]>
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; | |
6b2d7700 SV |
171 | |
172 | /* | |
173 | * shares assigned to a task group governs how much of cpu bandwidth | |
174 | * is allocated to the group. The more shares a group has, the more is | |
175 | * the cpu bandwidth allocated to it. | |
176 | * | |
177 | * For ex, lets say that there are three task groups, A, B and C which | |
178 | * have been assigned shares 1000, 2000 and 3000 respectively. Then, | |
179 | * cpu bandwidth allocated by the scheduler to task groups A, B and C | |
180 | * should be: | |
181 | * | |
182 | * Bw(A) = 1000/(1000+2000+3000) * 100 = 16.66% | |
183 | * Bw(B) = 2000/(1000+2000+3000) * 100 = 33.33% | |
184 | * Bw(C) = 3000/(1000+2000+3000) * 100 = 50% | |
185 | * | |
186 | * The weight assigned to a task group's schedulable entities on every | |
187 | * cpu (task_group.se[a_cpu]->load.weight) is derived from the task | |
188 | * group's shares. For ex: lets say that task group A has been | |
189 | * assigned shares of 1000 and there are two CPUs in a system. Then, | |
190 | * | |
191 | * tg_A->se[0]->load.weight = tg_A->se[1]->load.weight = 1000; | |
192 | * | |
193 | * Note: It's not necessary that each of a task's group schedulable | |
194 | * entity have the same weight on all CPUs. If the group | |
195 | * has 2 of its tasks on CPU0 and 1 task on CPU1, then a | |
196 | * better distribution of weight could be: | |
197 | * | |
198 | * tg_A->se[0]->load.weight = 2/3 * 2000 = 1333 | |
199 | * tg_A->se[1]->load.weight = 1/2 * 2000 = 667 | |
200 | * | |
201 | * rebalance_shares() is responsible for distributing the shares of a | |
202 | * task groups like this among the group's schedulable entities across | |
203 | * cpus. | |
204 | * | |
205 | */ | |
29f59db3 | 206 | unsigned long shares; |
6b2d7700 | 207 | |
ae8393e5 | 208 | struct rcu_head rcu; |
29f59db3 SV |
209 | }; |
210 | ||
211 | /* Default task group's sched entity on each cpu */ | |
212 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
213 | /* Default task group's cfs_rq on each cpu */ | |
214 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
215 | ||
9b5b7751 SV |
216 | static struct sched_entity *init_sched_entity_p[NR_CPUS]; |
217 | static struct cfs_rq *init_cfs_rq_p[NR_CPUS]; | |
29f59db3 | 218 | |
ec2c507f SV |
219 | /* task_group_mutex serializes add/remove of task groups and also changes to |
220 | * a task group's cpu shares. | |
221 | */ | |
222 | static DEFINE_MUTEX(task_group_mutex); | |
223 | ||
a1835615 SV |
224 | /* doms_cur_mutex serializes access to doms_cur[] array */ |
225 | static DEFINE_MUTEX(doms_cur_mutex); | |
226 | ||
6b2d7700 SV |
227 | #ifdef CONFIG_SMP |
228 | /* kernel thread that runs rebalance_shares() periodically */ | |
229 | static struct task_struct *lb_monitor_task; | |
230 | static int load_balance_monitor(void *unused); | |
231 | #endif | |
232 | ||
233 | static void set_se_shares(struct sched_entity *se, unsigned long shares); | |
234 | ||
29f59db3 | 235 | /* Default task group. |
3a252015 | 236 | * Every task in system belong to this group at bootup. |
29f59db3 | 237 | */ |
4cf86d77 | 238 | struct task_group init_task_group = { |
3a252015 IM |
239 | .se = init_sched_entity_p, |
240 | .cfs_rq = init_cfs_rq_p, | |
241 | }; | |
9b5b7751 | 242 | |
24e377a8 | 243 | #ifdef CONFIG_FAIR_USER_SCHED |
93f992cc | 244 | # define INIT_TASK_GROUP_LOAD 2*NICE_0_LOAD |
24e377a8 | 245 | #else |
93f992cc | 246 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
24e377a8 SV |
247 | #endif |
248 | ||
6b2d7700 SV |
249 | #define MIN_GROUP_SHARES 2 |
250 | ||
93f992cc | 251 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
29f59db3 SV |
252 | |
253 | /* return group to which a task belongs */ | |
4cf86d77 | 254 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 255 | { |
4cf86d77 | 256 | struct task_group *tg; |
9b5b7751 | 257 | |
24e377a8 SV |
258 | #ifdef CONFIG_FAIR_USER_SCHED |
259 | tg = p->user->tg; | |
68318b8e SV |
260 | #elif defined(CONFIG_FAIR_CGROUP_SCHED) |
261 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), | |
262 | struct task_group, css); | |
24e377a8 | 263 | #else |
41a2d6cf | 264 | tg = &init_task_group; |
24e377a8 | 265 | #endif |
9b5b7751 | 266 | return tg; |
29f59db3 SV |
267 | } |
268 | ||
269 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
ce96b5ac | 270 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 271 | { |
ce96b5ac DA |
272 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
273 | p->se.parent = task_group(p)->se[cpu]; | |
29f59db3 SV |
274 | } |
275 | ||
ec2c507f SV |
276 | static inline void lock_task_group_list(void) |
277 | { | |
278 | mutex_lock(&task_group_mutex); | |
279 | } | |
280 | ||
281 | static inline void unlock_task_group_list(void) | |
282 | { | |
283 | mutex_unlock(&task_group_mutex); | |
284 | } | |
285 | ||
a1835615 SV |
286 | static inline void lock_doms_cur(void) |
287 | { | |
288 | mutex_lock(&doms_cur_mutex); | |
289 | } | |
290 | ||
291 | static inline void unlock_doms_cur(void) | |
292 | { | |
293 | mutex_unlock(&doms_cur_mutex); | |
294 | } | |
295 | ||
29f59db3 SV |
296 | #else |
297 | ||
ce96b5ac | 298 | static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { } |
ec2c507f SV |
299 | static inline void lock_task_group_list(void) { } |
300 | static inline void unlock_task_group_list(void) { } | |
a1835615 SV |
301 | static inline void lock_doms_cur(void) { } |
302 | static inline void unlock_doms_cur(void) { } | |
29f59db3 SV |
303 | |
304 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
305 | ||
6aa645ea IM |
306 | /* CFS-related fields in a runqueue */ |
307 | struct cfs_rq { | |
308 | struct load_weight load; | |
309 | unsigned long nr_running; | |
310 | ||
6aa645ea | 311 | u64 exec_clock; |
e9acbff6 | 312 | u64 min_vruntime; |
6aa645ea IM |
313 | |
314 | struct rb_root tasks_timeline; | |
315 | struct rb_node *rb_leftmost; | |
316 | struct rb_node *rb_load_balance_curr; | |
6aa645ea IM |
317 | /* 'curr' points to currently running entity on this cfs_rq. |
318 | * It is set to NULL otherwise (i.e when none are currently running). | |
319 | */ | |
320 | struct sched_entity *curr; | |
ddc97297 PZ |
321 | |
322 | unsigned long nr_spread_over; | |
323 | ||
62160e3f | 324 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
325 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
326 | ||
41a2d6cf IM |
327 | /* |
328 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
329 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
330 | * (like users, containers etc.) | |
331 | * | |
332 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
333 | * list is used during load balance. | |
334 | */ | |
41a2d6cf IM |
335 | struct list_head leaf_cfs_rq_list; |
336 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
337 | #endif |
338 | }; | |
1da177e4 | 339 | |
6aa645ea IM |
340 | /* Real-Time classes' related field in a runqueue: */ |
341 | struct rt_rq { | |
342 | struct rt_prio_array active; | |
343 | int rt_load_balance_idx; | |
344 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; | |
345 | }; | |
346 | ||
1da177e4 LT |
347 | /* |
348 | * This is the main, per-CPU runqueue data structure. | |
349 | * | |
350 | * Locking rule: those places that want to lock multiple runqueues | |
351 | * (such as the load balancing or the thread migration code), lock | |
352 | * acquire operations must be ordered by ascending &runqueue. | |
353 | */ | |
70b97a7f | 354 | struct rq { |
d8016491 IM |
355 | /* runqueue lock: */ |
356 | spinlock_t lock; | |
1da177e4 LT |
357 | |
358 | /* | |
359 | * nr_running and cpu_load should be in the same cacheline because | |
360 | * remote CPUs use both these fields when doing load calculation. | |
361 | */ | |
362 | unsigned long nr_running; | |
6aa645ea IM |
363 | #define CPU_LOAD_IDX_MAX 5 |
364 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 365 | unsigned char idle_at_tick; |
46cb4b7c SS |
366 | #ifdef CONFIG_NO_HZ |
367 | unsigned char in_nohz_recently; | |
368 | #endif | |
d8016491 IM |
369 | /* capture load from *all* tasks on this cpu: */ |
370 | struct load_weight load; | |
6aa645ea IM |
371 | unsigned long nr_load_updates; |
372 | u64 nr_switches; | |
373 | ||
374 | struct cfs_rq cfs; | |
375 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
d8016491 IM |
376 | /* list of leaf cfs_rq on this cpu: */ |
377 | struct list_head leaf_cfs_rq_list; | |
1da177e4 | 378 | #endif |
41a2d6cf | 379 | struct rt_rq rt; |
1da177e4 LT |
380 | |
381 | /* | |
382 | * This is part of a global counter where only the total sum | |
383 | * over all CPUs matters. A task can increase this counter on | |
384 | * one CPU and if it got migrated afterwards it may decrease | |
385 | * it on another CPU. Always updated under the runqueue lock: | |
386 | */ | |
387 | unsigned long nr_uninterruptible; | |
388 | ||
36c8b586 | 389 | struct task_struct *curr, *idle; |
c9819f45 | 390 | unsigned long next_balance; |
1da177e4 | 391 | struct mm_struct *prev_mm; |
6aa645ea | 392 | |
6aa645ea IM |
393 | u64 clock, prev_clock_raw; |
394 | s64 clock_max_delta; | |
395 | ||
396 | unsigned int clock_warps, clock_overflows; | |
2aa44d05 IM |
397 | u64 idle_clock; |
398 | unsigned int clock_deep_idle_events; | |
529c7726 | 399 | u64 tick_timestamp; |
6aa645ea | 400 | |
1da177e4 LT |
401 | atomic_t nr_iowait; |
402 | ||
403 | #ifdef CONFIG_SMP | |
404 | struct sched_domain *sd; | |
405 | ||
406 | /* For active balancing */ | |
407 | int active_balance; | |
408 | int push_cpu; | |
d8016491 IM |
409 | /* cpu of this runqueue: */ |
410 | int cpu; | |
1da177e4 | 411 | |
36c8b586 | 412 | struct task_struct *migration_thread; |
1da177e4 LT |
413 | struct list_head migration_queue; |
414 | #endif | |
415 | ||
416 | #ifdef CONFIG_SCHEDSTATS | |
417 | /* latency stats */ | |
418 | struct sched_info rq_sched_info; | |
419 | ||
420 | /* sys_sched_yield() stats */ | |
480b9434 KC |
421 | unsigned int yld_exp_empty; |
422 | unsigned int yld_act_empty; | |
423 | unsigned int yld_both_empty; | |
424 | unsigned int yld_count; | |
1da177e4 LT |
425 | |
426 | /* schedule() stats */ | |
480b9434 KC |
427 | unsigned int sched_switch; |
428 | unsigned int sched_count; | |
429 | unsigned int sched_goidle; | |
1da177e4 LT |
430 | |
431 | /* try_to_wake_up() stats */ | |
480b9434 KC |
432 | unsigned int ttwu_count; |
433 | unsigned int ttwu_local; | |
b8efb561 IM |
434 | |
435 | /* BKL stats */ | |
480b9434 | 436 | unsigned int bkl_count; |
1da177e4 | 437 | #endif |
fcb99371 | 438 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
439 | }; |
440 | ||
f34e3b61 | 441 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 442 | |
dd41f596 IM |
443 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
444 | { | |
445 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
446 | } | |
447 | ||
0a2966b4 CL |
448 | static inline int cpu_of(struct rq *rq) |
449 | { | |
450 | #ifdef CONFIG_SMP | |
451 | return rq->cpu; | |
452 | #else | |
453 | return 0; | |
454 | #endif | |
455 | } | |
456 | ||
20d315d4 | 457 | /* |
b04a0f4c IM |
458 | * Update the per-runqueue clock, as finegrained as the platform can give |
459 | * us, but without assuming monotonicity, etc.: | |
20d315d4 | 460 | */ |
b04a0f4c | 461 | static void __update_rq_clock(struct rq *rq) |
20d315d4 IM |
462 | { |
463 | u64 prev_raw = rq->prev_clock_raw; | |
464 | u64 now = sched_clock(); | |
465 | s64 delta = now - prev_raw; | |
466 | u64 clock = rq->clock; | |
467 | ||
b04a0f4c IM |
468 | #ifdef CONFIG_SCHED_DEBUG |
469 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
470 | #endif | |
20d315d4 IM |
471 | /* |
472 | * Protect against sched_clock() occasionally going backwards: | |
473 | */ | |
474 | if (unlikely(delta < 0)) { | |
475 | clock++; | |
476 | rq->clock_warps++; | |
477 | } else { | |
478 | /* | |
479 | * Catch too large forward jumps too: | |
480 | */ | |
529c7726 IM |
481 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
482 | if (clock < rq->tick_timestamp + TICK_NSEC) | |
483 | clock = rq->tick_timestamp + TICK_NSEC; | |
484 | else | |
485 | clock++; | |
20d315d4 IM |
486 | rq->clock_overflows++; |
487 | } else { | |
488 | if (unlikely(delta > rq->clock_max_delta)) | |
489 | rq->clock_max_delta = delta; | |
490 | clock += delta; | |
491 | } | |
492 | } | |
493 | ||
494 | rq->prev_clock_raw = now; | |
495 | rq->clock = clock; | |
b04a0f4c | 496 | } |
20d315d4 | 497 | |
b04a0f4c IM |
498 | static void update_rq_clock(struct rq *rq) |
499 | { | |
500 | if (likely(smp_processor_id() == cpu_of(rq))) | |
501 | __update_rq_clock(rq); | |
20d315d4 IM |
502 | } |
503 | ||
674311d5 NP |
504 | /* |
505 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 506 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
507 | * |
508 | * The domain tree of any CPU may only be accessed from within | |
509 | * preempt-disabled sections. | |
510 | */ | |
48f24c4d IM |
511 | #define for_each_domain(cpu, __sd) \ |
512 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
513 | |
514 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
515 | #define this_rq() (&__get_cpu_var(runqueues)) | |
516 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
517 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
518 | ||
bf5c91ba IM |
519 | /* |
520 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
521 | */ | |
522 | #ifdef CONFIG_SCHED_DEBUG | |
523 | # define const_debug __read_mostly | |
524 | #else | |
525 | # define const_debug static const | |
526 | #endif | |
527 | ||
528 | /* | |
529 | * Debugging: various feature bits | |
530 | */ | |
531 | enum { | |
bbdba7c0 | 532 | SCHED_FEAT_NEW_FAIR_SLEEPERS = 1, |
9612633a IM |
533 | SCHED_FEAT_WAKEUP_PREEMPT = 2, |
534 | SCHED_FEAT_START_DEBIT = 4, | |
41a2d6cf IM |
535 | SCHED_FEAT_TREE_AVG = 8, |
536 | SCHED_FEAT_APPROX_AVG = 16, | |
bf5c91ba IM |
537 | }; |
538 | ||
539 | const_debug unsigned int sysctl_sched_features = | |
8401f775 | 540 | SCHED_FEAT_NEW_FAIR_SLEEPERS * 1 | |
9612633a | 541 | SCHED_FEAT_WAKEUP_PREEMPT * 1 | |
8401f775 IM |
542 | SCHED_FEAT_START_DEBIT * 1 | |
543 | SCHED_FEAT_TREE_AVG * 0 | | |
9612633a | 544 | SCHED_FEAT_APPROX_AVG * 0; |
bf5c91ba IM |
545 | |
546 | #define sched_feat(x) (sysctl_sched_features & SCHED_FEAT_##x) | |
547 | ||
b82d9fdd PZ |
548 | /* |
549 | * Number of tasks to iterate in a single balance run. | |
550 | * Limited because this is done with IRQs disabled. | |
551 | */ | |
552 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
553 | ||
e436d800 IM |
554 | /* |
555 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
556 | * clock constructed from sched_clock(): | |
557 | */ | |
558 | unsigned long long cpu_clock(int cpu) | |
559 | { | |
e436d800 IM |
560 | unsigned long long now; |
561 | unsigned long flags; | |
b04a0f4c | 562 | struct rq *rq; |
e436d800 | 563 | |
2cd4d0ea | 564 | local_irq_save(flags); |
b04a0f4c | 565 | rq = cpu_rq(cpu); |
8ced5f69 IM |
566 | /* |
567 | * Only call sched_clock() if the scheduler has already been | |
568 | * initialized (some code might call cpu_clock() very early): | |
569 | */ | |
570 | if (rq->idle) | |
571 | update_rq_clock(rq); | |
b04a0f4c | 572 | now = rq->clock; |
2cd4d0ea | 573 | local_irq_restore(flags); |
e436d800 IM |
574 | |
575 | return now; | |
576 | } | |
a58f6f25 | 577 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 578 | |
1da177e4 | 579 | #ifndef prepare_arch_switch |
4866cde0 NP |
580 | # define prepare_arch_switch(next) do { } while (0) |
581 | #endif | |
582 | #ifndef finish_arch_switch | |
583 | # define finish_arch_switch(prev) do { } while (0) | |
584 | #endif | |
585 | ||
051a1d1a DA |
586 | static inline int task_current(struct rq *rq, struct task_struct *p) |
587 | { | |
588 | return rq->curr == p; | |
589 | } | |
590 | ||
4866cde0 | 591 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 592 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 593 | { |
051a1d1a | 594 | return task_current(rq, p); |
4866cde0 NP |
595 | } |
596 | ||
70b97a7f | 597 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
598 | { |
599 | } | |
600 | ||
70b97a7f | 601 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 602 | { |
da04c035 IM |
603 | #ifdef CONFIG_DEBUG_SPINLOCK |
604 | /* this is a valid case when another task releases the spinlock */ | |
605 | rq->lock.owner = current; | |
606 | #endif | |
8a25d5de IM |
607 | /* |
608 | * If we are tracking spinlock dependencies then we have to | |
609 | * fix up the runqueue lock - which gets 'carried over' from | |
610 | * prev into current: | |
611 | */ | |
612 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
613 | ||
4866cde0 NP |
614 | spin_unlock_irq(&rq->lock); |
615 | } | |
616 | ||
617 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 618 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
619 | { |
620 | #ifdef CONFIG_SMP | |
621 | return p->oncpu; | |
622 | #else | |
051a1d1a | 623 | return task_current(rq, p); |
4866cde0 NP |
624 | #endif |
625 | } | |
626 | ||
70b97a7f | 627 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
628 | { |
629 | #ifdef CONFIG_SMP | |
630 | /* | |
631 | * We can optimise this out completely for !SMP, because the | |
632 | * SMP rebalancing from interrupt is the only thing that cares | |
633 | * here. | |
634 | */ | |
635 | next->oncpu = 1; | |
636 | #endif | |
637 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
638 | spin_unlock_irq(&rq->lock); | |
639 | #else | |
640 | spin_unlock(&rq->lock); | |
641 | #endif | |
642 | } | |
643 | ||
70b97a7f | 644 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
645 | { |
646 | #ifdef CONFIG_SMP | |
647 | /* | |
648 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
649 | * We must ensure this doesn't happen until the switch is completely | |
650 | * finished. | |
651 | */ | |
652 | smp_wmb(); | |
653 | prev->oncpu = 0; | |
654 | #endif | |
655 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
656 | local_irq_enable(); | |
1da177e4 | 657 | #endif |
4866cde0 NP |
658 | } |
659 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 660 | |
b29739f9 IM |
661 | /* |
662 | * __task_rq_lock - lock the runqueue a given task resides on. | |
663 | * Must be called interrupts disabled. | |
664 | */ | |
70b97a7f | 665 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
666 | __acquires(rq->lock) |
667 | { | |
3a5c359a AK |
668 | for (;;) { |
669 | struct rq *rq = task_rq(p); | |
670 | spin_lock(&rq->lock); | |
671 | if (likely(rq == task_rq(p))) | |
672 | return rq; | |
b29739f9 | 673 | spin_unlock(&rq->lock); |
b29739f9 | 674 | } |
b29739f9 IM |
675 | } |
676 | ||
1da177e4 LT |
677 | /* |
678 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 679 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
680 | * explicitly disabling preemption. |
681 | */ | |
70b97a7f | 682 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
683 | __acquires(rq->lock) |
684 | { | |
70b97a7f | 685 | struct rq *rq; |
1da177e4 | 686 | |
3a5c359a AK |
687 | for (;;) { |
688 | local_irq_save(*flags); | |
689 | rq = task_rq(p); | |
690 | spin_lock(&rq->lock); | |
691 | if (likely(rq == task_rq(p))) | |
692 | return rq; | |
1da177e4 | 693 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 694 | } |
1da177e4 LT |
695 | } |
696 | ||
a9957449 | 697 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
698 | __releases(rq->lock) |
699 | { | |
700 | spin_unlock(&rq->lock); | |
701 | } | |
702 | ||
70b97a7f | 703 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
704 | __releases(rq->lock) |
705 | { | |
706 | spin_unlock_irqrestore(&rq->lock, *flags); | |
707 | } | |
708 | ||
1da177e4 | 709 | /* |
cc2a73b5 | 710 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 711 | */ |
a9957449 | 712 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
713 | __acquires(rq->lock) |
714 | { | |
70b97a7f | 715 | struct rq *rq; |
1da177e4 LT |
716 | |
717 | local_irq_disable(); | |
718 | rq = this_rq(); | |
719 | spin_lock(&rq->lock); | |
720 | ||
721 | return rq; | |
722 | } | |
723 | ||
1b9f19c2 | 724 | /* |
2aa44d05 | 725 | * We are going deep-idle (irqs are disabled): |
1b9f19c2 | 726 | */ |
2aa44d05 | 727 | void sched_clock_idle_sleep_event(void) |
1b9f19c2 | 728 | { |
2aa44d05 IM |
729 | struct rq *rq = cpu_rq(smp_processor_id()); |
730 | ||
731 | spin_lock(&rq->lock); | |
732 | __update_rq_clock(rq); | |
733 | spin_unlock(&rq->lock); | |
734 | rq->clock_deep_idle_events++; | |
735 | } | |
736 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); | |
737 | ||
738 | /* | |
739 | * We just idled delta nanoseconds (called with irqs disabled): | |
740 | */ | |
741 | void sched_clock_idle_wakeup_event(u64 delta_ns) | |
742 | { | |
743 | struct rq *rq = cpu_rq(smp_processor_id()); | |
744 | u64 now = sched_clock(); | |
1b9f19c2 | 745 | |
2bacec8c | 746 | touch_softlockup_watchdog(); |
2aa44d05 IM |
747 | rq->idle_clock += delta_ns; |
748 | /* | |
749 | * Override the previous timestamp and ignore all | |
750 | * sched_clock() deltas that occured while we idled, | |
751 | * and use the PM-provided delta_ns to advance the | |
752 | * rq clock: | |
753 | */ | |
754 | spin_lock(&rq->lock); | |
755 | rq->prev_clock_raw = now; | |
756 | rq->clock += delta_ns; | |
757 | spin_unlock(&rq->lock); | |
1b9f19c2 | 758 | } |
2aa44d05 | 759 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
1b9f19c2 | 760 | |
c24d20db IM |
761 | /* |
762 | * resched_task - mark a task 'to be rescheduled now'. | |
763 | * | |
764 | * On UP this means the setting of the need_resched flag, on SMP it | |
765 | * might also involve a cross-CPU call to trigger the scheduler on | |
766 | * the target CPU. | |
767 | */ | |
768 | #ifdef CONFIG_SMP | |
769 | ||
770 | #ifndef tsk_is_polling | |
771 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
772 | #endif | |
773 | ||
774 | static void resched_task(struct task_struct *p) | |
775 | { | |
776 | int cpu; | |
777 | ||
778 | assert_spin_locked(&task_rq(p)->lock); | |
779 | ||
780 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) | |
781 | return; | |
782 | ||
783 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); | |
784 | ||
785 | cpu = task_cpu(p); | |
786 | if (cpu == smp_processor_id()) | |
787 | return; | |
788 | ||
789 | /* NEED_RESCHED must be visible before we test polling */ | |
790 | smp_mb(); | |
791 | if (!tsk_is_polling(p)) | |
792 | smp_send_reschedule(cpu); | |
793 | } | |
794 | ||
795 | static void resched_cpu(int cpu) | |
796 | { | |
797 | struct rq *rq = cpu_rq(cpu); | |
798 | unsigned long flags; | |
799 | ||
800 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
801 | return; | |
802 | resched_task(cpu_curr(cpu)); | |
803 | spin_unlock_irqrestore(&rq->lock, flags); | |
804 | } | |
805 | #else | |
806 | static inline void resched_task(struct task_struct *p) | |
807 | { | |
808 | assert_spin_locked(&task_rq(p)->lock); | |
809 | set_tsk_need_resched(p); | |
810 | } | |
811 | #endif | |
812 | ||
45bf76df IM |
813 | #if BITS_PER_LONG == 32 |
814 | # define WMULT_CONST (~0UL) | |
815 | #else | |
816 | # define WMULT_CONST (1UL << 32) | |
817 | #endif | |
818 | ||
819 | #define WMULT_SHIFT 32 | |
820 | ||
194081eb IM |
821 | /* |
822 | * Shift right and round: | |
823 | */ | |
cf2ab469 | 824 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 825 | |
cb1c4fc9 | 826 | static unsigned long |
45bf76df IM |
827 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
828 | struct load_weight *lw) | |
829 | { | |
830 | u64 tmp; | |
831 | ||
832 | if (unlikely(!lw->inv_weight)) | |
194081eb | 833 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
45bf76df IM |
834 | |
835 | tmp = (u64)delta_exec * weight; | |
836 | /* | |
837 | * Check whether we'd overflow the 64-bit multiplication: | |
838 | */ | |
194081eb | 839 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 840 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
841 | WMULT_SHIFT/2); |
842 | else | |
cf2ab469 | 843 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 844 | |
ecf691da | 845 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
846 | } |
847 | ||
848 | static inline unsigned long | |
849 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
850 | { | |
851 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
852 | } | |
853 | ||
1091985b | 854 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
855 | { |
856 | lw->weight += inc; | |
45bf76df IM |
857 | } |
858 | ||
1091985b | 859 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
860 | { |
861 | lw->weight -= dec; | |
45bf76df IM |
862 | } |
863 | ||
2dd73a4f PW |
864 | /* |
865 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
866 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
867 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 868 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
869 | * scaled version of the new time slice allocation that they receive on time |
870 | * slice expiry etc. | |
871 | */ | |
872 | ||
dd41f596 IM |
873 | #define WEIGHT_IDLEPRIO 2 |
874 | #define WMULT_IDLEPRIO (1 << 31) | |
875 | ||
876 | /* | |
877 | * Nice levels are multiplicative, with a gentle 10% change for every | |
878 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
879 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
880 | * that remained on nice 0. | |
881 | * | |
882 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
883 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
884 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
885 | * If a task goes up by ~10% and another task goes down by ~10% then | |
886 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
887 | */ |
888 | static const int prio_to_weight[40] = { | |
254753dc IM |
889 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
890 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
891 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
892 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
893 | /* 0 */ 1024, 820, 655, 526, 423, | |
894 | /* 5 */ 335, 272, 215, 172, 137, | |
895 | /* 10 */ 110, 87, 70, 56, 45, | |
896 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
897 | }; |
898 | ||
5714d2de IM |
899 | /* |
900 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
901 | * | |
902 | * In cases where the weight does not change often, we can use the | |
903 | * precalculated inverse to speed up arithmetics by turning divisions | |
904 | * into multiplications: | |
905 | */ | |
dd41f596 | 906 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
907 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
908 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
909 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
910 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
911 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
912 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
913 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
914 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 915 | }; |
2dd73a4f | 916 | |
dd41f596 IM |
917 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
918 | ||
919 | /* | |
920 | * runqueue iterator, to support SMP load-balancing between different | |
921 | * scheduling classes, without having to expose their internal data | |
922 | * structures to the load-balancing proper: | |
923 | */ | |
924 | struct rq_iterator { | |
925 | void *arg; | |
926 | struct task_struct *(*start)(void *); | |
927 | struct task_struct *(*next)(void *); | |
928 | }; | |
929 | ||
e1d1484f PW |
930 | #ifdef CONFIG_SMP |
931 | static unsigned long | |
932 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
933 | unsigned long max_load_move, struct sched_domain *sd, | |
934 | enum cpu_idle_type idle, int *all_pinned, | |
935 | int *this_best_prio, struct rq_iterator *iterator); | |
936 | ||
937 | static int | |
938 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
939 | struct sched_domain *sd, enum cpu_idle_type idle, | |
940 | struct rq_iterator *iterator); | |
e1d1484f | 941 | #endif |
dd41f596 | 942 | |
d842de87 SV |
943 | #ifdef CONFIG_CGROUP_CPUACCT |
944 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
945 | #else | |
946 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
947 | #endif | |
948 | ||
58e2d4ca SV |
949 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
950 | { | |
951 | update_load_add(&rq->load, load); | |
952 | } | |
953 | ||
954 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
955 | { | |
956 | update_load_sub(&rq->load, load); | |
957 | } | |
958 | ||
dd41f596 | 959 | #include "sched_stats.h" |
dd41f596 | 960 | #include "sched_idletask.c" |
5522d5d5 IM |
961 | #include "sched_fair.c" |
962 | #include "sched_rt.c" | |
dd41f596 IM |
963 | #ifdef CONFIG_SCHED_DEBUG |
964 | # include "sched_debug.c" | |
965 | #endif | |
966 | ||
967 | #define sched_class_highest (&rt_sched_class) | |
968 | ||
e5fa2237 | 969 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
970 | { |
971 | rq->nr_running++; | |
9c217245 IM |
972 | } |
973 | ||
db53181e | 974 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
975 | { |
976 | rq->nr_running--; | |
9c217245 IM |
977 | } |
978 | ||
45bf76df IM |
979 | static void set_load_weight(struct task_struct *p) |
980 | { | |
981 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
982 | p->se.load.weight = prio_to_weight[0] * 2; |
983 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
984 | return; | |
985 | } | |
45bf76df | 986 | |
dd41f596 IM |
987 | /* |
988 | * SCHED_IDLE tasks get minimal weight: | |
989 | */ | |
990 | if (p->policy == SCHED_IDLE) { | |
991 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
992 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
993 | return; | |
994 | } | |
71f8bd46 | 995 | |
dd41f596 IM |
996 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
997 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
998 | } |
999 | ||
8159f87e | 1000 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1001 | { |
dd41f596 | 1002 | sched_info_queued(p); |
fd390f6a | 1003 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1004 | p->se.on_rq = 1; |
71f8bd46 IM |
1005 | } |
1006 | ||
69be72c1 | 1007 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1008 | { |
f02231e5 | 1009 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1010 | p->se.on_rq = 0; |
71f8bd46 IM |
1011 | } |
1012 | ||
14531189 | 1013 | /* |
dd41f596 | 1014 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1015 | */ |
14531189 IM |
1016 | static inline int __normal_prio(struct task_struct *p) |
1017 | { | |
dd41f596 | 1018 | return p->static_prio; |
14531189 IM |
1019 | } |
1020 | ||
b29739f9 IM |
1021 | /* |
1022 | * Calculate the expected normal priority: i.e. priority | |
1023 | * without taking RT-inheritance into account. Might be | |
1024 | * boosted by interactivity modifiers. Changes upon fork, | |
1025 | * setprio syscalls, and whenever the interactivity | |
1026 | * estimator recalculates. | |
1027 | */ | |
36c8b586 | 1028 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1029 | { |
1030 | int prio; | |
1031 | ||
e05606d3 | 1032 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1033 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1034 | else | |
1035 | prio = __normal_prio(p); | |
1036 | return prio; | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Calculate the current priority, i.e. the priority | |
1041 | * taken into account by the scheduler. This value might | |
1042 | * be boosted by RT tasks, or might be boosted by | |
1043 | * interactivity modifiers. Will be RT if the task got | |
1044 | * RT-boosted. If not then it returns p->normal_prio. | |
1045 | */ | |
36c8b586 | 1046 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1047 | { |
1048 | p->normal_prio = normal_prio(p); | |
1049 | /* | |
1050 | * If we are RT tasks or we were boosted to RT priority, | |
1051 | * keep the priority unchanged. Otherwise, update priority | |
1052 | * to the normal priority: | |
1053 | */ | |
1054 | if (!rt_prio(p->prio)) | |
1055 | return p->normal_prio; | |
1056 | return p->prio; | |
1057 | } | |
1058 | ||
1da177e4 | 1059 | /* |
dd41f596 | 1060 | * activate_task - move a task to the runqueue. |
1da177e4 | 1061 | */ |
dd41f596 | 1062 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1063 | { |
dd41f596 IM |
1064 | if (p->state == TASK_UNINTERRUPTIBLE) |
1065 | rq->nr_uninterruptible--; | |
1da177e4 | 1066 | |
8159f87e | 1067 | enqueue_task(rq, p, wakeup); |
e5fa2237 | 1068 | inc_nr_running(p, rq); |
1da177e4 LT |
1069 | } |
1070 | ||
1da177e4 LT |
1071 | /* |
1072 | * deactivate_task - remove a task from the runqueue. | |
1073 | */ | |
2e1cb74a | 1074 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1075 | { |
dd41f596 IM |
1076 | if (p->state == TASK_UNINTERRUPTIBLE) |
1077 | rq->nr_uninterruptible++; | |
1078 | ||
69be72c1 | 1079 | dequeue_task(rq, p, sleep); |
db53181e | 1080 | dec_nr_running(p, rq); |
1da177e4 LT |
1081 | } |
1082 | ||
1da177e4 LT |
1083 | /** |
1084 | * task_curr - is this task currently executing on a CPU? | |
1085 | * @p: the task in question. | |
1086 | */ | |
36c8b586 | 1087 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1088 | { |
1089 | return cpu_curr(task_cpu(p)) == p; | |
1090 | } | |
1091 | ||
2dd73a4f PW |
1092 | /* Used instead of source_load when we know the type == 0 */ |
1093 | unsigned long weighted_cpuload(const int cpu) | |
1094 | { | |
495eca49 | 1095 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1096 | } |
1097 | ||
1098 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1099 | { | |
ce96b5ac | 1100 | set_task_cfs_rq(p, cpu); |
dd41f596 | 1101 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1102 | /* |
1103 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1104 | * successfuly executed on another CPU. We must ensure that updates of | |
1105 | * per-task data have been completed by this moment. | |
1106 | */ | |
1107 | smp_wmb(); | |
dd41f596 | 1108 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1109 | #endif |
2dd73a4f PW |
1110 | } |
