||
532 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 533 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
534 | { |
535 | #ifdef CONFIG_SMP | |
536 | return p->oncpu; | |
537 | #else | |
538 | return rq->curr == p; | |
539 | #endif | |
540 | } | |
541 | ||
70b97a7f | 542 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
543 | { |
544 | #ifdef CONFIG_SMP | |
545 | /* | |
546 | * We can optimise this out completely for !SMP, because the | |
547 | * SMP rebalancing from interrupt is the only thing that cares | |
548 | * here. | |
549 | */ | |
550 | next->oncpu = 1; | |
551 | #endif | |
552 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
553 | spin_unlock_irq(&rq->lock); | |
554 | #else | |
555 | spin_unlock(&rq->lock); | |
556 | #endif | |
557 | } | |
558 | ||
70b97a7f | 559 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
560 | { |
561 | #ifdef CONFIG_SMP | |
562 | /* | |
563 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
564 | * We must ensure this doesn't happen until the switch is completely | |
565 | * finished. | |
566 | */ | |
567 | smp_wmb(); | |
568 | prev->oncpu = 0; | |
569 | #endif | |
570 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
571 | local_irq_enable(); | |
1da177e4 | 572 | #endif |
4866cde0 NP |
573 | } |
574 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 575 | |
b29739f9 IM |
576 | /* |
577 | * __task_rq_lock - lock the runqueue a given task resides on. | |
578 | * Must be called interrupts disabled. | |
579 | */ | |
70b97a7f | 580 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
581 | __acquires(rq->lock) |
582 | { | |
3a5c359a AK |
583 | for (;;) { |
584 | struct rq *rq = task_rq(p); | |
585 | spin_lock(&rq->lock); | |
586 | if (likely(rq == task_rq(p))) | |
587 | return rq; | |
b29739f9 | 588 | spin_unlock(&rq->lock); |
b29739f9 | 589 | } |
b29739f9 IM |
590 | } |
591 | ||
1da177e4 LT |
592 | /* |
593 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 594 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
595 | * explicitly disabling preemption. |
596 | */ | |
70b97a7f | 597 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
598 | __acquires(rq->lock) |
599 | { | |
70b97a7f | 600 | struct rq *rq; |
1da177e4 | 601 | |
3a5c359a AK |
602 | for (;;) { |
603 | local_irq_save(*flags); | |
604 | rq = task_rq(p); | |
605 | spin_lock(&rq->lock); | |
606 | if (likely(rq == task_rq(p))) | |
607 | return rq; | |
1da177e4 | 608 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 609 | } |
1da177e4 LT |
610 | } |
611 | ||
a9957449 | 612 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
613 | __releases(rq->lock) |
614 | { | |
615 | spin_unlock(&rq->lock); | |
616 | } | |
617 | ||
70b97a7f | 618 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
619 | __releases(rq->lock) |
620 | { | |
621 | spin_unlock_irqrestore(&rq->lock, *flags); | |
622 | } | |
623 | ||
1da177e4 | 624 | /* |
cc2a73b5 | 625 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 626 | */ |
a9957449 | 627 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
628 | __acquires(rq->lock) |
629 | { | |
70b97a7f | 630 | struct rq *rq; |
1da177e4 LT |
631 | |
632 | local_irq_disable(); | |
633 | rq = this_rq(); | |
634 | spin_lock(&rq->lock); | |
635 | ||
636 | return rq; | |
637 | } | |
638 | ||
1b9f19c2 | 639 | /* |
2aa44d05 | 640 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 641 | */ |
2aa44d05 | 642 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 643 | { |
2aa44d05 IM |
644 | struct rq *rq = cpu_rq(smp_processor_id()); |
645 | ||
646 | spin_lock(&rq->lock); | |
647 | __update_rq_clock(rq); | |
648 | spin_unlock(&rq->lock); | |
649 | rq->clock_deep_idle_events++; | |
650 | } | |
651 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
652 | ||
653 | /* | |
654 | * We just idled delta nanoseconds (called with irqs disabled): | |
655 | */ | |
656 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
657 | { | |
658 | struct rq *rq = cpu_rq(smp_processor_id()); | |
659 | u64 now = sched_clock(); | |
1b9f19c2 | 660 | |
2aa44d05 IM |
661 | rq->idle_clock += delta_ns; |
662 | /* | |
663 | * Override the previous timestamp and ignore all | |
664 | * sched_clock() deltas that occured while we idled, | |
665 | * and use the PM-provided delta_ns to advance the | |
666 | * rq clock: | |
667 | */ | |
668 | spin_lock(&rq->lock); | |
669 | rq->prev_clock_raw = now; | |
670 | rq->clock += delta_ns; | |
671 | spin_unlock(&rq->lock); | |
1b9f19c2 | 672 | } |
2aa44d05 | 673 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 674 | |
c24d20db IM |
675 | /* |
676 | * resched_task - mark a task 'to be rescheduled now'. | |
677 | * | |
678 | * On UP this means the setting of the need_resched flag, on SMP it | |
679 | * might also involve a cross-CPU call to trigger the scheduler on | |
680 | * the target CPU. | |
681 | */ | |
682 | #ifdef CONFIG_SMP | |
683 | ||
684 | #ifndef tsk_is_polling | |
685 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
686 | #endif | |
687 | ||
688 | static void resched_task(struct task_struct *p) | |
689 | { | |
690 | int cpu; | |
691 | ||
692 | assert_spin_locked(&task_rq(p)->lock); | |
693 | ||
694 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
695 | return; | |
696 | ||
697 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
698 | ||
699 | cpu = task_cpu(p); | |
700 | if (cpu == smp_processor_id()) | |
701 | return; | |
702 | ||
703 | /* NEED_RESCHED must be visible before we test polling */ | |
704 | smp_mb(); | |
705 | if (!tsk_is_polling(p)) | |
706 | smp_send_reschedule(cpu); | |
707 | } | |
708 | ||
709 | static void resched_cpu(int cpu) | |
710 | { | |
711 | struct rq *rq = cpu_rq(cpu); | |
712 | unsigned long flags; | |
713 | ||
714 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
715 | return; | |
716 | resched_task(cpu_curr(cpu)); | |
717 | spin_unlock_irqrestore(&rq->lock, flags); | |
718 | } | |
719 | #else | |
720 | static inline void resched_task(struct task_struct *p) | |
721 | { | |
722 | assert_spin_locked(&task_rq(p)->lock); | |
723 | set_tsk_need_resched(p); | |
724 | } | |
725 | #endif | |
726 | ||
45bf76df IM |
727 | #if BITS_PER_LONG == 32 |
728 | # define WMULT_CONST (~0UL) | |
729 | #else | |
730 | # define WMULT_CONST (1UL << 32) | |
731 | #endif | |
732 | ||
733 | #define WMULT_SHIFT 32 | |
734 | ||
194081eb IM |
735 | /* |
736 | * Shift right and round: | |
737 | */ | |
cf2ab469 | 738 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 739 | |
cb1c4fc9 | 740 | static unsigned long |
45bf76df IM |
741 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
742 | struct load_weight *lw) | |
743 | { | |
744 | u64 tmp; | |
745 | ||
746 | if (unlikely(!lw->inv_weight)) | |
194081eb | 747 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
748 | |
749 | tmp = (u64)delta_exec * weight; | |
750 | /* | |
751 | * Check whether we'd overflow the 64-bit multiplication: | |
752 | */ | |
194081eb | 753 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 754 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
755 | WMULT_SHIFT/2); |
756 | else | |
cf2ab469 | 757 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 758 | |
ecf691da | 759 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
760 | } |
761 | ||
762 | static inline unsigned long | |
763 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
764 | { | |
765 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
766 | } | |
767 | ||
1091985b | 768 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
769 | { |
770 | lw->weight += inc; | |
45bf76df IM |
771 | } |
772 | ||
1091985b | 773 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
774 | { |
775 | lw->weight -= dec; | |
45bf76df IM |
776 | } |
777 | ||
2dd73a4f PW |
778 | /* |
779 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
780 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
781 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 782 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
783 | * scaled version of the new time slice allocation that they receive on time |
784 | * slice expiry etc. | |
785 | */ | |
786 | ||
dd41f596 IM |
787 | #define WEIGHT_IDLEPRIO 2 |
788 | #define WMULT_IDLEPRIO (1 << 31) | |
789 | ||
790 | /* | |
791 | * Nice levels are multiplicative, with a gentle 10% change for every | |
792 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
793 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
794 | * that remained on nice 0. | |
795 | * | |
796 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
797 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
798 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
799 | * If a task goes up by ~10% and another task goes down by ~10% then | |
800 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
801 | */ |
802 | static const int prio_to_weight[40] = { | |
254753dc IM |
803 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
804 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
805 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
806 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
807 | /* 0 */ 1024, 820, 655, 526, 423, | |
808 | /* 5 */ 335, 272, 215, 172, 137, | |
809 | /* 10 */ 110, 87, 70, 56, 45, | |
810 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
811 | }; |
812 | ||
5714d2de IM |
813 | /* |
814 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
815 | * | |
816 | * In cases where the weight does not change often, we can use the | |
817 | * precalculated inverse to speed up arithmetics by turning divisions | |
818 | * into multiplications: | |
819 | */ | |
dd41f596 | 820 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
821 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
822 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
823 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
824 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
825 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
826 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
827 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
828 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 829 | }; |
2dd73a4f | 830 | |
dd41f596 IM |
831 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
832 | ||
833 | /* | |
834 | * runqueue iterator, to support SMP load-balancing between different | |
835 | * scheduling classes, without having to expose their internal data | |
836 | * structures to the load-balancing proper: | |
837 | */ | |
838 | struct rq_iterator { | |
839 | void *arg; | |
840 | struct task_struct *(*start)(void *); | |
841 | struct task_struct *(*next)(void *); | |
842 | }; | |
843 | ||
e1d1484f PW |
844 | #ifdef CONFIG_SMP |
845 | static unsigned long | |
846 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
847 | unsigned long max_load_move, struct sched_domain *sd, | |
848 | enum cpu_idle_type idle, int *all_pinned, | |
849 | int *this_best_prio, struct rq_iterator *iterator); | |
850 | ||
851 | static int | |
852 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
853 | struct sched_domain *sd, enum cpu_idle_type idle, | |
854 | struct rq_iterator *iterator); | |
e1d1484f | 855 | #endif |
dd41f596 | 856 | |
d842de87 SV |
857 | #ifdef CONFIG_CGROUP_CPUACCT |
858 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
859 | #else | |
860 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
861 | #endif | |
862 | ||
dd41f596 | 863 | #include "sched_stats.h" |
dd41f596 | 864 | #include "sched_idletask.c" |
5522d5d5 IM |
865 | #include "sched_fair.c" |
866 | #include "sched_rt.c" | |
dd41f596 IM |
867 | #ifdef CONFIG_SCHED_DEBUG |
868 | # include "sched_debug.c" | |
869 | #endif | |
870 | ||
871 | #define sched_class_highest (&rt_sched_class) | |
872 | ||
9c217245 IM |
873 | /* |
874 | * Update delta_exec, delta_fair fields for rq. | |
875 | * | |
876 | * delta_fair clock advances at a rate inversely proportional to | |
495eca49 | 877 | * total load (rq->load.weight) on the runqueue, while |
9c217245 IM |
878 | * delta_exec advances at the same rate as wall-clock (provided |
879 | * cpu is not idle). | |
880 | * | |
881 | * delta_exec / delta_fair is a measure of the (smoothened) load on this | |
882 | * runqueue over any given interval. This (smoothened) load is used | |
883 | * during load balance. | |
884 | * | |
495eca49 | 885 | * This function is called /before/ updating rq->load |
9c217245 IM |
886 | * and when switching tasks. |
887 | */ | |
29b4b623 | 888 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 889 | { |
495eca49 | 890 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
891 | } |
892 | ||
79b5dddf | 893 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 894 | { |
495eca49 | 895 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
896 | } |
897 | ||
e5fa2237 | 898 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
899 | { |
900 | rq->nr_running++; | |
29b4b623 | 901 | inc_load(rq, p); |
9c217245 IM |
902 | } |
903 | ||
db53181e | 904 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
905 | { |
906 | rq->nr_running--; | |
79b5dddf | 907 | dec_load(rq, p); |
9c217245 IM |
908 | } |
909 | ||
45bf76df IM |
910 | static void set_load_weight(struct task_struct *p) |
911 | { | |
912 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
913 | p->se.load.weight = prio_to_weight[0] * 2; |
914 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
915 | return; | |
916 | } | |
45bf76df | 917 | |
dd41f596 IM |
918 | /* |
919 | * SCHED_IDLE tasks get minimal weight: | |
920 | */ | |
921 | if (p->policy == SCHED_IDLE) { | |
922 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
923 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
924 | return; | |
925 | } | |
71f8bd46 | 926 | |
dd41f596 IM |
927 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
928 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
929 | } |
930 | ||
8159f87e | 931 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 932 | { |
dd41f596 | 933 | sched_info_queued(p); |
fd390f6a | 934 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 935 | p->se.on_rq = 1; |
71f8bd46 IM |
936 | } |
937 | ||
69be72c1 | 938 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 939 | { |
f02231e5 | 940 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 941 | p->se.on_rq = 0; |
71f8bd46 IM |
942 | } |
943 | ||
14531189 | 944 | /* |
dd41f596 | 945 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 946 | */ |
14531189 IM |
947 | static inline int __normal_prio(struct task_struct *p) |
948 | { | |
dd41f596 | 949 | return p->static_prio; |
14531189 IM |
950 | } |
951 | ||
b29739f9 IM |
952 | /* |
953 | * Calculate the expected normal priority: i.e. priority | |
954 | * without taking RT-inheritance into account. Might be | |
955 | * boosted by interactivity modifiers. Changes upon fork, | |
956 | * setprio syscalls, and whenever the interactivity | |
957 | * estimator recalculates. | |
958 | */ | |
36c8b586 | 959 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
960 | { |
961 | int prio; | |
962 | ||
e05606d3 | 963 | if (task_has_rt_policy(p)) |
b29739f9 IM |
964 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
965 | else | |
966 | prio = __normal_prio(p); | |
967 | return prio; | |
968 | } | |
969 | ||
970 | /* | |
971 | * Calculate the current priority, i.e. the priority | |
972 | * taken into account by the scheduler. This value might | |
973 | * be boosted by RT tasks, or might be boosted by | |
974 | * interactivity modifiers. Will be RT if the task got | |
975 | * RT-boosted. If not then it returns p->normal_prio. | |
976 | */ | |
36c8b586 | 977 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
978 | { |
979 | p->normal_prio = normal_prio(p); | |
980 | /* | |
981 | * If we are RT tasks or we were boosted to RT priority, | |
982 | * keep the priority unchanged. Otherwise, update priority | |
983 | * to the normal priority: | |
984 | */ | |
985 | if (!rt_prio(p->prio)) | |
986 | return p->normal_prio; | |
987 | return p->prio; | |
988 | } | |
989 | ||
1da177e4 | 990 | /* |
dd41f596 | 991 | * activate_task - move a task to the runqueue. |
1da177e4 | 992 | */ |
dd41f596 | 993 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 994 | { |
d9514f6c | 995 | if (task_contributes_to_load(p)) |
dd41f596 | 996 | rq->nr_uninterruptible--; |
1da177e4 | 997 | |
8159f87e | 998 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 999 | inc_nr_running(p, rq); |
1da177e4 LT |
1000 | } |
1001 | ||
1da177e4 LT |
1002 | /* |
1003 | * deactivate_task - remove a task from the runqueue. | |
1004 | */ | |
2e1cb74a | 1005 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1006 | { |
d9514f6c | 1007 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1008 | rq->nr_uninterruptible++; |
1009 | ||
69be72c1 | 1010 | dequeue_task(rq, p, sleep); |
db53181e | 1011 | dec_nr_running(p, rq); |
1da177e4 LT |
1012 | } |
1013 | ||
1da177e4 LT |
1014 | /** |
1015 | * task_curr - is this task currently executing on a CPU? | |
1016 | * @p: the task in question. | |
1017 | */ | |
36c8b586 | 1018 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1019 | { |
1020 | return cpu_curr(task_cpu(p)) == p; | |
1021 | } | |
1022 | ||
2dd73a4f PW |
1023 | /* Used instead of source_load when we know the type == 0 */ |
1024 | unsigned long weighted_cpuload(const int cpu) | |
1025 | { | |
495eca49 | 1026 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1027 | } |
1028 | ||
1029 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1030 | { | |
ce96b5ac | 1031 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1032 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1033 | /* |
1034 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1035 | * successfuly executed on another CPU. We must ensure that updates of | |
1036 | * per-task data have been completed by this moment. | |
1037 | */ | |
1038 | smp_wmb(); | |
dd41f596 | 1039 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1040 | #endif |
2dd73a4f PW |
1041 | } |
1042 | ||
1da177e4 | 1043 | #ifdef CONFIG_SMP |
c65cc870 | 1044 | |
cc367732 IM |
1045 | /* |
1046 | * Is this task likely cache-hot: | |
1047 | */ | |
1048 | static inline int | |
1049 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1050 | { | |
1051 | s64 delta; | |
1052 | ||
1053 | if (p->sched_class != &fair_sched_class) | |
1054 | return 0; | |
1055 | ||
6bc1665b IM |
1056 | if (sysctl_sched_migration_cost == -1) |
1057 | return 1; | |
1058 | if (sysctl_sched_migration_cost == 0) | |
1059 | return 0; | |
1060 | ||
cc367732 IM |
1061 | delta = now - p->se.exec_start; |
1062 | ||
1063 | return delta < (s64)sysctl_sched_migration_cost; | |
1064 | } | |
1065 | ||
1066 | ||
dd41f596 | 1067 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1068 | { |
dd41f596 IM |
1069 | int old_cpu = task_cpu(p); |
1070 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1071 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1072 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1073 | u64 clock_offset; |
dd41f596 IM |
1074 | |
1075 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1076 | |
1077 | #ifdef CONFIG_SCHEDSTATS | |
1078 | if (p->se.wait_start) | |
1079 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1080 | if (p->se.sleep_start) |
1081 | p->se.sleep_start -= clock_offset; | |
1082 | if (p->se.block_start) | |
1083 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1084 | if (old_cpu != new_cpu) { |
1085 | schedstat_inc(p, se.nr_migrations); | |
1086 | if (task_hot(p, old_rq->clock, NULL)) | |
1087 | schedstat_inc(p, se.nr_forced2_migrations); | |
1088 | } | |
6cfb0d5d | 1089 | #endif |
2830cf8c SV |
1090 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1091 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1092 | |
1093 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1094 | } |
1095 | ||
70b97a7f | 1096 | struct migration_req { |
1da177e4 | 1097 | struct list_head list; |
1da177e4 | 1098 | |
36c8b586 | 1099 | struct task_struct *task; |
1da177e4 LT |
1100 | int dest_cpu; |
1101 | ||
1da177e4 | 1102 | struct completion done; |
70b97a7f | 1103 | }; |
1da177e4 LT |
1104 | |
1105 | /* | |
1106 | * The task's runqueue lock must be held. | |
1107 | * Returns true if you have to wait for migration thread. | |
1108 | */ | |
36c8b586 | 1109 | static int |
70b97a7f | 1110 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1111 | { |
70b97a7f | 1112 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1113 | |
1114 | /* | |
1115 | * If the task is not on a runqueue (and not running), then | |
1116 | * it is sufficient to simply update the task's cpu field. | |
1117 | */ | |
dd41f596 | 1118 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1119 | set_task_cpu(p, dest_cpu); |
1120 | return 0; | |
1121 | } | |
1122 | ||
1123 | init_completion(&req->done); | |
1da177e4 LT |
1124 | req->task = p; |
1125 | req->dest_cpu = dest_cpu; | |
1126 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1127 | |
1da177e4 LT |
1128 | return 1; |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * wait_task_inactive - wait for a thread to unschedule. | |
1133 | * | |
1134 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1135 | * else this function might spin for a *long* time. This function can't | |
1136 | * be called with interrupts off, or it may introduce deadlock with | |
1137 | * smp_call_function() if an IPI is sent by the same process we are | |
1138 | * waiting to become inactive. | |
1139 | */ | |
36c8b586 | 1140 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1141 | { |
1142 | unsigned long flags; | |
dd41f596 | 1143 | int running, on_rq; |
70b97a7f | 1144 | struct rq *rq; |
1da177e4 | 1145 | |
3a5c359a AK |
1146 | for (;;) { |
1147 | /* | |
1148 | * We do the initial early heuristics without holding | |
1149 | * any task-queue locks at all. We'll only try to get | |
1150 | * the runqueue lock when things look like they will | |
1151 | * work out! | |
1152 | */ | |
1153 | rq = task_rq(p); | |
fa490cfd | 1154 | |
3a5c359a AK |
1155 | /* |
1156 | * If the task is actively running on another CPU | |
1157 | * still, just relax and busy-wait without holding | |
1158 | * any locks. | |
1159 | * | |
1160 | * NOTE! Since we don't hold any locks, it's not | |
1161 | * even sure that "rq" stays as the right runqueue! | |
1162 | * But we don't care, since "task_running()" will | |
1163 | * return false if the runqueue has changed and p | |
1164 | * is actually now running somewhere else! | |
1165 | */ | |
1166 | while (task_running(rq, p)) | |
1167 | cpu_relax(); | |
fa490cfd | 1168 | |
3a5c359a AK |
1169 | /* |
1170 | * Ok, time to look more closely! We need the rq | |
1171 | * lock now, to be *sure*. If we're wrong, we'll | |
1172 | * just go back and repeat. | |
1173 | */ | |
1174 | rq = task_rq_lock(p, &flags); | |
1175 | running = task_running(rq, p); | |
1176 | on_rq = p->se.on_rq; | |
1177 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1178 | |
3a5c359a AK |
1179 | /* |
1180 | * Was it really running after all now that we | |
1181 | * checked with the proper locks actually held? | |
1182 | * | |
1183 | * Oops. Go back and try again.. | |
1184 | */ | |
1185 | if (unlikely(running)) { | |
1186 | cpu_relax(); | |
1187 | continue; | |
1188 | } | |
fa490cfd | 1189 | |
3a5c359a AK |
1190 | /* |
1191 | * It's not enough that it's not actively running, | |
1192 | * it must be off the runqueue _entirely_, and not | |
1193 | * preempted! | |
1194 | * | |
1195 | * So if it wa still runnable (but just not actively | |
1196 | * running right now), it's preempted, and we should | |
1197 | * yield - it could be a while. | |
1198 | */ | |
1199 | if (unlikely(on_rq)) { | |
1200 | schedule_timeout_uninterruptible(1); | |
1201 | continue; | |
1202 | } | |
fa490cfd | 1203 | |
3a5c359a AK |
1204 | /* |
1205 | * Ahh, all good. It wasn't running, and it wasn't | |
1206 | * runnable, which means that it will never become | |
1207 | * running in the future either. We're all done! | |
1208 | */ | |
1209 | break; | |
1210 | } | |
1da177e4 LT |
1211 | } |
1212 | ||
1213 | /*** | |
1214 | * kick_process - kick a running thread to enter/exit the kernel | |
1215 | * @p: the to-be-kicked thread | |
1216 | * | |
1217 | * Cause a process which is running on another CPU to enter | |
1218 | * kernel-mode, without any delay. (to get signals handled.) | |
1219 | * | |
1220 | * NOTE: this function doesnt have to take the runqueue lock, | |
1221 | * because all it wants to ensure is that the remote task enters | |
1222 | * the kernel. If the IPI races and the task has been migrated | |
1223 | * to another CPU then no harm is done and the purpose has been | |
1224 | * achieved as well. | |
1225 | */ | |
36c8b586 | 1226 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1227 | { |
1228 | int cpu; | |
1229 | ||
1230 | preempt_disable(); | |
1231 | cpu = task_cpu(p); | |
1232 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1233 | smp_send_reschedule(cpu); | |
1234 | preempt_enable(); | |
1235 | } | |
1236 | ||
1237 | /* | |
2dd73a4f PW |
1238 | * Return a low guess at the load of a migration-source cpu weighted |
1239 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1240 | * |
1241 | * We want to under-estimate the load of migration sources, to | |
1242 | * balance conservatively. | |
1243 | */ | |
a9957449 | 1244 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1245 | { |
70b97a7f | 1246 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1247 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1248 | |
3b0bd9bc | 1249 | if (type == 0) |
dd41f596 | 1250 | return total; |
b910472d | 1251 | |
dd41f596 | 1252 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | /* | |
2dd73a4f PW |
1256 | * Return a high guess at the load of a migration-target cpu weighted |
1257 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1258 | */ |
a9957449 | 1259 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1260 | { |
70b97a7f | 1261 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1262 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1263 | |
7897986b | 1264 | if (type == 0) |
dd41f596 | 1265 | return total; |
3b0bd9bc | 1266 | |
dd41f596 | 1267 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1268 | } |
1269 | ||
1270 | /* | |
1271 | * Return the average load per task on the cpu's run queue | |
1272 | */ | |
1273 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1274 | { | |
70b97a7f | 1275 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1276 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1277 | unsigned long n = rq->nr_running; |
1278 | ||
dd41f596 | 1279 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1280 | } |
1281 | ||
147cbb4b NP |
1282 | /* |
1283 | * find_idlest_group finds and returns the least busy CPU group within the | |
1284 | * domain. | |
1285 | */ | |
1286 | static struct sched_group * | |
1287 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1288 | { | |
1289 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1290 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1291 | int load_idx = sd->forkexec_idx; | |
1292 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1293 | ||
1294 | do { | |
1295 | unsigned long load, avg_load; | |
1296 | int local_group; | |
1297 | int i; | |
1298 | ||
da5a5522 BD |
1299 | /* Skip over this group if it has no CPUs allowed */ |
1300 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1301 | continue; |
da5a5522 | 1302 | |
147cbb4b | 1303 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1304 | |
1305 | /* Tally up the load of all CPUs in the group */ | |
1306 | avg_load = 0; | |
1307 | ||
1308 | for_each_cpu_mask(i, group->cpumask) { | |
1309 | /* Bias balancing toward cpus of our domain */ | |
1310 | if (local_group) | |
1311 | load = source_load(i, load_idx); | |
1312 | else | |
1313 | load = target_load(i, load_idx); | |
1314 | ||
1315 | avg_load += load; | |
1316 | } | |
1317 | ||
1318 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1319 | avg_load = sg_div_cpu_power(group, |
1320 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1321 | |
1322 | if (local_group) { | |
1323 | this_load = avg_load; | |
1324 | this = group; | |
1325 | } else if (avg_load < min_load) { | |
1326 | min_load = avg_load; | |
1327 | idlest = group; | |
1328 | } | |
3a5c359a | 1329 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1330 | |
1331 | if (!idlest || 100*this_load < imbalance*min_load) | |
1332 | return NULL; | |
1333 | return idlest; | |
1334 | } | |
1335 | ||
1336 | /* | |
0feaece9 | 1337 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1338 | */ |
95cdf3b7 IM |
1339 | static int |
1340 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1341 | { |
da5a5522 | 1342 | cpumask_t tmp; |
147cbb4b NP |
1343 | unsigned long load, min_load = ULONG_MAX; |
1344 | int idlest = -1; | |
1345 | int i; | |
1346 | ||
da5a5522 BD |
1347 | /* Traverse only the allowed CPUs */ |
1348 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1349 | ||
1350 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1351 | load = weighted_cpuload(i); |
147cbb4b NP |
1352 | |
1353 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1354 | min_load = load; | |
1355 | idlest = i; | |
1356 | } | |
1357 | } | |
1358 | ||
1359 | return idlest; | |
1360 | } | |
1361 | ||
476d139c NP |
1362 | /* |
1363 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1364 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1365 | * SD_BALANCE_EXEC. | |
1366 | * | |
1367 | * Balance, ie. select the least loaded group. | |
1368 | * | |
1369 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1370 | * | |
1371 | * preempt must be disabled. | |
1372 | */ | |
1373 | static int sched_balance_self(int cpu, int flag) | |
1374 | { | |
1375 | struct task_struct *t = current; | |
1376 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1377 | |
c96d145e | 1378 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1379 | /* |
1380 | * If power savings logic is enabled for a domain, stop there. | |
1381 | */ | |
5c45bf27 SS |
1382 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1383 | break; | |
476d139c NP |
1384 | if (tmp->flags & flag) |
1385 | sd = tmp; | |
c96d145e | 1386 | } |
476d139c NP |
1387 | |
1388 | while (sd) { | |
1389 | cpumask_t span; | |
1390 | struct sched_group *group; | |
1a848870 SS |
1391 | int new_cpu, weight; |
1392 | ||
1393 | if (!(sd->flags & flag)) { | |
1394 | sd = sd->child; | |
1395 | continue; | |
1396 | } | |
476d139c NP |
1397 | |
1398 | span = sd->span; | |
1399 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1400 | if (!group) { |
1401 | sd = sd->child; | |
1402 | continue; | |
1403 | } | |
476d139c | 1404 | |
da5a5522 | 1405 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1406 | if (new_cpu == -1 || new_cpu == cpu) { |
