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 |
93f992cc | 194 | # define INIT_TASK_GROUP_LOAD 2*NICE_0_LOAD |
24e377a8 | 195 | #else |
93f992cc | 196 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
24e377a8 SV |
197 | #endif |
198 | ||
93f992cc | 199 | static int init_task_group_load = INIT_TASK_GROUP_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 | 490 | rq = cpu_rq(cpu); |
8ced5f69 IM |
491 | /* |
492 | * Only call sched_clock() if the scheduler has already been | |
493 | * initialized (some code might call cpu_clock() very early): | |
494 | */ | |
495 | if (rq->idle) | |
496 | update_rq_clock(rq); | |
b04a0f4c | 497 | now = rq->clock; |
2cd4d0ea | 498 | local_irq_restore(flags); |
e436d800 IM |
499 | |
500 | return now; | |
501 | } | |
a58f6f25 | 502 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 503 | |
1da177e4 | 504 | #ifndef prepare_arch_switch |
4866cde0 NP |
505 | # define prepare_arch_switch(next) do { } while (0) |
506 | #endif | |
507 | #ifndef finish_arch_switch | |
508 | # define finish_arch_switch(prev) do { } while (0) | |
509 | #endif | |
510 | ||
051a1d1a DA |
511 | static inline int task_current(struct rq *rq, struct task_struct *p) |
512 | { | |
513 | return rq->curr == p; | |
514 | } | |
515 | ||
4866cde0 | 516 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 517 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 518 | { |
051a1d1a | 519 | return task_current(rq, p); |
4866cde0 NP |
520 | } |
521 | ||
70b97a7f | 522 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
523 | { |
524 | } | |
525 | ||
70b97a7f | 526 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 527 | { |
da04c035 IM |
528 | #ifdef CONFIG_DEBUG_SPINLOCK |
529 | /* this is a valid case when another task releases the spinlock */ | |
530 | rq->lock.owner = current; | |
531 | #endif | |
8a25d5de IM |
532 | /* |
533 | * If we are tracking spinlock dependencies then we have to | |
534 | * fix up the runqueue lock - which gets 'carried over' from | |
535 | * prev into current: | |
536 | */ | |
537 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
538 | ||
4866cde0 NP |
539 | spin_unlock_irq(&rq->lock); |
540 | } | |
541 | ||
542 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 543 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
544 | { |
545 | #ifdef CONFIG_SMP | |
546 | return p->oncpu; | |
547 | #else | |
051a1d1a | 548 | return task_current(rq, p); |
4866cde0 NP |
549 | #endif |
550 | } | |
551 | ||
70b97a7f | 552 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
553 | { |
554 | #ifdef CONFIG_SMP | |
555 | /* | |
556 | * We can optimise this out completely for !SMP, because the | |
557 | * SMP rebalancing from interrupt is the only thing that cares | |
558 | * here. | |
559 | */ | |
560 | next->oncpu = 1; | |
561 | #endif | |
562 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
563 | spin_unlock_irq(&rq->lock); | |
564 | #else | |
565 | spin_unlock(&rq->lock); | |
566 | #endif | |
567 | } | |
568 | ||
70b97a7f | 569 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
570 | { |
571 | #ifdef CONFIG_SMP | |
572 | /* | |
573 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
574 | * We must ensure this doesn't happen until the switch is completely | |
575 | * finished. | |
576 | */ | |
577 | smp_wmb(); | |
578 | prev->oncpu = 0; | |
579 | #endif | |
580 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
581 | local_irq_enable(); | |
1da177e4 | 582 | #endif |
4866cde0 NP |
583 | } |
584 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 585 | |
b29739f9 IM |
586 | /* |
587 | * __task_rq_lock - lock the runqueue a given task resides on. | |
588 | * Must be called interrupts disabled. | |
589 | */ | |
70b97a7f | 590 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
591 | __acquires(rq->lock) |
592 | { | |
3a5c359a AK |
593 | for (;;) { |
594 | struct rq *rq = task_rq(p); | |
595 | spin_lock(&rq->lock); | |
596 | if (likely(rq == task_rq(p))) | |
597 | return rq; | |
b29739f9 | 598 | spin_unlock(&rq->lock); |
b29739f9 | 599 | } |
b29739f9 IM |
600 | } |
601 | ||
1da177e4 LT |
602 | /* |
603 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 604 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
605 | * explicitly disabling preemption. |
606 | */ | |
70b97a7f | 607 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
608 | __acquires(rq->lock) |
609 | { | |
70b97a7f | 610 | struct rq *rq; |
1da177e4 | 611 | |
3a5c359a AK |
612 | for (;;) { |
613 | local_irq_save(*flags); | |
614 | rq = task_rq(p); | |
615 | spin_lock(&rq->lock); | |
616 | if (likely(rq == task_rq(p))) | |
617 | return rq; | |
1da177e4 | 618 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 619 | } |
1da177e4 LT |
620 | } |
621 | ||
a9957449 | 622 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
623 | __releases(rq->lock) |
624 | { | |
625 | spin_unlock(&rq->lock); | |
626 | } | |
627 | ||
70b97a7f | 628 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
629 | __releases(rq->lock) |
630 | { | |
631 | spin_unlock_irqrestore(&rq->lock, *flags); | |
632 | } | |
633 | ||
1da177e4 | 634 | /* |
cc2a73b5 | 635 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 636 | */ |
a9957449 | 637 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
638 | __acquires(rq->lock) |
639 | { | |
70b97a7f | 640 | struct rq *rq; |
1da177e4 LT |
641 | |
642 | local_irq_disable(); | |
643 | rq = this_rq(); | |
644 | spin_lock(&rq->lock); | |
645 | ||
646 | return rq; | |
647 | } | |
648 | ||
1b9f19c2 | 649 | /* |
2aa44d05 | 650 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 651 | */ |
2aa44d05 | 652 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 653 | { |
2aa44d05 IM |
654 | struct rq *rq = cpu_rq(smp_processor_id()); |
655 | ||
656 | spin_lock(&rq->lock); | |
657 | __update_rq_clock(rq); | |
658 | spin_unlock(&rq->lock); | |
659 | rq->clock_deep_idle_events++; | |
660 | } | |
661 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
662 | ||
663 | /* | |
664 | * We just idled delta nanoseconds (called with irqs disabled): | |
665 | */ | |
666 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
667 | { | |
668 | struct rq *rq = cpu_rq(smp_processor_id()); | |
669 | u64 now = sched_clock(); | |
1b9f19c2 | 670 | |
2bacec8c | 671 | touch_softlockup_watchdog(); |
2aa44d05 IM |
672 | rq->idle_clock += delta_ns; |
673 | /* | |
674 | * Override the previous timestamp and ignore all | |
675 | * sched_clock() deltas that occured while we idled, | |
676 | * and use the PM-provided delta_ns to advance the | |
677 | * rq clock: | |
678 | */ | |
679 | spin_lock(&rq->lock); | |
680 | rq->prev_clock_raw = now; | |
681 | rq->clock += delta_ns; | |
682 | spin_unlock(&rq->lock); | |
1b9f19c2 | 683 | } |
2aa44d05 | 684 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 685 | |
c24d20db IM |
686 | /* |
687 | * resched_task - mark a task 'to be rescheduled now'. | |
688 | * | |
689 | * On UP this means the setting of the need_resched flag, on SMP it | |
690 | * might also involve a cross-CPU call to trigger the scheduler on | |
691 | * the target CPU. | |
692 | */ | |
693 | #ifdef CONFIG_SMP | |
694 | ||
695 | #ifndef tsk_is_polling | |
696 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
697 | #endif | |
698 | ||
699 | static void resched_task(struct task_struct *p) | |
700 | { | |
701 | int cpu; | |
702 | ||
703 | assert_spin_locked(&task_rq(p)->lock); | |
704 | ||
705 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
706 | return; | |
707 | ||
708 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
709 | ||
710 | cpu = task_cpu(p); | |
711 | if (cpu == smp_processor_id()) | |
712 | return; | |
713 | ||
714 | /* NEED_RESCHED must be visible before we test polling */ | |
715 | smp_mb(); | |
716 | if (!tsk_is_polling(p)) | |
717 | smp_send_reschedule(cpu); | |
718 | } | |
719 | ||
720 | static void resched_cpu(int cpu) | |
721 | { | |
722 | struct rq *rq = cpu_rq(cpu); | |
723 | unsigned long flags; | |
724 | ||
725 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
726 | return; | |
727 | resched_task(cpu_curr(cpu)); | |
728 | spin_unlock_irqrestore(&rq->lock, flags); | |
729 | } | |
730 | #else | |
731 | static inline void resched_task(struct task_struct *p) | |
732 | { | |
733 | assert_spin_locked(&task_rq(p)->lock); | |
734 | set_tsk_need_resched(p); | |
735 | } | |
736 | #endif | |
737 | ||
45bf76df IM |
738 | #if BITS_PER_LONG == 32 |
739 | # define WMULT_CONST (~0UL) | |
740 | #else | |
741 | # define WMULT_CONST (1UL << 32) | |
742 | #endif | |
743 | ||
744 | #define WMULT_SHIFT 32 | |
745 | ||
194081eb IM |
746 | /* |
747 | * Shift right and round: | |
748 | */ | |
cf2ab469 | 749 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 750 | |
cb1c4fc9 | 751 | static unsigned long |
45bf76df IM |
752 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
753 | struct load_weight *lw) | |
754 | { | |
755 | u64 tmp; | |
756 | ||
757 | if (unlikely(!lw->inv_weight)) | |
194081eb | 758 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
759 | |
760 | tmp = (u64)delta_exec * weight; | |
761 | /* | |
762 | * Check whether we'd overflow the 64-bit multiplication: | |
763 | */ | |
194081eb | 764 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 765 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
766 | WMULT_SHIFT/2); |
767 | else | |
cf2ab469 | 768 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 769 | |
ecf691da | 770 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
771 | } |
772 | ||
773 | static inline unsigned long | |
774 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
775 | { | |
776 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
777 | } | |
778 | ||
1091985b | 779 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
780 | { |
781 | lw->weight += inc; | |
45bf76df IM |
782 | } |
783 | ||
1091985b | 784 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
785 | { |
786 | lw->weight -= dec; | |
45bf76df IM |
787 | } |
788 | ||
2dd73a4f PW |
789 | /* |
790 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
791 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
792 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 793 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
794 | * scaled version of the new time slice allocation that they receive on time |
795 | * slice expiry etc. | |
796 | */ | |
797 | ||
dd41f596 IM |
798 | #define WEIGHT_IDLEPRIO 2 |
799 | #define WMULT_IDLEPRIO (1 << 31) | |
800 | ||
801 | /* | |
802 | * Nice levels are multiplicative, with a gentle 10% change for every | |
803 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
804 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
805 | * that remained on nice 0. | |
806 | * | |
807 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
808 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
809 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
810 | * If a task goes up by ~10% and another task goes down by ~10% then | |
811 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
812 | */ |
813 | static const int prio_to_weight[40] = { | |
254753dc IM |
814 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
815 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
816 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
817 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
818 | /* 0 */ 1024, 820, 655, 526, 423, | |
819 | /* 5 */ 335, 272, 215, 172, 137, | |
820 | /* 10 */ 110, 87, 70, 56, 45, | |
821 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
822 | }; |
823 | ||
5714d2de IM |
824 | /* |
825 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
826 | * | |
827 | * In cases where the weight does not change often, we can use the | |
828 | * precalculated inverse to speed up arithmetics by turning divisions | |
829 | * into multiplications: | |
830 | */ | |
dd41f596 | 831 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
832 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
833 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
834 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
835 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
836 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
837 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
838 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
839 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 840 | }; |
2dd73a4f | 841 | |
dd41f596 IM |
842 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
843 | ||
844 | /* | |
845 | * runqueue iterator, to support SMP load-balancing between different | |
846 | * scheduling classes, without having to expose their internal data | |
847 | * structures to the load-balancing proper: | |
848 | */ | |
849 | struct rq_iterator { | |
850 | void *arg; | |
851 | struct task_struct *(*start)(void *); | |
852 | struct task_struct *(*next)(void *); | |
853 | }; | |
854 | ||
e1d1484f PW |
855 | #ifdef CONFIG_SMP |
856 | static unsigned long | |
857 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
858 | unsigned long max_load_move, struct sched_domain *sd, | |
859 | enum cpu_idle_type idle, int *all_pinned, | |
860 | int *this_best_prio, struct rq_iterator *iterator); | |
861 | ||
862 | static int | |
863 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
864 | struct sched_domain *sd, enum cpu_idle_type idle, | |
865 | struct rq_iterator *iterator); | |
e1d1484f | 866 | #endif |
dd41f596 | 867 | |
d842de87 SV |
868 | #ifdef CONFIG_CGROUP_CPUACCT |
869 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
870 | #else | |
871 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
872 | #endif | |
873 | ||
dd41f596 | 874 | #include "sched_stats.h" |
dd41f596 | 875 | #include "sched_idletask.c" |
5522d5d5 IM |
876 | #include "sched_fair.c" |
877 | #include "sched_rt.c" | |
dd41f596 IM |
878 | #ifdef CONFIG_SCHED_DEBUG |
879 | # include "sched_debug.c" | |
880 | #endif | |
881 | ||
882 | #define sched_class_highest (&rt_sched_class) | |
883 | ||
29b4b623 | 884 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 885 | { |
495eca49 | 886 | update_load_add(&rq->load, p->se.load.weight); |
9c217245 IM |
887 | } |
888 | ||
79b5dddf | 889 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
9c217245 | 890 | { |
495eca49 | 891 | update_load_sub(&rq->load, p->se.load.weight); |
9c217245 IM |
892 | } |
893 | ||
e5fa2237 | 894 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
895 | { |
896 | rq->nr_running++; | |
29b4b623 | 897 | inc_load(rq, p); |
9c217245 IM |
898 | } |
899 | ||
db53181e | 900 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
901 | { |
902 | rq->nr_running--; | |
79b5dddf | 903 | dec_load(rq, p); |
9c217245 IM |
904 | } |
905 | ||
45bf76df IM |
906 | static void set_load_weight(struct task_struct *p) |
907 | { | |
908 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
909 | p->se.load.weight = prio_to_weight[0] * 2; |
910 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
911 | return; | |
912 | } | |
45bf76df | 913 | |
dd41f596 IM |
914 | /* |
915 | * SCHED_IDLE tasks get minimal weight: | |
916 | */ | |
917 | if (p->policy == SCHED_IDLE) { | |
918 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
919 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
920 | return; | |
921 | } | |
71f8bd46 | 922 | |
dd41f596 IM |
923 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
924 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
925 | } |
926 | ||
8159f87e | 927 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 928 | { |
dd41f596 | 929 | sched_info_queued(p); |
fd390f6a | 930 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 931 | p->se.on_rq = 1; |
71f8bd46 IM |
932 | } |
933 | ||
69be72c1 | 934 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 935 | { |
f02231e5 | 936 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 937 | p->se.on_rq = 0; |
71f8bd46 IM |
938 | } |
939 | ||
14531189 | 940 | /* |
dd41f596 | 941 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 942 | */ |
14531189 IM |
943 | static inline int __normal_prio(struct task_struct *p) |
944 | { | |
dd41f596 | 945 | return p->static_prio; |
14531189 IM |
946 | } |
947 | ||
b29739f9 IM |
948 | /* |
949 | * Calculate the expected normal priority: i.e. priority | |
950 | * without taking RT-inheritance into account. Might be | |
951 | * boosted by interactivity modifiers. Changes upon fork, | |
952 | * setprio syscalls, and whenever the interactivity | |
953 | * estimator recalculates. | |
954 | */ | |
36c8b586 | 955 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
956 | { |
957 | int prio; | |
958 | ||
e05606d3 | 959 | if (task_has_rt_policy(p)) |
b29739f9 IM |
960 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
961 | else | |
962 | prio = __normal_prio(p); | |
963 | return prio; | |
964 | } | |
965 | ||
966 | /* | |
967 | * Calculate the current priority, i.e. the priority | |
968 | * taken into account by the scheduler. This value might | |
969 | * be boosted by RT tasks, or might be boosted by | |
970 | * interactivity modifiers. Will be RT if the task got | |
971 | * RT-boosted. If not then it returns p->normal_prio. | |
972 | */ | |
36c8b586 | 973 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
974 | { |
975 | p->normal_prio = normal_prio(p); | |
976 | /* | |
977 | * If we are RT tasks or we were boosted to RT priority, | |
978 | * keep the priority unchanged. Otherwise, update priority | |
979 | * to the normal priority: | |
980 | */ | |
981 | if (!rt_prio(p->prio)) | |
982 | return p->normal_prio; | |
983 | return p->prio; | |
984 | } | |
985 | ||
1da177e4 | 986 | /* |
dd41f596 | 987 | * activate_task - move a task to the runqueue. |
1da177e4 | 988 | */ |
dd41f596 | 989 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 990 | { |
dd41f596 IM |
991 | if (p->state == TASK_UNINTERRUPTIBLE) |
992 | rq->nr_uninterruptible--; | |
1da177e4 | 993 | |
8159f87e | 994 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 995 | inc_nr_running(p, rq); |
1da177e4 LT |
996 | } |
997 | ||
1da177e4 LT |
998 | /* |
999 | * deactivate_task - remove a task from the runqueue. | |
1000 | */ | |
2e1cb74a | 1001 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1002 | { |
dd41f596 IM |
1003 | if (p->state == TASK_UNINTERRUPTIBLE) |
1004 | rq->nr_uninterruptible++; | |
1005 | ||
69be72c1 | 1006 | dequeue_task(rq, p, sleep); |
db53181e | 1007 | dec_nr_running(p, rq); |
1da177e4 LT |
1008 | } |
1009 | ||
1da177e4 LT |
1010 | /** |
1011 | * task_curr - is this task currently executing on a CPU? | |
1012 | * @p: the task in question. | |
1013 | */ | |
36c8b586 | 1014 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1015 | { |
1016 | return cpu_curr(task_cpu(p)) == p; | |
1017 | } | |
1018 | ||
2dd73a4f PW |
1019 | /* Used instead of source_load when we know the type == 0 */ |
1020 | unsigned long weighted_cpuload(const int cpu) | |
1021 | { | |
495eca49 | 1022 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1023 | } |
1024 | ||
1025 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1026 | { | |
ce96b5ac | 1027 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1028 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1029 | /* |
1030 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1031 | * successfuly executed on another CPU. We must ensure that updates of | |
1032 | * per-task data have been completed by this moment. | |
1033 | */ | |
1034 | smp_wmb(); | |
dd41f596 | 1035 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1036 | #endif |
2dd73a4f PW |
1037 | } |
1038 | ||
1da177e4 | 1039 | #ifdef CONFIG_SMP |
c65cc870 | 1040 | |
cc367732 IM |
1041 | /* |
1042 | * Is this task likely cache-hot: | |
1043 | */ | |
1044 | static inline int | |
1045 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1046 | { | |
1047 | s64 delta; | |
1048 | ||
1049 | if (p->sched_class != &fair_sched_class) | |
1050 | return 0; | |
1051 | ||
6bc1665b IM |
1052 | if (sysctl_sched_migration_cost == -1) |
1053 | return 1; | |
1054 | if (sysctl_sched_migration_cost == 0) | |
1055 | return 0; | |
1056 | ||
cc367732 IM |
1057 | delta = now - p->se.exec_start; |
1058 | ||
1059 | return delta < (s64)sysctl_sched_migration_cost; | |
1060 | } | |
1061 | ||
1062 | ||
dd41f596 | 1063 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1064 | { |
dd41f596 IM |
1065 | int old_cpu = task_cpu(p); |
1066 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1067 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1068 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1069 | u64 clock_offset; |
dd41f596 IM |
1070 | |
1071 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1072 | |
1073 | #ifdef CONFIG_SCHEDSTATS | |
1074 | if (p->se.wait_start) | |
1075 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1076 | if (p->se.sleep_start) |
1077 | p->se.sleep_start -= clock_offset; | |
1078 | if (p->se.block_start) | |
1079 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1080 | if (old_cpu != new_cpu) { |
1081 | schedstat_inc(p, se.nr_migrations); | |
1082 | if (task_hot(p, old_rq->clock, NULL)) | |
1083 | schedstat_inc(p, se.nr_forced2_migrations); | |
1084 | } | |
6cfb0d5d | 1085 | #endif |
2830cf8c SV |
1086 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1087 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1088 | |
1089 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1090 | } |
1091 | ||
70b97a7f | 1092 | struct migration_req { |
1da177e4 | 1093 | struct list_head list; |
1da177e4 | 1094 | |
36c8b586 | 1095 | struct task_struct *task; |
1da177e4 LT |
1096 | int dest_cpu; |
1097 | ||
1da177e4 | 1098 | struct completion done; |
70b97a7f | 1099 | }; |
1da177e4 LT |
1100 | |
1101 | /* | |
1102 | * The task's runqueue lock must be held. | |
1103 | * Returns true if you have to wait for migration thread. | |
1104 | */ | |
36c8b586 | 1105 | static int |
70b97a7f | 1106 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1107 | { |
70b97a7f | 1108 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1109 | |
1110 | /* | |
1111 | * If the task is not on a runqueue (and not running), then | |
1112 | * it is sufficient to simply update the task's cpu field. | |
1113 | */ | |
dd41f596 | 1114 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1115 | set_task_cpu(p, dest_cpu); |
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | init_completion(&req->done); | |
1da177e4 LT |
1120 | req->task = p; |
1121 | req->dest_cpu = dest_cpu; | |
1122 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1123 | |
1da177e4 LT |
1124 | return 1; |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * wait_task_inactive - wait for a thread to unschedule. | |
1129 | * | |
1130 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1131 | * else this function might spin for a *long* time. This function can't | |
1132 | * be called with interrupts off, or it may introduce deadlock with | |
1133 | * smp_call_function() if an IPI is sent by the same process we are | |
1134 | * waiting to become inactive. | |
1135 | */ | |
36c8b586 | 1136 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1137 | { |
1138 | unsigned long flags; | |
dd41f596 | 1139 | int running, on_rq; |
70b97a7f | 1140 | struct rq *rq; |
1da177e4 | 1141 | |
3a5c359a AK |
1142 | for (;;) { |
1143 | /* | |
1144 | * We do the initial early heuristics without holding | |
1145 | * any task-queue locks at all. We'll only try to get | |
1146 | * the runqueue lock when things look like they will | |
1147 | * work out! | |
1148 | */ | |
1149 | rq = task_rq(p); | |
fa490cfd | 1150 | |
3a5c359a AK |
1151 | /* |
1152 | * If the task is actively running on another CPU | |
1153 | * still, just relax and busy-wait without holding | |
1154 | * any locks. | |
1155 | * | |
1156 | * NOTE! Since we don't hold any locks, it's not | |
1157 | * even sure that "rq" stays as the right runqueue! | |
1158 | * But we don't care, since "task_running()" will | |
1159 | * return false if the runqueue has changed and p | |
1160 | * is actually now running somewhere else! | |
1161 | */ | |
1162 | while (task_running(rq, p)) | |
1163 | cpu_relax(); | |
fa490cfd | 1164 | |
3a5c359a AK |
1165 | /* |
1166 | * Ok, time to look more closely! We need the rq | |
1167 | * lock now, to be *sure*. If we're wrong, we'll | |
1168 | * just go back and repeat. | |
1169 | */ | |
1170 | rq = task_rq_lock(p, &flags); | |
1171 | running = task_running(rq, p); | |
1172 | on_rq = p->se.on_rq; | |
1173 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1174 | |
3a5c359a AK |
1175 | /* |
1176 | * Was it really running after all now that we | |
1177 | * checked with the proper locks actually held? | |
1178 | * | |
1179 | * Oops. Go back and try again.. | |
1180 | */ | |
1181 | if (unlikely(running)) { | |
1182 | cpu_relax(); | |
1183 | continue; | |
1184 | } | |
fa490cfd | 1185 | |
3a5c359a AK |
1186 | /* |
1187 | * It's not enough that it's not actively running, | |
1188 | * it must be off the runqueue _entirely_, and not | |
1189 | * preempted! | |
1190 | * | |
1191 | * So if it wa still runnable (but just not actively | |
1192 | * running right now), it's preempted, and we should | |
1193 | * yield - it could be a while. | |
1194 | */ | |
1195 | if (unlikely(on_rq)) { | |
1196 | schedule_timeout_uninterruptible(1); | |
1197 | continue; | |
1198 | } | |
fa490cfd | 1199 | |
3a5c359a AK |
1200 | /* |
1201 | * Ahh, all good. It wasn't running, and it wasn't | |
1202 | * runnable, which means that it will never become | |
1203 | * running in the future either. We're all done! | |
1204 | */ | |
1205 | break; | |
1206 | } | |
1da177e4 LT |
1207 | } |
1208 | ||
1209 | /*** | |
1210 | * kick_process - kick a running thread to enter/exit the kernel | |
1211 | * @p: the to-be-kicked thread | |
1212 | * | |
1213 | * Cause a process which is running on another CPU to enter | |
1214 | * kernel-mode, without any delay. (to get signals handled.) | |
1215 | * | |
1216 | * NOTE: this function doesnt have to take the runqueue lock, | |
1217 | * because all it wants to ensure is that the remote task enters | |
1218 | * the kernel. If the IPI races and the task has been migrated | |
1219 | * to another CPU then no harm is done and the purpose has been | |
1220 | * achieved as well. | |
1221 | */ | |
36c8b586 | 1222 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1223 | { |
1224 | int cpu; | |
1225 | ||
1226 | preempt_disable(); | |
1227 | cpu = task_cpu(p); | |
1228 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1229 | smp_send_reschedule(cpu); | |
1230 | preempt_enable(); | |
1231 | } | |
1232 | ||
1233 | /* | |
2dd73a4f PW |
1234 | * Return a low guess at the load of a migration-source cpu weighted |
1235 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1236 | * |
1237 | * We want to under-estimate the load of migration sources, to | |
1238 | * balance conservatively. | |
1239 | */ | |
a9957449 | 1240 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1241 | { |
70b97a7f | 1242 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1243 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1244 | |
3b0bd9bc | 1245 | if (type == 0) |
dd41f596 | 1246 | return total; |
b910472d | 1247 | |
dd41f596 | 1248 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1249 | } |
1250 | ||
1251 | /* | |
2dd73a4f PW |
1252 | * Return a high guess at the load of a migration-target cpu weighted |
1253 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1254 | */ |
a9957449 | 1255 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1256 | { |
70b97a7f | 1257 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1258 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1259 | |
7897986b | 1260 | if (type == 0) |
dd41f596 | 1261 | return total; |
3b0bd9bc | 1262 | |
dd41f596 | 1263 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1264 | } |
1265 | ||
1266 | /* | |
1267 | * Return the average load per task on the cpu's run queue | |
1268 | */ | |
1269 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1270 | { | |
70b97a7f | 1271 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1272 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1273 | unsigned long n = rq->nr_running; |
1274 | ||
dd41f596 | 1275 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1276 | } |
1277 | ||
147cbb4b NP |
1278 | /* |
1279 | * find_idlest_group finds and returns the least busy CPU group within the | |
1280 | * domain. | |
1281 | */ | |
1282 | static struct sched_group * | |
1283 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1284 | { | |
1285 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1286 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1287 | int load_idx = sd->forkexec_idx; | |
1288 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1289 | ||
1290 | do { | |
1291 | unsigned long load, avg_load; | |
1292 | int local_group; | |
1293 | int i; | |
1294 | ||
da5a5522 BD |
1295 | /* Skip over this group if it has no CPUs allowed */ |
1296 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1297 | continue; |
da5a5522 | 1298 | |
147cbb4b | 1299 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1300 | |
1301 | /* Tally up the load of all CPUs in the group */ | |
1302 | avg_load = 0; | |
1303 | ||
1304 | for_each_cpu_mask(i, group->cpumask) { | |
1305 | /* Bias balancing toward cpus of our domain */ | |
1306 | if (local_group) | |
1307 | load = source_load(i, load_idx); | |
1308 | else | |
1309 | load = target_load(i, load_idx); | |
1310 | ||
1311 | avg_load += load; | |
1312 | } | |
1313 | ||
1314 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1315 | avg_load = sg_div_cpu_power(group, |
1316 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1317 | |
1318 | if (local_group) { | |
1319 | this_load = avg_load; | |
1320 | this = group; | |
1321 | } else if (avg_load < min_load) { | |
1322 | min_load = avg_load; | |
1323 | idlest = group; | |
1324 | } | |
3a5c359a | 1325 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1326 | |
1327 | if (!idlest || 100*this_load < imbalance*min_load) | |
1328 | return NULL; | |
1329 | return idlest; | |
1330 | } | |
1331 | ||
1332 | /* | |
0feaece9 | 1333 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1334 | */ |
95cdf3b7 IM |
1335 | static int |
1336 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1337 | { |
da5a5522 | 1338 | cpumask_t tmp; |
147cbb4b NP |
1339 | unsigned long load, min_load = ULONG_MAX; |
1340 | int idlest = -1; | |
1341 | int i; | |
1342 | ||
da5a5522 BD |
1343 | /* Traverse only the allowed CPUs */ |
1344 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1345 | ||
1346 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1347 | load = weighted_cpuload(i); |
147cbb4b NP |
1348 | |
1349 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1350 | min_load = load; | |
1351 | idlest = i; | |
1352 | } | |
1353 | } | |
1354 | ||
1355 | return idlest; | |
1356 | } | |
1357 | ||
476d139c NP |
1358 | /* |
1359 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1360 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1361 | * SD_BALANCE_EXEC. | |
1362 | * | |