1111 | ||
1da177e4 | 1112 | #ifdef CONFIG_SMP |
c65cc870 | 1113 | |
cc367732 IM |
1114 | /* |
1115 | * Is this task likely cache-hot: | |
1116 | */ | |
1117 | static inline int | |
1118 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) | |
1119 | { | |
1120 | s64 delta; | |
1121 | ||
1122 | if (p->sched_class != &fair_sched_class) | |
1123 | return 0; | |
1124 | ||
6bc1665b IM |
1125 | if (sysctl_sched_migration_cost == -1) |
1126 | return 1; | |
1127 | if (sysctl_sched_migration_cost == 0) | |
1128 | return 0; | |
1129 | ||
cc367732 IM |
1130 | delta = now - p->se.exec_start; |
1131 | ||
1132 | return delta < (s64)sysctl_sched_migration_cost; | |
1133 | } | |
1134 | ||
1135 | ||
dd41f596 | 1136 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1137 | { |
dd41f596 IM |
1138 | int old_cpu = task_cpu(p); |
1139 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1140 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1141 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1142 | u64 clock_offset; |
dd41f596 IM |
1143 | |
1144 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1145 | |
1146 | #ifdef CONFIG_SCHEDSTATS | |
1147 | if (p->se.wait_start) | |
1148 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1149 | if (p->se.sleep_start) |
1150 | p->se.sleep_start -= clock_offset; | |
1151 | if (p->se.block_start) | |
1152 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1153 | if (old_cpu != new_cpu) { |
1154 | schedstat_inc(p, se.nr_migrations); | |
1155 | if (task_hot(p, old_rq->clock, NULL)) | |
1156 | schedstat_inc(p, se.nr_forced2_migrations); | |
1157 | } | |
6cfb0d5d | 1158 | #endif |
2830cf8c SV |
1159 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1160 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1161 | |
1162 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1163 | } |
1164 | ||
70b97a7f | 1165 | struct migration_req { |
1da177e4 | 1166 | struct list_head list; |
1da177e4 | 1167 | |
36c8b586 | 1168 | struct task_struct *task; |
1da177e4 LT |
1169 | int dest_cpu; |
1170 | ||
1da177e4 | 1171 | struct completion done; |
70b97a7f | 1172 | }; |
1da177e4 LT |
1173 | |
1174 | /* | |
1175 | * The task's runqueue lock must be held. | |
1176 | * Returns true if you have to wait for migration thread. | |
1177 | */ | |
36c8b586 | 1178 | static int |
70b97a7f | 1179 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1180 | { |
70b97a7f | 1181 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1182 | |
1183 | /* | |
1184 | * If the task is not on a runqueue (and not running), then | |
1185 | * it is sufficient to simply update the task's cpu field. | |
1186 | */ | |
dd41f596 | 1187 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1188 | set_task_cpu(p, dest_cpu); |
1189 | return 0; | |
1190 | } | |
1191 | ||
1192 | init_completion(&req->done); | |
1da177e4 LT |
1193 | req->task = p; |
1194 | req->dest_cpu = dest_cpu; | |
1195 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1196 | |
1da177e4 LT |
1197 | return 1; |
1198 | } | |
1199 | ||
1200 | /* | |
1201 | * wait_task_inactive - wait for a thread to unschedule. | |
1202 | * | |
1203 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1204 | * else this function might spin for a *long* time. This function can't | |
1205 | * be called with interrupts off, or it may introduce deadlock with | |
1206 | * smp_call_function() if an IPI is sent by the same process we are | |
1207 | * waiting to become inactive. | |
1208 | */ | |
36c8b586 | 1209 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1210 | { |
1211 | unsigned long flags; | |
dd41f596 | 1212 | int running, on_rq; |
70b97a7f | 1213 | struct rq *rq; |
1da177e4 | 1214 | |
3a5c359a AK |
1215 | for (;;) { |
1216 | /* | |
1217 | * We do the initial early heuristics without holding | |
1218 | * any task-queue locks at all. We'll only try to get | |
1219 | * the runqueue lock when things look like they will | |
1220 | * work out! | |
1221 | */ | |
1222 | rq = task_rq(p); | |
fa490cfd | 1223 | |
3a5c359a AK |
1224 | /* |
1225 | * If the task is actively running on another CPU | |
1226 | * still, just relax and busy-wait without holding | |
1227 | * any locks. | |
1228 | * | |
1229 | * NOTE! Since we don't hold any locks, it's not | |
1230 | * even sure that "rq" stays as the right runqueue! | |
1231 | * But we don't care, since "task_running()" will | |
1232 | * return false if the runqueue has changed and p | |
1233 | * is actually now running somewhere else! | |
1234 | */ | |
1235 | while (task_running(rq, p)) | |
1236 | cpu_relax(); | |
fa490cfd | 1237 | |
3a5c359a AK |
1238 | /* |
1239 | * Ok, time to look more closely! We need the rq | |
1240 | * lock now, to be *sure*. If we're wrong, we'll | |
1241 | * just go back and repeat. | |
1242 | */ | |
1243 | rq = task_rq_lock(p, &flags); | |
1244 | running = task_running(rq, p); | |
1245 | on_rq = p->se.on_rq; | |
1246 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1247 | |
3a5c359a AK |
1248 | /* |
1249 | * Was it really running after all now that we | |
1250 | * checked with the proper locks actually held? | |
1251 | * | |
1252 | * Oops. Go back and try again.. | |
1253 | */ | |
1254 | if (unlikely(running)) { | |
1255 | cpu_relax(); | |
1256 | continue; | |
1257 | } | |
fa490cfd | 1258 | |
3a5c359a AK |
1259 | /* |
1260 | * It's not enough that it's not actively running, | |
1261 | * it must be off the runqueue _entirely_, and not | |
1262 | * preempted! | |
1263 | * | |
1264 | * So if it wa still runnable (but just not actively | |
1265 | * running right now), it's preempted, and we should | |
1266 | * yield - it could be a while. | |
1267 | */ | |
1268 | if (unlikely(on_rq)) { | |
1269 | schedule_timeout_uninterruptible(1); | |
1270 | continue; | |
1271 | } | |
fa490cfd | 1272 | |
3a5c359a AK |
1273 | /* |
1274 | * Ahh, all good. It wasn't running, and it wasn't | |
1275 | * runnable, which means that it will never become | |
1276 | * running in the future either. We're all done! | |
1277 | */ | |
1278 | break; | |
1279 | } | |
1da177e4 LT |
1280 | } |
1281 | ||
1282 | /*** | |
1283 | * kick_process - kick a running thread to enter/exit the kernel | |
1284 | * @p: the to-be-kicked thread | |
1285 | * | |
1286 | * Cause a process which is running on another CPU to enter | |
1287 | * kernel-mode, without any delay. (to get signals handled.) | |
1288 | * | |
1289 | * NOTE: this function doesnt have to take the runqueue lock, | |
1290 | * because all it wants to ensure is that the remote task enters | |
1291 | * the kernel. If the IPI races and the task has been migrated | |
1292 | * to another CPU then no harm is done and the purpose has been | |
1293 | * achieved as well. | |
1294 | */ | |
36c8b586 | 1295 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1296 | { |
1297 | int cpu; | |
1298 | ||
1299 | preempt_disable(); | |
1300 | cpu = task_cpu(p); | |
1301 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1302 | smp_send_reschedule(cpu); | |
1303 | preempt_enable(); | |
1304 | } | |
1305 | ||
1306 | /* | |
2dd73a4f PW |
1307 | * Return a low guess at the load of a migration-source cpu weighted |
1308 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1309 | * |
1310 | * We want to under-estimate the load of migration sources, to | |
1311 | * balance conservatively. | |
1312 | */ | |
a9957449 | 1313 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1314 | { |
70b97a7f | 1315 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1316 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1317 | |
3b0bd9bc | 1318 | if (type == 0) |
dd41f596 | 1319 | return total; |
b910472d | 1320 | |
dd41f596 | 1321 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1322 | } |
1323 | ||
1324 | /* | |
2dd73a4f PW |
1325 | * Return a high guess at the load of a migration-target cpu weighted |
1326 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1327 | */ |
a9957449 | 1328 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1329 | { |
70b97a7f | 1330 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1331 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1332 | |
7897986b | 1333 | if (type == 0) |
dd41f596 | 1334 | return total; |
3b0bd9bc | 1335 | |
dd41f596 | 1336 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1337 | } |
1338 | ||
1339 | /* | |
1340 | * Return the average load per task on the cpu's run queue | |
1341 | */ | |
1342 | static inline unsigned long cpu_avg_load_per_task(int cpu) | |
1343 | { | |
70b97a7f | 1344 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1345 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1346 | unsigned long n = rq->nr_running; |
1347 | ||
dd41f596 | 1348 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1349 | } |
1350 | ||
147cbb4b NP |
1351 | /* |
1352 | * find_idlest_group finds and returns the least busy CPU group within the | |
1353 | * domain. | |
1354 | */ | |
1355 | static struct sched_group * | |
1356 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1357 | { | |
1358 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1359 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1360 | int load_idx = sd->forkexec_idx; | |
1361 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1362 | ||
1363 | do { | |
1364 | unsigned long load, avg_load; | |
1365 | int local_group; | |
1366 | int i; | |
1367 | ||
da5a5522 BD |
1368 | /* Skip over this group if it has no CPUs allowed */ |
1369 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1370 | continue; |
da5a5522 | 1371 | |
147cbb4b | 1372 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1373 | |
1374 | /* Tally up the load of all CPUs in the group */ | |
1375 | avg_load = 0; | |
1376 | ||
1377 | for_each_cpu_mask(i, group->cpumask) { | |
1378 | /* Bias balancing toward cpus of our domain */ | |
1379 | if (local_group) | |
1380 | load = source_load(i, load_idx); | |
1381 | else | |
1382 | load = target_load(i, load_idx); | |
1383 | ||
1384 | avg_load += load; | |
1385 | } | |
1386 | ||
1387 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1388 | avg_load = sg_div_cpu_power(group, |
1389 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1390 | |
1391 | if (local_group) { | |
1392 | this_load = avg_load; | |
1393 | this = group; | |
1394 | } else if (avg_load < min_load) { | |
1395 | min_load = avg_load; | |
1396 | idlest = group; | |
1397 | } | |
3a5c359a | 1398 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1399 | |
1400 | if (!idlest || 100*this_load < imbalance*min_load) | |
1401 | return NULL; | |
1402 | return idlest; | |
1403 | } | |
1404 | ||
1405 | /* | |
0feaece9 | 1406 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1407 | */ |
95cdf3b7 IM |
1408 | static int |
1409 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) | |
147cbb4b | 1410 | { |
da5a5522 | 1411 | cpumask_t tmp; |
147cbb4b NP |
1412 | unsigned long load, min_load = ULONG_MAX; |
1413 | int idlest = -1; | |
1414 | int i; | |
1415 | ||
da5a5522 BD |
1416 | /* Traverse only the allowed CPUs */ |
1417 | cpus_and(tmp, group->cpumask, p->cpus_allowed); | |
1418 | ||
1419 | for_each_cpu_mask(i, tmp) { | |
2dd73a4f | 1420 | load = weighted_cpuload(i); |
147cbb4b NP |
1421 | |
1422 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
1423 | min_load = load; | |
1424 | idlest = i; | |
1425 | } | |
1426 | } | |
1427 | ||
1428 | return idlest; | |
1429 | } | |
1430 | ||
476d139c NP |
1431 | /* |
1432 | * sched_balance_self: balance the current task (running on cpu) in domains | |
1433 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
1434 | * SD_BALANCE_EXEC. | |
1435 | * | |
1436 | * Balance, ie. select the least loaded group. | |
1437 | * | |
1438 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
1439 | * | |
1440 | * preempt must be disabled. | |
1441 | */ | |
1442 | static int sched_balance_self(int cpu, int flag) | |
1443 | { | |
1444 | struct task_struct *t = current; | |
1445 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 1446 | |
c96d145e | 1447 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
1448 | /* |
1449 | * If power savings logic is enabled for a domain, stop there. | |
1450 | */ | |
5c45bf27 SS |
1451 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
1452 | break; | |
476d139c NP |
1453 | if (tmp->flags & flag) |
1454 | sd = tmp; | |
c96d145e | 1455 | } |
476d139c NP |
1456 | |
1457 | while (sd) { | |
1458 | cpumask_t span; | |
1459 | struct sched_group *group; | |
1a848870 SS |
1460 | int new_cpu, weight; |
1461 | ||
1462 | if (!(sd->flags & flag)) { | |
1463 | sd = sd->child; | |
1464 | continue; | |
1465 | } | |
476d139c NP |
1466 | |
1467 | span = sd->span; | |
1468 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
1469 | if (!group) { |
1470 | sd = sd->child; | |
1471 | continue; | |
1472 | } | |
476d139c | 1473 | |
da5a5522 | 1474 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
1475 | if (new_cpu == -1 || new_cpu == cpu) { |
1476 | /* Now try balancing at a lower domain level of cpu */ | |
1477 | sd = sd->child; | |
1478 | continue; | |
1479 | } | |
476d139c | 1480 | |
1a848870 | 1481 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 1482 | cpu = new_cpu; |
476d139c NP |
1483 | sd = NULL; |
1484 | weight = cpus_weight(span); | |
1485 | for_each_domain(cpu, tmp) { | |
1486 | if (weight <= cpus_weight(tmp->span)) | |
1487 | break; | |
1488 | if (tmp->flags & flag) | |
1489 | sd = tmp; | |
1490 | } | |
1491 | /* while loop will break here if sd == NULL */ | |
1492 | } | |
1493 | ||
1494 | return cpu; | |
1495 | } | |
1496 | ||
1497 | #endif /* CONFIG_SMP */ | |
1da177e4 LT |
1498 | |
1499 | /* | |
1500 | * wake_idle() will wake a task on an idle cpu if task->cpu is | |
1501 | * not idle and an idle cpu is available. The span of cpus to | |
1502 | * search starts with cpus closest then further out as needed, | |
1503 | * so we always favor a closer, idle cpu. | |
1504 | * | |
1505 | * Returns the CPU we should wake onto. | |
1506 | */ | |
1507 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) | |
36c8b586 | 1508 | static int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1509 | { |
1510 | cpumask_t tmp; | |
1511 | struct sched_domain *sd; | |
1512 | int i; | |
1513 | ||
4953198b SS |
1514 | /* |
1515 | * If it is idle, then it is the best cpu to run this task. | |
1516 | * | |
1517 | * This cpu is also the best, if it has more than one task already. | |
1518 | * Siblings must be also busy(in most cases) as they didn't already | |
1519 | * pickup the extra load from this cpu and hence we need not check | |
1520 | * sibling runqueue info. This will avoid the checks and cache miss | |
1521 | * penalities associated with that. | |
1522 | */ | |
1523 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) | |
1da177e4 LT |
1524 | return cpu; |
1525 | ||
1526 | for_each_domain(cpu, sd) { | |
1527 | if (sd->flags & SD_WAKE_IDLE) { | |
e0f364f4 | 1528 | cpus_and(tmp, sd->span, p->cpus_allowed); |
1da177e4 | 1529 | for_each_cpu_mask(i, tmp) { |
cc367732 IM |
1530 | if (idle_cpu(i)) { |
1531 | if (i != task_cpu(p)) { | |
1532 | schedstat_inc(p, | |
1533 | se.nr_wakeups_idle); | |
1534 | } | |
1da177e4 | 1535 | return i; |
cc367732 | 1536 | } |
1da177e4 | 1537 | } |
9761eea8 | 1538 | } else { |
e0f364f4 | 1539 | break; |
9761eea8 | 1540 | } |
1da177e4 LT |
1541 | } |
1542 | return cpu; | |
1543 | } | |
1544 | #else | |
36c8b586 | 1545 | static inline int wake_idle(int cpu, struct task_struct *p) |
1da177e4 LT |
1546 | { |
1547 | return cpu; | |
1548 | } | |
1549 | #endif | |
1550 | ||
1551 | /*** | |
1552 | * try_to_wake_up - wake up a thread | |
1553 | * @p: the to-be-woken-up thread | |
1554 | * @state: the mask of task states that can be woken | |
1555 | * @sync: do a synchronous wakeup? | |
1556 | * | |
1557 | * Put it on the run-queue if it's not already there. The "current" | |
1558 | * thread is always on the run-queue (except when the actual | |
1559 | * re-schedule is in progress), and as such you're allowed to do | |
1560 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
1561 | * runnable without the overhead of this. | |
1562 | * | |
1563 | * returns failure only if the task is already active. | |
1564 | */ | |
36c8b586 | 1565 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 1566 | { |
cc367732 | 1567 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
1568 | unsigned long flags; |
1569 | long old_state; | |
70b97a7f | 1570 | struct rq *rq; |
1da177e4 | 1571 | #ifdef CONFIG_SMP |
7897986b | 1572 | struct sched_domain *sd, *this_sd = NULL; |
70b97a7f | 1573 | unsigned long load, this_load; |
1da177e4 LT |
1574 | int new_cpu; |
1575 | #endif | |
1576 | ||
1577 | rq = task_rq_lock(p, &flags); | |
1578 | old_state = p->state; | |
1579 | if (!(old_state & state)) | |
1580 | goto out; | |
1581 | ||
dd41f596 | 1582 | if (p->se.on_rq) |
1da177e4 LT |
1583 | goto out_running; |
1584 | ||
1585 | cpu = task_cpu(p); | |
cc367732 | 1586 | orig_cpu = cpu; |
1da177e4 LT |
1587 | this_cpu = smp_processor_id(); |
1588 | ||
1589 | #ifdef CONFIG_SMP | |
1590 | if (unlikely(task_running(rq, p))) | |
1591 | goto out_activate; | |
1592 | ||
7897986b NP |
1593 | new_cpu = cpu; |
1594 | ||
2d72376b | 1595 | schedstat_inc(rq, ttwu_count); |
1da177e4 LT |
1596 | if (cpu == this_cpu) { |
1597 | schedstat_inc(rq, ttwu_local); | |
7897986b NP |
1598 | goto out_set_cpu; |
1599 | } | |
1600 | ||
1601 | for_each_domain(this_cpu, sd) { | |
1602 | if (cpu_isset(cpu, sd->span)) { | |
1603 | schedstat_inc(sd, ttwu_wake_remote); | |
1604 | this_sd = sd; | |
1605 | break; | |
1da177e4 LT |
1606 | } |
1607 | } | |
1da177e4 | 1608 | |
7897986b | 1609 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
1da177e4 LT |
1610 | goto out_set_cpu; |
1611 | ||
1da177e4 | 1612 | /* |
7897986b | 1613 | * Check for affine wakeup and passive balancing possibilities. |
1da177e4 | 1614 | */ |
7897986b NP |
1615 | if (this_sd) { |
1616 | int idx = this_sd->wake_idx; | |
1617 | unsigned int imbalance; | |
1da177e4 | 1618 | |
a3f21bce NP |
1619 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
1620 | ||
7897986b NP |
1621 | load = source_load(cpu, idx); |
1622 | this_load = target_load(this_cpu, idx); | |
1da177e4 | 1623 | |
7897986b NP |
1624 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
1625 | ||
a3f21bce NP |
1626 | if (this_sd->flags & SD_WAKE_AFFINE) { |
1627 | unsigned long tl = this_load; | |
33859f7f MOS |
1628 | unsigned long tl_per_task; |
1629 | ||
71e20f18 IM |
1630 | /* |
1631 | * Attract cache-cold tasks on sync wakeups: | |
1632 | */ | |
1633 | if (sync && !task_hot(p, rq->clock, this_sd)) | |
1634 | goto out_set_cpu; | |
1635 | ||
cc367732 | 1636 | schedstat_inc(p, se.nr_wakeups_affine_attempts); |
33859f7f | 1637 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
2dd73a4f | 1638 | |
1da177e4 | 1639 | /* |
a3f21bce NP |
1640 | * If sync wakeup then subtract the (maximum possible) |
1641 | * effect of the currently running task from the load | |
1642 | * of the current CPU: | |
1da177e4 | 1643 | */ |
a3f21bce | 1644 | if (sync) |
dd41f596 | 1645 | tl -= current->se.load.weight; |
a3f21bce NP |
1646 | |
1647 | if ((tl <= load && | |
2dd73a4f | 1648 | tl + target_load(cpu, idx) <= tl_per_task) || |
dd41f596 | 1649 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
a3f21bce NP |
1650 | /* |
1651 | * This domain has SD_WAKE_AFFINE and | |
1652 | * p is cache cold in this domain, and | |
1653 | * there is no bad imbalance. | |
1654 | */ | |
1655 | schedstat_inc(this_sd, ttwu_move_affine); | |
cc367732 | 1656 | schedstat_inc(p, se.nr_wakeups_affine); |
a3f21bce NP |
1657 | goto out_set_cpu; |
1658 | } | |
1659 | } | |
1660 | ||
1661 | /* | |
1662 | * Start passive balancing when half the imbalance_pct | |
1663 | * limit is reached. | |
1664 | */ | |
1665 | if (this_sd->flags & SD_WAKE_BALANCE) { | |
1666 | if (imbalance*this_load <= 100*load) { | |
1667 | schedstat_inc(this_sd, ttwu_move_balance); | |
cc367732 | 1668 | schedstat_inc(p, se.nr_wakeups_passive); |
a3f21bce NP |
1669 | goto out_set_cpu; |
1670 | } | |
1da177e4 LT |
1671 | } |
1672 | } | |
1673 | ||
1674 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ | |
1675 | out_set_cpu: | |
1676 | new_cpu = wake_idle(new_cpu, p); | |
1677 | if (new_cpu != cpu) { | |
1678 | set_task_cpu(p, new_cpu); | |
1679 | task_rq_unlock(rq, &flags); | |
1680 | /* might preempt at this point */ | |
1681 | rq = task_rq_lock(p, &flags); | |
1682 | old_state = p->state; | |
1683 | if (!(old_state & state)) | |
1684 | goto out; | |
dd41f596 | 1685 | if (p->se.on_rq) |
1da177e4 LT |
1686 | goto out_running; |
1687 | ||
1688 | this_cpu = smp_processor_id(); | |
1689 | cpu = task_cpu(p); | |
1690 | } | |
1691 | ||
1692 | out_activate: | |
1693 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
1694 | schedstat_inc(p, se.nr_wakeups); |
1695 | if (sync) | |
1696 | schedstat_inc(p, se.nr_wakeups_sync); | |
1697 | if (orig_cpu != cpu) | |
1698 | schedstat_inc(p, se.nr_wakeups_migrate); | |
1699 | if (cpu == this_cpu) | |
1700 | schedstat_inc(p, se.nr_wakeups_local); | |
1701 | else | |
1702 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 1703 | update_rq_clock(rq); |
dd41f596 | 1704 | activate_task(rq, p, 1); |
9c63d9c0 | 1705 | check_preempt_curr(rq, p); |
1da177e4 LT |
1706 | success = 1; |
1707 | ||
1708 | out_running: | |
1709 | p->state = TASK_RUNNING; | |
1710 | out: | |
1711 | task_rq_unlock(rq, &flags); | |
1712 | ||
1713 | return success; | |
1714 | } | |
1715 | ||
36c8b586 | 1716 | int fastcall wake_up_process(struct task_struct *p) |
1da177e4 LT |
1717 | { |
1718 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | | |
1719 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); | |
1720 | } | |
1da177e4 LT |
1721 | EXPORT_SYMBOL(wake_up_process); |
1722 | ||
36c8b586 | 1723 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
1724 | { |
1725 | return try_to_wake_up(p, state, 0); | |
1726 | } | |
1727 | ||
1da177e4 LT |
1728 | /* |
1729 | * Perform scheduler related setup for a newly forked process p. | |
1730 | * p is forked by current. | |
dd41f596 IM |
1731 | * |
1732 | * __sched_fork() is basic setup used by init_idle() too: | |
1733 | */ | |
1734 | static void __sched_fork(struct task_struct *p) | |
1735 | { | |
dd41f596 IM |
1736 | p->se.exec_start = 0; |
1737 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 1738 | p->se.prev_sum_exec_runtime = 0; |
6cfb0d5d IM |
1739 | |
1740 | #ifdef CONFIG_SCHEDSTATS | |
1741 | p->se.wait_start = 0; | |
dd41f596 IM |
1742 | p->se.sum_sleep_runtime = 0; |
1743 | p->se.sleep_start = 0; | |
dd41f596 IM |
1744 | p->se.block_start = 0; |
1745 | p->se.sleep_max = 0; | |
1746 | p->se.block_max = 0; | |
1747 | p->se.exec_max = 0; | |
eba1ed4b | 1748 | p->se.slice_max = 0; |
dd41f596 | 1749 | p->se.wait_max = 0; |
6cfb0d5d | 1750 | #endif |
476d139c | 1751 | |
dd41f596 IM |
1752 | INIT_LIST_HEAD(&p->run_list); |
1753 | p->se.on_rq = 0; | |
476d139c | 1754 | |
e107be36 AK |
1755 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1756 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
1757 | #endif | |
1758 | ||
1da177e4 LT |
1759 | /* |
1760 | * We mark the process as running here, but have not actually | |
1761 | * inserted it onto the runqueue yet. This guarantees that | |
1762 | * nobody will actually run it, and a signal or other external | |
1763 | * event cannot wake it up and insert it on the runqueue either. | |
1764 | */ | |
1765 | p->state = TASK_RUNNING; | |
dd41f596 IM |
1766 | } |
1767 | ||
1768 | /* | |
1769 | * fork()/clone()-time setup: | |
1770 | */ | |
1771 | void sched_fork(struct task_struct *p, int clone_flags) | |
1772 | { | |
1773 | int cpu = get_cpu(); | |
1774 | ||
1775 | __sched_fork(p); | |
1776 | ||
1777 | #ifdef CONFIG_SMP | |
1778 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
1779 | #endif | |
02e4bac2 | 1780 | set_task_cpu(p, cpu); |
b29739f9 IM |
1781 | |
1782 | /* | |
1783 | * Make sure we do not leak PI boosting priority to the child: | |
1784 | */ | |
1785 | p->prio = current->normal_prio; | |
2ddbf952 HS |
1786 | if (!rt_prio(p->prio)) |
1787 | p->sched_class = &fair_sched_class; | |
b29739f9 | 1788 | |
52f17b6c | 1789 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 1790 | if (likely(sched_info_on())) |
52f17b6c | 1791 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 1792 | #endif |
d6077cb8 | 1793 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
1794 | p->oncpu = 0; |
1795 | #endif | |
1da177e4 | 1796 | #ifdef CONFIG_PREEMPT |
4866cde0 | 1797 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 1798 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 1799 | #endif |
476d139c | 1800 | put_cpu(); |
1da177e4 LT |
1801 | } |
1802 | ||
1803 | /* | |
1804 | * wake_up_new_task - wake up a newly created task for the first time. | |
1805 | * | |
1806 | * This function will do some initial scheduler statistics housekeeping | |
1807 | * that must be done for every newly created context, then puts the task | |
1808 | * on the runqueue and wakes it. | |
1809 | */ | |
36c8b586 | 1810 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
1811 | { |
1812 | unsigned long flags; | |
dd41f596 | 1813 | struct rq *rq; |
1da177e4 LT |
1814 | |
1815 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 1816 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 1817 | update_rq_clock(rq); |
1da177e4 LT |
1818 | |
1819 | p->prio = effective_prio(p); | |
1820 | ||
b9dca1e0 | 1821 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 1822 | activate_task(rq, p, 0); |
1da177e4 | 1823 | } else { |
1da177e4 | 1824 | /* |
dd41f596 IM |
1825 | * Let the scheduling class do new task startup |
1826 | * management (if any): | |
1da177e4 | 1827 | */ |
ee0827d8 | 1828 | p->sched_class->task_new(rq, p); |
e5fa2237 | 1829 | inc_nr_running(p, rq); |
1da177e4 | 1830 | } |
dd41f596 IM |
1831 | check_preempt_curr(rq, p); |
1832 | task_rq_unlock(rq, &flags); | |
1da177e4 LT |
1833 | } |
1834 | ||
e107be36 AK |
1835 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
1836 | ||
1837 | /** | |
421cee29 RD |
1838 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
1839 | * @notifier: notifier struct to register | |
e107be36 AK |
1840 | */ |
1841 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
1842 | { | |
1843 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
1844 | } | |
1845 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
1846 | ||
1847 | /** | |
1848 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 1849 | * @notifier: notifier struct to unregister |
e107be36 AK |
1850 | * |
1851 | * This is safe to call from within a preemption notifier. | |
1852 | */ | |
1853 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
1854 | { | |
1855 | hlist_del(¬ifier->link); | |
1856 | } | |
1857 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
1858 | ||
1859 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1860 | { | |
1861 | struct preempt_notifier *notifier; | |
1862 | struct hlist_node *node; | |
1863 | ||
1864 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1865 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
1866 | } | |
1867 | ||
1868 | static void | |
1869 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1870 | struct task_struct *next) | |
1871 | { | |
1872 | struct preempt_notifier *notifier; | |
1873 | struct hlist_node *node; | |
1874 | ||
1875 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
1876 | notifier->ops->sched_out(notifier, next); | |
1877 | } | |
1878 | ||
1879 | #else | |
1880 | ||
1881 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
1882 | { | |
1883 | } | |
1884 | ||
1885 | static void | |
1886 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
1887 | struct task_struct *next) | |
1888 | { | |
1889 | } | |
1890 | ||
1891 | #endif | |
1892 | ||
4866cde0 NP |
1893 | /** |
1894 | * prepare_task_switch - prepare to switch tasks | |
1895 | * @rq: the runqueue preparing to switch | |
421cee29 | 1896 | * @prev: the current task that is being switched out |
4866cde0 NP |
1897 | * @next: the task we are going to switch to. |
1898 | * | |
1899 | * This is called with the rq lock held and interrupts off. It must | |
1900 | * be paired with a subsequent finish_task_switch after the context | |
1901 | * switch. | |
1902 | * | |
1903 | * prepare_task_switch sets up locking and calls architecture specific | |
1904 | * hooks. | |
1905 | */ | |
e107be36 AK |
1906 | static inline void |
1907 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
1908 | struct task_struct *next) | |
4866cde0 | 1909 | { |
e107be36 | 1910 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
1911 | prepare_lock_switch(rq, next); |
1912 | prepare_arch_switch(next); | |
1913 | } | |
1914 | ||
1da177e4 LT |
1915 | /** |
1916 | * finish_task_switch - clean up after a task-switch | |
344babaa | 1917 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
1918 | * @prev: the thread we just switched away from. |
1919 | * | |
4866cde0 NP |
1920 | * finish_task_switch must be called after the context switch, paired |
1921 | * with a prepare_task_switch call before the context switch. | |
1922 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
1923 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
1924 | * |
1925 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 1926 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
1927 | * with the lock held can cause deadlocks; see schedule() for |
1928 | * details.) | |
1929 | */ | |
a9957449 | 1930 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
1931 | __releases(rq->lock) |
1932 | { | |
1da177e4 | 1933 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 1934 | long prev_state; |
1da177e4 LT |
1935 | |
1936 | rq->prev_mm = NULL; | |
1937 | ||
1938 | /* | |
1939 | * A task struct has one reference for the use as "current". | |
c394cc9f | 1940 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
1941 | * schedule one last time. The schedule call will never return, and |
1942 | * the scheduled task must drop that reference. | |
c394cc9f | 1943 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
1944 | * still held, otherwise prev could be scheduled on another cpu, die |
1945 | * there before we look at prev->state, and then the reference would | |
1946 | * be dropped twice. | |
1947 | * Manfred Spraul <manfred@colorfullife.com> | |
1948 | */ | |
55a101f8 | 1949 | prev_state = prev->state; |
4866cde0 NP |
1950 | finish_arch_switch(prev); |
1951 | finish_lock_switch(rq, prev); | |
e107be36 | 1952 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
1953 | if (mm) |
1954 | mmdrop(mm); | |
c394cc9f | 1955 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 1956 | /* |
1957 | * Remove function-return probe instances associated with this | |
1958 | * task and put them back on the free list. | |
9761eea8 | 1959 | */ |
c6fd91f0 | 1960 | kprobe_flush_task(prev); |
1da177e4 | 1961 | put_task_struct(prev); |
c6fd91f0 | 1962 | } |
1da177e4 LT |
1963 | } |
1964 | ||
1965 | /** | |
1966 | * schedule_tail - first thing a freshly forked thread must call. | |
1967 | * @prev: the thread we just switched away from. | |
1968 | */ | |
36c8b586 | 1969 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
1970 | __releases(rq->lock) |
1971 | { | |
70b97a7f IM |
1972 | struct rq *rq = this_rq(); |
1973 | ||
4866cde0 NP |
1974 | finish_task_switch(rq, prev); |
1975 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
1976 | /* In this case, finish_task_switch does not reenable preemption */ | |
1977 | preempt_enable(); | |
1978 | #endif | |
1da177e4 | 1979 | if (current->set_child_tid) |
b488893a | 1980 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | /* | |
1984 | * context_switch - switch to the new MM and the new | |
1985 | * thread's register state. | |
1986 | */ | |
dd41f596 | 1987 | static inline void |
70b97a7f | 1988 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 1989 | struct task_struct *next) |
1da177e4 | 1990 | { |
dd41f596 | 1991 | struct mm_struct *mm, *oldmm; |
1da177e4 | 1992 | |
e107be36 | 1993 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
1994 | mm = next->mm; |
1995 | oldmm = prev->active_mm; | |
9226d125 ZA |
1996 | /* |
1997 | * For paravirt, this is coupled with an exit in switch_to to | |
1998 | * combine the page table reload and the switch backend into | |
1999 | * one hypercall. | |
2000 | */ | |
2001 | arch_enter_lazy_cpu_mode(); | |
2002 | ||
dd41f596 | 2003 | if (unlikely(!mm)) { |
1da177e4 LT |
2004 | next->active_mm = oldmm; |
2005 | atomic_inc(&oldmm->mm_count); | |
2006 | enter_lazy_tlb(oldmm, next); | |
2007 | } else | |
2008 | switch_mm(oldmm, mm, next); | |
2009 | ||
dd41f596 | 2010 | if (unlikely(!prev->mm)) { |
1da177e4 | 2011 | prev->active_mm = NULL; |
1da177e4 LT |
2012 | rq->prev_mm = oldmm; |
2013 | } | |
3a5f5e48 IM |
2014 | /* |
2015 | * Since the runqueue lock will be released by the next | |
2016 | * task (which is an invalid locking op but in the case | |
2017 | * of the scheduler it's an obvious special-case), so we | |
2018 | * do an early lockdep release here: | |
2019 | */ | |
2020 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2021 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2022 | #endif |
1da177e4 LT |
2023 | |
2024 | /* Here we just switch the register state and the stack. */ | |
2025 | switch_to(prev, next, prev); | |
2026 | ||
dd41f596 IM |
2027 | barrier(); |
2028 | /* | |
2029 | * this_rq must be evaluated again because prev may have moved | |
2030 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2031 | * frame will be invalid. | |
2032 | */ | |
2033 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2034 | } |
2035 | ||
2036 | /* | |
2037 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2038 | * | |
2039 | * externally visible scheduler statistics: current number of runnable | |
2040 | * threads, current number of uninterruptible-sleeping threads, total | |
2041 | * number of context switches performed since bootup. | |
2042 | */ | |