1407 | /* Now try balancing at a lower domain level of cpu */ | |
1408 | sd = sd->child; | |
1409 | continue; | |
1410 | } | |
476d139c | 1411 | |
1a848870 | 1412 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1413 | cpu = new_cpu; |
476d139c NP |
1414 | sd = NULL; |
1415 | weight = cpus_weight(span); | |
1416 | for_each_domain(cpu, tmp) { | |
1417 | if (weight <= cpus_weight(tmp->span)) | |
1418 | break; | |
1419 | if (tmp->flags & flag) | |
1420 | sd = tmp; | |
1421 | } | |
1422 | /* while loop will break here if sd == NULL */ | |
1423 | } | |
1424 | ||
1425 | return cpu; | |
1426 | } | |
1427 | ||
1428 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1429 | |
1430 | /* | |
1431 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1432 | * not idle and an idle cpu is available. The span of cpus to | |
1433 | * search starts with cpus closest then further out as needed, | |
1434 | * so we always favor a closer, idle cpu. | |
1435 | * | |
1436 | * Returns the CPU we should wake onto. | |
1437 | */ | |
1438 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1439 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1440 | { |
1441 | cpumask_t tmp; | |
1442 | struct sched_domain *sd; | |
1443 | int i; | |
1444 | ||
4953198b SS |
1445 | /* |
1446 | * If it is idle, then it is the best cpu to run this task. | |
1447 | * | |
1448 | * This cpu is also the best, if it has more than one task already. | |
1449 | * Siblings must be also busy(in most cases) as they didn't already | |
1450 | * pickup the extra load from this cpu and hence we need not check | |
1451 | * sibling runqueue info. This will avoid the checks and cache miss | |
1452 | * penalities associated with that. | |
1453 | */ | |
1454 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1455 | return cpu; |
1456 | ||
1457 | for_each_domain(cpu, sd) { | |
1458 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1459 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1460 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1461 | if (idle_cpu(i)) { |
1462 | if (i != task_cpu(p)) { | |
1463 | schedstat_inc(p, | |
1464 | se.nr_wakeups_idle); | |
1465 | } | |
1da177e4 | 1466 | return i; |
cc367732 | 1467 | } |
1da177e4 | 1468 | } |
9761eea8 | 1469 | } else { |
e0f364f4 | 1470 | break; |
9761eea8 | 1471 | } |
1da177e4 LT |
1472 | } |
1473 | return cpu; | |
1474 | } | |
1475 | #else | |
36c8b586 | 1476 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1477 | { |
1478 | return cpu; | |
1479 | } | |
1480 | #endif | |
1481 | ||
1482 | /*** | |
1483 | * try_to_wake_up - wake up a thread | |
1484 | * @p: the to-be-woken-up thread | |
1485 | * @state: the mask of task states that can be woken | |
1486 | * @sync: do a synchronous wakeup? | |
1487 | * | |
1488 | * Put it on the run-queue if it's not already there. The "current" | |
1489 | * thread is always on the run-queue (except when the actual | |
1490 | * re-schedule is in progress), and as such you're allowed to do | |
1491 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1492 | * runnable without the overhead of this. | |
1493 | * | |
1494 | * returns failure only if the task is already active. | |
1495 | */ | |
36c8b586 | 1496 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1497 | { |
cc367732 | 1498 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1499 | unsigned long flags; |
1500 | long old_state; | |
70b97a7f | 1501 | struct rq *rq; |
1da177e4 | 1502 | #ifdef CONFIG_SMP |
7897986b | 1503 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1504 | unsigned long load, this_load; |
1da177e4 LT |
1505 | int new_cpu; |
1506 | #endif | |
1507 | ||
1508 | rq = task_rq_lock(p, &flags); | |
1509 | old_state = p->state; | |
1510 | if (!(old_state & state)) | |
1511 | goto out; | |
1512 | ||
dd41f596 | 1513 | if (p->se.on_rq) |
1da177e4 LT |
1514 | goto out_running; |
1515 | ||
1516 | cpu = task_cpu(p); | |
cc367732 | 1517 | orig_cpu = cpu; |
1da177e4 LT |
1518 | this_cpu = smp_processor_id(); |
1519 | ||
1520 | #ifdef CONFIG_SMP | |
1521 | if (unlikely(task_running(rq, p))) | |
1522 | goto out_activate; | |
1523 | ||
7897986b NP |
1524 | new_cpu = cpu; |
1525 | ||
2d72376b | 1526 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1527 | if (cpu == this_cpu) { |
1528 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1529 | goto out_set_cpu; |
1530 | } | |
1531 | ||
1532 | for_each_domain(this_cpu, sd) { | |
1533 | if (cpu_isset(cpu, sd->span)) { | |
1534 | schedstat_inc(sd, ttwu_wake_remote); | |
1535 | this_sd = sd; | |
1536 | break; | |
1da177e4 LT |
1537 | } |
1538 | } | |
1da177e4 | 1539 | |
7897986b | 1540 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1541 | goto out_set_cpu; |
1542 | ||
1da177e4 | 1543 | /* |
7897986b | 1544 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1545 | */ |
7897986b NP |
1546 | if (this_sd) { |
1547 | int idx = this_sd->wake_idx; | |
1548 | unsigned int imbalance; | |
1da177e4 | 1549 | |
a3f21bce NP |
1550 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1551 | ||
7897986b NP |
1552 | load = source_load(cpu, idx); |
1553 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1554 | |
7897986b NP |
1555 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1556 | ||
a3f21bce NP |
1557 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1558 | unsigned long tl = this_load; | |
33859f7f MOS |
1559 | unsigned long tl_per_task; |
1560 | ||
71e20f18 IM |
1561 | /* |
1562 | * Attract cache-cold tasks on sync wakeups: | |
1563 | */ | |
1564 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1565 | goto out_set_cpu; | |
1566 | ||
cc367732 | 1567 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1568 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1569 | |
1da177e4 | 1570 | /* |
a3f21bce NP |
1571 | * If sync wakeup then subtract the (maximum possible) |
1572 | * effect of the currently running task from the load | |
1573 | * of the current CPU: | |
1da177e4 | 1574 | */ |
a3f21bce | 1575 | if (sync) |
dd41f596 | 1576 | tl -= current->se.load.weight; |
a3f21bce NP |
1577 | |
1578 | if ((tl <= load && | |
2dd73a4f | 1579 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1580 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1581 | /* |
1582 | * This domain has SD_WAKE_AFFINE and | |
1583 | * p is cache cold in this domain, and | |
1584 | * there is no bad imbalance. | |
1585 | */ | |
1586 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1587 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1588 | goto out_set_cpu; |
1589 | } | |
1590 | } | |
1591 | ||
1592 | /* | |
1593 | * Start passive balancing when half the imbalance_pct | |
1594 | * limit is reached. | |
1595 | */ | |
1596 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1597 | if (imbalance*this_load <= 100*load) { | |
1598 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1599 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1600 | goto out_set_cpu; |
1601 | } | |
1da177e4 LT |
1602 | } |
1603 | } | |
1604 | ||
1605 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1606 | out_set_cpu: | |
1607 | new_cpu = wake_idle(new_cpu, p); | |
1608 | if (new_cpu != cpu) { | |
1609 | set_task_cpu(p, new_cpu); | |
1610 | task_rq_unlock(rq, &flags); | |
1611 | /* might preempt at this point */ | |
1612 | rq = task_rq_lock(p, &flags); | |
1613 | old_state = p->state; | |
1614 | if (!(old_state & state)) | |
1615 | goto out; | |
dd41f596 | 1616 | if (p->se.on_rq) |
1da177e4 LT |
1617 | goto out_running; |
1618 | ||
1619 | this_cpu = smp_processor_id(); | |
1620 | cpu = task_cpu(p); | |
1621 | } | |
1622 | ||
1623 | out_activate: | |
1624 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1625 | schedstat_inc(p, se.nr_wakeups); |
1626 | if (sync) | |
1627 | schedstat_inc(p, se.nr_wakeups_sync); | |
1628 | if (orig_cpu != cpu) | |
1629 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1630 | if (cpu == this_cpu) | |
1631 | schedstat_inc(p, se.nr_wakeups_local); | |
1632 | else | |
1633 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1634 | update_rq_clock(rq); |
dd41f596 | 1635 | activate_task(rq, p, 1); |
9c63d9c0 | 1636 | check_preempt_curr(rq, p); |
1da177e4 LT |
1637 | success = 1; |
1638 | ||
1639 | out_running: | |
1640 | p->state = TASK_RUNNING; | |
1641 | out: | |
1642 | task_rq_unlock(rq, &flags); | |
1643 | ||
1644 | return success; | |
1645 | } | |
1646 | ||
36c8b586 | 1647 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 | 1648 | { |
d9514f6c | 1649 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 1650 | } |
1da177e4 LT |
1651 | EXPORT_SYMBOL(wake_up_process); |
1652 | ||
36c8b586 | 1653 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1654 | { |
1655 | return try_to_wake_up(p, state, 0); | |
1656 | } | |
1657 | ||
1da177e4 LT |
1658 | /* |
1659 | * Perform scheduler related setup for a newly forked process p. | |
1660 | * p is forked by current. | |
dd41f596 IM |
1661 | * |
1662 | * __sched_fork() is basic setup used by init_idle() too: | |
1663 | */ | |
1664 | static void __sched_fork(struct task_struct *p) | |
1665 | { | |
dd41f596 IM |
1666 | p->se.exec_start = 0; |
1667 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1668 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1669 | |
1670 | #ifdef CONFIG_SCHEDSTATS | |
1671 | p->se.wait_start = 0; | |
dd41f596 IM |
1672 | p->se.sum_sleep_runtime = 0; |
1673 | p->se.sleep_start = 0; | |
dd41f596 IM |
1674 | p->se.block_start = 0; |
1675 | p->se.sleep_max = 0; | |
1676 | p->se.block_max = 0; | |
1677 | p->se.exec_max = 0; | |
eba1ed4b | 1678 | p->se.slice_max = 0; |
dd41f596 | 1679 | p->se.wait_max = 0; |
6cfb0d5d | 1680 | #endif |
476d139c | 1681 | |
dd41f596 IM |
1682 | INIT_LIST_HEAD(&p->run_list); |
1683 | p->se.on_rq = 0; | |
476d139c | 1684 | |
e107be36 AK |
1685 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1686 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1687 | #endif | |
1688 | ||
1da177e4 LT |
1689 | /* |
1690 | * We mark the process as running here, but have not actually | |
1691 | * inserted it onto the runqueue yet. This guarantees that | |
1692 | * nobody will actually run it, and a signal or other external | |
1693 | * event cannot wake it up and insert it on the runqueue either. | |
1694 | */ | |
1695 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1696 | } |
1697 | ||
1698 | /* | |
1699 | * fork()/clone()-time setup: | |
1700 | */ | |
1701 | void sched_fork(struct task_struct *p, int clone_flags) | |
1702 | { | |
1703 | int cpu = get_cpu(); | |
1704 | ||
1705 | __sched_fork(p); | |
1706 | ||
1707 | #ifdef CONFIG_SMP | |
1708 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1709 | #endif | |
02e4bac2 | 1710 | set_task_cpu(p, cpu); |
b29739f9 IM |
1711 | |
1712 | /* | |
1713 | * Make sure we do not leak PI boosting priority to the child: | |
1714 | */ | |
1715 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1716 | if (!rt_prio(p->prio)) |
1717 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1718 | |
52f17b6c | 1719 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1720 | if (likely(sched_info_on())) |
52f17b6c | 1721 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1722 | #endif |
d6077cb8 | 1723 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1724 | p->oncpu = 0; |
1725 | #endif | |
1da177e4 | 1726 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1727 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1728 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1729 | #endif |
476d139c | 1730 | put_cpu(); |
1da177e4 LT |
1731 | } |
1732 | ||
1733 | /* | |
1734 | * wake_up_new_task - wake up a newly created task for the first time. | |
1735 | * | |
1736 | * This function will do some initial scheduler statistics housekeeping | |
1737 | * that must be done for every newly created context, then puts the task | |
1738 | * on the runqueue and wakes it. | |
1739 | */ | |
36c8b586 | 1740 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1741 | { |
1742 | unsigned long flags; | |
dd41f596 | 1743 | struct rq *rq; |
1da177e4 LT |
1744 | |
1745 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1746 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1747 | update_rq_clock(rq); |
1da177e4 LT |
1748 | |
1749 | p->prio = effective_prio(p); | |
1750 | ||
b9dca1e0 | 1751 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1752 | activate_task(rq, p, 0); |
1da177e4 | 1753 | } else { |
1da177e4 | 1754 | /* |
dd41f596 IM |
1755 | * Let the scheduling class do new task startup |
1756 | * management (if any): | |
1da177e4 | 1757 | */ |
ee0827d8 | 1758 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1759 | inc_nr_running(p, rq); |
1da177e4 | 1760 | } |
dd41f596 IM |
1761 | check_preempt_curr(rq, p); |
1762 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1763 | } |
1764 | ||
e107be36 AK |
1765 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1766 | ||
1767 | /** | |
421cee29 RD |
1768 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1769 | * @notifier: notifier struct to register | |
e107be36 AK |
1770 | */ |
1771 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1772 | { | |
1773 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1774 | } | |
1775 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1776 | ||
1777 | /** | |
1778 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1779 | * @notifier: notifier struct to unregister |
e107be36 AK |
1780 | * |
1781 | * This is safe to call from within a preemption notifier. | |
1782 | */ | |
1783 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1784 | { | |
1785 | hlist_del(¬ifier->link); | |
1786 | } | |
1787 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1788 | ||
1789 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1790 | { | |
1791 | struct preempt_notifier *notifier; | |
1792 | struct hlist_node *node; | |
1793 | ||
1794 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1795 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1796 | } | |
1797 | ||
1798 | static void | |
1799 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1800 | struct task_struct *next) | |
1801 | { | |
1802 | struct preempt_notifier *notifier; | |
1803 | struct hlist_node *node; | |
1804 | ||
1805 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1806 | notifier->ops->sched_out(notifier, next); | |
1807 | } | |
1808 | ||
1809 | #else | |
1810 | ||
1811 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1812 | { | |
1813 | } | |
1814 | ||
1815 | static void | |
1816 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1817 | struct task_struct *next) | |
1818 | { | |
1819 | } | |
1820 | ||
1821 | #endif | |
1822 | ||
4866cde0 NP |
1823 | /** |
1824 | * prepare_task_switch - prepare to switch tasks | |
1825 | * @rq: the runqueue preparing to switch | |
421cee29 | 1826 | * @prev: the current task that is being switched out |
4866cde0 NP |
1827 | * @next: the task we are going to switch to. |
1828 | * | |
1829 | * This is called with the rq lock held and interrupts off. It must | |
1830 | * be paired with a subsequent finish_task_switch after the context | |
1831 | * switch. | |
1832 | * | |
1833 | * prepare_task_switch sets up locking and calls architecture specific | |
1834 | * hooks. | |
1835 | */ | |
e107be36 AK |
1836 | static inline void |
1837 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1838 | struct task_struct *next) | |
4866cde0 | 1839 | { |
e107be36 | 1840 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1841 | prepare_lock_switch(rq, next); |
1842 | prepare_arch_switch(next); | |
1843 | } | |
1844 | ||
1da177e4 LT |
1845 | /** |
1846 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1847 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1848 | * @prev: the thread we just switched away from. |
1849 | * | |
4866cde0 NP |
1850 | * finish_task_switch must be called after the context switch, paired |
1851 | * with a prepare_task_switch call before the context switch. | |
1852 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1853 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1854 | * |
1855 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1856 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1857 | * with the lock held can cause deadlocks; see schedule() for |
1858 | * details.) | |
1859 | */ | |
a9957449 | 1860 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1861 | __releases(rq->lock) |
1862 | { | |
1da177e4 | 1863 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1864 | long prev_state; |
1da177e4 LT |
1865 | |
1866 | rq->prev_mm = NULL; | |
1867 | ||
1868 | /* | |
1869 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1870 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1871 | * schedule one last time. The schedule call will never return, and |
1872 | * the scheduled task must drop that reference. | |
c394cc9f | 1873 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1874 | * still held, otherwise prev could be scheduled on another cpu, die |
1875 | * there before we look at prev->state, and then the reference would | |
1876 | * be dropped twice. | |
1877 | * Manfred Spraul <manfred@colorfullife.com> | |
1878 | */ | |
55a101f8 | 1879 | prev_state = prev->state; |
4866cde0 NP |
1880 | finish_arch_switch(prev); |
1881 | finish_lock_switch(rq, prev); | |
e107be36 | 1882 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1883 | if (mm) |
1884 | mmdrop(mm); | |
c394cc9f | 1885 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1886 | /* |
1887 | * Remove function-return probe instances associated with this | |
1888 | * task and put them back on the free list. | |
9761eea8 | 1889 | */ |
c6fd91f0 | 1890 | kprobe_flush_task(prev); |
1da177e4 | 1891 | put_task_struct(prev); |
c6fd91f0 | 1892 | } |
1da177e4 LT |
1893 | } |
1894 | ||
1895 | /** | |
1896 | * schedule_tail - first thing a freshly forked thread must call. | |
1897 | * @prev: the thread we just switched away from. | |
1898 | */ | |
36c8b586 | 1899 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1900 | __releases(rq->lock) |
1901 | { | |
70b97a7f IM |
1902 | struct rq *rq = this_rq(); |
1903 | ||
4866cde0 NP |
1904 | finish_task_switch(rq, prev); |
1905 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1906 | /* In this case, finish_task_switch does not reenable preemption */ | |
1907 | preempt_enable(); | |
1908 | #endif | |
1da177e4 | 1909 | if (current->set_child_tid) |
b488893a | 1910 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1911 | } |
1912 | ||
1913 | /* | |
1914 | * context_switch - switch to the new MM and the new | |
1915 | * thread's register state. | |
1916 | */ | |
dd41f596 | 1917 | static inline void |
70b97a7f | 1918 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1919 | struct task_struct *next) |
1da177e4 | 1920 | { |
dd41f596 | 1921 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1922 | |
e107be36 | 1923 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1924 | mm = next->mm; |
1925 | oldmm = prev->active_mm; | |
9226d125 ZA |
1926 | /* |
1927 | * For paravirt, this is coupled with an exit in switch_to to | |
1928 | * combine the page table reload and the switch backend into | |
1929 | * one hypercall. | |
1930 | */ | |
1931 | arch_enter_lazy_cpu_mode(); | |
1932 | ||
dd41f596 | 1933 | if (unlikely(!mm)) { |
1da177e4 LT |
1934 | next->active_mm = oldmm; |
1935 | atomic_inc(&oldmm->mm_count); | |
1936 | enter_lazy_tlb(oldmm, next); | |
1937 | } else | |
1938 | switch_mm(oldmm, mm, next); | |
1939 | ||
dd41f596 | 1940 | if (unlikely(!prev->mm)) { |
1da177e4 | 1941 | prev->active_mm = NULL; |
1da177e4 LT |
1942 | rq->prev_mm = oldmm; |
1943 | } | |
3a5f5e48 IM |
1944 | /* |
1945 | * Since the runqueue lock will be released by the next | |
1946 | * task (which is an invalid locking op but in the case | |
1947 | * of the scheduler it's an obvious special-case), so we | |
1948 | * do an early lockdep release here: | |
1949 | */ | |
1950 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1951 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1952 | #endif |
1da177e4 LT |
1953 | |
1954 | /* Here we just switch the register state and the stack. */ | |
1955 | switch_to(prev, next, prev); | |
1956 | ||
dd41f596 IM |
1957 | barrier(); |
1958 | /* | |
1959 | * this_rq must be evaluated again because prev may have moved | |
1960 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1961 | * frame will be invalid. | |
1962 | */ | |
1963 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1964 | } |
1965 | ||
1966 | /* | |
1967 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1968 | * | |
1969 | * externally visible scheduler statistics: current number of runnable | |
1970 | * threads, current number of uninterruptible-sleeping threads, total | |
1971 | * number of context switches performed since bootup. | |
1972 | */ | |
1973 | unsigned long nr_running(void) | |
1974 | { | |
1975 | unsigned long i, sum = 0; | |
1976 | ||
1977 | for_each_online_cpu(i) | |
1978 | sum += cpu_rq(i)->nr_running; | |
1979 | ||
1980 | return sum; | |
1981 | } | |
1982 | ||
1983 | unsigned long nr_uninterruptible(void) | |
1984 | { | |
1985 | unsigned long i, sum = 0; | |
1986 | ||
0a945022 | 1987 | for_each_possible_cpu(i) |
1da177e4 LT |
1988 | sum += cpu_rq(i)->nr_uninterruptible; |
1989 | ||
1990 | /* | |
1991 | * Since we read the counters lockless, it might be slightly | |
1992 | * inaccurate. Do not allow it to go below zero though: | |
1993 | */ | |
1994 | if (unlikely((long)sum < 0)) | |
1995 | sum = 0; | |
1996 | ||
1997 | return sum; | |
1998 | } | |
1999 | ||
2000 | unsigned long long nr_context_switches(void) | |
2001 | { | |
cc94abfc SR |
2002 | int i; |
2003 | unsigned long long sum = 0; | |
1da177e4 | 2004 | |
0a945022 | 2005 | for_each_possible_cpu(i) |
1da177e4 LT |
2006 | sum += cpu_rq(i)->nr_switches; |
2007 | ||
2008 | return sum; | |
2009 | } | |
2010 | ||
2011 | unsigned long nr_iowait(void) | |
2012 | { | |
2013 | unsigned long i, sum = 0; | |
2014 | ||
0a945022 | 2015 | for_each_possible_cpu(i) |
1da177e4 LT |
2016 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2017 | ||
2018 | return sum; | |
2019 | } | |
2020 | ||
db1b1fef JS |
2021 | unsigned long nr_active(void) |
2022 | { | |
2023 | unsigned long i, running = 0, uninterruptible = 0; | |
2024 | ||
2025 | for_each_online_cpu(i) { | |
2026 | running += cpu_rq(i)->nr_running; | |
2027 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2028 | } | |
2029 | ||
2030 | if (unlikely((long)uninterruptible < 0)) | |
2031 | uninterruptible = 0; | |
2032 | ||
2033 | return running + uninterruptible; | |
2034 | } | |
2035 | ||
48f24c4d | 2036 | /* |
dd41f596 IM |
2037 | * Update rq->cpu_load[] statistics. This function is usually called every |
2038 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2039 | */ |
dd41f596 | 2040 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2041 | { |
495eca49 | 2042 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2043 | int i, scale; |
2044 | ||
2045 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2046 | |
2047 | /* Update our load: */ | |
2048 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2049 | unsigned long old_load, new_load; | |
2050 | ||
2051 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2052 | ||
2053 | old_load = this_rq->cpu_load[i]; | |
2054 | new_load = this_load; | |
a25707f3 IM |
2055 | /* |
2056 | * Round up the averaging division if load is increasing. This | |
2057 | * prevents us from getting stuck on 9 if the load is 10, for | |
2058 | * example. | |
2059 | */ | |
2060 | if (new_load > old_load) | |
2061 | new_load += scale-1; | |
dd41f596 IM |
2062 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2063 | } | |
48f24c4d IM |
2064 | } |
2065 | ||
dd41f596 IM |
2066 | #ifdef CONFIG_SMP |
2067 | ||
1da177e4 LT |
2068 | /* |
2069 | * double_rq_lock - safely lock two runqueues | |
2070 | * | |
2071 | * Note this does not disable interrupts like task_rq_lock, | |
2072 | * you need to do so manually before calling. | |
2073 | */ | |
70b97a7f | 2074 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2075 | __acquires(rq1->lock) |
2076 | __acquires(rq2->lock) | |
2077 | { | |
054b9108 | 2078 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2079 | if (rq1 == rq2) { |
2080 | spin_lock(&rq1->lock); | |
2081 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2082 | } else { | |
c96d145e | 2083 | if (rq1 < rq2) { |
1da177e4 LT |
2084 | spin_lock(&rq1->lock); |
2085 | spin_lock(&rq2->lock); | |
2086 | } else { | |
2087 | spin_lock(&rq2->lock); | |
2088 | spin_lock(&rq1->lock); | |
2089 | } | |
2090 | } | |
6e82a3be IM |
2091 | update_rq_clock(rq1); |
2092 | update_rq_clock(rq2); | |
1da177e4 LT |
2093 | } |
2094 | ||
2095 | /* | |
2096 | * double_rq_unlock - safely unlock two runqueues | |
2097 | * | |
2098 | * Note this does not restore interrupts like task_rq_unlock, | |
2099 | * you need to do so manually after calling. | |
2100 | */ | |
70b97a7f | 2101 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2102 | __releases(rq1->lock) |
2103 | __releases(rq2->lock) | |
2104 | { | |
2105 | spin_unlock(&rq1->lock); | |
2106 | if (rq1 != rq2) | |
2107 | spin_unlock(&rq2->lock); | |
2108 | else | |
2109 | __release(rq2->lock); | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2114 | */ | |
70b97a7f | 2115 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2116 | __releases(this_rq->lock) |
2117 | __acquires(busiest->lock) | |
2118 | __acquires(this_rq->lock) | |
2119 | { | |
054b9108 KK |
2120 | if (unlikely(!irqs_disabled())) { |
2121 | /* printk() doesn't work good under rq->lock */ | |
2122 | spin_unlock(&this_rq->lock); | |
2123 | BUG_ON(1); | |
2124 | } | |
1da177e4 | 2125 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2126 | if (busiest < this_rq) { |
1da177e4 LT |
2127 | spin_unlock(&this_rq->lock); |
2128 | spin_lock(&busiest->lock); | |
2129 | spin_lock(&this_rq->lock); | |
2130 | } else | |
2131 | spin_lock(&busiest->lock); | |
2132 | } | |
2133 | } | |
2134 | ||
1da177e4 LT |
2135 | /* |
2136 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2137 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2138 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2139 | * the cpu_allowed mask is restored. |
2140 | */ | |
36c8b586 | 2141 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2142 | { |
70b97a7f | 2143 | struct migration_req req; |
1da177e4 | 2144 | unsigned long flags; |
70b97a7f | 2145 | struct rq *rq; |
1da177e4 LT |
2146 | |
2147 | rq = task_rq_lock(p, &flags); | |
2148 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2149 | || unlikely(cpu_is_offline(dest_cpu))) | |
2150 | goto out; | |
2151 | ||
2152 | /* force the process onto the specified CPU */ | |
2153 | if (migrate_task(p, dest_cpu, &req)) { | |
2154 | /* Need to wait for migration thread (might exit: take ref). */ | |
2155 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2156 | |
1da177e4 LT |
2157 | get_task_struct(mt); |
2158 | task_rq_unlock(rq, &flags); | |
2159 | wake_up_process(mt); | |
2160 | put_task_struct(mt); | |
2161 | wait_for_completion(&req.done); | |
36c8b586 | 2162 | |
1da177e4 LT |
2163 | return; |
2164 | } | |
2165 | out: | |
2166 | task_rq_unlock(rq, &flags); | |
2167 | } | |
2168 | ||
2169 | /* | |
476d139c NP |
2170 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2171 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2172 | */ |
2173 | void sched_exec(void) | |
2174 | { | |
1da177e4 | 2175 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2176 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2177 | put_cpu(); |
476d139c NP |
2178 | if (new_cpu != this_cpu) |
2179 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2180 | } |
2181 | ||
2182 | /* | |
2183 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2184 | * Both runqueues must be locked. | |
2185 | */ | |
dd41f596 IM |
2186 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2187 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2188 | { |
2e1cb74a | 2189 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2190 | set_task_cpu(p, this_cpu); |
dd41f596 | 2191 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2192 | /* |
2193 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2194 | * to be always true for them. | |
2195 | */ | |
dd41f596 | 2196 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2197 | } |
2198 | ||
2199 | /* | |
2200 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2201 | */ | |
858119e1 | 2202 | static |
70b97a7f | 2203 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2204 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2205 | int *all_pinned) |
1da177e4 LT |
2206 | { |
2207 | /* | |
2208 | * We do not migrate tasks that are: | |
2209 | * 1) running (obviously), or | |
2210 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2211 | * 3) are cache-hot on their current CPU. | |
2212 | */ | |
cc367732 IM |
2213 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2214 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2215 | return 0; |
cc367732 | 2216 | } |
81026794 NP |
2217 | *all_pinned = 0; |
2218 | ||
cc367732 IM |
2219 | if (task_running(rq, p)) { |
2220 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2221 | return 0; |
cc367732 | 2222 | } |
1da177e4 | 2223 | |
da84d961 IM |
2224 | /* |
2225 | * Aggressive migration if: | |
2226 | * 1) task is cache cold, or | |
2227 | * 2) too many balance attempts have failed. | |
2228 | */ | |
2229 | ||
6bc1665b IM |
2230 | if (!task_hot(p, rq->clock, sd) || |
2231 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2232 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2233 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2234 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2235 | schedstat_inc(p, se.nr_forced_migrations); |
2236 | } | |
da84d961 IM |
2237 | #endif |
2238 | return 1; | |
2239 | } | |
2240 | ||
cc367732 IM |
2241 | if (task_hot(p, rq->clock, sd)) { |
2242 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2243 | return 0; |
cc367732 | 2244 | } |
1da177e4 LT |
2245 | return 1; |
2246 | } | |
2247 | ||
e1d1484f PW |
2248 | static unsigned long |
2249 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2250 | unsigned long max_load_move, struct sched_domain *sd, | |
2251 | enum cpu_idle_type idle, int *all_pinned, | |
2252 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2253 | { |
b82d9fdd | 2254 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2255 | struct task_struct *p; |
2256 | long rem_load_move = max_load_move; | |
1da177e4 | 2257 | |
e1d1484f | 2258 | if (max_load_move == 0) |
1da177e4 LT |
2259 | goto out; |
2260 | ||
81026794 NP |
2261 | pinned = 1; |
2262 | ||
1da177e4 | 2263 | /* |