1363 | * Balance, ie. select the least loaded group. | |
1364 | * | |
1365 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1366 | * | |
1367 | * preempt must be disabled. | |
1368 | */ | |
1369 | static int sched_balance_self(int cpu, int flag) | |
1370 | { | |
1371 | struct task_struct *t = current; | |
1372 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1373 | |
c96d145e | 1374 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1375 | /* |
1376 | * If power savings logic is enabled for a domain, stop there. | |
1377 | */ | |
5c45bf27 SS |
1378 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1379 | break; | |
476d139c NP |
1380 | if (tmp->flags & flag) |
1381 | sd = tmp; | |
c96d145e | 1382 | } |
476d139c NP |
1383 | |
1384 | while (sd) { | |
1385 | cpumask_t span; | |
1386 | struct sched_group *group; | |
1a848870 SS |
1387 | int new_cpu, weight; |
1388 | ||
1389 | if (!(sd->flags & flag)) { | |
1390 | sd = sd->child; | |
1391 | continue; | |
1392 | } | |
476d139c NP |
1393 | |
1394 | span = sd->span; | |
1395 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1396 | if (!group) { |
1397 | sd = sd->child; | |
1398 | continue; | |
1399 | } | |
476d139c | 1400 | |
da5a5522 | 1401 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1402 | if (new_cpu == -1 || new_cpu == cpu) { |
1403 | /* Now try balancing at a lower domain level of cpu */ | |
1404 | sd = sd->child; | |
1405 | continue; | |
1406 | } | |
476d139c | 1407 | |
1a848870 | 1408 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1409 | cpu = new_cpu; |
476d139c NP |
1410 | sd = NULL; |
1411 | weight = cpus_weight(span); | |
1412 | for_each_domain(cpu, tmp) { | |
1413 | if (weight <= cpus_weight(tmp->span)) | |
1414 | break; | |
1415 | if (tmp->flags & flag) | |
1416 | sd = tmp; | |
1417 | } | |
1418 | /* while loop will break here if sd == NULL */ | |
1419 | } | |
1420 | ||
1421 | return cpu; | |
1422 | } | |
1423 | ||
1424 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1425 | |
1426 | /* | |
1427 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1428 | * not idle and an idle cpu is available. The span of cpus to | |
1429 | * search starts with cpus closest then further out as needed, | |
1430 | * so we always favor a closer, idle cpu. | |
1431 | * | |
1432 | * Returns the CPU we should wake onto. | |
1433 | */ | |
1434 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1435 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1436 | { |
1437 | cpumask_t tmp; | |
1438 | struct sched_domain *sd; | |
1439 | int i; | |
1440 | ||
4953198b SS |
1441 | /* |
1442 | * If it is idle, then it is the best cpu to run this task. | |
1443 | * | |
1444 | * This cpu is also the best, if it has more than one task already. | |
1445 | * Siblings must be also busy(in most cases) as they didn't already | |
1446 | * pickup the extra load from this cpu and hence we need not check | |
1447 | * sibling runqueue info. This will avoid the checks and cache miss | |
1448 | * penalities associated with that. | |
1449 | */ | |
1450 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1451 | return cpu; |
1452 | ||
1453 | for_each_domain(cpu, sd) { | |
1454 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1455 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1456 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1457 | if (idle_cpu(i)) { |
1458 | if (i != task_cpu(p)) { | |
1459 | schedstat_inc(p, | |
1460 | se.nr_wakeups_idle); | |
1461 | } | |
1da177e4 | 1462 | return i; |
cc367732 | 1463 | } |
1da177e4 | 1464 | } |
9761eea8 | 1465 | } else { |
e0f364f4 | 1466 | break; |
9761eea8 | 1467 | } |
1da177e4 LT |
1468 | } |
1469 | return cpu; | |
1470 | } | |
1471 | #else | |
36c8b586 | 1472 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1473 | { |
1474 | return cpu; | |
1475 | } | |
1476 | #endif | |
1477 | ||
1478 | /*** | |
1479 | * try_to_wake_up - wake up a thread | |
1480 | * @p: the to-be-woken-up thread | |
1481 | * @state: the mask of task states that can be woken | |
1482 | * @sync: do a synchronous wakeup? | |
1483 | * | |
1484 | * Put it on the run-queue if it's not already there. The "current" | |
1485 | * thread is always on the run-queue (except when the actual | |
1486 | * re-schedule is in progress), and as such you're allowed to do | |
1487 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1488 | * runnable without the overhead of this. | |
1489 | * | |
1490 | * returns failure only if the task is already active. | |
1491 | */ | |
36c8b586 | 1492 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1493 | { |
cc367732 | 1494 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1495 | unsigned long flags; |
1496 | long old_state; | |
70b97a7f | 1497 | struct rq *rq; |
1da177e4 | 1498 | #ifdef CONFIG_SMP |
7897986b | 1499 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1500 | unsigned long load, this_load; |
1da177e4 LT |
1501 | int new_cpu; |
1502 | #endif | |
1503 | ||
1504 | rq = task_rq_lock(p, &flags); | |
1505 | old_state = p->state; | |
1506 | if (!(old_state & state)) | |
1507 | goto out; | |
1508 | ||
dd41f596 | 1509 | if (p->se.on_rq) |
1da177e4 LT |
1510 | goto out_running; |
1511 | ||
1512 | cpu = task_cpu(p); | |
cc367732 | 1513 | orig_cpu = cpu; |
1da177e4 LT |
1514 | this_cpu = smp_processor_id(); |
1515 | ||
1516 | #ifdef CONFIG_SMP | |
1517 | if (unlikely(task_running(rq, p))) | |
1518 | goto out_activate; | |
1519 | ||
7897986b NP |
1520 | new_cpu = cpu; |
1521 | ||
2d72376b | 1522 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1523 | if (cpu == this_cpu) { |
1524 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1525 | goto out_set_cpu; |
1526 | } | |
1527 | ||
1528 | for_each_domain(this_cpu, sd) { | |
1529 | if (cpu_isset(cpu, sd->span)) { | |
1530 | schedstat_inc(sd, ttwu_wake_remote); | |
1531 | this_sd = sd; | |
1532 | break; | |
1da177e4 LT |
1533 | } |
1534 | } | |
1da177e4 | 1535 | |
7897986b | 1536 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1537 | goto out_set_cpu; |
1538 | ||
1da177e4 | 1539 | /* |
7897986b | 1540 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1541 | */ |
7897986b NP |
1542 | if (this_sd) { |
1543 | int idx = this_sd->wake_idx; | |
1544 | unsigned int imbalance; | |
1da177e4 | 1545 | |
a3f21bce NP |
1546 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1547 | ||
7897986b NP |
1548 | load = source_load(cpu, idx); |
1549 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1550 | |
7897986b NP |
1551 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1552 | ||
a3f21bce NP |
1553 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1554 | unsigned long tl = this_load; | |
33859f7f MOS |
1555 | unsigned long tl_per_task; |
1556 | ||
71e20f18 IM |
1557 | /* |
1558 | * Attract cache-cold tasks on sync wakeups: | |
1559 | */ | |
1560 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1561 | goto out_set_cpu; | |
1562 | ||
cc367732 | 1563 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1564 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1565 | |
1da177e4 | 1566 | /* |
a3f21bce NP |
1567 | * If sync wakeup then subtract the (maximum possible) |
1568 | * effect of the currently running task from the load | |
1569 | * of the current CPU: | |
1da177e4 | 1570 | */ |
a3f21bce | 1571 | if (sync) |
dd41f596 | 1572 | tl -= current->se.load.weight; |
a3f21bce NP |
1573 | |
1574 | if ((tl <= load && | |
2dd73a4f | 1575 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1576 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1577 | /* |
1578 | * This domain has SD_WAKE_AFFINE and | |
1579 | * p is cache cold in this domain, and | |
1580 | * there is no bad imbalance. | |
1581 | */ | |
1582 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1583 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1584 | goto out_set_cpu; |
1585 | } | |
1586 | } | |
1587 | ||
1588 | /* | |
1589 | * Start passive balancing when half the imbalance_pct | |
1590 | * limit is reached. | |
1591 | */ | |
1592 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1593 | if (imbalance*this_load <= 100*load) { | |
1594 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1595 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1596 | goto out_set_cpu; |
1597 | } | |
1da177e4 LT |
1598 | } |
1599 | } | |
1600 | ||
1601 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1602 | out_set_cpu: | |
1603 | new_cpu = wake_idle(new_cpu, p); | |
1604 | if (new_cpu != cpu) { | |
1605 | set_task_cpu(p, new_cpu); | |
1606 | task_rq_unlock(rq, &flags); | |
1607 | /* might preempt at this point */ | |
1608 | rq = task_rq_lock(p, &flags); | |
1609 | old_state = p->state; | |
1610 | if (!(old_state & state)) | |
1611 | goto out; | |
dd41f596 | 1612 | if (p->se.on_rq) |
1da177e4 LT |
1613 | goto out_running; |
1614 | ||
1615 | this_cpu = smp_processor_id(); | |
1616 | cpu = task_cpu(p); | |
1617 | } | |
1618 | ||
1619 | out_activate: | |
1620 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1621 | schedstat_inc(p, se.nr_wakeups); |
1622 | if (sync) | |
1623 | schedstat_inc(p, se.nr_wakeups_sync); | |
1624 | if (orig_cpu != cpu) | |
1625 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1626 | if (cpu == this_cpu) | |
1627 | schedstat_inc(p, se.nr_wakeups_local); | |
1628 | else | |
1629 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1630 | update_rq_clock(rq); |
dd41f596 | 1631 | activate_task(rq, p, 1); |
9c63d9c0 | 1632 | check_preempt_curr(rq, p); |
1da177e4 LT |
1633 | success = 1; |
1634 | ||
1635 | out_running: | |
1636 | p->state = TASK_RUNNING; | |
1637 | out: | |
1638 | task_rq_unlock(rq, &flags); | |
1639 | ||
1640 | return success; | |
1641 | } | |
1642 | ||
36c8b586 | 1643 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1644 | { |
1645 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1646 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1647 | } | |
1da177e4 LT |
1648 | EXPORT_SYMBOL(wake_up_process); |
1649 | ||
36c8b586 | 1650 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1651 | { |
1652 | return try_to_wake_up(p, state, 0); | |
1653 | } | |
1654 | ||
1da177e4 LT |
1655 | /* |
1656 | * Perform scheduler related setup for a newly forked process p. | |
1657 | * p is forked by current. | |
dd41f596 IM |
1658 | * |
1659 | * __sched_fork() is basic setup used by init_idle() too: | |
1660 | */ | |
1661 | static void __sched_fork(struct task_struct *p) | |
1662 | { | |
dd41f596 IM |
1663 | p->se.exec_start = 0; |
1664 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1665 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1666 | |
1667 | #ifdef CONFIG_SCHEDSTATS | |
1668 | p->se.wait_start = 0; | |
dd41f596 IM |
1669 | p->se.sum_sleep_runtime = 0; |
1670 | p->se.sleep_start = 0; | |
dd41f596 IM |
1671 | p->se.block_start = 0; |
1672 | p->se.sleep_max = 0; | |
1673 | p->se.block_max = 0; | |
1674 | p->se.exec_max = 0; | |
eba1ed4b | 1675 | p->se.slice_max = 0; |
dd41f596 | 1676 | p->se.wait_max = 0; |
6cfb0d5d | 1677 | #endif |
476d139c | 1678 | |
dd41f596 IM |
1679 | INIT_LIST_HEAD(&p->run_list); |
1680 | p->se.on_rq = 0; | |
476d139c | 1681 | |
e107be36 AK |
1682 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1683 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1684 | #endif | |
1685 | ||
1da177e4 LT |
1686 | /* |
1687 | * We mark the process as running here, but have not actually | |
1688 | * inserted it onto the runqueue yet. This guarantees that | |
1689 | * nobody will actually run it, and a signal or other external | |
1690 | * event cannot wake it up and insert it on the runqueue either. | |
1691 | */ | |
1692 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1693 | } |
1694 | ||
1695 | /* | |
1696 | * fork()/clone()-time setup: | |
1697 | */ | |
1698 | void sched_fork(struct task_struct *p, int clone_flags) | |
1699 | { | |
1700 | int cpu = get_cpu(); | |
1701 | ||
1702 | __sched_fork(p); | |
1703 | ||
1704 | #ifdef CONFIG_SMP | |
1705 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1706 | #endif | |
02e4bac2 | 1707 | set_task_cpu(p, cpu); |
b29739f9 IM |
1708 | |
1709 | /* | |
1710 | * Make sure we do not leak PI boosting priority to the child: | |
1711 | */ | |
1712 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1713 | if (!rt_prio(p->prio)) |
1714 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1715 | |
52f17b6c | 1716 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1717 | if (likely(sched_info_on())) |
52f17b6c | 1718 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1719 | #endif |
d6077cb8 | 1720 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1721 | p->oncpu = 0; |
1722 | #endif | |
1da177e4 | 1723 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1724 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1725 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1726 | #endif |
476d139c | 1727 | put_cpu(); |
1da177e4 LT |
1728 | } |
1729 | ||
1730 | /* | |
1731 | * wake_up_new_task - wake up a newly created task for the first time. | |
1732 | * | |
1733 | * This function will do some initial scheduler statistics housekeeping | |
1734 | * that must be done for every newly created context, then puts the task | |
1735 | * on the runqueue and wakes it. | |
1736 | */ | |
36c8b586 | 1737 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1738 | { |
1739 | unsigned long flags; | |
dd41f596 | 1740 | struct rq *rq; |
1da177e4 LT |
1741 | |
1742 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1743 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1744 | update_rq_clock(rq); |
1da177e4 LT |
1745 | |
1746 | p->prio = effective_prio(p); | |
1747 | ||
b9dca1e0 | 1748 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1749 | activate_task(rq, p, 0); |
1da177e4 | 1750 | } else { |
1da177e4 | 1751 | /* |
dd41f596 IM |
1752 | * Let the scheduling class do new task startup |
1753 | * management (if any): | |
1da177e4 | 1754 | */ |
ee0827d8 | 1755 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1756 | inc_nr_running(p, rq); |
1da177e4 | 1757 | } |
dd41f596 IM |
1758 | check_preempt_curr(rq, p); |
1759 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1760 | } |
1761 | ||
e107be36 AK |
1762 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1763 | ||
1764 | /** | |
421cee29 RD |
1765 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1766 | * @notifier: notifier struct to register | |
e107be36 AK |
1767 | */ |
1768 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1769 | { | |
1770 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1771 | } | |
1772 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1773 | ||
1774 | /** | |
1775 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1776 | * @notifier: notifier struct to unregister |
e107be36 AK |
1777 | * |
1778 | * This is safe to call from within a preemption notifier. | |
1779 | */ | |
1780 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1781 | { | |
1782 | hlist_del(¬ifier->link); | |
1783 | } | |
1784 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1785 | ||
1786 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1787 | { | |
1788 | struct preempt_notifier *notifier; | |
1789 | struct hlist_node *node; | |
1790 | ||
1791 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1792 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1793 | } | |
1794 | ||
1795 | static void | |
1796 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1797 | struct task_struct *next) | |
1798 | { | |
1799 | struct preempt_notifier *notifier; | |
1800 | struct hlist_node *node; | |
1801 | ||
1802 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1803 | notifier->ops->sched_out(notifier, next); | |
1804 | } | |
1805 | ||
1806 | #else | |
1807 | ||
1808 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1809 | { | |
1810 | } | |
1811 | ||
1812 | static void | |
1813 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1814 | struct task_struct *next) | |
1815 | { | |
1816 | } | |
1817 | ||
1818 | #endif | |
1819 | ||
4866cde0 NP |
1820 | /** |
1821 | * prepare_task_switch - prepare to switch tasks | |
1822 | * @rq: the runqueue preparing to switch | |
421cee29 | 1823 | * @prev: the current task that is being switched out |
4866cde0 NP |
1824 | * @next: the task we are going to switch to. |
1825 | * | |
1826 | * This is called with the rq lock held and interrupts off. It must | |
1827 | * be paired with a subsequent finish_task_switch after the context | |
1828 | * switch. | |
1829 | * | |
1830 | * prepare_task_switch sets up locking and calls architecture specific | |
1831 | * hooks. | |
1832 | */ | |
e107be36 AK |
1833 | static inline void |
1834 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1835 | struct task_struct *next) | |
4866cde0 | 1836 | { |
e107be36 | 1837 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1838 | prepare_lock_switch(rq, next); |
1839 | prepare_arch_switch(next); | |
1840 | } | |
1841 | ||
1da177e4 LT |
1842 | /** |
1843 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1844 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1845 | * @prev: the thread we just switched away from. |
1846 | * | |
4866cde0 NP |
1847 | * finish_task_switch must be called after the context switch, paired |
1848 | * with a prepare_task_switch call before the context switch. | |
1849 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1850 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1851 | * |
1852 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1853 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1854 | * with the lock held can cause deadlocks; see schedule() for |
1855 | * details.) | |
1856 | */ | |
a9957449 | 1857 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1858 | __releases(rq->lock) |
1859 | { | |
1da177e4 | 1860 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1861 | long prev_state; |
1da177e4 LT |
1862 | |
1863 | rq->prev_mm = NULL; | |
1864 | ||
1865 | /* | |
1866 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1867 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1868 | * schedule one last time. The schedule call will never return, and |
1869 | * the scheduled task must drop that reference. | |
c394cc9f | 1870 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1871 | * still held, otherwise prev could be scheduled on another cpu, die |
1872 | * there before we look at prev->state, and then the reference would | |
1873 | * be dropped twice. | |
1874 | * Manfred Spraul <manfred@colorfullife.com> | |
1875 | */ | |
55a101f8 | 1876 | prev_state = prev->state; |
4866cde0 NP |
1877 | finish_arch_switch(prev); |
1878 | finish_lock_switch(rq, prev); | |
e107be36 | 1879 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1880 | if (mm) |
1881 | mmdrop(mm); | |
c394cc9f | 1882 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1883 | /* |
1884 | * Remove function-return probe instances associated with this | |
1885 | * task and put them back on the free list. | |
9761eea8 | 1886 | */ |
c6fd91f0 | 1887 | kprobe_flush_task(prev); |
1da177e4 | 1888 | put_task_struct(prev); |
c6fd91f0 | 1889 | } |
1da177e4 LT |
1890 | } |
1891 | ||
1892 | /** | |
1893 | * schedule_tail - first thing a freshly forked thread must call. | |
1894 | * @prev: the thread we just switched away from. | |
1895 | */ | |
36c8b586 | 1896 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1897 | __releases(rq->lock) |
1898 | { | |
70b97a7f IM |
1899 | struct rq *rq = this_rq(); |
1900 | ||
4866cde0 NP |
1901 | finish_task_switch(rq, prev); |
1902 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1903 | /* In this case, finish_task_switch does not reenable preemption */ | |
1904 | preempt_enable(); | |
1905 | #endif | |
1da177e4 | 1906 | if (current->set_child_tid) |
b488893a | 1907 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1908 | } |
1909 | ||
1910 | /* | |
1911 | * context_switch - switch to the new MM and the new | |
1912 | * thread's register state. | |
1913 | */ | |
dd41f596 | 1914 | static inline void |
70b97a7f | 1915 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1916 | struct task_struct *next) |
1da177e4 | 1917 | { |
dd41f596 | 1918 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1919 | |
e107be36 | 1920 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1921 | mm = next->mm; |
1922 | oldmm = prev->active_mm; | |
9226d125 ZA |
1923 | /* |
1924 | * For paravirt, this is coupled with an exit in switch_to to | |
1925 | * combine the page table reload and the switch backend into | |
1926 | * one hypercall. | |
1927 | */ | |
1928 | arch_enter_lazy_cpu_mode(); | |
1929 | ||
dd41f596 | 1930 | if (unlikely(!mm)) { |
1da177e4 LT |
1931 | next->active_mm = oldmm; |
1932 | atomic_inc(&oldmm->mm_count); | |
1933 | enter_lazy_tlb(oldmm, next); | |
1934 | } else | |
1935 | switch_mm(oldmm, mm, next); | |
1936 | ||
dd41f596 | 1937 | if (unlikely(!prev->mm)) { |
1da177e4 | 1938 | prev->active_mm = NULL; |
1da177e4 LT |
1939 | rq->prev_mm = oldmm; |
1940 | } | |
3a5f5e48 IM |
1941 | /* |
1942 | * Since the runqueue lock will be released by the next | |
1943 | * task (which is an invalid locking op but in the case | |
1944 | * of the scheduler it's an obvious special-case), so we | |
1945 | * do an early lockdep release here: | |
1946 | */ | |
1947 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 1948 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 1949 | #endif |
1da177e4 LT |
1950 | |
1951 | /* Here we just switch the register state and the stack. */ | |
1952 | switch_to(prev, next, prev); | |
1953 | ||
dd41f596 IM |
1954 | barrier(); |
1955 | /* | |
1956 | * this_rq must be evaluated again because prev may have moved | |
1957 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
1958 | * frame will be invalid. | |
1959 | */ | |
1960 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
1961 | } |
1962 | ||
1963 | /* | |
1964 | * nr_running, nr_uninterruptible and nr_context_switches: | |
1965 | * | |
1966 | * externally visible scheduler statistics: current number of runnable | |
1967 | * threads, current number of uninterruptible-sleeping threads, total | |
1968 | * number of context switches performed since bootup. | |
1969 | */ | |
1970 | unsigned long nr_running(void) | |
1971 | { | |
1972 | unsigned long i, sum = 0; | |
1973 | ||
1974 | for_each_online_cpu(i) | |
1975 | sum += cpu_rq(i)->nr_running; | |
1976 | ||
1977 | return sum; | |
1978 | } | |
1979 | ||
1980 | unsigned long nr_uninterruptible(void) | |
1981 | { | |
1982 | unsigned long i, sum = 0; | |
1983 | ||
0a945022 | 1984 | for_each_possible_cpu(i) |
1da177e4 LT |
1985 | sum += cpu_rq(i)->nr_uninterruptible; |
1986 | ||
1987 | /* | |
1988 | * Since we read the counters lockless, it might be slightly | |
1989 | * inaccurate. Do not allow it to go below zero though: | |
1990 | */ | |
1991 | if (unlikely((long)sum < 0)) | |
1992 | sum = 0; | |
1993 | ||
1994 | return sum; | |
1995 | } | |
1996 | ||
1997 | unsigned long long nr_context_switches(void) | |
1998 | { | |
cc94abfc SR |
1999 | int i; |
2000 | unsigned long long sum = 0; | |
1da177e4 | 2001 | |
0a945022 | 2002 | for_each_possible_cpu(i) |
1da177e4 LT |
2003 | sum += cpu_rq(i)->nr_switches; |
2004 | ||
2005 | return sum; | |
2006 | } | |
2007 | ||
2008 | unsigned long nr_iowait(void) | |
2009 | { | |
2010 | unsigned long i, sum = 0; | |
2011 | ||
0a945022 | 2012 | for_each_possible_cpu(i) |
1da177e4 LT |
2013 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2014 | ||
2015 | return sum; | |
2016 | } | |
2017 | ||
db1b1fef JS |
2018 | unsigned long nr_active(void) |
2019 | { | |
2020 | unsigned long i, running = 0, uninterruptible = 0; | |
2021 | ||
2022 | for_each_online_cpu(i) { | |
2023 | running += cpu_rq(i)->nr_running; | |
2024 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2025 | } | |
2026 | ||
2027 | if (unlikely((long)uninterruptible < 0)) | |
2028 | uninterruptible = 0; | |
2029 | ||
2030 | return running + uninterruptible; | |
2031 | } | |
2032 | ||
48f24c4d | 2033 | /* |
dd41f596 IM |
2034 | * Update rq->cpu_load[] statistics. This function is usually called every |
2035 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2036 | */ |
dd41f596 | 2037 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2038 | { |
495eca49 | 2039 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2040 | int i, scale; |
2041 | ||
2042 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2043 | |
2044 | /* Update our load: */ | |
2045 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2046 | unsigned long old_load, new_load; | |
2047 | ||
2048 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2049 | ||
2050 | old_load = this_rq->cpu_load[i]; | |
2051 | new_load = this_load; | |
a25707f3 IM |
2052 | /* |
2053 | * Round up the averaging division if load is increasing. This | |
2054 | * prevents us from getting stuck on 9 if the load is 10, for | |
2055 | * example. | |
2056 | */ | |
2057 | if (new_load > old_load) | |
2058 | new_load += scale-1; | |
dd41f596 IM |
2059 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2060 | } | |
48f24c4d IM |
2061 | } |
2062 | ||
dd41f596 IM |
2063 | #ifdef CONFIG_SMP |
2064 | ||
1da177e4 LT |
2065 | /* |
2066 | * double_rq_lock - safely lock two runqueues | |
2067 | * | |
2068 | * Note this does not disable interrupts like task_rq_lock, | |
2069 | * you need to do so manually before calling. | |
2070 | */ | |
70b97a7f | 2071 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2072 | __acquires(rq1->lock) |
2073 | __acquires(rq2->lock) | |
2074 | { | |
054b9108 | 2075 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2076 | if (rq1 == rq2) { |
2077 | spin_lock(&rq1->lock); | |
2078 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2079 | } else { | |
c96d145e | 2080 | if (rq1 < rq2) { |
1da177e4 LT |
2081 | spin_lock(&rq1->lock); |
2082 | spin_lock(&rq2->lock); | |
2083 | } else { | |
2084 | spin_lock(&rq2->lock); | |
2085 | spin_lock(&rq1->lock); | |
2086 | } | |
2087 | } | |
6e82a3be IM |
2088 | update_rq_clock(rq1); |
2089 | update_rq_clock(rq2); | |
1da177e4 LT |
2090 | } |
2091 | ||
2092 | /* | |
2093 | * double_rq_unlock - safely unlock two runqueues | |
2094 | * | |
2095 | * Note this does not restore interrupts like task_rq_unlock, | |
2096 | * you need to do so manually after calling. | |
2097 | */ | |
70b97a7f | 2098 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2099 | __releases(rq1->lock) |
2100 | __releases(rq2->lock) | |
2101 | { | |
2102 | spin_unlock(&rq1->lock); | |
2103 | if (rq1 != rq2) | |
2104 | spin_unlock(&rq2->lock); | |
2105 | else | |
2106 | __release(rq2->lock); | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2111 | */ | |
70b97a7f | 2112 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2113 | __releases(this_rq->lock) |
2114 | __acquires(busiest->lock) | |
2115 | __acquires(this_rq->lock) | |
2116 | { | |
054b9108 KK |
2117 | if (unlikely(!irqs_disabled())) { |
2118 | /* printk() doesn't work good under rq->lock */ | |
2119 | spin_unlock(&this_rq->lock); | |
2120 | BUG_ON(1); | |
2121 | } | |
1da177e4 | 2122 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2123 | if (busiest < this_rq) { |
1da177e4 LT |
2124 | spin_unlock(&this_rq->lock); |
2125 | spin_lock(&busiest->lock); | |
2126 | spin_lock(&this_rq->lock); | |
2127 | } else | |
2128 | spin_lock(&busiest->lock); | |
2129 | } | |
2130 | } | |
2131 | ||
1da177e4 LT |
2132 | /* |
2133 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2134 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2135 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2136 | * the cpu_allowed mask is restored. |
2137 | */ | |
36c8b586 | 2138 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2139 | { |
70b97a7f | 2140 | struct migration_req req; |
1da177e4 | 2141 | unsigned long flags; |
70b97a7f | 2142 | struct rq *rq; |
1da177e4 LT |
2143 | |
2144 | rq = task_rq_lock(p, &flags); | |
2145 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2146 | || unlikely(cpu_is_offline(dest_cpu))) | |
2147 | goto out; | |
2148 | ||
2149 | /* force the process onto the specified CPU */ | |
2150 | if (migrate_task(p, dest_cpu, &req)) { | |
2151 | /* Need to wait for migration thread (might exit: take ref). */ | |
2152 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2153 | |
1da177e4 LT |
2154 | get_task_struct(mt); |
2155 | task_rq_unlock(rq, &flags); | |
2156 | wake_up_process(mt); | |
2157 | put_task_struct(mt); | |
2158 | wait_for_completion(&req.done); | |
36c8b586 | 2159 | |
1da177e4 LT |
2160 | return; |
2161 | } | |
2162 | out: | |
2163 | task_rq_unlock(rq, &flags); | |
2164 | } | |
2165 | ||
2166 | /* | |
476d139c NP |
2167 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2168 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2169 | */ |
2170 | void sched_exec(void) | |
2171 | { | |
1da177e4 | 2172 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2173 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2174 | put_cpu(); |
476d139c NP |
2175 | if (new_cpu != this_cpu) |
2176 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2177 | } |
2178 | ||
2179 | /* | |
2180 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2181 | * Both runqueues must be locked. | |
2182 | */ | |
dd41f596 IM |
2183 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2184 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2185 | { |
2e1cb74a | 2186 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2187 | set_task_cpu(p, this_cpu); |
dd41f596 | 2188 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2189 | /* |
2190 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2191 | * to be always true for them. | |
2192 | */ | |
dd41f596 | 2193 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2194 | } |
2195 | ||
2196 | /* | |
2197 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2198 | */ | |
858119e1 | 2199 | static |
70b97a7f | 2200 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2201 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2202 | int *all_pinned) |
1da177e4 LT |
2203 | { |
2204 | /* | |
2205 | * We do not migrate tasks that are: | |
2206 | * 1) running (obviously), or | |
2207 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2208 | * 3) are cache-hot on their current CPU. | |
2209 | */ | |
cc367732 IM |
2210 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2211 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2212 | return 0; |
cc367732 | 2213 | } |
81026794 NP |
2214 | *all_pinned = 0; |
2215 | ||
cc367732 IM |
2216 | if (task_running(rq, p)) { |
2217 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2218 | return 0; |
cc367732 | 2219 | } |
1da177e4 | 2220 | |
da84d961 IM |
2221 | /* |
2222 | * Aggressive migration if: | |
2223 | * 1) task is cache cold, or | |
2224 | * 2) too many balance attempts have failed. | |
2225 | */ | |
2226 | ||
6bc1665b IM |
2227 | if (!task_hot(p, rq->clock, sd) || |
2228 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2229 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2230 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2231 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2232 | schedstat_inc(p, se.nr_forced_migrations); |