2043 | unsigned long nr_running(void) | |
2044 | { | |
2045 | unsigned long i, sum = 0; | |
2046 | ||
2047 | for_each_online_cpu(i) | |
2048 | sum += cpu_rq(i)->nr_running; | |
2049 | ||
2050 | return sum; | |
2051 | } | |
2052 | ||
2053 | unsigned long nr_uninterruptible(void) | |
2054 | { | |
2055 | unsigned long i, sum = 0; | |
2056 | ||
0a945022 | 2057 | for_each_possible_cpu(i) |
1da177e4 LT |
2058 | sum += cpu_rq(i)->nr_uninterruptible; |
2059 | ||
2060 | /* | |
2061 | * Since we read the counters lockless, it might be slightly | |
2062 | * inaccurate. Do not allow it to go below zero though: | |
2063 | */ | |
2064 | if (unlikely((long)sum < 0)) | |
2065 | sum = 0; | |
2066 | ||
2067 | return sum; | |
2068 | } | |
2069 | ||
2070 | unsigned long long nr_context_switches(void) | |
2071 | { | |
cc94abfc SR |
2072 | int i; |
2073 | unsigned long long sum = 0; | |
1da177e4 | 2074 | |
0a945022 | 2075 | for_each_possible_cpu(i) |
1da177e4 LT |
2076 | sum += cpu_rq(i)->nr_switches; |
2077 | ||
2078 | return sum; | |
2079 | } | |
2080 | ||
2081 | unsigned long nr_iowait(void) | |
2082 | { | |
2083 | unsigned long i, sum = 0; | |
2084 | ||
0a945022 | 2085 | for_each_possible_cpu(i) |
1da177e4 LT |
2086 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2087 | ||
2088 | return sum; | |
2089 | } | |
2090 | ||
db1b1fef JS |
2091 | unsigned long nr_active(void) |
2092 | { | |
2093 | unsigned long i, running = 0, uninterruptible = 0; | |
2094 | ||
2095 | for_each_online_cpu(i) { | |
2096 | running += cpu_rq(i)->nr_running; | |
2097 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2098 | } | |
2099 | ||
2100 | if (unlikely((long)uninterruptible < 0)) | |
2101 | uninterruptible = 0; | |
2102 | ||
2103 | return running + uninterruptible; | |
2104 | } | |
2105 | ||
48f24c4d | 2106 | /* |
dd41f596 IM |
2107 | * Update rq->cpu_load[] statistics. This function is usually called every |
2108 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2109 | */ |
dd41f596 | 2110 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2111 | { |
495eca49 | 2112 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2113 | int i, scale; |
2114 | ||
2115 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2116 | |
2117 | /* Update our load: */ | |
2118 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2119 | unsigned long old_load, new_load; | |
2120 | ||
2121 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2122 | ||
2123 | old_load = this_rq->cpu_load[i]; | |
2124 | new_load = this_load; | |
a25707f3 IM |
2125 | /* |
2126 | * Round up the averaging division if load is increasing. This | |
2127 | * prevents us from getting stuck on 9 if the load is 10, for | |
2128 | * example. | |
2129 | */ | |
2130 | if (new_load > old_load) | |
2131 | new_load += scale-1; | |
dd41f596 IM |
2132 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2133 | } | |
48f24c4d IM |
2134 | } |
2135 | ||
dd41f596 IM |
2136 | #ifdef CONFIG_SMP |
2137 | ||
1da177e4 LT |
2138 | /* |
2139 | * double_rq_lock - safely lock two runqueues | |
2140 | * | |
2141 | * Note this does not disable interrupts like task_rq_lock, | |
2142 | * you need to do so manually before calling. | |
2143 | */ | |
70b97a7f | 2144 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2145 | __acquires(rq1->lock) |
2146 | __acquires(rq2->lock) | |
2147 | { | |
054b9108 | 2148 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2149 | if (rq1 == rq2) { |
2150 | spin_lock(&rq1->lock); | |
2151 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2152 | } else { | |
c96d145e | 2153 | if (rq1 < rq2) { |
1da177e4 LT |
2154 | spin_lock(&rq1->lock); |
2155 | spin_lock(&rq2->lock); | |
2156 | } else { | |
2157 | spin_lock(&rq2->lock); | |
2158 | spin_lock(&rq1->lock); | |
2159 | } | |
2160 | } | |
6e82a3be IM |
2161 | update_rq_clock(rq1); |
2162 | update_rq_clock(rq2); | |
1da177e4 LT |
2163 | } |
2164 | ||
2165 | /* | |
2166 | * double_rq_unlock - safely unlock two runqueues | |
2167 | * | |
2168 | * Note this does not restore interrupts like task_rq_unlock, | |
2169 | * you need to do so manually after calling. | |
2170 | */ | |
70b97a7f | 2171 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2172 | __releases(rq1->lock) |
2173 | __releases(rq2->lock) | |
2174 | { | |
2175 | spin_unlock(&rq1->lock); | |
2176 | if (rq1 != rq2) | |
2177 | spin_unlock(&rq2->lock); | |
2178 | else | |
2179 | __release(rq2->lock); | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2184 | */ | |
70b97a7f | 2185 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2186 | __releases(this_rq->lock) |
2187 | __acquires(busiest->lock) | |
2188 | __acquires(this_rq->lock) | |
2189 | { | |
054b9108 KK |
2190 | if (unlikely(!irqs_disabled())) { |
2191 | /* printk() doesn't work good under rq->lock */ | |
2192 | spin_unlock(&this_rq->lock); | |
2193 | BUG_ON(1); | |
2194 | } | |
1da177e4 | 2195 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2196 | if (busiest < this_rq) { |
1da177e4 LT |
2197 | spin_unlock(&this_rq->lock); |
2198 | spin_lock(&busiest->lock); | |
2199 | spin_lock(&this_rq->lock); | |
2200 | } else | |
2201 | spin_lock(&busiest->lock); | |
2202 | } | |
2203 | } | |
2204 | ||
1da177e4 LT |
2205 | /* |
2206 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2207 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2208 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2209 | * the cpu_allowed mask is restored. |
2210 | */ | |
36c8b586 | 2211 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2212 | { |
70b97a7f | 2213 | struct migration_req req; |
1da177e4 | 2214 | unsigned long flags; |
70b97a7f | 2215 | struct rq *rq; |
1da177e4 LT |
2216 | |
2217 | rq = task_rq_lock(p, &flags); | |
2218 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2219 | || unlikely(cpu_is_offline(dest_cpu))) | |
2220 | goto out; | |
2221 | ||
2222 | /* force the process onto the specified CPU */ | |
2223 | if (migrate_task(p, dest_cpu, &req)) { | |
2224 | /* Need to wait for migration thread (might exit: take ref). */ | |
2225 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2226 | |
1da177e4 LT |
2227 | get_task_struct(mt); |
2228 | task_rq_unlock(rq, &flags); | |
2229 | wake_up_process(mt); | |
2230 | put_task_struct(mt); | |
2231 | wait_for_completion(&req.done); | |
36c8b586 | 2232 | |
1da177e4 LT |
2233 | return; |
2234 | } | |
2235 | out: | |
2236 | task_rq_unlock(rq, &flags); | |
2237 | } | |
2238 | ||
2239 | /* | |
476d139c NP |
2240 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2241 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2242 | */ |
2243 | void sched_exec(void) | |
2244 | { | |
1da177e4 | 2245 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2246 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2247 | put_cpu(); |
476d139c NP |
2248 | if (new_cpu != this_cpu) |
2249 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2250 | } |
2251 | ||
2252 | /* | |
2253 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2254 | * Both runqueues must be locked. | |
2255 | */ | |
dd41f596 IM |
2256 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2257 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2258 | { |
2e1cb74a | 2259 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2260 | set_task_cpu(p, this_cpu); |
dd41f596 | 2261 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2262 | /* |
2263 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2264 | * to be always true for them. | |
2265 | */ | |
dd41f596 | 2266 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2267 | } |
2268 | ||
2269 | /* | |
2270 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2271 | */ | |
858119e1 | 2272 | static |
70b97a7f | 2273 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2274 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2275 | int *all_pinned) |
1da177e4 LT |
2276 | { |
2277 | /* | |
2278 | * We do not migrate tasks that are: | |
2279 | * 1) running (obviously), or | |
2280 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2281 | * 3) are cache-hot on their current CPU. | |
2282 | */ | |
cc367732 IM |
2283 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2284 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2285 | return 0; |
cc367732 | 2286 | } |
81026794 NP |
2287 | *all_pinned = 0; |
2288 | ||
cc367732 IM |
2289 | if (task_running(rq, p)) { |
2290 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2291 | return 0; |
cc367732 | 2292 | } |
1da177e4 | 2293 | |
da84d961 IM |
2294 | /* |
2295 | * Aggressive migration if: | |
2296 | * 1) task is cache cold, or | |
2297 | * 2) too many balance attempts have failed. | |
2298 | */ | |
2299 | ||
6bc1665b IM |
2300 | if (!task_hot(p, rq->clock, sd) || |
2301 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2302 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2303 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2304 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2305 | schedstat_inc(p, se.nr_forced_migrations); |
2306 | } | |
da84d961 IM |
2307 | #endif |
2308 | return 1; | |
2309 | } | |
2310 | ||
cc367732 IM |
2311 | if (task_hot(p, rq->clock, sd)) { |
2312 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2313 | return 0; |
cc367732 | 2314 | } |
1da177e4 LT |
2315 | return 1; |
2316 | } | |
2317 | ||
e1d1484f PW |
2318 | static unsigned long |
2319 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2320 | unsigned long max_load_move, struct sched_domain *sd, | |
2321 | enum cpu_idle_type idle, int *all_pinned, | |
2322 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2323 | { |
b82d9fdd | 2324 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2325 | struct task_struct *p; |
2326 | long rem_load_move = max_load_move; | |
1da177e4 | 2327 | |
e1d1484f | 2328 | if (max_load_move == 0) |
1da177e4 LT |
2329 | goto out; |
2330 | ||
81026794 NP |
2331 | pinned = 1; |
2332 | ||
1da177e4 | 2333 | /* |
dd41f596 | 2334 | * Start the load-balancing iterator: |
1da177e4 | 2335 | */ |
dd41f596 IM |
2336 | p = iterator->start(iterator->arg); |
2337 | next: | |
b82d9fdd | 2338 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2339 | goto out; |
50ddd969 | 2340 | /* |
b82d9fdd | 2341 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2342 | * skip a task if it will be the highest priority task (i.e. smallest |
2343 | * prio value) on its new queue regardless of its load weight | |
2344 | */ | |
dd41f596 IM |
2345 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2346 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2347 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2348 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2349 | p = iterator->next(iterator->arg); |
2350 | goto next; | |
1da177e4 LT |
2351 | } |
2352 | ||
dd41f596 | 2353 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2354 | pulled++; |
dd41f596 | 2355 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2356 | |
2dd73a4f | 2357 | /* |
b82d9fdd | 2358 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2359 | */ |
e1d1484f | 2360 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2361 | if (p->prio < *this_best_prio) |
2362 | *this_best_prio = p->prio; | |
dd41f596 IM |
2363 | p = iterator->next(iterator->arg); |
2364 | goto next; | |
1da177e4 LT |
2365 | } |
2366 | out: | |
2367 | /* | |
e1d1484f | 2368 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2369 | * so we can safely collect pull_task() stats here rather than |
2370 | * inside pull_task(). | |
2371 | */ | |
2372 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2373 | |
2374 | if (all_pinned) | |
2375 | *all_pinned = pinned; | |
e1d1484f PW |
2376 | |
2377 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2378 | } |
2379 | ||
dd41f596 | 2380 | /* |
43010659 PW |
2381 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2382 | * this_rq, as part of a balancing operation within domain "sd". | |
2383 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2384 | * |
2385 | * Called with both runqueues locked. | |
2386 | */ | |
2387 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2388 | unsigned long max_load_move, |
dd41f596 IM |
2389 | struct sched_domain *sd, enum cpu_idle_type idle, |
2390 | int *all_pinned) | |
2391 | { | |
5522d5d5 | 2392 | const struct sched_class *class = sched_class_highest; |
43010659 | 2393 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2394 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2395 | |
2396 | do { | |
43010659 PW |
2397 | total_load_moved += |
2398 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2399 | max_load_move - total_load_moved, |
a4ac01c3 | 2400 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2401 | class = class->next; |
43010659 | 2402 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2403 | |
43010659 PW |
2404 | return total_load_moved > 0; |
2405 | } | |
2406 | ||
e1d1484f PW |
2407 | static int |
2408 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2409 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2410 | struct rq_iterator *iterator) | |
2411 | { | |
2412 | struct task_struct *p = iterator->start(iterator->arg); | |
2413 | int pinned = 0; | |
2414 | ||
2415 | while (p) { | |
2416 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2417 | pull_task(busiest, p, this_rq, this_cpu); | |
2418 | /* | |
2419 | * Right now, this is only the second place pull_task() | |
2420 | * is called, so we can safely collect pull_task() | |
2421 | * stats here rather than inside pull_task(). | |
2422 | */ | |
2423 | schedstat_inc(sd, lb_gained[idle]); | |
2424 | ||
2425 | return 1; | |
2426 | } | |
2427 | p = iterator->next(iterator->arg); | |
2428 | } | |
2429 | ||
2430 | return 0; | |
2431 | } | |
2432 | ||
43010659 PW |
2433 | /* |
2434 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2435 | * part of active balancing operations within "domain". | |
2436 | * Returns 1 if successful and 0 otherwise. | |
2437 | * | |
2438 | * Called with both runqueues locked. | |
2439 | */ | |
2440 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2441 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2442 | { | |
5522d5d5 | 2443 | const struct sched_class *class; |
43010659 PW |
2444 | |
2445 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2446 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2447 | return 1; |
2448 | ||
2449 | return 0; | |
dd41f596 IM |
2450 | } |
2451 | ||
1da177e4 LT |
2452 | /* |
2453 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2454 | * domain. It calculates and returns the amount of weighted load which |
2455 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2456 | */ |
2457 | static struct sched_group * | |
2458 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 IM |
2459 | unsigned long *imbalance, enum cpu_idle_type idle, |
2460 | int *sd_idle, cpumask_t *cpus, int *balance) | |
1da177e4 LT |
2461 | { |
2462 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2463 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2464 | unsigned long max_pull; |
2dd73a4f PW |
2465 | unsigned long busiest_load_per_task, busiest_nr_running; |
2466 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2467 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2468 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2469 | int power_savings_balance = 1; | |
2470 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2471 | unsigned long min_nr_running = ULONG_MAX; | |
2472 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2473 | #endif | |
1da177e4 LT |
2474 | |
2475 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2476 | busiest_load_per_task = busiest_nr_running = 0; |
2477 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2478 | if (idle == CPU_NOT_IDLE) |
7897986b | 2479 | load_idx = sd->busy_idx; |
d15bcfdb | 2480 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2481 | load_idx = sd->newidle_idx; |
2482 | else | |
2483 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2484 | |
2485 | do { | |
908a7c1b | 2486 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2487 | int local_group; |
2488 | int i; | |
908a7c1b | 2489 | int __group_imb = 0; |
783609c6 | 2490 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2491 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2492 | |
2493 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2494 | ||
783609c6 SS |
2495 | if (local_group) |
2496 | balance_cpu = first_cpu(group->cpumask); | |
2497 | ||
1da177e4 | 2498 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2499 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2500 | max_cpu_load = 0; |
2501 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2502 | |
2503 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2504 | struct rq *rq; |
2505 | ||
2506 | if (!cpu_isset(i, *cpus)) | |
2507 | continue; | |
2508 | ||
2509 | rq = cpu_rq(i); | |
2dd73a4f | 2510 | |
9439aab8 | 2511 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
2512 | *sd_idle = 0; |
2513 | ||
1da177e4 | 2514 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
2515 | if (local_group) { |
2516 | if (idle_cpu(i) && !first_idle_cpu) { | |
2517 | first_idle_cpu = 1; | |
2518 | balance_cpu = i; | |
2519 | } | |
2520 | ||
a2000572 | 2521 | load = target_load(i, load_idx); |
908a7c1b | 2522 | } else { |
a2000572 | 2523 | load = source_load(i, load_idx); |
908a7c1b KC |
2524 | if (load > max_cpu_load) |
2525 | max_cpu_load = load; | |
2526 | if (min_cpu_load > load) | |
2527 | min_cpu_load = load; | |
2528 | } | |
1da177e4 LT |
2529 | |
2530 | avg_load += load; | |
2dd73a4f | 2531 | sum_nr_running += rq->nr_running; |
dd41f596 | 2532 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
2533 | } |
2534 | ||
783609c6 SS |
2535 | /* |
2536 | * First idle cpu or the first cpu(busiest) in this sched group | |
2537 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
2538 | * domains. In the newly idle case, we will allow all the cpu's |
2539 | * to do the newly idle load balance. | |
783609c6 | 2540 | */ |
9439aab8 SS |
2541 | if (idle != CPU_NEWLY_IDLE && local_group && |
2542 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
2543 | *balance = 0; |
2544 | goto ret; | |
2545 | } | |
2546 | ||
1da177e4 | 2547 | total_load += avg_load; |
5517d86b | 2548 | total_pwr += group->__cpu_power; |
1da177e4 LT |
2549 | |
2550 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2551 | avg_load = sg_div_cpu_power(group, |
2552 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 2553 | |
908a7c1b KC |
2554 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
2555 | __group_imb = 1; | |
2556 | ||
5517d86b | 2557 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 2558 | |
1da177e4 LT |
2559 | if (local_group) { |
2560 | this_load = avg_load; | |
2561 | this = group; | |
2dd73a4f PW |
2562 | this_nr_running = sum_nr_running; |
2563 | this_load_per_task = sum_weighted_load; | |
2564 | } else if (avg_load > max_load && | |
908a7c1b | 2565 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
2566 | max_load = avg_load; |
2567 | busiest = group; | |
2dd73a4f PW |
2568 | busiest_nr_running = sum_nr_running; |
2569 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 2570 | group_imb = __group_imb; |
1da177e4 | 2571 | } |
5c45bf27 SS |
2572 | |
2573 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
2574 | /* | |
2575 | * Busy processors will not participate in power savings | |
2576 | * balance. | |
2577 | */ | |
dd41f596 IM |
2578 | if (idle == CPU_NOT_IDLE || |
2579 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
2580 | goto group_next; | |
5c45bf27 SS |
2581 | |
2582 | /* | |
2583 | * If the local group is idle or completely loaded | |
2584 | * no need to do power savings balance at this domain | |
2585 | */ | |
2586 | if (local_group && (this_nr_running >= group_capacity || | |
2587 | !this_nr_running)) | |
2588 | power_savings_balance = 0; | |
2589 | ||
dd41f596 | 2590 | /* |
5c45bf27 SS |
2591 | * If a group is already running at full capacity or idle, |
2592 | * don't include that group in power savings calculations | |
dd41f596 IM |
2593 | */ |
2594 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 2595 | || !sum_nr_running) |
dd41f596 | 2596 | goto group_next; |
5c45bf27 | 2597 | |
dd41f596 | 2598 | /* |
5c45bf27 | 2599 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
2600 | * This is the group from where we need to pick up the load |
2601 | * for saving power | |
2602 | */ | |
2603 | if ((sum_nr_running < min_nr_running) || | |
2604 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
2605 | first_cpu(group->cpumask) < |
2606 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
2607 | group_min = group; |
2608 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
2609 | min_load_per_task = sum_weighted_load / |
2610 | sum_nr_running; | |
dd41f596 | 2611 | } |
5c45bf27 | 2612 | |
dd41f596 | 2613 | /* |
5c45bf27 | 2614 | * Calculate the group which is almost near its |
dd41f596 IM |
2615 | * capacity but still has some space to pick up some load |
2616 | * from other group and save more power | |
2617 | */ | |
2618 | if (sum_nr_running <= group_capacity - 1) { | |
2619 | if (sum_nr_running > leader_nr_running || | |
2620 | (sum_nr_running == leader_nr_running && | |
2621 | first_cpu(group->cpumask) > | |
2622 | first_cpu(group_leader->cpumask))) { | |
2623 | group_leader = group; | |
2624 | leader_nr_running = sum_nr_running; | |
2625 | } | |
48f24c4d | 2626 | } |
5c45bf27 SS |
2627 | group_next: |
2628 | #endif | |
1da177e4 LT |
2629 | group = group->next; |
2630 | } while (group != sd->groups); | |
2631 | ||
2dd73a4f | 2632 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
2633 | goto out_balanced; |
2634 | ||
2635 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
2636 | ||
2637 | if (this_load >= avg_load || | |
2638 | 100*max_load <= sd->imbalance_pct*this_load) | |
2639 | goto out_balanced; | |
2640 | ||
2dd73a4f | 2641 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
2642 | if (group_imb) |
2643 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
2644 | ||
1da177e4 LT |
2645 | /* |
2646 | * We're trying to get all the cpus to the average_load, so we don't | |
2647 | * want to push ourselves above the average load, nor do we wish to | |
2648 | * reduce the max loaded cpu below the average load, as either of these | |
2649 | * actions would just result in more rebalancing later, and ping-pong | |
2650 | * tasks around. Thus we look for the minimum possible imbalance. | |
2651 | * Negative imbalances (*we* are more loaded than anyone else) will | |
2652 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 2653 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
2654 | * appear as very large values with unsigned longs. |
2655 | */ | |
2dd73a4f PW |
2656 | if (max_load <= busiest_load_per_task) |
2657 | goto out_balanced; | |
2658 | ||
2659 | /* | |
2660 | * In the presence of smp nice balancing, certain scenarios can have | |
2661 | * max load less than avg load(as we skip the groups at or below | |
2662 | * its cpu_power, while calculating max_load..) | |
2663 | */ | |
2664 | if (max_load < avg_load) { | |
2665 | *imbalance = 0; | |
2666 | goto small_imbalance; | |
2667 | } | |
0c117f1b SS |
2668 | |
2669 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 2670 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 2671 | |
1da177e4 | 2672 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
2673 | *imbalance = min(max_pull * busiest->__cpu_power, |
2674 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
2675 | / SCHED_LOAD_SCALE; |
2676 | ||
2dd73a4f PW |
2677 | /* |
2678 | * if *imbalance is less than the average load per runnable task | |
2679 | * there is no gaurantee that any tasks will be moved so we'll have | |
2680 | * a think about bumping its value to force at least one task to be | |
2681 | * moved | |
2682 | */ | |
7fd0d2dd | 2683 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 2684 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
2685 | unsigned int imbn; |
2686 | ||
2687 | small_imbalance: | |
2688 | pwr_move = pwr_now = 0; | |
2689 | imbn = 2; | |
2690 | if (this_nr_running) { | |
2691 | this_load_per_task /= this_nr_running; | |
2692 | if (busiest_load_per_task > this_load_per_task) | |
2693 | imbn = 1; | |
2694 | } else | |
2695 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 2696 | |
dd41f596 IM |
2697 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
2698 | busiest_load_per_task * imbn) { | |
2dd73a4f | 2699 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
2700 | return busiest; |
2701 | } | |
2702 | ||
2703 | /* | |
2704 | * OK, we don't have enough imbalance to justify moving tasks, | |
2705 | * however we may be able to increase total CPU power used by | |
2706 | * moving them. | |
2707 | */ | |
2708 | ||
5517d86b ED |
2709 | pwr_now += busiest->__cpu_power * |
2710 | min(busiest_load_per_task, max_load); | |
2711 | pwr_now += this->__cpu_power * | |
2712 | min(this_load_per_task, this_load); | |
1da177e4 LT |
2713 | pwr_now /= SCHED_LOAD_SCALE; |
2714 | ||
2715 | /* Amount of load we'd subtract */ | |
5517d86b ED |
2716 | tmp = sg_div_cpu_power(busiest, |
2717 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 2718 | if (max_load > tmp) |
5517d86b | 2719 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 2720 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
2721 | |
2722 | /* Amount of load we'd add */ | |
5517d86b | 2723 | if (max_load * busiest->__cpu_power < |
33859f7f | 2724 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
2725 | tmp = sg_div_cpu_power(this, |
2726 | max_load * busiest->__cpu_power); | |
1da177e4 | 2727 | else |
5517d86b ED |
2728 | tmp = sg_div_cpu_power(this, |
2729 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
2730 | pwr_move += this->__cpu_power * | |
2731 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
2732 | pwr_move /= SCHED_LOAD_SCALE; |
2733 | ||
2734 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
2735 | if (pwr_move > pwr_now) |
2736 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
2737 | } |
2738 | ||
1da177e4 LT |
2739 | return busiest; |
2740 | ||
2741 | out_balanced: | |
5c45bf27 | 2742 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 2743 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 2744 | goto ret; |
1da177e4 | 2745 | |
5c45bf27 SS |
2746 | if (this == group_leader && group_leader != group_min) { |
2747 | *imbalance = min_load_per_task; | |
2748 | return group_min; | |
2749 | } | |
5c45bf27 | 2750 | #endif |
783609c6 | 2751 | ret: |
1da177e4 LT |
2752 | *imbalance = 0; |
2753 | return NULL; | |
2754 | } | |
2755 | ||
2756 | /* | |
2757 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
2758 | */ | |
70b97a7f | 2759 | static struct rq * |
d15bcfdb | 2760 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
0a2966b4 | 2761 | unsigned long imbalance, cpumask_t *cpus) |
1da177e4 | 2762 | { |
70b97a7f | 2763 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 2764 | unsigned long max_load = 0; |
1da177e4 LT |
2765 | int i; |
2766 | ||
2767 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 2768 | unsigned long wl; |
0a2966b4 CL |
2769 | |
2770 | if (!cpu_isset(i, *cpus)) | |
2771 | continue; | |
2772 | ||
48f24c4d | 2773 | rq = cpu_rq(i); |
dd41f596 | 2774 | wl = weighted_cpuload(i); |
2dd73a4f | 2775 | |
dd41f596 | 2776 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 2777 | continue; |
1da177e4 | 2778 | |
dd41f596 IM |
2779 | if (wl > max_load) { |
2780 | max_load = wl; | |
48f24c4d | 2781 | busiest = rq; |
1da177e4 LT |
2782 | } |
2783 | } | |
2784 | ||
2785 | return busiest; | |
2786 | } | |
2787 | ||
77391d71 NP |
2788 | /* |
2789 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
2790 | * so long as it is large enough. | |
2791 | */ | |
2792 | #define MAX_PINNED_INTERVAL 512 | |
2793 | ||
1da177e4 LT |
2794 | /* |
2795 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2796 | * tasks if there is an imbalance. | |
1da177e4 | 2797 | */ |
70b97a7f | 2798 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 2799 | struct sched_domain *sd, enum cpu_idle_type idle, |
783609c6 | 2800 | int *balance) |
1da177e4 | 2801 | { |
43010659 | 2802 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 2803 | struct sched_group *group; |
1da177e4 | 2804 | unsigned long imbalance; |
70b97a7f | 2805 | struct rq *busiest; |
0a2966b4 | 2806 | cpumask_t cpus = CPU_MASK_ALL; |
fe2eea3f | 2807 | unsigned long flags; |
5969fe06 | 2808 | |
89c4710e SS |
2809 | /* |
2810 | * When power savings policy is enabled for the parent domain, idle | |
2811 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 2812 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 2813 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 2814 | */ |
d15bcfdb | 2815 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2816 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2817 | sd_idle = 1; |
1da177e4 | 2818 | |
2d72376b | 2819 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 2820 | |
0a2966b4 CL |
2821 | redo: |
2822 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
783609c6 SS |
2823 | &cpus, balance); |
2824 | ||
06066714 | 2825 | if (*balance == 0) |
783609c6 | 2826 | goto out_balanced; |
783609c6 | 2827 | |
1da177e4 LT |
2828 | if (!group) { |
2829 | schedstat_inc(sd, lb_nobusyg[idle]); | |
2830 | goto out_balanced; | |
2831 | } | |
2832 | ||
0a2966b4 | 2833 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
1da177e4 LT |
2834 | if (!busiest) { |
2835 | schedstat_inc(sd, lb_nobusyq[idle]); | |
2836 | goto out_balanced; | |
2837 | } | |
2838 | ||
db935dbd | 2839 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
2840 | |
2841 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
2842 | ||
43010659 | 2843 | ld_moved = 0; |
1da177e4 LT |
2844 | if (busiest->nr_running > 1) { |
2845 | /* | |
2846 | * Attempt to move tasks. If find_busiest_group has found | |
2847 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 2848 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
2849 | * correctly treated as an imbalance. |
2850 | */ | |
fe2eea3f | 2851 | local_irq_save(flags); |
e17224bf | 2852 | double_rq_lock(this_rq, busiest); |
43010659 | 2853 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 2854 | imbalance, sd, idle, &all_pinned); |
e17224bf | 2855 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 2856 | local_irq_restore(flags); |
81026794 | 2857 | |
46cb4b7c SS |
2858 | /* |
2859 | * some other cpu did the load balance for us. | |
2860 | */ | |
43010659 | 2861 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
2862 | resched_cpu(this_cpu); |
2863 | ||
81026794 | 2864 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 CL |
2865 | if (unlikely(all_pinned)) { |
2866 | cpu_clear(cpu_of(busiest), cpus); | |
2867 | if (!cpus_empty(cpus)) | |
2868 | goto redo; | |
81026794 | 2869 | goto out_balanced; |
0a2966b4 | 2870 | } |
1da177e4 | 2871 | } |
81026794 | 2872 | |
43010659 | 2873 | if (!ld_moved) { |
1da177e4 LT |
2874 | schedstat_inc(sd, lb_failed[idle]); |
2875 | sd->nr_balance_failed++; | |
2876 | ||
2877 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 2878 | |
fe2eea3f | 2879 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
2880 | |
2881 | /* don't kick the migration_thread, if the curr | |
2882 | * task on busiest cpu can't be moved to this_cpu | |
2883 | */ | |
2884 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 2885 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
2886 | all_pinned = 1; |
2887 | goto out_one_pinned; | |
2888 | } | |
2889 | ||
1da177e4 LT |
2890 | if (!busiest->active_balance) { |
2891 | busiest->active_balance = 1; | |
2892 | busiest->push_cpu = this_cpu; | |
81026794 | 2893 | active_balance = 1; |
1da177e4 | 2894 | } |
fe2eea3f | 2895 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 2896 | if (active_balance) |
1da177e4 LT |
2897 | wake_up_process(busiest->migration_thread); |
2898 | ||
2899 | /* | |
2900 | * We've kicked active balancing, reset the failure | |
2901 | * counter. | |
2902 | */ | |
39507451 | 2903 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 2904 | } |
81026794 | 2905 | } else |
1da177e4 LT |
2906 | sd->nr_balance_failed = 0; |
2907 | ||
81026794 | 2908 | if (likely(!active_balance)) { |
1da177e4 LT |
2909 | /* We were unbalanced, so reset the balancing interval */ |
2910 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
2911 | } else { |
2912 | /* | |
2913 | * If we've begun active balancing, start to back off. This | |
2914 | * case may not be covered by the all_pinned logic if there | |
2915 | * is only 1 task on the busy runqueue (because we don't call | |
2916 | * move_tasks). | |
2917 | */ | |
2918 | if (sd->balance_interval < sd->max_interval) | |
2919 | sd->balance_interval *= 2; | |
1da177e4 LT |
2920 | } |
2921 | ||
43010659 | 2922 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2923 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2924 | return -1; |
43010659 | 2925 | return ld_moved; |
1da177e4 LT |
2926 | |
2927 | out_balanced: | |
1da177e4 LT |
2928 | schedstat_inc(sd, lb_balanced[idle]); |
2929 | ||
16cfb1c0 | 2930 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
2931 | |
2932 | out_one_pinned: | |
1da177e4 | 2933 | /* tune up the balancing interval */ |
77391d71 NP |
2934 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
2935 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
2936 | sd->balance_interval *= 2; |
2937 | ||
48f24c4d | 2938 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 2939 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 2940 | return -1; |
1da177e4 LT |
2941 | return 0; |
2942 | } | |
2943 | ||
2944 | /* | |
2945 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
2946 | * tasks if there is an imbalance. | |
2947 | * | |
d15bcfdb | 2948 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
2949 | * this_rq is locked. |
2950 | */ | |
48f24c4d | 2951 | static int |