dd41f596 | 2264 | * Start the load-balancing iterator: |
1da177e4 | 2265 | */ |
dd41f596 IM |
2266 | p = iterator->start(iterator->arg); |
2267 | next: | |
b82d9fdd | 2268 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2269 | goto out; |
50ddd969 | 2270 | /* |
b82d9fdd | 2271 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2272 | * skip a task if it will be the highest priority task (i.e. smallest |
2273 | * prio value) on its new queue regardless of its load weight | |
2274 | */ | |
dd41f596 IM |
2275 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2276 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2277 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2278 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2279 | p = iterator->next(iterator->arg); |
2280 | goto next; | |
1da177e4 LT |
2281 | } |
2282 | ||
dd41f596 | 2283 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2284 | pulled++; |
dd41f596 | 2285 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2286 | |
2dd73a4f | 2287 | /* |
b82d9fdd | 2288 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2289 | */ |
e1d1484f | 2290 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2291 | if (p->prio < *this_best_prio) |
2292 | *this_best_prio = p->prio; | |
dd41f596 IM |
2293 | p = iterator->next(iterator->arg); |
2294 | goto next; | |
1da177e4 LT |
2295 | } |
2296 | out: | |
2297 | /* | |
e1d1484f | 2298 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2299 | * so we can safely collect pull_task() stats here rather than |
2300 | * inside pull_task(). | |
2301 | */ | |
2302 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2303 | |
2304 | if (all_pinned) | |
2305 | *all_pinned = pinned; | |
e1d1484f PW |
2306 | |
2307 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2308 | } |
2309 | ||
dd41f596 | 2310 | /* |
43010659 PW |
2311 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2312 | * this_rq, as part of a balancing operation within domain "sd". | |
2313 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2314 | * |
2315 | * Called with both runqueues locked. | |
2316 | */ | |
2317 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2318 | unsigned long max_load_move, |
dd41f596 IM |
2319 | struct sched_domain *sd, enum cpu_idle_type idle, |
2320 | int *all_pinned) | |
2321 | { | |
5522d5d5 | 2322 | const struct sched_class *class = sched_class_highest; |
43010659 | 2323 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2324 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2325 | |
2326 | do { | |
43010659 PW |
2327 | total_load_moved += |
2328 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2329 | max_load_move - total_load_moved, |
a4ac01c3 | 2330 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2331 | class = class->next; |
43010659 | 2332 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2333 | |
43010659 PW |
2334 | return total_load_moved > 0; |
2335 | } | |
2336 | ||
e1d1484f PW |
2337 | static int |
2338 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2339 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2340 | struct rq_iterator *iterator) | |
2341 | { | |
2342 | struct task_struct *p = iterator->start(iterator->arg); | |
2343 | int pinned = 0; | |
2344 | ||
2345 | while (p) { | |
2346 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2347 | pull_task(busiest, p, this_rq, this_cpu); | |
2348 | /* | |
2349 | * Right now, this is only the second place pull_task() | |
2350 | * is called, so we can safely collect pull_task() | |
2351 | * stats here rather than inside pull_task(). | |
2352 | */ | |
2353 | schedstat_inc(sd, lb_gained[idle]); | |
2354 | ||
2355 | return 1; | |
2356 | } | |
2357 | p = iterator->next(iterator->arg); | |
2358 | } | |
2359 | ||
2360 | return 0; | |
2361 | } | |
2362 | ||
43010659 PW |
2363 | /* |
2364 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2365 | * part of active balancing operations within "domain". | |
2366 | * Returns 1 if successful and 0 otherwise. | |
2367 | * | |
2368 | * Called with both runqueues locked. | |
2369 | */ | |
2370 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2371 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2372 | { | |
5522d5d5 | 2373 | const struct sched_class *class; |
43010659 PW |
2374 | |
2375 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2376 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2377 | return 1; |
2378 | ||
2379 | return 0; | |
dd41f596 IM |
2380 | } |
2381 | ||
1da177e4 LT |
2382 | /* |
2383 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2384 | * domain. It calculates and returns the amount of weighted load which |
2385 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2386 | */ |
2387 | static struct sched_group * | |
2388 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2389 | unsigned long *imbalance, enum cpu_idle_type idle, |
2390 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2391 | { |
2392 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2393 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2394 | unsigned long max_pull; |
2dd73a4f PW |
2395 | unsigned long busiest_load_per_task, busiest_nr_running; |
2396 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2397 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2398 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2399 | int power_savings_balance = 1; | |
2400 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2401 | unsigned long min_nr_running = ULONG_MAX; | |
2402 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2403 | #endif | |
1da177e4 LT |
2404 | |
2405 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2406 | busiest_load_per_task = busiest_nr_running = 0; |
2407 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2408 | if (idle == CPU_NOT_IDLE) |
7897986b | 2409 | load_idx = sd->busy_idx; |
d15bcfdb | 2410 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2411 | load_idx = sd->newidle_idx; |
2412 | else | |
2413 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2414 | |
2415 | do { | |
908a7c1b | 2416 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2417 | int local_group; |
2418 | int i; | |
908a7c1b | 2419 | int __group_imb = 0; |
783609c6 | 2420 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2421 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2422 | |
2423 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2424 | ||
783609c6 SS |
2425 | if (local_group) |
2426 | balance_cpu = first_cpu(group->cpumask); | |
2427 | ||
1da177e4 | 2428 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2429 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2430 | max_cpu_load = 0; |
2431 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2432 | |
2433 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2434 | struct rq *rq; |
2435 | ||
2436 | if (!cpu_isset(i, *cpus)) | |
2437 | continue; | |
2438 | ||
2439 | rq = cpu_rq(i); | |
2dd73a4f | 2440 | |
9439aab8 | 2441 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2442 | *sd_idle = 0; |
2443 | ||
1da177e4 | 2444 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2445 | if (local_group) { |
2446 | if (idle_cpu(i) && !first_idle_cpu) { | |
2447 | first_idle_cpu = 1; | |
2448 | balance_cpu = i; | |
2449 | } | |
2450 | ||
a2000572 | 2451 | load = target_load(i, load_idx); |
908a7c1b | 2452 | } else { |
a2000572 | 2453 | load = source_load(i, load_idx); |
908a7c1b KC |
2454 | if (load > max_cpu_load) |
2455 | max_cpu_load = load; | |
2456 | if (min_cpu_load > load) | |
2457 | min_cpu_load = load; | |
2458 | } | |
1da177e4 LT |
2459 | |
2460 | avg_load += load; | |
2dd73a4f | 2461 | sum_nr_running += rq->nr_running; |
dd41f596 | 2462 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2463 | } |
2464 | ||
783609c6 SS |
2465 | /* |
2466 | * First idle cpu or the first cpu(busiest) in this sched group | |
2467 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2468 | * domains. In the newly idle case, we will allow all the cpu's |
2469 | * to do the newly idle load balance. | |
783609c6 | 2470 | */ |
9439aab8 SS |
2471 | if (idle != CPU_NEWLY_IDLE && local_group && |
2472 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2473 | *balance = 0; |
2474 | goto ret; | |
2475 | } | |
2476 | ||
1da177e4 | 2477 | total_load += avg_load; |
5517d86b | 2478 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2479 | |
2480 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2481 | avg_load = sg_div_cpu_power(group, |
2482 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2483 | |
908a7c1b KC |
2484 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2485 | __group_imb = 1; | |
2486 | ||
5517d86b | 2487 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2488 | |
1da177e4 LT |
2489 | if (local_group) { |
2490 | this_load = avg_load; | |
2491 | this = group; | |
2dd73a4f PW |
2492 | this_nr_running = sum_nr_running; |
2493 | this_load_per_task = sum_weighted_load; | |
2494 | } else if (avg_load > max_load && | |
908a7c1b | 2495 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2496 | max_load = avg_load; |
2497 | busiest = group; | |
2dd73a4f PW |
2498 | busiest_nr_running = sum_nr_running; |
2499 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2500 | group_imb = __group_imb; |
1da177e4 | 2501 | } |
5c45bf27 SS |
2502 | |
2503 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2504 | /* | |
2505 | * Busy processors will not participate in power savings | |
2506 | * balance. | |
2507 | */ | |
dd41f596 IM |
2508 | if (idle == CPU_NOT_IDLE || |
2509 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2510 | goto group_next; | |
5c45bf27 SS |
2511 | |
2512 | /* | |
2513 | * If the local group is idle or completely loaded | |
2514 | * no need to do power savings balance at this domain | |
2515 | */ | |
2516 | if (local_group && (this_nr_running >= group_capacity || | |
2517 | !this_nr_running)) | |
2518 | power_savings_balance = 0; | |
2519 | ||
dd41f596 | 2520 | /* |
5c45bf27 SS |
2521 | * If a group is already running at full capacity or idle, |
2522 | * don't include that group in power savings calculations | |
dd41f596 IM |
2523 | */ |
2524 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2525 | || !sum_nr_running) |
dd41f596 | 2526 | goto group_next; |
5c45bf27 | 2527 | |
dd41f596 | 2528 | /* |
5c45bf27 | 2529 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2530 | * This is the group from where we need to pick up the load |
2531 | * for saving power | |
2532 | */ | |
2533 | if ((sum_nr_running < min_nr_running) || | |
2534 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2535 | first_cpu(group->cpumask) < |
2536 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2537 | group_min = group; |
2538 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2539 | min_load_per_task = sum_weighted_load / |
2540 | sum_nr_running; | |
dd41f596 | 2541 | } |
5c45bf27 | 2542 | |
dd41f596 | 2543 | /* |
5c45bf27 | 2544 | * Calculate the group which is almost near its |
dd41f596 IM |
2545 | * capacity but still has some space to pick up some load |
2546 | * from other group and save more power | |
2547 | */ | |
2548 | if (sum_nr_running <= group_capacity - 1) { | |
2549 | if (sum_nr_running > leader_nr_running || | |
2550 | (sum_nr_running == leader_nr_running && | |
2551 | first_cpu(group->cpumask) > | |
2552 | first_cpu(group_leader->cpumask))) { | |
2553 | group_leader = group; | |
2554 | leader_nr_running = sum_nr_running; | |
2555 | } | |
48f24c4d | 2556 | } |
5c45bf27 SS |
2557 | group_next: |
2558 | #endif | |
1da177e4 LT |
2559 | group = group->next; |
2560 | } while (group != sd->groups); | |
2561 | ||
2dd73a4f | 2562 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2563 | goto out_balanced; |
2564 | ||
2565 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2566 | ||
2567 | if (this_load >= avg_load || | |
2568 | 100*max_load <= sd->imbalance_pct*this_load) | |
2569 | goto out_balanced; | |
2570 | ||
2dd73a4f | 2571 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2572 | if (group_imb) |
2573 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2574 | ||
1da177e4 LT |
2575 | /* |
2576 | * We're trying to get all the cpus to the average_load, so we don't | |
2577 | * want to push ourselves above the average load, nor do we wish to | |
2578 | * reduce the max loaded cpu below the average load, as either of these | |
2579 | * actions would just result in more rebalancing later, and ping-pong | |
2580 | * tasks around. Thus we look for the minimum possible imbalance. | |
2581 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2582 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2583 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2584 | * appear as very large values with unsigned longs. |
2585 | */ | |
2dd73a4f PW |
2586 | if (max_load <= busiest_load_per_task) |
2587 | goto out_balanced; | |
2588 | ||
2589 | /* | |
2590 | * In the presence of smp nice balancing, certain scenarios can have | |
2591 | * max load less than avg load(as we skip the groups at or below | |
2592 | * its cpu_power, while calculating max_load..) | |
2593 | */ | |
2594 | if (max_load < avg_load) { | |
2595 | *imbalance = 0; | |
2596 | goto small_imbalance; | |
2597 | } | |
0c117f1b SS |
2598 | |
2599 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2600 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2601 | |
1da177e4 | 2602 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2603 | *imbalance = min(max_pull * busiest->__cpu_power, |
2604 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2605 | / SCHED_LOAD_SCALE; |
2606 | ||
2dd73a4f PW |
2607 | /* |
2608 | * if *imbalance is less than the average load per runnable task | |
2609 | * there is no gaurantee that any tasks will be moved so we'll have | |
2610 | * a think about bumping its value to force at least one task to be | |
2611 | * moved | |
2612 | */ | |
7fd0d2dd | 2613 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2614 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2615 | unsigned int imbn; |
2616 | ||
2617 | small_imbalance: | |
2618 | pwr_move = pwr_now = 0; | |
2619 | imbn = 2; | |
2620 | if (this_nr_running) { | |
2621 | this_load_per_task /= this_nr_running; | |
2622 | if (busiest_load_per_task > this_load_per_task) | |
2623 | imbn = 1; | |
2624 | } else | |
2625 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2626 | |
dd41f596 IM |
2627 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2628 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2629 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2630 | return busiest; |
2631 | } | |
2632 | ||
2633 | /* | |
2634 | * OK, we don't have enough imbalance to justify moving tasks, | |
2635 | * however we may be able to increase total CPU power used by | |
2636 | * moving them. | |
2637 | */ | |
2638 | ||
5517d86b ED |
2639 | pwr_now += busiest->__cpu_power * |
2640 | min(busiest_load_per_task, max_load); | |
2641 | pwr_now += this->__cpu_power * | |
2642 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2643 | pwr_now /= SCHED_LOAD_SCALE; |
2644 | ||
2645 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2646 | tmp = sg_div_cpu_power(busiest, |
2647 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2648 | if (max_load > tmp) |
5517d86b | 2649 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2650 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2651 | |
2652 | /* Amount of load we'd add */ | |
5517d86b | 2653 | if (max_load * busiest->__cpu_power < |
33859f7f | 2654 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2655 | tmp = sg_div_cpu_power(this, |
2656 | max_load * busiest->__cpu_power); | |
1da177e4 | 2657 | else |
5517d86b ED |
2658 | tmp = sg_div_cpu_power(this, |
2659 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2660 | pwr_move += this->__cpu_power * | |
2661 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2662 | pwr_move /= SCHED_LOAD_SCALE; |
2663 | ||
2664 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2665 | if (pwr_move > pwr_now) |
2666 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2667 | } |
2668 | ||
1da177e4 LT |
2669 | return busiest; |
2670 | ||
2671 | out_balanced: | |
5c45bf27 | 2672 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2673 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2674 | goto ret; |
1da177e4 | 2675 | |
5c45bf27 SS |
2676 | if (this == group_leader && group_leader != group_min) { |
2677 | *imbalance = min_load_per_task; | |
2678 | return group_min; | |
2679 | } | |
5c45bf27 | 2680 | #endif |
783609c6 | 2681 | ret: |
1da177e4 LT |
2682 | *imbalance = 0; |
2683 | return NULL; | |
2684 | } | |
2685 | ||
2686 | /* | |
2687 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2688 | */ | |
70b97a7f | 2689 | static struct rq * |
d15bcfdb | 2690 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2691 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2692 | { |
70b97a7f | 2693 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2694 | unsigned long max_load = 0; |
1da177e4 LT |
2695 | int i; |
2696 | ||
2697 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2698 | unsigned long wl; |
0a2966b4 CL |
2699 | |
2700 | if (!cpu_isset(i, *cpus)) | |
2701 | continue; | |
2702 | ||
48f24c4d | 2703 | rq = cpu_rq(i); |
dd41f596 | 2704 | wl = weighted_cpuload(i); |
2dd73a4f | 2705 | |
dd41f596 | 2706 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2707 | continue; |
1da177e4 | 2708 | |
dd41f596 IM |
2709 | if (wl > max_load) { |
2710 | max_load = wl; | |
48f24c4d | 2711 | busiest = rq; |
1da177e4 LT |
2712 | } |
2713 | } | |
2714 | ||
2715 | return busiest; | |
2716 | } | |
2717 | ||
77391d71 NP |
2718 | /* |
2719 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2720 | * so long as it is large enough. | |
2721 | */ | |
2722 | #define MAX_PINNED_INTERVAL 512 | |
2723 | ||
1da177e4 LT |
2724 | /* |
2725 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2726 | * tasks if there is an imbalance. | |
1da177e4 | 2727 | */ |
70b97a7f | 2728 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2729 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2730 | int *balance) |
1da177e4 | 2731 | { |
43010659 | 2732 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2733 | struct sched_group *group; |
1da177e4 | 2734 | unsigned long imbalance; |
70b97a7f | 2735 | struct rq *busiest; |
0a2966b4 | 2736 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2737 | unsigned long flags; |
5969fe06 | 2738 | |
89c4710e SS |
2739 | /* |
2740 | * When power savings policy is enabled for the parent domain, idle | |
2741 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2742 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2743 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2744 | */ |
d15bcfdb | 2745 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2746 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2747 | sd_idle = 1; |
1da177e4 | 2748 | |
2d72376b | 2749 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2750 | |
0a2966b4 CL |
2751 | redo: |
2752 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2753 | &cpus, balance); |
2754 | ||
06066714 | 2755 | if (*balance == 0) |
783609c6 | 2756 | goto out_balanced; |
783609c6 | 2757 | |
1da177e4 LT |
2758 | if (!group) { |
2759 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2760 | goto out_balanced; | |
2761 | } | |
2762 | ||
0a2966b4 | 2763 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2764 | if (!busiest) { |
2765 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2766 | goto out_balanced; | |
2767 | } | |
2768 | ||
db935dbd | 2769 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2770 | |
2771 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2772 | ||
43010659 | 2773 | ld_moved = 0; |
1da177e4 LT |
2774 | if (busiest->nr_running > 1) { |
2775 | /* | |
2776 | * Attempt to move tasks. If find_busiest_group has found | |
2777 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2778 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2779 | * correctly treated as an imbalance. |
2780 | */ | |
fe2eea3f | 2781 | local_irq_save(flags); |
e17224bf | 2782 | double_rq_lock(this_rq, busiest); |
43010659 | 2783 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2784 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2785 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2786 | local_irq_restore(flags); |
81026794 | 2787 | |
46cb4b7c SS |
2788 | /* |
2789 | * some other cpu did the load balance for us. | |
2790 | */ | |
43010659 | 2791 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2792 | resched_cpu(this_cpu); |
2793 | ||
81026794 | 2794 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2795 | if (unlikely(all_pinned)) { |
2796 | cpu_clear(cpu_of(busiest), cpus); | |
2797 | if (!cpus_empty(cpus)) | |
2798 | goto redo; | |
81026794 | 2799 | goto out_balanced; |
0a2966b4 | 2800 | } |
1da177e4 | 2801 | } |
81026794 | 2802 | |
43010659 | 2803 | if (!ld_moved) { |
1da177e4 LT |
2804 | schedstat_inc(sd, lb_failed[idle]); |
2805 | sd->nr_balance_failed++; | |
2806 | ||
2807 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2808 | |
fe2eea3f | 2809 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2810 | |
2811 | /* don't kick the migration_thread, if the curr | |
2812 | * task on busiest cpu can't be moved to this_cpu | |
2813 | */ | |
2814 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2815 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2816 | all_pinned = 1; |
2817 | goto out_one_pinned; | |
2818 | } | |
2819 | ||
1da177e4 LT |
2820 | if (!busiest->active_balance) { |
2821 | busiest->active_balance = 1; | |
2822 | busiest->push_cpu = this_cpu; | |
81026794 | 2823 | active_balance = 1; |
1da177e4 | 2824 | } |
fe2eea3f | 2825 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2826 | if (active_balance) |
1da177e4 LT |
2827 | wake_up_process(busiest->migration_thread); |
2828 | ||
2829 | /* | |
2830 | * We've kicked active balancing, reset the failure | |
2831 | * counter. | |
2832 | */ | |
39507451 | 2833 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2834 | } |
81026794 | 2835 | } else |
1da177e4 LT |
2836 | sd->nr_balance_failed = 0; |
2837 | ||
81026794 | 2838 | if (likely(!active_balance)) { |
1da177e4 LT |
2839 | /* We were unbalanced, so reset the balancing interval */ |
2840 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2841 | } else { |
2842 | /* | |
2843 | * If we've begun active balancing, start to back off. This | |
2844 | * case may not be covered by the all_pinned logic if there | |
2845 | * is only 1 task on the busy runqueue (because we don't call | |
2846 | * move_tasks). | |
2847 | */ | |
2848 | if (sd->balance_interval < sd->max_interval) | |
2849 | sd->balance_interval *= 2; | |
1da177e4 LT |
2850 | } |
2851 | ||
43010659 | 2852 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2853 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2854 | return -1; |
43010659 | 2855 | return ld_moved; |
1da177e4 LT |
2856 | |
2857 | out_balanced: | |
1da177e4 LT |
2858 | schedstat_inc(sd, lb_balanced[idle]); |
2859 | ||
16cfb1c0 | 2860 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2861 | |
2862 | out_one_pinned: | |
1da177e4 | 2863 | /* tune up the balancing interval */ |
77391d71 NP |
2864 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2865 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2866 | sd->balance_interval *= 2; |
2867 | ||
48f24c4d | 2868 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2869 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2870 | return -1; |
1da177e4 LT |
2871 | return 0; |
2872 | } | |
2873 | ||
2874 | /* | |
2875 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2876 | * tasks if there is an imbalance. | |
2877 | * | |
d15bcfdb | 2878 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2879 | * this_rq is locked. |
2880 | */ | |
48f24c4d | 2881 | static int |
70b97a7f | 2882 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2883 | { |
2884 | struct sched_group *group; | |
70b97a7f | 2885 | struct rq *busiest = NULL; |
1da177e4 | 2886 | unsigned long imbalance; |
43010659 | 2887 | int ld_moved = 0; |
5969fe06 | 2888 | int sd_idle = 0; |
969bb4e4 | 2889 | int all_pinned = 0; |
0a2966b4 | 2890 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2891 | |
89c4710e SS |
2892 | /* |
2893 | * When power savings policy is enabled for the parent domain, idle | |
2894 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2895 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2896 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2897 | */ |
2898 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2899 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2900 | sd_idle = 1; |
1da177e4 | 2901 | |
2d72376b | 2902 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2903 | redo: |
d15bcfdb | 2904 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2905 | &sd_idle, &cpus, NULL); |
1da177e4 | 2906 | if (!group) { |
d15bcfdb | 2907 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2908 | goto out_balanced; |
1da177e4 LT |
2909 | } |
2910 | ||
d15bcfdb | 2911 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2912 | &cpus); |
db935dbd | 2913 | if (!busiest) { |
d15bcfdb | 2914 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2915 | goto out_balanced; |
1da177e4 LT |
2916 | } |
2917 | ||
db935dbd NP |
2918 | BUG_ON(busiest == this_rq); |
2919 | ||
d15bcfdb | 2920 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2921 | |
43010659 | 2922 | ld_moved = 0; |
d6d5cfaf NP |
2923 | if (busiest->nr_running > 1) { |
2924 | /* Attempt to move tasks */ | |
2925 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2926 | /* this_rq->clock is already updated */ |
2927 | update_rq_clock(busiest); | |
43010659 | 2928 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2929 | imbalance, sd, CPU_NEWLY_IDLE, |
2930 | &all_pinned); | |
d6d5cfaf | 2931 | spin_unlock(&busiest->lock); |
0a2966b4 | 2932 | |
969bb4e4 | 2933 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2934 | cpu_clear(cpu_of(busiest), cpus); |
2935 | if (!cpus_empty(cpus)) | |
2936 | goto redo; | |
2937 | } | |
d6d5cfaf NP |
2938 | } |
2939 | ||
43010659 | 2940 | if (!ld_moved) { |
d15bcfdb | 2941 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2942 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2943 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2944 | return -1; |
2945 | } else | |
16cfb1c0 | 2946 | sd->nr_balance_failed = 0; |
1da177e4 | 2947 | |
43010659 | 2948 | return ld_moved; |
16cfb1c0 NP |
2949 | |
2950 | out_balanced: | |
d15bcfdb | 2951 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2952 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2953 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2954 | return -1; |
16cfb1c0 | 2955 | sd->nr_balance_failed = 0; |
48f24c4d | 2956 | |
16cfb1c0 | 2957 | return 0; |
1da177e4 LT |
2958 | } |
2959 | ||
2960 | /* | |
2961 | * idle_balance is called by schedule() if this_cpu is about to become | |
2962 | * idle. Attempts to pull tasks from other CPUs. | |
2963 | */ | |
70b97a7f | 2964 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2965 | { |
2966 | struct sched_domain *sd; | |
dd41f596 IM |
2967 | int pulled_task = -1; |
2968 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2969 | |
2970 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2971 | unsigned long interval; |
2972 | ||
2973 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2974 | continue; | |
2975 | ||
2976 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2977 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2978 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2979 | this_rq, sd); |
2980 | ||
2981 | interval = msecs_to_jiffies(sd->balance_interval); | |
2982 | if (time_after(next_balance, sd->last_balance + interval)) | |
2983 | next_balance = sd->last_balance + interval; | |
2984 | if (pulled_task) | |
2985 | break; | |
1da177e4 | 2986 | } |
dd41f596 | 2987 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2988 | /* |
2989 | * We are going idle. next_balance may be set based on | |
2990 | * a busy processor. So reset next_balance. | |
2991 | */ | |
2992 | this_rq->next_balance = next_balance; | |
dd41f596 | 2993 | } |
1da177e4 LT |
2994 | } |
2995 | ||
2996 | /* | |
2997 | * active_load_balance is run by migration threads. It pushes running tasks | |
2998 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2999 | * running on each physical CPU where possible, and avoids physical / | |
3000 | * logical imbalances. | |
3001 | * | |
3002 | * Called with busiest_rq locked. | |
3003 | */ | |
70b97a7f | 3004 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3005 | { |
39507451 | 3006 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3007 | struct sched_domain *sd; |
3008 | struct rq *target_rq; | |
39507451 | 3009 | |
48f24c4d | 3010 | /* Is there any task to move? */ |
39507451 | 3011 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3012 | return; |
3013 | ||
3014 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3015 | |
3016 | /* | |
39507451 | 3017 | * This condition is "impossible", if it occurs |
41a2d6cf | 3018 | * we need to fix it. Originally reported by |
39507451 | 3019 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3020 | */ |
39507451 | 3021 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3022 | |
39507451 NP |
3023 | /* move a task from busiest_rq to target_rq */ |
3024 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3025 | update_rq_clock(busiest_rq); |
3026 | update_rq_clock(target_rq); | |
39507451 NP |
3027 | |
3028 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3029 | for_each_domain(target_cpu, sd) { |
39507451 | 3030 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3031 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3032 | break; |
c96d145e | 3033 | } |
39507451 | 3034 | |
48f24c4d | 3035 | if (likely(sd)) { |
2d72376b | 3036 | schedstat_inc(sd, alb_count); |
39507451 | 3037 | |
43010659 PW |
3038 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3039 | sd, CPU_IDLE)) | |
48f24c4d IM |
3040 | schedstat_inc(sd, alb_pushed); |
3041 | else | |
3042 | schedstat_inc(sd, alb_failed); | |
3043 | } | |
39507451 | 3044 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3045 | } |
3046 | ||
46cb4b7c SS |
3047 | #ifdef CONFIG_NO_HZ |
3048 | static struct { | |
3049 | atomic_t load_balancer; | |
41a2d6cf | 3050 | cpumask_t cpu_mask; |
46cb4b7c SS |
3051 | } nohz ____cacheline_aligned = { |
3052 | .load_balancer = ATOMIC_INIT(-1), | |
3053 | .cpu_mask = CPU_MASK_NONE, | |
3054 | }; | |
3055 | ||
7835b98b | 3056 | /* |
46cb4b7c SS |
3057 | * This routine will try to nominate the ilb (idle load balancing) |