2233 | } | |
da84d961 IM |
2234 | #endif |
2235 | return 1; | |
2236 | } | |
2237 | ||
cc367732 IM |
2238 | if (task_hot(p, rq->clock, sd)) { |
2239 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2240 | return 0; |
cc367732 | 2241 | } |
1da177e4 LT |
2242 | return 1; |
2243 | } | |
2244 | ||
e1d1484f PW |
2245 | static unsigned long |
2246 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2247 | unsigned long max_load_move, struct sched_domain *sd, | |
2248 | enum cpu_idle_type idle, int *all_pinned, | |
2249 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2250 | { |
b82d9fdd | 2251 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2252 | struct task_struct *p; |
2253 | long rem_load_move = max_load_move; | |
1da177e4 | 2254 | |
e1d1484f | 2255 | if (max_load_move == 0) |
1da177e4 LT |
2256 | goto out; |
2257 | ||
81026794 NP |
2258 | pinned = 1; |
2259 | ||
1da177e4 | 2260 | /* |
dd41f596 | 2261 | * Start the load-balancing iterator: |
1da177e4 | 2262 | */ |
dd41f596 IM |
2263 | p = iterator->start(iterator->arg); |
2264 | next: | |
b82d9fdd | 2265 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2266 | goto out; |
50ddd969 | 2267 | /* |
b82d9fdd | 2268 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2269 | * skip a task if it will be the highest priority task (i.e. smallest |
2270 | * prio value) on its new queue regardless of its load weight | |
2271 | */ | |
dd41f596 IM |
2272 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2273 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2274 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2275 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2276 | p = iterator->next(iterator->arg); |
2277 | goto next; | |
1da177e4 LT |
2278 | } |
2279 | ||
dd41f596 | 2280 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2281 | pulled++; |
dd41f596 | 2282 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2283 | |
2dd73a4f | 2284 | /* |
b82d9fdd | 2285 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2286 | */ |
e1d1484f | 2287 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2288 | if (p->prio < *this_best_prio) |
2289 | *this_best_prio = p->prio; | |
dd41f596 IM |
2290 | p = iterator->next(iterator->arg); |
2291 | goto next; | |
1da177e4 LT |
2292 | } |
2293 | out: | |
2294 | /* | |
e1d1484f | 2295 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2296 | * so we can safely collect pull_task() stats here rather than |
2297 | * inside pull_task(). | |
2298 | */ | |
2299 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2300 | |
2301 | if (all_pinned) | |
2302 | *all_pinned = pinned; | |
e1d1484f PW |
2303 | |
2304 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2305 | } |
2306 | ||
dd41f596 | 2307 | /* |
43010659 PW |
2308 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2309 | * this_rq, as part of a balancing operation within domain "sd". | |
2310 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2311 | * |
2312 | * Called with both runqueues locked. | |
2313 | */ | |
2314 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2315 | unsigned long max_load_move, |
dd41f596 IM |
2316 | struct sched_domain *sd, enum cpu_idle_type idle, |
2317 | int *all_pinned) | |
2318 | { | |
5522d5d5 | 2319 | const struct sched_class *class = sched_class_highest; |
43010659 | 2320 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2321 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2322 | |
2323 | do { | |
43010659 PW |
2324 | total_load_moved += |
2325 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2326 | max_load_move - total_load_moved, |
a4ac01c3 | 2327 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2328 | class = class->next; |
43010659 | 2329 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2330 | |
43010659 PW |
2331 | return total_load_moved > 0; |
2332 | } | |
2333 | ||
e1d1484f PW |
2334 | static int |
2335 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2336 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2337 | struct rq_iterator *iterator) | |
2338 | { | |
2339 | struct task_struct *p = iterator->start(iterator->arg); | |
2340 | int pinned = 0; | |
2341 | ||
2342 | while (p) { | |
2343 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2344 | pull_task(busiest, p, this_rq, this_cpu); | |
2345 | /* | |
2346 | * Right now, this is only the second place pull_task() | |
2347 | * is called, so we can safely collect pull_task() | |
2348 | * stats here rather than inside pull_task(). | |
2349 | */ | |
2350 | schedstat_inc(sd, lb_gained[idle]); | |
2351 | ||
2352 | return 1; | |
2353 | } | |
2354 | p = iterator->next(iterator->arg); | |
2355 | } | |
2356 | ||
2357 | return 0; | |
2358 | } | |
2359 | ||
43010659 PW |
2360 | /* |
2361 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2362 | * part of active balancing operations within "domain". | |
2363 | * Returns 1 if successful and 0 otherwise. | |
2364 | * | |
2365 | * Called with both runqueues locked. | |
2366 | */ | |
2367 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2368 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2369 | { | |
5522d5d5 | 2370 | const struct sched_class *class; |
43010659 PW |
2371 | |
2372 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2373 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2374 | return 1; |
2375 | ||
2376 | return 0; | |
dd41f596 IM |
2377 | } |
2378 | ||
1da177e4 LT |
2379 | /* |
2380 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2381 | * domain. It calculates and returns the amount of weighted load which |
2382 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2383 | */ |
2384 | static struct sched_group * | |
2385 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2386 | unsigned long *imbalance, enum cpu_idle_type idle, |
2387 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2388 | { |
2389 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2390 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2391 | unsigned long max_pull; |
2dd73a4f PW |
2392 | unsigned long busiest_load_per_task, busiest_nr_running; |
2393 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2394 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2395 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2396 | int power_savings_balance = 1; | |
2397 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2398 | unsigned long min_nr_running = ULONG_MAX; | |
2399 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2400 | #endif | |
1da177e4 LT |
2401 | |
2402 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2403 | busiest_load_per_task = busiest_nr_running = 0; |
2404 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2405 | if (idle == CPU_NOT_IDLE) |
7897986b | 2406 | load_idx = sd->busy_idx; |
d15bcfdb | 2407 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2408 | load_idx = sd->newidle_idx; |
2409 | else | |
2410 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2411 | |
2412 | do { | |
908a7c1b | 2413 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2414 | int local_group; |
2415 | int i; | |
908a7c1b | 2416 | int __group_imb = 0; |
783609c6 | 2417 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2418 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2419 | |
2420 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2421 | ||
783609c6 SS |
2422 | if (local_group) |
2423 | balance_cpu = first_cpu(group->cpumask); | |
2424 | ||
1da177e4 | 2425 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2426 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2427 | max_cpu_load = 0; |
2428 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2429 | |
2430 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2431 | struct rq *rq; |
2432 | ||
2433 | if (!cpu_isset(i, *cpus)) | |
2434 | continue; | |
2435 | ||
2436 | rq = cpu_rq(i); | |
2dd73a4f | 2437 | |
9439aab8 | 2438 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2439 | *sd_idle = 0; |
2440 | ||
1da177e4 | 2441 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2442 | if (local_group) { |
2443 | if (idle_cpu(i) && !first_idle_cpu) { | |
2444 | first_idle_cpu = 1; | |
2445 | balance_cpu = i; | |
2446 | } | |
2447 | ||
a2000572 | 2448 | load = target_load(i, load_idx); |
908a7c1b | 2449 | } else { |
a2000572 | 2450 | load = source_load(i, load_idx); |
908a7c1b KC |
2451 | if (load > max_cpu_load) |
2452 | max_cpu_load = load; | |
2453 | if (min_cpu_load > load) | |
2454 | min_cpu_load = load; | |
2455 | } | |
1da177e4 LT |
2456 | |
2457 | avg_load += load; | |
2dd73a4f | 2458 | sum_nr_running += rq->nr_running; |
dd41f596 | 2459 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2460 | } |
2461 | ||
783609c6 SS |
2462 | /* |
2463 | * First idle cpu or the first cpu(busiest) in this sched group | |
2464 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2465 | * domains. In the newly idle case, we will allow all the cpu's |
2466 | * to do the newly idle load balance. | |
783609c6 | 2467 | */ |
9439aab8 SS |
2468 | if (idle != CPU_NEWLY_IDLE && local_group && |
2469 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2470 | *balance = 0; |
2471 | goto ret; | |
2472 | } | |
2473 | ||
1da177e4 | 2474 | total_load += avg_load; |
5517d86b | 2475 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2476 | |
2477 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2478 | avg_load = sg_div_cpu_power(group, |
2479 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2480 | |
908a7c1b KC |
2481 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2482 | __group_imb = 1; | |
2483 | ||
5517d86b | 2484 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2485 | |
1da177e4 LT |
2486 | if (local_group) { |
2487 | this_load = avg_load; | |
2488 | this = group; | |
2dd73a4f PW |
2489 | this_nr_running = sum_nr_running; |
2490 | this_load_per_task = sum_weighted_load; | |
2491 | } else if (avg_load > max_load && | |
908a7c1b | 2492 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2493 | max_load = avg_load; |
2494 | busiest = group; | |
2dd73a4f PW |
2495 | busiest_nr_running = sum_nr_running; |
2496 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2497 | group_imb = __group_imb; |
1da177e4 | 2498 | } |
5c45bf27 SS |
2499 | |
2500 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2501 | /* | |
2502 | * Busy processors will not participate in power savings | |
2503 | * balance. | |
2504 | */ | |
dd41f596 IM |
2505 | if (idle == CPU_NOT_IDLE || |
2506 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2507 | goto group_next; | |
5c45bf27 SS |
2508 | |
2509 | /* | |
2510 | * If the local group is idle or completely loaded | |
2511 | * no need to do power savings balance at this domain | |
2512 | */ | |
2513 | if (local_group && (this_nr_running >= group_capacity || | |
2514 | !this_nr_running)) | |
2515 | power_savings_balance = 0; | |
2516 | ||
dd41f596 | 2517 | /* |
5c45bf27 SS |
2518 | * If a group is already running at full capacity or idle, |
2519 | * don't include that group in power savings calculations | |
dd41f596 IM |
2520 | */ |
2521 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2522 | || !sum_nr_running) |
dd41f596 | 2523 | goto group_next; |
5c45bf27 | 2524 | |
dd41f596 | 2525 | /* |
5c45bf27 | 2526 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2527 | * This is the group from where we need to pick up the load |
2528 | * for saving power | |
2529 | */ | |
2530 | if ((sum_nr_running < min_nr_running) || | |
2531 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2532 | first_cpu(group->cpumask) < |
2533 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2534 | group_min = group; |
2535 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2536 | min_load_per_task = sum_weighted_load / |
2537 | sum_nr_running; | |
dd41f596 | 2538 | } |
5c45bf27 | 2539 | |
dd41f596 | 2540 | /* |
5c45bf27 | 2541 | * Calculate the group which is almost near its |
dd41f596 IM |
2542 | * capacity but still has some space to pick up some load |
2543 | * from other group and save more power | |
2544 | */ | |
2545 | if (sum_nr_running <= group_capacity - 1) { | |
2546 | if (sum_nr_running > leader_nr_running || | |
2547 | (sum_nr_running == leader_nr_running && | |
2548 | first_cpu(group->cpumask) > | |
2549 | first_cpu(group_leader->cpumask))) { | |
2550 | group_leader = group; | |
2551 | leader_nr_running = sum_nr_running; | |
2552 | } | |
48f24c4d | 2553 | } |
5c45bf27 SS |
2554 | group_next: |
2555 | #endif | |
1da177e4 LT |
2556 | group = group->next; |
2557 | } while (group != sd->groups); | |
2558 | ||
2dd73a4f | 2559 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2560 | goto out_balanced; |
2561 | ||
2562 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2563 | ||
2564 | if (this_load >= avg_load || | |
2565 | 100*max_load <= sd->imbalance_pct*this_load) | |
2566 | goto out_balanced; | |
2567 | ||
2dd73a4f | 2568 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2569 | if (group_imb) |
2570 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2571 | ||
1da177e4 LT |
2572 | /* |
2573 | * We're trying to get all the cpus to the average_load, so we don't | |
2574 | * want to push ourselves above the average load, nor do we wish to | |
2575 | * reduce the max loaded cpu below the average load, as either of these | |
2576 | * actions would just result in more rebalancing later, and ping-pong | |
2577 | * tasks around. Thus we look for the minimum possible imbalance. | |
2578 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2579 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2580 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2581 | * appear as very large values with unsigned longs. |
2582 | */ | |
2dd73a4f PW |
2583 | if (max_load <= busiest_load_per_task) |
2584 | goto out_balanced; | |
2585 | ||
2586 | /* | |
2587 | * In the presence of smp nice balancing, certain scenarios can have | |
2588 | * max load less than avg load(as we skip the groups at or below | |
2589 | * its cpu_power, while calculating max_load..) | |
2590 | */ | |
2591 | if (max_load < avg_load) { | |
2592 | *imbalance = 0; | |
2593 | goto small_imbalance; | |
2594 | } | |
0c117f1b SS |
2595 | |
2596 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2597 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2598 | |
1da177e4 | 2599 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2600 | *imbalance = min(max_pull * busiest->__cpu_power, |
2601 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2602 | / SCHED_LOAD_SCALE; |
2603 | ||
2dd73a4f PW |
2604 | /* |
2605 | * if *imbalance is less than the average load per runnable task | |
2606 | * there is no gaurantee that any tasks will be moved so we'll have | |
2607 | * a think about bumping its value to force at least one task to be | |
2608 | * moved | |
2609 | */ | |
7fd0d2dd | 2610 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2611 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2612 | unsigned int imbn; |
2613 | ||
2614 | small_imbalance: | |
2615 | pwr_move = pwr_now = 0; | |
2616 | imbn = 2; | |
2617 | if (this_nr_running) { | |
2618 | this_load_per_task /= this_nr_running; | |
2619 | if (busiest_load_per_task > this_load_per_task) | |
2620 | imbn = 1; | |
2621 | } else | |
2622 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2623 | |
dd41f596 IM |
2624 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2625 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2626 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2627 | return busiest; |
2628 | } | |
2629 | ||
2630 | /* | |
2631 | * OK, we don't have enough imbalance to justify moving tasks, | |
2632 | * however we may be able to increase total CPU power used by | |
2633 | * moving them. | |
2634 | */ | |
2635 | ||
5517d86b ED |
2636 | pwr_now += busiest->__cpu_power * |
2637 | min(busiest_load_per_task, max_load); | |
2638 | pwr_now += this->__cpu_power * | |
2639 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2640 | pwr_now /= SCHED_LOAD_SCALE; |
2641 | ||
2642 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2643 | tmp = sg_div_cpu_power(busiest, |
2644 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2645 | if (max_load > tmp) |
5517d86b | 2646 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2647 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2648 | |
2649 | /* Amount of load we'd add */ | |
5517d86b | 2650 | if (max_load * busiest->__cpu_power < |
33859f7f | 2651 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2652 | tmp = sg_div_cpu_power(this, |
2653 | max_load * busiest->__cpu_power); | |
1da177e4 | 2654 | else |
5517d86b ED |
2655 | tmp = sg_div_cpu_power(this, |
2656 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2657 | pwr_move += this->__cpu_power * | |
2658 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2659 | pwr_move /= SCHED_LOAD_SCALE; |
2660 | ||
2661 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2662 | if (pwr_move > pwr_now) |
2663 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2664 | } |
2665 | ||
1da177e4 LT |
2666 | return busiest; |
2667 | ||
2668 | out_balanced: | |
5c45bf27 | 2669 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2670 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2671 | goto ret; |
1da177e4 | 2672 | |
5c45bf27 SS |
2673 | if (this == group_leader && group_leader != group_min) { |
2674 | *imbalance = min_load_per_task; | |
2675 | return group_min; | |
2676 | } | |
5c45bf27 | 2677 | #endif |
783609c6 | 2678 | ret: |
1da177e4 LT |
2679 | *imbalance = 0; |
2680 | return NULL; | |
2681 | } | |
2682 | ||
2683 | /* | |
2684 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2685 | */ | |
70b97a7f | 2686 | static struct rq * |
d15bcfdb | 2687 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2688 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2689 | { |
70b97a7f | 2690 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2691 | unsigned long max_load = 0; |
1da177e4 LT |
2692 | int i; |
2693 | ||
2694 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2695 | unsigned long wl; |
0a2966b4 CL |
2696 | |
2697 | if (!cpu_isset(i, *cpus)) | |
2698 | continue; | |
2699 | ||
48f24c4d | 2700 | rq = cpu_rq(i); |
dd41f596 | 2701 | wl = weighted_cpuload(i); |
2dd73a4f | 2702 | |
dd41f596 | 2703 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2704 | continue; |
1da177e4 | 2705 | |
dd41f596 IM |
2706 | if (wl > max_load) { |
2707 | max_load = wl; | |
48f24c4d | 2708 | busiest = rq; |
1da177e4 LT |
2709 | } |
2710 | } | |
2711 | ||
2712 | return busiest; | |
2713 | } | |
2714 | ||
77391d71 NP |
2715 | /* |
2716 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2717 | * so long as it is large enough. | |
2718 | */ | |
2719 | #define MAX_PINNED_INTERVAL 512 | |
2720 | ||
1da177e4 LT |
2721 | /* |
2722 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2723 | * tasks if there is an imbalance. | |
1da177e4 | 2724 | */ |
70b97a7f | 2725 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2726 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2727 | int *balance) |
1da177e4 | 2728 | { |
43010659 | 2729 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2730 | struct sched_group *group; |
1da177e4 | 2731 | unsigned long imbalance; |
70b97a7f | 2732 | struct rq *busiest; |
0a2966b4 | 2733 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2734 | unsigned long flags; |
5969fe06 | 2735 | |
89c4710e SS |
2736 | /* |
2737 | * When power savings policy is enabled for the parent domain, idle | |
2738 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2739 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2740 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2741 | */ |
d15bcfdb | 2742 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2743 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2744 | sd_idle = 1; |
1da177e4 | 2745 | |
2d72376b | 2746 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2747 | |
0a2966b4 CL |
2748 | redo: |
2749 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2750 | &cpus, balance); |
2751 | ||
06066714 | 2752 | if (*balance == 0) |
783609c6 | 2753 | goto out_balanced; |
783609c6 | 2754 | |
1da177e4 LT |
2755 | if (!group) { |
2756 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2757 | goto out_balanced; | |
2758 | } | |
2759 | ||
0a2966b4 | 2760 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2761 | if (!busiest) { |
2762 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2763 | goto out_balanced; | |
2764 | } | |
2765 | ||
db935dbd | 2766 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2767 | |
2768 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2769 | ||
43010659 | 2770 | ld_moved = 0; |
1da177e4 LT |
2771 | if (busiest->nr_running > 1) { |
2772 | /* | |
2773 | * Attempt to move tasks. If find_busiest_group has found | |
2774 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2775 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2776 | * correctly treated as an imbalance. |
2777 | */ | |
fe2eea3f | 2778 | local_irq_save(flags); |
e17224bf | 2779 | double_rq_lock(this_rq, busiest); |
43010659 | 2780 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2781 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2782 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2783 | local_irq_restore(flags); |
81026794 | 2784 | |
46cb4b7c SS |
2785 | /* |
2786 | * some other cpu did the load balance for us. | |
2787 | */ | |
43010659 | 2788 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2789 | resched_cpu(this_cpu); |
2790 | ||
81026794 | 2791 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2792 | if (unlikely(all_pinned)) { |
2793 | cpu_clear(cpu_of(busiest), cpus); | |
2794 | if (!cpus_empty(cpus)) | |
2795 | goto redo; | |
81026794 | 2796 | goto out_balanced; |
0a2966b4 | 2797 | } |
1da177e4 | 2798 | } |
81026794 | 2799 | |
43010659 | 2800 | if (!ld_moved) { |
1da177e4 LT |
2801 | schedstat_inc(sd, lb_failed[idle]); |
2802 | sd->nr_balance_failed++; | |
2803 | ||
2804 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2805 | |
fe2eea3f | 2806 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2807 | |
2808 | /* don't kick the migration_thread, if the curr | |
2809 | * task on busiest cpu can't be moved to this_cpu | |
2810 | */ | |
2811 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2812 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2813 | all_pinned = 1; |
2814 | goto out_one_pinned; | |
2815 | } | |
2816 | ||
1da177e4 LT |
2817 | if (!busiest->active_balance) { |
2818 | busiest->active_balance = 1; | |
2819 | busiest->push_cpu = this_cpu; | |
81026794 | 2820 | active_balance = 1; |
1da177e4 | 2821 | } |
fe2eea3f | 2822 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2823 | if (active_balance) |
1da177e4 LT |
2824 | wake_up_process(busiest->migration_thread); |
2825 | ||
2826 | /* | |
2827 | * We've kicked active balancing, reset the failure | |
2828 | * counter. | |
2829 | */ | |
39507451 | 2830 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2831 | } |
81026794 | 2832 | } else |
1da177e4 LT |
2833 | sd->nr_balance_failed = 0; |
2834 | ||
81026794 | 2835 | if (likely(!active_balance)) { |
1da177e4 LT |
2836 | /* We were unbalanced, so reset the balancing interval */ |
2837 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2838 | } else { |
2839 | /* | |
2840 | * If we've begun active balancing, start to back off. This | |
2841 | * case may not be covered by the all_pinned logic if there | |
2842 | * is only 1 task on the busy runqueue (because we don't call | |
2843 | * move_tasks). | |
2844 | */ | |
2845 | if (sd->balance_interval < sd->max_interval) | |
2846 | sd->balance_interval *= 2; | |
1da177e4 LT |
2847 | } |
2848 | ||
43010659 | 2849 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2850 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2851 | return -1; |
43010659 | 2852 | return ld_moved; |
1da177e4 LT |
2853 | |
2854 | out_balanced: | |
1da177e4 LT |
2855 | schedstat_inc(sd, lb_balanced[idle]); |
2856 | ||
16cfb1c0 | 2857 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2858 | |
2859 | out_one_pinned: | |
1da177e4 | 2860 | /* tune up the balancing interval */ |
77391d71 NP |
2861 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2862 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2863 | sd->balance_interval *= 2; |
2864 | ||
48f24c4d | 2865 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2866 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2867 | return -1; |
1da177e4 LT |
2868 | return 0; |
2869 | } | |
2870 | ||
2871 | /* | |
2872 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2873 | * tasks if there is an imbalance. | |
2874 | * | |
d15bcfdb | 2875 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2876 | * this_rq is locked. |
2877 | */ | |
48f24c4d | 2878 | static int |
70b97a7f | 2879 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2880 | { |
2881 | struct sched_group *group; | |
70b97a7f | 2882 | struct rq *busiest = NULL; |
1da177e4 | 2883 | unsigned long imbalance; |
43010659 | 2884 | int ld_moved = 0; |
5969fe06 | 2885 | int sd_idle = 0; |
969bb4e4 | 2886 | int all_pinned = 0; |
0a2966b4 | 2887 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2888 | |
89c4710e SS |
2889 | /* |
2890 | * When power savings policy is enabled for the parent domain, idle | |
2891 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2892 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2893 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2894 | */ |
2895 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2896 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2897 | sd_idle = 1; |
1da177e4 | 2898 | |
2d72376b | 2899 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2900 | redo: |
d15bcfdb | 2901 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2902 | &sd_idle, &cpus, NULL); |
1da177e4 | 2903 | if (!group) { |
d15bcfdb | 2904 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2905 | goto out_balanced; |
1da177e4 LT |
2906 | } |
2907 | ||
d15bcfdb | 2908 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2909 | &cpus); |
db935dbd | 2910 | if (!busiest) { |
d15bcfdb | 2911 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2912 | goto out_balanced; |
1da177e4 LT |
2913 | } |
2914 | ||
db935dbd NP |
2915 | BUG_ON(busiest == this_rq); |
2916 | ||
d15bcfdb | 2917 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2918 | |
43010659 | 2919 | ld_moved = 0; |
d6d5cfaf NP |
2920 | if (busiest->nr_running > 1) { |
2921 | /* Attempt to move tasks */ | |
2922 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2923 | /* this_rq->clock is already updated */ |
2924 | update_rq_clock(busiest); | |
43010659 | 2925 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2926 | imbalance, sd, CPU_NEWLY_IDLE, |
2927 | &all_pinned); | |
d6d5cfaf | 2928 | spin_unlock(&busiest->lock); |
0a2966b4 | 2929 | |
969bb4e4 | 2930 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
2931 | cpu_clear(cpu_of(busiest), cpus); |
2932 | if (!cpus_empty(cpus)) | |
2933 | goto redo; | |
2934 | } | |
d6d5cfaf NP |
2935 | } |
2936 | ||
43010659 | 2937 | if (!ld_moved) { |
d15bcfdb | 2938 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
2939 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
2940 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
2941 | return -1; |
2942 | } else | |
16cfb1c0 | 2943 | sd->nr_balance_failed = 0; |
1da177e4 | 2944 | |
43010659 | 2945 | return ld_moved; |
16cfb1c0 NP |
2946 | |
2947 | out_balanced: | |
d15bcfdb | 2948 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 2949 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2950 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2951 | return -1; |
16cfb1c0 | 2952 | sd->nr_balance_failed = 0; |
48f24c4d | 2953 | |
16cfb1c0 | 2954 | return 0; |
1da177e4 LT |
2955 | } |
2956 | ||
2957 | /* | |
2958 | * idle_balance is called by schedule() if this_cpu is about to become | |
2959 | * idle. Attempts to pull tasks from other CPUs. | |
2960 | */ | |
70b97a7f | 2961 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
2962 | { |
2963 | struct sched_domain *sd; | |
dd41f596 IM |
2964 | int pulled_task = -1; |
2965 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
2966 | |
2967 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
2968 | unsigned long interval; |
2969 | ||
2970 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
2971 | continue; | |
2972 | ||
2973 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 2974 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 2975 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
2976 | this_rq, sd); |
2977 | ||
2978 | interval = msecs_to_jiffies(sd->balance_interval); | |
2979 | if (time_after(next_balance, sd->last_balance + interval)) | |
2980 | next_balance = sd->last_balance + interval; | |
2981 | if (pulled_task) | |
2982 | break; | |
1da177e4 | 2983 | } |
dd41f596 | 2984 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
2985 | /* |
2986 | * We are going idle. next_balance may be set based on | |
2987 | * a busy processor. So reset next_balance. | |
2988 | */ | |
2989 | this_rq->next_balance = next_balance; | |
dd41f596 | 2990 | } |
1da177e4 LT |
2991 | } |
2992 | ||
2993 | /* | |
2994 | * active_load_balance is run by migration threads. It pushes running tasks | |
2995 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
2996 | * running on each physical CPU where possible, and avoids physical / | |
2997 | * logical imbalances. | |
2998 | * | |
2999 | * Called with busiest_rq locked. | |
3000 | */ | |
70b97a7f | 3001 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3002 | { |
39507451 | 3003 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3004 | struct sched_domain *sd; |
3005 | struct rq *target_rq; | |
39507451 | 3006 | |
48f24c4d | 3007 | /* Is there any task to move? */ |
39507451 | 3008 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3009 | return; |
3010 | ||
3011 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3012 | |
3013 | /* | |
39507451 | 3014 | * This condition is "impossible", if it occurs |
41a2d6cf | 3015 | * we need to fix it. Originally reported by |
39507451 | 3016 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3017 | */ |
39507451 | 3018 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3019 | |
39507451 NP |
3020 | /* move a task from busiest_rq to target_rq */ |
3021 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3022 | update_rq_clock(busiest_rq); |
3023 | update_rq_clock(target_rq); | |
39507451 NP |
3024 | |
3025 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3026 | for_each_domain(target_cpu, sd) { |
39507451 | 3027 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3028 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3029 | break; |
c96d145e | 3030 | } |
39507451 | 3031 | |
48f24c4d | 3032 | if (likely(sd)) { |
2d72376b | 3033 | schedstat_inc(sd, alb_count); |
39507451 | 3034 | |
43010659 PW |