70b97a7f | 2952 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
2953 | { |
2954 | struct sched_group *group; | |
70b97a7f | 2955 | struct rq *busiest = NULL; |
1da177e4 | 2956 | unsigned long imbalance; |
43010659 | 2957 | int ld_moved = 0; |
5969fe06 | 2958 | int sd_idle = 0; |
969bb4e4 | 2959 | int all_pinned = 0; |
0a2966b4 | 2960 | cpumask_t cpus = CPU_MASK_ALL; |
5969fe06 | 2961 | |
89c4710e SS |
2962 | /* |
2963 | * When power savings policy is enabled for the parent domain, idle | |
2964 | * sibling can pick up load irrespective of busy siblings. In this case, | |
2965 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 2966 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
2967 | */ |
2968 | if (sd->flags & SD_SHARE_CPUPOWER && | |
2969 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 2970 | sd_idle = 1; |
1da177e4 | 2971 | |
2d72376b | 2972 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 2973 | redo: |
d15bcfdb | 2974 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
783609c6 | 2975 | &sd_idle, &cpus, NULL); |
1da177e4 | 2976 | if (!group) { |
d15bcfdb | 2977 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2978 | goto out_balanced; |
1da177e4 LT |
2979 | } |
2980 | ||
d15bcfdb | 2981 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
0a2966b4 | 2982 | &cpus); |
db935dbd | 2983 | if (!busiest) { |
d15bcfdb | 2984 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 2985 | goto out_balanced; |
1da177e4 LT |
2986 | } |
2987 | ||
db935dbd NP |
2988 | BUG_ON(busiest == this_rq); |
2989 | ||
d15bcfdb | 2990 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 2991 | |
43010659 | 2992 | ld_moved = 0; |
d6d5cfaf NP |
2993 | if (busiest->nr_running > 1) { |
2994 | /* Attempt to move tasks */ | |
2995 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
2996 | /* this_rq->clock is already updated */ |
2997 | update_rq_clock(busiest); | |
43010659 | 2998 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
2999 | imbalance, sd, CPU_NEWLY_IDLE, |
3000 | &all_pinned); | |
d6d5cfaf | 3001 | spin_unlock(&busiest->lock); |
0a2966b4 | 3002 | |
969bb4e4 | 3003 | if (unlikely(all_pinned)) { |
0a2966b4 CL |
3004 | cpu_clear(cpu_of(busiest), cpus); |
3005 | if (!cpus_empty(cpus)) | |
3006 | goto redo; | |
3007 | } | |
d6d5cfaf NP |
3008 | } |
3009 | ||
43010659 | 3010 | if (!ld_moved) { |
d15bcfdb | 3011 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3012 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3013 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3014 | return -1; |
3015 | } else | |
16cfb1c0 | 3016 | sd->nr_balance_failed = 0; |
1da177e4 | 3017 | |
43010659 | 3018 | return ld_moved; |
16cfb1c0 NP |
3019 | |
3020 | out_balanced: | |
d15bcfdb | 3021 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3022 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3023 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3024 | return -1; |
16cfb1c0 | 3025 | sd->nr_balance_failed = 0; |
48f24c4d | 3026 | |
16cfb1c0 | 3027 | return 0; |
1da177e4 LT |
3028 | } |
3029 | ||
3030 | /* | |
3031 | * idle_balance is called by schedule() if this_cpu is about to become | |
3032 | * idle. Attempts to pull tasks from other CPUs. | |
3033 | */ | |
70b97a7f | 3034 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3035 | { |
3036 | struct sched_domain *sd; | |
dd41f596 IM |
3037 | int pulled_task = -1; |
3038 | unsigned long next_balance = jiffies + HZ; | |
1da177e4 LT |
3039 | |
3040 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3041 | unsigned long interval; |
3042 | ||
3043 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3044 | continue; | |
3045 | ||
3046 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3047 | /* If we've pulled tasks over stop searching: */ |
1bd77f2d | 3048 | pulled_task = load_balance_newidle(this_cpu, |
92c4ca5c CL |
3049 | this_rq, sd); |
3050 | ||
3051 | interval = msecs_to_jiffies(sd->balance_interval); | |
3052 | if (time_after(next_balance, sd->last_balance + interval)) | |
3053 | next_balance = sd->last_balance + interval; | |
3054 | if (pulled_task) | |
3055 | break; | |
1da177e4 | 3056 | } |
dd41f596 | 3057 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3058 | /* |
3059 | * We are going idle. next_balance may be set based on | |
3060 | * a busy processor. So reset next_balance. | |
3061 | */ | |
3062 | this_rq->next_balance = next_balance; | |
dd41f596 | 3063 | } |
1da177e4 LT |
3064 | } |
3065 | ||
3066 | /* | |
3067 | * active_load_balance is run by migration threads. It pushes running tasks | |
3068 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3069 | * running on each physical CPU where possible, and avoids physical / | |
3070 | * logical imbalances. | |
3071 | * | |
3072 | * Called with busiest_rq locked. | |
3073 | */ | |
70b97a7f | 3074 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3075 | { |
39507451 | 3076 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3077 | struct sched_domain *sd; |
3078 | struct rq *target_rq; | |
39507451 | 3079 | |
48f24c4d | 3080 | /* Is there any task to move? */ |
39507451 | 3081 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3082 | return; |
3083 | ||
3084 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3085 | |
3086 | /* | |
39507451 | 3087 | * This condition is "impossible", if it occurs |
41a2d6cf | 3088 | * we need to fix it. Originally reported by |
39507451 | 3089 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3090 | */ |
39507451 | 3091 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3092 | |
39507451 NP |
3093 | /* move a task from busiest_rq to target_rq */ |
3094 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3095 | update_rq_clock(busiest_rq); |
3096 | update_rq_clock(target_rq); | |
39507451 NP |
3097 | |
3098 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3099 | for_each_domain(target_cpu, sd) { |
39507451 | 3100 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3101 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3102 | break; |
c96d145e | 3103 | } |
39507451 | 3104 | |
48f24c4d | 3105 | if (likely(sd)) { |
2d72376b | 3106 | schedstat_inc(sd, alb_count); |
39507451 | 3107 | |
43010659 PW |
3108 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3109 | sd, CPU_IDLE)) | |
48f24c4d IM |
3110 | schedstat_inc(sd, alb_pushed); |
3111 | else | |
3112 | schedstat_inc(sd, alb_failed); | |
3113 | } | |
39507451 | 3114 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3115 | } |
3116 | ||
46cb4b7c SS |
3117 | #ifdef CONFIG_NO_HZ |
3118 | static struct { | |
3119 | atomic_t load_balancer; | |
41a2d6cf | 3120 | cpumask_t cpu_mask; |
46cb4b7c SS |
3121 | } nohz ____cacheline_aligned = { |
3122 | .load_balancer = ATOMIC_INIT(-1), | |
3123 | .cpu_mask = CPU_MASK_NONE, | |
3124 | }; | |
3125 | ||
7835b98b | 3126 | /* |
46cb4b7c SS |
3127 | * This routine will try to nominate the ilb (idle load balancing) |
3128 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3129 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3130 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3131 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3132 | * arrives... | |
3133 | * | |
3134 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3135 | * for idle load balancing. ilb owner will still be part of | |
3136 | * nohz.cpu_mask.. | |
7835b98b | 3137 | * |
46cb4b7c SS |
3138 | * While stopping the tick, this cpu will become the ilb owner if there |
3139 | * is no other owner. And will be the owner till that cpu becomes busy | |
3140 | * or if all cpus in the system stop their ticks at which point | |
3141 | * there is no need for ilb owner. | |
3142 | * | |
3143 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3144 | * next busy scheduler_tick() | |
3145 | */ | |
3146 | int select_nohz_load_balancer(int stop_tick) | |
3147 | { | |
3148 | int cpu = smp_processor_id(); | |
3149 | ||
3150 | if (stop_tick) { | |
3151 | cpu_set(cpu, nohz.cpu_mask); | |
3152 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3153 | ||
3154 | /* | |
3155 | * If we are going offline and still the leader, give up! | |
3156 | */ | |
3157 | if (cpu_is_offline(cpu) && | |
3158 | atomic_read(&nohz.load_balancer) == cpu) { | |
3159 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3160 | BUG(); | |
3161 | return 0; | |
3162 | } | |
3163 | ||
3164 | /* time for ilb owner also to sleep */ | |
3165 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3166 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3167 | atomic_set(&nohz.load_balancer, -1); | |
3168 | return 0; | |
3169 | } | |
3170 | ||
3171 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3172 | /* make me the ilb owner */ | |
3173 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3174 | return 1; | |
3175 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3176 | return 1; | |
3177 | } else { | |
3178 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3179 | return 0; | |
3180 | ||
3181 | cpu_clear(cpu, nohz.cpu_mask); | |
3182 | ||
3183 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3184 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3185 | BUG(); | |
3186 | } | |
3187 | return 0; | |
3188 | } | |
3189 | #endif | |
3190 | ||
3191 | static DEFINE_SPINLOCK(balancing); | |
3192 | ||
3193 | /* | |
7835b98b CL |
3194 | * It checks each scheduling domain to see if it is due to be balanced, |
3195 | * and initiates a balancing operation if so. | |
3196 | * | |
3197 | * Balancing parameters are set up in arch_init_sched_domains. | |
3198 | */ | |
a9957449 | 3199 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3200 | { |
46cb4b7c SS |
3201 | int balance = 1; |
3202 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3203 | unsigned long interval; |
3204 | struct sched_domain *sd; | |
46cb4b7c | 3205 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3206 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3207 | int update_next_balance = 0; |
1da177e4 | 3208 | |
46cb4b7c | 3209 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3210 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3211 | continue; | |
3212 | ||
3213 | interval = sd->balance_interval; | |
d15bcfdb | 3214 | if (idle != CPU_IDLE) |
1da177e4 LT |
3215 | interval *= sd->busy_factor; |
3216 | ||
3217 | /* scale ms to jiffies */ | |
3218 | interval = msecs_to_jiffies(interval); | |
3219 | if (unlikely(!interval)) | |
3220 | interval = 1; | |
dd41f596 IM |
3221 | if (interval > HZ*NR_CPUS/10) |
3222 | interval = HZ*NR_CPUS/10; | |
3223 | ||
1da177e4 | 3224 | |
08c183f3 CL |
3225 | if (sd->flags & SD_SERIALIZE) { |
3226 | if (!spin_trylock(&balancing)) | |
3227 | goto out; | |
3228 | } | |
3229 | ||
c9819f45 | 3230 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
46cb4b7c | 3231 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
3232 | /* |
3233 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3234 | * longer idle, or one of our SMT siblings is |
3235 | * not idle. | |
3236 | */ | |
d15bcfdb | 3237 | idle = CPU_NOT_IDLE; |
1da177e4 | 3238 | } |
1bd77f2d | 3239 | sd->last_balance = jiffies; |
1da177e4 | 3240 | } |
08c183f3 CL |
3241 | if (sd->flags & SD_SERIALIZE) |
3242 | spin_unlock(&balancing); | |
3243 | out: | |
f549da84 | 3244 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3245 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3246 | update_next_balance = 1; |
3247 | } | |
783609c6 SS |
3248 | |
3249 | /* | |
3250 | * Stop the load balance at this level. There is another | |
3251 | * CPU in our sched group which is doing load balancing more | |
3252 | * actively. | |
3253 | */ | |
3254 | if (!balance) | |
3255 | break; | |
1da177e4 | 3256 | } |
f549da84 SS |
3257 | |
3258 | /* | |
3259 | * next_balance will be updated only when there is a need. | |
3260 | * When the cpu is attached to null domain for ex, it will not be | |
3261 | * updated. | |
3262 | */ | |
3263 | if (likely(update_next_balance)) | |
3264 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3265 | } |
3266 | ||
3267 | /* | |
3268 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3269 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3270 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3271 | */ | |
3272 | static void run_rebalance_domains(struct softirq_action *h) | |
3273 | { | |
dd41f596 IM |
3274 | int this_cpu = smp_processor_id(); |
3275 | struct rq *this_rq = cpu_rq(this_cpu); | |
3276 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3277 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3278 | |
dd41f596 | 3279 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3280 | |
3281 | #ifdef CONFIG_NO_HZ | |
3282 | /* | |
3283 | * If this cpu is the owner for idle load balancing, then do the | |
3284 | * balancing on behalf of the other idle cpus whose ticks are | |
3285 | * stopped. | |
3286 | */ | |
dd41f596 IM |
3287 | if (this_rq->idle_at_tick && |
3288 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3289 | cpumask_t cpus = nohz.cpu_mask; |
3290 | struct rq *rq; | |
3291 | int balance_cpu; | |
3292 | ||
dd41f596 | 3293 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3294 | for_each_cpu_mask(balance_cpu, cpus) { |
3295 | /* | |
3296 | * If this cpu gets work to do, stop the load balancing | |
3297 | * work being done for other cpus. Next load | |
3298 | * balancing owner will pick it up. | |
3299 | */ | |
3300 | if (need_resched()) | |
3301 | break; | |
3302 | ||
de0cf899 | 3303 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3304 | |
3305 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3306 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3307 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3308 | } |
3309 | } | |
3310 | #endif | |
3311 | } | |
3312 | ||
3313 | /* | |
3314 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3315 | * | |
3316 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3317 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3318 | * if the whole system is idle. | |
3319 | */ | |
dd41f596 | 3320 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3321 | { |
46cb4b7c SS |
3322 | #ifdef CONFIG_NO_HZ |
3323 | /* | |
3324 | * If we were in the nohz mode recently and busy at the current | |
3325 | * scheduler tick, then check if we need to nominate new idle | |
3326 | * load balancer. | |
3327 | */ | |
3328 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3329 | rq->in_nohz_recently = 0; | |
3330 | ||
3331 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3332 | cpu_clear(cpu, nohz.cpu_mask); | |
3333 | atomic_set(&nohz.load_balancer, -1); | |
3334 | } | |
3335 | ||
3336 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3337 | /* | |
3338 | * simple selection for now: Nominate the | |
3339 | * first cpu in the nohz list to be the next | |
3340 | * ilb owner. | |
3341 | * | |
3342 | * TBD: Traverse the sched domains and nominate | |
3343 | * the nearest cpu in the nohz.cpu_mask. | |
3344 | */ | |
3345 | int ilb = first_cpu(nohz.cpu_mask); | |
3346 | ||
3347 | if (ilb != NR_CPUS) | |
3348 | resched_cpu(ilb); | |
3349 | } | |
3350 | } | |
3351 | ||
3352 | /* | |
3353 | * If this cpu is idle and doing idle load balancing for all the | |
3354 | * cpus with ticks stopped, is it time for that to stop? | |
3355 | */ | |
3356 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3357 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3358 | resched_cpu(cpu); | |
3359 | return; | |
3360 | } | |
3361 | ||
3362 | /* | |
3363 | * If this cpu is idle and the idle load balancing is done by | |
3364 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3365 | */ | |
3366 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3367 | cpu_isset(cpu, nohz.cpu_mask)) | |
3368 | return; | |
3369 | #endif | |
3370 | if (time_after_eq(jiffies, rq->next_balance)) | |
3371 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3372 | } |
dd41f596 IM |
3373 | |
3374 | #else /* CONFIG_SMP */ | |
3375 | ||
1da177e4 LT |
3376 | /* |
3377 | * on UP we do not need to balance between CPUs: | |
3378 | */ | |
70b97a7f | 3379 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3380 | { |
3381 | } | |
dd41f596 | 3382 | |
1da177e4 LT |
3383 | #endif |
3384 | ||
1da177e4 LT |
3385 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3386 | ||
3387 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3388 | ||
3389 | /* | |
41b86e9c IM |
3390 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3391 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3392 | */ |
41b86e9c | 3393 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3394 | { |
1da177e4 | 3395 | unsigned long flags; |
41b86e9c IM |
3396 | u64 ns, delta_exec; |
3397 | struct rq *rq; | |
48f24c4d | 3398 | |
41b86e9c IM |
3399 | rq = task_rq_lock(p, &flags); |
3400 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3401 | if (task_current(rq, p)) { |
a8e504d2 IM |
3402 | update_rq_clock(rq); |
3403 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3404 | if ((s64)delta_exec > 0) |
3405 | ns += delta_exec; | |
3406 | } | |
3407 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3408 | |
1da177e4 LT |
3409 | return ns; |
3410 | } | |
3411 | ||
1da177e4 LT |
3412 | /* |
3413 | * Account user cpu time to a process. | |
3414 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3415 | * @cputime: the cpu time spent in user space since the last update |
3416 | */ | |
3417 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3418 | { | |
3419 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3420 | cputime64_t tmp; | |
3421 | ||
3422 | p->utime = cputime_add(p->utime, cputime); | |
3423 | ||
3424 | /* Add user time to cpustat. */ | |
3425 | tmp = cputime_to_cputime64(cputime); | |
3426 | if (TASK_NICE(p) > 0) | |
3427 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3428 | else | |
3429 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3430 | } | |
3431 | ||
94886b84 LV |
3432 | /* |
3433 | * Account guest cpu time to a process. | |
3434 | * @p: the process that the cpu time gets accounted to | |
3435 | * @cputime: the cpu time spent in virtual machine since the last update | |
3436 | */ | |
f7402e03 | 3437 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3438 | { |
3439 | cputime64_t tmp; | |
3440 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3441 | ||
3442 | tmp = cputime_to_cputime64(cputime); | |
3443 | ||
3444 | p->utime = cputime_add(p->utime, cputime); | |
3445 | p->gtime = cputime_add(p->gtime, cputime); | |
3446 | ||
3447 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3448 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3449 | } | |
3450 | ||
c66f08be MN |
3451 | /* |
3452 | * Account scaled user cpu time to a process. | |
3453 | * @p: the process that the cpu time gets accounted to | |
3454 | * @cputime: the cpu time spent in user space since the last update | |
3455 | */ | |
3456 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3457 | { | |
3458 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3459 | } | |
3460 | ||
1da177e4 LT |
3461 | /* |
3462 | * Account system cpu time to a process. | |
3463 | * @p: the process that the cpu time gets accounted to | |
3464 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3465 | * @cputime: the cpu time spent in kernel space since the last update | |
3466 | */ | |
3467 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3468 | cputime_t cputime) | |
3469 | { | |
3470 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3471 | struct rq *rq = this_rq(); |
1da177e4 LT |
3472 | cputime64_t tmp; |
3473 | ||
9778385d CB |
3474 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) |
3475 | return account_guest_time(p, cputime); | |
94886b84 | 3476 | |
1da177e4 LT |
3477 | p->stime = cputime_add(p->stime, cputime); |
3478 | ||
3479 | /* Add system time to cpustat. */ | |
3480 | tmp = cputime_to_cputime64(cputime); | |
3481 | if (hardirq_count() - hardirq_offset) | |
3482 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3483 | else if (softirq_count()) | |
3484 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3485 | else if (p != rq->idle) |
1da177e4 | 3486 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3487 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3488 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3489 | else | |
3490 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3491 | /* Account for system time used */ | |
3492 | acct_update_integrals(p); | |
1da177e4 LT |
3493 | } |
3494 | ||
c66f08be MN |
3495 | /* |
3496 | * Account scaled system cpu time to a process. | |
3497 | * @p: the process that the cpu time gets accounted to | |
3498 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3499 | * @cputime: the cpu time spent in kernel space since the last update | |
3500 | */ | |
3501 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
3502 | { | |
3503 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
3504 | } | |
3505 | ||
1da177e4 LT |
3506 | /* |
3507 | * Account for involuntary wait time. | |
3508 | * @p: the process from which the cpu time has been stolen | |
3509 | * @steal: the cpu time spent in involuntary wait | |
3510 | */ | |
3511 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
3512 | { | |
3513 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3514 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 3515 | struct rq *rq = this_rq(); |
1da177e4 LT |
3516 | |
3517 | if (p == rq->idle) { | |
3518 | p->stime = cputime_add(p->stime, steal); | |
3519 | if (atomic_read(&rq->nr_iowait) > 0) | |
3520 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
3521 | else | |
3522 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 3523 | } else |
1da177e4 LT |
3524 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
3525 | } | |
3526 | ||
7835b98b CL |
3527 | /* |
3528 | * This function gets called by the timer code, with HZ frequency. | |
3529 | * We call it with interrupts disabled. | |
3530 | * | |
3531 | * It also gets called by the fork code, when changing the parent's | |
3532 | * timeslices. | |
3533 | */ | |
3534 | void scheduler_tick(void) | |
3535 | { | |
7835b98b CL |
3536 | int cpu = smp_processor_id(); |
3537 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3538 | struct task_struct *curr = rq->curr; |
529c7726 | 3539 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
dd41f596 IM |
3540 | |
3541 | spin_lock(&rq->lock); | |
546fe3c9 | 3542 | __update_rq_clock(rq); |
529c7726 IM |
3543 | /* |
3544 | * Let rq->clock advance by at least TICK_NSEC: | |
3545 | */ | |
3546 | if (unlikely(rq->clock < next_tick)) | |
3547 | rq->clock = next_tick; | |
3548 | rq->tick_timestamp = rq->clock; | |
f1a438d8 | 3549 | update_cpu_load(rq); |
dd41f596 IM |
3550 | if (curr != rq->idle) /* FIXME: needed? */ |
3551 | curr->sched_class->task_tick(rq, curr); | |
dd41f596 | 3552 | spin_unlock(&rq->lock); |
7835b98b | 3553 | |
e418e1c2 | 3554 | #ifdef CONFIG_SMP |
dd41f596 IM |
3555 | rq->idle_at_tick = idle_cpu(cpu); |
3556 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3557 | #endif |
1da177e4 LT |
3558 | } |
3559 | ||
1da177e4 LT |
3560 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
3561 | ||
3562 | void fastcall add_preempt_count(int val) | |
3563 | { | |
3564 | /* | |
3565 | * Underflow? | |
3566 | */ | |
9a11b49a IM |
3567 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3568 | return; | |
1da177e4 LT |
3569 | preempt_count() += val; |
3570 | /* | |
3571 | * Spinlock count overflowing soon? | |
3572 | */ | |
33859f7f MOS |
3573 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3574 | PREEMPT_MASK - 10); | |
1da177e4 LT |
3575 | } |
3576 | EXPORT_SYMBOL(add_preempt_count); | |
3577 | ||
3578 | void fastcall sub_preempt_count(int val) | |
3579 | { | |
3580 | /* | |
3581 | * Underflow? | |
3582 | */ | |
9a11b49a IM |
3583 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
3584 | return; | |
1da177e4 LT |
3585 | /* |
3586 | * Is the spinlock portion underflowing? | |
3587 | */ | |
9a11b49a IM |
3588 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3589 | !(preempt_count() & PREEMPT_MASK))) | |
3590 | return; | |
3591 | ||
1da177e4 LT |
3592 | preempt_count() -= val; |
3593 | } | |
3594 | EXPORT_SYMBOL(sub_preempt_count); | |
3595 | ||
3596 | #endif | |
3597 | ||
3598 | /* | |
dd41f596 | 3599 | * Print scheduling while atomic bug: |
1da177e4 | 3600 | */ |
dd41f596 | 3601 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3602 | { |
838225b4 SS |
3603 | struct pt_regs *regs = get_irq_regs(); |
3604 | ||
3605 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
3606 | prev->comm, prev->pid, preempt_count()); | |
3607 | ||
dd41f596 IM |
3608 | debug_show_held_locks(prev); |
3609 | if (irqs_disabled()) | |
3610 | print_irqtrace_events(prev); | |
838225b4 SS |
3611 | |
3612 | if (regs) | |
3613 | show_regs(regs); | |
3614 | else | |
3615 | dump_stack(); | |
dd41f596 | 3616 | } |
1da177e4 | 3617 | |
dd41f596 IM |
3618 | /* |
3619 | * Various schedule()-time debugging checks and statistics: | |
3620 | */ | |
3621 | static inline void schedule_debug(struct task_struct *prev) | |
3622 | { | |
1da177e4 | 3623 | /* |
41a2d6cf | 3624 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3625 | * schedule() atomically, we ignore that path for now. |
3626 | * Otherwise, whine if we are scheduling when we should not be. | |
3627 | */ | |
dd41f596 IM |
3628 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
3629 | __schedule_bug(prev); | |
3630 | ||
1da177e4 LT |
3631 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3632 | ||
2d72376b | 3633 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3634 | #ifdef CONFIG_SCHEDSTATS |
3635 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3636 | schedstat_inc(this_rq(), bkl_count); |
3637 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3638 | } |
3639 | #endif | |
dd41f596 IM |
3640 | } |
3641 | ||
3642 | /* | |
3643 | * Pick up the highest-prio task: | |
3644 | */ | |
3645 | static inline struct task_struct * | |
ff95f3df | 3646 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 3647 | { |
5522d5d5 | 3648 | const struct sched_class *class; |
dd41f596 | 3649 | struct task_struct *p; |
1da177e4 LT |
3650 | |
3651 | /* | |
dd41f596 IM |
3652 | * Optimization: we know that if all tasks are in |
3653 | * the fair class we can call that function directly: | |
1da177e4 | 3654 | */ |
dd41f596 | 3655 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3656 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3657 | if (likely(p)) |
3658 | return p; | |
1da177e4 LT |
3659 | } |
3660 | ||
dd41f596 IM |
3661 | class = sched_class_highest; |
3662 | for ( ; ; ) { | |
fb8d4724 | 3663 | p = class->pick_next_task(rq); |
dd41f596 IM |
3664 | if (p) |
3665 | return p; | |
3666 | /* | |
3667 | * Will never be NULL as the idle class always | |
3668 | * returns a non-NULL p: | |
3669 | */ | |
3670 | class = class->next; | |
3671 | } | |
3672 | } | |
1da177e4 | 3673 | |
dd41f596 IM |
3674 | /* |
3675 | * schedule() is the main scheduler function. | |
3676 | */ | |
3677 | asmlinkage void __sched schedule(void) | |
3678 | { | |
3679 | struct task_struct *prev, *next; | |
3680 | long *switch_count; | |
3681 | struct rq *rq; | |
dd41f596 IM |
3682 | int cpu; |
3683 | ||
3684 | need_resched: | |
3685 | preempt_disable(); | |
3686 | cpu = smp_processor_id(); | |
3687 | rq = cpu_rq(cpu); | |
3688 | rcu_qsctr_inc(cpu); | |
3689 | prev = rq->curr; | |
3690 | switch_count = &prev->nivcsw; | |
3691 | ||
3692 | release_kernel_lock(prev); | |
3693 | need_resched_nonpreemptible: | |
3694 | ||
3695 | schedule_debug(prev); | |
1da177e4 | 3696 | |
1e819950 IM |
3697 | /* |
3698 | * Do the rq-clock update outside the rq lock: | |
3699 | */ | |
3700 | local_irq_disable(); | |
c1b3da3e | 3701 | __update_rq_clock(rq); |
1e819950 IM |
3702 | spin_lock(&rq->lock); |
3703 | clear_tsk_need_resched(prev); | |
1da177e4 | 3704 | |
1da177e4 | 3705 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
1da177e4 | 3706 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
dd41f596 | 3707 | unlikely(signal_pending(prev)))) { |
1da177e4 | 3708 | prev->state = TASK_RUNNING; |
dd41f596 | 3709 | } else { |
2e1cb74a | 3710 | deactivate_task(rq, prev, 1); |
1da177e4 | 3711 | } |
dd41f596 | 3712 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3713 | } |
3714 | ||
dd41f596 | 3715 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3716 | idle_balance(cpu, rq); |
1da177e4 | 3717 | |
31ee529c | 3718 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 3719 | next = pick_next_task(rq, prev); |
1da177e4 LT |
3720 | |
3721 | sched_info_switch(prev, next); | |
dd41f596 | 3722 | |
1da177e4 | 3723 | if (likely(prev != next)) { |
1da177e4 LT |
3724 | rq->nr_switches++; |
3725 | rq->curr = next; | |
3726 | ++*switch_count; | |
3727 | ||
dd41f596 | 3728 | context_switch(rq, prev, next); /* unlocks the rq */ |
1da177e4 LT |
3729 | } else |
3730 | spin_unlock_irq(&rq->lock); | |
3731 | ||
dd41f596 IM |
3732 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3733 | cpu = smp_processor_id(); | |
3734 | rq = cpu_rq(cpu); | |
1da177e4 | 3735 | goto need_resched_nonpreemptible; |
dd41f596 | 3736 | } |
1da177e4 LT |
3737 | preempt_enable_no_resched(); |
3738 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
3739 | goto need_resched; | |
3740 | } | |
1da177e4 LT |
3741 | EXPORT_SYMBOL(schedule); |
3742 | ||
3743 | #ifdef CONFIG_PREEMPT | |
3744 | /* | |
2ed6e34f | 3745 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3746 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3747 | * occur there and call schedule directly. |
3748 | */ | |
3749 | asmlinkage void __sched preempt_schedule(void) | |
3750 | { | |
3751 | struct thread_info *ti = current_thread_info(); | |
3752 | #ifdef CONFIG_PREEMPT_BKL | |
3753 | struct task_struct *task = current; | |
3754 | int saved_lock_depth; | |
3755 | #endif | |
3756 | /* | |
3757 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3758 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3759 | */ |
beed33a8 | 3760 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3761 | return; |
3762 | ||
3a5c359a AK |
3763 | do { |
3764 | add_preempt_count(PREEMPT_ACTIVE); | |
3765 | ||
3766 | /* | |
3767 | * We keep the big kernel semaphore locked, but we | |
3768 | * clear ->lock_depth so that schedule() doesnt | |
3769 | * auto-release the semaphore: | |
3770 | */ | |
1da177e4 | 3771 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3772 | saved_lock_depth = task->lock_depth; |
3773 | task->lock_depth = -1; | |
1da177e4 | 3774 | #endif |
3a5c359a | 3775 | schedule(); |
1da177e4 | 3776 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3777 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3778 | #endif |
3a5c359a | 3779 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3780 | |
3a5c359a AK |
3781 | /* |
3782 | * Check again in case we missed a preemption opportunity | |
3783 | * between schedule and now. | |
3784 | */ | |
3785 | barrier(); | |
3786 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 3787 | } |
1da177e4 LT |
3788 | EXPORT_SYMBOL(preempt_schedule); |
3789 | ||
3790 | /* | |
2ed6e34f | 3791 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3792 | * off of irq context. |
3793 | * Note, that this is called and return with irqs disabled. This will | |
3794 | * protect us against recursive calling from irq. | |
3795 | */ | |
3796 | asmlinkage void __sched preempt_schedule_irq(void) | |
3797 | { | |
3798 | struct thread_info *ti = current_thread_info(); | |
3799 | #ifdef CONFIG_PREEMPT_BKL | |
3800 | struct task_struct *task = current; | |
3801 | int saved_lock_depth; | |
3802 | #endif | |
2ed6e34f | 3803 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3804 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3805 | ||
3a5c359a AK |
3806 | do { |
3807 | add_preempt_count(PREEMPT_ACTIVE); | |
3808 | ||
3809 | /* | |
3810 | * We keep the big kernel semaphore locked, but we | |
3811 | * clear ->lock_depth so that schedule() doesnt | |
3812 | * auto-release the semaphore: | |
3813 | */ | |
1da177e4 | 3814 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a AK |
3815 | saved_lock_depth = task->lock_depth; |
3816 | task->lock_depth = -1; | |
1da177e4 | 3817 | #endif |
3a5c359a AK |
3818 | local_irq_enable(); |
3819 | schedule(); | |
3820 | local_irq_disable(); | |
1da177e4 | 3821 | #ifdef CONFIG_PREEMPT_BKL |
3a5c359a | 3822 | task->lock_depth = saved_lock_depth; |
1da177e4 | 3823 | #endif |
3a5c359a | 3824 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3825 | |
3a5c359a AK |
3826 | /* |
3827 | * Check again in case we missed a preemption opportunity | |
3828 | * between schedule and now. | |
3829 | */ | |
3830 | barrier(); | |
3831 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
3832 | } |
3833 | ||
3834 | #endif /* CONFIG_PREEMPT */ | |
3835 | ||
95cdf3b7 IM |
3836 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
3837 | void *key) | |
1da177e4 | 3838 | { |
48f24c4d | 3839 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 3840 | } |
1da177e4 LT |
3841 | EXPORT_SYMBOL(default_wake_function); |
3842 | ||
3843 | /* | |
41a2d6cf IM |
3844 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3845 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3846 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3847 | * | |
3848 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3849 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3850 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3851 | */ | |
3852 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
3853 | int nr_exclusive, int sync, void *key) | |
3854 | { | |
2e45874c | 3855 | wait_queue_t *curr, *next; |
1da177e4 | 3856 | |
2e45874c | 3857 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3858 | unsigned flags = curr->flags; |
3859 | ||
1da177e4 | 3860 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 3861 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3862 | break; |
3863 | } | |