3058 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3059 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3060 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3061 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3062 | * arrives... | |
3063 | * | |
3064 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3065 | * for idle load balancing. ilb owner will still be part of | |
3066 | * nohz.cpu_mask.. | |
7835b98b | 3067 | * |
46cb4b7c SS |
3068 | * While stopping the tick, this cpu will become the ilb owner if there |
3069 | * is no other owner. And will be the owner till that cpu becomes busy | |
3070 | * or if all cpus in the system stop their ticks at which point | |
3071 | * there is no need for ilb owner. | |
3072 | * | |
3073 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3074 | * next busy scheduler_tick() | |
3075 | */ | |
3076 | int select_nohz_load_balancer(int stop_tick) | |
3077 | { | |
3078 | int cpu = smp_processor_id(); | |
3079 | ||
3080 | if (stop_tick) { | |
3081 | cpu_set(cpu, nohz.cpu_mask); | |
3082 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3083 | ||
3084 | /* | |
3085 | * If we are going offline and still the leader, give up! | |
3086 | */ | |
3087 | if (cpu_is_offline(cpu) && | |
3088 | atomic_read(&nohz.load_balancer) == cpu) { | |
3089 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3090 | BUG(); | |
3091 | return 0; | |
3092 | } | |
3093 | ||
3094 | /* time for ilb owner also to sleep */ | |
3095 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3096 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3097 | atomic_set(&nohz.load_balancer, -1); | |
3098 | return 0; | |
3099 | } | |
3100 | ||
3101 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3102 | /* make me the ilb owner */ | |
3103 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3104 | return 1; | |
3105 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3106 | return 1; | |
3107 | } else { | |
3108 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3109 | return 0; | |
3110 | ||
3111 | cpu_clear(cpu, nohz.cpu_mask); | |
3112 | ||
3113 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3114 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3115 | BUG(); | |
3116 | } | |
3117 | return 0; | |
3118 | } | |
3119 | #endif | |
3120 | ||
3121 | static DEFINE_SPINLOCK(balancing); | |
3122 | ||
3123 | /* | |
7835b98b CL |
3124 | * It checks each scheduling domain to see if it is due to be balanced, |
3125 | * and initiates a balancing operation if so. | |
3126 | * | |
3127 | * Balancing parameters are set up in arch_init_sched_domains. | |
3128 | */ | |
a9957449 | 3129 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3130 | { |
46cb4b7c SS |
3131 | int balance = 1; |
3132 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3133 | unsigned long interval; |
3134 | struct sched_domain *sd; | |
46cb4b7c | 3135 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3136 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3137 | int update_next_balance = 0; |
1da177e4 | 3138 | |
46cb4b7c | 3139 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3140 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3141 | continue; | |
3142 | ||
3143 | interval = sd->balance_interval; | |
d15bcfdb | 3144 | if (idle != CPU_IDLE) |
1da177e4 LT |
3145 | interval *= sd->busy_factor; |
3146 | ||
3147 | /* scale ms to jiffies */ | |
3148 | interval = msecs_to_jiffies(interval); | |
3149 | if (unlikely(!interval)) | |
3150 | interval = 1; | |
dd41f596 IM |
3151 | if (interval > HZ*NR_CPUS/10) |
3152 | interval = HZ*NR_CPUS/10; | |
3153 | ||
1da177e4 | 3154 | |
08c183f3 CL |
3155 | if (sd->flags & SD_SERIALIZE) { |
3156 | if (!spin_trylock(&balancing)) | |
3157 | goto out; | |
3158 | } | |
3159 | ||
c9819f45 | 3160 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3161 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3162 | /* |
3163 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3164 | * longer idle, or one of our SMT siblings is |
3165 | * not idle. | |
3166 | */ | |
d15bcfdb | 3167 | idle = CPU_NOT_IDLE; |
1da177e4 | 3168 | } |
1bd77f2d | 3169 | sd->last_balance = jiffies; |
1da177e4 | 3170 | } |
08c183f3 CL |
3171 | if (sd->flags & SD_SERIALIZE) |
3172 | spin_unlock(&balancing); | |
3173 | out: | |
f549da84 | 3174 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3175 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3176 | update_next_balance = 1; |
3177 | } | |
783609c6 SS |
3178 | |
3179 | /* | |
3180 | * Stop the load balance at this level. There is another | |
3181 | * CPU in our sched group which is doing load balancing more | |
3182 | * actively. | |
3183 | */ | |
3184 | if (!balance) | |
3185 | break; | |
1da177e4 | 3186 | } |
f549da84 SS |
3187 | |
3188 | /* | |
3189 | * next_balance will be updated only when there is a need. | |
3190 | * When the cpu is attached to null domain for ex, it will not be | |
3191 | * updated. | |
3192 | */ | |
3193 | if (likely(update_next_balance)) | |
3194 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3195 | } |
3196 | ||
3197 | /* | |
3198 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3199 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3200 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3201 | */ | |
3202 | static void run_rebalance_domains(struct softirq_action *h) | |
3203 | { | |
dd41f596 IM |
3204 | int this_cpu = smp_processor_id(); |
3205 | struct rq *this_rq = cpu_rq(this_cpu); | |
3206 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3207 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3208 | |
dd41f596 | 3209 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3210 | |
3211 | #ifdef CONFIG_NO_HZ | |
3212 | /* | |
3213 | * If this cpu is the owner for idle load balancing, then do the | |
3214 | * balancing on behalf of the other idle cpus whose ticks are | |
3215 | * stopped. | |
3216 | */ | |
dd41f596 IM |
3217 | if (this_rq->idle_at_tick && |
3218 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3219 | cpumask_t cpus = nohz.cpu_mask; |
3220 | struct rq *rq; | |
3221 | int balance_cpu; | |
3222 | ||
dd41f596 | 3223 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3224 | for_each_cpu_mask(balance_cpu, cpus) { |
3225 | /* | |
3226 | * If this cpu gets work to do, stop the load balancing | |
3227 | * work being done for other cpus. Next load | |
3228 | * balancing owner will pick it up. | |
3229 | */ | |
3230 | if (need_resched()) | |
3231 | break; | |
3232 | ||
de0cf899 | 3233 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3234 | |
3235 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3236 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3237 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3238 | } |
3239 | } | |
3240 | #endif | |
3241 | } | |
3242 | ||
3243 | /* | |
3244 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3245 | * | |
3246 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3247 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3248 | * if the whole system is idle. | |
3249 | */ | |
dd41f596 | 3250 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3251 | { |
46cb4b7c SS |
3252 | #ifdef CONFIG_NO_HZ |
3253 | /* | |
3254 | * If we were in the nohz mode recently and busy at the current | |
3255 | * scheduler tick, then check if we need to nominate new idle | |
3256 | * load balancer. | |
3257 | */ | |
3258 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3259 | rq->in_nohz_recently = 0; | |
3260 | ||
3261 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3262 | cpu_clear(cpu, nohz.cpu_mask); | |
3263 | atomic_set(&nohz.load_balancer, -1); | |
3264 | } | |
3265 | ||
3266 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3267 | /* | |
3268 | * simple selection for now: Nominate the | |
3269 | * first cpu in the nohz list to be the next | |
3270 | * ilb owner. | |
3271 | * | |
3272 | * TBD: Traverse the sched domains and nominate | |
3273 | * the nearest cpu in the nohz.cpu_mask. | |
3274 | */ | |
3275 | int ilb = first_cpu(nohz.cpu_mask); | |
3276 | ||
3277 | if (ilb != NR_CPUS) | |
3278 | resched_cpu(ilb); | |
3279 | } | |
3280 | } | |
3281 | ||
3282 | /* | |
3283 | * If this cpu is idle and doing idle load balancing for all the | |
3284 | * cpus with ticks stopped, is it time for that to stop? | |
3285 | */ | |
3286 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3287 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3288 | resched_cpu(cpu); | |
3289 | return; | |
3290 | } | |
3291 | ||
3292 | /* | |
3293 | * If this cpu is idle and the idle load balancing is done by | |
3294 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3295 | */ | |
3296 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3297 | cpu_isset(cpu, nohz.cpu_mask)) | |
3298 | return; | |
3299 | #endif | |
3300 | if (time_after_eq(jiffies, rq->next_balance)) | |
3301 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3302 | } |
dd41f596 IM |
3303 | |
3304 | #else /* CONFIG_SMP */ | |
3305 | ||
1da177e4 LT |
3306 | /* |
3307 | * on UP we do not need to balance between CPUs: | |
3308 | */ | |
70b97a7f | 3309 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3310 | { |
3311 | } | |
dd41f596 | 3312 | |
1da177e4 LT |
3313 | #endif |
3314 | ||
1da177e4 LT |
3315 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3316 | ||
3317 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3318 | ||
3319 | /* | |
41b86e9c IM |
3320 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3321 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3322 | */ |
41b86e9c | 3323 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3324 | { |
1da177e4 | 3325 | unsigned long flags; |
41b86e9c IM |
3326 | u64 ns, delta_exec; |
3327 | struct rq *rq; | |
48f24c4d | 3328 | |
41b86e9c IM |
3329 | rq = task_rq_lock(p, &flags); |
3330 | ns = p->se.sum_exec_runtime; | |
3331 | if (rq->curr == p) { | |
a8e504d2 IM |
3332 | update_rq_clock(rq); |
3333 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3334 | if ((s64)delta_exec > 0) |
3335 | ns += delta_exec; | |
3336 | } | |
3337 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3338 | |
1da177e4 LT |
3339 | return ns; |
3340 | } | |
3341 | ||
1da177e4 LT |
3342 | /* |
3343 | * Account user cpu time to a process. | |
3344 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3345 | * @cputime: the cpu time spent in user space since the last update |
3346 | */ | |
3347 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3348 | { | |
3349 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3350 | cputime64_t tmp; | |
3351 | ||
3352 | p->utime = cputime_add(p->utime, cputime); | |
3353 | ||
3354 | /* Add user time to cpustat. */ | |
3355 | tmp = cputime_to_cputime64(cputime); | |
3356 | if (TASK_NICE(p) > 0) | |
3357 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3358 | else | |
3359 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3360 | } | |
3361 | ||
94886b84 LV |
3362 | /* |
3363 | * Account guest cpu time to a process. | |
3364 | * @p: the process that the cpu time gets accounted to | |
3365 | * @cputime: the cpu time spent in virtual machine since the last update | |
3366 | */ | |
f7402e03 | 3367 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3368 | { |
3369 | cputime64_t tmp; | |
3370 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3371 | ||
3372 | tmp = cputime_to_cputime64(cputime); | |
3373 | ||
3374 | p->utime = cputime_add(p->utime, cputime); | |
3375 | p->gtime = cputime_add(p->gtime, cputime); | |
3376 | ||
3377 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3378 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3379 | } | |
3380 | ||
c66f08be MN |
3381 | /* |
3382 | * Account scaled user cpu time to a process. | |
3383 | * @p: the process that the cpu time gets accounted to | |
3384 | * @cputime: the cpu time spent in user space since the last update | |
3385 | */ | |
3386 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3387 | { | |
3388 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3389 | } | |
3390 | ||
1da177e4 LT |
3391 | /* |
3392 | * Account system cpu time to a process. | |
3393 | * @p: the process that the cpu time gets accounted to | |
3394 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3395 | * @cputime: the cpu time spent in kernel space since the last update | |
3396 | */ | |
3397 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3398 | cputime_t cputime) | |
3399 | { | |
3400 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3401 | struct rq *rq = this_rq(); |
1da177e4 LT |
3402 | cputime64_t tmp; |
3403 | ||
9778385d CB |
3404 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3405 | return account_guest_time(p, cputime); | |
94886b84 | 3406 | |
1da177e4 LT |
3407 | p->stime = cputime_add(p->stime, cputime); |
3408 | ||
3409 | /* Add system time to cpustat. */ | |
3410 | tmp = cputime_to_cputime64(cputime); | |
3411 | if (hardirq_count() - hardirq_offset) | |
3412 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3413 | else if (softirq_count()) | |
3414 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3415 | else if (p != rq->idle) |
1da177e4 | 3416 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3417 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3418 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3419 | else | |
3420 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3421 | /* Account for system time used */ | |
3422 | acct_update_integrals(p); | |
1da177e4 LT |
3423 | } |
3424 | ||
c66f08be MN |
3425 | /* |
3426 | * Account scaled system cpu time to a process. | |
3427 | * @p: the process that the cpu time gets accounted to | |
3428 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3429 | * @cputime: the cpu time spent in kernel space since the last update | |
3430 | */ | |
3431 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3432 | { | |
3433 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3434 | } | |
3435 | ||
1da177e4 LT |
3436 | /* |
3437 | * Account for involuntary wait time. | |
3438 | * @p: the process from which the cpu time has been stolen | |
3439 | * @steal: the cpu time spent in involuntary wait | |
3440 | */ | |
3441 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3442 | { | |
3443 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3444 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3445 | struct rq *rq = this_rq(); |
1da177e4 LT |
3446 | |
3447 | if (p == rq->idle) { | |
3448 | p->stime = cputime_add(p->stime, steal); | |
3449 | if (atomic_read(&rq->nr_iowait) > 0) | |
3450 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3451 | else | |
3452 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3453 | } else |
1da177e4 LT |
3454 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3455 | } | |
3456 | ||
7835b98b CL |
3457 | /* |
3458 | * This function gets called by the timer code, with HZ frequency. | |
3459 | * We call it with interrupts disabled. | |
3460 | * | |
3461 | * It also gets called by the fork code, when changing the parent's | |
3462 | * timeslices. | |
3463 | */ | |
3464 | void scheduler_tick(void) | |
3465 | { | |
7835b98b CL |
3466 | int cpu = smp_processor_id(); |
3467 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3468 | struct task_struct *curr = rq->curr; |
529c7726 | 3469 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3470 | |
3471 | spin_lock(&rq->lock); | |
546fe3c9 | 3472 | __update_rq_clock(rq); |
529c7726 IM |
3473 | /* |
3474 | * Let rq->clock advance by at least TICK_NSEC: | |
3475 | */ | |
3476 | if (unlikely(rq->clock < next_tick)) | |
3477 | rq->clock = next_tick; | |
3478 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3479 | update_cpu_load(rq); |
dd41f596 IM |
3480 | if (curr != rq->idle) /* FIXME: needed? */ |
3481 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3482 | spin_unlock(&rq->lock); |
7835b98b | 3483 | |
e418e1c2 | 3484 | #ifdef CONFIG_SMP |
dd41f596 IM |
3485 | rq->idle_at_tick = idle_cpu(cpu); |
3486 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3487 | #endif |
1da177e4 LT |
3488 | } |
3489 | ||
1da177e4 LT |
3490 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3491 | ||
3492 | void fastcall add_preempt_count(int val) | |
3493 | { | |
3494 | /* | |
3495 | * Underflow? | |
3496 | */ | |
9a11b49a IM |
3497 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3498 | return; | |
1da177e4 LT |
3499 | preempt_count() += val; |
3500 | /* | |
3501 | * Spinlock count overflowing soon? | |
3502 | */ | |
33859f7f MOS |
3503 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3504 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3505 | } |
3506 | EXPORT_SYMBOL(add_preempt_count); | |
3507 | ||
3508 | void fastcall sub_preempt_count(int val) | |
3509 | { | |
3510 | /* | |
3511 | * Underflow? | |
3512 | */ | |
9a11b49a IM |
3513 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3514 | return; | |
1da177e4 LT |
3515 | /* |
3516 | * Is the spinlock portion underflowing? | |
3517 | */ | |
9a11b49a IM |
3518 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3519 | !(preempt_count() & PREEMPT_MASK))) | |
3520 | return; | |
3521 | ||
1da177e4 LT |
3522 | preempt_count() -= val; |
3523 | } | |
3524 | EXPORT_SYMBOL(sub_preempt_count); | |
3525 | ||
3526 | #endif | |
3527 | ||
3528 | /* | |
dd41f596 | 3529 | * Print scheduling while atomic bug: |
1da177e4 | 3530 | */ |
dd41f596 | 3531 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3532 | { |
838225b4 SS |
3533 | struct pt_regs *regs = get_irq_regs(); |
3534 | ||
3535 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3536 | prev->comm, prev->pid, preempt_count()); | |
3537 | ||
dd41f596 IM |
3538 | debug_show_held_locks(prev); |
3539 | if (irqs_disabled()) | |
3540 | print_irqtrace_events(prev); | |
838225b4 SS |
3541 | |
3542 | if (regs) | |
3543 | show_regs(regs); | |
3544 | else | |
3545 | dump_stack(); | |
dd41f596 | 3546 | } |
1da177e4 | 3547 | |
dd41f596 IM |
3548 | /* |
3549 | * Various schedule()-time debugging checks and statistics: | |
3550 | */ | |
3551 | static inline void schedule_debug(struct task_struct *prev) | |
3552 | { | |
1da177e4 | 3553 | /* |
41a2d6cf | 3554 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3555 | * schedule() atomically, we ignore that path for now. |
3556 | * Otherwise, whine if we are scheduling when we should not be. | |
3557 | */ | |
dd41f596 IM |
3558 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3559 | __schedule_bug(prev); | |
3560 | ||
1da177e4 LT |
3561 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3562 | ||
2d72376b | 3563 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3564 | #ifdef CONFIG_SCHEDSTATS |
3565 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3566 | schedstat_inc(this_rq(), bkl_count); |
3567 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3568 | } |
3569 | #endif | |
dd41f596 IM |
3570 | } |
3571 | ||
3572 | /* | |
3573 | * Pick up the highest-prio task: | |
3574 | */ | |
3575 | static inline struct task_struct * | |
ff95f3df | 3576 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3577 | { |
5522d5d5 | 3578 | const struct sched_class *class; |
dd41f596 | 3579 | struct task_struct *p; |
1da177e4 LT |
3580 | |
3581 | /* | |
dd41f596 IM |
3582 | * Optimization: we know that if all tasks are in |
3583 | * the fair class we can call that function directly: | |
1da177e4 | 3584 | */ |
dd41f596 | 3585 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3586 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3587 | if (likely(p)) |
3588 | return p; | |
1da177e4 LT |
3589 | } |
3590 | ||
dd41f596 IM |
3591 | class = sched_class_highest; |
3592 | for ( ; ; ) { | |
fb8d4724 | 3593 | p = class->pick_next_task(rq); |
dd41f596 IM |
3594 | if (p) |
3595 | return p; | |
3596 | /* | |
3597 | * Will never be NULL as the idle class always | |
3598 | * returns a non-NULL p: | |
3599 | */ | |
3600 | class = class->next; | |
3601 | } | |
3602 | } | |
1da177e4 | 3603 | |
dd41f596 IM |
3604 | /* |
3605 | * schedule() is the main scheduler function. | |
3606 | */ | |
3607 | asmlinkage void __sched schedule(void) | |
3608 | { | |
3609 | struct task_struct *prev, *next; | |
3610 | long *switch_count; | |
3611 | struct rq *rq; | |
dd41f596 IM |
3612 | int cpu; |
3613 | ||
3614 | need_resched: | |
3615 | preempt_disable(); | |
3616 | cpu = smp_processor_id(); | |
3617 | rq = cpu_rq(cpu); | |
3618 | rcu_qsctr_inc(cpu); | |
3619 | prev = rq->curr; | |
3620 | switch_count = &prev->nivcsw; | |
3621 | ||
3622 | release_kernel_lock(prev); | |
3623 | need_resched_nonpreemptible: | |
3624 | ||
3625 | schedule_debug(prev); | |
1da177e4 | 3626 | |
1e819950 IM |
3627 | /* |
3628 | * Do the rq-clock update outside the rq lock: | |
3629 | */ | |
3630 | local_irq_disable(); | |
c1b3da3e | 3631 | __update_rq_clock(rq); |
1e819950 IM |
3632 | spin_lock(&rq->lock); |
3633 | clear_tsk_need_resched(prev); | |
1da177e4 | 3634 | |
1da177e4 | 3635 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3636 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3637 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3638 | prev->state = TASK_RUNNING; |
dd41f596 | 3639 | } else { |
2e1cb74a | 3640 | deactivate_task(rq, prev, 1); |
1da177e4 | 3641 | } |
dd41f596 | 3642 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3643 | } |
3644 | ||
dd41f596 | 3645 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3646 | idle_balance(cpu, rq); |
1da177e4 | 3647 | |
31ee529c | 3648 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3649 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3650 | |
3651 | sched_info_switch(prev, next); | |
dd41f596 | 3652 | |
1da177e4 | 3653 | if (likely(prev != next)) { |
1da177e4 LT |
3654 | rq->nr_switches++; |
3655 | rq->curr = next; | |
3656 | ++*switch_count; | |
3657 | ||
dd41f596 | 3658 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3659 | } else |
3660 | spin_unlock_irq(&rq->lock); | |
3661 | ||
dd41f596 IM |
3662 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3663 | cpu = smp_processor_id(); | |
3664 | rq = cpu_rq(cpu); | |
1da177e4 | 3665 | goto need_resched_nonpreemptible; |
dd41f596 | 3666 | } |
1da177e4 LT |
3667 | preempt_enable_no_resched(); |
3668 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3669 | goto need_resched; | |
3670 | } | |
1da177e4 LT |
3671 | EXPORT_SYMBOL(schedule); |
3672 | ||
3673 | #ifdef CONFIG_PREEMPT | |
3674 | /* | |
2ed6e34f | 3675 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3676 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3677 | * occur there and call schedule directly. |
3678 | */ | |
3679 | asmlinkage void __sched preempt_schedule(void) | |
3680 | { | |
3681 | struct thread_info *ti = current_thread_info(); | |
3682 | #ifdef CONFIG_PREEMPT_BKL | |
3683 | struct task_struct *task = current; | |
3684 | int saved_lock_depth; | |
3685 | #endif | |
3686 | /* | |
3687 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3688 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3689 | */ |
beed33a8 | 3690 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3691 | return; |
3692 | ||
3a5c359a AK |
3693 | do { |
3694 | add_preempt_count(PREEMPT_ACTIVE); | |
3695 | ||
3696 | /* | |
3697 | * We keep the big kernel semaphore locked, but we | |
3698 | * clear ->lock_depth so that schedule() doesnt | |
3699 | * auto-release the semaphore: | |
3700 | */ | |
1da177e4 | 3701 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3702 | saved_lock_depth = task->lock_depth; |
3703 | task->lock_depth = -1; | |
1da177e4 | 3704 | #endif |
3a5c359a | 3705 | schedule(); |
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 | 3717 | } |
1da177e4 LT |
3718 | EXPORT_SYMBOL(preempt_schedule); |
3719 | ||
3720 | /* | |
2ed6e34f | 3721 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3722 | * off of irq context. |
3723 | * Note, that this is called and return with irqs disabled. This will | |
3724 | * protect us against recursive calling from irq. | |
3725 | */ | |
3726 | asmlinkage void __sched preempt_schedule_irq(void) | |
3727 | { | |
3728 | struct thread_info *ti = current_thread_info(); | |
3729 | #ifdef CONFIG_PREEMPT_BKL | |
3730 | struct task_struct *task = current; | |
3731 | int saved_lock_depth; | |
3732 | #endif | |
2ed6e34f | 3733 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3734 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3735 | ||
3a5c359a AK |
3736 | do { |
3737 | add_preempt_count(PREEMPT_ACTIVE); | |
3738 | ||
3739 | /* | |
3740 | * We keep the big kernel semaphore locked, but we | |
3741 | * clear ->lock_depth so that schedule() doesnt | |
3742 | * auto-release the semaphore: | |
3743 | */ | |
1da177e4 | 3744 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3745 | saved_lock_depth = task->lock_depth; |
3746 | task->lock_depth = -1; | |
1da177e4 | 3747 | #endif |
3a5c359a AK |
3748 | local_irq_enable(); |
3749 | schedule(); | |
3750 | local_irq_disable(); | |
1da177e4 | 3751 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3752 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3753 | #endif |
3a5c359a | 3754 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3755 | |
3a5c359a AK |
3756 | /* |
3757 | * Check again in case we missed a preemption opportunity | |
3758 | * between schedule and now. | |
3759 | */ | |
3760 | barrier(); | |
3761 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3762 | } |
3763 | ||
3764 | #endif /* CONFIG_PREEMPT */ | |
3765 | ||
95cdf3b7 IM |
3766 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3767 | void *key) | |
1da177e4 | 3768 | { |
48f24c4d | 3769 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3770 | } |
1da177e4 LT |
3771 | EXPORT_SYMBOL(default_wake_function); |
3772 | ||
3773 | /* | |
41a2d6cf IM |
3774 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3775 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3776 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3777 | * | |
3778 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3779 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3780 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3781 | */ | |
3782 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3783 | int nr_exclusive, int sync, void *key) | |
3784 | { | |
2e45874c | 3785 | wait_queue_t *curr, *next; |
1da177e4 | 3786 | |
2e45874c | 3787 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3788 | unsigned flags = curr->flags; |
3789 | ||
1da177e4 | 3790 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3791 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3792 | break; |
3793 | } | |
3794 | } | |
3795 | ||
3796 | /** | |
3797 | * __wake_up - wake up threads blocked on a waitqueue. | |
3798 | * @q: the waitqueue | |
3799 | * @mode: which threads | |
3800 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3801 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3802 | */ |
3803 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3804 | int nr_exclusive, void *key) |
1da177e4 LT |
3805 | { |
3806 | unsigned long flags; | |
3807 | ||
3808 | spin_lock_irqsave(&q->lock, flags); | |
3809 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3810 | spin_unlock_irqrestore(&q->lock, flags); | |
3811 | } | |
1da177e4 LT |
3812 | EXPORT_SYMBOL(__wake_up); |
3813 | ||
3814 | /* | |
3815 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3816 | */ | |
3817 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3818 | { | |
3819 | __wake_up_common(q, mode, 1, 0, NULL); | |
3820 | } | |
3821 | ||
3822 | /** | |
67be2dd1 | 3823 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3824 | * @q: the waitqueue |
3825 | * @mode: which threads | |
3826 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3827 | * | |
3828 | * The sync wakeup differs that the waker knows that it will schedule | |
3829 | * away soon, so while the target thread will be woken up, it will not | |
3830 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3831 | * with each other. This can prevent needless bouncing between CPUs. | |
3832 | * | |
3833 | * On UP it can prevent extra preemption. | |
3834 | */ | |
95cdf3b7 IM |
3835 | void fastcall |
3836 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3837 | { |
3838 | unsigned long flags; | |
3839 | int sync = 1; | |
3840 | ||
3841 | if (unlikely(!q)) | |
3842 | return; | |
3843 | ||
3844 | if (unlikely(!nr_exclusive)) | |
3845 | sync = 0; | |
3846 | ||
3847 | spin_lock_irqsave(&q->lock, flags); | |
3848 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3849 | spin_unlock_irqrestore(&q->lock, flags); | |
3850 | } | |
3851 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3852 | ||
b15136e9 | 3853 | void complete(struct completion *x) |
1da177e4 LT |
3854 | { |
3855 | unsigned long flags; | |
3856 | ||
3857 | spin_lock_irqsave(&x->wait.lock, flags); | |
3858 | x->done++; | |
d9514f6c | 3859 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3860 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3861 | } | |
3862 | EXPORT_SYMBOL(complete); | |
3863 | ||
b15136e9 | 3864 | void complete_all(struct completion *x) |
1da177e4 LT |
3865 | { |
3866 | unsigned long flags; | |
3867 | ||
3868 | spin_lock_irqsave(&x->wait.lock, flags); | |
3869 | x->done += UINT_MAX/2; | |
d9514f6c | 3870 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3871 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3872 | } | |
3873 | EXPORT_SYMBOL(complete_all); | |
3874 | ||
8cbbe86d AK |
3875 | static inline long __sched |
3876 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3877 | { |
1da177e4 LT |
3878 | if (!x->done) { |
3879 | DECLARE_WAITQUEUE(wait, current); | |
3880 | ||
3881 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3882 | __add_wait_queue_tail(&x->wait, &wait); | |
3883 | do { | |
8cbbe86d AK |
3884 | if (state == TASK_INTERRUPTIBLE && |
3885 | signal_pending(current)) { | |
3886 | __remove_wait_queue(&x->wait, &wait); | |
3887 | return -ERESTARTSYS; | |
3888 | } | |
3889 | __set_current_state(state); | |
1da177e4 LT |
3890 | spin_unlock_irq(&x->wait.lock); |
3891 | timeout = schedule_timeout(timeout); | |
3892 | spin_lock_irq(&x->wait.lock); | |
3893 | if (!timeout) { | |
3894 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3895 | return timeout; |
1da177e4 LT |
3896 | } |
3897 | } while (!x->done); | |
3898 | __remove_wait_queue(&x->wait, &wait); | |
3899 | } | |
3900 | x->done--; | |
1da177e4 LT |
3901 | return timeout; |
3902 | } | |
1da177e4 | 3903 | |
8cbbe86d AK |
3904 | static long __sched |
3905 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3906 | { |
1da177e4 LT |
3907 | might_sleep(); |
3908 | ||
3909 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3910 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3911 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3912 | return timeout; |