3035 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3036 | sd, CPU_IDLE)) | |
48f24c4d IM |
3037 | schedstat_inc(sd, alb_pushed); |
3038 | else | |
3039 | schedstat_inc(sd, alb_failed); | |
3040 | } | |
39507451 | 3041 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3042 | } |
3043 | ||
46cb4b7c SS |
3044 | #ifdef CONFIG_NO_HZ |
3045 | static struct { | |
3046 | atomic_t load_balancer; | |
41a2d6cf | 3047 | cpumask_t cpu_mask; |
46cb4b7c SS |
3048 | } nohz ____cacheline_aligned = { |
3049 | .load_balancer = ATOMIC_INIT(-1), | |
3050 | .cpu_mask = CPU_MASK_NONE, | |
3051 | }; | |
3052 | ||
7835b98b | 3053 | /* |
46cb4b7c SS |
3054 | * This routine will try to nominate the ilb (idle load balancing) |
3055 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3056 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3057 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3058 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3059 | * arrives... | |
3060 | * | |
3061 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3062 | * for idle load balancing. ilb owner will still be part of | |
3063 | * nohz.cpu_mask.. | |
7835b98b | 3064 | * |
46cb4b7c SS |
3065 | * While stopping the tick, this cpu will become the ilb owner if there |
3066 | * is no other owner. And will be the owner till that cpu becomes busy | |
3067 | * or if all cpus in the system stop their ticks at which point | |
3068 | * there is no need for ilb owner. | |
3069 | * | |
3070 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3071 | * next busy scheduler_tick() | |
3072 | */ | |
3073 | int select_nohz_load_balancer(int stop_tick) | |
3074 | { | |
3075 | int cpu = smp_processor_id(); | |
3076 | ||
3077 | if (stop_tick) { | |
3078 | cpu_set(cpu, nohz.cpu_mask); | |
3079 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3080 | ||
3081 | /* | |
3082 | * If we are going offline and still the leader, give up! | |
3083 | */ | |
3084 | if (cpu_is_offline(cpu) && | |
3085 | atomic_read(&nohz.load_balancer) == cpu) { | |
3086 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3087 | BUG(); | |
3088 | return 0; | |
3089 | } | |
3090 | ||
3091 | /* time for ilb owner also to sleep */ | |
3092 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3093 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3094 | atomic_set(&nohz.load_balancer, -1); | |
3095 | return 0; | |
3096 | } | |
3097 | ||
3098 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3099 | /* make me the ilb owner */ | |
3100 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3101 | return 1; | |
3102 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3103 | return 1; | |
3104 | } else { | |
3105 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3106 | return 0; | |
3107 | ||
3108 | cpu_clear(cpu, nohz.cpu_mask); | |
3109 | ||
3110 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3111 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3112 | BUG(); | |
3113 | } | |
3114 | return 0; | |
3115 | } | |
3116 | #endif | |
3117 | ||
3118 | static DEFINE_SPINLOCK(balancing); | |
3119 | ||
3120 | /* | |
7835b98b CL |
3121 | * It checks each scheduling domain to see if it is due to be balanced, |
3122 | * and initiates a balancing operation if so. | |
3123 | * | |
3124 | * Balancing parameters are set up in arch_init_sched_domains. | |
3125 | */ | |
a9957449 | 3126 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3127 | { |
46cb4b7c SS |
3128 | int balance = 1; |
3129 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3130 | unsigned long interval; |
3131 | struct sched_domain *sd; | |
46cb4b7c | 3132 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3133 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3134 | int update_next_balance = 0; |
1da177e4 | 3135 | |
46cb4b7c | 3136 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3137 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3138 | continue; | |
3139 | ||
3140 | interval = sd->balance_interval; | |
d15bcfdb | 3141 | if (idle != CPU_IDLE) |
1da177e4 LT |
3142 | interval *= sd->busy_factor; |
3143 | ||
3144 | /* scale ms to jiffies */ | |
3145 | interval = msecs_to_jiffies(interval); | |
3146 | if (unlikely(!interval)) | |
3147 | interval = 1; | |
dd41f596 IM |
3148 | if (interval > HZ*NR_CPUS/10) |
3149 | interval = HZ*NR_CPUS/10; | |
3150 | ||
1da177e4 | 3151 | |
08c183f3 CL |
3152 | if (sd->flags & SD_SERIALIZE) { |
3153 | if (!spin_trylock(&balancing)) | |
3154 | goto out; | |
3155 | } | |
3156 | ||
c9819f45 | 3157 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3158 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3159 | /* |
3160 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3161 | * longer idle, or one of our SMT siblings is |
3162 | * not idle. | |
3163 | */ | |
d15bcfdb | 3164 | idle = CPU_NOT_IDLE; |
1da177e4 | 3165 | } |
1bd77f2d | 3166 | sd->last_balance = jiffies; |
1da177e4 | 3167 | } |
08c183f3 CL |
3168 | if (sd->flags & SD_SERIALIZE) |
3169 | spin_unlock(&balancing); | |
3170 | out: | |
f549da84 | 3171 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3172 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3173 | update_next_balance = 1; |
3174 | } | |
783609c6 SS |
3175 | |
3176 | /* | |
3177 | * Stop the load balance at this level. There is another | |
3178 | * CPU in our sched group which is doing load balancing more | |
3179 | * actively. | |
3180 | */ | |
3181 | if (!balance) | |
3182 | break; | |
1da177e4 | 3183 | } |
f549da84 SS |
3184 | |
3185 | /* | |
3186 | * next_balance will be updated only when there is a need. | |
3187 | * When the cpu is attached to null domain for ex, it will not be | |
3188 | * updated. | |
3189 | */ | |
3190 | if (likely(update_next_balance)) | |
3191 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3192 | } |
3193 | ||
3194 | /* | |
3195 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3196 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3197 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3198 | */ | |
3199 | static void run_rebalance_domains(struct softirq_action *h) | |
3200 | { | |
dd41f596 IM |
3201 | int this_cpu = smp_processor_id(); |
3202 | struct rq *this_rq = cpu_rq(this_cpu); | |
3203 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3204 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3205 | |
dd41f596 | 3206 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3207 | |
3208 | #ifdef CONFIG_NO_HZ | |
3209 | /* | |
3210 | * If this cpu is the owner for idle load balancing, then do the | |
3211 | * balancing on behalf of the other idle cpus whose ticks are | |
3212 | * stopped. | |
3213 | */ | |
dd41f596 IM |
3214 | if (this_rq->idle_at_tick && |
3215 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3216 | cpumask_t cpus = nohz.cpu_mask; |
3217 | struct rq *rq; | |
3218 | int balance_cpu; | |
3219 | ||
dd41f596 | 3220 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3221 | for_each_cpu_mask(balance_cpu, cpus) { |
3222 | /* | |
3223 | * If this cpu gets work to do, stop the load balancing | |
3224 | * work being done for other cpus. Next load | |
3225 | * balancing owner will pick it up. | |
3226 | */ | |
3227 | if (need_resched()) | |
3228 | break; | |
3229 | ||
de0cf899 | 3230 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3231 | |
3232 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3233 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3234 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3235 | } |
3236 | } | |
3237 | #endif | |
3238 | } | |
3239 | ||
3240 | /* | |
3241 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3242 | * | |
3243 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3244 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3245 | * if the whole system is idle. | |
3246 | */ | |
dd41f596 | 3247 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3248 | { |
46cb4b7c SS |
3249 | #ifdef CONFIG_NO_HZ |
3250 | /* | |
3251 | * If we were in the nohz mode recently and busy at the current | |
3252 | * scheduler tick, then check if we need to nominate new idle | |
3253 | * load balancer. | |
3254 | */ | |
3255 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3256 | rq->in_nohz_recently = 0; | |
3257 | ||
3258 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3259 | cpu_clear(cpu, nohz.cpu_mask); | |
3260 | atomic_set(&nohz.load_balancer, -1); | |
3261 | } | |
3262 | ||
3263 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3264 | /* | |
3265 | * simple selection for now: Nominate the | |
3266 | * first cpu in the nohz list to be the next | |
3267 | * ilb owner. | |
3268 | * | |
3269 | * TBD: Traverse the sched domains and nominate | |
3270 | * the nearest cpu in the nohz.cpu_mask. | |
3271 | */ | |
3272 | int ilb = first_cpu(nohz.cpu_mask); | |
3273 | ||
3274 | if (ilb != NR_CPUS) | |
3275 | resched_cpu(ilb); | |
3276 | } | |
3277 | } | |
3278 | ||
3279 | /* | |
3280 | * If this cpu is idle and doing idle load balancing for all the | |
3281 | * cpus with ticks stopped, is it time for that to stop? | |
3282 | */ | |
3283 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3284 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3285 | resched_cpu(cpu); | |
3286 | return; | |
3287 | } | |
3288 | ||
3289 | /* | |
3290 | * If this cpu is idle and the idle load balancing is done by | |
3291 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3292 | */ | |
3293 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3294 | cpu_isset(cpu, nohz.cpu_mask)) | |
3295 | return; | |
3296 | #endif | |
3297 | if (time_after_eq(jiffies, rq->next_balance)) | |
3298 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3299 | } |
dd41f596 IM |
3300 | |
3301 | #else /* CONFIG_SMP */ | |
3302 | ||
1da177e4 LT |
3303 | /* |
3304 | * on UP we do not need to balance between CPUs: | |
3305 | */ | |
70b97a7f | 3306 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3307 | { |
3308 | } | |
dd41f596 | 3309 | |
1da177e4 LT |
3310 | #endif |
3311 | ||
1da177e4 LT |
3312 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3313 | ||
3314 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3315 | ||
3316 | /* | |
41b86e9c IM |
3317 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3318 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3319 | */ |
41b86e9c | 3320 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3321 | { |
1da177e4 | 3322 | unsigned long flags; |
41b86e9c IM |
3323 | u64 ns, delta_exec; |
3324 | struct rq *rq; | |
48f24c4d | 3325 | |
41b86e9c IM |
3326 | rq = task_rq_lock(p, &flags); |
3327 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3328 | if (task_current(rq, p)) { |
a8e504d2 IM |
3329 | update_rq_clock(rq); |
3330 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3331 | if ((s64)delta_exec > 0) |
3332 | ns += delta_exec; | |
3333 | } | |
3334 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3335 | |
1da177e4 LT |
3336 | return ns; |
3337 | } | |
3338 | ||
1da177e4 LT |
3339 | /* |
3340 | * Account user cpu time to a process. | |
3341 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3342 | * @cputime: the cpu time spent in user space since the last update |
3343 | */ | |
3344 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3345 | { | |
3346 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3347 | cputime64_t tmp; | |
3348 | ||
3349 | p->utime = cputime_add(p->utime, cputime); | |
3350 | ||
3351 | /* Add user time to cpustat. */ | |
3352 | tmp = cputime_to_cputime64(cputime); | |
3353 | if (TASK_NICE(p) > 0) | |
3354 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3355 | else | |
3356 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3357 | } | |
3358 | ||
94886b84 LV |
3359 | /* |
3360 | * Account guest cpu time to a process. | |
3361 | * @p: the process that the cpu time gets accounted to | |
3362 | * @cputime: the cpu time spent in virtual machine since the last update | |
3363 | */ | |
f7402e03 | 3364 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3365 | { |
3366 | cputime64_t tmp; | |
3367 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3368 | ||
3369 | tmp = cputime_to_cputime64(cputime); | |
3370 | ||
3371 | p->utime = cputime_add(p->utime, cputime); | |
3372 | p->gtime = cputime_add(p->gtime, cputime); | |
3373 | ||
3374 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3375 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3376 | } | |
3377 | ||
c66f08be MN |
3378 | /* |
3379 | * Account scaled user cpu time to a process. | |
3380 | * @p: the process that the cpu time gets accounted to | |
3381 | * @cputime: the cpu time spent in user space since the last update | |
3382 | */ | |
3383 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3384 | { | |
3385 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3386 | } | |
3387 | ||
1da177e4 LT |
3388 | /* |
3389 | * Account system cpu time to a process. | |
3390 | * @p: the process that the cpu time gets accounted to | |
3391 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3392 | * @cputime: the cpu time spent in kernel space since the last update | |
3393 | */ | |
3394 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3395 | cputime_t cputime) | |
3396 | { | |
3397 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3398 | struct rq *rq = this_rq(); |
1da177e4 LT |
3399 | cputime64_t tmp; |
3400 | ||
9778385d CB |
3401 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3402 | return account_guest_time(p, cputime); | |
94886b84 | 3403 | |
1da177e4 LT |
3404 | p->stime = cputime_add(p->stime, cputime); |
3405 | ||
3406 | /* Add system time to cpustat. */ | |
3407 | tmp = cputime_to_cputime64(cputime); | |
3408 | if (hardirq_count() - hardirq_offset) | |
3409 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3410 | else if (softirq_count()) | |
3411 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3412 | else if (p != rq->idle) |
1da177e4 | 3413 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3414 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3415 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3416 | else | |
3417 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3418 | /* Account for system time used */ | |
3419 | acct_update_integrals(p); | |
1da177e4 LT |
3420 | } |
3421 | ||
c66f08be MN |
3422 | /* |
3423 | * Account scaled system cpu time to a process. | |
3424 | * @p: the process that the cpu time gets accounted to | |
3425 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3426 | * @cputime: the cpu time spent in kernel space since the last update | |
3427 | */ | |
3428 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3429 | { | |
3430 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3431 | } | |
3432 | ||
1da177e4 LT |
3433 | /* |
3434 | * Account for involuntary wait time. | |
3435 | * @p: the process from which the cpu time has been stolen | |
3436 | * @steal: the cpu time spent in involuntary wait | |
3437 | */ | |
3438 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3439 | { | |
3440 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3441 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3442 | struct rq *rq = this_rq(); |
1da177e4 LT |
3443 | |
3444 | if (p == rq->idle) { | |
3445 | p->stime = cputime_add(p->stime, steal); | |
3446 | if (atomic_read(&rq->nr_iowait) > 0) | |
3447 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3448 | else | |
3449 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3450 | } else |
1da177e4 LT |
3451 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3452 | } | |
3453 | ||
7835b98b CL |
3454 | /* |
3455 | * This function gets called by the timer code, with HZ frequency. | |
3456 | * We call it with interrupts disabled. | |
3457 | * | |
3458 | * It also gets called by the fork code, when changing the parent's | |
3459 | * timeslices. | |
3460 | */ | |
3461 | void scheduler_tick(void) | |
3462 | { | |
7835b98b CL |
3463 | int cpu = smp_processor_id(); |
3464 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3465 | struct task_struct *curr = rq->curr; |
529c7726 | 3466 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3467 | |
3468 | spin_lock(&rq->lock); | |
546fe3c9 | 3469 | __update_rq_clock(rq); |
529c7726 IM |
3470 | /* |
3471 | * Let rq->clock advance by at least TICK_NSEC: | |
3472 | */ | |
3473 | if (unlikely(rq->clock < next_tick)) | |
3474 | rq->clock = next_tick; | |
3475 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3476 | update_cpu_load(rq); |
dd41f596 IM |
3477 | if (curr != rq->idle) /* FIXME: needed? */ |
3478 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3479 | spin_unlock(&rq->lock); |
7835b98b | 3480 | |
e418e1c2 | 3481 | #ifdef CONFIG_SMP |
dd41f596 IM |
3482 | rq->idle_at_tick = idle_cpu(cpu); |
3483 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3484 | #endif |
1da177e4 LT |
3485 | } |
3486 | ||
1da177e4 LT |
3487 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3488 | ||
3489 | void fastcall add_preempt_count(int val) | |
3490 | { | |
3491 | /* | |
3492 | * Underflow? | |
3493 | */ | |
9a11b49a IM |
3494 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3495 | return; | |
1da177e4 LT |
3496 | preempt_count() += val; |
3497 | /* | |
3498 | * Spinlock count overflowing soon? | |
3499 | */ | |
33859f7f MOS |
3500 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3501 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3502 | } |
3503 | EXPORT_SYMBOL(add_preempt_count); | |
3504 | ||
3505 | void fastcall sub_preempt_count(int val) | |
3506 | { | |
3507 | /* | |
3508 | * Underflow? | |
3509 | */ | |
9a11b49a IM |
3510 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3511 | return; | |
1da177e4 LT |
3512 | /* |
3513 | * Is the spinlock portion underflowing? | |
3514 | */ | |
9a11b49a IM |
3515 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3516 | !(preempt_count() & PREEMPT_MASK))) | |
3517 | return; | |
3518 | ||
1da177e4 LT |
3519 | preempt_count() -= val; |
3520 | } | |
3521 | EXPORT_SYMBOL(sub_preempt_count); | |
3522 | ||
3523 | #endif | |
3524 | ||
3525 | /* | |
dd41f596 | 3526 | * Print scheduling while atomic bug: |
1da177e4 | 3527 | */ |
dd41f596 | 3528 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3529 | { |
838225b4 SS |
3530 | struct pt_regs *regs = get_irq_regs(); |
3531 | ||
3532 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3533 | prev->comm, prev->pid, preempt_count()); | |
3534 | ||
dd41f596 IM |
3535 | debug_show_held_locks(prev); |
3536 | if (irqs_disabled()) | |
3537 | print_irqtrace_events(prev); | |
838225b4 SS |
3538 | |
3539 | if (regs) | |
3540 | show_regs(regs); | |
3541 | else | |
3542 | dump_stack(); | |
dd41f596 | 3543 | } |
1da177e4 | 3544 | |
dd41f596 IM |
3545 | /* |
3546 | * Various schedule()-time debugging checks and statistics: | |
3547 | */ | |
3548 | static inline void schedule_debug(struct task_struct *prev) | |
3549 | { | |
1da177e4 | 3550 | /* |
41a2d6cf | 3551 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3552 | * schedule() atomically, we ignore that path for now. |
3553 | * Otherwise, whine if we are scheduling when we should not be. | |
3554 | */ | |
dd41f596 IM |
3555 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3556 | __schedule_bug(prev); | |
3557 | ||
1da177e4 LT |
3558 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3559 | ||
2d72376b | 3560 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3561 | #ifdef CONFIG_SCHEDSTATS |
3562 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3563 | schedstat_inc(this_rq(), bkl_count); |
3564 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3565 | } |
3566 | #endif | |
dd41f596 IM |
3567 | } |
3568 | ||
3569 | /* | |
3570 | * Pick up the highest-prio task: | |
3571 | */ | |
3572 | static inline struct task_struct * | |
ff95f3df | 3573 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3574 | { |
5522d5d5 | 3575 | const struct sched_class *class; |
dd41f596 | 3576 | struct task_struct *p; |
1da177e4 LT |
3577 | |
3578 | /* | |
dd41f596 IM |
3579 | * Optimization: we know that if all tasks are in |
3580 | * the fair class we can call that function directly: | |
1da177e4 | 3581 | */ |
dd41f596 | 3582 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3583 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3584 | if (likely(p)) |
3585 | return p; | |
1da177e4 LT |
3586 | } |
3587 | ||
dd41f596 IM |
3588 | class = sched_class_highest; |
3589 | for ( ; ; ) { | |
fb8d4724 | 3590 | p = class->pick_next_task(rq); |
dd41f596 IM |
3591 | if (p) |
3592 | return p; | |
3593 | /* | |
3594 | * Will never be NULL as the idle class always | |
3595 | * returns a non-NULL p: | |
3596 | */ | |
3597 | class = class->next; | |
3598 | } | |
3599 | } | |
1da177e4 | 3600 | |
dd41f596 IM |
3601 | /* |
3602 | * schedule() is the main scheduler function. | |
3603 | */ | |
3604 | asmlinkage void __sched schedule(void) | |
3605 | { | |
3606 | struct task_struct *prev, *next; | |
3607 | long *switch_count; | |
3608 | struct rq *rq; | |
dd41f596 IM |
3609 | int cpu; |
3610 | ||
3611 | need_resched: | |
3612 | preempt_disable(); | |
3613 | cpu = smp_processor_id(); | |
3614 | rq = cpu_rq(cpu); | |
3615 | rcu_qsctr_inc(cpu); | |
3616 | prev = rq->curr; | |
3617 | switch_count = &prev->nivcsw; | |
3618 | ||
3619 | release_kernel_lock(prev); | |
3620 | need_resched_nonpreemptible: | |
3621 | ||
3622 | schedule_debug(prev); | |
1da177e4 | 3623 | |
1e819950 IM |
3624 | /* |
3625 | * Do the rq-clock update outside the rq lock: | |
3626 | */ | |
3627 | local_irq_disable(); | |
c1b3da3e | 3628 | __update_rq_clock(rq); |
1e819950 IM |
3629 | spin_lock(&rq->lock); |
3630 | clear_tsk_need_resched(prev); | |
1da177e4 | 3631 | |
1da177e4 | 3632 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3633 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3634 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3635 | prev->state = TASK_RUNNING; |
dd41f596 | 3636 | } else { |
2e1cb74a | 3637 | deactivate_task(rq, prev, 1); |
1da177e4 | 3638 | } |
dd41f596 | 3639 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3640 | } |
3641 | ||
dd41f596 | 3642 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3643 | idle_balance(cpu, rq); |
1da177e4 | 3644 | |
31ee529c | 3645 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3646 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3647 | |
3648 | sched_info_switch(prev, next); | |
dd41f596 | 3649 | |
1da177e4 | 3650 | if (likely(prev != next)) { |
1da177e4 LT |
3651 | rq->nr_switches++; |
3652 | rq->curr = next; | |
3653 | ++*switch_count; | |
3654 | ||
dd41f596 | 3655 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3656 | } else |
3657 | spin_unlock_irq(&rq->lock); | |
3658 | ||
dd41f596 IM |
3659 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3660 | cpu = smp_processor_id(); | |
3661 | rq = cpu_rq(cpu); | |
1da177e4 | 3662 | goto need_resched_nonpreemptible; |
dd41f596 | 3663 | } |
1da177e4 LT |
3664 | preempt_enable_no_resched(); |
3665 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3666 | goto need_resched; | |
3667 | } | |
1da177e4 LT |
3668 | EXPORT_SYMBOL(schedule); |
3669 | ||
3670 | #ifdef CONFIG_PREEMPT | |
3671 | /* | |
2ed6e34f | 3672 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3673 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3674 | * occur there and call schedule directly. |
3675 | */ | |
3676 | asmlinkage void __sched preempt_schedule(void) | |
3677 | { | |
3678 | struct thread_info *ti = current_thread_info(); | |
3679 | #ifdef CONFIG_PREEMPT_BKL | |
3680 | struct task_struct *task = current; | |
3681 | int saved_lock_depth; | |
3682 | #endif | |
3683 | /* | |
3684 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3685 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3686 | */ |
beed33a8 | 3687 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3688 | return; |
3689 | ||
3a5c359a AK |
3690 | do { |
3691 | add_preempt_count(PREEMPT_ACTIVE); | |
3692 | ||
3693 | /* | |
3694 | * We keep the big kernel semaphore locked, but we | |
3695 | * clear ->lock_depth so that schedule() doesnt | |
3696 | * auto-release the semaphore: | |
3697 | */ | |
1da177e4 | 3698 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3699 | saved_lock_depth = task->lock_depth; |
3700 | task->lock_depth = -1; | |
1da177e4 | 3701 | #endif |
3a5c359a | 3702 | schedule(); |
1da177e4 | 3703 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3704 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3705 | #endif |
3a5c359a | 3706 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3707 | |
3a5c359a AK |
3708 | /* |
3709 | * Check again in case we missed a preemption opportunity | |
3710 | * between schedule and now. | |
3711 | */ | |
3712 | barrier(); | |
3713 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3714 | } |
1da177e4 LT |
3715 | EXPORT_SYMBOL(preempt_schedule); |
3716 | ||
3717 | /* | |
2ed6e34f | 3718 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3719 | * off of irq context. |
3720 | * Note, that this is called and return with irqs disabled. This will | |
3721 | * protect us against recursive calling from irq. | |
3722 | */ | |
3723 | asmlinkage void __sched preempt_schedule_irq(void) | |
3724 | { | |
3725 | struct thread_info *ti = current_thread_info(); | |
3726 | #ifdef CONFIG_PREEMPT_BKL | |
3727 | struct task_struct *task = current; | |
3728 | int saved_lock_depth; | |
3729 | #endif | |
2ed6e34f | 3730 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3731 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3732 | ||
3a5c359a AK |
3733 | do { |
3734 | add_preempt_count(PREEMPT_ACTIVE); | |
3735 | ||
3736 | /* | |
3737 | * We keep the big kernel semaphore locked, but we | |
3738 | * clear ->lock_depth so that schedule() doesnt | |
3739 | * auto-release the semaphore: | |
3740 | */ | |
1da177e4 | 3741 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3742 | saved_lock_depth = task->lock_depth; |
3743 | task->lock_depth = -1; | |
1da177e4 | 3744 | #endif |
3a5c359a AK |
3745 | local_irq_enable(); |
3746 | schedule(); | |
3747 | local_irq_disable(); | |
1da177e4 | 3748 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3749 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3750 | #endif |
3a5c359a | 3751 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3752 | |
3a5c359a AK |
3753 | /* |
3754 | * Check again in case we missed a preemption opportunity | |
3755 | * between schedule and now. | |
3756 | */ | |
3757 | barrier(); | |
3758 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3759 | } |
3760 | ||
3761 | #endif /* CONFIG_PREEMPT */ | |
3762 | ||
95cdf3b7 IM |
3763 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3764 | void *key) | |
1da177e4 | 3765 | { |
48f24c4d | 3766 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3767 | } |
1da177e4 LT |
3768 | EXPORT_SYMBOL(default_wake_function); |
3769 | ||
3770 | /* | |
41a2d6cf IM |
3771 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3772 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3773 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3774 | * | |
3775 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3776 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3777 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3778 | */ | |
3779 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3780 | int nr_exclusive, int sync, void *key) | |
3781 | { | |
2e45874c | 3782 | wait_queue_t *curr, *next; |
1da177e4 | 3783 | |
2e45874c | 3784 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3785 | unsigned flags = curr->flags; |
3786 | ||
1da177e4 | 3787 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3788 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3789 | break; |
3790 | } | |
3791 | } | |
3792 | ||
3793 | /** | |
3794 | * __wake_up - wake up threads blocked on a waitqueue. | |
3795 | * @q: the waitqueue | |
3796 | * @mode: which threads | |
3797 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3798 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3799 | */ |
3800 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3801 | int nr_exclusive, void *key) |
1da177e4 LT |
3802 | { |
3803 | unsigned long flags; | |
3804 | ||
3805 | spin_lock_irqsave(&q->lock, flags); | |
3806 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3807 | spin_unlock_irqrestore(&q->lock, flags); | |
3808 | } | |
1da177e4 LT |
3809 | EXPORT_SYMBOL(__wake_up); |
3810 | ||
3811 | /* | |
3812 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3813 | */ | |
3814 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3815 | { | |
3816 | __wake_up_common(q, mode, 1, 0, NULL); | |
3817 | } | |
3818 | ||
3819 | /** | |
67be2dd1 | 3820 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3821 | * @q: the waitqueue |
3822 | * @mode: which threads | |
3823 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3824 | * | |
3825 | * The sync wakeup differs that the waker knows that it will schedule | |
3826 | * away soon, so while the target thread will be woken up, it will not | |
3827 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3828 | * with each other. This can prevent needless bouncing between CPUs. | |
3829 | * | |
3830 | * On UP it can prevent extra preemption. | |
3831 | */ | |
95cdf3b7 IM |
3832 | void fastcall |
3833 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3834 | { |
3835 | unsigned long flags; | |
3836 | int sync = 1; | |
3837 | ||
3838 | if (unlikely(!q)) | |
3839 | return; | |
3840 | ||
3841 | if (unlikely(!nr_exclusive)) | |
3842 | sync = 0; | |
3843 | ||
3844 | spin_lock_irqsave(&q->lock, flags); | |
3845 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3846 | spin_unlock_irqrestore(&q->lock, flags); | |
3847 | } | |
3848 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3849 | ||
b15136e9 | 3850 | void complete(struct completion *x) |
1da177e4 LT |
3851 | { |
3852 | unsigned long flags; | |
3853 | ||
3854 | spin_lock_irqsave(&x->wait.lock, flags); | |
3855 | x->done++; | |
3856 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3857 | 1, 0, NULL); | |
3858 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3859 | } | |
3860 | EXPORT_SYMBOL(complete); | |
3861 | ||
b15136e9 | 3862 | void complete_all(struct completion *x) |
1da177e4 LT |
3863 | { |
3864 | unsigned long flags; | |
3865 | ||
3866 | spin_lock_irqsave(&x->wait.lock, flags); | |
3867 | x->done += UINT_MAX/2; | |
3868 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3869 | 0, 0, NULL); | |
3870 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3871 | } | |
3872 | EXPORT_SYMBOL(complete_all); | |
3873 | ||
8cbbe86d AK |
3874 | static inline long __sched |
3875 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3876 | { |
1da177e4 LT |
3877 | if (!x->done) { |
3878 | DECLARE_WAITQUEUE(wait, current); | |
3879 | ||
3880 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3881 | __add_wait_queue_tail(&x->wait, &wait); | |
3882 | do { | |
8cbbe86d AK |
3883 | if (state == TASK_INTERRUPTIBLE && |
3884 | signal_pending(current)) { | |
3885 | __remove_wait_queue(&x->wait, &wait); | |
3886 | return -ERESTARTSYS; | |