3864 | } | |
3865 | ||
3866 | /** | |
3867 | * __wake_up - wake up threads blocked on a waitqueue. | |
3868 | * @q: the waitqueue | |
3869 | * @mode: which threads | |
3870 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3871 | * @key: is directly passed to the wakeup function |
1da177e4 LT |
3872 | */ |
3873 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, | |
95cdf3b7 | 3874 | int nr_exclusive, void *key) |
1da177e4 LT |
3875 | { |
3876 | unsigned long flags; | |
3877 | ||
3878 | spin_lock_irqsave(&q->lock, flags); | |
3879 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3880 | spin_unlock_irqrestore(&q->lock, flags); | |
3881 | } | |
1da177e4 LT |
3882 | EXPORT_SYMBOL(__wake_up); |
3883 | ||
3884 | /* | |
3885 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3886 | */ | |
3887 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) | |
3888 | { | |
3889 | __wake_up_common(q, mode, 1, 0, NULL); | |
3890 | } | |
3891 | ||
3892 | /** | |
67be2dd1 | 3893 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3894 | * @q: the waitqueue |
3895 | * @mode: which threads | |
3896 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
3897 | * | |
3898 | * The sync wakeup differs that the waker knows that it will schedule | |
3899 | * away soon, so while the target thread will be woken up, it will not | |
3900 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3901 | * with each other. This can prevent needless bouncing between CPUs. | |
3902 | * | |
3903 | * On UP it can prevent extra preemption. | |
3904 | */ | |
95cdf3b7 IM |
3905 | void fastcall |
3906 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
1da177e4 LT |
3907 | { |
3908 | unsigned long flags; | |
3909 | int sync = 1; | |
3910 | ||
3911 | if (unlikely(!q)) | |
3912 | return; | |
3913 | ||
3914 | if (unlikely(!nr_exclusive)) | |
3915 | sync = 0; | |
3916 | ||
3917 | spin_lock_irqsave(&q->lock, flags); | |
3918 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
3919 | spin_unlock_irqrestore(&q->lock, flags); | |
3920 | } | |
3921 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
3922 | ||
b15136e9 | 3923 | void complete(struct completion *x) |
1da177e4 LT |
3924 | { |
3925 | unsigned long flags; | |
3926 | ||
3927 | spin_lock_irqsave(&x->wait.lock, flags); | |
3928 | x->done++; | |
3929 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3930 | 1, 0, NULL); | |
3931 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3932 | } | |
3933 | EXPORT_SYMBOL(complete); | |
3934 | ||
b15136e9 | 3935 | void complete_all(struct completion *x) |
1da177e4 LT |
3936 | { |
3937 | unsigned long flags; | |
3938 | ||
3939 | spin_lock_irqsave(&x->wait.lock, flags); | |
3940 | x->done += UINT_MAX/2; | |
3941 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, | |
3942 | 0, 0, NULL); | |
3943 | spin_unlock_irqrestore(&x->wait.lock, flags); | |
3944 | } | |
3945 | EXPORT_SYMBOL(complete_all); | |
3946 | ||
8cbbe86d AK |
3947 | static inline long __sched |
3948 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3949 | { |
1da177e4 LT |
3950 | if (!x->done) { |
3951 | DECLARE_WAITQUEUE(wait, current); | |
3952 | ||
3953 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3954 | __add_wait_queue_tail(&x->wait, &wait); | |
3955 | do { | |
8cbbe86d AK |
3956 | if (state == TASK_INTERRUPTIBLE && |
3957 | signal_pending(current)) { | |
3958 | __remove_wait_queue(&x->wait, &wait); | |
3959 | return -ERESTARTSYS; | |
3960 | } | |
3961 | __set_current_state(state); | |
1da177e4 LT |
3962 | spin_unlock_irq(&x->wait.lock); |
3963 | timeout = schedule_timeout(timeout); | |
3964 | spin_lock_irq(&x->wait.lock); | |
3965 | if (!timeout) { | |
3966 | __remove_wait_queue(&x->wait, &wait); | |
8cbbe86d | 3967 | return timeout; |
1da177e4 LT |
3968 | } |
3969 | } while (!x->done); | |
3970 | __remove_wait_queue(&x->wait, &wait); | |
3971 | } | |
3972 | x->done--; | |
1da177e4 LT |
3973 | return timeout; |
3974 | } | |
1da177e4 | 3975 | |
8cbbe86d AK |
3976 | static long __sched |
3977 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3978 | { |
1da177e4 LT |
3979 | might_sleep(); |
3980 | ||
3981 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3982 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3983 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3984 | return timeout; |
3985 | } | |
1da177e4 | 3986 | |
b15136e9 | 3987 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3988 | { |
3989 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3990 | } |
8cbbe86d | 3991 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3992 | |
b15136e9 | 3993 | unsigned long __sched |
8cbbe86d | 3994 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 3995 | { |
8cbbe86d | 3996 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 3997 | } |
8cbbe86d | 3998 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 3999 | |
8cbbe86d | 4000 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4001 | { |
51e97990 AK |
4002 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4003 | if (t == -ERESTARTSYS) | |
4004 | return t; | |
4005 | return 0; | |
0fec171c | 4006 | } |
8cbbe86d | 4007 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4008 | |
b15136e9 | 4009 | unsigned long __sched |
8cbbe86d AK |
4010 | wait_for_completion_interruptible_timeout(struct completion *x, |
4011 | unsigned long timeout) | |
0fec171c | 4012 | { |
8cbbe86d | 4013 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4014 | } |
8cbbe86d | 4015 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4016 | |
8cbbe86d AK |
4017 | static long __sched |
4018 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4019 | { |
0fec171c IM |
4020 | unsigned long flags; |
4021 | wait_queue_t wait; | |
4022 | ||
4023 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4024 | |
8cbbe86d | 4025 | __set_current_state(state); |
1da177e4 | 4026 | |
8cbbe86d AK |
4027 | spin_lock_irqsave(&q->lock, flags); |
4028 | __add_wait_queue(q, &wait); | |
4029 | spin_unlock(&q->lock); | |
4030 | timeout = schedule_timeout(timeout); | |
4031 | spin_lock_irq(&q->lock); | |
4032 | __remove_wait_queue(q, &wait); | |
4033 | spin_unlock_irqrestore(&q->lock, flags); | |
4034 | ||
4035 | return timeout; | |
4036 | } | |
4037 | ||
4038 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4039 | { | |
4040 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4041 | } |
1da177e4 LT |
4042 | EXPORT_SYMBOL(interruptible_sleep_on); |
4043 | ||
0fec171c | 4044 | long __sched |
95cdf3b7 | 4045 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4046 | { |
8cbbe86d | 4047 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4048 | } |
1da177e4 LT |
4049 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4050 | ||
0fec171c | 4051 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4052 | { |
8cbbe86d | 4053 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4054 | } |
1da177e4 LT |
4055 | EXPORT_SYMBOL(sleep_on); |
4056 | ||
0fec171c | 4057 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4058 | { |
8cbbe86d | 4059 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4060 | } |
1da177e4 LT |
4061 | EXPORT_SYMBOL(sleep_on_timeout); |
4062 | ||
b29739f9 IM |
4063 | #ifdef CONFIG_RT_MUTEXES |
4064 | ||
4065 | /* | |
4066 | * rt_mutex_setprio - set the current priority of a task | |
4067 | * @p: task | |
4068 | * @prio: prio value (kernel-internal form) | |
4069 | * | |
4070 | * This function changes the 'effective' priority of a task. It does | |
4071 | * not touch ->normal_prio like __setscheduler(). | |
4072 | * | |
4073 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4074 | */ | |
36c8b586 | 4075 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4076 | { |
4077 | unsigned long flags; | |
83b699ed | 4078 | int oldprio, on_rq, running; |
70b97a7f | 4079 | struct rq *rq; |
b29739f9 IM |
4080 | |
4081 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4082 | ||
4083 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4084 | update_rq_clock(rq); |
b29739f9 | 4085 | |
d5f9f942 | 4086 | oldprio = p->prio; |
dd41f596 | 4087 | on_rq = p->se.on_rq; |
051a1d1a | 4088 | running = task_current(rq, p); |
83b699ed | 4089 | if (on_rq) { |
69be72c1 | 4090 | dequeue_task(rq, p, 0); |
83b699ed SV |
4091 | if (running) |
4092 | p->sched_class->put_prev_task(rq, p); | |
4093 | } | |
dd41f596 IM |
4094 | |
4095 | if (rt_prio(prio)) | |
4096 | p->sched_class = &rt_sched_class; | |
4097 | else | |
4098 | p->sched_class = &fair_sched_class; | |
4099 | ||
b29739f9 IM |
4100 | p->prio = prio; |
4101 | ||
dd41f596 | 4102 | if (on_rq) { |
83b699ed SV |
4103 | if (running) |
4104 | p->sched_class->set_curr_task(rq); | |
8159f87e | 4105 | enqueue_task(rq, p, 0); |
b29739f9 IM |
4106 | /* |
4107 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4108 | * our priority decreased, or if we are not currently running on |
4109 | * this runqueue and our priority is higher than the current's | |
b29739f9 | 4110 | */ |
83b699ed | 4111 | if (running) { |
d5f9f942 AM |
4112 | if (p->prio > oldprio) |
4113 | resched_task(rq->curr); | |
dd41f596 IM |
4114 | } else { |
4115 | check_preempt_curr(rq, p); | |
4116 | } | |
b29739f9 IM |
4117 | } |
4118 | task_rq_unlock(rq, &flags); | |
4119 | } | |
4120 | ||
4121 | #endif | |
4122 | ||
36c8b586 | 4123 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4124 | { |
dd41f596 | 4125 | int old_prio, delta, on_rq; |
1da177e4 | 4126 | unsigned long flags; |
70b97a7f | 4127 | struct rq *rq; |
1da177e4 LT |
4128 | |
4129 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4130 | return; | |
4131 | /* | |
4132 | * We have to be careful, if called from sys_setpriority(), | |
4133 | * the task might be in the middle of scheduling on another CPU. | |
4134 | */ | |
4135 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4136 | update_rq_clock(rq); |
1da177e4 LT |
4137 | /* |
4138 | * The RT priorities are set via sched_setscheduler(), but we still | |
4139 | * allow the 'normal' nice value to be set - but as expected | |
4140 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4141 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4142 | */ |
e05606d3 | 4143 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4144 | p->static_prio = NICE_TO_PRIO(nice); |
4145 | goto out_unlock; | |
4146 | } | |
dd41f596 | 4147 | on_rq = p->se.on_rq; |
58e2d4ca | 4148 | if (on_rq) |
69be72c1 | 4149 | dequeue_task(rq, p, 0); |
1da177e4 | 4150 | |
1da177e4 | 4151 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4152 | set_load_weight(p); |
b29739f9 IM |
4153 | old_prio = p->prio; |
4154 | p->prio = effective_prio(p); | |
4155 | delta = p->prio - old_prio; | |
1da177e4 | 4156 | |
dd41f596 | 4157 | if (on_rq) { |
8159f87e | 4158 | enqueue_task(rq, p, 0); |
1da177e4 | 4159 | /* |
d5f9f942 AM |
4160 | * If the task increased its priority or is running and |
4161 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4162 | */ |
d5f9f942 | 4163 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4164 | resched_task(rq->curr); |
4165 | } | |
4166 | out_unlock: | |
4167 | task_rq_unlock(rq, &flags); | |
4168 | } | |
1da177e4 LT |
4169 | EXPORT_SYMBOL(set_user_nice); |
4170 | ||
e43379f1 MM |
4171 | /* |
4172 | * can_nice - check if a task can reduce its nice value | |
4173 | * @p: task | |
4174 | * @nice: nice value | |
4175 | */ | |
36c8b586 | 4176 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4177 | { |
024f4747 MM |
4178 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4179 | int nice_rlim = 20 - nice; | |
48f24c4d | 4180 | |
e43379f1 MM |
4181 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4182 | capable(CAP_SYS_NICE)); | |
4183 | } | |
4184 | ||
1da177e4 LT |
4185 | #ifdef __ARCH_WANT_SYS_NICE |
4186 | ||
4187 | /* | |
4188 | * sys_nice - change the priority of the current process. | |
4189 | * @increment: priority increment | |
4190 | * | |
4191 | * sys_setpriority is a more generic, but much slower function that | |
4192 | * does similar things. | |
4193 | */ | |
4194 | asmlinkage long sys_nice(int increment) | |
4195 | { | |
48f24c4d | 4196 | long nice, retval; |
1da177e4 LT |
4197 | |
4198 | /* | |
4199 | * Setpriority might change our priority at the same moment. | |
4200 | * We don't have to worry. Conceptually one call occurs first | |
4201 | * and we have a single winner. | |
4202 | */ | |
e43379f1 MM |
4203 | if (increment < -40) |
4204 | increment = -40; | |
1da177e4 LT |
4205 | if (increment > 40) |
4206 | increment = 40; | |
4207 | ||
4208 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4209 | if (nice < -20) | |
4210 | nice = -20; | |
4211 | if (nice > 19) | |
4212 | nice = 19; | |
4213 | ||
e43379f1 MM |
4214 | if (increment < 0 && !can_nice(current, nice)) |
4215 | return -EPERM; | |
4216 | ||
1da177e4 LT |
4217 | retval = security_task_setnice(current, nice); |
4218 | if (retval) | |
4219 | return retval; | |
4220 | ||
4221 | set_user_nice(current, nice); | |
4222 | return 0; | |
4223 | } | |
4224 | ||
4225 | #endif | |
4226 | ||
4227 | /** | |
4228 | * task_prio - return the priority value of a given task. | |
4229 | * @p: the task in question. | |
4230 | * | |
4231 | * This is the priority value as seen by users in /proc. | |
4232 | * RT tasks are offset by -200. Normal tasks are centered | |
4233 | * around 0, value goes from -16 to +15. | |
4234 | */ | |
36c8b586 | 4235 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4236 | { |
4237 | return p->prio - MAX_RT_PRIO; | |
4238 | } | |
4239 | ||
4240 | /** | |
4241 | * task_nice - return the nice value of a given task. | |
4242 | * @p: the task in question. | |
4243 | */ | |
36c8b586 | 4244 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4245 | { |
4246 | return TASK_NICE(p); | |
4247 | } | |
1da177e4 | 4248 | EXPORT_SYMBOL_GPL(task_nice); |
1da177e4 LT |
4249 | |
4250 | /** | |
4251 | * idle_cpu - is a given cpu idle currently? | |
4252 | * @cpu: the processor in question. | |
4253 | */ | |
4254 | int idle_cpu(int cpu) | |
4255 | { | |
4256 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4257 | } | |
4258 | ||
1da177e4 LT |
4259 | /** |
4260 | * idle_task - return the idle task for a given cpu. | |
4261 | * @cpu: the processor in question. | |
4262 | */ | |
36c8b586 | 4263 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4264 | { |
4265 | return cpu_rq(cpu)->idle; | |
4266 | } | |
4267 | ||
4268 | /** | |
4269 | * find_process_by_pid - find a process with a matching PID value. | |
4270 | * @pid: the pid in question. | |
4271 | */ | |
a9957449 | 4272 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4273 | { |
228ebcbe | 4274 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4275 | } |
4276 | ||
4277 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4278 | static void |
4279 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4280 | { |
dd41f596 | 4281 | BUG_ON(p->se.on_rq); |
48f24c4d | 4282 | |
1da177e4 | 4283 | p->policy = policy; |
dd41f596 IM |
4284 | switch (p->policy) { |
4285 | case SCHED_NORMAL: | |
4286 | case SCHED_BATCH: | |
4287 | case SCHED_IDLE: | |
4288 | p->sched_class = &fair_sched_class; | |
4289 | break; | |
4290 | case SCHED_FIFO: | |
4291 | case SCHED_RR: | |
4292 | p->sched_class = &rt_sched_class; | |
4293 | break; | |
4294 | } | |
4295 | ||
1da177e4 | 4296 | p->rt_priority = prio; |
b29739f9 IM |
4297 | p->normal_prio = normal_prio(p); |
4298 | /* we are holding p->pi_lock already */ | |
4299 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4300 | set_load_weight(p); |
1da177e4 LT |
4301 | } |
4302 | ||
4303 | /** | |
72fd4a35 | 4304 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4305 | * @p: the task in question. |
4306 | * @policy: new policy. | |
4307 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4308 | * |
72fd4a35 | 4309 | * NOTE that the task may be already dead. |
1da177e4 | 4310 | */ |
95cdf3b7 IM |
4311 | int sched_setscheduler(struct task_struct *p, int policy, |
4312 | struct sched_param *param) | |
1da177e4 | 4313 | { |
83b699ed | 4314 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4315 | unsigned long flags; |
70b97a7f | 4316 | struct rq *rq; |
1da177e4 | 4317 | |
66e5393a SR |
4318 | /* may grab non-irq protected spin_locks */ |
4319 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4320 | recheck: |
4321 | /* double check policy once rq lock held */ | |
4322 | if (policy < 0) | |
4323 | policy = oldpolicy = p->policy; | |
4324 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4325 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4326 | policy != SCHED_IDLE) | |
b0a9499c | 4327 | return -EINVAL; |
1da177e4 LT |
4328 | /* |
4329 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4330 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4331 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4332 | */ |
4333 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4334 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4335 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4336 | return -EINVAL; |
e05606d3 | 4337 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4338 | return -EINVAL; |
4339 | ||
37e4ab3f OC |
4340 | /* |
4341 | * Allow unprivileged RT tasks to decrease priority: | |
4342 | */ | |
4343 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4344 | if (rt_policy(policy)) { |
8dc3e909 | 4345 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4346 | |
4347 | if (!lock_task_sighand(p, &flags)) | |
4348 | return -ESRCH; | |
4349 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4350 | unlock_task_sighand(p, &flags); | |
4351 | ||
4352 | /* can't set/change the rt policy */ | |
4353 | if (policy != p->policy && !rlim_rtprio) | |
4354 | return -EPERM; | |
4355 | ||
4356 | /* can't increase priority */ | |
4357 | if (param->sched_priority > p->rt_priority && | |
4358 | param->sched_priority > rlim_rtprio) | |
4359 | return -EPERM; | |
4360 | } | |
dd41f596 IM |
4361 | /* |
4362 | * Like positive nice levels, dont allow tasks to | |
4363 | * move out of SCHED_IDLE either: | |
4364 | */ | |
4365 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4366 | return -EPERM; | |
5fe1d75f | 4367 | |
37e4ab3f OC |
4368 | /* can't change other user's priorities */ |
4369 | if ((current->euid != p->euid) && | |
4370 | (current->euid != p->uid)) | |
4371 | return -EPERM; | |
4372 | } | |
1da177e4 LT |
4373 | |
4374 | retval = security_task_setscheduler(p, policy, param); | |
4375 | if (retval) | |
4376 | return retval; | |
b29739f9 IM |
4377 | /* |
4378 | * make sure no PI-waiters arrive (or leave) while we are | |
4379 | * changing the priority of the task: | |
4380 | */ | |
4381 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4382 | /* |
4383 | * To be able to change p->policy safely, the apropriate | |
4384 | * runqueue lock must be held. | |
4385 | */ | |
b29739f9 | 4386 | rq = __task_rq_lock(p); |
1da177e4 LT |
4387 | /* recheck policy now with rq lock held */ |
4388 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4389 | policy = oldpolicy = -1; | |
b29739f9 IM |
4390 | __task_rq_unlock(rq); |
4391 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4392 | goto recheck; |
4393 | } | |
2daa3577 | 4394 | update_rq_clock(rq); |
dd41f596 | 4395 | on_rq = p->se.on_rq; |
051a1d1a | 4396 | running = task_current(rq, p); |
83b699ed | 4397 | if (on_rq) { |
2e1cb74a | 4398 | deactivate_task(rq, p, 0); |
83b699ed SV |
4399 | if (running) |
4400 | p->sched_class->put_prev_task(rq, p); | |
4401 | } | |
f6b53205 | 4402 | |
1da177e4 | 4403 | oldprio = p->prio; |
dd41f596 | 4404 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4405 | |
dd41f596 | 4406 | if (on_rq) { |
83b699ed SV |
4407 | if (running) |
4408 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4409 | activate_task(rq, p, 0); |
1da177e4 LT |
4410 | /* |
4411 | * Reschedule if we are currently running on this runqueue and | |
d5f9f942 AM |
4412 | * our priority decreased, or if we are not currently running on |
4413 | * this runqueue and our priority is higher than the current's | |
1da177e4 | 4414 | */ |
83b699ed | 4415 | if (running) { |
d5f9f942 AM |
4416 | if (p->prio > oldprio) |
4417 | resched_task(rq->curr); | |
dd41f596 IM |
4418 | } else { |
4419 | check_preempt_curr(rq, p); | |
4420 | } | |
1da177e4 | 4421 | } |
b29739f9 IM |
4422 | __task_rq_unlock(rq); |
4423 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4424 | ||
95e02ca9 TG |
4425 | rt_mutex_adjust_pi(p); |
4426 | ||
1da177e4 LT |
4427 | return 0; |
4428 | } | |
4429 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4430 | ||
95cdf3b7 IM |
4431 | static int |
4432 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4433 | { |
1da177e4 LT |
4434 | struct sched_param lparam; |
4435 | struct task_struct *p; | |
36c8b586 | 4436 | int retval; |
1da177e4 LT |
4437 | |
4438 | if (!param || pid < 0) | |
4439 | return -EINVAL; | |
4440 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4441 | return -EFAULT; | |
5fe1d75f ON |
4442 | |
4443 | rcu_read_lock(); | |
4444 | retval = -ESRCH; | |
1da177e4 | 4445 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4446 | if (p != NULL) |
4447 | retval = sched_setscheduler(p, policy, &lparam); | |
4448 | rcu_read_unlock(); | |
36c8b586 | 4449 | |
1da177e4 LT |
4450 | return retval; |
4451 | } | |
4452 | ||
4453 | /** | |
4454 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4455 | * @pid: the pid in question. | |
4456 | * @policy: new policy. | |
4457 | * @param: structure containing the new RT priority. | |
4458 | */ | |
41a2d6cf IM |
4459 | asmlinkage long |
4460 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4461 | { |
c21761f1 JB |
4462 | /* negative values for policy are not valid */ |
4463 | if (policy < 0) | |
4464 | return -EINVAL; | |
4465 | ||
1da177e4 LT |
4466 | return do_sched_setscheduler(pid, policy, param); |
4467 | } | |
4468 | ||
4469 | /** | |
4470 | * sys_sched_setparam - set/change the RT priority of a thread | |
4471 | * @pid: the pid in question. | |
4472 | * @param: structure containing the new RT priority. | |
4473 | */ | |
4474 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4475 | { | |
4476 | return do_sched_setscheduler(pid, -1, param); | |
4477 | } | |
4478 | ||
4479 | /** | |
4480 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4481 | * @pid: the pid in question. | |
4482 | */ | |
4483 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4484 | { | |
36c8b586 | 4485 | struct task_struct *p; |
3a5c359a | 4486 | int retval; |
1da177e4 LT |
4487 | |
4488 | if (pid < 0) | |
3a5c359a | 4489 | return -EINVAL; |
1da177e4 LT |
4490 | |
4491 | retval = -ESRCH; | |
4492 | read_lock(&tasklist_lock); | |
4493 | p = find_process_by_pid(pid); | |
4494 | if (p) { | |
4495 | retval = security_task_getscheduler(p); | |
4496 | if (!retval) | |
4497 | retval = p->policy; | |
4498 | } | |
4499 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4500 | return retval; |
4501 | } | |
4502 | ||
4503 | /** | |
4504 | * sys_sched_getscheduler - get the RT priority of a thread | |
4505 | * @pid: the pid in question. | |
4506 | * @param: structure containing the RT priority. | |
4507 | */ | |
4508 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
4509 | { | |
4510 | struct sched_param lp; | |
36c8b586 | 4511 | struct task_struct *p; |
3a5c359a | 4512 | int retval; |
1da177e4 LT |
4513 | |
4514 | if (!param || pid < 0) | |
3a5c359a | 4515 | return -EINVAL; |
1da177e4 LT |
4516 | |
4517 | read_lock(&tasklist_lock); | |
4518 | p = find_process_by_pid(pid); | |
4519 | retval = -ESRCH; | |
4520 | if (!p) | |
4521 | goto out_unlock; | |
4522 | ||
4523 | retval = security_task_getscheduler(p); | |
4524 | if (retval) | |
4525 | goto out_unlock; | |
4526 | ||
4527 | lp.sched_priority = p->rt_priority; | |
4528 | read_unlock(&tasklist_lock); | |
4529 | ||
4530 | /* | |
4531 | * This one might sleep, we cannot do it with a spinlock held ... | |
4532 | */ | |
4533 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4534 | ||
1da177e4 LT |
4535 | return retval; |
4536 | ||
4537 | out_unlock: | |
4538 | read_unlock(&tasklist_lock); | |
4539 | return retval; | |
4540 | } | |
4541 | ||
4542 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) | |
4543 | { | |
1da177e4 | 4544 | cpumask_t cpus_allowed; |
36c8b586 IM |
4545 | struct task_struct *p; |
4546 | int retval; | |
1da177e4 | 4547 | |
95402b38 | 4548 | get_online_cpus(); |
1da177e4 LT |
4549 | read_lock(&tasklist_lock); |
4550 | ||
4551 | p = find_process_by_pid(pid); | |
4552 | if (!p) { | |
4553 | read_unlock(&tasklist_lock); | |
95402b38 | 4554 | put_online_cpus(); |
1da177e4 LT |
4555 | return -ESRCH; |
4556 | } | |
4557 | ||
4558 | /* | |
4559 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 4560 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
4561 | * usage count and then drop tasklist_lock. |
4562 | */ | |
4563 | get_task_struct(p); | |
4564 | read_unlock(&tasklist_lock); | |
4565 | ||
4566 | retval = -EPERM; | |
4567 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
4568 | !capable(CAP_SYS_NICE)) | |
4569 | goto out_unlock; | |
4570 | ||
e7834f8f DQ |
4571 | retval = security_task_setscheduler(p, 0, NULL); |
4572 | if (retval) | |
4573 | goto out_unlock; | |
4574 | ||
1da177e4 LT |
4575 | cpus_allowed = cpuset_cpus_allowed(p); |
4576 | cpus_and(new_mask, new_mask, cpus_allowed); | |
8707d8b8 | 4577 | again: |
1da177e4 LT |
4578 | retval = set_cpus_allowed(p, new_mask); |
4579 | ||
8707d8b8 PM |
4580 | if (!retval) { |
4581 | cpus_allowed = cpuset_cpus_allowed(p); | |
4582 | if (!cpus_subset(new_mask, cpus_allowed)) { | |
4583 | /* | |
4584 | * We must have raced with a concurrent cpuset | |
4585 | * update. Just reset the cpus_allowed to the | |
4586 | * cpuset's cpus_allowed | |
4587 | */ | |
4588 | new_mask = cpus_allowed; | |
4589 | goto again; | |
4590 | } | |
4591 | } | |
1da177e4 LT |
4592 | out_unlock: |
4593 | put_task_struct(p); | |
95402b38 | 4594 | put_online_cpus(); |
1da177e4 LT |
4595 | return retval; |
4596 | } | |
4597 | ||
4598 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
4599 | cpumask_t *new_mask) | |
4600 | { | |
4601 | if (len < sizeof(cpumask_t)) { | |
4602 | memset(new_mask, 0, sizeof(cpumask_t)); | |
4603 | } else if (len > sizeof(cpumask_t)) { | |
4604 | len = sizeof(cpumask_t); | |
4605 | } | |
4606 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
4607 | } | |
4608 | ||
4609 | /** | |
4610 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4611 | * @pid: pid of the process | |
4612 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4613 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4614 | */ | |
4615 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
4616 | unsigned long __user *user_mask_ptr) | |
4617 | { | |
4618 | cpumask_t new_mask; | |
4619 | int retval; | |
4620 | ||
4621 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
4622 | if (retval) | |
4623 | return retval; | |
4624 | ||
4625 | return sched_setaffinity(pid, new_mask); | |
4626 | } | |
4627 | ||
4628 | /* | |
4629 | * Represents all cpu's present in the system | |
4630 | * In systems capable of hotplug, this map could dynamically grow | |
4631 | * as new cpu's are detected in the system via any platform specific | |
4632 | * method, such as ACPI for e.g. | |
4633 | */ | |
4634 | ||
4cef0c61 | 4635 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
4636 | EXPORT_SYMBOL(cpu_present_map); |
4637 | ||
4638 | #ifndef CONFIG_SMP | |
4cef0c61 | 4639 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
4640 | EXPORT_SYMBOL(cpu_online_map); |
4641 | ||
4cef0c61 | 4642 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 4643 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
4644 | #endif |
4645 | ||
4646 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
4647 | { | |
36c8b586 | 4648 | struct task_struct *p; |
1da177e4 | 4649 | int retval; |
1da177e4 | 4650 | |
95402b38 | 4651 | get_online_cpus(); |
1da177e4 LT |
4652 | read_lock(&tasklist_lock); |
4653 | ||
4654 | retval = -ESRCH; | |
4655 | p = find_process_by_pid(pid); | |
4656 | if (!p) | |
4657 | goto out_unlock; | |
4658 | ||
e7834f8f DQ |
4659 | retval = security_task_getscheduler(p); |
4660 | if (retval) | |
4661 | goto out_unlock; | |
4662 | ||
2f7016d9 | 4663 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
4664 | |
4665 | out_unlock: | |
4666 | read_unlock(&tasklist_lock); | |
95402b38 | 4667 | put_online_cpus(); |
1da177e4 | 4668 | |
9531b62f | 4669 | return retval; |
1da177e4 LT |
4670 | } |
4671 | ||
4672 | /** | |
4673 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4674 | * @pid: pid of the process | |
4675 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4676 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4677 | */ | |
4678 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
4679 | unsigned long __user *user_mask_ptr) | |
4680 | { | |
4681 | int ret; | |
4682 | cpumask_t mask; | |
4683 | ||
4684 | if (len < sizeof(cpumask_t)) | |
4685 | return -EINVAL; | |
4686 | ||
4687 | ret = sched_getaffinity(pid, &mask); | |
4688 | if (ret < 0) | |
4689 | return ret; | |
4690 | ||
4691 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
4692 | return -EFAULT; | |
4693 | ||
4694 | return sizeof(cpumask_t); | |
4695 | } | |
4696 | ||
4697 | /** | |
4698 | * sys_sched_yield - yield the current processor to other threads. | |
4699 | * | |
dd41f596 IM |
4700 | * This function yields the current CPU to other tasks. If there are no |
4701 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
4702 | */ |
4703 | asmlinkage long sys_sched_yield(void) | |
4704 | { | |
70b97a7f | 4705 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4706 | |
2d72376b | 4707 | schedstat_inc(rq, yld_count); |
4530d7ab | 4708 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4709 | |
4710 | /* | |
4711 | * Since we are going to call schedule() anyway, there's | |
4712 | * no need to preempt or enable interrupts: | |
4713 | */ | |
4714 | __release(rq->lock); | |
8a25d5de | 4715 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4716 | _raw_spin_unlock(&rq->lock); |
4717 | preempt_enable_no_resched(); | |
4718 | ||
4719 | schedule(); | |
4720 | ||
4721 | return 0; | |
4722 | } | |
4723 | ||
e7b38404 | 4724 | static void __cond_resched(void) |
1da177e4 | 4725 | { |
8e0a43d8 IM |
4726 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
4727 | __might_sleep(__FILE__, __LINE__); | |
4728 | #endif | |
5bbcfd90 IM |
4729 | /* |
4730 | * The BKS might be reacquired before we have dropped | |
4731 | * PREEMPT_ACTIVE, which could trigger a second | |
4732 | * cond_resched() call. | |
4733 | */ | |
1da177e4 LT |
4734 | do { |
4735 | add_preempt_count(PREEMPT_ACTIVE); | |
4736 | schedule(); | |
4737 | sub_preempt_count(PREEMPT_ACTIVE); | |
4738 | } while (need_resched()); | |
4739 | } | |
4740 | ||
4741 | int __sched cond_resched(void) | |
4742 | { | |
9414232f IM |
4743 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
4744 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
4745 | __cond_resched(); |
4746 | return 1; | |
4747 | } | |
4748 | return 0; | |
4749 | } | |
1da177e4 LT |
4750 | EXPORT_SYMBOL(cond_resched); |
4751 | ||
4752 | /* | |
4753 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
4754 | * call schedule, and on return reacquire the lock. | |
4755 | * | |
41a2d6cf | 4756 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4757 | * operations here to prevent schedule() from being called twice (once via |
4758 | * spin_unlock(), once by hand). | |
4759 | */ | |
95cdf3b7 | 4760 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4761 | { |
6df3cecb JK |
4762 | int ret = 0; |
4763 | ||
1da177e4 LT |
4764 | if (need_lockbreak(lock)) { |
4765 | spin_unlock(lock); | |
4766 | cpu_relax(); | |
6df3cecb | 4767 | ret = 1; |
1da177e4 LT |
4768 | spin_lock(lock); |
4769 | } | |
9414232f | 4770 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
8a25d5de | 4771 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
1da177e4 LT |
4772 | _raw_spin_unlock(lock); |
4773 | preempt_enable_no_resched(); | |
4774 | __cond_resched(); | |
6df3cecb | 4775 | ret = 1; |
1da177e4 | 4776 | spin_lock(lock); |
1da177e4 | 4777 | } |
6df3cecb | 4778 | return ret; |
1da177e4 | 4779 | } |
1da177e4 LT |
4780 | EXPORT_SYMBOL(cond_resched_lock); |
4781 | ||
4782 | int __sched cond_resched_softirq(void) | |
4783 | { | |
4784 | BUG_ON(!in_softirq()); | |
4785 | ||
9414232f | 4786 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 4787 | local_bh_enable(); |
1da177e4 LT |
4788 | __cond_resched(); |
4789 | local_bh_disable(); | |
4790 | return 1; | |
4791 | } | |
4792 | return 0; | |
4793 | } | |
1da177e4 LT |
4794 | EXPORT_SYMBOL(cond_resched_softirq); |
4795 | ||
1da177e4 LT |
4796 | /** |
4797 | * yield - yield the current processor to other threads. | |
4798 | * | |
72fd4a35 | 4799 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4800 | * thread runnable and calls sys_sched_yield(). |
4801 | */ | |
4802 | void __sched yield(void) | |
4803 | { | |
4804 | set_current_state(TASK_RUNNING); | |
4805 | sys_sched_yield(); | |
4806 | } | |
1da177e4 LT |
4807 | EXPORT_SYMBOL(yield); |
4808 | ||
4809 | /* | |
41a2d6cf | 4810 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
4811 | * that process accounting knows that this is a task in IO wait state. |
4812 | * | |
4813 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
4814 | * has set its backing_dev_info: the queue against which it should throttle) | |
4815 | */ | |
4816 | void __sched io_schedule(void) | |
4817 | { | |
70b97a7f | 4818 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 4819 | |
0ff92245 | 4820 | delayacct_blkio_start(); |
1da177e4 LT |
4821 | atomic_inc(&rq->nr_iowait); |
4822 | schedule(); | |
4823 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4824 | delayacct_blkio_end(); |
1da177e4 | 4825 | } |
1da177e4 LT |
4826 | EXPORT_SYMBOL(io_schedule); |
4827 | ||
4828 | long __sched io_schedule_timeout(long timeout) | |
4829 | { | |
70b97a7f | 4830 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