3913 | } | |
1da177e4 | 3914 | |
b15136e9 | 3915 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3916 | { |
3917 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3918 | } |
8cbbe86d | 3919 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3920 | |
b15136e9 | 3921 | unsigned long __sched |
8cbbe86d | 3922 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3923 | { |
8cbbe86d | 3924 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3925 | } |
8cbbe86d | 3926 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3927 | |
8cbbe86d | 3928 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3929 | { |
51e97990 AK |
3930 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3931 | if (t == -ERESTARTSYS) | |
3932 | return t; | |
3933 | return 0; | |
0fec171c | 3934 | } |
8cbbe86d | 3935 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3936 | |
b15136e9 | 3937 | unsigned long __sched |
8cbbe86d AK |
3938 | wait_for_completion_interruptible_timeout(struct completion *x, |
3939 | unsigned long timeout) | |
0fec171c | 3940 | { |
8cbbe86d | 3941 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3942 | } |
8cbbe86d | 3943 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3944 | |
8cbbe86d AK |
3945 | static long __sched |
3946 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3947 | { |
0fec171c IM |
3948 | unsigned long flags; |
3949 | wait_queue_t wait; | |
3950 | ||
3951 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3952 | |
8cbbe86d | 3953 | __set_current_state(state); |
1da177e4 | 3954 | |
8cbbe86d AK |
3955 | spin_lock_irqsave(&q->lock, flags); |
3956 | __add_wait_queue(q, &wait); | |
3957 | spin_unlock(&q->lock); | |
3958 | timeout = schedule_timeout(timeout); | |
3959 | spin_lock_irq(&q->lock); | |
3960 | __remove_wait_queue(q, &wait); | |
3961 | spin_unlock_irqrestore(&q->lock, flags); | |
3962 | ||
3963 | return timeout; | |
3964 | } | |
3965 | ||
3966 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3967 | { | |
3968 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3969 | } |
1da177e4 LT |
3970 | EXPORT_SYMBOL(interruptible_sleep_on); |
3971 | ||
0fec171c | 3972 | long __sched |
95cdf3b7 | 3973 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3974 | { |
8cbbe86d | 3975 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3976 | } |
1da177e4 LT |
3977 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3978 | ||
0fec171c | 3979 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3980 | { |
8cbbe86d | 3981 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3982 | } |
1da177e4 LT |
3983 | EXPORT_SYMBOL(sleep_on); |
3984 | ||
0fec171c | 3985 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3986 | { |
8cbbe86d | 3987 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3988 | } |
1da177e4 LT |
3989 | EXPORT_SYMBOL(sleep_on_timeout); |
3990 | ||
b29739f9 IM |
3991 | #ifdef CONFIG_RT_MUTEXES |
3992 | ||
3993 | /* | |
3994 | * rt_mutex_setprio - set the current priority of a task | |
3995 | * @p: task | |
3996 | * @prio: prio value (kernel-internal form) | |
3997 | * | |
3998 | * This function changes the 'effective' priority of a task. It does | |
3999 | * not touch ->normal_prio like __setscheduler(). | |
4000 | * | |
4001 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4002 | */ | |
36c8b586 | 4003 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4004 | { |
4005 | unsigned long flags; | |
83b699ed | 4006 | int oldprio, on_rq, running; |
70b97a7f | 4007 | struct rq *rq; |
b29739f9 IM |
4008 | |
4009 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4010 | ||
4011 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4012 | update_rq_clock(rq); |
b29739f9 | 4013 | |
d5f9f942 | 4014 | oldprio = p->prio; |
dd41f596 | 4015 | on_rq = p->se.on_rq; |
83b699ed SV |
4016 | running = task_running(rq, p); |
4017 | if (on_rq) { | |
69be72c1 | 4018 | dequeue_task(rq, p, 0); |
83b699ed SV |
4019 | if (running) |
4020 | p->sched_class->put_prev_task(rq, p); | |
4021 | } | |
dd41f596 IM |
4022 | |
4023 | if (rt_prio(prio)) | |
4024 | p->sched_class = &rt_sched_class; | |
4025 | else | |
4026 | p->sched_class = &fair_sched_class; | |
4027 | ||
b29739f9 IM |
4028 | p->prio = prio; |
4029 | ||
dd41f596 | 4030 | if (on_rq) { |
83b699ed SV |
4031 | if (running) |
4032 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4033 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4034 | /* |
4035 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4036 | * our priority decreased, or if we are not currently running on |
4037 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4038 | */ |
83b699ed | 4039 | if (running) { |
d5f9f942 AM |
4040 | if (p->prio > oldprio) |
4041 | resched_task(rq->curr); | |
dd41f596 IM |
4042 | } else { |
4043 | check_preempt_curr(rq, p); | |
4044 | } | |
b29739f9 IM |
4045 | } |
4046 | task_rq_unlock(rq, &flags); | |
4047 | } | |
4048 | ||
4049 | #endif | |
4050 | ||
36c8b586 | 4051 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4052 | { |
dd41f596 | 4053 | int old_prio, delta, on_rq; |
1da177e4 | 4054 | unsigned long flags; |
70b97a7f | 4055 | struct rq *rq; |
1da177e4 LT |
4056 | |
4057 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4058 | return; | |
4059 | /* | |
4060 | * We have to be careful, if called from sys_setpriority(), | |
4061 | * the task might be in the middle of scheduling on another CPU. | |
4062 | */ | |
4063 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4064 | update_rq_clock(rq); |
1da177e4 LT |
4065 | /* |
4066 | * The RT priorities are set via sched_setscheduler(), but we still | |
4067 | * allow the 'normal' nice value to be set - but as expected | |
4068 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4069 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4070 | */ |
e05606d3 | 4071 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4072 | p->static_prio = NICE_TO_PRIO(nice); |
4073 | goto out_unlock; | |
4074 | } | |
dd41f596 IM |
4075 | on_rq = p->se.on_rq; |
4076 | if (on_rq) { | |
69be72c1 | 4077 | dequeue_task(rq, p, 0); |
79b5dddf | 4078 | dec_load(rq, p); |
2dd73a4f | 4079 | } |
1da177e4 | 4080 | |
1da177e4 | 4081 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4082 | set_load_weight(p); |
b29739f9 IM |
4083 | old_prio = p->prio; |
4084 | p->prio = effective_prio(p); | |
4085 | delta = p->prio - old_prio; | |
1da177e4 | 4086 | |
dd41f596 | 4087 | if (on_rq) { |
8159f87e | 4088 | enqueue_task(rq, p, 0); |
29b4b623 | 4089 | inc_load(rq, p); |
1da177e4 | 4090 | /* |
d5f9f942 AM |
4091 | * If the task increased its priority or is running and |
4092 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4093 | */ |
d5f9f942 | 4094 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4095 | resched_task(rq->curr); |
4096 | } | |
4097 | out_unlock: | |
4098 | task_rq_unlock(rq, &flags); | |
4099 | } | |
1da177e4 LT |
4100 | EXPORT_SYMBOL(set_user_nice); |
4101 | ||
e43379f1 MM |
4102 | /* |
4103 | * can_nice - check if a task can reduce its nice value | |
4104 | * @p: task | |
4105 | * @nice: nice value | |
4106 | */ | |
36c8b586 | 4107 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4108 | { |
024f4747 MM |
4109 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4110 | int nice_rlim = 20 - nice; | |
48f24c4d | 4111 | |
e43379f1 MM |
4112 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4113 | capable(CAP_SYS_NICE)); | |
4114 | } | |
4115 | ||
1da177e4 LT |
4116 | #ifdef __ARCH_WANT_SYS_NICE |
4117 | ||
4118 | /* | |
4119 | * sys_nice - change the priority of the current process. | |
4120 | * @increment: priority increment | |
4121 | * | |
4122 | * sys_setpriority is a more generic, but much slower function that | |
4123 | * does similar things. | |
4124 | */ | |
4125 | asmlinkage long sys_nice(int increment) | |
4126 | { | |
48f24c4d | 4127 | long nice, retval; |
1da177e4 LT |
4128 | |
4129 | /* | |
4130 | * Setpriority might change our priority at the same moment. | |
4131 | * We don't have to worry. Conceptually one call occurs first | |
4132 | * and we have a single winner. | |
4133 | */ | |
e43379f1 MM |
4134 | if (increment < -40) |
4135 | increment = -40; | |
1da177e4 LT |
4136 | if (increment > 40) |
4137 | increment = 40; | |
4138 | ||
4139 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4140 | if (nice < -20) | |
4141 | nice = -20; | |
4142 | if (nice > 19) | |
4143 | nice = 19; | |
4144 | ||
e43379f1 MM |
4145 | if (increment < 0 && !can_nice(current, nice)) |
4146 | return -EPERM; | |
4147 | ||
1da177e4 LT |
4148 | retval = security_task_setnice(current, nice); |
4149 | if (retval) | |
4150 | return retval; | |
4151 | ||
4152 | set_user_nice(current, nice); | |
4153 | return 0; | |
4154 | } | |
4155 | ||
4156 | #endif | |
4157 | ||
4158 | /** | |
4159 | * task_prio - return the priority value of a given task. | |
4160 | * @p: the task in question. | |
4161 | * | |
4162 | * This is the priority value as seen by users in /proc. | |
4163 | * RT tasks are offset by -200. Normal tasks are centered | |
4164 | * around 0, value goes from -16 to +15. | |
4165 | */ | |
36c8b586 | 4166 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4167 | { |
4168 | return p->prio - MAX_RT_PRIO; | |
4169 | } | |
4170 | ||
4171 | /** | |
4172 | * task_nice - return the nice value of a given task. | |
4173 | * @p: the task in question. | |
4174 | */ | |
36c8b586 | 4175 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4176 | { |
4177 | return TASK_NICE(p); | |
4178 | } | |
1da177e4 | 4179 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4180 | |
4181 | /** | |
4182 | * idle_cpu - is a given cpu idle currently? | |
4183 | * @cpu: the processor in question. | |
4184 | */ | |
4185 | int idle_cpu(int cpu) | |
4186 | { | |
4187 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4188 | } | |
4189 | ||
1da177e4 LT |
4190 | /** |
4191 | * idle_task - return the idle task for a given cpu. | |
4192 | * @cpu: the processor in question. | |
4193 | */ | |
36c8b586 | 4194 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4195 | { |
4196 | return cpu_rq(cpu)->idle; | |
4197 | } | |
4198 | ||
4199 | /** | |
4200 | * find_process_by_pid - find a process with a matching PID value. | |
4201 | * @pid: the pid in question. | |
4202 | */ | |
a9957449 | 4203 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4204 | { |
228ebcbe | 4205 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4206 | } |
4207 | ||
4208 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4209 | static void |
4210 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4211 | { |
dd41f596 | 4212 | BUG_ON(p->se.on_rq); |
48f24c4d | 4213 | |
1da177e4 | 4214 | p->policy = policy; |
dd41f596 IM |
4215 | switch (p->policy) { |
4216 | case SCHED_NORMAL: | |
4217 | case SCHED_BATCH: | |
4218 | case SCHED_IDLE: | |
4219 | p->sched_class = &fair_sched_class; | |
4220 | break; | |
4221 | case SCHED_FIFO: | |
4222 | case SCHED_RR: | |
4223 | p->sched_class = &rt_sched_class; | |
4224 | break; | |
4225 | } | |
4226 | ||
1da177e4 | 4227 | p->rt_priority = prio; |
b29739f9 IM |
4228 | p->normal_prio = normal_prio(p); |
4229 | /* we are holding p->pi_lock already */ | |
4230 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4231 | set_load_weight(p); |
1da177e4 LT |
4232 | } |
4233 | ||
4234 | /** | |
72fd4a35 | 4235 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4236 | * @p: the task in question. |
4237 | * @policy: new policy. | |
4238 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4239 | * |
72fd4a35 | 4240 | * NOTE that the task may be already dead. |
1da177e4 | 4241 | */ |
95cdf3b7 IM |
4242 | int sched_setscheduler(struct task_struct *p, int policy, |
4243 | struct sched_param *param) | |
1da177e4 | 4244 | { |
83b699ed | 4245 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4246 | unsigned long flags; |
70b97a7f | 4247 | struct rq *rq; |
1da177e4 | 4248 | |
66e5393a SR |
4249 | /* may grab non-irq protected spin_locks */ |
4250 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4251 | recheck: |
4252 | /* double check policy once rq lock held */ | |
4253 | if (policy < 0) | |
4254 | policy = oldpolicy = p->policy; | |
4255 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4256 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4257 | policy != SCHED_IDLE) | |
b0a9499c | 4258 | return -EINVAL; |
1da177e4 LT |
4259 | /* |
4260 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4261 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4262 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4263 | */ |
4264 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4265 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4266 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4267 | return -EINVAL; |
e05606d3 | 4268 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4269 | return -EINVAL; |
4270 | ||
37e4ab3f OC |
4271 | /* |
4272 | * Allow unprivileged RT tasks to decrease priority: | |
4273 | */ | |
4274 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4275 | if (rt_policy(policy)) { |
8dc3e909 | 4276 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4277 | |
4278 | if (!lock_task_sighand(p, &flags)) | |
4279 | return -ESRCH; | |
4280 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4281 | unlock_task_sighand(p, &flags); | |
4282 | ||
4283 | /* can't set/change the rt policy */ | |
4284 | if (policy != p->policy && !rlim_rtprio) | |
4285 | return -EPERM; | |
4286 | ||
4287 | /* can't increase priority */ | |
4288 | if (param->sched_priority > p->rt_priority && | |
4289 | param->sched_priority > rlim_rtprio) | |
4290 | return -EPERM; | |
4291 | } | |
dd41f596 IM |
4292 | /* |
4293 | * Like positive nice levels, dont allow tasks to | |
4294 | * move out of SCHED_IDLE either: | |
4295 | */ | |
4296 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4297 | return -EPERM; | |
5fe1d75f | 4298 | |
37e4ab3f OC |
4299 | /* can't change other user's priorities */ |
4300 | if ((current->euid != p->euid) && | |
4301 | (current->euid != p->uid)) | |
4302 | return -EPERM; | |
4303 | } | |
1da177e4 LT |
4304 | |
4305 | retval = security_task_setscheduler(p, policy, param); | |
4306 | if (retval) | |
4307 | return retval; | |
b29739f9 IM |
4308 | /* |
4309 | * make sure no PI-waiters arrive (or leave) while we are | |
4310 | * changing the priority of the task: | |
4311 | */ | |
4312 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4313 | /* |
4314 | * To be able to change p->policy safely, the apropriate | |
4315 | * runqueue lock must be held. | |
4316 | */ | |
b29739f9 | 4317 | rq = __task_rq_lock(p); |
1da177e4 LT |
4318 | /* recheck policy now with rq lock held */ |
4319 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4320 | policy = oldpolicy = -1; | |
b29739f9 IM |
4321 | __task_rq_unlock(rq); |
4322 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4323 | goto recheck; |
4324 | } | |
2daa3577 | 4325 | update_rq_clock(rq); |
dd41f596 | 4326 | on_rq = p->se.on_rq; |
83b699ed SV |
4327 | running = task_running(rq, p); |
4328 | if (on_rq) { | |
2e1cb74a | 4329 | deactivate_task(rq, p, 0); |
83b699ed SV |
4330 | if (running) |
4331 | p->sched_class->put_prev_task(rq, p); | |
4332 | } | |
f6b53205 | 4333 | |
1da177e4 | 4334 | oldprio = p->prio; |
dd41f596 | 4335 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4336 | |
dd41f596 | 4337 | if (on_rq) { |
83b699ed SV |
4338 | if (running) |
4339 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4340 | activate_task(rq, p, 0); |
1da177e4 LT |
4341 | /* |
4342 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4343 | * our priority decreased, or if we are not currently running on |
4344 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4345 | */ |
83b699ed | 4346 | if (running) { |
d5f9f942 AM |
4347 | if (p->prio > oldprio) |
4348 | resched_task(rq->curr); | |
dd41f596 IM |
4349 | } else { |
4350 | check_preempt_curr(rq, p); | |
4351 | } | |
1da177e4 | 4352 | } |
b29739f9 IM |
4353 | __task_rq_unlock(rq); |
4354 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4355 | ||
95e02ca9 TG |
4356 | rt_mutex_adjust_pi(p); |
4357 | ||
1da177e4 LT |
4358 | return 0; |
4359 | } | |
4360 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4361 | ||
95cdf3b7 IM |
4362 | static int |
4363 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4364 | { |
1da177e4 LT |
4365 | struct sched_param lparam; |
4366 | struct task_struct *p; | |
36c8b586 | 4367 | int retval; |
1da177e4 LT |
4368 | |
4369 | if (!param || pid < 0) | |
4370 | return -EINVAL; | |
4371 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4372 | return -EFAULT; | |
5fe1d75f ON |
4373 | |
4374 | rcu_read_lock(); | |
4375 | retval = -ESRCH; | |
1da177e4 | 4376 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4377 | if (p != NULL) |
4378 | retval = sched_setscheduler(p, policy, &lparam); | |
4379 | rcu_read_unlock(); | |
36c8b586 | 4380 | |
1da177e4 LT |
4381 | return retval; |
4382 | } | |
4383 | ||
4384 | /** | |
4385 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4386 | * @pid: the pid in question. | |
4387 | * @policy: new policy. | |
4388 | * @param: structure containing the new RT priority. | |
4389 | */ | |
41a2d6cf IM |
4390 | asmlinkage long |
4391 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4392 | { |
c21761f1 JB |
4393 | /* negative values for policy are not valid */ |
4394 | if (policy < 0) | |
4395 | return -EINVAL; | |
4396 | ||
1da177e4 LT |
4397 | return do_sched_setscheduler(pid, policy, param); |
4398 | } | |
4399 | ||
4400 | /** | |
4401 | * sys_sched_setparam - set/change the RT priority of a thread | |
4402 | * @pid: the pid in question. | |
4403 | * @param: structure containing the new RT priority. | |
4404 | */ | |
4405 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4406 | { | |
4407 | return do_sched_setscheduler(pid, -1, param); | |
4408 | } | |
4409 | ||
4410 | /** | |
4411 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4412 | * @pid: the pid in question. | |
4413 | */ | |
4414 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4415 | { | |
36c8b586 | 4416 | struct task_struct *p; |
3a5c359a | 4417 | int retval; |
1da177e4 LT |
4418 | |
4419 | if (pid < 0) | |
3a5c359a | 4420 | return -EINVAL; |
1da177e4 LT |
4421 | |
4422 | retval = -ESRCH; | |
4423 | read_lock(&tasklist_lock); | |
4424 | p = find_process_by_pid(pid); | |
4425 | if (p) { | |
4426 | retval = security_task_getscheduler(p); | |
4427 | if (!retval) | |
4428 | retval = p->policy; | |
4429 | } | |
4430 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4431 | return retval; |
4432 | } | |
4433 | ||
4434 | /** | |
4435 | * sys_sched_getscheduler - get the RT priority of a thread | |
4436 | * @pid: the pid in question. | |
4437 | * @param: structure containing the RT priority. | |
4438 | */ | |
4439 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4440 | { | |
4441 | struct sched_param lp; | |
36c8b586 | 4442 | struct task_struct *p; |
3a5c359a | 4443 | int retval; |
1da177e4 LT |
4444 | |
4445 | if (!param || pid < 0) | |
3a5c359a | 4446 | return -EINVAL; |
1da177e4 LT |
4447 | |
4448 | read_lock(&tasklist_lock); | |
4449 | p = find_process_by_pid(pid); | |
4450 | retval = -ESRCH; | |
4451 | if (!p) | |
4452 | goto out_unlock; | |
4453 | ||
4454 | retval = security_task_getscheduler(p); | |
4455 | if (retval) | |
4456 | goto out_unlock; | |
4457 | ||
4458 | lp.sched_priority = p->rt_priority; | |
4459 | read_unlock(&tasklist_lock); | |
4460 | ||
4461 | /* | |
4462 | * This one might sleep, we cannot do it with a spinlock held ... | |
4463 | */ | |
4464 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4465 | ||
1da177e4 LT |
4466 | return retval; |
4467 | ||
4468 | out_unlock: | |
4469 | read_unlock(&tasklist_lock); | |
4470 | return retval; | |
4471 | } | |
4472 | ||
4473 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4474 | { | |
1da177e4 | 4475 | cpumask_t cpus_allowed; |
36c8b586 IM |
4476 | struct task_struct *p; |
4477 | int retval; | |
1da177e4 | 4478 | |
5be9361c | 4479 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4480 | read_lock(&tasklist_lock); |
4481 | ||
4482 | p = find_process_by_pid(pid); | |
4483 | if (!p) { | |
4484 | read_unlock(&tasklist_lock); | |
5be9361c | 4485 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4486 | return -ESRCH; |
4487 | } | |
4488 | ||
4489 | /* | |
4490 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4491 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4492 | * usage count and then drop tasklist_lock. |
4493 | */ | |
4494 | get_task_struct(p); | |
4495 | read_unlock(&tasklist_lock); | |
4496 | ||
4497 | retval = -EPERM; | |
4498 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4499 | !capable(CAP_SYS_NICE)) | |
4500 | goto out_unlock; | |
4501 | ||
e7834f8f DQ |
4502 | retval = security_task_setscheduler(p, 0, NULL); |
4503 | if (retval) | |
4504 | goto out_unlock; | |
4505 | ||
1da177e4 LT |
4506 | cpus_allowed = cpuset_cpus_allowed(p); |
4507 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4508 | again: |
1da177e4 LT |
4509 | retval = set_cpus_allowed(p, new_mask); |
4510 | ||
8707d8b8 PM |
4511 | if (!retval) { |
4512 | cpus_allowed = cpuset_cpus_allowed(p); | |
4513 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4514 | /* | |
4515 | * We must have raced with a concurrent cpuset | |
4516 | * update. Just reset the cpus_allowed to the | |
4517 | * cpuset's cpus_allowed | |
4518 | */ | |
4519 | new_mask = cpus_allowed; | |
4520 | goto again; | |
4521 | } | |
4522 | } | |
1da177e4 LT |
4523 | out_unlock: |
4524 | put_task_struct(p); | |
5be9361c | 4525 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4526 | return retval; |
4527 | } | |
4528 | ||
4529 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4530 | cpumask_t *new_mask) | |
4531 | { | |
4532 | if (len < sizeof(cpumask_t)) { | |
4533 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4534 | } else if (len > sizeof(cpumask_t)) { | |
4535 | len = sizeof(cpumask_t); | |
4536 | } | |
4537 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4538 | } | |
4539 | ||
4540 | /** | |
4541 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4542 | * @pid: pid of the process | |
4543 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4544 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4545 | */ | |
4546 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4547 | unsigned long __user *user_mask_ptr) | |
4548 | { | |
4549 | cpumask_t new_mask; | |
4550 | int retval; | |
4551 | ||
4552 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4553 | if (retval) | |
4554 | return retval; | |
4555 | ||
4556 | return sched_setaffinity(pid, new_mask); | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * Represents all cpu's present in the system | |
4561 | * In systems capable of hotplug, this map could dynamically grow | |
4562 | * as new cpu's are detected in the system via any platform specific | |
4563 | * method, such as ACPI for e.g. | |
4564 | */ | |
4565 | ||
4cef0c61 | 4566 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4567 | EXPORT_SYMBOL(cpu_present_map); |
4568 | ||
4569 | #ifndef CONFIG_SMP | |
4cef0c61 | 4570 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4571 | EXPORT_SYMBOL(cpu_online_map); |
4572 | ||
4cef0c61 | 4573 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4574 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4575 | #endif |
4576 | ||
4577 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4578 | { | |
36c8b586 | 4579 | struct task_struct *p; |
1da177e4 | 4580 | int retval; |
1da177e4 | 4581 | |
5be9361c | 4582 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4583 | read_lock(&tasklist_lock); |
4584 | ||
4585 | retval = -ESRCH; | |
4586 | p = find_process_by_pid(pid); | |
4587 | if (!p) | |
4588 | goto out_unlock; | |
4589 | ||
e7834f8f DQ |
4590 | retval = security_task_getscheduler(p); |
4591 | if (retval) | |
4592 | goto out_unlock; | |
4593 | ||
2f7016d9 | 4594 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4595 | |
4596 | out_unlock: | |
4597 | read_unlock(&tasklist_lock); | |
5be9361c | 4598 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4599 | |
9531b62f | 4600 | return retval; |
1da177e4 LT |
4601 | } |
4602 | ||
4603 | /** | |
4604 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4605 | * @pid: pid of the process | |
4606 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4607 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4608 | */ | |
4609 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4610 | unsigned long __user *user_mask_ptr) | |
4611 | { | |
4612 | int ret; | |
4613 | cpumask_t mask; | |
4614 | ||
4615 | if (len < sizeof(cpumask_t)) | |
4616 | return -EINVAL; | |
4617 | ||
4618 | ret = sched_getaffinity(pid, &mask); | |
4619 | if (ret < 0) | |
4620 | return ret; | |
4621 | ||
4622 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4623 | return -EFAULT; | |
4624 | ||
4625 | return sizeof(cpumask_t); | |
4626 | } | |
4627 | ||
4628 | /** | |
4629 | * sys_sched_yield - yield the current processor to other threads. | |
4630 | * | |
dd41f596 IM |
4631 | * This function yields the current CPU to other tasks. If there are no |
4632 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4633 | */ |
4634 | asmlinkage long sys_sched_yield(void) | |
4635 | { | |
70b97a7f | 4636 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4637 | |
2d72376b | 4638 | schedstat_inc(rq, yld_count); |
4530d7ab | 4639 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4640 | |
4641 | /* | |
4642 | * Since we are going to call schedule() anyway, there's | |
4643 | * no need to preempt or enable interrupts: | |
4644 | */ | |
4645 | __release(rq->lock); | |
8a25d5de | 4646 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4647 | _raw_spin_unlock(&rq->lock); |
4648 | preempt_enable_no_resched(); | |
4649 | ||
4650 | schedule(); | |
4651 | ||
4652 | return 0; | |
4653 | } | |
4654 | ||
e7b38404 | 4655 | static void __cond_resched(void) |
1da177e4 | 4656 | { |
8e0a43d8 IM |
4657 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4658 | __might_sleep(__FILE__, __LINE__); | |
4659 | #endif | |
5bbcfd90 IM |
4660 | /* |
4661 | * The BKS might be reacquired before we have dropped | |
4662 | * PREEMPT_ACTIVE, which could trigger a second | |
4663 | * cond_resched() call. | |
4664 | */ | |
1da177e4 LT |
4665 | do { |
4666 | add_preempt_count(PREEMPT_ACTIVE); | |
4667 | schedule(); | |
4668 | sub_preempt_count(PREEMPT_ACTIVE); | |
4669 | } while (need_resched()); | |
4670 | } | |
4671 | ||
4672 | int __sched cond_resched(void) | |
4673 | { | |
9414232f IM |
4674 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4675 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4676 | __cond_resched(); |
4677 | return 1; | |
4678 | } | |
4679 | return 0; | |
4680 | } | |
1da177e4 LT |
4681 | EXPORT_SYMBOL(cond_resched); |
4682 | ||
4683 | /* | |
4684 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4685 | * call schedule, and on return reacquire the lock. | |
4686 | * | |
41a2d6cf | 4687 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4688 | * operations here to prevent schedule() from being called twice (once via |
4689 | * spin_unlock(), once by hand). | |
4690 | */ | |
95cdf3b7 | 4691 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4692 | { |
6df3cecb JK |
4693 | int ret = 0; |
4694 | ||
1da177e4 LT |
4695 | if (need_lockbreak(lock)) { |
4696 | spin_unlock(lock); | |
4697 | cpu_relax(); | |
6df3cecb | 4698 | ret = 1; |
1da177e4 LT |
4699 | spin_lock(lock); |
4700 | } | |
9414232f | 4701 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4702 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4703 | _raw_spin_unlock(lock); |
4704 | preempt_enable_no_resched(); | |
4705 | __cond_resched(); | |
6df3cecb | 4706 | ret = 1; |
1da177e4 | 4707 | spin_lock(lock); |
1da177e4 | 4708 | } |
6df3cecb | 4709 | return ret; |
1da177e4 | 4710 | } |
1da177e4 LT |
4711 | EXPORT_SYMBOL(cond_resched_lock); |
4712 | ||
4713 | int __sched cond_resched_softirq(void) | |
4714 | { | |
4715 | BUG_ON(!in_softirq()); | |
4716 | ||
9414232f | 4717 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4718 | local_bh_enable(); |
1da177e4 LT |
4719 | __cond_resched(); |
4720 | local_bh_disable(); | |
4721 | return 1; | |
4722 | } | |
4723 | return 0; | |
4724 | } | |
1da177e4 LT |
4725 | EXPORT_SYMBOL(cond_resched_softirq); |
4726 | ||
1da177e4 LT |
4727 | /** |
4728 | * yield - yield the current processor to other threads. | |
4729 | * | |
72fd4a35 | 4730 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4731 | * thread runnable and calls sys_sched_yield(). |
4732 | */ | |
4733 | void __sched yield(void) | |
4734 | { | |
4735 | set_current_state(TASK_RUNNING); | |
4736 | sys_sched_yield(); | |
4737 | } | |
1da177e4 LT |
4738 | EXPORT_SYMBOL(yield); |
4739 | ||
4740 | /* | |
41a2d6cf | 4741 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4742 | * that process accounting knows that this is a task in IO wait state. |
4743 | * | |
4744 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4745 | * has set its backing_dev_info: the queue against which it should throttle) | |
4746 | */ | |
4747 | void __sched io_schedule(void) | |
4748 | { | |
70b97a7f | 4749 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4750 | |
0ff92245 | 4751 | delayacct_blkio_start(); |
1da177e4 LT |
4752 | atomic_inc(&rq->nr_iowait); |
4753 | schedule(); | |
4754 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4755 | delayacct_blkio_end(); |
1da177e4 | 4756 | } |
1da177e4 LT |
4757 | EXPORT_SYMBOL(io_schedule); |
4758 | ||
4759 | long __sched io_schedule_timeout(long timeout) | |
4760 | { | |
70b97a7f | 4761 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4762 | long ret; |
4763 | ||
0ff92245 | 4764 | delayacct_blkio_start(); |
1da177e4 LT |
4765 | atomic_inc(&rq->nr_iowait); |
4766 | ret = schedule_timeout(timeout); | |
4767 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4768 | delayacct_blkio_end(); |
1da177e4 LT |
4769 | return ret; |
4770 | } | |
4771 | ||
4772 | /** | |
4773 | * sys_sched_get_priority_max - return maximum RT priority. | |
4774 | * @policy: scheduling class. | |
4775 | * | |
4776 | * this syscall returns the maximum rt_priority that can be used | |
4777 | * by a given scheduling class. | |
4778 | */ | |
4779 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4780 | { | |
4781 | int ret = -EINVAL; | |
4782 | ||
4783 | switch (policy) { | |
4784 | case SCHED_FIFO: | |
4785 | case SCHED_RR: | |
4786 | ret = MAX_USER_RT_PRIO-1; | |
4787 | break; | |
4788 | case SCHED_NORMAL: | |
b0a9499c | 4789 | case SCHED_BATCH: |
dd41f596 | 4790 | case SCHED_IDLE: |
1da177e4 LT |