3887 | } | |
3888 | __set_current_state(state); | |
1da177e4 LT |
3889 | spin_unlock_irq(&x->wait.lock); |
3890 | timeout = schedule_timeout(timeout); | |
3891 | spin_lock_irq(&x->wait.lock); | |
3892 | if (!timeout) { | |
3893 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3894 | return timeout; |
1da177e4 LT |
3895 | } |
3896 | } while (!x->done); | |
3897 | __remove_wait_queue(&x->wait, &wait); | |
3898 | } | |
3899 | x->done--; | |
1da177e4 LT |
3900 | return timeout; |
3901 | } | |
1da177e4 | 3902 | |
8cbbe86d AK |
3903 | static long __sched |
3904 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3905 | { |
1da177e4 LT |
3906 | might_sleep(); |
3907 | ||
3908 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3909 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3910 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3911 | return timeout; |
3912 | } | |
1da177e4 | 3913 | |
b15136e9 | 3914 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3915 | { |
3916 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3917 | } |
8cbbe86d | 3918 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3919 | |
b15136e9 | 3920 | unsigned long __sched |
8cbbe86d | 3921 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3922 | { |
8cbbe86d | 3923 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3924 | } |
8cbbe86d | 3925 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3926 | |
8cbbe86d | 3927 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 3928 | { |
51e97990 AK |
3929 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
3930 | if (t == -ERESTARTSYS) | |
3931 | return t; | |
3932 | return 0; | |
0fec171c | 3933 | } |
8cbbe86d | 3934 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 3935 | |
b15136e9 | 3936 | unsigned long __sched |
8cbbe86d AK |
3937 | wait_for_completion_interruptible_timeout(struct completion *x, |
3938 | unsigned long timeout) | |
0fec171c | 3939 | { |
8cbbe86d | 3940 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 3941 | } |
8cbbe86d | 3942 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 3943 | |
8cbbe86d AK |
3944 | static long __sched |
3945 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 3946 | { |
0fec171c IM |
3947 | unsigned long flags; |
3948 | wait_queue_t wait; | |
3949 | ||
3950 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 3951 | |
8cbbe86d | 3952 | __set_current_state(state); |
1da177e4 | 3953 | |
8cbbe86d AK |
3954 | spin_lock_irqsave(&q->lock, flags); |
3955 | __add_wait_queue(q, &wait); | |
3956 | spin_unlock(&q->lock); | |
3957 | timeout = schedule_timeout(timeout); | |
3958 | spin_lock_irq(&q->lock); | |
3959 | __remove_wait_queue(q, &wait); | |
3960 | spin_unlock_irqrestore(&q->lock, flags); | |
3961 | ||
3962 | return timeout; | |
3963 | } | |
3964 | ||
3965 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
3966 | { | |
3967 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 3968 | } |
1da177e4 LT |
3969 | EXPORT_SYMBOL(interruptible_sleep_on); |
3970 | ||
0fec171c | 3971 | long __sched |
95cdf3b7 | 3972 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3973 | { |
8cbbe86d | 3974 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 3975 | } |
1da177e4 LT |
3976 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
3977 | ||
0fec171c | 3978 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 3979 | { |
8cbbe86d | 3980 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 3981 | } |
1da177e4 LT |
3982 | EXPORT_SYMBOL(sleep_on); |
3983 | ||
0fec171c | 3984 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 3985 | { |
8cbbe86d | 3986 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 3987 | } |
1da177e4 LT |
3988 | EXPORT_SYMBOL(sleep_on_timeout); |
3989 | ||
b29739f9 IM |
3990 | #ifdef CONFIG_RT_MUTEXES |
3991 | ||
3992 | /* | |
3993 | * rt_mutex_setprio - set the current priority of a task | |
3994 | * @p: task | |
3995 | * @prio: prio value (kernel-internal form) | |
3996 | * | |
3997 | * This function changes the 'effective' priority of a task. It does | |
3998 | * not touch ->normal_prio like __setscheduler(). | |
3999 | * | |
4000 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4001 | */ | |
36c8b586 | 4002 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4003 | { |
4004 | unsigned long flags; | |
83b699ed | 4005 | int oldprio, on_rq, running; |
70b97a7f | 4006 | struct rq *rq; |
b29739f9 IM |
4007 | |
4008 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4009 | ||
4010 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4011 | update_rq_clock(rq); |
b29739f9 | 4012 | |
d5f9f942 | 4013 | oldprio = p->prio; |
dd41f596 | 4014 | on_rq = p->se.on_rq; |
051a1d1a | 4015 | running = task_current(rq, p); |
83b699ed | 4016 | if (on_rq) { |
69be72c1 | 4017 | dequeue_task(rq, p, 0); |
83b699ed SV |
4018 | if (running) |
4019 | p->sched_class->put_prev_task(rq, p); | |
4020 | } | |
dd41f596 IM |
4021 | |
4022 | if (rt_prio(prio)) | |
4023 | p->sched_class = &rt_sched_class; | |
4024 | else | |
4025 | p->sched_class = &fair_sched_class; | |
4026 | ||
b29739f9 IM |
4027 | p->prio = prio; |
4028 | ||
dd41f596 | 4029 | if (on_rq) { |
83b699ed SV |
4030 | if (running) |
4031 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4032 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4033 | /* |
4034 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4035 | * our priority decreased, or if we are not currently running on |
4036 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4037 | */ |
83b699ed | 4038 | if (running) { |
d5f9f942 AM |
4039 | if (p->prio > oldprio) |
4040 | resched_task(rq->curr); | |
dd41f596 IM |
4041 | } else { |
4042 | check_preempt_curr(rq, p); | |
4043 | } | |
b29739f9 IM |
4044 | } |
4045 | task_rq_unlock(rq, &flags); | |
4046 | } | |
4047 | ||
4048 | #endif | |
4049 | ||
36c8b586 | 4050 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4051 | { |
dd41f596 | 4052 | int old_prio, delta, on_rq; |
1da177e4 | 4053 | unsigned long flags; |
70b97a7f | 4054 | struct rq *rq; |
1da177e4 LT |
4055 | |
4056 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4057 | return; | |
4058 | /* | |
4059 | * We have to be careful, if called from sys_setpriority(), | |
4060 | * the task might be in the middle of scheduling on another CPU. | |
4061 | */ | |
4062 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4063 | update_rq_clock(rq); |
1da177e4 LT |
4064 | /* |
4065 | * The RT priorities are set via sched_setscheduler(), but we still | |
4066 | * allow the 'normal' nice value to be set - but as expected | |
4067 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4068 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4069 | */ |
e05606d3 | 4070 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4071 | p->static_prio = NICE_TO_PRIO(nice); |
4072 | goto out_unlock; | |
4073 | } | |
dd41f596 IM |
4074 | on_rq = p->se.on_rq; |
4075 | if (on_rq) { | |
69be72c1 | 4076 | dequeue_task(rq, p, 0); |
79b5dddf | 4077 | dec_load(rq, p); |
2dd73a4f | 4078 | } |
1da177e4 | 4079 | |
1da177e4 | 4080 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4081 | set_load_weight(p); |
b29739f9 IM |
4082 | old_prio = p->prio; |
4083 | p->prio = effective_prio(p); | |
4084 | delta = p->prio - old_prio; | |
1da177e4 | 4085 | |
dd41f596 | 4086 | if (on_rq) { |
8159f87e | 4087 | enqueue_task(rq, p, 0); |
29b4b623 | 4088 | inc_load(rq, p); |
1da177e4 | 4089 | /* |
d5f9f942 AM |
4090 | * If the task increased its priority or is running and |
4091 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4092 | */ |
d5f9f942 | 4093 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4094 | resched_task(rq->curr); |
4095 | } | |
4096 | out_unlock: | |
4097 | task_rq_unlock(rq, &flags); | |
4098 | } | |
1da177e4 LT |
4099 | EXPORT_SYMBOL(set_user_nice); |
4100 | ||
e43379f1 MM |
4101 | /* |
4102 | * can_nice - check if a task can reduce its nice value | |
4103 | * @p: task | |
4104 | * @nice: nice value | |
4105 | */ | |
36c8b586 | 4106 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4107 | { |
024f4747 MM |
4108 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4109 | int nice_rlim = 20 - nice; | |
48f24c4d | 4110 | |
e43379f1 MM |
4111 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4112 | capable(CAP_SYS_NICE)); | |
4113 | } | |
4114 | ||
1da177e4 LT |
4115 | #ifdef __ARCH_WANT_SYS_NICE |
4116 | ||
4117 | /* | |
4118 | * sys_nice - change the priority of the current process. | |
4119 | * @increment: priority increment | |
4120 | * | |
4121 | * sys_setpriority is a more generic, but much slower function that | |
4122 | * does similar things. | |
4123 | */ | |
4124 | asmlinkage long sys_nice(int increment) | |
4125 | { | |
48f24c4d | 4126 | long nice, retval; |
1da177e4 LT |
4127 | |
4128 | /* | |
4129 | * Setpriority might change our priority at the same moment. | |
4130 | * We don't have to worry. Conceptually one call occurs first | |
4131 | * and we have a single winner. | |
4132 | */ | |
e43379f1 MM |
4133 | if (increment < -40) |
4134 | increment = -40; | |
1da177e4 LT |
4135 | if (increment > 40) |
4136 | increment = 40; | |
4137 | ||
4138 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4139 | if (nice < -20) | |
4140 | nice = -20; | |
4141 | if (nice > 19) | |
4142 | nice = 19; | |
4143 | ||
e43379f1 MM |
4144 | if (increment < 0 && !can_nice(current, nice)) |
4145 | return -EPERM; | |
4146 | ||
1da177e4 LT |
4147 | retval = security_task_setnice(current, nice); |
4148 | if (retval) | |
4149 | return retval; | |
4150 | ||
4151 | set_user_nice(current, nice); | |
4152 | return 0; | |
4153 | } | |
4154 | ||
4155 | #endif | |
4156 | ||
4157 | /** | |
4158 | * task_prio - return the priority value of a given task. | |
4159 | * @p: the task in question. | |
4160 | * | |
4161 | * This is the priority value as seen by users in /proc. | |
4162 | * RT tasks are offset by -200. Normal tasks are centered | |
4163 | * around 0, value goes from -16 to +15. | |
4164 | */ | |
36c8b586 | 4165 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4166 | { |
4167 | return p->prio - MAX_RT_PRIO; | |
4168 | } | |
4169 | ||
4170 | /** | |
4171 | * task_nice - return the nice value of a given task. | |
4172 | * @p: the task in question. | |
4173 | */ | |
36c8b586 | 4174 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4175 | { |
4176 | return TASK_NICE(p); | |
4177 | } | |
1da177e4 | 4178 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4179 | |
4180 | /** | |
4181 | * idle_cpu - is a given cpu idle currently? | |
4182 | * @cpu: the processor in question. | |
4183 | */ | |
4184 | int idle_cpu(int cpu) | |
4185 | { | |
4186 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4187 | } | |
4188 | ||
1da177e4 LT |
4189 | /** |
4190 | * idle_task - return the idle task for a given cpu. | |
4191 | * @cpu: the processor in question. | |
4192 | */ | |
36c8b586 | 4193 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4194 | { |
4195 | return cpu_rq(cpu)->idle; | |
4196 | } | |
4197 | ||
4198 | /** | |
4199 | * find_process_by_pid - find a process with a matching PID value. | |
4200 | * @pid: the pid in question. | |
4201 | */ | |
a9957449 | 4202 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4203 | { |
228ebcbe | 4204 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4205 | } |
4206 | ||
4207 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4208 | static void |
4209 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4210 | { |
dd41f596 | 4211 | BUG_ON(p->se.on_rq); |
48f24c4d | 4212 | |
1da177e4 | 4213 | p->policy = policy; |
dd41f596 IM |
4214 | switch (p->policy) { |
4215 | case SCHED_NORMAL: | |
4216 | case SCHED_BATCH: | |
4217 | case SCHED_IDLE: | |
4218 | p->sched_class = &fair_sched_class; | |
4219 | break; | |
4220 | case SCHED_FIFO: | |
4221 | case SCHED_RR: | |
4222 | p->sched_class = &rt_sched_class; | |
4223 | break; | |
4224 | } | |
4225 | ||
1da177e4 | 4226 | p->rt_priority = prio; |
b29739f9 IM |
4227 | p->normal_prio = normal_prio(p); |
4228 | /* we are holding p->pi_lock already */ | |
4229 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4230 | set_load_weight(p); |
1da177e4 LT |
4231 | } |
4232 | ||
4233 | /** | |
72fd4a35 | 4234 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4235 | * @p: the task in question. |
4236 | * @policy: new policy. | |
4237 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4238 | * |
72fd4a35 | 4239 | * NOTE that the task may be already dead. |
1da177e4 | 4240 | */ |
95cdf3b7 IM |
4241 | int sched_setscheduler(struct task_struct *p, int policy, |
4242 | struct sched_param *param) | |
1da177e4 | 4243 | { |
83b699ed | 4244 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4245 | unsigned long flags; |
70b97a7f | 4246 | struct rq *rq; |
1da177e4 | 4247 | |
66e5393a SR |
4248 | /* may grab non-irq protected spin_locks */ |
4249 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4250 | recheck: |
4251 | /* double check policy once rq lock held */ | |
4252 | if (policy < 0) | |
4253 | policy = oldpolicy = p->policy; | |
4254 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4255 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4256 | policy != SCHED_IDLE) | |
b0a9499c | 4257 | return -EINVAL; |
1da177e4 LT |
4258 | /* |
4259 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4260 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4261 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4262 | */ |
4263 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4264 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4265 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4266 | return -EINVAL; |
e05606d3 | 4267 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4268 | return -EINVAL; |
4269 | ||
37e4ab3f OC |
4270 | /* |
4271 | * Allow unprivileged RT tasks to decrease priority: | |
4272 | */ | |
4273 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4274 | if (rt_policy(policy)) { |
8dc3e909 | 4275 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4276 | |
4277 | if (!lock_task_sighand(p, &flags)) | |
4278 | return -ESRCH; | |
4279 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4280 | unlock_task_sighand(p, &flags); | |
4281 | ||
4282 | /* can't set/change the rt policy */ | |
4283 | if (policy != p->policy && !rlim_rtprio) | |
4284 | return -EPERM; | |
4285 | ||
4286 | /* can't increase priority */ | |
4287 | if (param->sched_priority > p->rt_priority && | |
4288 | param->sched_priority > rlim_rtprio) | |
4289 | return -EPERM; | |
4290 | } | |
dd41f596 IM |
4291 | /* |
4292 | * Like positive nice levels, dont allow tasks to | |
4293 | * move out of SCHED_IDLE either: | |
4294 | */ | |
4295 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4296 | return -EPERM; | |
5fe1d75f | 4297 | |
37e4ab3f OC |
4298 | /* can't change other user's priorities */ |
4299 | if ((current->euid != p->euid) && | |
4300 | (current->euid != p->uid)) | |
4301 | return -EPERM; | |
4302 | } | |
1da177e4 LT |
4303 | |
4304 | retval = security_task_setscheduler(p, policy, param); | |
4305 | if (retval) | |
4306 | return retval; | |
b29739f9 IM |
4307 | /* |
4308 | * make sure no PI-waiters arrive (or leave) while we are | |
4309 | * changing the priority of the task: | |
4310 | */ | |
4311 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4312 | /* |
4313 | * To be able to change p->policy safely, the apropriate | |
4314 | * runqueue lock must be held. | |
4315 | */ | |
b29739f9 | 4316 | rq = __task_rq_lock(p); |
1da177e4 LT |
4317 | /* recheck policy now with rq lock held */ |
4318 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4319 | policy = oldpolicy = -1; | |
b29739f9 IM |
4320 | __task_rq_unlock(rq); |
4321 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4322 | goto recheck; |
4323 | } | |
2daa3577 | 4324 | update_rq_clock(rq); |
dd41f596 | 4325 | on_rq = p->se.on_rq; |
051a1d1a | 4326 | running = task_current(rq, p); |
83b699ed | 4327 | if (on_rq) { |
2e1cb74a | 4328 | deactivate_task(rq, p, 0); |
83b699ed SV |
4329 | if (running) |
4330 | p->sched_class->put_prev_task(rq, p); | |
4331 | } | |
f6b53205 | 4332 | |
1da177e4 | 4333 | oldprio = p->prio; |
dd41f596 | 4334 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4335 | |
dd41f596 | 4336 | if (on_rq) { |
83b699ed SV |
4337 | if (running) |
4338 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4339 | activate_task(rq, p, 0); |
1da177e4 LT |
4340 | /* |
4341 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4342 | * our priority decreased, or if we are not currently running on |
4343 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4344 | */ |
83b699ed | 4345 | if (running) { |
d5f9f942 AM |
4346 | if (p->prio > oldprio) |
4347 | resched_task(rq->curr); | |
dd41f596 IM |
4348 | } else { |
4349 | check_preempt_curr(rq, p); | |
4350 | } | |
1da177e4 | 4351 | } |
b29739f9 IM |
4352 | __task_rq_unlock(rq); |
4353 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4354 | ||
95e02ca9 TG |
4355 | rt_mutex_adjust_pi(p); |
4356 | ||
1da177e4 LT |
4357 | return 0; |
4358 | } | |
4359 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4360 | ||
95cdf3b7 IM |
4361 | static int |
4362 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4363 | { |
1da177e4 LT |
4364 | struct sched_param lparam; |
4365 | struct task_struct *p; | |
36c8b586 | 4366 | int retval; |
1da177e4 LT |
4367 | |
4368 | if (!param || pid < 0) | |
4369 | return -EINVAL; | |
4370 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4371 | return -EFAULT; | |
5fe1d75f ON |
4372 | |
4373 | rcu_read_lock(); | |
4374 | retval = -ESRCH; | |
1da177e4 | 4375 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4376 | if (p != NULL) |
4377 | retval = sched_setscheduler(p, policy, &lparam); | |
4378 | rcu_read_unlock(); | |
36c8b586 | 4379 | |
1da177e4 LT |
4380 | return retval; |
4381 | } | |
4382 | ||
4383 | /** | |
4384 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4385 | * @pid: the pid in question. | |
4386 | * @policy: new policy. | |
4387 | * @param: structure containing the new RT priority. | |
4388 | */ | |
41a2d6cf IM |
4389 | asmlinkage long |
4390 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4391 | { |
c21761f1 JB |
4392 | /* negative values for policy are not valid */ |
4393 | if (policy < 0) | |
4394 | return -EINVAL; | |
4395 | ||
1da177e4 LT |
4396 | return do_sched_setscheduler(pid, policy, param); |
4397 | } | |
4398 | ||
4399 | /** | |
4400 | * sys_sched_setparam - set/change the RT priority of a thread | |
4401 | * @pid: the pid in question. | |
4402 | * @param: structure containing the new RT priority. | |
4403 | */ | |
4404 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4405 | { | |
4406 | return do_sched_setscheduler(pid, -1, param); | |
4407 | } | |
4408 | ||
4409 | /** | |
4410 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4411 | * @pid: the pid in question. | |
4412 | */ | |
4413 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4414 | { | |
36c8b586 | 4415 | struct task_struct *p; |
3a5c359a | 4416 | int retval; |
1da177e4 LT |
4417 | |
4418 | if (pid < 0) | |
3a5c359a | 4419 | return -EINVAL; |
1da177e4 LT |
4420 | |
4421 | retval = -ESRCH; | |
4422 | read_lock(&tasklist_lock); | |
4423 | p = find_process_by_pid(pid); | |
4424 | if (p) { | |
4425 | retval = security_task_getscheduler(p); | |
4426 | if (!retval) | |
4427 | retval = p->policy; | |
4428 | } | |
4429 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4430 | return retval; |
4431 | } | |
4432 | ||
4433 | /** | |
4434 | * sys_sched_getscheduler - get the RT priority of a thread | |
4435 | * @pid: the pid in question. | |
4436 | * @param: structure containing the RT priority. | |
4437 | */ | |
4438 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4439 | { | |
4440 | struct sched_param lp; | |
36c8b586 | 4441 | struct task_struct *p; |
3a5c359a | 4442 | int retval; |
1da177e4 LT |
4443 | |
4444 | if (!param || pid < 0) | |
3a5c359a | 4445 | return -EINVAL; |
1da177e4 LT |
4446 | |
4447 | read_lock(&tasklist_lock); | |
4448 | p = find_process_by_pid(pid); | |
4449 | retval = -ESRCH; | |
4450 | if (!p) | |
4451 | goto out_unlock; | |
4452 | ||
4453 | retval = security_task_getscheduler(p); | |
4454 | if (retval) | |
4455 | goto out_unlock; | |
4456 | ||
4457 | lp.sched_priority = p->rt_priority; | |
4458 | read_unlock(&tasklist_lock); | |
4459 | ||
4460 | /* | |
4461 | * This one might sleep, we cannot do it with a spinlock held ... | |
4462 | */ | |
4463 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4464 | ||
1da177e4 LT |
4465 | return retval; |
4466 | ||
4467 | out_unlock: | |
4468 | read_unlock(&tasklist_lock); | |
4469 | return retval; | |
4470 | } | |
4471 | ||
4472 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4473 | { | |
1da177e4 | 4474 | cpumask_t cpus_allowed; |
36c8b586 IM |
4475 | struct task_struct *p; |
4476 | int retval; | |
1da177e4 | 4477 | |
5be9361c | 4478 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4479 | read_lock(&tasklist_lock); |
4480 | ||
4481 | p = find_process_by_pid(pid); | |
4482 | if (!p) { | |
4483 | read_unlock(&tasklist_lock); | |
5be9361c | 4484 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4485 | return -ESRCH; |
4486 | } | |
4487 | ||
4488 | /* | |
4489 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4490 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4491 | * usage count and then drop tasklist_lock. |
4492 | */ | |
4493 | get_task_struct(p); | |
4494 | read_unlock(&tasklist_lock); | |
4495 | ||
4496 | retval = -EPERM; | |
4497 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4498 | !capable(CAP_SYS_NICE)) | |
4499 | goto out_unlock; | |
4500 | ||
e7834f8f DQ |
4501 | retval = security_task_setscheduler(p, 0, NULL); |
4502 | if (retval) | |
4503 | goto out_unlock; | |
4504 | ||
1da177e4 LT |
4505 | cpus_allowed = cpuset_cpus_allowed(p); |
4506 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4507 | again: |
1da177e4 LT |
4508 | retval = set_cpus_allowed(p, new_mask); |
4509 | ||
8707d8b8 PM |
4510 | if (!retval) { |
4511 | cpus_allowed = cpuset_cpus_allowed(p); | |
4512 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4513 | /* | |
4514 | * We must have raced with a concurrent cpuset | |
4515 | * update. Just reset the cpus_allowed to the | |
4516 | * cpuset's cpus_allowed | |
4517 | */ | |
4518 | new_mask = cpus_allowed; | |
4519 | goto again; | |
4520 | } | |
4521 | } | |
1da177e4 LT |
4522 | out_unlock: |
4523 | put_task_struct(p); | |
5be9361c | 4524 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
4525 | return retval; |
4526 | } | |
4527 | ||
4528 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4529 | cpumask_t *new_mask) | |
4530 | { | |
4531 | if (len < sizeof(cpumask_t)) { | |
4532 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4533 | } else if (len > sizeof(cpumask_t)) { | |
4534 | len = sizeof(cpumask_t); | |
4535 | } | |
4536 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4537 | } | |
4538 | ||
4539 | /** | |
4540 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4541 | * @pid: pid of the process | |
4542 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4543 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4544 | */ | |
4545 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4546 | unsigned long __user *user_mask_ptr) | |
4547 | { | |
4548 | cpumask_t new_mask; | |
4549 | int retval; | |
4550 | ||
4551 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4552 | if (retval) | |
4553 | return retval; | |
4554 | ||
4555 | return sched_setaffinity(pid, new_mask); | |
4556 | } | |
4557 | ||
4558 | /* | |
4559 | * Represents all cpu's present in the system | |
4560 | * In systems capable of hotplug, this map could dynamically grow | |
4561 | * as new cpu's are detected in the system via any platform specific | |
4562 | * method, such as ACPI for e.g. | |
4563 | */ | |
4564 | ||
4cef0c61 | 4565 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4566 | EXPORT_SYMBOL(cpu_present_map); |
4567 | ||
4568 | #ifndef CONFIG_SMP | |
4cef0c61 | 4569 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4570 | EXPORT_SYMBOL(cpu_online_map); |
4571 | ||
4cef0c61 | 4572 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4573 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4574 | #endif |
4575 | ||
4576 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4577 | { | |
36c8b586 | 4578 | struct task_struct *p; |
1da177e4 | 4579 | int retval; |
1da177e4 | 4580 | |
5be9361c | 4581 | mutex_lock(&sched_hotcpu_mutex); |
1da177e4 LT |
4582 | read_lock(&tasklist_lock); |
4583 | ||
4584 | retval = -ESRCH; | |
4585 | p = find_process_by_pid(pid); | |
4586 | if (!p) | |
4587 | goto out_unlock; | |
4588 | ||
e7834f8f DQ |
4589 | retval = security_task_getscheduler(p); |
4590 | if (retval) | |
4591 | goto out_unlock; | |
4592 | ||
2f7016d9 | 4593 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4594 | |
4595 | out_unlock: | |
4596 | read_unlock(&tasklist_lock); | |
5be9361c | 4597 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 | 4598 | |
9531b62f | 4599 | return retval; |
1da177e4 LT |
4600 | } |
4601 | ||
4602 | /** | |
4603 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4604 | * @pid: pid of the process | |
4605 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4606 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4607 | */ | |
4608 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4609 | unsigned long __user *user_mask_ptr) | |
4610 | { | |
4611 | int ret; | |
4612 | cpumask_t mask; | |
4613 | ||
4614 | if (len < sizeof(cpumask_t)) | |
4615 | return -EINVAL; | |
4616 | ||
4617 | ret = sched_getaffinity(pid, &mask); | |
4618 | if (ret < 0) | |
4619 | return ret; | |
4620 | ||
4621 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4622 | return -EFAULT; | |
4623 | ||
4624 | return sizeof(cpumask_t); | |
4625 | } | |
4626 | ||
4627 | /** | |
4628 | * sys_sched_yield - yield the current processor to other threads. | |
4629 | * | |
dd41f596 IM |
4630 | * This function yields the current CPU to other tasks. If there are no |
4631 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4632 | */ |
4633 | asmlinkage long sys_sched_yield(void) | |
4634 | { | |
70b97a7f | 4635 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4636 | |
2d72376b | 4637 | schedstat_inc(rq, yld_count); |
4530d7ab | 4638 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4639 | |
4640 | /* | |
4641 | * Since we are going to call schedule() anyway, there's | |
4642 | * no need to preempt or enable interrupts: | |
4643 | */ | |
4644 | __release(rq->lock); | |
8a25d5de | 4645 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4646 | _raw_spin_unlock(&rq->lock); |
4647 | preempt_enable_no_resched(); | |
4648 | ||
4649 | schedule(); | |
4650 | ||
4651 | return 0; | |
4652 | } | |
4653 | ||
e7b38404 | 4654 | static void __cond_resched(void) |
1da177e4 | 4655 | { |
8e0a43d8 IM |
4656 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4657 | __might_sleep(__FILE__, __LINE__); | |
4658 | #endif | |
5bbcfd90 IM |
4659 | /* |
4660 | * The BKS might be reacquired before we have dropped | |
4661 | * PREEMPT_ACTIVE, which could trigger a second | |
4662 | * cond_resched() call. | |
4663 | */ | |
1da177e4 LT |
4664 | do { |
4665 | add_preempt_count(PREEMPT_ACTIVE); | |
4666 | schedule(); | |
4667 | sub_preempt_count(PREEMPT_ACTIVE); | |
4668 | } while (need_resched()); | |
4669 | } | |
4670 | ||
4671 | int __sched cond_resched(void) | |
4672 | { | |
9414232f IM |
4673 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4674 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4675 | __cond_resched(); |
4676 | return 1; | |
4677 | } | |
4678 | return 0; | |
4679 | } | |
1da177e4 LT |
4680 | EXPORT_SYMBOL(cond_resched); |
4681 | ||
4682 | /* | |
4683 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4684 | * call schedule, and on return reacquire the lock. | |
4685 | * | |
41a2d6cf | 4686 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4687 | * operations here to prevent schedule() from being called twice (once via |
4688 | * spin_unlock(), once by hand). | |
4689 | */ | |
95cdf3b7 | 4690 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4691 | { |
6df3cecb JK |
4692 | int ret = 0; |
4693 | ||
1da177e4 LT |
4694 | if (need_lockbreak(lock)) { |
4695 | spin_unlock(lock); | |
4696 | cpu_relax(); | |
6df3cecb | 4697 | ret = 1; |
1da177e4 LT |
4698 | spin_lock(lock); |
4699 | } | |
9414232f | 4700 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4701 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4702 | _raw_spin_unlock(lock); |
4703 | preempt_enable_no_resched(); | |
4704 | __cond_resched(); | |
6df3cecb | 4705 | ret = 1; |
1da177e4 | 4706 | spin_lock(lock); |
1da177e4 | 4707 | } |
6df3cecb | 4708 | return ret; |
1da177e4 | 4709 | } |
1da177e4 LT |
4710 | EXPORT_SYMBOL(cond_resched_lock); |
4711 | ||
4712 | int __sched cond_resched_softirq(void) | |
4713 | { | |
4714 | BUG_ON(!in_softirq()); | |
4715 | ||
9414232f | 4716 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4717 | local_bh_enable(); |
1da177e4 LT |
4718 | __cond_resched(); |
4719 | local_bh_disable(); | |
4720 | return 1; | |
4721 | } | |
4722 | return 0; | |
4723 | } | |
1da177e4 LT |
4724 | EXPORT_SYMBOL(cond_resched_softirq); |
4725 | ||
1da177e4 LT |
4726 | /** |
4727 | * yield - yield the current processor to other threads. | |
4728 | * | |
72fd4a35 | 4729 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4730 | * thread runnable and calls sys_sched_yield(). |
4731 | */ | |
4732 | void __sched yield(void) | |
4733 | { | |
4734 | set_current_state(TASK_RUNNING); | |
4735 | sys_sched_yield(); | |
4736 | } | |
1da177e4 LT |
4737 | EXPORT_SYMBOL(yield); |
4738 | ||
4739 | /* | |
41a2d6cf | 4740 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4741 | * that process accounting knows that this is a task in IO wait state. |
4742 | * | |
4743 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4744 | * has set its backing_dev_info: the queue against which it should throttle) | |
4745 | */ | |
4746 | void __sched io_schedule(void) | |
4747 | { | |
70b97a7f | 4748 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4749 | |
0ff92245 | 4750 | delayacct_blkio_start(); |
1da177e4 LT |
4751 | atomic_inc(&rq->nr_iowait); |
4752 | schedule(); | |
4753 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4754 | delayacct_blkio_end(); |
1da177e4 | 4755 | } |
1da177e4 LT |
4756 | EXPORT_SYMBOL(io_schedule); |
4757 | ||
4758 | long __sched io_schedule_timeout(long timeout) | |
4759 | { | |
70b97a7f | 4760 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4761 | long ret; |
4762 | ||
0ff92245 | 4763 | delayacct_blkio_start(); |
1da177e4 LT |
4764 | atomic_inc(&rq->nr_iowait); |
4765 | ret = schedule_timeout(timeout); | |
4766 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4767 | delayacct_blkio_end(); |
1da177e4 LT |
4768 | return ret; |
4769 | } | |
4770 | ||
4771 | /** | |
4772 | * sys_sched_get_priority_max - return maximum RT priority. | |
4773 | * @policy: scheduling class. | |