4831 | long ret; |
4832 | ||
0ff92245 | 4833 | delayacct_blkio_start(); |
1da177e4 LT |
4834 | atomic_inc(&rq->nr_iowait); |
4835 | ret = schedule_timeout(timeout); | |
4836 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 4837 | delayacct_blkio_end(); |
1da177e4 LT |
4838 | return ret; |
4839 | } | |
4840 | ||
4841 | /** | |
4842 | * sys_sched_get_priority_max - return maximum RT priority. | |
4843 | * @policy: scheduling class. | |
4844 | * | |
4845 | * this syscall returns the maximum rt_priority that can be used | |
4846 | * by a given scheduling class. | |
4847 | */ | |
4848 | asmlinkage long sys_sched_get_priority_max(int policy) | |
4849 | { | |
4850 | int ret = -EINVAL; | |
4851 | ||
4852 | switch (policy) { | |
4853 | case SCHED_FIFO: | |
4854 | case SCHED_RR: | |
4855 | ret = MAX_USER_RT_PRIO-1; | |
4856 | break; | |
4857 | case SCHED_NORMAL: | |
b0a9499c | 4858 | case SCHED_BATCH: |
dd41f596 | 4859 | case SCHED_IDLE: |
1da177e4 LT |
4860 | ret = 0; |
4861 | break; | |
4862 | } | |
4863 | return ret; | |
4864 | } | |
4865 | ||
4866 | /** | |
4867 | * sys_sched_get_priority_min - return minimum RT priority. | |
4868 | * @policy: scheduling class. | |
4869 | * | |
4870 | * this syscall returns the minimum rt_priority that can be used | |
4871 | * by a given scheduling class. | |
4872 | */ | |
4873 | asmlinkage long sys_sched_get_priority_min(int policy) | |
4874 | { | |
4875 | int ret = -EINVAL; | |
4876 | ||
4877 | switch (policy) { | |
4878 | case SCHED_FIFO: | |
4879 | case SCHED_RR: | |
4880 | ret = 1; | |
4881 | break; | |
4882 | case SCHED_NORMAL: | |
b0a9499c | 4883 | case SCHED_BATCH: |
dd41f596 | 4884 | case SCHED_IDLE: |
1da177e4 LT |
4885 | ret = 0; |
4886 | } | |
4887 | return ret; | |
4888 | } | |
4889 | ||
4890 | /** | |
4891 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
4892 | * @pid: pid of the process. | |
4893 | * @interval: userspace pointer to the timeslice value. | |
4894 | * | |
4895 | * this syscall writes the default timeslice value of a given process | |
4896 | * into the user-space timespec buffer. A value of '0' means infinity. | |
4897 | */ | |
4898 | asmlinkage | |
4899 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
4900 | { | |
36c8b586 | 4901 | struct task_struct *p; |
a4ec24b4 | 4902 | unsigned int time_slice; |
3a5c359a | 4903 | int retval; |
1da177e4 | 4904 | struct timespec t; |
1da177e4 LT |
4905 | |
4906 | if (pid < 0) | |
3a5c359a | 4907 | return -EINVAL; |
1da177e4 LT |
4908 | |
4909 | retval = -ESRCH; | |
4910 | read_lock(&tasklist_lock); | |
4911 | p = find_process_by_pid(pid); | |
4912 | if (!p) | |
4913 | goto out_unlock; | |
4914 | ||
4915 | retval = security_task_getscheduler(p); | |
4916 | if (retval) | |
4917 | goto out_unlock; | |
4918 | ||
77034937 IM |
4919 | /* |
4920 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
4921 | * tasks that are on an otherwise idle runqueue: | |
4922 | */ | |
4923 | time_slice = 0; | |
4924 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 4925 | time_slice = DEF_TIMESLICE; |
77034937 | 4926 | } else { |
a4ec24b4 DA |
4927 | struct sched_entity *se = &p->se; |
4928 | unsigned long flags; | |
4929 | struct rq *rq; | |
4930 | ||
4931 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
4932 | if (rq->cfs.load.weight) |
4933 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
4934 | task_rq_unlock(rq, &flags); |
4935 | } | |
1da177e4 | 4936 | read_unlock(&tasklist_lock); |
a4ec24b4 | 4937 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 4938 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 4939 | return retval; |
3a5c359a | 4940 | |
1da177e4 LT |
4941 | out_unlock: |
4942 | read_unlock(&tasklist_lock); | |
4943 | return retval; | |
4944 | } | |
4945 | ||
2ed6e34f | 4946 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 IM |
4947 | |
4948 | static void show_task(struct task_struct *p) | |
1da177e4 | 4949 | { |
1da177e4 | 4950 | unsigned long free = 0; |
36c8b586 | 4951 | unsigned state; |
1da177e4 | 4952 | |
1da177e4 | 4953 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 4954 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 4955 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 4956 | #if BITS_PER_LONG == 32 |
1da177e4 | 4957 | if (state == TASK_RUNNING) |
cc4ea795 | 4958 | printk(KERN_CONT " running "); |
1da177e4 | 4959 | else |
cc4ea795 | 4960 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
4961 | #else |
4962 | if (state == TASK_RUNNING) | |
cc4ea795 | 4963 | printk(KERN_CONT " running task "); |
1da177e4 | 4964 | else |
cc4ea795 | 4965 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
4966 | #endif |
4967 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
4968 | { | |
10ebffde | 4969 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
4970 | while (!*n) |
4971 | n++; | |
10ebffde | 4972 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
4973 | } |
4974 | #endif | |
ba25f9dc | 4975 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 4976 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 LT |
4977 | |
4978 | if (state != TASK_RUNNING) | |
4979 | show_stack(p, NULL); | |
4980 | } | |
4981 | ||
e59e2ae2 | 4982 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 4983 | { |
36c8b586 | 4984 | struct task_struct *g, *p; |
1da177e4 | 4985 | |
4bd77321 IM |
4986 | #if BITS_PER_LONG == 32 |
4987 | printk(KERN_INFO | |
4988 | " task PC stack pid father\n"); | |
1da177e4 | 4989 | #else |
4bd77321 IM |
4990 | printk(KERN_INFO |
4991 | " task PC stack pid father\n"); | |
1da177e4 LT |
4992 | #endif |
4993 | read_lock(&tasklist_lock); | |
4994 | do_each_thread(g, p) { | |
4995 | /* | |
4996 | * reset the NMI-timeout, listing all files on a slow | |
4997 | * console might take alot of time: | |
4998 | */ | |
4999 | touch_nmi_watchdog(); | |
39bc89fd | 5000 | if (!state_filter || (p->state & state_filter)) |
e59e2ae2 | 5001 | show_task(p); |
1da177e4 LT |
5002 | } while_each_thread(g, p); |
5003 | ||
04c9167f JF |
5004 | touch_all_softlockup_watchdogs(); |
5005 | ||
dd41f596 IM |
5006 | #ifdef CONFIG_SCHED_DEBUG |
5007 | sysrq_sched_debug_show(); | |
5008 | #endif | |
1da177e4 | 5009 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5010 | /* |
5011 | * Only show locks if all tasks are dumped: | |
5012 | */ | |
5013 | if (state_filter == -1) | |
5014 | debug_show_all_locks(); | |
1da177e4 LT |
5015 | } |
5016 | ||
1df21055 IM |
5017 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5018 | { | |
dd41f596 | 5019 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5020 | } |
5021 | ||
f340c0d1 IM |
5022 | /** |
5023 | * init_idle - set up an idle thread for a given CPU | |
5024 | * @idle: task in question | |
5025 | * @cpu: cpu the idle task belongs to | |
5026 | * | |
5027 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5028 | * flag, to make booting more robust. | |
5029 | */ | |
5c1e1767 | 5030 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5031 | { |
70b97a7f | 5032 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5033 | unsigned long flags; |
5034 | ||
dd41f596 IM |
5035 | __sched_fork(idle); |
5036 | idle->se.exec_start = sched_clock(); | |
5037 | ||
b29739f9 | 5038 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5039 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5040 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5041 | |
5042 | spin_lock_irqsave(&rq->lock, flags); | |
5043 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5044 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5045 | idle->oncpu = 1; | |
5046 | #endif | |
1da177e4 LT |
5047 | spin_unlock_irqrestore(&rq->lock, flags); |
5048 | ||
5049 | /* Set the preempt count _outside_ the spinlocks! */ | |
5050 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) | |
a1261f54 | 5051 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
1da177e4 | 5052 | #else |
a1261f54 | 5053 | task_thread_info(idle)->preempt_count = 0; |
1da177e4 | 5054 | #endif |
dd41f596 IM |
5055 | /* |
5056 | * The idle tasks have their own, simple scheduling class: | |
5057 | */ | |
5058 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5059 | } |
5060 | ||
5061 | /* | |
5062 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5063 | * indicates which cpus entered this state. This is used | |
5064 | * in the rcu update to wait only for active cpus. For system | |
5065 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5066 | * always be CPU_MASK_NONE. | |
5067 | */ | |
5068 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5069 | ||
19978ca6 IM |
5070 | /* |
5071 | * Increase the granularity value when there are more CPUs, | |
5072 | * because with more CPUs the 'effective latency' as visible | |
5073 | * to users decreases. But the relationship is not linear, | |
5074 | * so pick a second-best guess by going with the log2 of the | |
5075 | * number of CPUs. | |
5076 | * | |
5077 | * This idea comes from the SD scheduler of Con Kolivas: | |
5078 | */ | |
5079 | static inline void sched_init_granularity(void) | |
5080 | { | |
5081 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5082 | const unsigned long limit = 200000000; | |
5083 | ||
5084 | sysctl_sched_min_granularity *= factor; | |
5085 | if (sysctl_sched_min_granularity > limit) | |
5086 | sysctl_sched_min_granularity = limit; | |
5087 | ||
5088 | sysctl_sched_latency *= factor; | |
5089 | if (sysctl_sched_latency > limit) | |
5090 | sysctl_sched_latency = limit; | |
5091 | ||
5092 | sysctl_sched_wakeup_granularity *= factor; | |
5093 | sysctl_sched_batch_wakeup_granularity *= factor; | |
5094 | } | |
5095 | ||
1da177e4 LT |
5096 | #ifdef CONFIG_SMP |
5097 | /* | |
5098 | * This is how migration works: | |
5099 | * | |
70b97a7f | 5100 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5101 | * runqueue and wake up that CPU's migration thread. |
5102 | * 2) we down() the locked semaphore => thread blocks. | |
5103 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5104 | * thread off the CPU) | |
5105 | * 4) it gets the migration request and checks whether the migrated | |
5106 | * task is still in the wrong runqueue. | |
5107 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5108 | * it and puts it into the right queue. | |
5109 | * 6) migration thread up()s the semaphore. | |
5110 | * 7) we wake up and the migration is done. | |
5111 | */ | |
5112 | ||
5113 | /* | |
5114 | * Change a given task's CPU affinity. Migrate the thread to a | |
5115 | * proper CPU and schedule it away if the CPU it's executing on | |
5116 | * is removed from the allowed bitmask. | |
5117 | * | |
5118 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5119 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5120 | * call is not atomic; no spinlocks may be held. |
5121 | */ | |
36c8b586 | 5122 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
1da177e4 | 5123 | { |
70b97a7f | 5124 | struct migration_req req; |
1da177e4 | 5125 | unsigned long flags; |
70b97a7f | 5126 | struct rq *rq; |
48f24c4d | 5127 | int ret = 0; |
1da177e4 LT |
5128 | |
5129 | rq = task_rq_lock(p, &flags); | |
5130 | if (!cpus_intersects(new_mask, cpu_online_map)) { | |
5131 | ret = -EINVAL; | |
5132 | goto out; | |
5133 | } | |
5134 | ||
5135 | p->cpus_allowed = new_mask; | |
5136 | /* Can the task run on the task's current CPU? If so, we're done */ | |
5137 | if (cpu_isset(task_cpu(p), new_mask)) | |
5138 | goto out; | |
5139 | ||
5140 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { | |
5141 | /* Need help from migration thread: drop lock and wait. */ | |
5142 | task_rq_unlock(rq, &flags); | |
5143 | wake_up_process(rq->migration_thread); | |
5144 | wait_for_completion(&req.done); | |
5145 | tlb_migrate_finish(p->mm); | |
5146 | return 0; | |
5147 | } | |
5148 | out: | |
5149 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5150 | |
1da177e4 LT |
5151 | return ret; |
5152 | } | |
1da177e4 LT |
5153 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
5154 | ||
5155 | /* | |
41a2d6cf | 5156 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5157 | * this because either it can't run here any more (set_cpus_allowed() |
5158 | * away from this CPU, or CPU going down), or because we're | |
5159 | * attempting to rebalance this task on exec (sched_exec). | |
5160 | * | |
5161 | * So we race with normal scheduler movements, but that's OK, as long | |
5162 | * as the task is no longer on this CPU. | |
efc30814 KK |
5163 | * |
5164 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5165 | */ |
efc30814 | 5166 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5167 | { |
70b97a7f | 5168 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5169 | int ret = 0, on_rq; |
1da177e4 LT |
5170 | |
5171 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5172 | return ret; |
1da177e4 LT |
5173 | |
5174 | rq_src = cpu_rq(src_cpu); | |
5175 | rq_dest = cpu_rq(dest_cpu); | |
5176 | ||
5177 | double_rq_lock(rq_src, rq_dest); | |
5178 | /* Already moved. */ | |
5179 | if (task_cpu(p) != src_cpu) | |
5180 | goto out; | |
5181 | /* Affinity changed (again). */ | |
5182 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5183 | goto out; | |
5184 | ||
dd41f596 | 5185 | on_rq = p->se.on_rq; |
6e82a3be | 5186 | if (on_rq) |
2e1cb74a | 5187 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5188 | |
1da177e4 | 5189 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5190 | if (on_rq) { |
5191 | activate_task(rq_dest, p, 0); | |
5192 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5193 | } |
efc30814 | 5194 | ret = 1; |
1da177e4 LT |
5195 | out: |
5196 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5197 | return ret; |
1da177e4 LT |
5198 | } |
5199 | ||
5200 | /* | |
5201 | * migration_thread - this is a highprio system thread that performs | |
5202 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5203 | * another runqueue. | |
5204 | */ | |
95cdf3b7 | 5205 | static int migration_thread(void *data) |
1da177e4 | 5206 | { |
1da177e4 | 5207 | int cpu = (long)data; |
70b97a7f | 5208 | struct rq *rq; |
1da177e4 LT |
5209 | |
5210 | rq = cpu_rq(cpu); | |
5211 | BUG_ON(rq->migration_thread != current); | |
5212 | ||
5213 | set_current_state(TASK_INTERRUPTIBLE); | |
5214 | while (!kthread_should_stop()) { | |
70b97a7f | 5215 | struct migration_req *req; |
1da177e4 | 5216 | struct list_head *head; |
1da177e4 | 5217 | |
1da177e4 LT |
5218 | spin_lock_irq(&rq->lock); |
5219 | ||
5220 | if (cpu_is_offline(cpu)) { | |
5221 | spin_unlock_irq(&rq->lock); | |
5222 | goto wait_to_die; | |
5223 | } | |
5224 | ||
5225 | if (rq->active_balance) { | |
5226 | active_load_balance(rq, cpu); | |
5227 | rq->active_balance = 0; | |
5228 | } | |
5229 | ||
5230 | head = &rq->migration_queue; | |
5231 | ||
5232 | if (list_empty(head)) { | |
5233 | spin_unlock_irq(&rq->lock); | |
5234 | schedule(); | |
5235 | set_current_state(TASK_INTERRUPTIBLE); | |
5236 | continue; | |
5237 | } | |
70b97a7f | 5238 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5239 | list_del_init(head->next); |
5240 | ||
674311d5 NP |
5241 | spin_unlock(&rq->lock); |
5242 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5243 | local_irq_enable(); | |
1da177e4 LT |
5244 | |
5245 | complete(&req->done); | |
5246 | } | |
5247 | __set_current_state(TASK_RUNNING); | |
5248 | return 0; | |
5249 | ||
5250 | wait_to_die: | |
5251 | /* Wait for kthread_stop */ | |
5252 | set_current_state(TASK_INTERRUPTIBLE); | |
5253 | while (!kthread_should_stop()) { | |
5254 | schedule(); | |
5255 | set_current_state(TASK_INTERRUPTIBLE); | |
5256 | } | |
5257 | __set_current_state(TASK_RUNNING); | |
5258 | return 0; | |
5259 | } | |
5260 | ||
5261 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5262 | |
5263 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5264 | { | |
5265 | int ret; | |
5266 | ||
5267 | local_irq_disable(); | |
5268 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5269 | local_irq_enable(); | |
5270 | return ret; | |
5271 | } | |
5272 | ||
054b9108 | 5273 | /* |
3a4fa0a2 | 5274 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5275 | * NOTE: interrupts should be disabled by the caller |
5276 | */ | |
48f24c4d | 5277 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5278 | { |
efc30814 | 5279 | unsigned long flags; |
1da177e4 | 5280 | cpumask_t mask; |
70b97a7f IM |
5281 | struct rq *rq; |
5282 | int dest_cpu; | |
1da177e4 | 5283 | |
3a5c359a AK |
5284 | do { |
5285 | /* On same node? */ | |
5286 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5287 | cpus_and(mask, mask, p->cpus_allowed); | |
5288 | dest_cpu = any_online_cpu(mask); | |
5289 | ||
5290 | /* On any allowed CPU? */ | |
5291 | if (dest_cpu == NR_CPUS) | |
5292 | dest_cpu = any_online_cpu(p->cpus_allowed); | |
5293 | ||
5294 | /* No more Mr. Nice Guy. */ | |
5295 | if (dest_cpu == NR_CPUS) { | |
470fd646 CW |
5296 | cpumask_t cpus_allowed = cpuset_cpus_allowed_locked(p); |
5297 | /* | |
5298 | * Try to stay on the same cpuset, where the | |
5299 | * current cpuset may be a subset of all cpus. | |
5300 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5301 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5302 | * called within calls to cpuset_lock/cpuset_unlock. |
5303 | */ | |
3a5c359a | 5304 | rq = task_rq_lock(p, &flags); |
470fd646 | 5305 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5306 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5307 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5308 | |
3a5c359a AK |
5309 | /* |
5310 | * Don't tell them about moving exiting tasks or | |
5311 | * kernel threads (both mm NULL), since they never | |
5312 | * leave kernel. | |
5313 | */ | |
41a2d6cf | 5314 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5315 | printk(KERN_INFO "process %d (%s) no " |
5316 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5317 | task_pid_nr(p), p->comm, dead_cpu); |
5318 | } | |
3a5c359a | 5319 | } |
f7b4cddc | 5320 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5321 | } |
5322 | ||
5323 | /* | |
5324 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5325 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5326 | * for performance reasons the counter is not stricly tracking tasks to | |
5327 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5328 | * to keep the global sum constant after CPU-down: | |
5329 | */ | |
70b97a7f | 5330 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5331 | { |
70b97a7f | 5332 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
1da177e4 LT |
5333 | unsigned long flags; |
5334 | ||
5335 | local_irq_save(flags); | |
5336 | double_rq_lock(rq_src, rq_dest); | |
5337 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5338 | rq_src->nr_uninterruptible = 0; | |
5339 | double_rq_unlock(rq_src, rq_dest); | |
5340 | local_irq_restore(flags); | |
5341 | } | |
5342 | ||
5343 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5344 | static void migrate_live_tasks(int src_cpu) | |
5345 | { | |
48f24c4d | 5346 | struct task_struct *p, *t; |
1da177e4 | 5347 | |
f7b4cddc | 5348 | read_lock(&tasklist_lock); |
1da177e4 | 5349 | |
48f24c4d IM |
5350 | do_each_thread(t, p) { |
5351 | if (p == current) | |
1da177e4 LT |
5352 | continue; |
5353 | ||
48f24c4d IM |
5354 | if (task_cpu(p) == src_cpu) |
5355 | move_task_off_dead_cpu(src_cpu, p); | |
5356 | } while_each_thread(t, p); | |
1da177e4 | 5357 | |
f7b4cddc | 5358 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5359 | } |
5360 | ||
dd41f596 IM |
5361 | /* |
5362 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5363 | * It does so by boosting its priority to highest possible. |
5364 | * Used by CPU offline code. | |
1da177e4 LT |
5365 | */ |
5366 | void sched_idle_next(void) | |
5367 | { | |
48f24c4d | 5368 | int this_cpu = smp_processor_id(); |
70b97a7f | 5369 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5370 | struct task_struct *p = rq->idle; |
5371 | unsigned long flags; | |
5372 | ||
5373 | /* cpu has to be offline */ | |
48f24c4d | 5374 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5375 | |
48f24c4d IM |
5376 | /* |
5377 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5378 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5379 | */ |
5380 | spin_lock_irqsave(&rq->lock, flags); | |
5381 | ||
dd41f596 | 5382 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5383 | |
94bc9a7b DA |
5384 | update_rq_clock(rq); |
5385 | activate_task(rq, p, 0); | |
1da177e4 LT |
5386 | |
5387 | spin_unlock_irqrestore(&rq->lock, flags); | |
5388 | } | |
5389 | ||
48f24c4d IM |
5390 | /* |
5391 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5392 | * offline. |
5393 | */ | |
5394 | void idle_task_exit(void) | |
5395 | { | |
5396 | struct mm_struct *mm = current->active_mm; | |
5397 | ||
5398 | BUG_ON(cpu_online(smp_processor_id())); | |
5399 | ||
5400 | if (mm != &init_mm) | |
5401 | switch_mm(mm, &init_mm, current); | |
5402 | mmdrop(mm); | |
5403 | } | |
5404 | ||
054b9108 | 5405 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5406 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5407 | { |
70b97a7f | 5408 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5409 | |
5410 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5411 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5412 | |
5413 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5414 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5415 | |
48f24c4d | 5416 | get_task_struct(p); |
1da177e4 LT |
5417 | |
5418 | /* | |
5419 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5420 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5421 | * fine. |
5422 | */ | |
f7b4cddc | 5423 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5424 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5425 | spin_lock_irq(&rq->lock); |
1da177e4 | 5426 | |
48f24c4d | 5427 | put_task_struct(p); |
1da177e4 LT |
5428 | } |
5429 | ||
5430 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5431 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5432 | { | |
70b97a7f | 5433 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5434 | struct task_struct *next; |
48f24c4d | 5435 | |
dd41f596 IM |
5436 | for ( ; ; ) { |
5437 | if (!rq->nr_running) | |
5438 | break; | |
a8e504d2 | 5439 | update_rq_clock(rq); |
ff95f3df | 5440 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5441 | if (!next) |
5442 | break; | |
5443 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5444 | |
1da177e4 LT |
5445 | } |
5446 | } | |
5447 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5448 | ||
e692ab53 NP |
5449 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5450 | ||
5451 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5452 | { |
5453 | .procname = "sched_domain", | |
c57baf1e | 5454 | .mode = 0555, |
e0361851 | 5455 | }, |
38605cae | 5456 | {0, }, |
e692ab53 NP |
5457 | }; |
5458 | ||
5459 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5460 | { |
c57baf1e | 5461 | .ctl_name = CTL_KERN, |
e0361851 | 5462 | .procname = "kernel", |
c57baf1e | 5463 | .mode = 0555, |
e0361851 AD |
5464 | .child = sd_ctl_dir, |
5465 | }, | |
38605cae | 5466 | {0, }, |
e692ab53 NP |
5467 | }; |
5468 | ||
5469 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5470 | { | |
5471 | struct ctl_table *entry = | |
5cf9f062 | 5472 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5473 | |
e692ab53 NP |
5474 | return entry; |
5475 | } | |
5476 | ||
6382bc90 MM |
5477 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5478 | { | |
cd790076 | 5479 | struct ctl_table *entry; |
6382bc90 | 5480 | |
cd790076 MM |
5481 | /* |
5482 | * In the intermediate directories, both the child directory and | |
5483 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5484 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5485 | * static strings and all have proc handlers. |
5486 | */ | |
5487 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5488 | if (entry->child) |
5489 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5490 | if (entry->proc_handler == NULL) |
5491 | kfree(entry->procname); | |
5492 | } | |
6382bc90 MM |
5493 | |
5494 | kfree(*tablep); | |
5495 | *tablep = NULL; | |
5496 | } | |
5497 | ||
e692ab53 | 5498 | static void |
e0361851 | 5499 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5500 | const char *procname, void *data, int maxlen, |
5501 | mode_t mode, proc_handler *proc_handler) | |
5502 | { | |
e692ab53 NP |
5503 | entry->procname = procname; |
5504 | entry->data = data; | |
5505 | entry->maxlen = maxlen; | |
5506 | entry->mode = mode; | |
5507 | entry->proc_handler = proc_handler; | |
5508 | } | |
5509 | ||
5510 | static struct ctl_table * | |
5511 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5512 | { | |
ace8b3d6 | 5513 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 5514 | |
ad1cdc1d MM |
5515 | if (table == NULL) |
5516 | return NULL; | |
5517 | ||
e0361851 | 5518 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5519 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5520 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5521 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5522 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5523 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5524 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5525 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5526 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5527 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5528 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5529 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5530 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5531 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5532 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5533 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5534 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5535 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5536 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5537 | &sd->cache_nice_tries, |
5538 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5539 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5540 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 5541 | /* &table[11] is terminator */ |
e692ab53 NP |
5542 | |
5543 | return table; | |
5544 | } | |
5545 | ||
9a4e7159 | 5546 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5547 | { |
5548 | struct ctl_table *entry, *table; | |
5549 | struct sched_domain *sd; | |
5550 | int domain_num = 0, i; | |
5551 | char buf[32]; | |
5552 | ||
5553 | for_each_domain(cpu, sd) | |
5554 | domain_num++; | |
5555 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5556 | if (table == NULL) |
5557 | return NULL; | |
e692ab53 NP |
5558 | |
5559 | i = 0; | |
5560 | for_each_domain(cpu, sd) { | |
5561 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5562 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5563 | entry->mode = 0555; |
e692ab53 NP |
5564 | entry->child = sd_alloc_ctl_domain_table(sd); |
5565 | entry++; | |
5566 | i++; | |
5567 | } | |
5568 | return table; | |
5569 | } | |
5570 | ||
5571 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5572 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
5573 | { |
5574 | int i, cpu_num = num_online_cpus(); | |
5575 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
5576 | char buf[32]; | |
5577 | ||
7378547f MM |
5578 | WARN_ON(sd_ctl_dir[0].child); |
5579 | sd_ctl_dir[0].child = entry; | |
5580 | ||
ad1cdc1d MM |
5581 | if (entry == NULL) |
5582 | return; | |
5583 | ||
97b6ea7b | 5584 | for_each_online_cpu(i) { |
e692ab53 | 5585 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5586 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5587 | entry->mode = 0555; |
e692ab53 | 5588 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5589 | entry++; |
e692ab53 | 5590 | } |
7378547f MM |
5591 | |
5592 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5593 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5594 | } | |
6382bc90 | 5595 | |
7378547f | 5596 | /* may be called multiple times per register */ |
6382bc90 MM |
5597 | static void unregister_sched_domain_sysctl(void) |
5598 | { | |
7378547f MM |
5599 | if (sd_sysctl_header) |
5600 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5601 | sd_sysctl_header = NULL; |
7378547f MM |
5602 | if (sd_ctl_dir[0].child) |
5603 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5604 | } |
e692ab53 | 5605 | #else |
6382bc90 MM |
5606 | static void register_sched_domain_sysctl(void) |
5607 | { | |
5608 | } | |
5609 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5610 | { |
5611 | } | |
5612 | #endif | |
5613 | ||
1da177e4 LT |
5614 | /* |
5615 | * migration_call - callback that gets triggered when a CPU is added. | |
5616 | * Here we can start up the necessary migration thread for the new CPU. | |
5617 | */ | |
48f24c4d IM |
5618 | static int __cpuinit |
5619 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5620 | { |
1da177e4 | 5621 | struct task_struct *p; |
48f24c4d | 5622 | int cpu = (long)hcpu; |
1da177e4 | 5623 | unsigned long flags; |
70b97a7f | 5624 | struct rq *rq; |
1da177e4 LT |
5625 | |
5626 | switch (action) { | |
5be9361c | 5627 | |
1da177e4 | 5628 | case CPU_UP_PREPARE: |
8bb78442 | 5629 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5630 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5631 | if (IS_ERR(p)) |
5632 | return NOTIFY_BAD; | |
1da177e4 LT |
5633 | kthread_bind(p, cpu); |
5634 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5635 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5636 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
5637 | task_rq_unlock(rq, &flags); |
5638 | cpu_rq(cpu)->migration_thread = p; | |
5639 | break; | |
48f24c4d | 5640 | |
1da177e4 | 5641 | case CPU_ONLINE: |
8bb78442 | 5642 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5643 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 LT |
5644 | wake_up_process(cpu_rq(cpu)->migration_thread); |
5645 | break; | |
48f24c4d | 5646 | |
1da177e4 LT |
5647 | #ifdef CONFIG_HOTPLUG_CPU |
5648 | case CPU_UP_CANCELED: | |
8bb78442 | 5649 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5650 | if (!cpu_rq(cpu)->migration_thread) |
5651 | break; | |
41a2d6cf | 5652 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
5653 | kthread_bind(cpu_rq(cpu)->migration_thread, |
5654 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
5655 | kthread_stop(cpu_rq(cpu)->migration_thread); |
5656 | cpu_rq(cpu)->migration_thread = NULL; | |
5657 | break; | |
48f24c4d | 5658 | |
1da177e4 | 5659 | case CPU_DEAD: |
8bb78442 | 5660 | case CPU_DEAD_FROZEN: |
470fd646 | 5661 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5662 | migrate_live_tasks(cpu); |
5663 | rq = cpu_rq(cpu); | |
5664 | kthread_stop(rq->migration_thread); | |
5665 | rq->migration_thread = NULL; | |
5666 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 5667 | spin_lock_irq(&rq->lock); |
a8e504d2 | 5668 | update_rq_clock(rq); |
2e1cb74a | 5669 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 5670 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
5671 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5672 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5673 | migrate_dead_tasks(cpu); |
d2da272a | 5674 | spin_unlock_irq(&rq->lock); |
470fd646 | 5675 | cpuset_unlock(); |
1da177e4 LT |
5676 | migrate_nr_uninterruptible(rq); |
5677 | BUG_ON(rq->nr_running != 0); | |
5678 | ||
41a2d6cf IM |
5679 | /* |
5680 | * No need to migrate the tasks: it was best-effort if | |
5681 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5682 | * the requestors. | |
5683 | */ | |
1da177e4 LT |
5684 | spin_lock_irq(&rq->lock); |
5685 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
5686 | struct migration_req *req; |
5687 | ||
1da177e4 | 5688 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5689 | struct migration_req, list); |
1da177e4 LT |
5690 | list_del_init(&req->list); |
5691 | complete(&req->done); | |
5692 | } | |
5693 | spin_unlock_irq(&rq->lock); | |
5694 | break; | |
5695 | #endif | |
5696 | } | |
5697 | return NOTIFY_OK; | |
5698 | } | |
5699 | ||
5700 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
5701 | * happens before everything else. | |
5702 | */ | |
26c2143b | 5703 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5704 | .notifier_call = migration_call, |
5705 | .priority = 10 | |
5706 | }; | |
5707 | ||
e6fe6649 | 5708 | void __init migration_init(void) |
1da177e4 LT |
5709 | { |
5710 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5711 | int err; |
48f24c4d IM |
5712 | |
5713 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5714 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5715 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5716 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5717 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
5718 | } |
5719 | #endif | |
5720 | ||
5721 | #ifdef CONFIG_SMP | |
476f3534 CL |
5722 | |
5723 | /* Number of possible processor ids */ | |
5724 | int nr_cpu_ids __read_mostly = NR_CPUS; | |
5725 | EXPORT_SYMBOL(nr_cpu_ids); | |
5726 | ||
3e9830dc | 5727 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff IM |
5728 | |
5729 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level) | |
1da177e4 | 5730 | { |
4dcf6aff IM |
5731 | struct sched_group *group = sd->groups; |
5732 | cpumask_t groupmask; | |
5733 | char str[NR_CPUS]; | |
1da177e4 | 5734 | |
4dcf6aff IM |
5735 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
5736 | cpus_clear(groupmask); | |
5737 | ||
5738 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5739 | ||
5740 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
5741 | printk("does not load-balance\n"); | |
5742 | if (sd->parent) | |
5743 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
5744 | " has parent"); | |
5745 | return -1; | |
41c7ce9a NP |
5746 | } |
5747 | ||
4dcf6aff IM |
5748 | printk(KERN_CONT "span %s\n", str); |
5749 | ||