4791 | ret = 0; |
4792 | break; | |
4793 | } | |
4794 | return ret; | |
4795 | } | |
4796 | ||
4797 | /** | |
4798 | * sys_sched_get_priority_min - return minimum RT priority. | |
4799 | * @policy: scheduling class. | |
4800 | * | |
4801 | * this syscall returns the minimum rt_priority that can be used | |
4802 | * by a given scheduling class. | |
4803 | */ | |
4804 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4805 | { | |
4806 | int ret = -EINVAL; | |
4807 | ||
4808 | switch (policy) { | |
4809 | case SCHED_FIFO: | |
4810 | case SCHED_RR: | |
4811 | ret = 1; | |
4812 | break; | |
4813 | case SCHED_NORMAL: | |
b0a9499c | 4814 | case SCHED_BATCH: |
dd41f596 | 4815 | case SCHED_IDLE: |
1da177e4 LT |
4816 | ret = 0; |
4817 | } | |
4818 | return ret; | |
4819 | } | |
4820 | ||
4821 | /** | |
4822 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4823 | * @pid: pid of the process. | |
4824 | * @interval: userspace pointer to the timeslice value. | |
4825 | * | |
4826 | * this syscall writes the default timeslice value of a given process | |
4827 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4828 | */ | |
4829 | asmlinkage | |
4830 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4831 | { | |
36c8b586 | 4832 | struct task_struct *p; |
a4ec24b4 | 4833 | unsigned int time_slice; |
3a5c359a | 4834 | int retval; |
1da177e4 | 4835 | struct timespec t; |
1da177e4 LT |
4836 | |
4837 | if (pid < 0) | |
3a5c359a | 4838 | return -EINVAL; |
1da177e4 LT |
4839 | |
4840 | retval = -ESRCH; | |
4841 | read_lock(&tasklist_lock); | |
4842 | p = find_process_by_pid(pid); | |
4843 | if (!p) | |
4844 | goto out_unlock; | |
4845 | ||
4846 | retval = security_task_getscheduler(p); | |
4847 | if (retval) | |
4848 | goto out_unlock; | |
4849 | ||
77034937 IM |
4850 | /* |
4851 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
4852 | * tasks that are on an otherwise idle runqueue: | |
4853 | */ | |
4854 | time_slice = 0; | |
4855 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 4856 | time_slice = DEF_TIMESLICE; |
77034937 | 4857 | } else { |
a4ec24b4 DA |
4858 | struct sched_entity *se = &p->se; |
4859 | unsigned long flags; | |
4860 | struct rq *rq; | |
4861 | ||
4862 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
4863 | if (rq->cfs.load.weight) |
4864 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
4865 | task_rq_unlock(rq, &flags); |
4866 | } | |
1da177e4 | 4867 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4868 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4869 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4870 | return retval; |
3a5c359a | 4871 | |
1da177e4 LT |
4872 | out_unlock: |
4873 | read_unlock(&tasklist_lock); | |
4874 | return retval; | |
4875 | } | |
4876 | ||
2ed6e34f | 4877 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4878 | |
4879 | static void show_task(struct task_struct *p) | |
1da177e4 | 4880 | { |
1da177e4 | 4881 | unsigned long free = 0; |
36c8b586 | 4882 | unsigned state; |
1da177e4 | 4883 | |
1da177e4 | 4884 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4885 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4886 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4887 | #if BITS_PER_LONG == 32 |
1da177e4 | 4888 | if (state == TASK_RUNNING) |
cc4ea795 | 4889 | printk(KERN_CONT " running "); |
1da177e4 | 4890 | else |
cc4ea795 | 4891 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4892 | #else |
4893 | if (state == TASK_RUNNING) | |
cc4ea795 | 4894 | printk(KERN_CONT " running task "); |
1da177e4 | 4895 | else |
cc4ea795 | 4896 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4897 | #endif |
4898 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4899 | { | |
10ebffde | 4900 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4901 | while (!*n) |
4902 | n++; | |
10ebffde | 4903 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4904 | } |
4905 | #endif | |
ba25f9dc PE |
4906 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
4907 | task_pid_nr(p), task_pid_nr(p->parent)); | |
1da177e4 LT |
4908 | |
4909 | if (state != TASK_RUNNING) | |
4910 | show_stack(p, NULL); | |
4911 | } | |
4912 | ||
e59e2ae2 | 4913 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4914 | { |
36c8b586 | 4915 | struct task_struct *g, *p; |
1da177e4 | 4916 | |
4bd77321 IM |
4917 | #if BITS_PER_LONG == 32 |
4918 | printk(KERN_INFO | |
4919 | " task PC stack pid father\n"); | |
1da177e4 | 4920 | #else |
4bd77321 IM |
4921 | printk(KERN_INFO |
4922 | " task PC stack pid father\n"); | |
1da177e4 LT |
4923 | #endif |
4924 | read_lock(&tasklist_lock); | |
4925 | do_each_thread(g, p) { | |
4926 | /* | |
4927 | * reset the NMI-timeout, listing all files on a slow | |
4928 | * console might take alot of time: | |
4929 | */ | |
4930 | touch_nmi_watchdog(); | |
39bc89fd | 4931 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4932 | show_task(p); |
1da177e4 LT |
4933 | } while_each_thread(g, p); |
4934 | ||
04c9167f JF |
4935 | touch_all_softlockup_watchdogs(); |
4936 | ||
dd41f596 IM |
4937 | #ifdef CONFIG_SCHED_DEBUG |
4938 | sysrq_sched_debug_show(); | |
4939 | #endif | |
1da177e4 | 4940 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4941 | /* |
4942 | * Only show locks if all tasks are dumped: | |
4943 | */ | |
4944 | if (state_filter == -1) | |
4945 | debug_show_all_locks(); | |
1da177e4 LT |
4946 | } |
4947 | ||
1df21055 IM |
4948 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4949 | { | |
dd41f596 | 4950 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4951 | } |
4952 | ||
f340c0d1 IM |
4953 | /** |
4954 | * init_idle - set up an idle thread for a given CPU | |
4955 | * @idle: task in question | |
4956 | * @cpu: cpu the idle task belongs to | |
4957 | * | |
4958 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4959 | * flag, to make booting more robust. | |
4960 | */ | |
5c1e1767 | 4961 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4962 | { |
70b97a7f | 4963 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4964 | unsigned long flags; |
4965 | ||
dd41f596 IM |
4966 | __sched_fork(idle); |
4967 | idle->se.exec_start = sched_clock(); | |
4968 | ||
b29739f9 | 4969 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4970 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4971 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4972 | |
4973 | spin_lock_irqsave(&rq->lock, flags); | |
4974 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4975 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4976 | idle->oncpu = 1; | |
4977 | #endif | |
1da177e4 LT |
4978 | spin_unlock_irqrestore(&rq->lock, flags); |
4979 | ||
4980 | /* Set the preempt count _outside_ the spinlocks! */ | |
4981 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4982 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4983 | #else |
a1261f54 | 4984 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4985 | #endif |
dd41f596 IM |
4986 | /* |
4987 | * The idle tasks have their own, simple scheduling class: | |
4988 | */ | |
4989 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4990 | } |
4991 | ||
4992 | /* | |
4993 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4994 | * indicates which cpus entered this state. This is used | |
4995 | * in the rcu update to wait only for active cpus. For system | |
4996 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4997 | * always be CPU_MASK_NONE. | |
4998 | */ | |
4999 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5000 | ||
19978ca6 IM |
5001 | /* |
5002 | * Increase the granularity value when there are more CPUs, | |
5003 | * because with more CPUs the 'effective latency' as visible | |
5004 | * to users decreases. But the relationship is not linear, | |
5005 | * so pick a second-best guess by going with the log2 of the | |
5006 | * number of CPUs. | |
5007 | * | |
5008 | * This idea comes from the SD scheduler of Con Kolivas: | |
5009 | */ | |
5010 | static inline void sched_init_granularity(void) | |
5011 | { | |
5012 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5013 | const unsigned long limit = 200000000; | |
5014 | ||
5015 | sysctl_sched_min_granularity *= factor; | |
5016 | if (sysctl_sched_min_granularity > limit) | |
5017 | sysctl_sched_min_granularity = limit; | |
5018 | ||
5019 | sysctl_sched_latency *= factor; | |
5020 | if (sysctl_sched_latency > limit) | |
5021 | sysctl_sched_latency = limit; | |
5022 | ||
5023 | sysctl_sched_wakeup_granularity *= factor; | |
5024 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5025 | } | |
5026 | ||
1da177e4 LT |
5027 | #ifdef CONFIG_SMP |
5028 | /* | |
5029 | * This is how migration works: | |
5030 | * | |
70b97a7f | 5031 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5032 | * runqueue and wake up that CPU's migration thread. |
5033 | * 2) we down() the locked semaphore => thread blocks. | |
5034 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5035 | * thread off the CPU) | |
5036 | * 4) it gets the migration request and checks whether the migrated | |
5037 | * task is still in the wrong runqueue. | |
5038 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5039 | * it and puts it into the right queue. | |
5040 | * 6) migration thread up()s the semaphore. | |
5041 | * 7) we wake up and the migration is done. | |
5042 | */ | |
5043 | ||
5044 | /* | |
5045 | * Change a given task's CPU affinity. Migrate the thread to a | |
5046 | * proper CPU and schedule it away if the CPU it's executing on | |
5047 | * is removed from the allowed bitmask. | |
5048 | * | |
5049 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5050 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5051 | * call is not atomic; no spinlocks may be held. |
5052 | */ | |
36c8b586 | 5053 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5054 | { |
70b97a7f | 5055 | struct migration_req req; |
1da177e4 | 5056 | unsigned long flags; |
70b97a7f | 5057 | struct rq *rq; |
48f24c4d | 5058 | int ret = 0; |
1da177e4 LT |
5059 | |
5060 | rq = task_rq_lock(p, &flags); | |
5061 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5062 | ret = -EINVAL; | |
5063 | goto out; | |
5064 | } | |
5065 | ||
5066 | p->cpus_allowed = new_mask; | |
5067 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5068 | if (cpu_isset(task_cpu(p), new_mask)) | |
5069 | goto out; | |
5070 | ||
5071 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5072 | /* Need help from migration thread: drop lock and wait. */ | |
5073 | task_rq_unlock(rq, &flags); | |
5074 | wake_up_process(rq->migration_thread); | |
5075 | wait_for_completion(&req.done); | |
5076 | tlb_migrate_finish(p->mm); | |
5077 | return 0; | |
5078 | } | |
5079 | out: | |
5080 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5081 | |
1da177e4 LT |
5082 | return ret; |
5083 | } | |
1da177e4 LT |
5084 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5085 | ||
5086 | /* | |
41a2d6cf | 5087 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5088 | * this because either it can't run here any more (set_cpus_allowed() |
5089 | * away from this CPU, or CPU going down), or because we're | |
5090 | * attempting to rebalance this task on exec (sched_exec). | |
5091 | * | |
5092 | * So we race with normal scheduler movements, but that's OK, as long | |
5093 | * as the task is no longer on this CPU. | |
efc30814 KK |
5094 | * |
5095 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5096 | */ |
efc30814 | 5097 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5098 | { |
70b97a7f | 5099 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5100 | int ret = 0, on_rq; |
1da177e4 LT |
5101 | |
5102 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5103 | return ret; |
1da177e4 LT |
5104 | |
5105 | rq_src = cpu_rq(src_cpu); | |
5106 | rq_dest = cpu_rq(dest_cpu); | |
5107 | ||
5108 | double_rq_lock(rq_src, rq_dest); | |
5109 | /* Already moved. */ | |
5110 | if (task_cpu(p) != src_cpu) | |
5111 | goto out; | |
5112 | /* Affinity changed (again). */ | |
5113 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5114 | goto out; | |
5115 | ||
dd41f596 | 5116 | on_rq = p->se.on_rq; |
6e82a3be | 5117 | if (on_rq) |
2e1cb74a | 5118 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5119 | |
1da177e4 | 5120 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5121 | if (on_rq) { |
5122 | activate_task(rq_dest, p, 0); | |
5123 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5124 | } |
efc30814 | 5125 | ret = 1; |
1da177e4 LT |
5126 | out: |
5127 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5128 | return ret; |
1da177e4 LT |
5129 | } |
5130 | ||
5131 | /* | |
5132 | * migration_thread - this is a highprio system thread that performs | |
5133 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5134 | * another runqueue. | |
5135 | */ | |
95cdf3b7 | 5136 | static int migration_thread(void *data) |
1da177e4 | 5137 | { |
1da177e4 | 5138 | int cpu = (long)data; |
70b97a7f | 5139 | struct rq *rq; |
1da177e4 LT |
5140 | |
5141 | rq = cpu_rq(cpu); | |
5142 | BUG_ON(rq->migration_thread != current); | |
5143 | ||
5144 | set_current_state(TASK_INTERRUPTIBLE); | |
5145 | while (!kthread_should_stop()) { | |
70b97a7f | 5146 | struct migration_req *req; |
1da177e4 | 5147 | struct list_head *head; |
1da177e4 | 5148 | |
1da177e4 LT |
5149 | spin_lock_irq(&rq->lock); |
5150 | ||
5151 | if (cpu_is_offline(cpu)) { | |
5152 | spin_unlock_irq(&rq->lock); | |
5153 | goto wait_to_die; | |
5154 | } | |
5155 | ||
5156 | if (rq->active_balance) { | |
5157 | active_load_balance(rq, cpu); | |
5158 | rq->active_balance = 0; | |
5159 | } | |
5160 | ||
5161 | head = &rq->migration_queue; | |
5162 | ||
5163 | if (list_empty(head)) { | |
5164 | spin_unlock_irq(&rq->lock); | |
5165 | schedule(); | |
5166 | set_current_state(TASK_INTERRUPTIBLE); | |
5167 | continue; | |
5168 | } | |
70b97a7f | 5169 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5170 | list_del_init(head->next); |
5171 | ||
674311d5 NP |
5172 | spin_unlock(&rq->lock); |
5173 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5174 | local_irq_enable(); | |
1da177e4 LT |
5175 | |
5176 | complete(&req->done); | |
5177 | } | |
5178 | __set_current_state(TASK_RUNNING); | |
5179 | return 0; | |
5180 | ||
5181 | wait_to_die: | |
5182 | /* Wait for kthread_stop */ | |
5183 | set_current_state(TASK_INTERRUPTIBLE); | |
5184 | while (!kthread_should_stop()) { | |
5185 | schedule(); | |
5186 | set_current_state(TASK_INTERRUPTIBLE); | |
5187 | } | |
5188 | __set_current_state(TASK_RUNNING); | |
5189 | return 0; | |
5190 | } | |
5191 | ||
5192 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5193 | |
5194 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5195 | { | |
5196 | int ret; | |
5197 | ||
5198 | local_irq_disable(); | |
5199 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5200 | local_irq_enable(); | |
5201 | return ret; | |
5202 | } | |
5203 | ||
054b9108 | 5204 | /* |
3a4fa0a2 | 5205 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5206 | * NOTE: interrupts should be disabled by the caller |
5207 | */ | |
48f24c4d | 5208 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5209 | { |
efc30814 | 5210 | unsigned long flags; |
1da177e4 | 5211 | cpumask_t mask; |
70b97a7f IM |
5212 | struct rq *rq; |
5213 | int dest_cpu; | |
1da177e4 | 5214 | |
3a5c359a AK |
5215 | do { |
5216 | /* On same node? */ | |
5217 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5218 | cpus_and(mask, mask, p->cpus_allowed); | |
5219 | dest_cpu = any_online_cpu(mask); | |
5220 | ||
5221 | /* On any allowed CPU? */ | |
5222 | if (dest_cpu == NR_CPUS) | |
5223 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5224 | ||
5225 | /* No more Mr. Nice Guy. */ | |
5226 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5227 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5228 | /* | |
5229 | * Try to stay on the same cpuset, where the | |
5230 | * current cpuset may be a subset of all cpus. | |
5231 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5232 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5233 | * called within calls to cpuset_lock/cpuset_unlock. |
5234 | */ | |
3a5c359a | 5235 | rq = task_rq_lock(p, &flags); |
470fd646 | 5236 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5237 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5238 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5239 | |
3a5c359a AK |
5240 | /* |
5241 | * Don't tell them about moving exiting tasks or | |
5242 | * kernel threads (both mm NULL), since they never | |
5243 | * leave kernel. | |
5244 | */ | |
41a2d6cf | 5245 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5246 | printk(KERN_INFO "process %d (%s) no " |
5247 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5248 | task_pid_nr(p), p->comm, dead_cpu); |
5249 | } | |
3a5c359a | 5250 | } |
f7b4cddc | 5251 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5252 | } |
5253 | ||
5254 | /* | |
5255 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5256 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5257 | * for performance reasons the counter is not stricly tracking tasks to | |
5258 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5259 | * to keep the global sum constant after CPU-down: | |
5260 | */ | |
70b97a7f | 5261 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5262 | { |
70b97a7f | 5263 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5264 | unsigned long flags; |
5265 | ||
5266 | local_irq_save(flags); | |
5267 | double_rq_lock(rq_src, rq_dest); | |
5268 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5269 | rq_src->nr_uninterruptible = 0; | |
5270 | double_rq_unlock(rq_src, rq_dest); | |
5271 | local_irq_restore(flags); | |
5272 | } | |
5273 | ||
5274 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5275 | static void migrate_live_tasks(int src_cpu) | |
5276 | { | |
48f24c4d | 5277 | struct task_struct *p, *t; |
1da177e4 | 5278 | |
f7b4cddc | 5279 | read_lock(&tasklist_lock); |
1da177e4 | 5280 | |
48f24c4d IM |
5281 | do_each_thread(t, p) { |
5282 | if (p == current) | |
1da177e4 LT |
5283 | continue; |
5284 | ||
48f24c4d IM |
5285 | if (task_cpu(p) == src_cpu) |
5286 | move_task_off_dead_cpu(src_cpu, p); | |
5287 | } while_each_thread(t, p); | |
1da177e4 | 5288 | |
f7b4cddc | 5289 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5290 | } |
5291 | ||
dd41f596 IM |
5292 | /* |
5293 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5294 | * It does so by boosting its priority to highest possible. |
5295 | * Used by CPU offline code. | |
1da177e4 LT |
5296 | */ |
5297 | void sched_idle_next(void) | |
5298 | { | |
48f24c4d | 5299 | int this_cpu = smp_processor_id(); |
70b97a7f | 5300 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5301 | struct task_struct *p = rq->idle; |
5302 | unsigned long flags; | |
5303 | ||
5304 | /* cpu has to be offline */ | |
48f24c4d | 5305 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5306 | |
48f24c4d IM |
5307 | /* |
5308 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5309 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5310 | */ |
5311 | spin_lock_irqsave(&rq->lock, flags); | |
5312 | ||
dd41f596 | 5313 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5314 | |
94bc9a7b DA |
5315 | update_rq_clock(rq); |
5316 | activate_task(rq, p, 0); | |
1da177e4 LT |
5317 | |
5318 | spin_unlock_irqrestore(&rq->lock, flags); | |
5319 | } | |
5320 | ||
48f24c4d IM |
5321 | /* |
5322 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5323 | * offline. |
5324 | */ | |
5325 | void idle_task_exit(void) | |
5326 | { | |
5327 | struct mm_struct *mm = current->active_mm; | |
5328 | ||
5329 | BUG_ON(cpu_online(smp_processor_id())); | |
5330 | ||
5331 | if (mm != &init_mm) | |
5332 | switch_mm(mm, &init_mm, current); | |
5333 | mmdrop(mm); | |
5334 | } | |
5335 | ||
054b9108 | 5336 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5337 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5338 | { |
70b97a7f | 5339 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5340 | |
5341 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5342 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5343 | |
5344 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5345 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5346 | |
48f24c4d | 5347 | get_task_struct(p); |
1da177e4 LT |
5348 | |
5349 | /* | |
5350 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5351 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5352 | * fine. |
5353 | */ | |
f7b4cddc | 5354 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5355 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5356 | spin_lock_irq(&rq->lock); |
1da177e4 | 5357 | |
48f24c4d | 5358 | put_task_struct(p); |
1da177e4 LT |
5359 | } |
5360 | ||
5361 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5362 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5363 | { | |
70b97a7f | 5364 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5365 | struct task_struct *next; |
48f24c4d | 5366 | |
dd41f596 IM |
5367 | for ( ; ; ) { |
5368 | if (!rq->nr_running) | |
5369 | break; | |
a8e504d2 | 5370 | update_rq_clock(rq); |
ff95f3df | 5371 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5372 | if (!next) |
5373 | break; | |
5374 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5375 | |
1da177e4 LT |
5376 | } |
5377 | } | |
5378 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5379 | ||
e692ab53 NP |
5380 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5381 | ||
5382 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5383 | { |
5384 | .procname = "sched_domain", | |
c57baf1e | 5385 | .mode = 0555, |
e0361851 | 5386 | }, |
38605cae | 5387 | {0, }, |
e692ab53 NP |
5388 | }; |
5389 | ||
5390 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5391 | { |
c57baf1e | 5392 | .ctl_name = CTL_KERN, |
e0361851 | 5393 | .procname = "kernel", |
c57baf1e | 5394 | .mode = 0555, |
e0361851 AD |
5395 | .child = sd_ctl_dir, |
5396 | }, | |
38605cae | 5397 | {0, }, |
e692ab53 NP |
5398 | }; |
5399 | ||
5400 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5401 | { | |
5402 | struct ctl_table *entry = | |
5cf9f062 | 5403 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5404 | |
e692ab53 NP |
5405 | return entry; |
5406 | } | |
5407 | ||
6382bc90 MM |
5408 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5409 | { | |
cd790076 | 5410 | struct ctl_table *entry; |
6382bc90 | 5411 | |
cd790076 MM |
5412 | /* |
5413 | * In the intermediate directories, both the child directory and | |
5414 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5415 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5416 | * static strings and all have proc handlers. |
5417 | */ | |
5418 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5419 | if (entry->child) |
5420 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5421 | if (entry->proc_handler == NULL) |
5422 | kfree(entry->procname); | |
5423 | } | |
6382bc90 MM |
5424 | |
5425 | kfree(*tablep); | |
5426 | *tablep = NULL; | |
5427 | } | |
5428 | ||
e692ab53 | 5429 | static void |
e0361851 | 5430 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5431 | const char *procname, void *data, int maxlen, |
5432 | mode_t mode, proc_handler *proc_handler) | |
5433 | { | |
e692ab53 NP |
5434 | entry->procname = procname; |
5435 | entry->data = data; | |
5436 | entry->maxlen = maxlen; | |
5437 | entry->mode = mode; | |
5438 | entry->proc_handler = proc_handler; | |
5439 | } | |
5440 | ||
5441 | static struct ctl_table * | |
5442 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5443 | { | |
ace8b3d6 | 5444 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5445 | |
ad1cdc1d MM |
5446 | if (table == NULL) |
5447 | return NULL; | |
5448 | ||
e0361851 | 5449 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5450 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5451 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5452 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5453 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5454 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5455 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5456 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5457 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5458 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5459 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5460 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5461 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5462 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5463 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5464 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5465 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5466 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5467 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5468 | &sd->cache_nice_tries, |
5469 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5470 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5471 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5472 | /* &table[11] is terminator */ |
e692ab53 NP |
5473 | |
5474 | return table; | |
5475 | } | |
5476 | ||
9a4e7159 | 5477 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5478 | { |
5479 | struct ctl_table *entry, *table; | |
5480 | struct sched_domain *sd; | |
5481 | int domain_num = 0, i; | |
5482 | char buf[32]; | |
5483 | ||
5484 | for_each_domain(cpu, sd) | |
5485 | domain_num++; | |
5486 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5487 | if (table == NULL) |
5488 | return NULL; | |
e692ab53 NP |
5489 | |
5490 | i = 0; | |
5491 | for_each_domain(cpu, sd) { | |
5492 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5493 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5494 | entry->mode = 0555; |
e692ab53 NP |
5495 | entry->child = sd_alloc_ctl_domain_table(sd); |
5496 | entry++; | |
5497 | i++; | |
5498 | } | |
5499 | return table; | |
5500 | } | |
5501 | ||
5502 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5503 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5504 | { |
5505 | int i, cpu_num = num_online_cpus(); | |
5506 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5507 | char buf[32]; | |
5508 | ||
7378547f MM |
5509 | WARN_ON(sd_ctl_dir[0].child); |
5510 | sd_ctl_dir[0].child = entry; | |
5511 | ||
ad1cdc1d MM |
5512 | if (entry == NULL) |
5513 | return; | |
5514 | ||
97b6ea7b | 5515 | for_each_online_cpu(i) { |
e692ab53 | 5516 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5517 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5518 | entry->mode = 0555; |
e692ab53 | 5519 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5520 | entry++; |
e692ab53 | 5521 | } |
7378547f MM |
5522 | |
5523 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5524 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5525 | } | |
6382bc90 | 5526 | |
7378547f | 5527 | /* may be called multiple times per register */ |
6382bc90 MM |
5528 | static void unregister_sched_domain_sysctl(void) |
5529 | { | |
7378547f MM |
5530 | if (sd_sysctl_header) |
5531 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5532 | sd_sysctl_header = NULL; |
7378547f MM |
5533 | if (sd_ctl_dir[0].child) |
5534 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5535 | } |
e692ab53 | 5536 | #else |
6382bc90 MM |
5537 | static void register_sched_domain_sysctl(void) |
5538 | { | |
5539 | } | |
5540 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5541 | { |
5542 | } | |
5543 | #endif | |
5544 | ||
1da177e4 LT |
5545 | /* |
5546 | * migration_call - callback that gets triggered when a CPU is added. | |
5547 | * Here we can start up the necessary migration thread for the new CPU. | |
5548 | */ | |
48f24c4d IM |
5549 | static int __cpuinit |
5550 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5551 | { |
1da177e4 | 5552 | struct task_struct *p; |
48f24c4d | 5553 | int cpu = (long)hcpu; |
1da177e4 | 5554 | unsigned long flags; |
70b97a7f | 5555 | struct rq *rq; |
1da177e4 LT |
5556 | |
5557 | switch (action) { | |
5be9361c GS |
5558 | case CPU_LOCK_ACQUIRE: |
5559 | mutex_lock(&sched_hotcpu_mutex); | |
5560 | break; | |
5561 | ||
1da177e4 | 5562 | case CPU_UP_PREPARE: |
8bb78442 | 5563 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5564 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5565 | if (IS_ERR(p)) |
5566 | return NOTIFY_BAD; | |
1da177e4 LT |
5567 | kthread_bind(p, cpu); |
5568 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5569 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5570 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5571 | task_rq_unlock(rq, &flags); |
5572 | cpu_rq(cpu)->migration_thread = p; | |
5573 | break; | |
48f24c4d | 5574 | |
1da177e4 | 5575 | case CPU_ONLINE: |
8bb78442 | 5576 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5577 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 LT |
5578 | wake_up_process(cpu_rq(cpu)->migration_thread); |
5579 | break; | |
48f24c4d | 5580 | |
1da177e4 LT |
5581 | #ifdef CONFIG_HOTPLUG_CPU |
5582 | case CPU_UP_CANCELED: | |
8bb78442 | 5583 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5584 | if (!cpu_rq(cpu)->migration_thread) |
5585 | break; | |
41a2d6cf | 5586 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5587 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5588 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5589 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5590 | cpu_rq(cpu)->migration_thread = NULL; | |
5591 | break; | |
48f24c4d | 5592 | |
1da177e4 | 5593 | case CPU_DEAD: |
8bb78442 | 5594 | case CPU_DEAD_FROZEN: |
470fd646 | 5595 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5596 | migrate_live_tasks(cpu); |
5597 | rq = cpu_rq(cpu); | |
5598 | kthread_stop(rq->migration_thread); | |
5599 | rq->migration_thread = NULL; | |
5600 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5601 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5602 | update_rq_clock(rq); |
2e1cb74a | 5603 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5604 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5605 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5606 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5607 | migrate_dead_tasks(cpu); |
d2da272a | 5608 | spin_unlock_irq(&rq->lock); |
470fd646 | 5609 | cpuset_unlock(); |
1da177e4 LT |
5610 | migrate_nr_uninterruptible(rq); |
5611 | BUG_ON(rq->nr_running != 0); | |
5612 | ||
41a2d6cf IM |
5613 | /* |
5614 | * No need to migrate the tasks: it was best-effort if | |
5615 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5616 | * the requestors. | |
5617 | */ | |
1da177e4 LT |
5618 | spin_lock_irq(&rq->lock); |
5619 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5620 | struct migration_req *req; |