4774 | * | |
4775 | * this syscall returns the maximum rt_priority that can be used | |
4776 | * by a given scheduling class. | |
4777 | */ | |
4778 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4779 | { | |
4780 | int ret = -EINVAL; | |
4781 | ||
4782 | switch (policy) { | |
4783 | case SCHED_FIFO: | |
4784 | case SCHED_RR: | |
4785 | ret = MAX_USER_RT_PRIO-1; | |
4786 | break; | |
4787 | case SCHED_NORMAL: | |
b0a9499c | 4788 | case SCHED_BATCH: |
dd41f596 | 4789 | case SCHED_IDLE: |
1da177e4 LT |
4790 | ret = 0; |
4791 | break; | |
4792 | } | |
4793 | return ret; | |
4794 | } | |
4795 | ||
4796 | /** | |
4797 | * sys_sched_get_priority_min - return minimum RT priority. | |
4798 | * @policy: scheduling class. | |
4799 | * | |
4800 | * this syscall returns the minimum rt_priority that can be used | |
4801 | * by a given scheduling class. | |
4802 | */ | |
4803 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4804 | { | |
4805 | int ret = -EINVAL; | |
4806 | ||
4807 | switch (policy) { | |
4808 | case SCHED_FIFO: | |
4809 | case SCHED_RR: | |
4810 | ret = 1; | |
4811 | break; | |
4812 | case SCHED_NORMAL: | |
b0a9499c | 4813 | case SCHED_BATCH: |
dd41f596 | 4814 | case SCHED_IDLE: |
1da177e4 LT |
4815 | ret = 0; |
4816 | } | |
4817 | return ret; | |
4818 | } | |
4819 | ||
4820 | /** | |
4821 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4822 | * @pid: pid of the process. | |
4823 | * @interval: userspace pointer to the timeslice value. | |
4824 | * | |
4825 | * this syscall writes the default timeslice value of a given process | |
4826 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4827 | */ | |
4828 | asmlinkage | |
4829 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4830 | { | |
36c8b586 | 4831 | struct task_struct *p; |
a4ec24b4 | 4832 | unsigned int time_slice; |
3a5c359a | 4833 | int retval; |
1da177e4 | 4834 | struct timespec t; |
1da177e4 LT |
4835 | |
4836 | if (pid < 0) | |
3a5c359a | 4837 | return -EINVAL; |
1da177e4 LT |
4838 | |
4839 | retval = -ESRCH; | |
4840 | read_lock(&tasklist_lock); | |
4841 | p = find_process_by_pid(pid); | |
4842 | if (!p) | |
4843 | goto out_unlock; | |
4844 | ||
4845 | retval = security_task_getscheduler(p); | |
4846 | if (retval) | |
4847 | goto out_unlock; | |
4848 | ||
77034937 IM |
4849 | /* |
4850 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
4851 | * tasks that are on an otherwise idle runqueue: | |
4852 | */ | |
4853 | time_slice = 0; | |
4854 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 4855 | time_slice = DEF_TIMESLICE; |
77034937 | 4856 | } else { |
a4ec24b4 DA |
4857 | struct sched_entity *se = &p->se; |
4858 | unsigned long flags; | |
4859 | struct rq *rq; | |
4860 | ||
4861 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
4862 | if (rq->cfs.load.weight) |
4863 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
4864 | task_rq_unlock(rq, &flags); |
4865 | } | |
1da177e4 | 4866 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4867 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4868 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4869 | return retval; |
3a5c359a | 4870 | |
1da177e4 LT |
4871 | out_unlock: |
4872 | read_unlock(&tasklist_lock); | |
4873 | return retval; | |
4874 | } | |
4875 | ||
2ed6e34f | 4876 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4877 | |
4878 | static void show_task(struct task_struct *p) | |
1da177e4 | 4879 | { |
1da177e4 | 4880 | unsigned long free = 0; |
36c8b586 | 4881 | unsigned state; |
1da177e4 | 4882 | |
1da177e4 | 4883 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4884 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4885 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4886 | #if BITS_PER_LONG == 32 |
1da177e4 | 4887 | if (state == TASK_RUNNING) |
cc4ea795 | 4888 | printk(KERN_CONT " running "); |
1da177e4 | 4889 | else |
cc4ea795 | 4890 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4891 | #else |
4892 | if (state == TASK_RUNNING) | |
cc4ea795 | 4893 | printk(KERN_CONT " running task "); |
1da177e4 | 4894 | else |
cc4ea795 | 4895 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4896 | #endif |
4897 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4898 | { | |
10ebffde | 4899 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4900 | while (!*n) |
4901 | n++; | |
10ebffde | 4902 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4903 | } |
4904 | #endif | |
ba25f9dc | 4905 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 4906 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 LT |
4907 | |
4908 | if (state != TASK_RUNNING) | |
4909 | show_stack(p, NULL); | |
4910 | } | |
4911 | ||
e59e2ae2 | 4912 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4913 | { |
36c8b586 | 4914 | struct task_struct *g, *p; |
1da177e4 | 4915 | |
4bd77321 IM |
4916 | #if BITS_PER_LONG == 32 |
4917 | printk(KERN_INFO | |
4918 | " task PC stack pid father\n"); | |
1da177e4 | 4919 | #else |
4bd77321 IM |
4920 | printk(KERN_INFO |
4921 | " task PC stack pid father\n"); | |
1da177e4 LT |
4922 | #endif |
4923 | read_lock(&tasklist_lock); | |
4924 | do_each_thread(g, p) { | |
4925 | /* | |
4926 | * reset the NMI-timeout, listing all files on a slow | |
4927 | * console might take alot of time: | |
4928 | */ | |
4929 | touch_nmi_watchdog(); | |
39bc89fd | 4930 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 4931 | show_task(p); |
1da177e4 LT |
4932 | } while_each_thread(g, p); |
4933 | ||
04c9167f JF |
4934 | touch_all_softlockup_watchdogs(); |
4935 | ||
dd41f596 IM |
4936 | #ifdef CONFIG_SCHED_DEBUG |
4937 | sysrq_sched_debug_show(); | |
4938 | #endif | |
1da177e4 | 4939 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
4940 | /* |
4941 | * Only show locks if all tasks are dumped: | |
4942 | */ | |
4943 | if (state_filter == -1) | |
4944 | debug_show_all_locks(); | |
1da177e4 LT |
4945 | } |
4946 | ||
1df21055 IM |
4947 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
4948 | { | |
dd41f596 | 4949 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
4950 | } |
4951 | ||
f340c0d1 IM |
4952 | /** |
4953 | * init_idle - set up an idle thread for a given CPU | |
4954 | * @idle: task in question | |
4955 | * @cpu: cpu the idle task belongs to | |
4956 | * | |
4957 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
4958 | * flag, to make booting more robust. | |
4959 | */ | |
5c1e1767 | 4960 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 4961 | { |
70b97a7f | 4962 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
4963 | unsigned long flags; |
4964 | ||
dd41f596 IM |
4965 | __sched_fork(idle); |
4966 | idle->se.exec_start = sched_clock(); | |
4967 | ||
b29739f9 | 4968 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 4969 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 4970 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
4971 | |
4972 | spin_lock_irqsave(&rq->lock, flags); | |
4973 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
4974 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4975 | idle->oncpu = 1; | |
4976 | #endif | |
1da177e4 LT |
4977 | spin_unlock_irqrestore(&rq->lock, flags); |
4978 | ||
4979 | /* Set the preempt count _outside_ the spinlocks! */ | |
4980 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 4981 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 4982 | #else |
a1261f54 | 4983 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 4984 | #endif |
dd41f596 IM |
4985 | /* |
4986 | * The idle tasks have their own, simple scheduling class: | |
4987 | */ | |
4988 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
4989 | } |
4990 | ||
4991 | /* | |
4992 | * In a system that switches off the HZ timer nohz_cpu_mask | |
4993 | * indicates which cpus entered this state. This is used | |
4994 | * in the rcu update to wait only for active cpus. For system | |
4995 | * which do not switch off the HZ timer nohz_cpu_mask should | |
4996 | * always be CPU_MASK_NONE. | |
4997 | */ | |
4998 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
4999 | ||
19978ca6 IM |
5000 | /* |
5001 | * Increase the granularity value when there are more CPUs, | |
5002 | * because with more CPUs the 'effective latency' as visible | |
5003 | * to users decreases. But the relationship is not linear, | |
5004 | * so pick a second-best guess by going with the log2 of the | |
5005 | * number of CPUs. | |
5006 | * | |
5007 | * This idea comes from the SD scheduler of Con Kolivas: | |
5008 | */ | |
5009 | static inline void sched_init_granularity(void) | |
5010 | { | |
5011 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5012 | const unsigned long limit = 200000000; | |
5013 | ||
5014 | sysctl_sched_min_granularity *= factor; | |
5015 | if (sysctl_sched_min_granularity > limit) | |
5016 | sysctl_sched_min_granularity = limit; | |
5017 | ||
5018 | sysctl_sched_latency *= factor; | |
5019 | if (sysctl_sched_latency > limit) | |
5020 | sysctl_sched_latency = limit; | |
5021 | ||
5022 | sysctl_sched_wakeup_granularity *= factor; | |
5023 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5024 | } | |
5025 | ||
1da177e4 LT |
5026 | #ifdef CONFIG_SMP |
5027 | /* | |
5028 | * This is how migration works: | |
5029 | * | |
70b97a7f | 5030 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5031 | * runqueue and wake up that CPU's migration thread. |
5032 | * 2) we down() the locked semaphore => thread blocks. | |
5033 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5034 | * thread off the CPU) | |
5035 | * 4) it gets the migration request and checks whether the migrated | |
5036 | * task is still in the wrong runqueue. | |
5037 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5038 | * it and puts it into the right queue. | |
5039 | * 6) migration thread up()s the semaphore. | |
5040 | * 7) we wake up and the migration is done. | |
5041 | */ | |
5042 | ||
5043 | /* | |
5044 | * Change a given task's CPU affinity. Migrate the thread to a | |
5045 | * proper CPU and schedule it away if the CPU it's executing on | |
5046 | * is removed from the allowed bitmask. | |
5047 | * | |
5048 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5049 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5050 | * call is not atomic; no spinlocks may be held. |
5051 | */ | |
36c8b586 | 5052 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5053 | { |
70b97a7f | 5054 | struct migration_req req; |
1da177e4 | 5055 | unsigned long flags; |
70b97a7f | 5056 | struct rq *rq; |
48f24c4d | 5057 | int ret = 0; |
1da177e4 LT |
5058 | |
5059 | rq = task_rq_lock(p, &flags); | |
5060 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5061 | ret = -EINVAL; | |
5062 | goto out; | |
5063 | } | |
5064 | ||
5065 | p->cpus_allowed = new_mask; | |
5066 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5067 | if (cpu_isset(task_cpu(p), new_mask)) | |
5068 | goto out; | |
5069 | ||
5070 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5071 | /* Need help from migration thread: drop lock and wait. */ | |
5072 | task_rq_unlock(rq, &flags); | |
5073 | wake_up_process(rq->migration_thread); | |
5074 | wait_for_completion(&req.done); | |
5075 | tlb_migrate_finish(p->mm); | |
5076 | return 0; | |
5077 | } | |
5078 | out: | |
5079 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5080 | |
1da177e4 LT |
5081 | return ret; |
5082 | } | |
1da177e4 LT |
5083 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5084 | ||
5085 | /* | |
41a2d6cf | 5086 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5087 | * this because either it can't run here any more (set_cpus_allowed() |
5088 | * away from this CPU, or CPU going down), or because we're | |
5089 | * attempting to rebalance this task on exec (sched_exec). | |
5090 | * | |
5091 | * So we race with normal scheduler movements, but that's OK, as long | |
5092 | * as the task is no longer on this CPU. | |
efc30814 KK |
5093 | * |
5094 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5095 | */ |
efc30814 | 5096 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5097 | { |
70b97a7f | 5098 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5099 | int ret = 0, on_rq; |
1da177e4 LT |
5100 | |
5101 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5102 | return ret; |
1da177e4 LT |
5103 | |
5104 | rq_src = cpu_rq(src_cpu); | |
5105 | rq_dest = cpu_rq(dest_cpu); | |
5106 | ||
5107 | double_rq_lock(rq_src, rq_dest); | |
5108 | /* Already moved. */ | |
5109 | if (task_cpu(p) != src_cpu) | |
5110 | goto out; | |
5111 | /* Affinity changed (again). */ | |
5112 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5113 | goto out; | |
5114 | ||
dd41f596 | 5115 | on_rq = p->se.on_rq; |
6e82a3be | 5116 | if (on_rq) |
2e1cb74a | 5117 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5118 | |
1da177e4 | 5119 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5120 | if (on_rq) { |
5121 | activate_task(rq_dest, p, 0); | |
5122 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5123 | } |
efc30814 | 5124 | ret = 1; |
1da177e4 LT |
5125 | out: |
5126 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5127 | return ret; |
1da177e4 LT |
5128 | } |
5129 | ||
5130 | /* | |
5131 | * migration_thread - this is a highprio system thread that performs | |
5132 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5133 | * another runqueue. | |
5134 | */ | |
95cdf3b7 | 5135 | static int migration_thread(void *data) |
1da177e4 | 5136 | { |
1da177e4 | 5137 | int cpu = (long)data; |
70b97a7f | 5138 | struct rq *rq; |
1da177e4 LT |
5139 | |
5140 | rq = cpu_rq(cpu); | |
5141 | BUG_ON(rq->migration_thread != current); | |
5142 | ||
5143 | set_current_state(TASK_INTERRUPTIBLE); | |
5144 | while (!kthread_should_stop()) { | |
70b97a7f | 5145 | struct migration_req *req; |
1da177e4 | 5146 | struct list_head *head; |
1da177e4 | 5147 | |
1da177e4 LT |
5148 | spin_lock_irq(&rq->lock); |
5149 | ||
5150 | if (cpu_is_offline(cpu)) { | |
5151 | spin_unlock_irq(&rq->lock); | |
5152 | goto wait_to_die; | |
5153 | } | |
5154 | ||
5155 | if (rq->active_balance) { | |
5156 | active_load_balance(rq, cpu); | |
5157 | rq->active_balance = 0; | |
5158 | } | |
5159 | ||
5160 | head = &rq->migration_queue; | |
5161 | ||
5162 | if (list_empty(head)) { | |
5163 | spin_unlock_irq(&rq->lock); | |
5164 | schedule(); | |
5165 | set_current_state(TASK_INTERRUPTIBLE); | |
5166 | continue; | |
5167 | } | |
70b97a7f | 5168 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5169 | list_del_init(head->next); |
5170 | ||
674311d5 NP |
5171 | spin_unlock(&rq->lock); |
5172 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5173 | local_irq_enable(); | |
1da177e4 LT |
5174 | |
5175 | complete(&req->done); | |
5176 | } | |
5177 | __set_current_state(TASK_RUNNING); | |
5178 | return 0; | |
5179 | ||
5180 | wait_to_die: | |
5181 | /* Wait for kthread_stop */ | |
5182 | set_current_state(TASK_INTERRUPTIBLE); | |
5183 | while (!kthread_should_stop()) { | |
5184 | schedule(); | |
5185 | set_current_state(TASK_INTERRUPTIBLE); | |
5186 | } | |
5187 | __set_current_state(TASK_RUNNING); | |
5188 | return 0; | |
5189 | } | |
5190 | ||
5191 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5192 | |
5193 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5194 | { | |
5195 | int ret; | |
5196 | ||
5197 | local_irq_disable(); | |
5198 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5199 | local_irq_enable(); | |
5200 | return ret; | |
5201 | } | |
5202 | ||
054b9108 | 5203 | /* |
3a4fa0a2 | 5204 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5205 | * NOTE: interrupts should be disabled by the caller |
5206 | */ | |
48f24c4d | 5207 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5208 | { |
efc30814 | 5209 | unsigned long flags; |
1da177e4 | 5210 | cpumask_t mask; |
70b97a7f IM |
5211 | struct rq *rq; |
5212 | int dest_cpu; | |
1da177e4 | 5213 | |
3a5c359a AK |
5214 | do { |
5215 | /* On same node? */ | |
5216 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5217 | cpus_and(mask, mask, p->cpus_allowed); | |
5218 | dest_cpu = any_online_cpu(mask); | |
5219 | ||
5220 | /* On any allowed CPU? */ | |
5221 | if (dest_cpu == NR_CPUS) | |
5222 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5223 | ||
5224 | /* No more Mr. Nice Guy. */ | |
5225 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5226 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5227 | /* | |
5228 | * Try to stay on the same cpuset, where the | |
5229 | * current cpuset may be a subset of all cpus. | |
5230 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5231 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5232 | * called within calls to cpuset_lock/cpuset_unlock. |
5233 | */ | |
3a5c359a | 5234 | rq = task_rq_lock(p, &flags); |
470fd646 | 5235 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5236 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5237 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5238 | |
3a5c359a AK |
5239 | /* |
5240 | * Don't tell them about moving exiting tasks or | |
5241 | * kernel threads (both mm NULL), since they never | |
5242 | * leave kernel. | |
5243 | */ | |
41a2d6cf | 5244 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5245 | printk(KERN_INFO "process %d (%s) no " |
5246 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5247 | task_pid_nr(p), p->comm, dead_cpu); |
5248 | } | |
3a5c359a | 5249 | } |
f7b4cddc | 5250 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5251 | } |
5252 | ||
5253 | /* | |
5254 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5255 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5256 | * for performance reasons the counter is not stricly tracking tasks to | |
5257 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5258 | * to keep the global sum constant after CPU-down: | |
5259 | */ | |
70b97a7f | 5260 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5261 | { |
70b97a7f | 5262 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5263 | unsigned long flags; |
5264 | ||
5265 | local_irq_save(flags); | |
5266 | double_rq_lock(rq_src, rq_dest); | |
5267 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5268 | rq_src->nr_uninterruptible = 0; | |
5269 | double_rq_unlock(rq_src, rq_dest); | |
5270 | local_irq_restore(flags); | |
5271 | } | |
5272 | ||
5273 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5274 | static void migrate_live_tasks(int src_cpu) | |
5275 | { | |
48f24c4d | 5276 | struct task_struct *p, *t; |
1da177e4 | 5277 | |
f7b4cddc | 5278 | read_lock(&tasklist_lock); |
1da177e4 | 5279 | |
48f24c4d IM |
5280 | do_each_thread(t, p) { |
5281 | if (p == current) | |
1da177e4 LT |
5282 | continue; |
5283 | ||
48f24c4d IM |
5284 | if (task_cpu(p) == src_cpu) |
5285 | move_task_off_dead_cpu(src_cpu, p); | |
5286 | } while_each_thread(t, p); | |
1da177e4 | 5287 | |
f7b4cddc | 5288 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5289 | } |
5290 | ||
dd41f596 IM |
5291 | /* |
5292 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5293 | * It does so by boosting its priority to highest possible. |
5294 | * Used by CPU offline code. | |
1da177e4 LT |
5295 | */ |
5296 | void sched_idle_next(void) | |
5297 | { | |
48f24c4d | 5298 | int this_cpu = smp_processor_id(); |
70b97a7f | 5299 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5300 | struct task_struct *p = rq->idle; |
5301 | unsigned long flags; | |
5302 | ||
5303 | /* cpu has to be offline */ | |
48f24c4d | 5304 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5305 | |
48f24c4d IM |
5306 | /* |
5307 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5308 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5309 | */ |
5310 | spin_lock_irqsave(&rq->lock, flags); | |
5311 | ||
dd41f596 | 5312 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5313 | |
94bc9a7b DA |
5314 | update_rq_clock(rq); |
5315 | activate_task(rq, p, 0); | |
1da177e4 LT |
5316 | |
5317 | spin_unlock_irqrestore(&rq->lock, flags); | |
5318 | } | |
5319 | ||
48f24c4d IM |
5320 | /* |
5321 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5322 | * offline. |
5323 | */ | |
5324 | void idle_task_exit(void) | |
5325 | { | |
5326 | struct mm_struct *mm = current->active_mm; | |
5327 | ||
5328 | BUG_ON(cpu_online(smp_processor_id())); | |
5329 | ||
5330 | if (mm != &init_mm) | |
5331 | switch_mm(mm, &init_mm, current); | |
5332 | mmdrop(mm); | |
5333 | } | |
5334 | ||
054b9108 | 5335 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5336 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5337 | { |
70b97a7f | 5338 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5339 | |
5340 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5341 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5342 | |
5343 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5344 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5345 | |
48f24c4d | 5346 | get_task_struct(p); |
1da177e4 LT |
5347 | |
5348 | /* | |
5349 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5350 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5351 | * fine. |
5352 | */ | |
f7b4cddc | 5353 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5354 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5355 | spin_lock_irq(&rq->lock); |
1da177e4 | 5356 | |
48f24c4d | 5357 | put_task_struct(p); |
1da177e4 LT |
5358 | } |
5359 | ||
5360 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5361 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5362 | { | |
70b97a7f | 5363 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5364 | struct task_struct *next; |
48f24c4d | 5365 | |
dd41f596 IM |
5366 | for ( ; ; ) { |
5367 | if (!rq->nr_running) | |
5368 | break; | |
a8e504d2 | 5369 | update_rq_clock(rq); |
ff95f3df | 5370 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5371 | if (!next) |
5372 | break; | |
5373 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5374 | |
1da177e4 LT |
5375 | } |
5376 | } | |
5377 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5378 | ||
e692ab53 NP |
5379 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5380 | ||
5381 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5382 | { |
5383 | .procname = "sched_domain", | |
c57baf1e | 5384 | .mode = 0555, |
e0361851 | 5385 | }, |
38605cae | 5386 | {0, }, |
e692ab53 NP |
5387 | }; |
5388 | ||
5389 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5390 | { |
c57baf1e | 5391 | .ctl_name = CTL_KERN, |
e0361851 | 5392 | .procname = "kernel", |
c57baf1e | 5393 | .mode = 0555, |
e0361851 AD |
5394 | .child = sd_ctl_dir, |
5395 | }, | |
38605cae | 5396 | {0, }, |
e692ab53 NP |
5397 | }; |
5398 | ||
5399 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5400 | { | |
5401 | struct ctl_table *entry = | |
5cf9f062 | 5402 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5403 | |
e692ab53 NP |
5404 | return entry; |
5405 | } | |
5406 | ||
6382bc90 MM |
5407 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5408 | { | |
cd790076 | 5409 | struct ctl_table *entry; |
6382bc90 | 5410 | |
cd790076 MM |
5411 | /* |
5412 | * In the intermediate directories, both the child directory and | |
5413 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5414 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5415 | * static strings and all have proc handlers. |
5416 | */ | |
5417 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5418 | if (entry->child) |
5419 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5420 | if (entry->proc_handler == NULL) |
5421 | kfree(entry->procname); | |
5422 | } | |
6382bc90 MM |
5423 | |
5424 | kfree(*tablep); | |
5425 | *tablep = NULL; | |
5426 | } | |
5427 | ||
e692ab53 | 5428 | static void |
e0361851 | 5429 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5430 | const char *procname, void *data, int maxlen, |
5431 | mode_t mode, proc_handler *proc_handler) | |
5432 | { | |
e692ab53 NP |
5433 | entry->procname = procname; |
5434 | entry->data = data; | |
5435 | entry->maxlen = maxlen; | |
5436 | entry->mode = mode; | |
5437 | entry->proc_handler = proc_handler; | |
5438 | } | |
5439 | ||
5440 | static struct ctl_table * | |
5441 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5442 | { | |
ace8b3d6 | 5443 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5444 | |
ad1cdc1d MM |
5445 | if (table == NULL) |
5446 | return NULL; | |
5447 | ||
e0361851 | 5448 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5449 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5450 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5451 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5452 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5453 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5454 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5455 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5456 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5457 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5458 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5459 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5460 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5461 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5462 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5463 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5464 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5465 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5466 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5467 | &sd->cache_nice_tries, |
5468 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5469 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5470 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5471 | /* &table[11] is terminator */ |
e692ab53 NP |
5472 | |
5473 | return table; | |
5474 | } | |
5475 | ||
9a4e7159 | 5476 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5477 | { |
5478 | struct ctl_table *entry, *table; | |
5479 | struct sched_domain *sd; | |
5480 | int domain_num = 0, i; | |
5481 | char buf[32]; | |
5482 | ||
5483 | for_each_domain(cpu, sd) | |
5484 | domain_num++; | |
5485 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5486 | if (table == NULL) |
5487 | return NULL; | |
e692ab53 NP |
5488 | |
5489 | i = 0; | |
5490 | for_each_domain(cpu, sd) { | |
5491 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5492 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5493 | entry->mode = 0555; |
e692ab53 NP |
5494 | entry->child = sd_alloc_ctl_domain_table(sd); |
5495 | entry++; | |
5496 | i++; | |
5497 | } | |
5498 | return table; | |
5499 | } | |
5500 | ||
5501 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5502 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5503 | { |
5504 | int i, cpu_num = num_online_cpus(); | |
5505 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5506 | char buf[32]; | |
5507 | ||
7378547f MM |
5508 | WARN_ON(sd_ctl_dir[0].child); |
5509 | sd_ctl_dir[0].child = entry; | |
5510 | ||
ad1cdc1d MM |
5511 | if (entry == NULL) |
5512 | return; | |
5513 | ||
97b6ea7b | 5514 | for_each_online_cpu(i) { |
e692ab53 | 5515 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5516 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5517 | entry->mode = 0555; |
e692ab53 | 5518 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5519 | entry++; |
e692ab53 | 5520 | } |
7378547f MM |
5521 | |
5522 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5523 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5524 | } | |
6382bc90 | 5525 | |
7378547f | 5526 | /* may be called multiple times per register */ |
6382bc90 MM |
5527 | static void unregister_sched_domain_sysctl(void) |
5528 | { | |
7378547f MM |
5529 | if (sd_sysctl_header) |
5530 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5531 | sd_sysctl_header = NULL; |
7378547f MM |
5532 | if (sd_ctl_dir[0].child) |
5533 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5534 | } |
e692ab53 | 5535 | #else |
6382bc90 MM |
5536 | static void register_sched_domain_sysctl(void) |
5537 | { | |
5538 | } | |
5539 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5540 | { |
5541 | } | |
5542 | #endif | |
5543 | ||
1da177e4 LT |
5544 | /* |
5545 | * migration_call - callback that gets triggered when a CPU is added. | |
5546 | * Here we can start up the necessary migration thread for the new CPU. | |
5547 | */ | |
48f24c4d IM |
5548 | static int __cpuinit |
5549 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5550 | { |
1da177e4 | 5551 | struct task_struct *p; |
48f24c4d | 5552 | int cpu = (long)hcpu; |
1da177e4 | 5553 | unsigned long flags; |
70b97a7f | 5554 | struct rq *rq; |
1da177e4 LT |
5555 | |
5556 | switch (action) { | |
5be9361c GS |
5557 | case CPU_LOCK_ACQUIRE: |
5558 | mutex_lock(&sched_hotcpu_mutex); | |
5559 | break; | |
5560 | ||
1da177e4 | 5561 | case CPU_UP_PREPARE: |
8bb78442 | 5562 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5563 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5564 | if (IS_ERR(p)) |
5565 | return NOTIFY_BAD; | |
1da177e4 LT |
5566 | kthread_bind(p, cpu); |
5567 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5568 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5569 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5570 | task_rq_unlock(rq, &flags); |
5571 | cpu_rq(cpu)->migration_thread = p; | |
5572 | break; | |
48f24c4d | 5573 | |
1da177e4 | 5574 | case CPU_ONLINE: |
8bb78442 | 5575 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5576 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 LT |
5577 | wake_up_process(cpu_rq(cpu)->migration_thread); |
5578 | break; | |
48f24c4d | 5579 | |
1da177e4 LT |
5580 | #ifdef CONFIG_HOTPLUG_CPU |
5581 | case CPU_UP_CANCELED: | |
8bb78442 | 5582 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5583 | if (!cpu_rq(cpu)->migration_thread) |
5584 | break; | |
41a2d6cf | 5585 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5586 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5587 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5588 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5589 | cpu_rq(cpu)->migration_thread = NULL; | |
5590 | break; | |
48f24c4d | 5591 | |
1da177e4 | 5592 | case CPU_DEAD: |
8bb78442 | 5593 | case CPU_DEAD_FROZEN: |
470fd646 | 5594 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5595 | migrate_live_tasks(cpu); |
5596 | rq = cpu_rq(cpu); | |
5597 | kthread_stop(rq->migration_thread); | |
5598 | rq->migration_thread = NULL; | |
5599 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5600 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5601 | update_rq_clock(rq); |
2e1cb74a | 5602 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5603 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5604 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5605 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5606 | migrate_dead_tasks(cpu); |
d2da272a | 5607 | spin_unlock_irq(&rq->lock); |
470fd646 | 5608 | cpuset_unlock(); |
1da177e4 LT |
5609 | migrate_nr_uninterruptible(rq); |
5610 | BUG_ON(rq->nr_running != 0); | |
5611 | ||
41a2d6cf IM |
5612 | /* |
5613 | * No need to migrate the tasks: it was best-effort if | |