5750 | if (!cpu_isset(cpu, sd->span)) { | |
5751 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
5752 | "CPU%d\n", cpu); | |
5753 | } | |
5754 | if (!cpu_isset(cpu, group->cpumask)) { | |
5755 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
5756 | " CPU%d\n", cpu); | |
5757 | } | |
1da177e4 | 5758 | |
4dcf6aff | 5759 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5760 | do { |
4dcf6aff IM |
5761 | if (!group) { |
5762 | printk("\n"); | |
5763 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5764 | break; |
5765 | } | |
5766 | ||
4dcf6aff IM |
5767 | if (!group->__cpu_power) { |
5768 | printk(KERN_CONT "\n"); | |
5769 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5770 | "set\n"); | |
5771 | break; | |
5772 | } | |
1da177e4 | 5773 | |
4dcf6aff IM |
5774 | if (!cpus_weight(group->cpumask)) { |
5775 | printk(KERN_CONT "\n"); | |
5776 | printk(KERN_ERR "ERROR: empty group\n"); | |
5777 | break; | |
5778 | } | |
1da177e4 | 5779 | |
4dcf6aff IM |
5780 | if (cpus_intersects(groupmask, group->cpumask)) { |
5781 | printk(KERN_CONT "\n"); | |
5782 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
5783 | break; | |
5784 | } | |
1da177e4 | 5785 | |
4dcf6aff | 5786 | cpus_or(groupmask, groupmask, group->cpumask); |
1da177e4 | 5787 | |
4dcf6aff IM |
5788 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
5789 | printk(KERN_CONT " %s", str); | |
1da177e4 | 5790 | |
4dcf6aff IM |
5791 | group = group->next; |
5792 | } while (group != sd->groups); | |
5793 | printk(KERN_CONT "\n"); | |
1da177e4 | 5794 | |
4dcf6aff IM |
5795 | if (!cpus_equal(sd->span, groupmask)) |
5796 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); | |
1da177e4 | 5797 | |
4dcf6aff IM |
5798 | if (sd->parent && !cpus_subset(groupmask, sd->parent->span)) |
5799 | printk(KERN_ERR "ERROR: parent span is not a superset " | |
5800 | "of domain->span\n"); | |
5801 | return 0; | |
5802 | } | |
1da177e4 | 5803 | |
4dcf6aff IM |
5804 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5805 | { | |
5806 | int level = 0; | |
1da177e4 | 5807 | |
4dcf6aff IM |
5808 | if (!sd) { |
5809 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5810 | return; | |
5811 | } | |
1da177e4 | 5812 | |
4dcf6aff IM |
5813 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5814 | ||
5815 | for (;;) { | |
5816 | if (sched_domain_debug_one(sd, cpu, level)) | |
5817 | break; | |
1da177e4 LT |
5818 | level++; |
5819 | sd = sd->parent; | |
33859f7f | 5820 | if (!sd) |
4dcf6aff IM |
5821 | break; |
5822 | } | |
1da177e4 LT |
5823 | } |
5824 | #else | |
48f24c4d | 5825 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
5826 | #endif |
5827 | ||
1a20ff27 | 5828 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
5829 | { |
5830 | if (cpus_weight(sd->span) == 1) | |
5831 | return 1; | |
5832 | ||
5833 | /* Following flags need at least 2 groups */ | |
5834 | if (sd->flags & (SD_LOAD_BALANCE | | |
5835 | SD_BALANCE_NEWIDLE | | |
5836 | SD_BALANCE_FORK | | |
89c4710e SS |
5837 | SD_BALANCE_EXEC | |
5838 | SD_SHARE_CPUPOWER | | |
5839 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5840 | if (sd->groups != sd->groups->next) |
5841 | return 0; | |
5842 | } | |
5843 | ||
5844 | /* Following flags don't use groups */ | |
5845 | if (sd->flags & (SD_WAKE_IDLE | | |
5846 | SD_WAKE_AFFINE | | |
5847 | SD_WAKE_BALANCE)) | |
5848 | return 0; | |
5849 | ||
5850 | return 1; | |
5851 | } | |
5852 | ||
48f24c4d IM |
5853 | static int |
5854 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5855 | { |
5856 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5857 | ||
5858 | if (sd_degenerate(parent)) | |
5859 | return 1; | |
5860 | ||
5861 | if (!cpus_equal(sd->span, parent->span)) | |
5862 | return 0; | |
5863 | ||
5864 | /* Does parent contain flags not in child? */ | |
5865 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
5866 | if (cflags & SD_WAKE_AFFINE) | |
5867 | pflags &= ~SD_WAKE_BALANCE; | |
5868 | /* Flags needing groups don't count if only 1 group in parent */ | |
5869 | if (parent->groups == parent->groups->next) { | |
5870 | pflags &= ~(SD_LOAD_BALANCE | | |
5871 | SD_BALANCE_NEWIDLE | | |
5872 | SD_BALANCE_FORK | | |
89c4710e SS |
5873 | SD_BALANCE_EXEC | |
5874 | SD_SHARE_CPUPOWER | | |
5875 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
5876 | } |
5877 | if (~cflags & pflags) | |
5878 | return 0; | |
5879 | ||
5880 | return 1; | |
5881 | } | |
5882 | ||
1da177e4 LT |
5883 | /* |
5884 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must | |
5885 | * hold the hotplug lock. | |
5886 | */ | |
9c1cfda2 | 5887 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
1da177e4 | 5888 | { |
70b97a7f | 5889 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
5890 | struct sched_domain *tmp; |
5891 | ||
5892 | /* Remove the sched domains which do not contribute to scheduling. */ | |
5893 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
5894 | struct sched_domain *parent = tmp->parent; | |
5895 | if (!parent) | |
5896 | break; | |
1a848870 | 5897 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 5898 | tmp->parent = parent->parent; |
1a848870 SS |
5899 | if (parent->parent) |
5900 | parent->parent->child = tmp; | |
5901 | } | |
245af2c7 SS |
5902 | } |
5903 | ||
1a848870 | 5904 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 5905 | sd = sd->parent; |
1a848870 SS |
5906 | if (sd) |
5907 | sd->child = NULL; | |
5908 | } | |
1da177e4 LT |
5909 | |
5910 | sched_domain_debug(sd, cpu); | |
5911 | ||
674311d5 | 5912 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
5913 | } |
5914 | ||
5915 | /* cpus with isolated domains */ | |
67af63a6 | 5916 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
5917 | |
5918 | /* Setup the mask of cpus configured for isolated domains */ | |
5919 | static int __init isolated_cpu_setup(char *str) | |
5920 | { | |
5921 | int ints[NR_CPUS], i; | |
5922 | ||
5923 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
5924 | cpus_clear(cpu_isolated_map); | |
5925 | for (i = 1; i <= ints[0]; i++) | |
5926 | if (ints[i] < NR_CPUS) | |
5927 | cpu_set(ints[i], cpu_isolated_map); | |
5928 | return 1; | |
5929 | } | |
5930 | ||
8927f494 | 5931 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
5932 | |
5933 | /* | |
6711cab4 SS |
5934 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
5935 | * to a function which identifies what group(along with sched group) a CPU | |
5936 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
5937 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
5938 | * |
5939 | * init_sched_build_groups will build a circular linked list of the groups | |
5940 | * covered by the given span, and will set each group's ->cpumask correctly, | |
5941 | * and ->cpu_power to 0. | |
5942 | */ | |
a616058b | 5943 | static void |
6711cab4 SS |
5944 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
5945 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, | |
5946 | struct sched_group **sg)) | |
1da177e4 LT |
5947 | { |
5948 | struct sched_group *first = NULL, *last = NULL; | |
5949 | cpumask_t covered = CPU_MASK_NONE; | |
5950 | int i; | |
5951 | ||
5952 | for_each_cpu_mask(i, span) { | |
6711cab4 SS |
5953 | struct sched_group *sg; |
5954 | int group = group_fn(i, cpu_map, &sg); | |
1da177e4 LT |
5955 | int j; |
5956 | ||
5957 | if (cpu_isset(i, covered)) | |
5958 | continue; | |
5959 | ||
5960 | sg->cpumask = CPU_MASK_NONE; | |
5517d86b | 5961 | sg->__cpu_power = 0; |
1da177e4 LT |
5962 | |
5963 | for_each_cpu_mask(j, span) { | |
6711cab4 | 5964 | if (group_fn(j, cpu_map, NULL) != group) |
1da177e4 LT |
5965 | continue; |
5966 | ||
5967 | cpu_set(j, covered); | |
5968 | cpu_set(j, sg->cpumask); | |
5969 | } | |
5970 | if (!first) | |
5971 | first = sg; | |
5972 | if (last) | |
5973 | last->next = sg; | |
5974 | last = sg; | |
5975 | } | |
5976 | last->next = first; | |
5977 | } | |
5978 | ||
9c1cfda2 | 5979 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 5980 | |
9c1cfda2 | 5981 | #ifdef CONFIG_NUMA |
198e2f18 | 5982 | |
9c1cfda2 JH |
5983 | /** |
5984 | * find_next_best_node - find the next node to include in a sched_domain | |
5985 | * @node: node whose sched_domain we're building | |
5986 | * @used_nodes: nodes already in the sched_domain | |
5987 | * | |
41a2d6cf | 5988 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
5989 | * finds the closest node not already in the @used_nodes map. |
5990 | * | |
5991 | * Should use nodemask_t. | |
5992 | */ | |
5993 | static int find_next_best_node(int node, unsigned long *used_nodes) | |
5994 | { | |
5995 | int i, n, val, min_val, best_node = 0; | |
5996 | ||
5997 | min_val = INT_MAX; | |
5998 | ||
5999 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6000 | /* Start at @node */ | |
6001 | n = (node + i) % MAX_NUMNODES; | |
6002 | ||
6003 | if (!nr_cpus_node(n)) | |
6004 | continue; | |
6005 | ||
6006 | /* Skip already used nodes */ | |
6007 | if (test_bit(n, used_nodes)) | |
6008 | continue; | |
6009 | ||
6010 | /* Simple min distance search */ | |
6011 | val = node_distance(node, n); | |
6012 | ||
6013 | if (val < min_val) { | |
6014 | min_val = val; | |
6015 | best_node = n; | |
6016 | } | |
6017 | } | |
6018 | ||
6019 | set_bit(best_node, used_nodes); | |
6020 | return best_node; | |
6021 | } | |
6022 | ||
6023 | /** | |
6024 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6025 | * @node: node whose cpumask we're constructing | |
6026 | * @size: number of nodes to include in this span | |
6027 | * | |
41a2d6cf | 6028 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6029 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6030 | * out optimally. | |
6031 | */ | |
6032 | static cpumask_t sched_domain_node_span(int node) | |
6033 | { | |
9c1cfda2 | 6034 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
48f24c4d IM |
6035 | cpumask_t span, nodemask; |
6036 | int i; | |
9c1cfda2 JH |
6037 | |
6038 | cpus_clear(span); | |
6039 | bitmap_zero(used_nodes, MAX_NUMNODES); | |
6040 | ||
6041 | nodemask = node_to_cpumask(node); | |
6042 | cpus_or(span, span, nodemask); | |
6043 | set_bit(node, used_nodes); | |
6044 | ||
6045 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
6046 | int next_node = find_next_best_node(node, used_nodes); | |
48f24c4d | 6047 | |
9c1cfda2 JH |
6048 | nodemask = node_to_cpumask(next_node); |
6049 | cpus_or(span, span, nodemask); | |
6050 | } | |
6051 | ||
6052 | return span; | |
6053 | } | |
6054 | #endif | |
6055 | ||
5c45bf27 | 6056 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6057 | |
9c1cfda2 | 6058 | /* |
48f24c4d | 6059 | * SMT sched-domains: |
9c1cfda2 | 6060 | */ |
1da177e4 LT |
6061 | #ifdef CONFIG_SCHED_SMT |
6062 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6063 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6064 | |
41a2d6cf IM |
6065 | static int |
6066 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6067 | { |
6711cab4 SS |
6068 | if (sg) |
6069 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6070 | return cpu; |
6071 | } | |
6072 | #endif | |
6073 | ||
48f24c4d IM |
6074 | /* |
6075 | * multi-core sched-domains: | |
6076 | */ | |
1e9f28fa SS |
6077 | #ifdef CONFIG_SCHED_MC |
6078 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6079 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6080 | #endif |
6081 | ||
6082 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf IM |
6083 | static int |
6084 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6085 | { |
6711cab4 | 6086 | int group; |
d5a7430d | 6087 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6088 | cpus_and(mask, mask, *cpu_map); |
6711cab4 SS |
6089 | group = first_cpu(mask); |
6090 | if (sg) | |
6091 | *sg = &per_cpu(sched_group_core, group); | |
6092 | return group; | |
1e9f28fa SS |
6093 | } |
6094 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf IM |
6095 | static int |
6096 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1e9f28fa | 6097 | { |
6711cab4 SS |
6098 | if (sg) |
6099 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6100 | return cpu; |
6101 | } | |
6102 | #endif | |
6103 | ||
1da177e4 | 6104 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6105 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6106 | |
41a2d6cf IM |
6107 | static int |
6108 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg) | |
1da177e4 | 6109 | { |
6711cab4 | 6110 | int group; |
48f24c4d | 6111 | #ifdef CONFIG_SCHED_MC |
1e9f28fa | 6112 | cpumask_t mask = cpu_coregroup_map(cpu); |
a616058b | 6113 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6114 | group = first_cpu(mask); |
1e9f28fa | 6115 | #elif defined(CONFIG_SCHED_SMT) |
d5a7430d | 6116 | cpumask_t mask = per_cpu(cpu_sibling_map, cpu); |
a616058b | 6117 | cpus_and(mask, mask, *cpu_map); |
6711cab4 | 6118 | group = first_cpu(mask); |
1da177e4 | 6119 | #else |
6711cab4 | 6120 | group = cpu; |
1da177e4 | 6121 | #endif |
6711cab4 SS |
6122 | if (sg) |
6123 | *sg = &per_cpu(sched_group_phys, group); | |
6124 | return group; | |
1da177e4 LT |
6125 | } |
6126 | ||
6127 | #ifdef CONFIG_NUMA | |
1da177e4 | 6128 | /* |
9c1cfda2 JH |
6129 | * The init_sched_build_groups can't handle what we want to do with node |
6130 | * groups, so roll our own. Now each node has its own list of groups which | |
6131 | * gets dynamically allocated. | |
1da177e4 | 6132 | */ |
9c1cfda2 | 6133 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
d1b55138 | 6134 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
1da177e4 | 6135 | |
9c1cfda2 | 6136 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6137 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6138 | |
6711cab4 SS |
6139 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
6140 | struct sched_group **sg) | |
9c1cfda2 | 6141 | { |
6711cab4 SS |
6142 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6143 | int group; | |
6144 | ||
6145 | cpus_and(nodemask, nodemask, *cpu_map); | |
6146 | group = first_cpu(nodemask); | |
6147 | ||
6148 | if (sg) | |
6149 | *sg = &per_cpu(sched_group_allnodes, group); | |
6150 | return group; | |
1da177e4 | 6151 | } |
6711cab4 | 6152 | |
08069033 SS |
6153 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6154 | { | |
6155 | struct sched_group *sg = group_head; | |
6156 | int j; | |
6157 | ||
6158 | if (!sg) | |
6159 | return; | |
3a5c359a AK |
6160 | do { |
6161 | for_each_cpu_mask(j, sg->cpumask) { | |
6162 | struct sched_domain *sd; | |
08069033 | 6163 | |
3a5c359a AK |
6164 | sd = &per_cpu(phys_domains, j); |
6165 | if (j != first_cpu(sd->groups->cpumask)) { | |
6166 | /* | |
6167 | * Only add "power" once for each | |
6168 | * physical package. | |
6169 | */ | |
6170 | continue; | |
6171 | } | |
08069033 | 6172 | |
3a5c359a AK |
6173 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6174 | } | |
6175 | sg = sg->next; | |
6176 | } while (sg != group_head); | |
08069033 | 6177 | } |
1da177e4 LT |
6178 | #endif |
6179 | ||
a616058b | 6180 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6181 | /* Free memory allocated for various sched_group structures */ |
6182 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6183 | { | |
a616058b | 6184 | int cpu, i; |
51888ca2 SV |
6185 | |
6186 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6187 | struct sched_group **sched_group_nodes |
6188 | = sched_group_nodes_bycpu[cpu]; | |
6189 | ||
51888ca2 SV |
6190 | if (!sched_group_nodes) |
6191 | continue; | |
6192 | ||
6193 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6194 | cpumask_t nodemask = node_to_cpumask(i); | |
6195 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; | |
6196 | ||
6197 | cpus_and(nodemask, nodemask, *cpu_map); | |
6198 | if (cpus_empty(nodemask)) | |
6199 | continue; | |
6200 | ||
6201 | if (sg == NULL) | |
6202 | continue; | |
6203 | sg = sg->next; | |
6204 | next_sg: | |
6205 | oldsg = sg; | |
6206 | sg = sg->next; | |
6207 | kfree(oldsg); | |
6208 | if (oldsg != sched_group_nodes[i]) | |
6209 | goto next_sg; | |
6210 | } | |
6211 | kfree(sched_group_nodes); | |
6212 | sched_group_nodes_bycpu[cpu] = NULL; | |
6213 | } | |
51888ca2 | 6214 | } |
a616058b SS |
6215 | #else |
6216 | static void free_sched_groups(const cpumask_t *cpu_map) | |
6217 | { | |
6218 | } | |
6219 | #endif | |
51888ca2 | 6220 | |
89c4710e SS |
6221 | /* |
6222 | * Initialize sched groups cpu_power. | |
6223 | * | |
6224 | * cpu_power indicates the capacity of sched group, which is used while | |
6225 | * distributing the load between different sched groups in a sched domain. | |
6226 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6227 | * there are asymmetries in the topology. If there are asymmetries, group | |
6228 | * having more cpu_power will pickup more load compared to the group having | |
6229 | * less cpu_power. | |
6230 | * | |
6231 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6232 | * the maximum number of tasks a group can handle in the presence of other idle | |
6233 | * or lightly loaded groups in the same sched domain. | |
6234 | */ | |
6235 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6236 | { | |
6237 | struct sched_domain *child; | |
6238 | struct sched_group *group; | |
6239 | ||
6240 | WARN_ON(!sd || !sd->groups); | |
6241 | ||
6242 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6243 | return; | |
6244 | ||
6245 | child = sd->child; | |
6246 | ||
5517d86b ED |
6247 | sd->groups->__cpu_power = 0; |
6248 | ||
89c4710e SS |
6249 | /* |
6250 | * For perf policy, if the groups in child domain share resources | |
6251 | * (for example cores sharing some portions of the cache hierarchy | |
6252 | * or SMT), then set this domain groups cpu_power such that each group | |
6253 | * can handle only one task, when there are other idle groups in the | |
6254 | * same sched domain. | |
6255 | */ | |
6256 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6257 | (child->flags & | |
6258 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6259 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6260 | return; |
6261 | } | |
6262 | ||
89c4710e SS |
6263 | /* |
6264 | * add cpu_power of each child group to this groups cpu_power | |
6265 | */ | |
6266 | group = child->groups; | |
6267 | do { | |
5517d86b | 6268 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6269 | group = group->next; |
6270 | } while (group != child->groups); | |
6271 | } | |
6272 | ||
1da177e4 | 6273 | /* |
1a20ff27 DG |
6274 | * Build sched domains for a given set of cpus and attach the sched domains |
6275 | * to the individual cpus | |
1da177e4 | 6276 | */ |
51888ca2 | 6277 | static int build_sched_domains(const cpumask_t *cpu_map) |
1da177e4 LT |
6278 | { |
6279 | int i; | |
d1b55138 JH |
6280 | #ifdef CONFIG_NUMA |
6281 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6282 | int sd_allnodes = 0; |
d1b55138 JH |
6283 | |
6284 | /* | |
6285 | * Allocate the per-node list of sched groups | |
6286 | */ | |
5cf9f062 | 6287 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6288 | GFP_KERNEL); |
d1b55138 JH |
6289 | if (!sched_group_nodes) { |
6290 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6291 | return -ENOMEM; |
d1b55138 JH |
6292 | } |
6293 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
6294 | #endif | |
1da177e4 LT |
6295 | |
6296 | /* | |
1a20ff27 | 6297 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6298 | */ |
1a20ff27 | 6299 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6300 | struct sched_domain *sd = NULL, *p; |
6301 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); | |
6302 | ||
1a20ff27 | 6303 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6304 | |
6305 | #ifdef CONFIG_NUMA | |
dd41f596 IM |
6306 | if (cpus_weight(*cpu_map) > |
6307 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { | |
9c1cfda2 JH |
6308 | sd = &per_cpu(allnodes_domains, i); |
6309 | *sd = SD_ALLNODES_INIT; | |
6310 | sd->span = *cpu_map; | |
6711cab4 | 6311 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
9c1cfda2 | 6312 | p = sd; |
6711cab4 | 6313 | sd_allnodes = 1; |
9c1cfda2 JH |
6314 | } else |
6315 | p = NULL; | |
6316 | ||
1da177e4 | 6317 | sd = &per_cpu(node_domains, i); |
1da177e4 | 6318 | *sd = SD_NODE_INIT; |
9c1cfda2 JH |
6319 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
6320 | sd->parent = p; | |
1a848870 SS |
6321 | if (p) |
6322 | p->child = sd; | |
9c1cfda2 | 6323 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
6324 | #endif |
6325 | ||
6326 | p = sd; | |
6327 | sd = &per_cpu(phys_domains, i); | |
1da177e4 LT |
6328 | *sd = SD_CPU_INIT; |
6329 | sd->span = nodemask; | |
6330 | sd->parent = p; | |
1a848870 SS |
6331 | if (p) |
6332 | p->child = sd; | |
6711cab4 | 6333 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
1da177e4 | 6334 | |
1e9f28fa SS |
6335 | #ifdef CONFIG_SCHED_MC |
6336 | p = sd; | |
6337 | sd = &per_cpu(core_domains, i); | |
1e9f28fa SS |
6338 | *sd = SD_MC_INIT; |
6339 | sd->span = cpu_coregroup_map(i); | |
6340 | cpus_and(sd->span, sd->span, *cpu_map); | |
6341 | sd->parent = p; | |
1a848870 | 6342 | p->child = sd; |
6711cab4 | 6343 | cpu_to_core_group(i, cpu_map, &sd->groups); |
1e9f28fa SS |
6344 | #endif |
6345 | ||
1da177e4 LT |
6346 | #ifdef CONFIG_SCHED_SMT |
6347 | p = sd; | |
6348 | sd = &per_cpu(cpu_domains, i); | |
1da177e4 | 6349 | *sd = SD_SIBLING_INIT; |
d5a7430d | 6350 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6351 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 6352 | sd->parent = p; |
1a848870 | 6353 | p->child = sd; |
6711cab4 | 6354 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
1da177e4 LT |
6355 | #endif |
6356 | } | |
6357 | ||
6358 | #ifdef CONFIG_SCHED_SMT | |
6359 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 6360 | for_each_cpu_mask(i, *cpu_map) { |
d5a7430d | 6361 | cpumask_t this_sibling_map = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 6362 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
1da177e4 LT |
6363 | if (i != first_cpu(this_sibling_map)) |
6364 | continue; | |
6365 | ||
dd41f596 IM |
6366 | init_sched_build_groups(this_sibling_map, cpu_map, |
6367 | &cpu_to_cpu_group); | |
1da177e4 LT |
6368 | } |
6369 | #endif | |
6370 | ||
1e9f28fa SS |
6371 | #ifdef CONFIG_SCHED_MC |
6372 | /* Set up multi-core groups */ | |
6373 | for_each_cpu_mask(i, *cpu_map) { | |
6374 | cpumask_t this_core_map = cpu_coregroup_map(i); | |
6375 | cpus_and(this_core_map, this_core_map, *cpu_map); | |
6376 | if (i != first_cpu(this_core_map)) | |
6377 | continue; | |
dd41f596 IM |
6378 | init_sched_build_groups(this_core_map, cpu_map, |
6379 | &cpu_to_core_group); | |
1e9f28fa SS |
6380 | } |
6381 | #endif | |
6382 | ||
1da177e4 LT |
6383 | /* Set up physical groups */ |
6384 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6385 | cpumask_t nodemask = node_to_cpumask(i); | |
6386 | ||
1a20ff27 | 6387 | cpus_and(nodemask, nodemask, *cpu_map); |
1da177e4 LT |
6388 | if (cpus_empty(nodemask)) |
6389 | continue; | |
6390 | ||
6711cab4 | 6391 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
1da177e4 LT |
6392 | } |
6393 | ||
6394 | #ifdef CONFIG_NUMA | |
6395 | /* Set up node groups */ | |
6711cab4 | 6396 | if (sd_allnodes) |
dd41f596 IM |
6397 | init_sched_build_groups(*cpu_map, cpu_map, |
6398 | &cpu_to_allnodes_group); | |
9c1cfda2 JH |
6399 | |
6400 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6401 | /* Set up node groups */ | |
6402 | struct sched_group *sg, *prev; | |
6403 | cpumask_t nodemask = node_to_cpumask(i); | |
6404 | cpumask_t domainspan; | |
6405 | cpumask_t covered = CPU_MASK_NONE; | |
6406 | int j; | |
6407 | ||
6408 | cpus_and(nodemask, nodemask, *cpu_map); | |
d1b55138 JH |
6409 | if (cpus_empty(nodemask)) { |
6410 | sched_group_nodes[i] = NULL; | |
9c1cfda2 | 6411 | continue; |
d1b55138 | 6412 | } |
9c1cfda2 JH |
6413 | |
6414 | domainspan = sched_domain_node_span(i); | |
6415 | cpus_and(domainspan, domainspan, *cpu_map); | |
6416 | ||
15f0b676 | 6417 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
6418 | if (!sg) { |
6419 | printk(KERN_WARNING "Can not alloc domain group for " | |
6420 | "node %d\n", i); | |
6421 | goto error; | |
6422 | } | |
9c1cfda2 JH |
6423 | sched_group_nodes[i] = sg; |
6424 | for_each_cpu_mask(j, nodemask) { | |
6425 | struct sched_domain *sd; | |
9761eea8 | 6426 | |
9c1cfda2 JH |
6427 | sd = &per_cpu(node_domains, j); |
6428 | sd->groups = sg; | |
9c1cfda2 | 6429 | } |
5517d86b | 6430 | sg->__cpu_power = 0; |
9c1cfda2 | 6431 | sg->cpumask = nodemask; |
51888ca2 | 6432 | sg->next = sg; |
9c1cfda2 JH |
6433 | cpus_or(covered, covered, nodemask); |
6434 | prev = sg; | |
6435 | ||
6436 | for (j = 0; j < MAX_NUMNODES; j++) { | |
6437 | cpumask_t tmp, notcovered; | |
6438 | int n = (i + j) % MAX_NUMNODES; | |
6439 | ||
6440 | cpus_complement(notcovered, covered); | |
6441 | cpus_and(tmp, notcovered, *cpu_map); | |
6442 | cpus_and(tmp, tmp, domainspan); | |
6443 | if (cpus_empty(tmp)) | |
6444 | break; | |
6445 | ||
6446 | nodemask = node_to_cpumask(n); | |
6447 | cpus_and(tmp, tmp, nodemask); | |
6448 | if (cpus_empty(tmp)) | |
6449 | continue; | |
6450 | ||
15f0b676 SV |
6451 | sg = kmalloc_node(sizeof(struct sched_group), |
6452 | GFP_KERNEL, i); | |
9c1cfda2 JH |
6453 | if (!sg) { |
6454 | printk(KERN_WARNING | |
6455 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 6456 | goto error; |
9c1cfda2 | 6457 | } |
5517d86b | 6458 | sg->__cpu_power = 0; |
9c1cfda2 | 6459 | sg->cpumask = tmp; |
51888ca2 | 6460 | sg->next = prev->next; |
9c1cfda2 JH |
6461 | cpus_or(covered, covered, tmp); |
6462 | prev->next = sg; | |
6463 | prev = sg; | |
6464 | } | |
9c1cfda2 | 6465 | } |
1da177e4 LT |
6466 | #endif |
6467 | ||
6468 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6469 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 6470 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6471 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
6472 | ||
89c4710e | 6473 | init_sched_groups_power(i, sd); |
5c45bf27 | 6474 | } |
1da177e4 | 6475 | #endif |
1e9f28fa | 6476 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 6477 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6478 | struct sched_domain *sd = &per_cpu(core_domains, i); |
6479 | ||
89c4710e | 6480 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6481 | } |
6482 | #endif | |
1e9f28fa | 6483 | |
5c45bf27 | 6484 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
6485 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
6486 | ||
89c4710e | 6487 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6488 | } |
6489 | ||
9c1cfda2 | 6490 | #ifdef CONFIG_NUMA |
08069033 SS |
6491 | for (i = 0; i < MAX_NUMNODES; i++) |
6492 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 6493 | |
6711cab4 SS |
6494 | if (sd_allnodes) { |
6495 | struct sched_group *sg; | |
f712c0c7 | 6496 | |
6711cab4 | 6497 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
f712c0c7 SS |
6498 | init_numa_sched_groups_power(sg); |
6499 | } | |
9c1cfda2 JH |
6500 | #endif |
6501 | ||
1da177e4 | 6502 | /* Attach the domains */ |
1a20ff27 | 6503 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
6504 | struct sched_domain *sd; |
6505 | #ifdef CONFIG_SCHED_SMT | |
6506 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
6507 | #elif defined(CONFIG_SCHED_MC) |
6508 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
6509 | #else |
6510 | sd = &per_cpu(phys_domains, i); | |
6511 | #endif | |
6512 | cpu_attach_domain(sd, i); | |
6513 | } | |
51888ca2 SV |
6514 | |
6515 | return 0; | |
6516 | ||
a616058b | 6517 | #ifdef CONFIG_NUMA |
51888ca2 SV |
6518 | error: |
6519 | free_sched_groups(cpu_map); | |
6520 | return -ENOMEM; | |
a616058b | 6521 | #endif |
1da177e4 | 6522 | } |
029190c5 PJ |
6523 | |
6524 | static cpumask_t *doms_cur; /* current sched domains */ | |
6525 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
6526 | ||
6527 | /* | |
6528 | * Special case: If a kmalloc of a doms_cur partition (array of | |
6529 | * cpumask_t) fails, then fallback to a single sched domain, | |
6530 | * as determined by the single cpumask_t fallback_doms. | |
6531 | */ | |
6532 | static cpumask_t fallback_doms; | |
6533 | ||
1a20ff27 | 6534 | /* |
41a2d6cf | 6535 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
6536 | * For now this just excludes isolated cpus, but could be used to |
6537 | * exclude other special cases in the future. | |
1a20ff27 | 6538 | */ |
51888ca2 | 6539 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 6540 | { |
7378547f MM |
6541 | int err; |
6542 | ||
029190c5 PJ |
6543 | ndoms_cur = 1; |
6544 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
6545 | if (!doms_cur) | |
6546 | doms_cur = &fallback_doms; | |
6547 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
7378547f | 6548 | err = build_sched_domains(doms_cur); |
6382bc90 | 6549 | register_sched_domain_sysctl(); |
7378547f MM |
6550 | |
6551 | return err; | |
1a20ff27 DG |
6552 | } |
6553 | ||
6554 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) | |
1da177e4 | 6555 | { |
51888ca2 | 6556 | free_sched_groups(cpu_map); |
9c1cfda2 | 6557 | } |
1da177e4 | 6558 | |
1a20ff27 DG |
6559 | /* |
6560 | * Detach sched domains from a group of cpus specified in cpu_map | |
6561 | * These cpus will now be attached to the NULL domain | |
6562 | */ | |
858119e1 | 6563 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 DG |
6564 | { |
6565 | int i; | |
6566 | ||
6382bc90 MM |
6567 | unregister_sched_domain_sysctl(); |
6568 | ||
1a20ff27 DG |
6569 | for_each_cpu_mask(i, *cpu_map) |
6570 | cpu_attach_domain(NULL, i); | |
6571 | synchronize_sched(); | |
6572 | arch_destroy_sched_domains(cpu_map); | |
6573 | } | |
6574 | ||
029190c5 PJ |
6575 | /* |
6576 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 6577 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
6578 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
6579 | * It destroys each deleted domain and builds each new domain. | |
6580 | * | |
6581 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
6582 | * The masks don't intersect (don't overlap.) We should setup one |
6583 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
6584 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
6585 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
6586 | * it as it is. | |
6587 | * | |
41a2d6cf IM |
6588 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
6589 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
6590 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
6591 | * and partition_sched_domains() will fallback to the single partition | |
6592 | * 'fallback_doms'. | |
6593 | * | |
6594 | * Call with hotplug lock held | |
6595 | */ | |
6596 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new) | |
6597 | { | |
6598 | int i, j; | |
6599 | ||
a1835615 SV |
6600 | lock_doms_cur(); |
6601 | ||
7378547f MM |
6602 | /* always unregister in case we don't destroy any domains */ |
6603 | unregister_sched_domain_sysctl(); | |
6604 | ||
029190c5 PJ |
6605 | if (doms_new == NULL) { |
6606 | ndoms_new = 1; | |
6607 | doms_new = &fallback_doms; | |
6608 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
6609 | } | |
6610 | ||
6611 | /* Destroy deleted domains */ | |
6612 | for (i = 0; i < ndoms_cur; i++) { | |
6613 | for (j = 0; j < ndoms_new; j++) { | |
6614 | if (cpus_equal(doms_cur[i], doms_new[j])) | |
6615 | goto match1; | |
6616 | } | |
6617 | /* no match - a current sched domain not in new doms_new[] */ | |
6618 | detach_destroy_domains(doms_cur + i); | |
6619 | match1: | |
6620 | ; | |
6621 | } | |
6622 | ||
6623 | /* Build new domains */ | |
6624 | for (i = 0; i < ndoms_new; i++) { | |
6625 | for (j = 0; j < ndoms_cur; j++) { | |
6626 | if (cpus_equal(doms_new[i], doms_cur[j])) | |
6627 | goto match2; | |
6628 | } | |
6629 | /* no match - add a new doms_new */ | |
6630 | build_sched_domains(doms_new + i); | |
6631 | match2: | |
6632 | ; | |
6633 | } | |
6634 | ||
6635 | /* Remember the new sched domains */ | |
6636 | if (doms_cur != &fallback_doms) | |
6637 | kfree(doms_cur); | |
6638 | doms_cur = doms_new; | |
6639 | ndoms_cur = ndoms_new; | |
7378547f MM |
6640 | |
6641 | register_sched_domain_sysctl(); | |
a1835615 SV |
6642 | |
6643 | unlock_doms_cur(); | |
029190c5 PJ |
6644 | } |
6645 | ||
5c45bf27 | 6646 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
6707de00 | 6647 | static int arch_reinit_sched_domains(void) |
5c45bf27 SS |
6648 | { |
6649 | int err; | |
6650 | ||
95402b38 | 6651 | get_online_cpus(); |
5c45bf27 SS |
6652 | detach_destroy_domains(&cpu_online_map); |
6653 | err = arch_init_sched_domains(&cpu_online_map); | |
95402b38 | 6654 | put_online_cpus(); |
5c45bf27 SS |
6655 | |
6656 | return err; | |
6657 | } | |
6658 | ||
6659 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
6660 | { | |
6661 | int ret; | |
6662 | ||
6663 | if (buf[0] != '0' && buf[0] != '1') | |
6664 | return -EINVAL; | |
6665 | ||
6666 | if (smt) | |
6667 | sched_smt_power_savings = (buf[0] == '1'); | |
6668 | else | |
6669 | sched_mc_power_savings = (buf[0] == '1'); | |
6670 | ||
6671 | ret = arch_reinit_sched_domains(); | |
6672 | ||
6673 | return ret ? ret : count; | |
6674 | } | |
6675 | ||
5c45bf27 SS |
6676 | #ifdef CONFIG_SCHED_MC |
6677 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
6678 | { | |
6679 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
6680 | } | |
48f24c4d IM |
6681 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
6682 | const char *buf, size_t count) | |
5c45bf27 SS |
6683 | { |
6684 | return sched_power_savings_store(buf, count, 0); | |
6685 | } | |
6707de00 AB |
6686 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
6687 | sched_mc_power_savings_store); | |
5c45bf27 SS |
6688 | #endif |
6689 | ||
6690 | #ifdef CONFIG_SCHED_SMT | |
6691 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
6692 | { | |
6693 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
6694 | } | |
48f24c4d IM |
6695 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
6696 | const char *buf, size_t count) | |
5c45bf27 SS |
6697 | { |