5621 | ||
1da177e4 | 5622 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5623 | struct migration_req, list); |
1da177e4 LT |
5624 | list_del_init(&req->list); |
5625 | complete(&req->done); | |
5626 | } | |
5627 | spin_unlock_irq(&rq->lock); | |
5628 | break; | |
5629 | #endif | |
5be9361c GS |
5630 | case CPU_LOCK_RELEASE: |
5631 | mutex_unlock(&sched_hotcpu_mutex); | |
5632 | break; | |
1da177e4 LT |
5633 | } |
5634 | return NOTIFY_OK; | |
5635 | } | |
5636 | ||
5637 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5638 | * happens before everything else. | |
5639 | */ | |
26c2143b | 5640 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5641 | .notifier_call = migration_call, |
5642 | .priority = 10 | |
5643 | }; | |
5644 | ||
e6fe6649 | 5645 | void __init migration_init(void) |
1da177e4 LT |
5646 | { |
5647 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5648 | int err; |
48f24c4d IM |
5649 | |
5650 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5651 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5652 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5653 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5654 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5655 | } |
5656 | #endif | |
5657 | ||
5658 | #ifdef CONFIG_SMP | |
476f3534 CL |
5659 | |
5660 | /* Number of possible processor ids */ | |
5661 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5662 | EXPORT_SYMBOL(nr_cpu_ids); | |
5663 | ||
3e9830dc | 5664 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5665 | |
5666 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5667 | { |
4dcf6aff IM |
5668 | struct sched_group *group = sd->groups; |
5669 | cpumask_t groupmask; | |
5670 | char str[NR_CPUS]; | |
1da177e4 | 5671 | |
4dcf6aff IM |
5672 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5673 | cpus_clear(groupmask); | |
5674 | ||
5675 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5676 | ||
5677 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5678 | printk("does not load-balance\n"); | |
5679 | if (sd->parent) | |
5680 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5681 | " has parent"); | |
5682 | return -1; | |
41c7ce9a NP |
5683 | } |
5684 | ||
4dcf6aff IM |
5685 | printk(KERN_CONT "span %s\n", str); |
5686 | ||
5687 | if (!cpu_isset(cpu, sd->span)) { | |
5688 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5689 | "CPU%d\n", cpu); | |
5690 | } | |
5691 | if (!cpu_isset(cpu, group->cpumask)) { | |
5692 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5693 | " CPU%d\n", cpu); | |
5694 | } | |
1da177e4 | 5695 | |
4dcf6aff | 5696 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5697 | do { |
4dcf6aff IM |
5698 | if (!group) { |
5699 | printk("\n"); | |
5700 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5701 | break; |
5702 | } | |
5703 | ||
4dcf6aff IM |
5704 | if (!group->__cpu_power) { |
5705 | printk(KERN_CONT "\n"); | |
5706 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5707 | "set\n"); | |
5708 | break; | |
5709 | } | |
1da177e4 | 5710 | |
4dcf6aff IM |
5711 | if (!cpus_weight(group->cpumask)) { |
5712 | printk(KERN_CONT "\n"); | |
5713 | printk(KERN_ERR "ERROR: empty group\n"); | |
5714 | break; | |
5715 | } | |
1da177e4 | 5716 | |
4dcf6aff IM |
5717 | if (cpus_intersects(groupmask, group->cpumask)) { |
5718 | printk(KERN_CONT "\n"); | |
5719 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5720 | break; | |
5721 | } | |
1da177e4 | 5722 | |
4dcf6aff | 5723 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5724 | |
4dcf6aff IM |
5725 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5726 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5727 | |
4dcf6aff IM |
5728 | group = group->next; |
5729 | } while (group != sd->groups); | |
5730 | printk(KERN_CONT "\n"); | |
1da177e4 | 5731 | |
4dcf6aff IM |
5732 | if (!cpus_equal(sd->span, groupmask)) |
5733 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5734 | |
4dcf6aff IM |
5735 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5736 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5737 | "of domain->span\n"); | |
5738 | return 0; | |
5739 | } | |
1da177e4 | 5740 | |
4dcf6aff IM |
5741 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5742 | { | |
5743 | int level = 0; | |
1da177e4 | 5744 | |
4dcf6aff IM |
5745 | if (!sd) { |
5746 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5747 | return; | |
5748 | } | |
1da177e4 | 5749 | |
4dcf6aff IM |
5750 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5751 | ||
5752 | for (;;) { | |
5753 | if (sched_domain_debug_one(sd, cpu, level)) | |
5754 | break; | |
1da177e4 LT |
5755 | level++; |
5756 | sd = sd->parent; | |
33859f7f | 5757 | if (!sd) |
4dcf6aff IM |
5758 | break; |
5759 | } | |
1da177e4 LT |
5760 | } |
5761 | #else | |
48f24c4d | 5762 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5763 | #endif |
5764 | ||
1a20ff27 | 5765 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5766 | { |
5767 | if (cpus_weight(sd->span) == 1) | |
5768 | return 1; | |
5769 | ||
5770 | /* Following flags need at least 2 groups */ | |
5771 | if (sd->flags & (SD_LOAD_BALANCE | | |
5772 | SD_BALANCE_NEWIDLE | | |
5773 | SD_BALANCE_FORK | | |
89c4710e SS |
5774 | SD_BALANCE_EXEC | |
5775 | SD_SHARE_CPUPOWER | | |
5776 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5777 | if (sd->groups != sd->groups->next) |
5778 | return 0; | |
5779 | } | |
5780 | ||
5781 | /* Following flags don't use groups */ | |
5782 | if (sd->flags & (SD_WAKE_IDLE | | |
5783 | SD_WAKE_AFFINE | | |
5784 | SD_WAKE_BALANCE)) | |
5785 | return 0; | |
5786 | ||
5787 | return 1; | |
5788 | } | |
5789 | ||
48f24c4d IM |
5790 | static int |
5791 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5792 | { |
5793 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5794 | ||
5795 | if (sd_degenerate(parent)) | |
5796 | return 1; | |
5797 | ||
5798 | if (!cpus_equal(sd->span, parent->span)) | |
5799 | return 0; | |
5800 | ||
5801 | /* Does parent contain flags not in child? */ | |
5802 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5803 | if (cflags & SD_WAKE_AFFINE) | |
5804 | pflags &= ~SD_WAKE_BALANCE; | |
5805 | /* Flags needing groups don't count if only 1 group in parent */ | |
5806 | if (parent->groups == parent->groups->next) { | |
5807 | pflags &= ~(SD_LOAD_BALANCE | | |
5808 | SD_BALANCE_NEWIDLE | | |
5809 | SD_BALANCE_FORK | | |
89c4710e SS |
5810 | SD_BALANCE_EXEC | |
5811 | SD_SHARE_CPUPOWER | | |
5812 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5813 | } |
5814 | if (~cflags & pflags) | |
5815 | return 0; | |
5816 | ||
5817 | return 1; | |
5818 | } | |
5819 | ||
1da177e4 LT |
5820 | /* |
5821 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5822 | * hold the hotplug lock. | |
5823 | */ | |
9c1cfda2 | 5824 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5825 | { |
70b97a7f | 5826 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5827 | struct sched_domain *tmp; |
5828 | ||
5829 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5830 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5831 | struct sched_domain *parent = tmp->parent; | |
5832 | if (!parent) | |
5833 | break; | |
1a848870 | 5834 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5835 | tmp->parent = parent->parent; |
1a848870 SS |
5836 | if (parent->parent) |
5837 | parent->parent->child = tmp; | |
5838 | } | |
245af2c7 SS |
5839 | } |
5840 | ||
1a848870 | 5841 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5842 | sd = sd->parent; |
1a848870 SS |
5843 | if (sd) |
5844 | sd->child = NULL; | |
5845 | } | |
1da177e4 LT |
5846 | |
5847 | sched_domain_debug(sd, cpu); | |
5848 | ||
674311d5 | 5849 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5850 | } |
5851 | ||
5852 | /* cpus with isolated domains */ | |
67af63a6 | 5853 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5854 | |
5855 | /* Setup the mask of cpus configured for isolated domains */ | |
5856 | static int __init isolated_cpu_setup(char *str) | |
5857 | { | |
5858 | int ints[NR_CPUS], i; | |
5859 | ||
5860 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5861 | cpus_clear(cpu_isolated_map); | |
5862 | for (i = 1; i <= ints[0]; i++) | |
5863 | if (ints[i] < NR_CPUS) | |
5864 | cpu_set(ints[i], cpu_isolated_map); | |
5865 | return 1; | |
5866 | } | |
5867 | ||
8927f494 | 5868 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5869 | |
5870 | /* | |
6711cab4 SS |
5871 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5872 | * to a function which identifies what group(along with sched group) a CPU | |
5873 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5874 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5875 | * |
5876 | * init_sched_build_groups will build a circular linked list of the groups | |
5877 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5878 | * and ->cpu_power to 0. | |
5879 | */ | |
a616058b | 5880 | static void |
6711cab4 SS |
5881 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5882 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5883 | struct sched_group **sg)) | |
1da177e4 LT |
5884 | { |
5885 | struct sched_group *first = NULL, *last = NULL; | |
5886 | cpumask_t covered = CPU_MASK_NONE; | |
5887 | int i; | |
5888 | ||
5889 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5890 | struct sched_group *sg; |
5891 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5892 | int j; |
5893 | ||
5894 | if (cpu_isset(i, covered)) | |
5895 | continue; | |
5896 | ||
5897 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5898 | sg->__cpu_power = 0; |
1da177e4 LT |
5899 | |
5900 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5901 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5902 | continue; |
5903 | ||
5904 | cpu_set(j, covered); | |
5905 | cpu_set(j, sg->cpumask); | |
5906 | } | |
5907 | if (!first) | |
5908 | first = sg; | |
5909 | if (last) | |
5910 | last->next = sg; | |
5911 | last = sg; | |
5912 | } | |
5913 | last->next = first; | |
5914 | } | |
5915 | ||
9c1cfda2 | 5916 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5917 | |
9c1cfda2 | 5918 | #ifdef CONFIG_NUMA |
198e2f18 | 5919 | |
9c1cfda2 JH |
5920 | /** |
5921 | * find_next_best_node - find the next node to include in a sched_domain | |
5922 | * @node: node whose sched_domain we're building | |
5923 | * @used_nodes: nodes already in the sched_domain | |
5924 | * | |
41a2d6cf | 5925 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5926 | * finds the closest node not already in the @used_nodes map. |
5927 | * | |
5928 | * Should use nodemask_t. | |
5929 | */ | |
5930 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5931 | { | |
5932 | int i, n, val, min_val, best_node = 0; | |
5933 | ||
5934 | min_val = INT_MAX; | |
5935 | ||
5936 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5937 | /* Start at @node */ | |
5938 | n = (node + i) % MAX_NUMNODES; | |
5939 | ||
5940 | if (!nr_cpus_node(n)) | |
5941 | continue; | |
5942 | ||
5943 | /* Skip already used nodes */ | |
5944 | if (test_bit(n, used_nodes)) | |
5945 | continue; | |
5946 | ||
5947 | /* Simple min distance search */ | |
5948 | val = node_distance(node, n); | |
5949 | ||
5950 | if (val < min_val) { | |
5951 | min_val = val; | |
5952 | best_node = n; | |
5953 | } | |
5954 | } | |
5955 | ||
5956 | set_bit(best_node, used_nodes); | |
5957 | return best_node; | |
5958 | } | |
5959 | ||
5960 | /** | |
5961 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5962 | * @node: node whose cpumask we're constructing | |
5963 | * @size: number of nodes to include in this span | |
5964 | * | |
41a2d6cf | 5965 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5966 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5967 | * out optimally. | |
5968 | */ | |
5969 | static cpumask_t sched_domain_node_span(int node) | |
5970 | { | |
9c1cfda2 | 5971 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5972 | cpumask_t span, nodemask; |
5973 | int i; | |
9c1cfda2 JH |
5974 | |
5975 | cpus_clear(span); | |
5976 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5977 | ||
5978 | nodemask = node_to_cpumask(node); | |
5979 | cpus_or(span, span, nodemask); | |
5980 | set_bit(node, used_nodes); | |
5981 | ||
5982 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5983 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5984 | |
9c1cfda2 JH |
5985 | nodemask = node_to_cpumask(next_node); |
5986 | cpus_or(span, span, nodemask); | |
5987 | } | |
5988 | ||
5989 | return span; | |
5990 | } | |
5991 | #endif | |
5992 | ||
5c45bf27 | 5993 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5994 | |
9c1cfda2 | 5995 | /* |
48f24c4d | 5996 | * SMT sched-domains: |
9c1cfda2 | 5997 | */ |
1da177e4 LT |
5998 | #ifdef CONFIG_SCHED_SMT |
5999 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6000 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6001 | |
41a2d6cf IM |
6002 | static int |
6003 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6004 | { |
6711cab4 SS |
6005 | if (sg) |
6006 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6007 | return cpu; |
6008 | } | |
6009 | #endif | |
6010 | ||
48f24c4d IM |
6011 | /* |
6012 | * multi-core sched-domains: | |
6013 | */ | |
1e9f28fa SS |
6014 | #ifdef CONFIG_SCHED_MC |
6015 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6016 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6017 | #endif |
6018 | ||
6019 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6020 | static int |
6021 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6022 | { |
6711cab4 | 6023 | int group; |
d5a7430d | 6024 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6025 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6026 | group = first_cpu(mask); |
6027 | if (sg) | |
6028 | *sg = &per_cpu(sched_group_core, group); | |
6029 | return group; | |
1e9f28fa SS |
6030 | } |
6031 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6032 | static int |
6033 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6034 | { |
6711cab4 SS |
6035 | if (sg) |
6036 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6037 | return cpu; |
6038 | } | |
6039 | #endif | |
6040 | ||
1da177e4 | 6041 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6042 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6043 | |
41a2d6cf IM |
6044 | static int |
6045 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6046 | { |
6711cab4 | 6047 | int group; |
48f24c4d | 6048 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6049 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6050 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6051 | group = first_cpu(mask); |
1e9f28fa | 6052 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6053 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6054 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6055 | group = first_cpu(mask); |
1da177e4 | 6056 | #else |
6711cab4 | 6057 | group = cpu; |
1da177e4 | 6058 | #endif |
6711cab4 SS |
6059 | if (sg) |
6060 | *sg = &per_cpu(sched_group_phys, group); | |
6061 | return group; | |
1da177e4 LT |
6062 | } |
6063 | ||
6064 | #ifdef CONFIG_NUMA | |
1da177e4 | 6065 | /* |
9c1cfda2 JH |
6066 | * The init_sched_build_groups can't handle what we want to do with node |
6067 | * groups, so roll our own. Now each node has its own list of groups which | |
6068 | * gets dynamically allocated. | |
1da177e4 | 6069 | */ |
9c1cfda2 | 6070 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6071 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6072 | |
9c1cfda2 | 6073 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6074 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6075 | |
6711cab4 SS |
6076 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6077 | struct sched_group **sg) | |
9c1cfda2 | 6078 | { |
6711cab4 SS |
6079 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6080 | int group; | |
6081 | ||
6082 | cpus_and(nodemask, nodemask, *cpu_map); | |
6083 | group = first_cpu(nodemask); | |
6084 | ||
6085 | if (sg) | |
6086 | *sg = &per_cpu(sched_group_allnodes, group); | |
6087 | return group; | |
1da177e4 | 6088 | } |
6711cab4 | 6089 | |
08069033 SS |
6090 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6091 | { | |
6092 | struct sched_group *sg = group_head; | |
6093 | int j; | |
6094 | ||
6095 | if (!sg) | |
6096 | return; | |
3a5c359a AK |
6097 | do { |
6098 | for_each_cpu_mask(j, sg->cpumask) { | |
6099 | struct sched_domain *sd; | |
08069033 | 6100 | |
3a5c359a AK |
6101 | sd = &per_cpu(phys_domains, j); |
6102 | if (j != first_cpu(sd->groups->cpumask)) { | |
6103 | /* | |
6104 | * Only add "power" once for each | |
6105 | * physical package. | |
6106 | */ | |
6107 | continue; | |
6108 | } | |
08069033 | 6109 | |
3a5c359a AK |
6110 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6111 | } | |
6112 | sg = sg->next; | |
6113 | } while (sg != group_head); | |
08069033 | 6114 | } |
1da177e4 LT |
6115 | #endif |
6116 | ||
a616058b | 6117 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6118 | /* Free memory allocated for various sched_group structures */ |
6119 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6120 | { | |
a616058b | 6121 | int cpu, i; |
51888ca2 SV |
6122 | |
6123 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6124 | struct sched_group **sched_group_nodes |
6125 | = sched_group_nodes_bycpu[cpu]; | |
6126 | ||
51888ca2 SV |
6127 | if (!sched_group_nodes) |
6128 | continue; | |
6129 | ||
6130 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6131 | cpumask_t nodemask = node_to_cpumask(i); | |
6132 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6133 | ||
6134 | cpus_and(nodemask, nodemask, *cpu_map); | |
6135 | if (cpus_empty(nodemask)) | |
6136 | continue; | |
6137 | ||
6138 | if (sg == NULL) | |
6139 | continue; | |
6140 | sg = sg->next; | |
6141 | next_sg: | |
6142 | oldsg = sg; | |
6143 | sg = sg->next; | |
6144 | kfree(oldsg); | |
6145 | if (oldsg != sched_group_nodes[i]) | |
6146 | goto next_sg; | |
6147 | } | |
6148 | kfree(sched_group_nodes); | |
6149 | sched_group_nodes_bycpu[cpu] = NULL; | |
6150 | } | |
51888ca2 | 6151 | } |
a616058b SS |
6152 | #else |
6153 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6154 | { | |
6155 | } | |
6156 | #endif | |
51888ca2 | 6157 | |
89c4710e SS |
6158 | /* |
6159 | * Initialize sched groups cpu_power. | |
6160 | * | |
6161 | * cpu_power indicates the capacity of sched group, which is used while | |
6162 | * distributing the load between different sched groups in a sched domain. | |
6163 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6164 | * there are asymmetries in the topology. If there are asymmetries, group | |
6165 | * having more cpu_power will pickup more load compared to the group having | |
6166 | * less cpu_power. | |
6167 | * | |
6168 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6169 | * the maximum number of tasks a group can handle in the presence of other idle | |
6170 | * or lightly loaded groups in the same sched domain. | |
6171 | */ | |
6172 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6173 | { | |
6174 | struct sched_domain *child; | |
6175 | struct sched_group *group; | |
6176 | ||
6177 | WARN_ON(!sd || !sd->groups); | |
6178 | ||
6179 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6180 | return; | |
6181 | ||
6182 | child = sd->child; | |
6183 | ||
5517d86b ED |
6184 | sd->groups->__cpu_power = 0; |
6185 | ||
89c4710e SS |
6186 | /* |
6187 | * For perf policy, if the groups in child domain share resources | |
6188 | * (for example cores sharing some portions of the cache hierarchy | |
6189 | * or SMT), then set this domain groups cpu_power such that each group | |
6190 | * can handle only one task, when there are other idle groups in the | |
6191 | * same sched domain. | |
6192 | */ | |
6193 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6194 | (child->flags & | |
6195 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6196 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6197 | return; |
6198 | } | |
6199 | ||
89c4710e SS |
6200 | /* |
6201 | * add cpu_power of each child group to this groups cpu_power | |
6202 | */ | |
6203 | group = child->groups; | |
6204 | do { | |
5517d86b | 6205 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6206 | group = group->next; |
6207 | } while (group != child->groups); | |
6208 | } | |
6209 | ||
1da177e4 | 6210 | /* |
1a20ff27 DG |
6211 | * Build sched domains for a given set of cpus and attach the sched domains |
6212 | * to the individual cpus | |
1da177e4 | 6213 | */ |
51888ca2 | 6214 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6215 | { |
6216 | int i; | |
d1b55138 JH |
6217 | #ifdef CONFIG_NUMA |
6218 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6219 | int sd_allnodes = 0; |
d1b55138 JH |
6220 | |
6221 | /* | |
6222 | * Allocate the per-node list of sched groups | |
6223 | */ | |
5cf9f062 | 6224 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6225 | GFP_KERNEL); |
d1b55138 JH |
6226 | if (!sched_group_nodes) { |
6227 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6228 | return -ENOMEM; |
d1b55138 JH |
6229 | } |
6230 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6231 | #endif | |
1da177e4 LT |
6232 | |
6233 | /* | |
1a20ff27 | 6234 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6235 | */ |
1a20ff27 | 6236 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6237 | struct sched_domain *sd = NULL, *p; |
6238 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6239 | ||
1a20ff27 | 6240 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6241 | |
6242 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6243 | if (cpus_weight(*cpu_map) > |
6244 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6245 | sd = &per_cpu(allnodes_domains, i); |
6246 | *sd = SD_ALLNODES_INIT; | |
6247 | sd->span = *cpu_map; | |
6711cab4 | 6248 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6249 | p = sd; |
6711cab4 | 6250 | sd_allnodes = 1; |
9c1cfda2 JH |
6251 | } else |
6252 | p = NULL; | |
6253 | ||
1da177e4 | 6254 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6255 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6256 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6257 | sd->parent = p; | |
1a848870 SS |
6258 | if (p) |
6259 | p->child = sd; | |
9c1cfda2 | 6260 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6261 | #endif |
6262 | ||
6263 | p = sd; | |
6264 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6265 | *sd = SD_CPU_INIT; |
6266 | sd->span = nodemask; | |
6267 | sd->parent = p; | |
1a848870 SS |
6268 | if (p) |
6269 | p->child = sd; | |
6711cab4 | 6270 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6271 | |
1e9f28fa SS |
6272 | #ifdef CONFIG_SCHED_MC |
6273 | p = sd; | |
6274 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6275 | *sd = SD_MC_INIT; |
6276 | sd->span = cpu_coregroup_map(i); | |
6277 | cpus_and(sd->span, sd->span, *cpu_map); | |
6278 | sd->parent = p; | |
1a848870 | 6279 | p->child = sd; |
6711cab4 | 6280 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6281 | #endif |
6282 | ||
1da177e4 LT |
6283 | #ifdef CONFIG_SCHED_SMT |
6284 | p = sd; | |
6285 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6286 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6287 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6288 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6289 | sd->parent = p; |
1a848870 | 6290 | p->child = sd; |
6711cab4 | 6291 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6292 | #endif |
6293 | } | |
6294 | ||
6295 | #ifdef CONFIG_SCHED_SMT | |
6296 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6297 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6298 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6299 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6300 | if (i != first_cpu(this_sibling_map)) |
6301 | continue; | |
6302 | ||
dd41f596 IM |
6303 | init_sched_build_groups(this_sibling_map, cpu_map, |
6304 | &cpu_to_cpu_group); | |
1da177e4 LT |
6305 | } |
6306 | #endif | |
6307 | ||
1e9f28fa SS |
6308 | #ifdef CONFIG_SCHED_MC |
6309 | /* Set up multi-core groups */ | |
6310 | for_each_cpu_mask(i, *cpu_map) { | |
6311 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6312 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6313 | if (i != first_cpu(this_core_map)) | |
6314 | continue; | |
dd41f596 IM |
6315 | init_sched_build_groups(this_core_map, cpu_map, |
6316 | &cpu_to_core_group); | |
1e9f28fa SS |
6317 | } |
6318 | #endif | |
6319 | ||
1da177e4 LT |
6320 | /* Set up physical groups */ |
6321 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6322 | cpumask_t nodemask = node_to_cpumask(i); | |
6323 | ||
1a20ff27 | 6324 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6325 | if (cpus_empty(nodemask)) |
6326 | continue; | |
6327 | ||
6711cab4 | 6328 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6329 | } |
6330 | ||
6331 | #ifdef CONFIG_NUMA | |
6332 | /* Set up node groups */ | |
6711cab4 | 6333 | if (sd_allnodes) |
dd41f596 IM |
6334 | init_sched_build_groups(*cpu_map, cpu_map, |
6335 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6336 | |
6337 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6338 | /* Set up node groups */ | |
6339 | struct sched_group *sg, *prev; | |
6340 | cpumask_t nodemask = node_to_cpumask(i); | |
6341 | cpumask_t domainspan; | |
6342 | cpumask_t covered = CPU_MASK_NONE; | |
6343 | int j; | |
6344 | ||
6345 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6346 | if (cpus_empty(nodemask)) { |
6347 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6348 | continue; |
d1b55138 | 6349 | } |
9c1cfda2 JH |
6350 | |
6351 | domainspan = sched_domain_node_span(i); | |
6352 | cpus_and(domainspan, domainspan, *cpu_map); | |
6353 | ||
15f0b676 | 6354 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6355 | if (!sg) { |
6356 | printk(KERN_WARNING "Can not alloc domain group for " | |
6357 | "node %d\n", i); | |
6358 | goto error; | |
6359 | } | |
9c1cfda2 JH |
6360 | sched_group_nodes[i] = sg; |
6361 | for_each_cpu_mask(j, nodemask) { | |
6362 | struct sched_domain *sd; | |
9761eea8 | 6363 | |
9c1cfda2 JH |
6364 | sd = &per_cpu(node_domains, j); |
6365 | sd->groups = sg; | |
9c1cfda2 | 6366 | } |
5517d86b | 6367 | sg->__cpu_power = 0; |
9c1cfda2 | 6368 | sg->cpumask = nodemask; |
51888ca2 | 6369 | sg->next = sg; |
9c1cfda2 JH |
6370 | cpus_or(covered, covered, nodemask); |
6371 | prev = sg; | |
6372 | ||
6373 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6374 | cpumask_t tmp, notcovered; | |
6375 | int n = (i + j) % MAX_NUMNODES; | |
6376 | ||
6377 | cpus_complement(notcovered, covered); | |
6378 | cpus_and(tmp, notcovered, *cpu_map); | |
6379 | cpus_and(tmp, tmp, domainspan); | |
6380 | if (cpus_empty(tmp)) | |
6381 | break; | |
6382 | ||
6383 | nodemask = node_to_cpumask(n); | |
6384 | cpus_and(tmp, tmp, nodemask); | |
6385 | if (cpus_empty(tmp)) | |
6386 | continue; | |
6387 | ||
15f0b676 SV |
6388 | sg = kmalloc_node(sizeof(struct sched_group), |
6389 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6390 | if (!sg) { |
6391 | printk(KERN_WARNING | |
6392 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6393 | goto error; |
9c1cfda2 | 6394 | } |
5517d86b | 6395 | sg->__cpu_power = 0; |
9c1cfda2 | 6396 | sg->cpumask = tmp; |
51888ca2 | 6397 | sg->next = prev->next; |
9c1cfda2 JH |
6398 | cpus_or(covered, covered, tmp); |
6399 | prev->next = sg; | |
6400 | prev = sg; | |
6401 | } | |
9c1cfda2 | 6402 | } |
1da177e4 LT |
6403 | #endif |
6404 | ||
6405 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6406 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6407 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6408 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6409 | ||
89c4710e | 6410 | init_sched_groups_power(i, sd); |
5c45bf27 | 6411 | } |
1da177e4 | 6412 | #endif |
1e9f28fa | 6413 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6414 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6415 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6416 | ||
89c4710e | 6417 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6418 | } |
6419 | #endif | |
1e9f28fa | 6420 | |
5c45bf27 | 6421 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6422 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6423 | ||
89c4710e | 6424 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6425 | } |
6426 | ||
9c1cfda2 | 6427 | #ifdef CONFIG_NUMA |
08069033 SS |
6428 | for (i = 0; i < MAX_NUMNODES; i++) |
6429 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6430 | |
6711cab4 SS |
6431 | if (sd_allnodes) { |
6432 | struct sched_group *sg; | |
f712c0c7 | 6433 | |
6711cab4 | 6434 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6435 | init_numa_sched_groups_power(sg); |
6436 | } | |
9c1cfda2 JH |
6437 | #endif |
6438 | ||
1da177e4 | 6439 | /* Attach the domains */ |
1a20ff27 | 6440 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6441 | struct sched_domain *sd; |
6442 | #ifdef CONFIG_SCHED_SMT | |
6443 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6444 | #elif defined(CONFIG_SCHED_MC) |
6445 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6446 | #else |
6447 | sd = &per_cpu(phys_domains, i); | |
6448 | #endif | |
6449 | cpu_attach_domain(sd, i); | |
6450 | } | |
51888ca2 SV |
6451 | |
6452 | return 0; | |
6453 | ||
a616058b | 6454 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6455 | error: |
6456 | free_sched_groups(cpu_map); | |
6457 | return -ENOMEM; | |
a616058b | 6458 | #endif |
1da177e4 | 6459 | } |
029190c5 PJ |
6460 | |
6461 | static cpumask_t *doms_cur; /* current sched domains */ | |
6462 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6463 | ||
6464 | /* | |
6465 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6466 | * cpumask_t) fails, then fallback to a single sched domain, | |
6467 | * as determined by the single cpumask_t fallback_doms. | |
6468 | */ | |
6469 | static cpumask_t fallback_doms; | |
6470 | ||
1a20ff27 | 6471 | /* |
41a2d6cf | 6472 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6473 | * For now this just excludes isolated cpus, but could be used to |
6474 | * exclude other special cases in the future. | |
1a20ff27 | 6475 | */ |
51888ca2 | 6476 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6477 | { |
7378547f MM |
6478 | int err; |
6479 | ||
029190c5 PJ |
6480 | ndoms_cur = 1; |
6481 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6482 | if (!doms_cur) | |
6483 | doms_cur = &fallback_doms; | |
6484 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6485 | err = build_sched_domains(doms_cur); |
6382bc90 | 6486 | register_sched_domain_sysctl(); |
7378547f MM |
6487 | |
6488 | return err; | |
1a20ff27 DG |