5614 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5615 | * the requestors. | |
5616 | */ | |
1da177e4 LT |
5617 | spin_lock_irq(&rq->lock); |
5618 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5619 | struct migration_req *req; |
5620 | ||
1da177e4 | 5621 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5622 | struct migration_req, list); |
1da177e4 LT |
5623 | list_del_init(&req->list); |
5624 | complete(&req->done); | |
5625 | } | |
5626 | spin_unlock_irq(&rq->lock); | |
5627 | break; | |
5628 | #endif | |
5be9361c GS |
5629 | case CPU_LOCK_RELEASE: |
5630 | mutex_unlock(&sched_hotcpu_mutex); | |
5631 | break; | |
1da177e4 LT |
5632 | } |
5633 | return NOTIFY_OK; | |
5634 | } | |
5635 | ||
5636 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5637 | * happens before everything else. | |
5638 | */ | |
26c2143b | 5639 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5640 | .notifier_call = migration_call, |
5641 | .priority = 10 | |
5642 | }; | |
5643 | ||
e6fe6649 | 5644 | void __init migration_init(void) |
1da177e4 LT |
5645 | { |
5646 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5647 | int err; |
48f24c4d IM |
5648 | |
5649 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5650 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5651 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5652 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5653 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5654 | } |
5655 | #endif | |
5656 | ||
5657 | #ifdef CONFIG_SMP | |
476f3534 CL |
5658 | |
5659 | /* Number of possible processor ids */ | |
5660 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5661 | EXPORT_SYMBOL(nr_cpu_ids); | |
5662 | ||
3e9830dc | 5663 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5664 | |
5665 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5666 | { |
4dcf6aff IM |
5667 | struct sched_group *group = sd->groups; |
5668 | cpumask_t groupmask; | |
5669 | char str[NR_CPUS]; | |
1da177e4 | 5670 | |
4dcf6aff IM |
5671 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5672 | cpus_clear(groupmask); | |
5673 | ||
5674 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5675 | ||
5676 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5677 | printk("does not load-balance\n"); | |
5678 | if (sd->parent) | |
5679 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5680 | " has parent"); | |
5681 | return -1; | |
41c7ce9a NP |
5682 | } |
5683 | ||
4dcf6aff IM |
5684 | printk(KERN_CONT "span %s\n", str); |
5685 | ||
5686 | if (!cpu_isset(cpu, sd->span)) { | |
5687 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5688 | "CPU%d\n", cpu); | |
5689 | } | |
5690 | if (!cpu_isset(cpu, group->cpumask)) { | |
5691 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5692 | " CPU%d\n", cpu); | |
5693 | } | |
1da177e4 | 5694 | |
4dcf6aff | 5695 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5696 | do { |
4dcf6aff IM |
5697 | if (!group) { |
5698 | printk("\n"); | |
5699 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5700 | break; |
5701 | } | |
5702 | ||
4dcf6aff IM |
5703 | if (!group->__cpu_power) { |
5704 | printk(KERN_CONT "\n"); | |
5705 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5706 | "set\n"); | |
5707 | break; | |
5708 | } | |
1da177e4 | 5709 | |
4dcf6aff IM |
5710 | if (!cpus_weight(group->cpumask)) { |
5711 | printk(KERN_CONT "\n"); | |
5712 | printk(KERN_ERR "ERROR: empty group\n"); | |
5713 | break; | |
5714 | } | |
1da177e4 | 5715 | |
4dcf6aff IM |
5716 | if (cpus_intersects(groupmask, group->cpumask)) { |
5717 | printk(KERN_CONT "\n"); | |
5718 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5719 | break; | |
5720 | } | |
1da177e4 | 5721 | |
4dcf6aff | 5722 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5723 | |
4dcf6aff IM |
5724 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5725 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5726 | |
4dcf6aff IM |
5727 | group = group->next; |
5728 | } while (group != sd->groups); | |
5729 | printk(KERN_CONT "\n"); | |
1da177e4 | 5730 | |
4dcf6aff IM |
5731 | if (!cpus_equal(sd->span, groupmask)) |
5732 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5733 | |
4dcf6aff IM |
5734 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5735 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5736 | "of domain->span\n"); | |
5737 | return 0; | |
5738 | } | |
1da177e4 | 5739 | |
4dcf6aff IM |
5740 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5741 | { | |
5742 | int level = 0; | |
1da177e4 | 5743 | |
4dcf6aff IM |
5744 | if (!sd) { |
5745 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5746 | return; | |
5747 | } | |
1da177e4 | 5748 | |
4dcf6aff IM |
5749 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5750 | ||
5751 | for (;;) { | |
5752 | if (sched_domain_debug_one(sd, cpu, level)) | |
5753 | break; | |
1da177e4 LT |
5754 | level++; |
5755 | sd = sd->parent; | |
33859f7f | 5756 | if (!sd) |
4dcf6aff IM |
5757 | break; |
5758 | } | |
1da177e4 LT |
5759 | } |
5760 | #else | |
48f24c4d | 5761 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5762 | #endif |
5763 | ||
1a20ff27 | 5764 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5765 | { |
5766 | if (cpus_weight(sd->span) == 1) | |
5767 | return 1; | |
5768 | ||
5769 | /* Following flags need at least 2 groups */ | |
5770 | if (sd->flags & (SD_LOAD_BALANCE | | |
5771 | SD_BALANCE_NEWIDLE | | |
5772 | SD_BALANCE_FORK | | |
89c4710e SS |
5773 | SD_BALANCE_EXEC | |
5774 | SD_SHARE_CPUPOWER | | |
5775 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5776 | if (sd->groups != sd->groups->next) |
5777 | return 0; | |
5778 | } | |
5779 | ||
5780 | /* Following flags don't use groups */ | |
5781 | if (sd->flags & (SD_WAKE_IDLE | | |
5782 | SD_WAKE_AFFINE | | |
5783 | SD_WAKE_BALANCE)) | |
5784 | return 0; | |
5785 | ||
5786 | return 1; | |
5787 | } | |
5788 | ||
48f24c4d IM |
5789 | static int |
5790 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5791 | { |
5792 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5793 | ||
5794 | if (sd_degenerate(parent)) | |
5795 | return 1; | |
5796 | ||
5797 | if (!cpus_equal(sd->span, parent->span)) | |
5798 | return 0; | |
5799 | ||
5800 | /* Does parent contain flags not in child? */ | |
5801 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5802 | if (cflags & SD_WAKE_AFFINE) | |
5803 | pflags &= ~SD_WAKE_BALANCE; | |
5804 | /* Flags needing groups don't count if only 1 group in parent */ | |
5805 | if (parent->groups == parent->groups->next) { | |
5806 | pflags &= ~(SD_LOAD_BALANCE | | |
5807 | SD_BALANCE_NEWIDLE | | |
5808 | SD_BALANCE_FORK | | |
89c4710e SS |
5809 | SD_BALANCE_EXEC | |
5810 | SD_SHARE_CPUPOWER | | |
5811 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5812 | } |
5813 | if (~cflags & pflags) | |
5814 | return 0; | |
5815 | ||
5816 | return 1; | |
5817 | } | |
5818 | ||
1da177e4 LT |
5819 | /* |
5820 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5821 | * hold the hotplug lock. | |
5822 | */ | |
9c1cfda2 | 5823 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5824 | { |
70b97a7f | 5825 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5826 | struct sched_domain *tmp; |
5827 | ||
5828 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5829 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5830 | struct sched_domain *parent = tmp->parent; | |
5831 | if (!parent) | |
5832 | break; | |
1a848870 | 5833 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5834 | tmp->parent = parent->parent; |
1a848870 SS |
5835 | if (parent->parent) |
5836 | parent->parent->child = tmp; | |
5837 | } | |
245af2c7 SS |
5838 | } |
5839 | ||
1a848870 | 5840 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5841 | sd = sd->parent; |
1a848870 SS |
5842 | if (sd) |
5843 | sd->child = NULL; | |
5844 | } | |
1da177e4 LT |
5845 | |
5846 | sched_domain_debug(sd, cpu); | |
5847 | ||
674311d5 | 5848 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5849 | } |
5850 | ||
5851 | /* cpus with isolated domains */ | |
67af63a6 | 5852 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5853 | |
5854 | /* Setup the mask of cpus configured for isolated domains */ | |
5855 | static int __init isolated_cpu_setup(char *str) | |
5856 | { | |
5857 | int ints[NR_CPUS], i; | |
5858 | ||
5859 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5860 | cpus_clear(cpu_isolated_map); | |
5861 | for (i = 1; i <= ints[0]; i++) | |
5862 | if (ints[i] < NR_CPUS) | |
5863 | cpu_set(ints[i], cpu_isolated_map); | |
5864 | return 1; | |
5865 | } | |
5866 | ||
8927f494 | 5867 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5868 | |
5869 | /* | |
6711cab4 SS |
5870 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5871 | * to a function which identifies what group(along with sched group) a CPU | |
5872 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5873 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5874 | * |
5875 | * init_sched_build_groups will build a circular linked list of the groups | |
5876 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5877 | * and ->cpu_power to 0. | |
5878 | */ | |
a616058b | 5879 | static void |
6711cab4 SS |
5880 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5881 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5882 | struct sched_group **sg)) | |
1da177e4 LT |
5883 | { |
5884 | struct sched_group *first = NULL, *last = NULL; | |
5885 | cpumask_t covered = CPU_MASK_NONE; | |
5886 | int i; | |
5887 | ||
5888 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5889 | struct sched_group *sg; |
5890 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5891 | int j; |
5892 | ||
5893 | if (cpu_isset(i, covered)) | |
5894 | continue; | |
5895 | ||
5896 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5897 | sg->__cpu_power = 0; |
1da177e4 LT |
5898 | |
5899 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5900 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5901 | continue; |
5902 | ||
5903 | cpu_set(j, covered); | |
5904 | cpu_set(j, sg->cpumask); | |
5905 | } | |
5906 | if (!first) | |
5907 | first = sg; | |
5908 | if (last) | |
5909 | last->next = sg; | |
5910 | last = sg; | |
5911 | } | |
5912 | last->next = first; | |
5913 | } | |
5914 | ||
9c1cfda2 | 5915 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5916 | |
9c1cfda2 | 5917 | #ifdef CONFIG_NUMA |
198e2f18 | 5918 | |
9c1cfda2 JH |
5919 | /** |
5920 | * find_next_best_node - find the next node to include in a sched_domain | |
5921 | * @node: node whose sched_domain we're building | |
5922 | * @used_nodes: nodes already in the sched_domain | |
5923 | * | |
41a2d6cf | 5924 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5925 | * finds the closest node not already in the @used_nodes map. |
5926 | * | |
5927 | * Should use nodemask_t. | |
5928 | */ | |
5929 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5930 | { | |
5931 | int i, n, val, min_val, best_node = 0; | |
5932 | ||
5933 | min_val = INT_MAX; | |
5934 | ||
5935 | for (i = 0; i < MAX_NUMNODES; i++) { | |
5936 | /* Start at @node */ | |
5937 | n = (node + i) % MAX_NUMNODES; | |
5938 | ||
5939 | if (!nr_cpus_node(n)) | |
5940 | continue; | |
5941 | ||
5942 | /* Skip already used nodes */ | |
5943 | if (test_bit(n, used_nodes)) | |
5944 | continue; | |
5945 | ||
5946 | /* Simple min distance search */ | |
5947 | val = node_distance(node, n); | |
5948 | ||
5949 | if (val < min_val) { | |
5950 | min_val = val; | |
5951 | best_node = n; | |
5952 | } | |
5953 | } | |
5954 | ||
5955 | set_bit(best_node, used_nodes); | |
5956 | return best_node; | |
5957 | } | |
5958 | ||
5959 | /** | |
5960 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
5961 | * @node: node whose cpumask we're constructing | |
5962 | * @size: number of nodes to include in this span | |
5963 | * | |
41a2d6cf | 5964 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
5965 | * should be one that prevents unnecessary balancing, but also spreads tasks |
5966 | * out optimally. | |
5967 | */ | |
5968 | static cpumask_t sched_domain_node_span(int node) | |
5969 | { | |
9c1cfda2 | 5970 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
5971 | cpumask_t span, nodemask; |
5972 | int i; | |
9c1cfda2 JH |
5973 | |
5974 | cpus_clear(span); | |
5975 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
5976 | ||
5977 | nodemask = node_to_cpumask(node); | |
5978 | cpus_or(span, span, nodemask); | |
5979 | set_bit(node, used_nodes); | |
5980 | ||
5981 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
5982 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 5983 | |
9c1cfda2 JH |
5984 | nodemask = node_to_cpumask(next_node); |
5985 | cpus_or(span, span, nodemask); | |
5986 | } | |
5987 | ||
5988 | return span; | |
5989 | } | |
5990 | #endif | |
5991 | ||
5c45bf27 | 5992 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 5993 | |
9c1cfda2 | 5994 | /* |
48f24c4d | 5995 | * SMT sched-domains: |
9c1cfda2 | 5996 | */ |
1da177e4 LT |
5997 | #ifdef CONFIG_SCHED_SMT |
5998 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 5999 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6000 | |
41a2d6cf IM |
6001 | static int |
6002 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6003 | { |
6711cab4 SS |
6004 | if (sg) |
6005 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6006 | return cpu; |
6007 | } | |
6008 | #endif | |
6009 | ||
48f24c4d IM |
6010 | /* |
6011 | * multi-core sched-domains: | |
6012 | */ | |
1e9f28fa SS |
6013 | #ifdef CONFIG_SCHED_MC |
6014 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6015 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6016 | #endif |
6017 | ||
6018 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6019 | static int |
6020 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6021 | { |
6711cab4 | 6022 | int group; |
d5a7430d | 6023 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6024 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6025 | group = first_cpu(mask); |
6026 | if (sg) | |
6027 | *sg = &per_cpu(sched_group_core, group); | |
6028 | return group; | |
1e9f28fa SS |
6029 | } |
6030 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6031 | static int |
6032 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6033 | { |
6711cab4 SS |
6034 | if (sg) |
6035 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6036 | return cpu; |
6037 | } | |
6038 | #endif | |
6039 | ||
1da177e4 | 6040 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6041 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6042 | |
41a2d6cf IM |
6043 | static int |
6044 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6045 | { |
6711cab4 | 6046 | int group; |
48f24c4d | 6047 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6048 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6049 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6050 | group = first_cpu(mask); |
1e9f28fa | 6051 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6052 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6053 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6054 | group = first_cpu(mask); |
1da177e4 | 6055 | #else |
6711cab4 | 6056 | group = cpu; |
1da177e4 | 6057 | #endif |
6711cab4 SS |
6058 | if (sg) |
6059 | *sg = &per_cpu(sched_group_phys, group); | |
6060 | return group; | |
1da177e4 LT |
6061 | } |
6062 | ||
6063 | #ifdef CONFIG_NUMA | |
1da177e4 | 6064 | /* |
9c1cfda2 JH |
6065 | * The init_sched_build_groups can't handle what we want to do with node |
6066 | * groups, so roll our own. Now each node has its own list of groups which | |
6067 | * gets dynamically allocated. | |
1da177e4 | 6068 | */ |
9c1cfda2 | 6069 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6070 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6071 | |
9c1cfda2 | 6072 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6073 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6074 | |
6711cab4 SS |
6075 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6076 | struct sched_group **sg) | |
9c1cfda2 | 6077 | { |
6711cab4 SS |
6078 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6079 | int group; | |
6080 | ||
6081 | cpus_and(nodemask, nodemask, *cpu_map); | |
6082 | group = first_cpu(nodemask); | |
6083 | ||
6084 | if (sg) | |
6085 | *sg = &per_cpu(sched_group_allnodes, group); | |
6086 | return group; | |
1da177e4 | 6087 | } |
6711cab4 | 6088 | |
08069033 SS |
6089 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6090 | { | |
6091 | struct sched_group *sg = group_head; | |
6092 | int j; | |
6093 | ||
6094 | if (!sg) | |
6095 | return; | |
3a5c359a AK |
6096 | do { |
6097 | for_each_cpu_mask(j, sg->cpumask) { | |
6098 | struct sched_domain *sd; | |
08069033 | 6099 | |
3a5c359a AK |
6100 | sd = &per_cpu(phys_domains, j); |
6101 | if (j != first_cpu(sd->groups->cpumask)) { | |
6102 | /* | |
6103 | * Only add "power" once for each | |
6104 | * physical package. | |
6105 | */ | |
6106 | continue; | |
6107 | } | |
08069033 | 6108 | |
3a5c359a AK |
6109 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6110 | } | |
6111 | sg = sg->next; | |
6112 | } while (sg != group_head); | |
08069033 | 6113 | } |
1da177e4 LT |
6114 | #endif |
6115 | ||
a616058b | 6116 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6117 | /* Free memory allocated for various sched_group structures */ |
6118 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6119 | { | |
a616058b | 6120 | int cpu, i; |
51888ca2 SV |
6121 | |
6122 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6123 | struct sched_group **sched_group_nodes |
6124 | = sched_group_nodes_bycpu[cpu]; | |
6125 | ||
51888ca2 SV |
6126 | if (!sched_group_nodes) |
6127 | continue; | |
6128 | ||
6129 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6130 | cpumask_t nodemask = node_to_cpumask(i); | |
6131 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6132 | ||
6133 | cpus_and(nodemask, nodemask, *cpu_map); | |
6134 | if (cpus_empty(nodemask)) | |
6135 | continue; | |
6136 | ||
6137 | if (sg == NULL) | |
6138 | continue; | |
6139 | sg = sg->next; | |
6140 | next_sg: | |
6141 | oldsg = sg; | |
6142 | sg = sg->next; | |
6143 | kfree(oldsg); | |
6144 | if (oldsg != sched_group_nodes[i]) | |
6145 | goto next_sg; | |
6146 | } | |
6147 | kfree(sched_group_nodes); | |
6148 | sched_group_nodes_bycpu[cpu] = NULL; | |
6149 | } | |
51888ca2 | 6150 | } |
a616058b SS |
6151 | #else |
6152 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6153 | { | |
6154 | } | |
6155 | #endif | |
51888ca2 | 6156 | |
89c4710e SS |
6157 | /* |
6158 | * Initialize sched groups cpu_power. | |
6159 | * | |
6160 | * cpu_power indicates the capacity of sched group, which is used while | |
6161 | * distributing the load between different sched groups in a sched domain. | |
6162 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6163 | * there are asymmetries in the topology. If there are asymmetries, group | |
6164 | * having more cpu_power will pickup more load compared to the group having | |
6165 | * less cpu_power. | |
6166 | * | |
6167 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6168 | * the maximum number of tasks a group can handle in the presence of other idle | |
6169 | * or lightly loaded groups in the same sched domain. | |
6170 | */ | |
6171 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6172 | { | |
6173 | struct sched_domain *child; | |
6174 | struct sched_group *group; | |
6175 | ||
6176 | WARN_ON(!sd || !sd->groups); | |
6177 | ||
6178 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6179 | return; | |
6180 | ||
6181 | child = sd->child; | |
6182 | ||
5517d86b ED |
6183 | sd->groups->__cpu_power = 0; |
6184 | ||
89c4710e SS |
6185 | /* |
6186 | * For perf policy, if the groups in child domain share resources | |
6187 | * (for example cores sharing some portions of the cache hierarchy | |
6188 | * or SMT), then set this domain groups cpu_power such that each group | |
6189 | * can handle only one task, when there are other idle groups in the | |
6190 | * same sched domain. | |
6191 | */ | |
6192 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6193 | (child->flags & | |
6194 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6195 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6196 | return; |
6197 | } | |
6198 | ||
89c4710e SS |
6199 | /* |
6200 | * add cpu_power of each child group to this groups cpu_power | |
6201 | */ | |
6202 | group = child->groups; | |
6203 | do { | |
5517d86b | 6204 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6205 | group = group->next; |
6206 | } while (group != child->groups); | |
6207 | } | |
6208 | ||
1da177e4 | 6209 | /* |
1a20ff27 DG |
6210 | * Build sched domains for a given set of cpus and attach the sched domains |
6211 | * to the individual cpus | |
1da177e4 | 6212 | */ |
51888ca2 | 6213 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6214 | { |
6215 | int i; | |
d1b55138 JH |
6216 | #ifdef CONFIG_NUMA |
6217 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6218 | int sd_allnodes = 0; |
d1b55138 JH |
6219 | |
6220 | /* | |
6221 | * Allocate the per-node list of sched groups | |
6222 | */ | |
5cf9f062 | 6223 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6224 | GFP_KERNEL); |
d1b55138 JH |
6225 | if (!sched_group_nodes) { |
6226 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6227 | return -ENOMEM; |
d1b55138 JH |
6228 | } |
6229 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6230 | #endif | |
1da177e4 LT |
6231 | |
6232 | /* | |
1a20ff27 | 6233 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6234 | */ |
1a20ff27 | 6235 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6236 | struct sched_domain *sd = NULL, *p; |
6237 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6238 | ||
1a20ff27 | 6239 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6240 | |
6241 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6242 | if (cpus_weight(*cpu_map) > |
6243 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6244 | sd = &per_cpu(allnodes_domains, i); |
6245 | *sd = SD_ALLNODES_INIT; | |
6246 | sd->span = *cpu_map; | |
6711cab4 | 6247 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6248 | p = sd; |
6711cab4 | 6249 | sd_allnodes = 1; |
9c1cfda2 JH |
6250 | } else |
6251 | p = NULL; | |
6252 | ||
1da177e4 | 6253 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6254 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6255 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6256 | sd->parent = p; | |
1a848870 SS |
6257 | if (p) |
6258 | p->child = sd; | |
9c1cfda2 | 6259 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6260 | #endif |
6261 | ||
6262 | p = sd; | |
6263 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6264 | *sd = SD_CPU_INIT; |
6265 | sd->span = nodemask; | |
6266 | sd->parent = p; | |
1a848870 SS |
6267 | if (p) |
6268 | p->child = sd; | |
6711cab4 | 6269 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6270 | |
1e9f28fa SS |
6271 | #ifdef CONFIG_SCHED_MC |
6272 | p = sd; | |
6273 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6274 | *sd = SD_MC_INIT; |
6275 | sd->span = cpu_coregroup_map(i); | |
6276 | cpus_and(sd->span, sd->span, *cpu_map); | |
6277 | sd->parent = p; | |
1a848870 | 6278 | p->child = sd; |
6711cab4 | 6279 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6280 | #endif |
6281 | ||
1da177e4 LT |
6282 | #ifdef CONFIG_SCHED_SMT |
6283 | p = sd; | |
6284 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6285 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6286 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6287 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6288 | sd->parent = p; |
1a848870 | 6289 | p->child = sd; |
6711cab4 | 6290 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6291 | #endif |
6292 | } | |
6293 | ||
6294 | #ifdef CONFIG_SCHED_SMT | |
6295 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6296 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6297 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6298 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6299 | if (i != first_cpu(this_sibling_map)) |
6300 | continue; | |
6301 | ||
dd41f596 IM |
6302 | init_sched_build_groups(this_sibling_map, cpu_map, |
6303 | &cpu_to_cpu_group); | |
1da177e4 LT |
6304 | } |
6305 | #endif | |
6306 | ||
1e9f28fa SS |
6307 | #ifdef CONFIG_SCHED_MC |
6308 | /* Set up multi-core groups */ | |
6309 | for_each_cpu_mask(i, *cpu_map) { | |
6310 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6311 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6312 | if (i != first_cpu(this_core_map)) | |
6313 | continue; | |
dd41f596 IM |
6314 | init_sched_build_groups(this_core_map, cpu_map, |
6315 | &cpu_to_core_group); | |
1e9f28fa SS |
6316 | } |
6317 | #endif | |
6318 | ||
1da177e4 LT |
6319 | /* Set up physical groups */ |
6320 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6321 | cpumask_t nodemask = node_to_cpumask(i); | |
6322 | ||
1a20ff27 | 6323 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6324 | if (cpus_empty(nodemask)) |
6325 | continue; | |
6326 | ||
6711cab4 | 6327 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6328 | } |
6329 | ||
6330 | #ifdef CONFIG_NUMA | |
6331 | /* Set up node groups */ | |
6711cab4 | 6332 | if (sd_allnodes) |
dd41f596 IM |
6333 | init_sched_build_groups(*cpu_map, cpu_map, |
6334 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6335 | |
6336 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6337 | /* Set up node groups */ | |
6338 | struct sched_group *sg, *prev; | |
6339 | cpumask_t nodemask = node_to_cpumask(i); | |
6340 | cpumask_t domainspan; | |
6341 | cpumask_t covered = CPU_MASK_NONE; | |
6342 | int j; | |
6343 | ||
6344 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6345 | if (cpus_empty(nodemask)) { |
6346 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6347 | continue; |
d1b55138 | 6348 | } |
9c1cfda2 JH |
6349 | |
6350 | domainspan = sched_domain_node_span(i); | |
6351 | cpus_and(domainspan, domainspan, *cpu_map); | |
6352 | ||
15f0b676 | 6353 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6354 | if (!sg) { |
6355 | printk(KERN_WARNING "Can not alloc domain group for " | |
6356 | "node %d\n", i); | |
6357 | goto error; | |
6358 | } | |
9c1cfda2 JH |
6359 | sched_group_nodes[i] = sg; |
6360 | for_each_cpu_mask(j, nodemask) { | |
6361 | struct sched_domain *sd; | |
9761eea8 | 6362 | |
9c1cfda2 JH |
6363 | sd = &per_cpu(node_domains, j); |
6364 | sd->groups = sg; | |
9c1cfda2 | 6365 | } |
5517d86b | 6366 | sg->__cpu_power = 0; |
9c1cfda2 | 6367 | sg->cpumask = nodemask; |
51888ca2 | 6368 | sg->next = sg; |
9c1cfda2 JH |
6369 | cpus_or(covered, covered, nodemask); |
6370 | prev = sg; | |
6371 | ||
6372 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6373 | cpumask_t tmp, notcovered; | |
6374 | int n = (i + j) % MAX_NUMNODES; | |
6375 | ||
6376 | cpus_complement(notcovered, covered); | |
6377 | cpus_and(tmp, notcovered, *cpu_map); | |
6378 | cpus_and(tmp, tmp, domainspan); | |
6379 | if (cpus_empty(tmp)) | |
6380 | break; | |
6381 | ||
6382 | nodemask = node_to_cpumask(n); | |
6383 | cpus_and(tmp, tmp, nodemask); | |
6384 | if (cpus_empty(tmp)) | |
6385 | continue; | |
6386 | ||
15f0b676 SV |
6387 | sg = kmalloc_node(sizeof(struct sched_group), |
6388 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6389 | if (!sg) { |
6390 | printk(KERN_WARNING | |
6391 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6392 | goto error; |
9c1cfda2 | 6393 | } |
5517d86b | 6394 | sg->__cpu_power = 0; |
9c1cfda2 | 6395 | sg->cpumask = tmp; |
51888ca2 | 6396 | sg->next = prev->next; |
9c1cfda2 JH |
6397 | cpus_or(covered, covered, tmp); |
6398 | prev->next = sg; | |
6399 | prev = sg; | |
6400 | } | |
9c1cfda2 | 6401 | } |
1da177e4 LT |
6402 | #endif |
6403 | ||
6404 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6405 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6406 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6407 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6408 | ||
89c4710e | 6409 | init_sched_groups_power(i, sd); |
5c45bf27 | 6410 | } |
1da177e4 | 6411 | #endif |
1e9f28fa | 6412 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6413 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6414 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6415 | ||
89c4710e | 6416 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6417 | } |
6418 | #endif | |
1e9f28fa | 6419 | |
5c45bf27 | 6420 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6421 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6422 | ||
89c4710e | 6423 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6424 | } |
6425 | ||
9c1cfda2 | 6426 | #ifdef CONFIG_NUMA |
08069033 SS |
6427 | for (i = 0; i < MAX_NUMNODES; i++) |
6428 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6429 | |
6711cab4 SS |
6430 | if (sd_allnodes) { |
6431 | struct sched_group *sg; | |
f712c0c7 | 6432 | |
6711cab4 | 6433 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6434 | init_numa_sched_groups_power(sg); |
6435 | } | |
9c1cfda2 JH |
6436 | #endif |
6437 | ||
1da177e4 | 6438 | /* Attach the domains */ |
1a20ff27 | 6439 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6440 | struct sched_domain *sd; |
6441 | #ifdef CONFIG_SCHED_SMT | |
6442 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6443 | #elif defined(CONFIG_SCHED_MC) |
6444 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6445 | #else |
6446 | sd = &per_cpu(phys_domains, i); | |
6447 | #endif | |
6448 | cpu_attach_domain(sd, i); | |
6449 | } | |
51888ca2 SV |
6450 | |
6451 | return 0; | |
6452 | ||
a616058b | 6453 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6454 | error: |
6455 | free_sched_groups(cpu_map); | |
6456 | return -ENOMEM; | |
a616058b | 6457 | #endif |
1da177e4 | 6458 | } |
029190c5 PJ |
6459 | |
6460 | static cpumask_t *doms_cur; /* current sched domains */ | |
6461 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6462 | ||
6463 | /* | |
6464 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6465 | * cpumask_t) fails, then fallback to a single sched domain, | |
6466 | * as determined by the single cpumask_t fallback_doms. | |
6467 | */ | |
6468 | static cpumask_t fallback_doms; | |
6469 | ||
1a20ff27 | 6470 | /* |
41a2d6cf | 6471 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6472 | * For now this just excludes isolated cpus, but could be used to |
6473 | * exclude other special cases in the future. | |
1a20ff27 | 6474 | */ |
51888ca2 | 6475 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6476 | { |
7378547f MM |
6477 | int err; |
6478 | ||
029190c5 PJ |
6479 | ndoms_cur = 1; |
6480 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6481 | if (!doms_cur) | |
6482 | doms_cur = &fallback_doms; | |
6483 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6484 | err = build_sched_domains(doms_cur); |
6382bc90 | 6485 | register_sched_domain_sysctl(); |
7378547f MM |
6486 | |
6487 | return err; | |
1a20ff27 DG |
6488 | } |
6489 | ||
6490 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6491 | { |