6698 | return sched_power_savings_store(buf, count, 1); | |
6699 | } | |
6707de00 AB |
6700 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
6701 | sched_smt_power_savings_store); | |
6702 | #endif | |
6703 | ||
6704 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
6705 | { | |
6706 | int err = 0; | |
6707 | ||
6708 | #ifdef CONFIG_SCHED_SMT | |
6709 | if (smt_capable()) | |
6710 | err = sysfs_create_file(&cls->kset.kobj, | |
6711 | &attr_sched_smt_power_savings.attr); | |
6712 | #endif | |
6713 | #ifdef CONFIG_SCHED_MC | |
6714 | if (!err && mc_capable()) | |
6715 | err = sysfs_create_file(&cls->kset.kobj, | |
6716 | &attr_sched_mc_power_savings.attr); | |
6717 | #endif | |
6718 | return err; | |
6719 | } | |
5c45bf27 SS |
6720 | #endif |
6721 | ||
1da177e4 | 6722 | /* |
41a2d6cf | 6723 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 6724 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 6725 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
6726 | * which will prevent rebalancing while the sched domains are recalculated. |
6727 | */ | |
6728 | static int update_sched_domains(struct notifier_block *nfb, | |
6729 | unsigned long action, void *hcpu) | |
6730 | { | |
1da177e4 LT |
6731 | switch (action) { |
6732 | case CPU_UP_PREPARE: | |
8bb78442 | 6733 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 6734 | case CPU_DOWN_PREPARE: |
8bb78442 | 6735 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 6736 | detach_destroy_domains(&cpu_online_map); |
1da177e4 LT |
6737 | return NOTIFY_OK; |
6738 | ||
6739 | case CPU_UP_CANCELED: | |
8bb78442 | 6740 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 6741 | case CPU_DOWN_FAILED: |
8bb78442 | 6742 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 6743 | case CPU_ONLINE: |
8bb78442 | 6744 | case CPU_ONLINE_FROZEN: |
1da177e4 | 6745 | case CPU_DEAD: |
8bb78442 | 6746 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
6747 | /* |
6748 | * Fall through and re-initialise the domains. | |
6749 | */ | |
6750 | break; | |
6751 | default: | |
6752 | return NOTIFY_DONE; | |
6753 | } | |
6754 | ||
6755 | /* The hotplug lock is already held by cpu_up/cpu_down */ | |
1a20ff27 | 6756 | arch_init_sched_domains(&cpu_online_map); |
1da177e4 LT |
6757 | |
6758 | return NOTIFY_OK; | |
6759 | } | |
1da177e4 LT |
6760 | |
6761 | void __init sched_init_smp(void) | |
6762 | { | |
5c1e1767 NP |
6763 | cpumask_t non_isolated_cpus; |
6764 | ||
95402b38 | 6765 | get_online_cpus(); |
1a20ff27 | 6766 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 6767 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
6768 | if (cpus_empty(non_isolated_cpus)) |
6769 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
95402b38 | 6770 | put_online_cpus(); |
1da177e4 LT |
6771 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
6772 | hotcpu_notifier(update_sched_domains, 0); | |
5c1e1767 NP |
6773 | |
6774 | /* Move init over to a non-isolated CPU */ | |
6775 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) | |
6776 | BUG(); | |
19978ca6 | 6777 | sched_init_granularity(); |
6b2d7700 SV |
6778 | |
6779 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6780 | if (nr_cpu_ids == 1) | |
6781 | return; | |
6782 | ||
6783 | lb_monitor_task = kthread_create(load_balance_monitor, NULL, | |
6784 | "group_balance"); | |
6785 | if (!IS_ERR(lb_monitor_task)) { | |
6786 | lb_monitor_task->flags |= PF_NOFREEZE; | |
6787 | wake_up_process(lb_monitor_task); | |
6788 | } else { | |
6789 | printk(KERN_ERR "Could not create load balance monitor thread" | |
6790 | "(error = %ld) \n", PTR_ERR(lb_monitor_task)); | |
6791 | } | |
6792 | #endif | |
1da177e4 LT |
6793 | } |
6794 | #else | |
6795 | void __init sched_init_smp(void) | |
6796 | { | |
19978ca6 | 6797 | sched_init_granularity(); |
1da177e4 LT |
6798 | } |
6799 | #endif /* CONFIG_SMP */ | |
6800 | ||
6801 | int in_sched_functions(unsigned long addr) | |
6802 | { | |
1da177e4 LT |
6803 | return in_lock_functions(addr) || |
6804 | (addr >= (unsigned long)__sched_text_start | |
6805 | && addr < (unsigned long)__sched_text_end); | |
6806 | } | |
6807 | ||
a9957449 | 6808 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
6809 | { |
6810 | cfs_rq->tasks_timeline = RB_ROOT; | |
dd41f596 IM |
6811 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6812 | cfs_rq->rq = rq; | |
6813 | #endif | |
67e9fb2a | 6814 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
6815 | } |
6816 | ||
1da177e4 LT |
6817 | void __init sched_init(void) |
6818 | { | |
476f3534 | 6819 | int highest_cpu = 0; |
dd41f596 IM |
6820 | int i, j; |
6821 | ||
0a945022 | 6822 | for_each_possible_cpu(i) { |
dd41f596 | 6823 | struct rt_prio_array *array; |
70b97a7f | 6824 | struct rq *rq; |
1da177e4 LT |
6825 | |
6826 | rq = cpu_rq(i); | |
6827 | spin_lock_init(&rq->lock); | |
fcb99371 | 6828 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 6829 | rq->nr_running = 0; |
dd41f596 IM |
6830 | rq->clock = 1; |
6831 | init_cfs_rq(&rq->cfs, rq); | |
6832 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
6833 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); | |
3a252015 IM |
6834 | { |
6835 | struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i); | |
6836 | struct sched_entity *se = | |
6837 | &per_cpu(init_sched_entity, i); | |
6838 | ||
6839 | init_cfs_rq_p[i] = cfs_rq; | |
6840 | init_cfs_rq(cfs_rq, rq); | |
4cf86d77 | 6841 | cfs_rq->tg = &init_task_group; |
3a252015 | 6842 | list_add(&cfs_rq->leaf_cfs_rq_list, |
29f59db3 SV |
6843 | &rq->leaf_cfs_rq_list); |
6844 | ||
3a252015 IM |
6845 | init_sched_entity_p[i] = se; |
6846 | se->cfs_rq = &rq->cfs; | |
6847 | se->my_q = cfs_rq; | |
4cf86d77 | 6848 | se->load.weight = init_task_group_load; |
9b5b7751 | 6849 | se->load.inv_weight = |
4cf86d77 | 6850 | div64_64(1ULL<<32, init_task_group_load); |
3a252015 IM |
6851 | se->parent = NULL; |
6852 | } | |
4cf86d77 | 6853 | init_task_group.shares = init_task_group_load; |
dd41f596 | 6854 | #endif |
1da177e4 | 6855 | |
dd41f596 IM |
6856 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
6857 | rq->cpu_load[j] = 0; | |
1da177e4 | 6858 | #ifdef CONFIG_SMP |
41c7ce9a | 6859 | rq->sd = NULL; |
1da177e4 | 6860 | rq->active_balance = 0; |
dd41f596 | 6861 | rq->next_balance = jiffies; |
1da177e4 | 6862 | rq->push_cpu = 0; |
0a2966b4 | 6863 | rq->cpu = i; |
1da177e4 LT |
6864 | rq->migration_thread = NULL; |
6865 | INIT_LIST_HEAD(&rq->migration_queue); | |
6866 | #endif | |
6867 | atomic_set(&rq->nr_iowait, 0); | |
6868 | ||
dd41f596 IM |
6869 | array = &rq->rt.active; |
6870 | for (j = 0; j < MAX_RT_PRIO; j++) { | |
6871 | INIT_LIST_HEAD(array->queue + j); | |
6872 | __clear_bit(j, array->bitmap); | |
1da177e4 | 6873 | } |
476f3534 | 6874 | highest_cpu = i; |
dd41f596 IM |
6875 | /* delimiter for bitsearch: */ |
6876 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
1da177e4 LT |
6877 | } |
6878 | ||
2dd73a4f | 6879 | set_load_weight(&init_task); |
b50f60ce | 6880 | |
e107be36 AK |
6881 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
6882 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
6883 | #endif | |
6884 | ||
c9819f45 | 6885 | #ifdef CONFIG_SMP |
476f3534 | 6886 | nr_cpu_ids = highest_cpu + 1; |
c9819f45 CL |
6887 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
6888 | #endif | |
6889 | ||
b50f60ce HC |
6890 | #ifdef CONFIG_RT_MUTEXES |
6891 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
6892 | #endif | |
6893 | ||
1da177e4 LT |
6894 | /* |
6895 | * The boot idle thread does lazy MMU switching as well: | |
6896 | */ | |
6897 | atomic_inc(&init_mm.mm_count); | |
6898 | enter_lazy_tlb(&init_mm, current); | |
6899 | ||
6900 | /* | |
6901 | * Make us the idle thread. Technically, schedule() should not be | |
6902 | * called from this thread, however somewhere below it might be, | |
6903 | * but because we are the idle thread, we just pick up running again | |
6904 | * when this runqueue becomes "idle". | |
6905 | */ | |
6906 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
6907 | /* |
6908 | * During early bootup we pretend to be a normal task: | |
6909 | */ | |
6910 | current->sched_class = &fair_sched_class; | |
1da177e4 LT |
6911 | } |
6912 | ||
6913 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
6914 | void __might_sleep(char *file, int line) | |
6915 | { | |
48f24c4d | 6916 | #ifdef in_atomic |
1da177e4 LT |
6917 | static unsigned long prev_jiffy; /* ratelimiting */ |
6918 | ||
6919 | if ((in_atomic() || irqs_disabled()) && | |
6920 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
6921 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
6922 | return; | |
6923 | prev_jiffy = jiffies; | |
91368d73 | 6924 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
6925 | " context at %s:%d\n", file, line); |
6926 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
6927 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 6928 | debug_show_held_locks(current); |
3117df04 IM |
6929 | if (irqs_disabled()) |
6930 | print_irqtrace_events(current); | |
1da177e4 LT |
6931 | dump_stack(); |
6932 | } | |
6933 | #endif | |
6934 | } | |
6935 | EXPORT_SYMBOL(__might_sleep); | |
6936 | #endif | |
6937 | ||
6938 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
6939 | static void normalize_task(struct rq *rq, struct task_struct *p) |
6940 | { | |
6941 | int on_rq; | |
6942 | update_rq_clock(rq); | |
6943 | on_rq = p->se.on_rq; | |
6944 | if (on_rq) | |
6945 | deactivate_task(rq, p, 0); | |
6946 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
6947 | if (on_rq) { | |
6948 | activate_task(rq, p, 0); | |
6949 | resched_task(rq->curr); | |
6950 | } | |
6951 | } | |
6952 | ||
1da177e4 LT |
6953 | void normalize_rt_tasks(void) |
6954 | { | |
a0f98a1c | 6955 | struct task_struct *g, *p; |
1da177e4 | 6956 | unsigned long flags; |
70b97a7f | 6957 | struct rq *rq; |
1da177e4 LT |
6958 | |
6959 | read_lock_irq(&tasklist_lock); | |
a0f98a1c | 6960 | do_each_thread(g, p) { |
178be793 IM |
6961 | /* |
6962 | * Only normalize user tasks: | |
6963 | */ | |
6964 | if (!p->mm) | |
6965 | continue; | |
6966 | ||
6cfb0d5d | 6967 | p->se.exec_start = 0; |
6cfb0d5d | 6968 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 6969 | p->se.wait_start = 0; |
dd41f596 | 6970 | p->se.sleep_start = 0; |
dd41f596 | 6971 | p->se.block_start = 0; |
6cfb0d5d | 6972 | #endif |
dd41f596 IM |
6973 | task_rq(p)->clock = 0; |
6974 | ||
6975 | if (!rt_task(p)) { | |
6976 | /* | |
6977 | * Renice negative nice level userspace | |
6978 | * tasks back to 0: | |
6979 | */ | |
6980 | if (TASK_NICE(p) < 0 && p->mm) | |
6981 | set_user_nice(p, 0); | |
1da177e4 | 6982 | continue; |
dd41f596 | 6983 | } |
1da177e4 | 6984 | |
b29739f9 IM |
6985 | spin_lock_irqsave(&p->pi_lock, flags); |
6986 | rq = __task_rq_lock(p); | |
1da177e4 | 6987 | |
178be793 | 6988 | normalize_task(rq, p); |
3a5e4dc1 | 6989 | |
b29739f9 IM |
6990 | __task_rq_unlock(rq); |
6991 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
a0f98a1c IM |
6992 | } while_each_thread(g, p); |
6993 | ||
1da177e4 LT |
6994 | read_unlock_irq(&tasklist_lock); |
6995 | } | |
6996 | ||
6997 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
6998 | |
6999 | #ifdef CONFIG_IA64 | |
7000 | /* | |
7001 | * These functions are only useful for the IA64 MCA handling. | |
7002 | * | |
7003 | * They can only be called when the whole system has been | |
7004 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7005 | * activity can take place. Using them for anything else would | |
7006 | * be a serious bug, and as a result, they aren't even visible | |
7007 | * under any other configuration. | |
7008 | */ | |
7009 | ||
7010 | /** | |
7011 | * curr_task - return the current task for a given cpu. | |
7012 | * @cpu: the processor in question. | |
7013 | * | |
7014 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7015 | */ | |
36c8b586 | 7016 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7017 | { |
7018 | return cpu_curr(cpu); | |
7019 | } | |
7020 | ||
7021 | /** | |
7022 | * set_curr_task - set the current task for a given cpu. | |
7023 | * @cpu: the processor in question. | |
7024 | * @p: the task pointer to set. | |
7025 | * | |
7026 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7027 | * are serviced on a separate stack. It allows the architecture to switch the |
7028 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7029 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7030 | * and caller must save the original value of the current task (see | |
7031 | * curr_task() above) and restore that value before reenabling interrupts and | |
7032 | * re-starting the system. | |
7033 | * | |
7034 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7035 | */ | |
36c8b586 | 7036 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7037 | { |
7038 | cpu_curr(cpu) = p; | |
7039 | } | |
7040 | ||
7041 | #endif | |
29f59db3 SV |
7042 | |
7043 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7044 | ||
6b2d7700 SV |
7045 | #ifdef CONFIG_SMP |
7046 | /* | |
7047 | * distribute shares of all task groups among their schedulable entities, | |
7048 | * to reflect load distrbution across cpus. | |
7049 | */ | |
7050 | static int rebalance_shares(struct sched_domain *sd, int this_cpu) | |
7051 | { | |
7052 | struct cfs_rq *cfs_rq; | |
7053 | struct rq *rq = cpu_rq(this_cpu); | |
7054 | cpumask_t sdspan = sd->span; | |
7055 | int balanced = 1; | |
7056 | ||
7057 | /* Walk thr' all the task groups that we have */ | |
7058 | for_each_leaf_cfs_rq(rq, cfs_rq) { | |
7059 | int i; | |
7060 | unsigned long total_load = 0, total_shares; | |
7061 | struct task_group *tg = cfs_rq->tg; | |
7062 | ||
7063 | /* Gather total task load of this group across cpus */ | |
7064 | for_each_cpu_mask(i, sdspan) | |
7065 | total_load += tg->cfs_rq[i]->load.weight; | |
7066 | ||
7067 | /* Nothing to do if this group has no load */ | |
7068 | if (!total_load) | |
7069 | continue; | |
7070 | ||
7071 | /* | |
7072 | * tg->shares represents the number of cpu shares the task group | |
7073 | * is eligible to hold on a single cpu. On N cpus, it is | |
7074 | * eligible to hold (N * tg->shares) number of cpu shares. | |
7075 | */ | |
7076 | total_shares = tg->shares * cpus_weight(sdspan); | |
7077 | ||
7078 | /* | |
7079 | * redistribute total_shares across cpus as per the task load | |
7080 | * distribution. | |
7081 | */ | |
7082 | for_each_cpu_mask(i, sdspan) { | |
7083 | unsigned long local_load, local_shares; | |
7084 | ||
7085 | local_load = tg->cfs_rq[i]->load.weight; | |
7086 | local_shares = (local_load * total_shares) / total_load; | |
7087 | if (!local_shares) | |
7088 | local_shares = MIN_GROUP_SHARES; | |
7089 | if (local_shares == tg->se[i]->load.weight) | |
7090 | continue; | |
7091 | ||
7092 | spin_lock_irq(&cpu_rq(i)->lock); | |
7093 | set_se_shares(tg->se[i], local_shares); | |
7094 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7095 | balanced = 0; | |
7096 | } | |
7097 | } | |
7098 | ||
7099 | return balanced; | |
7100 | } | |
7101 | ||
7102 | /* | |
7103 | * How frequently should we rebalance_shares() across cpus? | |
7104 | * | |
7105 | * The more frequently we rebalance shares, the more accurate is the fairness | |
7106 | * of cpu bandwidth distribution between task groups. However higher frequency | |
7107 | * also implies increased scheduling overhead. | |
7108 | * | |
7109 | * sysctl_sched_min_bal_int_shares represents the minimum interval between | |
7110 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7111 | * | |
7112 | * sysctl_sched_max_bal_int_shares represents the maximum interval between | |
7113 | * consecutive calls to rebalance_shares() in the same sched domain. | |
7114 | * | |
7115 | * These settings allows for the appropriate tradeoff between accuracy of | |
7116 | * fairness and the associated overhead. | |
7117 | * | |
7118 | */ | |
7119 | ||
7120 | /* default: 8ms, units: milliseconds */ | |
7121 | const_debug unsigned int sysctl_sched_min_bal_int_shares = 8; | |
7122 | ||
7123 | /* default: 128ms, units: milliseconds */ | |
7124 | const_debug unsigned int sysctl_sched_max_bal_int_shares = 128; | |
7125 | ||
7126 | /* kernel thread that runs rebalance_shares() periodically */ | |
7127 | static int load_balance_monitor(void *unused) | |
7128 | { | |
7129 | unsigned int timeout = sysctl_sched_min_bal_int_shares; | |
7130 | struct sched_param schedparm; | |
7131 | int ret; | |
7132 | ||
7133 | /* | |
7134 | * We don't want this thread's execution to be limited by the shares | |
7135 | * assigned to default group (init_task_group). Hence make it run | |
7136 | * as a SCHED_RR RT task at the lowest priority. | |
7137 | */ | |
7138 | schedparm.sched_priority = 1; | |
7139 | ret = sched_setscheduler(current, SCHED_RR, &schedparm); | |
7140 | if (ret) | |
7141 | printk(KERN_ERR "Couldn't set SCHED_RR policy for load balance" | |
7142 | " monitor thread (error = %d) \n", ret); | |
7143 | ||
7144 | while (!kthread_should_stop()) { | |
7145 | int i, cpu, balanced = 1; | |
7146 | ||
7147 | /* Prevent cpus going down or coming up */ | |
86ef5c9a | 7148 | get_online_cpus(); |
6b2d7700 SV |
7149 | /* lockout changes to doms_cur[] array */ |
7150 | lock_doms_cur(); | |
7151 | /* | |
7152 | * Enter a rcu read-side critical section to safely walk rq->sd | |
7153 | * chain on various cpus and to walk task group list | |
7154 | * (rq->leaf_cfs_rq_list) in rebalance_shares(). | |
7155 | */ | |
7156 | rcu_read_lock(); | |
7157 | ||
7158 | for (i = 0; i < ndoms_cur; i++) { | |
7159 | cpumask_t cpumap = doms_cur[i]; | |
7160 | struct sched_domain *sd = NULL, *sd_prev = NULL; | |
7161 | ||
7162 | cpu = first_cpu(cpumap); | |
7163 | ||
7164 | /* Find the highest domain at which to balance shares */ | |
7165 | for_each_domain(cpu, sd) { | |
7166 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
7167 | continue; | |
7168 | sd_prev = sd; | |
7169 | } | |
7170 | ||
7171 | sd = sd_prev; | |
7172 | /* sd == NULL? No load balance reqd in this domain */ | |
7173 | if (!sd) | |
7174 | continue; | |
7175 | ||
7176 | balanced &= rebalance_shares(sd, cpu); | |
7177 | } | |
7178 | ||
7179 | rcu_read_unlock(); | |
7180 | ||
7181 | unlock_doms_cur(); | |
86ef5c9a | 7182 | put_online_cpus(); |
6b2d7700 SV |
7183 | |
7184 | if (!balanced) | |
7185 | timeout = sysctl_sched_min_bal_int_shares; | |
7186 | else if (timeout < sysctl_sched_max_bal_int_shares) | |
7187 | timeout *= 2; | |
7188 | ||
7189 | msleep_interruptible(timeout); | |
7190 | } | |
7191 | ||
7192 | return 0; | |
7193 | } | |
7194 | #endif /* CONFIG_SMP */ | |
7195 | ||
29f59db3 | 7196 | /* allocate runqueue etc for a new task group */ |
4cf86d77 | 7197 | struct task_group *sched_create_group(void) |
29f59db3 | 7198 | { |
4cf86d77 | 7199 | struct task_group *tg; |
29f59db3 SV |
7200 | struct cfs_rq *cfs_rq; |
7201 | struct sched_entity *se; | |
9b5b7751 | 7202 | struct rq *rq; |
29f59db3 SV |
7203 | int i; |
7204 | ||
29f59db3 SV |
7205 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); |
7206 | if (!tg) | |
7207 | return ERR_PTR(-ENOMEM); | |
7208 | ||
9b5b7751 | 7209 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7210 | if (!tg->cfs_rq) |
7211 | goto err; | |
9b5b7751 | 7212 | tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL); |
29f59db3 SV |
7213 | if (!tg->se) |
7214 | goto err; | |
7215 | ||
7216 | for_each_possible_cpu(i) { | |
9b5b7751 | 7217 | rq = cpu_rq(i); |
29f59db3 SV |
7218 | |
7219 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL, | |
7220 | cpu_to_node(i)); | |
7221 | if (!cfs_rq) | |
7222 | goto err; | |
7223 | ||
7224 | se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL, | |
7225 | cpu_to_node(i)); | |
7226 | if (!se) | |
7227 | goto err; | |
7228 | ||
7229 | memset(cfs_rq, 0, sizeof(struct cfs_rq)); | |
7230 | memset(se, 0, sizeof(struct sched_entity)); | |
7231 | ||
7232 | tg->cfs_rq[i] = cfs_rq; | |
7233 | init_cfs_rq(cfs_rq, rq); | |
7234 | cfs_rq->tg = tg; | |
29f59db3 SV |
7235 | |
7236 | tg->se[i] = se; | |
7237 | se->cfs_rq = &rq->cfs; | |
7238 | se->my_q = cfs_rq; | |
7239 | se->load.weight = NICE_0_LOAD; | |
7240 | se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD); | |
7241 | se->parent = NULL; | |
7242 | } | |
7243 | ||
ec2c507f SV |
7244 | tg->shares = NICE_0_LOAD; |
7245 | ||
7246 | lock_task_group_list(); | |
9b5b7751 SV |
7247 | for_each_possible_cpu(i) { |
7248 | rq = cpu_rq(i); | |
7249 | cfs_rq = tg->cfs_rq[i]; | |
7250 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7251 | } | |
ec2c507f | 7252 | unlock_task_group_list(); |
29f59db3 | 7253 | |
9b5b7751 | 7254 | return tg; |
29f59db3 SV |
7255 | |
7256 | err: | |
7257 | for_each_possible_cpu(i) { | |
a65914b3 | 7258 | if (tg->cfs_rq) |
29f59db3 | 7259 | kfree(tg->cfs_rq[i]); |
a65914b3 | 7260 | if (tg->se) |
29f59db3 SV |
7261 | kfree(tg->se[i]); |
7262 | } | |
a65914b3 IM |
7263 | kfree(tg->cfs_rq); |
7264 | kfree(tg->se); | |
7265 | kfree(tg); | |
29f59db3 SV |
7266 | |
7267 | return ERR_PTR(-ENOMEM); | |
7268 | } | |
7269 | ||
9b5b7751 SV |
7270 | /* rcu callback to free various structures associated with a task group */ |
7271 | static void free_sched_group(struct rcu_head *rhp) | |
29f59db3 | 7272 | { |
ae8393e5 SV |
7273 | struct task_group *tg = container_of(rhp, struct task_group, rcu); |
7274 | struct cfs_rq *cfs_rq; | |
29f59db3 SV |
7275 | struct sched_entity *se; |
7276 | int i; | |
7277 | ||
29f59db3 SV |
7278 | /* now it should be safe to free those cfs_rqs */ |
7279 | for_each_possible_cpu(i) { | |
7280 | cfs_rq = tg->cfs_rq[i]; | |
7281 | kfree(cfs_rq); | |
7282 | ||
7283 | se = tg->se[i]; | |
7284 | kfree(se); | |
7285 | } | |
7286 | ||
7287 | kfree(tg->cfs_rq); | |
7288 | kfree(tg->se); | |
7289 | kfree(tg); | |
7290 | } | |
7291 | ||
9b5b7751 | 7292 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 7293 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 7294 | { |
7bae49d4 | 7295 | struct cfs_rq *cfs_rq = NULL; |
9b5b7751 | 7296 | int i; |
29f59db3 | 7297 | |
ec2c507f | 7298 | lock_task_group_list(); |
9b5b7751 SV |
7299 | for_each_possible_cpu(i) { |
7300 | cfs_rq = tg->cfs_rq[i]; | |
7301 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7302 | } | |
ec2c507f | 7303 | unlock_task_group_list(); |
9b5b7751 | 7304 | |
7bae49d4 | 7305 | BUG_ON(!cfs_rq); |
9b5b7751 SV |
7306 | |
7307 | /* wait for possible concurrent references to cfs_rqs complete */ | |
ae8393e5 | 7308 | call_rcu(&tg->rcu, free_sched_group); |
29f59db3 SV |
7309 | } |
7310 | ||
9b5b7751 | 7311 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
7312 | * The caller of this function should have put the task in its new group |
7313 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
7314 | * reflect its new group. | |
9b5b7751 SV |
7315 | */ |
7316 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
7317 | { |
7318 | int on_rq, running; | |
7319 | unsigned long flags; | |
7320 | struct rq *rq; | |
7321 | ||
7322 | rq = task_rq_lock(tsk, &flags); | |
7323 | ||
dae51f56 | 7324 | if (tsk->sched_class != &fair_sched_class) { |
ce96b5ac | 7325 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7326 | goto done; |
dae51f56 | 7327 | } |
29f59db3 SV |
7328 | |
7329 | update_rq_clock(rq); | |
7330 | ||
051a1d1a | 7331 | running = task_current(rq, tsk); |
29f59db3 SV |
7332 | on_rq = tsk->se.on_rq; |
7333 | ||
83b699ed | 7334 | if (on_rq) { |
29f59db3 | 7335 | dequeue_task(rq, tsk, 0); |
83b699ed SV |
7336 | if (unlikely(running)) |
7337 | tsk->sched_class->put_prev_task(rq, tsk); | |
7338 | } | |
29f59db3 | 7339 | |
ce96b5ac | 7340 | set_task_cfs_rq(tsk, task_cpu(tsk)); |
29f59db3 | 7341 | |
83b699ed SV |
7342 | if (on_rq) { |
7343 | if (unlikely(running)) | |
7344 | tsk->sched_class->set_curr_task(rq); | |
7074badb | 7345 | enqueue_task(rq, tsk, 0); |
83b699ed | 7346 | } |
29f59db3 SV |
7347 | |
7348 | done: | |
7349 | task_rq_unlock(rq, &flags); | |
7350 | } | |
7351 | ||
6b2d7700 | 7352 | /* rq->lock to be locked by caller */ |
29f59db3 SV |
7353 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
7354 | { | |
7355 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
7356 | struct rq *rq = cfs_rq->rq; | |
7357 | int on_rq; | |
7358 | ||
6b2d7700 SV |
7359 | if (!shares) |
7360 | shares = MIN_GROUP_SHARES; | |
29f59db3 SV |
7361 | |
7362 | on_rq = se->on_rq; | |
6b2d7700 | 7363 | if (on_rq) { |
29f59db3 | 7364 | dequeue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7365 | dec_cpu_load(rq, se->load.weight); |
7366 | } | |
29f59db3 SV |
7367 | |
7368 | se->load.weight = shares; | |
7369 | se->load.inv_weight = div64_64((1ULL<<32), shares); | |
7370 | ||
6b2d7700 | 7371 | if (on_rq) { |
29f59db3 | 7372 | enqueue_entity(cfs_rq, se, 0); |
6b2d7700 SV |
7373 | inc_cpu_load(rq, se->load.weight); |
7374 | } | |
29f59db3 SV |
7375 | } |
7376 | ||
4cf86d77 | 7377 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
7378 | { |
7379 | int i; | |
6b2d7700 SV |
7380 | struct cfs_rq *cfs_rq; |
7381 | struct rq *rq; | |
c61935fd | 7382 | |
ec2c507f | 7383 | lock_task_group_list(); |
9b5b7751 | 7384 | if (tg->shares == shares) |
5cb350ba | 7385 | goto done; |
29f59db3 | 7386 | |
6b2d7700 SV |
7387 | if (shares < MIN_GROUP_SHARES) |
7388 | shares = MIN_GROUP_SHARES; | |
7389 | ||
7390 | /* | |
7391 | * Prevent any load balance activity (rebalance_shares, | |
7392 | * load_balance_fair) from referring to this group first, | |
7393 | * by taking it off the rq->leaf_cfs_rq_list on each cpu. | |
7394 | */ | |
7395 | for_each_possible_cpu(i) { | |
7396 | cfs_rq = tg->cfs_rq[i]; | |
7397 | list_del_rcu(&cfs_rq->leaf_cfs_rq_list); | |
7398 | } | |
7399 | ||
7400 | /* wait for any ongoing reference to this group to finish */ | |
7401 | synchronize_sched(); | |
7402 | ||
7403 | /* | |
7404 | * Now we are free to modify the group's share on each cpu | |
7405 | * w/o tripping rebalance_share or load_balance_fair. | |
7406 | */ | |
9b5b7751 | 7407 | tg->shares = shares; |
6b2d7700 SV |
7408 | for_each_possible_cpu(i) { |
7409 | spin_lock_irq(&cpu_rq(i)->lock); | |
9b5b7751 | 7410 | set_se_shares(tg->se[i], shares); |
6b2d7700 SV |
7411 | spin_unlock_irq(&cpu_rq(i)->lock); |
7412 | } | |
29f59db3 | 7413 | |
6b2d7700 SV |
7414 | /* |
7415 | * Enable load balance activity on this group, by inserting it back on | |
7416 | * each cpu's rq->leaf_cfs_rq_list. | |
7417 | */ | |
7418 | for_each_possible_cpu(i) { | |
7419 | rq = cpu_rq(i); | |
7420 | cfs_rq = tg->cfs_rq[i]; | |
7421 | list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7422 | } | |
5cb350ba | 7423 | done: |
ec2c507f | 7424 | unlock_task_group_list(); |
9b5b7751 | 7425 | return 0; |
29f59db3 SV |
7426 | } |
7427 | ||
5cb350ba DG |
7428 | unsigned long sched_group_shares(struct task_group *tg) |
7429 | { | |
7430 | return tg->shares; | |
7431 | } | |
7432 | ||
3a252015 | 7433 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e SV |
7434 | |
7435 | #ifdef CONFIG_FAIR_CGROUP_SCHED | |
7436 | ||
7437 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 7438 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 7439 | { |
2b01dfe3 PM |
7440 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
7441 | struct task_group, css); | |
68318b8e SV |
7442 | } |
7443 | ||
7444 | static struct cgroup_subsys_state * | |
2b01dfe3 | 7445 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e SV |
7446 | { |
7447 | struct task_group *tg; | |
7448 | ||
2b01dfe3 | 7449 | if (!cgrp->parent) { |
68318b8e | 7450 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 7451 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
7452 | return &init_task_group.css; |
7453 | } | |
7454 | ||
7455 | /* we support only 1-level deep hierarchical scheduler atm */ | |
2b01dfe3 | 7456 | if (cgrp->parent->parent) |
68318b8e SV |
7457 | return ERR_PTR(-EINVAL); |
7458 | ||
7459 | tg = sched_create_group(); | |
7460 | if (IS_ERR(tg)) | |
7461 | return ERR_PTR(-ENOMEM); | |
7462 | ||
7463 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 7464 | tg->css.cgroup = cgrp; |
68318b8e SV |
7465 | |
7466 | return &tg->css; | |
7467 | } | |
7468 | ||
41a2d6cf IM |
7469 | static void |
7470 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 7471 | { |
2b01dfe3 | 7472 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7473 | |
7474 | sched_destroy_group(tg); | |
7475 | } | |
7476 | ||
41a2d6cf IM |
7477 | static int |
7478 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
7479 | struct task_struct *tsk) | |
68318b8e SV |
7480 | { |
7481 | /* We don't support RT-tasks being in separate groups */ | |
7482 | if (tsk->sched_class != &fair_sched_class) | |
7483 | return -EINVAL; | |
7484 | ||
7485 | return 0; | |
7486 | } | |
7487 | ||
7488 | static void | |
2b01dfe3 | 7489 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
7490 | struct cgroup *old_cont, struct task_struct *tsk) |
7491 | { | |
7492 | sched_move_task(tsk); | |
7493 | } | |
7494 | ||
2b01dfe3 PM |
7495 | static int cpu_shares_write_uint(struct cgroup *cgrp, struct cftype *cftype, |
7496 | u64 shareval) | |
68318b8e | 7497 | { |
2b01dfe3 | 7498 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
7499 | } |
7500 | ||
2b01dfe3 | 7501 | static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 7502 | { |
2b01dfe3 | 7503 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
7504 | |
7505 | return (u64) tg->shares; | |
7506 | } | |
7507 | ||
fe5c7cc2 PM |
7508 | static struct cftype cpu_files[] = { |
7509 | { | |
7510 | .name = "shares", | |
7511 | .read_uint = cpu_shares_read_uint, | |
7512 | .write_uint = cpu_shares_write_uint, | |
7513 | }, | |
68318b8e SV |
7514 | }; |
7515 | ||
7516 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7517 | { | |
fe5c7cc2 | 7518 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
7519 | } |
7520 | ||
7521 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
7522 | .name = "cpu", |
7523 | .create = cpu_cgroup_create, | |
7524 | .destroy = cpu_cgroup_destroy, | |
7525 | .can_attach = cpu_cgroup_can_attach, | |
7526 | .attach = cpu_cgroup_attach, | |
7527 | .populate = cpu_cgroup_populate, | |
7528 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
7529 | .early_init = 1, |
7530 | }; | |
7531 | ||
7532 | #endif /* CONFIG_FAIR_CGROUP_SCHED */ | |
d842de87 SV |
7533 | |
7534 | #ifdef CONFIG_CGROUP_CPUACCT | |
7535 | ||
7536 | /* | |
7537 | * CPU accounting code for task groups. | |
7538 | * | |
7539 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
7540 | * (balbir@in.ibm.com). | |
7541 | */ | |
7542 | ||
7543 | /* track cpu usage of a group of tasks */ | |
7544 | struct cpuacct { | |
7545 | struct cgroup_subsys_state css; | |
7546 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
7547 | u64 *cpuusage; | |
7548 | }; | |
7549 | ||
7550 | struct cgroup_subsys cpuacct_subsys; | |
7551 | ||
7552 | /* return cpu accounting group corresponding to this container */ | |
7553 | static inline struct cpuacct *cgroup_ca(struct cgroup *cont) | |
7554 | { | |
7555 | return container_of(cgroup_subsys_state(cont, cpuacct_subsys_id), | |
7556 | struct cpuacct, css); | |
7557 | } | |
7558 | ||
7559 | /* return cpu accounting group to which this task belongs */ | |
7560 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
7561 | { | |
7562 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
7563 | struct cpuacct, css); | |
7564 | } | |
7565 | ||
7566 | /* create a new cpu accounting group */ | |
7567 | static struct cgroup_subsys_state *cpuacct_create( | |
7568 | struct cgroup_subsys *ss, struct cgroup *cont) | |
7569 | { | |
7570 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
7571 | ||
7572 | if (!ca) | |
7573 | return ERR_PTR(-ENOMEM); | |
7574 | ||
7575 | ca->cpuusage = alloc_percpu(u64); | |
7576 | if (!ca->cpuusage) { | |
7577 | kfree(ca); | |
7578 | return ERR_PTR(-ENOMEM); | |
7579 | } | |
7580 | ||
7581 | return &ca->css; | |
7582 | } | |
7583 | ||
7584 | /* destroy an existing cpu accounting group */ | |
41a2d6cf IM |
7585 | static void |
7586 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cont) | |
d842de87 SV |
7587 | { |
7588 | struct cpuacct *ca = cgroup_ca(cont); | |
7589 | ||
7590 | free_percpu(ca->cpuusage); | |
7591 | kfree(ca); | |
7592 | } | |
7593 | ||
7594 | /* return total cpu usage (in nanoseconds) of a group */ | |
7595 | static u64 cpuusage_read(struct cgroup *cont, struct cftype *cft) | |
7596 | { | |
7597 | struct cpuacct *ca = cgroup_ca(cont); | |
7598 | u64 totalcpuusage = 0; | |
7599 | int i; | |
7600 | ||
7601 | for_each_possible_cpu(i) { | |
7602 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
7603 | ||
7604 | /* | |
7605 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
7606 | * platforms. | |
7607 | */ | |
7608 | spin_lock_irq(&cpu_rq(i)->lock); | |
7609 | totalcpuusage += *cpuusage; | |
7610 | spin_unlock_irq(&cpu_rq(i)->lock); | |
7611 | } | |
7612 | ||
7613 | return totalcpuusage; | |
7614 | } | |
7615 | ||
7616 | static struct cftype files[] = { | |
7617 | { | |
7618 | .name = "usage", | |
7619 | .read_uint = cpuusage_read, | |
7620 | }, | |
7621 | }; | |
7622 | ||
7623 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
7624 | { | |
7625 | return cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); | |
7626 | } | |
7627 | ||
7628 | /* | |
7629 | * charge this task's execution time to its accounting group. | |
7630 | * | |
7631 | * called with rq->lock held. | |
7632 | */ | |
7633 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
7634 | { | |
7635 | struct cpuacct *ca; | |
7636 | ||
7637 | if (!cpuacct_subsys.active) | |
7638 | return; | |
7639 | ||
7640 | ca = task_ca(tsk); | |
7641 | if (ca) { | |
7642 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
7643 | ||
7644 | *cpuusage += cputime; | |
7645 | } | |
7646 | } | |
7647 | ||
7648 | struct cgroup_subsys cpuacct_subsys = { | |
7649 | .name = "cpuacct", | |
7650 | .create = cpuacct_create, | |
7651 | .destroy = cpuacct_destroy, | |
7652 | .populate = cpuacct_populate, | |
7653 | .subsys_id = cpuacct_subsys_id, | |
7654 | }; | |
7655 | #endif /* CONFIG_CGROUP_CPUACCT */ |