6489 | } |
6490 | ||
6491 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6492 | { |
51888ca2 | 6493 | free_sched_groups(cpu_map); |
9c1cfda2 | 6494 | } |
1da177e4 | 6495 | |
1a20ff27 DG |
6496 | /* |
6497 | * Detach sched domains from a group of cpus specified in cpu_map | |
6498 | * These cpus will now be attached to the NULL domain | |
6499 | */ | |
858119e1 | 6500 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6501 | { |
6502 | int i; | |
6503 | ||
6382bc90 MM |
6504 | unregister_sched_domain_sysctl(); |
6505 | ||
1a20ff27 DG |
6506 | for_each_cpu_mask(i, *cpu_map) |
6507 | cpu_attach_domain(NULL, i); | |
6508 | synchronize_sched(); | |
6509 | arch_destroy_sched_domains(cpu_map); | |
6510 | } | |
6511 | ||
029190c5 PJ |
6512 | /* |
6513 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6514 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6515 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6516 | * It destroys each deleted domain and builds each new domain. | |
6517 | * | |
6518 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6519 | * The masks don't intersect (don't overlap.) We should setup one |
6520 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6521 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6522 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6523 | * it as it is. | |
6524 | * | |
41a2d6cf IM |
6525 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6526 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6527 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6528 | * and partition_sched_domains() will fallback to the single partition | |
6529 | * 'fallback_doms'. | |
6530 | * | |
6531 | * Call with hotplug lock held | |
6532 | */ | |
6533 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6534 | { | |
6535 | int i, j; | |
6536 | ||
7378547f MM |
6537 | /* always unregister in case we don't destroy any domains */ |
6538 | unregister_sched_domain_sysctl(); | |
6539 | ||
029190c5 PJ |
6540 | if (doms_new == NULL) { |
6541 | ndoms_new = 1; | |
6542 | doms_new = &fallback_doms; | |
6543 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6544 | } | |
6545 | ||
6546 | /* Destroy deleted domains */ | |
6547 | for (i = 0; i < ndoms_cur; i++) { | |
6548 | for (j = 0; j < ndoms_new; j++) { | |
6549 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6550 | goto match1; | |
6551 | } | |
6552 | /* no match - a current sched domain not in new doms_new[] */ | |
6553 | detach_destroy_domains(doms_cur + i); | |
6554 | match1: | |
6555 | ; | |
6556 | } | |
6557 | ||
6558 | /* Build new domains */ | |
6559 | for (i = 0; i < ndoms_new; i++) { | |
6560 | for (j = 0; j < ndoms_cur; j++) { | |
6561 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6562 | goto match2; | |
6563 | } | |
6564 | /* no match - add a new doms_new */ | |
6565 | build_sched_domains(doms_new + i); | |
6566 | match2: | |
6567 | ; | |
6568 | } | |
6569 | ||
6570 | /* Remember the new sched domains */ | |
6571 | if (doms_cur != &fallback_doms) | |
6572 | kfree(doms_cur); | |
6573 | doms_cur = doms_new; | |
6574 | ndoms_cur = ndoms_new; | |
7378547f MM |
6575 | |
6576 | register_sched_domain_sysctl(); | |
029190c5 PJ |
6577 | } |
6578 | ||
5c45bf27 | 6579 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6580 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6581 | { |
6582 | int err; | |
6583 | ||
5be9361c | 6584 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6585 | detach_destroy_domains(&cpu_online_map); |
6586 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6587 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6588 | |
6589 | return err; | |
6590 | } | |
6591 | ||
6592 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6593 | { | |
6594 | int ret; | |
6595 | ||
6596 | if (buf[0] != '0' && buf[0] != '1') | |
6597 | return -EINVAL; | |
6598 | ||
6599 | if (smt) | |
6600 | sched_smt_power_savings = (buf[0] == '1'); | |
6601 | else | |
6602 | sched_mc_power_savings = (buf[0] == '1'); | |
6603 | ||
6604 | ret = arch_reinit_sched_domains(); | |
6605 | ||
6606 | return ret ? ret : count; | |
6607 | } | |
6608 | ||
5c45bf27 SS |
6609 | #ifdef CONFIG_SCHED_MC |
6610 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6611 | { | |
6612 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6613 | } | |
48f24c4d IM |
6614 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6615 | const char *buf, size_t count) | |
5c45bf27 SS |
6616 | { |
6617 | return sched_power_savings_store(buf, count, 0); | |
6618 | } | |
6707de00 AB |
6619 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6620 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6621 | #endif |
6622 | ||
6623 | #ifdef CONFIG_SCHED_SMT | |
6624 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6625 | { | |
6626 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6627 | } | |
48f24c4d IM |
6628 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6629 | const char *buf, size_t count) | |
5c45bf27 SS |
6630 | { |
6631 | return sched_power_savings_store(buf, count, 1); | |
6632 | } | |
6707de00 AB |
6633 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6634 | sched_smt_power_savings_store); | |
6635 | #endif | |
6636 | ||
6637 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6638 | { | |
6639 | int err = 0; | |
6640 | ||
6641 | #ifdef CONFIG_SCHED_SMT | |
6642 | if (smt_capable()) | |
6643 | err = sysfs_create_file(&cls->kset.kobj, | |
6644 | &attr_sched_smt_power_savings.attr); | |
6645 | #endif | |
6646 | #ifdef CONFIG_SCHED_MC | |
6647 | if (!err && mc_capable()) | |
6648 | err = sysfs_create_file(&cls->kset.kobj, | |
6649 | &attr_sched_mc_power_savings.attr); | |
6650 | #endif | |
6651 | return err; | |
6652 | } | |
5c45bf27 SS |
6653 | #endif |
6654 | ||
1da177e4 | 6655 | /* |
41a2d6cf | 6656 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6657 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6658 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6659 | * which will prevent rebalancing while the sched domains are recalculated. |
6660 | */ | |
6661 | static int update_sched_domains(struct notifier_block *nfb, | |
6662 | unsigned long action, void *hcpu) | |
6663 | { | |
1da177e4 LT |
6664 | switch (action) { |
6665 | case CPU_UP_PREPARE: | |
8bb78442 | 6666 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6667 | case CPU_DOWN_PREPARE: |
8bb78442 | 6668 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6669 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6670 | return NOTIFY_OK; |
6671 | ||
6672 | case CPU_UP_CANCELED: | |
8bb78442 | 6673 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6674 | case CPU_DOWN_FAILED: |
8bb78442 | 6675 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6676 | case CPU_ONLINE: |
8bb78442 | 6677 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6678 | case CPU_DEAD: |
8bb78442 | 6679 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6680 | /* |
6681 | * Fall through and re-initialise the domains. | |
6682 | */ | |
6683 | break; | |
6684 | default: | |
6685 | return NOTIFY_DONE; | |
6686 | } | |
6687 | ||
6688 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6689 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6690 | |
6691 | return NOTIFY_OK; | |
6692 | } | |
1da177e4 LT |
6693 | |
6694 | void __init sched_init_smp(void) | |
6695 | { | |
5c1e1767 NP |
6696 | cpumask_t non_isolated_cpus; |
6697 | ||
5be9361c | 6698 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6699 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6700 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6701 | if (cpus_empty(non_isolated_cpus)) |
6702 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6703 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6704 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6705 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6706 | |
6707 | /* Move init over to a non-isolated CPU */ | |
6708 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6709 | BUG(); | |
19978ca6 | 6710 | sched_init_granularity(); |
1da177e4 LT |
6711 | } |
6712 | #else | |
6713 | void __init sched_init_smp(void) | |
6714 | { | |
19978ca6 | 6715 | sched_init_granularity(); |
1da177e4 LT |
6716 | } |
6717 | #endif /* CONFIG_SMP */ | |
6718 | ||
6719 | int in_sched_functions(unsigned long addr) | |
6720 | { | |
1da177e4 LT |
6721 | return in_lock_functions(addr) || |
6722 | (addr >= (unsigned long)__sched_text_start | |
6723 | && addr < (unsigned long)__sched_text_end); | |
6724 | } | |
6725 | ||
a9957449 | 6726 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6727 | { |
6728 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6729 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6730 | cfs_rq->rq = rq; | |
6731 | #endif | |
67e9fb2a | 6732 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6733 | } |
6734 | ||
1da177e4 LT |
6735 | void __init sched_init(void) |
6736 | { | |
476f3534 | 6737 | int highest_cpu = 0; |
dd41f596 IM |
6738 | int i, j; |
6739 | ||
0a945022 | 6740 | for_each_possible_cpu(i) { |
dd41f596 | 6741 | struct rt_prio_array *array; |
70b97a7f | 6742 | struct rq *rq; |
1da177e4 LT |
6743 | |
6744 | rq = cpu_rq(i); | |
6745 | spin_lock_init(&rq->lock); | |
fcb99371 | 6746 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6747 | rq->nr_running = 0; |
dd41f596 IM |
6748 | rq->clock = 1; |
6749 | init_cfs_rq(&rq->cfs, rq); | |
6750 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6751 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6752 | { |
6753 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6754 | struct sched_entity *se = | |
6755 | &per_cpu(init_sched_entity, i); | |
6756 | ||
6757 | init_cfs_rq_p[i] = cfs_rq; | |
6758 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6759 | cfs_rq->tg = &init_task_group; |
3a252015 | 6760 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6761 | &rq->leaf_cfs_rq_list); |
6762 | ||
3a252015 IM |
6763 | init_sched_entity_p[i] = se; |
6764 | se->cfs_rq = &rq->cfs; | |
6765 | se->my_q = cfs_rq; | |
4cf86d77 | 6766 | se->load.weight = init_task_group_load; |
9b5b7751 | 6767 | se->load.inv_weight = |
4cf86d77 | 6768 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6769 | se->parent = NULL; |
6770 | } | |
4cf86d77 | 6771 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6772 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6773 | #endif |
1da177e4 | 6774 | |
dd41f596 IM |
6775 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6776 | rq->cpu_load[j] = 0; | |
1da177e4 | 6777 | #ifdef CONFIG_SMP |
41c7ce9a | 6778 | rq->sd = NULL; |
1da177e4 | 6779 | rq->active_balance = 0; |
dd41f596 | 6780 | rq->next_balance = jiffies; |
1da177e4 | 6781 | rq->push_cpu = 0; |
0a2966b4 | 6782 | rq->cpu = i; |
1da177e4 LT |
6783 | rq->migration_thread = NULL; |
6784 | INIT_LIST_HEAD(&rq->migration_queue); | |
6785 | #endif | |
6786 | atomic_set(&rq->nr_iowait, 0); | |
6787 | ||
dd41f596 IM |
6788 | array = &rq->rt.active; |
6789 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6790 | INIT_LIST_HEAD(array->queue + j); | |
6791 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6792 | } |
476f3534 | 6793 | highest_cpu = i; |
dd41f596 IM |
6794 | /* delimiter for bitsearch: */ |
6795 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6796 | } |
6797 | ||
2dd73a4f | 6798 | set_load_weight(&init_task); |
b50f60ce | 6799 | |
e107be36 AK |
6800 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6801 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6802 | #endif | |
6803 | ||
c9819f45 | 6804 | #ifdef CONFIG_SMP |
476f3534 | 6805 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6806 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6807 | #endif | |
6808 | ||
b50f60ce HC |
6809 | #ifdef CONFIG_RT_MUTEXES |
6810 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6811 | #endif | |
6812 | ||
1da177e4 LT |
6813 | /* |
6814 | * The boot idle thread does lazy MMU switching as well: | |
6815 | */ | |
6816 | atomic_inc(&init_mm.mm_count); | |
6817 | enter_lazy_tlb(&init_mm, current); | |
6818 | ||
6819 | /* | |
6820 | * Make us the idle thread. Technically, schedule() should not be | |
6821 | * called from this thread, however somewhere below it might be, | |
6822 | * but because we are the idle thread, we just pick up running again | |
6823 | * when this runqueue becomes "idle". | |
6824 | */ | |
6825 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6826 | /* |
6827 | * During early bootup we pretend to be a normal task: | |
6828 | */ | |
6829 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6830 | } |
6831 | ||
6832 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6833 | void __might_sleep(char *file, int line) | |
6834 | { | |
48f24c4d | 6835 | #ifdef in_atomic |
1da177e4 LT |
6836 | static unsigned long prev_jiffy; /* ratelimiting */ |
6837 | ||
6838 | if ((in_atomic() || irqs_disabled()) && | |
6839 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6840 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6841 | return; | |
6842 | prev_jiffy = jiffies; | |
91368d73 | 6843 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6844 | " context at %s:%d\n", file, line); |
6845 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6846 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6847 | debug_show_held_locks(current); |
3117df04 IM |
6848 | if (irqs_disabled()) |
6849 | print_irqtrace_events(current); | |
1da177e4 LT |
6850 | dump_stack(); |
6851 | } | |
6852 | #endif | |
6853 | } | |
6854 | EXPORT_SYMBOL(__might_sleep); | |
6855 | #endif | |
6856 | ||
6857 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6858 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6859 | { | |
6860 | int on_rq; | |
6861 | update_rq_clock(rq); | |
6862 | on_rq = p->se.on_rq; | |
6863 | if (on_rq) | |
6864 | deactivate_task(rq, p, 0); | |
6865 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6866 | if (on_rq) { | |
6867 | activate_task(rq, p, 0); | |
6868 | resched_task(rq->curr); | |
6869 | } | |
6870 | } | |
6871 | ||
1da177e4 LT |
6872 | void normalize_rt_tasks(void) |
6873 | { | |
a0f98a1c | 6874 | struct task_struct *g, *p; |
1da177e4 | 6875 | unsigned long flags; |
70b97a7f | 6876 | struct rq *rq; |
1da177e4 LT |
6877 | |
6878 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6879 | do_each_thread(g, p) { |
178be793 IM |
6880 | /* |
6881 | * Only normalize user tasks: | |
6882 | */ | |
6883 | if (!p->mm) | |
6884 | continue; | |
6885 | ||
6cfb0d5d | 6886 | p->se.exec_start = 0; |
6cfb0d5d | 6887 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6888 | p->se.wait_start = 0; |
dd41f596 | 6889 | p->se.sleep_start = 0; |
dd41f596 | 6890 | p->se.block_start = 0; |
6cfb0d5d | 6891 | #endif |
dd41f596 IM |
6892 | task_rq(p)->clock = 0; |
6893 | ||
6894 | if (!rt_task(p)) { | |
6895 | /* | |
6896 | * Renice negative nice level userspace | |
6897 | * tasks back to 0: | |
6898 | */ | |
6899 | if (TASK_NICE(p) < 0 && p->mm) | |
6900 | set_user_nice(p, 0); | |
1da177e4 | 6901 | continue; |
dd41f596 | 6902 | } |
1da177e4 | 6903 | |
b29739f9 IM |
6904 | spin_lock_irqsave(&p->pi_lock, flags); |
6905 | rq = __task_rq_lock(p); | |
1da177e4 | 6906 | |
178be793 | 6907 | normalize_task(rq, p); |
3a5e4dc1 | 6908 | |
b29739f9 IM |
6909 | __task_rq_unlock(rq); |
6910 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6911 | } while_each_thread(g, p); |
6912 | ||
1da177e4 LT |
6913 | read_unlock_irq(&tasklist_lock); |
6914 | } | |
6915 | ||
6916 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6917 | |
6918 | #ifdef CONFIG_IA64 | |
6919 | /* | |
6920 | * These functions are only useful for the IA64 MCA handling. | |
6921 | * | |
6922 | * They can only be called when the whole system has been | |
6923 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6924 | * activity can take place. Using them for anything else would | |
6925 | * be a serious bug, and as a result, they aren't even visible | |
6926 | * under any other configuration. | |
6927 | */ | |
6928 | ||
6929 | /** | |
6930 | * curr_task - return the current task for a given cpu. | |
6931 | * @cpu: the processor in question. | |
6932 | * | |
6933 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6934 | */ | |
36c8b586 | 6935 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6936 | { |
6937 | return cpu_curr(cpu); | |
6938 | } | |
6939 | ||
6940 | /** | |
6941 | * set_curr_task - set the current task for a given cpu. | |
6942 | * @cpu: the processor in question. | |
6943 | * @p: the task pointer to set. | |
6944 | * | |
6945 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
6946 | * are serviced on a separate stack. It allows the architecture to switch the |
6947 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
6948 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6949 | * and caller must save the original value of the current task (see | |
6950 | * curr_task() above) and restore that value before reenabling interrupts and | |
6951 | * re-starting the system. | |
6952 | * | |
6953 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6954 | */ | |
36c8b586 | 6955 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6956 | { |
6957 | cpu_curr(cpu) = p; | |
6958 | } | |
6959 | ||
6960 | #endif | |
29f59db3 SV |
6961 | |
6962 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6963 | ||
29f59db3 | 6964 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6965 | struct task_group *sched_create_group(void) |
29f59db3 | 6966 | { |
4cf86d77 | 6967 | struct task_group *tg; |
29f59db3 SV |
6968 | struct cfs_rq *cfs_rq; |
6969 | struct sched_entity *se; | |
9b5b7751 | 6970 | struct rq *rq; |
29f59db3 SV |
6971 | int i; |
6972 | ||
29f59db3 SV |
6973 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6974 | if (!tg) | |
6975 | return ERR_PTR(-ENOMEM); | |
6976 | ||
9b5b7751 | 6977 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6978 | if (!tg->cfs_rq) |
6979 | goto err; | |
9b5b7751 | 6980 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6981 | if (!tg->se) |
6982 | goto err; | |
6983 | ||
6984 | for_each_possible_cpu(i) { | |
9b5b7751 | 6985 | rq = cpu_rq(i); |
29f59db3 SV |
6986 | |
6987 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6988 | cpu_to_node(i)); | |
6989 | if (!cfs_rq) | |
6990 | goto err; | |
6991 | ||
6992 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6993 | cpu_to_node(i)); | |
6994 | if (!se) | |
6995 | goto err; | |
6996 | ||
6997 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6998 | memset(se, 0, sizeof(struct sched_entity)); | |
6999 | ||
7000 | tg->cfs_rq[i] = cfs_rq; | |
7001 | init_cfs_rq(cfs_rq, rq); | |
7002 | cfs_rq->tg = tg; | |
29f59db3 SV |
7003 | |
7004 | tg->se[i] = se; | |
7005 | se->cfs_rq = &rq->cfs; | |
7006 | se->my_q = cfs_rq; | |
7007 | se->load.weight = NICE_0_LOAD; | |
7008 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7009 | se->parent = NULL; | |
7010 | } | |
7011 | ||
9b5b7751 SV |
7012 | for_each_possible_cpu(i) { |
7013 | rq = cpu_rq(i); | |
7014 | cfs_rq = tg->cfs_rq[i]; | |
7015 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7016 | } | |
29f59db3 | 7017 | |
9b5b7751 | 7018 | tg->shares = NICE_0_LOAD; |
5cb350ba | 7019 | spin_lock_init(&tg->lock); |
29f59db3 | 7020 | |
9b5b7751 | 7021 | return tg; |
29f59db3 SV |
7022 | |
7023 | err: | |
7024 | for_each_possible_cpu(i) { | |
a65914b3 | 7025 | if (tg->cfs_rq) |
29f59db3 | 7026 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7027 | if (tg->se) |
29f59db3 SV |
7028 | kfree(tg->se[i]); |
7029 | } | |
a65914b3 IM |
7030 | kfree(tg->cfs_rq); |
7031 | kfree(tg->se); | |
7032 | kfree(tg); | |
29f59db3 SV |
7033 | |
7034 | return ERR_PTR(-ENOMEM); | |
7035 | } | |
7036 | ||
9b5b7751 SV |
7037 | /* rcu callback to free various structures associated with a task group */ |
7038 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7039 | { |
ae8393e5 SV |
7040 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7041 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7042 | struct sched_entity *se; |
7043 | int i; | |
7044 | ||
29f59db3 SV |
7045 | /* now it should be safe to free those cfs_rqs */ |
7046 | for_each_possible_cpu(i) { | |
7047 | cfs_rq = tg->cfs_rq[i]; | |
7048 | kfree(cfs_rq); | |
7049 | ||
7050 | se = tg->se[i]; | |
7051 | kfree(se); | |
7052 | } | |
7053 | ||
7054 | kfree(tg->cfs_rq); | |
7055 | kfree(tg->se); | |
7056 | kfree(tg); | |
7057 | } | |
7058 | ||
9b5b7751 | 7059 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7060 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7061 | { |
7bae49d4 | 7062 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7063 | int i; |
29f59db3 | 7064 | |
9b5b7751 SV |
7065 | for_each_possible_cpu(i) { |
7066 | cfs_rq = tg->cfs_rq[i]; | |
7067 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7068 | } | |
7069 | ||
7bae49d4 | 7070 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7071 | |
7072 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7073 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7074 | } |
7075 | ||
9b5b7751 | 7076 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7077 | * The caller of this function should have put the task in its new group |
7078 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7079 | * reflect its new group. | |
9b5b7751 SV |
7080 | */ |
7081 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7082 | { |
7083 | int on_rq, running; | |
7084 | unsigned long flags; | |
7085 | struct rq *rq; | |
7086 | ||
7087 | rq = task_rq_lock(tsk, &flags); | |
7088 | ||
dae51f56 | 7089 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7090 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7091 | goto done; |
dae51f56 | 7092 | } |
29f59db3 SV |
7093 | |
7094 | update_rq_clock(rq); | |
7095 | ||
7096 | running = task_running(rq, tsk); | |
7097 | on_rq = tsk->se.on_rq; | |
7098 | ||
83b699ed | 7099 | if (on_rq) { |
29f59db3 | 7100 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7101 | if (unlikely(running)) |
7102 | tsk->sched_class->put_prev_task(rq, tsk); | |
7103 | } | |
29f59db3 | 7104 | |
ce96b5ac | 7105 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7106 | |
83b699ed SV |
7107 | if (on_rq) { |
7108 | if (unlikely(running)) | |
7109 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7110 | enqueue_task(rq, tsk, 0); |
83b699ed | 7111 | } |
29f59db3 SV |
7112 | |
7113 | done: | |
7114 | task_rq_unlock(rq, &flags); | |
7115 | } | |
7116 | ||
7117 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
7118 | { | |
7119 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7120 | struct rq *rq = cfs_rq->rq; | |
7121 | int on_rq; | |
7122 | ||
7123 | spin_lock_irq(&rq->lock); | |
7124 | ||
7125 | on_rq = se->on_rq; | |
7126 | if (on_rq) | |
7127 | dequeue_entity(cfs_rq, se, 0); | |
7128 | ||
7129 | se->load.weight = shares; | |
7130 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7131 | ||
7132 | if (on_rq) | |
7133 | enqueue_entity(cfs_rq, se, 0); | |
7134 | ||
7135 | spin_unlock_irq(&rq->lock); | |
7136 | } | |
7137 | ||
4cf86d77 | 7138 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7139 | { |
7140 | int i; | |
29f59db3 | 7141 | |
5cb350ba | 7142 | spin_lock(&tg->lock); |
9b5b7751 | 7143 | if (tg->shares == shares) |
5cb350ba | 7144 | goto done; |
29f59db3 | 7145 | |
9b5b7751 | 7146 | tg->shares = shares; |
29f59db3 | 7147 | for_each_possible_cpu(i) |
9b5b7751 | 7148 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7149 | |
5cb350ba DG |
7150 | done: |
7151 | spin_unlock(&tg->lock); | |
9b5b7751 | 7152 | return 0; |
29f59db3 SV |
7153 | } |
7154 | ||
5cb350ba DG |
7155 | unsigned long sched_group_shares(struct task_group *tg) |
7156 | { | |
7157 | return tg->shares; | |
7158 | } | |
7159 | ||
3a252015 | 7160 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7161 | |
7162 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7163 | ||
7164 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7165 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7166 | { |
2b01dfe3 PM |
7167 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7168 | struct task_group, css); | |
68318b8e SV |
7169 | } |
7170 | ||
7171 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7172 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7173 | { |
7174 | struct task_group *tg; | |
7175 | ||
2b01dfe3 | 7176 | if (!cgrp->parent) { |
68318b8e | 7177 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7178 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7179 | return &init_task_group.css; |
7180 | } | |
7181 | ||
7182 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7183 | if (cgrp->parent->parent) |
68318b8e SV |
7184 | return ERR_PTR(-EINVAL); |
7185 | ||
7186 | tg = sched_create_group(); | |
7187 | if (IS_ERR(tg)) | |
7188 | return ERR_PTR(-ENOMEM); | |
7189 | ||
7190 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7191 | tg->css.cgroup = cgrp; |
68318b8e SV |
7192 | |
7193 | return &tg->css; | |
7194 | } | |
7195 | ||
41a2d6cf IM |
7196 | static void |
7197 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7198 | { |
2b01dfe3 | 7199 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7200 | |
7201 | sched_destroy_group(tg); | |
7202 | } | |
7203 | ||
41a2d6cf IM |
7204 | static int |
7205 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7206 | struct task_struct *tsk) | |
68318b8e SV |
7207 | { |
7208 | /* We don't support RT-tasks being in separate groups */ | |
7209 | if (tsk->sched_class != &fair_sched_class) | |
7210 | return -EINVAL; | |
7211 | ||
7212 | return 0; | |
7213 | } | |
7214 | ||
7215 | static void | |
2b01dfe3 | 7216 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7217 | struct cgroup *old_cont, struct task_struct *tsk) |
7218 | { | |
7219 | sched_move_task(tsk); | |
7220 | } | |
7221 | ||
2b01dfe3 PM |
7222 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7223 | u64 shareval) | |
68318b8e | 7224 | { |
2b01dfe3 | 7225 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7226 | } |
7227 | ||
2b01dfe3 | 7228 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7229 | { |
2b01dfe3 | 7230 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7231 | |
7232 | return (u64) tg->shares; | |
7233 | } | |
7234 | ||
fe5c7cc2 PM |
7235 | static struct cftype cpu_files[] = { |
7236 | { | |
7237 | .name = "shares", | |
7238 | .read_uint = cpu_shares_read_uint, | |
7239 | .write_uint = cpu_shares_write_uint, | |
7240 | }, | |
68318b8e SV |
7241 | }; |
7242 | ||
7243 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7244 | { | |
fe5c7cc2 | 7245 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7246 | } |
7247 | ||
7248 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7249 | .name = "cpu", |
7250 | .create = cpu_cgroup_create, | |
7251 | .destroy = cpu_cgroup_destroy, | |
7252 | .can_attach = cpu_cgroup_can_attach, | |
7253 | .attach = cpu_cgroup_attach, | |
7254 | .populate = cpu_cgroup_populate, | |
7255 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7256 | .early_init = 1, |
7257 | }; | |
7258 | ||
7259 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7260 | |
7261 | #ifdef CONFIG_CGROUP_CPUACCT | |
7262 | ||
7263 | /* | |
7264 | * CPU accounting code for task groups. | |
7265 | * | |
7266 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7267 | * (balbir@in.ibm.com). | |
7268 | */ | |
7269 | ||
7270 | /* track cpu usage of a group of tasks */ | |
7271 | struct cpuacct { | |
7272 | struct cgroup_subsys_state css; | |
7273 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7274 | u64 *cpuusage; | |
7275 | }; | |
7276 | ||
7277 | struct cgroup_subsys cpuacct_subsys; | |
7278 | ||
7279 | /* return cpu accounting group corresponding to this container */ | |
7280 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7281 | { | |
7282 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7283 | struct cpuacct, css); | |
7284 | } | |
7285 | ||
7286 | /* return cpu accounting group to which this task belongs */ | |
7287 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7288 | { | |
7289 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7290 | struct cpuacct, css); | |
7291 | } | |
7292 | ||
7293 | /* create a new cpu accounting group */ | |
7294 | static struct cgroup_subsys_state *cpuacct_create( | |
7295 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7296 | { | |
7297 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7298 | ||
7299 | if (!ca) | |
7300 | return ERR_PTR(-ENOMEM); | |
7301 | ||
7302 | ca->cpuusage = alloc_percpu(u64); | |
7303 | if (!ca->cpuusage) { | |
7304 | kfree(ca); | |
7305 | return ERR_PTR(-ENOMEM); | |
7306 | } | |
7307 | ||
7308 | return &ca->css; | |
7309 | } | |
7310 | ||
7311 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7312 | static void |
7313 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7314 | { |
7315 | struct cpuacct *ca = cgroup_ca(cont); | |
7316 | ||
7317 | free_percpu(ca->cpuusage); | |
7318 | kfree(ca); | |
7319 | } | |
7320 | ||
7321 | /* return total cpu usage (in nanoseconds) of a group */ | |
7322 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7323 | { | |
7324 | struct cpuacct *ca = cgroup_ca(cont); | |
7325 | u64 totalcpuusage = 0; | |
7326 | int i; | |
7327 | ||
7328 | for_each_possible_cpu(i) { | |
7329 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7330 | ||
7331 | /* | |
7332 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7333 | * platforms. | |
7334 | */ | |
7335 | spin_lock_irq(&cpu_rq(i)->lock); | |
7336 | totalcpuusage += *cpuusage; | |
7337 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7338 | } | |
7339 | ||
7340 | return totalcpuusage; | |
7341 | } | |
7342 | ||
7343 | static struct cftype files[] = { | |
7344 | { | |
7345 | .name = "usage", | |
7346 | .read_uint = cpuusage_read, | |
7347 | }, | |
7348 | }; | |
7349 | ||
7350 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7351 | { | |
7352 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7353 | } | |
7354 | ||
7355 | /* | |
7356 | * charge this task's execution time to its accounting group. | |
7357 | * | |
7358 | * called with rq->lock held. | |
7359 | */ | |
7360 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7361 | { | |
7362 | struct cpuacct *ca; | |
7363 | ||
7364 | if (!cpuacct_subsys.active) | |
7365 | return; | |
7366 | ||
7367 | ca = task_ca(tsk); | |
7368 | if (ca) { | |
7369 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7370 | ||
7371 | *cpuusage += cputime; | |
7372 | } | |
7373 | } | |
7374 | ||
7375 | struct cgroup_subsys cpuacct_subsys = { | |
7376 | .name = "cpuacct", | |
7377 | .create = cpuacct_create, | |
7378 | .destroy = cpuacct_destroy, | |
7379 | .populate = cpuacct_populate, | |
7380 | .subsys_id = cpuacct_subsys_id, | |
7381 | }; | |
7382 | #endif /* CONFIG_CGROUP_CPUACCT */ |