51888ca2 | 6492 | free_sched_groups(cpu_map); |
9c1cfda2 | 6493 | } |
1da177e4 | 6494 | |
1a20ff27 DG |
6495 | /* |
6496 | * Detach sched domains from a group of cpus specified in cpu_map | |
6497 | * These cpus will now be attached to the NULL domain | |
6498 | */ | |
858119e1 | 6499 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6500 | { |
6501 | int i; | |
6502 | ||
6382bc90 MM |
6503 | unregister_sched_domain_sysctl(); |
6504 | ||
1a20ff27 DG |
6505 | for_each_cpu_mask(i, *cpu_map) |
6506 | cpu_attach_domain(NULL, i); | |
6507 | synchronize_sched(); | |
6508 | arch_destroy_sched_domains(cpu_map); | |
6509 | } | |
6510 | ||
029190c5 PJ |
6511 | /* |
6512 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6513 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6514 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6515 | * It destroys each deleted domain and builds each new domain. | |
6516 | * | |
6517 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6518 | * The masks don't intersect (don't overlap.) We should setup one |
6519 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6520 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6521 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6522 | * it as it is. | |
6523 | * | |
41a2d6cf IM |
6524 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6525 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6526 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6527 | * and partition_sched_domains() will fallback to the single partition | |
6528 | * 'fallback_doms'. | |
6529 | * | |
6530 | * Call with hotplug lock held | |
6531 | */ | |
6532 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6533 | { | |
6534 | int i, j; | |
6535 | ||
7378547f MM |
6536 | /* always unregister in case we don't destroy any domains */ |
6537 | unregister_sched_domain_sysctl(); | |
6538 | ||
029190c5 PJ |
6539 | if (doms_new == NULL) { |
6540 | ndoms_new = 1; | |
6541 | doms_new = &fallback_doms; | |
6542 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6543 | } | |
6544 | ||
6545 | /* Destroy deleted domains */ | |
6546 | for (i = 0; i < ndoms_cur; i++) { | |
6547 | for (j = 0; j < ndoms_new; j++) { | |
6548 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6549 | goto match1; | |
6550 | } | |
6551 | /* no match - a current sched domain not in new doms_new[] */ | |
6552 | detach_destroy_domains(doms_cur + i); | |
6553 | match1: | |
6554 | ; | |
6555 | } | |
6556 | ||
6557 | /* Build new domains */ | |
6558 | for (i = 0; i < ndoms_new; i++) { | |
6559 | for (j = 0; j < ndoms_cur; j++) { | |
6560 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6561 | goto match2; | |
6562 | } | |
6563 | /* no match - add a new doms_new */ | |
6564 | build_sched_domains(doms_new + i); | |
6565 | match2: | |
6566 | ; | |
6567 | } | |
6568 | ||
6569 | /* Remember the new sched domains */ | |
6570 | if (doms_cur != &fallback_doms) | |
6571 | kfree(doms_cur); | |
6572 | doms_cur = doms_new; | |
6573 | ndoms_cur = ndoms_new; | |
7378547f MM |
6574 | |
6575 | register_sched_domain_sysctl(); | |
029190c5 PJ |
6576 | } |
6577 | ||
5c45bf27 | 6578 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6579 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6580 | { |
6581 | int err; | |
6582 | ||
5be9361c | 6583 | mutex_lock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6584 | detach_destroy_domains(&cpu_online_map); |
6585 | err = arch_init_sched_domains(&cpu_online_map); | |
5be9361c | 6586 | mutex_unlock(&sched_hotcpu_mutex); |
5c45bf27 SS |
6587 | |
6588 | return err; | |
6589 | } | |
6590 | ||
6591 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6592 | { | |
6593 | int ret; | |
6594 | ||
6595 | if (buf[0] != '0' && buf[0] != '1') | |
6596 | return -EINVAL; | |
6597 | ||
6598 | if (smt) | |
6599 | sched_smt_power_savings = (buf[0] == '1'); | |
6600 | else | |
6601 | sched_mc_power_savings = (buf[0] == '1'); | |
6602 | ||
6603 | ret = arch_reinit_sched_domains(); | |
6604 | ||
6605 | return ret ? ret : count; | |
6606 | } | |
6607 | ||
5c45bf27 SS |
6608 | #ifdef CONFIG_SCHED_MC |
6609 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6610 | { | |
6611 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6612 | } | |
48f24c4d IM |
6613 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6614 | const char *buf, size_t count) | |
5c45bf27 SS |
6615 | { |
6616 | return sched_power_savings_store(buf, count, 0); | |
6617 | } | |
6707de00 AB |
6618 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6619 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6620 | #endif |
6621 | ||
6622 | #ifdef CONFIG_SCHED_SMT | |
6623 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6624 | { | |
6625 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6626 | } | |
48f24c4d IM |
6627 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6628 | const char *buf, size_t count) | |
5c45bf27 SS |
6629 | { |
6630 | return sched_power_savings_store(buf, count, 1); | |
6631 | } | |
6707de00 AB |
6632 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6633 | sched_smt_power_savings_store); | |
6634 | #endif | |
6635 | ||
6636 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6637 | { | |
6638 | int err = 0; | |
6639 | ||
6640 | #ifdef CONFIG_SCHED_SMT | |
6641 | if (smt_capable()) | |
6642 | err = sysfs_create_file(&cls->kset.kobj, | |
6643 | &attr_sched_smt_power_savings.attr); | |
6644 | #endif | |
6645 | #ifdef CONFIG_SCHED_MC | |
6646 | if (!err && mc_capable()) | |
6647 | err = sysfs_create_file(&cls->kset.kobj, | |
6648 | &attr_sched_mc_power_savings.attr); | |
6649 | #endif | |
6650 | return err; | |
6651 | } | |
5c45bf27 SS |
6652 | #endif |
6653 | ||
1da177e4 | 6654 | /* |
41a2d6cf | 6655 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6656 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6657 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6658 | * which will prevent rebalancing while the sched domains are recalculated. |
6659 | */ | |
6660 | static int update_sched_domains(struct notifier_block *nfb, | |
6661 | unsigned long action, void *hcpu) | |
6662 | { | |
1da177e4 LT |
6663 | switch (action) { |
6664 | case CPU_UP_PREPARE: | |
8bb78442 | 6665 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6666 | case CPU_DOWN_PREPARE: |
8bb78442 | 6667 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6668 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6669 | return NOTIFY_OK; |
6670 | ||
6671 | case CPU_UP_CANCELED: | |
8bb78442 | 6672 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6673 | case CPU_DOWN_FAILED: |
8bb78442 | 6674 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6675 | case CPU_ONLINE: |
8bb78442 | 6676 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6677 | case CPU_DEAD: |
8bb78442 | 6678 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6679 | /* |
6680 | * Fall through and re-initialise the domains. | |
6681 | */ | |
6682 | break; | |
6683 | default: | |
6684 | return NOTIFY_DONE; | |
6685 | } | |
6686 | ||
6687 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6688 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6689 | |
6690 | return NOTIFY_OK; | |
6691 | } | |
1da177e4 LT |
6692 | |
6693 | void __init sched_init_smp(void) | |
6694 | { | |
5c1e1767 NP |
6695 | cpumask_t non_isolated_cpus; |
6696 | ||
5be9361c | 6697 | mutex_lock(&sched_hotcpu_mutex); |
1a20ff27 | 6698 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6699 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6700 | if (cpus_empty(non_isolated_cpus)) |
6701 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
5be9361c | 6702 | mutex_unlock(&sched_hotcpu_mutex); |
1da177e4 LT |
6703 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6704 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6705 | |
6706 | /* Move init over to a non-isolated CPU */ | |
6707 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6708 | BUG(); | |
19978ca6 | 6709 | sched_init_granularity(); |
1da177e4 LT |
6710 | } |
6711 | #else | |
6712 | void __init sched_init_smp(void) | |
6713 | { | |
19978ca6 | 6714 | sched_init_granularity(); |
1da177e4 LT |
6715 | } |
6716 | #endif /* CONFIG_SMP */ | |
6717 | ||
6718 | int in_sched_functions(unsigned long addr) | |
6719 | { | |
1da177e4 LT |
6720 | return in_lock_functions(addr) || |
6721 | (addr >= (unsigned long)__sched_text_start | |
6722 | && addr < (unsigned long)__sched_text_end); | |
6723 | } | |
6724 | ||
a9957449 | 6725 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6726 | { |
6727 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6728 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6729 | cfs_rq->rq = rq; | |
6730 | #endif | |
67e9fb2a | 6731 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6732 | } |
6733 | ||
1da177e4 LT |
6734 | void __init sched_init(void) |
6735 | { | |
476f3534 | 6736 | int highest_cpu = 0; |
dd41f596 IM |
6737 | int i, j; |
6738 | ||
0a945022 | 6739 | for_each_possible_cpu(i) { |
dd41f596 | 6740 | struct rt_prio_array *array; |
70b97a7f | 6741 | struct rq *rq; |
1da177e4 LT |
6742 | |
6743 | rq = cpu_rq(i); | |
6744 | spin_lock_init(&rq->lock); | |
fcb99371 | 6745 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6746 | rq->nr_running = 0; |
dd41f596 IM |
6747 | rq->clock = 1; |
6748 | init_cfs_rq(&rq->cfs, rq); | |
6749 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6750 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6751 | { |
6752 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6753 | struct sched_entity *se = | |
6754 | &per_cpu(init_sched_entity, i); | |
6755 | ||
6756 | init_cfs_rq_p[i] = cfs_rq; | |
6757 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6758 | cfs_rq->tg = &init_task_group; |
3a252015 | 6759 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6760 | &rq->leaf_cfs_rq_list); |
6761 | ||
3a252015 IM |
6762 | init_sched_entity_p[i] = se; |
6763 | se->cfs_rq = &rq->cfs; | |
6764 | se->my_q = cfs_rq; | |
4cf86d77 | 6765 | se->load.weight = init_task_group_load; |
9b5b7751 | 6766 | se->load.inv_weight = |
4cf86d77 | 6767 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6768 | se->parent = NULL; |
6769 | } | |
4cf86d77 | 6770 | init_task_group.shares = init_task_group_load; |
5cb350ba | 6771 | spin_lock_init(&init_task_group.lock); |
dd41f596 | 6772 | #endif |
1da177e4 | 6773 | |
dd41f596 IM |
6774 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6775 | rq->cpu_load[j] = 0; | |
1da177e4 | 6776 | #ifdef CONFIG_SMP |
41c7ce9a | 6777 | rq->sd = NULL; |
1da177e4 | 6778 | rq->active_balance = 0; |
dd41f596 | 6779 | rq->next_balance = jiffies; |
1da177e4 | 6780 | rq->push_cpu = 0; |
0a2966b4 | 6781 | rq->cpu = i; |
1da177e4 LT |
6782 | rq->migration_thread = NULL; |
6783 | INIT_LIST_HEAD(&rq->migration_queue); | |
6784 | #endif | |
6785 | atomic_set(&rq->nr_iowait, 0); | |
6786 | ||
dd41f596 IM |
6787 | array = &rq->rt.active; |
6788 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6789 | INIT_LIST_HEAD(array->queue + j); | |
6790 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6791 | } |
476f3534 | 6792 | highest_cpu = i; |
dd41f596 IM |
6793 | /* delimiter for bitsearch: */ |
6794 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6795 | } |
6796 | ||
2dd73a4f | 6797 | set_load_weight(&init_task); |
b50f60ce | 6798 | |
e107be36 AK |
6799 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6800 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6801 | #endif | |
6802 | ||
c9819f45 | 6803 | #ifdef CONFIG_SMP |
476f3534 | 6804 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6805 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6806 | #endif | |
6807 | ||
b50f60ce HC |
6808 | #ifdef CONFIG_RT_MUTEXES |
6809 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6810 | #endif | |
6811 | ||
1da177e4 LT |
6812 | /* |
6813 | * The boot idle thread does lazy MMU switching as well: | |
6814 | */ | |
6815 | atomic_inc(&init_mm.mm_count); | |
6816 | enter_lazy_tlb(&init_mm, current); | |
6817 | ||
6818 | /* | |
6819 | * Make us the idle thread. Technically, schedule() should not be | |
6820 | * called from this thread, however somewhere below it might be, | |
6821 | * but because we are the idle thread, we just pick up running again | |
6822 | * when this runqueue becomes "idle". | |
6823 | */ | |
6824 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6825 | /* |
6826 | * During early bootup we pretend to be a normal task: | |
6827 | */ | |
6828 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6829 | } |
6830 | ||
6831 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6832 | void __might_sleep(char *file, int line) | |
6833 | { | |
48f24c4d | 6834 | #ifdef in_atomic |
1da177e4 LT |
6835 | static unsigned long prev_jiffy; /* ratelimiting */ |
6836 | ||
6837 | if ((in_atomic() || irqs_disabled()) && | |
6838 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6839 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6840 | return; | |
6841 | prev_jiffy = jiffies; | |
91368d73 | 6842 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6843 | " context at %s:%d\n", file, line); |
6844 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6845 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6846 | debug_show_held_locks(current); |
3117df04 IM |
6847 | if (irqs_disabled()) |
6848 | print_irqtrace_events(current); | |
1da177e4 LT |
6849 | dump_stack(); |
6850 | } | |
6851 | #endif | |
6852 | } | |
6853 | EXPORT_SYMBOL(__might_sleep); | |
6854 | #endif | |
6855 | ||
6856 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6857 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6858 | { | |
6859 | int on_rq; | |
6860 | update_rq_clock(rq); | |
6861 | on_rq = p->se.on_rq; | |
6862 | if (on_rq) | |
6863 | deactivate_task(rq, p, 0); | |
6864 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6865 | if (on_rq) { | |
6866 | activate_task(rq, p, 0); | |
6867 | resched_task(rq->curr); | |
6868 | } | |
6869 | } | |
6870 | ||
1da177e4 LT |
6871 | void normalize_rt_tasks(void) |
6872 | { | |
a0f98a1c | 6873 | struct task_struct *g, *p; |
1da177e4 | 6874 | unsigned long flags; |
70b97a7f | 6875 | struct rq *rq; |
1da177e4 LT |
6876 | |
6877 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6878 | do_each_thread(g, p) { |
178be793 IM |
6879 | /* |
6880 | * Only normalize user tasks: | |
6881 | */ | |
6882 | if (!p->mm) | |
6883 | continue; | |
6884 | ||
6cfb0d5d | 6885 | p->se.exec_start = 0; |
6cfb0d5d | 6886 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6887 | p->se.wait_start = 0; |
dd41f596 | 6888 | p->se.sleep_start = 0; |
dd41f596 | 6889 | p->se.block_start = 0; |
6cfb0d5d | 6890 | #endif |
dd41f596 IM |
6891 | task_rq(p)->clock = 0; |
6892 | ||
6893 | if (!rt_task(p)) { | |
6894 | /* | |
6895 | * Renice negative nice level userspace | |
6896 | * tasks back to 0: | |
6897 | */ | |
6898 | if (TASK_NICE(p) < 0 && p->mm) | |
6899 | set_user_nice(p, 0); | |
1da177e4 | 6900 | continue; |
dd41f596 | 6901 | } |
1da177e4 | 6902 | |
b29739f9 IM |
6903 | spin_lock_irqsave(&p->pi_lock, flags); |
6904 | rq = __task_rq_lock(p); | |
1da177e4 | 6905 | |
178be793 | 6906 | normalize_task(rq, p); |
3a5e4dc1 | 6907 | |
b29739f9 IM |
6908 | __task_rq_unlock(rq); |
6909 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6910 | } while_each_thread(g, p); |
6911 | ||
1da177e4 LT |
6912 | read_unlock_irq(&tasklist_lock); |
6913 | } | |
6914 | ||
6915 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6916 | |
6917 | #ifdef CONFIG_IA64 | |
6918 | /* | |
6919 | * These functions are only useful for the IA64 MCA handling. | |
6920 | * | |
6921 | * They can only be called when the whole system has been | |
6922 | * stopped - every CPU needs to be quiescent, and no scheduling | |
6923 | * activity can take place. Using them for anything else would | |
6924 | * be a serious bug, and as a result, they aren't even visible | |
6925 | * under any other configuration. | |
6926 | */ | |
6927 | ||
6928 | /** | |
6929 | * curr_task - return the current task for a given cpu. | |
6930 | * @cpu: the processor in question. | |
6931 | * | |
6932 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6933 | */ | |
36c8b586 | 6934 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
6935 | { |
6936 | return cpu_curr(cpu); | |
6937 | } | |
6938 | ||
6939 | /** | |
6940 | * set_curr_task - set the current task for a given cpu. | |
6941 | * @cpu: the processor in question. | |
6942 | * @p: the task pointer to set. | |
6943 | * | |
6944 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
6945 | * are serviced on a separate stack. It allows the architecture to switch the |
6946 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
6947 | * must be called with all CPU's synchronized, and interrupts disabled, the |
6948 | * and caller must save the original value of the current task (see | |
6949 | * curr_task() above) and restore that value before reenabling interrupts and | |
6950 | * re-starting the system. | |
6951 | * | |
6952 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
6953 | */ | |
36c8b586 | 6954 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
6955 | { |
6956 | cpu_curr(cpu) = p; | |
6957 | } | |
6958 | ||
6959 | #endif | |
29f59db3 SV |
6960 | |
6961 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6962 | ||
29f59db3 | 6963 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 6964 | struct task_group *sched_create_group(void) |
29f59db3 | 6965 | { |
4cf86d77 | 6966 | struct task_group *tg; |
29f59db3 SV |
6967 | struct cfs_rq *cfs_rq; |
6968 | struct sched_entity *se; | |
9b5b7751 | 6969 | struct rq *rq; |
29f59db3 SV |
6970 | int i; |
6971 | ||
29f59db3 SV |
6972 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
6973 | if (!tg) | |
6974 | return ERR_PTR(-ENOMEM); | |
6975 | ||
9b5b7751 | 6976 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6977 | if (!tg->cfs_rq) |
6978 | goto err; | |
9b5b7751 | 6979 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
6980 | if (!tg->se) |
6981 | goto err; | |
6982 | ||
6983 | for_each_possible_cpu(i) { | |
9b5b7751 | 6984 | rq = cpu_rq(i); |
29f59db3 SV |
6985 | |
6986 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
6987 | cpu_to_node(i)); | |
6988 | if (!cfs_rq) | |
6989 | goto err; | |
6990 | ||
6991 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
6992 | cpu_to_node(i)); | |
6993 | if (!se) | |
6994 | goto err; | |
6995 | ||
6996 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
6997 | memset(se, 0, sizeof(struct sched_entity)); | |
6998 | ||
6999 | tg->cfs_rq[i] = cfs_rq; | |
7000 | init_cfs_rq(cfs_rq, rq); | |
7001 | cfs_rq->tg = tg; | |
29f59db3 SV |
7002 | |
7003 | tg->se[i] = se; | |
7004 | se->cfs_rq = &rq->cfs; | |
7005 | se->my_q = cfs_rq; | |
7006 | se->load.weight = NICE_0_LOAD; | |
7007 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7008 | se->parent = NULL; | |
7009 | } | |
7010 | ||
9b5b7751 SV |
7011 | for_each_possible_cpu(i) { |
7012 | rq = cpu_rq(i); | |
7013 | cfs_rq = tg->cfs_rq[i]; | |
7014 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7015 | } | |
29f59db3 | 7016 | |
9b5b7751 | 7017 | tg->shares = NICE_0_LOAD; |
5cb350ba | 7018 | spin_lock_init(&tg->lock); |
29f59db3 | 7019 | |
9b5b7751 | 7020 | return tg; |
29f59db3 SV |
7021 | |
7022 | err: | |
7023 | for_each_possible_cpu(i) { | |
a65914b3 | 7024 | if (tg->cfs_rq) |
29f59db3 | 7025 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7026 | if (tg->se) |
29f59db3 SV |
7027 | kfree(tg->se[i]); |
7028 | } | |
a65914b3 IM |
7029 | kfree(tg->cfs_rq); |
7030 | kfree(tg->se); | |
7031 | kfree(tg); | |
29f59db3 SV |
7032 | |
7033 | return ERR_PTR(-ENOMEM); | |
7034 | } | |
7035 | ||
9b5b7751 SV |
7036 | /* rcu callback to free various structures associated with a task group */ |
7037 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7038 | { |
ae8393e5 SV |
7039 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7040 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7041 | struct sched_entity *se; |
7042 | int i; | |
7043 | ||
29f59db3 SV |
7044 | /* now it should be safe to free those cfs_rqs */ |
7045 | for_each_possible_cpu(i) { | |
7046 | cfs_rq = tg->cfs_rq[i]; | |
7047 | kfree(cfs_rq); | |
7048 | ||
7049 | se = tg->se[i]; | |
7050 | kfree(se); | |
7051 | } | |
7052 | ||
7053 | kfree(tg->cfs_rq); | |
7054 | kfree(tg->se); | |
7055 | kfree(tg); | |
7056 | } | |
7057 | ||
9b5b7751 | 7058 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7059 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7060 | { |
7bae49d4 | 7061 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7062 | int i; |
29f59db3 | 7063 | |
9b5b7751 SV |
7064 | for_each_possible_cpu(i) { |
7065 | cfs_rq = tg->cfs_rq[i]; | |
7066 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7067 | } | |
7068 | ||
7bae49d4 | 7069 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7070 | |
7071 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7072 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7073 | } |
7074 | ||
9b5b7751 | 7075 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7076 | * The caller of this function should have put the task in its new group |
7077 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7078 | * reflect its new group. | |
9b5b7751 SV |
7079 | */ |
7080 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7081 | { |
7082 | int on_rq, running; | |
7083 | unsigned long flags; | |
7084 | struct rq *rq; | |
7085 | ||
7086 | rq = task_rq_lock(tsk, &flags); | |
7087 | ||
dae51f56 | 7088 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7089 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7090 | goto done; |
dae51f56 | 7091 | } |
29f59db3 SV |
7092 | |
7093 | update_rq_clock(rq); | |
7094 | ||
051a1d1a | 7095 | running = task_current(rq, tsk); |
29f59db3 SV |
7096 | on_rq = tsk->se.on_rq; |
7097 | ||
83b699ed | 7098 | if (on_rq) { |
29f59db3 | 7099 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7100 | if (unlikely(running)) |
7101 | tsk->sched_class->put_prev_task(rq, tsk); | |
7102 | } | |
29f59db3 | 7103 | |
ce96b5ac | 7104 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7105 | |
83b699ed SV |
7106 | if (on_rq) { |
7107 | if (unlikely(running)) | |
7108 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7109 | enqueue_task(rq, tsk, 0); |
83b699ed | 7110 | } |
29f59db3 SV |
7111 | |
7112 | done: | |
7113 | task_rq_unlock(rq, &flags); | |
7114 | } | |
7115 | ||
7116 | static void set_se_shares(struct sched_entity *se, unsigned long shares) | |
7117 | { | |
7118 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7119 | struct rq *rq = cfs_rq->rq; | |
7120 | int on_rq; | |
7121 | ||
7122 | spin_lock_irq(&rq->lock); | |
7123 | ||
7124 | on_rq = se->on_rq; | |
7125 | if (on_rq) | |
7126 | dequeue_entity(cfs_rq, se, 0); | |
7127 | ||
7128 | se->load.weight = shares; | |
7129 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7130 | ||
7131 | if (on_rq) | |
7132 | enqueue_entity(cfs_rq, se, 0); | |
7133 | ||
7134 | spin_unlock_irq(&rq->lock); | |
7135 | } | |
7136 | ||
4cf86d77 | 7137 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7138 | { |
7139 | int i; | |
29f59db3 | 7140 | |
c61935fd IM |
7141 | /* |
7142 | * A weight of 0 or 1 can cause arithmetics problems. | |
7143 | * (The default weight is 1024 - so there's no practical | |
7144 | * limitation from this.) | |
7145 | */ | |
7146 | if (shares < 2) | |
7147 | shares = 2; | |
7148 | ||
5cb350ba | 7149 | spin_lock(&tg->lock); |
9b5b7751 | 7150 | if (tg->shares == shares) |
5cb350ba | 7151 | goto done; |
29f59db3 | 7152 | |
9b5b7751 | 7153 | tg->shares = shares; |
29f59db3 | 7154 | for_each_possible_cpu(i) |
9b5b7751 | 7155 | set_se_shares(tg->se[i], shares); |
29f59db3 | 7156 | |
5cb350ba DG |
7157 | done: |
7158 | spin_unlock(&tg->lock); | |
9b5b7751 | 7159 | return 0; |
29f59db3 SV |
7160 | } |
7161 | ||
5cb350ba DG |
7162 | unsigned long sched_group_shares(struct task_group *tg) |
7163 | { | |
7164 | return tg->shares; | |
7165 | } | |
7166 | ||
3a252015 | 7167 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7168 | |
7169 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7170 | ||
7171 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7172 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7173 | { |
2b01dfe3 PM |
7174 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7175 | struct task_group, css); | |
68318b8e SV |
7176 | } |
7177 | ||
7178 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7179 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7180 | { |
7181 | struct task_group *tg; | |
7182 | ||
2b01dfe3 | 7183 | if (!cgrp->parent) { |
68318b8e | 7184 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7185 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7186 | return &init_task_group.css; |
7187 | } | |
7188 | ||
7189 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7190 | if (cgrp->parent->parent) |
68318b8e SV |
7191 | return ERR_PTR(-EINVAL); |
7192 | ||
7193 | tg = sched_create_group(); | |
7194 | if (IS_ERR(tg)) | |
7195 | return ERR_PTR(-ENOMEM); | |
7196 | ||
7197 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7198 | tg->css.cgroup = cgrp; |
68318b8e SV |
7199 | |
7200 | return &tg->css; | |
7201 | } | |
7202 | ||
41a2d6cf IM |
7203 | static void |
7204 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7205 | { |
2b01dfe3 | 7206 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7207 | |
7208 | sched_destroy_group(tg); | |
7209 | } | |
7210 | ||
41a2d6cf IM |
7211 | static int |
7212 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7213 | struct task_struct *tsk) | |
68318b8e SV |
7214 | { |
7215 | /* We don't support RT-tasks being in separate groups */ | |
7216 | if (tsk->sched_class != &fair_sched_class) | |
7217 | return -EINVAL; | |
7218 | ||
7219 | return 0; | |
7220 | } | |
7221 | ||
7222 | static void | |
2b01dfe3 | 7223 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7224 | struct cgroup *old_cont, struct task_struct *tsk) |
7225 | { | |
7226 | sched_move_task(tsk); | |
7227 | } | |
7228 | ||
2b01dfe3 PM |
7229 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7230 | u64 shareval) | |
68318b8e | 7231 | { |
2b01dfe3 | 7232 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7233 | } |
7234 | ||
2b01dfe3 | 7235 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7236 | { |
2b01dfe3 | 7237 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7238 | |
7239 | return (u64) tg->shares; | |
7240 | } | |
7241 | ||
fe5c7cc2 PM |
7242 | static struct cftype cpu_files[] = { |
7243 | { | |
7244 | .name = "shares", | |
7245 | .read_uint = cpu_shares_read_uint, | |
7246 | .write_uint = cpu_shares_write_uint, | |
7247 | }, | |
68318b8e SV |
7248 | }; |
7249 | ||
7250 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7251 | { | |
fe5c7cc2 | 7252 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7253 | } |
7254 | ||
7255 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7256 | .name = "cpu", |
7257 | .create = cpu_cgroup_create, | |
7258 | .destroy = cpu_cgroup_destroy, | |
7259 | .can_attach = cpu_cgroup_can_attach, | |
7260 | .attach = cpu_cgroup_attach, | |
7261 | .populate = cpu_cgroup_populate, | |
7262 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7263 | .early_init = 1, |
7264 | }; | |
7265 | ||
7266 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7267 | |
7268 | #ifdef CONFIG_CGROUP_CPUACCT | |
7269 | ||
7270 | /* | |
7271 | * CPU accounting code for task groups. | |
7272 | * | |
7273 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7274 | * (balbir@in.ibm.com). | |
7275 | */ | |
7276 | ||
7277 | /* track cpu usage of a group of tasks */ | |
7278 | struct cpuacct { | |
7279 | struct cgroup_subsys_state css; | |
7280 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7281 | u64 *cpuusage; | |
7282 | }; | |
7283 | ||
7284 | struct cgroup_subsys cpuacct_subsys; | |
7285 | ||
7286 | /* return cpu accounting group corresponding to this container */ | |
7287 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7288 | { | |
7289 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7290 | struct cpuacct, css); | |
7291 | } | |
7292 | ||
7293 | /* return cpu accounting group to which this task belongs */ | |
7294 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7295 | { | |
7296 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7297 | struct cpuacct, css); | |
7298 | } | |
7299 | ||
7300 | /* create a new cpu accounting group */ | |
7301 | static struct cgroup_subsys_state *cpuacct_create( | |
7302 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7303 | { | |
7304 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7305 | ||
7306 | if (!ca) | |
7307 | return ERR_PTR(-ENOMEM); | |
7308 | ||
7309 | ca->cpuusage = alloc_percpu(u64); | |
7310 | if (!ca->cpuusage) { | |
7311 | kfree(ca); | |
7312 | return ERR_PTR(-ENOMEM); | |
7313 | } | |
7314 | ||
7315 | return &ca->css; | |
7316 | } | |
7317 | ||
7318 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7319 | static void |
7320 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7321 | { |
7322 | struct cpuacct *ca = cgroup_ca(cont); | |
7323 | ||
7324 | free_percpu(ca->cpuusage); | |
7325 | kfree(ca); | |
7326 | } | |
7327 | ||
7328 | /* return total cpu usage (in nanoseconds) of a group */ | |
7329 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7330 | { | |
7331 | struct cpuacct *ca = cgroup_ca(cont); | |
7332 | u64 totalcpuusage = 0; | |
7333 | int i; | |
7334 | ||
7335 | for_each_possible_cpu(i) { | |
7336 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7337 | ||
7338 | /* | |
7339 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7340 | * platforms. | |
7341 | */ | |
7342 | spin_lock_irq(&cpu_rq(i)->lock); | |
7343 | totalcpuusage += *cpuusage; | |
7344 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7345 | } | |
7346 | ||
7347 | return totalcpuusage; | |
7348 | } | |
7349 | ||
7350 | static struct cftype files[] = { | |
7351 | { | |
7352 | .name = "usage", | |
7353 | .read_uint = cpuusage_read, | |
7354 | }, | |
7355 | }; | |
7356 | ||
7357 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7358 | { | |
7359 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7360 | } | |
7361 | ||
7362 | /* | |
7363 | * charge this task's execution time to its accounting group. | |
7364 | * | |
7365 | * called with rq->lock held. | |
7366 | */ | |
7367 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7368 | { | |
7369 | struct cpuacct *ca; | |
7370 | ||
7371 | if (!cpuacct_subsys.active) | |
7372 | return; | |
7373 | ||
7374 | ca = task_ca(tsk); | |
7375 | if (ca) { | |
7376 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7377 | ||
7378 | *cpuusage += cputime; | |
7379 | } | |
7380 | } | |
7381 | ||
7382 | struct cgroup_subsys cpuacct_subsys = { | |
7383 | .name = "cpuacct", | |
7384 | .create = cpuacct_create, | |
7385 | .destroy = cpuacct_destroy, | |
7386 | .populate = cpuacct_populate, | |
7387 | .subsys_id = cpuacct_subsys_id, | |
7388 | }; | |
7389 | #endif /* CONFIG_CGROUP_CPUACCT */ |